commit e904b667c67c7a8c023fde2f281823f8d8b1a535
Author: wehub-resource-sync <resource-sync@wehub.local>
Date:   Mon Jul 13 11:59:26 2026 +0800

    chore: import upstream snapshot with attribution

diff --git a/.github/ISSUE_TEMPLATE/bug-report.yml b/.github/ISSUE_TEMPLATE/bug-report.yml
new file mode 100644
index 0000000..612c1b6
--- /dev/null
+++ b/.github/ISSUE_TEMPLATE/bug-report.yml
@@ -0,0 +1,62 @@
+# PaddleOCR
+
+name: 🐛 Bug Report
+description: Problems with PaddleOCR
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Thank you for submitting a PaddleOCR 🐛 Bug Report!
+
+  - type: checkboxes
+    attributes:
+      label: 🔎 Search before asking
+      description: >
+        Please search the PaddleOCR [Docs](https://paddlepaddle.github.io/PaddleOCR/), [Issues](https://github.com/PaddlePaddle/PaddleOCR/issues) and [Discussions](https://github.com/PaddlePaddle/PaddleOCR/discussions) to see if a similar bug report already exists.
+      options:
+        - label: I have searched the PaddleOCR [Docs](https://paddlepaddle.github.io/PaddleOCR/) and found no similar bug report.
+          required: true
+        - label: I have searched the PaddleOCR [Issues](https://github.com/PaddlePaddle/PaddleOCR/issues) and found no similar bug report.
+          required: true
+        - label: I have searched the PaddleOCR [Discussions](https://github.com/PaddlePaddle/PaddleOCR/discussions) and found no similar bug report.
+          required: true
+
+  - type: textarea
+    attributes:
+      label: 🐛 Bug (问题描述)
+      description: Provide console output with error messages and/or screenshots of the bug. (请提供详细报错信息或者截图)
+      placeholder: |
+        💡 ProTip! Include as much information as possible (screenshots, logs, tracebacks etc.) to receive the most helpful response.
+    validations:
+      required: true
+
+  - type: textarea
+    attributes:
+      label: 🏃‍♂️ Environment (运行环境)
+      description: Please specify the software and hardware you used to produce the bug. (请给出详细依赖包信息，便于复现问题)
+      placeholder: |
+        ```bash
+        OS                  macOS-13.5.2
+        Environment         Jupyter
+        Python              3.11.2
+        PaddleOCR           2.8.1
+        Install             git
+        RAM                 16.00 GB
+        CPU                 Apple M2
+        CUDA                None
+        ```
+    validations:
+      required: true
+
+  - type: textarea
+    attributes:
+      label: 🌰 Minimal Reproducible Example (最小可复现问题的Demo)
+      description: >
+        When asking a question, people will be better able to provide help if you provide code that they can easily understand and use to **reproduce** the problem.
+        This is referred to by community members as creating a [minimal reproducible example](https://stackoverflow.com/help/minimal-reproducible-example). (请务必提供该Demo，这样节省大家时间)
+      placeholder: |
+        ```bash
+        # Code to reproduce your issue here
+        ```
+    validations:
+      required: true
diff --git a/.github/ISSUE_TEMPLATE/config.yml b/.github/ISSUE_TEMPLATE/config.yml
new file mode 100644
index 0000000..e427482
--- /dev/null
+++ b/.github/ISSUE_TEMPLATE/config.yml
@@ -0,0 +1,11 @@
+blank_issues_enabled: false
+contact_links:
+  - name: 🙏 Q&A
+    url: https://github.com/PaddlePaddle/PaddleOCR/discussions/categories/q-a
+    about: Ask the community for help
+  - name: 💡 Feature requests and ideas
+    url: https://github.com/PaddlePaddle/PaddleOCR/discussions/categories/ideas
+    about: Share ideas for new features
+  - name: 🙌 Show and tell
+    url: https://github.com/PaddlePaddle/PaddleOCR/discussions/categories/show-and-tell
+    about: Show off something you've made
diff --git a/.github/actions/detect-docs-only/action.yml b/.github/actions/detect-docs-only/action.yml
new file mode 100644
index 0000000..b09905f
--- /dev/null
+++ b/.github/actions/detect-docs-only/action.yml
@@ -0,0 +1,55 @@
+name: Detect Docs-Only Change
+description: >
+  Output docs_only=true if every changed file in the current pull_request
+  matches docs/**, **/*.md, or .github/**. On push or workflow_dispatch,
+  always output docs_only=false.
+outputs:
+  docs_only:
+    description: "true if change is docs-only, otherwise false"
+    value: ${{ steps.compute.outputs.docs_only }}
+runs:
+  using: composite
+  steps:
+    - id: compute
+      shell: bash
+      env:
+        GITHUB_EVENT_NAME: ${{ github.event_name }}
+        GITHUB_BASE_REF: ${{ github.base_ref }}
+      run: |
+        set -euo pipefail
+        if [ "${GITHUB_EVENT_NAME}" != "pull_request" ]; then
+          echo "docs_only=false" >> "$GITHUB_OUTPUT"
+          exit 0
+        fi
+        git fetch origin "${GITHUB_BASE_REF}" --depth=1 >/dev/null 2>&1 || true
+        CHANGED_FILES="$(git diff --name-only "origin/${GITHUB_BASE_REF}...HEAD")"
+        # >>> matcher-begin
+        shopt -s globstar extglob nullglob
+        is_docs_only() {
+          local files="$1"
+          if [ -z "$files" ]; then
+            echo "false"; return 0
+          fi
+          while IFS= read -r f; do
+            [ -z "$f" ] && continue
+            case "$f" in
+              docs/*) ;;
+              .github/*) ;;
+              *.md) ;;
+              *)
+                # also accept nested */**/*.md via case glob
+                case "$f" in
+                  */*.md) ;;
+                  *) echo "false"; return 0 ;;
+                esac
+                ;;
+            esac
+          done <<< "$files"
+          echo "true"
+        }
+        result="$(is_docs_only "${CHANGED_FILES}")"
+        if [ -n "${GITHUB_OUTPUT:-}" ] && [ "${GITHUB_OUTPUT}" != "/dev/null" ]; then
+          echo "docs_only=${result}" >> "$GITHUB_OUTPUT"
+        fi
+        echo "${result}"
+        # <<< matcher-end
diff --git a/.github/dependabot.yml b/.github/dependabot.yml
new file mode 100644
index 0000000..be006de
--- /dev/null
+++ b/.github/dependabot.yml
@@ -0,0 +1,13 @@
+# Keep GitHub Actions up to date with GitHub's Dependabot...
+# https://docs.github.com/en/code-security/dependabot/working-with-dependabot/keeping-your-actions-up-to-date-with-dependabot
+# https://docs.github.com/en/code-security/dependabot/dependabot-version-updates/configuration-options-for-the-dependabot.yml-file#package-ecosystem
+version: 2
+updates:
+  - package-ecosystem: github-actions
+    directory: /
+    groups:
+      github-actions:
+        patterns:
+          - "*"  # Group all Actions updates into a single larger pull request
+    schedule:
+      interval: weekly
diff --git a/.github/workflows/build_publish_develop_docs.yml b/.github/workflows/build_publish_develop_docs.yml
new file mode 100644
index 0000000..159f41a
--- /dev/null
+++ b/.github/workflows/build_publish_develop_docs.yml
@@ -0,0 +1,39 @@
+name: Build/Publish Develop Docs
+on:
+  push:
+    branches:
+      - master
+      - main
+permissions:
+  contents: write
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+      - name: Configure Git Credentials
+        run: |
+          git config user.name github-actions[bot]
+          git config user.email 41898282+github-actions[bot]@users.noreply.github.com
+      - uses: actions/setup-python@v6
+        with:
+          python-version: 3.x
+      - run: echo "cache_id=$(date --utc '+%V')" >> $GITHUB_ENV
+      - uses: actions/cache@v5
+        with:
+          key: mkdocs-material-${{ env.cache_id }}
+          path: .cache
+          restore-keys: |
+            mkdocs-material-
+      - run: pip install mike mkdocs-material jieba mkdocs-git-revision-date-localized-plugin mkdocs-git-committers-plugin-2 mkdocs-static-i18n markdown-callouts
+      - name: Check docs GitHub source links
+        run: python tools/check_docs_github_links.py --repo-slug PaddlePaddle/PaddleOCR --forbidden-ref main --forbidden-ref master
+      - name: Resolve docs GitHub source refs
+        run: python tools/resolve_doc_github_refs.py --placeholder '{{PADDLEOCR_GITHUB_REF}}' --source-ref "${REF_NAME}"
+        env:
+          REF_NAME: ${{ github.ref_name }}
+      - run: |
+          git fetch origin gh-pages --depth=1
+          mike deploy --push --update-aliases main latest
+        env:
+          DOCS_EDIT_URI: edit/${{ github.ref_name }}/docs/
diff --git a/.github/workflows/build_publish_release_docs.yml b/.github/workflows/build_publish_release_docs.yml
new file mode 100644
index 0000000..6d91593
--- /dev/null
+++ b/.github/workflows/build_publish_release_docs.yml
@@ -0,0 +1,41 @@
+name: Build/Publish Release Docs
+on:
+  push:
+    tags:
+      - v*
+
+permissions:
+  contents: write
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+      - name: Configure Git Credentials
+        run: |
+          git config user.name github-actions[bot]
+          git config user.email github-actions[bot]@users.noreply.github.com
+      - uses: actions/setup-python@v6
+        with:
+          python-version: 3.x
+      - run: echo "cache_id=$(date --utc '+%V')" >> $GITHUB_ENV
+      - uses: actions/cache@v5
+        with:
+          key: mkdocs-material-${{ env.cache_id }}
+          path: .cache
+          restore-keys: |
+            mkdocs-material-
+      - run: pip install mike mkdocs-material jieba mkdocs-git-revision-date-localized-plugin mkdocs-git-committers-plugin-2 mkdocs-static-i18n markdown-callouts
+      - name: Check docs GitHub source links
+        run: python tools/check_docs_github_links.py --repo-slug PaddlePaddle/PaddleOCR --forbidden-ref main --forbidden-ref master
+      - name: Resolve docs GitHub source refs
+        run: python tools/resolve_doc_github_refs.py --placeholder '{{PADDLEOCR_GITHUB_REF}}' --source-ref "${REF_NAME}"
+        env:
+          REF_NAME: ${{ github.ref_name }}
+      - run: |
+          git fetch origin gh-pages --depth=1
+          mike deploy --push "${REF_NAME}"
+
+        env:
+          REF_NAME: ${{ github.ref_name }}
+          DOCS_EDIT_URI: blob/${{ github.ref_name }}/docs/
diff --git a/.github/workflows/close_inactive_issues.yml b/.github/workflows/close_inactive_issues.yml
new file mode 100644
index 0000000..5046e62
--- /dev/null
+++ b/.github/workflows/close_inactive_issues.yml
@@ -0,0 +1,23 @@
+name: Close inactive issues
+on:
+  schedule:
+    - cron: "30 1 * * *"
+
+jobs:
+  close-issues:
+    runs-on: ubuntu-latest
+    permissions:
+      issues: write
+      pull-requests: write
+    steps:
+      - uses: actions/stale@v10
+        with:
+          days-before-issue-stale: 90
+          days-before-issue-close: 14
+          stale-issue-label: "stale"
+          stale-issue-message: "This issue is stale because it has been open for 90 days with no activity."
+          close-issue-message: "This issue was closed because it has been inactive for 14 days since being marked as stale."
+          days-before-pr-stale: -1
+          days-before-pr-close: -1
+          repo-token: ${{ secrets.GITHUB_TOKEN }}
+    
diff --git a/.github/workflows/codestyle.yml b/.github/workflows/codestyle.yml
new file mode 100644
index 0000000..4e074bd
--- /dev/null
+++ b/.github/workflows/codestyle.yml
@@ -0,0 +1,41 @@
+name: PaddleOCR Code Style Check
+
+# NOTE: Job name `check-code-style` is the required status check context
+# configured in branch protection. Do not rename without updating settings.
+
+on:
+  pull_request: {}
+  push:
+    branches: ['main', 'release/*']
+
+jobs:
+  check-code-style:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+        with:
+          ref: ${{ github.ref }}
+          fetch-depth: 0
+      - id: detect
+        uses: ./.github/actions/detect-docs-only
+      - uses: actions/setup-python@v6
+        if: steps.detect.outputs.docs_only != 'true'
+        with:
+          python-version: '3.10'
+      - name: Cache Python dependencies
+        if: steps.detect.outputs.docs_only != 'true'
+        uses: actions/cache@v5
+        with:
+          path: ~/.cache/pip
+          key: ${{ runner.os }}-pip-${{ hashFiles('**/requirements.txt') }}
+          restore-keys: |
+            ${{ runner.os }}-pip-
+      - name: Install Dependencies for Python
+        if: steps.detect.outputs.docs_only != 'true'
+        run: |
+          python -m pip install --upgrade pip
+          pip install "clang-format==13.0.0"
+      - uses: pre-commit/action@2c7b3805fd2a0fd8c1884dcaebf91fc102a13ecd # v3.0.1
+        if: steps.detect.outputs.docs_only != 'true'
+        with:
+          extra_args: '--all-files'
diff --git a/.github/workflows/docs_anchor_check.yml b/.github/workflows/docs_anchor_check.yml
new file mode 100644
index 0000000..dc341db
--- /dev/null
+++ b/.github/workflows/docs_anchor_check.yml
@@ -0,0 +1,39 @@
+name: Docs Anchor Link Check
+
+on:
+  pull_request:
+    paths:
+      - 'docs/**'
+      - 'mkdocs.yml'
+      - 'mkdocs-ci.yml'
+      - 'overrides/**'
+
+jobs:
+  check-anchor-links:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+        with:
+          fetch-depth: 1
+
+      - uses: actions/setup-python@v6
+        with:
+          python-version: '3.x'
+
+      - uses: actions/cache@v5
+        with:
+          path: ~/.cache/pip
+          key: ${{ runner.os }}-pip-mkdocs-${{ hashFiles('mkdocs.yml') }}
+          restore-keys: |
+            ${{ runner.os }}-pip-mkdocs-
+
+      - name: Install dependencies
+        run: pip install mike mkdocs-material jieba mkdocs-git-revision-date-localized-plugin mkdocs-git-committers-plugin-2 mkdocs-static-i18n markdown-callouts
+
+      - name: Resolve docs GitHub source refs
+        run: python tools/resolve_doc_github_refs.py --placeholder '{{PADDLEOCR_GITHUB_REF}}' --source-ref main
+
+      - name: Check for broken anchor links
+        env:
+          ENABLE_GIT_PLUGINS: 'false'
+        run: mkdocs build -f mkdocs-ci.yml
diff --git a/.github/workflows/link_check.yml b/.github/workflows/link_check.yml
new file mode 100644
index 0000000..6b6ab34
--- /dev/null
+++ b/.github/workflows/link_check.yml
@@ -0,0 +1,42 @@
+name: Link Checker
+
+on:
+  repository_dispatch:
+  workflow_dispatch:
+  # push:
+  #   branches:
+  #     - main
+  # schedule:
+  #   - cron: "00 18 * * 6"
+
+jobs:
+  linkChecker:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+
+      - uses: actions/setup-python@v6
+        with:
+          python-version: '3.x'
+
+      - name: Resolve docs GitHub source refs
+        run: python tools/resolve_doc_github_refs.py --placeholder '{{PADDLEOCR_GITHUB_REF}}' --source-ref main
+
+      - name: Link Checker
+        id: lychee
+        uses: lycheeverse/lychee-action@8646ba30535128ac92d33dfc9133794bfdd9b411 # v2
+        env:
+          GITHUB_TOKEN: ${{secrets.GITHUB_TOKEN}}
+        with:
+            args: --exclude 'docs/index/*.md' 'docs/update/*.md' --verbose --no-progress --max-redirects 8 'docs/**/*.md'
+            format: markdown
+            fail: false
+            output: lychee/results.md
+
+      - name: Create Issue From File
+        if: steps.lychee.outputs.exit_code != 0
+        uses: peter-evans/create-issue-from-file@fca9117c27cdc29c6c4db3b86c48e4115a786710 # v6
+        with:
+          title: Link Checker Report
+          content-filepath: ./lychee/results.md
+          labels: report, automated issue
diff --git a/.github/workflows/python_publish.yml b/.github/workflows/python_publish.yml
new file mode 100644
index 0000000..67b4ca9
--- /dev/null
+++ b/.github/workflows/python_publish.yml
@@ -0,0 +1,39 @@
+# This workflow will upload a Python Package using Twine when a release is created
+# For more information see: https://docs.github.com/en/actions/automating-builds-and-tests/building-and-testing-python#publishing-to-package-registries
+
+# This workflow uses actions that are not certified by GitHub.
+# They are provided by a third-party and are governed by
+# separate terms of service, privacy policy, and support
+# documentation.
+
+name: Upload Python Package
+
+on:
+  release:
+    types: [published]
+
+permissions:
+  contents: read
+
+jobs:
+  deploy:
+
+    runs-on: ubuntu-latest
+
+    steps:
+    - uses: actions/checkout@v6
+    - name: Set up Python
+      uses: actions/setup-python@v6
+      with:
+        python-version: '3.x'
+    - name: Install dependencies
+      run: |
+        python -m pip install --upgrade pip
+        pip install build==1.2.2
+    - name: Build package
+      run: python -m build
+    - name: Publish package
+      uses: pypa/gh-action-pypi-publish@ed0c53931b1dc9bd32cbe73a98c7f6766f8a527e  # v1.13.0
+      with:
+        user: __token__
+        password: ${{ secrets.PYPI_API_TOKEN }}
diff --git a/.github/workflows/test_gpu.yml b/.github/workflows/test_gpu.yml
new file mode 100644
index 0000000..e4c4654
--- /dev/null
+++ b/.github/workflows/test_gpu.yml
@@ -0,0 +1,117 @@
+name: PaddleOCR PR Tests GPU
+
+on:
+  push:
+    branches: ["main"]
+  pull_request:
+    branches: ["main"]
+  workflow_dispatch:
+env:
+  PR_ID: ${{ github.event.pull_request.number }}
+  COMMIT_ID: ${{ github.event.pull_request.head.sha || github.sha }}
+  work_dir: /workspace/PaddleOCR
+  PADDLENLP_ROOT: /workspace/PaddleOCR
+  TASK: paddleocr-CI-${{ github.event.pull_request.number }}
+  BRANCH: ${{ github.event.pull_request.base.ref }}
+  AGILE_COMPILE_BRANCH: ${{ github.event.pull_request.base.ref }}
+  DIR_NAME: ${{ github.repository }}
+permissions:
+  contents: read
+
+jobs:
+  detect-changes:
+    runs-on: ubuntu-latest
+    outputs:
+      docs_only: ${{ steps.detect.outputs.docs_only }}
+    steps:
+      - uses: actions/checkout@v6
+        with:
+          fetch-depth: 0
+      - id: detect
+        uses: ./.github/actions/detect-docs-only
+
+  test-pr-gpu-impl:
+    runs-on: [self-hosted, GPU-2Card-OCR]
+    needs: detect-changes
+    if: needs.detect-changes.outputs.docs_only != 'true'
+    steps:
+      - name: run test
+        env:
+          py_version: "3.10"
+          paddle_whl: https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-GpuSome-LinuxCentos-Gcc82-Cuda118-Cudnn86-Trt85-Py310-CINN-Compile/latest/paddlepaddle_gpu-0.0.0-cp310-cp310-linux_x86_64.whl
+          docker_image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:latest-dev-cuda11.8-cudnn8.6-trt8.5-gcc82
+        run: |
+          work_dir=$RANDOM
+          mkdir $work_dir
+          cd $work_dir
+          git clone --depth=1 https://github.com/PaddlePaddle/PaddleOCR.git -b main
+          cd PaddleOCR
+          if [ -n "${PR_ID}" ]; then
+            git fetch origin pull/${PR_ID}/head:ci_build
+            git checkout ci_build
+          else
+            git fetch --depth=1 origin "${COMMIT_ID}"
+            git checkout "${COMMIT_ID}"
+          fi
+          docker run --gpus all --rm -i --name PaddleOCR_CI_$RANDOM \
+            --shm-size=128g --net=host \
+            -v $PWD:/workspace -w /workspace  \
+            -e "py_version=${py_version}" \
+            -e "paddle_whl=${paddle_whl}" \
+            ${docker_image} /bin/bash -c '
+            ldconfig;
+            nvidia-smi
+            df -hl
+
+            echo ${py_version}
+            rm -rf run_env
+            mkdir run_env
+            ln -s $(which python${py_version}) run_env/python
+            ln -s $(which python${py_version}) run_env/python3
+            ln -s $(which pip${py_version}) run_env/pip
+            export PATH=$PWD/run_env:${PATH}
+
+            git config --global --add safe.directory /workspace
+
+            python -m pip install paddlepaddle-gpu==3.1.0 -i https://www.paddlepaddle.org.cn/packages/stable/cu118/
+            python -c "import paddle; paddle.version.show()"
+            python -m pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple
+            python -m pip install pytest
+            if [ -f requirements.txt ]; then python -m pip install -r requirements.txt; fi
+            PADDLEX_SERIES=$(grep -oE "paddlex\[[^]]+\]>=[0-9]+\.[0-9]+" pyproject.toml | head -1 | grep -oE "[0-9]+\.[0-9]+")
+            if [ -z "$PADDLEX_SERIES" ]; then
+              echo "Failed to determine PaddleX version requirement from pyproject.toml" >&2
+              exit 1
+            fi
+            PADDLEX_BRANCH="release/${PADDLEX_SERIES}"
+            echo "Installing PaddleX from branch: ${PADDLEX_BRANCH}"
+            python -m pip install -e ".[all]" "paddlex@git+https://github.com/PaddlePaddle/PaddleX.git@${PADDLEX_BRANCH}"
+            python -c "import paddlex; print(f\"Installed paddlex version: {paddlex.__version__}\")"
+            python -m pytest --verbose tests/
+            '
+
+  # Aggregator: produces a single check named `test-pr-gpu` so branch
+  # protection (which requires the literal context `test-pr-gpu`) is satisfied
+  # even when the heavy GPU job is skipped (e.g. docs-only changes). Without
+  # this, a SKIPPED conclusion on the GPU job leaves the required check
+  # unsatisfied and blocks merge.
+  test-pr-gpu:
+    runs-on: ubuntu-latest
+    needs: [detect-changes, test-pr-gpu-impl]
+    if: always()
+    steps:
+      - name: Verify required jobs
+        run: |
+          if [ "${{ needs.detect-changes.result }}" != "success" ]; then
+            echo "detect-changes did not succeed: ${{ needs.detect-changes.result }}"
+            exit 1
+          fi
+          if [ "${{ needs.detect-changes.outputs.docs_only }}" = "true" ]; then
+            echo "Docs-only change; treating GPU tests as not required."
+            exit 0
+          fi
+          if [ "${{ needs.test-pr-gpu-impl.result }}" != "success" ]; then
+            echo "test-pr-gpu-impl concluded with: ${{ needs.test-pr-gpu-impl.result }}"
+            exit 1
+          fi
+          echo "GPU tests passed."
diff --git a/.github/workflows/tests.yml b/.github/workflows/tests.yml
new file mode 100644
index 0000000..4e5c561
--- /dev/null
+++ b/.github/workflows/tests.yml
@@ -0,0 +1,108 @@
+name: PaddleOCR PR Tests
+
+on:
+  push:
+    branches: ["main", "release/*"]
+  pull_request:
+    branches: ["main", "release/*"]
+
+permissions:
+  contents: read
+
+jobs:
+  detect-changes:
+    runs-on: ubuntu-latest
+    outputs:
+      docs_only: ${{ steps.detect.outputs.docs_only }}
+    steps:
+      - uses: actions/checkout@v6
+        with:
+          fetch-depth: 0
+      - id: detect
+        uses: ./.github/actions/detect-docs-only
+
+  test-pr-python:
+    runs-on: ubuntu-latest
+    needs: detect-changes
+    if: needs.detect-changes.outputs.docs_only != 'true'
+    strategy:
+      matrix:
+        python-version: ["3.8", "3.9", "3.13"]
+
+    steps:
+    - uses: actions/checkout@v6
+    - name: Set up Python ${{ matrix.python-version }}
+      uses: actions/setup-python@v6
+      with:
+        python-version: ${{ matrix.python-version }}
+
+    - name: Cache dependencies
+      uses: actions/cache@v5
+      with:
+        path: |
+          ~/.cache/pip
+          ~/.local/lib/python${{ matrix.python-version }}/site-packages
+          ~/.paddleocr/
+        key: ${{ runner.os }}-dependencies-${{ matrix.python-version }}-${{ hashFiles('**/requirements.txt', 'pyproject.toml') }}
+        restore-keys: |
+          ${{ runner.os }}-dependencies-
+
+    - name: Install dependencies
+      run: |
+        python -m pip install --upgrade pip
+        pip install pytest
+        if [ -f requirements.txt ]; then pip install -r requirements.txt; fi
+        # On py3.8 several paddlex transitive deps require py3.9+, so only
+        # paddleocr[doc2md] is installable. See installation.md for the
+        # supported extras-by-Python-version matrix.
+        if [[ "${{ matrix.python-version }}" == "3.8" ]]; then
+          python -m pip install paddlepaddle==3.0.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+          PADDLEOCR_EXTRAS="[doc2md]"
+        else
+          python -m pip install paddlepaddle==3.1.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+          PADDLEOCR_EXTRAS="[all]"
+        fi
+        PADDLEX_SERIES=$(grep -oE 'paddlex\[[^]]+\]>=[0-9]+\.[0-9]+' pyproject.toml | head -1 | grep -oE '[0-9]+\.[0-9]+')
+        if [ -z "$PADDLEX_SERIES" ]; then
+          echo "Failed to determine PaddleX version requirement from pyproject.toml" >&2
+          exit 1
+        fi
+
+        python -m pip install "paddlex>=3.7.0,<3.8.0"
+        PADDLEX_BRANCH="release/${PADDLEX_SERIES}"
+        echo "Installing PaddleX from branch: ${PADDLEX_BRANCH}"
+        python -m pip install -e ".${PADDLEOCR_EXTRAS}" "paddlex@git+https://github.com/PaddlePaddle/PaddleX.git@${PADDLEX_BRANCH}"
+        python -c "import paddlex; print(f'Installed paddlex version: {paddlex.__version__}')"
+    - name: Test with pytest
+      run: |
+        # Skip py38_incompatible tests on py3.8.
+        if [[ "${{ matrix.python-version }}" == "3.8" ]]; then
+          pytest --verbose tests/ -m "not resource_intensive and not py38_incompatible"
+        else
+          pytest --verbose tests/
+        fi
+
+  # Aggregator: produces a single check named `test-pr` so branch protection
+  # (which requires the literal context `test-pr`) is satisfied. Without this,
+  # the matrix above only emits `test-pr (3.x)` checks and the required
+  # `test-pr` context never reports.
+  test-pr:
+    runs-on: ubuntu-latest
+    needs: [detect-changes, test-pr-python]
+    if: always()
+    steps:
+      - name: Verify required jobs
+        run: |
+          if [ "${{ needs.detect-changes.result }}" != "success" ]; then
+            echo "detect-changes did not succeed: ${{ needs.detect-changes.result }}"
+            exit 1
+          fi
+          if [ "${{ needs.detect-changes.outputs.docs_only }}" = "true" ]; then
+            echo "Docs-only change; skipping python tests."
+            exit 0
+          fi
+          if [ "${{ needs.test-pr-python.result }}" != "success" ]; then
+            echo "test-pr-python concluded with: ${{ needs.test-pr-python.result }}"
+            exit 1
+          fi
+          echo "All python matrix variants passed."
diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..7930be0
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,40 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+.ipynb_checkpoints/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+/inference/
+/inference_results/
+/output/
+/train_data/
+/log/
+
+*.DS_Store
+*.vs
+*.user
+*~
+*.vscode
+*.idea
+
+*.log
+.clang-format
+.clang_format.hook
+
+build/
+dist/
+*.egg-info/
+
+/deploy/android_demo/app/OpenCV/
+/deploy/android_demo/app/PaddleLite/
+/deploy/android_demo/app/.cxx/
+/deploy/android_demo/app/cache/
+
+test_tipc/web/models/
+test_tipc/web/node_modules/
+
+.venv/
+.worktrees/
diff --git a/.lycheeignore b/.lycheeignore
new file mode 100644
index 0000000..a0761f9
--- /dev/null
+++ b/.lycheeignore
@@ -0,0 +1,5 @@
+zhuanlan.zhihu.com/*
+https://demo.doctrp.top/
+http://127.0.0.1:8001/
+http://localhost:9003
+https://rrc.cvc.uab.es/
diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
new file mode 100644
index 0000000..c2c52b4
--- /dev/null
+++ b/.pre-commit-config.yaml
@@ -0,0 +1,46 @@
+exclude: ^(langchain-paddleocr/|paddleocr-js/)
+repos:
+-   repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v5.0.0
+    hooks:
+    -   id: check-added-large-files
+        args: ['--maxkb=512']
+    -   id: check-case-conflict
+    -   id: check-merge-conflict
+    -   id: check-symlinks
+    -   id: detect-private-key
+    -   id: end-of-file-fixer
+    -   id: trailing-whitespace
+        files: \.(c|cc|cxx|cpp|cu|h|hpp|hxx|py)$
+-   repo: https://github.com/Lucas-C/pre-commit-hooks
+    rev: v1.5.5
+    hooks:
+    -   id: remove-crlf
+    -   id: remove-tabs
+        files: \.(c|cc|cxx|cpp|cu|h|hpp|hxx|py)$
+-   repo: local
+    hooks:
+    -   id: clang-format
+        name: clang-format
+        description: Format files with ClangFormat
+        entry: bash .clang_format.hook -i
+        language: system
+        files: \.(c|cc|cxx|cpp|cu|h|hpp|hxx|cuh|proto)$
+# For Python files
+-   repo: https://github.com/psf/black.git
+    rev: 24.10.0
+    hooks:
+    -   id: black
+        files: (.*\.(py|pyi|bzl)|BUILD|.*\.BUILD|WORKSPACE)$
+
+# Flake8
+-   repo: https://github.com/pycqa/flake8
+    rev: 7.1.1
+    hooks:
+    -   id: flake8
+        args:
+            - --count
+            - --select=E9,F63,F7,F82,E721
+            - --show-source
+            - --statistics
+        exclude: ^benchmark/|^test_tipc/
diff --git a/.style.yapf b/.style.yapf
new file mode 100644
index 0000000..4741fb4
--- /dev/null
+++ b/.style.yapf
@@ -0,0 +1,3 @@
+[style]
+based_on_style = pep8
+column_limit = 80
diff --git a/CNAME b/CNAME
new file mode 100644
index 0000000..ac7eea8
--- /dev/null
+++ b/CNAME
@@ -0,0 +1 @@
+www.paddleocr.ai
diff --git a/LICENSE b/LICENSE
new file mode 100644
index 0000000..430edfa
--- /dev/null
+++ b/LICENSE
@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
diff --git a/MANIFEST.in b/MANIFEST.in
new file mode 100644
index 0000000..5e6f4f2
--- /dev/null
+++ b/MANIFEST.in
@@ -0,0 +1,22 @@
+prune .github
+prune applications
+prune benchmark
+prune configs
+prune deploy
+prune doc
+prune docs
+prune overrides
+prune ppocr/ext_op
+prune ppocr/losses
+prune ppocr/metrics
+prune ppocr/modeling
+prune ppocr/optimizer
+prune ppstructure/docs
+prune test_tipc
+prune tests
+exclude .clang_format.hook
+exclude .gitignore
+exclude .pre-commit-config.yaml
+exclude .style.yapf
+exclude mkdocs.yml
+exclude train.sh
diff --git a/README.md b/README.md
new file mode 100644
index 0000000..6458242
--- /dev/null
+++ b/README.md
@@ -0,0 +1,316 @@
+
+<div align="center">
+  <p>
+      <img width="800" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Banner.png" alt="Star-history">
+  </p>
+
+
+
+<h3>Global Leading OCR Toolkit & Document AI Engine</h3>
+
+English | [简体中文](./readme/README_cn.md) | [繁體中文](./readme/README_tcn.md) | [日本語](./readme/README_ja.md) | [한국어](./readme/README_ko.md) | [Français](./readme/README_fr.md) | [Русский](./readme/README_ru.md) | [Español](./readme/README_es.md) | [العربية](./readme/README_ar.md)
+
+<!-- icon -->
+
+[![PyPI Downloads](https://static.pepy.tech/badge/paddleocr)](https://pepy.tech/projects/paddleocr)
+[![Used by](https://img.shields.io/badge/Used%20by-6k%2B%20repositories-blue)](https://github.com/PaddlePaddle/PaddleOCR/network/dependents)
+![python](https://img.shields.io/badge/python-3.8~3.12-aff.svg)
+![os](https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-pink.svg)
+![hardware](https://img.shields.io/badge/hardware-cpu%2C%20gpu%2C%20xpu%2C%20npu-yellow.svg)
+
+[![AI Studio](https://img.shields.io/badge/PaddleOCR-_Offiical_Website-1927BA?logo=data:image/png;base64,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&labelColor=white)](https://www.paddleocr.com)
+[![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/PaddlePaddle/PaddleOCR)
+[![License](https://img.shields.io/badge/license-Apache_2.0-green)](../LICENSE)
+
+</div>
+
+
+
+
+
+
+**PaddleOCR converts PDF documents and images into structured, LLM-ready data (JSON/Markdown) with industry-leading accuracy. With 70k+ Stars and trusted by top-tier projects like Dify, RAGFlow, and Cherry Studio, PaddleOCR is the bedrock for building intelligent RAG and Agentic applications.**
+
+
+## 🚀 Key Features
+
+### 📄 Intelligent Document Parsing (LLM-Ready)
+> *Transforming messy visuals into structured data for the LLM era.*
+
+* **SOTA Document VLM**: Featuring **PaddleOCR-VL-1.6 (0.9B)**, the industry's leading lightweight vision-language model for document parsing. It achieves 96.3% accuracy on OmniDocBench v1.6, leads in text, formula, and table recognition, and shows significantly enhanced capabilities in ancient documents, rare characters, seals, and charts, with structured outputs in **Markdown** and **JSON** formats.
+* **Structure-Aware Conversion**: Powered by **PP-StructureV3**, seamlessly convert complex PDFs and images into **Markdown** or **JSON**. Unlike the PaddleOCR-VL series models, it provides more fine-grained coordinate information, including table cell coordinates, text coordinates, and more.
+* **Production-Ready Efficiency**: Achieve commercial-grade accuracy with an ultra-small footprint. Outperforms numerous closed-source solutions in public benchmarks while remaining resource-efficient for edge/cloud deployment.
+
+### 🔍 Universal Text Recognition (Scene OCR)
+> *The global gold standard for high-speed, multilingual text spotting.*
+
+* **100+ Languages Supported**: Native recognition for a vast global library. **PP-OCRv6** supports 50 languages with a single unified model (Chinese, English, Japanese, and 46 Latin-script languages) — no model switching needed for multilingual documents.
+* **Complex Element Mastery**: Beyond standard text recognition, we support **natural scene text spotting** across a wide range of environments, including IDs, street views, books, and industrial components
+* **Performance Leap**: PP-OCRv6 achieves **+4.6% detection** and **+5.1% recognition** accuracy over PP-OCRv5, surpassing mainstream Vision-Language Models. 5.2× CPU inference speedup end-to-end.
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Arch.jpg" alt="PaddleOCR Architecture">
+  </p>
+</div>
+
+### 🛠️ Developer-Centric Ecosystem
+* **Seamless Integration**: The premier choice for the AI Agent ecosystem—deeply integrated with **Dify, RAGFlow, Pathway, and Cherry Studio**.
+* **LLM Data Flywheel**: A complete pipeline to build high-quality datasets, providing a sustainable "Data Engine" for fine-tuning Large Language Models.
+* **One-Click Deployment**: Supports various hardware backends (NVIDIA GPU, Intel CPU, Kunlunxin XPU, and diverse AI Accelerators).
+
+
+## 📣 Recent updates
+
+### 🔥 2026.06.11: Release of PaddleOCR 3.7.0
+- PP-OCRv6 highlights:
+
+    - **Accuracy boost**: Medium tier achieves +4.6% detection and +5.1% recognition over PP-OCRv5_server, surpassing mainstream VLMs (Qwen3-VL-235B, GPT-5.5) with only 34.5M parameters.
+    - **50 languages unified**: Single model covers Chinese, English, Japanese, and 46 Latin-script languages — no model switching needed.
+    - **Specialized scenarios**: Major improvements in digital displays, dot-matrix characters, tire prints, and industrial text recognition.
+    - **Faster inference**: 5.2× CPU speedup (OpenVINO), 6.1× on Apple M4 (tiny), 0.13s on A100 GPU.
+    - **Three tiers for all scenarios**: tiny (1.5M) / small (7.7M) / medium (34.5M) for edge, mobile, and server deployment.
+    - **Model availability**: All models are available on [HuggingFace](https://huggingface.co/collections/PaddlePaddle/pp-ocrv6) and [ModelScope](https://www.modelscope.cn/collections/PaddlePaddle/PP-OCRv6).
+
+<details>
+<summary><strong>2026.05.28: Release of PaddleOCR 3.6.0</strong></summary>
+
+- PaddleOCR-VL-1.6 highlights:
+
+    - **New SOTA Accuracy**: Achieves over 96.3% on OmniDocBench v1.6, also sets new SOTA on OmniDocBench v1.5 and Real5-OmniDocBench, leading both open-source and proprietary solutions in text, formula, and table recognition.
+    - **Comprehensive Capability Upgrade**: Significant improvements in table, ancient document, and rare character recognition, with notably enhanced seal recognition, spotting, and chart understanding across multiple scenarios.
+    - **Seamless Migration**: Model architecture is fully consistent with PaddleOCR-VL-1.5, enabling zero-cost adaptation—swap and go.
+    - **Try it now**: Available on [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.6) or our [Official Website](https://www.paddleocr.com).
+
+</details>
+
+<details>
+<summary><strong>2026.04.21: Release of PaddleOCR 3.5.0</strong></summary>
+
+* **Flexible inference backends**: Seamlessly switch between Paddle static graph, Paddle dynamic graph, or Transformers. PaddleOCR is now deeply integrated with the Hugging Face ecosystem, and 20 major models support Transformers as the inference backend.
+* **Office documents to Markdown**: Convert common document formats such as Word, Excel, and PowerPoint into Markdown.
+* **DOCX export for parsed results**: The `PaddleOCR-VL` series, `PP-StructureV3`, and `PP-DocTranslation` now support exporting parsed results to DOCX for convenient viewing and editing in Microsoft Word.
+* **Official browser inference SDK**: Released `PaddleOCR.js`, the official browser inference SDK that supports running `PP-OCRv5` directly in the browser.
+
+</details>
+
+<details>
+<summary><strong>2026.01.29: Release of PaddleOCR 3.4.0</strong></summary>
+
+* PaddleOCR-VL-1.5 (SOTA 0.9B VLM): Our latest flagship model for document parsing is now live!
+    * **94.5% Accuracy on OmniDocBench**: Surpassing top-tier general large models and specialized document parsers.
+    * **Real-World Robustness**: First to introduce the **PP-DocLayoutV3** algorithm for irregular shape positioning, mastering 5 tough scenarios: *Skew, Warping, Scanning, Illumination, and Screen Photography*.
+    * **Capability Expansion**: Now supports **Seal Recognition**, **Text Spotting**, and expands to **111 languages** (including China’s Tibetan script and Bengali).
+    * **Long Document Mastery**: Supports automatic cross-page table merging and hierarchical heading identification.
+    * **Try it now**: Available on [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.5) or our [Official Website](https://www.paddleocr.com).
+
+</details>
+
+<details>
+<summary><strong>2025.10.16: Release of PaddleOCR 3.3.0</strong></summary>
+
+- Released PaddleOCR-VL:
+    - **Model Introduction**:
+        - **PaddleOCR-VL** is a SOTA and resource-efficient model tailored for document parsing. Its core component is PaddleOCR-VL-0.9B, a compact yet powerful vision-language model (VLM) that integrates a NaViT-style dynamic resolution visual encoder with the ERNIE-4.5-0.3B language model to enable accurate element recognition. **This innovative model efficiently supports 109 languages and excels in recognizing complex elements (e.g., text, tables, formulas, and charts), while maintaining minimal resource consumption**. Through comprehensive evaluations on widely used public benchmarks and in-house benchmarks, PaddleOCR-VL achieves SOTA performance in both page-level document parsing and element-level recognition. It significantly outperforms existing solutions, exhibits strong competitiveness against top-tier VLMs, and delivers fast inference speeds. These strengths make it highly suitable for practical deployment in real-world scenarios. The model has been released on [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL). Everyone is welcome to download and use it! More introduction information can be found in [PaddleOCR-VL](https://www.paddleocr.ai/latest/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL.html).
+
+    - **Core Features**:
+        - **Compact yet Powerful VLM Architecture**: We present a novel vision-language model that is specifically designed for resource-efficient inference, achieving outstanding performance in element recognition. By integrating a NaViT-style dynamic high-resolution visual encoder with the lightweight ERNIE-4.5-0.3B language model, we significantly enhance the model’s recognition capabilities and decoding efficiency. This integration maintains high accuracy while reducing computational demands, making it well-suited for efficient and practical document processing applications.
+        - **SOTA Performance on Document Parsing**: PaddleOCR-VL achieves state-of-the-art performance in both page-level document parsing and element-level recognition. It significantly outperforms existing pipeline-based solutions and exhibiting strong competitiveness against leading vision-language models (VLMs) in document parsing. Moreover, it excels in recognizing complex document elements, such as text, tables, formulas, and charts, making it suitable for a wide range of challenging content types, including handwritten text and historical documents. This makes it highly versatile and suitable for a wide range of document types and scenarios.
+        - **Multilingual Support**: PaddleOCR-VL Supports 109 languages, covering major global languages, including but not limited to Chinese, English, Japanese, Latin, and Korean, as well as languages with different scripts and structures, such as Russian (Cyrillic script), Arabic, Hindi (Devanagari script), and Thai. This broad language coverage substantially enhances the applicability of our system to multilingual and globalized document processing scenarios.
+
+- Released PP-OCRv5 Multilingual Recognition Model:
+    - Improved the accuracy and coverage of Latin script recognition; added support for Cyrillic, Arabic, Devanagari, Telugu, Tamil, and other language systems, covering recognition of 109 languages. The model has only 2M parameters, and the accuracy of some models has increased by over 40% compared to the previous generation.
+
+</details>
+
+
+<details>
+<summary><strong>2025.08.21: Release of PaddleOCR 3.2.0</strong></summary>
+
+- **Significant Model Additions:**
+    - Introduced training, inference, and deployment for PP-OCRv5 recognition models in English, Thai, and Greek. **The PP-OCRv5 English model delivers an 11% improvement in English scenarios compared to the main PP-OCRv5 model, with the Thai and Greek recognition models achieving accuracies of 82.68% and 89.28%, respectively.**
+
+- **Deployment Capability Upgrades:**
+    - **Full support for PaddlePaddle framework versions 3.1.0 and 3.1.1.**
+    - **Comprehensive upgrade of the PP-OCRv5 C++ local deployment solution, now supporting both Linux and Windows, with feature parity and identical accuracy to the Python implementation.**
+    - **High-performance inference now supports CUDA 12, and inference can be performed using either the Paddle Inference or ONNX Runtime backends.**
+    - **The high-stability service-oriented deployment solution is now fully open-sourced, allowing users to customize Docker images and SDKs as required.**
+    - The high-stability service-oriented deployment solution also supports invocation via manually constructed HTTP requests, enabling client-side code development in any programming language.
+
+- **Benchmark Support:**
+    - **All production lines now support fine-grained benchmarking, enabling measurement of end-to-end inference time as well as per-layer and per-module latency data to assist with performance analysis. [Here's](docs/version3.x/pipeline_usage/instructions/benchmark.en.md) how to set up and use the benchmark feature.**
+    - **Documentation has been updated to include key metrics for commonly used configurations on mainstream hardware, such as inference latency and memory usage, providing deployment references for users.**
+
+- **Bug Fixes:**
+    - Resolved the issue of failed log saving during model training.
+    - Upgraded the data augmentation component for formula models for compatibility with newer versions of the albumentations dependency, and fixed deadlock warnings when using the tokenizers package in multi-process scenarios.
+    - Fixed inconsistencies in switch behaviors (e.g., `use_chart_parsing`) in the PP-StructureV3 configuration files compared to other pipelines.
+
+- **Other Enhancements:**
+    - **Separated core and optional dependencies. Only minimal core dependencies are required for basic text recognition; additional dependencies for document parsing and information extraction can be installed as needed.**
+    - **Enabled support for NVIDIA RTX 50 series graphics cards on Windows; users can refer to the [installation guide](docs/version3.x/installation.en.md) for the corresponding PaddlePaddle framework versions.**
+    - **PP-OCR series models now support returning single-character coordinates.**
+    - Added AIStudio, ModelScope, and other model download sources, allowing users to specify the source for model downloads.
+    - Added support for chart-to-table conversion via the PP-Chart2Table module.
+    - Optimized documentation descriptions to improve usability.
+</details>
+
+
+[History Log](https://paddlepaddle.github.io/PaddleOCR/latest/en/update/update.html)
+
+
+## 🚀 Quick Start
+
+### Step 1: Try Online
+PaddleOCR official website provides interactive **Experience Center** and **APIs**—no setup required, just one click to experience.
+
+👉 [Visit Official Website](https://www.paddleocr.com)
+
+### Step 2: Local Deployment
+For local usage, please refer to the following documentation based on your needs:
+
+- **PP-OCR Series**: See [PP-OCR Documentation](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/OCR.html)
+- **PaddleOCR-VL Series**: See [PaddleOCR-VL Documentation](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PaddleOCR-VL.html)
+- **PP-StructureV3**: See [PP-StructureV3 Documentation](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PP-StructureV3.html)
+- **More Capabilities**: See [More Capabilities Documentation](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/pipeline_overview.html)
+
+
+## 🧩 More Features
+
+- Convert models to ONNX format: [Obtaining ONNX Models](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/others/obtaining_onnx_models.html).
+- Accelerate inference using engines like OpenVINO, ONNX Runtime, TensorRT, or perform inference using ONNX format models: [High-Performance Inference](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/local_inference/high_performance_inference.html).
+- Accelerate inference using multi-GPU and multi-process: [Parallel Inference for Pipelines](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/pipeline_usage/instructions/parallel_inference.html).
+- Integrate PaddleOCR into applications written in C++, C#, Java, etc.: [Serving](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/serving/serving.html).
+
+## 🔄 Quick Overview of Execution Results
+
+### PP-OCRv5
+
+<div align="center">
+  <p>
+       <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-OCRv5_demo.gif" alt="PP-OCRv5 Demo">
+  </p>
+</div>
+
+
+
+### PP-StructureV3
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-StructureV3_demo.gif" alt="PP-StructureV3 Demo">
+  </p>
+</div>
+
+### PaddleOCR-VL
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PaddleOCR-VL_demo.gif" alt="PP-StructureV3 Demo">
+  </p>
+</div>
+
+
+## ✨ Stay Tuned
+
+⭐ **Star this repository to keep up with exciting updates and new releases, including powerful OCR and document parsing capabilities!** ⭐
+
+<div align="center">
+  <p>
+       <img width="1200" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/star_paddleocr2.en.gif" alt="Star-Project">
+  </p>
+</div>
+
+
+## 👩‍👩‍👧‍👦 Community
+
+<div align="center">
+
+| PaddlePaddle WeChat official account |  Join the tech discussion group |
+| :---: | :---: |
+| <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qrcode_for_paddlepaddle_official_account.jpg" width="150"> | <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qr_code_for_the_questionnaire.jpg" width="150"> |
+</div>
+
+
+## 😃 Awesome Projects Leveraging PaddleOCR
+PaddleOCR wouldn't be where it is today without its incredible community! 💗 A massive thank you to all our longtime partners, new collaborators, and everyone who's poured their passion into PaddleOCR — whether we've named you or not. Your support fuels our fire!
+
+<div align="center">
+
+| Project Name | Description |
+| ------------ | ----------- |
+| [Dify](https://github.com/langgenius/dify) <a href="https://github.com/langgenius/dify"><img src="https://img.shields.io/github/stars/langgenius/dify"></a>|Production-ready platform for agentic workflow development.|
+| [RAGFlow](https://github.com/infiniflow/ragflow) <a href="https://github.com/infiniflow/ragflow"><img src="https://img.shields.io/github/stars/infiniflow/ragflow"></a>|RAG engine based on deep document understanding.|
+| [pathway](https://github.com/pathwaycom/pathway) <a href="https://github.com/pathwaycom/pathway"><img src="https://img.shields.io/github/stars/pathwaycom/pathway"></a>|Python ETL framework for stream processing, real-time analytics, LLM pipelines, and RAG.|
+| [MinerU](https://github.com/opendatalab/MinerU) <a href="https://github.com/opendatalab/MinerU"><img src="https://img.shields.io/github/stars/opendatalab/MinerU"></a>|Multi-type Document to Markdown Conversion Tool|
+| [Umi-OCR](https://github.com/hiroi-sora/Umi-OCR) <a href="https://github.com/hiroi-sora/Umi-OCR"><img src="https://img.shields.io/github/stars/hiroi-sora/Umi-OCR"></a>|Free, Open-source, Batch Offline OCR Software.|
+| [cherry-studio](https://github.com/CherryHQ/cherry-studio) <a href="https://github.com/CherryHQ/cherry-studio"><img src="https://img.shields.io/github/stars/CherryHQ/cherry-studio"></a>|A desktop client that supports for multiple LLM providers.|
+| [haystack](https://github.com/deepset-ai/haystack)<a href="https://github.com/deepset-ai/haystack"><img src="https://img.shields.io/github/stars/deepset-ai/haystack"></a> |AI orchestration framework to build customizable, production-ready LLM applications.|
+| [OmniParser](https://github.com/microsoft/OmniParser)<a href="https://github.com/microsoft/OmniParser"><img src="https://img.shields.io/github/stars/microsoft/OmniParser"></a> |OmniParser: Screen Parsing tool for Pure Vision Based GUI Agent.|
+| [QAnything](https://github.com/netease-youdao/QAnything)<a href="https://github.com/netease-youdao/QAnything"><img src="https://img.shields.io/github/stars/netease-youdao/QAnything"></a> |Question and Answer based on Anything.|
+| [Learn more projects](./awesome_projects.md) | [More projects based on PaddleOCR](./awesome_projects.md)|
+</div>
+
+## 👩‍👩‍👧‍👦 Contributors
+
+<div align="center">
+<a href="https://github.com/PaddlePaddle/PaddleOCR/graphs/contributors">
+  <img src="https://contrib.rocks/image?repo=PaddlePaddle/PaddleOCR&max=400&columns=20"  width="800"/>
+</a>
+</div>
+
+## 🌟 Star
+
+<div align="center">
+  <p>
+      <img width="800" src="https://api.star-history.com/svg?repos=PaddlePaddle/PaddleOCR&type=Date" alt="Star-history">
+  </p>
+</div>
+
+
+## 📄 License
+This project is released under the [Apache 2.0 license](LICENSE).
+
+## 🎓 Citation
+
+```bibtex
+@misc{cui2025paddleocr30technicalreport,
+      title={PaddleOCR 3.0 Technical Report}, 
+      author={Cheng Cui and Ting Sun and Manhui Lin and Tingquan Gao and Yubo Zhang and Jiaxuan Liu and Xueqing Wang and Zelun Zhang and Changda Zhou and Hongen Liu and Yue Zhang and Wenyu Lv and Kui Huang and Yichao Zhang and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2507.05595},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2507.05595}, 
+}
+
+@misc{cui2025paddleocrvlboostingmultilingualdocument,
+      title={PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model}, 
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Handong Zheng and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2510.14528},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2510.14528}, 
+}
+
+@misc{cui2026paddleocrvl15multitask09bvlm,
+      title={PaddleOCR-VL-1.5: Towards a Multi-Task 0.9B VLM for Robust In-the-Wild Document Parsing}, 
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2026},
+      eprint={2601.21957},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2601.21957}, 
+}
+
+@misc{zhang2026paddleocrvl16expandingfrontierdocument,
+      title={PaddleOCR-VL-1.6: Expanding the Frontier of Document Parsing with Under-Optimized Region Refinement and Progressive Post-Training}, 
+      author={Zelun Zhang and Hongen Liu and Suyin Liang and Yubo Zhang and Yiqing Xiang and Jiaxuan Liu and Ting Sun and Manhui Lin and Yue Zhang and Changda Zhou and Tingquan Gao and Cheng Cui and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2026},
+      eprint={2606.03264},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2606.03264}, 
+}
+```
diff --git a/README.wehub.md b/README.wehub.md
new file mode 100644
index 0000000..91d80c3
--- /dev/null
+++ b/README.wehub.md
@@ -0,0 +1,7 @@
+# WeHub 来源说明
+
+- 原始项目：`PaddlePaddle/PaddleOCR`
+- 原始仓库：https://github.com/PaddlePaddle/PaddleOCR
+- 导入方式：上游默认分支的最新快照
+- 原作者、版权和许可证信息以原始仓库及本仓库 LICENSE 为准
+- 本文件仅用于记录来源，不代表 WeHub 是原项目作者
diff --git a/RELEASING.md b/RELEASING.md
new file mode 100644
index 0000000..fe260a2
--- /dev/null
+++ b/RELEASING.md
@@ -0,0 +1,194 @@
+# PaddleOCR Release SOP
+
+[中文](./RELEASING_cn.md)
+
+## Scope
+
+This document describes the standard release process for PaddleOCR.
+
+## Release Types
+
+The current process supports the following two release types:
+
+- `bump patch`: for example, `3.4.0 -> 3.4.1`
+- `bump minor`: for example, `3.4.x -> 3.5.0`
+
+For `bump major`:
+
+- The current process does not directly support this scenario
+- A major release process should be discussed and designed separately, for example by introducing an additional branch or a new preparation flow
+
+## Release Principles
+
+- Daily development happens on `main`
+- Official releases are made only from `release/X.Y`
+- Official tags must use `vX.Y.Z`
+- Do not create official release tags on `main`
+
+## Standard Release Process
+
+### 1. Confirm the release target
+
+First confirm which release line this release belongs to and what the target version is.
+
+Examples:
+
+- Release `3.4.1` on `release/3.4`
+- Release `3.5.0` on `release/3.5`
+
+### 2. Create or switch to the release branch
+
+If the release line does not exist yet, create the corresponding `release/X.Y` branch from `main`.  
+If it already exists, switch to that branch and continue release preparation there.
+
+Requirements:
+
+- One minor release line corresponds to one fixed `release/X.Y` branch
+- That branch should contain only the release content and patch fixes for that line
+
+### 3. Pick release content from main
+
+Based on the release scope, `cherry-pick` the required commits from `main` into `release/X.Y` until the release content is ready.
+
+Requirements:
+
+- Only pick what is needed for the current release
+- Avoid bringing unrelated new features into the release branch
+- If there are release-specific fixes on the release branch, keep them limited to the current release scope
+
+### 4. Complete pre-release checks
+
+Complete pre-release checks on `release/X.Y`.
+
+At minimum, this should include:
+
+- The current branch is correct
+- The working tree is clean
+- The version is as expected
+- Key functionality is verified
+- Required tests, builds, packaging, and regression checks have passed
+- Release notes are ready
+
+### 5. Create the official tag
+
+Once the release is ready, create the official tag on `release/X.Y`:
+
+- The tag format must be `vX.Y.Z`
+
+Examples:
+
+- `v3.4.1`
+- `v3.5.0`
+
+Requirements:
+
+- The tag must be created on the release branch for the current official release
+- Do not use development tags as official release tags
+
+### 6. Publish the GitHub Release
+
+Create a GitHub Release based on the official tag for this release.
+
+### 7. Update dependency constraints and release notes
+
+If this release is the first release of a new minor line, or if it changes PaddleX dependency requirements, update the related release materials before or after the official tag is published.
+
+At minimum, this should include:
+
+- Checking whether the `paddlex` dependency constraints in `pyproject.toml` match the target release
+- Checking whether installation docs, upgrade notes, and release notes are aligned with the release version
+
+Completion criteria:
+
+- The `paddlex` dependency constraints match the target release
+- The version information in the documentation matches the released version
+
+### 8. Sync the release branch lineage back to main
+
+After the first official release of a new `release/X.Y` line is completed, sync the lineage of that `release/X.Y` branch back to `main`.
+
+This is a fixed step in the current workflow and should be done at least once for each new minor release line.
+
+Purpose:
+
+- Ensure `main` correctly reflects that the release line has produced an official version
+- Keep subsequent development and release cadence aligned
+
+Requirements:
+
+- Perform this once after the first official release of each new `release/X.Y`
+- For later patch releases on the same `release/X.Y`, it is usually not necessary to repeat it
+
+### 9. Move on to the next development cycle or patch release
+
+After the release:
+
+- `main` continues with ongoing development
+- `release/X.Y` continues to maintain that release line
+
+If more patch releases are needed later on the same line:
+
+- Continue preparing patches on `release/X.Y`
+- `cherry-pick` from `main` as needed
+- Repeat the relevant steps in this SOP
+
+## How to Handle Different Bump Types
+
+### Patch Release
+
+Applicable scenarios:
+
+- Fixing production issues
+- Small compatibility fixes
+- Documentation, dependency, or stability patches
+
+How to handle it:
+
+- Continue preparing changes on the existing `release/X.Y`
+- Create the next patch tag, for example `v3.4.2`
+
+### Minor Release
+
+Applicable scenarios:
+
+- Starting a new release line
+- Releasing the next stable version, for example `3.5.0`
+
+How to handle it:
+
+- Create a new `release/X.Y` from `main`
+- Prepare the release following the standard process
+- Create the first official tag, for example `v3.5.0`
+- Update `paddlex` dependency constraints if needed
+- Sync the lineage of that release branch back to `main` after the release
+
+### Major Release
+
+Current conclusion:
+
+- It is not included in this SOP for now
+- A separate process needs to be discussed and designed later
+
+Before a dedicated solution is defined, do not directly reuse the current minor/patch process for a major release.
+
+## Release Checklist
+
+Before each release, confirm the following:
+
+- The target version and corresponding release branch have been confirmed
+- The content on the release branch is ready
+- The release scope has been finalized
+- Key tests and regression checks have passed
+- Release notes are ready
+- The official tag uses the `vX.Y.Z` format
+- The GitHub Release has been created
+- If this is the first official release on `release/X.Y`, its lineage has been synced back to `main`
+
+## Daily Maintenance Recommendations
+
+- New features should go to `main` first
+- Official releases should always be performed through `release/X.Y`
+- Patch releases should always be maintained on the corresponding `release/X.Y`
+- For each new `release/X.Y`, sync its lineage once after the first official release
+
+If the current process changes, this document should be updated accordingly.
diff --git a/RELEASING_cn.md b/RELEASING_cn.md
new file mode 100644
index 0000000..8310f4b
--- /dev/null
+++ b/RELEASING_cn.md
@@ -0,0 +1,194 @@
+# PaddleOCR Release SOP
+
+[English](./RELEASING.md)
+
+## 适用范围
+
+本文档用于说明 PaddleOCR 的标准发版流程。
+
+## 版本类型
+
+当前流程支持以下两类发布：
+
+- `bump patch`：例如 `3.4.0 -> 3.4.1`
+- `bump minor`：例如 `3.4.x -> 3.5.0`
+
+对于 `bump major`：
+
+- 当前流程暂不支持直接覆盖这类场景
+- 如需发布大版本，需要单独讨论并设计额外流程，例如引入额外分支或新的版本准备方式
+
+## 发版原则
+
+- 日常开发在 `main` 分支进行
+- 正式发布只在 `release/X.Y` 分支进行
+- 正式标签只使用 `vX.Y.Z`
+- 不在 `main` 分支打正式发布标签
+
+## 标准发版流程
+
+### 1. 确认发布目标
+
+先确认本次发布属于哪一条版本线，以及目标版本号。
+
+示例：
+
+- 发布 `3.4.1`，对应分支为 `release/3.4`
+- 发布 `3.5.0`，对应分支为 `release/3.5`
+
+### 2. 创建或切换到 release 分支
+
+如果该版本线尚未建立，则从 `main` 创建对应的 `release/X.Y` 分支。  
+如果该版本线已存在，则直接切换到该分支继续发版准备。
+
+要求：
+
+- 一个 minor 版本线对应一个固定的 `release/X.Y` 分支
+- 该分支只承载该版本线的发布内容和补丁修复
+
+### 3. 从 main 挑选发布内容
+
+根据本次发布范围，从 `main` 将需要发布的提交按需 `cherry-pick` 到 `release/X.Y`，直到版本内容 ready。
+
+要求：
+
+- 只挑选本次发布需要的内容
+- 避免把与本次发布无关的新功能带入 release 分支
+- 如果 release 分支上出现专门的发布修复，也应仅限于本次发布范围
+
+### 4. 完成发布前检查
+
+在 `release/X.Y` 上完成发布前检查。
+
+至少应包括：
+
+- 当前分支正确
+- 工作区干净
+- 版本号符合预期
+- 关键功能验证通过
+- 必要的测试、构建、打包和回归通过
+- 发布说明已经准备好
+
+### 5. 打正式标签
+
+确认版本 ready 后，在 `release/X.Y` 分支上打正式标签：
+
+- 标签格式必须为 `vX.Y.Z`
+
+示例：
+
+- `v3.4.1`
+- `v3.5.0`
+
+要求：
+
+- 标签必须打在本次正式发布对应的 release 分支上
+- 不使用开发态标签作为正式发布标签
+
+### 6. 发布 GitHub Release
+
+在 GitHub 上基于本次正式标签创建 Release。
+
+### 7. 更新依赖约束与发布说明
+
+如果本次发布是新的 minor 版本首发，或本次发布涉及 PaddleX 依赖变化，则在正式标签发布前后同步完成相关更新。
+
+至少应包括：
+
+- 检查 `pyproject.toml` 中 `paddlex` 相关依赖约束是否与本次发布版本匹配
+- 检查安装说明、升级说明和发布说明中的版本信息是否与本次发布一致
+
+完成要求：
+
+- `paddlex` 相关依赖约束与本次发布目标一致
+- 文档中的版本信息与本次发布版本一致
+
+### 8. 将 release 分支的 lineage 同步回 main
+
+当一条新的 `release/X.Y` 发布线完成首个正式版本发布后，需要将该 `release/X.Y` 的 lineage 同步回 `main`。
+
+这一步是当前流程中的固定步骤，每条新的 minor 发布线至少执行一次。
+
+目的：
+
+- 让 `main` 正确感知该发布线已经完成的正式版本
+- 保持后续开发与发布节奏一致
+
+要求：
+
+- 每条新的 `release/X.Y` 在首个正式版本发布后执行一次
+- 同一条 `release/X.Y` 后续如果继续发布补丁版本，通常不要求重复执行
+
+### 9. 进入下一轮开发或补丁发布
+
+发布完成后：
+
+- `main` 继续进行后续开发
+- `release/X.Y` 继续维护该版本线
+
+如果后续还要发布该版本线的补丁版本：
+
+- 继续在 `release/X.Y` 上准备补丁
+- 从 `main` 按需 `cherry-pick`
+- 重复执行本 SOP 中的发布步骤
+
+## 不同 bump 类型的处理方式
+
+### Patch 发布
+
+适用场景：
+
+- 修复线上问题
+- 小范围兼容性修复
+- 文档、依赖或稳定性补丁
+
+处理方式：
+
+- 在现有 `release/X.Y` 分支上继续准备内容
+- 打下一个 patch 标签，例如 `v3.4.2`
+
+### Minor 发布
+
+适用场景：
+
+- 开启一条新的发布线
+- 发布下一阶段稳定版本，例如 `3.5.0`
+
+处理方式：
+
+- 从 `main` 建立新的 `release/X.Y`
+- 按标准流程准备版本内容
+- 打首个正式标签，例如 `v3.5.0`
+- 如有需要，同步更新 `paddlex` 相关依赖约束
+- 发布后将该 release 分支的 lineage 同步回 `main`
+
+### Major 发布
+
+当前结论：
+
+- 暂不纳入本 SOP
+- 后续需要单独讨论并设计流程
+
+在新的方案明确之前，不建议直接套用当前 minor/patch 的做法处理 major 发布。
+
+## 发布检查清单
+
+每次发版前，请确认以下事项：
+
+- 已确认目标版本号和对应 release 分支
+- release 分支中的内容已经 ready
+- 发布范围已经收敛
+- 关键测试和回归已通过
+- 发布说明已准备完成
+- 正式标签格式为 `vX.Y.Z`
+- GitHub Release 已创建
+- 如本次是该 `release/X.Y` 的首个正式版本，发布完成后已将 lineage 同步回 `main`
+
+## 日常维护建议
+
+- 新功能优先进入 `main`
+- 正式发布始终通过 `release/X.Y` 执行
+- 补丁版本始终在对应的 `release/X.Y` 上维护
+- 每条新的 `release/X.Y` 在首个正式版本发布后执行一次 lineage 同步
+
+如当前流程发生调整，应同步更新本文档。
diff --git a/TEST_REPORT.md b/TEST_REPORT.md
new file mode 100644
index 0000000..35ac392
--- /dev/null
+++ b/TEST_REPORT.md
@@ -0,0 +1,71 @@
+# PaddleOCR API SDK 集成测试报告
+
+**测试分支**: feature/api-sdk (PR #18049)  
+**测试环境**: macOS Darwin 24.3.0, Python 3.9.6, requests 2.32.5  
+**API Endpoint**: https://paddleocr.aistudio-app.com/api/v2/ocr/jobs
+
+---
+
+## 测试结果总览
+
+| #   | 测试项　　　　　　　　　　　　　　　　| 结果　 | 耗时 | 说明　　　　　　　　　　　　　　　　　　　　　　　|
+| -----| ---------------------------------------| --------| ------| ---------------------------------------------------|
+| 1   | OCR URL (PP-OCRv5)　　　　　　　　　　| ✅ PASS | 3.6s | URL 输入，默认参数　　　　　　　　　　　　　　　　|
+| 2   | OCR URL + 自定义 Options　　　　　　　| ✅ PASS | 3.6s | 设置 use_doc_orientation_classify=True　　　　　　|
+| 3   | Doc Parsing URL (PP-StructureV3)　　　| ✅ PASS | 3.7s | 文档版面解析　　　　　　　　　　　　　　　　　　　|
+| 4   | Submit + Poll 分步调用　　　　　　　　| ✅ PASS | 3.7s | 非阻塞 API：submit → get_result → wait_for_result |
+| 5   | OCR 本地文件上传　　　　　　　　　　　| ✅ PASS | 4.3s | file_path 模式　　　　　　　　　　　　　　　　　　|
+| 6   | 错误处理 (无效 token)　　　　　　　　 | ✅ PASS | 0.2s | 正确抛出 AuthError　　　　　　　　　　　　　　　　|
+| 7   | 输入校验　　　　　　　　　　　　　　　| ✅ PASS | 0.0s | 缺少输入 / 互斥参数均正确拦截　　　　　　　　　　 |
+| 8   | Context Manager (with)　　　　　　　　| ✅ PASS | 3.6s | with 语句正常工作　　　　　　　　　　　　　　　　 |
+| 9   | PaddleOCR-VL 模型　　　　　　　　　　 | ✅ PASS | 3.6s | VL 模型正常返回 markdown　　　　　　　　　　　　　|
+| 10  | PaddleOCR-VL-1.5 模型　　　　　　　　 | ✅ PASS | 3.6s | VL-1.5 模型正常返回 markdown　　　　　　　　　　　|
+| 11  | Doc Parsing 文件上传 (PP-StructureV3) | ✅ PASS | 4.3s | 本地文件上传 + 文档解析　　　　　　　　　　　　　 |
+
+**总计: 11 passed, 0 failed**
+
+---
+
+## 发现的 Bug（已修复验证）
+
+### 🔴 阻塞性 Bug: `fetch_jsonl` 请求 BOS 签名 URL 时携带了多余的 Authorization header
+
+**文件**: `paddleocr/_api_client/_http.py` (同步) + `async_client.py` (异步)
+
+**现象**: 任务提交和轮询均成功，但在最后一步下载 JSONL 结果文件时，SDK 使用带有 `Authorization: bearer <paddle_token>` 的 session 去请求百度 BOS 对象存储的预签名 URL。BOS 不认识这个 header，返回 400 Bad Request。
+
+**根因**: `HTTPClient.fetch_jsonl()` 使用 `self._session.get(url)` 发起请求，而 session 在初始化时设置了 `Authorization` header。BOS 的预签名 URL 已经包含了自己的鉴权参数（`authorization=bce-auth-v1/...`），额外的 Authorization header 导致冲突。
+
+**修复**: `fetch_jsonl` 应使用不带 auth header 的独立请求：
+```python
+# 修复前
+resp = self._session.get(url, timeout=self._timeout)
+
+# 修复后
+resp = requests.get(url, timeout=self._timeout)
+```
+
+**影响范围**: 所有实际 API 调用（OCR、Doc Parsing、所有模型）在修复前均无法获取结果。这是一个 **必须在合入前修复的阻塞性 bug**。
+
+---
+
+## 其他已知问题（非阻塞，可后续迭代）
+
+| 严重度 | 语言 | 问题 |
+|--------|------|------|
+| 中 | Python | AsyncAPIClient._poll_until_done 使用硬编码 DEFAULT_MAX_WAIT_TIME，忽略用户设置的 timeout |
+| 中 | Go | submitURL/submitFile/getJobStatus 未使用 http.NewRequestWithContext，context 取消无法中断 HTTP 请求 |
+| 中 | TypeScript | poller.ts 的 sleep 方法 abort listener 未设置 `{ once: true }`，长轮询会泄漏 listener |
+| 中 | TypeScript | http.ts 的 fetchJsonl 未检查 resp.ok（同样的 BOS auth 问题可能也存在于 TS SDK） |
+| 低 | 全部 | 轮询循环先 sleep 再 check，对已完成任务多等 3 秒 |
+| 低 | Python | CLI argparse 的 store_true 传 False 而非 None，导致多余字段发送给 API |
+
+---
+
+## 结论
+
+**修复 `fetch_jsonl` 的 auth header 问题后，Python SDK 的核心功能全部正常工作。** 4 个模型（PP-OCRv5、PP-StructureV3、PaddleOCR-VL、PaddleOCR-VL-1.5）均可正常调用，URL 输入和文件上传两种模式均可用，错误处理和输入校验逻辑正确。
+
+**建议**: 
+1. ⚠️ **合入前必须修复** `fetch_jsonl` 的 BOS auth header bug（Python 同步 + 异步，以及 Go/TypeScript SDK 中的同类问题）
+2. 其他问题可以合入后迭代修复
diff --git a/api_sdk/README.md b/api_sdk/README.md
new file mode 100644
index 0000000..729cdfe
--- /dev/null
+++ b/api_sdk/README.md
@@ -0,0 +1,50 @@
+# PaddleOCR official API SDKs
+
+English | [简体中文](README_cn.md)
+
+This directory contains source-adjacent maintainer files for the PaddleOCR
+official API SDKs. The SDKs call hosted PaddleOCR official API services; they do
+not run local PaddleOCR inference or load local models.
+
+The official user documentation:
+
+- [Overview](../docs/version3.x/inference_deployment/serving/paddleocr_official_api/overview.md)
+- [Python SDK](../docs/version3.x/inference_deployment/serving/paddleocr_official_api/python.md)
+- [TypeScript SDK](../docs/version3.x/inference_deployment/serving/paddleocr_official_api/typescript.md)
+- [Go SDK](../docs/version3.x/inference_deployment/serving/paddleocr_official_api/go.md)
+- [CLI](../docs/version3.x/inference_deployment/serving/paddleocr_official_api/cli.md)
+
+## Maintainer Files
+
+| File | Purpose |
+| --- | --- |
+| [`typescript/README.md`](typescript/README.md) | Package-level README for the TypeScript SDK. |
+| [`go/README.md`](go/README.md) | Package-level README for the Go SDK. |
+
+The Python SDK is part of the main `paddleocr` package.
+
+## Package Locations
+
+| Language | Source location | User docs |
+| --- | --- | --- |
+| Python | [`../paddleocr`](../paddleocr) | [Python SDK](../docs/version3.x/inference_deployment/serving/paddleocr_official_api/python.md) |
+| TypeScript | [`typescript`](typescript) | [TypeScript SDK](../docs/version3.x/inference_deployment/serving/paddleocr_official_api/typescript.md) |
+| Go | [`go`](go) | [Go SDK](../docs/version3.x/inference_deployment/serving/paddleocr_official_api/go.md) |
+
+## Validation
+
+Run these from the PaddleOCR repo root directory:
+
+```bash
+# Python
+python -m pytest tests/api_client/
+
+# TypeScript
+cd api_sdk/typescript
+npm run lint
+npm test
+
+# Go
+cd ../go
+go test ./...
+```
diff --git a/api_sdk/README_cn.md b/api_sdk/README_cn.md
new file mode 100644
index 0000000..b5a9d3e
--- /dev/null
+++ b/api_sdk/README_cn.md
@@ -0,0 +1,48 @@
+# PaddleOCR 官方 API SDK
+
+[English](README.md) | 简体中文
+
+本目录包含 PaddleOCR 官方 API SDK 的源码相邻维护文档。SDK 调用 PaddleOCR 官方 API 托管服务；它们不在本地执行 PaddleOCR 推理，也不加载本地模型。
+
+正式用户文档：
+
+- [总览](../docs/version3.x/inference_deployment/serving/paddleocr_official_api/overview.md)
+- [Python SDK](../docs/version3.x/inference_deployment/serving/paddleocr_official_api/python.md)
+- [TypeScript SDK](../docs/version3.x/inference_deployment/serving/paddleocr_official_api/typescript.md)
+- [Go SDK](../docs/version3.x/inference_deployment/serving/paddleocr_official_api/go.md)
+- [CLI](../docs/version3.x/inference_deployment/serving/paddleocr_official_api/cli.md)
+
+## 维护者文件
+
+| 文件 | 作用 |
+| --- | --- |
+| [`typescript/README_cn.md`](typescript/README_cn.md) | TypeScript SDK 的包级 README。 |
+| [`go/README_cn.md`](go/README_cn.md) | Go SDK 的包级 README。 |
+
+Python SDK 是主 `paddleocr` 包的一部分。
+
+## 包位置
+
+| 语言 | 源码位置 | 用户文档 |
+| --- | --- | --- |
+| Python | [`../paddleocr`](../paddleocr) | [Python SDK](../docs/version3.x/inference_deployment/serving/paddleocr_official_api/python.md) |
+| TypeScript | [`typescript`](typescript) | [TypeScript SDK](../docs/version3.x/inference_deployment/serving/paddleocr_official_api/typescript.md) |
+| Go | [`go`](go) | [Go SDK](../docs/version3.x/inference_deployment/serving/paddleocr_official_api/go.md) |
+
+## 验证
+
+在 PaddleOCR 仓库根目录执行：
+
+```bash
+# Python
+python -m pytest tests/api_client/
+
+# TypeScript
+cd api_sdk/typescript
+npm run lint
+npm test
+
+# Go
+cd ../go
+go test ./...
+```
diff --git a/api_sdk/go/README.md b/api_sdk/go/README.md
new file mode 100644
index 0000000..f964c1f
--- /dev/null
+++ b/api_sdk/go/README.md
@@ -0,0 +1,72 @@
+# PaddleOCR Go SDK
+
+English | [简体中文](README_cn.md)
+
+Go client for the PaddleOCR official API. It submits OCR and document parsing
+jobs to hosted PaddleOCR services; it does not run local PaddleOCR inference or
+load local models.
+
+Official user docs:
+
+- [Go SDK](../../docs/version3.x/inference_deployment/serving/paddleocr_official_api/go.md)
+- [Go SDK (English)](../../docs/version3.x/inference_deployment/serving/paddleocr_official_api/go.en.md)
+
+## Install
+
+```bash
+go get github.com/PaddlePaddle/PaddleOCR/api_sdk/go
+```
+
+Versioned releases use submodule tags such as `api_sdk/go/v0.1.0`.
+
+## Minimal Usage
+
+Set `PADDLEOCR_ACCESS_TOKEN` or pass `WithToken` when constructing the client:
+
+```bash
+export PADDLEOCR_ACCESS_TOKEN="your-access-token"
+```
+
+```go
+client, err := paddleocr.NewClient()
+if err != nil {
+	return err
+}
+
+result, err := client.OCR(ctx, &paddleocr.OCRRequest{
+	Model:   paddleocr.PPOCRv5,
+	FileURL: "https://example.com/invoice.pdf",
+})
+if err != nil {
+	return err
+}
+fmt.Println(result.JobID, len(result.Pages))
+```
+
+Set `Model: paddleocr.PPOCRv6` (or `"PP-OCRv6"`) to use the PP-OCRv6 hosted OCR model.
+Set `Model: paddleocr.PPOCRv5Latin` (or `"PP-OCRv5-latin"`) to use the PP-OCRv5 Latin-script hosted OCR model.
+
+Document parsing defaults to PaddleOCR-VL-1.6:
+
+```go
+doc, err := client.ParseDocument(ctx, &paddleocr.DocParsingRequest{
+	FilePath: "./report.pdf",
+	Options: &paddleocr.PaddleOCRVLOptions{
+		UseChartRecognition: paddleocr.Bool(true),
+	},
+})
+if err != nil {
+	return err
+}
+fmt.Println(doc.JobID, len(doc.Pages))
+```
+
+## Build And Test
+
+```bash
+go test ./...
+go vet ./...
+go test -race ./...
+```
+
+`go test -race ./...` is recommended before public release.
diff --git a/api_sdk/go/README_cn.md b/api_sdk/go/README_cn.md
new file mode 100644
index 0000000..f718a12
--- /dev/null
+++ b/api_sdk/go/README_cn.md
@@ -0,0 +1,70 @@
+# PaddleOCR Go SDK
+
+[English](README.md) | 简体中文
+
+面向 PaddleOCR 官方 API 的 Go 客户端。它会把 OCR 和文档解析任务提交到 PaddleOCR 官方托管服务；不会运行本地 PaddleOCR 推理，也不会加载本地模型。
+
+正式用户文档：
+
+- [Go SDK](../../docs/version3.x/inference_deployment/serving/paddleocr_official_api/go.md)
+- [Go SDK（英文）](../../docs/version3.x/inference_deployment/serving/paddleocr_official_api/go.en.md)
+
+## 安装
+
+```bash
+go get github.com/PaddlePaddle/PaddleOCR/api_sdk/go
+```
+
+版本化发布使用 `api_sdk/go/v0.1.0` 这类子目录 module tag。
+
+## 最小示例
+
+设置 `PADDLEOCR_ACCESS_TOKEN`，或在构造客户端时传入 `WithToken`：
+
+```bash
+export PADDLEOCR_ACCESS_TOKEN="your-access-token"
+```
+
+```go
+client, err := paddleocr.NewClient()
+if err != nil {
+	return err
+}
+
+result, err := client.OCR(ctx, &paddleocr.OCRRequest{
+	Model:   paddleocr.PPOCRv5,
+	FileURL: "https://example.com/invoice.pdf",
+})
+if err != nil {
+	return err
+}
+fmt.Println(result.JobID, len(result.Pages))
+```
+
+将 `Model` 设为 `paddleocr.PPOCRv6`（或 `"PP-OCRv6"`）可使用 PP-OCRv6 云端 OCR 模型。
+将 `Model` 设为 `paddleocr.PPOCRv5Latin`（或 `"PP-OCRv5-latin"`）可使用 PP-OCRv5 拉丁语系云端 OCR 模型。
+
+文档解析默认使用 PaddleOCR-VL-1.6：
+
+```go
+doc, err := client.ParseDocument(ctx, &paddleocr.DocParsingRequest{
+	FilePath: "./report.pdf",
+	Options: &paddleocr.PaddleOCRVLOptions{
+		UseChartRecognition: paddleocr.Bool(true),
+	},
+})
+if err != nil {
+	return err
+}
+fmt.Println(doc.JobID, len(doc.Pages))
+```
+
+## 构建与测试
+
+```bash
+go test ./...
+go vet ./...
+go test -race ./...
+```
+
+公开发布前建议运行 `go test -race ./...`。
diff --git a/api_sdk/go/client.go b/api_sdk/go/client.go
new file mode 100644
index 0000000..4df6c0f
--- /dev/null
+++ b/api_sdk/go/client.go
@@ -0,0 +1,66 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package paddleocr
+
+import (
+	"net/http"
+	"os"
+	"strings"
+	"time"
+)
+
+type Client struct {
+	token          string
+	baseURL        string
+	jobsURL        string
+	requestTimeout time.Duration
+	pollTimeout    time.Duration
+	clientPlatform string
+	httpClient     *http.Client
+}
+
+func NewClient(opts ...ClientOption) (*Client, error) {
+	c := &Client{
+		requestTimeout: 5 * time.Minute,
+		pollTimeout:    10 * time.Minute,
+	}
+	for _, opt := range opts {
+		opt(c)
+	}
+	if c.token == "" {
+		c.token = os.Getenv("PADDLEOCR_ACCESS_TOKEN")
+	}
+	if c.token == "" {
+		return nil, &AuthError{PaddleOCRAPIError{Message: "Token is required. Set PADDLEOCR_ACCESS_TOKEN or use WithToken()."}}
+	}
+	if c.baseURL == "" {
+		c.baseURL = os.Getenv("PADDLEOCR_BASE_URL")
+	}
+	if c.baseURL == "" {
+		c.baseURL = DefaultBaseURL
+	}
+	c.baseURL = strings.TrimRight(c.baseURL, "/")
+	c.jobsURL = c.baseURL + apiPath
+	if c.httpClient == nil {
+		c.httpClient = &http.Client{Timeout: c.requestTimeout}
+	}
+	return c, nil
+}
+
+func (c *Client) setClientPlatformHeader(req *http.Request) {
+	if c.clientPlatform != "" {
+		req.Header.Set("Client-Platform", c.clientPlatform)
+	}
+}
diff --git a/api_sdk/go/client_test.go b/api_sdk/go/client_test.go
new file mode 100644
index 0000000..7434e3c
--- /dev/null
+++ b/api_sdk/go/client_test.go
@@ -0,0 +1,235 @@
+package paddleocr
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+)
+
+func TestClientPlatformHeader(t *testing.T) {
+	var got string
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		got = r.Header.Get("Client-Platform")
+		w.Header().Set("Content-Type", "application/json")
+		_ = json.NewEncoder(w).Encode(map[string]interface{}{
+			"code": 0,
+			"data": map[string]string{"jobId": "job-1"},
+		})
+	}))
+	defer server.Close()
+
+	client, err := NewClient(
+		WithToken("token"),
+		WithBaseURL(server.URL),
+		WithClientPlatform("my-app"),
+	)
+	if err != nil {
+		t.Fatalf("NewClient() error = %v", err)
+	}
+
+	job, err := client.SubmitOCR(context.Background(), &OCRRequest{
+		FileURL: "https://example.test/input.pdf",
+	})
+	if err != nil {
+		t.Fatalf("SubmitOCR() error = %v", err)
+	}
+	if job.JobID != "job-1" {
+		t.Fatalf("JobID = %q, want job-1", job.JobID)
+	}
+	if got != "my-app" {
+		t.Fatalf("Client-Platform = %q, want my-app", got)
+	}
+}
+
+func TestDocumentParsingOptionsIncludeCurrentAndFutureServiceParameters(t *testing.T) {
+	trueValue := true
+	falseValue := false
+	var got map[string]interface{}
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		var body struct {
+			OptionalPayload map[string]interface{} `json:"optionalPayload"`
+		}
+		if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
+			t.Fatalf("Decode request body error = %v", err)
+		}
+		got = body.OptionalPayload
+		w.Header().Set("Content-Type", "application/json")
+		_ = json.NewEncoder(w).Encode(map[string]interface{}{
+			"code": 0,
+			"data": map[string]string{"jobId": "job-doc"},
+		})
+	}))
+	defer server.Close()
+
+	client, err := NewClient(
+		WithToken("token"),
+		WithBaseURL(server.URL),
+	)
+	if err != nil {
+		t.Fatalf("NewClient() error = %v", err)
+	}
+
+	_, err = client.SubmitDocumentParsing(context.Background(), &DocParsingRequest{
+		Model:   PaddleOCRVL16,
+		FileURL: "https://example.test/doc.pdf",
+		Options: &PaddleOCRVLOptions{
+			UseOcrForImageBlock:  &trueValue,
+			FormatBlockContent:   &trueValue,
+			MarkdownIgnoreLabels: []string{"image"},
+			VlmExtraArgs:         map[string]interface{}{"temperature": 0.1},
+			ReturnMarkdownImages: &falseValue,
+			OutputFormats:        []string{"docx"},
+			ExtraOptions:         map[string]interface{}{"futureOption": "enabled"},
+		},
+	})
+	if err != nil {
+		t.Fatalf("SubmitDocumentParsing() error = %v", err)
+	}
+
+	if got["useOcrForImageBlock"] != true || got["formatBlockContent"] != true {
+		t.Fatalf("got boolean options %#v", got)
+	}
+	if got["returnMarkdownImages"] != false || got["futureOption"] != "enabled" {
+		t.Fatalf("got passthrough options %#v", got)
+	}
+}
+
+func TestResultParsersPreserveRawFieldsAndDataInfo(t *testing.T) {
+	ocrLine := map[string]interface{}{
+		"result": map[string]interface{}{
+			"dataInfo": map[string]interface{}{"numPages": float64(1)},
+			"ocrResults": []interface{}{
+				map[string]interface{}{
+					"prunedResult":          map[string]interface{}{"text": "hello"},
+					"ocrImage":              "ocr.png",
+					"docPreprocessingImage": "pre.png",
+					"inputImage":            "input.png",
+				},
+			},
+		},
+	}
+	ocrResult, err := parseOCRResult("job-ocr", []map[string]interface{}{ocrLine})
+	if err != nil {
+		t.Fatalf("parseOCRResult() error = %v", err)
+	}
+	if ocrResult.DataInfo["numPages"] != float64(1) {
+		t.Fatalf("OCR metadata not preserved: %#v", ocrResult)
+	}
+	if ocrResult.Pages[0].DocPreprocessingImageURL != "pre.png" || ocrResult.Pages[0].InputImageURL != "input.png" {
+		t.Fatalf("OCR page image URLs not preserved: %#v", ocrResult.Pages[0])
+	}
+	if ocrResult.Pages[0].Raw["ocrImage"] != "ocr.png" {
+		t.Fatalf("OCR raw page not preserved: %#v", ocrResult.Pages[0].Raw)
+	}
+
+	docPage := map[string]interface{}{
+		"prunedResult": map[string]interface{}{"blocks": []interface{}{map[string]interface{}{"label": "text"}}},
+		"markdown":     map[string]interface{}{"text": "hello", "images": map[string]interface{}{"figure.png": "figure-url"}, "isStart": true},
+		"outputImages": map[string]interface{}{"page.png": "page-url"},
+		"inputImage":   "input.png",
+		"exports":      map[string]interface{}{"docx": "docx-url"},
+	}
+	docLine := map[string]interface{}{
+		"result": map[string]interface{}{
+			"dataInfo":             map[string]interface{}{"numPages": float64(1)},
+			"layoutParsingResults": []interface{}{docPage},
+		},
+	}
+	docResult, err := parseDocParsingResult("job-doc", []map[string]interface{}{docLine})
+	if err != nil {
+		t.Fatalf("parseDocParsingResult() error = %v", err)
+	}
+	if docResult.DataInfo["numPages"] != float64(1) {
+		t.Fatalf("document metadata not preserved: %#v", docResult)
+	}
+	if docResult.Pages[0].PrunedResult == nil || docResult.Pages[0].Raw["inputImage"] != "input.png" {
+		t.Fatalf("document page raw fields not preserved: %#v", docResult.Pages[0])
+	}
+	if docResult.Pages[0].Exports["docx"] != "docx-url" || docResult.Pages[0].Markdown["isStart"] != true {
+		t.Fatalf("document structured fields not preserved: %#v", docResult.Pages[0])
+	}
+}
+
+func TestSubmitOCRAcceptsOfficialModelNameString(t *testing.T) {
+	var got string
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		var body struct {
+			Model string `json:"model"`
+		}
+		if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
+			t.Fatalf("Decode request body error = %v", err)
+		}
+		got = body.Model
+		w.Header().Set("Content-Type", "application/json")
+		_ = json.NewEncoder(w).Encode(map[string]interface{}{
+			"code": 0,
+			"data": map[string]string{"jobId": "job-1"},
+		})
+	}))
+	defer server.Close()
+
+	client, err := NewClient(
+		WithToken("token"),
+		WithBaseURL(server.URL),
+	)
+	if err != nil {
+		t.Fatalf("NewClient() error = %v", err)
+	}
+
+	job, err := client.SubmitOCR(context.Background(), &OCRRequest{
+		Model:   "PP-OCRv5",
+		FileURL: "https://example.test/input.pdf",
+	})
+	if err != nil {
+		t.Fatalf("SubmitOCR() error = %v", err)
+	}
+	if job.Model != "PP-OCRv5" {
+		t.Fatalf("Job model = %q, want PP-OCRv5", job.Model)
+	}
+	if got != "PP-OCRv5" {
+		t.Fatalf("Request model = %q, want PP-OCRv5", got)
+	}
+}
+
+func TestSubmitOCRAcceptsPPOCRv5LatinModelNameString(t *testing.T) {
+	var got string
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		var body struct {
+			Model string `json:"model"`
+		}
+		if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
+			t.Fatalf("Decode request body error = %v", err)
+		}
+		got = body.Model
+		w.Header().Set("Content-Type", "application/json")
+		_ = json.NewEncoder(w).Encode(map[string]interface{}{
+			"code": 0,
+			"data": map[string]string{"jobId": "job-latin"},
+		})
+	}))
+	defer server.Close()
+
+	client, err := NewClient(
+		WithToken("token"),
+		WithBaseURL(server.URL),
+	)
+	if err != nil {
+		t.Fatalf("NewClient() error = %v", err)
+	}
+
+	job, err := client.SubmitOCR(context.Background(), &OCRRequest{
+		Model:   PPOCRv5Latin,
+		FileURL: "https://example.test/latin.pdf",
+	})
+	if err != nil {
+		t.Fatalf("SubmitOCR() error = %v", err)
+	}
+	if job.Model != PPOCRv5Latin {
+		t.Fatalf("Job model = %q, want %s", job.Model, PPOCRv5Latin)
+	}
+	if got != PPOCRv5Latin {
+		t.Fatalf("Request model = %q, want %s", got, PPOCRv5Latin)
+	}
+}
diff --git a/api_sdk/go/doc.go b/api_sdk/go/doc.go
new file mode 100644
index 0000000..67dc185
--- /dev/null
+++ b/api_sdk/go/doc.go
@@ -0,0 +1,30 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Package paddleocr provides a Go client for the PaddleOCR official API.
+//
+// Create a client with NewClient and authenticate with WithToken or the
+// PADDLEOCR_ACCESS_TOKEN environment variable. Use OCR for supported OCR models
+// and ParseDocument for document parsing models.
+// SubmitOCR and SubmitDocumentParsing return an Operation for non-blocking
+// status checks with Poll or typed waits with WaitOCR and WaitDocumentParsing.
+// SaveResource downloads one result resource URL. SaveOCRResultResources and
+// SaveDocumentParsingResultResources save resources from typed result objects
+// into an existing directory.
+//
+// Request timeout and polling timeout are configured separately with
+// WithRequestTimeout and WithPollTimeout. Errors are exposed as typed values,
+// such as AuthError, InvalidRequestError, APIError, ResponseFormatError, and
+// ResultParseError, and are suitable for errors.As.
+package paddleocr
diff --git a/api_sdk/go/errors.go b/api_sdk/go/errors.go
new file mode 100644
index 0000000..dd4e484
--- /dev/null
+++ b/api_sdk/go/errors.go
@@ -0,0 +1,103 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package paddleocr
+
+import "fmt"
+
+type PaddleOCRAPIError struct {
+	Message string
+	Cause   error
+}
+
+func (e *PaddleOCRAPIError) Error() string {
+	if e.Message == "" && e.Cause != nil {
+		return e.Cause.Error()
+	}
+	return e.Message
+}
+
+func (e *PaddleOCRAPIError) Unwrap() error {
+	return e.Cause
+}
+
+type AuthError struct {
+	PaddleOCRAPIError
+}
+
+type InvalidRequestError struct {
+	PaddleOCRAPIError
+}
+
+type APIError struct {
+	StatusCode int
+	PaddleOCRAPIError
+}
+
+func (e *APIError) Error() string {
+	return fmt.Sprintf("HTTP %d: %s", e.StatusCode, e.Message)
+}
+
+type RateLimitError struct {
+	APIError
+}
+
+type ServiceUnavailableError struct {
+	APIError
+}
+
+type JobFailedError struct {
+	JobID    string
+	ErrorMsg string
+	PaddleOCRAPIError
+}
+
+func (e *JobFailedError) Error() string {
+	return fmt.Sprintf("Job %s failed: %s", e.JobID, e.ErrorMsg)
+}
+
+type RequestTimeoutError struct {
+	PaddleOCRAPIError
+}
+
+type PollTimeoutError struct {
+	JobID   string
+	Elapsed float64
+	PaddleOCRAPIError
+}
+
+func (e *PollTimeoutError) Error() string {
+	return fmt.Sprintf("Timed out after %.1fs waiting for job %s", e.Elapsed, e.JobID)
+}
+
+type NetworkError struct {
+	PaddleOCRAPIError
+}
+
+type FileNotFoundError struct {
+	Path string
+	PaddleOCRAPIError
+}
+
+func (e *FileNotFoundError) Error() string {
+	return fmt.Sprintf("File not found: %s", e.Path)
+}
+
+type ResponseFormatError struct {
+	PaddleOCRAPIError
+}
+
+type ResultParseError struct {
+	PaddleOCRAPIError
+}
diff --git a/api_sdk/go/examples/doc_parsing_file/main.go b/api_sdk/go/examples/doc_parsing_file/main.go
new file mode 100644
index 0000000..09081a6
--- /dev/null
+++ b/api_sdk/go/examples/doc_parsing_file/main.go
@@ -0,0 +1,61 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package main
+
+import (
+	"context"
+	"fmt"
+	"log"
+
+	paddleocr "github.com/PaddlePaddle/PaddleOCR/api_sdk/go"
+)
+
+func main() {
+	client, err := paddleocr.NewClient()
+	if err != nil {
+		log.Fatal(err)
+	}
+	ctx := context.Background()
+
+	// Convenience method (blocks until done)
+	result, err := client.ParseDocument(ctx, &paddleocr.DocParsingRequest{
+		Model:    paddleocr.PPStructureV3,
+		FilePath: "./sample.pdf",
+		Options:  &paddleocr.PPStructureV3Options{UseChartRecognition: paddleocr.Bool(true)},
+	})
+	if err != nil {
+		log.Fatal(err)
+	}
+	for i, page := range result.Pages {
+		fmt.Printf("Page %d:\n%s\n", i+1, page.MarkdownText)
+	}
+
+	// Manual control with typed job metadata and typed wait methods.
+	ocrJob, _ := client.SubmitOCR(ctx, &paddleocr.OCRRequest{FileURL: "https://example.com/f1.pdf"})
+	docJob, _ := client.SubmitDocumentParsing(ctx, &paddleocr.DocParsingRequest{
+		Model: paddleocr.PPStructureV3, FilePath: "./sample.pdf",
+	})
+
+	ocrResult, err := client.WaitOCRResult(ctx, ocrJob.JobID)
+	if err != nil {
+		log.Printf("OCR job error: %v", err)
+	}
+	docResult, err := client.WaitDocumentParsingResult(ctx, docJob.JobID)
+	if err != nil {
+		log.Printf("document parsing job error: %v", err)
+	}
+	fmt.Printf("OCR done: %v\n", ocrResult)
+	fmt.Printf("Document parsing done: %v\n", docResult)
+}
diff --git a/api_sdk/go/examples/ocr_url/main.go b/api_sdk/go/examples/ocr_url/main.go
new file mode 100644
index 0000000..fe21c35
--- /dev/null
+++ b/api_sdk/go/examples/ocr_url/main.go
@@ -0,0 +1,43 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package main
+
+import (
+	"context"
+	"fmt"
+	"log"
+
+	paddleocr "github.com/PaddlePaddle/PaddleOCR/api_sdk/go"
+)
+
+func main() {
+	client, err := paddleocr.NewClient()
+	if err != nil {
+		log.Fatal(err)
+	}
+	ctx := context.Background()
+
+	result, err := client.OCR(ctx, &paddleocr.OCRRequest{
+		FileURL: "https://example.com/invoice.pdf",
+	})
+	if err != nil {
+		log.Fatal(err)
+	}
+
+	for i, page := range result.Pages {
+		fmt.Printf("Page %d: %v\n", i+1, page.PrunedResult)
+		fmt.Printf("  Image URL: %s\n", page.OCRImageURL)
+	}
+}
diff --git a/api_sdk/go/go.mod b/api_sdk/go/go.mod
new file mode 100644
index 0000000..355dd87
--- /dev/null
+++ b/api_sdk/go/go.mod
@@ -0,0 +1,3 @@
+module github.com/PaddlePaddle/PaddleOCR/api_sdk/go
+
+go 1.21
diff --git a/api_sdk/go/models.go b/api_sdk/go/models.go
new file mode 100644
index 0000000..168ae67
--- /dev/null
+++ b/api_sdk/go/models.go
@@ -0,0 +1,159 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package paddleocr
+
+const (
+	PPOCRv5       = "PP-OCRv5"
+	PPOCRv5Latin  = "PP-OCRv5-latin"
+	PPOCRv6       = "PP-OCRv6"
+	PPStructureV3 = "PP-StructureV3"
+	PaddleOCRVL   = "PaddleOCR-VL"
+	PaddleOCRVL15 = "PaddleOCR-VL-1.5"
+	PaddleOCRVL16 = "PaddleOCR-VL-1.6"
+)
+
+// IsOCRModel reports whether model is supported by OCR APIs.
+func IsOCRModel(model string) bool {
+	return model == PPOCRv5 || model == PPOCRv5Latin || model == PPOCRv6
+}
+
+// IsDocumentParsingModel reports whether model is supported by document parsing APIs.
+func IsDocumentParsingModel(model string) bool {
+	switch model {
+	case PPStructureV3, PaddleOCRVL, PaddleOCRVL15, PaddleOCRVL16:
+		return true
+	default:
+		return false
+	}
+}
+
+// IsVLModel reports whether model is a PaddleOCR-VL family model.
+func IsVLModel(model string) bool {
+	switch model {
+	case PaddleOCRVL, PaddleOCRVL15, PaddleOCRVL16:
+		return true
+	default:
+		return false
+	}
+}
+
+type OCROptions struct {
+	UseDocOrientationClassify *bool                  `json:"useDocOrientationClassify,omitempty"`
+	UseDocUnwarping           *bool                  `json:"useDocUnwarping,omitempty"`
+	UseTextlineOrientation    *bool                  `json:"useTextlineOrientation,omitempty"`
+	TextDetLimitSideLen       *int                   `json:"textDetLimitSideLen,omitempty"`
+	TextDetLimitType          *string                `json:"textDetLimitType,omitempty"`
+	TextDetThresh             *float64               `json:"textDetThresh,omitempty"`
+	TextDetBoxThresh          *float64               `json:"textDetBoxThresh,omitempty"`
+	TextDetUnclipRatio        *float64               `json:"textDetUnclipRatio,omitempty"`
+	TextRecScoreThresh        *float64               `json:"textRecScoreThresh,omitempty"`
+	Visualize                 *bool                  `json:"visualize,omitempty"`
+	ExtraOptions              map[string]interface{} `json:"-"`
+}
+
+type PPStructureV3Options struct {
+	UseDocOrientationClassify        *bool                  `json:"useDocOrientationClassify,omitempty"`
+	UseDocUnwarping                  *bool                  `json:"useDocUnwarping,omitempty"`
+	UseTextlineOrientation           *bool                  `json:"useTextlineOrientation,omitempty"`
+	UseSealRecognition               *bool                  `json:"useSealRecognition,omitempty"`
+	UseTableRecognition              *bool                  `json:"useTableRecognition,omitempty"`
+	UseFormulaRecognition            *bool                  `json:"useFormulaRecognition,omitempty"`
+	UseChartRecognition              *bool                  `json:"useChartRecognition,omitempty"`
+	UseRegionDetection               *bool                  `json:"useRegionDetection,omitempty"`
+	LayoutThreshold                  interface{}            `json:"layoutThreshold,omitempty"`
+	LayoutNms                        *bool                  `json:"layoutNms,omitempty"`
+	LayoutUnclipRatio                interface{}            `json:"layoutUnclipRatio,omitempty"`
+	LayoutMergeBboxesMode            interface{}            `json:"layoutMergeBboxesMode,omitempty"`
+	FormatBlockContent               *bool                  `json:"formatBlockContent,omitempty"`
+	TextDetLimitSideLen              *int                   `json:"textDetLimitSideLen,omitempty"`
+	TextDetLimitType                 *string                `json:"textDetLimitType,omitempty"`
+	TextDetThresh                    *float64               `json:"textDetThresh,omitempty"`
+	TextDetBoxThresh                 *float64               `json:"textDetBoxThresh,omitempty"`
+	TextDetUnclipRatio               *float64               `json:"textDetUnclipRatio,omitempty"`
+	TextRecScoreThresh               *float64               `json:"textRecScoreThresh,omitempty"`
+	UseWiredTableCellsTransToHtml    *bool                  `json:"useWiredTableCellsTransToHtml,omitempty"`
+	UseWirelessTableCellsTransToHtml *bool                  `json:"useWirelessTableCellsTransToHtml,omitempty"`
+	UseTableOrientationClassify      *bool                  `json:"useTableOrientationClassify,omitempty"`
+	UseOcrResultsWithTableCells      *bool                  `json:"useOcrResultsWithTableCells,omitempty"`
+	UseE2eWiredTableRecModel         *bool                  `json:"useE2eWiredTableRecModel,omitempty"`
+	UseE2eWirelessTableRecModel      *bool                  `json:"useE2eWirelessTableRecModel,omitempty"`
+	MarkdownIgnoreLabels             []string               `json:"markdownIgnoreLabels,omitempty"`
+	PrettifyMarkdown                 *bool                  `json:"prettifyMarkdown,omitempty"`
+	ShowFormulaNumber                *bool                  `json:"showFormulaNumber,omitempty"`
+	ReturnMarkdownImages             *bool                  `json:"returnMarkdownImages,omitempty"`
+	OutputFormats                    []string               `json:"outputFormats,omitempty"`
+	Visualize                        *bool                  `json:"visualize,omitempty"`
+	ExtraOptions                     map[string]interface{} `json:"-"`
+}
+
+type PaddleOCRVLOptions struct {
+	UseDocOrientationClassify *bool                  `json:"useDocOrientationClassify,omitempty"`
+	UseDocUnwarping           *bool                  `json:"useDocUnwarping,omitempty"`
+	UseLayoutDetection        *bool                  `json:"useLayoutDetection,omitempty"`
+	UseChartRecognition       *bool                  `json:"useChartRecognition,omitempty"`
+	UseSealRecognition        *bool                  `json:"useSealRecognition,omitempty"`
+	UseOcrForImageBlock       *bool                  `json:"useOcrForImageBlock,omitempty"`
+	LayoutThreshold           interface{}            `json:"layoutThreshold,omitempty"`
+	LayoutNms                 *bool                  `json:"layoutNms,omitempty"`
+	LayoutUnclipRatio         interface{}            `json:"layoutUnclipRatio,omitempty"`
+	LayoutMergeBboxesMode     interface{}            `json:"layoutMergeBboxesMode,omitempty"`
+	LayoutShapeMode           *string                `json:"layoutShapeMode,omitempty"`
+	PromptLabel               *string                `json:"promptLabel,omitempty"`
+	FormatBlockContent        *bool                  `json:"formatBlockContent,omitempty"`
+	RepetitionPenalty         *float64               `json:"repetitionPenalty,omitempty"`
+	Temperature               *float64               `json:"temperature,omitempty"`
+	TopP                      *float64               `json:"topP,omitempty"`
+	MinPixels                 *int                   `json:"minPixels,omitempty"`
+	MaxPixels                 *int                   `json:"maxPixels,omitempty"`
+	MaxNewTokens              *int                   `json:"maxNewTokens,omitempty"`
+	VlmExtraArgs              map[string]interface{} `json:"vlmExtraArgs,omitempty"`
+	MergeLayoutBlocks         *bool                  `json:"mergeLayoutBlocks,omitempty"`
+	MarkdownIgnoreLabels      []string               `json:"markdownIgnoreLabels,omitempty"`
+	PrettifyMarkdown          *bool                  `json:"prettifyMarkdown,omitempty"`
+	ShowFormulaNumber         *bool                  `json:"showFormulaNumber,omitempty"`
+	RestructurePages          *bool                  `json:"restructurePages,omitempty"`
+	MergeTables               *bool                  `json:"mergeTables,omitempty"`
+	RelevelTitles             *bool                  `json:"relevelTitles,omitempty"`
+	ReturnMarkdownImages      *bool                  `json:"returnMarkdownImages,omitempty"`
+	OutputFormats             []string               `json:"outputFormats,omitempty"`
+	Visualize                 *bool                  `json:"visualize,omitempty"`
+	ExtraOptions              map[string]interface{} `json:"-"`
+}
+
+// DocParsingOptionsProvider marks document parsing option structs.
+type DocParsingOptionsProvider interface {
+	isDocParsingOptions()
+}
+
+func (*PPStructureV3Options) isDocParsingOptions() {}
+func (*PaddleOCRVLOptions) isDocParsingOptions()   {}
+
+type OCRRequest struct {
+	Model      string
+	FileURL    string
+	FilePath   string
+	PageRanges string
+	BatchID    string
+	Options    *OCROptions
+}
+
+type DocParsingRequest struct {
+	Model      string
+	FileURL    string
+	FilePath   string
+	PageRanges string
+	BatchID    string
+	Options    DocParsingOptionsProvider
+}
diff --git a/api_sdk/go/ocr.go b/api_sdk/go/ocr.go
new file mode 100644
index 0000000..b7fd845
--- /dev/null
+++ b/api_sdk/go/ocr.go
@@ -0,0 +1,317 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package paddleocr
+
+import (
+	"context"
+	"encoding/json"
+)
+
+func (c *Client) OCR(ctx context.Context, req *OCRRequest) (*OCRResult, error) {
+	job, err := c.SubmitOCR(ctx, req)
+	if err != nil {
+		return nil, err
+	}
+	return c.WaitOCRResult(ctx, job.JobID)
+}
+
+// ParseDocument performs document parsing. Blocks until result is ready.
+func (c *Client) ParseDocument(ctx context.Context, req *DocParsingRequest) (*DocParsingResult, error) {
+	job, err := c.SubmitDocumentParsing(ctx, req)
+	if err != nil {
+		return nil, err
+	}
+	return c.WaitDocumentParsingResult(ctx, job.JobID)
+}
+
+// SubmitOCR submits an OCR job and returns job metadata for tracking.
+func (c *Client) SubmitOCR(ctx context.Context, req *OCRRequest) (*Job, error) {
+	if req == nil {
+		return nil, &InvalidRequestError{PaddleOCRAPIError{Message: "OCR request is nil"}}
+	}
+	model := req.Model
+	if model == "" {
+		model = PPOCRv6
+	}
+	if !IsOCRModel(model) {
+		return nil, &InvalidRequestError{PaddleOCRAPIError{Message: "model is not an OCR model: " + model}}
+	}
+	jobID, err := c.submit(ctx, model, req.FileURL, req.FilePath, req.Options, req.PageRanges, req.BatchID)
+	if err != nil {
+		return nil, err
+	}
+	return &Job{JobID: jobID, Model: model, Task: "ocr", PageRanges: req.PageRanges, BatchID: req.BatchID}, nil
+}
+
+// SubmitDocumentParsing submits a document parsing job and returns job metadata for tracking.
+func (c *Client) SubmitDocumentParsing(ctx context.Context, req *DocParsingRequest) (*Job, error) {
+	if req == nil {
+		return nil, &InvalidRequestError{PaddleOCRAPIError{Message: "document parsing request is nil"}}
+	}
+	model := req.Model
+	if model == "" {
+		model = PaddleOCRVL16
+	}
+	if !IsDocumentParsingModel(model) {
+		return nil, &InvalidRequestError{PaddleOCRAPIError{Message: "model is not a document parsing model: " + model}}
+	}
+	jobID, err := c.submit(ctx, model, req.FileURL, req.FilePath, req.Options, req.PageRanges, req.BatchID)
+	if err != nil {
+		return nil, err
+	}
+	return &Job{JobID: jobID, Model: model, Task: "document_parsing", PageRanges: req.PageRanges, BatchID: req.BatchID}, nil
+}
+
+func (c *Client) WaitOCRResult(ctx context.Context, jobID string) (*OCRResult, error) {
+	jsonlData, err := c.pollUntilDone(ctx, jobID)
+	if err != nil {
+		return nil, err
+	}
+	return parseOCRResult(jobID, jsonlData)
+}
+
+func (c *Client) WaitDocumentParsingResult(ctx context.Context, jobID string) (*DocParsingResult, error) {
+	jsonlData, err := c.pollUntilDone(ctx, jobID)
+	if err != nil {
+		return nil, err
+	}
+	return parseDocParsingResult(jobID, jsonlData)
+}
+
+func (c *Client) GetStatus(ctx context.Context, jobID string) (*JobStatus, error) {
+	status, err := c.getJobStatus(ctx, jobID)
+	if err != nil {
+		return nil, err
+	}
+	return normalizeStatus(jobID, status)
+}
+
+func (c *Client) GetBatchStatus(ctx context.Context, batchID string) (*BatchStatus, error) {
+	if batchID == "" {
+		return nil, &InvalidRequestError{PaddleOCRAPIError{Message: "batchID is required"}}
+	}
+	return c.getBatchStatus(ctx, batchID)
+}
+
+func (c *Client) submit(ctx context.Context, model, fileURL, filePath string, options interface{}, pageRanges, batchID string) (string, error) {
+	if fileURL == "" && filePath == "" {
+		return "", &InvalidRequestError{PaddleOCRAPIError{Message: "Either FileURL or FilePath is required."}}
+	}
+	if fileURL != "" && filePath != "" {
+		return "", &InvalidRequestError{PaddleOCRAPIError{Message: "FileURL and FilePath are mutually exclusive."}}
+	}
+
+	payload := defaultPayload(model, options)
+
+	if fileURL != "" {
+		return c.submitURL(ctx, model, fileURL, payload, pageRanges, batchID)
+	}
+	return c.submitFile(ctx, model, filePath, payload, pageRanges, batchID)
+}
+
+func defaultPayload(model string, options interface{}) interface{} {
+	switch typed := options.(type) {
+	case *OCROptions:
+		if typed != nil {
+			return payloadWithExtraOptions(typed)
+		}
+	case *PPStructureV3Options:
+		if typed != nil {
+			return payloadWithExtraOptions(typed)
+		}
+	case *PaddleOCRVLOptions:
+		if typed != nil {
+			return payloadWithExtraOptions(typed)
+		}
+	default:
+		if options != nil {
+			return payloadWithExtraOptions(options)
+		}
+	}
+	if IsOCRModel(model) {
+		return payloadWithExtraOptions(&OCROptions{})
+	}
+	if IsVLModel(model) {
+		return payloadWithExtraOptions(&PaddleOCRVLOptions{})
+	}
+	return payloadWithExtraOptions(&PPStructureV3Options{})
+}
+
+func payloadWithExtraOptions(options interface{}) interface{} {
+	payloadBytes, _ := json.Marshal(options)
+	payload := map[string]interface{}{}
+	_ = json.Unmarshal(payloadBytes, &payload)
+
+	var extraOptions map[string]interface{}
+	switch typed := options.(type) {
+	case *OCROptions:
+		extraOptions = typed.ExtraOptions
+	case *PPStructureV3Options:
+		extraOptions = typed.ExtraOptions
+	case *PaddleOCRVLOptions:
+		extraOptions = typed.ExtraOptions
+	}
+	for key, value := range extraOptions {
+		payload[key] = value
+	}
+	return payload
+}
+
+func parseOCRResult(jobID string, jsonlData []map[string]interface{}) (*OCRResult, error) {
+	result := &OCRResult{JobID: jobID, DataInfo: map[string]interface{}{}}
+	for _, lineObj := range jsonlData {
+		resultData, ok := lineObj["result"].(map[string]interface{})
+		if !ok {
+			return nil, &ResultParseError{PaddleOCRAPIError{Message: "OCR result item is missing result"}}
+		}
+		if dataInfo, ok := resultData["dataInfo"].(map[string]interface{}); ok {
+			for key, value := range dataInfo {
+				result.DataInfo[key] = value
+			}
+		}
+		ocrResults, ok := resultData["ocrResults"].([]interface{})
+		if !ok {
+			return nil, &ResultParseError{PaddleOCRAPIError{Message: "OCR result item is missing ocrResults"}}
+		}
+		for _, item := range ocrResults {
+			itemMap, ok := item.(map[string]interface{})
+			if !ok {
+				return nil, &ResultParseError{PaddleOCRAPIError{Message: "OCR result page is malformed"}}
+			}
+			if _, ok := itemMap["prunedResult"]; !ok {
+				return nil, &ResultParseError{PaddleOCRAPIError{Message: "OCR result page is missing prunedResult"}}
+			}
+			page := OCRPage{
+				PrunedResult:             itemMap["prunedResult"],
+				OCRImageURL:              getString(itemMap, "ocrImage"),
+				DocPreprocessingImageURL: getString(itemMap, "docPreprocessingImage"),
+				InputImageURL:            getString(itemMap, "inputImage"),
+				Raw:                      itemMap,
+			}
+			result.Pages = append(result.Pages, page)
+		}
+	}
+	return result, nil
+}
+
+func parseDocParsingResult(jobID string, jsonlData []map[string]interface{}) (*DocParsingResult, error) {
+	result := &DocParsingResult{JobID: jobID, DataInfo: map[string]interface{}{}}
+	for _, lineObj := range jsonlData {
+		resultData, ok := lineObj["result"].(map[string]interface{})
+		if !ok {
+			return nil, &ResultParseError{PaddleOCRAPIError{Message: "document parsing result item is missing result"}}
+		}
+		if dataInfo, ok := resultData["dataInfo"].(map[string]interface{}); ok {
+			for key, value := range dataInfo {
+				result.DataInfo[key] = value
+			}
+		}
+		lpResults, ok := resultData["layoutParsingResults"].([]interface{})
+		if !ok {
+			return nil, &ResultParseError{PaddleOCRAPIError{Message: "document parsing result item is missing layoutParsingResults"}}
+		}
+		for _, item := range lpResults {
+			itemMap, ok := item.(map[string]interface{})
+			if !ok {
+				return nil, &ResultParseError{PaddleOCRAPIError{Message: "document parsing result page is malformed"}}
+			}
+			markdown, ok := itemMap["markdown"].(map[string]interface{})
+			if !ok || getString(markdown, "text") == "" {
+				return nil, &ResultParseError{PaddleOCRAPIError{Message: "document parsing result page is missing markdown.text"}}
+			}
+			page := DocParsingPage{
+				MarkdownText:   getString(markdown, "text"),
+				MarkdownImages: getStringMap(markdown, "images"),
+				OutputImages:   getStringMap(itemMap, "outputImages"),
+				PrunedResult:   itemMap["prunedResult"],
+				InputImageURL:  getString(itemMap, "inputImage"),
+				Exports:        getMap(itemMap, "exports"),
+				Markdown:       markdown,
+				Raw:            itemMap,
+			}
+			result.Pages = append(result.Pages, page)
+		}
+	}
+	return result, nil
+}
+
+func getString(m map[string]interface{}, key string) string {
+	if m == nil {
+		return ""
+	}
+	v, _ := m[key].(string)
+	return v
+}
+
+func getMap(m map[string]interface{}, key string) map[string]interface{} {
+	value, ok := m[key].(map[string]interface{})
+	if !ok {
+		return map[string]interface{}{}
+	}
+	return value
+}
+
+func getStringMap(m map[string]interface{}, key string) map[string]string {
+	if m == nil {
+		return nil
+	}
+	raw, ok := m[key]
+	if !ok {
+		return nil
+	}
+	switch v := raw.(type) {
+	case map[string]interface{}:
+		result := make(map[string]string, len(v))
+		for k, val := range v {
+			if s, ok := val.(string); ok {
+				result[k] = s
+			}
+		}
+		return result
+	default:
+		b, _ := json.Marshal(raw)
+		var result map[string]string
+		json.Unmarshal(b, &result)
+		return result
+	}
+}
+
+func normalizeStatus(jobID string, status *jobStatusResponse) (*JobStatus, error) {
+	if status == nil || status.State == "" {
+		return nil, &ResponseFormatError{PaddleOCRAPIError{Message: "status response is missing state"}}
+	}
+	if status.State != "pending" && status.State != "running" && status.State != "done" && status.State != "failed" {
+		return nil, &ResponseFormatError{PaddleOCRAPIError{Message: "unknown job state: " + status.State}}
+	}
+	result := &JobStatus{
+		JobID:     jobID,
+		State:     status.State,
+		ResultURL: status.ResultURL,
+		ErrorMsg:  status.ErrorMsg,
+	}
+	if len(status.ExtractProgress) != 0 && string(status.ExtractProgress) != "null" {
+		var progress extractProgress
+		if err := json.Unmarshal(status.ExtractProgress, &progress); err != nil {
+			return nil, &ResponseFormatError{PaddleOCRAPIError{Message: "status progress is malformed", Cause: err}}
+		}
+		result.Progress = &Progress{
+			TotalPages:     progress.TotalPages,
+			ExtractedPages: progress.ExtractedPages,
+			StartTime:      progress.StartTime,
+			EndTime:        progress.EndTime,
+		}
+	}
+	return result, nil
+}
diff --git a/api_sdk/go/operation.go b/api_sdk/go/operation.go
new file mode 100644
index 0000000..86a558f
--- /dev/null
+++ b/api_sdk/go/operation.go
@@ -0,0 +1,69 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package paddleocr
+
+import (
+	"context"
+	"encoding/json"
+)
+
+// Operation represents an in-progress job. Use Wait() to block until done,
+// or Poll() to check status without blocking.
+type Operation struct {
+	client *Client
+	JobID  string
+	model  string
+}
+
+// Wait blocks until the job completes and returns the parsed result.
+func (op *Operation) Wait(ctx context.Context) (interface{}, error) {
+	jsonlData, err := op.client.pollUntilDone(ctx, op.JobID)
+	if err != nil {
+		return nil, err
+	}
+	if IsOCRModel(op.model) {
+		return parseOCRResult(op.JobID, jsonlData)
+	}
+	return parseDocParsingResult(op.JobID, jsonlData)
+}
+
+// Poll checks the current job status without waiting.
+// Returns the status, whether the job is done, and any error.
+func (op *Operation) Poll(ctx context.Context) (*JobStatus, bool, error) {
+	status, err := op.client.getJobStatus(ctx, op.JobID)
+	if err != nil {
+		return nil, false, err
+	}
+
+	js := &JobStatus{
+		JobID:    op.JobID,
+		State:    status.State,
+		ErrorMsg: status.ErrorMsg,
+	}
+
+	if status.ExtractProgress != nil {
+		var ep extractProgress
+		if err := json.Unmarshal(status.ExtractProgress, &ep); err == nil {
+			js.Progress = &Progress{
+				TotalPages:     ep.TotalPages,
+				ExtractedPages: ep.ExtractedPages,
+				StartTime:      ep.StartTime,
+				EndTime:        ep.EndTime,
+			}
+		}
+	}
+
+	return js, status.State == "done", nil
+}
diff --git a/api_sdk/go/options.go b/api_sdk/go/options.go
new file mode 100644
index 0000000..6b958ea
--- /dev/null
+++ b/api_sdk/go/options.go
@@ -0,0 +1,73 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package paddleocr
+
+import (
+	"net/http"
+	"time"
+)
+
+const DefaultBaseURL = "https://paddleocr.aistudio-app.com"
+
+const apiPath = "/api/v2/ocr/jobs"
+
+type ClientOption func(*Client)
+
+func WithToken(token string) ClientOption {
+	return func(c *Client) {
+		c.token = token
+	}
+}
+
+func WithBaseURL(url string) ClientOption {
+	return func(c *Client) {
+		c.baseURL = url
+	}
+}
+
+func WithTimeout(d time.Duration) ClientOption {
+	return func(c *Client) {
+		c.requestTimeout = d
+		c.pollTimeout = d
+	}
+}
+
+func WithRequestTimeout(d time.Duration) ClientOption {
+	return func(c *Client) {
+		c.requestTimeout = d
+	}
+}
+
+func WithPollTimeout(d time.Duration) ClientOption {
+	return func(c *Client) {
+		c.pollTimeout = d
+	}
+}
+
+func WithClientPlatform(clientPlatform string) ClientOption {
+	return func(c *Client) {
+		c.clientPlatform = clientPlatform
+	}
+}
+
+func WithHTTPClient(hc *http.Client) ClientOption {
+	return func(c *Client) {
+		c.httpClient = hc
+	}
+}
+
+func Bool(v bool) *bool {
+	return &v
+}
diff --git a/api_sdk/go/poller.go b/api_sdk/go/poller.go
new file mode 100644
index 0000000..4a4ce85
--- /dev/null
+++ b/api_sdk/go/poller.go
@@ -0,0 +1,90 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package paddleocr
+
+import (
+	"context"
+	"fmt"
+	"time"
+)
+
+const (
+	initialInterval = 3 * time.Second
+	multiplier      = 1.5
+	maxInterval     = 15 * time.Second
+)
+
+func (c *Client) pollUntilDone(ctx context.Context, jobID string) ([]map[string]interface{}, error) {
+	interval := initialInterval
+	deadline := time.Now().Add(c.pollTimeout)
+	pollCtx, cancel := context.WithDeadline(ctx, deadline)
+	defer cancel()
+
+	for time.Now().Before(deadline) {
+		remaining := time.Until(deadline)
+		if remaining <= 0 {
+			break
+		}
+		if interval > remaining {
+			interval = remaining
+		}
+		timer := time.NewTimer(interval)
+		select {
+		case <-pollCtx.Done():
+			timer.Stop()
+			if ctx.Err() != nil {
+				return nil, ctx.Err()
+			}
+			return nil, &PollTimeoutError{
+				JobID:             jobID,
+				Elapsed:           c.pollTimeout.Seconds(),
+				PaddleOCRAPIError: PaddleOCRAPIError{Message: fmt.Sprintf("Timed out after %.1fs", c.pollTimeout.Seconds())},
+			}
+		case <-timer.C:
+		}
+
+		status, err := c.getJobStatus(pollCtx, jobID)
+		if err != nil {
+			return nil, err
+		}
+
+		switch status.State {
+		case "done":
+			jsonURL := status.ResultURL["jsonUrl"]
+			if jsonURL == "" {
+				return nil, &ResponseFormatError{PaddleOCRAPIError{Message: "done job response is missing resultUrl.jsonUrl"}}
+			}
+			return c.fetchJSONL(pollCtx, jsonURL)
+		case "failed":
+			return nil, &JobFailedError{
+				JobID:             jobID,
+				ErrorMsg:          status.ErrorMsg,
+				PaddleOCRAPIError: PaddleOCRAPIError{Message: status.ErrorMsg},
+			}
+		}
+
+		next := time.Duration(float64(interval) * multiplier)
+		if next > maxInterval {
+			next = maxInterval
+		}
+		interval = next
+	}
+
+	return nil, &PollTimeoutError{
+		JobID:             jobID,
+		Elapsed:           c.pollTimeout.Seconds(),
+		PaddleOCRAPIError: PaddleOCRAPIError{Message: fmt.Sprintf("Timed out after %.1fs", c.pollTimeout.Seconds())},
+	}
+}
diff --git a/api_sdk/go/resource.go b/api_sdk/go/resource.go
new file mode 100644
index 0000000..bef6841
--- /dev/null
+++ b/api_sdk/go/resource.go
@@ -0,0 +1,267 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package paddleocr
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"net/url"
+	"os"
+	"path"
+	"path/filepath"
+	"sort"
+	"strings"
+)
+
+type SaveResourceOption func(*saveResourceOptions)
+
+type saveResourceOptions struct {
+	overwrite bool
+}
+
+func WithOverwrite(overwrite bool) SaveResourceOption {
+	return func(o *saveResourceOptions) {
+		o.overwrite = overwrite
+	}
+}
+
+// SaveResource downloads one result resource URL.
+func (c *Client) SaveResource(ctx context.Context, resourceURL, dest string, opts ...SaveResourceOption) (string, error) {
+	if resourceURL == "" {
+		return "", &InvalidRequestError{PaddleOCRAPIError{Message: "resource URL is required"}}
+	}
+	if dest == "" {
+		return "", &InvalidRequestError{PaddleOCRAPIError{Message: "destination path is required"}}
+	}
+	options := saveResourceOptions{}
+	for _, opt := range opts {
+		opt(&options)
+	}
+	savedPath, err := resolveSavePath(resourceURL, dest)
+	if err != nil {
+		return "", err
+	}
+	parent := filepath.Dir(savedPath)
+	if st, err := os.Stat(parent); err != nil {
+		if errors.Is(err, os.ErrNotExist) {
+			return "", &FileNotFoundError{Path: parent, PaddleOCRAPIError: PaddleOCRAPIError{Message: "destination parent not found: " + parent, Cause: err}}
+		}
+		return "", err
+	} else if !st.IsDir() {
+		return "", &FileNotFoundError{Path: parent, PaddleOCRAPIError: PaddleOCRAPIError{Message: "destination parent is not a directory: " + parent}}
+	}
+	if !options.overwrite {
+		if _, err := os.Stat(savedPath); err == nil {
+			return "", &InvalidRequestError{PaddleOCRAPIError{Message: "destination already exists: " + savedPath}}
+		} else if err != nil && !errors.Is(err, os.ErrNotExist) {
+			return "", err
+		}
+	}
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, resourceURL, nil)
+	if err != nil {
+		return "", &NetworkError{PaddleOCRAPIError{Message: err.Error(), Cause: err}}
+	}
+	resp, err := c.httpClient.Do(req)
+	if err != nil {
+		return "", classifyHTTPError(err)
+	}
+	defer resp.Body.Close()
+	if err := raiseForResourceResponse(resp); err != nil {
+		return "", err
+	}
+
+	tmp, err := os.CreateTemp(parent, "."+filepath.Base(savedPath)+".tmp-*")
+	if err != nil {
+		return "", err
+	}
+	tmpPath := tmp.Name()
+	cleanupTemp := true
+	defer func() {
+		if cleanupTemp {
+			_ = os.Remove(tmpPath)
+		}
+	}()
+	if _, err := io.Copy(tmp, resp.Body); err != nil {
+		_ = tmp.Close()
+		return "", err
+	}
+	if err := tmp.Close(); err != nil {
+		return "", err
+	}
+	if options.overwrite {
+		if err := os.Rename(tmpPath, savedPath); err != nil {
+			return "", err
+		}
+	} else {
+		if err := os.Link(tmpPath, savedPath); err != nil {
+			if errors.Is(err, os.ErrExist) {
+				return "", &InvalidRequestError{PaddleOCRAPIError{Message: "destination already exists: " + savedPath, Cause: err}}
+			}
+			return "", err
+		}
+		if err := os.Remove(tmpPath); err != nil {
+			return "", err
+		}
+	}
+	cleanupTemp = false
+	return savedPath, nil
+}
+
+func (c *Client) SaveOCRResultResources(ctx context.Context, result *OCRResult, destDir string, opts ...SaveResourceOption) ([]string, error) {
+	if result == nil {
+		return nil, &InvalidRequestError{PaddleOCRAPIError{Message: "OCR result is required"}}
+	}
+	if err := requireExistingDirectory(destDir); err != nil {
+		return nil, err
+	}
+	saved := make([]string, 0, len(result.Pages))
+	for i, page := range result.Pages {
+		if page.OCRImageURL == "" {
+			continue
+		}
+		name := fmt.Sprintf("ocr-page-%d%s", i+1, safeResourceExtension(page.OCRImageURL))
+		if err := validateResultResourceFilename(name); err != nil {
+			return nil, err
+		}
+		path, err := c.SaveResource(ctx, page.OCRImageURL, filepath.Join(destDir, name), opts...)
+		if err != nil {
+			return nil, err
+		}
+		saved = append(saved, path)
+	}
+	return saved, nil
+}
+
+func (c *Client) SaveDocumentParsingResultResources(ctx context.Context, result *DocParsingResult, destDir string, opts ...SaveResourceOption) ([]string, error) {
+	if result == nil {
+		return nil, &InvalidRequestError{PaddleOCRAPIError{Message: "document parsing result is required"}}
+	}
+	if err := requireExistingDirectory(destDir); err != nil {
+		return nil, err
+	}
+	var saved []string
+	for _, page := range result.Pages {
+		paths, err := c.saveNamedResourceMap(ctx, page.MarkdownImages, destDir, opts...)
+		if err != nil {
+			return nil, err
+		}
+		saved = append(saved, paths...)
+		paths, err = c.saveNamedResourceMap(ctx, page.OutputImages, destDir, opts...)
+		if err != nil {
+			return nil, err
+		}
+		saved = append(saved, paths...)
+	}
+	return saved, nil
+}
+
+func (c *Client) saveNamedResourceMap(ctx context.Context, resources map[string]string, destDir string, opts ...SaveResourceOption) ([]string, error) {
+	keys := make([]string, 0, len(resources))
+	for key := range resources {
+		keys = append(keys, key)
+	}
+	sort.Strings(keys)
+	saved := make([]string, 0, len(keys))
+	for _, key := range keys {
+		resourceURL := resources[key]
+		if resourceURL == "" {
+			continue
+		}
+		name, err := resultResourceFilename(key)
+		if err != nil {
+			return nil, err
+		}
+		path, err := c.SaveResource(ctx, resourceURL, filepath.Join(destDir, name), opts...)
+		if err != nil {
+			return nil, err
+		}
+		saved = append(saved, path)
+	}
+	return saved, nil
+}
+
+func resolveSavePath(resourceURL, dest string) (string, error) {
+	if st, err := os.Stat(dest); err == nil && st.IsDir() {
+		name := "resource"
+		u, parseErr := url.Parse(resourceURL)
+		if parseErr == nil {
+			if base := path.Base(u.Path); base != "." && base != "/" && base != "" {
+				name = base
+			}
+		}
+		return filepath.Join(dest, name), nil
+	} else if err != nil && !errors.Is(err, os.ErrNotExist) {
+		return "", err
+	}
+	return dest, nil
+}
+
+func raiseForResourceResponse(resp *http.Response) error {
+	if resp.StatusCode >= 200 && resp.StatusCode < 300 {
+		return nil
+	}
+	body, _ := io.ReadAll(resp.Body)
+	return &APIError{StatusCode: resp.StatusCode, PaddleOCRAPIError: PaddleOCRAPIError{Message: string(body)}}
+}
+
+func requireExistingDirectory(destDir string) error {
+	if destDir == "" {
+		return &InvalidRequestError{PaddleOCRAPIError{Message: "destination directory is required"}}
+	}
+	st, err := os.Stat(destDir)
+	if err != nil {
+		if errors.Is(err, os.ErrNotExist) {
+			return &FileNotFoundError{Path: destDir, PaddleOCRAPIError: PaddleOCRAPIError{Message: "destination directory not found: " + destDir, Cause: err}}
+		}
+		return err
+	}
+	if !st.IsDir() {
+		return &FileNotFoundError{Path: destDir, PaddleOCRAPIError: PaddleOCRAPIError{Message: "destination is not a directory: " + destDir}}
+	}
+	return nil
+}
+
+func resultResourceFilename(key string) (string, error) {
+	if err := validateResultResourceFilename(key); err != nil {
+		return "", err
+	}
+	return key, nil
+}
+
+func validateResultResourceFilename(name string) error {
+	if name == "" || name == "." || name == ".." || filepath.IsAbs(name) || path.IsAbs(name) || strings.ContainsAny(name, `/\`) {
+		return &InvalidRequestError{PaddleOCRAPIError{Message: "unsafe resource filename: " + name}}
+	}
+	return nil
+}
+
+func safeResourceExtension(resourceURL string) string {
+	u, err := url.Parse(resourceURL)
+	if err != nil {
+		return ""
+	}
+	ext := path.Ext(path.Base(u.Path))
+	if len(ext) <= 1 {
+		return ""
+	}
+	if err := validateResultResourceFilename("resource" + ext); err != nil {
+		return ""
+	}
+	return ext
+}
diff --git a/api_sdk/go/results.go b/api_sdk/go/results.go
new file mode 100644
index 0000000..81a0d14
--- /dev/null
+++ b/api_sdk/go/results.go
@@ -0,0 +1,74 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package paddleocr
+
+type OCRPage struct {
+	PrunedResult             interface{}            `json:"prunedResult"`
+	OCRImageURL              string                 `json:"ocrImageUrl,omitempty"`
+	DocPreprocessingImageURL string                 `json:"docPreprocessingImageUrl,omitempty"`
+	InputImageURL            string                 `json:"inputImageUrl,omitempty"`
+	Raw                      map[string]interface{} `json:"raw,omitempty"`
+}
+
+type DocParsingPage struct {
+	MarkdownText   string                 `json:"markdownText"`
+	MarkdownImages map[string]string      `json:"markdownImages"`
+	OutputImages   map[string]string      `json:"outputImages"`
+	PrunedResult   interface{}            `json:"prunedResult,omitempty"`
+	InputImageURL  string                 `json:"inputImageUrl,omitempty"`
+	Exports        map[string]interface{} `json:"exports,omitempty"`
+	Markdown       map[string]interface{} `json:"markdown,omitempty"`
+	Raw            map[string]interface{} `json:"raw,omitempty"`
+}
+
+type OCRResult struct {
+	JobID    string                 `json:"jobId"`
+	Pages    []OCRPage              `json:"pages"`
+	DataInfo map[string]interface{} `json:"dataInfo,omitempty"`
+}
+
+type DocParsingResult struct {
+	JobID    string                 `json:"jobId"`
+	Pages    []DocParsingPage       `json:"pages"`
+	DataInfo map[string]interface{} `json:"dataInfo,omitempty"`
+}
+
+type Progress struct {
+	TotalPages     int    `json:"totalPages"`
+	ExtractedPages int    `json:"extractedPages"`
+	StartTime      string `json:"startTime,omitempty"`
+	EndTime        string `json:"endTime,omitempty"`
+}
+
+type Job struct {
+	JobID      string `json:"jobId"`
+	Model      string `json:"model"`
+	Task       string `json:"task"`
+	PageRanges string `json:"pageRanges,omitempty"`
+	BatchID    string `json:"batchId,omitempty"`
+}
+
+type JobStatus struct {
+	JobID     string            `json:"jobId"`
+	State     string            `json:"state"`
+	Progress  *Progress         `json:"progress,omitempty"`
+	ResultURL map[string]string `json:"resultUrl,omitempty"`
+	ErrorMsg  string            `json:"errorMsg,omitempty"`
+}
+
+type BatchStatus struct {
+	BatchID string       `json:"batchId"`
+	Jobs    []*JobStatus `json:"jobs"`
+}
diff --git a/api_sdk/go/transport.go b/api_sdk/go/transport.go
new file mode 100644
index 0000000..b2f01a3
--- /dev/null
+++ b/api_sdk/go/transport.go
@@ -0,0 +1,329 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package paddleocr
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"io"
+	"mime/multipart"
+	"net"
+	"net/http"
+	"os"
+	"path/filepath"
+	"strings"
+)
+
+type apiResponse struct {
+	Code int             `json:"code"`
+	Msg  string          `json:"msg"`
+	Data json.RawMessage `json:"data"`
+}
+
+type submitResponse struct {
+	JobID string `json:"jobId"`
+}
+
+type jobStatusResponse struct {
+	JobID           string            `json:"jobId"`
+	State           string            `json:"state"`
+	ExtractProgress json.RawMessage   `json:"extractProgress"`
+	ResultURL       map[string]string `json:"resultUrl"`
+	ErrorMsg        string            `json:"errorMsg"`
+}
+
+type extractProgress struct {
+	TotalPages     int    `json:"totalPages"`
+	ExtractedPages int    `json:"extractedPages"`
+	StartTime      string `json:"startTime"`
+	EndTime        string `json:"endTime"`
+}
+
+func (c *Client) submitURL(ctx context.Context, model, fileURL string, payload interface{}, pageRanges, batchID string) (string, error) {
+	body := map[string]interface{}{
+		"fileUrl":         fileURL,
+		"model":           model,
+		"optionalPayload": payload,
+	}
+	if pageRanges != "" {
+		body["pageRanges"] = pageRanges
+	}
+	if batchID != "" {
+		body["batchId"] = batchID
+	}
+	jsonBody, err := json.Marshal(body)
+	if err != nil {
+		return "", &ResponseFormatError{PaddleOCRAPIError{Message: "failed to encode submit request", Cause: err}}
+	}
+
+	req, err := http.NewRequestWithContext(ctx, "POST", c.jobsURL, bytes.NewReader(jsonBody))
+	if err != nil {
+		return "", &NetworkError{PaddleOCRAPIError{Message: err.Error()}}
+	}
+	req.Header.Set("Authorization", "Bearer "+c.token)
+	req.Header.Set("Content-Type", "application/json")
+	c.setClientPlatformHeader(req)
+
+	resp, err := c.httpClient.Do(req)
+	if err != nil {
+		return "", classifyHTTPError(err)
+	}
+	defer resp.Body.Close()
+
+	if err := raiseForResponse(resp); err != nil {
+		return "", err
+	}
+
+	apiResp, err := decodeAPIResponse(resp)
+	if err != nil {
+		return "", err
+	}
+	var sr submitResponse
+	if err := json.Unmarshal(apiResp.Data, &sr); err != nil {
+		return "", &ResponseFormatError{PaddleOCRAPIError{Message: "submit response is malformed", Cause: err}}
+	}
+	if sr.JobID == "" {
+		return "", &ResponseFormatError{PaddleOCRAPIError{Message: "submit response is missing jobId"}}
+	}
+	return sr.JobID, nil
+}
+
+func (c *Client) submitFile(ctx context.Context, model, filePath string, payload interface{}, pageRanges, batchID string) (string, error) {
+	if _, err := os.Stat(filePath); os.IsNotExist(err) {
+		return "", &FileNotFoundError{Path: filePath, PaddleOCRAPIError: PaddleOCRAPIError{Message: "File not found: " + filePath}}
+	}
+
+	var buf bytes.Buffer
+	w := multipart.NewWriter(&buf)
+
+	_ = w.WriteField("model", model)
+	payloadJSON, err := json.Marshal(payload)
+	if err != nil {
+		return "", &ResponseFormatError{PaddleOCRAPIError{Message: "failed to encode optionalPayload", Cause: err}}
+	}
+	if err := w.WriteField("optionalPayload", string(payloadJSON)); err != nil {
+		return "", &ResponseFormatError{PaddleOCRAPIError{Message: "failed to prepare multipart body", Cause: err}}
+	}
+	if pageRanges != "" {
+		_ = w.WriteField("pageRanges", pageRanges)
+	}
+	if batchID != "" {
+		_ = w.WriteField("batchId", batchID)
+	}
+
+	file, err := os.Open(filePath)
+	if err != nil {
+		return "", &FileNotFoundError{Path: filePath, PaddleOCRAPIError: PaddleOCRAPIError{Message: "Failed to open file: " + filePath, Cause: err}}
+	}
+	defer file.Close()
+
+	fw, err := w.CreateFormFile("file", filepath.Base(filePath))
+	if err != nil {
+		return "", &ResponseFormatError{PaddleOCRAPIError{Message: "failed to prepare multipart body", Cause: err}}
+	}
+	if _, err := io.Copy(fw, file); err != nil {
+		return "", &NetworkError{PaddleOCRAPIError{Message: "failed to read file content", Cause: err}}
+	}
+	if err := w.Close(); err != nil {
+		return "", &ResponseFormatError{PaddleOCRAPIError{Message: "failed to finalize multipart body", Cause: err}}
+	}
+
+	req, err := http.NewRequestWithContext(ctx, "POST", c.jobsURL, &buf)
+	if err != nil {
+		return "", &NetworkError{PaddleOCRAPIError{Message: err.Error()}}
+	}
+	req.Header.Set("Authorization", "Bearer "+c.token)
+	req.Header.Set("Content-Type", w.FormDataContentType())
+	c.setClientPlatformHeader(req)
+
+	resp, err := c.httpClient.Do(req)
+	if err != nil {
+		return "", classifyHTTPError(err)
+	}
+	defer resp.Body.Close()
+
+	if err := raiseForResponse(resp); err != nil {
+		return "", err
+	}
+
+	apiResp, err := decodeAPIResponse(resp)
+	if err != nil {
+		return "", err
+	}
+	var sr submitResponse
+	if err := json.Unmarshal(apiResp.Data, &sr); err != nil {
+		return "", &ResponseFormatError{PaddleOCRAPIError{Message: "submit response is malformed", Cause: err}}
+	}
+	if sr.JobID == "" {
+		return "", &ResponseFormatError{PaddleOCRAPIError{Message: "submit response is missing jobId"}}
+	}
+	return sr.JobID, nil
+}
+
+func (c *Client) getJobStatus(ctx context.Context, jobID string) (*jobStatusResponse, error) {
+	req, err := http.NewRequestWithContext(ctx, "GET", c.jobsURL+"/"+jobID, nil)
+	if err != nil {
+		return nil, &NetworkError{PaddleOCRAPIError{Message: err.Error()}}
+	}
+	req.Header.Set("Authorization", "Bearer "+c.token)
+	c.setClientPlatformHeader(req)
+
+	resp, err := c.httpClient.Do(req)
+	if err != nil {
+		return nil, classifyHTTPError(err)
+	}
+	defer resp.Body.Close()
+
+	if err := raiseForResponse(resp); err != nil {
+		return nil, err
+	}
+
+	apiResp, err := decodeAPIResponse(resp)
+	if err != nil {
+		return nil, err
+	}
+	var status jobStatusResponse
+	if err := json.Unmarshal(apiResp.Data, &status); err != nil {
+		return nil, &ResponseFormatError{PaddleOCRAPIError{Message: "status response is malformed", Cause: err}}
+	}
+	if status.State == "" {
+		return nil, &ResponseFormatError{PaddleOCRAPIError{Message: "status response is missing state"}}
+	}
+	return &status, nil
+}
+
+func (c *Client) getBatchStatus(ctx context.Context, batchID string) (*BatchStatus, error) {
+	req, err := http.NewRequestWithContext(ctx, "GET", c.jobsURL+"/batch/"+batchID, nil)
+	if err != nil {
+		return nil, &NetworkError{PaddleOCRAPIError{Message: err.Error(), Cause: err}}
+	}
+	req.Header.Set("Authorization", "Bearer "+c.token)
+	c.setClientPlatformHeader(req)
+
+	resp, err := c.httpClient.Do(req)
+	if err != nil {
+		return nil, classifyHTTPError(err)
+	}
+	defer resp.Body.Close()
+
+	if err := raiseForResponse(resp); err != nil {
+		return nil, err
+	}
+	apiResp, err := decodeAPIResponse(resp)
+	if err != nil {
+		return nil, err
+	}
+	var payload struct {
+		ExtractResult []jobStatusResponse `json:"extractResult"`
+	}
+	if err := json.Unmarshal(apiResp.Data, &payload); err != nil {
+		return nil, &ResponseFormatError{PaddleOCRAPIError{Message: "batch response is malformed", Cause: err}}
+	}
+	result := &BatchStatus{BatchID: batchID}
+	for _, item := range payload.ExtractResult {
+		jobID := item.JobID
+		if jobID == "" {
+			return nil, &ResponseFormatError{PaddleOCRAPIError{Message: "batch response item is missing jobId"}}
+		}
+		status, err := normalizeStatus(jobID, &item)
+		if err != nil {
+			return nil, err
+		}
+		result.Jobs = append(result.Jobs, status)
+	}
+	return result, nil
+}
+
+func (c *Client) fetchJSONL(ctx context.Context, url string) ([]map[string]interface{}, error) {
+	req, err := http.NewRequestWithContext(ctx, "GET", url, nil)
+	if err != nil {
+		return nil, &NetworkError{PaddleOCRAPIError{Message: err.Error()}}
+	}
+	resp, err := c.httpClient.Do(req)
+	if err != nil {
+		return nil, classifyHTTPError(err)
+	}
+	defer resp.Body.Close()
+	if err := raiseForResponse(resp); err != nil {
+		return nil, err
+	}
+
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, err
+	}
+
+	lines := strings.Split(strings.TrimSpace(string(body)), "\n")
+	var results []map[string]interface{}
+	for _, line := range lines {
+		line = strings.TrimSpace(line)
+		if line == "" {
+			continue
+		}
+		var obj map[string]interface{}
+		if err := json.Unmarshal([]byte(line), &obj); err != nil {
+			return nil, &ResultParseError{PaddleOCRAPIError{Message: "failed to parse JSONL result payload", Cause: err}}
+		}
+		results = append(results, obj)
+	}
+	return results, nil
+}
+
+func raiseForResponse(resp *http.Response) error {
+	if resp.StatusCode >= 200 && resp.StatusCode < 300 {
+		return nil
+	}
+	body, _ := io.ReadAll(resp.Body)
+	msg := string(body)
+
+	switch {
+	case resp.StatusCode == 401 || resp.StatusCode == 403:
+		return &AuthError{PaddleOCRAPIError{Message: "Authentication failed: " + msg}}
+	case resp.StatusCode == 400:
+		return &InvalidRequestError{PaddleOCRAPIError{Message: "Bad request: " + msg}}
+	case resp.StatusCode == 429:
+		return &RateLimitError{APIError{StatusCode: 429, PaddleOCRAPIError: PaddleOCRAPIError{Message: "Rate limit exceeded: " + msg}}}
+	case resp.StatusCode == 503 || resp.StatusCode == 504:
+		return &ServiceUnavailableError{APIError{StatusCode: resp.StatusCode, PaddleOCRAPIError: PaddleOCRAPIError{Message: "Service unavailable: " + msg}}}
+	default:
+		return &APIError{StatusCode: resp.StatusCode, PaddleOCRAPIError: PaddleOCRAPIError{Message: msg}}
+	}
+}
+
+func decodeAPIResponse(resp *http.Response) (*apiResponse, error) {
+	var apiResp apiResponse
+	if err := json.NewDecoder(resp.Body).Decode(&apiResp); err != nil {
+		return nil, &ResponseFormatError{PaddleOCRAPIError{Message: "expected a JSON response body", Cause: err}}
+	}
+	if apiResp.Code != 0 {
+		msg := apiResp.Msg
+		if msg == "" {
+			msg = "PaddleOCR official API request failed"
+		}
+		return nil, &APIError{StatusCode: resp.StatusCode, PaddleOCRAPIError: PaddleOCRAPIError{Message: msg}}
+	}
+	if len(apiResp.Data) == 0 || string(apiResp.Data) == "null" {
+		return nil, &ResponseFormatError{PaddleOCRAPIError{Message: "response body is missing data"}}
+	}
+	return &apiResp, nil
+}
+
+func classifyHTTPError(err error) error {
+	if ne, ok := err.(net.Error); ok && ne.Timeout() {
+		return &RequestTimeoutError{PaddleOCRAPIError: PaddleOCRAPIError{Message: err.Error(), Cause: err}}
+	}
+	return &NetworkError{PaddleOCRAPIError{Message: err.Error(), Cause: err}}
+}
diff --git a/api_sdk/typescript/.gitignore b/api_sdk/typescript/.gitignore
new file mode 100644
index 0000000..a45cc03
--- /dev/null
+++ b/api_sdk/typescript/.gitignore
@@ -0,0 +1,9 @@
+node_modules/
+dist/
+.vite/
+coverage/
+.nyc_output/
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+pnpm-debug.log*
diff --git a/api_sdk/typescript/README.md b/api_sdk/typescript/README.md
new file mode 100644
index 0000000..06216f6
--- /dev/null
+++ b/api_sdk/typescript/README.md
@@ -0,0 +1,72 @@
+# PaddleOCR TypeScript SDK
+
+English | [简体中文](README_cn.md)
+
+TypeScript client for the PaddleOCR official API. It submits OCR and document
+parsing jobs to hosted PaddleOCR services; it does not perform local OCR
+inference.
+
+Official user docs:
+
+- [TypeScript SDK](../../docs/version3.x/inference_deployment/serving/paddleocr_official_api/typescript.md)
+- [TypeScript SDK (English)](../../docs/version3.x/inference_deployment/serving/paddleocr_official_api/typescript.en.md)
+
+## Install
+
+```bash
+npm install @paddleocr/api-sdk
+```
+
+This package follows SemVer and is published as a public scoped npm package.
+
+For local development:
+
+```bash
+npm install
+npm run build
+```
+
+## Minimal Usage
+
+Set `PADDLEOCR_ACCESS_TOKEN` or pass `token` to the client:
+
+```bash
+export PADDLEOCR_ACCESS_TOKEN="your-access-token"
+```
+
+```ts
+import { Model, PaddleOCRClient } from "@paddleocr/api-sdk";
+
+const client = new PaddleOCRClient();
+const result = await client.ocr({
+  model: Model.PPOCRv5,
+  fileUrl: "https://example.com/invoice.pdf",
+});
+
+console.log(result.jobId, result.pages.length);
+```
+
+Specify `model: Model.PPOCRv6` (or `"PP-OCRv6"`) to use the PP-OCRv6 hosted OCR model.
+Use `Model.PPOCRv5Latin` (or `"PP-OCRv5-latin"`) for the PP-OCRv5 Latin-script hosted OCR model.
+
+Document parsing defaults to PaddleOCR-VL-1.6:
+
+```ts
+const doc = await client.parseDocument({
+  filePath: "./report.pdf",
+  options: {
+    useChartRecognition: true,
+  },
+});
+
+console.log(doc.jobId, doc.pages.length);
+```
+
+## Build And Test
+
+```bash
+npm run lint
+npm run build
+npm test
+npm audit --audit-level=moderate
+```
diff --git a/api_sdk/typescript/README_cn.md b/api_sdk/typescript/README_cn.md
new file mode 100644
index 0000000..ef156f9
--- /dev/null
+++ b/api_sdk/typescript/README_cn.md
@@ -0,0 +1,70 @@
+# PaddleOCR TypeScript SDK
+
+[English](README.md) | 简体中文
+
+面向 PaddleOCR 官方 API 的 TypeScript 客户端。它会把 OCR 和文档解析任务提交到 PaddleOCR 官方托管服务；不会在本地执行 OCR 推理。
+
+正式用户文档：
+
+- [TypeScript SDK](../../docs/version3.x/inference_deployment/serving/paddleocr_official_api/typescript.md)
+- [TypeScript SDK（英文）](../../docs/version3.x/inference_deployment/serving/paddleocr_official_api/typescript.en.md)
+
+## 安装
+
+```bash
+npm install @paddleocr/api-sdk
+```
+
+该包遵循语义化版本，并作为公开 scoped npm 包发布。
+
+本地开发：
+
+```bash
+npm install
+npm run build
+```
+
+## 最小示例
+
+设置 `PADDLEOCR_ACCESS_TOKEN`，或在构造客户端时传入 `token`：
+
+```bash
+export PADDLEOCR_ACCESS_TOKEN="your-access-token"
+```
+
+```ts
+import { Model, PaddleOCRClient } from "@paddleocr/api-sdk";
+
+const client = new PaddleOCRClient();
+const result = await client.ocr({
+  model: Model.PPOCRv5,
+  fileUrl: "https://example.com/invoice.pdf",
+});
+
+console.log(result.jobId, result.pages.length);
+```
+
+通过 `model: Model.PPOCRv6`（或 `"PP-OCRv6"`）可指定 PP-OCRv6 云端 OCR 模型。
+使用 `Model.PPOCRv5Latin`（或 `"PP-OCRv5-latin"`）可指定 PP-OCRv5 拉丁语系云端 OCR 模型。
+
+文档解析默认使用 PaddleOCR-VL-1.6：
+
+```ts
+const doc = await client.parseDocument({
+  filePath: "./report.pdf",
+  options: {
+    useChartRecognition: true,
+  },
+});
+
+console.log(doc.jobId, doc.pages.length);
+```
+
+## 构建与测试
+
+```bash
+npm run lint
+npm run build
+npm test
+npm audit --audit-level=moderate
+```
diff --git a/api_sdk/typescript/examples/doc-parsing-file.ts b/api_sdk/typescript/examples/doc-parsing-file.ts
new file mode 100644
index 0000000..7cd751b
--- /dev/null
+++ b/api_sdk/typescript/examples/doc-parsing-file.ts
@@ -0,0 +1,47 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import { PaddleOCRClient, Model } from "@paddleocr/api-sdk";
+
+async function main() {
+  const client = new PaddleOCRClient();
+
+  // Convenience: doc parsing with local file
+  const result = await client.parseDocument({
+    model: Model.PPStructureV3,
+    filePath: "./sample.pdf",
+    options: { useChartRecognition: true },
+  });
+
+  for (const page of result.pages) {
+    console.log(page.markdownText);
+  }
+
+  // Manual control: submit + concurrent wait
+  const job1 = await client.submitOcr({ fileUrl: "https://example.com/f1.pdf" });
+  const job2 = await client.submitDocumentParsing({
+    model: Model.PPStructureV3,
+    filePath: "./sample.pdf",
+  });
+
+  const [r1, r2] = await Promise.all([
+    client.waitOcrResult(job1.jobId),
+    client.waitDocumentParsingResult(job2.jobId),
+  ]);
+
+  console.log("Job1 done:", r1);
+  console.log("Job2 done:", r2);
+}
+
+main().catch(console.error);
diff --git a/api_sdk/typescript/examples/ocr-url.ts b/api_sdk/typescript/examples/ocr-url.ts
new file mode 100644
index 0000000..430334f
--- /dev/null
+++ b/api_sdk/typescript/examples/ocr-url.ts
@@ -0,0 +1,31 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import { PaddleOCRClient, Model } from "@paddleocr/api-sdk";
+
+async function main() {
+  const client = new PaddleOCRClient();
+
+  // Convenience: OCR with URL
+  const ocrResult = await client.ocr({
+    fileUrl: "https://example.com/invoice.pdf",
+  });
+
+  for (const page of ocrResult.pages) {
+    console.log("Result:", page.prunedResult);
+    console.log("Image:", page.ocrImageUrl);
+  }
+}
+
+main().catch(console.error);
diff --git a/api_sdk/typescript/package-lock.json b/api_sdk/typescript/package-lock.json
new file mode 100644
index 0000000..a3060ed
--- /dev/null
+++ b/api_sdk/typescript/package-lock.json
@@ -0,0 +1,2640 @@
+{
+  "name": "@paddleocr/api-sdk",
+  "version": "0.2.2",
+  "lockfileVersion": 3,
+  "requires": true,
+  "packages": {
+    "": {
+      "name": "@paddleocr/api-sdk",
+      "version": "0.2.2",
+      "license": "Apache-2.0",
+      "devDependencies": {
+        "@types/node": "^25.9.1",
+        "tsup": "^8.0.0",
+        "typescript": "^5.3.0",
+        "vitest": "^4.1.7"
+      },
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@emnapi/core": {
+      "version": "1.10.0",
+      "resolved": "https://registry.npmjs.org/@emnapi/core/-/core-1.10.0.tgz",
+      "integrity": "sha512-yq6OkJ4p82CAfPl0u9mQebQHKPJkY7WrIuk205cTYnYe+k2Z8YBh11FrbRG/H6ihirqcacOgl2BIO8oyMQLeXw==",
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "@emnapi/wasi-threads": "1.2.1",
+        "tslib": "^2.4.0"
+      }
+    },
+    "node_modules/@emnapi/runtime": {
+      "version": "1.10.0",
+      "resolved": "https://registry.npmjs.org/@emnapi/runtime/-/runtime-1.10.0.tgz",
+      "integrity": "sha512-ewvYlk86xUoGI0zQRNq/mC+16R1QeDlKQy21Ki3oSYXNgLb45GV1P6A0M+/s6nyCuNDqe5VpaY84BzXGwVbwFA==",
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "tslib": "^2.4.0"
+      }
+    },
+    "node_modules/@emnapi/wasi-threads": {
+      "version": "1.2.1",
+      "resolved": "https://registry.npmjs.org/@emnapi/wasi-threads/-/wasi-threads-1.2.1.tgz",
+      "integrity": "sha512-uTII7OYF+/Mes/MrcIOYp5yOtSMLBWSIoLPpcgwipoiKbli6k322tcoFsxoIIxPDqW01SQGAgko4EzZi2BNv2w==",
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "tslib": "^2.4.0"
+      }
+    },
+    "node_modules/@esbuild/aix-ppc64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/aix-ppc64/-/aix-ppc64-0.27.7.tgz",
+      "integrity": "sha512-EKX3Qwmhz1eMdEJokhALr0YiD0lhQNwDqkPYyPhiSwKrh7/4KRjQc04sZ8db+5DVVnZ1LmbNDI1uAMPEUBnQPg==",
+      "cpu": [
+        "ppc64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "aix"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/android-arm": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/android-arm/-/android-arm-0.27.7.tgz",
+      "integrity": "sha512-jbPXvB4Yj2yBV7HUfE2KHe4GJX51QplCN1pGbYjvsyCZbQmies29EoJbkEc+vYuU5o45AfQn37vZlyXy4YJ8RQ==",
+      "cpu": [
+        "arm"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "android"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/android-arm64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/android-arm64/-/android-arm64-0.27.7.tgz",
+      "integrity": "sha512-62dPZHpIXzvChfvfLJow3q5dDtiNMkwiRzPylSCfriLvZeq0a1bWChrGx/BbUbPwOrsWKMn8idSllklzBy+dgQ==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "android"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/android-x64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/android-x64/-/android-x64-0.27.7.tgz",
+      "integrity": "sha512-x5VpMODneVDb70PYV2VQOmIUUiBtY3D3mPBG8NxVk5CogneYhkR7MmM3yR/uMdITLrC1ml/NV1rj4bMJuy9MCg==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "android"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/darwin-arm64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/darwin-arm64/-/darwin-arm64-0.27.7.tgz",
+      "integrity": "sha512-5lckdqeuBPlKUwvoCXIgI2D9/ABmPq3Rdp7IfL70393YgaASt7tbju3Ac+ePVi3KDH6N2RqePfHnXkaDtY9fkw==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "darwin"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/darwin-x64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/darwin-x64/-/darwin-x64-0.27.7.tgz",
+      "integrity": "sha512-rYnXrKcXuT7Z+WL5K980jVFdvVKhCHhUwid+dDYQpH+qu+TefcomiMAJpIiC2EM3Rjtq0sO3StMV/+3w3MyyqQ==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "darwin"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/freebsd-arm64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/freebsd-arm64/-/freebsd-arm64-0.27.7.tgz",
+      "integrity": "sha512-B48PqeCsEgOtzME2GbNM2roU29AMTuOIN91dsMO30t+Ydis3z/3Ngoj5hhnsOSSwNzS+6JppqWsuhTp6E82l2w==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "freebsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/freebsd-x64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/freebsd-x64/-/freebsd-x64-0.27.7.tgz",
+      "integrity": "sha512-jOBDK5XEjA4m5IJK3bpAQF9/Lelu/Z9ZcdhTRLf4cajlB+8VEhFFRjWgfy3M1O4rO2GQ/b2dLwCUGpiF/eATNQ==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "freebsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-arm": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-arm/-/linux-arm-0.27.7.tgz",
+      "integrity": "sha512-RkT/YXYBTSULo3+af8Ib0ykH8u2MBh57o7q/DAs3lTJlyVQkgQvlrPTnjIzzRPQyavxtPtfg0EopvDyIt0j1rA==",
+      "cpu": [
+        "arm"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-arm64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-arm64/-/linux-arm64-0.27.7.tgz",
+      "integrity": "sha512-RZPHBoxXuNnPQO9rvjh5jdkRmVizktkT7TCDkDmQ0W2SwHInKCAV95GRuvdSvA7w4VMwfCjUiPwDi0ZO6Nfe9A==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-ia32": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-ia32/-/linux-ia32-0.27.7.tgz",
+      "integrity": "sha512-GA48aKNkyQDbd3KtkplYWT102C5sn/EZTY4XROkxONgruHPU72l+gW+FfF8tf2cFjeHaRbWpOYa/uRBz/Xq1Pg==",
+      "cpu": [
+        "ia32"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-loong64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-loong64/-/linux-loong64-0.27.7.tgz",
+      "integrity": "sha512-a4POruNM2oWsD4WKvBSEKGIiWQF8fZOAsycHOt6JBpZ+JN2n2JH9WAv56SOyu9X5IqAjqSIPTaJkqN8F7XOQ5Q==",
+      "cpu": [
+        "loong64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-mips64el": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-mips64el/-/linux-mips64el-0.27.7.tgz",
+      "integrity": "sha512-KabT5I6StirGfIz0FMgl1I+R1H73Gp0ofL9A3nG3i/cYFJzKHhouBV5VWK1CSgKvVaG4q1RNpCTR2LuTVB3fIw==",
+      "cpu": [
+        "mips64el"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-ppc64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-ppc64/-/linux-ppc64-0.27.7.tgz",
+      "integrity": "sha512-gRsL4x6wsGHGRqhtI+ifpN/vpOFTQtnbsupUF5R5YTAg+y/lKelYR1hXbnBdzDjGbMYjVJLJTd2OFmMewAgwlQ==",
+      "cpu": [
+        "ppc64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-riscv64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-riscv64/-/linux-riscv64-0.27.7.tgz",
+      "integrity": "sha512-hL25LbxO1QOngGzu2U5xeXtxXcW+/GvMN3ejANqXkxZ/opySAZMrc+9LY/WyjAan41unrR3YrmtTsUpwT66InQ==",
+      "cpu": [
+        "riscv64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-s390x": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-s390x/-/linux-s390x-0.27.7.tgz",
+      "integrity": "sha512-2k8go8Ycu1Kb46vEelhu1vqEP+UeRVj2zY1pSuPdgvbd5ykAw82Lrro28vXUrRmzEsUV0NzCf54yARIK8r0fdw==",
+      "cpu": [
+        "s390x"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-x64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-x64/-/linux-x64-0.27.7.tgz",
+      "integrity": "sha512-hzznmADPt+OmsYzw1EE33ccA+HPdIqiCRq7cQeL1Jlq2gb1+OyWBkMCrYGBJ+sxVzve2ZJEVeePbLM2iEIZSxA==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/netbsd-arm64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/netbsd-arm64/-/netbsd-arm64-0.27.7.tgz",
+      "integrity": "sha512-b6pqtrQdigZBwZxAn1UpazEisvwaIDvdbMbmrly7cDTMFnw/+3lVxxCTGOrkPVnsYIosJJXAsILG9XcQS+Yu6w==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "netbsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/netbsd-x64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/netbsd-x64/-/netbsd-x64-0.27.7.tgz",
+      "integrity": "sha512-OfatkLojr6U+WN5EDYuoQhtM+1xco+/6FSzJJnuWiUw5eVcicbyK3dq5EeV/QHT1uy6GoDhGbFpprUiHUYggrw==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "netbsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/openbsd-arm64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/openbsd-arm64/-/openbsd-arm64-0.27.7.tgz",
+      "integrity": "sha512-AFuojMQTxAz75Fo8idVcqoQWEHIXFRbOc1TrVcFSgCZtQfSdc1RXgB3tjOn/krRHENUB4j00bfGjyl2mJrU37A==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "openbsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/openbsd-x64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/openbsd-x64/-/openbsd-x64-0.27.7.tgz",
+      "integrity": "sha512-+A1NJmfM8WNDv5CLVQYJ5PshuRm/4cI6WMZRg1by1GwPIQPCTs1GLEUHwiiQGT5zDdyLiRM/l1G0Pv54gvtKIg==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "openbsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/openharmony-arm64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/openharmony-arm64/-/openharmony-arm64-0.27.7.tgz",
+      "integrity": "sha512-+KrvYb/C8zA9CU/g0sR6w2RBw7IGc5J2BPnc3dYc5VJxHCSF1yNMxTV5LQ7GuKteQXZtspjFbiuW5/dOj7H4Yw==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "openharmony"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/sunos-x64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/sunos-x64/-/sunos-x64-0.27.7.tgz",
+      "integrity": "sha512-ikktIhFBzQNt/QDyOL580ti9+5mL/YZeUPKU2ivGtGjdTYoqz6jObj6nOMfhASpS4GU4Q/Clh1QtxWAvcYKamA==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "sunos"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/win32-arm64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/win32-arm64/-/win32-arm64-0.27.7.tgz",
+      "integrity": "sha512-7yRhbHvPqSpRUV7Q20VuDwbjW5kIMwTHpptuUzV+AA46kiPze5Z7qgt6CLCK3pWFrHeNfDd1VKgyP4O+ng17CA==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/win32-ia32": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/win32-ia32/-/win32-ia32-0.27.7.tgz",
+      "integrity": "sha512-SmwKXe6VHIyZYbBLJrhOoCJRB/Z1tckzmgTLfFYOfpMAx63BJEaL9ExI8x7v0oAO3Zh6D/Oi1gVxEYr5oUCFhw==",
+      "cpu": [
+        "ia32"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/win32-x64": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/@esbuild/win32-x64/-/win32-x64-0.27.7.tgz",
+      "integrity": "sha512-56hiAJPhwQ1R4i+21FVF7V8kSD5zZTdHcVuRFMW0hn753vVfQN8xlx4uOPT4xoGH0Z/oVATuR82AiqSTDIpaHg==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@jridgewell/gen-mapping": {
+      "version": "0.3.13",
+      "resolved": "https://registry.npmjs.org/@jridgewell/gen-mapping/-/gen-mapping-0.3.13.tgz",
+      "integrity": "sha512-2kkt/7niJ6MgEPxF0bYdQ6etZaA+fQvDcLKckhy1yIQOzaoKjBBjSj63/aLVjYE3qhRt5dvM+uUyfCg6UKCBbA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@jridgewell/sourcemap-codec": "^1.5.0",
+        "@jridgewell/trace-mapping": "^0.3.24"
+      }
+    },
+    "node_modules/@jridgewell/resolve-uri": {
+      "version": "3.1.2",
+      "resolved": "https://registry.npmjs.org/@jridgewell/resolve-uri/-/resolve-uri-3.1.2.tgz",
+      "integrity": "sha512-bRISgCIjP20/tbWSPWMEi54QVPRZExkuD9lJL+UIxUKtwVJA8wW1Trb1jMs1RFXo1CBTNZ/5hpC9QvmKWdopKw==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=6.0.0"
+      }
+    },
+    "node_modules/@jridgewell/sourcemap-codec": {
+      "version": "1.5.5",
+      "resolved": "https://registry.npmjs.org/@jridgewell/sourcemap-codec/-/sourcemap-codec-1.5.5.tgz",
+      "integrity": "sha512-cYQ9310grqxueWbl+WuIUIaiUaDcj7WOq5fVhEljNVgRfOUhY9fy2zTvfoqWsnebh8Sl70VScFbICvJnLKB0Og==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/@jridgewell/trace-mapping": {
+      "version": "0.3.31",
+      "resolved": "https://registry.npmjs.org/@jridgewell/trace-mapping/-/trace-mapping-0.3.31.tgz",
+      "integrity": "sha512-zzNR+SdQSDJzc8joaeP8QQoCQr8NuYx2dIIytl1QeBEZHJ9uW6hebsrYgbz8hJwUQao3TWCMtmfV8Nu1twOLAw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@jridgewell/resolve-uri": "^3.1.0",
+        "@jridgewell/sourcemap-codec": "^1.4.14"
+      }
+    },
+    "node_modules/@napi-rs/wasm-runtime": {
+      "version": "1.1.4",
+      "resolved": "https://registry.npmjs.org/@napi-rs/wasm-runtime/-/wasm-runtime-1.1.4.tgz",
+      "integrity": "sha512-3NQNNgA1YSlJb/kMH1ildASP9HW7/7kYnRI2szWJaofaS1hWmbGI4H+d3+22aGzXXN9IJ+n+GiFVcGipJP18ow==",
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "@tybys/wasm-util": "^0.10.1"
+      },
+      "funding": {
+        "type": "github",
+        "url": "https://github.com/sponsors/Brooooooklyn"
+      },
+      "peerDependencies": {
+        "@emnapi/core": "^1.7.1",
+        "@emnapi/runtime": "^1.7.1"
+      }
+    },
+    "node_modules/@oxc-project/types": {
+      "version": "0.132.0",
+      "resolved": "https://registry.npmjs.org/@oxc-project/types/-/types-0.132.0.tgz",
+      "integrity": "sha512-FESMOxil5Se014ui/Eq8fT5uHJo6nIRwH0PfJrZJXs6Gek3ZVFOrpUv3YIZT20m+extU98Hg1Ym72U58rlsxUQ==",
+      "dev": true,
+      "license": "MIT",
+      "funding": {
+        "url": "https://github.com/sponsors/Boshen"
+      }
+    },
+    "node_modules/@rolldown/binding-android-arm64": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-android-arm64/-/binding-android-arm64-1.0.2.tgz",
+      "integrity": "sha512-ZS4D1JPGn/MYQN/SYDWftIE/nVsM8j/AFOYEzAoOE2O3NktQOZru+/vYXGbR/qtdLdIfGCP0lcoJiYVzsEz+iQ==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "android"
+      ],
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      }
+    },
+    "node_modules/@rolldown/binding-darwin-arm64": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-darwin-arm64/-/binding-darwin-arm64-1.0.2.tgz",
+      "integrity": "sha512-vdFA9+C/rekyGce7WqHs/xoT0ioZEWaOFyZLIV1mEeNFaFDUQrPIo8Vs2GvJ6eetb3rzDUtUBgzto3ExpXJB3w==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "darwin"
+      ],
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      }
+    },
+    "node_modules/@rolldown/binding-darwin-x64": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-darwin-x64/-/binding-darwin-x64-1.0.2.tgz",
+      "integrity": "sha512-BewSOwTHazv77DTYiAZXSqqKZ4KP/KonFisDMVU7PImxoWfB2aepnPhd2E4SWz3zDzYgDNbs6jBmTdgNnF02GA==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "darwin"
+      ],
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      }
+    },
+    "node_modules/@rolldown/binding-freebsd-x64": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-freebsd-x64/-/binding-freebsd-x64-1.0.2.tgz",
+      "integrity": "sha512-m41o7M0YWtUdqk61Tb+jnKb2rN++iRdIASlExkUoKfIAH30DOHCB8fVLzSUpbWHHU8esmEioY62PxzexE8MBuA==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "freebsd"
+      ],
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      }
+    },
+    "node_modules/@rolldown/binding-linux-arm-gnueabihf": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-arm-gnueabihf/-/binding-linux-arm-gnueabihf-1.0.2.tgz",
+      "integrity": "sha512-jcojB9H7W/jS29pMKWAK1N+fU99vXodHDTatS3b3y/XSOCiHo0kkA74pL3jJmkoQtYpOCxDvaKs1fo2Ij/1X5w==",
+      "cpu": [
+        "arm"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      }
+    },
+    "node_modules/@rolldown/binding-linux-arm64-gnu": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-arm64-gnu/-/binding-linux-arm64-gnu-1.0.2.tgz",
+      "integrity": "sha512-1jn6qDU5iiOgFgygDzKUuKP0maTi0/f1+sBLgvij/76C77Nm3ts6ufz9Bjg5q5dduxiUIxtq86JIoBvo1xQ4Ig==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      }
+    },
+    "node_modules/@rolldown/binding-linux-arm64-musl": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-arm64-musl/-/binding-linux-arm64-musl-1.0.2.tgz",
+      "integrity": "sha512-QVLO/czFMdoMFSqlX3bcswcJNm/23r+qoa/jgtmFc/qEp6/jXmIkDjF/XIo8dPfGaiwy1xfQn8o77L79GeXFgw==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      }
+    },
+    "node_modules/@rolldown/binding-linux-ppc64-gnu": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-ppc64-gnu/-/binding-linux-ppc64-gnu-1.0.2.tgz",
+      "integrity": "sha512-hgO5Abm0w5UL6FEa2iFnZqo2KlK7TQ5QhV5x09hujBf7t5KzHQ1VmfPuTpqRy/rNlSxua3eWH374xxiVrP+lcA==",
+      "cpu": [
+        "ppc64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      }
+    },
+    "node_modules/@rolldown/binding-linux-s390x-gnu": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-s390x-gnu/-/binding-linux-s390x-gnu-1.0.2.tgz",
+      "integrity": "sha512-fy8rXxuYEu602abC8MUNaPjYLIFzReOaEIEMKMUa0rFEUxNpVXhs15KSSQ4qlqSaM7B6rcj9rDZgADh/IGDzLQ==",
+      "cpu": [
+        "s390x"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      }
+    },
+    "node_modules/@rolldown/binding-linux-x64-gnu": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-x64-gnu/-/binding-linux-x64-gnu-1.0.2.tgz",
+      "integrity": "sha512-0+bOkiQ779+r1WpoHOWHqncvyySci0vKph+myNDYb+im6meJAzHQXay6oEgnkHuUGouM1LKTZwqKpBow6Kj7CQ==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      }
+    },
+    "node_modules/@rolldown/binding-linux-x64-musl": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-linux-x64-musl/-/binding-linux-x64-musl-1.0.2.tgz",
+      "integrity": "sha512-mjSkrzZK5Qsl0a9d1JgILOiuZOSDTVdKENcSXBoqbzSrspLR/4/IRVDo5wd2GgZjNss/viBFJdeq+j7qH2nypw==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      }
+    },
+    "node_modules/@rolldown/binding-openharmony-arm64": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-openharmony-arm64/-/binding-openharmony-arm64-1.0.2.tgz",
+      "integrity": "sha512-1v5vHasdfQAZoEHakBV72LIFAC9JjnymsiKxp+GEr/ma3+NJCPSaYK+qavInOovJkgwFrs7GccX2d6IgDA3Z5w==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "openharmony"
+      ],
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      }
+    },
+    "node_modules/@rolldown/binding-wasm32-wasi": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-wasm32-wasi/-/binding-wasm32-wasi-1.0.2.tgz",
+      "integrity": "sha512-mb1VobWn6NheziTk5/WEaR6AKVbrwT5sOi6C7zk3gy/pD1qtJfU1j4PgTo2NJnOtbL9Dl3Aeei8w9jJ7qC2jZQ==",
+      "cpu": [
+        "wasm32"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "@emnapi/core": "1.10.0",
+        "@emnapi/runtime": "1.10.0",
+        "@napi-rs/wasm-runtime": "^1.1.4"
+      },
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      }
+    },
+    "node_modules/@rolldown/binding-win32-arm64-msvc": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-win32-arm64-msvc/-/binding-win32-arm64-msvc-1.0.2.tgz",
+      "integrity": "sha512-SqKonF56vA/L2yHwHYcEp2P34URpOZ7d1fS635cTkpDnUtEGdUbhI6NzsPdqeSWvAAeGDrxjWjNmibDIdFf9/A==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ],
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      }
+    },
+    "node_modules/@rolldown/binding-win32-x64-msvc": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@rolldown/binding-win32-x64-msvc/-/binding-win32-x64-msvc-1.0.2.tgz",
+      "integrity": "sha512-v7qRI7gXLRINcOGXt+7YmAZ6iFuyZVMIoXAxhd8oP+DR9dLfL9GfNIx7PLMxmhZdvq8waUJBQiWN9EKNy+TRBQ==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ],
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      }
+    },
+    "node_modules/@rolldown/pluginutils": {
+      "version": "1.0.1",
+      "resolved": "https://registry.npmjs.org/@rolldown/pluginutils/-/pluginutils-1.0.1.tgz",
+      "integrity": "sha512-2j9bGt5Jh8hj+vPtgzPtl72j0yRxHAyumoo6TNfAjsLB04UtpSvPbPcDcBMxz7n+9CYB0c1GxQFxYRg2jimqGw==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/@rollup/rollup-android-arm-eabi": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-android-arm-eabi/-/rollup-android-arm-eabi-4.60.4.tgz",
+      "integrity": "sha512-F5QXMSiFebS9hKZj02XhWLLnRpJ3B3AROP0tWbFBSj+6kCbg5m9j5JoHKd4mmSVy5mS/IMQloYgYxCuJC0fxEQ==",
+      "cpu": [
+        "arm"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "android"
+      ]
+    },
+    "node_modules/@rollup/rollup-android-arm64": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-android-arm64/-/rollup-android-arm64-4.60.4.tgz",
+      "integrity": "sha512-GxxTKApUpzRhof7poWvCJHRF51C67u1R7D6DiluBE8wKU1u5GWE8t+v81JvJYtbawoBFX1hLv5Ei4eVjkWokaw==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "android"
+      ]
+    },
+    "node_modules/@rollup/rollup-darwin-arm64": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-darwin-arm64/-/rollup-darwin-arm64-4.60.4.tgz",
+      "integrity": "sha512-tua0TaJxMOB1R0V0RS1jFZ/RpURFDJIOR2A6jWwQeawuFyS4gBW+rntLRaQd0EQ4bd6Vp44Z2rXW+YYDBsj6IA==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "darwin"
+      ]
+    },
+    "node_modules/@rollup/rollup-darwin-x64": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-darwin-x64/-/rollup-darwin-x64-4.60.4.tgz",
+      "integrity": "sha512-CSKq7MsP+5PFIcydhAiR1K0UhEI1A2jWXVKHPCBZ151yOutENwvnPocgVHkivu2kviURtCEB6zUQw0vs8RrhMg==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "darwin"
+      ]
+    },
+    "node_modules/@rollup/rollup-freebsd-arm64": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-freebsd-arm64/-/rollup-freebsd-arm64-4.60.4.tgz",
+      "integrity": "sha512-+O8OkVdyvXMtJEciu2wS/pzm1IxntEEQx3z5TAVy4l32G0etZn+RsA48ARRrFm6Ri8fvqPQfgrvNxSjKAbnd3g==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "freebsd"
+      ]
+    },
+    "node_modules/@rollup/rollup-freebsd-x64": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-freebsd-x64/-/rollup-freebsd-x64-4.60.4.tgz",
+      "integrity": "sha512-Iw3oMskH3AfNuhU0MSN7vNbdi4me/NiYo2azqPz/Le16zHSa+3RRmliCMWWQmh4lcndccU40xcJuTYJZxNo/lw==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "freebsd"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-arm-gnueabihf": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-arm-gnueabihf/-/rollup-linux-arm-gnueabihf-4.60.4.tgz",
+      "integrity": "sha512-EIPRXTVQpHyF8WOo219AD2yEltPehLTcTMz2fn6JsatLYSzQf00hj3rulF+yauOlF9/FtM2WpkT/hJh/KJFGhA==",
+      "cpu": [
+        "arm"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-arm-musleabihf": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-arm-musleabihf/-/rollup-linux-arm-musleabihf-4.60.4.tgz",
+      "integrity": "sha512-J3Yh9PzzF1Ovah2At+lHiGQdsYgArxBbXv/zHfSyaiFQEqvNv7DcW98pCrmdjCZBrqBiKrKKe2V+aaSGWuBe/w==",
+      "cpu": [
+        "arm"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-arm64-gnu": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-arm64-gnu/-/rollup-linux-arm64-gnu-4.60.4.tgz",
+      "integrity": "sha512-BFDEZMYfUvLn37ONE1yMBojPxnMlTFsdyNoqncT0qFq1mAfllL+ATMMJd8TeuVMiX84s1KbcxcZbXInmcO2mRg==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-arm64-musl": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-arm64-musl/-/rollup-linux-arm64-musl-4.60.4.tgz",
+      "integrity": "sha512-pc9EYOSlOgdQ2uPl1o9PF6/kLSgaUosia7gOuS8mB69IxJvlclko1MECXysjs5ryez1/5zjYqx3+xYU0TU6R1A==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-loong64-gnu": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-loong64-gnu/-/rollup-linux-loong64-gnu-4.60.4.tgz",
+      "integrity": "sha512-NxnomyxYerDh5n4iLrNa+sH+Z+U4BMEE46V2PgQ/hoB909i8gV1M5wPojWg9fk1jWpO3IQnOs20K4wyZuFLEFQ==",
+      "cpu": [
+        "loong64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-loong64-musl": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-loong64-musl/-/rollup-linux-loong64-musl-4.60.4.tgz",
+      "integrity": "sha512-nbJnQ8a3z1mtmrwImCYhc6BGpThAyYVRQxw9uKSKG4wR6aAYno9sVjJ0zaZcW9BPJX1GbrDPf+SvdWjgTuDmnw==",
+      "cpu": [
+        "loong64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-ppc64-gnu": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-ppc64-gnu/-/rollup-linux-ppc64-gnu-4.60.4.tgz",
+      "integrity": "sha512-2EU6acNrQLd8tYvo/LXW535wupT3m6fo7HKo6lr7ktQoItxTyOL1ZCR/GfGCuXl2vR+zmfI6eRXkSemafv+iVg==",
+      "cpu": [
+        "ppc64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-ppc64-musl": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-ppc64-musl/-/rollup-linux-ppc64-musl-4.60.4.tgz",
+      "integrity": "sha512-WeBtoMuaMxiiIrO2IYP3xs6GMWkJP2C0EoT8beTLkUPmzV1i/UcOSVw1d5r9KBODtHKilG5yFxsGRnBbK3wJ4A==",
+      "cpu": [
+        "ppc64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-riscv64-gnu": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-riscv64-gnu/-/rollup-linux-riscv64-gnu-4.60.4.tgz",
+      "integrity": "sha512-FJHFfqpKUI3A10WrWKiFbBZ7yVbGT4q4B5o1qKFFojqpaYoh9LrQgqWCmmcxQzVSXYtyB5bzkXrYzlHTs21MYA==",
+      "cpu": [
+        "riscv64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-riscv64-musl": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-riscv64-musl/-/rollup-linux-riscv64-musl-4.60.4.tgz",
+      "integrity": "sha512-mcEl6CUT5IAUmQf1m9FYSmVqCJlpQ8r8eyftFUHG8i9OhY7BkBXSUdnLH5DOf0wCOjcP9v/QO93zpmF1SptCCw==",
+      "cpu": [
+        "riscv64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-s390x-gnu": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-s390x-gnu/-/rollup-linux-s390x-gnu-4.60.4.tgz",
+      "integrity": "sha512-ynt3JxVd2w2buzoKDWIyiV1pJW93xlQic1THVLXilz429oijRpSHivZAgp65KBu+cMcgf1eVVjdnTLvPxgCuoQ==",
+      "cpu": [
+        "s390x"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-x64-gnu": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-x64-gnu/-/rollup-linux-x64-gnu-4.60.4.tgz",
+      "integrity": "sha512-Boiz5+MsaROEWDf+GGEwF8VMHGhlUoQMtIPjOgA5fv4osupqTVnJteQNKJwUcnUog2G55jYXH7KZFFiJe0TEzQ==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-x64-musl": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-x64-musl/-/rollup-linux-x64-musl-4.60.4.tgz",
+      "integrity": "sha512-+qfSY27qIrFfI/Hom04KYFw3GKZSGU4lXus51wsb5EuySfFlWRwjkKWoE9emgRw/ukoT4Udsj4W/+xxG8VbPKg==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-openbsd-x64": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-openbsd-x64/-/rollup-openbsd-x64-4.60.4.tgz",
+      "integrity": "sha512-VpTfOPHgVXEBeeR8hZ2O0F3aSso+JDWqTWmTmzcQKted54IAdUVbxE+j/MVxUsKa8L20HJhv3vUezVPoquqWjA==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "openbsd"
+      ]
+    },
+    "node_modules/@rollup/rollup-openharmony-arm64": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-openharmony-arm64/-/rollup-openharmony-arm64-4.60.4.tgz",
+      "integrity": "sha512-IPOsh5aRYuLv/nkU51X10Bf75Bsf6+gZdx1X+QP5QM6lIJFHHqbHLG0uJn/hWthzo13UAc2umiUorqZy3axoZg==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "openharmony"
+      ]
+    },
+    "node_modules/@rollup/rollup-win32-arm64-msvc": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-win32-arm64-msvc/-/rollup-win32-arm64-msvc-4.60.4.tgz",
+      "integrity": "sha512-4QzE9E81OohJ/HKzHhsqU+zcYYojVOXlFMs1DdyMT6qXl/niOH7AVElmmEdUNHHS/oRkc++d5k6Vy85zFs0DEw==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ]
+    },
+    "node_modules/@rollup/rollup-win32-ia32-msvc": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-win32-ia32-msvc/-/rollup-win32-ia32-msvc-4.60.4.tgz",
+      "integrity": "sha512-zTPgT1YuHHcd+Tmx7h8aml0FWFVelV5N54oHow9SLj+GfoDy/huQ+UV396N/C7KpMDMiPspRktzM1/0r1usYEA==",
+      "cpu": [
+        "ia32"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ]
+    },
+    "node_modules/@rollup/rollup-win32-x64-gnu": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-win32-x64-gnu/-/rollup-win32-x64-gnu-4.60.4.tgz",
+      "integrity": "sha512-DRS4G7mi9lJxqEDezIkKCaUIKCrLUUDCUaCsTPCi/rtqaC6D/jjwslMQyiDU50Ka0JKpeXeRBFBAXwArY52vBw==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ]
+    },
+    "node_modules/@rollup/rollup-win32-x64-msvc": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-win32-x64-msvc/-/rollup-win32-x64-msvc-4.60.4.tgz",
+      "integrity": "sha512-QVTUovf40zgTqlFVrKA1uXMVvU2QWEFWfAH8Wdc48IxLvrJMQVMBRjuQyUpzZCDkakImib9eVazbWlC6ksWtJw==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ]
+    },
+    "node_modules/@standard-schema/spec": {
+      "version": "1.1.0",
+      "resolved": "https://registry.npmjs.org/@standard-schema/spec/-/spec-1.1.0.tgz",
+      "integrity": "sha512-l2aFy5jALhniG5HgqrD6jXLi/rUWrKvqN/qJx6yoJsgKhblVd+iqqU4RCXavm/jPityDo5TCvKMnpjKnOriy0w==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/@tybys/wasm-util": {
+      "version": "0.10.2",
+      "resolved": "https://registry.npmjs.org/@tybys/wasm-util/-/wasm-util-0.10.2.tgz",
+      "integrity": "sha512-RoBvJ2X0wuKlWFIjrwffGw1IqZHKQqzIchKaadZZfnNpsAYp2mM0h36JtPCjNDAHGgYez/15uMBpfGwchhiMgg==",
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "tslib": "^2.4.0"
+      }
+    },
+    "node_modules/@types/chai": {
+      "version": "5.2.3",
+      "resolved": "https://registry.npmjs.org/@types/chai/-/chai-5.2.3.tgz",
+      "integrity": "sha512-Mw558oeA9fFbv65/y4mHtXDs9bPnFMZAL/jxdPFUpOHHIXX91mcgEHbS5Lahr+pwZFR8A7GQleRWeI6cGFC2UA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@types/deep-eql": "*",
+        "assertion-error": "^2.0.1"
+      }
+    },
+    "node_modules/@types/deep-eql": {
+      "version": "4.0.2",
+      "resolved": "https://registry.npmjs.org/@types/deep-eql/-/deep-eql-4.0.2.tgz",
+      "integrity": "sha512-c9h9dVVMigMPc4bwTvC5dxqtqJZwQPePsWjPlpSOnojbor6pGqdk541lfA7AqFQr5pB1BRdq0juY9db81BwyFw==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/@types/estree": {
+      "version": "1.0.8",
+      "resolved": "https://registry.npmjs.org/@types/estree/-/estree-1.0.8.tgz",
+      "integrity": "sha512-dWHzHa2WqEXI/O1E9OjrocMTKJl2mSrEolh1Iomrv6U+JuNwaHXsXx9bLu5gG7BUWFIN0skIQJQ/L1rIex4X6w==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/@types/node": {
+      "version": "25.9.1",
+      "resolved": "https://registry.npmjs.org/@types/node/-/node-25.9.1.tgz",
+      "integrity": "sha512-xfrlY7UD5rMJk3ZVJP8BNzS28J36YJg+xp+LPXV1TdWxr8uMH5A860QNxYDGQe/ylDSgjxE52Q9VnO7p75tJxg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "undici-types": ">=7.24.0 <7.24.7"
+      }
+    },
+    "node_modules/@vitest/expect": {
+      "version": "4.1.7",
+      "resolved": "https://registry.npmjs.org/@vitest/expect/-/expect-4.1.7.tgz",
+      "integrity": "sha512-1R+tw0ortHEbZDGMymm+pN7/AFQ/RkFFdtd7EN+VBpynKmLbP8A3rpEXdshBJ7+8hQ9zBJh/i1s0yKNtxAnU7w==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@standard-schema/spec": "^1.1.0",
+        "@types/chai": "^5.2.2",
+        "@vitest/spy": "4.1.7",
+        "@vitest/utils": "4.1.7",
+        "chai": "^6.2.2",
+        "tinyrainbow": "^3.1.0"
+      },
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      }
+    },
+    "node_modules/@vitest/mocker": {
+      "version": "4.1.7",
+      "resolved": "https://registry.npmjs.org/@vitest/mocker/-/mocker-4.1.7.tgz",
+      "integrity": "sha512-vY7nuamKgfvpA1Koa3oYIw/k7D6kZnpGyNMZW8loow2bsBYla1TFdqTaXncWdRn4pgwNs+90RhnXhJScDwQeJA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@vitest/spy": "4.1.7",
+        "estree-walker": "^3.0.3",
+        "magic-string": "^0.30.21"
+      },
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      },
+      "peerDependencies": {
+        "msw": "^2.4.9",
+        "vite": "^6.0.0 || ^7.0.0 || ^8.0.0"
+      },
+      "peerDependenciesMeta": {
+        "msw": {
+          "optional": true
+        },
+        "vite": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/@vitest/pretty-format": {
+      "version": "4.1.7",
+      "resolved": "https://registry.npmjs.org/@vitest/pretty-format/-/pretty-format-4.1.7.tgz",
+      "integrity": "sha512-umgCarTOYQWIaDMvGDRZij+6b9oVeLIyJzfN+AS88e0ZOU3QTgNNSTtjQOpcvWr3np1N0j4WgZj+sb3oYBDscw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "tinyrainbow": "^3.1.0"
+      },
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      }
+    },
+    "node_modules/@vitest/runner": {
+      "version": "4.1.7",
+      "resolved": "https://registry.npmjs.org/@vitest/runner/-/runner-4.1.7.tgz",
+      "integrity": "sha512-BapjmAQ2aI78WdMEfeUWivnfVzB+VPGwWRQcJE0OUq7qEeEcBsCSf+0T5iREBNE5nBb4wA5Ya0W6IA+sghdEFw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@vitest/utils": "4.1.7",
+        "pathe": "^2.0.3"
+      },
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      }
+    },
+    "node_modules/@vitest/snapshot": {
+      "version": "4.1.7",
+      "resolved": "https://registry.npmjs.org/@vitest/snapshot/-/snapshot-4.1.7.tgz",
+      "integrity": "sha512-ZacLzja+TmJeZ1h14xW2FB/WpeimUD3haBXQPyJqxvo8jQTmfeA8zv58mtjN2C7EHXZDYVcVYdYmAxjkWVvKCw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@vitest/pretty-format": "4.1.7",
+        "@vitest/utils": "4.1.7",
+        "magic-string": "^0.30.21",
+        "pathe": "^2.0.3"
+      },
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      }
+    },
+    "node_modules/@vitest/spy": {
+      "version": "4.1.7",
+      "resolved": "https://registry.npmjs.org/@vitest/spy/-/spy-4.1.7.tgz",
+      "integrity": "sha512-kbkI5LMWakyuTIvs6fUJ5qdIVb1XVKsYJAT4OJ938cHMROYMSfmoQdZy0aaAnjbbc8F61vkoTqz/Az+/HiIu5Q==",
+      "dev": true,
+      "license": "MIT",
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      }
+    },
+    "node_modules/@vitest/utils": {
+      "version": "4.1.7",
+      "resolved": "https://registry.npmjs.org/@vitest/utils/-/utils-4.1.7.tgz",
+      "integrity": "sha512-T532WBu791cBxJlCl6SO+J14l81DQx6uQHm1bQbmCDY7nqlEIgkza/UFnSBNaUtSf41unldDFjdOBYEQC4b5Hw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@vitest/pretty-format": "4.1.7",
+        "convert-source-map": "^2.0.0",
+        "tinyrainbow": "^3.1.0"
+      },
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      }
+    },
+    "node_modules/acorn": {
+      "version": "8.16.0",
+      "resolved": "https://registry.npmjs.org/acorn/-/acorn-8.16.0.tgz",
+      "integrity": "sha512-UVJyE9MttOsBQIDKw1skb9nAwQuR5wuGD3+82K6JgJlm/Y+KI92oNsMNGZCYdDsVtRHSak0pcV5Dno5+4jh9sw==",
+      "dev": true,
+      "license": "MIT",
+      "bin": {
+        "acorn": "bin/acorn"
+      },
+      "engines": {
+        "node": ">=0.4.0"
+      }
+    },
+    "node_modules/any-promise": {
+      "version": "1.3.0",
+      "resolved": "https://registry.npmjs.org/any-promise/-/any-promise-1.3.0.tgz",
+      "integrity": "sha512-7UvmKalWRt1wgjL1RrGxoSJW/0QZFIegpeGvZG9kjp8vrRu55XTHbwnqq2GpXm9uLbcuhxm3IqX9OB4MZR1b2A==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/assertion-error": {
+      "version": "2.0.1",
+      "resolved": "https://registry.npmjs.org/assertion-error/-/assertion-error-2.0.1.tgz",
+      "integrity": "sha512-Izi8RQcffqCeNVgFigKli1ssklIbpHnCYc6AknXGYoB6grJqyeby7jv12JUQgmTAnIDnbck1uxksT4dzN3PWBA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=12"
+      }
+    },
+    "node_modules/bundle-require": {
+      "version": "5.1.0",
+      "resolved": "https://registry.npmjs.org/bundle-require/-/bundle-require-5.1.0.tgz",
+      "integrity": "sha512-3WrrOuZiyaaZPWiEt4G3+IffISVC9HYlWueJEBWED4ZH4aIAC2PnkdnuRrR94M+w6yGWn4AglWtJtBI8YqvgoA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "load-tsconfig": "^0.2.3"
+      },
+      "engines": {
+        "node": "^12.20.0 || ^14.13.1 || >=16.0.0"
+      },
+      "peerDependencies": {
+        "esbuild": ">=0.18"
+      }
+    },
+    "node_modules/cac": {
+      "version": "6.7.14",
+      "resolved": "https://registry.npmjs.org/cac/-/cac-6.7.14.tgz",
+      "integrity": "sha512-b6Ilus+c3RrdDk+JhLKUAQfzzgLEPy6wcXqS7f/xe1EETvsDP6GORG7SFuOs6cID5YkqchW/LXZbX5bc8j7ZcQ==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/chai": {
+      "version": "6.2.2",
+      "resolved": "https://registry.npmjs.org/chai/-/chai-6.2.2.tgz",
+      "integrity": "sha512-NUPRluOfOiTKBKvWPtSD4PhFvWCqOi0BGStNWs57X9js7XGTprSmFoz5F0tWhR4WPjNeR9jXqdC7/UpSJTnlRg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/chokidar": {
+      "version": "4.0.3",
+      "resolved": "https://registry.npmjs.org/chokidar/-/chokidar-4.0.3.tgz",
+      "integrity": "sha512-Qgzu8kfBvo+cA4962jnP1KkS6Dop5NS6g7R5LFYJr4b8Ub94PPQXUksCw9PvXoeXPRRddRNC5C1JQUR2SMGtnA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "readdirp": "^4.0.1"
+      },
+      "engines": {
+        "node": ">= 14.16.0"
+      },
+      "funding": {
+        "url": "https://paulmillr.com/funding/"
+      }
+    },
+    "node_modules/commander": {
+      "version": "4.1.1",
+      "resolved": "https://registry.npmjs.org/commander/-/commander-4.1.1.tgz",
+      "integrity": "sha512-NOKm8xhkzAjzFx8B2v5OAHT+u5pRQc2UCa2Vq9jYL/31o2wi9mxBA7LIFs3sV5VSC49z6pEhfbMULvShKj26WA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">= 6"
+      }
+    },
+    "node_modules/confbox": {
+      "version": "0.1.8",
+      "resolved": "https://registry.npmjs.org/confbox/-/confbox-0.1.8.tgz",
+      "integrity": "sha512-RMtmw0iFkeR4YV+fUOSucriAQNb9g8zFR52MWCtl+cCZOFRNL6zeB395vPzFhEjjn4fMxXudmELnl/KF/WrK6w==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/consola": {
+      "version": "3.4.2",
+      "resolved": "https://registry.npmjs.org/consola/-/consola-3.4.2.tgz",
+      "integrity": "sha512-5IKcdX0nnYavi6G7TtOhwkYzyjfJlatbjMjuLSfE2kYT5pMDOilZ4OvMhi637CcDICTmz3wARPoyhqyX1Y+XvA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": "^14.18.0 || >=16.10.0"
+      }
+    },
+    "node_modules/convert-source-map": {
+      "version": "2.0.0",
+      "resolved": "https://registry.npmjs.org/convert-source-map/-/convert-source-map-2.0.0.tgz",
+      "integrity": "sha512-Kvp459HrV2FEJ1CAsi1Ku+MY3kasH19TFykTz2xWmMeq6bk2NU3XXvfJ+Q61m0xktWwt+1HSYf3JZsTms3aRJg==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/debug": {
+      "version": "4.4.3",
+      "resolved": "https://registry.npmjs.org/debug/-/debug-4.4.3.tgz",
+      "integrity": "sha512-RGwwWnwQvkVfavKVt22FGLw+xYSdzARwm0ru6DhTVA3umU5hZc28V3kO4stgYryrTlLpuvgI9GiijltAjNbcqA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "ms": "^2.1.3"
+      },
+      "engines": {
+        "node": ">=6.0"
+      },
+      "peerDependenciesMeta": {
+        "supports-color": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/detect-libc": {
+      "version": "2.1.2",
+      "resolved": "https://registry.npmjs.org/detect-libc/-/detect-libc-2.1.2.tgz",
+      "integrity": "sha512-Btj2BOOO83o3WyH59e8MgXsxEQVcarkUOpEYrubB0urwnN10yQ364rsiByU11nZlqWYZm05i/of7io4mzihBtQ==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/es-module-lexer": {
+      "version": "2.1.0",
+      "resolved": "https://registry.npmjs.org/es-module-lexer/-/es-module-lexer-2.1.0.tgz",
+      "integrity": "sha512-n27zTYMjYu1aj4MjCWzSP7G9r75utsaoc8m61weK+W8JMBGGQybd43GstCXZ3WNmSFtGT9wi59qQTW6mhTR5LQ==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/esbuild": {
+      "version": "0.27.7",
+      "resolved": "https://registry.npmjs.org/esbuild/-/esbuild-0.27.7.tgz",
+      "integrity": "sha512-IxpibTjyVnmrIQo5aqNpCgoACA/dTKLTlhMHihVHhdkxKyPO1uBBthumT0rdHmcsk9uMonIWS0m4FljWzILh3w==",
+      "dev": true,
+      "hasInstallScript": true,
+      "license": "MIT",
+      "bin": {
+        "esbuild": "bin/esbuild"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "optionalDependencies": {
+        "@esbuild/aix-ppc64": "0.27.7",
+        "@esbuild/android-arm": "0.27.7",
+        "@esbuild/android-arm64": "0.27.7",
+        "@esbuild/android-x64": "0.27.7",
+        "@esbuild/darwin-arm64": "0.27.7",
+        "@esbuild/darwin-x64": "0.27.7",
+        "@esbuild/freebsd-arm64": "0.27.7",
+        "@esbuild/freebsd-x64": "0.27.7",
+        "@esbuild/linux-arm": "0.27.7",
+        "@esbuild/linux-arm64": "0.27.7",
+        "@esbuild/linux-ia32": "0.27.7",
+        "@esbuild/linux-loong64": "0.27.7",
+        "@esbuild/linux-mips64el": "0.27.7",
+        "@esbuild/linux-ppc64": "0.27.7",
+        "@esbuild/linux-riscv64": "0.27.7",
+        "@esbuild/linux-s390x": "0.27.7",
+        "@esbuild/linux-x64": "0.27.7",
+        "@esbuild/netbsd-arm64": "0.27.7",
+        "@esbuild/netbsd-x64": "0.27.7",
+        "@esbuild/openbsd-arm64": "0.27.7",
+        "@esbuild/openbsd-x64": "0.27.7",
+        "@esbuild/openharmony-arm64": "0.27.7",
+        "@esbuild/sunos-x64": "0.27.7",
+        "@esbuild/win32-arm64": "0.27.7",
+        "@esbuild/win32-ia32": "0.27.7",
+        "@esbuild/win32-x64": "0.27.7"
+      }
+    },
+    "node_modules/estree-walker": {
+      "version": "3.0.3",
+      "resolved": "https://registry.npmjs.org/estree-walker/-/estree-walker-3.0.3.tgz",
+      "integrity": "sha512-7RUKfXgSMMkzt6ZuXmqapOurLGPPfgj6l9uRZ7lRGolvk0y2yocc35LdcxKC5PQZdn2DMqioAQ2NoWcrTKmm6g==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@types/estree": "^1.0.0"
+      }
+    },
+    "node_modules/expect-type": {
+      "version": "1.3.0",
+      "resolved": "https://registry.npmjs.org/expect-type/-/expect-type-1.3.0.tgz",
+      "integrity": "sha512-knvyeauYhqjOYvQ66MznSMs83wmHrCycNEN6Ao+2AeYEfxUIkuiVxdEa1qlGEPK+We3n0THiDciYSsCcgW/DoA==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "engines": {
+        "node": ">=12.0.0"
+      }
+    },
+    "node_modules/fdir": {
+      "version": "6.5.0",
+      "resolved": "https://registry.npmjs.org/fdir/-/fdir-6.5.0.tgz",
+      "integrity": "sha512-tIbYtZbucOs0BRGqPJkshJUYdL+SDH7dVM8gjy+ERp3WAUjLEFJE+02kanyHtwjWOnwrKYBiwAmM0p4kLJAnXg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=12.0.0"
+      },
+      "peerDependencies": {
+        "picomatch": "^3 || ^4"
+      },
+      "peerDependenciesMeta": {
+        "picomatch": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/fix-dts-default-cjs-exports": {
+      "version": "1.0.1",
+      "resolved": "https://registry.npmjs.org/fix-dts-default-cjs-exports/-/fix-dts-default-cjs-exports-1.0.1.tgz",
+      "integrity": "sha512-pVIECanWFC61Hzl2+oOCtoJ3F17kglZC/6N94eRWycFgBH35hHx0Li604ZIzhseh97mf2p0cv7vVrOZGoqhlEg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "magic-string": "^0.30.17",
+        "mlly": "^1.7.4",
+        "rollup": "^4.34.8"
+      }
+    },
+    "node_modules/fsevents": {
+      "version": "2.3.3",
+      "resolved": "https://registry.npmjs.org/fsevents/-/fsevents-2.3.3.tgz",
+      "integrity": "sha512-5xoDfX+fL7faATnagmWPpbFtwh/R77WmMMqqHGS65C3vvB0YHrgF+B1YmZ3441tMj5n63k0212XNoJwzlhffQw==",
+      "dev": true,
+      "hasInstallScript": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "darwin"
+      ],
+      "engines": {
+        "node": "^8.16.0 || ^10.6.0 || >=11.0.0"
+      }
+    },
+    "node_modules/joycon": {
+      "version": "3.1.1",
+      "resolved": "https://registry.npmjs.org/joycon/-/joycon-3.1.1.tgz",
+      "integrity": "sha512-34wB/Y7MW7bzjKRjUKTa46I2Z7eV62Rkhva+KkopW7Qvv/OSWBqvkSY7vusOPrNuZcUG3tApvdVgNB8POj3SPw==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=10"
+      }
+    },
+    "node_modules/lightningcss": {
+      "version": "1.32.0",
+      "resolved": "https://registry.npmjs.org/lightningcss/-/lightningcss-1.32.0.tgz",
+      "integrity": "sha512-NXYBzinNrblfraPGyrbPoD19C1h9lfI/1mzgWYvXUTe414Gz/X1FD2XBZSZM7rRTrMA8JL3OtAaGifrIKhQ5yQ==",
+      "dev": true,
+      "license": "MPL-2.0",
+      "dependencies": {
+        "detect-libc": "^2.0.3"
+      },
+      "engines": {
+        "node": ">= 12.0.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/parcel"
+      },
+      "optionalDependencies": {
+        "lightningcss-android-arm64": "1.32.0",
+        "lightningcss-darwin-arm64": "1.32.0",
+        "lightningcss-darwin-x64": "1.32.0",
+        "lightningcss-freebsd-x64": "1.32.0",
+        "lightningcss-linux-arm-gnueabihf": "1.32.0",
+        "lightningcss-linux-arm64-gnu": "1.32.0",
+        "lightningcss-linux-arm64-musl": "1.32.0",
+        "lightningcss-linux-x64-gnu": "1.32.0",
+        "lightningcss-linux-x64-musl": "1.32.0",
+        "lightningcss-win32-arm64-msvc": "1.32.0",
+        "lightningcss-win32-x64-msvc": "1.32.0"
+      }
+    },
+    "node_modules/lightningcss-android-arm64": {
+      "version": "1.32.0",
+      "resolved": "https://registry.npmjs.org/lightningcss-android-arm64/-/lightningcss-android-arm64-1.32.0.tgz",
+      "integrity": "sha512-YK7/ClTt4kAK0vo6w3X+Pnm0D2cf2vPHbhOXdoNti1Ga0al1P4TBZhwjATvjNwLEBCnKvjJc2jQgHXH0NEwlAg==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MPL-2.0",
+      "optional": true,
+      "os": [
+        "android"
+      ],
+      "engines": {
+        "node": ">= 12.0.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/parcel"
+      }
+    },
+    "node_modules/lightningcss-darwin-arm64": {
+      "version": "1.32.0",
+      "resolved": "https://registry.npmjs.org/lightningcss-darwin-arm64/-/lightningcss-darwin-arm64-1.32.0.tgz",
+      "integrity": "sha512-RzeG9Ju5bag2Bv1/lwlVJvBE3q6TtXskdZLLCyfg5pt+HLz9BqlICO7LZM7VHNTTn/5PRhHFBSjk5lc4cmscPQ==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MPL-2.0",
+      "optional": true,
+      "os": [
+        "darwin"
+      ],
+      "engines": {
+        "node": ">= 12.0.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/parcel"
+      }
+    },
+    "node_modules/lightningcss-darwin-x64": {
+      "version": "1.32.0",
+      "resolved": "https://registry.npmjs.org/lightningcss-darwin-x64/-/lightningcss-darwin-x64-1.32.0.tgz",
+      "integrity": "sha512-U+QsBp2m/s2wqpUYT/6wnlagdZbtZdndSmut/NJqlCcMLTWp5muCrID+K5UJ6jqD2BFshejCYXniPDbNh73V8w==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MPL-2.0",
+      "optional": true,
+      "os": [
+        "darwin"
+      ],
+      "engines": {
+        "node": ">= 12.0.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/parcel"
+      }
+    },
+    "node_modules/lightningcss-freebsd-x64": {
+      "version": "1.32.0",
+      "resolved": "https://registry.npmjs.org/lightningcss-freebsd-x64/-/lightningcss-freebsd-x64-1.32.0.tgz",
+      "integrity": "sha512-JCTigedEksZk3tHTTthnMdVfGf61Fky8Ji2E4YjUTEQX14xiy/lTzXnu1vwiZe3bYe0q+SpsSH/CTeDXK6WHig==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MPL-2.0",
+      "optional": true,
+      "os": [
+        "freebsd"
+      ],
+      "engines": {
+        "node": ">= 12.0.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/parcel"
+      }
+    },
+    "node_modules/lightningcss-linux-arm-gnueabihf": {
+      "version": "1.32.0",
+      "resolved": "https://registry.npmjs.org/lightningcss-linux-arm-gnueabihf/-/lightningcss-linux-arm-gnueabihf-1.32.0.tgz",
+      "integrity": "sha512-x6rnnpRa2GL0zQOkt6rts3YDPzduLpWvwAF6EMhXFVZXD4tPrBkEFqzGowzCsIWsPjqSK+tyNEODUBXeeVHSkw==",
+      "cpu": [
+        "arm"
+      ],
+      "dev": true,
+      "license": "MPL-2.0",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">= 12.0.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/parcel"
+      }
+    },
+    "node_modules/lightningcss-linux-arm64-gnu": {
+      "version": "1.32.0",
+      "resolved": "https://registry.npmjs.org/lightningcss-linux-arm64-gnu/-/lightningcss-linux-arm64-gnu-1.32.0.tgz",
+      "integrity": "sha512-0nnMyoyOLRJXfbMOilaSRcLH3Jw5z9HDNGfT/gwCPgaDjnx0i8w7vBzFLFR1f6CMLKF8gVbebmkUN3fa/kQJpQ==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MPL-2.0",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">= 12.0.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/parcel"
+      }
+    },
+    "node_modules/lightningcss-linux-arm64-musl": {
+      "version": "1.32.0",
+      "resolved": "https://registry.npmjs.org/lightningcss-linux-arm64-musl/-/lightningcss-linux-arm64-musl-1.32.0.tgz",
+      "integrity": "sha512-UpQkoenr4UJEzgVIYpI80lDFvRmPVg6oqboNHfoH4CQIfNA+HOrZ7Mo7KZP02dC6LjghPQJeBsvXhJod/wnIBg==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MPL-2.0",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">= 12.0.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/parcel"
+      }
+    },
+    "node_modules/lightningcss-linux-x64-gnu": {
+      "version": "1.32.0",
+      "resolved": "https://registry.npmjs.org/lightningcss-linux-x64-gnu/-/lightningcss-linux-x64-gnu-1.32.0.tgz",
+      "integrity": "sha512-V7Qr52IhZmdKPVr+Vtw8o+WLsQJYCTd8loIfpDaMRWGUZfBOYEJeyJIkqGIDMZPwPx24pUMfwSxxI8phr/MbOA==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MPL-2.0",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">= 12.0.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/parcel"
+      }
+    },
+    "node_modules/lightningcss-linux-x64-musl": {
+      "version": "1.32.0",
+      "resolved": "https://registry.npmjs.org/lightningcss-linux-x64-musl/-/lightningcss-linux-x64-musl-1.32.0.tgz",
+      "integrity": "sha512-bYcLp+Vb0awsiXg/80uCRezCYHNg1/l3mt0gzHnWV9XP1W5sKa5/TCdGWaR/zBM2PeF/HbsQv/j2URNOiVuxWg==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MPL-2.0",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">= 12.0.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/parcel"
+      }
+    },
+    "node_modules/lightningcss-win32-arm64-msvc": {
+      "version": "1.32.0",
+      "resolved": "https://registry.npmjs.org/lightningcss-win32-arm64-msvc/-/lightningcss-win32-arm64-msvc-1.32.0.tgz",
+      "integrity": "sha512-8SbC8BR40pS6baCM8sbtYDSwEVQd4JlFTOlaD3gWGHfThTcABnNDBda6eTZeqbofalIJhFx0qKzgHJmcPTnGdw==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MPL-2.0",
+      "optional": true,
+      "os": [
+        "win32"
+      ],
+      "engines": {
+        "node": ">= 12.0.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/parcel"
+      }
+    },
+    "node_modules/lightningcss-win32-x64-msvc": {
+      "version": "1.32.0",
+      "resolved": "https://registry.npmjs.org/lightningcss-win32-x64-msvc/-/lightningcss-win32-x64-msvc-1.32.0.tgz",
+      "integrity": "sha512-Amq9B/SoZYdDi1kFrojnoqPLxYhQ4Wo5XiL8EVJrVsB8ARoC1PWW6VGtT0WKCemjy8aC+louJnjS7U18x3b06Q==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MPL-2.0",
+      "optional": true,
+      "os": [
+        "win32"
+      ],
+      "engines": {
+        "node": ">= 12.0.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/parcel"
+      }
+    },
+    "node_modules/lilconfig": {
+      "version": "3.1.3",
+      "resolved": "https://registry.npmjs.org/lilconfig/-/lilconfig-3.1.3.tgz",
+      "integrity": "sha512-/vlFKAoH5Cgt3Ie+JLhRbwOsCQePABiU3tJ1egGvyQ+33R/vcwM2Zl2QR/LzjsBeItPt3oSVXapn+m4nQDvpzw==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=14"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/antonk52"
+      }
+    },
+    "node_modules/lines-and-columns": {
+      "version": "1.2.4",
+      "resolved": "https://registry.npmjs.org/lines-and-columns/-/lines-and-columns-1.2.4.tgz",
+      "integrity": "sha512-7ylylesZQ/PV29jhEDl3Ufjo6ZX7gCqJr5F7PKrqc93v7fzSymt1BpwEU8nAUXs8qzzvqhbjhK5QZg6Mt/HkBg==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/load-tsconfig": {
+      "version": "0.2.5",
+      "resolved": "https://registry.npmjs.org/load-tsconfig/-/load-tsconfig-0.2.5.tgz",
+      "integrity": "sha512-IXO6OCs9yg8tMKzfPZ1YmheJbZCiEsnBdcB03l0OcfK9prKnJb96siuHCr5Fl37/yo9DnKU+TLpxzTUspw9shg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": "^12.20.0 || ^14.13.1 || >=16.0.0"
+      }
+    },
+    "node_modules/magic-string": {
+      "version": "0.30.21",
+      "resolved": "https://registry.npmjs.org/magic-string/-/magic-string-0.30.21.tgz",
+      "integrity": "sha512-vd2F4YUyEXKGcLHoq+TEyCjxueSeHnFxyyjNp80yg0XV4vUhnDer/lvvlqM/arB5bXQN5K2/3oinyCRyx8T2CQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@jridgewell/sourcemap-codec": "^1.5.5"
+      }
+    },
+    "node_modules/mlly": {
+      "version": "1.8.2",
+      "resolved": "https://registry.npmjs.org/mlly/-/mlly-1.8.2.tgz",
+      "integrity": "sha512-d+ObxMQFmbt10sretNDytwt85VrbkhhUA/JBGm1MPaWJ65Cl4wOgLaB1NYvJSZ0Ef03MMEU/0xpPMXUIQ29UfA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "acorn": "^8.16.0",
+        "pathe": "^2.0.3",
+        "pkg-types": "^1.3.1",
+        "ufo": "^1.6.3"
+      }
+    },
+    "node_modules/ms": {
+      "version": "2.1.3",
+      "resolved": "https://registry.npmjs.org/ms/-/ms-2.1.3.tgz",
+      "integrity": "sha512-6FlzubTLZG3J2a/NVCAleEhjzq5oxgHyaCU9yYXvcLsvoVaHJq/s5xXI6/XXP6tz7R9xAOtHnSO/tXtF3WRTlA==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/mz": {
+      "version": "2.7.0",
+      "resolved": "https://registry.npmjs.org/mz/-/mz-2.7.0.tgz",
+      "integrity": "sha512-z81GNO7nnYMEhrGh9LeymoE4+Yr0Wn5McHIZMK5cfQCl+NDX08sCZgUc9/6MHni9IWuFLm1Z3HTCXu2z9fN62Q==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "any-promise": "^1.0.0",
+        "object-assign": "^4.0.1",
+        "thenify-all": "^1.0.0"
+      }
+    },
+    "node_modules/nanoid": {
+      "version": "3.3.12",
+      "resolved": "https://registry.npmjs.org/nanoid/-/nanoid-3.3.12.tgz",
+      "integrity": "sha512-ZB9RH/39qpq5Vu6Y+NmUaFhQR6pp+M2Xt76XBnEwDaGcVAqhlvxrl3B2bKS5D3NH3QR76v3aSrKaF/Kiy7lEtQ==",
+      "dev": true,
+      "funding": [
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/ai"
+        }
+      ],
+      "license": "MIT",
+      "bin": {
+        "nanoid": "bin/nanoid.cjs"
+      },
+      "engines": {
+        "node": "^10 || ^12 || ^13.7 || ^14 || >=15.0.1"
+      }
+    },
+    "node_modules/object-assign": {
+      "version": "4.1.1",
+      "resolved": "https://registry.npmjs.org/object-assign/-/object-assign-4.1.1.tgz",
+      "integrity": "sha512-rJgTQnkUnH1sFw8yT6VSU3zD3sWmu6sZhIseY8VX+GRu3P6F7Fu+JNDoXfklElbLJSnc3FUQHVe4cU5hj+BcUg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=0.10.0"
+      }
+    },
+    "node_modules/obug": {
+      "version": "2.1.1",
+      "resolved": "https://registry.npmjs.org/obug/-/obug-2.1.1.tgz",
+      "integrity": "sha512-uTqF9MuPraAQ+IsnPf366RG4cP9RtUi7MLO1N3KEc+wb0a6yKpeL0lmk2IB1jY5KHPAlTc6T/JRdC/YqxHNwkQ==",
+      "dev": true,
+      "funding": [
+        "https://github.com/sponsors/sxzz",
+        "https://opencollective.com/debug"
+      ],
+      "license": "MIT"
+    },
+    "node_modules/pathe": {
+      "version": "2.0.3",
+      "resolved": "https://registry.npmjs.org/pathe/-/pathe-2.0.3.tgz",
+      "integrity": "sha512-WUjGcAqP1gQacoQe+OBJsFA7Ld4DyXuUIjZ5cc75cLHvJ7dtNsTugphxIADwspS+AraAUePCKrSVtPLFj/F88w==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/picocolors": {
+      "version": "1.1.1",
+      "resolved": "https://registry.npmjs.org/picocolors/-/picocolors-1.1.1.tgz",
+      "integrity": "sha512-xceH2snhtb5M9liqDsmEw56le376mTZkEX/jEb/RxNFyegNul7eNslCXP9FDj/Lcu0X8KEyMceP2ntpaHrDEVA==",
+      "dev": true,
+      "license": "ISC"
+    },
+    "node_modules/picomatch": {
+      "version": "4.0.4",
+      "resolved": "https://registry.npmjs.org/picomatch/-/picomatch-4.0.4.tgz",
+      "integrity": "sha512-QP88BAKvMam/3NxH6vj2o21R6MjxZUAd6nlwAS/pnGvN9IVLocLHxGYIzFhg6fUQ+5th6P4dv4eW9jX3DSIj7A==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=12"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/jonschlinkert"
+      }
+    },
+    "node_modules/pirates": {
+      "version": "4.0.7",
+      "resolved": "https://registry.npmjs.org/pirates/-/pirates-4.0.7.tgz",
+      "integrity": "sha512-TfySrs/5nm8fQJDcBDuUng3VOUKsd7S+zqvbOTiGXHfxX4wK31ard+hoNuvkicM/2YFzlpDgABOevKSsB4G/FA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">= 6"
+      }
+    },
+    "node_modules/pkg-types": {
+      "version": "1.3.1",
+      "resolved": "https://registry.npmjs.org/pkg-types/-/pkg-types-1.3.1.tgz",
+      "integrity": "sha512-/Jm5M4RvtBFVkKWRu2BLUTNP8/M2a+UwuAX+ae4770q1qVGtfjG+WTCupoZixokjmHiry8uI+dlY8KXYV5HVVQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "confbox": "^0.1.8",
+        "mlly": "^1.7.4",
+        "pathe": "^2.0.1"
+      }
+    },
+    "node_modules/postcss": {
+      "version": "8.5.15",
+      "resolved": "https://registry.npmjs.org/postcss/-/postcss-8.5.15.tgz",
+      "integrity": "sha512-FfR8sjd4em2T6fb3I2MwAJU7HWVMr9zba+enmQeeWFfCbm+UOC/0X4DS8XtpUTMwWMGbjKYP7xjfNekzyGmB3A==",
+      "dev": true,
+      "funding": [
+        {
+          "type": "opencollective",
+          "url": "https://opencollective.com/postcss/"
+        },
+        {
+          "type": "tidelift",
+          "url": "https://tidelift.com/funding/github/npm/postcss"
+        },
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/ai"
+        }
+      ],
+      "license": "MIT",
+      "dependencies": {
+        "nanoid": "^3.3.12",
+        "picocolors": "^1.1.1",
+        "source-map-js": "^1.2.1"
+      },
+      "engines": {
+        "node": "^10 || ^12 || >=14"
+      }
+    },
+    "node_modules/postcss-load-config": {
+      "version": "6.0.1",
+      "resolved": "https://registry.npmjs.org/postcss-load-config/-/postcss-load-config-6.0.1.tgz",
+      "integrity": "sha512-oPtTM4oerL+UXmx+93ytZVN82RrlY/wPUV8IeDxFrzIjXOLF1pN+EmKPLbubvKHT2HC20xXsCAH2Z+CKV6Oz/g==",
+      "dev": true,
+      "funding": [
+        {
+          "type": "opencollective",
+          "url": "https://opencollective.com/postcss/"
+        },
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/ai"
+        }
+      ],
+      "license": "MIT",
+      "dependencies": {
+        "lilconfig": "^3.1.1"
+      },
+      "engines": {
+        "node": ">= 18"
+      },
+      "peerDependencies": {
+        "jiti": ">=1.21.0",
+        "postcss": ">=8.0.9",
+        "tsx": "^4.8.1",
+        "yaml": "^2.4.2"
+      },
+      "peerDependenciesMeta": {
+        "jiti": {
+          "optional": true
+        },
+        "postcss": {
+          "optional": true
+        },
+        "tsx": {
+          "optional": true
+        },
+        "yaml": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/readdirp": {
+      "version": "4.1.2",
+      "resolved": "https://registry.npmjs.org/readdirp/-/readdirp-4.1.2.tgz",
+      "integrity": "sha512-GDhwkLfywWL2s6vEjyhri+eXmfH6j1L7JE27WhqLeYzoh/A3DBaYGEj2H/HFZCn/kMfim73FXxEJTw06WtxQwg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">= 14.18.0"
+      },
+      "funding": {
+        "type": "individual",
+        "url": "https://paulmillr.com/funding/"
+      }
+    },
+    "node_modules/resolve-from": {
+      "version": "5.0.0",
+      "resolved": "https://registry.npmjs.org/resolve-from/-/resolve-from-5.0.0.tgz",
+      "integrity": "sha512-qYg9KP24dD5qka9J47d0aVky0N+b4fTU89LN9iDnjB5waksiC49rvMB0PrUJQGoTmH50XPiqOvAjDfaijGxYZw==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/rolldown": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/rolldown/-/rolldown-1.0.2.tgz",
+      "integrity": "sha512-oZx5zVDtVB44AW3eaifgDml1gWRDZGvjcfdxonE4swNPG98PrrXjaO/KrnUjzlMnztCCRVlUueA1kCXhARGk6g==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@oxc-project/types": "=0.132.0",
+        "@rolldown/pluginutils": "^1.0.0"
+      },
+      "bin": {
+        "rolldown": "bin/cli.mjs"
+      },
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      },
+      "optionalDependencies": {
+        "@rolldown/binding-android-arm64": "1.0.2",
+        "@rolldown/binding-darwin-arm64": "1.0.2",
+        "@rolldown/binding-darwin-x64": "1.0.2",
+        "@rolldown/binding-freebsd-x64": "1.0.2",
+        "@rolldown/binding-linux-arm-gnueabihf": "1.0.2",
+        "@rolldown/binding-linux-arm64-gnu": "1.0.2",
+        "@rolldown/binding-linux-arm64-musl": "1.0.2",
+        "@rolldown/binding-linux-ppc64-gnu": "1.0.2",
+        "@rolldown/binding-linux-s390x-gnu": "1.0.2",
+        "@rolldown/binding-linux-x64-gnu": "1.0.2",
+        "@rolldown/binding-linux-x64-musl": "1.0.2",
+        "@rolldown/binding-openharmony-arm64": "1.0.2",
+        "@rolldown/binding-wasm32-wasi": "1.0.2",
+        "@rolldown/binding-win32-arm64-msvc": "1.0.2",
+        "@rolldown/binding-win32-x64-msvc": "1.0.2"
+      }
+    },
+    "node_modules/rollup": {
+      "version": "4.60.4",
+      "resolved": "https://registry.npmjs.org/rollup/-/rollup-4.60.4.tgz",
+      "integrity": "sha512-WHeFSbZYsPu3+bLoNRUuAO+wavNlocOPf3wSHTP7hcFKVnJeWsYlCDbr3mTS14FCizf9ccIxXA8sGL8zKeQN3g==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@types/estree": "1.0.8"
+      },
+      "bin": {
+        "rollup": "dist/bin/rollup"
+      },
+      "engines": {
+        "node": ">=18.0.0",
+        "npm": ">=8.0.0"
+      },
+      "optionalDependencies": {
+        "@rollup/rollup-android-arm-eabi": "4.60.4",
+        "@rollup/rollup-android-arm64": "4.60.4",
+        "@rollup/rollup-darwin-arm64": "4.60.4",
+        "@rollup/rollup-darwin-x64": "4.60.4",
+        "@rollup/rollup-freebsd-arm64": "4.60.4",
+        "@rollup/rollup-freebsd-x64": "4.60.4",
+        "@rollup/rollup-linux-arm-gnueabihf": "4.60.4",
+        "@rollup/rollup-linux-arm-musleabihf": "4.60.4",
+        "@rollup/rollup-linux-arm64-gnu": "4.60.4",
+        "@rollup/rollup-linux-arm64-musl": "4.60.4",
+        "@rollup/rollup-linux-loong64-gnu": "4.60.4",
+        "@rollup/rollup-linux-loong64-musl": "4.60.4",
+        "@rollup/rollup-linux-ppc64-gnu": "4.60.4",
+        "@rollup/rollup-linux-ppc64-musl": "4.60.4",
+        "@rollup/rollup-linux-riscv64-gnu": "4.60.4",
+        "@rollup/rollup-linux-riscv64-musl": "4.60.4",
+        "@rollup/rollup-linux-s390x-gnu": "4.60.4",
+        "@rollup/rollup-linux-x64-gnu": "4.60.4",
+        "@rollup/rollup-linux-x64-musl": "4.60.4",
+        "@rollup/rollup-openbsd-x64": "4.60.4",
+        "@rollup/rollup-openharmony-arm64": "4.60.4",
+        "@rollup/rollup-win32-arm64-msvc": "4.60.4",
+        "@rollup/rollup-win32-ia32-msvc": "4.60.4",
+        "@rollup/rollup-win32-x64-gnu": "4.60.4",
+        "@rollup/rollup-win32-x64-msvc": "4.60.4",
+        "fsevents": "~2.3.2"
+      }
+    },
+    "node_modules/siginfo": {
+      "version": "2.0.0",
+      "resolved": "https://registry.npmjs.org/siginfo/-/siginfo-2.0.0.tgz",
+      "integrity": "sha512-ybx0WO1/8bSBLEWXZvEd7gMW3Sn3JFlW3TvX1nREbDLRNQNaeNN8WK0meBwPdAaOI7TtRRRJn/Es1zhrrCHu7g==",
+      "dev": true,
+      "license": "ISC"
+    },
+    "node_modules/source-map": {
+      "version": "0.7.6",
+      "resolved": "https://registry.npmjs.org/source-map/-/source-map-0.7.6.tgz",
+      "integrity": "sha512-i5uvt8C3ikiWeNZSVZNWcfZPItFQOsYTUAOkcUPGd8DqDy1uOUikjt5dG+uRlwyvR108Fb9DOd4GvXfT0N2/uQ==",
+      "dev": true,
+      "license": "BSD-3-Clause",
+      "engines": {
+        "node": ">= 12"
+      }
+    },
+    "node_modules/source-map-js": {
+      "version": "1.2.1",
+      "resolved": "https://registry.npmjs.org/source-map-js/-/source-map-js-1.2.1.tgz",
+      "integrity": "sha512-UXWMKhLOwVKb728IUtQPXxfYU+usdybtUrK/8uGE8CQMvrhOpwvzDBwj0QhSL7MQc7vIsISBG8VQ8+IDQxpfQA==",
+      "dev": true,
+      "license": "BSD-3-Clause",
+      "engines": {
+        "node": ">=0.10.0"
+      }
+    },
+    "node_modules/stackback": {
+      "version": "0.0.2",
+      "resolved": "https://registry.npmjs.org/stackback/-/stackback-0.0.2.tgz",
+      "integrity": "sha512-1XMJE5fQo1jGH6Y/7ebnwPOBEkIEnT4QF32d5R1+VXdXveM0IBMJt8zfaxX1P3QhVwrYe+576+jkANtSS2mBbw==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/std-env": {
+      "version": "4.1.0",
+      "resolved": "https://registry.npmjs.org/std-env/-/std-env-4.1.0.tgz",
+      "integrity": "sha512-Rq7ybcX2RuC55r9oaPVEW7/xu3tj8u4GeBYHBWCychFtzMIr86A7e3PPEBPT37sHStKX3+TiX/Fr/ACmJLVlLQ==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/sucrase": {
+      "version": "3.35.1",
+      "resolved": "https://registry.npmjs.org/sucrase/-/sucrase-3.35.1.tgz",
+      "integrity": "sha512-DhuTmvZWux4H1UOnWMB3sk0sbaCVOoQZjv8u1rDoTV0HTdGem9hkAZtl4JZy8P2z4Bg0nT+YMeOFyVr4zcG5Tw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@jridgewell/gen-mapping": "^0.3.2",
+        "commander": "^4.0.0",
+        "lines-and-columns": "^1.1.6",
+        "mz": "^2.7.0",
+        "pirates": "^4.0.1",
+        "tinyglobby": "^0.2.11",
+        "ts-interface-checker": "^0.1.9"
+      },
+      "bin": {
+        "sucrase": "bin/sucrase",
+        "sucrase-node": "bin/sucrase-node"
+      },
+      "engines": {
+        "node": ">=16 || 14 >=14.17"
+      }
+    },
+    "node_modules/thenify": {
+      "version": "3.3.1",
+      "resolved": "https://registry.npmjs.org/thenify/-/thenify-3.3.1.tgz",
+      "integrity": "sha512-RVZSIV5IG10Hk3enotrhvz0T9em6cyHBLkH/YAZuKqd8hRkKhSfCGIcP2KUY0EPxndzANBmNllzWPwak+bheSw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "any-promise": "^1.0.0"
+      }
+    },
+    "node_modules/thenify-all": {
+      "version": "1.6.0",
+      "resolved": "https://registry.npmjs.org/thenify-all/-/thenify-all-1.6.0.tgz",
+      "integrity": "sha512-RNxQH/qI8/t3thXJDwcstUO4zeqo64+Uy/+sNVRBx4Xn2OX+OZ9oP+iJnNFqplFra2ZUVeKCSa2oVWi3T4uVmA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "thenify": ">= 3.1.0 < 4"
+      },
+      "engines": {
+        "node": ">=0.8"
+      }
+    },
+    "node_modules/tinybench": {
+      "version": "2.9.0",
+      "resolved": "https://registry.npmjs.org/tinybench/-/tinybench-2.9.0.tgz",
+      "integrity": "sha512-0+DUvqWMValLmha6lr4kD8iAMK1HzV0/aKnCtWb9v9641TnP/MFb7Pc2bxoxQjTXAErryXVgUOfv2YqNllqGeg==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/tinyexec": {
+      "version": "0.3.2",
+      "resolved": "https://registry.npmjs.org/tinyexec/-/tinyexec-0.3.2.tgz",
+      "integrity": "sha512-KQQR9yN7R5+OSwaK0XQoj22pwHoTlgYqmUscPYoknOoWCWfj/5/ABTMRi69FrKU5ffPVh5QcFikpWJI/P1ocHA==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/tinyglobby": {
+      "version": "0.2.16",
+      "resolved": "https://registry.npmjs.org/tinyglobby/-/tinyglobby-0.2.16.tgz",
+      "integrity": "sha512-pn99VhoACYR8nFHhxqix+uvsbXineAasWm5ojXoN8xEwK5Kd3/TrhNn1wByuD52UxWRLy8pu+kRMniEi6Eq9Zg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "fdir": "^6.5.0",
+        "picomatch": "^4.0.4"
+      },
+      "engines": {
+        "node": ">=12.0.0"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/SuperchupuDev"
+      }
+    },
+    "node_modules/tinyrainbow": {
+      "version": "3.1.0",
+      "resolved": "https://registry.npmjs.org/tinyrainbow/-/tinyrainbow-3.1.0.tgz",
+      "integrity": "sha512-Bf+ILmBgretUrdJxzXM0SgXLZ3XfiaUuOj/IKQHuTXip+05Xn+uyEYdVg0kYDipTBcLrCVyUzAPz7QmArb0mmw==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=14.0.0"
+      }
+    },
+    "node_modules/tree-kill": {
+      "version": "1.2.2",
+      "resolved": "https://registry.npmjs.org/tree-kill/-/tree-kill-1.2.2.tgz",
+      "integrity": "sha512-L0Orpi8qGpRG//Nd+H90vFB+3iHnue1zSSGmNOOCh1GLJ7rUKVwV2HvijphGQS2UmhUZewS9VgvxYIdgr+fG1A==",
+      "dev": true,
+      "license": "MIT",
+      "bin": {
+        "tree-kill": "cli.js"
+      }
+    },
+    "node_modules/ts-interface-checker": {
+      "version": "0.1.13",
+      "resolved": "https://registry.npmjs.org/ts-interface-checker/-/ts-interface-checker-0.1.13.tgz",
+      "integrity": "sha512-Y/arvbn+rrz3JCKl9C4kVNfTfSm2/mEp5FSz5EsZSANGPSlQrpRI5M4PKF+mJnE52jOO90PnPSc3Ur3bTQw0gA==",
+      "dev": true,
+      "license": "Apache-2.0"
+    },
+    "node_modules/tslib": {
+      "version": "2.8.1",
+      "resolved": "https://registry.npmjs.org/tslib/-/tslib-2.8.1.tgz",
+      "integrity": "sha512-oJFu94HQb+KVduSUQL7wnpmqnfmLsOA/nAh6b6EH0wCEoK0/mPeXU6c3wKDV83MkOuHPRHtSXKKU99IBazS/2w==",
+      "dev": true,
+      "license": "0BSD",
+      "optional": true
+    },
+    "node_modules/tsup": {
+      "version": "8.5.1",
+      "resolved": "https://registry.npmjs.org/tsup/-/tsup-8.5.1.tgz",
+      "integrity": "sha512-xtgkqwdhpKWr3tKPmCkvYmS9xnQK3m3XgxZHwSUjvfTjp7YfXe5tT3GgWi0F2N+ZSMsOeWeZFh7ZZFg5iPhing==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "bundle-require": "^5.1.0",
+        "cac": "^6.7.14",
+        "chokidar": "^4.0.3",
+        "consola": "^3.4.0",
+        "debug": "^4.4.0",
+        "esbuild": "^0.27.0",
+        "fix-dts-default-cjs-exports": "^1.0.0",
+        "joycon": "^3.1.1",
+        "picocolors": "^1.1.1",
+        "postcss-load-config": "^6.0.1",
+        "resolve-from": "^5.0.0",
+        "rollup": "^4.34.8",
+        "source-map": "^0.7.6",
+        "sucrase": "^3.35.0",
+        "tinyexec": "^0.3.2",
+        "tinyglobby": "^0.2.11",
+        "tree-kill": "^1.2.2"
+      },
+      "bin": {
+        "tsup": "dist/cli-default.js",
+        "tsup-node": "dist/cli-node.js"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "peerDependencies": {
+        "@microsoft/api-extractor": "^7.36.0",
+        "@swc/core": "^1",
+        "postcss": "^8.4.12",
+        "typescript": ">=4.5.0"
+      },
+      "peerDependenciesMeta": {
+        "@microsoft/api-extractor": {
+          "optional": true
+        },
+        "@swc/core": {
+          "optional": true
+        },
+        "postcss": {
+          "optional": true
+        },
+        "typescript": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/typescript": {
+      "version": "5.9.3",
+      "resolved": "https://registry.npmjs.org/typescript/-/typescript-5.9.3.tgz",
+      "integrity": "sha512-jl1vZzPDinLr9eUt3J/t7V6FgNEw9QjvBPdysz9KfQDD41fQrC2Y4vKQdiaUpFT4bXlb1RHhLpp8wtm6M5TgSw==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "bin": {
+        "tsc": "bin/tsc",
+        "tsserver": "bin/tsserver"
+      },
+      "engines": {
+        "node": ">=14.17"
+      }
+    },
+    "node_modules/ufo": {
+      "version": "1.6.4",
+      "resolved": "https://registry.npmjs.org/ufo/-/ufo-1.6.4.tgz",
+      "integrity": "sha512-JFNbkD1Svwe0KvGi8GOeLcP4kAWQ609twvCdcHxq1oSL8svv39ZuSvajcD8B+5D0eL4+s1Is2D/O6KN3qcTeRA==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/undici-types": {
+      "version": "7.24.6",
+      "resolved": "https://registry.npmjs.org/undici-types/-/undici-types-7.24.6.tgz",
+      "integrity": "sha512-WRNW+sJgj5OBN4/0JpHFqtqzhpbnV0GuB+OozA9gCL7a993SmU+1JBZCzLNxYsbMfIeDL+lTsphD5jN5N+n0zg==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/vite": {
+      "version": "8.0.14",
+      "resolved": "https://registry.npmjs.org/vite/-/vite-8.0.14.tgz",
+      "integrity": "sha512-s4BJJ+5y1pYL6Otw51FHhVJQhPnuRinKig64g/1+EUNaJsd3gCKdD31IPFvswUgW9/60QT9oFHbZHbQK5imcxw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "lightningcss": "^1.32.0",
+        "picomatch": "^4.0.4",
+        "postcss": "^8.5.15",
+        "rolldown": "1.0.2",
+        "tinyglobby": "^0.2.16"
+      },
+      "bin": {
+        "vite": "bin/vite.js"
+      },
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      },
+      "funding": {
+        "url": "https://github.com/vitejs/vite?sponsor=1"
+      },
+      "optionalDependencies": {
+        "fsevents": "~2.3.3"
+      },
+      "peerDependencies": {
+        "@types/node": "^20.19.0 || >=22.12.0",
+        "@vitejs/devtools": "^0.1.18",
+        "esbuild": "^0.27.0 || ^0.28.0",
+        "jiti": ">=1.21.0",
+        "less": "^4.0.0",
+        "sass": "^1.70.0",
+        "sass-embedded": "^1.70.0",
+        "stylus": ">=0.54.8",
+        "sugarss": "^5.0.0",
+        "terser": "^5.16.0",
+        "tsx": "^4.8.1",
+        "yaml": "^2.4.2"
+      },
+      "peerDependenciesMeta": {
+        "@types/node": {
+          "optional": true
+        },
+        "@vitejs/devtools": {
+          "optional": true
+        },
+        "esbuild": {
+          "optional": true
+        },
+        "jiti": {
+          "optional": true
+        },
+        "less": {
+          "optional": true
+        },
+        "sass": {
+          "optional": true
+        },
+        "sass-embedded": {
+          "optional": true
+        },
+        "stylus": {
+          "optional": true
+        },
+        "sugarss": {
+          "optional": true
+        },
+        "terser": {
+          "optional": true
+        },
+        "tsx": {
+          "optional": true
+        },
+        "yaml": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/vitest": {
+      "version": "4.1.7",
+      "resolved": "https://registry.npmjs.org/vitest/-/vitest-4.1.7.tgz",
+      "integrity": "sha512-flYyaFd2CgoCoU+0UKt3pxksgC+S02iTDN0n3LtqaMeXsI9SBcdNujc2k0DeFLzUn/0k538yNjOSdwgCqcrwJA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@vitest/expect": "4.1.7",
+        "@vitest/mocker": "4.1.7",
+        "@vitest/pretty-format": "4.1.7",
+        "@vitest/runner": "4.1.7",
+        "@vitest/snapshot": "4.1.7",
+        "@vitest/spy": "4.1.7",
+        "@vitest/utils": "4.1.7",
+        "es-module-lexer": "^2.0.0",
+        "expect-type": "^1.3.0",
+        "magic-string": "^0.30.21",
+        "obug": "^2.1.1",
+        "pathe": "^2.0.3",
+        "picomatch": "^4.0.3",
+        "std-env": "^4.0.0-rc.1",
+        "tinybench": "^2.9.0",
+        "tinyexec": "^1.0.2",
+        "tinyglobby": "^0.2.15",
+        "tinyrainbow": "^3.1.0",
+        "vite": "^6.0.0 || ^7.0.0 || ^8.0.0",
+        "why-is-node-running": "^2.3.0"
+      },
+      "bin": {
+        "vitest": "vitest.mjs"
+      },
+      "engines": {
+        "node": "^20.0.0 || ^22.0.0 || >=24.0.0"
+      },
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      },
+      "peerDependencies": {
+        "@edge-runtime/vm": "*",
+        "@opentelemetry/api": "^1.9.0",
+        "@types/node": "^20.0.0 || ^22.0.0 || >=24.0.0",
+        "@vitest/browser-playwright": "4.1.7",
+        "@vitest/browser-preview": "4.1.7",
+        "@vitest/browser-webdriverio": "4.1.7",
+        "@vitest/coverage-istanbul": "4.1.7",
+        "@vitest/coverage-v8": "4.1.7",
+        "@vitest/ui": "4.1.7",
+        "happy-dom": "*",
+        "jsdom": "*",
+        "vite": "^6.0.0 || ^7.0.0 || ^8.0.0"
+      },
+      "peerDependenciesMeta": {
+        "@edge-runtime/vm": {
+          "optional": true
+        },
+        "@opentelemetry/api": {
+          "optional": true
+        },
+        "@types/node": {
+          "optional": true
+        },
+        "@vitest/browser-playwright": {
+          "optional": true
+        },
+        "@vitest/browser-preview": {
+          "optional": true
+        },
+        "@vitest/browser-webdriverio": {
+          "optional": true
+        },
+        "@vitest/coverage-istanbul": {
+          "optional": true
+        },
+        "@vitest/coverage-v8": {
+          "optional": true
+        },
+        "@vitest/ui": {
+          "optional": true
+        },
+        "happy-dom": {
+          "optional": true
+        },
+        "jsdom": {
+          "optional": true
+        },
+        "vite": {
+          "optional": false
+        }
+      }
+    },
+    "node_modules/vitest/node_modules/tinyexec": {
+      "version": "1.2.2",
+      "resolved": "https://registry.npmjs.org/tinyexec/-/tinyexec-1.2.2.tgz",
+      "integrity": "sha512-M/Q0B2cp4K7kynaT/vnED1j8TlLY+Pp7C6Wl2bl/7u/F0mUVwdyOpwomQb8JpYLitHUssAJRmLZdMCGsrx7i+g==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/why-is-node-running": {
+      "version": "2.3.0",
+      "resolved": "https://registry.npmjs.org/why-is-node-running/-/why-is-node-running-2.3.0.tgz",
+      "integrity": "sha512-hUrmaWBdVDcxvYqnyh09zunKzROWjbZTiNy8dBEjkS7ehEDQibXJ7XvlmtbwuTclUiIyN+CyXQD4Vmko8fNm8w==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "siginfo": "^2.0.0",
+        "stackback": "0.0.2"
+      },
+      "bin": {
+        "why-is-node-running": "cli.js"
+      },
+      "engines": {
+        "node": ">=8"
+      }
+    }
+  }
+}
diff --git a/api_sdk/typescript/package.json b/api_sdk/typescript/package.json
new file mode 100644
index 0000000..9252c3e
--- /dev/null
+++ b/api_sdk/typescript/package.json
@@ -0,0 +1,62 @@
+{
+  "name": "@paddleocr/api-sdk",
+  "version": "0.2.3",
+  "description": "TypeScript SDK for the PaddleOCR official API",
+  "type": "module",
+  "main": "./dist/index.cjs",
+  "module": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "exports": {
+    ".": {
+      "import": {
+        "types": "./dist/index.d.ts",
+        "default": "./dist/index.js"
+      },
+      "require": {
+        "types": "./dist/index.d.cts",
+        "default": "./dist/index.cjs"
+      }
+    }
+  },
+  "files": [
+    "dist",
+    "README.md",
+    "README_cn.md"
+  ],
+  "publishConfig": {
+    "access": "public"
+  },
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/PaddlePaddle/PaddleOCR.git",
+    "directory": "api_sdk/typescript"
+  },
+  "homepage": "https://github.com/PaddlePaddle/PaddleOCR/tree/main/api_sdk/typescript",
+  "bugs": {
+    "url": "https://github.com/PaddlePaddle/PaddleOCR/issues"
+  },
+  "keywords": [
+    "paddleocr",
+    "ocr",
+    "document-parsing",
+    "api-sdk",
+    "typescript",
+    "official-api"
+  ],
+  "scripts": {
+    "build": "tsup",
+    "test": "vitest run",
+    "lint": "tsc --noEmit",
+    "prepublishOnly": "npm run lint && npm run build && npm test"
+  },
+  "engines": {
+    "node": ">=18"
+  },
+  "license": "Apache-2.0",
+  "devDependencies": {
+    "@types/node": "^25.9.1",
+    "tsup": "^8.0.0",
+    "typescript": "^5.3.0",
+    "vitest": "^4.1.7"
+  }
+}
diff --git a/api_sdk/typescript/src/client.ts b/api_sdk/typescript/src/client.ts
new file mode 100644
index 0000000..96a287f
--- /dev/null
+++ b/api_sdk/typescript/src/client.ts
@@ -0,0 +1,394 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import { stat, writeFile } from "node:fs/promises";
+import { basename, dirname, extname, join } from "node:path";
+import { AuthError, FileNotFoundError, InvalidRequestError, ResultParseError } from "./errors.js";
+import { HttpClient } from "./internal/http.js";
+import { Poller } from "./internal/poller.js";
+import type { ClientOptions, DocParsingRequest, OCRRequest, SaveResourceOptions } from "./models.js";
+import { isDocumentParsingModel, isOCRModel, Model } from "./models.js";
+import type { BatchStatus, DocParsingResult, Job, JobStatus, OCRResult } from "./results.js";
+
+const DEFAULT_BASE_URL = "https://paddleocr.aistudio-app.com";
+
+interface ResourceSavePlan {
+  resourceUrl: string;
+  filename: string;
+}
+
+export class PaddleOCRClient {
+  private http: HttpClient;
+  private poller: Poller;
+
+  constructor(options: ClientOptions = {}) {
+    const token = options.token || process.env.PADDLEOCR_ACCESS_TOKEN || "";
+    if (!token) {
+      throw new AuthError("Token is required. Set PADDLEOCR_ACCESS_TOKEN or pass token option.");
+    }
+    const baseUrl = options.baseUrl || process.env.PADDLEOCR_BASE_URL || DEFAULT_BASE_URL;
+    const requestTimeout = options.requestTimeout || options.timeout || 300000;
+    const pollTimeout = options.pollTimeout || options.timeout || 600000;
+
+    this.http = new HttpClient(
+      token,
+      baseUrl,
+      requestTimeout,
+      options.fetch,
+      options.clientPlatform,
+    );
+    this.poller = new Poller(this.http, pollTimeout);
+  }
+
+  async ocr(req: OCRRequest, options?: { signal?: AbortSignal }): Promise<OCRResult> {
+    const job = await this.submitOcr(req, options);
+    return this.waitOcrResult(job, options);
+  }
+
+  async parseDocument(req: DocParsingRequest, options?: { signal?: AbortSignal }): Promise<DocParsingResult> {
+    const job = await this.submitDocumentParsing(req, options);
+    return this.waitDocumentParsingResult(job, options);
+  }
+
+  async submitOcr(req: OCRRequest, options?: { signal?: AbortSignal }): Promise<Job> {
+    const model = req.model ?? Model.PPOCRv6;
+    const jobId = await this.submit(model, "ocr", req, options?.signal);
+    return { jobId, model, task: "ocr", pageRanges: req.pageRanges, batchId: req.batchId };
+  }
+
+  async submitDocumentParsing(req: DocParsingRequest, options?: { signal?: AbortSignal }): Promise<Job> {
+    const model = req.model ?? Model.PaddleOCRVL16;
+    if (!isDocumentParsingModel(model)) {
+      throw new InvalidRequestError(`Model ${model} is not a document parsing model.`);
+    }
+    const jobId = await this.submit(model, "document_parsing", req, options?.signal);
+    return { jobId, model, task: "document_parsing", pageRanges: req.pageRanges, batchId: req.batchId };
+  }
+
+  async waitOcrResult(job: Job | string, options?: { signal?: AbortSignal }): Promise<OCRResult> {
+    const resolved = this.resolveJob(job, "ocr");
+    const jsonlData = await this.poller.pollUntilDone(resolved.jobId, options?.signal);
+    return this.parseOCRResult(resolved.jobId, jsonlData);
+  }
+
+  async waitDocumentParsingResult(job: Job | string, options?: { signal?: AbortSignal }): Promise<DocParsingResult> {
+    const resolved = this.resolveJob(job, "document_parsing");
+    const jsonlData = await this.poller.pollUntilDone(resolved.jobId, options?.signal);
+    return this.parseDocParsingResult(resolved.jobId, jsonlData);
+  }
+
+  async getStatus(jobId: string, options?: { signal?: AbortSignal }): Promise<JobStatus> {
+    return this.poller.getStatus(jobId, options?.signal);
+  }
+
+  async getBatchStatus(batchId: string, options?: { signal?: AbortSignal }): Promise<BatchStatus> {
+    return this.poller.getBatchStatus(batchId, options?.signal);
+  }
+
+  async saveResource(
+    resourceUrl: string,
+    destination: string,
+    options: SaveResourceOptions = {},
+  ): Promise<string> {
+    return this.saveResourceUrl(resourceUrl, destination, options);
+  }
+
+  async saveOcrResultResources(
+    result: OCRResult,
+    destination: string,
+    options: SaveResourceOptions = {},
+  ): Promise<string[]> {
+    return this.saveResultResources(result, destination, options);
+  }
+
+  async saveDocumentParsingResultResources(
+    result: DocParsingResult,
+    destination: string,
+    options: SaveResourceOptions = {},
+  ): Promise<string[]> {
+    return this.saveResultResources(result, destination, options);
+  }
+
+  private async saveResourceUrl(resourceUrl: string, destination: string, options: SaveResourceOptions): Promise<string> {
+    if (!resourceUrl) {
+      throw new InvalidRequestError("resourceUrl is required.");
+    }
+    let url: URL;
+    try {
+      url = new URL(resourceUrl);
+    } catch (error) {
+      throw new InvalidRequestError(`Invalid resource URL: ${resourceUrl}`, { cause: error });
+    }
+    const target = await this.resolveDestination(url, destination, options);
+    const content = await this.http.fetchResource(resourceUrl);
+    await writeFile(target, Buffer.from(content), { flag: options.overwrite ? "w" : "wx" });
+    return target;
+  }
+
+  private async saveResultResources(
+    result: OCRResult | DocParsingResult,
+    destination: string,
+    options: SaveResourceOptions,
+  ): Promise<string[]> {
+    await this.requireExistingDirectory(destination);
+    const plans = this.collectResultResourcePlans(result);
+    const targets = plans.map((plan) => join(destination, plan.filename));
+    for (const target of targets) {
+      await this.requireWritableTarget(target, options);
+    }
+    this.requireUniqueTargets(targets, options);
+    const savedPaths: string[] = [];
+    for (const [index, plan] of plans.entries()) {
+      await this.saveResourceUrl(plan.resourceUrl, targets[index], options);
+      savedPaths.push(targets[index]);
+    }
+    return savedPaths;
+  }
+
+  private collectResultResourcePlans(result: OCRResult | DocParsingResult): ResourceSavePlan[] {
+    if (isDocParsingResult(result)) {
+      return result.pages.flatMap((page) => [
+        ...this.collectMappedResourcePlans(page.markdownImages),
+        ...this.collectMappedResourcePlans(page.outputImages),
+      ]);
+    }
+    return result.pages.flatMap((page, index) => {
+      if (!page.ocrImageUrl) {
+        return [];
+      }
+      return [{
+        resourceUrl: page.ocrImageUrl,
+        filename: `ocr-page-${index + 1}${resourceExtension(page.ocrImageUrl)}`,
+      }];
+    });
+  }
+
+  private collectMappedResourcePlans(resources: Record<string, string>): ResourceSavePlan[] {
+    return Object.keys(resources)
+      .sort()
+      .map((key) => ({
+        resourceUrl: resources[key],
+        filename: safeMapKeyFilename(key),
+      }));
+  }
+
+  private async requireExistingDirectory(destination: string): Promise<void> {
+    let destinationStat;
+    try {
+      destinationStat = await stat(destination);
+    } catch {
+      throw new FileNotFoundError(destination);
+    }
+    if (!destinationStat.isDirectory()) {
+      throw new InvalidRequestError(`Destination must be an existing directory: ${destination}`);
+    }
+  }
+
+  private async requireWritableTarget(target: string, options: SaveResourceOptions): Promise<void> {
+    try {
+      await stat(target);
+    } catch {
+      return;
+    }
+    if (!options.overwrite) {
+      throw new InvalidRequestError(`Destination already exists: ${target}`);
+    }
+  }
+
+  private requireUniqueTargets(targets: string[], options: SaveResourceOptions): void {
+    if (options.overwrite) {
+      return;
+    }
+    const seen = new Set<string>();
+    for (const target of targets) {
+      if (seen.has(target)) {
+        throw new InvalidRequestError(`Destination already exists: ${target}`);
+      }
+      seen.add(target);
+    }
+  }
+
+  private async resolveDestination(url: URL, destination: string, options: SaveResourceOptions): Promise<string> {
+    let destinationStat;
+    try {
+      destinationStat = await stat(destination);
+    } catch {
+      destinationStat = undefined;
+    }
+    const target = destinationStat?.isDirectory() ? join(destination, safeUrlBasename(url)) : destination;
+    await this.requireWritableTarget(target, options);
+    const parent = dirname(target);
+    try {
+      const parentStat = await stat(parent);
+      if (!parentStat.isDirectory()) {
+        throw new InvalidRequestError(`Destination parent must be a directory: ${parent}`);
+      }
+    } catch (error) {
+      if (error instanceof InvalidRequestError) {
+        throw error;
+      }
+      throw new FileNotFoundError(parent, { cause: error });
+    }
+    return target;
+  }
+
+  private async submit(
+    model: string,
+    task: Job["task"],
+    req: { fileUrl?: string; filePath?: string; pageRanges?: string; batchId?: string; options?: object },
+    signal?: AbortSignal,
+  ): Promise<string> {
+    if (!req.fileUrl && !req.filePath) {
+      throw new InvalidRequestError("Either fileUrl or filePath is required.");
+    }
+    if (req.fileUrl && req.filePath) {
+      throw new InvalidRequestError("fileUrl and filePath are mutually exclusive.");
+    }
+
+    this.validateModelForTask(model, task);
+    const payload = req.options || {};
+
+    if (req.fileUrl) {
+      return this.http.submitUrl(model, req.fileUrl, payload, {
+        pageRanges: req.pageRanges,
+        batchId: req.batchId,
+        signal,
+      });
+    }
+    return this.http.submitFile(model, req.filePath!, payload, {
+      pageRanges: req.pageRanges,
+      batchId: req.batchId,
+      signal,
+    });
+  }
+
+  private parseOCRResult(jobId: string, jsonlData: unknown[]): OCRResult {
+    const dataInfo: Record<string, unknown> = {};
+    const pages = jsonlData.flatMap((lineObj) => {
+      if (!isRecord(lineObj) || !isRecord(lineObj.result) || !Array.isArray(lineObj.result.ocrResults)) {
+        throw new ResultParseError("OCR result item is missing result.ocrResults.");
+      }
+      if (isRecord(lineObj.result.dataInfo)) {
+        Object.assign(dataInfo, lineObj.result.dataInfo);
+      }
+      return lineObj.result.ocrResults.map((item) => {
+        if (!isRecord(item) || !("prunedResult" in item)) {
+          throw new ResultParseError("OCR result page is missing prunedResult.");
+        }
+        return {
+          prunedResult: item.prunedResult,
+          ocrImageUrl: typeof item.ocrImage === "string" ? item.ocrImage : undefined,
+          docPreprocessingImageUrl: typeof item.docPreprocessingImage === "string" ? item.docPreprocessingImage : undefined,
+          inputImageUrl: typeof item.inputImage === "string" ? item.inputImage : undefined,
+          raw: item,
+        };
+      });
+    });
+    return { jobId, pages, dataInfo };
+  }
+
+  private parseDocParsingResult(jobId: string, jsonlData: unknown[]): DocParsingResult {
+    const dataInfo: Record<string, unknown> = {};
+    const pages = jsonlData.flatMap((lineObj) => {
+      if (!isRecord(lineObj) || !isRecord(lineObj.result) || !Array.isArray(lineObj.result.layoutParsingResults)) {
+        throw new ResultParseError("Document parsing result item is missing result.layoutParsingResults.");
+      }
+      if (isRecord(lineObj.result.dataInfo)) {
+        Object.assign(dataInfo, lineObj.result.dataInfo);
+      }
+      return lineObj.result.layoutParsingResults.map((item) => {
+        if (!isRecord(item) || !isRecord(item.markdown) || typeof item.markdown.text !== "string") {
+          throw new ResultParseError("Document parsing result page is missing markdown.text.");
+        }
+        return {
+          markdownText: item.markdown.text,
+          markdownImages: isRecord(item.markdown.images) ? stringMap(item.markdown.images) : {},
+          outputImages: isRecord(item.outputImages) ? stringMap(item.outputImages) : {},
+          prunedResult: item.prunedResult,
+          inputImageUrl: typeof item.inputImage === "string" ? item.inputImage : undefined,
+          exports: isRecord(item.exports) ? item.exports : {},
+          markdown: item.markdown,
+          raw: item,
+        };
+      });
+    });
+    return { jobId, pages, dataInfo };
+  }
+
+  private resolveJob(job: Job | string, expectedTask: Job["task"]): Job {
+    if (typeof job === "string") {
+      return {
+        jobId: job,
+        model: expectedTask === "ocr" ? Model.PPOCRv6 : Model.PaddleOCRVL16,
+        task: expectedTask,
+      };
+    }
+    if (job.task !== expectedTask) {
+      throw new InvalidRequestError(`Job ${job.jobId} is a ${job.task} job, not a ${expectedTask} job.`);
+    }
+    this.validateModelForTask(job.model, expectedTask);
+    return job;
+  }
+
+  private validateModelForTask(model: string, task: Job["task"]): void {
+    if (task === "ocr" && !isOCRModel(model)) {
+      throw new InvalidRequestError(`Model ${model} is not an OCR model.`);
+    }
+    if (task === "document_parsing" && !isDocumentParsingModel(model)) {
+      throw new InvalidRequestError(`Model ${model} is not a document parsing model.`);
+    }
+  }
+}
+
+function stringMap(value: Record<string, unknown>): Record<string, string> {
+  const result: Record<string, string> = {};
+  for (const [key, val] of Object.entries(value)) {
+    if (typeof val === "string") {
+      result[key] = val;
+    }
+  }
+  return result;
+}
+
+function isRecord(value: unknown): value is Record<string, unknown> {
+  return typeof value === "object" && value !== null;
+}
+
+function isDocParsingResult(result: OCRResult | DocParsingResult): result is DocParsingResult {
+  return result.pages.some((page) => "markdownText" in page);
+}
+
+function safeMapKeyFilename(key: string): string {
+  if (!key || key === "." || key === ".." || key.includes("/") || key.includes("\\") || key.startsWith(".")) {
+    throw new InvalidRequestError(`Unsafe resource filename: ${key}`);
+  }
+  return key;
+}
+
+function safeUrlBasename(url: URL): string {
+  const name = basename(url.pathname) || "resource";
+  if (name === "." || name === "..") {
+    return "resource";
+  }
+  return name;
+}
+
+function resourceExtension(resourceUrl: string): string {
+  try {
+    const url = new URL(resourceUrl);
+    return extname(url.pathname);
+  } catch {
+    return "";
+  }
+}
diff --git a/api_sdk/typescript/src/errors.ts b/api_sdk/typescript/src/errors.ts
new file mode 100644
index 0000000..76a6d44
--- /dev/null
+++ b/api_sdk/typescript/src/errors.ts
@@ -0,0 +1,119 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+export class PaddleOCRAPIError extends Error {
+  constructor(message: string, options?: ErrorOptions) {
+    super(message);
+    this.name = "PaddleOCRAPIError";
+    this.cause = options?.cause;
+  }
+}
+
+export class AuthError extends PaddleOCRAPIError {
+  constructor(message: string, options?: ErrorOptions) {
+    super(message, options);
+    this.name = "AuthError";
+  }
+}
+
+export class InvalidRequestError extends PaddleOCRAPIError {
+  constructor(message: string, options?: ErrorOptions) {
+    super(message, options);
+    this.name = "InvalidRequestError";
+  }
+}
+
+export class APIError extends PaddleOCRAPIError {
+  statusCode: number;
+  constructor(statusCode: number, message: string, options?: ErrorOptions) {
+    super(`HTTP ${statusCode}: ${message}`, options);
+    this.name = "APIError";
+    this.statusCode = statusCode;
+  }
+}
+
+export class RateLimitError extends APIError {
+  constructor(message: string, options?: ErrorOptions) {
+    super(429, message, options);
+    this.name = "RateLimitError";
+  }
+}
+
+export class ServiceUnavailableError extends APIError {
+  constructor(statusCode: number, message: string, options?: ErrorOptions) {
+    super(statusCode, message, options);
+    this.name = "ServiceUnavailableError";
+  }
+}
+
+export class JobFailedError extends PaddleOCRAPIError {
+  jobId: string;
+  errorMsg: string;
+  constructor(jobId: string, errorMsg: string, options?: ErrorOptions) {
+    super(`Job ${jobId} failed: ${errorMsg}`, options);
+    this.name = "JobFailedError";
+    this.jobId = jobId;
+    this.errorMsg = errorMsg;
+  }
+}
+
+export class RequestTimeoutError extends PaddleOCRAPIError {
+  timeoutMs: number;
+  constructor(timeoutMs: number, options?: ErrorOptions) {
+    super(`Request timed out after ${timeoutMs}ms`, options);
+    this.name = "RequestTimeoutError";
+    this.timeoutMs = timeoutMs;
+  }
+}
+
+export class PollTimeoutError extends PaddleOCRAPIError {
+  jobId: string;
+  timeoutMs: number;
+  constructor(jobId: string, timeoutMs: number, options?: ErrorOptions) {
+    super(`Timed out after ${timeoutMs}ms waiting for job ${jobId}`, options);
+    this.name = "PollTimeoutError";
+    this.jobId = jobId;
+    this.timeoutMs = timeoutMs;
+  }
+}
+
+export class NetworkError extends PaddleOCRAPIError {
+  constructor(message: string, options?: ErrorOptions) {
+    super(message, options);
+    this.name = "NetworkError";
+  }
+}
+
+export class FileNotFoundError extends PaddleOCRAPIError {
+  path: string;
+  constructor(path: string, options?: ErrorOptions) {
+    super(`File not found: ${path}`, options);
+    this.name = "FileNotFoundError";
+    this.path = path;
+  }
+}
+
+export class ResponseFormatError extends PaddleOCRAPIError {
+  constructor(message: string, options?: ErrorOptions) {
+    super(message, options);
+    this.name = "ResponseFormatError";
+  }
+}
+
+export class ResultParseError extends PaddleOCRAPIError {
+  constructor(message: string, options?: ErrorOptions) {
+    super(message, options);
+    this.name = "ResultParseError";
+  }
+}
diff --git a/api_sdk/typescript/src/index.ts b/api_sdk/typescript/src/index.ts
new file mode 100644
index 0000000..4e57a82
--- /dev/null
+++ b/api_sdk/typescript/src/index.ts
@@ -0,0 +1,51 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+export { PaddleOCRClient } from "./client.js";
+export { Model, isDocumentParsingModel, isOCRModel, isVLModel } from "./models.js";
+export type {
+  ClientOptions,
+  DocParsingOptions,
+  DocParsingRequest,
+  OCROptions,
+  OCRRequest,
+  PaddleOCRVLOptions,
+  PPStructureV3Options,
+  SaveResourceOptions,
+} from "./models.js";
+export type {
+  BatchStatus,
+  DocParsingPage,
+  DocParsingResult,
+  Job,
+  JobStatus,
+  OCRPage,
+  OCRResult,
+  Progress,
+} from "./results.js";
+export {
+  APIError,
+  AuthError,
+  FileNotFoundError,
+  InvalidRequestError,
+  JobFailedError,
+  NetworkError,
+  PaddleOCRAPIError,
+  PollTimeoutError,
+  RateLimitError,
+  RequestTimeoutError,
+  ResponseFormatError,
+  ResultParseError,
+  ServiceUnavailableError,
+} from "./errors.js";
diff --git a/api_sdk/typescript/src/internal/abort.ts b/api_sdk/typescript/src/internal/abort.ts
new file mode 100644
index 0000000..8839143
--- /dev/null
+++ b/api_sdk/typescript/src/internal/abort.ts
@@ -0,0 +1,25 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+/** Value to reject/throw when an operation was aborted via `AbortSignal` (aligns with `fetch` + `DOMException`). */
+export function userAbortReason(signal: AbortSignal): unknown {
+  return signal.reason !== undefined
+    ? signal.reason
+    : new DOMException("The user aborted a request.", "AbortError");
+}
+
+export function throwIfAborted(signal?: AbortSignal): void {
+  if (!signal?.aborted) return;
+  throw userAbortReason(signal);
+}
diff --git a/api_sdk/typescript/src/internal/http.ts b/api_sdk/typescript/src/internal/http.ts
new file mode 100644
index 0000000..fe40700
--- /dev/null
+++ b/api_sdk/typescript/src/internal/http.ts
@@ -0,0 +1,255 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import {
+  APIError,
+  AuthError,
+  InvalidRequestError,
+  NetworkError,
+  RateLimitError,
+  RequestTimeoutError,
+  ResponseFormatError,
+  ResultParseError,
+  ServiceUnavailableError,
+} from "../errors.js";
+import { userAbortReason } from "./abort.js";
+
+const DEFAULT_BASE_URL = "https://paddleocr.aistudio-app.com";
+const API_PATH = "/api/v2/ocr/jobs";
+
+interface SubmitOptions {
+  pageRanges?: string;
+  batchId?: string;
+  signal?: AbortSignal;
+}
+
+interface APIResponse<T> {
+  code?: number;
+  msg?: string;
+  data: T;
+}
+
+interface SubmitResponse {
+  jobId: string;
+}
+
+export class HttpClient {
+  private baseUrl: string;
+  private jobsUrl: string;
+  private token: string;
+  private requestTimeout: number;
+  private fetchImpl: typeof fetch;
+  private clientPlatform?: string;
+
+  constructor(
+    token: string,
+    baseUrl: string = DEFAULT_BASE_URL,
+    requestTimeout: number = 300000,
+    fetchImpl: typeof fetch = fetch,
+    clientPlatform?: string,
+  ) {
+    this.token = token;
+    this.baseUrl = baseUrl.replace(/\/+$/, "");
+    this.jobsUrl = `${this.baseUrl}${API_PATH}`;
+    this.requestTimeout = requestTimeout;
+    this.fetchImpl = fetchImpl;
+    this.clientPlatform = clientPlatform;
+  }
+
+  async submitUrl(model: string, fileUrl: string, optionalPayload: object, options: SubmitOptions = {}): Promise<string> {
+    const body: Record<string, unknown> = { fileUrl, model, optionalPayload };
+    if (options.pageRanges !== undefined) {
+      body.pageRanges = options.pageRanges;
+    }
+    if (options.batchId !== undefined) {
+      body.batchId = options.batchId;
+    }
+    const data = await this.fetchJson<SubmitResponse>(this.jobsUrl, {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify(body),
+    }, options.signal);
+    return requireJobId(data);
+  }
+
+  async submitFile(model: string, filePath: string, optionalPayload: object, options: SubmitOptions = {}): Promise<string> {
+    const fs = await import("fs");
+    const path = await import("path");
+
+    if (!fs.existsSync(filePath)) {
+      const { FileNotFoundError } = await import("../errors.js");
+      throw new FileNotFoundError(filePath);
+    }
+
+    const form = new FormData();
+    form.append("model", model);
+    form.append("optionalPayload", JSON.stringify(optionalPayload));
+    if (options.pageRanges !== undefined) {
+      form.append("pageRanges", options.pageRanges);
+    }
+    if (options.batchId !== undefined) {
+      form.append("batchId", options.batchId);
+    }
+
+    const fileBuffer = fs.readFileSync(filePath);
+    const blob = new Blob([fileBuffer]);
+    form.append("file", blob, path.basename(filePath));
+
+    const data = await this.fetchJson<SubmitResponse>(this.jobsUrl, {
+      method: "POST",
+      body: form,
+    }, options.signal);
+    return requireJobId(data);
+  }
+
+  async getJobStatus(jobId: string, signal?: AbortSignal, timeoutMs?: number): Promise<unknown> {
+    return this.fetchJson<unknown>(
+      `${this.jobsUrl}/${encodeURIComponent(jobId)}`,
+      { method: "GET" },
+      signal,
+      true,
+      timeoutMs,
+    );
+  }
+
+  async getBatchStatus(batchId: string, signal?: AbortSignal, timeoutMs?: number): Promise<unknown> {
+    return this.fetchJson<unknown>(
+      `${this.jobsUrl}/batch/${encodeURIComponent(batchId)}`,
+      { method: "GET" },
+      signal,
+      true,
+      timeoutMs,
+    );
+  }
+
+  async fetchJsonl(url: string, signal?: AbortSignal, timeoutMs?: number): Promise<unknown[]> {
+    const resp = await this.fetch(url, { method: "GET" }, signal, false, timeoutMs);
+    const text = await resp.text();
+    try {
+      return text
+        .trim()
+        .split("\n")
+        .filter((line) => line.trim())
+        .map((line) => JSON.parse(line) as unknown);
+    } catch (error) {
+      throw new ResultParseError("Failed to parse JSONL result payload.", { cause: error });
+    }
+  }
+
+  async fetchResource(url: string, signal?: AbortSignal, timeoutMs?: number): Promise<ArrayBuffer> {
+    const resp = await this.fetch(url, { method: "GET" }, signal, false, timeoutMs);
+    return resp.arrayBuffer();
+  }
+
+  private async fetchJson<T>(
+    url: string,
+    init: RequestInit,
+    signal?: AbortSignal,
+    withAuth: boolean = true,
+    timeoutMs?: number,
+  ): Promise<T> {
+    const resp = await this.fetch(url, init, signal, withAuth, timeoutMs);
+    let payload: APIResponse<T>;
+    try {
+      payload = await resp.json() as APIResponse<T>;
+    } catch (error) {
+      throw new ResponseFormatError("Expected a JSON response body.", { cause: error });
+    }
+    if (payload.code !== undefined && payload.code !== 0) {
+      throw new APIError(resp.status, payload.msg || "PaddleOCR official API request failed.");
+    }
+    if (!payload || typeof payload !== "object" || !("data" in payload)) {
+      throw new ResponseFormatError("Response body is missing data.");
+    }
+    return payload.data;
+  }
+
+  private async fetch(
+    url: string,
+    init: RequestInit,
+    signal?: AbortSignal,
+    withAuth: boolean = true,
+    timeoutMs?: number,
+  ): Promise<Response> {
+    const headers: Record<string, string> = {
+      ...(init.headers as Record<string, string> || {}),
+    };
+    if (withAuth) {
+      headers.Authorization = `Bearer ${this.token}`;
+      if (this.clientPlatform) {
+        headers["Client-Platform"] = this.clientPlatform;
+      }
+    }
+
+    let resp: Response;
+    const timeoutController = new AbortController();
+    const effectiveTimeout = Math.max(0, Math.min(this.requestTimeout, timeoutMs ?? this.requestTimeout));
+    const timeoutID = setTimeout(() => timeoutController.abort(), effectiveTimeout);
+    const abortController = new AbortController();
+    const abort = () => abortController.abort();
+    timeoutController.signal.addEventListener("abort", abort, { once: true });
+    if (signal?.aborted) {
+      abort();
+    } else {
+      signal?.addEventListener("abort", abort, { once: true });
+    }
+    try {
+      resp = await this.fetchImpl(url, {
+        ...init,
+        headers,
+        signal: abortController.signal,
+      });
+    } catch (e: unknown) {
+      if (signal?.aborted) {
+        throw userAbortReason(signal);
+      }
+      if (timeoutController.signal.aborted) {
+        throw new RequestTimeoutError(effectiveTimeout, { cause: e });
+      }
+      const message = e instanceof Error ? e.message : String(e);
+      throw new NetworkError(`Connection failed: ${message}`);
+    } finally {
+      clearTimeout(timeoutID);
+      signal?.removeEventListener("abort", abort);
+    }
+
+    if (resp.ok) return resp;
+
+    let text = await resp.text();
+    try {
+      const payload = JSON.parse(text) as { msg?: string; message?: string; errorMsg?: string };
+      text = payload.msg || payload.message || payload.errorMsg || text;
+    } catch {
+      // Keep raw response text.
+    }
+    if (resp.status === 401 || resp.status === 403) {
+      throw new AuthError(`Authentication failed: ${text}`);
+    } else if (resp.status === 400) {
+      throw new InvalidRequestError(`Bad request: ${text}`);
+    } else if (resp.status === 429) {
+      throw new RateLimitError(`Rate limit exceeded: ${text}`);
+    } else if (resp.status === 503 || resp.status === 504) {
+      throw new ServiceUnavailableError(resp.status, `Service unavailable: ${text}`);
+    } else {
+      throw new APIError(resp.status, text);
+    }
+  }
+}
+
+function requireJobId(data: SubmitResponse): string {
+  if (!data || typeof data.jobId !== "string" || data.jobId.length === 0) {
+    throw new ResponseFormatError("Submit response is missing jobId.");
+  }
+  return data.jobId;
+}
diff --git a/api_sdk/typescript/src/internal/poller.ts b/api_sdk/typescript/src/internal/poller.ts
new file mode 100644
index 0000000..dc6f0b2
--- /dev/null
+++ b/api_sdk/typescript/src/internal/poller.ts
@@ -0,0 +1,162 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import { JobFailedError, PollTimeoutError, RequestTimeoutError, ResponseFormatError } from "../errors.js";
+import type { BatchStatus, JobStatus, Progress } from "../results.js";
+import { throwIfAborted, userAbortReason } from "./abort.js";
+import { HttpClient } from "./http.js";
+
+const INITIAL_INTERVAL = 3000;
+const MULTIPLIER = 1.5;
+const MAX_INTERVAL = 15000;
+const MAX_WAIT_TIME = 600000;
+
+export class Poller {
+  private http: HttpClient;
+  private maxWaitTime: number;
+
+  constructor(http: HttpClient, maxWaitTime: number = MAX_WAIT_TIME) {
+    this.http = http;
+    this.maxWaitTime = maxWaitTime;
+  }
+
+  async pollUntilDone(jobId: string, signal?: AbortSignal): Promise<unknown[]> {
+    let interval = INITIAL_INTERVAL;
+    const deadline = Date.now() + this.maxWaitTime;
+
+    while (Date.now() < deadline) {
+      throwIfAborted(signal);
+
+      const remaining = deadline - Date.now();
+      const data = await this.withPollTimeout(jobId, remaining, () => this.http.getJobStatus(jobId, signal, remaining));
+      const status = normalizeStatus(jobId, data);
+
+      if (status.state === "done") {
+        const jsonUrl = resultJsonUrl(data);
+        const resultRemaining = deadline - Date.now();
+        return await this.withPollTimeout(jobId, resultRemaining, () => this.http.fetchJsonl(jsonUrl, signal, resultRemaining));
+      }
+
+      if (status.state === "failed") {
+        throw new JobFailedError(jobId, status.errorMsg || "Unknown error");
+      }
+
+      await this.sleep(Math.min(interval, Math.max(0, deadline - Date.now())), signal);
+      interval = Math.min(interval * MULTIPLIER, MAX_INTERVAL);
+    }
+
+    throw new PollTimeoutError(jobId, this.maxWaitTime);
+  }
+
+  async getStatus(jobId: string, signal?: AbortSignal): Promise<JobStatus> {
+    return normalizeStatus(jobId, await this.http.getJobStatus(jobId, signal));
+  }
+
+  async getBatchStatus(batchId: string, signal?: AbortSignal): Promise<BatchStatus> {
+    const data = await this.http.getBatchStatus(batchId, signal);
+    if (!isRecord(data) || !Array.isArray(data.extractResult)) {
+      throw new ResponseFormatError("Batch response is missing extractResult.");
+    }
+    return {
+      batchId,
+      jobs: data.extractResult.map((item) => {
+        if (!isRecord(item) || typeof item.jobId !== "string") {
+          throw new ResponseFormatError("Batch extractResult item is missing jobId.");
+        }
+        return normalizeStatus(item.jobId, item);
+      }),
+    };
+  }
+
+  private sleep(ms: number, signal?: AbortSignal): Promise<void> {
+    return new Promise((resolve, reject) => {
+      const timer = setTimeout(resolve, ms);
+      if (!signal) return;
+      signal.addEventListener(
+        "abort",
+        () => {
+          clearTimeout(timer);
+          reject(userAbortReason(signal));
+        },
+        { once: true },
+      );
+    });
+  }
+
+  private async withPollTimeout<T>(jobId: string, remainingMs: number, operation: () => Promise<T>): Promise<T> {
+    if (remainingMs <= 0) {
+      throw new PollTimeoutError(jobId, this.maxWaitTime);
+    }
+    try {
+      return await operation();
+    } catch (error) {
+      if (error instanceof RequestTimeoutError && error.timeoutMs === remainingMs) {
+        throw new PollTimeoutError(jobId, this.maxWaitTime, { cause: error });
+      }
+      throw error;
+    }
+  }
+}
+
+function normalizeStatus(jobId: string, data: unknown): JobStatus {
+  if (!isRecord(data) || typeof data.state !== "string") {
+    throw new ResponseFormatError("Status response is missing state.");
+  }
+  if (!["pending", "running", "done", "failed"].includes(data.state)) {
+    throw new ResponseFormatError(`Unknown job state: ${data.state}`);
+  }
+  return {
+    jobId,
+    state: data.state as JobStatus["state"],
+    progress: normalizeProgress(data.extractProgress),
+    resultUrl: isRecord(data.resultUrl) ? stringMap(data.resultUrl) : undefined,
+    errorMsg: typeof data.errorMsg === "string" ? data.errorMsg : undefined,
+  };
+}
+
+function normalizeProgress(progress: unknown): Progress | undefined {
+  if (progress === undefined || progress === null) {
+    return undefined;
+  }
+  if (!isRecord(progress)) {
+    throw new ResponseFormatError("Status progress must be an object.");
+  }
+  return {
+    totalPages: typeof progress.totalPages === "number" ? progress.totalPages : 0,
+    extractedPages: typeof progress.extractedPages === "number" ? progress.extractedPages : 0,
+    startTime: typeof progress.startTime === "string" ? progress.startTime : undefined,
+    endTime: typeof progress.endTime === "string" ? progress.endTime : undefined,
+  };
+}
+
+function resultJsonUrl(data: unknown): string {
+  if (!isRecord(data) || !isRecord(data.resultUrl) || typeof data.resultUrl.jsonUrl !== "string") {
+    throw new ResponseFormatError("Done job response is missing resultUrl.jsonUrl.");
+  }
+  return data.resultUrl.jsonUrl;
+}
+
+function stringMap(value: Record<string, unknown>): Record<string, string> {
+  const result: Record<string, string> = {};
+  for (const [key, val] of Object.entries(value)) {
+    if (typeof val === "string") {
+      result[key] = val;
+    }
+  }
+  return result;
+}
+
+function isRecord(value: unknown): value is Record<string, unknown> {
+  return typeof value === "object" && value !== null;
+}
diff --git a/api_sdk/typescript/src/models.ts b/api_sdk/typescript/src/models.ts
new file mode 100644
index 0000000..6cf7d7d
--- /dev/null
+++ b/api_sdk/typescript/src/models.ts
@@ -0,0 +1,168 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+export enum Model {
+  PPOCRv5 = "PP-OCRv5",
+  PPOCRv5Latin = "PP-OCRv5-latin",
+  PPOCRv6 = "PP-OCRv6",
+  PPStructureV3 = "PP-StructureV3",
+  PaddleOCRVL = "PaddleOCR-VL",
+  PaddleOCRVL15 = "PaddleOCR-VL-1.5",
+  PaddleOCRVL16 = "PaddleOCR-VL-1.6",
+}
+
+const OCR_MODELS = new Set<string>([Model.PPOCRv5, Model.PPOCRv5Latin, Model.PPOCRv6]);
+const DOCUMENT_PARSING_MODELS = new Set<string>([
+  Model.PPStructureV3,
+  Model.PaddleOCRVL,
+  Model.PaddleOCRVL15,
+  Model.PaddleOCRVL16,
+]);
+const VL_MODELS = new Set<string>([
+  Model.PaddleOCRVL,
+  Model.PaddleOCRVL15,
+  Model.PaddleOCRVL16,
+]);
+
+export function isOCRModel(
+  model: string
+): model is Model.PPOCRv5 | Model.PPOCRv5Latin | Model.PPOCRv6 {
+  return OCR_MODELS.has(model);
+}
+
+export function isDocumentParsingModel(model: string): boolean {
+  return DOCUMENT_PARSING_MODELS.has(model);
+}
+
+export function isVLModel(model: string): boolean {
+  return VL_MODELS.has(model);
+}
+
+export interface OCROptions {
+  useDocOrientationClassify?: boolean;
+  useDocUnwarping?: boolean;
+  useTextlineOrientation?: boolean;
+  textDetLimitSideLen?: number;
+  textDetLimitType?: string;
+  textDetThresh?: number;
+  textDetBoxThresh?: number;
+  textDetUnclipRatio?: number;
+  textRecScoreThresh?: number;
+  visualize?: boolean;
+  [key: string]: unknown;
+}
+
+export interface PPStructureV3Options {
+  useDocOrientationClassify?: boolean;
+  useDocUnwarping?: boolean;
+  useTextlineOrientation?: boolean;
+  useSealRecognition?: boolean;
+  useTableRecognition?: boolean;
+  useFormulaRecognition?: boolean;
+  useChartRecognition?: boolean;
+  useRegionDetection?: boolean;
+  layoutThreshold?: number | Record<string, number>;
+  layoutNms?: boolean;
+  layoutUnclipRatio?: number | number[] | Record<string, number>;
+  layoutMergeBboxesMode?: string | Record<string, string>;
+  formatBlockContent?: boolean;
+  textDetLimitSideLen?: number;
+  textDetLimitType?: string;
+  textDetThresh?: number;
+  textDetBoxThresh?: number;
+  textDetUnclipRatio?: number;
+  textRecScoreThresh?: number;
+  useWiredTableCellsTransToHtml?: boolean;
+  useWirelessTableCellsTransToHtml?: boolean;
+  useTableOrientationClassify?: boolean;
+  useOcrResultsWithTableCells?: boolean;
+  useE2eWiredTableRecModel?: boolean;
+  useE2eWirelessTableRecModel?: boolean;
+  markdownIgnoreLabels?: string[];
+  prettifyMarkdown?: boolean;
+  showFormulaNumber?: boolean;
+  returnMarkdownImages?: boolean;
+  outputFormats?: string[];
+  visualize?: boolean;
+  [key: string]: unknown;
+}
+
+export interface PaddleOCRVLOptions {
+  useDocOrientationClassify?: boolean;
+  useDocUnwarping?: boolean;
+  useLayoutDetection?: boolean;
+  useChartRecognition?: boolean;
+  useSealRecognition?: boolean;
+  useOcrForImageBlock?: boolean;
+  layoutThreshold?: number | Record<string, number>;
+  layoutNms?: boolean;
+  layoutUnclipRatio?: number | number[] | Record<string, number>;
+  layoutMergeBboxesMode?: string | Record<string, string>;
+  layoutShapeMode?: "rect" | "quad" | "poly" | "auto";
+  promptLabel?: "ocr" | "formula" | "table" | "chart" | "seal" | "spotting";
+  formatBlockContent?: boolean;
+  repetitionPenalty?: number;
+  temperature?: number;
+  topP?: number;
+  minPixels?: number;
+  maxPixels?: number;
+  maxNewTokens?: number;
+  vlmExtraArgs?: Record<string, unknown>;
+  mergeLayoutBlocks?: boolean;
+  markdownIgnoreLabels?: string[];
+  prettifyMarkdown?: boolean;
+  showFormulaNumber?: boolean;
+  restructurePages?: boolean;
+  mergeTables?: boolean;
+  relevelTitles?: boolean;
+  returnMarkdownImages?: boolean;
+  outputFormats?: string[];
+  visualize?: boolean;
+  [key: string]: unknown;
+}
+
+export type DocParsingOptions = PPStructureV3Options | PaddleOCRVLOptions;
+
+export interface OCRRequest {
+  model?: Model | string;
+  fileUrl?: string;
+  filePath?: string;
+  pageRanges?: string;
+  batchId?: string;
+  options?: OCROptions;
+}
+
+export interface DocParsingRequest {
+  model?: Model | string;
+  fileUrl?: string;
+  filePath?: string;
+  pageRanges?: string;
+  batchId?: string;
+  options?: DocParsingOptions;
+}
+
+export interface ClientOptions {
+  token?: string;
+  baseUrl?: string;
+  timeout?: number;
+  requestTimeout?: number;
+  pollTimeout?: number;
+  clientPlatform?: string;
+  fetch?: typeof fetch;
+}
+
+export interface SaveResourceOptions {
+  overwrite?: boolean;
+  filename?: string;
+}
diff --git a/api_sdk/typescript/src/results.ts b/api_sdk/typescript/src/results.ts
new file mode 100644
index 0000000..60c01ce
--- /dev/null
+++ b/api_sdk/typescript/src/results.ts
@@ -0,0 +1,72 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+export interface OCRPage {
+  prunedResult: unknown;
+  ocrImageUrl?: string;
+  docPreprocessingImageUrl?: string;
+  inputImageUrl?: string;
+  raw?: unknown;
+}
+
+export interface DocParsingPage {
+  markdownText: string;
+  markdownImages: Record<string, string>;
+  outputImages: Record<string, string>;
+  prunedResult?: unknown;
+  inputImageUrl?: string;
+  exports?: Record<string, unknown>;
+  markdown?: Record<string, unknown>;
+  raw?: unknown;
+}
+
+export interface OCRResult {
+  jobId: string;
+  pages: OCRPage[];
+  dataInfo?: Record<string, unknown>;
+}
+
+export interface DocParsingResult {
+  jobId: string;
+  pages: DocParsingPage[];
+  dataInfo?: Record<string, unknown>;
+}
+
+export interface Progress {
+  totalPages: number;
+  extractedPages: number;
+  startTime?: string;
+  endTime?: string;
+}
+
+export interface Job {
+  jobId: string;
+  model: string;
+  task: "ocr" | "document_parsing";
+  pageRanges?: string;
+  batchId?: string;
+}
+
+export interface JobStatus {
+  jobId: string;
+  state: "pending" | "running" | "done" | "failed";
+  progress?: Progress;
+  resultUrl?: Record<string, string>;
+  errorMsg?: string;
+}
+
+export interface BatchStatus {
+  batchId: string;
+  jobs: JobStatus[];
+}
diff --git a/api_sdk/typescript/tests/client.test.ts b/api_sdk/typescript/tests/client.test.ts
new file mode 100644
index 0000000..1167c67
--- /dev/null
+++ b/api_sdk/typescript/tests/client.test.ts
@@ -0,0 +1,703 @@
+import { mkdtemp, readFile, rm, stat, writeFile } from "node:fs/promises";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, describe, expect, test, vi } from "vitest";
+import {
+  APIError,
+  AuthError,
+  FileNotFoundError,
+  InvalidRequestError,
+  JobFailedError,
+  Model,
+  PaddleOCRClient,
+  PollTimeoutError,
+  RequestTimeoutError,
+  ResponseFormatError,
+  ResultParseError,
+} from "../src/index.js";
+import type { DocParsingResult, Job, OCRResult } from "../src/index.js";
+
+type FetchCall = {
+  url: string;
+  init: RequestInit;
+};
+
+function jsonResponse(body: unknown, status = 200): Response {
+  return new Response(JSON.stringify(body), {
+    status,
+    headers: { "Content-Type": "application/json" },
+  });
+}
+
+function textResponse(body: string, status = 200): Response {
+  return new Response(body, { status });
+}
+
+function createClient(fetchImpl: typeof fetch, options: Partial<ConstructorParameters<typeof PaddleOCRClient>[0]> = {}) {
+  return new PaddleOCRClient({
+    token: "test-token",
+    baseUrl: "https://api.example.test",
+    requestTimeout: 100,
+    pollTimeout: 100,
+    fetch: fetchImpl,
+    ...options,
+  });
+}
+
+function captureFetch(responses: Array<Response | Error | Promise<Response>>): { fetch: typeof fetch; calls: FetchCall[] } {
+  const calls: FetchCall[] = [];
+  const fetchImpl = vi.fn(async (input: RequestInfo | URL, init?: RequestInit) => {
+    calls.push({ url: String(input), init: init ?? {} });
+    const next = responses.shift();
+    if (next instanceof Error) {
+      throw next;
+    }
+    if (!next) {
+      throw new Error("No mocked response remaining");
+    }
+    return next;
+  }) as unknown as typeof fetch;
+  return { fetch: fetchImpl, calls };
+}
+
+afterEach(() => {
+  vi.useRealTimers();
+  vi.restoreAllMocks();
+});
+
+describe("PaddleOCRClient public contract", () => {
+  test("requires token at construction and reads PADDLEOCR_ACCESS_TOKEN fallback", () => {
+    const previous = process.env.PADDLEOCR_ACCESS_TOKEN;
+    delete process.env.PADDLEOCR_ACCESS_TOKEN;
+    expect(() => new PaddleOCRClient()).toThrow(AuthError);
+
+    process.env.PADDLEOCR_ACCESS_TOKEN = "env-token";
+    const { fetch } = captureFetch([jsonResponse({ data: { state: "running" } })]);
+    expect(() => new PaddleOCRClient({ baseUrl: "https://api.example.test", fetch })).not.toThrow();
+
+    if (previous === undefined) {
+      delete process.env.PADDLEOCR_ACCESS_TOKEN;
+    } else {
+      process.env.PADDLEOCR_ACCESS_TOKEN = previous;
+    }
+  });
+
+  test("exposes contract names and omits unpublished draft aliases", () => {
+    const { fetch } = captureFetch([]);
+    const client = createClient(fetch);
+    expect(client.getStatus).toBeTypeOf("function");
+    expect(client.parseDocument).toBeTypeOf("function");
+    expect(client.submitDocumentParsing).toBeTypeOf("function");
+    expect(client.waitOcrResult).toBeTypeOf("function");
+    expect(client.waitDocumentParsingResult).toBeTypeOf("function");
+    expect(client.saveResource).toBeTypeOf("function");
+    expect(client.saveOcrResultResources).toBeTypeOf("function");
+    expect(client.saveDocumentParsingResultResources).toBeTypeOf("function");
+
+    expect("getResult" in client).toBe(false);
+    expect("waitForResult" in client).toBe(false);
+    expect("docParsing" in client).toBe(false);
+    expect("submitDocParsing" in client).toBe(false);
+  });
+
+  test("submitOcr sends contract body and returns job metadata", async () => {
+    const { fetch, calls } = captureFetch([jsonResponse({ data: { jobId: "job-1" } })]);
+    const client = createClient(fetch);
+
+    const job = await client.submitOcr({
+      fileUrl: "https://files.example.test/invoice.pdf",
+      pageRanges: "1-2",
+      batchId: "batch-1",
+      options: { visualize: true },
+    });
+
+    expect(job).toEqual({
+      jobId: "job-1",
+      model: Model.PPOCRv6,
+      task: "ocr",
+      pageRanges: "1-2",
+      batchId: "batch-1",
+    });
+    expect(calls[0].url).toBe("https://api.example.test/api/v2/ocr/jobs");
+    expect(calls[0].init.method).toBe("POST");
+    expect(calls[0].init.headers).toMatchObject({
+      Authorization: "Bearer test-token",
+      "Content-Type": "application/json",
+    });
+    expect(JSON.parse(String(calls[0].init.body))).toEqual({
+      fileUrl: "https://files.example.test/invoice.pdf",
+      model: Model.PPOCRv6,
+      optionalPayload: { visualize: true },
+      pageRanges: "1-2",
+      batchId: "batch-1",
+    });
+  });
+
+  test("clientPlatform is sent as Client-Platform header on API requests", async () => {
+    const { fetch, calls } = captureFetch([jsonResponse({ data: { jobId: "job-1" } })]);
+    const client = createClient(fetch, { clientPlatform: "my-app" });
+
+    await client.submitOcr({
+      fileUrl: "https://files.example.test/invoice.pdf",
+    });
+
+    expect(calls[0].init.headers).toMatchObject({
+      "Client-Platform": "my-app",
+    });
+  });
+
+  test("submitOcr propagates explicit OCR model and defaults to PP-OCRv6", async () => {
+    const { fetch, calls } = captureFetch([
+      jsonResponse({ data: { jobId: "job-explicit" } }),
+      jsonResponse({ data: { jobId: "job-default" } }),
+    ]);
+    const client = createClient(fetch);
+
+    const explicit = await client.submitOcr({
+      model: Model.PPOCRv5,
+      fileUrl: "https://files.example.test/explicit.pdf",
+    });
+    const implicit = await client.submitOcr({
+      fileUrl: "https://files.example.test/default.pdf",
+    });
+
+    expect(explicit.model).toBe(Model.PPOCRv5);
+    expect(implicit.model).toBe(Model.PPOCRv6);
+    expect(JSON.parse(String(calls[0].init.body)).model).toBe(Model.PPOCRv5);
+    expect(JSON.parse(String(calls[1].init.body)).model).toBe(Model.PPOCRv6);
+  });
+
+  test("submitOcr accepts PP-OCRv6 model name", async () => {
+    const { fetch, calls } = captureFetch([jsonResponse({ data: { jobId: "job-v6" } })]);
+    const client = createClient(fetch);
+
+    const job = await client.submitOcr({
+      model: Model.PPOCRv6,
+      fileUrl: "https://files.example.test/v6.pdf",
+    });
+
+    expect(job.model).toBe(Model.PPOCRv6);
+    expect(JSON.parse(String(calls[0].init.body)).model).toBe(Model.PPOCRv6);
+  });
+
+  test("submitOcr accepts PP-OCRv5-latin model name", async () => {
+    const { fetch, calls } = captureFetch([jsonResponse({ data: { jobId: "job-latin" } })]);
+    const client = createClient(fetch);
+
+    const job = await client.submitOcr({
+      model: Model.PPOCRv5Latin,
+      fileUrl: "https://files.example.test/latin.pdf",
+    });
+
+    expect(job.model).toBe(Model.PPOCRv5Latin);
+    expect(JSON.parse(String(calls[0].init.body)).model).toBe(Model.PPOCRv5Latin);
+  });
+
+  test("submitOcr and submitDocumentParsing accept official model name strings", async () => {
+    const { fetch, calls } = captureFetch([
+      jsonResponse({ data: { jobId: "job-ocr" } }),
+      jsonResponse({ data: { jobId: "job-doc" } }),
+    ]);
+    const client = createClient(fetch);
+
+    const ocrJob = await client.submitOcr({
+      model: "PP-OCRv5",
+      fileUrl: "https://files.example.test/ocr.pdf",
+    });
+    const docJob = await client.submitDocumentParsing({
+      model: "PaddleOCR-VL-1.6",
+      fileUrl: "https://files.example.test/doc.pdf",
+    });
+
+    expect(ocrJob.model).toBe("PP-OCRv5");
+    expect(docJob.model).toBe("PaddleOCR-VL-1.6");
+    expect(JSON.parse(String(calls[0].init.body)).model).toBe("PP-OCRv5");
+    expect(JSON.parse(String(calls[1].init.body)).model).toBe("PaddleOCR-VL-1.6");
+  });
+
+  test("model helpers classify current OCR and document parsing models", async () => {
+    const mod = await import("../src/index.js");
+
+    expect(mod.isOCRModel(Model.PPOCRv5)).toBe(true);
+    expect(mod.isOCRModel(Model.PPOCRv5Latin)).toBe(true);
+    expect(mod.isOCRModel(Model.PPOCRv6)).toBe(true);
+    expect(mod.isOCRModel(Model.PPStructureV3)).toBe(false);
+    expect(mod.isOCRModel("future-unknown-model")).toBe(false);
+    expect(mod.isDocumentParsingModel(Model.PaddleOCRVL)).toBe(true);
+    expect(mod.isDocumentParsingModel(Model.PaddleOCRVL16)).toBe(true);
+    expect(mod.isDocumentParsingModel(Model.PPOCRv5)).toBe(false);
+  });
+
+  test("submitDocumentParsing passes current and future document parsing options", async () => {
+    const { fetch, calls } = captureFetch([jsonResponse({ data: { jobId: "job-doc" } })]);
+    const client = createClient(fetch);
+
+    await client.submitDocumentParsing({
+      model: Model.PaddleOCRVL16,
+      fileUrl: "https://files.example.test/doc.pdf",
+      options: {
+        useOcrForImageBlock: true,
+        formatBlockContent: true,
+        markdownIgnoreLabels: ["image"],
+        vlmExtraArgs: { temperature: 0.1 },
+        returnMarkdownImages: false,
+        outputFormats: ["docx"],
+        futureOption: "enabled",
+      },
+    });
+
+    expect(JSON.parse(String(calls[0].init.body)).optionalPayload).toEqual({
+      useOcrForImageBlock: true,
+      formatBlockContent: true,
+      markdownIgnoreLabels: ["image"],
+      vlmExtraArgs: { temperature: 0.1 },
+      returnMarkdownImages: false,
+      outputFormats: ["docx"],
+      futureOption: "enabled",
+    });
+  });
+
+  test("submitDocumentParsing defaults to PaddleOCR-VL-1.6", async () => {
+    const { fetch, calls } = captureFetch([jsonResponse({ data: { jobId: "job-doc" } })]);
+    const client = createClient(fetch);
+
+    const job = await client.submitDocumentParsing({
+      fileUrl: "https://files.example.test/doc.pdf",
+    });
+
+    expect(job.model).toBe(Model.PaddleOCRVL16);
+    expect(JSON.parse(String(calls[0].init.body)).model).toBe(Model.PaddleOCRVL16);
+  });
+
+  test("submitOcr rejects non-OCR models before network calls", async () => {
+    const { fetch } = captureFetch([]);
+    const client = createClient(fetch);
+
+    await expect(
+      client.submitOcr({
+        model: Model.PPStructureV3,
+        fileUrl: "https://files.example.test/not-ocr.pdf",
+      }),
+    ).rejects.toThrow(InvalidRequestError);
+    expect(fetch).not.toHaveBeenCalled();
+  });
+
+  test("validates submit requests before network calls", async () => {
+    const { fetch } = captureFetch([]);
+    const client = createClient(fetch);
+
+    await expect(client.submitOcr({})).rejects.toThrow(InvalidRequestError);
+    await expect(
+      client.submitOcr({ fileUrl: "https://files.example.test/a.pdf", filePath: "./a.pdf" }),
+    ).rejects.toThrow(InvalidRequestError);
+    expect(fetch).not.toHaveBeenCalled();
+  });
+
+  test("getStatus performs one non-blocking status request", async () => {
+    const { fetch, calls } = captureFetch([
+      jsonResponse({
+        data: {
+          state: "running",
+          extractProgress: { totalPages: 3, extractedPages: 1 },
+        },
+      }),
+    ]);
+    const client = createClient(fetch);
+
+    await expect(client.getStatus("job-1")).resolves.toEqual({
+      jobId: "job-1",
+      state: "running",
+      progress: { totalPages: 3, extractedPages: 1 },
+      errorMsg: undefined,
+    });
+    expect(calls).toHaveLength(1);
+    expect(calls[0].url).toBe("https://api.example.test/api/v2/ocr/jobs/job-1");
+  });
+
+  test("normalizes multiple trailing baseUrl slashes", async () => {
+    const { fetch, calls } = captureFetch([jsonResponse({ data: { state: "running" } })]);
+    const client = createClient(fetch, { baseUrl: "https://api.example.test///" });
+
+    await client.getStatus("job-1");
+
+    expect(calls[0].url).toBe("https://api.example.test/api/v2/ocr/jobs/job-1");
+  });
+
+  test("typed wait accepts bare jobId, fetches JSONL without Authorization, and parses OCR results", async () => {
+    const { fetch, calls } = captureFetch([
+      jsonResponse({ data: { state: "done", resultUrl: { jsonUrl: "https://storage.example.test/job-1.jsonl" } } }),
+      textResponse(
+        JSON.stringify({
+          result: {
+            dataInfo: { numPages: 1 },
+            ocrResults: [
+              {
+                prunedResult: { text: "hello" },
+                ocrImage: "img.png",
+                docPreprocessingImage: "pre.png",
+                inputImage: "input.png",
+              },
+            ],
+          },
+        }),
+      ),
+    ]);
+    const client = createClient(fetch);
+
+    await expect(client.waitOcrResult("job-1")).resolves.toMatchObject({
+      jobId: "job-1",
+      dataInfo: { numPages: 1 },
+      pages: [
+        {
+          prunedResult: { text: "hello" },
+          ocrImageUrl: "img.png",
+          docPreprocessingImageUrl: "pre.png",
+          inputImageUrl: "input.png",
+          raw: {
+            prunedResult: { text: "hello" },
+            ocrImage: "img.png",
+            docPreprocessingImage: "pre.png",
+            inputImage: "input.png",
+          },
+        },
+      ],
+    });
+    expect(calls[1].url).toBe("https://storage.example.test/job-1.jsonl");
+    expect((calls[1].init.headers as Record<string, string>).Authorization).toBeUndefined();
+  });
+
+  test("typed waits reject mismatched Job task before polling", async () => {
+    const { fetch } = captureFetch([]);
+    const client = createClient(fetch);
+    const docJob: Job = { jobId: "job-1", model: Model.PPStructureV3, task: "document_parsing" };
+
+    await expect(client.waitOcrResult(docJob)).rejects.toThrow(InvalidRequestError);
+    expect(fetch).not.toHaveBeenCalled();
+  });
+
+  test("typed waits reject mismatched Job model before polling", async () => {
+    const { fetch } = captureFetch([]);
+    const client = createClient(fetch);
+    const ocrTaskWithDocumentModel: Job = { jobId: "job-1", model: Model.PPStructureV3, task: "ocr" };
+    const documentTaskWithOcrModel: Job = { jobId: "job-2", model: Model.PPOCRv5, task: "document_parsing" };
+
+    await expect(client.waitOcrResult(ocrTaskWithDocumentModel)).rejects.toThrow(InvalidRequestError);
+    await expect(client.waitDocumentParsingResult(documentTaskWithOcrModel)).rejects.toThrow(InvalidRequestError);
+    expect(fetch).not.toHaveBeenCalled();
+  });
+
+  test("polling maps failed, timeout, unknown state, and missing result URL", async () => {
+    const failed = createClient(captureFetch([jsonResponse({ data: { state: "failed", errorMsg: "boom" } })]).fetch);
+    await expect(failed.waitOcrResult("job-1")).rejects.toThrow(JobFailedError);
+
+    const unknown = createClient(captureFetch([jsonResponse({ data: { state: "paused" } })]).fetch);
+    await expect(unknown.waitOcrResult("job-1")).rejects.toThrow(ResponseFormatError);
+
+    const missingResult = createClient(captureFetch([jsonResponse({ data: { state: "done" } })]).fetch);
+    await expect(missingResult.waitOcrResult("job-1")).rejects.toThrow(ResponseFormatError);
+
+    vi.useFakeTimers();
+    const timeoutFetch = vi.fn(async () => jsonResponse({ data: { state: "running" } })) as unknown as typeof fetch;
+    const timeout = createClient(timeoutFetch, { pollTimeout: 1 });
+    const promise = expect(timeout.waitOcrResult("job-1")).rejects.toThrow(PollTimeoutError);
+    await vi.runAllTimersAsync();
+    await promise;
+  });
+
+  test("malformed successful responses and malformed JSONL use dedicated errors", async () => {
+    const malformedSubmit = createClient(captureFetch([jsonResponse({ data: {} })]).fetch);
+    await expect(malformedSubmit.submitOcr({ fileUrl: "https://files.example.test/a.pdf" })).rejects.toThrow(
+      ResponseFormatError,
+    );
+
+    const malformedJsonl = createClient(
+      captureFetch([
+        jsonResponse({ data: { state: "done", resultUrl: { jsonUrl: "https://storage.example.test/job.jsonl" } } }),
+        textResponse("{not-json}\n"),
+      ]).fetch,
+    );
+    await expect(malformedJsonl.waitOcrResult("job-1")).rejects.toThrow(ResultParseError);
+  });
+
+  test("malformed fetched OCR result records use ResultParseError", async () => {
+    const client = createClient(
+      captureFetch([
+        jsonResponse({ data: { state: "done", resultUrl: { jsonUrl: "https://storage.example.test/job.jsonl" } } }),
+        textResponse(JSON.stringify({ notResult: {} })),
+      ]).fetch,
+    );
+
+    await expect(client.waitOcrResult("job-1")).rejects.toThrow(ResultParseError);
+  });
+
+  test.each([
+    ["missing ocrResults", { result: {} }],
+    ["missing prunedResult", { result: { ocrResults: [{ ocrImage: "img.png" }] } }],
+  ])("malformed OCR payload with %s uses ResultParseError", async (_name, payload) => {
+    const client = createClient(
+      captureFetch([
+        jsonResponse({ data: { state: "done", resultUrl: { jsonUrl: "https://storage.example.test/job.jsonl" } } }),
+        textResponse(JSON.stringify(payload)),
+      ]).fetch,
+    );
+
+    await expect(client.waitOcrResult("job-1")).rejects.toThrow(ResultParseError);
+  });
+
+  test("waitDocumentParsingResult preserves raw page data and dataInfo", async () => {
+    const page = {
+      prunedResult: { blocks: [{ label: "text", content: "hello" }] },
+      markdown: { text: "hello", images: { "figure.png": "https://example.test/figure.png" }, isStart: true, isEnd: true },
+      outputImages: { "page.png": "https://example.test/page.png" },
+      inputImage: "https://example.test/input.png",
+      exports: { docx: "https://example.test/result.docx" },
+    };
+    const { fetch } = captureFetch([
+      jsonResponse({ data: { state: "done", resultUrl: { jsonUrl: "https://storage.example.test/job.jsonl" } } }),
+      textResponse(JSON.stringify({ result: { dataInfo: { numPages: 1 }, layoutParsingResults: [page] } })),
+    ]);
+    const client = createClient(fetch);
+
+    await expect(client.waitDocumentParsingResult("job-1")).resolves.toMatchObject({
+      jobId: "job-1",
+      dataInfo: { numPages: 1 },
+      pages: [
+        {
+          markdownText: "hello",
+          markdownImages: { "figure.png": "https://example.test/figure.png" },
+          outputImages: { "page.png": "https://example.test/page.png" },
+          prunedResult: { blocks: [{ label: "text", content: "hello" }] },
+          inputImageUrl: "https://example.test/input.png",
+          exports: { docx: "https://example.test/result.docx" },
+          markdown: page.markdown,
+          raw: page,
+        },
+      ],
+    });
+  });
+
+  test("malformed fetched document result records use ResultParseError", async () => {
+    const client = createClient(
+      captureFetch([
+        jsonResponse({ data: { state: "done", resultUrl: { jsonUrl: "https://storage.example.test/job.jsonl" } } }),
+        textResponse(JSON.stringify({ notResult: {} })),
+      ]).fetch,
+    );
+
+    await expect(client.waitDocumentParsingResult("job-1")).rejects.toThrow(ResultParseError);
+  });
+
+  test.each([
+    ["missing layoutParsingResults", { result: {} }],
+    ["missing markdown", { result: { layoutParsingResults: [{}] } }],
+    ["missing markdown.text", { result: { layoutParsingResults: [{ markdown: { images: {} } }] } }],
+  ])("malformed document payload with %s uses ResultParseError", async (_name, payload) => {
+    const client = createClient(
+      captureFetch([
+        jsonResponse({ data: { state: "done", resultUrl: { jsonUrl: "https://storage.example.test/job.jsonl" } } }),
+        textResponse(JSON.stringify(payload)),
+      ]).fetch,
+    );
+
+    await expect(client.waitDocumentParsingResult("job-1")).rejects.toThrow(ResultParseError);
+  });
+
+  test("HTTP status and timeout failures map to contract errors", async () => {
+    await expect(createClient(captureFetch([textResponse("bad auth", 401)]).fetch).getStatus("job-1")).rejects.toThrow(
+      AuthError,
+    );
+    await expect(createClient(captureFetch([textResponse("bad request", 400)]).fetch).getStatus("job-1")).rejects.toThrow(
+      InvalidRequestError,
+    );
+    await expect(createClient(captureFetch([textResponse("server error", 500)]).fetch).getStatus("job-1")).rejects.toThrow(
+      APIError,
+    );
+
+    const timeoutFetch = vi.fn((_input: RequestInfo | URL, init?: RequestInit) =>
+      new Promise<Response>((_resolve, reject) => {
+        init?.signal?.addEventListener("abort", () => reject(new DOMException("aborted", "AbortError")));
+      }),
+    ) as unknown as typeof fetch;
+    await expect(createClient(timeoutFetch, { requestTimeout: 1 }).getStatus("job-1")).rejects.toThrow(
+      RequestTimeoutError,
+    );
+  });
+
+  test("submitFile reports missing local files before network calls", async () => {
+    const { fetch } = captureFetch([]);
+    const client = createClient(fetch);
+
+    await expect(client.submitOcr({ filePath: "/definitely/missing/file.pdf" })).rejects.toThrow(FileNotFoundError);
+    expect(fetch).not.toHaveBeenCalled();
+  });
+
+  test("saveResource writes URL downloads without auth and requires overwrite opt-in", async () => {
+    const dir = await mkdtemp(join(tmpdir(), "paddleocr-api-sdk-"));
+    try {
+      const { fetch, calls } = captureFetch([textResponse("content"), textResponse("replacement")]);
+      const client = createClient(fetch);
+
+      const saved = await client.saveResource("https://storage.example.test/a/b/c.png", dir);
+      expect(saved).toBe(join(dir, "c.png"));
+      await expect(readFile(join(dir, "c.png"), "utf8")).resolves.toBe("content");
+      expect((calls[0].init.headers as Record<string, string>).Authorization).toBeUndefined();
+
+      await expect(client.saveResource("https://storage.example.test/a/b/c.png", dir)).rejects.toThrow(
+        InvalidRequestError,
+      );
+
+      await client.saveResource("https://storage.example.test/a/b/c.png", dir, { overwrite: true });
+      await expect(readFile(join(dir, "c.png"), "utf8")).resolves.toBe("replacement");
+      await expect(stat(join(dir, "c.png"))).resolves.toBeTruthy();
+    } finally {
+      await rm(dir, { recursive: true, force: true });
+    }
+  });
+
+  test("saveOcrResultResources saves OCR result image URLs with stable page filenames", async () => {
+    const dir = await mkdtemp(join(tmpdir(), "paddleocr-api-sdk-"));
+    try {
+      const { fetch, calls } = captureFetch([textResponse("page 1"), textResponse("page 2")]);
+      const client = createClient(fetch);
+      const result: OCRResult = {
+        jobId: "job-ocr",
+        pages: [
+          { prunedResult: { text: "one" }, ocrImageUrl: "https://storage.example.test/results/page-a.png?sig=opaque" },
+          { prunedResult: { text: "two" }, ocrImageUrl: "https://storage.example.test/results/page-b.jpg" },
+        ],
+      };
+
+      const saved = await client.saveOcrResultResources(result, dir);
+
+      expect(saved).toEqual([join(dir, "ocr-page-1.png"), join(dir, "ocr-page-2.jpg")]);
+      await expect(readFile(join(dir, "ocr-page-1.png"), "utf8")).resolves.toBe("page 1");
+      await expect(readFile(join(dir, "ocr-page-2.jpg"), "utf8")).resolves.toBe("page 2");
+      expect(calls).toHaveLength(2);
+      expect(calls.map((call) => (call.init.headers as Record<string, string>).Authorization)).toEqual([
+        undefined,
+        undefined,
+      ]);
+    } finally {
+      await rm(dir, { recursive: true, force: true });
+    }
+  });
+
+  test("saveDocumentParsingResultResources preserves safe document parsing map keys over opaque URL basenames", async () => {
+    const dir = await mkdtemp(join(tmpdir(), "paddleocr-api-sdk-"));
+    try {
+      const { fetch } = captureFetch([textResponse("markdown image"), textResponse("output image")]);
+      const client = createClient(fetch);
+      const result: DocParsingResult = {
+        jobId: "job-doc",
+        pages: [
+          {
+            markdownText: "![figure](figure 1.png)",
+            markdownImages: {
+              "figure 1.png": "https://storage.example.test/download?id=markdown-image",
+            },
+            outputImages: {
+              "rendered-page.jpg": "https://storage.example.test/blob?id=output-image",
+            },
+          },
+        ],
+      };
+
+      const saved = await client.saveDocumentParsingResultResources(result, dir);
+
+      expect(saved).toEqual([join(dir, "figure 1.png"), join(dir, "rendered-page.jpg")]);
+      await expect(readFile(join(dir, "figure 1.png"), "utf8")).resolves.toBe("markdown image");
+      await expect(readFile(join(dir, "rendered-page.jpg"), "utf8")).resolves.toBe("output image");
+    } finally {
+      await rm(dir, { recursive: true, force: true });
+    }
+  });
+
+  test("saveDocumentParsingResultResources sorts document parsing map keys for deterministic output", async () => {
+    const dir = await mkdtemp(join(tmpdir(), "paddleocr-api-sdk-"));
+    try {
+      const { fetch, calls } = captureFetch([textResponse("alpha"), textResponse("bravo"), textResponse("charlie")]);
+      const client = createClient(fetch);
+      const result: DocParsingResult = {
+        jobId: "job-doc",
+        pages: [
+          {
+            markdownText: "",
+            markdownImages: {
+              "charlie.png": "https://storage.example.test/charlie",
+              "alpha.png": "https://storage.example.test/alpha",
+              "bravo.png": "https://storage.example.test/bravo",
+            },
+            outputImages: {},
+          },
+        ],
+      };
+
+      const saved = await client.saveDocumentParsingResultResources(result, dir);
+
+      expect(saved).toEqual([join(dir, "alpha.png"), join(dir, "bravo.png"), join(dir, "charlie.png")]);
+      expect(calls.map((call) => call.url)).toEqual([
+        "https://storage.example.test/alpha",
+        "https://storage.example.test/bravo",
+        "https://storage.example.test/charlie",
+      ]);
+      await expect(readFile(join(dir, "alpha.png"), "utf8")).resolves.toBe("alpha");
+      await expect(readFile(join(dir, "bravo.png"), "utf8")).resolves.toBe("bravo");
+      await expect(readFile(join(dir, "charlie.png"), "utf8")).resolves.toBe("charlie");
+    } finally {
+      await rm(dir, { recursive: true, force: true });
+    }
+  });
+
+  test("saveOcrResultResources requires overwrite opt-in for bulk result files", async () => {
+    const dir = await mkdtemp(join(tmpdir(), "paddleocr-api-sdk-"));
+    try {
+      const result: OCRResult = {
+        jobId: "job-ocr",
+        pages: [{ prunedResult: {}, ocrImageUrl: "https://storage.example.test/page.png" }],
+      };
+
+      await writeFile(join(dir, "ocr-page-1.png"), "existing");
+      const blocked = captureFetch([textResponse("unexpected")]);
+      await expect(createClient(blocked.fetch).saveOcrResultResources(result, dir)).rejects.toThrow(InvalidRequestError);
+      expect(blocked.calls).toHaveLength(0);
+
+      const replacement = captureFetch([textResponse("replacement")]);
+      await createClient(replacement.fetch).saveOcrResultResources(result, dir, { overwrite: true });
+      await expect(readFile(join(dir, "ocr-page-1.png"), "utf8")).resolves.toBe("replacement");
+    } finally {
+      await rm(dir, { recursive: true, force: true });
+    }
+  });
+
+  test.each([
+    ["empty", ""],
+    ["absolute", "/absolute.png"],
+    ["parent traversal", "../escape.png"],
+    ["traversal segment", ".."],
+    ["forward separator", "nested/escape.png"],
+    ["backslash separator", "nested\\escape.png"],
+  ])("saveDocumentParsingResultResources rejects unsafe document parsing map key: %s", async (_name, unsafeKey) => {
+    const dir = await mkdtemp(join(tmpdir(), "paddleocr-api-sdk-"));
+    try {
+      const { fetch } = captureFetch([]);
+      const client = createClient(fetch);
+      const result: DocParsingResult = {
+        jobId: "job-doc",
+        pages: [
+          {
+            markdownText: "",
+            markdownImages: {
+              [unsafeKey]: "https://storage.example.test/escape.png",
+            },
+            outputImages: {},
+          },
+        ],
+      };
+
+      await expect(client.saveDocumentParsingResultResources(result, dir)).rejects.toThrow(InvalidRequestError);
+      expect(fetch).not.toHaveBeenCalled();
+    } finally {
+      await rm(dir, { recursive: true, force: true });
+    }
+  });
+});
diff --git a/api_sdk/typescript/tsconfig.json b/api_sdk/typescript/tsconfig.json
new file mode 100644
index 0000000..9eb3140
--- /dev/null
+++ b/api_sdk/typescript/tsconfig.json
@@ -0,0 +1,21 @@
+{
+  "compilerOptions": {
+    "target": "ES2022",
+    "module": "ESNext",
+    "moduleResolution": "bundler",
+    "lib": ["ES2022", "DOM"],
+    "types": ["node"],
+    "outDir": "./dist",
+    "declaration": true,
+    "strict": true,
+    "esModuleInterop": true,
+    "baseUrl": ".",
+    "paths": {
+      "@paddleocr/api-sdk": ["src/index.ts"]
+    },
+    "skipLibCheck": true,
+    "forceConsistentCasingInFileNames": true
+  },
+  "include": ["src/**/*"],
+  "exclude": ["node_modules", "dist", "tests", "examples"]
+}
diff --git a/api_sdk/typescript/tsup.config.ts b/api_sdk/typescript/tsup.config.ts
new file mode 100644
index 0000000..328eeed
--- /dev/null
+++ b/api_sdk/typescript/tsup.config.ts
@@ -0,0 +1,9 @@
+import { defineConfig } from "tsup";
+
+export default defineConfig({
+  entry: ["src/index.ts"],
+  format: ["esm", "cjs"],
+  dts: true,
+  clean: true,
+  splitting: false,
+});
diff --git a/applications/README.md b/applications/README.md
new file mode 100644
index 0000000..a0a7595
--- /dev/null
+++ b/applications/README.md
@@ -0,0 +1 @@
+移步[docs](https://www.paddleocr.ai/v2.10.0/applications/overview.html)
diff --git a/awesome_projects.md b/awesome_projects.md
new file mode 100644
index 0000000..ac0c87c
--- /dev/null
+++ b/awesome_projects.md
@@ -0,0 +1,30 @@
+## 😃 Awesome projects based on PaddleOCR
+💗 PaddleOCR wouldn’t be where it is today without its incredible community! A massive 🙌 thank you 🙌 to all our longtime partners, new collaborators, and everyone who’s poured their passion into PaddleOCR — whether we’ve named you or not. Your support fuels our fire! 🔥
+| Project Name | Description |
+| ------------ | ----------- |
+| [Umi-OCR](https://github.com/hiroi-sora/Umi-OCR) <a href="https://github.com/hiroi-sora/Umi-OCR"><img src="https://img.shields.io/github/stars/hiroi-sora/Umi-OCR"></a>|Free, Open-source, Batch Offline OCR Software.|
+| [LearnOpenCV](http://github.com/spmallick/learnopencv) <a href="http://github.com/spmallick/learnopencv"><img src="https://img.shields.io/github/stars/spmallick/learnopencv"></a> | code for Computer Vision, Deep learning, and AI research articles.|
+| [OmniParser](https://github.com/microsoft/OmniParser)<a href="https://github.com/microsoft/OmniParser"><img src="https://img.shields.io/github/stars/microsoft/OmniParser"></a> |OmniParser: Screen Parsing tool for Pure Vision Based GUI Agent.|
+| [QAnything](https://github.com/netease-youdao/QAnything)<a href="https://github.com/netease-youdao/QAnything"><img src="https://img.shields.io/github/stars/netease-youdao/QAnything"></a> |Question and Answer based on Anything.|
+| [PaddleHub](https://github.com/PaddlePaddle/PaddleHub)<a href="https://github.com/PaddlePaddle/PaddleHub"><img src="https://img.shields.io/github/stars/PaddlePaddle/PaddleHub"></a> |400+ AI Models: Rich, high-quality AI models, including CV, NLP, Speech, Video and Cross-Modal.|
+| [PaddleNLP](https://github.com/PaddlePaddle/PaddleNLP)<a href="https://github.com/PaddlePaddle/PaddleNLP"><img src="https://img.shields.io/github/stars/PaddlePaddle/PaddleNLP"></a> |A Large Language Model (LLM) development suite based on the PaddlePaddle.|
+| [Rerun](https://github.com/rerun-io/rerun) <a href="https://github.com/rerun-io/rerun"><img src="https://img.shields.io/github/stars/rerun-io/rerun"></a> | Rerun is building the multimodal data stack to model, ingest, store, query and view robotics-style data |
+| [Dango-Translator](https://github.com/PantsuDango/Dango-Translator) <a href="https://github.com/PantsuDango/Dango-Translator"><img src="https://img.shields.io/github/stars/PantsuDango/Dango-Translator"></a> | Recognize text on the screen, translate it and show the translation results in real time.|
+| [PDF-Extract-Kit](https://github.com/opendatalab/PDF-Extract-Kit) <a href="https://github.com/opendatalab/PDF-Extract-Kit"><img src="https://img.shields.io/github/stars/opendatalab/PDF-Extract-Kit"></a> | PDF-Extract-Kit is a powerful open-source toolkit designed to efficiently extract high-quality content from complex and diverse PDF documents.  |
+| [manga-image-translator](https://github.com/zyddnys/manga-image-translator) <a href="https://github.com/zyddnys/manga-image-translator"><img src="https://img.shields.io/github/stars/zyddnys/manga-image-translator"></a> | Translate texts in manga/images.|
+| [March7thAssistant](https://github.com/moesnow/March7thAssistant) <a href="https://github.com/moesnow/March7thAssistant"><img src="https://img.shields.io/github/stars/moesnow/March7thAssistant"></a> | Daily Tasks: Stamina recovery, daily training, claiming rewards, commissions, and farming. |
+| [PaddlePaddle/models](https://github.com/PaddlePaddle/models) <a href="https://github.com/PaddlePaddle/models"><img src="https://img.shields.io/github/stars/PaddlePaddle/models"></a> |PaddlePaddle's industrial-grade model zoo.|
+| [katanaml/sparrow](https://github.com/katanaml/sparrow) <a href="https://github.com/katanaml/sparrow"><img src="https://img.shields.io/github/stars/katanaml/sparrow"></a> | Sparrow is an innovative open-source solution for efficient data extraction and processing from various documents and images. |
+| [RapidOCR](https://github.com/RapidAI/RapidOCR) <a href="https://github.com/RapidAI/RapidOCR"><img src="https://img.shields.io/github/stars/RapidAI/RapidOCR"></a> | Awesome OCR multiple programming languages toolkits based on ONNXRuntime, OpenVINO, PaddlePaddle and PyTorch |
+| [autoMate](https://github.com/yuruotong1/autoMate) <a href="https://github.com/yuruotong1/autoMate"><img src="https://img.shields.io/github/stars/yuruotong1/autoMate"></a> | AI-Powered Local Automation Tool & Let Your Computer Work for You. |
+| [Agent-S](https://github.com/simular-ai/Agent-S) <a href="https://github.com/simular-ai/Agent-S"><img src="https://img.shields.io/github/stars/simular-ai/Agent-S"></a> | A Compositional Generalist-Specialist Framework for Computer Use Agents.  |
+| [pdf-craft](https://github.com/oomol-lab/pdf-craft) <a href="https://github.com/oomol-lab/pdf-craft"><img src="https://img.shields.io/github/stars/oomol-lab/pdf-craft"></a> | PDF Craft can convert PDF files into various other formats. |
+| [VV](https://github.com/Cicada000/VV) <a href="https://github.com/Cicada000/VV"><img src="https://img.shields.io/github/stars/Cicada000/VV"></a> | Zhang Weiwei Quotations Search Project. |
+| [docetl](https://github.com/ucbepic/docetl) <a href="https://github.com/ucbepic/docetl"><img src="https://img.shields.io/github/stars/ucbepic/docetl"></a> | DocETL is a tool for creating and executing data processing pipelines, especially suited for complex document processing tasks. |
+| [ZenlessZoneZero-Auto](https://github.com/sMythicalBird/ZenlessZoneZero-Auto) <a href="https://github.com/sMythicalBird/ZenlessZoneZero-Auto"><img src="https://img.shields.io/github/stars/sMythicalBird/ZenlessZoneZero-Auto"></a> | Zenless Zone Zero Automation Framework. |
+| [Yuxi-Know](https://github.com/xerrors/Yuxi-Know) <a href="https://github.com/xerrors/Yuxi-Know"><img src="https://img.shields.io/github/stars/xerrors/Yuxi-Know"></a> | Knowledge graph question answering system based on LLMs. |
+| [PaddleSharp](https://github.com/sdcb/PaddleSharp) <a href="https://github.com/sdcb/PaddleSharp"><img src="https://img.shields.io/github/stars/sdcb/PaddleSharp"></a>|.NET/C# binding for Baidu paddle inference library and PaddleOCR |
+| [python-office](https://github.com/CoderWanFeng/python-office) <a href="https://github.com/CoderWanFeng/python-office"><img src="https://img.shields.io/github/stars/CoderWanFeng/python-office"></a> | Python tool for office works. |
+| [OnnxOCR](https://github.com/jingsongliujing/OnnxOCR) <a href="https://github.com/jingsongliujing/OnnxOCR"><img src="https://img.shields.io/github/stars/jingsongliujing/OnnxOCR"></a>|A lightweight OCR system based on PaddleOCR, decoupled from the PaddlePaddle deep learning training framework, with ultra-fast inference speed |
+| [Frigate](https://github.com/blakeblackshear/frigate) <a href="https://github.com/blakeblackshear/frigate"><img src="https://img.shields.io/github/stars/blakeblackshear/frigate"></a> | Real-time NVR system with AI-powered object detection and License Plate Recognition (LPR) using PaddleOCR. |
+| ... |... |
diff --git a/benchmark/PaddleOCR_DBNet/.gitattributes b/benchmark/PaddleOCR_DBNet/.gitattributes
new file mode 100644
index 0000000..b4419d4
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/.gitattributes
@@ -0,0 +1,2 @@
+*.html linguist-language=python
+*.ipynb linguist-language=python
diff --git a/benchmark/PaddleOCR_DBNet/.gitignore b/benchmark/PaddleOCR_DBNet/.gitignore
new file mode 100644
index 0000000..f18fe10
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/.gitignore
@@ -0,0 +1,16 @@
+.DS_Store
+*.pth
+*.pyc
+*.pyo
+*.log
+*.tmp
+*.pkl
+__pycache__/
+.idea/
+output/
+test/*.jpg
+datasets/
+index/
+train_log/
+log/
+profiling_log/
diff --git a/benchmark/PaddleOCR_DBNet/LICENSE.md b/benchmark/PaddleOCR_DBNet/LICENSE.md
new file mode 100644
index 0000000..b09cd78
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/LICENSE.md
@@ -0,0 +1,201 @@
+Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
diff --git a/benchmark/PaddleOCR_DBNet/README.MD b/benchmark/PaddleOCR_DBNet/README.MD
new file mode 100644
index 0000000..81e64bc
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/README.MD
@@ -0,0 +1,132 @@
+# Real-time Scene Text Detection with Differentiable Binarization
+
+**note**: some code is inherited from [WenmuZhou/DBNet.pytorch](https://github.com/WenmuZhou/DBNet.pytorch)
+
+[中文解读](https://zhuanlan.zhihu.com/p/94677957)
+
+![network](imgs/paper/db.jpg)
+
+## update 
+2020-06-07: 添加灰度图训练，训练灰度图时需要在配置里移除`dataset.args.transforms.Normalize`
+
+## Install Using Conda
+```
+conda env create -f environment.yml
+git clone https://github.com/WenmuZhou/DBNet.paddle.git
+cd DBNet.paddle/
+```
+
+or
+## Install Manually 
+```bash
+conda create -n dbnet python=3.6
+conda activate dbnet
+
+conda install ipython pip
+
+# python dependencies
+pip install -r requirement.txt
+
+# clone repo
+git clone https://github.com/WenmuZhou/DBNet.paddle.git
+cd DBNet.paddle/
+
+```
+
+## Requirements
+* paddlepaddle 2.4+
+
+## Download
+
+TBD
+
+## Data Preparation
+
+Training data: prepare a text `train.txt` in the following format, use '\t' as a separator
+```
+./datasets/train/img/001.jpg	./datasets/train/gt/001.txt
+```
+
+Validation data: prepare a text `test.txt` in the following format, use '\t' as a separator
+```
+./datasets/test/img/001.jpg	./datasets/test/gt/001.txt
+```
+- Store images in the `img` folder
+- Store groundtruth in the `gt` folder
+
+The groundtruth can be `.txt` files, with the following format:
+```
+x1, y1, x2, y2, x3, y3, x4, y4, annotation
+```
+
+
+## Train
+1. config the `dataset['train']['dataset'['data_path']'`,`dataset['validate']['dataset'['data_path']`in [config/icdar2015_resnet18_fpn_DBhead_polyLR.yaml](cconfig/icdar2015_resnet18_fpn_DBhead_polyLR.yaml)
+* . single gpu train
+```bash
+bash single_gpu_train.sh
+```
+* . Multi-gpu training
+```bash
+bash multi_gpu_train.sh
+```
+## Test
+
+[eval.py](tools/eval.py) is used to test model on test dataset
+
+1. config `model_path` in [eval.sh](eval.sh)
+2. use following script to test
+```bash
+bash eval.sh
+```
+
+## Predict 
+[predict.py](tools/predict.py) Can be used to inference on all images in a folder
+1. config `model_path`,`input_folder`,`output_folder` in [predict.sh](predict.sh)
+2. use following script to predict
+```
+bash predict.sh
+```
+You can change the `model_path` in the `predict.sh` file to your model location. 
+
+tips: if result is not good, you can change `thre` in [predict.sh](predict.sh) 
+    
+## Export Model
+
+[export_model.py](tools/export_model.py) Can be used to inference on all images in a folder
+
+use following script to export inference model
+```
+python tools/export_model.py --config_file config/icdar2015_resnet50_FPN_DBhead_polyLR.yaml -o trainer.resume_checkpoint=model_best.pth trainer.output_dir=output/infer
+```
+
+## Paddle Inference infer
+
+[infer.py](tools/infer.py) Can be used to inference on all images in a folder
+
+use following script to export inference model
+```
+python tools/infer.py --model-dir=output/infer/ --img-path imgs/paper/db.jpg 
+```
+
+<h2 id="Performance">Performance</h2>
+
+### [ICDAR 2015](http://rrc.cvc.uab.es/?ch=4)
+only train on ICDAR2015 dataset
+
+| Method                   | image size (short size) |learning rate | Precision (%) | Recall (%) | F-measure (%) | FPS |
+|:--------------------------:|:-------:|:--------:|:--------:|:------------:|:---------------:|:-----:|
+| ImageNet-resnet50-FPN-DBHead（torch）  |736 |1e-3|90.19 | 78.14 | 83.88 | 27 |
+| ImageNet-resnet50-FPN-DBHead（paddle）  |736 |1e-3| 89.47 | 79.03 | 83.92 | 27 |
+| ImageNet-resnet50-FPN-DBHead（paddle_amp）  |736 |1e-3| 88.62 | 79.95 | 84.06 | 27 |
+
+
+### examples
+TBD
+
+
+### reference
+1. https://arxiv.org/pdf/1911.08947.pdf
+2. https://github.com/WenmuZhou/DBNet.pytorch
+
+**If this repository helps you，please star it. Thanks.**
diff --git a/benchmark/PaddleOCR_DBNet/base/__init__.py b/benchmark/PaddleOCR_DBNet/base/__init__.py
new file mode 100644
index 0000000..5d7b417
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/base/__init__.py
@@ -0,0 +1,2 @@
+from .base_trainer import BaseTrainer
+from .base_dataset import BaseDataSet
diff --git a/benchmark/PaddleOCR_DBNet/base/base_dataset.py b/benchmark/PaddleOCR_DBNet/base/base_dataset.py
new file mode 100644
index 0000000..e9f6c4d
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/base/base_dataset.py
@@ -0,0 +1,86 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/12/4 13:12
+# @Author  : zhoujun
+import copy
+from paddle.io import Dataset
+from data_loader.modules import *
+
+
+class BaseDataSet(Dataset):
+    def __init__(
+        self,
+        data_path: str,
+        img_mode,
+        pre_processes,
+        filter_keys,
+        ignore_tags,
+        transform=None,
+        target_transform=None,
+    ):
+        assert img_mode in ["RGB", "BRG", "GRAY"]
+        self.ignore_tags = ignore_tags
+        self.data_list = self.load_data(data_path)
+        item_keys = ["img_path", "img_name", "text_polys", "texts", "ignore_tags"]
+        for item in item_keys:
+            assert (
+                item in self.data_list[0]
+            ), "data_list from load_data must contains {}".format(item_keys)
+        self.img_mode = img_mode
+        self.filter_keys = filter_keys
+        self.transform = transform
+        self.target_transform = target_transform
+        self._init_pre_processes(pre_processes)
+
+    def _init_pre_processes(self, pre_processes):
+        self.aug = []
+        if pre_processes is not None:
+            for aug in pre_processes:
+                if "args" not in aug:
+                    args = {}
+                else:
+                    args = aug["args"]
+                if isinstance(args, dict):
+                    cls = eval(aug["type"])(**args)
+                else:
+                    cls = eval(aug["type"])(args)
+                self.aug.append(cls)
+
+    def load_data(self, data_path: str) -> list:
+        """
+        把数据加载为一个list：
+        :params data_path: 存储数据的文件夹或者文件
+        return a dict ,包含了，'img_path','img_name','text_polys','texts','ignore_tags'
+        """
+        raise NotImplementedError
+
+    def apply_pre_processes(self, data):
+        for aug in self.aug:
+            data = aug(data)
+        return data
+
+    def __getitem__(self, index):
+        try:
+            data = copy.deepcopy(self.data_list[index])
+            im = cv2.imread(data["img_path"], 1 if self.img_mode != "GRAY" else 0)
+            if self.img_mode == "RGB":
+                im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB)
+            data["img"] = im
+            data["shape"] = [im.shape[0], im.shape[1]]
+            data = self.apply_pre_processes(data)
+
+            if self.transform:
+                data["img"] = self.transform(data["img"])
+            data["text_polys"] = data["text_polys"].tolist()
+            if len(self.filter_keys):
+                data_dict = {}
+                for k, v in data.items():
+                    if k not in self.filter_keys:
+                        data_dict[k] = v
+                return data_dict
+            else:
+                return data
+        except:
+            return self.__getitem__(np.random.randint(self.__len__()))
+
+    def __len__(self):
+        return len(self.data_list)
diff --git a/benchmark/PaddleOCR_DBNet/base/base_trainer.py b/benchmark/PaddleOCR_DBNet/base/base_trainer.py
new file mode 100644
index 0000000..f0d7f74
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/base/base_trainer.py
@@ -0,0 +1,269 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/8/23 21:50
+# @Author  : zhoujun
+
+import os
+import pathlib
+import shutil
+from pprint import pformat
+
+import anyconfig
+import paddle
+import numpy as np
+import random
+from paddle.jit import to_static
+from paddle.static import InputSpec
+
+from utils import setup_logger
+
+
+class BaseTrainer:
+    def __init__(
+        self,
+        config,
+        model,
+        criterion,
+        train_loader,
+        validate_loader,
+        metric_cls,
+        post_process=None,
+    ):
+        config["trainer"]["output_dir"] = os.path.join(
+            str(pathlib.Path(os.path.abspath(__name__)).parent),
+            config["trainer"]["output_dir"],
+        )
+        config["name"] = config["name"] + "_" + model.name
+        self.save_dir = config["trainer"]["output_dir"]
+        self.checkpoint_dir = os.path.join(self.save_dir, "checkpoint")
+
+        os.makedirs(self.checkpoint_dir, exist_ok=True)
+
+        self.global_step = 0
+        self.start_epoch = 0
+        self.config = config
+        self.criterion = criterion
+        # logger and tensorboard
+        self.visualdl_enable = self.config["trainer"].get("visual_dl", False)
+        self.epochs = self.config["trainer"]["epochs"]
+        self.log_iter = self.config["trainer"]["log_iter"]
+        if paddle.distributed.get_rank() == 0:
+            anyconfig.dump(config, os.path.join(self.save_dir, "config.yaml"))
+            self.logger = setup_logger(os.path.join(self.save_dir, "train.log"))
+            self.logger_info(pformat(self.config))
+
+        self.model = self.apply_to_static(model)
+
+        # device
+        if (
+            paddle.device.cuda.device_count() > 0
+            and paddle.device.is_compiled_with_cuda()
+        ):
+            self.with_cuda = True
+            random.seed(self.config["trainer"]["seed"])
+            np.random.seed(self.config["trainer"]["seed"])
+            paddle.seed(self.config["trainer"]["seed"])
+        else:
+            self.with_cuda = False
+        self.logger_info("train with and paddle {}".format(paddle.__version__))
+        # metrics
+        self.metrics = {
+            "recall": 0,
+            "precision": 0,
+            "hmean": 0,
+            "train_loss": float("inf"),
+            "best_model_epoch": 0,
+        }
+
+        self.train_loader = train_loader
+        if validate_loader is not None:
+            assert post_process is not None and metric_cls is not None
+        self.validate_loader = validate_loader
+        self.post_process = post_process
+        self.metric_cls = metric_cls
+        self.train_loader_len = len(train_loader)
+
+        if self.validate_loader is not None:
+            self.logger_info(
+                "train dataset has {} samples,{} in dataloader, validate dataset has {} samples,{} in dataloader".format(
+                    len(self.train_loader.dataset),
+                    self.train_loader_len,
+                    len(self.validate_loader.dataset),
+                    len(self.validate_loader),
+                )
+            )
+        else:
+            self.logger_info(
+                "train dataset has {} samples,{} in dataloader".format(
+                    len(self.train_loader.dataset), self.train_loader_len
+                )
+            )
+
+        self._initialize_scheduler()
+
+        self._initialize_optimizer()
+
+        # resume or finetune
+        if self.config["trainer"]["resume_checkpoint"] != "":
+            self._load_checkpoint(
+                self.config["trainer"]["resume_checkpoint"], resume=True
+            )
+        elif self.config["trainer"]["finetune_checkpoint"] != "":
+            self._load_checkpoint(
+                self.config["trainer"]["finetune_checkpoint"], resume=False
+            )
+
+        if self.visualdl_enable and paddle.distributed.get_rank() == 0:
+            from visualdl import LogWriter
+
+            self.writer = LogWriter(self.save_dir)
+
+        # 混合精度训练
+        self.amp = self.config.get("amp", None)
+        if self.amp == "None":
+            self.amp = None
+        if self.amp:
+            self.amp["scaler"] = paddle.amp.GradScaler(
+                init_loss_scaling=self.amp.get("scale_loss", 1024),
+                use_dynamic_loss_scaling=self.amp.get("use_dynamic_loss_scaling", True),
+            )
+            self.model, self.optimizer = paddle.amp.decorate(
+                models=self.model,
+                optimizers=self.optimizer,
+                level=self.amp.get("amp_level", "O2"),
+            )
+
+        # 分布式训练
+        if paddle.device.cuda.device_count() > 1:
+            self.model = paddle.DataParallel(self.model)
+        # make inverse Normalize
+        self.UN_Normalize = False
+        for t in self.config["dataset"]["train"]["dataset"]["args"]["transforms"]:
+            if t["type"] == "Normalize":
+                self.normalize_mean = t["args"]["mean"]
+                self.normalize_std = t["args"]["std"]
+                self.UN_Normalize = True
+
+    def apply_to_static(self, model):
+        support_to_static = self.config["trainer"].get("to_static", False)
+        if support_to_static:
+            specs = None
+            print("static")
+            specs = [InputSpec([None, 3, -1, -1])]
+            model = to_static(model, input_spec=specs)
+            self.logger_info(
+                "Successfully to apply @to_static with specs: {}".format(specs)
+            )
+        return model
+
+    def train(self):
+        """
+        Full training logic
+        """
+        for epoch in range(self.start_epoch + 1, self.epochs + 1):
+            self.epoch_result = self._train_epoch(epoch)
+            self._on_epoch_finish()
+        if paddle.distributed.get_rank() == 0 and self.visualdl_enable:
+            self.writer.close()
+        self._on_train_finish()
+
+    def _train_epoch(self, epoch):
+        """
+        Training logic for an epoch
+
+        :param epoch: Current epoch number
+        """
+        raise NotImplementedError
+
+    def _eval(self, epoch):
+        """
+        eval logic for an epoch
+
+        :param epoch: Current epoch number
+        """
+        raise NotImplementedError
+
+    def _on_epoch_finish(self):
+        raise NotImplementedError
+
+    def _on_train_finish(self):
+        raise NotImplementedError
+
+    def _save_checkpoint(self, epoch, file_name):
+        """
+        Saving checkpoints
+
+        :param epoch: current epoch number
+        :param log: logging information of the epoch
+        :param save_best: if True, rename the saved checkpoint to 'model_best.pth.tar'
+        """
+        state_dict = self.model.state_dict()
+        state = {
+            "epoch": epoch,
+            "global_step": self.global_step,
+            "state_dict": state_dict,
+            "optimizer": self.optimizer.state_dict(),
+            "config": self.config,
+            "metrics": self.metrics,
+        }
+        filename = os.path.join(self.checkpoint_dir, file_name)
+        paddle.save(state, filename)
+
+    def _load_checkpoint(self, checkpoint_path, resume):
+        """
+        Resume from saved checkpoints
+        :param checkpoint_path: Checkpoint path to be resumed
+        """
+        self.logger_info("Loading checkpoint: {} ...".format(checkpoint_path))
+        checkpoint = paddle.load(checkpoint_path)
+        self.model.set_state_dict(checkpoint["state_dict"])
+        if resume:
+            self.global_step = checkpoint["global_step"]
+            self.start_epoch = checkpoint["epoch"]
+            self.config["lr_scheduler"]["args"]["last_epoch"] = self.start_epoch
+            # self.scheduler.load_state_dict(checkpoint['scheduler'])
+            self.optimizer.set_state_dict(checkpoint["optimizer"])
+            if "metrics" in checkpoint:
+                self.metrics = checkpoint["metrics"]
+            self.logger_info(
+                "resume from checkpoint {} (epoch {})".format(
+                    checkpoint_path, self.start_epoch
+                )
+            )
+        else:
+            self.logger_info("finetune from checkpoint {}".format(checkpoint_path))
+
+    def _initialize(self, name, module, *args, **kwargs):
+        module_name = self.config[name]["type"]
+        module_args = self.config[name].get("args", {})
+        assert all(
+            [k not in module_args for k in kwargs]
+        ), "Overwriting kwargs given in config file is not allowed"
+        module_args.update(kwargs)
+        return getattr(module, module_name)(*args, **module_args)
+
+    def _initialize_scheduler(self):
+        self.lr_scheduler = self._initialize("lr_scheduler", paddle.optimizer.lr)
+
+    def _initialize_optimizer(self):
+        self.optimizer = self._initialize(
+            "optimizer",
+            paddle.optimizer,
+            parameters=self.model.parameters(),
+            learning_rate=self.lr_scheduler,
+        )
+
+    def inverse_normalize(self, batch_img):
+        if self.UN_Normalize:
+            batch_img[:, 0, :, :] = (
+                batch_img[:, 0, :, :] * self.normalize_std[0] + self.normalize_mean[0]
+            )
+            batch_img[:, 1, :, :] = (
+                batch_img[:, 1, :, :] * self.normalize_std[1] + self.normalize_mean[1]
+            )
+            batch_img[:, 2, :, :] = (
+                batch_img[:, 2, :, :] * self.normalize_std[2] + self.normalize_mean[2]
+            )
+
+    def logger_info(self, s):
+        if paddle.distributed.get_rank() == 0:
+            self.logger.info(s)
diff --git a/benchmark/PaddleOCR_DBNet/config/SynthText.yaml b/benchmark/PaddleOCR_DBNet/config/SynthText.yaml
new file mode 100644
index 0000000..8fd511c
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/config/SynthText.yaml
@@ -0,0 +1,40 @@
+name: DBNet
+dataset:
+  train:
+    dataset:
+      type: SynthTextDataset # 数据集类型
+      args:
+        data_path: ''# SynthTextDataset 根目录
+        pre_processes: # 数据的预处理过程，包含augment和标签制作
+          - type: IaaAugment # 使用imgaug进行变换
+            args:
+              - {'type':Fliplr, 'args':{'p':0.5}}
+              - {'type': Affine, 'args':{'rotate':[-10,10]}}
+              - {'type':Resize,'args':{'size':[0.5,3]}}
+          - type: EastRandomCropData
+            args:
+              size: [640,640]
+              max_tries: 50
+              keep_ratio: true
+          - type: MakeBorderMap
+            args:
+              shrink_ratio: 0.4
+          - type: MakeShrinkMap
+            args:
+              shrink_ratio: 0.4
+              min_text_size: 8
+        transforms: # 对图片进行的变换方式
+          - type: ToTensor
+            args: {}
+          - type: Normalize
+            args:
+              mean: [0.485, 0.456, 0.406]
+              std: [0.229, 0.224, 0.225]
+        img_mode: RGB
+        filter_keys: ['img_path','img_name','text_polys','texts','ignore_tags','shape'] # 返回数据之前，从数据字典里删除的key
+        ignore_tags: ['*', '###']
+    loader:
+      batch_size: 1
+      shuffle: true
+      num_workers: 0
+      collate_fn: ''
diff --git a/benchmark/PaddleOCR_DBNet/config/SynthText_resnet18_FPN_DBhead_polyLR.yaml b/benchmark/PaddleOCR_DBNet/config/SynthText_resnet18_FPN_DBhead_polyLR.yaml
new file mode 100644
index 0000000..c285d36
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/config/SynthText_resnet18_FPN_DBhead_polyLR.yaml
@@ -0,0 +1,65 @@
+name: DBNet
+base: ['config/SynthText.yaml']
+arch:
+  type: Model
+  backbone:
+    type: resnet18
+    pretrained: true
+  neck:
+    type: FPN
+    inner_channels: 256
+  head:
+    type: DBHead
+    out_channels: 2
+    k: 50
+post_processing:
+  type: SegDetectorRepresenter
+  args:
+    thresh: 0.3
+    box_thresh: 0.7
+    max_candidates: 1000
+    unclip_ratio: 1.5 # from paper
+metric:
+  type: QuadMetric
+  args:
+    is_output_polygon: false
+loss:
+  type: DBLoss
+  alpha: 1
+  beta: 10
+  ohem_ratio: 3
+optimizer:
+  type: Adam
+  args:
+    lr: 0.001
+    weight_decay: 0
+    amsgrad: true
+lr_scheduler:
+  type: WarmupPolyLR
+  args:
+    warmup_epoch: 3
+trainer:
+  seed: 2
+  epochs: 1200
+  log_iter: 10
+  show_images_iter: 50
+  resume_checkpoint: ''
+  finetune_checkpoint: ''
+  output_dir: output
+  visual_dl: false
+amp:
+    scale_loss: 1024
+    amp_level: O2
+    custom_white_list: []
+    custom_black_list: ['exp', 'sigmoid', 'concat']
+dataset:
+  train:
+    dataset:
+      args:
+        data_path: ./datasets/SynthText
+        img_mode: RGB
+    loader:
+      batch_size: 2
+      shuffle: true
+      num_workers: 6
+      collate_fn: ''
diff --git a/benchmark/PaddleOCR_DBNet/config/icdar2015.yaml b/benchmark/PaddleOCR_DBNet/config/icdar2015.yaml
new file mode 100644
index 0000000..4233d3e
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/config/icdar2015.yaml
@@ -0,0 +1,69 @@
+name: DBNet
+dataset:
+  train:
+    dataset:
+      type: ICDAR2015Dataset # 数据集类型
+      args:
+        data_path: # 一个存放 img_path \t gt_path的文件
+          - ''
+        pre_processes: # 数据的预处理过程，包含augment和标签制作
+          - type: IaaAugment # 使用imgaug进行变换
+            args:
+              - {'type':Fliplr, 'args':{'p':0.5}}
+              - {'type': Affine, 'args':{'rotate':[-10,10]}}
+              - {'type':Resize,'args':{'size':[0.5,3]}}
+          - type: EastRandomCropData
+            args:
+              size: [640,640]
+              max_tries: 50
+              keep_ratio: true
+          - type: MakeBorderMap
+            args:
+              shrink_ratio: 0.4
+              thresh_min: 0.3
+              thresh_max: 0.7
+          - type: MakeShrinkMap
+            args:
+              shrink_ratio: 0.4
+              min_text_size: 8
+        transforms: # 对图片进行的变换方式
+          - type: ToTensor
+            args: {}
+          - type: Normalize
+            args:
+              mean: [0.485, 0.456, 0.406]
+              std: [0.229, 0.224, 0.225]
+        img_mode: RGB
+        filter_keys: [img_path,img_name,text_polys,texts,ignore_tags,shape] # 返回数据之前，从数据字典里删除的key
+        ignore_tags: ['*', '###']
+    loader:
+      batch_size: 1
+      shuffle: true
+      num_workers: 0
+      collate_fn: ''
+  validate:
+    dataset:
+      type: ICDAR2015Dataset
+      args:
+        data_path:
+          - ''
+        pre_processes:
+          - type: ResizeShortSize
+            args:
+              short_size: 736
+              resize_text_polys: false
+        transforms:
+          - type: ToTensor
+            args: {}
+          - type: Normalize
+            args:
+              mean: [0.485, 0.456, 0.406]
+              std: [0.229, 0.224, 0.225]
+        img_mode: RGB
+        filter_keys: []
+        ignore_tags: ['*', '###']
+    loader:
+      batch_size: 1
+      shuffle: true
+      num_workers: 0
+      collate_fn: ICDARCollectFN
diff --git a/benchmark/PaddleOCR_DBNet/config/icdar2015_dcn_resnet18_FPN_DBhead_polyLR.yaml b/benchmark/PaddleOCR_DBNet/config/icdar2015_dcn_resnet18_FPN_DBhead_polyLR.yaml
new file mode 100644
index 0000000..3e24428
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/config/icdar2015_dcn_resnet18_FPN_DBhead_polyLR.yaml
@@ -0,0 +1,82 @@
+name: DBNet
+base: ['config/icdar2015.yaml']
+arch:
+  type: Model
+  backbone:
+    type: deformable_resnet18
+    pretrained: true
+  neck:
+    type: FPN
+    inner_channels: 256
+  head:
+    type: DBHead
+    out_channels: 2
+    k: 50
+post_processing:
+  type: SegDetectorRepresenter
+  args:
+    thresh: 0.3
+    box_thresh: 0.7
+    max_candidates: 1000
+    unclip_ratio: 1.5 # from paper
+metric:
+  type: QuadMetric
+  args:
+    is_output_polygon: false
+loss:
+  type: DBLoss
+  alpha: 1
+  beta: 10
+  ohem_ratio: 3
+optimizer:
+  type: Adam
+  args:
+    lr: 0.001
+    weight_decay: 0
+    amsgrad: true
+lr_scheduler:
+  type: WarmupPolyLR
+  args:
+    warmup_epoch: 3
+trainer:
+  seed: 2
+  epochs: 1200
+  log_iter: 10
+  show_images_iter: 50
+  resume_checkpoint: ''
+  finetune_checkpoint: ''
+  output_dir: output
+  visual_dl: false
+amp:
+    scale_loss: 1024
+    amp_level: O2
+    custom_white_list: []
+    custom_black_list: ['exp', 'sigmoid', 'concat']
+dataset:
+  train:
+    dataset:
+      args:
+        data_path:
+          - ./datasets/train.txt
+        img_mode: RGB
+    loader:
+      batch_size: 1
+      shuffle: true
+      num_workers: 6
+      collate_fn: ''
+  validate:
+    dataset:
+      args:
+        data_path:
+          - ./datasets/test.txt
+        pre_processes:
+          - type: ResizeShortSize
+            args:
+              short_size: 736
+              resize_text_polys: false
+        img_mode: RGB
+    loader:
+      batch_size: 1
+      shuffle: true
+      num_workers: 6
+      collate_fn: ICDARCollectFN
diff --git a/benchmark/PaddleOCR_DBNet/config/icdar2015_resnet18_FPN_DBhead_polyLR.yaml b/benchmark/PaddleOCR_DBNet/config/icdar2015_resnet18_FPN_DBhead_polyLR.yaml
new file mode 100644
index 0000000..62c392b
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/config/icdar2015_resnet18_FPN_DBhead_polyLR.yaml
@@ -0,0 +1,82 @@
+name: DBNet
+base: ['config/icdar2015.yaml']
+arch:
+  type: Model
+  backbone:
+    type: resnet18
+    pretrained: true
+  neck:
+    type: FPN
+    inner_channels: 256
+  head:
+    type: DBHead
+    out_channels: 2
+    k: 50
+post_processing:
+  type: SegDetectorRepresenter
+  args:
+    thresh: 0.3
+    box_thresh: 0.7
+    max_candidates: 1000
+    unclip_ratio: 1.5 # from paper
+metric:
+  type: QuadMetric
+  args:
+    is_output_polygon: false
+loss:
+  type: DBLoss
+  alpha: 1
+  beta: 10
+  ohem_ratio: 3
+optimizer:
+  type: Adam
+  args:
+    lr: 0.001
+    weight_decay: 0
+    amsgrad: true
+lr_scheduler:
+  type: WarmupPolyLR
+  args:
+    warmup_epoch: 3
+trainer:
+  seed: 2
+  epochs: 1200
+  log_iter: 10
+  show_images_iter: 50
+  resume_checkpoint: ''
+  finetune_checkpoint: ''
+  output_dir: output
+  visual_dl: false
+amp:
+    scale_loss: 1024
+    amp_level: O2
+    custom_white_list: []
+    custom_black_list: ['exp', 'sigmoid', 'concat']
+dataset:
+  train:
+    dataset:
+      args:
+        data_path:
+          - ./datasets/train.txt
+        img_mode: RGB
+    loader:
+      batch_size: 1
+      shuffle: true
+      num_workers: 6
+      collate_fn: ''
+  validate:
+    dataset:
+      args:
+        data_path:
+          - ./datasets/test.txt
+        pre_processes:
+          - type: ResizeShortSize
+            args:
+              short_size: 736
+              resize_text_polys: false
+        img_mode: RGB
+    loader:
+      batch_size: 1
+      shuffle: true
+      num_workers: 6
+      collate_fn: ICDARCollectFN
diff --git a/benchmark/PaddleOCR_DBNet/config/icdar2015_resnet18_FPN_DBhead_polyLR_finetune.yaml b/benchmark/PaddleOCR_DBNet/config/icdar2015_resnet18_FPN_DBhead_polyLR_finetune.yaml
new file mode 100644
index 0000000..9b018d5
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/config/icdar2015_resnet18_FPN_DBhead_polyLR_finetune.yaml
@@ -0,0 +1,83 @@
+name: DBNet
+base: ['config/icdar2015.yaml']
+arch:
+  type: Model
+  backbone:
+    type: resnet18
+    pretrained: true
+  neck:
+    type: FPN
+    inner_channels: 256
+  head:
+    type: DBHead
+    out_channels: 2
+    k: 50
+post_processing:
+  type: SegDetectorRepresenter
+  args:
+    thresh: 0.3
+    box_thresh: 0.7
+    max_candidates: 1000
+    unclip_ratio: 1.5 # from paper
+metric:
+  type: QuadMetric
+  args:
+    is_output_polygon: false
+loss:
+  type: DBLoss
+  alpha: 1
+  beta: 10
+  ohem_ratio: 3
+optimizer:
+  type: Adam
+  args:
+    lr: 0.001
+    weight_decay: 0
+    amsgrad: true
+lr_scheduler:
+  type: StepLR
+  args:
+    step_size: 10
+    gama: 0.8
+trainer:
+  seed: 2
+  epochs: 500
+  log_iter: 10
+  show_images_iter: 50
+  resume_checkpoint: ''
+  finetune_checkpoint: ''
+  output_dir: output
+  visual_dl: false
+amp:
+    scale_loss: 1024
+    amp_level: O2
+    custom_white_list: []
+    custom_black_list: ['exp', 'sigmoid', 'concat']
+dataset:
+  train:
+    dataset:
+      args:
+        data_path:
+          - ./datasets/train.txt
+        img_mode: RGB
+    loader:
+      batch_size: 1
+      shuffle: true
+      num_workers: 6
+      collate_fn: ''
+  validate:
+    dataset:
+      args:
+        data_path:
+          - ./datasets/test.txt
+        pre_processes:
+          - type: ResizeShortSize
+            args:
+              short_size: 736
+              resize_text_polys: false
+        img_mode: RGB
+    loader:
+      batch_size: 1
+      shuffle: true
+      num_workers: 6
+      collate_fn: ICDARCollectFN
diff --git a/benchmark/PaddleOCR_DBNet/config/icdar2015_resnet50_FPN_DBhead_polyLR.yaml b/benchmark/PaddleOCR_DBNet/config/icdar2015_resnet50_FPN_DBhead_polyLR.yaml
new file mode 100644
index 0000000..2a870fd
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/config/icdar2015_resnet50_FPN_DBhead_polyLR.yaml
@@ -0,0 +1,79 @@
+name: DBNet
+base: ['config/icdar2015.yaml']
+arch:
+  type: Model
+  backbone:
+    type: resnet50
+    pretrained: true
+  neck:
+    type: FPN
+    inner_channels: 256
+  head:
+    type: DBHead
+    out_channels: 2
+    k: 50
+post_processing:
+  type: SegDetectorRepresenter
+  args:
+    thresh: 0.3
+    box_thresh: 0.7
+    max_candidates: 1000
+    unclip_ratio: 1.5 # from paper
+metric:
+  type: QuadMetric
+  args:
+    is_output_polygon: false
+loss:
+  type: DBLoss
+  alpha: 1
+  beta: 10
+  ohem_ratio: 3
+optimizer:
+  type: Adam
+lr_scheduler:
+  type: Polynomial
+  args:
+    learning_rate: 0.001
+    warmup_epoch: 3
+trainer:
+  seed: 2
+  epochs: 1200
+  log_iter: 10
+  show_images_iter: 50
+  resume_checkpoint: ''
+  finetune_checkpoint: ''
+  output_dir: output/fp16_o2
+  visual_dl: false
+amp:
+    scale_loss: 1024
+    amp_level: O2
+    custom_white_list: []
+    custom_black_list: ['exp', 'sigmoid', 'concat']
+dataset:
+  train:
+    dataset:
+      args:
+        data_path:
+          - ./datasets/train.txt
+        img_mode: RGB
+    loader:
+      batch_size: 16
+      shuffle: true
+      num_workers: 6
+      collate_fn: ''
+  validate:
+    dataset:
+      args:
+        data_path:
+          - ./datasets/test.txt
+        pre_processes:
+          - type: ResizeShortSize
+            args:
+              short_size: 736
+              resize_text_polys: false
+        img_mode: RGB
+    loader:
+      batch_size: 1
+      shuffle: true
+      num_workers: 6
+      collate_fn: ICDARCollectFN
diff --git a/benchmark/PaddleOCR_DBNet/config/open_dataset.yaml b/benchmark/PaddleOCR_DBNet/config/open_dataset.yaml
new file mode 100644
index 0000000..05ece6e
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/config/open_dataset.yaml
@@ -0,0 +1,73 @@
+name: DBNet
+dataset:
+  train:
+    dataset:
+      type: DetDataset # 数据集类型
+      args:
+        data_path: # 一个存放 img_path \t gt_path的文件
+          - ''
+        pre_processes: # 数据的预处理过程，包含augment和标签制作
+          - type: IaaAugment # 使用imgaug进行变换
+            args:
+              - {'type':Fliplr, 'args':{'p':0.5}}
+              - {'type': Affine, 'args':{'rotate':[-10,10]}}
+              - {'type':Resize,'args':{'size':[0.5,3]}}
+          - type: EastRandomCropData
+            args:
+              size: [640,640]
+              max_tries: 50
+              keep_ratio: true
+          - type: MakeBorderMap
+            args:
+              shrink_ratio: 0.4
+              thresh_min: 0.3
+              thresh_max: 0.7
+          - type: MakeShrinkMap
+            args:
+              shrink_ratio: 0.4
+              min_text_size: 8
+        transforms: # 对图片进行的变换方式
+          - type: ToTensor
+            args: {}
+          - type: Normalize
+            args:
+              mean: [0.485, 0.456, 0.406]
+              std: [0.229, 0.224, 0.225]
+        img_mode: RGB
+        load_char_annotation: false
+        expand_one_char: false
+        filter_keys: [img_path,img_name,text_polys,texts,ignore_tags,shape] # 返回数据之前，从数据字典里删除的key
+        ignore_tags: ['*', '###']
+    loader:
+      batch_size: 1
+      shuffle: true
+      num_workers: 0
+      collate_fn: ''
+  validate:
+    dataset:
+      type: DetDataset
+      args:
+        data_path:
+          - ''
+        pre_processes:
+          - type: ResizeShortSize
+            args:
+              short_size: 736
+              resize_text_polys: false
+        transforms:
+          - type: ToTensor
+            args: {}
+          - type: Normalize
+            args:
+              mean: [0.485, 0.456, 0.406]
+              std: [0.229, 0.224, 0.225]
+        img_mode: RGB
+        load_char_annotation: false # 是否加载字符级标注
+        expand_one_char: false # 是否对只有一个字符的框进行宽度扩充，扩充后w = w+h
+        filter_keys: []
+        ignore_tags: ['*', '###']
+    loader:
+      batch_size: 1
+      shuffle: true
+      num_workers: 0
+      collate_fn: ICDARCollectFN
diff --git a/benchmark/PaddleOCR_DBNet/config/open_dataset_dcn_resnet50_FPN_DBhead_polyLR.yaml b/benchmark/PaddleOCR_DBNet/config/open_dataset_dcn_resnet50_FPN_DBhead_polyLR.yaml
new file mode 100644
index 0000000..6c81738
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/config/open_dataset_dcn_resnet50_FPN_DBhead_polyLR.yaml
@@ -0,0 +1,86 @@
+name: DBNet
+base: ['config/open_dataset.yaml']
+arch:
+  type: Model
+  backbone:
+    type: deformable_resnet18
+    pretrained: true
+  neck:
+    type: FPN
+    inner_channels: 256
+  head:
+    type: DBHead
+    out_channels: 2
+    k: 50
+post_processing:
+  type: SegDetectorRepresenter
+  args:
+    thresh: 0.3
+    box_thresh: 0.7
+    max_candidates: 1000
+    unclip_ratio: 1.5 # from paper
+metric:
+  type: QuadMetric
+  args:
+    is_output_polygon: false
+loss:
+  type: DBLoss
+  alpha: 1
+  beta: 10
+  ohem_ratio: 3
+optimizer:
+  type: Adam
+  args:
+    lr: 0.001
+    weight_decay: 0
+    amsgrad: true
+lr_scheduler:
+  type: WarmupPolyLR
+  args:
+    warmup_epoch: 3
+trainer:
+  seed: 2
+  epochs: 1200
+  log_iter: 1
+  show_images_iter: 1
+  resume_checkpoint: ''
+  finetune_checkpoint: ''
+  output_dir: output
+  visual_dl: false
+amp:
+    scale_loss: 1024
+    amp_level: O2
+    custom_white_list: []
+    custom_black_list: ['exp', 'sigmoid', 'concat']
+dataset:
+  train:
+    dataset:
+      args:
+        data_path:
+          - ./datasets/train.json
+        img_mode: RGB
+        load_char_annotation: false
+        expand_one_char: false
+    loader:
+      batch_size: 2
+      shuffle: true
+      num_workers: 6
+      collate_fn: ''
+  validate:
+    dataset:
+      args:
+        data_path:
+          - ./datasets/test.json
+        pre_processes:
+          - type: ResizeShortSize
+            args:
+              short_size: 736
+              resize_text_polys: false
+        img_mode: RGB
+        load_char_annotation: false
+        expand_one_char: false
+    loader:
+      batch_size: 1
+      shuffle: true
+      num_workers: 6
+      collate_fn: ICDARCollectFN
diff --git a/benchmark/PaddleOCR_DBNet/config/open_dataset_resnest50_FPN_DBhead_polyLR.yaml b/benchmark/PaddleOCR_DBNet/config/open_dataset_resnest50_FPN_DBhead_polyLR.yaml
new file mode 100644
index 0000000..d47ab06
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/config/open_dataset_resnest50_FPN_DBhead_polyLR.yaml
@@ -0,0 +1,86 @@
+name: DBNet
+base: ['config/open_dataset.yaml']
+arch:
+  type: Model
+  backbone:
+    type: resnest50
+    pretrained: true
+  neck:
+    type: FPN
+    inner_channels: 256
+  head:
+    type: DBHead
+    out_channels: 2
+    k: 50
+post_processing:
+  type: SegDetectorRepresenter
+  args:
+    thresh: 0.3
+    box_thresh: 0.7
+    max_candidates: 1000
+    unclip_ratio: 1.5 # from paper
+metric:
+  type: QuadMetric
+  args:
+    is_output_polygon: false
+loss:
+  type: DBLoss
+  alpha: 1
+  beta: 10
+  ohem_ratio: 3
+optimizer:
+  type: Adam
+  args:
+    lr: 0.001
+    weight_decay: 0
+    amsgrad: true
+lr_scheduler:
+  type: WarmupPolyLR
+  args:
+    warmup_epoch: 3
+trainer:
+  seed: 2
+  epochs: 1200
+  log_iter: 1
+  show_images_iter: 1
+  resume_checkpoint: ''
+  finetune_checkpoint: ''
+  output_dir: output
+  visual_dl: false
+amp:
+    scale_loss: 1024
+    amp_level: O2
+    custom_white_list: []
+    custom_black_list: ['exp', 'sigmoid', 'concat']
+dataset:
+  train:
+    dataset:
+      args:
+        data_path:
+          - ./datasets/train.json
+        img_mode: RGB
+        load_char_annotation: false
+        expand_one_char: false
+    loader:
+      batch_size: 2
+      shuffle: true
+      num_workers: 6
+      collate_fn: ''
+  validate:
+    dataset:
+      args:
+        data_path:
+          - ./datasets/test.json
+        pre_processes:
+          - type: ResizeShortSize
+            args:
+              short_size: 736
+              resize_text_polys: false
+        img_mode: RGB
+        load_char_annotation: false
+        expand_one_char: false
+    loader:
+      batch_size: 1
+      shuffle: true
+      num_workers: 6
+      collate_fn: ICDARCollectFN
diff --git a/benchmark/PaddleOCR_DBNet/config/open_dataset_resnet18_FPN_DBhead_polyLR.yaml b/benchmark/PaddleOCR_DBNet/config/open_dataset_resnet18_FPN_DBhead_polyLR.yaml
new file mode 100644
index 0000000..ff16ddb
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/config/open_dataset_resnet18_FPN_DBhead_polyLR.yaml
@@ -0,0 +1,93 @@
+name: DBNet
+base: ['config/open_dataset.yaml']
+arch:
+  type: Model
+  backbone:
+    type: resnet18
+    pretrained: true
+  neck:
+    type: FPN
+    inner_channels: 256
+  head:
+    type: DBHead
+    out_channels: 2
+    k: 50
+post_processing:
+  type: SegDetectorRepresenter
+  args:
+    thresh: 0.3
+    box_thresh: 0.7
+    max_candidates: 1000
+    unclip_ratio: 1.5 # from paper
+metric:
+  type: QuadMetric
+  args:
+    is_output_polygon: false
+loss:
+  type: DBLoss
+  alpha: 1
+  beta: 10
+  ohem_ratio: 3
+optimizer:
+  type: Adam
+  args:
+    lr: 0.001
+    weight_decay: 0
+    amsgrad: true
+lr_scheduler:
+  type: WarmupPolyLR
+  args:
+    warmup_epoch: 3
+trainer:
+  seed: 2
+  epochs: 1200
+  log_iter: 1
+  show_images_iter: 1
+  resume_checkpoint: ''
+  finetune_checkpoint: ''
+  output_dir: output
+  visual_dl: false
+amp:
+    scale_loss: 1024
+    amp_level: O2
+    custom_white_list: []
+    custom_black_list: ['exp', 'sigmoid', 'concat']
+dataset:
+  train:
+    dataset:
+      args:
+        data_path:
+          - ./datasets/train.json
+        transforms: # 对图片进行的变换方式
+          - type: ToTensor
+            args: {}
+          - type: Normalize
+            args:
+              mean: [0.485, 0.456, 0.406]
+              std: [0.229, 0.224, 0.225]
+        img_mode: RGB
+        load_char_annotation: false
+        expand_one_char: false
+    loader:
+      batch_size: 2
+      shuffle: true
+      num_workers: 6
+      collate_fn: ''
+  validate:
+    dataset:
+      args:
+        data_path:
+          - ./datasets/test.json
+        pre_processes:
+          - type: ResizeShortSize
+            args:
+              short_size: 736
+              resize_text_polys: false
+        img_mode: RGB
+        load_char_annotation: false
+        expand_one_char: false
+    loader:
+      batch_size: 1
+      shuffle: true
+      num_workers: 6
+      collate_fn: ICDARCollectFN
diff --git a/benchmark/PaddleOCR_DBNet/data_loader/__init__.py b/benchmark/PaddleOCR_DBNet/data_loader/__init__.py
new file mode 100644
index 0000000..96c473f
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/data_loader/__init__.py
@@ -0,0 +1,114 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/8/23 21:52
+# @Author  : zhoujun
+import copy
+
+import PIL
+import numpy as np
+import paddle
+from paddle.io import DataLoader, DistributedBatchSampler, BatchSampler
+
+from paddle.vision import transforms
+
+
+def get_dataset(data_path, module_name, transform, dataset_args):
+    """
+    获取训练dataset
+    :param data_path: dataset文件列表，每个文件内以如下格式存储 ‘path/to/img\tlabel’
+    :param module_name: 所使用的自定义dataset名称，目前只支持data_loaders.ImageDataset
+    :param transform: 该数据集使用的transforms
+    :param dataset_args: module_name的参数
+    :return: 如果data_path列表不为空，返回对于的ConcatDataset对象，否则None
+    """
+    from . import dataset
+
+    s_dataset = getattr(dataset, module_name)(
+        transform=transform, data_path=data_path, **dataset_args
+    )
+    return s_dataset
+
+
+def get_transforms(transforms_config):
+    tr_list = []
+    for item in transforms_config:
+        if "args" not in item:
+            args = {}
+        else:
+            args = item["args"]
+        cls = getattr(transforms, item["type"])(**args)
+        tr_list.append(cls)
+    tr_list = transforms.Compose(tr_list)
+    return tr_list
+
+
+class ICDARCollectFN:
+    def __init__(self, *args, **kwargs):
+        pass
+
+    def __call__(self, batch):
+        data_dict = {}
+        to_tensor_keys = []
+        for sample in batch:
+            for k, v in sample.items():
+                if k not in data_dict:
+                    data_dict[k] = []
+                if isinstance(v, (np.ndarray, paddle.Tensor, PIL.Image.Image)):
+                    if k not in to_tensor_keys:
+                        to_tensor_keys.append(k)
+                data_dict[k].append(v)
+        for k in to_tensor_keys:
+            data_dict[k] = paddle.stack(data_dict[k], 0)
+        return data_dict
+
+
+def get_dataloader(module_config, distributed=False):
+    if module_config is None:
+        return None
+    config = copy.deepcopy(module_config)
+    dataset_args = config["dataset"]["args"]
+    if "transforms" in dataset_args:
+        img_transforms = get_transforms(dataset_args.pop("transforms"))
+    else:
+        img_transforms = None
+    # 创建数据集
+    dataset_name = config["dataset"]["type"]
+    data_path = dataset_args.pop("data_path")
+    if data_path == None:
+        return None
+
+    data_path = [x for x in data_path if x is not None]
+    if len(data_path) == 0:
+        return None
+    if (
+        "collate_fn" not in config["loader"]
+        or config["loader"]["collate_fn"] is None
+        or len(config["loader"]["collate_fn"]) == 0
+    ):
+        config["loader"]["collate_fn"] = None
+    else:
+        config["loader"]["collate_fn"] = eval(config["loader"]["collate_fn"])()
+
+    _dataset = get_dataset(
+        data_path=data_path,
+        module_name=dataset_name,
+        transform=img_transforms,
+        dataset_args=dataset_args,
+    )
+    sampler = None
+    if distributed:
+        # 3）使用DistributedSampler
+        batch_sampler = DistributedBatchSampler(
+            dataset=_dataset,
+            batch_size=config["loader"].pop("batch_size"),
+            shuffle=config["loader"].pop("shuffle"),
+        )
+    else:
+        batch_sampler = BatchSampler(
+            dataset=_dataset,
+            batch_size=config["loader"].pop("batch_size"),
+            shuffle=config["loader"].pop("shuffle"),
+        )
+    loader = DataLoader(
+        dataset=_dataset, batch_sampler=batch_sampler, **config["loader"]
+    )
+    return loader
diff --git a/benchmark/PaddleOCR_DBNet/data_loader/dataset.py b/benchmark/PaddleOCR_DBNet/data_loader/dataset.py
new file mode 100644
index 0000000..5353ec8
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/data_loader/dataset.py
@@ -0,0 +1,190 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/8/23 21:54
+# @Author  : zhoujun
+import pathlib
+import os
+import cv2
+import numpy as np
+import scipy.io as sio
+from tqdm.auto import tqdm
+
+from base import BaseDataSet
+from utils import order_points_clockwise, get_datalist, load, expand_polygon
+
+
+class ICDAR2015Dataset(BaseDataSet):
+    def __init__(
+        self,
+        data_path: str,
+        img_mode,
+        pre_processes,
+        filter_keys,
+        ignore_tags,
+        transform=None,
+        **kwargs,
+    ):
+        super().__init__(
+            data_path, img_mode, pre_processes, filter_keys, ignore_tags, transform
+        )
+
+    def load_data(self, data_path: str) -> list:
+        data_list = get_datalist(data_path)
+        t_data_list = []
+        for img_path, label_path in data_list:
+            data = self._get_annotation(label_path)
+            if len(data["text_polys"]) > 0:
+                item = {"img_path": img_path, "img_name": pathlib.Path(img_path).stem}
+                item.update(data)
+                t_data_list.append(item)
+            else:
+                print("there is no suit bbox in {}".format(label_path))
+        return t_data_list
+
+    def _get_annotation(self, label_path: str) -> dict:
+        boxes = []
+        texts = []
+        ignores = []
+        with open(label_path, encoding="utf-8", mode="r") as f:
+            for line in f.readlines():
+                params = line.strip().strip("\ufeff").strip("\xef\xbb\xbf").split(",")
+                try:
+                    box = order_points_clockwise(
+                        np.array(list(map(float, params[:8]))).reshape(-1, 2)
+                    )
+                    if cv2.contourArea(box) > 0:
+                        boxes.append(box)
+                        label = params[8]
+                        texts.append(label)
+                        ignores.append(label in self.ignore_tags)
+                except:
+                    print("load label failed on {}".format(label_path))
+        data = {
+            "text_polys": np.array(boxes),
+            "texts": texts,
+            "ignore_tags": ignores,
+        }
+        return data
+
+
+class DetDataset(BaseDataSet):
+    def __init__(
+        self,
+        data_path: str,
+        img_mode,
+        pre_processes,
+        filter_keys,
+        ignore_tags,
+        transform=None,
+        **kwargs,
+    ):
+        self.load_char_annotation = kwargs["load_char_annotation"]
+        self.expand_one_char = kwargs["expand_one_char"]
+        super().__init__(
+            data_path, img_mode, pre_processes, filter_keys, ignore_tags, transform
+        )
+
+    def load_data(self, data_path: str) -> list:
+        """
+        从json文件中读取出 文本行的坐标和gt，字符的坐标和gt
+        :param data_path:
+        :return:
+        """
+        data_list = []
+        for path in data_path:
+            content = load(path)
+            for gt in tqdm(content["data_list"], desc="read file {}".format(path)):
+                img_path = os.path.join(content["data_root"], gt["img_name"])
+                polygons = []
+                texts = []
+                illegibility_list = []
+                language_list = []
+                for annotation in gt["annotations"]:
+                    if len(annotation["polygon"]) == 0 or len(annotation["text"]) == 0:
+                        continue
+                    if len(annotation["text"]) > 1 and self.expand_one_char:
+                        annotation["polygon"] = expand_polygon(annotation["polygon"])
+                    polygons.append(annotation["polygon"])
+                    texts.append(annotation["text"])
+                    illegibility_list.append(annotation["illegibility"])
+                    language_list.append(annotation["language"])
+                    if self.load_char_annotation:
+                        for char_annotation in annotation["chars"]:
+                            if (
+                                len(char_annotation["polygon"]) == 0
+                                or len(char_annotation["char"]) == 0
+                            ):
+                                continue
+                            polygons.append(char_annotation["polygon"])
+                            texts.append(char_annotation["char"])
+                            illegibility_list.append(char_annotation["illegibility"])
+                            language_list.append(char_annotation["language"])
+                data_list.append(
+                    {
+                        "img_path": img_path,
+                        "img_name": gt["img_name"],
+                        "text_polys": np.array(polygons),
+                        "texts": texts,
+                        "ignore_tags": illegibility_list,
+                    }
+                )
+        return data_list
+
+
+class SynthTextDataset(BaseDataSet):
+    def __init__(
+        self,
+        data_path: str,
+        img_mode,
+        pre_processes,
+        filter_keys,
+        transform=None,
+        **kwargs,
+    ):
+        self.transform = transform
+        self.dataRoot = pathlib.Path(data_path)
+        if not self.dataRoot.exists():
+            raise FileNotFoundError("Dataset folder is not exist.")
+
+        self.targetFilePath = self.dataRoot / "gt.mat"
+        if not self.targetFilePath.exists():
+            raise FileExistsError("Target file is not exist.")
+        targets = {}
+        sio.loadmat(
+            self.targetFilePath,
+            targets,
+            squeeze_me=True,
+            struct_as_record=False,
+            variable_names=["imnames", "wordBB", "txt"],
+        )
+
+        self.imageNames = targets["imnames"]
+        self.wordBBoxes = targets["wordBB"]
+        self.transcripts = targets["txt"]
+        super().__init__(data_path, img_mode, pre_processes, filter_keys, transform)
+
+    def load_data(self, data_path: str) -> list:
+        t_data_list = []
+        for imageName, wordBBoxes, texts in zip(
+            self.imageNames, self.wordBBoxes, self.transcripts
+        ):
+            item = {}
+            wordBBoxes = (
+                np.expand_dims(wordBBoxes, axis=2)
+                if (wordBBoxes.ndim == 2)
+                else wordBBoxes
+            )
+            _, _, numOfWords = wordBBoxes.shape
+            text_polys = wordBBoxes.reshape(
+                [8, numOfWords], order="F"
+            ).T  # num_words * 8
+            text_polys = text_polys.reshape(numOfWords, 4, 2)  # num_of_words * 4 * 2
+            transcripts = [word for line in texts for word in line.split()]
+            if numOfWords != len(transcripts):
+                continue
+            item["img_path"] = str(self.dataRoot / imageName)
+            item["img_name"] = (self.dataRoot / imageName).stem
+            item["text_polys"] = text_polys
+            item["texts"] = transcripts
+            item["ignore_tags"] = [x in self.ignore_tags for x in transcripts]
+            t_data_list.append(item)
+        return t_data_list
diff --git a/benchmark/PaddleOCR_DBNet/data_loader/modules/__init__.py b/benchmark/PaddleOCR_DBNet/data_loader/modules/__init__.py
new file mode 100644
index 0000000..bc055da
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/data_loader/modules/__init__.py
@@ -0,0 +1,8 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/12/4 10:53
+# @Author  : zhoujun
+from .iaa_augment import IaaAugment
+from .augment import *
+from .random_crop_data import EastRandomCropData, PSERandomCrop
+from .make_border_map import MakeBorderMap
+from .make_shrink_map import MakeShrinkMap
diff --git a/benchmark/PaddleOCR_DBNet/data_loader/modules/augment.py b/benchmark/PaddleOCR_DBNet/data_loader/modules/augment.py
new file mode 100644
index 0000000..3b08a02
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/data_loader/modules/augment.py
@@ -0,0 +1,308 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/8/23 21:52
+# @Author  : zhoujun
+
+import math
+import numbers
+import random
+
+import cv2
+import numpy as np
+from skimage.util import random_noise
+
+
+class RandomNoise:
+    def __init__(self, random_rate):
+        self.random_rate = random_rate
+
+    def __call__(self, data: dict):
+        """
+        对图片加噪声
+        :param data: {'img':,'text_polys':,'texts':,'ignore_tags':}
+        :return:
+        """
+        if random.random() > self.random_rate:
+            return data
+        data["img"] = (
+            random_noise(data["img"], mode="gaussian", clip=True) * 255
+        ).astype(data["img"].dtype)
+        return data
+
+
+class RandomScale:
+    def __init__(self, scales, random_rate):
+        """
+        :param scales: 尺度
+        :param random_rate: 随机系数
+        :return:
+        """
+        self.random_rate = random_rate
+        self.scales = scales
+
+    def __call__(self, data: dict) -> dict:
+        """
+        从scales中随机选择一个尺度，对图片和文本框进行缩放
+        :param data: {'img':,'text_polys':,'texts':,'ignore_tags':}
+        :return:
+        """
+        if random.random() > self.random_rate:
+            return data
+        im = data["img"]
+        text_polys = data["text_polys"]
+
+        tmp_text_polys = text_polys.copy()
+        rd_scale = float(np.random.choice(self.scales))
+        im = cv2.resize(im, dsize=None, fx=rd_scale, fy=rd_scale)
+        tmp_text_polys *= rd_scale
+
+        data["img"] = im
+        data["text_polys"] = tmp_text_polys
+        return data
+
+
+class RandomRotateImgBox:
+    def __init__(self, degrees, random_rate, same_size=False):
+        """
+        :param degrees: 角度，可以是一个数值或者list
+        :param random_rate: 随机系数
+        :param same_size: 是否保持和原图一样大
+        :return:
+        """
+        if isinstance(degrees, numbers.Number):
+            if degrees < 0:
+                raise ValueError("If degrees is a single number, it must be positive.")
+            degrees = (-degrees, degrees)
+        elif (
+            isinstance(degrees, list)
+            or isinstance(degrees, tuple)
+            or isinstance(degrees, np.ndarray)
+        ):
+            if len(degrees) != 2:
+                raise ValueError("If degrees is a sequence, it must be of len 2.")
+            degrees = degrees
+        else:
+            raise Exception("degrees must in Number or list or tuple or np.ndarray")
+        self.degrees = degrees
+        self.same_size = same_size
+        self.random_rate = random_rate
+
+    def __call__(self, data: dict) -> dict:
+        """
+        从scales中随机选择一个尺度，对图片和文本框进行缩放
+        :param data: {'img':,'text_polys':,'texts':,'ignore_tags':}
+        :return:
+        """
+        if random.random() > self.random_rate:
+            return data
+        im = data["img"]
+        text_polys = data["text_polys"]
+
+        # ---------------------- 旋转图像 ----------------------
+        w = im.shape[1]
+        h = im.shape[0]
+        angle = np.random.uniform(self.degrees[0], self.degrees[1])
+
+        if self.same_size:
+            nw = w
+            nh = h
+        else:
+            # 角度变弧度
+            rangle = np.deg2rad(angle)
+            # 计算旋转之后图像的w, h
+            nw = abs(np.sin(rangle) * h) + abs(np.cos(rangle) * w)
+            nh = abs(np.cos(rangle) * h) + abs(np.sin(rangle) * w)
+        # 构造仿射矩阵
+        rot_mat = cv2.getRotationMatrix2D((nw * 0.5, nh * 0.5), angle, 1)
+        # 计算原图中心点到新图中心点的偏移量
+        rot_move = np.dot(rot_mat, np.array([(nw - w) * 0.5, (nh - h) * 0.5, 0]))
+        # 更新仿射矩阵
+        rot_mat[0, 2] += rot_move[0]
+        rot_mat[1, 2] += rot_move[1]
+        # 仿射变换
+        rot_img = cv2.warpAffine(
+            im,
+            rot_mat,
+            (int(math.ceil(nw)), int(math.ceil(nh))),
+            flags=cv2.INTER_LANCZOS4,
+        )
+
+        # ---------------------- 矫正bbox坐标 ----------------------
+        # rot_mat是最终的旋转矩阵
+        # 获取原始bbox的四个中点，然后将这四个点转换到旋转后的坐标系下
+        rot_text_polys = list()
+        for bbox in text_polys:
+            point1 = np.dot(rot_mat, np.array([bbox[0, 0], bbox[0, 1], 1]))
+            point2 = np.dot(rot_mat, np.array([bbox[1, 0], bbox[1, 1], 1]))
+            point3 = np.dot(rot_mat, np.array([bbox[2, 0], bbox[2, 1], 1]))
+            point4 = np.dot(rot_mat, np.array([bbox[3, 0], bbox[3, 1], 1]))
+            rot_text_polys.append([point1, point2, point3, point4])
+        data["img"] = rot_img
+        data["text_polys"] = np.array(rot_text_polys)
+        return data
+
+
+class RandomResize:
+    def __init__(self, size, random_rate, keep_ratio=False):
+        """
+        :param input_size: resize尺寸,数字或者list的形式，如果为list形式，就是[w,h]
+        :param random_rate: 随机系数
+        :param keep_ratio: 是否保持长宽比
+        :return:
+        """
+        if isinstance(size, numbers.Number):
+            if size < 0:
+                raise ValueError(
+                    "If input_size is a single number, it must be positive."
+                )
+            size = (size, size)
+        elif (
+            isinstance(size, list)
+            or isinstance(size, tuple)
+            or isinstance(size, np.ndarray)
+        ):
+            if len(size) != 2:
+                raise ValueError("If input_size is a sequence, it must be of len 2.")
+            size = (size[0], size[1])
+        else:
+            raise Exception("input_size must in Number or list or tuple or np.ndarray")
+        self.size = size
+        self.keep_ratio = keep_ratio
+        self.random_rate = random_rate
+
+    def __call__(self, data: dict) -> dict:
+        """
+        从scales中随机选择一个尺度，对图片和文本框进行缩放
+        :param data: {'img':,'text_polys':,'texts':,'ignore_tags':}
+        :return:
+        """
+        if random.random() > self.random_rate:
+            return data
+        im = data["img"]
+        text_polys = data["text_polys"]
+
+        if self.keep_ratio:
+            # 将图片短边pad到和长边一样
+            h, w, c = im.shape
+            max_h = max(h, self.size[0])
+            max_w = max(w, self.size[1])
+            im_padded = np.zeros((max_h, max_w, c), dtype=np.uint8)
+            im_padded[:h, :w] = im.copy()
+            im = im_padded
+        text_polys = text_polys.astype(np.float32)
+        h, w, _ = im.shape
+        im = cv2.resize(im, self.size)
+        w_scale = self.size[0] / float(w)
+        h_scale = self.size[1] / float(h)
+        text_polys[:, :, 0] *= w_scale
+        text_polys[:, :, 1] *= h_scale
+
+        data["img"] = im
+        data["text_polys"] = text_polys
+        return data
+
+
+def resize_image(img, short_size):
+    height, width, _ = img.shape
+    if height < width:
+        new_height = short_size
+        new_width = new_height / height * width
+    else:
+        new_width = short_size
+        new_height = new_width / width * height
+    new_height = int(round(new_height / 32) * 32)
+    new_width = int(round(new_width / 32) * 32)
+    resized_img = cv2.resize(img, (new_width, new_height))
+    return resized_img, (new_width / width, new_height / height)
+
+
+class ResizeShortSize:
+    def __init__(self, short_size, resize_text_polys=True):
+        """
+        :param size: resize尺寸,数字或者list的形式，如果为list形式，就是[w,h]
+        :return:
+        """
+        self.short_size = short_size
+        self.resize_text_polys = resize_text_polys
+
+    def __call__(self, data: dict) -> dict:
+        """
+        对图片和文本框进行缩放
+        :param data: {'img':,'text_polys':,'texts':,'ignore_tags':}
+        :return:
+        """
+        im = data["img"]
+        text_polys = data["text_polys"]
+
+        h, w, _ = im.shape
+        short_edge = min(h, w)
+        if short_edge < self.short_size:
+            # 保证短边 >= short_size
+            scale = self.short_size / short_edge
+            im = cv2.resize(im, dsize=None, fx=scale, fy=scale)
+            scale = (scale, scale)
+            # im, scale = resize_image(im, self.short_size)
+            if self.resize_text_polys:
+                # text_polys *= scale
+                text_polys[:, 0] *= scale[0]
+                text_polys[:, 1] *= scale[1]
+
+        data["img"] = im
+        data["text_polys"] = text_polys
+        return data
+
+
+class HorizontalFlip:
+    def __init__(self, random_rate):
+        """
+
+        :param random_rate: 随机系数
+        """
+        self.random_rate = random_rate
+
+    def __call__(self, data: dict) -> dict:
+        """
+        从scales中随机选择一个尺度，对图片和文本框进行缩放
+        :param data: {'img':,'text_polys':,'texts':,'ignore_tags':}
+        :return:
+        """
+        if random.random() > self.random_rate:
+            return data
+        im = data["img"]
+        text_polys = data["text_polys"]
+
+        flip_text_polys = text_polys.copy()
+        flip_im = cv2.flip(im, 1)
+        h, w, _ = flip_im.shape
+        flip_text_polys[:, :, 0] = w - flip_text_polys[:, :, 0]
+
+        data["img"] = flip_im
+        data["text_polys"] = flip_text_polys
+        return data
+
+
+class VerticalFlip:
+    def __init__(self, random_rate):
+        """
+
+        :param random_rate: 随机系数
+        """
+        self.random_rate = random_rate
+
+    def __call__(self, data: dict) -> dict:
+        """
+        从scales中随机选择一个尺度，对图片和文本框进行缩放
+        :param data: {'img':,'text_polys':,'texts':,'ignore_tags':}
+        :return:
+        """
+        if random.random() > self.random_rate:
+            return data
+        im = data["img"]
+        text_polys = data["text_polys"]
+
+        flip_text_polys = text_polys.copy()
+        flip_im = cv2.flip(im, 0)
+        h, w, _ = flip_im.shape
+        flip_text_polys[:, :, 1] = h - flip_text_polys[:, :, 1]
+        data["img"] = flip_im
+        data["text_polys"] = flip_text_polys
+        return data
diff --git a/benchmark/PaddleOCR_DBNet/data_loader/modules/iaa_augment.py b/benchmark/PaddleOCR_DBNet/data_loader/modules/iaa_augment.py
new file mode 100644
index 0000000..1f0fa19
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/data_loader/modules/iaa_augment.py
@@ -0,0 +1,68 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/12/4 18:06
+# @Author  : zhoujun
+import numpy as np
+import imgaug
+import imgaug.augmenters as iaa
+
+
+class AugmenterBuilder(object):
+    def __init__(self):
+        pass
+
+    def build(self, args, root=True):
+        if args is None or len(args) == 0:
+            return None
+        elif isinstance(args, list):
+            if root:
+                sequence = [self.build(value, root=False) for value in args]
+                return iaa.Sequential(sequence)
+            else:
+                return getattr(iaa, args[0])(
+                    *[self.to_tuple_if_list(a) for a in args[1:]]
+                )
+        elif isinstance(args, dict):
+            cls = getattr(iaa, args["type"])
+            return cls(**{k: self.to_tuple_if_list(v) for k, v in args["args"].items()})
+        else:
+            raise RuntimeError("unknown augmenter arg: " + str(args))
+
+    def to_tuple_if_list(self, obj):
+        if isinstance(obj, list):
+            return tuple(obj)
+        return obj
+
+
+class IaaAugment:
+    def __init__(self, augmenter_args):
+        self.augmenter_args = augmenter_args
+        self.augmenter = AugmenterBuilder().build(self.augmenter_args)
+
+    def __call__(self, data):
+        image = data["img"]
+        shape = image.shape
+
+        if self.augmenter:
+            aug = self.augmenter.to_deterministic()
+            data["img"] = aug.augment_image(image)
+            data = self.may_augment_annotation(aug, data, shape)
+        return data
+
+    def may_augment_annotation(self, aug, data, shape):
+        if aug is None:
+            return data
+
+        line_polys = []
+        for poly in data["text_polys"]:
+            new_poly = self.may_augment_poly(aug, shape, poly)
+            line_polys.append(new_poly)
+        data["text_polys"] = np.array(line_polys)
+        return data
+
+    def may_augment_poly(self, aug, img_shape, poly):
+        keypoints = [imgaug.Keypoint(p[0], p[1]) for p in poly]
+        keypoints = aug.augment_keypoints(
+            [imgaug.KeypointsOnImage(keypoints, shape=img_shape)]
+        )[0].keypoints
+        poly = [(p.x, p.y) for p in keypoints]
+        return poly
diff --git a/benchmark/PaddleOCR_DBNet/data_loader/modules/make_border_map.py b/benchmark/PaddleOCR_DBNet/data_loader/modules/make_border_map.py
new file mode 100644
index 0000000..28b9ac9
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/data_loader/modules/make_border_map.py
@@ -0,0 +1,159 @@
+import cv2
+import numpy as np
+
+np.seterr(divide="ignore", invalid="ignore")
+import pyclipper
+from shapely.geometry import Polygon
+
+
+class MakeBorderMap:
+    def __init__(self, shrink_ratio=0.4, thresh_min=0.3, thresh_max=0.7):
+        self.shrink_ratio = shrink_ratio
+        self.thresh_min = thresh_min
+        self.thresh_max = thresh_max
+
+    def __call__(self, data: dict) -> dict:
+        """
+        从scales中随机选择一个尺度，对图片和文本框进行缩放
+        :param data: {'img':,'text_polys':,'texts':,'ignore_tags':}
+        :return:
+        """
+        im = data["img"]
+        text_polys = data["text_polys"]
+        ignore_tags = data["ignore_tags"]
+
+        canvas = np.zeros(im.shape[:2], dtype=np.float32)
+        mask = np.zeros(im.shape[:2], dtype=np.float32)
+
+        for i in range(len(text_polys)):
+            if ignore_tags[i]:
+                continue
+            self.draw_border_map(text_polys[i], canvas, mask=mask)
+        canvas = canvas * (self.thresh_max - self.thresh_min) + self.thresh_min
+
+        data["threshold_map"] = canvas
+        data["threshold_mask"] = mask
+        return data
+
+    def draw_border_map(self, polygon, canvas, mask):
+        polygon = np.array(polygon)
+        assert polygon.ndim == 2
+        assert polygon.shape[1] == 2
+
+        polygon_shape = Polygon(polygon)
+        if polygon_shape.area <= 0:
+            return
+        distance = (
+            polygon_shape.area
+            * (1 - np.power(self.shrink_ratio, 2))
+            / polygon_shape.length
+        )
+        subject = [tuple(l) for l in polygon]
+        padding = pyclipper.PyclipperOffset()
+        padding.AddPath(subject, pyclipper.JT_ROUND, pyclipper.ET_CLOSEDPOLYGON)
+
+        padded_polygon = np.array(padding.Execute(distance)[0])
+        cv2.fillPoly(mask, [padded_polygon.astype(np.int32)], 1.0)
+
+        xmin = padded_polygon[:, 0].min()
+        xmax = padded_polygon[:, 0].max()
+        ymin = padded_polygon[:, 1].min()
+        ymax = padded_polygon[:, 1].max()
+        width = xmax - xmin + 1
+        height = ymax - ymin + 1
+
+        polygon[:, 0] = polygon[:, 0] - xmin
+        polygon[:, 1] = polygon[:, 1] - ymin
+
+        xs = np.broadcast_to(
+            np.linspace(0, width - 1, num=width).reshape(1, width), (height, width)
+        )
+        ys = np.broadcast_to(
+            np.linspace(0, height - 1, num=height).reshape(height, 1), (height, width)
+        )
+
+        distance_map = np.zeros((polygon.shape[0], height, width), dtype=np.float32)
+        for i in range(polygon.shape[0]):
+            j = (i + 1) % polygon.shape[0]
+            absolute_distance = self.distance(xs, ys, polygon[i], polygon[j])
+            distance_map[i] = np.clip(absolute_distance / distance, 0, 1)
+        distance_map = distance_map.min(axis=0)
+
+        xmin_valid = min(max(0, xmin), canvas.shape[1] - 1)
+        xmax_valid = min(max(0, xmax), canvas.shape[1] - 1)
+        ymin_valid = min(max(0, ymin), canvas.shape[0] - 1)
+        ymax_valid = min(max(0, ymax), canvas.shape[0] - 1)
+        canvas[ymin_valid : ymax_valid + 1, xmin_valid : xmax_valid + 1] = np.fmax(
+            1
+            - distance_map[
+                ymin_valid - ymin : ymax_valid - ymax + height,
+                xmin_valid - xmin : xmax_valid - xmax + width,
+            ],
+            canvas[ymin_valid : ymax_valid + 1, xmin_valid : xmax_valid + 1],
+        )
+
+    def distance(self, xs, ys, point_1, point_2):
+        """
+        compute the distance from point to a line
+        ys: coordinates in the first axis
+        xs: coordinates in the second axis
+        point_1, point_2: (x, y), the end of the line
+        """
+        height, width = xs.shape[:2]
+        square_distance_1 = np.square(xs - point_1[0]) + np.square(ys - point_1[1])
+        square_distance_2 = np.square(xs - point_2[0]) + np.square(ys - point_2[1])
+        square_distance = np.square(point_1[0] - point_2[0]) + np.square(
+            point_1[1] - point_2[1]
+        )
+
+        cosin = (square_distance - square_distance_1 - square_distance_2) / (
+            2 * np.sqrt(square_distance_1 * square_distance_2)
+        )
+        square_sin = 1 - np.square(cosin)
+        square_sin = np.nan_to_num(square_sin)
+
+        result = np.sqrt(
+            square_distance_1 * square_distance_2 * square_sin / square_distance
+        )
+        result[cosin < 0] = np.sqrt(np.fmin(square_distance_1, square_distance_2))[
+            cosin < 0
+        ]
+        # self.extend_line(point_1, point_2, result)
+        return result
+
+    def extend_line(self, point_1, point_2, result):
+        ex_point_1 = (
+            int(
+                round(point_1[0] + (point_1[0] - point_2[0]) * (1 + self.shrink_ratio))
+            ),
+            int(
+                round(point_1[1] + (point_1[1] - point_2[1]) * (1 + self.shrink_ratio))
+            ),
+        )
+        cv2.line(
+            result,
+            tuple(ex_point_1),
+            tuple(point_1),
+            4096.0,
+            1,
+            lineType=cv2.LINE_AA,
+            shift=0,
+        )
+        ex_point_2 = (
+            int(
+                round(point_2[0] + (point_2[0] - point_1[0]) * (1 + self.shrink_ratio))
+            ),
+            int(
+                round(point_2[1] + (point_2[1] - point_1[1]) * (1 + self.shrink_ratio))
+            ),
+        )
+        cv2.line(
+            result,
+            tuple(ex_point_2),
+            tuple(point_2),
+            4096.0,
+            1,
+            lineType=cv2.LINE_AA,
+            shift=0,
+        )
+        return ex_point_1, ex_point_2
diff --git a/benchmark/PaddleOCR_DBNet/data_loader/modules/make_shrink_map.py b/benchmark/PaddleOCR_DBNet/data_loader/modules/make_shrink_map.py
new file mode 100644
index 0000000..000ffed
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/data_loader/modules/make_shrink_map.py
@@ -0,0 +1,129 @@
+import numpy as np
+import cv2
+
+
+def shrink_polygon_py(polygon, shrink_ratio):
+    """
+    对框进行缩放，返回去的比例为1/shrink_ratio 即可
+    """
+    cx = polygon[:, 0].mean()
+    cy = polygon[:, 1].mean()
+    polygon[:, 0] = cx + (polygon[:, 0] - cx) * shrink_ratio
+    polygon[:, 1] = cy + (polygon[:, 1] - cy) * shrink_ratio
+    return polygon
+
+
+def shrink_polygon_pyclipper(polygon, shrink_ratio):
+    from shapely.geometry import Polygon
+    import pyclipper
+
+    polygon_shape = Polygon(polygon)
+    distance = (
+        polygon_shape.area * (1 - np.power(shrink_ratio, 2)) / polygon_shape.length
+    )
+    subject = [tuple(l) for l in polygon]
+    padding = pyclipper.PyclipperOffset()
+    padding.AddPath(subject, pyclipper.JT_ROUND, pyclipper.ET_CLOSEDPOLYGON)
+    shrunk = padding.Execute(-distance)
+    if shrunk == []:
+        shrunk = np.array(shrunk)
+    else:
+        shrunk = np.array(shrunk[0]).reshape(-1, 2)
+    return shrunk
+
+
+class MakeShrinkMap:
+    r"""
+    Making binary mask from detection data with ICDAR format.
+    Typically following the process of class `MakeICDARData`.
+    """
+
+    def __init__(self, min_text_size=8, shrink_ratio=0.4, shrink_type="pyclipper"):
+        shrink_func_dict = {
+            "py": shrink_polygon_py,
+            "pyclipper": shrink_polygon_pyclipper,
+        }
+        self.shrink_func = shrink_func_dict[shrink_type]
+        self.min_text_size = min_text_size
+        self.shrink_ratio = shrink_ratio
+
+    def __call__(self, data: dict) -> dict:
+        """
+        从scales中随机选择一个尺度，对图片和文本框进行缩放
+        :param data: {'img':,'text_polys':,'texts':,'ignore_tags':}
+        :return:
+        """
+        image = data["img"]
+        text_polys = data["text_polys"]
+        ignore_tags = data["ignore_tags"]
+
+        h, w = image.shape[:2]
+        text_polys, ignore_tags = self.validate_polygons(text_polys, ignore_tags, h, w)
+        gt = np.zeros((h, w), dtype=np.float32)
+        mask = np.ones((h, w), dtype=np.float32)
+        for i in range(len(text_polys)):
+            polygon = text_polys[i]
+            height = max(polygon[:, 1]) - min(polygon[:, 1])
+            width = max(polygon[:, 0]) - min(polygon[:, 0])
+            if ignore_tags[i] or min(height, width) < self.min_text_size:
+                cv2.fillPoly(mask, polygon.astype(np.int32)[np.newaxis, :, :], 0)
+                ignore_tags[i] = True
+            else:
+                shrunk = self.shrink_func(polygon, self.shrink_ratio)
+                if shrunk.size == 0:
+                    cv2.fillPoly(mask, polygon.astype(np.int32)[np.newaxis, :, :], 0)
+                    ignore_tags[i] = True
+                    continue
+                cv2.fillPoly(gt, [shrunk.astype(np.int32)], 1)
+
+        data["shrink_map"] = gt
+        data["shrink_mask"] = mask
+        return data
+
+    def validate_polygons(self, polygons, ignore_tags, h, w):
+        """
+        polygons (numpy.array, required): of shape (num_instances, num_points, 2)
+        """
+        if len(polygons) == 0:
+            return polygons, ignore_tags
+        assert len(polygons) == len(ignore_tags)
+        for polygon in polygons:
+            polygon[:, 0] = np.clip(polygon[:, 0], 0, w - 1)
+            polygon[:, 1] = np.clip(polygon[:, 1], 0, h - 1)
+
+        for i in range(len(polygons)):
+            area = self.polygon_area(polygons[i])
+            if abs(area) < 1:
+                ignore_tags[i] = True
+            if area > 0:
+                polygons[i] = polygons[i][::-1, :]
+        return polygons, ignore_tags
+
+    def polygon_area(self, polygon):
+        return cv2.contourArea(polygon)
+        # edge = 0
+        # for i in range(polygon.shape[0]):
+        #     next_index = (i + 1) % polygon.shape[0]
+        #     edge += (polygon[next_index, 0] - polygon[i, 0]) * (polygon[next_index, 1] - polygon[i, 1])
+        #
+        # return edge / 2.
+
+
+if __name__ == "__main__":
+    from shapely.geometry import Polygon
+    import pyclipper
+
+    polygon = np.array([[0, 0], [100, 10], [100, 100], [10, 90]])
+    a = shrink_polygon_py(polygon, 0.4)
+    print(a)
+    print(shrink_polygon_py(a, 1 / 0.4))
+    b = shrink_polygon_pyclipper(polygon, 0.4)
+    print(b)
+    poly = Polygon(b)
+    distance = poly.area * 1.5 / poly.length
+    offset = pyclipper.PyclipperOffset()
+    offset.AddPath(b, pyclipper.JT_ROUND, pyclipper.ET_CLOSEDPOLYGON)
+    expanded = np.array(offset.Execute(distance))
+    bounding_box = cv2.minAreaRect(expanded)
+    points = cv2.boxPoints(bounding_box)
+    print(points)
diff --git a/benchmark/PaddleOCR_DBNet/data_loader/modules/random_crop_data.py b/benchmark/PaddleOCR_DBNet/data_loader/modules/random_crop_data.py
new file mode 100644
index 0000000..8c6f656
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/data_loader/modules/random_crop_data.py
@@ -0,0 +1,211 @@
+import random
+
+import cv2
+import numpy as np
+
+
+# random crop algorithm similar to https://github.com/argman/EAST
+class EastRandomCropData:
+    def __init__(
+        self,
+        size=(640, 640),
+        max_tries=50,
+        min_crop_side_ratio=0.1,
+        require_original_image=False,
+        keep_ratio=True,
+    ):
+        self.size = size
+        self.max_tries = max_tries
+        self.min_crop_side_ratio = min_crop_side_ratio
+        self.require_original_image = require_original_image
+        self.keep_ratio = keep_ratio
+
+    def __call__(self, data: dict) -> dict:
+        """
+        从scales中随机选择一个尺度，对图片和文本框进行缩放
+        :param data: {'img':,'text_polys':,'texts':,'ignore_tags':}
+        :return:
+        """
+        im = data["img"]
+        text_polys = data["text_polys"]
+        ignore_tags = data["ignore_tags"]
+        texts = data["texts"]
+        all_care_polys = [text_polys[i] for i, tag in enumerate(ignore_tags) if not tag]
+        # 计算crop区域
+        crop_x, crop_y, crop_w, crop_h = self.crop_area(im, all_care_polys)
+        # crop 图片 保持比例填充
+        scale_w = self.size[0] / crop_w
+        scale_h = self.size[1] / crop_h
+        scale = min(scale_w, scale_h)
+        h = int(crop_h * scale)
+        w = int(crop_w * scale)
+        if self.keep_ratio:
+            if len(im.shape) == 3:
+                padimg = np.zeros((self.size[1], self.size[0], im.shape[2]), im.dtype)
+            else:
+                padimg = np.zeros((self.size[1], self.size[0]), im.dtype)
+            padimg[:h, :w] = cv2.resize(
+                im[crop_y : crop_y + crop_h, crop_x : crop_x + crop_w], (w, h)
+            )
+            img = padimg
+        else:
+            img = cv2.resize(
+                im[crop_y : crop_y + crop_h, crop_x : crop_x + crop_w], tuple(self.size)
+            )
+        # crop 文本框
+        text_polys_crop = []
+        ignore_tags_crop = []
+        texts_crop = []
+        for poly, text, tag in zip(text_polys, texts, ignore_tags):
+            poly = ((poly - (crop_x, crop_y)) * scale).tolist()
+            if not self.is_poly_outside_rect(poly, 0, 0, w, h):
+                text_polys_crop.append(poly)
+                ignore_tags_crop.append(tag)
+                texts_crop.append(text)
+        data["img"] = img
+        data["text_polys"] = np.float32(text_polys_crop)
+        data["ignore_tags"] = ignore_tags_crop
+        data["texts"] = texts_crop
+        return data
+
+    def is_poly_in_rect(self, poly, x, y, w, h):
+        poly = np.array(poly)
+        if poly[:, 0].min() < x or poly[:, 0].max() > x + w:
+            return False
+        if poly[:, 1].min() < y or poly[:, 1].max() > y + h:
+            return False
+        return True
+
+    def is_poly_outside_rect(self, poly, x, y, w, h):
+        poly = np.array(poly)
+        if poly[:, 0].max() < x or poly[:, 0].min() > x + w:
+            return True
+        if poly[:, 1].max() < y or poly[:, 1].min() > y + h:
+            return True
+        return False
+
+    def split_regions(self, axis):
+        regions = []
+        min_axis = 0
+        for i in range(1, axis.shape[0]):
+            if axis[i] != axis[i - 1] + 1:
+                region = axis[min_axis:i]
+                min_axis = i
+                regions.append(region)
+        return regions
+
+    def random_select(self, axis, max_size):
+        xx = np.random.choice(axis, size=2)
+        xmin = np.min(xx)
+        xmax = np.max(xx)
+        xmin = np.clip(xmin, 0, max_size - 1)
+        xmax = np.clip(xmax, 0, max_size - 1)
+        return xmin, xmax
+
+    def region_wise_random_select(self, regions, max_size):
+        selected_index = list(np.random.choice(len(regions), 2))
+        selected_values = []
+        for index in selected_index:
+            axis = regions[index]
+            xx = int(np.random.choice(axis, size=1))
+            selected_values.append(xx)
+        xmin = min(selected_values)
+        xmax = max(selected_values)
+        return xmin, xmax
+
+    def crop_area(self, im, text_polys):
+        h, w = im.shape[:2]
+        h_array = np.zeros(h, dtype=np.int32)
+        w_array = np.zeros(w, dtype=np.int32)
+        for points in text_polys:
+            points = np.round(points, decimals=0).astype(np.int32)
+            minx = np.min(points[:, 0])
+            maxx = np.max(points[:, 0])
+            w_array[minx:maxx] = 1
+            miny = np.min(points[:, 1])
+            maxy = np.max(points[:, 1])
+            h_array[miny:maxy] = 1
+        # ensure the cropped area not across a text
+        h_axis = np.where(h_array == 0)[0]
+        w_axis = np.where(w_array == 0)[0]
+
+        if len(h_axis) == 0 or len(w_axis) == 0:
+            return 0, 0, w, h
+
+        h_regions = self.split_regions(h_axis)
+        w_regions = self.split_regions(w_axis)
+
+        for i in range(self.max_tries):
+            if len(w_regions) > 1:
+                xmin, xmax = self.region_wise_random_select(w_regions, w)
+            else:
+                xmin, xmax = self.random_select(w_axis, w)
+            if len(h_regions) > 1:
+                ymin, ymax = self.region_wise_random_select(h_regions, h)
+            else:
+                ymin, ymax = self.random_select(h_axis, h)
+
+            if (
+                xmax - xmin < self.min_crop_side_ratio * w
+                or ymax - ymin < self.min_crop_side_ratio * h
+            ):
+                # area too small
+                continue
+            num_poly_in_rect = 0
+            for poly in text_polys:
+                if not self.is_poly_outside_rect(
+                    poly, xmin, ymin, xmax - xmin, ymax - ymin
+                ):
+                    num_poly_in_rect += 1
+                    break
+
+            if num_poly_in_rect > 0:
+                return xmin, ymin, xmax - xmin, ymax - ymin
+
+        return 0, 0, w, h
+
+
+class PSERandomCrop:
+    def __init__(self, size):
+        self.size = size
+
+    def __call__(self, data):
+        imgs = data["imgs"]
+
+        h, w = imgs[0].shape[0:2]
+        th, tw = self.size
+        if w == tw and h == th:
+            return imgs
+
+        # label中存在文本实例，并且按照概率进行裁剪，使用threshold_label_map控制
+        if np.max(imgs[2]) > 0 and random.random() > 3 / 8:
+            # 文本实例的左上角点
+            tl = np.min(np.where(imgs[2] > 0), axis=1) - self.size
+            tl[tl < 0] = 0
+            # 文本实例的右下角点
+            br = np.max(np.where(imgs[2] > 0), axis=1) - self.size
+            br[br < 0] = 0
+            # 保证选到右下角点时，有足够的距离进行crop
+            br[0] = min(br[0], h - th)
+            br[1] = min(br[1], w - tw)
+
+            for _ in range(50000):
+                i = random.randint(tl[0], br[0])
+                j = random.randint(tl[1], br[1])
+                # 保证shrink_label_map有文本
+                if imgs[1][i : i + th, j : j + tw].sum() <= 0:
+                    continue
+                else:
+                    break
+        else:
+            i = random.randint(0, h - th)
+            j = random.randint(0, w - tw)
+
+        # return i, j, th, tw
+        for idx in range(len(imgs)):
+            if len(imgs[idx].shape) == 3:
+                imgs[idx] = imgs[idx][i : i + th, j : j + tw, :]
+            else:
+                imgs[idx] = imgs[idx][i : i + th, j : j + tw]
+        data["imgs"] = imgs
+        return data
diff --git a/benchmark/PaddleOCR_DBNet/environment.yml b/benchmark/PaddleOCR_DBNet/environment.yml
new file mode 100644
index 0000000..571dbf2
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/environment.yml
@@ -0,0 +1,21 @@
+name: dbnet
+channels:
+  - conda-forge
+  - defaults
+dependencies:
+  - anyconfig==0.9.10
+  - future==0.18.2
+  - imgaug==0.4.0
+  - matplotlib==3.1.2
+  - numpy==1.17.4
+  - opencv
+  - pyclipper
+  - PyYAML==5.2
+  - scikit-image==0.16.2
+  - Shapely==1.6.4
+  - tensorboard=2
+  - tqdm==4.40.1
+  - ipython
+  - pip
+  - pip:
+    - polygon3
diff --git a/benchmark/PaddleOCR_DBNet/eval.sh b/benchmark/PaddleOCR_DBNet/eval.sh
new file mode 100644
index 0000000..7520a73
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/eval.sh
@@ -0,0 +1 @@
+CUDA_VISIBLE_DEVICES=0 python3 tools/eval.py --model_path ''
diff --git a/benchmark/PaddleOCR_DBNet/generate_lists.sh b/benchmark/PaddleOCR_DBNet/generate_lists.sh
new file mode 100644
index 0000000..84f408c
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/generate_lists.sh
@@ -0,0 +1,17 @@
+#Only use if your file names of the images and txts are identical
+rm ./datasets/train_img.txt
+rm ./datasets/train_gt.txt
+rm ./datasets/test_img.txt
+rm ./datasets/test_gt.txt
+rm ./datasets/train.txt
+rm ./datasets/test.txt
+ls ./datasets/train/img/*.jpg > ./datasets/train_img.txt
+ls ./datasets/train/gt/*.txt > ./datasets/train_gt.txt
+ls ./datasets/test/img/*.jpg > ./datasets/test_img.txt
+ls ./datasets/test/gt/*.txt > ./datasets/test_gt.txt
+paste ./datasets/train_img.txt ./datasets/train_gt.txt > ./datasets/train.txt
+paste ./datasets/test_img.txt ./datasets/test_gt.txt > ./datasets/test.txt
+rm ./datasets/train_img.txt
+rm ./datasets/train_gt.txt
+rm ./datasets/test_img.txt
+rm ./datasets/test_gt.txt
diff --git a/benchmark/PaddleOCR_DBNet/imgs/paper/db.jpg b/benchmark/PaddleOCR_DBNet/imgs/paper/db.jpg
new file mode 100644
index 0000000..aa6c7e9
Binary files /dev/null and b/benchmark/PaddleOCR_DBNet/imgs/paper/db.jpg differ
diff --git a/benchmark/PaddleOCR_DBNet/models/__init__.py b/benchmark/PaddleOCR_DBNet/models/__init__.py
new file mode 100644
index 0000000..a2669bf
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/models/__init__.py
@@ -0,0 +1,22 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/8/23 21:55
+# @Author  : zhoujun
+import copy
+from .model import Model
+from .losses import build_loss
+
+__all__ = ["build_loss", "build_model"]
+support_model = ["Model"]
+
+
+def build_model(config):
+    """
+    get architecture model class
+    """
+    copy_config = copy.deepcopy(config)
+    arch_type = copy_config.pop("type")
+    assert (
+        arch_type in support_model
+    ), f"{arch_type} is not developed yet!, only {support_model} are support now"
+    arch_model = eval(arch_type)(copy_config)
+    return arch_model
diff --git a/benchmark/PaddleOCR_DBNet/models/backbone/__init__.py b/benchmark/PaddleOCR_DBNet/models/backbone/__init__.py
new file mode 100644
index 0000000..5d0d8a2
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/models/backbone/__init__.py
@@ -0,0 +1,25 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/8/23 21:54
+# @Author  : zhoujun
+
+from .resnet import *
+
+__all__ = ["build_backbone"]
+
+support_backbone = [
+    "resnet18",
+    "deformable_resnet18",
+    "deformable_resnet50",
+    "resnet50",
+    "resnet34",
+    "resnet101",
+    "resnet152",
+]
+
+
+def build_backbone(backbone_name, **kwargs):
+    assert (
+        backbone_name in support_backbone
+    ), f"all support backbone is {support_backbone}"
+    backbone = eval(backbone_name)(**kwargs)
+    return backbone
diff --git a/benchmark/PaddleOCR_DBNet/models/backbone/resnet.py b/benchmark/PaddleOCR_DBNet/models/backbone/resnet.py
new file mode 100644
index 0000000..9aa886d
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/models/backbone/resnet.py
@@ -0,0 +1,366 @@
+import math
+import paddle
+from paddle import nn
+
+BatchNorm2d = nn.BatchNorm2D
+
+__all__ = [
+    "ResNet",
+    "resnet18",
+    "resnet34",
+    "resnet50",
+    "resnet101",
+    "deformable_resnet18",
+    "deformable_resnet50",
+    "resnet152",
+]
+
+model_urls = {
+    "resnet18": "https://download.pytorch.org/models/resnet18-5c106cde.pth",
+    "resnet34": "https://download.pytorch.org/models/resnet34-333f7ec4.pth",
+    "resnet50": "https://download.pytorch.org/models/resnet50-19c8e357.pth",
+    "resnet101": "https://download.pytorch.org/models/resnet101-5d3b4d8f.pth",
+    "resnet152": "https://download.pytorch.org/models/resnet152-b121ed2d.pth",
+}
+
+
+def constant_init(module, constant, bias=0):
+    module.weight = paddle.create_parameter(
+        shape=module.weight.shape,
+        dtype="float32",
+        default_initializer=paddle.nn.initializer.Constant(constant),
+    )
+    if hasattr(module, "bias"):
+        module.bias = paddle.create_parameter(
+            shape=module.bias.shape,
+            dtype="float32",
+            default_initializer=paddle.nn.initializer.Constant(bias),
+        )
+
+
+def conv3x3(in_planes, out_planes, stride=1):
+    """3x3 convolution with padding"""
+    return nn.Conv2D(
+        in_planes, out_planes, kernel_size=3, stride=stride, padding=1, bias_attr=False
+    )
+
+
+class BasicBlock(nn.Layer):
+    expansion = 1
+
+    def __init__(self, inplanes, planes, stride=1, downsample=None, dcn=None):
+        super(BasicBlock, self).__init__()
+        self.with_dcn = dcn is not None
+        self.conv1 = conv3x3(inplanes, planes, stride)
+        self.bn1 = BatchNorm2d(planes, momentum=0.1)
+        self.relu = nn.ReLU()
+        self.with_modulated_dcn = False
+        if not self.with_dcn:
+            self.conv2 = nn.Conv2D(
+                planes, planes, kernel_size=3, padding=1, bias_attr=False
+            )
+        else:
+            from paddle.vision.ops import DeformConv2D
+
+            deformable_groups = dcn.get("deformable_groups", 1)
+            offset_channels = 18
+            self.conv2_offset = nn.Conv2D(
+                planes, deformable_groups * offset_channels, kernel_size=3, padding=1
+            )
+            self.conv2 = DeformConv2D(
+                planes, planes, kernel_size=3, padding=1, bias_attr=False
+            )
+        self.bn2 = BatchNorm2d(planes, momentum=0.1)
+        self.downsample = downsample
+        self.stride = stride
+
+    def forward(self, x):
+        residual = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        # out = self.conv2(out)
+        if not self.with_dcn:
+            out = self.conv2(out)
+        else:
+            offset = self.conv2_offset(out)
+            out = self.conv2(out, offset)
+        out = self.bn2(out)
+
+        if self.downsample is not None:
+            residual = self.downsample(x)
+
+        out += residual
+        out = self.relu(out)
+
+        return out
+
+
+class Bottleneck(nn.Layer):
+    expansion = 4
+
+    def __init__(self, inplanes, planes, stride=1, downsample=None, dcn=None):
+        super(Bottleneck, self).__init__()
+        self.with_dcn = dcn is not None
+        self.conv1 = nn.Conv2D(inplanes, planes, kernel_size=1, bias_attr=False)
+        self.bn1 = BatchNorm2d(planes, momentum=0.1)
+        self.with_modulated_dcn = False
+        if not self.with_dcn:
+            self.conv2 = nn.Conv2D(
+                planes, planes, kernel_size=3, stride=stride, padding=1, bias_attr=False
+            )
+        else:
+            deformable_groups = dcn.get("deformable_groups", 1)
+            from paddle.vision.ops import DeformConv2D
+
+            offset_channels = 18
+            self.conv2_offset = nn.Conv2D(
+                planes,
+                deformable_groups * offset_channels,
+                stride=stride,
+                kernel_size=3,
+                padding=1,
+            )
+            self.conv2 = DeformConv2D(
+                planes, planes, kernel_size=3, padding=1, stride=stride, bias_attr=False
+            )
+        self.bn2 = BatchNorm2d(planes, momentum=0.1)
+        self.conv3 = nn.Conv2D(planes, planes * 4, kernel_size=1, bias_attr=False)
+        self.bn3 = BatchNorm2d(planes * 4, momentum=0.1)
+        self.relu = nn.ReLU()
+        self.downsample = downsample
+        self.stride = stride
+        self.dcn = dcn
+        self.with_dcn = dcn is not None
+
+    def forward(self, x):
+        residual = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        # out = self.conv2(out)
+        if not self.with_dcn:
+            out = self.conv2(out)
+        else:
+            offset = self.conv2_offset(out)
+            out = self.conv2(out, offset)
+        out = self.bn2(out)
+        out = self.relu(out)
+
+        out = self.conv3(out)
+        out = self.bn3(out)
+
+        if self.downsample is not None:
+            residual = self.downsample(x)
+
+        out += residual
+        out = self.relu(out)
+
+        return out
+
+
+class ResNet(nn.Layer):
+    def __init__(self, block, layers, in_channels=3, dcn=None):
+        self.dcn = dcn
+        self.inplanes = 64
+        super(ResNet, self).__init__()
+        self.out_channels = []
+        self.conv1 = nn.Conv2D(
+            in_channels, 64, kernel_size=7, stride=2, padding=3, bias_attr=False
+        )
+        self.bn1 = BatchNorm2d(64, momentum=0.1)
+        self.relu = nn.ReLU()
+        self.maxpool = nn.MaxPool2D(kernel_size=3, stride=2, padding=1)
+        self.layer1 = self._make_layer(block, 64, layers[0])
+        self.layer2 = self._make_layer(block, 128, layers[1], stride=2, dcn=dcn)
+        self.layer3 = self._make_layer(block, 256, layers[2], stride=2, dcn=dcn)
+        self.layer4 = self._make_layer(block, 512, layers[3], stride=2, dcn=dcn)
+
+        if self.dcn is not None:
+            for m in self.modules():
+                if isinstance(m, Bottleneck) or isinstance(m, BasicBlock):
+                    if hasattr(m, "conv2_offset"):
+                        constant_init(m.conv2_offset, 0)
+
+    def _make_layer(self, block, planes, blocks, stride=1, dcn=None):
+        downsample = None
+        if stride != 1 or self.inplanes != planes * block.expansion:
+            downsample = nn.Sequential(
+                nn.Conv2D(
+                    self.inplanes,
+                    planes * block.expansion,
+                    kernel_size=1,
+                    stride=stride,
+                    bias_attr=False,
+                ),
+                BatchNorm2d(planes * block.expansion, momentum=0.1),
+            )
+
+        layers = []
+        layers.append(block(self.inplanes, planes, stride, downsample, dcn=dcn))
+        self.inplanes = planes * block.expansion
+        for i in range(1, blocks):
+            layers.append(block(self.inplanes, planes, dcn=dcn))
+        self.out_channels.append(planes * block.expansion)
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = self.relu(x)
+        x = self.maxpool(x)
+
+        x2 = self.layer1(x)
+        x3 = self.layer2(x2)
+        x4 = self.layer3(x3)
+        x5 = self.layer4(x4)
+
+        return x2, x3, x4, x5
+
+
+def load_torch_params(paddle_model, torch_patams):
+    paddle_params = paddle_model.state_dict()
+
+    fc_names = ["classifier"]
+    for key, torch_value in torch_patams.items():
+        if "num_batches_tracked" in key:
+            continue
+        key = (
+            key.replace("running_var", "_variance")
+            .replace("running_mean", "_mean")
+            .replace("module.", "")
+        )
+        torch_value = torch_value.detach().cpu().numpy()
+        if key in paddle_params:
+            flag = [i in key for i in fc_names]
+            if any(flag) and "weight" in key:  # ignore bias
+                new_shape = [1, 0] + list(range(2, torch_value.ndim))
+                print(
+                    f"name: {key}, ori shape: {torch_value.shape}, new shape: {torch_value.transpose(new_shape).shape}"
+                )
+                torch_value = torch_value.transpose(new_shape)
+            paddle_params[key] = torch_value
+        else:
+            print(f"{key} not in paddle")
+    paddle_model.set_state_dict(paddle_params)
+
+
+def load_models(model, model_name):
+    import torch.utils.model_zoo as model_zoo
+
+    torch_patams = model_zoo.load_url(model_urls[model_name])
+    load_torch_params(model, torch_patams)
+
+
+def resnet18(pretrained=True, **kwargs):
+    """Constructs a ResNet-18 model.
+    Args:
+        pretrained (bool): If True, returns a model pre-trained on ImageNet
+    """
+    model = ResNet(BasicBlock, [2, 2, 2, 2], **kwargs)
+    if pretrained:
+        assert (
+            kwargs.get("in_channels", 3) == 3
+        ), "in_channels must be 3 when pretrained is True"
+        print("load from imagenet")
+        load_models(model, "resnet18")
+    return model
+
+
+def deformable_resnet18(pretrained=True, **kwargs):
+    """Constructs a ResNet-18 model.
+    Args:
+        pretrained (bool): If True, returns a model pre-trained on ImageNet
+    """
+    model = ResNet(BasicBlock, [2, 2, 2, 2], dcn=dict(deformable_groups=1), **kwargs)
+    if pretrained:
+        assert (
+            kwargs.get("in_channels", 3) == 3
+        ), "in_channels must be 3 when pretrained is True"
+        print("load from imagenet")
+        model.load_state_dict(model_zoo.load_url(model_urls["resnet18"]), strict=False)
+    return model
+
+
+def resnet34(pretrained=True, **kwargs):
+    """Constructs a ResNet-34 model.
+    Args:
+        pretrained (bool): If True, returns a model pre-trained on ImageNet
+    """
+    model = ResNet(BasicBlock, [3, 4, 6, 3], **kwargs)
+    if pretrained:
+        assert (
+            kwargs.get("in_channels", 3) == 3
+        ), "in_channels must be 3 when pretrained is True"
+        model.load_state_dict(model_zoo.load_url(model_urls["resnet34"]), strict=False)
+    return model
+
+
+def resnet50(pretrained=True, **kwargs):
+    """Constructs a ResNet-50 model.
+    Args:
+        pretrained (bool): If True, returns a model pre-trained on ImageNet
+    """
+    model = ResNet(Bottleneck, [3, 4, 6, 3], **kwargs)
+    if pretrained:
+        assert (
+            kwargs.get("in_channels", 3) == 3
+        ), "in_channels must be 3 when pretrained is True"
+        load_models(model, "resnet50")
+    return model
+
+
+def deformable_resnet50(pretrained=True, **kwargs):
+    """Constructs a ResNet-50 model with deformable conv.
+    Args:
+        pretrained (bool): If True, returns a model pre-trained on ImageNet
+    """
+    model = ResNet(Bottleneck, [3, 4, 6, 3], dcn=dict(deformable_groups=1), **kwargs)
+    if pretrained:
+        assert (
+            kwargs.get("in_channels", 3) == 3
+        ), "in_channels must be 3 when pretrained is True"
+        model.load_state_dict(model_zoo.load_url(model_urls["resnet50"]), strict=False)
+    return model
+
+
+def resnet101(pretrained=True, **kwargs):
+    """Constructs a ResNet-101 model.
+    Args:
+        pretrained (bool): If True, returns a model pre-trained on ImageNet
+    """
+    model = ResNet(Bottleneck, [3, 4, 23, 3], **kwargs)
+    if pretrained:
+        assert (
+            kwargs.get("in_channels", 3) == 3
+        ), "in_channels must be 3 when pretrained is True"
+        model.load_state_dict(model_zoo.load_url(model_urls["resnet101"]), strict=False)
+    return model
+
+
+def resnet152(pretrained=True, **kwargs):
+    """Constructs a ResNet-152 model.
+    Args:
+        pretrained (bool): If True, returns a model pre-trained on ImageNet
+    """
+    model = ResNet(Bottleneck, [3, 8, 36, 3], **kwargs)
+    if pretrained:
+        assert (
+            kwargs.get("in_channels", 3) == 3
+        ), "in_channels must be 3 when pretrained is True"
+        model.load_state_dict(model_zoo.load_url(model_urls["resnet152"]), strict=False)
+    return model
+
+
+if __name__ == "__main__":
+    x = paddle.zeros([2, 3, 640, 640])
+    net = resnet50(pretrained=True)
+    y = net(x)
+    for u in y:
+        print(u.shape)
+
+    print(net.out_channels)
diff --git a/benchmark/PaddleOCR_DBNet/models/basic.py b/benchmark/PaddleOCR_DBNet/models/basic.py
new file mode 100644
index 0000000..49d6a6a
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/models/basic.py
@@ -0,0 +1,40 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/12/6 11:19
+# @Author  : zhoujun
+from paddle import nn
+
+
+class ConvBnRelu(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride=1,
+        padding=0,
+        dilation=1,
+        groups=1,
+        bias=True,
+        padding_mode="zeros",
+        inplace=True,
+    ):
+        super().__init__()
+        self.conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            dilation=dilation,
+            groups=groups,
+            bias_attr=bias,
+            padding_mode=padding_mode,
+        )
+        self.bn = nn.BatchNorm2D(out_channels)
+        self.relu = nn.ReLU()
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        x = self.relu(x)
+        return x
diff --git a/benchmark/PaddleOCR_DBNet/models/head/DBHead.py b/benchmark/PaddleOCR_DBNet/models/head/DBHead.py
new file mode 100644
index 0000000..3a57914
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/models/head/DBHead.py
@@ -0,0 +1,132 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/12/4 14:54
+# @Author  : zhoujun
+import paddle
+from paddle import nn, ParamAttr
+
+
+class DBHead(nn.Layer):
+    def __init__(self, in_channels, out_channels, k=50):
+        super().__init__()
+        self.k = k
+        self.binarize = nn.Sequential(
+            nn.Conv2D(
+                in_channels,
+                in_channels // 4,
+                3,
+                padding=1,
+                weight_attr=ParamAttr(initializer=nn.initializer.KaimingNormal()),
+            ),
+            nn.BatchNorm2D(
+                in_channels // 4,
+                weight_attr=ParamAttr(initializer=nn.initializer.Constant(1)),
+                bias_attr=ParamAttr(initializer=nn.initializer.Constant(1e-4)),
+            ),
+            nn.ReLU(),
+            nn.Conv2DTranspose(
+                in_channels // 4,
+                in_channels // 4,
+                2,
+                2,
+                weight_attr=ParamAttr(initializer=nn.initializer.KaimingNormal()),
+            ),
+            nn.BatchNorm2D(
+                in_channels // 4,
+                weight_attr=ParamAttr(initializer=nn.initializer.Constant(1)),
+                bias_attr=ParamAttr(initializer=nn.initializer.Constant(1e-4)),
+            ),
+            nn.ReLU(),
+            nn.Conv2DTranspose(
+                in_channels // 4, 1, 2, 2, weight_attr=nn.initializer.KaimingNormal()
+            ),
+            nn.Sigmoid(),
+        )
+
+        self.thresh = self._init_thresh(in_channels)
+
+    def forward(self, x):
+        shrink_maps = self.binarize(x)
+        threshold_maps = self.thresh(x)
+        if self.training:
+            binary_maps = self.step_function(shrink_maps, threshold_maps)
+            y = paddle.concat((shrink_maps, threshold_maps, binary_maps), axis=1)
+        else:
+            y = paddle.concat((shrink_maps, threshold_maps), axis=1)
+        return y
+
+    def _init_thresh(self, inner_channels, serial=False, smooth=False, bias=False):
+        in_channels = inner_channels
+        if serial:
+            in_channels += 1
+        self.thresh = nn.Sequential(
+            nn.Conv2D(
+                in_channels,
+                inner_channels // 4,
+                3,
+                padding=1,
+                bias_attr=bias,
+                weight_attr=ParamAttr(initializer=nn.initializer.KaimingNormal()),
+            ),
+            nn.BatchNorm2D(
+                inner_channels // 4,
+                weight_attr=ParamAttr(initializer=nn.initializer.Constant(1)),
+                bias_attr=ParamAttr(initializer=nn.initializer.Constant(1e-4)),
+            ),
+            nn.ReLU(),
+            self._init_upsample(
+                inner_channels // 4, inner_channels // 4, smooth=smooth, bias=bias
+            ),
+            nn.BatchNorm2D(
+                inner_channels // 4,
+                weight_attr=ParamAttr(initializer=nn.initializer.Constant(1)),
+                bias_attr=ParamAttr(initializer=nn.initializer.Constant(1e-4)),
+            ),
+            nn.ReLU(),
+            self._init_upsample(inner_channels // 4, 1, smooth=smooth, bias=bias),
+            nn.Sigmoid(),
+        )
+        return self.thresh
+
+    def _init_upsample(self, in_channels, out_channels, smooth=False, bias=False):
+        if smooth:
+            inter_out_channels = out_channels
+            if out_channels == 1:
+                inter_out_channels = in_channels
+            module_list = [
+                nn.Upsample(scale_factor=2, mode="nearest"),
+                nn.Conv2D(
+                    in_channels,
+                    inter_out_channels,
+                    3,
+                    1,
+                    1,
+                    bias_attr=bias,
+                    weight_attr=ParamAttr(initializer=nn.initializer.KaimingNormal()),
+                ),
+            ]
+            if out_channels == 1:
+                module_list.append(
+                    nn.Conv2D(
+                        in_channels,
+                        out_channels,
+                        kernel_size=1,
+                        stride=1,
+                        padding=1,
+                        bias_attr=True,
+                        weight_attr=ParamAttr(
+                            initializer=nn.initializer.KaimingNormal()
+                        ),
+                    )
+                )
+            return nn.Sequential(module_list)
+        else:
+            return nn.Conv2DTranspose(
+                in_channels,
+                out_channels,
+                2,
+                2,
+                weight_attr=ParamAttr(initializer=nn.initializer.KaimingNormal()),
+            )
+
+    def step_function(self, x, y):
+        return paddle.reciprocal(1 + paddle.exp(-self.k * (x - y)))
diff --git a/benchmark/PaddleOCR_DBNet/models/head/__init__.py b/benchmark/PaddleOCR_DBNet/models/head/__init__.py
new file mode 100644
index 0000000..708ea9a
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/models/head/__init__.py
@@ -0,0 +1,13 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2020/6/5 11:35
+# @Author  : zhoujun
+from .DBHead import DBHead
+
+__all__ = ["build_head"]
+support_head = ["DBHead"]
+
+
+def build_head(head_name, **kwargs):
+    assert head_name in support_head, f"all support head is {support_head}"
+    head = eval(head_name)(**kwargs)
+    return head
diff --git a/benchmark/PaddleOCR_DBNet/models/losses/DB_loss.py b/benchmark/PaddleOCR_DBNet/models/losses/DB_loss.py
new file mode 100644
index 0000000..bad05b6
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/models/losses/DB_loss.py
@@ -0,0 +1,50 @@
+import paddle
+from models.losses.basic_loss import BalanceCrossEntropyLoss, MaskL1Loss, DiceLoss
+
+
+class DBLoss(paddle.nn.Layer):
+    def __init__(self, alpha=1.0, beta=10, ohem_ratio=3, reduction="mean", eps=1e-06):
+        """
+        Implement PSE Loss.
+        :param alpha: binary_map loss 前面的系数
+        :param beta: threshold_map loss 前面的系数
+        :param ohem_ratio: OHEM的比例
+        :param reduction: 'mean' or 'sum'对 batch里的loss 算均值或求和
+        """
+        super().__init__()
+        assert reduction in ["mean", "sum"], " reduction must in ['mean','sum']"
+        self.alpha = alpha
+        self.beta = beta
+        self.bce_loss = BalanceCrossEntropyLoss(negative_ratio=ohem_ratio)
+        self.dice_loss = DiceLoss(eps=eps)
+        self.l1_loss = MaskL1Loss(eps=eps)
+        self.ohem_ratio = ohem_ratio
+        self.reduction = reduction
+
+    def forward(self, pred, batch):
+        shrink_maps = pred[:, 0, :, :]
+        threshold_maps = pred[:, 1, :, :]
+        binary_maps = pred[:, 2, :, :]
+        loss_shrink_maps = self.bce_loss(
+            shrink_maps, batch["shrink_map"], batch["shrink_mask"]
+        )
+        loss_threshold_maps = self.l1_loss(
+            threshold_maps, batch["threshold_map"], batch["threshold_mask"]
+        )
+        metrics = dict(
+            loss_shrink_maps=loss_shrink_maps, loss_threshold_maps=loss_threshold_maps
+        )
+        if pred.shape[1] > 2:
+            loss_binary_maps = self.dice_loss(
+                binary_maps, batch["shrink_map"], batch["shrink_mask"]
+            )
+            metrics["loss_binary_maps"] = loss_binary_maps
+            loss_all = (
+                self.alpha * loss_shrink_maps
+                + self.beta * loss_threshold_maps
+                + loss_binary_maps
+            )
+            metrics["loss"] = loss_all
+        else:
+            metrics["loss"] = loss_shrink_maps
+        return metrics
diff --git a/benchmark/PaddleOCR_DBNet/models/losses/__init__.py b/benchmark/PaddleOCR_DBNet/models/losses/__init__.py
new file mode 100644
index 0000000..e783f7c
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/models/losses/__init__.py
@@ -0,0 +1,16 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2020/6/5 11:36
+# @Author  : zhoujun
+import copy
+from .DB_loss import DBLoss
+
+__all__ = ["build_loss"]
+support_loss = ["DBLoss"]
+
+
+def build_loss(config):
+    copy_config = copy.deepcopy(config)
+    loss_type = copy_config.pop("type")
+    assert loss_type in support_loss, f"all support loss is {support_loss}"
+    criterion = eval(loss_type)(**copy_config)
+    return criterion
diff --git a/benchmark/PaddleOCR_DBNet/models/losses/basic_loss.py b/benchmark/PaddleOCR_DBNet/models/losses/basic_loss.py
new file mode 100644
index 0000000..5f6fb5e
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/models/losses/basic_loss.py
@@ -0,0 +1,101 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/12/4 14:39
+# @Author  : zhoujun
+import paddle
+import paddle.nn as nn
+
+
+class BalanceCrossEntropyLoss(nn.Layer):
+    """
+    Balanced cross entropy loss.
+    Shape:
+        - Input: :math:`(N, 1, H, W)`
+        - GT: :math:`(N, 1, H, W)`, same shape as the input
+        - Mask: :math:`(N, H, W)`, same spatial shape as the input
+        - Output: scalar.
+
+    """
+
+    def __init__(self, negative_ratio=3.0, eps=1e-6):
+        super(BalanceCrossEntropyLoss, self).__init__()
+        self.negative_ratio = negative_ratio
+        self.eps = eps
+
+    def forward(
+        self,
+        pred: paddle.Tensor,
+        gt: paddle.Tensor,
+        mask: paddle.Tensor,
+        return_origin=False,
+    ):
+        """
+        Args:
+            pred: shape :math:`(N, 1, H, W)`, the prediction of network
+            gt: shape :math:`(N, 1, H, W)`, the target
+            mask: shape :math:`(N, H, W)`, the mask indicates positive regions
+        """
+        positive = gt * mask
+        negative = (1 - gt) * mask
+        positive_count = int(positive.sum())
+        negative_count = min(
+            int(negative.sum()), int(positive_count * self.negative_ratio)
+        )
+        loss = nn.functional.binary_cross_entropy(pred, gt, reduction="none")
+        positive_loss = loss * positive
+        negative_loss = loss * negative
+        negative_loss, _ = negative_loss.reshape([-1]).topk(negative_count)
+
+        balance_loss = (positive_loss.sum() + negative_loss.sum()) / (
+            positive_count + negative_count + self.eps
+        )
+
+        if return_origin:
+            return balance_loss, loss
+        return balance_loss
+
+
+class DiceLoss(nn.Layer):
+    """
+    Loss function from https://arxiv.org/abs/1707.03237,
+    where iou computation is introduced heatmap manner to measure the
+    diversity between tow heatmaps.
+    """
+
+    def __init__(self, eps=1e-6):
+        super(DiceLoss, self).__init__()
+        self.eps = eps
+
+    def forward(self, pred: paddle.Tensor, gt, mask, weights=None):
+        """
+        pred: one or two heatmaps of shape (N, 1, H, W),
+            the losses of tow heatmaps are added together.
+        gt: (N, 1, H, W)
+        mask: (N, H, W)
+        """
+        return self._compute(pred, gt, mask, weights)
+
+    def _compute(self, pred, gt, mask, weights):
+        if len(pred.shape) == 4:
+            pred = pred[:, 0, :, :]
+            gt = gt[:, 0, :, :]
+        assert pred.shape == gt.shape
+        assert pred.shape == mask.shape
+        if weights is not None:
+            assert weights.shape == mask.shape
+            mask = weights * mask
+        intersection = (pred * gt * mask).sum()
+
+        union = (pred * mask).sum() + (gt * mask).sum() + self.eps
+        loss = 1 - 2.0 * intersection / union
+        assert loss <= 1
+        return loss
+
+
+class MaskL1Loss(nn.Layer):
+    def __init__(self, eps=1e-6):
+        super(MaskL1Loss, self).__init__()
+        self.eps = eps
+
+    def forward(self, pred: paddle.Tensor, gt, mask):
+        loss = (paddle.abs(pred - gt) * mask).sum() / (mask.sum() + self.eps)
+        return loss
diff --git a/benchmark/PaddleOCR_DBNet/models/model.py b/benchmark/PaddleOCR_DBNet/models/model.py
new file mode 100644
index 0000000..ed36f90
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/models/model.py
@@ -0,0 +1,39 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/8/23 21:57
+# @Author  : zhoujun
+from addict import Dict
+from paddle import nn
+import paddle.nn.functional as F
+
+from models.backbone import build_backbone
+from models.neck import build_neck
+from models.head import build_head
+
+
+class Model(nn.Layer):
+    def __init__(self, model_config: dict):
+        """
+        PANnet
+        :param model_config: 模型配置
+        """
+        super().__init__()
+        model_config = Dict(model_config)
+        backbone_type = model_config.backbone.pop("type")
+        neck_type = model_config.neck.pop("type")
+        head_type = model_config.head.pop("type")
+        self.backbone = build_backbone(backbone_type, **model_config.backbone)
+        self.neck = build_neck(
+            neck_type, in_channels=self.backbone.out_channels, **model_config.neck
+        )
+        self.head = build_head(
+            head_type, in_channels=self.neck.out_channels, **model_config.head
+        )
+        self.name = f"{backbone_type}_{neck_type}_{head_type}"
+
+    def forward(self, x):
+        _, _, H, W = x.shape
+        backbone_out = self.backbone(x)
+        neck_out = self.neck(backbone_out)
+        y = self.head(neck_out)
+        y = F.interpolate(y, size=(H, W), mode="bilinear", align_corners=True)
+        return y
diff --git a/benchmark/PaddleOCR_DBNet/models/neck/FPN.py b/benchmark/PaddleOCR_DBNet/models/neck/FPN.py
new file mode 100644
index 0000000..3c49adf
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/models/neck/FPN.py
@@ -0,0 +1,75 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/9/13 10:29
+# @Author  : zhoujun
+import paddle
+import paddle.nn.functional as F
+from paddle import nn
+
+from models.basic import ConvBnRelu
+
+
+class FPN(nn.Layer):
+    def __init__(self, in_channels, inner_channels=256, **kwargs):
+        """
+        :param in_channels: 基础网络输出的维度
+        :param kwargs:
+        """
+        super().__init__()
+        inplace = True
+        self.conv_out = inner_channels
+        inner_channels = inner_channels // 4
+        # reduce layers
+        self.reduce_conv_c2 = ConvBnRelu(
+            in_channels[0], inner_channels, kernel_size=1, inplace=inplace
+        )
+        self.reduce_conv_c3 = ConvBnRelu(
+            in_channels[1], inner_channels, kernel_size=1, inplace=inplace
+        )
+        self.reduce_conv_c4 = ConvBnRelu(
+            in_channels[2], inner_channels, kernel_size=1, inplace=inplace
+        )
+        self.reduce_conv_c5 = ConvBnRelu(
+            in_channels[3], inner_channels, kernel_size=1, inplace=inplace
+        )
+        # Smooth layers
+        self.smooth_p4 = ConvBnRelu(
+            inner_channels, inner_channels, kernel_size=3, padding=1, inplace=inplace
+        )
+        self.smooth_p3 = ConvBnRelu(
+            inner_channels, inner_channels, kernel_size=3, padding=1, inplace=inplace
+        )
+        self.smooth_p2 = ConvBnRelu(
+            inner_channels, inner_channels, kernel_size=3, padding=1, inplace=inplace
+        )
+
+        self.conv = nn.Sequential(
+            nn.Conv2D(self.conv_out, self.conv_out, kernel_size=3, padding=1, stride=1),
+            nn.BatchNorm2D(self.conv_out),
+            nn.ReLU(),
+        )
+        self.out_channels = self.conv_out
+
+    def forward(self, x):
+        c2, c3, c4, c5 = x
+        # Top-down
+        p5 = self.reduce_conv_c5(c5)
+        p4 = self._upsample_add(p5, self.reduce_conv_c4(c4))
+        p4 = self.smooth_p4(p4)
+        p3 = self._upsample_add(p4, self.reduce_conv_c3(c3))
+        p3 = self.smooth_p3(p3)
+        p2 = self._upsample_add(p3, self.reduce_conv_c2(c2))
+        p2 = self.smooth_p2(p2)
+
+        x = self._upsample_cat(p2, p3, p4, p5)
+        x = self.conv(x)
+        return x
+
+    def _upsample_add(self, x, y):
+        return F.interpolate(x, size=y.shape[2:]) + y
+
+    def _upsample_cat(self, p2, p3, p4, p5):
+        h, w = p2.shape[2:]
+        p3 = F.interpolate(p3, size=(h, w))
+        p4 = F.interpolate(p4, size=(h, w))
+        p5 = F.interpolate(p5, size=(h, w))
+        return paddle.concat([p2, p3, p4, p5], axis=1)
diff --git a/benchmark/PaddleOCR_DBNet/models/neck/__init__.py b/benchmark/PaddleOCR_DBNet/models/neck/__init__.py
new file mode 100644
index 0000000..d63b8b5
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/models/neck/__init__.py
@@ -0,0 +1,13 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2020/6/5 11:34
+# @Author  : zhoujun
+from .FPN import FPN
+
+__all__ = ["build_neck"]
+support_neck = ["FPN"]
+
+
+def build_neck(neck_name, **kwargs):
+    assert neck_name in support_neck, f"all support neck is {support_neck}"
+    neck = eval(neck_name)(**kwargs)
+    return neck
diff --git a/benchmark/PaddleOCR_DBNet/multi_gpu_train.sh b/benchmark/PaddleOCR_DBNet/multi_gpu_train.sh
new file mode 100644
index 0000000..4b9a915
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/multi_gpu_train.sh
@@ -0,0 +1,2 @@
+# export NCCL_P2P_DISABLE=1
+CUDA_VISIBLE_DEVICES=0,1,2,3 python3 -m paddle.distributed.launch tools/train.py --config_file "config/icdar2015_resnet50_FPN_DBhead_polyLR.yaml"
diff --git a/benchmark/PaddleOCR_DBNet/post_processing/__init__.py b/benchmark/PaddleOCR_DBNet/post_processing/__init__.py
new file mode 100644
index 0000000..7d3a448
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/post_processing/__init__.py
@@ -0,0 +1,13 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/12/5 15:17
+# @Author  : zhoujun
+
+from .seg_detector_representer import SegDetectorRepresenter
+
+
+def get_post_processing(config):
+    try:
+        cls = eval(config["type"])(**config["args"])
+        return cls
+    except:
+        return None
diff --git a/benchmark/PaddleOCR_DBNet/post_processing/seg_detector_representer.py b/benchmark/PaddleOCR_DBNet/post_processing/seg_detector_representer.py
new file mode 100644
index 0000000..6a394fb
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/post_processing/seg_detector_representer.py
@@ -0,0 +1,191 @@
+import cv2
+import numpy as np
+import pyclipper
+import paddle
+from shapely.geometry import Polygon
+
+
+class SegDetectorRepresenter:
+    def __init__(
+        self, thresh=0.3, box_thresh=0.7, max_candidates=1000, unclip_ratio=1.5
+    ):
+        self.min_size = 3
+        self.thresh = thresh
+        self.box_thresh = box_thresh
+        self.max_candidates = max_candidates
+        self.unclip_ratio = unclip_ratio
+
+    def __call__(self, batch, pred, is_output_polygon=False):
+        """
+        batch: (image, polygons, ignore_tags
+        batch: a dict produced by dataloaders.
+            image: tensor of shape (N, C, H, W).
+            polygons: tensor of shape (N, K, 4, 2), the polygons of objective regions.
+            ignore_tags: tensor of shape (N, K), indicates whether a region is ignorable or not.
+            shape: the original shape of images.
+            filename: the original filenames of images.
+        pred:
+            binary: text region segmentation map, with shape (N, H, W)
+            thresh: [if exists] thresh hold prediction with shape (N, H, W)
+            thresh_binary: [if exists] binarized with threshold, (N, H, W)
+        """
+        if isinstance(pred, paddle.Tensor):
+            pred = pred.numpy()
+        pred = pred[:, 0, :, :]
+        segmentation = self.binarize(pred)
+        boxes_batch = []
+        scores_batch = []
+        for batch_index in range(pred.shape[0]):
+            height, width = batch["shape"][batch_index]
+            if is_output_polygon:
+                boxes, scores = self.polygons_from_bitmap(
+                    pred[batch_index], segmentation[batch_index], width, height
+                )
+            else:
+                boxes, scores = self.boxes_from_bitmap(
+                    pred[batch_index], segmentation[batch_index], width, height
+                )
+            boxes_batch.append(boxes)
+            scores_batch.append(scores)
+        return boxes_batch, scores_batch
+
+    def binarize(self, pred):
+        return pred > self.thresh
+
+    def polygons_from_bitmap(self, pred, _bitmap, dest_width, dest_height):
+        """
+        _bitmap: single map with shape (H, W),
+            whose values are binarized as {0, 1}
+        """
+
+        assert len(_bitmap.shape) == 2
+        bitmap = _bitmap  # The first channel
+        height, width = bitmap.shape
+        boxes = []
+        scores = []
+
+        contours, _ = cv2.findContours(
+            (bitmap * 255).astype(np.uint8), cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE
+        )
+
+        for contour in contours[: self.max_candidates]:
+            epsilon = 0.005 * cv2.arcLength(contour, True)
+            approx = cv2.approxPolyDP(contour, epsilon, True)
+            points = approx.reshape((-1, 2))
+            if points.shape[0] < 4:
+                continue
+            # _, sside = self.get_mini_boxes(contour)
+            # if sside < self.min_size:
+            #     continue
+            score = self.box_score_fast(pred, contour.squeeze(1))
+            if self.box_thresh > score:
+                continue
+
+            if points.shape[0] > 2:
+                box = self.unclip(points, unclip_ratio=self.unclip_ratio)
+                if len(box) > 1:
+                    continue
+            else:
+                continue
+            box = box.reshape(-1, 2)
+            _, sside = self.get_mini_boxes(box.reshape((-1, 1, 2)))
+            if sside < self.min_size + 2:
+                continue
+
+            if not isinstance(dest_width, int):
+                dest_width = dest_width.item()
+                dest_height = dest_height.item()
+
+            box[:, 0] = np.clip(np.round(box[:, 0] / width * dest_width), 0, dest_width)
+            box[:, 1] = np.clip(
+                np.round(box[:, 1] / height * dest_height), 0, dest_height
+            )
+            boxes.append(box)
+            scores.append(score)
+        return boxes, scores
+
+    def boxes_from_bitmap(self, pred, _bitmap, dest_width, dest_height):
+        """
+        _bitmap: single map with shape (H, W),
+            whose values are binarized as {0, 1}
+        """
+
+        assert len(_bitmap.shape) == 2
+        bitmap = _bitmap  # The first channel
+        height, width = bitmap.shape
+        contours, _ = cv2.findContours(
+            (bitmap * 255).astype(np.uint8), cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE
+        )
+        num_contours = min(len(contours), self.max_candidates)
+        boxes = np.zeros((num_contours, 4, 2), dtype=np.int16)
+        scores = np.zeros((num_contours,), dtype=np.float32)
+
+        for index in range(num_contours):
+            contour = contours[index].squeeze(1)
+            points, sside = self.get_mini_boxes(contour)
+            if sside < self.min_size:
+                continue
+            points = np.array(points)
+            score = self.box_score_fast(pred, contour)
+            if self.box_thresh > score:
+                continue
+
+            box = self.unclip(points, unclip_ratio=self.unclip_ratio).reshape(-1, 1, 2)
+            box, sside = self.get_mini_boxes(box)
+            if sside < self.min_size + 2:
+                continue
+            box = np.array(box)
+            if not isinstance(dest_width, int):
+                dest_width = dest_width.item()
+                dest_height = dest_height.item()
+
+            box[:, 0] = np.clip(np.round(box[:, 0] / width * dest_width), 0, dest_width)
+            box[:, 1] = np.clip(
+                np.round(box[:, 1] / height * dest_height), 0, dest_height
+            )
+            boxes[index, :, :] = box.astype(np.int16)
+            scores[index] = score
+        return boxes, scores
+
+    def unclip(self, box, unclip_ratio=1.5):
+        poly = Polygon(box)
+        distance = poly.area * unclip_ratio / poly.length
+        offset = pyclipper.PyclipperOffset()
+        offset.AddPath(box, pyclipper.JT_ROUND, pyclipper.ET_CLOSEDPOLYGON)
+        expanded = np.array(offset.Execute(distance))
+        return expanded
+
+    def get_mini_boxes(self, contour):
+        bounding_box = cv2.minAreaRect(contour)
+        points = sorted(list(cv2.boxPoints(bounding_box)), key=lambda x: x[0])
+
+        index_1, index_2, index_3, index_4 = 0, 1, 2, 3
+        if points[1][1] > points[0][1]:
+            index_1 = 0
+            index_4 = 1
+        else:
+            index_1 = 1
+            index_4 = 0
+        if points[3][1] > points[2][1]:
+            index_2 = 2
+            index_3 = 3
+        else:
+            index_2 = 3
+            index_3 = 2
+
+        box = [points[index_1], points[index_2], points[index_3], points[index_4]]
+        return box, min(bounding_box[1])
+
+    def box_score_fast(self, bitmap, _box):
+        h, w = bitmap.shape[:2]
+        box = _box.copy()
+        xmin = np.clip(np.floor(box[:, 0].min()).astype(np.int32), 0, w - 1)
+        xmax = np.clip(np.ceil(box[:, 0].max()).astype(np.int32), 0, w - 1)
+        ymin = np.clip(np.floor(box[:, 1].min()).astype(np.int32), 0, h - 1)
+        ymax = np.clip(np.ceil(box[:, 1].max()).astype(np.int32), 0, h - 1)
+
+        mask = np.zeros((ymax - ymin + 1, xmax - xmin + 1), dtype=np.uint8)
+        box[:, 0] = box[:, 0] - xmin
+        box[:, 1] = box[:, 1] - ymin
+        cv2.fillPoly(mask, box.reshape(1, -1, 2).astype(np.int32), 1)
+        return cv2.mean(bitmap[ymin : ymax + 1, xmin : xmax + 1], mask)[0]
diff --git a/benchmark/PaddleOCR_DBNet/predict.sh b/benchmark/PaddleOCR_DBNet/predict.sh
new file mode 100644
index 0000000..a4b9bfa
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/predict.sh
@@ -0,0 +1 @@
+CUDA_VISIBLE_DEVICES=0 python tools/predict.py --model_path model_best.pth --input_folder ./input --output_folder ./output --thre 0.7 --polygon --show --save_result
diff --git a/benchmark/PaddleOCR_DBNet/requirement.txt b/benchmark/PaddleOCR_DBNet/requirement.txt
new file mode 100644
index 0000000..23240ac
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/requirement.txt
@@ -0,0 +1,14 @@
+anyconfig
+future
+imgaug
+matplotlib
+numpy
+opencv-python
+Polygon3
+pyclipper
+PyYAML
+scikit-image
+Shapely
+tqdm
+addict
+visualdl
diff --git a/benchmark/PaddleOCR_DBNet/single_gpu_train.sh b/benchmark/PaddleOCR_DBNet/single_gpu_train.sh
new file mode 100644
index 0000000..6803201
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/single_gpu_train.sh
@@ -0,0 +1 @@
+CUDA_VISIBLE_DEVICES=0 python3 tools/train.py --config_file "config/icdar2015_resnet50_FPN_DBhead_polyLR.yaml" 
diff --git a/benchmark/PaddleOCR_DBNet/test/README.MD b/benchmark/PaddleOCR_DBNet/test/README.MD
new file mode 100644
index 0000000..3e2bb1a
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/test/README.MD
@@ -0,0 +1,8 @@
+Place the images that you want to detect here. You better named them as such:
+img_10.jpg
+img_11.jpg
+img_{img_id}.jpg
+
+For predicting single images, you can change the `img_path` in the `/tools/predict.py` to your image number.
+
+The result will be saved in the output_folder(default is test/output) you give in predict.sh
diff --git a/benchmark/PaddleOCR_DBNet/test_tipc/benchmark_train.sh b/benchmark/PaddleOCR_DBNet/test_tipc/benchmark_train.sh
new file mode 100644
index 0000000..d94dac2
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/test_tipc/benchmark_train.sh
@@ -0,0 +1,287 @@
+#!/bin/bash
+source test_tipc/common_func.sh
+
+# run benchmark sh 
+# Usage:
+# bash run_benchmark_train.sh config.txt params
+# or 
+# bash run_benchmark_train.sh config.txt
+
+function func_parser_params(){
+    strs=$1
+    IFS="="
+    array=(${strs})
+    tmp=${array[1]}
+    echo ${tmp}
+}
+
+function set_dynamic_epoch(){
+    string=$1
+    num=$2
+    _str=${string:1:6}
+    IFS="C"
+    arr=(${_str})
+    M=${arr[0]}
+    P=${arr[1]}
+    ep=`expr $num \* $M \* $P`
+    echo $ep
+}
+
+function func_sed_params(){
+    filename=$1
+    line=$2
+    param_value=$3
+    params=`sed -n "${line}p" $filename`
+    IFS=":"
+    array=(${params})
+    key=${array[0]}
+    value=${array[1]}
+
+    new_params="${key}:${param_value}"
+    IFS=";"
+    cmd="sed -i '${line}s/.*/${new_params}/' '${filename}'"
+    eval $cmd
+}
+
+function set_gpu_id(){
+    string=$1
+    _str=${string:1:6}
+    IFS="C"
+    arr=(${_str})
+    M=${arr[0]}
+    P=${arr[1]}
+    gn=`expr $P - 1`
+    gpu_num=`expr $gn / $M`
+    seq=`seq -s "," 0 $gpu_num`
+    echo $seq
+}
+
+function get_repo_name(){
+    IFS=";"
+    cur_dir=$(pwd)
+    IFS="/"
+    arr=(${cur_dir})
+    echo ${arr[-1]}
+}
+
+FILENAME=$1
+# copy FILENAME as new
+new_filename="./test_tipc/benchmark_train.txt"
+cmd=`yes|cp $FILENAME $new_filename`
+FILENAME=$new_filename
+# MODE must be one of ['benchmark_train']
+MODE=$2
+PARAMS=$3
+
+to_static=""
+# parse "to_static" options and modify trainer into "to_static_trainer"
+if [[ $PARAMS =~ "dynamicTostatic" ]] ;then
+   to_static="d2sT_"
+   sed -i 's/trainer:norm_train/trainer:to_static_train/g' $FILENAME
+   # clear PARAM contents
+   if [ $PARAMS = "to_static" ] ;then
+    PARAMS=""
+   fi
+fi
+# bash test_tipc/benchmark_train.sh test_tipc/configs/det_mv3_db_v2_0/train_benchmark.txt  benchmark_train dynamic_bs8_fp32_DP_N1C8
+# bash test_tipc/benchmark_train.sh test_tipc/configs/det_mv3_db_v2_0/train_benchmark.txt  benchmark_train dynamicTostatic_bs8_fp32_DP_N1C8
+# bash test_tipc/benchmark_train.sh test_tipc/configs/det_mv3_db_v2_0/train_benchmark.txt  benchmark_train dynamic_bs8_null_DP_N1C1
+IFS=$'\n'
+# parser params from train_benchmark.txt
+dataline=`cat $FILENAME`
+# parser params
+IFS=$'\n'
+lines=(${dataline})
+model_name=$(func_parser_value "${lines[1]}")
+python_name=$(func_parser_value "${lines[2]}")
+
+# set env
+python=${python_name}
+export str_tmp=$(echo `pip list|grep paddlepaddle-gpu|awk -F ' ' '{print $2}'`)
+export frame_version=${str_tmp%%.post*}
+export frame_commit=$(echo `${python} -c "import paddle;print(paddle.version.commit)"`)
+
+# 获取benchmark_params所在的行数
+line_num=`grep -n -w "train_benchmark_params" $FILENAME  | cut -d ":" -f 1`
+# for train log parser
+batch_size=$(func_parser_value "${lines[line_num]}")
+line_num=`expr $line_num + 1`
+fp_items=$(func_parser_value "${lines[line_num]}")
+line_num=`expr $line_num + 1`
+epoch=$(func_parser_value "${lines[line_num]}")
+
+line_num=`expr $line_num + 1`
+profile_option_key=$(func_parser_key "${lines[line_num]}")
+profile_option_params=$(func_parser_value "${lines[line_num]}")
+profile_option="${profile_option_key}:${profile_option_params}"
+
+line_num=`expr $line_num + 1`
+flags_value=$(func_parser_value "${lines[line_num]}")
+# set flags
+IFS=";"
+flags_list=(${flags_value})
+for _flag in ${flags_list[*]}; do
+    cmd="export ${_flag}"
+    eval $cmd
+done
+
+# set log_name
+repo_name=$(get_repo_name )
+SAVE_LOG=${BENCHMARK_LOG_DIR:-$(pwd)}   # */benchmark_log
+mkdir -p "${SAVE_LOG}/benchmark_log/"
+status_log="${SAVE_LOG}/benchmark_log/results.log"
+
+# The number of lines in which train params can be replaced.
+line_python=3
+line_gpuid=4
+line_precision=6
+line_epoch=7
+line_batchsize=9
+line_profile=13
+line_eval_py=24
+line_export_py=30
+
+func_sed_params "$FILENAME" "${line_eval_py}" "null"
+func_sed_params "$FILENAME" "${line_export_py}" "null"
+func_sed_params "$FILENAME" "${line_python}"  "$python"
+
+# if params
+if  [ ! -n "$PARAMS" ] ;then
+    # PARAMS input is not a word.
+    IFS="|"
+    batch_size_list=(${batch_size})
+    fp_items_list=(${fp_items})
+    device_num_list=(N1C4)
+    run_mode="DP"
+elif [[ ${PARAMS} = "dynamicTostatic" ]];then
+    IFS="|"
+    model_type=$PARAMS
+    batch_size_list=(${batch_size})
+    fp_items_list=(${fp_items})
+    device_num_list=(N1C4)
+    run_mode="DP"
+else
+    # parser params from input: modeltype_bs${bs_item}_${fp_item}_${run_mode}_${device_num}
+    IFS="_"
+    params_list=(${PARAMS})
+    model_type=${params_list[0]}
+    batch_size=${params_list[1]}
+    batch_size=`echo  ${batch_size} | tr -cd "[0-9]" `
+    precision=${params_list[2]}
+    run_mode=${params_list[3]}
+    device_num=${params_list[4]}
+    IFS=";"
+
+    if [ ${precision} = "fp16" ];then
+        precision="amp"
+    fi
+
+    epoch=$(set_dynamic_epoch $device_num $epoch)
+    fp_items_list=($precision)
+    batch_size_list=($batch_size)
+    device_num_list=($device_num)
+fi
+
+IFS="|"
+for batch_size in ${batch_size_list[*]}; do 
+    for train_precision in ${fp_items_list[*]}; do
+        for device_num in ${device_num_list[*]}; do
+            # sed batchsize and precision
+            if [ ${train_precision} = "amp" ];then
+                precision="fp16"
+            else
+                precision="fp32"
+            fi
+
+            func_sed_params "$FILENAME" "${line_precision}" "$train_precision"
+            func_sed_params "$FILENAME" "${line_batchsize}" "$MODE=$batch_size"
+            func_sed_params "$FILENAME" "${line_epoch}" "$MODE=$epoch"
+            gpu_id=$(set_gpu_id $device_num)
+
+            if [ ${#gpu_id} -le 1 ];then
+                log_path="$SAVE_LOG/profiling_log"
+                mkdir -p $log_path
+                log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_mode}_${device_num}_${to_static}profiling"
+                func_sed_params "$FILENAME" "${line_gpuid}" "0"  # sed used gpu_id 
+                # set profile_option params
+                tmp=`sed -i "${line_profile}s/.*/${profile_option}/" "${FILENAME}"`
+
+                # run test_train_inference_python.sh
+                cmd="bash test_tipc/test_train_inference_python.sh ${FILENAME} benchmark_train > ${log_path}/${log_name} 2>&1 "
+                echo $cmd
+                eval $cmd
+                eval "cat ${log_path}/${log_name}"
+
+                # without profile
+                log_path="$SAVE_LOG/train_log"
+                speed_log_path="$SAVE_LOG/index"
+                mkdir -p $log_path
+                mkdir -p $speed_log_path
+                log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_mode}_${device_num}_${to_static}log"
+                speed_log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_mode}_${device_num}_${to_static}speed"
+                func_sed_params "$FILENAME" "${line_profile}" "null"  # sed profile_id as null
+                cmd="bash test_tipc/test_train_inference_python.sh ${FILENAME} benchmark_train > ${log_path}/${log_name} 2>&1 "
+                echo $cmd
+                job_bt=`date '+%Y%m%d%H%M%S'`
+                eval $cmd
+                job_et=`date '+%Y%m%d%H%M%S'`
+                export model_run_time=$((${job_et}-${job_bt}))
+                eval "cat ${log_path}/${log_name}"
+
+                # parser log
+                _model_name="${model_name}_bs${batch_size}_${precision}_${run_mode}"
+                cmd="${python} ${BENCHMARK_ROOT}/scripts/analysis.py --filename ${log_path}/${log_name} \
+                        --speed_log_file '${speed_log_path}/${speed_log_name}' \
+                        --model_name ${_model_name} \
+                        --base_batch_size ${batch_size} \
+                        --run_mode ${run_mode} \
+                        --fp_item ${precision} \
+                        --keyword ips: \
+                        --skip_steps 2 \
+                        --device_num ${device_num} \
+                        --speed_unit samples/s \
+                        --convergence_key loss: "
+                echo $cmd
+                eval $cmd
+                last_status=${PIPESTATUS[0]}
+                status_check $last_status "${cmd}" "${status_log}"
+            else
+                IFS=";"
+                unset_env=`unset CUDA_VISIBLE_DEVICES`
+                log_path="$SAVE_LOG/train_log"
+                speed_log_path="$SAVE_LOG/index"
+                mkdir -p $log_path
+                mkdir -p $speed_log_path
+                log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_mode}_${device_num}_${to_static}log"
+                speed_log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_mode}_${device_num}_${to_static}speed"
+                func_sed_params "$FILENAME" "${line_gpuid}" "$gpu_id"  # sed used gpu_id 
+                func_sed_params "$FILENAME" "${line_profile}" "null"  # sed --profile_option as null
+                cmd="bash test_tipc/test_train_inference_python.sh ${FILENAME} benchmark_train > ${log_path}/${log_name} 2>&1 "
+                echo $cmd
+                job_bt=`date '+%Y%m%d%H%M%S'`
+                eval $cmd
+                job_et=`date '+%Y%m%d%H%M%S'`
+                export model_run_time=$((${job_et}-${job_bt}))
+                eval "cat ${log_path}/${log_name}"
+                # parser log
+                _model_name="${model_name}_bs${batch_size}_${precision}_${run_mode}"
+                
+                cmd="${python} ${BENCHMARK_ROOT}/scripts/analysis.py --filename ${log_path}/${log_name} \
+                        --speed_log_file '${speed_log_path}/${speed_log_name}' \
+                        --model_name ${_model_name} \
+                        --base_batch_size ${batch_size} \
+                        --run_mode ${run_mode} \
+                        --fp_item ${precision} \
+                        --keyword ips: \
+                        --skip_steps 2 \
+                        --device_num ${device_num} \
+                        --speed_unit images/s \
+                        --convergence_key loss: "
+                echo $cmd
+                eval $cmd
+                last_status=${PIPESTATUS[0]}
+                status_check $last_status "${cmd}" "${status_log}"
+            fi
+        done
+    done
+done
diff --git a/benchmark/PaddleOCR_DBNet/test_tipc/common_func.sh b/benchmark/PaddleOCR_DBNet/test_tipc/common_func.sh
new file mode 100644
index 0000000..9ec22f0
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/test_tipc/common_func.sh
@@ -0,0 +1,67 @@
+#!/bin/bash
+
+function func_parser_key(){
+    strs=$1
+    IFS=":"
+    array=(${strs})
+    tmp=${array[0]}
+    echo ${tmp}
+}
+
+function func_parser_value(){
+    strs=$1
+    IFS=":"
+    array=(${strs})
+    tmp=${array[1]}
+    echo ${tmp}
+}
+
+function func_set_params(){
+    key=$1
+    value=$2
+    if [ ${key}x = "null"x ];then
+        echo " "
+    elif [[ ${value} = "null" ]] || [[ ${value} = " " ]] || [ ${#value} -le 0 ];then
+        echo " "
+    else 
+        echo "${key}=${value}"
+    fi
+}
+
+function func_parser_params(){
+    strs=$1
+    MODE=$2
+    IFS=":"
+    array=(${strs})
+    key=${array[0]}
+    tmp=${array[1]}
+    IFS="|"
+    res=""
+    for _params in ${tmp[*]}; do
+        IFS="="
+        array=(${_params})
+        mode=${array[0]}
+        value=${array[1]}
+        if [[ ${mode} = ${MODE} ]]; then
+            IFS="|"
+            #echo $(func_set_params "${mode}" "${value}")
+            echo $value
+            break
+        fi
+        IFS="|"
+    done
+    echo ${res}
+}
+
+function status_check(){
+    last_status=$1   # the exit code
+    run_command=$2
+    run_log=$3
+    model_name=$4
+    log_path=$5
+    if [ $last_status -eq 0 ]; then
+        echo -e "\033[33m Run successfully with command - ${model_name} - ${run_command} - ${log_path} \033[0m" | tee -a ${run_log}
+    else
+        echo -e "\033[33m Run failed with command - ${model_name} - ${run_command} - ${log_path} \033[0m" | tee -a ${run_log}
+    fi
+}
diff --git a/benchmark/PaddleOCR_DBNet/test_tipc/configs/det_res50_db/train_infer_python.txt b/benchmark/PaddleOCR_DBNet/test_tipc/configs/det_res50_db/train_infer_python.txt
new file mode 100644
index 0000000..7dc3da3
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/test_tipc/configs/det_res50_db/train_infer_python.txt
@@ -0,0 +1,61 @@
+===========================train_params===========================
+model_name:det_res50_db
+python:python
+gpu_list:0|0,1
+trainer.use_gpu:True|True
+amp:null
+trainer.epochs:lite_train_lite_infer=1|whole_train_whole_infer=300
+trainer.output_dir:./output/
+dataset.train.loader.batch_size:lite_train_lite_infer=8|whole_train_lite_infer=8
+trainer.finetune_checkpoint:null
+train_model_name:checkpoint/model_latest.pth
+train_infer_img_dir:imgs/paper/db.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py --config_file config/icdar2015_resnet50_FPN_DBhead_polyLR.yaml -o trainer.log_iter=1 trainer.enable_eval=False dataset.train.loader.shuffle=false arch.backbone.pretrained=False
+quant_export:null
+fpgm_export:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:null
+null:null
+##
+===========================infer_params===========================
+trainer.output_dir:./output/
+trainer.resume_checkpoint:
+norm_export:tools/export_model.py --config_file config/icdar2015_resnet50_FPN_DBhead_polyLR.yaml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/det_r50_vd_db_v2.0_train/best_accuracy
+infer_export:tools/export_model.py --config_file config/icdar2015_resnet50_FPN_DBhead_polyLR.yaml -o
+infer_quant:False
+inference:tools/infer.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--batch_size:1
+--use_tensorrt:False
+--precision:fp32
+--model_dir:
+--img_path:imgs/paper/db.jpg
+--save_log_path:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
+===========================train_benchmark_params==========================
+batch_size:8
+fp_items:fp32|fp16
+epoch:2
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
+===========================to_static_train_benchmark_params===========================
+to_static_train:trainer.to_static=true
diff --git a/benchmark/PaddleOCR_DBNet/test_tipc/prepare.sh b/benchmark/PaddleOCR_DBNet/test_tipc/prepare.sh
new file mode 100644
index 0000000..a961603
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/test_tipc/prepare.sh
@@ -0,0 +1,54 @@
+#!/bin/bash
+source test_tipc/common_func.sh
+
+FILENAME=$1
+
+# MODE be one of ['lite_train_lite_infer' 'lite_train_whole_infer' 'whole_train_whole_infer',  
+#                 'whole_infer', 'klquant_whole_infer',
+#                 'cpp_infer', 'serving_infer']
+
+MODE=$2
+
+dataline=$(cat ${FILENAME})
+
+# parser params
+IFS=$'\n'
+lines=(${dataline})
+
+# The training params
+model_name=$(func_parser_value "${lines[1]}")
+
+trainer_list=$(func_parser_value "${lines[14]}")
+
+if [ ${MODE} = "lite_train_lite_infer" ];then
+    python_name_list=$(func_parser_value "${lines[2]}")
+    array=(${python_name_list}) 
+    python_name=${array[0]}
+    ${python_name} -m pip install -r requirement.txt
+    if [[ ${model_name} =~ "det_res50_db" ]];then
+        wget -nc https://paddle-wheel.bj.bcebos.com/benchmark/resnet50-19c8e357.pth -O /root/.cache/torch/hub/checkpoints/resnet50-19c8e357.pth
+
+        # 下载数据集并解压
+        rm -rf datasets
+        wget -nc https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/benchmark_train/datasets.tar
+        tar xf datasets.tar
+    fi
+elif [ ${MODE} = "benchmark_train" ];then
+    python_name_list=$(func_parser_value "${lines[2]}")
+    array=(${python_name_list}) 
+    python_name=${array[0]}
+    ${python_name} -m pip install -r requirement.txt
+    if [[ ${model_name} =~ "det_res50_db" ]];then
+        wget -nc https://paddle-wheel.bj.bcebos.com/benchmark/resnet50-19c8e357.pth -O /root/.cache/torch/hub/checkpoints/resnet50-19c8e357.pth
+
+        # 下载数据集并解压
+        rm -rf datasets
+        wget -nc https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/benchmark_train/datasets.tar
+        tar xf datasets.tar
+        # expand gt.txt 2 times
+        # cd ./train_data/icdar2015/text_localization
+        # for i in `seq 2`;do cp train_icdar2015_label.txt dup$i.txt;done
+        # cat dup* > train_icdar2015_label.txt && rm -rf dup*
+        # cd ../../../
+    fi
+fi
diff --git a/benchmark/PaddleOCR_DBNet/test_tipc/test_train_inference_python.sh b/benchmark/PaddleOCR_DBNet/test_tipc/test_train_inference_python.sh
new file mode 100644
index 0000000..33c81e1
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/test_tipc/test_train_inference_python.sh
@@ -0,0 +1,343 @@
+#!/bin/bash
+source test_tipc/common_func.sh
+
+FILENAME=$1
+# MODE be one of ['lite_train_lite_infer' 'lite_train_whole_infer' 'whole_train_whole_infer', 'whole_infer']
+MODE=$2
+
+dataline=$(awk 'NR>=1{print}'  $FILENAME)
+
+# parser params
+IFS=$'\n'
+lines=(${dataline})
+
+# The training params
+model_name=$(func_parser_value "${lines[1]}")
+python=$(func_parser_value "${lines[2]}")
+gpu_list=$(func_parser_value "${lines[3]}")
+train_use_gpu_key=$(func_parser_key "${lines[4]}")
+train_use_gpu_value=$(func_parser_value "${lines[4]}")
+autocast_list=$(func_parser_value "${lines[5]}")
+autocast_key=$(func_parser_key "${lines[5]}")
+epoch_key=$(func_parser_key "${lines[6]}")
+epoch_num=$(func_parser_params "${lines[6]}" "${MODE}")
+save_model_key=$(func_parser_key "${lines[7]}")
+train_batch_key=$(func_parser_key "${lines[8]}")
+train_batch_value=$(func_parser_params "${lines[8]}" "${MODE}")
+pretrain_model_key=$(func_parser_key "${lines[9]}")
+pretrain_model_value=$(func_parser_value "${lines[9]}")
+train_model_name=$(func_parser_value "${lines[10]}")
+train_infer_img_dir=$(func_parser_value "${lines[11]}")
+train_param_key1=$(func_parser_key "${lines[12]}")
+train_param_value1=$(func_parser_value "${lines[12]}")
+
+trainer_list=$(func_parser_value "${lines[14]}")
+trainer_norm=$(func_parser_key "${lines[15]}")
+norm_trainer=$(func_parser_value "${lines[15]}")
+pact_key=$(func_parser_key "${lines[16]}")
+pact_trainer=$(func_parser_value "${lines[16]}")
+fpgm_key=$(func_parser_key "${lines[17]}")
+fpgm_trainer=$(func_parser_value "${lines[17]}")
+distill_key=$(func_parser_key "${lines[18]}")
+distill_trainer=$(func_parser_value "${lines[18]}")
+trainer_key1=$(func_parser_key "${lines[19]}")
+trainer_value1=$(func_parser_value "${lines[19]}")
+trainer_key2=$(func_parser_key "${lines[20]}")
+trainer_value2=$(func_parser_value "${lines[20]}")
+
+eval_py=$(func_parser_value "${lines[23]}")
+eval_key1=$(func_parser_key "${lines[24]}")
+eval_value1=$(func_parser_value "${lines[24]}")
+
+save_infer_key=$(func_parser_key "${lines[27]}")
+export_weight=$(func_parser_key "${lines[28]}")
+norm_export=$(func_parser_value "${lines[29]}")
+pact_export=$(func_parser_value "${lines[30]}")
+fpgm_export=$(func_parser_value "${lines[31]}")
+distill_export=$(func_parser_value "${lines[32]}")
+export_key1=$(func_parser_key "${lines[33]}")
+export_value1=$(func_parser_value "${lines[33]}")
+export_key2=$(func_parser_key "${lines[34]}")
+export_value2=$(func_parser_value "${lines[34]}")
+inference_dir=$(func_parser_value "${lines[35]}")
+
+# parser inference model 
+infer_model_dir_list=$(func_parser_value "${lines[36]}")
+infer_export_list=$(func_parser_value "${lines[37]}")
+infer_is_quant=$(func_parser_value "${lines[38]}")
+# parser inference 
+inference_py=$(func_parser_value "${lines[39]}")
+use_gpu_key=$(func_parser_key "${lines[40]}")
+use_gpu_list=$(func_parser_value "${lines[40]}")
+use_mkldnn_key=$(func_parser_key "${lines[41]}")
+use_mkldnn_list=$(func_parser_value "${lines[41]}")
+cpu_threads_key=$(func_parser_key "${lines[42]}")
+cpu_threads_list=$(func_parser_value "${lines[42]}")
+batch_size_key=$(func_parser_key "${lines[43]}")
+batch_size_list=$(func_parser_value "${lines[43]}")
+use_trt_key=$(func_parser_key "${lines[44]}")
+use_trt_list=$(func_parser_value "${lines[44]}")
+precision_key=$(func_parser_key "${lines[45]}")
+precision_list=$(func_parser_value "${lines[45]}")
+infer_model_key=$(func_parser_key "${lines[46]}")
+image_dir_key=$(func_parser_key "${lines[47]}")
+infer_img_dir=$(func_parser_value "${lines[47]}")
+save_log_key=$(func_parser_key "${lines[48]}")
+benchmark_key=$(func_parser_key "${lines[49]}")
+benchmark_value=$(func_parser_value "${lines[49]}")
+infer_key1=$(func_parser_key "${lines[50]}")
+infer_value1=$(func_parser_value "${lines[50]}")
+
+LOG_PATH="./test_tipc/output/${model_name}/${MODE}"
+mkdir -p ${LOG_PATH}
+status_log="${LOG_PATH}/results_python.log"
+
+line_num=`grep -n -w "to_static_train_benchmark_params" $FILENAME  | cut -d ":" -f 1`
+to_static_key=$(func_parser_key "${lines[line_num]}")
+to_static_trainer=$(func_parser_value "${lines[line_num]}")
+
+function func_inference(){
+    IFS='|'
+    _python=$1
+    _script=$2
+    _model_dir=$3
+    _log_path=$4
+    _img_dir=$5
+    _flag_quant=$6
+    _gpu=$7
+    # inference 
+    for use_gpu in ${use_gpu_list[*]}; do
+        if [ ${use_gpu} = "False" ] || [ ${use_gpu} = "cpu" ]; then
+            for use_mkldnn in ${use_mkldnn_list[*]}; do
+                # if [ ${use_mkldnn} = "False" ] && [ ${_flag_quant} = "True" ]; then
+                #     continue
+                # fi
+                for threads in ${cpu_threads_list[*]}; do
+                    for batch_size in ${batch_size_list[*]}; do
+                        for precision in ${precision_list[*]}; do
+                            if [ ${use_mkldnn} = "False" ] && [ ${precision} = "fp16" ]; then
+                                continue
+                            fi # skip when enable fp16 but disable mkldnn
+                            if [ ${_flag_quant} = "True" ] && [ ${precision} != "int8" ]; then
+                                continue
+                            fi # skip when quant model inference but precision is not int8
+                            set_precision=$(func_set_params "${precision_key}" "${precision}")
+                            
+                            _save_log_path="${_log_path}/python_infer_cpu_gpus_${_gpu}_usemkldnn_${use_mkldnn}_threads_${threads}_precision_${precision}_batchsize_${batch_size}.log"
+                            set_infer_data=$(func_set_params "${image_dir_key}" "${_img_dir}")
+                            set_benchmark=$(func_set_params "${benchmark_key}" "${benchmark_value}")
+                            set_batchsize=$(func_set_params "${batch_size_key}" "${batch_size}")
+                            set_mkldnn=$(func_set_params "${use_mkldnn_key}" "${use_mkldnn}")
+                            set_cpu_threads=$(func_set_params "${cpu_threads_key}" "${threads}")
+                            set_model_dir=$(func_set_params "${infer_model_key}" "${_model_dir}")
+                            set_infer_params0=$(func_set_params "${save_log_key}" "${save_log_value}")
+                            set_infer_params1=$(func_set_params "${infer_key1}" "${infer_value1}")
+                            command="${_python} ${_script} ${use_gpu_key}=${use_gpu} ${set_mkldnn} ${set_cpu_threads} ${set_model_dir} ${set_batchsize} ${set_infer_params0} ${set_infer_data} ${set_benchmark} ${set_precision} ${set_infer_params1} > ${_save_log_path} 2>&1 "
+                            eval $command
+                            last_status=${PIPESTATUS[0]}
+                            eval "cat ${_save_log_path}"
+                            status_check $last_status "${command}" "${status_log}" "${model_name}" "${_save_log_path}"
+                        done
+                    done
+                done
+            done
+        elif [ ${use_gpu} = "True" ] || [ ${use_gpu} = "gpu" ]; then
+            for use_trt in ${use_trt_list[*]}; do
+                for precision in ${precision_list[*]}; do
+                    if [[ ${_flag_quant} = "False" ]] && [[ ${precision} =~ "int8" ]]; then
+                        continue
+                    fi 
+                    if [[ ${precision} =~ "fp16" || ${precision} =~ "int8" ]] && [ ${use_trt} = "False" ]; then
+                        continue
+                    fi
+                    if [[ ${use_trt} = "False" && ${precision} =~ "int8" ]] && [ ${_flag_quant} = "True" ]; then
+                        continue
+                    fi
+                    for batch_size in ${batch_size_list[*]}; do
+                        _save_log_path="${_log_path}/python_infer_gpu_gpus_${_gpu}_usetrt_${use_trt}_precision_${precision}_batchsize_${batch_size}.log"
+                        set_infer_data=$(func_set_params "${image_dir_key}" "${_img_dir}")
+                        set_benchmark=$(func_set_params "${benchmark_key}" "${benchmark_value}")
+                        set_batchsize=$(func_set_params "${batch_size_key}" "${batch_size}")
+                        set_tensorrt=$(func_set_params "${use_trt_key}" "${use_trt}")
+                        set_precision=$(func_set_params "${precision_key}" "${precision}")
+                        set_model_dir=$(func_set_params "${infer_model_key}" "${_model_dir}")
+                        set_infer_params0=$(func_set_params "${save_log_key}" "${save_log_value}")
+                        set_infer_params1=$(func_set_params "${infer_key1}" "${infer_value1}")
+                        command="${_python} ${_script} ${use_gpu_key}=${use_gpu} ${set_tensorrt} ${set_precision} ${set_model_dir} ${set_batchsize} ${set_infer_data} ${set_benchmark} ${set_infer_params1} ${set_infer_params0} > ${_save_log_path} 2>&1 "
+                        eval $command
+                        last_status=${PIPESTATUS[0]}
+                        eval "cat ${_save_log_path}"
+                        status_check $last_status "${command}" "${status_log}" "${model_name}" "${_save_log_path}"
+                        
+                    done
+                done
+            done
+        else
+            echo "Does not support hardware other than CPU and GPU Currently!"
+        fi
+    done
+}
+
+if [ ${MODE} = "whole_infer" ]; then
+    GPUID=$3
+    if [ ${#GPUID} -le 0 ];then
+        env=" "
+    else
+        env="export CUDA_VISIBLE_DEVICES=${GPUID}"
+    fi
+    # set CUDA_VISIBLE_DEVICES
+    eval $env
+    export Count=0
+    gpu=0
+    IFS="|"
+    infer_run_exports=(${infer_export_list})
+    infer_quant_flag=(${infer_is_quant})
+    for infer_model in ${infer_model_dir_list[*]}; do
+        # run export
+        if [ ${infer_run_exports[Count]} != "null" ];then 
+            save_infer_dir="${infer_model}"
+            set_export_weight=$(func_set_params "${export_weight}" "${infer_model}")
+            set_save_infer_key=$(func_set_params "${save_infer_key}" "${save_infer_dir}")
+            export_log_path="${LOG_PATH}_export_${Count}.log"
+            export_cmd="${python} ${infer_run_exports[Count]} ${set_export_weight} ${set_save_infer_key} > ${export_log_path} 2>&1 "
+            echo ${infer_run_exports[Count]} 
+            echo $export_cmd
+            eval $export_cmd
+            status_export=$?
+            status_check $status_export "${export_cmd}" "${status_log}" "${model_name}" "${export_log_path}"
+        else
+            save_infer_dir=${infer_model}
+        fi
+        #run inference
+        is_quant=${infer_quant_flag[Count]}
+        func_inference "${python}" "${inference_py}" "${save_infer_dir}" "${LOG_PATH}" "${infer_img_dir}" ${is_quant} "${gpu}"
+        Count=$(($Count + 1))
+    done
+else
+    IFS="|"
+    export Count=0
+    USE_GPU_KEY=(${train_use_gpu_value})
+    for gpu in ${gpu_list[*]}; do
+        train_use_gpu=${USE_GPU_KEY[Count]}
+        Count=$(($Count + 1))
+        ips=""
+        if [ ${gpu} = "-1" ];then
+            env=""
+        elif [ ${#gpu} -le 1 ];then
+            env="export CUDA_VISIBLE_DEVICES=${gpu}"
+        elif [ ${#gpu} -le 15 ];then
+            IFS=","
+            array=(${gpu})
+            env="export CUDA_VISIBLE_DEVICES=${array[0]}"
+            IFS="|"
+        else
+            IFS=";"
+            array=(${gpu})
+            ips=${array[0]}
+            gpu=${array[1]}
+            IFS="|"
+            env=" "
+        fi
+        for autocast in ${autocast_list[*]}; do 
+            if [ ${autocast} = "amp" ]; then
+                set_amp_config="amp.scale_loss=1024.0 amp.use_dynamic_loss_scaling=True amp.amp_level=O2"
+            else
+                set_amp_config="amp=None"
+            fi          
+            for trainer in ${trainer_list[*]}; do 
+                flag_quant=False
+                if [ ${trainer} = ${pact_key} ]; then
+                    run_train=${pact_trainer}
+                    run_export=${pact_export}
+                    flag_quant=True
+                elif [ ${trainer} = "${fpgm_key}" ]; then
+                    run_train=${fpgm_trainer}
+                    run_export=${fpgm_export}
+                elif [ ${trainer} = "${distill_key}" ]; then
+                    run_train=${distill_trainer}
+                    run_export=${distill_export}
+                elif [ ${trainer} = "${to_static_key}" ]; then
+                    run_train="${norm_trainer}  ${to_static_trainer}"
+                    run_export=${norm_export}
+                elif [[ ${trainer} = ${trainer_key2} ]]; then
+                    run_train=${trainer_value2}
+                    run_export=${export_value2}
+                else
+                    run_train=${norm_trainer}
+                    run_export=${norm_export}
+                fi
+
+                if [ ${run_train} = "null" ]; then
+                    continue
+                fi
+
+                set_epoch=$(func_set_params "${epoch_key}" "${epoch_num}")
+                set_pretrain=$(func_set_params "${pretrain_model_key}" "${pretrain_model_value}")
+                set_batchsize=$(func_set_params "${train_batch_key}" "${train_batch_value}")
+                set_train_params1=$(func_set_params "${train_param_key1}" "${train_param_value1}")
+                set_use_gpu=$(func_set_params "${train_use_gpu_key}" "${train_use_gpu}")
+                # if length of ips >= 15, then it is seen as multi-machine
+                # 15 is the min length of ips info for multi-machine: 0.0.0.0,0.0.0.0
+                if [ ${#ips} -le 15 ];then
+                    save_log="${LOG_PATH}/${trainer}_gpus_${gpu}_autocast_${autocast}"
+                    nodes=1
+                else
+                    IFS=","
+                    ips_array=(${ips})
+                    IFS="|"
+                    nodes=${#ips_array[@]}
+                    save_log="${LOG_PATH}/${trainer}_gpus_${gpu}_autocast_${autocast}_nodes_${nodes}"
+                fi
+
+
+                set_save_model=$(func_set_params "${save_model_key}" "${save_log}")
+                if [ ${#gpu} -le 2 ];then  # train with cpu or single gpu
+                    cmd="${python} ${run_train} ${set_use_gpu}  ${set_save_model} ${set_epoch} ${set_pretrain}  ${set_batchsize} ${set_amp_config} ${set_train_params1}"
+                elif [ ${#ips} -le 15 ];then  # train with multi-gpu
+                    cmd="${python} -m paddle.distributed.launch --gpus=${gpu} ${run_train} ${set_use_gpu} ${set_save_model} ${set_epoch} ${set_pretrain}  ${set_batchsize} ${set_amp_config} ${set_train_params1}"
+                else     # train with multi-machine
+                    cmd="${python} -m paddle.distributed.launch --ips=${ips} --gpus=${gpu} ${run_train} ${set_use_gpu} ${set_save_model} ${set_pretrain} ${set_epoch}  ${set_batchsize} ${set_amp_config} ${set_train_params1}"
+                fi
+                # run train
+                eval $cmd
+                eval "cat ${save_log}/train.log >> ${save_log}.log"
+                status_check $? "${cmd}" "${status_log}" "${model_name}" "${save_log}.log"
+
+                set_eval_pretrain=$(func_set_params "${pretrain_model_key}" "${save_log}/${train_model_name}")
+
+                # run eval 
+                if [ ${eval_py} != "null" ]; then
+                    eval ${env}
+                    set_eval_params1=$(func_set_params "${eval_key1}" "${eval_value1}")
+                    eval_log_path="${LOG_PATH}/${trainer}_gpus_${gpu}_autocast_${autocast}_nodes_${nodes}_eval.log"
+                    eval_cmd="${python} ${eval_py} ${set_eval_pretrain} ${set_use_gpu} ${set_eval_params1} > ${eval_log_path} 2>&1 " 
+                    eval $eval_cmd
+                    status_check $? "${eval_cmd}" "${status_log}" "${model_name}" "${eval_log_path}"
+                fi
+                # run export model
+                if [ ${run_export} != "null" ]; then 
+                    # run export model
+                    save_infer_path="${save_log}"
+                    export_log_path="${LOG_PATH}/${trainer}_gpus_${gpu}_autocast_${autocast}_nodes_${nodes}_export.log"
+                    set_export_weight=$(func_set_params "${export_weight}" "${save_log}/${train_model_name}")
+                    set_save_infer_key=$(func_set_params "${save_infer_key}" "${save_infer_path}")
+                    export_cmd="${python} ${run_export} ${set_export_weight} ${set_save_infer_key} > ${export_log_path} 2>&1 "
+                    eval $export_cmd
+                    status_check $? "${export_cmd}" "${status_log}" "${model_name}" "${export_log_path}"
+
+                    #run inference
+                    eval $env
+                    save_infer_path="${save_log}"
+                    if [[ ${inference_dir} != "null" ]] && [[ ${inference_dir} != '##' ]]; then
+                        infer_model_dir="${save_infer_path}/${inference_dir}"
+                    else
+                        infer_model_dir=${save_infer_path}
+                    fi
+                    func_inference "${python}" "${inference_py}" "${infer_model_dir}" "${LOG_PATH}" "${train_infer_img_dir}" "${flag_quant}" "${gpu}"
+                    
+                    eval "unset CUDA_VISIBLE_DEVICES"
+                fi
+            done  # done with:    for trainer in ${trainer_list[*]}; do 
+        done      # done with:    for autocast in ${autocast_list[*]}; do 
+    done          # done with:    for gpu in ${gpu_list[*]}; do
+fi  # end if [ ${MODE} = "infer" ]; then
diff --git a/benchmark/PaddleOCR_DBNet/tools/__init__.py b/benchmark/PaddleOCR_DBNet/tools/__init__.py
new file mode 100644
index 0000000..aa5fe6c
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/tools/__init__.py
@@ -0,0 +1,3 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/12/8 13:14
+# @Author  : zhoujun
diff --git a/benchmark/PaddleOCR_DBNet/tools/eval.py b/benchmark/PaddleOCR_DBNet/tools/eval.py
new file mode 100644
index 0000000..21a5a4a
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/tools/eval.py
@@ -0,0 +1,93 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2018/6/11 15:54
+# @Author  : zhoujun
+import os
+import sys
+import pathlib
+
+__dir__ = pathlib.Path(os.path.abspath(__file__))
+sys.path.append(str(__dir__))
+sys.path.append(str(__dir__.parent.parent))
+
+import argparse
+import time
+import paddle
+from tqdm.auto import tqdm
+
+
+class EVAL:
+    def __init__(self, model_path, gpu_id=0):
+        from models import build_model
+        from data_loader import get_dataloader
+        from post_processing import get_post_processing
+        from utils import get_metric
+
+        self.gpu_id = gpu_id
+        if (
+            self.gpu_id is not None
+            and isinstance(self.gpu_id, int)
+            and paddle.device.is_compiled_with_cuda()
+        ):
+            paddle.device.set_device("gpu:{}".format(self.gpu_id))
+        else:
+            paddle.device.set_device("cpu")
+        checkpoint = paddle.load(model_path)
+        config = checkpoint["config"]
+        config["arch"]["backbone"]["pretrained"] = False
+
+        self.validate_loader = get_dataloader(
+            config["dataset"]["validate"], config["distributed"]
+        )
+
+        self.model = build_model(config["arch"])
+        self.model.set_state_dict(checkpoint["state_dict"])
+
+        self.post_process = get_post_processing(config["post_processing"])
+        self.metric_cls = get_metric(config["metric"])
+
+    def eval(self):
+        self.model.eval()
+        raw_metrics = []
+        total_frame = 0.0
+        total_time = 0.0
+        for i, batch in tqdm(
+            enumerate(self.validate_loader),
+            total=len(self.validate_loader),
+            desc="test model",
+        ):
+            with paddle.no_grad():
+                start = time.time()
+                preds = self.model(batch["img"])
+                boxes, scores = self.post_process(
+                    batch, preds, is_output_polygon=self.metric_cls.is_output_polygon
+                )
+                total_frame += batch["img"].shape[0]
+                total_time += time.time() - start
+                raw_metric = self.metric_cls.validate_measure(batch, (boxes, scores))
+                raw_metrics.append(raw_metric)
+        metrics = self.metric_cls.gather_measure(raw_metrics)
+        print("FPS:{}".format(total_frame / total_time))
+        return {
+            "recall": metrics["recall"].avg,
+            "precision": metrics["precision"].avg,
+            "fmeasure": metrics["fmeasure"].avg,
+        }
+
+
+def init_args():
+    parser = argparse.ArgumentParser(description="DBNet.paddle")
+    parser.add_argument(
+        "--model_path",
+        required=False,
+        default="output/DBNet_resnet18_FPN_DBHead/checkpoint/1.pth",
+        type=str,
+    )
+    args = parser.parse_args()
+    return args
+
+
+if __name__ == "__main__":
+    args = init_args()
+    eval = EVAL(args.model_path)
+    result = eval.eval()
+    print(result)
diff --git a/benchmark/PaddleOCR_DBNet/tools/export_model.py b/benchmark/PaddleOCR_DBNet/tools/export_model.py
new file mode 100644
index 0000000..71a33d0
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/tools/export_model.py
@@ -0,0 +1,57 @@
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "..")))
+
+import argparse
+
+import paddle
+from paddle.jit import to_static
+
+from models import build_model
+from utils import Config, ArgsParser
+
+
+def init_args():
+    parser = ArgsParser()
+    args = parser.parse_args()
+    return args
+
+
+def load_checkpoint(model, checkpoint_path):
+    """
+    load checkpoints
+    :param checkpoint_path: Checkpoint path to be loaded
+    """
+    checkpoint = paddle.load(checkpoint_path)
+    model.set_state_dict(checkpoint["state_dict"])
+    print("load checkpoint from {}".format(checkpoint_path))
+
+
+def main(config):
+    model = build_model(config["arch"])
+    load_checkpoint(model, config["trainer"]["resume_checkpoint"])
+    model.eval()
+
+    save_path = config["trainer"]["output_dir"]
+    save_path = os.path.join(save_path, "inference")
+    infer_shape = [3, -1, -1]
+    model = to_static(
+        model,
+        input_spec=[
+            paddle.static.InputSpec(shape=[None] + infer_shape, dtype="float32")
+        ],
+    )
+
+    paddle.jit.save(model, save_path)
+    print("inference model is saved to {}".format(save_path))
+
+
+if __name__ == "__main__":
+    args = init_args()
+    assert os.path.exists(args.config_file)
+    config = Config(args.config_file)
+    config.merge_dict(args.opt)
+    main(config.cfg)
diff --git a/benchmark/PaddleOCR_DBNet/tools/infer.py b/benchmark/PaddleOCR_DBNet/tools/infer.py
new file mode 100644
index 0000000..fb42536
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/tools/infer.py
@@ -0,0 +1,315 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import sys
+import pathlib
+
+__dir__ = pathlib.Path(os.path.abspath(__file__))
+sys.path.append(str(__dir__))
+sys.path.append(str(__dir__.parent.parent))
+
+import cv2
+import paddle
+from paddle import inference
+import numpy as np
+from PIL import Image
+
+from paddle.vision import transforms
+from tools.predict import resize_image
+from post_processing import get_post_processing
+from utils.util import draw_bbox, save_result
+
+
+class InferenceEngine(object):
+    """InferenceEngine
+
+    Inference engine class which contains preprocess, run, postprocess
+    """
+
+    def __init__(self, args):
+        """
+        Args:
+            args: Parameters generated using argparser.
+        Returns: None
+        """
+        super().__init__()
+        self.args = args
+
+        # init inference engine
+        (
+            self.predictor,
+            self.config,
+            self.input_tensor,
+            self.output_tensor,
+        ) = self.load_predictor(
+            os.path.join(args.model_dir, "inference.pdmodel"),
+            os.path.join(args.model_dir, "inference.pdiparams"),
+        )
+
+        # build transforms
+        self.transforms = transforms.Compose(
+            [
+                transforms.ToTensor(),
+                transforms.Normalize(
+                    mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]
+                ),
+            ]
+        )
+
+        # wamrup
+        if self.args.warmup > 0:
+            for idx in range(args.warmup):
+                print(idx)
+                x = np.random.rand(
+                    1, 3, self.args.crop_size, self.args.crop_size
+                ).astype("float32")
+                self.input_tensor.copy_from_cpu(x)
+                self.predictor.run()
+                self.output_tensor.copy_to_cpu()
+
+        self.post_process = get_post_processing(
+            {
+                "type": "SegDetectorRepresenter",
+                "args": {
+                    "thresh": 0.3,
+                    "box_thresh": 0.7,
+                    "max_candidates": 1000,
+                    "unclip_ratio": 1.5,
+                },
+            }
+        )
+
+    def load_predictor(self, model_file_path, params_file_path):
+        """load_predictor
+        initialize the inference engine
+        Args:
+            model_file_path: inference model path (*.pdmodel)
+            model_file_path: inference parameter path (*.pdiparams)
+        Return:
+            predictor: Predictor created using Paddle Inference.
+            config: Configuration of the predictor.
+            input_tensor: Input tensor of the predictor.
+            output_tensor: Output tensor of the predictor.
+        """
+        args = self.args
+        config = inference.Config(model_file_path, params_file_path)
+        if args.use_gpu:
+            config.enable_use_gpu(1000, 0)
+            if args.use_tensorrt:
+                config.enable_tensorrt_engine(
+                    workspace_size=1 << 30,
+                    precision_mode=precision,
+                    max_batch_size=args.max_batch_size,
+                    min_subgraph_size=args.min_subgraph_size,  # skip the minimum trt subgraph
+                    use_calib_mode=False,
+                )
+
+                # collect shape
+                trt_shape_f = os.path.join(model_dir, "_trt_dynamic_shape.txt")
+
+                if not os.path.exists(trt_shape_f):
+                    config.collect_shape_range_info(trt_shape_f)
+                    logger.info(f"collect dynamic shape info into : {trt_shape_f}")
+                try:
+                    config.enable_tuned_tensorrt_dynamic_shape(trt_shape_f, True)
+                except Exception as E:
+                    logger.info(E)
+                    logger.info("Please keep your paddlepaddle-gpu >= 2.3.0!")
+        else:
+            config.disable_gpu()
+            # The thread num should not be greater than the number of cores in the CPU.
+            if args.enable_mkldnn:
+                # cache 10 different shapes for mkldnn to avoid memory leak
+                config.set_mkldnn_cache_capacity(10)
+                config.enable_mkldnn()
+                if args.precision == "fp16":
+                    config.enable_mkldnn_bfloat16()
+                if hasattr(args, "cpu_threads"):
+                    config.set_cpu_math_library_num_threads(args.cpu_threads)
+                else:
+                    # default cpu threads as 10
+                    config.set_cpu_math_library_num_threads(10)
+
+        # enable memory optim
+        config.enable_memory_optim()
+        config.disable_glog_info()
+
+        config.switch_use_feed_fetch_ops(False)
+        config.switch_ir_optim(True)
+
+        # create predictor
+        predictor = inference.create_predictor(config)
+
+        # get input and output tensor property
+        input_names = predictor.get_input_names()
+        input_tensor = predictor.get_input_handle(input_names[0])
+
+        output_names = predictor.get_output_names()
+        output_tensor = predictor.get_output_handle(output_names[0])
+
+        return predictor, config, input_tensor, output_tensor
+
+    def preprocess(self, img_path, short_size):
+        """preprocess
+        Preprocess to the input.
+        Args:
+            img_path: Image path.
+        Returns: Input data after preprocess.
+        """
+        img = cv2.imread(img_path, 1)
+        img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
+        h, w = img.shape[:2]
+        img = resize_image(img, short_size)
+        img = self.transforms(img)
+        img = np.expand_dims(img, axis=0)
+        shape_info = {"shape": [(h, w)]}
+        return img, shape_info
+
+    def postprocess(self, x, shape_info, is_output_polygon):
+        """postprocess
+        Postprocess to the inference engine output.
+        Args:
+            x: Inference engine output.
+        Returns: Output data after argmax.
+        """
+        box_list, score_list = self.post_process(
+            shape_info, x, is_output_polygon=is_output_polygon
+        )
+        box_list, score_list = box_list[0], score_list[0]
+        if len(box_list) > 0:
+            if is_output_polygon:
+                idx = [x.sum() > 0 for x in box_list]
+                box_list = [box_list[i] for i, v in enumerate(idx) if v]
+                score_list = [score_list[i] for i, v in enumerate(idx) if v]
+            else:
+                idx = (
+                    box_list.reshape(box_list.shape[0], -1).sum(axis=1) > 0
+                )  # 去掉全为0的框
+                box_list, score_list = box_list[idx], score_list[idx]
+        else:
+            box_list, score_list = [], []
+        return box_list, score_list
+
+    def run(self, x):
+        """run
+        Inference process using inference engine.
+        Args:
+            x: Input data after preprocess.
+        Returns: Inference engine output
+        """
+        self.input_tensor.copy_from_cpu(x)
+        self.predictor.run()
+        output = self.output_tensor.copy_to_cpu()
+        return output
+
+
+def get_args(add_help=True):
+    """
+    parse args
+    """
+    import argparse
+
+    def str2bool(v):
+        return v.lower() in ("true", "t", "1")
+
+    parser = argparse.ArgumentParser(
+        description="PaddlePaddle Classification Training", add_help=add_help
+    )
+
+    parser.add_argument("--model_dir", default=None, help="inference model dir")
+    parser.add_argument("--batch_size", type=int, default=1)
+    parser.add_argument("--short_size", default=1024, type=int, help="short size")
+    parser.add_argument("--img_path", default="./images/demo.jpg")
+
+    parser.add_argument("--benchmark", default=False, type=str2bool, help="benchmark")
+    parser.add_argument("--warmup", default=0, type=int, help="warmup iter")
+    parser.add_argument("--polygon", action="store_true", help="output polygon or box")
+
+    parser.add_argument("--use_gpu", type=str2bool, default=True)
+    parser.add_argument("--use_tensorrt", type=str2bool, default=False)
+    parser.add_argument("--precision", type=str, default="fp32")
+    parser.add_argument("--gpu_mem", type=int, default=500)
+    parser.add_argument("--gpu_id", type=int, default=0)
+    parser.add_argument("--enable_mkldnn", type=str2bool, default=False)
+    parser.add_argument("--cpu_threads", type=int, default=10)
+
+    args = parser.parse_args()
+    return args
+
+
+def main(args):
+    """
+    Main inference function.
+    Args:
+        args: Parameters generated using argparser.
+    Returns:
+        class_id: Class index of the input.
+        prob: : Probability of the input.
+    """
+    inference_engine = InferenceEngine(args)
+
+    # init benchmark
+    if args.benchmark:
+        import auto_log
+
+        autolog = auto_log.AutoLogger(
+            model_name="db",
+            batch_size=args.batch_size,
+            inference_config=inference_engine.config,
+            gpu_ids="auto" if args.use_gpu else None,
+        )
+
+    # enable benchmark
+    if args.benchmark:
+        autolog.times.start()
+
+    # preprocess
+    img, shape_info = inference_engine.preprocess(args.img_path, args.short_size)
+
+    if args.benchmark:
+        autolog.times.stamp()
+
+    output = inference_engine.run(img)
+
+    if args.benchmark:
+        autolog.times.stamp()
+
+    # postprocess
+    box_list, score_list = inference_engine.postprocess(
+        output, shape_info, args.polygon
+    )
+
+    if args.benchmark:
+        autolog.times.stamp()
+        autolog.times.end(stamp=True)
+        autolog.report()
+
+    img = draw_bbox(cv2.imread(args.img_path)[:, :, ::-1], box_list)
+    # 保存结果到路径
+    os.makedirs("output", exist_ok=True)
+    img_path = pathlib.Path(args.img_path)
+    output_path = os.path.join("output", img_path.stem + "_infer_result.jpg")
+    cv2.imwrite(output_path, img[:, :, ::-1])
+    save_result(
+        output_path.replace("_infer_result.jpg", ".txt"),
+        box_list,
+        score_list,
+        args.polygon,
+    )
+
+
+if __name__ == "__main__":
+    args = get_args()
+    main(args)
diff --git a/benchmark/PaddleOCR_DBNet/tools/predict.py b/benchmark/PaddleOCR_DBNet/tools/predict.py
new file mode 100644
index 0000000..3d6a089
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/tools/predict.py
@@ -0,0 +1,175 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/8/24 12:06
+# @Author  : zhoujun
+
+import os
+import sys
+import pathlib
+
+__dir__ = pathlib.Path(os.path.abspath(__file__))
+sys.path.append(str(__dir__))
+sys.path.append(str(__dir__.parent.parent))
+
+import time
+import cv2
+import paddle
+
+from data_loader import get_transforms
+from models import build_model
+from post_processing import get_post_processing
+
+
+def resize_image(img, short_size):
+    height, width, _ = img.shape
+    if height < width:
+        new_height = short_size
+        new_width = new_height / height * width
+    else:
+        new_width = short_size
+        new_height = new_width / width * height
+    new_height = int(round(new_height / 32) * 32)
+    new_width = int(round(new_width / 32) * 32)
+    resized_img = cv2.resize(img, (new_width, new_height))
+    return resized_img
+
+
+class PaddleModel:
+    def __init__(self, model_path, post_p_thre=0.7, gpu_id=None):
+        """
+        初始化模型
+        :param model_path: 模型地址(可以是模型的参数或者参数和计算图一起保存的文件)
+        :param gpu_id: 在哪一块gpu上运行
+        """
+        self.gpu_id = gpu_id
+
+        if (
+            self.gpu_id is not None
+            and isinstance(self.gpu_id, int)
+            and paddle.device.is_compiled_with_cuda()
+        ):
+            paddle.device.set_device("gpu:{}".format(self.gpu_id))
+        else:
+            paddle.device.set_device("cpu")
+        checkpoint = paddle.load(model_path)
+
+        config = checkpoint["config"]
+        config["arch"]["backbone"]["pretrained"] = False
+        self.model = build_model(config["arch"])
+        self.post_process = get_post_processing(config["post_processing"])
+        self.post_process.box_thresh = post_p_thre
+        self.img_mode = config["dataset"]["train"]["dataset"]["args"]["img_mode"]
+        self.model.set_state_dict(checkpoint["state_dict"])
+        self.model.eval()
+
+        self.transform = []
+        for t in config["dataset"]["train"]["dataset"]["args"]["transforms"]:
+            if t["type"] in ["ToTensor", "Normalize"]:
+                self.transform.append(t)
+        self.transform = get_transforms(self.transform)
+
+    def predict(self, img_path: str, is_output_polygon=False, short_size: int = 1024):
+        """
+        对传入的图像进行预测，支持图像地址,opencv 读取图片，偏慢
+        :param img_path: 图像地址
+        :param is_numpy:
+        :return:
+        """
+        assert os.path.exists(img_path), "file is not exists"
+        img = cv2.imread(img_path, 1 if self.img_mode != "GRAY" else 0)
+        if self.img_mode == "RGB":
+            img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
+        h, w = img.shape[:2]
+        img = resize_image(img, short_size)
+        # 将图片由(w,h)变为(1,img_channel,h,w)
+        tensor = self.transform(img)
+        tensor = tensor.unsqueeze_(0)
+
+        batch = {"shape": [(h, w)]}
+        with paddle.no_grad():
+            start = time.time()
+            preds = self.model(tensor)
+            box_list, score_list = self.post_process(
+                batch, preds, is_output_polygon=is_output_polygon
+            )
+            box_list, score_list = box_list[0], score_list[0]
+            if len(box_list) > 0:
+                if is_output_polygon:
+                    idx = [x.sum() > 0 for x in box_list]
+                    box_list = [box_list[i] for i, v in enumerate(idx) if v]
+                    score_list = [score_list[i] for i, v in enumerate(idx) if v]
+                else:
+                    idx = (
+                        box_list.reshape(box_list.shape[0], -1).sum(axis=1) > 0
+                    )  # 去掉全为0的框
+                    box_list, score_list = box_list[idx], score_list[idx]
+            else:
+                box_list, score_list = [], []
+            t = time.time() - start
+        return preds[0, 0, :, :].detach().cpu().numpy(), box_list, score_list, t
+
+
+def save_depoly(net, input, save_path):
+    input_spec = [paddle.static.InputSpec(shape=[None, 3, None, None], dtype="float32")]
+    net = paddle.jit.to_static(net, input_spec=input_spec)
+
+    # save static model for inference directly
+    paddle.jit.save(net, save_path)
+
+
+def init_args():
+    import argparse
+
+    parser = argparse.ArgumentParser(description="DBNet.paddle")
+    parser.add_argument("--model_path", default=r"model_best.pth", type=str)
+    parser.add_argument(
+        "--input_folder", default="./test/input", type=str, help="img path for predict"
+    )
+    parser.add_argument(
+        "--output_folder", default="./test/output", type=str, help="img path for output"
+    )
+    parser.add_argument("--gpu", default=0, type=int, help="gpu for inference")
+    parser.add_argument(
+        "--thre", default=0.3, type=float, help="the thresh of post_processing"
+    )
+    parser.add_argument("--polygon", action="store_true", help="output polygon or box")
+    parser.add_argument("--show", action="store_true", help="show result")
+    parser.add_argument(
+        "--save_result", action="store_true", help="save box and score to txt file"
+    )
+    args = parser.parse_args()
+    return args
+
+
+if __name__ == "__main__":
+    import pathlib
+    from tqdm import tqdm
+    import matplotlib.pyplot as plt
+    from utils.util import show_img, draw_bbox, save_result, get_image_file_list
+
+    args = init_args()
+    print(args)
+    # 初始化网络
+    model = PaddleModel(args.model_path, post_p_thre=args.thre, gpu_id=args.gpu)
+    img_folder = pathlib.Path(args.input_folder)
+    for img_path in tqdm(get_image_file_list(args.input_folder)):
+        preds, boxes_list, score_list, t = model.predict(
+            img_path, is_output_polygon=args.polygon
+        )
+        img = draw_bbox(cv2.imread(img_path)[:, :, ::-1], boxes_list)
+        if args.show:
+            show_img(preds)
+            show_img(img, title=os.path.basename(img_path))
+            plt.show()
+        # 保存结果到路径
+        os.makedirs(args.output_folder, exist_ok=True)
+        img_path = pathlib.Path(img_path)
+        output_path = os.path.join(args.output_folder, img_path.stem + "_result.jpg")
+        pred_path = os.path.join(args.output_folder, img_path.stem + "_pred.jpg")
+        cv2.imwrite(output_path, img[:, :, ::-1])
+        cv2.imwrite(pred_path, preds * 255)
+        save_result(
+            output_path.replace("_result.jpg", ".txt"),
+            boxes_list,
+            score_list,
+            args.polygon,
+        )
diff --git a/benchmark/PaddleOCR_DBNet/tools/train.py b/benchmark/PaddleOCR_DBNet/tools/train.py
new file mode 100644
index 0000000..170eebd
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/tools/train.py
@@ -0,0 +1,64 @@
+import os
+import sys
+import pathlib
+
+__dir__ = pathlib.Path(os.path.abspath(__file__))
+sys.path.append(str(__dir__))
+sys.path.append(str(__dir__.parent.parent))
+
+import paddle
+import paddle.distributed as dist
+from utils import Config, ArgsParser
+
+
+def init_args():
+    parser = ArgsParser()
+    args = parser.parse_args()
+    return args
+
+
+def main(config, profiler_options):
+    from models import build_model, build_loss
+    from data_loader import get_dataloader
+    from trainer import Trainer
+    from post_processing import get_post_processing
+    from utils import get_metric
+
+    if paddle.device.cuda.device_count() > 1:
+        dist.init_parallel_env()
+        config["distributed"] = True
+    else:
+        config["distributed"] = False
+    train_loader = get_dataloader(config["dataset"]["train"], config["distributed"])
+    assert train_loader is not None
+    if "validate" in config["dataset"]:
+        validate_loader = get_dataloader(config["dataset"]["validate"], False)
+    else:
+        validate_loader = None
+    criterion = build_loss(config["loss"])
+    config["arch"]["backbone"]["in_channels"] = (
+        3 if config["dataset"]["train"]["dataset"]["args"]["img_mode"] != "GRAY" else 1
+    )
+    model = build_model(config["arch"])
+    # set @to_static for benchmark, skip this by default.
+    post_p = get_post_processing(config["post_processing"])
+    metric = get_metric(config["metric"])
+    trainer = Trainer(
+        config=config,
+        model=model,
+        criterion=criterion,
+        train_loader=train_loader,
+        post_process=post_p,
+        metric_cls=metric,
+        validate_loader=validate_loader,
+        profiler_options=profiler_options,
+    )
+    trainer.train()
+
+
+if __name__ == "__main__":
+    args = init_args()
+    assert os.path.exists(args.config_file)
+    config = Config(args.config_file)
+    config.merge_dict(args.opt)
+    main(config.cfg, args.profiler_options)
diff --git a/benchmark/PaddleOCR_DBNet/trainer/__init__.py b/benchmark/PaddleOCR_DBNet/trainer/__init__.py
new file mode 100644
index 0000000..e5b2234
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/trainer/__init__.py
@@ -0,0 +1,4 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/8/23 21:58
+# @Author  : zhoujun
+from .trainer import Trainer
diff --git a/benchmark/PaddleOCR_DBNet/trainer/trainer.py b/benchmark/PaddleOCR_DBNet/trainer/trainer.py
new file mode 100644
index 0000000..077d39c
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/trainer/trainer.py
@@ -0,0 +1,256 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/8/23 21:58
+# @Author  : zhoujun
+import time
+
+import paddle
+from tqdm import tqdm
+
+from base import BaseTrainer
+from utils import runningScore, cal_text_score, Polynomial, profiler
+
+
+class Trainer(BaseTrainer):
+    def __init__(
+        self,
+        config,
+        model,
+        criterion,
+        train_loader,
+        validate_loader,
+        metric_cls,
+        post_process=None,
+        profiler_options=None,
+    ):
+        super(Trainer, self).__init__(
+            config,
+            model,
+            criterion,
+            train_loader,
+            validate_loader,
+            metric_cls,
+            post_process,
+        )
+        self.profiler_options = profiler_options
+        self.enable_eval = config["trainer"].get("enable_eval", True)
+
+    def _train_epoch(self, epoch):
+        self.model.train()
+        total_samples = 0
+        train_reader_cost = 0.0
+        train_batch_cost = 0.0
+        reader_start = time.time()
+        epoch_start = time.time()
+        train_loss = 0.0
+        running_metric_text = runningScore(2)
+
+        for i, batch in enumerate(self.train_loader):
+            profiler.add_profiler_step(self.profiler_options)
+            if i >= self.train_loader_len:
+                break
+            self.global_step += 1
+            lr = self.optimizer.get_lr()
+
+            cur_batch_size = batch["img"].shape[0]
+
+            train_reader_cost += time.time() - reader_start
+            if self.amp:
+                with paddle.amp.auto_cast(
+                    enable="gpu" in paddle.device.get_device(),
+                    custom_white_list=self.amp.get("custom_white_list", []),
+                    custom_black_list=self.amp.get("custom_black_list", []),
+                    level=self.amp.get("level", "O2"),
+                ):
+                    preds = self.model(batch["img"])
+                loss_dict = self.criterion(preds.astype(paddle.float32), batch)
+                scaled_loss = self.amp["scaler"].scale(loss_dict["loss"])
+                scaled_loss.backward()
+                self.amp["scaler"].minimize(self.optimizer, scaled_loss)
+            else:
+                preds = self.model(batch["img"])
+                loss_dict = self.criterion(preds, batch)
+                # backward
+                loss_dict["loss"].backward()
+                self.optimizer.step()
+            self.lr_scheduler.step()
+            self.optimizer.clear_grad()
+
+            train_batch_time = time.time() - reader_start
+            train_batch_cost += train_batch_time
+            total_samples += cur_batch_size
+
+            # acc iou
+            score_shrink_map = cal_text_score(
+                preds[:, 0, :, :],
+                batch["shrink_map"],
+                batch["shrink_mask"],
+                running_metric_text,
+                thred=self.config["post_processing"]["args"]["thresh"],
+            )
+
+            # loss 和 acc 记录到日志
+            loss_str = "loss: {:.4f}, ".format(loss_dict["loss"].item())
+            for idx, (key, value) in enumerate(loss_dict.items()):
+                loss_dict[key] = value.item()
+                if key == "loss":
+                    continue
+                loss_str += "{}: {:.4f}".format(key, loss_dict[key])
+                if idx < len(loss_dict) - 1:
+                    loss_str += ", "
+
+            train_loss += loss_dict["loss"]
+            acc = score_shrink_map["Mean Acc"]
+            iou_shrink_map = score_shrink_map["Mean IoU"]
+
+            if self.global_step % self.log_iter == 0:
+                self.logger_info(
+                    "[{}/{}], [{}/{}], global_step: {}, ips: {:.1f} samples/sec, avg_reader_cost: {:.5f} s, avg_batch_cost: {:.5f} s, avg_samples: {}, acc: {:.4f}, iou_shrink_map: {:.4f}, {}lr:{:.6}, time:{:.2f}".format(
+                        epoch,
+                        self.epochs,
+                        i + 1,
+                        self.train_loader_len,
+                        self.global_step,
+                        total_samples / train_batch_cost,
+                        train_reader_cost / self.log_iter,
+                        train_batch_cost / self.log_iter,
+                        total_samples / self.log_iter,
+                        acc,
+                        iou_shrink_map,
+                        loss_str,
+                        lr,
+                        train_batch_cost,
+                    )
+                )
+                total_samples = 0
+                train_reader_cost = 0.0
+                train_batch_cost = 0.0
+
+            if self.visualdl_enable and paddle.distributed.get_rank() == 0:
+                # write tensorboard
+                for key, value in loss_dict.items():
+                    self.writer.add_scalar(
+                        "TRAIN/LOSS/{}".format(key), value, self.global_step
+                    )
+                self.writer.add_scalar("TRAIN/ACC_IOU/acc", acc, self.global_step)
+                self.writer.add_scalar(
+                    "TRAIN/ACC_IOU/iou_shrink_map", iou_shrink_map, self.global_step
+                )
+                self.writer.add_scalar("TRAIN/lr", lr, self.global_step)
+            reader_start = time.time()
+        return {
+            "train_loss": train_loss / self.train_loader_len,
+            "lr": lr,
+            "time": time.time() - epoch_start,
+            "epoch": epoch,
+        }
+
+    def _eval(self, epoch):
+        self.model.eval()
+        raw_metrics = []
+        total_frame = 0.0
+        total_time = 0.0
+        for i, batch in tqdm(
+            enumerate(self.validate_loader),
+            total=len(self.validate_loader),
+            desc="test model",
+        ):
+            with paddle.no_grad():
+                start = time.time()
+                if self.amp:
+                    with paddle.amp.auto_cast(
+                        enable="gpu" in paddle.device.get_device(),
+                        custom_white_list=self.amp.get("custom_white_list", []),
+                        custom_black_list=self.amp.get("custom_black_list", []),
+                        level=self.amp.get("level", "O2"),
+                    ):
+                        preds = self.model(batch["img"])
+                    preds = preds.astype(paddle.float32)
+                else:
+                    preds = self.model(batch["img"])
+                boxes, scores = self.post_process(
+                    batch, preds, is_output_polygon=self.metric_cls.is_output_polygon
+                )
+                total_frame += batch["img"].shape[0]
+                total_time += time.time() - start
+                raw_metric = self.metric_cls.validate_measure(batch, (boxes, scores))
+                raw_metrics.append(raw_metric)
+        metrics = self.metric_cls.gather_measure(raw_metrics)
+        self.logger_info("FPS:{}".format(total_frame / total_time))
+        return metrics["recall"].avg, metrics["precision"].avg, metrics["fmeasure"].avg
+
+    def _on_epoch_finish(self):
+        self.logger_info(
+            "[{}/{}], train_loss: {:.4f}, time: {:.4f}, lr: {}".format(
+                self.epoch_result["epoch"],
+                self.epochs,
+                self.epoch_result["train_loss"],
+                self.epoch_result["time"],
+                self.epoch_result["lr"],
+            )
+        )
+        net_save_path = "{}/model_latest.pth".format(self.checkpoint_dir)
+        net_save_path_best = "{}/model_best.pth".format(self.checkpoint_dir)
+
+        if paddle.distributed.get_rank() == 0:
+            self._save_checkpoint(self.epoch_result["epoch"], net_save_path)
+            save_best = False
+            if (
+                self.validate_loader is not None
+                and self.metric_cls is not None
+                and self.enable_eval
+            ):  # 使用f1作为最优模型指标
+                recall, precision, hmean = self._eval(self.epoch_result["epoch"])
+
+                if self.visualdl_enable:
+                    self.writer.add_scalar("EVAL/recall", recall, self.global_step)
+                    self.writer.add_scalar(
+                        "EVAL/precision", precision, self.global_step
+                    )
+                    self.writer.add_scalar("EVAL/hmean", hmean, self.global_step)
+                self.logger_info(
+                    "test: recall: {:.6f}, precision: {:.6f}, hmean: {:.6f}".format(
+                        recall, precision, hmean
+                    )
+                )
+
+                if hmean >= self.metrics["hmean"]:
+                    save_best = True
+                    self.metrics["train_loss"] = self.epoch_result["train_loss"]
+                    self.metrics["hmean"] = hmean
+                    self.metrics["precision"] = precision
+                    self.metrics["recall"] = recall
+                    self.metrics["best_model_epoch"] = self.epoch_result["epoch"]
+            else:
+                if self.epoch_result["train_loss"] <= self.metrics["train_loss"]:
+                    save_best = True
+                    self.metrics["train_loss"] = self.epoch_result["train_loss"]
+                    self.metrics["best_model_epoch"] = self.epoch_result["epoch"]
+            best_str = "current best, "
+            for k, v in self.metrics.items():
+                best_str += "{}: {:.6f}, ".format(k, v)
+            self.logger_info(best_str)
+            if save_best:
+                import shutil
+
+                shutil.copy(net_save_path, net_save_path_best)
+                self.logger_info("Saving current best: {}".format(net_save_path_best))
+            else:
+                self.logger_info("Saving checkpoint: {}".format(net_save_path))
+
+    def _on_train_finish(self):
+        if self.enable_eval:
+            for k, v in self.metrics.items():
+                self.logger_info("{}:{}".format(k, v))
+        self.logger_info("finish train")
+
+    def _initialize_scheduler(self):
+        if self.config["lr_scheduler"]["type"] == "Polynomial":
+            self.config["lr_scheduler"]["args"]["epochs"] = self.config["trainer"][
+                "epochs"
+            ]
+            self.config["lr_scheduler"]["args"]["step_each_epoch"] = len(
+                self.train_loader
+            )
+            self.lr_scheduler = Polynomial(**self.config["lr_scheduler"]["args"])()
+        else:
+            self.lr_scheduler = self._initialize("lr_scheduler", paddle.optimizer.lr)
diff --git a/benchmark/PaddleOCR_DBNet/utils/__init__.py b/benchmark/PaddleOCR_DBNet/utils/__init__.py
new file mode 100644
index 0000000..c7bf6b9
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/utils/__init__.py
@@ -0,0 +1,8 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/8/23 21:58
+# @Author  : zhoujun
+from .util import *
+from .metrics import *
+from .schedulers import *
+from .cal_recall.script import cal_recall_precision_f1
+from .ocr_metric import get_metric
diff --git a/benchmark/PaddleOCR_DBNet/utils/cal_recall/__init__.py b/benchmark/PaddleOCR_DBNet/utils/cal_recall/__init__.py
new file mode 100644
index 0000000..b5c75e4
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/utils/cal_recall/__init__.py
@@ -0,0 +1,6 @@
+# -*- coding: utf-8 -*-
+# @Time    : 1/16/19 6:40 AM
+# @Author  : zhoujun
+from .script import cal_recall_precision_f1
+
+__all__ = ["cal_recall_precision_f1"]
diff --git a/benchmark/PaddleOCR_DBNet/utils/cal_recall/rrc_evaluation_funcs.py b/benchmark/PaddleOCR_DBNet/utils/cal_recall/rrc_evaluation_funcs.py
new file mode 100644
index 0000000..426549b
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/utils/cal_recall/rrc_evaluation_funcs.py
@@ -0,0 +1,494 @@
+#!/usr/bin/env python2
+# encoding: UTF-8
+import json
+import sys
+
+sys.path.append("./")
+import zipfile
+import re
+import sys
+import os
+import codecs
+import traceback
+import numpy as np
+from utils import order_points_clockwise
+
+
+def print_help():
+    sys.stdout.write(
+        "Usage: python %s.py -g=<gtFile> -s=<submFile> [-o=<outputFolder> -p=<jsonParams>]"
+        % sys.argv[0]
+    )
+    sys.exit(2)
+
+
+def load_zip_file_keys(file, fileNameRegExp=""):
+    """
+    Returns an array with the entries of the ZIP file that match with the regular expression.
+    The key's are the names or the file or the capturing group defined in the fileNameRegExp
+    """
+    try:
+        archive = zipfile.ZipFile(file, mode="r", allowZip64=True)
+    except:
+        raise Exception("Error loading the ZIP archive.")
+
+    pairs = []
+
+    for name in archive.namelist():
+        addFile = True
+        keyName = name
+        if fileNameRegExp != "":
+            m = re.match(fileNameRegExp, name)
+            if m == None:
+                addFile = False
+            else:
+                if len(m.groups()) > 0:
+                    keyName = m.group(1)
+
+        if addFile:
+            pairs.append(keyName)
+
+    return pairs
+
+
+def load_zip_file(file, fileNameRegExp="", allEntries=False):
+    """
+    Returns an array with the contents (filtered by fileNameRegExp) of a ZIP file.
+    The key's are the names or the file or the capturing group defined in the fileNameRegExp
+    allEntries validates that all entries in the ZIP file pass the fileNameRegExp
+    """
+    try:
+        archive = zipfile.ZipFile(file, mode="r", allowZip64=True)
+    except:
+        raise Exception("Error loading the ZIP archive")
+
+    pairs = []
+    for name in archive.namelist():
+        addFile = True
+        keyName = name
+        if fileNameRegExp != "":
+            m = re.match(fileNameRegExp, name)
+            if m == None:
+                addFile = False
+            else:
+                if len(m.groups()) > 0:
+                    keyName = m.group(1)
+
+        if addFile:
+            pairs.append([keyName, archive.read(name)])
+        else:
+            if allEntries:
+                raise Exception("ZIP entry not valid: %s" % name)
+
+    return dict(pairs)
+
+
+def load_folder_file(file, fileNameRegExp="", allEntries=False):
+    """
+    Returns an array with the contents (filtered by fileNameRegExp) of a ZIP file.
+    The key's are the names or the file or the capturing group defined in the fileNameRegExp
+    allEntries validates that all entries in the ZIP file pass the fileNameRegExp
+    """
+    pairs = []
+    for name in os.listdir(file):
+        addFile = True
+        keyName = name
+        if fileNameRegExp != "":
+            m = re.match(fileNameRegExp, name)
+            if m == None:
+                addFile = False
+            else:
+                if len(m.groups()) > 0:
+                    keyName = m.group(1)
+
+        if addFile:
+            pairs.append([keyName, open(os.path.join(file, name)).read()])
+        else:
+            if allEntries:
+                raise Exception("ZIP entry not valid: %s" % name)
+
+    return dict(pairs)
+
+
+def decode_utf8(raw):
+    """
+    Returns a Unicode object on success, or None on failure
+    """
+    try:
+        raw = codecs.decode(raw, "utf-8", "replace")
+        # extracts BOM if exists
+        raw = raw.encode("utf8")
+        if raw.startswith(codecs.BOM_UTF8):
+            raw = raw.replace(codecs.BOM_UTF8, "", 1)
+        return raw.decode("utf-8")
+    except:
+        return None
+
+
+def validate_lines_in_file(
+    fileName,
+    file_contents,
+    CRLF=True,
+    LTRB=True,
+    withTranscription=False,
+    withConfidence=False,
+    imWidth=0,
+    imHeight=0,
+):
+    """
+    This function validates that all lines of the file calling the Line validation function for each line
+    """
+    utf8File = decode_utf8(file_contents)
+    if utf8File is None:
+        raise Exception("The file %s is not UTF-8" % fileName)
+
+    lines = utf8File.split("\r\n" if CRLF else "\n")
+    for line in lines:
+        line = line.replace("\r", "").replace("\n", "")
+        if line != "":
+            try:
+                validate_tl_line(
+                    line, LTRB, withTranscription, withConfidence, imWidth, imHeight
+                )
+            except Exception as e:
+                raise Exception(
+                    (
+                        "Line in sample not valid. Sample: %s Line: %s Error: %s"
+                        % (fileName, line, str(e))
+                    ).encode("utf-8", "replace")
+                )
+
+
+def validate_tl_line(
+    line, LTRB=True, withTranscription=True, withConfidence=True, imWidth=0, imHeight=0
+):
+    """
+    Validate the format of the line. If the line is not valid an exception will be raised.
+    If maxWidth and maxHeight are specified, all points must be inside the image bounds.
+    Possible values are:
+    LTRB=True: xmin,ymin,xmax,ymax[,confidence][,transcription]
+    LTRB=False: x1,y1,x2,y2,x3,y3,x4,y4[,confidence][,transcription]
+    """
+    get_tl_line_values(line, LTRB, withTranscription, withConfidence, imWidth, imHeight)
+
+
+def get_tl_line_values(
+    line,
+    LTRB=True,
+    withTranscription=False,
+    withConfidence=False,
+    imWidth=0,
+    imHeight=0,
+):
+    """
+    Validate the format of the line. If the line is not valid an exception will be raised.
+    If maxWidth and maxHeight are specified, all points must be inside the image bounds.
+    Possible values are:
+    LTRB=True: xmin,ymin,xmax,ymax[,confidence][,transcription]
+    LTRB=False: x1,y1,x2,y2,x3,y3,x4,y4[,confidence][,transcription]
+    Returns values from a textline. Points , [Confidences], [Transcriptions]
+    """
+    confidence = 0.0
+    transcription = ""
+    points = []
+
+    numPoints = 4
+
+    if LTRB:
+        numPoints = 4
+
+        if withTranscription and withConfidence:
+            m = re.match(
+                r"^\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*([0-9]+)\s*,\s*([0-9]+)\s*,\s*([0-1].?[0-9]*)\s*,(.*)$",
+                line,
+            )
+            if m == None:
+                m = re.match(
+                    r"^\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*([0-9]+)\s*,\s*([0-9]+)\s*,\s*([0-1].?[0-9]*)\s*,(.*)$",
+                    line,
+                )
+                raise Exception(
+                    "Format incorrect. Should be: xmin,ymin,xmax,ymax,confidence,transcription"
+                )
+        elif withConfidence:
+            m = re.match(
+                r"^\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*([0-9]+)\s*,\s*([0-9]+)\s*,\s*([0-1].?[0-9]*)\s*$",
+                line,
+            )
+            if m == None:
+                raise Exception(
+                    "Format incorrect. Should be: xmin,ymin,xmax,ymax,confidence"
+                )
+        elif withTranscription:
+            m = re.match(
+                r"^\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*([0-9]+)\s*,\s*([0-9]+)\s*,(.*)$",
+                line,
+            )
+            if m == None:
+                raise Exception(
+                    "Format incorrect. Should be: xmin,ymin,xmax,ymax,transcription"
+                )
+        else:
+            m = re.match(
+                r"^\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*([0-9]+)\s*,\s*([0-9]+)\s*,?\s*$",
+                line,
+            )
+            if m == None:
+                raise Exception("Format incorrect. Should be: xmin,ymin,xmax,ymax")
+
+        xmin = int(m.group(1))
+        ymin = int(m.group(2))
+        xmax = int(m.group(3))
+        ymax = int(m.group(4))
+        if xmax < xmin:
+            raise Exception("Xmax value (%s) not valid (Xmax < Xmin)." % (xmax))
+        if ymax < ymin:
+            raise Exception("Ymax value (%s)  not valid (Ymax < Ymin)." % (ymax))
+
+        points = [float(m.group(i)) for i in range(1, (numPoints + 1))]
+
+        if imWidth > 0 and imHeight > 0:
+            validate_point_inside_bounds(xmin, ymin, imWidth, imHeight)
+            validate_point_inside_bounds(xmax, ymax, imWidth, imHeight)
+
+    else:
+        numPoints = 8
+
+        if withTranscription and withConfidence:
+            m = re.match(
+                r"^\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*([0-1].?[0-9]*)\s*,(.*)$",
+                line,
+            )
+            if m == None:
+                raise Exception(
+                    "Format incorrect. Should be: x1,y1,x2,y2,x3,y3,x4,y4,confidence,transcription"
+                )
+        elif withConfidence:
+            m = re.match(
+                r"^\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*([0-1].?[0-9]*)\s*$",
+                line,
+            )
+            if m == None:
+                raise Exception(
+                    "Format incorrect. Should be: x1,y1,x2,y2,x3,y3,x4,y4,confidence"
+                )
+        elif withTranscription:
+            m = re.match(
+                r"^\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,(.*)$",
+                line,
+            )
+            if m == None:
+                raise Exception(
+                    "Format incorrect. Should be: x1,y1,x2,y2,x3,y3,x4,y4,transcription"
+                )
+        else:
+            m = re.match(
+                r"^\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*,\s*(-?[0-9]+)\s*$",
+                line,
+            )
+            if m == None:
+                raise Exception("Format incorrect. Should be: x1,y1,x2,y2,x3,y3,x4,y4")
+
+        points = [float(m.group(i)) for i in range(1, (numPoints + 1))]
+
+        points = order_points_clockwise(np.array(points).reshape(-1, 2)).reshape(-1)
+        validate_clockwise_points(points)
+
+        if imWidth > 0 and imHeight > 0:
+            validate_point_inside_bounds(points[0], points[1], imWidth, imHeight)
+            validate_point_inside_bounds(points[2], points[3], imWidth, imHeight)
+            validate_point_inside_bounds(points[4], points[5], imWidth, imHeight)
+            validate_point_inside_bounds(points[6], points[7], imWidth, imHeight)
+
+    if withConfidence:
+        try:
+            confidence = float(m.group(numPoints + 1))
+        except ValueError:
+            raise Exception("Confidence value must be a float")
+
+    if withTranscription:
+        posTranscription = numPoints + (2 if withConfidence else 1)
+        transcription = m.group(posTranscription)
+        m2 = re.match(r"^\s*\"(.*)\"\s*$", transcription)
+        if (
+            m2 != None
+        ):  # Transcription with double quotes, we extract the value and replace escaped characters
+            transcription = m2.group(1).replace("\\\\", "\\").replace('\\"', '"')
+
+    return points, confidence, transcription
+
+
+def validate_point_inside_bounds(x, y, imWidth, imHeight):
+    if x < 0 or x > imWidth:
+        raise Exception(
+            "X value (%s) not valid. Image dimensions: (%s,%s)"
+            % (xmin, imWidth, imHeight)
+        )
+    if y < 0 or y > imHeight:
+        raise Exception(
+            "Y value (%s)  not valid. Image dimensions: (%s,%s) Sample: %s Line:%s"
+            % (ymin, imWidth, imHeight)
+        )
+
+
+def validate_clockwise_points(points):
+    """
+    Validates that the points that the 4 points that dlimite a polygon are in clockwise order.
+    """
+
+    if len(points) != 8:
+        raise Exception("Points list not valid." + str(len(points)))
+
+    point = [
+        [int(points[0]), int(points[1])],
+        [int(points[2]), int(points[3])],
+        [int(points[4]), int(points[5])],
+        [int(points[6]), int(points[7])],
+    ]
+    edge = [
+        (point[1][0] - point[0][0]) * (point[1][1] + point[0][1]),
+        (point[2][0] - point[1][0]) * (point[2][1] + point[1][1]),
+        (point[3][0] - point[2][0]) * (point[3][1] + point[2][1]),
+        (point[0][0] - point[3][0]) * (point[0][1] + point[3][1]),
+    ]
+
+    summatory = edge[0] + edge[1] + edge[2] + edge[3]
+    if summatory > 0:
+        raise Exception(
+            "Points are not clockwise. The coordinates of bounding quadrilaterals have to be given in clockwise order. Regarding the correct interpretation of 'clockwise' remember that the image coordinate system used is the standard one, with the image origin at the upper left, the X axis extending to the right and Y axis extending downwards."
+        )
+
+
+def get_tl_line_values_from_file_contents(
+    content,
+    CRLF=True,
+    LTRB=True,
+    withTranscription=False,
+    withConfidence=False,
+    imWidth=0,
+    imHeight=0,
+    sort_by_confidences=True,
+):
+    """
+    Returns all points, confindences and transcriptions of a file in lists. Valid line formats:
+    xmin,ymin,xmax,ymax,[confidence],[transcription]
+    x1,y1,x2,y2,x3,y3,x4,y4,[confidence],[transcription]
+    """
+    pointsList = []
+    transcriptionsList = []
+    confidencesList = []
+
+    lines = content.split("\r\n" if CRLF else "\n")
+    for line in lines:
+        line = line.replace("\r", "").replace("\n", "")
+        if line != "":
+            points, confidence, transcription = get_tl_line_values(
+                line, LTRB, withTranscription, withConfidence, imWidth, imHeight
+            )
+            pointsList.append(points)
+            transcriptionsList.append(transcription)
+            confidencesList.append(confidence)
+
+    if withConfidence and len(confidencesList) > 0 and sort_by_confidences:
+        import numpy as np
+
+        sorted_ind = np.argsort(-np.array(confidencesList))
+        confidencesList = [confidencesList[i] for i in sorted_ind]
+        pointsList = [pointsList[i] for i in sorted_ind]
+        transcriptionsList = [transcriptionsList[i] for i in sorted_ind]
+
+    return pointsList, confidencesList, transcriptionsList
+
+
+def main_evaluation(
+    p,
+    default_evaluation_params_fn,
+    validate_data_fn,
+    evaluate_method_fn,
+    show_result=True,
+    per_sample=True,
+):
+    """
+    This process validates a method, evaluates it and if it succeed generates a ZIP file with a JSON entry for each sample.
+    Params:
+    p: Dictionary of parameters with the GT/submission locations. If None is passed, the parameters send by the system are used.
+    default_evaluation_params_fn: points to a function that returns a dictionary with the default parameters used for the evaluation
+    validate_data_fn: points to a method that validates the correct format of the submission
+    evaluate_method_fn: points to a function that evaluated the submission and return a Dictionary with the results
+    """
+    evalParams = default_evaluation_params_fn()
+    if "p" in p.keys():
+        evalParams.update(
+            p["p"] if isinstance(p["p"], dict) else json.loads(p["p"][1:-1])
+        )
+
+    resDict = {"calculated": True, "Message": "", "method": "{}", "per_sample": "{}"}
+    try:
+        # validate_data_fn(p['g'], p['s'], evalParams)
+        evalData = evaluate_method_fn(p["g"], p["s"], evalParams)
+        resDict.update(evalData)
+
+    except Exception as e:
+        traceback.print_exc()
+        resDict["Message"] = str(e)
+        resDict["calculated"] = False
+
+    if "o" in p:
+        if not os.path.exists(p["o"]):
+            os.makedirs(p["o"])
+
+        resultsOutputname = p["o"] + "/results.zip"
+        outZip = zipfile.ZipFile(resultsOutputname, mode="w", allowZip64=True)
+
+        del resDict["per_sample"]
+        if "output_items" in resDict.keys():
+            del resDict["output_items"]
+
+        outZip.writestr("method.json", json.dumps(resDict))
+
+    if not resDict["calculated"]:
+        if show_result:
+            sys.stderr.write("Error!\n" + resDict["Message"] + "\n\n")
+        if "o" in p:
+            outZip.close()
+        return resDict
+
+    if "o" in p:
+        if per_sample == True:
+            for k, v in evalData["per_sample"].items():
+                outZip.writestr(k + ".json", json.dumps(v))
+
+            if "output_items" in evalData.keys():
+                for k, v in evalData["output_items"].items():
+                    outZip.writestr(k, v)
+
+        outZip.close()
+
+    if show_result:
+        sys.stdout.write("Calculated!")
+        sys.stdout.write(json.dumps(resDict["method"]))
+
+    return resDict
+
+
+def main_validation(default_evaluation_params_fn, validate_data_fn):
+    """
+    This process validates a method
+    Params:
+    default_evaluation_params_fn: points to a function that returns a dictionary with the default parameters used for the evaluation
+    validate_data_fn: points to a method that validates the correct format of the submission
+    """
+    try:
+        p = dict([s[1:].split("=") for s in sys.argv[1:]])
+        evalParams = default_evaluation_params_fn()
+        if "p" in p.keys():
+            evalParams.update(
+                p["p"] if isinstance(p["p"], dict) else json.loads(p["p"][1:-1])
+            )
+
+        validate_data_fn(p["g"], p["s"], evalParams)
+        print("SUCCESS")
+        sys.exit(0)
+    except Exception as e:
+        print(str(e))
+        sys.exit(101)
diff --git a/benchmark/PaddleOCR_DBNet/utils/cal_recall/script.py b/benchmark/PaddleOCR_DBNet/utils/cal_recall/script.py
new file mode 100644
index 0000000..5e1552d
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/utils/cal_recall/script.py
@@ -0,0 +1,402 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+from collections import namedtuple
+from . import rrc_evaluation_funcs
+import Polygon as plg
+import numpy as np
+
+
+def default_evaluation_params():
+    """
+    default_evaluation_params: Default parameters to use for the validation and evaluation.
+    """
+    return {
+        "IOU_CONSTRAINT": 0.5,
+        "AREA_PRECISION_CONSTRAINT": 0.5,
+        "GT_SAMPLE_NAME_2_ID": "gt_img_([0-9]+).txt",
+        "DET_SAMPLE_NAME_2_ID": "res_img_([0-9]+).txt",
+        "LTRB": False,  # LTRB:2points(left,top,right,bottom) or 4 points(x1,y1,x2,y2,x3,y3,x4,y4)
+        "CRLF": False,  # Lines are delimited by Windows CRLF format
+        "CONFIDENCES": False,  # Detections must include confidence value. AP will be calculated
+        "PER_SAMPLE_RESULTS": True,  # Generate per sample results and produce data for visualization
+    }
+
+
+def validate_data(gtFilePath, submFilePath, evaluationParams):
+    """
+    Method validate_data: validates that all files in the results folder are correct (have the correct name contents).
+                            Validates also that there are no missing files in the folder.
+                            If some error detected, the method raises the error
+    """
+    gt = rrc_evaluation_funcs.load_folder_file(
+        gtFilePath, evaluationParams["GT_SAMPLE_NAME_2_ID"]
+    )
+
+    subm = rrc_evaluation_funcs.load_folder_file(
+        submFilePath, evaluationParams["DET_SAMPLE_NAME_2_ID"], True
+    )
+
+    # Validate format of GroundTruth
+    for k in gt:
+        rrc_evaluation_funcs.validate_lines_in_file(
+            k, gt[k], evaluationParams["CRLF"], evaluationParams["LTRB"], True
+        )
+
+    # Validate format of results
+    for k in subm:
+        if (k in gt) == False:
+            raise Exception("The sample %s not present in GT" % k)
+
+        rrc_evaluation_funcs.validate_lines_in_file(
+            k,
+            subm[k],
+            evaluationParams["CRLF"],
+            evaluationParams["LTRB"],
+            False,
+            evaluationParams["CONFIDENCES"],
+        )
+
+
+def evaluate_method(gtFilePath, submFilePath, evaluationParams):
+    """
+    Method evaluate_method: evaluate method and returns the results
+        Results. Dictionary with the following values:
+        - method (required)  Global method metrics. Ex: { 'Precision':0.8,'Recall':0.9 }
+        - samples (optional) Per sample metrics. Ex: {'sample1' : { 'Precision':0.8,'Recall':0.9 } , 'sample2' : { 'Precision':0.8,'Recall':0.9 }
+    """
+
+    def polygon_from_points(points):
+        """
+        Returns a Polygon object to use with the Polygon2 class from a list of 8 points: x1,y1,x2,y2,x3,y3,x4,y4
+        """
+        resBoxes = np.empty([1, 8], dtype="int32")
+        resBoxes[0, 0] = int(points[0])
+        resBoxes[0, 4] = int(points[1])
+        resBoxes[0, 1] = int(points[2])
+        resBoxes[0, 5] = int(points[3])
+        resBoxes[0, 2] = int(points[4])
+        resBoxes[0, 6] = int(points[5])
+        resBoxes[0, 3] = int(points[6])
+        resBoxes[0, 7] = int(points[7])
+        pointMat = resBoxes[0].reshape([2, 4]).T
+        return plg.Polygon(pointMat)
+
+    def rectangle_to_polygon(rect):
+        resBoxes = np.empty([1, 8], dtype="int32")
+        resBoxes[0, 0] = int(rect.xmin)
+        resBoxes[0, 4] = int(rect.ymax)
+        resBoxes[0, 1] = int(rect.xmin)
+        resBoxes[0, 5] = int(rect.ymin)
+        resBoxes[0, 2] = int(rect.xmax)
+        resBoxes[0, 6] = int(rect.ymin)
+        resBoxes[0, 3] = int(rect.xmax)
+        resBoxes[0, 7] = int(rect.ymax)
+
+        pointMat = resBoxes[0].reshape([2, 4]).T
+
+        return plg.Polygon(pointMat)
+
+    def rectangle_to_points(rect):
+        points = [
+            int(rect.xmin),
+            int(rect.ymax),
+            int(rect.xmax),
+            int(rect.ymax),
+            int(rect.xmax),
+            int(rect.ymin),
+            int(rect.xmin),
+            int(rect.ymin),
+        ]
+        return points
+
+    def get_union(pD, pG):
+        areaA = pD.area()
+        areaB = pG.area()
+        return areaA + areaB - get_intersection(pD, pG)
+
+    def get_intersection_over_union(pD, pG):
+        try:
+            return get_intersection(pD, pG) / get_union(pD, pG)
+        except:
+            return 0
+
+    def get_intersection(pD, pG):
+        pInt = pD & pG
+        if len(pInt) == 0:
+            return 0
+        return pInt.area()
+
+    def compute_ap(confList, matchList, numGtCare):
+        correct = 0
+        AP = 0
+        if len(confList) > 0:
+            confList = np.array(confList)
+            matchList = np.array(matchList)
+            sorted_ind = np.argsort(-confList)
+            confList = confList[sorted_ind]
+            matchList = matchList[sorted_ind]
+            for n in range(len(confList)):
+                match = matchList[n]
+                if match:
+                    correct += 1
+                    AP += float(correct) / (n + 1)
+
+            if numGtCare > 0:
+                AP /= numGtCare
+
+        return AP
+
+    perSampleMetrics = {}
+
+    matchedSum = 0
+
+    Rectangle = namedtuple("Rectangle", "xmin ymin xmax ymax")
+
+    gt = rrc_evaluation_funcs.load_folder_file(
+        gtFilePath, evaluationParams["GT_SAMPLE_NAME_2_ID"]
+    )
+    subm = rrc_evaluation_funcs.load_folder_file(
+        submFilePath, evaluationParams["DET_SAMPLE_NAME_2_ID"], True
+    )
+
+    numGlobalCareGt = 0
+    numGlobalCareDet = 0
+
+    arrGlobalConfidences = []
+    arrGlobalMatches = []
+
+    for resFile in gt:
+        gtFile = gt[resFile]  # rrc_evaluation_funcs.decode_utf8(gt[resFile])
+        recall = 0
+        precision = 0
+        hmean = 0
+
+        detMatched = 0
+
+        iouMat = np.empty([1, 1])
+
+        gtPols = []
+        detPols = []
+
+        gtPolPoints = []
+        detPolPoints = []
+
+        # Array of Ground Truth Polygons' keys marked as don't Care
+        gtDontCarePolsNum = []
+        # Array of Detected Polygons' matched with a don't Care GT
+        detDontCarePolsNum = []
+
+        pairs = []
+        detMatchedNums = []
+
+        arrSampleConfidences = []
+        arrSampleMatch = []
+        sampleAP = 0
+
+        evaluationLog = ""
+
+        (
+            pointsList,
+            _,
+            transcriptionsList,
+        ) = rrc_evaluation_funcs.get_tl_line_values_from_file_contents(
+            gtFile, evaluationParams["CRLF"], evaluationParams["LTRB"], True, False
+        )
+        for n in range(len(pointsList)):
+            points = pointsList[n]
+            transcription = transcriptionsList[n]
+            dontCare = transcription == "###"
+            if evaluationParams["LTRB"]:
+                gtRect = Rectangle(*points)
+                gtPol = rectangle_to_polygon(gtRect)
+            else:
+                gtPol = polygon_from_points(points)
+            gtPols.append(gtPol)
+            gtPolPoints.append(points)
+            if dontCare:
+                gtDontCarePolsNum.append(len(gtPols) - 1)
+
+        evaluationLog += (
+            "GT polygons: "
+            + str(len(gtPols))
+            + (
+                " (" + str(len(gtDontCarePolsNum)) + " don't care)\n"
+                if len(gtDontCarePolsNum) > 0
+                else "\n"
+            )
+        )
+
+        if resFile in subm:
+            detFile = subm[resFile]  # rrc_evaluation_funcs.decode_utf8(subm[resFile])
+
+            (
+                pointsList,
+                confidencesList,
+                _,
+            ) = rrc_evaluation_funcs.get_tl_line_values_from_file_contents(
+                detFile,
+                evaluationParams["CRLF"],
+                evaluationParams["LTRB"],
+                False,
+                evaluationParams["CONFIDENCES"],
+            )
+            for n in range(len(pointsList)):
+                points = pointsList[n]
+
+                if evaluationParams["LTRB"]:
+                    detRect = Rectangle(*points)
+                    detPol = rectangle_to_polygon(detRect)
+                else:
+                    detPol = polygon_from_points(points)
+                detPols.append(detPol)
+                detPolPoints.append(points)
+                if len(gtDontCarePolsNum) > 0:
+                    for dontCarePol in gtDontCarePolsNum:
+                        dontCarePol = gtPols[dontCarePol]
+                        intersected_area = get_intersection(dontCarePol, detPol)
+                        pdDimensions = detPol.area()
+                        precision = (
+                            0 if pdDimensions == 0 else intersected_area / pdDimensions
+                        )
+                        if precision > evaluationParams["AREA_PRECISION_CONSTRAINT"]:
+                            detDontCarePolsNum.append(len(detPols) - 1)
+                            break
+
+            evaluationLog += (
+                "DET polygons: "
+                + str(len(detPols))
+                + (
+                    " (" + str(len(detDontCarePolsNum)) + " don't care)\n"
+                    if len(detDontCarePolsNum) > 0
+                    else "\n"
+                )
+            )
+
+            if len(gtPols) > 0 and len(detPols) > 0:
+                # Calculate IoU and precision matrixs
+                outputShape = [len(gtPols), len(detPols)]
+                iouMat = np.empty(outputShape)
+                gtRectMat = np.zeros(len(gtPols), np.int8)
+                detRectMat = np.zeros(len(detPols), np.int8)
+                for gtNum in range(len(gtPols)):
+                    for detNum in range(len(detPols)):
+                        pG = gtPols[gtNum]
+                        pD = detPols[detNum]
+                        iouMat[gtNum, detNum] = get_intersection_over_union(pD, pG)
+
+                for gtNum in range(len(gtPols)):
+                    for detNum in range(len(detPols)):
+                        if (
+                            gtRectMat[gtNum] == 0
+                            and detRectMat[detNum] == 0
+                            and gtNum not in gtDontCarePolsNum
+                            and detNum not in detDontCarePolsNum
+                        ):
+                            if (
+                                iouMat[gtNum, detNum]
+                                > evaluationParams["IOU_CONSTRAINT"]
+                            ):
+                                gtRectMat[gtNum] = 1
+                                detRectMat[detNum] = 1
+                                detMatched += 1
+                                pairs.append({"gt": gtNum, "det": detNum})
+                                detMatchedNums.append(detNum)
+                                evaluationLog += (
+                                    "Match GT #"
+                                    + str(gtNum)
+                                    + " with Det #"
+                                    + str(detNum)
+                                    + "\n"
+                                )
+
+            if evaluationParams["CONFIDENCES"]:
+                for detNum in range(len(detPols)):
+                    if detNum not in detDontCarePolsNum:
+                        # we exclude the don't care detections
+                        match = detNum in detMatchedNums
+
+                        arrSampleConfidences.append(confidencesList[detNum])
+                        arrSampleMatch.append(match)
+
+                        arrGlobalConfidences.append(confidencesList[detNum])
+                        arrGlobalMatches.append(match)
+
+        numGtCare = len(gtPols) - len(gtDontCarePolsNum)
+        numDetCare = len(detPols) - len(detDontCarePolsNum)
+        if numGtCare == 0:
+            recall = float(1)
+            precision = float(0) if numDetCare > 0 else float(1)
+            sampleAP = precision
+        else:
+            recall = float(detMatched) / numGtCare
+            precision = 0 if numDetCare == 0 else float(detMatched) / numDetCare
+            if (
+                evaluationParams["CONFIDENCES"]
+                and evaluationParams["PER_SAMPLE_RESULTS"]
+            ):
+                sampleAP = compute_ap(arrSampleConfidences, arrSampleMatch, numGtCare)
+
+        hmean = (
+            0
+            if (precision + recall) == 0
+            else 2.0 * precision * recall / (precision + recall)
+        )
+
+        matchedSum += detMatched
+        numGlobalCareGt += numGtCare
+        numGlobalCareDet += numDetCare
+
+        if evaluationParams["PER_SAMPLE_RESULTS"]:
+            perSampleMetrics[resFile] = {
+                "precision": precision,
+                "recall": recall,
+                "hmean": hmean,
+                "pairs": pairs,
+                "AP": sampleAP,
+                "iouMat": [] if len(detPols) > 100 else iouMat.tolist(),
+                "gtPolPoints": gtPolPoints,
+                "detPolPoints": detPolPoints,
+                "gtDontCare": gtDontCarePolsNum,
+                "detDontCare": detDontCarePolsNum,
+                "evaluationParams": evaluationParams,
+                "evaluationLog": evaluationLog,
+            }
+
+    # Compute MAP and MAR
+    AP = 0
+    if evaluationParams["CONFIDENCES"]:
+        AP = compute_ap(arrGlobalConfidences, arrGlobalMatches, numGlobalCareGt)
+
+    methodRecall = 0 if numGlobalCareGt == 0 else float(matchedSum) / numGlobalCareGt
+    methodPrecision = (
+        0 if numGlobalCareDet == 0 else float(matchedSum) / numGlobalCareDet
+    )
+    methodHmean = (
+        0
+        if methodRecall + methodPrecision == 0
+        else 2 * methodRecall * methodPrecision / (methodRecall + methodPrecision)
+    )
+
+    methodMetrics = {
+        "precision": methodPrecision,
+        "recall": methodRecall,
+        "hmean": methodHmean,
+        "AP": AP,
+    }
+
+    resDict = {
+        "calculated": True,
+        "Message": "",
+        "method": methodMetrics,
+        "per_sample": perSampleMetrics,
+    }
+
+    return resDict
+
+
+def cal_recall_precision_f1(gt_path, result_path, show_result=False):
+    p = {"g": gt_path, "s": result_path}
+    result = rrc_evaluation_funcs.main_evaluation(
+        p, default_evaluation_params, validate_data, evaluate_method, show_result
+    )
+    return result["method"]
diff --git a/benchmark/PaddleOCR_DBNet/utils/compute_mean_std.py b/benchmark/PaddleOCR_DBNet/utils/compute_mean_std.py
new file mode 100644
index 0000000..c76015e
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/utils/compute_mean_std.py
@@ -0,0 +1,47 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/12/7 14:46
+# @Author  : zhoujun
+
+import numpy as np
+import cv2
+import os
+import random
+from tqdm import tqdm
+
+# calculate means and std
+train_txt_path = "./train_val_list.txt"
+
+CNum = 10000  # 挑选多少图片进行计算
+
+img_h, img_w = 640, 640
+imgs = np.zeros([img_w, img_h, 3, 1])
+means, stdevs = [], []
+
+with open(train_txt_path, "r") as f:
+    lines = f.readlines()
+    random.shuffle(lines)  # shuffle , 随机挑选图片
+
+    for i in tqdm(range(CNum)):
+        img_path = lines[i].split("\t")[0]
+
+        img = cv2.imread(img_path)
+        img = cv2.resize(img, (img_h, img_w))
+        img = img[:, :, :, np.newaxis]
+
+        imgs = np.concatenate((imgs, img), axis=3)
+#         print(i)
+
+imgs = imgs.astype(np.float32) / 255.0
+
+for i in tqdm(range(3)):
+    pixels = imgs[:, :, i, :].ravel()  # 拉成一行
+    means.append(np.mean(pixels))
+    stdevs.append(np.std(pixels))
+
+# cv2 读取的图像格式为BGR，PIL/Skimage读取到的都是RGB不用转
+means.reverse()  # BGR --> RGB
+stdevs.reverse()
+
+print("normMean = {}".format(means))
+print("normStd = {}".format(stdevs))
+print("transforms.Normalize(normMean = {}, normStd = {})".format(means, stdevs))
diff --git a/benchmark/PaddleOCR_DBNet/utils/make_trainfile.py b/benchmark/PaddleOCR_DBNet/utils/make_trainfile.py
new file mode 100644
index 0000000..3232b01
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/utils/make_trainfile.py
@@ -0,0 +1,21 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/8/24 12:06
+# @Author  : zhoujun
+import os
+import glob
+import pathlib
+
+data_path = r"test"
+# data_path/img 存放图片
+# data_path/gt 存放标签文件
+
+f_w = open(os.path.join(data_path, "test.txt"), "w", encoding="utf8")
+for img_path in glob.glob(data_path + "/img/*.jpg", recursive=True):
+    d = pathlib.Path(img_path)
+    label_path = os.path.join(data_path, "gt", ("gt_" + str(d.stem) + ".txt"))
+    if os.path.exists(img_path) and os.path.exists(label_path):
+        print(img_path, label_path)
+    else:
+        print("不存在", img_path, label_path)
+    f_w.write("{}\t{}\n".format(img_path, label_path))
+f_w.close()
diff --git a/benchmark/PaddleOCR_DBNet/utils/metrics.py b/benchmark/PaddleOCR_DBNet/utils/metrics.py
new file mode 100644
index 0000000..81aa9bb
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/utils/metrics.py
@@ -0,0 +1,60 @@
+# Adapted from score written by wkentaro
+# https://github.com/wkentaro/pytorch-fcn/blob/master/torchfcn/utils.py
+
+import numpy as np
+
+
+class runningScore(object):
+    def __init__(self, n_classes):
+        self.n_classes = n_classes
+        self.confusion_matrix = np.zeros((n_classes, n_classes))
+
+    def _fast_hist(self, label_true, label_pred, n_class):
+        mask = (label_true >= 0) & (label_true < n_class)
+
+        if np.sum((label_pred[mask] < 0)) > 0:
+            print(label_pred[label_pred < 0])
+        hist = np.bincount(
+            n_class * label_true[mask].astype(int) + label_pred[mask],
+            minlength=n_class**2,
+        ).reshape(n_class, n_class)
+        return hist
+
+    def update(self, label_trues, label_preds):
+        # print label_trues.dtype, label_preds.dtype
+        for lt, lp in zip(label_trues, label_preds):
+            try:
+                self.confusion_matrix += self._fast_hist(
+                    lt.flatten(), lp.flatten(), self.n_classes
+                )
+            except:
+                pass
+
+    def get_scores(self):
+        """Returns accuracy score evaluation result.
+        - overall accuracy
+        - mean accuracy
+        - mean IU
+        - fwavacc
+        """
+        hist = self.confusion_matrix
+        acc = np.diag(hist).sum() / (hist.sum() + 0.0001)
+        acc_cls = np.diag(hist) / (hist.sum(axis=1) + 0.0001)
+        acc_cls = np.nanmean(acc_cls)
+        iu = np.diag(hist) / (
+            hist.sum(axis=1) + hist.sum(axis=0) - np.diag(hist) + 0.0001
+        )
+        mean_iu = np.nanmean(iu)
+        freq = hist.sum(axis=1) / (hist.sum() + 0.0001)
+        fwavacc = (freq[freq > 0] * iu[freq > 0]).sum()
+        cls_iu = dict(zip(range(self.n_classes), iu))
+
+        return {
+            "Overall Acc": acc,
+            "Mean Acc": acc_cls,
+            "FreqW Acc": fwavacc,
+            "Mean IoU": mean_iu,
+        }, cls_iu
+
+    def reset(self):
+        self.confusion_matrix = np.zeros((self.n_classes, self.n_classes))
diff --git a/benchmark/PaddleOCR_DBNet/utils/ocr_metric/__init__.py b/benchmark/PaddleOCR_DBNet/utils/ocr_metric/__init__.py
new file mode 100644
index 0000000..005f39c
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/utils/ocr_metric/__init__.py
@@ -0,0 +1,19 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/12/5 15:36
+# @Author  : zhoujun
+from .icdar2015 import QuadMetric
+
+
+def get_metric(config):
+    try:
+        if "args" not in config:
+            args = {}
+        else:
+            args = config["args"]
+        if isinstance(args, dict):
+            cls = eval(config["type"])(**args)
+        else:
+            cls = eval(config["type"])(args)
+        return cls
+    except:
+        return None
diff --git a/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/__init__.py b/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/__init__.py
new file mode 100644
index 0000000..5fd0e45
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/__init__.py
@@ -0,0 +1,5 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/12/5 15:36
+# @Author  : zhoujun
+
+from .quad_metric import QuadMetric
diff --git a/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/detection/__init__.py b/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/detection/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/detection/deteval.py b/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/detection/deteval.py
new file mode 100644
index 0000000..7b799ec
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/detection/deteval.py
@@ -0,0 +1,474 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+import math
+from collections import namedtuple
+import numpy as np
+from shapely.geometry import Polygon
+
+
+class DetectionDetEvalEvaluator(object):
+    def __init__(
+        self,
+        area_recall_constraint=0.8,
+        area_precision_constraint=0.4,
+        ev_param_ind_center_diff_thr=1,
+        mtype_oo_o=1.0,
+        mtype_om_o=0.8,
+        mtype_om_m=1.0,
+    ):
+        self.area_recall_constraint = area_recall_constraint
+        self.area_precision_constraint = area_precision_constraint
+        self.ev_param_ind_center_diff_thr = ev_param_ind_center_diff_thr
+        self.mtype_oo_o = mtype_oo_o
+        self.mtype_om_o = mtype_om_o
+        self.mtype_om_m = mtype_om_m
+
+    def evaluate_image(self, gt, pred):
+        def get_union(pD, pG):
+            return Polygon(pD).union(Polygon(pG)).area
+
+        def get_intersection_over_union(pD, pG):
+            return get_intersection(pD, pG) / get_union(pD, pG)
+
+        def get_intersection(pD, pG):
+            return Polygon(pD).intersection(Polygon(pG)).area
+
+        def one_to_one_match(row, col):
+            cont = 0
+            for j in range(len(recallMat[0])):
+                if (
+                    recallMat[row, j] >= self.area_recall_constraint
+                    and precisionMat[row, j] >= self.area_precision_constraint
+                ):
+                    cont = cont + 1
+            if cont != 1:
+                return False
+            cont = 0
+            for i in range(len(recallMat)):
+                if (
+                    recallMat[i, col] >= self.area_recall_constraint
+                    and precisionMat[i, col] >= self.area_precision_constraint
+                ):
+                    cont = cont + 1
+            if cont != 1:
+                return False
+
+            if (
+                recallMat[row, col] >= self.area_recall_constraint
+                and precisionMat[row, col] >= self.area_precision_constraint
+            ):
+                return True
+            return False
+
+        def num_overlaps_gt(gtNum):
+            cont = 0
+            for detNum in range(len(detRects)):
+                if detNum not in detDontCareRectsNum:
+                    if recallMat[gtNum, detNum] > 0:
+                        cont = cont + 1
+            return cont
+
+        def num_overlaps_det(detNum):
+            cont = 0
+            for gtNum in range(len(recallMat)):
+                if gtNum not in gtDontCareRectsNum:
+                    if recallMat[gtNum, detNum] > 0:
+                        cont = cont + 1
+            return cont
+
+        def is_single_overlap(row, col):
+            if num_overlaps_gt(row) == 1 and num_overlaps_det(col) == 1:
+                return True
+            else:
+                return False
+
+        def one_to_many_match(gtNum):
+            many_sum = 0
+            detRects = []
+            for detNum in range(len(recallMat[0])):
+                if (
+                    gtRectMat[gtNum] == 0
+                    and detRectMat[detNum] == 0
+                    and detNum not in detDontCareRectsNum
+                ):
+                    if precisionMat[gtNum, detNum] >= self.area_precision_constraint:
+                        many_sum += recallMat[gtNum, detNum]
+                        detRects.append(detNum)
+            if round(many_sum, 4) >= self.area_recall_constraint:
+                return True, detRects
+            else:
+                return False, []
+
+        def many_to_one_match(detNum):
+            many_sum = 0
+            gtRects = []
+            for gtNum in range(len(recallMat)):
+                if (
+                    gtRectMat[gtNum] == 0
+                    and detRectMat[detNum] == 0
+                    and gtNum not in gtDontCareRectsNum
+                ):
+                    if recallMat[gtNum, detNum] >= self.area_recall_constraint:
+                        many_sum += precisionMat[gtNum, detNum]
+                        gtRects.append(gtNum)
+            if round(many_sum, 4) >= self.area_precision_constraint:
+                return True, gtRects
+            else:
+                return False, []
+
+        def center_distance(r1, r2):
+            return ((np.mean(r1, axis=0) - np.mean(r2, axis=0)) ** 2).sum() ** 0.5
+
+        def diag(r):
+            r = np.array(r)
+            return (
+                (r[:, 0].max() - r[:, 0].min()) ** 2
+                + (r[:, 1].max() - r[:, 1].min()) ** 2
+            ) ** 0.5
+
+        perSampleMetrics = {}
+
+        recall = 0
+        precision = 0
+        hmean = 0
+        recallAccum = 0.0
+        precisionAccum = 0.0
+        gtRects = []
+        detRects = []
+        gtPolPoints = []
+        detPolPoints = []
+        gtDontCareRectsNum = (
+            []
+        )  # Array of Ground Truth Rectangles' keys marked as don't Care
+        detDontCareRectsNum = (
+            []
+        )  # Array of Detected Rectangles' matched with a don't Care GT
+        pairs = []
+        evaluationLog = ""
+
+        recallMat = np.empty([1, 1])
+        precisionMat = np.empty([1, 1])
+
+        for n in range(len(gt)):
+            points = gt[n]["points"]
+            # transcription = gt[n]['text']
+            dontCare = gt[n]["ignore"]
+
+            if not Polygon(points).is_valid or not Polygon(points).is_simple:
+                continue
+
+            gtRects.append(points)
+            gtPolPoints.append(points)
+            if dontCare:
+                gtDontCareRectsNum.append(len(gtRects) - 1)
+
+        evaluationLog += (
+            "GT rectangles: "
+            + str(len(gtRects))
+            + (
+                " (" + str(len(gtDontCareRectsNum)) + " don't care)\n"
+                if len(gtDontCareRectsNum) > 0
+                else "\n"
+            )
+        )
+
+        for n in range(len(pred)):
+            points = pred[n]["points"]
+
+            if not Polygon(points).is_valid or not Polygon(points).is_simple:
+                continue
+
+            detRect = points
+            detRects.append(detRect)
+            detPolPoints.append(points)
+            if len(gtDontCareRectsNum) > 0:
+                for dontCareRectNum in gtDontCareRectsNum:
+                    dontCareRect = gtRects[dontCareRectNum]
+                    intersected_area = get_intersection(dontCareRect, detRect)
+                    rdDimensions = Polygon(detRect).area
+                    if rdDimensions == 0:
+                        precision = 0
+                    else:
+                        precision = intersected_area / rdDimensions
+                    if precision > self.area_precision_constraint:
+                        detDontCareRectsNum.append(len(detRects) - 1)
+                        break
+
+        evaluationLog += (
+            "DET rectangles: "
+            + str(len(detRects))
+            + (
+                " (" + str(len(detDontCareRectsNum)) + " don't care)\n"
+                if len(detDontCareRectsNum) > 0
+                else "\n"
+            )
+        )
+
+        if len(gtRects) == 0:
+            recall = 1
+            precision = 0 if len(detRects) > 0 else 1
+
+        if len(detRects) > 0:
+            # Calculate recall and precision matrixes
+            outputShape = [len(gtRects), len(detRects)]
+            recallMat = np.empty(outputShape)
+            precisionMat = np.empty(outputShape)
+            gtRectMat = np.zeros(len(gtRects), np.int8)
+            detRectMat = np.zeros(len(detRects), np.int8)
+            for gtNum in range(len(gtRects)):
+                for detNum in range(len(detRects)):
+                    rG = gtRects[gtNum]
+                    rD = detRects[detNum]
+                    intersected_area = get_intersection(rG, rD)
+                    rgDimensions = Polygon(rG).area
+                    rdDimensions = Polygon(rD).area
+                    recallMat[gtNum, detNum] = (
+                        0 if rgDimensions == 0 else intersected_area / rgDimensions
+                    )
+                    precisionMat[gtNum, detNum] = (
+                        0 if rdDimensions == 0 else intersected_area / rdDimensions
+                    )
+
+            # Find one-to-one matches
+            evaluationLog += "Find one-to-one matches\n"
+            for gtNum in range(len(gtRects)):
+                for detNum in range(len(detRects)):
+                    if (
+                        gtRectMat[gtNum] == 0
+                        and detRectMat[detNum] == 0
+                        and gtNum not in gtDontCareRectsNum
+                        and detNum not in detDontCareRectsNum
+                    ):
+                        match = one_to_one_match(gtNum, detNum)
+                        if match is True:
+                            # in deteval we have to make other validation before mark as one-to-one
+                            if is_single_overlap(gtNum, detNum) is True:
+                                rG = gtRects[gtNum]
+                                rD = detRects[detNum]
+                                normDist = center_distance(rG, rD)
+                                normDist /= diag(rG) + diag(rD)
+                                normDist *= 2.0
+                                if normDist < self.ev_param_ind_center_diff_thr:
+                                    gtRectMat[gtNum] = 1
+                                    detRectMat[detNum] = 1
+                                    recallAccum += self.mtype_oo_o
+                                    precisionAccum += self.mtype_oo_o
+                                    pairs.append(
+                                        {"gt": gtNum, "det": detNum, "type": "OO"}
+                                    )
+                                    evaluationLog += (
+                                        "Match GT #"
+                                        + str(gtNum)
+                                        + " with Det #"
+                                        + str(detNum)
+                                        + "\n"
+                                    )
+                                else:
+                                    evaluationLog += (
+                                        "Match Discarded GT #"
+                                        + str(gtNum)
+                                        + " with Det #"
+                                        + str(detNum)
+                                        + " normDist: "
+                                        + str(normDist)
+                                        + " \n"
+                                    )
+                            else:
+                                evaluationLog += (
+                                    "Match Discarded GT #"
+                                    + str(gtNum)
+                                    + " with Det #"
+                                    + str(detNum)
+                                    + " not single overlap\n"
+                                )
+            # Find one-to-many matches
+            evaluationLog += "Find one-to-many matches\n"
+            for gtNum in range(len(gtRects)):
+                if gtNum not in gtDontCareRectsNum:
+                    match, matchesDet = one_to_many_match(gtNum)
+                    if match is True:
+                        evaluationLog += "num_overlaps_gt=" + str(
+                            num_overlaps_gt(gtNum)
+                        )
+                        # in deteval we have to make other validation before mark as one-to-one
+                        if num_overlaps_gt(gtNum) >= 2:
+                            gtRectMat[gtNum] = 1
+                            recallAccum += (
+                                self.mtype_oo_o
+                                if len(matchesDet) == 1
+                                else self.mtype_om_o
+                            )
+                            precisionAccum += (
+                                self.mtype_oo_o
+                                if len(matchesDet) == 1
+                                else self.mtype_om_o * len(matchesDet)
+                            )
+                            pairs.append(
+                                {
+                                    "gt": gtNum,
+                                    "det": matchesDet,
+                                    "type": "OO" if len(matchesDet) == 1 else "OM",
+                                }
+                            )
+                            for detNum in matchesDet:
+                                detRectMat[detNum] = 1
+                            evaluationLog += (
+                                "Match GT #"
+                                + str(gtNum)
+                                + " with Det #"
+                                + str(matchesDet)
+                                + "\n"
+                            )
+                        else:
+                            evaluationLog += (
+                                "Match Discarded GT #"
+                                + str(gtNum)
+                                + " with Det #"
+                                + str(matchesDet)
+                                + " not single overlap\n"
+                            )
+
+            # Find many-to-one matches
+            evaluationLog += "Find many-to-one matches\n"
+            for detNum in range(len(detRects)):
+                if detNum not in detDontCareRectsNum:
+                    match, matchesGt = many_to_one_match(detNum)
+                    if match is True:
+                        # in deteval we have to make other validation before mark as one-to-one
+                        if num_overlaps_det(detNum) >= 2:
+                            detRectMat[detNum] = 1
+                            recallAccum += (
+                                self.mtype_oo_o
+                                if len(matchesGt) == 1
+                                else self.mtype_om_m * len(matchesGt)
+                            )
+                            precisionAccum += (
+                                self.mtype_oo_o
+                                if len(matchesGt) == 1
+                                else self.mtype_om_m
+                            )
+                            pairs.append(
+                                {
+                                    "gt": matchesGt,
+                                    "det": detNum,
+                                    "type": "OO" if len(matchesGt) == 1 else "MO",
+                                }
+                            )
+                            for gtNum in matchesGt:
+                                gtRectMat[gtNum] = 1
+                            evaluationLog += (
+                                "Match GT #"
+                                + str(matchesGt)
+                                + " with Det #"
+                                + str(detNum)
+                                + "\n"
+                            )
+                        else:
+                            evaluationLog += (
+                                "Match Discarded GT #"
+                                + str(matchesGt)
+                                + " with Det #"
+                                + str(detNum)
+                                + " not single overlap\n"
+                            )
+
+            numGtCare = len(gtRects) - len(gtDontCareRectsNum)
+            if numGtCare == 0:
+                recall = float(1)
+                precision = float(0) if len(detRects) > 0 else float(1)
+            else:
+                recall = float(recallAccum) / numGtCare
+                precision = (
+                    float(0)
+                    if (len(detRects) - len(detDontCareRectsNum)) == 0
+                    else float(precisionAccum)
+                    / (len(detRects) - len(detDontCareRectsNum))
+                )
+            hmean = (
+                0
+                if (precision + recall) == 0
+                else 2.0 * precision * recall / (precision + recall)
+            )
+
+        numGtCare = len(gtRects) - len(gtDontCareRectsNum)
+        numDetCare = len(detRects) - len(detDontCareRectsNum)
+
+        perSampleMetrics = {
+            "precision": precision,
+            "recall": recall,
+            "hmean": hmean,
+            "pairs": pairs,
+            "recallMat": [] if len(detRects) > 100 else recallMat.tolist(),
+            "precisionMat": [] if len(detRects) > 100 else precisionMat.tolist(),
+            "gtPolPoints": gtPolPoints,
+            "detPolPoints": detPolPoints,
+            "gtCare": numGtCare,
+            "detCare": numDetCare,
+            "gtDontCare": gtDontCareRectsNum,
+            "detDontCare": detDontCareRectsNum,
+            "recallAccum": recallAccum,
+            "precisionAccum": precisionAccum,
+            "evaluationLog": evaluationLog,
+        }
+
+        return perSampleMetrics
+
+    def combine_results(self, results):
+        numGt = 0
+        numDet = 0
+        methodRecallSum = 0
+        methodPrecisionSum = 0
+
+        for result in results:
+            numGt += result["gtCare"]
+            numDet += result["detCare"]
+            methodRecallSum += result["recallAccum"]
+            methodPrecisionSum += result["precisionAccum"]
+
+        methodRecall = 0 if numGt == 0 else methodRecallSum / numGt
+        methodPrecision = 0 if numDet == 0 else methodPrecisionSum / numDet
+        methodHmean = (
+            0
+            if methodRecall + methodPrecision == 0
+            else 2 * methodRecall * methodPrecision / (methodRecall + methodPrecision)
+        )
+
+        methodMetrics = {
+            "precision": methodPrecision,
+            "recall": methodRecall,
+            "hmean": methodHmean,
+        }
+
+        return methodMetrics
+
+
+if __name__ == "__main__":
+    evaluator = DetectionDetEvalEvaluator()
+    gts = [
+        [
+            {
+                "points": [(0, 0), (1, 0), (1, 1), (0, 1)],
+                "text": 1234,
+                "ignore": False,
+            },
+            {
+                "points": [(2, 2), (3, 2), (3, 3), (2, 3)],
+                "text": 5678,
+                "ignore": True,
+            },
+        ]
+    ]
+    preds = [
+        [
+            {
+                "points": [(0.1, 0.1), (1, 0), (1, 1), (0, 1)],
+                "text": 123,
+                "ignore": False,
+            }
+        ]
+    ]
+    results = []
+    for gt, pred in zip(gts, preds):
+        results.append(evaluator.evaluate_image(gt, pred))
+    metrics = evaluator.combine_results(results)
+    print(metrics)
diff --git a/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/detection/icdar2013.py b/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/detection/icdar2013.py
new file mode 100644
index 0000000..1fd67b5
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/detection/icdar2013.py
@@ -0,0 +1,417 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+import math
+from collections import namedtuple
+import numpy as np
+from shapely.geometry import Polygon
+
+
+class DetectionICDAR2013Evaluator(object):
+    def __init__(
+        self,
+        area_recall_constraint=0.8,
+        area_precision_constraint=0.4,
+        ev_param_ind_center_diff_thr=1,
+        mtype_oo_o=1.0,
+        mtype_om_o=0.8,
+        mtype_om_m=1.0,
+    ):
+        self.area_recall_constraint = area_recall_constraint
+        self.area_precision_constraint = area_precision_constraint
+        self.ev_param_ind_center_diff_thr = ev_param_ind_center_diff_thr
+        self.mtype_oo_o = mtype_oo_o
+        self.mtype_om_o = mtype_om_o
+        self.mtype_om_m = mtype_om_m
+
+    def evaluate_image(self, gt, pred):
+        def get_union(pD, pG):
+            return Polygon(pD).union(Polygon(pG)).area
+
+        def get_intersection_over_union(pD, pG):
+            return get_intersection(pD, pG) / get_union(pD, pG)
+
+        def get_intersection(pD, pG):
+            return Polygon(pD).intersection(Polygon(pG)).area
+
+        def one_to_one_match(row, col):
+            cont = 0
+            for j in range(len(recallMat[0])):
+                if (
+                    recallMat[row, j] >= self.area_recall_constraint
+                    and precisionMat[row, j] >= self.area_precision_constraint
+                ):
+                    cont = cont + 1
+            if cont != 1:
+                return False
+            cont = 0
+            for i in range(len(recallMat)):
+                if (
+                    recallMat[i, col] >= self.area_recall_constraint
+                    and precisionMat[i, col] >= self.area_precision_constraint
+                ):
+                    cont = cont + 1
+            if cont != 1:
+                return False
+
+            if (
+                recallMat[row, col] >= self.area_recall_constraint
+                and precisionMat[row, col] >= self.area_precision_constraint
+            ):
+                return True
+            return False
+
+        def one_to_many_match(gtNum):
+            many_sum = 0
+            detRects = []
+            for detNum in range(len(recallMat[0])):
+                if (
+                    gtRectMat[gtNum] == 0
+                    and detRectMat[detNum] == 0
+                    and detNum not in detDontCareRectsNum
+                ):
+                    if precisionMat[gtNum, detNum] >= self.area_precision_constraint:
+                        many_sum += recallMat[gtNum, detNum]
+                        detRects.append(detNum)
+            if round(many_sum, 4) >= self.area_recall_constraint:
+                return True, detRects
+            else:
+                return False, []
+
+        def many_to_one_match(detNum):
+            many_sum = 0
+            gtRects = []
+            for gtNum in range(len(recallMat)):
+                if (
+                    gtRectMat[gtNum] == 0
+                    and detRectMat[detNum] == 0
+                    and gtNum not in gtDontCareRectsNum
+                ):
+                    if recallMat[gtNum, detNum] >= self.area_recall_constraint:
+                        many_sum += precisionMat[gtNum, detNum]
+                        gtRects.append(gtNum)
+            if round(many_sum, 4) >= self.area_precision_constraint:
+                return True, gtRects
+            else:
+                return False, []
+
+        def center_distance(r1, r2):
+            return ((np.mean(r1, axis=0) - np.mean(r2, axis=0)) ** 2).sum() ** 0.5
+
+        def diag(r):
+            r = np.array(r)
+            return (
+                (r[:, 0].max() - r[:, 0].min()) ** 2
+                + (r[:, 1].max() - r[:, 1].min()) ** 2
+            ) ** 0.5
+
+        perSampleMetrics = {}
+
+        recall = 0
+        precision = 0
+        hmean = 0
+        recallAccum = 0.0
+        precisionAccum = 0.0
+        gtRects = []
+        detRects = []
+        gtPolPoints = []
+        detPolPoints = []
+        gtDontCareRectsNum = (
+            []
+        )  # Array of Ground Truth Rectangles' keys marked as don't Care
+        detDontCareRectsNum = (
+            []
+        )  # Array of Detected Rectangles' matched with a don't Care GT
+        pairs = []
+        evaluationLog = ""
+
+        recallMat = np.empty([1, 1])
+        precisionMat = np.empty([1, 1])
+
+        for n in range(len(gt)):
+            points = gt[n]["points"]
+            # transcription = gt[n]['text']
+            dontCare = gt[n]["ignore"]
+
+            if not Polygon(points).is_valid or not Polygon(points).is_simple:
+                continue
+
+            gtRects.append(points)
+            gtPolPoints.append(points)
+            if dontCare:
+                gtDontCareRectsNum.append(len(gtRects) - 1)
+
+        evaluationLog += (
+            "GT rectangles: "
+            + str(len(gtRects))
+            + (
+                " (" + str(len(gtDontCareRectsNum)) + " don't care)\n"
+                if len(gtDontCareRectsNum) > 0
+                else "\n"
+            )
+        )
+
+        for n in range(len(pred)):
+            points = pred[n]["points"]
+
+            if not Polygon(points).is_valid or not Polygon(points).is_simple:
+                continue
+
+            detRect = points
+            detRects.append(detRect)
+            detPolPoints.append(points)
+            if len(gtDontCareRectsNum) > 0:
+                for dontCareRectNum in gtDontCareRectsNum:
+                    dontCareRect = gtRects[dontCareRectNum]
+                    intersected_area = get_intersection(dontCareRect, detRect)
+                    rdDimensions = Polygon(detRect).area
+                    if rdDimensions == 0:
+                        precision = 0
+                    else:
+                        precision = intersected_area / rdDimensions
+                    if precision > self.area_precision_constraint:
+                        detDontCareRectsNum.append(len(detRects) - 1)
+                        break
+
+        evaluationLog += (
+            "DET rectangles: "
+            + str(len(detRects))
+            + (
+                " (" + str(len(detDontCareRectsNum)) + " don't care)\n"
+                if len(detDontCareRectsNum) > 0
+                else "\n"
+            )
+        )
+
+        if len(gtRects) == 0:
+            recall = 1
+            precision = 0 if len(detRects) > 0 else 1
+
+        if len(detRects) > 0:
+            # Calculate recall and precision matrixes
+            outputShape = [len(gtRects), len(detRects)]
+            recallMat = np.empty(outputShape)
+            precisionMat = np.empty(outputShape)
+            gtRectMat = np.zeros(len(gtRects), np.int8)
+            detRectMat = np.zeros(len(detRects), np.int8)
+            for gtNum in range(len(gtRects)):
+                for detNum in range(len(detRects)):
+                    rG = gtRects[gtNum]
+                    rD = detRects[detNum]
+                    intersected_area = get_intersection(rG, rD)
+                    rgDimensions = Polygon(rG).area
+                    rdDimensions = Polygon(rD).area
+                    recallMat[gtNum, detNum] = (
+                        0 if rgDimensions == 0 else intersected_area / rgDimensions
+                    )
+                    precisionMat[gtNum, detNum] = (
+                        0 if rdDimensions == 0 else intersected_area / rdDimensions
+                    )
+
+            # Find one-to-one matches
+            evaluationLog += "Find one-to-one matches\n"
+            for gtNum in range(len(gtRects)):
+                for detNum in range(len(detRects)):
+                    if (
+                        gtRectMat[gtNum] == 0
+                        and detRectMat[detNum] == 0
+                        and gtNum not in gtDontCareRectsNum
+                        and detNum not in detDontCareRectsNum
+                    ):
+                        match = one_to_one_match(gtNum, detNum)
+                        if match is True:
+                            # in deteval we have to make other validation before mark as one-to-one
+                            rG = gtRects[gtNum]
+                            rD = detRects[detNum]
+                            normDist = center_distance(rG, rD)
+                            normDist /= diag(rG) + diag(rD)
+                            normDist *= 2.0
+                            if normDist < self.ev_param_ind_center_diff_thr:
+                                gtRectMat[gtNum] = 1
+                                detRectMat[detNum] = 1
+                                recallAccum += self.mtype_oo_o
+                                precisionAccum += self.mtype_oo_o
+                                pairs.append({"gt": gtNum, "det": detNum, "type": "OO"})
+                                evaluationLog += (
+                                    "Match GT #"
+                                    + str(gtNum)
+                                    + " with Det #"
+                                    + str(detNum)
+                                    + "\n"
+                                )
+                            else:
+                                evaluationLog += (
+                                    "Match Discarded GT #"
+                                    + str(gtNum)
+                                    + " with Det #"
+                                    + str(detNum)
+                                    + " normDist: "
+                                    + str(normDist)
+                                    + " \n"
+                                )
+            # Find one-to-many matches
+            evaluationLog += "Find one-to-many matches\n"
+            for gtNum in range(len(gtRects)):
+                if gtNum not in gtDontCareRectsNum:
+                    match, matchesDet = one_to_many_match(gtNum)
+                    if match is True:
+                        evaluationLog += "num_overlaps_gt=" + str(
+                            num_overlaps_gt(gtNum)
+                        )
+                        gtRectMat[gtNum] = 1
+                        recallAccum += (
+                            self.mtype_oo_o if len(matchesDet) == 1 else self.mtype_om_o
+                        )
+                        precisionAccum += (
+                            self.mtype_oo_o
+                            if len(matchesDet) == 1
+                            else self.mtype_om_o * len(matchesDet)
+                        )
+                        pairs.append(
+                            {
+                                "gt": gtNum,
+                                "det": matchesDet,
+                                "type": "OO" if len(matchesDet) == 1 else "OM",
+                            }
+                        )
+                        for detNum in matchesDet:
+                            detRectMat[detNum] = 1
+                        evaluationLog += (
+                            "Match GT #"
+                            + str(gtNum)
+                            + " with Det #"
+                            + str(matchesDet)
+                            + "\n"
+                        )
+
+            # Find many-to-one matches
+            evaluationLog += "Find many-to-one matches\n"
+            for detNum in range(len(detRects)):
+                if detNum not in detDontCareRectsNum:
+                    match, matchesGt = many_to_one_match(detNum)
+                    if match is True:
+                        detRectMat[detNum] = 1
+                        recallAccum += (
+                            self.mtype_oo_o
+                            if len(matchesGt) == 1
+                            else self.mtype_om_m * len(matchesGt)
+                        )
+                        precisionAccum += (
+                            self.mtype_oo_o if len(matchesGt) == 1 else self.mtype_om_m
+                        )
+                        pairs.append(
+                            {
+                                "gt": matchesGt,
+                                "det": detNum,
+                                "type": "OO" if len(matchesGt) == 1 else "MO",
+                            }
+                        )
+                        for gtNum in matchesGt:
+                            gtRectMat[gtNum] = 1
+                        evaluationLog += (
+                            "Match GT #"
+                            + str(matchesGt)
+                            + " with Det #"
+                            + str(detNum)
+                            + "\n"
+                        )
+
+            numGtCare = len(gtRects) - len(gtDontCareRectsNum)
+            if numGtCare == 0:
+                recall = float(1)
+                precision = float(0) if len(detRects) > 0 else float(1)
+            else:
+                recall = float(recallAccum) / numGtCare
+                precision = (
+                    float(0)
+                    if (len(detRects) - len(detDontCareRectsNum)) == 0
+                    else float(precisionAccum)
+                    / (len(detRects) - len(detDontCareRectsNum))
+                )
+            hmean = (
+                0
+                if (precision + recall) == 0
+                else 2.0 * precision * recall / (precision + recall)
+            )
+
+        numGtCare = len(gtRects) - len(gtDontCareRectsNum)
+        numDetCare = len(detRects) - len(detDontCareRectsNum)
+
+        perSampleMetrics = {
+            "precision": precision,
+            "recall": recall,
+            "hmean": hmean,
+            "pairs": pairs,
+            "recallMat": [] if len(detRects) > 100 else recallMat.tolist(),
+            "precisionMat": [] if len(detRects) > 100 else precisionMat.tolist(),
+            "gtPolPoints": gtPolPoints,
+            "detPolPoints": detPolPoints,
+            "gtCare": numGtCare,
+            "detCare": numDetCare,
+            "gtDontCare": gtDontCareRectsNum,
+            "detDontCare": detDontCareRectsNum,
+            "recallAccum": recallAccum,
+            "precisionAccum": precisionAccum,
+            "evaluationLog": evaluationLog,
+        }
+
+        return perSampleMetrics
+
+    def combine_results(self, results):
+        numGt = 0
+        numDet = 0
+        methodRecallSum = 0
+        methodPrecisionSum = 0
+
+        for result in results:
+            numGt += result["gtCare"]
+            numDet += result["detCare"]
+            methodRecallSum += result["recallAccum"]
+            methodPrecisionSum += result["precisionAccum"]
+
+        methodRecall = 0 if numGt == 0 else methodRecallSum / numGt
+        methodPrecision = 0 if numDet == 0 else methodPrecisionSum / numDet
+        methodHmean = (
+            0
+            if methodRecall + methodPrecision == 0
+            else 2 * methodRecall * methodPrecision / (methodRecall + methodPrecision)
+        )
+
+        methodMetrics = {
+            "precision": methodPrecision,
+            "recall": methodRecall,
+            "hmean": methodHmean,
+        }
+
+        return methodMetrics
+
+
+if __name__ == "__main__":
+    evaluator = DetectionICDAR2013Evaluator()
+    gts = [
+        [
+            {
+                "points": [(0, 0), (1, 0), (1, 1), (0, 1)],
+                "text": 1234,
+                "ignore": False,
+            },
+            {
+                "points": [(2, 2), (3, 2), (3, 3), (2, 3)],
+                "text": 5678,
+                "ignore": True,
+            },
+        ]
+    ]
+    preds = [
+        [
+            {
+                "points": [(0.1, 0.1), (1, 0), (1, 1), (0, 1)],
+                "text": 123,
+                "ignore": False,
+            }
+        ]
+    ]
+    results = []
+    for gt, pred in zip(gts, preds):
+        results.append(evaluator.evaluate_image(gt, pred))
+    metrics = evaluator.combine_results(results)
+    print(metrics)
diff --git a/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/detection/iou.py b/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/detection/iou.py
new file mode 100644
index 0000000..85fde78
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/detection/iou.py
@@ -0,0 +1,300 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+from collections import namedtuple
+import numpy as np
+from shapely.geometry import Polygon
+import cv2
+
+
+def iou_rotate(box_a, box_b, method="union"):
+    rect_a = cv2.minAreaRect(box_a)
+    rect_b = cv2.minAreaRect(box_b)
+    r1 = cv2.rotatedRectangleIntersection(rect_a, rect_b)
+    if r1[0] == 0:
+        return 0
+    else:
+        inter_area = cv2.contourArea(r1[1])
+        area_a = cv2.contourArea(box_a)
+        area_b = cv2.contourArea(box_b)
+        union_area = area_a + area_b - inter_area
+        if union_area == 0 or inter_area == 0:
+            return 0
+        if method == "union":
+            iou = inter_area / union_area
+        elif method == "intersection":
+            iou = inter_area / min(area_a, area_b)
+        else:
+            raise NotImplementedError
+        return iou
+
+
+class DetectionIoUEvaluator(object):
+    def __init__(
+        self, is_output_polygon=False, iou_constraint=0.5, area_precision_constraint=0.5
+    ):
+        self.is_output_polygon = is_output_polygon
+        self.iou_constraint = iou_constraint
+        self.area_precision_constraint = area_precision_constraint
+
+    def evaluate_image(self, gt, pred):
+        def get_union(pD, pG):
+            return Polygon(pD).union(Polygon(pG)).area
+
+        def get_intersection_over_union(pD, pG):
+            return get_intersection(pD, pG) / get_union(pD, pG)
+
+        def get_intersection(pD, pG):
+            return Polygon(pD).intersection(Polygon(pG)).area
+
+        def compute_ap(confList, matchList, numGtCare):
+            correct = 0
+            AP = 0
+            if len(confList) > 0:
+                confList = np.array(confList)
+                matchList = np.array(matchList)
+                sorted_ind = np.argsort(-confList)
+                confList = confList[sorted_ind]
+                matchList = matchList[sorted_ind]
+                for n in range(len(confList)):
+                    match = matchList[n]
+                    if match:
+                        correct += 1
+                        AP += float(correct) / (n + 1)
+
+                if numGtCare > 0:
+                    AP /= numGtCare
+
+            return AP
+
+        perSampleMetrics = {}
+
+        matchedSum = 0
+
+        Rectangle = namedtuple("Rectangle", "xmin ymin xmax ymax")
+
+        numGlobalCareGt = 0
+        numGlobalCareDet = 0
+
+        arrGlobalConfidences = []
+        arrGlobalMatches = []
+
+        recall = 0
+        precision = 0
+        hmean = 0
+
+        detMatched = 0
+
+        iouMat = np.empty([1, 1])
+
+        gtPols = []
+        detPols = []
+
+        gtPolPoints = []
+        detPolPoints = []
+
+        # Array of Ground Truth Polygons' keys marked as don't Care
+        gtDontCarePolsNum = []
+        # Array of Detected Polygons' matched with a don't Care GT
+        detDontCarePolsNum = []
+
+        pairs = []
+        detMatchedNums = []
+
+        arrSampleConfidences = []
+        arrSampleMatch = []
+
+        evaluationLog = ""
+
+        for n in range(len(gt)):
+            points = gt[n]["points"]
+            # transcription = gt[n]['text']
+            dontCare = gt[n]["ignore"]
+
+            if not Polygon(points).is_valid or not Polygon(points).is_simple:
+                continue
+
+            gtPol = points
+            gtPols.append(gtPol)
+            gtPolPoints.append(points)
+            if dontCare:
+                gtDontCarePolsNum.append(len(gtPols) - 1)
+
+        evaluationLog += (
+            "GT polygons: "
+            + str(len(gtPols))
+            + (
+                " (" + str(len(gtDontCarePolsNum)) + " don't care)\n"
+                if len(gtDontCarePolsNum) > 0
+                else "\n"
+            )
+        )
+
+        for n in range(len(pred)):
+            points = pred[n]["points"]
+            if not Polygon(points).is_valid or not Polygon(points).is_simple:
+                continue
+
+            detPol = points
+            detPols.append(detPol)
+            detPolPoints.append(points)
+            if len(gtDontCarePolsNum) > 0:
+                for dontCarePol in gtDontCarePolsNum:
+                    dontCarePol = gtPols[dontCarePol]
+                    intersected_area = get_intersection(dontCarePol, detPol)
+                    pdDimensions = Polygon(detPol).area
+                    precision = (
+                        0 if pdDimensions == 0 else intersected_area / pdDimensions
+                    )
+                    if precision > self.area_precision_constraint:
+                        detDontCarePolsNum.append(len(detPols) - 1)
+                        break
+
+        evaluationLog += (
+            "DET polygons: "
+            + str(len(detPols))
+            + (
+                " (" + str(len(detDontCarePolsNum)) + " don't care)\n"
+                if len(detDontCarePolsNum) > 0
+                else "\n"
+            )
+        )
+
+        if len(gtPols) > 0 and len(detPols) > 0:
+            # Calculate IoU and precision matrixs
+            outputShape = [len(gtPols), len(detPols)]
+            iouMat = np.empty(outputShape)
+            gtRectMat = np.zeros(len(gtPols), np.int8)
+            detRectMat = np.zeros(len(detPols), np.int8)
+            if self.is_output_polygon:
+                for gtNum in range(len(gtPols)):
+                    for detNum in range(len(detPols)):
+                        pG = gtPols[gtNum]
+                        pD = detPols[detNum]
+                        iouMat[gtNum, detNum] = get_intersection_over_union(pD, pG)
+            else:
+                # gtPols = np.float32(gtPols)
+                # detPols = np.float32(detPols)
+                for gtNum in range(len(gtPols)):
+                    for detNum in range(len(detPols)):
+                        pG = np.float32(gtPols[gtNum])
+                        pD = np.float32(detPols[detNum])
+                        iouMat[gtNum, detNum] = iou_rotate(pD, pG)
+            for gtNum in range(len(gtPols)):
+                for detNum in range(len(detPols)):
+                    if (
+                        gtRectMat[gtNum] == 0
+                        and detRectMat[detNum] == 0
+                        and gtNum not in gtDontCarePolsNum
+                        and detNum not in detDontCarePolsNum
+                    ):
+                        if iouMat[gtNum, detNum] > self.iou_constraint:
+                            gtRectMat[gtNum] = 1
+                            detRectMat[detNum] = 1
+                            detMatched += 1
+                            pairs.append({"gt": gtNum, "det": detNum})
+                            detMatchedNums.append(detNum)
+                            evaluationLog += (
+                                "Match GT #"
+                                + str(gtNum)
+                                + " with Det #"
+                                + str(detNum)
+                                + "\n"
+                            )
+
+        numGtCare = len(gtPols) - len(gtDontCarePolsNum)
+        numDetCare = len(detPols) - len(detDontCarePolsNum)
+        if numGtCare == 0:
+            recall = float(1)
+            precision = float(0) if numDetCare > 0 else float(1)
+        else:
+            recall = float(detMatched) / numGtCare
+            precision = 0 if numDetCare == 0 else float(detMatched) / numDetCare
+
+        hmean = (
+            0
+            if (precision + recall) == 0
+            else 2.0 * precision * recall / (precision + recall)
+        )
+
+        matchedSum += detMatched
+        numGlobalCareGt += numGtCare
+        numGlobalCareDet += numDetCare
+
+        perSampleMetrics = {
+            "precision": precision,
+            "recall": recall,
+            "hmean": hmean,
+            "pairs": pairs,
+            "iouMat": [] if len(detPols) > 100 else iouMat.tolist(),
+            "gtPolPoints": gtPolPoints,
+            "detPolPoints": detPolPoints,
+            "gtCare": numGtCare,
+            "detCare": numDetCare,
+            "gtDontCare": gtDontCarePolsNum,
+            "detDontCare": detDontCarePolsNum,
+            "detMatched": detMatched,
+            "evaluationLog": evaluationLog,
+        }
+
+        return perSampleMetrics
+
+    def combine_results(self, results):
+        numGlobalCareGt = 0
+        numGlobalCareDet = 0
+        matchedSum = 0
+        for result in results:
+            numGlobalCareGt += result["gtCare"]
+            numGlobalCareDet += result["detCare"]
+            matchedSum += result["detMatched"]
+
+        methodRecall = (
+            0 if numGlobalCareGt == 0 else float(matchedSum) / numGlobalCareGt
+        )
+        methodPrecision = (
+            0 if numGlobalCareDet == 0 else float(matchedSum) / numGlobalCareDet
+        )
+        methodHmean = (
+            0
+            if methodRecall + methodPrecision == 0
+            else 2 * methodRecall * methodPrecision / (methodRecall + methodPrecision)
+        )
+
+        methodMetrics = {
+            "precision": methodPrecision,
+            "recall": methodRecall,
+            "hmean": methodHmean,
+        }
+
+        return methodMetrics
+
+
+if __name__ == "__main__":
+    evaluator = DetectionIoUEvaluator()
+    preds = [
+        [
+            {
+                "points": [(0.1, 0.1), (0.5, 0), (0.5, 1), (0, 1)],
+                "text": 1234,
+                "ignore": False,
+            },
+            {
+                "points": [(0.5, 0.1), (1, 0), (1, 1), (0.5, 1)],
+                "text": 5678,
+                "ignore": False,
+            },
+        ]
+    ]
+    gts = [
+        [
+            {
+                "points": [(0.1, 0.1), (1, 0), (1, 1), (0, 1)],
+                "text": 123,
+                "ignore": False,
+            }
+        ]
+    ]
+    results = []
+    for gt, pred in zip(gts, preds):
+        results.append(evaluator.evaluate_image(gt, pred))
+    metrics = evaluator.combine_results(results)
+    print(metrics)
diff --git a/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/detection/mtwi2018.py b/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/detection/mtwi2018.py
new file mode 100644
index 0000000..51eccdc
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/detection/mtwi2018.py
@@ -0,0 +1,398 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+import math
+from collections import namedtuple
+import numpy as np
+from shapely.geometry import Polygon
+
+
+class DetectionMTWI2018Evaluator(object):
+    def __init__(
+        self,
+        area_recall_constraint=0.7,
+        area_precision_constraint=0.7,
+        ev_param_ind_center_diff_thr=1,
+    ):
+        self.area_recall_constraint = area_recall_constraint
+        self.area_precision_constraint = area_precision_constraint
+        self.ev_param_ind_center_diff_thr = ev_param_ind_center_diff_thr
+
+    def evaluate_image(self, gt, pred):
+        def get_union(pD, pG):
+            return Polygon(pD).union(Polygon(pG)).area
+
+        def get_intersection_over_union(pD, pG):
+            return get_intersection(pD, pG) / get_union(pD, pG)
+
+        def get_intersection(pD, pG):
+            return Polygon(pD).intersection(Polygon(pG)).area
+
+        def one_to_one_match(row, col):
+            cont = 0
+            for j in range(len(recallMat[0])):
+                if (
+                    recallMat[row, j] >= self.area_recall_constraint
+                    and precisionMat[row, j] >= self.area_precision_constraint
+                ):
+                    cont = cont + 1
+            if cont != 1:
+                return False
+            cont = 0
+            for i in range(len(recallMat)):
+                if (
+                    recallMat[i, col] >= self.area_recall_constraint
+                    and precisionMat[i, col] >= self.area_precision_constraint
+                ):
+                    cont = cont + 1
+            if cont != 1:
+                return False
+
+            if (
+                recallMat[row, col] >= self.area_recall_constraint
+                and precisionMat[row, col] >= self.area_precision_constraint
+            ):
+                return True
+            return False
+
+        def one_to_many_match(gtNum):
+            many_sum = 0
+            detRects = []
+            for detNum in range(len(recallMat[0])):
+                if (
+                    gtRectMat[gtNum] == 0
+                    and detRectMat[detNum] == 0
+                    and detNum not in detDontCareRectsNum
+                ):
+                    if precisionMat[gtNum, detNum] >= self.area_precision_constraint:
+                        many_sum += recallMat[gtNum, detNum]
+                        detRects.append(detNum)
+            if round(many_sum, 4) >= self.area_recall_constraint:
+                return True, detRects
+            else:
+                return False, []
+
+        def many_to_one_match(detNum):
+            many_sum = 0
+            gtRects = []
+            for gtNum in range(len(recallMat)):
+                if (
+                    gtRectMat[gtNum] == 0
+                    and detRectMat[detNum] == 0
+                    and gtNum not in gtDontCareRectsNum
+                ):
+                    if recallMat[gtNum, detNum] >= self.area_recall_constraint:
+                        many_sum += precisionMat[gtNum, detNum]
+                        gtRects.append(gtNum)
+            if round(many_sum, 4) >= self.area_precision_constraint:
+                return True, gtRects
+            else:
+                return False, []
+
+        def center_distance(r1, r2):
+            return ((np.mean(r1, axis=0) - np.mean(r2, axis=0)) ** 2).sum() ** 0.5
+
+        def diag(r):
+            r = np.array(r)
+            return (
+                (r[:, 0].max() - r[:, 0].min()) ** 2
+                + (r[:, 1].max() - r[:, 1].min()) ** 2
+            ) ** 0.5
+
+        perSampleMetrics = {}
+
+        recall = 0
+        precision = 0
+        hmean = 0
+        recallAccum = 0.0
+        precisionAccum = 0.0
+        gtRects = []
+        detRects = []
+        gtPolPoints = []
+        detPolPoints = []
+        gtDontCareRectsNum = (
+            []
+        )  # Array of Ground Truth Rectangles' keys marked as don't Care
+        detDontCareRectsNum = (
+            []
+        )  # Array of Detected Rectangles' matched with a don't Care GT
+        pairs = []
+        evaluationLog = ""
+
+        recallMat = np.empty([1, 1])
+        precisionMat = np.empty([1, 1])
+
+        for n in range(len(gt)):
+            points = gt[n]["points"]
+            # transcription = gt[n]['text']
+            dontCare = gt[n]["ignore"]
+
+            if not Polygon(points).is_valid or not Polygon(points).is_simple:
+                continue
+
+            gtRects.append(points)
+            gtPolPoints.append(points)
+            if dontCare:
+                gtDontCareRectsNum.append(len(gtRects) - 1)
+
+        evaluationLog += (
+            "GT rectangles: "
+            + str(len(gtRects))
+            + (
+                " (" + str(len(gtDontCareRectsNum)) + " don't care)\n"
+                if len(gtDontCareRectsNum) > 0
+                else "\n"
+            )
+        )
+
+        for n in range(len(pred)):
+            points = pred[n]["points"]
+
+            if not Polygon(points).is_valid or not Polygon(points).is_simple:
+                continue
+
+            detRect = points
+            detRects.append(detRect)
+            detPolPoints.append(points)
+            if len(gtDontCareRectsNum) > 0:
+                for dontCareRectNum in gtDontCareRectsNum:
+                    dontCareRect = gtRects[dontCareRectNum]
+                    intersected_area = get_intersection(dontCareRect, detRect)
+                    rdDimensions = Polygon(detRect).area
+                    if rdDimensions == 0:
+                        precision = 0
+                    else:
+                        precision = intersected_area / rdDimensions
+                    if precision > 0.5:
+                        detDontCareRectsNum.append(len(detRects) - 1)
+                        break
+
+        evaluationLog += (
+            "DET rectangles: "
+            + str(len(detRects))
+            + (
+                " (" + str(len(detDontCareRectsNum)) + " don't care)\n"
+                if len(detDontCareRectsNum) > 0
+                else "\n"
+            )
+        )
+
+        if len(gtRects) == 0:
+            recall = 1
+            precision = 0 if len(detRects) > 0 else 1
+
+        if len(detRects) > 0:
+            # Calculate recall and precision matrixs
+            outputShape = [len(gtRects), len(detRects)]
+            recallMat = np.empty(outputShape)
+            precisionMat = np.empty(outputShape)
+            gtRectMat = np.zeros(len(gtRects), np.int8)
+            detRectMat = np.zeros(len(detRects), np.int8)
+            for gtNum in range(len(gtRects)):
+                for detNum in range(len(detRects)):
+                    rG = gtRects[gtNum]
+                    rD = detRects[detNum]
+                    intersected_area = get_intersection(rG, rD)
+                    rgDimensions = Polygon(rG).area
+                    rdDimensions = Polygon(rD).area
+                    recallMat[gtNum, detNum] = (
+                        0 if rgDimensions == 0 else intersected_area / rgDimensions
+                    )
+                    precisionMat[gtNum, detNum] = (
+                        0 if rdDimensions == 0 else intersected_area / rdDimensions
+                    )
+
+            # Find one-to-one matches
+            evaluationLog += "Find one-to-one matches\n"
+            for gtNum in range(len(gtRects)):
+                for detNum in range(len(detRects)):
+                    if (
+                        gtRectMat[gtNum] == 0
+                        and detRectMat[detNum] == 0
+                        and gtNum not in gtDontCareRectsNum
+                        and detNum not in detDontCareRectsNum
+                    ):
+                        match = one_to_one_match(gtNum, detNum)
+                        if match is True:
+                            # in deteval we have to make other validation before mark as one-to-one
+                            rG = gtRects[gtNum]
+                            rD = detRects[detNum]
+                            normDist = center_distance(rG, rD)
+                            normDist /= diag(rG) + diag(rD)
+                            normDist *= 2.0
+                            if normDist < self.ev_param_ind_center_diff_thr:
+                                gtRectMat[gtNum] = 1
+                                detRectMat[detNum] = 1
+                                recallAccum += 1.0
+                                precisionAccum += 1.0
+                                pairs.append({"gt": gtNum, "det": detNum, "type": "OO"})
+                                evaluationLog += (
+                                    "Match GT #"
+                                    + str(gtNum)
+                                    + " with Det #"
+                                    + str(detNum)
+                                    + "\n"
+                                )
+                            else:
+                                evaluationLog += (
+                                    "Match Discarded GT #"
+                                    + str(gtNum)
+                                    + " with Det #"
+                                    + str(detNum)
+                                    + " normDist: "
+                                    + str(normDist)
+                                    + " \n"
+                                )
+            # Find one-to-many matches
+            evaluationLog += "Find one-to-many matches\n"
+            for gtNum in range(len(gtRects)):
+                if gtNum not in gtDontCareRectsNum:
+                    match, matchesDet = one_to_many_match(gtNum)
+                    if match is True:
+                        gtRectMat[gtNum] = 1
+                        recallAccum += 1.0
+                        precisionAccum += len(matchesDet) / (
+                            1 + math.log(len(matchesDet))
+                        )
+                        pairs.append(
+                            {
+                                "gt": gtNum,
+                                "det": matchesDet,
+                                "type": "OO" if len(matchesDet) == 1 else "OM",
+                            }
+                        )
+                        for detNum in matchesDet:
+                            detRectMat[detNum] = 1
+                        evaluationLog += (
+                            "Match GT #"
+                            + str(gtNum)
+                            + " with Det #"
+                            + str(matchesDet)
+                            + "\n"
+                        )
+
+            # Find many-to-one matches
+            evaluationLog += "Find many-to-one matches\n"
+            for detNum in range(len(detRects)):
+                if detNum not in detDontCareRectsNum:
+                    match, matchesGt = many_to_one_match(detNum)
+                    if match is True:
+                        detRectMat[detNum] = 1
+                        recallAccum += len(matchesGt) / (1 + math.log(len(matchesGt)))
+                        precisionAccum += 1.0
+                        pairs.append(
+                            {
+                                "gt": matchesGt,
+                                "det": detNum,
+                                "type": "OO" if len(matchesGt) == 1 else "MO",
+                            }
+                        )
+                        for gtNum in matchesGt:
+                            gtRectMat[gtNum] = 1
+                        evaluationLog += (
+                            "Match GT #"
+                            + str(matchesGt)
+                            + " with Det #"
+                            + str(detNum)
+                            + "\n"
+                        )
+
+            numGtCare = len(gtRects) - len(gtDontCareRectsNum)
+            if numGtCare == 0:
+                recall = float(1)
+                precision = float(0) if len(detRects) > 0 else float(1)
+            else:
+                recall = float(recallAccum) / numGtCare
+                precision = (
+                    float(0)
+                    if (len(detRects) - len(detDontCareRectsNum)) == 0
+                    else float(precisionAccum)
+                    / (len(detRects) - len(detDontCareRectsNum))
+                )
+            hmean = (
+                0
+                if (precision + recall) == 0
+                else 2.0 * precision * recall / (precision + recall)
+            )
+
+        numGtCare = len(gtRects) - len(gtDontCareRectsNum)
+        numDetCare = len(detRects) - len(detDontCareRectsNum)
+
+        perSampleMetrics = {
+            "precision": precision,
+            "recall": recall,
+            "hmean": hmean,
+            "pairs": pairs,
+            "recallMat": [] if len(detRects) > 100 else recallMat.tolist(),
+            "precisionMat": [] if len(detRects) > 100 else precisionMat.tolist(),
+            "gtPolPoints": gtPolPoints,
+            "detPolPoints": detPolPoints,
+            "gtCare": numGtCare,
+            "detCare": numDetCare,
+            "gtDontCare": gtDontCareRectsNum,
+            "detDontCare": detDontCareRectsNum,
+            "recallAccum": recallAccum,
+            "precisionAccum": precisionAccum,
+            "evaluationLog": evaluationLog,
+        }
+
+        return perSampleMetrics
+
+    def combine_results(self, results):
+        numGt = 0
+        numDet = 0
+        methodRecallSum = 0
+        methodPrecisionSum = 0
+
+        for result in results:
+            numGt += result["gtCare"]
+            numDet += result["detCare"]
+            methodRecallSum += result["recallAccum"]
+            methodPrecisionSum += result["precisionAccum"]
+
+        methodRecall = 0 if numGt == 0 else methodRecallSum / numGt
+        methodPrecision = 0 if numDet == 0 else methodPrecisionSum / numDet
+        methodHmean = (
+            0
+            if methodRecall + methodPrecision == 0
+            else 2 * methodRecall * methodPrecision / (methodRecall + methodPrecision)
+        )
+
+        methodMetrics = {
+            "precision": methodPrecision,
+            "recall": methodRecall,
+            "hmean": methodHmean,
+        }
+
+        return methodMetrics
+
+
+if __name__ == "__main__":
+    evaluator = DetectionICDAR2013Evaluator()
+    gts = [
+        [
+            {
+                "points": [(0, 0), (1, 0), (1, 1), (0, 1)],
+                "text": 1234,
+                "ignore": False,
+            },
+            {
+                "points": [(2, 2), (3, 2), (3, 3), (2, 3)],
+                "text": 5678,
+                "ignore": True,
+            },
+        ]
+    ]
+    preds = [
+        [
+            {
+                "points": [(0.1, 0.1), (1, 0), (1, 1), (0, 1)],
+                "text": 123,
+                "ignore": False,
+            }
+        ]
+    ]
+    results = []
+    for gt, pred in zip(gts, preds):
+        results.append(evaluator.evaluate_image(gt, pred))
+    metrics = evaluator.combine_results(results)
+    print(metrics)
diff --git a/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/quad_metric.py b/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/quad_metric.py
new file mode 100644
index 0000000..b4b6714
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/utils/ocr_metric/icdar2015/quad_metric.py
@@ -0,0 +1,100 @@
+import numpy as np
+
+from .detection.iou import DetectionIoUEvaluator
+
+
+class AverageMeter(object):
+    """Computes and stores the average and current value"""
+
+    def __init__(self):
+        self.reset()
+
+    def reset(self):
+        self.val = 0
+        self.avg = 0
+        self.sum = 0
+        self.count = 0
+
+    def update(self, val, n=1):
+        self.val = val
+        self.sum += val * n
+        self.count += n
+        self.avg = self.sum / self.count
+        return self
+
+
+class QuadMetric:
+    def __init__(self, is_output_polygon=False):
+        self.is_output_polygon = is_output_polygon
+        self.evaluator = DetectionIoUEvaluator(is_output_polygon=is_output_polygon)
+
+    def measure(self, batch, output, box_thresh=0.6):
+        """
+        batch: (image, polygons, ignore_tags
+        batch: a dict produced by dataloaders.
+            image: tensor of shape (N, C, H, W).
+            polygons: tensor of shape (N, K, 4, 2), the polygons of objective regions.
+            ignore_tags: tensor of shape (N, K), indicates whether a region is ignorable or not.
+            shape: the original shape of images.
+            filename: the original filenames of images.
+        output: (polygons, ...)
+        """
+        results = []
+        gt_polyons_batch = batch["text_polys"]
+        ignore_tags_batch = batch["ignore_tags"]
+        pred_polygons_batch = np.array(output[0])
+        pred_scores_batch = np.array(output[1])
+        for polygons, pred_polygons, pred_scores, ignore_tags in zip(
+            gt_polyons_batch, pred_polygons_batch, pred_scores_batch, ignore_tags_batch
+        ):
+            gt = [
+                dict(points=np.int64(polygons[i]), ignore=ignore_tags[i])
+                for i in range(len(polygons))
+            ]
+            if self.is_output_polygon:
+                pred = [
+                    dict(points=pred_polygons[i]) for i in range(len(pred_polygons))
+                ]
+            else:
+                pred = []
+                # print(pred_polygons.shape)
+                for i in range(pred_polygons.shape[0]):
+                    if pred_scores[i] >= box_thresh:
+                        # print(pred_polygons[i,:,:].tolist())
+                        pred.append(
+                            dict(points=pred_polygons[i, :, :].astype(np.int32))
+                        )
+                # pred = [dict(points=pred_polygons[i,:,:].tolist()) if pred_scores[i] >= box_thresh for i in range(pred_polygons.shape[0])]
+            results.append(self.evaluator.evaluate_image(gt, pred))
+        return results
+
+    def validate_measure(self, batch, output, box_thresh=0.6):
+        return self.measure(batch, output, box_thresh)
+
+    def evaluate_measure(self, batch, output):
+        return (
+            self.measure(batch, output),
+            np.linspace(0, batch["image"].shape[0]).tolist(),
+        )
+
+    def gather_measure(self, raw_metrics):
+        raw_metrics = [
+            image_metrics
+            for batch_metrics in raw_metrics
+            for image_metrics in batch_metrics
+        ]
+
+        result = self.evaluator.combine_results(raw_metrics)
+
+        precision = AverageMeter()
+        recall = AverageMeter()
+        fmeasure = AverageMeter()
+
+        precision.update(result["precision"], n=len(raw_metrics))
+        recall.update(result["recall"], n=len(raw_metrics))
+        fmeasure_score = (
+            2 * precision.val * recall.val / (precision.val + recall.val + 1e-8)
+        )
+        fmeasure.update(fmeasure_score)
+
+        return {"precision": precision, "recall": recall, "fmeasure": fmeasure}
diff --git a/benchmark/PaddleOCR_DBNet/utils/profiler.py b/benchmark/PaddleOCR_DBNet/utils/profiler.py
new file mode 100644
index 0000000..34fcea8
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/utils/profiler.py
@@ -0,0 +1,112 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import sys
+import paddle
+
+# A global variable to record the number of calling times for profiler
+# functions. It is used to specify the tracing range of training steps.
+_profiler_step_id = 0
+
+# A global variable to avoid parsing from string every time.
+_profiler_options = None
+
+
+class ProfilerOptions(object):
+    """
+    Use a string to initialize a ProfilerOptions.
+    The string should be in the format: "key1=value1;key2=value;key3=value3".
+    For example:
+      "profile_path=model.profile"
+      "batch_range=[50, 60]; profile_path=model.profile"
+      "batch_range=[50, 60]; tracer_option=OpDetail; profile_path=model.profile"
+    ProfilerOptions supports following key-value pair:
+      batch_range      - a integer list, e.g. [100, 110].
+      state            - a string, the optional values are 'CPU', 'GPU' or 'All'.
+      sorted_key       - a string, the optional values are 'calls', 'total',
+                         'max', 'min' or 'ave.
+      tracer_option    - a string, the optional values are 'Default', 'OpDetail',
+                         'AllOpDetail'.
+      profile_path     - a string, the path to save the serialized profile data,
+                         which can be used to generate a timeline.
+      exit_on_finished - a boolean.
+    """
+
+    def __init__(self, options_str):
+        assert isinstance(options_str, str)
+
+        self._options = {
+            "batch_range": [10, 20],
+            "state": "All",
+            "sorted_key": "total",
+            "tracer_option": "Default",
+            "profile_path": "/tmp/profile",
+            "exit_on_finished": True,
+        }
+        self._parse_from_string(options_str)
+
+    def _parse_from_string(self, options_str):
+        for kv in options_str.replace(" ", "").split(";"):
+            key, value = kv.split("=")
+            if key == "batch_range":
+                value_list = value.replace("[", "").replace("]", "").split(",")
+                value_list = list(map(int, value_list))
+                if (
+                    len(value_list) >= 2
+                    and value_list[0] >= 0
+                    and value_list[1] > value_list[0]
+                ):
+                    self._options[key] = value_list
+            elif key == "exit_on_finished":
+                self._options[key] = value.lower() in ("yes", "true", "t", "1")
+            elif key in ["state", "sorted_key", "tracer_option", "profile_path"]:
+                self._options[key] = value
+
+    def __getitem__(self, name):
+        if self._options.get(name, None) is None:
+            raise ValueError("ProfilerOptions does not have an option named %s." % name)
+        return self._options[name]
+
+
+def add_profiler_step(options_str=None):
+    """
+    Enable the operator-level timing using PaddlePaddle's profiler.
+    The profiler uses a independent variable to count the profiler steps.
+    One call of this function is treated as a profiler step.
+
+    Args:
+      profiler_options - a string to initialize the ProfilerOptions.
+                         Default is None, and the profiler is disabled.
+    """
+    if options_str is None:
+        return
+
+    global _profiler_step_id
+    global _profiler_options
+
+    if _profiler_options is None:
+        _profiler_options = ProfilerOptions(options_str)
+
+    if _profiler_step_id == _profiler_options["batch_range"][0]:
+        paddle.utils.profiler.start_profiler(
+            _profiler_options["state"], _profiler_options["tracer_option"]
+        )
+    elif _profiler_step_id == _profiler_options["batch_range"][1]:
+        paddle.utils.profiler.stop_profiler(
+            _profiler_options["sorted_key"], _profiler_options["profile_path"]
+        )
+        if _profiler_options["exit_on_finished"]:
+            sys.exit(0)
+
+    _profiler_step_id += 1
diff --git a/benchmark/PaddleOCR_DBNet/utils/schedulers.py b/benchmark/PaddleOCR_DBNet/utils/schedulers.py
new file mode 100644
index 0000000..e038ddc
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/utils/schedulers.py
@@ -0,0 +1,72 @@
+from paddle.optimizer import lr
+import logging
+
+__all__ = ["Polynomial"]
+
+
+class Polynomial(object):
+    """
+    Polynomial learning rate decay
+    Args:
+        learning_rate (float): The initial learning rate. It is a python float number.
+        epochs(int): The decay epoch size. It determines the decay cycle, when by_epoch is set to true, it will change to epochs=epochs*step_each_epoch.
+        step_each_epoch: all steps in each epoch.
+        end_lr(float, optional): The minimum final learning rate. Default: 0.0001.
+        power(float, optional): Power of polynomial. Default: 1.0.
+        warmup_epoch(int): The epoch numbers for LinearWarmup. Default: 0, , when by_epoch is set to true, it will change to warmup_epoch=warmup_epoch*step_each_epoch.
+        warmup_start_lr(float): Initial learning rate of warm up. Default: 0.0.
+        last_epoch (int, optional):  The index of last epoch. Can be set to restart training. Default: -1, means initial learning rate.
+        by_epoch: Whether the set parameter is based on epoch or iter, when set to true,, epochs and warmup_epoch will be automatically multiplied by step_each_epoch. Default: True
+    """
+
+    def __init__(
+        self,
+        learning_rate,
+        epochs,
+        step_each_epoch,
+        end_lr=0.0,
+        power=1.0,
+        warmup_epoch=0,
+        warmup_start_lr=0.0,
+        last_epoch=-1,
+        by_epoch=True,
+        **kwargs,
+    ):
+        super().__init__()
+        if warmup_epoch >= epochs:
+            msg = f'When using warm up, the value of "epochs" must be greater than value of "Optimizer.lr.warmup_epoch". The value of "Optimizer.lr.warmup_epoch" has been set to {epochs}.'
+            logging.warning(msg)
+            warmup_epoch = epochs
+        self.learning_rate = learning_rate
+        self.epochs = epochs
+        self.end_lr = end_lr
+        self.power = power
+        self.last_epoch = last_epoch
+        self.warmup_epoch = warmup_epoch
+        self.warmup_start_lr = warmup_start_lr
+
+        if by_epoch:
+            self.epochs *= step_each_epoch
+            self.warmup_epoch = int(self.warmup_epoch * step_each_epoch)
+
+    def __call__(self):
+        learning_rate = (
+            lr.PolynomialDecay(
+                learning_rate=self.learning_rate,
+                decay_steps=self.epochs,
+                end_lr=self.end_lr,
+                power=self.power,
+                last_epoch=self.last_epoch,
+            )
+            if self.epochs > 0
+            else self.learning_rate
+        )
+        if self.warmup_epoch > 0:
+            learning_rate = lr.LinearWarmup(
+                learning_rate=learning_rate,
+                warmup_steps=self.warmup_epoch,
+                start_lr=self.warmup_start_lr,
+                end_lr=self.learning_rate,
+                last_epoch=self.last_epoch,
+            )
+        return learning_rate
diff --git a/benchmark/PaddleOCR_DBNet/utils/util.py b/benchmark/PaddleOCR_DBNet/utils/util.py
new file mode 100644
index 0000000..e46caa1
--- /dev/null
+++ b/benchmark/PaddleOCR_DBNet/utils/util.py
@@ -0,0 +1,365 @@
+# -*- coding: utf-8 -*-
+# @Time    : 2019/8/23 21:59
+# @Author  : zhoujun
+import json
+import pathlib
+import time
+import os
+import glob
+import cv2
+import yaml
+from collections.abc import Mapping
+import matplotlib.pyplot as plt
+import numpy as np
+
+from argparse import ArgumentParser, RawDescriptionHelpFormatter
+
+
+def _check_image_file(path):
+    img_end = {"jpg", "bmp", "png", "jpeg", "rgb", "tif", "tiff", "gif", "pdf"}
+    return any([path.lower().endswith(e) for e in img_end])
+
+
+def get_image_file_list(img_file):
+    imgs_lists = []
+    if img_file is None or not os.path.exists(img_file):
+        raise Exception("not found any img file in {}".format(img_file))
+
+    img_end = {"jpg", "bmp", "png", "jpeg", "rgb", "tif", "tiff", "gif", "pdf"}
+    if os.path.isfile(img_file) and _check_image_file(img_file):
+        imgs_lists.append(img_file)
+    elif os.path.isdir(img_file):
+        for single_file in os.listdir(img_file):
+            file_path = os.path.join(img_file, single_file)
+            if os.path.isfile(file_path) and _check_image_file(file_path):
+                imgs_lists.append(file_path)
+    if len(imgs_lists) == 0:
+        raise Exception("not found any img file in {}".format(img_file))
+    imgs_lists = sorted(imgs_lists)
+    return imgs_lists
+
+
+def setup_logger(log_file_path: str = None):
+    import logging
+
+    logging._warn_preinit_stderr = 0
+    logger = logging.getLogger("DBNet.paddle")
+    formatter = logging.Formatter("%(asctime)s %(name)s %(levelname)s: %(message)s")
+    ch = logging.StreamHandler()
+    ch.setFormatter(formatter)
+    logger.addHandler(ch)
+    if log_file_path is not None:
+        file_handle = logging.FileHandler(log_file_path)
+        file_handle.setFormatter(formatter)
+        logger.addHandler(file_handle)
+    logger.setLevel(logging.DEBUG)
+    return logger
+
+
+# --exeTime
+def exe_time(func):
+    def newFunc(*args, **args2):
+        t0 = time.time()
+        back = func(*args, **args2)
+        print("{} cost {:.3f}s".format(func.__name__, time.time() - t0))
+        return back
+
+    return newFunc
+
+
+def load(file_path: str):
+    file_path = pathlib.Path(file_path)
+    func_dict = {".txt": _load_txt, ".json": _load_json, ".list": _load_txt}
+    assert file_path.suffix in func_dict
+    return func_dict[file_path.suffix](file_path)
+
+
+def _load_txt(file_path: str):
+    with open(file_path, "r", encoding="utf8") as f:
+        content = [
+            x.strip().strip("\ufeff").strip("\xef\xbb\xbf") for x in f.readlines()
+        ]
+    return content
+
+
+def _load_json(file_path: str):
+    with open(file_path, "r", encoding="utf8") as f:
+        content = json.load(f)
+    return content
+
+
+def save(data, file_path):
+    file_path = pathlib.Path(file_path)
+    func_dict = {".txt": _save_txt, ".json": _save_json}
+    assert file_path.suffix in func_dict
+    return func_dict[file_path.suffix](data, file_path)
+
+
+def _save_txt(data, file_path):
+    """
+    将一个list的数组写入txt文件里
+    :param data:
+    :param file_path:
+    :return:
+    """
+    if not isinstance(data, list):
+        data = [data]
+    with open(file_path, mode="w", encoding="utf8") as f:
+        f.write("\n".join(data))
+
+
+def _save_json(data, file_path):
+    with open(file_path, "w", encoding="utf-8") as json_file:
+        json.dump(data, json_file, ensure_ascii=False, indent=4)
+
+
+def show_img(imgs: np.ndarray, title="img"):
+    color = len(imgs.shape) == 3 and imgs.shape[-1] == 3
+    imgs = np.expand_dims(imgs, axis=0)
+    for i, img in enumerate(imgs):
+        plt.figure()
+        plt.title("{}_{}".format(title, i))
+        plt.imshow(img, cmap=None if color else "gray")
+    plt.show()
+
+
+def draw_bbox(img_path, result, color=(255, 0, 0), thickness=2):
+    if isinstance(img_path, str):
+        img_path = cv2.imread(img_path)
+        # img_path = cv2.cvtColor(img_path, cv2.COLOR_BGR2RGB)
+    img_path = img_path.copy()
+    for point in result:
+        point = point.astype(int)
+        cv2.polylines(img_path, [point], True, color, thickness)
+    return img_path
+
+
+def cal_text_score(texts, gt_texts, training_masks, running_metric_text, thred=0.5):
+    training_masks = training_masks.numpy()
+    pred_text = texts.numpy() * training_masks
+    pred_text[pred_text <= thred] = 0
+    pred_text[pred_text > thred] = 1
+    pred_text = pred_text.astype(np.int32)
+    gt_text = gt_texts.numpy() * training_masks
+    gt_text = gt_text.astype(np.int32)
+    running_metric_text.update(gt_text, pred_text)
+    score_text, _ = running_metric_text.get_scores()
+    return score_text
+
+
+def order_points_clockwise(pts):
+    rect = np.zeros((4, 2), dtype="float32")
+    s = pts.sum(axis=1)
+    rect[0] = pts[np.argmin(s)]
+    rect[2] = pts[np.argmax(s)]
+    diff = np.diff(pts, axis=1)
+    rect[1] = pts[np.argmin(diff)]
+    rect[3] = pts[np.argmax(diff)]
+    return rect
+
+
+def order_points_clockwise_list(pts):
+    pts = pts.tolist()
+    pts.sort(key=lambda x: (x[1], x[0]))
+    pts[:2] = sorted(pts[:2], key=lambda x: x[0])
+    pts[2:] = sorted(pts[2:], key=lambda x: -x[0])
+    pts = np.array(pts)
+    return pts
+
+
+def get_datalist(train_data_path):
+    """
+    获取训练和验证的数据list
+    :param train_data_path: 训练的dataset文件列表，每个文件内以如下格式存储 ‘path/to/img\tlabel’
+    :return:
+    """
+    train_data = []
+    for p in train_data_path:
+        with open(p, "r", encoding="utf-8") as f:
+            for line in f.readlines():
+                line = line.strip("\n").replace(".jpg ", ".jpg\t").split("\t")
+                if len(line) > 1:
+                    img_path = pathlib.Path(line[0].strip(" "))
+                    label_path = pathlib.Path(line[1].strip(" "))
+                    if (
+                        img_path.exists()
+                        and img_path.stat().st_size > 0
+                        and label_path.exists()
+                        and label_path.stat().st_size > 0
+                    ):
+                        train_data.append((str(img_path), str(label_path)))
+    return train_data
+
+
+def save_result(result_path, box_list, score_list, is_output_polygon):
+    if is_output_polygon:
+        with open(result_path, "wt") as res:
+            for i, box in enumerate(box_list):
+                box = box.reshape(-1).tolist()
+                result = ",".join([str(int(x)) for x in box])
+                score = score_list[i]
+                res.write(result + "," + str(score) + "\n")
+    else:
+        with open(result_path, "wt") as res:
+            for i, box in enumerate(box_list):
+                score = score_list[i]
+                box = box.reshape(-1).tolist()
+                result = ",".join([str(int(x)) for x in box])
+                res.write(result + "," + str(score) + "\n")
+
+
+def expand_polygon(polygon):
+    """
+    对只有一个字符的框进行扩充
+    """
+    (x, y), (w, h), angle = cv2.minAreaRect(np.float32(polygon))
+    if angle < -45:
+        w, h = h, w
+        angle += 90
+    new_w = w + h
+    box = ((x, y), (new_w, h), angle)
+    points = cv2.boxPoints(box)
+    return order_points_clockwise(points)
+
+
+def _merge_dict(config, merge_dct):
+    """Recursive dict merge. Inspired by :meth:``dict.update()``, instead of
+    updating only top-level keys, dict_merge recurses down into dicts nested
+    to an arbitrary depth, updating keys. The ``merge_dct`` is merged into
+    ``dct``.
+    Args:
+        config: dict onto which the merge is executed
+        merge_dct: dct merged into config
+    Returns: dct
+    """
+    for key, value in merge_dct.items():
+        sub_keys = key.split(".")
+        key = sub_keys[0]
+        if key in config and len(sub_keys) > 1:
+            _merge_dict(config[key], {".".join(sub_keys[1:]): value})
+        elif (
+            key in config
+            and isinstance(config[key], dict)
+            and isinstance(value, Mapping)
+        ):
+            _merge_dict(config[key], value)
+        else:
+            config[key] = value
+    return config
+
+
+def print_dict(cfg, print_func=print, delimiter=0):
+    """
+    Recursively visualize a dict and
+    indenting acrrording by the relationship of keys.
+    """
+    for k, v in sorted(cfg.items()):
+        if isinstance(v, dict):
+            print_func("{}{} : ".format(delimiter * " ", str(k)))
+            print_dict(v, print_func, delimiter + 4)
+        elif isinstance(v, list) and len(v) >= 1 and isinstance(v[0], dict):
+            print_func("{}{} : ".format(delimiter * " ", str(k)))
+            for value in v:
+                print_dict(value, print_func, delimiter + 4)
+        else:
+            print_func("{}{} : {}".format(delimiter * " ", k, v))
+
+
+class Config(object):
+    def __init__(self, config_path, BASE_KEY="base"):
+        self.BASE_KEY = BASE_KEY
+        self.cfg = self._load_config_with_base(config_path)
+
+    def _load_config_with_base(self, file_path):
+        """
+        Load config from file.
+        Args:
+            file_path (str): Path of the config file to be loaded.
+        Returns: global config
+        """
+        _, ext = os.path.splitext(file_path)
+        assert ext in [".yml", ".yaml"], "only support yaml files for now"
+
+        with open(file_path) as f:
+            file_cfg = yaml.load(f, Loader=yaml.SafeLoader)
+
+        # NOTE: cfgs outside have higher priority than cfgs in _BASE_
+        if self.BASE_KEY in file_cfg:
+            all_base_cfg = dict()
+            base_ymls = list(file_cfg[self.BASE_KEY])
+            for base_yml in base_ymls:
+                with open(base_yml) as f:
+                    base_cfg = self._load_config_with_base(base_yml)
+                    all_base_cfg = _merge_dict(all_base_cfg, base_cfg)
+
+            del file_cfg[self.BASE_KEY]
+            file_cfg = _merge_dict(all_base_cfg, file_cfg)
+        file_cfg["filename"] = os.path.splitext(os.path.split(file_path)[-1])[0]
+        return file_cfg
+
+    def merge_dict(self, args):
+        self.cfg = _merge_dict(self.cfg, args)
+
+    def print_cfg(self, print_func=print):
+        """
+        Recursively visualize a dict and
+        indenting according by the relationship of keys.
+        """
+        print_func("----------- Config -----------")
+        print_dict(self.cfg, print_func)
+        print_func("---------------------------------------------")
+
+    def save(self, p):
+        with open(p, "w") as f:
+            yaml.dump(dict(self.cfg), f, default_flow_style=False, sort_keys=False)
+
+
+class ArgsParser(ArgumentParser):
+    def __init__(self):
+        super(ArgsParser, self).__init__(formatter_class=RawDescriptionHelpFormatter)
+        self.add_argument("-c", "--config_file", help="configuration file to use")
+        self.add_argument("-o", "--opt", nargs="*", help="set configuration options")
+        self.add_argument(
+            "-p",
+            "--profiler_options",
+            type=str,
+            default=None,
+            help="The option of profiler, which should be in format "
+            '"key1=value1;key2=value2;key3=value3".',
+        )
+
+    def parse_args(self, argv=None):
+        args = super(ArgsParser, self).parse_args(argv)
+        assert (
+            args.config_file is not None
+        ), "Please specify --config_file=configure_file_path."
+        args.opt = self._parse_opt(args.opt)
+        return args
+
+    def _parse_opt(self, opts):
+        config = {}
+        if not opts:
+            return config
+        for s in opts:
+            s = s.strip()
+            k, v = s.split("=", 1)
+            if "." not in k:
+                config[k] = yaml.load(v, Loader=yaml.SafeLoader)
+            else:
+                keys = k.split(".")
+                if keys[0] not in config:
+                    config[keys[0]] = {}
+                cur = config[keys[0]]
+                for idx, key in enumerate(keys[1:]):
+                    if idx == len(keys) - 2:
+                        cur[key] = yaml.load(v, Loader=yaml.SafeLoader)
+                    else:
+                        cur[key] = {}
+                        cur = cur[key]
+        return config
+
+
+if __name__ == "__main__":
+    img = np.zeros((1, 3, 640, 640))
+    show_img(img[0][0])
+    plt.show()
diff --git a/benchmark/analysis.py b/benchmark/analysis.py
new file mode 100644
index 0000000..bf6233f
--- /dev/null
+++ b/benchmark/analysis.py
@@ -0,0 +1,354 @@
+# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import print_function
+
+import argparse
+import json
+import os
+import re
+import traceback
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument(
+        "--filename", type=str, help="The name of log which need to analysis."
+    )
+    parser.add_argument(
+        "--log_with_profiler", type=str, help="The path of train log with profiler"
+    )
+    parser.add_argument(
+        "--profiler_path", type=str, help="The path of profiler timeline log."
+    )
+    parser.add_argument("--keyword", type=str, help="Keyword to specify analysis data")
+    parser.add_argument(
+        "--separator",
+        type=str,
+        default=None,
+        help="Separator of different field in log",
+    )
+    parser.add_argument(
+        "--position", type=int, default=None, help="The position of data field"
+    )
+    parser.add_argument(
+        "--range", type=str, default="", help="The range of data field to intercept"
+    )
+    parser.add_argument("--base_batch_size", type=int, help="base_batch size on gpu")
+    parser.add_argument(
+        "--skip_steps", type=int, default=0, help="The number of steps to be skipped"
+    )
+    parser.add_argument(
+        "--model_mode", type=int, default=-1, help="Analysis mode, default value is -1"
+    )
+    parser.add_argument("--ips_unit", type=str, default=None, help="IPS unit")
+    parser.add_argument(
+        "--model_name",
+        type=str,
+        default=0,
+        help="training model_name, transformer_base",
+    )
+    parser.add_argument(
+        "--mission_name", type=str, default=0, help="training mission name"
+    )
+    parser.add_argument(
+        "--direction_id", type=int, default=0, help="training direction_id"
+    )
+    parser.add_argument(
+        "--run_mode", type=str, default="sp", help="multi process or single process"
+    )
+    parser.add_argument(
+        "--index",
+        type=int,
+        default=1,
+        help="{1: speed, 2:mem, 3:profiler, 6:max_batch_size}",
+    )
+    parser.add_argument("--gpu_num", type=int, default=1, help="nums of training gpus")
+    args = parser.parse_args()
+    args.separator = None if args.separator == "None" else args.separator
+    return args
+
+
+def _is_number(num):
+    pattern = re.compile(r"^[-+]?[-0-9]\d*\.\d*|[-+]?\.?[0-9]\d*$")
+    result = pattern.match(num)
+    if result:
+        return True
+    else:
+        return False
+
+
+class TimeAnalyzer(object):
+    def __init__(
+        self, filename, keyword=None, separator=None, position=None, range="-1"
+    ):
+        if filename is None:
+            raise Exception("Please specify the filename!")
+
+        if keyword is None:
+            raise Exception("Please specify the keyword!")
+
+        self.filename = filename
+        self.keyword = keyword
+        self.separator = separator
+        self.position = position
+        self.range = range
+        self.records = None
+        self._distil()
+
+    def _distil(self):
+        self.records = []
+        with open(self.filename, "r") as f_object:
+            lines = f_object.readlines()
+            for line in lines:
+                if self.keyword not in line:
+                    continue
+                try:
+                    result = None
+
+                    # Distil the string from a line.
+                    line = line.strip()
+                    line_words = (
+                        line.split(self.separator) if self.separator else line.split()
+                    )
+                    if args.position:
+                        result = line_words[self.position]
+                    else:
+                        # Distil the string following the keyword.
+                        for i in range(len(line_words) - 1):
+                            if line_words[i] == self.keyword:
+                                result = line_words[i + 1]
+                                break
+
+                    # Distil the result from the picked string.
+                    if not self.range:
+                        result = result[0:]
+                    elif _is_number(self.range):
+                        result = result[0 : int(self.range)]
+                    else:
+                        result = result[
+                            int(self.range.split(":")[0]) : int(
+                                self.range.split(":")[1]
+                            )
+                        ]
+                    self.records.append(float(result))
+                except Exception as exc:
+                    print(
+                        "line is: {}; separator={}; position={}".format(
+                            line, self.separator, self.position
+                        )
+                    )
+
+        print(
+            "Extract {} records: separator={}; position={}".format(
+                len(self.records), self.separator, self.position
+            )
+        )
+
+    def _get_fps(self, mode, batch_size, gpu_num, avg_of_records, run_mode, unit=None):
+        if mode == -1 and run_mode == "sp":
+            assert unit, "Please set the unit when mode is -1."
+            fps = gpu_num * avg_of_records
+        elif mode == -1 and run_mode == "mp":
+            assert unit, "Please set the unit when mode is -1."
+            fps = gpu_num * avg_of_records  # temporarily, not used now
+            print("------------this is mp")
+        elif mode == 0:
+            # s/step -> samples/s
+            fps = (batch_size * gpu_num) / avg_of_records
+            unit = "samples/s"
+        elif mode == 1:
+            # steps/s -> steps/s
+            fps = avg_of_records
+            unit = "steps/s"
+        elif mode == 2:
+            # s/step -> steps/s
+            fps = 1 / avg_of_records
+            unit = "steps/s"
+        elif mode == 3:
+            # steps/s -> samples/s
+            fps = batch_size * gpu_num * avg_of_records
+            unit = "samples/s"
+        elif mode == 4:
+            # s/epoch -> s/epoch
+            fps = avg_of_records
+            unit = "s/epoch"
+        else:
+            ValueError("Unsupported analysis mode.")
+
+        return fps, unit
+
+    def analysis(
+        self, batch_size, gpu_num=1, skip_steps=0, mode=-1, run_mode="sp", unit=None
+    ):
+        if batch_size <= 0:
+            print("base_batch_size should larger than 0.")
+            return 0, ""
+
+        if (
+            len(self.records) <= skip_steps
+        ):  # to address the condition which item of log equals to skip_steps
+            print("no records")
+            return 0, ""
+
+        sum_of_records = 0
+        sum_of_records_skipped = 0
+        skip_min = self.records[skip_steps]
+        skip_max = self.records[skip_steps]
+
+        count = len(self.records)
+        for i in range(count):
+            sum_of_records += self.records[i]
+            if i >= skip_steps:
+                sum_of_records_skipped += self.records[i]
+                if self.records[i] < skip_min:
+                    skip_min = self.records[i]
+                if self.records[i] > skip_max:
+                    skip_max = self.records[i]
+
+        avg_of_records = sum_of_records / float(count)
+        avg_of_records_skipped = sum_of_records_skipped / float(count - skip_steps)
+
+        fps, fps_unit = self._get_fps(
+            mode, batch_size, gpu_num, avg_of_records, run_mode, unit
+        )
+        fps_skipped, _ = self._get_fps(
+            mode, batch_size, gpu_num, avg_of_records_skipped, run_mode, unit
+        )
+        if mode == -1:
+            print("average ips of %d steps, skip 0 step:" % count)
+            print("\tAvg: %.3f %s" % (avg_of_records, fps_unit))
+            print("\tFPS: %.3f %s" % (fps, fps_unit))
+            if skip_steps > 0:
+                print("average ips of %d steps, skip %d steps:" % (count, skip_steps))
+                print("\tAvg: %.3f %s" % (avg_of_records_skipped, fps_unit))
+                print("\tMin: %.3f %s" % (skip_min, fps_unit))
+                print("\tMax: %.3f %s" % (skip_max, fps_unit))
+                print("\tFPS: %.3f %s" % (fps_skipped, fps_unit))
+        elif mode == 1 or mode == 3:
+            print("average latency of %d steps, skip 0 step:" % count)
+            print("\tAvg: %.3f steps/s" % avg_of_records)
+            print("\tFPS: %.3f %s" % (fps, fps_unit))
+            if skip_steps > 0:
+                print(
+                    "average latency of %d steps, skip %d steps:" % (count, skip_steps)
+                )
+                print("\tAvg: %.3f steps/s" % avg_of_records_skipped)
+                print("\tMin: %.3f steps/s" % skip_min)
+                print("\tMax: %.3f steps/s" % skip_max)
+                print("\tFPS: %.3f %s" % (fps_skipped, fps_unit))
+        elif mode == 0 or mode == 2:
+            print("average latency of %d steps, skip 0 step:" % count)
+            print("\tAvg: %.3f s/step" % avg_of_records)
+            print("\tFPS: %.3f %s" % (fps, fps_unit))
+            if skip_steps > 0:
+                print(
+                    "average latency of %d steps, skip %d steps:" % (count, skip_steps)
+                )
+                print("\tAvg: %.3f s/step" % avg_of_records_skipped)
+                print("\tMin: %.3f s/step" % skip_min)
+                print("\tMax: %.3f s/step" % skip_max)
+                print("\tFPS: %.3f %s" % (fps_skipped, fps_unit))
+
+        return round(fps_skipped, 3), fps_unit
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    run_info = dict()
+    run_info["log_file"] = args.filename
+    run_info["model_name"] = args.model_name
+    run_info["mission_name"] = args.mission_name
+    run_info["direction_id"] = args.direction_id
+    run_info["run_mode"] = args.run_mode
+    run_info["index"] = args.index
+    run_info["gpu_num"] = args.gpu_num
+    run_info["FINAL_RESULT"] = 0
+    run_info["JOB_FAIL_FLAG"] = 0
+
+    try:
+        if args.index == 1:
+            if args.gpu_num == 1:
+                run_info["log_with_profiler"] = args.log_with_profiler
+                run_info["profiler_path"] = args.profiler_path
+            analyzer = TimeAnalyzer(
+                args.filename, args.keyword, args.separator, args.position, args.range
+            )
+            run_info["FINAL_RESULT"], run_info["UNIT"] = analyzer.analysis(
+                batch_size=args.base_batch_size,
+                gpu_num=args.gpu_num,
+                skip_steps=args.skip_steps,
+                mode=args.model_mode,
+                run_mode=args.run_mode,
+                unit=args.ips_unit,
+            )
+            try:
+                if (
+                    int(os.getenv("job_fail_flag")) == 1
+                    or int(run_info["FINAL_RESULT"]) == 0
+                ):
+                    run_info["JOB_FAIL_FLAG"] = 1
+            except:
+                pass
+        elif args.index == 3:
+            run_info["FINAL_RESULT"] = {}
+            records_fo_total = TimeAnalyzer(
+                args.filename, "Framework overhead", None, 3, ""
+            ).records
+            records_fo_ratio = TimeAnalyzer(
+                args.filename, "Framework overhead", None, 5
+            ).records
+            records_ct_total = TimeAnalyzer(
+                args.filename, "Computation time", None, 3, ""
+            ).records
+            records_gm_total = TimeAnalyzer(
+                args.filename, "GpuMemcpy                Calls", None, 4, ""
+            ).records
+            records_gm_ratio = TimeAnalyzer(
+                args.filename, "GpuMemcpy                Calls", None, 6
+            ).records
+            records_gmas_total = TimeAnalyzer(
+                args.filename, "GpuMemcpyAsync         Calls", None, 4, ""
+            ).records
+            records_gms_total = TimeAnalyzer(
+                args.filename, "GpuMemcpySync          Calls", None, 4, ""
+            ).records
+            run_info["FINAL_RESULT"]["Framework_Total"] = (
+                records_fo_total[0] if records_fo_total else 0
+            )
+            run_info["FINAL_RESULT"]["Framework_Ratio"] = (
+                records_fo_ratio[0] if records_fo_ratio else 0
+            )
+            run_info["FINAL_RESULT"]["ComputationTime_Total"] = (
+                records_ct_total[0] if records_ct_total else 0
+            )
+            run_info["FINAL_RESULT"]["GpuMemcpy_Total"] = (
+                records_gm_total[0] if records_gm_total else 0
+            )
+            run_info["FINAL_RESULT"]["GpuMemcpy_Ratio"] = (
+                records_gm_ratio[0] if records_gm_ratio else 0
+            )
+            run_info["FINAL_RESULT"]["GpuMemcpyAsync_Total"] = (
+                records_gmas_total[0] if records_gmas_total else 0
+            )
+            run_info["FINAL_RESULT"]["GpuMemcpySync_Total"] = (
+                records_gms_total[0] if records_gms_total else 0
+            )
+        else:
+            print("Not support!")
+    except Exception:
+        traceback.print_exc()
+    print(
+        "{}".format(json.dumps(run_info))
+    )  # it's required, for the log file path  insert to the database
diff --git a/benchmark/readme.md b/benchmark/readme.md
new file mode 100644
index 0000000..d90d214
--- /dev/null
+++ b/benchmark/readme.md
@@ -0,0 +1,30 @@
+
+# PaddleOCR DB/EAST/PSE 算法训练benchmark测试
+
+PaddleOCR/benchmark目录下的文件用于获取并分析训练日志。
+训练采用icdar2015数据集，包括1000张训练图像和500张测试图像。模型配置采用resnet18_vd作为backbone，分别训练batch_size=8和batch_size=16的情况。
+
+## 运行训练benchmark
+
+benchmark/run_det.sh 中包含了三个过程：
+- 安装依赖
+- 下载数据
+- 执行训练
+- 日志分析获取IPS
+
+在执行训练部分，会执行单机单卡（默认0号卡）单机多卡训练，并分别执行batch_size=8和batch_size=16的情况。所以执行完后，每种模型会得到4个日志文件。
+
+run_det.sh 执行方式如下:
+
+```
+# cd PaddleOCR/
+bash benchmark/run_det.sh
+```
+
+以DB为例，将得到四个日志文件，如下：
+```
+det_res18_db_v2.0_sp_bs16_fp32_1
+det_res18_db_v2.0_sp_bs8_fp32_1
+det_res18_db_v2.0_mp_bs16_fp32_1
+det_res18_db_v2.0_mp_bs8_fp32_1
+```
diff --git a/benchmark/run_benchmark_det.sh b/benchmark/run_benchmark_det.sh
new file mode 100644
index 0000000..125d874
--- /dev/null
+++ b/benchmark/run_benchmark_det.sh
@@ -0,0 +1,61 @@
+#!/usr/bin/env bash
+# 运行示例：CUDA_VISIBLE_DEVICES=0 bash run_benchmark.sh ${run_mode} ${bs_item} ${fp_item} 500 ${model_mode}
+# 参数说明
+function _set_params(){
+    run_mode=${1:-"sp"}          # 单卡sp|多卡mp
+    batch_size=${2:-"64"}
+    fp_item=${3:-"fp32"}         # fp32|fp16
+    max_epoch=${4:-"10"}         # 可选，如果需要修改代码提前中断
+    model_item=${5:-"model_item"}
+    run_log_path=${TRAIN_LOG_DIR:-$(pwd)}  # TRAIN_LOG_DIR 后续QA设置该参数
+#   日志解析所需参数
+    base_batch_size=${batch_size}
+    mission_name="OCR"
+    direction_id="0"
+    ips_unit="images/sec"
+    skip_steps=2                 # 解析日志，有些模型前几个step耗时长，需要跳过                                    (必填)
+    keyword="ips:"               # 解析日志，筛选出数据所在行的关键字                                             (必填)
+    index="1"
+    model_name=${model_item}_bs${batch_size}_${fp_item}        # model_item 用于yml文件名匹配，model_name 用于数据入库前端展示
+#   以下不用修改   
+    device=${CUDA_VISIBLE_DEVICES//,/ }
+    arr=(${device})
+    num_gpu_devices=${#arr[*]}
+    log_file=${run_log_path}/${model_item}_${run_mode}_bs${batch_size}_${fp_item}_${num_gpu_devices}
+}
+function _train(){
+    echo "Train on ${num_gpu_devices} GPUs"
+    echo "current CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES, gpus=$num_gpu_devices, batch_size=$batch_size"
+
+    train_cmd="-c configs/det/${model_item}.yml -o Train.loader.batch_size_per_card=${batch_size} Global.epoch_num=${max_epoch} Global.eval_batch_step=[0,20000] Global.print_batch_step=2"
+    case ${run_mode} in
+      sp) 
+        train_cmd="python tools/train.py "${train_cmd}""
+        ;;
+      mp)
+        rm -rf ./mylog
+        train_cmd="python -m paddle.distributed.launch --log_dir=./mylog --gpus=$CUDA_VISIBLE_DEVICES tools/train.py ${train_cmd}"
+        ;;
+      *) echo "choose run_mode(sp or mp)"; exit 1;
+    esac
+# 以下不用修改
+    echo ${train_cmd}
+    timeout 15m ${train_cmd} > ${log_file} 2>&1
+    if [ $? -ne 0 ];then
+            echo -e "${model_name}, FAIL"
+        export job_fail_flag=1
+    else
+        echo -e "${model_name}, SUCCESS"
+        export job_fail_flag=0
+    fi
+
+    if [ $run_mode = "mp" -a -d mylog ]; then
+        rm ${log_file}
+        cp mylog/workerlog.0 ${log_file}
+    fi
+}
+
+source ${BENCHMARK_ROOT}/scripts/run_model.sh      # 在该脚本中会对符合benchmark规范的log使用analysis.py 脚本进行性能数据解析;该脚本在连调时可从benchmark repo中下载https://github.com/PaddlePaddle/benchmark/blob/master/scripts/run_model.sh;如果不联调只想要产出训练log可以注掉本行,提交时需打开
+_set_params $@
+#_train      # 如果只想产出训练log,不解析,可取消注释
+_run         # 该函数在run_model.sh中,执行时会调用_train; 如果不联调只想要产出训练log可以注掉本行,提交时需打开
diff --git a/benchmark/run_det.sh b/benchmark/run_det.sh
new file mode 100644
index 0000000..75b0f17
--- /dev/null
+++ b/benchmark/run_det.sh
@@ -0,0 +1,36 @@
+#!/bin/bash
+# 提供可稳定复现性能的脚本，默认在标准docker环境内py37执行： paddlepaddle/paddle:latest-gpu-cuda10.1-cudnn7  paddle=2.1.2  py=37
+# 执行目录: ./PaddleOCR
+# 1 安装该模型需要的依赖 (如需开启优化策略请注明)
+log_path=${LOG_PATH_INDEX_DIR:-$(pwd)}
+python -m pip install -r requirements.txt
+# 2 拷贝该模型需要数据、预训练模型
+wget -P ./train_data/  https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/icdar2015.tar && cd train_data  && tar xf icdar2015.tar && cd ../
+wget -P ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet50_vd_pretrained.pdparams
+wget -P ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet18_vd_pretrained.pdparams
+wget -P ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet50_vd_ssld_pretrained.pdparams
+# 3 批量运行（如不方便批量，1，2需放到单个模型中）
+
+model_mode_list=(det_res18_db_v2.0 det_r50_vd_east det_r50_vd_pse)
+fp_item_list=(fp32)
+for model_mode in ${model_mode_list[@]}; do
+      for fp_item in ${fp_item_list[@]}; do
+          if [ ${model_mode} == "det_r50_vd_east" ]; then
+              bs_list=(16)
+          else
+              bs_list=(8 16)
+          fi
+          for bs_item in ${bs_list[@]}; do
+            echo "index is speed, 1gpus, begin, ${model_name}"
+            run_mode=sp
+            log_name=ocr_${model_mode}_bs${bs_item}_${fp_item}_${run_mode}
+            CUDA_VISIBLE_DEVICES=0 bash benchmark/run_benchmark_det.sh ${run_mode} ${bs_item} ${fp_item} 1 ${model_mode} | tee ${log_path}/${log_name}_speed_1gpus 2>&1    #  (5min)
+            sleep 60
+            echo "index is speed, 8gpus, run_mode is multi_process, begin, ${model_name}"
+            run_mode=mp
+            log_name=ocr_${model_mode}_bs${bs_item}_${fp_item}_${run_mode}
+            CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 bash benchmark/run_benchmark_det.sh ${run_mode} ${bs_item} ${fp_item} 2 ${model_mode} | tee ${log_path}/${log_name}_speed_8gpus8p 2>&1
+            sleep 60
+            done
+      done
+done
diff --git a/configs/cls/ch_PP-OCRv3/ch_PP-OCRv3_rotnet.yml b/configs/cls/ch_PP-OCRv3/ch_PP-OCRv3_rotnet.yml
new file mode 100644
index 0000000..f7e327d
--- /dev/null
+++ b/configs/cls/ch_PP-OCRv3/ch_PP-OCRv3_rotnet.yml
@@ -0,0 +1,98 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 100
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_ppocr_v3_rotnet
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: 25
+  infer_mode: false
+  use_space_char: true
+  save_res_path: ./output/rec/predicts_chinese_lite_v2.0.txt
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: L2
+    factor: 1.0e-05
+Architecture:
+  model_type: cls
+  algorithm: CLS
+  Transform: null
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+    last_conv_stride: [1, 2]
+    last_pool_type: avg
+  Neck:
+  Head:
+    name: ClsHead
+    class_dim: 4
+
+Loss:
+  name: ClsLoss
+  main_indicator: acc
+
+PostProcess:
+  name: ClsPostProcess
+
+Metric:
+  name: ClsMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - BaseDataAugmentation:
+    - RandAugment:
+    - SSLRotateResize:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys: ["image", "label"]
+  loader:
+    collate_fn: "SSLRotateCollate"
+    shuffle: true
+    batch_size_per_card: 32
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - SSLRotateResize:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys: ["image", "label"]
+  loader:
+    collate_fn: "SSLRotateCollate"
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 64
+    num_workers: 8
+profiler_options: null
diff --git a/configs/cls/cls_mv3.yml b/configs/cls/cls_mv3.yml
new file mode 100644
index 0000000..0c46ff5
--- /dev/null
+++ b/configs/cls/cls_mv3.yml
@@ -0,0 +1,94 @@
+Global:
+  use_gpu: true
+  epoch_num: 100
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/cls/mv3/
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  label_list: ['0','180']
+
+Architecture:
+  model_type: cls
+  algorithm: CLS
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.35
+    model_name: small
+  Neck:
+  Head:
+    name: ClsHead
+    class_dim: 2
+
+Loss:
+  name: ClsLoss
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: ClsPostProcess
+
+Metric:
+  name: ClsMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/cls
+    label_file_list:
+      - ./train_data/cls/train.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ClsLabelEncode: # Class handling label
+      - BaseDataAugmentation:
+      - RandAugment:
+      - ClsResizeImg:
+          image_shape: [3, 48, 192]
+      - KeepKeys:
+          keep_keys: ['image', 'label'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 512
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/cls
+    label_file_list:
+      - ./train_data/cls/test.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ClsLabelEncode: # Class handling label
+      - ClsResizeImg:
+          image_shape: [3, 48, 192]
+      - KeepKeys:
+          keep_keys: ['image', 'label'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 512
+    num_workers: 4
diff --git a/configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml b/configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml
new file mode 100644
index 0000000..252d159
--- /dev/null
+++ b/configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml
@@ -0,0 +1,226 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/ch_PP-OCR_v3_det/
+  save_epoch_step: 100
+  eval_batch_step:
+  - 0
+  - 400
+  cal_metric_during_train: false
+  pretrained_model: null
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./checkpoints/det_db/predicts_db.txt
+  distributed: true
+  d2s_train_image_shape: [3, -1, -1]
+  amp_dtype: bfloat16
+
+Architecture:
+  name: DistillationModel
+  algorithm: Distillation
+  model_type: det
+  Models:
+    Student:
+      pretrained:
+      model_type: det
+      algorithm: DB
+      Transform: null
+      Backbone:
+        name: MobileNetV3
+        scale: 0.5
+        model_name: large
+        disable_se: true
+      Neck:
+        name: RSEFPN
+        out_channels: 96
+        shortcut: True
+      Head:
+        name: DBHead
+        k: 50
+    Student2:
+      pretrained:
+      model_type: det
+      algorithm: DB
+      Transform: null
+      Backbone:
+        name: MobileNetV3
+        scale: 0.5
+        model_name: large
+        disable_se: true
+      Neck:
+        name: RSEFPN
+        out_channels: 96
+        shortcut: True
+      Head:
+        name: DBHead
+        k: 50
+    Teacher:
+      freeze_params: true
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Backbone:
+        name: ResNet_vd
+        in_channels: 3
+        layers: 50
+      Neck:
+        name: LKPAN
+        out_channels: 256
+      Head:
+        name: DBHead
+        kernel_list: [7,2,2]
+        k: 50
+
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationDilaDBLoss:
+      weight: 1.0
+      model_name_pairs:
+      - ["Student", "Teacher"]
+      - ["Student2", "Teacher"]
+      key: maps
+      balance_loss: true
+      main_loss_type: DiceLoss
+      alpha: 5
+      beta: 10
+      ohem_ratio: 3
+  - DistillationDMLLoss:
+      model_name_pairs:
+      - ["Student", "Student2"]
+      maps_name: "thrink_maps"
+      weight: 1.0
+      model_name_pairs: ["Student", "Student2"]
+      key: maps
+  - DistillationDBLoss:
+      weight: 1.0
+      model_name_list: ["Student", "Student2"]
+      balance_loss: true
+      main_loss_type: DiceLoss
+      alpha: 5
+      beta: 10
+      ohem_ratio: 3
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 5.0e-05
+
+PostProcess:
+  name: DistillationDBPostProcess
+  model_name: ["Student"]
+  key: head_out
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DistillationMetric
+  base_metric_name: DetMetric
+  main_indicator: hmean
+  key: "Student"
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - CopyPaste:
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            rotate:
+            - -10
+            - 10
+        - type: Resize
+          args:
+            size:
+            - 0.5
+            - 3
+    - EastRandomCropData:
+        size:
+        - 960
+        - 960
+        max_tries: 50
+        keep_ratio: true
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/configs/det/PP-OCRv3/PP-OCRv3_det_dml.yml b/configs/det/PP-OCRv3/PP-OCRv3_det_dml.yml
new file mode 100644
index 0000000..0cb7a3c
--- /dev/null
+++ b/configs/det/PP-OCRv3/PP-OCRv3_det_dml.yml
@@ -0,0 +1,174 @@
+Global:
+  use_gpu: true
+  epoch_num: 1200
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/ch_db_mv3/
+  save_epoch_step: 1200
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [3000, 2000]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_db/predicts_db.txt
+  d2s_train_image_shape: [3, -1, -1]
+
+Architecture:
+  name: DistillationModel
+  algorithm: Distillation
+  model_type: det
+  Models:
+    Student:
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Backbone:
+        name: ResNet_vd
+        in_channels: 3
+        layers: 50
+      Neck:
+        name: LKPAN
+        out_channels: 256
+      Head:
+        name: DBHead
+        kernel_list: [7,2,2]
+        k: 50
+    Student2:
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Backbone:
+        name: ResNet_vd
+        in_channels: 3
+        layers: 50
+      Neck:
+        name: LKPAN
+        out_channels: 256
+      Head:
+        name: DBHead
+        kernel_list: [7,2,2]
+        k: 50
+
+
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationDMLLoss:
+      model_name_pairs:
+      - ["Student", "Student2"]
+      maps_name: "thrink_maps"
+      weight: 1.0
+      # act: None
+      model_name_pairs: ["Student", "Student2"]
+      key: maps
+  - DistillationDBLoss:
+      weight: 1.0
+      model_name_list: ["Student", "Student2"]
+      # key: maps
+      name: DBLoss
+      balance_loss: true
+      main_loss_type: DiceLoss
+      alpha: 5
+      beta: 10
+      ohem_ratio: 3
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 2
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: DistillationDBPostProcess
+  model_name: ["Student", "Student2"]
+  key: head_out
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DistillationMetric
+  base_metric_name: DetMetric
+  main_indicator: hmean
+  key: "Student"
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - CopyPaste:
+      - IaaAugment:
+          augmenter_args:
+            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
+            - { 'type': Affine, 'args': { 'rotate': [-10, 10] } }
+            - { 'type': Resize, 'args': { 'size': [0.5, 3] } }
+      - EastRandomCropData:
+          size: [960, 960]
+          max_tries: 50
+          keep_ratio: true
+      - MakeBorderMap:
+          shrink_ratio: 0.4
+          thresh_min: 0.3
+          thresh_max: 0.7
+      - MakeShrinkMap:
+          shrink_ratio: 0.4
+          min_text_size: 8
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+#           image_shape: [736, 1280]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml b/configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml
new file mode 100644
index 0000000..b7195ff
--- /dev/null
+++ b/configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml
@@ -0,0 +1,165 @@
+Global:
+  model_name: PP-OCRv3_mobile_det # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/PP-OCRv3_mobile_det/
+  save_epoch_step: 100
+  eval_batch_step:
+  - 0
+  - 400
+  cal_metric_during_train: false
+  pretrained_model: https://paddleocr.bj.bcebos.com/pretrained/MobileNetV3_large_x0_5_pretrained.pdparams
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./checkpoints/det_db/predicts_db.txt
+  distributed: true
+  d2s_train_image_shape: [3, -1, -1]
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+    disable_se: True
+  Neck:
+    name: RSEFPN
+    out_channels: 96
+    shortcut: True
+  Head:
+    name: DBHead
+    k: 50
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 5.0e-05
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            rotate:
+            - -10
+            - 10
+        - type: Resize
+          args:
+            size:
+            - 0.5
+            - 3
+    - EastRandomCropData:
+        size:
+        - 960
+        - 960
+        max_tries: 50
+        keep_ratio: true
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 8
+    num_workers: 4
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest: null
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 2
diff --git a/configs/det/PP-OCRv3/PP-OCRv3_server_det.yml b/configs/det/PP-OCRv3/PP-OCRv3_server_det.yml
new file mode 100644
index 0000000..d052ed2
--- /dev/null
+++ b/configs/det/PP-OCRv3/PP-OCRv3_server_det.yml
@@ -0,0 +1,160 @@
+Global:
+  model_name: PP-OCRv3_server_det # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/PP-OCRv3_server_det/
+  save_epoch_step: 100
+  eval_batch_step:
+  - 0
+  - 400
+  cal_metric_during_train: false
+  pretrained_model: https://paddleocr.bj.bcebos.com/pretrained/ResNet50_vd_ssld_pretrained.pdparams
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./checkpoints/det_db/predicts_db.txt
+  distributed: true
+  d2s_train_image_shape: [3, -1, -1]
+  amp_dtype: bfloat16
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Backbone:
+    name: ResNet_vd
+    in_channels: 3
+    layers: 50
+  Neck:
+    name: LKPAN
+    out_channels: 256
+  Head:
+    name: DBHead
+    kernel_list: [7,2,2]
+    k: 50
+
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 5.0e-05
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - CopyPaste:
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            rotate:
+            - -10
+            - 10
+        - type: Resize
+          args:
+            size:
+            - 0.5
+            - 3
+    - EastRandomCropData:
+        size:
+        - 960
+        - 960
+        max_tries: 50
+        keep_ratio: true
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
+profiler_options: null
diff --git a/configs/det/PP-OCRv4/PP-OCRv4_det_cml.yml b/configs/det/PP-OCRv4/PP-OCRv4_det_cml.yml
new file mode 100644
index 0000000..7e7ddcb
--- /dev/null
+++ b/configs/det/PP-OCRv4/PP-OCRv4_det_cml.yml
@@ -0,0 +1,240 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 20
+  save_model_dir: ./output/ch_PP-OCRv4
+  save_epoch_step: 50
+  eval_batch_step:
+  - 0
+  - 1000
+  cal_metric_during_train: false
+  checkpoints: null
+  pretrained_model: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./checkpoints/det_db/predicts_db.txt
+  d2s_train_image_shape: [3, 640, 640]
+  distributed: true
+Architecture:
+  name: DistillationModel
+  algorithm: Distillation
+  model_type: det
+  Models:
+    Student:
+      model_type: det
+      algorithm: DB
+      Transform: null
+      Backbone:
+        name: PPLCNetV3
+        scale: 0.75
+        pretrained: false
+        det: true
+      Neck:
+        name: RSEFPN
+        out_channels: 96
+        shortcut: true
+      Head:
+        name: DBHead
+        k: 50
+    Student2:
+      pretrained: null
+      model_type: det
+      algorithm: DB
+      Transform: null
+      Backbone:
+        name: PPLCNetV3
+        scale: 0.75
+        pretrained: true
+        det: true
+      Neck:
+        name: RSEFPN
+        out_channels: 96
+        shortcut: true
+      Head:
+        name: DBHead
+        k: 50
+    Teacher:
+      pretrained: https://paddleocr.bj.bcebos.com/PP-OCRv4/chinese/ch_PP-OCRv4_det_cml_teacher_pretrained/teacher.pdparams
+      freeze_params: true
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Backbone:
+        name: ResNet_vd
+        in_channels: 3
+        layers: 50
+      Neck:
+        name: LKPAN
+        out_channels: 256
+      Head:
+        name: DBHead
+        kernel_list:
+        - 7
+        - 2
+        - 2
+        k: 50
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationDilaDBLoss:
+      weight: 1.0
+      model_name_pairs:
+      - - Student
+        - Teacher
+      - - Student2
+        - Teacher
+      key: maps
+      balance_loss: true
+      main_loss_type: DiceLoss
+      alpha: 5
+      beta: 10
+      ohem_ratio: 3
+  - DistillationDMLLoss:
+      model_name_pairs:
+      - Student
+      - Student2
+      maps_name: thrink_maps
+      weight: 1.0
+      key: maps
+  - DistillationDBLoss:
+      weight: 1.0
+      model_name_list:
+      - Student
+      - Student2
+      balance_loss: true
+      main_loss_type: DiceLoss
+      alpha: 5
+      beta: 10
+      ohem_ratio: 3
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 5.0e-05
+PostProcess:
+  name: DistillationDBPostProcess
+  model_name:
+  - Student
+  key: head_out
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+Metric:
+  name: DistillationMetric
+  base_metric_name: DetMetric
+  main_indicator: hmean
+  key: Student
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            rotate:
+            - -10
+            - 10
+        - type: Resize
+          args:
+            size:
+            - 0.5
+            - 3
+    - EastRandomCropData:
+        size:
+        - 640
+        - 640
+        max_tries: 50
+        keep_ratio: true
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+        total_epoch: 500
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+        total_epoch: 500
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 16
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest: 
+        limit_side_len: 960
+        limit_type: max
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 2
+profiler_options: null
diff --git a/configs/det/PP-OCRv4/PP-OCRv4_mobile_det.yml b/configs/det/PP-OCRv4/PP-OCRv4_mobile_det.yml
new file mode 100644
index 0000000..f243d17
--- /dev/null
+++ b/configs/det/PP-OCRv4/PP-OCRv4_mobile_det.yml
@@ -0,0 +1,174 @@
+Global:
+  model_name: PP-OCRv4_mobile_det # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: &epoch_num 500
+  log_smooth_window: 20
+  print_batch_step: 100
+  save_model_dir: ./output/PP-OCRv4_mobile_det
+  save_epoch_step: 10
+  eval_batch_step:
+  - 0
+  - 1500
+  cal_metric_during_train: false
+  checkpoints:
+  pretrained_model: https://paddleocr.bj.bcebos.com/pretrained/PPLCNetV3_x0_75_ocr_det.pdparams
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./checkpoints/det_db/predicts_db.txt
+  d2s_train_image_shape: [3, 640, 640]
+  distributed: true
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform: null
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.75
+    det: True
+  Neck:
+    name: RSEFPN
+    out_channels: 96
+    shortcut: True
+  Head:
+    name: DBHead
+    k: 50
+    fix_nan: True
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001 #(8*8c)
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 5.0e-05
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - CopyPaste: null
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            rotate:
+            - -10
+            - 10
+        - type: Resize
+          args:
+            size:
+            - 0.5
+            - 3
+    - EastRandomCropData:
+        size:
+        - 640
+        - 640
+        max_tries: 50
+        keep_ratio: true
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+        total_epoch: *epoch_num
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+        total_epoch: *epoch_num
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 8
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest:
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 2
+profiler_options: null
diff --git a/configs/det/PP-OCRv4/PP-OCRv4_mobile_seal_det.yml b/configs/det/PP-OCRv4/PP-OCRv4_mobile_seal_det.yml
new file mode 100644
index 0000000..79a45cd
--- /dev/null
+++ b/configs/det/PP-OCRv4/PP-OCRv4_mobile_seal_det.yml
@@ -0,0 +1,171 @@
+Global:
+  model_name: PP-OCRv4_mobile_seal_det # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 100
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: output
+  save_epoch_step: 1
+  eval_batch_step:
+  - 0
+  - 100
+  cal_metric_during_train: false
+  checkpoints:
+  pretrained_model: https://paddleocr.bj.bcebos.com/pretrained/PPLCNetV3_x0_75_ocr_det.pdparams
+  save_inference_dir: null
+  use_visualdl: false
+  distributed: true
+  d2s_train_image_shape: [3, 640, 640]
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform: null
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.75
+    det: True
+  Neck:
+    name: RSEFPN
+    out_channels: 96
+    shortcut: True
+  Head:
+    name: DBHead
+    k: 50
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 1e-6
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.2
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 0.5
+  box_type: "poly"
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: TextDetDataset
+    data_dir: datasets/ICDAR2015
+    label_file_list:
+      - datasets/ICDAR2015/train.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            rotate:
+            - -10
+            - 10
+        - type: Resize
+          args:
+            size:
+            - 0.5
+            - 3
+    - EastRandomCropData:
+        size:
+        - 640
+        - 640
+        max_tries: 50
+        keep_ratio: true
+    - MakeBorderMap:
+        shrink_ratio: 0.8
+        thresh_min: 0.3
+        thresh_max: 0.7
+        total_epoch: 500
+    - MakeShrinkMap:
+        shrink_ratio: 0.8
+        min_text_size: 8
+        total_epoch: 500
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 8
+    num_workers: 3
+
+Eval:
+  dataset:
+    name: TextDetDataset
+    data_dir: datasets/ICDAR2015
+    label_file_list:
+      - datasets/ICDAR2015/val.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest:
+        resize_long: 736
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 0
+profiler_options: null
diff --git a/configs/det/PP-OCRv4/PP-OCRv4_server_det.yml b/configs/det/PP-OCRv4/PP-OCRv4_server_det.yml
new file mode 100644
index 0000000..bac6311
--- /dev/null
+++ b/configs/det/PP-OCRv4/PP-OCRv4_server_det.yml
@@ -0,0 +1,175 @@
+Global:
+  model_name: PP-OCRv4_server_det # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: &epoch_num 500
+  log_smooth_window: 20
+  print_batch_step: 100
+  save_model_dir: ./output/PP-OCRv4_server_det
+  save_epoch_step: 10
+  eval_batch_step:
+  - 0
+  - 1500
+  cal_metric_during_train: false
+  checkpoints:
+  pretrained_model: https://paddleocr.bj.bcebos.com/pretrained/PPHGNet_small_ocr_det.pdparams
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./checkpoints/det_db/predicts_db.txt
+  d2s_train_image_shape: [3, 640, 640]
+  distributed: true
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform: null
+  Backbone:
+    name: PPHGNet_small
+    det: True
+  Neck:
+    name: LKPAN
+    out_channels: 256
+    intracl: true
+  Head:
+    name: PFHeadLocal
+    k: 50
+    mode: "large"
+    fix_nan: True
+    
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001 #(8*8c)
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 1e-6
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - CopyPaste: null
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            rotate:
+            - -10
+            - 10
+        - type: Resize
+          args:
+            size:
+            - 0.5
+            - 3
+    - EastRandomCropData:
+        size:
+        - 640
+        - 640
+        max_tries: 50
+        keep_ratio: true
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+        total_epoch: *epoch_num
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+        total_epoch: *epoch_num
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 8
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest:
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 2
+profiler_options: null
diff --git a/configs/det/PP-OCRv4/PP-OCRv4_server_seal_det.yml b/configs/det/PP-OCRv4/PP-OCRv4_server_seal_det.yml
new file mode 100644
index 0000000..ebf1341
--- /dev/null
+++ b/configs/det/PP-OCRv4/PP-OCRv4_server_seal_det.yml
@@ -0,0 +1,171 @@
+Global:
+  model_name: PP-OCRv4_server_seal_det # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 100
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: output
+  save_epoch_step: 1
+  eval_batch_step:
+  - 0
+  - 100
+  cal_metric_during_train: false
+  checkpoints:
+  pretrained_model: https://paddleocr.bj.bcebos.com/pretrained/PPHGNet_small_ocr_det.pdparams
+  save_inference_dir: null
+  use_visualdl: false
+  distributed: true
+  d2s_train_image_shape: [3, 640, 640]
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform: null
+  Backbone:
+    name: PPHGNet_small
+    det: True
+  Neck:
+    name: LKPAN
+    out_channels: 256
+    intracl: true
+  Head:
+    name: PFHeadLocal
+    k: 50
+    mode: "large"
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 1e-6
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.2
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 0.5
+  box_type: "poly"
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: TextDetDataset
+    data_dir: datasets/ICDAR2015
+    label_file_list:
+      - datasets/ICDAR2015/train.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            rotate:
+            - -10
+            - 10
+        - type: Resize
+          args:
+            size:
+            - 0.5
+            - 3
+    - EastRandomCropData:
+        size:
+        - 640
+        - 640
+        max_tries: 50
+        keep_ratio: true
+    - MakeBorderMap:
+        shrink_ratio: 0.8
+        thresh_min: 0.3
+        thresh_max: 0.7
+        total_epoch: 500
+    - MakeShrinkMap:
+        shrink_ratio: 0.8
+        min_text_size: 8
+        total_epoch: 500
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 4
+    num_workers: 3
+
+Eval:
+  dataset:
+    name: TextDetDataset
+    data_dir: datasets/ICDAR2015
+    label_file_list:
+      - datasets/ICDAR2015/val.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest:
+        resize_long: 736
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 0
+profiler_options: null
diff --git a/configs/det/PP-OCRv5/PP-OCRv5_mobile_det.yml b/configs/det/PP-OCRv5/PP-OCRv5_mobile_det.yml
new file mode 100644
index 0000000..55d4da6
--- /dev/null
+++ b/configs/det/PP-OCRv5/PP-OCRv5_mobile_det.yml
@@ -0,0 +1,174 @@
+Global:
+  model_name: PP-OCRv5_mobile_det # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: &epoch_num 500
+  log_smooth_window: 20
+  print_batch_step: 100
+  save_model_dir: ./output/PP-OCRv5_mobile_det
+  save_epoch_step: 10
+  eval_batch_step:
+  - 0
+  - 1500
+  cal_metric_during_train: false
+  checkpoints:
+  pretrained_model: https://paddleocr.bj.bcebos.com/pretrained/PPLCNetV3_x0_75_ocr_det.pdparams
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./checkpoints/det_db/predicts_db.txt
+  d2s_train_image_shape: [3, 640, 640]
+  distributed: true
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform: null
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.75
+    det: True
+  Neck:
+    name: RSEFPN
+    out_channels: 96
+    shortcut: True
+  Head:
+    name: DBHead
+    k: 50
+    fix_nan: True
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001 #(8*8c)
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 5.0e-05
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - CopyPaste: null
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            rotate:
+            - -10
+            - 10
+        - type: Resize
+          args:
+            size:
+            - 0.5
+            - 3
+    - EastRandomCropData:
+        size:
+        - 640
+        - 640
+        max_tries: 50
+        keep_ratio: true
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+        total_epoch: *epoch_num
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+        total_epoch: *epoch_num
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 8
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest:
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 2
+profiler_options: null
diff --git a/configs/det/PP-OCRv5/PP-OCRv5_server_det.yml b/configs/det/PP-OCRv5/PP-OCRv5_server_det.yml
new file mode 100644
index 0000000..7c88fc8
--- /dev/null
+++ b/configs/det/PP-OCRv5/PP-OCRv5_server_det.yml
@@ -0,0 +1,173 @@
+Global:
+  model_name: PP-OCRv5_server_det # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: &epoch_num 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/PP-OCRv5_server_det
+  save_epoch_step: 10
+  eval_batch_step:
+  - 0
+  - 1500
+  cal_metric_during_train: false
+  checkpoints:
+  pretrained_model: https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PPHGNetV2_B4_ocr_det.pdparams
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./checkpoints/det_db/predicts_db.txt
+  distributed: true
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform: null
+  Backbone:
+    name: PPHGNetV2_B4
+    det: True
+  Neck:
+    name: LKPAN
+    out_channels: 256
+    intracl: true
+  Head:
+    name: PFHeadLocal
+    k: 50
+    mode: "large"
+    
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001 #(8*8c)
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 1e-6
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - CopyPaste: null
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            rotate:
+            - -10
+            - 10
+        - type: Resize
+          args:
+            size:
+            - 0.5
+            - 3
+    - EastRandomCropData:
+        size:
+        - 640
+        - 640
+        max_tries: 50
+        keep_ratio: true
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+        total_epoch: *epoch_num
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+        total_epoch: *epoch_num
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 8
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest:
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 2
+profiler_options: null
diff --git a/configs/det/PP-OCRv6/PP-OCRv6_medium_det.yml b/configs/det/PP-OCRv6/PP-OCRv6_medium_det.yml
new file mode 100644
index 0000000..19b2a4e
--- /dev/null
+++ b/configs/det/PP-OCRv6/PP-OCRv6_medium_det.yml
@@ -0,0 +1,195 @@
+Global:
+  model_name: PP-OCRv6_medium_det # To use static model for inference.
+  debug: false
+  use_gpu: true
+  use_ema: true
+  ema_decay: 0.9996
+  ema_decay_type: threshold
+  epoch_num: &epoch_num 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/PP-OCRv6_medium_det
+  save_epoch_step: 10
+  eval_batch_step:
+  - 0
+  - 1500
+  cal_metric_during_train: false
+  checkpoints: null
+  pretrained_model: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./checkpoints/det_db/predicts_db.txt
+  d2s_train_image_shape: [3, 640, 640]
+  distributed: true
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform: null
+  Backbone:
+    name: PPLCNetV4
+    det: true
+    model_size: medium
+  Neck:
+    name: RepLKPAN
+    out_channels: 256
+    intracl: true
+  Head:
+    name: DBHead
+    k: 50
+    fix_nan: true
+    aux_in_channels: 256
+
+Loss:
+  name: DBLoss
+  main_loss_type: DiceFocalLoss
+  alpha: 5
+  beta: 10
+  focal_alpha: 0.25
+  focal_gamma: 2.5
+  aux_weight_p4: 0.2
+  aux_weight_p3: 0.3
+  aux_weight_p2: 0.4
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001 #(8*8c)
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 1.0e-05
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.2
+  box_thresh: 0.45
+  max_candidates: 3000
+  unclip_ratio: 1.4
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - CopyPaste: null
+    - ColorJitter:
+        brightness: 0.4
+        contrast: 0.4
+        saturation: 0.7
+        hue: 0.1
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            p: 0.5
+            rotate:
+            - -45
+            - 45
+            fit_output: true
+        - type: Resize
+          args:
+            size:
+            - 0.1
+            - 2
+    - RandomPerspective:
+        prob: 0.3
+        scale: 0.0
+        shear: 20.0
+        perspective: 0.001
+        fit_output: true
+        fill_value: [0, 0, 0]
+    - RandomCrop:
+        size:
+        - 640
+        - 640
+        max_tries: 50
+        keep_ratio: true
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+        total_epoch: *epoch_num
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+        total_epoch: *epoch_num
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 8
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest: null
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 2
+profiler_options: null
diff --git a/configs/det/PP-OCRv6/PP-OCRv6_small_det.yml b/configs/det/PP-OCRv6/PP-OCRv6_small_det.yml
new file mode 100644
index 0000000..12b1327
--- /dev/null
+++ b/configs/det/PP-OCRv6/PP-OCRv6_small_det.yml
@@ -0,0 +1,184 @@
+Global:
+  model_name: PP-OCRv6_small_det # To use static model for inference.
+  debug: false
+  use_gpu: true
+  use_ema: true
+  ema_decay: 0.9997
+  ema_decay_type: threshold
+  epoch_num: &epoch_num 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/PP-OCRv6_small_det
+  save_epoch_step: 10
+  eval_batch_step:
+  - 0
+  - 1500
+  cal_metric_during_train: false
+  checkpoints: null
+  pretrained_model: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./checkpoints/det_db/predicts_db.txt
+  d2s_train_image_shape: [3, 640, 640]
+  distributed: true
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform: null
+  Backbone:
+    name: PPLCNetV4
+    det: true
+    model_size: small
+  Neck:
+    name: RepLKFPN
+    out_channels: 96
+    dilated_kernel_size: 7
+    shortcut: true
+  Head:
+    name: DBHead
+    k: 50
+    fix_nan: true
+    aux_in_channels: 96
+
+Loss:
+  name: DBLoss
+  main_loss_type: DiceFocalLoss
+  alpha: 5
+  beta: 10
+  focal_alpha: 0.25
+  focal_gamma: 2.5
+  aux_weight_p4: 0.2
+  aux_weight_p3: 0.3
+  aux_weight_p2: 0.4
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001 #(8*8c)
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 1.0e-05
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.2
+  box_thresh: 0.45
+  max_candidates: 3000
+  unclip_ratio: 1.4
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - CopyPaste: null
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            p: 0.5
+            rotate:
+            - -45
+            - 45
+            fit_output: true
+        - type: Resize
+          args:
+            size:
+            - 0.1
+            - 2
+    - RandomCrop:
+        size:
+        - 640
+        - 640
+        max_tries: 50
+        keep_ratio: true
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+        total_epoch: *epoch_num
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+        total_epoch: *epoch_num
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 8
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest: null
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 2
+profiler_options: null
diff --git a/configs/det/PP-OCRv6/PP-OCRv6_tiny_det.yml b/configs/det/PP-OCRv6/PP-OCRv6_tiny_det.yml
new file mode 100644
index 0000000..d6deb55
--- /dev/null
+++ b/configs/det/PP-OCRv6/PP-OCRv6_tiny_det.yml
@@ -0,0 +1,189 @@
+Global:
+  model_name: PP-OCRv6_tiny_det # To use static model for inference.
+  debug: false
+  use_gpu: true
+  use_ema: true
+  ema_decay: 0.9998
+  ema_decay_type: threshold
+  epoch_num: &epoch_num 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/PP-OCRv6_tiny_det
+  save_epoch_step: 10
+  eval_batch_step:
+  - 0
+  - 1500
+  cal_metric_during_train: false
+  checkpoints: null
+  pretrained_model: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./checkpoints/det_db/predicts_db.txt
+  d2s_train_image_shape: [3, 640, 640]
+  distributed: true
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform: null
+  Backbone:
+    name: PPLCNetV4
+    det: true
+    model_size: tiny
+  Neck:
+    name: RepLKFPN
+    out_channels: 64
+    dilated_kernel_size: 5
+    shortcut: true
+  Head:
+    name: DBHead
+    k: 50
+    fix_nan: true
+    aux_in_channels: 64
+
+Loss:
+  name: DBLoss
+  main_loss_type: DiceFocalLoss
+  alpha: 5
+  beta: 10
+  focal_alpha: 0.25
+  focal_gamma: 2.5
+  aux_weight_p4: 0.2
+  aux_weight_p3: 0.3
+  aux_weight_p2: 0.4
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001 #(8*8c)
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 1.0e-06
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.2
+  box_thresh: 0.4
+  max_candidates: 3000
+  unclip_ratio: 1.4
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - CopyPaste: null
+    - ColorJitter:
+        brightness: 0.4
+        contrast: 0.4
+        saturation: 0.7
+        hue: 0.1
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            p: 0.5
+            rotate:
+            - -45
+            - 45
+            fit_output: true
+        - type: Resize
+          args:
+            size:
+            - 0.3
+            - 2
+    - RandomCrop:
+        size:
+        - 640
+        - 640
+        max_tries: 50
+        keep_ratio: true
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+        total_epoch: *epoch_num
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+        total_epoch: *epoch_num
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 8
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest: null
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 2
+profiler_options: null
diff --git a/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml b/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml
new file mode 100644
index 0000000..d35e08a
--- /dev/null
+++ b/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml
@@ -0,0 +1,206 @@
+Global:
+  use_gpu: true
+  epoch_num: 1200
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/ch_db_mv3/
+  save_epoch_step: 1200
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [3000, 2000]
+  cal_metric_during_train: False
+  pretrained_model: 
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_db/predicts_db.txt
+  use_amp: False
+  amp_level: O2
+  amp_dtype: bfloat16
+
+Architecture:
+  name: DistillationModel
+  algorithm: Distillation
+  model_type: det
+  Models:
+    Teacher:
+      pretrained: ./pretrain_models/ch_ppocr_server_v2.0_det_train/best_accuracy
+      freeze_params: true
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Transform:
+      Backbone:
+        name: ResNet_vd
+        layers: 18
+      Neck:
+        name: DBFPN
+        out_channels: 256
+      Head:
+        name: DBHead
+        k: 50
+    Student:
+      pretrained:
+      freeze_params: false
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Backbone:
+        name: MobileNetV3
+        scale: 0.5
+        model_name: large
+        disable_se: True
+      Neck:
+        name: DBFPN
+        out_channels: 96
+      Head:
+        name: DBHead
+        k: 50
+    Student2:
+      pretrained:
+      freeze_params: false
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Transform:
+      Backbone:
+        name: MobileNetV3
+        scale: 0.5
+        model_name: large
+        disable_se: True
+      Neck:
+        name: DBFPN
+        out_channels: 96
+      Head:
+        name: DBHead
+        k: 50
+    
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationDilaDBLoss:
+      weight: 1.0
+      model_name_pairs:
+      - ["Student", "Teacher"]
+      - ["Student2", "Teacher"]
+      key: maps
+      balance_loss: true
+      main_loss_type: DiceLoss
+      alpha: 5
+      beta: 10
+      ohem_ratio: 3
+  - DistillationDMLLoss:
+      model_name_pairs:
+      - ["Student", "Student2"]
+      maps_name: "thrink_maps"
+      weight: 1.0
+      # act: None
+      model_name_pairs: ["Student", "Student2"]
+      key: maps
+  - DistillationDBLoss:
+      weight: 1.0
+      model_name_list: ["Student", "Student2"]
+      # key: maps
+      # name: DBLoss
+      balance_loss: true
+      main_loss_type: DiceLoss
+      alpha: 5
+      beta: 10
+      ohem_ratio: 3
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 2
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: DistillationDBPostProcess
+  model_name: ["Student", "Student2", "Teacher"]
+  # key: maps
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DistillationMetric
+  base_metric_name: DetMetric
+  main_indicator: hmean
+  key: "Student"
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - CopyPaste:
+      - IaaAugment:
+          augmenter_args:
+            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
+            - { 'type': Affine, 'args': { 'rotate': [-10, 10] } }
+            - { 'type': Resize, 'args': { 'size': [0.5, 3] } }
+      - EastRandomCropData:
+          size: [960, 960]
+          max_tries: 50
+          keep_ratio: true
+      - MakeBorderMap:
+          shrink_ratio: 0.4
+          thresh_min: 0.3
+          thresh_max: 0.7
+      - MakeShrinkMap:
+          shrink_ratio: 0.4
+          min_text_size: 8
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_distill.yml b/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_distill.yml
new file mode 100644
index 0000000..d24ee11
--- /dev/null
+++ b/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_distill.yml
@@ -0,0 +1,175 @@
+Global:
+  use_gpu: true
+  epoch_num: 1200
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/ch_db_mv3/
+  save_epoch_step: 1200
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [3000, 2000]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_db/predicts_db.txt
+
+Architecture:
+  name: DistillationModel
+  algorithm: Distillation
+  model_type: det
+  Models:
+    Student:
+      pretrained: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+      freeze_params: false
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Backbone:
+        name: MobileNetV3
+        scale: 0.5
+        model_name: large
+        disable_se: True
+      Neck:
+        name: DBFPN
+        out_channels: 96
+      Head:
+        name: DBHead
+        k: 50
+    Teacher:
+      pretrained: ./pretrain_models/ch_ppocr_server_v2.0_det_train/best_accuracy
+      freeze_params: true
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Transform:
+      Backbone:
+        name: ResNet_vd
+        layers: 18
+      Neck:
+        name: DBFPN
+        out_channels: 256
+      Head:
+        name: DBHead
+        k: 50
+
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationDilaDBLoss:
+      weight: 1.0
+      model_name_pairs:
+      - ["Student", "Teacher"]
+      key: maps
+      balance_loss: true
+      main_loss_type: DiceLoss
+      alpha: 5
+      beta: 10
+      ohem_ratio: 3
+  - DistillationDBLoss:
+      weight: 1.0
+      model_name_list: ["Student"]
+      name: DBLoss
+      balance_loss: true
+      main_loss_type: DiceLoss
+      alpha: 5
+      beta: 10
+      ohem_ratio: 3
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 2
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: DistillationDBPostProcess
+  model_name: ["Student"]
+  key: head_out
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DistillationMetric
+  base_metric_name: DetMetric
+  main_indicator: hmean
+  key: "Student"
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - CopyPaste:
+      - IaaAugment:
+          augmenter_args:
+            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
+            - { 'type': Affine, 'args': { 'rotate': [-10, 10] } }
+            - { 'type': Resize, 'args': { 'size': [0.5, 3] } }
+      - EastRandomCropData:
+          size: [960, 960]
+          max_tries: 50
+          keep_ratio: true
+      - MakeBorderMap:
+          shrink_ratio: 0.4
+          thresh_min: 0.3
+          thresh_max: 0.7
+      - MakeShrinkMap:
+          shrink_ratio: 0.4
+          min_text_size: 8
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+#           image_shape: [736, 1280]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_dml.yml b/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_dml.yml
new file mode 100644
index 0000000..8b160f6
--- /dev/null
+++ b/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_dml.yml
@@ -0,0 +1,178 @@
+Global:
+  use_gpu: true
+  epoch_num: 1200
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/ch_db_mv3/
+  save_epoch_step: 1200
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [3000, 2000]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_db/predicts_db.txt
+
+Architecture:
+  name: DistillationModel
+  algorithm: Distillation
+  model_type: det
+  Models:
+    Student:
+      pretrained: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+      freeze_params: false
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Backbone:
+        name: MobileNetV3
+        scale: 0.5
+        model_name: large
+        disable_se: True
+      Neck:
+        name: DBFPN
+        out_channels: 96
+      Head:
+        name: DBHead
+        k: 50
+    Teacher:
+      pretrained: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+      freeze_params: false
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Transform:
+      Backbone:
+        name: MobileNetV3
+        scale: 0.5
+        model_name: large
+        disable_se: True
+      Neck:
+        name: DBFPN
+        out_channels: 96
+      Head:
+        name: DBHead
+        k: 50
+
+
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationDMLLoss:
+      model_name_pairs:
+      - ["Student", "Teacher"]
+      maps_name: "thrink_maps"
+      weight: 1.0
+      # act: None
+      model_name_pairs: ["Student", "Teacher"]
+      key: maps
+  - DistillationDBLoss:
+      weight: 1.0
+      model_name_list: ["Student", "Teacher"]
+      # key: maps
+      name: DBLoss
+      balance_loss: true
+      main_loss_type: DiceLoss
+      alpha: 5
+      beta: 10
+      ohem_ratio: 3
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 2
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: DistillationDBPostProcess
+  model_name: ["Student", "Teacher"]
+  key: head_out
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DistillationMetric
+  base_metric_name: DetMetric
+  main_indicator: hmean
+  key: "Student"
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - CopyPaste:
+      - IaaAugment:
+          augmenter_args:
+            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
+            - { 'type': Affine, 'args': { 'rotate': [-10, 10] } }
+            - { 'type': Resize, 'args': { 'size': [0.5, 3] } }
+      - EastRandomCropData:
+          size: [960, 960]
+          max_tries: 50
+          keep_ratio: true
+      - MakeBorderMap:
+          shrink_ratio: 0.4
+          thresh_min: 0.3
+          thresh_max: 0.7
+      - MakeShrinkMap:
+          shrink_ratio: 0.4
+          min_text_size: 8
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+#           image_shape: [736, 1280]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_student.yml b/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_student.yml
new file mode 100644
index 0000000..cca2a59
--- /dev/null
+++ b/configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_student.yml
@@ -0,0 +1,132 @@
+Global:
+  use_gpu: true
+  epoch_num: 1200
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/ch_db_mv3/
+  save_epoch_step: 1200
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 400]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/student.pdparams
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_db/predicts_db.txt
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+    disable_se: True
+  Neck:
+    name: DBFPN
+    out_channels: 96
+  Head:
+    name: DBHead
+    k: 50
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 2
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - IaaAugment:
+          augmenter_args:
+            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
+            - { 'type': Affine, 'args': { 'rotate': [-10, 10] } }
+            - { 'type': Resize, 'args': { 'size': [0.5, 3] } }
+      - EastRandomCropData:
+          size: [960, 960]
+          max_tries: 50
+          keep_ratio: true
+      - MakeBorderMap:
+          shrink_ratio: 0.4
+          thresh_min: 0.3
+          thresh_max: 0.7
+      - MakeShrinkMap:
+          shrink_ratio: 0.4
+          min_text_size: 8
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+#           image_shape: [736, 1280]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml b/configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml
new file mode 100644
index 0000000..87718ca
--- /dev/null
+++ b/configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml
@@ -0,0 +1,132 @@
+Global:
+  use_gpu: true
+  epoch_num: 1200
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/ch_db_mv3/
+  save_epoch_step: 1200
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [3000, 2000]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_db/predicts_db.txt
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+    disable_se: True
+  Neck:
+    name: DBFPN
+    out_channels: 96
+  Head:
+    name: DBHead
+    k: 50
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 2
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - IaaAugment:
+          augmenter_args:
+            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
+            - { 'type': Affine, 'args': { 'rotate': [-10, 10] } }
+            - { 'type': Resize, 'args': { 'size': [0.5, 3] } }
+      - EastRandomCropData:
+          size: [960, 960]
+          max_tries: 50
+          keep_ratio: true
+      - MakeBorderMap:
+          shrink_ratio: 0.4
+          thresh_min: 0.3
+          thresh_max: 0.7
+      - MakeShrinkMap:
+          shrink_ratio: 0.4
+          min_text_size: 8
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+#           image_shape: [736, 1280]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/configs/det/ch_ppocr_v2.0/ch_det_res18_db_v2.0.yml b/configs/det/ch_ppocr_v2.0/ch_det_res18_db_v2.0.yml
new file mode 100644
index 0000000..e983c22
--- /dev/null
+++ b/configs/det/ch_ppocr_v2.0/ch_det_res18_db_v2.0.yml
@@ -0,0 +1,131 @@
+Global:
+  use_gpu: true
+  epoch_num: 1200
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/ch_db_res18/
+  save_epoch_step: 1200
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [3000, 2000]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/ResNet18_vd_pretrained
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_db/predicts_db.txt
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform:
+  Backbone:
+    name: ResNet_vd
+    layers: 18
+    disable_se: True
+  Neck:
+    name: DBFPN
+    out_channels: 256
+  Head:
+    name: DBHead
+    k: 50
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 2
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - IaaAugment:
+          augmenter_args:
+            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
+            - { 'type': Affine, 'args': { 'rotate': [-10, 10] } }
+            - { 'type': Resize, 'args': { 'size': [0.5, 3] } }
+      - EastRandomCropData:
+          size: [960, 960]
+          max_tries: 50
+          keep_ratio: true
+      - MakeBorderMap:
+          shrink_ratio: 0.4
+          thresh_min: 0.3
+          thresh_max: 0.7
+      - MakeShrinkMap:
+          shrink_ratio: 0.4
+          min_text_size: 8
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+#           image_shape: [736, 1280]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/configs/det/det_mv3_db.yml b/configs/det/det_mv3_db.yml
new file mode 100644
index 0000000..8f5685e
--- /dev/null
+++ b/configs/det/det_mv3_db.yml
@@ -0,0 +1,133 @@
+Global:
+  use_gpu: true
+  use_xpu: false
+  use_mlu: false
+  epoch_num: 1200
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/db_mv3/
+  save_epoch_step: 1200
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_db/predicts_db.txt
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+  Neck:
+    name: DBFPN
+    out_channels: 256
+  Head:
+    name: DBHead
+    k: 50
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - IaaAugment:
+          augmenter_args:
+            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
+            - { 'type': Affine, 'args': { 'rotate': [-10, 10] } }
+            - { 'type': Resize, 'args': { 'size': [0.5, 3] } }
+      - EastRandomCropData:
+          size: [640, 640]
+          max_tries: 50
+          keep_ratio: true
+      - MakeBorderMap:
+          shrink_ratio: 0.4
+          thresh_min: 0.3
+          thresh_max: 0.7
+      - MakeShrinkMap:
+          shrink_ratio: 0.4
+          min_text_size: 8
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 16
+    num_workers: 8
+    use_shared_memory: True
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          image_shape: [736, 1280]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 8
+    use_shared_memory: True
diff --git a/configs/det/det_mv3_east.yml b/configs/det/det_mv3_east.yml
new file mode 100644
index 0000000..461179e
--- /dev/null
+++ b/configs/det/det_mv3_east.yml
@@ -0,0 +1,109 @@
+Global:
+  use_gpu: true
+  epoch_num: 10000
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/east_mv3/
+  save_epoch_step: 1000
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [4000, 5000]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: 
+  save_res_path: ./output/det_east/predicts_east.txt
+
+Architecture:
+  model_type: det
+  algorithm: EAST
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+  Neck:
+    name: EASTFPN
+    model_name: small
+  Head:
+    name: EASTHead
+    model_name: small
+
+Loss:
+  name: EASTLoss
+  
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+  #  name: Cosine
+    learning_rate: 0.001
+  #  warmup_epoch: 0
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: EASTPostProcess
+  score_thresh: 0.8
+  cover_thresh: 0.1
+  nms_thresh: 0.2
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - EASTProcessTrain:
+          image_shape: [512, 512]
+          background_ratio: 0.125
+          min_crop_side_ratio: 0.1
+          min_text_size: 10
+      - KeepKeys:
+          keep_keys: ['image', 'score_map', 'geo_map', 'training_mask'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 16
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          limit_side_len: 2400
+          limit_type: max
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/configs/det/det_mv3_pse.yml b/configs/det/det_mv3_pse.yml
new file mode 100644
index 0000000..4b8c4be
--- /dev/null
+++ b/configs/det/det_mv3_pse.yml
@@ -0,0 +1,135 @@
+Global:
+  use_gpu: true
+  epoch_num: 600
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/det_mv3_pse/
+  save_epoch_step: 600
+  # evaluation is run every 63 iterations
+  eval_batch_step: [ 0,63 ]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+  checkpoints: #./output/det_r50_vd_pse_batch8_ColorJitter/best_accuracy
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_pse/predicts_pse.txt
+
+Architecture:
+  model_type: det
+  algorithm: PSE
+  Transform: null
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+  Neck:
+    name: FPN
+    out_channels: 96
+  Head:
+    name: PSEHead
+    hidden_dim: 96
+    out_channels: 7
+
+Loss:
+  name: PSELoss
+  alpha: 0.7
+  ohem_ratio: 3
+  kernel_sample_mask: pred
+  reduction: none
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Step
+    learning_rate: 0.001
+    step_size: 200
+    gamma: 0.1
+  regularizer:
+    name: 'L2'
+    factor: 0.0005
+
+PostProcess:
+  name: PSEPostProcess
+  thresh: 0
+  box_thresh: 0.85
+  min_area: 16
+  box_type: quad # 'quad' or 'poly'
+  scale: 1
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - ColorJitter:
+          brightness: 0.12549019607843137
+          saturation: 0.5
+      - IaaAugment:
+          augmenter_args:
+            - { 'type': Resize, 'args': { 'size': [ 0.5, 3 ] } }
+            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
+            - { 'type': Affine, 'args': { 'rotate': [ -10, 10 ] } }
+      - MakePseGt:
+          kernel_num: 7
+          min_shrink_ratio: 0.4
+          size: 640
+      - RandomCropImgMask:
+          size: [ 640,640 ]
+          main_key: gt_text
+          crop_keys: [ 'image', 'gt_text', 'gt_kernels', 'mask' ]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [ 0.485, 0.456, 0.406 ]
+          std: [ 0.229, 0.224, 0.225 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'image', 'gt_text', 'gt_kernels', 'mask' ] # the order of the dataloader list
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 16
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          limit_side_len: 736
+          limit_type: min
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [ 0.485, 0.456, 0.406 ]
+          std: [ 0.229, 0.224, 0.225 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'image', 'shape', 'polys', 'ignore_tags' ]
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 8
diff --git a/configs/det/det_r18_vd_ct.yml b/configs/det/det_r18_vd_ct.yml
new file mode 100644
index 0000000..42922df
--- /dev/null
+++ b/configs/det/det_r18_vd_ct.yml
@@ -0,0 +1,107 @@
+Global:
+  use_gpu: true
+  epoch_num: 600
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/det_ct/
+  save_epoch_step: 10
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0,1000]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/ResNet18_vd_pretrained.pdparams
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img623.jpg
+  save_res_path: ./output/det_ct/predicts_ct.txt
+
+Architecture:
+  model_type: det
+  algorithm: CT
+  Transform:
+  Backbone:
+    name: ResNet_vd
+    layers: 18
+  Neck:
+    name: CTFPN
+  Head:
+    name: CT_Head
+    in_channels: 512
+    hidden_dim: 128
+    num_classes: 3
+
+Loss:
+  name: CTLoss
+
+Optimizer:
+  name: Adam
+  lr:  #PolynomialDecay
+    name: Linear 
+    learning_rate: 0.001
+    end_lr: 0.
+    epochs: 600
+    step_each_epoch: 1254
+    power: 0.9
+
+PostProcess:
+  name: CTPostProcess
+  box_type: poly
+
+Metric:
+  name: CTMetric
+  main_indicator: f_score
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/total_text/train
+    label_file_list:
+      - ./train_data/total_text/train/train.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage:
+          img_mode: RGB
+          channel_first: False
+      - CTLabelEncode: # Class handling label
+      - RandomScale:
+      - MakeShrink:
+      - GroupRandomHorizontalFlip:
+      - GroupRandomRotate:
+      - GroupRandomCropPadding:
+      - MakeCentripetalShift:
+      - ColorJitter:
+          brightness: 0.125
+          saturation: 0.5 
+      - ToCHWImage: 
+      - NormalizeImage:
+      - KeepKeys:
+          keep_keys: ['image', 'gt_kernel', 'training_mask', 'gt_instance', 'gt_kernel_instance', 'training_mask_distance', 'gt_distance'] # the order of the dataloader list
+  loader:
+    shuffle: True
+    drop_last: True
+    batch_size_per_card: 4
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/total_text/test
+    label_file_list:
+      - ./train_data/total_text/test/test.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage:
+          img_mode: RGB
+          channel_first: False
+      - CTLabelEncode: # Class handling label
+      - ScaleAlignedShort:
+      - NormalizeImage:
+          order: 'hwc'
+      - ToCHWImage: 
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'texts'] # the order of the dataloader list          
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1
+    num_workers: 2
diff --git a/configs/det/det_r50_db++_icdar15.yml b/configs/det/det_r50_db++_icdar15.yml
new file mode 100644
index 0000000..2bb2cb8
--- /dev/null
+++ b/configs/det/det_r50_db++_icdar15.yml
@@ -0,0 +1,164 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 1000
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/det_r50_icdar15/
+  save_epoch_step: 200
+  eval_batch_step:
+  - 0
+  - 2000
+  cal_metric_during_train: false
+  pretrained_model: ./pretrain_models/ResNet50_dcn_asf_synthtext_pretrained
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./checkpoints/det_db/predicts_db.txt
+Architecture:
+  model_type: det
+  algorithm: DB++
+  Transform: null
+  Backbone:
+    name: ResNet
+    layers: 50
+    dcn_stage: [False, True, True, True]
+  Neck:
+    name: DBFPN
+    out_channels: 256
+    use_asf: True
+  Head:
+    name: DBHead
+    k: 50
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: BCELoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+Optimizer:
+  name: Momentum
+  momentum: 0.9
+  lr:
+    name: DecayLearningRate
+    learning_rate: 0.007
+    epochs: 1000
+    factor: 0.9
+    end_lr: 0
+  weight_decay: 0.0001
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+  det_box_type: 'quad' # 'quad' or 'poly'
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/                                                                        
+    label_file_list:                                                                                            
+    - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt                                                     
+    ratio_list:                                                                                                 
+    - 1.0 
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            rotate:
+            - -10
+            - 10
+        - type: Resize
+          args:
+            size:
+            - 0.5
+            - 3
+    - EastRandomCropData:
+        size:
+        - 640
+        - 640
+        max_tries: 10
+        keep_ratio: true
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.48109378172549
+        - 0.45752457890196
+        - 0.40787054090196
+        std:
+        - 1.0
+        - 1.0
+        - 1.0
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 4
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization                                                                          
+    label_file_list:                                                                                            
+    - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest:
+        image_shape:
+        - 1152
+        - 2048
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.48109378172549
+        - 0.45752457890196
+        - 0.40787054090196
+        std:
+        - 1.0
+        - 1.0
+        - 1.0
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 2
+profiler_options: null
diff --git a/configs/det/det_r50_db++_td_tr.yml b/configs/det/det_r50_db++_td_tr.yml
new file mode 100644
index 0000000..f3b02aa
--- /dev/null
+++ b/configs/det/det_r50_db++_td_tr.yml
@@ -0,0 +1,167 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 1000
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/det_r50_td_tr/
+  save_epoch_step: 200
+  eval_batch_step:
+  - 0
+  - 2000
+  cal_metric_during_train: false
+  pretrained_model: ./pretrain_models/ResNet50_dcn_asf_synthtext_pretrained
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./checkpoints/det_db/predicts_db.txt
+Architecture:
+  model_type: det
+  algorithm: DB++
+  Transform: null
+  Backbone:
+    name: ResNet
+    layers: 50
+    dcn_stage: [False, True, True, True]
+  Neck:
+    name: DBFPN
+    out_channels: 256
+    use_asf: True
+  Head:
+    name: DBHead
+    k: 50
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: BCELoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+Optimizer:
+  name: Momentum
+  momentum: 0.9
+  lr:
+    name: DecayLearningRate
+    learning_rate: 0.007
+    epochs: 1000
+    factor: 0.9
+    end_lr: 0
+  weight_decay: 0.0001
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.5
+  max_candidates: 1000
+  unclip_ratio: 1.5
+  det_box_type: 'quad' # 'quad' or 'poly'
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/                                                                                 
+    label_file_list:                                                                                            
+    - ./train_data/TD_TR/TD500/train_gt_labels.txt                                                           
+    - ./train_data/TD_TR/TR400/gt_labels.txt                                                                 
+    ratio_list:                                                                                                 
+    - 1.0                                                                                                       
+    - 1.0 
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            rotate:
+            - -10
+            - 10
+        - type: Resize
+          args:
+            size:
+            - 0.5
+            - 3
+    - EastRandomCropData:
+        size:
+        - 640
+        - 640
+        max_tries: 10
+        keep_ratio: true
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.48109378172549
+        - 0.45752457890196
+        - 0.40787054090196
+        std:
+        - 1.0
+        - 1.0
+        - 1.0
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 4
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/                                                                          
+    label_file_list:                                                                                            
+    - ./train_data/TD_TR/TD500/test_gt_labels.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest:
+        image_shape:
+        - 736
+        - 736
+        keep_ratio: True
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.48109378172549
+        - 0.45752457890196
+        - 0.40787054090196
+        std:
+        - 1.0
+        - 1.0
+        - 1.0
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 2
+profiler_options: null
diff --git a/configs/det/det_r50_drrg_ctw.yml b/configs/det/det_r50_drrg_ctw.yml
new file mode 100755
index 0000000..f56ac39
--- /dev/null
+++ b/configs/det/det_r50_drrg_ctw.yml
@@ -0,0 +1,133 @@
+Global:
+  use_gpu: true
+  epoch_num: 1200
+  log_smooth_window: 20
+  print_batch_step: 5
+  save_model_dir: ./output/det_r50_drrg_ctw/
+  save_epoch_step: 100
+  # evaluation is run every 1260 iterations
+  eval_batch_step: [37800, 1260]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/ResNet50_vd_ssld_pretrained.pdparams 
+  checkpoints: 
+  save_inference_dir: 
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_drrg/predicts_drrg.txt
+
+
+Architecture:
+  model_type: det
+  algorithm: DRRG
+  Transform:
+  Backbone:
+    name: ResNet_vd
+    layers: 50
+  Neck:
+    name: FPN_UNet
+    in_channels: [256, 512, 1024, 2048]
+    out_channels: 32
+  Head:
+    name: DRRGHead
+    in_channels: 32
+    text_region_thr: 0.3
+    center_region_thr: 0.4
+Loss:
+  name: DRRGLoss
+  
+Optimizer:
+  name: Momentum
+  momentum: 0.9
+  lr:
+    name: DecayLearningRate
+    learning_rate: 0.028
+    epochs: 1200
+    factor: 0.9
+    end_lr: 0.0000001
+  weight_decay: 0.0001
+
+PostProcess:
+  name: DRRGPostprocess
+  link_thr: 0.8
+
+Metric:
+  name: DetFCEMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ctw1500/imgs/
+    label_file_list: 
+      - ./train_data/ctw1500/imgs/training.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+          ignore_orientation: True
+      - DetLabelEncode: # Class handling label
+      - ColorJitter: 
+          brightness: 0.12549019607843137
+          saturation: 0.5
+      - RandomScaling: 
+      - RandomCropFlip:
+          crop_ratio: 0.5
+      - RandomCropPolyInstances:
+          crop_ratio: 0.8
+          min_side_ratio: 0.3
+      - RandomRotatePolyInstances:
+          rotate_ratio: 0.5
+          max_angle: 60
+          pad_with_fixed_color: False
+      - SquareResizePad:
+          target_size: 800
+          pad_ratio: 0.6
+      - IaaAugment:
+          augmenter_args:
+            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
+      - DRRGTargets:
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'gt_text_mask', 'gt_center_region_mask', 'gt_mask',
+            'gt_top_height_map', 'gt_bot_height_map', 'gt_sin_map',
+            'gt_cos_map', 'gt_comp_attribs'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 4
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ctw1500/imgs/
+    label_file_list:
+      - ./train_data/ctw1500/imgs/test.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+          ignore_orientation: True
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          limit_type: 'min'
+          limit_side_len: 640
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - Pad: 
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/configs/det/det_r50_vd_db.yml b/configs/det/det_r50_vd_db.yml
new file mode 100644
index 0000000..288dcc8
--- /dev/null
+++ b/configs/det/det_r50_vd_db.yml
@@ -0,0 +1,128 @@
+Global:
+  use_gpu: true
+  epoch_num: 1200
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/det_r50_vd/
+  save_epoch_step: 1200
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0,2000]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/ResNet50_vd_ssld_pretrained
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_db/predicts_db.txt
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform:
+  Backbone:
+    name: ResNet_vd
+    layers: 50
+  Neck:
+    name: DBFPN
+    out_channels: 256
+  Head:
+    name: DBHead
+    k: 50
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.7
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - IaaAugment:
+          augmenter_args:
+            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
+            - { 'type': Affine, 'args': { 'rotate': [-10, 10] } }
+            - { 'type': Resize, 'args': { 'size': [0.5, 3] } }
+      - EastRandomCropData:
+          size: [640, 640]
+          max_tries: 50
+          keep_ratio: true
+      - MakeBorderMap:
+          shrink_ratio: 0.4
+          thresh_min: 0.3
+          thresh_max: 0.7
+      - MakeShrinkMap:
+          shrink_ratio: 0.4
+          min_text_size: 8
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 16
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          image_shape: [736, 1280]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 8
diff --git a/configs/det/det_r50_vd_dcn_fce_ctw.yml b/configs/det/det_r50_vd_dcn_fce_ctw.yml
new file mode 100755
index 0000000..5e851d1
--- /dev/null
+++ b/configs/det/det_r50_vd_dcn_fce_ctw.yml
@@ -0,0 +1,139 @@
+Global:
+  use_gpu: true
+  epoch_num: 1500
+  log_smooth_window: 20
+  print_batch_step: 20
+  save_model_dir: ./output/det_r50_dcn_fce_ctw/
+  save_epoch_step: 100
+  # evaluation is run every 835 iterations
+  eval_batch_step: [0, 835]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/ResNet50_vd_ssld_pretrained 
+  checkpoints: 
+  save_inference_dir: 
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_fce/predicts_fce.txt
+
+
+Architecture:
+  model_type: det
+  algorithm: FCE
+  Transform:
+  Backbone:
+    name: ResNet_vd
+    layers: 50
+    dcn_stage: [False, True, True, True]
+    out_indices: [1,2,3]
+  Neck:
+    name: FCEFPN
+    out_channels: 256
+    has_extra_convs: False
+    extra_stage: 0
+  Head:
+    name: FCEHead
+    fourier_degree: 5
+Loss:
+  name: FCELoss
+  fourier_degree: 5
+  num_sample: 50
+  
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0001
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: FCEPostProcess
+  scales: [8, 16, 32]
+  alpha: 1.0
+  beta: 1.0
+  fourier_degree: 5
+  box_type: 'poly'
+
+Metric:
+  name: DetFCEMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ctw1500/imgs/
+    label_file_list: 
+      - ./train_data/ctw1500/imgs/training.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+          ignore_orientation: True
+      - DetLabelEncode: # Class handling label
+      - ColorJitter: 
+          brightness: 0.142
+          saturation: 0.5
+          contrast: 0.5
+      - RandomScaling: 
+      - RandomCropFlip:
+          crop_ratio: 0.5
+      - RandomCropPolyInstances:
+          crop_ratio: 0.8
+          min_side_ratio: 0.3
+      - RandomRotatePolyInstances:
+          rotate_ratio: 0.5
+          max_angle: 30
+          pad_with_fixed_color: False
+      - SquareResizePad:
+          target_size: 800
+          pad_ratio: 0.6
+      - IaaAugment:
+          augmenter_args:
+            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
+      - FCENetTargets:
+          fourier_degree: 5
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'p3_maps', 'p4_maps', 'p5_maps'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 6
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ctw1500/imgs/
+    label_file_list:
+      - ./train_data/ctw1500/imgs/test.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+          ignore_orientation: True
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          limit_type: 'min'
+          limit_side_len: 736
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - Pad: 
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/configs/det/det_r50_vd_east.yml b/configs/det/det_r50_vd_east.yml
new file mode 100644
index 0000000..5a488dd
--- /dev/null
+++ b/configs/det/det_r50_vd_east.yml
@@ -0,0 +1,108 @@
+Global:
+  use_gpu: true
+  epoch_num: 10000
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/east_r50_vd/
+  save_epoch_step: 1000
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [4000, 5000]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/ResNet50_vd_pretrained
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: 
+  save_res_path: ./output/det_east/predicts_east.txt
+
+Architecture:
+  model_type: det
+  algorithm: EAST
+  Transform:
+  Backbone:
+    name: ResNet_vd
+    layers: 50
+  Neck:
+    name: EASTFPN
+    model_name: large
+  Head:
+    name: EASTHead
+    model_name: large
+
+Loss:
+  name: EASTLoss
+  
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+  #  name: Cosine
+    learning_rate: 0.001
+  #  warmup_epoch: 0
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: EASTPostProcess
+  score_thresh: 0.8
+  cover_thresh: 0.1
+  nms_thresh: 0.2
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - EASTProcessTrain:
+          image_shape: [512, 512]
+          background_ratio: 0.125
+          min_crop_side_ratio: 0.1
+          min_text_size: 10
+      - KeepKeys:
+          keep_keys: ['image', 'score_map', 'geo_map', 'training_mask'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          limit_side_len: 2400
+          limit_type: max
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/configs/det/det_r50_vd_pse.yml b/configs/det/det_r50_vd_pse.yml
new file mode 100644
index 0000000..408e16d
--- /dev/null
+++ b/configs/det/det_r50_vd_pse.yml
@@ -0,0 +1,134 @@
+Global:
+  use_gpu: true
+  epoch_num: 600
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/det_r50_vd_pse/
+  save_epoch_step: 600
+  # evaluation is run every 125 iterations
+  eval_batch_step: [ 0,125 ]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/ResNet50_vd_ssld_pretrained
+  checkpoints: #./output/det_r50_vd_pse_batch8_ColorJitter/best_accuracy
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_pse/predicts_pse.txt
+
+Architecture:
+  model_type: det
+  algorithm: PSE
+  Transform:
+  Backbone:
+    name: ResNet_vd
+    layers: 50
+  Neck:
+    name: FPN
+    out_channels: 256
+  Head:
+    name: PSEHead
+    hidden_dim: 256
+    out_channels: 7
+
+Loss:
+  name: PSELoss
+  alpha: 0.7
+  ohem_ratio: 3
+  kernel_sample_mask: pred
+  reduction: none
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Step
+    learning_rate: 0.0001
+    step_size: 200
+    gamma: 0.1
+  regularizer:
+    name: 'L2'
+    factor: 0.0005
+
+PostProcess:
+  name: PSEPostProcess
+  thresh: 0
+  box_thresh: 0.85
+  min_area: 16
+  box_type: quad # 'quad' or 'poly'
+  scale: 1
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - ColorJitter:
+          brightness: 0.12549019607843137
+          saturation: 0.5
+      - IaaAugment:
+          augmenter_args:
+            - { 'type': Resize, 'args': { 'size': [ 0.5, 3 ] } }
+            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
+            - { 'type': Affine, 'args': { 'rotate': [ -10, 10 ] } }
+      - MakePseGt:
+          kernel_num: 7
+          min_shrink_ratio: 0.4
+          size: 640
+      - RandomCropImgMask:
+          size: [ 640,640 ]
+          main_key: gt_text
+          crop_keys: [ 'image', 'gt_text', 'gt_kernels', 'mask' ]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [ 0.485, 0.456, 0.406 ]
+          std: [ 0.229, 0.224, 0.225 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'image', 'gt_text', 'gt_kernels', 'mask' ] # the order of the dataloader list
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          limit_side_len: 736
+          limit_type: min
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [ 0.485, 0.456, 0.406 ]
+          std: [ 0.229, 0.224, 0.225 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'image', 'shape', 'polys', 'ignore_tags' ]
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 8
diff --git a/configs/det/det_r50_vd_sast_icdar15.yml b/configs/det/det_r50_vd_sast_icdar15.yml
new file mode 100755
index 0000000..e1bf6fa
--- /dev/null
+++ b/configs/det/det_r50_vd_sast_icdar15.yml
@@ -0,0 +1,109 @@
+Global:
+  use_gpu: true
+  epoch_num: 5000
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/sast_r50_vd_ic15/
+  save_epoch_step: 1000
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [4000, 5000]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/ResNet50_vd_ssld_pretrained
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img:
+  save_res_path: ./output/sast_r50_vd_ic15/predicts_sast.txt
+
+
+Architecture:
+  model_type: det
+  algorithm: SAST
+  Transform:
+  Backbone:
+    name: ResNet_SAST
+    layers: 50
+  Neck:
+    name: SASTFPN
+    with_cab: True
+  Head:
+    name: SASTHead
+
+Loss:
+  name: SASTLoss
+  
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+  #  name: Cosine
+    learning_rate: 0.001
+  #  warmup_epoch: 0
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: SASTPostProcess
+  score_thresh: 0.5
+  sample_pts_num: 2
+  nms_thresh: 0.2
+  expand_scale: 1.0
+  shrink_ratio_of_width: 0.3
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list: [./train_data/icdar2013/train_label_json.txt, ./train_data/icdar2015/train_label_json.txt, ./train_data/icdar17_mlt_latin/train_label_json.txt, ./train_data/coco_text_icdar_4pts/train_label_json.txt]
+    ratio_list: [0.1, 0.45, 0.3, 0.15]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - SASTProcessTrain:
+          image_shape: [512, 512]
+          min_crop_side_ratio: 0.3
+          min_crop_size: 24
+          min_text_size: 4
+          max_text_size: 512
+      - KeepKeys:
+          keep_keys: ['image', 'score_map', 'border_map', 'training_mask', 'tvo_map', 'tco_map'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 4
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          resize_long: 1536
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/configs/det/det_r50_vd_sast_totaltext.yml b/configs/det/det_r50_vd_sast_totaltext.yml
new file mode 100755
index 0000000..557ff8b
--- /dev/null
+++ b/configs/det/det_r50_vd_sast_totaltext.yml
@@ -0,0 +1,108 @@
+Global:
+  use_gpu: true
+  epoch_num: 5000
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/sast_r50_vd_tt/
+  save_epoch_step: 1000
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [4000, 5000]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/ResNet50_vd_ssld_pretrained
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: 
+  save_res_path: ./output/sast_r50_vd_tt/predicts_sast.txt
+
+Architecture:
+  model_type: det
+  algorithm: SAST
+  Transform:
+  Backbone:
+    name: ResNet_SAST
+    layers: 50
+  Neck:
+    name: SASTFPN
+    with_cab: True
+  Head:
+    name: SASTHead
+
+Loss:
+  name: SASTLoss
+  
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+  #  name: Cosine
+    learning_rate: 0.001
+  #  warmup_epoch: 0
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: SASTPostProcess
+  score_thresh: 0.5
+  sample_pts_num: 6
+  nms_thresh: 0.2
+  expand_scale: 1.2
+  shrink_ratio_of_width: 0.2
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list: [./train_data/art_latin_icdar_14pt/train_no_tt_test/train_label_json.txt, ./train_data/total_text_icdar_14pt/train_label_json.txt]
+    ratio_list: [0.5, 0.5]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - SASTProcessTrain:
+          image_shape: [512, 512]
+          min_crop_side_ratio: 0.3
+          min_crop_size: 24
+          min_text_size: 4
+          max_text_size: 512
+      - KeepKeys:
+          keep_keys: ['image', 'score_map', 'border_map', 'training_mask', 'tvo_map', 'tco_map'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 4
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list: 
+        - ./train_data/total_text_icdar_14pt/test_label_json.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          resize_long: 768
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/configs/det/det_repsvtr_db.yml b/configs/det/det_repsvtr_db.yml
new file mode 100644
index 0000000..8c4768e
--- /dev/null
+++ b/configs/det/det_repsvtr_db.yml
@@ -0,0 +1,169 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: &epoch_num 500
+  log_smooth_window: 20
+  print_batch_step: 100
+  save_model_dir: ./output/det_repsvtr_db
+  save_epoch_step: 10
+  eval_batch_step:
+  - 0
+  - 1000
+  cal_metric_during_train: false
+  checkpoints:
+  pretrained_model:
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./checkpoints/det_db/predicts_db.txt
+  distributed: true
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform: null
+  Backbone:
+    name: RepSVTR_det
+  Neck:
+    name: RSEFPN
+    out_channels: 96
+    shortcut: True
+  Head:
+    name: DBHead
+    k: 50
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001 #(8*8c)
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 5.0e-05
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - CopyPaste: null
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            rotate:
+            - -10
+            - 10
+        - type: Resize
+          args:
+            size:
+            - 0.5
+            - 3
+    - EastRandomCropData:
+        size:
+        - 640
+        - 640
+        max_tries: 50
+        keep_ratio: true
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+        total_epoch: *epoch_num
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+        total_epoch: *epoch_num
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 8
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest:
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 2
+profiler_options: null
diff --git a/configs/det/det_res18_db_v2.0.yml b/configs/det/det_res18_db_v2.0.yml
new file mode 100644
index 0000000..e983c22
--- /dev/null
+++ b/configs/det/det_res18_db_v2.0.yml
@@ -0,0 +1,131 @@
+Global:
+  use_gpu: true
+  epoch_num: 1200
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/ch_db_res18/
+  save_epoch_step: 1200
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [3000, 2000]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/ResNet18_vd_pretrained
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_db/predicts_db.txt
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform:
+  Backbone:
+    name: ResNet_vd
+    layers: 18
+    disable_se: True
+  Neck:
+    name: DBFPN
+    out_channels: 256
+  Head:
+    name: DBHead
+    k: 50
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 2
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - IaaAugment:
+          augmenter_args:
+            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
+            - { 'type': Affine, 'args': { 'rotate': [-10, 10] } }
+            - { 'type': Resize, 'args': { 'size': [0.5, 3] } }
+      - EastRandomCropData:
+          size: [960, 960]
+          max_tries: 50
+          keep_ratio: true
+      - MakeBorderMap:
+          shrink_ratio: 0.4
+          thresh_min: 0.3
+          thresh_max: 0.7
+      - MakeShrinkMap:
+          shrink_ratio: 0.4
+          min_text_size: 8
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+#           image_shape: [736, 1280]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/configs/e2e/e2e_r50_vd_pg.yml b/configs/e2e/e2e_r50_vd_pg.yml
new file mode 100644
index 0000000..4642f54
--- /dev/null
+++ b/configs/e2e/e2e_r50_vd_pg.yml
@@ -0,0 +1,121 @@
+Global:
+  use_gpu: True
+  epoch_num: 600
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/pgnet_r50_vd_totaltext/
+  save_epoch_step: 10
+  # evaluation is run every 0 iterationss after the 1000th iteration
+  eval_batch_step: [ 0, 1000 ]
+  cal_metric_during_train: False
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img:
+  infer_visual_type: EN # two mode: EN is for english datasets, CN is for chinese datasets
+  valid_set: totaltext # two mode: totaltext valid curved words, partvgg valid non-curved words
+  save_res_path: ./output/pgnet_r50_vd_totaltext/predicts_pgnet.txt
+  character_dict_path: ppocr/utils/ic15_dict.txt
+  character_type: EN
+  max_text_length: 50 # the max length in seq
+  max_text_nums: 30 # the max seq nums in a pic
+  tcl_len: 64
+
+Architecture:
+  model_type: e2e
+  algorithm: PGNet
+  Transform:
+  Backbone:
+    name: ResNet
+    layers: 50
+  Neck:
+    name: PGFPN
+  Head:
+    name: PGHead
+    character_dict_path: ppocr/utils/ic15_dict.txt # the same as Global:character_dict_path
+
+Loss:
+  name: PGLoss
+  tcl_bs: 64
+  max_text_length: 50 # the same as Global: max_text_length
+  max_text_nums: 30 # the same as Global：max_text_nums
+  pad_num: 36 # the length of dict for pad
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 50
+  regularizer:
+    name: 'L2'
+    factor: 0.0001
+
+PostProcess:
+  name: PGPostProcess
+  score_thresh: 0.5
+  mode: fast   # fast or slow two ways
+  point_gather_mode: align # same as PGProcessTrain: point_gather_mode
+
+Metric:
+  name: E2EMetric
+  mode: A   # two ways for eval, A: label from txt,  B: label from gt_mat
+  gt_mat_dir:  ./train_data/total_text/gt  # the dir of gt_mat
+  character_dict_path: ppocr/utils/ic15_dict.txt
+  main_indicator: f_score_e2e
+
+Train:
+  dataset:
+    name: PGDataSet
+    data_dir: ./train_data/total_text/train
+    label_file_list: [./train_data/total_text/train/train.txt]
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - E2ELabelEncodeTrain:
+      - PGProcessTrain:
+          batch_size: 14  # same as loader: batch_size_per_card
+          use_resize: True
+          use_random_crop: False
+          min_crop_size: 24
+          min_text_size: 4
+          max_text_size: 512
+          point_gather_mode: align # two mode: align and none, align mode is better than none mode
+      - KeepKeys:
+          keep_keys: [ 'images', 'tcl_maps', 'tcl_label_maps', 'border_maps','direction_maps', 'training_masks', 'label_list', 'pos_list', 'pos_mask' ] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: True
+    batch_size_per_card: 14
+    num_workers: 16
+
+Eval:
+  dataset:
+    name: PGDataSet
+    data_dir: ./train_data/total_text/test
+    label_file_list: [./train_data/total_text/test/test.txt]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - E2ELabelEncodeTest:
+      - E2EResizeForTest:
+          max_side_len: 768
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [ 0.485, 0.456, 0.406 ]
+          std: [ 0.229, 0.224, 0.225 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'image', 'shape', 'polys', 'texts', 'ignore_tags', 'img_id']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/configs/kie/layoutlm_series/re_layoutlmv2_xfund_zh.yml b/configs/kie/layoutlm_series/re_layoutlmv2_xfund_zh.yml
new file mode 100644
index 0000000..3e3578d
--- /dev/null
+++ b/configs/kie/layoutlm_series/re_layoutlmv2_xfund_zh.yml
@@ -0,0 +1,123 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 200
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/re_layoutlmv2_xfund_zh
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 19 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: ppstructure/docs/kie/input/zh_val_21.jpg
+  save_res_path: ./output/re_layoutlmv2_xfund_zh/res/
+
+Architecture:
+  model_type: kie
+  algorithm: &algorithm "LayoutLMv2"
+  Transform:
+  Backbone:
+    name: LayoutLMv2ForRe
+    pretrained: True
+    checkpoints:
+
+Loss:
+  name: LossFromOutput
+  key: loss
+  reduction: mean
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  clip_norm: 10
+  lr:
+    learning_rate: 0.00005
+    warmup_epoch: 10
+  regularizer:
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: VQAReTokenLayoutLMPostProcess
+
+Metric:
+  name: VQAReTokenMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_train/image
+    label_file_list: 
+      - train_data/XFUND/zh_train/train.json
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: True
+          algorithm: *algorithm
+          class_path: &class_path train_data/XFUND/class_list_xfun.txt
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQAReTokenRelation:
+      - VQAReTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids','image', 'entities', 'relations'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 8
+    collate_fn: ListCollator
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_val/image
+    label_file_list:
+      - train_data/XFUND/zh_val/val.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: True
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQAReTokenRelation:
+      - VQAReTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image','entities', 'relations'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 8
+    collate_fn: ListCollator
diff --git a/configs/kie/layoutlm_series/re_layoutxlm_xfund_zh.yml b/configs/kie/layoutlm_series/re_layoutxlm_xfund_zh.yml
new file mode 100644
index 0000000..99dc771
--- /dev/null
+++ b/configs/kie/layoutlm_series/re_layoutxlm_xfund_zh.yml
@@ -0,0 +1,123 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 130
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/re_layoutxlm_xfund_zh
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 19 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: ppstructure/docs/kie/input/zh_val_21.jpg
+  save_res_path: ./output/re_layoutxlm_xfund_zh/res/
+
+Architecture:
+  model_type: kie
+  algorithm: &algorithm "LayoutXLM"
+  Transform:
+  Backbone:
+    name: LayoutXLMForRe
+    pretrained: True
+    checkpoints:
+
+Loss:
+  name: LossFromOutput
+  key: loss
+  reduction: mean
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  clip_norm: 10
+  lr:
+    learning_rate: 0.00005
+    warmup_epoch: 10
+  regularizer:
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: VQAReTokenLayoutLMPostProcess
+
+Metric:
+  name: VQAReTokenMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_train/image
+    label_file_list: 
+      - train_data/XFUND/zh_train/train.json
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: True
+          algorithm: *algorithm
+          class_path: &class_path train_data/XFUND/class_list_xfun.txt
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQAReTokenRelation:
+      - VQAReTokenChunk:
+          max_seq_len: *max_seq_len
+      - TensorizeEntitiesRelations:
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox','attention_mask', 'token_type_ids', 'image', 'entities', 'relations'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 2
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_val/image
+    label_file_list:
+      - train_data/XFUND/zh_val/val.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: True
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQAReTokenRelation:
+      - VQAReTokenChunk:
+          max_seq_len: *max_seq_len
+      - TensorizeEntitiesRelations:
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'entities', 'relations'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 8
diff --git a/configs/kie/layoutlm_series/ser_layoutlm_xfund_zh.yml b/configs/kie/layoutlm_series/ser_layoutlm_xfund_zh.yml
new file mode 100644
index 0000000..5424d7e
--- /dev/null
+++ b/configs/kie/layoutlm_series/ser_layoutlm_xfund_zh.yml
@@ -0,0 +1,121 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 200
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/ser_layoutlm_xfund_zh
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 19 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: ppstructure/docs/kie/input/zh_val_42.jpg
+  save_res_path: ./output/re_layoutlm_xfund_zh/res
+
+Architecture:
+  model_type: kie
+  algorithm: &algorithm "LayoutLM"
+  Transform:
+  Backbone:
+    name: LayoutLMForSer
+    pretrained: True
+    checkpoints:
+    num_classes: &num_classes 7
+
+Loss:
+  name: VQASerTokenLayoutLMLoss
+  num_classes: *num_classes
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Linear
+    learning_rate: 0.00005
+    epochs: *epoch_num
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: VQASerTokenLayoutLMPostProcess
+  class_path: &class_path train_data/XFUND/class_list_xfun.txt
+
+Metric:
+  name: VQASerTokenMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_train/image
+    label_file_list: 
+      - train_data/XFUND/zh_train/train.json
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 16
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_val/image
+    label_file_list:
+      - train_data/XFUND/zh_val/val.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
diff --git a/configs/kie/layoutlm_series/ser_layoutlmv2_xfund_zh.yml b/configs/kie/layoutlm_series/ser_layoutlmv2_xfund_zh.yml
new file mode 100644
index 0000000..c5e8335
--- /dev/null
+++ b/configs/kie/layoutlm_series/ser_layoutlmv2_xfund_zh.yml
@@ -0,0 +1,122 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 200
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/ser_layoutlmv2_xfund_zh/
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 19 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: ppstructure/docs/kie/input/zh_val_42.jpg
+  save_res_path: ./output/ser_layoutlmv2_xfund_zh/res/
+
+Architecture:
+  model_type: kie
+  algorithm: &algorithm "LayoutLMv2"
+  Transform:
+  Backbone:
+    name: LayoutLMv2ForSer
+    pretrained: True
+    checkpoints:
+    num_classes: &num_classes 7
+
+Loss:
+  name: VQASerTokenLayoutLMLoss
+  num_classes: *num_classes
+  key: "backbone_out"
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Linear
+    learning_rate: 0.00005
+    epochs: *epoch_num
+    warmup_epoch: 2
+  regularizer:
+  
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: VQASerTokenLayoutLMPostProcess
+  class_path: &class_path train_data/XFUND/class_list_xfun.txt
+
+Metric:
+  name: VQASerTokenMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_train/image
+    label_file_list: 
+      - train_data/XFUND/zh_train/train.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_val/image
+    label_file_list:
+      - train_data/XFUND/zh_val/val.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
diff --git a/configs/kie/layoutlm_series/ser_layoutxlm_xfund_zh.yml b/configs/kie/layoutlm_series/ser_layoutxlm_xfund_zh.yml
new file mode 100644
index 0000000..abcfec2
--- /dev/null
+++ b/configs/kie/layoutlm_series/ser_layoutxlm_xfund_zh.yml
@@ -0,0 +1,122 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 200
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/ser_layoutxlm_xfund_zh
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 19 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: ppstructure/docs/kie/input/zh_val_42.jpg
+  save_res_path: ./output/ser_layoutxlm_xfund_zh/res
+
+Architecture:
+  model_type: kie
+  algorithm: &algorithm "LayoutXLM"
+  Transform:
+  Backbone:
+    name: LayoutXLMForSer
+    pretrained: True
+    checkpoints:
+    num_classes: &num_classes 7
+
+Loss:
+  name: VQASerTokenLayoutLMLoss
+  num_classes: *num_classes
+  key: "backbone_out"
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Linear
+    learning_rate: 0.00005
+    epochs: *epoch_num
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: VQASerTokenLayoutLMPostProcess
+  class_path: &class_path train_data/XFUND/class_list_xfun.txt
+
+Metric:
+  name: VQASerTokenMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_train/image
+    label_file_list: 
+      - train_data/XFUND/zh_train/train.json
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_val/image
+    label_file_list:
+      - train_data/XFUND/zh_val/val.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
diff --git a/configs/kie/sdmgr/kie_unet_sdmgr.yml b/configs/kie/sdmgr/kie_unet_sdmgr.yml
new file mode 100644
index 0000000..da2e4fd
--- /dev/null
+++ b/configs/kie/sdmgr/kie_unet_sdmgr.yml
@@ -0,0 +1,111 @@
+Global:
+  use_gpu: True
+  epoch_num: 60
+  log_smooth_window: 20
+  print_batch_step: 50
+  save_model_dir: ./output/kie_5/
+  save_epoch_step: 50
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [ 0, 80 ]
+  # 1. If pretrained_model is saved in static mode, such as classification pretrained model
+  #    from static branch, load_static_weights must be set as True.
+  # 2. If you want to finetune the pretrained models we provide in the docs,
+  #    you should set load_static_weights as False.
+  load_static_weights: False
+  cal_metric_during_train: False
+  pretrained_model: 
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  class_path: &class_path ./train_data/wildreceipt/class_list.txt
+  infer_img: ./train_data/wildreceipt/1.txt
+  save_res_path: ./output/sdmgr_kie/predicts_kie.txt
+  img_scale: [ 1024, 512 ]
+
+Architecture:
+  model_type: kie
+  algorithm: SDMGR
+  Transform:
+  Backbone:
+    name: Kie_backbone
+  Head:
+    name: SDMGRHead
+
+Loss:
+  name: SDMGRLoss
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Piecewise
+    learning_rate: 0.001
+    decay_epochs: [ 60, 80, 100]
+    values: [ 0.001, 0.0001, 0.00001]
+    warmup_epoch: 2
+  regularizer:
+    name: 'L2'
+    factor: 0.00005
+
+PostProcess:
+  name: None
+
+Metric:
+  name: KIEMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/wildreceipt/
+    label_file_list: [ './train_data/wildreceipt/wildreceipt_train.txt' ]
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - KieLabelEncode: # Class handling label
+          character_dict_path: ./train_data/wildreceipt/dict.txt
+          class_path: *class_path
+      - KieResize:
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'image', 'relations', 'texts', 'points', 'labels', 'tag', 'shape'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 4
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/wildreceipt
+    label_file_list:
+      - ./train_data/wildreceipt/wildreceipt_test.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - KieLabelEncode: # Class handling label
+          character_dict_path: ./train_data/wildreceipt/dict.txt
+      - KieResize:
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'image', 'relations', 'texts', 'points', 'labels', 'tag', 'ori_image', 'ori_boxes', 'shape']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 4
diff --git a/configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml b/configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml
new file mode 100644
index 0000000..e65af0a
--- /dev/null
+++ b/configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml
@@ -0,0 +1,130 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 130
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/re_vi_layoutxlm_xfund_zh
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 19 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: ppstructure/docs/kie/input/zh_val_21.jpg
+  save_res_path: ./output/re/xfund_zh/with_gt
+  kie_rec_model_dir: 
+  kie_det_model_dir:
+
+Architecture:
+  model_type: kie
+  algorithm: &algorithm "LayoutXLM"
+  Transform:
+  Backbone:
+    name: LayoutXLMForRe
+    pretrained: True
+    mode: vi
+    checkpoints:
+
+Loss:
+  name: LossFromOutput
+  key: loss
+  reduction: mean
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  clip_norm: 10
+  lr:
+    learning_rate: 0.00005
+    warmup_epoch: 10
+  regularizer:
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: VQAReTokenLayoutLMPostProcess
+
+Metric:
+  name: VQAReTokenMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_train/image
+    label_file_list: 
+      - train_data/XFUND/zh_train/train.json
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: True
+          algorithm: *algorithm
+          class_path: &class_path train_data/XFUND/class_list_xfun.txt
+          use_textline_bbox_info: &use_textline_bbox_info True
+          order_method: &order_method "tb-yx"
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQAReTokenRelation:
+      - VQAReTokenChunk:
+          max_seq_len: *max_seq_len
+      - TensorizeEntitiesRelations:
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox','attention_mask', 'token_type_ids', 'entities', 'relations'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 2
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_val/image
+    label_file_list:
+      - train_data/XFUND/zh_val/val.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: True
+          algorithm: *algorithm
+          class_path: *class_path
+          use_textline_bbox_info: *use_textline_bbox_info
+          order_method: *order_method
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQAReTokenRelation:
+      - VQAReTokenChunk:
+          max_seq_len: *max_seq_len
+      - TensorizeEntitiesRelations:
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'entities', 'relations'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 8
diff --git a/configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh_udml.yml b/configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh_udml.yml
new file mode 100644
index 0000000..eda3a2b
--- /dev/null
+++ b/configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh_udml.yml
@@ -0,0 +1,175 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 130
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/re_vi_layoutxlm_xfund_zh_udml
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 19 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: ppstructure/docs/kie/input/zh_val_21.jpg
+  save_res_path: ./output/re/xfund_zh/with_gt
+
+Architecture:
+  model_type: &model_type "kie"
+  name: DistillationModel
+  algorithm: Distillation
+  Models:
+    Teacher:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: &algorithm "LayoutXLM"
+      Transform:
+      Backbone:
+        name: LayoutXLMForRe
+        pretrained: True
+        mode: vi
+        checkpoints:
+    Student:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: *algorithm
+      Transform:
+      Backbone:
+        name: LayoutXLMForRe
+        pretrained: True
+        mode: vi
+        checkpoints:
+
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationLossFromOutput:
+      weight: 1.0
+      model_name_list: ["Student", "Teacher"]
+      key: loss
+      reduction: mean
+  - DistillationVQADistanceLoss:
+      weight: 0.5
+      mode: "l2"
+      model_name_pairs:
+        - ["Student", "Teacher"]
+      key: hidden_states
+      index: 5
+      name: "loss_5"
+  - DistillationVQADistanceLoss:
+      weight: 0.5
+      mode: "l2"
+      model_name_pairs:
+        - ["Student", "Teacher"]
+      key: hidden_states
+      index: 8
+      name: "loss_8"
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  clip_norm: 10
+  lr:
+    learning_rate: 0.00005
+    warmup_epoch: 10
+  regularizer:
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: DistillationRePostProcess
+  model_name: ["Student", "Teacher"]
+  key: null
+
+
+Metric:
+  name: DistillationMetric
+  base_metric_name: VQAReTokenMetric
+  main_indicator: hmean
+  key: "Student"
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_train/image
+    label_file_list: 
+      - train_data/XFUND/zh_train/train.json
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: True
+          algorithm: *algorithm
+          class_path: &class_path train_data/XFUND/class_list_xfun.txt
+          use_textline_bbox_info: &use_textline_bbox_info True
+          # [None, "tb-yx"]
+          order_method: &order_method "tb-yx"
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQAReTokenRelation:
+      - VQAReTokenChunk:
+          max_seq_len: *max_seq_len
+      - TensorizeEntitiesRelations:
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox','attention_mask', 'token_type_ids', 'entities', 'relations'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 2
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_val/image
+    label_file_list:
+      - train_data/XFUND/zh_val/val.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: True
+          algorithm: *algorithm
+          class_path: *class_path
+          use_textline_bbox_info: *use_textline_bbox_info
+          order_method: *order_method
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQAReTokenRelation:
+      - VQAReTokenChunk:
+          max_seq_len: *max_seq_len
+      - TensorizeEntitiesRelations:
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'entities', 'relations'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 8
diff --git a/configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml b/configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml
new file mode 100644
index 0000000..50b04ba
--- /dev/null
+++ b/configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml
@@ -0,0 +1,138 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 200
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/ser_vi_layoutxlm_xfund_zh
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 19 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: ppstructure/docs/kie/input/zh_val_42.jpg
+  d2s_train_image_shape: [3, 224, 224]
+  # if you want to predict using the groundtruth ocr info,
+  # you can use the following config
+  # infer_img: train_data/XFUND/zh_val/val.json
+  # infer_mode: False
+
+  save_res_path: ./output/ser/xfund_zh/res
+  kie_rec_model_dir: 
+  kie_det_model_dir:
+  amp_custom_white_list: ['scale', 'concat', 'elementwise_add']
+
+Architecture:
+  model_type: kie
+  algorithm: &algorithm "LayoutXLM"
+  Transform:
+  Backbone:
+    name: LayoutXLMForSer
+    pretrained: True
+    checkpoints:
+    # one of base or vi
+    mode: vi
+    num_classes: &num_classes 7
+
+Loss:
+  name: VQASerTokenLayoutLMLoss
+  num_classes: *num_classes
+  key: "backbone_out"
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Linear
+    learning_rate: 0.00005
+    epochs: *epoch_num
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: VQASerTokenLayoutLMPostProcess
+  class_path: &class_path train_data/XFUND/class_list_xfun.txt
+
+Metric:
+  name: VQASerTokenMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_train/image
+    label_file_list: 
+      - train_data/XFUND/zh_train/train.json
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+          use_textline_bbox_info: &use_textline_bbox_info True
+          # one of [None, "tb-yx"]
+          order_method: &order_method "tb-yx"
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_val/image
+    label_file_list:
+      - train_data/XFUND/zh_val/val.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+          use_textline_bbox_info: *use_textline_bbox_info
+          order_method: *order_method
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
diff --git a/configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh_udml.yml b/configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh_udml.yml
new file mode 100644
index 0000000..5166533
--- /dev/null
+++ b/configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh_udml.yml
@@ -0,0 +1,182 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 200
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/ser_vi_layoutxlm_xfund_zh_udml
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 19 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: ppstructure/docs/kie/input/zh_val_42.jpg
+  save_res_path: ./output/ser_layoutxlm_xfund_zh/res
+
+
+Architecture:
+  model_type: &model_type "kie"
+  name: DistillationModel
+  algorithm: Distillation
+  Models:
+    Teacher:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: &algorithm "LayoutXLM"
+      Transform:
+      Backbone:
+        name: LayoutXLMForSer
+        pretrained: True
+        # one of base or vi
+        mode: vi
+        checkpoints:
+        num_classes: &num_classes 7
+    Student:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: *algorithm
+      Transform:
+      Backbone:
+        name: LayoutXLMForSer
+        pretrained: True
+        # one of base or vi
+        mode: vi
+        checkpoints:
+        num_classes: *num_classes
+
+
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationVQASerTokenLayoutLMLoss:
+      weight: 1.0
+      model_name_list: ["Student", "Teacher"]
+      key: backbone_out
+      num_classes: *num_classes
+  - DistillationSERDMLLoss:
+      weight: 1.0
+      act: "softmax"
+      use_log: true
+      model_name_pairs:
+      - ["Student", "Teacher"]
+      key: backbone_out
+  - DistillationVQADistanceLoss:
+      weight: 0.5
+      mode: "l2"
+      model_name_pairs:
+        - ["Student", "Teacher"]
+      key: hidden_states_5
+      name: "loss_5"
+  - DistillationVQADistanceLoss:
+      weight: 0.5
+      mode: "l2"
+      model_name_pairs:
+        - ["Student", "Teacher"]
+      key: hidden_states_8
+      name: "loss_8"
+  
+  
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Linear
+    learning_rate: 0.00005
+    epochs: *epoch_num
+    warmup_epoch: 10
+  regularizer:
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: DistillationSerPostProcess
+  model_name: ["Student", "Teacher"]
+  key: backbone_out
+  class_path: &class_path train_data/XFUND/class_list_xfun.txt
+
+Metric:
+  name: DistillationMetric
+  base_metric_name: VQASerTokenMetric
+  main_indicator: hmean
+  key: "Student"
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_train/image
+    label_file_list: 
+      - train_data/XFUND/zh_train/train.json
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+          # one of [None, "tb-yx"]
+          order_method: &order_method "tb-yx"
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 4
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_val/image
+    label_file_list:
+      - train_data/XFUND/zh_val/val.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+          order_method: *order_method
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
diff --git a/configs/rec/LaTeX_OCR_rec.yaml b/configs/rec/LaTeX_OCR_rec.yaml
new file mode 100644
index 0000000..514f4e6
--- /dev/null
+++ b/configs/rec/LaTeX_OCR_rec.yaml
@@ -0,0 +1,131 @@
+Global:
+  model_name: LaTeX_OCR_rec # To use static model for inference.
+  use_gpu: True
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 100
+  save_model_dir: ./output/rec/latex_ocr/
+  save_epoch_step: 5
+  max_seq_len: 512
+  # evaluation is run every 60000 iterations (22 epoch)(batch_size = 56)
+  eval_batch_step: [0, 60000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/datasets/pme_demo/0000013.png
+  infer_mode: False
+  use_space_char: False
+  rec_char_dict_path:  ppocr/utils/dict/latex_ocr_tokenizer.json
+  save_res_path: ./output/rec/predicts_latexocr.txt
+  d2s_train_image_shape: [1,256,256]
+  find_unused_parameters: True
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Const
+    learning_rate: 0.0001
+
+Architecture:
+  model_type: rec
+  algorithm: LaTeXOCR
+  in_channels: 1
+  Transform:
+  Backbone:
+    name: HybridTransformer
+    img_size: [192, 672]
+    patch_size: 16
+    num_classes: 0
+    embed_dim: 256
+    depth: 4
+    num_heads: 8
+    input_channel: 1
+    is_predict: False
+    is_export: False
+  Head:
+    name: LaTeXOCRHead
+    pad_value: 0
+    is_export: False
+    decoder_args:
+      attn_on_attn: True
+      cross_attend: True
+      ff_glu: True
+      rel_pos_bias: False
+      use_scalenorm: False
+
+Loss:
+  name: LaTeXOCRLoss
+
+PostProcess:
+  name: LaTeXOCRDecode
+  rec_char_dict_path: ppocr/utils/dict/latex_ocr_tokenizer.json
+
+Metric:
+  name: LaTeXOCRMetric
+  main_indicator:  exp_rate
+  cal_bleu_score: True
+
+Train:
+  dataset:
+    name: LaTeXOCRDataSet
+    data_dir: ./train_data/LaTeXOCR/train
+    data: ./train_data/LaTeXOCR/latexocr_train.pkl
+    min_dimensions: [32, 32]
+    max_dimensions: [672, 192]
+    batch_size_per_pair: 56
+    keep_smaller_batches: False
+    transforms:
+      - DecodeImage:
+          channel_first: False
+      - MinMaxResize:
+          min_dimensions: [32, 32]
+          max_dimensions: [672, 192]        
+      - LatexTrainTransform:
+          bitmap_prob: .04
+      - NormalizeImage:
+          mean: [0.7931, 0.7931, 0.7931]
+          std: [0.1738, 0.1738, 0.1738]
+          order: 'hwc'
+      - LatexImageFormat:
+      - KeepKeys:
+          keep_keys: ['image']
+  loader:
+    shuffle: True
+    batch_size_per_card: 1
+    drop_last: False
+    num_workers: 0
+    collate_fn: LaTeXOCRCollator
+
+Eval:
+  dataset:
+    name: LaTeXOCRDataSet
+    data_dir: ./train_data/LaTeXOCR/val
+    data: ./train_data/LaTeXOCR/latexocr_val.pkl
+    min_dimensions: [32, 32]
+    max_dimensions: [672, 192]
+    batch_size_per_pair: 10
+    keep_smaller_batches: True
+    transforms:
+      - DecodeImage:
+          channel_first: False
+      - MinMaxResize:
+          min_dimensions: [32, 32]
+          max_dimensions: [672, 192]  
+      - LatexTestTransform:
+      - NormalizeImage:
+          mean: [0.7931, 0.7931, 0.7931]
+          std: [0.1738, 0.1738, 0.1738]
+          order: 'hwc'
+      - LatexImageFormat:
+      - KeepKeys:
+          keep_keys: ['image']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1
+    num_workers: 0
+    collate_fn: LaTeXOCRCollator
diff --git a/configs/rec/PP-FormuaNet/PP-FormulaNet-L.yaml b/configs/rec/PP-FormuaNet/PP-FormulaNet-L.yaml
new file mode 100644
index 0000000..6ecbb47
--- /dev/null
+++ b/configs/rec/PP-FormuaNet/PP-FormulaNet-L.yaml
@@ -0,0 +1,119 @@
+Global:
+  model_name: PP-FormulaNet-L # To use static model for inference.
+  use_gpu: True
+  epoch_num: 10
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/rec/pp_formulanet_l/
+  save_epoch_step: 2
+  # evaluation is run every  417  iterations (1 epoch)(batch_size = 24)   # max_seq_len: 1024
+  eval_batch_step: [0,  417 ]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/datasets/pme_demo/0000013.png
+  infer_mode: False
+  use_space_char: False
+  rec_char_dict_path: &rec_char_dict_path ppocr/utils/dict/unimernet_tokenizer
+  max_new_tokens: &max_new_tokens 1024
+  input_size: &input_size [768, 768]
+  save_res_path: ./output/rec/predicts_pp_formulanet_l.txt
+  allow_resize_largeImg: False
+  start_ema: True
+  d2s_train_image_shape: [1,768,768]
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  weight_decay: 0.05
+  lr:
+    name: LinearWarmupCosine
+    learning_rate: 0.0001
+
+Architecture:
+  model_type: rec
+  algorithm: PP-FormulaNet-L
+  in_channels: 3
+  Transform:
+  Backbone:
+    name: Vary_VIT_B_Formula
+    image_size: 768 
+    encoder_embed_dim: 768
+    encoder_depth: 12
+    encoder_num_heads: 12
+    encoder_global_attn_indexes: [2, 5, 8, 11]
+  Head:
+    name: PPFormulaNet_Head
+    max_new_tokens: *max_new_tokens
+    decoder_start_token_id: 0
+    decoder_ffn_dim: 2048
+    decoder_hidden_size: 512
+    decoder_layers: 8
+    temperature: 0.2
+    do_sample: False
+    top_p: 0.95 
+    encoder_hidden_size: 1024
+    is_export: False
+    length_aware: False 
+    use_parallel: False
+    parallel_step: 0
+
+Loss:
+  name: PPFormulaNet_L_Loss
+
+PostProcess:
+  name:  UniMERNetDecode
+  rec_char_dict_path:  *rec_char_dict_path
+
+Metric:
+  name: LaTeXOCRMetric
+  main_indicator:  exp_rate
+  cal_bleu_score: True
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./ocr_rec_latexocr_dataset_example
+    label_file_list: ["./ocr_rec_latexocr_dataset_example/train.txt"]
+    transforms:
+      - UniMERNetImgDecode:
+          input_size: *input_size
+      - UniMERNetTrainTransform: 
+      - LatexImageFormat:
+      - UniMERNetLabelEncode:
+          rec_char_dict_path: *rec_char_dict_path
+          max_seq_len:  *max_new_tokens
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'attention_mask']
+
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 6
+    num_workers: 0
+    collate_fn: UniMERNetCollator
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./ocr_rec_latexocr_dataset_example
+    label_file_list: ["./ocr_rec_latexocr_dataset_example/val.txt"]
+    transforms:
+      - UniMERNetImgDecode:
+          input_size: *input_size
+      - UniMERNetTestTransform:
+      - LatexImageFormat:
+      - UniMERNetLabelEncode:
+          max_seq_len:  *max_new_tokens
+          rec_char_dict_path: *rec_char_dict_path
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'attention_mask', 'filename']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 10
+    num_workers: 0
+    collate_fn: UniMERNetCollator
diff --git a/configs/rec/PP-FormuaNet/PP-FormulaNet-S.yaml b/configs/rec/PP-FormuaNet/PP-FormulaNet-S.yaml
new file mode 100644
index 0000000..3360c7d
--- /dev/null
+++ b/configs/rec/PP-FormuaNet/PP-FormulaNet-S.yaml
@@ -0,0 +1,117 @@
+Global:
+  model_name: PP-FormulaNet-S # To use static model for inference.
+  use_gpu: True
+  epoch_num: 20
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/rec/pp_formulanet_s/
+  save_epoch_step: 2
+  # evaluation is run every 179 iterations (1 epoch)(batch_size = 56)   # max_seq_len: 1024
+  eval_batch_step: [0, 179]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/datasets/pme_demo/0000013.png
+  infer_mode: False
+  use_space_char: False
+  rec_char_dict_path: &rec_char_dict_path  ppocr/utils/dict/unimernet_tokenizer
+  max_new_tokens: &max_new_tokens 1024
+  input_size: &input_size [384, 384]
+  save_res_path: ./output/rec/predicts_pp_formulanet_s.txt
+  allow_resize_largeImg: False
+  start_ema: True
+  d2s_train_image_shape: [1,384,384]
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  weight_decay: 0.05
+  lr:
+    name: LinearWarmupCosine
+    learning_rate: 0.0001
+
+Architecture:
+  model_type: rec
+  algorithm: PP-FormulaNet-S
+  in_channels: 3
+  Transform:
+  Backbone:
+    name: PPHGNetV2_B4_Formula
+    class_num: 1024
+
+  Head:
+    name: PPFormulaNet_Head
+    max_new_tokens:  *max_new_tokens
+    decoder_start_token_id: 0
+    decoder_ffn_dim: 1536
+    decoder_hidden_size: 384
+    decoder_layers: 2
+    temperature: 0.2
+    do_sample: False
+    top_p: 0.95 
+    encoder_hidden_size: 2048
+    is_export: False
+    length_aware: True 
+    use_parallel: True
+    parallel_step: 3
+
+Loss:
+  name: PPFormulaNet_S_Loss
+  parallel_step: 3
+
+PostProcess:
+  name:  UniMERNetDecode
+  rec_char_dict_path: *rec_char_dict_path
+
+Metric:
+  name: LaTeXOCRMetric
+  main_indicator:  exp_rate
+  cal_bleu_score: True
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./ocr_rec_latexocr_dataset_example
+    label_file_list: ["./ocr_rec_latexocr_dataset_example/train.txt"]
+    transforms:
+      - UniMERNetImgDecode:
+          input_size: *input_size
+      - UniMERNetTrainTransform: 
+      - LatexImageFormat:
+      - UniMERNetLabelEncode:
+          rec_char_dict_path: *rec_char_dict_path
+          max_seq_len: *max_new_tokens
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'attention_mask']
+
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 14
+    num_workers: 0
+    collate_fn: UniMERNetCollator
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./ocr_rec_latexocr_dataset_example
+    label_file_list: ["./ocr_rec_latexocr_dataset_example/val.txt"]
+    transforms:
+      - UniMERNetImgDecode:
+          input_size:  *input_size
+      - UniMERNetTestTransform:
+      - LatexImageFormat:
+      - UniMERNetLabelEncode:
+          max_seq_len: *max_new_tokens
+          rec_char_dict_path: *rec_char_dict_path
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'attention_mask', 'filename']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 30
+    num_workers: 0
+    collate_fn: UniMERNetCollator
diff --git a/configs/rec/PP-FormuaNet/PP-FormulaNet_plus-L.yaml b/configs/rec/PP-FormuaNet/PP-FormulaNet_plus-L.yaml
new file mode 100644
index 0000000..eb8d368
--- /dev/null
+++ b/configs/rec/PP-FormuaNet/PP-FormulaNet_plus-L.yaml
@@ -0,0 +1,122 @@
+Global:
+  model_name: PP-FormulaNet_plus-L # To use static model for inference.
+  use_gpu: True
+  epoch_num: 10
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/rec/pp_formulanet_plus_l/
+  save_epoch_step: 2
+  # evaluation is run every  417  iterations (1 epoch)(batch_size = 24)   # max_seq_len: 1024
+  eval_batch_step: [0,  417 ]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/datasets/pme_demo/0000013.png
+  infer_mode: False
+  use_space_char: False
+  rec_char_dict_path: &rec_char_dict_path ppocr/utils/dict/unimernet_tokenizer
+  max_new_tokens: &max_new_tokens 2560
+  input_size: &input_size [768, 768]
+  save_res_path: ./output/rec/predicts_pp_formulanet_plus_l.txt
+  allow_resize_largeImg: False
+  start_ema: True
+  d2s_train_image_shape: [1,768,768]
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  weight_decay: 0.05
+  lr:
+    name: LinearWarmupCosine
+    learning_rate: 0.0001
+
+Architecture:
+  model_type: rec
+  algorithm: PP-FormulaNet_plus-L
+  in_channels: 3
+  Transform:
+  Backbone:
+    name: Vary_VIT_B_Formula
+    image_size: 768 
+    encoder_embed_dim: 768
+    encoder_depth: 12
+    encoder_num_heads: 12
+    encoder_global_attn_indexes: [2, 5, 8, 11]
+  Head:
+    name: PPFormulaNet_Head
+    max_new_tokens: *max_new_tokens
+    decoder_start_token_id: 0
+    decoder_ffn_dim: 2048
+    decoder_hidden_size: 512
+    decoder_layers: 8
+    temperature: 0.2
+    do_sample: False
+    top_p: 0.95 
+    encoder_hidden_size: 1024
+    is_export: False
+    length_aware: False 
+    use_parallel: False
+    parallel_step: 0
+
+Loss:
+  name: PPFormulaNet_L_Loss
+
+PostProcess:
+  name:  UniMERNetDecode
+  rec_char_dict_path:  *rec_char_dict_path
+
+Metric:
+  name: LaTeXOCRMetric
+  main_indicator:  exp_rate
+  cal_bleu_score: True
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./ocr_rec_latexocr_dataset_example
+    label_file_list: ["./ocr_rec_latexocr_dataset_example/train.txt"]
+    transforms:
+      - UniMERNetImgDecode:
+          input_size: *input_size
+          random_padding: True
+          random_resize:  True
+          random_crop: True 
+      - UniMERNetTrainTransform: 
+      - LatexImageFormat:
+      - UniMERNetLabelEncode:
+          rec_char_dict_path: *rec_char_dict_path
+          max_seq_len:  *max_new_tokens
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'attention_mask']
+
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 3
+    num_workers: 0
+    collate_fn: UniMERNetCollator
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./ocr_rec_latexocr_dataset_example
+    label_file_list: ["./ocr_rec_latexocr_dataset_example/val.txt"]
+    transforms:
+      - UniMERNetImgDecode:
+          input_size: *input_size
+      - UniMERNetTestTransform:
+      - LatexImageFormat:
+      - UniMERNetLabelEncode:
+          max_seq_len:  *max_new_tokens
+          rec_char_dict_path: *rec_char_dict_path
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'attention_mask', 'filename']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 10
+    num_workers: 0
+    collate_fn: UniMERNetCollator
diff --git a/configs/rec/PP-FormuaNet/PP-FormulaNet_plus-M.yaml b/configs/rec/PP-FormuaNet/PP-FormulaNet_plus-M.yaml
new file mode 100644
index 0000000..fbaf356
--- /dev/null
+++ b/configs/rec/PP-FormuaNet/PP-FormulaNet_plus-M.yaml
@@ -0,0 +1,119 @@
+Global:
+  model_name: PP-FormulaNet_plus-M # To use static model for inference.
+  use_gpu: True
+  epoch_num: 20
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/rec/pp_formulanet_plus_m/
+  save_epoch_step: 2
+  # evaluation is run every 179 iterations (1 epoch)(batch_size = 56)   # max_seq_len: 1024
+  eval_batch_step: [0, 179]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/datasets/pme_demo/0000013.png
+  infer_mode: False
+  use_space_char: False
+  rec_char_dict_path: &rec_char_dict_path  ppocr/utils/dict/unimernet_tokenizer
+  max_new_tokens: &max_new_tokens 2560
+  input_size: &input_size [384, 384]
+  save_res_path: ./output/rec/predicts_pp_formulanet_plus_m.txt
+  allow_resize_largeImg: False
+  start_ema: True
+  d2s_train_image_shape: [1,384,384]
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  weight_decay: 0.05
+  lr:
+    name: LinearWarmupCosine
+    learning_rate: 0.0001
+
+Architecture:
+  model_type: rec
+  algorithm: PP-FormulaNet_plus-M
+  in_channels: 3
+  Transform:
+  Backbone:
+    name: PPHGNetV2_B6_Formula
+    class_num: 1024
+
+  Head:
+    name: PPFormulaNet_Head
+    max_new_tokens: *max_new_tokens
+    decoder_start_token_id: 0
+    decoder_ffn_dim: 2048
+    decoder_hidden_size: 512
+    decoder_layers: 6
+    temperature: 0.2
+    do_sample: False
+    top_p: 0.95 
+    encoder_hidden_size: 2048
+    is_export: False
+    length_aware: False 
+    use_parallel: False
+    parallel_step: 0
+
+Loss:
+  name: PPFormulaNet_L_Loss
+
+PostProcess:
+  name:  UniMERNetDecode
+  rec_char_dict_path: *rec_char_dict_path
+
+Metric:
+  name: LaTeXOCRMetric
+  main_indicator:  exp_rate
+  cal_bleu_score: True
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./ocr_rec_latexocr_dataset_example
+    label_file_list: ["./ocr_rec_latexocr_dataset_example/train.txt"]
+    transforms:
+      - UniMERNetImgDecode:
+          input_size: *input_size
+          random_padding: True
+          random_resize:  True
+          random_crop: True 
+      - UniMERNetTrainTransform: 
+      - LatexImageFormat:
+      - UniMERNetLabelEncode:
+          rec_char_dict_path: *rec_char_dict_path
+          max_seq_len: *max_new_tokens
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'attention_mask']
+
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 14
+    num_workers: 0
+    collate_fn: UniMERNetCollator
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./ocr_rec_latexocr_dataset_example
+    label_file_list: ["./ocr_rec_latexocr_dataset_example/val.txt"]
+    transforms:
+      - UniMERNetImgDecode:
+          input_size:  *input_size
+      - UniMERNetTestTransform:
+      - LatexImageFormat:
+      - UniMERNetLabelEncode:
+          max_seq_len: *max_new_tokens
+          rec_char_dict_path: *rec_char_dict_path
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'attention_mask', 'filename']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 30
+    num_workers: 0
+    collate_fn: UniMERNetCollator
diff --git a/configs/rec/PP-FormuaNet/PP-FormulaNet_plus-S.yaml b/configs/rec/PP-FormuaNet/PP-FormulaNet_plus-S.yaml
new file mode 100644
index 0000000..0b24316
--- /dev/null
+++ b/configs/rec/PP-FormuaNet/PP-FormulaNet_plus-S.yaml
@@ -0,0 +1,120 @@
+Global:
+  model_name: PP-FormulaNet_plus-S # To use static model for inference.
+  use_gpu: True
+  epoch_num: 20
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/rec/pp_formulanet_plus_s/
+  save_epoch_step: 2
+  # evaluation is run every 179 iterations (1 epoch)(batch_size = 56)   # max_seq_len: 1024
+  eval_batch_step: [0, 179]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/datasets/pme_demo/0000013.png
+  infer_mode: False
+  use_space_char: False
+  rec_char_dict_path: &rec_char_dict_path  ppocr/utils/dict/unimernet_tokenizer
+  max_new_tokens: &max_new_tokens 1024
+  input_size: &input_size [384, 384]
+  save_res_path: ./output/rec/predicts_pp_formulanet_plus_s.txt
+  allow_resize_largeImg: False
+  start_ema: True
+  d2s_train_image_shape: [1,384,384]
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  weight_decay: 0.05
+  lr:
+    name: LinearWarmupCosine
+    learning_rate: 0.0001
+
+Architecture:
+  model_type: rec
+  algorithm: PP-FormulaNet_plus-S
+  in_channels: 3
+  Transform:
+  Backbone:
+    name: PPHGNetV2_B4_Formula
+    class_num: 1024
+
+  Head:
+    name: PPFormulaNet_Head
+    max_new_tokens:  *max_new_tokens
+    decoder_start_token_id: 0
+    decoder_ffn_dim: 1536
+    decoder_hidden_size: 384
+    decoder_layers: 2
+    temperature: 0.2
+    do_sample: False
+    top_p: 0.95 
+    encoder_hidden_size: 2048
+    is_export: False
+    length_aware: True 
+    use_parallel: True,
+    parallel_step: 3
+
+Loss:
+  name: PPFormulaNet_S_Loss
+  parallel_step: 3
+
+PostProcess:
+  name:  UniMERNetDecode
+  rec_char_dict_path: *rec_char_dict_path
+
+Metric:
+  name: LaTeXOCRMetric
+  main_indicator:  exp_rate
+  cal_bleu_score: True
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./ocr_rec_latexocr_dataset_example
+    label_file_list: ["./ocr_rec_latexocr_dataset_example/train.txt"]
+    transforms:
+      - UniMERNetImgDecode:
+          input_size: *input_size
+          random_padding: True
+          random_resize:  True
+          random_crop: True 
+      - UniMERNetTrainTransform: 
+      - LatexImageFormat:
+      - UniMERNetLabelEncode:
+          rec_char_dict_path: *rec_char_dict_path
+          max_seq_len: *max_new_tokens
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'attention_mask']
+
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 14
+    num_workers: 0
+    collate_fn: UniMERNetCollator
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./ocr_rec_latexocr_dataset_example
+    label_file_list: ["./ocr_rec_latexocr_dataset_example/val.txt"]
+    transforms:
+      - UniMERNetImgDecode:
+          input_size:  *input_size
+      - UniMERNetTestTransform:
+      - LatexImageFormat:
+      - UniMERNetLabelEncode:
+          max_seq_len: *max_new_tokens
+          rec_char_dict_path: *rec_char_dict_path
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'attention_mask', 'filename']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 30
+    num_workers: 0
+    collate_fn: UniMERNetCollator
diff --git a/configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec.yml b/configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec.yml
new file mode 100644
index 0000000..a239288
--- /dev/null
+++ b/configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec.yml
@@ -0,0 +1,134 @@
+Global:
+  model_name: PP-OCRv3_mobile_rec  # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_ppocr_v3
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3.txt
+  d2s_train_image_shape: [3,48,320]
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+    last_conv_stride: [1, 2]
+    last_pool_type: avg
+    last_pool_kernel_size: [2, 2]
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 64
+            depth: 2
+            hidden_dims: 120
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - SARHead:
+          enc_dim: 512
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - SARLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_workers: 4
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec_distillation.yml b/configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec_distillation.yml
new file mode 100644
index 0000000..3b82ef8
--- /dev/null
+++ b/configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec_distillation.yml
@@ -0,0 +1,209 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 800
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_ppocr_v3_distillation
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3_distillation.txt
+  d2s_train_image_shape: [3, 48, -1]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Piecewise
+    decay_epochs : [700]
+    values : [0.0005, 0.00005]
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: &model_type "rec"
+  name: DistillationModel
+  algorithm: Distillation
+  Models:
+    Teacher:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: SVTR_LCNet
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+        last_conv_stride: [1, 2]
+        last_pool_type: avg
+        last_pool_kernel_size: [2, 2]
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 64
+                depth: 2
+                hidden_dims: 120
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - SARHead:
+              enc_dim: 512
+              max_text_length: *max_text_length
+    Student:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: SVTR_LCNet
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+        last_conv_stride: [1, 2]
+        last_pool_type: avg
+        last_pool_kernel_size: [2, 2]
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 64
+                depth: 2
+                hidden_dims: 120
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - SARHead:
+              enc_dim: 512
+              max_text_length: *max_text_length
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationDMLLoss:
+      weight: 1.0
+      act: "softmax"
+      use_log: true
+      model_name_pairs:
+      - ["Student", "Teacher"]
+      key: head_out
+      multi_head: True
+      dis_head: ctc
+      name: dml_ctc
+  - DistillationDMLLoss:
+      weight: 0.5
+      act: "softmax"
+      use_log: true
+      model_name_pairs:
+      - ["Student", "Teacher"]
+      key: head_out
+      multi_head: True
+      dis_head: sar
+      name: dml_sar
+  - DistillationDistanceLoss:
+      weight: 1.0
+      mode: "l2"
+      model_name_pairs:
+      - ["Student", "Teacher"]
+      key: backbone_out
+  - DistillationCTCLoss:
+      weight: 1.0
+      model_name_list: ["Student", "Teacher"]
+      key: head_out
+      multi_head: True
+  - DistillationSARLoss:
+      weight: 1.0
+      model_name_list: ["Student", "Teacher"]
+      key: head_out
+      multi_head: True
+
+PostProcess:
+  name: DistillationCTCLabelDecode
+  model_name: ["Student", "Teacher"]
+  key: head_out
+  multi_head: True
+
+Metric:
+  name: DistillationMetric
+  base_metric_name: RecMetric
+  main_indicator: acc
+  key: "Student"
+  ignore_space: False
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_workers: 4
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml b/configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml
new file mode 100644
index 0000000..7c21860
--- /dev/null
+++ b/configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml
@@ -0,0 +1,134 @@
+Global:
+  model_name: en_PP-OCRv3_mobile_rec  # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/v3_en_mobile
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/en_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3_en.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+    last_conv_stride: [1, 2]
+    last_pool_type: avg
+    last_pool_kernel_size: [2, 2]
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 64
+            depth: 2
+            hidden_dims: 120
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - SARHead:
+          enc_dim: 512
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - SARLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_workers: 4
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv3/multi_language/.gitkeep b/configs/rec/PP-OCRv3/multi_language/.gitkeep
new file mode 100644
index 0000000..e69de29
diff --git a/configs/rec/PP-OCRv3/multi_language/arabic_PP-OCRv3_mobile_rec.yml b/configs/rec/PP-OCRv3/multi_language/arabic_PP-OCRv3_mobile_rec.yml
new file mode 100644
index 0000000..5c82700
--- /dev/null
+++ b/configs/rec/PP-OCRv3/multi_language/arabic_PP-OCRv3_mobile_rec.yml
@@ -0,0 +1,133 @@
+Global:
+  model_name: arabic_PP-OCRv3_mobile_rec  # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/v3_arabic_mobile
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: ./doc/imgs_words/arabic/ar_2.jpg
+  character_dict_path: ppocr/utils/dict/arabic_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3_arabic.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+    last_conv_stride: [1, 2]
+    last_pool_type: avg
+    last_pool_kernel_size: [2, 2]
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 64
+            depth: 2
+            hidden_dims: 120
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - SARHead:
+          enc_dim: 512
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - SARLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+    - RecAug:
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_workers: 4
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv3/multi_language/chinese_cht_PP-OCRv3_mobile_rec.yaml b/configs/rec/PP-OCRv3/multi_language/chinese_cht_PP-OCRv3_mobile_rec.yaml
new file mode 100644
index 0000000..2ba7be6
--- /dev/null
+++ b/configs/rec/PP-OCRv3/multi_language/chinese_cht_PP-OCRv3_mobile_rec.yaml
@@ -0,0 +1,133 @@
+Global:
+  model_name: chinese_cht_PP-OCRv3_mobile_rec  # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/v3_chinese_cht_mobile
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/dict/chinese_cht_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3_chinese_cht.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+    last_conv_stride: [1, 2]
+    last_pool_type: avg
+    last_pool_kernel_size: [2, 2]
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 64
+            depth: 2
+            hidden_dims: 120
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - SARHead:
+          enc_dim: 512
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - SARLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+    - RecAug:
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_workers: 4
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv3/multi_language/cyrillic_PP-OCRv3_mobile_rec.yml b/configs/rec/PP-OCRv3/multi_language/cyrillic_PP-OCRv3_mobile_rec.yml
new file mode 100644
index 0000000..0991717
--- /dev/null
+++ b/configs/rec/PP-OCRv3/multi_language/cyrillic_PP-OCRv3_mobile_rec.yml
@@ -0,0 +1,133 @@
+Global:
+  model_name: cyrillic_PP-OCRv3_mobile_rec  # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/v3_cyrillic_mobile
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/dict/cyrillic_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3_cyrillic.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+    last_conv_stride: [1, 2]
+    last_pool_type: avg
+    last_pool_kernel_size: [2, 2]
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 64
+            depth: 2
+            hidden_dims: 120
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - SARHead:
+          enc_dim: 512
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - SARLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+    - RecAug:
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_workers: 4
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv3/multi_language/devanagari_PP-OCRv3_mobile_rec.yml b/configs/rec/PP-OCRv3/multi_language/devanagari_PP-OCRv3_mobile_rec.yml
new file mode 100644
index 0000000..8328757
--- /dev/null
+++ b/configs/rec/PP-OCRv3/multi_language/devanagari_PP-OCRv3_mobile_rec.yml
@@ -0,0 +1,133 @@
+Global:
+  model_name: devanagari_PP-OCRv3_mobile_rec  # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/v3_devanagari_mobile
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/dict/devanagari_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3_devanagari.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+    last_conv_stride: [1, 2]
+    last_pool_type: avg
+    last_pool_kernel_size: [2, 2]
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 64
+            depth: 2
+            hidden_dims: 120
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - SARHead:
+          enc_dim: 512
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - SARLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+    - RecAug:
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_workers: 4
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv3/multi_language/japan_PP-OCRv3_mobile_rec.yml b/configs/rec/PP-OCRv3/multi_language/japan_PP-OCRv3_mobile_rec.yml
new file mode 100644
index 0000000..98b7faa
--- /dev/null
+++ b/configs/rec/PP-OCRv3/multi_language/japan_PP-OCRv3_mobile_rec.yml
@@ -0,0 +1,133 @@
+Global:
+  model_name: japan_PP-OCRv3_mobile_rec  # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/v3_japan_mobile
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/dict/japan_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3_japan.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+    last_conv_stride: [1, 2]
+    last_pool_type: avg
+    last_pool_kernel_size: [2, 2]
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 64
+            depth: 2
+            hidden_dims: 120
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - SARHead:
+          enc_dim: 512
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - SARLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+    - RecAug:
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_workers: 4
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv3/multi_language/ka_PP-OCRv3_mobile_rec.yml b/configs/rec/PP-OCRv3/multi_language/ka_PP-OCRv3_mobile_rec.yml
new file mode 100644
index 0000000..47905e2
--- /dev/null
+++ b/configs/rec/PP-OCRv3/multi_language/ka_PP-OCRv3_mobile_rec.yml
@@ -0,0 +1,133 @@
+Global:
+  model_name: ka_PP-OCRv3_mobile_rec  # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/v3_ka_mobile
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/dict/ka_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3_ka.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+    last_conv_stride: [1, 2]
+    last_pool_type: avg
+    last_pool_kernel_size: [2, 2]
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 64
+            depth: 2
+            hidden_dims: 120
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - SARHead:
+          enc_dim: 512
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - SARLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+    - RecAug:
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_workers: 4
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv3/multi_language/korean_PP-OCRv3_mobile_rec.yml b/configs/rec/PP-OCRv3/multi_language/korean_PP-OCRv3_mobile_rec.yml
new file mode 100644
index 0000000..08f3501
--- /dev/null
+++ b/configs/rec/PP-OCRv3/multi_language/korean_PP-OCRv3_mobile_rec.yml
@@ -0,0 +1,133 @@
+Global:
+  model_name: korean_PP-OCRv3_mobile_rec  # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/v3_korean_mobile
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/dict/korean_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3_korean.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+    last_conv_stride: [1, 2]
+    last_pool_type: avg
+    last_pool_kernel_size: [2, 2]
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 64
+            depth: 2
+            hidden_dims: 120
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - SARHead:
+          enc_dim: 512
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - SARLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+    - RecAug:
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_workers: 4
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv3/multi_language/latin_PP-OCRv3_mobile_rec.yml b/configs/rec/PP-OCRv3/multi_language/latin_PP-OCRv3_mobile_rec.yml
new file mode 100644
index 0000000..5c2987a
--- /dev/null
+++ b/configs/rec/PP-OCRv3/multi_language/latin_PP-OCRv3_mobile_rec.yml
@@ -0,0 +1,133 @@
+Global:
+  model_name: latin_PP-OCRv3_mobile_rec  # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/v3_latin_mobile
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/dict/latin_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3_latin.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+    last_conv_stride: [1, 2]
+    last_pool_type: avg
+    last_pool_kernel_size: [2, 2]
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 64
+            depth: 2
+            hidden_dims: 120
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - SARHead:
+          enc_dim: 512
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - SARLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+    - RecAug:
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_workers: 4
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv3/multi_language/ta_PP-OCRv3_mobile_rec.yml b/configs/rec/PP-OCRv3/multi_language/ta_PP-OCRv3_mobile_rec.yml
new file mode 100644
index 0000000..3551eb0
--- /dev/null
+++ b/configs/rec/PP-OCRv3/multi_language/ta_PP-OCRv3_mobile_rec.yml
@@ -0,0 +1,133 @@
+Global:
+  model_name: ta_PP-OCRv3_mobile_rec  # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/v3_ta_mobile
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/dict/ta_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3_ta.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+    last_conv_stride: [1, 2]
+    last_pool_type: avg
+    last_pool_kernel_size: [2, 2]
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 64
+            depth: 2
+            hidden_dims: 120
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - SARHead:
+          enc_dim: 512
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - SARLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+    - RecAug:
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_workers: 4
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv3/multi_language/te_PP-OCRv3_mobile_rec.yml b/configs/rec/PP-OCRv3/multi_language/te_PP-OCRv3_mobile_rec.yml
new file mode 100644
index 0000000..fb7fe5e
--- /dev/null
+++ b/configs/rec/PP-OCRv3/multi_language/te_PP-OCRv3_mobile_rec.yml
@@ -0,0 +1,133 @@
+Global:
+  model_name: te_PP-OCRv3_mobile_rec  # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/v3_te_mobile
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/dict/te_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3_te.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+    last_conv_stride: [1, 2]
+    last_pool_type: avg
+    last_pool_kernel_size: [2, 2]
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 64
+            depth: 2
+            hidden_dims: 120
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - SARHead:
+          enc_dim: 512
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - SARLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+    - RecAug:
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_workers: 4
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec.yml b/configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec.yml
new file mode 100644
index 0000000..bd0369d
--- /dev/null
+++ b/configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec.yml
@@ -0,0 +1,140 @@
+Global:
+  model_name: PP-OCRv4_mobile_rec  # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 200
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_ppocr_v4
+  save_epoch_step: 10
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.95
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 192
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec_ampO2_ultra.yml b/configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec_ampO2_ultra.yml
new file mode 100644
index 0000000..475c551
--- /dev/null
+++ b/configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec_ampO2_ultra.yml
@@ -0,0 +1,140 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 200
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_ppocr_v4
+  save_epoch_step: 10
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3.txt
+  use_amp: True
+  amp_level: O2
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.95
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 384
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 16
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 16
diff --git a/configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec_distillation.yml b/configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec_distillation.yml
new file mode 100644
index 0000000..aa2a36d
--- /dev/null
+++ b/configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec_distillation.yml
@@ -0,0 +1,231 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 200
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_dkd_400w_svtr_ctc_lcnet_blank_dkd0.1/
+  save_epoch_step: 40
+  eval_batch_step:
+  - 0
+  - 2000
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints: ./output/rec_dkd_400w_svtr_ctc_lcnet_blank_dkd0.1/latest
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3.txt
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+Architecture:
+  model_type: rec
+  name: DistillationModel
+  algorithm: Distillation
+  Models:
+    Teacher:
+      pretrained: 
+      freeze_params: true
+      return_all_feats: true
+      model_type: rec
+      algorithm: SVTR
+      Transform: null
+      Backbone:
+        name: SVTRNet
+        img_size:
+        - 48
+        - 320
+        out_char_num: 40
+        out_channels: 192
+        patch_merging: Conv
+        embed_dim:
+        - 64
+        - 128
+        - 256
+        depth:
+        - 3
+        - 6
+        - 3
+        num_heads:
+        - 2
+        - 4
+        - 8
+        mixer:
+        - Conv
+        - Conv
+        - Conv
+        - Conv
+        - Conv
+        - Conv
+        - Global
+        - Global
+        - Global
+        - Global
+        - Global
+        - Global
+        local_mixer:
+        - - 5
+          - 5
+        - - 5
+          - 5
+        - - 5
+          - 5
+        last_stage: false
+        prenorm: true
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 120
+                depth: 2
+                hidden_dims: 120
+                kernel_size: [1, 3]
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - NRTRHead:
+              nrtr_dim: 384
+              max_text_length: *max_text_length
+    Student:
+      pretrained: 
+      freeze_params: false
+      return_all_feats: true
+      model_type: rec
+      algorithm: SVTR
+      Transform: null
+      Backbone:
+        name: PPLCNetV3
+        scale: 0.95
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 120
+                depth: 2
+                hidden_dims: 120
+                kernel_size: [1, 3]
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - NRTRHead:
+              nrtr_dim: 384
+              max_text_length: *max_text_length
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationDKDLoss:
+      weight: 0.1
+      model_name_pairs:
+      - - Student
+        - Teacher
+      key: head_out
+      multi_head: true
+      alpha: 1.0
+      beta: 2.0
+      dis_head: gtc
+      name: dkd
+  - DistillationCTCLoss:
+      weight: 1.0
+      model_name_list:
+      - Student
+      key: head_out
+      multi_head: true
+  - DistillationNRTRLoss:
+      weight: 1.0
+      smoothing: false
+      model_name_list:
+      - Student
+      key: head_out
+      multi_head: true
+  - DistillCTCLogits:
+      weight: 1.0
+      reduction: mean
+      model_name_pairs:
+      - - Student
+        - Teacher
+      key: head_out
+PostProcess:
+  name: DistillationCTCLabelDecode
+  model_name:
+  - Student
+  key: head_out
+  multi_head: true
+Metric:
+  name: DistillationMetric
+  base_metric_name: RecMetric
+  main_indicator: acc
+  key: Student
+  ignore_space: false
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/train_list.txt
+    ratio_list:
+    - 1.0
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_workers: 8
+    use_shared_memory: true
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
+profiler_options: null
diff --git a/configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec_fp32_ultra.yml b/configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec_fp32_ultra.yml
new file mode 100644
index 0000000..8c26730
--- /dev/null
+++ b/configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec_fp32_ultra.yml
@@ -0,0 +1,138 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 200
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_ppocr_v4
+  save_epoch_step: 10
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.95
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 192
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 16
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 16
diff --git a/configs/rec/PP-OCRv4/PP-OCRv4_server_rec.yml b/configs/rec/PP-OCRv4/PP-OCRv4_server_rec.yml
new file mode 100644
index 0000000..6ec4960
--- /dev/null
+++ b/configs/rec/PP-OCRv4/PP-OCRv4_server_rec.yml
@@ -0,0 +1,138 @@
+Global:
+  model_name: PP-OCRv4_server_rec # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 200
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_ppocr_v4_hgnet
+  save_epoch_step: 10
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_HGNet
+  Transform:
+  Backbone:
+    name: PPHGNet_small
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 128
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv4/PP-OCRv4_server_rec_ampO2_ultra.yml b/configs/rec/PP-OCRv4/PP-OCRv4_server_rec_ampO2_ultra.yml
new file mode 100644
index 0000000..4303521
--- /dev/null
+++ b/configs/rec/PP-OCRv4/PP-OCRv4_server_rec_ampO2_ultra.yml
@@ -0,0 +1,139 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 200
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_ppocr_v4_hgnet
+  save_epoch_step: 10
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3.txt
+  use_amp: True
+  amp_level: O2
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_HGNet
+  Transform:
+  Backbone:
+    name: PPHGNet_small
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 256
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 16
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 16
diff --git a/configs/rec/PP-OCRv4/PP-OCRv4_server_rec_doc.yml b/configs/rec/PP-OCRv4/PP-OCRv4_server_rec_doc.yml
new file mode 100644
index 0000000..46bf0fb
--- /dev/null
+++ b/configs/rec/PP-OCRv4/PP-OCRv4_server_rec_doc.yml
@@ -0,0 +1,138 @@
+Global:
+  model_name: PP-OCRv4_server_rec_doc # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 200
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_ppocr_v4_hgnet
+  save_epoch_step: 10
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/dict/ppocrv4_doc_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_HGNet
+  Transform:
+  Backbone:
+    name: PPHGNet_small
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 128
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv4/PP-OCRv4_server_rec_fp32_ultra.yml b/configs/rec/PP-OCRv4/PP-OCRv4_server_rec_fp32_ultra.yml
new file mode 100644
index 0000000..ee9ebca
--- /dev/null
+++ b/configs/rec/PP-OCRv4/PP-OCRv4_server_rec_fp32_ultra.yml
@@ -0,0 +1,137 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 200
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_ppocr_v4_hgnet
+  save_epoch_step: 10
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_HGNet
+  Transform:
+  Backbone:
+    name: PPHGNet_small
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 256
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 16
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 16
diff --git a/configs/rec/PP-OCRv4/ch_PP-OCRv4_rec_svtr_large.yml b/configs/rec/PP-OCRv4/ch_PP-OCRv4_rec_svtr_large.yml
new file mode 100644
index 0000000..525d1c0
--- /dev/null
+++ b/configs/rec/PP-OCRv4/ch_PP-OCRv4_rec_svtr_large.yml
@@ -0,0 +1,144 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 200
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/svtr_large/
+  save_epoch_step: 10
+  # evaluation is run every 2000 iterations after the 0th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: &max_text_length 40
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_svtr_large.txt
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.99
+  epsilon: 1.0e-08
+  weight_decay: 0.05
+  no_weight_decay_name: norm pos_embed char_node_embed pos_node_embed char_pos_embed vis_pos_embed
+  one_dim_param_no_weight_decay: true
+  lr:
+    name: Cosine
+    learning_rate: 0.00025 # 8gpus 64bs
+    warmup_epoch: 5
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform: null    
+  Backbone:
+    name: SVTRNet
+    img_size:
+    - 48
+    - 320
+    out_char_num: 40
+    out_channels: 512
+    patch_merging: Conv
+    embed_dim: [192, 256, 512]
+    depth: [6, 6, 9]
+    num_heads: [6, 8, 16]
+    mixer: ['Conv','Conv','Conv','Conv','Conv','Conv','Conv','Conv','Conv','Global','Global','Global','Global','Global','Global','Global','Global','Global','Global','Global','Global']
+    local_mixer: [[5, 5], [5, 5], [5, 5]]
+    last_stage: False
+    prenorm: True
+  Head:
+    name: MultiHead
+    use_pool: true
+    use_pos: true
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 256
+            depth: 2
+            hidden_dims: 256
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 512
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: true
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    batch_size_per_card: 64
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - SVTRRecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv4/en_PP-OCRv4_mobile_rec.yml b/configs/rec/PP-OCRv4/en_PP-OCRv4_mobile_rec.yml
new file mode 100644
index 0000000..ddbb982
--- /dev/null
+++ b/configs/rec/PP-OCRv4/en_PP-OCRv4_mobile_rec.yml
@@ -0,0 +1,151 @@
+Global:
+  model_name: en_PP-OCRv4_mobile_rec  # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 50
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_ppocr_v4
+  save_epoch_step: 10
+  eval_batch_step:
+  - 0
+  - 2000
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/en_dict.txt
+  max_text_length: 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3.txt
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform: null
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.95
+  Head:
+    name: MultiHead
+    head_list:
+    - CTCHead:
+        Neck:
+          name: svtr
+          dims: 120
+          depth: 2
+          hidden_dims: 120
+          kernel_size:
+          - 1
+          - 3
+          use_guide: true
+        Head:
+          fc_decay: 1.0e-05
+    - NRTRHead:
+        nrtr_dim: 384
+        max_text_length: 25
+Loss:
+  name: MultiLoss
+  loss_config_list:
+  - CTCLoss: null
+  - NRTRLoss: null
+PostProcess:
+  name: CTCLabelDecode
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: false
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape:
+        - 48
+        - 320
+        - 3
+        max_text_length: 25
+    - RecAug: null
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales:
+    - - 320
+      - 32
+    - - 320
+      - 48
+    - - 320
+      - 64
+    first_bs: 96
+    fix_bs: false
+    divided_factor:
+    - 8
+    - 16
+    is_training: true
+  loader:
+    shuffle: true
+    batch_size_per_card: 96
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape:
+        - 3
+        - 48
+        - 320
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
+profiler_options: null
diff --git a/configs/rec/PP-OCRv5/PP-OCRv5_mobile_rec.yml b/configs/rec/PP-OCRv5/PP-OCRv5_mobile_rec.yml
new file mode 100644
index 0000000..ad882db
--- /dev/null
+++ b/configs/rec/PP-OCRv5/PP-OCRv5_mobile_rec.yml
@@ -0,0 +1,140 @@
+Global:
+  model_name: PP-OCRv5_mobile_rec # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 75
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/PP-OCRv5_mobile_rec
+  save_epoch_step: 10
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ./ppocr/utils/dict/ppocrv5_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv5.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.95
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 128
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv5/PP-OCRv5_server_rec.yml b/configs/rec/PP-OCRv5/PP-OCRv5_server_rec.yml
new file mode 100644
index 0000000..abbda00
--- /dev/null
+++ b/configs/rec/PP-OCRv5/PP-OCRv5_server_rec.yml
@@ -0,0 +1,136 @@
+Global:
+  model_name: PP-OCRv5_server_rec # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 75
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/PP-OCRv5_server_rec
+  save_epoch_step: 1
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  calc_epoch_interval: 1
+  pretrained_model: 
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ./ppocr/utils/dict/ppocrv5_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv5.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 1
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_HGNet
+  Transform:
+  Backbone:
+    name: PPHGNetV2_B4
+    text_rec: True
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 128
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 16
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv5/multi_language/arabic_PP-OCRv5_mobile_rec.yaml b/configs/rec/PP-OCRv5/multi_language/arabic_PP-OCRv5_mobile_rec.yaml
new file mode 100644
index 0000000..c087ce5
--- /dev/null
+++ b/configs/rec/PP-OCRv5/multi_language/arabic_PP-OCRv5_mobile_rec.yaml
@@ -0,0 +1,141 @@
+Global:
+  model_name: arabic_PP-OCRv5_mobile_rec # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 75
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/arabic_rec_ppocr_v5
+  save_epoch_step: 10
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img:
+  character_dict_path: ./ppocr/utils/dict/ppocrv5_arabic_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_arabic_ppocrv5.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.95
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 128
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv5/multi_language/cyrillic_PP-OCRv5_mobile_rec.yaml b/configs/rec/PP-OCRv5/multi_language/cyrillic_PP-OCRv5_mobile_rec.yaml
new file mode 100644
index 0000000..0d887ea
--- /dev/null
+++ b/configs/rec/PP-OCRv5/multi_language/cyrillic_PP-OCRv5_mobile_rec.yaml
@@ -0,0 +1,141 @@
+Global:
+  model_name: cyrillic_PP-OCRv5_mobile_rec # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 75
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/cyrillic_rec_ppocr_v5
+  save_epoch_step: 10
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img:
+  character_dict_path: ./ppocr/utils/dict/ppocrv5_cyrillic_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_cyrillic_ppocrv5.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.95
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 128
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv5/multi_language/devanagari_PP-OCRv5_mobile_rec.yaml b/configs/rec/PP-OCRv5/multi_language/devanagari_PP-OCRv5_mobile_rec.yaml
new file mode 100644
index 0000000..dd85e51
--- /dev/null
+++ b/configs/rec/PP-OCRv5/multi_language/devanagari_PP-OCRv5_mobile_rec.yaml
@@ -0,0 +1,141 @@
+Global:
+  model_name: devanagari_PP-OCRv5_mobile_rec # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 75
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/devanagari_rec_ppocr_v5
+  save_epoch_step: 10
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img:
+  character_dict_path: ./ppocr/utils/dict/ppocrv5_devanagari_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_devanagari_ppocrv5.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.95
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 128
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv5/multi_language/el_PP-OCRv5_mobile_rec.yaml b/configs/rec/PP-OCRv5/multi_language/el_PP-OCRv5_mobile_rec.yaml
new file mode 100644
index 0000000..2c34e04
--- /dev/null
+++ b/configs/rec/PP-OCRv5/multi_language/el_PP-OCRv5_mobile_rec.yaml
@@ -0,0 +1,141 @@
+Global:
+  model_name: el_PP-OCRv5_mobile_rec # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 75
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/el_rec_ppocr_v5
+  save_epoch_step: 10
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img:
+  character_dict_path: ./ppocr/utils/dict/ppocrv5_el_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_el_ppocrv5.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.95
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 128
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv5/multi_language/en_PP-OCRv5_mobile_rec.yaml b/configs/rec/PP-OCRv5/multi_language/en_PP-OCRv5_mobile_rec.yaml
new file mode 100644
index 0000000..fba7afa
--- /dev/null
+++ b/configs/rec/PP-OCRv5/multi_language/en_PP-OCRv5_mobile_rec.yaml
@@ -0,0 +1,141 @@
+Global:
+  model_name: en_PP-OCRv5_mobile_rec # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 75
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/en_rec_ppocr_v5
+  save_epoch_step: 10
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img:
+  character_dict_path: ./ppocr/utils/dict/ppocrv5_en_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_en_ppocrv5.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.95
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 128
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv5/multi_language/eslav_PP-OCRv5_mobile_rec.yml b/configs/rec/PP-OCRv5/multi_language/eslav_PP-OCRv5_mobile_rec.yml
new file mode 100644
index 0000000..777b312
--- /dev/null
+++ b/configs/rec/PP-OCRv5/multi_language/eslav_PP-OCRv5_mobile_rec.yml
@@ -0,0 +1,141 @@
+Global:
+  model_name: eslav_PP-OCRv5_mobile_rec # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 75
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/eslav_rec_ppocr_v5
+  save_epoch_step: 10
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img:
+  character_dict_path: ./ppocr/utils/dict/ppocrv5_eslav_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv5.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.95
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 128
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv5/multi_language/korean_PP-OCRv5_mobile_rec.yml b/configs/rec/PP-OCRv5/multi_language/korean_PP-OCRv5_mobile_rec.yml
new file mode 100644
index 0000000..01244c4
--- /dev/null
+++ b/configs/rec/PP-OCRv5/multi_language/korean_PP-OCRv5_mobile_rec.yml
@@ -0,0 +1,141 @@
+Global:
+  model_name: korean_PP-OCRv5_mobile_rec # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 75
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/korean_rec_ppocr_v5
+  save_epoch_step: 10
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img:
+  character_dict_path: ./ppocr/utils/dict/ppocrv5_korean_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv5.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.95
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 128
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv5/multi_language/latin_PP-OCRv5_mobile_rec.yml b/configs/rec/PP-OCRv5/multi_language/latin_PP-OCRv5_mobile_rec.yml
new file mode 100644
index 0000000..6e5d4b9
--- /dev/null
+++ b/configs/rec/PP-OCRv5/multi_language/latin_PP-OCRv5_mobile_rec.yml
@@ -0,0 +1,143 @@
+Global:
+  model_name: latin_PP-OCRv5_mobile_rec # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 75
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/latin_rec_ppocr_v5
+  save_epoch_step: 10
+  eval_batch_step: [0, 500]
+  cal_metric_during_train: true
+  pretrained_model: PP-OCRv5_mobile_rec_pretrained.pdparams
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img:
+  character_dict_path: ./ppocr/utils/dict/ppocrv5_latin_dict.txt
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv5.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.95
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: 25
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: 25
+    - RecAug:
+    - MultiLabelEncode:
+        max_text_length: 25
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 128
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/eval_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        max_text_length: 1000
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        eval_mode: True
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv5/multi_language/ta_PP-OCRv5_mobile_rec.yaml b/configs/rec/PP-OCRv5/multi_language/ta_PP-OCRv5_mobile_rec.yaml
new file mode 100644
index 0000000..17f17e7
--- /dev/null
+++ b/configs/rec/PP-OCRv5/multi_language/ta_PP-OCRv5_mobile_rec.yaml
@@ -0,0 +1,141 @@
+Global:
+  model_name: ta_PP-OCRv5_mobile_rec # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 75
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/ta_rec_ppocr_v5
+  save_epoch_step: 10
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img:
+  character_dict_path: ./ppocr/utils/dict/ppocrv5_ta_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ta_ppocrv5.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.95
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 128
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv5/multi_language/te_PP-OCRv5_mobile_rec.yaml b/configs/rec/PP-OCRv5/multi_language/te_PP-OCRv5_mobile_rec.yaml
new file mode 100644
index 0000000..75c0fc9
--- /dev/null
+++ b/configs/rec/PP-OCRv5/multi_language/te_PP-OCRv5_mobile_rec.yaml
@@ -0,0 +1,141 @@
+Global:
+  model_name: te_PP-OCRv5_mobile_rec # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 75
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/te_rec_ppocr_v5
+  save_epoch_step: 10
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img:
+  character_dict_path: ./ppocr/utils/dict/ppocrv5_te_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_te_ppocrv5.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.95
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 128
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv5/multi_language/th_PP-OCRv5_mobile_rec.yaml b/configs/rec/PP-OCRv5/multi_language/th_PP-OCRv5_mobile_rec.yaml
new file mode 100644
index 0000000..305b4ad
--- /dev/null
+++ b/configs/rec/PP-OCRv5/multi_language/th_PP-OCRv5_mobile_rec.yaml
@@ -0,0 +1,141 @@
+Global:
+  model_name: th_PP-OCRv5_mobile_rec # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 75
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/th_rec_ppocr_v5
+  save_epoch_step: 10
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img:
+  character_dict_path: ./ppocr/utils/dict/ppocrv5_th_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_th_ppocrv5.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.95
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 128
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv6/PP-OCRv6_medium_rec.yml b/configs/rec/PP-OCRv6/PP-OCRv6_medium_rec.yml
new file mode 100644
index 0000000..d8149ed
--- /dev/null
+++ b/configs/rec/PP-OCRv6/PP-OCRv6_medium_rec.yml
@@ -0,0 +1,143 @@
+Global:
+  model_name: PP-OCRv6_medium_rec
+  debug: false
+  use_gpu: true
+  epoch_num: 100
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/PP-OCRv6_medium_rec
+  save_epoch_step: 10
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/dict/ppocrv6_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv6_medium.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV4
+    model_size: medium
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: lightsvtr
+            dims: 192
+            depth: 2
+            mlp_ratio: 4.0
+            local_kernel: 7
+            use_guide: false
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 512
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    ratio_list:
+    - 0.5
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 64
+    fix_bs: false
+    divided_factor: [8, 16]
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv6/PP-OCRv6_small_rec.yml b/configs/rec/PP-OCRv6/PP-OCRv6_small_rec.yml
new file mode 100644
index 0000000..cf2b417
--- /dev/null
+++ b/configs/rec/PP-OCRv6/PP-OCRv6_small_rec.yml
@@ -0,0 +1,140 @@
+Global:
+  model_name: PP-OCRv6_small_rec
+  debug: false
+  use_gpu: true
+  epoch_num: 100
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/PP-OCRv6_small_rec
+  save_epoch_step: 10
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/dict/ppocrv6_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv6_small.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV4
+    model_size: small
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: lightsvtr
+            dims: 120
+            depth: 2
+            mlp_ratio: 2.0
+            local_kernel: 7
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 128
+    fix_bs: false
+    divided_factor: [8, 16]
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/PP-OCRv6/PP-OCRv6_tiny_rec.yml b/configs/rec/PP-OCRv6/PP-OCRv6_tiny_rec.yml
new file mode 100644
index 0000000..c5486b8
--- /dev/null
+++ b/configs/rec/PP-OCRv6/PP-OCRv6_tiny_rec.yml
@@ -0,0 +1,138 @@
+Global:
+  model_name: PP-OCRv6_tiny_rec
+  debug: false
+  use_gpu: true
+  epoch_num: 100
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/PP-OCRv6_tiny_rec
+  save_epoch_step: 10
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/dict/ppocrv6_tiny_dict.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv6_tiny.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform:
+  Backbone:
+    name: PPLCNetV4
+    model_size: tiny
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: reshape
+          Head:
+            mid_channels: 80
+            use_guide: true
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 128
+    fix_bs: false
+    divided_factor: [8, 16]
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 8
diff --git a/configs/rec/SVTRv2/ch_RepSVTR_rec.yml b/configs/rec/SVTRv2/ch_RepSVTR_rec.yml
new file mode 100644
index 0000000..5c31916
--- /dev/null
+++ b/configs/rec/SVTRv2/ch_RepSVTR_rec.yml
@@ -0,0 +1,136 @@
+Global:
+  model_name: ch_RepSVTR_rec  # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 200
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/ch_RepSVTR_rec
+  save_epoch_step: 10
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: False
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_repsvtr.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1.e-8
+  weight_decay: 0.025
+  no_weight_decay_name: norm
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 0.001 # 8gpus 192bs
+    warmup_epoch: 5
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_HGNet
+  Transform:
+  Backbone:
+    name: RepSVTR
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 256
+            depth: 2
+            hidden_dims: 256
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+          num_decoder_layers: 2
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 192
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/SVTRv2/ch_RepSVTR_rec_gtc.yml b/configs/rec/SVTRv2/ch_RepSVTR_rec_gtc.yml
new file mode 100644
index 0000000..28e3611
--- /dev/null
+++ b/configs/rec/SVTRv2/ch_RepSVTR_rec_gtc.yml
@@ -0,0 +1,134 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 200
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/ch_RepSVTR_rec_gtc
+  save_epoch_step: 10
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: False
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_repsvtr.txt
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1.e-8
+  weight_decay: 0.025
+  no_weight_decay_name: norm
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 0.001 # 8gpus 192bs
+    warmup_epoch: 5
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_HGNet
+  Transform:
+  Backbone:
+    name: RepSVTR
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 256
+            depth: 2
+            hidden_dims: 256
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+          num_decoder_layers: 2
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 192
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/SVTRv2/ch_SVTRv2_rec.yml b/configs/rec/SVTRv2/ch_SVTRv2_rec.yml
new file mode 100644
index 0000000..7b3fe63
--- /dev/null
+++ b/configs/rec/SVTRv2/ch_SVTRv2_rec.yml
@@ -0,0 +1,144 @@
+Global:
+  model_name: ch_SVTRv2_rec  # To use static model for inference.
+  debug: false
+  use_gpu: true
+  epoch_num: 200
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/ch_SVTRv2_rec
+  save_epoch_step: 10
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: False
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_svrtv2.txt
+  d2s_train_image_shape: [3, 48, 320]
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1.e-8
+  weight_decay: 0.05
+  no_weight_decay_name: norm
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 0.001 # 8gpus 192bs
+    warmup_epoch: 5
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_HGNet
+  Transform:
+  Backbone:
+    name: SVTRv2
+    use_pos_embed: False
+    dims: [128, 256, 384]
+    depths: [6, 6, 6]
+    num_heads: [4, 8, 12]
+    mixer: [['Conv','Conv','Conv','Conv','Conv','Conv'],['Conv','Conv','Global','Global','Global','Global'],['Global','Global','Global','Global','Global','Global']]
+    local_k: [[5, 5], [5, 5], [-1, -1]]
+    sub_k: [[2, 1], [2, 1], [-1, -1]]
+    last_stage: False
+    use_pool: True
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 256
+            depth: 2
+            hidden_dims: 256
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+          num_decoder_layers: 2
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 192
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/SVTRv2/ch_SVTRv2_rec_distillation.yml b/configs/rec/SVTRv2/ch_SVTRv2_rec_distillation.yml
new file mode 100644
index 0000000..4c80148
--- /dev/null
+++ b/configs/rec/SVTRv2/ch_SVTRv2_rec_distillation.yml
@@ -0,0 +1,208 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 100
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/ch_SVTRv2_rec_distill_lr00002/
+  save_epoch_step: 5
+  eval_batch_step:
+  - 0
+  - 1000
+  cal_metric_during_train: False
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_svtrv2_ch_distill.txt
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.99
+  epsilon: 1.e-8
+  weight_decay: 0.05
+  no_weight_decay_name: norm pos_embed patch_embed downsample
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 0.0002 # 8gpus 192bs
+    warmup_epoch: 5
+Architecture:
+  model_type: rec
+  name: DistillationModel
+  algorithm: Distillation
+  Models:
+    Teacher:
+      pretrained: ./output/ch_SVTRv2_rec/best_accuracy
+      freeze_params: true
+      return_all_feats: true
+      model_type: rec
+      algorithm: SVTR_LCNet
+      Transform: null
+      Backbone:
+        name: SVTRv2
+        use_pos_embed: False
+        dims: [128, 256, 384]
+        depths: [6, 6, 6]
+        num_heads: [4, 8, 12]
+        mixer: [['Conv','Conv','Conv','Conv','Conv','Conv'],['Conv','Conv','Global','Global','Global','Global'],['Global','Global','Global','Global','Global','Global']]
+        local_k: [[5, 5], [5, 5], [-1, -1]]
+        sub_k: [[2, 1], [2, 1], [-1, -1]]
+        last_stage: False
+        use_pool: True
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 256
+                depth: 2
+                hidden_dims: 256
+                kernel_size: [1, 3]
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - NRTRHead:
+              nrtr_dim: 384
+              num_decoder_layers: 2
+              max_text_length: *max_text_length
+    Student:
+      pretrained: ./output/ch_RepSVTR_rec/best_accuracy
+      freeze_params: false
+      return_all_feats: true
+      model_type: rec
+      algorithm: SVTR_LCNet
+      Transform: null
+      Backbone:
+        name: RepSVTR
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 256
+                depth: 2
+                hidden_dims: 256
+                kernel_size: [1, 3]
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - NRTRHead:
+              nrtr_dim: 384
+              num_decoder_layers: 2
+              max_text_length: *max_text_length
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationDKDLoss:
+      weight: 0.1
+      model_name_pairs:
+      - - Student
+        - Teacher
+      key: head_out
+      multi_head: true
+      alpha: 1.0
+      beta: 2.0
+      dis_head: gtc
+      name: dkd
+  - DistillationCTCLoss:
+      weight: 1.0
+      model_name_list:
+      - Student
+      key: head_out
+      multi_head: true
+  - DistillationNRTRLoss:
+      weight: 1.0
+      smoothing: false
+      model_name_list:
+      - Student
+      key: head_out
+      multi_head: true
+  - DistillCTCLogits:
+      weight: 1.0
+      reduction: mean
+      model_name_pairs:
+      - - Student
+        - Teacher
+      key: head_out
+PostProcess:
+  name: DistillationCTCLabelDecode
+  model_name:
+  - Student
+  key: head_out
+  multi_head: true
+Metric:
+  name: DistillationMetric
+  base_metric_name: RecMetric
+  main_indicator: acc
+  key: Student
+
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 192
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/SVTRv2/ch_SVTRv2_rec_gtc.yml b/configs/rec/SVTRv2/ch_SVTRv2_rec_gtc.yml
new file mode 100644
index 0000000..bf7587f
--- /dev/null
+++ b/configs/rec/SVTRv2/ch_SVTRv2_rec_gtc.yml
@@ -0,0 +1,145 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 200
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/ch_SVTRv2_rec_gtc
+  save_epoch_step: 10
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: False
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_svrtv2.txt
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  epsilon: 1.e-8
+  weight_decay: 0.05
+  no_weight_decay_name: norm
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 0.001 # 8gpus 192bs
+    warmup_epoch: 5
+
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_HGNet
+  Transform:
+  Backbone:
+    name: SVTRv2
+    use_pos_embed: False
+    dims: [128, 256, 384]
+    depths: [6, 6, 6]
+    num_heads: [4, 8, 12]
+    mixer: [['Conv','Conv','Conv','Conv','Conv','Conv'],['Conv','Conv','Global','Global','Global','Global'],['Global','Global','Global','Global','Global','Global']]
+    local_k: [[5, 5], [5, 5], [-1, -1]]
+    sub_k: [[2, 1], [2, 1], [-1, -1]]
+    last_stage: False
+    use_pool: True
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 256
+            depth: 2
+            hidden_dims: 256
+            kernel_size: [1, 3]
+            use_guide: True
+          Head:
+            fc_decay: 0.00001
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: *max_text_length
+          num_decoder_layers: 2
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:  
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 192
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/SVTRv2/ch_SVTRv2_rec_gtc_distill.yml b/configs/rec/SVTRv2/ch_SVTRv2_rec_gtc_distill.yml
new file mode 100644
index 0000000..c588c3a
--- /dev/null
+++ b/configs/rec/SVTRv2/ch_SVTRv2_rec_gtc_distill.yml
@@ -0,0 +1,208 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 100
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/ch_SVTRv2_rec_gtc_distill_lr00002/
+  save_epoch_step: 5
+  eval_batch_step:
+  - 0
+  - 1000
+  cal_metric_during_train: False
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_svtrv2_gtc_distill.txt
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.99
+  epsilon: 1.e-8
+  weight_decay: 0.05
+  no_weight_decay_name: norm pos_embed patch_embed downsample
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 0.0002 # 8gpus 192bs
+    warmup_epoch: 5
+Architecture:
+  model_type: rec
+  name: DistillationModel
+  algorithm: Distillation
+  Models:
+    Teacher:
+      pretrained: ./output/ch_SVTRv2_rec_gtc/best_accuracy
+      freeze_params: true
+      return_all_feats: true
+      model_type: rec
+      algorithm: SVTR_LCNet
+      Transform: null
+      Backbone:
+        name: SVTRv2
+        use_pos_embed: False
+        dims: [128, 256, 384]
+        depths: [6, 6, 6]
+        num_heads: [4, 8, 12]
+        mixer: [['Conv','Conv','Conv','Conv','Conv','Conv'],['Conv','Conv','Global','Global','Global','Global'],['Global','Global','Global','Global','Global','Global']]
+        local_k: [[5, 5], [5, 5], [-1, -1]]
+        sub_k: [[2, 1], [2, 1], [-1, -1]]
+        last_stage: False
+        use_pool: True
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 256
+                depth: 2
+                hidden_dims: 256
+                kernel_size: [1, 3]
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - NRTRHead:
+              nrtr_dim: 384
+              num_decoder_layers: 2
+              max_text_length: *max_text_length
+    Student:
+      pretrained: ./output/ch_RepSVTR_rec_gtc/best_accuracy
+      freeze_params: false
+      return_all_feats: true
+      model_type: rec
+      algorithm: SVTR_LCNet
+      Transform: null
+      Backbone:
+        name: repvit_svtr
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 256
+                depth: 2
+                hidden_dims: 256
+                kernel_size: [1, 3]
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - NRTRHead:
+              nrtr_dim: 384
+              num_decoder_layers: 2
+              max_text_length: *max_text_length
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationDKDLoss:
+      weight: 0.1
+      model_name_pairs:
+      - - Student
+        - Teacher
+      key: head_out
+      multi_head: true
+      alpha: 1.0
+      beta: 2.0
+      dis_head: gtc
+      name: dkd
+  - DistillationCTCLoss:
+      weight: 1.0
+      model_name_list:
+      - Student
+      key: head_out
+      multi_head: true
+  - DistillationNRTRLoss:
+      weight: 1.0
+      smoothing: false
+      model_name_list:
+      - Student
+      key: head_out
+      multi_head: true
+  - DistillCTCLogits:
+      weight: 1.0
+      reduction: mean
+      model_name_pairs:
+      - - Student
+        - Teacher
+      key: head_out
+PostProcess:
+  name: DistillationCTCLabelDecode
+  model_name:
+  - Student
+  key: head_out
+  multi_head: true
+Metric:
+  name: DistillationMetric
+  base_metric_name: RecMetric
+  main_indicator: acc
+  key: Student
+
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: ./train_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 192
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/configs/rec/UniMERNet.yaml b/configs/rec/UniMERNet.yaml
new file mode 100644
index 0000000..b8befc8
--- /dev/null
+++ b/configs/rec/UniMERNet.yaml
@@ -0,0 +1,115 @@
+Global:
+  model_name: UniMERNet # To use static model for inference.
+  use_gpu: True
+  epoch_num: 40
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/rec/unimernet/
+  save_epoch_step: 5
+  # evaluation is run every 37880 iterations after the 0th iteration
+  eval_batch_step: [0, 37880]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/datasets/pme_demo/0000013.png
+  infer_mode: False
+  use_space_char: False
+  rec_char_dict_path: &rec_char_dict_path ppocr/utils/dict/unimernet_tokenizer
+  input_size: &input_size [192, 672]
+  max_seq_len: &max_seq_len 1024
+  save_res_path: ./output/rec/predicts_unimernet.txt
+  allow_resize_largeImg: False
+  d2s_train_image_shape: [1,192,672]
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  weight_decay: 0.05
+  lr:
+    name: LinearWarmupCosine
+    learning_rate: 1e-4
+    start_lr: 1e-5
+    min_lr: 1e-8
+    warmup_steps: 5000
+
+Architecture:
+  model_type: rec
+  algorithm: UniMERNet
+  in_channels: 3
+  Transform:
+  Backbone:
+    name: DonutSwinModel
+    hidden_size : 1024
+    num_layers: 4
+    num_heads: [4, 8, 16, 32]
+    add_pooling_layer: True
+    use_mask_token: False
+  Head:
+    name: UniMERNetHead
+    max_new_tokens: 1536
+    decoder_start_token_id: 0
+    temperature: 0.2
+    do_sample: False
+    top_p: 0.95 
+    encoder_hidden_size: 1024
+    is_export: False
+    length_aware: True 
+
+Loss:
+  name: UniMERNetLoss
+
+PostProcess:
+  name:  UniMERNetDecode
+  rec_char_dict_path: *rec_char_dict_path
+
+Metric:
+  name: LaTeXOCRMetric
+  main_indicator:  exp_rate
+  cal_bleu_score: True
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/UniMERNet/
+    label_file_list: ["./train_data/UniMERNet/train_unimernet_1M.txt"]
+    transforms:
+      - UniMERNetImgDecode:
+          input_size: *input_size
+      - UniMERNetTrainTransform: 
+      - UniMERNetImageFormat:
+      - UniMERNetLabelEncode:
+          rec_char_dict_path: *rec_char_dict_path
+          max_seq_len: *max_seq_len
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'attention_mask']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 7
+    num_workers: 0
+    collate_fn: UniMERNetCollator
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/UniMERNet/UniMER-Test/cpe
+    label_file_list: ["./train_data/UniMERNet/test_unimernet_cpe.txt"]
+    transforms:
+      - UniMERNetImgDecode:
+          input_size: *input_size
+      - UniMERNetTestTransform:
+      - UniMERNetImageFormat:
+      - UniMERNetLabelEncode:
+          max_seq_len: *max_seq_len
+          rec_char_dict_path: *rec_char_dict_path
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'attention_mask']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 30
+    num_workers: 0
+    collate_fn: UniMERNetCollator
diff --git a/configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml b/configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml
new file mode 100644
index 0000000..793bb5c
--- /dev/null
+++ b/configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml
@@ -0,0 +1,110 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 800
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_mobile_pp-OCRv2
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_mobile_pp-OCRv2.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Piecewise
+    decay_epochs : [700]
+    values : [0.001, 0.0001]
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 2.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 64
+  Head:
+    name: CTCHead
+    mid_channels: 96
+    fc_decay: 0.00002
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug:
+    - CTCLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 32, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - CTCLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 32, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 8
diff --git a/configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec_distillation.yml b/configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec_distillation.yml
new file mode 100644
index 0000000..3855005
--- /dev/null
+++ b/configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec_distillation.yml
@@ -0,0 +1,160 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 800
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_pp-OCRv2_distillation
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_pp-OCRv2_distillation.txt
+  amp_custom_black_list: ['matmul','matmul_v2','elementwise_add']
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Piecewise
+    decay_epochs : [700]
+    values : [0.001, 0.0001]
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 2.0e-05
+
+Architecture:
+  model_type: &model_type "rec"
+  name: DistillationModel
+  algorithm: Distillation
+  Models:
+    Teacher:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: CRNN
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+      Neck:
+        name: SequenceEncoder
+        encoder_type: rnn
+        hidden_size: 64
+      Head:
+        name: CTCHead
+        mid_channels: 96
+        fc_decay: 0.00002
+    Student:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: CRNN
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+      Neck:
+        name: SequenceEncoder
+        encoder_type: rnn
+        hidden_size: 64
+      Head:
+        name: CTCHead
+        mid_channels: 96
+        fc_decay: 0.00002
+  
+
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationCTCLoss:
+      weight: 1.0
+      model_name_list: ["Student", "Teacher"]
+      key: head_out
+  - DistillationDMLLoss:
+      weight: 1.0
+      act: "softmax"
+      use_log: true
+      model_name_pairs:
+      - ["Student", "Teacher"]
+      key: head_out
+  - DistillationDistanceLoss:
+      weight: 1.0
+      mode: "l2"
+      model_name_pairs:
+      - ["Student", "Teacher"]
+      key: backbone_out
+
+PostProcess:
+  name: DistillationCTCLabelDecode
+  model_name: ["Student", "Teacher"]
+  key: head_out
+
+Metric:
+  name: DistillationMetric
+  base_metric_name: RecMetric
+  main_indicator: acc
+  key: "Student"
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug:
+    - CTCLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 32, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_sections: 1
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - CTCLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 32, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 8
diff --git a/configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec_enhanced_ctc_loss.yml b/configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec_enhanced_ctc_loss.yml
new file mode 100644
index 0000000..ef0e893
--- /dev/null
+++ b/configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec_enhanced_ctc_loss.yml
@@ -0,0 +1,124 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 800
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_mobile_pp-OCRv2_enhanced_ctc_loss
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_mobile_pp-OCRv2_enhanced_ctc_loss.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Piecewise
+    decay_epochs : [700]
+    values : [0.001, 0.0001]
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 2.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 64
+  Head:
+    name: CTCHead
+    mid_channels: 96
+    fc_decay: 0.00002
+    return_feats: true
+
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - CTCLoss:
+      use_focal_loss: false
+      weight: 1.0
+  - CenterLoss:
+      weight: 0.05
+      num_classes: 6625
+      feat_dim: 96
+      center_file_path:
+  # you can also try to add ace loss on your own dataset
+  # - ACELoss:
+  #     weight: 0.1
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug:
+    - CTCLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 32, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+        - label_ace
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - CTCLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 32, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 8
diff --git a/configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0.yml b/configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0.yml
new file mode 100644
index 0000000..3e427b6
--- /dev/null
+++ b/configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0.yml
@@ -0,0 +1,100 @@
+Global:
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_chinese_common_v2.0
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  # for data or label process
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: True
+  save_res_path: ./output/rec/predicts_chinese_common_v2.0.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: 'L2'
+    factor: 0.00004
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: ResNet
+    layers: 34
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 256
+  Head:
+    name: CTCHead
+    fc_decay: 0.00004
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list: ["./train_data/train_list.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - RecAug: 
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list: ["./train_data/val_list.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml b/configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml
new file mode 100644
index 0000000..abd5cd9
--- /dev/null
+++ b/configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml
@@ -0,0 +1,102 @@
+Global:
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_chinese_lite_v2.0
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  # for data or label process
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: True
+  save_res_path: ./output/rec/predicts_chinese_lite_v2.0.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 5
+  regularizer:
+    name: 'L2'
+    factor: 0.00001
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: small
+    small_stride: [1, 2, 2, 2]
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 48
+  Head:
+    name: CTCHead
+    fc_decay: 0.00001
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list: ["./train_data/train_list.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - RecAug: 
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list: ["./train_data/val_list.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/multi_language/generate_multi_language_configs.py b/configs/rec/multi_language/generate_multi_language_configs.py
new file mode 100644
index 0000000..1c34045
--- /dev/null
+++ b/configs/rec/multi_language/generate_multi_language_configs.py
@@ -0,0 +1,307 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import yaml
+from argparse import ArgumentParser, RawDescriptionHelpFormatter
+import os.path
+import logging
+
+logger = logging.getLogger("ppocr")
+
+support_list = {
+    "it": "italian",
+    "xi": "spanish",
+    "pu": "portuguese",
+    "ru": "russian",
+    "ar": "arabic",
+    "ta": "tamil",
+    "ug": "uyghur",
+    "fa": "persian",
+    "ur": "urdu",
+    "rs": "serbian latin",
+    "oc": "occitan",
+    "rsc": "serbian cyrillic",
+    "bg": "bulgarian",
+    "uk": "ukranian",
+    "be": "belarusian",
+    "te": "telugu",
+    "ka": "kannada",
+    "chinese_cht": "chinese tradition",
+    "hi": "hindi",
+    "mr": "marathi",
+    "ne": "nepali",
+}
+
+latin_lang = [
+    "af",
+    "az",
+    "bs",
+    "cs",
+    "cy",
+    "da",
+    "de",
+    "es",
+    "et",
+    "fr",
+    "ga",
+    "hr",
+    "hu",
+    "id",
+    "is",
+    "it",
+    "ku",
+    "la",
+    "lt",
+    "lv",
+    "mi",
+    "ms",
+    "mt",
+    "nl",
+    "no",
+    "oc",
+    "pi",
+    "pl",
+    "pt",
+    "ro",
+    "rs_latin",
+    "sk",
+    "sl",
+    "sq",
+    "sv",
+    "sw",
+    "tl",
+    "tr",
+    "uz",
+    "vi",
+    "latin",
+]
+arabic_lang = ["ar", "fa", "ug", "ur"]
+cyrillic_lang = [
+    "ru",
+    "rs_cyrillic",
+    "be",
+    "bg",
+    "uk",
+    "mn",
+    "abq",
+    "ady",
+    "kbd",
+    "ava",
+    "dar",
+    "inh",
+    "che",
+    "lbe",
+    "lez",
+    "tab",
+    "cyrillic",
+]
+devanagari_lang = [
+    "hi",
+    "mr",
+    "ne",
+    "bh",
+    "mai",
+    "ang",
+    "bho",
+    "mah",
+    "sck",
+    "new",
+    "gom",
+    "sa",
+    "bgc",
+    "devanagari",
+]
+multi_lang = latin_lang + arabic_lang + cyrillic_lang + devanagari_lang
+
+assert os.path.isfile(
+    "./rec_multi_language_lite_train.yml"
+), "Loss basic configuration file rec_multi_language_lite_train.yml.\
+You can download it from \
+https://github.com/PaddlePaddle/PaddleOCR/tree/dygraph/configs/rec/multi_language/"
+
+global_config = yaml.load(
+    open("./rec_multi_language_lite_train.yml", "rb"), Loader=yaml.SafeLoader
+)
+project_path = os.path.abspath(os.path.join(os.getcwd(), "../../../"))
+
+
+class ArgsParser(ArgumentParser):
+    def __init__(self):
+        super(ArgsParser, self).__init__(formatter_class=RawDescriptionHelpFormatter)
+        self.add_argument("-o", "--opt", nargs="+", help="set configuration options")
+        self.add_argument(
+            "-l",
+            "--language",
+            nargs="+",
+            help="set language type, support {}".format(support_list),
+        )
+        self.add_argument(
+            "--train",
+            type=str,
+            help="you can use this command to change the train dataset default path",
+        )
+        self.add_argument(
+            "--val",
+            type=str,
+            help="you can use this command to change the eval dataset default path",
+        )
+        self.add_argument(
+            "--dict",
+            type=str,
+            help="you can use this command to change the dictionary default path",
+        )
+        self.add_argument(
+            "--data_dir",
+            type=str,
+            help="you can use this command to change the dataset default root path",
+        )
+
+    def parse_args(self, argv=None):
+        args = super(ArgsParser, self).parse_args(argv)
+        args.opt = self._parse_opt(args.opt)
+        args.language = self._set_language(args.language)
+        return args
+
+    def _parse_opt(self, opts):
+        config = {}
+        if not opts:
+            return config
+        for s in opts:
+            s = s.strip()
+            k, v = s.split("=")
+            config[k] = yaml.load(v, Loader=yaml.SafeLoader)
+        return config
+
+    def _set_language(self, type):
+        lang = type[0]
+        assert type, "please use -l or --language to choose language type"
+        assert lang in support_list.keys() or lang in multi_lang, (
+            "the sub_keys(-l or --language) can only be one of support list: \n{},\nbut get: {}, "
+            "please check your running command".format(multi_lang, type)
+        )
+        if lang in latin_lang:
+            lang = "latin"
+        elif lang in arabic_lang:
+            lang = "arabic"
+        elif lang in cyrillic_lang:
+            lang = "cyrillic"
+        elif lang in devanagari_lang:
+            lang = "devanagari"
+        global_config["Global"]["character_dict_path"] = (
+            "ppocr/utils/dict/{}_dict.txt".format(lang)
+        )
+        global_config["Global"]["save_model_dir"] = "./output/rec_{}_lite".format(lang)
+        global_config["Train"]["dataset"]["label_file_list"] = [
+            "train_data/{}_train.txt".format(lang)
+        ]
+        global_config["Eval"]["dataset"]["label_file_list"] = [
+            "train_data/{}_val.txt".format(lang)
+        ]
+        global_config["Global"]["character_type"] = lang
+        assert os.path.isfile(
+            os.path.join(project_path, global_config["Global"]["character_dict_path"])
+        ), "Loss default dictionary file {}_dict.txt.You can download it from \
+https://github.com/PaddlePaddle/PaddleOCR/tree/dygraph/ppocr/utils/dict/".format(
+            lang
+        )
+        return lang
+
+
+def merge_config(config):
+    """
+    Merge config into global config.
+    Args:
+        config (dict): Config to be merged.
+    Returns: global config
+    """
+    for key, value in config.items():
+        if "." not in key:
+            if isinstance(value, dict) and key in global_config:
+                global_config[key].update(value)
+            else:
+                global_config[key] = value
+        else:
+            sub_keys = key.split(".")
+            assert (
+                sub_keys[0] in global_config
+            ), "the sub_keys can only be one of global_config: {}, but get: {}, please check your running command".format(
+                global_config.keys(), sub_keys[0]
+            )
+            cur = global_config[sub_keys[0]]
+            for idx, sub_key in enumerate(sub_keys[1:]):
+                if idx == len(sub_keys) - 2:
+                    cur[sub_key] = value
+                else:
+                    cur = cur[sub_key]
+
+
+def loss_file(path):
+    assert os.path.exists(
+        path
+    ), "There is no such file:{},Please do not forget to put in the specified file".format(
+        path
+    )
+
+
+if __name__ == "__main__":
+    FLAGS = ArgsParser().parse_args()
+    merge_config(FLAGS.opt)
+    save_file_path = "rec_{}_lite_train.yml".format(FLAGS.language)
+    if os.path.isfile(save_file_path):
+        os.remove(save_file_path)
+
+    if FLAGS.train:
+        global_config["Train"]["dataset"]["label_file_list"] = [FLAGS.train]
+        train_label_path = os.path.join(project_path, FLAGS.train)
+        loss_file(train_label_path)
+    if FLAGS.val:
+        global_config["Eval"]["dataset"]["label_file_list"] = [FLAGS.val]
+        eval_label_path = os.path.join(project_path, FLAGS.val)
+        loss_file(eval_label_path)
+    if FLAGS.dict:
+        global_config["Global"]["character_dict_path"] = FLAGS.dict
+        dict_path = os.path.join(project_path, FLAGS.dict)
+        loss_file(dict_path)
+    if FLAGS.data_dir:
+        global_config["Eval"]["dataset"]["data_dir"] = FLAGS.data_dir
+        global_config["Train"]["dataset"]["data_dir"] = FLAGS.data_dir
+        data_dir = os.path.join(project_path, FLAGS.data_dir)
+        loss_file(data_dir)
+
+    with open(save_file_path, "w") as f:
+        yaml.dump(dict(global_config), f, default_flow_style=False, sort_keys=False)
+    logger.info("Project path is          :{}".format(project_path))
+    logger.info(
+        "Train list path set to   :{}".format(
+            global_config["Train"]["dataset"]["label_file_list"][0]
+        )
+    )
+    logger.info(
+        "Eval list path set to    :{}".format(
+            global_config["Eval"]["dataset"]["label_file_list"][0]
+        )
+    )
+    logger.info(
+        "Dataset root path set to :{}".format(
+            global_config["Eval"]["dataset"]["data_dir"]
+        )
+    )
+    logger.info(
+        "Dict path set to         :{}".format(
+            global_config["Global"]["character_dict_path"]
+        )
+    )
+    logger.info(
+        "Config file set to       :configs/rec/multi_language/{}".format(save_file_path)
+    )
diff --git a/configs/rec/multi_language/rec_arabic_lite_train.yml b/configs/rec/multi_language/rec_arabic_lite_train.yml
new file mode 100644
index 0000000..a746260
--- /dev/null
+++ b/configs/rec/multi_language/rec_arabic_lite_train.yml
@@ -0,0 +1,110 @@
+Global:
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_arabic_lite
+  save_epoch_step: 3
+  eval_batch_step:
+  - 0
+  - 2000
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: null
+  character_dict_path: ppocr/utils/dict/arabic_dict.txt
+  max_text_length: 25
+  infer_mode: false
+  use_space_char: true
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: L2
+    factor: 1.0e-05
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform: null
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: small
+    small_stride:
+    - 1
+    - 2
+    - 2
+    - 2
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 48
+  Head:
+    name: CTCHead
+    fc_decay: 1.0e-05
+Loss:
+  name: CTCLoss
+PostProcess:
+  name: CTCLabelDecode
+Metric:
+  name: RecMetric
+  main_indicator: acc
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/
+    label_file_list:
+    - train_data/arabic_train.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug: null
+    - CTCLabelEncode: null
+    - RecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: true
+    batch_size_per_card: 256
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/
+    label_file_list:
+    - train_data/arabic_val.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - CTCLabelEncode: null
+    - RecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/multi_language/rec_cyrillic_lite_train.yml b/configs/rec/multi_language/rec_cyrillic_lite_train.yml
new file mode 100644
index 0000000..98544f6
--- /dev/null
+++ b/configs/rec/multi_language/rec_cyrillic_lite_train.yml
@@ -0,0 +1,110 @@
+Global:
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_cyrillic_lite
+  save_epoch_step: 3
+  eval_batch_step:
+  - 0
+  - 2000
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: null
+  character_dict_path: ppocr/utils/dict/cyrillic_dict.txt
+  max_text_length: 25
+  infer_mode: false
+  use_space_char: true
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: L2
+    factor: 1.0e-05
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform: null
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: small
+    small_stride:
+    - 1
+    - 2
+    - 2
+    - 2
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 48
+  Head:
+    name: CTCHead
+    fc_decay: 1.0e-05
+Loss:
+  name: CTCLoss
+PostProcess:
+  name: CTCLabelDecode
+Metric:
+  name: RecMetric
+  main_indicator: acc
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/
+    label_file_list:
+    - train_data/cyrillic_train.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug: null
+    - CTCLabelEncode: null
+    - RecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: true
+    batch_size_per_card: 256
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/
+    label_file_list:
+    - train_data/cyrillic_val.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - CTCLabelEncode: null
+    - RecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/multi_language/rec_devanagari_lite_train.yml b/configs/rec/multi_language/rec_devanagari_lite_train.yml
new file mode 100644
index 0000000..518b9f1
--- /dev/null
+++ b/configs/rec/multi_language/rec_devanagari_lite_train.yml
@@ -0,0 +1,110 @@
+Global:
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_devanagari_lite
+  save_epoch_step: 3
+  eval_batch_step:
+  - 0
+  - 2000
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: null
+  character_dict_path: ppocr/utils/dict/devanagari_dict.txt
+  max_text_length: 25
+  infer_mode: false
+  use_space_char: true
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: L2
+    factor: 1.0e-05
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform: null
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: small
+    small_stride:
+    - 1
+    - 2
+    - 2
+    - 2
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 48
+  Head:
+    name: CTCHead
+    fc_decay: 1.0e-05
+Loss:
+  name: CTCLoss
+PostProcess:
+  name: CTCLabelDecode
+Metric:
+  name: RecMetric
+  main_indicator: acc
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/
+    label_file_list:
+    - train_data/devanagari_train.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug: null
+    - CTCLabelEncode: null
+    - RecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: true
+    batch_size_per_card: 256
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/
+    label_file_list:
+    - train_data/devanagari_val.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - CTCLabelEncode: null
+    - RecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/multi_language/rec_en_number_lite_train.yml b/configs/rec/multi_language/rec_en_number_lite_train.yml
new file mode 100644
index 0000000..ff1fb86
--- /dev/null
+++ b/configs/rec/multi_language/rec_en_number_lite_train.yml
@@ -0,0 +1,101 @@
+Global:
+  use_gpu: True
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_en_number_lite
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  # if pretrained_model is saved in static mode, load_static_weights must set to True
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img:
+  # for data or label process
+  character_dict_path: ppocr/utils/en_dict.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: True
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0.00001
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: small
+    small_stride: [1, 2, 2, 2]
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 48
+  Head:
+    name: CTCHead
+    fc_decay: 0.00001
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list: ["./train_data/train_list.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - RecAug: 
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list: ["./train_data/eval_list.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/multi_language/rec_french_lite_train.yml b/configs/rec/multi_language/rec_french_lite_train.yml
new file mode 100644
index 0000000..217369d
--- /dev/null
+++ b/configs/rec/multi_language/rec_french_lite_train.yml
@@ -0,0 +1,101 @@
+Global:
+  use_gpu: True
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_french_lite
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  # if pretrained_model is saved in static mode, load_static_weights must set to True
+  cal_metric_during_train: True
+  pretrained_model: 
+  checkpoints: 
+  save_inference_dir: 
+  use_visualdl: False
+  infer_img:
+  # for data or label process
+  character_dict_path: ppocr/utils/dict/french_dict.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0.00001
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: small
+    small_stride: [1, 2, 2, 2]
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 48
+  Head:
+    name: CTCHead
+    fc_decay: 0.00001
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list: ["./train_data/train_list.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - RecAug: 
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list: ["./train_data/eval_list.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/multi_language/rec_german_lite_train.yml b/configs/rec/multi_language/rec_german_lite_train.yml
new file mode 100644
index 0000000..67520f5
--- /dev/null
+++ b/configs/rec/multi_language/rec_german_lite_train.yml
@@ -0,0 +1,101 @@
+Global:
+  use_gpu: True
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_german_lite
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  # if pretrained_model is saved in static mode, load_static_weights must set to True
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img:
+  # for data or label process
+  character_dict_path: ppocr/utils/dict/german_dict.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0.00001
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: small
+    small_stride: [1, 2, 2, 2]
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 48
+  Head:
+    name: CTCHead
+    fc_decay: 0.00001
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list: ["./train_data/train_list.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - RecAug: 
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list: ["./train_data/eval_list.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/multi_language/rec_hebrew_lite_train.yml b/configs/rec/multi_language/rec_hebrew_lite_train.yml
new file mode 100644
index 0000000..056073e
--- /dev/null
+++ b/configs/rec/multi_language/rec_hebrew_lite_train.yml
@@ -0,0 +1,110 @@
+Global:
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_hebrew_lite
+  save_epoch_step: 3
+  eval_batch_step:
+  - 0
+  - 2000
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: null
+  character_dict_path: ppocr/utils/dict/hebrew_dict.txt
+  max_text_length: 25
+  infer_mode: false
+  use_space_char: true
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: L2
+    factor: 1.0e-05
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform: null
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: small
+    small_stride:
+    - 1
+    - 2
+    - 2
+    - 2
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 48
+  Head:
+    name: CTCHead
+    fc_decay: 1.0e-05
+Loss:
+  name: CTCLoss
+PostProcess:
+  name: CTCLabelDecode
+Metric:
+  name: RecMetric
+  main_indicator: acc
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/
+    label_file_list:
+    - train_data/hebrew_train.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug: null
+    - CTCLabelEncode: null
+    - RecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: true
+    batch_size_per_card: 256
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/
+    label_file_list:
+    - train_data/hebrew_val.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - CTCLabelEncode: null
+    - RecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/multi_language/rec_japan_lite_train.yml b/configs/rec/multi_language/rec_japan_lite_train.yml
new file mode 100644
index 0000000..448aff1
--- /dev/null
+++ b/configs/rec/multi_language/rec_japan_lite_train.yml
@@ -0,0 +1,101 @@
+Global:
+  use_gpu: True
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_japan_lite
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  # if pretrained_model is saved in static mode, load_static_weights must set to True
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img:
+  # for data or label process
+  character_dict_path: ppocr/utils/dict/japan_dict.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0.00001
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: small
+    small_stride: [1, 2, 2, 2]
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 48
+  Head:
+    name: CTCHead
+    fc_decay: 0.00001
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list: ["./train_data/train_list.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - RecAug: 
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list: ["./train_data/eval_list.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/multi_language/rec_korean_lite_train.yml b/configs/rec/multi_language/rec_korean_lite_train.yml
new file mode 100644
index 0000000..8118119
--- /dev/null
+++ b/configs/rec/multi_language/rec_korean_lite_train.yml
@@ -0,0 +1,101 @@
+Global:
+  use_gpu: True
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_korean_lite
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  # if pretrained_model is saved in static mode, load_static_weights must set to True
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img:
+  # for data or label process
+  character_dict_path: ppocr/utils/dict/korean_dict.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0.00001
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: small
+    small_stride: [1, 2, 2, 2]
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 48
+  Head:
+    name: CTCHead
+    fc_decay: 0.00001
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list: ["./train_data/train_list.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - RecAug: 
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list: ["./train_data/eval_list.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/multi_language/rec_latin_lite_train.yml b/configs/rec/multi_language/rec_latin_lite_train.yml
new file mode 100644
index 0000000..04fe6d1
--- /dev/null
+++ b/configs/rec/multi_language/rec_latin_lite_train.yml
@@ -0,0 +1,110 @@
+Global:
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_latin_lite
+  save_epoch_step: 3
+  eval_batch_step:
+  - 0
+  - 2000
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: null
+  character_dict_path: ppocr/utils/dict/latin_dict.txt
+  max_text_length: 25
+  infer_mode: false
+  use_space_char: true
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: L2
+    factor: 1.0e-05
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform: null
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: small
+    small_stride:
+    - 1
+    - 2
+    - 2
+    - 2
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 48
+  Head:
+    name: CTCHead
+    fc_decay: 1.0e-05
+Loss:
+  name: CTCLoss
+PostProcess:
+  name: CTCLabelDecode
+Metric:
+  name: RecMetric
+  main_indicator: acc
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/
+    label_file_list:
+    - train_data/latin_train.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug: null
+    - CTCLabelEncode: null
+    - RecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: true
+    batch_size_per_card: 256
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/
+    label_file_list:
+    - train_data/latin_val.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - CTCLabelEncode: null
+    - RecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/multi_language/rec_multi_language_lite_train.yml b/configs/rec/multi_language/rec_multi_language_lite_train.yml
new file mode 100644
index 0000000..c42a3d1
--- /dev/null
+++ b/configs/rec/multi_language/rec_multi_language_lite_train.yml
@@ -0,0 +1,103 @@
+Global:
+  use_gpu: True
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_multi_language_lite
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  # if pretrained_model is saved in static mode, load_static_weights must set to True
+  cal_metric_during_train: True
+  pretrained_model: 
+  checkpoints: 
+  save_inference_dir: 
+  use_visualdl: False
+  infer_img:
+  # for data or label process
+  character_dict_path: 
+  # Set the language of training, if set, select the default dictionary file
+  character_type: 
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: True
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0.00001
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: small
+    small_stride: [1, 2, 2, 2]
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 48
+  Head:
+    name: CTCHead
+    fc_decay: 0.00001
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/
+    label_file_list: ["./train_data/train_list.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - RecAug: 
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/
+    label_file_list: ["./train_data/val_list.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/multi_language/rec_samaritan_lite_train.yml b/configs/rec/multi_language/rec_samaritan_lite_train.yml
new file mode 100644
index 0000000..82f9f2c
--- /dev/null
+++ b/configs/rec/multi_language/rec_samaritan_lite_train.yml
@@ -0,0 +1,110 @@
+Global:
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_samaritan_lite
+  save_epoch_step: 3
+  eval_batch_step:
+  - 0
+  - 2000
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: null
+  character_dict_path: ppocr/utils/dict/samaritan_dict.txt
+  max_text_length: 25
+  infer_mode: false
+  use_space_char: true
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: L2
+    factor: 1.0e-05
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform: null
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: small
+    small_stride:
+    - 1
+    - 2
+    - 2
+    - 2
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 48
+  Head:
+    name: CTCHead
+    fc_decay: 1.0e-05
+Loss:
+  name: CTCLoss
+PostProcess:
+  name: CTCLabelDecode
+Metric:
+  name: RecMetric
+  main_indicator: acc
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/
+    label_file_list:
+    - train_data/samaritan_train.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug: null
+    - CTCLabelEncode: null
+    - RecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: true
+    batch_size_per_card: 256
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/
+    label_file_list:
+    - train_data/samaritan_val.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - CTCLabelEncode: null
+    - RecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/multi_language/rec_syriac_lite_train.yml b/configs/rec/multi_language/rec_syriac_lite_train.yml
new file mode 100644
index 0000000..8d7d22f
--- /dev/null
+++ b/configs/rec/multi_language/rec_syriac_lite_train.yml
@@ -0,0 +1,110 @@
+Global:
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_syriac_lite
+  save_epoch_step: 3
+  eval_batch_step:
+  - 0
+  - 2000
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: null
+  character_dict_path: ppocr/utils/dict/syriac_dict.txt
+  max_text_length: 25
+  infer_mode: false
+  use_space_char: true
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: L2
+    factor: 1.0e-05
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform: null
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: small
+    small_stride:
+    - 1
+    - 2
+    - 2
+    - 2
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 48
+  Head:
+    name: CTCHead
+    fc_decay: 1.0e-05
+Loss:
+  name: CTCLoss
+PostProcess:
+  name: CTCLabelDecode
+Metric:
+  name: RecMetric
+  main_indicator: acc
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/
+    label_file_list:
+    - train_data/syriac_train.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug: null
+    - CTCLabelEncode: null
+    - RecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: true
+    batch_size_per_card: 256
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/
+    label_file_list:
+    - train_data/syriac_val.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - CTCLabelEncode: null
+    - RecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/rec_d28_can.yml b/configs/rec/rec_d28_can.yml
new file mode 100644
index 0000000..7c3b0fd
--- /dev/null
+++ b/configs/rec/rec_d28_can.yml
@@ -0,0 +1,122 @@
+Global:
+  use_gpu: True
+  epoch_num: 240
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/can/
+  save_epoch_step: 1
+  # evaluation is run every 1105 iterations (1 epoch)(batch_size = 8)
+  eval_batch_step: [0, 1105]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/datasets/crohme_demo/hme_00.jpg
+  # for data or label process
+  character_dict_path: ppocr/utils/dict/latex_symbol_dict.txt
+  max_text_length: 36
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_can.txt
+
+Optimizer:
+  name: Momentum
+  momentum: 0.9
+  clip_norm_global: 100.0
+  lr:
+    name: TwoStepCosine
+    learning_rate: 0.01
+    warmup_epoch: 1
+  weight_decay: 0.0001
+
+Architecture:
+  model_type: rec
+  algorithm: CAN
+  in_channels: 1
+  Transform:
+  Backbone:
+    name: DenseNet 
+    growthRate: 24
+    reduction: 0.5
+    bottleneck: True
+    use_dropout: True
+    input_channel: 1 
+  Head:
+    name: CANHead
+    in_channel: 684
+    out_channel: 111
+    max_text_length: 36
+    ratio: 16
+    attdecoder:
+      is_train: True
+      input_size: 256
+      hidden_size: 256
+      encoder_out_channel: 684
+      dropout: True
+      dropout_ratio: 0.5
+      word_num: 111
+      counting_decoder_out_channel: 111
+      attention:
+        attention_dim: 512
+        word_conv_kernel: 1
+   
+Loss:
+  name: CANLoss
+
+PostProcess:
+  name: CANLabelDecode
+
+Metric:
+  name: CANMetric
+  main_indicator: exp_rate
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/CROHME/training/images/
+    label_file_list: ["./train_data/CROHME/training/labels.txt"]
+    transforms:
+      - DecodeImage:
+          channel_first: False
+      - NormalizeImage:
+          mean: [0,0,0]
+          std: [1,1,1]
+          order: 'hwc'
+      - GrayImageChannelFormat: 
+          inverse: True
+      - CANLabelEncode:
+          lower: False
+      - KeepKeys:
+          keep_keys: ['image', 'label']
+  loader:
+    shuffle: True
+    batch_size_per_card: 8
+    drop_last: False
+    num_workers: 4
+    collate_fn: DyMaskCollator
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/CROHME/evaluation/images/
+    label_file_list: ["./train_data/CROHME/evaluation/labels.txt"]
+    transforms: 
+      - DecodeImage:
+          channel_first: False
+      - NormalizeImage:
+          mean: [0,0,0]
+          std: [1,1,1]
+          order: 'hwc'
+      - GrayImageChannelFormat:
+          inverse: True
+      - CANLabelEncode:
+          lower: False
+      - KeepKeys:
+          keep_keys: ['image', 'label']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1
+    num_workers: 4
+    collate_fn: DyMaskCollator
diff --git a/configs/rec/rec_efficientb3_fpn_pren.yml b/configs/rec/rec_efficientb3_fpn_pren.yml
new file mode 100644
index 0000000..0fac6a7
--- /dev/null
+++ b/configs/rec/rec_efficientb3_fpn_pren.yml
@@ -0,0 +1,92 @@
+Global:
+  use_gpu: True
+  epoch_num: 8
+  log_smooth_window: 20
+  print_batch_step: 5
+  save_model_dir: ./output/rec/pren_new
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations after the 4000th iteration
+  eval_batch_step: [4000, 2000]
+  cal_metric_during_train: True
+  pretrained_model: 
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  # for data or label process
+  character_dict_path:
+  max_text_length: &max_text_length 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_pren.txt
+
+Optimizer:
+  name: Adadelta
+  lr:
+    name: Piecewise
+    decay_epochs: [2, 5, 7]
+    values: [0.5, 0.1, 0.01, 0.001]
+
+Architecture:
+  model_type: rec
+  algorithm: PREN
+  in_channels: 3
+  Backbone:
+    name: EfficientNetb3_PREN
+  Neck:
+    name: PRENFPN
+    n_r: 5
+    d_model: 384
+    max_len: *max_text_length
+    dropout: 0.1
+  Head:
+    name: PRENHead
+
+Loss:
+  name: PRENLoss
+
+PostProcess:
+  name: PRENLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: False
+      - PRENLabelEncode:
+      - RecAug:
+      - PRENResizeImg: 
+          image_shape: [64, 256]  # h,w
+      - KeepKeys:
+          keep_keys: ['image', 'label']
+  loader:
+    shuffle: True
+    batch_size_per_card: 128
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: False
+      - PRENLabelEncode:
+      - PRENResizeImg: 
+          image_shape: [64, 256]  # h,w
+      - KeepKeys:
+          keep_keys: ['image', 'label']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 64
+    num_workers: 8
diff --git a/configs/rec/rec_icdar15_train.yml b/configs/rec/rec_icdar15_train.yml
new file mode 100644
index 0000000..893f738
--- /dev/null
+++ b/configs/rec/rec_icdar15_train.yml
@@ -0,0 +1,99 @@
+Global:
+  use_gpu: true
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/ic15/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir: ./
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path: ppocr/utils/en_dict.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_ic15.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 96
+  Head:
+    name: CTCHead
+    fc_decay: 0
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+    use_shared_memory: False
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
+    use_shared_memory: False
diff --git a/configs/rec/rec_mtb_nrtr.yml b/configs/rec/rec_mtb_nrtr.yml
new file mode 100644
index 0000000..aa9a347
--- /dev/null
+++ b/configs/rec/rec_mtb_nrtr.yml
@@ -0,0 +1,101 @@
+Global:
+  use_gpu: True
+  epoch_num: 21
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/nrtr/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path: ppocr/utils/EN_symbol_dict.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_nrtr.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.99
+  clip_norm: 5.0
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 2
+  regularizer:
+    name: 'L2'
+    factor: 0.
+
+Architecture:
+  model_type: rec
+  algorithm: NRTR
+  in_channels: 1
+  Transform:
+  Backbone:
+    name: MTB
+    cnn_num: 2
+  Head:
+    name: Transformer
+    d_model: 512
+    num_encoder_layers: 6
+    beam_size: -1 # When Beam size is greater than 0, it means to use beam search when evaluation.
+    
+
+Loss:
+  name: CELoss
+  smoothing: True
+
+PostProcess:
+  name: NRTRLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - NRTRLabelEncode: # Class handling label
+      - GrayRecResizeImg:
+          image_shape: [100, 32] # W H
+          resize_type: PIL # PIL or OpenCV
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 512
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/evaluation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - NRTRLabelEncode: # Class handling label
+      - GrayRecResizeImg:
+          image_shape: [100, 32] # W H
+          resize_type: PIL # PIL or OpenCV
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
+    use_shared_memory: False
diff --git a/configs/rec/rec_mv3_none_bilstm_ctc.yml b/configs/rec/rec_mv3_none_bilstm_ctc.yml
new file mode 100644
index 0000000..9a95092
--- /dev/null
+++ b/configs/rec/rec_mv3_none_bilstm_ctc.yml
@@ -0,0 +1,95 @@
+Global:
+  use_gpu: True
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/mv3_none_bilstm_ctc/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_mv3_none_bilstm_ctc.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 96
+  Head:
+    name: CTCHead
+    fc_decay: 0
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
diff --git a/configs/rec/rec_mv3_none_none_ctc.yml b/configs/rec/rec_mv3_none_none_ctc.yml
new file mode 100644
index 0000000..28f0252
--- /dev/null
+++ b/configs/rec/rec_mv3_none_none_ctc.yml
@@ -0,0 +1,94 @@
+Global:
+  use_gpu: True
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/mv3_none_none_ctc/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_mv3_none_none_ctc.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: Rosetta
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+    fc_decay: 0.0004
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/rec_mv3_tps_bilstm_att.yml b/configs/rec/rec_mv3_tps_bilstm_att.yml
new file mode 100644
index 0000000..6c347e7
--- /dev/null
+++ b/configs/rec/rec_mv3_tps_bilstm_att.yml
@@ -0,0 +1,101 @@
+Global:
+  use_gpu: True
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/rec_mv3_tps_bilstm_att/
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_mv3_tps_bilstm_att.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0.00001
+
+Architecture:
+  model_type: rec
+  algorithm: RARE
+  Transform:
+    name: TPS
+    num_fiducial: 20
+    loc_lr: 0.1
+    model_name: small
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn 
+    hidden_size: 96
+  Head:
+    name: AttentionHead  
+    hidden_size: 96
+    
+
+Loss:
+  name: AttentionLoss
+
+PostProcess:
+  name: AttnLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - AttnLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - AttnLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 1
diff --git a/configs/rec/rec_mv3_tps_bilstm_ctc.yml b/configs/rec/rec_mv3_tps_bilstm_ctc.yml
new file mode 100644
index 0000000..9d1ebbe
--- /dev/null
+++ b/configs/rec/rec_mv3_tps_bilstm_ctc.yml
@@ -0,0 +1,99 @@
+Global:
+  use_gpu: True
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/mv3_tps_bilstm_ctc/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_mv3_tps_bilstm_ctc.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: STARNet
+  Transform:
+    name: TPS
+    num_fiducial: 20
+    loc_lr: 0.1
+    model_name: small
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 96
+  Head:
+    name: CTCHead
+    fc_decay: 0.0004
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
diff --git a/configs/rec/rec_r31_robustscanner.yml b/configs/rec/rec_r31_robustscanner.yml
new file mode 100644
index 0000000..54b69d4
--- /dev/null
+++ b/configs/rec/rec_r31_robustscanner.yml
@@ -0,0 +1,109 @@
+Global:
+  use_gpu: true
+  epoch_num: 5
+  log_smooth_window: 20
+  print_batch_step: 20
+  save_model_dir: ./output/rec/rec_r31_robustscanner/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path: ppocr/utils/dict90.txt
+  max_text_length: &max_text_length 40
+  infer_mode: False
+  use_space_char: False
+  rm_symbol: True
+  save_res_path: ./output/rec/predicts_robustscanner.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Piecewise
+    decay_epochs: [3, 4]
+    values: [0.001, 0.0001, 0.00001] 
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: RobustScanner
+  Transform:
+  Backbone:
+    name: ResNet31
+    init_type: KaimingNormal
+  Head:
+    name: RobustScannerHead
+    enc_outchannles: 128
+    hybrid_dec_rnn_layers: 2
+    hybrid_dec_dropout: 0
+    position_dec_rnn_layers: 2
+    start_idx: 91
+    mask: True
+    padding_idx: 92
+    encode_value: False
+    max_text_length: *max_text_length
+
+Loss:
+  name: SARLoss
+
+PostProcess:
+  name: SARLabelDecode
+
+Metric:
+  name: RecMetric
+  is_filter: True
+
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SARLabelEncode: # Class handling label
+      - RobustScannerRecResizeImg:
+          image_shape: [3, 48, 48, 160] # h:48 w:[48,160]
+          width_downsample_ratio: 0.25
+          max_text_length: *max_text_length
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'valid_ratio', 'word_positons'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 64
+    drop_last: True
+    num_workers: 8
+    use_shared_memory: False
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/evaluation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SARLabelEncode: # Class handling label
+      - RobustScannerRecResizeImg:
+          image_shape: [3, 48, 48, 160]
+          max_text_length: *max_text_length
+          width_downsample_ratio: 0.25
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'valid_ratio', 'word_positons'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 64
+    num_workers: 4
+    use_shared_memory: False
+  
diff --git a/configs/rec/rec_r31_sar.yml b/configs/rec/rec_r31_sar.yml
new file mode 100644
index 0000000..65e7877
--- /dev/null
+++ b/configs/rec/rec_r31_sar.yml
@@ -0,0 +1,98 @@
+Global:
+  use_gpu: true
+  epoch_num: 5
+  log_smooth_window: 20
+  print_batch_step: 20
+  save_model_dir: ./sar_rec
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: 
+  # for data or label process
+  character_dict_path: ppocr/utils/dict90.txt
+  max_text_length: 30
+  infer_mode: False
+  use_space_char: False
+  rm_symbol: True
+  save_res_path: ./output/rec/predicts_sar.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Piecewise
+    decay_epochs: [3, 4]
+    values: [0.001, 0.0001, 0.00001] 
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: SAR
+  Transform:
+  Backbone:
+    name: ResNet31
+  Head:
+    name: SARHead
+
+Loss:
+  name: SARLoss
+
+PostProcess:
+  name: SARLabelDecode
+
+Metric:
+  name: RecMetric
+
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    label_file_list: ['./train_data/train_list.txt']
+    data_dir: ./train_data/
+    ratio_list: 1.0
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SARLabelEncode: # Class handling label
+      - SARRecResizeImg:
+          image_shape: [3, 48, 48, 160] # h:48 w:[48,160]
+          width_downsample_ratio: 0.25
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'valid_ratio'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 64
+    drop_last: True
+    num_workers: 8
+    use_shared_memory: False
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/evaluation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SARLabelEncode: # Class handling label
+      - SARRecResizeImg:
+          image_shape: [3, 48, 48, 160]
+          width_downsample_ratio: 0.25
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'valid_ratio'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 64
+    num_workers: 4
+    use_shared_memory: False
+  
diff --git a/configs/rec/rec_r32_gaspin_bilstm_att.yml b/configs/rec/rec_r32_gaspin_bilstm_att.yml
new file mode 100644
index 0000000..91d3e10
--- /dev/null
+++ b/configs/rec/rec_r32_gaspin_bilstm_att.yml
@@ -0,0 +1,116 @@
+Global:
+  use_gpu: True
+  epoch_num: 6
+  log_smooth_window: 50
+  print_batch_step: 50
+  save_model_dir: ./output/rec/rec_r32_gaspin_bilstm_att/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path: ./ppocr/utils/dict/spin_dict.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_r32_gaspin_bilstm_att.txt
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Piecewise
+    decay_epochs: [3, 4, 5]
+    values: [0.001, 0.0003, 0.00009, 0.000027] 
+  clip_norm: 5
+
+Architecture:
+  model_type: rec
+  algorithm: SPIN
+  in_channels: 1
+  Transform:
+    name: GA_SPIN
+    offsets: True
+    default_type: 6
+    loc_lr: 0.1
+    stn: True
+  Backbone:
+    name: ResNet32
+    out_channels: 512
+  Neck:
+    name: SequenceEncoder
+    encoder_type: cascadernn 
+    hidden_size: 256
+    out_channels: [256, 512]
+    with_linear: True
+  Head:
+    name: SPINAttentionHead  
+    hidden_size: 256
+    
+
+Loss:
+  name: SPINAttentionLoss
+  ignore_index: 0
+
+PostProcess:
+  name: SPINLabelDecode
+  use_space_char: False
+
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  is_filter: True
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SPINLabelEncode: # Class handling label
+      - SPINRecResizeImg:
+          image_shape: [100, 32]
+          interpolation : 2
+          mean: [127.5]
+          std: [127.5]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 8
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SPINLabelEncode: # Class handling label
+      - SPINRecResizeImg:
+          image_shape: [100, 32]
+          interpolation : 2
+          mean: [127.5]
+          std: [127.5]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 2
diff --git a/configs/rec/rec_r34_vd_none_bilstm_ctc.yml b/configs/rec/rec_r34_vd_none_bilstm_ctc.yml
new file mode 100644
index 0000000..9fdb5e9
--- /dev/null
+++ b/configs/rec/rec_r34_vd_none_bilstm_ctc.yml
@@ -0,0 +1,94 @@
+Global:
+  use_gpu: true
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/r34_vd_none_bilstm_ctc/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_r34_vd_none_bilstm_ctc.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: ResNet
+    layers: 34
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 256
+  Head:
+    name: CTCHead
+    fc_decay: 0
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
diff --git a/configs/rec/rec_r34_vd_none_none_ctc.yml b/configs/rec/rec_r34_vd_none_none_ctc.yml
new file mode 100644
index 0000000..0af2b2f
--- /dev/null
+++ b/configs/rec/rec_r34_vd_none_none_ctc.yml
@@ -0,0 +1,92 @@
+Global:
+  use_gpu: true
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/r34_vd_none_none_ctc/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_r34_vd_none_none_ctc.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: Rosetta
+  Backbone:
+    name: ResNet
+    layers: 34
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+    fc_decay: 0.0004
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
diff --git a/configs/rec/rec_r34_vd_tps_bilstm_att.yml b/configs/rec/rec_r34_vd_tps_bilstm_att.yml
new file mode 100644
index 0000000..8919aae
--- /dev/null
+++ b/configs/rec/rec_r34_vd_tps_bilstm_att.yml
@@ -0,0 +1,100 @@
+Global:
+  use_gpu: True
+  epoch_num: 400
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/b3_rare_r34_none_gru/
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_b3_rare_r34_none_gru.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0.00000
+
+Architecture:
+  model_type: rec
+  algorithm: RARE
+  Transform:
+    name: TPS
+    num_fiducial: 20
+    loc_lr: 0.1
+    model_name: large
+  Backbone:
+    name: ResNet  
+    layers: 34
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn 
+    hidden_size: 256 #96
+  Head:
+    name: AttentionHead  # AttentionHead
+    hidden_size: 256 #
+    l2_decay: 0.00001
+
+Loss:
+  name: AttentionLoss
+
+PostProcess:
+  name: AttnLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - AttnLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - AttnLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/rec_r34_vd_tps_bilstm_ctc.yml b/configs/rec/rec_r34_vd_tps_bilstm_ctc.yml
new file mode 100644
index 0000000..c21fe61
--- /dev/null
+++ b/configs/rec/rec_r34_vd_tps_bilstm_ctc.yml
@@ -0,0 +1,98 @@
+Global:
+  use_gpu: true
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/r34_vd_tps_bilstm_ctc/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_r34_vd_tps_bilstm_ctc.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: STARNet
+  Transform:
+    name: TPS
+    num_fiducial: 20
+    loc_lr: 0.1
+    model_name: large
+  Backbone:
+    name: ResNet
+    layers: 34
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 256
+  Head:
+    name: CTCHead
+    fc_decay: 0
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
diff --git a/configs/rec/rec_r45_abinet.yml b/configs/rec/rec_r45_abinet.yml
new file mode 100644
index 0000000..db2f28f
--- /dev/null
+++ b/configs/rec/rec_r45_abinet.yml
@@ -0,0 +1,103 @@
+Global:
+  use_gpu: True
+  epoch_num: 10
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/r45_abinet/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model: ./pretrain_models/abinet_vl_pretrained
+  checkpoints:
+  save_inference_dir: ./output/rec/r45_abinet/infer
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  character_type: en
+  max_text_length: &max_text_length 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_abinet.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.99
+  clip_norm: 20.0
+  lr:
+    name: Piecewise
+    decay_epochs: [6]
+    values: [0.0001, 0.00001]
+  regularizer:
+    name: 'L2'
+    factor: 0.
+
+Architecture:
+  model_type: rec
+  algorithm: ABINet
+  in_channels: 3
+  Transform:
+  Backbone:
+    name: ResNet45
+  Head:
+    name: ABINetHead
+    use_lang: True
+    iter_size: 3
+    max_length: *max_text_length
+    image_size: [ &h 32, &w 128 ]  # [ h, w ]
+
+
+Loss:
+  name: CELoss
+  ignore_index: &ignore_index 100 # Must be greater than the number of character classes
+
+PostProcess:
+  name: ABINetLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - ABINetRecAug:
+      - ABINetLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - ABINetRecResizeImg:
+          image_shape: [3, *h, *w]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 96
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/evaluation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - ABINetLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - ABINetRecResizeImg:
+          image_shape: [3, *h, *w]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
+    use_shared_memory: False
diff --git a/configs/rec/rec_r45_visionlan.yml b/configs/rec/rec_r45_visionlan.yml
new file mode 100644
index 0000000..2501765
--- /dev/null
+++ b/configs/rec/rec_r45_visionlan.yml
@@ -0,0 +1,106 @@
+Global:
+  use_gpu: true
+  epoch_num: 8
+  log_smooth_window: 200
+  print_batch_step: 200
+  save_model_dir: ./output/rec/r45_visionlan
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: True
+  infer_img: doc/imgs_words/en/word_2.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: &max_text_length 25
+  training_step: &training_step LA
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_visionlan.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  clip_norm: 20.0
+  group_lr: true
+  training_step: *training_step
+  lr:
+    name: Piecewise
+    decay_epochs: [6]
+    values: [0.0001, 0.00001] 
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: VisionLAN
+  Transform:
+  Backbone:
+    name: ResNet45
+    strides: [2, 2, 2, 1, 1]
+  Head:
+    name: VLHead
+    n_layers: 3
+    n_position: 256
+    n_dim: 512
+    max_text_length: *max_text_length
+    training_step: *training_step
+
+Loss:
+  name: VLLoss
+  mode: *training_step
+  weight_res: 0.5
+  weight_mas: 0.5
+
+PostProcess:
+  name: VLLabelDecode
+
+Metric:
+  name: RecMetric
+  is_filter: true
+
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - ABINetRecAug:
+      - VLLabelEncode: # Class handling label
+      - VLRecResizeImg:
+          image_shape: [3, 64, 256]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'label_res', 'label_sub', 'label_id', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 220
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VLLabelEncode: # Class handling label
+      - VLRecResizeImg:
+          image_shape: [3, 64, 256]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'label_res', 'label_sub', 'label_id', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 64
+    num_workers: 4
+  
diff --git a/configs/rec/rec_r50_fpn_srn.yml b/configs/rec/rec_r50_fpn_srn.yml
new file mode 100644
index 0000000..b685362
--- /dev/null
+++ b/configs/rec/rec_r50_fpn_srn.yml
@@ -0,0 +1,106 @@
+Global:
+  use_gpu: True
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 5
+  save_model_dir: ./output/rec/srn_new
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 5000]
+  cal_metric_during_train: True
+  pretrained_model: 
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  num_heads: 8
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_srn.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  clip_norm: 10.0
+  lr:
+    learning_rate: 0.0001
+
+Architecture:
+  model_type: rec
+  algorithm: SRN
+  in_channels: 1
+  Transform:
+  Backbone:
+    name: ResNetFPN
+  Head:
+    name: SRNHead
+    max_text_length: 25
+    num_heads: 8
+    num_encoder_TUs: 2
+    num_decoder_TUs: 4
+    hidden_dims: 512
+
+Loss:
+  name: SRNLoss
+
+PostProcess:
+  name: SRNLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SRNLabelEncode: # Class handling label
+      - SRNRecResizeImg:
+          image_shape: [1, 64, 256]
+      - KeepKeys:
+          keep_keys: ['image',
+                      'label',
+                      'length',
+                      'encoder_word_pos',
+                      'gsrm_word_pos',
+                      'gsrm_slf_attn_bias1',
+                      'gsrm_slf_attn_bias2'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    batch_size_per_card: 64
+    drop_last: False
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SRNLabelEncode: # Class handling label
+      - SRNRecResizeImg:
+          image_shape: [1, 64, 256]
+      - KeepKeys:
+          keep_keys: ['image',
+                      'label',
+                      'length',
+                      'encoder_word_pos',
+                      'gsrm_word_pos',
+                      'gsrm_slf_attn_bias1',
+                      'gsrm_slf_attn_bias2'] 
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 32
+    num_workers: 4
diff --git a/configs/rec/rec_resnet_rfl_att.yml b/configs/rec/rec_resnet_rfl_att.yml
new file mode 100644
index 0000000..b9fb741
--- /dev/null
+++ b/configs/rec/rec_resnet_rfl_att.yml
@@ -0,0 +1,112 @@
+Global:
+  use_gpu: True
+  epoch_num: 6
+  log_smooth_window: 20
+  print_batch_step: 50
+  save_model_dir: ./output/rec/rec_resnet_rfl_att/
+  save_epoch_step: 1
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 5000]
+  cal_metric_during_train: True
+  pretrained_model: ./pretrain_models/rec_resnet_rfl_visual/best_accuracy.pdparams 
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/rec_resnet_rfl.txt
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  weight_decay: 0.0
+  clip_norm_global: 5.0
+  lr:
+    name: Piecewise
+    decay_epochs : [3, 4, 5]
+    values : [0.001, 0.0003, 0.00009, 0.000027]
+
+Architecture:
+  model_type: rec
+  algorithm: RFL
+  in_channels: 1
+  Transform:
+    name: TPS
+    num_fiducial: 20
+    loc_lr: 1.0
+    model_name: large
+  Backbone:
+    name: ResNetRFL
+    use_cnt: True
+    use_seq: True
+  Neck:
+    name: RFAdaptor
+    use_v2s: True
+    use_s2v: True
+  Head:
+    name: RFLHead  
+    in_channels: 512
+    hidden_size: 256
+    batch_max_legnth: 25
+    out_channels: 38
+    use_cnt: True
+    use_seq: True
+
+Loss:
+  name: RFLLoss
+  # ignore_index: 0
+
+PostProcess:
+  name: RFLLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - RFLLabelEncode: # Class handling label
+      - RFLRecResizeImg:
+          image_shape: [1, 32, 100]
+          padding: false
+          interpolation: 2
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length', 'cnt_label'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 64
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/validation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - RFLLabelEncode: # Class handling label
+      - RFLRecResizeImg:
+          image_shape: [1, 32, 100]
+          padding: false
+          interpolation: 2
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length', 'cnt_label'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/rec_resnet_rfl_visual.yml b/configs/rec/rec_resnet_rfl_visual.yml
new file mode 100644
index 0000000..5eaea08
--- /dev/null
+++ b/configs/rec/rec_resnet_rfl_visual.yml
@@ -0,0 +1,110 @@
+Global:
+  use_gpu: True
+  epoch_num: 6
+  log_smooth_window: 20
+  print_batch_step: 50
+  save_model_dir: ./output/rec/rec_resnet_rfl_visual/
+  save_epoch_step: 1
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 5000]
+  cal_metric_during_train: False
+  pretrained_model:
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/rec_resnet_rfl_visual.txt
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  weight_decay: 0.0
+  clip_norm_global: 5.0
+  lr:
+    name: Piecewise
+    decay_epochs : [3, 4, 5]
+    values : [0.001, 0.0003, 0.00009, 0.000027]
+
+Architecture:
+  model_type: rec
+  algorithm: RFL
+  in_channels: 1
+  Transform:
+    name: TPS
+    num_fiducial: 20
+    loc_lr: 1.0
+    model_name: large
+  Backbone:
+    name: ResNetRFL
+    use_cnt: True
+    use_seq: False
+  Neck:
+    name: RFAdaptor
+    use_v2s: False
+    use_s2v: False
+  Head:
+    name: RFLHead  
+    in_channels: 512
+    hidden_size: 256
+    batch_max_legnth: 25
+    out_channels: 38
+    use_cnt: True
+    use_seq: False
+Loss:
+  name: RFLLoss
+
+PostProcess:
+  name: RFLLabelDecode
+
+Metric:
+  name: CNTMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - RFLLabelEncode: # Class handling label
+      - RFLRecResizeImg:
+          image_shape: [1, 32, 100]
+          padding: false
+          interpolation: 2
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length', 'cnt_label'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 64
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/evaluation
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - RFLLabelEncode: # Class handling label
+      - RFLRecResizeImg:
+          image_shape: [1, 32, 100]
+          padding: false
+          interpolation: 2
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length', 'cnt_label'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/configs/rec/rec_resnet_stn_bilstm_att.yml b/configs/rec/rec_resnet_stn_bilstm_att.yml
new file mode 100644
index 0000000..20ed953
--- /dev/null
+++ b/configs/rec/rec_resnet_stn_bilstm_att.yml
@@ -0,0 +1,108 @@
+Global:
+  use_gpu: True
+  epoch_num: 6
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/seed
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path: ppocr/utils/EN_symbol_dict.txt
+  max_text_length: 100
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_seed.txt
+
+
+Optimizer:
+  name: Adadelta
+  weight_deacy: 0.0
+  momentum: 0.9
+  lr:
+    name: Piecewise
+    decay_epochs: [4, 5]
+    values: [1.0, 0.1, 0.01]
+  regularizer:
+    name: 'L2'
+    factor: 2.0e-05
+
+
+Architecture:
+  model_type: rec
+  algorithm: SEED
+  Transform:
+    name: STN_ON
+    tps_inputsize: [32, 64]
+    tps_outputsize: [32, 100]
+    num_control_points: 20
+    tps_margins: [0.05,0.05]
+    stn_activation: none
+  Backbone:
+    name: ResNet_ASTER
+  Head:
+    name: AsterHead  # AttentionHead
+    sDim: 512
+    attDim: 512
+    max_len_labels: 100
+
+Loss:
+  name: AsterLoss
+
+PostProcess:
+  name: SEEDLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  is_filter: True
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - Fasttext:
+          path: "./cc.en.300.bin"
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SEEDLabelEncode: # Class handling label
+      - RecResizeImg:
+          character_dict_path:
+          image_shape: [3, 64, 256]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length', 'fast_label'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 6
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/evaluation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SEEDLabelEncode: # Class handling label
+      - RecResizeImg:
+          character_dict_path:
+          image_shape: [3, 64, 256]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: True
+    batch_size_per_card: 256
+    num_workers: 4
diff --git a/configs/rec/rec_satrn.yml b/configs/rec/rec_satrn.yml
new file mode 100644
index 0000000..376c2cc
--- /dev/null
+++ b/configs/rec/rec_satrn.yml
@@ -0,0 +1,116 @@
+Global:
+  use_gpu: true
+  epoch_num: 5
+  log_smooth_window: 20
+  print_batch_step: 50
+  save_model_dir: ./output/rec/rec_satrn/
+  save_epoch_step: 1
+  # evaluation is run every 5000 iterations
+  eval_batch_step: [0, 5000]
+  cal_metric_during_train: False
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: 
+  # for data or label process
+  character_dict_path: ppocr/utils/dict90.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  rm_symbol: True
+  save_res_path: ./output/rec/predicts_satrn.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Piecewise
+    decay_epochs: [3, 4]
+    values: [0.0003, 0.00003, 0.000003] 
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: SATRN
+  Backbone:
+    name: ShallowCNN
+    in_channels: 3
+    hidden_dim: 256
+  Head:
+    name: SATRNHead
+    enc_cfg:
+      n_layers: 6
+      n_head: 8
+      d_k: 32
+      d_v: 32
+      d_model: 256
+      n_position: 100
+      d_inner: 1024
+      dropout: 0.1
+    dec_cfg:
+      n_layers: 6
+      d_embedding: 256
+      n_head: 8
+      d_model: 256
+      d_inner: 1024
+      d_k: 32
+      d_v: 32
+      max_seq_len: 25
+      start_idx: 91
+
+Loss:
+  name: SATRNLoss
+
+PostProcess:
+  name: SATRNLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SATRNLabelEncode: # Class handling label
+      - SVTRRecResizeImg:
+          image_shape: [3, 32, 100] 
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'valid_ratio'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 128
+    drop_last: True
+    num_workers: 8
+    use_shared_memory: False
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/evaluation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SATRNLabelEncode: # Class handling label
+      - SVTRRecResizeImg:
+          image_shape: [3, 32, 100] 
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'valid_ratio'] # dataloader will return list in this order
+  
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 128 
+    num_workers: 4
+    use_shared_memory: False
diff --git a/configs/rec/rec_svtrnet.yml b/configs/rec/rec_svtrnet.yml
new file mode 100644
index 0000000..4657c5c
--- /dev/null
+++ b/configs/rec/rec_svtrnet.yml
@@ -0,0 +1,118 @@
+Global:
+  use_gpu: True
+  epoch_num: 20
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/svtr/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations after the 0th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  character_type: en
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_svtr_tiny.txt
+  d2s_train_image_shape: [3, 64, 256]
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.99
+  epsilon: 1.e-8
+  weight_decay: 0.05
+  no_weight_decay_name: norm pos_embed
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 2
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR
+  Transform:
+    name: STN_ON
+    tps_inputsize: [32, 64]
+    tps_outputsize: [32, 100]
+    num_control_points: 20
+    tps_margins: [0.05,0.05]
+    stn_activation: none
+  Backbone:
+    name: SVTRNet
+    img_size: [32, 100]
+    out_char_num: 25 # W//4 or W//8 or W/12
+    out_channels: 192
+    patch_merging: 'Conv'
+    embed_dim: [64, 128, 256]
+    depth: [3, 6, 3]
+    num_heads: [2, 4, 8]
+    mixer: ['Local','Local','Local','Local','Local','Local','Global','Global','Global','Global','Global','Global']
+    local_mixer: [[7, 11], [7, 11], [7, 11]]
+    last_stage: True
+    prenorm: False
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SVTRRecAug:
+          aug_type: 0 # or 1
+      - CTCLabelEncode: # Class handling label
+      - SVTRRecResizeImg:
+          image_shape: [3, 64, 256]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 512
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/evaluation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - SVTRRecResizeImg:
+          image_shape: [3, 64, 256]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 2
diff --git a/configs/rec/rec_svtrnet_ch.yml b/configs/rec/rec_svtrnet_ch.yml
new file mode 100644
index 0000000..bb82cfc
--- /dev/null
+++ b/configs/rec/rec_svtrnet_ch.yml
@@ -0,0 +1,156 @@
+Global:
+  use_gpu: true
+  epoch_num: 100
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/svtr_ch_all/
+  save_epoch_step: 10
+  eval_batch_step:
+  - 0
+  - 2000
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: 25
+  infer_mode: false
+  use_space_char: true
+  save_res_path: ./output/rec/predicts_svtr_tiny_ch_all.txt
+  d2s_train_image_shape: [3, 32, 320]
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.99
+  epsilon: 1.0e-08
+  weight_decay: 0.05
+  no_weight_decay_name: norm pos_embed
+  one_dim_param_no_weight_decay: true
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 2
+Architecture:
+  model_type: rec
+  algorithm: SVTR
+  Transform: null
+  Backbone:
+    name: SVTRNet
+    img_size:
+    - 32
+    - 320
+    out_char_num: 40 # W//4 or W//8 or W/12
+    out_channels: 96
+    patch_merging: Conv
+    embed_dim:
+    - 64
+    - 128
+    - 256
+    depth:
+    - 3
+    - 6
+    - 3
+    num_heads:
+    - 2
+    - 4
+    - 8
+    mixer:
+    - Local
+    - Local
+    - Local
+    - Local
+    - Local
+    - Local
+    - Global
+    - Global
+    - Global
+    - Global
+    - Global
+    - Global
+    local_mixer:
+    - - 7
+      - 11
+    - - 7
+      - 11
+    - - 7
+      - 11
+    last_stage: true
+    prenorm: false
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+Loss:
+  name: CTCLoss
+PostProcess:
+  name: CTCLabelDecode
+Metric:
+  name: RecMetric
+  main_indicator: acc
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/train_list.txt
+    ext_op_transform_idx: 1
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape:
+        - 32
+        - 320
+        - 3
+    - RecAug: null
+    - CTCLabelEncode: null
+    - SVTRRecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+        padding: true
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: true
+    batch_size_per_card: 256
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data
+    label_file_list:
+    - ./train_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - CTCLabelEncode: null
+    - SVTRRecResizeImg:
+        image_shape:
+        - 3
+        - 32
+        - 320
+        padding: true
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 256
+    num_workers: 2
+profiler_options: null
diff --git a/configs/rec/rec_svtrnet_cppd_base_ch.yml b/configs/rec/rec_svtrnet_cppd_base_ch.yml
new file mode 100644
index 0000000..e2ed04c
--- /dev/null
+++ b/configs/rec/rec_svtrnet_cppd_base_ch.yml
@@ -0,0 +1,113 @@
+Global:
+  use_gpu: True
+  epoch_num: 100
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/svtr_cppd_base_ch/
+  save_epoch_step: 10
+  # evaluation is run every 2000 iterations after the 0th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_svtr_cppd_base_ch.txt
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.99
+  epsilon: 1.e-8
+  weight_decay: 0.05
+  no_weight_decay_name: norm pos_embed char_node_embed pos_node_embed char_pos_embed vis_pos_embed
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 0.0005 # 4gpus 128bs
+    warmup_epoch: 5
+
+Architecture:
+  model_type: rec
+  algorithm: CPPD
+  Transform:
+  Backbone:
+    name: SVTRNet
+    img_size: [32, 256]
+    patch_merging: 'Conv'
+    embed_dim: [128, 256, 384]
+    depth: [6, 6, 4]
+    num_heads: [4, 8, 12]
+    mixer: ['Conv','Conv','Conv','Conv','Conv','Conv', 'Conv','Conv', 'Global','Global','Global','Global','Global','Global','Global','Global','Global','Global']
+    local_mixer: [[5, 5], [5, 5], [5, 5]]
+    last_stage: False
+    prenorm: True
+  Head:
+    name: CPPDHead
+    dim: 384
+    vis_seq: 128
+    ch: &ch True
+
+Loss:
+  name: CPPDLoss
+  ignore_index: &ignore_index 7000 # must be greater than the number of character classes
+  smoothing: True
+  sideloss_weight: 1.0
+
+PostProcess:
+  name: CPPDLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CPPDLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+          ch: *ch
+      - SVTRRecResizeImg:
+          image_shape: [3, 32, 256]
+          padding: True
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'label_node', 'label_index', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/evaluation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CPPDLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+          ch: *ch
+      - SVTRRecResizeImg:
+          image_shape: [3, 32, 256]
+          padding: True
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'label_node','label_index','length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 2
diff --git a/configs/rec/rec_svtrnet_cppd_base_en.yml b/configs/rec/rec_svtrnet_cppd_base_en.yml
new file mode 100644
index 0000000..99885fb
--- /dev/null
+++ b/configs/rec/rec_svtrnet_cppd_base_en.yml
@@ -0,0 +1,112 @@
+Global:
+  use_gpu: True
+  epoch_num: 20
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/svtr_cppd_base/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations after the 0th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  character_type: en
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_svtr_cppd_base.txt
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.99
+  epsilon: 1.e-8
+  weight_decay: 0.05
+  no_weight_decay_name: norm pos_embed char_node_embed pos_node_embed char_pos_embed vis_pos_embed
+  one_dim_param_no_weight_decay: True
+  lr:
+    name: Cosine
+    learning_rate: 0.0005 # 4gpus 256bs
+    warmup_epoch: 2
+
+Architecture:
+  model_type: rec
+  algorithm: CPPD
+  Transform:
+  Backbone:
+    name: SVTRNet
+    img_size: [32, 100]
+    patch_merging: 'Conv'
+    embed_dim: [128, 256, 384]
+    depth: [6, 6, 4]
+    num_heads: [4, 8, 12]
+    mixer: ['Conv','Conv','Conv','Conv','Conv','Conv', 'Conv','Conv', 'Global','Global','Global','Global','Global','Global','Global','Global','Global','Global']
+    local_mixer: [[5, 5], [5, 5], [5, 5]]
+    last_stage: False
+    prenorm: True
+  Head:
+    name: CPPDHead
+    dim: 384
+    vis_seq: 50
+    num_layer: 3
+
+Loss:
+  name: CPPDLoss
+  ignore_index: &ignore_index 100 # must be greater than the number of character classes
+  smoothing: True
+  sideloss_weight: 1.0
+
+PostProcess:
+  name: CPPDLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CPPDLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - SVTRRecResizeImg:
+          image_shape: [3, 32, 100]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'label_node', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/evaluation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CPPDLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - SVTRRecResizeImg:
+          image_shape: [3, 32, 100]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'label_node','length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 2
diff --git a/configs/rec/rec_vit_parseq.yml b/configs/rec/rec_vit_parseq.yml
new file mode 100644
index 0000000..8ba99e6
--- /dev/null
+++ b/configs/rec/rec_vit_parseq.yml
@@ -0,0 +1,116 @@
+Global:
+  use_gpu: True
+  epoch_num: 20
+  log_smooth_window: 20
+  print_batch_step: 5
+  save_model_dir: ./output/rec/parseq
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 500]
+  cal_metric_during_train: True
+  pretrained_model: 
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path: ppocr/utils/dict/parseq_dict.txt
+  character_type: en
+  max_text_length: 25
+  num_heads: 8
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_parseq.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: OneCycle
+    max_lr: 0.0007
+
+Architecture:
+  model_type: rec
+  algorithm: ParseQ
+  in_channels: 3
+  Transform:
+  Backbone:
+    name: ViTParseQ
+    img_size: [32, 128]
+    patch_size: [4, 8]
+    embed_dim: 384
+    depth: 12
+    num_heads: 6
+    mlp_ratio: 4
+    in_channels: 3
+  Head:
+    name: ParseQHead
+    # Architecture
+    max_text_length: 25
+    embed_dim: 384
+    dec_num_heads: 12
+    dec_mlp_ratio: 4
+    dec_depth: 1
+    # Training
+    perm_num: 6
+    perm_forward: true
+    perm_mirrored: true
+    dropout: 0.1
+    # Decoding mode (test)
+    decode_ar: true
+    refine_iters: 1
+
+Loss:
+  name: ParseQLoss
+
+PostProcess:
+  name: ParseQLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  is_filter: True
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir:
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ParseQRecAug:
+          aug_type: 0 # or 1
+      - ParseQLabelEncode:
+      - SVTRRecResizeImg:
+          image_shape: [3, 32, 128]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 192
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir:
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ParseQLabelEncode: # Class handling label
+      - SVTRRecResizeImg:
+          image_shape: [3, 32, 128]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] 
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 384
+    num_workers: 4
diff --git a/configs/rec/rec_vitstr_none_ce.yml b/configs/rec/rec_vitstr_none_ce.yml
new file mode 100644
index 0000000..ebe304f
--- /dev/null
+++ b/configs/rec/rec_vitstr_none_ce.yml
@@ -0,0 +1,102 @@
+Global:
+  use_gpu: True
+  epoch_num: 20
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/vitstr_none_ce/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations after the 0th iteration#
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path: ppocr/utils/EN_symbol_dict.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_vitstr.txt
+
+
+Optimizer:
+  name: Adadelta
+  epsilon: 1.e-8
+  rho: 0.95
+  clip_norm: 5.0
+  lr:
+    learning_rate: 1.0
+
+Architecture:
+  model_type: rec
+  algorithm: ViTSTR
+  in_channels: 1
+  Transform:
+  Backbone:
+    name: ViTSTR
+    scale: tiny
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+
+Loss:
+  name: CELoss
+  with_all: True
+  ignore_index: &ignore_index 0 # Must be zero or greater than the number of character classes
+
+PostProcess:
+  name: ViTSTRLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/training/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ViTSTRLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - GrayRecResizeImg:
+          image_shape: [224, 224] # W H
+          resize_type: PIL # PIL or OpenCV
+          inter_type: 'Image.BICUBIC'
+          scale: false
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 48
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSet
+    data_dir: ./train_data/data_lmdb_release/evaluation/
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ViTSTRLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - GrayRecResizeImg:
+          image_shape: [224, 224] # W H
+          resize_type: PIL # PIL or OpenCV
+          inter_type: 'Image.BICUBIC'
+          scale: false
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 2
diff --git a/configs/sr/sr_telescope.yml b/configs/sr/sr_telescope.yml
new file mode 100644
index 0000000..ed25725
--- /dev/null
+++ b/configs/sr/sr_telescope.yml
@@ -0,0 +1,83 @@
+Global:
+  use_gpu: true
+  epoch_num: 100
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/sr/sr_telescope/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: False
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:  ./output/sr/sr_telescope/infer
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_52.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: 100
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/sr/predicts_telescope.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.5
+  beta2: 0.999
+  clip_norm: 0.25
+  lr:
+    learning_rate: 0.0001
+
+Architecture:
+  model_type: sr
+  algorithm: Telescope
+  Transform:
+    name: TBSRN
+    STN: True
+    infer_mode: False
+
+Loss:
+  name: TelescopeLoss
+  confuse_dict_path: ./ppocr/utils/dict/confuse.pkl
+
+
+PostProcess:
+  name: None
+
+Metric:
+  name: SRMetric
+  main_indicator: all
+
+Train:
+  dataset:
+    name: LMDBDataSetSR
+    data_dir: ./train_data/TextZoom/train
+    transforms:
+      - SRResize:
+          imgH: 32
+          imgW: 128
+          down_sample_scale: 2
+      - KeepKeys:
+          keep_keys: ['img_lr', 'img_hr', 'label'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    batch_size_per_card: 16
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: LMDBDataSetSR
+    data_dir: ./train_data/TextZoom/test
+    transforms:
+      - SRResize:
+          imgH: 32
+          imgW: 128
+          down_sample_scale: 2
+      - KeepKeys:
+          keep_keys: ['img_lr', 'img_hr', 'label'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 16
+    num_workers: 4
diff --git a/configs/sr/sr_tsrn_transformer_strock.yml b/configs/sr/sr_tsrn_transformer_strock.yml
new file mode 100644
index 0000000..627bf24
--- /dev/null
+++ b/configs/sr/sr_tsrn_transformer_strock.yml
@@ -0,0 +1,84 @@
+Global:
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/sr/sr_tsrn_transformer_strock/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: False
+  pretrained_model:
+  checkpoints:
+  save_inference_dir: sr_output
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_52.png
+  # for data or label process
+  character_dict_path: ./train_data/srdata/english_decomposition.txt
+  max_text_length: 100
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/sr/predicts_gestalt.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.5
+  beta2: 0.999
+  clip_norm: 0.25
+  lr:
+    learning_rate: 0.0001
+
+Architecture:
+  model_type: sr
+  algorithm: Gestalt
+  Transform:
+    name: TSRN
+    STN: True
+    infer_mode: False
+
+Loss:
+  name: StrokeFocusLoss
+  character_dict_path: ./train_data/srdata/english_decomposition.txt
+
+PostProcess:
+  name: None
+
+Metric:
+  name: SRMetric
+  main_indicator: all
+
+Train:
+  dataset:
+    name: LMDBDataSetSR
+    data_dir: ./train_data/srdata/train
+    transforms:
+      - SRResize:
+          imgH: 32
+          imgW: 128
+          down_sample_scale: 2
+      - SRLabelEncode: # Class handling label
+      - KeepKeys:
+          keep_keys: ['img_lr', 'img_hr', 'length', 'input_tensor', 'label'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    batch_size_per_card: 16
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: LMDBDataSetSR
+    data_dir: ./train_data/srdata/test
+    transforms:
+      - SRResize:
+          imgH: 32
+          imgW: 128
+          down_sample_scale: 2
+      - SRLabelEncode: # Class handling label
+      - KeepKeys:
+          keep_keys: ['img_lr', 'img_hr','length', 'input_tensor', 'label'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 16
+    num_workers: 4
diff --git a/configs/table/SLANeXt_wired.yml b/configs/table/SLANeXt_wired.yml
new file mode 100644
index 0000000..b6d5f6a
--- /dev/null
+++ b/configs/table/SLANeXt_wired.yml
@@ -0,0 +1,181 @@
+Global:
+  model_name: SLANeXt_wired # To use static model for inference.
+  use_gpu: true
+  epoch_num: 400
+  log_smooth_window: 20
+  print_batch_step: 20
+  save_model_dir: ./output/SLANeXt_wired
+  save_epoch_step: 400
+  eval_batch_step:
+  - 0
+  - 331
+  cal_metric_during_train: true
+  pretrained_model: "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wired_pretrained.pdparams"
+  checkpoints: null
+  save_inference_dir: ./output/SLANeXt_wired/infer
+  use_visualdl: false
+  infer_img: ppstructure/docs/table/table.jpg
+  character_dict_path: ppocr/utils/dict/table_structure_dict_ch.txt
+  character_type: en
+  max_text_length: 500
+  box_format: xyxyxyxy
+  infer_mode: false
+  use_sync_bn: true
+  save_res_path: output/infer
+  d2s_train_image_shape: [3, 512, 512]
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  clip_norm: 5.0
+  lr:
+    name: Cosine
+    learning_rate: 0.0001
+    warmup_epoch: 1
+  regularizer:
+    name: L2
+    factor: 0.0
+
+Architecture:
+  model_type: table
+  algorithm: SLANeXt
+  Backbone:
+    name: Vary_VIT_B
+    image_size: 512 
+    encoder_embed_dim: 768
+    encoder_depth: 12
+    encoder_num_heads: 12
+    encoder_global_attn_indexes: [2, 5, 8, 11]
+  Head:
+    name: SLAHead
+    hidden_size: 512
+    max_text_length: 500
+    loc_reg_num: 8
+
+Loss:
+  name: SLALoss
+  structure_weight: 1.0
+  # SLANeXt does not train the cell location task by default, set the loc_weight if needed.
+  loc_weight: 0.0
+  loc_loss: smooth_l1
+
+PostProcess:
+  name: TableLabelDecode
+  merge_no_span_structure: true
+
+Metric:
+  name: TableMetric
+  main_indicator: acc
+  compute_bbox_metric: false
+  loc_reg_num: 8
+  box_format: xyxyxyxy
+  del_thead_tbody: true
+
+Train:
+  dataset:
+    name: PubTabDataSet
+    data_dir: train_data/table/train/
+    label_file_list:
+    - train_data/table/train.txt
+    ratio_list:
+    - 1
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - TableLabelEncode:
+        learn_empty_box: false
+        merge_no_span_structure: true
+        replace_empty_cell_token: false
+        loc_reg_num: 8
+        max_text_length: 500
+    - TableBoxEncode:
+        in_box_format: xyxyxyxy
+        out_box_format: xyxyxyxy
+    - ResizeTableImage:
+        max_len: 512
+        resize_bboxes: true
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - PaddingTableImage:
+        size:
+        - 512
+        - 512
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - structure
+        - bboxes
+        - bbox_masks
+        - length
+        - shape
+  loader:
+    shuffle: true
+    batch_size_per_card: 48
+    drop_last: true
+    num_workers: 1
+
+Eval:
+  dataset:
+    name: PubTabDataSet
+    data_dir: train_data/table/val/
+    label_file_list:
+    - train_data/table/val.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - TableLabelEncode:
+        learn_empty_box: false
+        merge_no_span_structure: true
+        replace_empty_cell_token: false
+        loc_reg_num: 8
+        max_text_length: 500
+    - TableBoxEncode:
+        in_box_format: xyxyxyxy
+        out_box_format: xyxyxyxy
+    - ResizeTableImage:
+        max_len: 512
+        resize_bboxes: true
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - PaddingTableImage:
+        size:
+        - 512
+        - 512
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - structure
+        - bboxes
+        - bbox_masks
+        - length
+        - shape
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 48
+    num_workers: 1
+
+profiler_options: null
diff --git a/configs/table/SLANeXt_wireless.yml b/configs/table/SLANeXt_wireless.yml
new file mode 100644
index 0000000..9d18ba9
--- /dev/null
+++ b/configs/table/SLANeXt_wireless.yml
@@ -0,0 +1,181 @@
+Global:
+  model_name: SLANeXt_wireless # To use static model for inference.
+  use_gpu: true
+  epoch_num: 400
+  log_smooth_window: 20
+  print_batch_step: 20
+  save_model_dir: ./output/SLANeXt_wireless
+  save_epoch_step: 400
+  eval_batch_step:
+  - 0
+  - 331
+  cal_metric_during_train: true
+  pretrained_model: "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wireless_pretrained.pdparams"
+  checkpoints: null
+  save_inference_dir: ./output/SLANeXt_wireless/infer
+  use_visualdl: false
+  infer_img: ppstructure/docs/table/table.jpg
+  character_dict_path: ppocr/utils/dict/table_structure_dict_ch.txt
+  character_type: en
+  max_text_length: 500
+  box_format: xyxyxyxy
+  infer_mode: false
+  use_sync_bn: true
+  save_res_path: output/infer
+  d2s_train_image_shape: [3, 512, 512]
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  clip_norm: 5.0
+  lr:
+    name: Cosine
+    learning_rate: 0.0001
+    warmup_epoch: 1
+  regularizer:
+    name: L2
+    factor: 0.0
+
+Architecture:
+  model_type: table
+  algorithm: SLANeXt
+  Backbone:
+    name: Vary_VIT_B
+    image_size: 512 
+    encoder_embed_dim: 768
+    encoder_depth: 12
+    encoder_num_heads: 12
+    encoder_global_attn_indexes: [2, 5, 8, 11]
+  Head:
+    name: SLAHead
+    hidden_size: 512
+    max_text_length: 500
+    loc_reg_num: 8
+
+Loss:
+  name: SLALoss
+  structure_weight: 1.0
+  # SLANeXt does not train the cell location task by default, set the loc_weight if needed.
+  loc_weight: 0.0
+  loc_loss: smooth_l1
+
+PostProcess:
+  name: TableLabelDecode
+  merge_no_span_structure: true
+
+Metric:
+  name: TableMetric
+  main_indicator: acc
+  compute_bbox_metric: false
+  loc_reg_num: 8
+  box_format: xyxyxyxy
+  del_thead_tbody: true
+
+Train:
+  dataset:
+    name: PubTabDataSet
+    data_dir: train_data/table/train/
+    label_file_list:
+    - train_data/table/train.txt
+    ratio_list:
+    - 1
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - TableLabelEncode:
+        learn_empty_box: false
+        merge_no_span_structure: true
+        replace_empty_cell_token: false
+        loc_reg_num: 8
+        max_text_length: 500
+    - TableBoxEncode:
+        in_box_format: xyxyxyxy
+        out_box_format: xyxyxyxy
+    - ResizeTableImage:
+        max_len: 512
+        resize_bboxes: true
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - PaddingTableImage:
+        size:
+        - 512
+        - 512
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - structure
+        - bboxes
+        - bbox_masks
+        - length
+        - shape
+  loader:
+    shuffle: true
+    batch_size_per_card: 48
+    drop_last: true
+    num_workers: 1
+
+Eval:
+  dataset:
+    name: PubTabDataSet
+    data_dir: train_data/table/val/
+    label_file_list:
+    - train_data/table/val.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - TableLabelEncode:
+        learn_empty_box: false
+        merge_no_span_structure: true
+        replace_empty_cell_token: false
+        loc_reg_num: 8
+        max_text_length: 500
+    - TableBoxEncode:
+        in_box_format: xyxyxyxy
+        out_box_format: xyxyxyxy
+    - ResizeTableImage:
+        max_len: 512
+        resize_bboxes: true
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - PaddingTableImage:
+        size:
+        - 512
+        - 512
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - structure
+        - bboxes
+        - bbox_masks
+        - length
+        - shape
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 48
+    num_workers: 1
+
+profiler_options: null
diff --git a/configs/table/SLANet.yml b/configs/table/SLANet.yml
new file mode 100644
index 0000000..05da5a9
--- /dev/null
+++ b/configs/table/SLANet.yml
@@ -0,0 +1,146 @@
+Global:
+  model_name: SLANet # To use static model for inference.
+  use_gpu: true
+  epoch_num: 100
+  log_smooth_window: 20
+  print_batch_step: 20
+  save_model_dir: ./output/SLANet
+  save_epoch_step: 400
+  # evaluation is run every 1000 iterations after the 0th iteration
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: True
+  pretrained_model: "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_pretrained.pdparams"
+  checkpoints:
+  save_inference_dir: ./output/SLANet/infer
+  use_visualdl: False
+  infer_img: ppstructure/docs/table/table.jpg
+  # for data or label process
+  character_dict_path: ppocr/utils/dict/table_structure_dict.txt
+  character_type: en
+  max_text_length: &max_text_length 500
+  box_format: &box_format 'xyxy' # 'xywh', 'xyxy', 'xyxyxyxy'
+  infer_mode: False
+  use_sync_bn: True
+  save_res_path: 'output/infer'
+  d2s_train_image_shape: [3, -1, -1]
+  amp_custom_white_list: ['concat', 'elementwise_sub', 'set_value']
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  clip_norm: 5.0
+  lr:
+    name: Piecewise
+    learning_rate: 0.001
+    decay_epochs : [40, 50]
+    values : [0.001, 0.0001, 0.00005]
+  regularizer:
+    name: 'L2'
+    factor: 0.00000
+
+Architecture:
+  model_type: table
+  algorithm: SLANet
+  Backbone:
+    name: PPLCNet
+    scale: 1.0
+    pretrained: true
+    use_ssld: true
+  Neck:
+    name: CSPPAN
+    out_channels: 96
+  Head:
+    name: SLAHead
+    hidden_size: 256
+    max_text_length: *max_text_length
+    loc_reg_num: &loc_reg_num 4
+
+Loss:
+  name: SLALoss
+  structure_weight: 1.0
+  loc_weight: 2.0
+  loc_loss: smooth_l1
+
+PostProcess:
+  name: TableLabelDecode
+  merge_no_span_structure: &merge_no_span_structure True
+
+Metric:
+  name: TableMetric
+  main_indicator: acc
+  compute_bbox_metric: False
+  loc_reg_num: *loc_reg_num
+  box_format: *box_format
+
+Train:
+  dataset:
+    name: PubTabDataSet
+    data_dir: train_data/table/pubtabnet/train/
+    label_file_list: [train_data/table/pubtabnet/PubTabNet_2.0.0_train.jsonl]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - TableLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: *merge_no_span_structure
+          replace_empty_cell_token: False
+          loc_reg_num: *loc_reg_num
+          max_text_length: *max_text_length
+      - TableBoxEncode:
+          in_box_format: *box_format
+          out_box_format: *box_format
+      - ResizeTableImage:
+          max_len: 488
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - PaddingTableImage:
+          size: [488, 488]
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'structure', 'bboxes', 'bbox_masks', 'length', 'shape']
+  loader:
+    shuffle: True
+    batch_size_per_card: 48
+    drop_last: True
+    num_workers: 1
+
+Eval:
+  dataset:
+    name: PubTabDataSet
+    data_dir: train_data/table/pubtabnet/val/
+    label_file_list: [train_data/table/pubtabnet/PubTabNet_2.0.0_val.jsonl]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - TableLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: *merge_no_span_structure
+          replace_empty_cell_token: False
+          loc_reg_num: *loc_reg_num
+          max_text_length: *max_text_length
+      - TableBoxEncode:
+          in_box_format: *box_format
+          out_box_format: *box_format
+      - ResizeTableImage:
+          max_len: 488
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - PaddingTableImage:
+          size: [488, 488]
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'structure', 'bboxes', 'bbox_masks', 'length', 'shape']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 48
+    num_workers: 1
diff --git a/configs/table/SLANet_lcnetv2.yml b/configs/table/SLANet_lcnetv2.yml
new file mode 100644
index 0000000..df11acf
--- /dev/null
+++ b/configs/table/SLANet_lcnetv2.yml
@@ -0,0 +1,139 @@
+Global:
+  use_gpu: true
+  epoch_num: 50
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/SLANet_lcnetv2
+  save_epoch_step: 20
+  # evaluation is run every 1000 iterations after the 0th iteration
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir: ./SLANet_lcnetv2_infer
+  use_visualdl: False
+  infer_img: ppstructure/docs/table/table.jpg
+  # for data or label process
+  character_dict_path: ppocr/utils/dict/table_structure_dict.txt
+  character_type: en
+  max_text_length: &max_text_length 500
+  box_format: &box_format 'xyxy' # 'xywh', 'xyxy', 'xyxyxyxy'
+  infer_mode: False
+  use_sync_bn: True
+  save_res_path: 'output/infer'
+  d2s_train_image_shape: [3, -1, -1]
+  amp_custom_white_list: ['concat', 'elementwise_sub', 'set_value']
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  clip_norm: 5.0
+  lr:
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0.00000
+
+Architecture:
+  model_type: table
+  algorithm: SLANet
+  Backbone:
+    name: PPLCNetV2_base
+  Neck:
+    name: CSPPAN
+    out_channels: 96
+  Head:
+    name: SLAHead
+    hidden_size: 256
+    max_text_length: *max_text_length
+    loc_reg_num: &loc_reg_num 4
+
+Loss:
+  name: SLALoss
+  structure_weight: 1.0
+  loc_weight: 2.0
+  loc_loss: smooth_l1
+
+PostProcess:
+  name: TableLabelDecode
+  merge_no_span_structure: &merge_no_span_structure True
+
+Metric:
+  name: TableMetric
+  main_indicator: acc
+  compute_bbox_metric: False
+  loc_reg_num: *loc_reg_num
+  box_format: *box_format
+
+Train:
+  dataset:
+    name: PubTabDataSet
+    data_dir: ../table_data/pubtabnet/train/
+    label_file_list: [../table_data/pubtabnet/PubTabNet_2.0.0_train.jsonl]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - TableLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: *merge_no_span_structure
+          replace_empty_cell_token: False
+          loc_reg_num: *loc_reg_num
+          max_text_length: *max_text_length
+      - TableBoxEncode:
+          in_box_format: *box_format
+          out_box_format: *box_format
+      - ResizeTableImage:
+          max_len: 488
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - PaddingTableImage:
+          size: [488, 488]
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'image', 'structure', 'bboxes', 'bbox_masks', 'length', 'shape' ]
+  loader:
+    shuffle: True
+    batch_size_per_card: 24
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: PubTabDataSet
+    data_dir: ../table_data/pubtabnet/val/
+    label_file_list: [../table_data/pubtabnet/PubTabNet_2.0.0_val.jsonl]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - TableLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: *merge_no_span_structure
+          replace_empty_cell_token: False
+          loc_reg_num: *loc_reg_num
+          max_text_length: *max_text_length
+      - TableBoxEncode:
+          in_box_format: *box_format
+          out_box_format: *box_format
+      - ResizeTableImage:
+          max_len: 488
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - PaddingTableImage:
+          size: [488, 488]
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'image', 'structure', 'bboxes', 'bbox_masks', 'shape' ]
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 48
+    num_workers: 4
diff --git a/configs/table/SLANet_plus.yml b/configs/table/SLANet_plus.yml
new file mode 100644
index 0000000..e67ce66
--- /dev/null
+++ b/configs/table/SLANet_plus.yml
@@ -0,0 +1,143 @@
+Global:
+  model_name: SLANet_plus # To use static model for inference.
+  use_gpu: True
+  epoch_num: 400
+  log_smooth_window: 20
+  print_batch_step: 20
+  save_model_dir: ./output/SLANet_plus
+  save_epoch_step: 400
+  # evaluation is run every 331 iterations after the 0th iteration
+  eval_batch_step: [0, 331]
+  cal_metric_during_train: True
+  pretrained_model: "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_plus_pretrained.pdparams"
+  checkpoints: 
+  save_inference_dir: ./output/SLANet_plus/infer
+  use_visualdl: False
+  infer_img: ppstructure/docs/table/table.jpg
+  # for data or label process
+  character_dict_path: ppocr/utils/dict/table_structure_dict_ch.txt
+  character_type: en
+  max_text_length: &max_text_length 500
+  box_format: &box_format xyxyxyxy # 'xywh', 'xyxy', 'xyxyxyxy'
+  infer_mode: False
+  use_sync_bn: True
+  save_res_path: output/infer
+  d2s_train_image_shape: [3, 488, 488]
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  clip_norm: 5.0
+  lr:
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0.00000
+
+Architecture:
+  model_type: table
+  algorithm: SLANet
+  Backbone:
+    name: PPLCNet
+    scale: 1.0
+    pretrained: True
+    use_ssld: True
+  Neck:
+    name: CSPPAN
+    out_channels: 96
+  Head:
+    name: SLAHead
+    hidden_size: 256
+    max_text_length: *max_text_length
+    loc_reg_num: &loc_reg_num 8
+
+Loss:
+  name: SLALoss
+  structure_weight: 1.0
+  loc_weight: 2.0
+  loc_loss: smooth_l1
+
+PostProcess:
+  name: TableLabelDecode
+  merge_no_span_structure: &merge_no_span_structure True
+
+Metric:
+  name: TableMetric
+  main_indicator: acc
+  compute_bbox_metric: False
+  loc_reg_num: *loc_reg_num
+  box_format: *box_format
+  del_thead_tbody: True
+
+Train:
+  dataset:
+    name: PubTabDataSet
+    data_dir: train_data/table/train/
+    label_file_list: [train_data/table/train.txt]
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: False
+      - TableLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: *merge_no_span_structure
+          replace_empty_cell_token: False
+          loc_reg_num: *loc_reg_num
+          max_text_length: *max_text_length
+      - TableBoxEncode:
+          in_box_format: *box_format
+          out_box_format: *box_format
+      - ResizeTableImage:
+          max_len: 488
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - PaddingTableImage:
+          size: [488, 488]
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'structure', 'bboxes', 'bbox_masks', 'length', 'shape']
+  loader:
+    shuffle: True
+    batch_size_per_card: 48
+    drop_last: True
+    num_workers: 1
+
+Eval:
+  dataset:
+    name: PubTabDataSet
+    data_dir: train_data/table/val/
+    label_file_list: [train_data/table/val.txt]
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: False
+      - TableLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: *merge_no_span_structure
+          replace_empty_cell_token: False
+          loc_reg_num: *loc_reg_num
+          max_text_length: *max_text_length
+      - TableBoxEncode:
+          in_box_format: *box_format
+          out_box_format: *box_format
+      - ResizeTableImage:
+          max_len: 488
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - PaddingTableImage:
+          size: [488, 488]
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'structure', 'bboxes', 'bbox_masks', 'length', 'shape']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 48
+    num_workers: 1
diff --git a/configs/table/table_master.yml b/configs/table/table_master.yml
new file mode 100755
index 0000000..37cf44d
--- /dev/null
+++ b/configs/table/table_master.yml
@@ -0,0 +1,145 @@
+Global:
+  use_gpu: true
+  epoch_num: 17
+  log_smooth_window: 20
+  print_batch_step: 100
+  save_model_dir: ./output/table_master/
+  save_epoch_step: 17
+  eval_batch_step: [0,  6259]
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints:
+  save_inference_dir: output/table_master/infer
+  use_visualdl: false
+  infer_img: ppstructure/docs/table/table.jpg
+  save_res_path: ./output/table_master
+  character_dict_path: ppocr/utils/dict/table_master_structure_dict.txt
+  infer_mode: false
+  max_text_length: &max_text_length 500
+  box_format: &box_format 'xywh' # 'xywh', 'xyxy', 'xyxyxyxy'
+  d2s_train_image_shape: [3, 480, 480]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: MultiStepDecay
+    learning_rate: 0.001
+    milestones: [12, 15]
+    gamma: 0.1
+    warmup_epoch: 0.02
+  regularizer:
+    name: L2
+    factor: 0.0
+
+Architecture:
+  model_type: table
+  algorithm: TableMaster
+  Backbone:
+    name: TableResNetExtra
+    gcb_config:
+      ratio: 0.0625
+      headers: 1
+      att_scale: False
+      fusion_type: channel_add
+      layers: [False, True, True, True]
+    layers: [1,2,5,3]
+  Head:
+    name: TableMasterHead
+    hidden_size: 512
+    headers: 8
+    dropout: 0
+    d_ff: 2024
+    max_text_length: *max_text_length
+    loc_reg_num: &loc_reg_num 4
+
+Loss:
+  name: TableMasterLoss
+  ignore_index: 42 # set to len of dict + 3
+
+PostProcess:
+  name: TableMasterLabelDecode
+  box_shape: pad
+  merge_no_span_structure: &merge_no_span_structure True
+
+Metric:
+  name: TableMetric
+  main_indicator: acc
+  compute_bbox_metric: False
+  box_format: *box_format
+
+Train:
+  dataset:
+    name: PubTabDataSet
+    data_dir: train_data/table/pubtabnet/train/
+    label_file_list: [train_data/table/pubtabnet/PubTabNet_2.0.0_train.jsonl]
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: False
+      - TableMasterLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: *merge_no_span_structure
+          replace_empty_cell_token: True
+          loc_reg_num: *loc_reg_num
+          max_text_length: *max_text_length
+      - ResizeTableImage:
+          max_len: 480
+          resize_bboxes: True
+      - PaddingTableImage:
+          size: [480, 480]
+      - TableBoxEncode:
+          in_box_format: *box_format
+          out_box_format: *box_format
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.5, 0.5, 0.5]
+          std: [0.5, 0.5, 0.5]
+          order: hwc
+      - ToCHWImage: null
+      - KeepKeys:
+          keep_keys: [image, structure, bboxes, bbox_masks, shape]
+  loader:
+    shuffle: True
+    batch_size_per_card: 10
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: PubTabDataSet
+    data_dir: train_data/table/pubtabnet/val/
+    label_file_list: [train_data/table/pubtabnet/PubTabNet_2.0.0_val.jsonl]
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: False
+      - TableMasterLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: *merge_no_span_structure
+          replace_empty_cell_token: True
+          loc_reg_num: *loc_reg_num
+          max_text_length: *max_text_length
+      - ResizeTableImage:
+          max_len: 480
+          resize_bboxes: True
+      - PaddingTableImage:
+          size: [480, 480]
+      - TableBoxEncode:
+          in_box_format: *box_format
+          out_box_format: *box_format
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.5, 0.5, 0.5]
+          std: [0.5, 0.5, 0.5]
+          order: hwc
+      - ToCHWImage: null
+      - KeepKeys:
+          keep_keys: [image, structure, bboxes, bbox_masks, shape]
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 10
+    num_workers: 8
diff --git a/configs/table/table_mv3.yml b/configs/table/table_mv3.yml
new file mode 100755
index 0000000..50c8439
--- /dev/null
+++ b/configs/table/table_mv3.yml
@@ -0,0 +1,129 @@
+Global:
+  use_gpu: true
+  epoch_num: 400
+  log_smooth_window: 20
+  print_batch_step: 5
+  save_model_dir: ./output/table_mv3/
+  save_epoch_step: 400
+  # evaluation is run every 400 iterations after the 0th iteration
+  eval_batch_step: [0, 400]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: ppstructure/docs/table/table.jpg
+  save_res_path: output/table_mv3
+  # for data or label process
+  character_dict_path: ppocr/utils/dict/table_structure_dict.txt
+  character_type: en
+  max_text_length: &max_text_length 500
+  box_format: &box_format 'xyxy' # 'xywh', 'xyxy', 'xyxyxyxy'
+  infer_mode: False
+  amp_custom_black_list: ['matmul_v2','elementwise_add']
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  clip_norm: 5.0
+  lr:
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0.00000
+
+Architecture:
+  model_type: table
+  algorithm: TableAttn
+  Backbone:
+    name: MobileNetV3
+    scale: 1.0
+    model_name: small
+    disable_se: true
+  Head:
+    name: TableAttentionHead
+    hidden_size: 256
+    max_text_length: *max_text_length
+    loc_reg_num: &loc_reg_num 4
+
+Loss:
+  name: TableAttentionLoss
+  structure_weight: 100.0
+  loc_weight: 10000.0
+
+PostProcess:
+  name: TableLabelDecode
+
+Metric:
+  name: TableMetric
+  main_indicator: acc
+  compute_bbox_metric: false # cost many time, set False for training
+
+Train:
+  dataset:
+    name: PubTabDataSet
+    data_dir: train_data/table/pubtabnet/train/
+    label_file_list: [train_data/table/pubtabnet/PubTabNet_2.0.0_train.jsonl]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - TableLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: False
+          replace_empty_cell_token: False
+          loc_reg_num: *loc_reg_num
+          max_text_length: *max_text_length
+      - TableBoxEncode:
+      - ResizeTableImage:
+          max_len: 488
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - PaddingTableImage:
+          size: [488, 488]
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'image', 'structure', 'bboxes', 'bbox_masks', 'shape' ]
+  loader:
+    shuffle: True
+    batch_size_per_card: 48
+    drop_last: True
+    num_workers: 1
+
+Eval:
+  dataset:
+    name: PubTabDataSet
+    data_dir: train_data/table/pubtabnet/val/
+    label_file_list: [train_data/table/pubtabnet/PubTabNet_2.0.0_val.jsonl]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - TableLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: False
+          replace_empty_cell_token: False
+          loc_reg_num: *loc_reg_num
+          max_text_length: *max_text_length
+      - TableBoxEncode:
+      - ResizeTableImage:
+          max_len: 488
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - PaddingTableImage:
+          size: [488, 488]
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'image', 'structure', 'bboxes', 'bbox_masks', 'shape' ]
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 48
+    num_workers: 1
diff --git a/deploy/README.md b/deploy/README.md
new file mode 100644
index 0000000..43ee65e
--- /dev/null
+++ b/deploy/README.md
@@ -0,0 +1,30 @@
+English | [简体中文](README_ch.md)
+
+# PP-OCR Deployment
+
+- [PP-OCR Deployment](#pp-ocr-deployment)
+  - [Paddle Deployment Introduction](#paddle-deployment-introduction)
+  - [PP-OCR Deployment](#pp-ocr-deployment-1)
+
+<a name="1"></a>
+## Paddle Deployment Introduction
+
+Paddle provides a variety of deployment schemes to meet the deployment requirements of different scenarios. Please choose according to the actual situation:
+
+<div align="center">
+    <img src="../doc/deployment_en.png" width="800">
+</div>
+
+
+<a name="2"></a>
+## PP-OCR Deployment
+
+PP-OCR has supported multi deployment schemes. Click the link to get the specific tutorial.
+
+- [Python Inference](../doc/doc_en/inference_ppocr_en.md)
+- [C++ Inference](./cpp_infer/readme.md)
+- [Serving (Python/C++)](./pdserving/README.md)
+- [Paddle-Lite (ARM CPU/OpenCL ARM GPU)](./lite/readme.md)
+- [Paddle2ONNX](./paddle2onnx/readme.md)
+
+If you need the deployment tutorial of academic algorithm models other than PP-OCR, please directly enter the main page of corresponding algorithms, [entrance](../doc/doc_en/algorithm_overview_en.md)。
diff --git a/deploy/README_ch.md b/deploy/README_ch.md
new file mode 100644
index 0000000..97c7bf9
--- /dev/null
+++ b/deploy/README_ch.md
@@ -0,0 +1 @@
+See [link](https://www.paddleocr.ai/main/version3.x/inference_deployment/local_inference/high_performance_inference.html) for details.
diff --git a/deploy/android_demo/.gitignore b/deploy/android_demo/.gitignore
new file mode 100644
index 0000000..d77f574
--- /dev/null
+++ b/deploy/android_demo/.gitignore
@@ -0,0 +1,8 @@
+*.iml
+.gradle
+/local.properties
+/.idea/*
+.DS_Store
+/build
+/captures
+.externalNativeBuild
diff --git a/deploy/android_demo/README.md b/deploy/android_demo/README.md
new file mode 100644
index 0000000..3c59ca6
--- /dev/null
+++ b/deploy/android_demo/README.md
@@ -0,0 +1,118 @@
+- [Android Demo](#android-demo)
+  - [1. 简介](#1-简介)
+  - [2. 近期更新](#2-近期更新)
+  - [3. 快速使用](#3-快速使用)
+    - [3.1 环境准备](#31-环境准备)
+    - [3.2 导入项目](#32-导入项目)
+    - [3.3 运行demo](#33-运行demo)
+    - [3.4 运行模式](#34-运行模式)
+    - [3.5 设置](#35-设置)
+  - [4 更多支持](#4-更多支持)
+
+# Android Demo
+
+## 1. 简介
+此为PaddleOCR的Android Demo，目前支持文本检测，文本方向分类器和文本识别模型的使用。使用 [PaddleLite v2.10](https://github.com/PaddlePaddle/Paddle-Lite/tree/release/v2.10) 进行开发。
+
+## 2. 近期更新
+* 2022.02.27
+    * 预测库更新到PaddleLite v2.10
+    * 支持6种运行模式：
+      * 检测+分类+识别
+      * 检测+识别
+      * 分类+识别
+      * 检测
+      * 识别
+      * 分类
+
+## 3. 快速使用
+
+### 3.1 环境准备
+1. 在本地环境安装好 Android Studio 工具，详细安装方法请见[Android Studio 官网](https://developer.android.com/studio)。
+2. 准备一部 Android 手机，并开启 USB 调试模式。开启方法: `手机设置 -> 查找开发者选项 -> 打开开发者选项和 USB 调试模式`
+
+**注意**：如果您的 Android Studio 尚未配置 NDK ，请根据 Android Studio 用户指南中的[安装及配置 NDK 和 CMake ](https://developer.android.com/studio/projects/install-ndk)内容，预先配置好 NDK 。您可以选择最新的 NDK 版本，或者使用 Paddle Lite 预测库版本一样的 NDK
+
+### 3.2 导入项目
+
+点击 File->New->Import Project...， 然后跟着Android Studio的引导导入
+导入完成后呈现如下界面
+![](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/imgs/import_demo.jpg)
+
+### 3.3 运行demo
+将手机连接上电脑后，点击Android Studio工具栏中的运行按钮即可运行demo。在此过程中，手机会弹出"允许从 USB 安装软件权限"的弹窗，点击允许即可。
+
+软件安转到手机上后会在手机主屏最后一页看到如下app
+<div align="left">
+    <img src="https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/imgs/install_finish.jpeg" width="400">
+</div>
+
+点击app图标即可启动app，启动后app主页如下
+
+<div align="left">
+    <img src="https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/imgs/main_page.jpg" width="400">
+</div>
+
+app主页中有四个按钮，一个下拉列表和一个菜单按钮，他们的功能分别为
+
+* 运行模型：按照已选择的模式，运行对应的模型组合
+* 拍照识别：唤起手机相机拍照并获取拍照的图像，拍照完成后需要点击运行模型进行识别
+* 选取图片：唤起手机相册拍照选择图像，选择完成后需要点击运行模型进行识别
+* 清空绘图：清空当前显示图像上绘制的文本框，以便进行下一次识别(每次识别使用的图像都是当前显示的图像)
+* 下拉列表：进行运行模式的选择，目前包含6种运行模式，默认模式为**检测+分类+识别**详细说明见下一节。
+* 菜单按钮：点击后会进入菜单界面，进行模型和内置图像有关设置
+
+点击运行模型后，会按照所选择的模式运行对应的模型，**检测+分类+识别**模式下运行的模型结果如下所示：
+
+<img src="https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/imgs/run_det_cls_rec.jpg" width="400">
+
+模型运行完成后，模型和运行状态显示区`STATUS`字段显示了当前模型的运行状态，这里显示为`run model succeeded`表明模型运行成功。
+
+模型的运行结果显示在运行结果显示区，显示格式为
+```text
+序号：Det：(x1,y1)(x2,y2)(x3,y3)(x4,y4) Rec: 识别文本,识别置信度 Cls：分类类别,分类分时
+```
+
+### 3.4 运行模式
+
+PaddleOCR demo共提供了6种运行模式，如下图
+<div align="left">
+    <img src="https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/imgs/select_mode.jpg" width="400">
+</div>
+
+每种模式的运行结果如下表所示
+
+| 检测+分类+识别                                                                                       | 检测+识别                                                                                      | 分类+识别                                                                                      |
+|------------------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------|
+| <img src="https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/imgs/run_det_cls_rec.jpg" width="400"> | <img src="https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/imgs/run_det_rec.jpg" width="400"> | <img src="https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/imgs/run_cls_rec.jpg" width="400"> |
+
+
+| 检测                                                                                     | 识别                                                                                     | 分类                                                                                     |
+|----------------------------------------------------------------------------------------|----------------------------------------------------------------------------------------|----------------------------------------------------------------------------------------|
+| <img src="https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/imgs/run_det.jpg" width="400"> | <img src="https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/imgs/run_rec.jpg" width="400"> | <img src="https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/imgs/run_cls.jpg" width="400"> |
+
+### 3.5 设置
+
+设置界面如下
+
+<div align="left">
+    <img src="https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/imgs/settings.jpg" width="400">
+</div>
+
+在设置界面可以进行如下几项设定：
+1. 普通设置
+   * Enable custom settings: 选中状态下才能更改设置
+   * Model Path: 所运行的模型地址，使用默认值就好
+   * Label Path: 识别模型的字典
+   * Image Path: 进行识别的内置图像名
+2. 模型运行态设置，此项设置更改后返回主界面时，会自动重新加载模型
+   * CPU Thread Num: 模型运行使用的CPU核心数量
+   * CPU Power Mode: 模型运行模式，大小核设定
+3. 输入设置
+   * det long size: DB模型预处理时图像的长边长度，超过此长度resize到该值，短边进行等比例缩放，小于此长度不进行处理。
+4. 输出设置
+   * Score Threshold: DB模型后处理box的阈值，低于此阈值的box进行过滤，不显示。
+
+## 4 更多支持
+1. 实时识别，更新预测库可参考 https://github.com/PaddlePaddle/Paddle-Lite-Demo/tree/develop/ocr/android/app/cxx/ppocr_demo
+2. 更多Paddle-Lite相关问题可前往[Paddle-Lite](https://github.com/PaddlePaddle/Paddle-Lite) ，获得更多开发支持
diff --git a/deploy/android_demo/app/.gitignore b/deploy/android_demo/app/.gitignore
new file mode 100644
index 0000000..796b96d
--- /dev/null
+++ b/deploy/android_demo/app/.gitignore
@@ -0,0 +1 @@
+/build
diff --git a/deploy/android_demo/app/build.gradle b/deploy/android_demo/app/build.gradle
new file mode 100644
index 0000000..00fae70
--- /dev/null
+++ b/deploy/android_demo/app/build.gradle
@@ -0,0 +1,93 @@
+import java.security.MessageDigest
+
+apply plugin: 'com.android.application'
+
+android {
+    compileSdkVersion 29
+    defaultConfig {
+        applicationId "com.baidu.paddle.lite.demo.ocr"
+        minSdkVersion 23
+        targetSdkVersion 29
+        versionCode 2
+        versionName "2.0"
+        testInstrumentationRunner "android.support.test.runner.AndroidJUnitRunner"
+        externalNativeBuild {
+            cmake {
+                cppFlags "-std=c++11 -frtti -fexceptions -Wno-format"
+                arguments '-DANDROID_PLATFORM=android-23', '-DANDROID_STL=c++_shared' ,"-DANDROID_ARM_NEON=TRUE"
+            }
+        }
+    }
+    buildTypes {
+        release {
+            minifyEnabled false
+            proguardFiles getDefaultProguardFile('proguard-android-optimize.txt'), 'proguard-rules.pro'
+        }
+    }
+    externalNativeBuild {
+        cmake {
+            path "src/main/cpp/CMakeLists.txt"
+            version "3.10.2"
+        }
+    }
+}
+
+dependencies {
+    implementation fileTree(include: ['*.jar'], dir: 'libs')
+    implementation 'androidx.appcompat:appcompat:1.1.0'
+    implementation 'androidx.constraintlayout:constraintlayout:1.1.3'
+    testImplementation 'junit:junit:4.12'
+    androidTestImplementation 'com.android.support.test:runner:1.0.2'
+    androidTestImplementation 'com.android.support.test.espresso:espresso-core:3.0.2'
+}
+
+def archives = [
+        [
+                'src' : 'https://paddleocr.bj.bcebos.com/libs/paddle_lite_libs_v2_10.tar.gz',
+                'dest': 'PaddleLite'
+        ],
+        [
+                'src' : 'https://paddlelite-demo.bj.bcebos.com/libs/android/opencv-4.2.0-android-sdk.tar.gz',
+                'dest': 'OpenCV'
+        ],
+        [
+                'src' : 'https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2.tar.gz',
+                'dest' : 'src/main/assets/models'
+        ],
+        [
+                'src' : 'https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_dict.tar.gz',
+                'dest' : 'src/main/assets/labels'
+        ]
+]
+
+task downloadAndExtractArchives(type: DefaultTask) {
+    doFirst {
+        println "Downloading and extracting archives including libs and models"
+    }
+    doLast {
+        // Prepare cache folder for archives
+        String cachePath = "cache"
+        if (!file("${cachePath}").exists()) {
+            mkdir "${cachePath}"
+        }
+        archives.eachWithIndex { archive, index ->
+            MessageDigest messageDigest = MessageDigest.getInstance('MD5')
+            messageDigest.update(archive.src.bytes)
+            String cacheName = new BigInteger(1, messageDigest.digest()).toString(32)
+            // Download the target archive if not exists
+            boolean copyFiles = !file("${archive.dest}").exists()
+            if (!file("${cachePath}/${cacheName}.tar.gz").exists()) {
+                ant.get(src: archive.src, dest: file("${cachePath}/${cacheName}.tar.gz"))
+                copyFiles = true; // force to copy files from the latest archive files
+            }
+            // Extract the target archive if its dest path does not exists
+            if (copyFiles) {
+                copy {
+                    from tarTree("${cachePath}/${cacheName}.tar.gz")
+                    into "${archive.dest}"
+                }
+            }
+        }
+    }
+}
+preBuild.dependsOn downloadAndExtractArchives
diff --git a/deploy/android_demo/app/proguard-rules.pro b/deploy/android_demo/app/proguard-rules.pro
new file mode 100644
index 0000000..f1b4245
--- /dev/null
+++ b/deploy/android_demo/app/proguard-rules.pro
@@ -0,0 +1,21 @@
+# Add project specific ProGuard rules here.
+# You can control the set of applied configuration files using the
+# proguardFiles setting in build.gradle.
+#
+# For more details, see
+#   http://developer.android.com/guide/developing/tools/proguard.html
+
+# If your project uses WebView with JS, uncomment the following
+# and specify the fully qualified class name to the JavaScript interface
+# class:
+#-keepclassmembers class fqcn.of.javascript.interface.for.webview {
+#   public *;
+#}
+
+# Uncomment this to preserve the line number information for
+# debugging stack traces.
+#-keepattributes SourceFile,LineNumberTable
+
+# If you keep the line number information, uncomment this to
+# hide the original source file name.
+#-renamesourcefileattribute SourceFile
diff --git a/deploy/android_demo/app/src/androidTest/java/com/baidu/paddle/lite/demo/ocr/ExampleInstrumentedTest.java b/deploy/android_demo/app/src/androidTest/java/com/baidu/paddle/lite/demo/ocr/ExampleInstrumentedTest.java
new file mode 100644
index 0000000..77b179d
--- /dev/null
+++ b/deploy/android_demo/app/src/androidTest/java/com/baidu/paddle/lite/demo/ocr/ExampleInstrumentedTest.java
@@ -0,0 +1,26 @@
+package com.baidu.paddle.lite.demo.ocr;
+
+import android.content.Context;
+import android.support.test.InstrumentationRegistry;
+import android.support.test.runner.AndroidJUnit4;
+
+import org.junit.Test;
+import org.junit.runner.RunWith;
+
+import static org.junit.Assert.*;
+
+/**
+ * Instrumented test, which will execute on an Android device.
+ *
+ * @see <a href="http://d.android.com/tools/testing">Testing documentation</a>
+ */
+@RunWith(AndroidJUnit4.class)
+public class ExampleInstrumentedTest {
+    @Test
+    public void useAppContext() {
+        // Context of the app under test.
+        Context appContext = InstrumentationRegistry.getTargetContext();
+
+        assertEquals("com.baidu.paddle.lite.demo", appContext.getPackageName());
+    }
+}
diff --git a/deploy/android_demo/app/src/main/AndroidManifest.xml b/deploy/android_demo/app/src/main/AndroidManifest.xml
new file mode 100644
index 0000000..fef3a39
--- /dev/null
+++ b/deploy/android_demo/app/src/main/AndroidManifest.xml
@@ -0,0 +1,38 @@
+<?xml version="1.0" encoding="utf-8"?>
+<manifest xmlns:android="http://schemas.android.com/apk/res/android"
+          package="com.baidu.paddle.lite.demo.ocr">
+
+    <uses-permission android:name="android.permission.WRITE_EXTERNAL_STORAGE"/>
+    <uses-permission android:name="android.permission.READ_EXTERNAL_STORAGE"/>
+    <uses-permission android:name="android.permission.CAMERA"/>
+
+
+    <application
+            android:allowBackup="true"
+            android:icon="@mipmap/ic_launcher"
+            android:label="@string/app_name"
+            android:roundIcon="@mipmap/ic_launcher_round"
+            android:supportsRtl="true"
+            android:theme="@style/AppTheme">
+        <activity android:name="com.baidu.paddle.lite.demo.ocr.MainActivity">
+            <intent-filter>
+                <action android:name="android.intent.action.MAIN"/>
+                <category android:name="android.intent.category.LAUNCHER"/>
+            </intent-filter>
+        </activity>
+        <activity
+                android:name="com.baidu.paddle.lite.demo.ocr.SettingsActivity"
+                android:label="Settings">
+        </activity>
+        <provider
+            android:name="androidx.core.content.FileProvider"
+            android:authorities="com.baidu.paddle.lite.demo.ocr.fileprovider"
+            android:exported="false"
+            android:grantUriPermissions="true">
+            <meta-data
+                android:name="android.support.FILE_PROVIDER_PATHS"
+                android:resource="@xml/file_paths"></meta-data>
+        </provider>
+    </application>
+
+</manifest>
diff --git a/deploy/android_demo/app/src/main/assets/images/det_0.jpg b/deploy/android_demo/app/src/main/assets/images/det_0.jpg
new file mode 100644
index 0000000..8517e12
Binary files /dev/null and b/deploy/android_demo/app/src/main/assets/images/det_0.jpg differ
diff --git a/deploy/android_demo/app/src/main/assets/images/det_180.jpg b/deploy/android_demo/app/src/main/assets/images/det_180.jpg
new file mode 100644
index 0000000..b1bb8a4
Binary files /dev/null and b/deploy/android_demo/app/src/main/assets/images/det_180.jpg differ
diff --git a/deploy/android_demo/app/src/main/assets/images/det_270.jpg b/deploy/android_demo/app/src/main/assets/images/det_270.jpg
new file mode 100644
index 0000000..5687390
Binary files /dev/null and b/deploy/android_demo/app/src/main/assets/images/det_270.jpg differ
diff --git a/deploy/android_demo/app/src/main/assets/images/det_90.jpg b/deploy/android_demo/app/src/main/assets/images/det_90.jpg
new file mode 100644
index 0000000..49e949a
Binary files /dev/null and b/deploy/android_demo/app/src/main/assets/images/det_90.jpg differ
diff --git a/deploy/android_demo/app/src/main/assets/images/rec_0.jpg b/deploy/android_demo/app/src/main/assets/images/rec_0.jpg
new file mode 100644
index 0000000..2c34cd3
Binary files /dev/null and b/deploy/android_demo/app/src/main/assets/images/rec_0.jpg differ
diff --git a/deploy/android_demo/app/src/main/assets/images/rec_0_180.jpg b/deploy/android_demo/app/src/main/assets/images/rec_0_180.jpg
new file mode 100644
index 0000000..02bc3b9
Binary files /dev/null and b/deploy/android_demo/app/src/main/assets/images/rec_0_180.jpg differ
diff --git a/deploy/android_demo/app/src/main/assets/images/rec_1.jpg b/deploy/android_demo/app/src/main/assets/images/rec_1.jpg
new file mode 100644
index 0000000..22031ba
Binary files /dev/null and b/deploy/android_demo/app/src/main/assets/images/rec_1.jpg differ
diff --git a/deploy/android_demo/app/src/main/assets/images/rec_1_180.jpg b/deploy/android_demo/app/src/main/assets/images/rec_1_180.jpg
new file mode 100644
index 0000000..d745530
Binary files /dev/null and b/deploy/android_demo/app/src/main/assets/images/rec_1_180.jpg differ
diff --git a/deploy/android_demo/app/src/main/cpp/CMakeLists.txt b/deploy/android_demo/app/src/main/cpp/CMakeLists.txt
new file mode 100644
index 0000000..39b7102
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/CMakeLists.txt
@@ -0,0 +1,117 @@
+# For more information about using CMake with Android Studio, read the
+# documentation: https://d.android.com/studio/projects/add-native-code.html
+
+# Sets the minimum version of CMake required to build the native library.
+
+cmake_minimum_required(VERSION 3.4.1)
+
+# Creates and names a library, sets it as either STATIC or SHARED, and provides
+# the relative paths to its source code. You can define multiple libraries, and
+# CMake builds them for you. Gradle automatically packages shared libraries with
+# your APK.
+
+set(PaddleLite_DIR "${CMAKE_CURRENT_SOURCE_DIR}/../../../PaddleLite")
+include_directories(${PaddleLite_DIR}/cxx/include)
+
+set(OpenCV_DIR "${CMAKE_CURRENT_SOURCE_DIR}/../../../OpenCV/sdk/native/jni")
+message(STATUS "opencv dir: ${OpenCV_DIR}")
+find_package(OpenCV REQUIRED)
+message(STATUS "OpenCV libraries: ${OpenCV_LIBS}")
+include_directories(${OpenCV_INCLUDE_DIRS})
+aux_source_directory(. SOURCES)
+set(CMAKE_CXX_FLAGS
+        "${CMAKE_CXX_FLAGS} -ffast-math -Ofast -Os"
+        )
+set(CMAKE_CXX_FLAGS
+        "${CMAKE_CXX_FLAGS} -fvisibility=hidden -fvisibility-inlines-hidden -fdata-sections -ffunction-sections"
+        )
+set(CMAKE_SHARED_LINKER_FLAGS
+        "${CMAKE_SHARED_LINKER_FLAGS} -Wl,--gc-sections -Wl,-z,nocopyreloc")
+
+add_library(
+        # Sets the name of the library.
+        Native
+        # Sets the library as a shared library.
+        SHARED
+        # Provides a relative path to your source file(s).
+        ${SOURCES})
+
+find_library(
+        # Sets the name of the path variable.
+        log-lib
+        # Specifies the name of the NDK library that you want CMake to locate.
+        log)
+
+add_library(
+        # Sets the name of the library.
+        paddle_light_api_shared
+        # Sets the library as a shared library.
+        SHARED
+        # Provides a relative path to your source file(s).
+        IMPORTED)
+
+set_target_properties(
+        # Specifies the target library.
+        paddle_light_api_shared
+        # Specifies the parameter you want to define.
+        PROPERTIES
+        IMPORTED_LOCATION
+        ${PaddleLite_DIR}/cxx/libs/${ANDROID_ABI}/libpaddle_light_api_shared.so
+        # Provides the path to the library you want to import.
+)
+
+
+# Specifies libraries CMake should link to your target library. You can link
+# multiple libraries, such as libraries you define in this build script,
+# prebuilt third-party libraries, or system libraries.
+
+target_link_libraries(
+        # Specifies the target library.
+        Native
+        paddle_light_api_shared
+        ${OpenCV_LIBS}
+        GLESv2
+        EGL
+        jnigraphics
+        ${log-lib}
+)
+
+add_custom_command(
+        TARGET Native
+        POST_BUILD
+        COMMAND
+        ${CMAKE_COMMAND} -E copy
+        ${PaddleLite_DIR}/cxx/libs/${ANDROID_ABI}/libc++_shared.so
+        ${CMAKE_LIBRARY_OUTPUT_DIRECTORY}/libc++_shared.so)
+
+add_custom_command(
+        TARGET Native
+        POST_BUILD
+        COMMAND
+        ${CMAKE_COMMAND} -E copy
+        ${PaddleLite_DIR}/cxx/libs/${ANDROID_ABI}/libpaddle_light_api_shared.so
+        ${CMAKE_LIBRARY_OUTPUT_DIRECTORY}/libpaddle_light_api_shared.so)
+
+add_custom_command(
+        TARGET Native
+        POST_BUILD
+        COMMAND
+        ${CMAKE_COMMAND} -E copy
+        ${PaddleLite_DIR}/cxx/libs/${ANDROID_ABI}/libhiai.so
+        ${CMAKE_LIBRARY_OUTPUT_DIRECTORY}/libhiai.so)
+
+add_custom_command(
+        TARGET Native
+        POST_BUILD
+        COMMAND
+        ${CMAKE_COMMAND} -E copy
+        ${PaddleLite_DIR}/cxx/libs/${ANDROID_ABI}/libhiai_ir.so
+        ${CMAKE_LIBRARY_OUTPUT_DIRECTORY}/libhiai_ir.so)
+
+add_custom_command(
+        TARGET Native
+        POST_BUILD
+        COMMAND
+        ${CMAKE_COMMAND} -E copy
+        ${PaddleLite_DIR}/cxx/libs/${ANDROID_ABI}/libhiai_ir_build.so
+        ${CMAKE_LIBRARY_OUTPUT_DIRECTORY}/libhiai_ir_build.so)
diff --git a/deploy/android_demo/app/src/main/cpp/common.h b/deploy/android_demo/app/src/main/cpp/common.h
new file mode 100644
index 0000000..fc47407
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/common.h
@@ -0,0 +1,37 @@
+//
+// Created by fu on 4/25/18.
+//
+
+#pragma once
+#import <numeric>
+#import <vector>
+
+#ifdef __ANDROID__
+
+#include <android/log.h>
+
+#define LOG_TAG "OCR_NDK"
+
+#define LOGI(...) __android_log_print(ANDROID_LOG_INFO, LOG_TAG, __VA_ARGS__)
+#define LOGW(...) __android_log_print(ANDROID_LOG_WARN, LOG_TAG, __VA_ARGS__)
+#define LOGE(...) __android_log_print(ANDROID_LOG_ERROR, LOG_TAG, __VA_ARGS__)
+#else
+#include <stdio.h>
+#define LOGI(format, ...)                                                      \
+  fprintf(stdout, "[" LOG_TAG "]" format "\n", ##__VA_ARGS__)
+#define LOGW(format, ...)                                                      \
+  fprintf(stdout, "[" LOG_TAG "]" format "\n", ##__VA_ARGS__)
+#define LOGE(format, ...)                                                      \
+  fprintf(stderr, "[" LOG_TAG "]Error: " format "\n", ##__VA_ARGS__)
+#endif
+
+enum RETURN_CODE { RETURN_OK = 0 };
+
+enum NET_TYPE { NET_OCR = 900100, NET_OCR_INTERNAL = 991008 };
+
+template <typename T> inline T product(const std::vector<T> &vec) {
+  if (vec.empty()) {
+    return 0;
+  }
+  return std::accumulate(vec.begin(), vec.end(), 1, std::multiplies<T>());
+}
diff --git a/deploy/android_demo/app/src/main/cpp/native.cpp b/deploy/android_demo/app/src/main/cpp/native.cpp
new file mode 100644
index 0000000..5674f64
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/native.cpp
@@ -0,0 +1,120 @@
+//
+// Created by fujiayi on 2020/7/5.
+//
+
+#include "native.h"
+#include "ocr_ppredictor.h"
+#include <algorithm>
+#include <paddle_api.h>
+#include <string>
+
+static paddle::lite_api::PowerMode str_to_cpu_mode(const std::string &cpu_mode);
+
+extern "C" JNIEXPORT jlong JNICALL
+Java_com_baidu_paddle_lite_demo_ocr_OCRPredictorNative_init(
+    JNIEnv *env, jobject thiz, jstring j_det_model_path,
+    jstring j_rec_model_path, jstring j_cls_model_path, jint j_use_opencl,
+    jint j_thread_num, jstring j_cpu_mode) {
+  std::string det_model_path = jstring_to_cpp_string(env, j_det_model_path);
+  std::string rec_model_path = jstring_to_cpp_string(env, j_rec_model_path);
+  std::string cls_model_path = jstring_to_cpp_string(env, j_cls_model_path);
+  int thread_num = j_thread_num;
+  std::string cpu_mode = jstring_to_cpp_string(env, j_cpu_mode);
+  ppredictor::OCR_Config conf;
+  conf.use_opencl = j_use_opencl;
+  conf.thread_num = thread_num;
+  conf.mode = str_to_cpu_mode(cpu_mode);
+  ppredictor::OCR_PPredictor *orc_predictor =
+      new ppredictor::OCR_PPredictor{conf};
+  orc_predictor->init_from_file(det_model_path, rec_model_path, cls_model_path);
+  return reinterpret_cast<jlong>(orc_predictor);
+}
+
+/**
+ * "LITE_POWER_HIGH" convert to paddle::lite_api::LITE_POWER_HIGH
+ * @param cpu_mode
+ * @return
+ */
+static paddle::lite_api::PowerMode
+str_to_cpu_mode(const std::string &cpu_mode) {
+  static std::map<std::string, paddle::lite_api::PowerMode> cpu_mode_map{
+      {"LITE_POWER_HIGH", paddle::lite_api::LITE_POWER_HIGH},
+      {"LITE_POWER_LOW", paddle::lite_api::LITE_POWER_HIGH},
+      {"LITE_POWER_FULL", paddle::lite_api::LITE_POWER_FULL},
+      {"LITE_POWER_NO_BIND", paddle::lite_api::LITE_POWER_NO_BIND},
+      {"LITE_POWER_RAND_HIGH", paddle::lite_api::LITE_POWER_RAND_HIGH},
+      {"LITE_POWER_RAND_LOW", paddle::lite_api::LITE_POWER_RAND_LOW}};
+  std::string upper_key;
+  std::transform(cpu_mode.cbegin(), cpu_mode.cend(), upper_key.begin(),
+                 ::toupper);
+  auto index = cpu_mode_map.find(upper_key.c_str());
+  if (index == cpu_mode_map.end()) {
+    LOGE("cpu_mode not found %s", upper_key.c_str());
+    return paddle::lite_api::LITE_POWER_HIGH;
+  } else {
+    return index->second;
+  }
+}
+
+extern "C" JNIEXPORT jfloatArray JNICALL
+Java_com_baidu_paddle_lite_demo_ocr_OCRPredictorNative_forward(
+    JNIEnv *env, jobject thiz, jlong java_pointer, jobject original_image,
+    jint j_max_size_len, jint j_run_det, jint j_run_cls, jint j_run_rec) {
+  LOGI("begin to run native forward");
+  if (java_pointer == 0) {
+    LOGE("JAVA pointer is NULL");
+    return cpp_array_to_jfloatarray(env, nullptr, 0);
+  }
+
+  cv::Mat origin = bitmap_to_cv_mat(env, original_image);
+  if (origin.size == 0) {
+    LOGE("origin bitmap cannot convert to CV Mat");
+    return cpp_array_to_jfloatarray(env, nullptr, 0);
+  }
+
+  int max_size_len = j_max_size_len;
+  int run_det = j_run_det;
+  int run_cls = j_run_cls;
+  int run_rec = j_run_rec;
+
+  ppredictor::OCR_PPredictor *ppredictor =
+      (ppredictor::OCR_PPredictor *)java_pointer;
+  std::vector<int64_t> dims_arr;
+  std::vector<ppredictor::OCRPredictResult> results =
+      ppredictor->infer_ocr(origin, max_size_len, run_det, run_cls, run_rec);
+  LOGI("infer_ocr finished with boxes %ld", results.size());
+
+  // 这里将std::vector<ppredictor::OCRPredictResult> 序列化成
+  // float数组，传输到java层再反序列化
+  std::vector<float> float_arr;
+  for (const ppredictor::OCRPredictResult &r : results) {
+    float_arr.push_back(r.points.size());
+    float_arr.push_back(r.word_index.size());
+    float_arr.push_back(r.score);
+    // add det point
+    for (const std::vector<int> &point : r.points) {
+      float_arr.push_back(point.at(0));
+      float_arr.push_back(point.at(1));
+    }
+    // add rec word idx
+    for (int index : r.word_index) {
+      float_arr.push_back(index);
+    }
+    // add cls result
+    float_arr.push_back(r.cls_label);
+    float_arr.push_back(r.cls_score);
+  }
+  return cpp_array_to_jfloatarray(env, float_arr.data(), float_arr.size());
+}
+
+extern "C" JNIEXPORT void JNICALL
+Java_com_baidu_paddle_lite_demo_ocr_OCRPredictorNative_release(
+    JNIEnv *env, jobject thiz, jlong java_pointer) {
+  if (java_pointer == 0) {
+    LOGE("JAVA pointer is NULL");
+    return;
+  }
+  ppredictor::OCR_PPredictor *ppredictor =
+      (ppredictor::OCR_PPredictor *)java_pointer;
+  delete ppredictor;
+}
diff --git a/deploy/android_demo/app/src/main/cpp/native.h b/deploy/android_demo/app/src/main/cpp/native.h
new file mode 100644
index 0000000..9a87106
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/native.h
@@ -0,0 +1,146 @@
+//
+// Created by fujiayi on 2020/7/5.
+//
+
+#pragma once
+
+#include "common.h"
+#include <android/bitmap.h>
+#include <jni.h>
+#include <opencv2/opencv.hpp>
+#include <string>
+#include <vector>
+
+inline std::string jstring_to_cpp_string(JNIEnv *env, jstring jstr) {
+  // In java, a unicode char will be encoded using 2 bytes (utf16).
+  // so jstring will contain characters utf16. std::string in c++ is
+  // essentially a string of bytes, not characters, so if we want to
+  // pass jstring from JNI to c++, we have convert utf16 to bytes.
+  if (!jstr) {
+    return "";
+  }
+  const jclass stringClass = env->GetObjectClass(jstr);
+  const jmethodID getBytes =
+      env->GetMethodID(stringClass, "getBytes", "(Ljava/lang/String;)[B");
+  const jbyteArray stringJbytes = (jbyteArray)env->CallObjectMethod(
+      jstr, getBytes, env->NewStringUTF("UTF-8"));
+
+  size_t length = (size_t)env->GetArrayLength(stringJbytes);
+  jbyte *pBytes = env->GetByteArrayElements(stringJbytes, NULL);
+
+  std::string ret = std::string(reinterpret_cast<char *>(pBytes), length);
+  env->ReleaseByteArrayElements(stringJbytes, pBytes, JNI_ABORT);
+
+  env->DeleteLocalRef(stringJbytes);
+  env->DeleteLocalRef(stringClass);
+  return ret;
+}
+
+inline jstring cpp_string_to_jstring(JNIEnv *env, std::string str) {
+  auto *data = str.c_str();
+  jclass strClass = env->FindClass("java/lang/String");
+  jmethodID strClassInitMethodID =
+      env->GetMethodID(strClass, "<init>", "([BLjava/lang/String;)V");
+
+  jbyteArray bytes = env->NewByteArray(strlen(data));
+  env->SetByteArrayRegion(bytes, 0, strlen(data),
+                          reinterpret_cast<const jbyte *>(data));
+
+  jstring encoding = env->NewStringUTF("UTF-8");
+  jstring res = (jstring)(env->NewObject(strClass, strClassInitMethodID, bytes,
+                                         encoding));
+
+  env->DeleteLocalRef(strClass);
+  env->DeleteLocalRef(encoding);
+  env->DeleteLocalRef(bytes);
+
+  return res;
+}
+
+inline jfloatArray cpp_array_to_jfloatarray(JNIEnv *env, const float *buf,
+                                            int64_t len) {
+  if (len == 0) {
+    return env->NewFloatArray(0);
+  }
+  jfloatArray result = env->NewFloatArray(len);
+  env->SetFloatArrayRegion(result, 0, len, buf);
+  return result;
+}
+
+inline jintArray cpp_array_to_jintarray(JNIEnv *env, const int *buf,
+                                        int64_t len) {
+  jintArray result = env->NewIntArray(len);
+  env->SetIntArrayRegion(result, 0, len, buf);
+  return result;
+}
+
+inline jbyteArray cpp_array_to_jbytearray(JNIEnv *env, const int8_t *buf,
+                                          int64_t len) {
+  jbyteArray result = env->NewByteArray(len);
+  env->SetByteArrayRegion(result, 0, len, buf);
+  return result;
+}
+
+inline jlongArray int64_vector_to_jlongarray(JNIEnv *env,
+                                             const std::vector<int64_t> &vec) {
+  jlongArray result = env->NewLongArray(vec.size());
+  jlong *buf = new jlong[vec.size()];
+  for (size_t i = 0; i < vec.size(); ++i) {
+    buf[i] = (jlong)vec[i];
+  }
+  env->SetLongArrayRegion(result, 0, vec.size(), buf);
+  delete[] buf;
+  return result;
+}
+
+inline std::vector<int64_t> jlongarray_to_int64_vector(JNIEnv *env,
+                                                       jlongArray data) {
+  int data_size = env->GetArrayLength(data);
+  jlong *data_ptr = env->GetLongArrayElements(data, nullptr);
+  std::vector<int64_t> data_vec(data_ptr, data_ptr + data_size);
+  env->ReleaseLongArrayElements(data, data_ptr, 0);
+  return data_vec;
+}
+
+inline std::vector<float> jfloatarray_to_float_vector(JNIEnv *env,
+                                                      jfloatArray data) {
+  int data_size = env->GetArrayLength(data);
+  jfloat *data_ptr = env->GetFloatArrayElements(data, nullptr);
+  std::vector<float> data_vec(data_ptr, data_ptr + data_size);
+  env->ReleaseFloatArrayElements(data, data_ptr, 0);
+  return data_vec;
+}
+
+inline cv::Mat bitmap_to_cv_mat(JNIEnv *env, jobject bitmap) {
+  AndroidBitmapInfo info;
+  int result = AndroidBitmap_getInfo(env, bitmap, &info);
+  if (result != ANDROID_BITMAP_RESULT_SUCCESS) {
+    LOGE("AndroidBitmap_getInfo failed, result: %d", result);
+    return cv::Mat{};
+  }
+  if (info.format != ANDROID_BITMAP_FORMAT_RGBA_8888) {
+    LOGE("Bitmap format is not RGBA_8888 !");
+    return cv::Mat{};
+  }
+  if (info.width == 0 || info.height == 0 || info.stride < info.width * 4u) {
+    LOGE("Invalid bitmap dimensions or stride");
+    return cv::Mat{};
+  }
+  unsigned char *srcData = NULL;
+  AndroidBitmap_lockPixels(env, bitmap, (void **)&srcData);
+  cv::Mat mat = cv::Mat::zeros(info.height, info.width, CV_8UC4);
+  if (!mat.data) {
+    AndroidBitmap_unlockPixels(env, bitmap);
+    return cv::Mat{};
+  }
+  size_t copy_size = (size_t)info.height * info.width * 4;
+  memcpy(mat.data, srcData, copy_size);
+  AndroidBitmap_unlockPixels(env, bitmap);
+  cv::cvtColor(mat, mat, cv::COLOR_RGBA2BGR);
+  /**
+  if (!cv::imwrite("/sdcard/1/copy.jpg", mat)){
+      LOGE("Write image failed " );
+  }
+   */
+  return mat;
+}
diff --git a/deploy/android_demo/app/src/main/cpp/ocr_clipper.cpp b/deploy/android_demo/app/src/main/cpp/ocr_clipper.cpp
new file mode 100644
index 0000000..2d52d20
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/ocr_clipper.cpp
@@ -0,0 +1,4378 @@
+/*******************************************************************************
+ *                                                                              *
+ * Author    :  Angus Johnson * Version   :  6.4.2 * Date      :  27 February
+ *2017                                                * Website   :
+ *http://www.angusj.com                                           * Copyright :
+ *Angus Johnson 2010-2017                                         *
+ *                                                                              *
+ * License: * Use, modification & distribution is subject to Boost Software
+ *License Ver 1. * http://www.boost.org/LICENSE_1_0.txt *
+ *                                                                              *
+ * Attributions: * The code in this library is an extension of Bala Vatti's
+ *clipping algorithm: * "A generic solution to polygon clipping" *
+ * Communications of the ACM, Vol 35, Issue 7 (July 1992) pp 56-63. *
+ * http://portal.acm.org/citation.cfm?id=129906 *
+ *                                                                              *
+ * Computer graphics and geometric modeling: implementation and algorithms * By
+ *Max K. Agoston                                                            *
+ * Springer; 1 edition (January 4, 2005) *
+ * http://books.google.com/books?q=vatti+clipping+agoston *
+ *                                                                              *
+ * See also: * "Polygon Offsetting by Computing Winding Numbers" * Paper no.
+ *DETC2005-85513 pp. 565-575                                         * ASME 2005
+ *International Design Engineering Technical Conferences             * and
+ *Computers and Information in Engineering Conference (IDETC/CIE2005)      *
+ * September 24-28, 2005 , Long Beach, California, USA *
+ * http://www.me.berkeley.edu/~mcmains/pubs/DAC05OffsetPolygon.pdf *
+ *                                                                              *
+ *******************************************************************************/
+
+/*******************************************************************************
+ *                                                                              *
+ * This is a translation of the Delphi Clipper library and the naming style *
+ * used has retained a Delphi flavour. *
+ *                                                                              *
+ *******************************************************************************/
+
+#include "ocr_clipper.hpp"
+#include <algorithm>
+#include <cmath>
+#include <cstdlib>
+#include <cstring>
+#include <functional>
+#include <ostream>
+#include <stdexcept>
+#include <vector>
+
+namespace ClipperLib {
+
+static double const pi = 3.141592653589793238;
+static double const two_pi = pi * 2;
+static double const def_arc_tolerance = 0.25;
+
+enum Direction { dRightToLeft, dLeftToRight };
+
+static int const Unassigned = -1; // edge not currently 'owning' a solution
+static int const Skip = -2;       // edge that would otherwise close a path
+
+#define HORIZONTAL (-1.0E+40)
+#define TOLERANCE (1.0e-20)
+#define NEAR_ZERO(val) (((val) > -TOLERANCE) && ((val) < TOLERANCE))
+
+struct TEdge {
+  IntPoint Bot;
+  IntPoint Curr; // current (updated for every new scanbeam)
+  IntPoint Top;
+  double Dx;
+  PolyType PolyTyp;
+  EdgeSide Side; // side only refers to current side of solution poly
+  int WindDelta; // 1 or -1 depending on winding direction
+  int WindCnt;
+  int WindCnt2; // winding count of the opposite polytype
+  int OutIdx;
+  TEdge *Next;
+  TEdge *Prev;
+  TEdge *NextInLML;
+  TEdge *NextInAEL;
+  TEdge *PrevInAEL;
+  TEdge *NextInSEL;
+  TEdge *PrevInSEL;
+};
+
+struct IntersectNode {
+  TEdge *Edge1;
+  TEdge *Edge2;
+  IntPoint Pt;
+};
+
+struct LocalMinimum {
+  cInt Y;
+  TEdge *LeftBound;
+  TEdge *RightBound;
+};
+
+struct OutPt;
+
+// OutRec: contains a path in the clipping solution. Edges in the AEL will
+// carry a pointer to an OutRec when they are part of the clipping solution.
+struct OutRec {
+  int Idx;
+  bool IsHole;
+  bool IsOpen;
+  OutRec *FirstLeft; // see comments in clipper.pas
+  PolyNode *PolyNd;
+  OutPt *Pts;
+  OutPt *BottomPt;
+};
+
+struct OutPt {
+  int Idx;
+  IntPoint Pt;
+  OutPt *Next;
+  OutPt *Prev;
+};
+
+struct Join {
+  OutPt *OutPt1;
+  OutPt *OutPt2;
+  IntPoint OffPt;
+};
+
+struct LocMinSorter {
+  inline bool operator()(const LocalMinimum &locMin1,
+                         const LocalMinimum &locMin2) {
+    return locMin2.Y < locMin1.Y;
+  }
+};
+
+//------------------------------------------------------------------------------
+//------------------------------------------------------------------------------
+
+inline cInt Round(double val) {
+  if ((val < 0))
+    return static_cast<cInt>(val - 0.5);
+  else
+    return static_cast<cInt>(val + 0.5);
+}
+//------------------------------------------------------------------------------
+
+inline cInt Abs(cInt val) { return val < 0 ? -val : val; }
+
+//------------------------------------------------------------------------------
+// PolyTree methods ...
+//------------------------------------------------------------------------------
+
+void PolyTree::Clear() {
+  for (PolyNodes::size_type i = 0; i < AllNodes.size(); ++i)
+    delete AllNodes[i];
+  AllNodes.resize(0);
+  Children.resize(0);
+}
+//------------------------------------------------------------------------------
+
+PolyNode *PolyTree::GetFirst() const {
+  if (!Children.empty())
+    return Children[0];
+  else
+    return 0;
+}
+//------------------------------------------------------------------------------
+
+int PolyTree::Total() const {
+  int result = (int)AllNodes.size();
+  // with negative offsets, ignore the hidden outer polygon ...
+  if (result > 0 && Children[0] != AllNodes[0])
+    result--;
+  return result;
+}
+
+//------------------------------------------------------------------------------
+// PolyNode methods ...
+//------------------------------------------------------------------------------
+
+PolyNode::PolyNode() : Parent(0), Index(0), m_IsOpen(false) {}
+//------------------------------------------------------------------------------
+
+int PolyNode::ChildCount() const { return (int)Children.size(); }
+//------------------------------------------------------------------------------
+
+void PolyNode::AddChild(PolyNode &child) {
+  unsigned cnt = (unsigned)Children.size();
+  Children.push_back(&child);
+  child.Parent = this;
+  child.Index = cnt;
+}
+//------------------------------------------------------------------------------
+
+PolyNode *PolyNode::GetNext() const {
+  if (!Children.empty())
+    return Children[0];
+  else
+    return GetNextSiblingUp();
+}
+//------------------------------------------------------------------------------
+
+PolyNode *PolyNode::GetNextSiblingUp() const {
+  if (!Parent) // protects against PolyTree.GetNextSiblingUp()
+    return 0;
+  else if (Index == Parent->Children.size() - 1)
+    return Parent->GetNextSiblingUp();
+  else
+    return Parent->Children[Index + 1];
+}
+//------------------------------------------------------------------------------
+
+bool PolyNode::IsHole() const {
+  bool result = true;
+  PolyNode *node = Parent;
+  while (node) {
+    result = !result;
+    node = node->Parent;
+  }
+  return result;
+}
+//------------------------------------------------------------------------------
+
+bool PolyNode::IsOpen() const { return m_IsOpen; }
+//------------------------------------------------------------------------------
+
+#ifndef use_int32
+
+//------------------------------------------------------------------------------
+// Int128 class (enables safe math on signed 64bit integers)
+// eg Int128 val1((long64)9223372036854775807); //ie 2^63 -1
+//    Int128 val2((long64)9223372036854775807);
+//    Int128 val3 = val1 * val2;
+//    val3.AsString => "85070591730234615847396907784232501249" (8.5e+37)
+//------------------------------------------------------------------------------
+
+class Int128 {
+public:
+  ulong64 lo;
+  long64 hi;
+
+  Int128(long64 _lo = 0) {
+    lo = (ulong64)_lo;
+    if (_lo < 0)
+      hi = -1;
+    else
+      hi = 0;
+  }
+
+  Int128(const Int128 &val) : lo(val.lo), hi(val.hi) {}
+
+  Int128(const long64 &_hi, const ulong64 &_lo) : lo(_lo), hi(_hi) {}
+
+  Int128 &operator=(const long64 &val) {
+    lo = (ulong64)val;
+    if (val < 0)
+      hi = -1;
+    else
+      hi = 0;
+    return *this;
+  }
+
+  bool operator==(const Int128 &val) const {
+    return (hi == val.hi && lo == val.lo);
+  }
+
+  bool operator!=(const Int128 &val) const { return !(*this == val); }
+
+  bool operator>(const Int128 &val) const {
+    if (hi != val.hi)
+      return hi > val.hi;
+    else
+      return lo > val.lo;
+  }
+
+  bool operator<(const Int128 &val) const {
+    if (hi != val.hi)
+      return hi < val.hi;
+    else
+      return lo < val.lo;
+  }
+
+  bool operator>=(const Int128 &val) const { return !(*this < val); }
+
+  bool operator<=(const Int128 &val) const { return !(*this > val); }
+
+  Int128 &operator+=(const Int128 &rhs) {
+    hi += rhs.hi;
+    lo += rhs.lo;
+    if (lo < rhs.lo)
+      hi++;
+    return *this;
+  }
+
+  Int128 operator+(const Int128 &rhs) const {
+    Int128 result(*this);
+    result += rhs;
+    return result;
+  }
+
+  Int128 &operator-=(const Int128 &rhs) {
+    *this += -rhs;
+    return *this;
+  }
+
+  Int128 operator-(const Int128 &rhs) const {
+    Int128 result(*this);
+    result -= rhs;
+    return result;
+  }
+
+  Int128 operator-() const // unary negation
+  {
+    if (lo == 0)
+      return Int128(-hi, 0);
+    else
+      return Int128(~hi, ~lo + 1);
+  }
+
+  operator double() const {
+    const double shift64 = 18446744073709551616.0; // 2^64
+    if (hi < 0) {
+      if (lo == 0)
+        return (double)hi * shift64;
+      else
+        return -(double)(~lo + ~hi * shift64);
+    } else
+      return (double)(lo + hi * shift64);
+  }
+};
+//------------------------------------------------------------------------------
+
+Int128 Int128Mul(long64 lhs, long64 rhs) {
+  bool negate = (lhs < 0) != (rhs < 0);
+
+  if (lhs < 0)
+    lhs = -lhs;
+  ulong64 int1Hi = ulong64(lhs) >> 32;
+  ulong64 int1Lo = ulong64(lhs & 0xFFFFFFFF);
+
+  if (rhs < 0)
+    rhs = -rhs;
+  ulong64 int2Hi = ulong64(rhs) >> 32;
+  ulong64 int2Lo = ulong64(rhs & 0xFFFFFFFF);
+
+  // nb: see comments in clipper.pas
+  ulong64 a = int1Hi * int2Hi;
+  ulong64 b = int1Lo * int2Lo;
+  ulong64 c = int1Hi * int2Lo + int1Lo * int2Hi;
+
+  Int128 tmp;
+  tmp.hi = long64(a + (c >> 32));
+  tmp.lo = long64(c << 32);
+  tmp.lo += long64(b);
+  if (tmp.lo < b)
+    tmp.hi++;
+  if (negate)
+    tmp = -tmp;
+  return tmp;
+};
+#endif
+
+//------------------------------------------------------------------------------
+// Miscellaneous global functions
+//------------------------------------------------------------------------------
+
+bool Orientation(const Path &poly) { return Area(poly) >= 0; }
+//------------------------------------------------------------------------------
+
+double Area(const Path &poly) {
+  int size = (int)poly.size();
+  if (size < 3)
+    return 0;
+
+  double a = 0;
+  for (int i = 0, j = size - 1; i < size; ++i) {
+    a += ((double)poly[j].X + poly[i].X) * ((double)poly[j].Y - poly[i].Y);
+    j = i;
+  }
+  return -a * 0.5;
+}
+//------------------------------------------------------------------------------
+
+double Area(const OutPt *op) {
+  const OutPt *startOp = op;
+  if (!op)
+    return 0;
+  double a = 0;
+  do {
+    a += (double)(op->Prev->Pt.X + op->Pt.X) *
+         (double)(op->Prev->Pt.Y - op->Pt.Y);
+    op = op->Next;
+  } while (op != startOp);
+  return a * 0.5;
+}
+//------------------------------------------------------------------------------
+
+double Area(const OutRec &outRec) { return Area(outRec.Pts); }
+//------------------------------------------------------------------------------
+
+bool PointIsVertex(const IntPoint &Pt, OutPt *pp) {
+  OutPt *pp2 = pp;
+  do {
+    if (pp2->Pt == Pt)
+      return true;
+    pp2 = pp2->Next;
+  } while (pp2 != pp);
+  return false;
+}
+//------------------------------------------------------------------------------
+
+// See "The Point in Polygon Problem for Arbitrary Polygons" by Hormann &
+// Agathos
+// http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.88.5498&rep=rep1&type=pdf
+int PointInPolygon(const IntPoint &pt, const Path &path) {
+  // returns 0 if false, +1 if true, -1 if pt ON polygon boundary
+  int result = 0;
+  size_t cnt = path.size();
+  if (cnt < 3)
+    return 0;
+  IntPoint ip = path[0];
+  for (size_t i = 1; i <= cnt; ++i) {
+    IntPoint ipNext = (i == cnt ? path[0] : path[i]);
+    if (ipNext.Y == pt.Y) {
+      if ((ipNext.X == pt.X) ||
+          (ip.Y == pt.Y && ((ipNext.X > pt.X) == (ip.X < pt.X))))
+        return -1;
+    }
+    if ((ip.Y < pt.Y) != (ipNext.Y < pt.Y)) {
+      if (ip.X >= pt.X) {
+        if (ipNext.X > pt.X)
+          result = 1 - result;
+        else {
+          double d = (double)(ip.X - pt.X) * (ipNext.Y - pt.Y) -
+                     (double)(ipNext.X - pt.X) * (ip.Y - pt.Y);
+          if (!d)
+            return -1;
+          if ((d > 0) == (ipNext.Y > ip.Y))
+            result = 1 - result;
+        }
+      } else {
+        if (ipNext.X > pt.X) {
+          double d = (double)(ip.X - pt.X) * (ipNext.Y - pt.Y) -
+                     (double)(ipNext.X - pt.X) * (ip.Y - pt.Y);
+          if (!d)
+            return -1;
+          if ((d > 0) == (ipNext.Y > ip.Y))
+            result = 1 - result;
+        }
+      }
+    }
+    ip = ipNext;
+  }
+  return result;
+}
+//------------------------------------------------------------------------------
+
+int PointInPolygon(const IntPoint &pt, OutPt *op) {
+  // returns 0 if false, +1 if true, -1 if pt ON polygon boundary
+  int result = 0;
+  OutPt *startOp = op;
+  for (;;) {
+    if (op->Next->Pt.Y == pt.Y) {
+      if ((op->Next->Pt.X == pt.X) ||
+          (op->Pt.Y == pt.Y && ((op->Next->Pt.X > pt.X) == (op->Pt.X < pt.X))))
+        return -1;
+    }
+    if ((op->Pt.Y < pt.Y) != (op->Next->Pt.Y < pt.Y)) {
+      if (op->Pt.X >= pt.X) {
+        if (op->Next->Pt.X > pt.X)
+          result = 1 - result;
+        else {
+          double d = (double)(op->Pt.X - pt.X) * (op->Next->Pt.Y - pt.Y) -
+                     (double)(op->Next->Pt.X - pt.X) * (op->Pt.Y - pt.Y);
+          if (!d)
+            return -1;
+          if ((d > 0) == (op->Next->Pt.Y > op->Pt.Y))
+            result = 1 - result;
+        }
+      } else {
+        if (op->Next->Pt.X > pt.X) {
+          double d = (double)(op->Pt.X - pt.X) * (op->Next->Pt.Y - pt.Y) -
+                     (double)(op->Next->Pt.X - pt.X) * (op->Pt.Y - pt.Y);
+          if (!d)
+            return -1;
+          if ((d > 0) == (op->Next->Pt.Y > op->Pt.Y))
+            result = 1 - result;
+        }
+      }
+    }
+    op = op->Next;
+    if (startOp == op)
+      break;
+  }
+  return result;
+}
+//------------------------------------------------------------------------------
+
+bool Poly2ContainsPoly1(OutPt *OutPt1, OutPt *OutPt2) {
+  OutPt *op = OutPt1;
+  do {
+    // nb: PointInPolygon returns 0 if false, +1 if true, -1 if pt on polygon
+    int res = PointInPolygon(op->Pt, OutPt2);
+    if (res >= 0)
+      return res > 0;
+    op = op->Next;
+  } while (op != OutPt1);
+  return true;
+}
+//----------------------------------------------------------------------
+
+bool SlopesEqual(const TEdge &e1, const TEdge &e2, bool UseFullInt64Range) {
+#ifndef use_int32
+  if (UseFullInt64Range)
+    return Int128Mul(e1.Top.Y - e1.Bot.Y, e2.Top.X - e2.Bot.X) ==
+           Int128Mul(e1.Top.X - e1.Bot.X, e2.Top.Y - e2.Bot.Y);
+  else
+#endif
+    return (e1.Top.Y - e1.Bot.Y) * (e2.Top.X - e2.Bot.X) ==
+           (e1.Top.X - e1.Bot.X) * (e2.Top.Y - e2.Bot.Y);
+}
+//------------------------------------------------------------------------------
+
+bool SlopesEqual(const IntPoint pt1, const IntPoint pt2, const IntPoint pt3,
+                 bool UseFullInt64Range) {
+#ifndef use_int32
+  if (UseFullInt64Range)
+    return Int128Mul(pt1.Y - pt2.Y, pt2.X - pt3.X) ==
+           Int128Mul(pt1.X - pt2.X, pt2.Y - pt3.Y);
+  else
+#endif
+    return (pt1.Y - pt2.Y) * (pt2.X - pt3.X) ==
+           (pt1.X - pt2.X) * (pt2.Y - pt3.Y);
+}
+//------------------------------------------------------------------------------
+
+bool SlopesEqual(const IntPoint pt1, const IntPoint pt2, const IntPoint pt3,
+                 const IntPoint pt4, bool UseFullInt64Range) {
+#ifndef use_int32
+  if (UseFullInt64Range)
+    return Int128Mul(pt1.Y - pt2.Y, pt3.X - pt4.X) ==
+           Int128Mul(pt1.X - pt2.X, pt3.Y - pt4.Y);
+  else
+#endif
+    return (pt1.Y - pt2.Y) * (pt3.X - pt4.X) ==
+           (pt1.X - pt2.X) * (pt3.Y - pt4.Y);
+}
+//------------------------------------------------------------------------------
+
+inline bool IsHorizontal(TEdge &e) { return e.Dx == HORIZONTAL; }
+//------------------------------------------------------------------------------
+
+inline double GetDx(const IntPoint pt1, const IntPoint pt2) {
+  return (pt1.Y == pt2.Y) ? HORIZONTAL
+                          : (double)(pt2.X - pt1.X) / (pt2.Y - pt1.Y);
+}
+//---------------------------------------------------------------------------
+
+inline void SetDx(TEdge &e) {
+  cInt dy = (e.Top.Y - e.Bot.Y);
+  if (dy == 0)
+    e.Dx = HORIZONTAL;
+  else
+    e.Dx = (double)(e.Top.X - e.Bot.X) / dy;
+}
+//---------------------------------------------------------------------------
+
+inline void SwapSides(TEdge &Edge1, TEdge &Edge2) {
+  EdgeSide Side = Edge1.Side;
+  Edge1.Side = Edge2.Side;
+  Edge2.Side = Side;
+}
+//------------------------------------------------------------------------------
+
+inline void SwapPolyIndexes(TEdge &Edge1, TEdge &Edge2) {
+  int OutIdx = Edge1.OutIdx;
+  Edge1.OutIdx = Edge2.OutIdx;
+  Edge2.OutIdx = OutIdx;
+}
+//------------------------------------------------------------------------------
+
+inline cInt TopX(TEdge &edge, const cInt currentY) {
+  return (currentY == edge.Top.Y)
+             ? edge.Top.X
+             : edge.Bot.X + Round(edge.Dx * (currentY - edge.Bot.Y));
+}
+//------------------------------------------------------------------------------
+
+void IntersectPoint(TEdge &Edge1, TEdge &Edge2, IntPoint &ip) {
+#ifdef use_xyz
+  ip.Z = 0;
+#endif
+
+  double b1, b2;
+  if (Edge1.Dx == Edge2.Dx) {
+    ip.Y = Edge1.Curr.Y;
+    ip.X = TopX(Edge1, ip.Y);
+    return;
+  } else if (Edge1.Dx == 0) {
+    ip.X = Edge1.Bot.X;
+    if (IsHorizontal(Edge2))
+      ip.Y = Edge2.Bot.Y;
+    else {
+      b2 = Edge2.Bot.Y - (Edge2.Bot.X / Edge2.Dx);
+      ip.Y = Round(ip.X / Edge2.Dx + b2);
+    }
+  } else if (Edge2.Dx == 0) {
+    ip.X = Edge2.Bot.X;
+    if (IsHorizontal(Edge1))
+      ip.Y = Edge1.Bot.Y;
+    else {
+      b1 = Edge1.Bot.Y - (Edge1.Bot.X / Edge1.Dx);
+      ip.Y = Round(ip.X / Edge1.Dx + b1);
+    }
+  } else {
+    b1 = Edge1.Bot.X - Edge1.Bot.Y * Edge1.Dx;
+    b2 = Edge2.Bot.X - Edge2.Bot.Y * Edge2.Dx;
+    double q = (b2 - b1) / (Edge1.Dx - Edge2.Dx);
+    ip.Y = Round(q);
+    if (std::fabs(Edge1.Dx) < std::fabs(Edge2.Dx))
+      ip.X = Round(Edge1.Dx * q + b1);
+    else
+      ip.X = Round(Edge2.Dx * q + b2);
+  }
+
+  if (ip.Y < Edge1.Top.Y || ip.Y < Edge2.Top.Y) {
+    if (Edge1.Top.Y > Edge2.Top.Y)
+      ip.Y = Edge1.Top.Y;
+    else
+      ip.Y = Edge2.Top.Y;
+    if (std::fabs(Edge1.Dx) < std::fabs(Edge2.Dx))
+      ip.X = TopX(Edge1, ip.Y);
+    else
+      ip.X = TopX(Edge2, ip.Y);
+  }
+  // finally, don't allow 'ip' to be BELOW curr.Y (ie bottom of scanbeam) ...
+  if (ip.Y > Edge1.Curr.Y) {
+    ip.Y = Edge1.Curr.Y;
+    // use the more vertical edge to derive X ...
+    if (std::fabs(Edge1.Dx) > std::fabs(Edge2.Dx))
+      ip.X = TopX(Edge2, ip.Y);
+    else
+      ip.X = TopX(Edge1, ip.Y);
+  }
+}
+//------------------------------------------------------------------------------
+
+void ReversePolyPtLinks(OutPt *pp) {
+  if (!pp)
+    return;
+  OutPt *pp1, *pp2;
+  pp1 = pp;
+  do {
+    pp2 = pp1->Next;
+    pp1->Next = pp1->Prev;
+    pp1->Prev = pp2;
+    pp1 = pp2;
+  } while (pp1 != pp);
+}
+//------------------------------------------------------------------------------
+
+void DisposeOutPts(OutPt *&pp) {
+  if (pp == 0)
+    return;
+  pp->Prev->Next = 0;
+  while (pp) {
+    OutPt *tmpPp = pp;
+    pp = pp->Next;
+    delete tmpPp;
+  }
+}
+//------------------------------------------------------------------------------
+
+inline void InitEdge(TEdge *e, TEdge *eNext, TEdge *ePrev, const IntPoint &Pt) {
+  std::memset(e, 0, sizeof(TEdge));
+  e->Next = eNext;
+  e->Prev = ePrev;
+  e->Curr = Pt;
+  e->OutIdx = Unassigned;
+}
+//------------------------------------------------------------------------------
+
+void InitEdge2(TEdge &e, PolyType Pt) {
+  if (e.Curr.Y >= e.Next->Curr.Y) {
+    e.Bot = e.Curr;
+    e.Top = e.Next->Curr;
+  } else {
+    e.Top = e.Curr;
+    e.Bot = e.Next->Curr;
+  }
+  SetDx(e);
+  e.PolyTyp = Pt;
+}
+//------------------------------------------------------------------------------
+
+TEdge *RemoveEdge(TEdge *e) {
+  // removes e from double_linked_list (but without removing from memory)
+  e->Prev->Next = e->Next;
+  e->Next->Prev = e->Prev;
+  TEdge *result = e->Next;
+  e->Prev = 0; // flag as removed (see ClipperBase.Clear)
+  return result;
+}
+//------------------------------------------------------------------------------
+
+inline void ReverseHorizontal(TEdge &e) {
+  // swap horizontal edges' Top and Bottom x's so they follow the natural
+  // progression of the bounds - ie so their xbots will align with the
+  // adjoining lower edge. [Helpful in the ProcessHorizontal() method.]
+  std::swap(e.Top.X, e.Bot.X);
+#ifdef use_xyz
+  std::swap(e.Top.Z, e.Bot.Z);
+#endif
+}
+//------------------------------------------------------------------------------
+
+void SwapPoints(IntPoint &pt1, IntPoint &pt2) {
+  IntPoint tmp = pt1;
+  pt1 = pt2;
+  pt2 = tmp;
+}
+//------------------------------------------------------------------------------
+
+bool GetOverlapSegment(IntPoint pt1a, IntPoint pt1b, IntPoint pt2a,
+                       IntPoint pt2b, IntPoint &pt1, IntPoint &pt2) {
+  // precondition: segments are Collinear.
+  if (Abs(pt1a.X - pt1b.X) > Abs(pt1a.Y - pt1b.Y)) {
+    if (pt1a.X > pt1b.X)
+      SwapPoints(pt1a, pt1b);
+    if (pt2a.X > pt2b.X)
+      SwapPoints(pt2a, pt2b);
+    if (pt1a.X > pt2a.X)
+      pt1 = pt1a;
+    else
+      pt1 = pt2a;
+    if (pt1b.X < pt2b.X)
+      pt2 = pt1b;
+    else
+      pt2 = pt2b;
+    return pt1.X < pt2.X;
+  } else {
+    if (pt1a.Y < pt1b.Y)
+      SwapPoints(pt1a, pt1b);
+    if (pt2a.Y < pt2b.Y)
+      SwapPoints(pt2a, pt2b);
+    if (pt1a.Y < pt2a.Y)
+      pt1 = pt1a;
+    else
+      pt1 = pt2a;
+    if (pt1b.Y > pt2b.Y)
+      pt2 = pt1b;
+    else
+      pt2 = pt2b;
+    return pt1.Y > pt2.Y;
+  }
+}
+//------------------------------------------------------------------------------
+
+bool FirstIsBottomPt(const OutPt *btmPt1, const OutPt *btmPt2) {
+  OutPt *p = btmPt1->Prev;
+  while ((p->Pt == btmPt1->Pt) && (p != btmPt1))
+    p = p->Prev;
+  double dx1p = std::fabs(GetDx(btmPt1->Pt, p->Pt));
+  p = btmPt1->Next;
+  while ((p->Pt == btmPt1->Pt) && (p != btmPt1))
+    p = p->Next;
+  double dx1n = std::fabs(GetDx(btmPt1->Pt, p->Pt));
+
+  p = btmPt2->Prev;
+  while ((p->Pt == btmPt2->Pt) && (p != btmPt2))
+    p = p->Prev;
+  double dx2p = std::fabs(GetDx(btmPt2->Pt, p->Pt));
+  p = btmPt2->Next;
+  while ((p->Pt == btmPt2->Pt) && (p != btmPt2))
+    p = p->Next;
+  double dx2n = std::fabs(GetDx(btmPt2->Pt, p->Pt));
+
+  if (std::max(dx1p, dx1n) == std::max(dx2p, dx2n) &&
+      std::min(dx1p, dx1n) == std::min(dx2p, dx2n))
+    return Area(btmPt1) > 0; // if otherwise identical use orientation
+  else
+    return (dx1p >= dx2p && dx1p >= dx2n) || (dx1n >= dx2p && dx1n >= dx2n);
+}
+//------------------------------------------------------------------------------
+
+OutPt *GetBottomPt(OutPt *pp) {
+  OutPt *dups = 0;
+  OutPt *p = pp->Next;
+  while (p != pp) {
+    if (p->Pt.Y > pp->Pt.Y) {
+      pp = p;
+      dups = 0;
+    } else if (p->Pt.Y == pp->Pt.Y && p->Pt.X <= pp->Pt.X) {
+      if (p->Pt.X < pp->Pt.X) {
+        dups = 0;
+        pp = p;
+      } else {
+        if (p->Next != pp && p->Prev != pp)
+          dups = p;
+      }
+    }
+    p = p->Next;
+  }
+  if (dups) {
+    // there appears to be at least 2 vertices at BottomPt so ...
+    while (dups != p) {
+      if (!FirstIsBottomPt(p, dups))
+        pp = dups;
+      dups = dups->Next;
+      while (dups->Pt != pp->Pt)
+        dups = dups->Next;
+    }
+  }
+  return pp;
+}
+//------------------------------------------------------------------------------
+
+bool Pt2IsBetweenPt1AndPt3(const IntPoint pt1, const IntPoint pt2,
+                           const IntPoint pt3) {
+  if ((pt1 == pt3) || (pt1 == pt2) || (pt3 == pt2))
+    return false;
+  else if (pt1.X != pt3.X)
+    return (pt2.X > pt1.X) == (pt2.X < pt3.X);
+  else
+    return (pt2.Y > pt1.Y) == (pt2.Y < pt3.Y);
+}
+//------------------------------------------------------------------------------
+
+bool HorzSegmentsOverlap(cInt seg1a, cInt seg1b, cInt seg2a, cInt seg2b) {
+  if (seg1a > seg1b)
+    std::swap(seg1a, seg1b);
+  if (seg2a > seg2b)
+    std::swap(seg2a, seg2b);
+  return (seg1a < seg2b) && (seg2a < seg1b);
+}
+
+//------------------------------------------------------------------------------
+// ClipperBase class methods ...
+//------------------------------------------------------------------------------
+
+ClipperBase::ClipperBase() // constructor
+{
+  m_CurrentLM = m_MinimaList.begin(); // begin() == end() here
+  m_UseFullRange = false;
+}
+//------------------------------------------------------------------------------
+
+ClipperBase::~ClipperBase() // destructor
+{
+  Clear();
+}
+//------------------------------------------------------------------------------
+
+void RangeTest(const IntPoint &Pt, bool &useFullRange) {
+  if (useFullRange) {
+    if (Pt.X > hiRange || Pt.Y > hiRange || -Pt.X > hiRange || -Pt.Y > hiRange)
+      throw clipperException("Coordinate outside allowed range");
+  } else if (Pt.X > loRange || Pt.Y > loRange || -Pt.X > loRange ||
+             -Pt.Y > loRange) {
+    useFullRange = true;
+    RangeTest(Pt, useFullRange);
+  }
+}
+//------------------------------------------------------------------------------
+
+TEdge *FindNextLocMin(TEdge *E) {
+  for (;;) {
+    while (E->Bot != E->Prev->Bot || E->Curr == E->Top)
+      E = E->Next;
+    if (!IsHorizontal(*E) && !IsHorizontal(*E->Prev))
+      break;
+    while (IsHorizontal(*E->Prev))
+      E = E->Prev;
+    TEdge *E2 = E;
+    while (IsHorizontal(*E))
+      E = E->Next;
+    if (E->Top.Y == E->Prev->Bot.Y)
+      continue; // ie just an intermediate horz.
+    if (E2->Prev->Bot.X < E->Bot.X)
+      E = E2;
+    break;
+  }
+  return E;
+}
+//------------------------------------------------------------------------------
+
+TEdge *ClipperBase::ProcessBound(TEdge *E, bool NextIsForward) {
+  TEdge *Result = E;
+  TEdge *Horz = 0;
+
+  if (E->OutIdx == Skip) {
+    // if edges still remain in the current bound beyond the skip edge then
+    // create another LocMin and call ProcessBound once more
+    if (NextIsForward) {
+      while (E->Top.Y == E->Next->Bot.Y)
+        E = E->Next;
+      // don't include top horizontals when parsing a bound a second time,
+      // they will be contained in the opposite bound ...
+      while (E != Result && IsHorizontal(*E))
+        E = E->Prev;
+    } else {
+      while (E->Top.Y == E->Prev->Bot.Y)
+        E = E->Prev;
+      while (E != Result && IsHorizontal(*E))
+        E = E->Next;
+    }
+
+    if (E == Result) {
+      if (NextIsForward)
+        Result = E->Next;
+      else
+        Result = E->Prev;
+    } else {
+      // there are more edges in the bound beyond result starting with E
+      if (NextIsForward)
+        E = Result->Next;
+      else
+        E = Result->Prev;
+      MinimaList::value_type locMin;
+      locMin.Y = E->Bot.Y;
+      locMin.LeftBound = 0;
+      locMin.RightBound = E;
+      E->WindDelta = 0;
+      Result = ProcessBound(E, NextIsForward);
+      m_MinimaList.push_back(locMin);
+    }
+    return Result;
+  }
+
+  TEdge *EStart;
+
+  if (IsHorizontal(*E)) {
+    // We need to be careful with open paths because this may not be a
+    // true local minima (ie E may be following a skip edge).
+    // Also, consecutive horz. edges may start heading left before going right.
+    if (NextIsForward)
+      EStart = E->Prev;
+    else
+      EStart = E->Next;
+    if (IsHorizontal(*EStart)) // ie an adjoining horizontal skip edge
+    {
+      if (EStart->Bot.X != E->Bot.X && EStart->Top.X != E->Bot.X)
+        ReverseHorizontal(*E);
+    } else if (EStart->Bot.X != E->Bot.X)
+      ReverseHorizontal(*E);
+  }
+
+  EStart = E;
+  if (NextIsForward) {
+    while (Result->Top.Y == Result->Next->Bot.Y && Result->Next->OutIdx != Skip)
+      Result = Result->Next;
+    if (IsHorizontal(*Result) && Result->Next->OutIdx != Skip) {
+      // nb: at the top of a bound, horizontals are added to the bound
+      // only when the preceding edge attaches to the horizontal's left vertex
+      // unless a Skip edge is encountered when that becomes the top divide
+      Horz = Result;
+      while (IsHorizontal(*Horz->Prev))
+        Horz = Horz->Prev;
+      if (Horz->Prev->Top.X > Result->Next->Top.X)
+        Result = Horz->Prev;
+    }
+    while (E != Result) {
+      E->NextInLML = E->Next;
+      if (IsHorizontal(*E) && E != EStart && E->Bot.X != E->Prev->Top.X)
+        ReverseHorizontal(*E);
+      E = E->Next;
+    }
+    if (IsHorizontal(*E) && E != EStart && E->Bot.X != E->Prev->Top.X)
+      ReverseHorizontal(*E);
+    Result = Result->Next; // move to the edge just beyond current bound
+  } else {
+    while (Result->Top.Y == Result->Prev->Bot.Y && Result->Prev->OutIdx != Skip)
+      Result = Result->Prev;
+    if (IsHorizontal(*Result) && Result->Prev->OutIdx != Skip) {
+      Horz = Result;
+      while (IsHorizontal(*Horz->Next))
+        Horz = Horz->Next;
+      if (Horz->Next->Top.X == Result->Prev->Top.X ||
+          Horz->Next->Top.X > Result->Prev->Top.X)
+        Result = Horz->Next;
+    }
+
+    while (E != Result) {
+      E->NextInLML = E->Prev;
+      if (IsHorizontal(*E) && E != EStart && E->Bot.X != E->Next->Top.X)
+        ReverseHorizontal(*E);
+      E = E->Prev;
+    }
+    if (IsHorizontal(*E) && E != EStart && E->Bot.X != E->Next->Top.X)
+      ReverseHorizontal(*E);
+    Result = Result->Prev; // move to the edge just beyond current bound
+  }
+
+  return Result;
+}
+//------------------------------------------------------------------------------
+
+bool ClipperBase::AddPath(const Path &pg, PolyType PolyTyp, bool Closed) {
+#ifdef use_lines
+  if (!Closed && PolyTyp == ptClip)
+    throw clipperException("AddPath: Open paths must be subject.");
+#else
+  if (!Closed)
+    throw clipperException("AddPath: Open paths have been disabled.");
+#endif
+
+  int highI = (int)pg.size() - 1;
+  if (Closed)
+    while (highI > 0 && (pg[highI] == pg[0]))
+      --highI;
+  while (highI > 0 && (pg[highI] == pg[highI - 1]))
+    --highI;
+  if ((Closed && highI < 2) || (!Closed && highI < 1))
+    return false;
+
+  // create a new edge array ...
+  TEdge *edges = new TEdge[highI + 1];
+
+  bool IsFlat = true;
+  // 1. Basic (first) edge initialization ...
+  try {
+    edges[1].Curr = pg[1];
+    RangeTest(pg[0], m_UseFullRange);
+    RangeTest(pg[highI], m_UseFullRange);
+    InitEdge(&edges[0], &edges[1], &edges[highI], pg[0]);
+    InitEdge(&edges[highI], &edges[0], &edges[highI - 1], pg[highI]);
+    for (int i = highI - 1; i >= 1; --i) {
+      RangeTest(pg[i], m_UseFullRange);
+      InitEdge(&edges[i], &edges[i + 1], &edges[i - 1], pg[i]);
+    }
+  } catch (...) {
+    delete[] edges;
+    throw; // range test fails
+  }
+  TEdge *eStart = &edges[0];
+
+  // 2. Remove duplicate vertices, and (when closed) collinear edges ...
+  TEdge *E = eStart, *eLoopStop = eStart;
+  for (;;) {
+    // nb: allows matching start and end points when not Closed ...
+    if (E->Curr == E->Next->Curr && (Closed || E->Next != eStart)) {
+      if (E == E->Next)
+        break;
+      if (E == eStart)
+        eStart = E->Next;
+      E = RemoveEdge(E);
+      eLoopStop = E;
+      continue;
+    }
+    if (E->Prev == E->Next)
+      break; // only two vertices
+    else if (Closed &&
+             SlopesEqual(E->Prev->Curr, E->Curr, E->Next->Curr,
+                         m_UseFullRange) &&
+             (!m_PreserveCollinear ||
+              !Pt2IsBetweenPt1AndPt3(E->Prev->Curr, E->Curr, E->Next->Curr))) {
+      // Collinear edges are allowed for open paths but in closed paths
+      // the default is to merge adjacent collinear edges into a single edge.
+      // However, if the PreserveCollinear property is enabled, only overlapping
+      // collinear edges (ie spikes) will be removed from closed paths.
+      if (E == eStart)
+        eStart = E->Next;
+      E = RemoveEdge(E);
+      E = E->Prev;
+      eLoopStop = E;
+      continue;
+    }
+    E = E->Next;
+    if ((E == eLoopStop) || (!Closed && E->Next == eStart))
+      break;
+  }
+
+  if ((!Closed && (E == E->Next)) || (Closed && (E->Prev == E->Next))) {
+    delete[] edges;
+    return false;
+  }
+
+  if (!Closed) {
+    m_HasOpenPaths = true;
+    eStart->Prev->OutIdx = Skip;
+  }
+
+  // 3. Do second stage of edge initialization ...
+  E = eStart;
+  do {
+    InitEdge2(*E, PolyTyp);
+    E = E->Next;
+    if (IsFlat && E->Curr.Y != eStart->Curr.Y)
+      IsFlat = false;
+  } while (E != eStart);
+
+  // 4. Finally, add edge bounds to LocalMinima list ...
+
+  // Totally flat paths must be handled differently when adding them
+  // to LocalMinima list to avoid endless loops etc ...
+  if (IsFlat) {
+    if (Closed) {
+      delete[] edges;
+      return false;
+    }
+    E->Prev->OutIdx = Skip;
+    MinimaList::value_type locMin;
+    locMin.Y = E->Bot.Y;
+    locMin.LeftBound = 0;
+    locMin.RightBound = E;
+    locMin.RightBound->Side = esRight;
+    locMin.RightBound->WindDelta = 0;
+    for (;;) {
+      if (E->Bot.X != E->Prev->Top.X)
+        ReverseHorizontal(*E);
+      if (E->Next->OutIdx == Skip)
+        break;
+      E->NextInLML = E->Next;
+      E = E->Next;
+    }
+    m_MinimaList.push_back(locMin);
+    m_edges.push_back(edges);
+    return true;
+  }
+
+  m_edges.push_back(edges);
+  bool leftBoundIsForward;
+  TEdge *EMin = 0;
+
+  // workaround to avoid an endless loop in the while loop below when
+  // open paths have matching start and end points ...
+  if (E->Prev->Bot == E->Prev->Top)
+    E = E->Next;
+
+  for (;;) {
+    E = FindNextLocMin(E);
+    if (E == EMin)
+      break;
+    else if (!EMin)
+      EMin = E;
+
+    // E and E.Prev now share a local minima (left aligned if horizontal).
+    // Compare their slopes to find which starts which bound ...
+    MinimaList::value_type locMin;
+    locMin.Y = E->Bot.Y;
+    if (E->Dx < E->Prev->Dx) {
+      locMin.LeftBound = E->Prev;
+      locMin.RightBound = E;
+      leftBoundIsForward = false; // Q.nextInLML = Q.prev
+    } else {
+      locMin.LeftBound = E;
+      locMin.RightBound = E->Prev;
+      leftBoundIsForward = true; // Q.nextInLML = Q.next
+    }
+
+    if (!Closed)
+      locMin.LeftBound->WindDelta = 0;
+    else if (locMin.LeftBound->Next == locMin.RightBound)
+      locMin.LeftBound->WindDelta = -1;
+    else
+      locMin.LeftBound->WindDelta = 1;
+    locMin.RightBound->WindDelta = -locMin.LeftBound->WindDelta;
+
+    E = ProcessBound(locMin.LeftBound, leftBoundIsForward);
+    if (E->OutIdx == Skip)
+      E = ProcessBound(E, leftBoundIsForward);
+
+    TEdge *E2 = ProcessBound(locMin.RightBound, !leftBoundIsForward);
+    if (E2->OutIdx == Skip)
+      E2 = ProcessBound(E2, !leftBoundIsForward);
+
+    if (locMin.LeftBound->OutIdx == Skip)
+      locMin.LeftBound = 0;
+    else if (locMin.RightBound->OutIdx == Skip)
+      locMin.RightBound = 0;
+    m_MinimaList.push_back(locMin);
+    if (!leftBoundIsForward)
+      E = E2;
+  }
+  return true;
+}
+//------------------------------------------------------------------------------
+
+bool ClipperBase::AddPaths(const Paths &ppg, PolyType PolyTyp, bool Closed) {
+  bool result = false;
+  for (Paths::size_type i = 0; i < ppg.size(); ++i)
+    if (AddPath(ppg[i], PolyTyp, Closed))
+      result = true;
+  return result;
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::Clear() {
+  DisposeLocalMinimaList();
+  for (EdgeList::size_type i = 0; i < m_edges.size(); ++i) {
+    TEdge *edges = m_edges[i];
+    delete[] edges;
+  }
+  m_edges.clear();
+  m_UseFullRange = false;
+  m_HasOpenPaths = false;
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::Reset() {
+  m_CurrentLM = m_MinimaList.begin();
+  if (m_CurrentLM == m_MinimaList.end())
+    return; // ie nothing to process
+  std::sort(m_MinimaList.begin(), m_MinimaList.end(), LocMinSorter());
+
+  m_Scanbeam = ScanbeamList(); // clears/resets priority_queue
+  // reset all edges ...
+  for (MinimaList::iterator lm = m_MinimaList.begin(); lm != m_MinimaList.end();
+       ++lm) {
+    InsertScanbeam(lm->Y);
+    TEdge *e = lm->LeftBound;
+    if (e) {
+      e->Curr = e->Bot;
+      e->Side = esLeft;
+      e->OutIdx = Unassigned;
+    }
+
+    e = lm->RightBound;
+    if (e) {
+      e->Curr = e->Bot;
+      e->Side = esRight;
+      e->OutIdx = Unassigned;
+    }
+  }
+  m_ActiveEdges = 0;
+  m_CurrentLM = m_MinimaList.begin();
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::DisposeLocalMinimaList() {
+  m_MinimaList.clear();
+  m_CurrentLM = m_MinimaList.begin();
+}
+//------------------------------------------------------------------------------
+
+bool ClipperBase::PopLocalMinima(cInt Y, const LocalMinimum *&locMin) {
+  if (m_CurrentLM == m_MinimaList.end() || (*m_CurrentLM).Y != Y)
+    return false;
+  locMin = &(*m_CurrentLM);
+  ++m_CurrentLM;
+  return true;
+}
+//------------------------------------------------------------------------------
+
+IntRect ClipperBase::GetBounds() {
+  IntRect result;
+  MinimaList::iterator lm = m_MinimaList.begin();
+  if (lm == m_MinimaList.end()) {
+    result.left = result.top = result.right = result.bottom = 0;
+    return result;
+  }
+  result.left = lm->LeftBound->Bot.X;
+  result.top = lm->LeftBound->Bot.Y;
+  result.right = lm->LeftBound->Bot.X;
+  result.bottom = lm->LeftBound->Bot.Y;
+  while (lm != m_MinimaList.end()) {
+    // todo - needs fixing for open paths
+    result.bottom = std::max(result.bottom, lm->LeftBound->Bot.Y);
+    TEdge *e = lm->LeftBound;
+    for (;;) {
+      TEdge *bottomE = e;
+      while (e->NextInLML) {
+        if (e->Bot.X < result.left)
+          result.left = e->Bot.X;
+        if (e->Bot.X > result.right)
+          result.right = e->Bot.X;
+        e = e->NextInLML;
+      }
+      result.left = std::min(result.left, e->Bot.X);
+      result.right = std::max(result.right, e->Bot.X);
+      result.left = std::min(result.left, e->Top.X);
+      result.right = std::max(result.right, e->Top.X);
+      result.top = std::min(result.top, e->Top.Y);
+      if (bottomE == lm->LeftBound)
+        e = lm->RightBound;
+      else
+        break;
+    }
+    ++lm;
+  }
+  return result;
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::InsertScanbeam(const cInt Y) { m_Scanbeam.push(Y); }
+//------------------------------------------------------------------------------
+
+bool ClipperBase::PopScanbeam(cInt &Y) {
+  if (m_Scanbeam.empty())
+    return false;
+  Y = m_Scanbeam.top();
+  m_Scanbeam.pop();
+  while (!m_Scanbeam.empty() && Y == m_Scanbeam.top()) {
+    m_Scanbeam.pop();
+  } // Pop duplicates.
+  return true;
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::DisposeAllOutRecs() {
+  for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); ++i)
+    DisposeOutRec(i);
+  m_PolyOuts.clear();
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::DisposeOutRec(PolyOutList::size_type index) {
+  OutRec *outRec = m_PolyOuts[index];
+  if (outRec->Pts)
+    DisposeOutPts(outRec->Pts);
+  delete outRec;
+  m_PolyOuts[index] = 0;
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::DeleteFromAEL(TEdge *e) {
+  TEdge *AelPrev = e->PrevInAEL;
+  TEdge *AelNext = e->NextInAEL;
+  if (!AelPrev && !AelNext && (e != m_ActiveEdges))
+    return; // already deleted
+  if (AelPrev)
+    AelPrev->NextInAEL = AelNext;
+  else
+    m_ActiveEdges = AelNext;
+  if (AelNext)
+    AelNext->PrevInAEL = AelPrev;
+  e->NextInAEL = 0;
+  e->PrevInAEL = 0;
+}
+//------------------------------------------------------------------------------
+
+OutRec *ClipperBase::CreateOutRec() {
+  OutRec *result = new OutRec;
+  result->IsHole = false;
+  result->IsOpen = false;
+  result->FirstLeft = 0;
+  result->Pts = 0;
+  result->BottomPt = 0;
+  result->PolyNd = 0;
+  m_PolyOuts.push_back(result);
+  result->Idx = (int)m_PolyOuts.size() - 1;
+  return result;
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::SwapPositionsInAEL(TEdge *Edge1, TEdge *Edge2) {
+  // check that one or other edge hasn't already been removed from AEL ...
+  if (Edge1->NextInAEL == Edge1->PrevInAEL ||
+      Edge2->NextInAEL == Edge2->PrevInAEL)
+    return;
+
+  if (Edge1->NextInAEL == Edge2) {
+    TEdge *Next = Edge2->NextInAEL;
+    if (Next)
+      Next->PrevInAEL = Edge1;
+    TEdge *Prev = Edge1->PrevInAEL;
+    if (Prev)
+      Prev->NextInAEL = Edge2;
+    Edge2->PrevInAEL = Prev;
+    Edge2->NextInAEL = Edge1;
+    Edge1->PrevInAEL = Edge2;
+    Edge1->NextInAEL = Next;
+  } else if (Edge2->NextInAEL == Edge1) {
+    TEdge *Next = Edge1->NextInAEL;
+    if (Next)
+      Next->PrevInAEL = Edge2;
+    TEdge *Prev = Edge2->PrevInAEL;
+    if (Prev)
+      Prev->NextInAEL = Edge1;
+    Edge1->PrevInAEL = Prev;
+    Edge1->NextInAEL = Edge2;
+    Edge2->PrevInAEL = Edge1;
+    Edge2->NextInAEL = Next;
+  } else {
+    TEdge *Next = Edge1->NextInAEL;
+    TEdge *Prev = Edge1->PrevInAEL;
+    Edge1->NextInAEL = Edge2->NextInAEL;
+    if (Edge1->NextInAEL)
+      Edge1->NextInAEL->PrevInAEL = Edge1;
+    Edge1->PrevInAEL = Edge2->PrevInAEL;
+    if (Edge1->PrevInAEL)
+      Edge1->PrevInAEL->NextInAEL = Edge1;
+    Edge2->NextInAEL = Next;
+    if (Edge2->NextInAEL)
+      Edge2->NextInAEL->PrevInAEL = Edge2;
+    Edge2->PrevInAEL = Prev;
+    if (Edge2->PrevInAEL)
+      Edge2->PrevInAEL->NextInAEL = Edge2;
+  }
+
+  if (!Edge1->PrevInAEL)
+    m_ActiveEdges = Edge1;
+  else if (!Edge2->PrevInAEL)
+    m_ActiveEdges = Edge2;
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::UpdateEdgeIntoAEL(TEdge *&e) {
+  if (!e->NextInLML)
+    throw clipperException("UpdateEdgeIntoAEL: invalid call");
+
+  e->NextInLML->OutIdx = e->OutIdx;
+  TEdge *AelPrev = e->PrevInAEL;
+  TEdge *AelNext = e->NextInAEL;
+  if (AelPrev)
+    AelPrev->NextInAEL = e->NextInLML;
+  else
+    m_ActiveEdges = e->NextInLML;
+  if (AelNext)
+    AelNext->PrevInAEL = e->NextInLML;
+  e->NextInLML->Side = e->Side;
+  e->NextInLML->WindDelta = e->WindDelta;
+  e->NextInLML->WindCnt = e->WindCnt;
+  e->NextInLML->WindCnt2 = e->WindCnt2;
+  e = e->NextInLML;
+  e->Curr = e->Bot;
+  e->PrevInAEL = AelPrev;
+  e->NextInAEL = AelNext;
+  if (!IsHorizontal(*e))
+    InsertScanbeam(e->Top.Y);
+}
+//------------------------------------------------------------------------------
+
+bool ClipperBase::LocalMinimaPending() {
+  return (m_CurrentLM != m_MinimaList.end());
+}
+
+//------------------------------------------------------------------------------
+// TClipper methods ...
+//------------------------------------------------------------------------------
+
+Clipper::Clipper(int initOptions)
+    : ClipperBase() // constructor
+{
+  m_ExecuteLocked = false;
+  m_UseFullRange = false;
+  m_ReverseOutput = ((initOptions & ioReverseSolution) != 0);
+  m_StrictSimple = ((initOptions & ioStrictlySimple) != 0);
+  m_PreserveCollinear = ((initOptions & ioPreserveCollinear) != 0);
+  m_HasOpenPaths = false;
+#ifdef use_xyz
+  m_ZFill = 0;
+#endif
+}
+//------------------------------------------------------------------------------
+
+#ifdef use_xyz
+void Clipper::ZFillFunction(ZFillCallback zFillFunc) { m_ZFill = zFillFunc; }
+//------------------------------------------------------------------------------
+#endif
+
+bool Clipper::Execute(ClipType clipType, Paths &solution,
+                      PolyFillType fillType) {
+  return Execute(clipType, solution, fillType, fillType);
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::Execute(ClipType clipType, PolyTree &polytree,
+                      PolyFillType fillType) {
+  return Execute(clipType, polytree, fillType, fillType);
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::Execute(ClipType clipType, Paths &solution,
+                      PolyFillType subjFillType, PolyFillType clipFillType) {
+  if (m_ExecuteLocked)
+    return false;
+  if (m_HasOpenPaths)
+    throw clipperException(
+        "Error: PolyTree struct is needed for open path clipping.");
+  m_ExecuteLocked = true;
+  solution.resize(0);
+  m_SubjFillType = subjFillType;
+  m_ClipFillType = clipFillType;
+  m_ClipType = clipType;
+  m_UsingPolyTree = false;
+  bool succeeded = ExecuteInternal();
+  if (succeeded)
+    BuildResult(solution);
+  DisposeAllOutRecs();
+  m_ExecuteLocked = false;
+  return succeeded;
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::Execute(ClipType clipType, PolyTree &polytree,
+                      PolyFillType subjFillType, PolyFillType clipFillType) {
+  if (m_ExecuteLocked)
+    return false;
+  m_ExecuteLocked = true;
+  m_SubjFillType = subjFillType;
+  m_ClipFillType = clipFillType;
+  m_ClipType = clipType;
+  m_UsingPolyTree = true;
+  bool succeeded = ExecuteInternal();
+  if (succeeded)
+    BuildResult2(polytree);
+  DisposeAllOutRecs();
+  m_ExecuteLocked = false;
+  return succeeded;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::FixHoleLinkage(OutRec &outrec) {
+  // skip OutRecs that (a) contain outermost polygons or
+  //(b) already have the correct owner/child linkage ...
+  if (!outrec.FirstLeft ||
+      (outrec.IsHole != outrec.FirstLeft->IsHole && outrec.FirstLeft->Pts))
+    return;
+
+  OutRec *orfl = outrec.FirstLeft;
+  while (orfl && ((orfl->IsHole == outrec.IsHole) || !orfl->Pts))
+    orfl = orfl->FirstLeft;
+  outrec.FirstLeft = orfl;
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::ExecuteInternal() {
+  bool succeeded = true;
+  try {
+    Reset();
+    m_Maxima = MaximaList();
+    m_SortedEdges = 0;
+
+    succeeded = true;
+    cInt botY, topY;
+    if (!PopScanbeam(botY))
+      return false;
+    InsertLocalMinimaIntoAEL(botY);
+    while (PopScanbeam(topY) || LocalMinimaPending()) {
+      ProcessHorizontals();
+      ClearGhostJoins();
+      if (!ProcessIntersections(topY)) {
+        succeeded = false;
+        break;
+      }
+      ProcessEdgesAtTopOfScanbeam(topY);
+      botY = topY;
+      InsertLocalMinimaIntoAEL(botY);
+    }
+  } catch (...) {
+    succeeded = false;
+  }
+
+  if (succeeded) {
+    // fix orientations ...
+    for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); ++i) {
+      OutRec *outRec = m_PolyOuts[i];
+      if (!outRec->Pts || outRec->IsOpen)
+        continue;
+      if ((outRec->IsHole ^ m_ReverseOutput) == (Area(*outRec) > 0))
+        ReversePolyPtLinks(outRec->Pts);
+    }
+
+    if (!m_Joins.empty())
+      JoinCommonEdges();
+
+    // unfortunately FixupOutPolygon() must be done after JoinCommonEdges()
+    for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); ++i) {
+      OutRec *outRec = m_PolyOuts[i];
+      if (!outRec->Pts)
+        continue;
+      if (outRec->IsOpen)
+        FixupOutPolyline(*outRec);
+      else
+        FixupOutPolygon(*outRec);
+    }
+
+    if (m_StrictSimple)
+      DoSimplePolygons();
+  }
+
+  ClearJoins();
+  ClearGhostJoins();
+  return succeeded;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::SetWindingCount(TEdge &edge) {
+  TEdge *e = edge.PrevInAEL;
+  // find the edge of the same polytype that immediately precedes 'edge' in AEL
+  while (e && ((e->PolyTyp != edge.PolyTyp) || (e->WindDelta == 0)))
+    e = e->PrevInAEL;
+  if (!e) {
+    if (edge.WindDelta == 0) {
+      PolyFillType pft =
+          (edge.PolyTyp == ptSubject ? m_SubjFillType : m_ClipFillType);
+      edge.WindCnt = (pft == pftNegative ? -1 : 1);
+    } else
+      edge.WindCnt = edge.WindDelta;
+    edge.WindCnt2 = 0;
+    e = m_ActiveEdges; // ie get ready to calc WindCnt2
+  } else if (edge.WindDelta == 0 && m_ClipType != ctUnion) {
+    edge.WindCnt = 1;
+    edge.WindCnt2 = e->WindCnt2;
+    e = e->NextInAEL; // ie get ready to calc WindCnt2
+  } else if (IsEvenOddFillType(edge)) {
+    // EvenOdd filling ...
+    if (edge.WindDelta == 0) {
+      // are we inside a subj polygon ...
+      bool Inside = true;
+      TEdge *e2 = e->PrevInAEL;
+      while (e2) {
+        if (e2->PolyTyp == e->PolyTyp && e2->WindDelta != 0)
+          Inside = !Inside;
+        e2 = e2->PrevInAEL;
+      }
+      edge.WindCnt = (Inside ? 0 : 1);
+    } else {
+      edge.WindCnt = edge.WindDelta;
+    }
+    edge.WindCnt2 = e->WindCnt2;
+    e = e->NextInAEL; // ie get ready to calc WindCnt2
+  } else {
+    // nonZero, Positive or Negative filling ...
+    if (e->WindCnt * e->WindDelta < 0) {
+      // prev edge is 'decreasing' WindCount (WC) toward zero
+      // so we're outside the previous polygon ...
+      if (Abs(e->WindCnt) > 1) {
+        // outside prev poly but still inside another.
+        // when reversing direction of prev poly use the same WC
+        if (e->WindDelta * edge.WindDelta < 0)
+          edge.WindCnt = e->WindCnt;
+        // otherwise continue to 'decrease' WC ...
+        else
+          edge.WindCnt = e->WindCnt + edge.WindDelta;
+      } else
+        // now outside all polys of same polytype so set own WC ...
+        edge.WindCnt = (edge.WindDelta == 0 ? 1 : edge.WindDelta);
+    } else {
+      // prev edge is 'increasing' WindCount (WC) away from zero
+      // so we're inside the previous polygon ...
+      if (edge.WindDelta == 0)
+        edge.WindCnt = (e->WindCnt < 0 ? e->WindCnt - 1 : e->WindCnt + 1);
+      // if wind direction is reversing prev then use same WC
+      else if (e->WindDelta * edge.WindDelta < 0)
+        edge.WindCnt = e->WindCnt;
+      // otherwise add to WC ...
+      else
+        edge.WindCnt = e->WindCnt + edge.WindDelta;
+    }
+    edge.WindCnt2 = e->WindCnt2;
+    e = e->NextInAEL; // ie get ready to calc WindCnt2
+  }
+
+  // update WindCnt2 ...
+  if (IsEvenOddAltFillType(edge)) {
+    // EvenOdd filling ...
+    while (e != &edge) {
+      if (e->WindDelta != 0)
+        edge.WindCnt2 = (edge.WindCnt2 == 0 ? 1 : 0);
+      e = e->NextInAEL;
+    }
+  } else {
+    // nonZero, Positive or Negative filling ...
+    while (e != &edge) {
+      edge.WindCnt2 += e->WindDelta;
+      e = e->NextInAEL;
+    }
+  }
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::IsEvenOddFillType(const TEdge &edge) const {
+  if (edge.PolyTyp == ptSubject)
+    return m_SubjFillType == pftEvenOdd;
+  else
+    return m_ClipFillType == pftEvenOdd;
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::IsEvenOddAltFillType(const TEdge &edge) const {
+  if (edge.PolyTyp == ptSubject)
+    return m_ClipFillType == pftEvenOdd;
+  else
+    return m_SubjFillType == pftEvenOdd;
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::IsContributing(const TEdge &edge) const {
+  PolyFillType pft, pft2;
+  if (edge.PolyTyp == ptSubject) {
+    pft = m_SubjFillType;
+    pft2 = m_ClipFillType;
+  } else {
+    pft = m_ClipFillType;
+    pft2 = m_SubjFillType;
+  }
+
+  switch (pft) {
+  case pftEvenOdd:
+    // return false if a subj line has been flagged as inside a subj polygon
+    if (edge.WindDelta == 0 && edge.WindCnt != 1)
+      return false;
+    break;
+  case pftNonZero:
+    if (Abs(edge.WindCnt) != 1)
+      return false;
+    break;
+  case pftPositive:
+    if (edge.WindCnt != 1)
+      return false;
+    break;
+  default: // pftNegative
+    if (edge.WindCnt != -1)
+      return false;
+  }
+
+  switch (m_ClipType) {
+  case ctIntersection:
+    switch (pft2) {
+    case pftEvenOdd:
+    case pftNonZero:
+      return (edge.WindCnt2 != 0);
+    case pftPositive:
+      return (edge.WindCnt2 > 0);
+    default:
+      return (edge.WindCnt2 < 0);
+    }
+    break;
+  case ctUnion:
+    switch (pft2) {
+    case pftEvenOdd:
+    case pftNonZero:
+      return (edge.WindCnt2 == 0);
+    case pftPositive:
+      return (edge.WindCnt2 <= 0);
+    default:
+      return (edge.WindCnt2 >= 0);
+    }
+    break;
+  case ctDifference:
+    if (edge.PolyTyp == ptSubject)
+      switch (pft2) {
+      case pftEvenOdd:
+      case pftNonZero:
+        return (edge.WindCnt2 == 0);
+      case pftPositive:
+        return (edge.WindCnt2 <= 0);
+      default:
+        return (edge.WindCnt2 >= 0);
+      }
+    else
+      switch (pft2) {
+      case pftEvenOdd:
+      case pftNonZero:
+        return (edge.WindCnt2 != 0);
+      case pftPositive:
+        return (edge.WindCnt2 > 0);
+      default:
+        return (edge.WindCnt2 < 0);
+      }
+    break;
+  case ctXor:
+    if (edge.WindDelta == 0) // XOr always contributing unless open
+      switch (pft2) {
+      case pftEvenOdd:
+      case pftNonZero:
+        return (edge.WindCnt2 == 0);
+      case pftPositive:
+        return (edge.WindCnt2 <= 0);
+      default:
+        return (edge.WindCnt2 >= 0);
+      }
+    else
+      return true;
+    break;
+  default:
+    return true;
+  }
+}
+//------------------------------------------------------------------------------
+
+OutPt *Clipper::AddLocalMinPoly(TEdge *e1, TEdge *e2, const IntPoint &Pt) {
+  OutPt *result;
+  TEdge *e, *prevE;
+  if (IsHorizontal(*e2) || (e1->Dx > e2->Dx)) {
+    result = AddOutPt(e1, Pt);
+    e2->OutIdx = e1->OutIdx;
+    e1->Side = esLeft;
+    e2->Side = esRight;
+    e = e1;
+    if (e->PrevInAEL == e2)
+      prevE = e2->PrevInAEL;
+    else
+      prevE = e->PrevInAEL;
+  } else {
+    result = AddOutPt(e2, Pt);
+    e1->OutIdx = e2->OutIdx;
+    e1->Side = esRight;
+    e2->Side = esLeft;
+    e = e2;
+    if (e->PrevInAEL == e1)
+      prevE = e1->PrevInAEL;
+    else
+      prevE = e->PrevInAEL;
+  }
+
+  if (prevE && prevE->OutIdx >= 0 && prevE->Top.Y < Pt.Y && e->Top.Y < Pt.Y) {
+    cInt xPrev = TopX(*prevE, Pt.Y);
+    cInt xE = TopX(*e, Pt.Y);
+    if (xPrev == xE && (e->WindDelta != 0) && (prevE->WindDelta != 0) &&
+        SlopesEqual(IntPoint(xPrev, Pt.Y), prevE->Top, IntPoint(xE, Pt.Y),
+                    e->Top, m_UseFullRange)) {
+      OutPt *outPt = AddOutPt(prevE, Pt);
+      AddJoin(result, outPt, e->Top);
+    }
+  }
+  return result;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::AddLocalMaxPoly(TEdge *e1, TEdge *e2, const IntPoint &Pt) {
+  AddOutPt(e1, Pt);
+  if (e2->WindDelta == 0)
+    AddOutPt(e2, Pt);
+  if (e1->OutIdx == e2->OutIdx) {
+    e1->OutIdx = Unassigned;
+    e2->OutIdx = Unassigned;
+  } else if (e1->OutIdx < e2->OutIdx)
+    AppendPolygon(e1, e2);
+  else
+    AppendPolygon(e2, e1);
+}
+//------------------------------------------------------------------------------
+
+void Clipper::AddEdgeToSEL(TEdge *edge) {
+  // SEL pointers in PEdge are reused to build a list of horizontal edges.
+  // However, we don't need to worry about order with horizontal edge
+  // processing.
+  if (!m_SortedEdges) {
+    m_SortedEdges = edge;
+    edge->PrevInSEL = 0;
+    edge->NextInSEL = 0;
+  } else {
+    edge->NextInSEL = m_SortedEdges;
+    edge->PrevInSEL = 0;
+    m_SortedEdges->PrevInSEL = edge;
+    m_SortedEdges = edge;
+  }
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::PopEdgeFromSEL(TEdge *&edge) {
+  if (!m_SortedEdges)
+    return false;
+  edge = m_SortedEdges;
+  DeleteFromSEL(m_SortedEdges);
+  return true;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::CopyAELToSEL() {
+  TEdge *e = m_ActiveEdges;
+  m_SortedEdges = e;
+  while (e) {
+    e->PrevInSEL = e->PrevInAEL;
+    e->NextInSEL = e->NextInAEL;
+    e = e->NextInAEL;
+  }
+}
+//------------------------------------------------------------------------------
+
+void Clipper::AddJoin(OutPt *op1, OutPt *op2, const IntPoint OffPt) {
+  Join *j = new Join;
+  j->OutPt1 = op1;
+  j->OutPt2 = op2;
+  j->OffPt = OffPt;
+  m_Joins.push_back(j);
+}
+//------------------------------------------------------------------------------
+
+void Clipper::ClearJoins() {
+  for (JoinList::size_type i = 0; i < m_Joins.size(); i++)
+    delete m_Joins[i];
+  m_Joins.resize(0);
+}
+//------------------------------------------------------------------------------
+
+void Clipper::ClearGhostJoins() {
+  for (JoinList::size_type i = 0; i < m_GhostJoins.size(); i++)
+    delete m_GhostJoins[i];
+  m_GhostJoins.resize(0);
+}
+//------------------------------------------------------------------------------
+
+void Clipper::AddGhostJoin(OutPt *op, const IntPoint OffPt) {
+  Join *j = new Join;
+  j->OutPt1 = op;
+  j->OutPt2 = 0;
+  j->OffPt = OffPt;
+  m_GhostJoins.push_back(j);
+}
+//------------------------------------------------------------------------------
+
+void Clipper::InsertLocalMinimaIntoAEL(const cInt botY) {
+  const LocalMinimum *lm;
+  while (PopLocalMinima(botY, lm)) {
+    TEdge *lb = lm->LeftBound;
+    TEdge *rb = lm->RightBound;
+
+    OutPt *Op1 = 0;
+    if (!lb) {
+      // nb: don't insert LB into either AEL or SEL
+      InsertEdgeIntoAEL(rb, 0);
+      SetWindingCount(*rb);
+      if (IsContributing(*rb))
+        Op1 = AddOutPt(rb, rb->Bot);
+    } else if (!rb) {
+      InsertEdgeIntoAEL(lb, 0);
+      SetWindingCount(*lb);
+      if (IsContributing(*lb))
+        Op1 = AddOutPt(lb, lb->Bot);
+      InsertScanbeam(lb->Top.Y);
+    } else {
+      InsertEdgeIntoAEL(lb, 0);
+      InsertEdgeIntoAEL(rb, lb);
+      SetWindingCount(*lb);
+      rb->WindCnt = lb->WindCnt;
+      rb->WindCnt2 = lb->WindCnt2;
+      if (IsContributing(*lb))
+        Op1 = AddLocalMinPoly(lb, rb, lb->Bot);
+      InsertScanbeam(lb->Top.Y);
+    }
+
+    if (rb) {
+      if (IsHorizontal(*rb)) {
+        AddEdgeToSEL(rb);
+        if (rb->NextInLML)
+          InsertScanbeam(rb->NextInLML->Top.Y);
+      } else
+        InsertScanbeam(rb->Top.Y);
+    }
+
+    if (!lb || !rb)
+      continue;
+
+    // if any output polygons share an edge, they'll need joining later ...
+    if (Op1 && IsHorizontal(*rb) && m_GhostJoins.size() > 0 &&
+        (rb->WindDelta != 0)) {
+      for (JoinList::size_type i = 0; i < m_GhostJoins.size(); ++i) {
+        Join *jr = m_GhostJoins[i];
+        // if the horizontal Rb and a 'ghost' horizontal overlap, then convert
+        // the 'ghost' join to a real join ready for later ...
+        if (HorzSegmentsOverlap(jr->OutPt1->Pt.X, jr->OffPt.X, rb->Bot.X,
+                                rb->Top.X))
+          AddJoin(jr->OutPt1, Op1, jr->OffPt);
+      }
+    }
+
+    if (lb->OutIdx >= 0 && lb->PrevInAEL &&
+        lb->PrevInAEL->Curr.X == lb->Bot.X && lb->PrevInAEL->OutIdx >= 0 &&
+        SlopesEqual(lb->PrevInAEL->Bot, lb->PrevInAEL->Top, lb->Curr, lb->Top,
+                    m_UseFullRange) &&
+        (lb->WindDelta != 0) && (lb->PrevInAEL->WindDelta != 0)) {
+      OutPt *Op2 = AddOutPt(lb->PrevInAEL, lb->Bot);
+      AddJoin(Op1, Op2, lb->Top);
+    }
+
+    if (lb->NextInAEL != rb) {
+
+      if (rb->OutIdx >= 0 && rb->PrevInAEL->OutIdx >= 0 &&
+          SlopesEqual(rb->PrevInAEL->Curr, rb->PrevInAEL->Top, rb->Curr,
+                      rb->Top, m_UseFullRange) &&
+          (rb->WindDelta != 0) && (rb->PrevInAEL->WindDelta != 0)) {
+        OutPt *Op2 = AddOutPt(rb->PrevInAEL, rb->Bot);
+        AddJoin(Op1, Op2, rb->Top);
+      }
+
+      TEdge *e = lb->NextInAEL;
+      if (e) {
+        while (e != rb) {
+          // nb: For calculating winding counts etc, IntersectEdges() assumes
+          // that param1 will be to the Right of param2 ABOVE the intersection
+          // ...
+          IntersectEdges(rb, e, lb->Curr); // order important here
+          e = e->NextInAEL;
+        }
+      }
+    }
+  }
+}
+//------------------------------------------------------------------------------
+
+void Clipper::DeleteFromSEL(TEdge *e) {
+  TEdge *SelPrev = e->PrevInSEL;
+  TEdge *SelNext = e->NextInSEL;
+  if (!SelPrev && !SelNext && (e != m_SortedEdges))
+    return; // already deleted
+  if (SelPrev)
+    SelPrev->NextInSEL = SelNext;
+  else
+    m_SortedEdges = SelNext;
+  if (SelNext)
+    SelNext->PrevInSEL = SelPrev;
+  e->NextInSEL = 0;
+  e->PrevInSEL = 0;
+}
+//------------------------------------------------------------------------------
+
+#ifdef use_xyz
+void Clipper::SetZ(IntPoint &pt, TEdge &e1, TEdge &e2) {
+  if (pt.Z != 0 || !m_ZFill)
+    return;
+  else if (pt == e1.Bot)
+    pt.Z = e1.Bot.Z;
+  else if (pt == e1.Top)
+    pt.Z = e1.Top.Z;
+  else if (pt == e2.Bot)
+    pt.Z = e2.Bot.Z;
+  else if (pt == e2.Top)
+    pt.Z = e2.Top.Z;
+  else
+    (*m_ZFill)(e1.Bot, e1.Top, e2.Bot, e2.Top, pt);
+}
+//------------------------------------------------------------------------------
+#endif
+
+void Clipper::IntersectEdges(TEdge *e1, TEdge *e2, IntPoint &Pt) {
+  bool e1Contributing = (e1->OutIdx >= 0);
+  bool e2Contributing = (e2->OutIdx >= 0);
+
+#ifdef use_xyz
+  SetZ(Pt, *e1, *e2);
+#endif
+
+#ifdef use_lines
+  // if either edge is on an OPEN path ...
+  if (e1->WindDelta == 0 || e2->WindDelta == 0) {
+    // ignore subject-subject open path intersections UNLESS they
+    // are both open paths, AND they are both 'contributing maximas' ...
+    if (e1->WindDelta == 0 && e2->WindDelta == 0)
+      return;
+
+    // if intersecting a subj line with a subj poly ...
+    else if (e1->PolyTyp == e2->PolyTyp && e1->WindDelta != e2->WindDelta &&
+             m_ClipType == ctUnion) {
+      if (e1->WindDelta == 0) {
+        if (e2Contributing) {
+          AddOutPt(e1, Pt);
+          if (e1Contributing)
+            e1->OutIdx = Unassigned;
+        }
+      } else {
+        if (e1Contributing) {
+          AddOutPt(e2, Pt);
+          if (e2Contributing)
+            e2->OutIdx = Unassigned;
+        }
+      }
+    } else if (e1->PolyTyp != e2->PolyTyp) {
+      // toggle subj open path OutIdx on/off when Abs(clip.WndCnt) == 1 ...
+      if ((e1->WindDelta == 0) && abs(e2->WindCnt) == 1 &&
+          (m_ClipType != ctUnion || e2->WindCnt2 == 0)) {
+        AddOutPt(e1, Pt);
+        if (e1Contributing)
+          e1->OutIdx = Unassigned;
+      } else if ((e2->WindDelta == 0) && (abs(e1->WindCnt) == 1) &&
+                 (m_ClipType != ctUnion || e1->WindCnt2 == 0)) {
+        AddOutPt(e2, Pt);
+        if (e2Contributing)
+          e2->OutIdx = Unassigned;
+      }
+    }
+    return;
+  }
+#endif
+
+  // update winding counts...
+  // assumes that e1 will be to the Right of e2 ABOVE the intersection
+  if (e1->PolyTyp == e2->PolyTyp) {
+    if (IsEvenOddFillType(*e1)) {
+      int oldE1WindCnt = e1->WindCnt;
+      e1->WindCnt = e2->WindCnt;
+      e2->WindCnt = oldE1WindCnt;
+    } else {
+      if (e1->WindCnt + e2->WindDelta == 0)
+        e1->WindCnt = -e1->WindCnt;
+      else
+        e1->WindCnt += e2->WindDelta;
+      if (e2->WindCnt - e1->WindDelta == 0)
+        e2->WindCnt = -e2->WindCnt;
+      else
+        e2->WindCnt -= e1->WindDelta;
+    }
+  } else {
+    if (!IsEvenOddFillType(*e2))
+      e1->WindCnt2 += e2->WindDelta;
+    else
+      e1->WindCnt2 = (e1->WindCnt2 == 0) ? 1 : 0;
+    if (!IsEvenOddFillType(*e1))
+      e2->WindCnt2 -= e1->WindDelta;
+    else
+      e2->WindCnt2 = (e2->WindCnt2 == 0) ? 1 : 0;
+  }
+
+  PolyFillType e1FillType, e2FillType, e1FillType2, e2FillType2;
+  if (e1->PolyTyp == ptSubject) {
+    e1FillType = m_SubjFillType;
+    e1FillType2 = m_ClipFillType;
+  } else {
+    e1FillType = m_ClipFillType;
+    e1FillType2 = m_SubjFillType;
+  }
+  if (e2->PolyTyp == ptSubject) {
+    e2FillType = m_SubjFillType;
+    e2FillType2 = m_ClipFillType;
+  } else {
+    e2FillType = m_ClipFillType;
+    e2FillType2 = m_SubjFillType;
+  }
+
+  cInt e1Wc, e2Wc;
+  switch (e1FillType) {
+  case pftPositive:
+    e1Wc = e1->WindCnt;
+    break;
+  case pftNegative:
+    e1Wc = -e1->WindCnt;
+    break;
+  default:
+    e1Wc = Abs(e1->WindCnt);
+  }
+  switch (e2FillType) {
+  case pftPositive:
+    e2Wc = e2->WindCnt;
+    break;
+  case pftNegative:
+    e2Wc = -e2->WindCnt;
+    break;
+  default:
+    e2Wc = Abs(e2->WindCnt);
+  }
+
+  if (e1Contributing && e2Contributing) {
+    if ((e1Wc != 0 && e1Wc != 1) || (e2Wc != 0 && e2Wc != 1) ||
+        (e1->PolyTyp != e2->PolyTyp && m_ClipType != ctXor)) {
+      AddLocalMaxPoly(e1, e2, Pt);
+    } else {
+      AddOutPt(e1, Pt);
+      AddOutPt(e2, Pt);
+      SwapSides(*e1, *e2);
+      SwapPolyIndexes(*e1, *e2);
+    }
+  } else if (e1Contributing) {
+    if (e2Wc == 0 || e2Wc == 1) {
+      AddOutPt(e1, Pt);
+      SwapSides(*e1, *e2);
+      SwapPolyIndexes(*e1, *e2);
+    }
+  } else if (e2Contributing) {
+    if (e1Wc == 0 || e1Wc == 1) {
+      AddOutPt(e2, Pt);
+      SwapSides(*e1, *e2);
+      SwapPolyIndexes(*e1, *e2);
+    }
+  } else if ((e1Wc == 0 || e1Wc == 1) && (e2Wc == 0 || e2Wc == 1)) {
+    // neither edge is currently contributing ...
+
+    cInt e1Wc2, e2Wc2;
+    switch (e1FillType2) {
+    case pftPositive:
+      e1Wc2 = e1->WindCnt2;
+      break;
+    case pftNegative:
+      e1Wc2 = -e1->WindCnt2;
+      break;
+    default:
+      e1Wc2 = Abs(e1->WindCnt2);
+    }
+    switch (e2FillType2) {
+    case pftPositive:
+      e2Wc2 = e2->WindCnt2;
+      break;
+    case pftNegative:
+      e2Wc2 = -e2->WindCnt2;
+      break;
+    default:
+      e2Wc2 = Abs(e2->WindCnt2);
+    }
+
+    if (e1->PolyTyp != e2->PolyTyp) {
+      AddLocalMinPoly(e1, e2, Pt);
+    } else if (e1Wc == 1 && e2Wc == 1)
+      switch (m_ClipType) {
+      case ctIntersection:
+        if (e1Wc2 > 0 && e2Wc2 > 0)
+          AddLocalMinPoly(e1, e2, Pt);
+        break;
+      case ctUnion:
+        if (e1Wc2 <= 0 && e2Wc2 <= 0)
+          AddLocalMinPoly(e1, e2, Pt);
+        break;
+      case ctDifference:
+        if (((e1->PolyTyp == ptClip) && (e1Wc2 > 0) && (e2Wc2 > 0)) ||
+            ((e1->PolyTyp == ptSubject) && (e1Wc2 <= 0) && (e2Wc2 <= 0)))
+          AddLocalMinPoly(e1, e2, Pt);
+        break;
+      case ctXor:
+        AddLocalMinPoly(e1, e2, Pt);
+      }
+    else
+      SwapSides(*e1, *e2);
+  }
+}
+//------------------------------------------------------------------------------
+
+void Clipper::SetHoleState(TEdge *e, OutRec *outrec) {
+  TEdge *e2 = e->PrevInAEL;
+  TEdge *eTmp = 0;
+  while (e2) {
+    if (e2->OutIdx >= 0 && e2->WindDelta != 0) {
+      if (!eTmp)
+        eTmp = e2;
+      else if (eTmp->OutIdx == e2->OutIdx)
+        eTmp = 0;
+    }
+    e2 = e2->PrevInAEL;
+  }
+  if (!eTmp) {
+    outrec->FirstLeft = 0;
+    outrec->IsHole = false;
+  } else {
+    outrec->FirstLeft = m_PolyOuts[eTmp->OutIdx];
+    outrec->IsHole = !outrec->FirstLeft->IsHole;
+  }
+}
+//------------------------------------------------------------------------------
+
+OutRec *GetLowermostRec(OutRec *outRec1, OutRec *outRec2) {
+  // work out which polygon fragment has the correct hole state ...
+  if (!outRec1->BottomPt)
+    outRec1->BottomPt = GetBottomPt(outRec1->Pts);
+  if (!outRec2->BottomPt)
+    outRec2->BottomPt = GetBottomPt(outRec2->Pts);
+  OutPt *OutPt1 = outRec1->BottomPt;
+  OutPt *OutPt2 = outRec2->BottomPt;
+  if (OutPt1->Pt.Y > OutPt2->Pt.Y)
+    return outRec1;
+  else if (OutPt1->Pt.Y < OutPt2->Pt.Y)
+    return outRec2;
+  else if (OutPt1->Pt.X < OutPt2->Pt.X)
+    return outRec1;
+  else if (OutPt1->Pt.X > OutPt2->Pt.X)
+    return outRec2;
+  else if (OutPt1->Next == OutPt1)
+    return outRec2;
+  else if (OutPt2->Next == OutPt2)
+    return outRec1;
+  else if (FirstIsBottomPt(OutPt1, OutPt2))
+    return outRec1;
+  else
+    return outRec2;
+}
+//------------------------------------------------------------------------------
+
+bool OutRec1RightOfOutRec2(OutRec *outRec1, OutRec *outRec2) {
+  do {
+    outRec1 = outRec1->FirstLeft;
+    if (outRec1 == outRec2)
+      return true;
+  } while (outRec1);
+  return false;
+}
+//------------------------------------------------------------------------------
+
+OutRec *Clipper::GetOutRec(int Idx) {
+  OutRec *outrec = m_PolyOuts[Idx];
+  while (outrec != m_PolyOuts[outrec->Idx])
+    outrec = m_PolyOuts[outrec->Idx];
+  return outrec;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::AppendPolygon(TEdge *e1, TEdge *e2) {
+  // get the start and ends of both output polygons ...
+  OutRec *outRec1 = m_PolyOuts[e1->OutIdx];
+  OutRec *outRec2 = m_PolyOuts[e2->OutIdx];
+
+  OutRec *holeStateRec;
+  if (OutRec1RightOfOutRec2(outRec1, outRec2))
+    holeStateRec = outRec2;
+  else if (OutRec1RightOfOutRec2(outRec2, outRec1))
+    holeStateRec = outRec1;
+  else
+    holeStateRec = GetLowermostRec(outRec1, outRec2);
+
+  // get the start and ends of both output polygons and
+  // join e2 poly onto e1 poly and delete pointers to e2 ...
+
+  OutPt *p1_lft = outRec1->Pts;
+  OutPt *p1_rt = p1_lft->Prev;
+  OutPt *p2_lft = outRec2->Pts;
+  OutPt *p2_rt = p2_lft->Prev;
+
+  // join e2 poly onto e1 poly and delete pointers to e2 ...
+  if (e1->Side == esLeft) {
+    if (e2->Side == esLeft) {
+      // z y x a b c
+      ReversePolyPtLinks(p2_lft);
+      p2_lft->Next = p1_lft;
+      p1_lft->Prev = p2_lft;
+      p1_rt->Next = p2_rt;
+      p2_rt->Prev = p1_rt;
+      outRec1->Pts = p2_rt;
+    } else {
+      // x y z a b c
+      p2_rt->Next = p1_lft;
+      p1_lft->Prev = p2_rt;
+      p2_lft->Prev = p1_rt;
+      p1_rt->Next = p2_lft;
+      outRec1->Pts = p2_lft;
+    }
+  } else {
+    if (e2->Side == esRight) {
+      // a b c z y x
+      ReversePolyPtLinks(p2_lft);
+      p1_rt->Next = p2_rt;
+      p2_rt->Prev = p1_rt;
+      p2_lft->Next = p1_lft;
+      p1_lft->Prev = p2_lft;
+    } else {
+      // a b c x y z
+      p1_rt->Next = p2_lft;
+      p2_lft->Prev = p1_rt;
+      p1_lft->Prev = p2_rt;
+      p2_rt->Next = p1_lft;
+    }
+  }
+
+  outRec1->BottomPt = 0;
+  if (holeStateRec == outRec2) {
+    if (outRec2->FirstLeft != outRec1)
+      outRec1->FirstLeft = outRec2->FirstLeft;
+    outRec1->IsHole = outRec2->IsHole;
+  }
+  outRec2->Pts = 0;
+  outRec2->BottomPt = 0;
+  outRec2->FirstLeft = outRec1;
+
+  int OKIdx = e1->OutIdx;
+  int ObsoleteIdx = e2->OutIdx;
+
+  e1->OutIdx =
+      Unassigned; // nb: safe because we only get here via AddLocalMaxPoly
+  e2->OutIdx = Unassigned;
+
+  TEdge *e = m_ActiveEdges;
+  while (e) {
+    if (e->OutIdx == ObsoleteIdx) {
+      e->OutIdx = OKIdx;
+      e->Side = e1->Side;
+      break;
+    }
+    e = e->NextInAEL;
+  }
+
+  outRec2->Idx = outRec1->Idx;
+}
+//------------------------------------------------------------------------------
+
+OutPt *Clipper::AddOutPt(TEdge *e, const IntPoint &pt) {
+  if (e->OutIdx < 0) {
+    OutRec *outRec = CreateOutRec();
+    outRec->IsOpen = (e->WindDelta == 0);
+    OutPt *newOp = new OutPt;
+    outRec->Pts = newOp;
+    newOp->Idx = outRec->Idx;
+    newOp->Pt = pt;
+    newOp->Next = newOp;
+    newOp->Prev = newOp;
+    if (!outRec->IsOpen)
+      SetHoleState(e, outRec);
+    e->OutIdx = outRec->Idx;
+    return newOp;
+  } else {
+    OutRec *outRec = m_PolyOuts[e->OutIdx];
+    // OutRec.Pts is the 'Left-most' point & OutRec.Pts.Prev is the 'Right-most'
+    OutPt *op = outRec->Pts;
+
+    bool ToFront = (e->Side == esLeft);
+    if (ToFront && (pt == op->Pt))
+      return op;
+    else if (!ToFront && (pt == op->Prev->Pt))
+      return op->Prev;
+
+    OutPt *newOp = new OutPt;
+    newOp->Idx = outRec->Idx;
+    newOp->Pt = pt;
+    newOp->Next = op;
+    newOp->Prev = op->Prev;
+    newOp->Prev->Next = newOp;
+    op->Prev = newOp;
+    if (ToFront)
+      outRec->Pts = newOp;
+    return newOp;
+  }
+}
+//------------------------------------------------------------------------------
+
+OutPt *Clipper::GetLastOutPt(TEdge *e) {
+  OutRec *outRec = m_PolyOuts[e->OutIdx];
+  if (e->Side == esLeft)
+    return outRec->Pts;
+  else
+    return outRec->Pts->Prev;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::ProcessHorizontals() {
+  TEdge *horzEdge;
+  while (PopEdgeFromSEL(horzEdge))
+    ProcessHorizontal(horzEdge);
+}
+//------------------------------------------------------------------------------
+
+inline bool IsMinima(TEdge *e) {
+  return e && (e->Prev->NextInLML != e) && (e->Next->NextInLML != e);
+}
+//------------------------------------------------------------------------------
+
+inline bool IsMaxima(TEdge *e, const cInt Y) {
+  return e && e->Top.Y == Y && !e->NextInLML;
+}
+//------------------------------------------------------------------------------
+
+inline bool IsIntermediate(TEdge *e, const cInt Y) {
+  return e->Top.Y == Y && e->NextInLML;
+}
+//------------------------------------------------------------------------------
+
+TEdge *GetMaximaPair(TEdge *e) {
+  if ((e->Next->Top == e->Top) && !e->Next->NextInLML)
+    return e->Next;
+  else if ((e->Prev->Top == e->Top) && !e->Prev->NextInLML)
+    return e->Prev;
+  else
+    return 0;
+}
+//------------------------------------------------------------------------------
+
+TEdge *GetMaximaPairEx(TEdge *e) {
+  // as GetMaximaPair() but returns 0 if MaxPair isn't in AEL (unless it's
+  // horizontal)
+  TEdge *result = GetMaximaPair(e);
+  if (result &&
+      (result->OutIdx == Skip ||
+       (result->NextInAEL == result->PrevInAEL && !IsHorizontal(*result))))
+    return 0;
+  return result;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::SwapPositionsInSEL(TEdge *Edge1, TEdge *Edge2) {
+  if (!(Edge1->NextInSEL) && !(Edge1->PrevInSEL))
+    return;
+  if (!(Edge2->NextInSEL) && !(Edge2->PrevInSEL))
+    return;
+
+  if (Edge1->NextInSEL == Edge2) {
+    TEdge *Next = Edge2->NextInSEL;
+    if (Next)
+      Next->PrevInSEL = Edge1;
+    TEdge *Prev = Edge1->PrevInSEL;
+    if (Prev)
+      Prev->NextInSEL = Edge2;
+    Edge2->PrevInSEL = Prev;
+    Edge2->NextInSEL = Edge1;
+    Edge1->PrevInSEL = Edge2;
+    Edge1->NextInSEL = Next;
+  } else if (Edge2->NextInSEL == Edge1) {
+    TEdge *Next = Edge1->NextInSEL;
+    if (Next)
+      Next->PrevInSEL = Edge2;
+    TEdge *Prev = Edge2->PrevInSEL;
+    if (Prev)
+      Prev->NextInSEL = Edge1;
+    Edge1->PrevInSEL = Prev;
+    Edge1->NextInSEL = Edge2;
+    Edge2->PrevInSEL = Edge1;
+    Edge2->NextInSEL = Next;
+  } else {
+    TEdge *Next = Edge1->NextInSEL;
+    TEdge *Prev = Edge1->PrevInSEL;
+    Edge1->NextInSEL = Edge2->NextInSEL;
+    if (Edge1->NextInSEL)
+      Edge1->NextInSEL->PrevInSEL = Edge1;
+    Edge1->PrevInSEL = Edge2->PrevInSEL;
+    if (Edge1->PrevInSEL)
+      Edge1->PrevInSEL->NextInSEL = Edge1;
+    Edge2->NextInSEL = Next;
+    if (Edge2->NextInSEL)
+      Edge2->NextInSEL->PrevInSEL = Edge2;
+    Edge2->PrevInSEL = Prev;
+    if (Edge2->PrevInSEL)
+      Edge2->PrevInSEL->NextInSEL = Edge2;
+  }
+
+  if (!Edge1->PrevInSEL)
+    m_SortedEdges = Edge1;
+  else if (!Edge2->PrevInSEL)
+    m_SortedEdges = Edge2;
+}
+//------------------------------------------------------------------------------
+
+TEdge *GetNextInAEL(TEdge *e, Direction dir) {
+  return dir == dLeftToRight ? e->NextInAEL : e->PrevInAEL;
+}
+//------------------------------------------------------------------------------
+
+void GetHorzDirection(TEdge &HorzEdge, Direction &Dir, cInt &Left,
+                      cInt &Right) {
+  if (HorzEdge.Bot.X < HorzEdge.Top.X) {
+    Left = HorzEdge.Bot.X;
+    Right = HorzEdge.Top.X;
+    Dir = dLeftToRight;
+  } else {
+    Left = HorzEdge.Top.X;
+    Right = HorzEdge.Bot.X;
+    Dir = dRightToLeft;
+  }
+}
+//------------------------------------------------------------------------
+
+/*******************************************************************************
+ * Notes: Horizontal edges (HEs) at scanline intersections (ie at the Top or *
+ * Bottom of a scanbeam) are processed as if layered. The order in which HEs *
+ * are processed doesn't matter. HEs intersect with other HE Bot.Xs only [#] *
+ * (or they could intersect with Top.Xs only, ie EITHER Bot.Xs OR Top.Xs), * and
+ *with other non-horizontal edges [*]. Once these intersections are        *
+ * processed, intermediate HEs then 'promote' the Edge above (NextInLML) into *
+ * the AEL. These 'promoted' edges may in turn intersect [%] with other HEs. *
+ *******************************************************************************/
+
+void Clipper::ProcessHorizontal(TEdge *horzEdge) {
+  Direction dir;
+  cInt horzLeft, horzRight;
+  bool IsOpen = (horzEdge->WindDelta == 0);
+
+  GetHorzDirection(*horzEdge, dir, horzLeft, horzRight);
+
+  TEdge *eLastHorz = horzEdge, *eMaxPair = 0;
+  while (eLastHorz->NextInLML && IsHorizontal(*eLastHorz->NextInLML))
+    eLastHorz = eLastHorz->NextInLML;
+  if (!eLastHorz->NextInLML)
+    eMaxPair = GetMaximaPair(eLastHorz);
+
+  MaximaList::const_iterator maxIt;
+  MaximaList::const_reverse_iterator maxRit;
+  if (m_Maxima.size() > 0) {
+    // get the first maxima in range (X) ...
+    if (dir == dLeftToRight) {
+      maxIt = m_Maxima.begin();
+      while (maxIt != m_Maxima.end() && *maxIt <= horzEdge->Bot.X)
+        maxIt++;
+      if (maxIt != m_Maxima.end() && *maxIt >= eLastHorz->Top.X)
+        maxIt = m_Maxima.end();
+    } else {
+      maxRit = m_Maxima.rbegin();
+      while (maxRit != m_Maxima.rend() && *maxRit > horzEdge->Bot.X)
+        maxRit++;
+      if (maxRit != m_Maxima.rend() && *maxRit <= eLastHorz->Top.X)
+        maxRit = m_Maxima.rend();
+    }
+  }
+
+  OutPt *op1 = 0;
+
+  for (;;) // loop through consec. horizontal edges
+  {
+
+    bool IsLastHorz = (horzEdge == eLastHorz);
+    TEdge *e = GetNextInAEL(horzEdge, dir);
+    while (e) {
+
+      // this code block inserts extra coords into horizontal edges (in output
+      // polygons) wherever maxima touch these horizontal edges. This helps
+      //'simplifying' polygons (ie if the Simplify property is set).
+      if (m_Maxima.size() > 0) {
+        if (dir == dLeftToRight) {
+          while (maxIt != m_Maxima.end() && *maxIt < e->Curr.X) {
+            if (horzEdge->OutIdx >= 0 && !IsOpen)
+              AddOutPt(horzEdge, IntPoint(*maxIt, horzEdge->Bot.Y));
+            maxIt++;
+          }
+        } else {
+          while (maxRit != m_Maxima.rend() && *maxRit > e->Curr.X) {
+            if (horzEdge->OutIdx >= 0 && !IsOpen)
+              AddOutPt(horzEdge, IntPoint(*maxRit, horzEdge->Bot.Y));
+            maxRit++;
+          }
+        }
+      };
+
+      if ((dir == dLeftToRight && e->Curr.X > horzRight) ||
+          (dir == dRightToLeft && e->Curr.X < horzLeft))
+        break;
+
+      // Also break if we've got to the end of an intermediate horizontal edge
+      // ...
+      // nb: Smaller Dx's are to the right of larger Dx's ABOVE the horizontal.
+      if (e->Curr.X == horzEdge->Top.X && horzEdge->NextInLML &&
+          e->Dx < horzEdge->NextInLML->Dx)
+        break;
+
+      if (horzEdge->OutIdx >= 0 && !IsOpen) // note: may be done multiple times
+      {
+#ifdef use_xyz
+        if (dir == dLeftToRight)
+          SetZ(e->Curr, *horzEdge, *e);
+        else
+          SetZ(e->Curr, *e, *horzEdge);
+#endif
+        op1 = AddOutPt(horzEdge, e->Curr);
+        TEdge *eNextHorz = m_SortedEdges;
+        while (eNextHorz) {
+          if (eNextHorz->OutIdx >= 0 &&
+              HorzSegmentsOverlap(horzEdge->Bot.X, horzEdge->Top.X,
+                                  eNextHorz->Bot.X, eNextHorz->Top.X)) {
+            OutPt *op2 = GetLastOutPt(eNextHorz);
+            AddJoin(op2, op1, eNextHorz->Top);
+          }
+          eNextHorz = eNextHorz->NextInSEL;
+        }
+        AddGhostJoin(op1, horzEdge->Bot);
+      }
+
+      // OK, so far we're still in range of the horizontal Edge  but make sure
+      // we're at the last of consec. horizontals when matching with eMaxPair
+      if (e == eMaxPair && IsLastHorz) {
+        if (horzEdge->OutIdx >= 0)
+          AddLocalMaxPoly(horzEdge, eMaxPair, horzEdge->Top);
+        DeleteFromAEL(horzEdge);
+        DeleteFromAEL(eMaxPair);
+        return;
+      }
+
+      if (dir == dLeftToRight) {
+        IntPoint Pt = IntPoint(e->Curr.X, horzEdge->Curr.Y);
+        IntersectEdges(horzEdge, e, Pt);
+      } else {
+        IntPoint Pt = IntPoint(e->Curr.X, horzEdge->Curr.Y);
+        IntersectEdges(e, horzEdge, Pt);
+      }
+      TEdge *eNext = GetNextInAEL(e, dir);
+      SwapPositionsInAEL(horzEdge, e);
+      e = eNext;
+    } // end while(e)
+
+    // Break out of loop if HorzEdge.NextInLML is not also horizontal ...
+    if (!horzEdge->NextInLML || !IsHorizontal(*horzEdge->NextInLML))
+      break;
+
+    UpdateEdgeIntoAEL(horzEdge);
+    if (horzEdge->OutIdx >= 0)
+      AddOutPt(horzEdge, horzEdge->Bot);
+    GetHorzDirection(*horzEdge, dir, horzLeft, horzRight);
+
+  } // end for (;;)
+
+  if (horzEdge->OutIdx >= 0 && !op1) {
+    op1 = GetLastOutPt(horzEdge);
+    TEdge *eNextHorz = m_SortedEdges;
+    while (eNextHorz) {
+      if (eNextHorz->OutIdx >= 0 &&
+          HorzSegmentsOverlap(horzEdge->Bot.X, horzEdge->Top.X,
+                              eNextHorz->Bot.X, eNextHorz->Top.X)) {
+        OutPt *op2 = GetLastOutPt(eNextHorz);
+        AddJoin(op2, op1, eNextHorz->Top);
+      }
+      eNextHorz = eNextHorz->NextInSEL;
+    }
+    AddGhostJoin(op1, horzEdge->Top);
+  }
+
+  if (horzEdge->NextInLML) {
+    if (horzEdge->OutIdx >= 0) {
+      op1 = AddOutPt(horzEdge, horzEdge->Top);
+      UpdateEdgeIntoAEL(horzEdge);
+      if (horzEdge->WindDelta == 0)
+        return;
+      // nb: HorzEdge is no longer horizontal here
+      TEdge *ePrev = horzEdge->PrevInAEL;
+      TEdge *eNext = horzEdge->NextInAEL;
+      if (ePrev && ePrev->Curr.X == horzEdge->Bot.X &&
+          ePrev->Curr.Y == horzEdge->Bot.Y && ePrev->WindDelta != 0 &&
+          (ePrev->OutIdx >= 0 && ePrev->Curr.Y > ePrev->Top.Y &&
+           SlopesEqual(*horzEdge, *ePrev, m_UseFullRange))) {
+        OutPt *op2 = AddOutPt(ePrev, horzEdge->Bot);
+        AddJoin(op1, op2, horzEdge->Top);
+      } else if (eNext && eNext->Curr.X == horzEdge->Bot.X &&
+                 eNext->Curr.Y == horzEdge->Bot.Y && eNext->WindDelta != 0 &&
+                 eNext->OutIdx >= 0 && eNext->Curr.Y > eNext->Top.Y &&
+                 SlopesEqual(*horzEdge, *eNext, m_UseFullRange)) {
+        OutPt *op2 = AddOutPt(eNext, horzEdge->Bot);
+        AddJoin(op1, op2, horzEdge->Top);
+      }
+    } else
+      UpdateEdgeIntoAEL(horzEdge);
+  } else {
+    if (horzEdge->OutIdx >= 0)
+      AddOutPt(horzEdge, horzEdge->Top);
+    DeleteFromAEL(horzEdge);
+  }
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::ProcessIntersections(const cInt topY) {
+  if (!m_ActiveEdges)
+    return true;
+  try {
+    BuildIntersectList(topY);
+    size_t IlSize = m_IntersectList.size();
+    if (IlSize == 0)
+      return true;
+    if (IlSize == 1 || FixupIntersectionOrder())
+      ProcessIntersectList();
+    else
+      return false;
+  } catch (...) {
+    m_SortedEdges = 0;
+    DisposeIntersectNodes();
+    throw clipperException("ProcessIntersections error");
+  }
+  m_SortedEdges = 0;
+  return true;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::DisposeIntersectNodes() {
+  for (size_t i = 0; i < m_IntersectList.size(); ++i)
+    delete m_IntersectList[i];
+  m_IntersectList.clear();
+}
+//------------------------------------------------------------------------------
+
+void Clipper::BuildIntersectList(const cInt topY) {
+  if (!m_ActiveEdges)
+    return;
+
+  // prepare for sorting ...
+  TEdge *e = m_ActiveEdges;
+  m_SortedEdges = e;
+  while (e) {
+    e->PrevInSEL = e->PrevInAEL;
+    e->NextInSEL = e->NextInAEL;
+    e->Curr.X = TopX(*e, topY);
+    e = e->NextInAEL;
+  }
+
+  // bubblesort ...
+  bool isModified;
+  do {
+    isModified = false;
+    e = m_SortedEdges;
+    while (e->NextInSEL) {
+      TEdge *eNext = e->NextInSEL;
+      IntPoint Pt;
+      if (e->Curr.X > eNext->Curr.X) {
+        IntersectPoint(*e, *eNext, Pt);
+        if (Pt.Y < topY)
+          Pt = IntPoint(TopX(*e, topY), topY);
+        IntersectNode *newNode = new IntersectNode;
+        newNode->Edge1 = e;
+        newNode->Edge2 = eNext;
+        newNode->Pt = Pt;
+        m_IntersectList.push_back(newNode);
+
+        SwapPositionsInSEL(e, eNext);
+        isModified = true;
+      } else
+        e = eNext;
+    }
+    if (e->PrevInSEL)
+      e->PrevInSEL->NextInSEL = 0;
+    else
+      break;
+  } while (isModified);
+  m_SortedEdges = 0; // important
+}
+//------------------------------------------------------------------------------
+
+void Clipper::ProcessIntersectList() {
+  for (size_t i = 0; i < m_IntersectList.size(); ++i) {
+    IntersectNode *iNode = m_IntersectList[i];
+    {
+      IntersectEdges(iNode->Edge1, iNode->Edge2, iNode->Pt);
+      SwapPositionsInAEL(iNode->Edge1, iNode->Edge2);
+    }
+    delete iNode;
+  }
+  m_IntersectList.clear();
+}
+//------------------------------------------------------------------------------
+
+bool IntersectListSort(IntersectNode *node1, IntersectNode *node2) {
+  return node2->Pt.Y < node1->Pt.Y;
+}
+//------------------------------------------------------------------------------
+
+inline bool EdgesAdjacent(const IntersectNode &inode) {
+  return (inode.Edge1->NextInSEL == inode.Edge2) ||
+         (inode.Edge1->PrevInSEL == inode.Edge2);
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::FixupIntersectionOrder() {
+  // pre-condition: intersections are sorted Bottom-most first.
+  // Now it's crucial that intersections are made only between adjacent edges,
+  // so to ensure this the order of intersections may need adjusting ...
+  CopyAELToSEL();
+  std::sort(m_IntersectList.begin(), m_IntersectList.end(), IntersectListSort);
+  size_t cnt = m_IntersectList.size();
+  for (size_t i = 0; i < cnt; ++i) {
+    if (!EdgesAdjacent(*m_IntersectList[i])) {
+      size_t j = i + 1;
+      while (j < cnt && !EdgesAdjacent(*m_IntersectList[j]))
+        j++;
+      if (j == cnt)
+        return false;
+      std::swap(m_IntersectList[i], m_IntersectList[j]);
+    }
+    SwapPositionsInSEL(m_IntersectList[i]->Edge1, m_IntersectList[i]->Edge2);
+  }
+  return true;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::DoMaxima(TEdge *e) {
+  TEdge *eMaxPair = GetMaximaPairEx(e);
+  if (!eMaxPair) {
+    if (e->OutIdx >= 0)
+      AddOutPt(e, e->Top);
+    DeleteFromAEL(e);
+    return;
+  }
+
+  TEdge *eNext = e->NextInAEL;
+  while (eNext && eNext != eMaxPair) {
+    IntersectEdges(e, eNext, e->Top);
+    SwapPositionsInAEL(e, eNext);
+    eNext = e->NextInAEL;
+  }
+
+  if (e->OutIdx == Unassigned && eMaxPair->OutIdx == Unassigned) {
+    DeleteFromAEL(e);
+    DeleteFromAEL(eMaxPair);
+  } else if (e->OutIdx >= 0 && eMaxPair->OutIdx >= 0) {
+    if (e->OutIdx >= 0)
+      AddLocalMaxPoly(e, eMaxPair, e->Top);
+    DeleteFromAEL(e);
+    DeleteFromAEL(eMaxPair);
+  }
+#ifdef use_lines
+  else if (e->WindDelta == 0) {
+    if (e->OutIdx >= 0) {
+      AddOutPt(e, e->Top);
+      e->OutIdx = Unassigned;
+    }
+    DeleteFromAEL(e);
+
+    if (eMaxPair->OutIdx >= 0) {
+      AddOutPt(eMaxPair, e->Top);
+      eMaxPair->OutIdx = Unassigned;
+    }
+    DeleteFromAEL(eMaxPair);
+  }
+#endif
+  else
+    throw clipperException("DoMaxima error");
+}
+//------------------------------------------------------------------------------
+
+void Clipper::ProcessEdgesAtTopOfScanbeam(const cInt topY) {
+  TEdge *e = m_ActiveEdges;
+  while (e) {
+    // 1. process maxima, treating them as if they're 'bent' horizontal edges,
+    //   but exclude maxima with horizontal edges. nb: e can't be a horizontal.
+    bool IsMaximaEdge = IsMaxima(e, topY);
+
+    if (IsMaximaEdge) {
+      TEdge *eMaxPair = GetMaximaPairEx(e);
+      IsMaximaEdge = (!eMaxPair || !IsHorizontal(*eMaxPair));
+    }
+
+    if (IsMaximaEdge) {
+      if (m_StrictSimple)
+        m_Maxima.push_back(e->Top.X);
+      TEdge *ePrev = e->PrevInAEL;
+      DoMaxima(e);
+      if (!ePrev)
+        e = m_ActiveEdges;
+      else
+        e = ePrev->NextInAEL;
+    } else {
+      // 2. promote horizontal edges, otherwise update Curr.X and Curr.Y ...
+      if (IsIntermediate(e, topY) && IsHorizontal(*e->NextInLML)) {
+        UpdateEdgeIntoAEL(e);
+        if (e->OutIdx >= 0)
+          AddOutPt(e, e->Bot);
+        AddEdgeToSEL(e);
+      } else {
+        e->Curr.X = TopX(*e, topY);
+        e->Curr.Y = topY;
+#ifdef use_xyz
+        e->Curr.Z =
+            topY == e->Top.Y ? e->Top.Z : (topY == e->Bot.Y ? e->Bot.Z : 0);
+#endif
+      }
+
+      // When StrictlySimple and 'e' is being touched by another edge, then
+      // make sure both edges have a vertex here ...
+      if (m_StrictSimple) {
+        TEdge *ePrev = e->PrevInAEL;
+        if ((e->OutIdx >= 0) && (e->WindDelta != 0) && ePrev &&
+            (ePrev->OutIdx >= 0) && (ePrev->Curr.X == e->Curr.X) &&
+            (ePrev->WindDelta != 0)) {
+          IntPoint pt = e->Curr;
+#ifdef use_xyz
+          SetZ(pt, *ePrev, *e);
+#endif
+          OutPt *op = AddOutPt(ePrev, pt);
+          OutPt *op2 = AddOutPt(e, pt);
+          AddJoin(op, op2, pt); // StrictlySimple (type-3) join
+        }
+      }
+
+      e = e->NextInAEL;
+    }
+  }
+
+  // 3. Process horizontals at the Top of the scanbeam ...
+  m_Maxima.sort();
+  ProcessHorizontals();
+  m_Maxima.clear();
+
+  // 4. Promote intermediate vertices ...
+  e = m_ActiveEdges;
+  while (e) {
+    if (IsIntermediate(e, topY)) {
+      OutPt *op = 0;
+      if (e->OutIdx >= 0)
+        op = AddOutPt(e, e->Top);
+      UpdateEdgeIntoAEL(e);
+
+      // if output polygons share an edge, they'll need joining later ...
+      TEdge *ePrev = e->PrevInAEL;
+      TEdge *eNext = e->NextInAEL;
+      if (ePrev && ePrev->Curr.X == e->Bot.X && ePrev->Curr.Y == e->Bot.Y &&
+          op && ePrev->OutIdx >= 0 && ePrev->Curr.Y > ePrev->Top.Y &&
+          SlopesEqual(e->Curr, e->Top, ePrev->Curr, ePrev->Top,
+                      m_UseFullRange) &&
+          (e->WindDelta != 0) && (ePrev->WindDelta != 0)) {
+        OutPt *op2 = AddOutPt(ePrev, e->Bot);
+        AddJoin(op, op2, e->Top);
+      } else if (eNext && eNext->Curr.X == e->Bot.X &&
+                 eNext->Curr.Y == e->Bot.Y && op && eNext->OutIdx >= 0 &&
+                 eNext->Curr.Y > eNext->Top.Y &&
+                 SlopesEqual(e->Curr, e->Top, eNext->Curr, eNext->Top,
+                             m_UseFullRange) &&
+                 (e->WindDelta != 0) && (eNext->WindDelta != 0)) {
+        OutPt *op2 = AddOutPt(eNext, e->Bot);
+        AddJoin(op, op2, e->Top);
+      }
+    }
+    e = e->NextInAEL;
+  }
+}
+//------------------------------------------------------------------------------
+
+void Clipper::FixupOutPolyline(OutRec &outrec) {
+  OutPt *pp = outrec.Pts;
+  OutPt *lastPP = pp->Prev;
+  while (pp != lastPP) {
+    pp = pp->Next;
+    if (pp->Pt == pp->Prev->Pt) {
+      if (pp == lastPP)
+        lastPP = pp->Prev;
+      OutPt *tmpPP = pp->Prev;
+      tmpPP->Next = pp->Next;
+      pp->Next->Prev = tmpPP;
+      delete pp;
+      pp = tmpPP;
+    }
+  }
+
+  if (pp == pp->Prev) {
+    DisposeOutPts(pp);
+    outrec.Pts = 0;
+    return;
+  }
+}
+//------------------------------------------------------------------------------
+
+void Clipper::FixupOutPolygon(OutRec &outrec) {
+  // FixupOutPolygon() - removes duplicate points and simplifies consecutive
+  // parallel edges by removing the middle vertex.
+  OutPt *lastOK = 0;
+  outrec.BottomPt = 0;
+  OutPt *pp = outrec.Pts;
+  bool preserveCol = m_PreserveCollinear || m_StrictSimple;
+
+  for (;;) {
+    if (pp->Prev == pp || pp->Prev == pp->Next) {
+      DisposeOutPts(pp);
+      outrec.Pts = 0;
+      return;
+    }
+
+    // test for duplicate points and collinear edges ...
+    if ((pp->Pt == pp->Next->Pt) || (pp->Pt == pp->Prev->Pt) ||
+        (SlopesEqual(pp->Prev->Pt, pp->Pt, pp->Next->Pt, m_UseFullRange) &&
+         (!preserveCol ||
+          !Pt2IsBetweenPt1AndPt3(pp->Prev->Pt, pp->Pt, pp->Next->Pt)))) {
+      lastOK = 0;
+      OutPt *tmp = pp;
+      pp->Prev->Next = pp->Next;
+      pp->Next->Prev = pp->Prev;
+      pp = pp->Prev;
+      delete tmp;
+    } else if (pp == lastOK)
+      break;
+    else {
+      if (!lastOK)
+        lastOK = pp;
+      pp = pp->Next;
+    }
+  }
+  outrec.Pts = pp;
+}
+//------------------------------------------------------------------------------
+
+int PointCount(OutPt *Pts) {
+  if (!Pts)
+    return 0;
+  int result = 0;
+  OutPt *p = Pts;
+  do {
+    result++;
+    p = p->Next;
+  } while (p != Pts);
+  return result;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::BuildResult(Paths &polys) {
+  polys.reserve(m_PolyOuts.size());
+  for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); ++i) {
+    if (!m_PolyOuts[i]->Pts)
+      continue;
+    Path pg;
+    OutPt *p = m_PolyOuts[i]->Pts->Prev;
+    int cnt = PointCount(p);
+    if (cnt < 2)
+      continue;
+    pg.reserve(cnt);
+    for (int i = 0; i < cnt; ++i) {
+      pg.push_back(p->Pt);
+      p = p->Prev;
+    }
+    polys.push_back(pg);
+  }
+}
+//------------------------------------------------------------------------------
+
+void Clipper::BuildResult2(PolyTree &polytree) {
+  polytree.Clear();
+  polytree.AllNodes.reserve(m_PolyOuts.size());
+  // add each output polygon/contour to polytree ...
+  for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); i++) {
+    OutRec *outRec = m_PolyOuts[i];
+    int cnt = PointCount(outRec->Pts);
+    if ((outRec->IsOpen && cnt < 2) || (!outRec->IsOpen && cnt < 3))
+      continue;
+    FixHoleLinkage(*outRec);
+    PolyNode *pn = new PolyNode();
+    // nb: polytree takes ownership of all the PolyNodes
+    polytree.AllNodes.push_back(pn);
+    outRec->PolyNd = pn;
+    pn->Parent = 0;
+    pn->Index = 0;
+    pn->Contour.reserve(cnt);
+    OutPt *op = outRec->Pts->Prev;
+    for (int j = 0; j < cnt; j++) {
+      pn->Contour.push_back(op->Pt);
+      op = op->Prev;
+    }
+  }
+
+  // fixup PolyNode links etc ...
+  polytree.Children.reserve(m_PolyOuts.size());
+  for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); i++) {
+    OutRec *outRec = m_PolyOuts[i];
+    if (!outRec->PolyNd)
+      continue;
+    if (outRec->IsOpen) {
+      outRec->PolyNd->m_IsOpen = true;
+      polytree.AddChild(*outRec->PolyNd);
+    } else if (outRec->FirstLeft && outRec->FirstLeft->PolyNd)
+      outRec->FirstLeft->PolyNd->AddChild(*outRec->PolyNd);
+    else
+      polytree.AddChild(*outRec->PolyNd);
+  }
+}
+//------------------------------------------------------------------------------
+
+void SwapIntersectNodes(IntersectNode &int1, IntersectNode &int2) {
+  // just swap the contents (because fIntersectNodes is a single-linked-list)
+  IntersectNode inode = int1; // gets a copy of Int1
+  int1.Edge1 = int2.Edge1;
+  int1.Edge2 = int2.Edge2;
+  int1.Pt = int2.Pt;
+  int2.Edge1 = inode.Edge1;
+  int2.Edge2 = inode.Edge2;
+  int2.Pt = inode.Pt;
+}
+//------------------------------------------------------------------------------
+
+inline bool E2InsertsBeforeE1(TEdge &e1, TEdge &e2) {
+  if (e2.Curr.X == e1.Curr.X) {
+    if (e2.Top.Y > e1.Top.Y)
+      return e2.Top.X < TopX(e1, e2.Top.Y);
+    else
+      return e1.Top.X > TopX(e2, e1.Top.Y);
+  } else
+    return e2.Curr.X < e1.Curr.X;
+}
+//------------------------------------------------------------------------------
+
+bool GetOverlap(const cInt a1, const cInt a2, const cInt b1, const cInt b2,
+                cInt &Left, cInt &Right) {
+  if (a1 < a2) {
+    if (b1 < b2) {
+      Left = std::max(a1, b1);
+      Right = std::min(a2, b2);
+    } else {
+      Left = std::max(a1, b2);
+      Right = std::min(a2, b1);
+    }
+  } else {
+    if (b1 < b2) {
+      Left = std::max(a2, b1);
+      Right = std::min(a1, b2);
+    } else {
+      Left = std::max(a2, b2);
+      Right = std::min(a1, b1);
+    }
+  }
+  return Left < Right;
+}
+//------------------------------------------------------------------------------
+
+inline void UpdateOutPtIdxs(OutRec &outrec) {
+  OutPt *op = outrec.Pts;
+  do {
+    op->Idx = outrec.Idx;
+    op = op->Prev;
+  } while (op != outrec.Pts);
+}
+//------------------------------------------------------------------------------
+
+void Clipper::InsertEdgeIntoAEL(TEdge *edge, TEdge *startEdge) {
+  if (!m_ActiveEdges) {
+    edge->PrevInAEL = 0;
+    edge->NextInAEL = 0;
+    m_ActiveEdges = edge;
+  } else if (!startEdge && E2InsertsBeforeE1(*m_ActiveEdges, *edge)) {
+    edge->PrevInAEL = 0;
+    edge->NextInAEL = m_ActiveEdges;
+    m_ActiveEdges->PrevInAEL = edge;
+    m_ActiveEdges = edge;
+  } else {
+    if (!startEdge)
+      startEdge = m_ActiveEdges;
+    while (startEdge->NextInAEL &&
+           !E2InsertsBeforeE1(*startEdge->NextInAEL, *edge))
+      startEdge = startEdge->NextInAEL;
+    edge->NextInAEL = startEdge->NextInAEL;
+    if (startEdge->NextInAEL)
+      startEdge->NextInAEL->PrevInAEL = edge;
+    edge->PrevInAEL = startEdge;
+    startEdge->NextInAEL = edge;
+  }
+}
+//----------------------------------------------------------------------
+
+OutPt *DupOutPt(OutPt *outPt, bool InsertAfter) {
+  OutPt *result = new OutPt;
+  result->Pt = outPt->Pt;
+  result->Idx = outPt->Idx;
+  if (InsertAfter) {
+    result->Next = outPt->Next;
+    result->Prev = outPt;
+    outPt->Next->Prev = result;
+    outPt->Next = result;
+  } else {
+    result->Prev = outPt->Prev;
+    result->Next = outPt;
+    outPt->Prev->Next = result;
+    outPt->Prev = result;
+  }
+  return result;
+}
+//------------------------------------------------------------------------------
+
+bool JoinHorz(OutPt *op1, OutPt *op1b, OutPt *op2, OutPt *op2b,
+              const IntPoint Pt, bool DiscardLeft) {
+  Direction Dir1 = (op1->Pt.X > op1b->Pt.X ? dRightToLeft : dLeftToRight);
+  Direction Dir2 = (op2->Pt.X > op2b->Pt.X ? dRightToLeft : dLeftToRight);
+  if (Dir1 == Dir2)
+    return false;
+
+  // When DiscardLeft, we want Op1b to be on the Left of Op1, otherwise we
+  // want Op1b to be on the Right. (And likewise with Op2 and Op2b.)
+  // So, to facilitate this while inserting Op1b and Op2b ...
+  // when DiscardLeft, make sure we're AT or RIGHT of Pt before adding Op1b,
+  // otherwise make sure we're AT or LEFT of Pt. (Likewise with Op2b.)
+  if (Dir1 == dLeftToRight) {
+    while (op1->Next->Pt.X <= Pt.X && op1->Next->Pt.X >= op1->Pt.X &&
+           op1->Next->Pt.Y == Pt.Y)
+      op1 = op1->Next;
+    if (DiscardLeft && (op1->Pt.X != Pt.X))
+      op1 = op1->Next;
+    op1b = DupOutPt(op1, !DiscardLeft);
+    if (op1b->Pt != Pt) {
+      op1 = op1b;
+      op1->Pt = Pt;
+      op1b = DupOutPt(op1, !DiscardLeft);
+    }
+  } else {
+    while (op1->Next->Pt.X >= Pt.X && op1->Next->Pt.X <= op1->Pt.X &&
+           op1->Next->Pt.Y == Pt.Y)
+      op1 = op1->Next;
+    if (!DiscardLeft && (op1->Pt.X != Pt.X))
+      op1 = op1->Next;
+    op1b = DupOutPt(op1, DiscardLeft);
+    if (op1b->Pt != Pt) {
+      op1 = op1b;
+      op1->Pt = Pt;
+      op1b = DupOutPt(op1, DiscardLeft);
+    }
+  }
+
+  if (Dir2 == dLeftToRight) {
+    while (op2->Next->Pt.X <= Pt.X && op2->Next->Pt.X >= op2->Pt.X &&
+           op2->Next->Pt.Y == Pt.Y)
+      op2 = op2->Next;
+    if (DiscardLeft && (op2->Pt.X != Pt.X))
+      op2 = op2->Next;
+    op2b = DupOutPt(op2, !DiscardLeft);
+    if (op2b->Pt != Pt) {
+      op2 = op2b;
+      op2->Pt = Pt;
+      op2b = DupOutPt(op2, !DiscardLeft);
+    };
+  } else {
+    while (op2->Next->Pt.X >= Pt.X && op2->Next->Pt.X <= op2->Pt.X &&
+           op2->Next->Pt.Y == Pt.Y)
+      op2 = op2->Next;
+    if (!DiscardLeft && (op2->Pt.X != Pt.X))
+      op2 = op2->Next;
+    op2b = DupOutPt(op2, DiscardLeft);
+    if (op2b->Pt != Pt) {
+      op2 = op2b;
+      op2->Pt = Pt;
+      op2b = DupOutPt(op2, DiscardLeft);
+    };
+  };
+
+  if ((Dir1 == dLeftToRight) == DiscardLeft) {
+    op1->Prev = op2;
+    op2->Next = op1;
+    op1b->Next = op2b;
+    op2b->Prev = op1b;
+  } else {
+    op1->Next = op2;
+    op2->Prev = op1;
+    op1b->Prev = op2b;
+    op2b->Next = op1b;
+  }
+  return true;
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::JoinPoints(Join *j, OutRec *outRec1, OutRec *outRec2) {
+  OutPt *op1 = j->OutPt1, *op1b;
+  OutPt *op2 = j->OutPt2, *op2b;
+
+  // There are 3 kinds of joins for output polygons ...
+  // 1. Horizontal joins where Join.OutPt1 & Join.OutPt2 are vertices anywhere
+  // along (horizontal) collinear edges (& Join.OffPt is on the same
+  // horizontal).
+  // 2. Non-horizontal joins where Join.OutPt1 & Join.OutPt2 are at the same
+  // location at the Bottom of the overlapping segment (& Join.OffPt is above).
+  // 3. StrictSimple joins where edges touch but are not collinear and where
+  // Join.OutPt1, Join.OutPt2 & Join.OffPt all share the same point.
+  bool isHorizontal = (j->OutPt1->Pt.Y == j->OffPt.Y);
+
+  if (isHorizontal && (j->OffPt == j->OutPt1->Pt) &&
+      (j->OffPt == j->OutPt2->Pt)) {
+    // Strictly Simple join ...
+    if (outRec1 != outRec2)
+      return false;
+    op1b = j->OutPt1->Next;
+    while (op1b != op1 && (op1b->Pt == j->OffPt))
+      op1b = op1b->Next;
+    bool reverse1 = (op1b->Pt.Y > j->OffPt.Y);
+    op2b = j->OutPt2->Next;
+    while (op2b != op2 && (op2b->Pt == j->OffPt))
+      op2b = op2b->Next;
+    bool reverse2 = (op2b->Pt.Y > j->OffPt.Y);
+    if (reverse1 == reverse2)
+      return false;
+    if (reverse1) {
+      op1b = DupOutPt(op1, false);
+      op2b = DupOutPt(op2, true);
+      op1->Prev = op2;
+      op2->Next = op1;
+      op1b->Next = op2b;
+      op2b->Prev = op1b;
+      j->OutPt1 = op1;
+      j->OutPt2 = op1b;
+      return true;
+    } else {
+      op1b = DupOutPt(op1, true);
+      op2b = DupOutPt(op2, false);
+      op1->Next = op2;
+      op2->Prev = op1;
+      op1b->Prev = op2b;
+      op2b->Next = op1b;
+      j->OutPt1 = op1;
+      j->OutPt2 = op1b;
+      return true;
+    }
+  } else if (isHorizontal) {
+    // treat horizontal joins differently to non-horizontal joins since with
+    // them we're not yet sure where the overlapping is. OutPt1.Pt & OutPt2.Pt
+    // may be anywhere along the horizontal edge.
+    op1b = op1;
+    while (op1->Prev->Pt.Y == op1->Pt.Y && op1->Prev != op1b &&
+           op1->Prev != op2)
+      op1 = op1->Prev;
+    while (op1b->Next->Pt.Y == op1b->Pt.Y && op1b->Next != op1 &&
+           op1b->Next != op2)
+      op1b = op1b->Next;
+    if (op1b->Next == op1 || op1b->Next == op2)
+      return false; // a flat 'polygon'
+
+    op2b = op2;
+    while (op2->Prev->Pt.Y == op2->Pt.Y && op2->Prev != op2b &&
+           op2->Prev != op1b)
+      op2 = op2->Prev;
+    while (op2b->Next->Pt.Y == op2b->Pt.Y && op2b->Next != op2 &&
+           op2b->Next != op1)
+      op2b = op2b->Next;
+    if (op2b->Next == op2 || op2b->Next == op1)
+      return false; // a flat 'polygon'
+
+    cInt Left, Right;
+    // Op1 --> Op1b & Op2 --> Op2b are the extremites of the horizontal edges
+    if (!GetOverlap(op1->Pt.X, op1b->Pt.X, op2->Pt.X, op2b->Pt.X, Left, Right))
+      return false;
+
+    // DiscardLeftSide: when overlapping edges are joined, a spike will created
+    // which needs to be cleaned up. However, we don't want Op1 or Op2 caught up
+    // on the discard Side as either may still be needed for other joins ...
+    IntPoint Pt;
+    bool DiscardLeftSide;
+    if (op1->Pt.X >= Left && op1->Pt.X <= Right) {
+      Pt = op1->Pt;
+      DiscardLeftSide = (op1->Pt.X > op1b->Pt.X);
+    } else if (op2->Pt.X >= Left && op2->Pt.X <= Right) {
+      Pt = op2->Pt;
+      DiscardLeftSide = (op2->Pt.X > op2b->Pt.X);
+    } else if (op1b->Pt.X >= Left && op1b->Pt.X <= Right) {
+      Pt = op1b->Pt;
+      DiscardLeftSide = op1b->Pt.X > op1->Pt.X;
+    } else {
+      Pt = op2b->Pt;
+      DiscardLeftSide = (op2b->Pt.X > op2->Pt.X);
+    }
+    j->OutPt1 = op1;
+    j->OutPt2 = op2;
+    return JoinHorz(op1, op1b, op2, op2b, Pt, DiscardLeftSide);
+  } else {
+    // nb: For non-horizontal joins ...
+    //    1. Jr.OutPt1.Pt.Y == Jr.OutPt2.Pt.Y
+    //    2. Jr.OutPt1.Pt > Jr.OffPt.Y
+
+    // make sure the polygons are correctly oriented ...
+    op1b = op1->Next;
+    while ((op1b->Pt == op1->Pt) && (op1b != op1))
+      op1b = op1b->Next;
+    bool Reverse1 = ((op1b->Pt.Y > op1->Pt.Y) ||
+                     !SlopesEqual(op1->Pt, op1b->Pt, j->OffPt, m_UseFullRange));
+    if (Reverse1) {
+      op1b = op1->Prev;
+      while ((op1b->Pt == op1->Pt) && (op1b != op1))
+        op1b = op1b->Prev;
+      if ((op1b->Pt.Y > op1->Pt.Y) ||
+          !SlopesEqual(op1->Pt, op1b->Pt, j->OffPt, m_UseFullRange))
+        return false;
+    };
+    op2b = op2->Next;
+    while ((op2b->Pt == op2->Pt) && (op2b != op2))
+      op2b = op2b->Next;
+    bool Reverse2 = ((op2b->Pt.Y > op2->Pt.Y) ||
+                     !SlopesEqual(op2->Pt, op2b->Pt, j->OffPt, m_UseFullRange));
+    if (Reverse2) {
+      op2b = op2->Prev;
+      while ((op2b->Pt == op2->Pt) && (op2b != op2))
+        op2b = op2b->Prev;
+      if ((op2b->Pt.Y > op2->Pt.Y) ||
+          !SlopesEqual(op2->Pt, op2b->Pt, j->OffPt, m_UseFullRange))
+        return false;
+    }
+
+    if ((op1b == op1) || (op2b == op2) || (op1b == op2b) ||
+        ((outRec1 == outRec2) && (Reverse1 == Reverse2)))
+      return false;
+
+    if (Reverse1) {
+      op1b = DupOutPt(op1, false);
+      op2b = DupOutPt(op2, true);
+      op1->Prev = op2;
+      op2->Next = op1;
+      op1b->Next = op2b;
+      op2b->Prev = op1b;
+      j->OutPt1 = op1;
+      j->OutPt2 = op1b;
+      return true;
+    } else {
+      op1b = DupOutPt(op1, true);
+      op2b = DupOutPt(op2, false);
+      op1->Next = op2;
+      op2->Prev = op1;
+      op1b->Prev = op2b;
+      op2b->Next = op1b;
+      j->OutPt1 = op1;
+      j->OutPt2 = op1b;
+      return true;
+    }
+  }
+}
+//----------------------------------------------------------------------
+
+static OutRec *ParseFirstLeft(OutRec *FirstLeft) {
+  while (FirstLeft && !FirstLeft->Pts)
+    FirstLeft = FirstLeft->FirstLeft;
+  return FirstLeft;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::FixupFirstLefts1(OutRec *OldOutRec, OutRec *NewOutRec) {
+  // tests if NewOutRec contains the polygon before reassigning FirstLeft
+  for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); ++i) {
+    OutRec *outRec = m_PolyOuts[i];
+    OutRec *firstLeft = ParseFirstLeft(outRec->FirstLeft);
+    if (outRec->Pts && firstLeft == OldOutRec) {
+      if (Poly2ContainsPoly1(outRec->Pts, NewOutRec->Pts))
+        outRec->FirstLeft = NewOutRec;
+    }
+  }
+}
+//----------------------------------------------------------------------
+
+void Clipper::FixupFirstLefts2(OutRec *InnerOutRec, OutRec *OuterOutRec) {
+  // A polygon has split into two such that one is now the inner of the other.
+  // It's possible that these polygons now wrap around other polygons, so check
+  // every polygon that's also contained by OuterOutRec's FirstLeft container
+  //(including 0) to see if they've become inner to the new inner polygon ...
+  OutRec *orfl = OuterOutRec->FirstLeft;
+  for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); ++i) {
+    OutRec *outRec = m_PolyOuts[i];
+
+    if (!outRec->Pts || outRec == OuterOutRec || outRec == InnerOutRec)
+      continue;
+    OutRec *firstLeft = ParseFirstLeft(outRec->FirstLeft);
+    if (firstLeft != orfl && firstLeft != InnerOutRec &&
+        firstLeft != OuterOutRec)
+      continue;
+    if (Poly2ContainsPoly1(outRec->Pts, InnerOutRec->Pts))
+      outRec->FirstLeft = InnerOutRec;
+    else if (Poly2ContainsPoly1(outRec->Pts, OuterOutRec->Pts))
+      outRec->FirstLeft = OuterOutRec;
+    else if (outRec->FirstLeft == InnerOutRec ||
+             outRec->FirstLeft == OuterOutRec)
+      outRec->FirstLeft = orfl;
+  }
+}
+//----------------------------------------------------------------------
+void Clipper::FixupFirstLefts3(OutRec *OldOutRec, OutRec *NewOutRec) {
+  // reassigns FirstLeft WITHOUT testing if NewOutRec contains the polygon
+  for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); ++i) {
+    OutRec *outRec = m_PolyOuts[i];
+    OutRec *firstLeft = ParseFirstLeft(outRec->FirstLeft);
+    if (outRec->Pts && firstLeft == OldOutRec)
+      outRec->FirstLeft = NewOutRec;
+  }
+}
+//----------------------------------------------------------------------
+
+void Clipper::JoinCommonEdges() {
+  for (JoinList::size_type i = 0; i < m_Joins.size(); i++) {
+    Join *join = m_Joins[i];
+
+    OutRec *outRec1 = GetOutRec(join->OutPt1->Idx);
+    OutRec *outRec2 = GetOutRec(join->OutPt2->Idx);
+
+    if (!outRec1->Pts || !outRec2->Pts)
+      continue;
+    if (outRec1->IsOpen || outRec2->IsOpen)
+      continue;
+
+    // get the polygon fragment with the correct hole state (FirstLeft)
+    // before calling JoinPoints() ...
+    OutRec *holeStateRec;
+    if (outRec1 == outRec2)
+      holeStateRec = outRec1;
+    else if (OutRec1RightOfOutRec2(outRec1, outRec2))
+      holeStateRec = outRec2;
+    else if (OutRec1RightOfOutRec2(outRec2, outRec1))
+      holeStateRec = outRec1;
+    else
+      holeStateRec = GetLowermostRec(outRec1, outRec2);
+
+    if (!JoinPoints(join, outRec1, outRec2))
+      continue;
+
+    if (outRec1 == outRec2) {
+      // instead of joining two polygons, we've just created a new one by
+      // splitting one polygon into two.
+      outRec1->Pts = join->OutPt1;
+      outRec1->BottomPt = 0;
+      outRec2 = CreateOutRec();
+      outRec2->Pts = join->OutPt2;
+
+      // update all OutRec2.Pts Idx's ...
+      UpdateOutPtIdxs(*outRec2);
+
+      if (Poly2ContainsPoly1(outRec2->Pts, outRec1->Pts)) {
+        // outRec1 contains outRec2 ...
+        outRec2->IsHole = !outRec1->IsHole;
+        outRec2->FirstLeft = outRec1;
+
+        if (m_UsingPolyTree)
+          FixupFirstLefts2(outRec2, outRec1);
+
+        if ((outRec2->IsHole ^ m_ReverseOutput) == (Area(*outRec2) > 0))
+          ReversePolyPtLinks(outRec2->Pts);
+
+      } else if (Poly2ContainsPoly1(outRec1->Pts, outRec2->Pts)) {
+        // outRec2 contains outRec1 ...
+        outRec2->IsHole = outRec1->IsHole;
+        outRec1->IsHole = !outRec2->IsHole;
+        outRec2->FirstLeft = outRec1->FirstLeft;
+        outRec1->FirstLeft = outRec2;
+
+        if (m_UsingPolyTree)
+          FixupFirstLefts2(outRec1, outRec2);
+
+        if ((outRec1->IsHole ^ m_ReverseOutput) == (Area(*outRec1) > 0))
+          ReversePolyPtLinks(outRec1->Pts);
+      } else {
+        // the 2 polygons are completely separate ...
+        outRec2->IsHole = outRec1->IsHole;
+        outRec2->FirstLeft = outRec1->FirstLeft;
+
+        // fixup FirstLeft pointers that may need reassigning to OutRec2
+        if (m_UsingPolyTree)
+          FixupFirstLefts1(outRec1, outRec2);
+      }
+
+    } else {
+      // joined 2 polygons together ...
+
+      outRec2->Pts = 0;
+      outRec2->BottomPt = 0;
+      outRec2->Idx = outRec1->Idx;
+
+      outRec1->IsHole = holeStateRec->IsHole;
+      if (holeStateRec == outRec2)
+        outRec1->FirstLeft = outRec2->FirstLeft;
+      outRec2->FirstLeft = outRec1;
+
+      if (m_UsingPolyTree)
+        FixupFirstLefts3(outRec2, outRec1);
+    }
+  }
+}
+
+//------------------------------------------------------------------------------
+// ClipperOffset support functions ...
+//------------------------------------------------------------------------------
+
+DoublePoint GetUnitNormal(const IntPoint &pt1, const IntPoint &pt2) {
+  if (pt2.X == pt1.X && pt2.Y == pt1.Y)
+    return DoublePoint(0, 0);
+
+  double Dx = (double)(pt2.X - pt1.X);
+  double dy = (double)(pt2.Y - pt1.Y);
+  double f = 1 * 1.0 / std::sqrt(Dx * Dx + dy * dy);
+  Dx *= f;
+  dy *= f;
+  return DoublePoint(dy, -Dx);
+}
+
+//------------------------------------------------------------------------------
+// ClipperOffset class
+//------------------------------------------------------------------------------
+
+ClipperOffset::ClipperOffset(double miterLimit, double arcTolerance) {
+  this->MiterLimit = miterLimit;
+  this->ArcTolerance = arcTolerance;
+  m_lowest.X = -1;
+}
+//------------------------------------------------------------------------------
+
+ClipperOffset::~ClipperOffset() { Clear(); }
+//------------------------------------------------------------------------------
+
+void ClipperOffset::Clear() {
+  for (int i = 0; i < m_polyNodes.ChildCount(); ++i)
+    delete m_polyNodes.Children[i];
+  m_polyNodes.Children.clear();
+  m_lowest.X = -1;
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::AddPath(const Path &path, JoinType joinType,
+                            EndType endType) {
+  int highI = (int)path.size() - 1;
+  if (highI < 0)
+    return;
+  PolyNode *newNode = new PolyNode();
+  newNode->m_jointype = joinType;
+  newNode->m_endtype = endType;
+
+  // strip duplicate points from path and also get index to the lowest point ...
+  if (endType == etClosedLine || endType == etClosedPolygon)
+    while (highI > 0 && path[0] == path[highI])
+      highI--;
+  newNode->Contour.reserve(highI + 1);
+  newNode->Contour.push_back(path[0]);
+  int j = 0, k = 0;
+  for (int i = 1; i <= highI; i++)
+    if (newNode->Contour[j] != path[i]) {
+      j++;
+      newNode->Contour.push_back(path[i]);
+      if (path[i].Y > newNode->Contour[k].Y ||
+          (path[i].Y == newNode->Contour[k].Y &&
+           path[i].X < newNode->Contour[k].X))
+        k = j;
+    }
+  if (endType == etClosedPolygon && j < 2) {
+    delete newNode;
+    return;
+  }
+  m_polyNodes.AddChild(*newNode);
+
+  // if this path's lowest pt is lower than all the others then update m_lowest
+  if (endType != etClosedPolygon)
+    return;
+  if (m_lowest.X < 0)
+    m_lowest = IntPoint(m_polyNodes.ChildCount() - 1, k);
+  else {
+    IntPoint ip =
+        m_polyNodes.Children[(int)m_lowest.X]->Contour[(int)m_lowest.Y];
+    if (newNode->Contour[k].Y > ip.Y ||
+        (newNode->Contour[k].Y == ip.Y && newNode->Contour[k].X < ip.X))
+      m_lowest = IntPoint(m_polyNodes.ChildCount() - 1, k);
+  }
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::AddPaths(const Paths &paths, JoinType joinType,
+                             EndType endType) {
+  for (Paths::size_type i = 0; i < paths.size(); ++i)
+    AddPath(paths[i], joinType, endType);
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::FixOrientations() {
+  // fixup orientations of all closed paths if the orientation of the
+  // closed path with the lowermost vertex is wrong ...
+  if (m_lowest.X >= 0 &&
+      !Orientation(m_polyNodes.Children[(int)m_lowest.X]->Contour)) {
+    for (int i = 0; i < m_polyNodes.ChildCount(); ++i) {
+      PolyNode &node = *m_polyNodes.Children[i];
+      if (node.m_endtype == etClosedPolygon ||
+          (node.m_endtype == etClosedLine && Orientation(node.Contour)))
+        ReversePath(node.Contour);
+    }
+  } else {
+    for (int i = 0; i < m_polyNodes.ChildCount(); ++i) {
+      PolyNode &node = *m_polyNodes.Children[i];
+      if (node.m_endtype == etClosedLine && !Orientation(node.Contour))
+        ReversePath(node.Contour);
+    }
+  }
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::Execute(Paths &solution, double delta) {
+  solution.clear();
+  FixOrientations();
+  DoOffset(delta);
+
+  // now clean up 'corners' ...
+  Clipper clpr;
+  clpr.AddPaths(m_destPolys, ptSubject, true);
+  if (delta > 0) {
+    clpr.Execute(ctUnion, solution, pftPositive, pftPositive);
+  } else {
+    IntRect r = clpr.GetBounds();
+    Path outer(4);
+    outer[0] = IntPoint(r.left - 10, r.bottom + 10);
+    outer[1] = IntPoint(r.right + 10, r.bottom + 10);
+    outer[2] = IntPoint(r.right + 10, r.top - 10);
+    outer[3] = IntPoint(r.left - 10, r.top - 10);
+
+    clpr.AddPath(outer, ptSubject, true);
+    clpr.ReverseSolution(true);
+    clpr.Execute(ctUnion, solution, pftNegative, pftNegative);
+    if (solution.size() > 0)
+      solution.erase(solution.begin());
+  }
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::Execute(PolyTree &solution, double delta) {
+  solution.Clear();
+  FixOrientations();
+  DoOffset(delta);
+
+  // now clean up 'corners' ...
+  Clipper clpr;
+  clpr.AddPaths(m_destPolys, ptSubject, true);
+  if (delta > 0) {
+    clpr.Execute(ctUnion, solution, pftPositive, pftPositive);
+  } else {
+    IntRect r = clpr.GetBounds();
+    Path outer(4);
+    outer[0] = IntPoint(r.left - 10, r.bottom + 10);
+    outer[1] = IntPoint(r.right + 10, r.bottom + 10);
+    outer[2] = IntPoint(r.right + 10, r.top - 10);
+    outer[3] = IntPoint(r.left - 10, r.top - 10);
+
+    clpr.AddPath(outer, ptSubject, true);
+    clpr.ReverseSolution(true);
+    clpr.Execute(ctUnion, solution, pftNegative, pftNegative);
+    // remove the outer PolyNode rectangle ...
+    if (solution.ChildCount() == 1 && solution.Children[0]->ChildCount() > 0) {
+      PolyNode *outerNode = solution.Children[0];
+      solution.Children.reserve(outerNode->ChildCount());
+      solution.Children[0] = outerNode->Children[0];
+      solution.Children[0]->Parent = outerNode->Parent;
+      for (int i = 1; i < outerNode->ChildCount(); ++i)
+        solution.AddChild(*outerNode->Children[i]);
+    } else
+      solution.Clear();
+  }
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::DoOffset(double delta) {
+  m_destPolys.clear();
+  m_delta = delta;
+
+  // if Zero offset, just copy any CLOSED polygons to m_p and return ...
+  if (NEAR_ZERO(delta)) {
+    m_destPolys.reserve(m_polyNodes.ChildCount());
+    for (int i = 0; i < m_polyNodes.ChildCount(); i++) {
+      PolyNode &node = *m_polyNodes.Children[i];
+      if (node.m_endtype == etClosedPolygon)
+        m_destPolys.push_back(node.Contour);
+    }
+    return;
+  }
+
+  // see offset_triginometry3.svg in the documentation folder ...
+  if (MiterLimit > 2)
+    m_miterLim = 2 / (MiterLimit * MiterLimit);
+  else
+    m_miterLim = 0.5;
+
+  double y;
+  if (ArcTolerance <= 0.0)
+    y = def_arc_tolerance;
+  else if (ArcTolerance > std::fabs(delta) * def_arc_tolerance)
+    y = std::fabs(delta) * def_arc_tolerance;
+  else
+    y = ArcTolerance;
+  // see offset_triginometry2.svg in the documentation folder ...
+  double steps = pi / std::acos(1 - y / std::fabs(delta));
+  if (steps > std::fabs(delta) * pi)
+    steps = std::fabs(delta) * pi; // ie excessive precision check
+  m_sin = std::sin(two_pi / steps);
+  m_cos = std::cos(two_pi / steps);
+  m_StepsPerRad = steps / two_pi;
+  if (delta < 0.0)
+    m_sin = -m_sin;
+
+  m_destPolys.reserve(m_polyNodes.ChildCount() * 2);
+  for (int i = 0; i < m_polyNodes.ChildCount(); i++) {
+    PolyNode &node = *m_polyNodes.Children[i];
+    m_srcPoly = node.Contour;
+
+    int len = (int)m_srcPoly.size();
+    if (len == 0 ||
+        (delta <= 0 && (len < 3 || node.m_endtype != etClosedPolygon)))
+      continue;
+
+    m_destPoly.clear();
+    if (len == 1) {
+      if (node.m_jointype == jtRound) {
+        double X = 1.0, Y = 0.0;
+        for (cInt j = 1; j <= steps; j++) {
+          m_destPoly.push_back(IntPoint(Round(m_srcPoly[0].X + X * delta),
+                                        Round(m_srcPoly[0].Y + Y * delta)));
+          double X2 = X;
+          X = X * m_cos - m_sin * Y;
+          Y = X2 * m_sin + Y * m_cos;
+        }
+      } else {
+        double X = -1.0, Y = -1.0;
+        for (int j = 0; j < 4; ++j) {
+          m_destPoly.push_back(IntPoint(Round(m_srcPoly[0].X + X * delta),
+                                        Round(m_srcPoly[0].Y + Y * delta)));
+          if (X < 0)
+            X = 1;
+          else if (Y < 0)
+            Y = 1;
+          else
+            X = -1;
+        }
+      }
+      m_destPolys.push_back(m_destPoly);
+      continue;
+    }
+    // build m_normals ...
+    m_normals.clear();
+    m_normals.reserve(len);
+    for (int j = 0; j < len - 1; ++j)
+      m_normals.push_back(GetUnitNormal(m_srcPoly[j], m_srcPoly[j + 1]));
+    if (node.m_endtype == etClosedLine || node.m_endtype == etClosedPolygon)
+      m_normals.push_back(GetUnitNormal(m_srcPoly[len - 1], m_srcPoly[0]));
+    else
+      m_normals.push_back(DoublePoint(m_normals[len - 2]));
+
+    if (node.m_endtype == etClosedPolygon) {
+      int k = len - 1;
+      for (int j = 0; j < len; ++j)
+        OffsetPoint(j, k, node.m_jointype);
+      m_destPolys.push_back(m_destPoly);
+    } else if (node.m_endtype == etClosedLine) {
+      int k = len - 1;
+      for (int j = 0; j < len; ++j)
+        OffsetPoint(j, k, node.m_jointype);
+      m_destPolys.push_back(m_destPoly);
+      m_destPoly.clear();
+      // re-build m_normals ...
+      DoublePoint n = m_normals[len - 1];
+      for (int j = len - 1; j > 0; j--)
+        m_normals[j] = DoublePoint(-m_normals[j - 1].X, -m_normals[j - 1].Y);
+      m_normals[0] = DoublePoint(-n.X, -n.Y);
+      k = 0;
+      for (int j = len - 1; j >= 0; j--)
+        OffsetPoint(j, k, node.m_jointype);
+      m_destPolys.push_back(m_destPoly);
+    } else {
+      int k = 0;
+      for (int j = 1; j < len - 1; ++j)
+        OffsetPoint(j, k, node.m_jointype);
+
+      IntPoint pt1;
+      if (node.m_endtype == etOpenButt) {
+        int j = len - 1;
+        pt1 = IntPoint((cInt)Round(m_srcPoly[j].X + m_normals[j].X * delta),
+                       (cInt)Round(m_srcPoly[j].Y + m_normals[j].Y * delta));
+        m_destPoly.push_back(pt1);
+        pt1 = IntPoint((cInt)Round(m_srcPoly[j].X - m_normals[j].X * delta),
+                       (cInt)Round(m_srcPoly[j].Y - m_normals[j].Y * delta));
+        m_destPoly.push_back(pt1);
+      } else {
+        int j = len - 1;
+        k = len - 2;
+        m_sinA = 0;
+        m_normals[j] = DoublePoint(-m_normals[j].X, -m_normals[j].Y);
+        if (node.m_endtype == etOpenSquare)
+          DoSquare(j, k);
+        else
+          DoRound(j, k);
+      }
+
+      // re-build m_normals ...
+      for (int j = len - 1; j > 0; j--)
+        m_normals[j] = DoublePoint(-m_normals[j - 1].X, -m_normals[j - 1].Y);
+      m_normals[0] = DoublePoint(-m_normals[1].X, -m_normals[1].Y);
+
+      k = len - 1;
+      for (int j = k - 1; j > 0; --j)
+        OffsetPoint(j, k, node.m_jointype);
+
+      if (node.m_endtype == etOpenButt) {
+        pt1 = IntPoint((cInt)Round(m_srcPoly[0].X - m_normals[0].X * delta),
+                       (cInt)Round(m_srcPoly[0].Y - m_normals[0].Y * delta));
+        m_destPoly.push_back(pt1);
+        pt1 = IntPoint((cInt)Round(m_srcPoly[0].X + m_normals[0].X * delta),
+                       (cInt)Round(m_srcPoly[0].Y + m_normals[0].Y * delta));
+        m_destPoly.push_back(pt1);
+      } else {
+        k = 1;
+        m_sinA = 0;
+        if (node.m_endtype == etOpenSquare)
+          DoSquare(0, 1);
+        else
+          DoRound(0, 1);
+      }
+      m_destPolys.push_back(m_destPoly);
+    }
+  }
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::OffsetPoint(int j, int &k, JoinType jointype) {
+  // cross product ...
+  m_sinA = (m_normals[k].X * m_normals[j].Y - m_normals[j].X * m_normals[k].Y);
+  if (std::fabs(m_sinA * m_delta) < 1.0) {
+    // dot product ...
+    double cosA =
+        (m_normals[k].X * m_normals[j].X + m_normals[j].Y * m_normals[k].Y);
+    if (cosA > 0) // angle => 0 degrees
+    {
+      m_destPoly.push_back(
+          IntPoint(Round(m_srcPoly[j].X + m_normals[k].X * m_delta),
+                   Round(m_srcPoly[j].Y + m_normals[k].Y * m_delta)));
+      return;
+    }
+    // else angle => 180 degrees
+  } else if (m_sinA > 1.0)
+    m_sinA = 1.0;
+  else if (m_sinA < -1.0)
+    m_sinA = -1.0;
+
+  if (m_sinA * m_delta < 0) {
+    m_destPoly.push_back(
+        IntPoint(Round(m_srcPoly[j].X + m_normals[k].X * m_delta),
+                 Round(m_srcPoly[j].Y + m_normals[k].Y * m_delta)));
+    m_destPoly.push_back(m_srcPoly[j]);
+    m_destPoly.push_back(
+        IntPoint(Round(m_srcPoly[j].X + m_normals[j].X * m_delta),
+                 Round(m_srcPoly[j].Y + m_normals[j].Y * m_delta)));
+  } else
+    switch (jointype) {
+    case jtMiter: {
+      double r = 1 + (m_normals[j].X * m_normals[k].X +
+                      m_normals[j].Y * m_normals[k].Y);
+      if (r >= m_miterLim)
+        DoMiter(j, k, r);
+      else
+        DoSquare(j, k);
+      break;
+    }
+    case jtSquare:
+      DoSquare(j, k);
+      break;
+    case jtRound:
+      DoRound(j, k);
+      break;
+    }
+  k = j;
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::DoSquare(int j, int k) {
+  double dx = std::tan(std::atan2(m_sinA, m_normals[k].X * m_normals[j].X +
+                                              m_normals[k].Y * m_normals[j].Y) /
+                       4);
+  m_destPoly.push_back(IntPoint(
+      Round(m_srcPoly[j].X + m_delta * (m_normals[k].X - m_normals[k].Y * dx)),
+      Round(m_srcPoly[j].Y +
+            m_delta * (m_normals[k].Y + m_normals[k].X * dx))));
+  m_destPoly.push_back(IntPoint(
+      Round(m_srcPoly[j].X + m_delta * (m_normals[j].X + m_normals[j].Y * dx)),
+      Round(m_srcPoly[j].Y +
+            m_delta * (m_normals[j].Y - m_normals[j].X * dx))));
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::DoMiter(int j, int k, double r) {
+  double q = m_delta / r;
+  m_destPoly.push_back(
+      IntPoint(Round(m_srcPoly[j].X + (m_normals[k].X + m_normals[j].X) * q),
+               Round(m_srcPoly[j].Y + (m_normals[k].Y + m_normals[j].Y) * q)));
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::DoRound(int j, int k) {
+  double a = std::atan2(m_sinA, m_normals[k].X * m_normals[j].X +
+                                    m_normals[k].Y * m_normals[j].Y);
+  int steps = std::max((int)Round(m_StepsPerRad * std::fabs(a)), 1);
+
+  double X = m_normals[k].X, Y = m_normals[k].Y, X2;
+  for (int i = 0; i < steps; ++i) {
+    m_destPoly.push_back(IntPoint(Round(m_srcPoly[j].X + X * m_delta),
+                                  Round(m_srcPoly[j].Y + Y * m_delta)));
+    X2 = X;
+    X = X * m_cos - m_sin * Y;
+    Y = X2 * m_sin + Y * m_cos;
+  }
+  m_destPoly.push_back(
+      IntPoint(Round(m_srcPoly[j].X + m_normals[j].X * m_delta),
+               Round(m_srcPoly[j].Y + m_normals[j].Y * m_delta)));
+}
+
+//------------------------------------------------------------------------------
+// Miscellaneous public functions
+//------------------------------------------------------------------------------
+
+void Clipper::DoSimplePolygons() {
+  PolyOutList::size_type i = 0;
+  while (i < m_PolyOuts.size()) {
+    OutRec *outrec = m_PolyOuts[i++];
+    OutPt *op = outrec->Pts;
+    if (!op || outrec->IsOpen)
+      continue;
+    do // for each Pt in Polygon until duplicate found do ...
+    {
+      OutPt *op2 = op->Next;
+      while (op2 != outrec->Pts) {
+        if ((op->Pt == op2->Pt) && op2->Next != op && op2->Prev != op) {
+          // split the polygon into two ...
+          OutPt *op3 = op->Prev;
+          OutPt *op4 = op2->Prev;
+          op->Prev = op4;
+          op4->Next = op;
+          op2->Prev = op3;
+          op3->Next = op2;
+
+          outrec->Pts = op;
+          OutRec *outrec2 = CreateOutRec();
+          outrec2->Pts = op2;
+          UpdateOutPtIdxs(*outrec2);
+          if (Poly2ContainsPoly1(outrec2->Pts, outrec->Pts)) {
+            // OutRec2 is contained by OutRec1 ...
+            outrec2->IsHole = !outrec->IsHole;
+            outrec2->FirstLeft = outrec;
+            if (m_UsingPolyTree)
+              FixupFirstLefts2(outrec2, outrec);
+          } else if (Poly2ContainsPoly1(outrec->Pts, outrec2->Pts)) {
+            // OutRec1 is contained by OutRec2 ...
+            outrec2->IsHole = outrec->IsHole;
+            outrec->IsHole = !outrec2->IsHole;
+            outrec2->FirstLeft = outrec->FirstLeft;
+            outrec->FirstLeft = outrec2;
+            if (m_UsingPolyTree)
+              FixupFirstLefts2(outrec, outrec2);
+          } else {
+            // the 2 polygons are separate ...
+            outrec2->IsHole = outrec->IsHole;
+            outrec2->FirstLeft = outrec->FirstLeft;
+            if (m_UsingPolyTree)
+              FixupFirstLefts1(outrec, outrec2);
+          }
+          op2 = op; // ie get ready for the Next iteration
+        }
+        op2 = op2->Next;
+      }
+      op = op->Next;
+    } while (op != outrec->Pts);
+  }
+}
+//------------------------------------------------------------------------------
+
+void ReversePath(Path &p) { std::reverse(p.begin(), p.end()); }
+//------------------------------------------------------------------------------
+
+void ReversePaths(Paths &p) {
+  for (Paths::size_type i = 0; i < p.size(); ++i)
+    ReversePath(p[i]);
+}
+//------------------------------------------------------------------------------
+
+void SimplifyPolygon(const Path &in_poly, Paths &out_polys,
+                     PolyFillType fillType) {
+  Clipper c;
+  c.StrictlySimple(true);
+  c.AddPath(in_poly, ptSubject, true);
+  c.Execute(ctUnion, out_polys, fillType, fillType);
+}
+//------------------------------------------------------------------------------
+
+void SimplifyPolygons(const Paths &in_polys, Paths &out_polys,
+                      PolyFillType fillType) {
+  Clipper c;
+  c.StrictlySimple(true);
+  c.AddPaths(in_polys, ptSubject, true);
+  c.Execute(ctUnion, out_polys, fillType, fillType);
+}
+//------------------------------------------------------------------------------
+
+void SimplifyPolygons(Paths &polys, PolyFillType fillType) {
+  SimplifyPolygons(polys, polys, fillType);
+}
+//------------------------------------------------------------------------------
+
+inline double DistanceSqrd(const IntPoint &pt1, const IntPoint &pt2) {
+  double Dx = ((double)pt1.X - pt2.X);
+  double dy = ((double)pt1.Y - pt2.Y);
+  return (Dx * Dx + dy * dy);
+}
+//------------------------------------------------------------------------------
+
+double DistanceFromLineSqrd(const IntPoint &pt, const IntPoint &ln1,
+                            const IntPoint &ln2) {
+  // The equation of a line in general form (Ax + By + C = 0)
+  // given 2 points (x�,y�) & (x�,y�) is ...
+  //(y� - y�)x + (x� - x�)y + (y� - y�)x� - (x� - x�)y� = 0
+  // A = (y� - y�); B = (x� - x�); C = (y� - y�)x� - (x� - x�)y�
+  // perpendicular distance of point (x�,y�) = (Ax� + By� + C)/Sqrt(A� + B�)
+  // see http://en.wikipedia.org/wiki/Perpendicular_distance
+  double A = double(ln1.Y - ln2.Y);
+  double B = double(ln2.X - ln1.X);
+  double C = A * ln1.X + B * ln1.Y;
+  C = A * pt.X + B * pt.Y - C;
+  return (C * C) / (A * A + B * B);
+}
+//---------------------------------------------------------------------------
+
+bool SlopesNearCollinear(const IntPoint &pt1, const IntPoint &pt2,
+                         const IntPoint &pt3, double distSqrd) {
+  // this function is more accurate when the point that's geometrically
+  // between the other 2 points is the one that's tested for distance.
+  // ie makes it more likely to pick up 'spikes' ...
+  if (Abs(pt1.X - pt2.X) > Abs(pt1.Y - pt2.Y)) {
+    if ((pt1.X > pt2.X) == (pt1.X < pt3.X))
+      return DistanceFromLineSqrd(pt1, pt2, pt3) < distSqrd;
+    else if ((pt2.X > pt1.X) == (pt2.X < pt3.X))
+      return DistanceFromLineSqrd(pt2, pt1, pt3) < distSqrd;
+    else
+      return DistanceFromLineSqrd(pt3, pt1, pt2) < distSqrd;
+  } else {
+    if ((pt1.Y > pt2.Y) == (pt1.Y < pt3.Y))
+      return DistanceFromLineSqrd(pt1, pt2, pt3) < distSqrd;
+    else if ((pt2.Y > pt1.Y) == (pt2.Y < pt3.Y))
+      return DistanceFromLineSqrd(pt2, pt1, pt3) < distSqrd;
+    else
+      return DistanceFromLineSqrd(pt3, pt1, pt2) < distSqrd;
+  }
+}
+//------------------------------------------------------------------------------
+
+bool PointsAreClose(IntPoint pt1, IntPoint pt2, double distSqrd) {
+  double Dx = (double)pt1.X - pt2.X;
+  double dy = (double)pt1.Y - pt2.Y;
+  return ((Dx * Dx) + (dy * dy) <= distSqrd);
+}
+//------------------------------------------------------------------------------
+
+OutPt *ExcludeOp(OutPt *op) {
+  OutPt *result = op->Prev;
+  result->Next = op->Next;
+  op->Next->Prev = result;
+  result->Idx = 0;
+  return result;
+}
+//------------------------------------------------------------------------------
+
+void CleanPolygon(const Path &in_poly, Path &out_poly, double distance) {
+  // distance = proximity in units/pixels below which vertices
+  // will be stripped. Default ~= sqrt(2).
+
+  size_t size = in_poly.size();
+
+  if (size == 0) {
+    out_poly.clear();
+    return;
+  }
+
+  OutPt *outPts = new OutPt[size];
+  for (size_t i = 0; i < size; ++i) {
+    outPts[i].Pt = in_poly[i];
+    outPts[i].Next = &outPts[(i + 1) % size];
+    outPts[i].Next->Prev = &outPts[i];
+    outPts[i].Idx = 0;
+  }
+
+  double distSqrd = distance * distance;
+  OutPt *op = &outPts[0];
+  while (op->Idx == 0 && op->Next != op->Prev) {
+    if (PointsAreClose(op->Pt, op->Prev->Pt, distSqrd)) {
+      op = ExcludeOp(op);
+      size--;
+    } else if (PointsAreClose(op->Prev->Pt, op->Next->Pt, distSqrd)) {
+      ExcludeOp(op->Next);
+      op = ExcludeOp(op);
+      size -= 2;
+    } else if (SlopesNearCollinear(op->Prev->Pt, op->Pt, op->Next->Pt,
+                                   distSqrd)) {
+      op = ExcludeOp(op);
+      size--;
+    } else {
+      op->Idx = 1;
+      op = op->Next;
+    }
+  }
+
+  if (size < 3)
+    size = 0;
+  out_poly.resize(size);
+  for (size_t i = 0; i < size; ++i) {
+    out_poly[i] = op->Pt;
+    op = op->Next;
+  }
+  delete[] outPts;
+}
+//------------------------------------------------------------------------------
+
+void CleanPolygon(Path &poly, double distance) {
+  CleanPolygon(poly, poly, distance);
+}
+//------------------------------------------------------------------------------
+
+void CleanPolygons(const Paths &in_polys, Paths &out_polys, double distance) {
+  out_polys.resize(in_polys.size());
+  for (Paths::size_type i = 0; i < in_polys.size(); ++i)
+    CleanPolygon(in_polys[i], out_polys[i], distance);
+}
+//------------------------------------------------------------------------------
+
+void CleanPolygons(Paths &polys, double distance) {
+  CleanPolygons(polys, polys, distance);
+}
+//------------------------------------------------------------------------------
+
+void Minkowski(const Path &poly, const Path &path, Paths &solution, bool isSum,
+               bool isClosed) {
+  int delta = (isClosed ? 1 : 0);
+  size_t polyCnt = poly.size();
+  size_t pathCnt = path.size();
+  Paths pp;
+  pp.reserve(pathCnt);
+  if (isSum)
+    for (size_t i = 0; i < pathCnt; ++i) {
+      Path p;
+      p.reserve(polyCnt);
+      for (size_t j = 0; j < poly.size(); ++j)
+        p.push_back(IntPoint(path[i].X + poly[j].X, path[i].Y + poly[j].Y));
+      pp.push_back(p);
+    }
+  else
+    for (size_t i = 0; i < pathCnt; ++i) {
+      Path p;
+      p.reserve(polyCnt);
+      for (size_t j = 0; j < poly.size(); ++j)
+        p.push_back(IntPoint(path[i].X - poly[j].X, path[i].Y - poly[j].Y));
+      pp.push_back(p);
+    }
+
+  solution.clear();
+  solution.reserve((pathCnt + delta) * (polyCnt + 1));
+  for (size_t i = 0; i < pathCnt - 1 + delta; ++i)
+    for (size_t j = 0; j < polyCnt; ++j) {
+      Path quad;
+      quad.reserve(4);
+      quad.push_back(pp[i % pathCnt][j % polyCnt]);
+      quad.push_back(pp[(i + 1) % pathCnt][j % polyCnt]);
+      quad.push_back(pp[(i + 1) % pathCnt][(j + 1) % polyCnt]);
+      quad.push_back(pp[i % pathCnt][(j + 1) % polyCnt]);
+      if (!Orientation(quad))
+        ReversePath(quad);
+      solution.push_back(quad);
+    }
+}
+//------------------------------------------------------------------------------
+
+void MinkowskiSum(const Path &pattern, const Path &path, Paths &solution,
+                  bool pathIsClosed) {
+  Minkowski(pattern, path, solution, true, pathIsClosed);
+  Clipper c;
+  c.AddPaths(solution, ptSubject, true);
+  c.Execute(ctUnion, solution, pftNonZero, pftNonZero);
+}
+//------------------------------------------------------------------------------
+
+void TranslatePath(const Path &input, Path &output, const IntPoint delta) {
+  // precondition: input != output
+  output.resize(input.size());
+  for (size_t i = 0; i < input.size(); ++i)
+    output[i] = IntPoint(input[i].X + delta.X, input[i].Y + delta.Y);
+}
+//------------------------------------------------------------------------------
+
+void MinkowskiSum(const Path &pattern, const Paths &paths, Paths &solution,
+                  bool pathIsClosed) {
+  Clipper c;
+  for (size_t i = 0; i < paths.size(); ++i) {
+    Paths tmp;
+    Minkowski(pattern, paths[i], tmp, true, pathIsClosed);
+    c.AddPaths(tmp, ptSubject, true);
+    if (pathIsClosed) {
+      Path tmp2;
+      TranslatePath(paths[i], tmp2, pattern[0]);
+      c.AddPath(tmp2, ptClip, true);
+    }
+  }
+  c.Execute(ctUnion, solution, pftNonZero, pftNonZero);
+}
+//------------------------------------------------------------------------------
+
+void MinkowskiDiff(const Path &poly1, const Path &poly2, Paths &solution) {
+  Minkowski(poly1, poly2, solution, false, true);
+  Clipper c;
+  c.AddPaths(solution, ptSubject, true);
+  c.Execute(ctUnion, solution, pftNonZero, pftNonZero);
+}
+//------------------------------------------------------------------------------
+
+enum NodeType { ntAny, ntOpen, ntClosed };
+
+void AddPolyNodeToPaths(const PolyNode &polynode, NodeType nodetype,
+                        Paths &paths) {
+  bool match = true;
+  if (nodetype == ntClosed)
+    match = !polynode.IsOpen();
+  else if (nodetype == ntOpen)
+    return;
+
+  if (!polynode.Contour.empty() && match)
+    paths.push_back(polynode.Contour);
+  for (int i = 0; i < polynode.ChildCount(); ++i)
+    AddPolyNodeToPaths(*polynode.Children[i], nodetype, paths);
+}
+//------------------------------------------------------------------------------
+
+void PolyTreeToPaths(const PolyTree &polytree, Paths &paths) {
+  paths.resize(0);
+  paths.reserve(polytree.Total());
+  AddPolyNodeToPaths(polytree, ntAny, paths);
+}
+//------------------------------------------------------------------------------
+
+void ClosedPathsFromPolyTree(const PolyTree &polytree, Paths &paths) {
+  paths.resize(0);
+  paths.reserve(polytree.Total());
+  AddPolyNodeToPaths(polytree, ntClosed, paths);
+}
+//------------------------------------------------------------------------------
+
+void OpenPathsFromPolyTree(PolyTree &polytree, Paths &paths) {
+  paths.resize(0);
+  paths.reserve(polytree.Total());
+  // Open paths are top level only, so ...
+  for (int i = 0; i < polytree.ChildCount(); ++i)
+    if (polytree.Children[i]->IsOpen())
+      paths.push_back(polytree.Children[i]->Contour);
+}
+//------------------------------------------------------------------------------
+
+std::ostream &operator<<(std::ostream &s, const IntPoint &p) {
+  s << "(" << p.X << "," << p.Y << ")";
+  return s;
+}
+//------------------------------------------------------------------------------
+
+std::ostream &operator<<(std::ostream &s, const Path &p) {
+  if (p.empty())
+    return s;
+  Path::size_type last = p.size() - 1;
+  for (Path::size_type i = 0; i < last; i++)
+    s << "(" << p[i].X << "," << p[i].Y << "), ";
+  s << "(" << p[last].X << "," << p[last].Y << ")\n";
+  return s;
+}
+//------------------------------------------------------------------------------
+
+std::ostream &operator<<(std::ostream &s, const Paths &p) {
+  for (Paths::size_type i = 0; i < p.size(); i++)
+    s << p[i];
+  s << "\n";
+  return s;
+}
+//------------------------------------------------------------------------------
+
+} // namespace ClipperLib
diff --git a/deploy/android_demo/app/src/main/cpp/ocr_clipper.hpp b/deploy/android_demo/app/src/main/cpp/ocr_clipper.hpp
new file mode 100644
index 0000000..0d1ac24
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/ocr_clipper.hpp
@@ -0,0 +1,540 @@
+/*******************************************************************************
+ *                                                                              *
+ * Author    :  Angus Johnson * Version   :  6.4.2 * Date      :  27 February
+ *2017                                                * Website   :
+ *http://www.angusj.com                                           * Copyright :
+ *Angus Johnson 2010-2017                                         *
+ *                                                                              *
+ * License: * Use, modification & distribution is subject to Boost Software
+ *License Ver 1. * http://www.boost.org/LICENSE_1_0.txt *
+ *                                                                              *
+ * Attributions: * The code in this library is an extension of Bala Vatti's
+ *clipping algorithm: * "A generic solution to polygon clipping" *
+ * Communications of the ACM, Vol 35, Issue 7 (July 1992) pp 56-63. *
+ * http://portal.acm.org/citation.cfm?id=129906 *
+ *                                                                              *
+ * Computer graphics and geometric modeling: implementation and algorithms * By
+ *Max K. Agoston                                                            *
+ * Springer; 1 edition (January 4, 2005) *
+ * http://books.google.com/books?q=vatti+clipping+agoston *
+ *                                                                              *
+ * See also: * "Polygon Offsetting by Computing Winding Numbers" * Paper no.
+ *DETC2005-85513 pp. 565-575                                         * ASME 2005
+ *International Design Engineering Technical Conferences             * and
+ *Computers and Information in Engineering Conference (IDETC/CIE2005)      *
+ * September 24-28, 2005 , Long Beach, California, USA *
+ * http://www.me.berkeley.edu/~mcmains/pubs/DAC05OffsetPolygon.pdf *
+ *                                                                              *
+ *******************************************************************************/
+
+#ifndef clipper_hpp
+#define clipper_hpp
+
+#define CLIPPER_VERSION "6.4.2"
+
+// use_int32: When enabled 32bit ints are used instead of 64bit ints. This
+// improve performance but coordinate values are limited to the range +/- 46340
+//#define use_int32
+
+// use_xyz: adds a Z member to IntPoint. Adds a minor cost to performance.
+//#define use_xyz
+
+// use_lines: Enables line clipping. Adds a very minor cost to performance.
+#define use_lines
+
+// use_deprecated: Enables temporary support for the obsolete functions
+//#define use_deprecated
+
+#include <cstdlib>
+#include <cstring>
+#include <functional>
+#include <list>
+#include <ostream>
+#include <queue>
+#include <set>
+#include <stdexcept>
+#include <vector>
+
+namespace ClipperLib {
+
+enum ClipType { ctIntersection, ctUnion, ctDifference, ctXor };
+enum PolyType { ptSubject, ptClip };
+// By far the most widely used winding rules for polygon filling are
+// EvenOdd & NonZero (GDI, GDI+, XLib, OpenGL, Cairo, AGG, Quartz, SVG, Gr32)
+// Others rules include Positive, Negative and ABS_GTR_EQ_TWO (only in OpenGL)
+// see http://glprogramming.com/red/chapter11.html
+enum PolyFillType { pftEvenOdd, pftNonZero, pftPositive, pftNegative };
+
+#ifdef use_int32
+typedef int cInt;
+static cInt const loRange = 0x7FFF;
+static cInt const hiRange = 0x7FFF;
+#else
+typedef signed long long cInt;
+static cInt const loRange = 0x3FFFFFFF;
+static cInt const hiRange = 0x3FFFFFFFFFFFFFFFLL;
+typedef signed long long long64; // used by Int128 class
+typedef unsigned long long ulong64;
+
+#endif
+
+struct IntPoint {
+  cInt X;
+  cInt Y;
+#ifdef use_xyz
+  cInt Z;
+  IntPoint(cInt x = 0, cInt y = 0, cInt z = 0) : X(x), Y(y), Z(z){};
+#else
+
+  IntPoint(cInt x = 0, cInt y = 0) : X(x), Y(y){};
+#endif
+
+  friend inline bool operator==(const IntPoint &a, const IntPoint &b) {
+    return a.X == b.X && a.Y == b.Y;
+  }
+
+  friend inline bool operator!=(const IntPoint &a, const IntPoint &b) {
+    return a.X != b.X || a.Y != b.Y;
+  }
+};
+//------------------------------------------------------------------------------
+
+typedef std::vector<IntPoint> Path;
+typedef std::vector<Path> Paths;
+
+inline Path &operator<<(Path &poly, const IntPoint &p) {
+  poly.push_back(p);
+  return poly;
+}
+
+inline Paths &operator<<(Paths &polys, const Path &p) {
+  polys.push_back(p);
+  return polys;
+}
+
+std::ostream &operator<<(std::ostream &s, const IntPoint &p);
+
+std::ostream &operator<<(std::ostream &s, const Path &p);
+
+std::ostream &operator<<(std::ostream &s, const Paths &p);
+
+struct DoublePoint {
+  double X;
+  double Y;
+
+  DoublePoint(double x = 0, double y = 0) : X(x), Y(y) {}
+
+  DoublePoint(IntPoint ip) : X((double)ip.X), Y((double)ip.Y) {}
+};
+//------------------------------------------------------------------------------
+
+#ifdef use_xyz
+typedef void (*ZFillCallback)(IntPoint &e1bot, IntPoint &e1top, IntPoint &e2bot,
+                              IntPoint &e2top, IntPoint &pt);
+#endif
+
+enum InitOptions {
+  ioReverseSolution = 1,
+  ioStrictlySimple = 2,
+  ioPreserveCollinear = 4
+};
+enum JoinType { jtSquare, jtRound, jtMiter };
+enum EndType {
+  etClosedPolygon,
+  etClosedLine,
+  etOpenButt,
+  etOpenSquare,
+  etOpenRound
+};
+
+class PolyNode;
+
+typedef std::vector<PolyNode *> PolyNodes;
+
+class PolyNode {
+public:
+  PolyNode();
+
+  virtual ~PolyNode(){};
+  Path Contour;
+  PolyNodes Children;
+  PolyNode *Parent;
+
+  PolyNode *GetNext() const;
+
+  bool IsHole() const;
+
+  bool IsOpen() const;
+
+  int ChildCount() const;
+
+private:
+  // PolyNode& operator =(PolyNode& other);
+  unsigned Index; // node index in Parent.Children
+  bool m_IsOpen;
+  JoinType m_jointype;
+  EndType m_endtype;
+
+  PolyNode *GetNextSiblingUp() const;
+
+  void AddChild(PolyNode &child);
+
+  friend class Clipper; // to access Index
+  friend class ClipperOffset;
+};
+
+class PolyTree : public PolyNode {
+public:
+  ~PolyTree() { Clear(); };
+
+  PolyNode *GetFirst() const;
+
+  void Clear();
+
+  int Total() const;
+
+private:
+  // PolyTree& operator =(PolyTree& other);
+  PolyNodes AllNodes;
+
+  friend class Clipper; // to access AllNodes
+};
+
+bool Orientation(const Path &poly);
+
+double Area(const Path &poly);
+
+int PointInPolygon(const IntPoint &pt, const Path &path);
+
+void SimplifyPolygon(const Path &in_poly, Paths &out_polys,
+                     PolyFillType fillType = pftEvenOdd);
+
+void SimplifyPolygons(const Paths &in_polys, Paths &out_polys,
+                      PolyFillType fillType = pftEvenOdd);
+
+void SimplifyPolygons(Paths &polys, PolyFillType fillType = pftEvenOdd);
+
+void CleanPolygon(const Path &in_poly, Path &out_poly, double distance = 1.415);
+
+void CleanPolygon(Path &poly, double distance = 1.415);
+
+void CleanPolygons(const Paths &in_polys, Paths &out_polys,
+                   double distance = 1.415);
+
+void CleanPolygons(Paths &polys, double distance = 1.415);
+
+void MinkowskiSum(const Path &pattern, const Path &path, Paths &solution,
+                  bool pathIsClosed);
+
+void MinkowskiSum(const Path &pattern, const Paths &paths, Paths &solution,
+                  bool pathIsClosed);
+
+void MinkowskiDiff(const Path &poly1, const Path &poly2, Paths &solution);
+
+void PolyTreeToPaths(const PolyTree &polytree, Paths &paths);
+
+void ClosedPathsFromPolyTree(const PolyTree &polytree, Paths &paths);
+
+void OpenPathsFromPolyTree(PolyTree &polytree, Paths &paths);
+
+void ReversePath(Path &p);
+
+void ReversePaths(Paths &p);
+
+struct IntRect {
+  cInt left;
+  cInt top;
+  cInt right;
+  cInt bottom;
+};
+
+// enums that are used internally ...
+enum EdgeSide { esLeft = 1, esRight = 2 };
+
+// forward declarations (for stuff used internally) ...
+struct TEdge;
+struct IntersectNode;
+struct LocalMinimum;
+struct OutPt;
+struct OutRec;
+struct Join;
+
+typedef std::vector<OutRec *> PolyOutList;
+typedef std::vector<TEdge *> EdgeList;
+typedef std::vector<Join *> JoinList;
+typedef std::vector<IntersectNode *> IntersectList;
+
+//------------------------------------------------------------------------------
+
+// ClipperBase is the ancestor to the Clipper class. It should not be
+// instantiated directly. This class simply abstracts the conversion of sets of
+// polygon coordinates into edge objects that are stored in a LocalMinima list.
+class ClipperBase {
+public:
+  ClipperBase();
+
+  virtual ~ClipperBase();
+
+  virtual bool AddPath(const Path &pg, PolyType PolyTyp, bool Closed);
+
+  bool AddPaths(const Paths &ppg, PolyType PolyTyp, bool Closed);
+
+  virtual void Clear();
+
+  IntRect GetBounds();
+
+  bool PreserveCollinear() { return m_PreserveCollinear; };
+
+  void PreserveCollinear(bool value) { m_PreserveCollinear = value; };
+
+protected:
+  void DisposeLocalMinimaList();
+
+  TEdge *AddBoundsToLML(TEdge *e, bool IsClosed);
+
+  virtual void Reset();
+
+  TEdge *ProcessBound(TEdge *E, bool IsClockwise);
+
+  void InsertScanbeam(const cInt Y);
+
+  bool PopScanbeam(cInt &Y);
+
+  bool LocalMinimaPending();
+
+  bool PopLocalMinima(cInt Y, const LocalMinimum *&locMin);
+
+  OutRec *CreateOutRec();
+
+  void DisposeAllOutRecs();
+
+  void DisposeOutRec(PolyOutList::size_type index);
+
+  void SwapPositionsInAEL(TEdge *edge1, TEdge *edge2);
+
+  void DeleteFromAEL(TEdge *e);
+
+  void UpdateEdgeIntoAEL(TEdge *&e);
+
+  typedef std::vector<LocalMinimum> MinimaList;
+  MinimaList::iterator m_CurrentLM;
+  MinimaList m_MinimaList;
+
+  bool m_UseFullRange;
+  EdgeList m_edges;
+  bool m_PreserveCollinear;
+  bool m_HasOpenPaths;
+  PolyOutList m_PolyOuts;
+  TEdge *m_ActiveEdges;
+
+  typedef std::priority_queue<cInt> ScanbeamList;
+  ScanbeamList m_Scanbeam;
+};
+//------------------------------------------------------------------------------
+
+class Clipper : public virtual ClipperBase {
+public:
+  Clipper(int initOptions = 0);
+
+  bool Execute(ClipType clipType, Paths &solution,
+               PolyFillType fillType = pftEvenOdd);
+
+  bool Execute(ClipType clipType, Paths &solution, PolyFillType subjFillType,
+               PolyFillType clipFillType);
+
+  bool Execute(ClipType clipType, PolyTree &polytree,
+               PolyFillType fillType = pftEvenOdd);
+
+  bool Execute(ClipType clipType, PolyTree &polytree, PolyFillType subjFillType,
+               PolyFillType clipFillType);
+
+  bool ReverseSolution() { return m_ReverseOutput; };
+
+  void ReverseSolution(bool value) { m_ReverseOutput = value; };
+
+  bool StrictlySimple() { return m_StrictSimple; };
+
+  void StrictlySimple(bool value) { m_StrictSimple = value; };
+// set the callback function for z value filling on intersections (otherwise Z
+// is 0)
+#ifdef use_xyz
+  void ZFillFunction(ZFillCallback zFillFunc);
+#endif
+protected:
+  virtual bool ExecuteInternal();
+
+private:
+  JoinList m_Joins;
+  JoinList m_GhostJoins;
+  IntersectList m_IntersectList;
+  ClipType m_ClipType;
+  typedef std::list<cInt> MaximaList;
+  MaximaList m_Maxima;
+  TEdge *m_SortedEdges;
+  bool m_ExecuteLocked;
+  PolyFillType m_ClipFillType;
+  PolyFillType m_SubjFillType;
+  bool m_ReverseOutput;
+  bool m_UsingPolyTree;
+  bool m_StrictSimple;
+#ifdef use_xyz
+  ZFillCallback m_ZFill; // custom callback
+#endif
+
+  void SetWindingCount(TEdge &edge);
+
+  bool IsEvenOddFillType(const TEdge &edge) const;
+
+  bool IsEvenOddAltFillType(const TEdge &edge) const;
+
+  void InsertLocalMinimaIntoAEL(const cInt botY);
+
+  void InsertEdgeIntoAEL(TEdge *edge, TEdge *startEdge);
+
+  void AddEdgeToSEL(TEdge *edge);
+
+  bool PopEdgeFromSEL(TEdge *&edge);
+
+  void CopyAELToSEL();
+
+  void DeleteFromSEL(TEdge *e);
+
+  void SwapPositionsInSEL(TEdge *edge1, TEdge *edge2);
+
+  bool IsContributing(const TEdge &edge) const;
+
+  bool IsTopHorz(const cInt XPos);
+
+  void DoMaxima(TEdge *e);
+
+  void ProcessHorizontals();
+
+  void ProcessHorizontal(TEdge *horzEdge);
+
+  void AddLocalMaxPoly(TEdge *e1, TEdge *e2, const IntPoint &pt);
+
+  OutPt *AddLocalMinPoly(TEdge *e1, TEdge *e2, const IntPoint &pt);
+
+  OutRec *GetOutRec(int idx);
+
+  void AppendPolygon(TEdge *e1, TEdge *e2);
+
+  void IntersectEdges(TEdge *e1, TEdge *e2, IntPoint &pt);
+
+  OutPt *AddOutPt(TEdge *e, const IntPoint &pt);
+
+  OutPt *GetLastOutPt(TEdge *e);
+
+  bool ProcessIntersections(const cInt topY);
+
+  void BuildIntersectList(const cInt topY);
+
+  void ProcessIntersectList();
+
+  void ProcessEdgesAtTopOfScanbeam(const cInt topY);
+
+  void BuildResult(Paths &polys);
+
+  void BuildResult2(PolyTree &polytree);
+
+  void SetHoleState(TEdge *e, OutRec *outrec);
+
+  void DisposeIntersectNodes();
+
+  bool FixupIntersectionOrder();
+
+  void FixupOutPolygon(OutRec &outrec);
+
+  void FixupOutPolyline(OutRec &outrec);
+
+  bool IsHole(TEdge *e);
+
+  bool FindOwnerFromSplitRecs(OutRec &outRec, OutRec *&currOrfl);
+
+  void FixHoleLinkage(OutRec &outrec);
+
+  void AddJoin(OutPt *op1, OutPt *op2, const IntPoint offPt);
+
+  void ClearJoins();
+
+  void ClearGhostJoins();
+
+  void AddGhostJoin(OutPt *op, const IntPoint offPt);
+
+  bool JoinPoints(Join *j, OutRec *outRec1, OutRec *outRec2);
+
+  void JoinCommonEdges();
+
+  void DoSimplePolygons();
+
+  void FixupFirstLefts1(OutRec *OldOutRec, OutRec *NewOutRec);
+
+  void FixupFirstLefts2(OutRec *InnerOutRec, OutRec *OuterOutRec);
+
+  void FixupFirstLefts3(OutRec *OldOutRec, OutRec *NewOutRec);
+
+#ifdef use_xyz
+  void SetZ(IntPoint &pt, TEdge &e1, TEdge &e2);
+#endif
+};
+//------------------------------------------------------------------------------
+
+class ClipperOffset {
+public:
+  ClipperOffset(double miterLimit = 2.0, double roundPrecision = 0.25);
+
+  ~ClipperOffset();
+
+  void AddPath(const Path &path, JoinType joinType, EndType endType);
+
+  void AddPaths(const Paths &paths, JoinType joinType, EndType endType);
+
+  void Execute(Paths &solution, double delta);
+
+  void Execute(PolyTree &solution, double delta);
+
+  void Clear();
+
+  double MiterLimit;
+  double ArcTolerance;
+
+private:
+  Paths m_destPolys;
+  Path m_srcPoly;
+  Path m_destPoly;
+  std::vector<DoublePoint> m_normals;
+  double m_delta, m_sinA, m_sin, m_cos;
+  double m_miterLim, m_StepsPerRad;
+  IntPoint m_lowest;
+  PolyNode m_polyNodes;
+
+  void FixOrientations();
+
+  void DoOffset(double delta);
+
+  void OffsetPoint(int j, int &k, JoinType jointype);
+
+  void DoSquare(int j, int k);
+
+  void DoMiter(int j, int k, double r);
+
+  void DoRound(int j, int k);
+};
+//------------------------------------------------------------------------------
+
+class clipperException : public std::exception {
+public:
+  clipperException(const char *description) : m_descr(description) {}
+
+  virtual ~clipperException() throw() {}
+
+  virtual const char *what() const throw() { return m_descr.c_str(); }
+
+private:
+  std::string m_descr;
+};
+//------------------------------------------------------------------------------
+
+} // namespace ClipperLib
+
+#endif // clipper_hpp
diff --git a/deploy/android_demo/app/src/main/cpp/ocr_cls_process.cpp b/deploy/android_demo/app/src/main/cpp/ocr_cls_process.cpp
new file mode 100644
index 0000000..c3434d9
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/ocr_cls_process.cpp
@@ -0,0 +1,45 @@
+// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "ocr_cls_process.h"
+#include <cmath>
+#include <cstring>
+#include <fstream>
+#include <iostream>
+#include <vector>
+
+const std::vector<int> CLS_IMAGE_SHAPE = {3, 48, 192};
+
+cv::Mat cls_resize_img(const cv::Mat &img) {
+  int imgC = CLS_IMAGE_SHAPE[0];
+  int imgW = CLS_IMAGE_SHAPE[2];
+  int imgH = CLS_IMAGE_SHAPE[1];
+
+  float ratio = float(img.cols) / float(img.rows);
+  int resize_w = 0;
+  if (ceilf(imgH * ratio) > imgW)
+    resize_w = imgW;
+  else
+    resize_w = int(ceilf(imgH * ratio));
+
+  cv::Mat resize_img;
+  cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f,
+             cv::INTER_CUBIC);
+
+  if (resize_w < imgW) {
+    cv::copyMakeBorder(resize_img, resize_img, 0, 0, 0, int(imgW - resize_w),
+                       cv::BORDER_CONSTANT, {0, 0, 0});
+  }
+  return resize_img;
+}
diff --git a/deploy/android_demo/app/src/main/cpp/ocr_cls_process.h b/deploy/android_demo/app/src/main/cpp/ocr_cls_process.h
new file mode 100644
index 0000000..17480a9
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/ocr_cls_process.h
@@ -0,0 +1,23 @@
+// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include "common.h"
+#include <opencv2/opencv.hpp>
+#include <vector>
+
+extern const std::vector<int> CLS_IMAGE_SHAPE;
+
+cv::Mat cls_resize_img(const cv::Mat &img);
diff --git a/deploy/android_demo/app/src/main/cpp/ocr_crnn_process.cpp b/deploy/android_demo/app/src/main/cpp/ocr_crnn_process.cpp
new file mode 100644
index 0000000..7e61a33
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/ocr_crnn_process.cpp
@@ -0,0 +1,141 @@
+// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "ocr_crnn_process.h"
+#include <cmath>
+#include <cstring>
+#include <fstream>
+#include <iostream>
+#include <vector>
+
+const std::string CHARACTER_TYPE = "ch";
+const int MAX_DICT_LENGTH = 6624;
+const std::vector<int> REC_IMAGE_SHAPE = {3, 32, 320};
+
+static cv::Mat crnn_resize_norm_img(cv::Mat img, float wh_ratio) {
+  int imgC = REC_IMAGE_SHAPE[0];
+  int imgW = REC_IMAGE_SHAPE[2];
+  int imgH = REC_IMAGE_SHAPE[1];
+
+  if (CHARACTER_TYPE == "ch")
+    imgW = int(32 * wh_ratio);
+
+  float ratio = float(img.cols) / float(img.rows);
+  int resize_w = 0;
+  if (ceilf(imgH * ratio) > imgW)
+    resize_w = imgW;
+  else
+    resize_w = int(ceilf(imgH * ratio));
+  cv::Mat resize_img;
+  cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f,
+             cv::INTER_CUBIC);
+
+  resize_img.convertTo(resize_img, CV_32FC3, 1 / 255.f);
+
+  for (int h = 0; h < resize_img.rows; h++) {
+    for (int w = 0; w < resize_img.cols; w++) {
+      resize_img.at<cv::Vec3f>(h, w)[0] =
+          (resize_img.at<cv::Vec3f>(h, w)[0] - 0.5) * 2;
+      resize_img.at<cv::Vec3f>(h, w)[1] =
+          (resize_img.at<cv::Vec3f>(h, w)[1] - 0.5) * 2;
+      resize_img.at<cv::Vec3f>(h, w)[2] =
+          (resize_img.at<cv::Vec3f>(h, w)[2] - 0.5) * 2;
+    }
+  }
+
+  cv::Mat dist;
+  cv::copyMakeBorder(resize_img, dist, 0, 0, 0, int(imgW - resize_w),
+                     cv::BORDER_CONSTANT, {0, 0, 0});
+
+  return dist;
+}
+
+cv::Mat crnn_resize_img(const cv::Mat &img, float wh_ratio) {
+  int imgC = REC_IMAGE_SHAPE[0];
+  int imgW = REC_IMAGE_SHAPE[2];
+  int imgH = REC_IMAGE_SHAPE[1];
+
+  if (CHARACTER_TYPE == "ch") {
+    imgW = int(32 * wh_ratio);
+  }
+
+  float ratio = float(img.cols) / float(img.rows);
+  int resize_w = 0;
+  if (ceilf(imgH * ratio) > imgW)
+    resize_w = imgW;
+  else
+    resize_w = int(ceilf(imgH * ratio));
+  cv::Mat resize_img;
+  cv::resize(img, resize_img, cv::Size(resize_w, imgH));
+  return resize_img;
+}
+
+cv::Mat get_rotate_crop_image(const cv::Mat &srcimage,
+                              const std::vector<std::vector<int>> &box) {
+
+  std::vector<std::vector<int>> points = box;
+
+  int x_collect[4] = {box[0][0], box[1][0], box[2][0], box[3][0]};
+  int y_collect[4] = {box[0][1], box[1][1], box[2][1], box[3][1]};
+  int left = int(*std::min_element(x_collect, x_collect + 4));
+  int right = int(*std::max_element(x_collect, x_collect + 4));
+  int top = int(*std::min_element(y_collect, y_collect + 4));
+  int bottom = int(*std::max_element(y_collect, y_collect + 4));
+
+  cv::Mat img_crop;
+  srcimage(cv::Rect(left, top, right - left, bottom - top)).copyTo(img_crop);
+
+  for (int i = 0; i < points.size(); i++) {
+    points[i][0] -= left;
+    points[i][1] -= top;
+  }
+
+  int img_crop_width = int(sqrt(pow(points[0][0] - points[1][0], 2) +
+                                pow(points[0][1] - points[1][1], 2)));
+  int img_crop_height = int(sqrt(pow(points[0][0] - points[3][0], 2) +
+                                 pow(points[0][1] - points[3][1], 2)));
+
+  cv::Point2f pts_std[4];
+  pts_std[0] = cv::Point2f(0., 0.);
+  pts_std[1] = cv::Point2f(img_crop_width, 0.);
+  pts_std[2] = cv::Point2f(img_crop_width, img_crop_height);
+  pts_std[3] = cv::Point2f(0.f, img_crop_height);
+
+  cv::Point2f pointsf[4];
+  pointsf[0] = cv::Point2f(points[0][0], points[0][1]);
+  pointsf[1] = cv::Point2f(points[1][0], points[1][1]);
+  pointsf[2] = cv::Point2f(points[2][0], points[2][1]);
+  pointsf[3] = cv::Point2f(points[3][0], points[3][1]);
+
+  cv::Mat M = cv::getPerspectiveTransform(pointsf, pts_std);
+
+  cv::Mat dst_img;
+  cv::warpPerspective(img_crop, dst_img, M,
+                      cv::Size(img_crop_width, img_crop_height),
+                      cv::BORDER_REPLICATE);
+
+  if (float(dst_img.rows) >= float(dst_img.cols) * 1.5) {
+    /*
+    cv::Mat srcCopy = cv::Mat(dst_img.rows, dst_img.cols, dst_img.depth());
+    cv::transpose(dst_img, srcCopy);
+    cv::flip(srcCopy, srcCopy, 0);
+    return srcCopy;
+    */
+    cv::transpose(dst_img, dst_img);
+    cv::flip(dst_img, dst_img, 0);
+    return dst_img;
+  } else {
+    return dst_img;
+  }
+}
diff --git a/deploy/android_demo/app/src/main/cpp/ocr_crnn_process.h b/deploy/android_demo/app/src/main/cpp/ocr_crnn_process.h
new file mode 100644
index 0000000..6ce8128
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/ocr_crnn_process.h
@@ -0,0 +1,20 @@
+//
+// Created by fujiayi on 2020/7/3.
+//
+#pragma once
+
+#include "common.h"
+#include <opencv2/opencv.hpp>
+#include <vector>
+
+extern const std::vector<int> REC_IMAGE_SHAPE;
+
+cv::Mat get_rotate_crop_image(const cv::Mat &srcimage,
+                              const std::vector<std::vector<int>> &box);
+
+cv::Mat crnn_resize_img(const cv::Mat &img, float wh_ratio);
+
+template <class ForwardIterator>
+inline size_t argmax(ForwardIterator first, ForwardIterator last) {
+  return std::distance(first, std::max_element(first, last));
+}
diff --git a/deploy/android_demo/app/src/main/cpp/ocr_db_post_process.cpp b/deploy/android_demo/app/src/main/cpp/ocr_db_post_process.cpp
new file mode 100644
index 0000000..ec731dd
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/ocr_db_post_process.cpp
@@ -0,0 +1,342 @@
+// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "ocr_clipper.hpp"
+#include "opencv2/core.hpp"
+#include "opencv2/imgcodecs.hpp"
+#include "opencv2/imgproc.hpp"
+#include <iostream>
+#include <math.h>
+#include <vector>
+
+static void getcontourarea(float **box, float unclip_ratio, float &distance) {
+  int pts_num = 4;
+  float area = 0.0f;
+  float dist = 0.0f;
+  for (int i = 0; i < pts_num; i++) {
+    area += box[i][0] * box[(i + 1) % pts_num][1] -
+            box[i][1] * box[(i + 1) % pts_num][0];
+    dist += sqrtf((box[i][0] - box[(i + 1) % pts_num][0]) *
+                      (box[i][0] - box[(i + 1) % pts_num][0]) +
+                  (box[i][1] - box[(i + 1) % pts_num][1]) *
+                      (box[i][1] - box[(i + 1) % pts_num][1]));
+  }
+  area = fabs(float(area / 2.0));
+
+  distance = area * unclip_ratio / dist;
+}
+
+static cv::RotatedRect unclip(float **box) {
+  float unclip_ratio = 2.0;
+  float distance = 1.0;
+
+  getcontourarea(box, unclip_ratio, distance);
+
+  ClipperLib::ClipperOffset offset;
+  ClipperLib::Path p;
+  p << ClipperLib::IntPoint(int(box[0][0]), int(box[0][1]))
+    << ClipperLib::IntPoint(int(box[1][0]), int(box[1][1]))
+    << ClipperLib::IntPoint(int(box[2][0]), int(box[2][1]))
+    << ClipperLib::IntPoint(int(box[3][0]), int(box[3][1]));
+  offset.AddPath(p, ClipperLib::jtRound, ClipperLib::etClosedPolygon);
+
+  ClipperLib::Paths soln;
+  offset.Execute(soln, distance);
+  std::vector<cv::Point2f> points;
+
+  for (int j = 0; j < soln.size(); j++) {
+    for (int i = 0; i < soln[soln.size() - 1].size(); i++) {
+      points.emplace_back(soln[j][i].X, soln[j][i].Y);
+    }
+  }
+  cv::RotatedRect res = cv::minAreaRect(points);
+
+  return res;
+}
+
+static float **Mat2Vec(cv::Mat mat) {
+  auto **array = new float *[mat.rows];
+  for (int i = 0; i < mat.rows; ++i) {
+    array[i] = new float[mat.cols];
+  }
+  for (int i = 0; i < mat.rows; ++i) {
+    for (int j = 0; j < mat.cols; ++j) {
+      array[i][j] = mat.at<float>(i, j);
+    }
+  }
+
+  return array;
+}
+
+static void quickSort(float **s, int l, int r) {
+  if (l < r) {
+    int i = l, j = r;
+    float x = s[l][0];
+    float *xp = s[l];
+    while (i < j) {
+      while (i < j && s[j][0] >= x) {
+        j--;
+      }
+      if (i < j) {
+        std::swap(s[i++], s[j]);
+      }
+      while (i < j && s[i][0] < x) {
+        i++;
+      }
+      if (i < j) {
+        std::swap(s[j--], s[i]);
+      }
+    }
+    s[i] = xp;
+    quickSort(s, l, i - 1);
+    quickSort(s, i + 1, r);
+  }
+}
+
+static void quickSort_vector(std::vector<std::vector<int>> &box, int l, int r,
+                             int axis) {
+  if (l < r) {
+    int i = l, j = r;
+    int x = box[l][axis];
+    std::vector<int> xp(box[l]);
+    while (i < j) {
+      while (i < j && box[j][axis] >= x) {
+        j--;
+      }
+      if (i < j) {
+        std::swap(box[i++], box[j]);
+      }
+      while (i < j && box[i][axis] < x) {
+        i++;
+      }
+      if (i < j) {
+        std::swap(box[j--], box[i]);
+      }
+    }
+    box[i] = xp;
+    quickSort_vector(box, l, i - 1, axis);
+    quickSort_vector(box, i + 1, r, axis);
+  }
+}
+
+static std::vector<std::vector<int>>
+order_points_clockwise(std::vector<std::vector<int>> pts) {
+  std::vector<std::vector<int>> box = pts;
+  quickSort_vector(box, 0, int(box.size() - 1), 0);
+  std::vector<std::vector<int>> leftmost = {box[0], box[1]};
+  std::vector<std::vector<int>> rightmost = {box[2], box[3]};
+
+  if (leftmost[0][1] > leftmost[1][1]) {
+    std::swap(leftmost[0], leftmost[1]);
+  }
+
+  if (rightmost[0][1] > rightmost[1][1]) {
+    std::swap(rightmost[0], rightmost[1]);
+  }
+
+  std::vector<std::vector<int>> rect = {leftmost[0], rightmost[0], rightmost[1],
+                                        leftmost[1]};
+  return rect;
+}
+
+static float **get_mini_boxes(cv::RotatedRect box, float &ssid) {
+  ssid = box.size.width >= box.size.height ? box.size.height : box.size.width;
+
+  cv::Mat points;
+  cv::boxPoints(box, points);
+  // sorted box points
+  auto array = Mat2Vec(points);
+  quickSort(array, 0, 3);
+
+  float *idx1 = array[0], *idx2 = array[1], *idx3 = array[2], *idx4 = array[3];
+  if (array[3][1] <= array[2][1]) {
+    idx2 = array[3];
+    idx3 = array[2];
+  } else {
+    idx2 = array[2];
+    idx3 = array[3];
+  }
+  if (array[1][1] <= array[0][1]) {
+    idx1 = array[1];
+    idx4 = array[0];
+  } else {
+    idx1 = array[0];
+    idx4 = array[1];
+  }
+
+  array[0] = idx1;
+  array[1] = idx2;
+  array[2] = idx3;
+  array[3] = idx4;
+
+  return array;
+}
+
+template <class T> T clamp(T x, T min, T max) {
+  if (x > max) {
+    return max;
+  }
+  if (x < min) {
+    return min;
+  }
+  return x;
+}
+
+static float clampf(float x, float min, float max) {
+  if (x > max)
+    return max;
+  if (x < min)
+    return min;
+  return x;
+}
+
+float box_score_fast(float **box_array, cv::Mat pred) {
+  auto array = box_array;
+  int width = pred.cols;
+  int height = pred.rows;
+
+  float box_x[4] = {array[0][0], array[1][0], array[2][0], array[3][0]};
+  float box_y[4] = {array[0][1], array[1][1], array[2][1], array[3][1]};
+
+  int xmin = clamp(int(std::floorf(*(std::min_element(box_x, box_x + 4)))), 0,
+                   width - 1);
+  int xmax = clamp(int(std::ceilf(*(std::max_element(box_x, box_x + 4)))), 0,
+                   width - 1);
+  int ymin = clamp(int(std::floorf(*(std::min_element(box_y, box_y + 4)))), 0,
+                   height - 1);
+  int ymax = clamp(int(std::ceilf(*(std::max_element(box_y, box_y + 4)))), 0,
+                   height - 1);
+
+  cv::Mat mask;
+  mask = cv::Mat::zeros(ymax - ymin + 1, xmax - xmin + 1, CV_8UC1);
+
+  cv::Point root_point[4];
+  root_point[0] = cv::Point(int(array[0][0]) - xmin, int(array[0][1]) - ymin);
+  root_point[1] = cv::Point(int(array[1][0]) - xmin, int(array[1][1]) - ymin);
+  root_point[2] = cv::Point(int(array[2][0]) - xmin, int(array[2][1]) - ymin);
+  root_point[3] = cv::Point(int(array[3][0]) - xmin, int(array[3][1]) - ymin);
+  const cv::Point *ppt[1] = {root_point};
+  int npt[] = {4};
+  cv::fillPoly(mask, ppt, npt, 1, cv::Scalar(1));
+
+  cv::Mat croppedImg;
+  pred(cv::Rect(xmin, ymin, xmax - xmin + 1, ymax - ymin + 1))
+      .copyTo(croppedImg);
+
+  auto score = cv::mean(croppedImg, mask)[0];
+  return score;
+}
+
+std::vector<std::vector<std::vector<int>>>
+boxes_from_bitmap(const cv::Mat &pred, const cv::Mat &bitmap) {
+  const int min_size = 3;
+  const int max_candidates = 1000;
+  const float box_thresh = 0.5;
+
+  int width = bitmap.cols;
+  int height = bitmap.rows;
+
+  std::vector<std::vector<cv::Point>> contours;
+  std::vector<cv::Vec4i> hierarchy;
+
+  cv::findContours(bitmap, contours, hierarchy, cv::RETR_LIST,
+                   cv::CHAIN_APPROX_SIMPLE);
+
+  int num_contours =
+      contours.size() >= max_candidates ? max_candidates : contours.size();
+
+  std::vector<std::vector<std::vector<int>>> boxes;
+
+  for (int _i = 0; _i < num_contours; _i++) {
+    float ssid;
+    cv::RotatedRect box = cv::minAreaRect(contours[_i]);
+    auto array = get_mini_boxes(box, ssid);
+
+    auto box_for_unclip = array;
+    // end get_mini_box
+
+    if (ssid < min_size) {
+      continue;
+    }
+
+    float score;
+    score = box_score_fast(array, pred);
+    // end box_score_fast
+    if (score < box_thresh) {
+      continue;
+    }
+
+    // start for unclip
+    cv::RotatedRect points = unclip(box_for_unclip);
+    // end for unclip
+
+    cv::RotatedRect clipbox = points;
+    auto cliparray = get_mini_boxes(clipbox, ssid);
+
+    if (ssid < min_size + 2)
+      continue;
+
+    int dest_width = pred.cols;
+    int dest_height = pred.rows;
+    std::vector<std::vector<int>> intcliparray;
+
+    for (int num_pt = 0; num_pt < 4; num_pt++) {
+      std::vector<int> a{int(clampf(roundf(cliparray[num_pt][0] / float(width) *
+                                           float(dest_width)),
+                                    0, float(dest_width))),
+                         int(clampf(roundf(cliparray[num_pt][1] /
+                                           float(height) * float(dest_height)),
+                                    0, float(dest_height)))};
+      intcliparray.emplace_back(std::move(a));
+    }
+    boxes.emplace_back(std::move(intcliparray));
+
+  } // end for
+  return boxes;
+}
+
+int _max(int a, int b) { return a >= b ? a : b; }
+
+int _min(int a, int b) { return a >= b ? b : a; }
+
+std::vector<std::vector<std::vector<int>>>
+filter_tag_det_res(const std::vector<std::vector<std::vector<int>>> &o_boxes,
+                   float ratio_h, float ratio_w, const cv::Mat &srcimg) {
+  int oriimg_h = srcimg.rows;
+  int oriimg_w = srcimg.cols;
+  std::vector<std::vector<std::vector<int>>> boxes{o_boxes};
+  std::vector<std::vector<std::vector<int>>> root_points;
+  for (int n = 0; n < boxes.size(); n++) {
+    boxes[n] = order_points_clockwise(boxes[n]);
+    for (int m = 0; m < boxes[0].size(); m++) {
+      boxes[n][m][0] /= ratio_w;
+      boxes[n][m][1] /= ratio_h;
+
+      boxes[n][m][0] = int(_min(_max(boxes[n][m][0], 0), oriimg_w - 1));
+      boxes[n][m][1] = int(_min(_max(boxes[n][m][1], 0), oriimg_h - 1));
+    }
+  }
+
+  for (int n = 0; n < boxes.size(); n++) {
+    int rect_width, rect_height;
+    rect_width = int(sqrt(pow(boxes[n][0][0] - boxes[n][1][0], 2) +
+                          pow(boxes[n][0][1] - boxes[n][1][1], 2)));
+    rect_height = int(sqrt(pow(boxes[n][0][0] - boxes[n][3][0], 2) +
+                           pow(boxes[n][0][1] - boxes[n][3][1], 2)));
+    if (rect_width <= 10 || rect_height <= 10)
+      continue;
+    root_points.push_back(boxes[n]);
+  }
+  return root_points;
+}
diff --git a/deploy/android_demo/app/src/main/cpp/ocr_db_post_process.h b/deploy/android_demo/app/src/main/cpp/ocr_db_post_process.h
new file mode 100644
index 0000000..6d84762
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/ocr_db_post_process.h
@@ -0,0 +1,13 @@
+//
+// Created by fujiayi on 2020/7/2.
+//
+#pragma once
+#include <opencv2/opencv.hpp>
+#include <vector>
+
+std::vector<std::vector<std::vector<int>>>
+boxes_from_bitmap(const cv::Mat &pred, const cv::Mat &bitmap);
+
+std::vector<std::vector<std::vector<int>>>
+filter_tag_det_res(const std::vector<std::vector<std::vector<int>>> &o_boxes,
+                   float ratio_h, float ratio_w, const cv::Mat &srcimg);
diff --git a/deploy/android_demo/app/src/main/cpp/ocr_ppredictor.cpp b/deploy/android_demo/app/src/main/cpp/ocr_ppredictor.cpp
new file mode 100644
index 0000000..95b86a2
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/ocr_ppredictor.cpp
@@ -0,0 +1,354 @@
+//
+// Created by fujiayi on 2020/7/1.
+//
+
+#include "ocr_ppredictor.h"
+#include "common.h"
+#include "ocr_cls_process.h"
+#include "ocr_crnn_process.h"
+#include "ocr_db_post_process.h"
+#include "preprocess.h"
+
+namespace ppredictor {
+
+OCR_PPredictor::OCR_PPredictor(const OCR_Config &config) : _config(config) {}
+
+int OCR_PPredictor::init(const std::string &det_model_content,
+                         const std::string &rec_model_content,
+                         const std::string &cls_model_content) {
+  _det_predictor = std::unique_ptr<PPredictor>(new PPredictor{
+      _config.use_opencl, _config.thread_num, NET_OCR, _config.mode});
+  _det_predictor->init_nb(det_model_content);
+
+  _rec_predictor = std::unique_ptr<PPredictor>(new PPredictor{
+      _config.use_opencl, _config.thread_num, NET_OCR_INTERNAL, _config.mode});
+  _rec_predictor->init_nb(rec_model_content);
+
+  _cls_predictor = std::unique_ptr<PPredictor>(new PPredictor{
+      _config.use_opencl, _config.thread_num, NET_OCR_INTERNAL, _config.mode});
+  _cls_predictor->init_nb(cls_model_content);
+  return RETURN_OK;
+}
+
+int OCR_PPredictor::init_from_file(const std::string &det_model_path,
+                                   const std::string &rec_model_path,
+                                   const std::string &cls_model_path) {
+  _det_predictor = std::unique_ptr<PPredictor>(new PPredictor{
+      _config.use_opencl, _config.thread_num, NET_OCR, _config.mode});
+  _det_predictor->init_from_file(det_model_path);
+
+  _rec_predictor = std::unique_ptr<PPredictor>(new PPredictor{
+      _config.use_opencl, _config.thread_num, NET_OCR_INTERNAL, _config.mode});
+  _rec_predictor->init_from_file(rec_model_path);
+
+  _cls_predictor = std::unique_ptr<PPredictor>(new PPredictor{
+      _config.use_opencl, _config.thread_num, NET_OCR_INTERNAL, _config.mode});
+  _cls_predictor->init_from_file(cls_model_path);
+  return RETURN_OK;
+}
+/**
+ * for debug use, show result of First Step
+ * @param filter_boxes
+ * @param boxes
+ * @param srcimg
+ */
+static void
+visual_img(const std::vector<std::vector<std::vector<int>>> &filter_boxes,
+           const std::vector<std::vector<std::vector<int>>> &boxes,
+           const cv::Mat &srcimg) {
+  // visualization
+  cv::Point rook_points[filter_boxes.size()][4];
+  for (int n = 0; n < filter_boxes.size(); n++) {
+    for (int m = 0; m < filter_boxes[0].size(); m++) {
+      rook_points[n][m] =
+          cv::Point(int(filter_boxes[n][m][0]), int(filter_boxes[n][m][1]));
+    }
+  }
+
+  cv::Mat img_vis;
+  srcimg.copyTo(img_vis);
+  for (int n = 0; n < boxes.size(); n++) {
+    const cv::Point *ppt[1] = {rook_points[n]};
+    int npt[] = {4};
+    cv::polylines(img_vis, ppt, npt, 1, 1, CV_RGB(0, 255, 0), 2, 8, 0);
+  }
+  // 调试用，自行替换需要修改的路径
+  cv::imwrite("/sdcard/1/vis.png", img_vis);
+}
+
+std::vector<OCRPredictResult>
+OCR_PPredictor::infer_ocr(cv::Mat &origin, int max_size_len, int run_det,
+                          int run_cls, int run_rec) {
+  LOGI("ocr cpp start *****************");
+  LOGI("ocr cpp det: %d, cls: %d, rec: %d", run_det, run_cls, run_rec);
+  std::vector<OCRPredictResult> ocr_results;
+  if (run_det) {
+    infer_det(origin, max_size_len, ocr_results);
+  }
+  if (run_rec) {
+    if (ocr_results.size() == 0) {
+      OCRPredictResult res;
+      ocr_results.emplace_back(std::move(res));
+    }
+    for (int i = 0; i < ocr_results.size(); i++) {
+      infer_rec(origin, run_cls, ocr_results[i]);
+    }
+  } else if (run_cls) {
+    ClsPredictResult cls_res = infer_cls(origin);
+    OCRPredictResult res;
+    res.cls_score = cls_res.cls_score;
+    res.cls_label = cls_res.cls_label;
+    ocr_results.push_back(res);
+  }
+
+  LOGI("ocr cpp end *****************");
+  return ocr_results;
+}
+
+cv::Mat DetResizeImg(const cv::Mat img, int max_size_len,
+                     std::vector<float> &ratio_hw) {
+  int w = img.cols;
+  int h = img.rows;
+
+  float ratio = 1.f;
+  int max_wh = w >= h ? w : h;
+  if (max_wh > max_size_len) {
+    if (h > w) {
+      ratio = static_cast<float>(max_size_len) / static_cast<float>(h);
+    } else {
+      ratio = static_cast<float>(max_size_len) / static_cast<float>(w);
+    }
+  }
+
+  int resize_h = static_cast<int>(float(h) * ratio);
+  int resize_w = static_cast<int>(float(w) * ratio);
+  if (resize_h % 32 == 0)
+    resize_h = resize_h;
+  else if (resize_h / 32 < 1 + 1e-5)
+    resize_h = 32;
+  else
+    resize_h = (resize_h / 32 - 1) * 32;
+
+  if (resize_w % 32 == 0)
+    resize_w = resize_w;
+  else if (resize_w / 32 < 1 + 1e-5)
+    resize_w = 32;
+  else
+    resize_w = (resize_w / 32 - 1) * 32;
+
+  cv::Mat resize_img;
+  cv::resize(img, resize_img, cv::Size(resize_w, resize_h));
+
+  ratio_hw.push_back(static_cast<float>(resize_h) / static_cast<float>(h));
+  ratio_hw.push_back(static_cast<float>(resize_w) / static_cast<float>(w));
+  return resize_img;
+}
+
+void OCR_PPredictor::infer_det(cv::Mat &origin, int max_size_len,
+                               std::vector<OCRPredictResult> &ocr_results) {
+  std::vector<float> mean = {0.485f, 0.456f, 0.406f};
+  std::vector<float> scale = {1 / 0.229f, 1 / 0.224f, 1 / 0.225f};
+
+  PredictorInput input = _det_predictor->get_first_input();
+
+  std::vector<float> ratio_hw;
+  cv::Mat input_image = DetResizeImg(origin, max_size_len, ratio_hw);
+  input_image.convertTo(input_image, CV_32FC3, 1 / 255.0f);
+  const float *dimg = reinterpret_cast<const float *>(input_image.data);
+  int input_size = input_image.rows * input_image.cols;
+
+  input.set_dims({1, 3, input_image.rows, input_image.cols});
+
+  neon_mean_scale(dimg, input.get_mutable_float_data(), input_size, mean,
+                  scale);
+  LOGI("ocr cpp det shape %d,%d", input_image.rows, input_image.cols);
+  std::vector<PredictorOutput> results = _det_predictor->infer();
+  PredictorOutput &res = results.at(0);
+  std::vector<std::vector<std::vector<int>>> filtered_box =
+      calc_filtered_boxes(res.get_float_data(), res.get_size(),
+                          input_image.rows, input_image.cols, origin);
+  LOGI("ocr cpp det Filter_box size %ld", filtered_box.size());
+
+  for (int i = 0; i < filtered_box.size(); i++) {
+    LOGI("ocr cpp box  %d,%d,%d,%d,%d,%d,%d,%d", filtered_box[i][0][0],
+         filtered_box[i][0][1], filtered_box[i][1][0], filtered_box[i][1][1],
+         filtered_box[i][2][0], filtered_box[i][2][1], filtered_box[i][3][0],
+         filtered_box[i][3][1]);
+    OCRPredictResult res;
+    res.points = filtered_box[i];
+    ocr_results.push_back(res);
+  }
+}
+
+void OCR_PPredictor::infer_rec(const cv::Mat &origin_img, int run_cls,
+                               OCRPredictResult &ocr_result) {
+  std::vector<float> mean = {0.5f, 0.5f, 0.5f};
+  std::vector<float> scale = {1 / 0.5f, 1 / 0.5f, 1 / 0.5f};
+  std::vector<int64_t> dims = {1, 3, 0, 0};
+
+  PredictorInput input = _rec_predictor->get_first_input();
+
+  const std::vector<std::vector<int>> &box = ocr_result.points;
+  cv::Mat crop_img;
+  if (box.size() > 0) {
+    crop_img = get_rotate_crop_image(origin_img, box);
+  } else {
+    crop_img = origin_img;
+  }
+
+  if (run_cls) {
+    ClsPredictResult cls_res = infer_cls(crop_img);
+    crop_img = cls_res.img;
+    ocr_result.cls_score = cls_res.cls_score;
+    ocr_result.cls_label = cls_res.cls_label;
+  }
+
+  float wh_ratio = float(crop_img.cols) / float(crop_img.rows);
+  cv::Mat input_image = crnn_resize_img(crop_img, wh_ratio);
+  input_image.convertTo(input_image, CV_32FC3, 1 / 255.0f);
+  const float *dimg = reinterpret_cast<const float *>(input_image.data);
+  int input_size = input_image.rows * input_image.cols;
+
+  dims[2] = input_image.rows;
+  dims[3] = input_image.cols;
+  input.set_dims(dims);
+
+  neon_mean_scale(dimg, input.get_mutable_float_data(), input_size, mean,
+                  scale);
+
+  std::vector<PredictorOutput> results = _rec_predictor->infer();
+  const float *predict_batch = results.at(0).get_float_data();
+  const std::vector<int64_t> predict_shape = results.at(0).get_shape();
+
+  // ctc decode
+  int argmax_idx;
+  int last_index = 0;
+  float score = 0.f;
+  int count = 0;
+  float max_value = 0.0f;
+
+  for (int n = 0; n < predict_shape[1]; n++) {
+    argmax_idx = int(argmax(&predict_batch[n * predict_shape[2]],
+                            &predict_batch[(n + 1) * predict_shape[2]]));
+    max_value =
+        float(*std::max_element(&predict_batch[n * predict_shape[2]],
+                                &predict_batch[(n + 1) * predict_shape[2]]));
+    if (argmax_idx > 0 && (!(n > 0 && argmax_idx == last_index))) {
+      score += max_value;
+      count += 1;
+      ocr_result.word_index.push_back(argmax_idx);
+    }
+    last_index = argmax_idx;
+  }
+  score /= count;
+  ocr_result.score = score;
+  LOGI("ocr cpp rec word size %ld", count);
+}
+
+ClsPredictResult OCR_PPredictor::infer_cls(const cv::Mat &img, float thresh) {
+  std::vector<float> mean = {0.5f, 0.5f, 0.5f};
+  std::vector<float> scale = {1 / 0.5f, 1 / 0.5f, 1 / 0.5f};
+  std::vector<int64_t> dims = {1, 3, 0, 0};
+
+  PredictorInput input = _cls_predictor->get_first_input();
+
+  cv::Mat input_image = cls_resize_img(img);
+  input_image.convertTo(input_image, CV_32FC3, 1 / 255.0f);
+  const float *dimg = reinterpret_cast<const float *>(input_image.data);
+  int input_size = input_image.rows * input_image.cols;
+
+  dims[2] = input_image.rows;
+  dims[3] = input_image.cols;
+  input.set_dims(dims);
+
+  neon_mean_scale(dimg, input.get_mutable_float_data(), input_size, mean,
+                  scale);
+
+  std::vector<PredictorOutput> results = _cls_predictor->infer();
+
+  const float *scores = results.at(0).get_float_data();
+  float score = 0;
+  int label = 0;
+  for (int64_t i = 0; i < results.at(0).get_size(); i++) {
+    LOGI("ocr cpp cls output scores [%f]", scores[i]);
+    if (scores[i] > score) {
+      score = scores[i];
+      label = i;
+    }
+  }
+  cv::Mat srcimg;
+  img.copyTo(srcimg);
+  if (label % 2 == 1 && score > thresh) {
+    cv::rotate(srcimg, srcimg, 1);
+  }
+  ClsPredictResult res;
+  res.cls_label = label;
+  res.cls_score = score;
+  res.img = srcimg;
+  LOGI("ocr cpp cls word cls %ld, %f", label, score);
+  return res;
+}
+
+std::vector<std::vector<std::vector<int>>>
+OCR_PPredictor::calc_filtered_boxes(const float *pred, int pred_size,
+                                    int output_height, int output_width,
+                                    const cv::Mat &origin) {
+  const double threshold = 0.3;
+  const double maxvalue = 1;
+
+  cv::Mat pred_map = cv::Mat::zeros(output_height, output_width, CV_32F);
+  memcpy(pred_map.data, pred, pred_size * sizeof(float));
+  cv::Mat cbuf_map;
+  pred_map.convertTo(cbuf_map, CV_8UC1);
+
+  cv::Mat bit_map;
+  cv::threshold(cbuf_map, bit_map, threshold, maxvalue, cv::THRESH_BINARY);
+
+  std::vector<std::vector<std::vector<int>>> boxes =
+      boxes_from_bitmap(pred_map, bit_map);
+  float ratio_h = output_height * 1.0f / origin.rows;
+  float ratio_w = output_width * 1.0f / origin.cols;
+  std::vector<std::vector<std::vector<int>>> filter_boxes =
+      filter_tag_det_res(boxes, ratio_h, ratio_w, origin);
+  return filter_boxes;
+}
+
+std::vector<int>
+OCR_PPredictor::postprocess_rec_word_index(const PredictorOutput &res) {
+  const int *rec_idx = res.get_int_data();
+  const std::vector<std::vector<uint64_t>> rec_idx_lod = res.get_lod();
+
+  std::vector<int> pred_idx;
+  for (int n = int(rec_idx_lod[0][0]); n < int(rec_idx_lod[0][1] * 2); n += 2) {
+    pred_idx.emplace_back(rec_idx[n]);
+  }
+  return pred_idx;
+}
+
+float OCR_PPredictor::postprocess_rec_score(const PredictorOutput &res) {
+  const float *predict_batch = res.get_float_data();
+  const std::vector<int64_t> predict_shape = res.get_shape();
+  const std::vector<std::vector<uint64_t>> predict_lod = res.get_lod();
+  int blank = predict_shape[1];
+  float score = 0.f;
+  int count = 0;
+  for (int n = predict_lod[0][0]; n < predict_lod[0][1] - 1; n++) {
+    int argmax_idx = argmax(predict_batch + n * predict_shape[1],
+                            predict_batch + (n + 1) * predict_shape[1]);
+    float max_value = predict_batch[n * predict_shape[1] + argmax_idx];
+    if (blank - 1 - argmax_idx > 1e-5) {
+      score += max_value;
+      count += 1;
+    }
+  }
+  if (count == 0) {
+    LOGE("calc score count 0");
+  } else {
+    score /= count;
+  }
+  LOGI("calc score: %f", score);
+  return score;
+}
+
+NET_TYPE OCR_PPredictor::get_net_flag() const { return NET_OCR; }
+} // namespace ppredictor
diff --git a/deploy/android_demo/app/src/main/cpp/ocr_ppredictor.h b/deploy/android_demo/app/src/main/cpp/ocr_ppredictor.h
new file mode 100644
index 0000000..1567684
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/ocr_ppredictor.h
@@ -0,0 +1,131 @@
+//
+// Created by fujiayi on 2020/7/1.
+//
+
+#pragma once
+
+#include "ppredictor.h"
+#include <opencv2/opencv.hpp>
+#include <paddle_api.h>
+#include <string>
+
+namespace ppredictor {
+
+/**
+ * Config
+ */
+struct OCR_Config {
+  int use_opencl = 0;
+  int thread_num = 4; // Thread num
+  paddle::lite_api::PowerMode mode =
+      paddle::lite_api::LITE_POWER_HIGH; // PaddleLite Mode
+};
+
+/**
+ * Polygons Result
+ */
+struct OCRPredictResult {
+  std::vector<int> word_index;
+  std::vector<std::vector<int>> points;
+  float score;
+  float cls_score;
+  int cls_label = -1;
+};
+
+struct ClsPredictResult {
+  float cls_score;
+  int cls_label = -1;
+  cv::Mat img;
+};
+/**
+ * OCR there are 2 models
+ * 1. First model（det），select polygons to show where are the texts
+ * 2. crop from the origin images, use these polygons to infer
+ */
+class OCR_PPredictor : public PPredictor_Interface {
+public:
+  OCR_PPredictor(const OCR_Config &config);
+
+  virtual ~OCR_PPredictor() {}
+
+  /**
+   * 初始化二个模型的Predictor
+   * @param det_model_content
+   * @param rec_model_content
+   * @return
+   */
+  int init(const std::string &det_model_content,
+           const std::string &rec_model_content,
+           const std::string &cls_model_content);
+  int init_from_file(const std::string &det_model_path,
+                     const std::string &rec_model_path,
+                     const std::string &cls_model_path);
+  /**
+   * Return OCR result
+   * @param dims
+   * @param input_data
+   * @param input_len
+   * @param net_flag
+   * @param origin
+   * @return
+   */
+  virtual std::vector<OCRPredictResult> infer_ocr(cv::Mat &origin,
+                                                  int max_size_len, int run_det,
+                                                  int run_cls, int run_rec);
+
+  virtual NET_TYPE get_net_flag() const;
+
+private:
+  /**
+   * calculate polygons from the result image of first model
+   * @param pred
+   * @param output_height
+   * @param output_width
+   * @param origin
+   * @return
+   */
+  std::vector<std::vector<std::vector<int>>>
+  calc_filtered_boxes(const float *pred, int pred_size, int output_height,
+                      int output_width, const cv::Mat &origin);
+
+  void infer_det(cv::Mat &origin, int max_side_len,
+                 std::vector<OCRPredictResult> &ocr_results);
+  /**
+   * infer for rec model
+   *
+   * @param boxes
+   * @param origin
+   * @return
+   */
+  void infer_rec(const cv::Mat &origin, int run_cls,
+                 OCRPredictResult &ocr_result);
+
+  /**
+   * infer for cls model
+   *
+   * @param boxes
+   * @param origin
+   * @return
+   */
+  ClsPredictResult infer_cls(const cv::Mat &origin, float thresh = 0.9);
+
+  /**
+   * Postprocess or second model to extract text
+   * @param res
+   * @return
+   */
+  std::vector<int> postprocess_rec_word_index(const PredictorOutput &res);
+
+  /**
+   * calculate confidence of second model text result
+   * @param res
+   * @return
+   */
+  float postprocess_rec_score(const PredictorOutput &res);
+
+  std::unique_ptr<PPredictor> _det_predictor;
+  std::unique_ptr<PPredictor> _rec_predictor;
+  std::unique_ptr<PPredictor> _cls_predictor;
+  OCR_Config _config;
+};
+} // namespace ppredictor
diff --git a/deploy/android_demo/app/src/main/cpp/ppredictor.cpp b/deploy/android_demo/app/src/main/cpp/ppredictor.cpp
new file mode 100644
index 0000000..10b8d7e
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/ppredictor.cpp
@@ -0,0 +1,99 @@
+#include "ppredictor.h"
+#include "common.h"
+
+namespace ppredictor {
+PPredictor::PPredictor(int use_opencl, int thread_num, int net_flag,
+                       paddle::lite_api::PowerMode mode)
+    : _use_opencl(use_opencl), _thread_num(thread_num), _net_flag(net_flag),
+      _mode(mode) {}
+
+int PPredictor::init_nb(const std::string &model_content) {
+  paddle::lite_api::MobileConfig config;
+  config.set_model_from_buffer(model_content);
+  return _init(config);
+}
+
+int PPredictor::init_from_file(const std::string &model_content) {
+  paddle::lite_api::MobileConfig config;
+  config.set_model_from_file(model_content);
+  return _init(config);
+}
+
+template <typename ConfigT> int PPredictor::_init(ConfigT &config) {
+  bool is_opencl_backend_valid =
+      paddle::lite_api::IsOpenCLBackendValid(/*check_fp16_valid = false*/);
+  if (is_opencl_backend_valid) {
+    if (_use_opencl != 0) {
+      // Make sure you have write permission of the binary path.
+      // We strongly recommend each model has a unique binary name.
+      const std::string bin_path = "/data/local/tmp/";
+      const std::string bin_name = "lite_opencl_kernel.bin";
+      config.set_opencl_binary_path_name(bin_path, bin_name);
+
+      // opencl tune option
+      // CL_TUNE_NONE: 0
+      // CL_TUNE_RAPID: 1
+      // CL_TUNE_NORMAL: 2
+      // CL_TUNE_EXHAUSTIVE: 3
+      const std::string tuned_path = "/data/local/tmp/";
+      const std::string tuned_name = "lite_opencl_tuned.bin";
+      config.set_opencl_tune(paddle::lite_api::CL_TUNE_NORMAL, tuned_path,
+                             tuned_name);
+
+      // opencl precision option
+      // CL_PRECISION_AUTO: 0, first fp16 if valid, default
+      // CL_PRECISION_FP32: 1, force fp32
+      // CL_PRECISION_FP16: 2, force fp16
+      config.set_opencl_precision(paddle::lite_api::CL_PRECISION_FP32);
+      LOGI("ocr cpp device: running on gpu.");
+    }
+  } else {
+    LOGI("ocr cpp device: running on cpu.");
+    // you can give backup cpu nb model instead
+    // config.set_model_from_file(cpu_nb_model_dir);
+  }
+  config.set_threads(_thread_num);
+  config.set_power_mode(_mode);
+  _predictor = paddle::lite_api::CreatePaddlePredictor(config);
+  LOGI("ocr cpp paddle instance created");
+  return RETURN_OK;
+}
+
+PredictorInput PPredictor::get_input(int index) {
+  PredictorInput input{_predictor->GetInput(index), index, _net_flag};
+  _is_input_get = true;
+  return input;
+}
+
+std::vector<PredictorInput> PPredictor::get_inputs(int num) {
+  std::vector<PredictorInput> results;
+  for (int i = 0; i < num; i++) {
+    results.emplace_back(get_input(i));
+  }
+  return results;
+}
+
+PredictorInput PPredictor::get_first_input() { return get_input(0); }
+
+std::vector<PredictorOutput> PPredictor::infer() {
+  LOGI("ocr cpp infer Run start %d", _net_flag);
+  std::vector<PredictorOutput> results;
+  if (!_is_input_get) {
+    return results;
+  }
+  _predictor->Run();
+  LOGI("ocr cpp infer Run end");
+
+  for (int i = 0; i < _predictor->GetOutputNames().size(); i++) {
+    std::unique_ptr<const paddle::lite_api::Tensor> output_tensor =
+        _predictor->GetOutput(i);
+    LOGI("ocr cpp output tensor[%d] size %ld", i,
+         product(output_tensor->shape()));
+    PredictorOutput result{std::move(output_tensor), i, _net_flag};
+    results.emplace_back(std::move(result));
+  }
+  return results;
+}
+
+NET_TYPE PPredictor::get_net_flag() const { return (NET_TYPE)_net_flag; }
+} // namespace ppredictor
diff --git a/deploy/android_demo/app/src/main/cpp/ppredictor.h b/deploy/android_demo/app/src/main/cpp/ppredictor.h
new file mode 100644
index 0000000..230a84d
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/ppredictor.h
@@ -0,0 +1,64 @@
+#pragma once
+
+#include "paddle_api.h"
+#include "predictor_input.h"
+#include "predictor_output.h"
+
+namespace ppredictor {
+
+/**
+ * PaddleLite Preditor Common Interface
+ */
+class PPredictor_Interface {
+public:
+  virtual ~PPredictor_Interface() {}
+
+  virtual NET_TYPE get_net_flag() const = 0;
+};
+
+/**
+ * Common Predictor
+ */
+class PPredictor : public PPredictor_Interface {
+public:
+  PPredictor(
+      int use_opencl, int thread_num, int net_flag = 0,
+      paddle::lite_api::PowerMode mode = paddle::lite_api::LITE_POWER_HIGH);
+
+  virtual ~PPredictor() {}
+
+  /**
+   * init paddlitelite opt model，nb format ，or use ini_paddle
+   * @param model_content
+   * @return 0
+   */
+  virtual int init_nb(const std::string &model_content);
+
+  virtual int init_from_file(const std::string &model_content);
+
+  std::vector<PredictorOutput> infer();
+
+  std::shared_ptr<paddle::lite_api::PaddlePredictor> get_predictor() {
+    return _predictor;
+  }
+
+  virtual std::vector<PredictorInput> get_inputs(int num);
+
+  virtual PredictorInput get_input(int index);
+
+  virtual PredictorInput get_first_input();
+
+  virtual NET_TYPE get_net_flag() const;
+
+protected:
+  template <typename ConfigT> int _init(ConfigT &config);
+
+private:
+  int _use_opencl;
+  int _thread_num;
+  paddle::lite_api::PowerMode _mode;
+  std::shared_ptr<paddle::lite_api::PaddlePredictor> _predictor;
+  bool _is_input_get = false;
+  int _net_flag;
+};
+} // namespace ppredictor
diff --git a/deploy/android_demo/app/src/main/cpp/predictor_input.cpp b/deploy/android_demo/app/src/main/cpp/predictor_input.cpp
new file mode 100644
index 0000000..2d5666f
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/predictor_input.cpp
@@ -0,0 +1,28 @@
+#include "predictor_input.h"
+
+namespace ppredictor {
+
+void PredictorInput::set_dims(std::vector<int64_t> dims) {
+  // yolov3
+  if (_net_flag == 101 && _index == 1) {
+    _tensor->Resize({1, 2});
+    _tensor->mutable_data<int>()[0] = (int)dims.at(2);
+    _tensor->mutable_data<int>()[1] = (int)dims.at(3);
+  } else {
+    _tensor->Resize(dims);
+  }
+  _is_dims_set = true;
+}
+
+float *PredictorInput::get_mutable_float_data() {
+  if (!_is_dims_set) {
+    LOGE("PredictorInput::set_dims is not called");
+  }
+  return _tensor->mutable_data<float>();
+}
+
+void PredictorInput::set_data(const float *input_data, int input_float_len) {
+  float *input_raw_data = get_mutable_float_data();
+  memcpy(input_raw_data, input_data, input_float_len * sizeof(float));
+}
+} // namespace ppredictor
diff --git a/deploy/android_demo/app/src/main/cpp/predictor_input.h b/deploy/android_demo/app/src/main/cpp/predictor_input.h
new file mode 100644
index 0000000..185f081
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/predictor_input.h
@@ -0,0 +1,26 @@
+#pragma once
+
+#include "common.h"
+#include <paddle_api.h>
+#include <vector>
+
+namespace ppredictor {
+class PredictorInput {
+public:
+  PredictorInput(std::unique_ptr<paddle::lite_api::Tensor> &&tensor, int index,
+                 int net_flag)
+      : _tensor(std::move(tensor)), _index(index), _net_flag(net_flag) {}
+
+  void set_dims(std::vector<int64_t> dims);
+
+  float *get_mutable_float_data();
+
+  void set_data(const float *input_data, int input_float_len);
+
+private:
+  std::unique_ptr<paddle::lite_api::Tensor> _tensor;
+  bool _is_dims_set = false;
+  int _index;
+  int _net_flag;
+};
+} // namespace ppredictor
diff --git a/deploy/android_demo/app/src/main/cpp/predictor_output.cpp b/deploy/android_demo/app/src/main/cpp/predictor_output.cpp
new file mode 100644
index 0000000..2176ef5
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/predictor_output.cpp
@@ -0,0 +1,26 @@
+#include "predictor_output.h"
+namespace ppredictor {
+const float *PredictorOutput::get_float_data() const {
+  return _tensor->data<float>();
+}
+
+const int *PredictorOutput::get_int_data() const {
+  return _tensor->data<int>();
+}
+
+const std::vector<std::vector<uint64_t>> PredictorOutput::get_lod() const {
+  return _tensor->lod();
+}
+
+int64_t PredictorOutput::get_size() const {
+  if (_net_flag == NET_OCR) {
+    return _tensor->shape().at(2) * _tensor->shape().at(3);
+  } else {
+    return product(_tensor->shape());
+  }
+}
+
+const std::vector<int64_t> PredictorOutput::get_shape() const {
+  return _tensor->shape();
+}
+} // namespace ppredictor
diff --git a/deploy/android_demo/app/src/main/cpp/predictor_output.h b/deploy/android_demo/app/src/main/cpp/predictor_output.h
new file mode 100644
index 0000000..ce00af7
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/predictor_output.h
@@ -0,0 +1,31 @@
+#pragma once
+
+#include "common.h"
+#include <paddle_api.h>
+#include <vector>
+
+namespace ppredictor {
+class PredictorOutput {
+public:
+  PredictorOutput() {}
+  PredictorOutput(std::unique_ptr<const paddle::lite_api::Tensor> &&tensor,
+                  int index, int net_flag)
+      : _tensor(std::move(tensor)), _index(index), _net_flag(net_flag) {}
+
+  const float *get_float_data() const;
+  const int *get_int_data() const;
+  int64_t get_size() const;
+  const std::vector<std::vector<uint64_t>> get_lod() const;
+  const std::vector<int64_t> get_shape() const;
+
+  std::vector<float> data;    // return float, or use data_int
+  std::vector<int> data_int;  // several layers return int ，or use data
+  std::vector<int64_t> shape; // PaddleLite output shape
+  std::vector<std::vector<uint64_t>> lod; // PaddleLite output lod
+
+private:
+  std::unique_ptr<const paddle::lite_api::Tensor> _tensor;
+  int _index;
+  int _net_flag;
+};
+} // namespace ppredictor
diff --git a/deploy/android_demo/app/src/main/cpp/preprocess.cpp b/deploy/android_demo/app/src/main/cpp/preprocess.cpp
new file mode 100644
index 0000000..f3b29ec
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/preprocess.cpp
@@ -0,0 +1,82 @@
+#include "preprocess.h"
+#include <android/bitmap.h>
+
+cv::Mat bitmap_to_cv_mat(JNIEnv *env, jobject bitmap) {
+  AndroidBitmapInfo info;
+  int result = AndroidBitmap_getInfo(env, bitmap, &info);
+  if (result != ANDROID_BITMAP_RESULT_SUCCESS) {
+    LOGE("AndroidBitmap_getInfo failed, result: %d", result);
+    return cv::Mat{};
+  }
+  if (info.format != ANDROID_BITMAP_FORMAT_RGBA_8888) {
+    LOGE("Bitmap format is not RGBA_8888 !");
+    return cv::Mat{};
+  }
+  unsigned char *srcData = NULL;
+  AndroidBitmap_lockPixels(env, bitmap, (void **)&srcData);
+  cv::Mat mat = cv::Mat::zeros(info.height, info.width, CV_8UC4);
+  memcpy(mat.data, srcData, info.height * info.width * 4);
+  AndroidBitmap_unlockPixels(env, bitmap);
+  cv::cvtColor(mat, mat, cv::COLOR_RGBA2BGR);
+  /**
+  if (!cv::imwrite("/sdcard/1/copy.jpg", mat)){
+      LOGE("Write image failed " );
+  }
+   */
+
+  return mat;
+}
+
+cv::Mat resize_img(const cv::Mat &img, int height, int width) {
+  if (img.rows == height && img.cols == width) {
+    return img;
+  }
+  cv::Mat new_img;
+  cv::resize(img, new_img, cv::Size(height, width));
+  return new_img;
+}
+
+// fill tensor with mean and scale and trans layout: nhwc -> nchw, neon speed up
+void neon_mean_scale(const float *din, float *dout, int size,
+                     const std::vector<float> &mean,
+                     const std::vector<float> &scale) {
+  if (mean.size() != 3 || scale.size() != 3) {
+    LOGE("[ERROR] mean or scale size must equal to 3");
+    return;
+  }
+
+  float32x4_t vmean0 = vdupq_n_f32(mean[0]);
+  float32x4_t vmean1 = vdupq_n_f32(mean[1]);
+  float32x4_t vmean2 = vdupq_n_f32(mean[2]);
+  float32x4_t vscale0 = vdupq_n_f32(scale[0]);
+  float32x4_t vscale1 = vdupq_n_f32(scale[1]);
+  float32x4_t vscale2 = vdupq_n_f32(scale[2]);
+
+  float *dout_c0 = dout;
+  float *dout_c1 = dout + size;
+  float *dout_c2 = dout + size * 2;
+
+  int i = 0;
+  for (; i < size - 3; i += 4) {
+    float32x4x3_t vin3 = vld3q_f32(din);
+    float32x4_t vsub0 = vsubq_f32(vin3.val[0], vmean0);
+    float32x4_t vsub1 = vsubq_f32(vin3.val[1], vmean1);
+    float32x4_t vsub2 = vsubq_f32(vin3.val[2], vmean2);
+    float32x4_t vs0 = vmulq_f32(vsub0, vscale0);
+    float32x4_t vs1 = vmulq_f32(vsub1, vscale1);
+    float32x4_t vs2 = vmulq_f32(vsub2, vscale2);
+    vst1q_f32(dout_c0, vs0);
+    vst1q_f32(dout_c1, vs1);
+    vst1q_f32(dout_c2, vs2);
+
+    din += 12;
+    dout_c0 += 4;
+    dout_c1 += 4;
+    dout_c2 += 4;
+  }
+  for (; i < size; i++) {
+    *(dout_c0++) = (*(din++) - mean[0]) * scale[0];
+    *(dout_c1++) = (*(din++) - mean[1]) * scale[1];
+    *(dout_c2++) = (*(din++) - mean[2]) * scale[2];
+  }
+}
diff --git a/deploy/android_demo/app/src/main/cpp/preprocess.h b/deploy/android_demo/app/src/main/cpp/preprocess.h
new file mode 100644
index 0000000..7909152
--- /dev/null
+++ b/deploy/android_demo/app/src/main/cpp/preprocess.h
@@ -0,0 +1,12 @@
+#pragma once
+
+#include "common.h"
+#include <jni.h>
+#include <opencv2/opencv.hpp>
+cv::Mat bitmap_to_cv_mat(JNIEnv *env, jobject bitmap);
+
+cv::Mat resize_img(const cv::Mat &img, int height, int width);
+
+void neon_mean_scale(const float *din, float *dout, int size,
+                     const std::vector<float> &mean,
+                     const std::vector<float> &scale);
diff --git a/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/AppCompatPreferenceActivity.java b/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/AppCompatPreferenceActivity.java
new file mode 100644
index 0000000..49af0af
--- /dev/null
+++ b/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/AppCompatPreferenceActivity.java
@@ -0,0 +1,128 @@
+/*
+ * Copyright (C) 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package com.baidu.paddle.lite.demo.ocr;
+
+import android.content.res.Configuration;
+import android.os.Bundle;
+import android.preference.PreferenceActivity;
+import android.view.MenuInflater;
+import android.view.View;
+import android.view.ViewGroup;
+
+import androidx.annotation.LayoutRes;
+import androidx.annotation.Nullable;
+import androidx.appcompat.app.ActionBar;
+import androidx.appcompat.app.AppCompatDelegate;
+import androidx.appcompat.widget.Toolbar;
+
+/**
+ * A {@link PreferenceActivity} which implements and proxies the necessary calls
+ * to be used with AppCompat.
+ * <p>
+ * This technique can be used with an {@link android.app.Activity} class, not just
+ * {@link PreferenceActivity}.
+ */
+public abstract class AppCompatPreferenceActivity extends PreferenceActivity {
+    private AppCompatDelegate mDelegate;
+
+    @Override
+    protected void onCreate(Bundle savedInstanceState) {
+        getDelegate().installViewFactory();
+        getDelegate().onCreate(savedInstanceState);
+        super.onCreate(savedInstanceState);
+    }
+
+    @Override
+    protected void onPostCreate(Bundle savedInstanceState) {
+        super.onPostCreate(savedInstanceState);
+        getDelegate().onPostCreate(savedInstanceState);
+    }
+
+    public ActionBar getSupportActionBar() {
+        return getDelegate().getSupportActionBar();
+    }
+
+    public void setSupportActionBar(@Nullable Toolbar toolbar) {
+        getDelegate().setSupportActionBar(toolbar);
+    }
+
+    @Override
+    public MenuInflater getMenuInflater() {
+        return getDelegate().getMenuInflater();
+    }
+
+    @Override
+    public void setContentView(@LayoutRes int layoutResID) {
+        getDelegate().setContentView(layoutResID);
+    }
+
+    @Override
+    public void setContentView(View view) {
+        getDelegate().setContentView(view);
+    }
+
+    @Override
+    public void setContentView(View view, ViewGroup.LayoutParams params) {
+        getDelegate().setContentView(view, params);
+    }
+
+    @Override
+    public void addContentView(View view, ViewGroup.LayoutParams params) {
+        getDelegate().addContentView(view, params);
+    }
+
+    @Override
+    protected void onPostResume() {
+        super.onPostResume();
+        getDelegate().onPostResume();
+    }
+
+    @Override
+    protected void onTitleChanged(CharSequence title, int color) {
+        super.onTitleChanged(title, color);
+        getDelegate().setTitle(title);
+    }
+
+    @Override
+    public void onConfigurationChanged(Configuration newConfig) {
+        super.onConfigurationChanged(newConfig);
+        getDelegate().onConfigurationChanged(newConfig);
+    }
+
+    @Override
+    protected void onStop() {
+        super.onStop();
+        getDelegate().onStop();
+    }
+
+    @Override
+    protected void onDestroy() {
+        super.onDestroy();
+        getDelegate().onDestroy();
+    }
+
+    public void invalidateOptionsMenu() {
+        getDelegate().invalidateOptionsMenu();
+    }
+
+    private AppCompatDelegate getDelegate() {
+        if (mDelegate == null) {
+            mDelegate = AppCompatDelegate.create(this, null);
+        }
+        return mDelegate;
+    }
+}
diff --git a/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/MainActivity.java b/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/MainActivity.java
new file mode 100644
index 0000000..91a6d76
--- /dev/null
+++ b/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/MainActivity.java
@@ -0,0 +1,520 @@
+package com.baidu.paddle.lite.demo.ocr;
+
+import android.Manifest;
+import android.app.ProgressDialog;
+import android.content.ContentResolver;
+import android.content.Context;
+import android.content.Intent;
+import android.content.SharedPreferences;
+import android.content.pm.PackageManager;
+import android.database.Cursor;
+import android.graphics.Bitmap;
+import android.graphics.BitmapFactory;
+import android.graphics.drawable.BitmapDrawable;
+import android.media.ExifInterface;
+import android.content.res.AssetManager;
+import android.media.FaceDetector;
+import android.net.Uri;
+import android.os.Bundle;
+import android.os.Environment;
+import android.os.Handler;
+import android.os.HandlerThread;
+import android.os.Message;
+import android.preference.PreferenceManager;
+import android.provider.MediaStore;
+import android.text.method.ScrollingMovementMethod;
+import android.util.Log;
+import android.view.Menu;
+import android.view.MenuInflater;
+import android.view.MenuItem;
+import android.view.View;
+import android.widget.CheckBox;
+import android.widget.ImageView;
+import android.widget.Spinner;
+import android.widget.TextView;
+import android.widget.Toast;
+
+import androidx.annotation.NonNull;
+import androidx.appcompat.app.AppCompatActivity;
+import androidx.core.app.ActivityCompat;
+import androidx.core.content.ContextCompat;
+import androidx.core.content.FileProvider;
+
+import java.io.File;
+import java.io.IOException;
+import java.io.InputStream;
+import java.text.SimpleDateFormat;
+import java.util.Date;
+
+public class MainActivity extends AppCompatActivity {
+    private static final String TAG = MainActivity.class.getSimpleName();
+    public static final int OPEN_GALLERY_REQUEST_CODE = 0;
+    public static final int TAKE_PHOTO_REQUEST_CODE = 1;
+
+    public static final int REQUEST_LOAD_MODEL = 0;
+    public static final int REQUEST_RUN_MODEL = 1;
+    public static final int RESPONSE_LOAD_MODEL_SUCCESSED = 0;
+    public static final int RESPONSE_LOAD_MODEL_FAILED = 1;
+    public static final int RESPONSE_RUN_MODEL_SUCCESSED = 2;
+    public static final int RESPONSE_RUN_MODEL_FAILED = 3;
+
+    protected ProgressDialog pbLoadModel = null;
+    protected ProgressDialog pbRunModel = null;
+
+    protected Handler receiver = null; // Receive messages from worker thread
+    protected Handler sender = null; // Send command to worker thread
+    protected HandlerThread worker = null; // Worker thread to load&run model
+
+    // UI components of object detection
+    protected TextView tvInputSetting;
+    protected TextView tvStatus;
+    protected ImageView ivInputImage;
+    protected TextView tvOutputResult;
+    protected TextView tvInferenceTime;
+    protected CheckBox cbOpencl;
+    protected Spinner spRunMode;
+
+    // Model settings of ocr
+    protected String modelPath = "";
+    protected String labelPath = "";
+    protected String imagePath = "";
+    protected int cpuThreadNum = 1;
+    protected String cpuPowerMode = "";
+    protected int detLongSize = 960;
+    protected float scoreThreshold = 0.1f;
+    private String currentPhotoPath;
+    private AssetManager assetManager = null;
+
+    protected Predictor predictor = new Predictor();
+
+    private Bitmap cur_predict_image = null;
+
+    @Override
+    protected void onCreate(Bundle savedInstanceState) {
+        super.onCreate(savedInstanceState);
+        setContentView(R.layout.activity_main);
+
+        // Clear all setting items to avoid app crashing due to the incorrect settings
+        SharedPreferences sharedPreferences = PreferenceManager.getDefaultSharedPreferences(this);
+        SharedPreferences.Editor editor = sharedPreferences.edit();
+        editor.clear();
+        editor.apply();
+
+        // Setup the UI components
+        tvInputSetting = findViewById(R.id.tv_input_setting);
+        cbOpencl = findViewById(R.id.cb_opencl);
+        tvStatus = findViewById(R.id.tv_model_img_status);
+        ivInputImage = findViewById(R.id.iv_input_image);
+        tvInferenceTime = findViewById(R.id.tv_inference_time);
+        tvOutputResult = findViewById(R.id.tv_output_result);
+        spRunMode = findViewById(R.id.sp_run_mode);
+        tvInputSetting.setMovementMethod(ScrollingMovementMethod.getInstance());
+        tvOutputResult.setMovementMethod(ScrollingMovementMethod.getInstance());
+
+        // Prepare the worker thread for mode loading and inference
+        receiver = new Handler() {
+            @Override
+            public void handleMessage(Message msg) {
+                switch (msg.what) {
+                    case RESPONSE_LOAD_MODEL_SUCCESSED:
+                        if (pbLoadModel != null && pbLoadModel.isShowing()) {
+                            pbLoadModel.dismiss();
+                        }
+                        onLoadModelSuccessed();
+                        break;
+                    case RESPONSE_LOAD_MODEL_FAILED:
+                        if (pbLoadModel != null && pbLoadModel.isShowing()) {
+                            pbLoadModel.dismiss();
+                        }
+                        Toast.makeText(MainActivity.this, "Load model failed!", Toast.LENGTH_SHORT).show();
+                        onLoadModelFailed();
+                        break;
+                    case RESPONSE_RUN_MODEL_SUCCESSED:
+                        if (pbRunModel != null && pbRunModel.isShowing()) {
+                            pbRunModel.dismiss();
+                        }
+                        onRunModelSuccessed();
+                        break;
+                    case RESPONSE_RUN_MODEL_FAILED:
+                        if (pbRunModel != null && pbRunModel.isShowing()) {
+                            pbRunModel.dismiss();
+                        }
+                        Toast.makeText(MainActivity.this, "Run model failed!", Toast.LENGTH_SHORT).show();
+                        onRunModelFailed();
+                        break;
+                    default:
+                        break;
+                }
+            }
+        };
+
+        worker = new HandlerThread("Predictor Worker");
+        worker.start();
+        sender = new Handler(worker.getLooper()) {
+            public void handleMessage(Message msg) {
+                switch (msg.what) {
+                    case REQUEST_LOAD_MODEL:
+                        // Load model and reload test image
+                        if (onLoadModel()) {
+                            receiver.sendEmptyMessage(RESPONSE_LOAD_MODEL_SUCCESSED);
+                        } else {
+                            receiver.sendEmptyMessage(RESPONSE_LOAD_MODEL_FAILED);
+                        }
+                        break;
+                    case REQUEST_RUN_MODEL:
+                        // Run model if model is loaded
+                        if (onRunModel()) {
+                            receiver.sendEmptyMessage(RESPONSE_RUN_MODEL_SUCCESSED);
+                        } else {
+                            receiver.sendEmptyMessage(RESPONSE_RUN_MODEL_FAILED);
+                        }
+                        break;
+                    default:
+                        break;
+                }
+            }
+        };
+    }
+
+    @Override
+    protected void onResume() {
+        super.onResume();
+        SharedPreferences sharedPreferences = PreferenceManager.getDefaultSharedPreferences(this);
+        boolean settingsChanged = false;
+        boolean model_settingsChanged = false;
+        String model_path = sharedPreferences.getString(getString(R.string.MODEL_PATH_KEY),
+                getString(R.string.MODEL_PATH_DEFAULT));
+        String label_path = sharedPreferences.getString(getString(R.string.LABEL_PATH_KEY),
+                getString(R.string.LABEL_PATH_DEFAULT));
+        String image_path = sharedPreferences.getString(getString(R.string.IMAGE_PATH_KEY),
+                getString(R.string.IMAGE_PATH_DEFAULT));
+        model_settingsChanged |= !model_path.equalsIgnoreCase(modelPath);
+        settingsChanged |= !label_path.equalsIgnoreCase(labelPath);
+        settingsChanged |= !image_path.equalsIgnoreCase(imagePath);
+        int cpu_thread_num = Integer.parseInt(sharedPreferences.getString(getString(R.string.CPU_THREAD_NUM_KEY),
+                getString(R.string.CPU_THREAD_NUM_DEFAULT)));
+        model_settingsChanged |= cpu_thread_num != cpuThreadNum;
+        String cpu_power_mode =
+                sharedPreferences.getString(getString(R.string.CPU_POWER_MODE_KEY),
+                        getString(R.string.CPU_POWER_MODE_DEFAULT));
+        model_settingsChanged |= !cpu_power_mode.equalsIgnoreCase(cpuPowerMode);
+
+        int det_long_size = Integer.parseInt(sharedPreferences.getString(getString(R.string.DET_LONG_SIZE_KEY),
+                getString(R.string.DET_LONG_SIZE_DEFAULT)));
+        settingsChanged |= det_long_size != detLongSize;
+        float score_threshold =
+                Float.parseFloat(sharedPreferences.getString(getString(R.string.SCORE_THRESHOLD_KEY),
+                        getString(R.string.SCORE_THRESHOLD_DEFAULT)));
+        settingsChanged |= scoreThreshold != score_threshold;
+        if (settingsChanged) {
+            labelPath = label_path;
+            imagePath = image_path;
+            detLongSize = det_long_size;
+            scoreThreshold = score_threshold;
+            set_img();
+        }
+        if (model_settingsChanged) {
+            modelPath = model_path;
+            cpuThreadNum = cpu_thread_num;
+            cpuPowerMode = cpu_power_mode;
+            // Update UI
+            tvInputSetting.setText("Model: " + modelPath.substring(modelPath.lastIndexOf("/") + 1) + "\nOPENCL: " + cbOpencl.isChecked() + "\nCPU Thread Num: " + cpuThreadNum + "\nCPU Power Mode: " + cpuPowerMode);
+            tvInputSetting.scrollTo(0, 0);
+            // Reload model if configure has been changed
+            loadModel();
+        }
+    }
+
+    public void loadModel() {
+        pbLoadModel = ProgressDialog.show(this, "", "loading model...", false, false);
+        sender.sendEmptyMessage(REQUEST_LOAD_MODEL);
+    }
+
+    public void runModel() {
+        pbRunModel = ProgressDialog.show(this, "", "running model...", false, false);
+        sender.sendEmptyMessage(REQUEST_RUN_MODEL);
+    }
+
+    public boolean onLoadModel() {
+        if (predictor.isLoaded()) {
+            predictor.releaseModel();
+        }
+        return predictor.init(MainActivity.this, modelPath, labelPath, cbOpencl.isChecked() ? 1 : 0, cpuThreadNum,
+                cpuPowerMode,
+                detLongSize, scoreThreshold);
+    }
+
+    public boolean onRunModel() {
+        String run_mode = spRunMode.getSelectedItem().toString();
+        int run_det = run_mode.contains("检测") ? 1 : 0;
+        int run_cls = run_mode.contains("分类") ? 1 : 0;
+        int run_rec = run_mode.contains("识别") ? 1 : 0;
+        return predictor.isLoaded() && predictor.runModel(run_det, run_cls, run_rec);
+    }
+
+    public void onLoadModelSuccessed() {
+        // Load test image from path and run model
+        tvInputSetting.setText("Model: " + modelPath.substring(modelPath.lastIndexOf("/") + 1) + "\nOPENCL: " + cbOpencl.isChecked() + "\nCPU Thread Num: " + cpuThreadNum + "\nCPU Power Mode: " + cpuPowerMode);
+        tvInputSetting.scrollTo(0, 0);
+        tvStatus.setText("STATUS: load model succeeded");
+
+    }
+
+    public void onLoadModelFailed() {
+        tvStatus.setText("STATUS: load model failed");
+    }
+
+    public void onRunModelSuccessed() {
+        tvStatus.setText("STATUS: run model succeeded");
+        // Obtain results and update UI
+        tvInferenceTime.setText("Inference time: " + predictor.inferenceTime() + " ms");
+        Bitmap outputImage = predictor.outputImage();
+        if (outputImage != null) {
+            ivInputImage.setImageBitmap(outputImage);
+        }
+        tvOutputResult.setText(predictor.outputResult());
+        tvOutputResult.scrollTo(0, 0);
+    }
+
+    public void onRunModelFailed() {
+        tvStatus.setText("STATUS: run model failed");
+    }
+
+    public void set_img() {
+        // Load test image from path and run model
+        try {
+            assetManager = getAssets();
+            InputStream in = assetManager.open(imagePath);
+            Bitmap bmp = BitmapFactory.decodeStream(in);
+            cur_predict_image = bmp;
+            ivInputImage.setImageBitmap(bmp);
+        } catch (IOException e) {
+            Toast.makeText(MainActivity.this, "Load image failed!", Toast.LENGTH_SHORT).show();
+            e.printStackTrace();
+        }
+    }
+
+    public void onSettingsClicked() {
+        startActivity(new Intent(MainActivity.this, SettingsActivity.class));
+    }
+
+    @Override
+    public boolean onCreateOptionsMenu(Menu menu) {
+        MenuInflater inflater = getMenuInflater();
+        inflater.inflate(R.menu.menu_action_options, menu);
+        return true;
+    }
+
+    public boolean onPrepareOptionsMenu(Menu menu) {
+        boolean isLoaded = predictor.isLoaded();
+        return super.onPrepareOptionsMenu(menu);
+    }
+
+    @Override
+    public boolean onOptionsItemSelected(MenuItem item) {
+        switch (item.getItemId()) {
+            case android.R.id.home:
+                finish();
+                break;
+            case R.id.settings:
+                if (requestAllPermissions()) {
+                    // Make sure we have SDCard r&w permissions to load model from SDCard
+                    onSettingsClicked();
+                }
+                break;
+        }
+        return super.onOptionsItemSelected(item);
+    }
+
+    @Override
+    public void onRequestPermissionsResult(int requestCode, @NonNull String[] permissions,
+                                           @NonNull int[] grantResults) {
+        super.onRequestPermissionsResult(requestCode, permissions, grantResults);
+        if (grantResults[0] != PackageManager.PERMISSION_GRANTED || grantResults[1] != PackageManager.PERMISSION_GRANTED) {
+            Toast.makeText(this, "Permission Denied", Toast.LENGTH_SHORT).show();
+        }
+    }
+
+    private boolean requestAllPermissions() {
+        if (ContextCompat.checkSelfPermission(this, Manifest.permission.WRITE_EXTERNAL_STORAGE)
+                != PackageManager.PERMISSION_GRANTED || ContextCompat.checkSelfPermission(this,
+                Manifest.permission.CAMERA)
+                != PackageManager.PERMISSION_GRANTED) {
+            ActivityCompat.requestPermissions(this, new String[]{Manifest.permission.WRITE_EXTERNAL_STORAGE,
+                            Manifest.permission.CAMERA},
+                    0);
+            return false;
+        }
+        return true;
+    }
+
+    private void openGallery() {
+        Intent intent = new Intent(Intent.ACTION_PICK, null);
+        intent.setDataAndType(MediaStore.Images.Media.EXTERNAL_CONTENT_URI, "image/*");
+        startActivityForResult(intent, OPEN_GALLERY_REQUEST_CODE);
+    }
+
+    private void takePhoto() {
+        Intent takePictureIntent = new Intent(MediaStore.ACTION_IMAGE_CAPTURE);
+        // Ensure that there's a camera activity to handle the intent
+        if (takePictureIntent.resolveActivity(getPackageManager()) != null) {
+            // Create the File where the photo should go
+            File photoFile = null;
+            try {
+                photoFile = createImageFile();
+            } catch (IOException ex) {
+                Log.e("MainActitity", ex.getMessage(), ex);
+                Toast.makeText(MainActivity.this,
+                        "Create Camera temp file failed: " + ex.getMessage(), Toast.LENGTH_SHORT).show();
+            }
+            // Continue only if the File was successfully created
+            if (photoFile != null) {
+                Log.i(TAG, "FILEPATH " + getExternalFilesDir("Pictures").getAbsolutePath());
+                Uri photoURI = FileProvider.getUriForFile(this,
+                        "com.baidu.paddle.lite.demo.ocr.fileprovider",
+                        photoFile);
+                currentPhotoPath = photoFile.getAbsolutePath();
+                takePictureIntent.putExtra(MediaStore.EXTRA_OUTPUT, photoURI);
+                startActivityForResult(takePictureIntent, TAKE_PHOTO_REQUEST_CODE);
+                Log.i(TAG, "startActivityForResult finished");
+            }
+        }
+
+    }
+
+    private File createImageFile() throws IOException {
+        // Create an image file name
+        String timeStamp = new SimpleDateFormat("yyyyMMdd_HHmmss").format(new Date());
+        String imageFileName = "JPEG_" + timeStamp + "_";
+        File storageDir = getExternalFilesDir(Environment.DIRECTORY_PICTURES);
+        File image = File.createTempFile(
+                imageFileName,  /* prefix */
+                ".bmp",         /* suffix */
+                storageDir      /* directory */
+        );
+
+        return image;
+    }
+
+    @Override
+    protected void onActivityResult(int requestCode, int resultCode, Intent data) {
+        super.onActivityResult(requestCode, resultCode, data);
+        if (resultCode == RESULT_OK) {
+            switch (requestCode) {
+                case OPEN_GALLERY_REQUEST_CODE:
+                    if (data == null) {
+                        break;
+                    }
+                    try {
+                        ContentResolver resolver = getContentResolver();
+                        Uri uri = data.getData();
+                        Bitmap image = MediaStore.Images.Media.getBitmap(resolver, uri);
+                        String[] proj = {MediaStore.Images.Media.DATA};
+                        Cursor cursor = managedQuery(uri, proj, null, null, null);
+                        cursor.moveToFirst();
+                        if (image != null) {
+                            cur_predict_image = image;
+                            ivInputImage.setImageBitmap(image);
+                        }
+                    } catch (IOException e) {
+                        Log.e(TAG, e.toString());
+                    }
+                    break;
+                case TAKE_PHOTO_REQUEST_CODE:
+                    if (currentPhotoPath != null) {
+                        ExifInterface exif = null;
+                        try {
+                            exif = new ExifInterface(currentPhotoPath);
+                        } catch (IOException e) {
+                            e.printStackTrace();
+                        }
+                        int orientation = exif.getAttributeInt(ExifInterface.TAG_ORIENTATION,
+                                ExifInterface.ORIENTATION_UNDEFINED);
+                        Log.i(TAG, "rotation " + orientation);
+                        Bitmap image = BitmapFactory.decodeFile(currentPhotoPath);
+                        image = Utils.rotateBitmap(image, orientation);
+                        if (image != null) {
+                            cur_predict_image = image;
+                            ivInputImage.setImageBitmap(image);
+                        }
+                    } else {
+                        Log.e(TAG, "currentPhotoPath is null");
+                    }
+                    break;
+                default:
+                    break;
+            }
+        }
+    }
+
+    public void btn_reset_img_click(View view) {
+        ivInputImage.setImageBitmap(cur_predict_image);
+    }
+
+    public void cb_opencl_click(View view) {
+        tvStatus.setText("STATUS: load model ......");
+        loadModel();
+    }
+
+    public void btn_run_model_click(View view) {
+        Bitmap image = ((BitmapDrawable) ivInputImage.getDrawable()).getBitmap();
+        if (image == null) {
+            tvStatus.setText("STATUS: image is not exists");
+        } else if (!predictor.isLoaded()) {
+            tvStatus.setText("STATUS: model is not loaded");
+        } else {
+            tvStatus.setText("STATUS: run model ...... ");
+            predictor.setInputImage(image);
+            runModel();
+        }
+    }
+
+    public void btn_choice_img_click(View view) {
+        if (requestAllPermissions()) {
+            openGallery();
+        }
+    }
+
+    public void btn_take_photo_click(View view) {
+        if (requestAllPermissions()) {
+            takePhoto();
+        }
+    }
+
+    @Override
+    protected void onDestroy() {
+        if (predictor != null) {
+            predictor.releaseModel();
+        }
+        worker.quit();
+        super.onDestroy();
+    }
+
+    public int get_run_mode() {
+        String run_mode = spRunMode.getSelectedItem().toString();
+        int mode;
+        switch (run_mode) {
+            case "检测+分类+识别":
+                mode = 1;
+                break;
+            case "检测+识别":
+                mode = 2;
+                break;
+            case "识别+分类":
+                mode = 3;
+                break;
+            case "检测":
+                mode = 4;
+                break;
+            case "识别":
+                mode = 5;
+                break;
+            case "分类":
+                mode = 6;
+                break;
+            default:
+                mode = 1;
+        }
+        return mode;
+    }
+}
diff --git a/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/OCRPredictorNative.java b/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/OCRPredictorNative.java
new file mode 100644
index 0000000..c0bfd84
--- /dev/null
+++ b/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/OCRPredictorNative.java
@@ -0,0 +1,105 @@
+package com.baidu.paddle.lite.demo.ocr;
+
+import android.graphics.Bitmap;
+import android.util.Log;
+
+import java.util.ArrayList;
+import java.util.concurrent.atomic.AtomicBoolean;
+
+public class OCRPredictorNative {
+
+    private static final AtomicBoolean isSOLoaded = new AtomicBoolean();
+
+    public static void loadLibrary() throws RuntimeException {
+        if (!isSOLoaded.get() && isSOLoaded.compareAndSet(false, true)) {
+            try {
+                System.loadLibrary("Native");
+            } catch (Throwable e) {
+                RuntimeException exception = new RuntimeException(
+                        "Load libNative.so failed, please check it exists in apk file.", e);
+                throw exception;
+            }
+        }
+    }
+
+    private Config config;
+
+    private long nativePointer = 0;
+
+    public OCRPredictorNative(Config config) {
+        this.config = config;
+        loadLibrary();
+        nativePointer = init(config.detModelFilename, config.recModelFilename, config.clsModelFilename, config.useOpencl,
+                config.cpuThreadNum, config.cpuPower);
+        Log.i("OCRPredictorNative", "load success " + nativePointer);
+
+    }
+
+
+    public ArrayList<OcrResultModel> runImage(Bitmap originalImage, int max_size_len, int run_det, int run_cls, int run_rec) {
+        Log.i("OCRPredictorNative", "begin to run image ");
+        float[] rawResults = forward(nativePointer, originalImage, max_size_len, run_det, run_cls, run_rec);
+        ArrayList<OcrResultModel> results = postprocess(rawResults);
+        return results;
+    }
+
+    public static class Config {
+        public int useOpencl;
+        public int cpuThreadNum;
+        public String cpuPower;
+        public String detModelFilename;
+        public String recModelFilename;
+        public String clsModelFilename;
+
+    }
+
+    public void destroy() {
+        if (nativePointer != 0) {
+            release(nativePointer);
+            nativePointer = 0;
+        }
+    }
+
+    protected native long init(String detModelPath, String recModelPath, String clsModelPath, int useOpencl, int threadNum, String cpuMode);
+
+    protected native float[] forward(long pointer, Bitmap originalImage,int max_size_len, int run_det, int run_cls, int run_rec);
+
+    protected native void release(long pointer);
+
+    private ArrayList<OcrResultModel> postprocess(float[] raw) {
+        ArrayList<OcrResultModel> results = new ArrayList<OcrResultModel>();
+        int begin = 0;
+
+        while (begin < raw.length) {
+            int point_num = Math.round(raw[begin]);
+            int word_num = Math.round(raw[begin + 1]);
+            OcrResultModel res = parse(raw, begin + 2, point_num, word_num);
+            begin += 2 + 1 + point_num * 2 + word_num + 2;
+            results.add(res);
+        }
+
+        return results;
+    }
+
+    private OcrResultModel parse(float[] raw, int begin, int pointNum, int wordNum) {
+        int current = begin;
+        OcrResultModel res = new OcrResultModel();
+        res.setConfidence(raw[current]);
+        current++;
+        for (int i = 0; i < pointNum; i++) {
+            res.addPoints(Math.round(raw[current + i * 2]), Math.round(raw[current + i * 2 + 1]));
+        }
+        current += (pointNum * 2);
+        for (int i = 0; i < wordNum; i++) {
+            int index = Math.round(raw[current + i]);
+            res.addWordIndex(index);
+        }
+        current += wordNum;
+        res.setClsIdx(raw[current]);
+        res.setClsConfidence(raw[current + 1]);
+        Log.i("OCRPredictorNative", "word finished " + wordNum);
+        return res;
+    }
+
+
+}
diff --git a/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/OcrResultModel.java b/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/OcrResultModel.java
new file mode 100644
index 0000000..1bccbc7
--- /dev/null
+++ b/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/OcrResultModel.java
@@ -0,0 +1,79 @@
+package com.baidu.paddle.lite.demo.ocr;
+
+import android.graphics.Point;
+
+import java.util.ArrayList;
+import java.util.List;
+
+public class OcrResultModel {
+    private List<Point> points;
+    private List<Integer> wordIndex;
+    private String label;
+    private float confidence;
+    private float cls_idx;
+    private String cls_label;
+    private float cls_confidence;
+
+    public OcrResultModel() {
+        super();
+        points = new ArrayList<>();
+        wordIndex = new ArrayList<>();
+    }
+
+    public void addPoints(int x, int y) {
+        Point point = new Point(x, y);
+        points.add(point);
+    }
+
+    public void addWordIndex(int index) {
+        wordIndex.add(index);
+    }
+
+    public List<Point> getPoints() {
+        return points;
+    }
+
+    public List<Integer> getWordIndex() {
+        return wordIndex;
+    }
+
+    public String getLabel() {
+        return label;
+    }
+
+    public void setLabel(String label) {
+        this.label = label;
+    }
+
+    public float getConfidence() {
+        return confidence;
+    }
+
+    public void setConfidence(float confidence) {
+        this.confidence = confidence;
+    }
+
+    public float getClsIdx() {
+        return cls_idx;
+    }
+
+    public void setClsIdx(float idx) {
+        this.cls_idx = idx;
+    }
+
+    public String getClsLabel() {
+        return cls_label;
+    }
+
+    public void setClsLabel(String label) {
+        this.cls_label = label;
+    }
+
+    public float getClsConfidence() {
+        return cls_confidence;
+    }
+
+    public void setClsConfidence(float confidence) {
+        this.cls_confidence = confidence;
+    }
+}
diff --git a/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/Predictor.java b/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/Predictor.java
new file mode 100644
index 0000000..7bd6245
--- /dev/null
+++ b/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/Predictor.java
@@ -0,0 +1,278 @@
+package com.baidu.paddle.lite.demo.ocr;
+
+import android.content.Context;
+import android.graphics.Bitmap;
+import android.graphics.Canvas;
+import android.graphics.Color;
+import android.graphics.Paint;
+import android.graphics.Path;
+import android.graphics.Point;
+import android.util.Log;
+
+import java.io.File;
+import java.io.InputStream;
+import java.util.ArrayList;
+import java.util.Date;
+import java.util.List;
+import java.util.Vector;
+
+import static android.graphics.Color.*;
+
+public class Predictor {
+    private static final String TAG = Predictor.class.getSimpleName();
+    public boolean isLoaded = false;
+    public int warmupIterNum = 1;
+    public int inferIterNum = 1;
+    public int cpuThreadNum = 4;
+    public String cpuPowerMode = "LITE_POWER_HIGH";
+    public String modelPath = "";
+    public String modelName = "";
+    protected OCRPredictorNative paddlePredictor = null;
+    protected float inferenceTime = 0;
+    // Only for object detection
+    protected Vector<String> wordLabels = new Vector<String>();
+    protected int detLongSize = 960;
+    protected float scoreThreshold = 0.1f;
+    protected Bitmap inputImage = null;
+    protected Bitmap outputImage = null;
+    protected volatile String outputResult = "";
+    protected float postprocessTime = 0;
+
+
+    public Predictor() {
+    }
+
+    public boolean init(Context appCtx, String modelPath, String labelPath, int useOpencl, int cpuThreadNum, String cpuPowerMode) {
+        isLoaded = loadModel(appCtx, modelPath, useOpencl, cpuThreadNum, cpuPowerMode);
+        if (!isLoaded) {
+            return false;
+        }
+        isLoaded = loadLabel(appCtx, labelPath);
+        return isLoaded;
+    }
+
+
+    public boolean init(Context appCtx, String modelPath, String labelPath, int useOpencl, int cpuThreadNum, String cpuPowerMode,
+                        int detLongSize, float scoreThreshold) {
+        boolean isLoaded = init(appCtx, modelPath, labelPath, useOpencl, cpuThreadNum, cpuPowerMode);
+        if (!isLoaded) {
+            return false;
+        }
+        this.detLongSize = detLongSize;
+        this.scoreThreshold = scoreThreshold;
+        return true;
+    }
+
+    protected boolean loadModel(Context appCtx, String modelPath, int useOpencl, int cpuThreadNum, String cpuPowerMode) {
+        // Release model if exists
+        releaseModel();
+
+        // Load model
+        if (modelPath.isEmpty()) {
+            return false;
+        }
+        String realPath = modelPath;
+        if (!modelPath.substring(0, 1).equals("/")) {
+            // Read model files from custom path if the first character of mode path is '/'
+            // otherwise copy model to cache from assets
+            realPath = appCtx.getCacheDir() + "/" + modelPath;
+            Utils.copyDirectoryFromAssets(appCtx, modelPath, realPath);
+        }
+        if (realPath.isEmpty()) {
+            return false;
+        }
+
+        OCRPredictorNative.Config config = new OCRPredictorNative.Config();
+        config.useOpencl = useOpencl;
+        config.cpuThreadNum = cpuThreadNum;
+        config.cpuPower = cpuPowerMode;
+        config.detModelFilename = realPath + File.separator + "det_db.nb";
+        config.recModelFilename = realPath + File.separator + "rec_crnn.nb";
+        config.clsModelFilename = realPath + File.separator + "cls.nb";
+        Log.i("Predictor", "model path" + config.detModelFilename + " ; " + config.recModelFilename + ";" + config.clsModelFilename);
+        paddlePredictor = new OCRPredictorNative(config);
+
+        this.cpuThreadNum = cpuThreadNum;
+        this.cpuPowerMode = cpuPowerMode;
+        this.modelPath = realPath;
+        this.modelName = realPath.substring(realPath.lastIndexOf("/") + 1);
+        return true;
+    }
+
+    public void releaseModel() {
+        if (paddlePredictor != null) {
+            paddlePredictor.destroy();
+            paddlePredictor = null;
+        }
+        isLoaded = false;
+        cpuThreadNum = 1;
+        cpuPowerMode = "LITE_POWER_HIGH";
+        modelPath = "";
+        modelName = "";
+    }
+
+    protected boolean loadLabel(Context appCtx, String labelPath) {
+        wordLabels.clear();
+        wordLabels.add("black");
+        // Load word labels from file
+        try {
+            InputStream assetsInputStream = appCtx.getAssets().open(labelPath);
+            int available = assetsInputStream.available();
+            byte[] lines = new byte[available];
+            assetsInputStream.read(lines);
+            assetsInputStream.close();
+            String words = new String(lines);
+            String[] contents = words.split("\n");
+            for (String content : contents) {
+                wordLabels.add(content);
+            }
+            wordLabels.add(" ");
+            Log.i(TAG, "Word label size: " + wordLabels.size());
+        } catch (Exception e) {
+            Log.e(TAG, e.getMessage());
+            return false;
+        }
+        return true;
+    }
+
+
+    public boolean runModel(int run_det, int run_cls, int run_rec) {
+        if (inputImage == null || !isLoaded()) {
+            return false;
+        }
+
+        // Warm up
+        for (int i = 0; i < warmupIterNum; i++) {
+            paddlePredictor.runImage(inputImage, detLongSize, run_det, run_cls, run_rec);
+        }
+        warmupIterNum = 0; // do not need warm
+        // Run inference
+        Date start = new Date();
+        ArrayList<OcrResultModel> results = paddlePredictor.runImage(inputImage, detLongSize, run_det, run_cls, run_rec);
+        Date end = new Date();
+        inferenceTime = (end.getTime() - start.getTime()) / (float) inferIterNum;
+
+        results = postprocess(results);
+        Log.i(TAG, "[stat] Inference Time: " + inferenceTime + " ;Box Size " + results.size());
+        drawResults(results);
+
+        return true;
+    }
+
+    public boolean isLoaded() {
+        return paddlePredictor != null && isLoaded;
+    }
+
+    public String modelPath() {
+        return modelPath;
+    }
+
+    public String modelName() {
+        return modelName;
+    }
+
+    public int cpuThreadNum() {
+        return cpuThreadNum;
+    }
+
+    public String cpuPowerMode() {
+        return cpuPowerMode;
+    }
+
+    public float inferenceTime() {
+        return inferenceTime;
+    }
+
+    public Bitmap inputImage() {
+        return inputImage;
+    }
+
+    public Bitmap outputImage() {
+        return outputImage;
+    }
+
+    public String outputResult() {
+        return outputResult;
+    }
+
+    public float postprocessTime() {
+        return postprocessTime;
+    }
+
+
+    public void setInputImage(Bitmap image) {
+        if (image == null) {
+            return;
+        }
+        this.inputImage = image.copy(Bitmap.Config.ARGB_8888, true);
+    }
+
+    private ArrayList<OcrResultModel> postprocess(ArrayList<OcrResultModel> results) {
+        for (OcrResultModel r : results) {
+            StringBuffer word = new StringBuffer();
+            for (int index : r.getWordIndex()) {
+                if (index >= 0 && index < wordLabels.size()) {
+                    word.append(wordLabels.get(index));
+                } else {
+                    Log.e(TAG, "Word index is not in label list:" + index);
+                    word.append("×");
+                }
+            }
+            r.setLabel(word.toString());
+            r.setClsLabel(r.getClsIdx() == 1 ? "180" : "0");
+        }
+        return results;
+    }
+
+    private void drawResults(ArrayList<OcrResultModel> results) {
+        StringBuffer outputResultSb = new StringBuffer("");
+        for (int i = 0; i < results.size(); i++) {
+            OcrResultModel result = results.get(i);
+            StringBuilder sb = new StringBuilder("");
+            if(result.getPoints().size()>0){
+                sb.append("Det: ");
+                for (Point p : result.getPoints()) {
+                    sb.append("(").append(p.x).append(",").append(p.y).append(") ");
+                }
+            }
+            if(result.getLabel().length() > 0){
+                sb.append("\n Rec: ").append(result.getLabel());
+                sb.append(",").append(result.getConfidence());
+            }
+            if(result.getClsIdx()!=-1){
+                sb.append(" Cls: ").append(result.getClsLabel());
+                sb.append(",").append(result.getClsConfidence());
+            }
+            Log.i(TAG, sb.toString()); // show LOG in Logcat panel
+            outputResultSb.append(i + 1).append(": ").append(sb.toString()).append("\n");
+        }
+        outputResult = outputResultSb.toString();
+        outputImage = inputImage;
+        Canvas canvas = new Canvas(outputImage);
+        Paint paintFillAlpha = new Paint();
+        paintFillAlpha.setStyle(Paint.Style.FILL);
+        paintFillAlpha.setColor(Color.parseColor("#3B85F5"));
+        paintFillAlpha.setAlpha(50);
+
+        Paint paint = new Paint();
+        paint.setColor(Color.parseColor("#3B85F5"));
+        paint.setStrokeWidth(5);
+        paint.setStyle(Paint.Style.STROKE);
+
+        for (OcrResultModel result : results) {
+            Path path = new Path();
+            List<Point> points = result.getPoints();
+            if(points.size()==0){
+                continue;
+            }
+            path.moveTo(points.get(0).x, points.get(0).y);
+            for (int i = points.size() - 1; i >= 0; i--) {
+                Point p = points.get(i);
+                path.lineTo(p.x, p.y);
+            }
+            canvas.drawPath(path, paint);
+            canvas.drawPath(path, paintFillAlpha);
+        }
+    }
+
+}
diff --git a/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/SettingsActivity.java b/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/SettingsActivity.java
new file mode 100644
index 0000000..477cd5d
--- /dev/null
+++ b/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/SettingsActivity.java
@@ -0,0 +1,172 @@
+package com.baidu.paddle.lite.demo.ocr;
+
+import android.content.SharedPreferences;
+import android.os.Bundle;
+import android.preference.CheckBoxPreference;
+import android.preference.EditTextPreference;
+import android.preference.ListPreference;
+
+import androidx.appcompat.app.ActionBar;
+
+import java.util.ArrayList;
+import java.util.List;
+
+
+public class SettingsActivity extends AppCompatPreferenceActivity implements SharedPreferences.OnSharedPreferenceChangeListener {
+    ListPreference lpChoosePreInstalledModel = null;
+    CheckBoxPreference cbEnableCustomSettings = null;
+    EditTextPreference etModelPath = null;
+    EditTextPreference etLabelPath = null;
+    ListPreference etImagePath = null;
+    ListPreference lpCPUThreadNum = null;
+    ListPreference lpCPUPowerMode = null;
+    EditTextPreference etDetLongSize = null;
+    EditTextPreference etScoreThreshold = null;
+
+    List<String> preInstalledModelPaths = null;
+    List<String> preInstalledLabelPaths = null;
+    List<String> preInstalledImagePaths = null;
+    List<String> preInstalledDetLongSizes = null;
+    List<String> preInstalledCPUThreadNums = null;
+    List<String> preInstalledCPUPowerModes = null;
+    List<String> preInstalledInputColorFormats = null;
+    List<String> preInstalledInputMeans = null;
+    List<String> preInstalledInputStds = null;
+    List<String> preInstalledScoreThresholds = null;
+
+    @Override
+    public void onCreate(Bundle savedInstanceState) {
+        super.onCreate(savedInstanceState);
+        addPreferencesFromResource(R.xml.settings);
+        ActionBar supportActionBar = getSupportActionBar();
+        if (supportActionBar != null) {
+            supportActionBar.setDisplayHomeAsUpEnabled(true);
+        }
+
+        // Initialized pre-installed models
+        preInstalledModelPaths = new ArrayList<String>();
+        preInstalledLabelPaths = new ArrayList<String>();
+        preInstalledImagePaths = new ArrayList<String>();
+        preInstalledDetLongSizes = new ArrayList<String>();
+        preInstalledCPUThreadNums = new ArrayList<String>();
+        preInstalledCPUPowerModes = new ArrayList<String>();
+        preInstalledInputColorFormats = new ArrayList<String>();
+        preInstalledInputMeans = new ArrayList<String>();
+        preInstalledInputStds = new ArrayList<String>();
+        preInstalledScoreThresholds = new ArrayList<String>();
+        // Add ssd_mobilenet_v1_pascalvoc_for_cpu
+        preInstalledModelPaths.add(getString(R.string.MODEL_PATH_DEFAULT));
+        preInstalledLabelPaths.add(getString(R.string.LABEL_PATH_DEFAULT));
+        preInstalledImagePaths.add(getString(R.string.IMAGE_PATH_DEFAULT));
+        preInstalledCPUThreadNums.add(getString(R.string.CPU_THREAD_NUM_DEFAULT));
+        preInstalledCPUPowerModes.add(getString(R.string.CPU_POWER_MODE_DEFAULT));
+        preInstalledDetLongSizes.add(getString(R.string.DET_LONG_SIZE_DEFAULT));
+        preInstalledScoreThresholds.add(getString(R.string.SCORE_THRESHOLD_DEFAULT));
+
+        // Setup UI components
+        lpChoosePreInstalledModel =
+                (ListPreference) findPreference(getString(R.string.CHOOSE_PRE_INSTALLED_MODEL_KEY));
+        String[] preInstalledModelNames = new String[preInstalledModelPaths.size()];
+        for (int i = 0; i < preInstalledModelPaths.size(); i++) {
+            preInstalledModelNames[i] =
+                    preInstalledModelPaths.get(i).substring(preInstalledModelPaths.get(i).lastIndexOf("/") + 1);
+        }
+        lpChoosePreInstalledModel.setEntries(preInstalledModelNames);
+        lpChoosePreInstalledModel.setEntryValues(preInstalledModelPaths.toArray(new String[preInstalledModelPaths.size()]));
+        cbEnableCustomSettings =
+                (CheckBoxPreference) findPreference(getString(R.string.ENABLE_CUSTOM_SETTINGS_KEY));
+        etModelPath = (EditTextPreference) findPreference(getString(R.string.MODEL_PATH_KEY));
+        etModelPath.setTitle("Model Path (SDCard: " + Utils.getSDCardDirectory() + ")");
+        etLabelPath = (EditTextPreference) findPreference(getString(R.string.LABEL_PATH_KEY));
+        etImagePath = (ListPreference) findPreference(getString(R.string.IMAGE_PATH_KEY));
+        lpCPUThreadNum =
+                (ListPreference) findPreference(getString(R.string.CPU_THREAD_NUM_KEY));
+        lpCPUPowerMode =
+                (ListPreference) findPreference(getString(R.string.CPU_POWER_MODE_KEY));
+        etDetLongSize = (EditTextPreference) findPreference(getString(R.string.DET_LONG_SIZE_KEY));
+        etScoreThreshold = (EditTextPreference) findPreference(getString(R.string.SCORE_THRESHOLD_KEY));
+    }
+
+    private void reloadPreferenceAndUpdateUI() {
+        SharedPreferences sharedPreferences = getPreferenceScreen().getSharedPreferences();
+        boolean enableCustomSettings =
+                sharedPreferences.getBoolean(getString(R.string.ENABLE_CUSTOM_SETTINGS_KEY), false);
+        String modelPath = sharedPreferences.getString(getString(R.string.CHOOSE_PRE_INSTALLED_MODEL_KEY),
+                getString(R.string.MODEL_PATH_DEFAULT));
+        int modelIdx = lpChoosePreInstalledModel.findIndexOfValue(modelPath);
+        if (modelIdx >= 0 && modelIdx < preInstalledModelPaths.size()) {
+            if (!enableCustomSettings) {
+                SharedPreferences.Editor editor = sharedPreferences.edit();
+                editor.putString(getString(R.string.MODEL_PATH_KEY), preInstalledModelPaths.get(modelIdx));
+                editor.putString(getString(R.string.LABEL_PATH_KEY), preInstalledLabelPaths.get(modelIdx));
+                editor.putString(getString(R.string.IMAGE_PATH_KEY), preInstalledImagePaths.get(modelIdx));
+                editor.putString(getString(R.string.CPU_THREAD_NUM_KEY), preInstalledCPUThreadNums.get(modelIdx));
+                editor.putString(getString(R.string.CPU_POWER_MODE_KEY), preInstalledCPUPowerModes.get(modelIdx));
+                editor.putString(getString(R.string.DET_LONG_SIZE_KEY), preInstalledDetLongSizes.get(modelIdx));
+                editor.putString(getString(R.string.SCORE_THRESHOLD_KEY),
+                        preInstalledScoreThresholds.get(modelIdx));
+                editor.apply();
+            }
+            lpChoosePreInstalledModel.setSummary(modelPath);
+        }
+        cbEnableCustomSettings.setChecked(enableCustomSettings);
+        etModelPath.setEnabled(enableCustomSettings);
+        etLabelPath.setEnabled(enableCustomSettings);
+        etImagePath.setEnabled(enableCustomSettings);
+        lpCPUThreadNum.setEnabled(enableCustomSettings);
+        lpCPUPowerMode.setEnabled(enableCustomSettings);
+        etDetLongSize.setEnabled(enableCustomSettings);
+        etScoreThreshold.setEnabled(enableCustomSettings);
+        modelPath = sharedPreferences.getString(getString(R.string.MODEL_PATH_KEY),
+                getString(R.string.MODEL_PATH_DEFAULT));
+        String labelPath = sharedPreferences.getString(getString(R.string.LABEL_PATH_KEY),
+                getString(R.string.LABEL_PATH_DEFAULT));
+        String imagePath = sharedPreferences.getString(getString(R.string.IMAGE_PATH_KEY),
+                getString(R.string.IMAGE_PATH_DEFAULT));
+        String cpuThreadNum = sharedPreferences.getString(getString(R.string.CPU_THREAD_NUM_KEY),
+                getString(R.string.CPU_THREAD_NUM_DEFAULT));
+        String cpuPowerMode = sharedPreferences.getString(getString(R.string.CPU_POWER_MODE_KEY),
+                getString(R.string.CPU_POWER_MODE_DEFAULT));
+        String detLongSize = sharedPreferences.getString(getString(R.string.DET_LONG_SIZE_KEY),
+                getString(R.string.DET_LONG_SIZE_DEFAULT));
+        String scoreThreshold = sharedPreferences.getString(getString(R.string.SCORE_THRESHOLD_KEY),
+                getString(R.string.SCORE_THRESHOLD_DEFAULT));
+        etModelPath.setSummary(modelPath);
+        etModelPath.setText(modelPath);
+        etLabelPath.setSummary(labelPath);
+        etLabelPath.setText(labelPath);
+        etImagePath.setSummary(imagePath);
+        etImagePath.setValue(imagePath);
+        lpCPUThreadNum.setValue(cpuThreadNum);
+        lpCPUThreadNum.setSummary(cpuThreadNum);
+        lpCPUPowerMode.setValue(cpuPowerMode);
+        lpCPUPowerMode.setSummary(cpuPowerMode);
+        etDetLongSize.setSummary(detLongSize);
+        etDetLongSize.setText(detLongSize);
+        etScoreThreshold.setText(scoreThreshold);
+        etScoreThreshold.setSummary(scoreThreshold);
+    }
+
+    @Override
+    protected void onResume() {
+        super.onResume();
+        getPreferenceScreen().getSharedPreferences().registerOnSharedPreferenceChangeListener(this);
+        reloadPreferenceAndUpdateUI();
+    }
+
+    @Override
+    protected void onPause() {
+        super.onPause();
+        getPreferenceScreen().getSharedPreferences().unregisterOnSharedPreferenceChangeListener(this);
+    }
+
+    @Override
+    public void onSharedPreferenceChanged(SharedPreferences sharedPreferences, String key) {
+        if (key.equals(getString(R.string.CHOOSE_PRE_INSTALLED_MODEL_KEY))) {
+            SharedPreferences.Editor editor = sharedPreferences.edit();
+            editor.putBoolean(getString(R.string.ENABLE_CUSTOM_SETTINGS_KEY), false);
+            editor.commit();
+        }
+        reloadPreferenceAndUpdateUI();
+    }
+}
diff --git a/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/Utils.java b/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/Utils.java
new file mode 100644
index 0000000..ef46805
--- /dev/null
+++ b/deploy/android_demo/app/src/main/java/com/baidu/paddle/lite/demo/ocr/Utils.java
@@ -0,0 +1,159 @@
+package com.baidu.paddle.lite.demo.ocr;
+
+import android.content.Context;
+import android.graphics.Bitmap;
+import android.graphics.Matrix;
+import android.media.ExifInterface;
+import android.os.Environment;
+
+import java.io.*;
+
+public class Utils {
+    private static final String TAG = Utils.class.getSimpleName();
+
+    public static void copyFileFromAssets(Context appCtx, String srcPath, String dstPath) {
+        if (srcPath.isEmpty() || dstPath.isEmpty()) {
+            return;
+        }
+        InputStream is = null;
+        OutputStream os = null;
+        try {
+            is = new BufferedInputStream(appCtx.getAssets().open(srcPath));
+            os = new BufferedOutputStream(new FileOutputStream(new File(dstPath)));
+            byte[] buffer = new byte[1024];
+            int length = 0;
+            while ((length = is.read(buffer)) != -1) {
+                os.write(buffer, 0, length);
+            }
+        } catch (FileNotFoundException e) {
+            e.printStackTrace();
+        } catch (IOException e) {
+            e.printStackTrace();
+        } finally {
+            try {
+                os.close();
+                is.close();
+            } catch (IOException e) {
+                e.printStackTrace();
+            }
+        }
+    }
+
+    public static void copyDirectoryFromAssets(Context appCtx, String srcDir, String dstDir) {
+        if (srcDir.isEmpty() || dstDir.isEmpty()) {
+            return;
+        }
+        try {
+            if (!new File(dstDir).exists()) {
+                new File(dstDir).mkdirs();
+            }
+            for (String fileName : appCtx.getAssets().list(srcDir)) {
+                String srcSubPath = srcDir + File.separator + fileName;
+                String dstSubPath = dstDir + File.separator + fileName;
+                if (new File(srcSubPath).isDirectory()) {
+                    copyDirectoryFromAssets(appCtx, srcSubPath, dstSubPath);
+                } else {
+                    copyFileFromAssets(appCtx, srcSubPath, dstSubPath);
+                }
+            }
+        } catch (Exception e) {
+            e.printStackTrace();
+        }
+    }
+
+    public static float[] parseFloatsFromString(String string, String delimiter) {
+        String[] pieces = string.trim().toLowerCase().split(delimiter);
+        float[] floats = new float[pieces.length];
+        for (int i = 0; i < pieces.length; i++) {
+            floats[i] = Float.parseFloat(pieces[i].trim());
+        }
+        return floats;
+    }
+
+    public static long[] parseLongsFromString(String string, String delimiter) {
+        String[] pieces = string.trim().toLowerCase().split(delimiter);
+        long[] longs = new long[pieces.length];
+        for (int i = 0; i < pieces.length; i++) {
+            longs[i] = Long.parseLong(pieces[i].trim());
+        }
+        return longs;
+    }
+
+    public static String getSDCardDirectory() {
+        return Environment.getExternalStorageDirectory().getAbsolutePath();
+    }
+
+    public static boolean isSupportedNPU() {
+        return false;
+        // String hardware = android.os.Build.HARDWARE;
+        // return hardware.equalsIgnoreCase("kirin810") || hardware.equalsIgnoreCase("kirin990");
+    }
+
+    public static Bitmap resizeWithStep(Bitmap bitmap, int maxLength, int step) {
+        int width = bitmap.getWidth();
+        int height = bitmap.getHeight();
+        int maxWH = Math.max(width, height);
+        float ratio = 1;
+        int newWidth = width;
+        int newHeight = height;
+        if (maxWH > maxLength) {
+            ratio = maxLength * 1.0f / maxWH;
+            newWidth = (int) Math.floor(ratio * width);
+            newHeight = (int) Math.floor(ratio * height);
+        }
+
+        newWidth = newWidth - newWidth % step;
+        if (newWidth == 0) {
+            newWidth = step;
+        }
+        newHeight = newHeight - newHeight % step;
+        if (newHeight == 0) {
+            newHeight = step;
+        }
+        return Bitmap.createScaledBitmap(bitmap, newWidth, newHeight, true);
+    }
+
+    public static Bitmap rotateBitmap(Bitmap bitmap, int orientation) {
+
+        Matrix matrix = new Matrix();
+        switch (orientation) {
+            case ExifInterface.ORIENTATION_NORMAL:
+                return bitmap;
+            case ExifInterface.ORIENTATION_FLIP_HORIZONTAL:
+                matrix.setScale(-1, 1);
+                break;
+            case ExifInterface.ORIENTATION_ROTATE_180:
+                matrix.setRotate(180);
+                break;
+            case ExifInterface.ORIENTATION_FLIP_VERTICAL:
+                matrix.setRotate(180);
+                matrix.postScale(-1, 1);
+                break;
+            case ExifInterface.ORIENTATION_TRANSPOSE:
+                matrix.setRotate(90);
+                matrix.postScale(-1, 1);
+                break;
+            case ExifInterface.ORIENTATION_ROTATE_90:
+                matrix.setRotate(90);
+                break;
+            case ExifInterface.ORIENTATION_TRANSVERSE:
+                matrix.setRotate(-90);
+                matrix.postScale(-1, 1);
+                break;
+            case ExifInterface.ORIENTATION_ROTATE_270:
+                matrix.setRotate(-90);
+                break;
+            default:
+                return bitmap;
+        }
+        try {
+            Bitmap bmRotated = Bitmap.createBitmap(bitmap, 0, 0, bitmap.getWidth(), bitmap.getHeight(), matrix, true);
+            bitmap.recycle();
+            return bmRotated;
+        }
+        catch (OutOfMemoryError e) {
+            e.printStackTrace();
+            return null;
+        }
+    }
+}
diff --git a/deploy/android_demo/app/src/main/res/drawable-v24/ic_launcher_foreground.xml b/deploy/android_demo/app/src/main/res/drawable-v24/ic_launcher_foreground.xml
new file mode 100644
index 0000000..1f6bb29
--- /dev/null
+++ b/deploy/android_demo/app/src/main/res/drawable-v24/ic_launcher_foreground.xml
@@ -0,0 +1,34 @@
+<vector xmlns:android="http://schemas.android.com/apk/res/android"
+    xmlns:aapt="http://schemas.android.com/aapt"
+    android:width="108dp"
+    android:height="108dp"
+    android:viewportWidth="108"
+    android:viewportHeight="108">
+    <path
+        android:fillType="evenOdd"
+        android:pathData="M32,64C32,64 38.39,52.99 44.13,50.95C51.37,48.37 70.14,49.57 70.14,49.57L108.26,87.69L108,109.01L75.97,107.97L32,64Z"
+        android:strokeWidth="1"
+        android:strokeColor="#00000000">
+        <aapt:attr name="android:fillColor">
+            <gradient
+                android:endX="78.5885"
+                android:endY="90.9159"
+                android:startX="48.7653"
+                android:startY="61.0927"
+                android:type="linear">
+                <item
+                    android:color="#44000000"
+                    android:offset="0.0" />
+                <item
+                    android:color="#00000000"
+                    android:offset="1.0" />
+            </gradient>
+        </aapt:attr>
+    </path>
+    <path
+        android:fillColor="#FFFFFF"
+        android:fillType="nonZero"
+        android:pathData="M66.94,46.02L66.94,46.02C72.44,50.07 76,56.61 76,64L32,64C32,56.61 35.56,50.11 40.98,46.06L36.18,41.19C35.45,40.45 35.45,39.3 36.18,38.56C36.91,37.81 38.05,37.81 38.78,38.56L44.25,44.05C47.18,42.57 50.48,41.71 54,41.71C57.48,41.71 60.78,42.57 63.68,44.05L69.11,38.56C69.84,37.81 70.98,37.81 71.71,38.56C72.44,39.3 72.44,40.45 71.71,41.19L66.94,46.02ZM62.94,56.92C64.08,56.92 65,56.01 65,54.88C65,53.76 64.08,52.85 62.94,52.85C61.8,52.85 60.88,53.76 60.88,54.88C60.88,56.01 61.8,56.92 62.94,56.92ZM45.06,56.92C46.2,56.92 47.13,56.01 47.13,54.88C47.13,53.76 46.2,52.85 45.06,52.85C43.92,52.85 43,53.76 43,54.88C43,56.01 43.92,56.92 45.06,56.92Z"
+        android:strokeWidth="1"
+        android:strokeColor="#00000000" />
+</vector>
diff --git a/deploy/android_demo/app/src/main/res/drawable/ic_launcher_background.xml b/deploy/android_demo/app/src/main/res/drawable/ic_launcher_background.xml
new file mode 100644
index 0000000..0d025f9
--- /dev/null
+++ b/deploy/android_demo/app/src/main/res/drawable/ic_launcher_background.xml
@@ -0,0 +1,170 @@
+<?xml version="1.0" encoding="utf-8"?>
+<vector xmlns:android="http://schemas.android.com/apk/res/android"
+    android:width="108dp"
+    android:height="108dp"
+    android:viewportWidth="108"
+    android:viewportHeight="108">
+    <path
+        android:fillColor="#008577"
+        android:pathData="M0,0h108v108h-108z" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M9,0L9,108"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M19,0L19,108"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M29,0L29,108"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M39,0L39,108"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M49,0L49,108"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M59,0L59,108"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M69,0L69,108"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M79,0L79,108"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M89,0L89,108"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M99,0L99,108"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M0,9L108,9"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M0,19L108,19"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M0,29L108,29"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M0,39L108,39"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M0,49L108,49"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M0,59L108,59"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M0,69L108,69"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M0,79L108,79"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M0,89L108,89"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M0,99L108,99"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M19,29L89,29"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M19,39L89,39"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M19,49L89,49"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M19,59L89,59"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M19,69L89,69"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M19,79L89,79"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M29,19L29,89"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M39,19L39,89"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M49,19L49,89"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M59,19L59,89"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M69,19L69,89"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+    <path
+        android:fillColor="#00000000"
+        android:pathData="M79,19L79,89"
+        android:strokeWidth="0.8"
+        android:strokeColor="#33FFFFFF" />
+</vector>
diff --git a/deploy/android_demo/app/src/main/res/layout/activity_main.xml b/deploy/android_demo/app/src/main/res/layout/activity_main.xml
new file mode 100644
index 0000000..e90c99a
--- /dev/null
+++ b/deploy/android_demo/app/src/main/res/layout/activity_main.xml
@@ -0,0 +1,180 @@
+<?xml version="1.0" encoding="utf-8"?>
+<androidx.constraintlayout.widget.ConstraintLayout xmlns:android="http://schemas.android.com/apk/res/android"
+                                             xmlns:app="http://schemas.android.com/apk/res-auto"
+                                             xmlns:tools="http://schemas.android.com/tools"
+                                             android:layout_width="match_parent"
+                                             android:layout_height="match_parent"
+                                             tools:context=".MainActivity">
+
+    <RelativeLayout
+            android:layout_width="match_parent"
+            android:layout_height="match_parent">
+
+        <LinearLayout
+                android:id="@+id/v_input_info"
+                android:layout_width="fill_parent"
+                android:layout_height="wrap_content"
+                android:layout_alignParentTop="true"
+                android:orientation="vertical">
+
+            <LinearLayout
+                android:id="@+id/btn_layout"
+                android:layout_width="fill_parent"
+                android:layout_height="wrap_content"
+                android:orientation="horizontal">
+
+
+                <Button
+                    android:id="@+id/btn_run_model"
+                    android:layout_width="0dp"
+                    android:layout_height="wrap_content"
+                    android:layout_weight="1"
+                    android:onClick="btn_run_model_click"
+                    android:text="运行模型" />
+                <Button
+                    android:id="@+id/btn_take_photo"
+                    android:layout_width="0dp"
+                    android:layout_height="wrap_content"
+                    android:layout_weight="1"
+                    android:onClick="btn_take_photo_click"
+                    android:text="拍照识别" />
+                <Button
+                    android:id="@+id/btn_choice_img"
+                    android:layout_width="0dp"
+                    android:layout_height="wrap_content"
+                    android:layout_weight="1"
+                    android:onClick="btn_choice_img_click"
+                    android:text="选取图片" />
+
+                <Button
+                    android:id="@+id/btn_reset_img"
+                    android:layout_width="0dp"
+                    android:layout_height="wrap_content"
+                    android:layout_weight="1"
+                    android:onClick="btn_reset_img_click"
+                    android:text="清空绘图" />
+
+            </LinearLayout>
+            <LinearLayout
+                android:id="@+id/run_mode_layout"
+                android:layout_width="fill_parent"
+                android:layout_height="wrap_content"
+                android:orientation="horizontal">
+
+                <CheckBox
+                    android:id="@+id/cb_opencl"
+                    android:layout_width="0dp"
+                    android:layout_weight="1"
+                    android:layout_height="wrap_content"
+                    android:text="开启OPENCL"
+                    android:onClick="cb_opencl_click"
+                    android:visibility="gone"/>
+                <TextView
+                    android:layout_width="0dp"
+                    android:layout_weight="0.5"
+                    android:layout_height="wrap_content"
+                    android:text="运行模式:"/>
+
+                <Spinner
+                    android:id="@+id/sp_run_mode"
+                    android:layout_width="0dp"
+                    android:layout_weight="1.5"
+                    android:layout_height="wrap_content"
+                    android:entries="@array/run_Model"
+                    />
+
+            </LinearLayout>
+
+            <TextView
+                    android:id="@+id/tv_input_setting"
+                    android:layout_width="wrap_content"
+                    android:layout_height="wrap_content"
+                    android:scrollbars="vertical"
+                    android:layout_marginLeft="12dp"
+                    android:layout_marginRight="12dp"
+                    android:layout_marginTop="5dp"
+                    android:layout_marginBottom="5dp"
+                    android:lineSpacingExtra="4dp"
+                    android:singleLine="false"
+                    android:maxLines="6"
+                    android:text=""/>
+            <TextView
+                android:id="@+id/tv_model_img_status"
+                android:layout_width="wrap_content"
+                android:layout_height="wrap_content"
+                android:scrollbars="vertical"
+                android:layout_marginLeft="12dp"
+                android:layout_marginRight="12dp"
+                android:layout_marginTop="-5dp"
+                android:layout_marginBottom="5dp"
+                android:lineSpacingExtra="4dp"
+                android:singleLine="false"
+                android:maxLines="6"
+                android:text="STATUS: ok"/>
+
+        </LinearLayout>
+
+        <RelativeLayout
+                android:layout_width="match_parent"
+                android:layout_height="match_parent"
+                android:layout_above="@+id/v_output_info"
+                android:layout_below="@+id/v_input_info">
+
+            <ImageView
+                    android:id="@+id/iv_input_image"
+                    android:layout_width="400dp"
+                    android:layout_height="400dp"
+                    android:layout_centerHorizontal="true"
+                    android:layout_centerVertical="true"
+                    android:layout_marginLeft="12dp"
+                    android:layout_marginRight="12dp"
+                    android:layout_marginTop="5dp"
+                    android:layout_marginBottom="5dp"
+                    android:adjustViewBounds="true"
+                    android:scaleType="fitCenter"/>
+        </RelativeLayout>
+
+
+        <RelativeLayout
+                android:id="@+id/v_output_info"
+                android:layout_width="wrap_content"
+                android:layout_height="wrap_content"
+                android:layout_alignParentBottom="true"
+                android:layout_centerHorizontal="true">
+
+            <TextView
+                    android:id="@+id/tv_output_result"
+                    android:layout_width="wrap_content"
+                    android:layout_height="wrap_content"
+                    android:layout_alignParentTop="true"
+                    android:layout_centerHorizontal="true"
+                    android:layout_centerVertical="true"
+                    android:scrollbars="vertical"
+                    android:layout_marginLeft="12dp"
+                    android:layout_marginRight="12dp"
+                    android:layout_marginTop="5dp"
+                    android:layout_marginBottom="5dp"
+                    android:textAlignment="center"
+                    android:lineSpacingExtra="5dp"
+                    android:singleLine="false"
+                    android:maxLines="5"
+                    android:text=""/>
+
+            <TextView
+                    android:id="@+id/tv_inference_time"
+                    android:layout_width="wrap_content"
+                    android:layout_height="wrap_content"
+                    android:layout_below="@+id/tv_output_result"
+                    android:layout_centerHorizontal="true"
+                    android:layout_centerVertical="true"
+                    android:textAlignment="center"
+                    android:layout_marginLeft="12dp"
+                    android:layout_marginRight="12dp"
+                    android:layout_marginTop="5dp"
+                    android:layout_marginBottom="10dp"
+                    android:text=""/>
+
+        </RelativeLayout>
+
+    </RelativeLayout>
+</androidx.constraintlayout.widget.ConstraintLayout>
diff --git a/deploy/android_demo/app/src/main/res/menu/menu_action_options.xml b/deploy/android_demo/app/src/main/res/menu/menu_action_options.xml
new file mode 100644
index 0000000..049bd7e
--- /dev/null
+++ b/deploy/android_demo/app/src/main/res/menu/menu_action_options.xml
@@ -0,0 +1,10 @@
+<menu xmlns:android="http://schemas.android.com/apk/res/android"
+      xmlns:app="http://schemas.android.com/apk/res-auto">
+
+    <group>
+        <item
+                android:id="@+id/settings"
+                android:title="Settings..."
+                app:showAsAction="withText"/>
+    </group>
+</menu>
diff --git a/deploy/android_demo/app/src/main/res/mipmap-anydpi-v26/ic_launcher.xml b/deploy/android_demo/app/src/main/res/mipmap-anydpi-v26/ic_launcher.xml
new file mode 100644
index 0000000..6b78462
--- /dev/null
+++ b/deploy/android_demo/app/src/main/res/mipmap-anydpi-v26/ic_launcher.xml
@@ -0,0 +1,5 @@
+<?xml version="1.0" encoding="utf-8"?>
+<adaptive-icon xmlns:android="http://schemas.android.com/apk/res/android">
+    <background android:drawable="@drawable/ic_launcher_background" />
+    <foreground android:drawable="@drawable/ic_launcher_foreground" />
+</adaptive-icon>
diff --git a/deploy/android_demo/app/src/main/res/mipmap-anydpi-v26/ic_launcher_round.xml b/deploy/android_demo/app/src/main/res/mipmap-anydpi-v26/ic_launcher_round.xml
new file mode 100644
index 0000000..6b78462
--- /dev/null
+++ b/deploy/android_demo/app/src/main/res/mipmap-anydpi-v26/ic_launcher_round.xml
@@ -0,0 +1,5 @@
+<?xml version="1.0" encoding="utf-8"?>
+<adaptive-icon xmlns:android="http://schemas.android.com/apk/res/android">
+    <background android:drawable="@drawable/ic_launcher_background" />
+    <foreground android:drawable="@drawable/ic_launcher_foreground" />
+</adaptive-icon>
diff --git a/deploy/android_demo/app/src/main/res/mipmap-hdpi/ic_launcher.png b/deploy/android_demo/app/src/main/res/mipmap-hdpi/ic_launcher.png
new file mode 100644
index 0000000..898f3ed
Binary files /dev/null and b/deploy/android_demo/app/src/main/res/mipmap-hdpi/ic_launcher.png differ
diff --git a/deploy/android_demo/app/src/main/res/mipmap-hdpi/ic_launcher_round.png b/deploy/android_demo/app/src/main/res/mipmap-hdpi/ic_launcher_round.png
new file mode 100644
index 0000000..dffca36
Binary files /dev/null and b/deploy/android_demo/app/src/main/res/mipmap-hdpi/ic_launcher_round.png differ
diff --git a/deploy/android_demo/app/src/main/res/mipmap-mdpi/ic_launcher.png b/deploy/android_demo/app/src/main/res/mipmap-mdpi/ic_launcher.png
new file mode 100644
index 0000000..64ba76f
Binary files /dev/null and b/deploy/android_demo/app/src/main/res/mipmap-mdpi/ic_launcher.png differ
diff --git a/deploy/android_demo/app/src/main/res/mipmap-mdpi/ic_launcher_round.png b/deploy/android_demo/app/src/main/res/mipmap-mdpi/ic_launcher_round.png
new file mode 100644
index 0000000..dae5e08
Binary files /dev/null and b/deploy/android_demo/app/src/main/res/mipmap-mdpi/ic_launcher_round.png differ
diff --git a/deploy/android_demo/app/src/main/res/mipmap-xhdpi/ic_launcher.png b/deploy/android_demo/app/src/main/res/mipmap-xhdpi/ic_launcher.png
new file mode 100644
index 0000000..e5ed465
Binary files /dev/null and b/deploy/android_demo/app/src/main/res/mipmap-xhdpi/ic_launcher.png differ
diff --git a/deploy/android_demo/app/src/main/res/mipmap-xhdpi/ic_launcher_round.png b/deploy/android_demo/app/src/main/res/mipmap-xhdpi/ic_launcher_round.png
new file mode 100644
index 0000000..14ed0af
Binary files /dev/null and b/deploy/android_demo/app/src/main/res/mipmap-xhdpi/ic_launcher_round.png differ
diff --git a/deploy/android_demo/app/src/main/res/mipmap-xxhdpi/ic_launcher.png b/deploy/android_demo/app/src/main/res/mipmap-xxhdpi/ic_launcher.png
new file mode 100644
index 0000000..b0907ca
Binary files /dev/null and b/deploy/android_demo/app/src/main/res/mipmap-xxhdpi/ic_launcher.png differ
diff --git a/deploy/android_demo/app/src/main/res/mipmap-xxhdpi/ic_launcher_round.png b/deploy/android_demo/app/src/main/res/mipmap-xxhdpi/ic_launcher_round.png
new file mode 100644
index 0000000..d8ae031
Binary files /dev/null and b/deploy/android_demo/app/src/main/res/mipmap-xxhdpi/ic_launcher_round.png differ
diff --git a/deploy/android_demo/app/src/main/res/mipmap-xxxhdpi/ic_launcher.png b/deploy/android_demo/app/src/main/res/mipmap-xxxhdpi/ic_launcher.png
new file mode 100644
index 0000000..2c18de9
Binary files /dev/null and b/deploy/android_demo/app/src/main/res/mipmap-xxxhdpi/ic_launcher.png differ
diff --git a/deploy/android_demo/app/src/main/res/mipmap-xxxhdpi/ic_launcher_round.png b/deploy/android_demo/app/src/main/res/mipmap-xxxhdpi/ic_launcher_round.png
new file mode 100644
index 0000000..beed3cd
Binary files /dev/null and b/deploy/android_demo/app/src/main/res/mipmap-xxxhdpi/ic_launcher_round.png differ
diff --git a/deploy/android_demo/app/src/main/res/values/arrays.xml b/deploy/android_demo/app/src/main/res/values/arrays.xml
new file mode 100644
index 0000000..9805070
--- /dev/null
+++ b/deploy/android_demo/app/src/main/res/values/arrays.xml
@@ -0,0 +1,67 @@
+<?xml version="1.0" encoding="utf-8"?>
+<resources>
+    <string-array name="image_name_entries">
+        <item>det_0.jpg</item>
+        <item>det_90.jpg</item>
+        <item>det_180.jpg</item>
+        <item>det_270.jpg</item>
+        <item>rec_0.jpg</item>
+        <item>rec_0_180.jpg</item>
+        <item>rec_1.jpg</item>
+        <item>rec_1_180.jpg</item>
+    </string-array>
+    <string-array name="image_name_values">
+        <item>images/det_0.jpg</item>
+        <item>images/det_90.jpg</item>
+        <item>images/det_180.jpg</item>
+        <item>images/det_270.jpg</item>
+        <item>images/rec_0.jpg</item>
+        <item>images/rec_0_180.jpg</item>
+        <item>images/rec_1.jpg</item>
+        <item>images/rec_1_180.jpg</item>
+    </string-array>
+    <string-array name="cpu_thread_num_entries">
+        <item>1 threads</item>
+        <item>2 threads</item>
+        <item>4 threads</item>
+        <item>8 threads</item>
+    </string-array>
+    <string-array name="cpu_thread_num_values">
+        <item>1</item>
+        <item>2</item>
+        <item>4</item>
+        <item>8</item>
+    </string-array>
+    <string-array name="cpu_power_mode_entries">
+        <item>HIGH(only big cores)</item>
+        <item>LOW(only LITTLE cores)</item>
+        <item>FULL(all cores)</item>
+        <item>NO_BIND(depends on system)</item>
+        <item>RAND_HIGH</item>
+        <item>RAND_LOW</item>
+    </string-array>
+    <string-array name="cpu_power_mode_values">
+        <item>LITE_POWER_HIGH</item>
+        <item>LITE_POWER_LOW</item>
+        <item>LITE_POWER_FULL</item>
+        <item>LITE_POWER_NO_BIND</item>
+        <item>LITE_POWER_RAND_HIGH</item>
+        <item>LITE_POWER_RAND_LOW</item>
+    </string-array>
+    <string-array name="input_color_format_entries">
+        <item>BGR color format</item>
+        <item>RGB color format</item>
+    </string-array>
+    <string-array name="input_color_format_values">
+        <item>BGR</item>
+        <item>RGB</item>
+    </string-array>
+    <string-array name="run_Model">
+        <item>检测+分类+识别</item>
+        <item>检测+识别</item>
+        <item>分类+识别</item>
+        <item>检测</item>
+        <item>识别</item>
+        <item>分类</item>
+    </string-array>
+</resources>
diff --git a/deploy/android_demo/app/src/main/res/values/colors.xml b/deploy/android_demo/app/src/main/res/values/colors.xml
new file mode 100644
index 0000000..69b2233
--- /dev/null
+++ b/deploy/android_demo/app/src/main/res/values/colors.xml
@@ -0,0 +1,6 @@
+<?xml version="1.0" encoding="utf-8"?>
+<resources>
+    <color name="colorPrimary">#008577</color>
+    <color name="colorPrimaryDark">#00574B</color>
+    <color name="colorAccent">#D81B60</color>
+</resources>
diff --git a/deploy/android_demo/app/src/main/res/values/strings.xml b/deploy/android_demo/app/src/main/res/values/strings.xml
new file mode 100644
index 0000000..38b9a78
--- /dev/null
+++ b/deploy/android_demo/app/src/main/res/values/strings.xml
@@ -0,0 +1,19 @@
+<resources>
+    <string name="app_name">PaddleOCR</string>
+    <string name="CHOOSE_PRE_INSTALLED_MODEL_KEY">CHOOSE_PRE_INSTALLED_MODEL_KEY</string>
+    <string name="ENABLE_CUSTOM_SETTINGS_KEY">ENABLE_CUSTOM_SETTINGS_KEY</string>
+    <string name="MODEL_PATH_KEY">MODEL_PATH_KEY</string>
+    <string name="LABEL_PATH_KEY">LABEL_PATH_KEY</string>
+    <string name="IMAGE_PATH_KEY">IMAGE_PATH_KEY</string>
+    <string name="CPU_THREAD_NUM_KEY">CPU_THREAD_NUM_KEY</string>
+    <string name="CPU_POWER_MODE_KEY">CPU_POWER_MODE_KEY</string>
+    <string name="DET_LONG_SIZE_KEY">DET_LONG_SIZE_KEY</string>
+    <string name="SCORE_THRESHOLD_KEY">SCORE_THRESHOLD_KEY</string>
+    <string name="MODEL_PATH_DEFAULT">models/ch_PP-OCRv2</string>
+    <string name="LABEL_PATH_DEFAULT">labels/ppocr_keys_v1.txt</string>
+    <string name="IMAGE_PATH_DEFAULT">images/det_0.jpg</string>
+    <string name="CPU_THREAD_NUM_DEFAULT">4</string>
+    <string name="CPU_POWER_MODE_DEFAULT">LITE_POWER_HIGH</string>
+    <string name="DET_LONG_SIZE_DEFAULT">960</string>
+    <string name="SCORE_THRESHOLD_DEFAULT">0.1</string>
+</resources>
diff --git a/deploy/android_demo/app/src/main/res/values/styles.xml b/deploy/android_demo/app/src/main/res/values/styles.xml
new file mode 100644
index 0000000..8532620
--- /dev/null
+++ b/deploy/android_demo/app/src/main/res/values/styles.xml
@@ -0,0 +1,25 @@
+<resources>
+
+    <!-- Base application theme. -->
+    <style name="AppTheme" parent="Theme.AppCompat.Light.DarkActionBar">
+        <!-- Customize your theme here. -->
+        <item name="colorPrimary">@color/colorPrimary</item>
+        <item name="colorPrimaryDark">@color/colorPrimaryDark</item>
+        <item name="colorAccent">@color/colorAccent</item>
+        <item name="actionOverflowMenuStyle">@style/OverflowMenuStyle</item>
+    </style>
+
+    <style name="OverflowMenuStyle" parent="Widget.AppCompat.Light.PopupMenu.Overflow">
+        <item name="overlapAnchor">false</item>
+    </style>
+
+    <style name="AppTheme.NoActionBar">
+        <item name="windowActionBar">false</item>
+        <item name="windowNoTitle">true</item>
+    </style>
+
+    <style name="AppTheme.AppBarOverlay" parent="ThemeOverlay.AppCompat.Dark.ActionBar"/>
+
+    <style name="AppTheme.PopupOverlay" parent="ThemeOverlay.AppCompat.Light"/>
+
+</resources>
diff --git a/deploy/android_demo/app/src/main/res/xml/file_paths.xml b/deploy/android_demo/app/src/main/res/xml/file_paths.xml
new file mode 100644
index 0000000..0648268
--- /dev/null
+++ b/deploy/android_demo/app/src/main/res/xml/file_paths.xml
@@ -0,0 +1,4 @@
+<?xml version="1.0" encoding="utf-8"?>
+<paths xmlns:android="http://schemas.android.com/apk/res/android">
+    <external-files-path name="my_images" path="Pictures" />
+</paths>
diff --git a/deploy/android_demo/app/src/main/res/xml/settings.xml b/deploy/android_demo/app/src/main/res/xml/settings.xml
new file mode 100644
index 0000000..8c2ea62
--- /dev/null
+++ b/deploy/android_demo/app/src/main/res/xml/settings.xml
@@ -0,0 +1,61 @@
+<?xml version="1.0" encoding="utf-8"?>
+<PreferenceScreen xmlns:android="http://schemas.android.com/apk/res/android" >
+    <PreferenceCategory android:title="Model Settings">
+        <ListPreference
+                android:defaultValue="@string/MODEL_PATH_DEFAULT"
+                android:key="@string/CHOOSE_PRE_INSTALLED_MODEL_KEY"
+                android:negativeButtonText="@null"
+                android:positiveButtonText="@null"
+                android:title="Choose pre-installed models" />
+        <CheckBoxPreference
+                android:defaultValue="false"
+                android:key="@string/ENABLE_CUSTOM_SETTINGS_KEY"
+                android:summaryOn="Enable"
+                android:summaryOff="Disable"
+                android:title="Enable custom settings"/>
+        <EditTextPreference
+                android:key="@string/MODEL_PATH_KEY"
+                android:defaultValue="@string/MODEL_PATH_DEFAULT"
+                android:title="Model Path" />
+        <EditTextPreference
+                android:key="@string/LABEL_PATH_KEY"
+                android:defaultValue="@string/LABEL_PATH_DEFAULT"
+                android:title="Label Path" />
+        <ListPreference
+                android:key="@string/IMAGE_PATH_KEY"
+                android:defaultValue="@string/IMAGE_PATH_DEFAULT"
+                android:entries="@array/image_name_entries"
+                android:entryValues="@array/image_name_values"
+                android:title="Image Path" />
+    </PreferenceCategory>
+    <PreferenceCategory android:title="CPU Settings">
+        <ListPreference
+                android:defaultValue="@string/CPU_THREAD_NUM_DEFAULT"
+                android:key="@string/CPU_THREAD_NUM_KEY"
+                android:negativeButtonText="@null"
+                android:positiveButtonText="@null"
+                android:title="CPU Thread Num"
+                android:entries="@array/cpu_thread_num_entries"
+                android:entryValues="@array/cpu_thread_num_values"/>
+        <ListPreference
+                android:defaultValue="@string/CPU_POWER_MODE_DEFAULT"
+                android:key="@string/CPU_POWER_MODE_KEY"
+                android:negativeButtonText="@null"
+                android:positiveButtonText="@null"
+                android:title="CPU Power Mode"
+                android:entries="@array/cpu_power_mode_entries"
+                android:entryValues="@array/cpu_power_mode_values"/>
+    </PreferenceCategory>
+    <PreferenceCategory android:title="Input Settings">
+        <EditTextPreference
+                android:key="@string/DET_LONG_SIZE_KEY"
+                android:defaultValue="@string/DET_LONG_SIZE_DEFAULT"
+                android:title="det long size" />
+    </PreferenceCategory>
+    <PreferenceCategory android:title="Output Settings">
+        <EditTextPreference
+                android:key="@string/SCORE_THRESHOLD_KEY"
+                android:defaultValue="@string/SCORE_THRESHOLD_DEFAULT"
+                android:title="Score Threshold" />
+    </PreferenceCategory>
+</PreferenceScreen>
diff --git a/deploy/android_demo/app/src/test/java/com/baidu/paddle/lite/demo/ocr/ExampleUnitTest.java b/deploy/android_demo/app/src/test/java/com/baidu/paddle/lite/demo/ocr/ExampleUnitTest.java
new file mode 100644
index 0000000..ea4d51f
--- /dev/null
+++ b/deploy/android_demo/app/src/test/java/com/baidu/paddle/lite/demo/ocr/ExampleUnitTest.java
@@ -0,0 +1,17 @@
+package com.baidu.paddle.lite.demo.ocr;
+
+import org.junit.Test;
+
+import static org.junit.Assert.*;
+
+/**
+ * Example local unit test, which will execute on the development machine (host).
+ *
+ * @see <a href="http://d.android.com/tools/testing">Testing documentation</a>
+ */
+public class ExampleUnitTest {
+    @Test
+    public void addition_isCorrect() {
+        assertEquals(4, 2 + 2);
+    }
+}
diff --git a/deploy/android_demo/build.gradle b/deploy/android_demo/build.gradle
new file mode 100644
index 0000000..6cba4f3
--- /dev/null
+++ b/deploy/android_demo/build.gradle
@@ -0,0 +1,27 @@
+// Top-level build file where you can add configuration options common to all sub-projects/modules.
+
+buildscript {
+    repositories {
+        google()
+        jcenter()
+        
+    }
+    dependencies {
+        classpath 'com.android.tools.build:gradle:4.1.2'
+        
+        // NOTE: Do not place your application dependencies here; they belong
+        // in the individual module build.gradle files
+    }
+}
+
+allprojects {
+    repositories {
+        google()
+        jcenter()
+        
+    }
+}
+
+task clean(type: Delete) {
+    delete rootProject.buildDir
+}
diff --git a/deploy/android_demo/gradle.properties b/deploy/android_demo/gradle.properties
new file mode 100644
index 0000000..e59e7fc
--- /dev/null
+++ b/deploy/android_demo/gradle.properties
@@ -0,0 +1,15 @@
+# Project-wide Gradle settings.
+# IDE (e.g. Android Studio) users:
+# Gradle settings configured through the IDE *will override*
+# any settings specified in this file.
+# For more details on how to configure your build environment visit
+# http://www.gradle.org/docs/current/userguide/build_environment.html
+# Specifies the JVM arguments used for the daemon process.
+# The setting is particularly useful for tweaking memory settings.
+org.gradle.jvmargs=-Xmx1536m
+# When configured, Gradle will run in incubating parallel mode.
+# This option should only be used with decoupled projects. More details, visit
+# http://www.gradle.org/docs/current/userguide/multi_project_builds.html#sec:decoupled_projects
+# org.gradle.parallel=true
+
+android.useAndroidX=true
diff --git a/deploy/android_demo/gradle/wrapper/gradle-wrapper.jar b/deploy/android_demo/gradle/wrapper/gradle-wrapper.jar
new file mode 100644
index 0000000..f6b961f
Binary files /dev/null and b/deploy/android_demo/gradle/wrapper/gradle-wrapper.jar differ
diff --git a/deploy/android_demo/gradle/wrapper/gradle-wrapper.properties b/deploy/android_demo/gradle/wrapper/gradle-wrapper.properties
new file mode 100644
index 0000000..4bb3c2f
--- /dev/null
+++ b/deploy/android_demo/gradle/wrapper/gradle-wrapper.properties
@@ -0,0 +1,6 @@
+#Thu Feb 04 20:28:08 CST 2021
+distributionBase=GRADLE_USER_HOME
+distributionPath=wrapper/dists
+zipStoreBase=GRADLE_USER_HOME
+zipStorePath=wrapper/dists
+distributionUrl=https\://services.gradle.org/distributions/gradle-6.5-bin.zip
diff --git a/deploy/android_demo/gradlew b/deploy/android_demo/gradlew
new file mode 100644
index 0000000..dfc1f9b
--- /dev/null
+++ b/deploy/android_demo/gradlew
@@ -0,0 +1,170 @@
+#!/usr/bin/env sh
+
+##############################################################################
+##
+##  Gradle start up script for UN*X
+##
+##############################################################################
+
+# Attempt to set APP_HOME
+# Resolve links: $0 may be a link
+PRG="$0"
+# Need this for relative symlinks.
+while [ -h "$PRG" ] ; do
+    ls=`ls -ld "$PRG"`
+    link=`expr "$ls" : '.*-> \(.*\)$'`
+    if expr "$link" : '/.*' > /dev/null; then
+        PRG="$link"
+    else
+        PRG=`dirname "$PRG"`"/$link"
+    fi
+done
+SAVED="`pwd`"
+cd "`dirname \"$PRG\"`/" >/dev/null
+APP_HOME="`pwd -P`"
+cd "$SAVED" >/dev/null
+
+APP_NAME="Gradle"
+APP_BASE_NAME=`basename "$0"`
+
+# Add default JVM options here. You can also use JAVA_OPTS and GRADLE_OPTS to pass JVM options to this script.
+DEFAULT_JVM_OPTS=""
+
+# Use the maximum available, or set MAX_FD != -1 to use that value.
+MAX_FD="maximum"
+
+warn () {
+    echo "$*"
+}
+
+die () {
+    echo
+    echo "$*"
+    echo
+    exit 1
+}
+
+# OS specific support (must be 'true' or 'false').
+cygwin=false
+msys=false
+darwin=false
+nonstop=false
+case "`uname`" in
+  CYGWIN* )
+    cygwin=true
+    ;;
+  Darwin* )
+    darwin=true
+    ;;
+  MINGW* )
+    msys=true
+    ;;
+  NONSTOP* )
+    nonstop=true
+    ;;
+esac
+
+CLASSPATH=$APP_HOME/gradle/wrapper/gradle-wrapper.jar
+
+# Determine the Java command to use to start the JVM.
+if [ -n "$JAVA_HOME" ] ; then
+    if [ -x "$JAVA_HOME/jre/sh/java" ] ; then
+        # IBM's JDK on AIX uses strange locations for the executables
+        JAVACMD="$JAVA_HOME/jre/sh/java"
+    else
+        JAVACMD="$JAVA_HOME/bin/java"
+    fi
+    if [ ! -x "$JAVACMD" ] ; then
+    die "ERROR: JAVA_HOME is set to an invalid directory: $JAVA_HOME\n\nPlease set the JAVA_HOME variable in your environment to match the location of your Java installation."
+    fi
+
+else
+    JAVACMD="java"
+    which java >/dev/null 2>&1 || die "ERROR: JAVA_HOME is not set and no 'java' command could be found in your PATH.
+
+Please set the JAVA_HOME variable in your environment to match the
+location of your Java installation."
+fi
+
+# Increase the maximum file descriptors if we can.
+if [ "$cygwin" = "false" -a "$darwin" = "false" -a "$nonstop" = "false" ] ; then
+    MAX_FD_LIMIT=`ulimit -H -n`
+    if [ $? -eq 0 ] ; then
+        if [ "$MAX_FD" = "maximum" -o "$MAX_FD" = "max" ] ; then
+            MAX_FD="$MAX_FD_LIMIT"
+        fi
+        ulimit -n $MAX_FD
+        if [ $? -ne 0 ] ; then
+            warn "Could not set maximum file descriptor limit: $MAX_FD"
+        fi
+    else
+        warn "Could not query maximum file descriptor limit: $MAX_FD_LIMIT"
+    fi
+fi
+
+# For Darwin, add options to specify how the application appears in the dock
+if $darwin; then
+    GRADLE_OPTS="$GRADLE_OPTS \"-Xdock:name=$APP_NAME\" \"-Xdock:icon=$APP_HOME/media/gradle.icns\""
+fi
+
+# For Cygwin, switch paths to Windows format before running java
+if $cygwin ; then
+    APP_HOME=`cygpath --path --mixed "$APP_HOME"`
+    CLASSPATH=`cygpath --path --mixed "$CLASSPATH"`
+    JAVACMD=`cygpath --unix "$JAVACMD"`
+
+    # We build the pattern for arguments to be converted via cygpath
+    ROOTDIRSRAW=`find -L / -maxdepth 1 -mindepth 1 -type d 2>/dev/null`
+    SEP=""
+    for dir in $ROOTDIRSRAW ; do
+        ROOTDIRS="$ROOTDIRS$SEP$dir"
+        SEP="|"
+    done
+    OURCYGPATTERN="(^($ROOTDIRS))"
+    # Add a user-defined pattern to the cygpath arguments
+    if [ "$GRADLE_CYGPATTERN" != "" ] ; then
+        OURCYGPATTERN="$OURCYGPATTERN|($GRADLE_CYGPATTERN)"
+    fi
+    # Now convert the arguments - kludge to limit ourselves to /bin/sh
+    i=0
+    for arg in "$@" ; do
+        CHECK=`echo "$arg"|egrep -c "$OURCYGPATTERN" -`
+        CHECK2=`echo "$arg"|egrep -c "^-"`                                 ### Determine if an option
+
+        if [ $CHECK -ne 0 ] && [ $CHECK2 -eq 0 ] ; then                    ### Added a condition
+            eval `echo args$i`=`cygpath --path --ignore --mixed "$arg"`
+        else
+            eval `echo args$i`="\"$arg\""
+        fi
+        i=$((i+1))
+    done
+    case $i in
+        (0) set -- ;;
+        (1) set -- "$args0" ;;
+        (2) set -- "$args0" "$args1" ;;
+        (3) set -- "$args0" "$args1" "$args2" ;;
+        (4) set -- "$args0" "$args1" "$args2" "$args3" ;;
+        (5) set -- "$args0" "$args1" "$args2" "$args3" "$args4" ;;
+        (6) set -- "$args0" "$args1" "$args2" "$args3" "$args4" "$args5" ;;
+        (7) set -- "$args0" "$args1" "$args2" "$args3" "$args4" "$args5" "$args6" ;;
+        (8) set -- "$args0" "$args1" "$args2" "$args3" "$args4" "$args5" "$args6" "$args7" ;;
+        (9) set -- "$args0" "$args1" "$args2" "$args3" "$args4" "$args5" "$args6" "$args7" "$args8" ;;
+    esac
+fi
+
+# Escape application args
+save () {
+    for i do printf %s\\n "$i" | sed "s/'/'\\\\''/g;1s/^/'/;\$s/\$/' \\\\/" ; done
+    echo " "
+}
+APP_ARGS=$(save "$@")
+
+# Collect all arguments for the java command, following the shell quoting and substitution rules
+eval set -- $DEFAULT_JVM_OPTS $JAVA_OPTS $GRADLE_OPTS "\"-Dorg.gradle.appname=$APP_BASE_NAME\"" -classpath "\"$CLASSPATH\"" org.gradle.wrapper.GradleWrapperMain "$APP_ARGS"
+
+# by default we should be in the correct project dir, but when run from Finder on Mac, the cwd is wrong
+if [ "$(uname)" = "Darwin" ] && [ "$HOME" = "$PWD" ]; then
+  cd "$(dirname "$0")"
+fi
+
+exec "$JAVACMD" "$@"
diff --git a/deploy/android_demo/gradlew.bat b/deploy/android_demo/gradlew.bat
new file mode 100644
index 0000000..f955316
--- /dev/null
+++ b/deploy/android_demo/gradlew.bat
@@ -0,0 +1,84 @@
+@if "%DEBUG%" == "" @echo off
+@rem ##########################################################################
+@rem
+@rem  Gradle startup script for Windows
+@rem
+@rem ##########################################################################
+
+@rem Set local scope for the variables with windows NT shell
+if "%OS%"=="Windows_NT" setlocal
+
+set DIRNAME=%~dp0
+if "%DIRNAME%" == "" set DIRNAME=.
+set APP_BASE_NAME=%~n0
+set APP_HOME=%DIRNAME%
+
+@rem Add default JVM options here. You can also use JAVA_OPTS and GRADLE_OPTS to pass JVM options to this script.
+set DEFAULT_JVM_OPTS=
+
+@rem Find java.exe
+if defined JAVA_HOME goto findJavaFromJavaHome
+
+set JAVA_EXE=java.exe
+%JAVA_EXE% -version >NUL 2>&1
+if "%ERRORLEVEL%" == "0" goto init
+
+echo.
+echo ERROR: JAVA_HOME is not set and no 'java' command could be found in your PATH.
+echo.
+echo Please set the JAVA_HOME variable in your environment to match the
+echo location of your Java installation.
+
+goto fail
+
+:findJavaFromJavaHome
+set JAVA_HOME=%JAVA_HOME:"=%
+set JAVA_EXE=%JAVA_HOME%/bin/java.exe
+
+if exist "%JAVA_EXE%" goto init
+
+echo.
+echo ERROR: JAVA_HOME is set to an invalid directory: %JAVA_HOME%
+echo.
+echo Please set the JAVA_HOME variable in your environment to match the
+echo location of your Java installation.
+
+goto fail
+
+:init
+@rem Get command-line arguments, handling Windows variants
+
+if not "%OS%" == "Windows_NT" goto win9xME_args
+
+:win9xME_args
+@rem Slurp the command line arguments.
+set CMD_LINE_ARGS=
+set _SKIP=2
+
+:win9xME_args_slurp
+if "x%~1" == "x" goto execute
+
+set CMD_LINE_ARGS=%*
+
+:execute
+@rem Setup the command line
+
+set CLASSPATH=%APP_HOME%\gradle\wrapper\gradle-wrapper.jar
+
+@rem Execute Gradle
+"%JAVA_EXE%" %DEFAULT_JVM_OPTS% %JAVA_OPTS% %GRADLE_OPTS% "-Dorg.gradle.appname=%APP_BASE_NAME%" -classpath "%CLASSPATH%" org.gradle.wrapper.GradleWrapperMain %CMD_LINE_ARGS%
+
+:end
+@rem End local scope for the variables with windows NT shell
+if "%ERRORLEVEL%"=="0" goto mainEnd
+
+:fail
+rem Set variable GRADLE_EXIT_CONSOLE if you need the _script_ return code instead of
+rem the _cmd.exe /c_ return code!
+if  not "" == "%GRADLE_EXIT_CONSOLE%" exit 1
+exit /b 1
+
+:mainEnd
+if "%OS%"=="Windows_NT" endlocal
+
+:omega
diff --git a/deploy/android_demo/settings.gradle b/deploy/android_demo/settings.gradle
new file mode 100644
index 0000000..e7b4def
--- /dev/null
+++ b/deploy/android_demo/settings.gradle
@@ -0,0 +1 @@
+include ':app'
diff --git a/deploy/avh/.gitignore b/deploy/avh/.gitignore
new file mode 100644
index 0000000..a2f78a2
--- /dev/null
+++ b/deploy/avh/.gitignore
@@ -0,0 +1,5 @@
+include/inputs.h
+include/outputs.h
+
+__pycache__/
+build/
diff --git a/deploy/avh/Makefile b/deploy/avh/Makefile
new file mode 100644
index 0000000..cf7d375
--- /dev/null
+++ b/deploy/avh/Makefile
@@ -0,0 +1,129 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+# Makefile to build demo
+
+# Setup build environment
+BUILD_DIR := build
+
+ARM_CPU = ARMCM55
+ETHOSU_PATH = /opt/arm/ethosu
+CMSIS_PATH ?= ${ETHOSU_PATH}/cmsis
+ETHOSU_PLATFORM_PATH ?= ${ETHOSU_PATH}/core_platform
+STANDALONE_CRT_PATH := $(abspath $(BUILD_DIR))/runtime
+CORSTONE_300_PATH = ${ETHOSU_PLATFORM_PATH}/targets/corstone-300
+PKG_COMPILE_OPTS = -g -Wall -O2 -Wno-incompatible-pointer-types -Wno-format -mcpu=cortex-m55 -mthumb -mfloat-abi=hard -std=gnu99
+CMAKE ?= cmake
+CC = arm-none-eabi-gcc
+AR = arm-none-eabi-ar
+RANLIB = arm-none-eabi-ranlib
+PKG_CFLAGS = ${PKG_COMPILE_OPTS} \
+	-I${STANDALONE_CRT_PATH}/include \
+	-I${STANDALONE_CRT_PATH}/src/runtime/crt/include \
+	-I${PWD}/include \
+	-I${CORSTONE_300_PATH} \
+	-I${CMSIS_PATH}/Device/ARM/${ARM_CPU}/Include/ \
+	-I${CMSIS_PATH}/CMSIS/Core/Include \
+	-I${CMSIS_PATH}/CMSIS/NN/Include \
+	-I${CMSIS_PATH}/CMSIS/DSP/Include \
+	-I$(abspath $(BUILD_DIR))/codegen/host/include
+CMSIS_NN_CMAKE_FLAGS = -DCMAKE_TOOLCHAIN_FILE=$(abspath $(BUILD_DIR))/../arm-none-eabi-gcc.cmake \
+	-DTARGET_CPU=cortex-m55 \
+	-DBUILD_CMSIS_NN_FUNCTIONS=YES
+PKG_LDFLAGS = -lm -specs=nosys.specs -static -T corstone300.ld
+
+$(ifeq VERBOSE,1)
+QUIET ?=
+$(else)
+QUIET ?= @
+$(endif)
+
+DEMO_MAIN = src/demo_bare_metal.c
+CODEGEN_SRCS = $(wildcard $(abspath $(BUILD_DIR))/codegen/host/src/*.c)
+CODEGEN_OBJS = $(subst .c,.o,$(CODEGEN_SRCS))
+CMSIS_STARTUP_SRCS = $(wildcard ${CMSIS_PATH}/Device/ARM/${ARM_CPU}/Source/*.c)
+UART_SRCS = $(wildcard ${CORSTONE_300_PATH}/*.c)
+
+demo: $(BUILD_DIR)/demo
+
+$(BUILD_DIR)/stack_allocator.o: $(STANDALONE_CRT_PATH)/src/runtime/crt/memory/stack_allocator.c
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)$(CC) -c $(PKG_CFLAGS) -o $@  $^
+
+$(BUILD_DIR)/crt_backend_api.o: $(STANDALONE_CRT_PATH)/src/runtime/crt/common/crt_backend_api.c
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)$(CC) -c $(PKG_CFLAGS) -o $@  $^
+
+# Build generated code
+$(BUILD_DIR)/libcodegen.a: $(CODEGEN_SRCS)
+	$(QUIET)cd $(abspath $(BUILD_DIR)/codegen/host/src) && $(CC) -c $(PKG_CFLAGS) $(CODEGEN_SRCS)
+	$(QUIET)$(AR) -cr $(abspath $(BUILD_DIR)/libcodegen.a) $(CODEGEN_OBJS)
+	$(QUIET)$(RANLIB) $(abspath $(BUILD_DIR)/libcodegen.a)
+
+# Build CMSIS startup code
+${BUILD_DIR}/libcmsis_startup.a: $(CMSIS_STARTUP_SRCS)
+	$(QUIET)mkdir -p $(abspath $(BUILD_DIR)/libcmsis_startup)
+	$(QUIET)cd $(abspath $(BUILD_DIR)/libcmsis_startup) && $(CC) -c $(PKG_CFLAGS) -D${ARM_CPU} $^
+	$(QUIET)$(AR) -cr $(abspath $(BUILD_DIR)/libcmsis_startup.a) $(abspath $(BUILD_DIR))/libcmsis_startup/*.o
+	$(QUIET)$(RANLIB) $(abspath $(BUILD_DIR)/libcmsis_startup.a)
+
+CMSIS_SHA_FILE=${CMSIS_PATH}/977abe9849781a2e788b02282986480ff4e25ea6.sha
+ifneq ("$(wildcard $(CMSIS_SHA_FILE))","")
+${BUILD_DIR}/cmsis_nn/Source/libcmsis-nn.a:
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)cd $(CMSIS_PATH)/CMSIS/NN && $(CMAKE) -B $(abspath $(BUILD_DIR)/cmsis_nn) $(CMSIS_NN_CMAKE_FLAGS)
+	$(QUIET)cd $(abspath $(BUILD_DIR)/cmsis_nn) && $(MAKE) all
+else
+# Build CMSIS-NN
+${BUILD_DIR}/cmsis_nn/Source/SoftmaxFunctions/libCMSISNNSoftmax.a:
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)cd $(CMSIS_PATH)/CMSIS/NN && $(CMAKE) -B $(abspath $(BUILD_DIR)/cmsis_nn) $(CMSIS_NN_CMAKE_FLAGS)
+	$(QUIET)cd $(abspath $(BUILD_DIR)/cmsis_nn) && $(MAKE) all
+endif
+
+# Build demo application
+ifneq ("$(wildcard $(CMSIS_SHA_FILE))","")
+$(BUILD_DIR)/demo: $(DEMO_MAIN) $(UART_SRCS) $(BUILD_DIR)/stack_allocator.o $(BUILD_DIR)/crt_backend_api.o \
+	${BUILD_DIR}/libcodegen.a ${BUILD_DIR}/libcmsis_startup.a ${BUILD_DIR}/cmsis_nn/Source/libcmsis-nn.a
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)$(CC) $(PKG_CFLAGS) $(FREERTOS_FLAGS) -o $@ -Wl,--whole-archive $^ -Wl,--no-whole-archive $(PKG_LDFLAGS)
+else
+$(BUILD_DIR)/demo: $(DEMO_MAIN) $(UART_SRCS) $(BUILD_DIR)/stack_allocator.o $(BUILD_DIR)/crt_backend_api.o \
+       ${BUILD_DIR}/libcodegen.a ${BUILD_DIR}/libcmsis_startup.a \
+       ${BUILD_DIR}/cmsis_nn/Source/SoftmaxFunctions/libCMSISNNSoftmax.a \
+       ${BUILD_DIR}/cmsis_nn/Source/FullyConnectedFunctions/libCMSISNNFullyConnected.a \
+       ${BUILD_DIR}/cmsis_nn/Source/SVDFunctions/libCMSISNNSVDF.a \
+       ${BUILD_DIR}/cmsis_nn/Source/ReshapeFunctions/libCMSISNNReshape.a \
+       ${BUILD_DIR}/cmsis_nn/Source/ActivationFunctions/libCMSISNNActivation.a \
+       ${BUILD_DIR}/cmsis_nn/Source/NNSupportFunctions/libCMSISNNSupport.a \
+       ${BUILD_DIR}/cmsis_nn/Source/ConcatenationFunctions/libCMSISNNConcatenation.a \
+       ${BUILD_DIR}/cmsis_nn/Source/BasicMathFunctions/libCMSISNNBasicMaths.a \
+       ${BUILD_DIR}/cmsis_nn/Source/ConvolutionFunctions/libCMSISNNConvolutions.a \
+       ${BUILD_DIR}/cmsis_nn/Source/PoolingFunctions/libCMSISNNPooling.a
+	$(QUIET)mkdir -p $(@D)
+	$(QUIET)$(CC) $(PKG_CFLAGS) $(FREERTOS_FLAGS) -o $@ -Wl,--whole-archive $^ -Wl,--no-whole-archive $(PKG_LDFLAGS)
+endif
+
+clean:
+	$(QUIET)rm -rf $(BUILD_DIR)/codegen
+
+cleanall:
+	$(QUIET)rm -rf $(BUILD_DIR)
+
+.SUFFIXES:
+
+.DEFAULT: demo
diff --git a/deploy/avh/README.md b/deploy/avh/README.md
new file mode 100644
index 0000000..b366349
--- /dev/null
+++ b/deploy/avh/README.md
@@ -0,0 +1,122 @@
+<!--- Licensed to the Apache Software Foundation (ASF) under one -->
+<!--- or more contributor license agreements.  See the NOTICE file -->
+<!--- distributed with this work for additional information -->
+<!--- regarding copyright ownership.  The ASF licenses this file -->
+<!--- to you under the Apache License, Version 2.0 (the -->
+<!--- "License"); you may not use this file except in compliance -->
+<!--- with the License.  You may obtain a copy of the License at -->
+
+<!---   http://www.apache.org/licenses/LICENSE-2.0 -->
+
+<!--- Unless required by applicable law or agreed to in writing, -->
+<!--- software distributed under the License is distributed on an -->
+<!--- "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -->
+<!--- KIND, either express or implied.  See the License for the -->
+<!--- specific language governing permissions and limitations -->
+<!--- under the License. -->
+
+English | [简体中文](README_ch.md)
+
+Running PaddleOCR text recognition model on bare metal Arm(R) Cortex(R)-M55 CPU using Arm Virtual Hardware
+======================================================================
+
+This folder contains an example of how to run a PaddleOCR model on bare metal [Cortex(R)-M55 CPU](https://www.arm.com/products/silicon-ip-cpu/cortex-m/cortex-m55) using [Arm Virtual Hardware](https://www.arm.com/products/development-tools/simulation/virtual-hardware).
+
+Running environment and prerequisites
+-------------
+
+Case 1: If the demo is run in Arm Virtual Hardware Amazon Machine Image(AMI) instance hosted by [AWS](https://aws.amazon.com/marketplace/pp/prodview-urbpq7yo5va7g?sr=0-1&ref_=beagle&applicationId=AWSMPContessa)/[AWS China](https://awsmarketplace.amazonaws.cn/marketplace/pp/prodview-2y7nefntbmybu), the following software will be installed through [configure_avh.sh](./configure_avh.sh) script. It will install automatically when you run the application through [run_demo.sh](./run_demo.sh) script.
+You can refer to this [guide](https://arm-software.github.io/AVH/main/examples/html/MicroSpeech.html#amilaunch) to launch an Arm Virtual Hardware AMI instance.
+
+Case 2: If the demo is run in the [ci_cpu Docker container](https://github.com/apache/tvm/blob/main/docker/Dockerfile.ci_cpu) provided with [TVM](https://github.com/apache/tvm), then the following software will already be installed.
+
+Case 3: If the demo is not run in the ci_cpu Docker container, then you will need the following:
+
+- Software required to build and run the demo (These can all be installed by running
+  tvm/docker/install/ubuntu_install_ethosu_driver_stack.sh.)
+    - [Fixed Virtual Platform (FVP) based on Arm(R) Corstone(TM)-300 software](https://developer.arm.com/tools-and-software/open-source-software/arm-platforms-software/arm-ecosystem-fvps)
+    - [cmake 3.19.5](https://github.com/Kitware/CMake/releases/)
+    - [GCC toolchain from Arm(R)](https://developer.arm.com/-/media/Files/downloads/gnu-rm/10-2020q4/gcc-arm-none-eabi-10-2020-q4-major-x86_64-linux.tar.bz2)
+    - [Arm(R) Ethos(TM)-U NPU driver stack](https://review.mlplatform.org)
+    - [CMSIS](https://github.com/ARM-software/CMSIS_5)
+- The python libraries listed in the requirements.txt of this directory
+    - These can be installed by running the following from the current directory:
+
+    ```bash
+    pip install -r ./requirements.txt
+    ```
+
+In case2 and case3:
+
+You will need to update your PATH environment variable to include the path to cmake 3.19.5 and the FVP.
+For example if you've installed these in ```/opt/arm``` , then you would do the following:
+
+```bash
+export PATH=/opt/arm/FVP_Corstone_SSE-300/models/Linux64_GCC-6.4:/opt/arm/cmake/bin:$PATH
+```
+
+You will also need TVM which can either be:
+
+- Installed from TLCPack(see [TLCPack](https://tlcpack.ai/))
+- Built from source (see [Install from Source](https://tvm.apache.org/docs/install/from_source.html))
+    - When building from source, the following need to be set in config.cmake:
+        - set(USE_CMSISNN ON)
+        - set(USE_MICRO ON)
+        - set(USE_LLVM ON)
+
+Running the demo application
+----------------------------
+
+Type the following command to run the bare metal text recognition application ([src/demo_bare_metal.c](./src/demo_bare_metal.c)):
+
+```bash
+./run_demo.sh
+```
+
+If you are not able to use Arm Virtual Hardware Amazon Machine Image(AMI) instance hosted by AWS/AWS China, specify argument --enable_FVP to 1 to make the application run on local Fixed Virtual Platforms (FVPs) executables.
+
+```bash
+./run_demo.sh --enable_FVP 1
+```
+
+If the Ethos(TM)-U platform and/or CMSIS have not been installed in /opt/arm/ethosu then
+the locations for these can be specified as arguments to run_demo.sh, for example:
+
+```bash
+./run_demo.sh --cmsis_path /home/tvm-user/cmsis \
+--ethosu_platform_path /home/tvm-user/ethosu/core_platform
+```
+
+With [run_demo.sh](./run_demo.sh) to run the demo application, it will:
+
+- Set up running environment by installing the required prerequisites automatically if running in Arm Virtual Hardware Amazon AMI instance(not specify --enable_FVP to 1)
+- Download a PaddleOCR text recognition model
+- Use tvmc to compile the text recognition model for Cortex(R)-M55 CPU and CMSIS-NN
+- Create a C header file inputs.c containing the image data as a C array
+- Create a C header file outputs.c containing a C array where the output of inference will be stored
+- Build the demo application
+- Run the demo application on a Arm Virtual Hardware based on Arm(R) Corstone(TM)-300 software
+- The application will report the text on the image and the corresponding score.
+
+Using your own image
+--------------------
+
+The create_image.py script takes a single argument on the command line which is the path of the
+image to be converted into an array of bytes for consumption by the model.
+
+The demo can be modified to use an image of your choice by changing the following line in run_demo.sh
+
+```bash
+python3 ./convert_image.py path/to/image
+```
+
+Model description
+-----------------
+
+The example is built on [PP-OCRv3](https://github.com/PaddlePaddle/PaddleOCR/blob/main/docs/version2.x/ppocr/blog/PP-OCRv3_introduction.md) English recognition model released by [PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR). Since Arm(R) Cortex(R)-M55 CPU does not support rnn operator, we delete the unsupported operator based on the PP-OCRv3 text recognition model to obtain the current 2.7M English recognition model.
+
+PP-OCRv3 is the third version of the PP-OCR series model. This series of models has the following features:
+
+- PP-OCRv3: ultra-lightweight OCR system: detection (3.6M) + direction classifier (1.4M) + recognition (12M) = 17.0M
+- Support more than 80 kinds of multi-language recognition models, including English, Chinese, French, German, Arabic, Korean, Japanese and so on. For details
+- Support vertical text recognition, and long text recognition
diff --git a/deploy/avh/README_ch.md b/deploy/avh/README_ch.md
new file mode 100644
index 0000000..4377b4c
--- /dev/null
+++ b/deploy/avh/README_ch.md
@@ -0,0 +1,94 @@
+<!--- Licensed to the Apache Software Foundation (ASF) under one -->
+<!--- or more contributor license agreements.  See the NOTICE file -->
+<!--- distributed with this work for additional information -->
+<!--- regarding copyright ownership.  The ASF licenses this file -->
+<!--- to you under the Apache License, Version 2.0 (the -->
+<!--- "License"); you may not use this file except in compliance -->
+<!--- with the License.  You may obtain a copy of the License at -->
+
+<!---   http://www.apache.org/licenses/LICENSE-2.0 -->
+
+<!--- Unless required by applicable law or agreed to in writing, -->
+<!--- software distributed under the License is distributed on an -->
+<!--- "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY -->
+<!--- KIND, either express or implied.  See the License for the -->
+<!--- specific language governing permissions and limitations -->
+<!--- under the License. -->
+[English](README.md) | 简体中文
+
+通过TVM在 Arm(R) Cortex(R)-M55 CPU 上运行 PaddleOCR文 本能识别模型
+===============================================================
+
+此文件夹包含如何使用 TVM 在 Cortex(R)-M55 CPU 上运行 PaddleOCR 模型的示例。
+
+依赖
+-------------
+本demo运行在TVM提供的docker环境上，在该环境中已经安装好的必须的软件
+
+
+在非docker环境中，需要手动安装如下依赖项:
+
+- 软件可通过[安装脚本](https://github.com/apache/tvm/blob/v0.9.0/docker/install/ubuntu_install_ethosu_driver_stack.sh)一键安装
+  - [Fixed Virtual Platform (FVP) based on Arm(R) Corstone(TM)-300 software](https://developer.arm.com/tools-and-software/open-source-software/arm-platforms-software/arm-ecosystem-fvps)
+  - [cmake 3.19.5](https://github.com/Kitware/CMake/releases/)
+  - [GCC toolchain from Arm(R)](https://developer.arm.com/-/media/Files/downloads/gnu-rm/10-2020q4/gcc-arm-none-eabi-10-2020-q4-major-x86_64-linux.tar.bz2)
+  - [Arm(R) Ethos(TM)-U NPU driver stack](https://review.mlplatform.org)
+  - [CMSIS](https://github.com/ARM-software/CMSIS_5)
+- python 依赖
+  ```bash
+  pip install -r ./requirements.txt
+  ```
+- TVM
+  - 从源码安装([Install from Source](https://tvm.apache.org/docs/install/from_source.html))
+    从源码安装时，需要设置如下字段
+      - set(USE_CMSISNN ON)
+      - set(USE_MICRO ON)
+      - set(USE_LLVM ON)
+  - 从TLCPack 安装([TLCPack](https://tlcpack.ai/))
+
+安装完成后需要更新环境变量，以软件安装地址为`/opt/arm`为例：
+```bash
+export PATH=/opt/arm/FVP_Corstone_SSE-300/models/Linux64_GCC-6.4:/opt/arm/cmake/bin:$PATH
+```
+
+运行demo
+----------------------------
+使用如下命令可以一键运行demo
+
+```bash
+./run_demo.sh
+```
+
+如果 Ethos(TM)-U 平台或 CMSIS 没有安装在 `/opt/arm/ethosu` 中，可通过参数进行设置，例如：
+
+```bash
+./run_demo.sh --cmsis_path /home/tvm-user/cmsis \
+--ethosu_platform_path /home/tvm-user/ethosu/core_platform
+```
+
+`./run_demo.sh`脚本会执行如下步骤:
+- 下载 PaddleOCR 文字识别模型
+- 使用tvm将PaddleOCR 文字识别模型编译为 Cortex(R)-M55 CPU 和 CMSIS-NN 后端的可执行文件
+- 创建一个包含输入图像数据的头文件`inputs.c`
+- 创建一个包含输出tensor大小的头文件`outputs.c`
+- 编译可执行程序
+- 运行程序
+- 输出图片上的文字和置信度
+
+使用自己的图片
+--------------------
+替换 `run_demo.sh ` 中140行处的图片地址即可
+
+使用自己的模型
+--------------------
+替换 `run_demo.sh ` 中130行处的模型地址即可
+
+模型描述
+-----------------
+
+在这个demo中，我们使用的模型是基于[PP-OCRv3](https://github.com/PaddlePaddle/PaddleOCR/blob/main/docs/version2.x/ppocr/blog/PP-OCRv3_introduction.md)的英文识别模型。由于Arm(R) Cortex(R)-M55 CPU不支持rnn算子，我们在PP-OCRv3原始文本识别模型的基础上进行适配，最终模型大小为2.7M。
+
+PP-OCRv3是[PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR)发布的PP-OCR系列模型的第三个版本，该系列模型具有以下特点：
+   - 超轻量级OCR系统：检测（3.6M）+方向分类器（1.4M）+识别（12M）=17.0M。
+   - 支持80多种多语言识别模型，包括英文、中文、法文、德文、阿拉伯文、韩文、日文等。
+   - 支持竖排文本识别，长文本识别。
diff --git a/deploy/avh/arm-none-eabi-gcc.cmake b/deploy/avh/arm-none-eabi-gcc.cmake
new file mode 100644
index 0000000..415b313
--- /dev/null
+++ b/deploy/avh/arm-none-eabi-gcc.cmake
@@ -0,0 +1,79 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+if (__TOOLCHAIN_LOADED)
+    return()
+endif()
+set(__TOOLCHAIN_LOADED TRUE)
+
+set(CMAKE_SYSTEM_NAME Generic)
+set(CMAKE_C_COMPILER "arm-none-eabi-gcc")
+set(CMAKE_CXX_COMPILER "arm-none-eabi-g++")
+set(CMAKE_SYSTEM_PROCESSOR "cortex-m55" CACHE STRING "Select Arm(R) Cortex(R)-M architecture. (cortex-m0, cortex-m3, cortex-m33, cortex-m4, cortex-m55, cortex-m7, etc)")
+
+set(CMAKE_TRY_COMPILE_TARGET_TYPE STATIC_LIBRARY)
+
+SET(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
+SET(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
+SET(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)
+
+set(CMAKE_C_STANDARD 99)
+set(CMAKE_CXX_STANDARD 14)
+
+# The system processor could for example be set to cortex-m33+nodsp+nofp.
+set(__CPU_COMPILE_TARGET ${CMAKE_SYSTEM_PROCESSOR})
+string(REPLACE "+" ";" __CPU_FEATURES ${__CPU_COMPILE_TARGET})
+list(POP_FRONT __CPU_FEATURES CMAKE_SYSTEM_PROCESSOR)
+
+string(FIND ${__CPU_COMPILE_TARGET} "+" __OFFSET)
+if(__OFFSET GREATER_EQUAL 0)
+    string(SUBSTRING ${__CPU_COMPILE_TARGET} ${__OFFSET} -1 CPU_FEATURES)
+endif()
+
+# Add -mcpu to the compile options to override the -mcpu the CMake toolchain adds
+add_compile_options(-mcpu=${__CPU_COMPILE_TARGET})
+
+# Set floating point unit
+if("${__CPU_COMPILE_TARGET}" MATCHES "\\+fp")
+    set(FLOAT hard)
+elseif("${__CPU_COMPILE_TARGET}" MATCHES "\\+nofp")
+    set(FLOAT soft)
+elseif("${CMAKE_SYSTEM_PROCESSOR}" STREQUAL "cortex-m33" OR
+       "${CMAKE_SYSTEM_PROCESSOR}" STREQUAL "cortex-m55")
+    set(FLOAT hard)
+else()
+    set(FLOAT soft)
+endif()
+
+add_compile_options(-mfloat-abi=${FLOAT})
+add_link_options(-mfloat-abi=${FLOAT})
+
+# Link target
+add_link_options(-mcpu=${__CPU_COMPILE_TARGET})
+add_link_options(-Xlinker -Map=output.map)
+
+#
+# Compile options
+#
+set(cxx_flags "-fno-unwind-tables;-fno-rtti;-fno-exceptions")
+
+add_compile_options("-Wall;-Wextra;-Wsign-compare;-Wunused;-Wswitch-default;\
+-Wdouble-promotion;-Wredundant-decls;-Wshadow;-Wnull-dereference;\
+-Wno-format-extra-args;-Wno-unused-function;-Wno-unused-label;\
+-Wno-missing-field-initializers;-Wno-return-type;-Wno-format;-Wno-int-conversion"
+    "$<$<COMPILE_LANGUAGE:CXX>:${cxx_flags}>"
+)
diff --git a/deploy/avh/configure_avh.sh b/deploy/avh/configure_avh.sh
new file mode 100755
index 0000000..4789271
--- /dev/null
+++ b/deploy/avh/configure_avh.sh
@@ -0,0 +1,79 @@
+#!/bin/bash
+# Copyright (c) 2022 Arm Limited and Contributors. All rights reserved.
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+set -e
+set -u
+set -o pipefail
+
+# Show usage
+function show_usage() {
+    cat <<EOF
+Usage: Set up running environment by installing the required prerequisites.
+-h, --help
+    Display this help message.
+EOF
+}
+
+if [ "$#" -eq 1 ] && [ "$1" == "--help" -o "$1" == "-h" ]; then
+    show_usage
+    exit 0
+elif [ "$#" -ge 1 ]; then
+        show_usage
+        exit 1
+fi
+
+echo -e "\e[36mStart setting up running environment\e[0m"
+
+# Install CMSIS
+echo -e "\e[36mStart installing CMSIS\e[0m"
+CMSIS_PATH="/opt/arm/ethosu/cmsis" 
+mkdir -p "${CMSIS_PATH}"
+
+CMSIS_SHA="977abe9849781a2e788b02282986480ff4e25ea6"
+CMSIS_SHASUM="86c88d9341439fbb78664f11f3f25bc9fda3cd7de89359324019a4d87d169939eea85b7fdbfa6ad03aa428c6b515ef2f8cd52299ce1959a5444d4ac305f934cc"
+CMSIS_URL="http://github.com/ARM-software/CMSIS_5/archive/${CMSIS_SHA}.tar.gz"
+DOWNLOAD_PATH="/tmp/${CMSIS_SHA}.tar.gz"
+
+wget ${CMSIS_URL} -O "${DOWNLOAD_PATH}"
+echo "$CMSIS_SHASUM" ${DOWNLOAD_PATH} | sha512sum -c
+tar -xf "${DOWNLOAD_PATH}" -C "${CMSIS_PATH}" --strip-components=1
+touch "${CMSIS_PATH}"/"${CMSIS_SHA}".sha
+echo -e "\e[36mCMSIS Installation SUCCESS\e[0m"
+
+# Install Arm(R) Ethos(TM)-U NPU driver stack
+echo -e "\e[36mStart installing Arm(R) Ethos(TM)-U NPU driver stack\e[0m"
+git clone "https://review.mlplatform.org/ml/ethos-u/ethos-u-core-platform" /opt/arm/ethosu/core_platform 
+cd /opt/arm/ethosu/core_platform 
+git checkout tags/"21.11"
+echo -e "\e[36mArm(R) Ethos(TM)-U Core Platform Installation SUCCESS\e[0m"
+ 
+# Install Arm(R) GNU Toolchain
+echo -e "\e[36mStart installing Arm(R) GNU Toolchain\e[0m"
+mkdir -p /opt/arm/gcc-arm-none-eabi
+export gcc_arm_url='https://developer.arm.com/-/media/Files/downloads/gnu-rm/10-2020q4/gcc-arm-none-eabi-10-2020-q4-major-x86_64-linux.tar.bz2?revision=ca0cbf9c-9de2-491c-ac48-898b5bbc0443&la=en&hash=68760A8AE66026BCF99F05AC017A6A50C6FD832A'
+curl --retry 64 -sSL ${gcc_arm_url} | tar -C /opt/arm/gcc-arm-none-eabi --strip-components=1 -jx 
+export PATH=/opt/arm/gcc-arm-none-eabi/bin:$PATH 
+arm-none-eabi-gcc --version
+arm-none-eabi-g++ --version
+echo -e "\e[36mArm(R) Arm(R) GNU Toolchain Installation SUCCESS\e[0m"
+
+# Install TVM from TLCPack
+echo -e "\e[36mStart installing TVM\e[0m"
+pip install tlcpack-nightly -f https://tlcpack.ai/wheels
+echo -e "\e[36mTVM Installation SUCCESS\e[0m"
diff --git a/deploy/avh/convert_image.py b/deploy/avh/convert_image.py
new file mode 100755
index 0000000..7a78fac
--- /dev/null
+++ b/deploy/avh/convert_image.py
@@ -0,0 +1,102 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+import os
+import pathlib
+import re
+import sys
+import cv2
+import math
+from PIL import Image
+import numpy as np
+
+
+def resize_norm_img(img, image_shape, padding=True):
+    imgC, imgH, imgW = image_shape
+    h = img.shape[0]
+    w = img.shape[1]
+    if not padding:
+        resized_image = cv2.resize(img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
+        resized_w = imgW
+    else:
+        ratio = w / float(h)
+        if math.ceil(imgH * ratio) > imgW:
+            resized_w = imgW
+        else:
+            resized_w = int(math.ceil(imgH * ratio))
+        resized_image = cv2.resize(img, (resized_w, imgH))
+    resized_image = resized_image.astype("float32")
+    if image_shape[0] == 1:
+        resized_image = resized_image / 255
+        resized_image = resized_image[np.newaxis, :]
+    else:
+        resized_image = resized_image.transpose((2, 0, 1)) / 255
+    resized_image -= 0.5
+    resized_image /= 0.5
+    padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
+    padding_im[:, :, 0:resized_w] = resized_image
+    return padding_im
+
+
+def create_header_file(name, tensor_name, tensor_data, output_path):
+    """
+    This function generates a header file containing the data from the numpy array provided.
+    """
+    file_path = pathlib.Path(f"{output_path}/" + name).resolve()
+    # Create header file with npy_data as a C array
+    raw_path = file_path.with_suffix(".h").resolve()
+    with open(raw_path, "w") as header_file:
+        header_file.write(
+            "\n"
+            + f"const size_t {tensor_name}_len = {tensor_data.size};\n"
+            + f'__attribute__((section(".data.tvm"), aligned(16))) float {tensor_name}[] = '
+        )
+
+        header_file.write("{")
+        for i in np.ndindex(tensor_data.shape):
+            header_file.write(f"{tensor_data[i]}, ")
+        header_file.write("};\n\n")
+
+
+def create_headers(image_name):
+    """
+    This function generates C header files for the input and output arrays required to run inferences
+    """
+    img_path = os.path.join("./", f"{image_name}")
+
+    # Resize image to 32x320
+    img = cv2.imread(img_path)
+    img = resize_norm_img(img, [3, 32, 320])
+    img_data = img.astype("float32")
+
+    # # Add the batch dimension, as we are expecting 4-dimensional input: NCHW.
+    img_data = np.expand_dims(img_data, axis=0)
+
+    # Create input header file
+    create_header_file("inputs", "input", img_data, "./include")
+    # Create output header file
+    output_data = np.zeros([7760], np.float32)
+    create_header_file(
+        "outputs",
+        "output",
+        output_data,
+        "./include",
+    )
+
+
+if __name__ == "__main__":
+    create_headers(sys.argv[1])
diff --git a/deploy/avh/corstone300.ld b/deploy/avh/corstone300.ld
new file mode 100644
index 0000000..e52b23d
--- /dev/null
+++ b/deploy/avh/corstone300.ld
@@ -0,0 +1,295 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied.  See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*------------------ Reference System Memories -------------
+  +===================+============+=======+============+============+
+  | Memory            | Address    | Size  | CPU Access | NPU Access |
+  +===================+============+=======+============+============+
+  | ITCM              | 0x00000000 | 512KB | Yes (RO)   | No         |
+  +-------------------+------------+-------+------------+------------+
+  | DTCM              | 0x20000000 | 512KB | Yes (R/W)  | No         |
+  +-------------------+------------+-------+------------+------------+
+  | SSE-300 SRAM      | 0x21000000 |   2MB | Yes (R/W)  | Yes (R/W)  |
+  +-------------------+------------+-------+------------+------------+
+  | Data SRAM         | 0x01000000 |   2MB | Yes (R/W)  | Yes (R/W)  |
+  +-------------------+------------+-------+------------+------------+
+  | DDR               | 0x60000000 |  32MB | Yes (R/W)  | Yes (R/W)  |
+  +-------------------+------------+-------+------------+------------+ */
+
+/*---------------------- ITCM Configuration ----------------------------------
+  <h> Flash Configuration
+    <o0> Flash Base Address <0x0-0xFFFFFFFF:8>
+    <o1> Flash Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+  -----------------------------------------------------------------------------*/
+__ROM_BASE = 0x00000000;
+__ROM_SIZE = 0x00080000;
+
+/*--------------------- DTCM RAM Configuration ----------------------------
+  <h> RAM Configuration
+    <o0> RAM Base Address    <0x0-0xFFFFFFFF:8>
+    <o1> RAM Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+ -----------------------------------------------------------------------------*/
+__RAM_BASE = 0x20000000;
+__RAM_SIZE = 0x00080000;
+
+/*----------------------- Data SRAM Configuration ------------------------------
+  <h> Data SRAM Configuration
+    <o0> DATA_SRAM Base Address    <0x0-0xFFFFFFFF:8>
+    <o1> DATA_SRAM Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+ -----------------------------------------------------------------------------*/
+__DATA_SRAM_BASE = 0x01000000;
+__DATA_SRAM_SIZE = 0x00200000;
+
+/*--------------------- Embedded SRAM Configuration ----------------------------
+  <h> SRAM Configuration
+    <o0> SRAM Base Address    <0x0-0xFFFFFFFF:8>
+    <o1> SRAM Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+ -----------------------------------------------------------------------------*/
+__SRAM_BASE = 0x21000000;
+__SRAM_SIZE = 0x00200000;
+
+/*--------------------- Stack / Heap Configuration ----------------------------
+  <h> Stack / Heap Configuration
+    <o0> Stack Size (in Bytes) <0x0-0xFFFFFFFF:8>
+    <o1> Heap Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+  -----------------------------------------------------------------------------*/
+__STACK_SIZE = 0x00008000;
+__HEAP_SIZE  = 0x00008000;
+
+/*--------------------- Embedded RAM Configuration ----------------------------
+  <h> DDR Configuration
+    <o0> DDR Base Address    <0x0-0xFFFFFFFF:8>
+    <o1> DDR Size (in Bytes) <0x0-0xFFFFFFFF:8>
+  </h>
+ -----------------------------------------------------------------------------*/
+__DDR_BASE = 0x60000000;
+__DDR_SIZE = 0x02000000;
+
+/*
+ *-------------------- <<< end of configuration section >>> -------------------
+ */
+
+MEMORY
+{
+  ITCM       (rx)  : ORIGIN = __ROM_BASE, LENGTH = __ROM_SIZE
+  DTCM       (rwx) : ORIGIN = __RAM_BASE, LENGTH = __RAM_SIZE
+  DATA_SRAM  (rwx) : ORIGIN = __DATA_SRAM_BASE, LENGTH = __DATA_SRAM_SIZE
+  SRAM       (rwx) : ORIGIN = __SRAM_BASE, LENGTH = __SRAM_SIZE
+  DDR        (rwx) : ORIGIN = __DDR_BASE, LENGTH = __DDR_SIZE
+}
+
+/* Linker script to place sections and symbol values. Should be used together
+ * with other linker script that defines memory regions ITCM and RAM.
+ * It references following symbols, which must be defined in code:
+ *   Reset_Handler : Entry of reset handler
+ *
+ * It defines following symbols, which code can use without definition:
+ *   __exidx_start
+ *   __exidx_end
+ *   __copy_table_start__
+ *   __copy_table_end__
+ *   __zero_table_start__
+ *   __zero_table_end__
+ *   __etext
+ *   __data_start__
+ *   __preinit_array_start
+ *   __preinit_array_end
+ *   __init_array_start
+ *   __init_array_end
+ *   __fini_array_start
+ *   __fini_array_end
+ *   __data_end__
+ *   __bss_start__
+ *   __bss_end__
+ *   __end__
+ *   end
+ *   __HeapLimit
+ *   __StackLimit
+ *   __StackTop
+ *   __stack
+ */
+ENTRY(Reset_Handler)
+
+SECTIONS
+{
+  /* .ddr is placed before .text so that .rodata.tvm is encountered before .rodata* */
+  .ddr :
+  {
+    . = ALIGN (16);
+    *(.rodata.tvm)
+    . = ALIGN (16);
+    *(.data.tvm);
+    . = ALIGN(16);    
+  } > DDR
+
+  .text :
+  {
+    KEEP(*(.vectors))
+    *(.text*)
+
+    KEEP(*(.init))
+    KEEP(*(.fini))
+
+    /* .ctors */
+    *crtbegin.o(.ctors)
+    *crtbegin?.o(.ctors)
+    *(EXCLUDE_FILE(*crtend?.o *crtend.o) .ctors)
+    *(SORT(.ctors.*))
+    *(.ctors)
+
+    /* .dtors */
+    *crtbegin.o(.dtors)
+    *crtbegin?.o(.dtors)
+    *(EXCLUDE_FILE(*crtend?.o *crtend.o) .dtors)
+    *(SORT(.dtors.*))
+    *(.dtors)
+
+    *(.rodata*)
+
+    KEEP(*(.eh_frame*))
+  } > ITCM
+
+  .ARM.extab :
+  {
+    *(.ARM.extab* .gnu.linkonce.armextab.*)
+  } > ITCM
+
+  __exidx_start = .;
+  .ARM.exidx :
+  {
+    *(.ARM.exidx* .gnu.linkonce.armexidx.*)
+  } > ITCM
+  __exidx_end = .;
+
+  .copy.table :
+  {
+    . = ALIGN(4);
+    __copy_table_start__ = .;
+    LONG (__etext)
+    LONG (__data_start__)
+    LONG (__data_end__ - __data_start__)
+    /* Add each additional data section here */
+    __copy_table_end__ = .;
+  } > ITCM
+
+  .zero.table :
+  {
+    . = ALIGN(4);
+    __zero_table_start__ = .;
+    __zero_table_end__ = .;
+  } > ITCM
+
+  /**
+   * Location counter can end up 2byte aligned with narrow Thumb code but
+   * __etext is assumed by startup code to be the LMA of a section in DTCM
+   * which must be 4byte aligned
+   */
+  __etext = ALIGN (4);
+
+  .sram :
+  {
+    . = ALIGN(16);
+  } > SRAM AT > SRAM
+
+  .data : AT (__etext)
+  {
+    __data_start__ = .;
+    *(vtable)
+    *(.data)
+    *(.data.*)
+
+    . = ALIGN(4);
+    /* preinit data */
+    PROVIDE_HIDDEN (__preinit_array_start = .);
+    KEEP(*(.preinit_array))
+    PROVIDE_HIDDEN (__preinit_array_end = .);
+
+    . = ALIGN(4);
+    /* init data */
+    PROVIDE_HIDDEN (__init_array_start = .);
+    KEEP(*(SORT(.init_array.*)))
+    KEEP(*(.init_array))
+    PROVIDE_HIDDEN (__init_array_end = .);
+
+
+    . = ALIGN(4);
+    /* finit data */
+    PROVIDE_HIDDEN (__fini_array_start = .);
+    KEEP(*(SORT(.fini_array.*)))
+    KEEP(*(.fini_array))
+    PROVIDE_HIDDEN (__fini_array_end = .);
+
+    KEEP(*(.jcr*))
+    . = ALIGN(4);
+    /* All data end */
+    __data_end__ = .;
+
+  } > DTCM
+
+  .bss.noinit (NOLOAD):
+  {
+    . = ALIGN(16);
+    *(.bss.noinit.*)
+    . = ALIGN(16);
+  } > SRAM AT > SRAM
+
+  .bss :
+  {
+    . = ALIGN(4);
+    __bss_start__ = .;
+    *(.bss)
+    *(.bss.*)
+    *(COMMON)
+    . = ALIGN(4);
+    __bss_end__ = .;
+  } > DTCM AT > DTCM
+
+  .data_sram :
+  {
+    . = ALIGN(16);
+  } > DATA_SRAM
+
+  .heap (COPY) :
+  {
+    . = ALIGN(8);
+    __end__ = .;
+    PROVIDE(end = .);
+    . = . + __HEAP_SIZE;
+    . = ALIGN(8);
+    __HeapLimit = .;
+  } > DTCM
+
+  .stack (ORIGIN(DTCM) + LENGTH(DTCM) - __STACK_SIZE) (COPY) :
+  {
+    . = ALIGN(8);
+    __StackLimit = .;
+    . = . + __STACK_SIZE;
+    . = ALIGN(8);
+    __StackTop = .;
+  } > DTCM
+  PROVIDE(__stack = __StackTop);
+
+  /* Check if data + stack exceeds DTCM limit */
+  ASSERT(__StackLimit >= __bss_end__, "region DTCM overflowed with stack")
+}
diff --git a/deploy/avh/imgs_words_en/word_10.png b/deploy/avh/imgs_words_en/word_10.png
new file mode 100644
index 0000000..07370f7
Binary files /dev/null and b/deploy/avh/imgs_words_en/word_10.png differ
diff --git a/deploy/avh/imgs_words_en/word_116.png b/deploy/avh/imgs_words_en/word_116.png
new file mode 100644
index 0000000..fd000ff
Binary files /dev/null and b/deploy/avh/imgs_words_en/word_116.png differ
diff --git a/deploy/avh/include/crt_config.h b/deploy/avh/include/crt_config.h
new file mode 100644
index 0000000..2fd0ead
--- /dev/null
+++ b/deploy/avh/include/crt_config.h
@@ -0,0 +1,26 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied.  See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+#ifndef TVM_RUNTIME_CRT_CONFIG_H_
+#define TVM_RUNTIME_CRT_CONFIG_H_
+
+/*! Log level of the CRT runtime */
+#define TVM_CRT_LOG_LEVEL TVM_CRT_LOG_LEVEL_DEBUG
+
+#endif // TVM_RUNTIME_CRT_CONFIG_H_
diff --git a/deploy/avh/include/tvm_runtime.h b/deploy/avh/include/tvm_runtime.h
new file mode 100644
index 0000000..0978d7a
--- /dev/null
+++ b/deploy/avh/include/tvm_runtime.h
@@ -0,0 +1,59 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied.  See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+#include <stdarg.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <tvm/runtime/c_runtime_api.h>
+#include <tvm/runtime/crt/stack_allocator.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void __attribute__((noreturn)) TVMPlatformAbort(tvm_crt_error_t error_code) {
+  printf("TVMPlatformAbort: %d\n", error_code);
+  printf("EXITTHESIM\n");
+  exit(-1);
+}
+
+tvm_crt_error_t TVMPlatformMemoryAllocate(size_t num_bytes, DLDevice dev,
+                                          void **out_ptr) {
+  return kTvmErrorFunctionCallNotImplemented;
+}
+
+tvm_crt_error_t TVMPlatformMemoryFree(void *ptr, DLDevice dev) {
+  return kTvmErrorFunctionCallNotImplemented;
+}
+
+void TVMLogf(const char *msg, ...) {
+  va_list args;
+  va_start(args, msg);
+  vfprintf(stdout, msg, args);
+  va_end(args);
+}
+
+TVM_DLL int TVMFuncRegisterGlobal(const char *name, TVMFunctionHandle f,
+                                  int override) {
+  return 0;
+}
+
+#ifdef __cplusplus
+}
+#endif
diff --git a/deploy/avh/requirements.txt b/deploy/avh/requirements.txt
new file mode 100644
index 0000000..a1a8626
--- /dev/null
+++ b/deploy/avh/requirements.txt
@@ -0,0 +1,4 @@
+paddlepaddle
+numpy
+opencv-python
+typing-extensions
diff --git a/deploy/avh/run_demo.sh b/deploy/avh/run_demo.sh
new file mode 100755
index 0000000..63de91a
--- /dev/null
+++ b/deploy/avh/run_demo.sh
@@ -0,0 +1,184 @@
+#!/bin/bash
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+set -e
+set -u
+set -o pipefail
+
+# Show usage
+function show_usage() {
+    cat <<EOF
+Usage: run_demo.sh
+-h, --help
+	Display this help message.
+--cmsis_path CMSIS_PATH
+	Set path to CMSIS.
+--ethosu_platform_path ETHOSU_PLATFORM_PATH
+	Set path to Arm(R) Ethos(TM)-U core platform.
+--fvp_path FVP_PATH
+	Set path to FVP.
+--cmake_path
+	Set path to cmake.
+--enable_FVP
+	Set 1 to run application on local Fixed Virtual Platforms (FVPs) executables.
+EOF
+}
+
+# Configure environment variables
+FVP_enable=0
+export PATH=/opt/arm/gcc-arm-none-eabi/bin:$PATH
+
+# Install python libraries
+echo -e "\e[36mInstall python libraries\e[0m"
+sudo pip install -r ./requirements.txt
+
+# Parse arguments
+while (( $# )); do
+    case "$1" in
+        -h|--help)
+            show_usage
+            exit 0
+            ;;
+
+        --cmsis_path)
+            if [ $# -gt 1 ]
+            then
+                export CMSIS_PATH="$2"
+                shift 2
+            else
+                echo 'ERROR: --cmsis_path requires a non-empty argument' >&2
+                show_usage >&2
+                exit 1
+            fi
+            ;;
+
+        --ethosu_platform_path)
+            if [ $# -gt 1 ]
+            then
+                export ETHOSU_PLATFORM_PATH="$2"
+                shift 2
+            else
+                echo 'ERROR: --ethosu_platform_path requires a non-empty argument' >&2
+                show_usage >&2
+                exit 1
+            fi
+            ;;
+
+        --fvp_path)
+            if [ $# -gt 1 ]
+            then
+                export PATH="$2/models/Linux64_GCC-6.4:$PATH"
+                shift 2
+            else
+                echo 'ERROR: --fvp_path requires a non-empty argument' >&2
+                show_usage >&2
+                exit 1
+            fi
+            ;;
+
+        --cmake_path)
+            if [ $# -gt 1 ]
+            then
+                export CMAKE="$2"
+                shift 2
+            else
+                echo 'ERROR: --cmake_path requires a non-empty argument' >&2
+                show_usage >&2
+                exit 1
+            fi
+            ;;
+            
+        --enable_FVP)
+            if [ $# -gt 1 ] && [ "$2" == "1" -o "$2" == "0" ];
+            then
+                FVP_enable="$2"
+                shift 2
+            else
+                echo 'ERROR: --enable_FVP requires a right argument 1 or 0' >&2
+                show_usage >&2
+                exit 1
+            fi
+            ;;
+
+        -*|--*)
+            echo "Error: Unknown flag: $1" >&2
+            show_usage >&2
+            exit 1
+            ;;
+    esac
+done
+
+# Choose running environment: cloud(default) or local environment
+Platform="VHT_Corstone_SSE-300_Ethos-U55"
+if [ $FVP_enable == "1" ]; then
+	Platform="FVP_Corstone_SSE-300_Ethos-U55"
+	echo -e "\e[36mRun application on local Fixed Virtual Platforms (FVPs)\e[0m"
+else
+	if [ ! -d "/opt/arm/" ]; then
+		sudo ./configure_avh.sh
+	fi
+fi
+
+# Directories
+script_dir="$( cd "$( dirname "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )"
+
+# Make build directory
+make cleanall
+mkdir -p build
+cd build
+
+# Get PaddlePaddle inference model
+echo -e "\e[36mDownload PaddlePaddle inference model\e[0m"
+wget https://paddleocr.bj.bcebos.com/tvm/ocr_en.tar
+tar -xf ocr_en.tar
+
+# Compile model for Arm(R) Cortex(R)-M55 CPU and CMSIS-NN
+# An alternative to using "python3 -m tvm.driver.tvmc" is to call
+# "tvmc" directly once TVM has been pip installed.
+python3 -m tvm.driver.tvmc compile --target=cmsis-nn,c \
+    --target-cmsis-nn-mcpu=cortex-m55 \
+    --target-c-mcpu=cortex-m55 \
+    --runtime=crt \
+    --executor=aot \
+    --executor-aot-interface-api=c \
+    --executor-aot-unpacked-api=1 \
+    --pass-config tir.usmp.enable=1 \
+    --pass-config tir.usmp.algorithm=hill_climb \
+    --pass-config tir.disable_storage_rewrite=1 \
+    --pass-config tir.disable_vectorize=1 ocr_en/inference.pdmodel \
+    --output-format=mlf \
+    --model-format=paddle \
+    --module-name=rec \
+    --input-shapes x:[1,3,32,320] \
+    --output=rec.tar
+tar -xf rec.tar
+
+# Create C header files
+cd ..
+python3 ./convert_image.py imgs_words_en/word_116.png
+
+# Build demo executable
+cd ${script_dir}
+echo ${script_dir}
+make
+
+# Run demo executable on the AVH
+$Platform -C cpu0.CFGDTCMSZ=15 \
+-C cpu0.CFGITCMSZ=15 -C mps3_board.uart0.out_file=\"-\" -C mps3_board.uart0.shutdown_tag=\"EXITTHESIM\" \
+-C mps3_board.visualisation.disable-visualisation=1 -C mps3_board.telnetterminal0.start_telnet=0 \
+-C mps3_board.telnetterminal1.start_telnet=0 -C mps3_board.telnetterminal2.start_telnet=0 -C mps3_board.telnetterminal5.start_telnet=0 \
+./build/demo --stat
diff --git a/deploy/avh/src/demo_bare_metal.c b/deploy/avh/src/demo_bare_metal.c
new file mode 100644
index 0000000..c90e17f
--- /dev/null
+++ b/deploy/avh/src/demo_bare_metal.c
@@ -0,0 +1,78 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *   http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied.  See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+#include <stdio.h>
+#include <tvm_runtime.h>
+#include <tvmgen_rec.h>
+
+#include "uart.h"
+
+// Header files generated by convert_image.py
+#include "inputs.h"
+#include "outputs.h"
+
+int main(int argc, char **argv) {
+  char dict[] = {"#0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\\]^_`"
+                 "abcdefghijklmnopqrstuvwxyz{|}~!\"#$%&'()*+,-./  "};
+  int char_dict_nums = 97;
+  uart_init();
+  printf("Starting ocr rec inference\n");
+  struct tvmgen_rec_outputs rec_outputs = {
+      .output = output,
+  };
+  struct tvmgen_rec_inputs rec_inputs = {
+      .x = input,
+  };
+
+  tvmgen_rec_run(&rec_inputs, &rec_outputs);
+
+  // post process
+  int char_nums = output_len / char_dict_nums;
+
+  int last_index = 0;
+  float score = 0.f;
+  int count = 0;
+
+  printf("text: ");
+  for (int i = 0; i < char_nums; i++) {
+    int argmax_idx = 0;
+    float max_value = 0.0f;
+    for (int j = 0; j < char_dict_nums; j++) {
+      if (output[i * char_dict_nums + j] > max_value) {
+        max_value = output[i * char_dict_nums + j];
+        argmax_idx = j;
+      }
+    }
+    if (argmax_idx > 0 && (!(i > 0 && argmax_idx == last_index))) {
+      score += max_value;
+      count += 1;
+      // printf("%d，%f，%c\n", argmax_idx, max_value, dict[argmax_idx]);
+      printf("%c", dict[argmax_idx]);
+    }
+    last_index = argmax_idx;
+  }
+  score /= count;
+  printf(", score: %f\n", score);
+
+  // The FVP will shut down when it receives "EXITTHESIM" on the UART
+  printf("EXITTHESIM\n");
+  while (1 == 1)
+    ;
+  return 0;
+}
diff --git a/deploy/cpp_infer/CMakeLists.txt b/deploy/cpp_infer/CMakeLists.txt
new file mode 100644
index 0000000..c70bfe0
--- /dev/null
+++ b/deploy/cpp_infer/CMakeLists.txt
@@ -0,0 +1,277 @@
+cmake_minimum_required(VERSION 3.14)
+project(ppocr CXX)
+
+set(DEMO_NAME "ppocr")
+
+set(CMAKE_CXX_STANDARD 11)
+set(CMAKE_CXX_STANDARD_REQUIRED True)
+
+option(WITH_MKL        "Compile demo with MKL/OpenBlas support, default use MKL."       ON)
+option(WITH_GPU        "Compile demo with GPU/CPU, default use CPU."                    OFF)
+option(WITH_STATIC_LIB "Compile demo with static/shared library, default use static."   ON)
+option(USE_FREETYPE "Enable FreeType support" OFF)
+
+SET(PADDLE_LIB "" CACHE PATH "Location of libraries")
+SET(OPENCV_DIR "" CACHE PATH "Location of libraries")
+SET(CUDA_LIB "" CACHE PATH "Location of libraries")
+SET(CUDNN_LIB "" CACHE PATH "Location of libraries")
+
+macro(safe_set_static_flag)
+    foreach(flag_var
+        CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_DEBUG CMAKE_CXX_FLAGS_RELEASE
+        CMAKE_CXX_FLAGS_MINSIZEREL CMAKE_CXX_FLAGS_RELWITHDEBINFO)
+      if(${${flag_var}} MATCHES "/MD")
+        string(REGEX REPLACE "/MD" "/MT" ${flag_var} "${${flag_var}}")
+      endif()
+    endforeach(flag_var)
+endmacro()
+
+if (WITH_MKL)
+    ADD_DEFINITIONS(-DUSE_MKL)
+endif()
+
+if(NOT DEFINED PADDLE_LIB)
+  message(FATAL_ERROR "please set PADDLE_LIB with -DPADDLE_LIB=/path/paddle/lib")
+endif()
+
+if(NOT DEFINED OPENCV_DIR)
+    message(FATAL_ERROR "please set OPENCV_DIR with -DOPENCV_DIR=/path/opencv")
+endif()
+
+
+if (WIN32)
+  include_directories("${PADDLE_LIB}/paddle/include")
+  link_directories("${PADDLE_LIB}/paddle/lib")
+  set(CMAKE_CONFIGURATION_TYPES "Debug;Release" CACHE STRING "" FORCE)
+  set(OpenCV_DIR "${OPENCV_DIR}/x64/vc16/lib") 
+  find_package(OpenCV REQUIRED )
+  if(USE_FREETYPE)
+    if(NOT "opencv_freetype" IN_LIST OpenCV_LIBS)
+      message(FATAL_ERROR "OpenCV was not compiled with the freetype module (opencv_freetype) !")
+    endif()
+    add_definitions(-DUSE_FREETYPE)
+  endif()  
+
+else ()
+  set(OpenCV_DIR "${OPENCV_DIR}/lib64/cmake/opencv4") 
+  find_package(OpenCV REQUIRED)
+  if(USE_FREETYPE)
+    if(NOT "opencv_freetype" IN_LIST OpenCV_LIBS)
+      message(FATAL_ERROR "OpenCV was not compiled with the freetype module (opencv_freetype) !")
+    endif()
+    add_definitions(-DUSE_FREETYPE)
+  endif()  
+
+  include_directories("${PADDLE_LIB}/paddle/include")
+  link_directories("${PADDLE_LIB}/paddle/lib")
+endif ()
+include_directories(${OpenCV_INCLUDE_DIRS})
+
+if (WIN32)
+    add_definitions("/DGOOGLE_GLOG_DLL_DECL=")
+    if(WITH_MKL)
+        set(FLAG_OPENMP "/openmp")
+    endif()
+    set(CMAKE_C_FLAGS_DEBUG   "${CMAKE_C_FLAGS_DEBUG} /bigobj /MTd ${FLAG_OPENMP}")
+    set(CMAKE_C_FLAGS_RELEASE  "${CMAKE_C_FLAGS_RELEASE} /bigobj /MT ${FLAG_OPENMP}")
+    set(CMAKE_CXX_FLAGS_DEBUG  "${CMAKE_CXX_FLAGS_DEBUG} /bigobj /MTd ${FLAG_OPENMP}")
+    set(CMAKE_CXX_FLAGS_RELEASE   "${CMAKE_CXX_FLAGS_RELEASE} /bigobj /MT ${FLAG_OPENMP}")
+    if (WITH_STATIC_LIB)
+        safe_set_static_flag()
+        add_definitions(-DSTATIC_LIB)
+	add_definitions(-DYAML_CPP_STATIC_DEFINE)
+    endif()
+    message("cmake c debug flags " ${CMAKE_C_FLAGS_DEBUG})
+    message("cmake c release flags " ${CMAKE_C_FLAGS_RELEASE})
+    message("cmake cxx debug flags " ${CMAKE_CXX_FLAGS_DEBUG})
+    message("cmake cxx release flags " ${CMAKE_CXX_FLAGS_RELEASE})
+else()
+    if(WITH_MKL)
+        set(FLAG_OPENMP "-fopenmp")
+    endif()
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -g -O3 ${FLAG_OPENMP} -std=c++11")
+    set(CMAKE_STATIC_LIBRARY_PREFIX "")
+    message("cmake cxx flags" ${CMAKE_CXX_FLAGS})
+endif()
+
+if (WITH_GPU)
+    if (NOT DEFINED CUDA_LIB OR ${CUDA_LIB} STREQUAL "")
+        message(FATAL_ERROR "please set CUDA_LIB with -DCUDA_LIB=/path/cuda-8.0/lib64")
+    endif()
+    if (NOT WIN32)
+        if (NOT DEFINED CUDNN_LIB)
+            message(FATAL_ERROR "please set CUDNN_LIB with -DCUDNN_LIB=/path/cudnn_v7.4/cuda/lib64")
+        endif()
+    add_definitions(-DWITH_GPU)
+    endif(NOT WIN32)
+endif()
+
+include_directories("${PADDLE_LIB}/third_party/install/protobuf/include")
+include_directories("${PADDLE_LIB}/third_party/install/glog/include")
+include_directories("${PADDLE_LIB}/third_party/install/gflags/include")
+include_directories("${PADDLE_LIB}/third_party/install/xxhash/include")
+include_directories("${PADDLE_LIB}/third_party/install/zlib/include")
+include_directories("${PADDLE_LIB}/third_party/install/onnxruntime/include")
+include_directories("${PADDLE_LIB}/third_party/install/paddle2onnx/include")
+include_directories("${PADDLE_LIB}/third_party/install/yaml-cpp/include")
+include_directories("${PADDLE_LIB}/third_party/install/openvino/include")
+include_directories("${PADDLE_LIB}/third_party/install/tbb/include")
+include_directories("${PADDLE_LIB}/third_party/boost")
+include_directories("${PADDLE_LIB}/third_party/eigen3")
+include_directories("${PADDLE_LIB}/paddle/include/") 
+include_directories("${CMAKE_SOURCE_DIR}/")
+
+link_directories("${PADDLE_LIB}/third_party/install/zlib/lib")
+link_directories("${PADDLE_LIB}/third_party/install/protobuf/lib")
+link_directories("${PADDLE_LIB}/third_party/install/glog/lib")
+link_directories("${PADDLE_LIB}/third_party/install/gflags/lib")
+link_directories("${PADDLE_LIB}/third_party/install/xxhash/lib")
+link_directories("${PADDLE_LIB}/third_party/install/onnxruntime/lib")
+link_directories("${PADDLE_LIB}/third_party/install/paddle2onnx/lib")
+link_directories("${PADDLE_LIB}/third_party/install/yaml-cpp/lib")
+link_directories("${PADDLE_LIB}/third_party/install/openvino/intel64")
+link_directories("${PADDLE_LIB}/third_party/install/tbb/lib")
+link_directories("${PADDLE_LIB}/paddle/lib")
+
+if(WITH_MKL)
+  include_directories("${PADDLE_LIB}/third_party/install/mklml/include")
+  if (WIN32)
+    set(MATH_LIB ${PADDLE_LIB}/third_party/install/mklml/lib/mklml.lib
+            ${PADDLE_LIB}/third_party/install/mklml/lib/libiomp5md.lib)
+  else ()
+    set(MATH_LIB ${PADDLE_LIB}/third_party/install/mklml/lib/libmklml_intel${CMAKE_SHARED_LIBRARY_SUFFIX}
+            ${PADDLE_LIB}/third_party/install/mklml/lib/libiomp5${CMAKE_SHARED_LIBRARY_SUFFIX})
+    execute_process(COMMAND cp -r ${PADDLE_LIB}/third_party/install/mklml/lib/libmklml_intel${CMAKE_SHARED_LIBRARY_SUFFIX} /usr/lib)
+  endif ()
+  set(MKLDNN_PATH "${PADDLE_LIB}/third_party/install/onednn")
+  if(EXISTS ${MKLDNN_PATH})
+    include_directories("${MKLDNN_PATH}/include")
+    if (WIN32)
+      set(MKLDNN_LIB ${MKLDNN_PATH}/lib/mkldnn.lib)
+    else ()
+      set(MKLDNN_LIB ${MKLDNN_PATH}/lib/libdnnl.so.3)
+    endif ()
+  endif()
+else()
+  if (WIN32)
+    set(MATH_LIB ${PADDLE_LIB}/third_party/install/openblas/lib/openblas${CMAKE_STATIC_LIBRARY_SUFFIX})
+  else ()
+    set(MATH_LIB ${PADDLE_LIB}/third_party/install/openblas/lib/libopenblas${CMAKE_STATIC_LIBRARY_SUFFIX})
+  endif ()
+endif()
+
+# Note: libpaddle_inference_api.so/a must put before libpaddle_inference.so/a
+if(WITH_STATIC_LIB)
+  if(WIN32)
+    set(DEPS
+        ${PADDLE_LIB}/paddle/lib/paddle_inference${CMAKE_STATIC_LIBRARY_SUFFIX})
+  else()
+    set(DEPS
+        ${PADDLE_LIB}/paddle/lib/libpaddle_inference${CMAKE_STATIC_LIBRARY_SUFFIX})
+  endif()
+else()
+  if(WIN32)
+    set(DEPS
+        ${PADDLE_LIB}/paddle/lib/paddle_inference${CMAKE_SHARED_LIBRARY_SUFFIX})
+  else()
+    set(DEPS
+        ${PADDLE_LIB}/paddle/lib/libpaddle_inference${CMAKE_SHARED_LIBRARY_SUFFIX})
+  endif()
+endif(WITH_STATIC_LIB)
+
+if (NOT WIN32)
+    set(DEPS ${DEPS}
+        ${MATH_LIB} ${MKLDNN_LIB}
+        glog gflags protobuf z xxhash
+        )
+    if(EXISTS "${PADDLE_LIB}/third_party/install/snappystream/lib")
+        set(DEPS ${DEPS} snappystream)
+    endif()
+    if (EXISTS "${PADDLE_LIB}/third_party/install/snappy/lib")
+        set(DEPS ${DEPS} snappy)
+    endif()
+else()
+    set(DEPS ${DEPS}
+        ${MATH_LIB} ${MKLDNN_LIB}
+        glog gflags_static libprotobuf xxhash)
+    set(DEPS ${DEPS} libcmt shlwapi)
+    if (EXISTS "${PADDLE_LIB}/third_party/install/snappy/lib")
+        set(DEPS ${DEPS} snappy)
+    endif()
+    if(EXISTS "${PADDLE_LIB}/third_party/install/snappystream/lib")
+        set(DEPS ${DEPS} snappystream)
+    endif()
+endif(NOT WIN32)
+
+if (EXISTS "${PADDLE_LIB}/third_party/install/yaml-cpp/lib")
+  set(DEPS ${DEPS} yaml-cpp)
+endif()
+
+if(WITH_GPU)
+  if(NOT WIN32)
+    set(DEPS ${DEPS} ${CUDA_LIB}/libcudart${CMAKE_SHARED_LIBRARY_SUFFIX})
+    set(DEPS ${DEPS} ${CUDNN_LIB}/libcudnn${CMAKE_SHARED_LIBRARY_SUFFIX})
+  else()
+    set(DEPS ${DEPS} ${CUDA_LIB}/cudart${CMAKE_STATIC_LIBRARY_SUFFIX} )
+    message($DEPS)
+    set(DEPS ${DEPS} ${CUDA_LIB}/cublas${CMAKE_STATIC_LIBRARY_SUFFIX} )
+    set(DEPS ${DEPS} ${CUDNN_LIB}/cudnn${CMAKE_STATIC_LIBRARY_SUFFIX})
+  endif()
+endif()
+
+
+if (NOT WIN32)
+    set(EXTERNAL_LIB "-ldl -lrt -lgomp -lz -lm -lpthread")
+    set(DEPS ${DEPS} ${EXTERNAL_LIB})
+endif()
+
+set(THIRD_PARTY_PATH ${CMAKE_CURRENT_LIST_DIR}/third_party)
+function(download_and_decompress url filename decompress_dir)
+  if(NOT EXISTS "${filename}" AND NOT EXISTS "${decompress_dir}")
+    message("Downloading file from ${url} to ${filename} ...")
+    file(DOWNLOAD ${url} "${filename}.tmp" SHOW_PROGRESS)
+    file(RENAME "${filename}.tmp" ${filename})
+  endif()
+  if(NOT EXISTS ${decompress_dir})
+    file(MAKE_DIRECTORY ${decompress_dir})
+    message("Decompress file ${filename} ...")
+    execute_process(COMMAND ${CMAKE_COMMAND} -E tar -xf ${filename} WORKING_DIRECTORY ${decompress_dir})
+  endif()
+endfunction()
+
+
+set(PACKAGE_LIST abseil-cpp clipper_ver6.4.2 nlohmann)
+foreach(PKG ${PACKAGE_LIST})
+    set(PKG_URL "https://paddle-model-ecology.bj.bcebos.com/paddlex/cpp/libs/${PKG}.tgz")
+    set(PKG_TGZ_PATH "${CMAKE_CURRENT_BINARY_DIR}/${PKG}.tgz")
+    set(PKG_DST_PATH "${THIRD_PARTY_PATH}/${PKG}")
+    download_and_decompress(${PKG_URL} ${PKG_TGZ_PATH} ${PKG_DST_PATH})
+endforeach()
+
+add_subdirectory(third_party/abseil-cpp)
+add_subdirectory(third_party/clipper_ver6.4.2/cpp)
+include_directories(${POLYCLIPPING_INCLUDE_DIR})
+
+set(DEPS ${DEPS} ${OpenCV_LIBS})
+set(DEPS ${DEPS} absl::statusor)
+set(DEPS ${DEPS} polyclipping)
+
+if(UNIX)
+  find_package(Iconv REQUIRED) 
+endif()
+
+file(GLOB_RECURSE SRC_LIST "./src/*.cc")
+set(SRCS cli.cc )
+add_executable(${DEMO_NAME} ${SRCS} ${SRC_LIST} )
+target_link_libraries(${DEMO_NAME} ${DEPS} )
+
+if (WIN32 AND WITH_MKL)
+    add_custom_command(TARGET ${DEMO_NAME} POST_BUILD
+        COMMAND ${CMAKE_COMMAND} -E copy_if_different ${PADDLE_LIB}/third_party/install/mklml/lib/mklml.dll ./mklml.dll
+        COMMAND ${CMAKE_COMMAND} -E copy_if_different ${PADDLE_LIB}/third_party/install/mklml/lib/libiomp5md.dll ./libiomp5md.dll
+        COMMAND ${CMAKE_COMMAND} -E copy_if_different ${PADDLE_LIB}/third_party/install/onednn/lib/mkldnn.dll ./mkldnn.dll
+        COMMAND ${CMAKE_COMMAND} -E copy_if_different ${PADDLE_LIB}/third_party/install/mklml/lib/mklml.dll ./release/mklml.dll
+        COMMAND ${CMAKE_COMMAND} -E copy_if_different ${PADDLE_LIB}/third_party/install/mklml/lib/libiomp5md.dll ./release/libiomp5md.dll
+        COMMAND ${CMAKE_COMMAND} -E copy_if_different ${PADDLE_LIB}/third_party/install/onednn/lib/mkldnn.dll ./release/mkldnn.dll
+    )
+endif()
diff --git a/deploy/cpp_infer/THIRD_PARTY_LICENSES/tensorRT_Pro/LICENSE b/deploy/cpp_infer/THIRD_PARTY_LICENSES/tensorRT_Pro/LICENSE
new file mode 100644
index 0000000..c75021b
--- /dev/null
+++ b/deploy/cpp_infer/THIRD_PARTY_LICENSES/tensorRT_Pro/LICENSE
@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2022 TensorRTPro
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
diff --git a/deploy/cpp_infer/cli.cc b/deploy/cpp_infer/cli.cc
new file mode 100644
index 0000000..a5b526e
--- /dev/null
+++ b/deploy/cpp_infer/cli.cc
@@ -0,0 +1,327 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "src/api/models/doc_img_orientation_classification.h"
+#include "src/api/models/text_detection.h"
+#include "src/api/models/text_image_unwarping.h"
+#include "src/api/models/text_recognition.h"
+#include "src/api/models/textline_orientation_classification.h"
+#include "src/api/pipelines/doc_preprocessor.h"
+#include "src/api/pipelines/ocr.h"
+#include "src/utils/args.h"
+#include <functional>
+#include <iostream>
+#include <memory>
+#include <string>
+#include <tuple>
+#include <unordered_map>
+#include <vector>
+
+static const std::unordered_set<std::string> SUPPORT_MODE_PIPELINE = {
+    "ocr",
+    "doc_preprocessor",
+};
+
+static const std::unordered_set<std::string> SUPPORT_MODE_MODEL = {
+    "text_image_unwarping", "doc_img_orientation_classification",
+    "textline_orientation_classification", "text_detection",
+    "text_recognition"};
+
+void PrintErrorInfo(const std::string &msg, const std::string &main_mode = "") {
+  auto join_modes =
+      [](const std::unordered_set<std::string> &modes) -> std::string {
+    std::string result;
+    for (const auto &mode : modes) {
+      result += mode + ", ";
+    }
+    if (!result.empty()) {
+      result.pop_back();
+      result.pop_back();
+    }
+    return result;
+  };
+
+  std::string pipeline_modes = join_modes(SUPPORT_MODE_PIPELINE);
+  std::string model_modes = join_modes(SUPPORT_MODE_MODEL);
+
+  INFOE("%s%s", msg.c_str(),
+        main_mode.empty() ? "" : (": \"" + main_mode + "\"").c_str());
+  INFO("==========================================");
+  INFO("Supported pipeline : [%s]", pipeline_modes.c_str());
+  INFO("Supported model    : [%s]", model_modes.c_str());
+  INFO("==========================================");
+}
+
+std::tuple<PaddleOCRParams, DocPreprocessorParams,
+           DocImgOrientationClassificationParams, TextImageUnwarpingParams,
+           TextDetectionParams, TextLineOrientationClassificationParams,
+           TextRecognitionParams>
+GetPipelineMoudleParams() {
+  PaddleOCRParams ocr_params;
+  DocPreprocessorParams doc_pre_params;
+  DocImgOrientationClassificationParams doc_orient_params;
+  TextImageUnwarpingParams unwarp_params;
+  TextDetectionParams det_params;
+  TextLineOrientationClassificationParams teline_orient_params;
+  TextRecognitionParams rec_params;
+  if (!FLAGS_doc_orientation_classify_model_name.empty()) {
+    ocr_params.doc_orientation_classify_model_name =
+        FLAGS_doc_orientation_classify_model_name;
+    doc_pre_params.doc_orientation_classify_model_name =
+        FLAGS_doc_orientation_classify_model_name;
+    doc_orient_params.model_name = FLAGS_doc_orientation_classify_model_name;
+  }
+  if (!FLAGS_doc_orientation_classify_model_dir.empty()) {
+    ocr_params.doc_orientation_classify_model_dir =
+        FLAGS_doc_orientation_classify_model_dir;
+    doc_pre_params.doc_orientation_classify_model_dir =
+        FLAGS_doc_orientation_classify_model_dir;
+    doc_orient_params.model_dir = FLAGS_doc_orientation_classify_model_dir;
+  }
+  if (!FLAGS_doc_unwarping_model_name.empty()) {
+    ocr_params.doc_unwarping_model_name = FLAGS_doc_unwarping_model_name;
+    doc_pre_params.doc_unwarping_model_name = FLAGS_doc_unwarping_model_name;
+    unwarp_params.model_name = FLAGS_doc_unwarping_model_name;
+  }
+  if (!FLAGS_doc_unwarping_model_dir.empty()) {
+    ocr_params.doc_unwarping_model_dir = FLAGS_doc_unwarping_model_dir;
+    doc_pre_params.doc_unwarping_model_dir = FLAGS_doc_unwarping_model_dir;
+    unwarp_params.model_dir = FLAGS_doc_unwarping_model_dir;
+  }
+  if (!FLAGS_text_detection_model_name.empty()) {
+    ocr_params.text_detection_model_name = FLAGS_text_detection_model_name;
+    det_params.model_name = FLAGS_text_detection_model_name;
+  }
+  if (!FLAGS_text_detection_model_dir.empty()) {
+    ocr_params.text_detection_model_dir = FLAGS_text_detection_model_dir;
+    det_params.model_dir = FLAGS_text_detection_model_dir;
+  }
+  if (!FLAGS_textline_orientation_model_name.empty()) {
+    ocr_params.textline_orientation_model_name =
+        FLAGS_textline_orientation_model_name;
+    teline_orient_params.model_name = FLAGS_textline_orientation_model_name;
+  }
+  if (!FLAGS_textline_orientation_model_dir.empty()) {
+    ocr_params.textline_orientation_model_dir =
+        FLAGS_textline_orientation_model_dir;
+    teline_orient_params.model_dir = FLAGS_textline_orientation_model_dir;
+  }
+  if (!FLAGS_textline_orientation_batch_size.empty()) {
+    ocr_params.textline_orientation_batch_size =
+        std::stoi(FLAGS_textline_orientation_batch_size);
+  }
+  if (!FLAGS_text_recognition_model_name.empty()) {
+    ocr_params.text_recognition_model_name = FLAGS_text_recognition_model_name;
+    rec_params.model_name = FLAGS_text_recognition_model_name;
+  }
+  if (!FLAGS_text_recognition_model_dir.empty()) {
+    ocr_params.text_recognition_model_dir = FLAGS_text_recognition_model_dir;
+    rec_params.model_dir = FLAGS_text_recognition_model_dir;
+  }
+  if (!FLAGS_text_recognition_batch_size.empty()) {
+    ocr_params.text_recognition_batch_size =
+        std::stoi(FLAGS_text_recognition_batch_size);
+    rec_params.batch_size = std::stoi(FLAGS_text_recognition_batch_size);
+    rec_params.input_shape =
+        YamlConfig::SmartParseVector(FLAGS_text_rec_input_shape).vec_int;
+  }
+  if (!FLAGS_use_doc_orientation_classify.empty()) {
+    ocr_params.use_doc_orientation_classify =
+        Utility::StringToBool(FLAGS_use_doc_orientation_classify);
+    doc_pre_params.use_doc_orientation_classify =
+        Utility::StringToBool(FLAGS_use_doc_orientation_classify);
+  }
+  if (!FLAGS_use_doc_unwarping.empty()) {
+    ocr_params.use_doc_unwarping =
+        Utility::StringToBool(FLAGS_use_doc_unwarping);
+    doc_pre_params.use_doc_unwarping =
+        Utility::StringToBool(FLAGS_use_doc_unwarping);
+  }
+  if (!FLAGS_use_textline_orientation.empty()) {
+    ocr_params.use_textline_orientation =
+        Utility::StringToBool(FLAGS_use_textline_orientation);
+  }
+  if (!FLAGS_text_det_limit_side_len.empty()) {
+    ocr_params.text_det_limit_side_len =
+        std::stoi(FLAGS_text_det_limit_side_len);
+  }
+  if (!FLAGS_text_det_limit_type.empty()) {
+    ocr_params.text_det_limit_type = FLAGS_text_det_limit_type;
+    det_params.limit_type = FLAGS_text_det_limit_type;
+  }
+  if (!FLAGS_text_det_thresh.empty()) {
+    ocr_params.text_det_thresh = std::stof(FLAGS_text_det_thresh);
+    det_params.thresh = std::stof(FLAGS_text_det_thresh);
+  }
+  if (!FLAGS_text_det_box_thresh.empty()) {
+    ocr_params.text_det_box_thresh = std::stof(FLAGS_text_det_box_thresh);
+    det_params.box_thresh = std::stof(FLAGS_text_det_box_thresh);
+  }
+  if (!FLAGS_text_det_unclip_ratio.empty()) {
+    ocr_params.text_det_unclip_ratio = std::stof(FLAGS_text_det_unclip_ratio);
+    det_params.unclip_ratio = std::stof(FLAGS_text_det_unclip_ratio);
+  }
+  if (!FLAGS_text_det_input_shape.empty()) {
+    ocr_params.text_det_input_shape =
+        YamlConfig::SmartParseVector(FLAGS_text_det_input_shape).vec_int;
+    det_params.input_shape =
+        YamlConfig::SmartParseVector(FLAGS_text_det_input_shape).vec_int;
+  }
+  if (!FLAGS_text_rec_score_thresh.empty()) {
+    ocr_params.text_rec_score_thresh = std::stof(FLAGS_text_rec_score_thresh);
+  }
+  if (!FLAGS_text_rec_input_shape.empty()) {
+    ocr_params.text_rec_input_shape =
+        YamlConfig::SmartParseVector(FLAGS_text_rec_input_shape).vec_int;
+  }
+  if (!FLAGS_lang.empty()) {
+    ocr_params.lang = FLAGS_lang;
+  }
+  if (!FLAGS_ocr_version.empty()) {
+    ocr_params.ocr_version = FLAGS_ocr_version;
+  }
+  if (!FLAGS_vis_font_dir.empty()) {
+    ocr_params.vis_font_dir = FLAGS_vis_font_dir;
+    rec_params.vis_font_dir = FLAGS_vis_font_dir;
+  }
+  if (!FLAGS_device.empty()) {
+    ocr_params.device = FLAGS_device;
+    doc_pre_params.device = FLAGS_device;
+    doc_orient_params.device = FLAGS_device;
+    unwarp_params.device = FLAGS_device;
+    teline_orient_params.device = FLAGS_device;
+    det_params.device = FLAGS_device;
+    rec_params.device = FLAGS_device;
+  }
+  if (!FLAGS_precision.empty()) {
+    ocr_params.precision = FLAGS_precision;
+    doc_pre_params.precision = FLAGS_precision;
+    doc_orient_params.precision = FLAGS_precision;
+    unwarp_params.precision = FLAGS_precision;
+    teline_orient_params.precision = FLAGS_precision;
+    det_params.precision = FLAGS_precision;
+    rec_params.precision = FLAGS_precision;
+  }
+  if (!FLAGS_enable_mkldnn.empty()) {
+    ocr_params.enable_mkldnn = Utility::StringToBool(FLAGS_enable_mkldnn);
+    doc_pre_params.enable_mkldnn = Utility::StringToBool(FLAGS_enable_mkldnn);
+    doc_orient_params.enable_mkldnn =
+        Utility::StringToBool(FLAGS_enable_mkldnn);
+    unwarp_params.enable_mkldnn = Utility::StringToBool(FLAGS_enable_mkldnn);
+    teline_orient_params.enable_mkldnn =
+        Utility::StringToBool(FLAGS_enable_mkldnn);
+    det_params.enable_mkldnn = Utility::StringToBool(FLAGS_enable_mkldnn);
+    rec_params.enable_mkldnn = Utility::StringToBool(FLAGS_enable_mkldnn);
+  }
+  if (!FLAGS_mkldnn_cache_capacity.empty()) {
+    ocr_params.mkldnn_cache_capacity = std::stoi(FLAGS_mkldnn_cache_capacity);
+    doc_pre_params.mkldnn_cache_capacity =
+        std::stoi(FLAGS_mkldnn_cache_capacity);
+    doc_orient_params.mkldnn_cache_capacity =
+        std::stoi(FLAGS_mkldnn_cache_capacity);
+    unwarp_params.mkldnn_cache_capacity =
+        std::stoi(FLAGS_mkldnn_cache_capacity);
+    teline_orient_params.mkldnn_cache_capacity =
+        std::stoi(FLAGS_mkldnn_cache_capacity);
+    det_params.mkldnn_cache_capacity = std::stoi(FLAGS_mkldnn_cache_capacity);
+    rec_params.mkldnn_cache_capacity = std::stoi(FLAGS_mkldnn_cache_capacity);
+  }
+  if (!FLAGS_cpu_threads.empty()) {
+    ocr_params.cpu_threads = std::stoi(FLAGS_cpu_threads);
+    doc_pre_params.cpu_threads = std::stoi(FLAGS_cpu_threads);
+    doc_orient_params.cpu_threads = std::stoi(FLAGS_cpu_threads);
+    unwarp_params.cpu_threads = std::stoi(FLAGS_cpu_threads);
+    teline_orient_params.cpu_threads = std::stoi(FLAGS_cpu_threads);
+    det_params.cpu_threads = std::stoi(FLAGS_cpu_threads);
+    rec_params.cpu_threads = std::stoi(FLAGS_cpu_threads);
+  }
+  if (!FLAGS_thread_num.empty()) {
+    ocr_params.thread_num = std::stoi(FLAGS_thread_num);
+    doc_pre_params.thread_num = std::stoi(FLAGS_thread_num);
+  }
+  if (!FLAGS_paddlex_config.empty()) {
+    ocr_params.paddlex_config = FLAGS_paddlex_config;
+    doc_pre_params.paddlex_config = FLAGS_paddlex_config;
+  }
+  return std::make_tuple(ocr_params, doc_pre_params, doc_orient_params,
+                         unwarp_params, det_params, teline_orient_params,
+                         rec_params);
+}
+
+int main(int argc, char *argv[]) {
+  gflags::ParseCommandLineFlags(&argc, &argv, true);
+  if (FLAGS_input.empty()) {
+    INFOE("Require input, such as ./build/ppocr <pipeline_or_module> --input "
+          "your_image_path [--param1] [--param2] [...]");
+    exit(-1);
+  }
+  std::string main_mode = "";
+  if (argc > 1) {
+    main_mode = argv[1];
+    if (SUPPORT_MODE_PIPELINE.count(main_mode) == 0 &&
+        SUPPORT_MODE_MODEL.count(main_mode) == 0) {
+      PrintErrorInfo("ERROR: Unsupported pipeline or module", main_mode);
+      exit(-1);
+    }
+  } else {
+    PrintErrorInfo(
+        "Must provide pipeline or module name, such as ./build/ppocr "
+        "<pipeline_or_module> [--param1] [--param2] [...]");
+    exit(-1);
+  }
+  auto params = GetPipelineMoudleParams();
+  using PredFunc = std::function<std::vector<std::unique_ptr<BaseCVResult>>(
+      const std::string &)>;
+  std::unordered_map<std::string, PredFunc> pred_map = {
+      {"ocr",
+       [&params](const std::string &input) {
+         return PaddleOCR(std::get<0>(params)).Predict(input);
+       }},
+      {"doc_preprocessor",
+       [&params](const std::string &input) {
+         return DocPreprocessor(std::get<1>(params)).Predict(input);
+       }},
+      {"doc_img_orientation_classification",
+       [&params](const std::string &input) {
+         return DocImgOrientationClassification(std::get<2>(params))
+             .Predict(input);
+       }},
+      {"text_image_unwarping",
+       [&params](const std::string &input) {
+         return TextImageUnwarping(std::get<3>(params)).Predict(input);
+       }},
+      {"text_detection",
+       [&params](const std::string &input) {
+         return TextDetection(std::get<4>(params)).Predict(input);
+       }},
+      {"textline_orientation_classification",
+       [&params](const std::string &input) {
+         return TextLineOrientationClassification(std::get<5>(params))
+             .Predict(input);
+       }},
+      {"text_recognition",
+       [&params](const std::string &input) {
+         return TextRecognition(std::get<6>(params)).Predict(input);
+       }},
+
+  };
+  auto it = pred_map.find(main_mode);
+  auto outputs = it->second(FLAGS_input);
+  for (auto &output : outputs) {
+    output->Print();
+    output->SaveToImg(FLAGS_save_path);
+    output->SaveToJson(FLAGS_save_path);
+  }
+  return 0;
+}
diff --git a/deploy/cpp_infer/src/api/models/doc_img_orientation_classification.cc b/deploy/cpp_infer/src/api/models/doc_img_orientation_classification.cc
new file mode 100644
index 0000000..9c273ce
--- /dev/null
+++ b/deploy/cpp_infer/src/api/models/doc_img_orientation_classification.cc
@@ -0,0 +1,62 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "doc_img_orientation_classification.h"
+
+#include "src/utils/args.h"
+#include "src/utils/yaml_config.h"
+
+#define COPY_PARAMS(field) to.field = from.field;
+
+DocImgOrientationClassification::DocImgOrientationClassification(
+    const DocImgOrientationClassificationParams &params)
+    : params_(params) {
+  auto status = CheckParams();
+  if (!status.ok()) {
+    INFOE("Init DocImgOrientationClassification fail : %s",
+          status.ToString().c_str());
+    exit(-1);
+  }
+  CreateModel();
+};
+std::vector<std::unique_ptr<BaseCVResult>>
+DocImgOrientationClassification::Predict(
+    const std::vector<std::string> &input) {
+  return model_infer_->Predict(input);
+}
+void DocImgOrientationClassification::CreateModel() {
+  model_infer_ = std::unique_ptr<BasePredictor>(
+      new ClasPredictor(ToDocImgOrientationClassificationModelParams(params_)));
+}
+
+absl::Status DocImgOrientationClassification::CheckParams() {
+  if (!params_.model_dir.has_value()) {
+    return absl::NotFoundError("Require doc orientation classify model dir.");
+  }
+  return absl::OkStatus();
+}
+
+ClasPredictorParams
+DocImgOrientationClassification::ToDocImgOrientationClassificationModelParams(
+    const DocImgOrientationClassificationParams &from) {
+  ClasPredictorParams to;
+  COPY_PARAMS(model_name)
+  COPY_PARAMS(model_dir)
+  COPY_PARAMS(device)
+  COPY_PARAMS(enable_mkldnn)
+  COPY_PARAMS(mkldnn_cache_capacity)
+  COPY_PARAMS(precision)
+  COPY_PARAMS(cpu_threads)
+  return to;
+}
diff --git a/deploy/cpp_infer/src/api/models/doc_img_orientation_classification.h b/deploy/cpp_infer/src/api/models/doc_img_orientation_classification.h
new file mode 100644
index 0000000..3bf5864
--- /dev/null
+++ b/deploy/cpp_infer/src/api/models/doc_img_orientation_classification.h
@@ -0,0 +1,51 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include "src/modules/image_classification/predictor.h"
+
+struct DocImgOrientationClassificationParams {
+  absl::optional<std::string> model_name = absl::nullopt;
+  absl::optional<std::string> model_dir = absl::nullopt;
+  absl::optional<std::string> device = absl::nullopt;
+  std::string precision = "fp32";
+  bool enable_mkldnn = true;
+  int mkldnn_cache_capacity = 10;
+  int cpu_threads = 8;
+  int batch_size = 1;
+};
+
+class DocImgOrientationClassification {
+public:
+  DocImgOrientationClassification(
+      const DocImgOrientationClassificationParams &params =
+          DocImgOrientationClassificationParams());
+
+  std::vector<std::unique_ptr<BaseCVResult>> Predict(const std::string &input) {
+    std::vector<std::string> inputs = {input};
+    return Predict(inputs);
+  };
+  std::vector<std::unique_ptr<BaseCVResult>>
+  Predict(const std::vector<std::string> &input);
+
+  void CreateModel();
+  absl::Status CheckParams();
+  static ClasPredictorParams ToDocImgOrientationClassificationModelParams(
+      const DocImgOrientationClassificationParams &from);
+
+private:
+  DocImgOrientationClassificationParams params_;
+  std::unique_ptr<BasePredictor> model_infer_;
+};
diff --git a/deploy/cpp_infer/src/api/models/text_detection.cc b/deploy/cpp_infer/src/api/models/text_detection.cc
new file mode 100644
index 0000000..952aaea
--- /dev/null
+++ b/deploy/cpp_infer/src/api/models/text_detection.cc
@@ -0,0 +1,64 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "text_detection.h"
+
+#include "src/utils/args.h"
+#include "src/utils/yaml_config.h"
+
+#define COPY_PARAMS(field) to.field = from.field;
+
+TextDetection::TextDetection(const TextDetectionParams &params)
+    : params_(params) {
+  auto status = CheckParams();
+  if (!status.ok()) {
+    INFOE("Init TextDetection fail : %s", status.ToString().c_str());
+    exit(-1);
+  }
+  CreateModel();
+};
+std::vector<std::unique_ptr<BaseCVResult>>
+TextDetection::Predict(const std::vector<std::string> &input) {
+  return model_infer_->Predict(input);
+}
+void TextDetection::CreateModel() {
+  model_infer_ = std::unique_ptr<BasePredictor>(
+      new TextDetPredictor(ToTextDetectionModelParams(params_)));
+}
+
+absl::Status TextDetection::CheckParams() {
+  if (!params_.model_dir.has_value()) {
+    return absl::NotFoundError("Require text detection model dir.");
+  }
+  return absl::OkStatus();
+}
+
+TextDetPredictorParams
+TextDetection::ToTextDetectionModelParams(const TextDetectionParams &from) {
+  TextDetPredictorParams to;
+  COPY_PARAMS(model_name)
+  COPY_PARAMS(model_dir)
+  COPY_PARAMS(limit_side_len)
+  COPY_PARAMS(limit_type)
+  COPY_PARAMS(thresh)
+  COPY_PARAMS(box_thresh)
+  COPY_PARAMS(unclip_ratio)
+  COPY_PARAMS(input_shape)
+  COPY_PARAMS(device)
+  COPY_PARAMS(enable_mkldnn)
+  COPY_PARAMS(mkldnn_cache_capacity)
+  COPY_PARAMS(precision)
+  COPY_PARAMS(cpu_threads)
+  return to;
+}
diff --git a/deploy/cpp_infer/src/api/models/text_detection.h b/deploy/cpp_infer/src/api/models/text_detection.h
new file mode 100644
index 0000000..822c4b4
--- /dev/null
+++ b/deploy/cpp_infer/src/api/models/text_detection.h
@@ -0,0 +1,56 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include "src/modules/text_detection/predictor.h"
+
+struct TextDetectionParams {
+  absl::optional<std::string> model_name = absl::nullopt;
+  absl::optional<std::string> model_dir = absl::nullopt;
+  absl::optional<std::string> device = absl::nullopt;
+  std::string precision = "fp32";
+  bool enable_mkldnn = true;
+  int mkldnn_cache_capacity = 10;
+  int cpu_threads = 8;
+  int batch_size = 1;
+  absl::optional<int> limit_side_len = absl::nullopt;
+  absl::optional<std::string> limit_type = absl::nullopt;
+  absl::optional<int> max_side_limit = absl::nullopt;
+  absl::optional<float> thresh = absl::nullopt;
+  absl::optional<float> box_thresh = absl::nullopt;
+  absl::optional<float> unclip_ratio = absl::nullopt;
+  absl::optional<std::vector<int>> input_shape = absl::nullopt;
+};
+
+class TextDetection {
+public:
+  TextDetection(const TextDetectionParams &params = TextDetectionParams());
+
+  std::vector<std::unique_ptr<BaseCVResult>> Predict(const std::string &input) {
+    std::vector<std::string> inputs = {input};
+    return Predict(inputs);
+  };
+  std::vector<std::unique_ptr<BaseCVResult>>
+  Predict(const std::vector<std::string> &input);
+
+  void CreateModel();
+  absl::Status CheckParams();
+  static TextDetPredictorParams
+  ToTextDetectionModelParams(const TextDetectionParams &from);
+
+private:
+  TextDetectionParams params_;
+  std::unique_ptr<BasePredictor> model_infer_;
+};
diff --git a/deploy/cpp_infer/src/api/models/text_image_unwarping.cc b/deploy/cpp_infer/src/api/models/text_image_unwarping.cc
new file mode 100644
index 0000000..b0b4acc
--- /dev/null
+++ b/deploy/cpp_infer/src/api/models/text_image_unwarping.cc
@@ -0,0 +1,58 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "text_image_unwarping.h"
+
+#include "src/utils/args.h"
+#include "src/utils/yaml_config.h"
+
+#define COPY_PARAMS(field) to.field = from.field;
+
+TextImageUnwarping::TextImageUnwarping(const TextImageUnwarpingParams &params)
+    : params_(params) {
+  auto status = CheckParams();
+  if (!status.ok()) {
+    INFOE("Init TextImageUnwarping fail : %s", status.ToString().c_str());
+    exit(-1);
+  }
+  CreateModel();
+};
+std::vector<std::unique_ptr<BaseCVResult>>
+TextImageUnwarping::Predict(const std::vector<std::string> &input) {
+  return model_infer_->Predict(input);
+}
+void TextImageUnwarping::CreateModel() {
+  model_infer_ = std::unique_ptr<BasePredictor>(
+      new WarpPredictor(ToTextImageUnwarpingModelParams(params_)));
+}
+
+absl::Status TextImageUnwarping::CheckParams() {
+  if (!params_.model_dir.has_value()) {
+    return absl::NotFoundError("Require doc unwarping model dir.");
+  }
+  return absl::OkStatus();
+}
+
+WarpPredictorParams TextImageUnwarping::ToTextImageUnwarpingModelParams(
+    const TextImageUnwarpingParams &from) {
+  WarpPredictorParams to;
+  COPY_PARAMS(model_name)
+  COPY_PARAMS(model_dir)
+  COPY_PARAMS(device)
+  COPY_PARAMS(enable_mkldnn)
+  COPY_PARAMS(mkldnn_cache_capacity)
+  COPY_PARAMS(precision)
+  COPY_PARAMS(cpu_threads)
+  return to;
+}
diff --git a/deploy/cpp_infer/src/api/models/text_image_unwarping.h b/deploy/cpp_infer/src/api/models/text_image_unwarping.h
new file mode 100644
index 0000000..6d34612
--- /dev/null
+++ b/deploy/cpp_infer/src/api/models/text_image_unwarping.h
@@ -0,0 +1,51 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include "src/modules/image_unwarping/predictor.h"
+
+struct TextImageUnwarpingParams {
+  absl::optional<std::string> model_name = absl::nullopt;
+  absl::optional<std::string> model_dir = absl::nullopt;
+  absl::optional<std::string> device = absl::nullopt;
+  bool enable_mkldnn = true;
+  std::string precision = "fp32";
+  int mkldnn_cache_capacity = 10;
+  int cpu_threads = 8;
+
+  int batch_size = 1;
+};
+
+class TextImageUnwarping {
+public:
+  TextImageUnwarping(
+      const TextImageUnwarpingParams &params = TextImageUnwarpingParams());
+
+  std::vector<std::unique_ptr<BaseCVResult>> Predict(const std::string &input) {
+    std::vector<std::string> inputs = {input};
+    return Predict(inputs);
+  };
+  std::vector<std::unique_ptr<BaseCVResult>>
+  Predict(const std::vector<std::string> &input);
+
+  void CreateModel();
+  absl::Status CheckParams();
+  static WarpPredictorParams
+  ToTextImageUnwarpingModelParams(const TextImageUnwarpingParams &from);
+
+private:
+  TextImageUnwarpingParams params_;
+  std::unique_ptr<BasePredictor> model_infer_;
+};
diff --git a/deploy/cpp_infer/src/api/models/text_recognition.cc b/deploy/cpp_infer/src/api/models/text_recognition.cc
new file mode 100644
index 0000000..6ca09cf
--- /dev/null
+++ b/deploy/cpp_infer/src/api/models/text_recognition.cc
@@ -0,0 +1,61 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "text_recognition.h"
+
+#include "src/utils/args.h"
+#include "src/utils/yaml_config.h"
+
+#define COPY_PARAMS(field) to.field = from.field;
+
+TextRecognition::TextRecognition(const TextRecognitionParams &params)
+    : params_(params) {
+  auto status = CheckParams();
+  if (!status.ok()) {
+    INFOE("Init TextRecognition fail : %s", status.ToString().c_str());
+    exit(-1);
+  }
+  CreateModel();
+};
+std::vector<std::unique_ptr<BaseCVResult>>
+TextRecognition::Predict(const std::vector<std::string> &input) {
+  return model_infer_->Predict(input);
+}
+void TextRecognition::CreateModel() {
+  model_infer_ = std::unique_ptr<BasePredictor>(
+      new TextRecPredictor(ToTextRecognitionModelParams(params_)));
+}
+
+absl::Status TextRecognition::CheckParams() {
+  if (!params_.model_dir.has_value()) {
+    return absl::NotFoundError("Require text recognition model_dir.");
+  }
+  return absl::OkStatus();
+}
+
+TextRecPredictorParams TextRecognition::ToTextRecognitionModelParams(
+    const TextRecognitionParams &from) {
+  TextRecPredictorParams to;
+  COPY_PARAMS(model_name)
+  COPY_PARAMS(model_dir)
+  COPY_PARAMS(batch_size)
+  COPY_PARAMS(input_shape)
+  COPY_PARAMS(vis_font_dir)
+  COPY_PARAMS(device)
+  COPY_PARAMS(enable_mkldnn)
+  COPY_PARAMS(mkldnn_cache_capacity)
+  COPY_PARAMS(precision)
+  COPY_PARAMS(cpu_threads)
+  return to;
+}
diff --git a/deploy/cpp_infer/src/api/models/text_recognition.h b/deploy/cpp_infer/src/api/models/text_recognition.h
new file mode 100644
index 0000000..360e036
--- /dev/null
+++ b/deploy/cpp_infer/src/api/models/text_recognition.h
@@ -0,0 +1,54 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include "src/modules/text_recognition/predictor.h"
+
+struct TextRecognitionParams {
+  absl::optional<std::string> model_name = absl::nullopt;
+  absl::optional<std::string> model_dir = absl::nullopt;
+  absl::optional<std::string> lang = absl::nullopt;
+  absl::optional<std::string> ocr_version = absl::nullopt;
+  absl::optional<std::string> vis_font_dir = absl::nullopt;
+  absl::optional<std::string> device = absl::nullopt;
+  std::string precision = "fp32";
+  bool enable_mkldnn = true;
+  int mkldnn_cache_capacity = 10;
+  int cpu_threads = 8;
+  int batch_size = 1;
+  absl::optional<std::vector<int>> input_shape = absl::nullopt;
+};
+
+class TextRecognition {
+public:
+  TextRecognition(
+      const TextRecognitionParams &params = TextRecognitionParams());
+
+  std::vector<std::unique_ptr<BaseCVResult>> Predict(const std::string &input) {
+    std::vector<std::string> inputs = {input};
+    return Predict(inputs);
+  };
+  std::vector<std::unique_ptr<BaseCVResult>>
+  Predict(const std::vector<std::string> &input);
+
+  void CreateModel();
+  absl::Status CheckParams();
+  static TextRecPredictorParams
+  ToTextRecognitionModelParams(const TextRecognitionParams &from);
+
+private:
+  TextRecognitionParams params_;
+  std::unique_ptr<BasePredictor> model_infer_;
+};
diff --git a/deploy/cpp_infer/src/api/models/textline_orientation_classification.cc b/deploy/cpp_infer/src/api/models/textline_orientation_classification.cc
new file mode 100644
index 0000000..8156d77
--- /dev/null
+++ b/deploy/cpp_infer/src/api/models/textline_orientation_classification.cc
@@ -0,0 +1,63 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "textline_orientation_classification.h"
+
+#include "src/utils/args.h"
+#include "src/utils/yaml_config.h"
+
+#define COPY_PARAMS(field) to.field = from.field;
+
+TextLineOrientationClassification::TextLineOrientationClassification(
+    const TextLineOrientationClassificationParams &params)
+    : params_(params) {
+  auto status = CheckParams();
+  if (!status.ok()) {
+    INFOE("Init TextLineOrientationClassification fail : %s",
+          status.ToString().c_str());
+    exit(-1);
+  }
+  CreateModel();
+};
+std::vector<std::unique_ptr<BaseCVResult>>
+TextLineOrientationClassification::Predict(
+    const std::vector<std::string> &input) {
+  return model_infer_->Predict(input);
+}
+void TextLineOrientationClassification::CreateModel() {
+  model_infer_ = std::unique_ptr<BasePredictor>(new ClasPredictor(
+      ToTextLineOrientationClassificationModelParams(params_)));
+}
+
+absl::Status TextLineOrientationClassification::CheckParams() {
+  if (!params_.model_dir.has_value()) {
+    return absl::NotFoundError(
+        "Require textLine orientation classification model dir.");
+  }
+  return absl::OkStatus();
+}
+
+ClasPredictorParams TextLineOrientationClassification::
+    ToTextLineOrientationClassificationModelParams(
+        const TextLineOrientationClassificationParams &from) {
+  ClasPredictorParams to;
+  COPY_PARAMS(model_name)
+  COPY_PARAMS(model_dir)
+  COPY_PARAMS(device)
+  COPY_PARAMS(enable_mkldnn)
+  COPY_PARAMS(mkldnn_cache_capacity)
+  COPY_PARAMS(precision)
+  COPY_PARAMS(cpu_threads)
+  return to;
+}
diff --git a/deploy/cpp_infer/src/api/models/textline_orientation_classification.h b/deploy/cpp_infer/src/api/models/textline_orientation_classification.h
new file mode 100644
index 0000000..a40e7d4
--- /dev/null
+++ b/deploy/cpp_infer/src/api/models/textline_orientation_classification.h
@@ -0,0 +1,51 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include "src/modules/image_classification/predictor.h"
+
+struct TextLineOrientationClassificationParams {
+  absl::optional<std::string> model_name = absl::nullopt;
+  absl::optional<std::string> model_dir = absl::nullopt;
+  absl::optional<std::string> device = absl::nullopt;
+  std::string precision = "fp32";
+  bool enable_mkldnn = true;
+  int mkldnn_cache_capacity = 10;
+  int cpu_threads = 8;
+  int batch_size = 1;
+};
+
+class TextLineOrientationClassification {
+public:
+  TextLineOrientationClassification(
+      const TextLineOrientationClassificationParams &params =
+          TextLineOrientationClassificationParams());
+
+  std::vector<std::unique_ptr<BaseCVResult>> Predict(const std::string &input) {
+    std::vector<std::string> inputs = {input};
+    return Predict(inputs);
+  };
+  std::vector<std::unique_ptr<BaseCVResult>>
+  Predict(const std::vector<std::string> &input);
+
+  void CreateModel();
+  absl::Status CheckParams();
+  static ClasPredictorParams ToTextLineOrientationClassificationModelParams(
+      const TextLineOrientationClassificationParams &from);
+
+private:
+  TextLineOrientationClassificationParams params_;
+  std::unique_ptr<BasePredictor> model_infer_;
+};
diff --git a/deploy/cpp_infer/src/api/pipelines/base.h b/deploy/cpp_infer/src/api/pipelines/base.h
new file mode 100644
index 0000000..61e4a5c
--- /dev/null
+++ b/deploy/cpp_infer/src/api/pipelines/base.h
@@ -0,0 +1,25 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+#include <memory>
+
+#include "src/base/base_pipeline.h"
+
+class PaddleXPipelineWrapper {
+public:
+  virtual ~PaddleXPipelineWrapper() = default;
+  PaddleXPipelineWrapper() = delete;
+  virtual std::unique_ptr<BasePipeline> CreatePipeline() = 0;
+};
diff --git a/deploy/cpp_infer/src/api/pipelines/doc_preprocessor.cc b/deploy/cpp_infer/src/api/pipelines/doc_preprocessor.cc
new file mode 100644
index 0000000..123d7e1
--- /dev/null
+++ b/deploy/cpp_infer/src/api/pipelines/doc_preprocessor.cc
@@ -0,0 +1,71 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "doc_preprocessor.h"
+
+#include "src/utils/args.h"
+#include "src/utils/yaml_config.h"
+
+#define COPY_PARAMS(field) to.field = from.field;
+
+DocPreprocessor::DocPreprocessor(const DocPreprocessorParams &params)
+    : params_(params) {
+  auto status = CheckParams();
+  if (!status.ok()) {
+    INFOE("Init DocPreprocessor fail : %s", status.ToString().c_str());
+    exit(-1);
+  }
+  CreatePipeline();
+};
+std::vector<std::unique_ptr<BaseCVResult>>
+DocPreprocessor::Predict(const std::vector<std::string> &input) {
+  return pipeline_infer_->Predict(input);
+}
+void DocPreprocessor::CreatePipeline() {
+  pipeline_infer_ = std::unique_ptr<BasePipeline>(
+      new DocPreprocessorPipeline(ToDocPreprocessorPipelineParams(params_)));
+}
+
+absl::Status DocPreprocessor::CheckParams() {
+  if (!params_.doc_orientation_classify_model_dir.has_value() &&
+      !(params_.use_doc_orientation_classify.has_value() &&
+        !params_.use_doc_orientation_classify.value())) {
+    return absl::NotFoundError("Require doc orientation classify model dir.");
+  }
+  if (!params_.doc_unwarping_model_dir.has_value() &&
+      !(params_.use_doc_unwarping.has_value() &&
+        !params_.use_doc_unwarping.value())) {
+    return absl::NotFoundError("Require doc unwarping model dir.");
+  }
+  return absl::OkStatus();
+}
+
+DocPreprocessorPipelineParams DocPreprocessor::ToDocPreprocessorPipelineParams(
+    const DocPreprocessorParams &from) {
+  DocPreprocessorPipelineParams to;
+  COPY_PARAMS(doc_orientation_classify_model_name)
+  COPY_PARAMS(doc_orientation_classify_model_dir)
+  COPY_PARAMS(doc_unwarping_model_name)
+  COPY_PARAMS(doc_unwarping_model_dir)
+  COPY_PARAMS(use_doc_orientation_classify)
+  COPY_PARAMS(use_doc_unwarping)
+  COPY_PARAMS(device)
+  COPY_PARAMS(enable_mkldnn)
+  COPY_PARAMS(mkldnn_cache_capacity)
+  COPY_PARAMS(precision)
+  COPY_PARAMS(cpu_threads)
+  COPY_PARAMS(thread_num)
+  COPY_PARAMS(paddlex_config)
+  return to;
+}
diff --git a/deploy/cpp_infer/src/api/pipelines/doc_preprocessor.h b/deploy/cpp_infer/src/api/pipelines/doc_preprocessor.h
new file mode 100644
index 0000000..c5add06
--- /dev/null
+++ b/deploy/cpp_infer/src/api/pipelines/doc_preprocessor.h
@@ -0,0 +1,57 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include "src/pipelines/doc_preprocessor/pipeline.h"
+
+struct DocPreprocessorParams {
+  absl::optional<std::string> doc_orientation_classify_model_name =
+      absl::nullopt;
+  absl::optional<std::string> doc_orientation_classify_model_dir =
+      absl::nullopt;
+  absl::optional<std::string> doc_unwarping_model_name = absl::nullopt;
+  absl::optional<std::string> doc_unwarping_model_dir = absl::nullopt;
+  absl::optional<bool> use_doc_orientation_classify = absl::nullopt;
+  absl::optional<bool> use_doc_unwarping = absl::nullopt;
+  absl::optional<std::string> device = absl::nullopt;
+  bool enable_mkldnn = true;
+  int mkldnn_cache_capacity = 10;
+  std::string precision = "fp32";
+  int cpu_threads = 8;
+  int thread_num = 1;
+  absl::optional<Utility::PaddleXConfigVariant> paddlex_config = absl::nullopt;
+};
+
+class DocPreprocessor {
+public:
+  DocPreprocessor(
+      const DocPreprocessorParams &params = DocPreprocessorParams());
+
+  std::vector<std::unique_ptr<BaseCVResult>> Predict(const std::string &input) {
+    std::vector<std::string> inputs = {input};
+    return Predict(inputs);
+  };
+  std::vector<std::unique_ptr<BaseCVResult>>
+  Predict(const std::vector<std::string> &input);
+
+  void CreatePipeline();
+  absl::Status CheckParams();
+  static DocPreprocessorPipelineParams
+  ToDocPreprocessorPipelineParams(const DocPreprocessorParams &from);
+
+private:
+  DocPreprocessorParams params_;
+  std::unique_ptr<BasePipeline> pipeline_infer_;
+};
diff --git a/deploy/cpp_infer/src/api/pipelines/ocr.cc b/deploy/cpp_infer/src/api/pipelines/ocr.cc
new file mode 100644
index 0000000..b1f1737
--- /dev/null
+++ b/deploy/cpp_infer/src/api/pipelines/ocr.cc
@@ -0,0 +1,100 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "ocr.h"
+
+#include "src/utils/args.h"
+#include "src/utils/yaml_config.h"
+
+#define COPY_PARAMS(field) to.field = from.field;
+
+PaddleOCR::PaddleOCR(const PaddleOCRParams &params) : params_(params) {
+  auto status = CheckParams();
+  if (!status.ok()) {
+    INFOE("Init paddleOCR fail : %s", status.ToString().c_str());
+    exit(-1);
+  }
+  CreatePipeline();
+};
+std::vector<std::unique_ptr<BaseCVResult>>
+PaddleOCR::Predict(const std::vector<std::string> &input) {
+  return pipeline_infer_->Predict(input);
+}
+void PaddleOCR::CreatePipeline() {
+  pipeline_infer_ = std::unique_ptr<BasePipeline>(
+      new OCRPipeline(ToOCRPipelineParams(params_)));
+}
+
+absl::Status PaddleOCR::CheckParams() {
+  if (!params_.doc_orientation_classify_model_dir.has_value() &&
+      !(params_.use_doc_orientation_classify.has_value() &&
+        !params_.use_doc_orientation_classify.value())) {
+    return absl::NotFoundError("Require doc orientation classify model dir.");
+  }
+  if (!params_.doc_unwarping_model_dir.has_value() &&
+      !(params_.use_doc_unwarping.has_value() &&
+        !params_.use_doc_unwarping.value())) {
+    return absl::NotFoundError("Require doc unwarping model dir.");
+  }
+  if (!params_.textline_orientation_model_dir.has_value() &&
+      !(params_.use_textline_orientation.has_value() &&
+        !params_.use_textline_orientation.value())) {
+    return absl::NotFoundError("Require textline orientation model_dir.");
+  }
+  if (!params_.text_detection_model_dir.has_value()) {
+    return absl::NotFoundError("Require text detection model dir.");
+  }
+  if (!params_.text_recognition_model_dir.has_value()) {
+    return absl::NotFoundError("Require text recognition model_dir.");
+  }
+  return absl::OkStatus();
+}
+
+OCRPipelineParams PaddleOCR::ToOCRPipelineParams(const PaddleOCRParams &from) {
+  OCRPipelineParams to;
+  COPY_PARAMS(doc_orientation_classify_model_name)
+  COPY_PARAMS(doc_orientation_classify_model_dir)
+  COPY_PARAMS(doc_unwarping_model_name)
+  COPY_PARAMS(doc_unwarping_model_dir)
+  COPY_PARAMS(text_detection_model_name)
+  COPY_PARAMS(text_detection_model_dir)
+  COPY_PARAMS(textline_orientation_model_name)
+  COPY_PARAMS(textline_orientation_model_dir)
+  COPY_PARAMS(textline_orientation_batch_size)
+  COPY_PARAMS(text_recognition_model_name)
+  COPY_PARAMS(text_recognition_model_dir)
+  COPY_PARAMS(text_recognition_batch_size)
+  COPY_PARAMS(use_doc_orientation_classify)
+  COPY_PARAMS(use_doc_unwarping)
+  COPY_PARAMS(use_textline_orientation)
+  COPY_PARAMS(text_det_limit_side_len)
+  COPY_PARAMS(text_det_limit_type)
+  COPY_PARAMS(text_det_thresh)
+  COPY_PARAMS(text_det_box_thresh)
+  COPY_PARAMS(text_det_unclip_ratio)
+  COPY_PARAMS(text_det_input_shape)
+  COPY_PARAMS(text_rec_score_thresh)
+  COPY_PARAMS(text_rec_input_shape)
+  COPY_PARAMS(lang)
+  COPY_PARAMS(ocr_version)
+  COPY_PARAMS(vis_font_dir)
+  COPY_PARAMS(device)
+  COPY_PARAMS(enable_mkldnn)
+  COPY_PARAMS(mkldnn_cache_capacity)
+  COPY_PARAMS(precision)
+  COPY_PARAMS(cpu_threads)
+  COPY_PARAMS(thread_num)
+  COPY_PARAMS(paddlex_config)
+  return to;
+}
diff --git a/deploy/cpp_infer/src/api/pipelines/ocr.h b/deploy/cpp_infer/src/api/pipelines/ocr.h
new file mode 100644
index 0000000..c9f87e6
--- /dev/null
+++ b/deploy/cpp_infer/src/api/pipelines/ocr.h
@@ -0,0 +1,75 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include "src/pipelines/ocr/pipeline.h"
+
+struct PaddleOCRParams {
+  absl::optional<std::string> doc_orientation_classify_model_name =
+      absl::nullopt;
+  absl::optional<std::string> doc_orientation_classify_model_dir =
+      absl::nullopt;
+  absl::optional<std::string> doc_unwarping_model_name = absl::nullopt;
+  absl::optional<std::string> doc_unwarping_model_dir = absl::nullopt;
+  absl::optional<std::string> text_detection_model_name = absl::nullopt;
+  absl::optional<std::string> text_detection_model_dir = absl::nullopt;
+  absl::optional<std::string> textline_orientation_model_name = absl::nullopt;
+  absl::optional<std::string> textline_orientation_model_dir = absl::nullopt;
+  absl::optional<int> textline_orientation_batch_size = absl::nullopt;
+  absl::optional<std::string> text_recognition_model_name = absl::nullopt;
+  absl::optional<std::string> text_recognition_model_dir = absl::nullopt;
+  absl::optional<int> text_recognition_batch_size = absl::nullopt;
+  absl::optional<bool> use_doc_orientation_classify = absl::nullopt;
+  absl::optional<bool> use_doc_unwarping = absl::nullopt;
+  absl::optional<bool> use_textline_orientation = absl::nullopt;
+  absl::optional<int> text_det_limit_side_len = absl::nullopt;
+  absl::optional<std::string> text_det_limit_type = absl::nullopt;
+  absl::optional<float> text_det_thresh = absl::nullopt;
+  absl::optional<float> text_det_box_thresh = absl::nullopt;
+  absl::optional<float> text_det_unclip_ratio = absl::nullopt;
+  absl::optional<std::vector<int>> text_det_input_shape = absl::nullopt;
+  absl::optional<float> text_rec_score_thresh = absl::nullopt;
+  absl::optional<std::vector<int>> text_rec_input_shape = absl::nullopt;
+  absl::optional<std::string> lang = absl::nullopt;
+  absl::optional<std::string> ocr_version = absl::nullopt;
+  absl::optional<std::string> vis_font_dir = absl::nullopt;
+  absl::optional<std::string> device = absl::nullopt;
+  bool enable_mkldnn = true;
+  int mkldnn_cache_capacity = 10;
+  std::string precision = "fp32";
+  int cpu_threads = 8;
+  int thread_num = 1;
+  absl::optional<Utility::PaddleXConfigVariant> paddlex_config = absl::nullopt;
+};
+
+class PaddleOCR {
+public:
+  PaddleOCR(const PaddleOCRParams &params = PaddleOCRParams());
+
+  std::vector<std::unique_ptr<BaseCVResult>> Predict(const std::string &input) {
+    std::vector<std::string> inputs = {input};
+    return Predict(inputs);
+  };
+  std::vector<std::unique_ptr<BaseCVResult>>
+  Predict(const std::vector<std::string> &input);
+
+  void CreatePipeline();
+  absl::Status CheckParams();
+  static OCRPipelineParams ToOCRPipelineParams(const PaddleOCRParams &from);
+
+private:
+  PaddleOCRParams params_;
+  std::unique_ptr<BasePipeline> pipeline_infer_;
+};
diff --git a/deploy/cpp_infer/src/base/base_batch_sampler.cc b/deploy/cpp_infer/src/base/base_batch_sampler.cc
new file mode 100644
index 0000000..0050751
--- /dev/null
+++ b/deploy/cpp_infer/src/base/base_batch_sampler.cc
@@ -0,0 +1,106 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "base_batch_sampler.h"
+
+#include "src/utils/ilogger.h"
+#include "src/utils/utility.h"
+int BaseBatchSampler::BatchSize() const { return batch_size_; }
+
+absl::Status BaseBatchSampler::SetBatchSize(int batch_size) {
+  if (batch_size <= 0) {
+    return absl::InvalidArgumentError("Batch size must be greater than 0");
+  }
+  batch_size_ = batch_size;
+  return absl::OkStatus();
+}
+
+absl::StatusOr<std::vector<std::vector<std::string>>>
+BaseBatchSampler::SampleFromVectorToStringVector(
+    const std::vector<std::string> &inputs) {
+  std::vector<std::vector<std::string>> result;
+  std::vector<std::string> current_batch;
+
+  for (size_t i = 0; i < inputs.size(); ++i) {
+    const std::string &input = inputs[i];
+
+    if (Utility::IsDirectory(input)) {
+      absl::StatusOr<std::vector<std::string>> files_result =
+          GetFilesList(input);
+      if (!files_result.ok()) {
+        return files_result.status();
+      }
+
+      absl::StatusOr<std::vector<std::vector<std::string>>> sub_result =
+          SampleFromVectorToStringVector(files_result.value());
+      if (!sub_result.ok()) {
+        return sub_result.status();
+      }
+
+      const std::vector<std::vector<std::string>> &sub_batches =
+          sub_result.value();
+      for (size_t j = 0; j < sub_batches.size(); ++j) {
+        result.push_back(sub_batches[j]);
+      }
+    } else if (Utility::IsImageFile(input)) {
+      if (!Utility::FileExists(input).ok()) {
+        return absl::NotFoundError("File not found: " + input);
+      }
+      current_batch.push_back(input);
+
+      if (static_cast<int>(current_batch.size()) == batch_size_) {
+        result.push_back(current_batch);
+        current_batch.clear();
+      }
+    } else {
+      return absl::InvalidArgumentError("Unsupported file type: " + input);
+    }
+  }
+
+  if (!current_batch.empty()) {
+    result.push_back(current_batch); // last batch
+  }
+
+  return result;
+}
+
+absl::StatusOr<std::vector<std::vector<std::string>>>
+BaseBatchSampler::SampleFromStringToStringVector(const std::string &input) {
+  std::vector<std::string> inputs = {input};
+  return SampleFromVectorToStringVector(inputs);
+}
+
+absl::StatusOr<std::vector<std::string>>
+BaseBatchSampler::GetFilesList(const std::string &path) {
+  if (!Utility::FileExists(path).ok()) {
+    return absl::NotFoundError("Path not found: " + path);
+  }
+
+  std::vector<std::string> file_list;
+
+  if (!Utility::IsDirectory(path)) {
+    if (Utility::IsImageFile(path)) {
+      file_list.push_back(path);
+    }
+  } else {
+    Utility::GetFilesRecursive(path, file_list);
+  }
+
+  if (file_list.empty()) {
+    return absl::NotFoundError("No image files found in path: " + path);
+  }
+
+  std::sort(file_list.begin(), file_list.end());
+  return file_list;
+}
diff --git a/deploy/cpp_infer/src/base/base_batch_sampler.h b/deploy/cpp_infer/src/base/base_batch_sampler.h
new file mode 100644
index 0000000..12e20f6
--- /dev/null
+++ b/deploy/cpp_infer/src/base/base_batch_sampler.h
@@ -0,0 +1,89 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <opencv2/opencv.hpp>
+#include <string>
+#include <type_traits>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+
+class BaseBatchSampler {
+public:
+  explicit BaseBatchSampler(int batch_size) : batch_size_(batch_size) {}
+  virtual ~BaseBatchSampler() = default;
+
+  int BatchSize() const;
+  absl::Status SetBatchSize(int batch_size);
+
+  template <typename T>
+  absl::StatusOr<std::vector<std::vector<cv::Mat>>> Apply(const T &input);
+
+  template <typename T>
+  absl::StatusOr<std::vector<std::vector<cv::Mat>>> Sample(const T &input) {
+    return absl::InvalidArgumentError(
+        "Sample failed! Unsupported type for Sample");
+  }
+
+  virtual absl::StatusOr<std::vector<std::vector<cv::Mat>>>
+  SampleFromString(const std::string &input) = 0;
+
+  virtual absl::StatusOr<std::vector<std::vector<cv::Mat>>>
+  SampleFromVector(const std::vector<std::string> &inputs) = 0;
+  std::vector<std::string> InputPath() { return input_path_; };
+
+  virtual absl::StatusOr<std::vector<std::vector<cv::Mat>>>
+  SampleFromMatVector(const std::vector<cv::Mat> &inputs) = 0;
+
+  absl::StatusOr<std::vector<std::vector<std::string>>>
+  SampleFromStringToStringVector(const std::string &input);
+  absl::StatusOr<std::vector<std::vector<std::string>>>
+  SampleFromVectorToStringVector(const std::vector<std::string> &input);
+
+  absl::StatusOr<std::vector<std::string>>
+  GetFilesList(const std::string &path);
+
+protected:
+  int batch_size_ = 1;
+  std::vector<std::string> input_path_;
+};
+
+template <typename T>
+absl::StatusOr<std::vector<std::vector<cv::Mat>>>
+BaseBatchSampler::Apply(const T &input) {
+  return Sample(input);
+}
+
+template <>
+inline absl::StatusOr<std::vector<std::vector<cv::Mat>>>
+BaseBatchSampler::Sample<std::string>(const std::string &input) {
+  return SampleFromString(input);
+}
+
+template <>
+inline absl::StatusOr<std::vector<std::vector<cv::Mat>>>
+BaseBatchSampler::Sample<std::vector<std::string>>(
+    const std::vector<std::string> &input) {
+  return SampleFromVector(input);
+}
+
+template <>
+inline absl::StatusOr<std::vector<std::vector<cv::Mat>>>
+BaseBatchSampler::Sample<std::vector<cv::Mat>>(
+    const std::vector<cv::Mat> &input) {
+  return SampleFromMatVector(input);
+}
diff --git a/deploy/cpp_infer/src/base/base_cv_result.h b/deploy/cpp_infer/src/base/base_cv_result.h
new file mode 100644
index 0000000..0d1cabf
--- /dev/null
+++ b/deploy/cpp_infer/src/base/base_cv_result.h
@@ -0,0 +1,46 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <chrono>
+#include <iomanip>
+#include <opencv2/opencv.hpp>
+#include <string>
+#include <unordered_map>
+
+#include "absl/status/statusor.h"
+
+class ImageWriter {};
+
+class BaseCVResult {
+public:
+  BaseCVResult(const std::string &backend);
+  BaseCVResult() = default;
+  virtual ~BaseCVResult() = default;
+  std::string Str() const;
+  std::unordered_map<std::string, cv::Mat> Img() const;
+  // absl::Status Print() const;
+  absl::Status SaveToImg() const;
+
+  virtual void SaveToImg(const std::string &save_path) = 0;
+  virtual void Print() const = 0;
+  virtual void SaveToJson(const std::string &save_path) const = 0;
+
+protected:
+  std::unordered_map<std::string, std::string> res_;
+  ImageWriter img_writer_;
+  std::string ToStr() const;
+  // virtual std::unordered_map<std::string, cv::Mat> ToImg() const = 0;
+};
diff --git a/deploy/cpp_infer/src/base/base_pipeline.cc b/deploy/cpp_infer/src/base/base_pipeline.cc
new file mode 100644
index 0000000..7c9bf3a
--- /dev/null
+++ b/deploy/cpp_infer/src/base/base_pipeline.cc
@@ -0,0 +1,15 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "base_pipeline.h"
diff --git a/deploy/cpp_infer/src/base/base_pipeline.h b/deploy/cpp_infer/src/base/base_pipeline.h
new file mode 100644
index 0000000..6291be2
--- /dev/null
+++ b/deploy/cpp_infer/src/base/base_pipeline.h
@@ -0,0 +1,60 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <iostream>
+#include <memory>
+#include <opencv2/opencv.hpp>
+#include <string>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+#include "base_cv_result.h"
+#include "base_predictor.h"
+
+class BasePipeline {
+public:
+  BasePipeline() = default;
+  virtual ~BasePipeline() = default;
+
+  std::vector<std::unique_ptr<BaseCVResult>> Predict(const std::string &input) {
+    std::vector<std::string> inputs = {input};
+    return Predict(inputs);
+  }
+
+  virtual std::vector<std::unique_ptr<BaseCVResult>>
+  Predict(const std::vector<std::string> &input) = 0;
+
+  template <typename T, typename... Args>
+  std::unique_ptr<BasePredictor> CreateModule(Args &&...args);
+
+  template <typename T, typename... Args>
+  std::unique_ptr<BasePipeline> CreatePipeline(Args &&...args);
+};
+
+template <typename T, typename... Args>
+std::unique_ptr<BasePredictor> BasePipeline::CreateModule(Args &&...args) {
+  std::unique_ptr<BasePredictor> base_predictor =
+      std::unique_ptr<T>(new T(std::forward<Args>(args)...));
+  return base_predictor;
+}
+
+template <typename T, typename... Args>
+std::unique_ptr<BasePipeline> BasePipeline::CreatePipeline(Args &&...args) {
+  std::unique_ptr<BasePipeline> base_pipeline =
+      std::unique_ptr<T>(new T(std::forward<Args>(args)...));
+  return base_pipeline;
+}
diff --git a/deploy/cpp_infer/src/base/base_predictor.cc b/deploy/cpp_infer/src/base/base_predictor.cc
new file mode 100644
index 0000000..6d95dea
--- /dev/null
+++ b/deploy/cpp_infer/src/base/base_predictor.cc
@@ -0,0 +1,176 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "base_predictor.h"
+
+#include <yaml-cpp/yaml.h>
+
+#include <iostream>
+
+#include "base_batch_sampler.h"
+#include "src/common/image_batch_sampler.h"
+#include "src/utils/ilogger.h"
+#include "src/utils/pp_option.h"
+#include "src/utils/utility.h"
+
+BasePredictor::BasePredictor(const absl::optional<std::string> &model_dir,
+                             const absl::optional<std::string> &model_name,
+                             const absl::optional<std::string> &device,
+                             const std::string &precision,
+                             const bool enable_mkldnn,
+                             int mkldnn_cache_capacityint, int cpu_threads,
+                             int batch_size, const std::string sampler_type)
+    : model_dir_(model_dir), batch_size_(batch_size),
+      sampler_type_(sampler_type) {
+  if (model_dir_.has_value()) {
+    config_ = YamlConfig(model_dir_.value());
+  } else {
+    INFOE("Model dir is empty.");
+    exit(-1);
+  }
+  auto status_build = BuildBatchSampler();
+  if (!status_build.ok()) {
+    INFOE("Build sampler fail: %s", status_build.ToString().c_str());
+    exit(-1);
+  }
+  auto model_name_config = config_.GetString(std::string("Global.model_name"));
+  if (!model_name_config.ok()) {
+    INFOE(model_name_config.status().ToString().c_str());
+    exit(-1);
+  }
+  model_name_ = model_name_config.value();
+  if (model_name.has_value()) {
+    if (model_name_ != model_name.value()) {
+      INFOE(
+          "Model name mismatch, please input the correct model dir. model dir "
+          "is %s, but model name is %s",
+          model_dir_.value().c_str(), model_name.value().c_str());
+      exit(-1);
+    }
+  }
+  model_name_ = model_name.value_or(model_name_);
+  pp_option_ptr_.reset(new PaddlePredictorOption());
+  auto device_result = device.value_or(DEVICE);
+
+  size_t pos = device_result.find(':');
+  std::string device_type = "";
+  int device_id = 0;
+  if (pos != std::string::npos) {
+    device_type = device_result.substr(0, pos);
+    device_id = std::stoi(device_result.substr(pos + 1));
+  } else {
+    device_type = device_result;
+    device_id = 0;
+  }
+  auto status_device_type = pp_option_ptr_->SetDeviceType(device_type);
+  if (!status_device_type.ok()) {
+    INFOE("Failed to set device : %s", status_device_type.ToString().c_str());
+    exit(-1);
+    ;
+  }
+  auto status_device_id = pp_option_ptr_->SetDeviceId(device_id);
+  if (!status_device_id.ok()) {
+    INFOE("Failed to set device id: %s", status_device_id.ToString().c_str());
+    exit(-1);
+    ;
+  }
+
+  if (enable_mkldnn && device_type == "cpu") {
+    if (precision == "fp16") {
+      INFOW("When MKLDNN is enabled, FP16 precision is not supported.The "
+            "computation will proceed with FP32 instead.");
+    }
+    if (Utility::IsMkldnnAvailable()) {
+      auto status_mkldnn = pp_option_ptr_->SetRunMode("mkldnn");
+      if (!status_mkldnn.ok()) {
+        INFOE("Failed to set run mode: %s", status_mkldnn.ToString().c_str());
+        exit(-1);
+        ;
+      }
+    } else {
+      INFOW("Mkldnn is not available, using paddle instead!");
+      auto status_paddle = pp_option_ptr_->SetRunMode("paddle");
+      if (!status_paddle.ok()) {
+        INFOE("Failed to set run mode: %s", status_paddle.ToString().c_str());
+        exit(-1);
+      }
+    }
+  } else if (precision == "fp16") {
+    if (precision == "fp16") {
+      auto status_paddle_fp16 = pp_option_ptr_->SetRunMode("paddle_fp16");
+      if (!status_paddle_fp16.ok()) {
+        INFOE("Failed to set run mode: %s",
+              status_paddle_fp16.ToString().c_str());
+        exit(-1);
+        ;
+      }
+    }
+  } else {
+    auto status_paddle = pp_option_ptr_->SetRunMode("paddle");
+    if (!status_paddle.ok()) {
+      INFOE("Failed to set run mode: %s", status_paddle.ToString().c_str());
+      exit(-1);
+    }
+  }
+  auto status_mkldnn_cache_capacityint =
+      pp_option_ptr_->SetMkldnnCacheCapacity(mkldnn_cache_capacityint);
+  if (!status_mkldnn_cache_capacityint.ok()) {
+    INFOE("Set status_mkldnn_cache_capacityint fail : %s",
+          status_mkldnn_cache_capacityint.ToString().c_str());
+    exit(-1);
+  }
+  auto status_cpu_threads = pp_option_ptr_->SetCpuThreads(cpu_threads);
+  if (!status_cpu_threads.ok()) {
+    INFOE("Set cpu threads fail : %s", status_cpu_threads.ToString().c_str());
+    exit(-1);
+  }
+  if (print_flag) {
+    INFO(pp_option_ptr_->DebugString().c_str());
+    print_flag = false;
+  }
+  INFO("Create model: %s.", model_name_.c_str());
+}
+
+std::vector<std::unique_ptr<BaseCVResult>>
+BasePredictor::Predict(const std::string &input) {
+  std::vector<std::string> inputs = {input};
+  return Predict(inputs);
+}
+
+const PaddlePredictorOption &BasePredictor::PPOption() {
+  return *pp_option_ptr_;
+}
+
+void BasePredictor::SetBatchSize(int batch_size) { batch_size_ = batch_size; }
+
+std::unique_ptr<PaddleInfer> BasePredictor::CreateStaticInfer() {
+  return std::unique_ptr<PaddleInfer>(new PaddleInfer(
+      model_name_, model_dir_.value(), MODEL_FILE_PREFIX, PPOption()));
+}
+
+absl::Status BasePredictor::BuildBatchSampler() {
+  if (SAMPLER_TYPE.count(sampler_type_) == 0) {
+    return absl::InvalidArgumentError("Unsupported sampler type !");
+  } else if (sampler_type_ == "image") {
+    batch_sampler_ptr_ =
+        std::unique_ptr<BaseBatchSampler>(new ImageBatchSampler(batch_size_));
+  }
+  return absl::OkStatus();
+}
+
+const std::unordered_set<std::string> BasePredictor::SAMPLER_TYPE = {
+    "image",
+};
+
+bool BasePredictor::print_flag = true;
diff --git a/deploy/cpp_infer/src/base/base_predictor.h b/deploy/cpp_infer/src/base/base_predictor.h
new file mode 100644
index 0000000..ea2e3db
--- /dev/null
+++ b/deploy/cpp_infer/src/base/base_predictor.h
@@ -0,0 +1,107 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <memory>
+#include <string>
+#include <unordered_map>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+#include "absl/types/optional.h"
+#include "base_batch_sampler.h"
+#include "base_cv_result.h"
+#include "src/common/static_infer.h"
+#include "src/utils/func_register.h"
+#include "src/utils/pp_option.h"
+#include "src/utils/yaml_config.h"
+
+class BasePredictor {
+public:
+  BasePredictor(const absl::optional<std::string> &model_dir = absl::nullopt,
+                const absl::optional<std::string> &model_name = absl::nullopt,
+                const absl::optional<std::string> &device = absl::nullopt,
+                const std::string &precision = "fp32",
+                const bool enable_mkldnn = true,
+                int mkldnn_cache_capacityint = 10, int cpu_threads = 8,
+                int batch_size = 1, const std::string sample_type = "");
+  virtual ~BasePredictor() = default;
+  std::vector<std::unique_ptr<BaseCVResult>> Predict(const std::string &input);
+
+  template <typename T>
+  std::vector<std::unique_ptr<BaseCVResult>> Predict(const T &input);
+
+  std::unique_ptr<PaddleInfer> CreateStaticInfer();
+
+  const PaddlePredictorOption &PPOption();
+  absl::StatusOr<std::string> ModelName() { return model_name_; };
+  std::string ConfigPath() { return config_.ConfigYamlPath(); };
+
+  void SetBatchSize(int batch_size);
+
+  virtual std::vector<std::unique_ptr<BaseCVResult>>
+  Process(std::vector<cv::Mat> &batch_data) = 0;
+  virtual void ResetResult() = 0;
+  absl::Status BuildBatchSampler();
+
+  void SetInputPath(const std::vector<std::string> &input_path) {
+    input_path_ = input_path;
+  };
+
+  template <typename T, typename... Args>
+  void Register(const std::string &key, Args &&...args);
+
+  static constexpr const char *MODEL_FILE_PREFIX = "inference";
+  static const std::unordered_set<std::string> SAMPLER_TYPE;
+  static bool print_flag;
+
+protected:
+  absl::optional<std::string> model_dir_;
+  YamlConfig config_;
+  int batch_size_;
+  std::unique_ptr<BaseBatchSampler> batch_sampler_ptr_;
+  std::unique_ptr<PaddlePredictorOption> pp_option_ptr_;
+  std::vector<std::string> input_path_;
+  std::string model_name_;
+  std::string sampler_type_;
+  std::unordered_map<std::string, std::unique_ptr<BaseProcessor>> pre_op_;
+};
+
+template <typename T, typename... Args>
+void BasePredictor::Register(const std::string &key, Args &&...args) {
+  auto instance = std::unique_ptr<T>(new T(std::forward<Args>(args)...));
+  pre_op_[key] = std::move(instance);
+};
+
+template <typename T>
+std::vector<std::unique_ptr<BaseCVResult>>
+BasePredictor::Predict(const T &input) {
+  std::vector<std::unique_ptr<BaseCVResult>> result;
+  ResetResult();
+  auto batches = batch_sampler_ptr_->Apply(input);
+  if (!batches.ok()) {
+    INFOE("Get sample fail : %s", batches.status().ToString().c_str());
+    exit(-1);
+  }
+  input_path_ = batch_sampler_ptr_->InputPath();
+  for (auto &batch_data : batches.value()) {
+    auto predictions = Process(batch_data);
+    for (auto &prediction : predictions) {
+      result.emplace_back(std::move(prediction));
+    }
+  }
+  return result;
+}
diff --git a/deploy/cpp_infer/src/common/image_batch_sampler.cc b/deploy/cpp_infer/src/common/image_batch_sampler.cc
new file mode 100644
index 0000000..6f83dd6
--- /dev/null
+++ b/deploy/cpp_infer/src/common/image_batch_sampler.cc
@@ -0,0 +1,118 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "image_batch_sampler.h"
+
+#include <dirent.h>
+#include <sys/stat.h>
+
+#include <algorithm>
+#include <cctype>
+#include <iostream>
+
+#include "src/utils/ilogger.h"
+#include "src/utils/utility.h"
+
+const std::set<std::string> ImageBatchSampler::kImgSuffixes = {"jpg", "png",
+                                                               "jpeg", "bmp"};
+
+ImageBatchSampler::ImageBatchSampler(int batch_size)
+    : BaseBatchSampler(batch_size) {}
+
+absl::StatusOr<std::vector<std::vector<cv::Mat>>>
+ImageBatchSampler::SampleFromString(const std::string &input) {
+  std::vector<std::string> inputs = {input};
+  return SampleFromVector(inputs);
+}
+
+absl::StatusOr<std::vector<std::vector<cv::Mat>>>
+ImageBatchSampler::SampleFromVector(const std::vector<std::string> &inputs) {
+  std::vector<std::vector<cv::Mat>> results;
+  std::vector<cv::Mat> current_batch;
+  input_path_.clear();
+  for (size_t i = 0; i < inputs.size(); ++i) {
+    const std::string &input = inputs[i];
+
+    if (Utility::IsDirectory(input)) {
+      absl::StatusOr<std::vector<std::string>> files_result =
+          GetFilesList(input);
+      if (!files_result.ok()) {
+        return files_result.status();
+      }
+      input_path_.insert(input_path_.end(), files_result.value().begin(),
+                         files_result.value().end());
+      absl::StatusOr<std::vector<std::vector<cv::Mat>>> sub_result =
+          SampleFromVector(files_result.value());
+      if (!sub_result.ok()) {
+        return sub_result.status();
+      }
+
+      const std::vector<std::vector<cv::Mat>> &sub_batches = sub_result.value();
+      for (size_t j = 0; j < sub_batches.size(); ++j) {
+        results.push_back(sub_batches[j]);
+      }
+    } else if (Utility::IsImageFile(input)) {
+      if (!Utility::FileExists(input).ok()) {
+        return absl::NotFoundError("File not found: " + input);
+      }
+      input_path_.push_back(input);
+      absl::StatusOr<cv::Mat> image_result = Utility::MyLoadImage(input);
+      if (!image_result.ok()) {
+        return image_result.status();
+      }
+
+      current_batch.push_back(image_result.value());
+
+      if (static_cast<int>(current_batch.size()) == batch_size_) {
+        results.push_back(current_batch);
+        current_batch.clear();
+      }
+    } else {
+      return absl::InvalidArgumentError("Unsupported file type: " + input);
+    }
+  }
+
+  if (!current_batch.empty()) {
+    results.push_back(current_batch); // last batch
+  }
+  return results;
+}
+
+absl::StatusOr<std::vector<std::vector<cv::Mat>>>
+ImageBatchSampler::SampleFromMatVector(const std::vector<cv::Mat> &inputs) {
+  std::vector<std::vector<cv::Mat>> results;
+  std::vector<cv::Mat> current_batch;
+
+  for (size_t i = 0; i < inputs.size(); ++i) {
+    const cv::Mat &image = inputs[i];
+
+    if (image.empty()) {
+      return absl::InvalidArgumentError("Input image at index " +
+                                        std::to_string(i) + " is empty.");
+    }
+
+    current_batch.push_back(image);
+
+    if (static_cast<int>(current_batch.size()) == batch_size_) {
+      results.push_back(current_batch);
+      current_batch.clear();
+    }
+  }
+
+  if (!current_batch.empty()) {
+    results.push_back(current_batch);
+  }
+
+  return results;
+}
diff --git a/deploy/cpp_infer/src/common/image_batch_sampler.h b/deploy/cpp_infer/src/common/image_batch_sampler.h
new file mode 100644
index 0000000..7d82c2d
--- /dev/null
+++ b/deploy/cpp_infer/src/common/image_batch_sampler.h
@@ -0,0 +1,42 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <opencv2/opencv.hpp>
+#include <set>
+#include <string>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+#include "src/base/base_batch_sampler.h"
+
+class ImageBatchSampler : public BaseBatchSampler {
+public:
+  explicit ImageBatchSampler(int batch_size = 1);
+  virtual ~ImageBatchSampler() {} //这里还没调研怎么实现  ？？？
+
+  absl::StatusOr<std::vector<std::vector<cv::Mat>>>
+  SampleFromString(const std::string &input) override;
+
+  absl::StatusOr<std::vector<std::vector<cv::Mat>>>
+  SampleFromVector(const std::vector<std::string> &inputs) override;
+
+  absl::StatusOr<std::vector<std::vector<cv::Mat>>>
+  SampleFromMatVector(const std::vector<cv::Mat> &inputs) override;
+
+private:
+  static const std::set<std::string> kImgSuffixes;
+};
diff --git a/deploy/cpp_infer/src/common/parallel.h b/deploy/cpp_infer/src/common/parallel.h
new file mode 100644
index 0000000..06b3125
--- /dev/null
+++ b/deploy/cpp_infer/src/common/parallel.h
@@ -0,0 +1,229 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+#pragma once
+
+#include <atomic>
+#include <condition_variable>
+#include <iostream>
+#include <memory>
+#include <mutex>
+#include <queue>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+#include "src/base/base_pipeline.h"
+#include "thread_pool.h"
+
+template <typename Pipeline, typename PipelineParams, typename PipelineInput,
+          typename PipelineResult>
+class AutoParallelSimpleInferencePipeline : public BasePipeline {
+private:
+  struct InferenceInstance {
+    std::shared_ptr<BasePipeline> pipeline;
+    std::queue<PipelineInput> task_queue;
+    std::queue<std::promise<PipelineResult>> promise_queue;
+    std::mutex queue_mutex;
+    std::atomic<bool> is_busy{false};
+    int instance_id;
+  };
+
+public:
+  AutoParallelSimpleInferencePipeline(const PipelineParams &params);
+  absl::Status Init();
+
+  std::future<PipelineResult> PredictAsync(const PipelineInput &input);
+
+  absl::Status PredictThread(const PipelineInput &input);
+  absl::StatusOr<PipelineResult> GetResult();
+
+  virtual ~AutoParallelSimpleInferencePipeline();
+
+private:
+  void ProcessInstanceTasks(int instance_id);
+  PipelineParams params_;
+  int thread_num_;
+
+  std::atomic<int> round_robin_index_{0};
+  std::unique_ptr<PaddlePool::ThreadPool> pool_;
+  std::vector<std::unique_ptr<InferenceInstance>> instances_;
+
+  std::queue<std::future<PipelineResult>> legacy_results_;
+  std::mutex legacy_results_mutex_;
+};
+
+template <typename Pipeline, typename PipelineParams, typename PipelineInput,
+          typename PipelineResult>
+AutoParallelSimpleInferencePipeline<Pipeline, PipelineParams, PipelineInput,
+                                    PipelineResult>::
+    AutoParallelSimpleInferencePipeline(const PipelineParams &params)
+    : BasePipeline(), params_(params), thread_num_(params.thread_num) {
+  if (thread_num_ > 1) {
+    auto status = Init();
+    if (!status.ok()) {
+      INFOE("Pipeline pool init error : %s", status.ToString().c_str());
+      exit(-1);
+    }
+  }
+}
+
+template <typename Pipeline, typename PipelineParams, typename PipelineInput,
+          typename PipelineResult>
+absl::Status
+AutoParallelSimpleInferencePipeline<Pipeline, PipelineParams, PipelineInput,
+                                    PipelineResult>::Init() {
+  try {
+    pool_ = std::unique_ptr<PaddlePool::ThreadPool>(
+        new PaddlePool::ThreadPool(thread_num_));
+
+    for (int i = 0; i < thread_num_; i++) {
+      auto instance =
+          std::unique_ptr<InferenceInstance>(new InferenceInstance());
+      instance->instance_id = i;
+
+      instance->pipeline = std::shared_ptr<BasePipeline>(new Pipeline(params_));
+
+      instances_.push_back(std::move(instance));
+    }
+  } catch (const std::bad_alloc &e) {
+    return absl::ResourceExhaustedError(std::string("Out of memory: ") +
+                                        e.what());
+  } catch (const std::exception &e) {
+    return absl::InternalError(std::string("Init failed: ") + e.what());
+  }
+  return absl::OkStatus();
+}
+
+template <typename Pipeline, typename PipelineParams, typename PipelineInput,
+          typename PipelineResult>
+std::future<PipelineResult> AutoParallelSimpleInferencePipeline<
+    Pipeline, PipelineParams, PipelineInput,
+    PipelineResult>::PredictAsync(const PipelineInput &input) {
+  int instance_id = round_robin_index_.fetch_add(1) % thread_num_;
+  auto &instance = instances_[instance_id];
+
+  std::promise<PipelineResult> promise;
+  auto future = promise.get_future();
+
+  {
+    std::lock_guard<std::mutex> lock(instance->queue_mutex);
+    instance->task_queue.push(input);
+    instance->promise_queue.push(std::move(promise));
+  }
+
+  bool expected = false;
+  if (instance->is_busy.compare_exchange_strong(
+          expected, true)) { // one instance just process one input
+    pool_->submit([this, instance_id]() { ProcessInstanceTasks(instance_id); });
+  }
+
+  return future;
+}
+
+template <typename Pipeline, typename PipelineParams, typename PipelineInput,
+          typename PipelineResult>
+void AutoParallelSimpleInferencePipeline<
+    Pipeline, PipelineParams, PipelineInput,
+    PipelineResult>::ProcessInstanceTasks(int instance_id) {
+  auto &instance = instances_[instance_id];
+
+  while (true) {
+    std::vector<std::string> input;
+    std::promise<PipelineResult> promise;
+    {
+      std::lock_guard<std::mutex> lock(instance->queue_mutex);
+      if (instance->task_queue.empty()) {
+        instance->is_busy = false;
+
+        if (!instance->task_queue.empty()) {
+          bool expected = false;
+          if (instance->is_busy.compare_exchange_strong(expected, true)) {
+            continue;
+          }
+        }
+        return;
+      }
+      input = std::move(instance->task_queue.front());
+      instance->task_queue.pop();
+      promise = std::move(instance->promise_queue.front());
+      instance->promise_queue.pop();
+    }
+    try {
+      PipelineResult result = instance->pipeline->Predict(input);
+      promise.set_value(std::move(result));
+    } catch (const std::exception &e) {
+      promise.set_exception(std::current_exception());
+    }
+  }
+}
+
+template <typename Pipeline, typename PipelineParams, typename PipelineInput,
+          typename PipelineResult>
+absl::Status AutoParallelSimpleInferencePipeline<
+    Pipeline, PipelineParams, PipelineInput,
+    PipelineResult>::PredictThread(const PipelineInput &input) {
+  try {
+    auto future = PredictAsync(input);
+
+    std::lock_guard<std::mutex> lock(legacy_results_mutex_);
+    legacy_results_.push(std::move(future));
+
+    return absl::OkStatus();
+  } catch (const std::exception &e) {
+    return absl::InternalError(std::string("Failed to submit inference: ") +
+                               e.what());
+  }
+}
+
+template <typename Pipeline, typename PipelineParams, typename PipelineInput,
+          typename PipelineResult>
+absl::StatusOr<PipelineResult>
+AutoParallelSimpleInferencePipeline<Pipeline, PipelineParams, PipelineInput,
+                                    PipelineResult>::GetResult() {
+  std::lock_guard<std::mutex> lock(legacy_results_mutex_);
+
+  if (legacy_results_.empty())
+    return absl::NotFoundError("No inference result available");
+
+  try {
+    auto future = std::move(legacy_results_.front());
+    legacy_results_.pop();
+
+    PipelineResult result = future.get();
+    return result;
+  } catch (const std::exception &e) {
+    return absl::InternalError(std::string("Failed to get inference result: ") +
+                               e.what());
+  }
+}
+
+template <typename Pipeline, typename PipelineParams, typename PipelineInput,
+          typename PipelineResult>
+AutoParallelSimpleInferencePipeline<
+    Pipeline, PipelineParams, PipelineInput,
+    PipelineResult>::~AutoParallelSimpleInferencePipeline() {
+  while (!legacy_results_.empty()) {
+    try {
+      legacy_results_.front().get();
+    } catch (...) {
+    }
+    legacy_results_.pop();
+  }
+
+  for (auto &instance : instances_) {
+    while (instance->is_busy.load()) {
+      std::this_thread::sleep_for(std::chrono::milliseconds(10));
+    }
+  }
+}
diff --git a/deploy/cpp_infer/src/common/processors.cc b/deploy/cpp_infer/src/common/processors.cc
new file mode 100644
index 0000000..0116455
--- /dev/null
+++ b/deploy/cpp_infer/src/common/processors.cc
@@ -0,0 +1,826 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "processors.h"
+
+#include <algorithm>
+#include <cmath>
+#include <numeric>
+#include <stdexcept>
+#include <unordered_map>
+
+#include "src/utils/ilogger.h"
+#include "src/utils/utility.h"
+
+absl::StatusOr<int> Resize::GetInterp(const std::string &interp) {
+  static const std::unordered_map<std::string, int> interp_map = {
+      {"NEAREST", cv::INTER_NEAREST},
+      {"LINEAR", cv::INTER_LINEAR},
+      {"BICUBIC", cv::INTER_CUBIC},
+      {"AREA", cv::INTER_AREA},
+      {"LANCZOS4", cv::INTER_LANCZOS4}};
+  auto it = interp_map.find(interp);
+  if (it == interp_map.end())
+    return -1;
+  return it->second;
+}
+
+std::pair<std::vector<int>, double>
+Resize::RescaleSize(const std::vector<int> &img_size) const {
+  int img_w = img_size[0], img_h = img_size[1];
+  int target_w = target_size_[0], target_h = target_size_[1];
+  double scale = std::min(static_cast<double>(std::max(target_w, target_h)) /
+                              std::max(img_w, img_h),
+                          static_cast<double>(std::min(target_w, target_h)) /
+                              std::min(img_w, img_h));
+  std::vector<int> rescaled_size = {
+      static_cast<int>(std::round(img_w * scale)),
+      static_cast<int>(std::round(img_h * scale))};
+  return std::make_pair(rescaled_size, scale);
+}
+
+absl::Status Resize::CheckImageSize() const {
+  if (target_size_.size() != 2) {
+    return absl::InvalidArgumentError("Size must be a vector of two elements.");
+  }
+  if (target_size_[0] <= 0 || target_size_[1] <= 0) {
+    return absl::InvalidArgumentError("Width and height must be positive.");
+  }
+  return absl::OkStatus();
+}
+
+Resize::Resize(const std::vector<int> &target_size, bool keep_ratio,
+               int size_divisor, const std::string &interp)
+    : keep_ratio_(keep_ratio), size_divisor_(size_divisor) {
+  if (target_size.size() == 1) {
+    target_size_ = {target_size[0], target_size[0]};
+  } else {
+    target_size_ = target_size;
+  }
+  absl::Status status = CheckImageSize();
+  if (!status.ok()) {
+    INFOE("image check fail : %s", status.ToString().c_str());
+    exit(-1);
+  }
+  std::string interp_upper = interp;
+  std::transform(interp_upper.begin(), interp_upper.end(), interp_upper.begin(),
+                 ::toupper);
+
+  auto interp_value = GetInterp(interp_upper);
+  if (!interp_value.ok()) {
+    INFOE("Unknown type: %s", interp_value.status().ToString().c_str());
+    exit(-1);
+  }
+  interp_ = interp_value.value();
+}
+
+absl::StatusOr<std::vector<cv::Mat>> Resize::Apply(std::vector<cv::Mat> &input,
+                                                   const void *param) const {
+  std::vector<cv::Mat> out_imgs;
+  for (const auto &img : input) {
+    auto out = ResizeOne(img);
+    if (!out.ok())
+      return out.status();
+    out_imgs.push_back(std::move(out.value()));
+  }
+  return out_imgs;
+}
+
+absl::StatusOr<cv::Mat> Resize::ResizeOne(const cv::Mat &img) const {
+  if (img.empty()) {
+    return absl::InvalidArgumentError("Input image is empty.");
+  }
+
+  std::vector<int> cur_target = target_size_;
+  auto size_test = img.size();
+  cv::Size orig_size = img.size();
+  int orig_w = orig_size.width, orig_h = orig_size.height;
+
+  if (keep_ratio_) {
+    std::vector<int> wh = {orig_w, orig_h};
+    auto rescale = RescaleSize(wh);
+    cur_target = rescale.first;
+  }
+
+  if (size_divisor_ > 0) {
+    for (auto &x : cur_target) {
+      x = static_cast<int>(std::ceil(static_cast<double>(x) / size_divisor_)) *
+          size_divisor_;
+    }
+  }
+
+  cv::Mat out;
+  cv::resize(img, out, cv::Size(cur_target[0], cur_target[1]), 0, 0, interp_);
+  return out;
+}
+
+ResizeByShort::ResizeByShort(int target_short_edge, int size_divisor,
+                             const std::string &interp)
+    : target_short_edge_(target_short_edge), size_divisor_(size_divisor) {
+  std::string interp_upper = interp;
+  std::transform(interp_upper.begin(), interp_upper.end(), interp_upper.begin(),
+                 ::toupper);
+
+  auto interp_value = Resize::GetInterp(interp_upper);
+  if (!interp_value.ok()) {
+    INFOE("Unknown type: %s", interp_value.status().ToString().c_str());
+    exit(-1);
+  }
+  interp_ = interp_value.value();
+}
+absl::StatusOr<std::vector<cv::Mat>>
+ResizeByShort::Apply(std::vector<cv::Mat> &input, const void *param) const {
+  std::vector<cv::Mat> out_imgs;
+  for (auto &image : input) {
+    auto out = ResizeOne(image);
+    if (!out.ok())
+      return out.status();
+    out_imgs.push_back(std::move(out.value()));
+  }
+  return out_imgs;
+}
+
+absl::StatusOr<cv::Mat> ResizeByShort::ResizeOne(const cv::Mat &img) const {
+  if (img.empty()) {
+    return absl::InvalidArgumentError("Input image is empty.");
+  }
+  int h = img.size[0];
+  int w = img.size[1];
+  int short_edge = std::min(h, w);
+  float scale = static_cast<double>(target_short_edge_) / short_edge;
+  int h_resize = static_cast<int>(std::round(h * scale));
+  int w_resize = static_cast<int>(std::round(w * scale));
+
+  if (size_divisor_ > 0) {
+    h_resize = static_cast<int>(std::ceil(h_resize / (float)size_divisor_)) *
+               size_divisor_;
+    w_resize = static_cast<int>(std::ceil(w_resize / (float)size_divisor_)) *
+               size_divisor_;
+  }
+
+  cv::Mat dst;
+  cv::resize(img, dst, cv::Size(w_resize, h_resize), 0, 0, interp_);
+  return dst;
+}
+
+ReadImage::ReadImage(const std::string &format) {
+  auto fmt = StringToFormat(format);
+  if (!fmt.ok()) {
+    INFOE(fmt.status().ToString().c_str());
+    exit(-1);
+  }
+  format_ = *fmt;
+}
+
+absl::StatusOr<std::vector<cv::Mat>>
+ReadImage::Apply(std::vector<cv::Mat> &input, const void *param_ptr) const {
+  if (input.empty()) {
+    return absl::InvalidArgumentError("Input image vector is empty.");
+  }
+  std::vector<cv::Mat> output;
+  output.reserve(input.size());
+
+  for (size_t i = 0; i < input.size(); ++i) {
+    const cv::Mat &img = input[i];
+    if (img.empty()) {
+      return absl::InvalidArgumentError("Image at index " + std::to_string(i) +
+                                        " is empty.");
+    }
+
+    cv::Mat converted;
+    switch (format_) {
+    case Format::BGR:
+      if (img.channels() == 3) {
+        converted = img.clone();
+      } else if (img.channels() == 1) {
+        cv::cvtColor(img, converted, cv::COLOR_GRAY2BGR);
+      } else {
+        return absl::InvalidArgumentError("Image at index " +
+                                          std::to_string(i) +
+                                          " channel not supported for BGR.");
+      }
+      break;
+    case Format::RGB:
+      if (img.channels() == 3) {
+        cv::cvtColor(img, converted, cv::COLOR_BGR2RGB);
+      } else if (img.channels() == 1) {
+        cv::cvtColor(img, converted, cv::COLOR_GRAY2RGB);
+      } else {
+        return absl::InvalidArgumentError("Image at index " +
+                                          std::to_string(i) +
+                                          " channel not supported for RGB.");
+      }
+      break;
+    case Format::GRAY:
+      if (img.channels() == 3) {
+        cv::cvtColor(img, converted, cv::COLOR_BGR2GRAY);
+      } else if (img.channels() == 1) {
+        converted = img.clone();
+      } else {
+        return absl::InvalidArgumentError("Image at index " +
+                                          std::to_string(i) +
+                                          " channel not supported for GRAY.");
+      }
+      break;
+    default:
+      return absl::InvalidArgumentError("Unknown format.");
+    }
+    output.push_back(std::move(converted));
+  }
+  return output;
+}
+
+absl::StatusOr<ReadImage::Format>
+ReadImage::StringToFormat(const std::string &format) {
+  if (format == "BGR")
+    return Format::BGR;
+  if (format == "RGB")
+    return Format::RGB;
+  if (format == "GRAY")
+    return Format::GRAY;
+  return absl::InvalidArgumentError("Unsupported format: " + format);
+}
+
+absl::StatusOr<std::vector<cv::Mat>>
+ToCHWImage::operator()(const std::vector<cv::Mat> &imgs_batch) {
+  std::vector<std::vector<cv::Mat>> chw_imgs_batch;
+
+  std::vector<cv::Mat> chw_imgs;
+  for (const auto &img : imgs_batch) {
+    if (img.empty()) {
+      return absl::InvalidArgumentError("Input image is empty!");
+    }
+    if (img.channels() != 3) {
+      return absl::InvalidArgumentError(
+          "Input image must have 3 channels (HWC format)!");
+    }
+
+    cv::Mat chw_img(3, img.rows * img.cols, CV_32F);
+    float *ptr = chw_img.ptr<float>();
+
+    for (int h = 0; h < img.rows; ++h) {
+      for (int w = 0; w < img.cols; ++w) {
+        const cv::Vec3b &pixel = img.at<cv::Vec3b>(h, w);
+        ptr[0 * img.total() + h * img.cols + w] = pixel[0];
+        ptr[1 * img.total() + h * img.cols + w] = pixel[1];
+        ptr[2 * img.total() + h * img.cols + w] = pixel[2];
+      }
+    }
+
+    chw_imgs.push_back(chw_img);
+  }
+
+  return chw_imgs;
+};
+
+Normalize::Normalize(float scale, const std::vector<float> &mean,
+                     const std::vector<float> &std)
+    : alpha_(CHANNEL), beta_(CHANNEL) {
+  assert(mean.size() == CHANNEL && std.size() == CHANNEL);
+  for (size_t i = 0; i < CHANNEL; ++i) {
+    alpha_[i] = scale / std.at(i);
+    beta_[i] = -mean.at(i) / std.at(i);
+  }
+}
+Normalize::Normalize(float scale, const float &mean, const float &std)
+    : alpha_(CHANNEL), beta_(CHANNEL) {
+  for (size_t i = 0; i < CHANNEL; ++i) {
+    alpha_[i] = scale / std;
+    beta_[i] = -mean / std;
+  }
+}
+
+absl::StatusOr<cv::Mat> Normalize::NormalizeOne(const cv::Mat &image) const {
+  if (image.empty()) {
+    return absl::InvalidArgumentError("Input image is empty.");
+  }
+  if (image.channels() != CHANNEL) {
+    return absl::InvalidArgumentError("Input image must have 3 dims");
+  }
+  if (image.depth() != CV_8U && image.depth() != CV_32F) {
+    return absl::InvalidArgumentError("Input image must be CV_8U or CV_32F.");
+  }
+  cv::Mat input;
+  if (image.depth() == CV_8U) {
+    image.convertTo(input, CV_32F);
+  } else {
+    input = image.clone(); // note origin type is CV_8U
+  }
+  if (input.channels() == CHANNEL) {
+    cv::Mat processed = input;
+    std::vector<cv::Mat> channels(input.channels());
+    cv::split(processed, channels);
+
+    for (int c = 0; c < input.channels(); ++c) {
+      channels[c] = channels[c] * alpha_[c] + beta_[c];
+    }
+    cv::merge(channels, processed);
+    return processed;
+  } else { // dims >= 3
+    assert(input.isContinuous());
+    int total = 1;
+    for (int i = 0; i < input.dims - 1; i++) {
+      total *= input.size[i];
+    }
+    float *data = input.ptr<float>();
+    for (int i = 0; i < total; i++) {
+      float *group = data + i * CHANNEL;
+      for (int j = 0; j < CHANNEL; j++) {
+        group[j] = group[j] * alpha_[j] + beta_[j];
+      }
+    }
+    return input;
+  }
+}
+absl::StatusOr<std::vector<cv::Mat>>
+Normalize::Apply(std::vector<cv::Mat> &input, const void *param) const {
+  std::vector<cv::Mat> results_norm;
+  results_norm.reserve(input.size());
+  for (const auto &img : input) {
+    auto norm_single = NormalizeOne(img);
+    if (!norm_single.ok()) {
+      return norm_single.status();
+    }
+    results_norm.emplace_back(norm_single.value());
+  }
+  return results_norm;
+}
+
+NormalizeImage::NormalizeImage(float scale, const std::vector<float> &mean,
+                               const std::vector<float> &std)
+    : alpha_(CHANNEL), beta_(CHANNEL) {
+  assert(mean.size() == CHANNEL && std.size() == CHANNEL);
+  for (size_t i = 0; i < CHANNEL; ++i) {
+    alpha_[i] = scale / std.at(i);
+    beta_[i] = -mean.at(i) / std.at(i);
+  }
+}
+
+absl::StatusOr<cv::Mat> NormalizeImage::Normalize(const cv::Mat &img) const {
+  if (img.empty()) {
+    return absl::InvalidArgumentError("Input image is empty.");
+  }
+  if (img.channels() != CHANNEL) {
+    return absl::InvalidArgumentError("Input image must have 3 channels.");
+  }
+  if (img.depth() != CV_8U && img.depth() != CV_32F) {
+    return absl::InvalidArgumentError("Input image must be CV_8U or CV_32F.");
+  }
+
+  cv::Mat input;
+  if (img.depth() == CV_8U) {
+    img.convertTo(input, CV_32F);
+  } else {
+    input = img.clone();
+  }
+
+  cv::Mat processed = input;
+
+  std::vector<cv::Mat> channels(CHANNEL);
+
+  cv::split(processed, channels);
+
+  for (int c = 0; c < CHANNEL; ++c) {
+    channels[c] = channels[c] * alpha_[c] + beta_[c];
+  }
+
+  cv::merge(channels, processed);
+  return processed;
+}
+
+absl::StatusOr<std::vector<cv::Mat>>
+NormalizeImage::Apply(std::vector<cv::Mat> &imgs, const void *param) const {
+  std::vector<cv::Mat> results;
+  results.reserve(imgs.size());
+  for (const auto &img : imgs) {
+    auto normed = this->Normalize(img);
+    if (!normed.ok()) {
+      return normed.status();
+    }
+    results.push_back(std::move(normed).value());
+  }
+  return results;
+}
+
+// absl::StatusOr<std::vector<cv::Mat>> ToCHWImage::Apply(
+//   std::vector<cv::Mat>& input, const void* param) const {
+//   std::vector<cv::Mat> chw_imgs;
+//   for (const auto& img : input) {
+//     if (img.empty()) {
+//       return absl::InvalidArgumentError("Input image is empty!");
+//     }
+//     if (img.channels() != 3) {
+//       return absl::InvalidArgumentError(
+//           "Input image must have 3 channels (HWC format)!");
+//     }
+
+//     std::vector<int> shape_chw = {img.channels(), img.rows, img.cols};  //
+//     Define sizes for CHW cv::Mat chw_img(shape_chw.size(), shape_chw.data(),
+//     CV_32F); float* ptr = chw_img.ptr<float>(); for (int h = 0; h < img.rows;
+//     ++h) {
+//       for (int w = 0; w < img.cols; ++w) {
+//         const cv::Vec3f& pixel = img.at<cv::Vec3f>(h, w);
+//         ptr[0 * img.total() + h * img.cols + w] = pixel[0];
+//         ptr[1 * img.total() + h * img.cols + w] = pixel[1];
+//         ptr[2 * img.total() + h * img.cols + w] = pixel[2];
+//       }
+//     }
+
+//     chw_imgs.push_back(chw_img);
+//   }
+
+//   return chw_imgs;
+// }
+
+absl::StatusOr<std::vector<cv::Mat>>
+ToCHWImage::Apply(std::vector<cv::Mat> &input, const void *param) const {
+  std::vector<cv::Mat> chw_imgs;
+  for (const auto &img : input) {
+    if (img.empty()) {
+      return absl::InvalidArgumentError("Input image is empty!");
+    }
+    if (img.channels() != 3) {
+      return absl::InvalidArgumentError(
+          "Input image must have 3 channels (HWC format)!");
+    }
+
+    std::vector<cv::Mat> vec_split = {};
+    cv::split(img, vec_split);
+    cv::Mat chw_img;
+    for (auto &split : vec_split)
+      split = split.reshape(1, 1);
+    cv::hconcat(vec_split, chw_img);
+    std::vector<int> shape = {img.channels(), img.size[0], img.size[1]};
+    chw_img = chw_img.reshape(1, shape);
+    chw_imgs.push_back(chw_img);
+  }
+
+  return chw_imgs;
+}
+
+absl::StatusOr<std::vector<cv::Mat>>
+ToBatch::operator()(const std::vector<cv::Mat> &imgs) const {
+  if (imgs.empty()) {
+    return absl::InvalidArgumentError("Input image vector is empty.");
+  }
+  const int batch = imgs.size();
+  const int rows = imgs[0].rows;
+  const int cols = imgs[0].cols;
+  const int channels = imgs[0].channels();
+
+  for (size_t i = 0; i < imgs.size(); ++i) {
+    if (imgs[i].rows != rows || imgs[i].cols != cols ||
+        imgs[i].channels() != channels) {
+      return absl::InvalidArgumentError(
+          "All images must have the same size and number of channels.");
+    }
+  }
+
+  std::vector<int> sizes = {batch, rows, cols, channels};
+  cv::Mat out(4, sizes.data(), CV_32F);
+
+  for (int b = 0; b < batch; ++b) {
+    cv::Mat img_float;
+    if (imgs[b].depth() != CV_32F) {
+      imgs[b].convertTo(img_float, CV_32F);
+    } else {
+      img_float = imgs[b];
+    }
+
+    for (int r = 0; r < rows; ++r) {
+      for (int c = 0; c < cols; ++c) {
+        if (channels == 1) {
+          float v = img_float.at<float>(r, c);
+          int idx[4] = {b, r, c, 0};
+          out.at<float>(idx) = v;
+        } else if (channels == 3) {
+          cv::Vec3f v = img_float.at<cv::Vec3f>(r, c);
+          for (int ch = 0; ch < 3; ++ch) {
+            int idx[4] = {b, r, c, ch};
+            out.at<float>(idx) = v[ch];
+          }
+        } else {
+          const float *pix = img_float.ptr<float>(r, c);
+          for (int ch = 0; ch < channels; ++ch) {
+            int idx[4] = {b, r, c, ch};
+            out.at<float>(idx) = pix[ch];
+          }
+        }
+      }
+    }
+  }
+  std::vector<cv::Mat> result{out};
+  return result;
+}
+
+absl::StatusOr<std::vector<cv::Mat>> ToBatch::Apply(std::vector<cv::Mat> &input,
+                                                    const void *param) const {
+  if (input.empty()) {
+    return absl::InvalidArgumentError("Input image vector is empty.");
+  }
+
+  std::vector<int> batch_shape = {(int)input.size()};
+  for (const auto &image : input) {
+    if (image.dims != input[0].dims) {
+      return absl::InvalidArgumentError("All images must have the same dims.");
+    } else {
+      for (int i = 0; i < input[0].dims; i++) {
+        if (image.size[i] != input[0].size[i]) {
+          return absl::InvalidArgumentError(
+              "All images must have the same size and number of channels.");
+        }
+        if (&image == &(*std::begin(input)))
+          batch_shape.emplace_back(input[0].size[i]);
+      }
+    }
+  }
+  cv::Mat batch_out;
+  for (auto &image : input)
+    image = image.reshape(1, 1);
+  cv::vconcat(input, batch_out);
+  batch_out = batch_out.reshape(1, batch_shape);
+  std::vector<cv::Mat> out = {batch_out};
+  return out;
+}
+
+absl::StatusOr<cv::Mat> ComponentsProcessor::RotateImage(const cv::Mat &image,
+                                                         int angle) {
+  if (image.empty() || image.channels() != 3) {
+    return absl::InvalidArgumentError("image is invalid");
+  }
+  if (angle < 0 || angle >= 360) {
+    return absl::InvalidArgumentError("`angle` should be in range [0, 360)");
+  }
+  if (std::abs(angle) < 1e-7) {
+    return image.clone();
+  }
+
+  int h = image.rows;
+  int w = image.cols;
+  cv::Point2f center(w / 2.0f, h / 2.0f);
+  double scale = 1.0;
+  cv::Mat rot_mat = cv::getRotationMatrix2D(center, angle, scale);
+
+  double abs_cos = std::abs(rot_mat.at<double>(0, 0));
+  double abs_sin = std::abs(rot_mat.at<double>(0, 1));
+  int new_w = int(h * abs_sin + w * abs_cos);
+  int new_h = int(h * abs_cos + w * abs_sin);
+
+  rot_mat.at<double>(0, 2) += (new_w - w) / 2.0;
+  rot_mat.at<double>(1, 2) += (new_h - h) / 2.0;
+
+  cv::Mat rotated;
+  cv::warpAffine(image, rotated, rot_mat, cv::Size(new_w, new_h),
+                 cv::INTER_CUBIC);
+
+  return rotated;
+}
+
+std::vector<std::vector<cv::Point2f>> ComponentsProcessor::SortQuadBoxes(
+    const std::vector<std::vector<cv::Point2f>> &dt_polys) {
+  std::vector<std::vector<cv::Point2f>> dt_boxes = dt_polys;
+
+  std::sort(
+      dt_boxes.begin(), dt_boxes.end(),
+      [](const std::vector<cv::Point2f> &a, const std::vector<cv::Point2f> &b) {
+        return (a[0].y < b[0].y) || (a[0].y == b[0].y && a[0].x < b[0].x);
+      });
+
+  for (size_t i = 0; i < dt_boxes.size() - 1; ++i) {
+    for (size_t j = i + 1; j > 0; --j) {
+      if (std::abs(dt_boxes[j][0].y - dt_boxes[j - 1][0].y) < 10 &&
+          dt_boxes[j][0].x < dt_boxes[j - 1][0].x) {
+        std::swap(dt_boxes[j], dt_boxes[j - 1]);
+      } else {
+        break;
+      }
+    }
+  }
+  return dt_boxes;
+}
+
+std::vector<std::vector<cv::Point2f>> ComponentsProcessor::SortPolyBoxes(
+    const std::vector<std::vector<cv::Point2f>> &dt_polys) {
+  size_t num_boxes = dt_polys.size();
+  if (num_boxes == 0)
+    return dt_polys;
+  std::vector<int> y_min_list(num_boxes);
+  for (size_t i = 0; i < num_boxes; ++i) {
+    int y_min = dt_polys[i][0].y;
+    for (size_t j = 1; j < dt_polys[i].size(); ++j) {
+      if (dt_polys[i][j].y < y_min) {
+        y_min = dt_polys[i][j].y;
+      }
+    }
+    y_min_list[i] = y_min;
+  }
+  std::vector<size_t> rank(num_boxes);
+  std::iota(rank.begin(), rank.end(), 0);
+  std::sort(rank.begin(), rank.end(),
+            [&](size_t a, size_t b) { return y_min_list[a] < y_min_list[b]; });
+  std::vector<std::vector<cv::Point2f>> dt_polys_rank(num_boxes);
+  for (size_t i = 0; i < num_boxes; ++i) {
+    dt_polys_rank[i] = dt_polys[rank[i]];
+  }
+  return dt_polys_rank;
+}
+
+std::vector<std::array<float, 4>> ComponentsProcessor::ConvertPointsToBoxes(
+    const std::vector<std::vector<cv::Point2f>> &dt_polys) {
+  std::vector<std::array<float, 4>> dt_boxes;
+  for (const auto &poly : dt_polys) {
+    if (poly.empty()) {
+      continue;
+    }
+    float left = std::numeric_limits<float>::max();
+    float right = std::numeric_limits<float>::lowest();
+    float top = std::numeric_limits<float>::max();
+    float bottom = std::numeric_limits<float>::lowest();
+
+    for (const auto &pt : poly) {
+      if (pt.x < left)
+        left = pt.x;
+      if (pt.x > right)
+        right = pt.x;
+      if (pt.y < top)
+        top = pt.y;
+      if (pt.y > bottom)
+        bottom = pt.y;
+    }
+    dt_boxes.push_back({left, top, right, bottom});
+  }
+  return dt_boxes;
+}
+
+CropByPolys::CropByPolys(const std::string &box_type) {
+  assert(box_type == "quad" || box_type == "poly");
+  if (box_type == "quad") {
+    box_type_ = DetBoxType::kQuad;
+  } else {
+    box_type_ = DetBoxType::kPoly;
+  }
+}
+
+absl::StatusOr<std::vector<cv::Mat>>
+CropByPolys::operator()(const cv::Mat &img,
+                        const std::vector<std::vector<cv::Point2f>> &dt_polys) {
+  if (img.empty())
+    return absl::InvalidArgumentError("Input image is empty.");
+  std::vector<cv::Mat> output_list;
+  try {
+    if (box_type_ == DetBoxType::kQuad) {
+      for (const auto &poly : dt_polys) {
+        auto out = GetMinAreaRectCrop(img, poly);
+        if (!out.ok())
+          return out.status();
+        output_list.push_back(*out);
+      }
+    } else if (box_type_ == DetBoxType::kPoly) {
+      for (const auto &poly : dt_polys) {
+        auto out = GetPolyRectCrop(img, poly);
+        if (!out.ok())
+          return out.status();
+        output_list.push_back(*out);
+      }
+    } else {
+      return absl::UnimplementedError("Unknown box type.");
+    }
+  } catch (const std::exception &e) {
+    return absl::InternalError(std::string("Exception: ") + e.what());
+  }
+  return output_list;
+}
+
+absl::StatusOr<cv::Mat>
+CropByPolys::GetMinAreaRectCrop(const cv::Mat &img,
+                                const std::vector<cv::Point2f> &points) const {
+  if (points.size() < 4)
+    return absl::InvalidArgumentError("Less than 4 points for min area rect.");
+  std::vector<cv::Point2f> box = GetMinAreaRectPoints(points);
+  return GetRotateCropImage(img, box);
+}
+
+absl::StatusOr<cv::Mat>
+CropByPolys::GetRotateCropImage(const cv::Mat &img,
+                                const std::vector<cv::Point2f> &box) const {
+  if (box.size() != 4)
+    return absl::InvalidArgumentError("Box must have 4 points.");
+  float widthTop = cv::norm(box[0] - box[1]);
+  float widthBottom = cv::norm(box[2] - box[3]);
+  float maxWidth = std::max(widthTop, widthBottom);
+
+  float heightLeft = cv::norm(box[0] - box[3]);
+  float heightRight = cv::norm(box[1] - box[2]);
+  float maxHeight = std::max(heightLeft, heightRight);
+
+  std::vector<cv::Point2f> dst = {
+      cv::Point2f(0, 0), cv::Point2f(maxWidth - 1, 0),
+      cv::Point2f(maxWidth - 1, maxHeight - 1), cv::Point2f(0, maxHeight - 1)};
+  cv::Mat M = cv::getPerspectiveTransform(box, dst);
+  cv::Mat out;
+  cv::warpPerspective(img, out, M, cv::Size((int)maxWidth, (int)maxHeight),
+                      cv::INTER_CUBIC, cv::BORDER_REPLICATE);
+  if (out.rows != 0 && 1.0 * out.rows / out.cols >= 1.5)
+    cv::rotate(out, out, cv::ROTATE_90_COUNTERCLOCKWISE);
+  return out;
+}
+
+std::vector<cv::Point2f>
+CropByPolys::GetMinAreaRectPoints(const std::vector<cv::Point2f> &poly) const {
+  auto pts = poly;
+  if (pts.size() < 4)
+    return {};
+  cv::RotatedRect minRect = cv::minAreaRect(pts);
+  std::vector<cv::Point2f> box(4);
+  minRect.points(box.data());
+  std::sort(box.begin(), box.end(),
+            [](const cv::Point2f &a, const cv::Point2f &b) {
+              return a.x < b.x || (a.x == b.x && a.y < b.y);
+            });
+  size_t index_a = 0, index_d = 1;
+  if (box[1].y > box[0].y) {
+    index_a = 0;
+    index_d = 1;
+  } else {
+    index_a = 1;
+    index_d = 0;
+  }
+  size_t index_b = 2, index_c = 3;
+  if (box[3].y > box[2].y) {
+    index_b = 2;
+    index_c = 3;
+  } else {
+    index_b = 3;
+    index_c = 2;
+  }
+  return {box[index_a], box[index_b], box[index_c], box[index_d]};
+}
+
+absl::StatusOr<cv::Mat>
+CropByPolys::GetPolyRectCrop(const cv::Mat &img,
+                             const std::vector<cv::Point2f> &poly) const {
+  if (poly.size() < 4)
+    return absl::InvalidArgumentError(
+        "Less than 4 points for GetPolyRectCrop.");
+  // 对Poly和最小外接矩形做IoU判断
+  std::vector<cv::Point2f> minrect = GetMinAreaRectPoints(poly);
+  if (minrect.size() != 4)
+    return absl::InternalError("Failed to get minarea rect.");
+  double iou = IoU(poly, minrect);
+  // 若IoU>0.7则返回直接crop，否则可做更复杂处理，如透视矫正，可进一步实现自定义变形矫正
+  auto crop_result = GetRotateCropImage(img, minrect);
+  if (!crop_result.ok())
+    return crop_result.status();
+  // 测试下如果IoU很高就用直接的最小外接矩形crop，否则复杂矫正（本实现只用直接crop）
+  // 若需更强几何修复，可集成TPS、ThinPlateSpline或AutoRectifier
+  return *crop_result;
+}
+
+const double CropByPolys::SCALE = 10000.0;
+
+ClipperLib::Path
+CropByPolys::CvPolyToClipperPath(const std::vector<cv::Point2f> &poly) {
+  ClipperLib::Path path;
+  for (const auto &pt : poly)
+    path.emplace_back(static_cast<ClipperLib::cInt>(std::round(pt.x * SCALE)),
+                      static_cast<ClipperLib::cInt>(std::round(pt.y * SCALE)));
+  return path;
+}
+
+double CropByPolys::IoU(const std::vector<cv::Point2f> &poly1,
+                        const std::vector<cv::Point2f> &poly2) {
+  auto path1 = CvPolyToClipperPath(poly1);
+  auto path2 = CvPolyToClipperPath(poly2);
+  ClipperLib::Paths inter_solution, union_solution;
+  ClipperLib::Clipper c_inter, c_union;
+  c_inter.AddPath(path1, ClipperLib::ptSubject, true);
+  c_inter.AddPath(path2, ClipperLib::ptClip, true);
+  c_inter.Execute(ClipperLib::ctIntersection, inter_solution,
+                  ClipperLib::pftNonZero, ClipperLib::pftNonZero);
+  double area_inter = 0.0;
+  for (const auto &p : inter_solution)
+    area_inter += std::fabs(ClipperLib::Area(p));
+  c_union.AddPath(path1, ClipperLib::ptSubject, true);
+  c_union.AddPath(path2, ClipperLib::ptClip, true);
+  c_union.Execute(ClipperLib::ctUnion, union_solution, ClipperLib::pftNonZero,
+                  ClipperLib::pftNonZero);
+  double area_union = 0.0;
+  for (const auto &p : union_solution)
+    area_union += std::fabs(ClipperLib::Area(p));
+  area_inter /= (SCALE * SCALE);
+  area_union /= (SCALE * SCALE);
+  if (area_union < 1e-8)
+    return 0.0;
+  return area_inter / area_union;
+}
diff --git a/deploy/cpp_infer/src/common/processors.h b/deploy/cpp_infer/src/common/processors.h
new file mode 100644
index 0000000..c63f7ae
--- /dev/null
+++ b/deploy/cpp_infer/src/common/processors.h
@@ -0,0 +1,182 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <iostream>
+#include <opencv2/opencv.hpp>
+#include <string>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+#include "polyclipping/clipper.hpp"
+#include "src/utils/func_register.h"
+
+class Resize : public BaseProcessor {
+public:
+  Resize(const std::vector<int> &target_size, bool keep_ratio = false,
+         int size_divisor = 0, const std::string &interp = "LINEAR");
+  absl::Status CheckImageSize() const;
+  std::pair<std::vector<int>, double>
+  RescaleSize(const std::vector<int> &img_size) const;
+  absl::StatusOr<std::vector<cv::Mat>>
+  Apply(std::vector<cv::Mat> &input,
+        const void *param = nullptr) const override;
+  absl::StatusOr<cv::Mat> ResizeOne(const cv::Mat &img) const;
+  static absl::StatusOr<int> GetInterp(const std::string &interp);
+
+private:
+  std::vector<int> target_size_;
+  bool keep_ratio_;
+  int size_divisor_;
+  int interp_;
+};
+
+class ResizeByShort : public BaseProcessor {
+public:
+  ResizeByShort(int target_short_edge, int size_divisor = 0,
+                const std::string &interp = "LINEAR");
+  absl::StatusOr<std::vector<cv::Mat>>
+  Apply(std::vector<cv::Mat> &input,
+        const void *param = nullptr) const override;
+  absl::StatusOr<cv::Mat> ResizeOne(const cv::Mat &img) const;
+
+private:
+  int target_short_edge_;
+  int size_divisor_;
+  int interp_;
+};
+
+class ReadImage : public BaseProcessor {
+public:
+  enum class Format { BGR, RGB, GRAY };
+
+  ReadImage(const std::string &format = "RGB");
+
+  ReadImage(const ReadImage &) = delete;
+  ReadImage &operator=(const ReadImage &) = delete;
+
+  absl::StatusOr<std::vector<cv::Mat>>
+  Apply(std::vector<cv::Mat> &input,
+        const void *param_ptr = nullptr) const override;
+
+private:
+  static absl::StatusOr<Format> StringToFormat(const std::string &format);
+  Format format_;
+};
+
+class Normalize : public BaseProcessor {
+public:
+  Normalize(float scale = 1.0 / 255.0,
+            const std::vector<float> &mean = {0.5, 0.5, 0.5},
+            const std::vector<float> &std = {0.5, 0.5, 0.5});
+  Normalize(float scale = 1.0 / 255.0, const float &mean = 0.5,
+            const float &std = 0.5);
+  absl::StatusOr<cv::Mat> NormalizeOne(const cv::Mat &input) const;
+  absl::StatusOr<std::vector<cv::Mat>>
+  Apply(std::vector<cv::Mat> &input,
+        const void *param = nullptr) const override;
+  static constexpr int CHANNEL = 3;
+
+private:
+  std::vector<float> alpha_;
+  std::vector<float> beta_;
+};
+
+class NormalizeImage : public BaseProcessor {
+public:
+  NormalizeImage(float scale = 1.0 / 255.0,
+                 const std::vector<float> &mean = {0.485, 0.456, 0.406},
+                 const std::vector<float> &std = {0.229, 0.224, 0.225});
+
+  absl::StatusOr<std::vector<cv::Mat>>
+  Apply(std::vector<cv::Mat> &input,
+        const void *param = nullptr) const override;
+
+private:
+  std::vector<float> alpha_;
+  std::vector<float> beta_;
+
+  absl::StatusOr<cv::Mat> Normalize(const cv::Mat &img) const;
+  NormalizeImage(const NormalizeImage &) = delete;
+  NormalizeImage &operator=(const NormalizeImage &) = delete;
+  static constexpr int CHANNEL = 3;
+};
+
+class ToCHWImage : public BaseProcessor {
+public:
+  absl::StatusOr<std::vector<cv::Mat>>
+  operator()(const std::vector<cv::Mat> &imgs_batch);
+  absl::StatusOr<std::vector<cv::Mat>>
+  Apply(std::vector<cv::Mat> &input,
+        const void *param = nullptr) const override;
+};
+
+class ToBatch : public BaseProcessor {
+public:
+  absl::StatusOr<std::vector<cv::Mat>>
+  operator()(const std::vector<cv::Mat> &imgs) const;
+  absl::StatusOr<std::vector<cv::Mat>>
+  Apply(std::vector<cv::Mat> &input,
+        const void *param = nullptr) const override;
+};
+
+class ComponentsProcessor {
+public:
+  static absl::StatusOr<cv::Mat> RotateImage(const cv::Mat &image, int angle);
+  static std::vector<std::vector<cv::Point2f>>
+  SortQuadBoxes(const std::vector<std::vector<cv::Point2f>> &dt_polys);
+  static std::vector<std::vector<cv::Point2f>>
+  SortPolyBoxes(const std::vector<std::vector<cv::Point2f>> &dt_polys);
+  static std::vector<std::array<float, 4>>
+  ConvertPointsToBoxes(const std::vector<std::vector<cv::Point2f>> &dt_polys);
+};
+
+class CropByPolys {
+public:
+  enum class DetBoxType { kQuad, kPoly };
+
+  CropByPolys(const std::string &box_type = "quad");
+
+  absl::StatusOr<std::vector<cv::Mat>>
+  operator()(const cv::Mat &img,
+             const std::vector<std::vector<cv::Point2f>> &dt_polys);
+
+  absl::StatusOr<cv::Mat>
+  GetMinAreaRectCrop(const cv::Mat &img,
+                     const std::vector<cv::Point2f> &points) const;
+
+  absl::StatusOr<cv::Mat>
+  GetPolyRectCrop(const cv::Mat &img,
+                  const std::vector<cv::Point2f> &poly) const;
+
+  absl::StatusOr<cv::Mat>
+  GetRotateCropImage(const cv::Mat &img,
+                     const std::vector<cv::Point2f> &box) const;
+
+  std::vector<cv::Point2f>
+  GetMinAreaRectPoints(const std::vector<cv::Point2f> &poly) const;
+
+  static double IoU(const std::vector<cv::Point2f> &poly1,
+                    const std::vector<cv::Point2f> &poly2);
+
+  static ClipperLib::Path
+  CvPolyToClipperPath(const std::vector<cv::Point2f> &poly);
+
+  static const double SCALE;
+
+private:
+  DetBoxType box_type_;
+};
diff --git a/deploy/cpp_infer/src/common/static_infer.cc b/deploy/cpp_infer/src/common/static_infer.cc
new file mode 100644
index 0000000..f35e4d0
--- /dev/null
+++ b/deploy/cpp_infer/src/common/static_infer.cc
@@ -0,0 +1,197 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "static_infer.h"
+
+#include <fstream>
+
+#include "src/utils/ilogger.h"
+#include "src/utils/mkldnn_blocklist.h"
+#include "src/utils/utility.h"
+
+PaddleInfer::PaddleInfer(const std::string &model_name,
+                         const std::string &model_dir,
+                         const std::string &model_file_prefix,
+                         const PaddlePredictorOption &option)
+    : model_name_(model_name), model_dir_(model_dir),
+      model_file_prefix_(model_file_prefix), option_(option) {
+  auto result = Create();
+  if (!result.ok()) {
+    INFOE("Create predictor failed: %s", result.status().ToString().c_str());
+    exit(-1);
+  }
+
+  predictor_ = std::move(result.value());
+  auto input_names = predictor_->GetInputNames();
+  for (const auto &name : input_names) {
+    auto handle = predictor_->GetInputHandle(name);
+    input_handles_.emplace_back(std::move(handle));
+  }
+  auto output_names = predictor_->GetOutputNames();
+  for (const auto &name : output_names) {
+    auto handle = predictor_->GetOutputHandle(name);
+    output_handles_.emplace_back(std::move(handle));
+  }
+}
+
+absl::StatusOr<std::shared_ptr<paddle_infer::Predictor>> PaddleInfer::Create() {
+  auto model_paths = Utility::GetModelPaths(model_dir_, model_file_prefix_);
+  if (!model_paths.ok()) {
+    return model_paths.status();
+  }
+  if (model_paths->find("paddle") == model_paths->end()) {
+    return absl::NotFoundError("No valid PaddlePaddle model found");
+  }
+
+  auto result_check = CheckRunMode();
+  if (!result_check.ok()) {
+    return result_check;
+  }
+
+  auto model_files = model_paths.value()["paddle"];
+  std::string model_file = model_files.first;
+  std::string params_file = model_files.second;
+
+  if (option_.DeviceType() == "cpu" && option_.DeviceId() >= 0) {
+    auto result_set = option_.SetDeviceId(0);
+    if (!result_set.ok()) {
+      return result_set;
+    }
+    INFO("`device_id` has been set to nullptr");
+  }
+
+  if (option_.DeviceType() == "gpu" && option_.DeviceId() < 0) {
+    auto result_device_id = option_.SetDeviceId(0);
+    if (!result_device_id.ok()) {
+      return result_device_id;
+    }
+    INFO("`device_id` has been set to 0");
+  }
+
+  paddle_infer::Config config;
+  config.SetModel(model_file, params_file);
+
+  if (option_.DeviceType() == "gpu") {
+    std::unordered_set<std::string> mixed_op_set = {"feed", "fetch"};
+    config.Exp_DisableMixedPrecisionOps(mixed_op_set);
+
+    paddle_infer::PrecisionType precision =
+        paddle_infer::PrecisionType::kFloat32;
+    if (option_.RunMode() == "paddle_fp16") {
+      precision = paddle_infer::PrecisionType::kHalf;
+    }
+
+    config.DisableMKLDNN();
+    config.EnableUseGpu(100, option_.DeviceId(), precision);
+    config.EnableNewIR(option_.EnableNewIR());
+    if (option_.EnableNewIR() && option_.EnableCinn()) {
+      config.EnableCINN();
+    }
+    config.EnableNewExecutor();
+    config.SetOptimizationLevel(3);
+  } else if (option_.DeviceType() == "cpu") {
+    config.DisableGpu();
+    if (option_.RunMode().find("mkldnn") != std::string::npos) {
+      config.EnableMKLDNN();
+      if (option_.RunMode().find("bf16") != std::string::npos) {
+        config.EnableMkldnnBfloat16();
+      }
+      config.SetMkldnnCacheCapacity(option_.MkldnnCacheCapacity());
+    } else {
+      config.DisableMKLDNN();
+    }
+    config.SetCpuMathLibraryNumThreads(option_.CpuThreads());
+    config.EnableNewIR(option_.EnableNewIR());
+    config.EnableNewExecutor();
+    config.SetOptimizationLevel(3);
+  } else {
+    return absl::InvalidArgumentError("Not supported device type: " +
+                                      option_.DeviceType());
+  }
+
+  config.EnableMemoryOptim();
+  for (const auto &del_p : option_.DeletePass()) {
+    config.DeletePass(del_p);
+  }
+  config.DisableGlogInfo();
+
+  auto predictor_shared = paddle_infer::CreatePredictor(config);
+
+  return predictor_shared;
+};
+
+absl::StatusOr<std::vector<cv::Mat>>
+PaddleInfer::Apply(const std::vector<cv::Mat> &x) {
+  for (size_t i = 0; i < x.size(); ++i) {
+    auto &input_handle = input_handles_[i];
+    std::vector<int> input_shape(x[0].dims);
+    for (int i = 0; i < x[0].dims; i++) {
+      input_shape[i] = x[0].size[i];
+    }
+    input_handle->Reshape(input_shape);
+    input_handle->CopyFromCpu<float>((float *)x[i].data);
+  }
+  try {
+    predictor_->Run();
+  } catch (const std::exception &e) {
+    INFOE("static Infer fail: %s", e.what());
+    exit(-1);
+  }
+
+  std::vector<std::vector<float>> outputs;
+  std::vector<int> output_shape = {};
+  for (auto &output_handle : output_handles_) {
+    output_shape = output_handle->shape();
+    size_t numel = 1;
+    for (auto dim : output_shape)
+      numel *= dim;
+    std::vector<float> out_data(numel);
+    output_handle->CopyToCpu(out_data.data());
+    outputs.push_back(std::move(out_data));
+  }
+  auto size_v = outputs[0].size();
+  cv::Mat pred(output_shape.size(), output_shape.data(), CV_32F);
+  memcpy(pred.ptr<float>(), outputs[0].data(),
+         outputs[0].size() * sizeof(float));
+  std::vector<cv::Mat> pred_outputs = {pred};
+  return pred_outputs;
+};
+
+absl::Status PaddleInfer::CheckRunMode() {
+  if (option_.RunMode().rfind("mkldnn", 0) == 0 &&
+      Mkldnn::MKLDNN_BLOCKLIST.count(model_name_) > 0 &&
+      option_.DeviceType() == "cpu") {
+    INFOW("The model %s is not supported to run in MKLDNN mode! Using `paddle` "
+          "instead!",
+          model_name_.c_str());
+
+    auto result = option_.SetRunMode("paddle");
+    if (!result.ok()) {
+      return result;
+    }
+  }
+  if (model_name_ == "LaTeX_OCR_rec" && option_.DeviceType() == "cpu") {
+    if (Utility::IsMkldnnAvailable() && option_.RunMode() != "mkldnn") {
+      INFOE("Now, the `LaTeX_OCR_rec` model only support `mkldnn` mode when "
+            "running on Intel CPU devices. So using `mkldnn` instead.");
+      exit(-1);
+      auto result = option_.SetRunMode("mkldnn");
+      if (!result.ok()) {
+        return result;
+      }
+    }
+  }
+
+  return absl::OkStatus();
+};
diff --git a/deploy/cpp_infer/src/common/static_infer.h b/deploy/cpp_infer/src/common/static_infer.h
new file mode 100644
index 0000000..8b94c4b
--- /dev/null
+++ b/deploy/cpp_infer/src/common/static_infer.h
@@ -0,0 +1,49 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <opencv2/opencv.hpp>
+#include <string>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+#include "paddle_inference_api.h"
+#include "src/utils/ilogger.h"
+#include "src/utils/pp_option.h"
+class PaddleInfer {
+public:
+  explicit PaddleInfer(const std::string &model_name,
+                       const std::string &model_dir,
+                       const std::string &model_file_prefix,
+                       const PaddlePredictorOption &option);
+  ~PaddleInfer() = default;
+  absl::StatusOr<std::vector<cv::Mat>>
+  Apply(const std::vector<cv::Mat> &x); //***********
+
+private:
+  std::string model_dir_;
+  std::string model_file_prefix_;
+  std::string model_name_;
+  PaddlePredictorOption option_;
+  std::shared_ptr<paddle_infer::Predictor> predictor_;
+
+  std::vector<std::unique_ptr<paddle_infer::Tensor>> input_handles_;
+  std::vector<std::unique_ptr<paddle_infer::Tensor>> output_handles_;
+
+  absl::StatusOr<std::shared_ptr<paddle_infer::Predictor>> Create();
+
+  absl::Status CheckRunMode();
+};
diff --git a/deploy/cpp_infer/src/common/thread_pool.cc b/deploy/cpp_infer/src/common/thread_pool.cc
new file mode 100644
index 0000000..1798d7e
--- /dev/null
+++ b/deploy/cpp_infer/src/common/thread_pool.cc
@@ -0,0 +1,88 @@
+
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+#include "thread_pool.h"
+
+namespace PaddlePool {
+
+constexpr size_t ThreadPool::WAIT_SECONDS;
+
+ThreadPool::ThreadPool() : ThreadPool(Thread::hardware_concurrency()) {}
+
+ThreadPool::ThreadPool(size_t maxThreads)
+    : quit_(false), currentThreads_(0), idleThreads_(0),
+      maxThreads_(maxThreads) {}
+
+ThreadPool::~ThreadPool() {
+  {
+    MutexGuard guard(mutex_);
+    quit_ = true;
+  }
+  cv_.notify_all();
+
+  for (auto &elem : threads_) {
+    assert(elem.second.joinable());
+    elem.second.join();
+  }
+}
+
+size_t ThreadPool::threadsNum() const {
+  MutexGuard guard(mutex_);
+  return currentThreads_;
+}
+
+void ThreadPool::worker() {
+  while (true) {
+    Task task;
+    {
+      UniqueLock uniqueLock(mutex_);
+      ++idleThreads_;
+      auto hasTimedout =
+          !cv_.wait_for(uniqueLock, std::chrono::seconds(WAIT_SECONDS),
+                        [this]() { return quit_ || !tasks_.empty(); });
+      --idleThreads_;
+      if (tasks_.empty()) {
+        if (quit_) {
+          --currentThreads_;
+          return;
+        }
+        if (hasTimedout) {
+          --currentThreads_;
+          joinFinishedThreads();
+          finishedThreadIDs_.emplace(std::this_thread::get_id());
+          return;
+        }
+      }
+      task = std::move(tasks_.front());
+      tasks_.pop();
+    }
+    task();
+  }
+}
+
+void ThreadPool::joinFinishedThreads() {
+  while (!finishedThreadIDs_.empty()) {
+    auto id = std::move(finishedThreadIDs_.front());
+    finishedThreadIDs_.pop();
+    auto iter = threads_.find(id);
+
+    assert(iter != threads_.end());
+    assert(iter->second.joinable());
+
+    iter->second.join();
+    threads_.erase(iter);
+  }
+}
+
+} // namespace PaddlePool
diff --git a/deploy/cpp_infer/src/common/thread_pool.h b/deploy/cpp_infer/src/common/thread_pool.h
new file mode 100644
index 0000000..084154d
--- /dev/null
+++ b/deploy/cpp_infer/src/common/thread_pool.h
@@ -0,0 +1,99 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+#pragma once
+
+#include <cassert>
+#include <condition_variable>
+#include <functional>
+#include <future>
+#include <memory>
+#include <mutex>
+#include <queue>
+#include <thread>
+#include <unordered_map>
+
+namespace PaddlePool {
+
+class ThreadPool {
+public:
+  using MutexGuard = std::lock_guard<std::mutex>;
+  using UniqueLock = std::unique_lock<std::mutex>;
+  using Thread = std::thread;
+  using ThreadID = std::thread::id;
+  using Task = std::function<void()>;
+
+  ThreadPool();
+  explicit ThreadPool(size_t maxThreads);
+
+  ThreadPool(const ThreadPool &) = delete;
+  ThreadPool &operator=(const ThreadPool &) = delete;
+
+  ~ThreadPool();
+
+  template <typename Func, typename... Ts>
+  auto submit(Func &&func, Ts &&...params)
+      -> std::future<typename std::result_of<Func(Ts...)>::type>;
+
+  size_t threadsNum() const;
+
+private:
+  static constexpr size_t WAIT_SECONDS = 2;
+  void worker();
+  void joinFinishedThreads();
+
+  bool quit_;
+  size_t currentThreads_;
+  size_t idleThreads_;
+  size_t maxThreads_;
+
+  mutable std::mutex mutex_;
+  std::condition_variable cv_;
+  std::queue<Task> tasks_;
+  std::queue<ThreadID> finishedThreadIDs_;
+  std::unordered_map<ThreadID, Thread> threads_;
+};
+
+} // namespace PaddlePool
+
+namespace PaddlePool {
+
+template <typename Func, typename... Ts>
+auto ThreadPool::submit(Func &&func, Ts &&...params)
+    -> std::future<typename std::result_of<Func(Ts...)>::type> {
+  auto execute =
+      std::bind(std::forward<Func>(func), std::forward<Ts>(params)...);
+
+  using ReturnType = typename std::result_of<Func(Ts...)>::type;
+  using PackagedTask = std::packaged_task<ReturnType()>;
+
+  auto task = std::make_shared<PackagedTask>(std::move(execute));
+  auto result = task->get_future();
+
+  MutexGuard guard(mutex_);
+  assert(!quit_);
+
+  tasks_.emplace([task]() { (*task)(); });
+  if (idleThreads_ > 0) {
+    cv_.notify_one();
+  } else if (currentThreads_ < maxThreads_) {
+    Thread t(&ThreadPool::worker, this);
+    assert(threads_.find(t.get_id()) == threads_.end());
+    threads_[t.get_id()] = std::move(t);
+    ++currentThreads_;
+  }
+
+  return result;
+}
+
+} // namespace PaddlePool
diff --git a/deploy/cpp_infer/src/configs/OCR.yaml b/deploy/cpp_infer/src/configs/OCR.yaml
new file mode 100644
index 0000000..237c131
--- /dev/null
+++ b/deploy/cpp_infer/src/configs/OCR.yaml
@@ -0,0 +1,45 @@
+
+pipeline_name: OCR
+
+text_type: general
+
+use_doc_preprocessor: True
+use_textline_orientation: True
+
+SubPipelines:
+  DocPreprocessor:
+    pipeline_name: doc_preprocessor
+    use_doc_orientation_classify: True
+    use_doc_unwarping: True
+    SubModules:
+      DocOrientationClassify:
+        module_name: doc_text_orientation
+        model_name: PP-LCNet_x1_0_doc_ori
+        model_dir: null
+      DocUnwarping:
+        module_name: image_unwarping
+        model_name: UVDoc
+        model_dir: null
+
+SubModules:
+  TextDetection:
+    module_name: text_detection
+    model_name: PP-OCRv6_medium_det
+    model_dir: null
+    limit_side_len: 64
+    limit_type: min
+    max_side_limit: 4000
+    thresh: 0.3
+    box_thresh: 0.6
+    unclip_ratio: 1.5
+  TextLineOrientation:
+    module_name: textline_orientation
+    model_name: PP-LCNet_x1_0_textline_ori 
+    model_dir: null
+    batch_size: 6    
+  TextRecognition:
+    module_name: text_recognition
+    model_name: PP-OCRv6_medium_rec
+    model_dir: null
+    batch_size: 6
+    score_thresh: 0.0
diff --git a/deploy/cpp_infer/src/configs/doc_preprocessor.yaml b/deploy/cpp_infer/src/configs/doc_preprocessor.yaml
new file mode 100644
index 0000000..0222bf7
--- /dev/null
+++ b/deploy/cpp_infer/src/configs/doc_preprocessor.yaml
@@ -0,0 +1,15 @@
+
+pipeline_name: doc_preprocessor
+
+use_doc_orientation_classify: True
+use_doc_unwarping: True
+batch_size: 1
+SubModules:
+  DocOrientationClassify:
+    module_name: doc_text_orientation
+    model_name: PP-LCNet_x1_0_doc_ori
+    model_dir: null
+  DocUnwarping:
+    module_name: image_unwarping
+    model_name: UVDoc
+    model_dir: null
diff --git a/deploy/cpp_infer/src/modules/image_classification/predictor.cc b/deploy/cpp_infer/src/modules/image_classification/predictor.cc
new file mode 100644
index 0000000..a51e1fb
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/image_classification/predictor.cc
@@ -0,0 +1,158 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "predictor.h"
+
+#include "result.h"
+#include "src/common/image_batch_sampler.h"
+#include "src/utils/ilogger.h"
+
+ClasPredictor::ClasPredictor(const ClasPredictorParams &params)
+    : BasePredictor(params.model_dir, params.model_name, params.device,
+                    params.precision, params.enable_mkldnn,
+                    params.mkldnn_cache_capacity, params.cpu_threads,
+                    params.batch_size, "image"),
+      params_(params) {
+  auto status = Build();
+  if (!status.ok()) {
+    INFOE("Build fail: %s", status.ToString().c_str());
+    exit(-1);
+  }
+};
+
+absl::Status ClasPredictor::BuildResize() {
+  const auto &pre_params = config_.PreProcessOpInfo();
+  if (pre_params.find("ResizeImage.size") != pre_params.end()) {
+    Register<Resize>("Resize", YamlConfig::SmartParseVector(
+                                   pre_params.at("ResizeImage.size"))
+                                   .vec_int);
+  } else if (pre_params.find("ResizeImage.resize_short") != pre_params.end()) {
+    Register<ResizeByShort>(
+        "Resize", std::stoi(pre_params.at("ResizeImage.resize_short")));
+  } else {
+    return absl::NotFoundError("Resize must be provide param !");
+  }
+  return absl::OkStatus();
+}
+
+absl::Status ClasPredictor::Build() {
+  const auto &pre_params = config_.PreProcessOpInfo();
+  Register<ReadImage>("Read");
+
+  auto status = BuildResize();
+  if (!status.ok()) {
+    return absl::InternalError("build resize_op fail: " + status.ToString());
+  }
+  if (config_.FindKey("Crop").ok()) {
+    Register<Crop>("Crop",
+                   YamlConfig::SmartParseVector(pre_params.at("CropImage.size"))
+                       .vec_int); //*************
+  }
+  Register<NormalizeImage>(
+      "Normalize", std::stof(pre_params.at("NormalizeImage.scale")),
+      YamlConfig::SmartParseVector(pre_params.at("NormalizeImage.mean"))
+          .vec_float,
+      YamlConfig::SmartParseVector(pre_params.at("NormalizeImage.std"))
+          .vec_float);
+  Register<ToCHWImage>("ToCHW");
+  Register<ToBatch>("ToBatch");
+
+  infer_ptr_ = CreateStaticInfer();
+  const auto &post_params = config_.PostProcessOpInfo();
+  auto gsj = YamlConfig::SmartParseVector(
+      post_params.at("PostProcess.Topk.label_list"));
+  post_op_["Topk"] = std::unique_ptr<Topk>(
+      new Topk(YamlConfig::SmartParseVector(
+                   post_params.at("PostProcess.Topk.label_list"))
+                   .vec_string,
+               std::stoi(post_params.at("PostProcess.Topk.topk"))));
+  return absl::OkStatus();
+};
+
+std::vector<std::unique_ptr<BaseCVResult>>
+ClasPredictor::Process(std::vector<cv::Mat> &batch_data) {
+  std::vector<cv::Mat> origin_image = {};
+  origin_image.reserve(batch_data.size());
+  for (const auto &mat : batch_data) {
+    origin_image.push_back(mat.clone());
+  }
+  auto batch_read = pre_op_.at("Read")->Apply(batch_data);
+
+  if (!batch_read.ok()) {
+    INFOE(batch_read.status().ToString().c_str());
+    exit(-1);
+  }
+
+  auto batch_resize = pre_op_.at("Resize")->Apply(batch_read.value());
+  if (!batch_resize.ok()) {
+    INFOE(batch_resize.status().ToString().c_str());
+    exit(-1);
+  }
+  if (config_.FindKey("Crop").ok()) {
+    batch_resize = pre_op_.at("Crop")->Apply(batch_resize.value()); // **
+    if (!batch_resize.ok()) {
+      INFOE(batch_resize.status().ToString().c_str());
+      exit(-1);
+    }
+  }
+  auto batch_normalize = pre_op_.at("Normalize")->Apply(batch_resize.value());
+  if (!batch_normalize.ok()) {
+    INFOE(batch_normalize.status().ToString().c_str());
+    exit(-1);
+  }
+
+  auto batch_tochw = pre_op_.at("ToCHW")->Apply(batch_normalize.value());
+  if (!batch_tochw.ok()) {
+    INFOE(batch_tochw.status().ToString().c_str());
+    exit(-1);
+  }
+
+  auto batch_tobatch = pre_op_.at("ToBatch")->Apply(batch_tochw.value());
+  if (!batch_tobatch.ok()) {
+    INFOE(batch_tobatch.status().ToString().c_str());
+    exit(-1);
+  }
+
+  auto batch_infer = infer_ptr_->Apply(batch_tobatch.value());
+  if (!batch_infer.ok()) {
+    INFOE(batch_infer.status().ToString().c_str());
+    exit(-1);
+  }
+
+  auto cls_result = post_op_.at("Topk")->Apply(batch_infer.value()[0]);
+
+  if (!cls_result.ok()) {
+    INFOE(cls_result.status().ToString().c_str());
+    exit(-1);
+  }
+
+  std::vector<std::unique_ptr<BaseCVResult>> base_cv_result_ptr_vec = {};
+  for (int i = 0; i < cls_result.value().size(); i++, input_index_++) {
+    ClasPredictorResult predictor_result;
+    if (!input_path_.empty()) {
+      if (input_index_ == input_path_.size())
+        input_index_ = 0;
+      predictor_result.input_path = input_path_[input_index_];
+    }
+    predictor_result.input_image = origin_image[i];
+    predictor_result.class_ids = cls_result.value()[i].class_ids;
+    predictor_result.scores = cls_result.value()[i].scores;
+    predictor_result.label_names = cls_result.value()[i].label_names;
+    predictor_result_vec_.push_back(predictor_result);
+    base_cv_result_ptr_vec.push_back(
+        std::unique_ptr<BaseCVResult>(new TopkResult(predictor_result)));
+  }
+
+  return base_cv_result_ptr_vec;
+}
diff --git a/deploy/cpp_infer/src/modules/image_classification/predictor.h b/deploy/cpp_infer/src/modules/image_classification/predictor.h
new file mode 100644
index 0000000..0036863
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/image_classification/predictor.h
@@ -0,0 +1,67 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include "processors.h"
+#include "src/base/base_batch_sampler.h"
+#include "src/base/base_cv_result.h"
+#include "src/base/base_predictor.h"
+#include "src/common/processors.h"
+
+struct ClasPredictorParams {
+  absl::optional<std::string> model_name = absl::nullopt;
+  absl::optional<std::string> model_dir = absl::nullopt;
+  absl::optional<std::string> device = absl::nullopt;
+  std::string precision = "fp32";
+  bool enable_mkldnn = true;
+  int mkldnn_cache_capacity = 10;
+  int cpu_threads = 8;
+  int batch_size = 1;
+};
+
+struct ClasPredictorResult {
+  std::string input_path = "";
+  cv::Mat input_image;
+  std::vector<int> class_ids;
+  std::vector<float> scores;
+  std::vector<std::string> label_names;
+};
+
+class ClasPredictor : public BasePredictor {
+public:
+  explicit ClasPredictor(const ClasPredictorParams &params);
+
+  ClasPredictor() = delete;
+
+  absl::Status Build();
+
+  std::vector<std::unique_ptr<BaseCVResult>>
+  Process(std::vector<cv::Mat> &batch_data) override;
+
+  std::vector<ClasPredictorResult> PredictorResult() const {
+    return predictor_result_vec_;
+  };
+
+  void ResetResult() override { predictor_result_vec_.clear(); };
+
+  absl::Status BuildResize();
+
+private:
+  ClasPredictorParams params_;
+  std::unordered_map<std::string, std::unique_ptr<Topk>> post_op_;
+  std::vector<ClasPredictorResult> predictor_result_vec_;
+  std::unique_ptr<PaddleInfer> infer_ptr_;
+  int input_index_ = 0;
+};
diff --git a/deploy/cpp_infer/src/modules/image_classification/processors.cc b/deploy/cpp_infer/src/modules/image_classification/processors.cc
new file mode 100644
index 0000000..793e24f
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/image_classification/processors.cc
@@ -0,0 +1,133 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "processors.h"
+
+#include <algorithm>
+#include <stdexcept>
+
+#include "processors.h"
+#include "src/utils/utility.h"
+
+Crop::Crop(const std::vector<int> crop_size, const std::string &mode)
+    : mode_(mode) {
+  if (crop_size.size() == 1) {
+    crop_size_ = std::vector<int>(2, crop_size[0]);
+  } else {
+    crop_size_ = crop_size;
+  }
+  assert(mode_ == "Center" || mode_ == "TopLeft");
+  assert(crop_size_.size() == 2 && crop_size_[0] > 0 && crop_size_[1] > 0);
+}
+
+Crop::Crop(const int crop_size, const std::string &mode)
+    : crop_size_(2, crop_size), mode_(mode) {
+  assert(mode_ == "Center" || mode_ == "TopLeft");
+  assert(crop_size_.size() == 2 && crop_size_[0] > 0 && crop_size_[1] > 0);
+}
+
+absl::StatusOr<cv::Mat> Crop::CropImage(const cv::Mat &img) const {
+  int h = img.rows;
+  int w = img.cols;
+  int crop_width = crop_size_[0];
+  int crop_height = crop_size_[1];
+
+  if (w < crop_width || h < crop_height) {
+    return absl::InvalidArgumentError(
+        "Input image (" + std::to_string(w) + ", " + std::to_string(h) +
+        ") smaller than target size (" + std::to_string(crop_width) + ", " +
+        std::to_string(crop_height) + ").");
+  }
+  int x1 = 0, y1 = 0;
+  if (mode_ == "Center") {
+    x1 = std::max(0, (w - crop_width) / 2);
+    y1 = std::max(0, (h - crop_height) / 2);
+  } else if (mode_ == "TopLeft") {
+    x1 = 0;
+    y1 = 0;
+  } else {
+    return absl::InvalidArgumentError("Unsupported crop mode.");
+  }
+
+  cv::Rect roi(x1, y1, crop_width, crop_height);
+  return img(roi).clone();
+}
+
+absl::StatusOr<std::vector<cv::Mat>> Crop::Apply(std::vector<cv::Mat> &imgs,
+                                                 const void *param) const {
+  std::vector<cv::Mat> result;
+  result.reserve(imgs.size());
+  for (const auto &img : imgs) {
+    auto cropped = CropImage(img);
+    if (!cropped.ok())
+      return cropped.status();
+    result.push_back(cropped.value());
+  }
+  return result;
+}
+
+Topk::Topk(const std::vector<std::string> &class_names, const int topk)
+    : class_names_(class_names), topk_(topk) {}
+
+absl::StatusOr<std::vector<Topk::TopkOutput>> Topk::Apply(const cv::Mat &preds,
+                                                          const int topk) {
+  topk_ = topk;
+  auto preds_batch = Utility::SplitBatch(preds);
+  if (!preds_batch.ok()) {
+    return preds_batch.status();
+  }
+  std::vector<TopkOutput> topk_results = {};
+  topk_results.reserve(preds_batch.value().size());
+  for (const auto &pred : preds_batch.value()) {
+    auto topk_result = Process(pred);
+    if (!topk_result.ok()) {
+      return topk_result.status();
+    }
+    topk_results.push_back(topk_result.value());
+  }
+  return topk_results;
+}
+
+absl::StatusOr<Topk::TopkOutput> Topk::Process(const cv::Mat &pred) const {
+  if (pred.dims != 2 || pred.type() != CV_32F) {
+    return absl::InvalidArgumentError("Input scores must be 2-D float matrix.");
+  }
+
+  TopkOutput topk_result(pred.size[0]);
+  int num_classes = pred.size[1];
+  const float *row = pred.ptr<float>();
+
+  std::vector<std::pair<float, int>> score_idx;
+  for (int j = 0; j < num_classes; ++j) {
+    score_idx.emplace_back(row[j], j);
+  }
+  std::partial_sort(
+      score_idx.begin(), score_idx.begin() + topk_, score_idx.end(),
+      [](const std::pair<float, int> &a, const std::pair<float, int> &b) {
+        return a.first > b.first;
+      });
+
+  for (int t = 0; t < topk_; ++t) {
+    topk_result.class_ids.push_back(score_idx[t].second);
+    topk_result.scores.push_back(score_idx[t].first);
+    if (!class_names_.empty() &&
+        score_idx[t].second < (int)class_names_.size()) {
+      topk_result.label_names.push_back(class_names_[score_idx[t].second]);
+    } else {
+      topk_result.label_names.push_back(std::to_string(score_idx[t].second));
+    }
+  }
+
+  return topk_result;
+}
diff --git a/deploy/cpp_infer/src/modules/image_classification/processors.h b/deploy/cpp_infer/src/modules/image_classification/processors.h
new file mode 100644
index 0000000..ba2f8d4
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/image_classification/processors.h
@@ -0,0 +1,63 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <opencv2/opencv.hpp>
+#include <string>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+#include "src/utils/func_register.h"
+
+class Crop : public BaseProcessor {
+public:
+  explicit Crop(const std::vector<int> crop_size,
+                const std::string &mode = "Center");
+  explicit Crop(const int crop_size, const std::string &mode = "Center");
+  absl::StatusOr<cv::Mat> CropImage(const cv::Mat &img) const;
+  absl::StatusOr<std::vector<cv::Mat>>
+  Apply(std::vector<cv::Mat> &input,
+        const void *param = nullptr) const override;
+
+private:
+  std::vector<int> crop_size_;
+  std::string mode_;
+};
+
+class Topk {
+public:
+  struct TopkOutput {
+    TopkOutput(int batch) {
+      class_ids.reserve(batch);
+      scores.reserve(batch);
+      label_names.reserve(batch);
+    }
+    std::vector<int> class_ids;
+    std::vector<float> scores;
+    std::vector<std::string> label_names;
+  };
+  explicit Topk(
+      const std::vector<std::string> &class_names = std::vector<std::string>(),
+      int topk = 1);
+
+  absl::StatusOr<TopkOutput> Process(const cv::Mat &pred_data) const;
+  absl::StatusOr<std::vector<TopkOutput>> Apply(const cv::Mat &preds,
+                                                const int topk = 1);
+
+private:
+  std::vector<std::string> class_names_;
+  int topk_;
+};
diff --git a/deploy/cpp_infer/src/modules/image_classification/result.cc b/deploy/cpp_infer/src/modules/image_classification/result.cc
new file mode 100644
index 0000000..8262090
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/image_classification/result.cc
@@ -0,0 +1,160 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "result.h"
+
+#include <fstream>
+#include <string>
+
+#include "src/utils/utility.h"
+#include "third_party/nlohmann/json.hpp"
+
+using json = nlohmann::json;
+void TopkResult::SaveToImg(const std::string &save_path) {
+  cv::Mat img = predictor_result_.input_image.clone();
+
+  std::ostringstream oss;
+  oss << predictor_result_.label_names[0] << " " << std::fixed
+      << std::setprecision(2) << predictor_result_.scores[0];
+  std::string label_str = oss.str();
+
+  int imgWidth = img.cols;
+  int minFont = std::max(12, imgWidth * 2 / 100);
+  int maxFont = std::max(18, imgWidth * 5 / 100);
+  int baseline = 0;
+  int fontFace = 0;
+  double fontScale = getAdaptiveFontScale(
+      label_str, imgWidth, imgWidth, minFont, maxFont, 2, baseline, fontFace);
+
+  cv::Size textSize =
+      cv::getTextSize(label_str, fontFace, fontScale, 2, &baseline);
+
+  int rect_left = 3, rect_top = 3;
+  int rect_right = rect_left + textSize.width + 6;
+  int rect_bottom = rect_top + textSize.height + 6;
+  cv::Scalar bgColor(0, 0, 255);
+  cv::rectangle(img, cv::Point(rect_left, rect_top),
+                cv::Point(rect_right, rect_bottom), bgColor, cv::FILLED);
+
+  int text_x = rect_left + 3;
+  int text_y = rect_top + textSize.height + 2;
+  cv::Scalar fontColor(255, 255, 255);
+  cv::putText(img, label_str, cv::Point(text_x, text_y), fontFace, fontScale,
+              fontColor, 2, cv::LINE_AA);
+  absl::StatusOr<std::string> full_path;
+  if (predictor_result_.input_path.empty()) {
+    auto now = std::chrono::system_clock::now();
+    auto now_time = std::chrono::system_clock::to_time_t(now);
+    std::stringstream ss;
+    ss << "output_" << std::put_time(std::localtime(&now_time), "%Y%m%d_%H%M%S")
+       << ".jpg";
+    std::string timestamp_filename = ss.str();
+    INFOW("Input path is empty, will use %s instead!",
+          timestamp_filename.c_str());
+    predictor_result_.input_path = timestamp_filename;
+    full_path =
+        Utility::SmartCreateDirectoryForImage(save_path, timestamp_filename);
+  } else {
+    full_path = Utility::SmartCreateDirectoryForImage(
+        save_path, predictor_result_.input_path);
+  }
+  if (!full_path.ok()) {
+    INFOE(full_path.status().ToString().c_str());
+    exit(-1);
+  }
+  bool success = cv::imwrite(full_path.value(), img);
+  if (!success) {
+    INFOE("Failed to write the image %s ", full_path.value().c_str());
+    exit(-1);
+  }
+}
+
+void TopkResult::Print() const {
+  std::cout << "{\n  \"res\": {" << std::endl;
+
+  std::cout << "    \"input_path\": {" << predictor_result_.input_path << "},"
+            << std::endl;
+  std::cout << "    \"class_ids\": {" << predictor_result_.class_ids[0] << "},"
+            << std::endl;
+  std::cout << "    \"scores\": {" << predictor_result_.scores[0] << "},"
+            << std::endl;
+  std::cout << "    \"label_names\": {" << predictor_result_.label_names[0]
+            << "}," << std::endl;
+  std::cout << "}" << std::endl;
+}
+
+void TopkResult::SaveToJson(const std::string &save_path) const {
+  nlohmann::ordered_json j;
+
+  j["input_path"] = predictor_result_.input_path;
+  j["page_index"] = nullptr; //********
+
+  json class_ids = json::array();
+  for (const auto &item : predictor_result_.class_ids) {
+    class_ids.push_back(item);
+  }
+  json scores = json::array();
+  for (const auto &item : predictor_result_.scores) {
+    scores.push_back(item);
+  }
+  json label_names = json::array();
+  for (const auto &item : predictor_result_.label_names) {
+    label_names.push_back(item);
+  }
+
+  j["class_ids"] = class_ids;
+  j["scores"] = scores;
+  j["label_names"] = label_names;
+
+  auto full_path = Utility::SmartCreateDirectoryForJson(
+      save_path, predictor_result_.input_path);
+  if (!full_path.ok()) {
+    INFOE(full_path.status().ToString().c_str());
+    exit(-1);
+  }
+  std::ofstream file(full_path.value());
+  if (file.is_open()) {
+    file << j.dump(4);
+    file.close();
+  } else {
+    INFOE("Could not open file for writing: %s", save_path.c_str());
+    exit(-1);
+  }
+}
+
+int TopkResult::getAdaptiveFontScale(const std::string &text, int imgWidth,
+                                     int maxWidth, int minFont, int maxFont,
+                                     int thickness, int &outBaseline,
+                                     int &outFontFace) {
+  int fontFace = cv::FONT_HERSHEY_SIMPLEX;
+  double fontScale = 1.0;
+  int baseline = 0;
+  int bestFontSize = minFont;
+
+  for (int fontSize = maxFont; fontSize >= minFont; --fontSize) {
+    fontScale = fontSize / 20.0; // 20为基准比例，可调
+    int base;
+    cv::Size textSize =
+        cv::getTextSize(text, fontFace, fontScale, thickness, &base);
+    if (textSize.width <= maxWidth) {
+      bestFontSize = fontSize;
+      outBaseline = base;
+      outFontFace = fontFace;
+      return fontScale;
+    }
+  }
+  outBaseline = 0;
+  outFontFace = fontFace;
+  return minFont / 20.0;
+}
diff --git a/deploy/cpp_infer/src/modules/image_classification/result.h b/deploy/cpp_infer/src/modules/image_classification/result.h
new file mode 100644
index 0000000..0adba30
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/image_classification/result.h
@@ -0,0 +1,38 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <opencv2/opencv.hpp>
+#include <vector>
+
+#include "predictor.h"
+#include "src/base/base_cv_result.h"
+
+class TopkResult : public BaseCVResult {
+public:
+  TopkResult(ClasPredictorResult predictor_result)
+      : BaseCVResult(), predictor_result_(predictor_result){};
+  // std::unordered_map<std::string, cv::Mat> ToImg() const override;
+  void SaveToImg(const std::string &save_path) override;
+  void Print() const override;
+  void SaveToJson(const std::string &save_path) const override;
+  static int getAdaptiveFontScale(const std::string &text, int imgWidth,
+                                  int maxWidth, int minFont, int maxFont,
+                                  int thickness, int &outBaseline,
+                                  int &outFontFace);
+
+private:
+  ClasPredictorResult predictor_result_;
+};
diff --git a/deploy/cpp_infer/src/modules/image_unwarping/predictor.cc b/deploy/cpp_infer/src/modules/image_unwarping/predictor.cc
new file mode 100644
index 0000000..62c6ab3
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/image_unwarping/predictor.cc
@@ -0,0 +1,100 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "predictor.h"
+
+#include "result.h"
+#include "src/common/image_batch_sampler.h"
+
+WarpPredictor::WarpPredictor(const WarpPredictorParams &params)
+    : BasePredictor(params.model_dir, params.model_name, params.device,
+                    params.precision, params.enable_mkldnn,
+                    params.mkldnn_cache_capacity, params.cpu_threads,
+                    params.batch_size, "image"),
+      params_(params) {
+  auto status = Build();
+  if (!status.ok()) {
+    INFOE("Build fail: %s", status.ToString().c_str());
+    exit(-1);
+  }
+};
+
+absl::Status WarpPredictor::Build() {
+  const auto &pre_params = config_.PreProcessOpInfo();
+  Register<ReadImage>("Read", "BGR");
+  Register<Normalize>("Normalize", 1.0 / 255.0, 0.0, 1.0);
+  Register<ToCHWImage>("ToCHW");
+  Register<ToBatch>("ToBatch");
+
+  infer_ptr_ = CreateStaticInfer();
+  const auto &post_params = config_.PostProcessOpInfo();
+  post_op_["DocTr"] = std::unique_ptr<DocTrPostProcess>(new DocTrPostProcess());
+  return absl::OkStatus();
+};
+
+std::vector<std::unique_ptr<BaseCVResult>>
+WarpPredictor::Process(std::vector<cv::Mat> &batch_data) {
+  std::vector<cv::Mat> origin_image = {};
+  origin_image.reserve(batch_data.size());
+  for (const auto &mat : batch_data) {
+    origin_image.push_back(mat.clone());
+  }
+  auto batch_read = pre_op_.at("Read")->Apply(batch_data);
+  if (!batch_read.ok()) {
+    INFOE(batch_read.status().ToString().c_str());
+    exit(-1);
+  }
+
+  auto batch_normalize = pre_op_.at("Normalize")->Apply(batch_read.value());
+  if (!batch_normalize.ok()) {
+    INFOE(batch_normalize.status().ToString().c_str());
+    exit(-1);
+  }
+  auto batch_tochw = pre_op_.at("ToCHW")->Apply(batch_normalize.value());
+  if (!batch_tochw.ok()) {
+    INFOE(batch_tochw.status().ToString().c_str());
+    exit(-1);
+  }
+  auto batch_tobatch = pre_op_.at("ToBatch")->Apply(batch_tochw.value());
+  if (!batch_tobatch.ok()) {
+    INFOE(batch_tobatch.status().ToString().c_str());
+    exit(-1);
+  }
+  auto batch_infer = infer_ptr_->Apply(batch_tobatch.value());
+  if (!batch_infer.ok()) {
+    INFOE(batch_infer.status().ToString().c_str());
+    exit(-1);
+  }
+  auto warp_result = post_op_.at("DocTr")->Apply(batch_infer.value()[0]);
+
+  if (!warp_result.ok()) {
+    INFOE(warp_result.status().ToString().c_str());
+    exit(-1);
+  }
+  std::vector<std::unique_ptr<BaseCVResult>> base_cv_result_ptr_vec = {};
+  for (int i = 0; i < warp_result.value().size(); i++, input_index_++) {
+    WarpPredictorResult predictor_result;
+    if (!input_path_.empty()) {
+      if (input_index_ == input_path_.size())
+        input_index_ = 0;
+      predictor_result.input_path = input_path_[input_index_];
+    }
+    predictor_result.input_image = origin_image[i];
+    predictor_result.doctr_img = warp_result.value()[i];
+    predictor_result_vec_.push_back(predictor_result);
+    base_cv_result_ptr_vec.push_back(
+        std::unique_ptr<BaseCVResult>(new DocTrResult(predictor_result)));
+  }
+  return base_cv_result_ptr_vec;
+}
diff --git a/deploy/cpp_infer/src/modules/image_unwarping/predictor.h b/deploy/cpp_infer/src/modules/image_unwarping/predictor.h
new file mode 100644
index 0000000..cfe3264
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/image_unwarping/predictor.h
@@ -0,0 +1,61 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include "processors.h"
+#include "src/base/base_batch_sampler.h"
+#include "src/base/base_cv_result.h"
+#include "src/base/base_predictor.h"
+#include "src/common/processors.h"
+
+struct WarpPredictorParams {
+  absl::optional<std::string> model_name = absl::nullopt;
+  absl::optional<std::string> model_dir = absl::nullopt;
+  absl::optional<std::string> device = absl::nullopt;
+  bool enable_mkldnn = true;
+  std::string precision = "fp32";
+  int mkldnn_cache_capacity = 10;
+  int cpu_threads = 8;
+  int batch_size = 1;
+};
+
+struct WarpPredictorResult {
+  std::string input_path = "";
+  cv::Mat input_image;
+  cv::Mat doctr_img;
+};
+
+class WarpPredictor : public BasePredictor {
+public:
+  explicit WarpPredictor(const WarpPredictorParams &params);
+
+  absl::Status Build();
+
+  std::vector<std::unique_ptr<BaseCVResult>>
+  Process(std::vector<cv::Mat> &batch_data) override;
+
+  std::vector<WarpPredictorResult> PredictorResult() const {
+    return predictor_result_vec_;
+  };
+
+  void ResetResult() override { predictor_result_vec_.clear(); };
+
+private:
+  std::unordered_map<std::string, std::unique_ptr<DocTrPostProcess>> post_op_;
+  std::vector<WarpPredictorResult> predictor_result_vec_;
+  std::unique_ptr<PaddleInfer> infer_ptr_;
+  WarpPredictorParams params_;
+  int input_index_ = 0;
+};
diff --git a/deploy/cpp_infer/src/modules/image_unwarping/processors.cc b/deploy/cpp_infer/src/modules/image_unwarping/processors.cc
new file mode 100644
index 0000000..14f0518
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/image_unwarping/processors.cc
@@ -0,0 +1,73 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "processors.h"
+
+#include <sstream>
+#include <stdexcept>
+
+#include "src/utils/utility.h"
+
+DocTrPostProcess::DocTrPostProcess(double scale) : scale_(scale) {}
+
+absl::StatusOr<std::vector<cv::Mat>>
+DocTrPostProcess::Apply(const cv::Mat &preds) const {
+  auto preds_batch = Utility::SplitBatch(preds);
+  if (!preds_batch.ok()) {
+    return preds_batch.status();
+  }
+  std::vector<cv::Mat> doc_out;
+  doc_out.reserve(preds_batch.value().size());
+  for (auto &pred_data : preds_batch.value()) {
+    auto result = Process(pred_data);
+    if (!result.ok()) {
+      return result.status();
+    }
+    doc_out.push_back(result.value());
+  }
+  return doc_out;
+}
+
+absl::StatusOr<cv::Mat> DocTrPostProcess::Process(cv::Mat &pred_data) const {
+  if (pred_data.dims != 4) {
+    return absl::InvalidArgumentError("must have 4D"); //********
+  }
+  std::vector<int> shape = {};
+  for (int i = 1; i < pred_data.dims; i++) {
+    shape.push_back(pred_data.size[i]);
+  }
+  pred_data = pred_data.reshape(1, shape);
+  std::vector<cv::Range> ranges(pred_data.size[0]);
+  std::vector<cv::Mat> mat_split(pred_data.size[0]);
+
+  for (int i = 0; i < pred_data.size[0]; i++) {
+    ranges[0] = cv::Range(i, i + 1);
+    for (int j = 1; j < pred_data.dims; j++) {
+      ranges[j] = cv::Range::all();
+    }
+    mat_split[i] = pred_data(&ranges[0]);
+  }
+  for (auto &item : mat_split) {
+    std::vector<int> shape_item = {};
+    for (int i = 1; i < item.dims; i++) {
+      shape_item.push_back(item.size[i]);
+    }
+    item = item.reshape(1, shape_item);
+    item = item * scale_;
+  }
+  cv::Mat out_hwc;
+  cv::merge(mat_split, out_hwc);
+  out_hwc.convertTo(out_hwc, CV_8U);
+  return out_hwc;
+}
diff --git a/deploy/cpp_infer/src/modules/image_unwarping/processors.h b/deploy/cpp_infer/src/modules/image_unwarping/processors.h
new file mode 100644
index 0000000..9f85853
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/image_unwarping/processors.h
@@ -0,0 +1,35 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <opencv2/opencv.hpp>
+#include <string>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+#include "src/utils/func_register.h"
+
+class DocTrPostProcess {
+public:
+  explicit DocTrPostProcess(double scale = 255.0f);
+
+  absl::StatusOr<std::vector<cv::Mat>> Apply(const cv::Mat &preds) const;
+
+  absl::StatusOr<cv::Mat> Process(cv::Mat &pred_data) const;
+
+private:
+  double scale_;
+};
diff --git a/deploy/cpp_infer/src/modules/image_unwarping/result.cc b/deploy/cpp_infer/src/modules/image_unwarping/result.cc
new file mode 100644
index 0000000..8284228
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/image_unwarping/result.cc
@@ -0,0 +1,119 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "result.h"
+
+#include <fstream>
+#include <string>
+
+#include "src/utils/utility.h"
+#include "third_party/nlohmann/json.hpp"
+
+using json = nlohmann::json;
+void DocTrResult::SaveToImg(const std::string &save_path) {
+  absl::StatusOr<std::string> full_path;
+  if (predictor_result_.input_path.empty()) {
+    auto now = std::chrono::system_clock::now();
+    auto now_time = std::chrono::system_clock::to_time_t(now);
+    std::stringstream ss;
+    ss << "output_" << std::put_time(std::localtime(&now_time), "%Y%m%d_%H%M%S")
+       << ".jpg";
+    std::string timestamp_filename = ss.str();
+    INFOW("Input path is empty, will use %s instead!",
+          timestamp_filename.c_str());
+    predictor_result_.input_path = timestamp_filename;
+    full_path =
+        Utility::SmartCreateDirectoryForImage(save_path, timestamp_filename);
+  } else {
+    full_path = Utility::SmartCreateDirectoryForImage(
+        save_path, predictor_result_.input_path);
+  }
+  bool success = cv::imwrite(full_path.value(), predictor_result_.doctr_img);
+  if (!success) {
+    INFOE("Failed to write the image %s", full_path.value().c_str());
+    exit(-1);
+  }
+}
+
+void DocTrResult::Print() const {
+  std::cout << "{\n  \"res\": {" << std::endl;
+
+  std::cout << "    \"input_path\": {" << predictor_result_.input_path << "},"
+            << std::endl;
+  std::cout << "    \"doctr_img\": {"
+            << "..."
+            << "}" << std::endl;
+  std::cout << "}" << std::endl;
+}
+
+void DocTrResult::SaveToJson(const std::string &save_path) const {
+  nlohmann::ordered_json j;
+
+  j["input_path"] = predictor_result_.input_path;
+  j["page_index"] = nullptr; //********
+
+  nlohmann::json mat_array = nlohmann::json::array();
+
+  for (int i = 0; i < predictor_result_.doctr_img.rows; ++i) {
+    nlohmann::json row = nlohmann::json::array();
+    for (int j = 0; j < predictor_result_.doctr_img.cols; ++j) {
+      cv::Vec3b color = predictor_result_.doctr_img.at<cv::Vec3b>(i, j);
+      row.push_back({color[0], color[1], color[2]});
+    }
+    mat_array.push_back(row);
+  }
+  j["doctr_img"] = mat_array;
+
+  auto full_path = Utility::SmartCreateDirectoryForJson(
+      save_path, predictor_result_.input_path);
+
+  if (!full_path.ok()) {
+    INFOE(full_path.status().ToString().c_str());
+    exit(-1);
+  }
+  std::ofstream file(full_path.value());
+  if (file.is_open()) {
+    file << j.dump(4);
+    file.close();
+  } else {
+    INFOE("Could not open file for writing: %s", save_path.c_str());
+    exit(-1);
+  }
+}
+
+int DocTrResult::getAdaptiveFontScale(const std::string &text, int imgWidth,
+                                      int maxWidth, int minFont, int maxFont,
+                                      int thickness, int &outBaseline,
+                                      int &outFontFace) {
+  int fontFace = cv::FONT_HERSHEY_SIMPLEX;
+  double fontScale = 1.0;
+  int baseline = 0;
+  int bestFontSize = minFont;
+
+  for (int fontSize = maxFont; fontSize >= minFont; --fontSize) {
+    fontScale = fontSize / 20.0;
+    int base;
+    cv::Size textSize =
+        cv::getTextSize(text, fontFace, fontScale, thickness, &base);
+    if (textSize.width <= maxWidth) {
+      bestFontSize = fontSize;
+      outBaseline = base;
+      outFontFace = fontFace;
+      return fontScale;
+    }
+  }
+  outBaseline = 0;
+  outFontFace = fontFace;
+  return minFont / 20.0;
+}
diff --git a/deploy/cpp_infer/src/modules/image_unwarping/result.h b/deploy/cpp_infer/src/modules/image_unwarping/result.h
new file mode 100644
index 0000000..fff93c2
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/image_unwarping/result.h
@@ -0,0 +1,38 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <opencv2/opencv.hpp>
+#include <vector>
+
+#include "predictor.h"
+#include "src/base/base_cv_result.h"
+
+class DocTrResult : public BaseCVResult {
+public:
+  DocTrResult(WarpPredictorResult predictor_result)
+      : BaseCVResult(), predictor_result_(predictor_result){};
+  // std::unordered_map<std::string, cv::Mat> ToImg() const override;
+  void SaveToImg(const std::string &save_path) override;
+  void Print() const override;
+  void SaveToJson(const std::string &save_path) const override;
+  static int getAdaptiveFontScale(const std::string &text, int imgWidth,
+                                  int maxWidth, int minFont, int maxFont,
+                                  int thickness, int &outBaseline,
+                                  int &outFontFace);
+
+private:
+  WarpPredictorResult predictor_result_;
+};
diff --git a/deploy/cpp_infer/src/modules/text_detection/predictor.cc b/deploy/cpp_infer/src/modules/text_detection/predictor.cc
new file mode 100644
index 0000000..0337f71
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/text_detection/predictor.cc
@@ -0,0 +1,139 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "predictor.h"
+
+#include "result.h"
+#include "src/common/image_batch_sampler.h"
+
+TextDetPredictor::TextDetPredictor(const TextDetPredictorParams &params)
+    : BasePredictor(params.model_dir, params.model_name, params.device,
+                    params.precision, params.enable_mkldnn,
+                    params.mkldnn_cache_capacity, params.cpu_threads,
+                    params.batch_size, "image"),
+      params_(params) {
+  auto status = Build();
+  if (!status.ok()) {
+    INFOE("Build fail: %s", status.ToString().c_str());
+    exit(-1);
+  }
+};
+
+absl::Status TextDetPredictor::Build() {
+  const auto &pre_tfs = config_.PreProcessOpInfo();
+  // Register<ReadImage>("Read", pre_tfs.at("DecodeImage.img_mode"));
+  Register<ReadImage>("Read");
+  DetResizeForTestParam resize_param;
+  resize_param.input_shape = params_.input_shape;
+  resize_param.max_side_limit = params_.max_side_limit;
+  resize_param.limit_side_len = params_.limit_side_len;
+  resize_param.limit_type = params_.limit_type;
+  resize_param.max_side_limit = params_.max_side_limit;
+  resize_param.resize_long =
+      std::stoi(pre_tfs.at("DetResizeForTest.resize_long"));
+  Register<DetResizeForTest>("Resize", resize_param);
+  Register<NormalizeImage>("Normalize");
+  Register<ToCHWImage>("ToCHW");
+  Register<ToBatch>("ToBatch");
+  infer_ptr_ = CreateStaticInfer();
+  const auto &post_params = config_.PostProcessOpInfo();
+  DBPostProcessParams db_param;
+  db_param.thresh = params_.thresh.has_value()
+                        ? params_.thresh
+                        : std::stof(post_params.at("PostProcess.thresh"));
+  db_param.box_thresh =
+      params_.box_thresh.has_value()
+          ? params_.box_thresh
+          : std::stof(post_params.at("PostProcess.box_thresh"));
+  db_param.unclip_ratio =
+      params_.unclip_ratio.has_value()
+          ? params_.unclip_ratio
+          : std::stof(post_params.at("PostProcess.unclip_ratio"));
+  db_param.max_candidates =
+      std::stoi(post_params.at("PostProcess.max_candidates"));
+  post_op_["DBPostProcess"] =
+      std::unique_ptr<DBPostProcess>(new DBPostProcess(db_param));
+  return absl::OkStatus();
+};
+
+std::vector<std::unique_ptr<BaseCVResult>>
+TextDetPredictor::Process(std::vector<cv::Mat> &batch_data) {
+  std::vector<cv::Mat> origin_image = {};
+  origin_image.reserve(batch_data.size());
+  for (const auto &mat : batch_data) {
+    origin_image.push_back(mat.clone());
+  }
+  auto batch_raw_imgs = pre_op_.at("Read")->Apply(batch_data);
+  if (!batch_raw_imgs.ok()) {
+    INFOE(batch_raw_imgs.status().ToString().c_str());
+    exit(-1);
+  }
+  std::vector<int> origin_shape = {batch_raw_imgs.value()[0].rows,
+                                   batch_raw_imgs.value()[0].cols};
+
+  auto batch_imgs = pre_op_.at("Resize")->Apply(batch_raw_imgs.value());
+  if (!batch_imgs.ok()) {
+    INFOE(batch_imgs.status().ToString().c_str());
+    exit(-1);
+  }
+  auto batch_imgs_normalize =
+      pre_op_.at("Normalize")->Apply(batch_imgs.value());
+  if (!batch_imgs_normalize.ok()) {
+    INFOE(batch_imgs_normalize.status().ToString().c_str());
+    exit(-1);
+  }
+
+  auto batch_imgs_to_chw =
+      pre_op_.at("ToCHW")->Apply(batch_imgs_normalize.value());
+  if (!batch_imgs_to_chw.ok()) {
+    INFOE(batch_imgs_to_chw.status().ToString().c_str());
+    exit(-1);
+  }
+  auto batch_imgs_to_batch =
+      pre_op_.at("ToBatch")->Apply(batch_imgs_to_chw.value());
+  if (!batch_imgs_to_batch.ok()) {
+    INFOE(batch_imgs_to_batch.status().ToString().c_str());
+    exit(-1);
+  }
+  auto infer_result = infer_ptr_->Apply(batch_imgs_to_batch.value());
+  if (!infer_result.ok()) {
+    INFOE(infer_result.status().ToString().c_str());
+    exit(-1);
+  }
+  auto db_result = post_op_.at("DBPostProcess")
+                       ->Apply(infer_result.value()[0], origin_shape);
+
+  if (!db_result.ok()) {
+    INFOE(db_result.status().ToString().c_str());
+    exit(-1);
+  }
+
+  std::vector<std::unique_ptr<BaseCVResult>> base_cv_result_ptr_vec = {};
+  for (int i = 0; i < db_result.value().size(); i++, input_index_++) {
+    TextDetPredictorResult predictor_result;
+    if (!input_path_.empty()) {
+      if (input_index_ == input_path_.size())
+        input_index_ = 0;
+      predictor_result.input_path = input_path_[input_index_];
+    }
+    predictor_result.input_image = origin_image[i];
+    predictor_result.dt_polys = db_result.value()[i].first;
+    predictor_result.dt_scores = db_result.value()[i].second;
+    predictor_result_vec_.push_back(predictor_result);
+    base_cv_result_ptr_vec.push_back(
+        std::unique_ptr<BaseCVResult>(new TextDetResult(predictor_result)));
+  }
+
+  return base_cv_result_ptr_vec;
+}
diff --git a/deploy/cpp_infer/src/modules/text_detection/predictor.h b/deploy/cpp_infer/src/modules/text_detection/predictor.h
new file mode 100644
index 0000000..bdd28a0
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/text_detection/predictor.h
@@ -0,0 +1,69 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include "processors.h"
+#include "src/base/base_batch_sampler.h"
+#include "src/base/base_cv_result.h"
+#include "src/base/base_predictor.h"
+#include "src/common/processors.h"
+
+struct TextDetPredictorResult {
+  std::string input_path = "";
+  cv::Mat input_image;
+  std::vector<std::vector<cv::Point2f>> dt_polys = {};
+  std::vector<float> dt_scores = {};
+};
+
+struct TextDetPredictorParams {
+  absl::optional<std::string> model_name = absl::nullopt;
+  absl::optional<std::string> model_dir = absl::nullopt;
+  absl::optional<std::string> device = absl::nullopt;
+  std::string precision = "fp32";
+  bool enable_mkldnn = true;
+  int mkldnn_cache_capacity = 10;
+  int cpu_threads = 8;
+  int batch_size = 1;
+  absl::optional<int> limit_side_len = absl::nullopt;
+  absl::optional<std::string> limit_type = absl::nullopt;
+  absl::optional<int> max_side_limit = absl::nullopt;
+  absl::optional<float> thresh = absl::nullopt;
+  absl::optional<float> box_thresh = absl::nullopt;
+  absl::optional<float> unclip_ratio = absl::nullopt;
+  absl::optional<std::vector<int>> input_shape = absl::nullopt;
+};
+
+class TextDetPredictor : public BasePredictor {
+public:
+  TextDetPredictor(const TextDetPredictorParams &params);
+
+  std::vector<TextDetPredictorResult> PredictorResult() const {
+    return predictor_result_vec_;
+  };
+
+  void ResetResult() override { predictor_result_vec_.clear(); };
+
+  absl::Status Build();
+
+  std::vector<std::unique_ptr<BaseCVResult>>
+  Process(std::vector<cv::Mat> &batch_data) override;
+
+private:
+  TextDetPredictorParams params_;
+  std::unordered_map<std::string, std::unique_ptr<DBPostProcess>> post_op_;
+  std::vector<TextDetPredictorResult> predictor_result_vec_;
+  std::unique_ptr<PaddleInfer> infer_ptr_;
+  int input_index_ = 0;
+};
diff --git a/deploy/cpp_infer/src/modules/text_detection/processors.cc b/deploy/cpp_infer/src/modules/text_detection/processors.cc
new file mode 100644
index 0000000..21c8b72
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/text_detection/processors.cc
@@ -0,0 +1,651 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "processors.h"
+
+#include <stdexcept>
+
+#include "src/utils/utility.h"
+
+DetResizeForTest::DetResizeForTest(const DetResizeForTestParam &params) {
+  if (params.input_shape.has_value()) {
+    input_shape_ = params.input_shape.value();
+    resize_type_ = 3;
+  } else if (params.image_shape.has_value()) {
+    image_shape_ = params.image_shape.value();
+    resize_type_ = 1;
+    if (params.keep_ratio.has_value()) {
+      keep_ratio_ = params.keep_ratio.value();
+    }
+  } else if (params.limit_side_len.has_value()) {
+    limit_side_len_ = params.limit_side_len.value();
+    limit_type_ = params.limit_type.value_or("min");
+  } else if (params.resize_long.has_value()) {
+    resize_type_ = 2;
+    resize_long_ = params.resize_long.value_or(960);
+  } else {
+    limit_side_len_ = 736;
+    limit_type_ = "min";
+  }
+  if (params.max_side_limit.has_value()) {
+    max_side_limit_ = params.max_side_limit.value();
+  }
+}
+
+absl::StatusOr<std::vector<cv::Mat>>
+DetResizeForTest::Apply(std::vector<cv::Mat> &input,
+                        const void *param_ptr) const {
+  if (input.empty()) {
+    return absl::InvalidArgumentError("Input image vector is empty.");
+  }
+  std::vector<cv::Mat> results;
+  if (param_ptr != nullptr) {
+    const DetResizeForTestParam *param =
+        static_cast<const DetResizeForTestParam *>(param_ptr);
+    for (const auto &img : input) {
+      auto res = Resize(
+          img,
+          param->limit_side_len.has_value() ? param->limit_side_len.value()
+                                            : limit_side_len_,
+          param->limit_type.has_value() ? param->limit_type.value()
+                                        : limit_type_,
+          param->max_side_limit.has_value() ? param->max_side_limit.value()
+                                            : max_side_limit_);
+      if (!res.ok())
+        return res.status();
+      results.push_back(res.value());
+    }
+  } else {
+    for (const auto &img : input) {
+      auto res = Resize(img, limit_side_len_, limit_type_, max_side_limit_);
+      if (!res.ok())
+        return res.status();
+      results.push_back(res.value());
+    }
+  }
+  return results;
+}
+
+absl::StatusOr<cv::Mat> DetResizeForTest::Resize(const cv::Mat &img,
+                                                 int limit_side_len,
+                                                 const std::string &limit_type,
+                                                 int max_side_limit) const {
+  int src_h = img.rows;
+  int src_w = img.cols;
+  if (src_h + src_w < 64) {
+    cv::Mat padded = ImagePadding(img);
+    src_h = padded.rows;
+    src_w = padded.cols;
+    return Resize(padded, limit_side_len, limit_type, max_side_limit);
+  }
+
+  switch (resize_type_) {
+  case 0:
+    return ResizeImageType0(img, limit_side_len, limit_type, max_side_limit);
+  case 1:
+    return ResizeImageType1(img);
+  case 2:
+    return ResizeImageType2(img);
+  case 3:
+    return ResizeImageType3(img);
+  default:
+    return absl::InvalidArgumentError("Unknown resize_type: " +
+                                      std::to_string(resize_type_));
+  }
+}
+
+cv::Mat DetResizeForTest::ImagePadding(const cv::Mat &img, int value) const {
+  int h = img.rows, w = img.cols, c = img.channels();
+  int pad_h = std::max(32, h);
+  int pad_w = std::max(32, w);
+  cv::Mat im_pad = cv::Mat::zeros(pad_h, pad_w, img.type());
+  im_pad.setTo(cv::Scalar::all(value));
+  img.copyTo(im_pad(cv::Rect(0, 0, w, h)));
+  return im_pad;
+}
+
+absl::StatusOr<cv::Mat>
+DetResizeForTest::ResizeImageType0(const cv::Mat &img, int limit_side_len,
+                                   const std::string &limit_type,
+                                   int max_side_limit) const {
+  int h = img.rows, w = img.cols;
+  float ratio = 1.f;
+  if (limit_type == "max") {
+    if (std::max(h, w) > limit_side_len)
+      ratio = float(limit_side_len) / std::max(h, w);
+  } else if (limit_type == "min") {
+    if (std::min(h, w) < limit_side_len)
+      ratio = float(limit_side_len) / std::min(h, w);
+  } else if (limit_type == "resize_long") {
+    ratio = float(limit_side_len) / std::max(h, w);
+  } else {
+    return absl::InvalidArgumentError("Not supported limit_type: " +
+                                      limit_type);
+  }
+  int resize_h = int(h * ratio);
+  int resize_w = int(w * ratio);
+
+  if (std::max(resize_h, resize_w) > max_side_limit) {
+    ratio = float(max_side_limit) / std::max(resize_h, resize_w);
+    resize_h = int(resize_h * ratio);
+    resize_w = int(resize_w * ratio);
+  }
+  resize_h = std::max(int(std::round(resize_h / 32.0) * 32), 32);
+  resize_w = std::max(int(std::round(resize_w / 32.0) * 32), 32);
+
+  if (resize_h == h && resize_w == w)
+    return img;
+  if (resize_h <= 0 || resize_w <= 0)
+    return absl::InvalidArgumentError("resize_w/h <= 0");
+  cv::Mat resized;
+  cv::resize(img, resized, cv::Size(resize_w, resize_h));
+  return resized;
+}
+
+absl::StatusOr<cv::Mat>
+DetResizeForTest::ResizeImageType1(const cv::Mat &img) const {
+  int resize_h = image_shape_[0];
+  int resize_w = image_shape_[1];
+  int ori_h = img.rows, ori_w = img.cols;
+  if (keep_ratio_) {
+    resize_w = int(ori_w * (float(resize_h) / ori_h));
+    int N = int(std::ceil(resize_w / 32.0));
+    resize_w = N * 32;
+  }
+  if (resize_h == ori_h && resize_w == ori_w)
+    return img;
+  cv::Mat resized;
+  cv::resize(img, resized, cv::Size(resize_w, resize_h));
+  return resized;
+}
+
+absl::StatusOr<cv::Mat>
+DetResizeForTest::ResizeImageType2(const cv::Mat &img) const {
+  int h = img.rows, w = img.cols;
+  int resize_h = h, resize_w = w;
+  float ratio;
+  if (resize_h > resize_w)
+    ratio = float(resize_long_) / resize_h;
+  else
+    ratio = float(resize_long_) / resize_w;
+
+  resize_h = int(resize_h * ratio);
+  resize_w = int(resize_w * ratio);
+
+  int max_stride = 128;
+  resize_h = ((resize_h + max_stride - 1) / max_stride) * max_stride;
+  resize_w = ((resize_w + max_stride - 1) / max_stride) * max_stride;
+
+  if (resize_h == h && resize_w == w)
+    return img;
+  cv::Mat resized;
+  cv::resize(img, resized, cv::Size(resize_w, resize_h));
+  return resized;
+}
+
+absl::StatusOr<cv::Mat>
+DetResizeForTest::ResizeImageType3(const cv::Mat &img) const {
+  if (input_shape_.size() != INPUTSHAPE)
+    return absl::InvalidArgumentError("input_shape not set for type " +
+                                      std::to_string(INPUTSHAPE));
+  int resize_h = input_shape_[1];
+  int resize_w = input_shape_[2];
+  int ori_h = img.rows, ori_w = img.cols;
+  if (resize_h == ori_h && resize_w == ori_w)
+    return img;
+  cv::Mat resized;
+  cv::resize(img, resized, cv::Size(resize_w, resize_h));
+  return resized;
+}
+
+DBPostProcess::DBPostProcess(const DBPostProcessParams &params)
+    : thresh_(params.thresh.value_or(0.3)),
+      box_thresh_(params.box_thresh.value_or(0.7)),
+      unclip_ratio_(params.unclip_ratio.value_or(2.0)),
+      max_candidates_(params.max_candidates), min_size_(3),
+      use_dilation_(params.use_dilation), score_mode_(params.score_mode),
+      box_type_(params.box_type) {
+  assert(score_mode_ == "slow" || score_mode_ == "fast");
+  assert(box_type_ == "quad" || box_type_ == "poly");
+}
+
+absl::StatusOr<
+    std::pair<std::vector<std::vector<cv::Point2f>>, std::vector<float>>>
+DBPostProcess::operator()(const cv::Mat &preds,
+                          const std::vector<int> &img_shapes,
+                          absl::optional<float> thresh,
+                          absl::optional<float> box_thresh,
+                          absl::optional<float> unclip_ratio) {
+  std::vector<std::vector<cv::Point2f>> all_boxes;
+  std::vector<float> all_scores;
+  auto preds_batch = Utility::SplitBatch(preds);
+  if (!preds_batch.ok()) {
+    return preds_batch.status();
+  }
+  for (const auto &preds_data : *preds_batch) {
+    auto result = Process(preds_data, img_shapes, thresh.value_or(thresh_),
+                          box_thresh.value_or(box_thresh_),
+                          unclip_ratio.value_or(unclip_ratio_));
+
+    if (!result.ok()) {
+      return result.status();
+    }
+
+    auto boxes_result = *result;
+    auto boxes = boxes_result.first;
+    auto scores = boxes_result.second;
+    all_boxes.insert(all_boxes.end(), boxes.begin(), boxes.end());
+    all_scores.insert(all_scores.end(), scores.begin(), scores.end());
+  }
+
+  return std::make_pair(all_boxes, all_scores);
+}
+
+absl::StatusOr<std::vector<
+    std::pair<std::vector<std::vector<cv::Point2f>>, std::vector<float>>>>
+DBPostProcess::Apply(const cv::Mat &preds, const std::vector<int> &img_shapes,
+                     absl::optional<float> thresh,
+                     absl::optional<float> box_thresh,
+                     absl::optional<float> unclip_ratio) {
+  std::vector<
+      std::pair<std::vector<std::vector<cv::Point2f>>, std::vector<float>>>
+      db_result = {};
+
+  auto preds_batch = Utility::SplitBatch(preds);
+
+  if (!preds_batch.ok()) {
+    return preds_batch.status();
+  }
+  for (const auto &pred : preds_batch.value()) {
+    auto result = Process(pred, img_shapes, thresh.value_or(thresh_),
+                          box_thresh.value_or(box_thresh_),
+                          unclip_ratio.value_or(unclip_ratio_));
+
+    if (!result.ok()) {
+      return result.status();
+    }
+    db_result.push_back(result.value());
+  }
+
+  return db_result;
+}
+
+absl::StatusOr<
+    std::pair<std::vector<std::vector<cv::Point2f>>, std::vector<float>>>
+DBPostProcess::Process(const cv::Mat &pred, const std::vector<int> &img_shape,
+                       float thresh, float box_thresh, float unclip_ratio) {
+  cv::Mat pred_single = pred.clone();
+  std::vector<int> shape_pred = {pred_single.size[pred_single.dims - 2],
+                                 pred_single.size[pred_single.dims - 1]};
+  pred_single = pred_single.reshape(1, shape_pred);
+  cv::Mat segmentation = pred_single > thresh;
+  cv::Mat mask;
+  if (use_dilation_) {
+    cv::Mat kernel = (cv::Mat_<uchar>(2, 2) << 1, 1, 1, 1); //暂时未测试
+    cv::dilate(segmentation, mask, kernel);
+  } else {
+    mask = segmentation;
+  }
+
+  int src_h = img_shape[0];
+  int src_w = img_shape[1];
+
+  if (box_type_ == "poly") {
+    return PolygonsFromBitmap(pred_single, mask, src_w, src_h, box_thresh,
+                              unclip_ratio);
+  } else if (box_type_ == "quad") {
+    return BoxesFromBitmap(pred_single, mask, src_w, src_h, box_thresh,
+                           unclip_ratio);
+  }
+
+  return absl::InvalidArgumentError(
+      "box_type can only be one of ['quad', 'poly']");
+}
+
+absl::StatusOr<
+    std::pair<std::vector<std::vector<cv::Point2f>>, std::vector<float>>>
+DBPostProcess::PolygonsFromBitmap(const cv::Mat &pred, const cv::Mat &bitmap,
+                                  int dest_width, int dest_height,
+                                  float box_thresh, float unclip_ratio) {
+  std::vector<std::vector<cv::Point2f>> boxes;
+  std::vector<float> scores;
+
+  float width_scale = static_cast<float>(dest_width) / bitmap.cols;
+  float height_scale = static_cast<float>(dest_height) / bitmap.rows;
+
+  cv::Mat bitmap_uint8;
+  bitmap.convertTo(bitmap_uint8, CV_8UC1, 255.0);
+
+  std::vector<std::vector<cv::Point2f>> contours;
+  cv::findContours(bitmap_uint8, contours, cv::RETR_LIST,
+                   cv::CHAIN_APPROX_SIMPLE);
+
+  int num_contours =
+      std::min(static_cast<int>(contours.size()), max_candidates_);
+
+  for (int i = 0; i < num_contours; ++i) {
+    const auto &contour = contours[i];
+
+    std::vector<cv::Point2f> approx;
+    double epsilon = 0.002 * cv::arcLength(contour, true);
+    cv::approxPolyDP(contour, approx, epsilon, true);
+
+    if (approx.size() < 4) {
+      continue;
+    }
+
+    float score = BoxScoreFast(pred, approx);
+    if (box_thresh > score) {
+      continue;
+    }
+
+    std::vector<cv::Point2f> box;
+    if (approx.size() > 2) {
+      auto unclip_result = Unclip(approx, unclip_ratio);
+      if (!unclip_result.ok()) {
+        continue;
+      }
+      box = *unclip_result;
+      if (box.size() > 1) {
+        continue;
+      }
+    } else {
+      continue;
+    }
+
+    if (!box.empty()) {
+      auto min_box_result = GetMiniBoxes(box);
+      auto min_box = min_box_result.first;
+      auto sside = min_box_result.second;
+      if (sside < min_size_ + 2) {
+        continue;
+      }
+
+      for (auto &point : box) {
+        point.x = std::max(
+            0, std::min(static_cast<int>(std::round(point.x * width_scale)),
+                        dest_width - 1));
+        point.y = std::max(
+            0, std::min(static_cast<int>(std::round(point.y * height_scale)),
+                        dest_height - 1));
+      }
+
+      boxes.push_back(box);
+      scores.push_back(score);
+    }
+  }
+
+  return std::make_pair(boxes, scores);
+}
+
+absl::StatusOr<
+    std::pair<std::vector<std::vector<cv::Point2f>>, std::vector<float>>>
+DBPostProcess::BoxesFromBitmap(const cv::Mat &pred, const cv::Mat &bitmap,
+                               int dest_width, int dest_height,
+                               float box_thresh, float unclip_ratio) {
+  std::vector<std::vector<cv::Point2f>> boxes;
+  std::vector<float> scores;
+
+  float width_scale = static_cast<float>(dest_width) / bitmap.cols;
+  float height_scale = static_cast<float>(dest_height) / bitmap.rows;
+
+  cv::Mat bitmap_uint8;
+  bitmap.convertTo(bitmap_uint8, CV_8UC1, 255.0);
+
+  std::vector<std::vector<cv::Point>> contours_;
+  cv::findContours(bitmap_uint8, contours_, cv::RETR_LIST,
+                   cv::CHAIN_APPROX_SIMPLE);
+  std::vector<std::vector<cv::Point2f>> contours;
+  for (const auto &contour : contours_) { // 这里可以优化
+    std::vector<cv::Point2f> float_contour;
+    for (const auto &point : contour) {
+      float_contour.push_back(cv::Point2f(point.x, point.y));
+    }
+    contours.push_back(float_contour);
+  }
+  int num_contours =
+      std::min(static_cast<int>(contours.size()), max_candidates_);
+
+  for (int i = 0; i < num_contours; ++i) {
+    const auto &contour = contours[i];
+
+    auto contour_result = GetMiniBoxes(contour);
+    auto points = contour_result.first;
+    auto sside = contour_result.second;
+    if (sside < min_size_) {
+      continue;
+    }
+
+    float score = 0;
+    if (score_mode_ == "fast") {
+      score = BoxScoreFast(pred, points);
+    } else {
+      score = BoxScoreSlow(pred, contour);
+    }
+
+    if (box_thresh > score) {
+      continue;
+    }
+
+    auto unclip_result = Unclip(points, unclip_ratio);
+    if (!unclip_result.ok()) {
+      continue;
+    }
+
+    auto box = *unclip_result;
+    auto min_box_result = GetMiniBoxes(box);
+    auto min_box = min_box_result.first;
+    auto new_sside = min_box_result.second;
+    if (new_sside < min_size_ + 2) {
+      continue;
+    }
+
+    for (auto &point : min_box) {
+      point.x = std::max(
+          0, std::min(static_cast<int>(std::round(point.x * width_scale)),
+                      dest_width - 1));
+      point.y = std::max(
+          0, std::min(static_cast<int>(std::round(point.y * height_scale)),
+                      dest_height - 1));
+    }
+
+    boxes.push_back(min_box);
+    scores.push_back(score);
+  }
+
+  return std::make_pair(boxes, scores);
+}
+
+absl::StatusOr<std::vector<cv::Point2f>>
+DBPostProcess::Unclip(const std::vector<cv::Point2f> &box, float unclip_ratio) {
+  float area = cv::contourArea(box);
+  float length = cv::arcLength(box, true);
+  float distance = area * unclip_ratio / length;
+
+  ClipperLib::Path path;
+  for (const auto &point : box) {
+    path << ClipperLib::IntPoint(point.x, point.y);
+  }
+
+  ClipperLib::ClipperOffset co;
+  co.AddPath(path, ClipperLib::jtRound, ClipperLib::etClosedPolygon);
+
+  ClipperLib::Paths solution;
+  co.Execute(solution, distance);
+
+  if (solution.empty()) {
+    return absl::InternalError("Failed to unclip polygon");
+  }
+
+  std::vector<cv::Point2f> result;
+  for (const auto &p : solution[0]) {
+    result.emplace_back(p.X, p.Y);
+  }
+
+  return result;
+}
+
+std::pair<std::vector<cv::Point2f>, float>
+DBPostProcess::GetMiniBoxes(const std::vector<cv::Point2f> &contour) {
+  cv::RotatedRect box = cv::minAreaRect(contour);
+
+  std::vector<cv::Point2f> points(4);
+  box.points(points.data());
+
+  std::sort(
+      points.begin(), points.end(),
+      [](const cv::Point2f &a, const cv::Point2f &b) { return a.x < b.x; });
+
+  int index_1 = 0, index_2 = 1, index_3 = 2, index_4 = 3;
+  if (points[1].y > points[0].y) {
+    index_1 = 0;
+    index_4 = 1;
+  } else {
+    index_1 = 1;
+    index_4 = 0;
+  }
+
+  if (points[3].y > points[2].y) {
+    index_2 = 2;
+    index_3 = 3;
+  } else {
+    index_2 = 3;
+    index_3 = 2;
+  }
+
+  std::vector<cv::Point2f> box_points = {points[index_1], points[index_2],
+                                         points[index_3], points[index_4]};
+
+  float sside = std::min(box.size.width, box.size.height);
+  return std::make_pair(box_points, sside);
+}
+
+float DBPostProcess::BoxScoreFast(const cv::Mat &bitmap,
+                                  const std::vector<cv::Point2f> &contour) {
+  int h = bitmap.size[bitmap.dims - 2]; // must be CHW
+  int w = bitmap.size[bitmap.dims - 1];
+
+  std::vector<cv::Point2f> contour_copy = contour;
+
+  int xmin = std::max(
+      0, static_cast<int>(std::floor(
+             std::min_element(contour_copy.begin(), contour_copy.end(),
+                              [](const cv::Point2f &a, const cv::Point2f &b) {
+                                return a.x < b.x;
+                              })
+                 ->x)));
+  int xmax = std::max(
+      0, static_cast<int>(std::ceil(
+             std::max_element(contour_copy.begin(), contour_copy.end(),
+                              [](const cv::Point2f &a, const cv::Point2f &b) {
+                                return a.x < b.x;
+                              })
+                 ->x)));
+  int ymin = std::max(
+      0, static_cast<int>(std::floor(
+             std::min_element(contour_copy.begin(), contour_copy.end(),
+                              [](const cv::Point2f &a, const cv::Point2f &b) {
+                                return a.y < b.y;
+                              })
+                 ->y)));
+  int ymax = std::max(
+      0, static_cast<int>(std::ceil(
+             std::max_element(contour_copy.begin(), contour_copy.end(),
+                              [](const cv::Point2f &a, const cv::Point2f &b) {
+                                return a.y < b.y;
+                              })
+                 ->y)));
+
+  xmin = std::min(xmin, w - 1);
+  xmax = std::min(xmax, w - 1);
+  ymin = std::min(ymin, h - 1);
+  ymax = std::min(ymax, h - 1);
+
+  cv::Mat mask = cv::Mat::zeros(ymax - ymin + 1, xmax - xmin + 1, CV_8UC1);
+
+  std::vector<cv::Point> contour_copy_int;
+  for (auto &point : contour_copy) {
+    point.x -= xmin;
+    point.y -= ymin;
+    contour_copy_int.push_back(
+        cv::Point(static_cast<int>(point.x), static_cast<int>(point.y)));
+  }
+
+  std::vector<std::vector<cv::Point>> contours = {contour_copy_int};
+  cv::fillPoly(mask, contours, cv::Scalar(1));
+
+  cv::Mat roi = bitmap(cv::Rect(xmin, ymin, xmax - xmin + 1, ymax - ymin + 1));
+  cv::Scalar mean_val = cv::mean(roi, mask);
+
+  return mean_val[0];
+}
+
+float DBPostProcess::BoxScoreSlow(const cv::Mat &bitmap,
+                                  const std::vector<cv::Point2f> &contour) {
+  int h = bitmap.size[bitmap.dims - 2]; // must be CHW
+  int w = bitmap.size[bitmap.dims - 1];
+
+  std::vector<cv::Point2f> contour_copy = contour;
+
+  int xmin = std::max(
+      0, static_cast<int>(std::floor(
+             std::min_element(contour_copy.begin(), contour_copy.end(),
+                              [](const cv::Point2f &a, const cv::Point2f &b) {
+                                return a.x < b.x;
+                              })
+                 ->x)));
+  int xmax = std::max(
+      0, static_cast<int>(std::ceil(
+             std::max_element(contour_copy.begin(), contour_copy.end(),
+                              [](const cv::Point2f &a, const cv::Point2f &b) {
+                                return a.x < b.x;
+                              })
+                 ->x)));
+  int ymin = std::max(
+      0, static_cast<int>(std::floor(
+             std::min_element(contour_copy.begin(), contour_copy.end(),
+                              [](const cv::Point2f &a, const cv::Point2f &b) {
+                                return a.y < b.y;
+                              })
+                 ->y)));
+  int ymax = std::max(
+      0, static_cast<int>(std::ceil(
+             std::max_element(contour_copy.begin(), contour_copy.end(),
+                              [](const cv::Point2f &a, const cv::Point2f &b) {
+                                return a.y < b.y;
+                              })
+                 ->y)));
+
+  xmin = std::min(xmin, w - 1);
+  xmax = std::min(xmax, w - 1);
+  ymin = std::min(ymin, h - 1);
+  ymax = std::min(ymax, h - 1);
+
+  cv::Mat mask = cv::Mat::zeros(ymax - ymin + 1, xmax - xmin + 1, CV_8UC1);
+
+  std::vector<cv::Point> contour_copy_int;
+  for (auto &point : contour_copy) {
+    point.x -= xmin;
+    point.y -= ymin;
+    contour_copy_int.push_back(
+        cv::Point(static_cast<int>(point.x), static_cast<int>(point.y)));
+  }
+
+  std::vector<std::vector<cv::Point>> contours = {contour_copy_int};
+  cv::fillPoly(mask, contours, 1);
+
+  cv::Scalar mean = cv::mean(
+      bitmap(cv::Rect(xmin, ymin, xmax - xmin + 1, ymax - ymin + 1)), mask);
+  return static_cast<float>(mean[0]);
+}
diff --git a/deploy/cpp_infer/src/modules/text_detection/processors.h b/deploy/cpp_infer/src/modules/text_detection/processors.h
new file mode 100644
index 0000000..6f8d467
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/text_detection/processors.h
@@ -0,0 +1,135 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <iostream>
+#include <opencv2/opencv.hpp>
+#include <string>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+#include "absl/types/optional.h"
+#include "polyclipping/clipper.hpp"
+#include "src/utils/func_register.h"
+
+struct DetResizeForTestParam {
+  absl::optional<std::vector<int>> input_shape = absl::nullopt;
+  absl::optional<int> max_side_limit = absl::nullopt;
+  absl::optional<std::vector<int>> image_shape = absl::nullopt;
+  absl::optional<bool> keep_ratio = absl::nullopt;
+  absl::optional<int> limit_side_len = absl::nullopt;
+  absl::optional<std::string> limit_type = absl::nullopt;
+  absl::optional<int> resize_long = absl::nullopt;
+};
+
+class DetResizeForTest : public BaseProcessor {
+public:
+  DetResizeForTest(const DetResizeForTestParam &params);
+  absl::StatusOr<std::vector<cv::Mat>>
+  Apply(std::vector<cv::Mat> &input,
+        const void *param_ptr = nullptr) const override;
+
+private:
+  int resize_type_ = 0;
+  bool keep_ratio_ = false;
+  int resize_long_;
+  std::vector<int> input_shape_;
+  std::vector<int> image_shape_;
+  int limit_side_len_;
+  std::string limit_type_;
+  int max_side_limit_ = 4000;
+
+  absl::StatusOr<cv::Mat> Resize(const cv::Mat &img, int limit_side_len,
+                                 const std::string &limit_type,
+                                 int max_side_limit) const;
+
+  cv::Mat ImagePadding(const cv::Mat &img, int value = 0) const;
+
+  absl::StatusOr<cv::Mat> ResizeImageType0(const cv::Mat &img,
+                                           int limit_side_len,
+                                           const std::string &limit_type,
+                                           int max_side_limit) const;
+  absl::StatusOr<cv::Mat> ResizeImageType1(const cv::Mat &img) const;
+  absl::StatusOr<cv::Mat> ResizeImageType2(const cv::Mat &img) const;
+  absl::StatusOr<cv::Mat> ResizeImageType3(const cv::Mat &img) const;
+  static constexpr int INPUTSHAPE = 3;
+};
+
+struct DBPostProcessParams {
+  absl::optional<float> thresh = absl::nullopt;
+  absl::optional<float> box_thresh = absl::nullopt;
+  absl::optional<float> unclip_ratio = absl::nullopt;
+  int max_candidates = 1000;
+  bool use_dilation = false;
+  std::string score_mode = "fast";
+  std::string box_type = "quad";
+};
+
+class DBPostProcess {
+public:
+  DBPostProcess(const DBPostProcessParams &params);
+
+  absl::StatusOr<
+      std::pair<std::vector<std::vector<cv::Point2f>>, std::vector<float>>>
+  operator()(const cv::Mat &preds, const std::vector<int> &img_shapes,
+             absl::optional<float> thresh = absl::nullopt,
+             absl::optional<float> box_thresh = absl::nullopt,
+             absl::optional<float> unclip_ratio = absl::nullopt);
+  absl::StatusOr<std::vector<
+      std::pair<std::vector<std::vector<cv::Point2f>>, std::vector<float>>>>
+  Apply(const cv::Mat &preds, const std::vector<int> &img_shapes,
+        absl::optional<float> thresh = absl::nullopt,
+        absl::optional<float> box_thresh = absl::nullopt,
+        absl::optional<float> unclip_ratio = absl::nullopt);
+
+private:
+  absl::StatusOr<
+      std::pair<std::vector<std::vector<cv::Point2f>>, std::vector<float>>>
+  Process(const cv::Mat &pred, const std::vector<int> &img_shape, float thresh,
+          float box_thresh, float unclip_ratio);
+
+  absl::StatusOr<
+      std::pair<std::vector<std::vector<cv::Point2f>>, std::vector<float>>>
+  PolygonsFromBitmap(const cv::Mat &pred, const cv::Mat &bitmap, int dest_width,
+                     int dest_height, float box_thresh, float unclip_ratio);
+
+  absl::StatusOr<
+      std::pair<std::vector<std::vector<cv::Point2f>>, std::vector<float>>>
+  BoxesFromBitmap(const cv::Mat &pred, const cv::Mat &bitmap, int dest_width,
+                  int dest_height, float box_thresh, float unclip_ratio);
+
+  absl::StatusOr<std::vector<cv::Point2f>>
+  Unclip(const std::vector<cv::Point2f> &box, float unclip_ratio);
+
+  std::pair<std::vector<cv::Point2f>, float>
+  GetMiniBoxes(const std::vector<cv::Point2f> &contour);
+
+  float BoxScoreFast(const cv::Mat &bitmap,
+                     const std::vector<cv::Point2f> &contour);
+
+  float BoxScoreSlow(const cv::Mat &bitmap,
+                     const std::vector<cv::Point2f> &contour);
+
+private:
+  float thresh_;
+  float box_thresh_;
+  int max_candidates_;
+  float unclip_ratio_;
+  int min_size_;
+  bool use_dilation_;
+  std::string score_mode_;
+  std::string box_type_;
+};
diff --git a/deploy/cpp_infer/src/modules/text_detection/result.cc b/deploy/cpp_infer/src/modules/text_detection/result.cc
new file mode 100644
index 0000000..07a2a4f
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/text_detection/result.cc
@@ -0,0 +1,135 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "result.h"
+
+#include <fstream>
+#include <string>
+
+#include "src/utils/utility.h"
+#include "third_party/nlohmann/json.hpp"
+
+using json = nlohmann::json;
+void TextDetResult::SaveToImg(const std::string &save_path) {
+  cv::Mat img = predictor_result_.input_image.clone();
+
+  const auto &dt_polys = predictor_result_.dt_polys;
+  for (const auto &poly : dt_polys) {
+    std::vector<cv::Point> pts;
+    for (const auto &pt : poly) {
+      pts.emplace_back(cv::Point(cvRound(pt.x), cvRound(pt.y)));
+    }
+
+    const cv::Point *pts_ptr = pts.data();
+    int npts = pts.size();
+    cv::polylines(img, &pts_ptr, &npts, 1, true, cv::Scalar(0, 0, 255), 2);
+  }
+
+  absl::StatusOr<std::string> full_path;
+  if (predictor_result_.input_path.empty()) {
+    auto now = std::chrono::system_clock::now();
+    auto now_time = std::chrono::system_clock::to_time_t(now);
+    std::stringstream ss;
+    ss << "output_" << std::put_time(std::localtime(&now_time), "%Y%m%d_%H%M%S")
+       << ".jpg";
+    std::string timestamp_filename = ss.str();
+    INFOW("Input path is empty, will use %s instead!",
+          timestamp_filename.c_str());
+    predictor_result_.input_path = timestamp_filename;
+    full_path =
+        Utility::SmartCreateDirectoryForImage(save_path, timestamp_filename);
+  } else {
+    full_path = Utility::SmartCreateDirectoryForImage(
+        save_path, predictor_result_.input_path);
+  }
+  if (!full_path.ok()) {
+    INFOE(full_path.status().ToString().c_str());
+    exit(-1);
+  }
+  bool success = cv::imwrite(full_path.value(), img);
+  if (!success) {
+    INFOE("Failed to write the image : %s", full_path.value().c_str());
+    exit(-1);
+  }
+}
+
+void TextDetResult::Print() const {
+  std::cout << "{\n  \"res\": {" << std::endl;
+
+  std::cout << "    \"input_path\": {" << predictor_result_.input_path
+            << "    }," << std::endl;
+
+  std::cout << "    \"dt_polys\": [" << std::endl;
+  for (const auto &polygon : predictor_result_.dt_polys) {
+    std::cout << "        [";
+    for (size_t i = 0; i < polygon.size(); ++i) {
+      std::cout << "[" << static_cast<int>(polygon[i].x) << ", "
+                << static_cast<int>(polygon[i].y) << "]";
+      if (i < polygon.size() - 1)
+        std::cout << ", ";
+    }
+    std::cout << "]," << std::endl;
+  }
+
+  std::cout << "      ]}," << std::endl;
+
+  std::cout << "    \"dt_scores\": [" << std::endl;
+  for (auto it = predictor_result_.dt_scores.begin();
+       it != predictor_result_.dt_scores.end(); ++it) {
+    std::cout << *it;
+    if (it < predictor_result_.dt_scores.end() - 1)
+      std::cout << ", ";
+  }
+  std::cout << "]}" << std::endl;
+
+  std::cout << "    ]" << std::endl;
+
+  std::cout << "  }\n}" << std::endl;
+}
+
+void TextDetResult::SaveToJson(const std::string &save_path) const {
+  nlohmann::ordered_json j;
+
+  j["input_path"] = predictor_result_.input_path;
+
+  j["page_index"] = nullptr; //********
+  json polys_json = json::array();
+  for (const auto &polygon : predictor_result_.dt_polys) {
+    json poly_json = json::array();
+    for (const auto &point : polygon) {
+      poly_json.push_back(
+          {static_cast<int>(point.x), static_cast<int>(point.y)});
+    }
+    polys_json.push_back(poly_json);
+  }
+  j["dt_polys"] = polys_json;
+  j["dt_score"] = predictor_result_.dt_scores;
+
+  absl::StatusOr<std::string> full_path;
+
+  full_path = Utility::SmartCreateDirectoryForJson(
+      save_path, predictor_result_.input_path);
+  if (!full_path.ok()) {
+    INFOE(full_path.status().ToString().c_str());
+    exit(-1);
+  }
+  std::ofstream file(full_path.value());
+  if (file.is_open()) {
+    file << j.dump(4);
+    file.close();
+  } else {
+    INFOE("Could not open file for writing: %s", save_path.c_str());
+    exit(-1);
+  }
+}
diff --git a/deploy/cpp_infer/src/modules/text_detection/result.h b/deploy/cpp_infer/src/modules/text_detection/result.h
new file mode 100644
index 0000000..0c2417c
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/text_detection/result.h
@@ -0,0 +1,34 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <opencv2/opencv.hpp>
+#include <vector>
+
+#include "predictor.h"
+#include "src/base/base_cv_result.h"
+
+class TextDetResult : public BaseCVResult {
+public:
+  TextDetResult(TextDetPredictorResult predictor_result)
+      : BaseCVResult(), predictor_result_(predictor_result){};
+  // std::unordered_map<std::string, cv::Mat> ToImg() const override;
+  void SaveToImg(const std::string &save_path) override;
+  void Print() const override;
+  void SaveToJson(const std::string &save_path) const override;
+
+private:
+  TextDetPredictorResult predictor_result_;
+};
diff --git a/deploy/cpp_infer/src/modules/text_recognition/predictor.cc b/deploy/cpp_infer/src/modules/text_recognition/predictor.cc
new file mode 100644
index 0000000..08d0fe5
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/text_recognition/predictor.cc
@@ -0,0 +1,153 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "predictor.h"
+
+#include <algorithm>
+
+#include "result.h"
+#include "src/common/image_batch_sampler.h"
+
+TextRecPredictor::TextRecPredictor(const TextRecPredictorParams &params)
+    : BasePredictor(params.model_dir, params.model_name, params.device,
+                    params.precision, params.enable_mkldnn,
+                    params.mkldnn_cache_capacity, params.cpu_threads,
+                    params.batch_size, "image"),
+      params_(params) {
+  auto status = CheckRecModelParams();
+  auto status_build = Build();
+  if (!status_build.ok()) {
+    INFOE("Build fail: %s", status_build.ToString().c_str());
+    exit(-1);
+  }
+};
+
+absl::Status TextRecPredictor::Build() {
+  const auto &pre_params = config_.PreProcessOpInfo();
+  Register<ReadImage>("Read", "BGR"); //******
+  Register<OCRReisizeNormImg>("ReisizeNorm", params_.input_shape);
+  Register<ToBatchUniform>("ToBatch");
+  infer_ptr_ = CreateStaticInfer();
+  const auto &post_params = config_.PostProcessOpInfo();
+  post_op_["CTCLabelDecode"] = std::unique_ptr<CTCLabelDecode>(
+      new CTCLabelDecode(YamlConfig::SmartParseVector(
+                             post_params.at("PostProcess.character_dict"))
+                             .vec_string));
+  return absl::OkStatus();
+};
+
+std::vector<std::unique_ptr<BaseCVResult>>
+TextRecPredictor::Process(std::vector<cv::Mat> &batch_data) {
+  std::vector<cv::Mat> origin_image = {};
+  origin_image.reserve(batch_data.size());
+  for (const auto &mat : batch_data) {
+    origin_image.push_back(mat.clone());
+  }
+  auto batch_read = pre_op_.at("Read")->Apply(batch_data);
+  if (!batch_read.ok()) {
+    INFOE(batch_read.status().ToString().c_str());
+    exit(-1);
+  }
+
+  auto batch_resize_norm = pre_op_.at("ReisizeNorm")->Apply(batch_read.value());
+  if (!batch_resize_norm.ok()) {
+    INFOE(batch_resize_norm.status().ToString().c_str());
+    exit(-1);
+  }
+
+  auto batch_tobatch = pre_op_.at("ToBatch")->Apply(batch_resize_norm.value());
+  if (!batch_tobatch.ok()) {
+    INFOE(batch_tobatch.status().ToString().c_str());
+    exit(-1);
+  }
+  auto batch_infer = infer_ptr_->Apply(batch_tobatch.value());
+  if (!batch_infer.ok()) {
+    INFOE(batch_infer.status().ToString().c_str());
+    exit(-1);
+  }
+
+  auto ctc_result =
+      post_op_.at("CTCLabelDecode")->Apply(batch_infer.value()[0]);
+
+  if (!ctc_result.ok()) {
+    INFOE(ctc_result.status().ToString().c_str());
+    exit(-1);
+  }
+
+  std::vector<std::unique_ptr<BaseCVResult>> base_cv_result_ptr_vec = {};
+  for (int i = 0; i < ctc_result.value().size(); i++, input_index_++) {
+    TextRecPredictorResult predictor_result;
+    if (!input_path_.empty()) {
+      if (input_index_ == input_path_.size())
+        input_index_ = 0;
+      predictor_result.input_path = input_path_[input_index_];
+    }
+    predictor_result.input_image = origin_image[i];
+    predictor_result.rec_text = ctc_result.value()[i].first;
+    predictor_result.rec_score = ctc_result.value()[i].second;
+    predictor_result.vis_font = params_.vis_font_dir.value_or("");
+    predictor_result_vec_.push_back(predictor_result);
+    base_cv_result_ptr_vec.push_back(
+        std::unique_ptr<BaseCVResult>(new TextRecResult(predictor_result)));
+  }
+  return base_cv_result_ptr_vec;
+}
+
+absl::Status TextRecPredictor::CheckRecModelParams() {
+  auto result_models_check = Utility::GetOcrModelInfo(
+      params_.lang.value_or(""), params_.ocr_version.value_or(""));
+  if (!result_models_check.ok()) {
+    return absl::InvalidArgumentError("lang and ocr_version is invalid : " +
+                                      result_models_check.status().ToString());
+  }
+  auto result_model_name = ModelName();
+  if (!result_model_name.ok()) {
+    return absl::InternalError("Get model name fail : " +
+                               result_model_name.status().ToString());
+  }
+  size_t pos_model_name = result_model_name.value().find('_');
+  size_t pos_model_check = std::get<1>(result_models_check.value()).find('_');
+  std::string prefix_model_name =
+      result_model_name.value().substr(0, pos_model_name);
+  std::string prefix_model_check =
+      std::get<1>(result_models_check.value()).substr(0, pos_model_check);
+  auto result =
+      Utility::GetOcrModelInfo(params_.lang.value_or(""), prefix_model_name);
+  if (!result.ok()) {
+    return absl::InternalError("Model and lang do not match : " +
+                               result.status().ToString());
+  }
+  if (params_.ocr_version.has_value()) {
+    if (prefix_model_name != params_.ocr_version.value()) {
+      INFOW("Rec model ocr_version and ocr_verision params do not match");
+    }
+  }
+
+#ifdef USE_FREETYPE
+  if (!params_.vis_font_dir.has_value()) {
+    return absl::InvalidArgumentError(
+        "Visualization font path is empty, please provide " +
+        std::get<2>(result_models_check.value()) + " path.");
+  } else {
+    size_t pos = params_.vis_font_dir.value().find_last_of("/\\");
+    std::string filename = params_.vis_font_dir.value().substr(pos + 1);
+    if (filename != std::get<2>(result_models_check.value())) {
+      return absl::NotFoundError("Expected visualization font is " +
+                                 std::get<2>(result_models_check.value()) +
+                                 ", but get is " + filename);
+    }
+  }
+#endif
+  return absl::OkStatus();
+}
diff --git a/deploy/cpp_infer/src/modules/text_recognition/predictor.h b/deploy/cpp_infer/src/modules/text_recognition/predictor.h
new file mode 100644
index 0000000..41b003e
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/text_recognition/predictor.h
@@ -0,0 +1,69 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include "processors.h"
+#include "src/base/base_batch_sampler.h"
+#include "src/base/base_cv_result.h"
+#include "src/base/base_predictor.h"
+#include "src/common/processors.h"
+
+struct TextRecPredictorResult {
+  std::string input_path = "";
+  cv::Mat input_image;
+  std::string rec_text = "";
+  float rec_score = 0.0;
+  std::string vis_font = "";
+};
+
+struct TextRecPredictorParams {
+  absl::optional<std::string> model_name = absl::nullopt;
+  absl::optional<std::string> model_dir = absl::nullopt;
+  absl::optional<std::string> lang = absl::nullopt;
+  absl::optional<std::string> ocr_version = absl::nullopt;
+  absl::optional<std::string> vis_font_dir = absl::nullopt;
+  absl::optional<std::string> device = absl::nullopt;
+  std::string precision = "fp32";
+  bool enable_mkldnn = true;
+  int mkldnn_cache_capacity = 10;
+  int cpu_threads = 8;
+  int batch_size = 1;
+  absl::optional<std::vector<int>> input_shape = absl::nullopt;
+};
+
+class TextRecPredictor : public BasePredictor {
+public:
+  TextRecPredictor(const TextRecPredictorParams &params);
+
+  std::vector<TextRecPredictorResult> PredictorResult() const {
+    return predictor_result_vec_;
+  };
+
+  void ResetResult() override { predictor_result_vec_.clear(); };
+
+  absl::Status Build();
+
+  std::vector<std::unique_ptr<BaseCVResult>>
+  Process(std::vector<cv::Mat> &batch_data) override;
+
+  absl::Status CheckRecModelParams();
+
+private:
+  std::unordered_map<std::string, std::unique_ptr<CTCLabelDecode>> post_op_;
+  std::vector<TextRecPredictorResult> predictor_result_vec_;
+  std::unique_ptr<PaddleInfer> infer_ptr_;
+  TextRecPredictorParams params_;
+  int input_index_ = 0;
+};
diff --git a/deploy/cpp_infer/src/modules/text_recognition/processors.cc b/deploy/cpp_infer/src/modules/text_recognition/processors.cc
new file mode 100644
index 0000000..1f374b5
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/text_recognition/processors.cc
@@ -0,0 +1,267 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "processors.h"
+
+#include <numeric>
+#include <sstream>
+#include <stdexcept>
+#include <unordered_map>
+
+#include "src/utils/utility.h"
+
+absl::StatusOr<std::vector<cv::Mat>>
+OCRReisizeNormImg::Apply(std::vector<cv::Mat> &input, const void *param) const {
+  std::vector<cv::Mat> output = {};
+  output.reserve(input.size());
+  if (input_shape_.empty()) {
+    for (auto &image : input) {
+      auto result = Resize(image);
+      if (!result.ok()) {
+        return result.status();
+      }
+      output.push_back(result.value());
+    }
+  } else {
+    for (auto &image : input) {
+      auto result = StaticResize(image);
+      if (!result.ok()) {
+        return result.status();
+      }
+      output.push_back(result.value());
+    }
+  }
+  return output;
+}
+
+absl::StatusOr<cv::Mat> OCRReisizeNormImg::Resize(cv::Mat &image) const {
+  float rec_wh_ratio = (float)rec_image_shape_[2] / (float)rec_image_shape_[1];
+  float image_wh_ratio = (float)image.size[1] / (float)image.size[0];
+  float max_wh_ratio = std::max(rec_wh_ratio, image_wh_ratio);
+  auto image_result = ResizeNormImg(image, max_wh_ratio);
+  if (!image_result.ok()) {
+    return image_result.status();
+  }
+  return image_result.value();
+}
+
+absl::StatusOr<cv::Mat> OCRReisizeNormImg::StaticResize(cv::Mat &image) const {
+  cv::Mat resize_image;
+  int img_c = input_shape_[0];
+  int img_h = input_shape_[1];
+  int img_w = input_shape_[2];
+  cv::resize(image, resize_image, cv::Size(img_w, img_h));
+  resize_image.convertTo(resize_image, CV_32F);
+  std::vector<cv::Mat> mat_split(resize_image.channels());
+  cv::split(resize_image, mat_split);
+  for (auto &item : mat_split) {
+    item /= 255;
+    item -= 0.5;
+    item /= 0.5;
+    item = item.reshape(1, 1);
+  }
+  cv::Mat resize_image_process;
+  cv::hconcat(mat_split, resize_image_process);
+  std::vector<int> resize_shape = {img_c, img_h, img_w};
+  resize_image_process = resize_image_process.reshape(1, resize_shape);
+  return resize_image_process;
+}
+
+absl::StatusOr<cv::Mat>
+OCRReisizeNormImg::ResizeNormImg(cv::Mat &image, float max_wh_ratio) const {
+  assert(rec_image_shape_[0] == image.channels());
+  int rec_c = rec_image_shape_[0];
+  int rec_h = rec_image_shape_[1];
+  int rec_w = rec_image_shape_[2];
+
+  rec_w = rec_h * max_wh_ratio;
+  cv::Mat resize_image;
+  int resize_w = 0;
+  if (rec_w > MAX_IMG_W) {
+    rec_w = MAX_IMG_W;
+    resize_w = MAX_IMG_W;
+    cv::resize(image, resize_image, cv::Size(resize_w, rec_h));
+  } else {
+    float wh_ratio = (float)image.size[1] / (float)image.size[0];
+    if (std::ceil(rec_h * wh_ratio) > rec_w) {
+      resize_w = rec_w;
+    } else {
+      resize_w = std::ceil(rec_h * wh_ratio);
+    }
+    cv::resize(image, resize_image, cv::Size(resize_w, rec_h));
+  }
+  resize_image.convertTo(resize_image, CV_32F);
+  std::vector<cv::Mat> mat_split(resize_image.channels());
+  cv::split(resize_image, mat_split);
+  for (auto &item : mat_split) {
+    item /= 255;
+    item -= 0.5;
+    item /= 0.5;
+    item = item.reshape(1, 1);
+  }
+  cv::Mat resize_image_process;
+  cv::hconcat(mat_split, resize_image_process);
+  std::vector<int> resize_shape = {rec_c, rec_h, resize_w};
+  resize_image_process = resize_image_process.reshape(1, resize_shape);
+  std::vector<int> image_shape = {rec_c, rec_h, rec_w};
+  cv::Mat padding_im =
+      cv::Mat::zeros(image_shape.size(), &image_shape[0], CV_32F);
+  for (int c = 0; c < rec_c; ++c) {
+    for (int row = 0; row < rec_h; ++row) {
+      float *dst = padding_im.ptr<float>(c) + row * rec_w;
+      float *src = resize_image_process.ptr<float>(c) + row * resize_w;
+      std::copy(src, src + resize_w, dst);
+    }
+  }
+  return padding_im;
+}
+
+CTCLabelDecode::CTCLabelDecode(const std::vector<std::string> &character_list,
+                               bool use_space_char)
+    : character_list_(character_list), use_space_char_(use_space_char) {
+  if (character_list_.empty()) {
+    const std::string normal = "0123456789abcdefghijklmnopqrstuvwxyz";
+    for (const auto &item : normal) {
+      character_list_.emplace_back(std::string(1, item));
+    }
+  }
+  if (use_space_char) {
+    character_list_.emplace_back(std::string(" "));
+  }
+  AddSpecialChar();
+  dict_.reserve(character_list_.size());
+  for (int i = 0; i < character_list_.size(); i++) {
+    dict_[i] = character_list_[i];
+  }
+}
+
+absl::StatusOr<std::vector<std::pair<std::string, float>>>
+CTCLabelDecode::Apply(const cv::Mat &preds) const {
+  auto preds_batch = Utility::SplitBatch(preds);
+  std::vector<std::pair<std::string, float>> ctc_result = {};
+  ctc_result.reserve(preds_batch.value().size());
+  if (!preds_batch.ok()) {
+    return preds_batch.status();
+  }
+  for (const auto &pred : preds_batch.value()) {
+    auto result = Process(pred);
+    if (!result.ok()) {
+      return result.status();
+    }
+    ctc_result.push_back(result.value());
+  }
+  return ctc_result;
+}
+
+absl::StatusOr<std::pair<std::string, float>>
+CTCLabelDecode::Process(const cv::Mat &pred_data) const {
+  std::vector<int> shape_squeeze = {};
+  for (int i = 1; i < pred_data.dims; i++) {
+    shape_squeeze.push_back(pred_data.size[i]);
+  }
+  cv::Mat pred_data_process;
+  pred_data_process = pred_data.reshape(1, shape_squeeze);
+
+  int seq_len = pred_data_process.size[0];
+  int num_classes = pred_data_process.size[1];
+  std::list<int> text_index;
+  std::list<float> text_prob;
+  for (int t = 0; t < seq_len; ++t) {
+    const float *row_ptr = pred_data_process.ptr<float>(t);
+    float max_val = row_ptr[0];
+    int max_idx = 0;
+    for (int c = 1; c < num_classes; ++c) {
+      if (row_ptr[c] > max_val) {
+        max_val = row_ptr[c];
+        max_idx = c;
+      }
+    }
+    text_index.push_back(max_idx);
+    text_prob.push_back(max_val);
+  }
+  auto decode_result = Decode(text_index, text_prob, true);
+  if (!decode_result.ok()) {
+    return decode_result.status();
+  }
+  return decode_result.value();
+}
+
+absl::StatusOr<std::pair<std::string, float>>
+CTCLabelDecode::Decode(std::list<int> &text_index, std::list<float> &text_prob,
+                       bool is_remove_duplicate) const {
+  std::vector<bool> selection(text_index.size(), true);
+  if (is_remove_duplicate && text_index.size() > 1) {
+    auto prev = text_index.begin();
+    auto curr = std::next(prev);
+    size_t idx = 1;
+    for (; curr != text_index.end(); ++curr, ++prev, ++idx) {
+      if (*curr == *prev)
+        selection[idx] = false;
+    }
+  }
+  for (const auto &ignore_item : IGNORE_TOKEN) {
+    size_t idx = 0;
+    for (auto item_list = text_index.begin(); item_list != text_index.end();
+         ++item_list, idx++) {
+      if (*item_list == ignore_item) {
+        selection[idx] = false;
+      }
+    }
+  }
+  auto sel_it = selection.begin();
+  for (auto it = text_index.begin(); it != text_index.end();) {
+    if (!(*sel_it)) {
+      it = text_index.erase(it);
+    } else {
+      ++it;
+    }
+    ++sel_it;
+  }
+  auto sel_it_prob = selection.begin();
+  for (auto it = text_prob.begin(); it != text_prob.end();) {
+    if (!(*sel_it_prob)) {
+      it = text_prob.erase(it);
+    } else {
+      ++it;
+    }
+    ++sel_it_prob;
+  }
+  std::vector<std::string> char_list = {};
+  for (auto list_index = text_index.begin(); list_index != text_index.end();
+       ++list_index) {
+    if ((*list_index) < character_list_.size()) {
+      char_list.push_back(character_list_[*list_index]);
+    } else {
+      char_list.push_back(" ");
+    }
+  }
+  std::list<float> conf_list = {};
+  if (!text_prob.empty()) {
+    conf_list = text_prob;
+  } else {
+    conf_list = std::list<float>(selection.size(), 0);
+  }
+  std::string text;
+  for (const auto &item_char : char_list) {
+    text += item_char;
+  }
+  float sum = std::accumulate(conf_list.begin(), conf_list.end(), 0.0f);
+  float mean = sum / conf_list.size();
+
+  return std::pair<std::string, float>(text, mean);
+}
+
+void CTCLabelDecode::AddSpecialChar() {
+  character_list_.insert(character_list_.begin(), "blank");
+}
diff --git a/deploy/cpp_infer/src/modules/text_recognition/processors.h b/deploy/cpp_infer/src/modules/text_recognition/processors.h
new file mode 100644
index 0000000..76aef3a
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/text_recognition/processors.h
@@ -0,0 +1,125 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <algorithm>
+#include <opencv2/opencv.hpp>
+#include <string>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+#include "src/common/processors.h"
+#include "src/utils/func_register.h"
+
+class OCRReisizeNormImg : public BaseProcessor {
+public:
+  OCRReisizeNormImg(
+      absl::optional<std::vector<int>> input_shape = absl::nullopt,
+      std::vector<int> rec_image_shape = {3, 48, 320})
+      : rec_image_shape_(rec_image_shape),
+        input_shape_(input_shape.value_or(std::vector<int>())){};
+  absl::StatusOr<std::vector<cv::Mat>>
+  Apply(std::vector<cv::Mat> &input,
+        const void *param = nullptr) const override;
+  absl::StatusOr<cv::Mat> Resize(cv::Mat &image) const;
+  absl::StatusOr<cv::Mat> StaticResize(cv::Mat &image) const;
+  absl::StatusOr<cv::Mat> ResizeNormImg(cv::Mat &image,
+                                        float max_wh_ratio) const;
+  static constexpr int MAX_IMG_W = 3200;
+
+private:
+  std::vector<int> rec_image_shape_;
+  std::vector<int> input_shape_;
+};
+
+class CTCLabelDecode {
+public:
+  CTCLabelDecode(const std::vector<std::string> &character_list = {},
+                 bool use_space_char = true);
+  absl::StatusOr<std::vector<std::pair<std::string, float>>>
+  Apply(const cv::Mat &preds) const;
+  absl::StatusOr<std::pair<std::string, float>>
+  Process(const cv::Mat &pred_data) const;
+  absl::StatusOr<std::pair<std::string, float>>
+  Decode(std::list<int> &text_index, std::list<float> &text_prob,
+         bool is_remove_duplicate = false) const;
+  void AddSpecialChar();
+
+private:
+  std::vector<std::string> character_list_;
+  bool use_space_char_;
+  std::unordered_map<int, std::string> dict_;
+
+  const std::vector<int> IGNORE_TOKEN = {0};
+};
+
+class ToBatchUniform : public ToBatch {
+public:
+  absl::StatusOr<std::vector<cv::Mat>>
+  Apply(std::vector<cv::Mat> &input,
+        const void *param = nullptr) const override {
+    if (input.empty()) {
+      return absl::InvalidArgumentError("Input image vector is empty.");
+    }
+    int numDims = input[0].dims;
+    int dtype = input[0].type();
+
+    int maxWidth = 0;
+    for (const auto &img : input) {
+      if (img.dims != numDims || img.type() != dtype) {
+        return absl::InvalidArgumentError(
+            "All images must have the same number of dimensions and data type");
+      }
+
+      for (int i = 0; i < numDims - 1; ++i) {
+        if (img.size[i] != input[0].size[i]) {
+          return absl::InvalidArgumentError(
+              "All images must have the same dimensions except width");
+        }
+      }
+      maxWidth = std::max(maxWidth, img.size[numDims - 1]);
+    }
+
+    std::vector<cv::Mat> paddedImages;
+
+    for (const auto &img : input) {
+      int currentWidth = img.size[numDims - 1];
+
+      if (currentWidth == maxWidth) {
+        paddedImages.push_back(img.clone());
+        continue;
+      }
+
+      std::vector<int> newSizes(numDims);
+      for (int i = 0; i < numDims - 1; ++i) {
+        newSizes[i] = img.size[i];
+      }
+      newSizes[numDims - 1] = maxWidth;
+
+      cv::Mat paddedImg(numDims, newSizes.data(), dtype, cv::Scalar::all(0));
+
+      std::vector<cv::Range> srcRanges(numDims, cv::Range::all());
+
+      std::vector<cv::Range> dstRanges(numDims, cv::Range::all());
+      dstRanges[numDims - 1] = cv::Range(0, currentWidth);
+
+      img.copyTo(paddedImg(dstRanges));
+
+      paddedImages.push_back(paddedImg);
+    }
+    return ToBatch::Apply(paddedImages);
+  }
+};
diff --git a/deploy/cpp_infer/src/modules/text_recognition/result.cc b/deploy/cpp_infer/src/modules/text_recognition/result.cc
new file mode 100644
index 0000000..07c9903
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/text_recognition/result.cc
@@ -0,0 +1,138 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "result.h"
+
+#include <fstream>
+
+#ifdef USE_FREETYPE
+#include <opencv2/freetype.hpp>
+#endif
+#include <string>
+
+#include "src/utils/utility.h"
+#include "third_party/nlohmann/json.hpp"
+
+using json = nlohmann::json;
+
+#ifdef USE_FREETYPE
+void TextRecResult::SaveToImg(const std::string &save_path) {
+  int image_width = predictor_result_.input_image.size[1];
+  int image_height = predictor_result_.input_image.size[0];
+  std::string text = predictor_result_.rec_text + "(" +
+                     std::to_string(predictor_result_.rec_score) + ")";
+  int font = AdjustFontSize(image_width, text);
+  cv::Ptr<cv::freetype::FreeType2> ft2 = cv::freetype::createFreeType2();
+  ft2->loadFontData(predictor_result_.vis_font, 0);
+  int baseline = 0;
+  cv::Size text_size = ft2->getTextSize(text, font, -1, &baseline);
+  int row_height = text_size.height;
+  int new_image_height = image_height + static_cast<int>(row_height * 1.2);
+  cv::Mat new_image(new_image_height, image_width, CV_8UC3,
+                    cv::Scalar(255, 255, 255));
+  predictor_result_.input_image.copyTo(
+      new_image(cv::Rect(0, 0, image_width, image_height)));
+  cv::Point org(0, image_height + row_height);
+  ft2->putText(new_image, text, org, font, cv::Scalar(0, 0, 0), -1, cv::LINE_AA,
+               true);
+
+  absl::StatusOr<std::string> full_path;
+  if (predictor_result_.input_path.empty()) {
+    auto now = std::chrono::system_clock::now();
+    auto now_time = std::chrono::system_clock::to_time_t(now);
+    std::stringstream ss;
+    ss << "output_" << std::put_time(std::localtime(&now_time), "%Y%m%d_%H%M%S")
+       << ".jpg";
+    std::string timestamp_filename = ss.str();
+    INFOW("Input path is empty, will use %s instead!",
+          timestamp_filename.c_str());
+    predictor_result_.input_path = timestamp_filename;
+    full_path =
+        Utility::SmartCreateDirectoryForImage(save_path, timestamp_filename);
+  } else {
+    full_path = Utility::SmartCreateDirectoryForImage(
+        save_path, predictor_result_.input_path);
+  }
+  if (!full_path.ok()) {
+    INFOE(full_path.status().ToString().c_str());
+    exit(-1);
+  }
+  bool success = cv::imwrite(full_path.value(), new_image);
+  if (!success) {
+    INFOE("Error: Failed to write the image :%s ", full_path.value().c_str());
+    exit(-1);
+  }
+}
+int TextRecResult::AdjustFontSize(int image_width,
+                                  const std::string &text) const {
+  cv::Ptr<cv::freetype::FreeType2> ft2 = cv::freetype::createFreeType2();
+  int font_size = static_cast<int>(image_width * 0.06);
+
+  ft2->loadFontData(predictor_result_.vis_font, 0);
+
+  cv::Size text_size;
+  int baseline = 0;
+
+  do {
+    text_size = ft2->getTextSize(text, font_size, -1, &baseline);
+    if (text_size.width <= image_width)
+      break;
+    font_size--;
+  } while (font_size > 0);
+
+  return font_size;
+}
+#else
+void TextRecResult::SaveToImg(const std::string &save_path) {
+  INFOW(
+      "OpenCV was not compiled with the freetype module (opencv_freetype), rec "
+      "image will be not saved.");
+}
+#endif
+void TextRecResult::Print() const {
+  std::cout << "{\n  \"res\": {" << std::endl;
+
+  std::cout << "    \"input_path\": {" << predictor_result_.input_path << " },"
+            << std::endl;
+  std::cout << "    \"rec_text\": {" << predictor_result_.rec_text << " }"
+            << std::endl;
+  std::cout << "    \"rec_score\": {" << predictor_result_.rec_score << " }"
+            << std::endl;
+  std::cout << "}" << std::endl;
+}
+
+void TextRecResult::SaveToJson(const std::string &save_path) const {
+  nlohmann::ordered_json j;
+
+  j["input_path"] = predictor_result_.input_path;
+  j["page_index"] = nlohmann::json::value_t::null; //********
+
+  j["rec_text"] = predictor_result_.rec_text;
+  j["rec_score"] = predictor_result_.rec_score;
+
+  auto full_path = Utility::SmartCreateDirectoryForJson(
+      save_path, predictor_result_.input_path);
+  if (!full_path.ok()) {
+    INFOE(full_path.status().ToString().c_str());
+    exit(-1);
+  }
+  std::ofstream file(full_path.value());
+  if (file.is_open()) {
+    file << j.dump(4);
+    file.close();
+  } else {
+    INFOE("Could not open file for writing: %s", save_path.c_str());
+    exit(-1);
+  }
+}
diff --git a/deploy/cpp_infer/src/modules/text_recognition/result.h b/deploy/cpp_infer/src/modules/text_recognition/result.h
new file mode 100644
index 0000000..0ad165a
--- /dev/null
+++ b/deploy/cpp_infer/src/modules/text_recognition/result.h
@@ -0,0 +1,35 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <opencv2/opencv.hpp>
+#include <vector>
+
+#include "predictor.h"
+#include "src/base/base_cv_result.h"
+
+class TextRecResult : public BaseCVResult {
+public:
+  TextRecResult(TextRecPredictorResult predictor_result)
+      : BaseCVResult(), predictor_result_(predictor_result){};
+  // std::unordered_map<std::string, cv::Mat> ToImg() const override;
+  void SaveToImg(const std::string &save_path) override;
+  void Print() const override;
+  void SaveToJson(const std::string &save_path) const override;
+  int AdjustFontSize(int image_width, const std::string &text) const;
+
+private:
+  TextRecPredictorResult predictor_result_;
+};
diff --git a/deploy/cpp_infer/src/pipelines/doc_preprocessor/pipeline.cc b/deploy/cpp_infer/src/pipelines/doc_preprocessor/pipeline.cc
new file mode 100644
index 0000000..dfc7ce1
--- /dev/null
+++ b/deploy/cpp_infer/src/pipelines/doc_preprocessor/pipeline.cc
@@ -0,0 +1,367 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "pipeline.h"
+
+#include "result.h"
+#include "src/modules/image_classification/predictor.h"
+#include "src/modules/image_unwarping/predictor.h"
+
+_DocPreprocessorPipeline::_DocPreprocessorPipeline(
+    const DocPreprocessorPipelineParams &params)
+    : BasePipeline(), params_(params) {
+  if (params.paddlex_config.has_value()) {
+    if (params.paddlex_config.value().IsStr()) {
+      config_ = YamlConfig(params.paddlex_config.value().GetStr());
+    } else {
+      config_ = YamlConfig(params.paddlex_config.value().GetMap());
+    }
+  } else {
+    auto config_path = Utility::GetDefaultConfig("doc_preprocessor");
+    if (!config_path.ok()) {
+      INFOE("Could not find doc_preprocessors pipeline config file : %s",
+            config_path.status().ToString().c_str());
+      exit(-1);
+    }
+    config_ = YamlConfig(config_path.value());
+  }
+  OverrideConfig();
+  auto result_doc = config_.GetBool("use_doc_orientation_classify", true);
+  if (!result_doc.ok()) {
+    INFOE("use_doc_orientation_classify set fail : %s",
+          result_doc.status().ToString().c_str());
+    exit(-1);
+  }
+  use_doc_orientation_classify_ = result_doc.value();
+  auto result_batch = config_.GetInt("batch_size", 1);
+  if (!result_batch.ok()) {
+    INFOE("batch_size get fail: %s", result_batch.status().ToString().c_str());
+    exit(-1);
+  }
+
+  if (use_doc_orientation_classify_) {
+    ClasPredictorParams doc_ori_classify_params;
+    auto result_model_dir =
+        config_.GetString("DocOrientationClassify.model_dir");
+    if (!result_model_dir.ok()) {
+      INFOE("Could not find DocOrientationClassify model dir : %s",
+            result_model_dir.status().ToString().c_str());
+      exit(-1);
+    }
+    auto result_model_name =
+        config_.GetString("DocOrientationClassify.model_name");
+    if (!result_model_name.ok()) {
+      INFOE("Could not find DocOrientationClassify model name : %s",
+            result_model_name.status().ToString().c_str());
+      exit(-1);
+    }
+    doc_ori_classify_params.model_dir = result_model_dir.value();
+    doc_ori_classify_params.model_name = result_model_name.value();
+    doc_ori_classify_params.device = params_.device;
+    doc_ori_classify_params.precision = params_.precision;
+    doc_ori_classify_params.enable_mkldnn = params_.enable_mkldnn;
+    doc_ori_classify_params.mkldnn_cache_capacity =
+        params_.mkldnn_cache_capacity;
+    doc_ori_classify_params.cpu_threads = params_.cpu_threads;
+    doc_ori_classify_params.batch_size = result_batch.value();
+
+    doc_ori_classify_model_ =
+        CreateModule<ClasPredictor>(doc_ori_classify_params);
+  }
+
+  auto result_unwarping = config_.GetBool("use_doc_unwarping", true);
+  if (!result_unwarping.ok()) {
+    INFOE("use_doc_unwarping get fail:%s",
+          result_unwarping.status().ToString().c_str());
+    exit(-1);
+  }
+  use_doc_unwarping_ = result_unwarping.value();
+
+  if (use_doc_unwarping_) {
+    WarpPredictorParams doc_unwarping_params;
+    auto result_model_dir = config_.GetString("DocUnwarping.model_dir");
+    if (!result_model_dir.ok()) {
+      INFOE("Could not find DocUnwarping model dir : %s",
+            result_model_dir.status().ToString().c_str());
+      exit(-1);
+    }
+    auto result_model_name = config_.GetString("DocUnwarping.model_name");
+    if (!result_model_name.ok()) {
+      INFOE("Could not find DocUnwarping model name : %s",
+            result_model_name.status().ToString().c_str());
+      exit(-1);
+    }
+    doc_unwarping_params.model_dir = result_model_dir.value();
+    doc_unwarping_params.model_name = result_model_name.value();
+    doc_unwarping_params.device = params_.device;
+    doc_unwarping_params.precision = params_.precision;
+    doc_unwarping_params.enable_mkldnn = params_.enable_mkldnn;
+    doc_unwarping_params.mkldnn_cache_capacity = params_.mkldnn_cache_capacity;
+    doc_unwarping_params.cpu_threads = params_.cpu_threads;
+    doc_unwarping_params.batch_size = result_batch.value();
+
+    doc_unwarping_model_ = CreateModule<WarpPredictor>(doc_unwarping_params);
+  }
+
+  batch_sampler_ptr_ = std::unique_ptr<BaseBatchSampler>(
+      new ImageBatchSampler(result_batch.value()));
+};
+
+std::vector<std::unique_ptr<BaseCVResult>>
+_DocPreprocessorPipeline::Predict(const std::vector<std::string> &input) {
+  auto model_setting = GetModelSettings();
+  auto status = CheckModelSettingsVaild(model_setting);
+  if (!status.ok()) {
+    INFOE("the input params for model settings are invalid!: %s",
+          status.ToString().c_str());
+    exit(-1);
+  }
+  auto batches = batch_sampler_ptr_->Apply(input);
+  if (!batches.ok()) {
+    INFOE("pipeline get sample fail : %s", batches.status().ToString().c_str());
+    exit(-1);
+  }
+  auto input_path = batch_sampler_ptr_->InputPath();
+  int index = 0;
+  std::vector<cv::Mat> origin_image = {};
+
+  std::vector<std::unique_ptr<BaseCVResult>> base_cv_result_ptr_vec = {};
+  std::vector<DocPreprocessorPipelineResult> pipeline_result_vec = {};
+  pipeline_result_vec_.clear();
+  for (auto &batch_data : batches.value()) {
+    origin_image.reserve(batch_data.size());
+    for (const auto &mat : batch_data) {
+      origin_image.push_back(mat.clone());
+    }
+    std::vector<int> angles = {};
+    std::vector<cv::Mat> rotate_images = {};
+    if (model_setting["use_doc_orientation_classify"]) {
+      doc_ori_classify_model_->Predict(batch_data);
+      ClasPredictor *derived =
+          static_cast<ClasPredictor *>(doc_ori_classify_model_.get());
+      std::vector<ClasPredictorResult> preds = derived->PredictorResult();
+      for (auto &pred : preds) {
+        auto result_angle = Utility::StringToInt(pred.label_names[0]);
+        if (!result_angle.ok()) {
+          INFOE("angle is invalid : %s",
+                result_angle.status().ToString().c_str());
+          exit(-1);
+        }
+        angles.push_back(result_angle.value());
+        auto result_rotate = ComponentsProcessor::RotateImage(
+            pred.input_image, result_angle.value());
+        if (!result_rotate.ok()) {
+          INFOE("RotateImage fail : %s",
+                result_rotate.status().ToString().c_str());
+          exit(-1);
+        }
+        rotate_images.push_back(result_rotate.value());
+      }
+    } else {
+      angles = std::vector<int>(batch_data.size(), -1);
+      rotate_images = batch_data;
+    }
+    std::vector<cv::Mat> output_imgs = {};
+    if (model_setting["use_doc_unwarping"]) {
+      doc_unwarping_model_->Predict(rotate_images);
+      WarpPredictor *derived =
+          static_cast<WarpPredictor *>(doc_unwarping_model_.get());
+      std::vector<WarpPredictorResult> preds = derived->PredictorResult();
+      for (auto &pred : preds) {
+        output_imgs.push_back(pred.doctr_img); //***"RGB" "BGR"
+      }
+    } else {
+      output_imgs = rotate_images;
+    }
+
+    pipeline_result_vec.clear();
+    for (int i = 0; i < output_imgs.size(); i++, index++) {
+      DocPreprocessorPipelineResult pipeline_result;
+      pipeline_result.input_path = input_path[index];
+      pipeline_result.input_image = origin_image[i];
+      pipeline_result.model_settings = model_setting;
+      pipeline_result.angle = angles[i];
+      pipeline_result.rotate_image = rotate_images[i];
+      pipeline_result.output_image = output_imgs[i];
+      pipeline_result_vec.push_back(pipeline_result);
+    }
+    origin_image.clear();
+    pipeline_result_vec_.insert(pipeline_result_vec_.end(),
+                                pipeline_result_vec.begin(),
+                                pipeline_result_vec.end());
+    for (auto &pipeline_result : pipeline_result_vec) {
+      std::unique_ptr<BaseCVResult> base_cv_result_ptr =
+          std::unique_ptr<BaseCVResult>(
+              new DocPreprocessorResult(pipeline_result));
+      base_cv_result_ptr_vec.emplace_back(std::move(base_cv_result_ptr));
+    }
+  }
+  return base_cv_result_ptr_vec;
+};
+
+std::unordered_map<std::string, bool>
+_DocPreprocessorPipeline::GetModelSettings(
+    absl::optional<bool> use_doc_orientation_classify,
+    absl::optional<bool> use_doc_unwarping) const {
+  if (!use_doc_orientation_classify.has_value()) {
+    use_doc_orientation_classify = use_doc_orientation_classify_;
+  }
+  if (!use_doc_unwarping.has_value()) {
+    use_doc_unwarping = use_doc_unwarping_;
+  }
+  std::unordered_map<std::string, bool> model_settings = {};
+  model_settings["use_doc_orientation_classify"] =
+      use_doc_orientation_classify.value();
+  model_settings["use_doc_unwarping"] = use_doc_unwarping.value();
+  return model_settings;
+};
+
+absl::Status _DocPreprocessorPipeline::CheckModelSettingsVaild(
+    std::unordered_map<std::string, bool> model_settings) const {
+  if (model_settings["use_doc_orientation_classify"] &&
+      !use_doc_orientation_classify_) {
+    return absl::InvalidArgumentError(
+        "Set use_doc_orientation_classify, but the model for doc orientation "
+        "classify is not initialized.");
+  }
+
+  if (model_settings["use_doc_unwarping"] && !use_doc_unwarping_) {
+    return absl::InvalidArgumentError(
+        "Set use_doc_unwarping, but the model for doc unwarping is not "
+        "initialized.");
+  }
+  return absl::OkStatus();
+}
+
+std::vector<std::unique_ptr<BaseCVResult>>
+DocPreprocessorPipeline::Predict(const std::vector<std::string> &input) {
+  if (thread_num_ == 1) {
+    return infer_->Predict(input);
+  }
+  batch_sampler_ptr_ =
+      std::unique_ptr<BaseBatchSampler>(new ImageBatchSampler(1));
+  auto nomeaning = batch_sampler_ptr_->Apply(input);
+  int input_num = nomeaning.value().size();
+  if (thread_num_ > input_num) {
+    INFOW("thread num exceed input num, will set %d", input_num);
+    thread_num_ = input_num;
+  }
+  int infer_batch_num = input_num / thread_num_;
+  auto status = batch_sampler_ptr_->SetBatchSize(infer_batch_num);
+  if (!status.ok()) {
+    INFOE("Set batch size fail : %s", status.ToString().c_str());
+    exit(-1);
+  }
+  auto infer_batch_data =
+      batch_sampler_ptr_->SampleFromVectorToStringVector(input);
+  if (!infer_batch_data.ok()) {
+    INFOE("Get infer batch data fail : %s",
+          infer_batch_data.status().ToString().c_str());
+    exit(-1);
+  }
+  std::vector<std::unique_ptr<BaseCVResult>> results = {};
+  results.reserve(input_num);
+  for (auto &infer_data : infer_batch_data.value()) {
+    auto status =
+        AutoParallelSimpleInferencePipeline::PredictThread(infer_data);
+    if (!status.ok()) {
+      INFOE("Infer fail : %s", status.ToString().c_str());
+      exit(-1);
+    }
+  }
+  for (int i = 0; i < infer_batch_data.value().size(); i++) {
+    auto infer_data_result = GetResult();
+    if (!infer_data_result.ok()) {
+      INFOE("Get infer result fail : %s",
+            infer_batch_data.status().ToString().c_str());
+      exit(-1);
+    }
+    results.insert(results.end(),
+                   std::make_move_iterator(infer_data_result.value().begin()),
+                   std::make_move_iterator(infer_data_result.value().end()));
+  }
+  return results;
+}
+
+void _DocPreprocessorPipeline::OverrideConfig() {
+  auto &data = config_.Data();
+  if (params_.doc_orientation_classify_model_name.has_value()) {
+    auto it = config_.FindKey("DocOrientationClassify.model_name");
+    if (!it.ok()) {
+      data["DocPreprocessor.SubModules.DocOrientationClassify."
+           "model_name"] = params_.doc_orientation_classify_model_name.value();
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.doc_orientation_classify_model_name.value();
+    }
+  }
+  if (params_.doc_orientation_classify_model_dir.has_value()) {
+    auto it = config_.FindKey("DocOrientationClassify.model_dir");
+    if (!it.ok()) {
+      data["DocPreprocessor.SubModules.DocOrientationClassify."
+           "model_dir"] = params_.doc_orientation_classify_model_dir.value();
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.doc_orientation_classify_model_dir.value();
+    }
+  }
+  if (params_.doc_unwarping_model_name.has_value()) {
+    auto it = config_.FindKey("DocUnwarping.model_name");
+    if (!it.ok()) {
+      data["DocPreprocessor.SubModules.DocUnwarping.model_name"] =
+          params_.doc_unwarping_model_name.value();
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.doc_unwarping_model_name.value();
+    }
+  }
+  if (params_.doc_unwarping_model_dir.has_value()) {
+    auto it = config_.FindKey("DocUnwarping.model_dir");
+    if (!it.ok()) {
+      data["DocPreprocessor.SubModules.DocUnwarping.model_dir"] =
+          params_.doc_unwarping_model_dir.value();
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.doc_unwarping_model_dir.value();
+    }
+  }
+
+  if (params_.use_doc_orientation_classify.has_value()) {
+    auto it = config_.FindKey("DocPreprocessor.use_doc_orientation_classify");
+    if (!it.ok()) {
+      data["DocPreprocessor.use_doc_orientation_classify"] =
+          params_.use_doc_orientation_classify.value() ? "true" : "false";
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] =
+          params_.use_doc_orientation_classify.value() ? "true" : "false";
+    }
+  }
+  if (params_.use_doc_unwarping.has_value()) {
+    auto it = config_.FindKey("DocPreprocessor.use_doc_unwarping");
+    if (!it.ok()) {
+      data["DocPreprocessor.use_doc_unwarping"] =
+          params_.use_doc_unwarping.value() ? "true" : "false";
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.use_doc_unwarping.value() ? "true" : "false";
+    }
+  }
+}
diff --git a/deploy/cpp_infer/src/pipelines/doc_preprocessor/pipeline.h b/deploy/cpp_infer/src/pipelines/doc_preprocessor/pipeline.h
new file mode 100644
index 0000000..bc53236
--- /dev/null
+++ b/deploy/cpp_infer/src/pipelines/doc_preprocessor/pipeline.h
@@ -0,0 +1,115 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+#pragma once
+
+#include <iostream>
+#include <string>
+#include <unordered_map>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+#include "absl/types/optional.h"
+#include "src/base/base_pipeline.h"
+#include "src/common/image_batch_sampler.h"
+#include "src/common/parallel.h"
+#include "src/common/processors.h"
+#include "src/utils/ilogger.h"
+#include "src/utils/utility.h"
+
+struct DocPreprocessorPipelineResult {
+  std::string input_path = "";
+  cv::Mat input_image;
+  std::unordered_map<std::string, bool> model_settings;
+  int angle = 0;
+  cv::Mat rotate_image;
+  cv::Mat output_image;
+  cv::Mat image_all;
+};
+
+struct DocPreprocessorPipelineParams {
+  absl::optional<std::string> doc_orientation_classify_model_name =
+      absl::nullopt;
+  absl::optional<std::string> doc_orientation_classify_model_dir =
+      absl::nullopt;
+  absl::optional<std::string> doc_unwarping_model_name = absl::nullopt;
+  absl::optional<std::string> doc_unwarping_model_dir = absl::nullopt;
+  absl::optional<bool> use_doc_orientation_classify = absl::nullopt;
+  absl::optional<bool> use_doc_unwarping = absl::nullopt;
+  absl::optional<std::string> device = absl::nullopt;
+  bool enable_mkldnn = true;
+  int mkldnn_cache_capacity = 10;
+  std::string precision = "fp32";
+  int cpu_threads = 8;
+  int thread_num = 1;
+  absl::optional<Utility::PaddleXConfigVariant> paddlex_config = absl::nullopt;
+};
+
+class _DocPreprocessorPipeline : public BasePipeline {
+public:
+  explicit _DocPreprocessorPipeline(
+      const DocPreprocessorPipelineParams &params);
+  virtual ~_DocPreprocessorPipeline() = default;
+
+  _DocPreprocessorPipeline() = delete;
+
+  std::vector<std::unique_ptr<BaseCVResult>>
+  Predict(const std::vector<std::string> &input) override;
+
+  std::unordered_map<std::string, bool> GetModelSettings(
+      absl::optional<bool> use_doc_orientation_classify = absl::nullopt,
+      absl::optional<bool> use_doc_unwarping = absl::nullopt) const;
+  absl::Status CheckModelSettingsVaild(
+      std::unordered_map<std::string, bool> model_settings) const;
+
+  std::vector<DocPreprocessorPipelineResult> PipelineResult() const {
+    return pipeline_result_vec_;
+  };
+
+  void OverrideConfig();
+
+private:
+  bool use_doc_orientation_classify_;
+  bool use_doc_unwarping_;
+  std::unique_ptr<BasePredictor> doc_ori_classify_model_;
+  std::unique_ptr<BasePredictor> doc_unwarping_model_;
+  DocPreprocessorPipelineParams params_;
+  YamlConfig config_;
+  std::unique_ptr<BaseBatchSampler> batch_sampler_ptr_;
+  std::vector<DocPreprocessorPipelineResult> pipeline_result_vec_;
+};
+
+class DocPreprocessorPipeline
+    : public AutoParallelSimpleInferencePipeline<
+          _DocPreprocessorPipeline, DocPreprocessorPipelineParams,
+          std::vector<std::string>,
+          std::vector<std::unique_ptr<BaseCVResult>>> {
+public:
+  DocPreprocessorPipeline(const DocPreprocessorPipelineParams &params)
+      : AutoParallelSimpleInferencePipeline(params),
+        thread_num_(params.thread_num) {
+    if (thread_num_ == 1) {
+      infer_ =
+          std::unique_ptr<BasePipeline>(new _DocPreprocessorPipeline(params));
+    }
+  };
+
+  std::vector<std::unique_ptr<BaseCVResult>>
+  Predict(const std::vector<std::string> &input) override;
+
+private:
+  int thread_num_;
+  std::unique_ptr<BasePipeline> infer_;
+  std::unique_ptr<BaseBatchSampler> batch_sampler_ptr_;
+};
diff --git a/deploy/cpp_infer/src/pipelines/doc_preprocessor/result.cc b/deploy/cpp_infer/src/pipelines/doc_preprocessor/result.cc
new file mode 100644
index 0000000..dce2ab5
--- /dev/null
+++ b/deploy/cpp_infer/src/pipelines/doc_preprocessor/result.cc
@@ -0,0 +1,129 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "result.h"
+
+#include <algorithm>
+#include <fstream>
+#include <string>
+
+#include "src/utils/utility.h"
+#include "third_party/nlohmann/json.hpp"
+
+using json = nlohmann::json;
+void DocPreprocessorResult::SaveToImg(const std::string &save_path) {
+  cv::Mat input_img = pipeline_result_.input_image.clone();
+  cv::Mat rot_img = pipeline_result_.rotate_image.clone();
+  cv::Mat output_img = pipeline_result_.output_image.clone();
+  bool use_doc_orientation_classify =
+      pipeline_result_.model_settings.at("use_doc_orientation_classify");
+  bool use_doc_unwarping =
+      pipeline_result_.model_settings.at("use_doc_unwarping");
+  int angle = pipeline_result_.angle;
+  int h1 = input_img.size[0], w1 = input_img.size[1];
+  int h2 = rot_img.size[0], w2 = rot_img.size[1];
+  int h3 = output_img.size[0], w3 = output_img.size[1];
+  int h = std::max(h1, std::max(h2, h3));
+  int total_w = w1 + w2 + w3;
+  int final_h = h + 25;
+
+  cv::Mat img_show(final_h, total_w, CV_8UC3, cv::Scalar(255, 255, 255));
+
+  input_img.copyTo(img_show(cv::Rect(0, 0, w1, h1)));
+  rot_img.copyTo(img_show(cv::Rect(w1, 0, w2, h2)));
+  output_img.copyTo(img_show(cv::Rect(w1 + w2, 0, w3, h3)));
+  pipeline_result_.image_all = img_show.clone();
+  std::vector<std::string> txt_list = {
+      "Original Image",
+      "Rotated Image (" +
+          std::string(use_doc_orientation_classify ? "True" : "False") + ", " +
+          std::to_string(angle) + ")",
+      "Unwarping Image (" + std::string(use_doc_unwarping ? "True" : "False") +
+          ")"};
+  std::vector<int> region_w_list = {w1, w2, w3};
+  std::vector<int> beg_w_list = {0, w1, w1 + w2};
+  for (int tno = 0; tno < 3; ++tno) {
+    DrawText(img_show, txt_list[tno], beg_w_list[tno], h, region_w_list[tno]);
+  }
+  auto full_path = Utility::SmartCreateDirectoryForImage(
+      save_path, pipeline_result_.input_path);
+  if (!full_path.ok()) {
+    INFOE(full_path.status().ToString().c_str());
+    exit(-1);
+  }
+  bool success = cv::imwrite(full_path.value(), img_show);
+  if (!success) {
+    INFOE("Error: Failed to write the image : %s", full_path.value().c_str());
+    exit(-1);
+  }
+}
+
+void DocPreprocessorResult::Print() const {
+  std::cout << "{\n  \"res\": {" << std::endl;
+
+  std::cout << "    \"input_path\": {" << pipeline_result_.input_path << "},"
+            << std::endl;
+  std::cout << "    \"model_settings\": {"
+            << "use_doc_orientation_classify: " +
+                   std::string(pipeline_result_.model_settings.at(
+                                   "use_doc_orientation_classify")
+                                   ? "True"
+                                   : "False")
+            << ", use_doc_unwarping: " +
+                   std::string(
+                       pipeline_result_.model_settings.at("use_doc_unwarping")
+                           ? "True"
+                           : "False")
+            << "}," << std::endl;
+  std::cout << "    \"angle\": {" << pipeline_result_.angle << "},"
+            << std::endl;
+  std::cout << "}" << std::endl;
+}
+
+void DocPreprocessorResult::SaveToJson(const std::string &save_path) const {
+  nlohmann::ordered_json j;
+
+  j["input_path"] = pipeline_result_.input_path;
+  j["page_index"] = nullptr; //********
+  j["model_settings"] = pipeline_result_.model_settings;
+  j["angle"] = pipeline_result_.angle;
+
+  auto full_path = Utility::SmartCreateDirectoryForJson(
+      save_path, pipeline_result_.input_path);
+  if (!full_path.ok()) {
+    INFOE(full_path.status().ToString().c_str());
+    exit(-1);
+  }
+  std::ofstream file(full_path.value());
+  if (file.is_open()) {
+    file << j.dump(4);
+    file.close();
+  } else {
+    INFOE("Could not open file for writing : %s", save_path.c_str());
+    exit(-1);
+  }
+}
+
+void DocPreprocessorResult::DrawText(cv::Mat &img, const std::string &text,
+                                     int x, int y, int width) {
+  int fontFace = cv::FONT_HERSHEY_SIMPLEX;
+  double fontScale = 0.7;
+  int thickness = 2;
+  int baseline = 0;
+  cv::Size textSize =
+      cv::getTextSize(text, fontFace, fontScale, thickness, &baseline);
+
+  putText(img, text, cv::Point(x + 10, y + textSize.height + 2), fontFace,
+          fontScale, cv::Scalar(0, 0, 0), thickness, cv::LINE_AA);
+}
diff --git a/deploy/cpp_infer/src/pipelines/doc_preprocessor/result.h b/deploy/cpp_infer/src/pipelines/doc_preprocessor/result.h
new file mode 100644
index 0000000..de4ad93
--- /dev/null
+++ b/deploy/cpp_infer/src/pipelines/doc_preprocessor/result.h
@@ -0,0 +1,33 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include "pipeline.h"
+#include "src/base/base_cv_result.h"
+
+class DocPreprocessorResult : public BaseCVResult {
+public:
+  DocPreprocessorResult(DocPreprocessorPipelineResult pipeline_result_)
+      : BaseCVResult(), pipeline_result_(pipeline_result_){};
+
+  void SaveToImg(const std::string &save_path) override;
+  void Print() const override;
+  void SaveToJson(const std::string &save_path) const override;
+  static void DrawText(cv::Mat &img, const std::string &text, int x, int y,
+                       int width);
+
+private:
+  DocPreprocessorPipelineResult pipeline_result_;
+};
diff --git a/deploy/cpp_infer/src/pipelines/ocr/pipeline.cc b/deploy/cpp_infer/src/pipelines/ocr/pipeline.cc
new file mode 100644
index 0000000..0678db7
--- /dev/null
+++ b/deploy/cpp_infer/src/pipelines/ocr/pipeline.cc
@@ -0,0 +1,767 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "pipeline.h"
+
+#include "result.h"
+#include "src/utils/args.h"
+_OCRPipeline::_OCRPipeline(const OCRPipelineParams &params)
+    : BasePipeline(), params_(params) {
+  if (params.paddlex_config.has_value()) {
+    if (params.paddlex_config.value().IsStr()) {
+      config_ = YamlConfig(params.paddlex_config.value().GetStr());
+    } else {
+      config_ = YamlConfig(params.paddlex_config.value().GetMap());
+    }
+  } else {
+    auto config_path = Utility::GetDefaultConfig("OCR");
+    if (!config_path.ok()) {
+      INFOE("Could not find OCR pipeline config file: %s",
+            config_path.status().ToString().c_str());
+      exit(-1);
+    }
+    config_ = YamlConfig(config_path.value());
+  }
+  OverrideConfig();
+  auto result_use_doc_orientation_classify =
+      config_.GetBool("use_doc_orientation_classify", true);
+  if (!result_use_doc_orientation_classify.ok()) {
+    INFOE("use_doc_orientation_classify config error : %s",
+          result_use_doc_orientation_classify.status().ToString().c_str());
+    exit(-1);
+  }
+  auto result_use_use_doc_unwarping =
+      config_.GetBool("use_doc_unwarping", true);
+  if (!result_use_use_doc_unwarping.ok()) {
+    INFOE("use_doc_unwarping config error : %s",
+          result_use_use_doc_unwarping.status().ToString().c_str());
+    exit(-1);
+  }
+  if (result_use_doc_orientation_classify.value() ||
+      result_use_use_doc_unwarping.value()) {
+    use_doc_preprocessor_ = true;
+  } else {
+    use_doc_preprocessor_ = false;
+  }
+  if (use_doc_preprocessor_) {
+    auto result_doc_preprocessor_config = config_.GetSubModule("SubPipelines");
+    if (!result_doc_preprocessor_config.ok()) {
+      INFOE("Get doc preprocessors subpipelines config fail : ",
+            result_doc_preprocessor_config.status().ToString().c_str());
+      exit(-1);
+    }
+    DocPreprocessorPipelineParams params;
+    params.device = params_.device;
+    params.precision = params_.precision;
+    params.enable_mkldnn = params_.enable_mkldnn;
+    params.mkldnn_cache_capacity = params_.mkldnn_cache_capacity;
+    params.cpu_threads = params_.cpu_threads;
+    params.paddlex_config = result_doc_preprocessor_config.value();
+    doc_preprocessors_pipeline_ =
+        CreatePipeline<_DocPreprocessorPipeline>(params);
+
+    use_doc_orientation_classify_ =
+        config_.GetBool("DocPreprocessor.use_doc_orientation_classify", true)
+            .value();
+    use_doc_unwarping_ =
+        config_.GetBool("DocPreprocessor.use_doc_unwarping", true).value();
+  }
+  auto result_use_textline_orientation =
+      config_.GetBool("use_textline_orientation", true);
+  if (!result_use_textline_orientation.ok()) {
+    INFOE("use_textline_orientation config error : %s",
+          result_use_textline_orientation.status().ToString().c_str());
+    exit(-1);
+  }
+  use_textline_orientation_ = result_use_textline_orientation.value();
+  if (use_textline_orientation_) {
+    ClasPredictorParams params;
+    params.device = params_.device;
+    params.precision = params_.precision;
+    params.enable_mkldnn = params_.enable_mkldnn;
+    params.mkldnn_cache_capacity = params_.mkldnn_cache_capacity;
+    params.cpu_threads = params_.cpu_threads;
+    auto result_batch_size =
+        config_.GetInt("TextLineOrientation.batch_size", 1);
+    if (!result_batch_size.ok()) {
+      INFOE("Get TextLineOrientation batch size fail: %s",
+            result_batch_size.status().ToString().c_str());
+      exit(-1);
+    }
+    params.batch_size = result_batch_size.value();
+
+    auto result_model_name =
+        config_.GetString("TextLineOrientation.model_name");
+    if (!result_model_name.ok()) {
+      INFOE("Could not find TextLineOrientation model name : %s",
+            result_model_name.status().ToString().c_str());
+      exit(-1);
+    }
+    params.model_name = result_model_name.value();
+    auto result_model_dir = config_.GetString("TextLineOrientation.model_dir");
+    if (!result_model_dir.ok()) {
+      INFOE("Could not find TextLineOrientation model dir : %s",
+            result_model_dir.status().ToString().c_str());
+      exit(-1);
+    }
+    params.model_dir = result_model_dir.value();
+    textline_orientation_model_ = CreateModule<ClasPredictor>(params);
+  }
+  auto text_type = config_.GetString("text_type");
+  if (!text_type.ok()) {
+    INFOE("Get text type fail : %s", text_type.status().ToString().c_str());
+    exit(-1);
+  }
+  text_type_ = text_type.value();
+  TextDetPredictorParams params_det;
+  auto result_text_det_model_name =
+      config_.GetString("TextDetection.model_name");
+  if (!result_text_det_model_name.ok()) {
+    INFOE("Could not find TextDetection model name : %s",
+          result_text_det_model_name.status().ToString().c_str());
+    exit(-1);
+  }
+  params_det.model_name = result_text_det_model_name.value();
+  auto result_text_det_model_dir = config_.GetString("TextDetection.model_dir");
+  if (!result_text_det_model_dir.ok()) {
+    INFOE("Could not find TextDetection model dir : %s",
+          result_text_det_model_dir.status().ToString().c_str());
+    exit(-1);
+  }
+  params_det.model_dir = result_text_det_model_dir.value();
+  auto result_det_input_shape = config_.GetString("TextDetection.input_shape");
+  if (!result_det_input_shape.value().empty()) {
+    params_det.input_shape =
+        config_.SmartParseVector(result_det_input_shape.value()).vec_int;
+  }
+  params_det.device = params_.device;
+  params_det.precision = params_.precision;
+  params_det.enable_mkldnn = params_.enable_mkldnn;
+  params_det.mkldnn_cache_capacity = params_.mkldnn_cache_capacity;
+  params_det.cpu_threads = params_.cpu_threads;
+  params_det.batch_size = config_.GetInt("TextDetection.batch_size", 1).value();
+  if (text_type_ == "general") {
+    params_det.limit_side_len =
+        config_.GetInt("TextDetection.limit_side_len", 960).value();
+    params_det.limit_type =
+        config_.GetString("TextDetection.limit_type", "max").value();
+    params_det.max_side_limit =
+        config_.GetInt("TextDetection.max_side_limit", 4000).value();
+    params_det.thresh = config_.GetFloat("TextDetection.thresh", 0.3).value();
+    params_det.box_thresh =
+        config_.GetFloat("TextDetection.box_thresh", 0.6).value();
+    params_det.unclip_ratio =
+        config_.GetFloat("TextDetection.unclip_ratio", 2.0).value();
+    sort_boxes_ = ComponentsProcessor::SortQuadBoxes;
+    crop_by_polys_ = std::unique_ptr<CropByPolys>(new CropByPolys("quad"));
+  } else if (text_type_ == "seal") {
+    params_det.limit_side_len =
+        config_.GetInt("TextDetection.limit_side_len", 736).value();
+    params_det.limit_type =
+        config_.GetString("TextDetection.limit_type", "min").value();
+    params_det.max_side_limit =
+        config_.GetInt("TextDetection.max_side_limit", 4000).value();
+    params_det.thresh = config_.GetFloat("TextDetection.thresh", 0.2).value();
+    params_det.box_thresh =
+        config_.GetFloat("TextDetection.box_thresh", 0.6).value();
+    params_det.unclip_ratio =
+        config_.GetFloat("TextDetection.unclip_ratio", 0.5).value();
+    sort_boxes_ = ComponentsProcessor::SortPolyBoxes;
+    crop_by_polys_ = std::unique_ptr<CropByPolys>(new CropByPolys("poly"));
+  } else {
+    INFOE("Unsupported text type We %s", text_type.value().c_str());
+    exit(-1);
+  }
+  text_det_model_ = CreateModule<TextDetPredictor>(params_det);
+
+  text_det_params_.text_det_limit_side_len = params_det.limit_side_len.value();
+  text_det_params_.text_det_limit_type = params_det.limit_type.value();
+  text_det_params_.text_det_max_side_limit = params_det.max_side_limit.value();
+  text_det_params_.text_det_thresh = params_det.thresh.value();
+  text_det_params_.text_det_box_thresh = params_det.box_thresh.value();
+  text_det_params_.text_det_unclip_ratio = params_det.unclip_ratio.value();
+
+  TextRecPredictorParams params_rec;
+  auto result_text_rec_model_name =
+      config_.GetString("TextRecognition.model_name");
+  if (!result_text_rec_model_name.ok()) {
+    INFOE("Could not find TextRecognition model name : %s",
+          result_text_rec_model_name.status().ToString().c_str());
+    exit(-1);
+  }
+  params_rec.model_name = result_text_rec_model_name.value();
+  auto result_text_rec_model_dir =
+      config_.GetString("TextRecognition.model_dir");
+  if (!result_text_rec_model_dir.ok()) {
+    INFOE("Could not find TextRecognition model dir : %s",
+          result_text_rec_model_dir.status().ToString().c_str());
+    exit(-1);
+  }
+  auto result_rec_input_shape =
+      config_.GetString("TextRecognition.input_shape");
+  if (!result_rec_input_shape.value().empty()) {
+    params_rec.input_shape =
+        config_.SmartParseVector(result_rec_input_shape.value()).vec_int;
+  }
+  params_rec.model_dir = result_text_rec_model_dir.value();
+  params_rec.lang = params_.lang;
+  params_rec.ocr_version = params_.ocr_version;
+  params_rec.vis_font_dir = params_.vis_font_dir;
+  params_rec.device = params_.device;
+  params_rec.precision = params_.precision;
+  params_rec.enable_mkldnn = params_.enable_mkldnn;
+  params_rec.mkldnn_cache_capacity = params_.mkldnn_cache_capacity;
+  params_rec.cpu_threads = params_.cpu_threads;
+  params_rec.batch_size =
+      config_.GetInt("TextRecognition.batch_size", 1).value();
+
+  text_rec_model_ = CreateModule<TextRecPredictor>(params_rec);
+  text_rec_score_thresh_ =
+      config_.GetFloat("TextRecognition.score_thresh", 0.0).value();
+
+  batch_sampler_ptr_ = std::unique_ptr<BaseBatchSampler>(
+      new ImageBatchSampler(1)); //** pipeline batch_size
+};
+
+absl::StatusOr<std::vector<cv::Mat>>
+_OCRPipeline::RotateImage(const std::vector<cv::Mat> &image_array_list,
+                          const std::vector<int> &rotate_angle_list) {
+  if (image_array_list.size() != rotate_angle_list.size()) {
+    return absl::InvalidArgumentError(
+        "Length of image_array_list (" +
+        std::to_string(image_array_list.size()) +
+        ") must match length of rotate_angle_list (" +
+        std::to_string(rotate_angle_list.size()) + ")");
+  }
+  std::vector<cv::Mat> rotated_images;
+  rotated_images.reserve(image_array_list.size());
+  for (std::size_t i = 0; i < image_array_list.size(); ++i) {
+    int angle_indicator = rotate_angle_list[i];
+    if (angle_indicator != 0 && angle_indicator != 1) {
+      return absl::InvalidArgumentError(
+          "rotate_angle must be 0 or 1, now it's: " +
+          std::to_string(angle_indicator));
+    }
+    int rotate_angle = angle_indicator * 180;
+    auto result_rotated_image =
+        ComponentsProcessor::RotateImage(image_array_list[i], rotate_angle);
+    if (!result_rotated_image.ok()) {
+      return result_rotated_image.status();
+    }
+    cv::Mat rotated_image = result_rotated_image.value();
+    rotated_images.push_back(rotated_image);
+  }
+  return rotated_images;
+}
+
+std::unordered_map<std::string, bool> _OCRPipeline::GetModelSettings() const {
+  std::unordered_map<std::string, bool> model_settings = {};
+  model_settings["use_doc_preprocessor"] = use_doc_preprocessor_;
+  model_settings["use_textline_orientation"] = use_textline_orientation_;
+  return model_settings;
+}
+
+std::vector<std::unique_ptr<BaseCVResult>>
+_OCRPipeline::Predict(const std::vector<std::string> &input) {
+  auto model_settings = GetModelSettings();
+  auto batches = batch_sampler_ptr_->Apply(input);
+  auto batches_string =
+      batch_sampler_ptr_->SampleFromVectorToStringVector(input);
+  if (!batches.ok()) {
+    INFOE("pipeline get sample fail : %s", batches.status().ToString().c_str());
+    exit(-1);
+  }
+  if (!batches_string.ok()) {
+    INFOE("pipeline get sample fail : %s",
+          batches_string.status().ToString().c_str());
+    exit(-1);
+  }
+  auto input_path = batch_sampler_ptr_->InputPath();
+  int index = 0;
+  std::vector<cv::Mat> origin_image = {};
+  std::vector<std::unique_ptr<BaseCVResult>> base_results = {};
+  pipeline_result_vec_.clear();
+  for (int i = 0; i < batches.value().size(); i++) {
+    origin_image.reserve(batches.value()[i].size());
+    for (const auto &mat : batches.value()[i]) {
+      origin_image.push_back(mat.clone());
+    }
+    std::vector<DocPreprocessorPipelineResult>
+        doc_preprocessors_pipeline_results = {};
+    if (use_doc_preprocessor_) {
+      doc_preprocessors_pipeline_->Predict(batches_string.value()[i]);
+      doc_preprocessors_pipeline_results =
+          static_cast<_DocPreprocessorPipeline *>(
+              doc_preprocessors_pipeline_.get())
+              ->PipelineResult();
+    } else {
+      DocPreprocessorPipelineResult result;
+      for (auto &image : batches.value()[i]) {
+        result.output_image = image.clone();
+        doc_preprocessors_pipeline_results.push_back(result);
+      }
+    }
+    std::vector<cv::Mat> doc_preprocessor_pipeline_images = {};
+    std::vector<cv::Mat> doc_preprocessor_pipeline_images_copy = {};
+    for (auto &item : doc_preprocessors_pipeline_results) {
+      doc_preprocessor_pipeline_images.push_back(item.output_image);
+      doc_preprocessor_pipeline_images_copy.push_back(
+          item.output_image.clone());
+    }
+    text_det_model_->Predict(doc_preprocessor_pipeline_images_copy);
+    std::vector<TextDetPredictorResult> det_results =
+        static_cast<TextDetPredictor *>(text_det_model_.get())
+            ->PredictorResult();
+    std::vector<std::vector<std::vector<cv::Point2f>>> dt_polys_list = {};
+    for (auto &item : det_results) {
+      if (!item.dt_polys.empty()) {
+        auto sort_item = sort_boxes_(item.dt_polys);
+        dt_polys_list.push_back(sort_item);
+      } else {
+        dt_polys_list.push_back(std::vector<std::vector<cv::Point2f>>{});
+      }
+    }
+
+    std::vector<int> indices = {};
+    for (int j = 0; j < doc_preprocessor_pipeline_images.size(); j++) {
+      if (!dt_polys_list.empty() && !dt_polys_list[j].empty()) {
+        indices.push_back(j);
+      }
+    }
+    std::vector<OCRPipelineResult> results(
+        doc_preprocessor_pipeline_images.size());
+    for (int k = 0; k < results.size(); k++, index++) {
+      results[k].input_path = input_path[index];
+      results[k].doc_preprocessor_res = doc_preprocessors_pipeline_results[k];
+      results[k].dt_polys = dt_polys_list[k];
+      results[k].model_settings = model_settings;
+      results[k].text_det_params = text_det_params_;
+      results[k].text_type = text_type_;
+      results[k].text_rec_score_thresh = text_rec_score_thresh_;
+    }
+    if (!indices.empty()) {
+      std::vector<cv::Mat> all_subs_of_imgs = {};
+      std::vector<cv::Mat> all_subs_of_imgs_copy = {};
+      std::vector<int> chunk_indices(1, 0);
+      for (auto &idx : indices) {
+        auto result_all_subs_of_img = (*crop_by_polys_)(
+            doc_preprocessor_pipeline_images[idx], dt_polys_list[idx]);
+        if (!result_all_subs_of_img.ok()) {
+          INFOE("Split image fail : ",
+                result_all_subs_of_img.status().ToString().c_str());
+          exit(-1);
+        }
+        all_subs_of_imgs.insert(all_subs_of_imgs.end(),
+                                result_all_subs_of_img.value().begin(),
+                                result_all_subs_of_img.value().end());
+        chunk_indices.emplace_back(chunk_indices.back() +
+                                   result_all_subs_of_img.value().size());
+      }
+      for (auto &item : all_subs_of_imgs) {
+        all_subs_of_imgs_copy.push_back(item.clone());
+      }
+      std::vector<int> angles = {};
+      if (model_settings["use_textline_orientation"]) {
+        textline_orientation_model_->Predict(all_subs_of_imgs_copy);
+        auto textline_orientation_model_results =
+            static_cast<ClasPredictor *>(textline_orientation_model_.get())
+                ->PredictorResult();
+        textline_orientation_model_results[0].input_image;
+        for (auto &result_angle : textline_orientation_model_results) {
+          angles.push_back(result_angle.class_ids[0]);
+        }
+        auto result_all_subs_of_imgs = RotateImage(all_subs_of_imgs, angles);
+        if (!result_all_subs_of_imgs.ok()) {
+          INFOE("Rotate images fail : %s",
+                result_all_subs_of_imgs.status().ToString().c_str());
+          exit(-1);
+        }
+        all_subs_of_imgs = result_all_subs_of_imgs.value();
+      } else {
+        angles = std::vector<int>(all_subs_of_imgs.size(), -1);
+      }
+      for (int l = 0; l < indices.size(); l++) {
+        for (int m = chunk_indices[l]; m < chunk_indices[l + 1]; m++) {
+          results[indices[l]].textline_orientation_angles.push_back(angles[m]);
+        }
+      }
+      for (int l = 0; l < indices.size(); l++) {
+        std::vector<cv::Mat> all_subs_of_img = {};
+        for (int m = chunk_indices[l]; m < chunk_indices[l + 1]; m++) {
+          all_subs_of_img.push_back(all_subs_of_imgs[m]);
+        }
+        std::vector<std::pair<std::pair<int, float>, TextRecPredictorResult>>
+            sub_img_info_list = {};
+
+        for (int m = 0; m < all_subs_of_img.size(); m++) {
+          int sub_img_id = m;
+          float sub_img_ratio = (float)all_subs_of_img[m].size[1] /
+                                (float)all_subs_of_img[m].size[0];
+          TextRecPredictorResult result;
+          sub_img_info_list.push_back({{sub_img_id, sub_img_ratio}, result});
+        }
+        std::vector<std::pair<int, float>> sorted_subs_info = {};
+        for (auto &item : sub_img_info_list) {
+          sorted_subs_info.push_back(item.first);
+        }
+        std::sort(
+            sorted_subs_info.begin(), sorted_subs_info.end(),
+            [](const std::pair<int, float> &a, const std::pair<int, float> &b) {
+              return a.second < b.second;
+            });
+        std::vector<cv::Mat> sorted_subs_of_img = {};
+        for (auto &item : sorted_subs_info) {
+          sorted_subs_of_img.push_back(all_subs_of_img[item.first]);
+        }
+        text_rec_model_->Predict(sorted_subs_of_img);
+        auto text_rec_model_results =
+            static_cast<TextRecPredictor *>(text_rec_model_.get())
+                ->PredictorResult();
+        for (int m = 0; m < text_rec_model_results.size(); m++) {
+          int sub_img_id = sorted_subs_info[m].first;
+          sub_img_info_list[sub_img_id].second = text_rec_model_results[m];
+        }
+        for (int sno = 0; sno < sub_img_info_list.size(); sno++) {
+          auto rec_res = sub_img_info_list[sno].second;
+          if (rec_res.rec_score >= text_rec_score_thresh_) {
+            results[l].rec_texts.push_back(rec_res.rec_text);
+            results[l].rec_scores.push_back(rec_res.rec_score);
+            results[l].rec_polys.push_back(dt_polys_list[l][sno]);
+            results[l].vis_fonts = rec_res.vis_font;
+          }
+        }
+      }
+    }
+    for (auto &res : results) {
+      if (text_type_ == "general") {
+        res.rec_boxes =
+            ComponentsProcessor::ConvertPointsToBoxes(res.rec_polys);
+      }
+      pipeline_result_vec_.push_back(res);
+      base_results.push_back(std::unique_ptr<BaseCVResult>(new OCRResult(res)));
+    }
+  }
+  return base_results;
+}
+
+std::vector<std::unique_ptr<BaseCVResult>>
+OCRPipeline::Predict(const std::vector<std::string> &input) {
+  if (thread_num_ == 1) {
+    return infer_->Predict(input);
+  }
+  batch_sampler_ptr_ =
+      std::unique_ptr<BaseBatchSampler>(new ImageBatchSampler(1));
+  auto nomeaning = batch_sampler_ptr_->Apply(input);
+  int input_num = nomeaning.value().size();
+  if (thread_num_ > input_num) {
+    INFOW("thread num exceed input num, will set %d", input_num);
+    thread_num_ = input_num;
+  }
+  int infer_batch_num = input_num / thread_num_;
+  auto status = batch_sampler_ptr_->SetBatchSize(infer_batch_num);
+  if (!status.ok()) {
+    INFOE("Set batch size fail : %s", status.ToString().c_str());
+    exit(-1);
+  }
+  auto infer_batch_data =
+      batch_sampler_ptr_->SampleFromVectorToStringVector(input);
+  if (!infer_batch_data.ok()) {
+    INFOE("Get infer batch data fail : %s",
+          infer_batch_data.status().ToString().c_str());
+    exit(-1);
+  }
+  std::vector<std::unique_ptr<BaseCVResult>> results = {};
+  results.reserve(input_num);
+  for (auto &infer_data : infer_batch_data.value()) {
+    auto status =
+        AutoParallelSimpleInferencePipeline::PredictThread(infer_data);
+    if (!status.ok()) {
+      INFOE("Infer fail : %s", status.ToString().c_str());
+      exit(-1);
+    }
+  }
+  for (int i = 0; i < infer_batch_data.value().size(); i++) {
+    auto infer_data_result = GetResult();
+    if (!infer_data_result.ok()) {
+      INFOE("Get infer result fail : %s",
+            infer_batch_data.status().ToString().c_str());
+      exit(-1);
+    }
+    results.insert(results.end(),
+                   std::make_move_iterator(infer_data_result.value().begin()),
+                   std::make_move_iterator(infer_data_result.value().end()));
+  }
+  return results;
+}
+
+void _OCRPipeline::OverrideConfig() {
+  auto &data = config_.Data();
+  if (params_.doc_orientation_classify_model_name.has_value()) {
+    auto it = config_.FindKey("DocOrientationClassify.model_name");
+    if (!it.ok()) {
+      data["SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify."
+           "model_name"] = params_.doc_orientation_classify_model_name.value();
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.doc_orientation_classify_model_name.value();
+    }
+  }
+  if (params_.doc_orientation_classify_model_dir.has_value()) {
+    auto it = config_.FindKey("DocOrientationClassify.model_dir");
+    if (!it.ok()) {
+      data["SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify."
+           "model_dir"] = params_.doc_orientation_classify_model_dir.value();
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.doc_orientation_classify_model_dir.value();
+    }
+  }
+  if (params_.doc_unwarping_model_name.has_value()) {
+    auto it = config_.FindKey("DocUnwarping.model_name");
+    if (!it.ok()) {
+      data["SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_name"] =
+          params_.doc_unwarping_model_name.value();
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.doc_unwarping_model_name.value();
+    }
+  }
+  if (params_.doc_unwarping_model_dir.has_value()) {
+    auto it = config_.FindKey("DocUnwarping.model_dir");
+    if (!it.ok()) {
+      data["SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_dir"] =
+          params_.doc_unwarping_model_dir.value();
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.doc_unwarping_model_dir.value();
+    }
+  }
+  if (params_.text_detection_model_name.has_value()) {
+    auto it = config_.FindKey("TextDetection.model_name");
+    if (!it.ok()) {
+      data["SubModules.TextDetection.model_name"] =
+          params_.text_detection_model_name.value();
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.text_detection_model_name.value();
+    }
+  }
+  if (params_.text_detection_model_dir.has_value()) {
+    auto it = config_.FindKey("TextDetection.model_dir");
+    if (!it.ok()) {
+      data["SubModules.TextDetection.model_dir"] =
+          params_.text_detection_model_dir.value();
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.text_detection_model_dir.value();
+    }
+  }
+  if (params_.textline_orientation_model_name.has_value()) {
+    auto it = config_.FindKey("TextLineOrientation.model_name");
+    if (!it.ok()) {
+      data["SubModules.TextLineOrientation.model_name"] =
+          params_.textline_orientation_model_name.value();
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.textline_orientation_model_name.value();
+    }
+  }
+  if (params_.textline_orientation_model_dir.has_value()) {
+    auto it = config_.FindKey("TextLineOrientation.model_dir");
+    if (!it.ok()) {
+      data["SubModules.TextLineOrientation.model_dir"] =
+          params_.textline_orientation_model_dir.value();
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.textline_orientation_model_dir.value();
+    }
+  }
+  if (params_.textline_orientation_batch_size.has_value()) {
+    auto it = config_.FindKey("TextLineOrientation.batch_size");
+    if (!it.ok()) {
+      data["SubModules.TextLineOrientation.batch_size"] =
+          std::to_string(params_.textline_orientation_batch_size.value());
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] =
+          std::to_string(params_.textline_orientation_batch_size.value());
+    }
+  }
+
+  if (params_.text_recognition_model_name.has_value()) {
+    auto it = config_.FindKey("TextRecognition.model_name");
+    if (!it.ok()) {
+      data["SubModules.TextRecognition.model_name"] =
+          params_.text_recognition_model_name.value();
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.text_recognition_model_name.value();
+    }
+  }
+  if (params_.text_recognition_model_dir.has_value()) {
+    auto it = config_.FindKey("TextRecognition.model_dir");
+    if (!it.ok()) {
+      data["SubModules.TextRecognition.model_dir"] =
+          params_.text_recognition_model_dir.value();
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.text_recognition_model_dir.value();
+    }
+  }
+  if (params_.text_recognition_batch_size.has_value()) {
+    auto it = config_.FindKey("TextRecognition.batch_size");
+    if (!it.ok()) {
+      data["SubModules.TextRecognition.batch_size"] =
+          std::to_string(params_.text_recognition_batch_size.value());
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = std::to_string(params_.text_recognition_batch_size.value());
+    }
+  }
+
+  if (params_.use_doc_orientation_classify.has_value()) {
+    auto it = config_.FindKey("DocPreprocessor.use_doc_orientation_classify");
+    if (!it.ok()) {
+      data["SubPipelines.DocPreprocessor.use_doc_orientation_classify"] =
+          params_.use_doc_orientation_classify.value() ? "true" : "false";
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] =
+          params_.use_doc_orientation_classify.value() ? "true" : "false";
+    }
+  }
+  if (params_.use_doc_unwarping.has_value()) {
+    auto it = config_.FindKey("DocPreprocessor.use_doc_unwarping");
+    if (!it.ok()) {
+      data["SubPipelines.DocPreprocessor.use_doc_unwarping"] =
+          params_.use_doc_unwarping.value() ? "true" : "false";
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.use_doc_unwarping.value() ? "true" : "false";
+    }
+  }
+  if (params_.use_textline_orientation.has_value()) {
+    auto it = config_.FindKey("use_textline_orientation");
+    if (!it.ok()) {
+      data["use_textline_orientation"] =
+          params_.use_textline_orientation.value() ? "true" : "false";
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.use_textline_orientation.value() ? "true" : "false";
+    }
+  }
+  if (params_.text_det_limit_side_len.has_value()) {
+    auto it = config_.FindKey("TextDetection.limit_side_len");
+    if (!it.ok()) {
+      data["SubModules.TextDetection.limit_side_len"] =
+          std::to_string(params_.text_det_limit_side_len.value());
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = std::to_string(params_.text_det_limit_side_len.value());
+    }
+  }
+  if (params_.text_det_limit_type.has_value()) {
+    auto it = config_.FindKey("TextDetection.limit_type");
+    if (!it.ok()) {
+      data["SubModules.TextDetection.limit_type"] =
+          params_.text_det_limit_type.value();
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = params_.text_det_limit_type.value();
+    }
+  }
+  if (params_.text_det_thresh.has_value()) {
+    auto it = config_.FindKey("TextDetection.thresh");
+    if (!it.ok()) {
+      data["SubModules.TextDetection.thresh"] =
+          std::to_string(params_.text_det_thresh.value());
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = std::to_string(params_.text_det_thresh.value());
+    }
+  }
+  if (params_.text_det_box_thresh.has_value()) {
+    auto it = config_.FindKey("TextDetection.box_thresh");
+    if (!it.ok()) {
+      data["SubModules.TextDetection.box_thresh"] =
+          std::to_string(params_.text_det_box_thresh.value());
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = std::to_string(params_.text_det_box_thresh.value());
+    }
+  }
+  if (params_.text_det_unclip_ratio.has_value()) {
+    auto it = config_.FindKey("TextDetection.unclip_ratio");
+    if (!it.ok()) {
+      data["SubModules.TextDetection.unclip_ratio"] =
+          std::to_string(params_.text_det_unclip_ratio.value());
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = std::to_string(params_.text_det_unclip_ratio.value());
+    }
+  }
+  if (params_.text_det_input_shape.has_value()) {
+    auto it = config_.FindKey("TextDetection.input_shape");
+    if (!it.ok()) {
+      data["SubModules.TextDetection.input_shape"] =
+          Utility::VecToString(params_.text_det_input_shape.value());
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = Utility::VecToString(params_.text_det_input_shape.value());
+    }
+  }
+  if (params_.text_rec_score_thresh.has_value()) {
+    auto it = config_.FindKey("TextRecognition.score_thresh");
+    if (!it.ok()) {
+      data["SubModules.TextRecognition.score_thresh"] =
+          std::to_string(params_.text_rec_score_thresh.value());
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = std::to_string(params_.text_rec_score_thresh.value());
+    }
+  }
+  if (params_.text_rec_input_shape.has_value()) {
+    auto it = config_.FindKey("TextRecognition.input_shape");
+    if (!it.ok()) {
+      data["SubModules.TextRecognition.input_shape"] =
+          Utility::VecToString(params_.text_rec_input_shape.value());
+    } else {
+      auto key = it.value().first;
+      data.erase(data.find(key));
+      data[key] = Utility::VecToString(params_.text_rec_input_shape.value());
+    }
+  }
+}
diff --git a/deploy/cpp_infer/src/pipelines/ocr/pipeline.h b/deploy/cpp_infer/src/pipelines/ocr/pipeline.h
new file mode 100644
index 0000000..875c894
--- /dev/null
+++ b/deploy/cpp_infer/src/pipelines/ocr/pipeline.h
@@ -0,0 +1,162 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <functional>
+#include <iostream>
+#include <string>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+#include "absl/types/optional.h"
+#include "src/base/base_pipeline.h"
+#include "src/common/image_batch_sampler.h"
+#include "src/common/processors.h"
+#include "src/modules/image_classification/predictor.h"
+#include "src/modules/text_detection/predictor.h"
+#include "src/modules/text_recognition/predictor.h"
+#include "src/pipelines/doc_preprocessor/pipeline.h"
+#include "src/utils/ilogger.h"
+#include "src/utils/utility.h"
+
+struct TextDetParams {
+  int text_det_limit_side_len = -1;
+  std::string text_det_limit_type = "";
+  int text_det_max_side_limit = -1;
+  float text_det_thresh = -1;
+  float text_det_box_thresh = -1;
+  float text_det_unclip_ratio = -1;
+};
+
+struct OCRPipelineResult {
+  std::string input_path = "";
+  DocPreprocessorPipelineResult doc_preprocessor_res;
+  std::vector<std::vector<cv::Point2f>> dt_polys = {};
+  std::unordered_map<std::string, bool> model_settings = {};
+  TextDetParams text_det_params;
+  std::string text_type = "";
+  float text_rec_score_thresh = 0.0;
+  std::vector<std::string> rec_texts = {};
+  std::vector<float> rec_scores = {};
+  std::vector<int> textline_orientation_angles = {};
+  std::vector<std::vector<cv::Point2f>> rec_polys = {};
+  std::vector<std::array<float, 4>> rec_boxes = {};
+  std::string vis_fonts = "";
+};
+
+struct OCRPipelineParams {
+  absl::optional<std::string> doc_orientation_classify_model_name =
+      absl::nullopt;
+  absl::optional<std::string> doc_orientation_classify_model_dir =
+      absl::nullopt;
+  absl::optional<std::string> doc_unwarping_model_name = absl::nullopt;
+  absl::optional<std::string> doc_unwarping_model_dir = absl::nullopt;
+  absl::optional<std::string> text_detection_model_name = absl::nullopt;
+  absl::optional<std::string> text_detection_model_dir = absl::nullopt;
+  absl::optional<std::string> textline_orientation_model_name = absl::nullopt;
+  absl::optional<std::string> textline_orientation_model_dir = absl::nullopt;
+  absl::optional<int> textline_orientation_batch_size = absl::nullopt;
+  absl::optional<std::string> text_recognition_model_name = absl::nullopt;
+  absl::optional<std::string> text_recognition_model_dir = absl::nullopt;
+  absl::optional<int> text_recognition_batch_size = absl::nullopt;
+  absl::optional<bool> use_doc_orientation_classify = absl::nullopt;
+  absl::optional<bool> use_doc_unwarping = absl::nullopt;
+  absl::optional<bool> use_textline_orientation = absl::nullopt;
+  absl::optional<int> text_det_limit_side_len = absl::nullopt;
+  absl::optional<std::string> text_det_limit_type = absl::nullopt;
+  absl::optional<float> text_det_thresh = absl::nullopt;
+  absl::optional<float> text_det_box_thresh = absl::nullopt;
+  absl::optional<float> text_det_unclip_ratio = absl::nullopt;
+  absl::optional<std::vector<int>> text_det_input_shape = absl::nullopt;
+  absl::optional<float> text_rec_score_thresh = absl::nullopt;
+  absl::optional<std::vector<int>> text_rec_input_shape = absl::nullopt;
+  absl::optional<std::string> lang = absl::nullopt;
+  absl::optional<std::string> ocr_version = absl::nullopt;
+  absl::optional<std::string> vis_font_dir = absl::nullopt;
+  absl::optional<std::string> device = absl::nullopt;
+  bool enable_mkldnn = true;
+  int mkldnn_cache_capacity = 10;
+  std::string precision = "fp32";
+  int cpu_threads = 8;
+  int thread_num = 1;
+  absl::optional<Utility::PaddleXConfigVariant> paddlex_config = absl::nullopt;
+};
+
+class _OCRPipeline : public BasePipeline {
+public:
+  explicit _OCRPipeline(const OCRPipelineParams &params);
+  virtual ~_OCRPipeline() = default;
+  _OCRPipeline() = delete;
+
+  std::vector<std::unique_ptr<BaseCVResult>>
+  Predict(const std::vector<std::string> &input) override;
+
+  std::vector<OCRPipelineResult> PipelineResult() const {
+    return pipeline_result_vec_;
+  };
+
+  static absl::StatusOr<std::vector<cv::Mat>>
+  RotateImage(const std::vector<cv::Mat> &image_array_list,
+              const std::vector<int> &rotate_angle_list);
+
+  std::unordered_map<std::string, bool> GetModelSettings() const;
+  TextDetParams GetTextDetParams() const { return text_det_params_; };
+
+  void OverrideConfig();
+
+private:
+  OCRPipelineParams params_;
+  YamlConfig config_;
+  std::unique_ptr<BaseBatchSampler> batch_sampler_ptr_;
+  std::vector<OCRPipelineResult> pipeline_result_vec_;
+  bool use_doc_preprocessor_ = false;
+  bool use_doc_orientation_classify_ = false;
+  bool use_doc_unwarping_ = false;
+  std::unique_ptr<BasePipeline> doc_preprocessors_pipeline_;
+  bool use_textline_orientation_ = false;
+  std::unique_ptr<BasePredictor> textline_orientation_model_;
+  std::unique_ptr<BasePredictor> text_det_model_;
+  std::unique_ptr<BasePredictor> text_rec_model_;
+  std::unique_ptr<CropByPolys> crop_by_polys_;
+  std::function<std::vector<std::vector<cv::Point2f>>(
+      const std::vector<std::vector<cv::Point2f>> &)>
+      sort_boxes_;
+  float text_rec_score_thresh_ = 0.0;
+  std::string text_type_;
+  TextDetParams text_det_params_;
+};
+
+class OCRPipeline
+    : public AutoParallelSimpleInferencePipeline<
+          _OCRPipeline, OCRPipelineParams, std::vector<std::string>,
+          std::vector<std::unique_ptr<BaseCVResult>>> {
+public:
+  OCRPipeline(const OCRPipelineParams &params)
+      : AutoParallelSimpleInferencePipeline(params),
+        thread_num_(params.thread_num) {
+    if (thread_num_ == 1) {
+      infer_ = std::unique_ptr<BasePipeline>(new _OCRPipeline(params));
+    }
+  };
+
+  std::vector<std::unique_ptr<BaseCVResult>>
+  Predict(const std::vector<std::string> &input) override;
+
+private:
+  int thread_num_;
+  std::unique_ptr<BasePipeline> infer_;
+  std::unique_ptr<BaseBatchSampler> batch_sampler_ptr_;
+};
diff --git a/deploy/cpp_infer/src/pipelines/ocr/result.cc b/deploy/cpp_infer/src/pipelines/ocr/result.cc
new file mode 100644
index 0000000..f1fb4be
--- /dev/null
+++ b/deploy/cpp_infer/src/pipelines/ocr/result.cc
@@ -0,0 +1,539 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "result.h"
+
+#include <algorithm>
+#include <codecvt>
+#include <fstream>
+#include <locale>
+#include <random>
+#include <string>
+
+#include "src/utils/utility.h"
+#include "third_party/nlohmann/json.hpp"
+
+using json = nlohmann::json;
+
+void OCRResult::SaveToImg(const std::string &save_path) {
+  cv::Mat image = pipeline_result_.doc_preprocessor_res.output_image;
+  auto texts = pipeline_result_.rec_texts;
+  std::vector<std::vector<cv::Point>> boxes;
+  std::vector<std::vector<cv::Point2f>> boxes_float =
+      pipeline_result_.rec_polys;
+  for (const auto &floatPolygon : pipeline_result_.rec_polys) {
+    std::vector<cv::Point> intPolygon;
+    for (const auto &point : floatPolygon) {
+      intPolygon.push_back(cv::Point(cvRound(point.x), cvRound(point.y)));
+    }
+    boxes.push_back(intPolygon);
+  }
+
+  if (image.empty()) {
+    INFOE("Input image is empty.");
+    exit(-1);
+  }
+
+  int h = image.rows;
+  int w = image.cols;
+
+  cv::Mat img_left = image.clone();
+
+  cv::Mat img_right(h, w, CV_8UC3, cv::Scalar(255, 255, 255));
+
+  std::random_device rd;
+  std::mt19937 gen(rd());
+  std::uniform_int_distribution<> dis(0, 255);
+
+  for (size_t i = 0; i < boxes.size(); ++i) {
+    auto &box = boxes[i];
+    const auto &box_float = boxes_float[i];
+    const auto &text = texts[i];
+
+    cv::Scalar color(dis(gen), dis(gen), dis(gen));
+
+    if (box.size() > 4) {
+      const std::vector<std::vector<cv::Point>> polygons{box};
+      cv::fillPoly(img_left, polygons, color);
+      cv::polylines(img_left, polygons, true, color, 8);
+      box = GetMinareaRect(box);
+      std::vector<int> ys;
+      for (const auto &pt : box)
+        ys.push_back(pt.y);
+      int min_y = *std::min_element(ys.begin(), ys.end());
+      int max_y = *std::max_element(ys.begin(), ys.end());
+      int height = static_cast<int>(0.5 * (max_y - min_y));
+      double mean_y = std::accumulate(ys.begin(), ys.end(), 0.0) / ys.size();
+      if (box.size() >= 4) {
+        box[0].y = static_cast<int>(mean_y);
+        box[1].y = static_cast<int>(mean_y);
+        box[2].y = static_cast<int>(mean_y + std::min(20, height));
+        box[3].y = static_cast<int>(mean_y + std::min(20, height));
+      }
+    } else {
+      cv::fillPoly(img_left, std::vector<std::vector<cv::Point>>{box}, color);
+    }
+#ifdef USE_FREETYPE
+    cv::Mat img_right_text = DrawBoxTextFine(cv::Size(w, h), box_float, text,
+                                             pipeline_result_.vis_fonts);
+    cv::polylines(img_right_text, box, true, color, 1);
+    cv::bitwise_and(img_right, img_right_text, img_right);
+#endif
+  }
+  cv::Mat blended;
+  cv::addWeighted(image, 0.5, img_left, 0.5, 0, blended);
+#ifdef USE_FREETYPE
+  cv::Mat ocr_res_image(h, w * 2, CV_8UC3, cv::Scalar(255, 255, 255));
+  blended.copyTo(ocr_res_image(cv::Rect(0, 0, w, h)));
+  img_right.copyTo(ocr_res_image(cv::Rect(w, 0, w, h)));
+#else
+  cv::Mat ocr_res_image = blended;
+#endif
+
+  auto model_settings = pipeline_result_.model_settings;
+  std::unordered_map<std::string, cv::Mat> res_img_dict;
+  res_img_dict["ocr_res_img"] = ocr_res_image;
+
+  auto ocr_path = Utility::SmartCreateDirectoryForImage(
+      save_path, pipeline_result_.input_path, "_ocr_res_img");
+  if (!ocr_path.ok()) {
+    INFOE(ocr_path.status().ToString().c_str());
+    exit(-1);
+  }
+  auto doc_pre_path = Utility::SmartCreateDirectoryForImage(
+      save_path, pipeline_result_.input_path, "_doc_preprocessor_res");
+  if (!doc_pre_path.ok()) {
+    INFOE(doc_pre_path.status().ToString().c_str());
+    exit(-1);
+  }
+  cv::imwrite(ocr_path.value(), ocr_res_image);
+  if (model_settings["use_doc_preprocessor"]) {
+    int h1 = pipeline_result_.doc_preprocessor_res.input_image.size[0];
+    int w1 = pipeline_result_.doc_preprocessor_res.input_image.size[1];
+    int h2 = pipeline_result_.doc_preprocessor_res.rotate_image.size[0];
+    int w2 = pipeline_result_.doc_preprocessor_res.rotate_image.size[1];
+    int h3 = pipeline_result_.doc_preprocessor_res.output_image.size[0];
+    int w3 = pipeline_result_.doc_preprocessor_res.output_image.size[1];
+    int h_all = std::max(h1, std::max(h2, h3));
+    int total_w = w1 + w2 + w3;
+
+    cv::Mat doc_pre_res_image(h_all, total_w, CV_8UC3,
+                              cv::Scalar(255, 255, 255));
+
+    pipeline_result_.doc_preprocessor_res.input_image.copyTo(
+        doc_pre_res_image(cv::Rect(0, 0, w1, h1)));
+    pipeline_result_.doc_preprocessor_res.rotate_image.copyTo(
+        doc_pre_res_image(cv::Rect(w1, 0, w2, h2)));
+    pipeline_result_.doc_preprocessor_res.output_image.copyTo(
+        doc_pre_res_image(cv::Rect(w1 + w2, 0, w3, h3)));
+    cv::imwrite(doc_pre_path.value(), doc_pre_res_image);
+    res_img_dict["doc_preprocessor_res"] = doc_pre_res_image;
+  }
+}
+
+#ifdef USE_FREETYPE
+cv::Mat OCRResult::DrawBoxTextFine(const cv::Size &img_size,
+                                   const std::vector<cv::Point2f> &box,
+                                   const std::string &txt,
+                                   const std::string &vis_font) {
+  int box_height = cv::norm(box[0] - box[3]);
+  int box_width = cv::norm(box[0] - box[1]);
+  auto ft2 = cv::freetype::createFreeType2();
+  ft2->loadFontData(vis_font, 0);
+
+  bool vertical_mode = box_height > 2 * box_width && box_height > 30;
+  int n = std::max(int(txt.size()), 1);
+
+  int font_height = 10;
+  if (!txt.empty()) {
+    if (vertical_mode) {
+      font_height = CreateFontVertical(ft2, txt, box_height, box_width);
+    } else {
+      font_height = CreateFont(ft2, txt, box_height, box_width);
+    }
+  }
+  cv::Mat img_text(box_height, box_width, CV_8UC3, cv::Scalar(255, 255, 255));
+  int x = 0, y = 0;
+
+  if (!txt.empty()) {
+    if (vertical_mode) {
+      DrawVerticalText(ft2, img_text, txt, x, y, font_height,
+                       cv::Scalar(0, 0, 0));
+    } else {
+      int baseline = 0;
+      cv::Size textsize = ft2->getTextSize(txt, font_height, -1, &baseline);
+      x = (box_width - textsize.width) / 2;
+      y = (box_height + textsize.height) / 2 - baseline;
+      ft2->putText(img_text, txt, cv::Point(x, y), font_height,
+                   cv::Scalar(0, 0, 0), -1, cv::LINE_AA, true);
+    }
+  }
+  std::vector<cv::Point2f> src_pts = {{0, 0},
+                                      {float(box_width), 0},
+                                      {float(box_width), float(box_height)},
+                                      {0, float(box_height)}};
+  cv::Mat M = cv::getPerspectiveTransform(src_pts, box);
+
+  cv::Mat dst(img_size, CV_8UC3, cv::Scalar(255, 255, 255));
+  cv::warpPerspective(img_text, dst, M, img_size, cv::INTER_NEAREST,
+                      cv::BORDER_CONSTANT, cv::Scalar(255, 255, 255));
+  return dst;
+}
+cv::Size OCRResult::getActualCharSize(cv::Ptr<cv::freetype::FreeType2> &ft2,
+                                      const std::string &utf8_char,
+                                      int font_height) {
+  cv::Mat temp = cv::Mat::zeros(300, 300, CV_8UC1);
+
+  cv::Point pos(100, 150);
+
+  ft2->putText(temp, utf8_char, pos, font_height, cv::Scalar(255), -1,
+               cv::LINE_AA, false);
+
+  std::vector<cv::Point> nonZeroPoints;
+  cv::findNonZero(temp, nonZeroPoints);
+
+  if (nonZeroPoints.empty()) {
+    return cv::Size(0, 0);
+  }
+
+  cv::Rect boundingRect = cv::boundingRect(nonZeroPoints);
+  return cv::Size(boundingRect.width, boundingRect.height);
+}
+void OCRResult::DrawVerticalText(cv::Ptr<cv::freetype::FreeType2> &ft2,
+                                 cv::Mat &img, const std::string &text, int x,
+                                 int y, int font_height, cv::Scalar color,
+                                 float line_spacing) {
+  std::wstring wtext =
+      std::wstring_convert<std::codecvt_utf8<wchar_t>>().from_bytes(text);
+  for (size_t i = 0; i < wtext.size(); ++i) {
+    std::wstring single_char(1, wtext[i]);
+    std::string utf8_char =
+        std::wstring_convert<std::codecvt_utf8<wchar_t>>().to_bytes(
+            single_char);
+    ft2->putText(img, utf8_char, cv::Point(x, y), font_height, color, -1,
+                 cv::LINE_AA, true);
+    int baseline = 0;
+    cv::Size size = ft2->getTextSize(utf8_char, font_height, -1, &baseline);
+    size.height += baseline;
+    y += size.height * 1.1 + line_spacing;
+  }
+}
+int OCRResult::CreateFont(cv::Ptr<cv::freetype::FreeType2> &ft2,
+                          const std::string &text, int region_height,
+                          int region_width) {
+  int font_height = std::max(int(region_height * 0.8), 10);
+  int baseline = 0;
+  cv::Size text_size = ft2->getTextSize(text, font_height, -1, &baseline);
+  if (text_size.width > region_width) {
+    font_height =
+        static_cast<int>(font_height * region_width / text_size.width);
+    text_size = ft2->getTextSize(text, font_height, -1, &baseline);
+  }
+  return font_height;
+}
+int OCRResult::CreateFontVertical(cv::Ptr<cv::freetype::FreeType2> &ft2,
+                                  const std::string &text, int region_height,
+                                  int region_width, float scale) {
+  std::wstring_convert<std::codecvt_utf8<wchar_t>> conv;
+  std::wstring wtext = conv.from_bytes(text);
+  int n = static_cast<int>(wtext.length());
+  int baseFontSize = static_cast<int>(region_height / n * 0.8 * scale);
+  baseFontSize = std::max(baseFontSize, 10);
+
+  int maxCharWidth = 0;
+  for (size_t i = 0; i < wtext.length(); ++i) {
+    std::wstring singleChar(1, wtext[i]);
+    std::string utf8Char =
+        std::wstring_convert<std::codecvt_utf8<wchar_t>>().to_bytes(singleChar);
+    cv::Size textSize = getActualCharSize(ft2, utf8Char, baseFontSize);
+    maxCharWidth = std::max(maxCharWidth, textSize.width);
+  }
+
+  int finalFontSize = baseFontSize;
+  if (maxCharWidth > region_width) {
+    finalFontSize =
+        static_cast<int>(baseFontSize * region_width / maxCharWidth);
+    finalFontSize = std::max(finalFontSize, 10);
+  }
+
+  return finalFontSize;
+}
+#endif
+
+std::vector<cv::Point>
+OCRResult::GetMinareaRect(const std::vector<cv::Point> &points) {
+  cv::RotatedRect bounding_box = cv::minAreaRect(points);
+
+  cv::Point2f boxPts[4];
+  bounding_box.points(boxPts);
+  std::vector<cv::Point2f> ptsVec(boxPts, boxPts + 4);
+
+  std::sort(
+      ptsVec.begin(), ptsVec.end(),
+      [](const cv::Point2f &a, const cv::Point2f &b) { return a.x < b.x; });
+  int index_a, index_b, index_c, index_d;
+  if (ptsVec[1].y > ptsVec[0].y) {
+    index_a = 0;
+    index_d = 1;
+  } else {
+    index_a = 1;
+    index_d = 0;
+  }
+  if (ptsVec[3].y > ptsVec[2].y) {
+    index_b = 2;
+    index_c = 3;
+  } else {
+    index_b = 3;
+    index_c = 2;
+  }
+
+  std::vector<cv::Point> box = {ptsVec[index_a], ptsVec[index_b],
+                                ptsVec[index_c], ptsVec[index_d]};
+
+  for (auto &pt : box) {
+    pt.x = static_cast<int>(std::round(pt.x));
+    pt.y = static_cast<int>(std::round(pt.y));
+  }
+
+  return box;
+}
+
+void OCRResult::SaveToJson(const std::string &save_path) const {
+  nlohmann::ordered_json j;
+  j["input_path"] = pipeline_result_.input_path;
+
+  j["page_index"] = nullptr;
+
+  j["model_settings"] = pipeline_result_.model_settings;
+
+  auto it = pipeline_result_.model_settings.find("use_doc_preprocessor");
+  if (it != pipeline_result_.model_settings.end() && it->second) {
+    nlohmann::ordered_json j_doc_pre;
+    j_doc_pre["model_settings"] =
+        pipeline_result_.doc_preprocessor_res.model_settings;
+    j_doc_pre["angle"] = pipeline_result_.doc_preprocessor_res.angle;
+    j["doc_preprocessor_res"] = j_doc_pre;
+  }
+  json polys_json = json::array();
+  for (const auto &polygon : pipeline_result_.dt_polys) {
+    json poly_json = json::array();
+    for (const auto &point : polygon) {
+      poly_json.push_back(
+          {static_cast<int>(point.x), static_cast<int>(point.y)});
+    }
+    polys_json.push_back(poly_json);
+  }
+  j["dt_polys"] = polys_json;
+  nlohmann::ordered_json j_text_det_params;
+  j_text_det_params["limit_side_len"] =
+      pipeline_result_.text_det_params.text_det_limit_side_len;
+  j_text_det_params["limit_type"] =
+      pipeline_result_.text_det_params.text_det_limit_type;
+  j_text_det_params["thresh"] =
+      pipeline_result_.text_det_params.text_det_thresh;
+  j_text_det_params["max_side_limit"] =
+      pipeline_result_.text_det_params.text_det_max_side_limit;
+  j_text_det_params["box_thresh"] =
+      pipeline_result_.text_det_params.text_det_box_thresh;
+  j_text_det_params["unclip_ratio"] =
+      pipeline_result_.text_det_params.text_det_unclip_ratio;
+  j["text_det_params"] = j_text_det_params;
+  j["text_type"] = pipeline_result_.text_type;
+
+  if (!pipeline_result_.textline_orientation_angles.empty()) {
+    j["textline_orientation_angles"] =
+        pipeline_result_.textline_orientation_angles;
+  }
+  j["text_rec_score_thresh"] = pipeline_result_.text_rec_score_thresh;
+  j["rec_texts"] = pipeline_result_.rec_texts;
+  j["rec_scores"] = pipeline_result_.rec_scores;
+  json rec_polys_json = json::array();
+  for (const auto &polygon : pipeline_result_.rec_polys) {
+    json poly_json = json::array();
+    for (const auto &point : polygon) {
+      poly_json.push_back(
+          {static_cast<int>(point.x), static_cast<int>(point.y)});
+    }
+    rec_polys_json.push_back(poly_json);
+  }
+  j["rec_polys"] = rec_polys_json;
+
+  std::vector<std::array<int, 4>> int_vec;
+  int_vec.reserve(pipeline_result_.rec_boxes.size());
+
+  std::transform(pipeline_result_.rec_boxes.begin(),
+                 pipeline_result_.rec_boxes.end(), std::back_inserter(int_vec),
+                 [](const std::array<float, 4> &arr) {
+                   std::array<int, 4> res;
+                   for (size_t i = 0; i < 4; ++i) {
+                     res[i] = static_cast<int>(arr[i]);
+                   }
+                   return res;
+                 });
+  j["rec_boxes"] = int_vec;
+
+  absl::StatusOr<std::string> full_path;
+  if (pipeline_result_.input_path.empty()) {
+    INFOW("Input path is empty, will use output_res.json instead!");
+    full_path = Utility::SmartCreateDirectoryForJson(save_path, "output");
+  } else {
+    full_path = Utility::SmartCreateDirectoryForJson(
+        save_path, pipeline_result_.input_path);
+  }
+  if (!full_path.ok()) {
+    INFOE(full_path.status().ToString().c_str());
+    exit(-1);
+  }
+  std::ofstream file(full_path.value());
+  if (file.is_open()) {
+    file << j.dump(4);
+    file.close();
+  } else {
+    INFOE("Could not open file for writing: %s", save_path.c_str());
+    exit(-1);
+  }
+}
+
+void PrintDocPreprocessorPipelineResult(
+    const DocPreprocessorPipelineResult &doc) {
+  std::cout << "{\n";
+  std::cout << "    \"model_settings\": {";
+  bool first = true;
+  for (const auto &kv : doc.model_settings) {
+    if (!first)
+      std::cout << ", ";
+    std::cout << "\"" << kv.first << "\": " << (kv.second ? "true" : "false");
+    first = false;
+  }
+  std::cout << "},\n";
+  std::cout << "    \"angle\": " << doc.angle << "\n";
+  std::cout << "  }";
+}
+
+void PrintPolys(const std::vector<std::vector<cv::Point2f>> &polys) {
+  std::cout << "[";
+  for (size_t i = 0; i < polys.size(); ++i) {
+    if (i != 0)
+      std::cout << ",\n ";
+    std::cout << "[";
+    for (size_t j = 0; j < polys[i].size(); ++j) {
+      if (j != 0)
+        std::cout << ", ";
+      std::cout << "[" << polys[i][j].x << ", " << polys[i][j].y << "]";
+    }
+    std::cout << "]";
+  }
+  std::cout << "]";
+}
+
+void PrintModelSettings(const std::unordered_map<std::string, bool> &ms) {
+  std::cout << "{";
+  bool first = true;
+  for (const auto &kv : ms) {
+    if (!first)
+      std::cout << ", ";
+    std::cout << "\"" << kv.first << "\": " << (kv.second ? "true" : "false");
+    first = false;
+  }
+  std::cout << "}";
+}
+
+void PrintArray(const std::vector<float> &arr) {
+  std::cout << "[";
+  for (size_t i = 0; i < arr.size(); ++i) {
+    if (i != 0)
+      std::cout << ", ";
+    std::cout << arr[i];
+  }
+  std::cout << "]";
+}
+
+void PrintStringArray(const std::vector<std::string> &arr) {
+  std::cout << "[";
+  for (size_t i = 0; i < arr.size(); ++i) {
+    if (i != 0)
+      std::cout << ", ";
+    std::cout << "\"" << arr[i] << "\"";
+  }
+  std::cout << "]";
+}
+
+void PrintIntArray(const std::vector<int> &arr) {
+  std::cout << "[";
+  for (size_t i = 0; i < arr.size(); ++i) {
+    if (i != 0)
+      std::cout << ", ";
+    std::cout << arr[i];
+  }
+  std::cout << "]";
+}
+
+void PrintRecBoxes(const std::vector<std::array<float, 4>> &arr) {
+  std::cout << "[";
+  for (size_t i = 0; i < arr.size(); ++i) {
+    if (i != 0)
+      std::cout << ", ";
+    std::cout << "[" << arr[i][0] << ", " << arr[i][1] << ", " << arr[i][2]
+              << ", " << arr[i][3] << "]";
+  }
+  std::cout << "],";
+}
+
+void PrintTextDetParams(const TextDetParams &p) {
+  std::cout << "{";
+  std::cout << "\"limit_side_len\": " << p.text_det_limit_side_len << ", ";
+  std::cout << "\"limit_type\": \"" << p.text_det_limit_type << "\", ";
+  std::cout << "\"thresh\": " << p.text_det_thresh << ", ";
+  std::cout << "\"max_side_limit\": " << p.text_det_max_side_limit << ", ";
+  std::cout << "\"box_thresh\": " << p.text_det_box_thresh << ", ";
+  std::cout << "\"unclip_ratio\": " << p.text_det_unclip_ratio;
+  std::cout << "}";
+}
+
+void OCRResult::Print() const {
+  std::cout << "{\n";
+  std::cout << "  \"input_path\": \"" << pipeline_result_.input_path << "\",\n";
+  if (pipeline_result_.model_settings.at("use_doc_preprocessor")) {
+    std::cout << "  \"doc_preprocessor_res\": ";
+    PrintDocPreprocessorPipelineResult(pipeline_result_.doc_preprocessor_res);
+    std::cout << ",\n";
+  }
+  std::cout << "  \"dt_polys\": ";
+  PrintPolys(pipeline_result_.dt_polys);
+  std::cout << ",\n";
+  std::cout << "  \"model_settings\": ";
+  PrintModelSettings(pipeline_result_.model_settings);
+  std::cout << ",\n";
+  std::cout << "  \"text_det_params\": ";
+  PrintTextDetParams(pipeline_result_.text_det_params);
+  std::cout << ",\n";
+  std::cout << "  \"text_type\": \"" << pipeline_result_.text_type << "\",\n";
+  std::cout << "  \"text_rec_score_thresh\": "
+            << pipeline_result_.text_rec_score_thresh << ",\n";
+  std::cout << "  \"rec_texts\": ";
+  PrintStringArray(pipeline_result_.rec_texts);
+  std::cout << ",\n";
+  std::cout << "  \"rec_scores\": ";
+  PrintArray(pipeline_result_.rec_scores);
+  std::cout << ",\n";
+  std::cout << "  \"textline_orientation_angles\": ";
+  PrintIntArray(pipeline_result_.textline_orientation_angles);
+  std::cout << ",\n";
+  std::cout << "  \"rec_polys\": ";
+  PrintPolys(pipeline_result_.rec_polys);
+  std::cout << ",\n";
+  std::cout << "  \"rec_boxes\": ";
+  PrintRecBoxes(pipeline_result_.rec_boxes);
+  std::cout << "\n}" << std::endl;
+}
diff --git a/deploy/cpp_infer/src/pipelines/ocr/result.h b/deploy/cpp_infer/src/pipelines/ocr/result.h
new file mode 100644
index 0000000..8d3e3ed
--- /dev/null
+++ b/deploy/cpp_infer/src/pipelines/ocr/result.h
@@ -0,0 +1,58 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#ifdef USE_FREETYPE
+#include <opencv2/freetype.hpp>
+#endif
+#include "pipeline.h"
+#include "src/base/base_cv_result.h"
+
+class OCRResult : public BaseCVResult {
+public:
+  OCRResult(OCRPipelineResult pipeline_result_)
+      : BaseCVResult(), pipeline_result_(pipeline_result_){};
+
+  void SaveToImg(const std::string &save_path) override;
+  void Print() const override;
+  void SaveToJson(const std::string &save_path) const override;
+
+#ifdef USE_FREETYPE
+  static cv::Mat DrawBoxTextFine(const cv::Size &img_ize,
+                                 const std::vector<cv::Point2f> &box,
+                                 const std::string &txt,
+                                 const std::string &vis_font);
+
+  static void DrawVerticalText(cv::Ptr<cv::freetype::FreeType2> &ft2,
+                               cv::Mat &img, const std::string &text, int x,
+                               int y, int font_height, cv::Scalar color,
+                               float line_spacing = 2);
+  static int CreateFont(cv::Ptr<cv::freetype::FreeType2> &ft2,
+                        const std::string &text, int region_height,
+                        int region_width);
+
+  static int CreateFontVertical(cv::Ptr<cv::freetype::FreeType2> &ft2,
+                                const std::string &text, int region_height,
+                                int region_width, float scale = 1.2f);
+  static cv::Size getActualCharSize(cv::Ptr<cv::freetype::FreeType2> &ft2,
+                                    const std::string &utf8_char,
+                                    int font_height);
+#endif
+  static std::vector<cv::Point>
+  GetMinareaRect(const std::vector<cv::Point> &points);
+
+private:
+  OCRPipelineResult pipeline_result_;
+};
diff --git a/deploy/cpp_infer/src/utils/args.cc b/deploy/cpp_infer/src/utils/args.cc
new file mode 100644
index 0000000..a3611da
--- /dev/null
+++ b/deploy/cpp_infer/src/utils/args.cc
@@ -0,0 +1,101 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "args.h"
+
+DEFINE_string(input, "",
+              "Data to be predicted, required. Local path of an image file.");
+DEFINE_string(save_path, "./output/", "Path to save inference result files.");
+DEFINE_string(doc_orientation_classify_model_name, "PP-LCNet_x1_0_doc_ori",
+              "Name of the document image orientation classification model.");
+DEFINE_string(
+    doc_orientation_classify_model_dir, "",
+    "Path to the document image orientation classification model directory.");
+DEFINE_string(doc_unwarping_model_name, "UVDoc",
+              "Name of the text image unwarping model.");
+DEFINE_string(doc_unwarping_model_dir, "",
+              "Path to the image unwarping model directory.");
+DEFINE_string(text_detection_model_name, "PP-OCRv5_server_det",
+              "Name of the text detection model.");
+DEFINE_string(text_detection_model_dir, "",
+              "Path to the text detection model directory.");
+DEFINE_string(textline_orientation_model_name, "PP-LCNet_x1_0_textline_ori",
+              "Name of the text line orientation classification model.");
+DEFINE_string(
+    textline_orientation_model_dir, "",
+    "Path to the text line orientation classification model directory.");
+DEFINE_string(textline_orientation_batch_size, "6",
+              "Batch size for the text line orientation classification model.");
+DEFINE_string(text_recognition_model_name, "PP-OCRv5_server_rec",
+              "Name of the text recognition model.");
+DEFINE_string(text_recognition_model_dir, "",
+              "Path to the text recognition model directory.");
+DEFINE_string(text_recognition_batch_size, "6",
+              "Batch size for the text recognition model.");
+DEFINE_string(use_doc_orientation_classify, "true",
+              "Whether to use document image orientation classification.");
+DEFINE_string(use_doc_unwarping, "true",
+              "Whether to use text image unwarping.");
+DEFINE_string(use_textline_orientation, "true",
+              "Whether to use text line orientation classification.");
+DEFINE_string(text_det_limit_side_len, "64",
+              "This sets a limit on the side length of the input image for the "
+              "text detection model.");
+DEFINE_string(text_det_limit_type, "min",
+              "This determines how the side length limit is applied to the "
+              "input image before feeding it into the text detection model.");
+DEFINE_string(text_det_thresh, "0.3",
+              "Detection pixel threshold for the text detection model. Pixels "
+              "with scores greater than this threshold in the output "
+              "probability map are considered text pixels.");
+DEFINE_string(
+    text_det_box_thresh, "0.6",
+    "Detection box threshold for the text detection model. A detection result "
+    "is considered a text region if the average score of all pixels within the "
+    "border of the result is greater than this threshold.");
+DEFINE_string(
+    text_det_unclip_ratio, "1.5",
+    "Text detection expansion coefficient, which expands the text region using "
+    "this method. The larger the value, the larger the expansion area.");
+DEFINE_string(text_det_input_shape, "",
+              "Input shape of the text detection model.eg C,H,W");
+DEFINE_string(text_rec_score_thresh, "0",
+              "Text recognition threshold. Text results with scores greater "
+              "than this threshold are retained.");
+DEFINE_string(text_rec_input_shape, "",
+              "Input shape of the text recognition model.eg C,H,W");
+DEFINE_string(lang, "", "Language in the input image for OCR processing.");
+DEFINE_string(ocr_version, "", "PP-OCR version to use.");
+#ifdef WITH_GPU
+DEFINE_string(device, "gpu:0",
+              "Device for inference. Supports specifying a specific card "
+              "number: gpu:0.");
+#else
+DEFINE_string(device, "cpu",
+              "Device for inference. Supports specifying a specific card "
+              "number: gpu:0.");
+#endif
+DEFINE_string(vis_font_dir, "",
+              "When enable USE_FREETYPE, required. Path to the visualization "
+              "font, render the detected texts on images");
+DEFINE_string(precision, "fp32",
+              "Computational precision, such as fp32, fp16.");
+DEFINE_string(enable_mkldnn, "true", "enable_mkldnn");
+DEFINE_string(mkldnn_cache_capacity, "10", "MKL-DNN cache capacity.");
+DEFINE_string(cpu_threads, "8",
+              "Number of threads used for paddlepaddle inference on CPU.");
+DEFINE_string(thread_num, "1",
+              "Number of threads used for pipeline instance inference on CPU.");
+DEFINE_string(paddlex_config, "",
+              "Path to the PaddleX pipeline configuration file.");
diff --git a/deploy/cpp_infer/src/utils/args.h b/deploy/cpp_infer/src/utils/args.h
new file mode 100644
index 0000000..af8091d
--- /dev/null
+++ b/deploy/cpp_infer/src/utils/args.h
@@ -0,0 +1,53 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <gflags/gflags.h>
+
+DECLARE_string(input);
+DECLARE_string(save_path);
+DECLARE_string(doc_orientation_classify_model_name);
+DECLARE_string(doc_orientation_classify_model_dir);
+DECLARE_string(doc_unwarping_model_name);
+DECLARE_string(doc_unwarping_model_dir);
+DECLARE_string(text_detection_model_name);
+DECLARE_string(text_detection_model_dir);
+DECLARE_string(textline_orientation_model_name);
+DECLARE_string(textline_orientation_model_dir);
+DECLARE_string(textline_orientation_batch_size);
+DECLARE_string(text_recognition_model_name);
+DECLARE_string(text_recognition_model_dir);
+DECLARE_string(text_recognition_batch_size);
+DECLARE_string(use_doc_orientation_classify);
+DECLARE_string(use_doc_unwarping);
+DECLARE_string(use_textline_orientation);
+DECLARE_string(text_det_limit_side_len);
+DECLARE_string(text_det_limit_type);
+DECLARE_string(text_det_thresh);
+DECLARE_string(text_det_box_thresh);
+DECLARE_string(text_det_unclip_ratio);
+DECLARE_string(text_det_input_shape);
+DECLARE_string(text_rec_score_thresh);
+DECLARE_string(text_rec_input_shape);
+DECLARE_string(lang);
+DECLARE_string(ocr_version);
+DECLARE_string(device);
+DECLARE_string(vis_font_dir);
+DECLARE_string(precision);
+DECLARE_string(enable_mkldnn);
+DECLARE_string(mkldnn_cache_capacity);
+DECLARE_string(cpu_threads);
+DECLARE_string(thread_num);
+DECLARE_string(paddlex_config);
diff --git a/deploy/cpp_infer/src/utils/func_register.h b/deploy/cpp_infer/src/utils/func_register.h
new file mode 100644
index 0000000..bc466b1
--- /dev/null
+++ b/deploy/cpp_infer/src/utils/func_register.h
@@ -0,0 +1,32 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <iostream>
+#include <memory>
+#include <opencv2/opencv.hpp>
+#include <string>
+#include <unordered_map>
+
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+
+class BaseProcessor {
+public:
+  BaseProcessor() = default;
+  virtual ~BaseProcessor() = default;
+  virtual absl::StatusOr<std::vector<cv::Mat>>
+  Apply(std::vector<cv::Mat> &input, const void *param_ptr = nullptr) const = 0;
+};
diff --git a/deploy/cpp_infer/src/utils/ilogger.cc b/deploy/cpp_infer/src/utils/ilogger.cc
new file mode 100644
index 0000000..fef87f4
--- /dev/null
+++ b/deploy/cpp_infer/src/utils/ilogger.cc
@@ -0,0 +1,1238 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Based on https://github.com/shouxieai/tensorRT_Pro
+
+// Copyright (c) 2022 TensorRTPro
+
+// Permission is hereby granted, free of charge, to any person obtaining a copy
+// of this software && associated documentation files (the "Software"), to deal
+// in the Software without restriction, including without limitation the rights
+// to use, copy, modify, merge, publish, distribute, sublicense, &&/||  sell
+// copies of the Software, && to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+
+// The above copyright notice && this permission notice shall be included in
+// all copies ||  substantial portions of the Software.
+
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+
+#include "ilogger.h"
+
+#include <math.h>
+#include <signal.h>
+#include <stdarg.h>
+#include <string.h>
+#include <time.h>
+
+#include <algorithm>
+#include <atomic>
+#include <fstream>
+#include <functional>
+#include <memory>
+#include <mutex>
+#include <sstream>
+#include <stack>
+#include <string>
+#include <thread>
+#include <vector>
+
+#if defined(U_OS_WINDOWS)
+#define HAS_UUID
+#include <Shlwapi.h>
+#include <Windows.h>
+#include <wingdi.h>
+#pragma comment(lib, "shlwapi.lib")
+#pragma comment(lib, "ole32.lib")
+#pragma comment(lib, "gdi32.lib")
+#undef min
+#undef max
+#endif
+
+#if defined(U_OS_LINUX)
+//# include <sys/io.h>
+#include <dirent.h>
+#include <stdarg.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <unistd.h>
+#define strtok_s strtok_r
+#endif
+
+#if defined(U_OS_LINUX)
+#define __GetTimeBlock                                                         \
+  time_t timep;                                                                \
+  time(&timep);                                                                \
+  tm &t = *(tm *)localtime(&timep);
+#endif
+
+#if defined(U_OS_WINDOWS)
+#define __GetTimeBlock                                                         \
+  tm t;                                                                        \
+  _getsystime(&t);
+#endif
+
+namespace iLogger {
+
+using namespace std;
+
+const char *level_string(LogLevel level) {
+  switch (level) {
+  case LogLevel::Debug:
+    return "debug";
+  case LogLevel::Verbose:
+    return "verbo";
+  case LogLevel::Info:
+    return "info";
+  case LogLevel::Warning:
+    return "warn";
+  case LogLevel::Error:
+    return "error";
+  case LogLevel::Fatal:
+    return "fatal";
+  default:
+    return "unknown";
+  }
+}
+
+std::tuple<uint8_t, uint8_t, uint8_t> hsv2bgr(float h, float s, float v) {
+  const int h_i = static_cast<int>(h * 6);
+  const float f = h * 6 - h_i;
+  const float p = v * (1 - s);
+  const float q = v * (1 - f * s);
+  const float t = v * (1 - (1 - f) * s);
+  float r, g, b;
+  switch (h_i) {
+  case 0:
+    r = v;
+    g = t;
+    b = p;
+    break;
+  case 1:
+    r = q;
+    g = v;
+    b = p;
+    break;
+  case 2:
+    r = p;
+    g = v;
+    b = t;
+    break;
+  case 3:
+    r = p;
+    g = q;
+    b = v;
+    break;
+  case 4:
+    r = t;
+    g = p;
+    b = v;
+    break;
+  case 5:
+    r = v;
+    g = p;
+    b = q;
+    break;
+  default:
+    r = 1;
+    g = 1;
+    b = 1;
+    break;
+  }
+  return make_tuple(static_cast<uint8_t>(b * 255),
+                    static_cast<uint8_t>(g * 255),
+                    static_cast<uint8_t>(r * 255));
+}
+
+std::tuple<uint8_t, uint8_t, uint8_t> random_color(int id) {
+  float h_plane = ((((unsigned int)id << 2) ^ 0x937151) % 100) / 100.0f;
+  ;
+  float s_plane = ((((unsigned int)id << 3) ^ 0x315793) % 100) / 100.0f;
+  return hsv2bgr(h_plane, s_plane, 1);
+}
+
+string date_now() {
+  char time_string[20];
+  __GetTimeBlock;
+
+  sprintf(time_string, "%04d-%02d-%02d", t.tm_year + 1900, t.tm_mon + 1,
+          t.tm_mday);
+  return time_string;
+}
+
+string time_now() {
+  char time_string[20];
+  __GetTimeBlock;
+
+  sprintf(time_string, "%04d-%02d-%02d %02d:%02d:%02d", t.tm_year + 1900,
+          t.tm_mon + 1, t.tm_mday, t.tm_hour, t.tm_min, t.tm_sec);
+  return time_string;
+}
+
+size_t file_size(const string &file) {
+#if defined(U_OS_LINUX)
+  struct stat st;
+  stat(file.c_str(), &st);
+  return st.st_size;
+#elif defined(U_OS_WINDOWS)
+  WIN32_FIND_DATAA find_data;
+  HANDLE hFind = FindFirstFileA(file.c_str(), &find_data);
+  if (hFind == INVALID_HANDLE_VALUE)
+    return 0;
+
+  FindClose(hFind);
+  return (uint64_t)find_data.nFileSizeLow |
+         ((uint64_t)find_data.nFileSizeHigh << 32);
+#endif
+}
+
+time_t last_modify(const string &file) {
+#if defined(U_OS_LINUX)
+  struct stat st;
+  stat(file.c_str(), &st);
+  return st.st_mtim.tv_sec;
+#elif defined(U_OS_WINDOWS)
+  INFOW("LastModify has not support on windows os");
+  return 0;
+#endif
+}
+
+void sleep(int ms) { this_thread::sleep_for(std::chrono::milliseconds(ms)); }
+
+int get_month_by_name(char *month) {
+  if (strcmp(month, "Jan") == 0)
+    return 0;
+  if (strcmp(month, "Feb") == 0)
+    return 1;
+  if (strcmp(month, "Mar") == 0)
+    return 2;
+  if (strcmp(month, "Apr") == 0)
+    return 3;
+  if (strcmp(month, "May") == 0)
+    return 4;
+  if (strcmp(month, "Jun") == 0)
+    return 5;
+  if (strcmp(month, "Jul") == 0)
+    return 6;
+  if (strcmp(month, "Aug") == 0)
+    return 7;
+  if (strcmp(month, "Sep") == 0)
+    return 8;
+  if (strcmp(month, "Oct") == 0)
+    return 9;
+  if (strcmp(month, "Nov") == 0)
+    return 10;
+  if (strcmp(month, "Dec") == 0)
+    return 11;
+  return -1;
+}
+
+int get_week_day_by_name(char *wday) {
+  if (strcmp(wday, "Sun") == 0)
+    return 0;
+  if (strcmp(wday, "Mon") == 0)
+    return 1;
+  if (strcmp(wday, "Tue") == 0)
+    return 2;
+  if (strcmp(wday, "Wed") == 0)
+    return 3;
+  if (strcmp(wday, "Thu") == 0)
+    return 4;
+  if (strcmp(wday, "Fri") == 0)
+    return 5;
+  if (strcmp(wday, "Sat") == 0)
+    return 6;
+  return -1;
+}
+
+time_t gmtime2ctime(const string &gmt) {
+  char week[4] = {0};
+  char month[4] = {0};
+  tm date;
+  sscanf(gmt.c_str(), "%3s, %2d %3s %4d %2d:%2d:%2d GMT", week, &date.tm_mday,
+         month, &date.tm_year, &date.tm_hour, &date.tm_min, &date.tm_sec);
+  date.tm_mon = get_month_by_name(month);
+  date.tm_wday = get_week_day_by_name(week);
+  date.tm_year = date.tm_year - 1900;
+  return mktime(&date);
+}
+
+string gmtime(time_t t) {
+  t += 28800;
+  tm *gmt = ::gmtime(&t);
+
+  // http://en.cppreference.com/w/c/chrono/strftime
+  // e.g.: Sat, 22 Aug 2015 11:48:50 GMT
+  //       5+   3+4+   5+   9+       3   = 29
+  const char *fmt = "%a, %d %b %Y %H:%M:%S GMT";
+  char tstr[30];
+  strftime(tstr, sizeof(tstr), fmt, gmt);
+  return tstr;
+}
+
+string gmtime_now() { return gmtime(time(nullptr)); }
+
+bool mkdir(const string &path) {
+#ifdef U_OS_WINDOWS
+  return CreateDirectoryA(path.c_str(), nullptr);
+#else
+  return ::mkdir(path.c_str(), 0755) == 0;
+#endif
+}
+
+bool mkdirs(const string &path) {
+  if (path.empty())
+    return false;
+  if (exists(path))
+    return true;
+
+  string _path = path;
+  char *dir_ptr = (char *)_path.c_str();
+  char *iter_ptr = dir_ptr;
+
+  bool keep_going = *iter_ptr != 0;
+  while (keep_going) {
+    if (*iter_ptr == 0)
+      keep_going = false;
+
+#ifdef U_OS_WINDOWS
+    if (*iter_ptr == '/' || *iter_ptr == '\\' || *iter_ptr == 0) {
+#else
+    if ((*iter_ptr == '/' && iter_ptr != dir_ptr) || *iter_ptr == 0) {
+#endif
+      char old = *iter_ptr;
+      *iter_ptr = 0;
+      if (!exists(dir_ptr)) {
+        if (!mkdir(dir_ptr)) {
+          if (!exists(dir_ptr)) {
+            INFOE("mkdirs %s return false.", dir_ptr);
+            return false;
+          }
+        }
+      }
+      *iter_ptr = old;
+    }
+    iter_ptr++;
+  }
+  return true;
+}
+
+bool isfile(const string &file) {
+#if defined(U_OS_LINUX)
+  struct stat st;
+  stat(file.c_str(), &st);
+  return S_ISREG(st.st_mode);
+#elif defined(U_OS_WINDOWS)
+  INFOW("is_file has not support on windows os");
+  return 0;
+#endif
+}
+
+FILE *fopen_mkdirs(const string &path, const string &mode) {
+  FILE *f = fopen(path.c_str(), mode.c_str());
+  if (f)
+    return f;
+
+  int p = path.rfind('/');
+
+#if defined(U_OS_WINDOWS)
+  int e = path.rfind('\\');
+  p = std::max(p, e);
+#endif
+  if (p == -1)
+    return nullptr;
+
+  string directory = path.substr(0, p);
+  if (!mkdirs(directory))
+    return nullptr;
+
+  return fopen(path.c_str(), mode.c_str());
+}
+
+bool exists(const string &path) {
+#ifdef U_OS_WINDOWS
+  return ::PathFileExistsA(path.c_str());
+#elif defined(U_OS_LINUX)
+  return access(path.c_str(), R_OK) == 0;
+#endif
+}
+
+string format(const char *fmt, ...) {
+  va_list vl;
+  va_start(vl, fmt);
+  char buffer[2048];
+  vsnprintf(buffer, sizeof(buffer), fmt, vl);
+  return buffer;
+}
+
+string file_name(const string &path, bool include_suffix) {
+  if (path.empty())
+    return "";
+
+  int p = path.rfind('/');
+
+#ifdef U_OS_WINDOWS
+  int e = path.rfind('\\');
+  p = std::max(p, e);
+#endif
+  p += 1;
+
+  // include suffix
+  if (include_suffix)
+    return path.substr(p);
+
+  int u = path.rfind('.');
+  if (u == -1)
+    return path.substr(p);
+
+  if (u <= p)
+    u = path.size();
+  return path.substr(p, u - p);
+}
+
+string directory(const string &path) {
+  if (path.empty())
+    return ".";
+
+  int p = path.rfind('/');
+
+#ifdef U_OS_WINDOWS
+  int e = path.rfind('\\');
+  p = std::max(p, e);
+#endif
+  if (p == -1)
+    return ".";
+
+  return path.substr(0, p + 1);
+}
+
+bool begin_with(const string &str, const string &with) {
+  if (str.length() < with.length())
+    return false;
+  return strncmp(str.c_str(), with.c_str(), with.length()) == 0;
+}
+
+bool end_with(const string &str, const string &with) {
+  if (str.length() < with.length())
+    return false;
+
+  return strncmp(str.c_str() + str.length() - with.length(), with.c_str(),
+                 with.length()) == 0;
+}
+
+long long timestamp_now() {
+#ifdef _WIN32
+  FILETIME ft;
+  GetSystemTimeAsFileTime(&ft);
+  ULARGE_INTEGER li;
+  li.LowPart = ft.dwLowDateTime;
+  li.HighPart = ft.dwHighDateTime;
+  return (li.QuadPart - 116444736000000000LL) / 10000LL;
+#else
+  return chrono::duration_cast<chrono::milliseconds>(
+             chrono::system_clock::now().time_since_epoch())
+      .count();
+#endif
+}
+
+double timestamp_now_float() {
+#ifdef _WIN32
+  FILETIME ft;
+  GetSystemTimeAsFileTime(&ft);
+  ULARGE_INTEGER li;
+  li.LowPart = ft.dwLowDateTime;
+  li.HighPart = ft.dwHighDateTime;
+  return ((li.QuadPart - 116444736000000000LL) / 10000.0);
+#else
+  return chrono::duration_cast<chrono::microseconds>(
+             chrono::system_clock::now().time_since_epoch())
+             .count() /
+         1000.0;
+#endif
+}
+
+static struct Logger {
+  mutex logger_lock_;
+  string logger_directory;
+  LogLevel logger_level{LogLevel::Info};
+  vector<string> cache_, local_;
+  shared_ptr<thread> flush_thread_;
+  atomic<bool> keep_run_{false};
+  shared_ptr<FILE> handler;
+  bool logger_shutdown{false};
+
+  void write(const string &line) {
+    lock_guard<mutex> l(logger_lock_);
+    if (logger_shutdown)
+      return;
+
+    if (!keep_run_) {
+      if (flush_thread_)
+        return;
+
+      cache_.reserve(1000);
+      keep_run_ = true;
+      flush_thread_.reset(new thread(std::bind(&Logger::flush_job, this)));
+    }
+    cache_.emplace_back(line);
+  }
+
+  void flush() {
+    if (cache_.empty())
+      return;
+
+    {
+      std::lock_guard<mutex> l(logger_lock_);
+      std::swap(local_, cache_);
+    }
+
+    if (!local_.empty() && !logger_directory.empty()) {
+      auto now = date_now();
+      auto file = format("%s%s.txt", logger_directory.c_str(), now.c_str());
+      if (!exists(file)) {
+        handler.reset(fopen_mkdirs(file, "wb"), fclose);
+      } else if (!handler) {
+        handler.reset(fopen_mkdirs(file, "a+"), fclose);
+      }
+
+      if (handler) {
+        for (auto &line : local_)
+          fprintf(handler.get(), "%s\n", line.c_str());
+        fflush(handler.get());
+        handler.reset();
+      }
+    }
+    local_.clear();
+  }
+
+  void flush_job() {
+    auto tick_begin = timestamp_now();
+    std::vector<string> local;
+    while (keep_run_) {
+      if (timestamp_now() - tick_begin < 1000) {
+        this_thread::sleep_for(std::chrono::milliseconds(100));
+        continue;
+      }
+
+      tick_begin = timestamp_now();
+      flush();
+    }
+    flush();
+  }
+
+  void set_save_directory(const string &loggerDirectory) {
+    logger_directory = loggerDirectory;
+
+    if (logger_directory.empty())
+      logger_directory = ".";
+
+#if defined(U_OS_LINUX)
+    if (logger_directory.back() != '/') {
+      logger_directory.push_back('/');
+    }
+#endif
+
+#if defined(U_OS_WINDOWS)
+    if (logger_directory.back() != '/' && logger_directory.back() != '\\') {
+      logger_directory.push_back('/');
+    }
+#endif
+  }
+
+  void set_logger_level(LogLevel level) { logger_level = level; }
+
+  void close() {
+    {
+      lock_guard<mutex> l(logger_lock_);
+      if (logger_shutdown)
+        return;
+
+      logger_shutdown = true;
+    };
+
+    if (!keep_run_)
+      return;
+    keep_run_ = false;
+    flush_thread_->join();
+    flush_thread_.reset();
+    handler.reset();
+  }
+
+  virtual ~Logger() { close(); }
+} __g_logger;
+
+void destroy_logger() { __g_logger.close(); }
+
+static void remove_color_text(char *buffer) {
+  //"\033[31m%s\033[0m"
+  char *p = buffer;
+  while (*p) {
+    if (*p == 0x1B) {
+      char np = *(p + 1);
+      if (np == '[') {
+        // has token
+        char *t = p + 2;
+        while (*t) {
+          if (*t == 'm') {
+            t = t + 1;
+            char *k = p;
+            while (*t) {
+              *k++ = *t++;
+            }
+            *k = 0;
+            break;
+          }
+          t++;
+        }
+      }
+    }
+    p++;
+  }
+}
+
+void set_logger_save_directory(const string &loggerDirectory) {
+  __g_logger.set_save_directory(loggerDirectory);
+}
+
+void set_log_level(LogLevel level) { __g_logger.set_logger_level(level); }
+
+LogLevel get_log_level() { return __g_logger.logger_level; }
+
+void __log_func(const char *file, int line, LogLevel level, const char *fmt,
+                ...) {
+  if (level > __g_logger.logger_level)
+    return;
+
+  string now = time_now();
+  va_list vl;
+  va_start(vl, fmt);
+
+  char buffer[2048];
+  string filename = file_name(file, true);
+  int n = snprintf(buffer, sizeof(buffer), "[%s]", now.c_str());
+
+#if defined(U_OS_WINDOWS)
+  if (level == LogLevel::Fatal || level == LogLevel::Error) {
+    n += snprintf(buffer + n, sizeof(buffer) - n, "[%s]", level_string(level));
+  } else if (level == LogLevel::Warning) {
+    n += snprintf(buffer + n, sizeof(buffer) - n, "[%s]", level_string(level));
+  } else {
+    n += snprintf(buffer + n, sizeof(buffer) - n, "[%s]", level_string(level));
+  }
+#elif defined(U_OS_LINUX)
+  if (level == LogLevel::Fatal || level == LogLevel::Error) {
+    n += snprintf(buffer + n, sizeof(buffer) - n, "[\033[31m%s\033[0m]",
+                  level_string(level));
+  } else if (level == LogLevel::Warning) {
+    n += snprintf(buffer + n, sizeof(buffer) - n, "[\033[33m%s\033[0m]",
+                  level_string(level));
+  } else if (level == LogLevel::Info) {
+    n += snprintf(buffer + n, sizeof(buffer) - n, "[\033[35m%s\033[0m]",
+                  level_string(level));
+  } else if (level == LogLevel::Verbose) {
+    n += snprintf(buffer + n, sizeof(buffer) - n, "[\033[34m%s\033[0m]",
+                  level_string(level));
+  } else {
+    n += snprintf(buffer + n, sizeof(buffer) - n, "[%s]", level_string(level));
+  }
+#endif
+
+  n += snprintf(buffer + n, sizeof(buffer) - n, "[%s:%d]:", filename.c_str(),
+                line);
+  vsnprintf(buffer + n, sizeof(buffer) - n, fmt, vl);
+
+  if (level == LogLevel::Fatal || level == LogLevel::Error) {
+    fprintf(stderr, "%s\n", buffer);
+  } else if (level == LogLevel::Warning) {
+    fprintf(stdout, "%s\n", buffer);
+  } else {
+    fprintf(stdout, "%s\n", buffer);
+  }
+
+  if (!__g_logger.logger_directory.empty()) {
+#ifdef U_OS_LINUX
+    // remove save color txt
+    remove_color_text(buffer);
+#endif
+    __g_logger.write(buffer);
+    if (level == LogLevel::Fatal) {
+      __g_logger.flush();
+    }
+  }
+
+  if (level == LogLevel::Fatal) {
+    fflush(stdout);
+    abort();
+  }
+}
+
+string load_text_file(const string &file) {
+  ifstream in(file, ios::in | ios::binary);
+  if (!in.is_open())
+    return {};
+
+  in.seekg(0, ios::end);
+  size_t length = in.tellg();
+
+  string data;
+  if (length > 0) {
+    in.seekg(0, ios::beg);
+    data.resize(length);
+
+    in.read((char *)&data[0], length);
+  }
+  in.close();
+  return data;
+}
+
+std::vector<uint8_t> load_file(const string &file) {
+  ifstream in(file, ios::in | ios::binary);
+  if (!in.is_open())
+    return {};
+
+  in.seekg(0, ios::end);
+  size_t length = in.tellg();
+
+  std::vector<uint8_t> data;
+  if (length > 0) {
+    in.seekg(0, ios::beg);
+    data.resize(length);
+
+    in.read((char *)&data[0], length);
+  }
+  in.close();
+  return data;
+}
+
+bool alphabet_equal(char a, char b, bool ignore_case) {
+  if (ignore_case) {
+    a = a > 'a' && a < 'z' ? a - 'a' + 'A' : a;
+    b = b > 'a' && b < 'z' ? b - 'a' + 'A' : b;
+  }
+  return a == b;
+}
+
+static bool pattern_match_body(const char *str, const char *matcher,
+                               bool igrnoe_case) {
+  //   abcdefg.pnga          *.png      > false
+  //   abcdefg.png           *.png      > true
+  //   abcdefg.png          a?cdefg.png > true
+
+  if (!matcher || !*matcher || !str || !*str)
+    return false;
+
+  const char *ptr_matcher = matcher;
+  while (*str) {
+    if (*ptr_matcher == '?') {
+      ptr_matcher++;
+    } else if (*ptr_matcher == '*') {
+      if (*(ptr_matcher + 1)) {
+        if (pattern_match_body(str, ptr_matcher + 1, igrnoe_case))
+          return true;
+      } else {
+        return true;
+      }
+    } else if (!alphabet_equal(*ptr_matcher, *str, igrnoe_case)) {
+      return false;
+    } else {
+      if (*ptr_matcher)
+        ptr_matcher++;
+      else
+        return false;
+    }
+    str++;
+  }
+
+  while (*ptr_matcher) {
+    if (*ptr_matcher != '*')
+      return false;
+    ptr_matcher++;
+  }
+  return true;
+}
+
+bool pattern_match(const char *str, const char *matcher, bool igrnoe_case) {
+  //   abcdefg.pnga          *.png      > false
+  //   abcdefg.png           *.png      > true
+  //   abcdefg.png          a?cdefg.png > true
+
+  if (!matcher || !*matcher || !str || !*str)
+    return false;
+
+  char filter[500];
+  strcpy(filter, matcher);
+
+  vector<const char *> arr;
+  char *ptr_str = filter;
+  char *ptr_prev_str = ptr_str;
+  while (*ptr_str) {
+    if (*ptr_str == ';') {
+      *ptr_str = 0;
+      arr.push_back(ptr_prev_str);
+      ptr_prev_str = ptr_str + 1;
+    }
+    ptr_str++;
+  }
+
+  if (*ptr_prev_str)
+    arr.push_back(ptr_prev_str);
+
+  for (int i = 0; i < arr.size(); ++i) {
+    if (pattern_match_body(str, arr[i], igrnoe_case))
+      return true;
+  }
+  return false;
+}
+
+#ifdef U_OS_WINDOWS
+vector<string> find_files(const string &directory, const string &filter,
+                          bool findDirectory, bool includeSubDirectory) {
+  string realpath = directory;
+  if (realpath.empty())
+    realpath = "./";
+
+  char backchar = realpath.back();
+  if (backchar != '\\' && backchar != '/')
+    realpath += "/";
+
+  vector<string> out;
+  _WIN32_FIND_DATAA find_data;
+  stack<string> ps;
+  ps.push(realpath);
+
+  while (!ps.empty()) {
+    string search_path = ps.top();
+    ps.pop();
+
+    HANDLE hFind = FindFirstFileA((search_path + "*").c_str(), &find_data);
+    if (hFind != INVALID_HANDLE_VALUE) {
+      do {
+        if (strcmp(find_data.cFileName, ".") == 0 ||
+            strcmp(find_data.cFileName, "..") == 0)
+          continue;
+
+        if (!findDirectory &&
+                (find_data.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY) !=
+                    FILE_ATTRIBUTE_DIRECTORY ||
+            findDirectory &&
+                (find_data.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY) ==
+                    FILE_ATTRIBUTE_DIRECTORY) {
+          if (PathMatchSpecA(find_data.cFileName, filter.c_str()))
+            out.push_back(search_path + find_data.cFileName);
+        }
+
+        if (includeSubDirectory &&
+            (find_data.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY) ==
+                FILE_ATTRIBUTE_DIRECTORY)
+          ps.push(search_path + find_data.cFileName + "/");
+
+      } while (FindNextFileA(hFind, &find_data));
+      FindClose(hFind);
+    }
+  }
+  return out;
+}
+#endif
+
+#ifdef U_OS_LINUX
+vector<string> find_files(const string &directory, const string &filter,
+                          bool findDirectory, bool includeSubDirectory) {
+  string realpath = directory;
+  if (realpath.empty())
+    realpath = "./";
+
+  char backchar = realpath.back();
+  if (backchar != '\\' && backchar != '/')
+    realpath += "/";
+
+  struct dirent *fileinfo;
+  DIR *handle;
+  stack<string> ps;
+  vector<string> out;
+  ps.push(realpath);
+
+  while (!ps.empty()) {
+    string search_path = ps.top();
+    ps.pop();
+
+    handle = opendir(search_path.c_str());
+    if (handle != 0) {
+      while (fileinfo = readdir(handle)) {
+        struct stat file_stat;
+        if (strcmp(fileinfo->d_name, ".") == 0 ||
+            strcmp(fileinfo->d_name, "..") == 0)
+          continue;
+
+        if (lstat((search_path + fileinfo->d_name).c_str(), &file_stat) < 0)
+          continue;
+
+        if (!findDirectory && !S_ISDIR(file_stat.st_mode) ||
+            findDirectory && S_ISDIR(file_stat.st_mode)) {
+          if (pattern_match(fileinfo->d_name, filter.c_str()))
+            out.push_back(search_path + fileinfo->d_name);
+        }
+
+        if (includeSubDirectory && S_ISDIR(file_stat.st_mode))
+          ps.push(search_path + fileinfo->d_name + "/");
+      }
+      closedir(handle);
+    }
+  }
+  return out;
+}
+#endif
+
+string align_blank(const string &input, int align_size, char blank) {
+  if (input.size() >= align_size)
+    return input;
+  string output = input;
+  for (int i = 0; i < align_size - input.size(); ++i)
+    output.push_back(blank);
+  return output;
+}
+
+vector<string> split_string(const string &str, const std::string &spstr) {
+  vector<string> res;
+  if (str.empty())
+    return res;
+  if (spstr.empty())
+    return {str};
+
+  auto p = str.find(spstr);
+  if (p == string::npos)
+    return {str};
+
+  res.reserve(5);
+  string::size_type prev = 0;
+  int lent = spstr.length();
+  const char *ptr = str.c_str();
+
+  while (p != string::npos) {
+    int len = p - prev;
+    if (len > 0) {
+      res.emplace_back(str.substr(prev, len));
+    }
+    prev = p + lent;
+    p = str.find(spstr, prev);
+  }
+
+  int len = str.length() - prev;
+  if (len > 0) {
+    res.emplace_back(str.substr(prev, len));
+  }
+  return res;
+}
+
+string replace_string(const string &str, const string &token,
+                      const string &value, int nreplace, int *out_num_replace) {
+  if (nreplace == -1) {
+    nreplace = str.size();
+  }
+
+  if (nreplace == 0) {
+    return str;
+  }
+
+  string result;
+  result.resize(str.size());
+
+  char *dest = &result[0];
+  const char *src = str.c_str();
+  const char *value_ptr = value.c_str();
+  int old_nreplace = nreplace;
+  bool keep = true;
+  string::size_type pos = 0;
+  string::size_type prev = 0;
+  size_t token_length = token.length();
+  size_t value_length = value.length();
+
+  do {
+    pos = str.find(token, pos);
+    if (pos == string::npos) {
+      keep = false;
+      pos = str.length();
+    } else {
+      if (nreplace == 0) {
+        pos = str.length();
+        keep = false;
+      } else {
+        nreplace--;
+      }
+    }
+
+    size_t copy_length = pos - prev;
+    if (copy_length > 0) {
+      size_t dest_length = dest - &result[0];
+
+      // Extended memory space if needed.
+      if (dest_length + copy_length > result.size()) {
+        result.resize((result.size() + copy_length) * 1.2);
+        dest = &result[dest_length];
+      }
+
+      memcpy(dest, src + prev, copy_length);
+      dest += copy_length;
+    }
+
+    if (keep) {
+      pos += token_length;
+      prev = pos;
+
+      size_t dest_length = dest - &result[0];
+
+      // Extended memory space if needed.
+      if (dest_length + value_length > result.size()) {
+        result.resize((result.size() + value_length) * 1.2);
+        dest = &result[dest_length];
+      }
+      memcpy(dest, value_ptr, value_length);
+      dest += value_length;
+    }
+  } while (keep);
+
+  if (out_num_replace) {
+    *out_num_replace = old_nreplace - nreplace;
+  }
+
+  // Crop extra space
+  size_t valid_size = dest - &result[0];
+  result.resize(valid_size);
+  return result;
+}
+
+bool save_file(const string &file, const void *data, size_t length,
+               bool mk_dirs) {
+  if (mk_dirs) {
+    int p = (int)file.rfind('/');
+
+#ifdef U_OS_WINDOWS
+    int e = (int)file.rfind('\\');
+    p = std::max(p, e);
+#endif
+    if (p != -1) {
+      if (!mkdirs(file.substr(0, p)))
+        return false;
+    }
+  }
+
+  FILE *f = fopen(file.c_str(), "wb");
+  if (!f)
+    return false;
+
+  if (data && length > 0) {
+    if (fwrite(data, 1, length, f) != length) {
+      fclose(f);
+      return false;
+    }
+  }
+  fclose(f);
+  return true;
+}
+
+bool save_file(const string &file, const string &data, bool mk_dirs) {
+  return save_file(file, data.data(), data.size(), mk_dirs);
+}
+
+bool save_file(const string &file, const vector<uint8_t> &data, bool mk_dirs) {
+  return save_file(file, data.data(), data.size(), mk_dirs);
+}
+
+static volatile bool g_has_exit_signal = false;
+static int g_signum = 0;
+static void signal_callback_handler(int signum) {
+  INFO("Capture interrupt signal.");
+  g_has_exit_signal = true;
+  g_signum = signum;
+}
+
+int while_loop() {
+  signal(SIGINT, signal_callback_handler);
+  // signal(SIGQUIT, signal_callback_handler);
+  while (!g_has_exit_signal) {
+    this_thread::yield();
+  }
+  INFO("Loop over.");
+  return g_signum;
+}
+
+static unsigned char from_b64(unsigned char ch) {
+  /* Inverse lookup map */
+  static const unsigned char tab[128] = {
+      255, 255, 255, 255,
+      255, 255, 255, 255, /*  0 */
+      255, 255, 255, 255,
+      255, 255, 255, 255, /*  8 */
+      255, 255, 255, 255,
+      255, 255, 255, 255, /*  16 */
+      255, 255, 255, 255,
+      255, 255, 255, 255, /*  24 */
+      255, 255, 255, 255,
+      255, 255, 255, 255, /*  32 */
+      255, 255, 255, 62,
+      255, 255, 255, 63, /*  40 */
+      52,  53,  54,  55,
+      56,  57,  58,  59, /*  48 */
+      60,  61,  255, 255,
+      255, 200, 255, 255, /*  56   '=' is 200, on index 61 */
+      255, 0,   1,   2,
+      3,   4,   5,   6, /*  64 */
+      7,   8,   9,   10,
+      11,  12,  13,  14, /*  72 */
+      15,  16,  17,  18,
+      19,  20,  21,  22, /*  80 */
+      23,  24,  25,  255,
+      255, 255, 255, 255, /*  88 */
+      255, 26,  27,  28,
+      29,  30,  31,  32, /*  96 */
+      33,  34,  35,  36,
+      37,  38,  39,  40, /*  104 */
+      41,  42,  43,  44,
+      45,  46,  47,  48, /*  112 */
+      49,  50,  51,  255,
+      255, 255, 255, 255, /*  120 */
+  };
+  return tab[ch & 127];
+}
+
+string base64_decode(const string &base64) {
+  if (base64.empty())
+    return "";
+
+  int len = base64.size();
+  auto s = (const unsigned char *)base64.data();
+  unsigned char a, b, c, d;
+  int orig_len = len;
+  int dec_len = 0;
+  string out_data;
+
+  auto end_s = s + base64.size();
+  int count_eq = 0;
+  while (*--end_s == '=') {
+    count_eq++;
+  }
+  out_data.resize(len / 4 * 3 - count_eq);
+
+  char *dst = const_cast<char *>(out_data.data());
+  char *orig_dst = dst;
+  while (len >= 4 && (a = from_b64(s[0])) != 255 &&
+         (b = from_b64(s[1])) != 255 && (c = from_b64(s[2])) != 255 &&
+         (d = from_b64(s[3])) != 255) {
+    s += 4;
+    len -= 4;
+    if (a == 200 || b == 200)
+      break; /* '=' can't be there */
+    *dst++ = a << 2 | b >> 4;
+    if (c == 200)
+      break;
+    *dst++ = b << 4 | c >> 2;
+    if (d == 200)
+      break;
+    *dst++ = c << 6 | d;
+  }
+  dec_len = (dst - orig_dst);
+  return out_data;
+}
+
+string base64_encode(const void *data, size_t size) {
+  string encode_result;
+  encode_result.reserve(size / 3 * 4 + (size % 3 != 0 ? 4 : 0));
+
+  const unsigned char *current = static_cast<const unsigned char *>(data);
+  static const char *base64_table =
+      "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";
+  while (size > 2) {
+    encode_result += base64_table[current[0] >> 2];
+    encode_result +=
+        base64_table[((current[0] & 0x03) << 4) + (current[1] >> 4)];
+    encode_result +=
+        base64_table[((current[1] & 0x0f) << 2) + (current[2] >> 6)];
+    encode_result += base64_table[current[2] & 0x3f];
+
+    current += 3;
+    size -= 3;
+  }
+
+  if (size > 0) {
+    encode_result += base64_table[current[0] >> 2];
+    if (size % 3 == 1) {
+      encode_result += base64_table[(current[0] & 0x03) << 4];
+      encode_result += "==";
+    } else if (size % 3 == 2) {
+      encode_result +=
+          base64_table[((current[0] & 0x03) << 4) + (current[1] >> 4)];
+      encode_result += base64_table[(current[1] & 0x0f) << 2];
+      encode_result += "=";
+    }
+  }
+  return encode_result;
+}
+
+bool delete_file(const string &path) {
+#ifdef U_OS_WINDOWS
+  return DeleteFileA(path.c_str());
+#else
+  return ::remove(path.c_str()) == 0;
+#endif
+}
+
+bool rmtree(const string &directory, bool ignore_fail) {
+  if (directory.empty())
+    return false;
+  auto files = find_files(directory, "*", false);
+  auto dirs = find_files(directory, "*", true);
+
+  bool success = true;
+  for (int i = 0; i < files.size(); ++i) {
+    if (::remove(files[i].c_str()) != 0) {
+      success = false;
+
+      if (!ignore_fail) {
+        return false;
+      }
+    }
+  }
+
+  dirs.insert(dirs.begin(), directory);
+  for (int i = (int)dirs.size() - 1; i >= 0; --i) {
+#ifdef U_OS_WINDOWS
+    if (!::RemoveDirectoryA(dirs[i].c_str())) {
+#else
+    if (::rmdir(dirs[i].c_str()) != 0) {
+#endif
+      success = false;
+      if (!ignore_fail)
+        return false;
+    }
+  }
+  return success;
+}
+
+string join_dims(const vector<int64_t> &dims) {
+  stringstream output;
+  char buf[64];
+  const char *fmts[] = {"%d", " x %d"};
+  for (int i = 0; i < dims.size(); ++i) {
+    snprintf(buf, sizeof(buf), fmts[i != 0], dims[i]);
+    output << buf;
+  }
+  return output.str();
+}
+
+}; // namespace iLogger
diff --git a/deploy/cpp_infer/src/utils/ilogger.h b/deploy/cpp_infer/src/utils/ilogger.h
new file mode 100644
index 0000000..6efd453
--- /dev/null
+++ b/deploy/cpp_infer/src/utils/ilogger.h
@@ -0,0 +1,151 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// Based on https://github.com/shouxieai/tensorRT_Pro
+
+// Copyright (c) 2022 TensorRTPro
+
+// Permission is hereby granted, free of charge, to any person obtaining a copy
+// of this software && associated documentation files (the "Software"), to deal
+// in the Software without restriction, including without limitation the rights
+// to use, copy, modify, merge, publish, distribute, sublicense, &&/||  sell
+// copies of the Software, && to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions:
+
+// The above copyright notice && this permission notice shall be included in
+// all copies ||  substantial portions of the Software.
+
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+// SOFTWARE.
+
+#pragma once
+
+#include <time.h>
+
+#include <string>
+#include <tuple>
+#include <vector>
+
+#if defined(_WIN32)
+#define U_OS_WINDOWS
+#else
+#define U_OS_LINUX
+#endif
+
+namespace iLogger {
+
+using namespace std;
+
+enum class LogLevel : int {
+  Debug = 5,
+  Verbose = 4,
+  Info = 3,
+  Warning = 2,
+  Error = 1,
+  Fatal = 0
+};
+
+#define INFOD(...)                                                             \
+  iLogger::__log_func(__FILE__, __LINE__, iLogger::LogLevel::Debug, __VA_ARGS__)
+#define INFOV(...)                                                             \
+  iLogger::__log_func(__FILE__, __LINE__, iLogger::LogLevel::Verbose,          \
+                      __VA_ARGS__)
+#define INFO(...)                                                              \
+  iLogger::__log_func(__FILE__, __LINE__, iLogger::LogLevel::Info, __VA_ARGS__)
+#define INFOW(...)                                                             \
+  iLogger::__log_func(__FILE__, __LINE__, iLogger::LogLevel::Warning,          \
+                      __VA_ARGS__)
+#define INFOE(...)                                                             \
+  iLogger::__log_func(__FILE__, __LINE__, iLogger::LogLevel::Error, __VA_ARGS__)
+#define INFOF(...)                                                             \
+  iLogger::__log_func(__FILE__, __LINE__, iLogger::LogLevel::Fatal, __VA_ARGS__)
+
+string date_now();
+string time_now();
+string gmtime_now();
+string gmtime(time_t t);
+time_t gmtime2ctime(const string &gmt);
+void sleep(int ms);
+
+bool isfile(const string &file);
+bool mkdir(const string &path);
+bool mkdirs(const string &path);
+bool delete_file(const string &path);
+bool rmtree(const string &directory, bool ignore_fail = false);
+bool exists(const string &path);
+string format(const char *fmt, ...);
+FILE *fopen_mkdirs(const string &path, const string &mode);
+string file_name(const string &path, bool include_suffix = true);
+string directory(const string &path);
+long long timestamp_now();
+double timestamp_now_float();
+time_t last_modify(const string &file);
+vector<uint8_t> load_file(const string &file);
+string load_text_file(const string &file);
+size_t file_size(const string &file);
+
+bool begin_with(const string &str, const string &with);
+bool end_with(const string &str, const string &with);
+vector<string> split_string(const string &str, const std::string &spstr);
+string replace_string(const string &str, const string &token,
+                      const string &value, int nreplace = -1,
+                      int *out_num_replace = nullptr);
+
+// h[0-1], s[0-1], v[0-1]
+// return, 0-255, 0-255, 0-255
+tuple<uint8_t, uint8_t, uint8_t> hsv2rgb(float h, float s, float v);
+tuple<uint8_t, uint8_t, uint8_t> random_color(int id);
+
+//   abcdefg.pnga          *.png      > false
+//   abcdefg.png           *.png      > true
+//   abcdefg.png          a?cdefg.png > true
+bool pattern_match(const char *str, const char *matcher,
+                   bool igrnoe_case = true);
+vector<string> find_files(const string &directory, const string &filter = "*",
+                          bool findDirectory = false,
+                          bool includeSubDirectory = false);
+
+string align_blank(const string &input, int align_size, char blank = ' ');
+bool save_file(const string &file, const vector<uint8_t> &data,
+               bool mk_dirs = true);
+bool save_file(const string &file, const string &data, bool mk_dirs = true);
+bool save_file(const string &file, const void *data, size_t length,
+               bool mk_dirs = true);
+
+// 捕获：SIGINT(2)、SIGQUIT(3)
+int while_loop();
+
+// 关于logger的api
+const char *level_string(LogLevel level);
+void set_logger_save_directory(const string &loggerDirectory);
+
+void set_log_level(LogLevel level);
+LogLevel get_log_level();
+void __log_func(const char *file, int line, LogLevel level, const char *fmt,
+                ...);
+void destroy_logger();
+
+string base64_decode(const string &base64);
+string base64_encode(const void *data, size_t size);
+
+inline int upbound(int n, int align = 32) {
+  return (n + align - 1) / align * align;
+}
+string join_dims(const vector<int64_t> &dims);
+}; // namespace iLogger
diff --git a/deploy/cpp_infer/src/utils/mkldnn_blocklist.cc b/deploy/cpp_infer/src/utils/mkldnn_blocklist.cc
new file mode 100644
index 0000000..e70ffb6
--- /dev/null
+++ b/deploy/cpp_infer/src/utils/mkldnn_blocklist.cc
@@ -0,0 +1,63 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "mkldnn_blocklist.h"
+
+namespace Mkldnn {
+
+const std::unordered_set<std::string> MKLDNN_BLOCKLIST = {
+    "LaTeX_OCR_rec",
+    "PP-FormulaNet-L",
+    "PP-FormulaNet-S",
+    "UniMERNet",
+    "UVDoc",
+    "Cascade-MaskRCNN-ResNet50-FPN",
+    "Cascade-MaskRCNN-ResNet50-vd-SSLDv2-FPN",
+    "Mask-RT-DETR-M",
+    "Mask-RT-DETR-S",
+    "MaskRCNN-ResNeXt101-vd-FPN",
+    "MaskRCNN-ResNet101-FPN",
+    "MaskRCNN-ResNet101-vd-FPN",
+    "MaskRCNN-ResNet50-FPN",
+    "MaskRCNN-ResNet50-vd-FPN",
+    "MaskRCNN-ResNet50",
+    "SOLOv2",
+    "PP-TinyPose_128x96",
+    "PP-TinyPose_256x192",
+    "Cascade-FasterRCNN-ResNet50-FPN",
+    "Cascade-FasterRCNN-ResNet50-vd-SSLDv2-FPN",
+    "Co-DINO-Swin-L",
+    "Co-Deformable-DETR-Swin-T",
+    "FasterRCNN-ResNeXt101-vd-FPN",
+    "FasterRCNN-ResNet101-FPN",
+    "FasterRCNN-ResNet101",
+    "FasterRCNN-ResNet34-FPN",
+    "FasterRCNN-ResNet50-FPN",
+    "FasterRCNN-ResNet50-vd-FPN",
+    "FasterRCNN-ResNet50-vd-SSLDv2-FPN",
+    "FasterRCNN-ResNet50",
+    "FasterRCNN-Swin-Tiny-FPN",
+    "MaskFormer_small",
+    "MaskFormer_tiny",
+    "SLANeXt_wired",
+    "SLANeXt_wireless",
+    "SLANet",
+    "SLANet_plus",
+    "YOWO",
+    "SAM-H_box",
+    "SAM-H_point",
+    "PP-FormulaNet_plus-L",
+    "PP-FormulaNet_plus-M",
+    "PP-FormulaNet_plus-S"};
+}
diff --git a/deploy/cpp_infer/src/utils/mkldnn_blocklist.h b/deploy/cpp_infer/src/utils/mkldnn_blocklist.h
new file mode 100644
index 0000000..716812e
--- /dev/null
+++ b/deploy/cpp_infer/src/utils/mkldnn_blocklist.h
@@ -0,0 +1,23 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <string>
+#include <unordered_set>
+namespace Mkldnn {
+
+extern const std::unordered_set<std::string> MKLDNN_BLOCKLIST;
+
+}
diff --git a/deploy/cpp_infer/src/utils/pp_option.cc b/deploy/cpp_infer/src/utils/pp_option.cc
new file mode 100644
index 0000000..3a7c4f5
--- /dev/null
+++ b/deploy/cpp_infer/src/utils/pp_option.cc
@@ -0,0 +1,136 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "src/utils/pp_option.h"
+
+#include <algorithm>
+#include <sstream>
+#include <stdexcept>
+
+#include "absl/status/statusor.h"
+
+const std::string &PaddlePredictorOption::RunMode() const { return run_mode_; }
+
+const std::string &PaddlePredictorOption::DeviceType() const {
+  return device_type_;
+}
+
+int PaddlePredictorOption::DeviceId() const { return device_id_; }
+
+int PaddlePredictorOption::CpuThreads() const { return cpu_threads_; }
+
+const std::vector<std::string> &PaddlePredictorOption::DeletePass() const {
+  return delete_pass_;
+}
+
+bool PaddlePredictorOption::EnableNewIR() const { return enable_new_ir_; }
+
+bool PaddlePredictorOption::EnableCinn() const { return enable_cinn_; }
+
+int PaddlePredictorOption::MkldnnCacheCapacity() const {
+  return mkldnn_cache_capacity_;
+}
+
+const std::vector<std::string> &
+PaddlePredictorOption::GetSupportRunMode() const {
+  return SUPPORT_RUN_MODE;
+}
+
+const std::vector<std::string> &
+PaddlePredictorOption::GetSupportDevice() const {
+  return SUPPORT_DEVICE;
+}
+
+absl::Status PaddlePredictorOption::SetRunMode(const std::string &run_mode) {
+  if (std::find(SUPPORT_RUN_MODE.begin(), SUPPORT_RUN_MODE.end(), run_mode) ==
+      SUPPORT_RUN_MODE.end()) {
+    return absl::InvalidArgumentError("Unsupported run_mode: " + run_mode);
+  }
+  run_mode_ = run_mode;
+  return absl::OkStatus();
+}
+
+absl::Status
+PaddlePredictorOption::SetDeviceType(const std::string &device_type) {
+  if (std::find(SUPPORT_DEVICE.begin(), SUPPORT_DEVICE.end(), device_type) ==
+      SUPPORT_DEVICE.end()) {
+    return absl::InvalidArgumentError(
+        "SetDeviceType failed! Unsupported device_type: " + device_type);
+  }
+  device_type_ = device_type;
+  if (device_type_ == "cpu") {
+    device_id_ = 0;
+  }
+  return absl::OkStatus();
+}
+
+absl::Status PaddlePredictorOption::SetDeviceId(int device_id) {
+  if (device_id < 0) {
+    return absl::InvalidArgumentError(
+        "SetDeviceId failed! device_id must be >= 0");
+  }
+  device_id_ = device_id;
+  return absl::OkStatus();
+}
+
+absl::Status PaddlePredictorOption::SetCpuThreads(int cpu_threads) {
+  if (cpu_threads < 1) {
+    throw std::invalid_argument(
+        "SetCpuThreads failed! cpu_threads must be >= 1");
+  }
+  cpu_threads_ = cpu_threads;
+  return absl::OkStatus();
+}
+
+absl::Status
+PaddlePredictorOption::SetMkldnnCacheCapacity(int mkldnn_cache_capacity) {
+  if (mkldnn_cache_capacity < 1) {
+    throw std::invalid_argument(
+        "SetMkldnnCacheCapacity failed! mkldnn_cache_capacity must be >= 1");
+  }
+  mkldnn_cache_capacity_ = mkldnn_cache_capacity;
+  return absl::OkStatus();
+}
+
+void PaddlePredictorOption::SetDeletePass(
+    const std::vector<std::string> &delete_pass) {
+  delete_pass_ = delete_pass;
+}
+
+void PaddlePredictorOption::SetEnableNewIR(bool enable_new_ir) {
+  enable_new_ir_ = enable_new_ir;
+}
+
+void PaddlePredictorOption::SetEnableCinn(bool enable_cinn) {
+  enable_cinn_ = enable_cinn;
+}
+
+std::string PaddlePredictorOption::DebugString() const {
+  std::ostringstream oss;
+  oss << "run_mode: " << run_mode_ << ", "
+      << "device_type: " << device_type_ << ", "
+      << "device_id: " << device_id_ << ", "
+      << "cpu_threads: " << cpu_threads_ << ", "
+      << "delete_pass: [";
+  for (size_t i = 0; i < delete_pass_.size(); ++i) {
+    oss << delete_pass_[i];
+    if (i != delete_pass_.size() - 1)
+      oss << ", ";
+  }
+  oss << "], "
+      << "enable_new_ir: " << (enable_new_ir_ ? "true" : "false") << ", "
+      << "enable_cinn: " << (enable_cinn_ ? "true" : "false") << ", "
+      << "mkldnn_cache_capacity: " << mkldnn_cache_capacity_;
+  return oss.str();
+}
diff --git a/deploy/cpp_infer/src/utils/pp_option.h b/deploy/cpp_infer/src/utils/pp_option.h
new file mode 100644
index 0000000..a722bf1
--- /dev/null
+++ b/deploy/cpp_infer/src/utils/pp_option.h
@@ -0,0 +1,65 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <set>
+#include <string>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+#ifdef WITH_GPU
+static constexpr const char *DEVICE = "gpu:0";
+#else
+static constexpr const char *DEVICE = "cpu";
+#endif
+class PaddlePredictorOption {
+public:
+  const std::vector<std::string> SUPPORT_RUN_MODE = {"paddle", "paddle_fp16",
+                                                     "mkldnn", "mkldnn_bf16"};
+
+  const std::vector<std::string> SUPPORT_DEVICE = {"gpu", "cpu"};
+
+  const std::string &RunMode() const;
+  const std::string &DeviceType() const;
+  int DeviceId() const;
+  int CpuThreads() const;
+  const std::vector<std::string> &DeletePass() const;
+  bool EnableNewIR() const;
+  bool EnableCinn() const;
+  int MkldnnCacheCapacity() const;
+  const std::vector<std::string> &GetSupportRunMode() const;
+  const std::vector<std::string> &GetSupportDevice() const;
+  std::string DebugString() const;
+
+  absl::Status SetRunMode(const std::string &run_mode);
+  absl::Status SetDeviceType(const std::string &device_type);
+  absl::Status SetDeviceId(int device_id);
+  absl::Status SetCpuThreads(int cpu_threads);
+  absl::Status SetMkldnnCacheCapacity(int mkldnn_cache_capacity);
+  void SetDeletePass(const std::vector<std::string> &delete_pass);
+  void SetEnableNewIR(bool enable_new_ir);
+  void SetEnableCinn(bool enable_cinn);
+
+private:
+  std::string run_mode_ = "paddle";
+  std::string device_type_ = DEVICE;
+  int device_id_ = 0;
+  int cpu_threads_ = 10;
+  std::vector<std::string> delete_pass_ = {};
+  bool enable_new_ir_ = true;
+  bool enable_cinn_ = false;
+  int mkldnn_cache_capacity_ = 10;
+};
diff --git a/deploy/cpp_infer/src/utils/utility.cc b/deploy/cpp_infer/src/utils/utility.cc
new file mode 100644
index 0000000..4439c47
--- /dev/null
+++ b/deploy/cpp_infer/src/utils/utility.cc
@@ -0,0 +1,612 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "utility.h"
+
+#include <dirent.h>
+#include <sys/stat.h>
+
+#include <regex>
+
+#include "ilogger.h"
+
+absl::Status Utility::FileExists(const std::string &path) {
+  struct stat st;
+  if (stat(path.c_str(), &st) == 0) {
+    return absl::OkStatus();
+  } else {
+    return absl::NotFoundError("File is not exist:" + path);
+  }
+}
+absl::StatusOr<std::map<std::string, std::pair<std::string, std::string>>>
+Utility::GetModelPaths(const std::string &model_dir,
+                       const std::string &model_file_prefix) {
+  std::map<std::string, std::pair<std::string, std::string>> model_paths;
+  std::string model_path;
+
+  std::string json_path =
+      model_dir + PATH_SEPARATOR + model_file_prefix + ".json";
+  std::string pdmodel_path =
+      model_dir + PATH_SEPARATOR + model_file_prefix + ".pdmodel";
+  std::string params_path =
+      model_dir + PATH_SEPARATOR + model_file_prefix + ".pdiparams";
+  if (FileExists(json_path).ok()) {
+    model_path = json_path;
+  } else if (FileExists(pdmodel_path).ok()) {
+    model_path = pdmodel_path;
+  } else {
+    return absl::NotFoundError(FileExists(json_path).ToString() + " and " +
+                               FileExists(pdmodel_path).ToString());
+  }
+
+  if (model_path.empty()) {
+    return absl::NotFoundError(
+        "No PaddlePaddle model file (.json or .pdmodel) found!");
+  }
+
+  if (FileExists(params_path).ok()) {
+    model_paths["paddle"] = std::make_pair(model_path, params_path);
+  } else {
+    return absl::NotFoundError(
+        "No PaddlePaddle params file (.pdiparams) found!");
+  }
+
+  return model_paths;
+}
+
+absl::StatusOr<std::string>
+Utility::FindModelPath(const std::string &model_dir,
+                       const std::string &model_name) {
+  char last_char = model_dir.back();
+  std::string model_path;
+  if (last_char == PATH_SEPARATOR)
+    model_path = model_dir + model_name;
+  else
+    model_path = model_dir + PATH_SEPARATOR + model_name;
+  auto status = FileExists(model_path);
+  if (!status.ok()) {
+    return status;
+  }
+  return model_path;
+}
+absl::StatusOr<std::string>
+Utility::GetDefaultConfig(std::string pipeline_name) {
+  std::string current_path = __FILE__;
+  for (int i = 0; i < 2; i++) {
+    size_t pos = current_path.find_last_of(PATH_SEPARATOR);
+    if (pos == std::string::npos) {
+      return absl::NotFoundError("Could not find pipeline config yaml :" +
+                                 pipeline_name);
+    }
+    current_path = current_path.substr(0, pos);
+  }
+  std::string config_path_yaml = current_path + PATH_SEPARATOR + "configs" +
+                                 PATH_SEPARATOR + pipeline_name + ".yaml";
+  std::string config_path_yml = current_path + PATH_SEPARATOR + "configs" +
+                                PATH_SEPARATOR + pipeline_name + ".yml";
+  if (FileExists(config_path_yaml).ok()) {
+    return config_path_yaml;
+  } else if (FileExists(config_path_yml).ok()) {
+    return config_path_yml;
+  }
+  return absl::NotFoundError("Could not find pipeline config yaml :" +
+                             pipeline_name);
+}
+absl::StatusOr<std::string>
+Utility::GetConfigPaths(const std::string &model_dir,
+                        const std::string &model_file_prefix) {
+  std::string config_path = "";
+  std::string config_path_find =
+      model_dir + PATH_SEPARATOR + model_file_prefix + ".yml";
+  if (FileExists(config_path_find).ok()) {
+    config_path = config_path_find;
+  } else {
+    return FileExists(config_path_find);
+  }
+  return config_path;
+};
+
+bool Utility::IsMkldnnAvailable() {
+#ifdef _WIN32
+  SYSTEM_INFO si;
+  GetSystemInfo(&si);
+
+  char cpuBrand[0x40] = {0};
+  int cpuInfo[4] = {0};
+  __cpuid(cpuInfo, 0x80000002);
+  memcpy(cpuBrand, cpuInfo, sizeof(cpuInfo));
+  __cpuid(cpuInfo, 0x80000003);
+  memcpy(cpuBrand + 16, cpuInfo, sizeof(cpuInfo));
+  __cpuid(cpuInfo, 0x80000004);
+  memcpy(cpuBrand + 32, cpuInfo, sizeof(cpuInfo));
+  std::string brandStr(cpuBrand);
+  if (brandStr.find("Intel") != std::string::npos) {
+    return true;
+  }
+  return false;
+#else
+  std::ifstream cpuinfo("/proc/cpuinfo");
+  std::string line;
+  while (std::getline(cpuinfo, line)) {
+    if (line.find("vendor_id") != std::string::npos) {
+      auto pos = line.find(":");
+      if (pos != std::string::npos) {
+        if (line.substr(pos + 2).find("Intel") != std::string::npos) {
+          return true;
+        }
+      }
+    }
+  }
+  return false;
+#endif
+};
+
+void Utility::PrintShape(const cv::Mat &img) {
+  for (int i = 0; i < img.dims; i++) {
+    std::cout << img.size[i] << " ";
+  }
+  std::cout << std::endl;
+}
+
+absl::Status Utility::MyCreateDirectory(const std::string &path) {
+#ifdef _WIN32
+  int ret = _mkdir(path.c_str());
+#else
+  int ret = mkdir(path.c_str(), 0755);
+#endif
+  if (ret == 0) {
+    return absl::OkStatus();
+  }
+  if (errno == EEXIST) {
+    return absl::OkStatus();
+  }
+  return absl::ErrnoToStatus(errno, "Failed to create directory: " + path);
+}
+
+absl::Status Utility::MyCreatePath(const std::string &path) {
+  std::vector<std::string> paths;
+  std::string tmp;
+  for (size_t i = 0; i < path.size(); ++i) {
+    tmp += path[i];
+    if (path[i] == PATH_SEPARATOR) {
+      paths.push_back(tmp);
+    }
+  }
+  if (!tmp.empty() && tmp.back() != PATH_SEPARATOR)
+    paths.push_back(tmp);
+
+  std::string current;
+  for (size_t i = 0; i < paths.size(); ++i) {
+    current += paths[i];
+    absl::Status status = MyCreateDirectory(current);
+    if (!status.ok()) {
+      return status;
+    }
+  }
+  return absl::OkStatus();
+}
+
+absl::Status Utility::MyCreateFile(const std::string &filepath) {
+  std::ifstream infile(filepath.c_str());
+  if (infile.good()) {
+    return absl::OkStatus();
+  }
+
+  std::ofstream outfile(filepath.c_str(), std::ios::out | std::ios::trunc);
+  if (!outfile.is_open()) {
+    return absl::InternalError("Failed to create file: " + filepath);
+  }
+
+  outfile.close();
+  return absl::OkStatus();
+}
+
+absl::StatusOr<std::vector<cv::Mat>> Utility::SplitBatch(const cv::Mat &batch) {
+  if (batch.dims < 1) {
+    return absl::InvalidArgumentError(
+        "Input batch must have at least 1 dimension.");
+  }
+  if (batch.type() != CV_32F) {
+    return absl::InvalidArgumentError(
+        "Input batch must have CV_32F element type.");
+  }
+
+  std::vector<cv::Mat> split_mats;
+  int batch_size = batch.size[0];
+  std::vector<cv::Range> myranges(batch.dims);
+  for (int i = 0; i < batch_size; ++i) {
+    myranges[0] = cv::Range(i, i + 1);
+    for (int d = 1; d < batch.dims; ++d)
+      myranges[d] = cv::Range::all();
+    cv::Mat sub_mat = batch(&myranges[0]);
+
+    split_mats.push_back(sub_mat);
+  }
+
+  return split_mats;
+}
+
+std::string Utility::GetFileExtension(const std::string &file_path) {
+  size_t pos = file_path.find_last_of('.');
+  if (pos == std::string::npos || pos == file_path.length() - 1) {
+    return "";
+  }
+  return file_path.substr(pos + 1);
+}
+
+std::string Utility::ToLower(const std::string &str) {
+  std::string result = str;
+  std::transform(result.begin(), result.end(), result.begin(), ::tolower);
+  return result;
+}
+
+bool Utility::IsDirectory(const std::string &path) {
+  struct stat path_stat;
+  if (stat(path.c_str(), &path_stat) != 0) {
+    return false;
+  }
+  return S_ISDIR(path_stat.st_mode);
+}
+
+void Utility::GetFilesRecursive(const std::string &dir_path,
+                                std::vector<std::string> &file_list) {
+  DIR *dir = opendir(dir_path.c_str());
+  if (dir == NULL) {
+    return;
+  }
+
+  struct dirent *entry;
+  while ((entry = readdir(dir)) != NULL) {
+    std::string name = entry->d_name;
+    if (name == "." || name == "..") {
+      continue;
+    }
+
+    std::string full_path = "";
+    if (dir_path.back() == PATH_SEPARATOR) {
+      full_path = dir_path + name;
+    } else {
+      full_path = dir_path + PATH_SEPARATOR + name;
+    }
+
+    if (Utility::IsDirectory(full_path)) {
+      Utility::GetFilesRecursive(full_path, file_list);
+    } else if (IsImageFile(full_path)) {
+      file_list.push_back(full_path);
+    }
+  }
+
+  closedir(dir);
+}
+
+bool Utility::IsImageFile(const std::string &file_path) {
+  std::string extension = GetFileExtension(file_path);
+  std::string lower_ext = ToLower(extension);
+  return kImgSuffixes.find(lower_ext) != kImgSuffixes.end();
+}
+
+absl::StatusOr<cv::Mat> Utility::MyLoadImage(const std::string &file_path) {
+  cv::Mat image = cv::imread(file_path, cv::IMREAD_COLOR);
+  if (image.empty()) {
+    return absl::InvalidArgumentError("Failed to load image: " + file_path);
+  }
+  return image;
+}
+
+int Utility::MakeDir(const std::string &path) {
+#ifdef _WIN32
+  return _mkdir(path.c_str());
+#else
+  return mkdir(path.c_str(), 0755); // Linux/macOS 权限 755
+#endif
+}
+
+absl::Status Utility::CreateDirectoryRecursive(const std::string &path) {
+  if (path.empty()) {
+    return absl::InvalidArgumentError("Path cannot be empty");
+  }
+
+  size_t pos = 0;
+  std::string dir = path;
+#ifdef _WIN32
+#define ACCESS _access
+#define F_OK 0
+#else
+#define ACCESS access
+#endif
+  while (pos < dir.size()) {
+    pos = dir.find_first_of(PATH_SEPARATOR, pos + 1);
+    std::string subdir = (pos == std::string::npos) ? dir : dir.substr(0, pos);
+
+    if (!subdir.empty() && ACCESS(subdir.c_str(), F_OK) != 0) {
+      if (MakeDir(subdir) != 0) {
+        return absl::InternalError("Failed to create directory: " + subdir);
+      }
+    }
+
+    if (pos == std::string::npos) {
+      break;
+    }
+  }
+  return absl::OkStatus();
+}
+
+absl::Status Utility::CreateDirectoryForFile(const std::string &filePath) {
+  size_t found = filePath.find_last_of(PATH_SEPARATOR);
+  if (found != std::string::npos) {
+    std::string dirPath = filePath.substr(0, found);
+    if (!CreateDirectoryRecursive(dirPath).ok()) {
+      return absl::InternalError("Failed to create file: " + filePath);
+      ;
+    }
+  }
+  return absl::OkStatus();
+}
+
+absl::StatusOr<std::string>
+Utility::SmartCreateDirectoryForImage(std::string save_path,
+                                      const std::string &input_path,
+                                      const std::string &suffix) {
+  size_t pos = save_path.find_last_of("/\\");
+  std::string lastPart = save_path.substr(pos + 1);
+  if (lastPart.find(".") == std::string::npos) {
+    save_path += PATH_SEPARATOR;
+  }
+  std::string full_path = save_path;
+  auto status = CreateDirectoryForFile(save_path);
+  if (!status.ok()) {
+    return status;
+  }
+  if (Utility::IsDirectory(save_path)) {
+    auto file_path = input_path;
+    size_t pos = file_path.find_last_of("/\\");
+    std::string file_name =
+        (pos == std::string::npos) ? file_path : file_path.substr(pos + 1);
+    size_t dot_pos = file_name.find_last_of('.');
+    if (dot_pos == std::string::npos) {
+      file_name = file_name + suffix;
+    } else {
+      file_name.insert(dot_pos, suffix);
+    }
+    if (save_path.back() != PATH_SEPARATOR) {
+      full_path += PATH_SEPARATOR;
+    }
+    full_path += file_name;
+  }
+  return full_path;
+}
+
+absl::StatusOr<std::string>
+Utility::SmartCreateDirectoryForJson(const std::string &save_path,
+                                     const std::string &input_path,
+                                     const std::string &suffix) {
+  auto full_path = SmartCreateDirectoryForImage(save_path, input_path, suffix);
+  if (!full_path.ok()) {
+    return full_path.status();
+  }
+  size_t pos = full_path.value().rfind('.');
+  if (pos != std::string::npos) {
+    full_path.value().replace(pos, std::string::npos, ".json");
+  }
+  return full_path.value();
+}
+
+absl::StatusOr<int> Utility::StringToInt(std::string s) {
+  std::regex pattern("(\\d+)");
+  std::smatch match;
+  if (std::regex_search(s, match, pattern)) {
+    int value = std::stoi(match[1]);
+    return value;
+  } else {
+    return absl::NotFoundError("Could not find int !");
+  }
+}
+
+bool Utility::StringToBool(const std::string &str) {
+  std::string result = str;
+  std::transform(result.begin(), result.end(), result.begin(), ::tolower);
+  assert(result == "true" || result == "false");
+  if (result == "true") {
+    return true;
+  } else {
+    return false;
+  }
+}
+
+std::string Utility::VecToString(const std::vector<int> &input) {
+  std::string result;
+  for (auto it = input.begin(); it != input.end(); ++it) {
+    if (it != input.begin())
+      result += ",";
+    result += std::to_string(*it);
+  }
+  return result;
+}
+
+absl::StatusOr<std::tuple<std::string, std::string, std::string>>
+Utility::GetOcrModelInfo(std::string lang, std::string ppocr_version) {
+  // Font constants
+  const static std::string PINGFANG_FONT = "PingFang-SC-Regular.ttf";
+  const static std::string SIMFANG_FONT = "simfang.ttf";
+  const static std::string LATIN_FONT = "latin.ttf";
+  const static std::string KOREAN_FONT = "korean.ttf";
+  const static std::string ARABIC_FONT = "arabic.ttf";
+  const static std::string CYRILLIC_FONT = "cyrillic.ttf";
+  const static std::string KANNADA_FONT = "kannada.ttf";
+  const static std::string TELUGU_FONT = "telugu.ttf";
+  const static std::string TAMIL_FONT = "tamil.ttf";
+  const static std::string DEVANAGARI_FONT = "devanagari.ttf";
+
+  // Supported PP-OCR versions
+  const static std::unordered_set<std::string> SUPPORT_PPOCR_VERSION = {
+      "PP-OCRv5", "PP-OCRv4", "PP-OCRv3"};
+
+  // Language sets
+  const static std::unordered_set<std::string> LATIN_LANGS = {
+      "af", "az", "bs", "cs",       "cy",     "da",    "de", "es", "et",
+      "fr", "ga", "hr", "hu",       "id",     "is",    "it", "ku", "la",
+      "lt", "lv", "mi", "ms",       "mt",     "nl",    "no", "oc", "pi",
+      "pl", "pt", "ro", "rs_latin", "sk",     "sl",    "sq", "sv", "sw",
+      "tl", "tr", "uz", "vi",       "french", "german"};
+
+  const static std::unordered_set<std::string> ARABIC_LANGS = {"ar", "fa", "ug",
+                                                               "ur"};
+  const static std::unordered_set<std::string> ESLAV_LANGS = {"ru", "be", "uk"};
+  const static std::unordered_set<std::string> CYRILLIC_LANGS = {
+      "ru",  "rs_cyrillic", "be",  "bg",  "uk",  "mn",  "abq", "ady",
+      "kbd", "ava",         "dar", "inh", "che", "lbe", "lez", "tab"};
+  const static std::unordered_set<std::string> DEVANAGARI_LANGS = {
+      "hi",  "mr",  "ne",  "bh",  "mai", "ang", "bho",
+      "mah", "sck", "new", "gom", "sa",  "bgc"};
+  const static std::unordered_set<std::string> SPECIFIC_LANGS = {
+      "ch", "en", "korean", "japan", "chinese_cht", "te", "ka", "ta"};
+
+  // Validate input parameters
+  if (!ppocr_version.empty() &&
+      SUPPORT_PPOCR_VERSION.count(ppocr_version) == 0) {
+    return absl::InvalidArgumentError("Unsupported ppocr_version: " +
+                                      ppocr_version);
+  }
+
+  if (lang.empty())
+    lang = "ch";
+
+  // Create combined supported languages set
+  const static std::unordered_set<std::string> supported_langs = []() {
+    std::unordered_set<std::string> s;
+    s.insert(LATIN_LANGS.begin(), LATIN_LANGS.end());
+    s.insert(ARABIC_LANGS.begin(), ARABIC_LANGS.end());
+    s.insert(ESLAV_LANGS.begin(), ESLAV_LANGS.end());
+    s.insert(CYRILLIC_LANGS.begin(), CYRILLIC_LANGS.end());
+    s.insert(DEVANAGARI_LANGS.begin(), DEVANAGARI_LANGS.end());
+    s.insert(SPECIFIC_LANGS.begin(), SPECIFIC_LANGS.end());
+    s.insert("ch");
+    return s;
+  }();
+
+  if (supported_langs.count(lang) == 0) {
+    return absl::InvalidArgumentError("Unsupported lang: " + lang);
+  }
+
+  // Determine default ppocr_version if not specified
+  if (ppocr_version.empty()) {
+    std::unordered_set<std::string> v5_langs = {"ch", "chinese_cht", "en",
+                                                "japan", "korean"};
+    v5_langs.insert(LATIN_LANGS.begin(), LATIN_LANGS.end());
+    v5_langs.insert(ESLAV_LANGS.begin(), ESLAV_LANGS.end());
+
+    if (v5_langs.count(lang)) {
+      ppocr_version = "PP-OCRv5";
+    } else {
+      std::unordered_set<std::string> v3_langs = LATIN_LANGS;
+      v3_langs.insert(ARABIC_LANGS.begin(), ARABIC_LANGS.end());
+      v3_langs.insert(CYRILLIC_LANGS.begin(), CYRILLIC_LANGS.end());
+      v3_langs.insert(DEVANAGARI_LANGS.begin(), DEVANAGARI_LANGS.end());
+      v3_langs.insert(SPECIFIC_LANGS.begin(), SPECIFIC_LANGS.end());
+
+      if (v3_langs.count(lang)) {
+        ppocr_version = "PP-OCRv3";
+      } else {
+        return absl::InvalidArgumentError(
+            "Invalid lang and ocr_version combination!");
+      }
+    }
+  }
+
+  // Initialize return values
+  std::string det_model_name;
+  std::string rec_model_name;
+  std::string font_name = SIMFANG_FONT; // Default font
+
+  // Model and font selection logic
+  if (ppocr_version == "PP-OCRv5") {
+    det_model_name = "PP-OCRv5_server_det";
+    std::string rec_lang;
+
+    if (lang == "ch" || lang == "chinese_cht" || lang == "en" ||
+        lang == "japan") {
+      rec_model_name = "PP-OCRv5_server_rec";
+      font_name = SIMFANG_FONT;
+    } else if (LATIN_LANGS.count(lang)) {
+      rec_lang = "latin";
+      font_name = LATIN_FONT;
+    } else if (ESLAV_LANGS.count(lang)) {
+      rec_lang = "eslav";
+      font_name = CYRILLIC_FONT;
+    } else if (lang == "korean") {
+      rec_lang = "korean";
+      font_name = KOREAN_FONT;
+    }
+
+    if (!rec_lang.empty()) {
+      rec_model_name = rec_lang + "_PP-OCRv5_mobile_rec";
+    }
+  } else if (ppocr_version == "PP-OCRv4") {
+    if (lang == "ch") {
+      det_model_name = "PP-OCRv4_mobile_det";
+      rec_model_name = "PP-OCRv4_mobile_rec";
+      font_name = SIMFANG_FONT;
+    } else if (lang == "en") {
+      det_model_name = "PP-OCRv4_mobile_det";
+      rec_model_name = "en_PP-OCRv4_mobile_rec";
+      font_name = SIMFANG_FONT;
+    } else {
+      return absl::InvalidArgumentError(
+          "PP-OCRv4 only support ch and en languages!");
+    }
+  } else { // PP-OCRv3
+    det_model_name = "PP-OCRv3_mobile_det";
+    std::string rec_lang;
+
+    if (LATIN_LANGS.count(lang)) {
+      rec_lang = "latin";
+      font_name = LATIN_FONT;
+    } else if (ARABIC_LANGS.count(lang)) {
+      rec_lang = "arabic";
+      font_name = ARABIC_FONT;
+    } else if (CYRILLIC_LANGS.count(lang)) {
+      rec_lang = "cyrillic";
+      font_name = CYRILLIC_FONT;
+    } else if (DEVANAGARI_LANGS.count(lang)) {
+      rec_lang = "devanagari";
+      font_name = DEVANAGARI_FONT;
+    } else if (SPECIFIC_LANGS.count(lang)) {
+      rec_lang = lang;
+      if (lang == "ka") {
+        font_name = KANNADA_FONT;
+      } else if (lang == "te") {
+        font_name = TELUGU_FONT;
+      } else if (lang == "ta") {
+        font_name = TAMIL_FONT;
+      } else if (lang == "ch") {
+        font_name = SIMFANG_FONT;
+      }
+    }
+
+    if (rec_lang == "ch") {
+      rec_model_name = "PP-OCRv3_mobile_rec";
+    } else if (!rec_lang.empty()) {
+      rec_model_name = rec_lang + "_PP-OCRv3_mobile_rec";
+    }
+  }
+
+  if (rec_model_name.empty()) {
+    return absl::InvalidArgumentError(
+        "Invalid lang and ocr_version combination!");
+  }
+
+  return std::make_tuple(det_model_name, rec_model_name, font_name);
+}
+const std::set<std::string> Utility::kImgSuffixes = {"jpg", "png", "jpeg",
+                                                     "bmp"};
diff --git a/deploy/cpp_infer/src/utils/utility.h b/deploy/cpp_infer/src/utils/utility.h
new file mode 100644
index 0000000..5644ea3
--- /dev/null
+++ b/deploy/cpp_infer/src/utils/utility.h
@@ -0,0 +1,121 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <algorithm>
+#include <fstream>
+#include <map>
+#include <opencv2/opencv.hpp>
+#include <string>
+#include <unordered_set>
+#include <vector>
+
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+#ifdef _WIN32
+#include <direct.h>
+#include <io.h>
+#define mkdir _mkdir
+static const char PATH_SEPARATOR = '\\';
+#else
+#include <sys/stat.h>
+#include <sys/types.h>
+static const char PATH_SEPARATOR = '/';
+#endif
+
+#include <errno.h>
+
+class Utility {
+public:
+  struct PaddleXConfigVariant {
+    enum class Type { NONE, STR, MAP };
+    Type type;
+    std::string str_val;
+    std::unordered_map<std::string, std::string> map_val;
+    PaddleXConfigVariant() : type(Type::NONE) {}
+    PaddleXConfigVariant(const std::string &val)
+        : type(Type::STR), str_val(val) {}
+    PaddleXConfigVariant(const char *val)
+        : type(Type::STR), str_val(val ? val : "") {}
+    PaddleXConfigVariant(
+        const std::unordered_map<std::string, std::string> &val)
+        : type(Type::MAP), map_val(val) {}
+    bool IsStr() const { return type == Type::STR; }
+    bool IsMap() const { return type == Type::MAP; }
+    const std::string &GetStr() const {
+      assert(IsStr());
+      return str_val;
+    }
+    const std::unordered_map<std::string, std::string> &GetMap() const {
+      assert(IsMap());
+      return map_val;
+    }
+  };
+  static constexpr const char *MODEL_FILE_PREFIX = "inference";
+  static const std::set<std::string> kImgSuffixes;
+
+  static absl::StatusOr<
+      std::map<std::string, std::pair<std::string, std::string>>>
+  GetModelPaths(const std::string &model_dir,
+                const std::string &model_file_prefix = MODEL_FILE_PREFIX);
+
+  static absl::StatusOr<std::string>
+  FindModelPath(const std::string &model_dir, const std::string &model_name);
+  static absl::StatusOr<std::string>
+  GetConfigPaths(const std::string &model_dir,
+                 const std::string &model_file_prefix = MODEL_FILE_PREFIX);
+
+  static absl::StatusOr<std::string>
+  GetDefaultConfig(std::string pipeline_name);
+
+  static absl::Status FileExists(const std::string &path);
+
+  // TODO windows
+  static bool IsMkldnnAvailable();
+
+  static void PrintShape(const cv::Mat &img);
+
+  static absl::Status MyCreateDirectory(const std::string &path);
+  static absl::Status MyCreatePath(const std::string &path);
+  static absl::Status MyCreateFile(const std::string &filepath);
+
+  static absl::StatusOr<std::vector<cv::Mat>> SplitBatch(const cv::Mat &batch);
+
+  static absl::StatusOr<cv::Mat> MyLoadImage(const std::string &file_path);
+  static bool IsDirectory(const std::string &path);
+  static std::string GetFileExtension(const std::string &file_path);
+  static void GetFilesRecursive(const std::string &dir_path,
+                                std::vector<std::string> &file_list);
+  static std::string ToLower(const std::string &str);
+  static bool IsImageFile(const std::string &file_path);
+  static int MakeDir(const std::string &path);
+  static absl::Status CreateDirectoryRecursive(const std::string &path);
+  static absl::Status CreateDirectoryForFile(const std::string &filePath);
+  static absl::StatusOr<std::string>
+  SmartCreateDirectoryForImage(std::string save_path,
+                               const std::string &input_path,
+                               const std::string &suffix = "_res");
+  static absl::StatusOr<std::string>
+  SmartCreateDirectoryForJson(const std::string &save_path,
+                              const std::string &input_path,
+                              const std::string &suffix = "_res");
+
+  static absl::StatusOr<int> StringToInt(std::string s);
+  static bool StringToBool(const std::string &str);
+  static std::string VecToString(const std::vector<int> &input);
+
+  static absl::StatusOr<std::tuple<std::string, std::string, std::string>>
+  GetOcrModelInfo(std::string lang, std::string ppocr_version);
+};
diff --git a/deploy/cpp_infer/src/utils/yaml_config.cc b/deploy/cpp_infer/src/utils/yaml_config.cc
new file mode 100644
index 0000000..e1132c3
--- /dev/null
+++ b/deploy/cpp_infer/src/utils/yaml_config.cc
@@ -0,0 +1,406 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "yaml_config.h"
+
+#include <iostream>
+#include <sstream>
+
+#include "absl/strings/str_cat.h"
+#include "src/utils/ilogger.h"
+
+YamlConfig::YamlConfig(const std::string &model_dir) {
+  auto status_get = GetConfigYamlPaths(model_dir);
+  if (!status_get.ok()) {
+    INFOE("Could find files with the .yaml or .yml in %s %s", model_dir.c_str(),
+          status_get.ToString().c_str());
+    exit(-1);
+  }
+  auto status = LoadYamlFile();
+  if (!status.ok()) {
+    INFOE("Failed to load config: ", status.ToString().c_str());
+    exit(-1);
+  }
+  Init();
+}
+
+absl::Status YamlConfig::GetConfigYamlPaths(const std::string &model_dir) {
+  if (Utility::GetFileExtension(model_dir) == "yaml" ||
+      Utility::GetFileExtension(model_dir) == "yml") {
+    config_yaml_path_ = model_dir;
+    return absl::OkStatus();
+  }
+  std::string config_path_yml =
+      model_dir + "/" + Utility::MODEL_FILE_PREFIX + ".yml";
+  std::string config_path_yaml =
+      model_dir + "/" + Utility::MODEL_FILE_PREFIX + ".yaml";
+  if (Utility::FileExists(config_path_yml).ok()) {
+    config_yaml_path_ = config_path_yml;
+    return absl::OkStatus();
+  } else if (Utility::FileExists(config_path_yaml).ok()) {
+    config_yaml_path_ = config_path_yaml;
+    return absl::OkStatus();
+  } else {
+    return absl::NotFoundError("file is not exist!");
+  }
+};
+
+absl::Status YamlConfig::LoadYamlFile() {
+  try {
+    YAML::Node config = YAML::LoadFile(config_yaml_path_);
+    ParseNode(config);
+    return absl::OkStatus();
+  } catch (const YAML::BadFile &e) {
+    return absl::NotFoundError(
+        absl::StrCat("Failed to open YAML file: ", config_yaml_path_));
+  } catch (const YAML::ParserException &e) {
+    return absl::InvalidArgumentError(
+        absl::StrCat("Failed to parse YAML file: ", e.what()));
+  } catch (const YAML::Exception &e) {
+    return absl::InternalError(absl::StrCat("YAML error: ", e.what()));
+  } catch (const std::exception &e) {
+    return absl::InternalError(absl::StrCat("Unexpected error: ", e.what()));
+  }
+}
+
+void YamlConfig::Init() {
+  for (const auto &info : data_) {
+    if (info.first.find("DecodeImage.channel_first") != std::string::npos) {
+      pre_process_op_info_["DecodeImage.channel_first"] = info.second;
+    } else if (info.first.find("DecodeImage.img_mode") != std::string::npos) {
+      pre_process_op_info_["DecodeImage.img_mode"] = info.second;
+    } else if (info.first.find("DetLabelEncode") != std::string::npos) {
+      pre_process_op_info_["DetLabelEncode"] = info.second;
+    } else if (info.first.find("DetResizeForTest.resize_long") !=
+               std::string::npos) {
+      pre_process_op_info_["DetResizeForTest.resize_long"] = info.second;
+    } else if (info.first.find("NormalizeImage.mean") != std::string::npos) {
+      size_t pos = info.first.find("NormalizeImage.mean");
+      size_t after = pos + std::string("NormalizeImage.mean").size();
+      if (info.first[after] != '[') {
+        pre_process_op_info_["NormalizeImage.mean"] = info.second;
+      }
+    } else if (info.first.find("NormalizeImage.order") != std::string::npos) {
+      pre_process_op_info_["NormalizeImage.order"] = info.second;
+    } else if (info.first.find("NormalizeImage.scale") != std::string::npos) {
+      pre_process_op_info_["NormalizeImage.scale"] = info.second;
+    } else if (info.first.find("NormalizeImage.std") != std::string::npos) {
+      size_t pos = info.first.find("NormalizeImage.std");
+      size_t after = pos + std::string("NormalizeImage.std").size();
+      if (info.first[after] != '[') {
+        pre_process_op_info_["NormalizeImage.std"] = info.second;
+      }
+    } else if (info.first.find("ResizeImage.size") != std::string::npos) {
+      size_t pos = info.first.find("ResizeImage.size");
+      size_t after = pos + std::string("ResizeImage.size").size();
+      if (info.first[after] != '[') {
+        pre_process_op_info_["ResizeImage.size"] = info.second;
+      }
+    } else if (info.first.find("ResizeImage.resize_short") !=
+               std::string::npos) {
+      pre_process_op_info_["ResizeImage.resize_short"] = info.second;
+    } else if (info.first.find("CropImage.size") != std::string::npos) {
+      pre_process_op_info_["CropImage.size"] = info.second;
+    } else if (info.first.find("ToCHWImage") != std::string::npos) {
+      pre_process_op_info_["ToCHWImage"] = info.second;
+    } else if (info.first.find("KeepKeys.keep_keys") != std::string::npos) {
+      pre_process_op_info_["KeepKeys.keep_keys"] = info.second;
+    } else if (info.first.find("PostProcess.name") != std::string::npos) {
+      post_process_op_info_["PostProcess.name"] = info.second;
+    } else if (info.first.find("PostProcess.thresh") != std::string::npos) {
+      post_process_op_info_["PostProcess.thresh"] = info.second;
+    } else if (info.first.find("PostProcess.box_thresh") != std::string::npos) {
+      post_process_op_info_["PostProcess.box_thresh"] = info.second;
+    } else if (info.first.find("PostProcess.max_candidates") !=
+               std::string::npos) {
+      post_process_op_info_["PostProcess.max_candidates"] = info.second;
+    } else if (info.first.find("PostProcess.unclip_ratio") !=
+               std::string::npos) {
+      post_process_op_info_["PostProcess.unclip_ratio"] = info.second;
+    } else if (info.first.find("PostProcess.Topk.topk") != std::string::npos) {
+      post_process_op_info_["PostProcess.Topk.topk"] = info.second;
+    } else if (info.first.find("PostProcess.Topk.label_list") !=
+               std::string::npos) {
+      size_t pos = info.first.find("PostProcess.Topk.label_list");
+      size_t after = pos + std::string("PostProcess.Topk.label_list").size();
+      if (info.first[after] != '[') {
+        post_process_op_info_["PostProcess.Topk.label_list"] = info.second;
+      }
+    } else if (info.first.find("PostProcess.character_dict") !=
+               std::string::npos) {
+      size_t pos = info.first.find("PostProcess.character_dict");
+      size_t after = pos + std::string("PostProcess.character_dict").size();
+      if (info.first[after] != '[') {
+        post_process_op_info_["PostProcess.character_dict"] = info.second;
+      }
+    }
+  }
+}
+
+void YamlConfig::ParseNode(const YAML::Node &node, const std::string &prefix) {
+  if (node.IsMap()) {
+    for (auto it = node.begin(); it != node.end(); ++it) {
+      std::string key = prefix.empty()
+                            ? it->first.as<std::string>()
+                            : prefix + "." + it->first.as<std::string>();
+      ParseNode(it->second, key);
+    }
+  } else if (node.IsSequence()) {
+    std::stringstream ss;
+    ss << "[";
+    for (size_t i = 0; i < node.size(); ++i) {
+      std::string index_key = prefix + "[" + std::to_string(i) + "]";
+      if (node[i].IsScalar()) {
+        data_[index_key] = node[i].as<std::string>();
+        if (i > 0)
+          ss << ", ";
+        ss << node[i].as<std::string>();
+      } else {
+        ParseNode(node[i], index_key);
+      }
+    }
+    ss << "]";
+    data_[prefix] = ss.str();
+  } else if (node.IsScalar()) {
+    data_[prefix] = node.as<std::string>();
+  } else if (node.IsNull()) {
+    data_[prefix] = "null";
+  }
+}
+
+absl::StatusOr<std::string>
+YamlConfig::GetString(const std::string &key,
+                      const std::string &default_value) const {
+  for (const auto &info : data_) {
+    if (info.first.find(key) != std::string::npos) {
+      return info.second;
+    }
+  }
+  return default_value;
+}
+
+absl::StatusOr<int> YamlConfig::GetInt(const std::string &key,
+                                       int default_value) const {
+  for (const auto &info : data_) {
+    if (info.first.find(key) != std::string::npos) {
+      for (int i = 0; i < info.second.size(); i++) {
+        if (!std::isdigit(static_cast<uchar>(info.second[i]))) {
+          return absl::InvalidArgumentError("the " + key + " is not int type");
+        }
+      }
+      return std::stoi(info.second);
+    }
+  }
+  return default_value;
+}
+
+absl::StatusOr<float> YamlConfig::GetFloat(const std::string &key,
+                                           float default_value) const {
+  for (const auto &info : data_) {
+    if (info.first.find(key) != std::string::npos) {
+      return std::stof(info.second);
+    }
+  }
+  INFOW("Key not found %s,will use default value %f.", key.c_str(),
+        default_value);
+  return default_value;
+}
+
+absl::StatusOr<double> YamlConfig::GetDouble(const std::string &key) const {
+  auto it = data_.find(key);
+  if (it == data_.end()) {
+    return absl::NotFoundError(absl::StrCat("Key not found: ", key));
+  }
+  try {
+    return std::stod(it->second);
+  } catch (const std::invalid_argument &) {
+    return absl::InvalidArgumentError(
+        absl::StrCat("Invalid double value for key '", key, "': ", it->second));
+  } catch (const std::out_of_range &) {
+    return absl::OutOfRangeError(absl::StrCat(
+        "Double value out of range for key '", key, "': ", it->second));
+  }
+}
+
+absl::StatusOr<bool> YamlConfig::GetBool(const std::string &key,
+                                         bool default_value) const {
+  for (const auto &info : data_) {
+    if (info.first.find(key) != std::string::npos) {
+      if (Utility::ToLower(info.second) == "true") {
+        return true;
+      } else if (Utility::ToLower(info.second) == "false") {
+        return false;
+      } else {
+        return absl::InvalidArgumentError("the " + key + " is not bool type");
+      }
+    }
+  }
+  return default_value;
+}
+absl::StatusOr<std::unordered_map<std::string, std::string>>
+YamlConfig::GetSubModule(const std::string &key) const {
+  std::unordered_map<std::string, std::string> submodule_result = {};
+  for (const auto &info : data_) {
+    if (info.first.find(key) != std::string::npos) {
+      submodule_result[info.first] = info.second;
+    }
+  }
+  if (submodule_result.empty()) {
+    return absl::NotFoundError("the " + key + " is not exits!");
+  }
+  return submodule_result;
+}
+absl::Status YamlConfig::HasKey(const std::string &key) const {
+  if (data_.find(key) != data_.end()) {
+    return absl::OkStatus();
+  }
+  return absl::NotFoundError(absl::StrCat("Key not found: ", key));
+}
+
+absl::Status YamlConfig::PrintAll() const {
+  for (const auto &it : data_) {
+    std::cout << it.first << ": " << it.second << std::endl;
+  }
+  return absl::OkStatus();
+}
+
+absl::Status YamlConfig::PrintWithPrefix(const std::string &prefix) const {
+  for (const auto &it : data_) {
+    if (it.first.find(prefix) == 0) {
+      std::cout << it.first << ": " << it.second << std::endl;
+    }
+  }
+  return absl::OkStatus();
+}
+
+absl::Status YamlConfig::FindPreProcessOp(const std::string &prefix) const {
+  std::unordered_map<std::string, std::string> pre_process_op_info{};
+  for (const auto &it : data_) {
+    if (it.first.find(prefix) == 0) {
+      std::cout << it.first << ": " << it.second << std::endl;
+    }
+  }
+  return absl::OkStatus();
+}
+
+VectorVariant YamlConfig::SmartParseVector(const std::string &input) {
+  auto trimBracketAndSpace = [](const std::string &str) -> std::string {
+    std::string s = str;
+    s.erase(std::remove(s.begin(), s.end(), ' '), s.end());
+    if (!s.empty() && s.front() == '[')
+      s.erase(0, 1);
+    if (!s.empty() && s.back() == ']')
+      s.pop_back();
+    return s;
+  };
+  auto splitComma = [](const std::string &s) -> std::vector<std::string> {
+    std::vector<std::string> res;
+    std::string cur;
+    bool inQuotes = false;
+    for (size_t i = 0; i < s.size(); ++i) {
+      char ch = s[i];
+      if (ch == '"' && s[i + 1] != ',') {
+        inQuotes = !inQuotes;
+      }
+      if (ch == ',' && s[i - 1] == ',' && s[i + 1] == ',') {
+        cur += ch;
+        continue;
+      }
+      if (ch == ',' && !inQuotes) {
+        res.push_back(cur);
+        cur.clear();
+      } else {
+        cur += ch;
+      }
+    }
+    if (!cur.empty())
+      res.push_back(cur);
+    return res;
+  };
+  auto isInt = [](const std::string &s) -> bool {
+    if (s.empty())
+      return false;
+    size_t i = 0;
+    if (s[0] == '-' || s[0] == '+')
+      i = 1;
+    if (i == s.size())
+      return false;
+    for (; i < s.size(); ++i) {
+      if (!isdigit(s[i]))
+        return false;
+    }
+    return true;
+  };
+  auto isFloat = [](const std::string &s) -> bool {
+    std::istringstream iss(s);
+    float f;
+    char c;
+    return (iss >> f) && !(iss >> c);
+  };
+  VectorVariant result;
+  result.type = VECTOR_UNKNOWN;
+
+  std::string s = trimBracketAndSpace(input);
+  std::vector<std::string> items = splitComma(s);
+
+  bool allString = true, allInt = true, allFloat = true;
+  for (size_t i = 0; i < items.size(); ++i) {
+    std::string tmp = items[i];
+    if (tmp.size() >= 2 && tmp.front() == '"' && tmp.back() == '"') {
+      continue;
+    }
+    allString = false;
+    if (!isInt(tmp))
+      allInt = false;
+    if (!isFloat(tmp))
+      allFloat = false;
+  }
+
+  if (allString) {
+    result.type = VECTOR_STRING;
+    for (size_t i = 0; i < items.size(); ++i) {
+      std::string tmp = items[i];
+      result.vec_string.push_back(tmp.substr(1, tmp.size() - 2));
+    }
+  } else if (allInt) {
+    result.type = VECTOR_INT; // int maybe is string
+    for (size_t i = 0; i < items.size(); ++i) {
+      result.vec_int.push_back(std::stoi(items[i]));
+    }
+    for (size_t i = 0; i < items.size(); ++i) {
+      result.vec_string.push_back(items[i]);
+    }
+  } else if (allFloat) {
+    result.type = VECTOR_FLOAT;
+    for (size_t i = 0; i < items.size(); ++i) {
+      result.vec_float.push_back(std::stof(items[i]));
+    }
+  } else {
+    result.type = VECTOR_STRING;
+    for (size_t i = 0; i < items.size(); ++i) {
+      result.vec_string.push_back(items[i]);
+    }
+  }
+  return result;
+}
+
+absl::StatusOr<std::pair<std::string, std::string>>
+YamlConfig::FindKey(const std::string &key) {
+  for (const auto &info : data_) {
+    if (info.first.find(key) != std::string::npos) {
+      return info;
+    }
+  }
+  return absl::NotFoundError("Could find key " + key);
+}
diff --git a/deploy/cpp_infer/src/utils/yaml_config.h b/deploy/cpp_infer/src/utils/yaml_config.h
new file mode 100644
index 0000000..de4f827
--- /dev/null
+++ b/deploy/cpp_infer/src/utils/yaml_config.h
@@ -0,0 +1,82 @@
+// Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <yaml-cpp/yaml.h>
+
+#include <string>
+#include <unordered_map>
+
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+#include "utility.h"
+
+enum VectorType { VECTOR_INT, VECTOR_FLOAT, VECTOR_STRING, VECTOR_UNKNOWN };
+
+struct VectorVariant {
+  VectorType type;
+  std::vector<int> vec_int;
+  std::vector<float> vec_float;
+  std::vector<std::string> vec_string;
+};
+
+class YamlConfig {
+public:
+  YamlConfig(const std::unordered_map<std::string, std::string> &data)
+      : data_(data) {}
+  YamlConfig(const std::string &model_dir);
+  YamlConfig() = default;
+  ~YamlConfig() = default;
+
+  void Init();
+  std::unordered_map<std::string, std::string> PreProcessOpInfo() {
+    return pre_process_op_info_;
+  };
+  std::unordered_map<std::string, std::string> PostProcessOpInfo() {
+    return post_process_op_info_;
+  };
+  absl::StatusOr<std::string>
+  GetString(const std::string &key,
+            const std::string &default_value = "") const;
+  absl::StatusOr<int> GetInt(const std::string &key, int default_value) const;
+  absl::StatusOr<float> GetFloat(const std::string &key,
+                                 float default_value) const;
+  absl::StatusOr<double> GetDouble(const std::string &key) const;
+  absl::StatusOr<bool> GetBool(const std::string &key,
+                               bool default_value) const;
+  absl::StatusOr<std::unordered_map<std::string, std::string>>
+  GetSubModule(const std::string &key) const;
+
+  absl::Status HasKey(const std::string &key) const;
+
+  absl::Status PrintAll() const;
+  absl::Status PrintWithPrefix(const std::string &prefix) const;
+  absl::Status FindPreProcessOp(
+      const std::string &prefix = "PreProcess.transform_ops[0]") const;
+  std::unordered_map<std::string, std::string> &Data() { return data_; };
+  std::string ConfigYamlPath() { return config_yaml_path_; };
+  absl::Status GetConfigYamlPaths(const std::string &model_dir);
+  absl::Status LoadYamlFile();
+  absl::StatusOr<std::pair<std::string, std::string>>
+  FindKey(const std::string &key);
+  static VectorVariant SmartParseVector(const std::string &input);
+
+private:
+  std::string config_yaml_path_;
+  void ParseNode(const YAML::Node &node, const std::string &prefix = "");
+  std::unordered_map<std::string, std::string> data_;
+  std::unordered_map<std::string, std::string> pre_process_op_info_;
+  std::unordered_map<std::string, std::string> post_process_op_info_;
+};
diff --git a/deploy/cpp_infer/tools/build.sh b/deploy/cpp_infer/tools/build.sh
new file mode 100755
index 0000000..52d7967
--- /dev/null
+++ b/deploy/cpp_infer/tools/build.sh
@@ -0,0 +1,22 @@
+OPENCV_DIR=your_opencv_dir
+LIB_DIR=your_paddle_lib_dir
+CUDA_LIB_DIR=your_cuda_lib_dir
+CUDNN_LIB_DIR=your_cudnn_lib_dir
+
+BUILD_DIR=build
+rm -rf ${BUILD_DIR}
+mkdir ${BUILD_DIR}
+cd ${BUILD_DIR}
+
+cmake .. \
+    -DPADDLE_LIB=${LIB_DIR} \
+    -DWITH_MKL=ON \
+    -DWITH_GPU=OFF \
+    -DWITH_STATIC_LIB=OFF \
+    -DWITH_TENSORRT=OFF \
+    -DOPENCV_DIR=${OPENCV_DIR} \
+    -DCUDNN_LIB=${CUDNN_LIB_DIR} \
+    -DCUDA_LIB=${CUDA_LIB_DIR} \
+    -DUSE_FREETYPE=OFF
+
+make -j
diff --git a/deploy/cpp_infer/tools/build_opencv.sh b/deploy/cpp_infer/tools/build_opencv.sh
new file mode 100755
index 0000000..8cc03d0
--- /dev/null
+++ b/deploy/cpp_infer/tools/build_opencv.sh
@@ -0,0 +1,28 @@
+root_path="./third_party/opencv-4.7.0"
+install_path=${root_path}/opencv4
+build_dir=${root_path}/build
+
+rm -rf ${build_dir}
+mkdir ${build_dir}
+cd ${build_dir}
+
+cmake .. \
+    -DCMAKE_INSTALL_PREFIX=${install_path} \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DBUILD_SHARED_LIBS=OFF \
+    -DWITH_IPP=OFF \
+    -DBUILD_IPP_IW=OFF \
+    -DWITH_LAPACK=OFF \
+    -DWITH_EIGEN=OFF \
+    -DCMAKE_INSTALL_LIBDIR=lib64 \
+    -DWITH_ZLIB=ON \
+    -DBUILD_ZLIB=ON \
+    -DWITH_JPEG=ON \
+    -DBUILD_JPEG=ON \
+    -DWITH_PNG=ON \
+    -DBUILD_PNG=ON \
+    -DWITH_TIFF=ON \
+    -DBUILD_TIFF=ON 
+
+make -j4
+make install
diff --git a/deploy/docker/hubserving/README.md b/deploy/docker/hubserving/README.md
new file mode 100644
index 0000000..a9cfc89
--- /dev/null
+++ b/deploy/docker/hubserving/README.md
@@ -0,0 +1,54 @@
+English | [简体中文](README_cn.md)
+
+## Introduction
+Many users hope package the PaddleOCR service into a docker image, so that it can be quickly released and used in the docker or K8s environment.
+
+This page provides some standardized code to achieve this goal. You can quickly publish the PaddleOCR project into a callable Restful API service through the following steps. (At present, the deployment based on the HubServing mode is implemented first, and author plans to increase the deployment of the PaddleServing mode in the future)
+
+## 1. Prerequisites
+
+You need to install the following basic components first：
+a. Docker
+b. Graphics driver and CUDA 10.0+（GPU）
+c. NVIDIA Container Toolkit（GPU，Docker 19.03+ can skip this）
+d. cuDNN 7.6+（GPU）
+
+## 2. Build Image
+a. Go to Dockerfile directory（PS: Need to distinguish between CPU and GPU version, the following takes CPU as an example, GPU version needs to replace the keyword）
+```
+cd deploy/docker/hubserving/cpu
+```
+c. Build image
+```
+docker build -t paddleocr:cpu .
+```
+
+## 3. Start container
+a. CPU version
+```
+sudo docker run -dp 8868:8868 --name paddle_ocr paddleocr:cpu
+```
+b. GPU version (base on NVIDIA Container Toolkit)
+```
+sudo nvidia-docker run -dp 8868:8868 --name paddle_ocr paddleocr:gpu
+```
+c. GPU version (Docker 19.03++)
+```
+sudo docker run -dp 8868:8868 --gpus all --name paddle_ocr paddleocr:gpu
+```
+d. Check service status（If you can see the following statement then it means completed：Successfully installed ocr_system && Running on http://0.0.0.0:8868/）
+```
+docker logs -f paddle_ocr
+```
+
+## 4. Test
+a. Calculate the Base64 encoding of the picture to be recognized (For test purpose, you can use a free online tool such as https://freeonlinetools24.com/base64-image/ )
+b. Post a service request（sample request in sample_request.txt）
+
+```
+curl -H "Content-Type:application/json" -X POST --data "{\"images\": [\"Input image Base64 encode(need to delete the code 'data:image/jpg;base64,'）\"]}" http://localhost:8868/predict/ocr_system
+```
+c. Get response（If the call is successful, the following result will be returned）
+```
+{"msg":"","results":[[{"confidence":0.8403433561325073,"text":"约定","text_region":[[345,377],[641,390],[634,540],[339,528]]},{"confidence":0.8131805658340454,"text":"最终相遇","text_region":[[356,532],[624,530],[624,596],[356,598]]}]],"status":"0"}
+```
diff --git a/deploy/docker/hubserving/README_cn.md b/deploy/docker/hubserving/README_cn.md
new file mode 100644
index 0000000..b695b79
--- /dev/null
+++ b/deploy/docker/hubserving/README_cn.md
@@ -0,0 +1,53 @@
+[English](README.md) | 简体中文
+
+## Docker化部署服务
+在日常项目应用中，相信大家一般都会希望能通过Docker技术，把PaddleOCR服务打包成一个镜像，以便在Docker或k8s环境里，快速发布上线使用。
+
+本文将提供一些标准化的代码来实现这样的目标。大家通过如下步骤可以把PaddleOCR项目快速发布成可调用的Restful API服务。（目前暂时先实现了基于HubServing模式的部署，后续作者计划增加PaddleServing模式的部署）
+
+## 1.实施前提准备
+
+需要先完成如下基本组件的安装：
+a. Docker环境
+b. 显卡驱动和CUDA 10.0+（GPU）
+c. NVIDIA Container Toolkit（GPU，Docker 19.03以上版本可以跳过此步）
+d. cuDNN 7.6+（GPU）
+
+## 2.制作镜像
+a.切换至Dockerfile目录（注：需要区分cpu或gpu版本，下文以cpu为例，gpu版本需要替换一下关键字即可）
+```
+cd deploy/docker/hubserving/cpu
+```
+c.生成镜像
+```
+docker build -t paddleocr:cpu .
+```
+
+## 3.启动Docker容器
+a. CPU 版本
+```
+sudo docker run -dp 8868:8868 --name paddle_ocr paddleocr:cpu
+```
+b. GPU 版本 (通过NVIDIA Container Toolkit)
+```
+sudo nvidia-docker run -dp 8868:8868 --name paddle_ocr paddleocr:gpu
+```
+c. GPU 版本 (Docker 19.03以上版本，可以直接用如下命令)
+```
+sudo docker run -dp 8868:8869 --gpus all --name paddle_ocr paddleocr:gpu
+```
+d. 检查服务运行情况（出现：Successfully installed ocr_system和Running on http://0.0.0.0:8868 等信息，表示运行成功）
+```
+docker logs -f paddle_ocr
+```
+
+## 4.测试服务
+a. 计算待识别图片的Base64编码（如果只是测试一下效果，可以通过免费的在线工具实现，如：http://tool.chinaz.com/tools/imgtobase/
+b. 发送服务请求（可参见sample_request.txt中的值）
+```
+curl -H "Content-Type:application/json" -X POST --data "{\"images\": [\"填入图片Base64编码(需要删除'data:image/jpg;base64,'）\"]}" http://localhost:8868/predict/ocr_system
+```
+c. 返回结果（如果调用成功，会返回如下结果）
+```
+{"msg":"","results":[[{"confidence":0.8403433561325073,"text":"约定","text_region":[[345,377],[641,390],[634,540],[339,528]]},{"confidence":0.8131805658340454,"text":"最终相遇","text_region":[[356,532],[624,530],[624,596],[356,598]]}]],"status":"0"}
+```
diff --git a/deploy/docker/hubserving/cpu/Dockerfile b/deploy/docker/hubserving/cpu/Dockerfile
new file mode 100644
index 0000000..030ce24
--- /dev/null
+++ b/deploy/docker/hubserving/cpu/Dockerfile
@@ -0,0 +1,30 @@
+# Version: 2.0.0
+FROM registry.baidubce.com/paddlepaddle/paddle:2.0.0
+
+# PaddleOCR base on Python3.7
+RUN pip3.7 install --upgrade pip -i https://mirror.baidu.com/pypi/simple
+
+RUN pip3.7 install paddlehub --upgrade -i https://mirror.baidu.com/pypi/simple
+
+RUN git clone https://github.com/PaddlePaddle/PaddleOCR.git /PaddleOCR
+
+WORKDIR /PaddleOCR
+
+RUN pip3.7 install -r requirements.txt -i https://mirror.baidu.com/pypi/simple
+
+RUN mkdir -p /PaddleOCR/inference/
+# Download orc detect model(light version). if you want to change normal version, you can change ch_ppocr_mobile_v2.0_det_infer to ch_ppocr_server_v2.0_det_infer, also remember change det_model_dir in deploy/hubserving/ocr_system/params.py）
+ADD {link} /PaddleOCR/inference/
+RUN tar xf /PaddleOCR/inference/{file} -C /PaddleOCR/inference/
+
+# Download direction classifier(light version). If you want to change normal version, you can change ch_ppocr_mobile_v2.0_cls_infer to ch_ppocr_mobile_v2.0_cls_infer, also remember change cls_model_dir in deploy/hubserving/ocr_system/params.py）
+ADD {link} /PaddleOCR/inference/
+RUN tar xf /PaddleOCR/inference/{file}.tar -C /PaddleOCR/inference/
+
+# Download orc recognition model(light version). If you want to change normal version, you can change ch_ppocr_mobile_v2.0_rec_infer to ch_ppocr_server_v2.0_rec_infer, also remember change rec_model_dir in deploy/hubserving/ocr_system/params.py）
+ADD {link} /PaddleOCR/inference/
+RUN tar xf /PaddleOCR/inference/{file}.tar -C /PaddleOCR/inference/
+
+EXPOSE 8868
+
+CMD ["/bin/bash","-c","hub install deploy/hubserving/ocr_system/ && hub serving start -m ocr_system"]
diff --git a/deploy/docker/hubserving/gpu/Dockerfile b/deploy/docker/hubserving/gpu/Dockerfile
new file mode 100644
index 0000000..d51e3c3
--- /dev/null
+++ b/deploy/docker/hubserving/gpu/Dockerfile
@@ -0,0 +1,34 @@
+# Version: 2.0.0
+FROM registry.baidubce.com/paddlepaddle/paddle:2.0.0-gpu-cuda10.1-cudnn7
+
+# PaddleOCR base on Python3.7
+RUN pip3.7 install --upgrade pip -i https://mirror.baidu.com/pypi/simple
+
+RUN pip3.7 install paddlehub --upgrade -i https://mirror.baidu.com/pypi/simple
+
+# If you want to use PP-OCRv4. You need to modify the `model_dir` fields in the file of `deploy/hubserving/ocr_system/params.py`
+RUN git clone https://github.com/PaddlePaddle/PaddleOCR.git /PaddleOCR
+
+WORKDIR /PaddleOCR
+
+RUN pip3.7 install -r requirements.txt -i https://mirror.baidu.com/pypi/simple
+
+RUN mkdir -p /PaddleOCR/inference/
+
+# Tip: You can download models from https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.7/doc/doc_ch/models_list.md
+
+# Download orc detect model(light version). if you want to change normal version, you can change ch_ppocr_mobile_v2.0_det_infer to ch_ppocr_server_v2.0_det_infer, also remember change det_model_dir in deploy/hubserving/ocr_system/params.py）
+ADD {link} /PaddleOCR/inference/
+RUN tar xf /PaddleOCR/inference/{file}.tar -C /PaddleOCR/inference/
+
+# Download direction classifier(light version). If you want to change normal version, you can change ch_ppocr_mobile_v2.0_cls_infer to ch_ppocr_mobile_v2.0_cls_infer, also remember change cls_model_dir in deploy/hubserving/ocr_system/params.py）
+ADD {link} /PaddleOCR/inference/
+RUN tar xf /PaddleOCR/inference/{file}.tar -C /PaddleOCR/inference/
+
+# Download orc recognition model(light version). If you want to change normal version, you can change ch_ppocr_mobile_v2.0_rec_infer to ch_ppocr_server_v2.0_rec_infer, also remember change rec_model_dir in deploy/hubserving/ocr_system/params.py）
+ADD {link} /PaddleOCR/inference/
+RUN tar xf /PaddleOCR/inference/{file}.tar -C /PaddleOCR/inference/
+
+EXPOSE 8868
+
+CMD ["/bin/bash","-c","hub install deploy/hubserving/ocr_system/ && hub serving start -m ocr_system"]
diff --git a/deploy/docker/hubserving/sample_request.txt b/deploy/docker/hubserving/sample_request.txt
new file mode 100644
index 0000000..0b9ea36
--- /dev/null
+++ b/deploy/docker/hubserving/sample_request.txt
@@ -0,0 +1 @@
+curl -H "Content-Type:application/json" -X POST --data "{\"images\": [\"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\"]}" http://localhost:8866/predict/ocr_system
diff --git a/deploy/hubserving/kie_ser/__init__.py b/deploy/hubserving/kie_ser/__init__.py
new file mode 100644
index 0000000..97043fd
--- /dev/null
+++ b/deploy/hubserving/kie_ser/__init__.py
@@ -0,0 +1,13 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/deploy/hubserving/kie_ser/config.json b/deploy/hubserving/kie_ser/config.json
new file mode 100644
index 0000000..17b00d9
--- /dev/null
+++ b/deploy/hubserving/kie_ser/config.json
@@ -0,0 +1,15 @@
+{
+    "modules_info": {
+        "kie_ser": {
+            "init_args": {
+                "version": "1.0.0",
+                "use_gpu": true
+            },
+            "predict_args": {
+            }
+        }
+    },
+    "port": 8871,
+    "use_multiprocess": false,
+    "workers": 2
+}
diff --git a/deploy/hubserving/kie_ser/module.py b/deploy/hubserving/kie_ser/module.py
new file mode 100644
index 0000000..d55d3e5
--- /dev/null
+++ b/deploy/hubserving/kie_ser/module.py
@@ -0,0 +1,152 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+sys.path.insert(0, ".")
+import copy
+
+import time
+import paddlehub
+from paddlehub.common.logger import logger
+from paddlehub.module.module import moduleinfo, runnable, serving
+import cv2
+import numpy as np
+import paddlehub as hub
+
+from tools.infer.utility import base64_to_cv2
+from ppstructure.kie.predict_kie_token_ser import SerPredictor
+from ppstructure.utility import parse_args
+
+from deploy.hubserving.kie_ser.params import read_params
+
+
+@moduleinfo(
+    name="kie_ser",
+    version="1.0.0",
+    summary="kie ser service",
+    author="paddle-dev",
+    author_email="paddle-dev@baidu.com",
+    type="cv/KIE_SER",
+)
+class KIESer(hub.Module):
+    def _initialize(self, use_gpu=False, enable_mkldnn=False):
+        """
+        initialize with the necessary elements
+        """
+        cfg = self.merge_configs()
+
+        cfg.use_gpu = use_gpu
+        if use_gpu:
+            try:
+                _places = os.environ["CUDA_VISIBLE_DEVICES"]
+                int(_places[0])
+                print("use gpu: ", use_gpu)
+                print("CUDA_VISIBLE_DEVICES: ", _places)
+                cfg.gpu_mem = 8000
+            except:
+                raise RuntimeError(
+                    "Environment Variable CUDA_VISIBLE_DEVICES is not set correctly. If you wanna use gpu, please set CUDA_VISIBLE_DEVICES via export CUDA_VISIBLE_DEVICES=cuda_device_id."
+                )
+        cfg.ir_optim = True
+        cfg.enable_mkldnn = enable_mkldnn
+
+        self.ser_predictor = SerPredictor(cfg)
+
+    def merge_configs(
+        self,
+    ):
+        # default cfg
+        backup_argv = copy.deepcopy(sys.argv)
+        sys.argv = sys.argv[:1]
+        cfg = parse_args()
+
+        update_cfg_map = vars(read_params())
+
+        for key in update_cfg_map:
+            cfg.__setattr__(key, update_cfg_map[key])
+
+        sys.argv = copy.deepcopy(backup_argv)
+        return cfg
+
+    def read_images(self, paths=[]):
+        images = []
+        for img_path in paths:
+            assert os.path.isfile(img_path), "The {} isn't a valid file.".format(
+                img_path
+            )
+            img = cv2.imread(img_path)
+            if img is None:
+                logger.info("error in loading image:{}".format(img_path))
+                continue
+            images.append(img)
+        return images
+
+    def predict(self, images=[], paths=[]):
+        """
+        Get the chinese texts in the predicted images.
+        Args:
+            images (list(numpy.ndarray)): images data, shape of each is [H, W, C]. If images not paths
+            paths (list[str]): The paths of images. If paths not images
+        Returns:
+            res (list): The result of chinese texts and save path of images.
+        """
+        if images != [] and isinstance(images, list) and paths == []:
+            predicted_data = images
+        elif images == [] and isinstance(paths, list) and paths != []:
+            predicted_data = self.read_images(paths)
+        else:
+            raise TypeError("The input data is inconsistent with expectations.")
+
+        assert (
+            predicted_data != []
+        ), "There is not any image to be predicted. Please check the input data."
+
+        all_results = []
+        for img in predicted_data:
+            if img is None:
+                logger.info("error in loading image")
+                all_results.append([])
+                continue
+            starttime = time.time()
+            ser_res, _, elapse = self.ser_predictor(img)
+            elapse = time.time() - starttime
+            logger.info("Predict time: {}".format(elapse))
+            all_results.append(ser_res)
+        return all_results
+
+    @serving
+    def serving_method(self, images, **kwargs):
+        """
+        Run as a service.
+        """
+        images_decode = [base64_to_cv2(image) for image in images]
+        results = self.predict(images_decode, **kwargs)
+        return results
+
+
+if __name__ == "__main__":
+    ocr = KIESer()
+    ocr._initialize()
+    image_path = [
+        "./doc/imgs/11.jpg",
+        "./doc/imgs/12.jpg",
+    ]
+    res = ocr.predict(paths=image_path)
+    print(res)
diff --git a/deploy/hubserving/kie_ser/params.py b/deploy/hubserving/kie_ser/params.py
new file mode 100755
index 0000000..bdd04c7
--- /dev/null
+++ b/deploy/hubserving/kie_ser/params.py
@@ -0,0 +1,38 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from deploy.hubserving.ocr_system.params import read_params as pp_ocr_read_params
+
+
+class Config(object):
+    pass
+
+
+def read_params():
+    cfg = pp_ocr_read_params()
+
+    # SER params
+    cfg.kie_algorithm = "LayoutXLM"
+    cfg.use_visual_backbone = False
+
+    cfg.ser_model_dir = "./inference/ser_vi_layoutxlm_xfund_infer"
+    cfg.ser_dict_path = "train_data/XFUND/class_list_xfun.txt"
+    cfg.vis_font_path = "./doc/fonts/simfang.ttf"
+    cfg.ocr_order_method = "tb-yx"
+
+    return cfg
diff --git a/deploy/hubserving/kie_ser_re/__init__.py b/deploy/hubserving/kie_ser_re/__init__.py
new file mode 100644
index 0000000..97043fd
--- /dev/null
+++ b/deploy/hubserving/kie_ser_re/__init__.py
@@ -0,0 +1,13 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/deploy/hubserving/kie_ser_re/config.json b/deploy/hubserving/kie_ser_re/config.json
new file mode 100644
index 0000000..abc7897
--- /dev/null
+++ b/deploy/hubserving/kie_ser_re/config.json
@@ -0,0 +1,15 @@
+{
+    "modules_info": {
+        "kie_ser_re": {
+            "init_args": {
+                "version": "1.0.0",
+                "use_gpu": true
+            },
+            "predict_args": {
+            }
+        }
+    },
+    "port": 8872,
+    "use_multiprocess": false,
+    "workers": 2
+}
diff --git a/deploy/hubserving/kie_ser_re/module.py b/deploy/hubserving/kie_ser_re/module.py
new file mode 100644
index 0000000..cbd4c3a
--- /dev/null
+++ b/deploy/hubserving/kie_ser_re/module.py
@@ -0,0 +1,154 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+sys.path.insert(0, ".")
+import copy
+
+import time
+import paddlehub
+from paddlehub.common.logger import logger
+from paddlehub.module.module import moduleinfo, runnable, serving
+import cv2
+import numpy as np
+import paddlehub as hub
+
+from tools.infer.utility import base64_to_cv2
+from ppstructure.kie.predict_kie_token_ser_re import SerRePredictor
+from ppstructure.utility import parse_args
+
+from deploy.hubserving.kie_ser_re.params import read_params
+
+
+@moduleinfo(
+    name="kie_ser_re",
+    version="1.0.0",
+    summary="kie ser re service",
+    author="paddle-dev",
+    author_email="paddle-dev@baidu.com",
+    type="cv/KIE_SER_RE",
+)
+class KIESerRE(hub.Module):
+    def _initialize(self, use_gpu=False, enable_mkldnn=False):
+        """
+        initialize with the necessary elements
+        """
+        cfg = self.merge_configs()
+
+        cfg.use_gpu = use_gpu
+        if use_gpu:
+            try:
+                _places = os.environ["CUDA_VISIBLE_DEVICES"]
+                int(_places[0])
+                print("use gpu: ", use_gpu)
+                print("CUDA_VISIBLE_DEVICES: ", _places)
+                cfg.gpu_mem = 8000
+            except:
+                raise RuntimeError(
+                    "Environment Variable CUDA_VISIBLE_DEVICES is not set correctly. If you wanna use gpu, please set CUDA_VISIBLE_DEVICES via export CUDA_VISIBLE_DEVICES=cuda_device_id."
+                )
+        cfg.ir_optim = True
+        cfg.enable_mkldnn = enable_mkldnn
+
+        self.ser_re_predictor = SerRePredictor(cfg)
+
+    def merge_configs(
+        self,
+    ):
+        # default cfg
+        backup_argv = copy.deepcopy(sys.argv)
+        sys.argv = sys.argv[:1]
+        cfg = parse_args()
+
+        update_cfg_map = vars(read_params())
+
+        for key in update_cfg_map:
+            cfg.__setattr__(key, update_cfg_map[key])
+
+        sys.argv = copy.deepcopy(backup_argv)
+        return cfg
+
+    def read_images(self, paths=[]):
+        images = []
+        for img_path in paths:
+            assert os.path.isfile(img_path), "The {} isn't a valid file.".format(
+                img_path
+            )
+            img = cv2.imread(img_path)
+            if img is None:
+                logger.info("error in loading image:{}".format(img_path))
+                continue
+            images.append(img)
+        return images
+
+    def predict(self, images=[], paths=[]):
+        """
+        Get the chinese texts in the predicted images.
+        Args:
+            images (list(numpy.ndarray)): images data, shape of each is [H, W, C]. If images not paths
+            paths (list[str]): The paths of images. If paths not images
+        Returns:
+            res (list): The result of chinese texts and save path of images.
+        """
+        if images != [] and isinstance(images, list) and paths == []:
+            predicted_data = images
+        elif images == [] and isinstance(paths, list) and paths != []:
+            predicted_data = self.read_images(paths)
+        else:
+            raise TypeError("The input data is inconsistent with expectations.")
+
+        assert (
+            predicted_data != []
+        ), "There is not any image to be predicted. Please check the input data."
+
+        all_results = []
+        for img in predicted_data:
+            if img is None:
+                logger.info("error in loading image")
+                all_results.append([])
+                continue
+            print(img.shape)
+            starttime = time.time()
+            re_res, _ = self.ser_re_predictor(img)
+            print(re_res)
+            elapse = time.time() - starttime
+            logger.info("Predict time: {}".format(elapse))
+            all_results.append(re_res)
+        return all_results
+
+    @serving
+    def serving_method(self, images, **kwargs):
+        """
+        Run as a service.
+        """
+        images_decode = [base64_to_cv2(image) for image in images]
+        results = self.predict(images_decode, **kwargs)
+        return results
+
+
+if __name__ == "__main__":
+    ocr = KIESerRE()
+    ocr._initialize()
+    image_path = [
+        "./doc/imgs/11.jpg",
+        "./doc/imgs/12.jpg",
+    ]
+    res = ocr.predict(paths=image_path)
+    print(res)
diff --git a/deploy/hubserving/kie_ser_re/params.py b/deploy/hubserving/kie_ser_re/params.py
new file mode 100755
index 0000000..8c3214a
--- /dev/null
+++ b/deploy/hubserving/kie_ser_re/params.py
@@ -0,0 +1,40 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from deploy.hubserving.ocr_system.params import read_params as pp_ocr_read_params
+
+
+class Config(object):
+    pass
+
+
+def read_params():
+    cfg = pp_ocr_read_params()
+
+    # SER params
+    cfg.kie_algorithm = "LayoutXLM"
+    cfg.use_visual_backbone = False
+
+    cfg.ser_model_dir = "./inference/ser_vi_layoutxlm_xfund_infer"
+    cfg.re_model_dir = "./inference/re_vi_layoutxlm_xfund_infer"
+
+    cfg.ser_dict_path = "train_data/XFUND/class_list_xfun.txt"
+    cfg.vis_font_path = "./doc/fonts/simfang.ttf"
+    cfg.ocr_order_method = "tb-yx"
+
+    return cfg
diff --git a/deploy/hubserving/ocr_cls/__init__.py b/deploy/hubserving/ocr_cls/__init__.py
new file mode 100644
index 0000000..97043fd
--- /dev/null
+++ b/deploy/hubserving/ocr_cls/__init__.py
@@ -0,0 +1,13 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/deploy/hubserving/ocr_cls/config.json b/deploy/hubserving/ocr_cls/config.json
new file mode 100644
index 0000000..2ced861
--- /dev/null
+++ b/deploy/hubserving/ocr_cls/config.json
@@ -0,0 +1,15 @@
+{
+    "modules_info": {
+        "ocr_cls": {
+            "init_args": {
+                "version": "1.0.0",
+                "use_gpu": true
+            },
+            "predict_args": {
+            }
+        }
+    },
+    "port": 8866,
+    "use_multiprocess": false,
+    "workers": 2
+}
diff --git a/deploy/hubserving/ocr_cls/module.py b/deploy/hubserving/ocr_cls/module.py
new file mode 100644
index 0000000..9d33927
--- /dev/null
+++ b/deploy/hubserving/ocr_cls/module.py
@@ -0,0 +1,160 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+sys.path.insert(0, ".")
+import copy
+import paddlehub
+from paddlehub.common.logger import logger
+from paddlehub.module.module import moduleinfo, runnable, serving
+import cv2
+import paddlehub as hub
+
+from tools.infer.utility import base64_to_cv2
+from tools.infer.predict_cls import TextClassifier
+from tools.infer.utility import parse_args
+from deploy.hubserving.ocr_cls.params import read_params
+
+
+@moduleinfo(
+    name="ocr_cls",
+    version="1.0.0",
+    summary="ocr angle cls service",
+    author="paddle-dev",
+    author_email="paddle-dev@baidu.com",
+    type="cv/text_angle_cls",
+)
+class OCRCls(hub.Module):
+    def _initialize(self, use_gpu=False, enable_mkldnn=False):
+        """
+        initialize with the necessary elements
+        """
+        cfg = self.merge_configs()
+
+        cfg.use_gpu = use_gpu
+        if use_gpu:
+            try:
+                _places = os.environ["CUDA_VISIBLE_DEVICES"]
+                int(_places[0])
+                print("use gpu: ", use_gpu)
+                print("CUDA_VISIBLE_DEVICES: ", _places)
+                cfg.gpu_mem = 8000
+            except:
+                raise RuntimeError(
+                    "Environment Variable CUDA_VISIBLE_DEVICES is not set correctly. If you wanna use gpu, please set CUDA_VISIBLE_DEVICES via export CUDA_VISIBLE_DEVICES=cuda_device_id."
+                )
+        cfg.ir_optim = True
+        cfg.enable_mkldnn = enable_mkldnn
+
+        self.text_classifier = TextClassifier(cfg)
+
+    def merge_configs(
+        self,
+    ):
+        # default cfg
+        backup_argv = copy.deepcopy(sys.argv)
+        sys.argv = sys.argv[:1]
+        cfg = parse_args()
+
+        update_cfg_map = vars(read_params())
+
+        for key in update_cfg_map:
+            cfg.__setattr__(key, update_cfg_map[key])
+
+        sys.argv = copy.deepcopy(backup_argv)
+        return cfg
+
+    def read_images(self, paths=[]):
+        images = []
+        for img_path in paths:
+            assert os.path.isfile(img_path), "The {} isn't a valid file.".format(
+                img_path
+            )
+            img = cv2.imread(img_path)
+            if img is None:
+                logger.info("error in loading image:{}".format(img_path))
+                continue
+            images.append(img)
+        return images
+
+    def predict(self, images=[], paths=[]):
+        """
+        Get the text angle in the predicted images.
+        Args:
+            images (list(numpy.ndarray)): images data, shape of each is [H, W, C]. If images not paths
+            paths (list[str]): The paths of images. If paths not images
+        Returns:
+            res (list): The result of text detection box and save path of images.
+        """
+
+        if images != [] and isinstance(images, list) and paths == []:
+            predicted_data = images
+        elif images == [] and isinstance(paths, list) and paths != []:
+            predicted_data = self.read_images(paths)
+        else:
+            raise TypeError("The input data is inconsistent with expectations.")
+
+        assert (
+            predicted_data != []
+        ), "There is not any image to be predicted. Please check the input data."
+
+        img_list = []
+        for img in predicted_data:
+            if img is None:
+                continue
+            img_list.append(img)
+
+        rec_res_final = []
+        try:
+            img_list, cls_res, predict_time = self.text_classifier(img_list)
+            for dno in range(len(cls_res)):
+                angle, score = cls_res[dno]
+                rec_res_final.append(
+                    {
+                        "angle": angle,
+                        "confidence": float(score),
+                    }
+                )
+        except Exception as e:
+            print(e)
+            return [[]]
+
+        return [rec_res_final]
+
+    @serving
+    def serving_method(self, images, **kwargs):
+        """
+        Run as a service.
+        """
+        images_decode = [base64_to_cv2(image) for image in images]
+        results = self.predict(images_decode, **kwargs)
+        return results
+
+
+if __name__ == "__main__":
+    ocr = OCRCls()
+    ocr._initialize()
+    image_path = [
+        "./doc/imgs_words/ch/word_1.jpg",
+        "./doc/imgs_words/ch/word_2.jpg",
+        "./doc/imgs_words/ch/word_3.jpg",
+    ]
+    res = ocr.predict(paths=image_path)
+    print(res)
diff --git a/deploy/hubserving/ocr_cls/params.py b/deploy/hubserving/ocr_cls/params.py
new file mode 100755
index 0000000..589fd33
--- /dev/null
+++ b/deploy/hubserving/ocr_cls/params.py
@@ -0,0 +1,37 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+
+class Config(object):
+    pass
+
+
+def read_params():
+    cfg = Config()
+
+    # params for text classifier
+    cfg.cls_model_dir = "./inference/ch_ppocr_mobile_v2.0_cls_infer/"
+    cfg.cls_image_shape = "3, 48, 192"
+    cfg.label_list = ["0", "180"]
+    cfg.cls_batch_num = 30
+    cfg.cls_thresh = 0.9
+
+    cfg.use_pdserving = False
+    cfg.use_tensorrt = False
+
+    return cfg
diff --git a/deploy/hubserving/ocr_det/__init__.py b/deploy/hubserving/ocr_det/__init__.py
new file mode 100644
index 0000000..97043fd
--- /dev/null
+++ b/deploy/hubserving/ocr_det/__init__.py
@@ -0,0 +1,13 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/deploy/hubserving/ocr_det/config.json b/deploy/hubserving/ocr_det/config.json
new file mode 100644
index 0000000..6080d1c
--- /dev/null
+++ b/deploy/hubserving/ocr_det/config.json
@@ -0,0 +1,15 @@
+{
+    "modules_info": {
+        "ocr_det": {
+            "init_args": {
+                "version": "1.0.0",
+                "use_gpu": true
+            },
+            "predict_args": {
+            }
+        }
+    },
+    "port": 8865,
+    "use_multiprocess": false,
+    "workers": 2    
+}
diff --git a/deploy/hubserving/ocr_det/module.py b/deploy/hubserving/ocr_det/module.py
new file mode 100644
index 0000000..5fee28f
--- /dev/null
+++ b/deploy/hubserving/ocr_det/module.py
@@ -0,0 +1,155 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+sys.path.insert(0, ".")
+
+import copy
+import paddlehub
+from paddlehub.common.logger import logger
+from paddlehub.module.module import moduleinfo, runnable, serving
+import cv2
+import numpy as np
+import paddlehub as hub
+
+from tools.infer.utility import base64_to_cv2
+from tools.infer.predict_det import TextDetector
+from tools.infer.utility import parse_args
+from deploy.hubserving.ocr_system.params import read_params
+
+
+@moduleinfo(
+    name="ocr_det",
+    version="1.0.0",
+    summary="ocr detection service",
+    author="paddle-dev",
+    author_email="paddle-dev@baidu.com",
+    type="cv/text_detection",
+)
+class OCRDet(hub.Module):
+    def _initialize(self, use_gpu=False, enable_mkldnn=False):
+        """
+        initialize with the necessary elements
+        """
+        cfg = self.merge_configs()
+
+        cfg.use_gpu = use_gpu
+        if use_gpu:
+            try:
+                _places = os.environ["CUDA_VISIBLE_DEVICES"]
+                int(_places[0])
+                print("use gpu: ", use_gpu)
+                print("CUDA_VISIBLE_DEVICES: ", _places)
+                cfg.gpu_mem = 8000
+            except:
+                raise RuntimeError(
+                    "Environment Variable CUDA_VISIBLE_DEVICES is not set correctly. If you wanna use gpu, please set CUDA_VISIBLE_DEVICES via export CUDA_VISIBLE_DEVICES=cuda_device_id."
+                )
+        cfg.ir_optim = True
+        cfg.enable_mkldnn = enable_mkldnn
+
+        self.text_detector = TextDetector(cfg)
+
+    def merge_configs(
+        self,
+    ):
+        # default cfg
+        backup_argv = copy.deepcopy(sys.argv)
+        sys.argv = sys.argv[:1]
+        cfg = parse_args()
+
+        update_cfg_map = vars(read_params())
+
+        for key in update_cfg_map:
+            cfg.__setattr__(key, update_cfg_map[key])
+
+        sys.argv = copy.deepcopy(backup_argv)
+        return cfg
+
+    def read_images(self, paths=[]):
+        images = []
+        for img_path in paths:
+            assert os.path.isfile(img_path), "The {} isn't a valid file.".format(
+                img_path
+            )
+            img = cv2.imread(img_path)
+            if img is None:
+                logger.info("error in loading image:{}".format(img_path))
+                continue
+            images.append(img)
+        return images
+
+    def predict(self, images=[], paths=[]):
+        """
+        Get the text box in the predicted images.
+        Args:
+            images (list(numpy.ndarray)): images data, shape of each is [H, W, C]. If images not paths
+            paths (list[str]): The paths of images. If paths not images
+        Returns:
+            res (list): The result of text detection box and save path of images.
+        """
+
+        if images != [] and isinstance(images, list) and paths == []:
+            predicted_data = images
+        elif images == [] and isinstance(paths, list) and paths != []:
+            predicted_data = self.read_images(paths)
+        else:
+            raise TypeError("The input data is inconsistent with expectations.")
+
+        assert (
+            predicted_data != []
+        ), "There is not any image to be predicted. Please check the input data."
+
+        all_results = []
+        for img in predicted_data:
+            if img is None:
+                logger.info("error in loading image")
+                all_results.append([])
+                continue
+            dt_boxes, elapse = self.text_detector(img)
+            logger.info("Predict time : {}".format(elapse))
+
+            rec_res_final = []
+            for dno in range(len(dt_boxes)):
+                rec_res_final.append(
+                    {"text_region": dt_boxes[dno].astype(np.int32).tolist()}
+                )
+            all_results.append(rec_res_final)
+        return all_results
+
+    @serving
+    def serving_method(self, images, **kwargs):
+        """
+        Run as a service.
+        """
+        images_decode = [base64_to_cv2(image) for image in images]
+        results = self.predict(images_decode, **kwargs)
+        return results
+
+
+if __name__ == "__main__":
+    ocr = OCRDet()
+    ocr._initialize()
+    image_path = [
+        "./doc/imgs/11.jpg",
+        "./doc/imgs/12.jpg",
+    ]
+    res = ocr.predict(paths=image_path)
+    print(res)
diff --git a/deploy/hubserving/ocr_det/params.py b/deploy/hubserving/ocr_det/params.py
new file mode 100755
index 0000000..94115d6
--- /dev/null
+++ b/deploy/hubserving/ocr_det/params.py
@@ -0,0 +1,48 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+
+class Config(object):
+    pass
+
+
+def read_params():
+    cfg = Config()
+
+    # params for text detector
+    cfg.det_algorithm = "DB"
+    cfg.det_model_dir = "./inference/PP-OCRv3_mobile_det_infer/"
+    cfg.det_limit_side_len = 960
+    cfg.det_limit_type = "max"
+
+    # DB params
+    cfg.det_db_thresh = 0.3
+    cfg.det_db_box_thresh = 0.6
+    cfg.det_db_unclip_ratio = 1.5
+    cfg.use_dilation = False
+    cfg.det_db_score_mode = "fast"
+
+    # #EAST params
+    # cfg.det_east_score_thresh = 0.8
+    # cfg.det_east_cover_thresh = 0.1
+    # cfg.det_east_nms_thresh = 0.2
+
+    cfg.use_pdserving = False
+    cfg.use_tensorrt = False
+
+    return cfg
diff --git a/deploy/hubserving/ocr_rec/__init__.py b/deploy/hubserving/ocr_rec/__init__.py
new file mode 100644
index 0000000..97043fd
--- /dev/null
+++ b/deploy/hubserving/ocr_rec/__init__.py
@@ -0,0 +1,13 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/deploy/hubserving/ocr_rec/config.json b/deploy/hubserving/ocr_rec/config.json
new file mode 100644
index 0000000..59d54e6
--- /dev/null
+++ b/deploy/hubserving/ocr_rec/config.json
@@ -0,0 +1,15 @@
+{
+    "modules_info": {
+        "ocr_rec": {
+            "init_args": {
+                "version": "1.0.0",
+                "use_gpu": true
+            },
+            "predict_args": {
+            }
+        }
+    },
+    "port": 8867,
+    "use_multiprocess": false,
+    "workers": 2
+}
diff --git a/deploy/hubserving/ocr_rec/module.py b/deploy/hubserving/ocr_rec/module.py
new file mode 100644
index 0000000..7cbcfcd
--- /dev/null
+++ b/deploy/hubserving/ocr_rec/module.py
@@ -0,0 +1,160 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+sys.path.insert(0, ".")
+import copy
+import paddlehub
+from paddlehub.common.logger import logger
+from paddlehub.module.module import moduleinfo, runnable, serving
+import cv2
+import paddlehub as hub
+
+from tools.infer.utility import base64_to_cv2
+from tools.infer.predict_rec import TextRecognizer
+from tools.infer.utility import parse_args
+from deploy.hubserving.ocr_rec.params import read_params
+
+
+@moduleinfo(
+    name="ocr_rec",
+    version="1.0.0",
+    summary="ocr recognition service",
+    author="paddle-dev",
+    author_email="paddle-dev@baidu.com",
+    type="cv/text_recognition",
+)
+class OCRRec(hub.Module):
+    def _initialize(self, use_gpu=False, enable_mkldnn=False):
+        """
+        initialize with the necessary elements
+        """
+        cfg = self.merge_configs()
+
+        cfg.use_gpu = use_gpu
+        if use_gpu:
+            try:
+                _places = os.environ["CUDA_VISIBLE_DEVICES"]
+                int(_places[0])
+                print("use gpu: ", use_gpu)
+                print("CUDA_VISIBLE_DEVICES: ", _places)
+                cfg.gpu_mem = 8000
+            except:
+                raise RuntimeError(
+                    "Environment Variable CUDA_VISIBLE_DEVICES is not set correctly. If you wanna use gpu, please set CUDA_VISIBLE_DEVICES via export CUDA_VISIBLE_DEVICES=cuda_device_id."
+                )
+        cfg.ir_optim = True
+        cfg.enable_mkldnn = enable_mkldnn
+
+        self.text_recognizer = TextRecognizer(cfg)
+
+    def merge_configs(
+        self,
+    ):
+        # default cfg
+        backup_argv = copy.deepcopy(sys.argv)
+        sys.argv = sys.argv[:1]
+        cfg = parse_args()
+
+        update_cfg_map = vars(read_params())
+
+        for key in update_cfg_map:
+            cfg.__setattr__(key, update_cfg_map[key])
+
+        sys.argv = copy.deepcopy(backup_argv)
+        return cfg
+
+    def read_images(self, paths=[]):
+        images = []
+        for img_path in paths:
+            assert os.path.isfile(img_path), "The {} isn't a valid file.".format(
+                img_path
+            )
+            img = cv2.imread(img_path)
+            if img is None:
+                logger.info("error in loading image:{}".format(img_path))
+                continue
+            images.append(img)
+        return images
+
+    def predict(self, images=[], paths=[]):
+        """
+        Get the text box in the predicted images.
+        Args:
+            images (list(numpy.ndarray)): images data, shape of each is [H, W, C]. If images not paths
+            paths (list[str]): The paths of images. If paths not images
+        Returns:
+            res (list): The result of text detection box and save path of images.
+        """
+
+        if images != [] and isinstance(images, list) and paths == []:
+            predicted_data = images
+        elif images == [] and isinstance(paths, list) and paths != []:
+            predicted_data = self.read_images(paths)
+        else:
+            raise TypeError("The input data is inconsistent with expectations.")
+
+        assert (
+            predicted_data != []
+        ), "There is not any image to be predicted. Please check the input data."
+
+        img_list = []
+        for img in predicted_data:
+            if img is None:
+                continue
+            img_list.append(img)
+
+        rec_res_final = []
+        try:
+            rec_res, predict_time = self.text_recognizer(img_list)
+            for dno in range(len(rec_res)):
+                text, score = rec_res[dno]
+                rec_res_final.append(
+                    {
+                        "text": text,
+                        "confidence": float(score),
+                    }
+                )
+        except Exception as e:
+            print(e)
+            return [[]]
+
+        return [rec_res_final]
+
+    @serving
+    def serving_method(self, images, **kwargs):
+        """
+        Run as a service.
+        """
+        images_decode = [base64_to_cv2(image) for image in images]
+        results = self.predict(images_decode, **kwargs)
+        return results
+
+
+if __name__ == "__main__":
+    ocr = OCRRec()
+    ocr._initialize()
+    image_path = [
+        "./doc/imgs_words/ch/word_1.jpg",
+        "./doc/imgs_words/ch/word_2.jpg",
+        "./doc/imgs_words/ch/word_3.jpg",
+    ]
+    res = ocr.predict(paths=image_path)
+    print(res)
diff --git a/deploy/hubserving/ocr_rec/params.py b/deploy/hubserving/ocr_rec/params.py
new file mode 100644
index 0000000..0496b47
--- /dev/null
+++ b/deploy/hubserving/ocr_rec/params.py
@@ -0,0 +1,41 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+
+class Config(object):
+    pass
+
+
+def read_params():
+    cfg = Config()
+
+    # params for text recognizer
+    cfg.rec_algorithm = "CRNN"
+    cfg.rec_model_dir = "./inference/ch_PP-OCRv3_rec_infer/"
+
+    cfg.rec_image_shape = "3, 48, 320"
+    cfg.rec_batch_num = 6
+    cfg.max_text_length = 25
+
+    cfg.rec_char_dict_path = "./ppocr/utils/ppocr_keys_v1.txt"
+    cfg.use_space_char = True
+
+    cfg.use_pdserving = False
+    cfg.use_tensorrt = False
+
+    return cfg
diff --git a/deploy/hubserving/ocr_system/__init__.py b/deploy/hubserving/ocr_system/__init__.py
new file mode 100644
index 0000000..97043fd
--- /dev/null
+++ b/deploy/hubserving/ocr_system/__init__.py
@@ -0,0 +1,13 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/deploy/hubserving/ocr_system/config.json b/deploy/hubserving/ocr_system/config.json
new file mode 100644
index 0000000..fbde31d
--- /dev/null
+++ b/deploy/hubserving/ocr_system/config.json
@@ -0,0 +1,15 @@
+{
+    "modules_info": {
+        "ocr_system": {
+            "init_args": {
+                "version": "1.0.0",
+                "use_gpu": true
+            },
+            "predict_args": {
+            }
+        }
+    },
+    "port": 8868,
+    "use_multiprocess": false,
+    "workers": 2
+}
diff --git a/deploy/hubserving/ocr_system/module.py b/deploy/hubserving/ocr_system/module.py
new file mode 100644
index 0000000..48a33b7
--- /dev/null
+++ b/deploy/hubserving/ocr_system/module.py
@@ -0,0 +1,165 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+sys.path.insert(0, ".")
+import copy
+
+import time
+import paddlehub
+from paddlehub.common.logger import logger
+from paddlehub.module.module import moduleinfo, runnable, serving
+import cv2
+import numpy as np
+import paddlehub as hub
+
+from tools.infer.utility import base64_to_cv2
+from tools.infer.predict_system import TextSystem
+from tools.infer.utility import parse_args
+from deploy.hubserving.ocr_system.params import read_params
+
+
+@moduleinfo(
+    name="ocr_system",
+    version="1.0.0",
+    summary="ocr system service",
+    author="paddle-dev",
+    author_email="paddle-dev@baidu.com",
+    type="cv/PP-OCR_system",
+)
+class OCRSystem(hub.Module):
+    def _initialize(self, use_gpu=False, enable_mkldnn=False):
+        """
+        initialize with the necessary elements
+        """
+        cfg = self.merge_configs()
+
+        cfg.use_gpu = use_gpu
+        if use_gpu:
+            try:
+                _places = os.environ["CUDA_VISIBLE_DEVICES"]
+                int(_places[0])
+                print("use gpu: ", use_gpu)
+                print("CUDA_VISIBLE_DEVICES: ", _places)
+                cfg.gpu_mem = 8000
+            except:
+                raise RuntimeError(
+                    "Environment Variable CUDA_VISIBLE_DEVICES is not set correctly. If you wanna use gpu, please set CUDA_VISIBLE_DEVICES via export CUDA_VISIBLE_DEVICES=cuda_device_id."
+                )
+        cfg.ir_optim = True
+        cfg.enable_mkldnn = enable_mkldnn
+
+        self.text_sys = TextSystem(cfg)
+
+    def merge_configs(
+        self,
+    ):
+        # default cfg
+        backup_argv = copy.deepcopy(sys.argv)
+        sys.argv = sys.argv[:1]
+        cfg = parse_args()
+
+        update_cfg_map = vars(read_params())
+
+        for key in update_cfg_map:
+            cfg.__setattr__(key, update_cfg_map[key])
+
+        sys.argv = copy.deepcopy(backup_argv)
+        return cfg
+
+    def read_images(self, paths=[]):
+        images = []
+        for img_path in paths:
+            assert os.path.isfile(img_path), "The {} isn't a valid file.".format(
+                img_path
+            )
+            img = cv2.imread(img_path)
+            if img is None:
+                logger.info("error in loading image:{}".format(img_path))
+                continue
+            images.append(img)
+        return images
+
+    def predict(self, images=[], paths=[]):
+        """
+        Get the chinese texts in the predicted images.
+        Args:
+            images (list(numpy.ndarray)): images data, shape of each is [H, W, C]. If images not paths
+            paths (list[str]): The paths of images. If paths not images
+        Returns:
+            res (list): The result of chinese texts and save path of images.
+        """
+
+        if images != [] and isinstance(images, list) and paths == []:
+            predicted_data = images
+        elif images == [] and isinstance(paths, list) and paths != []:
+            predicted_data = self.read_images(paths)
+        else:
+            raise TypeError("The input data is inconsistent with expectations.")
+
+        assert (
+            predicted_data != []
+        ), "There is not any image to be predicted. Please check the input data."
+
+        all_results = []
+        for img in predicted_data:
+            if img is None:
+                logger.info("error in loading image")
+                all_results.append([])
+                continue
+            starttime = time.time()
+            dt_boxes, rec_res, _ = self.text_sys(img)
+            elapse = time.time() - starttime
+            logger.info("Predict time: {}".format(elapse))
+
+            dt_num = len(dt_boxes)
+            rec_res_final = []
+
+            for dno in range(dt_num):
+                text, score = rec_res[dno]
+                rec_res_final.append(
+                    {
+                        "text": text,
+                        "confidence": float(score),
+                        "text_region": dt_boxes[dno].astype(np.int32).tolist(),
+                    }
+                )
+            all_results.append(rec_res_final)
+        return all_results
+
+    @serving
+    def serving_method(self, images, **kwargs):
+        """
+        Run as a service.
+        """
+        images_decode = [base64_to_cv2(image) for image in images]
+        results = self.predict(images_decode, **kwargs)
+        return results
+
+
+if __name__ == "__main__":
+    ocr = OCRSystem()
+    ocr._initialize()
+    image_path = [
+        "./doc/imgs/11.jpg",
+        "./doc/imgs/12.jpg",
+    ]
+    res = ocr.predict(paths=image_path)
+    print(res)
diff --git a/deploy/hubserving/ocr_system/params.py b/deploy/hubserving/ocr_system/params.py
new file mode 100755
index 0000000..8733e07
--- /dev/null
+++ b/deploy/hubserving/ocr_system/params.py
@@ -0,0 +1,73 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+
+class Config(object):
+    pass
+
+
+def read_params():
+    cfg = Config()
+
+    # params for text detector
+    cfg.det_algorithm = "DB"
+    cfg.det_model_dir = "./inference/PP-OCRv3_mobile_det_infer/"
+    cfg.det_limit_side_len = 960
+    cfg.det_limit_type = "max"
+
+    # DB params
+    cfg.det_db_thresh = 0.3
+    cfg.det_db_box_thresh = 0.5
+    cfg.det_db_unclip_ratio = 1.6
+    cfg.use_dilation = False
+    cfg.det_db_score_mode = "fast"
+
+    # EAST params
+    cfg.det_east_score_thresh = 0.8
+    cfg.det_east_cover_thresh = 0.1
+    cfg.det_east_nms_thresh = 0.2
+
+    # params for text recognizer
+    cfg.rec_algorithm = "CRNN"
+    cfg.rec_model_dir = "./inference/ch_PP-OCRv3_rec_infer/"
+
+    cfg.rec_image_shape = "3, 48, 320"
+    cfg.rec_batch_num = 6
+    cfg.max_text_length = 25
+
+    cfg.rec_char_dict_path = "./ppocr/utils/ppocr_keys_v1.txt"
+    cfg.use_space_char = True
+
+    # params for text classifier
+    cfg.use_angle_cls = True
+    cfg.cls_model_dir = "./inference/ch_ppocr_mobile_v2.0_cls_infer/"
+    cfg.cls_image_shape = "3, 48, 192"
+    cfg.label_list = ["0", "180"]
+    cfg.cls_batch_num = 30
+    cfg.cls_thresh = 0.9
+
+    # Model download URLs using HTTPS for secure transmission
+    cfg.det_model_url = "https://paddle-ocr-models.bj.bcebos.com/dygraph_v2.0/ch/ch_pp-ocrv2_det_infer.tar"
+    cfg.rec_model_url = "https://paddle-ocr-models.bj.bcebos.com/dygraph_v2.0/ch/ch_pp-ocrv2_rec_infer.tar"
+    cfg.cls_model_url = "https://paddle-ocr-models.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar"
+
+    cfg.use_pdserving = False
+    cfg.use_tensorrt = False
+    cfg.drop_score = 0.5
+
+    return cfg
diff --git a/deploy/hubserving/readme.md b/deploy/hubserving/readme.md
new file mode 100755
index 0000000..47e781c
--- /dev/null
+++ b/deploy/hubserving/readme.md
@@ -0,0 +1,250 @@
+[English](readme_en.md) | 简体中文
+
+- [基于PaddleHub Serving的服务部署](#基于paddlehub-serving的服务部署)
+  - [1. 近期更新](#1-近期更新)
+  - [2. 快速启动服务](#2-快速启动服务)
+    - [2.1 安装PaddleHub](#21-安装PaddleHub)
+    - [2.2 下载推理模型](#22-下载推理模型)
+    - [2.3 安装服务模块](#23-安装服务模块)
+    - [2.4 启动服务](#24-启动服务)
+      - [2.4.1. 命令行命令启动（仅支持CPU）](#241-命令行命令启动仅支持cpu)
+      - [2.4.2 配置文件启动（支持CPU、GPU）](#242-配置文件启动支持cpugpu)
+  - [3. 发送预测请求](#3-发送预测请求)
+  - [4. 返回结果格式说明](#4-返回结果格式说明)
+  - [5. 自定义修改服务模块](#5-自定义修改服务模块)
+
+
+PaddleOCR提供2种服务部署方式：
+- 基于PaddleHub Serving的部署：代码路径为`./deploy/hubserving`，按照本教程使用；
+- 基于PaddleServing的部署：代码路径为`./deploy/pdserving`，使用方法参考[文档](../../deploy/pdserving/README_CN.md)。
+
+# 基于PaddleHub Serving的服务部署
+
+hubserving服务部署目录下包括文本检测、文本方向分类，文本识别、文本检测+文本方向分类+文本识别3阶段串联，版面分析、表格识别和PP-Structure七种服务包，请根据需求选择相应的服务包进行安装和启动。目录结构如下：
+```
+deploy/hubserving/
+  └─  ocr_cls     文本方向分类模块服务包
+  └─  ocr_det     文本检测模块服务包
+  └─  ocr_rec     文本识别模块服务包
+  └─  ocr_system  文本检测+文本方向分类+文本识别串联服务包
+  └─  structure_layout  版面分析服务包
+  └─  structure_table  表格识别服务包
+  └─  structure_system  PP-Structure服务包
+  └─  kie_ser  关键信息抽取-SER服务包
+  └─  kie_ser_re  关键信息抽取-SER+RE服务包
+```
+
+每个服务包下包含3个文件。以2阶段串联服务包为例，目录如下：
+```
+deploy/hubserving/ocr_system/
+  └─  __init__.py    空文件，必选
+  └─  config.json    配置文件，可选，使用配置启动服务时作为参数传入
+  └─  module.py      主模块，必选，包含服务的完整逻辑
+  └─  params.py      参数文件，必选，包含模型路径、前后处理参数等参数
+```
+## 1. 近期更新
+
+* 2022.10.09 新增关键信息抽取服务。
+* 2022.08.23 新增版面分析服务。
+* 2022.05.05 新增PP-OCRv3检测和识别模型。
+* 2022.03.30 新增PP-Structure和表格识别两种服务。
+
+## 2. 快速启动服务
+以下步骤以检测+识别2阶段串联服务为例，如果只需要检测服务或识别服务，替换相应文件路径即可。
+### 2.1 安装PaddleHub
+paddlehub 需要 python>3.6.2
+```bash
+pip3 install paddlehub==2.1.0 --upgrade -i https://mirror.baidu.com/pypi/simple
+```
+
+### 2.2 下载推理模型
+安装服务模块前，需要准备推理模型并放到正确路径。默认使用的是PP-OCRv3模型，默认模型路径为：
+| 模型 | 路径 |
+| ------- | - |
+| 检测模型 | `./inference/PP-OCRv3_mobile_det_infer/` |
+| 识别模型 | `./inference/ch_PP-OCRv3_rec_infer/` |
+| 方向分类器 | `./inference/ch_ppocr_mobile_v2.0_cls_infer/` |
+| 版面分析模型 | `./inference/picodet_lcnet_x1_0_fgd_layout_infer/` |
+| 表格结构识别模型 | `./inference/ch_ppstructure_mobile_v2.0_SLANet_infer/` |
+| 关键信息抽取SER模型 | `./inference/ser_vi_layoutxlm_xfund_infer/` |
+| 关键信息抽取RE模型 | `./inference/re_vi_layoutxlm_xfund_infer/` |
+
+**模型路径可在`params.py`中查看和修改。**
+
+更多模型可以从PaddleOCR提供的模型库[PP-OCR](../../docs/version2.x/ppstructure/models_list.md)和[PP-Structure](../../docs/version2.x/ppstructure/models_list.en.md)下载，也可以替换成自己训练转换好的模型。
+
+### 2.3 安装服务模块
+PaddleOCR提供5种服务模块，根据需要安装所需模块。
+
+在Linux环境（Windows环境请将`/`替换为`\`）下，安装模块命令如下表：
+| 服务模块 | 命令 |
+| ------- | - |
+| 检测 | `hub install deploy/hubserving/ocr_det` |
+| 分类 | `hub install deploy/hubserving/ocr_cls` |
+| 识别 | `hub install deploy/hubserving/ocr_rec` |
+| 检测+识别串联 | `hub install deploy/hubserving/ocr_system` |
+| 表格识别 | `hub install deploy/hubserving/structure_table` |
+| PP-Structure | `hub install deploy/hubserving/structure_system` |
+| 版面分析 | `hub install deploy/hubserving/structure_layout` |
+| 关键信息抽取SER | `hub install deploy/hubserving/kie_ser` |
+| 关键信息抽取SER+RE | `hub install deploy/hubserving/kie_ser_re` |
+
+### 2.4 启动服务
+#### 2.4.1. 命令行命令启动（仅支持CPU）
+**启动命令：**
+```bash
+hub serving start --modules Module1==Version1, Module2==Version2, ... \
+                  --port 8866 \
+                  --use_multiprocess \
+                  --workers \
+```
+
+**参数：**
+|参数|用途|
+|---|---|
+|`--modules`/`-m`|PaddleHub Serving预安装模型，以多个Module==Version键值对的形式列出<br>**当不指定Version时，默认选择最新版本**|
+|`--port`/`-p`|服务端口，默认为8866|
+|`--use_multiprocess`|是否启用并发方式，默认为单进程方式，推荐多核CPU机器使用此方式<br>**Windows操作系统只支持单进程方式**|
+|`--workers`|在并发方式下指定的并发任务数，默认为`2*cpu_count-1`，其中`cpu_count`为CPU核数|
+
+如启动串联服务：
+```bash
+hub serving start -m ocr_system
+```
+
+这样就完成了一个服务化API的部署，使用默认端口号8866。
+
+#### 2.4.2 配置文件启动（支持CPU、GPU）
+**启动命令：**
+```bash
+hub serving start -c config.json
+```
+
+其中，`config.json`格式如下：
+```json
+{
+    "modules_info": {
+        "ocr_system": {
+            "init_args": {
+                "version": "1.0.0",
+                "use_gpu": true
+            },
+            "predict_args": {
+            }
+        }
+    },
+    "port": 8868,
+    "use_multiprocess": false,
+    "workers": 2
+}
+```
+
+- `init_args`中的可配参数与`module.py`中的`_initialize`函数接口一致。
+
+   **当`use_gpu`为`true`时，表示使用GPU启动服务。**
+- `predict_args`中的可配参数与`module.py`中的`predict`函数接口一致。
+
+**注意：**
+- 使用配置文件启动服务时，其他参数会被忽略。
+- 如果使用GPU预测(即，`use_gpu`置为`true`)，则需要在启动服务之前，设置CUDA_VISIBLE_DEVICES环境变量，如：
+  ```bash
+  export CUDA_VISIBLE_DEVICES=0
+  ```
+- **`use_gpu`不可与`use_multiprocess`同时为`true`**。
+
+如，使用GPU 3号卡启动串联服务：
+```bash
+export CUDA_VISIBLE_DEVICES=3
+hub serving start -c deploy/hubserving/ocr_system/config.json
+```
+
+## 3. 发送预测请求
+配置好服务端，可使用以下命令发送预测请求，获取预测结果：
+```bash
+python tools/test_hubserving.py --server_url=server_url --image_dir=image_path
+```
+
+需要给脚本传递2个参数：
+- `server_url`：服务地址，格式为`http://[ip_address]:[port]/predict/[module_name]`
+
+   例如，如果使用配置文件启动分类，检测、识别，检测+分类+识别3阶段，表格识别和PP-Structure服务
+
+   并为每个服务修改了port，那么发送请求的url将分别是：
+   ```
+   http://127.0.0.1:8865/predict/ocr_det
+   http://127.0.0.1:8866/predict/ocr_cls
+   http://127.0.0.1:8867/predict/ocr_rec
+   http://127.0.0.1:8868/predict/ocr_system
+   http://127.0.0.1:8869/predict/structure_table
+   http://127.0.0.1:8870/predict/structure_system
+   http://127.0.0.1:8870/predict/structure_layout
+   http://127.0.0.1:8871/predict/kie_ser
+   http://127.0.0.1:8872/predict/kie_ser_re
+   ```
+- `image_dir`：测试图像路径，可以是单张图片路径，也可以是图像集合目录路径
+- `visualize`：是否可视化结果，默认为False
+- `output`：可视化结果保存路径，默认为`./hubserving_result`
+
+访问示例：
+```bash
+python tools/test_hubserving.py --server_url=http://127.0.0.1:8868/predict/ocr_system --image_dir=./doc/imgs/ --visualize=false
+```
+
+## 4. 返回结果格式说明
+返回结果为列表（list），列表中的每一项为词典（dict），词典一共可能包含3种字段，信息如下：
+|字段名称|数据类型|意义|
+|---|---|---|
+|angle|str|文本角度|
+|text|str|文本内容|
+|confidence|float| 文本识别置信度或文本角度分类置信度|
+|text_region|list|文本位置坐标|
+|html|str|表格的html字符串|
+|regions|list|版面分析+表格识别+OCR的结果，每一项为一个list<br>包含表示区域坐标的`bbox`，区域类型的`type`和区域结果的`res`三个字段|
+|layout|list|版面分析的结果，每一项一个dict，包含版面区域坐标的`bbox`，区域类型的`label`|
+
+不同模块返回的字段不同，如，文本识别服务模块返回结果不含`text_region`字段，具体信息如下：
+|字段名/模块名     |ocr_det |ocr_cls |ocr_rec |ocr_system |structure_table |structure_system |structure_layout |kie_ser |kie_re |
+|---             |---     |---     |---     |---        |---             |---              |---              |---     |---    |
+|angle           |        |✔       |        |✔          |                |                 |                 |
+|text            |        |        |✔       |✔          |                |✔                |                 |✔       |✔      |
+|confidence      |        |✔       |✔       |✔          |                |✔                |                 |✔       |✔      |
+|text_region     |✔       |        |        |✔          |                |✔                |                 |✔       |✔      |
+|html            |        |        |        |           |✔               |✔                |                 |        |       |
+|regions         |        |        |        |           |✔               |✔                |                 |        |       |
+|layout          |        |        |        |           |                |                 |✔                |        |       |
+|ser_res         |        |        |        |           |                |                 |                 |✔       |       |
+|re_res          |        |        |        |           |                |                 |                 |        |✔      |
+
+**说明：** 如果需要增加、删除、修改返回字段，可在相应模块的`module.py`文件中进行修改，完整流程参考下一节自定义修改服务模块。
+
+## 5. 自定义修改服务模块
+如果需要修改服务逻辑，一般需要操作以下步骤（以修改`deploy/hubserving/ocr_system`为例）：
+
+1. 停止服务：
+   ```bash
+   hub serving stop --port/-p XXXX
+   ```
+2. 到`deploy/hubserving/ocr_system`下的`module.py`和`params.py`等文件中根据实际需求修改代码。
+
+   例如，如果需要替换部署服务所用模型，则需要到`params.py`中修改模型路径参数`det_model_dir`和`rec_model_dir`，如果需要关闭文本方向分类器，则将参数`use_angle_cls`置为`False`
+
+   当然，同时可能还需要修改其他相关参数，请根据实际情况修改调试。
+
+   **强烈建议修改后先直接运行`module.py`调试，能正确运行预测后再启动服务测试。**
+
+   **注意：** PPOCR-v3识别模型使用的图片输入shape为`3,48,320`,因此需要修改`params.py`中的`cfg.rec_image_shape = "3, 48, 320"`，如果不使用PPOCR-v3识别模型，则无需修改该参数。
+3. （可选）如果想要重命名模块需要更改`module.py`文件中的以下行：
+   - [`from deploy.hubserving.ocr_system.params import read_params`中的`ocr_system`](https://github.com/PaddlePaddle/PaddleOCR/blob/a923f35de57b5e378f8dd16e54d0a3e4f51267fd/deploy/hubserving/ocr_system/module.py#L35)
+   - [`name="ocr_system",`中的`ocr_system`](https://github.com/PaddlePaddle/PaddleOCR/blob/a923f35de57b5e378f8dd16e54d0a3e4f51267fd/deploy/hubserving/ocr_system/module.py#L39)
+4. （可选）可能需要删除`__pycache__`目录以强制刷新CPython缓存：
+   ```bash
+   find deploy/hubserving/ocr_system -name '__pycache__' -exec rm -r {} \;
+   ```
+5. 安装修改后的新服务包：
+   ```bash
+   hub install deploy/hubserving/ocr_system
+   ```
+6. 重新启动服务：
+   ```bash
+   hub serving start -m ocr_system
+   ```
diff --git a/deploy/hubserving/readme_en.md b/deploy/hubserving/readme_en.md
new file mode 100755
index 0000000..dfd46ec
--- /dev/null
+++ b/deploy/hubserving/readme_en.md
@@ -0,0 +1,249 @@
+English | [简体中文](readme.md)
+
+- [Service deployment based on PaddleHub Serving](#service-deployment-based-on-paddlehub-serving)
+  - [1. Update](#1-update)
+  - [2. Quick start service](#2-quick-start-service)
+    - [2.1 Install PaddleHub](#21-install-paddlehub)
+    - [2.2 Download inference model](#22-download-inference-model)
+    - [2.3 Install Service Module](#23-install-service-module)
+    - [2.4 Start service](#24-start-service)
+      - [2.4.1 Start with command line parameters (CPU only)](#241-start-with-command-line-parameters-cpu-only)
+      - [2.4.2 Start with configuration file（CPU and GPU）](#242-start-with-configuration-filecpugpu)
+  - [3. Send prediction requests](#3-send-prediction-requests)
+  - [4. Returned result format](#4-returned-result-format)
+  - [5. User-defined service module modification](#5-user-defined-service-module-modification)
+
+PaddleOCR provides 2 service deployment methods:
+- Based on **PaddleHub Serving**: Code path is `./deploy/hubserving`. Please follow this tutorial.
+- Based on **PaddleServing**: Code path is `./deploy/pdserving`. Please refer to the [tutorial](../../deploy/pdserving/README.md) for usage.
+
+# Service deployment based on PaddleHub Serving
+
+The hubserving service deployment directory includes seven service packages: text detection, text angle class, text recognition, text detection+text angle class+text recognition three-stage series connection, layout analysis, table recognition, and PP-Structure. Please select the corresponding service package to install and start the service according to your needs. The directory is as follows:
+```
+deploy/hubserving/
+  └─  ocr_det     text detection module service package
+  └─  ocr_cls     text angle class module service package
+  └─  ocr_rec     text recognition module service package
+  └─  ocr_system  text detection+text angle class+text recognition three-stage series connection service package
+  └─  structure_layout  layout analysis service package
+  └─  structure_table  table recognition service package
+  └─  structure_system  PP-Structure service package
+  └─  kie_ser  KIE(SER) service package
+  └─  kie_ser_re  KIE(SER+RE) service package
+```
+
+Each service pack contains 3 files. Take the 2-stage series connection service package as an example, the directory is as follows:
+```
+deploy/hubserving/ocr_system/
+  └─  __init__.py    Empty file, required
+  └─  config.json    Configuration file, optional, passed in as a parameter when using configuration to start the service
+  └─  module.py      Main module file, required, contains the complete logic of the service
+  └─  params.py      Parameter file, required, including parameters such as model path, pre and post-processing parameters
+```
+## 1. Update
+
+* 2022.10.09 add KIE services.
+* 2022.08.23 add layout analysis services.
+* 2022.03.30 add PP-Structure and table recognition services.
+* 2022.05.05 add PP-OCRv3 text detection and recognition services.
+
+## 2. Quick start service
+The following steps take the 2-stage series service as an example. If only the detection service or recognition service is needed, replace the corresponding file path.
+
+### 2.1 Install PaddleHub
+```bash
+pip3 install paddlehub==2.1.0 --upgrade
+```
+
+### 2.2 Download inference model
+Before installing the service module, you need to prepare the inference model and put it in the correct path. By default, the PP-OCRv3 models are used, and the default model path is:
+| Model | Path |
+| ------- | - |
+| text detection model | ./inference/PP-OCRv3_mobile_det_infer/ |
+| text recognition model | ./inference/ch_PP-OCRv3_rec_infer/ |
+| text angle classifier | ./inference/ch_ppocr_mobile_v2.0_cls_infer/ |
+| layout parse model | ./inference/picodet_lcnet_x1_0_fgd_layout_infer/ |
+| tanle recognition | ./inference/ch_ppstructure_mobile_v2.0_SLANet_infer/ |
+| KIE(SER) | ./inference/ser_vi_layoutxlm_xfund_infer/ |
+| KIE(SER+RE) | ./inference/re_vi_layoutxlm_xfund_infer/ |
+
+**The model path can be found and modified in `params.py`.**
+More models provided by PaddleOCR can be obtained from the [model library](../../docs/version2.x/ppocr/model_list.en.md). You can also use models trained by yourself.
+
+### 2.3 Install Service Module
+PaddleOCR provides 5 kinds of service modules, install the required modules according to your needs.
+
+* On the Linux platform(replace `/` with `\` if using Windows), the examples are as the following table:
+| Service model | Command |
+| text detection | `hub install deploy/hubserving/ocr_det` |
+| text angle class: | `hub install deploy/hubserving/ocr_cls` |
+| text recognition: | `hub install deploy/hubserving/ocr_rec` |
+| 2-stage series: | `hub install deploy/hubserving/ocr_system` |
+| table recognition | `hub install deploy/hubserving/structure_table` |
+| PP-Structure | `hub install deploy/hubserving/structure_system` |
+| KIE(SER) | `hub install deploy/hubserving/kie_ser` |
+| KIE(SER+RE) | `hub install deploy/hubserving/kie_ser_re` |
+
+### 2.4 Start service
+#### 2.4.1 Start with command line parameters (CPU only)
+**start command:**
+```bash
+hub serving start --modules Module1==Version1, Module2==Version2, ... \
+                  --port 8866 \
+                  --use_multiprocess \
+                  --workers \
+```
+
+**Parameters:**
+|parameters|usage|
+|---|---|
+|`--modules`/`-m`|PaddleHub Serving pre-installed model, listed in the form of multiple Module==Version key-value pairs<br>**When Version is not specified, the latest version is selected by default**|
+|`--port`/`-p`|Service port, default is 8866|
+|`--use_multiprocess`|Enable concurrent mode, by default using the single-process mode, this mode is recommended for multi-core CPU machines<br>**Windows operating system only supports single-process mode**|
+|`--workers`|The number of concurrent tasks specified in concurrent mode, the default is `2*cpu_count-1`, where `cpu_count` is the number of CPU cores|
+
+For example, start the 2-stage series service:
+```bash
+hub serving start -m ocr_system
+```
+
+This completes the deployment of a service API, using the default port number 8866.
+
+#### 2.4.2 Start with configuration file（CPU and GPU）
+**start command:**
+```bash
+hub serving start --config/-c config.json
+```
+
+In which the format of `config.json` is as follows:
+```json
+{
+    "modules_info": {
+        "ocr_system": {
+            "init_args": {
+                "version": "1.0.0",
+                "use_gpu": true
+            },
+            "predict_args": {
+            }
+        }
+    },
+    "port": 8868,
+    "use_multiprocess": false,
+    "workers": 2
+}
+```
+- The configurable parameters in `init_args` are consistent with the `_initialize` function interface in `module.py`.
+
+  **When `use_gpu` is `true`, it means that the GPU is used to start the service**.
+- The configurable parameters in `predict_args` are consistent with the `predict` function interface in `module.py`.
+
+  **Note:**
+  - When using the configuration file to start the service, other parameters will be ignored.
+  - If you use GPU prediction (that is, `use_gpu` is set to `true`), you need to set the environment variable CUDA_VISIBLE_DEVICES before starting the service, such as:
+    ```bash
+    export CUDA_VISIBLE_DEVICES=0
+    ```
+  - **`use_gpu` and `use_multiprocess` cannot be `true` at the same time.**
+
+For example, use GPU card No. 3 to start the 2-stage series service:
+```bash
+export CUDA_VISIBLE_DEVICES=3
+hub serving start -c deploy/hubserving/ocr_system/config.json
+```
+
+## 3. Send prediction requests
+After the service starts, you can use the following command to send a prediction request to obtain the prediction result:
+```bash
+python tools/test_hubserving.py --server_url=server_url --image_dir=image_path
+```
+
+Two parameters need to be passed to the script:
+- **server_url**:service address, the format of which is
+`http://[ip_address]:[port]/predict/[module_name]`
+
+  For example, if using the configuration file to start the text angle classification, text detection, text recognition, detection+classification+recognition 3 stages, table recognition and PP-Structure service,
+
+  also modified the port for each service, then the `server_url` to send the request will be:
+
+  ```
+  http://127.0.0.1:8865/predict/ocr_det
+  http://127.0.0.1:8866/predict/ocr_cls
+  http://127.0.0.1:8867/predict/ocr_rec
+  http://127.0.0.1:8868/predict/ocr_system
+  http://127.0.0.1:8869/predict/structure_table
+  http://127.0.0.1:8870/predict/structure_system
+  http://127.0.0.1:8870/predict/structure_layout
+  http://127.0.0.1:8871/predict/kie_ser
+  http://127.0.0.1:8872/predict/kie_ser_re
+  ```
+- **image_dir**:Test image path, which can be a single image path or an image directory path
+- **visualize**:Whether to visualize the results, the default value is False
+- **output**:The folder to save the Visualization result, the default value is `./hubserving_result`
+
+Example:
+```bash
+python tools/test_hubserving.py --server_url=http://127.0.0.1:8868/predict/ocr_system --image_dir=./doc/imgs/ --visualize=false`
+```
+
+## 4. Returned result format
+The returned result is a list. Each item in the list is a dictionary which may contain three fields. The information is as follows:
+
+|field name|data type|description|
+|----|----|----|
+|angle|str|angle|
+|text|str|text content|
+|confidence|float|text recognition confidence|
+|text_region|list|text location coordinates|
+|html|str|table HTML string|
+|regions|list|The result of layout analysis + table recognition + OCR, each item is a list<br>including `bbox` indicating area coordinates, `type` of area type and `res` of area results|
+|layout|list|The result of layout analysis, each item is a dict, including `bbox` indicating area coordinates, `label` of area type|
+
+The fields returned by different modules are different. For example, the results returned by the text recognition service module do not contain `text_region`, detailed table is as follows:
+
+|field name/module name |ocr_det |ocr_cls |ocr_rec |ocr_system |structure_table |structure_system |structure_layout |kie_ser |kie_re |
+|---                    |---     |---     |---     |---        |---             |---              |---              |---     |---    |
+|angle                  |        |✔       |        |✔          |                |                 |                 |
+|text                   |        |        |✔       |✔          |                |✔                |                 |✔       |✔      |
+|confidence             |        |✔       |✔       |✔          |                |✔                |                 |✔       |✔      |
+|text_region            |✔       |        |        |✔          |                |✔                |                 |✔       |✔      |
+|html                   |        |        |        |           |✔               |✔                |                 |        |       |
+|regions                |        |        |        |           |✔               |✔                |                 |        |       |
+|layout                 |        |        |        |           |                |                 |✔                |        |       |
+|ser_res                |        |        |        |           |                |                 |                 |✔       |       |
+|re_res                 |        |        |        |           |                |                 |                 |        |✔      |
+
+**Note:** If you need to add, delete or modify the returned fields, you can modify the file `module.py` of the corresponding module. For the complete process, refer to the user-defined modification service module in the next section.
+
+## 5. User-defined service module modification
+If you need to modify the service logic, the following steps are generally required (take the modification of `deploy/hubserving/ocr_system` for example):
+
+1. Stop service:
+```bash
+hub serving stop --port/-p XXXX
+```
+2. Modify the code in the corresponding files under `deploy/hubserving/ocr_system`, such as `module.py` and `params.py`, to your actual needs.
+
+   For example, if you need to replace the model used by the deployed service, you need to modify model path parameters `det_model_dir` and `rec_model_dir` in `params.py`. If you want to turn off the text direction classifier, set the parameter `use_angle_cls` to `False`.
+
+   Of course, other related parameters may need to be modified at the same time. Please modify and debug according to the actual situation.
+
+   **It is suggested to run `module.py` directly for debugging after modification before starting the service test.**
+
+   **Note** The image input shape used by the PPOCR-v3 recognition model is `3, 48, 320`, so you need to modify `cfg.rec_image_shape = "3, 48, 320"` in `params.py`, if you do not use the PPOCR-v3 recognition model, then there is no need to modify this parameter.
+3. (Optional) If you want to rename the module, the following lines should be modified:
+   - [`ocr_system` within `from deploy.hubserving.ocr_system.params import read_params`](https://github.com/PaddlePaddle/PaddleOCR/blob/a923f35de57b5e378f8dd16e54d0a3e4f51267fd/deploy/hubserving/ocr_system/module.py#L35)
+   - [`ocr_system` within `name="ocr_system",`](https://github.com/PaddlePaddle/PaddleOCR/blob/a923f35de57b5e378f8dd16e54d0a3e4f51267fd/deploy/hubserving/ocr_system/module.py#L39)
+4. (Optional) It may require you to delete the directory `__pycache__` to force flush build cache of CPython:
+   ```bash
+   find deploy/hubserving/ocr_system -name '__pycache__' -exec rm -r {} \;
+   ```
+5. Install modified service module:
+   ```bash
+   hub install deploy/hubserving/ocr_system/
+   ```
+6. Restart service:
+   ```bash
+   hub serving start -m ocr_system
+   ```
diff --git a/deploy/hubserving/structure_layout/__init__.py b/deploy/hubserving/structure_layout/__init__.py
new file mode 100644
index 0000000..97043fd
--- /dev/null
+++ b/deploy/hubserving/structure_layout/__init__.py
@@ -0,0 +1,13 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/deploy/hubserving/structure_layout/config.json b/deploy/hubserving/structure_layout/config.json
new file mode 100644
index 0000000..43a5983
--- /dev/null
+++ b/deploy/hubserving/structure_layout/config.json
@@ -0,0 +1,15 @@
+{
+    "modules_info": {
+        "structure_layout": {
+            "init_args": {
+                "version": "1.0.0",
+                "use_gpu": true
+            },
+            "predict_args": {
+            }
+        }
+    },
+    "port": 8871,
+    "use_multiprocess": false,
+    "workers": 2
+}
diff --git a/deploy/hubserving/structure_layout/module.py b/deploy/hubserving/structure_layout/module.py
new file mode 100644
index 0000000..58cdb52
--- /dev/null
+++ b/deploy/hubserving/structure_layout/module.py
@@ -0,0 +1,148 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+sys.path.insert(0, ".")
+import copy
+
+import time
+import paddlehub
+from paddlehub.common.logger import logger
+from paddlehub.module.module import moduleinfo, runnable, serving
+import cv2
+import paddlehub as hub
+
+from tools.infer.utility import base64_to_cv2
+from ppstructure.layout.predict_layout import LayoutPredictor as _LayoutPredictor
+from ppstructure.utility import parse_args
+from deploy.hubserving.structure_layout.params import read_params
+
+
+@moduleinfo(
+    name="structure_layout",
+    version="1.0.0",
+    summary="PP-Structure layout service",
+    author="paddle-dev",
+    author_email="paddle-dev@baidu.com",
+    type="cv/structure_layout",
+)
+class LayoutPredictor(hub.Module):
+    def _initialize(self, use_gpu=False, enable_mkldnn=False):
+        """
+        initialize with the necessary elements
+        """
+        cfg = self.merge_configs()
+        cfg.use_gpu = use_gpu
+        if use_gpu:
+            try:
+                _places = os.environ["CUDA_VISIBLE_DEVICES"]
+                int(_places[0])
+                print("use gpu: ", use_gpu)
+                print("CUDA_VISIBLE_DEVICES: ", _places)
+                cfg.gpu_mem = 8000
+            except:
+                raise RuntimeError(
+                    "Environment Variable CUDA_VISIBLE_DEVICES is not set correctly. If you wanna use gpu, please set CUDA_VISIBLE_DEVICES via export CUDA_VISIBLE_DEVICES=cuda_device_id."
+                )
+        cfg.ir_optim = True
+        cfg.enable_mkldnn = enable_mkldnn
+
+        self.layout_predictor = _LayoutPredictor(cfg)
+
+    def merge_configs(self):
+        # default cfg
+        backup_argv = copy.deepcopy(sys.argv)
+        sys.argv = sys.argv[:1]
+        cfg = parse_args()
+
+        update_cfg_map = vars(read_params())
+
+        for key in update_cfg_map:
+            cfg.__setattr__(key, update_cfg_map[key])
+
+        sys.argv = copy.deepcopy(backup_argv)
+        return cfg
+
+    def read_images(self, paths=[]):
+        images = []
+        for img_path in paths:
+            assert os.path.isfile(img_path), "The {} isn't a valid file.".format(
+                img_path
+            )
+            img = cv2.imread(img_path)
+            if img is None:
+                logger.info("error in loading image:{}".format(img_path))
+                continue
+            images.append(img)
+        return images
+
+    def predict(self, images=[], paths=[]):
+        """
+        Get the chinese texts in the predicted images.
+        Args:
+            images (list(numpy.ndarray)): images data, shape of each is [H, W, C]. If images not paths
+            paths (list[str]): The paths of images. If paths not images
+        Returns:
+            res (list): The layout results of images.
+        """
+
+        if images != [] and isinstance(images, list) and paths == []:
+            predicted_data = images
+        elif images == [] and isinstance(paths, list) and paths != []:
+            predicted_data = self.read_images(paths)
+        else:
+            raise TypeError("The input data is inconsistent with expectations.")
+
+        assert (
+            predicted_data != []
+        ), "There is not any image to be predicted. Please check the input data."
+
+        all_results = []
+        for img in predicted_data:
+            if img is None:
+                logger.info("error in loading image")
+                all_results.append([])
+                continue
+            starttime = time.time()
+            res, _ = self.layout_predictor(img)
+            elapse = time.time() - starttime
+            logger.info("Predict time: {}".format(elapse))
+
+            for item in res:
+                item["bbox"] = item["bbox"].tolist()
+            all_results.append({"layout": res})
+        return all_results
+
+    @serving
+    def serving_method(self, images, **kwargs):
+        """
+        Run as a service.
+        """
+        images_decode = [base64_to_cv2(image) for image in images]
+        results = self.predict(images_decode, **kwargs)
+        return results
+
+
+if __name__ == "__main__":
+    layout = LayoutPredictor()
+    layout._initialize()
+    image_path = ["./ppstructure/docs/table/1.png"]
+    res = layout.predict(paths=image_path)
+    print(res)
diff --git a/deploy/hubserving/structure_layout/params.py b/deploy/hubserving/structure_layout/params.py
new file mode 100755
index 0000000..5c0b2e8
--- /dev/null
+++ b/deploy/hubserving/structure_layout/params.py
@@ -0,0 +1,32 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+
+class Config(object):
+    pass
+
+
+def read_params():
+    cfg = Config()
+
+    # params for layout analysis
+    cfg.layout_model_dir = "./inference/picodet_lcnet_x1_0_fgd_layout_infer/"
+    cfg.layout_dict_path = "./ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt"
+    cfg.layout_score_threshold = 0.5
+    cfg.layout_nms_threshold = 0.5
+    return cfg
diff --git a/deploy/hubserving/structure_system/__init__.py b/deploy/hubserving/structure_system/__init__.py
new file mode 100644
index 0000000..97043fd
--- /dev/null
+++ b/deploy/hubserving/structure_system/__init__.py
@@ -0,0 +1,13 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/deploy/hubserving/structure_system/config.json b/deploy/hubserving/structure_system/config.json
new file mode 100644
index 0000000..7b6c166
--- /dev/null
+++ b/deploy/hubserving/structure_system/config.json
@@ -0,0 +1,15 @@
+{
+    "modules_info": {
+        "structure_system": {
+            "init_args": {
+                "version": "1.0.0",
+                "use_gpu": true
+            },
+            "predict_args": {
+            }
+        }
+    },
+    "port": 8870,
+    "use_multiprocess": false,
+    "workers": 2
+}
diff --git a/deploy/hubserving/structure_system/module.py b/deploy/hubserving/structure_system/module.py
new file mode 100644
index 0000000..571cccd
--- /dev/null
+++ b/deploy/hubserving/structure_system/module.py
@@ -0,0 +1,154 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+sys.path.insert(0, ".")
+import copy
+
+import time
+import paddlehub
+from paddlehub.common.logger import logger
+from paddlehub.module.module import moduleinfo, runnable, serving
+import cv2
+import numpy as np
+import paddlehub as hub
+
+from tools.infer.utility import base64_to_cv2
+from ppstructure.predict_system import StructureSystem as PPStructureSystem
+from ppstructure.predict_system import save_structure_res
+from ppstructure.utility import parse_args
+from deploy.hubserving.structure_system.params import read_params
+
+
+@moduleinfo(
+    name="structure_system",
+    version="1.0.0",
+    summary="PP-Structure system service",
+    author="paddle-dev",
+    author_email="paddle-dev@baidu.com",
+    type="cv/structure_system",
+)
+class StructureSystem(hub.Module):
+    def _initialize(self, use_gpu=False, enable_mkldnn=False):
+        """
+        initialize with the necessary elements
+        """
+        cfg = self.merge_configs()
+
+        cfg.use_gpu = use_gpu
+        if use_gpu:
+            try:
+                _places = os.environ["CUDA_VISIBLE_DEVICES"]
+                int(_places[0])
+                print("use gpu: ", use_gpu)
+                print("CUDA_VISIBLE_DEVICES: ", _places)
+                cfg.gpu_mem = 8000
+            except:
+                raise RuntimeError(
+                    "Environment Variable CUDA_VISIBLE_DEVICES is not set correctly. If you wanna use gpu, please set CUDA_VISIBLE_DEVICES via export CUDA_VISIBLE_DEVICES=cuda_device_id."
+                )
+        cfg.ir_optim = True
+        cfg.enable_mkldnn = enable_mkldnn
+
+        self.table_sys = PPStructureSystem(cfg)
+
+    def merge_configs(self):
+        # default cfg
+        backup_argv = copy.deepcopy(sys.argv)
+        sys.argv = sys.argv[:1]
+        cfg = parse_args()
+
+        update_cfg_map = vars(read_params())
+
+        for key in update_cfg_map:
+            cfg.__setattr__(key, update_cfg_map[key])
+
+        sys.argv = copy.deepcopy(backup_argv)
+        return cfg
+
+    def read_images(self, paths=[]):
+        images = []
+        for img_path in paths:
+            assert os.path.isfile(img_path), "The {} isn't a valid file.".format(
+                img_path
+            )
+            img = cv2.imread(img_path)
+            if img is None:
+                logger.info("error in loading image:{}".format(img_path))
+                continue
+            images.append(img)
+        return images
+
+    def predict(self, images=[], paths=[]):
+        """
+        Get the chinese texts in the predicted images.
+        Args:
+            images (list(numpy.ndarray)): images data, shape of each is [H, W, C]. If images not paths
+            paths (list[str]): The paths of images. If paths not images
+        Returns:
+            res (list): The result of chinese texts and save path of images.
+        """
+
+        if images != [] and isinstance(images, list) and paths == []:
+            predicted_data = images
+        elif images == [] and isinstance(paths, list) and paths != []:
+            predicted_data = self.read_images(paths)
+        else:
+            raise TypeError("The input data is inconsistent with expectations.")
+
+        assert (
+            predicted_data != []
+        ), "There is not any image to be predicted. Please check the input data."
+
+        all_results = []
+        for img in predicted_data:
+            if img is None:
+                logger.info("error in loading image")
+                all_results.append([])
+                continue
+            starttime = time.time()
+            res, _ = self.table_sys(img)
+            elapse = time.time() - starttime
+            logger.info("Predict time: {}".format(elapse))
+
+            # parse result
+            res_final = []
+            for region in res:
+                region.pop("img")
+                res_final.append(region)
+            all_results.append({"regions": res_final})
+        return all_results
+
+    @serving
+    def serving_method(self, images, **kwargs):
+        """
+        Run as a service.
+        """
+        images_decode = [base64_to_cv2(image) for image in images]
+        results = self.predict(images_decode, **kwargs)
+        return results
+
+
+if __name__ == "__main__":
+    structure_system = StructureSystem()
+    structure_system._initialize()
+    image_path = ["./ppstructure/docs/table/1.png"]
+    res = structure_system.predict(paths=image_path)
+    print(res)
diff --git a/deploy/hubserving/structure_system/params.py b/deploy/hubserving/structure_system/params.py
new file mode 100755
index 0000000..c5d28de
--- /dev/null
+++ b/deploy/hubserving/structure_system/params.py
@@ -0,0 +1,33 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from deploy.hubserving.structure_table.params import read_params as table_read_params
+
+
+def read_params():
+    cfg = table_read_params()
+
+    # params for layout parser model
+    cfg.layout_model_dir = ""
+    cfg.layout_dict_path = "./ppocr/utils/dict/layout_publaynet_dict.txt"
+    cfg.layout_score_threshold = 0.5
+    cfg.layout_nms_threshold = 0.5
+
+    cfg.mode = "structure"
+    cfg.output = "./output"
+    return cfg
diff --git a/deploy/hubserving/structure_table/__init__.py b/deploy/hubserving/structure_table/__init__.py
new file mode 100644
index 0000000..97043fd
--- /dev/null
+++ b/deploy/hubserving/structure_table/__init__.py
@@ -0,0 +1,13 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/deploy/hubserving/structure_table/config.json b/deploy/hubserving/structure_table/config.json
new file mode 100644
index 0000000..9ce15fc
--- /dev/null
+++ b/deploy/hubserving/structure_table/config.json
@@ -0,0 +1,15 @@
+{
+    "modules_info": {
+        "structure_table": {
+            "init_args": {
+                "version": "1.0.0",
+                "use_gpu": true
+            },
+            "predict_args": {
+            }
+        }
+    },
+    "port": 8869,
+    "use_multiprocess": false,
+    "workers": 2
+}
diff --git a/deploy/hubserving/structure_table/module.py b/deploy/hubserving/structure_table/module.py
new file mode 100644
index 0000000..8d4d613
--- /dev/null
+++ b/deploy/hubserving/structure_table/module.py
@@ -0,0 +1,148 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+sys.path.insert(0, ".")
+import copy
+
+import time
+import paddlehub
+from paddlehub.common.logger import logger
+from paddlehub.module.module import moduleinfo, runnable, serving
+import cv2
+import numpy as np
+import paddlehub as hub
+
+from tools.infer.utility import base64_to_cv2
+from ppstructure.table.predict_table import TableSystem as _TableSystem
+from ppstructure.predict_system import save_structure_res
+from ppstructure.utility import parse_args
+from deploy.hubserving.structure_table.params import read_params
+
+
+@moduleinfo(
+    name="structure_table",
+    version="1.0.0",
+    summary="PP-Structure table service",
+    author="paddle-dev",
+    author_email="paddle-dev@baidu.com",
+    type="cv/structure_table",
+)
+class TableSystem(hub.Module):
+    def _initialize(self, use_gpu=False, enable_mkldnn=False):
+        """
+        initialize with the necessary elements
+        """
+        cfg = self.merge_configs()
+        cfg.use_gpu = use_gpu
+        if use_gpu:
+            try:
+                _places = os.environ["CUDA_VISIBLE_DEVICES"]
+                int(_places[0])
+                print("use gpu: ", use_gpu)
+                print("CUDA_VISIBLE_DEVICES: ", _places)
+                cfg.gpu_mem = 8000
+            except:
+                raise RuntimeError(
+                    "Environment Variable CUDA_VISIBLE_DEVICES is not set correctly. If you wanna use gpu, please set CUDA_VISIBLE_DEVICES via export CUDA_VISIBLE_DEVICES=cuda_device_id."
+                )
+        cfg.ir_optim = True
+        cfg.enable_mkldnn = enable_mkldnn
+
+        self.table_sys = _TableSystem(cfg)
+
+    def merge_configs(self):
+        # default cfg
+        backup_argv = copy.deepcopy(sys.argv)
+        sys.argv = sys.argv[:1]
+        cfg = parse_args()
+
+        update_cfg_map = vars(read_params())
+
+        for key in update_cfg_map:
+            cfg.__setattr__(key, update_cfg_map[key])
+
+        sys.argv = copy.deepcopy(backup_argv)
+        return cfg
+
+    def read_images(self, paths=[]):
+        images = []
+        for img_path in paths:
+            assert os.path.isfile(img_path), "The {} isn't a valid file.".format(
+                img_path
+            )
+            img = cv2.imread(img_path)
+            if img is None:
+                logger.info("error in loading image:{}".format(img_path))
+                continue
+            images.append(img)
+        return images
+
+    def predict(self, images=[], paths=[]):
+        """
+        Get the chinese texts in the predicted images.
+        Args:
+            images (list(numpy.ndarray)): images data, shape of each is [H, W, C]. If images not paths
+            paths (list[str]): The paths of images. If paths not images
+        Returns:
+            res (list): The result of chinese texts and save path of images.
+        """
+
+        if images != [] and isinstance(images, list) and paths == []:
+            predicted_data = images
+        elif images == [] and isinstance(paths, list) and paths != []:
+            predicted_data = self.read_images(paths)
+        else:
+            raise TypeError("The input data is inconsistent with expectations.")
+
+        assert (
+            predicted_data != []
+        ), "There is not any image to be predicted. Please check the input data."
+
+        all_results = []
+        for img in predicted_data:
+            if img is None:
+                logger.info("error in loading image")
+                all_results.append([])
+                continue
+            starttime = time.time()
+            res, _ = self.table_sys(img)
+            elapse = time.time() - starttime
+            logger.info("Predict time: {}".format(elapse))
+
+            all_results.append({"html": res["html"]})
+        return all_results
+
+    @serving
+    def serving_method(self, images, **kwargs):
+        """
+        Run as a service.
+        """
+        images_decode = [base64_to_cv2(image) for image in images]
+        results = self.predict(images_decode, **kwargs)
+        return results
+
+
+if __name__ == "__main__":
+    table_system = TableSystem()
+    table_system._initialize()
+    image_path = ["./ppstructure/docs/table/table.jpg"]
+    res = table_system.predict(paths=image_path)
+    print(res)
diff --git a/deploy/hubserving/structure_table/params.py b/deploy/hubserving/structure_table/params.py
new file mode 100755
index 0000000..1b0bf0c
--- /dev/null
+++ b/deploy/hubserving/structure_table/params.py
@@ -0,0 +1,30 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from deploy.hubserving.ocr_system.params import read_params as pp_ocr_read_params
+
+
+def read_params():
+    cfg = pp_ocr_read_params()
+
+    # params for table structure model
+    cfg.table_max_len = 488
+    cfg.table_model_dir = "./inference/en_ppocr_mobile_v2.0_table_structure_infer/"
+    cfg.table_char_dict_path = "./ppocr/utils/dict/table_structure_dict.txt"
+    cfg.show_log = False
+    return cfg
diff --git a/deploy/ios_demo/.gitignore b/deploy/ios_demo/.gitignore
new file mode 100644
index 0000000..6c55cd7
--- /dev/null
+++ b/deploy/ios_demo/.gitignore
@@ -0,0 +1,21 @@
+# CocoaPods
+Pods/
+
+# Xcode
+PaddleOCRDemo.xcworkspace/xcuserdata/
+PaddleOCRDemo.xcodeproj/xcuserdata/
+PaddleOCRDemo.xcodeproj/project.xcworkspace/xcuserdata/
+*.xcuserstate
+DerivedData/
+build/
+
+# fetch_ios_demo_models.sh download cache
+.fetch_ios_demo_models_work/
+
+# Models — ignore all files under Models/; track only `.gitkeep` so det/, rec/, … stay in the tree
+PaddleOCRDemo/Models/**
+!PaddleOCRDemo/Models/*/
+!PaddleOCRDemo/Models/*/.gitkeep
+
+# Local validation reports
+out/
diff --git a/deploy/ios_demo/NOTICE b/deploy/ios_demo/NOTICE
new file mode 100644
index 0000000..e0c1ff1
--- /dev/null
+++ b/deploy/ios_demo/NOTICE
@@ -0,0 +1,25 @@
+PaddleOCR iOS demo — third-party notices
+===============================================================================
+
+This demo application is part of the PaddleOCR project (Apache License 2.0).
+It may bundle or link third-party components in addition to the PaddleOCR
+repository LICENSE. The following components are included in **source** form
+in this tree:
+
+Clipper / polyclipping (polygon clipping and offsetting)
+--------------------------------------------------------------------------------------------
+  • Upstream: polyclipping clipper 6.4.2 (Angus Johnson)  
+  • Bundled:   `PaddleOCRDemo/ThirdParty/Clipper1/` (clipper.cpp, clipper.hpp)  
+  • SPDX:      BSL-1.0 (Boost Software License 1.0)  
+  • Full text: `PaddleOCRDemo/ThirdParty/Clipper1/LICENSE`  
+
+CocoaPods dependencies (onnxruntime-objc, OpenCV, Yams, etc.)
+-------------------------------------------------------------
+  • Versions and upstream packages are recorded in `Podfile.lock` after
+    `pod install`. Each pod is subject to its own license as published by
+    the respective project; consult the Pods project and upstream repos
+    for notices required by those licenses.
+
+If you distribute binaries built from this demo, ensure your product’s
+open-source / third-party attribution (e.g. in-app “Licenses” or
+distribution packaging) satisfies your legal and policy requirements.
diff --git a/deploy/ios_demo/PaddleOCRDemo.xcodeproj/project.pbxproj b/deploy/ios_demo/PaddleOCRDemo.xcodeproj/project.pbxproj
new file mode 100644
index 0000000..6bca78b
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo.xcodeproj/project.pbxproj
@@ -0,0 +1,822 @@
+// !$*UTF8*$!
+{
+	archiveVersion = 1;
+	classes = {
+	};
+	objectVersion = 56;
+	objects = {
+
+/* Begin PBXBuildFile section */
+		6F50EE661821421A98371746 /* Pods_PaddleOCRDemoTests.framework in Frameworks */ = {isa = PBXBuildFile; fileRef = 0A24555277519FE15EF10A5D /* Pods_PaddleOCRDemoTests.framework */; };
+		A1000001 /* PaddleOCRDemoApp.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000001 /* PaddleOCRDemoApp.swift */; };
+		A1000002 /* ContentView.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000002 /* ContentView.swift */; };
+		A1000003 /* Models in Resources */ = {isa = PBXBuildFile; fileRef = A2000006 /* Models */; };
+		A1000004 /* ModelConfig.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000011 /* ModelConfig.swift */; };
+		A1000005 /* ORTSessionManager.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000012 /* ORTSessionManager.swift */; };
+		A10000F0 /* ORTProfilingBridge.mm in Sources */ = {isa = PBXBuildFile; fileRef = A20000F1 /* ORTProfilingBridge.mm */; };
+		A1000007 /* InferenceConfig.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000015 /* InferenceConfig.swift */; };
+		A1000008 /* Preprocessing.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000016 /* Preprocessing.swift */; };
+		A1000011 /* ORTSessionManagerTests.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000014 /* ORTSessionManagerTests.swift */; };
+		A1000021 /* DBPostProcess.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000021 /* DBPostProcess.swift */; };
+		A1000022 /* OpenCVDBBridge.mm in Sources */ = {isa = PBXBuildFile; fileRef = A2000023 /* OpenCVDBBridge.mm */; };
+		A1000023 /* OpenCVImageBridge.mm in Sources */ = {isa = PBXBuildFile; fileRef = A2000025 /* OpenCVImageBridge.mm */; };
+		A10000E1 /* clipper.cpp in Sources */ = {isa = PBXBuildFile; fileRef = A20000E1 /* clipper.cpp */; };
+		A10000E2 /* PDBPolygonOffsetBridge.mm in Sources */ = {isa = PBXBuildFile; fileRef = A20000E2 /* PDBPolygonOffsetBridge.mm */; };
+		A10000C0 /* DemoSectionChrome.swift in Sources */ = {isa = PBXBuildFile; fileRef = A20000C0 /* DemoSectionChrome.swift */; };
+		A1000030 /* DetectionEngine.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000030 /* DetectionEngine.swift */; };
+		A1000040 /* RecPreprocessor.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000040 /* RecPreprocessor.swift */; };
+		A1000050 /* CTCDecoder.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000050 /* CTCDecoder.swift */; };
+		A1000060 /* RecognitionEngine.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000060 /* RecognitionEngine.swift */; };
+		A1000070 /* OCREngine.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000070 /* OCREngine.swift */; };
+		A1000071 /* BoxSorter.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000071 /* BoxSorter.swift */; };
+		A1000072 /* QuadTextCrop.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000072 /* QuadTextCrop.swift */; };
+		A1000074 /* EncodedImageCodec.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000074 /* EncodedImageCodec.swift */; };
+		A1000080 /* OCRViewModel.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000080 /* OCRViewModel.swift */; };
+		A1000090 /* ImagePickerSection.swift in Sources */ = {isa = PBXBuildFile; fileRef = A2000090 /* ImagePickerSection.swift */; };
+		A1000091 /* SampleImages in Resources */ = {isa = PBXBuildFile; fileRef = A2000091 /* SampleImages */; };
+		A10000A0 /* ResultImageView.swift in Sources */ = {isa = PBXBuildFile; fileRef = A20000A0 /* ResultImageView.swift */; };
+		A10000A1 /* TimingView.swift in Sources */ = {isa = PBXBuildFile; fileRef = A20000A1 /* TimingView.swift */; };
+		A10000A2 /* ResultsListView.swift in Sources */ = {isa = PBXBuildFile; fileRef = A20000A2 /* ResultsListView.swift */; };
+		A10000A3 /* ErrorView.swift in Sources */ = {isa = PBXBuildFile; fileRef = A20000A3 /* ErrorView.swift */; };
+		A10000A4 /* OCRParametersPanel.swift in Sources */ = {isa = PBXBuildFile; fileRef = A20000A4 /* OCRParametersPanel.swift */; };
+		A10000A5 /* OnnxRuntimeSettingsPanel.swift in Sources */ = {isa = PBXBuildFile; fileRef = A20000A5 /* OnnxRuntimeSettingsPanel.swift */; };
+		A10000A6 /* EngineSettingsSheet.swift in Sources */ = {isa = PBXBuildFile; fileRef = A20000A6 /* EngineSettingsSheet.swift */; };
+		A10000B3 /* OCRParameterResolverTests.swift in Sources */ = {isa = PBXBuildFile; fileRef = A20000B3 /* OCRParameterResolverTests.swift */; };
+		A10000C5 /* OCRBenchmarkTests.swift in Sources */ = {isa = PBXBuildFile; fileRef = A20000C5 /* OCRBenchmarkTests.swift */; };
+		A10000C6 /* BenchmarkPayloads.swift in Sources */ = {isa = PBXBuildFile; fileRef = A20000C6 /* BenchmarkPayloads.swift */; };
+		A10000D3 /* BenchmarkFixtures.swift in Sources */ = {isa = PBXBuildFile; fileRef = A20000D3 /* BenchmarkFixtures.swift */; };
+		A10000C7 /* Fixtures in Resources */ = {isa = PBXBuildFile; fileRef = A20000C7 /* Fixtures */; };
+		A10000B1 /* OCRParameterResolver.swift in Sources */ = {isa = PBXBuildFile; fileRef = A20000B1 /* OCRParameterResolver.swift */; };
+		A10000D0 /* Assets.xcassets in Resources */ = {isa = PBXBuildFile; fileRef = A20000D0 /* Assets.xcassets */; };
+		A10000D1 /* Bundle+BundledImage.swift in Sources */ = {isa = PBXBuildFile; fileRef = A20000D1 /* Bundle+BundledImage.swift */; };
+		A10000D2 /* DemoAssets.swift in Sources */ = {isa = PBXBuildFile; fileRef = A20000D2 /* DemoAssets.swift */; };
+		F9888FA77A2AF8AF45B50193 /* Pods_PaddleOCRDemo.framework in Frameworks */ = {isa = PBXBuildFile; fileRef = 3D0FFA11A4E43E5BD0F80219 /* Pods_PaddleOCRDemo.framework */; };
+/* End PBXBuildFile section */
+
+/* Begin PBXContainerItemProxy section */
+		A5000001 /* PBXContainerItemProxy */ = {
+			isa = PBXContainerItemProxy;
+			containerPortal = A3000001 /* Project object */;
+			proxyType = 1;
+			remoteGlobalIDString = A4000001;
+			remoteInfo = PaddleOCRDemo;
+		};
+/* End PBXContainerItemProxy section */
+
+/* Begin PBXFileReference section */
+		0A24555277519FE15EF10A5D /* Pods_PaddleOCRDemoTests.framework */ = {isa = PBXFileReference; explicitFileType = wrapper.framework; includeInIndex = 0; path = Pods_PaddleOCRDemoTests.framework; sourceTree = BUILT_PRODUCTS_DIR; };
+		3D0FFA11A4E43E5BD0F80219 /* Pods_PaddleOCRDemo.framework */ = {isa = PBXFileReference; explicitFileType = wrapper.framework; includeInIndex = 0; path = Pods_PaddleOCRDemo.framework; sourceTree = BUILT_PRODUCTS_DIR; };
+		49C7CEE60686CC98BE2EEE45 /* Pods-PaddleOCRDemoTests.release.xcconfig */ = {isa = PBXFileReference; includeInIndex = 1; lastKnownFileType = text.xcconfig; name = "Pods-PaddleOCRDemoTests.release.xcconfig"; path = "Target Support Files/Pods-PaddleOCRDemoTests/Pods-PaddleOCRDemoTests.release.xcconfig"; sourceTree = "<group>"; };
+		520BC3C496ED955BBB302DC4 /* Pods-PaddleOCRDemo.debug.xcconfig */ = {isa = PBXFileReference; includeInIndex = 1; lastKnownFileType = text.xcconfig; name = "Pods-PaddleOCRDemo.debug.xcconfig"; path = "Target Support Files/Pods-PaddleOCRDemo/Pods-PaddleOCRDemo.debug.xcconfig"; sourceTree = "<group>"; };
+		54F64B7C830508805D832F8A /* Pods-PaddleOCRDemo.release.xcconfig */ = {isa = PBXFileReference; includeInIndex = 1; lastKnownFileType = text.xcconfig; name = "Pods-PaddleOCRDemo.release.xcconfig"; path = "Target Support Files/Pods-PaddleOCRDemo/Pods-PaddleOCRDemo.release.xcconfig"; sourceTree = "<group>"; };
+		72FF5C269BF861629F89D6B5 /* Pods-PaddleOCRDemoTests.debug.xcconfig */ = {isa = PBXFileReference; includeInIndex = 1; lastKnownFileType = text.xcconfig; name = "Pods-PaddleOCRDemoTests.debug.xcconfig"; path = "Target Support Files/Pods-PaddleOCRDemoTests/Pods-PaddleOCRDemoTests.debug.xcconfig"; sourceTree = "<group>"; };
+		A2000001 /* PaddleOCRDemoApp.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = PaddleOCRDemoApp.swift; sourceTree = "<group>"; };
+		A2000002 /* ContentView.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = ContentView.swift; sourceTree = "<group>"; };
+		A2000003 /* PaddleOCRDemo-Bridging-Header.h */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.c.h; path = "PaddleOCRDemo-Bridging-Header.h"; sourceTree = "<group>"; };
+		A2000004 /* Info.plist */ = {isa = PBXFileReference; lastKnownFileType = text.plist.xml; path = Info.plist; sourceTree = "<group>"; };
+		A2000005 /* PaddleOCRDemo.app */ = {isa = PBXFileReference; explicitFileType = wrapper.application; includeInIndex = 0; path = PaddleOCRDemo.app; sourceTree = BUILT_PRODUCTS_DIR; };
+		A2000006 /* Models */ = {isa = PBXFileReference; lastKnownFileType = folder; path = Models; sourceTree = "<group>"; };
+		A2000009 /* PaddleOCRDemoTests.xctest */ = {isa = PBXFileReference; explicitFileType = wrapper.cfbundle; includeInIndex = 0; path = PaddleOCRDemoTests.xctest; sourceTree = BUILT_PRODUCTS_DIR; };
+		A2000011 /* ModelConfig.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = ModelConfig.swift; sourceTree = "<group>"; };
+		A2000012 /* ORTSessionManager.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = ORTSessionManager.swift; sourceTree = "<group>"; };
+		A20000F0 /* ORTProfilingBridge.h */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.c.h; path = ORTProfilingBridge.h; sourceTree = "<group>"; };
+		A20000F1 /* ORTProfilingBridge.mm */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.cpp.objcpp; path = ORTProfilingBridge.mm; sourceTree = "<group>"; };
+		A2000014 /* ORTSessionManagerTests.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = ORTSessionManagerTests.swift; sourceTree = "<group>"; };
+		A2000015 /* InferenceConfig.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = InferenceConfig.swift; sourceTree = "<group>"; };
+		A2000016 /* Preprocessing.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = Preprocessing.swift; sourceTree = "<group>"; };
+		A2000021 /* DBPostProcess.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = DBPostProcess.swift; sourceTree = "<group>"; };
+		A2000022 /* OpenCVDBBridge.h */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.c.h; path = OpenCVDBBridge.h; sourceTree = "<group>"; };
+		A2000023 /* OpenCVDBBridge.mm */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.cpp.objcpp; path = OpenCVDBBridge.mm; sourceTree = "<group>"; };
+		A2000024 /* OpenCVImageBridge.h */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.c.h; path = OpenCVImageBridge.h; sourceTree = "<group>"; };
+		A2000025 /* OpenCVImageBridge.mm */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.cpp.objcpp; path = OpenCVImageBridge.mm; sourceTree = "<group>"; };
+		A20000E1 /* clipper.cpp */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.cpp.cpp; path = clipper.cpp; sourceTree = "<group>"; };
+		A20000E2 /* PDBPolygonOffsetBridge.mm */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.cpp.objcpp; path = PDBPolygonOffsetBridge.mm; sourceTree = "<group>"; };
+		A20000E3 /* PDBPolygonOffsetBridge.h */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.c.h; path = PDBPolygonOffsetBridge.h; sourceTree = "<group>"; };
+		A2000030 /* DetectionEngine.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = DetectionEngine.swift; sourceTree = "<group>"; };
+		A2000040 /* RecPreprocessor.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = RecPreprocessor.swift; sourceTree = "<group>"; };
+		A2000050 /* CTCDecoder.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = CTCDecoder.swift; sourceTree = "<group>"; };
+		A2000060 /* RecognitionEngine.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = RecognitionEngine.swift; sourceTree = "<group>"; };
+		A2000070 /* OCREngine.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = OCREngine.swift; sourceTree = "<group>"; };
+		A2000071 /* BoxSorter.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = BoxSorter.swift; sourceTree = "<group>"; };
+		A2000072 /* QuadTextCrop.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = QuadTextCrop.swift; sourceTree = "<group>"; };
+		A2000074 /* EncodedImageCodec.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = EncodedImageCodec.swift; sourceTree = "<group>"; };
+		A2000080 /* OCRViewModel.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = OCRViewModel.swift; sourceTree = "<group>"; };
+		A2000090 /* ImagePickerSection.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = ImagePickerSection.swift; sourceTree = "<group>"; };
+		A2000091 /* SampleImages */ = {isa = PBXFileReference; lastKnownFileType = folder; path = SampleImages; sourceTree = "<group>"; };
+		A20000A0 /* ResultImageView.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = ResultImageView.swift; sourceTree = "<group>"; };
+		A20000A1 /* TimingView.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = TimingView.swift; sourceTree = "<group>"; };
+		A20000A2 /* ResultsListView.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = ResultsListView.swift; sourceTree = "<group>"; };
+		A20000A3 /* ErrorView.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = ErrorView.swift; sourceTree = "<group>"; };
+		A20000A4 /* OCRParametersPanel.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = OCRParametersPanel.swift; sourceTree = "<group>"; };
+		A20000A5 /* OnnxRuntimeSettingsPanel.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = OnnxRuntimeSettingsPanel.swift; sourceTree = "<group>"; };
+		A20000A6 /* EngineSettingsSheet.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = EngineSettingsSheet.swift; sourceTree = "<group>"; };
+		A20000B3 /* OCRParameterResolverTests.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = OCRParameterResolverTests.swift; sourceTree = "<group>"; };
+		A20000C5 /* OCRBenchmarkTests.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = OCRBenchmarkTests.swift; sourceTree = "<group>"; };
+		A20000C6 /* BenchmarkPayloads.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = BenchmarkPayloads.swift; sourceTree = "<group>"; };
+		A20000D3 /* BenchmarkFixtures.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = BenchmarkFixtures.swift; sourceTree = "<group>"; };
+		A20000C7 /* Fixtures */ = {isa = PBXFileReference; lastKnownFileType = folder; path = Fixtures; sourceTree = "<group>"; };
+		A20000B1 /* OCRParameterResolver.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = OCRParameterResolver.swift; sourceTree = "<group>"; };
+		A20000C0 /* DemoSectionChrome.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = DemoSectionChrome.swift; sourceTree = "<group>"; };
+		A20000D0 /* Assets.xcassets */ = {isa = PBXFileReference; lastKnownFileType = folder.assetcatalog; path = Assets.xcassets; sourceTree = "<group>"; };
+		A20000D1 /* Bundle+BundledImage.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = "Bundle+BundledImage.swift"; sourceTree = "<group>"; };
+		A20000D2 /* DemoAssets.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = DemoAssets.swift; sourceTree = "<group>"; };
+/* End PBXFileReference section */
+
+/* Begin PBXFrameworksBuildPhase section */
+		A6000001 /* Frameworks */ = {
+			isa = PBXFrameworksBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+				F9888FA77A2AF8AF45B50193 /* Pods_PaddleOCRDemo.framework in Frameworks */,
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+		A6000002 /* Frameworks */ = {
+			isa = PBXFrameworksBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+				6F50EE661821421A98371746 /* Pods_PaddleOCRDemoTests.framework in Frameworks */,
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+/* End PBXFrameworksBuildPhase section */
+
+/* Begin PBXGroup section */
+		2D2CDE3432BBDFC183DD82E7 /* Frameworks */ = {
+			isa = PBXGroup;
+			children = (
+				3D0FFA11A4E43E5BD0F80219 /* Pods_PaddleOCRDemo.framework */,
+				0A24555277519FE15EF10A5D /* Pods_PaddleOCRDemoTests.framework */,
+			);
+			name = Frameworks;
+			sourceTree = "<group>";
+		};
+		3E7A5973A61F2EB2D5542DE3 /* Pods */ = {
+			isa = PBXGroup;
+			children = (
+				520BC3C496ED955BBB302DC4 /* Pods-PaddleOCRDemo.debug.xcconfig */,
+				54F64B7C830508805D832F8A /* Pods-PaddleOCRDemo.release.xcconfig */,
+				72FF5C269BF861629F89D6B5 /* Pods-PaddleOCRDemoTests.debug.xcconfig */,
+				49C7CEE60686CC98BE2EEE45 /* Pods-PaddleOCRDemoTests.release.xcconfig */,
+			);
+			path = Pods;
+			sourceTree = "<group>";
+		};
+		A7000001 = {
+			isa = PBXGroup;
+			children = (
+				A7000002 /* PaddleOCRDemo */,
+				A7000005 /* PaddleOCRDemoTests */,
+				A7000003 /* Products */,
+				3E7A5973A61F2EB2D5542DE3 /* Pods */,
+				2D2CDE3432BBDFC183DD82E7 /* Frameworks */,
+			);
+			sourceTree = "<group>";
+		};
+		A7000002 /* PaddleOCRDemo */ = {
+			isa = PBXGroup;
+			children = (
+				A7000004 /* App */,
+				A7000007 /* Engine */,
+				A7000008 /* ViewModels */,
+				A7000009 /* Views */,
+				A7000006 /* Supporting */,
+				A7000010 /* Resources */,
+				A700000D /* ThirdParty */,
+				A2000006 /* Models */,
+			);
+			path = PaddleOCRDemo;
+			sourceTree = "<group>";
+		};
+		A7000003 /* Products */ = {
+			isa = PBXGroup;
+			children = (
+				A2000005 /* PaddleOCRDemo.app */,
+				A2000009 /* PaddleOCRDemoTests.xctest */,
+			);
+			name = Products;
+			sourceTree = "<group>";
+		};
+		A7000004 /* App */ = {
+			isa = PBXGroup;
+			children = (
+				A20000D1 /* Bundle+BundledImage.swift */,
+				A20000D2 /* DemoAssets.swift */,
+				A2000001 /* PaddleOCRDemoApp.swift */,
+				A2000002 /* ContentView.swift */,
+				A20000C0 /* DemoSectionChrome.swift */,
+			);
+			path = App;
+			sourceTree = "<group>";
+		};
+		A7000005 /* PaddleOCRDemoTests */ = {
+			isa = PBXGroup;
+			children = (
+				A2000014 /* ORTSessionManagerTests.swift */,
+				A20000B3 /* OCRParameterResolverTests.swift */,
+				A20000C5 /* OCRBenchmarkTests.swift */,
+				A20000C6 /* BenchmarkPayloads.swift */,
+				A20000D3 /* BenchmarkFixtures.swift */,
+				A20000C7 /* Fixtures */,
+			);
+			path = PaddleOCRDemoTests;
+			sourceTree = "<group>";
+		};
+		A7000006 /* Supporting */ = {
+			isa = PBXGroup;
+			children = (
+				A2000003 /* PaddleOCRDemo-Bridging-Header.h */,
+				A2000004 /* Info.plist */,
+			);
+			path = Supporting;
+			sourceTree = "<group>";
+		};
+		A7000007 /* Engine */ = {
+			isa = PBXGroup;
+			children = (
+				A2000071 /* BoxSorter.swift */,
+				A2000050 /* CTCDecoder.swift */,
+				A2000021 /* DBPostProcess.swift */,
+				A2000022 /* OpenCVDBBridge.h */,
+				A2000023 /* OpenCVDBBridge.mm */,
+				A2000024 /* OpenCVImageBridge.h */,
+				A2000025 /* OpenCVImageBridge.mm */,
+				A20000E2 /* PDBPolygonOffsetBridge.mm */,
+				A20000E3 /* PDBPolygonOffsetBridge.h */,
+				A2000030 /* DetectionEngine.swift */,
+				A2000015 /* InferenceConfig.swift */,
+				A20000B1 /* OCRParameterResolver.swift */,
+				A2000011 /* ModelConfig.swift */,
+				A2000070 /* OCREngine.swift */,
+				A2000012 /* ORTSessionManager.swift */,
+				A20000F0 /* ORTProfilingBridge.h */,
+				A20000F1 /* ORTProfilingBridge.mm */,
+				A2000072 /* QuadTextCrop.swift */,
+				A2000074 /* EncodedImageCodec.swift */,
+				A2000016 /* Preprocessing.swift */,
+				A2000040 /* RecPreprocessor.swift */,
+				A2000060 /* RecognitionEngine.swift */,
+			);
+			path = Engine;
+			sourceTree = "<group>";
+		};
+		A7000008 /* ViewModels */ = {
+			isa = PBXGroup;
+			children = (
+				A2000080 /* OCRViewModel.swift */,
+			);
+			path = ViewModels;
+			sourceTree = "<group>";
+		};
+		A7000009 /* Views */ = {
+			isa = PBXGroup;
+			children = (
+				A20000A3 /* ErrorView.swift */,
+				A2000090 /* ImagePickerSection.swift */,
+				A20000A5 /* OnnxRuntimeSettingsPanel.swift */,
+				A20000A4 /* OCRParametersPanel.swift */,
+				A20000A6 /* EngineSettingsSheet.swift */,
+				A20000A0 /* ResultImageView.swift */,
+				A20000A2 /* ResultsListView.swift */,
+				A20000A1 /* TimingView.swift */,
+			);
+			path = Views;
+			sourceTree = "<group>";
+		};
+		A700000D /* ThirdParty */ = {
+			isa = PBXGroup;
+			children = (
+				A700000E /* Clipper1 */,
+			);
+			path = ThirdParty;
+			sourceTree = "<group>";
+		};
+		A700000E /* Clipper1 */ = {
+			isa = PBXGroup;
+			children = (
+				A20000E1 /* clipper.cpp */,
+			);
+			path = Clipper1;
+			sourceTree = "<group>";
+		};
+		A7000010 /* Resources */ = {
+			isa = PBXGroup;
+			children = (
+				A20000D0 /* Assets.xcassets */,
+				A2000091 /* SampleImages */,
+			);
+			path = Resources;
+			sourceTree = "<group>";
+		};
+/* End PBXGroup section */
+
+/* Begin PBXNativeTarget section */
+		A4000001 /* PaddleOCRDemo */ = {
+			isa = PBXNativeTarget;
+			buildConfigurationList = A9000003 /* Build configuration list for PBXNativeTarget "PaddleOCRDemo" */;
+			buildPhases = (
+				A9D5C3BAF4FDCBE19D45BCA7 /* [CP] Check Pods Manifest.lock */,
+				A8000001 /* Sources */,
+				A6000001 /* Frameworks */,
+				A8000003 /* Resources */,
+				6530B8008C9EFD4F162C3E86 /* [CP] Embed Pods Frameworks */,
+			);
+			buildRules = (
+			);
+			dependencies = (
+			);
+			name = PaddleOCRDemo;
+			productName = PaddleOCRDemo;
+			productReference = A2000005 /* PaddleOCRDemo.app */;
+			productType = "com.apple.product-type.application";
+		};
+		A4000002 /* PaddleOCRDemoTests */ = {
+			isa = PBXNativeTarget;
+			buildConfigurationList = A9000006 /* Build configuration list for PBXNativeTarget "PaddleOCRDemoTests" */;
+			buildPhases = (
+				9FD4B8CB3C3E1C27991BDE0B /* [CP] Check Pods Manifest.lock */,
+				A8000002 /* Sources */,
+				A6000002 /* Frameworks */,
+				A8000004 /* Resources */,
+			);
+			buildRules = (
+			);
+			dependencies = (
+				A5000002 /* PBXTargetDependency */,
+			);
+			name = PaddleOCRDemoTests;
+			productName = PaddleOCRDemoTests;
+			productReference = A2000009 /* PaddleOCRDemoTests.xctest */;
+			productType = "com.apple.product-type.bundle.unit-test";
+		};
+/* End PBXNativeTarget section */
+
+/* Begin PBXProject section */
+		A3000001 /* Project object */ = {
+			isa = PBXProject;
+			attributes = {
+				BuildIndependentTargetsInParallel = 1;
+				LastSwiftUpdateCheck = 1540;
+				LastUpgradeCheck = 2640;
+				TargetAttributes = {
+					A4000001 = {
+						CreatedOnToolsVersion = 15.4;
+					};
+					A4000002 = {
+						CreatedOnToolsVersion = 15.4;
+						TestTargetID = A4000001;
+					};
+				};
+			};
+			buildConfigurationList = A9000001 /* Build configuration list for PBXProject "PaddleOCRDemo" */;
+			compatibilityVersion = "Xcode 14.0";
+			developmentRegion = en;
+			hasScannedForEncodings = 0;
+			knownRegions = (
+				en,
+				Base,
+			);
+			mainGroup = A7000001;
+			productRefGroup = A7000003 /* Products */;
+			projectDirPath = "";
+			projectRoot = "";
+			targets = (
+				A4000001 /* PaddleOCRDemo */,
+				A4000002 /* PaddleOCRDemoTests */,
+			);
+		};
+/* End PBXProject section */
+
+/* Begin PBXResourcesBuildPhase section */
+		A8000003 /* Resources */ = {
+			isa = PBXResourcesBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+				A10000D0 /* Assets.xcassets in Resources */,
+				A1000003 /* Models in Resources */,
+				A1000091 /* SampleImages in Resources */,
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+		A8000004 /* Resources */ = {
+			isa = PBXResourcesBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+				A10000C7 /* Fixtures in Resources */,
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+/* End PBXResourcesBuildPhase section */
+
+/* Begin PBXShellScriptBuildPhase section */
+		6530B8008C9EFD4F162C3E86 /* [CP] Embed Pods Frameworks */ = {
+			isa = PBXShellScriptBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+			);
+			inputFileListPaths = (
+				"${PODS_ROOT}/Target Support Files/Pods-PaddleOCRDemo/Pods-PaddleOCRDemo-frameworks-${CONFIGURATION}-input-files.xcfilelist",
+			);
+			name = "[CP] Embed Pods Frameworks";
+			outputFileListPaths = (
+				"${PODS_ROOT}/Target Support Files/Pods-PaddleOCRDemo/Pods-PaddleOCRDemo-frameworks-${CONFIGURATION}-output-files.xcfilelist",
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+			shellPath = /bin/sh;
+			shellScript = "\"${PODS_ROOT}/Target Support Files/Pods-PaddleOCRDemo/Pods-PaddleOCRDemo-frameworks.sh\"\n";
+			showEnvVarsInLog = 0;
+		};
+		9FD4B8CB3C3E1C27991BDE0B /* [CP] Check Pods Manifest.lock */ = {
+			isa = PBXShellScriptBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+			);
+			inputFileListPaths = (
+			);
+			inputPaths = (
+				"${PODS_PODFILE_DIR_PATH}/Podfile.lock",
+				"${PODS_ROOT}/Manifest.lock",
+			);
+			name = "[CP] Check Pods Manifest.lock";
+			outputFileListPaths = (
+			);
+			outputPaths = (
+				"$(DERIVED_FILE_DIR)/Pods-PaddleOCRDemoTests-checkManifestLockResult.txt",
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+			shellPath = /bin/sh;
+			shellScript = "diff \"${PODS_PODFILE_DIR_PATH}/Podfile.lock\" \"${PODS_ROOT}/Manifest.lock\" > /dev/null\nif [ $? != 0 ] ; then\n    # print error to STDERR\n    echo \"error: The sandbox is not in sync with the Podfile.lock. Run 'pod install' or update your CocoaPods installation.\" >&2\n    exit 1\nfi\n# This output is used by Xcode 'outputs' to avoid re-running this script phase.\necho \"SUCCESS\" > \"${SCRIPT_OUTPUT_FILE_0}\"\n";
+			showEnvVarsInLog = 0;
+		};
+		A9D5C3BAF4FDCBE19D45BCA7 /* [CP] Check Pods Manifest.lock */ = {
+			isa = PBXShellScriptBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+			);
+			inputFileListPaths = (
+			);
+			inputPaths = (
+				"${PODS_PODFILE_DIR_PATH}/Podfile.lock",
+				"${PODS_ROOT}/Manifest.lock",
+			);
+			name = "[CP] Check Pods Manifest.lock";
+			outputFileListPaths = (
+			);
+			outputPaths = (
+				"$(DERIVED_FILE_DIR)/Pods-PaddleOCRDemo-checkManifestLockResult.txt",
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+			shellPath = /bin/sh;
+			shellScript = "diff \"${PODS_PODFILE_DIR_PATH}/Podfile.lock\" \"${PODS_ROOT}/Manifest.lock\" > /dev/null\nif [ $? != 0 ] ; then\n    # print error to STDERR\n    echo \"error: The sandbox is not in sync with the Podfile.lock. Run 'pod install' or update your CocoaPods installation.\" >&2\n    exit 1\nfi\n# This output is used by Xcode 'outputs' to avoid re-running this script phase.\necho \"SUCCESS\" > \"${SCRIPT_OUTPUT_FILE_0}\"\n";
+			showEnvVarsInLog = 0;
+		};
+/* End PBXShellScriptBuildPhase section */
+
+/* Begin PBXSourcesBuildPhase section */
+		A8000001 /* Sources */ = {
+			isa = PBXSourcesBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+				A1000001 /* PaddleOCRDemoApp.swift in Sources */,
+				A10000D1 /* Bundle+BundledImage.swift in Sources */,
+				A10000D2 /* DemoAssets.swift in Sources */,
+				A1000002 /* ContentView.swift in Sources */,
+				A10000C0 /* DemoSectionChrome.swift in Sources */,
+				A1000021 /* DBPostProcess.swift in Sources */,
+				A1000022 /* OpenCVDBBridge.mm in Sources */,
+				A1000023 /* OpenCVImageBridge.mm in Sources */,
+				A10000E1 /* clipper.cpp in Sources */,
+				A10000E2 /* PDBPolygonOffsetBridge.mm in Sources */,
+				A1000030 /* DetectionEngine.swift in Sources */,
+				A1000004 /* ModelConfig.swift in Sources */,
+				A1000005 /* ORTSessionManager.swift in Sources */,
+				A10000F0 /* ORTProfilingBridge.mm in Sources */,
+				A1000007 /* InferenceConfig.swift in Sources */,
+				A10000B1 /* OCRParameterResolver.swift in Sources */,
+				A1000008 /* Preprocessing.swift in Sources */,
+				A1000040 /* RecPreprocessor.swift in Sources */,
+				A1000050 /* CTCDecoder.swift in Sources */,
+				A1000060 /* RecognitionEngine.swift in Sources */,
+				A1000070 /* OCREngine.swift in Sources */,
+				A1000071 /* BoxSorter.swift in Sources */,
+				A1000072 /* QuadTextCrop.swift in Sources */,
+				A1000074 /* EncodedImageCodec.swift in Sources */,
+				A1000080 /* OCRViewModel.swift in Sources */,
+				A1000090 /* ImagePickerSection.swift in Sources */,
+				A10000A0 /* ResultImageView.swift in Sources */,
+				A10000A1 /* TimingView.swift in Sources */,
+				A10000A2 /* ResultsListView.swift in Sources */,
+				A10000A3 /* ErrorView.swift in Sources */,
+				A10000A5 /* OnnxRuntimeSettingsPanel.swift in Sources */,
+				A10000A6 /* EngineSettingsSheet.swift in Sources */,
+				A10000A4 /* OCRParametersPanel.swift in Sources */,
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+		A8000002 /* Sources */ = {
+			isa = PBXSourcesBuildPhase;
+			buildActionMask = 2147483647;
+			files = (
+				A1000011 /* ORTSessionManagerTests.swift in Sources */,
+				A10000B3 /* OCRParameterResolverTests.swift in Sources */,
+				A10000C5 /* OCRBenchmarkTests.swift in Sources */,
+				A10000C6 /* BenchmarkPayloads.swift in Sources */,
+				A10000D3 /* BenchmarkFixtures.swift in Sources */,
+			);
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+/* End PBXSourcesBuildPhase section */
+
+/* Begin PBXTargetDependency section */
+		A5000002 /* PBXTargetDependency */ = {
+			isa = PBXTargetDependency;
+			target = A4000001 /* PaddleOCRDemo */;
+			targetProxy = A5000001 /* PBXContainerItemProxy */;
+		};
+/* End PBXTargetDependency section */
+
+/* Begin XCBuildConfiguration section */
+		AA000001 /* Debug */ = {
+			isa = XCBuildConfiguration;
+			buildSettings = {
+				ALWAYS_SEARCH_USER_PATHS = NO;
+				ASSETCATALOG_COMPILER_GENERATE_SWIFT_ASSET_SYMBOL_EXTENSIONS = YES;
+				CLANG_ANALYZER_NONNULL = YES;
+				CLANG_ANALYZER_NUMBER_OBJECT_CONVERSION = YES_AGGRESSIVE;
+				CLANG_CXX_LANGUAGE_STANDARD = "gnu++20";
+				CLANG_ENABLE_MODULES = YES;
+				CLANG_ENABLE_OBJC_ARC = YES;
+				CLANG_ENABLE_OBJC_WEAK = YES;
+				CLANG_WARN_BLOCK_CAPTURE_AUTORELEASING = YES;
+				CLANG_WARN_BOOL_CONVERSION = YES;
+				CLANG_WARN_COMMA = YES;
+				CLANG_WARN_CONSTANT_CONVERSION = YES;
+				CLANG_WARN_DEPRECATED_OBJC_IMPLEMENTATIONS = YES;
+				CLANG_WARN_DIRECT_OBJC_ISA_USAGE = YES_ERROR;
+				CLANG_WARN_DOCUMENTATION_COMMENTS = YES;
+				CLANG_WARN_EMPTY_BODY = YES;
+				CLANG_WARN_ENUM_CONVERSION = YES;
+				CLANG_WARN_INFINITE_RECURSION = YES;
+				CLANG_WARN_INT_CONVERSION = YES;
+				CLANG_WARN_NON_LITERAL_NULL_CONVERSION = YES;
+				CLANG_WARN_OBJC_IMPLICIT_RETAIN_SELF = YES;
+				CLANG_WARN_OBJC_LITERAL_CONVERSION = YES;
+				CLANG_WARN_OBJC_ROOT_CLASS = YES_ERROR;
+				CLANG_WARN_QUOTED_INCLUDE_IN_FRAMEWORK_HEADER = YES;
+				CLANG_WARN_RANGE_LOOP_ANALYSIS = YES;
+				CLANG_WARN_STRICT_PROTOTYPES = YES;
+				CLANG_WARN_SUSPICIOUS_MOVE = YES;
+				CLANG_WARN_UNGUARDED_AVAILABILITY = YES_AGGRESSIVE;
+				CLANG_WARN_UNREACHABLE_CODE = YES;
+				CLANG_WARN__DUPLICATE_METHOD_MATCH = YES;
+				COPY_PHASE_STRIP = NO;
+				DEBUG_INFORMATION_FORMAT = dwarf;
+				ENABLE_STRICT_OBJC_MSGSEND = YES;
+				ENABLE_TESTABILITY = YES;
+				ENABLE_USER_SCRIPT_SANDBOXING = NO;
+				GCC_C_LANGUAGE_STANDARD = gnu17;
+				GCC_DYNAMIC_NO_PIC = NO;
+				GCC_NO_COMMON_BLOCKS = YES;
+				GCC_OPTIMIZATION_LEVEL = 0;
+				GCC_PREPROCESSOR_DEFINITIONS = (
+					"DEBUG=1",
+					"$(inherited)",
+				);
+				GCC_WARN_64_TO_32_BIT_CONVERSION = YES;
+				GCC_WARN_ABOUT_RETURN_TYPE = YES_ERROR;
+				GCC_WARN_UNDECLARED_SELECTOR = YES;
+				GCC_WARN_UNINITIALIZED_AUTOS = YES_AGGRESSIVE;
+				GCC_WARN_UNUSED_FUNCTION = YES;
+				GCC_WARN_UNUSED_VARIABLE = YES;
+				IPHONEOS_DEPLOYMENT_TARGET = 16.0;
+				MTL_ENABLE_DEBUG_INFO = INCLUDE_SOURCE;
+				MTL_FAST_MATH = YES;
+				ONLY_ACTIVE_ARCH = YES;
+				SDKROOT = iphoneos;
+				STRING_CATALOG_GENERATE_SYMBOLS = YES;
+				SWIFT_ACTIVE_COMPILATION_CONDITIONS = "$(inherited) DEBUG";
+				SWIFT_OPTIMIZATION_LEVEL = "-Onone";
+			};
+			name = Debug;
+		};
+		AA000002 /* Release */ = {
+			isa = XCBuildConfiguration;
+			buildSettings = {
+				ALWAYS_SEARCH_USER_PATHS = NO;
+				ASSETCATALOG_COMPILER_GENERATE_SWIFT_ASSET_SYMBOL_EXTENSIONS = YES;
+				CLANG_ANALYZER_NONNULL = YES;
+				CLANG_ANALYZER_NUMBER_OBJECT_CONVERSION = YES_AGGRESSIVE;
+				CLANG_CXX_LANGUAGE_STANDARD = "gnu++20";
+				CLANG_ENABLE_MODULES = YES;
+				CLANG_ENABLE_OBJC_ARC = YES;
+				CLANG_ENABLE_OBJC_WEAK = YES;
+				CLANG_WARN_BLOCK_CAPTURE_AUTORELEASING = YES;
+				CLANG_WARN_BOOL_CONVERSION = YES;
+				CLANG_WARN_COMMA = YES;
+				CLANG_WARN_CONSTANT_CONVERSION = YES;
+				CLANG_WARN_DEPRECATED_OBJC_IMPLEMENTATIONS = YES;
+				CLANG_WARN_DIRECT_OBJC_ISA_USAGE = YES_ERROR;
+				CLANG_WARN_DOCUMENTATION_COMMENTS = YES;
+				CLANG_WARN_EMPTY_BODY = YES;
+				CLANG_WARN_ENUM_CONVERSION = YES;
+				CLANG_WARN_INFINITE_RECURSION = YES;
+				CLANG_WARN_INT_CONVERSION = YES;
+				CLANG_WARN_NON_LITERAL_NULL_CONVERSION = YES;
+				CLANG_WARN_OBJC_IMPLICIT_RETAIN_SELF = YES;
+				CLANG_WARN_OBJC_LITERAL_CONVERSION = YES;
+				CLANG_WARN_OBJC_ROOT_CLASS = YES_ERROR;
+				CLANG_WARN_QUOTED_INCLUDE_IN_FRAMEWORK_HEADER = YES;
+				CLANG_WARN_RANGE_LOOP_ANALYSIS = YES;
+				CLANG_WARN_STRICT_PROTOTYPES = YES;
+				CLANG_WARN_SUSPICIOUS_MOVE = YES;
+				CLANG_WARN_UNGUARDED_AVAILABILITY = YES_AGGRESSIVE;
+				CLANG_WARN_UNREACHABLE_CODE = YES;
+				CLANG_WARN__DUPLICATE_METHOD_MATCH = YES;
+				COPY_PHASE_STRIP = NO;
+				DEBUG_INFORMATION_FORMAT = "dwarf-with-dsym";
+				ENABLE_NS_ASSERTIONS = NO;
+				ENABLE_STRICT_OBJC_MSGSEND = YES;
+				ENABLE_USER_SCRIPT_SANDBOXING = NO;
+				GCC_C_LANGUAGE_STANDARD = gnu17;
+				GCC_NO_COMMON_BLOCKS = YES;
+				GCC_WARN_64_TO_32_BIT_CONVERSION = YES;
+				GCC_WARN_ABOUT_RETURN_TYPE = YES_ERROR;
+				GCC_WARN_UNDECLARED_SELECTOR = YES;
+				GCC_WARN_UNINITIALIZED_AUTOS = YES_AGGRESSIVE;
+				GCC_WARN_UNUSED_FUNCTION = YES;
+				GCC_WARN_UNUSED_VARIABLE = YES;
+				IPHONEOS_DEPLOYMENT_TARGET = 16.0;
+				MTL_ENABLE_DEBUG_INFO = NO;
+				MTL_FAST_MATH = YES;
+				SDKROOT = iphoneos;
+				STRING_CATALOG_GENERATE_SYMBOLS = YES;
+				SWIFT_COMPILATION_MODE = wholemodule;
+				VALIDATE_PRODUCT = YES;
+			};
+			name = Release;
+		};
+		AA000003 /* Debug */ = {
+			isa = XCBuildConfiguration;
+			baseConfigurationReference = 520BC3C496ED955BBB302DC4 /* Pods-PaddleOCRDemo.debug.xcconfig */;
+			buildSettings = {
+				ASSETCATALOG_COMPILER_APPICON_NAME = AppIcon;
+				ASSETCATALOG_COMPILER_GLOBAL_ACCENT_COLOR_NAME = AccentColor;
+				CODE_SIGN_STYLE = Automatic;
+				CURRENT_PROJECT_VERSION = 1;
+				DEVELOPMENT_TEAM = 76MLN428C9;
+				GENERATE_INFOPLIST_FILE = YES;
+				INFOPLIST_FILE = PaddleOCRDemo/Supporting/Info.plist;
+				INFOPLIST_KEY_CFBundleDisplayName = "PaddleOCR";
+				INFOPLIST_KEY_UIApplicationSceneManifest_Generation = YES;
+				INFOPLIST_KEY_UIApplicationSupportsIndirectInputEvents = YES;
+				INFOPLIST_KEY_UILaunchScreen_Generation = YES;
+				INFOPLIST_KEY_UISupportedInterfaceOrientations = UIInterfaceOrientationPortrait;
+				LD_RUNPATH_SEARCH_PATHS = (
+					"$(inherited)",
+					"@executable_path/Frameworks",
+				);
+				MARKETING_VERSION = 1.0;
+				OTHER_LDFLAGS = (
+					"$(inherited)",
+					"-Wl,-no_warn_duplicate_libraries",
+				);
+				PRODUCT_BUNDLE_IDENTIFIER = com.paddleocr.demo;
+				PRODUCT_NAME = "$(TARGET_NAME)";
+				SUPPORTED_PLATFORMS = "iphoneos iphonesimulator";
+				SUPPORTS_MACCATALYST = NO;
+				SUPPORTS_MAC_DESIGNED_FOR_IPHONE_IPAD = NO;
+				SUPPORTS_XR_DESIGNED_FOR_IPHONE_IPAD = NO;
+				SWIFT_EMIT_LOC_STRINGS = YES;
+				SWIFT_OBJC_BRIDGING_HEADER = "PaddleOCRDemo/Supporting/PaddleOCRDemo-Bridging-Header.h";
+				SWIFT_VERSION = 5.0;
+				TARGETED_DEVICE_FAMILY = 1;
+				USER_HEADER_SEARCH_PATHS = (
+					"$(inherited)",
+					"$(SRCROOT)/PaddleOCRDemo/Engine",
+					"$(SRCROOT)/PaddleOCRDemo/ThirdParty/Clipper1",
+					"$(PODS_ROOT)/onnxruntime-objc/objectivec",
+					"$(PODS_ROOT)/onnxruntime-objc/objectivec/include",
+				);
+			};
+			name = Debug;
+		};
+		AA000004 /* Release */ = {
+			isa = XCBuildConfiguration;
+			baseConfigurationReference = 54F64B7C830508805D832F8A /* Pods-PaddleOCRDemo.release.xcconfig */;
+			buildSettings = {
+				ASSETCATALOG_COMPILER_APPICON_NAME = AppIcon;
+				ASSETCATALOG_COMPILER_GLOBAL_ACCENT_COLOR_NAME = AccentColor;
+				CODE_SIGN_STYLE = Automatic;
+				CURRENT_PROJECT_VERSION = 1;
+				DEVELOPMENT_TEAM = 76MLN428C9;
+				ENABLE_TESTABILITY = YES;
+				GENERATE_INFOPLIST_FILE = YES;
+				INFOPLIST_FILE = PaddleOCRDemo/Supporting/Info.plist;
+				INFOPLIST_KEY_CFBundleDisplayName = "PaddleOCR";
+				INFOPLIST_KEY_UIApplicationSceneManifest_Generation = YES;
+				INFOPLIST_KEY_UIApplicationSupportsIndirectInputEvents = YES;
+				INFOPLIST_KEY_UILaunchScreen_Generation = YES;
+				INFOPLIST_KEY_UISupportedInterfaceOrientations = UIInterfaceOrientationPortrait;
+				LD_RUNPATH_SEARCH_PATHS = (
+					"$(inherited)",
+					"@executable_path/Frameworks",
+				);
+				MARKETING_VERSION = 1.0;
+				OTHER_LDFLAGS = (
+					"$(inherited)",
+					"-Wl,-no_warn_duplicate_libraries",
+				);
+				PRODUCT_BUNDLE_IDENTIFIER = com.paddleocr.demo;
+				PRODUCT_NAME = "$(TARGET_NAME)";
+				SUPPORTED_PLATFORMS = "iphoneos iphonesimulator";
+				SUPPORTS_MACCATALYST = NO;
+				SUPPORTS_MAC_DESIGNED_FOR_IPHONE_IPAD = NO;
+				SUPPORTS_XR_DESIGNED_FOR_IPHONE_IPAD = NO;
+				SWIFT_EMIT_LOC_STRINGS = YES;
+				SWIFT_OBJC_BRIDGING_HEADER = "PaddleOCRDemo/Supporting/PaddleOCRDemo-Bridging-Header.h";
+				SWIFT_VERSION = 5.0;
+				TARGETED_DEVICE_FAMILY = 1;
+				USER_HEADER_SEARCH_PATHS = (
+					"$(inherited)",
+					"$(SRCROOT)/PaddleOCRDemo/Engine",
+					"$(SRCROOT)/PaddleOCRDemo/ThirdParty/Clipper1",
+					"$(PODS_ROOT)/onnxruntime-objc/objectivec",
+					"$(PODS_ROOT)/onnxruntime-objc/objectivec/include",
+				);
+			};
+			name = Release;
+		};
+		AA000005 /* Debug */ = {
+			isa = XCBuildConfiguration;
+			baseConfigurationReference = 72FF5C269BF861629F89D6B5 /* Pods-PaddleOCRDemoTests.debug.xcconfig */;
+			buildSettings = {
+				BUNDLE_LOADER = "$(TEST_HOST)";
+				CODE_SIGN_STYLE = Automatic;
+				CURRENT_PROJECT_VERSION = 1;
+				DEVELOPMENT_TEAM = 76MLN428C9;
+				GENERATE_INFOPLIST_FILE = YES;
+				IPHONEOS_DEPLOYMENT_TARGET = 16.0;
+				MARKETING_VERSION = 1.0;
+				PRODUCT_BUNDLE_IDENTIFIER = com.paddleocr.demo.tests;
+				PRODUCT_NAME = "$(TARGET_NAME)";
+				SWIFT_EMIT_LOC_STRINGS = NO;
+				SWIFT_VERSION = 5.0;
+				TARGETED_DEVICE_FAMILY = "1,2";
+				TEST_HOST = "$(BUILT_PRODUCTS_DIR)/PaddleOCRDemo.app/$(BUNDLE_EXECUTABLE_FOLDER_PATH)/PaddleOCRDemo";
+			};
+			name = Debug;
+		};
+		AA000006 /* Release */ = {
+			isa = XCBuildConfiguration;
+			baseConfigurationReference = 49C7CEE60686CC98BE2EEE45 /* Pods-PaddleOCRDemoTests.release.xcconfig */;
+			buildSettings = {
+				BUNDLE_LOADER = "$(TEST_HOST)";
+				CODE_SIGN_STYLE = Automatic;
+				CURRENT_PROJECT_VERSION = 1;
+				DEVELOPMENT_TEAM = 76MLN428C9;
+				GENERATE_INFOPLIST_FILE = YES;
+				IPHONEOS_DEPLOYMENT_TARGET = 16.0;
+				MARKETING_VERSION = 1.0;
+				PRODUCT_BUNDLE_IDENTIFIER = com.paddleocr.demo.tests;
+				PRODUCT_NAME = "$(TARGET_NAME)";
+				SWIFT_EMIT_LOC_STRINGS = NO;
+				SWIFT_VERSION = 5.0;
+				TARGETED_DEVICE_FAMILY = "1,2";
+				TEST_HOST = "$(BUILT_PRODUCTS_DIR)/PaddleOCRDemo.app/$(BUNDLE_EXECUTABLE_FOLDER_PATH)/PaddleOCRDemo";
+			};
+			name = Release;
+		};
+/* End XCBuildConfiguration section */
+
+/* Begin XCConfigurationList section */
+		A9000001 /* Build configuration list for PBXProject "PaddleOCRDemo" */ = {
+			isa = XCConfigurationList;
+			buildConfigurations = (
+				AA000001 /* Debug */,
+				AA000002 /* Release */,
+			);
+			defaultConfigurationIsVisible = 0;
+			defaultConfigurationName = Release;
+		};
+		A9000003 /* Build configuration list for PBXNativeTarget "PaddleOCRDemo" */ = {
+			isa = XCConfigurationList;
+			buildConfigurations = (
+				AA000003 /* Debug */,
+				AA000004 /* Release */,
+			);
+			defaultConfigurationIsVisible = 0;
+			defaultConfigurationName = Release;
+		};
+		A9000006 /* Build configuration list for PBXNativeTarget "PaddleOCRDemoTests" */ = {
+			isa = XCConfigurationList;
+			buildConfigurations = (
+				AA000005 /* Debug */,
+				AA000006 /* Release */,
+			);
+			defaultConfigurationIsVisible = 0;
+			defaultConfigurationName = Release;
+		};
+/* End XCConfigurationList section */
+	};
+	rootObject = A3000001 /* Project object */;
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo.xcodeproj/xcshareddata/xcschemes/PaddleOCRDemo.xcscheme b/deploy/ios_demo/PaddleOCRDemo.xcodeproj/xcshareddata/xcschemes/PaddleOCRDemo.xcscheme
new file mode 100644
index 0000000..7e16e01
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo.xcodeproj/xcshareddata/xcschemes/PaddleOCRDemo.xcscheme
@@ -0,0 +1,92 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<Scheme
+   LastUpgradeVersion = "2640"
+   version = "1.8">
+   <BuildAction
+      parallelizeBuildables = "YES"
+      buildImplicitDependencies = "YES">
+      <BuildActionEntries>
+         <BuildActionEntry
+            buildForTesting = "YES"
+            buildForRunning = "YES"
+            buildForProfiling = "YES"
+            buildForArchiving = "YES"
+            buildForAnalyzing = "YES">
+            <BuildableReference
+               BuildableIdentifier = "primary"
+               BlueprintIdentifier = "A4000001"
+               BuildableName = "PaddleOCRDemo.app"
+               BlueprintName = "PaddleOCRDemo"
+               ReferencedContainer = "container:PaddleOCRDemo.xcodeproj">
+            </BuildableReference>
+         </BuildActionEntry>
+      </BuildActionEntries>
+   </BuildAction>
+   <TestAction
+      buildConfiguration = "Debug"
+      selectedDebuggerIdentifier = "Xcode.DebuggerFoundation.Debugger.LLDB"
+      selectedLauncherIdentifier = "Xcode.DebuggerFoundation.Launcher.LLDB"
+      shouldUseLaunchSchemeArgsEnv = "YES"
+      shouldAutocreateTestPlan = "YES">
+      <Testables>
+         <TestableReference
+            skipped = "NO"
+            parallelizable = "YES">
+            <BuildableReference
+               BuildableIdentifier = "primary"
+               BlueprintIdentifier = "A4000002"
+               BuildableName = "PaddleOCRDemoTests.xctest"
+               BlueprintName = "PaddleOCRDemoTests"
+               ReferencedContainer = "container:PaddleOCRDemo.xcodeproj">
+            </BuildableReference>
+         </TestableReference>
+      </Testables>
+   </TestAction>
+   <LaunchAction
+      buildConfiguration = "Release"
+      selectedDebuggerIdentifier = ""
+      selectedLauncherIdentifier = "Xcode.IDEFoundation.Launcher.PosixSpawn"
+      launchStyle = "0"
+      useCustomWorkingDirectory = "NO"
+      ignoresPersistentStateOnLaunch = "NO"
+      debugDocumentVersioning = "YES"
+      debugServiceExtension = "internal"
+      allowLocationSimulation = "YES">
+      <BuildableProductRunnable
+         reuseRunningArtifactsWhenBuildingTests = "YES"
+         runnableDebuggingMode = "0">
+         <BuildableReference
+            BuildableIdentifier = "primary"
+            BlueprintIdentifier = "A4000001"
+            BuildableName = "PaddleOCRDemo.app"
+            BlueprintName = "PaddleOCRDemo"
+            ReferencedContainer = "container:PaddleOCRDemo.xcodeproj">
+         </BuildableReference>
+      </BuildableProductRunnable>
+   </LaunchAction>
+   <ProfileAction
+      buildConfiguration = "Release"
+      shouldUseLaunchSchemeArgsEnv = "YES"
+      savedToolIdentifier = ""
+      useCustomWorkingDirectory = "NO"
+      debugDocumentVersioning = "YES">
+      <BuildableProductRunnable
+         reuseRunningArtifactsWhenBuildingTests = "YES"
+         runnableDebuggingMode = "0">
+         <BuildableReference
+            BuildableIdentifier = "primary"
+            BlueprintIdentifier = "A4000001"
+            BuildableName = "PaddleOCRDemo.app"
+            BlueprintName = "PaddleOCRDemo"
+            ReferencedContainer = "container:PaddleOCRDemo.xcodeproj">
+         </BuildableReference>
+      </BuildableProductRunnable>
+   </ProfileAction>
+   <AnalyzeAction
+      buildConfiguration = "Debug">
+   </AnalyzeAction>
+   <ArchiveAction
+      buildConfiguration = "Release"
+      revealArchiveInOrganizer = "YES">
+   </ArchiveAction>
+</Scheme>
diff --git a/deploy/ios_demo/PaddleOCRDemo.xcworkspace/contents.xcworkspacedata b/deploy/ios_demo/PaddleOCRDemo.xcworkspace/contents.xcworkspacedata
new file mode 100644
index 0000000..09f53bd
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo.xcworkspace/contents.xcworkspacedata
@@ -0,0 +1,10 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<Workspace
+   version = "1.0">
+   <FileRef
+      location = "group:PaddleOCRDemo.xcodeproj">
+   </FileRef>
+   <FileRef
+      location = "group:Pods/Pods.xcodeproj">
+   </FileRef>
+</Workspace>
diff --git a/deploy/ios_demo/PaddleOCRDemo/App/Bundle+BundledImage.swift b/deploy/ios_demo/PaddleOCRDemo/App/Bundle+BundledImage.swift
new file mode 100644
index 0000000..e9d9004
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/App/Bundle+BundledImage.swift
@@ -0,0 +1,48 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import Foundation
+
+extension Bundle {
+    /// Raster extensions tried when resolving a bundled image by stem, or after an explicit extension misses.
+    static let bundledRasterImageExtensions: [String] = [
+        "jpg", "jpeg", "png", "heic", "heif", "webp", "tiff", "tif", "bmp", "gif",
+    ]
+
+    /// Resolves a filesystem path for an image under `subdirectory`.
+    ///
+    /// `name` may be a resource stem (e.g. `photo`) or include a suffix (e.g. `photo.jpg`).
+    /// Extensions are matched case-insensitively when provided; otherwise each entry in
+    /// ``bundledRasterImageExtensions`` is tried in order.
+    func path(forBundledImageNamed name: String, subdirectory: String) -> String? {
+        let trimmed = name.trimmingCharacters(in: .whitespacesAndNewlines)
+        guard !trimmed.isEmpty else { return nil }
+        let ns = trimmed as NSString
+        let extGiven = ns.pathExtension.lowercased()
+        let stem = ns.deletingPathExtension
+        let resourceStem = stem.isEmpty ? trimmed : stem
+
+        if !extGiven.isEmpty {
+            if let path = path(forResource: resourceStem, ofType: extGiven, inDirectory: subdirectory) {
+                return path
+            }
+        }
+        for ext in Self.bundledRasterImageExtensions where ext != extGiven {
+            if let path = path(forResource: resourceStem, ofType: ext, inDirectory: subdirectory) {
+                return path
+            }
+        }
+        return nil
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/App/ContentView.swift b/deploy/ios_demo/PaddleOCRDemo/App/ContentView.swift
new file mode 100644
index 0000000..3ecd61b
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/App/ContentView.swift
@@ -0,0 +1,321 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import PhotosUI
+import SwiftUI
+
+/// Root view: routes `AppState` and lays out **input (photo/sample)** vs **output (preview + text)**; engine tuning lives in ``EngineSettingsSheet``.
+struct ContentView: View {
+    @Environment(\.horizontalSizeClass) private var horizontalSizeClass
+    @StateObject private var viewModel = OCRViewModel()
+    @State private var selectedItem: PhotosPickerItem?
+    @State private var showEngineSettings = false
+
+    private var navigationTitleDisplayMode: NavigationBarItem.TitleDisplayMode {
+        switch viewModel.state {
+        case .ready: return .large
+        case .error(let err) where !err.isModelError: return .large
+        default: return .inline
+        }
+    }
+
+    private var showsAdvancedToolbarItem: Bool {
+        switch viewModel.state {
+        case .loadingModels, .processing: return false
+        case .ready, .results, .error: return true
+        }
+    }
+
+    var body: some View {
+        NavigationStack {
+            ScrollView {
+                VStack(spacing: 0) {
+                    contentForState
+                }
+                .padding(.horizontal, 20)
+                .padding(.vertical, 20)
+            }
+            .background(Color(.systemGroupedBackground))
+            .navigationTitle("PaddleOCR Demo")
+            .navigationBarTitleDisplayMode(navigationTitleDisplayMode)
+            .toolbar {
+                if showsAdvancedToolbarItem {
+                    ToolbarItem(placement: .topBarTrailing) {
+                        Button {
+                            showEngineSettings = true
+                        } label: {
+                            Image(systemName: "slider.horizontal.3")
+                        }
+                        .accessibilityLabel("Advanced settings")
+                    }
+                }
+            }
+        }
+        .sheet(isPresented: $showEngineSettings) {
+            EngineSettingsSheet(viewModel: viewModel)
+        }
+        .task {
+            await viewModel.loadModels()
+        }
+        .onChange(of: selectedItem) { newItem in
+            guard let item = newItem else { return }
+            Task {
+                guard let data = try? await item.loadTransferable(type: Data.self),
+                      let uiImage = UIImage(data: data) else {
+                    viewModel.reportImageLoadFailed()
+                    return
+                }
+                selectedItem = nil
+                await viewModel.processImage(uiImage)
+            }
+        }
+    }
+
+    private func imageInputSection(title: String, subtitle: String) -> some View {
+        VStack(alignment: .leading, spacing: 10) {
+            DemoSectionHeader(title: title, subtitle: subtitle)
+            ImagePickerSection(
+                selectedItem: $selectedItem,
+                sampleImageNames: viewModel.sampleImageNames,
+                onSampleSelected: { name in
+                    Task { await viewModel.selectSampleImage(named: name) }
+                }
+            )
+        }
+    }
+
+    // MARK: - State Router
+
+    @ViewBuilder
+    private var contentForState: some View {
+        switch viewModel.state {
+        case .loadingModels:
+            loadingModelsView
+        case .ready:
+            readyView
+        case .processing(let image):
+            processingView(image: image)
+        case .results(let result, let image, _):
+            resultsView(result: result, image: image)
+        case .error(let error):
+            errorView(error: error)
+        }
+    }
+
+    // MARK: - Loading Models
+
+    private var loadingModelsView: some View {
+        VStack(spacing: 16) {
+            ProgressView()
+                .scaleEffect(1.3)
+                .tint(.accentColor)
+            Text("Loading models")
+                .font(.title3.weight(.semibold))
+            Text("Preparing detection and recognition…")
+                .font(.subheadline)
+                .foregroundStyle(.secondary)
+                .multilineTextAlignment(.center)
+        }
+        .frame(maxWidth: .infinity)
+        .padding(.top, 72)
+        .padding(.bottom, 40)
+    }
+
+    // MARK: - Ready
+
+    private var readyView: some View {
+        VStack(spacing: 22) {
+            statusPill(icon: "checkmark.circle.fill", tint: .green) {
+                VStack(alignment: .leading, spacing: 4) {
+                    Text("Ready")
+                        .font(.title3.weight(.semibold))
+                    Text("Choose a photo or open a sample. Engine and OCR tuning are under Advanced in the toolbar.")
+                        .font(.subheadline)
+                        .foregroundStyle(.secondary)
+                        .fixedSize(horizontal: false, vertical: true)
+                }
+            }
+
+            imageInputSection(
+                title: "Image",
+                subtitle: "Choose from your library or a bundled sample."
+            )
+        }
+    }
+
+    // MARK: - Processing
+
+    private func processingView(image: UIImage) -> some View {
+        VStack(alignment: .leading, spacing: 16) {
+            Text("Running OCR")
+                .font(.title3.weight(.semibold))
+            Text("Hold on — detection and recognition are in progress.")
+                .font(.subheadline)
+                .foregroundStyle(.secondary)
+
+            ZStack {
+                Image(uiImage: image)
+                    .resizable()
+                    .aspectRatio(contentMode: .fit)
+                    .frame(maxHeight: 360)
+                    .clipShape(RoundedRectangle(cornerRadius: 14, style: .continuous))
+
+                RoundedRectangle(cornerRadius: 14, style: .continuous)
+                    .fill(.ultraThinMaterial)
+
+                VStack(spacing: 10) {
+                    ProgressView()
+                        .scaleEffect(1.25)
+                        .tint(.accentColor)
+                    Text("Working…")
+                        .font(.headline)
+                }
+            }
+            .frame(maxWidth: .infinity)
+        }
+    }
+
+    // MARK: - Results
+
+    private func resultsView(result: OCRRunResult, image: UIImage) -> some View {
+        // Only use two columns when explicitly regular (e.g. iPad); treat nil as compact.
+        let isCompact = horizontalSizeClass != .regular
+        return VStack(spacing: 22) {
+            if isCompact {
+                previewBlock(result: result, image: image)
+                TimingView(result: result)
+                    .padding(.vertical, 4)
+            } else {
+                TimingView(result: result)
+                    .padding(.vertical, 4)
+                HStack(alignment: .top, spacing: 16) {
+                    previewBlock(result: result, image: image)
+                        .frame(maxWidth: .infinity)
+                        .layoutPriority(1)
+                    resultsListBlock(result: result)
+                        .frame(minWidth: 280, maxWidth: .infinity)
+                }
+            }
+
+            if isCompact {
+                resultsListBlock(result: result)
+            }
+
+            imageInputSection(
+                title: "New image",
+                subtitle: "Choose another photo or sample. Re-run uses the current image and OCR parameters."
+            )
+
+            rerunOCRButton()
+        }
+    }
+
+    @ViewBuilder
+    private func rerunOCRButton() -> some View {
+        if viewModel.hasPendingOcrChanges {
+            Button {
+                Task { await viewModel.rerunOCR() }
+            } label: {
+                Label("Re-run OCR on this image", systemImage: "arrow.clockwise.circle.fill")
+                    .frame(maxWidth: .infinity)
+                    .padding(.vertical, 4)
+            }
+            .buttonStyle(.borderedProminent)
+            .controlSize(.large)
+        } else {
+            Button {
+                Task { await viewModel.rerunOCR() }
+            } label: {
+                Label("Re-run OCR on this image", systemImage: "arrow.clockwise.circle.fill")
+                    .frame(maxWidth: .infinity)
+                    .padding(.vertical, 4)
+            }
+            .buttonStyle(.bordered)
+            .controlSize(.large)
+        }
+    }
+
+    private func previewBlock(result: OCRRunResult, image: UIImage) -> some View {
+        VStack(alignment: .leading, spacing: 10) {
+            DemoSectionHeader(
+                title: "Preview",
+                subtitle: "Accent overlay shows each detected text region."
+            )
+            DemoCard {
+                ResultImageView(image: image, results: result.results)
+                    .clipShape(RoundedRectangle(cornerRadius: 10, style: .continuous))
+            }
+        }
+    }
+
+    private func resultsListBlock(result: OCRRunResult) -> some View {
+        VStack(alignment: .leading, spacing: 10) {
+            DemoSectionHeader(title: "Results", subtitle: nil)
+            DemoCard {
+                ResultsListView(
+                    results: result.results,
+                    copiedFeedback: viewModel.copiedFeedback,
+                    onCopy: { viewModel.copyResultsToClipboard() }
+                )
+            }
+        }
+    }
+
+    // MARK: - Error
+
+    private func errorView(error: AppError) -> some View {
+        VStack(spacing: 22) {
+            ErrorView(error: error, onRetry: {
+                Task { await viewModel.retry() }
+            })
+
+            if !error.isModelError {
+                VStack(alignment: .leading, spacing: 10) {
+                    DemoSectionHeader(title: "Try another image", subtitle: nil)
+                    ImagePickerSection(
+                        selectedItem: $selectedItem,
+                        sampleImageNames: viewModel.sampleImageNames,
+                        onSampleSelected: { name in
+                            Task { await viewModel.selectSampleImage(named: name) }
+                        }
+                    )
+                }
+            }
+        }
+        .padding(.top, 24)
+    }
+
+    // MARK: - Small chrome
+
+    private func statusPill<Content: View>(icon: String, tint: Color, @ViewBuilder content: () -> Content) -> some View {
+        HStack(alignment: .top, spacing: 14) {
+            Image(systemName: icon)
+                .font(.system(size: 28))
+                .foregroundStyle(tint)
+                .symbolRenderingMode(.hierarchical)
+            content()
+                .frame(maxWidth: .infinity, alignment: .leading)
+        }
+        .padding(16)
+        .frame(maxWidth: .infinity, alignment: .leading)
+        .background {
+            RoundedRectangle(cornerRadius: 14, style: .continuous)
+                .fill(Color(.secondarySystemGroupedBackground))
+        }
+        .overlay {
+            RoundedRectangle(cornerRadius: 14, style: .continuous)
+                .strokeBorder(Color(.separator).opacity(0.35), lineWidth: 0.5)
+        }
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/App/DemoAssets.swift b/deploy/ios_demo/PaddleOCRDemo/App/DemoAssets.swift
new file mode 100644
index 0000000..24865e6
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/App/DemoAssets.swift
@@ -0,0 +1,21 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import Foundation
+
+/// Built-in demo images (bundle resource stems).
+enum DemoAssets {
+    /// Default bundled sample for the image picker.
+    static let defaultSampleImageStem: String = "general_ocr_002"
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/App/DemoSectionChrome.swift b/deploy/ios_demo/PaddleOCRDemo/App/DemoSectionChrome.swift
new file mode 100644
index 0000000..52c7b0b
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/App/DemoSectionChrome.swift
@@ -0,0 +1,62 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import SwiftUI
+
+// MARK: - Section chrome (shared by ContentView + panels)
+
+/// Vertical section title + optional subtitle for scan-friendly hierarchy.
+struct DemoSectionHeader: View {
+    let title: String
+    var subtitle: String?
+
+    var body: some View {
+        VStack(alignment: .leading, spacing: 4) {
+            Text(title)
+                .font(.footnote.weight(.semibold))
+                .foregroundStyle(.secondary)
+                .textCase(.uppercase)
+                .tracking(0.4)
+            if let subtitle {
+                Text(subtitle)
+                    .font(.subheadline)
+                    .foregroundStyle(.secondary)
+            }
+        }
+        .frame(maxWidth: .infinity, alignment: .leading)
+    }
+}
+
+/// Grouped card matching system settings style.
+struct DemoCard<Content: View>: View {
+    let content: Content
+
+    init(@ViewBuilder content: () -> Content) {
+        self.content = content()
+    }
+
+    var body: some View {
+        content
+            .padding(16)
+            .frame(maxWidth: .infinity, alignment: .leading)
+            .background {
+                RoundedRectangle(cornerRadius: 14, style: .continuous)
+                    .fill(Color(.secondarySystemGroupedBackground))
+            }
+            .overlay {
+                RoundedRectangle(cornerRadius: 14, style: .continuous)
+                    .strokeBorder(Color(.separator).opacity(0.35), lineWidth: 0.5)
+            }
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/App/PaddleOCRDemoApp.swift b/deploy/ios_demo/PaddleOCRDemo/App/PaddleOCRDemoApp.swift
new file mode 100644
index 0000000..6175c92
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/App/PaddleOCRDemoApp.swift
@@ -0,0 +1,24 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import SwiftUI
+
+@main
+struct PaddleOCRDemoApp: App {
+    var body: some Scene {
+        WindowGroup {
+            ContentView()
+        }
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/BoxSorter.swift b/deploy/ios_demo/PaddleOCRDemo/Engine/BoxSorter.swift
new file mode 100644
index 0000000..3d7b45a
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/BoxSorter.swift
@@ -0,0 +1,72 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import Foundation
+
+// MARK: - BoxSorter
+
+/// Sorts detection boxes in natural reading order: top-to-bottom, then
+/// left-to-right within the same text line.
+///
+/// Two-pass: sort by top-left (y then x), then swap same-line neighbors when
+/// the y-gap is below `yThreshold` but x is out of order.
+struct BoxSorter {
+
+    /// Y-coordinate threshold for same-line detection. Boxes whose top-left y
+    /// coordinates differ by less than this value are considered on the same line
+    /// and sorted left-to-right. Default: 10 pixels.
+    static let yThreshold: Int32 = 10
+
+    /// Sort detection boxes in reading order: top-to-bottom, then left-to-right
+    /// within the same line (defined by yThreshold).
+    ///
+    /// Algorithm:
+    /// 1. Sort all boxes by top-left corner: first by y, then by x.
+    /// 2. Insertion-sort pass: for each box, walk backward and swap with the
+    ///    previous box if they are on the same line (y-difference < 10) and
+    ///    the current box's x is smaller.
+    ///
+    /// - Parameter boxes: Unordered detection boxes from DBPostProcessor.
+    /// - Returns: Boxes sorted in reading order.
+    static func sortInReadingOrder(_ boxes: [DetectionBox]) -> [DetectionBox] {
+        guard boxes.count > 1 else { return boxes }
+
+        // Step 1: Sort by top-left point -- first by y, then by x
+        var sorted = boxes.sorted { a, b in
+            let ay = a.points[0][1], ax = a.points[0][0]
+            let by = b.points[0][1], bx = b.points[0][0]
+            if ay != by { return ay < by }
+            return ax < bx
+        }
+
+        // Step 2: Backward insertion sort for same-line boxes
+        for i in 0..<(sorted.count - 1) {
+            var j = i
+            while j >= 0 {
+                let nextY = sorted[j + 1].points[0][1]
+                let currY = sorted[j].points[0][1]
+                let nextX = sorted[j + 1].points[0][0]
+                let currX = sorted[j].points[0][0]
+                if abs(nextY - currY) < yThreshold && nextX < currX {
+                    sorted.swapAt(j, j + 1)
+                    j -= 1
+                } else {
+                    break
+                }
+            }
+        }
+
+        return sorted
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/CTCDecoder.swift b/deploy/ios_demo/PaddleOCRDemo/Engine/CTCDecoder.swift
new file mode 100644
index 0000000..60b654e
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/CTCDecoder.swift
@@ -0,0 +1,199 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import Foundation
+
+// MARK: - Errors
+
+enum CTCDecoderError: LocalizedError {
+    case emptyDictionary
+    case invalidOutputShape([Int])
+
+    var errorDescription: String? {
+        switch self {
+        case .emptyDictionary:
+            return "Character dictionary is empty"
+        case .invalidOutputShape(let shape):
+            return "Invalid recognition output shape: \(shape), expected [1, T, C]"
+        }
+    }
+}
+
+// MARK: - Recognition Result
+
+/// A single text recognition result from CTC decoding.
+struct RecognitionResult {
+    /// The decoded text string.
+    let text: String
+    /// Average confidence across all decoded characters (0.0 to 1.0).
+    let confidence: Float
+}
+
+// MARK: - CTCDecoder
+
+/// CTC label decoding for recognition output
+///
+/// Greedy CTC decoding with blank index 0 and duplicate collapse.
+/// Per timestep: take the best class and its score, then merge duplicates and drop blanks,
+/// then average the kept scores. Logits need not be softmax probabilities;
+/// `max` over the class dimension is used for confidence on exported logits.
+///
+/// The CTC (Connectionist Temporal Classification) decoding algorithm:
+/// 1. Argmax across the class dimension to get per-timestep predicted indices
+/// 2. Max across the class dimension to get per-timestep prediction probabilities
+/// 3. Remove consecutive duplicate indices (CTC merge rule)
+/// 4. Remove blank tokens (index 0)
+/// 5. Map remaining indices to characters via the dictionary
+/// 6. Confidence = mean of probabilities at selected positions
+///
+struct CTCDecoder {
+    private let characters: [String]
+
+    /// The blank token index (always 0 for CTC).
+    private let blankIndex: Int = 0
+
+    /// Creates a CTCDecoder from a recognition model's InferenceConfig.
+    ///
+    /// Reads `character_dict` from the model config file and prepends the blank token.
+    ///
+    /// - Parameter config: A parsed InferenceConfig from the recognition model's config file.
+    /// - Throws: `CTCDecoderError.emptyDictionary` if no character dict is found.
+    init(config: InferenceConfig) throws {
+        guard let dict = config.postProcess.characterDict, !dict.isEmpty else {
+            throw CTCDecoderError.emptyDictionary
+        }
+
+        // Build character list: dict from config, trailing space (default-on), blank at index 0.
+        var chars = ["blank"]
+        chars.append(contentsOf: dict)
+        chars.append(" ")
+        self.characters = chars
+    }
+
+    /// Decodes the raw recognition model output into text with confidence.
+    ///
+    /// The output tensor has shape `[1, T, C]` where:
+    /// - T = number of timesteps (sequence length)
+    /// - C = number of classes (character vocabulary size + 1 for blank)
+    ///
+    /// - Parameters:
+    ///   - outputData: Flat float32 array of the model output.
+    ///   - outputShape: Shape of the output tensor, expected `[1, T, C]`.
+    /// - Returns: A `RecognitionResult` with decoded text and confidence score.
+    func decode(outputData: [Float], outputShape: [Int]) throws -> RecognitionResult {
+        guard outputShape.count == 3, outputShape[0] == 1 else {
+            throw CTCDecoderError.invalidOutputShape(outputShape)
+        }
+
+        let timesteps = outputShape[1]
+        let numClasses = outputShape[2]
+
+        // Step 1 & 2: Argmax and max across the class dimension for each timestep
+        var predIndices = [Int](repeating: 0, count: timesteps)
+        var predProbs = [Float](repeating: 0, count: timesteps)
+
+        for t in 0..<timesteps {
+            let offset = t * numClasses
+            var maxVal: Float = outputData[offset]
+            var maxIdx = 0
+
+            for c in 1..<numClasses {
+                let val = outputData[offset + c]
+                if val > maxVal {
+                    maxVal = val
+                    maxIdx = c
+                }
+            }
+
+            predIndices[t] = maxIdx
+            predProbs[t] = maxVal
+        }
+
+        // Step 3: Build selection mask
+        // - Remove consecutive duplicates (keep first of each run)
+        // - Remove blank tokens (index 0)
+        //
+        // Drop consecutive duplicate indices, then drop blank (0).
+        var selection = [Bool](repeating: true, count: timesteps)
+
+        // Remove consecutive duplicates
+        for t in 1..<timesteps {
+            if predIndices[t] == predIndices[t - 1] {
+                selection[t] = false
+            }
+        }
+
+        // Remove blank tokens
+        for t in 0..<timesteps {
+            if predIndices[t] == blankIndex {
+                selection[t] = false
+            }
+        }
+
+        // Step 4: Collect selected characters and their probabilities
+        var charList: [String] = []
+        var confList: [Float] = []
+
+        for t in 0..<timesteps {
+            if selection[t] {
+                let idx = predIndices[t]
+                if idx < characters.count {
+                    charList.append(characters[idx])
+                }
+                confList.append(predProbs[t])
+            }
+        }
+
+        // Step 5: Join characters into text
+        let text = charList.joined()
+
+        // Step 6: Compute mean confidence
+        // Mean probability over kept timesteps; 0 if none kept.
+        let confidence: Float
+        if confList.isEmpty {
+            confidence = 0
+        } else {
+            confidence = confList.reduce(0, +) / Float(confList.count)
+        }
+
+        return RecognitionResult(text: text, confidence: confidence)
+    }
+
+    /// Decodes a batched output tensor with shape `[N, T, C]`.
+    func decodeBatch(outputData: [Float], outputShape: [Int]) throws -> [RecognitionResult] {
+        guard outputShape.count == 3 else {
+            throw CTCDecoderError.invalidOutputShape(outputShape)
+        }
+        let batchCount = outputShape[0]
+        let timesteps = outputShape[1]
+        let numClasses = outputShape[2]
+        guard batchCount >= 1, timesteps >= 1, numClasses >= 1 else {
+            throw CTCDecoderError.invalidOutputShape(outputShape)
+        }
+        let rowElements = timesteps * numClasses
+        let expected = batchCount * rowElements
+        guard outputData.count == expected else {
+            throw CTCDecoderError.invalidOutputShape(outputShape)
+        }
+        var results: [RecognitionResult] = []
+        results.reserveCapacity(batchCount)
+        for b in 0..<batchCount {
+            let start = b * rowElements
+            let slice = Array(outputData[start..<(start + rowElements)])
+            let one = try decode(outputData: slice, outputShape: [1, timesteps, numClasses])
+            results.append(one)
+        }
+        return results
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/DBPostProcess.swift b/deploy/ios_demo/PaddleOCRDemo/Engine/DBPostProcess.swift
new file mode 100644
index 0000000..b0c8dfd
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/DBPostProcess.swift
@@ -0,0 +1,250 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import Foundation
+import CoreGraphics
+
+// MARK: - Detection Result
+
+/// A detected text region with its bounding quadrilateral and confidence score.
+struct DetectionBox {
+    /// Four corner points of the bounding quadrilateral, each as [x, y].
+    /// Order: top-left, top-right, bottom-right, bottom-left (`getMiniBoxes` / min-area rect).
+    let points: [[Int32]]
+
+    /// Confidence score from `boxScoreFast` (mean probability in the box region).
+    let score: Float
+}
+
+// MARK: - DBPostProcessor
+
+/// DB (Differentiable Binarization) text detection postprocessor.
+///
+/// Maps raw model output probability map to
+/// bounding polygons (threshold → contours → min-area quads → score filter → unclip → scale).
+///
+/// Steps: threshold → contours → min-area quads → box score → polygon offset (unclip) → scale to original size.
+///
+struct DBPostProcessor {
+
+    /// Binary threshold for the probability map (pixels above this are foreground).
+    let thresh: Float
+
+    /// Minimum box confidence score to keep a detection.
+    let boxThresh: Float
+
+    /// Maximum number of contours to evaluate.
+    let maxCandidates: Int
+
+    /// Expansion ratio for polygon offset (unclip step).
+    let unclipRatio: Float
+
+    /// Minimum side length of a bounding box to be kept (pixels in probability map space).
+    let minSize: Float = 3.0
+
+    /// Score mode: only `"fast"` is implemented (`"slow"` is not supported).
+    let scoreMode: String = "fast"
+
+    // MARK: - Initializers
+
+    /// Initialize with explicit parameters.
+    init(thresh: Float = 0.3, boxThresh: Float = 0.6, maxCandidates: Int = 1000, unclipRatio: Float = 1.5) {
+        self.thresh = thresh
+        self.boxThresh = boxThresh
+        self.maxCandidates = maxCandidates
+        self.unclipRatio = unclipRatio
+    }
+
+    /// Initialize from a parsed ``PostProcessConfig`.
+    init(config: PostProcessConfig) {
+        self.thresh = config.thresh
+        self.boxThresh = config.boxThresh
+        self.maxCandidates = config.maxCandidates
+        self.unclipRatio = config.unclipRatio
+    }
+
+    // MARK: - Public API
+
+    /// Process raw ORT detection output into bounding polygons.
+    ///
+    /// - Parameters:
+    ///   - outputTensor: Raw float output from ONNX Runtime, shape [1, 1, H, W].
+    ///   - tensorHeight: Height of the output tensor (H).
+    ///   - tensorWidth: Width of the output tensor (W).
+    ///   - originalWidth: Width of the original input image.
+    ///   - originalHeight: Height of the original input image.
+    /// - Returns: Array of detected text boxes with confidence scores.
+    func process(
+        outputTensor: [Float],
+        tensorHeight: Int,
+        tensorWidth: Int,
+        originalWidth: Int,
+        originalHeight: Int
+    ) -> [DetectionBox] {
+        // 1. Extract probability map: pred[0, 0, :, :] from shape [1, 1, H, W]
+        let mapSize = tensorHeight * tensorWidth
+        let pred: [Float]
+        if outputTensor.count >= mapSize {
+            pred = Array(outputTensor.prefix(mapSize))
+        } else {
+            return []
+        }
+
+        // 2. Binary threshold: create mask where prob > thresh
+        var binaryMask = [UInt8](repeating: 0, count: mapSize)
+        for i in 0..<mapSize {
+            binaryMask[i] = pred[i] > thresh ? 1 : 0
+        }
+
+        // 3. Find contours on the binary mask
+        let contours = findContours(
+            binaryMask: binaryMask,
+            width: tensorWidth,
+            height: tensorHeight
+        )
+
+        // 4. Process each contour
+        let numContours = min(contours.count, maxCandidates)
+        var results: [DetectionBox] = []
+
+        for i in 0..<numContours {
+            let contour = contours[i]
+
+            // 4a. Get minimum area bounding rectangle
+            guard let (miniBox, sside) = getMiniBoxes(contour: contour) else {
+                continue
+            }
+            if sside < minSize {
+                continue
+            }
+
+            // 4b. Score the box using fast mode
+            let points = miniBox.map { [$0.x, $0.y] }
+            let score = boxScoreFast(
+                pred: pred,
+                predWidth: tensorWidth,
+                predHeight: tensorHeight,
+                box: points
+            )
+            if boxThresh > score {
+                continue
+            }
+
+            // 4c. Expand polygon (unclip) via integer-grid offset.
+            let nsMini = miniBox.map { NSValue(cgPoint: CGPoint(x: CGFloat($0.x), y: CGFloat($0.y))) }
+            let area = Double(PDBOpenCVDBBridge.contourArea(fromPoints: nsMini))
+            let length = Double(PDBOpenCVDBBridge.arcLength(fromPoints: nsMini, closed: true))
+            guard length > 0 else { continue }
+            let distance = area * Double(unclipRatio) / length
+
+            let rings = PDBPolygonOffsetBridge.inflateClosedPolygon(with: nsMini, distance: distance) as [[NSValue]]
+            guard rings.count == 1 else {
+                continue
+            }
+            let firstRing = rings[0]
+
+            // 4d. Second getMiniBoxes on the expanded polygon
+            let expandedFloats = firstRing.map { FloatPoint(x: Float($0.cgPointValue.x), y: Float($0.cgPointValue.y)) }
+            guard let (finalBox, finalSside) = getMiniBoxes(contour: expandedFloats) else {
+                continue
+            }
+            if finalSside < minSize + 2 {
+                continue
+            }
+
+            // 4e. Scale to original image dimensions
+            let w = Float(tensorWidth)
+            let h = Float(tensorHeight)
+            let dw = Float(originalWidth)
+            let dh = Float(originalHeight)
+
+            var scaledPoints: [[Int32]] = []
+            for pt in finalBox {
+                let sx = Int32(min(max((pt.x / w * dw).rounded(), 0), dw))
+                let sy = Int32(min(max((pt.y / h * dh).rounded(), 0), dh))
+                scaledPoints.append([sx, sy])
+            }
+
+            results.append(DetectionBox(points: scaledPoints, score: score))
+        }
+
+        return results
+    }
+}
+
+// MARK: - Internal Point Type
+
+/// Float 2D point for `DBPostProcessor` geometry helpers.
+internal struct FloatPoint {
+    var x: Float
+    var y: Float
+}
+
+// MARK: - OpenCV (CocoaPods OpenCV 4.3.x)
+
+extension DBPostProcessor {
+
+    /// Foreground pixels as 255 on an 8-bit mask; runs `findContours` with `RETR_LIST` and `CHAIN_APPROX_SIMPLE`.
+    func findContours(binaryMask: [UInt8], width: Int, height: Int) -> [[FloatPoint]] {
+        let data = Data(binaryMask)
+        let raw = PDBOpenCVDBBridge.findContours(
+            binaryMask: data,
+            width: width,
+            height: height,
+            maxCandidates: maxCandidates
+        )
+        return raw.map { contour in
+            contour.map { v in
+                let p = v.cgPointValue
+                return FloatPoint(x: Float(p.x), y: Float(p.y))
+            }
+        }
+    }
+
+    /// Minimum-area rectangle: `minAreaRect` on contour points, `boxPoints` for corners, then TL/TR/BR/BL ordering.
+    func getMiniBoxes(contour: [FloatPoint]) -> (box: [FloatPoint], minSide: Float)? {
+        guard contour.count >= 2 else { return nil }
+        let values = contour.map { NSValue(cgPoint: CGPoint(x: CGFloat($0.x), y: CGFloat($0.y))) }
+        guard let result = PDBOpenCVDBBridge.miniBox(fromPoints: values) else { return nil }
+        let corners = result.corners.map { v in
+            let p = v.cgPointValue
+            return FloatPoint(x: Float(p.x), y: Float(p.y))
+        }
+        return (corners, result.minSide)
+    }
+}
+
+// MARK: - Box Scoring
+
+extension DBPostProcessor {
+
+    /// Compute the mean probability score within a box region of the probability map.
+    ///
+    /// Mean probability inside the quad: `fillPoly` mask on the score map, then `mean` over the masked region.
+    func boxScoreFast(
+        pred: [Float],
+        predWidth: Int,
+        predHeight: Int,
+        box: [[Float]]
+    ) -> Float {
+        let predData = pred.withUnsafeBufferPointer { Data(buffer: $0) }
+        let boxNS: [[NSNumber]] = box.map { [NSNumber(value: $0[0]), NSNumber(value: $0[1])] }
+        return PDBOpenCVImageBridge.meanPredInQuad(
+            predData,
+            width: predWidth,
+            height: predHeight,
+            box: boxNS
+        )
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/DetectionEngine.swift b/deploy/ios_demo/PaddleOCRDemo/Engine/DetectionEngine.swift
new file mode 100644
index 0000000..ad9b28c
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/DetectionEngine.swift
@@ -0,0 +1,168 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import CoreGraphics
+import Foundation
+
+// MARK: - Detection Result
+
+/// The result of running text detection on an image.
+///
+/// Contains the detected text boxes and per-stage timing metrics for performance analysis.
+struct DetectionResult {
+    /// Detected text region bounding polygons with confidence scores.
+    let boxes: [DetectionBox]
+    /// Seconds spent in preprocessing (resize, normalize, HWC-to-CHW).
+    let preprocessTime: TimeInterval
+    /// Seconds spent in ONNX Runtime inference.
+    let inferenceTime: TimeInterval
+    /// Seconds spent in DB postprocessing (threshold, contours, polygon offset / unclip).
+    let postprocessTime: TimeInterval
+    /// Total detection time (preprocess + inference + postprocess).
+    let totalTime: TimeInterval
+    /// Detection model input tensor shape, e.g. [1, 3, H, W].
+    let inputTensorShape: [Int]
+}
+
+// MARK: - Detection Engine Errors
+
+enum DetectionEngineError: LocalizedError {
+    case noOutputTensor
+    case unexpectedOutputShape([Int])
+
+    var errorDescription: String? {
+        switch self {
+        case .noOutputTensor:
+            return "Detection model produced no output tensor"
+        case .unexpectedOutputShape(let shape):
+            return "Unexpected detection output shape: \(shape), expected [1, 1, H, W]"
+        }
+    }
+}
+
+// MARK: - DetectionEngine
+
+/// Orchestrates text detection: CGImage → preprocess → ORT inference → postprocess.
+///
+/// This is the integration layer that composes `DetPreprocessor`, `ORTSessionManager`, and
+/// `DBPostProcessor` into a single callable unit.
+///
+/// Usage:
+/// ```swift
+/// let engine = try DetectionEngine(sessionManager: manager)
+/// let result = try await engine.detect(cgImage)
+/// print("Found \(result.boxes.count) text regions in \(result.totalTime)s")
+/// ```
+class DetectionEngine {
+    private let sessionManager: ORTSessionManager
+    private let preprocessor: DetPreprocessor
+    private(set) var modelConfig: InferenceConfig
+
+    /// Initialize with an existing ORTSessionManager (models must already be loaded).
+    ///
+    /// Loads the detection model config file from the bundle to configure
+    /// the preprocessor and postprocessor.
+    ///
+    /// - Parameter sessionManager: A loaded ORTSessionManager with detection model ready.
+    /// - Throws: If the model config cannot be loaded or required fields are missing.
+    init(sessionManager: ORTSessionManager) throws {
+        self.sessionManager = sessionManager
+
+        // Load detection model config
+        let detPaths = try ModelConfig.detection()
+        let cfg = try InferenceConfig.load(from: detPaths.configPath)
+        self.modelConfig = cfg
+        self.preprocessor = try DetPreprocessor(config: cfg)
+    }
+
+    private func makePostProcessor(effective: EffectiveOCRParams) -> DBPostProcessor {
+        DBPostProcessor(
+            thresh: effective.detThresh,
+            boxThresh: effective.detBoxThresh,
+            maxCandidates: modelConfig.postProcess.maxCandidates,
+            unclipRatio: effective.detUnclipRatio
+        )
+    }
+
+    /// Run detection on a CGImage, returning bounding polygons with confidence scores.
+    ///
+    /// Steps: DetResizeForTest → NormalizeImage → ToCHW → ORT inference → DBPostProcess
+    ///
+    /// - Parameters:
+    ///   - image: The input image to detect text regions in.
+    ///   - runtimeParams: Detection postprocess / resize overrides (see ``OCRRuntimeParams``).
+    /// - Returns: A `DetectionResult` with boxes and per-stage timing metrics.
+    func detect(_ image: CGImage, runtimeParams: OCRRuntimeParams = .noOverrides) async throws -> DetectionResult {
+        let eff = OCRParameterResolver.effective(modelConfig: modelConfig, runtime: runtimeParams)
+        return try await detect(effective: eff) {
+            try preprocessor.preprocess(image, detResize: eff.detResize)
+        }
+    }
+
+    private func detect(
+        effective: EffectiveOCRParams,
+        preprocess: () throws -> PreprocessResult
+    ) async throws -> DetectionResult {
+        let postprocessor = makePostProcessor(effective: effective)
+        // Step 1: Preprocess
+        let preprocessStart = CFAbsoluteTimeGetCurrent()
+        let preprocessed = try preprocess()
+        let preprocessTime = CFAbsoluteTimeGetCurrent() - preprocessStart
+
+        // Step 2: Run ORT inference
+        let inferenceStart = CFAbsoluteTimeGetCurrent()
+        let outputs = try await sessionManager.runDetection(
+            inputData: preprocessed.tensorData,
+            shape: preprocessed.tensorShape
+        )
+        let inferenceTime = CFAbsoluteTimeGetCurrent() - inferenceStart
+
+        // Step 3: Extract the probability map from ORT output
+        // The det model outputs a single tensor with shape [1, 1, H, W]
+        guard let firstOutput = outputs.values.first else {
+            throw DetectionEngineError.noOutputTensor
+        }
+        let outputData = firstOutput.data
+        let outputShape = firstOutput.shape
+
+        // Output shape should be [1, 1, H, W]
+        guard outputShape.count == 4 else {
+            throw DetectionEngineError.unexpectedOutputShape(outputShape)
+        }
+        let tensorHeight = outputShape[2]
+        let tensorWidth = outputShape[3]
+
+        // Step 4: Postprocess
+        let postprocessStart = CFAbsoluteTimeGetCurrent()
+        let boxes = postprocessor.process(
+            outputTensor: outputData,
+            tensorHeight: tensorHeight,
+            tensorWidth: tensorWidth,
+            originalWidth: preprocessed.originalSize.width,
+            originalHeight: preprocessed.originalSize.height
+        )
+        let postprocessTime = CFAbsoluteTimeGetCurrent() - postprocessStart
+
+        let totalTime = preprocessTime + inferenceTime + postprocessTime
+
+        return DetectionResult(
+            boxes: boxes,
+            preprocessTime: preprocessTime,
+            inferenceTime: inferenceTime,
+            postprocessTime: postprocessTime,
+            totalTime: totalTime,
+            inputTensorShape: preprocessed.tensorShape
+        )
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/EncodedImageCodec.swift b/deploy/ios_demo/PaddleOCRDemo/Engine/EncodedImageCodec.swift
new file mode 100644
index 0000000..4346c79
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/EncodedImageCodec.swift
@@ -0,0 +1,51 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import CoreGraphics
+import Foundation
+import ImageIO
+
+enum EncodedImageCodecError: LocalizedError {
+    case emptyData
+    case createImageSourceFailed
+    case createImageFailed
+
+    var errorDescription: String? {
+        switch self {
+        case .emptyData:
+            return "Image data is empty"
+        case .createImageSourceFailed:
+            return "Could not create image source from data"
+        case .createImageFailed:
+            return "Could not decode image from data"
+        }
+    }
+}
+
+/// Decodes PNG/JPEG/WebP and other formats supported by Image I/O into a ``CGImage``.
+enum EncodedImageCodec {
+    static func cgImage(fromEncodedData data: Data) throws -> CGImage {
+        guard !data.isEmpty else {
+            throw EncodedImageCodecError.emptyData
+        }
+        guard let source = CGImageSourceCreateWithData(data as CFData, nil) else {
+            throw EncodedImageCodecError.createImageSourceFailed
+        }
+        let options: [CFString: Any] = [kCGImageSourceShouldCache: true]
+        guard let cgImage = CGImageSourceCreateImageAtIndex(source, 0, options as CFDictionary) else {
+            throw EncodedImageCodecError.createImageFailed
+        }
+        return cgImage
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/InferenceConfig.swift b/deploy/ios_demo/PaddleOCRDemo/Engine/InferenceConfig.swift
new file mode 100644
index 0000000..76ccd8c
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/InferenceConfig.swift
@@ -0,0 +1,322 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import Foundation
+import Yams
+
+// MARK: - Errors
+
+enum InferenceConfigError: LocalizedError {
+    case fileNotFound(String)
+    case parseError(String)
+    case missingField(String)
+
+    var errorDescription: String? {
+        switch self {
+        case .fileNotFound(let path):
+            return "Model config file not found: \(path)"
+        case .parseError(let detail):
+            return "Failed to parse model config: \(detail)"
+        case .missingField(let field):
+            return "Missing required field in model config: \(field)"
+        }
+    }
+}
+
+// MARK: - Transform Operations
+
+/// Channel order implied by `DecodeImage` in the model config file (`img_mode`: BGR or RGB).
+/// When `DecodeImage` is absent, defaults to **BGR** (typical decode path for three-channel inputs).
+enum InferenceImageChannelOrder: Equatable, Sendable {
+    case bgr
+    case rgb
+
+    /// Parses `DecodeImage.img_mode`; empty/missing defaults to **BGR**.
+    static func fromDecodeImage(imgMode: String?) -> InferenceImageChannelOrder {
+        guard let raw = imgMode?.trimmingCharacters(in: .whitespacesAndNewlines), !raw.isEmpty else {
+            return .bgr
+        }
+        switch raw.lowercased() {
+        case "rgb": return .rgb
+        case "bgr": return .bgr
+        default:
+            return .bgr
+        }
+    }
+}
+
+/// Parameters for `DetResizeForTest` from the model config file.
+struct DetResizeParams: Equatable {
+    var limitSideLen: Int
+    var limitType: String
+    var maxSideLimit: Int
+}
+
+/// Represents a single preprocessing transform operation parsed from the model config file.
+/// Each case carries associated parameters read from the model config file.
+enum TransformOp {
+    /// `DecodeImage` in the model config file — selects BGR vs RGB for decoded pixels.
+    case decodeImage(channelOrder: InferenceImageChannelOrder)
+    case detResizeForTest(DetResizeParams)
+    case normalizeImage(scale: Float, mean: [Float], std: [Float], order: String)
+    case toCHWImage
+    case recResizeImg(imageShape: [Int])
+    case unknown(name: String)
+}
+
+// MARK: - Config Structures
+
+struct PreProcessConfig {
+    let transformOps: [TransformOp]
+}
+
+struct PostProcessConfig {
+    let name: String
+    let thresh: Float
+    let boxThresh: Float
+    let maxCandidates: Int
+    let unclipRatio: Float
+    let characterDict: [String]?
+}
+
+struct InferenceConfig {
+    let modelName: String
+    let preProcess: PreProcessConfig
+    let postProcess: PostProcessConfig
+
+    // MARK: - Loading
+
+    /// Loads and parses a model config file into a typed `InferenceConfig`.
+    ///
+    /// - Parameter path: Absolute filesystem path to the model config file.
+    /// - Returns: A fully parsed `InferenceConfig` with typed transform operations.
+    static func load(from path: String) throws -> InferenceConfig {
+        guard FileManager.default.fileExists(atPath: path) else {
+            throw InferenceConfigError.fileNotFound(path)
+        }
+
+        let configFileText: String
+        do {
+            configFileText = try String(contentsOfFile: path, encoding: .utf8)
+        } catch {
+            throw InferenceConfigError.parseError("Cannot read file: \(error.localizedDescription)")
+        }
+
+        guard let root = try Yams.load(yaml: configFileText) as? [String: Any] else {
+            throw InferenceConfigError.parseError("Root element is not a dictionary")
+        }
+
+        // Parse Global.model_name
+        guard let global = root["Global"] as? [String: Any],
+              let modelName = global["model_name"] as? String else {
+            throw InferenceConfigError.missingField("Global.model_name")
+        }
+
+        // Parse PreProcess
+        guard let preProcessDict = root["PreProcess"] as? [String: Any],
+              let transformOpsRaw = preProcessDict["transform_ops"] as? [[String: Any?]] else {
+            throw InferenceConfigError.missingField("PreProcess.transform_ops")
+        }
+
+        let transformOps = transformOpsRaw.map { parseTransformOp($0) }
+        let preProcess = PreProcessConfig(transformOps: transformOps)
+
+        // Parse PostProcess
+        guard let postProcessDict = root["PostProcess"] as? [String: Any] else {
+            throw InferenceConfigError.missingField("PostProcess")
+        }
+        let postProcess = parsePostProcess(postProcessDict)
+
+        return InferenceConfig(
+            modelName: modelName,
+            preProcess: preProcess,
+            postProcess: postProcess
+        )
+    }
+}
+
+extension InferenceConfig {
+    // MARK: - Private Parsing Helpers
+
+    private enum ParseDefaults {
+        static let postThresh: Float = 0.3
+        static let postBoxThresh: Float = 0.6
+        static let postMaxCandidates: Int = 1000
+        static let postUnclipRatio: Float = 2.0
+        static let detResizeLimitSideLen: Int = 960
+        static let detResizeLimitType: String = "max"
+        static let detResizeMaxSideLimit: Int = 4000
+    }
+
+    private static let defaultDetResizeParams = DetResizeParams(
+        limitSideLen: ParseDefaults.detResizeLimitSideLen,
+        limitType: ParseDefaults.detResizeLimitType,
+        maxSideLimit: ParseDefaults.detResizeMaxSideLimit
+    )
+
+    /// Parses a single transform operation dictionary (one key = op name, value = params or null).
+    private static func parseTransformOp(_ dict: [String: Any?]) -> TransformOp {
+        guard let opName = dict.keys.first else {
+            return .unknown(name: "empty")
+        }
+
+        switch opName {
+        case "DecodeImage":
+            let params = dict[opName] as? [String: Any] ?? [:]
+            let imgMode = params["img_mode"] as? String
+            return .decodeImage(channelOrder: InferenceImageChannelOrder.fromDecodeImage(imgMode: imgMode))
+
+        case "DetResizeForTest":
+            let params = dict[opName] as? [String: Any] ?? [:]
+            let limitSideLen =
+                optionalIntValue(params["resize_long"])
+                ?? optionalIntValue(params["limit_side_len"])
+                ?? ParseDefaults.detResizeLimitSideLen
+            let limitType = params["limit_type"] as? String ?? ParseDefaults.detResizeLimitType
+            let maxSideLimit =
+                intValueForKeyIfPresent(params, key: "max_side_limit")
+                ?? ParseDefaults.detResizeMaxSideLimit
+            return .detResizeForTest(
+                DetResizeParams(
+                    limitSideLen: limitSideLen,
+                    limitType: limitType,
+                    maxSideLimit: maxSideLimit
+                )
+            )
+
+        case "NormalizeImage":
+            let params = dict[opName] as? [String: Any] ?? [:]
+            let scale = parseScale(params["scale"])
+            let mean = parseFloatArray(params["mean"]) ?? [0.485, 0.456, 0.406]
+            let std = parseFloatArray(params["std"]) ?? [0.229, 0.224, 0.225]
+            let order = params["order"] as? String ?? "hwc"
+            return .normalizeImage(scale: scale, mean: mean, std: std, order: order)
+
+        case "ToCHWImage":
+            return .toCHWImage
+
+        case "RecResizeImg":
+            let params = dict[opName] as? [String: Any] ?? [:]
+            let imageShape = (params["image_shape"] as? [Any])?.compactMap { toInt($0) } ?? [3, 48, 320]
+            return .recResizeImg(imageShape: imageShape)
+
+        default:
+            return .unknown(name: opName)
+        }
+    }
+
+    /// Parses the `scale` field which may be a numeric value or a fraction string like `"1./255."`.
+    /// Handles string division expressions by splitting on "/" and computing the result.
+    private static func parseScale(_ value: Any?) -> Float {
+        if let floatVal = value as? Double {
+            return Float(floatVal)
+        }
+        if let stringVal = value as? String {
+            return parseScaleString(stringVal)
+        }
+        // Default scale: 1/255
+        return 1.0 / 255.0
+    }
+
+    /// Evaluates a fraction string by splitting on `"/"` and dividing numerator by denominator.
+    private static func parseScaleString(_ s: String) -> Float {
+        if s.contains("/") {
+            let parts = s.split(separator: "/")
+            if parts.count == 2,
+               let numerator = Double(parts[0].trimmingCharacters(in: .init(charactersIn: "."))),
+               let denominator = Double(parts[1].trimmingCharacters(in: .init(charactersIn: "."))) {
+                return Float(numerator / denominator)
+            }
+            let rawParts = s.split(separator: "/")
+            if rawParts.count == 2,
+               let num = Double(rawParts[0]),
+               let den = Double(rawParts[1]) {
+                return Float(num / den)
+            }
+        }
+        if let val = Double(s) {
+            return Float(val)
+        }
+        return 1.0 / 255.0
+    }
+
+    private static func parseFloatArray(_ value: Any?) -> [Float]? {
+        guard let array = value as? [Any] else { return nil }
+        return array.compactMap { element -> Float? in
+            if let d = element as? Double { return Float(d) }
+            if let i = element as? Int { return Float(i) }
+            return nil
+        }
+    }
+
+    private static func parsePostProcess(_ dict: [String: Any]) -> PostProcessConfig {
+        let name = dict["name"] as? String ?? "Unknown"
+        let thresh = (dict["thresh"] as? Double).map { Float($0) } ?? ParseDefaults.postThresh
+        let boxThresh = (dict["box_thresh"] as? Double).map { Float($0) } ?? ParseDefaults.postBoxThresh
+        let maxCandidates = dict["max_candidates"] as? Int ?? ParseDefaults.postMaxCandidates
+        let unclipRatio = (dict["unclip_ratio"] as? Double).map { Float($0) } ?? ParseDefaults.postUnclipRatio
+        let characterDict = (dict["character_dict"] as? [Any])?.compactMap { $0 as? String }
+
+        return PostProcessConfig(
+            name: name,
+            thresh: thresh,
+            boxThresh: boxThresh,
+            maxCandidates: maxCandidates,
+            unclipRatio: unclipRatio,
+            characterDict: characterDict
+        )
+    }
+
+    private static func toInt(_ value: Any) -> Int? {
+        if let i = value as? Int { return i }
+        if let d = value as? Double { return Int(d) }
+        return nil
+    }
+
+    private static func optionalIntValue(_ value: Any?) -> Int? {
+        guard let value else { return nil }
+        return toInt(value)
+    }
+
+    /// Only returns a value when the key exists in `params` (a missing key yields `nil`).
+    private static func stringValueForKeyIfPresent(_ params: [String: Any], key: String) -> String? {
+        guard params.keys.contains(key) else { return nil }
+        return params[key] as? String
+    }
+
+    private static func intValueForKeyIfPresent(_ params: [String: Any], key: String) -> Int? {
+        guard params.keys.contains(key) else { return nil }
+        return optionalIntValue(params[key])
+    }
+
+    /// First `DetResizeForTest` block in the model config file next to the ONNX weights, if present.
+    var detResizeFromModel: DetResizeParams? {
+        for op in preProcess.transformOps {
+            if case .detResizeForTest(let p) = op { return p }
+        }
+        return nil
+    }
+
+    var detResizeForMerge: DetResizeParams {
+        detResizeFromModel ?? Self.defaultDetResizeParams
+    }
+
+    /// First `DecodeImage` entry in `transform_ops` and its `img_mode` (BGR or RGB). If absent, **BGR**.
+    var decodeImageChannelOrder: InferenceImageChannelOrder {
+        for op in preProcess.transformOps {
+            if case .decodeImage(let order) = op { return order }
+        }
+        return .bgr
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/ModelConfig.swift b/deploy/ios_demo/PaddleOCRDemo/Engine/ModelConfig.swift
new file mode 100644
index 0000000..657852d
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/ModelConfig.swift
@@ -0,0 +1,107 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import Foundation
+
+/// Resolves exported ONNX or ORT-format models and their **model config files** in the app bundle (`Models/det`, `Models/rec`).
+/// Resource lookup uses the conventional shared basename for the weight file and the model config file in each folder.
+/// Weights: prefers `inference*.ort` in the model directory (see ``bundledOrtPath``), otherwise `inference.onnx`.
+struct ModelConfig {
+    let modelPath: String
+    let configPath: String
+    /// From `Global.model_name` in the model config file (same directory as the ONNX weights).
+    let name: String
+
+    /// Human-readable label from `Global.model_name` in the model config file; `fallback` if the file cannot be read.
+    private static func displayName(fromModelConfigPath configPath: String, fallback: String) -> String {
+        guard let cfg = try? InferenceConfig.load(from: configPath) else {
+            return fallback
+        }
+        return cfg.modelName
+    }
+
+    /// Bundled model weights: any `inference*.ort` in the subdirectory (``ortBundledFileMatchOrder``), else `inference.onnx`.
+    ///
+    /// ONNX Runtime’s `convert_onnx_models_to_ort` may emit `inference.with_runtime_opt.ort` instead of `inference.ort`
+    /// when using Runtime optimization style; the bundle cannot resolve that with `forResource: "inference"`.
+    private static let ortResourceStems: [String] = [
+        "inference",
+        "inference.with_runtime_opt",
+    ]
+
+    private static func bundledOrtPath(inDirectory directory: String) -> String? {
+        for stem in ortResourceStems {
+            if let p = Bundle.main.path(forResource: stem, ofType: "ort", inDirectory: directory) {
+                return p
+            }
+        }
+        // Other ORT export suffixes (e.g. non-default optim levels): one `inference*.ort` in the folder.
+        guard let urls = Bundle.main.urls(forResourcesWithExtension: "ort", subdirectory: directory) else {
+            return nil
+        }
+        let matches = urls.filter { url in
+            let c = url.lastPathComponent
+            return c.hasPrefix("inference") && c.hasSuffix(".ort")
+        }
+        guard !matches.isEmpty else { return nil }
+        if matches.count == 1 { return matches[0].path }
+        // If multiple, prefer lexicographic first (deterministic); avoid shipping several exports in one dir.
+        return matches.sorted { $0.lastPathComponent < $1.lastPathComponent }.first?.path
+    }
+
+    private static func bundledWeightsPath(inDirectory directory: String, notFoundMessage: String) throws -> String {
+        if let ort = bundledOrtPath(inDirectory: directory) {
+            return ort
+        }
+        if let onnx = Bundle.main.path(forResource: "inference", ofType: "onnx", inDirectory: directory) {
+            return onnx
+        }
+        throw ModelConfigError.modelNotFound(notFoundMessage)
+    }
+
+    static func detection() throws -> ModelConfig {
+        let modelPath = try Self.bundledWeightsPath(
+            inDirectory: "Models/det",
+            notFoundMessage: "det/inference.ort or det/inference.onnx"
+        )
+        guard let configPath = Bundle.main.path(forResource: "inference", ofType: "yml", inDirectory: "Models/det") else {
+            throw ModelConfigError.modelNotFound("det model config file")
+        }
+        let name = displayName(fromModelConfigPath: configPath, fallback: "text_detection")
+        return ModelConfig(modelPath: modelPath, configPath: configPath, name: name)
+    }
+
+    static func recognition() throws -> ModelConfig {
+        let modelPath = try Self.bundledWeightsPath(
+            inDirectory: "Models/rec",
+            notFoundMessage: "rec/inference.ort or rec/inference.onnx"
+        )
+        guard let configPath = Bundle.main.path(forResource: "inference", ofType: "yml", inDirectory: "Models/rec") else {
+            throw ModelConfigError.modelNotFound("rec model config file")
+        }
+        let name = displayName(fromModelConfigPath: configPath, fallback: "text_recognition")
+        return ModelConfig(modelPath: modelPath, configPath: configPath, name: name)
+    }
+}
+
+enum ModelConfigError: LocalizedError {
+    case modelNotFound(String)
+
+    var errorDescription: String? {
+        switch self {
+        case .modelNotFound(let path):
+            return "Model file not found in bundle: \(path)"
+        }
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/OCREngine.swift b/deploy/ios_demo/PaddleOCRDemo/Engine/OCREngine.swift
new file mode 100644
index 0000000..c71c238
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/OCREngine.swift
@@ -0,0 +1,293 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import CoreGraphics
+import Foundation
+
+// MARK: - OCR run result types
+
+/// A single OCR result: one detected text region with its recognized text.
+struct OCRResult {
+    /// Four corner points of the bounding polygon [x, y] in original image coordinates.
+    /// Order: top-left, top-right, bottom-right, bottom-left.
+    let polygon: [[Int32]]
+    /// The recognized text string from CTC decoding.
+    let text: String
+    /// Recognition confidence score (0.0 to 1.0).
+    let confidence: Float
+}
+
+/// Result of one full OCR run on an image, with per-stage timing.
+struct OCRRunResult {
+    /// All detected and recognized text regions, in reading order.
+    let results: [OCRResult]
+    /// Total time spent in the detection stage (preprocess + inference + postprocess).
+    let detectionTime: TimeInterval
+    /// Seconds in detection preprocessing (resize, normalize, HWC-to-CHW).
+    let detectionPreprocessTime: TimeInterval
+    /// Seconds in detection ONNX inference.
+    let detectionInferenceTime: TimeInterval
+    /// Seconds in detection postprocessing (DB map, contours, polygons).
+    let detectionPostprocessTime: TimeInterval
+    /// Total time spent recognizing all text regions (sum of per-line totals).
+    let recognitionTime: TimeInterval
+    /// Sum of recognition preprocess time across all lines (batched work split per line).
+    let recognitionPreprocessTime: TimeInterval
+    /// Sum of recognition inference time across all lines.
+    let recognitionInferenceTime: TimeInterval
+    /// Sum of recognition postprocess (CTC decode) time across all lines.
+    let recognitionPostprocessTime: TimeInterval
+    /// Wall-clock time for the entire run (detect + sort + crop + recognize).
+    let totalTime: TimeInterval
+    /// Time not attributed to detection or recognition totals.
+    let pipelineOverheadTime: TimeInterval
+    /// Number of text lines from detection sent through recognition.
+    let recognitionLineCount: Int
+    /// Detection model input tensor shape for this OCR run, e.g. [1, 3, H, W].
+    let detectionInputTensorShape: [Int]
+    /// Recognition model input tensor shapes for each non-empty recognition batch.
+    let recognitionInputTensorShapes: [[Int]]
+    /// Per-line recognition ONNX inference times in **seconds**.
+    let lineRecognitionInferenceTimes: [TimeInterval]
+    /// Per-line recognition preprocess times in **seconds**.
+    let lineRecognitionPreprocessTimes: [TimeInterval]
+    /// Per-line recognition postprocess times in **seconds**.
+    let lineRecognitionPostprocessTimes: [TimeInterval]
+}
+
+// MARK: - OCR Engine Errors
+
+enum OCREngineError: LocalizedError {
+    case quadTextCropFailed(boxIndex: Int, underlying: Error)
+    case recognitionBatchSizeMismatch(expected: Int, actual: Int)
+
+    var errorDescription: String? {
+        switch self {
+        case .quadTextCropFailed(let idx, let err):
+            return "Quad text crop failed for box \(idx): \(err.localizedDescription)"
+        case .recognitionBatchSizeMismatch(let expected, let actual):
+            return "Recognition returned \(actual) results but \(expected) crops were sent"
+        }
+    }
+}
+
+// MARK: - OCREngine
+
+/// End-to-end OCR: detect → sort → crop → recognize.
+///
+/// Composes `DetectionEngine`, `BoxSorter`, `QuadTextCrop`, and `RecognitionEngine`
+/// into a single `run(CGImage)` call.
+///
+/// Runs entirely via async/await. Since `DetectionEngine` and
+/// `RecognitionEngine` delegate to `ORTSessionManager` (a Swift actor), all ORT
+/// calls are off the main thread.
+///
+/// Usage:
+/// ```swift
+/// let manager = ORTSessionManager()
+/// try await manager.loadModels()
+/// let engine = try OCREngine(sessionManager: manager)
+/// let result = try await engine.run(cgImage, params: .noOverrides)
+/// for item in result.results {
+///     print("\(item.text) (\(item.confidence))")
+/// }
+/// ```
+class OCREngine {
+    private let detectionEngine: DetectionEngine
+    private let recognitionEngine: RecognitionEngine
+
+    /// Initialize with an existing ORTSessionManager (models must already be loaded).
+    ///
+    /// Creates both DetectionEngine and RecognitionEngine.
+    ///
+    /// - Parameter sessionManager: A loaded ORTSessionManager.
+    /// - Throws: If either engine's model config cannot be loaded.
+    init(sessionManager: ORTSessionManager) throws {
+        self.detectionEngine = try DetectionEngine(sessionManager: sessionManager)
+        self.recognitionEngine = try RecognitionEngine(sessionManager: sessionManager)
+    }
+
+    func baselineRuntimeDefaults() -> ResolvedOCRRuntimeParams {
+        ResolvedOCRRuntimeParams.fromModelConfig(detectionEngine.modelConfig)
+    }
+
+    /// Run full OCR on an image.
+    ///
+    /// End-to-end OCR flow (detect → sort → crop → recognize):
+    /// 1. **Detect**: Run detection to get bounding polygons
+    /// 2. **Sort**: Sort boxes in reading order (top-to-bottom, left-to-right)
+    /// 3. **Crop + Recognize**: Crop each region, order crops by ascending width/height for batched
+    ///    recognition, then map results back to reading order
+    ///
+    /// - Parameters:
+    ///   - image: The input `CGImage` to process.
+    ///   - params: Optional runtime parameter overrides (see ``OCRRuntimeParams``).
+    /// - Returns: `OCRRunResult` with all line results and timing.
+    func run(_ image: CGImage, params: OCRRuntimeParams = .noOverrides) async throws -> OCRRunResult {
+        let runStart = CFAbsoluteTimeGetCurrent()
+
+        let resolved = params.resolved(detectionEngine.modelConfig)
+
+        let detResult = try await detectionEngine.detect(image, runtimeParams: params)
+        let sortedBoxes = BoxSorter.sortInReadingOrder(detResult.boxes)
+        let (ocrResults, recTiming, perLine) = try await recognizeSortedBoxes(
+            sortedBoxes,
+            sourceImage: image,
+            resolved: resolved
+        )
+
+        let totalTime = CFAbsoluteTimeGetCurrent() - runStart
+        let overhead = totalTime - detResult.totalTime - recTiming.total
+
+        return OCRRunResult(
+            results: ocrResults,
+            detectionTime: detResult.totalTime,
+            detectionPreprocessTime: detResult.preprocessTime,
+            detectionInferenceTime: detResult.inferenceTime,
+            detectionPostprocessTime: detResult.postprocessTime,
+            recognitionTime: recTiming.total,
+            recognitionPreprocessTime: recTiming.preprocess,
+            recognitionInferenceTime: recTiming.inference,
+            recognitionPostprocessTime: recTiming.postprocess,
+            totalTime: totalTime,
+            pipelineOverheadTime: max(0, overhead),
+            recognitionLineCount: perLine.count,
+            detectionInputTensorShape: detResult.inputTensorShape,
+            recognitionInputTensorShapes: perLine.inputTensorShapes,
+            lineRecognitionInferenceTimes: perLine.inference,
+            lineRecognitionPreprocessTimes: perLine.preprocess,
+            lineRecognitionPostprocessTimes: perLine.postprocess
+        )
+    }
+
+    /// Aggregated recognition timing across all lines (sums of per-line split batch times).
+    private struct RecognitionTimingAggregate {
+        let preprocess: TimeInterval
+        let inference: TimeInterval
+        let postprocess: TimeInterval
+        let total: TimeInterval
+    }
+
+    /// Per-line rec timings
+    private struct PerLineRecognitionTimes {
+        let count: Int
+        let inputTensorShapes: [[Int]]
+        let inference: [TimeInterval]
+        let preprocess: [TimeInterval]
+        let postprocess: [TimeInterval]
+    }
+
+    private func recognizeSortedBoxes(
+        _ sortedBoxes: [DetectionBox],
+        sourceImage: CGImage,
+        resolved: ResolvedOCRRuntimeParams
+    ) async throws -> ([OCRResult], RecognitionTimingAggregate, PerLineRecognitionTimes) {
+        struct LineCrop {
+            let lineIndex: Int
+            let aspectRatio: Float
+            let crop: CGImage
+            let polygon: [[Int32]]
+        }
+
+        var lines: [LineCrop] = []
+        lines.reserveCapacity(sortedBoxes.count)
+
+        for (index, box) in sortedBoxes.enumerated() {
+            let croppedImage: CGImage
+            do {
+                croppedImage = try QuadTextCrop.crop(sourceImage, polygon: box.points)
+            } catch {
+                throw OCREngineError.quadTextCropFailed(boxIndex: index, underlying: error)
+            }
+            let h = max(croppedImage.height, 1)
+            let aspect = Float(croppedImage.width) / Float(h)
+            lines.append(LineCrop(lineIndex: index, aspectRatio: aspect, crop: croppedImage, polygon: box.points))
+        }
+
+        let sortedForInference = lines.sorted { a, b in
+            if a.aspectRatio != b.aspectRatio {
+                return a.aspectRatio < b.aspectRatio
+            }
+            return a.lineIndex < b.lineIndex
+        }
+
+        let batchSize = max(1, resolved.textRecBatchSize)
+        var perLine: [RecognitionEngineResult?] = Array(repeating: nil, count: lines.count)
+        var batchInputShapes: [[Int]] = []
+
+        var chunkStart = 0
+        while chunkStart < sortedForInference.count {
+            let chunkEnd = min(chunkStart + batchSize, sortedForInference.count)
+            let chunk = Array(sortedForInference[chunkStart..<chunkEnd])
+            let crops = chunk.map(\.crop)
+            let recBatch = try await recognitionEngine.recognizeBatch(crops)
+            guard recBatch.count == chunk.count else {
+                throw OCREngineError.recognitionBatchSizeMismatch(expected: chunk.count, actual: recBatch.count)
+            }
+            if let first = recBatch.first {
+                batchInputShapes.append(first.inputTensorShape)
+            }
+            for (i, item) in chunk.enumerated() {
+                perLine[item.lineIndex] = recBatch[i]
+            }
+            chunkStart = chunkEnd
+        }
+
+        var ocrResults: [OCRResult] = []
+        var totalRecTime: TimeInterval = 0
+        var totalRecPre: TimeInterval = 0
+        var totalRecInf: TimeInterval = 0
+        var totalRecPost: TimeInterval = 0
+        var lineInf: [TimeInterval] = []
+        var linePre: [TimeInterval] = []
+        var linePost: [TimeInterval] = []
+        lineInf.reserveCapacity(lines.count)
+        linePre.reserveCapacity(lines.count)
+        linePost.reserveCapacity(lines.count)
+
+        for line in lines {
+            guard let recResult = perLine[line.lineIndex] else { continue }
+            linePre.append(recResult.preprocessTime)
+            lineInf.append(recResult.inferenceTime)
+            linePost.append(recResult.postprocessTime)
+            totalRecTime += recResult.totalTime
+            totalRecPre += recResult.preprocessTime
+            totalRecInf += recResult.inferenceTime
+            totalRecPost += recResult.postprocessTime
+            if recResult.confidence < resolved.textRecScoreThresh {
+                continue
+            }
+            ocrResults.append(OCRResult(
+                polygon: line.polygon,
+                text: recResult.text,
+                confidence: recResult.confidence
+            ))
+        }
+
+        let aggregate = RecognitionTimingAggregate(
+            preprocess: totalRecPre,
+            inference: totalRecInf,
+            postprocess: totalRecPost,
+            total: totalRecTime
+        )
+        let perLineOut = PerLineRecognitionTimes(
+            count: lines.count,
+            inputTensorShapes: batchInputShapes,
+            inference: lineInf,
+            preprocess: linePre,
+            postprocess: linePost
+        )
+        return (ocrResults, aggregate, perLineOut)
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/OCRParameterResolver.swift b/deploy/ios_demo/PaddleOCRDemo/Engine/OCRParameterResolver.swift
new file mode 100644
index 0000000..fee9c4a
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/OCRParameterResolver.swift
@@ -0,0 +1,238 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import Foundation
+
+// MARK: - Runtime (UI) parameters
+
+struct OCRRuntimeParams: Equatable, Sendable {
+    // Detection — preprocess
+    var textDetLimitSideLen: Int?
+    var textDetLimitType: String?
+    var textDetMaxSideLimit: Int?
+    // Detection — postprocess
+    var textDetThresh: Float?
+    var textDetBoxThresh: Float?
+    var textDetUnclipRatio: Float?
+    // Recognition
+    var textRecBatchSize: Int?
+    var textRecScoreThresh: Float?
+
+    static let noOverrides = OCRRuntimeParams(
+        textDetLimitSideLen: nil,
+        textDetLimitType: nil,
+        textDetMaxSideLimit: nil,
+        textDetThresh: nil,
+        textDetBoxThresh: nil,
+        textDetUnclipRatio: nil,
+        textRecBatchSize: nil,
+        textRecScoreThresh: nil
+    )
+}
+
+// MARK: - Resolved parameters (for UI baseline & filtering)
+
+struct ResolvedOCRRuntimeParams: Equatable, Sendable {
+    // Detection — preprocess
+    var textDetLimitSideLen: Int
+    var textDetLimitType: String
+    var textDetMaxSideLimit: Int
+    // Detection — postprocess
+    var textDetThresh: Float
+    var textDetBoxThresh: Float
+    var textDetUnclipRatio: Float
+    // Recognition
+    var textRecBatchSize: Int
+    var textRecScoreThresh: Float
+
+    init(
+        textDetLimitSideLen: Int,
+        textDetLimitType: String,
+        textDetMaxSideLimit: Int,
+        textDetThresh: Float,
+        textDetBoxThresh: Float,
+        textDetUnclipRatio: Float,
+        textRecBatchSize: Int,
+        textRecScoreThresh: Float,
+    ) {
+        self.textDetLimitSideLen = textDetLimitSideLen
+        self.textDetLimitType = textDetLimitType
+        self.textDetMaxSideLimit = textDetMaxSideLimit
+        self.textDetThresh = textDetThresh
+        self.textDetBoxThresh = textDetBoxThresh
+        self.textDetUnclipRatio = textDetUnclipRatio
+        self.textRecBatchSize = textRecBatchSize
+        self.textRecScoreThresh = textRecScoreThresh
+    }
+
+    static let fallbackForUI = ResolvedOCRRuntimeParams(
+        textDetLimitSideLen: OCRParameterResolver.textDetLimitSideLenAppDefault,
+        textDetLimitType: OCRParameterResolver.textDetLimitTypeAppDefault,
+        textDetMaxSideLimit: OCRParameterResolver.textDetMaxSideLimitAppDefault,
+        textDetThresh: OCRParameterResolver.textDetThreshAppDefault,
+        textDetBoxThresh: OCRParameterResolver.textDetBoxThreshAppDefault,
+        textDetUnclipRatio: OCRParameterResolver.textDetUnclipRatioAppDefault,
+        textRecBatchSize: OCRParameterResolver.textRecBatchSizeAppDefault,
+        textRecScoreThresh: OCRParameterResolver.textRecScoreThreshAppDefault
+    )
+}
+
+extension OCRRuntimeParams {
+    func resolved(_ modelConfig: InferenceConfig) -> ResolvedOCRRuntimeParams {
+        let base = ResolvedOCRRuntimeParams.fromModelConfig(modelConfig)
+        return ResolvedOCRRuntimeParams(
+            textDetLimitSideLen: textDetLimitSideLen ?? base.textDetLimitSideLen,
+            textDetLimitType: textDetLimitType ?? base.textDetLimitType,
+            textDetMaxSideLimit: textDetMaxSideLimit ?? base.textDetMaxSideLimit,
+            textDetThresh: textDetThresh ?? base.textDetThresh,
+            textDetBoxThresh: textDetBoxThresh ?? base.textDetBoxThresh,
+            textDetUnclipRatio: textDetUnclipRatio ?? base.textDetUnclipRatio,
+            textRecBatchSize: textRecBatchSize ?? base.textRecBatchSize,
+            textRecScoreThresh: textRecScoreThresh ?? base.textRecScoreThresh
+        )
+    }
+}
+
+extension ResolvedOCRRuntimeParams {
+    /// Baseline with no UI overrides, using ``OCRParameterResolver`` app-tier defaults.
+    static func fromModelConfig(_ modelConfig: InferenceConfig) -> ResolvedOCRRuntimeParams {
+        let o = OCRParameterResolver.effective(modelConfig: modelConfig, runtime: .noOverrides)
+        return ResolvedOCRRuntimeParams(
+            textDetLimitSideLen: o.mergedLimitSideLen,
+            textDetLimitType: o.mergedLimitType,
+            textDetMaxSideLimit: o.mergedMaxSideLimit,
+            textDetThresh: o.detThresh,
+            textDetBoxThresh: o.detBoxThresh,
+            textDetUnclipRatio: o.detUnclipRatio,
+            textRecBatchSize: o.textRecBatchSize,
+            textRecScoreThresh: o.recScoreThresh
+        )
+    }
+}
+
+// MARK: - Effective parameters (single merge pass)
+
+struct EffectiveOCRParams: Equatable, Sendable {
+    // Detection — preprocess
+    var detResize: DetResizeParams
+    var mergedLimitSideLen: Int
+    var mergedLimitType: String
+    var mergedMaxSideLimit: Int
+    // Detection — postprocess
+    var detThresh: Float
+    var detBoxThresh: Float
+    var detUnclipRatio: Float
+    // Recognition
+    var textRecBatchSize: Int
+    var recScoreThresh: Float
+}
+
+// MARK: - Resolver
+
+enum OCRParameterResolver {
+
+    // MARK: App tier
+
+    // Detection — preprocess
+    static let textDetLimitSideLenAppDefault: Int = 64
+    static let textDetLimitTypeAppDefault: String = "min"
+    static let textDetMaxSideLimitAppDefault: Int = 4000
+    // Detection — postprocess
+    static let textDetThreshAppDefault: Float = 0.3
+    static let textDetBoxThreshAppDefault: Float = 0.6
+    static let textDetUnclipRatioAppDefault: Float = 1.5
+    // Recognition
+    static let textRecBatchSizeAppDefault: Int = 1
+    static let textRecScoreThreshAppDefault: Float = 0.0
+
+    // MARK: Merge helpers
+
+    /// UI override if set; otherwise the app-tier default.
+    private static func mergeFloat(ui: Float?, app: Float) -> Float {
+        ui ?? app
+    }
+
+    private static func mergeInt(ui: Int?, app: Int) -> Int {
+        ui ?? app
+    }
+
+    private static func mergeString(ui: String?, app: String) -> String {
+        ui ?? app
+    }
+
+    private static func mergedLimitSideFields(runtime: OCRRuntimeParams) -> (
+        sideLen: Int, limitType: String, maxSide: Int
+    ) {
+        let sideLen = mergeInt(
+            ui: runtime.textDetLimitSideLen,
+            app: Self.textDetLimitSideLenAppDefault
+        )
+        let limitType = mergeString(
+            ui: runtime.textDetLimitType,
+            app: Self.textDetLimitTypeAppDefault
+        )
+        let maxSide = mergeInt(
+            ui: runtime.textDetMaxSideLimit,
+            app: Self.textDetMaxSideLimitAppDefault
+        )
+        return (sideLen, limitType, maxSide)
+    }
+
+    private static func limitSideParams(runtime: OCRRuntimeParams) -> DetResizeParams {
+        let m = mergedLimitSideFields(runtime: runtime)
+        return DetResizeParams(
+            limitSideLen: m.sideLen,
+            limitType: m.limitType,
+            maxSideLimit: m.maxSide
+        )
+    }
+
+    /// Merges ``OCRRuntimeParams`` with app-tier defaults (UI → app). The `modelConfig` argument is
+    /// unused at this moment.
+    static func effective(modelConfig _: InferenceConfig, runtime: OCRRuntimeParams) -> EffectiveOCRParams {
+        let detResize = limitSideParams(runtime: runtime)
+        let merged = mergedLimitSideFields(runtime: runtime)
+
+        let detThresh = mergeFloat(
+            ui: runtime.textDetThresh,
+            app: Self.textDetThreshAppDefault
+        )
+        let detBoxThresh = mergeFloat(
+            ui: runtime.textDetBoxThresh,
+            app: Self.textDetBoxThreshAppDefault
+        )
+        let detUnclipRatio = mergeFloat(
+            ui: runtime.textDetUnclipRatio,
+            app: Self.textDetUnclipRatioAppDefault
+        )
+
+        let textRecBatchSize = mergeInt(
+            ui: runtime.textRecBatchSize,
+            app: Self.textRecBatchSizeAppDefault
+        )
+        let recScoreThresh = runtime.textRecScoreThresh ?? Self.textRecScoreThreshAppDefault
+
+        return EffectiveOCRParams(
+            detResize: detResize,
+            mergedLimitSideLen: merged.sideLen,
+            mergedLimitType: merged.limitType,
+            mergedMaxSideLimit: merged.maxSide,
+            detThresh: detThresh,
+            detBoxThresh: detBoxThresh,
+            detUnclipRatio: detUnclipRatio,
+            textRecBatchSize: textRecBatchSize,
+            recScoreThresh: recScoreThresh
+        )
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/ORTProfilingBridge.h b/deploy/ios_demo/PaddleOCRDemo/Engine/ORTProfilingBridge.h
new file mode 100644
index 0000000..6353309
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/ORTProfilingBridge.h
@@ -0,0 +1,45 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#import <Foundation/Foundation.h>
+
+@class ORTSession;
+@class ORTSessionOptions;
+
+NS_ASSUME_NONNULL_BEGIN
+
+/// Bridges ONNX Runtime C++ profiling APIs that are not exposed on
+/// `onnxruntime-objc`'s public surface.
+///
+/// - Warning: `endProfilingOnSession` reads the private `_session` ivar on
+/// `ORTSession` (layout depends on the
+///   installed `onnxruntime-objc` version). Re-verify after pod upgrades.
+@interface ORTProfilingBridge : NSObject
+
++ (BOOL)enableProfilingOnSessionOptions:(ORTSessionOptions *)options
+                             pathPrefix:(NSString *)pathPrefix
+                                  error:(NSError *_Nullable __autoreleasing *)
+                                            error
+    NS_SWIFT_NAME(enableProfiling(sessionOptions:pathPrefix:));
+
+/// Ends profiling and returns the path to the written JSON (see ORT
+/// `Session::EndProfilingAllocated`).
++ (nullable NSString *)
+    endProfilingOnSession:(ORTSession *)session
+                    error:(NSError *_Nullable __autoreleasing *)error
+    NS_SWIFT_NAME(endProfiling(session:));
+
+@end
+
+NS_ASSUME_NONNULL_END
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/ORTProfilingBridge.mm b/deploy/ios_demo/PaddleOCRDemo/Engine/ORTProfilingBridge.mm
new file mode 100644
index 0000000..8b20c94
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/ORTProfilingBridge.mm
@@ -0,0 +1,72 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#import "ORTProfilingBridge.h"
+
+#import "ort_session_internal.h"
+
+#import "cxx_api.h"
+#import "error_utils.h"
+
+#import <objc/runtime.h>
+
+#include <optional>
+
+@implementation ORTProfilingBridge
+
++ (BOOL)enableProfilingOnSessionOptions:(ORTSessionOptions *)options
+                             pathPrefix:(NSString *)pathPrefix
+                                  error:(NSError **)error {
+  if (options == nil || pathPrefix.length == 0) {
+    ORTSaveCodeAndDescriptionToError(ORT_INVALID_ARGUMENT, @"ORTProfilingBridge: options and non-empty pathPrefix required", error);
+    return NO;
+  }
+  try {
+    [options CXXAPIOrtSessionOptions].EnableProfiling(pathPrefix.fileSystemRepresentation);
+    return YES;
+  }
+  ORT_OBJC_API_IMPL_CATCH_RETURNING_BOOL(error);
+}
+
++ (NSString *)endProfilingOnSession:(ORTSession *)session error:(NSError **)error {
+  if (session == nil) {
+    ORTSaveCodeAndDescriptionToError(ORT_INVALID_ARGUMENT, @"ORTProfilingBridge: session is nil", error);
+    return nil;
+  }
+  try {
+    Ivar ivar = class_getInstanceVariable([ORTSession class], "_session");
+    if (ivar == nullptr) {
+      ORTSaveCodeAndDescriptionToError(ORT_FAIL,
+                                     @"ORTProfilingBridge: ORTSession has no _session ivar — onnxruntime-objc mismatch?",
+                                     error);
+      return nil;
+    }
+    uintptr_t base = reinterpret_cast<uintptr_t>(session);
+    auto *opt = reinterpret_cast<std::optional<Ort::Session> *>(base + ivar_getOffset(ivar));
+    if (opt == nullptr || !opt->has_value()) {
+      ORTSaveCodeAndDescriptionToError(ORT_FAIL, @"ORTProfilingBridge: ORTSession is not ready", error);
+      return nil;
+    }
+    Ort::AllocatorWithDefaultOptions allocator;
+    Ort::AllocatedStringPtr path = opt->value().EndProfilingAllocated(allocator);
+    const char *cstr = path.get();
+    if (cstr == nullptr) {
+      return nil;
+    }
+    return [NSString stringWithUTF8String:cstr];
+  }
+  ORT_OBJC_API_IMPL_CATCH_RETURNING_NULLABLE(error);
+}
+
+@end
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/ORTSessionManager.swift b/deploy/ios_demo/PaddleOCRDemo/Engine/ORTSessionManager.swift
new file mode 100644
index 0000000..4e8a098
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/ORTSessionManager.swift
@@ -0,0 +1,338 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import Foundation
+
+/// Preset for which ONNX Runtime execution provider is registered first on the session chain in this demo.
+enum ORTPrimaryExecutionProvider: String, CaseIterable, Identifiable, Sendable {
+    /// Default CPU execution. Default for this demo.
+    case cpu = "cpu"
+    /// Registers the XNNPACK EP; nodes it cannot run still use the CPU EP.
+    case xnnpack = "xnnpack"
+    /// Registers the Core ML EP; remaining subgraphs use other registered EPs (commonly CPU).
+    case coreML = "core_ml"
+
+    var id: String { rawValue }
+
+    var displayTitle: String {
+        switch self {
+        case .cpu: return "CPU"
+        case .xnnpack: return "XNNPACK"
+        case .coreML: return "Core ML"
+        }
+    }
+}
+
+enum ORTCoreMLProviderOption: String, CaseIterable, Identifiable, Sendable {
+    case enableOnSubgraphs
+    case createMLProgram
+    case staticInputShapes
+    case cpuOnly
+    case cpuAndGPU
+    case aneOnly
+
+    var id: String { rawValue }
+
+    var displayTitle: String {
+        switch self {
+        case .enableOnSubgraphs: return "Enable on subgraphs"
+        case .createMLProgram: return "Use ML program"
+        case .staticInputShapes: return "Fixed tensor shapes only"
+        case .cpuOnly: return "Core ML CPU only"
+        case .cpuAndGPU: return "Core ML CPU and GPU"
+        case .aneOnly: return "Only on devices with ANE"
+        }
+    }
+
+    var detailCaption: String {
+        switch self {
+        case .enableOnSubgraphs: return "Compile subgraphs independently for Core ML."
+        case .createMLProgram: return "Prefer MLProgram over NeuralNetwork (when supported)."
+        case .staticInputShapes:
+            return "Core ML tries to run only ops whose input tensors have fixed shapes at inference time; parts with varying dimensions typically stay on the CPU EP instead."
+        case .cpuOnly: return "Forces Core ML compute on CPU. Mutually exclusive with CPU and GPU."
+        case .cpuAndGPU: return "Allows Core ML to use CPU and GPU. Mutually exclusive with CPU only."
+        case .aneOnly: return "Only enable Core ML EP on devices that have Apple Neural Engine hardware."
+        }
+    }
+}
+
+enum ORTSessionManagerError: LocalizedError {
+    case modelNotFound(String)
+    case sessionCreationFailed(String)
+    case inferenceFailed(String)
+    case inputTensorElementCountMismatch(expected: Int, actual: Int, modelName: String)
+    case outputContainsNaN(String)
+    case ortProfilingFailed(String)
+
+    var errorDescription: String? {
+        switch self {
+        case .modelNotFound(let name): return "Model not found: \(name)"
+        case .sessionCreationFailed(let detail): return "Session creation failed: \(detail)"
+        case .inferenceFailed(let detail): return "Inference failed: \(detail)"
+        case .inputTensorElementCountMismatch(let expected, let actual, let modelName):
+            return "\(modelName): input tensor size mismatch (expected \(expected) floats, got \(actual))"
+        case .outputContainsNaN(let name): return "Output tensor '\(name)' contains NaN values"
+        case .ortProfilingFailed(let detail): return "ONNX Runtime profiling: \(detail)"
+        }
+    }
+}
+
+/// File URLs of ONNX Runtime JSON profiles after ``ORTSessionManager/finalizeORTProfiling()`` (separate file per sub-model).
+struct ORTProfilingOutput: Sendable {
+    var detectionProfileJSON: URL
+    var recognitionProfileJSON: URL
+}
+
+/// Tuning options for ORT session creation.
+struct ORTSessionTuningOptions: Sendable {
+    /// Session-level intra-op thread count. 0 = ORT default.
+    var intraOpThreads: Int = 0
+    /// XNNPACK EP thread count. 0 = ORT default.
+    var xnnpackThreads: Int = 0
+    /// CoreML EP flags: enableOnSubgraphs, createMLProgram, staticInputShapes, cpuOnly, cpuAndGPU, aneOnly.
+    var coreMLFlags: Set<String> = []
+
+    static let `default` = ORTSessionTuningOptions()
+}
+
+extension ORTSessionTuningOptions: Equatable {}
+
+/// At most one placement-related Core ML EP option is applied. Priority matches
+/// ``OnnxRuntimeSettingsPanel`` when several keys appear (e.g. benchmark env / JSON).
+private func applyCoreMLPlacementFlags(from flags: Set<String>, to coreml: ORTCoreMLExecutionProviderOptions) {
+    if flags.contains(ORTCoreMLProviderOption.cpuOnly.rawValue) {
+        coreml.useCPUOnly = true
+    } else if flags.contains(ORTCoreMLProviderOption.cpuAndGPU.rawValue) {
+        coreml.useCPUAndGPU = true
+    } else if flags.contains(ORTCoreMLProviderOption.aneOnly.rawValue) {
+        coreml.onlyEnableForDevicesWithANE = true
+    }
+}
+
+actor ORTSessionManager {
+    private struct SessionIONames {
+        let inputName: String
+        let outputNames: Set<String>
+    }
+
+    private var env: ORTEnv?
+    private var detSession: ORTSession?
+    private var recSession: ORTSession?
+    private var detIO: SessionIONames?
+    private var recIO: SessionIONames?
+    private var ortProfilingPendingFinalize: Bool = false
+
+    /// Load both models. Creates one ORTEnv (per ORT docs: one per process),
+    /// configures execution providers per ``ORTPrimaryExecutionProvider``,
+    /// and creates sessions for detection and recognition models.
+    ///
+    /// - Parameter ortProfiling: When `true`, enables ONNX Runtime session profiling (written as JSON on finalize).
+    ///   Profiling adds overhead; do not treat wall-clock times from the same run as a clean latency benchmark.
+    /// - Parameter tuning: Session tuning options (threads, EP flags). Default reads from environment.
+    func loadModels(executionProvider: ORTPrimaryExecutionProvider = .cpu, ortProfiling: Bool = false, tuning: ORTSessionTuningOptions = .default) async throws {
+        let env = try ORTEnv(loggingLevel: .warning)
+        self.env = env
+        ortProfilingPendingFinalize = ortProfiling
+
+        let detConfig = try ModelConfig.detection()
+        let recConfig = try ModelConfig.recognition()
+
+        if ortProfiling {
+            let baseDir = try ortProfilingDirectoryURL()
+            let detPrefix = baseDir.appendingPathComponent("paddle_ort_det", isDirectory: false).path
+            let recPrefix = baseDir.appendingPathComponent("paddle_ort_rec", isDirectory: false).path
+            let detOptions = try makeSessionOptions(executionProvider: executionProvider, ortProfilePathPrefix: detPrefix, tuning: tuning)
+            let recOptions = try makeSessionOptions(executionProvider: executionProvider, ortProfilePathPrefix: recPrefix, tuning: tuning)
+            detSession = try ORTSession(env: env, modelPath: detConfig.modelPath, sessionOptions: detOptions)
+            recSession = try ORTSession(env: env, modelPath: recConfig.modelPath, sessionOptions: recOptions)
+        } else {
+            let options = try makeSessionOptions(executionProvider: executionProvider, ortProfilePathPrefix: nil, tuning: tuning)
+            detSession = try ORTSession(env: env, modelPath: detConfig.modelPath, sessionOptions: options)
+            recSession = try ORTSession(env: env, modelPath: recConfig.modelPath, sessionOptions: options)
+        }
+        if let detSession {
+            detIO = try makeSessionIONames(session: detSession, modelName: "det")
+        }
+        if let recSession {
+            recIO = try makeSessionIONames(session: recSession, modelName: "rec")
+        }
+    }
+
+    /// Flushes ORT profiling and returns the JSON file paths. Call once after the last inference when
+    /// ``loadModels(ortProfiling:)`` was `true`. Safe to call when profiling was disabled (returns `nil`).
+    func finalizeORTProfiling() throws -> ORTProfilingOutput? {
+        guard ortProfilingPendingFinalize else { return nil }
+        ortProfilingPendingFinalize = false
+        guard let det = detSession, let rec = recSession else {
+            throw ORTSessionManagerError.sessionCreationFailed("Cannot finalize ORT profiling: session missing")
+        }
+        let detPath = try ORTProfilingBridge.endProfiling(session: det)
+        let recPath = try ORTProfilingBridge.endProfiling(session: rec)
+        if detPath.isEmpty || recPath.isEmpty {
+            throw ORTSessionManagerError.ortProfilingFailed("EndProfiling returned an empty path")
+        }
+        return ORTProfilingOutput(
+            detectionProfileJSON: URL(fileURLWithPath: detPath),
+            recognitionProfileJSON: URL(fileURLWithPath: recPath)
+        )
+    }
+
+    private func ortProfilingDirectoryURL() throws -> URL {
+        let base = try FileManager.default.url(
+            for: .cachesDirectory,
+            in: .userDomainMask,
+            appropriateFor: nil,
+            create: true
+        )
+        let dir = base.appendingPathComponent("PaddleOCRORTProfiling", isDirectory: true)
+        try FileManager.default.createDirectory(at: dir, withIntermediateDirectories: true)
+        return dir
+    }
+
+    private func makeSessionOptions(executionProvider: ORTPrimaryExecutionProvider, ortProfilePathPrefix: String?, tuning: ORTSessionTuningOptions) throws -> ORTSessionOptions {
+        let options = try ORTSessionOptions()
+        try options.setGraphOptimizationLevel(.all)
+        if tuning.intraOpThreads > 0 {
+            try options.setIntraOpNumThreads(Int32(tuning.intraOpThreads))
+        }
+        if let prefix = ortProfilePathPrefix {
+            try ORTProfilingBridge.enableProfiling(sessionOptions: options, pathPrefix: prefix)
+        }
+        switch executionProvider {
+        case .cpu:
+            break
+        case .xnnpack:
+            let xnnpackOptions = ORTXnnpackExecutionProviderOptions()
+            if tuning.xnnpackThreads > 0 {
+                xnnpackOptions.intra_op_num_threads = Int32(tuning.xnnpackThreads)
+            }
+            try options.appendXnnpackExecutionProvider(with: xnnpackOptions)
+        case .coreML:
+            let coremlOptions = ORTCoreMLExecutionProviderOptions()
+            let flags = tuning.coreMLFlags
+            if flags.contains("enableOnSubgraphs") { coremlOptions.enableOnSubgraphs = true }
+            if flags.contains("createMLProgram") { coremlOptions.createMLProgram = true }
+            if flags.contains("staticInputShapes") { coremlOptions.onlyAllowStaticInputShapes = true }
+            applyCoreMLPlacementFlags(from: flags, to: coremlOptions)
+            try options.appendCoreMLExecutionProvider(with: coremlOptions)
+        }
+        return options
+    }
+
+    private func makeSessionIONames(session: ORTSession, modelName: String) throws -> SessionIONames {
+        let inputNames = try session.inputNames()
+        let outputNamesList = try session.outputNames()
+        guard let firstInputName = inputNames.first else {
+            throw ORTSessionManagerError.inferenceFailed("\(modelName): no input names found")
+        }
+        return SessionIONames(inputName: firstInputName, outputNames: Set(outputNamesList))
+    }
+
+    /// Run detection inference with real preprocessed input data.
+    ///
+    /// - Parameters:
+    ///   - inputData: Float32 array in CHW layout; length must equal the product of `shape`.
+    ///   - shape: Tensor shape, e.g. [1, 3, 896, 960].
+    /// - Returns: Dictionary mapping output name to (data as [Float], shape as [Int]).
+    func runDetection(inputData: [Float], shape: [Int]) async throws -> [String: (data: [Float], shape: [Int])] {
+        guard let session = detSession else {
+            throw ORTSessionManagerError.sessionCreationFailed("Detection session not loaded")
+        }
+        guard let io = detIO else {
+            throw ORTSessionManagerError.sessionCreationFailed("Detection session IO names not loaded")
+        }
+        return try runInference(session: session, io: io, modelName: "det", inputData: inputData, shape: shape)
+    }
+
+    /// Run recognition inference with preprocessed input data.
+    ///
+    /// The recognition model accepts dynamic batch and width: shape `[N, 3, H, W]` with `N ≥ 1`,
+    /// fixed `H` (e.g. 48), and `W` shared after padding within each batch. A single line uses `N = 1`.
+    ///
+    /// - Parameters:
+    ///   - inputData: Float32 row-major tensor; length must equal the product of `shape`.
+    ///   - shape: e.g. `[1, 3, 48, 320]` or `[6, 3, 48, 640]` for a batch of six lines.
+    /// - Returns: Dictionary mapping output name to (data as [Float], shape as [Int]).
+    func runRecognition(inputData: [Float], shape: [Int]) async throws -> [String: (data: [Float], shape: [Int])] {
+        guard let session = recSession else {
+            throw ORTSessionManagerError.sessionCreationFailed("Recognition session not loaded")
+        }
+        guard let io = recIO else {
+            throw ORTSessionManagerError.sessionCreationFailed("Recognition session IO names not loaded")
+        }
+        return try runInference(session: session, io: io, modelName: "rec", inputData: inputData, shape: shape)
+    }
+
+    /// Shared inference logic used by both detection and recognition.
+    ///
+    /// Discovers input/output names at runtime, creates a float32 tensor from the
+    /// provided data and shape, runs the session, extracts output arrays, and
+    /// validates that no output contains NaN values.
+    private func runInference(
+        session: ORTSession,
+        io: SessionIONames,
+        modelName: String,
+        inputData: [Float],
+        shape: [Int]
+    ) throws -> [String: (data: [Float], shape: [Int])] {
+        let expectedElements = shape.reduce(1, *)
+        guard expectedElements == inputData.count else {
+            throw ORTSessionManagerError.inputTensorElementCountMismatch(
+                expected: expectedElements,
+                actual: inputData.count,
+                modelName: modelName
+            )
+        }
+
+        // Create ORT tensor from input data
+        let nsShape = shape.map { NSNumber(value: $0) }
+        var data = inputData
+        let tensorData = NSMutableData(
+            bytes: &data,
+            length: inputData.count * MemoryLayout<Float>.stride
+        )
+        let inputTensor = try ORTValue(
+            tensorData: tensorData,
+            elementType: .float,
+            shape: nsShape
+        )
+
+        let inputs: [String: ORTValue] = [io.inputName: inputTensor]
+        let outputs = try session.run(
+            withInputs: inputs,
+            outputNames: io.outputNames,
+            runOptions: nil
+        )
+
+        // Extract output tensors as Float arrays
+        var result: [String: (data: [Float], shape: [Int])] = [:]
+        for (name, value) in outputs {
+            let info = try value.tensorTypeAndShapeInfo()
+            let outputShape = info.shape.map { $0.intValue }
+            let outputData = try value.tensorData() as Data
+            let floats: [Float] = outputData.withUnsafeBytes { buffer in
+                Array(buffer.bindMemory(to: Float.self))
+            }
+
+            if floats.contains(where: \.isNaN) {
+                throw ORTSessionManagerError.outputContainsNaN(name)
+            }
+
+            result[name] = (data: floats, shape: outputShape)
+        }
+
+        return result
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/OpenCVDBBridge.h b/deploy/ios_demo/PaddleOCRDemo/Engine/OpenCVDBBridge.h
new file mode 100644
index 0000000..3745813
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/OpenCVDBBridge.h
@@ -0,0 +1,45 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#import <Foundation/Foundation.h>
+
+NS_ASSUME_NONNULL_BEGIN
+
+@interface PDBMiniBoxResult : NSObject
+@property(nonatomic) NSArray<NSValue *> *corners;
+@property(nonatomic) float minSide;
+@end
+
+@interface PDBOpenCVDBBridge : NSObject
+
++ (NSArray<NSArray<NSValue *> *> *)findContoursWithBinaryMask:(NSData *)data
+                                                        width:(NSInteger)width
+                                                       height:(NSInteger)height
+                                                maxCandidates:
+                                                    (NSInteger)maxCandidates
+    NS_SWIFT_NAME(findContours(binaryMask:width:height:maxCandidates:));
+
++ (nullable PDBMiniBoxResult *)miniBoxFromPoints:(NSArray<NSValue *> *)points
+    NS_SWIFT_NAME(miniBox(fromPoints:));
+
++ (float)contourAreaFromPoints:(NSArray<NSValue *> *)points
+    NS_SWIFT_NAME(contourArea(fromPoints:));
+
++ (float)arcLengthFromPoints:(NSArray<NSValue *> *)points
+                      closed:(BOOL)closed
+    NS_SWIFT_NAME(arcLength(fromPoints:closed:));
+
+@end
+
+NS_ASSUME_NONNULL_END
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/OpenCVDBBridge.mm b/deploy/ios_demo/PaddleOCRDemo/Engine/OpenCVDBBridge.mm
new file mode 100644
index 0000000..c214ff7
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/OpenCVDBBridge.mm
@@ -0,0 +1,147 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#import "OpenCVDBBridge.h"
+// Include imgproc before UIKit: Objective-C's `NO` macro breaks OpenCV enums named `NO` in opencv.hpp.
+#import <opencv2/core.hpp>
+#import <opencv2/imgproc.hpp>
+#import <UIKit/UIKit.h>
+#include <algorithm>
+
+@implementation PDBMiniBoxResult
+@end
+
+@implementation PDBOpenCVDBBridge
+
++ (NSArray<NSArray<NSValue *> *> *)findContoursWithBinaryMask:(NSData *)data
+                                                         width:(NSInteger)width
+                                                        height:(NSInteger)height
+                                                 maxCandidates:(NSInteger)maxCandidates {
+    if (width <= 0 || height <= 0 || data.length < (NSUInteger)(width * height)) {
+        return @[];
+    }
+
+    const uint8_t *src = static_cast<const uint8_t *>(data.bytes);
+    cv::Mat mat((int)height, (int)width, CV_8UC1);
+    for (int y = 0; y < (int)height; y++) {
+        uint8_t *row = mat.ptr<uint8_t>(y);
+        for (int x = 0; x < (int)width; x++) {
+            row[x] = src[y * (int)width + x] ? 255 : 0;
+        }
+    }
+
+    std::vector<std::vector<cv::Point>> contours;
+    cv::findContours(mat, contours, cv::RETR_LIST, cv::CHAIN_APPROX_SIMPLE);
+
+    const NSInteger n = std::min((NSInteger)contours.size(), maxCandidates);
+    NSMutableArray *out = [NSMutableArray arrayWithCapacity:(NSUInteger)n];
+    for (NSInteger i = 0; i < n; i++) {
+        const auto &c = contours[(size_t)i];
+        NSMutableArray *pts = [NSMutableArray arrayWithCapacity:c.size()];
+        for (const auto &pt : c) {
+            [pts addObject:[NSValue valueWithCGPoint:CGPointMake(pt.x, pt.y)]];
+        }
+        [out addObject:pts];
+    }
+    return out;
+}
+
++ (nullable PDBMiniBoxResult *)miniBoxFromPoints:(NSArray<NSValue *> *)points {
+    if (points.count < 2) {
+        return nil;
+    }
+
+    std::vector<cv::Point2f> pts;
+    pts.reserve(points.count);
+    for (NSValue *v in points) {
+        CGPoint p = v.CGPointValue;
+        pts.emplace_back((float)p.x, (float)p.y);
+    }
+
+    cv::RotatedRect rr = cv::minAreaRect(pts);
+    float w = rr.size.width;
+    float h = rr.size.height;
+    if (w <= 0 && h <= 0) {
+        return nil;
+    }
+
+    cv::Point2f cvPts[4];
+    rr.points(cvPts);
+
+    // Stable sort by x only (min-area box corner order).
+    int ord[4] = {0, 1, 2, 3};
+    std::stable_sort(ord, ord + 4, [&](int a, int b) { return cvPts[a].x < cvPts[b].x; });
+    cv::Point2f sortedPts[4];
+    for (int i = 0; i < 4; i++) {
+        sortedPts[i] = cvPts[ord[i]];
+    }
+
+    NSInteger index1, index2, index3, index4;
+    cv::Point2f p0 = sortedPts[0];
+    cv::Point2f p1 = sortedPts[1];
+    if (p1.y > p0.y) {
+        index1 = 0;
+        index4 = 1;
+    } else {
+        index1 = 1;
+        index4 = 0;
+    }
+    cv::Point2f p2 = sortedPts[2];
+    cv::Point2f p3 = sortedPts[3];
+    if (p3.y > p2.y) {
+        index2 = 2;
+        index3 = 3;
+    } else {
+        index2 = 3;
+        index3 = 2;
+    }
+
+    PDBMiniBoxResult *res = [[PDBMiniBoxResult alloc] init];
+    res.corners = @[
+        [NSValue valueWithCGPoint:CGPointMake(sortedPts[index1].x, sortedPts[index1].y)],
+        [NSValue valueWithCGPoint:CGPointMake(sortedPts[index2].x, sortedPts[index2].y)],
+        [NSValue valueWithCGPoint:CGPointMake(sortedPts[index3].x, sortedPts[index3].y)],
+        [NSValue valueWithCGPoint:CGPointMake(sortedPts[index4].x, sortedPts[index4].y)],
+    ];
+    res.minSide = std::min(w, h);
+    return res;
+}
+
++ (float)contourAreaFromPoints:(NSArray<NSValue *> *)points {
+    if (points.count < 2) {
+        return 0;
+    }
+    std::vector<cv::Point2f> pts;
+    pts.reserve(points.count);
+    for (NSValue *v in points) {
+        CGPoint p = v.CGPointValue;
+        pts.emplace_back((float)p.x, (float)p.y);
+    }
+    return cv::contourArea(pts);
+}
+
++ (float)arcLengthFromPoints:(NSArray<NSValue *> *)points closed:(BOOL)closed {
+    if (points.count < 2) {
+        return 0;
+    }
+    std::vector<cv::Point2f> pts;
+    pts.reserve(points.count);
+    for (NSValue *v in points) {
+        CGPoint p = v.CGPointValue;
+        pts.emplace_back((float)p.x, (float)p.y);
+    }
+    return (float)cv::arcLength(pts, closed);
+}
+
+@end
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/OpenCVImageBridge.h b/deploy/ios_demo/PaddleOCRDemo/Engine/OpenCVImageBridge.h
new file mode 100644
index 0000000..0954205
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/OpenCVImageBridge.h
@@ -0,0 +1,63 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#import <Foundation/Foundation.h>
+
+NS_ASSUME_NONNULL_BEGIN
+
+@interface PDBPerspectiveCropOutput : NSObject
+@property(nonatomic) NSData *rgbData;
+@property(nonatomic) NSInteger width;
+@property(nonatomic) NSInteger height;
+@end
+
+@interface PDBOpenCVImageBridge : NSObject
+
+/// `cv::resize(..., INTER_LINEAR)` on 8-bit **3-channel** row-major data
+/// (`CV_8UC3`; stores BGR channel order for OpenCV).
++ (NSData *)resizeRGBU8:(NSData *)data
+               srcWidth:(NSInteger)srcWidth
+              srcHeight:(NSInteger)srcHeight
+               dstWidth:(NSInteger)dstWidth
+              dstHeight:(NSInteger)dstHeight NS_SWIFT_NAME(resizeRGBU8(_:srcWidth:srcHeight:dstWidth:dstHeight:));
+
+/// Pad RGB image to `padWidth` x `padHeight` with zeros on bottom/right.
++ (NSData *)padRGBU8:(NSData *)data
+               width:(NSInteger)width
+              height:(NSInteger)height
+            padWidth:(NSInteger)padWidth
+           padHeight:(NSInteger)padHeight
+    NS_SWIFT_NAME(padRGBU8(_:width:height:padWidth:padHeight:));
+
+/// Minimum-area rectangle, corner ordering, perspective warp (`INTER_CUBIC`,
+/// `BORDER_REPLICATE`), then 90° CCW rotation when height/width ≥ 1.5. Input
+/// buffer is row-major **BGR**.
++ (nullable PDBPerspectiveCropOutput *)
+    quadTextLineCropBGR:(NSData *)src
+               srcWidth:(NSInteger)srcWidth
+              srcHeight:(NSInteger)srcHeight
+                   quad:(NSArray<NSValue *> *)quad
+    NS_SWIFT_NAME(quadTextLineCropBGR(_:srcWidth:srcHeight:quad:));
+
+/// Mean of the float probability map inside the quadrilateral: `fillPoly` mask
+/// then `cv::mean`.
++ (float)meanPredInQuad:(NSData *)pred
+                  width:(NSInteger)width
+                 height:(NSInteger)height
+                    box:(NSArray<NSArray<NSNumber *> *> *)box
+    NS_SWIFT_NAME(meanPredInQuad(_:width:height:box:));
+
+@end
+
+NS_ASSUME_NONNULL_END
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/OpenCVImageBridge.mm b/deploy/ios_demo/PaddleOCRDemo/Engine/OpenCVImageBridge.mm
new file mode 100644
index 0000000..a349c47
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/OpenCVImageBridge.mm
@@ -0,0 +1,205 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#import "OpenCVImageBridge.h"
+#import <algorithm>
+#import <cfloat>
+#import <cmath>
+#import <opencv2/core.hpp>
+#import <opencv2/imgproc.hpp>
+#import <UIKit/UIKit.h>
+
+@implementation PDBPerspectiveCropOutput
+@end
+
+@implementation PDBOpenCVImageBridge
+
++ (NSData *)resizeRGBU8:(NSData *)data
+               srcWidth:(NSInteger)srcWidth
+              srcHeight:(NSInteger)srcHeight
+               dstWidth:(NSInteger)dstWidth
+              dstHeight:(NSInteger)dstHeight {
+    if (srcWidth <= 0 || srcHeight <= 0 || dstWidth <= 0 || dstHeight <= 0) {
+        return [NSData data];
+    }
+    if (data.length < (NSUInteger)(srcWidth * srcHeight * 3)) {
+        return [NSData data];
+    }
+    cv::Mat src((int)srcHeight, (int)srcWidth, CV_8UC3, (void *)data.bytes);
+    cv::Mat dst;
+    cv::resize(src, dst, cv::Size((int)dstWidth, (int)dstHeight), 0, 0, cv::INTER_LINEAR);
+    return [NSData dataWithBytes:dst.data length:(NSUInteger)(dst.rows * dst.cols * 3)];
+}
+
++ (NSData *)padRGBU8:(NSData *)data
+               width:(NSInteger)width
+              height:(NSInteger)height
+            padWidth:(NSInteger)padWidth
+           padHeight:(NSInteger)padHeight {
+    if (width <= 0 || height <= 0 || padWidth < width || padHeight < height) {
+        return data ?: [NSData data];
+    }
+    if (data.length < (NSUInteger)(width * height * 3)) {
+        return [NSData data];
+    }
+    cv::Mat src((int)height, (int)width, CV_8UC3, (void *)data.bytes);
+    int bot = (int)(padHeight - height);
+    int right = (int)(padWidth - width);
+    cv::Mat dst;
+    cv::copyMakeBorder(src, dst, 0, bot, 0, right, cv::BORDER_CONSTANT, cv::Scalar(0, 0, 0));
+    return [NSData dataWithBytes:dst.data length:(NSUInteger)(dst.rows * dst.cols * 3)];
+}
+
++ (nullable PDBPerspectiveCropOutput *)quadTextLineCropBGR:(NSData *)src
+                                                 srcWidth:(NSInteger)srcWidth
+                                                srcHeight:(NSInteger)srcHeight
+                                                     quad:(NSArray<NSValue *> *)quad {
+    if (srcWidth <= 0 || srcHeight <= 0) {
+        return nil;
+    }
+    if (quad.count != 4 || src.length < (NSUInteger)(srcWidth * srcHeight * 3)) {
+        return nil;
+    }
+
+    cv::Mat im((int)srcHeight, (int)srcWidth, CV_8UC3, (void *)src.bytes);
+
+    std::vector<cv::Point2f> raw(4);
+    for (int i = 0; i < 4; i++) {
+        CGPoint p = [quad[i] CGPointValue];
+        raw[i] = cv::Point2f((float)(int32_t)p.x, (float)(int32_t)p.y);
+    }
+
+    cv::RotatedRect rr = cv::minAreaRect(raw);
+    cv::Point2f boxPts[4];
+    rr.points(boxPts);
+    std::vector<cv::Point2f> sorted(boxPts, boxPts + 4);
+    std::sort(sorted.begin(), sorted.end(),
+              [](const cv::Point2f &a, const cv::Point2f &b) { return a.x < b.x; });
+
+    int index_a, index_b, index_c, index_d;
+    if (sorted[1].y > sorted[0].y) {
+        index_a = 0;
+        index_d = 1;
+    } else {
+        index_a = 1;
+        index_d = 0;
+    }
+    if (sorted[3].y > sorted[2].y) {
+        index_b = 2;
+        index_c = 3;
+    } else {
+        index_b = 3;
+        index_c = 2;
+    }
+
+    cv::Point2f srcPts[4] = {sorted[index_a], sorted[index_b], sorted[index_c], sorted[index_d]};
+
+    double w1 = cv::norm(srcPts[0] - srcPts[1]);
+    double w2 = cv::norm(srcPts[2] - srcPts[3]);
+    double h1 = cv::norm(srcPts[0] - srcPts[3]);
+    double h2 = cv::norm(srcPts[1] - srcPts[2]);
+
+    int img_crop_width = (int)std::max(w1, w2);
+    int img_crop_height = (int)std::max(h1, h2);
+    if (img_crop_width <= 0 || img_crop_height <= 0) {
+        return nil;
+    }
+
+    cv::Point2f dstPts[4] = {
+        cv::Point2f(0, 0),
+        cv::Point2f((float)img_crop_width, 0),
+        cv::Point2f((float)img_crop_width, (float)img_crop_height),
+        cv::Point2f(0, (float)img_crop_height),
+    };
+
+    cv::Mat M = cv::getPerspectiveTransform(srcPts, dstPts);
+    cv::Mat warped;
+    cv::warpPerspective(im, warped, M, cv::Size(img_crop_width, img_crop_height), cv::INTER_CUBIC,
+                          cv::BORDER_REPLICATE);
+
+    cv::Mat outMat;
+    double ratio = (double)warped.rows / (double)warped.cols;
+    if (ratio >= 1.5) {
+        cv::rotate(warped, outMat, cv::ROTATE_90_COUNTERCLOCKWISE);
+    } else {
+        outMat = warped;
+    }
+
+    PDBPerspectiveCropOutput *result = [[PDBPerspectiveCropOutput alloc] init];
+    result.rgbData = [NSData dataWithBytes:outMat.data length:(NSUInteger)(outMat.rows * outMat.cols * 3)];
+    result.width = outMat.cols;
+    result.height = outMat.rows;
+    return result;
+}
+
++ (float)meanPredInQuad:(NSData *)pred
+                  width:(NSInteger)width
+                 height:(NSInteger)height
+                    box:(NSArray<NSArray<NSNumber *> *> *)box {
+    if (width <= 0 || height <= 0 || box.count != 4) {
+        return 0;
+    }
+    size_t need = (size_t)(width * height) * sizeof(float);
+    if (pred.length < need) {
+        return 0;
+    }
+
+    float xminF = FLT_MAX, xmaxF = -FLT_MAX, yminF = FLT_MAX, ymaxF = -FLT_MAX;
+    for (NSArray<NSNumber *> *pt in box) {
+        if (pt.count < 2) {
+            return 0;
+        }
+        float x = [pt[0] floatValue];
+        float y = [pt[1] floatValue];
+        xminF = std::min(xminF, x);
+        xmaxF = std::max(xmaxF, x);
+        yminF = std::min(yminF, y);
+        ymaxF = std::max(ymaxF, y);
+    }
+
+    int w = (int)width;
+    int h = (int)height;
+    // Clamp ROI to bitmap bounds before masking and mean (avoids out-of-range reads).
+    int xmin = std::max(0, std::min((int)std::floor(xminF), w - 1));
+    int xmax = std::max(0, std::min((int)std::ceil(xmaxF), w - 1));
+    int ymin = std::max(0, std::min((int)std::floor(yminF), h - 1));
+    int ymax = std::max(0, std::min((int)std::ceil(ymaxF), h - 1));
+    if (xmax < xmin || ymax < ymin) {
+        return 0;
+    }
+
+    const float *p = (const float *)pred.bytes;
+    cv::Mat bitmap(h, w, CV_32FC1, (void *)const_cast<float *>(p));
+    cv::Rect roi(xmin, ymin, xmax - xmin + 1, ymax - ymin + 1);
+    cv::Mat crop = bitmap(roi);
+
+    cv::Mat mask = cv::Mat::zeros(crop.rows, crop.cols, CV_8UC1);
+    std::vector<cv::Point> pts;
+    pts.reserve(4);
+    for (NSArray<NSNumber *> *pt in box) {
+        float fx = [pt[0] floatValue] - static_cast<float>(xmin);
+        float fy = [pt[1] floatValue] - static_cast<float>(ymin);
+        int px = static_cast<int>(fx);
+        int py = static_cast<int>(fy);
+        pts.push_back(cv::Point(px, py));
+    }
+    std::vector<std::vector<cv::Point>> contours;
+    contours.push_back(pts);
+    cv::fillPoly(mask, contours, cv::Scalar(1));
+
+    cv::Scalar m = cv::mean(crop, mask);
+    return (float)m[0];
+}
+
+@end
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/PDBPolygonOffsetBridge.h b/deploy/ios_demo/PaddleOCRDemo/Engine/PDBPolygonOffsetBridge.h
new file mode 100644
index 0000000..4f5de3c
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/PDBPolygonOffsetBridge.h
@@ -0,0 +1,29 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#import <Foundation/Foundation.h>
+
+NS_ASSUME_NONNULL_BEGIN
+
+/// Polygon inflate (unclip) via Clipper 6.x (`ClipperLib`) for closed paths.
+@interface PDBPolygonOffsetBridge : NSObject
+
++ (NSArray<NSArray<NSValue *> *> *)inflateClosedPolygonWith:
+                                       (NSArray<NSValue *> *)points
+                                                   distance:(double)distance
+    NS_SWIFT_NAME(inflateClosedPolygon(with:distance:));
+
+@end
+
+NS_ASSUME_NONNULL_END
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/PDBPolygonOffsetBridge.mm b/deploy/ios_demo/PaddleOCRDemo/Engine/PDBPolygonOffsetBridge.mm
new file mode 100644
index 0000000..75196cd
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/PDBPolygonOffsetBridge.mm
@@ -0,0 +1,49 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#import "PDBPolygonOffsetBridge.h"
+#import <UIKit/UIKit.h>
+#include <cmath>
+#include "clipper.hpp"
+@implementation PDBPolygonOffsetBridge
++ (NSArray<NSArray<NSValue *> *> *)inflateClosedPolygonWith:(NSArray<NSValue *> *)points
+                                                   distance:(double)distance {
+    if (points.count < 2) {
+        return @[];
+    }
+    using namespace ClipperLib;
+    // Default miter limit and arc tolerance for ClipperOffset (round join).
+    ClipperOffset co(2.0, 0.25);
+    Path path;
+    path.reserve(points.count);
+    for (NSValue *v in points) {
+        CGPoint p = v.CGPointValue;
+        cInt ix = static_cast<cInt>(std::lround((double)p.x));
+        cInt iy = static_cast<cInt>(std::lround((double)p.y));
+        path.push_back(IntPoint(ix, iy));
+    }
+    co.AddPath(path, jtRound, etClosedPolygon);
+    Paths solution;
+    co.Execute(solution, distance);
+    if (solution.empty()) {
+        return @[];
+    }
+    const Path &out = solution[0];
+    NSMutableArray *ring = [NSMutableArray arrayWithCapacity:out.size()];
+    for (const IntPoint &pt : out) {
+        [ring addObject:[NSValue valueWithCGPoint:CGPointMake((CGFloat)pt.X, (CGFloat)pt.Y)]];
+    }
+    return @[ ring ];
+}
+@end
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/Preprocessing.swift b/deploy/ios_demo/PaddleOCRDemo/Engine/Preprocessing.swift
new file mode 100644
index 0000000..8bf7b31
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/Preprocessing.swift
@@ -0,0 +1,386 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import CoreGraphics
+import Foundation
+
+// MARK: - Errors
+
+enum DetPreprocessorError: LocalizedError {
+    case invalidImage
+    case configMissing(String)
+    case pixelExtractionFailed
+
+    var errorDescription: String? {
+        switch self {
+        case .invalidImage:
+            return "Invalid input image: width or height is zero"
+        case .configMissing(let detail):
+            return "Required preprocessing config missing: \(detail)"
+        case .pixelExtractionFailed:
+            return "Failed to extract pixel data from image"
+        }
+    }
+}
+
+// MARK: - Preprocessing Result
+
+/// The output of detection preprocessing: a float32 tensor and metadata needed for postprocessing.
+struct PreprocessResult {
+    /// Flat CHW array of shape [1, 3, resizedH, resizedW].
+    let tensorData: [Float]
+    /// Tensor dimensions: [1, 3, resizedH, resizedW].
+    let tensorShape: [Int]
+    /// Original image dimensions before any resizing.
+    let originalSize: (width: Int, height: Int)
+    /// Image dimensions after resize and padding to stride multiples.
+    let resizedSize: (width: Int, height: Int)
+    /// Vertical resize ratio: resizedH / originalH.
+    let ratioH: Float
+    /// Horizontal resize ratio: resizedW / originalW.
+    let ratioW: Float
+}
+
+// MARK: - DetPreprocessor
+
+/// Implements detection preprocessing: **DetResizeForTest** →
+/// **NormalizeImage** (from the model config file) -> **ToCHWImage**.
+///
+/// Rasterization uses CoreGraphics; resize/pad use OpenCV (`INTER_LINEAR`, `copyMakeBorder`).
+/// HWC channel order follows `DecodeImage.img_mode` (`InferenceConfig.decodeImageChannelOrder`).
+struct DetPreprocessor {
+    private enum ResizeRule {
+        case limitSide(len: Int, limitType: String, maxSideLimit: Int)
+        case longEdge(resizeLong: Int)
+    }
+
+    private let scale: Float
+    private let mean: [Float]
+    private let std: [Float]
+    private let channelOrder: InferenceImageChannelOrder
+
+    init(config: InferenceConfig) throws {
+        var foundScale: Float?
+        var foundMean: [Float]?
+        var foundStd: [Float]?
+
+        for op in config.preProcess.transformOps {
+            switch op {
+            case .decodeImage:
+                break
+            case .detResizeForTest(_):
+                break
+            case .normalizeImage(let s, let m, let st, _):
+                foundScale = s
+                foundMean = m
+                foundStd = st
+            case .toCHWImage, .recResizeImg, .unknown:
+                break
+            }
+        }
+
+        guard let scale = foundScale, let mean = foundMean, let std = foundStd else {
+            throw DetPreprocessorError.configMissing("NormalizeImage with scale, mean, std")
+        }
+        guard mean.count == 3, std.count == 3 else {
+            throw DetPreprocessorError.configMissing("NormalizeImage mean/std must have exactly 3 values")
+        }
+
+        self.scale = scale
+        self.mean = mean
+        self.std = std
+        self.channelOrder = config.decodeImageChannelOrder
+    }
+
+    private static func resizeRule(from params: DetResizeParams) throws -> ResizeRule {
+        .limitSide(
+            len: params.limitSideLen,
+            limitType: params.limitType,
+            maxSideLimit: params.maxSideLimit
+        )
+    }
+
+    /// Rasterizes ``CGImage`` to HWC uint8 (`BGR` or `RGB` per `DecodeImage.img_mode`).
+    func preprocess(_ image: CGImage, detResize: DetResizeParams) throws -> PreprocessResult {
+        let originalW = image.width
+        let originalH = image.height
+
+        guard originalW > 0, originalH > 0 else {
+            throw DetPreprocessorError.invalidImage
+        }
+
+        let pixelBytes = try rasterizeImageIntoHWCBuffer(
+            from: image, width: originalW, height: originalH, order: channelOrder
+        )
+        return try preprocessFromHWCPixelBuffer(
+            pixelBytes: pixelBytes,
+            originalWidth: originalW,
+            originalHeight: originalH,
+            detResize: detResize
+        )
+    }
+
+    private func preprocessFromHWCPixelBuffer(
+        pixelBytes initialPixels: [UInt8],
+        originalWidth: Int,
+        originalHeight: Int,
+        detResize: DetResizeParams
+    ) throws -> PreprocessResult {
+        let resizeRule = try Self.resizeRule(from: detResize)
+        var pixelBytes = initialPixels
+        var currentW = originalWidth
+        var currentH = originalHeight
+
+        if currentH + currentW < 64 {
+            let paddedH = max(32, currentH)
+            let paddedW = max(32, currentW)
+            pixelBytes = padImage(pixelBytes, fromW: currentW, fromH: currentH, toW: paddedW, toH: paddedH)
+            currentW = paddedW
+            currentH = paddedH
+        }
+
+        let (resizeW, resizeH, ratioW, ratioH): (Int, Int, Float, Float)
+        switch resizeRule {
+        case .limitSide(let len, let limitType, let maxSideLimit):
+            (resizeW, resizeH, ratioW, ratioH) = computeResizeLimitSide(
+                srcW: currentW,
+                srcH: currentH,
+                limitSideLen: len,
+                limitType: limitType,
+                maxSideLimit: maxSideLimit
+            )
+        case .longEdge(let resizeLong):
+            (resizeW, resizeH, ratioW, ratioH) = computeResizeLongEdge(
+                srcW: currentW,
+                srcH: currentH,
+                resizeLong: resizeLong
+            )
+        }
+
+        let resizedPixels = resizeImageBuffer(
+            pixelBytes, srcW: currentW, srcH: currentH, dstW: resizeW, dstH: resizeH
+        )
+
+        let normalizedHWC = normalizePixels(resizedPixels, width: resizeW, height: resizeH)
+        let chwData = hwcToCHW(normalizedHWC, width: resizeW, height: resizeH)
+
+        return PreprocessResult(
+            tensorData: chwData,
+            tensorShape: [1, 3, resizeH, resizeW],
+            originalSize: (width: originalWidth, height: originalHeight),
+            resizedSize: (width: resizeW, height: resizeH),
+            ratioH: ratioH,
+            ratioW: ratioW
+        )
+    }
+
+    // MARK: - Rasterization
+
+    /// Row-major HWC uint8: **BGR** or **RGB** per `DecodeImage.img_mode` in the model config file.
+    private func rasterizeImageIntoHWCBuffer(
+        from image: CGImage,
+        width: Int,
+        height: Int,
+        order: InferenceImageChannelOrder
+    ) throws -> [UInt8] {
+        let bytesPerPixel = 4
+        let bytesPerRow = width * bytesPerPixel
+        var rgbaData = [UInt8](repeating: 0, count: height * bytesPerRow)
+
+        guard let colorSpace = CGColorSpace(name: CGColorSpace.sRGB),
+              let context = CGContext(
+                  data: &rgbaData,
+                  width: width,
+                  height: height,
+                  bitsPerComponent: 8,
+                  bytesPerRow: bytesPerRow,
+                  space: colorSpace,
+                  bitmapInfo: CGImageAlphaInfo.noneSkipLast.rawValue
+              ) else {
+            throw DetPreprocessorError.pixelExtractionFailed
+        }
+
+        context.draw(image, in: CGRect(x: 0, y: 0, width: width, height: height))
+
+        var out = [UInt8](repeating: 0, count: height * width * 3)
+        for i in 0..<(height * width) {
+            let r = rgbaData[i * 4]
+            let g = rgbaData[i * 4 + 1]
+            let b = rgbaData[i * 4 + 2]
+            switch order {
+            case .bgr:
+                out[i * 3] = b
+                out[i * 3 + 1] = g
+                out[i * 3 + 2] = r
+            case .rgb:
+                out[i * 3] = r
+                out[i * 3 + 1] = g
+                out[i * 3 + 2] = b
+            }
+        }
+        return out
+    }
+
+    private func padImage(_ pixels: [UInt8], fromW: Int, fromH: Int, toW: Int, toH: Int) -> [UInt8] {
+        if fromW == toW && fromH == toH {
+            return pixels
+        }
+        let out = PDBOpenCVImageBridge.padRGBU8(
+            Data(pixels), width: fromW, height: fromH, padWidth: toW, padHeight: toH
+        )
+        let expected = toW * toH * 3
+        guard out.count == expected else {
+            return [UInt8](repeating: 0, count: expected)
+        }
+        return [UInt8](out)
+    }
+
+    // MARK: - Resize geometry
+
+    /// Limit-side resize with multiple-of-32 dimensions, optional cap on the larger side.
+    private func computeResizeLimitSide(
+        srcW: Int,
+        srcH: Int,
+        limitSideLen: Int,
+        limitType: String,
+        maxSideLimit: Int
+    ) -> (width: Int, height: Int, ratioW: Float, ratioH: Float) {
+        let h = Float(srcH)
+        let w = Float(srcW)
+        let ratio: Float
+        switch limitType {
+        case "max":
+            if max(srcH, srcW) > limitSideLen {
+                if srcH > srcW {
+                    ratio = Float(limitSideLen) / h
+                } else {
+                    ratio = Float(limitSideLen) / w
+                }
+            } else {
+                ratio = 1.0
+            }
+        case "min":
+            if min(srcH, srcW) < limitSideLen {
+                if srcH < srcW {
+                    ratio = Float(limitSideLen) / h
+                } else {
+                    ratio = Float(limitSideLen) / w
+                }
+            } else {
+                ratio = 1.0
+            }
+        case "resize_long":
+            ratio = Float(limitSideLen) / Float(max(srcH, srcW))
+        default:
+            ratio = 1.0
+        }
+
+        var resizeH = Int(Float(srcH) * ratio)
+        var resizeW = Int(Float(srcW) * ratio)
+
+        if max(resizeH, resizeW) > maxSideLimit {
+            let shrink = Float(maxSideLimit) / Float(max(resizeH, resizeW))
+            resizeH = Int(Float(resizeH) * shrink)
+            resizeW = Int(Float(resizeW) * shrink)
+        }
+
+        resizeH = max(Int((Double(resizeH) / 32.0).rounded(.toNearestOrEven)) * 32, 32)
+        resizeW = max(Int((Double(resizeW) / 32.0).rounded(.toNearestOrEven)) * 32, 32)
+
+        if resizeH == srcH && resizeW == srcW {
+            return (srcW, srcH, 1.0, 1.0)
+        }
+
+        let ratioH = Float(resizeH) / Float(srcH)
+        let ratioW = Float(resizeW) / Float(srcW)
+        return (resizeW, resizeH, ratioW, ratioH)
+    }
+
+    /// Long-edge resize with stride 128 on both sides.
+    private func computeResizeLongEdge(
+        srcW: Int,
+        srcH: Int,
+        resizeLong: Int
+    ) -> (width: Int, height: Int, ratioW: Float, ratioH: Float) {
+        let ratio: Float
+        if srcH > srcW {
+            ratio = Float(resizeLong) / Float(srcH)
+        } else {
+            ratio = Float(resizeLong) / Float(srcW)
+        }
+
+        var resizeH = Int(Float(srcH) * ratio)
+        var resizeW = Int(Float(srcW) * ratio)
+
+        let stride = 128
+        resizeH = ((resizeH + stride - 1) / stride) * stride
+        resizeW = ((resizeW + stride - 1) / stride) * stride
+
+        let ratioH = Float(resizeH) / Float(srcH)
+        let ratioW = Float(resizeW) / Float(srcW)
+        return (resizeW, resizeH, ratioW, ratioH)
+    }
+
+    private func resizeImageBuffer(
+        _ pixels: [UInt8], srcW: Int, srcH: Int, dstW: Int, dstH: Int
+    ) -> [UInt8] {
+        if srcW == dstW && srcH == dstH {
+            return pixels
+        }
+        let out = PDBOpenCVImageBridge.resizeRGBU8(
+            Data(pixels), srcWidth: srcW, srcHeight: srcH, dstWidth: dstW, dstHeight: dstH
+        )
+        let expected = dstW * dstH * 3
+        guard out.count == expected else {
+            return [UInt8](repeating: 0, count: expected)
+        }
+        return [UInt8](out)
+    }
+
+    // MARK: - NormalizeImage
+
+    private func normalizePixels(_ pixels: [UInt8], width: Int, height: Int) -> [Float] {
+        let pixelCount = width * height
+        var result = [Float](repeating: 0, count: pixelCount * 3)
+
+        let scaleOverStd = [scale / std[0], scale / std[1], scale / std[2]]
+        let meanOverStd = [mean[0] / std[0], mean[1] / std[1], mean[2] / std[2]]
+
+        for i in 0..<pixelCount {
+            let baseIdx = i * 3
+            result[baseIdx] = Float(pixels[baseIdx]) * scaleOverStd[0] - meanOverStd[0]
+            result[baseIdx + 1] = Float(pixels[baseIdx + 1]) * scaleOverStd[1] - meanOverStd[1]
+            result[baseIdx + 2] = Float(pixels[baseIdx + 2]) * scaleOverStd[2] - meanOverStd[2]
+        }
+
+        return result
+    }
+
+    private func hwcToCHW(_ hwcData: [Float], width: Int, height: Int) -> [Float] {
+        let channelSize = height * width
+        var chwData = [Float](repeating: 0, count: 3 * channelSize)
+
+        for y in 0..<height {
+            for x in 0..<width {
+                let hwcIdx = (y * width + x) * 3
+                let pixelOffset = y * width + x
+                chwData[0 * channelSize + pixelOffset] = hwcData[hwcIdx]
+                chwData[1 * channelSize + pixelOffset] = hwcData[hwcIdx + 1]
+                chwData[2 * channelSize + pixelOffset] = hwcData[hwcIdx + 2]
+            }
+        }
+
+        return chwData
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/QuadTextCrop.swift b/deploy/ios_demo/PaddleOCRDemo/Engine/QuadTextCrop.swift
new file mode 100644
index 0000000..ebe2a32
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/QuadTextCrop.swift
@@ -0,0 +1,152 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import CoreGraphics
+import Foundation
+
+// MARK: - BGR ↔ CGImage
+
+/// Builds a ``CGImage`` from row-major BGR uint8 (8-bit three-channel).
+enum BGRImageCodec {
+    static func makeCGImageFromBGR(pixels: [UInt8], width: Int, height: Int) throws -> CGImage {
+        let bytesPerPixel = 4
+        let bytesPerRow = width * bytesPerPixel
+
+        let pixelCount = width * height
+        guard pixels.count == pixelCount * 3 else {
+            throw QuadTextCropError.imageCreationFailed
+        }
+
+        var rgbaData = [UInt8](repeating: 255, count: pixelCount * bytesPerPixel)
+        for i in 0..<pixelCount {
+            let b = pixels[i * 3]
+            let g = pixels[i * 3 + 1]
+            let r = pixels[i * 3 + 2]
+            rgbaData[i * 4] = r
+            rgbaData[i * 4 + 1] = g
+            rgbaData[i * 4 + 2] = b
+        }
+
+        guard let colorSpace = CGColorSpace(name: CGColorSpace.sRGB),
+              let context = CGContext(
+                  data: &rgbaData,
+                  width: width,
+                  height: height,
+                  bitsPerComponent: 8,
+                  bytesPerRow: bytesPerRow,
+                  space: colorSpace,
+                  bitmapInfo: CGImageAlphaInfo.noneSkipLast.rawValue
+              ),
+              let cgImage = context.makeImage() else {
+            throw QuadTextCropError.imageCreationFailed
+        }
+
+        return cgImage
+    }
+}
+
+// MARK: - Errors
+
+enum QuadTextCropError: LocalizedError {
+    case invalidPolygon(String)
+    case imageCreationFailed
+    case openCVFailed
+
+    var errorDescription: String? {
+        switch self {
+        case .invalidPolygon(let detail):
+            return "Invalid polygon for quad text crop: \(detail)"
+        case .imageCreationFailed:
+            return "Failed to create output CGImage"
+        case .openCVFailed:
+            return "OpenCV quad text-line crop failed"
+        }
+    }
+}
+
+// MARK: - QuadTextCrop
+
+/// Minimum-area quadrilateral, corner ordering, perspective warp, then optional 90° CCW rotation when height/width ≥ 1.5.
+struct QuadTextCrop {
+
+    /// Crop a text region from the source image using the detector quadrilateral.
+    static func crop(_ image: CGImage, polygon: [[Int32]]) throws -> CGImage {
+        guard polygon.count == 4, polygon.allSatisfy({ $0.count == 2 }) else {
+            throw QuadTextCropError.invalidPolygon(
+                "Expected 4 points with 2 coordinates each, got \(polygon.count) points"
+            )
+        }
+
+        let srcPixels = try extractBGRPixels(from: image)
+        let srcW = image.width
+        let srcH = image.height
+
+        let polyValues: [NSValue] = polygon.map {
+            NSValue(cgPoint: CGPoint(x: CGFloat($0[0]), y: CGFloat($0[1])))
+        }
+        guard let out = PDBOpenCVImageBridge.quadTextLineCropBGR(
+            Data(srcPixels),
+            srcWidth: srcW,
+            srcHeight: srcH,
+            quad: polyValues
+        ) else {
+            throw QuadTextCropError.openCVFailed
+        }
+
+        let bytes = [UInt8](out.rgbData)
+        return try BGRImageCodec.makeCGImageFromBGR(pixels: bytes, width: Int(out.width), height: Int(out.height))
+    }
+}
+
+// MARK: - Pixel Extraction & Image Creation
+
+private extension QuadTextCrop {
+
+    /// Row-major HWC **BGR**
+    static func extractBGRPixels(from image: CGImage) throws -> [UInt8] {
+        let width = image.width
+        let height = image.height
+        let bytesPerPixel = 4
+        let bytesPerRow = width * bytesPerPixel
+        var rgbaData = [UInt8](repeating: 0, count: height * bytesPerRow)
+
+        guard let colorSpace = CGColorSpace(name: CGColorSpace.sRGB),
+              let context = CGContext(
+                  data: &rgbaData,
+                  width: width,
+                  height: height,
+                  bitsPerComponent: 8,
+                  bytesPerRow: bytesPerRow,
+                  space: colorSpace,
+                  bitmapInfo: CGImageAlphaInfo.noneSkipLast.rawValue
+              ) else {
+            throw QuadTextCropError.imageCreationFailed
+        }
+
+        context.draw(image, in: CGRect(x: 0, y: 0, width: width, height: height))
+
+        let pixelCount = width * height
+        var bgrData = [UInt8](repeating: 0, count: pixelCount * 3)
+        for i in 0..<pixelCount {
+            let r = rgbaData[i * 4]
+            let g = rgbaData[i * 4 + 1]
+            let b = rgbaData[i * 4 + 2]
+            bgrData[i * 3] = b
+            bgrData[i * 3 + 1] = g
+            bgrData[i * 3 + 2] = r
+        }
+
+        return bgrData
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/RecPreprocessor.swift b/deploy/ios_demo/PaddleOCRDemo/Engine/RecPreprocessor.swift
new file mode 100644
index 0000000..73d3f75
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/RecPreprocessor.swift
@@ -0,0 +1,380 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import CoreGraphics
+import Foundation
+
+// MARK: - Errors
+
+enum RecPreprocessorError: LocalizedError {
+    case invalidImage
+    case configMissing(String)
+    case pixelExtractionFailed
+
+    var errorDescription: String? {
+        switch self {
+        case .invalidImage:
+            return "Invalid input image: width or height is zero"
+        case .configMissing(let detail):
+            return "Required recognition preprocessing config missing: \(detail)"
+        case .pixelExtractionFailed:
+            return "Failed to extract pixel data from image"
+        }
+    }
+}
+
+// MARK: - Preprocessing Result
+
+/// The output of recognition preprocessing: a float32 tensor and metadata.
+struct RecPreprocessResult {
+    /// Flat CHW array of shape [1, imgC, imgH, targetW].
+    let tensorData: [Float]
+    /// Tensor dimensions: [1, imgC, imgH, targetW].
+    let tensorShape: [Int]
+    /// Original image dimensions before any resizing.
+    let originalSize: (width: Int, height: Int)
+    /// Actual content width before zero-padding (the resized image width).
+    let resizedWidth: Int
+}
+
+/// Batched recognition input: N × C × H × W with a common width after padding.
+struct RecBatchPreprocessResult {
+    let tensorData: [Float]
+    let tensorShape: [Int]
+}
+
+// MARK: - RecPreprocessor
+
+/// Recognition resize + normalize + pad: dynamic width, fixed height, scale to [-1, 1],
+/// CHW layout, right-pad to target width.
+///
+/// Recognition resize, per-channel normalization, HWC→CHW, and width padding.
+/// `DecodeImage.img_mode` is applied via `InferenceConfig.decodeImageChannelOrder`. Resize uses OpenCV `INTER_LINEAR`.
+///
+/// Set environment variable `PADDLEOCR_REC_ALIGN_WIDTH_TO_8` to `1`, `true`, or `yes` to round the
+/// canvas width up to a multiple of 8 (off by default).
+struct RecPreprocessor {
+    /// Channel count from `RecResizeImg.image_shape` (3).
+    private let imgC: Int
+    /// Target height from config RecResizeImg.image_shape (e.g., 48).
+    private let imgH: Int
+    /// Default width from config RecResizeImg.image_shape (e.g., 320).
+    private let imgW: Int
+    /// Absolute maximum width cap to prevent excessive memory usage.
+    private let maxImgW: Int = 3200
+    /// Channel order from `DecodeImage.img_mode` in the model config file.
+    private let channelOrder: InferenceImageChannelOrder
+
+    /// Creates a RecPreprocessor by extracting image_shape from the RecResizeImg transform op.
+    ///
+    /// - Parameter config: A parsed InferenceConfig from the recognition model's config file.
+    /// - Throws: `RecPreprocessorError.configMissing` if RecResizeImg with image_shape is absent.
+    init(config: InferenceConfig) throws {
+        var foundImageShape: [Int]?
+
+        for op in config.preProcess.transformOps {
+            switch op {
+            case .decodeImage, .detResizeForTest, .normalizeImage, .toCHWImage, .unknown:
+                break
+            case .recResizeImg(let imageShape):
+                foundImageShape = imageShape
+            }
+        }
+
+        guard let imageShape = foundImageShape, imageShape.count >= 3 else {
+            throw RecPreprocessorError.configMissing("RecResizeImg with image_shape")
+        }
+
+        self.imgC = imageShape[0]
+        self.imgH = imageShape[1]
+        self.imgW = imageShape[2]
+        self.channelOrder = config.decodeImageChannelOrder
+    }
+
+    /// Runs recognition preprocessing on a `CGImage`.
+    ///
+    /// Steps: extract pixels → compute dimensions → resize → normalize → HWC-to-CHW → pad
+    ///
+    /// - Parameter image: The input image (typically a cropped text region).
+    /// - Returns: A `RecPreprocessResult` with the float32 CHW tensor and metadata.
+    func preprocess(_ image: CGImage) throws -> RecPreprocessResult {
+        let originalW = image.width
+        let originalH = image.height
+
+        guard originalW > 0, originalH > 0 else {
+            throw RecPreprocessorError.invalidImage
+        }
+
+        let pixelBytes = try extractHWCPixels(from: image, width: originalW, height: originalH, order: channelOrder)
+        let (paddedData, canvasW, resizedW) = try lineResizeNormalizeCHW(
+            pixelBytes: pixelBytes,
+            originalW: originalW,
+            originalH: originalH,
+            alignCanvasTo8: Self.isRecognitionCanvasWidthAlignedTo8Enabled()
+        )
+
+        return RecPreprocessResult(
+            tensorData: paddedData,
+            tensorShape: [1, imgC, imgH, canvasW],
+            originalSize: (width: originalW, height: originalH),
+            resizedWidth: resizedW
+        )
+    }
+
+    /// Builds a batched tensor: each line is resized and padded to its canvas width, then all rows
+    /// are padded on the right to the maximum canvas width in the batch (shape `[N, C, H, W]`).
+    func preprocessBatch(_ images: [CGImage]) throws -> RecBatchPreprocessResult {
+        guard !images.isEmpty else {
+            throw RecPreprocessorError.invalidImage
+        }
+
+        let align8 = Self.isRecognitionCanvasWidthAlignedTo8Enabled()
+        var rowCHW: [[Float]] = []
+        var canvasWidths: [Int] = []
+
+        for image in images {
+            let originalW = image.width
+            let originalH = image.height
+            guard originalW > 0, originalH > 0 else {
+                throw RecPreprocessorError.invalidImage
+            }
+            let pixelBytes = try extractHWCPixels(from: image, width: originalW, height: originalH, order: channelOrder)
+            let (chw, canvasW, _) = try lineResizeNormalizeCHW(
+                pixelBytes: pixelBytes,
+                originalW: originalW,
+                originalH: originalH,
+                alignCanvasTo8: align8
+            )
+            rowCHW.append(chw)
+            canvasWidths.append(canvasW)
+        }
+
+        guard let maxCanvasW = canvasWidths.max() else {
+            throw RecPreprocessorError.invalidImage
+        }
+
+        let n = rowCHW.count
+        let channelSize = imgH * maxCanvasW
+        var flat = [Float](repeating: 0, count: n * imgC * channelSize)
+
+        for (b, row) in rowCHW.enumerated() {
+            let w = canvasWidths[b]
+            let paddedRow: [Float]
+            if w == maxCanvasW {
+                paddedRow = row
+            } else {
+                let contentW = w
+                paddedRow = padToTargetWidth(
+                    row,
+                    contentWidth: contentW,
+                    targetWidth: maxCanvasW,
+                    height: imgH,
+                    channels: imgC
+                )
+            }
+            let batchOffset = b * imgC * channelSize
+            for i in 0..<paddedRow.count {
+                flat[batchOffset + i] = paddedRow[i]
+            }
+        }
+
+        return RecBatchPreprocessResult(
+            tensorData: flat,
+            tensorShape: [n, imgC, imgH, maxCanvasW]
+        )
+    }
+
+    /// When `PADDLEOCR_REC_ALIGN_WIDTH_TO_8` is set to a truthy value, the canvas width is rounded up to a multiple of 8.
+    private static func isRecognitionCanvasWidthAlignedTo8Enabled() -> Bool {
+        let v = ProcessInfo.processInfo.environment["PADDLEOCR_REC_ALIGN_WIDTH_TO_8"]?.lowercased()
+        return v == "1" || v == "true" || v == "yes"
+    }
+
+    /// One text line: HWC uint8 → resize → normalize → CHW, padded to the computed canvas width.
+    private func lineResizeNormalizeCHW(
+        pixelBytes: [UInt8],
+        originalW: Int,
+        originalH: Int,
+        alignCanvasTo8: Bool
+    ) throws -> (chw: [Float], canvasW: Int, resizedW: Int) {
+        let w = originalW
+        let h = originalH
+        let whRatio = Float(w) / Float(h)
+        let defaultRatio = Float(imgW) / Float(imgH)
+        let maxWhRatio = max(defaultRatio, whRatio)
+        var canvasW = Int(Float(imgH) * maxWhRatio)
+        let resizedW: Int
+
+        if canvasW > maxImgW {
+            canvasW = maxImgW
+            resizedW = maxImgW
+        } else {
+            let ratio = Float(w) / Float(h)
+            if Int(ceil(Float(imgH) * ratio)) > canvasW {
+                resizedW = canvasW
+            } else {
+                resizedW = Int(ceil(Float(imgH) * ratio))
+            }
+        }
+
+        if alignCanvasTo8 {
+            canvasW = min(max(roundUpToMultipleOf(canvasW, 8), resizedW), maxImgW)
+        }
+
+        let resizedPixels = resizeWithOpenCV(pixelBytes, srcW: w, srcH: h, dstW: resizedW, dstH: imgH)
+        let normalizedHWC = normalizePixels(resizedPixels, width: resizedW, height: imgH)
+        let chwData = hwcToCHW(normalizedHWC, width: resizedW, height: imgH)
+        let paddedData = padToTargetWidth(
+            chwData,
+            contentWidth: resizedW,
+            targetWidth: canvasW,
+            height: imgH,
+            channels: imgC
+        )
+        return (paddedData, canvasW, resizedW)
+    }
+
+    // MARK: - Step 0: Pixel extraction (BGR / RGB)
+
+    private func extractHWCPixels(
+        from image: CGImage,
+        width: Int,
+        height: Int,
+        order: InferenceImageChannelOrder
+    ) throws -> [UInt8] {
+        let bytesPerPixel = 4
+        let bytesPerRow = width * bytesPerPixel
+        var rgbaData = [UInt8](repeating: 0, count: height * bytesPerRow)
+
+        guard let colorSpace = CGColorSpace(name: CGColorSpace.sRGB),
+              let context = CGContext(
+                  data: &rgbaData,
+                  width: width,
+                  height: height,
+                  bitsPerComponent: 8,
+                  bytesPerRow: bytesPerRow,
+                  space: colorSpace,
+                  bitmapInfo: CGImageAlphaInfo.noneSkipLast.rawValue
+              ) else {
+            throw RecPreprocessorError.pixelExtractionFailed
+        }
+
+        context.draw(image, in: CGRect(x: 0, y: 0, width: width, height: height))
+
+        var out = [UInt8](repeating: 0, count: height * width * 3)
+        for i in 0..<(height * width) {
+            let r = rgbaData[i * 4]
+            let g = rgbaData[i * 4 + 1]
+            let b = rgbaData[i * 4 + 2]
+            switch order {
+            case .bgr:
+                out[i * 3] = b
+                out[i * 3 + 1] = g
+                out[i * 3 + 2] = r
+            case .rgb:
+                out[i * 3] = r
+                out[i * 3 + 1] = g
+                out[i * 3 + 2] = b
+            }
+        }
+        return out
+    }
+
+    // MARK: - Resize (OpenCV)
+
+    private func resizeWithOpenCV(
+        _ pixels: [UInt8], srcW: Int, srcH: Int, dstW: Int, dstH: Int
+    ) -> [UInt8] {
+        if srcW == dstW && srcH == dstH {
+            return pixels
+        }
+        let out = PDBOpenCVImageBridge.resizeRGBU8(
+            Data(pixels), srcWidth: srcW, srcHeight: srcH, dstWidth: dstW, dstHeight: dstH
+        )
+        let expected = dstW * dstH * 3
+        guard out.count == expected else {
+            return [UInt8](repeating: 0, count: expected)
+        }
+        return [UInt8](out)
+    }
+
+    // MARK: - Normalize (recognition)
+
+    /// Maps pixel values from [0, 255] to [-1, 1] via `pixel / 127.5 - 1.0`.
+    private func normalizePixels(_ pixels: [UInt8], width: Int, height: Int) -> [Float] {
+        let count = width * height * 3
+        var result = [Float](repeating: 0, count: count)
+
+        for i in 0..<count {
+            result[i] = Float(pixels[i]) / 127.5 - 1.0
+        }
+
+        return result
+    }
+
+    // MARK: - HWC → CHW
+
+    private func hwcToCHW(_ hwcData: [Float], width: Int, height: Int) -> [Float] {
+        let channelSize = height * width
+        var chwData = [Float](repeating: 0, count: 3 * channelSize)
+
+        for y in 0..<height {
+            for x in 0..<width {
+                let hwcIdx = (y * width + x) * 3
+                let pixelOffset = y * width + x
+                chwData[0 * channelSize + pixelOffset] = hwcData[hwcIdx]
+                chwData[1 * channelSize + pixelOffset] = hwcData[hwcIdx + 1]
+                chwData[2 * channelSize + pixelOffset] = hwcData[hwcIdx + 2]
+            }
+        }
+
+        return chwData
+    }
+
+    // MARK: - Zero-Padding
+
+    /// Right-pads a CHW tensor from contentWidth to targetWidth with zeros.
+    private func padToTargetWidth(
+        _ chwData: [Float],
+        contentWidth: Int,
+        targetWidth: Int,
+        height: Int,
+        channels: Int
+    ) -> [Float] {
+        if contentWidth == targetWidth {
+            return chwData
+        }
+
+        let targetChannelSize = height * targetWidth
+        let sourceChannelSize = height * contentWidth
+        var padded = [Float](repeating: 0, count: channels * targetChannelSize)
+
+        for c in 0..<channels {
+            for y in 0..<height {
+                let srcOffset = c * sourceChannelSize + y * contentWidth
+                let dstOffset = c * targetChannelSize + y * targetWidth
+                for x in 0..<contentWidth {
+                    padded[dstOffset + x] = chwData[srcOffset + x]
+                }
+            }
+        }
+
+        return padded
+    }
+}
+
+private func roundUpToMultipleOf(_ value: Int, _ multiple: Int) -> Int {
+    guard multiple > 0 else { return value }
+    return (value + multiple - 1) / multiple * multiple
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Engine/RecognitionEngine.swift b/deploy/ios_demo/PaddleOCRDemo/Engine/RecognitionEngine.swift
new file mode 100644
index 0000000..9b7713e
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Engine/RecognitionEngine.swift
@@ -0,0 +1,165 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import CoreGraphics
+import Foundation
+
+// MARK: - Recognition Engine Result
+
+/// The result of running text recognition on a cropped text image.
+///
+/// Contains the decoded text, confidence score, and per-stage timing metrics.
+struct RecognitionEngineResult {
+    /// The decoded text string from CTC decoding.
+    let text: String
+    /// Average confidence across all decoded characters (0.0 to 1.0).
+    let confidence: Float
+    /// Seconds spent in preprocessing (resize, normalize, HWC-to-CHW, pad).
+    let preprocessTime: TimeInterval
+    /// Seconds spent in ONNX Runtime inference.
+    let inferenceTime: TimeInterval
+    /// Seconds spent in CTC decoding (best-class pick, dedup, char mapping).
+    let postprocessTime: TimeInterval
+    /// Total recognition time (preprocess + inference + postprocess).
+    let totalTime: TimeInterval
+    /// Recognition model input tensor shape for the batch this line belonged to.
+    let inputTensorShape: [Int]
+}
+
+// MARK: - Recognition Engine Errors
+
+enum RecognitionEngineError: LocalizedError {
+    case noOutputTensor
+    case unexpectedOutputShape([Int])
+    case batchOutputCountMismatch(expectedBatch: Int, shape: [Int])
+
+    var errorDescription: String? {
+        switch self {
+        case .noOutputTensor:
+            return "Recognition model produced no output tensor"
+        case .unexpectedOutputShape(let shape):
+            return "Unexpected recognition output shape: \(shape), expected [1, T, C] or [N, T, C]"
+        case .batchOutputCountMismatch(let expectedBatch, let shape):
+            return "Recognition batch size mismatch: expected batch dimension \(expectedBatch), got shape \(shape)"
+        }
+    }
+}
+
+// MARK: - RecognitionEngine
+
+/// Orchestrates text recognition:
+/// CGImage → RecPreprocessor → ORT inference → CTCDecoder.
+///
+/// This is the integration layer that composes `RecPreprocessor`, `ORTSessionManager`,
+/// and `CTCDecoder` into a single callable unit. All preprocessing parameters are read
+/// from the recognition model config file at initialization time.
+///
+/// Usage:
+/// ```swift
+/// let engine = try RecognitionEngine(sessionManager: manager)
+/// let result = try await engine.recognize(croppedTextImage)
+/// print("Text: \(result.text) (confidence: \(result.confidence))")
+/// ```
+class RecognitionEngine {
+    private let sessionManager: ORTSessionManager
+    private let preprocessor: RecPreprocessor
+    private let decoder: CTCDecoder
+    private(set) var modelConfig: InferenceConfig
+
+    /// Initialize with an existing ORTSessionManager (models must already be loaded).
+    ///
+    /// Loads the recognition model config file from the bundle to configure
+    /// the preprocessor and CTC decoder.
+    ///
+    /// - Parameter sessionManager: A loaded ORTSessionManager with recognition model ready.
+    /// - Throws: If the model config cannot be loaded or required fields are missing.
+    init(sessionManager: ORTSessionManager) throws {
+        self.sessionManager = sessionManager
+
+        let recPaths = try ModelConfig.recognition()
+        let cfg = try InferenceConfig.load(from: recPaths.configPath)
+        self.modelConfig = cfg
+
+        self.preprocessor = try RecPreprocessor(config: cfg)
+        self.decoder = try CTCDecoder(config: cfg)
+    }
+
+    /// Run recognition on a cropped text region CGImage, returning decoded text with confidence.
+    ///
+    /// Steps: RecPreprocessor → ORT inference → CTCDecoder
+    ///
+    /// - Parameter image: A cropped text region image.
+    /// - Returns: A `RecognitionEngineResult` with text, confidence, and per-stage timing.
+    func recognize(_ image: CGImage) async throws -> RecognitionEngineResult {
+        let batch = try await recognizeBatch([image])
+        guard let first = batch.first else {
+            throw RecognitionEngineError.noOutputTensor
+        }
+        return first
+    }
+
+    /// Runs recognition on multiple cropped lines in one batched inference call.
+    ///
+    /// - Parameter images: Cropped text regions in inference order (caller controls ordering).
+    /// - Returns: One result per input image, same order.
+    func recognizeBatch(_ images: [CGImage]) async throws -> [RecognitionEngineResult] {
+        guard !images.isEmpty else { return [] }
+
+        let preprocessStart = CFAbsoluteTimeGetCurrent()
+        let batchPreprocessed = try preprocessor.preprocessBatch(images)
+        let preprocessTime = CFAbsoluteTimeGetCurrent() - preprocessStart
+
+        let inferenceStart = CFAbsoluteTimeGetCurrent()
+        let outputs = try await sessionManager.runRecognition(
+            inputData: batchPreprocessed.tensorData,
+            shape: batchPreprocessed.tensorShape
+        )
+        let inferenceTime = CFAbsoluteTimeGetCurrent() - inferenceStart
+
+        guard let firstOutput = outputs.values.first else {
+            throw RecognitionEngineError.noOutputTensor
+        }
+        let outputData = firstOutput.data
+        let outputShape = firstOutput.shape
+
+        guard outputShape.count == 3, outputShape[0] == images.count else {
+            if outputShape.count == 3 {
+                throw RecognitionEngineError.batchOutputCountMismatch(expectedBatch: images.count, shape: outputShape)
+            }
+            throw RecognitionEngineError.unexpectedOutputShape(outputShape)
+        }
+
+        let postprocessStart = CFAbsoluteTimeGetCurrent()
+        let decoded = try decoder.decodeBatch(outputData: outputData, outputShape: outputShape)
+        let postprocessTime = CFAbsoluteTimeGetCurrent() - postprocessStart
+
+        let n = Double(images.count)
+        let preprocessPer = preprocessTime / n
+        let inferencePer = inferenceTime / n
+        let postPer = postprocessTime / n
+        let totalPer = (preprocessTime + inferenceTime + postprocessTime) / n
+
+        return decoded.map { d in
+            RecognitionEngineResult(
+                text: d.text,
+                confidence: d.confidence,
+                preprocessTime: preprocessPer,
+                inferenceTime: inferencePer,
+                postprocessTime: postPer,
+                totalTime: totalPer,
+                inputTensorShape: batchPreprocessed.tensorShape
+            )
+        }
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AccentColor.colorset/Contents.json b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AccentColor.colorset/Contents.json
new file mode 100644
index 0000000..075c56a
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AccentColor.colorset/Contents.json
@@ -0,0 +1,20 @@
+{
+  "colors" : [
+    {
+      "color" : {
+        "color-space" : "srgb",
+        "components" : {
+          "alpha" : "1.000",
+          "blue" : "0.753",
+          "green" : "0.396",
+          "red" : "0.082"
+        }
+      },
+      "idiom" : "universal"
+    }
+  ],
+  "info" : {
+    "author" : "xcode",
+    "version" : 1
+  }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Contents.json b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Contents.json
new file mode 100644
index 0000000..015c64f
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Contents.json
@@ -0,0 +1,62 @@
+{
+  "images": [
+    {
+      "filename": "Icon-20@2x.png",
+      "idiom": "iphone",
+      "scale": "2x",
+      "size": "20x20"
+    },
+    {
+      "filename": "Icon-20@3x.png",
+      "idiom": "iphone",
+      "scale": "3x",
+      "size": "20x20"
+    },
+    {
+      "filename": "Icon-29@2x.png",
+      "idiom": "iphone",
+      "scale": "2x",
+      "size": "29x29"
+    },
+    {
+      "filename": "Icon-29@3x.png",
+      "idiom": "iphone",
+      "scale": "3x",
+      "size": "29x29"
+    },
+    {
+      "filename": "Icon-40@2x.png",
+      "idiom": "iphone",
+      "scale": "2x",
+      "size": "40x40"
+    },
+    {
+      "filename": "Icon-40@3x.png",
+      "idiom": "iphone",
+      "scale": "3x",
+      "size": "40x40"
+    },
+    {
+      "filename": "Icon-60@2x.png",
+      "idiom": "iphone",
+      "scale": "2x",
+      "size": "60x60"
+    },
+    {
+      "filename": "Icon-60@3x.png",
+      "idiom": "iphone",
+      "scale": "3x",
+      "size": "60x60"
+    },
+    {
+      "filename": "Icon-1024.png",
+      "idiom": "universal",
+      "platform": "ios",
+      "size": "1024x1024"
+    }
+  ],
+  "info": {
+    "author": "xcode",
+    "version": 1
+  }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-1024.png b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-1024.png
new file mode 100644
index 0000000..5cbc32e
Binary files /dev/null and b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-1024.png differ
diff --git a/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-20@2x.png b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-20@2x.png
new file mode 100644
index 0000000..8de1117
Binary files /dev/null and b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-20@2x.png differ
diff --git a/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-20@3x.png b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-20@3x.png
new file mode 100644
index 0000000..3aa6a12
Binary files /dev/null and b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-20@3x.png differ
diff --git a/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-29@2x.png b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-29@2x.png
new file mode 100644
index 0000000..78920df
Binary files /dev/null and b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-29@2x.png differ
diff --git a/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-29@3x.png b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-29@3x.png
new file mode 100644
index 0000000..5161539
Binary files /dev/null and b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-29@3x.png differ
diff --git a/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-40@2x.png b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-40@2x.png
new file mode 100644
index 0000000..3fb3e0e
Binary files /dev/null and b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-40@2x.png differ
diff --git a/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-40@3x.png b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-40@3x.png
new file mode 100644
index 0000000..efdf5d1
Binary files /dev/null and b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-40@3x.png differ
diff --git a/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-60@2x.png b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-60@2x.png
new file mode 100644
index 0000000..efdf5d1
Binary files /dev/null and b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-60@2x.png differ
diff --git a/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-60@3x.png b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-60@3x.png
new file mode 100644
index 0000000..01e2fc9
Binary files /dev/null and b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/AppIcon.appiconset/Icon-60@3x.png differ
diff --git a/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/Contents.json b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/Contents.json
new file mode 100644
index 0000000..73c0059
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Resources/Assets.xcassets/Contents.json
@@ -0,0 +1,6 @@
+{
+  "info" : {
+    "author" : "xcode",
+    "version" : 1
+  }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Resources/SampleImages/general_ocr_002.jpg b/deploy/ios_demo/PaddleOCRDemo/Resources/SampleImages/general_ocr_002.jpg
new file mode 100644
index 0000000..5c3329a
Binary files /dev/null and b/deploy/ios_demo/PaddleOCRDemo/Resources/SampleImages/general_ocr_002.jpg differ
diff --git a/deploy/ios_demo/PaddleOCRDemo/Supporting/Info.plist b/deploy/ios_demo/PaddleOCRDemo/Supporting/Info.plist
new file mode 100644
index 0000000..915af7e
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Supporting/Info.plist
@@ -0,0 +1,37 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+	<key>CFBundleDevelopmentRegion</key>
+	<string>en</string>
+	<key>CFBundleExecutable</key>
+	<string>$(EXECUTABLE_NAME)</string>
+	<key>CFBundleIdentifier</key>
+	<string>$(PRODUCT_BUNDLE_IDENTIFIER)</string>
+	<key>CFBundleInfoDictionaryVersion</key>
+	<string>6.0</string>
+	<key>CFBundleDisplayName</key>
+	<string>PaddleOCR</string>
+	<key>CFBundleName</key>
+	<string>$(PRODUCT_NAME)</string>
+	<key>CFBundlePackageType</key>
+	<string>APPL</string>
+	<key>CFBundleShortVersionString</key>
+	<string>1.0</string>
+	<key>CFBundleVersion</key>
+	<string>1</string>
+	<key>LSRequiresIPhoneOS</key>
+	<true/>
+	<key>UIApplicationSceneManifest</key>
+	<dict>
+		<key>UIApplicationSupportsMultipleScenes</key>
+		<false/>
+	</dict>
+	<key>UILaunchScreen</key>
+	<dict/>
+	<key>UISupportedInterfaceOrientations</key>
+	<array>
+		<string>UIInterfaceOrientationPortrait</string>
+	</array>
+</dict>
+</plist>
diff --git a/deploy/ios_demo/PaddleOCRDemo/Supporting/PaddleOCRDemo-Bridging-Header.h b/deploy/ios_demo/PaddleOCRDemo/Supporting/PaddleOCRDemo-Bridging-Header.h
new file mode 100644
index 0000000..f563b6b
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Supporting/PaddleOCRDemo-Bridging-Header.h
@@ -0,0 +1,24 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#ifndef BridgingHeader_h
+#define BridgingHeader_h
+
+#import "ORTProfilingBridge.h"
+#import "OpenCVDBBridge.h"
+#import "OpenCVImageBridge.h"
+#import "PDBPolygonOffsetBridge.h"
+#import <onnxruntime.h>
+
+#endif
diff --git a/deploy/ios_demo/PaddleOCRDemo/ThirdParty/Clipper1/LICENSE b/deploy/ios_demo/PaddleOCRDemo/ThirdParty/Clipper1/LICENSE
new file mode 100644
index 0000000..a17d377
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/ThirdParty/Clipper1/LICENSE
@@ -0,0 +1,23 @@
+Boost Software License - Version 1.0 - August 17th, 2003
+
+Permission is hereby granted, free of charge, to any person or organization
+obtaining a copy of the software and accompanying documentation covered by
+this license (the "Software") to use, reproduce, display, distribute,
+execute, and transmit the Software, and to prepare derivative works of the
+Software, and to permit third parties to whom the Software is furnished to
+do so, all subject to the following:
+
+The copyright notices in the Software and this entire statement, including
+the above license grant, this restriction and the following disclaimer,
+must be included in all copies of the Software, in whole or in part, and
+all derivative works of the Software, unless such copies or derivative
+works are solely in the form of machine-executable object code generated by
+a source language processor.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE, TITLE AND NON-INFRINGEMENT. IN NO EVENT
+SHALL THE COPYRIGHT HOLDERS OR ANYONE DISTRIBUTING THE SOFTWARE BE LIABLE
+FOR ANY DAMAGES OR OTHER LIABILITY, WHETHER IN CONTRACT, TORT OR OTHERWISE,
+ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+DEALINGS IN THE SOFTWARE.
diff --git a/deploy/ios_demo/PaddleOCRDemo/ThirdParty/Clipper1/clipper.cpp b/deploy/ios_demo/PaddleOCRDemo/ThirdParty/Clipper1/clipper.cpp
new file mode 100644
index 0000000..2f7c050
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/ThirdParty/Clipper1/clipper.cpp
@@ -0,0 +1,4377 @@
+/*******************************************************************************
+ *                                                                              *
+ * Author    :  Angus Johnson * Version   :  6.4.2 * Date      :  27 February
+ *2017                                                * Website   :
+ *http://www.angusj.com                                           * Copyright :
+ *Angus Johnson 2010-2017                                         *
+ *                                                                              *
+ * License: * Use, modification & distribution is subject to Boost Software
+ *License Ver 1. * http://www.boost.org/LICENSE_1_0.txt *
+ *                                                                              *
+ * Attributions: * The code in this library is an extension of Bala Vatti's
+ *clipping algorithm: * "A generic solution to polygon clipping" *
+ * Communications of the ACM, Vol 35, Issue 7 (July 1992) pp 56-63. *
+ * http://portal.acm.org/citation.cfm?id=129906 *
+ *                                                                              *
+ * Computer graphics and geometric modeling: implementation and algorithms * By
+ *Max K. Agoston                                                            *
+ * Springer; 1 edition (January 4, 2005) *
+ * http://books.google.com/books?q=vatti+clipping+agoston *
+ *                                                                              *
+ * See also: * "Polygon Offsetting by Computing Winding Numbers" * Paper no.
+ *DETC2005-85513 pp. 565-575                                         * ASME 2005
+ *International Design Engineering Technical Conferences             * and
+ *Computers and Information in Engineering Conference (IDETC/CIE2005)      *
+ * September 24-28, 2005 , Long Beach, California, USA *
+ * http://www.me.berkeley.edu/~mcmains/pubs/DAC05OffsetPolygon.pdf *
+ *                                                                              *
+ *******************************************************************************/
+
+/*******************************************************************************
+ *                                                                              *
+ * This is a translation of the Delphi Clipper library and the naming style *
+ * used has retained a Delphi flavour. *
+ *                                                                              *
+ *******************************************************************************/
+
+#include "clipper.hpp"
+#include <algorithm>
+#include <cmath>
+#include <cstdlib>
+#include <cstring>
+#include <functional>
+#include <ostream>
+#include <stdexcept>
+#include <vector>
+
+namespace ClipperLib {
+
+static double const pi = 3.141592653589793238;
+static double const two_pi = pi * 2;
+static double const def_arc_tolerance = 0.25;
+
+enum Direction { dRightToLeft, dLeftToRight };
+
+static int const Unassigned = -1; // edge not currently 'owning' a solution
+static int const Skip = -2;       // edge that would otherwise close a path
+
+#define HORIZONTAL (-1.0E+40)
+#define TOLERANCE (1.0e-20)
+#define NEAR_ZERO(val) (((val) > -TOLERANCE) && ((val) < TOLERANCE))
+
+struct TEdge {
+  IntPoint Bot;
+  IntPoint Curr; // current (updated for every new scanbeam)
+  IntPoint Top;
+  double Dx;
+  PolyType PolyTyp;
+  EdgeSide Side; // side only refers to current side of solution poly
+  int WindDelta; // 1 or -1 depending on winding direction
+  int WindCnt;
+  int WindCnt2; // winding count of the opposite polytype
+  int OutIdx;
+  TEdge *Next;
+  TEdge *Prev;
+  TEdge *NextInLML;
+  TEdge *NextInAEL;
+  TEdge *PrevInAEL;
+  TEdge *NextInSEL;
+  TEdge *PrevInSEL;
+};
+
+struct IntersectNode {
+  TEdge *Edge1;
+  TEdge *Edge2;
+  IntPoint Pt;
+};
+
+struct LocalMinimum {
+  cInt Y;
+  TEdge *LeftBound;
+  TEdge *RightBound;
+};
+
+struct OutPt;
+
+// OutRec: contains a path in the clipping solution. Edges in the AEL will
+// carry a pointer to an OutRec when they are part of the clipping solution.
+struct OutRec {
+  int Idx;
+  bool IsHole;
+  bool IsOpen;
+  OutRec *FirstLeft; // see comments in clipper.pas
+  PolyNode *PolyNd;
+  OutPt *Pts;
+  OutPt *BottomPt;
+};
+
+struct OutPt {
+  int Idx;
+  IntPoint Pt;
+  OutPt *Next;
+  OutPt *Prev;
+};
+
+struct Join {
+  OutPt *OutPt1;
+  OutPt *OutPt2;
+  IntPoint OffPt;
+};
+
+struct LocMinSorter {
+  inline bool operator()(const LocalMinimum &locMin1,
+                         const LocalMinimum &locMin2) {
+    return locMin2.Y < locMin1.Y;
+  }
+};
+
+//------------------------------------------------------------------------------
+//------------------------------------------------------------------------------
+
+inline cInt Round(double val) {
+  if ((val < 0))
+    return static_cast<cInt>(val - 0.5);
+  else
+    return static_cast<cInt>(val + 0.5);
+}
+//------------------------------------------------------------------------------
+
+inline cInt Abs(cInt val) { return val < 0 ? -val : val; }
+
+//------------------------------------------------------------------------------
+// PolyTree methods ...
+//------------------------------------------------------------------------------
+
+void PolyTree::Clear() {
+  for (PolyNodes::size_type i = 0; i < AllNodes.size(); ++i)
+    delete AllNodes[i];
+  AllNodes.resize(0);
+  Childs.resize(0);
+}
+//------------------------------------------------------------------------------
+
+PolyNode *PolyTree::GetFirst() const {
+  if (!Childs.empty())
+    return Childs[0];
+  else
+    return 0;
+}
+//------------------------------------------------------------------------------
+
+int PolyTree::Total() const {
+  int result = (int)AllNodes.size();
+  // with negative offsets, ignore the hidden outer polygon ...
+  if (result > 0 && Childs[0] != AllNodes[0])
+    result--;
+  return result;
+}
+
+//------------------------------------------------------------------------------
+// PolyNode methods ...
+//------------------------------------------------------------------------------
+
+PolyNode::PolyNode() : Parent(0), Index(0), m_IsOpen(false) {}
+//------------------------------------------------------------------------------
+
+int PolyNode::ChildCount() const { return (int)Childs.size(); }
+//------------------------------------------------------------------------------
+
+void PolyNode::AddChild(PolyNode &child) {
+  unsigned cnt = (unsigned)Childs.size();
+  Childs.push_back(&child);
+  child.Parent = this;
+  child.Index = cnt;
+}
+//------------------------------------------------------------------------------
+
+PolyNode *PolyNode::GetNext() const {
+  if (!Childs.empty())
+    return Childs[0];
+  else
+    return GetNextSiblingUp();
+}
+//------------------------------------------------------------------------------
+
+PolyNode *PolyNode::GetNextSiblingUp() const {
+  if (!Parent) // protects against PolyTree.GetNextSiblingUp()
+    return 0;
+  else if (Index == Parent->Childs.size() - 1)
+    return Parent->GetNextSiblingUp();
+  else
+    return Parent->Childs[Index + 1];
+}
+//------------------------------------------------------------------------------
+
+bool PolyNode::IsHole() const {
+  bool result = true;
+  PolyNode *node = Parent;
+  while (node) {
+    result = !result;
+    node = node->Parent;
+  }
+  return result;
+}
+//------------------------------------------------------------------------------
+
+bool PolyNode::IsOpen() const { return m_IsOpen; }
+//------------------------------------------------------------------------------
+
+#ifndef use_int32
+
+//------------------------------------------------------------------------------
+// Int128 class (enables safe math on signed 64bit integers)
+// eg Int128 val1((long64)9223372036854775807); //ie 2^63 -1
+//    Int128 val2((long64)9223372036854775807);
+//    Int128 val3 = val1 * val2;
+//    val3.AsString => "85070591730234615847396907784232501249" (8.5e+37)
+//------------------------------------------------------------------------------
+
+class Int128 {
+public:
+  ulong64 lo;
+  long64 hi;
+
+  Int128(long64 _lo = 0) {
+    lo = (ulong64)_lo;
+    if (_lo < 0)
+      hi = -1;
+    else
+      hi = 0;
+  }
+
+  Int128(const Int128 &val) : lo(val.lo), hi(val.hi) {}
+
+  Int128(const long64 &_hi, const ulong64 &_lo) : lo(_lo), hi(_hi) {}
+
+  Int128 &operator=(const long64 &val) {
+    lo = (ulong64)val;
+    if (val < 0)
+      hi = -1;
+    else
+      hi = 0;
+    return *this;
+  }
+
+  bool operator==(const Int128 &val) const {
+    return (hi == val.hi && lo == val.lo);
+  }
+
+  bool operator!=(const Int128 &val) const { return !(*this == val); }
+
+  bool operator>(const Int128 &val) const {
+    if (hi != val.hi)
+      return hi > val.hi;
+    else
+      return lo > val.lo;
+  }
+
+  bool operator<(const Int128 &val) const {
+    if (hi != val.hi)
+      return hi < val.hi;
+    else
+      return lo < val.lo;
+  }
+
+  bool operator>=(const Int128 &val) const { return !(*this < val); }
+
+  bool operator<=(const Int128 &val) const { return !(*this > val); }
+
+  Int128 &operator+=(const Int128 &rhs) {
+    hi += rhs.hi;
+    lo += rhs.lo;
+    if (lo < rhs.lo)
+      hi++;
+    return *this;
+  }
+
+  Int128 operator+(const Int128 &rhs) const {
+    Int128 result(*this);
+    result += rhs;
+    return result;
+  }
+
+  Int128 &operator-=(const Int128 &rhs) {
+    *this += -rhs;
+    return *this;
+  }
+
+  Int128 operator-(const Int128 &rhs) const {
+    Int128 result(*this);
+    result -= rhs;
+    return result;
+  }
+
+  Int128 operator-() const // unary negation
+  {
+    if (lo == 0)
+      return Int128(-hi, 0);
+    else
+      return Int128(~hi, ~lo + 1);
+  }
+
+  operator double() const {
+    const double shift64 = 18446744073709551616.0; // 2^64
+    if (hi < 0) {
+      if (lo == 0)
+        return (double)hi * shift64;
+      else
+        return -(double)(~lo + ~hi * shift64);
+    } else
+      return (double)(lo + hi * shift64);
+  }
+};
+//------------------------------------------------------------------------------
+
+Int128 Int128Mul(long64 lhs, long64 rhs) {
+  bool negate = (lhs < 0) != (rhs < 0);
+
+  if (lhs < 0)
+    lhs = -lhs;
+  ulong64 int1Hi = ulong64(lhs) >> 32;
+  ulong64 int1Lo = ulong64(lhs & 0xFFFFFFFF);
+
+  if (rhs < 0)
+    rhs = -rhs;
+  ulong64 int2Hi = ulong64(rhs) >> 32;
+  ulong64 int2Lo = ulong64(rhs & 0xFFFFFFFF);
+
+  // nb: see comments in clipper.pas
+  ulong64 a = int1Hi * int2Hi;
+  ulong64 b = int1Lo * int2Lo;
+  ulong64 c = int1Hi * int2Lo + int1Lo * int2Hi;
+
+  Int128 tmp;
+  tmp.hi = long64(a + (c >> 32));
+  tmp.lo = long64(c << 32);
+  tmp.lo += long64(b);
+  if (tmp.lo < b)
+    tmp.hi++;
+  if (negate)
+    tmp = -tmp;
+  return tmp;
+};
+#endif
+
+//------------------------------------------------------------------------------
+// Miscellaneous global functions
+//------------------------------------------------------------------------------
+
+bool Orientation(const Path &poly) { return Area(poly) >= 0; }
+//------------------------------------------------------------------------------
+
+double Area(const Path &poly) {
+  int size = (int)poly.size();
+  if (size < 3)
+    return 0;
+
+  double a = 0;
+  for (int i = 0, j = size - 1; i < size; ++i) {
+    a += ((double)poly[j].X + poly[i].X) * ((double)poly[j].Y - poly[i].Y);
+    j = i;
+  }
+  return -a * 0.5;
+}
+//------------------------------------------------------------------------------
+
+double Area(const OutPt *op) {
+  const OutPt *startOp = op;
+  if (!op)
+    return 0;
+  double a = 0;
+  do {
+    a += (double)(op->Prev->Pt.X + op->Pt.X) *
+         (double)(op->Prev->Pt.Y - op->Pt.Y);
+    op = op->Next;
+  } while (op != startOp);
+  return a * 0.5;
+}
+//------------------------------------------------------------------------------
+
+double Area(const OutRec &outRec) { return Area(outRec.Pts); }
+//------------------------------------------------------------------------------
+
+bool PointIsVertex(const IntPoint &Pt, OutPt *pp) {
+  OutPt *pp2 = pp;
+  do {
+    if (pp2->Pt == Pt)
+      return true;
+    pp2 = pp2->Next;
+  } while (pp2 != pp);
+  return false;
+}
+//------------------------------------------------------------------------------
+
+// See "The Point in Polygon Problem for Arbitrary Polygons" by Hormann &
+// Agathos
+// http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.88.5498&rep=rep1&type=pdf
+int PointInPolygon(const IntPoint &pt, const Path &path) {
+  // returns 0 if false, +1 if true, -1 if pt ON polygon boundary
+  int result = 0;
+  size_t cnt = path.size();
+  if (cnt < 3)
+    return 0;
+  IntPoint ip = path[0];
+  for (size_t i = 1; i <= cnt; ++i) {
+    IntPoint ipNext = (i == cnt ? path[0] : path[i]);
+    if (ipNext.Y == pt.Y) {
+      if ((ipNext.X == pt.X) ||
+          (ip.Y == pt.Y && ((ipNext.X > pt.X) == (ip.X < pt.X))))
+        return -1;
+    }
+    if ((ip.Y < pt.Y) != (ipNext.Y < pt.Y)) {
+      if (ip.X >= pt.X) {
+        if (ipNext.X > pt.X)
+          result = 1 - result;
+        else {
+          double d = (double)(ip.X - pt.X) * (ipNext.Y - pt.Y) -
+                     (double)(ipNext.X - pt.X) * (ip.Y - pt.Y);
+          if (!d)
+            return -1;
+          if ((d > 0) == (ipNext.Y > ip.Y))
+            result = 1 - result;
+        }
+      } else {
+        if (ipNext.X > pt.X) {
+          double d = (double)(ip.X - pt.X) * (ipNext.Y - pt.Y) -
+                     (double)(ipNext.X - pt.X) * (ip.Y - pt.Y);
+          if (!d)
+            return -1;
+          if ((d > 0) == (ipNext.Y > ip.Y))
+            result = 1 - result;
+        }
+      }
+    }
+    ip = ipNext;
+  }
+  return result;
+}
+//------------------------------------------------------------------------------
+
+int PointInPolygon(const IntPoint &pt, OutPt *op) {
+  // returns 0 if false, +1 if true, -1 if pt ON polygon boundary
+  int result = 0;
+  OutPt *startOp = op;
+  for (;;) {
+    if (op->Next->Pt.Y == pt.Y) {
+      if ((op->Next->Pt.X == pt.X) ||
+          (op->Pt.Y == pt.Y && ((op->Next->Pt.X > pt.X) == (op->Pt.X < pt.X))))
+        return -1;
+    }
+    if ((op->Pt.Y < pt.Y) != (op->Next->Pt.Y < pt.Y)) {
+      if (op->Pt.X >= pt.X) {
+        if (op->Next->Pt.X > pt.X)
+          result = 1 - result;
+        else {
+          double d = (double)(op->Pt.X - pt.X) * (op->Next->Pt.Y - pt.Y) -
+                     (double)(op->Next->Pt.X - pt.X) * (op->Pt.Y - pt.Y);
+          if (!d)
+            return -1;
+          if ((d > 0) == (op->Next->Pt.Y > op->Pt.Y))
+            result = 1 - result;
+        }
+      } else {
+        if (op->Next->Pt.X > pt.X) {
+          double d = (double)(op->Pt.X - pt.X) * (op->Next->Pt.Y - pt.Y) -
+                     (double)(op->Next->Pt.X - pt.X) * (op->Pt.Y - pt.Y);
+          if (!d)
+            return -1;
+          if ((d > 0) == (op->Next->Pt.Y > op->Pt.Y))
+            result = 1 - result;
+        }
+      }
+    }
+    op = op->Next;
+    if (startOp == op)
+      break;
+  }
+  return result;
+}
+//------------------------------------------------------------------------------
+
+bool Poly2ContainsPoly1(OutPt *OutPt1, OutPt *OutPt2) {
+  OutPt *op = OutPt1;
+  do {
+    // nb: PointInPolygon returns 0 if false, +1 if true, -1 if pt on polygon
+    int res = PointInPolygon(op->Pt, OutPt2);
+    if (res >= 0)
+      return res > 0;
+    op = op->Next;
+  } while (op != OutPt1);
+  return true;
+}
+//----------------------------------------------------------------------
+
+bool SlopesEqual(const TEdge &e1, const TEdge &e2, bool UseFullInt64Range) {
+#ifndef use_int32
+  if (UseFullInt64Range)
+    return Int128Mul(e1.Top.Y - e1.Bot.Y, e2.Top.X - e2.Bot.X) ==
+           Int128Mul(e1.Top.X - e1.Bot.X, e2.Top.Y - e2.Bot.Y);
+  else
+#endif
+    return (e1.Top.Y - e1.Bot.Y) * (e2.Top.X - e2.Bot.X) ==
+           (e1.Top.X - e1.Bot.X) * (e2.Top.Y - e2.Bot.Y);
+}
+//------------------------------------------------------------------------------
+
+bool SlopesEqual(const IntPoint pt1, const IntPoint pt2, const IntPoint pt3,
+                 bool UseFullInt64Range) {
+#ifndef use_int32
+  if (UseFullInt64Range)
+    return Int128Mul(pt1.Y - pt2.Y, pt2.X - pt3.X) ==
+           Int128Mul(pt1.X - pt2.X, pt2.Y - pt3.Y);
+  else
+#endif
+    return (pt1.Y - pt2.Y) * (pt2.X - pt3.X) ==
+           (pt1.X - pt2.X) * (pt2.Y - pt3.Y);
+}
+//------------------------------------------------------------------------------
+
+bool SlopesEqual(const IntPoint pt1, const IntPoint pt2, const IntPoint pt3,
+                 const IntPoint pt4, bool UseFullInt64Range) {
+#ifndef use_int32
+  if (UseFullInt64Range)
+    return Int128Mul(pt1.Y - pt2.Y, pt3.X - pt4.X) ==
+           Int128Mul(pt1.X - pt2.X, pt3.Y - pt4.Y);
+  else
+#endif
+    return (pt1.Y - pt2.Y) * (pt3.X - pt4.X) ==
+           (pt1.X - pt2.X) * (pt3.Y - pt4.Y);
+}
+//------------------------------------------------------------------------------
+
+inline bool IsHorizontal(TEdge &e) { return e.Dx == HORIZONTAL; }
+//------------------------------------------------------------------------------
+
+inline double GetDx(const IntPoint pt1, const IntPoint pt2) {
+  return (pt1.Y == pt2.Y) ? HORIZONTAL
+                          : (double)(pt2.X - pt1.X) / (pt2.Y - pt1.Y);
+}
+//---------------------------------------------------------------------------
+
+inline void SetDx(TEdge &e) {
+  cInt dy = (e.Top.Y - e.Bot.Y);
+  if (dy == 0)
+    e.Dx = HORIZONTAL;
+  else
+    e.Dx = (double)(e.Top.X - e.Bot.X) / dy;
+}
+//---------------------------------------------------------------------------
+
+inline void SwapSides(TEdge &Edge1, TEdge &Edge2) {
+  EdgeSide Side = Edge1.Side;
+  Edge1.Side = Edge2.Side;
+  Edge2.Side = Side;
+}
+//------------------------------------------------------------------------------
+
+inline void SwapPolyIndexes(TEdge &Edge1, TEdge &Edge2) {
+  int OutIdx = Edge1.OutIdx;
+  Edge1.OutIdx = Edge2.OutIdx;
+  Edge2.OutIdx = OutIdx;
+}
+//------------------------------------------------------------------------------
+
+inline cInt TopX(TEdge &edge, const cInt currentY) {
+  return (currentY == edge.Top.Y)
+             ? edge.Top.X
+             : edge.Bot.X + Round(edge.Dx * (currentY - edge.Bot.Y));
+}
+//------------------------------------------------------------------------------
+
+void IntersectPoint(TEdge &Edge1, TEdge &Edge2, IntPoint &ip) {
+#ifdef use_xyz
+  ip.Z = 0;
+#endif
+
+  double b1, b2;
+  if (Edge1.Dx == Edge2.Dx) {
+    ip.Y = Edge1.Curr.Y;
+    ip.X = TopX(Edge1, ip.Y);
+    return;
+  } else if (Edge1.Dx == 0) {
+    ip.X = Edge1.Bot.X;
+    if (IsHorizontal(Edge2))
+      ip.Y = Edge2.Bot.Y;
+    else {
+      b2 = Edge2.Bot.Y - (Edge2.Bot.X / Edge2.Dx);
+      ip.Y = Round(ip.X / Edge2.Dx + b2);
+    }
+  } else if (Edge2.Dx == 0) {
+    ip.X = Edge2.Bot.X;
+    if (IsHorizontal(Edge1))
+      ip.Y = Edge1.Bot.Y;
+    else {
+      b1 = Edge1.Bot.Y - (Edge1.Bot.X / Edge1.Dx);
+      ip.Y = Round(ip.X / Edge1.Dx + b1);
+    }
+  } else {
+    b1 = Edge1.Bot.X - Edge1.Bot.Y * Edge1.Dx;
+    b2 = Edge2.Bot.X - Edge2.Bot.Y * Edge2.Dx;
+    double q = (b2 - b1) / (Edge1.Dx - Edge2.Dx);
+    ip.Y = Round(q);
+    if (std::fabs(Edge1.Dx) < std::fabs(Edge2.Dx))
+      ip.X = Round(Edge1.Dx * q + b1);
+    else
+      ip.X = Round(Edge2.Dx * q + b2);
+  }
+
+  if (ip.Y < Edge1.Top.Y || ip.Y < Edge2.Top.Y) {
+    if (Edge1.Top.Y > Edge2.Top.Y)
+      ip.Y = Edge1.Top.Y;
+    else
+      ip.Y = Edge2.Top.Y;
+    if (std::fabs(Edge1.Dx) < std::fabs(Edge2.Dx))
+      ip.X = TopX(Edge1, ip.Y);
+    else
+      ip.X = TopX(Edge2, ip.Y);
+  }
+  // finally, don't allow 'ip' to be BELOW curr.Y (ie bottom of scanbeam) ...
+  if (ip.Y > Edge1.Curr.Y) {
+    ip.Y = Edge1.Curr.Y;
+    // use the more vertical edge to derive X ...
+    if (std::fabs(Edge1.Dx) > std::fabs(Edge2.Dx))
+      ip.X = TopX(Edge2, ip.Y);
+    else
+      ip.X = TopX(Edge1, ip.Y);
+  }
+}
+//------------------------------------------------------------------------------
+
+void ReversePolyPtLinks(OutPt *pp) {
+  if (!pp)
+    return;
+  OutPt *pp1, *pp2;
+  pp1 = pp;
+  do {
+    pp2 = pp1->Next;
+    pp1->Next = pp1->Prev;
+    pp1->Prev = pp2;
+    pp1 = pp2;
+  } while (pp1 != pp);
+}
+//------------------------------------------------------------------------------
+
+void DisposeOutPts(OutPt *&pp) {
+  if (pp == 0)
+    return;
+  pp->Prev->Next = 0;
+  while (pp) {
+    OutPt *tmpPp = pp;
+    pp = pp->Next;
+    delete tmpPp;
+  }
+}
+//------------------------------------------------------------------------------
+
+inline void InitEdge(TEdge *e, TEdge *eNext, TEdge *ePrev, const IntPoint &Pt) {
+  std::memset(e, 0, sizeof(TEdge));
+  e->Next = eNext;
+  e->Prev = ePrev;
+  e->Curr = Pt;
+  e->OutIdx = Unassigned;
+}
+//------------------------------------------------------------------------------
+
+void InitEdge2(TEdge &e, PolyType Pt) {
+  if (e.Curr.Y >= e.Next->Curr.Y) {
+    e.Bot = e.Curr;
+    e.Top = e.Next->Curr;
+  } else {
+    e.Top = e.Curr;
+    e.Bot = e.Next->Curr;
+  }
+  SetDx(e);
+  e.PolyTyp = Pt;
+}
+//------------------------------------------------------------------------------
+
+TEdge *RemoveEdge(TEdge *e) {
+  // removes e from double_linked_list (but without removing from memory)
+  e->Prev->Next = e->Next;
+  e->Next->Prev = e->Prev;
+  TEdge *result = e->Next;
+  e->Prev = 0; // flag as removed (see ClipperBase.Clear)
+  return result;
+}
+//------------------------------------------------------------------------------
+
+inline void ReverseHorizontal(TEdge &e) {
+  // swap horizontal edges' Top and Bottom x's so they follow the natural
+  // progression of the bounds - ie so their xbots will align with the
+  // adjoining lower edge. [Helpful in the ProcessHorizontal() method.]
+  std::swap(e.Top.X, e.Bot.X);
+#ifdef use_xyz
+  std::swap(e.Top.Z, e.Bot.Z);
+#endif
+}
+//------------------------------------------------------------------------------
+
+void SwapPoints(IntPoint &pt1, IntPoint &pt2) {
+  IntPoint tmp = pt1;
+  pt1 = pt2;
+  pt2 = tmp;
+}
+//------------------------------------------------------------------------------
+
+bool GetOverlapSegment(IntPoint pt1a, IntPoint pt1b, IntPoint pt2a,
+                       IntPoint pt2b, IntPoint &pt1, IntPoint &pt2) {
+  // precondition: segments are Collinear.
+  if (Abs(pt1a.X - pt1b.X) > Abs(pt1a.Y - pt1b.Y)) {
+    if (pt1a.X > pt1b.X)
+      SwapPoints(pt1a, pt1b);
+    if (pt2a.X > pt2b.X)
+      SwapPoints(pt2a, pt2b);
+    if (pt1a.X > pt2a.X)
+      pt1 = pt1a;
+    else
+      pt1 = pt2a;
+    if (pt1b.X < pt2b.X)
+      pt2 = pt1b;
+    else
+      pt2 = pt2b;
+    return pt1.X < pt2.X;
+  } else {
+    if (pt1a.Y < pt1b.Y)
+      SwapPoints(pt1a, pt1b);
+    if (pt2a.Y < pt2b.Y)
+      SwapPoints(pt2a, pt2b);
+    if (pt1a.Y < pt2a.Y)
+      pt1 = pt1a;
+    else
+      pt1 = pt2a;
+    if (pt1b.Y > pt2b.Y)
+      pt2 = pt1b;
+    else
+      pt2 = pt2b;
+    return pt1.Y > pt2.Y;
+  }
+}
+//------------------------------------------------------------------------------
+
+bool FirstIsBottomPt(const OutPt *btmPt1, const OutPt *btmPt2) {
+  OutPt *p = btmPt1->Prev;
+  while ((p->Pt == btmPt1->Pt) && (p != btmPt1))
+    p = p->Prev;
+  double dx1p = std::fabs(GetDx(btmPt1->Pt, p->Pt));
+  p = btmPt1->Next;
+  while ((p->Pt == btmPt1->Pt) && (p != btmPt1))
+    p = p->Next;
+  double dx1n = std::fabs(GetDx(btmPt1->Pt, p->Pt));
+
+  p = btmPt2->Prev;
+  while ((p->Pt == btmPt2->Pt) && (p != btmPt2))
+    p = p->Prev;
+  double dx2p = std::fabs(GetDx(btmPt2->Pt, p->Pt));
+  p = btmPt2->Next;
+  while ((p->Pt == btmPt2->Pt) && (p != btmPt2))
+    p = p->Next;
+  double dx2n = std::fabs(GetDx(btmPt2->Pt, p->Pt));
+
+  if (std::max(dx1p, dx1n) == std::max(dx2p, dx2n) &&
+      std::min(dx1p, dx1n) == std::min(dx2p, dx2n))
+    return Area(btmPt1) > 0; // if otherwise identical use orientation
+  else
+    return (dx1p >= dx2p && dx1p >= dx2n) || (dx1n >= dx2p && dx1n >= dx2n);
+}
+//------------------------------------------------------------------------------
+
+OutPt *GetBottomPt(OutPt *pp) {
+  OutPt *dups = 0;
+  OutPt *p = pp->Next;
+  while (p != pp) {
+    if (p->Pt.Y > pp->Pt.Y) {
+      pp = p;
+      dups = 0;
+    } else if (p->Pt.Y == pp->Pt.Y && p->Pt.X <= pp->Pt.X) {
+      if (p->Pt.X < pp->Pt.X) {
+        dups = 0;
+        pp = p;
+      } else {
+        if (p->Next != pp && p->Prev != pp)
+          dups = p;
+      }
+    }
+    p = p->Next;
+  }
+  if (dups) {
+    // there appears to be at least 2 vertices at BottomPt so ...
+    while (dups != p) {
+      if (!FirstIsBottomPt(p, dups))
+        pp = dups;
+      dups = dups->Next;
+      while (dups->Pt != pp->Pt)
+        dups = dups->Next;
+    }
+  }
+  return pp;
+}
+//------------------------------------------------------------------------------
+
+bool Pt2IsBetweenPt1AndPt3(const IntPoint pt1, const IntPoint pt2,
+                           const IntPoint pt3) {
+  if ((pt1 == pt3) || (pt1 == pt2) || (pt3 == pt2))
+    return false;
+  else if (pt1.X != pt3.X)
+    return (pt2.X > pt1.X) == (pt2.X < pt3.X);
+  else
+    return (pt2.Y > pt1.Y) == (pt2.Y < pt3.Y);
+}
+//------------------------------------------------------------------------------
+
+bool HorzSegmentsOverlap(cInt seg1a, cInt seg1b, cInt seg2a, cInt seg2b) {
+  if (seg1a > seg1b)
+    std::swap(seg1a, seg1b);
+  if (seg2a > seg2b)
+    std::swap(seg2a, seg2b);
+  return (seg1a < seg2b) && (seg2a < seg1b);
+}
+
+//------------------------------------------------------------------------------
+// ClipperBase class methods ...
+//------------------------------------------------------------------------------
+
+ClipperBase::ClipperBase() // constructor
+{
+  m_CurrentLM = m_MinimaList.begin(); // begin() == end() here
+  m_UseFullRange = false;
+}
+//------------------------------------------------------------------------------
+
+ClipperBase::~ClipperBase() // destructor
+{
+  Clear();
+}
+//------------------------------------------------------------------------------
+
+void RangeTest(const IntPoint &Pt, bool &useFullRange) {
+  if (useFullRange) {
+    if (Pt.X > hiRange || Pt.Y > hiRange || -Pt.X > hiRange || -Pt.Y > hiRange)
+      throw clipperException("Coordinate outside allowed range");
+  } else if (Pt.X > loRange || Pt.Y > loRange || -Pt.X > loRange ||
+             -Pt.Y > loRange) {
+    useFullRange = true;
+    RangeTest(Pt, useFullRange);
+  }
+}
+//------------------------------------------------------------------------------
+
+TEdge *FindNextLocMin(TEdge *E) {
+  for (;;) {
+    while (E->Bot != E->Prev->Bot || E->Curr == E->Top)
+      E = E->Next;
+    if (!IsHorizontal(*E) && !IsHorizontal(*E->Prev))
+      break;
+    while (IsHorizontal(*E->Prev))
+      E = E->Prev;
+    TEdge *E2 = E;
+    while (IsHorizontal(*E))
+      E = E->Next;
+    if (E->Top.Y == E->Prev->Bot.Y)
+      continue; // ie just an intermediate horz.
+    if (E2->Prev->Bot.X < E->Bot.X)
+      E = E2;
+    break;
+  }
+  return E;
+}
+//------------------------------------------------------------------------------
+
+TEdge *ClipperBase::ProcessBound(TEdge *E, bool NextIsForward) {
+  TEdge *Result = E;
+  TEdge *Horz = 0;
+
+  if (E->OutIdx == Skip) {
+    // if edges still remain in the current bound beyond the skip edge then
+    // create another LocMin and call ProcessBound once more
+    if (NextIsForward) {
+      while (E->Top.Y == E->Next->Bot.Y)
+        E = E->Next;
+      // don't include top horizontals when parsing a bound a second time,
+      // they will be contained in the opposite bound ...
+      while (E != Result && IsHorizontal(*E))
+        E = E->Prev;
+    } else {
+      while (E->Top.Y == E->Prev->Bot.Y)
+        E = E->Prev;
+      while (E != Result && IsHorizontal(*E))
+        E = E->Next;
+    }
+
+    if (E == Result) {
+      if (NextIsForward)
+        Result = E->Next;
+      else
+        Result = E->Prev;
+    } else {
+      // there are more edges in the bound beyond result starting with E
+      if (NextIsForward)
+        E = Result->Next;
+      else
+        E = Result->Prev;
+      MinimaList::value_type locMin;
+      locMin.Y = E->Bot.Y;
+      locMin.LeftBound = 0;
+      locMin.RightBound = E;
+      E->WindDelta = 0;
+      Result = ProcessBound(E, NextIsForward);
+      m_MinimaList.push_back(locMin);
+    }
+    return Result;
+  }
+
+  TEdge *EStart;
+
+  if (IsHorizontal(*E)) {
+    // We need to be careful with open paths because this may not be a
+    // true local minima (ie E may be following a skip edge).
+    // Also, consecutive horz. edges may start heading left before going right.
+    if (NextIsForward)
+      EStart = E->Prev;
+    else
+      EStart = E->Next;
+    if (IsHorizontal(*EStart)) // ie an adjoining horizontal skip edge
+    {
+      if (EStart->Bot.X != E->Bot.X && EStart->Top.X != E->Bot.X)
+        ReverseHorizontal(*E);
+    } else if (EStart->Bot.X != E->Bot.X)
+      ReverseHorizontal(*E);
+  }
+
+  EStart = E;
+  if (NextIsForward) {
+    while (Result->Top.Y == Result->Next->Bot.Y && Result->Next->OutIdx != Skip)
+      Result = Result->Next;
+    if (IsHorizontal(*Result) && Result->Next->OutIdx != Skip) {
+      // nb: at the top of a bound, horizontals are added to the bound
+      // only when the preceding edge attaches to the horizontal's left vertex
+      // unless a Skip edge is encountered when that becomes the top divide
+      Horz = Result;
+      while (IsHorizontal(*Horz->Prev))
+        Horz = Horz->Prev;
+      if (Horz->Prev->Top.X > Result->Next->Top.X)
+        Result = Horz->Prev;
+    }
+    while (E != Result) {
+      E->NextInLML = E->Next;
+      if (IsHorizontal(*E) && E != EStart && E->Bot.X != E->Prev->Top.X)
+        ReverseHorizontal(*E);
+      E = E->Next;
+    }
+    if (IsHorizontal(*E) && E != EStart && E->Bot.X != E->Prev->Top.X)
+      ReverseHorizontal(*E);
+    Result = Result->Next; // move to the edge just beyond current bound
+  } else {
+    while (Result->Top.Y == Result->Prev->Bot.Y && Result->Prev->OutIdx != Skip)
+      Result = Result->Prev;
+    if (IsHorizontal(*Result) && Result->Prev->OutIdx != Skip) {
+      Horz = Result;
+      while (IsHorizontal(*Horz->Next))
+        Horz = Horz->Next;
+      if (Horz->Next->Top.X == Result->Prev->Top.X ||
+          Horz->Next->Top.X > Result->Prev->Top.X)
+        Result = Horz->Next;
+    }
+
+    while (E != Result) {
+      E->NextInLML = E->Prev;
+      if (IsHorizontal(*E) && E != EStart && E->Bot.X != E->Next->Top.X)
+        ReverseHorizontal(*E);
+      E = E->Prev;
+    }
+    if (IsHorizontal(*E) && E != EStart && E->Bot.X != E->Next->Top.X)
+      ReverseHorizontal(*E);
+    Result = Result->Prev; // move to the edge just beyond current bound
+  }
+
+  return Result;
+}
+//------------------------------------------------------------------------------
+
+bool ClipperBase::AddPath(const Path &pg, PolyType PolyTyp, bool Closed) {
+#ifdef use_lines
+  if (!Closed && PolyTyp == ptClip)
+    throw clipperException("AddPath: Open paths must be subject.");
+#else
+  if (!Closed)
+    throw clipperException("AddPath: Open paths have been disabled.");
+#endif
+
+  int highI = (int)pg.size() - 1;
+  if (Closed)
+    while (highI > 0 && (pg[highI] == pg[0]))
+      --highI;
+  while (highI > 0 && (pg[highI] == pg[highI - 1]))
+    --highI;
+  if ((Closed && highI < 2) || (!Closed && highI < 1))
+    return false;
+
+  // create a new edge array ...
+  TEdge *edges = new TEdge[highI + 1];
+
+  bool IsFlat = true;
+  // 1. Basic (first) edge initialization ...
+  try {
+    edges[1].Curr = pg[1];
+    RangeTest(pg[0], m_UseFullRange);
+    RangeTest(pg[highI], m_UseFullRange);
+    InitEdge(&edges[0], &edges[1], &edges[highI], pg[0]);
+    InitEdge(&edges[highI], &edges[0], &edges[highI - 1], pg[highI]);
+    for (int i = highI - 1; i >= 1; --i) {
+      RangeTest(pg[i], m_UseFullRange);
+      InitEdge(&edges[i], &edges[i + 1], &edges[i - 1], pg[i]);
+    }
+  } catch (...) {
+    delete[] edges;
+    throw; // range test fails
+  }
+  TEdge *eStart = &edges[0];
+
+  // 2. Remove duplicate vertices, and (when closed) collinear edges ...
+  TEdge *E = eStart, *eLoopStop = eStart;
+  for (;;) {
+    // nb: allows matching start and end points when not Closed ...
+    if (E->Curr == E->Next->Curr && (Closed || E->Next != eStart)) {
+      if (E == E->Next)
+        break;
+      if (E == eStart)
+        eStart = E->Next;
+      E = RemoveEdge(E);
+      eLoopStop = E;
+      continue;
+    }
+    if (E->Prev == E->Next)
+      break; // only two vertices
+    else if (Closed &&
+             SlopesEqual(E->Prev->Curr, E->Curr, E->Next->Curr,
+                         m_UseFullRange) &&
+             (!m_PreserveCollinear ||
+              !Pt2IsBetweenPt1AndPt3(E->Prev->Curr, E->Curr, E->Next->Curr))) {
+      // Collinear edges are allowed for open paths but in closed paths
+      // the default is to merge adjacent collinear edges into a single edge.
+      // However, if the PreserveCollinear property is enabled, only overlapping
+      // collinear edges (ie spikes) will be removed from closed paths.
+      if (E == eStart)
+        eStart = E->Next;
+      E = RemoveEdge(E);
+      E = E->Prev;
+      eLoopStop = E;
+      continue;
+    }
+    E = E->Next;
+    if ((E == eLoopStop) || (!Closed && E->Next == eStart))
+      break;
+  }
+
+  if ((!Closed && (E == E->Next)) || (Closed && (E->Prev == E->Next))) {
+    delete[] edges;
+    return false;
+  }
+
+  if (!Closed) {
+    m_HasOpenPaths = true;
+    eStart->Prev->OutIdx = Skip;
+  }
+
+  // 3. Do second stage of edge initialization ...
+  E = eStart;
+  do {
+    InitEdge2(*E, PolyTyp);
+    E = E->Next;
+    if (IsFlat && E->Curr.Y != eStart->Curr.Y)
+      IsFlat = false;
+  } while (E != eStart);
+
+  // 4. Finally, add edge bounds to LocalMinima list ...
+
+  // Totally flat paths must be handled differently when adding them
+  // to LocalMinima list to avoid endless loops etc ...
+  if (IsFlat) {
+    if (Closed) {
+      delete[] edges;
+      return false;
+    }
+    E->Prev->OutIdx = Skip;
+    MinimaList::value_type locMin;
+    locMin.Y = E->Bot.Y;
+    locMin.LeftBound = 0;
+    locMin.RightBound = E;
+    locMin.RightBound->Side = esRight;
+    locMin.RightBound->WindDelta = 0;
+    for (;;) {
+      if (E->Bot.X != E->Prev->Top.X)
+        ReverseHorizontal(*E);
+      if (E->Next->OutIdx == Skip)
+        break;
+      E->NextInLML = E->Next;
+      E = E->Next;
+    }
+    m_MinimaList.push_back(locMin);
+    m_edges.push_back(edges);
+    return true;
+  }
+
+  m_edges.push_back(edges);
+  bool leftBoundIsForward;
+  TEdge *EMin = 0;
+
+  // workaround to avoid an endless loop in the while loop below when
+  // open paths have matching start and end points ...
+  if (E->Prev->Bot == E->Prev->Top)
+    E = E->Next;
+
+  for (;;) {
+    E = FindNextLocMin(E);
+    if (E == EMin)
+      break;
+    else if (!EMin)
+      EMin = E;
+
+    // E and E.Prev now share a local minima (left aligned if horizontal).
+    // Compare their slopes to find which starts which bound ...
+    MinimaList::value_type locMin;
+    locMin.Y = E->Bot.Y;
+    if (E->Dx < E->Prev->Dx) {
+      locMin.LeftBound = E->Prev;
+      locMin.RightBound = E;
+      leftBoundIsForward = false; // Q.nextInLML = Q.prev
+    } else {
+      locMin.LeftBound = E;
+      locMin.RightBound = E->Prev;
+      leftBoundIsForward = true; // Q.nextInLML = Q.next
+    }
+
+    if (!Closed)
+      locMin.LeftBound->WindDelta = 0;
+    else if (locMin.LeftBound->Next == locMin.RightBound)
+      locMin.LeftBound->WindDelta = -1;
+    else
+      locMin.LeftBound->WindDelta = 1;
+    locMin.RightBound->WindDelta = -locMin.LeftBound->WindDelta;
+
+    E = ProcessBound(locMin.LeftBound, leftBoundIsForward);
+    if (E->OutIdx == Skip)
+      E = ProcessBound(E, leftBoundIsForward);
+
+    TEdge *E2 = ProcessBound(locMin.RightBound, !leftBoundIsForward);
+    if (E2->OutIdx == Skip)
+      E2 = ProcessBound(E2, !leftBoundIsForward);
+
+    if (locMin.LeftBound->OutIdx == Skip)
+      locMin.LeftBound = 0;
+    else if (locMin.RightBound->OutIdx == Skip)
+      locMin.RightBound = 0;
+    m_MinimaList.push_back(locMin);
+    if (!leftBoundIsForward)
+      E = E2;
+  }
+  return true;
+}
+//------------------------------------------------------------------------------
+
+bool ClipperBase::AddPaths(const Paths &ppg, PolyType PolyTyp, bool Closed) {
+  bool result = false;
+  for (Paths::size_type i = 0; i < ppg.size(); ++i)
+    if (AddPath(ppg[i], PolyTyp, Closed))
+      result = true;
+  return result;
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::Clear() {
+  DisposeLocalMinimaList();
+  for (EdgeList::size_type i = 0; i < m_edges.size(); ++i) {
+    TEdge *edges = m_edges[i];
+    delete[] edges;
+  }
+  m_edges.clear();
+  m_UseFullRange = false;
+  m_HasOpenPaths = false;
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::Reset() {
+  m_CurrentLM = m_MinimaList.begin();
+  if (m_CurrentLM == m_MinimaList.end())
+    return; // ie nothing to process
+  std::sort(m_MinimaList.begin(), m_MinimaList.end(), LocMinSorter());
+
+  m_Scanbeam = ScanbeamList(); // clears/resets priority_queue
+  // reset all edges ...
+  for (MinimaList::iterator lm = m_MinimaList.begin(); lm != m_MinimaList.end();
+       ++lm) {
+    InsertScanbeam(lm->Y);
+    TEdge *e = lm->LeftBound;
+    if (e) {
+      e->Curr = e->Bot;
+      e->Side = esLeft;
+      e->OutIdx = Unassigned;
+    }
+
+    e = lm->RightBound;
+    if (e) {
+      e->Curr = e->Bot;
+      e->Side = esRight;
+      e->OutIdx = Unassigned;
+    }
+  }
+  m_ActiveEdges = 0;
+  m_CurrentLM = m_MinimaList.begin();
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::DisposeLocalMinimaList() {
+  m_MinimaList.clear();
+  m_CurrentLM = m_MinimaList.begin();
+}
+//------------------------------------------------------------------------------
+
+bool ClipperBase::PopLocalMinima(cInt Y, const LocalMinimum *&locMin) {
+  if (m_CurrentLM == m_MinimaList.end() || (*m_CurrentLM).Y != Y)
+    return false;
+  locMin = &(*m_CurrentLM);
+  ++m_CurrentLM;
+  return true;
+}
+//------------------------------------------------------------------------------
+
+IntRect ClipperBase::GetBounds() {
+  IntRect result;
+  MinimaList::iterator lm = m_MinimaList.begin();
+  if (lm == m_MinimaList.end()) {
+    result.left = result.top = result.right = result.bottom = 0;
+    return result;
+  }
+  result.left = lm->LeftBound->Bot.X;
+  result.top = lm->LeftBound->Bot.Y;
+  result.right = lm->LeftBound->Bot.X;
+  result.bottom = lm->LeftBound->Bot.Y;
+  while (lm != m_MinimaList.end()) {
+    // todo - needs fixing for open paths
+    result.bottom = std::max(result.bottom, lm->LeftBound->Bot.Y);
+    TEdge *e = lm->LeftBound;
+    for (;;) {
+      TEdge *bottomE = e;
+      while (e->NextInLML) {
+        if (e->Bot.X < result.left)
+          result.left = e->Bot.X;
+        if (e->Bot.X > result.right)
+          result.right = e->Bot.X;
+        e = e->NextInLML;
+      }
+      result.left = std::min(result.left, e->Bot.X);
+      result.right = std::max(result.right, e->Bot.X);
+      result.left = std::min(result.left, e->Top.X);
+      result.right = std::max(result.right, e->Top.X);
+      result.top = std::min(result.top, e->Top.Y);
+      if (bottomE == lm->LeftBound)
+        e = lm->RightBound;
+      else
+        break;
+    }
+    ++lm;
+  }
+  return result;
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::InsertScanbeam(const cInt Y) { m_Scanbeam.push(Y); }
+//------------------------------------------------------------------------------
+
+bool ClipperBase::PopScanbeam(cInt &Y) {
+  if (m_Scanbeam.empty())
+    return false;
+  Y = m_Scanbeam.top();
+  m_Scanbeam.pop();
+  while (!m_Scanbeam.empty() && Y == m_Scanbeam.top()) {
+    m_Scanbeam.pop();
+  } // Pop duplicates.
+  return true;
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::DisposeAllOutRecs() {
+  for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); ++i)
+    DisposeOutRec(i);
+  m_PolyOuts.clear();
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::DisposeOutRec(PolyOutList::size_type index) {
+  OutRec *outRec = m_PolyOuts[index];
+  if (outRec->Pts)
+    DisposeOutPts(outRec->Pts);
+  delete outRec;
+  m_PolyOuts[index] = 0;
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::DeleteFromAEL(TEdge *e) {
+  TEdge *AelPrev = e->PrevInAEL;
+  TEdge *AelNext = e->NextInAEL;
+  if (!AelPrev && !AelNext && (e != m_ActiveEdges))
+    return; // already deleted
+  if (AelPrev)
+    AelPrev->NextInAEL = AelNext;
+  else
+    m_ActiveEdges = AelNext;
+  if (AelNext)
+    AelNext->PrevInAEL = AelPrev;
+  e->NextInAEL = 0;
+  e->PrevInAEL = 0;
+}
+//------------------------------------------------------------------------------
+
+OutRec *ClipperBase::CreateOutRec() {
+  OutRec *result = new OutRec;
+  result->IsHole = false;
+  result->IsOpen = false;
+  result->FirstLeft = 0;
+  result->Pts = 0;
+  result->BottomPt = 0;
+  result->PolyNd = 0;
+  m_PolyOuts.push_back(result);
+  result->Idx = (int)m_PolyOuts.size() - 1;
+  return result;
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::SwapPositionsInAEL(TEdge *Edge1, TEdge *Edge2) {
+  // check that one or other edge hasn't already been removed from AEL ...
+  if (Edge1->NextInAEL == Edge1->PrevInAEL ||
+      Edge2->NextInAEL == Edge2->PrevInAEL)
+    return;
+
+  if (Edge1->NextInAEL == Edge2) {
+    TEdge *Next = Edge2->NextInAEL;
+    if (Next)
+      Next->PrevInAEL = Edge1;
+    TEdge *Prev = Edge1->PrevInAEL;
+    if (Prev)
+      Prev->NextInAEL = Edge2;
+    Edge2->PrevInAEL = Prev;
+    Edge2->NextInAEL = Edge1;
+    Edge1->PrevInAEL = Edge2;
+    Edge1->NextInAEL = Next;
+  } else if (Edge2->NextInAEL == Edge1) {
+    TEdge *Next = Edge1->NextInAEL;
+    if (Next)
+      Next->PrevInAEL = Edge2;
+    TEdge *Prev = Edge2->PrevInAEL;
+    if (Prev)
+      Prev->NextInAEL = Edge1;
+    Edge1->PrevInAEL = Prev;
+    Edge1->NextInAEL = Edge2;
+    Edge2->PrevInAEL = Edge1;
+    Edge2->NextInAEL = Next;
+  } else {
+    TEdge *Next = Edge1->NextInAEL;
+    TEdge *Prev = Edge1->PrevInAEL;
+    Edge1->NextInAEL = Edge2->NextInAEL;
+    if (Edge1->NextInAEL)
+      Edge1->NextInAEL->PrevInAEL = Edge1;
+    Edge1->PrevInAEL = Edge2->PrevInAEL;
+    if (Edge1->PrevInAEL)
+      Edge1->PrevInAEL->NextInAEL = Edge1;
+    Edge2->NextInAEL = Next;
+    if (Edge2->NextInAEL)
+      Edge2->NextInAEL->PrevInAEL = Edge2;
+    Edge2->PrevInAEL = Prev;
+    if (Edge2->PrevInAEL)
+      Edge2->PrevInAEL->NextInAEL = Edge2;
+  }
+
+  if (!Edge1->PrevInAEL)
+    m_ActiveEdges = Edge1;
+  else if (!Edge2->PrevInAEL)
+    m_ActiveEdges = Edge2;
+}
+//------------------------------------------------------------------------------
+
+void ClipperBase::UpdateEdgeIntoAEL(TEdge *&e) {
+  if (!e->NextInLML)
+    throw clipperException("UpdateEdgeIntoAEL: invalid call");
+
+  e->NextInLML->OutIdx = e->OutIdx;
+  TEdge *AelPrev = e->PrevInAEL;
+  TEdge *AelNext = e->NextInAEL;
+  if (AelPrev)
+    AelPrev->NextInAEL = e->NextInLML;
+  else
+    m_ActiveEdges = e->NextInLML;
+  if (AelNext)
+    AelNext->PrevInAEL = e->NextInLML;
+  e->NextInLML->Side = e->Side;
+  e->NextInLML->WindDelta = e->WindDelta;
+  e->NextInLML->WindCnt = e->WindCnt;
+  e->NextInLML->WindCnt2 = e->WindCnt2;
+  e = e->NextInLML;
+  e->Curr = e->Bot;
+  e->PrevInAEL = AelPrev;
+  e->NextInAEL = AelNext;
+  if (!IsHorizontal(*e))
+    InsertScanbeam(e->Top.Y);
+}
+//------------------------------------------------------------------------------
+
+bool ClipperBase::LocalMinimaPending() {
+  return (m_CurrentLM != m_MinimaList.end());
+}
+
+//------------------------------------------------------------------------------
+// TClipper methods ...
+//------------------------------------------------------------------------------
+
+Clipper::Clipper(int initOptions)
+    : ClipperBase() // constructor
+{
+  m_ExecuteLocked = false;
+  m_UseFullRange = false;
+  m_ReverseOutput = ((initOptions & ioReverseSolution) != 0);
+  m_StrictSimple = ((initOptions & ioStrictlySimple) != 0);
+  m_PreserveCollinear = ((initOptions & ioPreserveCollinear) != 0);
+  m_HasOpenPaths = false;
+#ifdef use_xyz
+  m_ZFill = 0;
+#endif
+}
+//------------------------------------------------------------------------------
+
+#ifdef use_xyz
+void Clipper::ZFillFunction(ZFillCallback zFillFunc) { m_ZFill = zFillFunc; }
+//------------------------------------------------------------------------------
+#endif
+
+bool Clipper::Execute(ClipType clipType, Paths &solution,
+                      PolyFillType fillType) {
+  return Execute(clipType, solution, fillType, fillType);
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::Execute(ClipType clipType, PolyTree &polytree,
+                      PolyFillType fillType) {
+  return Execute(clipType, polytree, fillType, fillType);
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::Execute(ClipType clipType, Paths &solution,
+                      PolyFillType subjFillType, PolyFillType clipFillType) {
+  if (m_ExecuteLocked)
+    return false;
+  if (m_HasOpenPaths)
+    throw clipperException(
+        "Error: PolyTree struct is needed for open path clipping.");
+  m_ExecuteLocked = true;
+  solution.resize(0);
+  m_SubjFillType = subjFillType;
+  m_ClipFillType = clipFillType;
+  m_ClipType = clipType;
+  m_UsingPolyTree = false;
+  bool succeeded = ExecuteInternal();
+  if (succeeded)
+    BuildResult(solution);
+  DisposeAllOutRecs();
+  m_ExecuteLocked = false;
+  return succeeded;
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::Execute(ClipType clipType, PolyTree &polytree,
+                      PolyFillType subjFillType, PolyFillType clipFillType) {
+  if (m_ExecuteLocked)
+    return false;
+  m_ExecuteLocked = true;
+  m_SubjFillType = subjFillType;
+  m_ClipFillType = clipFillType;
+  m_ClipType = clipType;
+  m_UsingPolyTree = true;
+  bool succeeded = ExecuteInternal();
+  if (succeeded)
+    BuildResult2(polytree);
+  DisposeAllOutRecs();
+  m_ExecuteLocked = false;
+  return succeeded;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::FixHoleLinkage(OutRec &outrec) {
+  // skip OutRecs that (a) contain outermost polygons or
+  //(b) already have the correct owner/child linkage ...
+  if (!outrec.FirstLeft ||
+      (outrec.IsHole != outrec.FirstLeft->IsHole && outrec.FirstLeft->Pts))
+    return;
+
+  OutRec *orfl = outrec.FirstLeft;
+  while (orfl && ((orfl->IsHole == outrec.IsHole) || !orfl->Pts))
+    orfl = orfl->FirstLeft;
+  outrec.FirstLeft = orfl;
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::ExecuteInternal() {
+  bool succeeded = true;
+  try {
+    Reset();
+    m_Maxima = MaximaList();
+    m_SortedEdges = 0;
+
+    succeeded = true;
+    cInt botY, topY;
+    if (!PopScanbeam(botY))
+      return false;
+    InsertLocalMinimaIntoAEL(botY);
+    while (PopScanbeam(topY) || LocalMinimaPending()) {
+      ProcessHorizontals();
+      ClearGhostJoins();
+      if (!ProcessIntersections(topY)) {
+        succeeded = false;
+        break;
+      }
+      ProcessEdgesAtTopOfScanbeam(topY);
+      botY = topY;
+      InsertLocalMinimaIntoAEL(botY);
+    }
+  } catch (...) {
+    succeeded = false;
+  }
+
+  if (succeeded) {
+    // fix orientations ...
+    for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); ++i) {
+      OutRec *outRec = m_PolyOuts[i];
+      if (!outRec->Pts || outRec->IsOpen)
+        continue;
+      if ((outRec->IsHole ^ m_ReverseOutput) == (Area(*outRec) > 0))
+        ReversePolyPtLinks(outRec->Pts);
+    }
+
+    if (!m_Joins.empty())
+      JoinCommonEdges();
+
+    // unfortunately FixupOutPolygon() must be done after JoinCommonEdges()
+    for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); ++i) {
+      OutRec *outRec = m_PolyOuts[i];
+      if (!outRec->Pts)
+        continue;
+      if (outRec->IsOpen)
+        FixupOutPolyline(*outRec);
+      else
+        FixupOutPolygon(*outRec);
+    }
+
+    if (m_StrictSimple)
+      DoSimplePolygons();
+  }
+
+  ClearJoins();
+  ClearGhostJoins();
+  return succeeded;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::SetWindingCount(TEdge &edge) {
+  TEdge *e = edge.PrevInAEL;
+  // find the edge of the same polytype that immediately preceeds 'edge' in AEL
+  while (e && ((e->PolyTyp != edge.PolyTyp) || (e->WindDelta == 0)))
+    e = e->PrevInAEL;
+  if (!e) {
+    if (edge.WindDelta == 0) {
+      PolyFillType pft =
+          (edge.PolyTyp == ptSubject ? m_SubjFillType : m_ClipFillType);
+      edge.WindCnt = (pft == pftNegative ? -1 : 1);
+    } else
+      edge.WindCnt = edge.WindDelta;
+    edge.WindCnt2 = 0;
+    e = m_ActiveEdges; // ie get ready to calc WindCnt2
+  } else if (edge.WindDelta == 0 && m_ClipType != ctUnion) {
+    edge.WindCnt = 1;
+    edge.WindCnt2 = e->WindCnt2;
+    e = e->NextInAEL; // ie get ready to calc WindCnt2
+  } else if (IsEvenOddFillType(edge)) {
+    // EvenOdd filling ...
+    if (edge.WindDelta == 0) {
+      // are we inside a subj polygon ...
+      bool Inside = true;
+      TEdge *e2 = e->PrevInAEL;
+      while (e2) {
+        if (e2->PolyTyp == e->PolyTyp && e2->WindDelta != 0)
+          Inside = !Inside;
+        e2 = e2->PrevInAEL;
+      }
+      edge.WindCnt = (Inside ? 0 : 1);
+    } else {
+      edge.WindCnt = edge.WindDelta;
+    }
+    edge.WindCnt2 = e->WindCnt2;
+    e = e->NextInAEL; // ie get ready to calc WindCnt2
+  } else {
+    // nonZero, Positive or Negative filling ...
+    if (e->WindCnt * e->WindDelta < 0) {
+      // prev edge is 'decreasing' WindCount (WC) toward zero
+      // so we're outside the previous polygon ...
+      if (Abs(e->WindCnt) > 1) {
+        // outside prev poly but still inside another.
+        // when reversing direction of prev poly use the same WC
+        if (e->WindDelta * edge.WindDelta < 0)
+          edge.WindCnt = e->WindCnt;
+        // otherwise continue to 'decrease' WC ...
+        else
+          edge.WindCnt = e->WindCnt + edge.WindDelta;
+      } else
+        // now outside all polys of same polytype so set own WC ...
+        edge.WindCnt = (edge.WindDelta == 0 ? 1 : edge.WindDelta);
+    } else {
+      // prev edge is 'increasing' WindCount (WC) away from zero
+      // so we're inside the previous polygon ...
+      if (edge.WindDelta == 0)
+        edge.WindCnt = (e->WindCnt < 0 ? e->WindCnt - 1 : e->WindCnt + 1);
+      // if wind direction is reversing prev then use same WC
+      else if (e->WindDelta * edge.WindDelta < 0)
+        edge.WindCnt = e->WindCnt;
+      // otherwise add to WC ...
+      else
+        edge.WindCnt = e->WindCnt + edge.WindDelta;
+    }
+    edge.WindCnt2 = e->WindCnt2;
+    e = e->NextInAEL; // ie get ready to calc WindCnt2
+  }
+
+  // update WindCnt2 ...
+  if (IsEvenOddAltFillType(edge)) {
+    // EvenOdd filling ...
+    while (e != &edge) {
+      if (e->WindDelta != 0)
+        edge.WindCnt2 = (edge.WindCnt2 == 0 ? 1 : 0);
+      e = e->NextInAEL;
+    }
+  } else {
+    // nonZero, Positive or Negative filling ...
+    while (e != &edge) {
+      edge.WindCnt2 += e->WindDelta;
+      e = e->NextInAEL;
+    }
+  }
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::IsEvenOddFillType(const TEdge &edge) const {
+  if (edge.PolyTyp == ptSubject)
+    return m_SubjFillType == pftEvenOdd;
+  else
+    return m_ClipFillType == pftEvenOdd;
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::IsEvenOddAltFillType(const TEdge &edge) const {
+  if (edge.PolyTyp == ptSubject)
+    return m_ClipFillType == pftEvenOdd;
+  else
+    return m_SubjFillType == pftEvenOdd;
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::IsContributing(const TEdge &edge) const {
+  PolyFillType pft, pft2;
+  if (edge.PolyTyp == ptSubject) {
+    pft = m_SubjFillType;
+    pft2 = m_ClipFillType;
+  } else {
+    pft = m_ClipFillType;
+    pft2 = m_SubjFillType;
+  }
+
+  switch (pft) {
+  case pftEvenOdd:
+    // return false if a subj line has been flagged as inside a subj polygon
+    if (edge.WindDelta == 0 && edge.WindCnt != 1)
+      return false;
+    break;
+  case pftNonZero:
+    if (Abs(edge.WindCnt) != 1)
+      return false;
+    break;
+  case pftPositive:
+    if (edge.WindCnt != 1)
+      return false;
+    break;
+  default: // pftNegative
+    if (edge.WindCnt != -1)
+      return false;
+  }
+
+  switch (m_ClipType) {
+  case ctIntersection:
+    switch (pft2) {
+    case pftEvenOdd:
+    case pftNonZero:
+      return (edge.WindCnt2 != 0);
+    case pftPositive:
+      return (edge.WindCnt2 > 0);
+    default:
+      return (edge.WindCnt2 < 0);
+    }
+    break;
+  case ctUnion:
+    switch (pft2) {
+    case pftEvenOdd:
+    case pftNonZero:
+      return (edge.WindCnt2 == 0);
+    case pftPositive:
+      return (edge.WindCnt2 <= 0);
+    default:
+      return (edge.WindCnt2 >= 0);
+    }
+    break;
+  case ctDifference:
+    if (edge.PolyTyp == ptSubject)
+      switch (pft2) {
+      case pftEvenOdd:
+      case pftNonZero:
+        return (edge.WindCnt2 == 0);
+      case pftPositive:
+        return (edge.WindCnt2 <= 0);
+      default:
+        return (edge.WindCnt2 >= 0);
+      }
+    else
+      switch (pft2) {
+      case pftEvenOdd:
+      case pftNonZero:
+        return (edge.WindCnt2 != 0);
+      case pftPositive:
+        return (edge.WindCnt2 > 0);
+      default:
+        return (edge.WindCnt2 < 0);
+      }
+    break;
+  case ctXor:
+    if (edge.WindDelta == 0) // XOr always contributing unless open
+      switch (pft2) {
+      case pftEvenOdd:
+      case pftNonZero:
+        return (edge.WindCnt2 == 0);
+      case pftPositive:
+        return (edge.WindCnt2 <= 0);
+      default:
+        return (edge.WindCnt2 >= 0);
+      }
+    else
+      return true;
+    break;
+  default:
+    return true;
+  }
+}
+//------------------------------------------------------------------------------
+
+OutPt *Clipper::AddLocalMinPoly(TEdge *e1, TEdge *e2, const IntPoint &Pt) {
+  OutPt *result;
+  TEdge *e, *prevE;
+  if (IsHorizontal(*e2) || (e1->Dx > e2->Dx)) {
+    result = AddOutPt(e1, Pt);
+    e2->OutIdx = e1->OutIdx;
+    e1->Side = esLeft;
+    e2->Side = esRight;
+    e = e1;
+    if (e->PrevInAEL == e2)
+      prevE = e2->PrevInAEL;
+    else
+      prevE = e->PrevInAEL;
+  } else {
+    result = AddOutPt(e2, Pt);
+    e1->OutIdx = e2->OutIdx;
+    e1->Side = esRight;
+    e2->Side = esLeft;
+    e = e2;
+    if (e->PrevInAEL == e1)
+      prevE = e1->PrevInAEL;
+    else
+      prevE = e->PrevInAEL;
+  }
+
+  if (prevE && prevE->OutIdx >= 0 && prevE->Top.Y < Pt.Y && e->Top.Y < Pt.Y) {
+    cInt xPrev = TopX(*prevE, Pt.Y);
+    cInt xE = TopX(*e, Pt.Y);
+    if (xPrev == xE && (e->WindDelta != 0) && (prevE->WindDelta != 0) &&
+        SlopesEqual(IntPoint(xPrev, Pt.Y), prevE->Top, IntPoint(xE, Pt.Y),
+                    e->Top, m_UseFullRange)) {
+      OutPt *outPt = AddOutPt(prevE, Pt);
+      AddJoin(result, outPt, e->Top);
+    }
+  }
+  return result;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::AddLocalMaxPoly(TEdge *e1, TEdge *e2, const IntPoint &Pt) {
+  AddOutPt(e1, Pt);
+  if (e2->WindDelta == 0)
+    AddOutPt(e2, Pt);
+  if (e1->OutIdx == e2->OutIdx) {
+    e1->OutIdx = Unassigned;
+    e2->OutIdx = Unassigned;
+  } else if (e1->OutIdx < e2->OutIdx)
+    AppendPolygon(e1, e2);
+  else
+    AppendPolygon(e2, e1);
+}
+//------------------------------------------------------------------------------
+
+void Clipper::AddEdgeToSEL(TEdge *edge) {
+  // SEL pointers in PEdge are reused to build a list of horizontal edges.
+  // However, we don't need to worry about order with horizontal edge
+  // processing.
+  if (!m_SortedEdges) {
+    m_SortedEdges = edge;
+    edge->PrevInSEL = 0;
+    edge->NextInSEL = 0;
+  } else {
+    edge->NextInSEL = m_SortedEdges;
+    edge->PrevInSEL = 0;
+    m_SortedEdges->PrevInSEL = edge;
+    m_SortedEdges = edge;
+  }
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::PopEdgeFromSEL(TEdge *&edge) {
+  if (!m_SortedEdges)
+    return false;
+  edge = m_SortedEdges;
+  DeleteFromSEL(m_SortedEdges);
+  return true;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::CopyAELToSEL() {
+  TEdge *e = m_ActiveEdges;
+  m_SortedEdges = e;
+  while (e) {
+    e->PrevInSEL = e->PrevInAEL;
+    e->NextInSEL = e->NextInAEL;
+    e = e->NextInAEL;
+  }
+}
+//------------------------------------------------------------------------------
+
+void Clipper::AddJoin(OutPt *op1, OutPt *op2, const IntPoint OffPt) {
+  Join *j = new Join;
+  j->OutPt1 = op1;
+  j->OutPt2 = op2;
+  j->OffPt = OffPt;
+  m_Joins.push_back(j);
+}
+//------------------------------------------------------------------------------
+
+void Clipper::ClearJoins() {
+  for (JoinList::size_type i = 0; i < m_Joins.size(); i++)
+    delete m_Joins[i];
+  m_Joins.resize(0);
+}
+//------------------------------------------------------------------------------
+
+void Clipper::ClearGhostJoins() {
+  for (JoinList::size_type i = 0; i < m_GhostJoins.size(); i++)
+    delete m_GhostJoins[i];
+  m_GhostJoins.resize(0);
+}
+//------------------------------------------------------------------------------
+
+void Clipper::AddGhostJoin(OutPt *op, const IntPoint OffPt) {
+  Join *j = new Join;
+  j->OutPt1 = op;
+  j->OutPt2 = 0;
+  j->OffPt = OffPt;
+  m_GhostJoins.push_back(j);
+}
+//------------------------------------------------------------------------------
+
+void Clipper::InsertLocalMinimaIntoAEL(const cInt botY) {
+  const LocalMinimum *lm;
+  while (PopLocalMinima(botY, lm)) {
+    TEdge *lb = lm->LeftBound;
+    TEdge *rb = lm->RightBound;
+
+    OutPt *Op1 = 0;
+    if (!lb) {
+      // nb: don't insert LB into either AEL or SEL
+      InsertEdgeIntoAEL(rb, 0);
+      SetWindingCount(*rb);
+      if (IsContributing(*rb))
+        Op1 = AddOutPt(rb, rb->Bot);
+    } else if (!rb) {
+      InsertEdgeIntoAEL(lb, 0);
+      SetWindingCount(*lb);
+      if (IsContributing(*lb))
+        Op1 = AddOutPt(lb, lb->Bot);
+      InsertScanbeam(lb->Top.Y);
+    } else {
+      InsertEdgeIntoAEL(lb, 0);
+      InsertEdgeIntoAEL(rb, lb);
+      SetWindingCount(*lb);
+      rb->WindCnt = lb->WindCnt;
+      rb->WindCnt2 = lb->WindCnt2;
+      if (IsContributing(*lb))
+        Op1 = AddLocalMinPoly(lb, rb, lb->Bot);
+      InsertScanbeam(lb->Top.Y);
+    }
+
+    if (rb) {
+      if (IsHorizontal(*rb)) {
+        AddEdgeToSEL(rb);
+        if (rb->NextInLML)
+          InsertScanbeam(rb->NextInLML->Top.Y);
+      } else
+        InsertScanbeam(rb->Top.Y);
+    }
+
+    if (!lb || !rb)
+      continue;
+
+    // if any output polygons share an edge, they'll need joining later ...
+    if (Op1 && IsHorizontal(*rb) && m_GhostJoins.size() > 0 &&
+        (rb->WindDelta != 0)) {
+      for (JoinList::size_type i = 0; i < m_GhostJoins.size(); ++i) {
+        Join *jr = m_GhostJoins[i];
+        // if the horizontal Rb and a 'ghost' horizontal overlap, then convert
+        // the 'ghost' join to a real join ready for later ...
+        if (HorzSegmentsOverlap(jr->OutPt1->Pt.X, jr->OffPt.X, rb->Bot.X,
+                                rb->Top.X))
+          AddJoin(jr->OutPt1, Op1, jr->OffPt);
+      }
+    }
+
+    if (lb->OutIdx >= 0 && lb->PrevInAEL &&
+        lb->PrevInAEL->Curr.X == lb->Bot.X && lb->PrevInAEL->OutIdx >= 0 &&
+        SlopesEqual(lb->PrevInAEL->Bot, lb->PrevInAEL->Top, lb->Curr, lb->Top,
+                    m_UseFullRange) &&
+        (lb->WindDelta != 0) && (lb->PrevInAEL->WindDelta != 0)) {
+      OutPt *Op2 = AddOutPt(lb->PrevInAEL, lb->Bot);
+      AddJoin(Op1, Op2, lb->Top);
+    }
+
+    if (lb->NextInAEL != rb) {
+
+      if (rb->OutIdx >= 0 && rb->PrevInAEL->OutIdx >= 0 &&
+          SlopesEqual(rb->PrevInAEL->Curr, rb->PrevInAEL->Top, rb->Curr,
+                      rb->Top, m_UseFullRange) &&
+          (rb->WindDelta != 0) && (rb->PrevInAEL->WindDelta != 0)) {
+        OutPt *Op2 = AddOutPt(rb->PrevInAEL, rb->Bot);
+        AddJoin(Op1, Op2, rb->Top);
+      }
+
+      TEdge *e = lb->NextInAEL;
+      if (e) {
+        while (e != rb) {
+          // nb: For calculating winding counts etc, IntersectEdges() assumes
+          // that param1 will be to the Right of param2 ABOVE the intersection
+          // ...
+          IntersectEdges(rb, e, lb->Curr); // order important here
+          e = e->NextInAEL;
+        }
+      }
+    }
+  }
+}
+//------------------------------------------------------------------------------
+
+void Clipper::DeleteFromSEL(TEdge *e) {
+  TEdge *SelPrev = e->PrevInSEL;
+  TEdge *SelNext = e->NextInSEL;
+  if (!SelPrev && !SelNext && (e != m_SortedEdges))
+    return; // already deleted
+  if (SelPrev)
+    SelPrev->NextInSEL = SelNext;
+  else
+    m_SortedEdges = SelNext;
+  if (SelNext)
+    SelNext->PrevInSEL = SelPrev;
+  e->NextInSEL = 0;
+  e->PrevInSEL = 0;
+}
+//------------------------------------------------------------------------------
+
+#ifdef use_xyz
+void Clipper::SetZ(IntPoint &pt, TEdge &e1, TEdge &e2) {
+  if (pt.Z != 0 || !m_ZFill)
+    return;
+  else if (pt == e1.Bot)
+    pt.Z = e1.Bot.Z;
+  else if (pt == e1.Top)
+    pt.Z = e1.Top.Z;
+  else if (pt == e2.Bot)
+    pt.Z = e2.Bot.Z;
+  else if (pt == e2.Top)
+    pt.Z = e2.Top.Z;
+  else
+    (*m_ZFill)(e1.Bot, e1.Top, e2.Bot, e2.Top, pt);
+}
+//------------------------------------------------------------------------------
+#endif
+
+void Clipper::IntersectEdges(TEdge *e1, TEdge *e2, IntPoint &Pt) {
+  bool e1Contributing = (e1->OutIdx >= 0);
+  bool e2Contributing = (e2->OutIdx >= 0);
+
+#ifdef use_xyz
+  SetZ(Pt, *e1, *e2);
+#endif
+
+#ifdef use_lines
+  // if either edge is on an OPEN path ...
+  if (e1->WindDelta == 0 || e2->WindDelta == 0) {
+    // ignore subject-subject open path intersections UNLESS they
+    // are both open paths, AND they are both 'contributing maximas' ...
+    if (e1->WindDelta == 0 && e2->WindDelta == 0)
+      return;
+
+    // if intersecting a subj line with a subj poly ...
+    else if (e1->PolyTyp == e2->PolyTyp && e1->WindDelta != e2->WindDelta &&
+             m_ClipType == ctUnion) {
+      if (e1->WindDelta == 0) {
+        if (e2Contributing) {
+          AddOutPt(e1, Pt);
+          if (e1Contributing)
+            e1->OutIdx = Unassigned;
+        }
+      } else {
+        if (e1Contributing) {
+          AddOutPt(e2, Pt);
+          if (e2Contributing)
+            e2->OutIdx = Unassigned;
+        }
+      }
+    } else if (e1->PolyTyp != e2->PolyTyp) {
+      // toggle subj open path OutIdx on/off when Abs(clip.WndCnt) == 1 ...
+      if ((e1->WindDelta == 0) && abs(e2->WindCnt) == 1 &&
+          (m_ClipType != ctUnion || e2->WindCnt2 == 0)) {
+        AddOutPt(e1, Pt);
+        if (e1Contributing)
+          e1->OutIdx = Unassigned;
+      } else if ((e2->WindDelta == 0) && (abs(e1->WindCnt) == 1) &&
+                 (m_ClipType != ctUnion || e1->WindCnt2 == 0)) {
+        AddOutPt(e2, Pt);
+        if (e2Contributing)
+          e2->OutIdx = Unassigned;
+      }
+    }
+    return;
+  }
+#endif
+
+  // update winding counts...
+  // assumes that e1 will be to the Right of e2 ABOVE the intersection
+  if (e1->PolyTyp == e2->PolyTyp) {
+    if (IsEvenOddFillType(*e1)) {
+      int oldE1WindCnt = e1->WindCnt;
+      e1->WindCnt = e2->WindCnt;
+      e2->WindCnt = oldE1WindCnt;
+    } else {
+      if (e1->WindCnt + e2->WindDelta == 0)
+        e1->WindCnt = -e1->WindCnt;
+      else
+        e1->WindCnt += e2->WindDelta;
+      if (e2->WindCnt - e1->WindDelta == 0)
+        e2->WindCnt = -e2->WindCnt;
+      else
+        e2->WindCnt -= e1->WindDelta;
+    }
+  } else {
+    if (!IsEvenOddFillType(*e2))
+      e1->WindCnt2 += e2->WindDelta;
+    else
+      e1->WindCnt2 = (e1->WindCnt2 == 0) ? 1 : 0;
+    if (!IsEvenOddFillType(*e1))
+      e2->WindCnt2 -= e1->WindDelta;
+    else
+      e2->WindCnt2 = (e2->WindCnt2 == 0) ? 1 : 0;
+  }
+
+  PolyFillType e1FillType, e2FillType, e1FillType2, e2FillType2;
+  if (e1->PolyTyp == ptSubject) {
+    e1FillType = m_SubjFillType;
+    e1FillType2 = m_ClipFillType;
+  } else {
+    e1FillType = m_ClipFillType;
+    e1FillType2 = m_SubjFillType;
+  }
+  if (e2->PolyTyp == ptSubject) {
+    e2FillType = m_SubjFillType;
+    e2FillType2 = m_ClipFillType;
+  } else {
+    e2FillType = m_ClipFillType;
+    e2FillType2 = m_SubjFillType;
+  }
+
+  cInt e1Wc, e2Wc;
+  switch (e1FillType) {
+  case pftPositive:
+    e1Wc = e1->WindCnt;
+    break;
+  case pftNegative:
+    e1Wc = -e1->WindCnt;
+    break;
+  default:
+    e1Wc = Abs(e1->WindCnt);
+  }
+  switch (e2FillType) {
+  case pftPositive:
+    e2Wc = e2->WindCnt;
+    break;
+  case pftNegative:
+    e2Wc = -e2->WindCnt;
+    break;
+  default:
+    e2Wc = Abs(e2->WindCnt);
+  }
+
+  if (e1Contributing && e2Contributing) {
+    if ((e1Wc != 0 && e1Wc != 1) || (e2Wc != 0 && e2Wc != 1) ||
+        (e1->PolyTyp != e2->PolyTyp && m_ClipType != ctXor)) {
+      AddLocalMaxPoly(e1, e2, Pt);
+    } else {
+      AddOutPt(e1, Pt);
+      AddOutPt(e2, Pt);
+      SwapSides(*e1, *e2);
+      SwapPolyIndexes(*e1, *e2);
+    }
+  } else if (e1Contributing) {
+    if (e2Wc == 0 || e2Wc == 1) {
+      AddOutPt(e1, Pt);
+      SwapSides(*e1, *e2);
+      SwapPolyIndexes(*e1, *e2);
+    }
+  } else if (e2Contributing) {
+    if (e1Wc == 0 || e1Wc == 1) {
+      AddOutPt(e2, Pt);
+      SwapSides(*e1, *e2);
+      SwapPolyIndexes(*e1, *e2);
+    }
+  } else if ((e1Wc == 0 || e1Wc == 1) && (e2Wc == 0 || e2Wc == 1)) {
+    // neither edge is currently contributing ...
+
+    cInt e1Wc2, e2Wc2;
+    switch (e1FillType2) {
+    case pftPositive:
+      e1Wc2 = e1->WindCnt2;
+      break;
+    case pftNegative:
+      e1Wc2 = -e1->WindCnt2;
+      break;
+    default:
+      e1Wc2 = Abs(e1->WindCnt2);
+    }
+    switch (e2FillType2) {
+    case pftPositive:
+      e2Wc2 = e2->WindCnt2;
+      break;
+    case pftNegative:
+      e2Wc2 = -e2->WindCnt2;
+      break;
+    default:
+      e2Wc2 = Abs(e2->WindCnt2);
+    }
+
+    if (e1->PolyTyp != e2->PolyTyp) {
+      AddLocalMinPoly(e1, e2, Pt);
+    } else if (e1Wc == 1 && e2Wc == 1)
+      switch (m_ClipType) {
+      case ctIntersection:
+        if (e1Wc2 > 0 && e2Wc2 > 0)
+          AddLocalMinPoly(e1, e2, Pt);
+        break;
+      case ctUnion:
+        if (e1Wc2 <= 0 && e2Wc2 <= 0)
+          AddLocalMinPoly(e1, e2, Pt);
+        break;
+      case ctDifference:
+        if (((e1->PolyTyp == ptClip) && (e1Wc2 > 0) && (e2Wc2 > 0)) ||
+            ((e1->PolyTyp == ptSubject) && (e1Wc2 <= 0) && (e2Wc2 <= 0)))
+          AddLocalMinPoly(e1, e2, Pt);
+        break;
+      case ctXor:
+        AddLocalMinPoly(e1, e2, Pt);
+      }
+    else
+      SwapSides(*e1, *e2);
+  }
+}
+//------------------------------------------------------------------------------
+
+void Clipper::SetHoleState(TEdge *e, OutRec *outrec) {
+  TEdge *e2 = e->PrevInAEL;
+  TEdge *eTmp = 0;
+  while (e2) {
+    if (e2->OutIdx >= 0 && e2->WindDelta != 0) {
+      if (!eTmp)
+        eTmp = e2;
+      else if (eTmp->OutIdx == e2->OutIdx)
+        eTmp = 0;
+    }
+    e2 = e2->PrevInAEL;
+  }
+  if (!eTmp) {
+    outrec->FirstLeft = 0;
+    outrec->IsHole = false;
+  } else {
+    outrec->FirstLeft = m_PolyOuts[eTmp->OutIdx];
+    outrec->IsHole = !outrec->FirstLeft->IsHole;
+  }
+}
+//------------------------------------------------------------------------------
+
+OutRec *GetLowermostRec(OutRec *outRec1, OutRec *outRec2) {
+  // work out which polygon fragment has the correct hole state ...
+  if (!outRec1->BottomPt)
+    outRec1->BottomPt = GetBottomPt(outRec1->Pts);
+  if (!outRec2->BottomPt)
+    outRec2->BottomPt = GetBottomPt(outRec2->Pts);
+  OutPt *OutPt1 = outRec1->BottomPt;
+  OutPt *OutPt2 = outRec2->BottomPt;
+  if (OutPt1->Pt.Y > OutPt2->Pt.Y)
+    return outRec1;
+  else if (OutPt1->Pt.Y < OutPt2->Pt.Y)
+    return outRec2;
+  else if (OutPt1->Pt.X < OutPt2->Pt.X)
+    return outRec1;
+  else if (OutPt1->Pt.X > OutPt2->Pt.X)
+    return outRec2;
+  else if (OutPt1->Next == OutPt1)
+    return outRec2;
+  else if (OutPt2->Next == OutPt2)
+    return outRec1;
+  else if (FirstIsBottomPt(OutPt1, OutPt2))
+    return outRec1;
+  else
+    return outRec2;
+}
+//------------------------------------------------------------------------------
+
+bool OutRec1RightOfOutRec2(OutRec *outRec1, OutRec *outRec2) {
+  do {
+    outRec1 = outRec1->FirstLeft;
+    if (outRec1 == outRec2)
+      return true;
+  } while (outRec1);
+  return false;
+}
+//------------------------------------------------------------------------------
+
+OutRec *Clipper::GetOutRec(int Idx) {
+  OutRec *outrec = m_PolyOuts[Idx];
+  while (outrec != m_PolyOuts[outrec->Idx])
+    outrec = m_PolyOuts[outrec->Idx];
+  return outrec;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::AppendPolygon(TEdge *e1, TEdge *e2) {
+  // get the start and ends of both output polygons ...
+  OutRec *outRec1 = m_PolyOuts[e1->OutIdx];
+  OutRec *outRec2 = m_PolyOuts[e2->OutIdx];
+
+  OutRec *holeStateRec;
+  if (OutRec1RightOfOutRec2(outRec1, outRec2))
+    holeStateRec = outRec2;
+  else if (OutRec1RightOfOutRec2(outRec2, outRec1))
+    holeStateRec = outRec1;
+  else
+    holeStateRec = GetLowermostRec(outRec1, outRec2);
+
+  // get the start and ends of both output polygons and
+  // join e2 poly onto e1 poly and delete pointers to e2 ...
+
+  OutPt *p1_lft = outRec1->Pts;
+  OutPt *p1_rt = p1_lft->Prev;
+  OutPt *p2_lft = outRec2->Pts;
+  OutPt *p2_rt = p2_lft->Prev;
+
+  // join e2 poly onto e1 poly and delete pointers to e2 ...
+  if (e1->Side == esLeft) {
+    if (e2->Side == esLeft) {
+      // z y x a b c
+      ReversePolyPtLinks(p2_lft);
+      p2_lft->Next = p1_lft;
+      p1_lft->Prev = p2_lft;
+      p1_rt->Next = p2_rt;
+      p2_rt->Prev = p1_rt;
+      outRec1->Pts = p2_rt;
+    } else {
+      // x y z a b c
+      p2_rt->Next = p1_lft;
+      p1_lft->Prev = p2_rt;
+      p2_lft->Prev = p1_rt;
+      p1_rt->Next = p2_lft;
+      outRec1->Pts = p2_lft;
+    }
+  } else {
+    if (e2->Side == esRight) {
+      // a b c z y x
+      ReversePolyPtLinks(p2_lft);
+      p1_rt->Next = p2_rt;
+      p2_rt->Prev = p1_rt;
+      p2_lft->Next = p1_lft;
+      p1_lft->Prev = p2_lft;
+    } else {
+      // a b c x y z
+      p1_rt->Next = p2_lft;
+      p2_lft->Prev = p1_rt;
+      p1_lft->Prev = p2_rt;
+      p2_rt->Next = p1_lft;
+    }
+  }
+
+  outRec1->BottomPt = 0;
+  if (holeStateRec == outRec2) {
+    if (outRec2->FirstLeft != outRec1)
+      outRec1->FirstLeft = outRec2->FirstLeft;
+    outRec1->IsHole = outRec2->IsHole;
+  }
+  outRec2->Pts = 0;
+  outRec2->BottomPt = 0;
+  outRec2->FirstLeft = outRec1;
+
+  int OKIdx = e1->OutIdx;
+  int ObsoleteIdx = e2->OutIdx;
+
+  e1->OutIdx =
+      Unassigned; // nb: safe because we only get here via AddLocalMaxPoly
+  e2->OutIdx = Unassigned;
+
+  TEdge *e = m_ActiveEdges;
+  while (e) {
+    if (e->OutIdx == ObsoleteIdx) {
+      e->OutIdx = OKIdx;
+      e->Side = e1->Side;
+      break;
+    }
+    e = e->NextInAEL;
+  }
+
+  outRec2->Idx = outRec1->Idx;
+}
+//------------------------------------------------------------------------------
+
+OutPt *Clipper::AddOutPt(TEdge *e, const IntPoint &pt) {
+  if (e->OutIdx < 0) {
+    OutRec *outRec = CreateOutRec();
+    outRec->IsOpen = (e->WindDelta == 0);
+    OutPt *newOp = new OutPt;
+    outRec->Pts = newOp;
+    newOp->Idx = outRec->Idx;
+    newOp->Pt = pt;
+    newOp->Next = newOp;
+    newOp->Prev = newOp;
+    if (!outRec->IsOpen)
+      SetHoleState(e, outRec);
+    e->OutIdx = outRec->Idx;
+    return newOp;
+  } else {
+    OutRec *outRec = m_PolyOuts[e->OutIdx];
+    // OutRec.Pts is the 'Left-most' point & OutRec.Pts.Prev is the 'Right-most'
+    OutPt *op = outRec->Pts;
+
+    bool ToFront = (e->Side == esLeft);
+    if (ToFront && (pt == op->Pt))
+      return op;
+    else if (!ToFront && (pt == op->Prev->Pt))
+      return op->Prev;
+
+    OutPt *newOp = new OutPt;
+    newOp->Idx = outRec->Idx;
+    newOp->Pt = pt;
+    newOp->Next = op;
+    newOp->Prev = op->Prev;
+    newOp->Prev->Next = newOp;
+    op->Prev = newOp;
+    if (ToFront)
+      outRec->Pts = newOp;
+    return newOp;
+  }
+}
+//------------------------------------------------------------------------------
+
+OutPt *Clipper::GetLastOutPt(TEdge *e) {
+  OutRec *outRec = m_PolyOuts[e->OutIdx];
+  if (e->Side == esLeft)
+    return outRec->Pts;
+  else
+    return outRec->Pts->Prev;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::ProcessHorizontals() {
+  TEdge *horzEdge;
+  while (PopEdgeFromSEL(horzEdge))
+    ProcessHorizontal(horzEdge);
+}
+//------------------------------------------------------------------------------
+
+inline bool IsMinima(TEdge *e) {
+  return e && (e->Prev->NextInLML != e) && (e->Next->NextInLML != e);
+}
+//------------------------------------------------------------------------------
+
+inline bool IsMaxima(TEdge *e, const cInt Y) {
+  return e && e->Top.Y == Y && !e->NextInLML;
+}
+//------------------------------------------------------------------------------
+
+inline bool IsIntermediate(TEdge *e, const cInt Y) {
+  return e->Top.Y == Y && e->NextInLML;
+}
+//------------------------------------------------------------------------------
+
+TEdge *GetMaximaPair(TEdge *e) {
+  if ((e->Next->Top == e->Top) && !e->Next->NextInLML)
+    return e->Next;
+  else if ((e->Prev->Top == e->Top) && !e->Prev->NextInLML)
+    return e->Prev;
+  else
+    return 0;
+}
+//------------------------------------------------------------------------------
+
+TEdge *GetMaximaPairEx(TEdge *e) {
+  // as GetMaximaPair() but returns 0 if MaxPair isn't in AEL (unless it's
+  // horizontal)
+  TEdge *result = GetMaximaPair(e);
+  if (result &&
+      (result->OutIdx == Skip ||
+       (result->NextInAEL == result->PrevInAEL && !IsHorizontal(*result))))
+    return 0;
+  return result;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::SwapPositionsInSEL(TEdge *Edge1, TEdge *Edge2) {
+  if (!(Edge1->NextInSEL) && !(Edge1->PrevInSEL))
+    return;
+  if (!(Edge2->NextInSEL) && !(Edge2->PrevInSEL))
+    return;
+
+  if (Edge1->NextInSEL == Edge2) {
+    TEdge *Next = Edge2->NextInSEL;
+    if (Next)
+      Next->PrevInSEL = Edge1;
+    TEdge *Prev = Edge1->PrevInSEL;
+    if (Prev)
+      Prev->NextInSEL = Edge2;
+    Edge2->PrevInSEL = Prev;
+    Edge2->NextInSEL = Edge1;
+    Edge1->PrevInSEL = Edge2;
+    Edge1->NextInSEL = Next;
+  } else if (Edge2->NextInSEL == Edge1) {
+    TEdge *Next = Edge1->NextInSEL;
+    if (Next)
+      Next->PrevInSEL = Edge2;
+    TEdge *Prev = Edge2->PrevInSEL;
+    if (Prev)
+      Prev->NextInSEL = Edge1;
+    Edge1->PrevInSEL = Prev;
+    Edge1->NextInSEL = Edge2;
+    Edge2->PrevInSEL = Edge1;
+    Edge2->NextInSEL = Next;
+  } else {
+    TEdge *Next = Edge1->NextInSEL;
+    TEdge *Prev = Edge1->PrevInSEL;
+    Edge1->NextInSEL = Edge2->NextInSEL;
+    if (Edge1->NextInSEL)
+      Edge1->NextInSEL->PrevInSEL = Edge1;
+    Edge1->PrevInSEL = Edge2->PrevInSEL;
+    if (Edge1->PrevInSEL)
+      Edge1->PrevInSEL->NextInSEL = Edge1;
+    Edge2->NextInSEL = Next;
+    if (Edge2->NextInSEL)
+      Edge2->NextInSEL->PrevInSEL = Edge2;
+    Edge2->PrevInSEL = Prev;
+    if (Edge2->PrevInSEL)
+      Edge2->PrevInSEL->NextInSEL = Edge2;
+  }
+
+  if (!Edge1->PrevInSEL)
+    m_SortedEdges = Edge1;
+  else if (!Edge2->PrevInSEL)
+    m_SortedEdges = Edge2;
+}
+//------------------------------------------------------------------------------
+
+TEdge *GetNextInAEL(TEdge *e, Direction dir) {
+  return dir == dLeftToRight ? e->NextInAEL : e->PrevInAEL;
+}
+//------------------------------------------------------------------------------
+
+void GetHorzDirection(TEdge &HorzEdge, Direction &Dir, cInt &Left,
+                      cInt &Right) {
+  if (HorzEdge.Bot.X < HorzEdge.Top.X) {
+    Left = HorzEdge.Bot.X;
+    Right = HorzEdge.Top.X;
+    Dir = dLeftToRight;
+  } else {
+    Left = HorzEdge.Top.X;
+    Right = HorzEdge.Bot.X;
+    Dir = dRightToLeft;
+  }
+}
+//------------------------------------------------------------------------
+
+/*******************************************************************************
+ * Notes: Horizontal edges (HEs) at scanline intersections (ie at the Top or *
+ * Bottom of a scanbeam) are processed as if layered. The order in which HEs *
+ * are processed doesn't matter. HEs intersect with other HE Bot.Xs only [#] *
+ * (or they could intersect with Top.Xs only, ie EITHER Bot.Xs OR Top.Xs), * and
+ *with other non-horizontal edges [*]. Once these intersections are        *
+ * processed, intermediate HEs then 'promote' the Edge above (NextInLML) into *
+ * the AEL. These 'promoted' edges may in turn intersect [%] with other HEs. *
+ *******************************************************************************/
+
+void Clipper::ProcessHorizontal(TEdge *horzEdge) {
+  Direction dir;
+  cInt horzLeft, horzRight;
+  bool IsOpen = (horzEdge->WindDelta == 0);
+
+  GetHorzDirection(*horzEdge, dir, horzLeft, horzRight);
+
+  TEdge *eLastHorz = horzEdge, *eMaxPair = 0;
+  while (eLastHorz->NextInLML && IsHorizontal(*eLastHorz->NextInLML))
+    eLastHorz = eLastHorz->NextInLML;
+  if (!eLastHorz->NextInLML)
+    eMaxPair = GetMaximaPair(eLastHorz);
+
+  MaximaList::const_iterator maxIt;
+  MaximaList::const_reverse_iterator maxRit;
+  if (m_Maxima.size() > 0) {
+    // get the first maxima in range (X) ...
+    if (dir == dLeftToRight) {
+      maxIt = m_Maxima.begin();
+      while (maxIt != m_Maxima.end() && *maxIt <= horzEdge->Bot.X)
+        maxIt++;
+      if (maxIt != m_Maxima.end() && *maxIt >= eLastHorz->Top.X)
+        maxIt = m_Maxima.end();
+    } else {
+      maxRit = m_Maxima.rbegin();
+      while (maxRit != m_Maxima.rend() && *maxRit > horzEdge->Bot.X)
+        maxRit++;
+      if (maxRit != m_Maxima.rend() && *maxRit <= eLastHorz->Top.X)
+        maxRit = m_Maxima.rend();
+    }
+  }
+
+  OutPt *op1 = 0;
+
+  for (;;) // loop through consec. horizontal edges
+  {
+
+    bool IsLastHorz = (horzEdge == eLastHorz);
+    TEdge *e = GetNextInAEL(horzEdge, dir);
+    while (e) {
+
+      // this code block inserts extra coords into horizontal edges (in output
+      // polygons) whereever maxima touch these horizontal edges. This helps
+      //'simplifying' polygons (ie if the Simplify property is set).
+      if (m_Maxima.size() > 0) {
+        if (dir == dLeftToRight) {
+          while (maxIt != m_Maxima.end() && *maxIt < e->Curr.X) {
+            if (horzEdge->OutIdx >= 0 && !IsOpen)
+              AddOutPt(horzEdge, IntPoint(*maxIt, horzEdge->Bot.Y));
+            maxIt++;
+          }
+        } else {
+          while (maxRit != m_Maxima.rend() && *maxRit > e->Curr.X) {
+            if (horzEdge->OutIdx >= 0 && !IsOpen)
+              AddOutPt(horzEdge, IntPoint(*maxRit, horzEdge->Bot.Y));
+            maxRit++;
+          }
+        }
+      };
+
+      if ((dir == dLeftToRight && e->Curr.X > horzRight) ||
+          (dir == dRightToLeft && e->Curr.X < horzLeft))
+        break;
+
+      // Also break if we've got to the end of an intermediate horizontal edge
+      // ... nb: Smaller Dx's are to the right of larger Dx's ABOVE the
+      // horizontal.
+      if (e->Curr.X == horzEdge->Top.X && horzEdge->NextInLML &&
+          e->Dx < horzEdge->NextInLML->Dx)
+        break;
+
+      if (horzEdge->OutIdx >= 0 && !IsOpen) // note: may be done multiple times
+      {
+#ifdef use_xyz
+        if (dir == dLeftToRight)
+          SetZ(e->Curr, *horzEdge, *e);
+        else
+          SetZ(e->Curr, *e, *horzEdge);
+#endif
+        op1 = AddOutPt(horzEdge, e->Curr);
+        TEdge *eNextHorz = m_SortedEdges;
+        while (eNextHorz) {
+          if (eNextHorz->OutIdx >= 0 &&
+              HorzSegmentsOverlap(horzEdge->Bot.X, horzEdge->Top.X,
+                                  eNextHorz->Bot.X, eNextHorz->Top.X)) {
+            OutPt *op2 = GetLastOutPt(eNextHorz);
+            AddJoin(op2, op1, eNextHorz->Top);
+          }
+          eNextHorz = eNextHorz->NextInSEL;
+        }
+        AddGhostJoin(op1, horzEdge->Bot);
+      }
+
+      // OK, so far we're still in range of the horizontal Edge  but make sure
+      // we're at the last of consec. horizontals when matching with eMaxPair
+      if (e == eMaxPair && IsLastHorz) {
+        if (horzEdge->OutIdx >= 0)
+          AddLocalMaxPoly(horzEdge, eMaxPair, horzEdge->Top);
+        DeleteFromAEL(horzEdge);
+        DeleteFromAEL(eMaxPair);
+        return;
+      }
+
+      if (dir == dLeftToRight) {
+        IntPoint Pt = IntPoint(e->Curr.X, horzEdge->Curr.Y);
+        IntersectEdges(horzEdge, e, Pt);
+      } else {
+        IntPoint Pt = IntPoint(e->Curr.X, horzEdge->Curr.Y);
+        IntersectEdges(e, horzEdge, Pt);
+      }
+      TEdge *eNext = GetNextInAEL(e, dir);
+      SwapPositionsInAEL(horzEdge, e);
+      e = eNext;
+    } // end while(e)
+
+    // Break out of loop if HorzEdge.NextInLML is not also horizontal ...
+    if (!horzEdge->NextInLML || !IsHorizontal(*horzEdge->NextInLML))
+      break;
+
+    UpdateEdgeIntoAEL(horzEdge);
+    if (horzEdge->OutIdx >= 0)
+      AddOutPt(horzEdge, horzEdge->Bot);
+    GetHorzDirection(*horzEdge, dir, horzLeft, horzRight);
+
+  } // end for (;;)
+
+  if (horzEdge->OutIdx >= 0 && !op1) {
+    op1 = GetLastOutPt(horzEdge);
+    TEdge *eNextHorz = m_SortedEdges;
+    while (eNextHorz) {
+      if (eNextHorz->OutIdx >= 0 &&
+          HorzSegmentsOverlap(horzEdge->Bot.X, horzEdge->Top.X,
+                              eNextHorz->Bot.X, eNextHorz->Top.X)) {
+        OutPt *op2 = GetLastOutPt(eNextHorz);
+        AddJoin(op2, op1, eNextHorz->Top);
+      }
+      eNextHorz = eNextHorz->NextInSEL;
+    }
+    AddGhostJoin(op1, horzEdge->Top);
+  }
+
+  if (horzEdge->NextInLML) {
+    if (horzEdge->OutIdx >= 0) {
+      op1 = AddOutPt(horzEdge, horzEdge->Top);
+      UpdateEdgeIntoAEL(horzEdge);
+      if (horzEdge->WindDelta == 0)
+        return;
+      // nb: HorzEdge is no longer horizontal here
+      TEdge *ePrev = horzEdge->PrevInAEL;
+      TEdge *eNext = horzEdge->NextInAEL;
+      if (ePrev && ePrev->Curr.X == horzEdge->Bot.X &&
+          ePrev->Curr.Y == horzEdge->Bot.Y && ePrev->WindDelta != 0 &&
+          (ePrev->OutIdx >= 0 && ePrev->Curr.Y > ePrev->Top.Y &&
+           SlopesEqual(*horzEdge, *ePrev, m_UseFullRange))) {
+        OutPt *op2 = AddOutPt(ePrev, horzEdge->Bot);
+        AddJoin(op1, op2, horzEdge->Top);
+      } else if (eNext && eNext->Curr.X == horzEdge->Bot.X &&
+                 eNext->Curr.Y == horzEdge->Bot.Y && eNext->WindDelta != 0 &&
+                 eNext->OutIdx >= 0 && eNext->Curr.Y > eNext->Top.Y &&
+                 SlopesEqual(*horzEdge, *eNext, m_UseFullRange)) {
+        OutPt *op2 = AddOutPt(eNext, horzEdge->Bot);
+        AddJoin(op1, op2, horzEdge->Top);
+      }
+    } else
+      UpdateEdgeIntoAEL(horzEdge);
+  } else {
+    if (horzEdge->OutIdx >= 0)
+      AddOutPt(horzEdge, horzEdge->Top);
+    DeleteFromAEL(horzEdge);
+  }
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::ProcessIntersections(const cInt topY) {
+  if (!m_ActiveEdges)
+    return true;
+  try {
+    BuildIntersectList(topY);
+    size_t IlSize = m_IntersectList.size();
+    if (IlSize == 0)
+      return true;
+    if (IlSize == 1 || FixupIntersectionOrder())
+      ProcessIntersectList();
+    else
+      return false;
+  } catch (...) {
+    m_SortedEdges = 0;
+    DisposeIntersectNodes();
+    throw clipperException("ProcessIntersections error");
+  }
+  m_SortedEdges = 0;
+  return true;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::DisposeIntersectNodes() {
+  for (size_t i = 0; i < m_IntersectList.size(); ++i)
+    delete m_IntersectList[i];
+  m_IntersectList.clear();
+}
+//------------------------------------------------------------------------------
+
+void Clipper::BuildIntersectList(const cInt topY) {
+  if (!m_ActiveEdges)
+    return;
+
+  // prepare for sorting ...
+  TEdge *e = m_ActiveEdges;
+  m_SortedEdges = e;
+  while (e) {
+    e->PrevInSEL = e->PrevInAEL;
+    e->NextInSEL = e->NextInAEL;
+    e->Curr.X = TopX(*e, topY);
+    e = e->NextInAEL;
+  }
+
+  // bubblesort ...
+  bool isModified;
+  do {
+    isModified = false;
+    e = m_SortedEdges;
+    while (e->NextInSEL) {
+      TEdge *eNext = e->NextInSEL;
+      IntPoint Pt;
+      if (e->Curr.X > eNext->Curr.X) {
+        IntersectPoint(*e, *eNext, Pt);
+        if (Pt.Y < topY)
+          Pt = IntPoint(TopX(*e, topY), topY);
+        IntersectNode *newNode = new IntersectNode;
+        newNode->Edge1 = e;
+        newNode->Edge2 = eNext;
+        newNode->Pt = Pt;
+        m_IntersectList.push_back(newNode);
+
+        SwapPositionsInSEL(e, eNext);
+        isModified = true;
+      } else
+        e = eNext;
+    }
+    if (e->PrevInSEL)
+      e->PrevInSEL->NextInSEL = 0;
+    else
+      break;
+  } while (isModified);
+  m_SortedEdges = 0; // important
+}
+//------------------------------------------------------------------------------
+
+void Clipper::ProcessIntersectList() {
+  for (size_t i = 0; i < m_IntersectList.size(); ++i) {
+    IntersectNode *iNode = m_IntersectList[i];
+    {
+      IntersectEdges(iNode->Edge1, iNode->Edge2, iNode->Pt);
+      SwapPositionsInAEL(iNode->Edge1, iNode->Edge2);
+    }
+    delete iNode;
+  }
+  m_IntersectList.clear();
+}
+//------------------------------------------------------------------------------
+
+bool IntersectListSort(IntersectNode *node1, IntersectNode *node2) {
+  return node2->Pt.Y < node1->Pt.Y;
+}
+//------------------------------------------------------------------------------
+
+inline bool EdgesAdjacent(const IntersectNode &inode) {
+  return (inode.Edge1->NextInSEL == inode.Edge2) ||
+         (inode.Edge1->PrevInSEL == inode.Edge2);
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::FixupIntersectionOrder() {
+  // pre-condition: intersections are sorted Bottom-most first.
+  // Now it's crucial that intersections are made only between adjacent edges,
+  // so to ensure this the order of intersections may need adjusting ...
+  CopyAELToSEL();
+  std::sort(m_IntersectList.begin(), m_IntersectList.end(), IntersectListSort);
+  size_t cnt = m_IntersectList.size();
+  for (size_t i = 0; i < cnt; ++i) {
+    if (!EdgesAdjacent(*m_IntersectList[i])) {
+      size_t j = i + 1;
+      while (j < cnt && !EdgesAdjacent(*m_IntersectList[j]))
+        j++;
+      if (j == cnt)
+        return false;
+      std::swap(m_IntersectList[i], m_IntersectList[j]);
+    }
+    SwapPositionsInSEL(m_IntersectList[i]->Edge1, m_IntersectList[i]->Edge2);
+  }
+  return true;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::DoMaxima(TEdge *e) {
+  TEdge *eMaxPair = GetMaximaPairEx(e);
+  if (!eMaxPair) {
+    if (e->OutIdx >= 0)
+      AddOutPt(e, e->Top);
+    DeleteFromAEL(e);
+    return;
+  }
+
+  TEdge *eNext = e->NextInAEL;
+  while (eNext && eNext != eMaxPair) {
+    IntersectEdges(e, eNext, e->Top);
+    SwapPositionsInAEL(e, eNext);
+    eNext = e->NextInAEL;
+  }
+
+  if (e->OutIdx == Unassigned && eMaxPair->OutIdx == Unassigned) {
+    DeleteFromAEL(e);
+    DeleteFromAEL(eMaxPair);
+  } else if (e->OutIdx >= 0 && eMaxPair->OutIdx >= 0) {
+    if (e->OutIdx >= 0)
+      AddLocalMaxPoly(e, eMaxPair, e->Top);
+    DeleteFromAEL(e);
+    DeleteFromAEL(eMaxPair);
+  }
+#ifdef use_lines
+  else if (e->WindDelta == 0) {
+    if (e->OutIdx >= 0) {
+      AddOutPt(e, e->Top);
+      e->OutIdx = Unassigned;
+    }
+    DeleteFromAEL(e);
+
+    if (eMaxPair->OutIdx >= 0) {
+      AddOutPt(eMaxPair, e->Top);
+      eMaxPair->OutIdx = Unassigned;
+    }
+    DeleteFromAEL(eMaxPair);
+  }
+#endif
+  else
+    throw clipperException("DoMaxima error");
+}
+//------------------------------------------------------------------------------
+
+void Clipper::ProcessEdgesAtTopOfScanbeam(const cInt topY) {
+  TEdge *e = m_ActiveEdges;
+  while (e) {
+    // 1. process maxima, treating them as if they're 'bent' horizontal edges,
+    //    but exclude maxima with horizontal edges. nb: e can't be a horizontal.
+    bool IsMaximaEdge = IsMaxima(e, topY);
+
+    if (IsMaximaEdge) {
+      TEdge *eMaxPair = GetMaximaPairEx(e);
+      IsMaximaEdge = (!eMaxPair || !IsHorizontal(*eMaxPair));
+    }
+
+    if (IsMaximaEdge) {
+      if (m_StrictSimple)
+        m_Maxima.push_back(e->Top.X);
+      TEdge *ePrev = e->PrevInAEL;
+      DoMaxima(e);
+      if (!ePrev)
+        e = m_ActiveEdges;
+      else
+        e = ePrev->NextInAEL;
+    } else {
+      // 2. promote horizontal edges, otherwise update Curr.X and Curr.Y ...
+      if (IsIntermediate(e, topY) && IsHorizontal(*e->NextInLML)) {
+        UpdateEdgeIntoAEL(e);
+        if (e->OutIdx >= 0)
+          AddOutPt(e, e->Bot);
+        AddEdgeToSEL(e);
+      } else {
+        e->Curr.X = TopX(*e, topY);
+        e->Curr.Y = topY;
+#ifdef use_xyz
+        e->Curr.Z =
+            topY == e->Top.Y ? e->Top.Z : (topY == e->Bot.Y ? e->Bot.Z : 0);
+#endif
+      }
+
+      // When StrictlySimple and 'e' is being touched by another edge, then
+      // make sure both edges have a vertex here ...
+      if (m_StrictSimple) {
+        TEdge *ePrev = e->PrevInAEL;
+        if ((e->OutIdx >= 0) && (e->WindDelta != 0) && ePrev &&
+            (ePrev->OutIdx >= 0) && (ePrev->Curr.X == e->Curr.X) &&
+            (ePrev->WindDelta != 0)) {
+          IntPoint pt = e->Curr;
+#ifdef use_xyz
+          SetZ(pt, *ePrev, *e);
+#endif
+          OutPt *op = AddOutPt(ePrev, pt);
+          OutPt *op2 = AddOutPt(e, pt);
+          AddJoin(op, op2, pt); // StrictlySimple (type-3) join
+        }
+      }
+
+      e = e->NextInAEL;
+    }
+  }
+
+  // 3. Process horizontals at the Top of the scanbeam ...
+  m_Maxima.sort();
+  ProcessHorizontals();
+  m_Maxima.clear();
+
+  // 4. Promote intermediate vertices ...
+  e = m_ActiveEdges;
+  while (e) {
+    if (IsIntermediate(e, topY)) {
+      OutPt *op = 0;
+      if (e->OutIdx >= 0)
+        op = AddOutPt(e, e->Top);
+      UpdateEdgeIntoAEL(e);
+
+      // if output polygons share an edge, they'll need joining later ...
+      TEdge *ePrev = e->PrevInAEL;
+      TEdge *eNext = e->NextInAEL;
+      if (ePrev && ePrev->Curr.X == e->Bot.X && ePrev->Curr.Y == e->Bot.Y &&
+          op && ePrev->OutIdx >= 0 && ePrev->Curr.Y > ePrev->Top.Y &&
+          SlopesEqual(e->Curr, e->Top, ePrev->Curr, ePrev->Top,
+                      m_UseFullRange) &&
+          (e->WindDelta != 0) && (ePrev->WindDelta != 0)) {
+        OutPt *op2 = AddOutPt(ePrev, e->Bot);
+        AddJoin(op, op2, e->Top);
+      } else if (eNext && eNext->Curr.X == e->Bot.X &&
+                 eNext->Curr.Y == e->Bot.Y && op && eNext->OutIdx >= 0 &&
+                 eNext->Curr.Y > eNext->Top.Y &&
+                 SlopesEqual(e->Curr, e->Top, eNext->Curr, eNext->Top,
+                             m_UseFullRange) &&
+                 (e->WindDelta != 0) && (eNext->WindDelta != 0)) {
+        OutPt *op2 = AddOutPt(eNext, e->Bot);
+        AddJoin(op, op2, e->Top);
+      }
+    }
+    e = e->NextInAEL;
+  }
+}
+//------------------------------------------------------------------------------
+
+void Clipper::FixupOutPolyline(OutRec &outrec) {
+  OutPt *pp = outrec.Pts;
+  OutPt *lastPP = pp->Prev;
+  while (pp != lastPP) {
+    pp = pp->Next;
+    if (pp->Pt == pp->Prev->Pt) {
+      if (pp == lastPP)
+        lastPP = pp->Prev;
+      OutPt *tmpPP = pp->Prev;
+      tmpPP->Next = pp->Next;
+      pp->Next->Prev = tmpPP;
+      delete pp;
+      pp = tmpPP;
+    }
+  }
+
+  if (pp == pp->Prev) {
+    DisposeOutPts(pp);
+    outrec.Pts = 0;
+    return;
+  }
+}
+//------------------------------------------------------------------------------
+
+void Clipper::FixupOutPolygon(OutRec &outrec) {
+  // FixupOutPolygon() - removes duplicate points and simplifies consecutive
+  // parallel edges by removing the middle vertex.
+  OutPt *lastOK = 0;
+  outrec.BottomPt = 0;
+  OutPt *pp = outrec.Pts;
+  bool preserveCol = m_PreserveCollinear || m_StrictSimple;
+
+  for (;;) {
+    if (pp->Prev == pp || pp->Prev == pp->Next) {
+      DisposeOutPts(pp);
+      outrec.Pts = 0;
+      return;
+    }
+
+    // test for duplicate points and collinear edges ...
+    if ((pp->Pt == pp->Next->Pt) || (pp->Pt == pp->Prev->Pt) ||
+        (SlopesEqual(pp->Prev->Pt, pp->Pt, pp->Next->Pt, m_UseFullRange) &&
+         (!preserveCol ||
+          !Pt2IsBetweenPt1AndPt3(pp->Prev->Pt, pp->Pt, pp->Next->Pt)))) {
+      lastOK = 0;
+      OutPt *tmp = pp;
+      pp->Prev->Next = pp->Next;
+      pp->Next->Prev = pp->Prev;
+      pp = pp->Prev;
+      delete tmp;
+    } else if (pp == lastOK)
+      break;
+    else {
+      if (!lastOK)
+        lastOK = pp;
+      pp = pp->Next;
+    }
+  }
+  outrec.Pts = pp;
+}
+//------------------------------------------------------------------------------
+
+int PointCount(OutPt *Pts) {
+  if (!Pts)
+    return 0;
+  int result = 0;
+  OutPt *p = Pts;
+  do {
+    result++;
+    p = p->Next;
+  } while (p != Pts);
+  return result;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::BuildResult(Paths &polys) {
+  polys.reserve(m_PolyOuts.size());
+  for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); ++i) {
+    if (!m_PolyOuts[i]->Pts)
+      continue;
+    Path pg;
+    OutPt *p = m_PolyOuts[i]->Pts->Prev;
+    int cnt = PointCount(p);
+    if (cnt < 2)
+      continue;
+    pg.reserve(cnt);
+    for (int i = 0; i < cnt; ++i) {
+      pg.push_back(p->Pt);
+      p = p->Prev;
+    }
+    polys.push_back(pg);
+  }
+}
+//------------------------------------------------------------------------------
+
+void Clipper::BuildResult2(PolyTree &polytree) {
+  polytree.Clear();
+  polytree.AllNodes.reserve(m_PolyOuts.size());
+  // add each output polygon/contour to polytree ...
+  for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); i++) {
+    OutRec *outRec = m_PolyOuts[i];
+    int cnt = PointCount(outRec->Pts);
+    if ((outRec->IsOpen && cnt < 2) || (!outRec->IsOpen && cnt < 3))
+      continue;
+    FixHoleLinkage(*outRec);
+    PolyNode *pn = new PolyNode();
+    // nb: polytree takes ownership of all the PolyNodes
+    polytree.AllNodes.push_back(pn);
+    outRec->PolyNd = pn;
+    pn->Parent = 0;
+    pn->Index = 0;
+    pn->Contour.reserve(cnt);
+    OutPt *op = outRec->Pts->Prev;
+    for (int j = 0; j < cnt; j++) {
+      pn->Contour.push_back(op->Pt);
+      op = op->Prev;
+    }
+  }
+
+  // fixup PolyNode links etc ...
+  polytree.Childs.reserve(m_PolyOuts.size());
+  for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); i++) {
+    OutRec *outRec = m_PolyOuts[i];
+    if (!outRec->PolyNd)
+      continue;
+    if (outRec->IsOpen) {
+      outRec->PolyNd->m_IsOpen = true;
+      polytree.AddChild(*outRec->PolyNd);
+    } else if (outRec->FirstLeft && outRec->FirstLeft->PolyNd)
+      outRec->FirstLeft->PolyNd->AddChild(*outRec->PolyNd);
+    else
+      polytree.AddChild(*outRec->PolyNd);
+  }
+}
+//------------------------------------------------------------------------------
+
+void SwapIntersectNodes(IntersectNode &int1, IntersectNode &int2) {
+  // just swap the contents (because fIntersectNodes is a single-linked-list)
+  IntersectNode inode = int1; // gets a copy of Int1
+  int1.Edge1 = int2.Edge1;
+  int1.Edge2 = int2.Edge2;
+  int1.Pt = int2.Pt;
+  int2.Edge1 = inode.Edge1;
+  int2.Edge2 = inode.Edge2;
+  int2.Pt = inode.Pt;
+}
+//------------------------------------------------------------------------------
+
+inline bool E2InsertsBeforeE1(TEdge &e1, TEdge &e2) {
+  if (e2.Curr.X == e1.Curr.X) {
+    if (e2.Top.Y > e1.Top.Y)
+      return e2.Top.X < TopX(e1, e2.Top.Y);
+    else
+      return e1.Top.X > TopX(e2, e1.Top.Y);
+  } else
+    return e2.Curr.X < e1.Curr.X;
+}
+//------------------------------------------------------------------------------
+
+bool GetOverlap(const cInt a1, const cInt a2, const cInt b1, const cInt b2,
+                cInt &Left, cInt &Right) {
+  if (a1 < a2) {
+    if (b1 < b2) {
+      Left = std::max(a1, b1);
+      Right = std::min(a2, b2);
+    } else {
+      Left = std::max(a1, b2);
+      Right = std::min(a2, b1);
+    }
+  } else {
+    if (b1 < b2) {
+      Left = std::max(a2, b1);
+      Right = std::min(a1, b2);
+    } else {
+      Left = std::max(a2, b2);
+      Right = std::min(a1, b1);
+    }
+  }
+  return Left < Right;
+}
+//------------------------------------------------------------------------------
+
+inline void UpdateOutPtIdxs(OutRec &outrec) {
+  OutPt *op = outrec.Pts;
+  do {
+    op->Idx = outrec.Idx;
+    op = op->Prev;
+  } while (op != outrec.Pts);
+}
+//------------------------------------------------------------------------------
+
+void Clipper::InsertEdgeIntoAEL(TEdge *edge, TEdge *startEdge) {
+  if (!m_ActiveEdges) {
+    edge->PrevInAEL = 0;
+    edge->NextInAEL = 0;
+    m_ActiveEdges = edge;
+  } else if (!startEdge && E2InsertsBeforeE1(*m_ActiveEdges, *edge)) {
+    edge->PrevInAEL = 0;
+    edge->NextInAEL = m_ActiveEdges;
+    m_ActiveEdges->PrevInAEL = edge;
+    m_ActiveEdges = edge;
+  } else {
+    if (!startEdge)
+      startEdge = m_ActiveEdges;
+    while (startEdge->NextInAEL &&
+           !E2InsertsBeforeE1(*startEdge->NextInAEL, *edge))
+      startEdge = startEdge->NextInAEL;
+    edge->NextInAEL = startEdge->NextInAEL;
+    if (startEdge->NextInAEL)
+      startEdge->NextInAEL->PrevInAEL = edge;
+    edge->PrevInAEL = startEdge;
+    startEdge->NextInAEL = edge;
+  }
+}
+//----------------------------------------------------------------------
+
+OutPt *DupOutPt(OutPt *outPt, bool InsertAfter) {
+  OutPt *result = new OutPt;
+  result->Pt = outPt->Pt;
+  result->Idx = outPt->Idx;
+  if (InsertAfter) {
+    result->Next = outPt->Next;
+    result->Prev = outPt;
+    outPt->Next->Prev = result;
+    outPt->Next = result;
+  } else {
+    result->Prev = outPt->Prev;
+    result->Next = outPt;
+    outPt->Prev->Next = result;
+    outPt->Prev = result;
+  }
+  return result;
+}
+//------------------------------------------------------------------------------
+
+bool JoinHorz(OutPt *op1, OutPt *op1b, OutPt *op2, OutPt *op2b,
+              const IntPoint Pt, bool DiscardLeft) {
+  Direction Dir1 = (op1->Pt.X > op1b->Pt.X ? dRightToLeft : dLeftToRight);
+  Direction Dir2 = (op2->Pt.X > op2b->Pt.X ? dRightToLeft : dLeftToRight);
+  if (Dir1 == Dir2)
+    return false;
+
+  // When DiscardLeft, we want Op1b to be on the Left of Op1, otherwise we
+  // want Op1b to be on the Right. (And likewise with Op2 and Op2b.)
+  // So, to facilitate this while inserting Op1b and Op2b ...
+  // when DiscardLeft, make sure we're AT or RIGHT of Pt before adding Op1b,
+  // otherwise make sure we're AT or LEFT of Pt. (Likewise with Op2b.)
+  if (Dir1 == dLeftToRight) {
+    while (op1->Next->Pt.X <= Pt.X && op1->Next->Pt.X >= op1->Pt.X &&
+           op1->Next->Pt.Y == Pt.Y)
+      op1 = op1->Next;
+    if (DiscardLeft && (op1->Pt.X != Pt.X))
+      op1 = op1->Next;
+    op1b = DupOutPt(op1, !DiscardLeft);
+    if (op1b->Pt != Pt) {
+      op1 = op1b;
+      op1->Pt = Pt;
+      op1b = DupOutPt(op1, !DiscardLeft);
+    }
+  } else {
+    while (op1->Next->Pt.X >= Pt.X && op1->Next->Pt.X <= op1->Pt.X &&
+           op1->Next->Pt.Y == Pt.Y)
+      op1 = op1->Next;
+    if (!DiscardLeft && (op1->Pt.X != Pt.X))
+      op1 = op1->Next;
+    op1b = DupOutPt(op1, DiscardLeft);
+    if (op1b->Pt != Pt) {
+      op1 = op1b;
+      op1->Pt = Pt;
+      op1b = DupOutPt(op1, DiscardLeft);
+    }
+  }
+
+  if (Dir2 == dLeftToRight) {
+    while (op2->Next->Pt.X <= Pt.X && op2->Next->Pt.X >= op2->Pt.X &&
+           op2->Next->Pt.Y == Pt.Y)
+      op2 = op2->Next;
+    if (DiscardLeft && (op2->Pt.X != Pt.X))
+      op2 = op2->Next;
+    op2b = DupOutPt(op2, !DiscardLeft);
+    if (op2b->Pt != Pt) {
+      op2 = op2b;
+      op2->Pt = Pt;
+      op2b = DupOutPt(op2, !DiscardLeft);
+    };
+  } else {
+    while (op2->Next->Pt.X >= Pt.X && op2->Next->Pt.X <= op2->Pt.X &&
+           op2->Next->Pt.Y == Pt.Y)
+      op2 = op2->Next;
+    if (!DiscardLeft && (op2->Pt.X != Pt.X))
+      op2 = op2->Next;
+    op2b = DupOutPt(op2, DiscardLeft);
+    if (op2b->Pt != Pt) {
+      op2 = op2b;
+      op2->Pt = Pt;
+      op2b = DupOutPt(op2, DiscardLeft);
+    };
+  };
+
+  if ((Dir1 == dLeftToRight) == DiscardLeft) {
+    op1->Prev = op2;
+    op2->Next = op1;
+    op1b->Next = op2b;
+    op2b->Prev = op1b;
+  } else {
+    op1->Next = op2;
+    op2->Prev = op1;
+    op1b->Prev = op2b;
+    op2b->Next = op1b;
+  }
+  return true;
+}
+//------------------------------------------------------------------------------
+
+bool Clipper::JoinPoints(Join *j, OutRec *outRec1, OutRec *outRec2) {
+  OutPt *op1 = j->OutPt1, *op1b;
+  OutPt *op2 = j->OutPt2, *op2b;
+
+  // There are 3 kinds of joins for output polygons ...
+  // 1. Horizontal joins where Join.OutPt1 & Join.OutPt2 are vertices anywhere
+  // along (horizontal) collinear edges (& Join.OffPt is on the same
+  // horizontal).
+  // 2. Non-horizontal joins where Join.OutPt1 & Join.OutPt2 are at the same
+  // location at the Bottom of the overlapping segment (& Join.OffPt is above).
+  // 3. StrictSimple joins where edges touch but are not collinear and where
+  // Join.OutPt1, Join.OutPt2 & Join.OffPt all share the same point.
+  bool isHorizontal = (j->OutPt1->Pt.Y == j->OffPt.Y);
+
+  if (isHorizontal && (j->OffPt == j->OutPt1->Pt) &&
+      (j->OffPt == j->OutPt2->Pt)) {
+    // Strictly Simple join ...
+    if (outRec1 != outRec2)
+      return false;
+    op1b = j->OutPt1->Next;
+    while (op1b != op1 && (op1b->Pt == j->OffPt))
+      op1b = op1b->Next;
+    bool reverse1 = (op1b->Pt.Y > j->OffPt.Y);
+    op2b = j->OutPt2->Next;
+    while (op2b != op2 && (op2b->Pt == j->OffPt))
+      op2b = op2b->Next;
+    bool reverse2 = (op2b->Pt.Y > j->OffPt.Y);
+    if (reverse1 == reverse2)
+      return false;
+    if (reverse1) {
+      op1b = DupOutPt(op1, false);
+      op2b = DupOutPt(op2, true);
+      op1->Prev = op2;
+      op2->Next = op1;
+      op1b->Next = op2b;
+      op2b->Prev = op1b;
+      j->OutPt1 = op1;
+      j->OutPt2 = op1b;
+      return true;
+    } else {
+      op1b = DupOutPt(op1, true);
+      op2b = DupOutPt(op2, false);
+      op1->Next = op2;
+      op2->Prev = op1;
+      op1b->Prev = op2b;
+      op2b->Next = op1b;
+      j->OutPt1 = op1;
+      j->OutPt2 = op1b;
+      return true;
+    }
+  } else if (isHorizontal) {
+    // treat horizontal joins differently to non-horizontal joins since with
+    // them we're not yet sure where the overlapping is. OutPt1.Pt & OutPt2.Pt
+    // may be anywhere along the horizontal edge.
+    op1b = op1;
+    while (op1->Prev->Pt.Y == op1->Pt.Y && op1->Prev != op1b &&
+           op1->Prev != op2)
+      op1 = op1->Prev;
+    while (op1b->Next->Pt.Y == op1b->Pt.Y && op1b->Next != op1 &&
+           op1b->Next != op2)
+      op1b = op1b->Next;
+    if (op1b->Next == op1 || op1b->Next == op2)
+      return false; // a flat 'polygon'
+
+    op2b = op2;
+    while (op2->Prev->Pt.Y == op2->Pt.Y && op2->Prev != op2b &&
+           op2->Prev != op1b)
+      op2 = op2->Prev;
+    while (op2b->Next->Pt.Y == op2b->Pt.Y && op2b->Next != op2 &&
+           op2b->Next != op1)
+      op2b = op2b->Next;
+    if (op2b->Next == op2 || op2b->Next == op1)
+      return false; // a flat 'polygon'
+
+    cInt Left, Right;
+    // Op1 --> Op1b & Op2 --> Op2b are the extremites of the horizontal edges
+    if (!GetOverlap(op1->Pt.X, op1b->Pt.X, op2->Pt.X, op2b->Pt.X, Left, Right))
+      return false;
+
+    // DiscardLeftSide: when overlapping edges are joined, a spike will created
+    // which needs to be cleaned up. However, we don't want Op1 or Op2 caught up
+    // on the discard Side as either may still be needed for other joins ...
+    IntPoint Pt;
+    bool DiscardLeftSide;
+    if (op1->Pt.X >= Left && op1->Pt.X <= Right) {
+      Pt = op1->Pt;
+      DiscardLeftSide = (op1->Pt.X > op1b->Pt.X);
+    } else if (op2->Pt.X >= Left && op2->Pt.X <= Right) {
+      Pt = op2->Pt;
+      DiscardLeftSide = (op2->Pt.X > op2b->Pt.X);
+    } else if (op1b->Pt.X >= Left && op1b->Pt.X <= Right) {
+      Pt = op1b->Pt;
+      DiscardLeftSide = op1b->Pt.X > op1->Pt.X;
+    } else {
+      Pt = op2b->Pt;
+      DiscardLeftSide = (op2b->Pt.X > op2->Pt.X);
+    }
+    j->OutPt1 = op1;
+    j->OutPt2 = op2;
+    return JoinHorz(op1, op1b, op2, op2b, Pt, DiscardLeftSide);
+  } else {
+    // nb: For non-horizontal joins ...
+    //     1. Jr.OutPt1.Pt.Y == Jr.OutPt2.Pt.Y
+    //     2. Jr.OutPt1.Pt > Jr.OffPt.Y
+
+    // make sure the polygons are correctly oriented ...
+    op1b = op1->Next;
+    while ((op1b->Pt == op1->Pt) && (op1b != op1))
+      op1b = op1b->Next;
+    bool Reverse1 = ((op1b->Pt.Y > op1->Pt.Y) ||
+                     !SlopesEqual(op1->Pt, op1b->Pt, j->OffPt, m_UseFullRange));
+    if (Reverse1) {
+      op1b = op1->Prev;
+      while ((op1b->Pt == op1->Pt) && (op1b != op1))
+        op1b = op1b->Prev;
+      if ((op1b->Pt.Y > op1->Pt.Y) ||
+          !SlopesEqual(op1->Pt, op1b->Pt, j->OffPt, m_UseFullRange))
+        return false;
+    };
+    op2b = op2->Next;
+    while ((op2b->Pt == op2->Pt) && (op2b != op2))
+      op2b = op2b->Next;
+    bool Reverse2 = ((op2b->Pt.Y > op2->Pt.Y) ||
+                     !SlopesEqual(op2->Pt, op2b->Pt, j->OffPt, m_UseFullRange));
+    if (Reverse2) {
+      op2b = op2->Prev;
+      while ((op2b->Pt == op2->Pt) && (op2b != op2))
+        op2b = op2b->Prev;
+      if ((op2b->Pt.Y > op2->Pt.Y) ||
+          !SlopesEqual(op2->Pt, op2b->Pt, j->OffPt, m_UseFullRange))
+        return false;
+    }
+
+    if ((op1b == op1) || (op2b == op2) || (op1b == op2b) ||
+        ((outRec1 == outRec2) && (Reverse1 == Reverse2)))
+      return false;
+
+    if (Reverse1) {
+      op1b = DupOutPt(op1, false);
+      op2b = DupOutPt(op2, true);
+      op1->Prev = op2;
+      op2->Next = op1;
+      op1b->Next = op2b;
+      op2b->Prev = op1b;
+      j->OutPt1 = op1;
+      j->OutPt2 = op1b;
+      return true;
+    } else {
+      op1b = DupOutPt(op1, true);
+      op2b = DupOutPt(op2, false);
+      op1->Next = op2;
+      op2->Prev = op1;
+      op1b->Prev = op2b;
+      op2b->Next = op1b;
+      j->OutPt1 = op1;
+      j->OutPt2 = op1b;
+      return true;
+    }
+  }
+}
+//----------------------------------------------------------------------
+
+static OutRec *ParseFirstLeft(OutRec *FirstLeft) {
+  while (FirstLeft && !FirstLeft->Pts)
+    FirstLeft = FirstLeft->FirstLeft;
+  return FirstLeft;
+}
+//------------------------------------------------------------------------------
+
+void Clipper::FixupFirstLefts1(OutRec *OldOutRec, OutRec *NewOutRec) {
+  // tests if NewOutRec contains the polygon before reassigning FirstLeft
+  for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); ++i) {
+    OutRec *outRec = m_PolyOuts[i];
+    OutRec *firstLeft = ParseFirstLeft(outRec->FirstLeft);
+    if (outRec->Pts && firstLeft == OldOutRec) {
+      if (Poly2ContainsPoly1(outRec->Pts, NewOutRec->Pts))
+        outRec->FirstLeft = NewOutRec;
+    }
+  }
+}
+//----------------------------------------------------------------------
+
+void Clipper::FixupFirstLefts2(OutRec *InnerOutRec, OutRec *OuterOutRec) {
+  // A polygon has split into two such that one is now the inner of the other.
+  // It's possible that these polygons now wrap around other polygons, so check
+  // every polygon that's also contained by OuterOutRec's FirstLeft container
+  //(including 0) to see if they've become inner to the new inner polygon ...
+  OutRec *orfl = OuterOutRec->FirstLeft;
+  for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); ++i) {
+    OutRec *outRec = m_PolyOuts[i];
+
+    if (!outRec->Pts || outRec == OuterOutRec || outRec == InnerOutRec)
+      continue;
+    OutRec *firstLeft = ParseFirstLeft(outRec->FirstLeft);
+    if (firstLeft != orfl && firstLeft != InnerOutRec &&
+        firstLeft != OuterOutRec)
+      continue;
+    if (Poly2ContainsPoly1(outRec->Pts, InnerOutRec->Pts))
+      outRec->FirstLeft = InnerOutRec;
+    else if (Poly2ContainsPoly1(outRec->Pts, OuterOutRec->Pts))
+      outRec->FirstLeft = OuterOutRec;
+    else if (outRec->FirstLeft == InnerOutRec ||
+             outRec->FirstLeft == OuterOutRec)
+      outRec->FirstLeft = orfl;
+  }
+}
+//----------------------------------------------------------------------
+void Clipper::FixupFirstLefts3(OutRec *OldOutRec, OutRec *NewOutRec) {
+  // reassigns FirstLeft WITHOUT testing if NewOutRec contains the polygon
+  for (PolyOutList::size_type i = 0; i < m_PolyOuts.size(); ++i) {
+    OutRec *outRec = m_PolyOuts[i];
+    OutRec *firstLeft = ParseFirstLeft(outRec->FirstLeft);
+    if (outRec->Pts && firstLeft == OldOutRec)
+      outRec->FirstLeft = NewOutRec;
+  }
+}
+//----------------------------------------------------------------------
+
+void Clipper::JoinCommonEdges() {
+  for (JoinList::size_type i = 0; i < m_Joins.size(); i++) {
+    Join *join = m_Joins[i];
+
+    OutRec *outRec1 = GetOutRec(join->OutPt1->Idx);
+    OutRec *outRec2 = GetOutRec(join->OutPt2->Idx);
+
+    if (!outRec1->Pts || !outRec2->Pts)
+      continue;
+    if (outRec1->IsOpen || outRec2->IsOpen)
+      continue;
+
+    // get the polygon fragment with the correct hole state (FirstLeft)
+    // before calling JoinPoints() ...
+    OutRec *holeStateRec;
+    if (outRec1 == outRec2)
+      holeStateRec = outRec1;
+    else if (OutRec1RightOfOutRec2(outRec1, outRec2))
+      holeStateRec = outRec2;
+    else if (OutRec1RightOfOutRec2(outRec2, outRec1))
+      holeStateRec = outRec1;
+    else
+      holeStateRec = GetLowermostRec(outRec1, outRec2);
+
+    if (!JoinPoints(join, outRec1, outRec2))
+      continue;
+
+    if (outRec1 == outRec2) {
+      // instead of joining two polygons, we've just created a new one by
+      // splitting one polygon into two.
+      outRec1->Pts = join->OutPt1;
+      outRec1->BottomPt = 0;
+      outRec2 = CreateOutRec();
+      outRec2->Pts = join->OutPt2;
+
+      // update all OutRec2.Pts Idx's ...
+      UpdateOutPtIdxs(*outRec2);
+
+      if (Poly2ContainsPoly1(outRec2->Pts, outRec1->Pts)) {
+        // outRec1 contains outRec2 ...
+        outRec2->IsHole = !outRec1->IsHole;
+        outRec2->FirstLeft = outRec1;
+
+        if (m_UsingPolyTree)
+          FixupFirstLefts2(outRec2, outRec1);
+
+        if ((outRec2->IsHole ^ m_ReverseOutput) == (Area(*outRec2) > 0))
+          ReversePolyPtLinks(outRec2->Pts);
+
+      } else if (Poly2ContainsPoly1(outRec1->Pts, outRec2->Pts)) {
+        // outRec2 contains outRec1 ...
+        outRec2->IsHole = outRec1->IsHole;
+        outRec1->IsHole = !outRec2->IsHole;
+        outRec2->FirstLeft = outRec1->FirstLeft;
+        outRec1->FirstLeft = outRec2;
+
+        if (m_UsingPolyTree)
+          FixupFirstLefts2(outRec1, outRec2);
+
+        if ((outRec1->IsHole ^ m_ReverseOutput) == (Area(*outRec1) > 0))
+          ReversePolyPtLinks(outRec1->Pts);
+      } else {
+        // the 2 polygons are completely separate ...
+        outRec2->IsHole = outRec1->IsHole;
+        outRec2->FirstLeft = outRec1->FirstLeft;
+
+        // fixup FirstLeft pointers that may need reassigning to OutRec2
+        if (m_UsingPolyTree)
+          FixupFirstLefts1(outRec1, outRec2);
+      }
+
+    } else {
+      // joined 2 polygons together ...
+
+      outRec2->Pts = 0;
+      outRec2->BottomPt = 0;
+      outRec2->Idx = outRec1->Idx;
+
+      outRec1->IsHole = holeStateRec->IsHole;
+      if (holeStateRec == outRec2)
+        outRec1->FirstLeft = outRec2->FirstLeft;
+      outRec2->FirstLeft = outRec1;
+
+      if (m_UsingPolyTree)
+        FixupFirstLefts3(outRec2, outRec1);
+    }
+  }
+}
+
+//------------------------------------------------------------------------------
+// ClipperOffset support functions ...
+//------------------------------------------------------------------------------
+
+DoublePoint GetUnitNormal(const IntPoint &pt1, const IntPoint &pt2) {
+  if (pt2.X == pt1.X && pt2.Y == pt1.Y)
+    return DoublePoint(0, 0);
+
+  double Dx = (double)(pt2.X - pt1.X);
+  double dy = (double)(pt2.Y - pt1.Y);
+  double f = 1 * 1.0 / std::sqrt(Dx * Dx + dy * dy);
+  Dx *= f;
+  dy *= f;
+  return DoublePoint(dy, -Dx);
+}
+
+//------------------------------------------------------------------------------
+// ClipperOffset class
+//------------------------------------------------------------------------------
+
+ClipperOffset::ClipperOffset(double miterLimit, double arcTolerance) {
+  this->MiterLimit = miterLimit;
+  this->ArcTolerance = arcTolerance;
+  m_lowest.X = -1;
+}
+//------------------------------------------------------------------------------
+
+ClipperOffset::~ClipperOffset() { Clear(); }
+//------------------------------------------------------------------------------
+
+void ClipperOffset::Clear() {
+  for (int i = 0; i < m_polyNodes.ChildCount(); ++i)
+    delete m_polyNodes.Childs[i];
+  m_polyNodes.Childs.clear();
+  m_lowest.X = -1;
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::AddPath(const Path &path, JoinType joinType,
+                            EndType endType) {
+  int highI = (int)path.size() - 1;
+  if (highI < 0)
+    return;
+  PolyNode *newNode = new PolyNode();
+  newNode->m_jointype = joinType;
+  newNode->m_endtype = endType;
+
+  // strip duplicate points from path and also get index to the lowest point ...
+  if (endType == etClosedLine || endType == etClosedPolygon)
+    while (highI > 0 && path[0] == path[highI])
+      highI--;
+  newNode->Contour.reserve(highI + 1);
+  newNode->Contour.push_back(path[0]);
+  int j = 0, k = 0;
+  for (int i = 1; i <= highI; i++)
+    if (newNode->Contour[j] != path[i]) {
+      j++;
+      newNode->Contour.push_back(path[i]);
+      if (path[i].Y > newNode->Contour[k].Y ||
+          (path[i].Y == newNode->Contour[k].Y &&
+           path[i].X < newNode->Contour[k].X))
+        k = j;
+    }
+  if (endType == etClosedPolygon && j < 2) {
+    delete newNode;
+    return;
+  }
+  m_polyNodes.AddChild(*newNode);
+
+  // if this path's lowest pt is lower than all the others then update m_lowest
+  if (endType != etClosedPolygon)
+    return;
+  if (m_lowest.X < 0)
+    m_lowest = IntPoint(m_polyNodes.ChildCount() - 1, k);
+  else {
+    IntPoint ip = m_polyNodes.Childs[(int)m_lowest.X]->Contour[(int)m_lowest.Y];
+    if (newNode->Contour[k].Y > ip.Y ||
+        (newNode->Contour[k].Y == ip.Y && newNode->Contour[k].X < ip.X))
+      m_lowest = IntPoint(m_polyNodes.ChildCount() - 1, k);
+  }
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::AddPaths(const Paths &paths, JoinType joinType,
+                             EndType endType) {
+  for (Paths::size_type i = 0; i < paths.size(); ++i)
+    AddPath(paths[i], joinType, endType);
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::FixOrientations() {
+  // fixup orientations of all closed paths if the orientation of the
+  // closed path with the lowermost vertex is wrong ...
+  if (m_lowest.X >= 0 &&
+      !Orientation(m_polyNodes.Childs[(int)m_lowest.X]->Contour)) {
+    for (int i = 0; i < m_polyNodes.ChildCount(); ++i) {
+      PolyNode &node = *m_polyNodes.Childs[i];
+      if (node.m_endtype == etClosedPolygon ||
+          (node.m_endtype == etClosedLine && Orientation(node.Contour)))
+        ReversePath(node.Contour);
+    }
+  } else {
+    for (int i = 0; i < m_polyNodes.ChildCount(); ++i) {
+      PolyNode &node = *m_polyNodes.Childs[i];
+      if (node.m_endtype == etClosedLine && !Orientation(node.Contour))
+        ReversePath(node.Contour);
+    }
+  }
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::Execute(Paths &solution, double delta) {
+  solution.clear();
+  FixOrientations();
+  DoOffset(delta);
+
+  // now clean up 'corners' ...
+  Clipper clpr;
+  clpr.AddPaths(m_destPolys, ptSubject, true);
+  if (delta > 0) {
+    clpr.Execute(ctUnion, solution, pftPositive, pftPositive);
+  } else {
+    IntRect r = clpr.GetBounds();
+    Path outer(4);
+    outer[0] = IntPoint(r.left - 10, r.bottom + 10);
+    outer[1] = IntPoint(r.right + 10, r.bottom + 10);
+    outer[2] = IntPoint(r.right + 10, r.top - 10);
+    outer[3] = IntPoint(r.left - 10, r.top - 10);
+
+    clpr.AddPath(outer, ptSubject, true);
+    clpr.ReverseSolution(true);
+    clpr.Execute(ctUnion, solution, pftNegative, pftNegative);
+    if (solution.size() > 0)
+      solution.erase(solution.begin());
+  }
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::Execute(PolyTree &solution, double delta) {
+  solution.Clear();
+  FixOrientations();
+  DoOffset(delta);
+
+  // now clean up 'corners' ...
+  Clipper clpr;
+  clpr.AddPaths(m_destPolys, ptSubject, true);
+  if (delta > 0) {
+    clpr.Execute(ctUnion, solution, pftPositive, pftPositive);
+  } else {
+    IntRect r = clpr.GetBounds();
+    Path outer(4);
+    outer[0] = IntPoint(r.left - 10, r.bottom + 10);
+    outer[1] = IntPoint(r.right + 10, r.bottom + 10);
+    outer[2] = IntPoint(r.right + 10, r.top - 10);
+    outer[3] = IntPoint(r.left - 10, r.top - 10);
+
+    clpr.AddPath(outer, ptSubject, true);
+    clpr.ReverseSolution(true);
+    clpr.Execute(ctUnion, solution, pftNegative, pftNegative);
+    // remove the outer PolyNode rectangle ...
+    if (solution.ChildCount() == 1 && solution.Childs[0]->ChildCount() > 0) {
+      PolyNode *outerNode = solution.Childs[0];
+      solution.Childs.reserve(outerNode->ChildCount());
+      solution.Childs[0] = outerNode->Childs[0];
+      solution.Childs[0]->Parent = outerNode->Parent;
+      for (int i = 1; i < outerNode->ChildCount(); ++i)
+        solution.AddChild(*outerNode->Childs[i]);
+    } else
+      solution.Clear();
+  }
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::DoOffset(double delta) {
+  m_destPolys.clear();
+  m_delta = delta;
+
+  // if Zero offset, just copy any CLOSED polygons to m_p and return ...
+  if (NEAR_ZERO(delta)) {
+    m_destPolys.reserve(m_polyNodes.ChildCount());
+    for (int i = 0; i < m_polyNodes.ChildCount(); i++) {
+      PolyNode &node = *m_polyNodes.Childs[i];
+      if (node.m_endtype == etClosedPolygon)
+        m_destPolys.push_back(node.Contour);
+    }
+    return;
+  }
+
+  // see offset_triginometry3.svg in the documentation folder ...
+  if (MiterLimit > 2)
+    m_miterLim = 2 / (MiterLimit * MiterLimit);
+  else
+    m_miterLim = 0.5;
+
+  double y;
+  if (ArcTolerance <= 0.0)
+    y = def_arc_tolerance;
+  else if (ArcTolerance > std::fabs(delta) * def_arc_tolerance)
+    y = std::fabs(delta) * def_arc_tolerance;
+  else
+    y = ArcTolerance;
+  // see offset_triginometry2.svg in the documentation folder ...
+  double steps = pi / std::acos(1 - y / std::fabs(delta));
+  if (steps > std::fabs(delta) * pi)
+    steps = std::fabs(delta) * pi; // ie excessive precision check
+  m_sin = std::sin(two_pi / steps);
+  m_cos = std::cos(two_pi / steps);
+  m_StepsPerRad = steps / two_pi;
+  if (delta < 0.0)
+    m_sin = -m_sin;
+
+  m_destPolys.reserve(m_polyNodes.ChildCount() * 2);
+  for (int i = 0; i < m_polyNodes.ChildCount(); i++) {
+    PolyNode &node = *m_polyNodes.Childs[i];
+    m_srcPoly = node.Contour;
+
+    int len = (int)m_srcPoly.size();
+    if (len == 0 ||
+        (delta <= 0 && (len < 3 || node.m_endtype != etClosedPolygon)))
+      continue;
+
+    m_destPoly.clear();
+    if (len == 1) {
+      if (node.m_jointype == jtRound) {
+        double X = 1.0, Y = 0.0;
+        for (cInt j = 1; j <= steps; j++) {
+          m_destPoly.push_back(IntPoint(Round(m_srcPoly[0].X + X * delta),
+                                        Round(m_srcPoly[0].Y + Y * delta)));
+          double X2 = X;
+          X = X * m_cos - m_sin * Y;
+          Y = X2 * m_sin + Y * m_cos;
+        }
+      } else {
+        double X = -1.0, Y = -1.0;
+        for (int j = 0; j < 4; ++j) {
+          m_destPoly.push_back(IntPoint(Round(m_srcPoly[0].X + X * delta),
+                                        Round(m_srcPoly[0].Y + Y * delta)));
+          if (X < 0)
+            X = 1;
+          else if (Y < 0)
+            Y = 1;
+          else
+            X = -1;
+        }
+      }
+      m_destPolys.push_back(m_destPoly);
+      continue;
+    }
+    // build m_normals ...
+    m_normals.clear();
+    m_normals.reserve(len);
+    for (int j = 0; j < len - 1; ++j)
+      m_normals.push_back(GetUnitNormal(m_srcPoly[j], m_srcPoly[j + 1]));
+    if (node.m_endtype == etClosedLine || node.m_endtype == etClosedPolygon)
+      m_normals.push_back(GetUnitNormal(m_srcPoly[len - 1], m_srcPoly[0]));
+    else
+      m_normals.push_back(DoublePoint(m_normals[len - 2]));
+
+    if (node.m_endtype == etClosedPolygon) {
+      int k = len - 1;
+      for (int j = 0; j < len; ++j)
+        OffsetPoint(j, k, node.m_jointype);
+      m_destPolys.push_back(m_destPoly);
+    } else if (node.m_endtype == etClosedLine) {
+      int k = len - 1;
+      for (int j = 0; j < len; ++j)
+        OffsetPoint(j, k, node.m_jointype);
+      m_destPolys.push_back(m_destPoly);
+      m_destPoly.clear();
+      // re-build m_normals ...
+      DoublePoint n = m_normals[len - 1];
+      for (int j = len - 1; j > 0; j--)
+        m_normals[j] = DoublePoint(-m_normals[j - 1].X, -m_normals[j - 1].Y);
+      m_normals[0] = DoublePoint(-n.X, -n.Y);
+      k = 0;
+      for (int j = len - 1; j >= 0; j--)
+        OffsetPoint(j, k, node.m_jointype);
+      m_destPolys.push_back(m_destPoly);
+    } else {
+      int k = 0;
+      for (int j = 1; j < len - 1; ++j)
+        OffsetPoint(j, k, node.m_jointype);
+
+      IntPoint pt1;
+      if (node.m_endtype == etOpenButt) {
+        int j = len - 1;
+        pt1 = IntPoint((cInt)Round(m_srcPoly[j].X + m_normals[j].X * delta),
+                       (cInt)Round(m_srcPoly[j].Y + m_normals[j].Y * delta));
+        m_destPoly.push_back(pt1);
+        pt1 = IntPoint((cInt)Round(m_srcPoly[j].X - m_normals[j].X * delta),
+                       (cInt)Round(m_srcPoly[j].Y - m_normals[j].Y * delta));
+        m_destPoly.push_back(pt1);
+      } else {
+        int j = len - 1;
+        k = len - 2;
+        m_sinA = 0;
+        m_normals[j] = DoublePoint(-m_normals[j].X, -m_normals[j].Y);
+        if (node.m_endtype == etOpenSquare)
+          DoSquare(j, k);
+        else
+          DoRound(j, k);
+      }
+
+      // re-build m_normals ...
+      for (int j = len - 1; j > 0; j--)
+        m_normals[j] = DoublePoint(-m_normals[j - 1].X, -m_normals[j - 1].Y);
+      m_normals[0] = DoublePoint(-m_normals[1].X, -m_normals[1].Y);
+
+      k = len - 1;
+      for (int j = k - 1; j > 0; --j)
+        OffsetPoint(j, k, node.m_jointype);
+
+      if (node.m_endtype == etOpenButt) {
+        pt1 = IntPoint((cInt)Round(m_srcPoly[0].X - m_normals[0].X * delta),
+                       (cInt)Round(m_srcPoly[0].Y - m_normals[0].Y * delta));
+        m_destPoly.push_back(pt1);
+        pt1 = IntPoint((cInt)Round(m_srcPoly[0].X + m_normals[0].X * delta),
+                       (cInt)Round(m_srcPoly[0].Y + m_normals[0].Y * delta));
+        m_destPoly.push_back(pt1);
+      } else {
+        k = 1;
+        m_sinA = 0;
+        if (node.m_endtype == etOpenSquare)
+          DoSquare(0, 1);
+        else
+          DoRound(0, 1);
+      }
+      m_destPolys.push_back(m_destPoly);
+    }
+  }
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::OffsetPoint(int j, int &k, JoinType jointype) {
+  // cross product ...
+  m_sinA = (m_normals[k].X * m_normals[j].Y - m_normals[j].X * m_normals[k].Y);
+  if (std::fabs(m_sinA * m_delta) < 1.0) {
+    // dot product ...
+    double cosA =
+        (m_normals[k].X * m_normals[j].X + m_normals[j].Y * m_normals[k].Y);
+    if (cosA > 0) // angle => 0 degrees
+    {
+      m_destPoly.push_back(
+          IntPoint(Round(m_srcPoly[j].X + m_normals[k].X * m_delta),
+                   Round(m_srcPoly[j].Y + m_normals[k].Y * m_delta)));
+      return;
+    }
+    // else angle => 180 degrees
+  } else if (m_sinA > 1.0)
+    m_sinA = 1.0;
+  else if (m_sinA < -1.0)
+    m_sinA = -1.0;
+
+  if (m_sinA * m_delta < 0) {
+    m_destPoly.push_back(
+        IntPoint(Round(m_srcPoly[j].X + m_normals[k].X * m_delta),
+                 Round(m_srcPoly[j].Y + m_normals[k].Y * m_delta)));
+    m_destPoly.push_back(m_srcPoly[j]);
+    m_destPoly.push_back(
+        IntPoint(Round(m_srcPoly[j].X + m_normals[j].X * m_delta),
+                 Round(m_srcPoly[j].Y + m_normals[j].Y * m_delta)));
+  } else
+    switch (jointype) {
+    case jtMiter: {
+      double r = 1 + (m_normals[j].X * m_normals[k].X +
+                      m_normals[j].Y * m_normals[k].Y);
+      if (r >= m_miterLim)
+        DoMiter(j, k, r);
+      else
+        DoSquare(j, k);
+      break;
+    }
+    case jtSquare:
+      DoSquare(j, k);
+      break;
+    case jtRound:
+      DoRound(j, k);
+      break;
+    }
+  k = j;
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::DoSquare(int j, int k) {
+  double dx = std::tan(std::atan2(m_sinA, m_normals[k].X * m_normals[j].X +
+                                              m_normals[k].Y * m_normals[j].Y) /
+                       4);
+  m_destPoly.push_back(IntPoint(
+      Round(m_srcPoly[j].X + m_delta * (m_normals[k].X - m_normals[k].Y * dx)),
+      Round(m_srcPoly[j].Y +
+            m_delta * (m_normals[k].Y + m_normals[k].X * dx))));
+  m_destPoly.push_back(IntPoint(
+      Round(m_srcPoly[j].X + m_delta * (m_normals[j].X + m_normals[j].Y * dx)),
+      Round(m_srcPoly[j].Y +
+            m_delta * (m_normals[j].Y - m_normals[j].X * dx))));
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::DoMiter(int j, int k, double r) {
+  double q = m_delta / r;
+  m_destPoly.push_back(
+      IntPoint(Round(m_srcPoly[j].X + (m_normals[k].X + m_normals[j].X) * q),
+               Round(m_srcPoly[j].Y + (m_normals[k].Y + m_normals[j].Y) * q)));
+}
+//------------------------------------------------------------------------------
+
+void ClipperOffset::DoRound(int j, int k) {
+  double a = std::atan2(m_sinA, m_normals[k].X * m_normals[j].X +
+                                    m_normals[k].Y * m_normals[j].Y);
+  int steps = std::max((int)Round(m_StepsPerRad * std::fabs(a)), 1);
+
+  double X = m_normals[k].X, Y = m_normals[k].Y, X2;
+  for (int i = 0; i < steps; ++i) {
+    m_destPoly.push_back(IntPoint(Round(m_srcPoly[j].X + X * m_delta),
+                                  Round(m_srcPoly[j].Y + Y * m_delta)));
+    X2 = X;
+    X = X * m_cos - m_sin * Y;
+    Y = X2 * m_sin + Y * m_cos;
+  }
+  m_destPoly.push_back(
+      IntPoint(Round(m_srcPoly[j].X + m_normals[j].X * m_delta),
+               Round(m_srcPoly[j].Y + m_normals[j].Y * m_delta)));
+}
+
+//------------------------------------------------------------------------------
+// Miscellaneous public functions
+//------------------------------------------------------------------------------
+
+void Clipper::DoSimplePolygons() {
+  PolyOutList::size_type i = 0;
+  while (i < m_PolyOuts.size()) {
+    OutRec *outrec = m_PolyOuts[i++];
+    OutPt *op = outrec->Pts;
+    if (!op || outrec->IsOpen)
+      continue;
+    do // for each Pt in Polygon until duplicate found do ...
+    {
+      OutPt *op2 = op->Next;
+      while (op2 != outrec->Pts) {
+        if ((op->Pt == op2->Pt) && op2->Next != op && op2->Prev != op) {
+          // split the polygon into two ...
+          OutPt *op3 = op->Prev;
+          OutPt *op4 = op2->Prev;
+          op->Prev = op4;
+          op4->Next = op;
+          op2->Prev = op3;
+          op3->Next = op2;
+
+          outrec->Pts = op;
+          OutRec *outrec2 = CreateOutRec();
+          outrec2->Pts = op2;
+          UpdateOutPtIdxs(*outrec2);
+          if (Poly2ContainsPoly1(outrec2->Pts, outrec->Pts)) {
+            // OutRec2 is contained by OutRec1 ...
+            outrec2->IsHole = !outrec->IsHole;
+            outrec2->FirstLeft = outrec;
+            if (m_UsingPolyTree)
+              FixupFirstLefts2(outrec2, outrec);
+          } else if (Poly2ContainsPoly1(outrec->Pts, outrec2->Pts)) {
+            // OutRec1 is contained by OutRec2 ...
+            outrec2->IsHole = outrec->IsHole;
+            outrec->IsHole = !outrec2->IsHole;
+            outrec2->FirstLeft = outrec->FirstLeft;
+            outrec->FirstLeft = outrec2;
+            if (m_UsingPolyTree)
+              FixupFirstLefts2(outrec, outrec2);
+          } else {
+            // the 2 polygons are separate ...
+            outrec2->IsHole = outrec->IsHole;
+            outrec2->FirstLeft = outrec->FirstLeft;
+            if (m_UsingPolyTree)
+              FixupFirstLefts1(outrec, outrec2);
+          }
+          op2 = op; // ie get ready for the Next iteration
+        }
+        op2 = op2->Next;
+      }
+      op = op->Next;
+    } while (op != outrec->Pts);
+  }
+}
+//------------------------------------------------------------------------------
+
+void ReversePath(Path &p) { std::reverse(p.begin(), p.end()); }
+//------------------------------------------------------------------------------
+
+void ReversePaths(Paths &p) {
+  for (Paths::size_type i = 0; i < p.size(); ++i)
+    ReversePath(p[i]);
+}
+//------------------------------------------------------------------------------
+
+void SimplifyPolygon(const Path &in_poly, Paths &out_polys,
+                     PolyFillType fillType) {
+  Clipper c;
+  c.StrictlySimple(true);
+  c.AddPath(in_poly, ptSubject, true);
+  c.Execute(ctUnion, out_polys, fillType, fillType);
+}
+//------------------------------------------------------------------------------
+
+void SimplifyPolygons(const Paths &in_polys, Paths &out_polys,
+                      PolyFillType fillType) {
+  Clipper c;
+  c.StrictlySimple(true);
+  c.AddPaths(in_polys, ptSubject, true);
+  c.Execute(ctUnion, out_polys, fillType, fillType);
+}
+//------------------------------------------------------------------------------
+
+void SimplifyPolygons(Paths &polys, PolyFillType fillType) {
+  SimplifyPolygons(polys, polys, fillType);
+}
+//------------------------------------------------------------------------------
+
+inline double DistanceSqrd(const IntPoint &pt1, const IntPoint &pt2) {
+  double Dx = ((double)pt1.X - pt2.X);
+  double dy = ((double)pt1.Y - pt2.Y);
+  return (Dx * Dx + dy * dy);
+}
+//------------------------------------------------------------------------------
+
+double DistanceFromLineSqrd(const IntPoint &pt, const IntPoint &ln1,
+                            const IntPoint &ln2) {
+  // The equation of a line in general form (Ax + By + C = 0)
+  // given 2 points (x�,y�) & (x�,y�) is ...
+  //(y� - y�)x + (x� - x�)y + (y� - y�)x� - (x� - x�)y� = 0
+  // A = (y� - y�); B = (x� - x�); C = (y� - y�)x� - (x� - x�)y�
+  // perpendicular distance of point (x�,y�) = (Ax� + By� + C)/Sqrt(A� + B�)
+  // see http://en.wikipedia.org/wiki/Perpendicular_distance
+  double A = double(ln1.Y - ln2.Y);
+  double B = double(ln2.X - ln1.X);
+  double C = A * ln1.X + B * ln1.Y;
+  C = A * pt.X + B * pt.Y - C;
+  return (C * C) / (A * A + B * B);
+}
+//---------------------------------------------------------------------------
+
+bool SlopesNearCollinear(const IntPoint &pt1, const IntPoint &pt2,
+                         const IntPoint &pt3, double distSqrd) {
+  // this function is more accurate when the point that's geometrically
+  // between the other 2 points is the one that's tested for distance.
+  // ie makes it more likely to pick up 'spikes' ...
+  if (Abs(pt1.X - pt2.X) > Abs(pt1.Y - pt2.Y)) {
+    if ((pt1.X > pt2.X) == (pt1.X < pt3.X))
+      return DistanceFromLineSqrd(pt1, pt2, pt3) < distSqrd;
+    else if ((pt2.X > pt1.X) == (pt2.X < pt3.X))
+      return DistanceFromLineSqrd(pt2, pt1, pt3) < distSqrd;
+    else
+      return DistanceFromLineSqrd(pt3, pt1, pt2) < distSqrd;
+  } else {
+    if ((pt1.Y > pt2.Y) == (pt1.Y < pt3.Y))
+      return DistanceFromLineSqrd(pt1, pt2, pt3) < distSqrd;
+    else if ((pt2.Y > pt1.Y) == (pt2.Y < pt3.Y))
+      return DistanceFromLineSqrd(pt2, pt1, pt3) < distSqrd;
+    else
+      return DistanceFromLineSqrd(pt3, pt1, pt2) < distSqrd;
+  }
+}
+//------------------------------------------------------------------------------
+
+bool PointsAreClose(IntPoint pt1, IntPoint pt2, double distSqrd) {
+  double Dx = (double)pt1.X - pt2.X;
+  double dy = (double)pt1.Y - pt2.Y;
+  return ((Dx * Dx) + (dy * dy) <= distSqrd);
+}
+//------------------------------------------------------------------------------
+
+OutPt *ExcludeOp(OutPt *op) {
+  OutPt *result = op->Prev;
+  result->Next = op->Next;
+  op->Next->Prev = result;
+  result->Idx = 0;
+  return result;
+}
+//------------------------------------------------------------------------------
+
+void CleanPolygon(const Path &in_poly, Path &out_poly, double distance) {
+  // distance = proximity in units/pixels below which vertices
+  // will be stripped. Default ~= sqrt(2).
+
+  size_t size = in_poly.size();
+
+  if (size == 0) {
+    out_poly.clear();
+    return;
+  }
+
+  OutPt *outPts = new OutPt[size];
+  for (size_t i = 0; i < size; ++i) {
+    outPts[i].Pt = in_poly[i];
+    outPts[i].Next = &outPts[(i + 1) % size];
+    outPts[i].Next->Prev = &outPts[i];
+    outPts[i].Idx = 0;
+  }
+
+  double distSqrd = distance * distance;
+  OutPt *op = &outPts[0];
+  while (op->Idx == 0 && op->Next != op->Prev) {
+    if (PointsAreClose(op->Pt, op->Prev->Pt, distSqrd)) {
+      op = ExcludeOp(op);
+      size--;
+    } else if (PointsAreClose(op->Prev->Pt, op->Next->Pt, distSqrd)) {
+      ExcludeOp(op->Next);
+      op = ExcludeOp(op);
+      size -= 2;
+    } else if (SlopesNearCollinear(op->Prev->Pt, op->Pt, op->Next->Pt,
+                                   distSqrd)) {
+      op = ExcludeOp(op);
+      size--;
+    } else {
+      op->Idx = 1;
+      op = op->Next;
+    }
+  }
+
+  if (size < 3)
+    size = 0;
+  out_poly.resize(size);
+  for (size_t i = 0; i < size; ++i) {
+    out_poly[i] = op->Pt;
+    op = op->Next;
+  }
+  delete[] outPts;
+}
+//------------------------------------------------------------------------------
+
+void CleanPolygon(Path &poly, double distance) {
+  CleanPolygon(poly, poly, distance);
+}
+//------------------------------------------------------------------------------
+
+void CleanPolygons(const Paths &in_polys, Paths &out_polys, double distance) {
+  out_polys.resize(in_polys.size());
+  for (Paths::size_type i = 0; i < in_polys.size(); ++i)
+    CleanPolygon(in_polys[i], out_polys[i], distance);
+}
+//------------------------------------------------------------------------------
+
+void CleanPolygons(Paths &polys, double distance) {
+  CleanPolygons(polys, polys, distance);
+}
+//------------------------------------------------------------------------------
+
+void Minkowski(const Path &poly, const Path &path, Paths &solution, bool isSum,
+               bool isClosed) {
+  int delta = (isClosed ? 1 : 0);
+  size_t polyCnt = poly.size();
+  size_t pathCnt = path.size();
+  Paths pp;
+  pp.reserve(pathCnt);
+  if (isSum)
+    for (size_t i = 0; i < pathCnt; ++i) {
+      Path p;
+      p.reserve(polyCnt);
+      for (size_t j = 0; j < poly.size(); ++j)
+        p.push_back(IntPoint(path[i].X + poly[j].X, path[i].Y + poly[j].Y));
+      pp.push_back(p);
+    }
+  else
+    for (size_t i = 0; i < pathCnt; ++i) {
+      Path p;
+      p.reserve(polyCnt);
+      for (size_t j = 0; j < poly.size(); ++j)
+        p.push_back(IntPoint(path[i].X - poly[j].X, path[i].Y - poly[j].Y));
+      pp.push_back(p);
+    }
+
+  solution.clear();
+  solution.reserve((pathCnt + delta) * (polyCnt + 1));
+  for (size_t i = 0; i < pathCnt - 1 + delta; ++i)
+    for (size_t j = 0; j < polyCnt; ++j) {
+      Path quad;
+      quad.reserve(4);
+      quad.push_back(pp[i % pathCnt][j % polyCnt]);
+      quad.push_back(pp[(i + 1) % pathCnt][j % polyCnt]);
+      quad.push_back(pp[(i + 1) % pathCnt][(j + 1) % polyCnt]);
+      quad.push_back(pp[i % pathCnt][(j + 1) % polyCnt]);
+      if (!Orientation(quad))
+        ReversePath(quad);
+      solution.push_back(quad);
+    }
+}
+//------------------------------------------------------------------------------
+
+void MinkowskiSum(const Path &pattern, const Path &path, Paths &solution,
+                  bool pathIsClosed) {
+  Minkowski(pattern, path, solution, true, pathIsClosed);
+  Clipper c;
+  c.AddPaths(solution, ptSubject, true);
+  c.Execute(ctUnion, solution, pftNonZero, pftNonZero);
+}
+//------------------------------------------------------------------------------
+
+void TranslatePath(const Path &input, Path &output, const IntPoint delta) {
+  // precondition: input != output
+  output.resize(input.size());
+  for (size_t i = 0; i < input.size(); ++i)
+    output[i] = IntPoint(input[i].X + delta.X, input[i].Y + delta.Y);
+}
+//------------------------------------------------------------------------------
+
+void MinkowskiSum(const Path &pattern, const Paths &paths, Paths &solution,
+                  bool pathIsClosed) {
+  Clipper c;
+  for (size_t i = 0; i < paths.size(); ++i) {
+    Paths tmp;
+    Minkowski(pattern, paths[i], tmp, true, pathIsClosed);
+    c.AddPaths(tmp, ptSubject, true);
+    if (pathIsClosed) {
+      Path tmp2;
+      TranslatePath(paths[i], tmp2, pattern[0]);
+      c.AddPath(tmp2, ptClip, true);
+    }
+  }
+  c.Execute(ctUnion, solution, pftNonZero, pftNonZero);
+}
+//------------------------------------------------------------------------------
+
+void MinkowskiDiff(const Path &poly1, const Path &poly2, Paths &solution) {
+  Minkowski(poly1, poly2, solution, false, true);
+  Clipper c;
+  c.AddPaths(solution, ptSubject, true);
+  c.Execute(ctUnion, solution, pftNonZero, pftNonZero);
+}
+//------------------------------------------------------------------------------
+
+enum NodeType { ntAny, ntOpen, ntClosed };
+
+void AddPolyNodeToPaths(const PolyNode &polynode, NodeType nodetype,
+                        Paths &paths) {
+  bool match = true;
+  if (nodetype == ntClosed)
+    match = !polynode.IsOpen();
+  else if (nodetype == ntOpen)
+    return;
+
+  if (!polynode.Contour.empty() && match)
+    paths.push_back(polynode.Contour);
+  for (int i = 0; i < polynode.ChildCount(); ++i)
+    AddPolyNodeToPaths(*polynode.Childs[i], nodetype, paths);
+}
+//------------------------------------------------------------------------------
+
+void PolyTreeToPaths(const PolyTree &polytree, Paths &paths) {
+  paths.resize(0);
+  paths.reserve(polytree.Total());
+  AddPolyNodeToPaths(polytree, ntAny, paths);
+}
+//------------------------------------------------------------------------------
+
+void ClosedPathsFromPolyTree(const PolyTree &polytree, Paths &paths) {
+  paths.resize(0);
+  paths.reserve(polytree.Total());
+  AddPolyNodeToPaths(polytree, ntClosed, paths);
+}
+//------------------------------------------------------------------------------
+
+void OpenPathsFromPolyTree(PolyTree &polytree, Paths &paths) {
+  paths.resize(0);
+  paths.reserve(polytree.Total());
+  // Open paths are top level only, so ...
+  for (int i = 0; i < polytree.ChildCount(); ++i)
+    if (polytree.Childs[i]->IsOpen())
+      paths.push_back(polytree.Childs[i]->Contour);
+}
+//------------------------------------------------------------------------------
+
+std::ostream &operator<<(std::ostream &s, const IntPoint &p) {
+  s << "(" << p.X << "," << p.Y << ")";
+  return s;
+}
+//------------------------------------------------------------------------------
+
+std::ostream &operator<<(std::ostream &s, const Path &p) {
+  if (p.empty())
+    return s;
+  Path::size_type last = p.size() - 1;
+  for (Path::size_type i = 0; i < last; i++)
+    s << "(" << p[i].X << "," << p[i].Y << "), ";
+  s << "(" << p[last].X << "," << p[last].Y << ")\n";
+  return s;
+}
+//------------------------------------------------------------------------------
+
+std::ostream &operator<<(std::ostream &s, const Paths &p) {
+  for (Paths::size_type i = 0; i < p.size(); i++)
+    s << p[i];
+  s << "\n";
+  return s;
+}
+//------------------------------------------------------------------------------
+
+} // namespace ClipperLib
diff --git a/deploy/ios_demo/PaddleOCRDemo/ThirdParty/Clipper1/clipper.hpp b/deploy/ios_demo/PaddleOCRDemo/ThirdParty/Clipper1/clipper.hpp
new file mode 100644
index 0000000..45cd12a
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/ThirdParty/Clipper1/clipper.hpp
@@ -0,0 +1,419 @@
+/*******************************************************************************
+ *                                                                              *
+ * Author    :  Angus Johnson * Version   :  6.4.2 * Date      :  27 February
+ *2017                                                * Website   :
+ *http://www.angusj.com                                           * Copyright :
+ *Angus Johnson 2010-2017                                         *
+ *                                                                              *
+ * License: * Use, modification & distribution is subject to Boost Software
+ *License Ver 1. * http://www.boost.org/LICENSE_1_0.txt *
+ *                                                                              *
+ * Attributions: * The code in this library is an extension of Bala Vatti's
+ *clipping algorithm: * "A generic solution to polygon clipping" *
+ * Communications of the ACM, Vol 35, Issue 7 (July 1992) pp 56-63. *
+ * http://portal.acm.org/citation.cfm?id=129906 *
+ *                                                                              *
+ * Computer graphics and geometric modeling: implementation and algorithms * By
+ *Max K. Agoston                                                            *
+ * Springer; 1 edition (January 4, 2005) *
+ * http://books.google.com/books?q=vatti+clipping+agoston *
+ *                                                                              *
+ * See also: * "Polygon Offsetting by Computing Winding Numbers" * Paper no.
+ *DETC2005-85513 pp. 565-575                                         * ASME 2005
+ *International Design Engineering Technical Conferences             * and
+ *Computers and Information in Engineering Conference (IDETC/CIE2005)      *
+ * September 24-28, 2005 , Long Beach, California, USA *
+ * http://www.me.berkeley.edu/~mcmains/pubs/DAC05OffsetPolygon.pdf *
+ *                                                                              *
+ *******************************************************************************/
+
+#ifndef clipper_hpp
+#define clipper_hpp
+
+#define CLIPPER_VERSION "6.4.2"
+
+// use_int32: When enabled 32bit ints are used instead of 64bit ints. This
+// improve performance but coordinate values are limited to the range +/- 46340
+// #define use_int32
+
+// use_xyz: adds a Z member to IntPoint. Adds a minor cost to perfomance.
+// #define use_xyz
+
+// use_lines: Enables line clipping. Adds a very minor cost to performance.
+#define use_lines
+
+// use_deprecated: Enables temporary support for the obsolete functions
+// #define use_deprecated
+
+#include <cstdlib>
+#include <cstring>
+#include <functional>
+#include <list>
+#include <ostream>
+#include <queue>
+#include <set>
+#include <stdexcept>
+#include <vector>
+
+namespace ClipperLib {
+
+enum ClipType { ctIntersection, ctUnion, ctDifference, ctXor };
+enum PolyType { ptSubject, ptClip };
+// By far the most widely used winding rules for polygon filling are
+// EvenOdd & NonZero (GDI, GDI+, XLib, OpenGL, Cairo, AGG, Quartz, SVG, Gr32)
+// Others rules include Positive, Negative and ABS_GTR_EQ_TWO (only in OpenGL)
+// see http://glprogramming.com/red/chapter11.html
+enum PolyFillType { pftEvenOdd, pftNonZero, pftPositive, pftNegative };
+
+#ifdef use_int32
+typedef int cInt;
+static cInt const loRange = 0x7FFF;
+static cInt const hiRange = 0x7FFF;
+#else
+typedef signed long long cInt;
+static cInt const loRange = 0x3FFFFFFF;
+static cInt const hiRange = 0x3FFFFFFFFFFFFFFFLL;
+typedef signed long long long64; // used by Int128 class
+typedef unsigned long long ulong64;
+
+#endif
+
+struct IntPoint {
+  cInt X;
+  cInt Y;
+#ifdef use_xyz
+  cInt Z;
+  IntPoint(cInt x = 0, cInt y = 0, cInt z = 0) : X(x), Y(y), Z(z){};
+#else
+  IntPoint(cInt x = 0, cInt y = 0) : X(x), Y(y){};
+#endif
+
+  friend inline bool operator==(const IntPoint &a, const IntPoint &b) {
+    return a.X == b.X && a.Y == b.Y;
+  }
+  friend inline bool operator!=(const IntPoint &a, const IntPoint &b) {
+    return a.X != b.X || a.Y != b.Y;
+  }
+};
+//------------------------------------------------------------------------------
+
+typedef std::vector<IntPoint> Path;
+typedef std::vector<Path> Paths;
+
+inline Path &operator<<(Path &poly, const IntPoint &p) {
+  poly.push_back(p);
+  return poly;
+}
+inline Paths &operator<<(Paths &polys, const Path &p) {
+  polys.push_back(p);
+  return polys;
+}
+
+std::ostream &operator<<(std::ostream &s, const IntPoint &p);
+std::ostream &operator<<(std::ostream &s, const Path &p);
+std::ostream &operator<<(std::ostream &s, const Paths &p);
+
+struct DoublePoint {
+  double X;
+  double Y;
+  DoublePoint(double x = 0, double y = 0) : X(x), Y(y) {}
+  DoublePoint(IntPoint ip) : X((double)ip.X), Y((double)ip.Y) {}
+};
+//------------------------------------------------------------------------------
+
+#ifdef use_xyz
+typedef void (*ZFillCallback)(IntPoint &e1bot, IntPoint &e1top, IntPoint &e2bot,
+                              IntPoint &e2top, IntPoint &pt);
+#endif
+
+enum InitOptions {
+  ioReverseSolution = 1,
+  ioStrictlySimple = 2,
+  ioPreserveCollinear = 4
+};
+enum JoinType { jtSquare, jtRound, jtMiter };
+enum EndType {
+  etClosedPolygon,
+  etClosedLine,
+  etOpenButt,
+  etOpenSquare,
+  etOpenRound
+};
+
+class PolyNode;
+typedef std::vector<PolyNode *> PolyNodes;
+
+class PolyNode {
+public:
+  PolyNode();
+  virtual ~PolyNode(){};
+  Path Contour;
+  PolyNodes Childs;
+  PolyNode *Parent;
+  PolyNode *GetNext() const;
+  bool IsHole() const;
+  bool IsOpen() const;
+  int ChildCount() const;
+
+private:
+  // PolyNode& operator =(PolyNode& other);
+  unsigned Index; // node index in Parent.Childs
+  bool m_IsOpen;
+  JoinType m_jointype;
+  EndType m_endtype;
+  PolyNode *GetNextSiblingUp() const;
+  void AddChild(PolyNode &child);
+  friend class Clipper; // to access Index
+  friend class ClipperOffset;
+};
+
+class PolyTree : public PolyNode {
+public:
+  ~PolyTree() { Clear(); };
+  PolyNode *GetFirst() const;
+  void Clear();
+  int Total() const;
+
+private:
+  // PolyTree& operator =(PolyTree& other);
+  PolyNodes AllNodes;
+  friend class Clipper; // to access AllNodes
+};
+
+bool Orientation(const Path &poly);
+double Area(const Path &poly);
+int PointInPolygon(const IntPoint &pt, const Path &path);
+
+void SimplifyPolygon(const Path &in_poly, Paths &out_polys,
+                     PolyFillType fillType = pftEvenOdd);
+void SimplifyPolygons(const Paths &in_polys, Paths &out_polys,
+                      PolyFillType fillType = pftEvenOdd);
+void SimplifyPolygons(Paths &polys, PolyFillType fillType = pftEvenOdd);
+
+void CleanPolygon(const Path &in_poly, Path &out_poly, double distance = 1.415);
+void CleanPolygon(Path &poly, double distance = 1.415);
+void CleanPolygons(const Paths &in_polys, Paths &out_polys,
+                   double distance = 1.415);
+void CleanPolygons(Paths &polys, double distance = 1.415);
+
+void MinkowskiSum(const Path &pattern, const Path &path, Paths &solution,
+                  bool pathIsClosed);
+void MinkowskiSum(const Path &pattern, const Paths &paths, Paths &solution,
+                  bool pathIsClosed);
+void MinkowskiDiff(const Path &poly1, const Path &poly2, Paths &solution);
+
+void PolyTreeToPaths(const PolyTree &polytree, Paths &paths);
+void ClosedPathsFromPolyTree(const PolyTree &polytree, Paths &paths);
+void OpenPathsFromPolyTree(PolyTree &polytree, Paths &paths);
+
+void ReversePath(Path &p);
+void ReversePaths(Paths &p);
+
+struct IntRect {
+  cInt left;
+  cInt top;
+  cInt right;
+  cInt bottom;
+};
+
+// enums that are used internally ...
+enum EdgeSide { esLeft = 1, esRight = 2 };
+
+// forward declarations (for stuff used internally) ...
+struct TEdge;
+struct IntersectNode;
+struct LocalMinimum;
+struct OutPt;
+struct OutRec;
+struct Join;
+
+typedef std::vector<OutRec *> PolyOutList;
+typedef std::vector<TEdge *> EdgeList;
+typedef std::vector<Join *> JoinList;
+typedef std::vector<IntersectNode *> IntersectList;
+
+//------------------------------------------------------------------------------
+
+// ClipperBase is the ancestor to the Clipper class. It should not be
+// instantiated directly. This class simply abstracts the conversion of sets of
+// polygon coordinates into edge objects that are stored in a LocalMinima list.
+class ClipperBase {
+public:
+  ClipperBase();
+  virtual ~ClipperBase();
+  virtual bool AddPath(const Path &pg, PolyType PolyTyp, bool Closed);
+  bool AddPaths(const Paths &ppg, PolyType PolyTyp, bool Closed);
+  virtual void Clear();
+  IntRect GetBounds();
+  bool PreserveCollinear() { return m_PreserveCollinear; };
+  void PreserveCollinear(bool value) { m_PreserveCollinear = value; };
+
+protected:
+  void DisposeLocalMinimaList();
+  TEdge *AddBoundsToLML(TEdge *e, bool IsClosed);
+  virtual void Reset();
+  TEdge *ProcessBound(TEdge *E, bool IsClockwise);
+  void InsertScanbeam(const cInt Y);
+  bool PopScanbeam(cInt &Y);
+  bool LocalMinimaPending();
+  bool PopLocalMinima(cInt Y, const LocalMinimum *&locMin);
+  OutRec *CreateOutRec();
+  void DisposeAllOutRecs();
+  void DisposeOutRec(PolyOutList::size_type index);
+  void SwapPositionsInAEL(TEdge *edge1, TEdge *edge2);
+  void DeleteFromAEL(TEdge *e);
+  void UpdateEdgeIntoAEL(TEdge *&e);
+
+  typedef std::vector<LocalMinimum> MinimaList;
+  MinimaList::iterator m_CurrentLM;
+  MinimaList m_MinimaList;
+
+  bool m_UseFullRange;
+  EdgeList m_edges;
+  bool m_PreserveCollinear;
+  bool m_HasOpenPaths;
+  PolyOutList m_PolyOuts;
+  TEdge *m_ActiveEdges;
+
+  typedef std::priority_queue<cInt> ScanbeamList;
+  ScanbeamList m_Scanbeam;
+};
+//------------------------------------------------------------------------------
+
+class Clipper : public virtual ClipperBase {
+public:
+  Clipper(int initOptions = 0);
+  bool Execute(ClipType clipType, Paths &solution,
+               PolyFillType fillType = pftEvenOdd);
+  bool Execute(ClipType clipType, Paths &solution, PolyFillType subjFillType,
+               PolyFillType clipFillType);
+  bool Execute(ClipType clipType, PolyTree &polytree,
+               PolyFillType fillType = pftEvenOdd);
+  bool Execute(ClipType clipType, PolyTree &polytree, PolyFillType subjFillType,
+               PolyFillType clipFillType);
+  bool ReverseSolution() { return m_ReverseOutput; };
+  void ReverseSolution(bool value) { m_ReverseOutput = value; };
+  bool StrictlySimple() { return m_StrictSimple; };
+  void StrictlySimple(bool value) { m_StrictSimple = value; };
+    // set the callback function for z value filling on intersections (otherwise
+    // Z is 0)
+#ifdef use_xyz
+  void ZFillFunction(ZFillCallback zFillFunc);
+#endif
+protected:
+  virtual bool ExecuteInternal();
+
+private:
+  JoinList m_Joins;
+  JoinList m_GhostJoins;
+  IntersectList m_IntersectList;
+  ClipType m_ClipType;
+  typedef std::list<cInt> MaximaList;
+  MaximaList m_Maxima;
+  TEdge *m_SortedEdges;
+  bool m_ExecuteLocked;
+  PolyFillType m_ClipFillType;
+  PolyFillType m_SubjFillType;
+  bool m_ReverseOutput;
+  bool m_UsingPolyTree;
+  bool m_StrictSimple;
+#ifdef use_xyz
+  ZFillCallback m_ZFill; // custom callback
+#endif
+  void SetWindingCount(TEdge &edge);
+  bool IsEvenOddFillType(const TEdge &edge) const;
+  bool IsEvenOddAltFillType(const TEdge &edge) const;
+  void InsertLocalMinimaIntoAEL(const cInt botY);
+  void InsertEdgeIntoAEL(TEdge *edge, TEdge *startEdge);
+  void AddEdgeToSEL(TEdge *edge);
+  bool PopEdgeFromSEL(TEdge *&edge);
+  void CopyAELToSEL();
+  void DeleteFromSEL(TEdge *e);
+  void SwapPositionsInSEL(TEdge *edge1, TEdge *edge2);
+  bool IsContributing(const TEdge &edge) const;
+  bool IsTopHorz(const cInt XPos);
+  void DoMaxima(TEdge *e);
+  void ProcessHorizontals();
+  void ProcessHorizontal(TEdge *horzEdge);
+  void AddLocalMaxPoly(TEdge *e1, TEdge *e2, const IntPoint &pt);
+  OutPt *AddLocalMinPoly(TEdge *e1, TEdge *e2, const IntPoint &pt);
+  OutRec *GetOutRec(int idx);
+  void AppendPolygon(TEdge *e1, TEdge *e2);
+  void IntersectEdges(TEdge *e1, TEdge *e2, IntPoint &pt);
+  OutPt *AddOutPt(TEdge *e, const IntPoint &pt);
+  OutPt *GetLastOutPt(TEdge *e);
+  bool ProcessIntersections(const cInt topY);
+  void BuildIntersectList(const cInt topY);
+  void ProcessIntersectList();
+  void ProcessEdgesAtTopOfScanbeam(const cInt topY);
+  void BuildResult(Paths &polys);
+  void BuildResult2(PolyTree &polytree);
+  void SetHoleState(TEdge *e, OutRec *outrec);
+  void DisposeIntersectNodes();
+  bool FixupIntersectionOrder();
+  void FixupOutPolygon(OutRec &outrec);
+  void FixupOutPolyline(OutRec &outrec);
+  bool IsHole(TEdge *e);
+  bool FindOwnerFromSplitRecs(OutRec &outRec, OutRec *&currOrfl);
+  void FixHoleLinkage(OutRec &outrec);
+  void AddJoin(OutPt *op1, OutPt *op2, const IntPoint offPt);
+  void ClearJoins();
+  void ClearGhostJoins();
+  void AddGhostJoin(OutPt *op, const IntPoint offPt);
+  bool JoinPoints(Join *j, OutRec *outRec1, OutRec *outRec2);
+  void JoinCommonEdges();
+  void DoSimplePolygons();
+  void FixupFirstLefts1(OutRec *OldOutRec, OutRec *NewOutRec);
+  void FixupFirstLefts2(OutRec *InnerOutRec, OutRec *OuterOutRec);
+  void FixupFirstLefts3(OutRec *OldOutRec, OutRec *NewOutRec);
+#ifdef use_xyz
+  void SetZ(IntPoint &pt, TEdge &e1, TEdge &e2);
+#endif
+};
+//------------------------------------------------------------------------------
+
+class ClipperOffset {
+public:
+  ClipperOffset(double miterLimit = 2.0, double roundPrecision = 0.25);
+  ~ClipperOffset();
+  void AddPath(const Path &path, JoinType joinType, EndType endType);
+  void AddPaths(const Paths &paths, JoinType joinType, EndType endType);
+  void Execute(Paths &solution, double delta);
+  void Execute(PolyTree &solution, double delta);
+  void Clear();
+  double MiterLimit;
+  double ArcTolerance;
+
+private:
+  Paths m_destPolys;
+  Path m_srcPoly;
+  Path m_destPoly;
+  std::vector<DoublePoint> m_normals;
+  double m_delta, m_sinA, m_sin, m_cos;
+  double m_miterLim, m_StepsPerRad;
+  IntPoint m_lowest;
+  PolyNode m_polyNodes;
+
+  void FixOrientations();
+  void DoOffset(double delta);
+  void OffsetPoint(int j, int &k, JoinType jointype);
+  void DoSquare(int j, int k);
+  void DoMiter(int j, int k, double r);
+  void DoRound(int j, int k);
+};
+//------------------------------------------------------------------------------
+
+class clipperException : public std::exception {
+public:
+  clipperException(const char *description) : m_descr(description) {}
+  virtual ~clipperException() throw() {}
+  virtual const char *what() const throw() { return m_descr.c_str(); }
+
+private:
+  std::string m_descr;
+};
+//------------------------------------------------------------------------------
+
+} // namespace ClipperLib
+
+#endif // clipper_hpp
diff --git a/deploy/ios_demo/PaddleOCRDemo/ViewModels/OCRViewModel.swift b/deploy/ios_demo/PaddleOCRDemo/ViewModels/OCRViewModel.swift
new file mode 100644
index 0000000..e318532
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/ViewModels/OCRViewModel.swift
@@ -0,0 +1,275 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import Foundation
+import UIKit
+
+// MARK: - App State
+
+/// The complete set of states the app can be in.
+/// Drives all UI rendering via a single `@Published` property on `OCRViewModel`.
+enum AppState: Equatable {
+    /// Models are being loaded from the bundle.
+    case loadingModels
+    /// Models are loaded and ready; waiting for the user to select an image.
+    case ready
+    /// An image is being processed by OCR.
+    case processing(UIImage)
+    /// OCR completed successfully with results and the processed image.
+    /// `runId` distinguishes successive successful runs for ``Equatable`` without requiring ``OCRRunResult`` to conform.
+    case results(OCRRunResult, UIImage, runId: UUID)
+    /// An error occurred during model loading or inference.
+    case error(AppError)
+
+    static func == (lhs: AppState, rhs: AppState) -> Bool {
+        switch (lhs, rhs) {
+        case (.loadingModels, .loadingModels): return true
+        case (.ready, .ready): return true
+        case (.processing, .processing): return true
+        case (.results(_, _, let runIdL), .results(_, _, let runIdR)): return runIdL == runIdR
+        case (.error(let a), .error(let b)): return a == b
+        default: return false
+        }
+    }
+}
+
+// MARK: - App Error
+
+/// User-facing errors with recovery guidance.
+enum AppError: Equatable, LocalizedError {
+    case modelLoadFailed(String)
+    case inferenceFailed(String)
+    case imageLoadFailed
+
+    var errorDescription: String? {
+        switch self {
+        case .modelLoadFailed(let msg): return msg
+        case .inferenceFailed(let msg): return msg
+        case .imageLoadFailed: return "Could not load the selected image. Try a different one."
+        }
+    }
+
+    /// Whether this error is recoverable by retrying model load (vs just selecting a new image).
+    var isModelError: Bool {
+        if case .modelLoadFailed = self { return true }
+        return false
+    }
+}
+
+// MARK: - OCRViewModel
+
+/// Central view model that manages the complete app lifecycle.
+///
+/// State transitions:
+/// ```
+/// loadingModels --> ready --> processing --> results
+///       |            ^           |            |
+///       v            |           v            v
+///     error ----retry---->    error       ready (reset)
+/// ```
+///
+/// All state mutations happen on `@MainActor` through the single
+/// `@Published var state` property, giving SwiftUI a single source of truth.
+@MainActor
+class OCRViewModel: ObservableObject {
+    @Published var state: AppState = .loadingModels
+    @Published var copiedFeedback: Bool = false
+    @Published var runtimeParams = OCRRuntimeParams.noOverrides
+    /// ONNX Runtime presets and tuning shown in UI (draft). Use **Apply & reload** in Advanced settings to reload sessions (`loadModels()`).
+    @Published var ortPrimaryExecutionProvider: ORTPrimaryExecutionProvider = .cpu
+    @Published var ortSessionTuning = ORTSessionTuningOptions.default
+    /// Snapshot of Ort settings actually used by the loaded ``ocrEngine``.
+    private var ortLoadedPrimaryExecutionProvider: ORTPrimaryExecutionProvider = .cpu
+    private var ortLoadedSessionTuning: ORTSessionTuningOptions = .default
+    @Published private(set) var resolvedRuntimeBaseline: ResolvedOCRRuntimeParams?
+
+    private var sessionManager: ORTSessionManager?
+    private var ocrEngine: OCREngine?
+    /// Last image run through OCR (normalized orientation), for **Re-run OCR** without re-picking.
+    private var lastNormalizedImage: UIImage?
+    /// OCR runtime parameters used for the last successful run (drives ``hasPendingOcrChanges``).
+    private var lastRunRuntimeParams: OCRRuntimeParams?
+
+    /// Bundled sample resource names (stem or `stem.ext`) for ``selectSampleImage(named:)``.
+    let sampleImageNames: [String] = [DemoAssets.defaultSampleImageStem]
+
+    /// Draft Ort settings differ from the configuration used by loaded sessions.
+    var hasUnappliedOrtSettings: Bool {
+        ortPrimaryExecutionProvider != ortLoadedPrimaryExecutionProvider
+            || ortSessionTuning != ortLoadedSessionTuning
+    }
+
+    /// `runtimeParams` differs from the last successful OCR run (only meaningful in ``AppState/results``).
+    var hasPendingOcrChanges: Bool {
+        guard case .results = state else { return false }
+        guard let last = lastRunRuntimeParams else { return false }
+        return runtimeParams != last
+    }
+
+    /// When `false`, disable Apply/Revert (reload would conflict with loading or OCR in flight).
+    var allowsOrtApplyControls: Bool {
+        switch state {
+        case .loadingModels, .processing: return false
+        case .ready, .results, .error: return true
+        }
+    }
+
+    /// Restores ONNX Runtime controls to match the currently loaded sessions.
+    func revertOrtDraftToLoaded() {
+        ortPrimaryExecutionProvider = ortLoadedPrimaryExecutionProvider
+        ortSessionTuning = ortLoadedSessionTuning
+    }
+
+    /// Photo picker or file load produced unusable image data.
+    func reportImageLoadFailed() {
+        state = .error(.imageLoadFailed)
+    }
+
+    // MARK: - Lifecycle
+
+    /// Load ORT models from the app bundle.
+    ///
+    /// Creates an `ORTSessionManager`, loads both detection and recognition models,
+    /// then creates the `OCREngine` for end-to-end OCR.
+    func loadModels() async {
+        state = .loadingModels
+        do {
+            let manager = ORTSessionManager()
+            try await manager.loadModels(executionProvider: ortPrimaryExecutionProvider, tuning: ortSessionTuning)
+            let engine = try OCREngine(sessionManager: manager)
+            self.sessionManager = manager
+            self.ocrEngine = engine
+            self.resolvedRuntimeBaseline = engine.baselineRuntimeDefaults()
+            ortLoadedPrimaryExecutionProvider = ortPrimaryExecutionProvider
+            ortLoadedSessionTuning = ortSessionTuning
+            state = .ready
+        } catch {
+            state = .error(.modelLoadFailed(error.localizedDescription))
+        }
+    }
+
+    // MARK: - Image Processing
+
+    /// Run full OCR on a user-selected image.
+    ///
+    /// Normalizes orientation first (camera photos often have EXIF rotation),
+    /// then delegates to `OCREngine.run(_:)`.
+    func processImage(_ uiImage: UIImage) async {
+        let normalized = normalizeOrientation(uiImage)
+        lastNormalizedImage = normalized
+        state = .processing(normalized)
+        guard let cgImage = normalized.cgImage else {
+            state = .error(.imageLoadFailed)
+            return
+        }
+        guard let engine = ocrEngine else {
+            state = .error(.modelLoadFailed("OCR engine is not ready."))
+            return
+        }
+        do {
+            let result = try await engine.run(cgImage, params: runtimeParams)
+            lastRunRuntimeParams = runtimeParams
+            state = .results(result, normalized, runId: UUID())
+        } catch {
+            lastNormalizedImage = nil
+            state = .error(.inferenceFailed(error.localizedDescription))
+        }
+    }
+
+    /// Runs OCR again on the **last image** using the current `runtimeParams` (e.g. after tuning sliders).
+    func rerunOCR() async {
+        guard let image = lastNormalizedImage else { return }
+        await processImage(image)
+    }
+
+    /// Load a bundled sample image by name and run OCR on it.
+    func selectSampleImage(named name: String) async {
+        guard let path = Bundle.main.path(forBundledImageNamed: name, subdirectory: "SampleImages") else {
+            state = .error(.imageLoadFailed)
+            return
+        }
+        do {
+            guard let uiFromFile = UIImage(contentsOfFile: path) else {
+                state = .error(.imageLoadFailed)
+                return
+            }
+            let normalized = normalizeOrientation(uiFromFile)
+            lastNormalizedImage = normalized
+            state = .processing(normalized)
+            guard let cgImage = normalized.cgImage else {
+                state = .error(.imageLoadFailed)
+                return
+            }
+            guard let engine = ocrEngine else {
+                state = .error(.modelLoadFailed("OCR engine is not ready."))
+                return
+            }
+            let result = try await engine.run(cgImage, params: runtimeParams)
+            lastRunRuntimeParams = runtimeParams
+            state = .results(result, normalized, runId: UUID())
+        } catch {
+            lastNormalizedImage = nil
+            state = .error(.inferenceFailed(error.localizedDescription))
+        }
+    }
+
+    // MARK: - Result Actions
+
+    /// Copy all recognized text to the system clipboard, with brief visual feedback.
+    func copyResultsToClipboard() {
+        guard case .results(let result, _, _) = state else { return }
+        let allText = result.results.map(\.text).joined(separator: "\n")
+        UIPasteboard.general.string = allText
+        copiedFeedback = true
+        Task {
+            try? await Task.sleep(nanoseconds: 2_000_000_000)
+            copiedFeedback = false
+        }
+    }
+
+    // MARK: - Error Recovery
+
+    /// Retry after an error. Model errors reload models; other errors return to ready state.
+    func retry() async {
+        guard case .error(let error) = state else { return }
+        if error.isModelError {
+            await loadModels()
+        } else {
+            state = .ready
+        }
+    }
+
+    /// Return to the ready state (e.g. after viewing results, user wants to pick a new image).
+    func reset() {
+        lastNormalizedImage = nil
+        lastRunRuntimeParams = nil
+        state = .ready
+    }
+}
+
+// MARK: - Image Orientation
+
+/// Normalize EXIF orientation by redrawing the image in `.up` orientation.
+///
+/// Camera photos from the photo picker often carry non-`.up` orientation metadata.
+/// CoreGraphics ignores EXIF orientation, so we must flatten it before OCR processing.
+private func normalizeOrientation(_ image: UIImage) -> UIImage {
+    guard image.imageOrientation != .up else { return image }
+    let format = UIGraphicsImageRendererFormat()
+    format.scale = image.scale
+    let renderer = UIGraphicsImageRenderer(size: image.size, format: format)
+    return renderer.image { _ in
+        image.draw(in: CGRect(origin: .zero, size: image.size))
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Views/EngineSettingsSheet.swift b/deploy/ios_demo/PaddleOCRDemo/Views/EngineSettingsSheet.swift
new file mode 100644
index 0000000..f6a83aa
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Views/EngineSettingsSheet.swift
@@ -0,0 +1,106 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import SwiftUI
+
+/// Advanced ONNX Runtime and OCR parameters, presented as a sheet from the main screen toolbar.
+struct EngineSettingsSheet: View {
+    @Environment(\.dismiss) private var dismiss
+    @ObservedObject var viewModel: OCRViewModel
+    @State private var ortSessionReloadConfirmed = false
+
+    var body: some View {
+        NavigationStack {
+            ScrollView {
+                VStack(alignment: .leading, spacing: 20) {
+                    DemoSectionHeader(
+                        title: "ONNX Runtime",
+                        subtitle: "Choose execution provider and options, then reload to apply."
+                    )
+                    DemoCard {
+                        OnnxRuntimeSettingsPanel(
+                            primaryExecutionProvider: $viewModel.ortPrimaryExecutionProvider,
+                            tuning: $viewModel.ortSessionTuning
+                        )
+                    }
+
+                    DemoSectionHeader(
+                        title: "OCR parameters",
+                        subtitle: "These apply to the next OCR run."
+                    )
+                    DemoCard {
+                        OCRParametersPanel(
+                            params: $viewModel.runtimeParams,
+                            resolvedBaseline: viewModel.resolvedRuntimeBaseline ?? .fallbackForUI
+                        )
+                    }
+                }
+                .padding(20)
+                .padding(.bottom, viewModel.hasUnappliedOrtSettings ? 120 : 24)
+            }
+            .background(Color(.systemGroupedBackground))
+            .navigationTitle("Engine & parameters")
+            .navigationBarTitleDisplayMode(.inline)
+            .toolbar {
+                ToolbarItem(placement: .confirmationAction) {
+                    Button("Done") {
+                        dismiss()
+                    }
+                }
+            }
+            .safeAreaInset(edge: .bottom, spacing: 0) {
+                if viewModel.hasUnappliedOrtSettings {
+                    VStack(alignment: .leading, spacing: 12) {
+                        Divider()
+                        Text("Sessions still use the previous ONNX Runtime choices until you reload.")
+                            .font(.footnote)
+                            .foregroundStyle(Color.orange.opacity(0.95))
+                            .fixedSize(horizontal: false, vertical: true)
+
+                        HStack(spacing: 12) {
+                            Button("Revert draft") {
+                                viewModel.revertOrtDraftToLoaded()
+                            }
+                            .disabled(!viewModel.allowsOrtApplyControls || !viewModel.hasUnappliedOrtSettings)
+
+                            Button {
+                                ortSessionReloadConfirmed = true
+                            } label: {
+                                Label("Apply & reload", systemImage: "arrow.clockwise.circle")
+                                    .frame(maxWidth: .infinity)
+                            }
+                            .buttonStyle(.borderedProminent)
+                            .disabled(!viewModel.allowsOrtApplyControls || !viewModel.hasUnappliedOrtSettings)
+                        }
+                    }
+                    .padding(.horizontal, 20)
+                    .padding(.vertical, 12)
+                    .background(.ultraThinMaterial)
+                }
+            }
+        }
+        .confirmationDialog(
+            "Rebuild ONNX Runtime sessions?",
+            isPresented: $ortSessionReloadConfirmed,
+            titleVisibility: .visible
+        ) {
+            Button("Reload models") {
+                Task { await viewModel.loadModels() }
+            }
+            Button("Cancel", role: .cancel) {}
+        } message: {
+            Text("Sessions are recreated from the bundle. Expect a short pause.")
+        }
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Views/ErrorView.swift b/deploy/ios_demo/PaddleOCRDemo/Views/ErrorView.swift
new file mode 100644
index 0000000..4da7783
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Views/ErrorView.swift
@@ -0,0 +1,60 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import SwiftUI
+
+/// Displays an error state with icon, message, and retry button.
+struct ErrorView: View {
+    let error: AppError
+    let onRetry: () -> Void
+
+    var body: some View {
+        VStack(spacing: 8) {
+            Image(systemName: "exclamationmark.triangle.fill")
+                .font(.system(size: 48))
+                .symbolRenderingMode(.hierarchical)
+                .foregroundStyle(errorSeverityColor)
+
+            Text(errorTitle)
+                .font(.title2)
+                .fontWeight(.semibold)
+
+            Text(error.localizedDescription)
+                .font(.body)
+                .foregroundStyle(.secondary)
+                .multilineTextAlignment(.center)
+
+            Button(action: onRetry) {
+                Label("Retry", systemImage: "arrow.clockwise")
+            }
+            .buttonStyle(.borderedProminent)
+            .padding(.top, 8)
+        }
+    }
+
+    private var errorTitle: String {
+        switch error {
+        case .modelLoadFailed: return "Model Loading Failed"
+        case .inferenceFailed: return "OCR Failed"
+        case .imageLoadFailed: return "Image Error"
+        }
+    }
+
+    private var errorSeverityColor: Color {
+        switch error {
+        case .imageLoadFailed: return Color(.systemOrange)
+        case .modelLoadFailed, .inferenceFailed: return Color(.systemRed)
+        }
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Views/ImagePickerSection.swift b/deploy/ios_demo/PaddleOCRDemo/Views/ImagePickerSection.swift
new file mode 100644
index 0000000..ea044ee
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Views/ImagePickerSection.swift
@@ -0,0 +1,83 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import PhotosUI
+import SwiftUI
+
+/// Image selection UI: a prominent "Choose Photo" button backed by PhotosPicker,
+/// plus a row of sample image thumbnails for quick demo access.
+struct ImagePickerSection: View {
+    @Binding var selectedItem: PhotosPickerItem?
+    let sampleImageNames: [String]
+    let onSampleSelected: (String) -> Void
+
+    var body: some View {
+        VStack(alignment: .leading, spacing: 12) {
+            // Primary CTA: system photo picker
+            PhotosPicker(
+                selection: $selectedItem,
+                matching: .images,
+                photoLibrary: .shared()
+            ) {
+                Label("Choose Photo", systemImage: "photo.on.rectangle")
+                    .frame(maxWidth: .infinity)
+            }
+            .buttonStyle(.borderedProminent)
+
+            // Sample images row
+            if !sampleImageNames.isEmpty {
+                VStack(alignment: .leading, spacing: 8) {
+                    Text("Or try a sample:")
+                        .font(.subheadline)
+                        .foregroundStyle(.secondary)
+
+                    HStack(spacing: 8) {
+                        ForEach(sampleImageNames, id: \.self) { name in
+                            Button {
+                                onSampleSelected(name)
+                            } label: {
+                                sampleThumbnail(named: name)
+                            }
+                            .buttonStyle(.plain)
+                            .frame(minWidth: 44, minHeight: 44)
+                            .contentShape(Rectangle())
+                            .accessibilityLabel("Sample image: \(name)")
+                        }
+                    }
+                }
+            }
+        }
+    }
+
+    @ViewBuilder
+    private func sampleThumbnail(named name: String) -> some View {
+        if let path = Bundle.main.path(forBundledImageNamed: name, subdirectory: "SampleImages"),
+           let uiImage = UIImage(contentsOfFile: path) {
+            Image(uiImage: uiImage)
+                .resizable()
+                .aspectRatio(contentMode: .fill)
+                .frame(width: 60, height: 60)
+                .clipShape(RoundedRectangle(cornerRadius: 8))
+        } else {
+            // Fallback for when image can't be loaded
+            RoundedRectangle(cornerRadius: 8)
+                .fill(Color(.tertiarySystemFill))
+                .frame(width: 60, height: 60)
+                .overlay(
+                    Image(systemName: "photo")
+                        .foregroundStyle(.secondary)
+                )
+        }
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Views/OCRParametersPanel.swift b/deploy/ios_demo/PaddleOCRDemo/Views/OCRParametersPanel.swift
new file mode 100644
index 0000000..caca216
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Views/OCRParametersPanel.swift
@@ -0,0 +1,226 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import SwiftUI
+
+/// Sliders for detection / recognition runtime parameters (content only; wrap with `DemoCard` in parent).
+struct OCRParametersPanel: View {
+    @Binding var params: OCRRuntimeParams
+    let resolvedBaseline: ResolvedOCRRuntimeParams
+
+    var body: some View {
+        VStack(alignment: .leading, spacing: 0) {
+            groupTitle("Detection preprocess")
+            intParamRow(
+                title: "Limit side length",
+                caption: "Side length limit for detection resize",
+                range: 32...2048,
+                value: intOverrideBinding(\.textDetLimitSideLen, fallback: resolvedBaseline.textDetLimitSideLen)
+            )
+            limitTypeRow(
+                selection: stringOverrideBinding(\.textDetLimitType, fallback: resolvedBaseline.textDetLimitType)
+            )
+            intParamRow(
+                title: "Max side length cap",
+                caption: "After limit-side resize, caps the longer side if it is still above this",
+                range: 500...8000,
+                value: intOverrideBinding(\.textDetMaxSideLimit, fallback: resolvedBaseline.textDetMaxSideLimit)
+            )
+
+            groupTitle("Detection postprocess")
+            paramRow(
+                title: "Map threshold",
+                caption: "Binarization on the detector heatmap",
+                range: 0.05...0.95,
+                value: floatOverrideBinding(\.textDetThresh, fallback: resolvedBaseline.textDetThresh)
+            )
+            paramRow(
+                title: "Box score threshold",
+                caption: "Minimum score to keep a box",
+                range: 0.1...0.95,
+                value: floatOverrideBinding(\.textDetBoxThresh, fallback: resolvedBaseline.textDetBoxThresh)
+            )
+            paramRow(
+                title: "Unclip ratio",
+                caption: "Expand recovered polygons",
+                range: 0.5...3.0,
+                value: floatOverrideBinding(\.textDetUnclipRatio, fallback: resolvedBaseline.textDetUnclipRatio)
+            )
+
+            Divider()
+                .padding(.vertical, 12)
+
+            groupTitle("Recognition")
+            intParamRow(
+                title: "Recognition batch size",
+                caption: "Number of text lines per recognition run",
+                range: 1...32,
+                value: intOverrideBinding(\.textRecBatchSize, fallback: resolvedBaseline.textRecBatchSize)
+            )
+            paramRow(
+                title: "Line confidence",
+                caption: "Discard lines below this score",
+                range: 0...1,
+                value: floatOverrideBinding(\.textRecScoreThresh, fallback: resolvedBaseline.textRecScoreThresh)
+            )
+
+            HStack {
+                Spacer()
+                Button("Reset to defaults") {
+                    params = .noOverrides
+                }
+                .font(.subheadline.weight(.medium))
+            }
+            .padding(.top, 4)
+        }
+    }
+
+    private func floatOverrideBinding(
+        _ keyPath: WritableKeyPath<OCRRuntimeParams, Float?>,
+        fallback: Float
+    ) -> Binding<Float> {
+        Binding(
+            get: { params[keyPath: keyPath] ?? fallback },
+            set: { newValue in
+                var next = params
+                next[keyPath: keyPath] = newValue
+                params = next
+            }
+        )
+    }
+
+    private func intOverrideBinding(
+        _ keyPath: WritableKeyPath<OCRRuntimeParams, Int?>,
+        fallback: Int
+    ) -> Binding<Int> {
+        Binding(
+            get: { params[keyPath: keyPath] ?? fallback },
+            set: { newValue in
+                var next = params
+                next[keyPath: keyPath] = newValue
+                params = next
+            }
+        )
+    }
+
+    private func stringOverrideBinding(
+        _ keyPath: WritableKeyPath<OCRRuntimeParams, String?>,
+        fallback: String
+    ) -> Binding<String> {
+        Binding(
+            get: { params[keyPath: keyPath] ?? fallback },
+            set: { newValue in
+                var next = params
+                next[keyPath: keyPath] = newValue
+                params = next
+            }
+        )
+    }
+
+    private func groupTitle(_ text: String) -> some View {
+        Text(text)
+            .font(.subheadline.weight(.semibold))
+            .foregroundStyle(.primary)
+            .padding(.bottom, 8)
+    }
+
+    private func paramRow(
+        title: String,
+        caption: String,
+        range: ClosedRange<Float>,
+        value: Binding<Float>
+    ) -> some View {
+        VStack(alignment: .leading, spacing: 6) {
+            HStack(alignment: .firstTextBaseline) {
+                VStack(alignment: .leading, spacing: 2) {
+                    Text(title)
+                        .font(.subheadline)
+                    Text(caption)
+                        .font(.caption)
+                        .foregroundStyle(.secondary)
+                }
+                Spacer()
+                Text(String(format: "%.3f", value.wrappedValue))
+                    .font(.caption.monospacedDigit())
+                    .foregroundStyle(.secondary)
+                    .padding(.horizontal, 8)
+                    .padding(.vertical, 4)
+                    .background(Color(.tertiarySystemFill), in: Capsule())
+            }
+            Slider(
+                value: Binding(
+                    get: { Double(value.wrappedValue) },
+                    set: { value.wrappedValue = Float($0) }
+                ),
+                in: Double(range.lowerBound)...Double(range.upperBound)
+            )
+            .tint(Color.accentColor)
+        }
+        .padding(.bottom, 14)
+    }
+
+    private func intParamRow(
+        title: String,
+        caption: String,
+        range: ClosedRange<Int>,
+        value: Binding<Int>
+    ) -> some View {
+        VStack(alignment: .leading, spacing: 6) {
+            HStack(alignment: .firstTextBaseline) {
+                VStack(alignment: .leading, spacing: 2) {
+                    Text(title)
+                        .font(.subheadline)
+                    Text(caption)
+                        .font(.caption)
+                        .foregroundStyle(.secondary)
+                }
+                Spacer()
+                Text("\(value.wrappedValue)")
+                    .font(.caption.monospacedDigit())
+                    .foregroundStyle(.secondary)
+                    .padding(.horizontal, 8)
+                    .padding(.vertical, 4)
+                    .background(Color(.tertiarySystemFill), in: Capsule())
+            }
+            Slider(
+                value: Binding(
+                    get: { Double(value.wrappedValue) },
+                    set: { value.wrappedValue = Int($0.rounded()) }
+                ),
+                in: Double(range.lowerBound)...Double(range.upperBound),
+                step: 1
+            )
+            .tint(Color.accentColor)
+        }
+        .padding(.bottom, 14)
+    }
+
+    private func limitTypeRow(selection: Binding<String>) -> some View {
+        VStack(alignment: .leading, spacing: 6) {
+            VStack(alignment: .leading, spacing: 2) {
+                Text("Limit type")
+                    .font(.subheadline)
+                Text("min: shortest side ≥ limit length; max: longest side ≤ limit length")
+                    .font(.caption)
+                    .foregroundStyle(.secondary)
+            }
+            Picker("Limit type", selection: selection) {
+                Text("min").tag("min")
+                Text("max").tag("max")
+            }
+            .pickerStyle(.segmented)
+        }
+        .padding(.bottom, 14)
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Views/OnnxRuntimeSettingsPanel.swift b/deploy/ios_demo/PaddleOCRDemo/Views/OnnxRuntimeSettingsPanel.swift
new file mode 100644
index 0000000..303b583
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Views/OnnxRuntimeSettingsPanel.swift
@@ -0,0 +1,225 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import SwiftUI
+
+private enum CoreMLComputeMode: String, CaseIterable, Identifiable {
+    case automatic
+    case cpuOnly
+    case cpuAndGPU
+    case devicesWithANE
+
+    var id: String { rawValue }
+
+    var segmentLabel: String {
+        switch self {
+        case .automatic: return "Auto"
+        case .cpuOnly: return "CPU"
+        case .cpuAndGPU: return "CPU+GPU"
+        case .devicesWithANE: return "ANE"
+        }
+    }
+
+    static func from(flags: Set<String>) -> CoreMLComputeMode {
+        let cpuOnly = ORTCoreMLProviderOption.cpuOnly.rawValue
+        let cpuAndGPU = ORTCoreMLProviderOption.cpuAndGPU.rawValue
+        let aneOnly = ORTCoreMLProviderOption.aneOnly.rawValue
+        if flags.contains(cpuOnly) { return .cpuOnly }
+        if flags.contains(cpuAndGPU) { return .cpuAndGPU }
+        if flags.contains(aneOnly) { return .devicesWithANE }
+        return .automatic
+    }
+
+    func merge(into flags: inout Set<String>) {
+        let cpuOnly = ORTCoreMLProviderOption.cpuOnly.rawValue
+        let cpuAndGPU = ORTCoreMLProviderOption.cpuAndGPU.rawValue
+        let aneOnly = ORTCoreMLProviderOption.aneOnly.rawValue
+        flags.remove(cpuOnly)
+        flags.remove(cpuAndGPU)
+        flags.remove(aneOnly)
+        switch self {
+        case .automatic: break
+        case .cpuOnly: flags.insert(cpuOnly)
+        case .cpuAndGPU: flags.insert(cpuAndGPU)
+        case .devicesWithANE: flags.insert(aneOnly)
+        }
+    }
+}
+
+/// ONNX Runtime: preferred execution provider (EP) preset plus session tuning. Card content only.
+struct OnnxRuntimeSettingsPanel: View {
+    private static let coreMLExtraToggleOptions: [ORTCoreMLProviderOption] =
+        [.enableOnSubgraphs, .createMLProgram, .staticInputShapes]
+
+    @Binding var primaryExecutionProvider: ORTPrimaryExecutionProvider
+    @Binding var tuning: ORTSessionTuningOptions
+
+    var body: some View {
+        VStack(alignment: .leading, spacing: 14) {
+            Text("All inference uses ONNX Runtime. Choose which EP is preferred first for this demo preset; unsupported operators still run on the CPU EP.")
+                .font(.caption)
+                .foregroundStyle(.secondary)
+
+            Picker("Execution provider", selection: $primaryExecutionProvider) {
+                ForEach(ORTPrimaryExecutionProvider.allCases) { provider in
+                    Text(provider.displayTitle).tag(provider)
+                }
+            }
+            .pickerStyle(.segmented)
+
+            Divider()
+                .padding(.vertical, 2)
+
+            Text("Session options")
+                .font(.subheadline.weight(.semibold))
+
+            labeledStepper(
+                title: "Intra-op threads",
+                caption: "Session-wide CPU parallelism. Use 0 to leave ONNX Runtime’s default (no explicit limit).",
+                value: intraOpThreadsBinding,
+                range: 0...8
+            )
+
+            if primaryExecutionProvider == .xnnpack {
+                labeledStepper(
+                    title: "XNNPACK intra-op threads",
+                    caption: "0 leaves the XNNPACK EP default.",
+                    value: xnnpackThreadsBinding,
+                    range: 0...8
+                )
+            }
+
+            if primaryExecutionProvider == .coreML {
+                VStack(alignment: .leading, spacing: 10) {
+                    Text("Core ML EP options")
+                        .font(.subheadline.weight(.semibold))
+                        .padding(.top, 4)
+
+                    VStack(alignment: .leading, spacing: 6) {
+                        Text("Core ML placement")
+                            .font(.body)
+                        Picker("Core ML placement", selection: coreMLComputeModeBinding) {
+                            ForEach(CoreMLComputeMode.allCases) { mode in
+                                Text(mode.segmentLabel).tag(mode)
+                            }
+                        }
+                        .pickerStyle(.segmented)
+                        Text(coreMLComputeModeCaption)
+                            .font(.caption)
+                            .foregroundStyle(.secondary)
+                    }
+
+                    ForEach(Self.coreMLExtraToggleOptions) { option in
+                        Toggle(isOn: coreMLFlagBinding(option)) {
+                            VStack(alignment: .leading, spacing: 2) {
+                                Text(option.displayTitle)
+                                Text(option.detailCaption)
+                                    .font(.caption)
+                                    .foregroundStyle(.secondary)
+                            }
+                        }
+                    }
+                }
+            }
+
+            HStack {
+                Spacer()
+                Button("Reset session options") {
+                    tuning = .default
+                }
+                .font(.subheadline.weight(.medium))
+            }
+            .padding(.top, 4)
+        }
+    }
+
+    private var coreMLComputeModeCaption: String {
+        switch CoreMLComputeMode.from(flags: tuning.coreMLFlags) {
+        case .automatic:
+            return "ORT defaults — no cpuOnly/cpuAndGPU/device ANE restriction."
+        case .cpuOnly:
+            return "useCPUOnly — Core ML constrained to CPU."
+        case .cpuAndGPU:
+            return "useCPUAndGPU — CPU and GPU paths; excludes ANE in Core ML."
+        case .devicesWithANE:
+            return "Only enable Core ML EP on devices that have a Neural Engine."
+        }
+    }
+
+    private var intraOpThreadsBinding: Binding<Int> {
+        Binding(
+            get: { tuning.intraOpThreads },
+            set: { newValue in
+                var next = tuning
+                next.intraOpThreads = newValue
+                tuning = next
+            }
+        )
+    }
+
+    private var xnnpackThreadsBinding: Binding<Int> {
+        Binding(
+            get: { tuning.xnnpackThreads },
+            set: { newValue in
+                var next = tuning
+                next.xnnpackThreads = newValue
+                tuning = next
+            }
+        )
+    }
+
+    private var coreMLComputeModeBinding: Binding<CoreMLComputeMode> {
+        Binding(
+            get: { CoreMLComputeMode.from(flags: tuning.coreMLFlags) },
+            set: { mode in
+                var next = tuning
+                mode.merge(into: &next.coreMLFlags)
+                tuning = next
+            }
+        )
+    }
+
+    private func coreMLFlagBinding(_ option: ORTCoreMLProviderOption) -> Binding<Bool> {
+        let key = option.rawValue
+        return Binding(
+            get: { tuning.coreMLFlags.contains(key) },
+            set: { on in
+                var next = tuning
+                if on {
+                    next.coreMLFlags.insert(key)
+                } else {
+                    next.coreMLFlags.remove(key)
+                }
+                tuning = next
+            }
+        )
+    }
+
+    private func labeledStepper(title: String, caption: String, value: Binding<Int>, range: ClosedRange<Int>) -> some View {
+        VStack(alignment: .leading, spacing: 6) {
+            HStack(alignment: .firstTextBaseline) {
+                Text(title)
+                Spacer()
+                Stepper(value: value, in: range) {
+                    Text("\(value.wrappedValue)")
+                        .font(.body.monospacedDigit())
+                        .frame(minWidth: 24, alignment: .trailing)
+                }
+            }
+            Text(caption)
+                .font(.caption)
+                .foregroundStyle(.secondary)
+        }
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Views/ResultImageView.swift b/deploy/ios_demo/PaddleOCRDemo/Views/ResultImageView.swift
new file mode 100644
index 0000000..12bb749
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Views/ResultImageView.swift
@@ -0,0 +1,79 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import SwiftUI
+
+/// Displays the source image with bounding box polygon overlays drawn on a Canvas.
+///
+/// Polygon coordinates from `OCRResult.polygon` are in original image pixel space.
+/// This view computes the scale factor and offset to map them to the displayed
+/// image area, accounting for `.aspectFit` rendering within the available frame.
+struct ResultImageView: View {
+    let image: UIImage
+    let results: [OCRResult]
+
+    var body: some View {
+        GeometryReader { geo in
+            let imageSize = image.size
+            let imageAspect = imageSize.width / imageSize.height
+            let viewAspect = geo.size.width / geo.size.height
+
+            // Compute the actual rendered image rect within the GeometryReader frame
+            let fitSize: CGSize = imageAspect > viewAspect
+                ? CGSize(width: geo.size.width, height: geo.size.width / imageAspect)
+                : CGSize(width: geo.size.height * imageAspect, height: geo.size.height)
+            let offset = CGPoint(
+                x: (geo.size.width - fitSize.width) / 2,
+                y: (geo.size.height - fitSize.height) / 2
+            )
+            let scale = fitSize.width / imageSize.width
+
+            ZStack {
+                Image(uiImage: image)
+                    .resizable()
+                    .aspectRatio(contentMode: .fit)
+
+                Canvas { context, size in
+                    for result in results {
+                        let points = result.polygon.compactMap { coords -> CGPoint? in
+                            guard coords.count >= 2 else { return nil }
+                            return CGPoint(
+                                x: CGFloat(coords[0]) * scale + offset.x,
+                                y: CGFloat(coords[1]) * scale + offset.y
+                            )
+                        }
+                        guard points.count >= 4 else { continue }
+
+                        var path = Path()
+                        path.move(to: points[0])
+                        for pt in points.dropFirst() {
+                            path.addLine(to: pt)
+                        }
+                        path.closeSubpath()
+
+                        let strokeColor = Color.accentColor.opacity(0.9)
+                        let fillColor = Color.accentColor.opacity(0.12)
+                        context.stroke(path, with: .color(strokeColor), lineWidth: 2)
+                        context.fill(path, with: .color(fillColor))
+                    }
+                }
+            }
+            .frame(width: geo.size.width, height: geo.size.height)
+        }
+        .aspectRatio(image.size.width / image.size.height, contentMode: .fit)
+        .frame(maxHeight: 400)
+        .accessibilityElement(children: .combine)
+        .accessibilityLabel("Image preview with \(results.count) detected text regions")
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Views/ResultsListView.swift b/deploy/ios_demo/PaddleOCRDemo/Views/ResultsListView.swift
new file mode 100644
index 0000000..d69e643
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Views/ResultsListView.swift
@@ -0,0 +1,98 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import SwiftUI
+
+/// Recognized lines + copy action (intended to sit inside `DemoCard` from the parent).
+struct ResultsListView: View {
+    let results: [OCRResult]
+    let copiedFeedback: Bool
+    let onCopy: () -> Void
+
+    var body: some View {
+        VStack(alignment: .leading, spacing: 0) {
+            HStack(alignment: .firstTextBaseline) {
+                Text("Recognized text")
+                    .font(.headline)
+                Spacer()
+                Button(action: onCopy) {
+                    Label(
+                        copiedFeedback ? "Copied" : "Copy all",
+                        systemImage: copiedFeedback ? "checkmark.circle.fill" : "doc.on.doc"
+                    )
+                    .labelStyle(.titleAndIcon)
+                }
+                .font(.subheadline.weight(.medium))
+                .buttonStyle(.bordered)
+                .controlSize(.small)
+                .disabled(results.isEmpty)
+            }
+            .padding(.bottom, 12)
+
+            if results.isEmpty {
+                emptyState
+            } else {
+                VStack(alignment: .leading, spacing: 0) {
+                    ForEach(Array(results.enumerated()), id: \.offset) { index, result in
+                        resultRow(index: index, result: result)
+                        if index < results.count - 1 {
+                            Divider()
+                                .padding(.vertical, 10)
+                        }
+                    }
+                }
+            }
+        }
+    }
+
+    private var emptyState: some View {
+        VStack(spacing: 12) {
+            Image(systemName: "text.magnifyingglass")
+                .font(.system(size: 32))
+                .foregroundStyle(.secondary)
+                .symbolRenderingMode(.hierarchical)
+            Text("No text to show")
+                .font(.subheadline.weight(.semibold))
+            Text("Try lowering thresholds or use an image with clearer text.")
+                .font(.caption)
+                .foregroundStyle(.secondary)
+                .multilineTextAlignment(.center)
+        }
+        .frame(maxWidth: .infinity)
+        .padding(.vertical, 20)
+    }
+
+    private func resultRow(index: Int, result: OCRResult) -> some View {
+        HStack(alignment: .top, spacing: 12) {
+            Text("\(index + 1)")
+                .font(.caption.weight(.bold))
+                .foregroundStyle(.secondary)
+                .frame(width: 22, height: 22)
+                .background(Color(.tertiarySystemFill), in: Circle())
+
+            Text(result.text)
+                .font(.body)
+                .foregroundStyle(.primary)
+                .frame(maxWidth: .infinity, alignment: .leading)
+                .lineLimit(5)
+
+            Text(String(format: "%.0f%%", result.confidence * 100))
+                .font(.caption.monospacedDigit())
+                .foregroundStyle(.secondary)
+                .padding(.horizontal, 6)
+                .padding(.vertical, 2)
+                .background(Color(.tertiarySystemFill), in: Capsule())
+        }
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemo/Views/TimingView.swift b/deploy/ios_demo/PaddleOCRDemo/Views/TimingView.swift
new file mode 100644
index 0000000..021aef6
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemo/Views/TimingView.swift
@@ -0,0 +1,54 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import SwiftUI
+
+/// Compact per-stage timing chips (fits the “output” section below the preview).
+struct TimingView: View {
+    let result: OCRRunResult
+
+    var body: some View {
+        HStack(spacing: 10) {
+            chip(title: "Detect", ms: result.detectionTime)
+            chip(title: "Recognize", ms: result.recognitionTime)
+            chip(title: "Total", ms: result.totalTime, emphasized: true)
+        }
+        .frame(maxWidth: .infinity)
+    }
+
+    private func chip(title: String, ms: TimeInterval, emphasized: Bool = false) -> some View {
+        VStack(spacing: 6) {
+            Text(title)
+                .font(.caption2.weight(.semibold))
+                .foregroundStyle(.secondary)
+            Text(String(format: "%.0f", ms * 1000))
+                .font(.system(.body, design: .rounded).weight(emphasized ? .semibold : .regular))
+                .monospacedDigit()
+            Text("ms")
+                .font(.caption2)
+                .foregroundStyle(.tertiary)
+        }
+        .frame(maxWidth: .infinity)
+        .padding(.vertical, 12)
+        .padding(.horizontal, 8)
+        .background {
+            RoundedRectangle(cornerRadius: 12, style: .continuous)
+                .fill(emphasized ? Color.accentColor.opacity(0.12) : Color(.tertiarySystemFill))
+        }
+        .overlay {
+            RoundedRectangle(cornerRadius: 12, style: .continuous)
+                .strokeBorder(Color(.separator).opacity(0.25), lineWidth: 0.5)
+        }
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemoTests/BenchmarkFixtures.swift b/deploy/ios_demo/PaddleOCRDemoTests/BenchmarkFixtures.swift
new file mode 100644
index 0000000..a804213
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemoTests/BenchmarkFixtures.swift
@@ -0,0 +1,25 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import Foundation
+
+/// Default bundled image file names for benchmark tests.
+enum BenchmarkFixtures {
+    static let defaultReferenceImageStem: String = "ios_ocr_benchmark_reference"
+    static let defaultReferenceImageFileExtension: String = "jpg"
+
+    static var defaultReferenceImageFileName: String {
+        "\(defaultReferenceImageStem).\(defaultReferenceImageFileExtension)"
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemoTests/BenchmarkPayloads.swift b/deploy/ios_demo/PaddleOCRDemoTests/BenchmarkPayloads.swift
new file mode 100644
index 0000000..ae146ab
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemoTests/BenchmarkPayloads.swift
@@ -0,0 +1,95 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+
+import Foundation
+
+@testable import PaddleOCRDemo
+
+/// Names for JSON written as `XCTAttachment`s; keep in sync with whatever exports attachments from test results.
+enum BenchmarkArtifact: String {
+    case iOSExport = "ios-ocr-export.json"
+    case onDevicePerformance = "on-device-performance.json"
+}
+
+// MARK: - Accuracy export payload
+
+struct OCRExportPayload: Codable {
+    var schemaVersion: Int = 1
+    var source: String
+    var items: [OCRExportItem]
+}
+
+struct OCRExportItem: Codable {
+    var polygon: [[Int]]
+    var text: String
+    var score: Double?
+}
+
+// MARK: - On-device performance payload
+
+struct TimingSummary: Codable {
+    var mean: Double
+    var stdev: Double
+    var p90: Double
+}
+
+/// Pooled: one sample per det line per measured run (all lines concatenated, ms).
+struct RecognitionPerLinePooledMs: Codable {
+    var count: Int
+    var inferenceMs: TimingSummary
+    var preprocessMs: TimingSummary
+    var postprocessMs: TimingSummary
+    var totalMs: TimingSummary
+}
+
+struct RecognitionPerLineBlock: Codable {
+    var pooled: RecognitionPerLinePooledMs
+}
+
+struct BenchmarkTensorShapeSample: Codable {
+    var shape: [Int]
+    var count: Int
+}
+
+struct BenchmarkInputShapeDistribution: Codable {
+    var detection: [BenchmarkTensorShapeSample]
+    var recognition: [BenchmarkTensorShapeSample]
+}
+
+struct OCRDeviceBenchmarkPayload: Codable {
+    var schemaVersion: Int = 1
+    var buildConfiguration: String
+    var deviceModel: String
+    var osVersion: String
+    var isSimulator: Bool
+    var ortExecutionProvider: String
+    var ortProfilingEnabled: Bool
+    var warmupIterations: Int
+    var measuredIterations: Int
+    var inputShapeDistribution: BenchmarkInputShapeDistribution
+    var firstMeasuredLineCount: Int
+    var coldLoadTimeMs: Double
+    var totalTimeMs: TimingSummary
+    var detectionTimeMs: TimingSummary
+    var detectionPreprocessTimeMs: TimingSummary
+    var detectionInferenceTimeMs: TimingSummary
+    var detectionPostprocessTimeMs: TimingSummary
+    var recognitionTimeMs: TimingSummary
+    var recognitionPreprocessTimeMs: TimingSummary
+    var recognitionInferenceTimeMs: TimingSummary
+    var recognitionPostprocessTimeMs: TimingSummary
+    var recognitionPerLine: RecognitionPerLineBlock?
+    var pipelineOverheadTimeMs: TimingSummary
+    var memoryFootprintBeforeLoadBytes: UInt64?
+    var memoryFootprintAfterLoadBytes: UInt64
+    var thermalState: String?
+}
diff --git a/deploy/ios_demo/PaddleOCRDemoTests/Fixtures/.gitignore b/deploy/ios_demo/PaddleOCRDemoTests/Fixtures/.gitignore
new file mode 100644
index 0000000..e7f372f
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemoTests/Fixtures/.gitignore
@@ -0,0 +1,2 @@
+# Ad-hoc copies from ./scripts/run_validation.sh --image (local-*); not committed.
+local-*
diff --git a/deploy/ios_demo/PaddleOCRDemoTests/Fixtures/ios_ocr_validation_reference.jpg b/deploy/ios_demo/PaddleOCRDemoTests/Fixtures/ios_ocr_validation_reference.jpg
new file mode 100644
index 0000000..5c3329a
Binary files /dev/null and b/deploy/ios_demo/PaddleOCRDemoTests/Fixtures/ios_ocr_validation_reference.jpg differ
diff --git a/deploy/ios_demo/PaddleOCRDemoTests/OCRBenchmarkTests.swift b/deploy/ios_demo/PaddleOCRDemoTests/OCRBenchmarkTests.swift
new file mode 100644
index 0000000..94d57d7
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemoTests/OCRBenchmarkTests.swift
@@ -0,0 +1,562 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import Darwin
+import Foundation
+
+import XCTest
+
+@testable import PaddleOCRDemo
+
+import UIKit
+
+// MARK: - Tests
+
+final class OCRBenchmarkTests: XCTestCase {
+
+    /// `PADDLEOCR_BENCHMARK_IMAGE_NAME`: bundled image stem or `stem.ext` (optional).
+    private static let imageNameEnvKey = "PADDLEOCR_BENCHMARK_IMAGE_NAME"
+
+    /// Optional non-negative int; default `3`. Used only by `testOnDevicePerformanceMetrics`.
+    private static let warmupIterationsEnvKey = "PADDLEOCR_BENCHMARK_WARMUP_ITERATIONS"
+
+    /// Optional non-negative int; default `10`. Used only by `testOnDevicePerformanceMetrics`.
+    private static let measuredIterationsEnvKey = "PADDLEOCR_BENCHMARK_MEASURED_ITERATIONS"
+
+    /// `CPU` (default), `XNNPACK`, or `CORE_ML` — ONNX Runtime execution provider preset for benchmarks.
+    private static let ortExecutionProviderEnvKey = "PADDLEOCR_BENCHMARK_ORT_EXECUTION_PROVIDER"
+
+    /// `1` / `true` / `yes` / `on` enables ONNX Runtime JSON profiling (see ``ORTSessionManager/finalizeORTProfiling()``).
+    /// When set, on-device performance timings are not representative of a clean latency benchmark.
+    private static let ortProfilingEnvKey = "PADDLEOCR_BENCHMARK_ORT_PROFILING"
+
+    private static let buildConfigurationEnvKey = "PADDLEOCR_BENCHMARK_BUILD_CONFIGURATION"
+
+    private static let intraOpThreadsEnvKey = "PADDLEOCR_BENCHMARK_INTRA_OP_THREADS"
+    private static let xnnpackThreadsEnvKey = "PADDLEOCR_BENCHMARK_XNNPACK_THREADS"
+    private static let coremlOptionsEnvKey = "PADDLEOCR_BENCHMARK_COREML_OPTIONS"
+    private static let recBatchSizeEnvKey = "PADDLEOCR_BENCHMARK_REC_BATCH_SIZE"
+
+    func testOCRExportJSONSchema() async throws {
+        let cgImage = try resolveBenchmarkImage()
+        let executionProvider = try resolveBenchmarkExecutionProvider()
+        let manager = ORTSessionManager()
+        try await manager.loadModels(executionProvider: executionProvider, ortProfiling: resolveORTProfilingEnabled(), tuning: resolveSessionTuningOptions())
+        let engine = try OCREngine(sessionManager: manager)
+        let run = try await engine.run(cgImage, params: .noOverrides)
+
+        let payload = OCRExportPayload(
+            source: "ios_ocr_demo",
+            items: run.results.map { r in
+                OCRExportItem(
+                    polygon: r.polygon.map { $0.map { Int($0) } },
+                    text: r.text,
+                    score: Double(r.confidence)
+                )
+            }
+        )
+
+        let encoder = JSONEncoder()
+        encoder.outputFormatting = [.sortedKeys]
+        let data = try encoder.encode(payload)
+        _ = try JSONDecoder().decode(OCRExportPayload.self, from: data)
+        attachJSON(data, artifact: .iOSExport)
+
+        if resolveORTProfilingEnabled(), let out = try await manager.finalizeORTProfiling() {
+            attachProfileFile(out.detectionProfileJSON, name: "ort_profile_detection")
+            attachProfileFile(out.recognitionProfileJSON, name: "ort_profile_recognition")
+        }
+    }
+
+    func testOnDeviceLatencyBenchmark() async throws {
+        let executionProvider = try resolveBenchmarkExecutionProvider()
+        let ortProfiling = resolveORTProfilingEnabled()
+        let context = try await makeBenchmarkContext(executionProvider: executionProvider, ortProfiling: ortProfiling)
+        let warmup = try parseNonNegativeIntEnv(Self.warmupIterationsEnvKey, defaultValue: 3)
+        let iterations = max(try parseNonNegativeIntEnv(Self.measuredIterationsEnvKey, defaultValue: 10), 1)
+        try await runWarmup(engine: context.engine, image: context.cgImage, iterations: warmup)
+
+        var totals: [Double] = []
+        var dets: [Double] = []
+        var detPre: [Double] = []
+        var detInf: [Double] = []
+        var detPost: [Double] = []
+        var recs: [Double] = []
+        var recPre: [Double] = []
+        var recInf: [Double] = []
+        var recPost: [Double] = []
+        var overheads: [Double] = []
+        totals.reserveCapacity(iterations)
+        dets.reserveCapacity(iterations)
+        detPre.reserveCapacity(iterations)
+        detInf.reserveCapacity(iterations)
+        detPost.reserveCapacity(iterations)
+        recs.reserveCapacity(iterations)
+        recPre.reserveCapacity(iterations)
+        recInf.reserveCapacity(iterations)
+        recPost.reserveCapacity(iterations)
+        overheads.reserveCapacity(iterations)
+
+        var lineInferenceMsPooled: [Double] = []
+        var linePreprocessMsPooled: [Double] = []
+        var linePostprocessMsPooled: [Double] = []
+        var lineTotalMsPooled: [Double] = []
+        var firstMeasuredLineCount = 0
+        var detectionInputShapes: [[Int]] = []
+        var recognitionInputShapes: [[Int]] = []
+
+        let benchmarkParams = resolveBenchmarkRuntimeParams()
+
+        for index in 0..<iterations {
+            let run = try await context.engine.run(context.cgImage, params: benchmarkParams)
+            if index == 0 {
+                firstMeasuredLineCount = run.recognitionLineCount
+            }
+            detectionInputShapes.append(run.detectionInputTensorShape)
+            recognitionInputShapes.append(contentsOf: run.recognitionInputTensorShapes)
+            appendRun(
+                run,
+                totals: &totals,
+                dets: &dets,
+                detPre: &detPre,
+                detInf: &detInf,
+                detPost: &detPost,
+                recs: &recs,
+                recPre: &recPre,
+                recInf: &recInf,
+                recPost: &recPost,
+                overheads: &overheads,
+                lineInferenceMsPooled: &lineInferenceMsPooled,
+                linePreprocessMsPooled: &linePreprocessMsPooled,
+                linePostprocessMsPooled: &linePostprocessMsPooled,
+                lineTotalMsPooled: &lineTotalMsPooled
+            )
+        }
+
+        let perLine: RecognitionPerLineBlock? = lineInferenceMsPooled.isEmpty
+            ? nil
+            : RecognitionPerLineBlock(
+                pooled: RecognitionPerLinePooledMs(
+                    count: lineInferenceMsPooled.count,
+                    inferenceMs: summarizeMs(lineInferenceMsPooled),
+                    preprocessMs: summarizeMs(linePreprocessMsPooled),
+                    postprocessMs: summarizeMs(linePostprocessMsPooled),
+                    totalMs: summarizeMs(lineTotalMsPooled)
+                )
+            )
+
+        let stats = OCRDeviceBenchmarkPayload(
+            schemaVersion: 1,
+            buildConfiguration: resolveBuildConfiguration(),
+            deviceModel: deviceModelName(),
+            osVersion: ProcessInfo.processInfo.operatingSystemVersionString,
+            isSimulator: isRunningOnSimulator(),
+            ortExecutionProvider: executionProvider.rawValue,
+            ortProfilingEnabled: ortProfiling,
+            warmupIterations: warmup,
+            measuredIterations: iterations,
+            inputShapeDistribution: BenchmarkInputShapeDistribution(
+                detection: summarizeShapes(detectionInputShapes),
+                recognition: summarizeShapes(recognitionInputShapes)
+            ),
+            firstMeasuredLineCount: firstMeasuredLineCount,
+            coldLoadTimeMs: context.coldLoadTime * 1000,
+            totalTimeMs: summarizeMs(totals),
+            detectionTimeMs: summarizeMs(dets),
+            detectionPreprocessTimeMs: summarizeMs(detPre),
+            detectionInferenceTimeMs: summarizeMs(detInf),
+            detectionPostprocessTimeMs: summarizeMs(detPost),
+            recognitionTimeMs: summarizeMs(recs),
+            recognitionPreprocessTimeMs: summarizeMs(recPre),
+            recognitionInferenceTimeMs: summarizeMs(recInf),
+            recognitionPostprocessTimeMs: summarizeMs(recPost),
+            recognitionPerLine: perLine,
+            pipelineOverheadTimeMs: summarizeMs(overheads),
+            memoryFootprintBeforeLoadBytes: context.memoryBeforeLoad,
+            memoryFootprintAfterLoadBytes: context.memoryAfterLoad,
+            thermalState: String(describing: ProcessInfo.processInfo.thermalState)
+        )
+
+        let encoder = JSONEncoder()
+        encoder.outputFormatting = [.sortedKeys]
+        let encoded = try encoder.encode(stats)
+        let json = String(data: encoded, encoding: .utf8) ?? ""
+        XCTAssertFalse(json.isEmpty, "on-device performance stats should encode")
+
+        attachJSON(encoded, artifact: .onDevicePerformance)
+
+        if ortProfiling, let out = try await context.manager.finalizeORTProfiling() {
+            attachProfileFile(out.detectionProfileJSON, name: "ort_profile_detection")
+            attachProfileFile(out.recognitionProfileJSON, name: "ort_profile_recognition")
+        }
+    }
+
+    func testOnDeviceMemoryBenchmark() async throws {
+        let executionProvider = try resolveBenchmarkExecutionProvider()
+        let context = try await makeBenchmarkContext(executionProvider: executionProvider, ortProfiling: false)
+        let warmup = try parseNonNegativeIntEnv(Self.warmupIterationsEnvKey, defaultValue: 3)
+        let iterations = max(try parseNonNegativeIntEnv(Self.measuredIterationsEnvKey, defaultValue: 10), 1)
+        try await runWarmup(engine: context.engine, image: context.cgImage, iterations: warmup)
+
+        var measuredError: Error?
+        let options = XCTMeasureOptions()
+        options.iterationCount = iterations
+        measure(metrics: [XCTMemoryMetric()], options: options) {
+            guard measuredError == nil else { return }
+            do {
+                _ = try waitForAsync {
+                    try await context.engine.run(context.cgImage, params: self.resolveBenchmarkRuntimeParams())
+                }
+            } catch {
+                measuredError = error
+            }
+        }
+        if let measuredError {
+            throw measuredError
+        }
+    }
+
+    // MARK: - Helpers
+
+    private struct BenchmarkContext {
+        let cgImage: CGImage
+        let manager: ORTSessionManager
+        let engine: OCREngine
+        let memoryBeforeLoad: UInt64
+        let memoryAfterLoad: UInt64
+        let coldLoadTime: TimeInterval
+    }
+
+    private func makeBenchmarkContext(executionProvider: ORTPrimaryExecutionProvider, ortProfiling: Bool) async throws -> BenchmarkContext {
+        let cgImage = try resolveBenchmarkImage()
+        let memoryBeforeLoad = physicalFootprintBytes()
+        let manager = ORTSessionManager()
+        let loadStart = CFAbsoluteTimeGetCurrent()
+        try await manager.loadModels(executionProvider: executionProvider, ortProfiling: ortProfiling, tuning: resolveSessionTuningOptions())
+        let coldLoadTime = CFAbsoluteTimeGetCurrent() - loadStart
+        let memoryAfterLoad = physicalFootprintBytes()
+        let engine = try OCREngine(sessionManager: manager)
+        return BenchmarkContext(
+            cgImage: cgImage,
+            manager: manager,
+            engine: engine,
+            memoryBeforeLoad: memoryBeforeLoad,
+            memoryAfterLoad: memoryAfterLoad,
+            coldLoadTime: coldLoadTime
+        )
+    }
+
+    private func summarizeShapes(_ shapes: [[Int]]) -> [BenchmarkTensorShapeSample] {
+        let counts = shapes.reduce(into: [[Int]: Int]()) { partial, shape in
+            partial[shape, default: 0] += 1
+        }
+        return counts
+            .map { BenchmarkTensorShapeSample(shape: $0.key, count: $0.value) }
+            .sorted { lhs, rhs in
+                if lhs.shape != rhs.shape {
+                    return lhs.shape.lexicographicallyPrecedes(rhs.shape)
+                }
+                return lhs.count > rhs.count
+            }
+    }
+
+    private func runWarmup(engine: OCREngine, image: CGImage, iterations: Int) async throws {
+        let params = resolveBenchmarkRuntimeParams()
+        for _ in 0..<iterations {
+            _ = try await engine.run(image, params: params)
+        }
+    }
+
+    private func appendRun(
+        _ run: OCRRunResult,
+        totals: inout [Double],
+        dets: inout [Double],
+        detPre: inout [Double],
+        detInf: inout [Double],
+        detPost: inout [Double],
+        recs: inout [Double],
+        recPre: inout [Double],
+        recInf: inout [Double],
+        recPost: inout [Double],
+        overheads: inout [Double],
+        lineInferenceMsPooled: inout [Double],
+        linePreprocessMsPooled: inout [Double],
+        linePostprocessMsPooled: inout [Double],
+        lineTotalMsPooled: inout [Double]
+    ) {
+        totals.append(run.totalTime * 1000)
+        dets.append(run.detectionTime * 1000)
+        detPre.append(run.detectionPreprocessTime * 1000)
+        detInf.append(run.detectionInferenceTime * 1000)
+        detPost.append(run.detectionPostprocessTime * 1000)
+        recs.append(run.recognitionTime * 1000)
+        recPre.append(run.recognitionPreprocessTime * 1000)
+        recInf.append(run.recognitionInferenceTime * 1000)
+        recPost.append(run.recognitionPostprocessTime * 1000)
+        overheads.append(run.pipelineOverheadTime * 1000)
+
+        let n = run.recognitionLineCount
+        if n > 0 {
+            let infMs = run.lineRecognitionInferenceTimes.map { $0 * 1000.0 }
+            let preMs = run.lineRecognitionPreprocessTimes.map { $0 * 1000.0 }
+            let postMs = run.lineRecognitionPostprocessTimes.map { $0 * 1000.0 }
+            lineInferenceMsPooled.append(contentsOf: infMs)
+            linePreprocessMsPooled.append(contentsOf: preMs)
+            linePostprocessMsPooled.append(contentsOf: postMs)
+            for i in 0..<n {
+                lineTotalMsPooled.append(preMs[i] + infMs[i] + postMs[i])
+            }
+        }
+    }
+
+    private func resolveBuildConfiguration() -> String {
+        let raw =
+            ProcessInfo.processInfo.environment[Self.buildConfigurationEnvKey]?
+                .trimmingCharacters(in: .whitespacesAndNewlines) ?? ""
+        return raw.isEmpty ? "Release" : raw
+    }
+
+    private func isRunningOnSimulator() -> Bool {
+        #if targetEnvironment(simulator)
+        return true
+        #else
+        return false
+        #endif
+    }
+
+    private func deviceModelName() -> String {
+        var systemInfo = utsname()
+        uname(&systemInfo)
+        let mirror = Mirror(reflecting: systemInfo.machine)
+        return mirror.children.reduce(into: "") { identifier, element in
+            guard let value = element.value as? Int8, value != 0 else { return }
+            identifier.append(String(UnicodeScalar(UInt8(value))))
+        }
+    }
+
+    private func resolveORTProfilingEnabled() -> Bool {
+        let raw =
+            ProcessInfo.processInfo.environment[Self.ortProfilingEnvKey]?
+                .trimmingCharacters(in: .whitespacesAndNewlines) ?? ""
+        if raw.isEmpty { return false }
+        switch raw.lowercased() {
+        case "1", "true", "yes", "on":
+            return true
+        default:
+            return false
+        }
+    }
+
+    private func resolveSessionTuningOptions() -> ORTSessionTuningOptions {
+        let env = ProcessInfo.processInfo.environment
+        var opts = ORTSessionTuningOptions()
+        if let v = env[Self.intraOpThreadsEnvKey], let n = Int(v), n > 0 {
+            opts.intraOpThreads = n
+        }
+        if let v = env[Self.xnnpackThreadsEnvKey], let n = Int(v), n > 0 {
+            opts.xnnpackThreads = n
+        }
+        if let v = env[Self.coremlOptionsEnvKey], !v.isEmpty {
+            opts.coreMLFlags = Set(v.split(separator: ",").map { String($0).trimmingCharacters(in: .whitespaces) })
+        }
+        return opts
+    }
+
+    private func attachProfileFile(_ url: URL, name: String) {
+        let attachment = XCTAttachment(contentsOfFile: url)
+        attachment.name = name
+        attachment.lifetime = XCTAttachment.Lifetime.keepAlways
+        add(attachment)
+    }
+
+    private func resolveBenchmarkExecutionProvider() throws -> ORTPrimaryExecutionProvider {
+        let raw =
+            ProcessInfo.processInfo.environment[Self.ortExecutionProviderEnvKey]?
+                .trimmingCharacters(in: .whitespacesAndNewlines) ?? ""
+        if raw.isEmpty {
+            return .cpu
+        }
+        if let exact = ORTPrimaryExecutionProvider(rawValue: raw) {
+            return exact
+        }
+        switch raw.lowercased() {
+        case "core_ml":
+            return .coreML
+        case "xnnpack":
+            return .xnnpack
+        case "cpu":
+            return .cpu
+        default:
+            throw NSError(
+                domain: "OCRBenchmarkTests",
+                code: 5,
+                userInfo: [
+                    NSLocalizedDescriptionKey:
+                        "Invalid \(Self.ortExecutionProviderEnvKey): \"\(raw)\". "
+                        + "Use CPU (default), XNNPACK, CORE_ML, or Swift raw values "
+                        + "\(ORTPrimaryExecutionProvider.cpu.rawValue) / "
+                        + "\(ORTPrimaryExecutionProvider.xnnpack.rawValue) / "
+                        + "\(ORTPrimaryExecutionProvider.coreML.rawValue).",
+                ]
+            )
+        }
+    }
+
+    private func resolveBenchmarkImage() throws -> CGImage {
+        let envRaw =
+            ProcessInfo.processInfo.environment[Self.imageNameEnvKey]?
+                .trimmingCharacters(in: .whitespacesAndNewlines) ?? ""
+        let raw = envRaw.isEmpty ? BenchmarkFixtures.defaultReferenceImageStem : envRaw
+        let bundle = Bundle(for: Self.self)
+        guard let path = bundle.path(forBundledImageNamed: raw, subdirectory: "Fixtures") else {
+            throw NSError(
+                domain: "OCRBenchmarkTests",
+                code: 3,
+                userInfo: [
+                    NSLocalizedDescriptionKey:
+                        "Image \"\(raw)\" not found in the test bundle. Check the name and that the file is included in the test target resources (e.g. \"\(BenchmarkFixtures.defaultReferenceImageStem)\").",
+                ]
+            )
+        }
+        let url = URL(fileURLWithPath: path)
+        guard let ui = UIImage(contentsOfFile: url.path),
+              let cg = normalizeOrientation(ui).cgImage
+        else {
+            throw NSError(
+                domain: "OCRBenchmarkTests",
+                code: 4,
+                userInfo: [NSLocalizedDescriptionKey: "Could not decode image at \(url.path)."]
+            )
+        }
+        return cg
+    }
+
+    private func attachJSON(_ data: Data, artifact: BenchmarkArtifact) {
+        let attachment = XCTAttachment(data: data, uniformTypeIdentifier: "public.json")
+        attachment.name = artifact.rawValue
+        attachment.lifetime = .keepAlways
+        add(attachment)
+    }
+
+    private func waitForAsync<T>(_ operation: @escaping () async throws -> T) throws -> T {
+        let semaphore = DispatchSemaphore(value: 0)
+        let lock = NSLock()
+        var capturedResult: Result<T, Error>?
+        Task {
+            do {
+                let value = try await operation()
+                lock.lock()
+                capturedResult = .success(value)
+                lock.unlock()
+            } catch {
+                lock.lock()
+                capturedResult = .failure(error)
+                lock.unlock()
+            }
+            semaphore.signal()
+        }
+        semaphore.wait()
+        lock.lock()
+        let result = capturedResult
+        lock.unlock()
+        switch result {
+        case .success(let value):
+            return value
+        case .failure(let error):
+            throw error
+        case .none:
+            throw NSError(
+                domain: "OCRBenchmarkTests",
+                code: 6,
+                userInfo: [NSLocalizedDescriptionKey: "Async operation completed without a result."]
+            )
+        }
+    }
+
+    /// Parses a non-negative integer from the environment, or returns `defaultValue` when unset/blank.
+    private func parseNonNegativeIntEnv(_ key: String, defaultValue: Int) throws -> Int {
+        let raw =
+            ProcessInfo.processInfo.environment[key]?.trimmingCharacters(in: .whitespacesAndNewlines) ?? ""
+        if raw.isEmpty {
+            return defaultValue
+        }
+        guard let n = Int(raw), n >= 0 else {
+            throw NSError(
+                domain: "OCRBenchmarkTests",
+                code: 3,
+                userInfo: [
+                    NSLocalizedDescriptionKey:
+                        "Invalid \(key): expected a non-negative integer (default \(defaultValue) if unset), got \(raw).",
+                ]
+            )
+        }
+        return n
+    }
+
+    private func resolveBenchmarkRuntimeParams() -> OCRRuntimeParams {
+        let env = ProcessInfo.processInfo.environment
+        if let v = env[Self.recBatchSizeEnvKey], let n = Int(v), n >= 1 {
+            return OCRRuntimeParams(
+                textDetLimitSideLen: nil,
+                textDetLimitType: nil,
+                textDetMaxSideLimit: nil,
+                textDetThresh: nil,
+                textDetBoxThresh: nil,
+                textDetUnclipRatio: nil,
+                textRecBatchSize: n,
+                textRecScoreThresh: nil
+            )
+        }
+        return .noOverrides
+    }
+
+    /// Physical memory footprint (bytes) — matches Xcode Memory gauge more closely than RSS alone.
+    /// Uses `task_vm_info_data_t.phys_footprint` via `task_info(TASK_VM_INFO, ...)`.
+    private func physicalFootprintBytes() -> UInt64 {
+        var info = task_vm_info_data_t()
+        var count = mach_msg_type_number_t(MemoryLayout<task_vm_info_data_t>.size / MemoryLayout<natural_t>.size)
+        let kerr = withUnsafeMutablePointer(to: &info) { p in
+            p.withMemoryRebound(to: integer_t.self, capacity: Int(count)) {
+                task_info(mach_task_self_, task_flavor_t(TASK_VM_INFO), $0, &count)
+            }
+        }
+        guard kerr == KERN_SUCCESS else { return 0 }
+        return UInt64(info.phys_footprint)
+    }
+
+    private func summarizeMs(_ samples: [Double]) -> TimingSummary {
+        let sorted = samples.sorted()
+        let mean = samples.reduce(0, +) / Double(max(samples.count, 1))
+        let variance =
+            samples.isEmpty
+                ? 0
+                : samples.map { pow($0 - mean, 2) }.reduce(0, +) / Double(samples.count)
+        let stdev = sqrt(variance)
+        let p90Idx = max(0, min(sorted.count - 1, Int(floor(0.9 * Double(max(sorted.count - 1, 0))))))
+        let p90 = sorted.isEmpty ? 0 : sorted[p90Idx]
+        return TimingSummary(mean: mean, stdev: stdev, p90: p90)
+    }
+}
+
+// MARK: - Orientation (match OCRViewModel)
+
+private func normalizeOrientation(_ image: UIImage) -> UIImage {
+    guard image.imageOrientation != .up else { return image }
+    let format = UIGraphicsImageRendererFormat()
+    format.scale = image.scale
+    let renderer = UIGraphicsImageRenderer(size: image.size, format: format)
+    return renderer.image { _ in
+        image.draw(in: CGRect(origin: .zero, size: image.size))
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemoTests/OCRParameterResolverTests.swift b/deploy/ios_demo/PaddleOCRDemoTests/OCRParameterResolverTests.swift
new file mode 100644
index 0000000..42aaf67
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemoTests/OCRParameterResolverTests.swift
@@ -0,0 +1,118 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import XCTest
+@testable import PaddleOCRDemo
+
+/// Verifies ``OCRParameterResolver``
+final class OCRParameterResolverTests: XCTestCase {
+
+    func testDetResizeMergePrefersAppDefaultsOverModelWhenUINil() throws {
+        let dir = FileManager.default.temporaryDirectory
+        let url = dir.appendingPathComponent("merge_test_det_\(UUID().uuidString).yaml")
+        let yaml = """
+        Global:
+          model_name: PP-OCRv5_mobile_det
+        PreProcess:
+          transform_ops:
+          - DetResizeForTest:
+              resize_long: 960
+        PostProcess:
+          name: DBPostProcess
+        """
+        try yaml.write(to: url, atomically: true, encoding: .utf8)
+        defer { try? FileManager.default.removeItem(at: url) }
+
+        let det = try InferenceConfig.load(from: url.path)
+        let m = det.detResizeFromModel
+        XCTAssertNotNil(m)
+        XCTAssertEqual(m?.limitSideLen, 960)
+
+        let eff = OCRParameterResolver.effective(modelConfig: det, runtime: .noOverrides)
+        let app = OCRParameterResolver.textDetLimitSideLenAppDefault
+        XCTAssertEqual(eff.detResize.limitSideLen, app)
+        XCTAssertEqual(eff.mergedLimitSideLen, app)
+    }
+
+    func testEffectiveDetPostprocessMergesThresholdsOnly() throws {
+        let dir = FileManager.default.temporaryDirectory
+        let url = dir.appendingPathComponent("merge_test_post_\(UUID().uuidString).yaml")
+        let yaml = """
+        Global:
+          model_name: PP-OCRv5_mobile_det
+        PreProcess:
+          transform_ops:
+          - DetResizeForTest:
+              resize_long: 960
+        PostProcess:
+          name: DBPostProcess
+          thresh: 0.99
+          box_thresh: 0.99
+          unclip_ratio: 9.9
+          max_candidates: 7
+        """
+        try yaml.write(to: url, atomically: true, encoding: .utf8)
+        defer { try? FileManager.default.removeItem(at: url) }
+
+        let det = try InferenceConfig.load(from: url.path)
+        let post = OCRParameterResolver.effective(modelConfig: det, runtime: .noOverrides)
+        XCTAssertEqual(post.detThresh, 0.3, accuracy: 0.0001)
+        XCTAssertEqual(post.detBoxThresh, 0.6, accuracy: 0.0001)
+        XCTAssertEqual(post.detUnclipRatio, 1.5, accuracy: 0.0001)
+        XCTAssertEqual(det.postProcess.maxCandidates, 7)
+    }
+
+    func testDecodeImageImgModeParsedAsRGB() throws {
+        let dir = FileManager.default.temporaryDirectory
+        let url = dir.appendingPathComponent("decode_rgb_\(UUID().uuidString).yaml")
+        let yaml = """
+        Global:
+          model_name: PP-OCRv5_mobile_det
+        PreProcess:
+          transform_ops:
+          - DecodeImage:
+              channel_first: false
+              img_mode: RGB
+          - DetResizeForTest:
+              resize_long: 960
+        PostProcess:
+          name: DBPostProcess
+        """
+        try yaml.write(to: url, atomically: true, encoding: .utf8)
+        defer { try? FileManager.default.removeItem(at: url) }
+
+        let cfg = try InferenceConfig.load(from: url.path)
+        XCTAssertEqual(cfg.decodeImageChannelOrder, .rgb)
+    }
+
+    func testDecodeImageMissingDefaultsToBGR() throws {
+        let dir = FileManager.default.temporaryDirectory
+        let url = dir.appendingPathComponent("decode_default_bgr_\(UUID().uuidString).yaml")
+        let yaml = """
+        Global:
+          model_name: PP-OCRv5_mobile_det
+        PreProcess:
+          transform_ops:
+          - DetResizeForTest:
+              resize_long: 960
+        PostProcess:
+          name: DBPostProcess
+        """
+        try yaml.write(to: url, atomically: true, encoding: .utf8)
+        defer { try? FileManager.default.removeItem(at: url) }
+
+        let cfg = try InferenceConfig.load(from: url.path)
+        XCTAssertEqual(cfg.decodeImageChannelOrder, .bgr)
+    }
+}
diff --git a/deploy/ios_demo/PaddleOCRDemoTests/ORTSessionManagerTests.swift b/deploy/ios_demo/PaddleOCRDemoTests/ORTSessionManagerTests.swift
new file mode 100644
index 0000000..c1c854f
--- /dev/null
+++ b/deploy/ios_demo/PaddleOCRDemoTests/ORTSessionManagerTests.swift
@@ -0,0 +1,55 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import XCTest
+@testable import PaddleOCRDemo
+
+final class ORTSessionManagerTests: XCTestCase {
+
+    func testLoadModels() async throws {
+        let manager = ORTSessionManager()
+        try await manager.loadModels(executionProvider: .coreML)
+    }
+
+    func testLoadModelsCPU() async throws {
+        let manager = ORTSessionManager()
+        try await manager.loadModels(executionProvider: .cpu)
+    }
+
+    func testModelConfigResolution() throws {
+        do {
+            let detConfig = try ModelConfig.detection()
+            XCTAssertFalse(detConfig.modelPath.isEmpty, "Det model path should not be empty")
+            let expected = try InferenceConfig.load(from: detConfig.configPath).modelName
+            XCTAssertEqual(detConfig.name, expected, "Display name should equal Global.model_name in the det model config")
+        } catch let error as ModelConfigError {
+            switch error {
+            case .modelNotFound(let path):
+                XCTAssertTrue(path.contains("det"), "Error should reference det model path")
+            }
+        }
+
+        do {
+            let recConfig = try ModelConfig.recognition()
+            XCTAssertFalse(recConfig.modelPath.isEmpty, "Rec model path should not be empty")
+            let expected = try InferenceConfig.load(from: recConfig.configPath).modelName
+            XCTAssertEqual(recConfig.name, expected, "Display name should equal Global.model_name in the rec model config")
+        } catch let error as ModelConfigError {
+            switch error {
+            case .modelNotFound(let path):
+                XCTAssertTrue(path.contains("rec"), "Error should reference rec model path")
+            }
+        }
+    }
+}
diff --git a/deploy/ios_demo/Podfile b/deploy/ios_demo/Podfile
new file mode 100644
index 0000000..ee0a4ce
--- /dev/null
+++ b/deploy/ios_demo/Podfile
@@ -0,0 +1,30 @@
+platform :ios, '16.0'
+
+target 'PaddleOCRDemo' do
+  use_frameworks!
+
+  # ONNX Runtime with ObjC API — includes Core ML EP + XNNPACK EP pre-built
+  pod 'onnxruntime-objc', '~> 1.24'
+
+  # YAML parsing for model config files
+  pod 'Yams', '~> 5.0'
+
+  # OpenCV (imgproc): preprocessing resize/pad, recognition resize, quad crop, DB contours — see OpenCV*Bridge.
+  pod 'OpenCV', '~> 4.3.0'
+
+  # Unit tests `@testable import PaddleOCRDemo` need the same Swift / header search paths
+  # as the app without duplicating pod linkage.
+  target 'PaddleOCRDemoTests' do
+    inherit! :search_paths
+  end
+end
+
+# onnxruntime-objc ships headers that use #include "..." inside framework headers; newer Xcode
+# treats that as an error (CLANG_WARN_QUOTED_INCLUDE_IN_FRAMEWORK_HEADER). Allow Pods to build.
+post_install do |installer|
+  installer.pods_project.targets.each do |target|
+    target.build_configurations.each do |config|
+      config.build_settings['CLANG_WARN_QUOTED_INCLUDE_IN_FRAMEWORK_HEADER'] = 'NO'
+    end
+  end
+end
diff --git a/deploy/ios_demo/Podfile.lock b/deploy/ios_demo/Podfile.lock
new file mode 100644
index 0000000..5828f34
--- /dev/null
+++ b/deploy/ios_demo/Podfile.lock
@@ -0,0 +1,30 @@
+PODS:
+  - onnxruntime-c (1.24.3)
+  - onnxruntime-objc (1.24.3):
+    - onnxruntime-objc/Core (= 1.24.3)
+  - onnxruntime-objc/Core (1.24.3):
+    - onnxruntime-c (= 1.24.3)
+  - OpenCV (4.3.0)
+  - Yams (5.0.6)
+
+DEPENDENCIES:
+  - onnxruntime-objc (~> 1.24)
+  - OpenCV (~> 4.3.0)
+  - Yams (~> 5.0)
+
+SPEC REPOS:
+  trunk:
+    - onnxruntime-c
+    - onnxruntime-objc
+    - OpenCV
+    - Yams
+
+SPEC CHECKSUMS:
+  onnxruntime-c: 4a1a1a81da2087869dc9b6357f182952db2df29c
+  onnxruntime-objc: f9310465f0a284682811d071b7eb8e93f10517b8
+  OpenCV: 681e0451c1e54c5930a8922277f817ed5d5ab459
+  Yams: e10dae147f517ed57ecae37c5e8681bdf8fcab65
+
+PODFILE CHECKSUM: 24111b5e94db1b6a72362b62148f587abc1a06be
+
+COCOAPODS: 1.16.2
diff --git a/deploy/ios_demo/README.md b/deploy/ios_demo/README.md
new file mode 100644
index 0000000..eba5448
--- /dev/null
+++ b/deploy/ios_demo/README.md
@@ -0,0 +1,5 @@
+# PaddleOCR iOS Demo
+
+中文 | [English](./README_en.md)
+
+请查看 [文档](../../docs/version3.x/inference_deployment/cross_platform/ios_deployment.md)。
diff --git a/deploy/ios_demo/README_en.md b/deploy/ios_demo/README_en.md
new file mode 100644
index 0000000..99640bc
--- /dev/null
+++ b/deploy/ios_demo/README_en.md
@@ -0,0 +1,5 @@
+# PaddleOCR iOS Demo
+
+[中文](./README.md) | English
+
+Please refer to the [documentation](../../docs/version3.x/inference_deployment/cross_platform/ios_deployment.en.md).
diff --git a/deploy/ios_demo/requirements-accuracy.txt b/deploy/ios_demo/requirements-accuracy.txt
new file mode 100644
index 0000000..c94a0f4
--- /dev/null
+++ b/deploy/ios_demo/requirements-accuracy.txt
@@ -0,0 +1,2 @@
+PyYAML>=6.0
+shapely>=2.0
diff --git a/deploy/ios_demo/requirements-build-calib.txt b/deploy/ios_demo/requirements-build-calib.txt
new file mode 100644
index 0000000..a8ee356
--- /dev/null
+++ b/deploy/ios_demo/requirements-build-calib.txt
@@ -0,0 +1,4 @@
+numpy
+paddlex>=3.5,<3.6
+onnxruntime>=1.18
+opencv-contrib-python>=4.5
diff --git a/deploy/ios_demo/requirements-onnx-convert-ort.txt b/deploy/ios_demo/requirements-onnx-convert-ort.txt
new file mode 100644
index 0000000..7dcc071
--- /dev/null
+++ b/deploy/ios_demo/requirements-onnx-convert-ort.txt
@@ -0,0 +1,3 @@
+numpy>=1.20
+onnx>=1.16,<1.21
+onnxruntime>=1.18
diff --git a/deploy/ios_demo/requirements-onnx-quantize.txt b/deploy/ios_demo/requirements-onnx-quantize.txt
new file mode 100644
index 0000000..db6ca76
--- /dev/null
+++ b/deploy/ios_demo/requirements-onnx-quantize.txt
@@ -0,0 +1,5 @@
+# Keep onnx and onnxruntime within a range that works together (avoids IR / schema skew on older CPU wheels).
+# For best parity with the app, install a onnxruntime major version close to onnxruntime-objc in the Podfile.
+numpy>=1.20
+onnx>=1.16,<1.21
+onnxruntime>=1.18
diff --git a/deploy/ios_demo/scripts/build_onnx_calib_npy.py b/deploy/ios_demo/scripts/build_onnx_calib_npy.py
new file mode 100755
index 0000000..bececcd
--- /dev/null
+++ b/deploy/ios_demo/scripts/build_onnx_calib_npy.py
@@ -0,0 +1,397 @@
+#!/usr/bin/env python3
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Build float32 NCHW ``.npy`` tensors for static ONNX quantization.
+
+Note:
+    Uses undocumented PaddleX predictor internals (e.g. ``pre_tfs``). May break when
+    upgrading PaddleX.
+
+Examples:
+
+  # Detection model
+  python3 scripts/build_onnx_calib_npy.py --task det \\
+    --model-dir PaddleOCRDemo/Models/det --image-dir /path/to/images --output-dir calib_det
+
+  # Recognition model (one tensor per full image)
+  python3 scripts/build_onnx_calib_npy.py --task rec \\
+    --model-dir PaddleOCRDemo/Models/rec --image-dir /path/to/images --output-dir calib_rec
+
+  # Recognition: full-page images — run det to crop line boxes, one tensor per crop
+  python3 scripts/build_onnx_calib_npy.py --task rec \\
+    --model-dir PaddleOCRDemo/Models/rec --det-model-dir PaddleOCRDemo/Models/det \\
+    --image-dir /path/to/pages --output-dir calib_rec_crops
+"""
+
+from __future__ import annotations
+
+import argparse
+import re
+import sys
+from pathlib import Path
+from typing import Any, Iterator
+
+import cv2
+import numpy as np
+from paddlex import create_predictor
+
+_script_dir = Path(__file__).resolve().parent
+if str(_script_dir) not in sys.path:
+    sys.path.insert(0, str(_script_dir))
+
+from utils import die
+
+
+def _read_model_name(model_dir: Path) -> str:
+    try:
+        import yaml
+    except ImportError as e:
+        die(f"PyYAML is required: {e}")
+    p = model_dir / "inference.yml"
+    if not p.is_file():
+        die(f"missing {p}")
+    with p.open("r", encoding="utf-8") as f:
+        data = yaml.safe_load(f)
+    name = (data or {}).get("Global", {}).get("model_name")
+    if not name:
+        die("Global.model_name not found in inference.yml")
+    return str(name)
+
+
+def _collect_images(image_dir: Path) -> list[Path]:
+    exts = {".png", ".jpg", ".jpeg", ".bmp", ".webp", ".tif", ".tiff"}
+    out: list[Path] = []
+    for p in sorted(image_dir.iterdir()):
+        if p.is_file() and p.suffix.lower() in exts:
+            out.append(p)
+    if not out:
+        die(f"no images found under {image_dir}")
+    return out
+
+
+def _safe_stem(p: Path, idx: int) -> str:
+    s = p.stem
+    s = re.sub(r"[^\w\-.]+", "_", s)[:80]
+    return f"{idx:04d}_{s}" if s else f"{idx:04d}"
+
+
+def _iter_text_boxes(dt_polys: Any) -> Iterator[np.ndarray]:
+    """Turn ``dt_polys`` from a TextDetResult into a sequence of (N,2) point arrays."""
+    if dt_polys is None:
+        return
+    if isinstance(dt_polys, np.ndarray):
+        if dt_polys.size == 0:
+            return
+        if dt_polys.ndim == 2 and dt_polys.shape[0] >= 3:
+            yield np.array(dt_polys, dtype=np.float32, copy=True)
+        elif dt_polys.ndim == 3:
+            for i in range(dt_polys.shape[0]):
+                box = np.array(dt_polys[i], dtype=np.float32, copy=True)
+                if box.shape[0] >= 3:
+                    yield box
+        return
+    for box in dt_polys:
+        b = np.array(box, dtype=np.float32, copy=True)
+        if b.size >= 6 and b.ndim == 2 and b.shape[0] >= 3:
+            yield b
+
+
+def _minarea_rect_rotate_crop(img: np.ndarray, points: np.ndarray) -> np.ndarray:
+    """
+    Min-area rect crop aligned with PP-OCR (see PaddleX CropByPolys, quad).
+    *img* is RGB (same as TextDet Read output).
+    """
+    tmp_box = np.array(points, dtype=np.float32, copy=True)
+    bounding_box = cv2.minAreaRect(tmp_box.astype(np.int32))
+    pts = sorted(list(cv2.boxPoints(bounding_box)), key=lambda x: x[0])
+    if pts[1][1] > pts[0][1]:
+        index_a, index_d = 0, 1
+    else:
+        index_a, index_d = 1, 0
+    if pts[3][1] > pts[2][1]:
+        index_b, index_c = 2, 3
+    else:
+        index_b, index_c = 3, 2
+    box = [pts[index_a], pts[index_b], pts[index_c], pts[index_d]]
+    pts4 = [np.array(p, dtype=np.float32) for p in box]
+    assert len(pts4) == 4, "shape of points must be 4*2"
+    img_crop_width = int(
+        max(
+            np.linalg.norm(pts4[0] - pts4[1]),
+            np.linalg.norm(pts4[2] - pts4[3]),
+        )
+    )
+    img_crop_height = int(
+        max(
+            np.linalg.norm(pts4[0] - pts4[3]),
+            np.linalg.norm(pts4[1] - pts4[2]),
+        )
+    )
+    if img_crop_width < 1 or img_crop_height < 1:
+        raise ValueError("degenerate text crop (zero width/height)")
+    pts_std = np.float32(
+        [
+            [0, 0],
+            [img_crop_width, 0],
+            [img_crop_width, img_crop_height],
+            [0, img_crop_height],
+        ]
+    )
+    pstack = np.stack(pts4, axis=0)
+    m = cv2.getPerspectiveTransform(pstack, pts_std)
+    dst = cv2.warpPerspective(
+        img,
+        m,
+        (img_crop_width, img_crop_height),
+        borderMode=cv2.BORDER_REPLICATE,
+        flags=cv2.INTER_CUBIC,
+    )
+    dh, dw = dst.shape[0:2]
+    if dh * 1.0 / max(dw, 1) >= 1.5:
+        dst = np.rot90(dst)
+    return dst
+
+
+def _tensor_det(pred, image_path: str) -> np.ndarray:
+    """Mirror TextDetRunnerPredictor.process() preprocess, return (1,C,H,W) float32."""
+    batches = list(pred.batch_sampler([image_path]))
+    b = batches[0]
+    batch_raw_imgs = pred.pre_tfs["Read"](imgs=b.instances)
+    batch_imgs, _ = pred.pre_tfs["Resize"](
+        imgs=batch_raw_imgs,
+        limit_side_len=pred.limit_side_len,
+        limit_type=pred.limit_type,
+        max_side_limit=pred.max_side_limit,
+    )
+    batch_imgs = pred.pre_tfs["Normalize"](imgs=batch_imgs)
+    batch_imgs = pred.pre_tfs["ToCHW"](imgs=batch_imgs)
+    x = pred.pre_tfs["ToBatch"](imgs=batch_imgs)
+    if not x:
+        die("ToBatch returned empty (det)")
+    return np.ascontiguousarray(x[0], dtype=np.float32)
+
+
+def _tensor_rec(pred, image_path: str) -> np.ndarray:
+    from paddlex.inference.models.text_recognition.processors import (
+        validate_text_rec_image_array,
+    )
+
+    batches = list(pred.batch_sampler([image_path]))
+    b = batches[0]
+    batch_raw_imgs = pred.pre_tfs["Read"](imgs=b.instances)
+    for i, img in enumerate(batch_raw_imgs):
+        validate_text_rec_image_array(img, index=i)
+    batch_imgs = pred.pre_tfs["ReisizeNorm"](imgs=batch_raw_imgs)
+    x = pred.pre_tfs["ToBatch"](imgs=batch_imgs)
+    if not x:
+        die("ToBatch returned empty (rec)")
+    return np.ascontiguousarray(x[0], dtype=np.float32)
+
+
+def _tensor_rec_from_rgb_ndarray(rec_pred, img_rgb: np.ndarray) -> np.ndarray:
+    """Same preprocessing as *path* input, for an RGB array (e.g. det crop on full page)."""
+    from paddlex.inference.models.text_recognition.processors import (
+        validate_text_rec_image_array,
+    )
+
+    bgr = cv2.cvtColor(img_rgb, cv2.COLOR_RGB2BGR)
+    batch_raw_imgs = rec_pred.pre_tfs["Read"](imgs=[bgr])
+    for i, img in enumerate(batch_raw_imgs):
+        validate_text_rec_image_array(img, index=i)
+    batch_imgs = rec_pred.pre_tfs["ReisizeNorm"](imgs=batch_raw_imgs)
+    x = rec_pred.pre_tfs["ToBatch"](imgs=batch_imgs)
+    if not x:
+        die("ToBatch returned empty (rec, ndarray)")
+    return np.ascontiguousarray(x[0], dtype=np.float32)
+
+
+def main() -> None:
+    ap = argparse.ArgumentParser(
+        description="Build float32 NCHW .npy files for static ONNX quantize."
+    )
+    ap.add_argument(
+        "--image-dir",
+        type=Path,
+        required=True,
+        help="Directory of input images.",
+    )
+    ap.add_argument(
+        "--output-dir",
+        type=Path,
+        required=True,
+        help="Output directory for .npy files (created if missing).",
+    )
+    ap.add_argument(
+        "--model-dir",
+        type=Path,
+        required=True,
+        help="Model directory.",
+    )
+    ap.add_argument(
+        "--task",
+        choices=("det", "rec"),
+        required=True,
+        help="Model role: det = text detection, rec = text recognition.",
+    )
+    ap.add_argument(
+        "--det-model-dir",
+        type=Path,
+        default=None,
+        help=(
+            "For --task rec only: text-detection model directory. "
+            "When set, each full image is run through det; each text box is cropped and one "
+            "rec calibration tensor is written per crop (instead of one tensor per file)."
+        ),
+    )
+    ap.add_argument(
+        "--max-crops-per-image",
+        type=int,
+        default=0,
+        help="With --det-model-dir, cap how many box crops to use per input image (0 = no cap).",
+    )
+    ap.add_argument("--device", default="cpu", help="Device string, e.g. cpu or gpu:0.")
+    args = ap.parse_args()
+
+    if not args.image_dir.is_dir():
+        die(f"not a directory: {args.image_dir}")
+    if not (args.model_dir / "inference.onnx").is_file():
+        die(f"missing {args.model_dir / 'inference.onnx'}")
+
+    if args.det_model_dir is not None:
+        if args.task != "rec":
+            die("--det-model-dir is only valid with --task rec")
+        if not (args.det_model_dir / "inference.onnx").is_file():
+            die(f"missing {args.det_model_dir / 'inference.onnx'}")
+
+    model_name = _read_model_name(args.model_dir)
+    if args.task == "det" and "_rec" in model_name and "_det" not in model_name:
+        print(
+            "warning: --task det but model_name looks like recognition; check --model-dir",
+            file=sys.stderr,
+        )
+    if args.task == "rec" and "_det" in model_name and "_rec" not in model_name:
+        print(
+            "warning: --task rec but model_name looks like detection; check --model-dir",
+            file=sys.stderr,
+        )
+
+    try:
+        pred = create_predictor(
+            model_name,
+            model_dir=str(args.model_dir.resolve()),
+            device=args.device,
+            engine="onnxruntime",
+            batch_size=1,
+        )
+    except Exception as e:
+        die(f"create_predictor failed: {e}")
+
+    det_pred = None
+    det_name = None
+    if args.det_model_dir is not None:
+        det_name = _read_model_name(args.det_model_dir)
+        if "_rec" in det_name and "_det" not in det_name:
+            print(
+                "warning: --det-model-dir model_name looks like recognition; check path",
+                file=sys.stderr,
+            )
+        try:
+            det_pred = create_predictor(
+                det_name,
+                model_dir=str(args.det_model_dir.resolve()),
+                device=args.device,
+                engine="onnxruntime",
+                batch_size=1,
+            )
+        except Exception as e:
+            die(f"create_predictor (det) failed: {e}")
+
+    out_dir = args.output_dir
+    out_dir.mkdir(parents=True, exist_ok=True)
+
+    images = _collect_images(args.image_dir)
+    written = 0
+    for idx, path in enumerate(images):
+        pstr = str(path.resolve())
+        if args.task == "det":
+            try:
+                arr = _tensor_det(pred, pstr)
+            except Exception as e:
+                die(f"preprocess failed for {path}: {e}")
+            if arr.dtype != np.float32:
+                arr = arr.astype(np.float32)
+            name = _safe_stem(path, idx) + ".npy"
+            out_path = out_dir / name
+            np.save(str(out_path), arr)
+            print(f"wrote {out_path}  shape={arr.shape}  dtype={arr.dtype}")
+            written += 1
+            continue
+
+        # rec
+        if det_pred is None:
+            try:
+                arr = _tensor_rec(pred, pstr)
+            except Exception as e:
+                die(f"preprocess failed for {path}: {e}")
+            if arr.dtype != np.float32:
+                arr = arr.astype(np.float32)
+            out_path = out_dir / (_safe_stem(path, idx) + ".npy")
+            np.save(str(out_path), arr)
+            print(f"wrote {out_path}  shape={arr.shape}  dtype={arr.dtype}")
+            written += 1
+            continue
+
+        # rec + det: crops from full page
+        try:
+            det_res = next(det_pred(pstr))  # type: ignore[union-attr]
+        except Exception as e:
+            die(f"detection failed for {path}: {e}")
+        polys = list(_iter_text_boxes(det_res["dt_polys"]))
+        if not polys:
+            print(
+                f"warning: no text boxes for {path}; skipping (no .npy written)",
+                file=sys.stderr,
+            )
+            continue
+        if args.max_crops_per_image > 0:
+            polys = polys[: args.max_crops_per_image]
+        full_rgb = det_res["input_img"]
+        if isinstance(full_rgb, (list, tuple)) and len(full_rgb) == 1:
+            full_rgb = full_rgb[0]
+        for box_i, quad in enumerate(polys):
+            try:
+                line_rgb = _minarea_rect_rotate_crop(full_rgb, quad)
+            except Exception as e:
+                print(
+                    f"warning: skip box {box_i} in {path}: {e}",
+                    file=sys.stderr,
+                )
+                continue
+            try:
+                arr = _tensor_rec_from_rgb_ndarray(pred, line_rgb)
+            except Exception as e:
+                die(f"rec preprocess failed for {path} box {box_i}: {e}")
+            if arr.dtype != np.float32:
+                arr = arr.astype(np.float32)
+            out_path = out_dir / (f"{_safe_stem(path, idx)}_box{box_i:03d}.npy")
+            np.save(str(out_path), arr)
+            print(f"wrote {out_path}  shape={arr.shape}  dtype={arr.dtype}")
+            written += 1
+
+    print(f"done. {written} .npy file(s) -> {out_dir}")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/deploy/ios_demo/scripts/compare_ocr_json.py b/deploy/ios_demo/scripts/compare_ocr_json.py
new file mode 100755
index 0000000..9c4ec72
--- /dev/null
+++ b/deploy/ios_demo/scripts/compare_ocr_json.py
@@ -0,0 +1,223 @@
+#!/usr/bin/env python3
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Compare two OCR JSON files (reference vs device) using polygon IoU + CER.
+
+Default thresholds (see argparse defaults) target a stricter CI gate than a loose
+0.5 IoU baseline: tighter box alignment, ~95% char-level headroom on mean CER,
+and limited reference-side misses.
+
+Example:
+  python compare_ocr_json.py ref.json ios.json \\
+    --iou-threshold 0.65 --cer-threshold 0.05 --max-unmatched-ratio 0.1
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import sys
+from pathlib import Path
+from typing import Any, Dict, List, Sequence, Tuple
+
+
+def _levenshtein(a: str, b: str) -> int:
+    if a == b:
+        return 0
+    if not a:
+        return len(b)
+    if not b:
+        return len(a)
+    prev = list(range(len(b) + 1))
+    for i, ca in enumerate(a, 1):
+        cur = [i]
+        for j, cb in enumerate(b, 1):
+            ins = cur[j - 1] + 1
+            delete = prev[j] + 1
+            sub = prev[j - 1] + (ca != cb)
+            cur.append(min(ins, delete, sub))
+        prev = cur
+    return prev[-1]
+
+
+def _cer(ref: str, hyp: str) -> float:
+    if not ref and not hyp:
+        return 0.0
+    if not ref:
+        return 1.0
+    return _levenshtein(ref, hyp) / max(len(ref), 1)
+
+
+def _polygon_iou(
+    poly_a: Sequence[Sequence[float]], poly_b: Sequence[Sequence[float]]
+) -> float:
+    try:
+        from shapely.geometry import Polygon
+    except ImportError as exc:
+        raise RuntimeError(
+            "compare_ocr_json.py requires `pip install shapely`"
+        ) from exc
+
+    def _to_poly(p: Sequence[Sequence[float]]) -> Polygon:
+        pts = [(float(x), float(y)) for x, y in p]
+        if len(pts) < 3:
+            return Polygon()
+        if pts[0] != pts[-1]:
+            pts = pts + [pts[0]]
+        return Polygon(pts)
+
+    p1 = _to_poly(poly_a)
+    p2 = _to_poly(poly_b)
+    if p1.is_empty or p2.is_empty or not p1.is_valid or not p2.is_valid:
+        return 0.0
+    inter = p1.intersection(p2).area
+    union = p1.union(p2).area
+    if union <= 0:
+        return 0.0
+    return float(inter / union)
+
+
+def _load_items(path: Path) -> List[Dict[str, Any]]:
+    with path.open("r", encoding="utf-8") as f:
+        data = json.load(f)
+    items = data.get("items")
+    if not isinstance(items, list):
+        raise ValueError(f"{path}: missing 'items' array")
+    return items
+
+
+def _greedy_match(
+    ref_items: List[Dict[str, Any]],
+    hyp_items: List[Dict[str, Any]],
+    iou_threshold: float,
+) -> Tuple[List[Tuple[int, int, float]], List[int], List[int]]:
+    """Return (pairs as ref_idx, hyp_idx, iou), unmatched_ref, unmatched_hyp."""
+    candidates: List[Tuple[float, int, int]] = []
+    for i, ri in enumerate(ref_items):
+        ra = ri.get("polygon")
+        if not isinstance(ra, list):
+            continue
+        for j, hj in enumerate(hyp_items):
+            ha = hj.get("polygon")
+            if not isinstance(ha, list):
+                continue
+            iou = _polygon_iou(ra, ha)
+            candidates.append((iou, i, j))
+
+    candidates.sort(key=lambda t: t[0], reverse=True)
+    used_r = set()
+    used_h = set()
+    pairs: List[Tuple[int, int, float]] = []
+    for iou, i, j in candidates:
+        if iou < iou_threshold:
+            break
+        if i in used_r or j in used_h:
+            continue
+        used_r.add(i)
+        used_h.add(j)
+        pairs.append((i, j, iou))
+
+    unmatched_r = [i for i in range(len(ref_items)) if i not in used_r]
+    unmatched_h = [j for j in range(len(hyp_items)) if j not in used_h]
+    return pairs, unmatched_r, unmatched_h
+
+
+def main(argv: Sequence[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(description="Compare OCR JSON outputs")
+    parser.add_argument(
+        "reference", type=Path, help="Reference JSON (e.g. paddleocr_reference)"
+    )
+    parser.add_argument("hypothesis", type=Path, help="Device / second JSON")
+    parser.add_argument(
+        "--iou-threshold",
+        type=float,
+        default=0.65,
+        help="Min IoU to pair quadrilateral boxes (stricter than common 0.5 VOC-style cutoff)",
+    )
+    parser.add_argument(
+        "--cer-threshold",
+        type=float,
+        default=0.05,
+        help="Fail if mean CER on matched pairs exceeds this (e.g. 0.05 ≈ 95%% char accuracy headroom)",
+    )
+    parser.add_argument(
+        "--max-unmatched-ratio",
+        type=float,
+        default=0.1,
+        help="Fail if unmatched ref lines / len(ref) exceeds this",
+    )
+    parser.add_argument(
+        "--json-summary-out",
+        type=Path,
+        default=None,
+        help="Write the same JSON as stdout to this path (PASS or FAIL) for generate_benchmark_report.py",
+    )
+    args = parser.parse_args(list(argv) if argv is not None else None)
+
+    ref_items = _load_items(args.reference)
+    hyp_items = _load_items(args.hypothesis)
+
+    pairs, unmatched_r, unmatched_h = _greedy_match(
+        ref_items, hyp_items, args.iou_threshold
+    )
+
+    cers: List[float] = []
+    for ri, hj, _ in pairs:
+        rt = str(ref_items[ri].get("text", ""))
+        ht = str(hyp_items[hj].get("text", ""))
+        cers.append(_cer(rt, ht))
+
+    mean_cer = sum(cers) / len(cers) if cers else 0.0
+    nref = max(len(ref_items), 1)
+    unmatched_ratio = len(unmatched_r) / nref
+
+    report = {
+        "matched_pairs": len(pairs),
+        "reference_count": len(ref_items),
+        "hypothesis_count": len(hyp_items),
+        "unmatched_reference": len(unmatched_r),
+        "unmatched_hypothesis": len(unmatched_h),
+        "unmatched_reference_ratio": unmatched_ratio,
+        "mean_cer_matched": mean_cer,
+        "iou_threshold": args.iou_threshold,
+        "cer_threshold": args.cer_threshold,
+        "max_unmatched_ratio": args.max_unmatched_ratio,
+    }
+    failed = False
+    if mean_cer > args.cer_threshold:
+        print(f"FAIL: mean CER {mean_cer:.4f} > {args.cer_threshold}", file=sys.stderr)
+        failed = True
+    if unmatched_ratio > args.max_unmatched_ratio:
+        print(
+            f"FAIL: unmatched ref ratio {unmatched_ratio:.4f} > {args.max_unmatched_ratio}",
+            file=sys.stderr,
+        )
+        failed = True
+
+    report["pass"] = not failed
+    out_txt = json.dumps(report, indent=2, ensure_ascii=False)
+    print(out_txt)
+
+    if args.json_summary_out is not None:
+        args.json_summary_out.parent.mkdir(parents=True, exist_ok=True)
+        args.json_summary_out.write_text(out_txt + "\n", encoding="utf-8")
+
+    if not failed:
+        print("PASS", file=sys.stderr)
+    return 1 if failed else 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/deploy/ios_demo/scripts/convert_onnx_to_ort.sh b/deploy/ios_demo/scripts/convert_onnx_to_ort.sh
new file mode 100755
index 0000000..e3c70ab
--- /dev/null
+++ b/deploy/ios_demo/scripts/convert_onnx_to_ort.sh
@@ -0,0 +1,184 @@
+#!/usr/bin/env bash
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Convert ONNX models to ORT format for ONNX Runtime.
+# See: https://onnxruntime.ai/docs/performance/model-optimizations/ort-format-models.html
+#
+# Requires: python3
+#
+# Usage:
+#   ./scripts/convert_onnx_to_ort.sh
+#   ./scripts/convert_onnx_to_ort.sh --output-dir ./out/ort_bundles
+#   ./scripts/convert_onnx_to_ort.sh --input-dir /path/to/models --output-dir /tmp/ort
+#   ./scripts/convert_onnx_to_ort.sh --optimization-style Runtime
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+IOS_DEMO_ROOT="$(cd "${SCRIPT_DIR}/.." && pwd)"
+DEFAULT_INPUT_DIR="${IOS_DEMO_ROOT}/PaddleOCRDemo/Models"
+
+INPUT_DIR="${DEFAULT_INPUT_DIR}"
+ORT_OUTPUT_DIR=""
+KEEP_OPERATOR_CONFIG=false
+EXPLICIT_OPTIMIZATION_STYLE=false
+OPTIMIZATION_STYLE_VALUES=()
+
+usage() {
+  cat <<EOF
+Usage: ./scripts/convert_onnx_to_ort.sh [OPTIONS] [ORT_CONVERTER_ARGS...]
+
+Options:
+  --input-dir <dir>     Root directory to scan for *.onnx (recursive). Default: PaddleOCRDemo/Models.
+  --output-dir <dir>    If set, write .ort files under this directory, mirroring relative paths from
+                        the input tree; also copies inference.yml beside each converted bundle (same
+                        layout as under --input-dir) so you can swap in the whole output folder. If
+                        omitted, write next to each .onnx in-place.
+  --keep-operator-config
+                        Keep required_operators*.config files emitted by the ORT converter (for
+                        minimal ORT builds). By default these are removed after a successful run.
+  --optimization-style <style>...
+                        ORT conversion style(s): Fixed and/or Runtime (forwarded as Python
+                        --optimization_style). One or more values after the flag, e.g.
+                        --optimization-style Runtime  or  --optimization-style Fixed Runtime
+                        The flag may appear more than once; values accumulate.
+                        Alias: --optimization_style
+                        Default when omitted: Fixed (single .ort per ONNX; upstream defaults to both).
+  -h, --help            Show this help
+
+ORT_CONVERTER_ARGS are forwarded to:
+  python3 -m onnxruntime.tools.convert_onnx_models_to_ort
+(documentation: https://onnxruntime.ai/ — place this script’s options first, e.g. --output-dir
+before other converter flags such as --enable_type_reduction.)
+
+Examples:
+  ./scripts/convert_onnx_to_ort.sh
+  ./scripts/convert_onnx_to_ort.sh --output-dir ./out/ort_only
+  ./scripts/convert_onnx_to_ort.sh --input-dir ./PaddleOCRDemo/Models --output-dir ./out/ort_bundles
+  ./scripts/convert_onnx_to_ort.sh --output-dir /tmp/ort --optimization-style Runtime
+  ./scripts/convert_onnx_to_ort.sh --optimization-style Fixed Runtime --output-dir ./out/both
+  ./scripts/convert_onnx_to_ort.sh --output-dir ./out/ort_only --keep-operator-config
+EOF
+}
+
+die() {
+  echo "error: $*" >&2
+  exit 1
+}
+
+# Remove ORT converter operator-config files.
+prune_operator_config_files() {
+  local root="$1"
+  local n=0
+  while IFS= read -r -d '' f; do
+    rm -f "${f}"
+    n=$((n + 1))
+  done < <(find "${root}" -name 'required_operators*.config' -type f -print0 2>/dev/null || true)
+  if [[ "$n" -gt 0 ]]; then
+    echo "Removed ${n} required_operators*.config file(s) (minimal-build manifests; use --keep-operator-config to retain)."
+  fi
+}
+
+copy_inference_yml_mirrored() {
+  local src_root="$1"
+  local dst_root="$2"
+  local yml
+  local n=0
+  while IFS= read -r -d '' yml; do
+    local rel="${yml#"${src_root}/"}"
+    local dest="${dst_root}/${rel}"
+    mkdir -p "$(dirname "${dest}")"
+    cp -f "${yml}" "${dest}"
+    n=$((n + 1))
+  done < <(find "${src_root}" -name 'inference.yml' -type f -print0)
+  echo "Copied ${n} inference.yml into --output-dir (same relative paths as under --input-dir)."
+}
+
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    --input-dir)
+      [[ -n "${2:-}" ]] || die "--input-dir requires a directory"
+      INPUT_DIR="$2"
+      shift 2
+      ;;
+    --output-dir)
+      [[ -n "${2:-}" ]] || die "--output-dir requires a directory"
+      ORT_OUTPUT_DIR="$2"
+      shift 2
+      ;;
+    --keep-operator-config)
+      KEEP_OPERATOR_CONFIG=true
+      shift
+      ;;
+    --optimization-style|--optimization_style)
+      shift
+      EXPLICIT_OPTIMIZATION_STYLE=true
+      chunk=()
+      while [[ $# -gt 0 ]]; do
+        case "$1" in
+          Fixed|Runtime)
+            chunk+=("$1")
+            shift
+            ;;
+          *)
+            break
+            ;;
+        esac
+      done
+      [[ ${#chunk[@]} -gt 0 ]] || die "--optimization-style requires one or more of: Fixed Runtime"
+      OPTIMIZATION_STYLE_VALUES+=("${chunk[@]}")
+      ;;
+    -h | --help)
+      usage
+      exit 0
+      ;;
+    *)
+      break
+      ;;
+  esac
+done
+
+CONVERTER_EXTRA=("$@")
+
+if [[ "$EXPLICIT_OPTIMIZATION_STYLE" == false ]]; then
+  OPTIMIZATION_STYLE_VALUES=(Fixed)
+fi
+
+ORT_OPTIMIZATION_ARGS=(--optimization_style "${OPTIMIZATION_STYLE_VALUES[@]}")
+
+command -v python3 >/dev/null || die "python3 is required"
+[[ -d "${INPUT_DIR}" ]] || die "input directory not found: ${INPUT_DIR}"
+INPUT_ABS="$(cd "${INPUT_DIR}" && pwd)"
+
+if [[ -n "${ORT_OUTPUT_DIR}" ]]; then
+  mkdir -p "${ORT_OUTPUT_DIR}" || die "cannot create --output-dir: ${ORT_OUTPUT_DIR}"
+  ORT_OUT_ABS="$(cd "${ORT_OUTPUT_DIR}" && pwd)"
+  echo "Converting .onnx -> .ort: --input-dir ${INPUT_ABS} -> --output-dir ${ORT_OUT_ABS} ..."
+  python3 -m onnxruntime.tools.convert_onnx_models_to_ort \
+    --output_dir "${ORT_OUT_ABS}" \
+    "${ORT_OPTIMIZATION_ARGS[@]}" \
+    "${CONVERTER_EXTRA[@]}" \
+    -- \
+    "${INPUT_ABS}"
+  copy_inference_yml_mirrored "${INPUT_ABS}" "${ORT_OUT_ABS}"
+  if [[ "$KEEP_OPERATOR_CONFIG" == false ]]; then
+    prune_operator_config_files "${ORT_OUT_ABS}"
+  fi
+else
+  echo "Converting .onnx -> .ort in-place under ${INPUT_ABS} ..."
+  python3 -m onnxruntime.tools.convert_onnx_models_to_ort \
+    "${ORT_OPTIMIZATION_ARGS[@]}" \
+    "${CONVERTER_EXTRA[@]}" \
+    -- \
+    "${INPUT_ABS}"
+  if [[ "$KEEP_OPERATOR_CONFIG" == false ]]; then
+    prune_operator_config_files "${INPUT_ABS}"
+  fi
+fi
+echo "Done."
diff --git a/deploy/ios_demo/scripts/debug_onnx_qdq.py b/deploy/ios_demo/scripts/debug_onnx_qdq.py
new file mode 100755
index 0000000..a1c3469
--- /dev/null
+++ b/deploy/ios_demo/scripts/debug_onnx_qdq.py
@@ -0,0 +1,321 @@
+#!/usr/bin/env python3
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""QDQ quantization debug: align with ONNX Runtime ``qdq_loss_debug`` (official example flow).
+
+This mirrors the *Debugging* section of:
+https://github.com/microsoft/onnxruntime-inference-examples/blob/main/quantization/image_classification/cpu/ReadMe.md
+
+Workflow:
+1. Augment float32 and QDQ models to expose intermediate tensors
+   (``modify_model_output_intermediate_tensors``).
+2. Run both with the same calibration ``.npy`` inputs
+   (``collect_activations``), matching ``quantize_onnx_model`` / ``build_onnx_calib_npy`` data.
+3. ``create_activation_matching`` + SQNR; ``create_weight_matching`` + per-weight SQNR.
+4. Print a report and optionally write JSON (scalar metrics only).
+
+*float* should be the **same** float graph you quantized (ideally the ORT
+``quant_pre_process`` output if you used ``--ort-preprocess`` in ``quantize_onnx_model.py``),
+or the graph is unlikely to line-tensor match the QDQ model.
+
+Example:
+  python3 scripts/debug_onnx_qdq.py \\
+    --float-model /path/to/float_infer.onnx \\
+    --qdq-model /path/to/inference_int8.onnx \\
+    --calib-data-dir /path/to/calib_npy \\
+    --max-samples 3 \\
+    --json-report qdq_debug.json
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import logging
+import sys
+import tempfile
+from pathlib import Path
+from typing import Any
+
+_script_dir = Path(__file__).resolve().parent
+if str(_script_dir) not in sys.path:
+    sys.path.insert(0, str(_script_dir))
+
+from utils import build_npy_dir_reader, die
+
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
+
+
+def _parse_op_types(s: str | None) -> list[str] | None:
+    if not s or not s.strip():
+        return None
+    return [t.strip() for t in s.split(",") if t.strip()]
+
+
+def _safe_compute_weight_error(
+    wmatch: dict[str, dict[str, Any]],
+    err_func: Any,
+) -> dict[str, float]:
+    """Like ORT ``compute_weight_error``, but coerce scalars to 0-d arrays.
+
+    ``create_weight_matching`` can set ``dequantized`` to a bare ``numpy.float32``; ORT
+    ``compute_signal_to_quantization_noice_ratio`` then does ``len(ylist)`` on that scalar
+    and raises ``TypeError``.
+    """
+    import numpy as np
+
+    out: dict[str, float] = {}
+    for name, m in wmatch.items():
+        try:
+            xf = np.asarray(m["float"])
+            yd = np.asarray(m["dequantized"])
+        except (KeyError, TypeError, ValueError) as e:
+            logger.warning("weight SQNR skip %r: %s", name, e)
+            continue
+        if xf.size == 0 and yd.size == 0:
+            continue
+        if xf.size != yd.size:
+            logger.warning(
+                "weight SQNR skip %r: size mismatch %s vs %s",
+                name,
+                getattr(xf, "shape", xf),
+                getattr(yd, "shape", yd),
+            )
+            continue
+        try:
+            out[name] = float(err_func(xf, yd))
+        except (TypeError, ValueError, RuntimeError) as e:
+            logger.warning("weight SQNR skip %r: %s", name, e)
+    return out
+
+
+def _safe_compute_activation_error(
+    match: dict[str, dict[str, Any]],
+    err_func: Any,
+) -> dict[str, dict[str, float]]:
+    """Like ORT ``compute_activation_error``, but skip missing *float* entries (avoids KeyError)."""
+    result: dict[str, dict[str, float]] = {}
+    for name, m in match.items():
+        e: dict[str, float] = {
+            "qdq_err": float(err_func(m["pre_qdq"], m["post_qdq"])),
+        }
+        if m.get("float"):
+            e["xmodel_err"] = float(err_func(m["float"], m["post_qdq"]))
+        result[name] = e
+    return result
+
+
+def _sqnr_key(entry: dict[str, float], prefer_xmodel: bool) -> float:
+    if prefer_xmodel and "xmodel_err" in entry:
+        return entry["xmodel_err"]
+    return entry.get("qdq_err", float("nan"))
+
+
+def _run(
+    float_model: Path,
+    qdq_model: Path,
+    calib: Path,
+    max_samples: int,
+    op_types: list[str] | None,
+    json_report: Path | None,
+    top_n: int,
+    skip_weights: bool,
+    skip_activations: bool,
+) -> None:
+    from onnxruntime.quantization.qdq_loss_debug import (
+        collect_activations,
+        compute_signal_to_quantization_noice_ratio,
+        create_activation_matching,
+        create_weight_matching,
+        modify_model_output_intermediate_tensors,
+    )
+
+    if not float_model.is_file():
+        die(f"not a file: {float_model}")
+    if not qdq_model.is_file():
+        die(f"not a file: {qdq_model}")
+    if not skip_activations:
+        if not calib.is_dir():
+            die(f"not a directory: {calib}")
+
+    out_weights: dict[str, float] = {}
+    out_act: dict[str, dict[str, float]] = {}
+
+    if not skip_weights:
+        logger.info("weight matching (no inference run)…")
+        try:
+            wmatch = create_weight_matching(str(float_model), str(qdq_model))
+        except Exception as e:
+            die(f"create_weight_matching failed: {e}")
+        out_weights = _safe_compute_weight_error(
+            wmatch, compute_signal_to_quantization_noice_ratio
+        )
+        print("\n--- Weight SQNR (dB, higher is better) ---\n")
+        for name in sorted(out_weights, key=lambda x: out_weights[x]):
+            print(f"  {name}: {out_weights[name]:.4f}")
+
+    if not skip_activations:
+        with tempfile.TemporaryDirectory(prefix="ort.qdq.debug.") as td:
+            tmp = Path(td)
+            aug_f = tmp / "augmented_float.onnx"
+            aug_q = tmp / "augmented_qdq.onnx"
+            op_arg = op_types if op_types is not None else []
+            logger.info("augmenting float model → %s", aug_f)
+            try:
+                modify_model_output_intermediate_tensors(
+                    str(float_model), str(aug_f), op_types_for_saving=op_arg
+                )
+            except Exception as e:
+                die(f"modify_model_output_intermediate_tensors (float) failed: {e}")
+            logger.info("augmenting QDQ model → %s", aug_q)
+            try:
+                modify_model_output_intermediate_tensors(
+                    str(qdq_model), str(aug_q), op_types_for_saving=op_arg
+                )
+            except Exception as e:
+                die(f"modify_model_output_intermediate_tensors (qdq) failed: {e}")
+
+            r_float = build_npy_dir_reader(float_model, calib, max_samples=max_samples)
+            r_qdq = build_npy_dir_reader(qdq_model, calib, max_samples=max_samples)
+            try:
+                logger.info("collect_activations (float)…")
+                act_f = collect_activations(str(aug_f), r_float)
+                logger.info("collect_activations (QDQ)…")
+                act_q = collect_activations(str(aug_q), r_qdq)
+            except Exception as e:
+                die(f"collect_activations failed: {e}")
+
+        match = create_activation_matching(act_q, act_f)
+        if not match:
+            print(
+                "\nwarning: empty activation match — is the quantized model QDQ (not QOperator)?",
+                file=sys.stderr,
+            )
+        out_act = _safe_compute_activation_error(
+            match, compute_signal_to_quantization_noice_ratio
+        )
+        has_xm = any("xmodel_err" in v for v in out_act.values())
+        scored = sorted(
+            out_act.items(),
+            key=lambda kv: _sqnr_key(kv[1], prefer_xmodel=has_xm),
+        )[: top_n if top_n > 0 else len(out_act)]
+        which = "xmodel_err (float vs post_qdq)" if has_xm else "qdq_err (q dq pair)"
+        print(
+            f"\n--- Top activation SQNR (dB, {which}; lower = worse) — showing "
+            f"{len(scored)} of {len(out_act)} —\n"
+        )
+        for name, m in scored:
+            if has_xm and "xmodel_err" in m:
+                print(
+                    f"  {name}: pre/post_qdq {m['qdq_err']:.4f}  "
+                    f"float vs post_qdq {m['xmodel_err']:.4f}"
+                )
+            else:
+                print(f"  {name}: pre/post_qdq {m['qdq_err']:.4f}")
+
+    if json_report is not None:
+        payload: dict[str, Any] = {
+            "weight_sqnr_db": out_weights,
+            "activation_sqnr_db": out_act,
+        }
+        json_report = json_report.resolve()
+        json_report.parent.mkdir(parents=True, exist_ok=True)
+        with json_report.open("w", encoding="utf-8") as f:
+            json.dump(payload, f, indent=2, sort_keys=True)
+        print(f"\nwrote {json_report}")
+
+
+def main() -> None:
+    ap = argparse.ArgumentParser(
+        description="ORT-style QDQ debug: float vs QDQ activations/weights (same .npy calib as quantize).",
+    )
+    ap.add_argument(
+        "--float-model",
+        type=Path,
+        required=True,
+        help="Float32 ONNX (same topology as the graph you quantized; use ORT preprocessed model if you used it before quantize).",
+    )
+    ap.add_argument(
+        "--qdq-model",
+        type=Path,
+        required=True,
+        help="Statically quantized (QDQ) ONNX model, e.g. from quantize_onnx_model.py.",
+    )
+    ap.add_argument(
+        "--calib-data-dir",
+        type=Path,
+        default=None,
+        help="Directory of float32 .npy used for calibration (required unless --skip-activations).",
+    )
+    ap.add_argument(
+        "--max-samples",
+        type=int,
+        default=5,
+        help="Max number of .npy files to use (default 5; full runs can be slow and memory-heavy).",
+    )
+    ap.add_argument(
+        "--op-types",
+        type=str,
+        default="",
+        help="Comma-separated op types to save intermediates for (e.g. Conv,Add). Empty = all float tensors (very slow and large).",
+    )
+    ap.add_argument(
+        "--json-report",
+        type=Path,
+        default=None,
+        help="Write metrics-only JSON to this path.",
+    )
+    ap.add_argument(
+        "--top",
+        type=int,
+        default=40,
+        help="How many activations to print (sorted worst-first by SQNR). 0 = print all.",
+    )
+    ap.add_argument(
+        "--skip-weights",
+        action="store_true",
+        help="Only run activation path (faster if you only care about activations).",
+    )
+    ap.add_argument(
+        "--skip-activations",
+        action="store_true",
+        help="Only run weight matching (no collect_activations; ignores calib).",
+    )
+    args = ap.parse_args()
+
+    if args.skip_weights and args.skip_activations:
+        die("cannot set both --skip-weights and --skip-activations")
+    if not args.skip_activations:
+        if args.calib_data_dir is None:
+            die("activation path requires --calib-data-dir")
+        if args.max_samples < 1:
+            die("--max-samples must be >= 1 for activation path")
+
+    _run(
+        float_model=args.float_model,
+        qdq_model=args.qdq_model,
+        calib=args.calib_data_dir or Path(),
+        max_samples=args.max_samples,
+        op_types=_parse_op_types(args.op_types),
+        json_report=args.json_report,
+        top_n=args.top,
+        skip_weights=args.skip_weights,
+        skip_activations=args.skip_activations,
+    )
+
+
+if __name__ == "__main__":
+    main()
diff --git a/deploy/ios_demo/scripts/extract_xcresult_attachments.py b/deploy/ios_demo/scripts/extract_xcresult_attachments.py
new file mode 100755
index 0000000..ad8c7e9
--- /dev/null
+++ b/deploy/ios_demo/scripts/extract_xcresult_attachments.py
@@ -0,0 +1,380 @@
+#!/usr/bin/env python3
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Extract named XCTAttachments from an .xcresult bundle into files.
+
+Uses `xcrun xcresulttool` only. Prerequisites are probed against the **currently
+selected** Xcode (`xcode-select`): see `_check_xcresulttool_capability`.
+
+**Discovery (documented CLI behavior)**
+
+- ``xcresulttool get test-results tests`` — “Get all tests from test report.”
+  (``xcrun xcresulttool help get test-results tests``). We walk Test Case nodes
+  and read each ``nodeIdentifierURL`` as ``--test-id``.
+
+- ``xcresulttool get test-results activities`` — “Get the activity trees for the
+  specified test.” (``help get test-results activities``). Attachment records
+  (``name``, ``payloadId``) are taken from activity trees under ``testRuns``.
+
+- ``xcresulttool export attachments`` — “An additional manifest.json file will be
+  generated… contains a list of attachment details for each test.”
+  (``help export attachments``). We resolve on-disk filenames via
+  ``exportedFileName`` / ``suggestedHumanReadableName`` in that manifest when
+  present.
+
+**Logical filename matching (repo heuristic, not from Apple help)**
+
+We match ``--name`` to each activities record’s ``name`` by **exact string**, or
+by **stem + underscore-suffixed** basename before the extension. Adjust
+``_matches_wanted_attachment`` if your tests use different attachment titles.
+
+If ``activities`` yields no attachments, we fall back to walking the ``tests``
+JSON for ``attachments`` on nodes (empty after the activities pass in some
+bundles).
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import shutil
+import subprocess
+import sys
+import tempfile
+from pathlib import Path
+from typing import Any, List, Optional, Set, Tuple
+
+AttachmentRow = Tuple[str, str, str]  # stored_name, test_id_url, payload_id
+
+
+def _die(what: str, where: str, next_step: str, code: int = 1) -> int:
+    sys.stderr.write(
+        f"[extract_xcresult_attachments] FAIL: {what}\n  Where: {where}\n  Next:  {next_step}\n"
+    )
+    return code
+
+
+def _run(cmd: List[str]) -> subprocess.CompletedProcess:
+    return subprocess.run(cmd, capture_output=True, check=False)
+
+
+def _check_xcresulttool_capability() -> None:
+    """Require `xcrun xcresulttool help get` to list `test-results` (command exists)."""
+    r = _run(["xcrun", "xcresulttool", "help", "get"])
+    text = (r.stdout + r.stderr).decode(errors="replace")
+    if "test-results" not in text:
+        raise SystemExit(
+            _die(
+                "`xcrun xcresulttool help get` has no `test-results` subcommand in this Xcode.",
+                "xcrun xcresulttool help get",
+                "Install a newer Xcode or `xcode-select` a toolchain that ships this subcommand.",
+            )
+        )
+
+
+def _load_tests_json(result_path: Path) -> dict:
+    r = _run(
+        [
+            "xcrun",
+            "xcresulttool",
+            "get",
+            "test-results",
+            "tests",
+            "--path",
+            str(result_path),
+            "--format",
+            "json",
+        ]
+    )
+    if r.returncode != 0:
+        raise SystemExit(
+            _die(
+                f"xcresulttool get test-results failed with exit {r.returncode}.",
+                str(result_path),
+                "Confirm the .xcresult was produced by a completed `xcodebuild test`.",
+            )
+        )
+    try:
+        return json.loads(r.stdout or b"{}")
+    except json.JSONDecodeError as e:
+        raise SystemExit(
+            _die(
+                f"Could not parse xcresulttool JSON: {e}.",
+                str(result_path),
+                "Re-run xcodebuild test to regenerate the .xcresult.",
+            )
+        )
+
+
+def _collect_test_case_urls(node: dict) -> List[str]:
+    urls: List[str] = []
+    if node.get("nodeType") == "Test Case":
+        u = node.get("nodeIdentifierURL")
+        if isinstance(u, str) and u.startswith("test://"):
+            urls.append(u)
+    for ch in node.get("children") or []:
+        urls.extend(_collect_test_case_urls(ch))
+    return urls
+
+
+def _walk_activity_tree(act: dict, acc: List[AttachmentRow], test_url: str) -> None:
+    for att in act.get("attachments") or []:
+        name = att.get("name")
+        pid = att.get("payloadId")
+        if isinstance(name, str) and isinstance(pid, str):
+            acc.append((name, test_url, pid))
+    for ch in act.get("childActivities") or []:
+        _walk_activity_tree(ch, acc, test_url)
+
+
+def _attachments_from_activities(
+    result_path: Path, test_url: str
+) -> List[AttachmentRow]:
+    r = _run(
+        [
+            "xcrun",
+            "xcresulttool",
+            "get",
+            "test-results",
+            "activities",
+            "--path",
+            str(result_path),
+            "--test-id",
+            test_url,
+            "--format",
+            "json",
+        ]
+    )
+    if r.returncode != 0:
+        return []
+    try:
+        parsed = json.loads(r.stdout or b"{}")
+    except json.JSONDecodeError:
+        return []
+    acc: List[AttachmentRow] = []
+    for tr in parsed.get("testRuns") or []:
+        for act in tr.get("activities") or []:
+            _walk_activity_tree(act, acc, test_url)
+    return acc
+
+
+def _walk_attachments_legacy(node: dict, acc: List[AttachmentRow]) -> None:
+    """Fallback: walk ``tests`` JSON for ``attachments`` on nodes (used if activities path is empty)."""
+    identifier = node.get("identifier") or node.get("nodeIdentifierURL") or ""
+    for att in node.get("attachments") or []:
+        name = att.get("name")
+        pid = att.get("payloadId")
+        if isinstance(name, str) and isinstance(pid, str):
+            acc.append((name, str(identifier), pid))
+    for child in node.get("children") or []:
+        _walk_attachments_legacy(child, acc)
+
+
+def _enumerate(result_path: Path) -> List[AttachmentRow]:
+    parsed = _load_tests_json(result_path)
+    acc: List[AttachmentRow] = []
+    urls: List[str] = []
+    for top in parsed.get("testNodes") or []:
+        urls.extend(_collect_test_case_urls(top))
+    for url in urls:
+        acc.extend(_attachments_from_activities(result_path, url))
+    if acc:
+        return acc
+    for top in parsed.get("testNodes") or []:
+        _walk_attachments_legacy(top, acc)
+    return acc
+
+
+def _matches_wanted_attachment(stored: str, wanted: str) -> bool:
+    if stored == wanted:
+        return True
+    stem = Path(wanted).stem
+    suffix = Path(wanted).suffix if Path(wanted).suffix else ".json"
+    return stored.startswith(stem + "_") and stored.endswith(suffix)
+
+
+def _pick_exported_file(
+    td_path: Path, wanted_out_name: str, stored_name: str
+) -> Optional[Path]:
+    """Pick the exported file using ``manifest.json`` when `xcresulttool export` writes it (see ``help export``)."""
+    manifest_path = td_path / "manifest.json"
+    if manifest_path.is_file():
+        try:
+            raw: Any = json.loads(manifest_path.read_text(encoding="utf-8"))
+        except json.JSONDecodeError:
+            raw = []
+        blocks: List[dict] = raw if isinstance(raw, list) else []
+        for block in blocks:
+            for att in block.get("attachments") or []:
+                if not isinstance(att, dict):
+                    continue
+                sug = att.get("suggestedHumanReadableName") or ""
+                fn = att.get("exportedFileName")
+                if not isinstance(fn, str):
+                    continue
+                candidate = td_path / fn
+                if not candidate.is_file():
+                    continue
+                if sug == stored_name or _matches_wanted_attachment(
+                    sug, wanted_out_name
+                ):
+                    return candidate
+
+    direct = td_path / stored_name
+    if direct.is_file():
+        return direct
+
+    for p in sorted(td_path.iterdir()):
+        if not p.is_file() or p.name == "manifest.json":
+            continue
+        if _matches_wanted_attachment(p.name, wanted_out_name):
+            return p
+
+    orphans = [
+        p
+        for p in td_path.iterdir()
+        if p.is_file() and p.suffix == ".json" and p.name != "manifest.json"
+    ]
+    if len(orphans) == 1:
+        return orphans[0]
+
+    return None
+
+
+def _export_one(
+    result_path: Path,
+    test_identifier: str,
+    wanted_out_name: str,
+    stored_name: str,
+    out_path: Path,
+) -> None:
+    with tempfile.TemporaryDirectory() as td:
+        r = _run(
+            [
+                "xcrun",
+                "xcresulttool",
+                "export",
+                "attachments",
+                "--path",
+                str(result_path),
+                "--test-id",
+                test_identifier,
+                "--output-path",
+                td,
+            ]
+        )
+        if r.returncode != 0:
+            raise SystemExit(
+                _die(
+                    f"xcresulttool export attachments failed with exit {r.returncode}.",
+                    f"test-id={test_identifier}",
+                    "Inspect stderr above.",
+                )
+            )
+        td_path = Path(td)
+        src = _pick_exported_file(td_path, wanted_out_name, stored_name)
+        if src is None or not src.is_file():
+            raise SystemExit(
+                _die(
+                    f"Could not resolve exported file (wanted {wanted_out_name!r}, stored {stored_name!r}).",
+                    f"export dir: {td}",
+                    "Check `manifest.json` and `xcresulttool export attachments` stderr.",
+                )
+            )
+        out_path.parent.mkdir(parents=True, exist_ok=True)
+        shutil.move(str(src), str(out_path))
+
+
+def main(argv: List[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument(
+        "--result", required=True, type=Path, help="Path to .xcresult bundle."
+    )
+    parser.add_argument(
+        "--output-dir",
+        required=True,
+        type=Path,
+        help="Directory to write exported attachments into.",
+    )
+    parser.add_argument(
+        "--name",
+        action="append",
+        default=[],
+        help="Attachment name to extract (may be repeated). Fails the run if missing.",
+    )
+    parser.add_argument(
+        "--optional-name",
+        action="append",
+        default=[],
+        dest="optional_name",
+        help="Like --name, but a missing attachment is skipped with a message (exit still 0).",
+    )
+    args = parser.parse_args(argv)
+    if not args.name and not args.optional_name:
+        parser.error("at least one of --name or --optional-name is required")
+
+    _check_xcresulttool_capability()
+
+    if not args.result.exists():
+        return _die(
+            ".xcresult bundle does not exist.",
+            str(args.result),
+            "Run `xcodebuild test -resultBundlePath <path>` first.",
+        )
+
+    attachments = _enumerate(args.result)
+
+    errors = 0
+
+    def _extract_wanted(wanted: str, optional: bool) -> None:
+        nonlocal errors
+        hits = [
+            (tid, sn)
+            for sn, tid, _pid in attachments
+            if _matches_wanted_attachment(sn, wanted)
+        ]
+        uniq: Set[Tuple[str, str]] = set(hits)
+        if len(uniq) == 0:
+            if optional:
+                sys.stderr.write(
+                    f"[extract_xcresult_attachments] optional attachment not found, skipping: {wanted!r}\n"
+                )
+                return
+            errors += _die(
+                f"Attachment not found: {wanted}.",
+                str(args.result),
+                "Confirm the test run attached JSON and inspect `get test-results activities` / `export attachments` for this bundle.",
+            )
+            return
+        if len(uniq) > 1:
+            errors += _die(
+                f"Attachment name collision: {wanted} matches {len(uniq)} exports.",
+                str(args.result),
+                "Ensure attachment base names are unique across tests.",
+            )
+            return
+        test_id, stored = next(iter(uniq))
+        _export_one(args.result, test_id, wanted, stored, args.out_dir / wanted)
+
+    for wanted in args.name:
+        _extract_wanted(wanted, optional=False)
+    for wanted in args.optional_name:
+        _extract_wanted(wanted, optional=True)
+
+    return 1 if errors else 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/deploy/ios_demo/scripts/extract_xctest_metrics.py b/deploy/ios_demo/scripts/extract_xctest_metrics.py
new file mode 100755
index 0000000..24b9e37
--- /dev/null
+++ b/deploy/ios_demo/scripts/extract_xctest_metrics.py
@@ -0,0 +1,129 @@
+#!/usr/bin/env python3
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Export XCTest performance metrics from an xcresult bundle."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import subprocess
+import sys
+from pathlib import Path
+from typing import Any, Dict, List, Optional, Sequence
+
+
+def _die(message: str, code: int = 1) -> int:
+    sys.stderr.write(f"[extract_xctest_metrics] {message}\n")
+    return code
+
+
+def _metric_summary(values: List[float]) -> Dict[str, float]:
+    if not values:
+        return {}
+    return {
+        "count": len(values),
+        "min": min(values),
+        "max": max(values),
+        "mean": sum(values) / len(values),
+    }
+
+
+def export_metrics(result: Path) -> Dict[str, Any]:
+    proc = subprocess.run(
+        [
+            "xcrun",
+            "xcresulttool",
+            "get",
+            "test-results",
+            "metrics",
+            "--path",
+            str(result),
+            "--compact",
+        ],
+        text=True,
+        capture_output=True,
+        check=False,
+    )
+    if proc.returncode != 0:
+        raise RuntimeError(proc.stderr.strip() or "xcresulttool metrics export failed")
+    raw = json.loads(proc.stdout or "[]")
+    tests = raw if isinstance(raw, list) else []
+    memory_metrics: List[Dict[str, Any]] = []
+    for test in tests:
+        if not isinstance(test, dict):
+            continue
+        for run in test.get("testRuns") or []:
+            if not isinstance(run, dict):
+                continue
+            device = run.get("device") if isinstance(run.get("device"), dict) else {}
+            config = (
+                run.get("testPlanConfiguration")
+                if isinstance(run.get("testPlanConfiguration"), dict)
+                else {}
+            )
+            for metric in run.get("metrics") or []:
+                if not isinstance(metric, dict):
+                    continue
+                display_name = str(metric.get("displayName") or "")
+                identifier = str(metric.get("identifier") or "")
+                unit = str(metric.get("unitOfMeasurement") or "")
+                haystack = f"{display_name} {identifier} {unit}".lower()
+                if "memory" not in haystack:
+                    continue
+                values = [
+                    float(v)
+                    for v in metric.get("measurements") or []
+                    if isinstance(v, (int, float))
+                ]
+                memory_metrics.append(
+                    {
+                        "testIdentifier": test.get("testIdentifier"),
+                        "testIdentifierURL": test.get("testIdentifierURL"),
+                        "device": device,
+                        "configuration": config,
+                        "displayName": display_name,
+                        "identifier": identifier or None,
+                        "unitOfMeasurement": unit,
+                        "measurements": values,
+                        "summary": _metric_summary(values),
+                    }
+                )
+    return {
+        "schemaVersion": 1,
+        "source": "xcresulttool get test-results metrics",
+        "metrics": memory_metrics,
+    }
+
+
+def main(argv: Optional[Sequence[str]] = None) -> int:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--result", type=Path, required=True)
+    parser.add_argument("--output", type=Path, required=True)
+    args = parser.parse_args(list(argv) if argv is not None else None)
+    try:
+        doc = export_metrics(args.result)
+    except (OSError, RuntimeError, json.JSONDecodeError) as exc:
+        return _die(str(exc))
+    args.output.parent.mkdir(parents=True, exist_ok=True)
+    args.output.write_text(
+        json.dumps(doc, indent=2, ensure_ascii=False), encoding="utf-8"
+    )
+    print(f"Wrote {args.output}")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/deploy/ios_demo/scripts/fetch_ios_demo_models.sh b/deploy/ios_demo/scripts/fetch_ios_demo_models.sh
new file mode 100755
index 0000000..84bd6dc
--- /dev/null
+++ b/deploy/ios_demo/scripts/fetch_ios_demo_models.sh
@@ -0,0 +1,151 @@
+#!/usr/bin/env bash
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+# Fetch ONNX detection + recognition model bundles for the iOS demo.
+#
+# Requires: bash, curl, tar
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+IOS_DEMO_ROOT="$(cd "${SCRIPT_DIR}/.." && pwd)"
+
+BASE_URL="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0"
+ALLOWED_PRESETS=(PP-OCRv6_small PP-OCRv6_tiny PP-OCRv5_mobile)
+DEFAULT_PRESET="PP-OCRv6_small"
+DEST_DET="${IOS_DEMO_ROOT}/PaddleOCRDemo/Models/det"
+DEST_REC="${IOS_DEMO_ROOT}/PaddleOCRDemo/Models/rec"
+
+WORKDIR="${IOS_DEMO_ROOT}/.fetch_ios_demo_models_work"
+
+usage() {
+  cat <<EOF
+Usage: ./scripts/fetch_ios_demo_models.sh [-h|--help] [preset]
+
+Downloads ONNX bundles into:
+  ${DEST_DET}
+  ${DEST_REC}
+
+Options:
+  -h, --help    Show this help
+
+Positional:
+  preset        Model preset; one of: ${ALLOWED_PRESETS[*]} (default: ${DEFAULT_PRESET})
+EOF
+}
+
+preset="${DEFAULT_PRESET}"
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    -h|--help)
+      usage
+      exit 0
+      ;;
+    -*)
+      echo "error: unknown option: $1" >&2
+      exit 1
+      ;;
+    *)
+      break
+      ;;
+  esac
+done
+
+args=()
+while [[ $# -gt 0 ]]; do
+  args+=("$1")
+  shift
+done
+
+if [[ ${#args[@]} -gt 1 ]]; then
+  echo "error: at most one model preset argument is allowed" >&2
+  exit 1
+fi
+if [[ ${#args[@]} -eq 1 ]]; then
+  preset="${args[0]}"
+fi
+
+die() {
+  echo "error: $*" >&2
+  exit 1
+}
+
+preset_allowed() {
+  local v="$1"
+  local x
+  for x in "${ALLOWED_PRESETS[@]}"; do
+    if [[ "$x" == "$v" ]]; then
+      return 0
+    fi
+  done
+  return 1
+}
+
+command -v curl >/dev/null || die "curl is required"
+command -v tar >/dev/null || die "tar is required"
+
+if ! preset_allowed "${preset}"; then
+  die "unsupported model preset: ${preset} (allowed: ${ALLOWED_PRESETS[*]})"
+fi
+
+DET_TAR="${preset}_det_onnx_infer.tar"
+REC_TAR="${preset}_rec_onnx_infer.tar"
+
+mkdir -p "${WORKDIR}"
+
+fetch() {
+  local name="$1"
+  local url="${BASE_URL}/${name}"
+  local out="${WORKDIR}/${name}"
+  if [[ -f "${out}" ]]; then
+    echo "Using cached ${out}"
+  else
+    echo "Downloading ${url}"
+    curl -fL --retry 3 --retry-delay 2 -o "${out}" "${url}"
+  fi
+}
+
+extract() {
+  local tar_path="$1"
+  tar -xf "${tar_path}" -C "${WORKDIR}"
+}
+
+install_extracted_dir() {
+  local src_dir="$1"
+  local dest_dir="$2"
+  local label="$3"
+  [[ -d "${src_dir}" ]] || die "expected directory ${src_dir} (${label})"
+  mkdir -p "$(dirname "${dest_dir}")"
+  rm -rf "${dest_dir}"
+  mv "${src_dir}" "${dest_dir}"
+}
+
+echo "=== ONNX models (preset: ${preset}) ==="
+echo "Work directory: ${WORKDIR}"
+fetch "${DET_TAR}"
+fetch "${REC_TAR}"
+
+extract "${WORKDIR}/${DET_TAR}"
+extract "${WORKDIR}/${REC_TAR}"
+
+DET_SRC="${WORKDIR}/$(basename "${DET_TAR}" .tar)"
+REC_SRC="${WORKDIR}/$(basename "${REC_TAR}" .tar)"
+
+install_extracted_dir "${DET_SRC}" "${DEST_DET}" "det"
+install_extracted_dir "${REC_SRC}" "${DEST_REC}" "rec"
+
+echo "Installed:"
+echo "  ${DEST_DET}"
+echo "  ${DEST_REC}"
diff --git a/deploy/ios_demo/scripts/generate_benchmark_report.py b/deploy/ios_demo/scripts/generate_benchmark_report.py
new file mode 100755
index 0000000..9f4791a
--- /dev/null
+++ b/deploy/ios_demo/scripts/generate_benchmark_report.py
@@ -0,0 +1,697 @@
+#!/usr/bin/env python3
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Merge optional benchmark JSONs into a local Markdown report.
+
+Inputs:
+    --accuracy-summary       Optional JSON from compare_ocr_json.py
+    --on-device-performance-json   JSON attachment "on-device-performance.json"
+    --run-status             JSON from run_benchmark.sh
+    --xctest-metrics-json    JSON from extract_xctest_metrics.py
+
+All inputs are optional; missing sections render placeholders.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import subprocess
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any, Dict, List, Optional, Sequence
+
+
+def _run_text(cmd: List[str]) -> Optional[str]:
+    try:
+        return subprocess.check_output(
+            cmd, stderr=subprocess.DEVNULL, text=True
+        ).strip()
+    except (OSError, subprocess.CalledProcessError):
+        return None
+
+
+def _git_short() -> Optional[str]:
+    return _run_text(["git", "rev-parse", "--short", "HEAD"])
+
+
+def _xcode_version() -> Optional[str]:
+    return _run_text(["xcodebuild", "-version"])
+
+
+def _sw_vers() -> Optional[str]:
+    return _run_text(["sw_vers", "-productVersion"])
+
+
+def _load_json(path: Path) -> Dict[str, Any]:
+    with path.open("r", encoding="utf-8") as f:
+        data = json.load(f)
+    return data if isinstance(data, dict) else {}
+
+
+def _fmt_bytes(n: Optional[int]) -> str:
+    if n is None:
+        return "— (path missing)"
+    if n < 1024:
+        return f"{n} B"
+    if n < 1024 * 1024:
+        return f"{n / 1024:.1f} KiB"
+    return f"{n / (1024 * 1024):.2f} MiB"
+
+
+def _fmt_duration_ms(raw: Any) -> str:
+    if raw is None:
+        return "—"
+    try:
+        ms = float(raw)
+    except (TypeError, ValueError):
+        return "—"
+    if ms < 1000:
+        return f"{ms:.0f} ms"
+    return f"{ms / 1000:.2f} s"
+
+
+def _fmt_ms(raw: Any) -> str:
+    try:
+        return f"{float(raw):.4f} ms"
+    except (TypeError, ValueError):
+        return "—"
+
+
+def _fmt_shape(raw: Any) -> str:
+    if isinstance(raw, list) and raw:
+        return "`[" + ", ".join(str(x) for x in raw) + "]`"
+    return "—"
+
+
+def _truthy(value: Any) -> bool:
+    if isinstance(value, bool):
+        return value
+    if value is None:
+        return False
+    return str(value).strip().lower() in {"1", "true", "yes", "on"}
+
+
+def _benchmark_value(
+    status: Dict[str, Any],
+    perf_data: Optional[Dict[str, Any]],
+    key: str,
+) -> Any:
+    benchmark = (
+        status.get("benchmark") if isinstance(status.get("benchmark"), dict) else {}
+    )
+    if key in benchmark:
+        return benchmark.get(key)
+    if isinstance(perf_data, dict):
+        return perf_data.get(key)
+    return None
+
+
+def _benchmark_warnings(
+    status: Dict[str, Any],
+    perf_data: Optional[Dict[str, Any]],
+) -> List[str]:
+    warnings: List[str] = []
+    build = _benchmark_value(status, perf_data, "buildConfiguration")
+    is_sim = _benchmark_value(status, perf_data, "isSimulator")
+    profiling = _benchmark_value(status, perf_data, "ortProfilingEnabled")
+    if build and str(build) != "Release":
+        warnings.append(f"build configuration is `{build}`, not `Release`")
+    if _truthy(is_sim):
+        warnings.append("destination is simulator")
+    if _truthy(profiling):
+        warnings.append("ORT profiling is enabled")
+    return warnings
+
+
+def _benchmark_metadata_markdown(
+    status: Dict[str, Any],
+    perf_data: Optional[Dict[str, Any]],
+) -> str:
+    rows: List[str] = ["| Metric | Value |", "|--------|-------|"]
+    ep = _benchmark_value(status, perf_data, "ortExecutionProvider")
+    if ep is not None:
+        rows.append(f"| ORT execution provider | `{ep}` |")
+    for key, label in [
+        ("buildConfiguration", "Build configuration"),
+        ("deviceModel", "Device model"),
+        ("osVersion", "OS version"),
+        ("isSimulator", "Simulator"),
+        ("modelPreset", "Model preset"),
+        ("ortProfilingEnabled", "ORT profiling"),
+    ]:
+        value = _benchmark_value(status, perf_data, key)
+        if value is None:
+            continue
+        rows.append(f"| {label} | `{value}` |")
+    warnings = _benchmark_warnings(status, perf_data)
+    if warnings:
+        rows.append("")
+        rows.append(
+            "> Not publishable: "
+            + "; ".join(warnings)
+            + ". Only Release runs on a real device with ORT profiling disabled should be used as publishable latency numbers."
+        )
+    else:
+        rows.append("")
+        rows.append(
+            "> Publishable baseline: Release run on a real device with ORT profiling disabled."
+        )
+    return "\n".join(rows)
+
+
+def _model_artifacts_markdown(status: Dict[str, Any]) -> str:
+    benchmark = (
+        status.get("benchmark") if isinstance(status.get("benchmark"), dict) else {}
+    )
+    model_files = (
+        benchmark.get("modelFiles")
+        if isinstance(benchmark.get("modelFiles"), dict)
+        else {}
+    )
+    det = (
+        model_files.get("detection")
+        if isinstance(model_files.get("detection"), dict)
+        else {}
+    )
+    rec = (
+        model_files.get("recognition")
+        if isinstance(model_files.get("recognition"), dict)
+        else {}
+    )
+    rows = ["| Artifact | Format | Size |", "|----------|--------|------|"]
+    rows.append(
+        f"| Detection model | `{det.get('format') or '—'}` | {_fmt_bytes(det.get('sizeBytes'))} |"
+    )
+    rows.append(
+        f"| Recognition model | `{rec.get('format') or '—'}` | {_fmt_bytes(rec.get('sizeBytes'))} |"
+    )
+    rows.append(
+        f"| Total model weights | — | {_fmt_bytes(model_files.get('totalSizeBytes'))} |"
+    )
+    rows.append(
+        f"| App executable | — | {_fmt_bytes(benchmark.get('appBinarySizeBytes'))} |"
+    )
+    return "\n".join(rows)
+
+
+# ---------- run-status ----------
+
+
+def _run_status_section(status: Dict[str, Any]) -> str:
+    lines: List[str] = ["## Run status", ""]
+    lines.append(f"- **Exit code:** `{status.get('exitCode', '—')}`")
+    image = status.get("image") or {}
+    if image:
+        lines.append(
+            f"- **Image:** `{image.get('name', '—')}` ({image.get('source', '—')})"
+        )
+    dest = status.get("destination")
+    if dest:
+        lines.append(f"- **Destination:** `{dest}`")
+    benchmark = (
+        status.get("benchmark") if isinstance(status.get("benchmark"), dict) else {}
+    )
+    if benchmark:
+        if build := benchmark.get("buildConfiguration"):
+            lines.append(f"- **Build configuration:** `{build}`")
+        ep = benchmark.get("ortExecutionProvider")
+        if ep:
+            lines.append(f"- **ORT execution provider:** `{ep}`")
+        if scope := benchmark.get("testingScope"):
+            lines.append(f"- **Testing scope:** `{scope}`")
+    started = status.get("runStartedAt")
+    finished = status.get("runFinishedAt")
+    if started and finished:
+        lines.append(f"- **Run window:** {started} → {finished}")
+    lines.append("")
+    lines.append("| Step | Status | Duration | Notes |")
+    lines.append("|------|--------|----------|-------|")
+    for step in status.get("steps") or []:
+        name = step.get("name", "?")
+        st = (step.get("status") or "?").upper()
+        dur = _fmt_duration_ms(step.get("durationMs"))
+        note = step.get("reason") or ""
+        if step.get("logPath"):
+            note = (note + " · " if note else "") + f"see `{step['logPath']}`"
+        if step.get("exitCode") not in (None, 0):
+            note = (note + " · " if note else "") + f"exit {step['exitCode']}"
+        lines.append(f"| {name} | {st} | {dur} | {note} |")
+    return "\n".join(lines)
+
+
+def _stale_data_banner(overall: str) -> str:
+    return (
+        f"> This run ended with status **{overall}**; the numbers below may be partial or stale. "
+        "Do not use them as a baseline.\n"
+    )
+
+
+def _performance_section_banner(overall: Optional[str]) -> str:
+    if overall and overall != "COMPLETED":
+        return _stale_data_banner(overall)
+    return ""
+
+
+# ---------- timing + memory (on-device performance) ----------
+
+
+def _timing_markdown(data: Dict[str, Any]) -> str:
+    lines = ["| Metric | mean | stdev | p90 |", "|--------|------|-------|-----|"]
+    for key, label in [
+        ("totalTimeMs", "Full OCR latency (ms)"),
+        ("detectionTimeMs", "Detection stage latency (ms)"),
+        ("detectionPreprocessTimeMs", "Detection preprocess latency (ms)"),
+        ("detectionInferenceTimeMs", "Detection inference segment latency (ms)"),
+        ("detectionPostprocessTimeMs", "Detection postprocess latency (ms)"),
+        ("recognitionTimeMs", "Recognition stage latency (ms)"),
+        ("recognitionPreprocessTimeMs", "Recognition preprocess latency (ms)"),
+        ("recognitionInferenceTimeMs", "Recognition inference segment latency (ms)"),
+        ("recognitionPostprocessTimeMs", "Recognition postprocess latency (ms)"),
+    ]:
+        block = data.get(key)
+        if not isinstance(block, dict):
+            continue
+        lines.append(
+            "| {label} | {mean:.4f} | {stdev:.4f} | {p90:.4f} |".format(
+                label=label,
+                mean=float(block.get("mean", 0)),
+                stdev=float(block.get("stdev", 0)),
+                p90=float(block.get("p90", 0)),
+            )
+        )
+    ovh = data.get("pipelineOverheadTimeMs")
+    if isinstance(ovh, dict):
+        lines.append(
+            "| Pipeline overhead latency (ms) | {mean:.4f} | {stdev:.4f} | {p90:.4f} |".format(
+                mean=float(ovh.get("mean", 0)),
+                stdev=float(ovh.get("stdev", 0)),
+                p90=float(ovh.get("p90", 0)),
+            )
+        )
+    w = data.get("warmupIterations")
+    m = data.get("measuredIterations")
+    if w is not None or m is not None:
+        lines.append("")
+        lines.append(f"Warmup iterations: `{w}` · Measured iterations: `{m}`")
+    lines.append("")
+    lines.append(
+        "*Inference segment latency includes the tensor input/output handling required by the ORT wrapper, "
+        "`session.run`, output reads, and validation. It is not pure kernel latency or isolated `session.run` latency.*"
+    )
+
+    per_line_extra = _recognition_per_line_markdown(data)
+    if per_line_extra:
+        lines.append("")
+        lines.append(per_line_extra)
+    return "\n".join(lines)
+
+
+def _input_profile_markdown(data: Dict[str, Any]) -> str:
+    dist = (
+        data.get("inputShapeDistribution")
+        if isinstance(data.get("inputShapeDistribution"), dict)
+        else {}
+    )
+    rows = [
+        "| Model | Input tensor shape | Count |",
+        "|-------|--------------------|-------|",
+    ]
+    for model_key, label in [
+        ("detection", "Detection"),
+        ("recognition", "Recognition"),
+    ]:
+        samples = dist.get(model_key)
+        if not isinstance(samples, list) or not samples:
+            rows.append(f"| {label} | — | `0` |")
+            continue
+        for sample in samples:
+            if not isinstance(sample, dict):
+                continue
+            rows.append(
+                f"| {label} | {_fmt_shape(sample.get('shape'))} | `{sample.get('count', '—')}` |"
+            )
+    rows.append("")
+    rows.append("| Metric | Value |")
+    rows.append("|--------|-------|")
+    rows.append(
+        f"| First measured run line count | `{data.get('firstMeasuredLineCount', '—')}` |"
+    )
+    rows.append(f"| Cold model load time | {_fmt_ms(data.get('coldLoadTimeMs'))} |")
+    rows.append("")
+    rows.append(
+        "*Input tensor shape counts are counted per model invocation in the measured loop.*"
+    )
+    return "\n".join(rows)
+
+
+def _recognition_per_line_markdown(data: Dict[str, Any]) -> str:
+    """Renders pooled `recognitionPerLine` samples."""
+    pl = data.get("recognitionPerLine")
+    if not isinstance(pl, dict):
+        return ""
+    out: List[str] = []
+    pooled = pl.get("pooled")
+    any_pooled = False
+    pooled_lines: List[str] = [
+        "| Pooled (all lines × measured runs) | count | mean | stdev | p90 |",
+        "|------------------------------------|------|------|-------|-----|",
+    ]
+    for key, label in [
+        ("preprocessMs", "Per-line recognition preprocess pooled (ms)"),
+        ("inferenceMs", "Per-line recognition inference segment pooled (ms)"),
+        ("postprocessMs", "Per-line recognition postprocess pooled (ms)"),
+        ("totalMs", "Per-line recognition total pooled (ms)"),
+    ]:
+        if not isinstance(pooled, dict):
+            break
+        b = pooled.get(key)
+        if not isinstance(b, dict):
+            continue
+        c = pooled.get("count") or b.get("count")
+        if c is None or c == 0:
+            continue
+        any_pooled = True
+        pooled_lines.append(
+            "| {label} | {c} | {me:.4f} | {sd:.4f} | {p90:.4f} |".format(
+                label=label,
+                c=int(c),
+                me=float(b.get("mean", 0)),
+                sd=float(b.get("stdev", 0)),
+                p90=float(b.get("p90", 0)),
+            )
+        )
+    if any_pooled:
+        out.append("#### Per-line Recognition Latency (ms)")
+        out.append("")
+        out.append(
+            "*Pooled* merges every detected line from every measured full-image run."
+        )
+        out.append("")
+        out.extend(pooled_lines)
+
+    return "\n".join(out) if out else ""
+
+
+def _fmt_metric_number(value: Any, unit: str) -> str:
+    try:
+        number = float(value)
+    except (TypeError, ValueError):
+        return "—"
+    if unit.lower() == "kb":
+        return f"{number / 1024:.2f} MB"
+    if "byte" in unit.lower():
+        return _fmt_bytes(int(number))
+    if abs(number) >= 1000:
+        text = f"{number:.0f}"
+    else:
+        text = f"{number:.4f}".rstrip("0").rstrip(".")
+    return f"{text} {unit}".rstrip()
+
+
+def _xctest_memory_rows(xctest_metrics: Optional[Dict[str, Any]]) -> List[str]:
+    rows: List[str] = []
+    metrics = (
+        xctest_metrics.get("metrics") if isinstance(xctest_metrics, dict) else None
+    )
+    if not isinstance(metrics, list) or not metrics:
+        rows.append("| XCTest memory metric | `not available` |")
+        return rows
+    for metric in metrics:
+        if not isinstance(metric, dict):
+            continue
+        summary = (
+            metric.get("summary") if isinstance(metric.get("summary"), dict) else {}
+        )
+        unit = str(metric.get("unitOfMeasurement") or "")
+        raw_label = str(
+            metric.get("displayName") or metric.get("identifier") or "memory"
+        )
+        label = {
+            "Memory Physical": "physical memory delta",
+            "Memory Peak Physical": "physical memory peak",
+        }.get(raw_label, raw_label)
+        test_id = metric.get("testIdentifier")
+        prefix = f"XCTest {label}"
+        if test_id:
+            prefix += f" ({test_id})"
+        rows.append(f"| {prefix} count | `{int(summary.get('count', 0))}` |")
+        rows.append(
+            f"| {prefix} min | {_fmt_metric_number(summary.get('min'), unit)} |"
+        )
+        rows.append(
+            f"| {prefix} max | {_fmt_metric_number(summary.get('max'), unit)} |"
+        )
+        rows.append(
+            f"| {prefix} mean | {_fmt_metric_number(summary.get('mean'), unit)} |"
+        )
+    return rows
+
+
+def _memory_markdown(
+    data: Dict[str, Any],
+    xctest_metrics: Optional[Dict[str, Any]],
+) -> str:
+    if data.get("memoryFootprintAfterLoadBytes") is None and xctest_metrics is None:
+        return ""
+    rows = [
+        "| Metric | Value |",
+        "|--------|-------|",
+    ]
+    rows.extend(_xctest_memory_rows(xctest_metrics))
+    before = data.get("memoryFootprintBeforeLoadBytes")
+    if before is not None:
+        rows.append(f"| Footprint before load | {_fmt_bytes(int(before))} |")
+    after = data.get("memoryFootprintAfterLoadBytes")
+    if after is not None:
+        rows.append(f"| Footprint after model load | {_fmt_bytes(int(after))} |")
+    ts = data.get("thermalState")
+    if ts is not None:
+        rows.append(f"| Thermal state | `{ts}` |")
+    rows.append("")
+    rows.append(
+        "`physical memory delta` is a per-iteration net change, so negative values mean memory was released during that measured block. "
+        "`task_vm_info.phys_footprint` rows are diagnostic snapshots before and after model loading, not continuous peak measurements.*"
+    )
+    return "\n".join(rows)
+
+
+def _on_device_section(
+    data: Dict[str, Any],
+    xctest_metrics: Optional[Dict[str, Any]],
+    status: Optional[Dict[str, Any]] = None,
+) -> str:
+    parts: List[str] = []
+    ver = data.get("schemaVersion")
+    if ver is not None:
+        parts.append(f"Schema version: `{ver}`")
+        parts.append("")
+    parts.append("### Benchmark metadata")
+    parts.append("")
+    parts.append(_benchmark_metadata_markdown(status or {}, data))
+    parts.append("")
+    parts.append(
+        "*Latency and memory are both runtime performance attributes, but they are sampled by separate benchmark paths.*"
+    )
+    parts.append("")
+    parts.append("### Model and app artifacts")
+    parts.append("")
+    parts.append(_model_artifacts_markdown(status or {}))
+    parts.append("")
+    parts.append("### Input profile")
+    parts.append("")
+    parts.append(_input_profile_markdown(data))
+    parts.append("")
+    parts.append("### Latency")
+    parts.append("")
+    parts.append(_timing_markdown(data))
+    mem = _memory_markdown(data, xctest_metrics)
+    if mem:
+        parts.append("")
+        parts.append("### Memory (resource footprint)")
+        parts.append("")
+        parts.append(mem)
+    return "\n".join(parts)
+
+
+# ---------- accuracy ----------
+
+
+def _compare_section(data: Dict[str, Any]) -> str:
+    passed = data.get("pass")
+    status = "**PASS**" if passed is True else ("**FAIL**" if passed is False else "—")
+    return "\n".join(
+        [
+            f"Status: {status}",
+            "",
+            "```json",
+            json.dumps(data, indent=2, ensure_ascii=False),
+            "```",
+        ]
+    )
+
+
+def _accuracy_section(
+    summary_data: Optional[Dict[str, Any]],
+    run_status: Dict[str, Any],
+) -> str:
+    accuracy = (
+        run_status.get("accuracy")
+        if isinstance(run_status.get("accuracy"), dict)
+        else {}
+    )
+    enabled = bool(accuracy.get("enabled")) or summary_data is not None
+    if not enabled:
+        return ""
+
+    parts: List[str] = ["## Accuracy check", ""]
+    status = str(accuracy.get("status") or "").upper()
+    reason = accuracy.get("reason")
+    if status:
+        parts.append(f"Precheck status: **{status}**")
+        if reason:
+            parts.append("")
+            parts.append(f"> {reason}")
+        parts.append("")
+    parts.append("")
+    if summary_data is not None:
+        parts.append(_compare_section(summary_data))
+    elif enabled:
+        parts.append("*No accuracy summary JSON was produced.*")
+    return "\n".join(parts)
+
+
+# ---------- driver ----------
+
+
+def _placeholder_from_status(status: Dict[str, Any], for_step: str) -> str:
+    for step in status.get("steps") or []:
+        if step.get("status") in ("fail", "skipped"):
+            reason = step.get("reason") or ""
+            if step.get("name") == for_step or reason:
+                return (
+                    f"*Unavailable — step `{step.get('name', '?')}` ended in "
+                    f"`{step.get('status')}`"
+                    + (f"; reason: {reason}" if reason else "")
+                    + ".*"
+                )
+    return "*Unavailable — no artifact and no run-status reason.*"
+
+
+def generate_report(
+    *,
+    out_path: Path,
+    on_device_performance_json: Optional[Path],
+    accuracy_summary: Optional[Path],
+    run_status: Optional[Path],
+    xctest_metrics_json: Optional[Path],
+) -> None:
+    status: Dict[str, Any] = {}
+    if run_status and run_status.is_file():
+        status = _load_json(run_status)
+
+    overall = (status.get("overall") or "").upper() or None
+
+    accuracy_data = (
+        _load_json(accuracy_summary)
+        if accuracy_summary and accuracy_summary.is_file()
+        else None
+    )
+    perf_data = (
+        _load_json(on_device_performance_json)
+        if on_device_performance_json and on_device_performance_json.is_file()
+        else None
+    )
+    xctest_metrics_data = (
+        _load_json(xctest_metrics_json)
+        if xctest_metrics_json and xctest_metrics_json.is_file()
+        else None
+    )
+
+    head_lines: List[str] = ["# iOS demo — benchmark report", ""]
+    if overall:
+        head_lines.append(f"**Overall:** {overall}")
+        head_lines.append("")
+
+    parts: List[str] = ["\n".join(head_lines)]
+
+    if status:
+        parts.append(_run_status_section(status))
+        parts.append("")
+
+    perf_banner = _performance_section_banner(overall)
+
+    accuracy_section = _accuracy_section(accuracy_data, status)
+    if accuracy_section:
+        parts.append(accuracy_section)
+        parts.append("")
+
+    parts.append("## On-device runtime performance")
+    parts.append("")
+    if perf_banner:
+        parts.append(perf_banner)
+    if perf_data is not None:
+        parts.append(_on_device_section(perf_data, xctest_metrics_data, status))
+    else:
+        parts.append(
+            _placeholder_from_status(status, "extract-artifacts")
+            if status
+            else "*No performance JSON provided.*"
+        )
+    parts.append("")
+
+    parts.append("## Metadata")
+    parts.append("")
+    parts.append(
+        f"- **Generated:** {datetime.now(timezone.utc).strftime('%Y-%m-%d %H:%M:%S UTC')}"
+    )
+    if v := _sw_vers():
+        parts.append(f"- **macOS:** {v}")
+    if v := _xcode_version():
+        parts.append("- **Xcode:**\n```\n" + v + "\n```")
+    if g := _git_short():
+        parts.append(f"- **Git (short):** `{g}`")
+    parts.append("")
+    parts.append("*Generated by `scripts/generate_benchmark_report.py`.*")
+
+    out_path.parent.mkdir(parents=True, exist_ok=True)
+    out_path.write_text("\n".join(parts), encoding="utf-8")
+    print(f"Wrote {out_path}")
+
+
+def main(argv: Optional[Sequence[str]] = None) -> int:
+    root = Path(__file__).resolve().parent.parent
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument(
+        "--output", type=Path, default=root / "out" / "benchmark-report.md"
+    )
+    parser.add_argument("--on-device-performance-json", type=Path, default=None)
+    parser.add_argument("--accuracy-summary", type=Path, default=None)
+    parser.add_argument("--run-status", type=Path, default=None)
+    parser.add_argument("--xctest-metrics-json", type=Path, default=None)
+    args = parser.parse_args(list(argv) if argv is not None else None)
+
+    generate_report(
+        out_path=args.output.resolve(),
+        on_device_performance_json=args.on_device_performance_json,
+        accuracy_summary=args.accuracy_summary,
+        run_status=args.run_status,
+        xctest_metrics_json=args.xctest_metrics_json,
+    )
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/deploy/ios_demo/scripts/ocr_reference_run.py b/deploy/ios_demo/scripts/ocr_reference_run.py
new file mode 100755
index 0000000..6b553d9
--- /dev/null
+++ b/deploy/ios_demo/scripts/ocr_reference_run.py
@@ -0,0 +1,213 @@
+#!/usr/bin/env python3
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Generate a JSON reference OCR result for comparison with the iOS demo.
+
+Uses ``PaddleOCR`` with ``engine="onnxruntime"`` and no doc-orientation /
+unwarping / textline-orientation modules.
+
+Examples:
+  python ocr_reference_run.py --image a.png --output ref.json \\
+    --device cpu --align-ios-defaults
+
+  # Explicit models path
+  python ocr_reference_run.py --image a.png --output ref.json \\
+    --ios-models-root /path/to/PaddleOCRDemo/Models --device cpu --align-ios-defaults
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import sys
+from pathlib import Path
+from typing import Any, Dict, List, Sequence
+
+
+def _paddleocr_package_root() -> Path:
+    return Path(__file__).resolve().parents[3]
+
+
+def _default_ios_models_root() -> Path:
+    return Path(__file__).resolve().parent.parent / "PaddleOCRDemo" / "Models"
+
+
+def _load_yaml_model_name(path: Path) -> str:
+    try:
+        import yaml
+    except ImportError as exc:
+        raise RuntimeError("PyYAML is required to read inference.yml") from exc
+    with path.open("r", encoding="utf-8") as f:
+        data = yaml.safe_load(f)
+    return data["Global"]["model_name"]
+
+
+def _numpy_to_python(obj: Any) -> Any:
+    if obj is None:
+        return None
+    if hasattr(obj, "tolist"):
+        return obj.tolist()
+    if isinstance(obj, (list, tuple)):
+        return [_numpy_to_python(x) for x in obj]
+    if isinstance(obj, dict):
+        return {k: _numpy_to_python(v) for k, v in obj.items()}
+    elif isinstance(obj, (str, int, float, bool)):
+        return obj
+    return str(obj)
+
+
+def _extract_items(result_obj: Any) -> List[Dict[str, Any]]:
+    """Build a list of {polygon, text, score} from an OCR pipeline result."""
+    if isinstance(result_obj, dict):
+        res = result_obj
+    elif hasattr(result_obj, "__getitem__"):
+        res = dict(result_obj)
+    else:
+        raise TypeError(f"Unsupported result type: {type(result_obj)}")
+    return _extract_items_from_res_dict(res)
+
+
+def _extract_items_from_res_dict(res: Dict[str, Any]) -> List[Dict[str, Any]]:
+    texts = res.get("rec_texts") or []
+    scores = res.get("rec_scores")
+    polys = res.get("rec_polys") or res.get("dt_polys") or []
+    if hasattr(scores, "tolist"):
+        scores = scores.tolist()
+    items: List[Dict[str, Any]] = []
+    n = min(len(texts), len(polys))
+    for i in range(n):
+        poly = polys[i]
+        if hasattr(poly, "tolist"):
+            poly = poly.tolist()
+        score = float(scores[i]) if scores is not None and i < len(scores) else None
+        items.append(
+            {
+                "polygon": _numpy_to_python(poly),
+                "text": str(texts[i]),
+                "score": score,
+            }
+        )
+    return items
+
+
+def _build_ocr(
+    *,
+    device: str,
+    text_detection_model_name: str | None,
+    text_recognition_model_name: str | None,
+    text_detection_model_dir: str | None,
+    text_recognition_model_dir: str | None,
+) -> Any:
+    from paddleocr import PaddleOCR
+
+    kwargs: Dict[str, Any] = dict(
+        use_doc_orientation_classify=False,
+        use_doc_unwarping=False,
+        use_textline_orientation=False,
+        device=device,
+        engine="onnxruntime",
+    )
+    if text_detection_model_name:
+        kwargs["text_detection_model_name"] = text_detection_model_name
+    if text_recognition_model_name:
+        kwargs["text_recognition_model_name"] = text_recognition_model_name
+    if text_detection_model_dir:
+        kwargs["text_detection_model_dir"] = text_detection_model_dir
+    if text_recognition_model_dir:
+        kwargs["text_recognition_model_dir"] = text_recognition_model_dir
+    return PaddleOCR(**kwargs)
+
+
+def main(argv: Sequence[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(
+        description="OCR reference JSON for iOS demo comparison"
+    )
+    parser.add_argument("--image", required=True, help="Input image path")
+    parser.add_argument("--output", required=True, help="Output JSON path")
+    parser.add_argument("--device", default="cpu", help="Device, e.g. cpu or gpu:0")
+    parser.add_argument(
+        "--ios-models-root",
+        type=Path,
+        default=_default_ios_models_root(),
+        help="Path to models directory",
+    )
+    parser.add_argument(
+        "--align-ios-defaults",
+        action="store_true",
+        help="Match OCRParameterResolver app-tier defaults (Swift demo)",
+    )
+    args = parser.parse_args(list(argv) if argv is not None else None)
+
+    image_path = Path(args.image).resolve()
+    if not image_path.is_file():
+        print(f"error: image not found: {image_path}", file=sys.stderr)
+        return 2
+
+    root = Path(args.ios_models_root).resolve()
+    det_yml = root / "det" / "inference.yml"
+    rec_yml = root / "rec" / "inference.yml"
+    if not det_yml.is_file() or not rec_yml.is_file():
+        print(f"error: expected {det_yml} and {rec_yml}", file=sys.stderr)
+        return 2
+    text_detection_model_name = _load_yaml_model_name(det_yml)
+    text_recognition_model_name = _load_yaml_model_name(rec_yml)
+    text_detection_model_dir = str(root / "det")
+    text_recognition_model_dir = str(root / "rec")
+
+    ocr = _build_ocr(
+        device=args.device,
+        text_detection_model_name=text_detection_model_name,
+        text_recognition_model_name=text_recognition_model_name,
+        text_detection_model_dir=text_detection_model_dir,
+        text_recognition_model_dir=text_recognition_model_dir,
+    )
+
+    predict_kwargs: Dict[str, Any] = {}
+    if args.align_ios_defaults:
+        predict_kwargs.update(
+            text_det_limit_side_len=64,
+            text_det_limit_type="min",
+            text_det_thresh=0.3,
+            text_det_box_thresh=0.6,
+            text_det_unclip_ratio=1.5,
+            text_rec_score_thresh=0.0,
+        )
+
+    results = ocr.predict(str(image_path), **predict_kwargs)
+    if not results:
+        print("error: empty predict() result", file=sys.stderr)
+        return 3
+
+    items = _extract_items(results[0])
+    payload: Dict[str, Any] = {
+        "schema_version": 1,
+        "source": "paddleocr_reference",
+        "engine": "onnxruntime",
+        "image": str(image_path),
+        "items": items,
+    }
+
+    out_path = Path(args.output).resolve()
+    out_path.parent.mkdir(parents=True, exist_ok=True)
+    with out_path.open("w", encoding="utf-8") as f:
+        json.dump(payload, f, ensure_ascii=False, indent=2)
+
+    print(f"Wrote {len(items)} lines to {out_path}")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.path.insert(0, str(_paddleocr_package_root()))
+    raise SystemExit(main())
diff --git a/deploy/ios_demo/scripts/quantize_onnx_model.py b/deploy/ios_demo/scripts/quantize_onnx_model.py
new file mode 100755
index 0000000..71a592a
--- /dev/null
+++ b/deploy/ios_demo/scripts/quantize_onnx_model.py
@@ -0,0 +1,385 @@
+#!/usr/bin/env python3
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Quantize a bundled PaddleOCR ONNX model.
+
+Reads ``inference.onnx`` from ``--input-model-dir``, writes quantized ``inference.onnx`` to
+``--output-model-dir``. When input and output directories differ, copies ``inference.yml`` from
+source to destination after a successful quantize. In-place mode (same directory) only
+replaces ``inference.onnx``.
+
+Examples:
+  # Dynamic (weights int8); no calibration data
+  python quantize_onnx_model.py --input-model-dir ./PaddleOCRDemo/Models/det \\
+    --output-model-dir ./out/det_q --mode dynamic
+
+  # Static; one .npy file per calibration sample (float32, model input shape)
+  python quantize_onnx_model.py --input-model-dir ./Models/rec \\
+    --output-model-dir ./Models/rec_int8 --mode static \\
+    --calib-data-dir ./my_calib_npy
+"""
+
+from __future__ import annotations
+
+import argparse
+import os
+import shutil
+import sys
+import tempfile
+from pathlib import Path
+from typing import Any
+
+_script_dir = Path(__file__).resolve().parent
+if str(_script_dir) not in sys.path:
+    sys.path.insert(0, str(_script_dir))
+
+from utils import build_npy_dir_reader, die
+
+
+def _same_file(a: Path, b: Path) -> bool:
+    try:
+        return a.resolve() == b.resolve()
+    except OSError:
+        return False
+
+
+def _run_ort_quant_pre_process(src_onnx: Path, work_dir: Path) -> Path:
+    """Run ORT *quant_pre_process*; write a new ONNX with ``onnx.quant.pre_process`` metadata.
+
+    Tries, in order: default → ``skip_symbolic_shape`` (some det/NMS graphs fail symbolic infer)
+    → also ``skip_optimization`` (last resort). Removes any failed output before retrying.
+    """
+    try:
+        from onnxruntime.quantization import quant_pre_process
+    except ImportError as e:
+        die(
+            f"onnxruntime.quantization.quant_pre_process is required for --ort-preprocess: {e}"
+        )
+
+    fd, raw = tempfile.mkstemp(
+        suffix=".pre.onnx",
+        prefix="inference.ort_",
+        dir=str(work_dir),
+    )
+    os.close(fd)
+    out = Path(raw)
+
+    attempts: list[tuple[str, dict[str, bool]]] = [
+        (
+            "(full symbolic + ORT optimize)",
+            {"skip_symbolic_shape": False, "skip_optimization": False},
+        ),
+        (
+            "(skip symbolic shape, keep ORT optimize)",
+            {"skip_symbolic_shape": True, "skip_optimization": False},
+        ),
+        (
+            "(skip symbolic shape, skip ORT graph optimization)",
+            {"skip_symbolic_shape": True, "skip_optimization": True},
+        ),
+    ]
+    last_err: Exception | None = None
+    for label, kwargs in attempts:
+        out.unlink(missing_ok=True)
+        try:
+            quant_pre_process(
+                input_model=str(src_onnx),
+                output_model_path=str(out),
+                **kwargs,
+            )
+        except Exception as e:
+            last_err = e
+            continue
+        if label != attempts[0][0]:
+            print(
+                f"warning: ORT quant_pre_process succeeded with {label}.",
+                file=sys.stderr,
+            )
+        return out
+
+    out.unlink(missing_ok=True)
+    die(
+        f"ORT quant_pre_process failed after {len(attempts)} attempt(s). Last error: {last_err!r}. "
+        "You can try again with --no-ort-preprocess to quantize the original model only."
+    )
+
+
+def _quantize_dynamic(src_onnx: Path, dst_onnx: Path, per_channel: bool) -> None:
+    from onnxruntime.quantization import QuantType, quantize_dynamic
+
+    quantize_dynamic(
+        model_input=str(src_onnx),
+        model_output=str(dst_onnx),
+        weight_type=QuantType.QInt8,
+        per_channel=per_channel,
+    )
+
+
+def _quantize_static(
+    src_onnx: Path,
+    dst_onnx: Path,
+    calib_dir: Path,
+    per_channel: bool,
+    calibrate_method_name: str,
+) -> None:
+    from onnxruntime.quantization import (
+        CalibrationMethod,
+        QuantFormat,
+        QuantType,
+        quantize_static,
+    )
+
+    reader = build_npy_dir_reader(src_onnx, calib_dir)
+    try:
+        method = getattr(CalibrationMethod, calibrate_method_name)
+    except AttributeError:
+        die(
+            f"unknown calibration method {calibrate_method_name!r}; "
+            f"valid names: {', '.join(CalibrationMethod.__members__)}"
+        )
+    quantize_static(
+        model_input=str(src_onnx),
+        model_output=str(dst_onnx),
+        calibration_data_reader=reader,
+        quant_format=QuantFormat.QDQ,
+        activation_type=QuantType.QUInt8,
+        weight_type=QuantType.QInt8,
+        per_channel=per_channel,
+        calibrate_method=method,
+    )
+
+
+def _default_domain_onnx_opset(m: Any) -> int:
+    """Max declared opset for default / ``ai.onnx`` imports."""
+    v = 0
+    for oi in m.opset_import:
+        dom = oi.domain or ""
+        if dom in ("", "ai.onnx"):
+            v = max(v, int(oi.version))
+    return v
+
+
+def _try_onnx_opset_via_version_converter(path: Path, target_opset: int) -> str:
+    import onnx
+    from onnx import version_converter
+
+    m = onnx.load(str(path), load_external_data=True)
+    cur = _default_domain_onnx_opset(m)
+    if cur >= target_opset:
+        return "skipped_already_ge_target"
+    try:
+        m2 = version_converter.convert_version(m, target_opset)
+    except Exception as e:
+        print(
+            f"warning: onnx.version_converter.convert_version(..., {target_opset}) failed: {e!r}. "
+            "Full checker may still fail. Try a newer `onnx`, or use --no-verify if ORT loads the model.",
+            file=sys.stderr,
+        )
+        return "convert_failed"
+    try:
+        onnx.save(m2, str(path))
+    except Exception as e:
+        die(f"failed to write ONNX after version conversion: {e}")
+    return "converted"
+
+
+def _verify_onnx_file(path: Path) -> None:
+    """Validate the output with the ONNX checker (avoids ORT IR / build skew in the host venv)."""
+    import onnx
+
+    m = onnx.load(str(path), load_external_data=True)
+    try:
+        onnx.checker.check_model(m, full_check=True)
+    except Exception as e:
+        die(f"output model failed ONNX checker validation: {e}")
+
+
+def _atomic_replace(src: Path, dst: Path) -> None:
+    os.replace(str(src), str(dst))
+
+
+def main() -> None:
+    p = argparse.ArgumentParser(description="Quantize PaddleOCR ONNX model.")
+    p.add_argument(
+        "--input-model-dir",
+        required=True,
+        type=Path,
+        help="Input model directory",
+    )
+    p.add_argument(
+        "--output-model-dir",
+        required=True,
+        type=Path,
+        help="Output model directory",
+    )
+    p.add_argument(
+        "--mode",
+        required=True,
+        choices=("dynamic", "static"),
+        help="dynamic: weight-only (quantize_dynamic). static: QDQ (quantize_static, needs --calib-data-dir).",
+    )
+    p.add_argument(
+        "--calib-data-dir",
+        type=Path,
+        default=None,
+        help="Directory of float32 .npy calibration samples (static mode only; one tensor per file).",
+    )
+    p.add_argument(
+        "--per-channel",
+        action=argparse.BooleanOptionalAction,
+        default=True,
+        help="Per-channel weight quantization (default: true).",
+    )
+    p.add_argument(
+        "--calibration-method",
+        default="MinMax",
+        help="ORT CalibrationMethod name (e.g. MinMax, Entropy, Percentile).",
+    )
+    p.add_argument(
+        "--no-verify",
+        action="store_true",
+        help="Skip ONNX checker validation of the output model after quantization.",
+    )
+    p.add_argument(
+        "--ort-preprocess",
+        action=argparse.BooleanOptionalAction,
+        default=True,
+        help=(
+            "Before quantize_static / quantize_dynamic, run ORT quant_pre_process (shape infer + "
+            "optional graph optimization) and attach onnx.quant metadata."
+        ),
+    )
+    p.add_argument(
+        "--onnx-opset-convert",
+        action=argparse.BooleanOptionalAction,
+        default=True,
+        help=(
+            "After ORT writes the output, run onnx.version_converter when the graph's declared "
+            "default-domain opset is below --onnx-target-opset. Use --no-onnx-opset-convert to keep "
+            "raw ORT output."
+        ),
+    )
+    p.add_argument(
+        "--onnx-target-opset",
+        type=int,
+        default=13,
+        metavar="N",
+        help=(
+            "Target ONNX opset for --onnx-opset-convert (default: 13). Ignored when conversion is off."
+        ),
+    )
+    args = p.parse_args()
+
+    if args.mode == "static" and args.calib_data_dir is None:
+        die("static mode requires --calib-data-dir")
+    if args.mode == "dynamic" and args.calib_data_dir is not None:
+        print("warning: --calib-data-dir is ignored for dynamic mode", file=sys.stderr)
+    if args.onnx_target_opset < 1:
+        die("--onnx-target-opset must be >= 1")
+
+    input_model_dir: Path = args.input_model_dir
+    output_model_dir: Path = args.output_model_dir
+    if not input_model_dir.is_dir():
+        die(f"input directory does not exist: {input_model_dir}")
+
+    src_onnx = input_model_dir / "inference.onnx"
+    src_yml = input_model_dir / "inference.yml"
+    if not src_onnx.is_file():
+        die(f"missing {src_onnx}")
+    if not src_yml.is_file():
+        die(f"missing {src_yml} (expected alongside inference.onnx)")
+
+    out_onnx = output_model_dir / "inference.onnx"
+    in_place = _same_file(input_model_dir, output_model_dir)
+    if not in_place:
+        output_model_dir.mkdir(parents=True, exist_ok=True)
+
+    pre_onnx: Path | None = None
+    if args.ort_preprocess:
+        pre_onnx = _run_ort_quant_pre_process(src_onnx, input_model_dir)
+    quant_src = pre_onnx if pre_onnx is not None else src_onnx
+
+    try:
+        if in_place:
+            fd, tmp_name = tempfile.mkstemp(
+                prefix="inference.onnx.",
+                suffix=".tmp",
+                dir=str(input_model_dir),
+            )
+            os.close(fd)
+            tmp_path = Path(tmp_name)
+            try:
+                if args.mode == "dynamic":
+                    _quantize_dynamic(quant_src, tmp_path, per_channel=args.per_channel)
+                else:
+                    assert args.calib_data_dir is not None
+                    _quantize_static(
+                        quant_src,
+                        tmp_path,
+                        args.calib_data_dir,
+                        per_channel=args.per_channel,
+                        calibrate_method_name=args.calibration_method,
+                    )
+                _atomic_replace(tmp_path, out_onnx)
+            finally:
+                if tmp_path.is_file() and not _same_file(tmp_path, out_onnx):
+                    try:
+                        tmp_path.unlink()
+                    except OSError:
+                        pass
+        else:
+            try:
+                if args.mode == "dynamic":
+                    _quantize_dynamic(quant_src, out_onnx, per_channel=args.per_channel)
+                else:
+                    assert args.calib_data_dir is not None
+                    _quantize_static(
+                        quant_src,
+                        out_onnx,
+                        args.calib_data_dir,
+                        per_channel=args.per_channel,
+                        calibrate_method_name=args.calibration_method,
+                    )
+                shutil.copy2(src_yml, output_model_dir / "inference.yml")
+            except Exception:
+                if out_onnx.is_file():
+                    try:
+                        out_onnx.unlink()
+                    except OSError:
+                        pass
+                raise
+    finally:
+        if pre_onnx is not None and pre_onnx.is_file():
+            try:
+                pre_onnx.unlink()
+            except OSError:
+                pass
+
+    if args.onnx_opset_convert:
+        st = _try_onnx_opset_via_version_converter(out_onnx, args.onnx_target_opset)
+        if st == "converted":
+            print(
+                f"note: applied onnx.version_converter to opset {args.onnx_target_opset}.",
+                file=sys.stderr,
+            )
+    if not args.no_verify:
+        _verify_onnx_file(out_onnx)
+
+    extra = f" and copied inference.yml" if not in_place else ""
+    print(f"Wrote {out_onnx}{extra}")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/deploy/ios_demo/scripts/run_benchmark.sh b/deploy/ios_demo/scripts/run_benchmark.sh
new file mode 100755
index 0000000..6797084
--- /dev/null
+++ b/deploy/ios_demo/scripts/run_benchmark.sh
@@ -0,0 +1,896 @@
+#!/usr/bin/env bash
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+# End-to-end benchmark runner for the iOS demo.
+#
+# Pipeline: preflight → resolve-image → resolve-destination →
+#           optional accuracy-precheck → benchmark xcodebuild-test → extract-artifacts → report
+#
+# Requires: bash 3.2+, xcodebuild, xcrun, python3.
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+IOS_DEMO_ROOT="$(cd "${SCRIPT_DIR}/.." && pwd)"
+# Fixtures/: images shipped for benchmark runs; `local-*` entries come from --image; auto-pick ignores `local-*`.
+FIXTURES_DIR="${IOS_DEMO_ROOT}/PaddleOCRDemoTests/Fixtures"
+DEFAULT_SIMULATOR="iPhone 16"
+BUILD_CONFIGURATION="Release"
+# Comma-separated xcodebuild -only-testing targets.
+TESTING_SCOPE="${PADDLEOCR_BENCHMARK_ONLY_TESTING_SCOPE:-PaddleOCRDemoTests/OCRBenchmarkTests/testOnDeviceLatencyBenchmark,PaddleOCRDemoTests/OCRBenchmarkTests/testOnDeviceMemoryBenchmark}"
+XCODE_TEST_SINGLE_RUN_FLAGS=(
+  -parallel-testing-enabled NO
+  -collect-test-diagnostics never
+)
+
+UDID=""
+SIMULATOR=""
+IMAGE=""
+FIXTURE_ARG=""
+WARMUP_CLI=""
+MEASURED_CLI=""
+ORT_EP_CLI=""
+THREADS_CLI=""
+XNNPACK_THREADS_CLI=""
+REC_BATCH_SIZE_CLI=""
+COREML_OPTIONS_CLI=""
+ORT_PROFILING_CLI=0
+ACCURACY_CHECK=0
+ACCURACY_STOP_ON_FAILURE=0
+ACCURACY_REFERENCE_JSON=""
+OUT_DIR="${IOS_DEMO_ROOT}/out"
+CLEAN=0
+
+usage() {
+  cat <<EOF
+Usage: ./scripts/run_benchmark.sh [OPTIONS]
+
+Options:
+  --udid <id>           Real-device UDID (preferred real-device path)
+  --simulator <name>    Simulator name (default: ${DEFAULT_SIMULATOR})
+  --image <path>        Ad-hoc benchmark image; copied as Fixtures/local-<basename>.
+                        Without --image/--fixture, see image selection below.
+  --fixture <name>      Use a file already under PaddleOCRDemoTests/Fixtures/ (stem or
+                        file name, e.g. ios_ocr_benchmark_reference or .jpg). Useful when
+                        several fixtures exist.
+  --warmup <n>          Benchmark warmup iterations (non-negative int).
+  --measured-iterations <n>  Timed benchmark iterations.
+  --ort-execution-provider <NAME>  Preferred ONNX Runtime EP preset for tests:
+                        CPU, XNNPACK, or CORE_ML.
+  --threads <n>         Session-level intra-op thread count (positive int).
+  --xnnpack-threads <n> XNNPACK EP thread count (positive int).
+  --rec-batch-size <n>  Recognition batch size override (positive int).
+  --coreml-options <flags>  Comma-separated CoreML EP flags: enableOnSubgraphs,
+                        createMLProgram, staticInputShapes, cpuOnly, cpuAndGPU, aneOnly.
+  --ort-profiling       Enable ONNX Runtime session profiling.
+  --accuracy-check      Run an accuracy precheck before benchmark tests.
+  --stop-on-accuracy-failure
+                        With --accuracy-check, treat accuracy FAIL as blocking.
+  --accuracy-reference-json <path>
+                        Existing OCR reference JSON for --accuracy-check. If omitted,
+                        the script generates accuracy-reference.json under --output-dir.
+  --output-dir <dir>       Output directory (default: out/)
+  --clean               Delete Fixtures/local-* and prior non-log artifacts under <output-dir>
+                        before running.
+  -h, --help            Show help
+
+Environment (optional; CLI wins when both are set):
+  PADDLEOCR_BENCHMARK_IMAGE_NAME   Select which bundled fixture to use (same as --fixture)
+  PADDLEOCR_BENCHMARK_WARMUP_ITERATIONS
+  PADDLEOCR_BENCHMARK_MEASURED_ITERATIONS
+  PADDLEOCR_BENCHMARK_ORT_EXECUTION_PROVIDER  CPU, XNNPACK, or CORE_ML
+  PADDLEOCR_BENCHMARK_INTRA_OP_THREADS    Session-level thread count
+  PADDLEOCR_BENCHMARK_XNNPACK_THREADS     XNNPACK EP thread count
+  PADDLEOCR_BENCHMARK_REC_BATCH_SIZE      Recognition batch size
+  PADDLEOCR_BENCHMARK_COREML_OPTIONS      Comma-separated CoreML EP flags
+  PADDLEOCR_BENCHMARK_ORT_PROFILING   Set to 1/true/yes/on to emit ORT JSON profiles
+  PADDLEOCR_BENCHMARK_ONLY_TESTING_SCOPE   Optional comma-separated xcodebuild -only-testing scopes.
+
+Image selection when --image is not used:
+  1) --fixture <name>, if given, else PADDLEOCR_BENCHMARK_IMAGE_NAME, if set
+  2) otherwise exactly one non-local-* file in Fixtures/
+EOF
+}
+
+die() {
+  printf "[run_benchmark] ERROR: %s\n" "$1" >&2
+  printf "  Where: %s\n" "$2" >&2
+  printf "  Next:  %s\n" "$3" >&2
+  exit 1
+}
+
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    --udid) UDID="$2"; shift 2 ;;
+    --simulator) SIMULATOR="$2"; shift 2 ;;
+    --image) IMAGE="$2"; shift 2 ;;
+    --fixture) FIXTURE_ARG="$2"; shift 2 ;;
+    --warmup) WARMUP_CLI="$2"; shift 2 ;;
+    --measured-iterations) MEASURED_CLI="$2"; shift 2 ;;
+    --ort-execution-provider) ORT_EP_CLI="$2"; shift 2 ;;
+    --threads) THREADS_CLI="$2"; shift 2 ;;
+    --xnnpack-threads) XNNPACK_THREADS_CLI="$2"; shift 2 ;;
+    --rec-batch-size) REC_BATCH_SIZE_CLI="$2"; shift 2 ;;
+    --coreml-options) COREML_OPTIONS_CLI="$2"; shift 2 ;;
+    --ort-profiling) ORT_PROFILING_CLI=1; shift ;;
+    --accuracy-check) ACCURACY_CHECK=1; shift ;;
+    --stop-on-accuracy-failure) ACCURACY_STOP_ON_FAILURE=1; shift ;;
+    --accuracy-reference-json) ACCURACY_REFERENCE_JSON="$2"; shift 2 ;;
+    --output-dir) OUT_DIR="$2"; shift 2 ;;
+    --clean) CLEAN=1; shift ;;
+    -h|--help) usage; exit 0 ;;
+    *) die "Unknown option: $1" "argument parsing" "See --help." ;;
+  esac
+done
+
+if [[ "${ACCURACY_STOP_ON_FAILURE}" -eq 1 ]]; then
+  [[ "${ACCURACY_CHECK}" -eq 1 ]] \
+    || die "--stop-on-accuracy-failure requires --accuracy-check" "argument parsing" "Add --accuracy-check, or remove --stop-on-accuracy-failure."
+fi
+
+if [[ -n "${ACCURACY_REFERENCE_JSON}" ]]; then
+  [[ "${ACCURACY_CHECK}" -eq 1 ]] \
+    || die "--accuracy-reference-json requires --accuracy-check" "argument parsing" "Add --accuracy-check, or remove --accuracy-reference-json."
+  [[ -f "${ACCURACY_REFERENCE_JSON}" ]] \
+    || die "Accuracy reference JSON not found: ${ACCURACY_REFERENCE_JSON}" "argument parsing" "Pass an existing OCR JSON reference file."
+fi
+
+# Fixture label: --fixture overrides PADDLEOCR_BENCHMARK_IMAGE_NAME.
+EXPLICIT_FIXTURE="${FIXTURE_ARG:-${PADDLEOCR_BENCHMARK_IMAGE_NAME:-}}"
+
+# Warmup / measured: CLI over env; forward to xcodebuild only when non-empty.
+WARMUP_MERGED="${WARMUP_CLI:-${PADDLEOCR_BENCHMARK_WARMUP_ITERATIONS:-}}"
+MEASURED_MERGED="${MEASURED_CLI:-${PADDLEOCR_BENCHMARK_MEASURED_ITERATIONS:-}}"
+if [[ -n "${WARMUP_MERGED}" ]]; then
+  [[ "${WARMUP_MERGED}" =~ ^[0-9]+$ ]] \
+    || die "Invalid --warmup / PADDLEOCR_BENCHMARK_WARMUP_ITERATIONS: ${WARMUP_MERGED}" "argument parsing" "Use a non-negative integer."
+fi
+if [[ -n "${MEASURED_MERGED}" ]]; then
+  [[ "${MEASURED_MERGED}" =~ ^[0-9]+$ ]] \
+    || die "Invalid --measured-iterations / PADDLEOCR_BENCHMARK_MEASURED_ITERATIONS: ${MEASURED_MERGED}" "argument parsing" "Use a non-negative integer."
+fi
+
+# ORT execution provider preset: CPU, XNNPACK, or CORE_ML; forward when set (flag or env).
+ORT_EP_MERGED="${ORT_EP_CLI:-${PADDLEOCR_BENCHMARK_ORT_EXECUTION_PROVIDER:-}}"
+ORT_EP_CANON=""
+if [[ -n "${ORT_EP_MERGED}" ]]; then
+  _ort_ep_lc="$(printf '%s' "${ORT_EP_MERGED}" | tr '[:upper:]' '[:lower:]')"
+  case "${_ort_ep_lc}" in
+    core_ml) ORT_EP_CANON="CORE_ML" ;;
+    xnnpack) ORT_EP_CANON="XNNPACK" ;;
+    cpu) ORT_EP_CANON="CPU" ;;
+    *)
+      die "Invalid --ort-execution-provider / PADDLEOCR_BENCHMARK_ORT_EXECUTION_PROVIDER: ${ORT_EP_MERGED}" "argument parsing" "Use CPU, XNNPACK, or CORE_ML."
+      ;;
+  esac
+fi
+
+# Tuning parameters.
+THREADS_MERGED="${THREADS_CLI:-${PADDLEOCR_BENCHMARK_INTRA_OP_THREADS:-}}"
+XNNPACK_THREADS_MERGED="${XNNPACK_THREADS_CLI:-${PADDLEOCR_BENCHMARK_XNNPACK_THREADS:-}}"
+REC_BATCH_SIZE_MERGED="${REC_BATCH_SIZE_CLI:-${PADDLEOCR_BENCHMARK_REC_BATCH_SIZE:-}}"
+COREML_OPTIONS_MERGED="${COREML_OPTIONS_CLI:-${PADDLEOCR_BENCHMARK_COREML_OPTIONS:-}}"
+if [[ -n "${THREADS_MERGED}" ]]; then
+  [[ "${THREADS_MERGED}" =~ ^[0-9]+$ && "${THREADS_MERGED}" -gt 0 ]] \
+    || die "Invalid --threads: ${THREADS_MERGED}" "argument parsing" "Use a positive integer."
+fi
+if [[ -n "${XNNPACK_THREADS_MERGED}" ]]; then
+  [[ "${XNNPACK_THREADS_MERGED}" =~ ^[0-9]+$ && "${XNNPACK_THREADS_MERGED}" -gt 0 ]] \
+    || die "Invalid --xnnpack-threads: ${XNNPACK_THREADS_MERGED}" "argument parsing" "Use a positive integer."
+fi
+if [[ -n "${REC_BATCH_SIZE_MERGED}" ]]; then
+  [[ "${REC_BATCH_SIZE_MERGED}" =~ ^[0-9]+$ && "${REC_BATCH_SIZE_MERGED}" -gt 0 ]] \
+    || die "Invalid --rec-batch-size: ${REC_BATCH_SIZE_MERGED}" "argument parsing" "Use a positive integer."
+fi
+
+if [[ "${ORT_PROFILING_CLI}" -eq 1 ]]; then
+  ORT_PROFILING_MERGED="1"
+else
+  ORT_PROFILING_MERGED="${PADDLEOCR_BENCHMARK_ORT_PROFILING:-}"
+fi
+
+LOGS_DIR="${OUT_DIR}/logs"
+STATUS_PATH="${OUT_DIR}/run-status.json"
+REPORT_PATH="${OUT_DIR}/benchmark-report.md"
+XCTEST_METRICS_PATH="${OUT_DIR}/xctest-memory-metrics.json"
+ACCURACY_RESULT_PATH="${OUT_DIR}/accuracy-result.xcresult"
+LATENCY_RESULT_PATH="${OUT_DIR}/latency-result.xcresult"
+MEMORY_RESULT_PATH="${OUT_DIR}/memory-result.xcresult"
+ACCURACY_REFERENCE_OUT="${OUT_DIR}/accuracy-reference.json"
+ACCURACY_IOS_EXPORT_PATH="${OUT_DIR}/ios-ocr-export.json"
+ACCURACY_SUMMARY_PATH="${OUT_DIR}/accuracy-summary.json"
+
+mkdir -p "${OUT_DIR}" "${LOGS_DIR}"
+
+if [[ "${CLEAN}" -eq 1 ]]; then
+  rm -f "${FIXTURES_DIR}"/local-*
+  find "${OUT_DIR}" -mindepth 1 -not -path "${LOGS_DIR}" -not -path "${LOGS_DIR}/*" -exec rm -rf {} + 2>/dev/null || true
+  mkdir -p "${LOGS_DIR}"
+fi
+
+# ---------- Step state (parallel indexed arrays) ----------
+
+STEP_NAMES=(preflight resolve-image resolve-destination accuracy-precheck xcodebuild-test extract-artifacts report)
+STEP_STATUS=()
+STEP_DURATION_MS=()
+STEP_EXIT=()
+STEP_REASON=()
+STEP_LOG=()
+
+for i in "${!STEP_NAMES[@]}"; do
+  STEP_STATUS[$i]=pending
+  STEP_DURATION_MS[$i]=""
+  STEP_EXIT[$i]=""
+  STEP_REASON[$i]=""
+  STEP_LOG[$i]=""
+done
+
+step_index() {
+  local want=$1 i
+  for i in "${!STEP_NAMES[@]}"; do
+    [[ "${STEP_NAMES[$i]}" == "$want" ]] && { echo "$i"; return 0; }
+  done
+  return 1
+}
+
+step_status_set() { local i; i=$(step_index "$1") || return 1; STEP_STATUS[$i]="$2"; }
+step_status_get() { local i; i=$(step_index "$1") || return 1; echo "${STEP_STATUS[$i]}"; }
+step_duration_set() { local i; i=$(step_index "$1") || return 1; STEP_DURATION_MS[$i]="$2"; }
+step_exit_set() { local i; i=$(step_index "$1") || return 1; STEP_EXIT[$i]="$2"; }
+step_reason_set() { local i; i=$(step_index "$1") || return 1; STEP_REASON[$i]="$2"; }
+step_log_set() { local i; i=$(step_index "$1") || return 1; STEP_LOG[$i]="$2"; }
+
+HALTED=0
+HALT_REASON=""
+CURRENT_STEP_NAME=""
+CURRENT_STEP_START_MS=""
+ACTIVE_CHILD_PID=""
+ACTIVE_CHILD_LABEL=""
+IMAGE_NAME=""
+IMAGE_SRC=""
+IMAGE_SOURCE=""
+DEST=""
+IS_SIMULATOR=1
+RUN_STARTED="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
+ACCURACY_STATUS="SKIPPED"
+ACCURACY_REFERENCE_USED=""
+ACCURACY_SUMMARY_USED=""
+ACCURACY_REASON=""
+MODEL_PRESET="unknown"
+DET_MODEL_PATH=""
+DET_MODEL_FORMAT=""
+DET_MODEL_SIZE_BYTES=""
+REC_MODEL_PATH=""
+REC_MODEL_FORMAT=""
+REC_MODEL_SIZE_BYTES=""
+TOTAL_MODEL_SIZE_BYTES=""
+APP_BINARY_SIZE_BYTES=""
+
+now_ms() {
+  python3 -c 'import time; print(int(time.time()*1000))'
+}
+
+file_size_bytes() {
+  [[ -f "$1" ]] || { echo ""; return 0; }
+  stat -f%z "$1" 2>/dev/null || echo ""
+}
+
+first_model_file() {
+  local dir="$1" f matches
+  for f in "${dir}/inference.ort" "${dir}/inference.with_runtime_opt.ort"; do
+    [[ -f "$f" ]] && { echo "$f"; return 0; }
+  done
+  matches=("${dir}"/inference*.ort)
+  if [[ -f "${matches[0]:-}" ]]; then
+    printf "%s\n" "${matches[@]}" | sort | head -n 1
+    return 0
+  fi
+  [[ -f "${dir}/inference.onnx" ]] && { echo "${dir}/inference.onnx"; return 0; }
+  echo ""
+}
+
+yml_model_name() {
+  local path="$1"
+  python3 - "$path" <<'PY'
+import re
+import sys
+from pathlib import Path
+
+path = Path(sys.argv[1])
+if not path.is_file():
+    print("")
+    raise SystemExit(0)
+for line in path.read_text(encoding="utf-8", errors="replace").splitlines():
+    m = re.match(r"\s*model_name:\s*['\"]?([^'\"]+?)['\"]?\s*$", line)
+    if m:
+        print(m.group(1).strip())
+        break
+else:
+    print("")
+PY
+}
+
+infer_model_preset() {
+  local det_name="$1" rec_name="$2" joined
+  joined="$(printf "%s %s" "$det_name" "$rec_name" | tr '[:upper:]' '[:lower:]')"
+  case "$joined" in
+    *small*) echo "PP-OCRv6_small" ;;
+    *tiny*) echo "PP-OCRv6_tiny" ;;
+    *mobile*) echo "PP-OCRv5_mobile" ;;
+    *) echo "unknown" ;;
+  esac
+}
+
+collect_model_metadata() {
+  local det_name rec_name
+  DET_MODEL_PATH="$(first_model_file "${IOS_DEMO_ROOT}/PaddleOCRDemo/Models/det")"
+  REC_MODEL_PATH="$(first_model_file "${IOS_DEMO_ROOT}/PaddleOCRDemo/Models/rec")"
+  DET_MODEL_FORMAT="${DET_MODEL_PATH##*.}"
+  REC_MODEL_FORMAT="${REC_MODEL_PATH##*.}"
+  DET_MODEL_SIZE_BYTES="$(file_size_bytes "$DET_MODEL_PATH")"
+  REC_MODEL_SIZE_BYTES="$(file_size_bytes "$REC_MODEL_PATH")"
+  if [[ -n "$DET_MODEL_SIZE_BYTES" && -n "$REC_MODEL_SIZE_BYTES" ]]; then
+    TOTAL_MODEL_SIZE_BYTES=$((DET_MODEL_SIZE_BYTES + REC_MODEL_SIZE_BYTES))
+  fi
+  det_name="$(yml_model_name "${IOS_DEMO_ROOT}/PaddleOCRDemo/Models/det/inference.yml")"
+  rec_name="$(yml_model_name "${IOS_DEMO_ROOT}/PaddleOCRDemo/Models/rec/inference.yml")"
+  MODEL_PRESET="$(infer_model_preset "$det_name" "$rec_name")"
+}
+
+collect_app_binary_size() {
+  local settings target_build_dir executable_path app_binary
+  settings="$(xcodebuild -showBuildSettings -workspace "${IOS_DEMO_ROOT}/PaddleOCRDemo.xcworkspace" -scheme PaddleOCRDemo -configuration "${BUILD_CONFIGURATION}" -destination "${DEST}" 2>/dev/null || true)"
+  target_build_dir="$(printf "%s\n" "$settings" | awk -F'= ' '/ TARGET_BUILD_DIR = / {print $2; exit}')"
+  executable_path="$(printf "%s\n" "$settings" | awk -F'= ' '/ EXECUTABLE_PATH = / {print $2; exit}')"
+  if [[ -n "$target_build_dir" && -n "$executable_path" ]]; then
+    app_binary="${target_build_dir}/${executable_path}"
+    APP_BINARY_SIZE_BYTES="$(file_size_bytes "$app_binary")"
+  fi
+}
+
+record_step_ok() {
+  local name="$1" duration="$2"
+  step_status_set "$name" ok
+  step_duration_set "$name" "$duration"
+  step_exit_set "$name" 0
+  printf "  -> OK (%sms)\n" "$duration"
+}
+
+record_step_fail() {
+  local name="$1" duration="$2" rc="$3" reason="$4"
+  step_status_set "$name" fail
+  step_duration_set "$name" "$duration"
+  step_exit_set "$name" "$rc"
+  step_reason_set "$name" "$reason"
+  step_log_set "$name" "logs/${name}.log"
+  HALTED=1
+  HALT_REASON="$reason"
+  printf "  -> ERROR (%sms, exit %s): %s\n" "$duration" "$rc" "$reason" >&2
+}
+
+terminate_process_tree() {
+  local pid="$1" sig="$2" child
+  while IFS= read -r child; do
+    [[ -n "$child" ]] && terminate_process_tree "$child" "$sig"
+  done < <(pgrep -P "$pid" 2>/dev/null || true)
+  kill "-${sig}" "$pid" 2>/dev/null || true
+}
+
+handle_signal() {
+  local sig="$1" rc=130 reason end duration
+  [[ "$sig" == "TERM" ]] && rc=143
+  trap - INT TERM
+  reason="interrupted by SIG${sig}"
+  printf "\n[run_benchmark] %s\n" "$reason" >&2
+  if [[ -n "${ACTIVE_CHILD_PID}" ]] && kill -0 "${ACTIVE_CHILD_PID}" 2>/dev/null; then
+    printf "[run_benchmark] stopping %s (pid %s)\n" "${ACTIVE_CHILD_LABEL:-active child}" "${ACTIVE_CHILD_PID}" >&2
+    terminate_process_tree "${ACTIVE_CHILD_PID}" "$sig"
+    sleep 1
+    if kill -0 "${ACTIVE_CHILD_PID}" 2>/dev/null; then
+      terminate_process_tree "${ACTIVE_CHILD_PID}" TERM
+    fi
+  fi
+  if [[ -n "${CURRENT_STEP_NAME}" ]]; then
+    end="$(now_ms)"; duration=$((end - CURRENT_STEP_START_MS))
+    record_step_fail "${CURRENT_STEP_NAME}" "$duration" "$rc" "$reason"
+    mark_remaining_skipped "$reason"
+  else
+    HALTED=1
+    HALT_REASON="$reason"
+  fi
+  exit "$rc"
+}
+
+mark_remaining_skipped() {
+  local reason="$1" skipped=0 i st
+  for i in "${!STEP_NAMES[@]}"; do
+    st="${STEP_STATUS[$i]}"
+    if [[ "$st" == "pending" || "$st" == "running" ]]; then
+      if [[ "${STEP_NAMES[$i]}" != "report" ]]; then
+        STEP_STATUS[$i]="skipped"
+        STEP_REASON[$i]="$reason"
+        skipped=1
+      fi
+    fi
+  done
+  [[ $skipped -eq 1 ]] && echo "  (remaining steps marked skipped)" >&2 || true
+}
+
+write_status_json() {
+  local overall="$1" exit_code="$2"
+  local finished
+  finished="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
+  python3 - "$STATUS_PATH" "$overall" "$exit_code" "$RUN_STARTED" "$finished" \
+    "$IMAGE_NAME" "$IMAGE_SOURCE" "$DEST" "$BUILD_CONFIGURATION" "$TESTING_SCOPE" \
+    "$IS_SIMULATOR" "${ORT_EP_CANON:-CPU}" "${WARMUP_MERGED:-}" "${MEASURED_MERGED:-}" \
+    "$MODEL_PRESET" "$DET_MODEL_FORMAT" "$DET_MODEL_SIZE_BYTES" "$REC_MODEL_FORMAT" "$REC_MODEL_SIZE_BYTES" "$TOTAL_MODEL_SIZE_BYTES" "$APP_BINARY_SIZE_BYTES" \
+    "${ORT_PROFILING_MERGED:-}" "$ACCURACY_CHECK" "$ACCURACY_STATUS" \
+    "$ACCURACY_STOP_ON_FAILURE" "$ACCURACY_REFERENCE_USED" "$ACCURACY_IOS_EXPORT_PATH" "$ACCURACY_SUMMARY_USED" "$ACCURACY_REASON" \
+    "${#STEP_NAMES[@]}" \
+    "${STEP_NAMES[@]}" \
+    "${STEP_STATUS[@]}" \
+    "${STEP_DURATION_MS[@]}" \
+    "${STEP_EXIT[@]}" \
+    "${STEP_REASON[@]}" \
+    "${STEP_LOG[@]}" <<'PY'
+import json
+import sys
+
+args = sys.argv[1:]
+(
+    path,
+    overall,
+    exit_code,
+    started,
+    finished,
+    image_name,
+    image_source,
+    destination,
+    build_configuration,
+    testing_scope,
+    is_simulator,
+    ort_ep_preset,
+    warmup_iterations,
+    measured_iterations,
+    model_preset,
+    det_model_format,
+    det_model_size_bytes,
+    rec_model_format,
+    rec_model_size_bytes,
+    total_model_size_bytes,
+    app_binary_size_bytes,
+    ort_profiling,
+    accuracy_enabled,
+    accuracy_status,
+    accuracy_stop_on_failure,
+    accuracy_reference,
+    accuracy_ios_export,
+    accuracy_summary,
+    accuracy_reason,
+) = args[:29]
+n = int(args[29])
+base = 30
+names  = args[base + 0 * n : base + 1 * n]
+status = args[base + 1 * n : base + 2 * n]
+dur    = args[base + 2 * n : base + 3 * n]
+exitc  = args[base + 3 * n : base + 4 * n]
+reason = args[base + 4 * n : base + 5 * n]
+logp   = args[base + 5 * n : base + 6 * n]
+steps = []
+for i in range(n):
+    row = {"name": names[i], "status": status[i] or "pending"}
+    if dur[i]:
+        row["durationMs"] = int(dur[i])
+    if exitc[i]:
+        row["exitCode"] = int(exitc[i])
+    if reason[i]:
+        row["reason"] = reason[i]
+    if logp[i]:
+        row["logPath"] = logp[i]
+    steps.append(row)
+doc = {
+    "schemaVersion": 1,
+    "overall": overall,
+    "exitCode": int(exit_code),
+    "image": {"name": image_name, "source": image_source},
+    "destination": destination,
+    "benchmark": {
+        "buildConfiguration": build_configuration,
+        "testingScope": testing_scope,
+        "isSimulator": is_simulator == "1",
+        "ortExecutionProvider": ort_ep_preset,
+        "warmupIterations": int(warmup_iterations) if warmup_iterations else None,
+        "measuredIterations": int(measured_iterations) if measured_iterations else None,
+        "modelPreset": model_preset or "unknown",
+        "modelFiles": {
+            "detection": {
+                "format": det_model_format or None,
+                "sizeBytes": int(det_model_size_bytes) if det_model_size_bytes else None,
+            },
+            "recognition": {
+                "format": rec_model_format or None,
+                "sizeBytes": int(rec_model_size_bytes) if rec_model_size_bytes else None,
+            },
+            "totalSizeBytes": int(total_model_size_bytes) if total_model_size_bytes else None,
+        },
+        "appBinarySizeBytes": int(app_binary_size_bytes) if app_binary_size_bytes else None,
+        "ortProfilingEnabled": ort_profiling.lower() in ("1", "true", "yes", "on"),
+    },
+    "accuracy": {
+        "enabled": accuracy_enabled == "1",
+        "status": accuracy_status,
+        "stopOnFailure": accuracy_stop_on_failure == "1",
+        "referenceJson": accuracy_reference or None,
+        "iosExportJson": accuracy_ios_export if accuracy_enabled == "1" else None,
+        "summaryJson": accuracy_summary or None,
+        "reason": accuracy_reason or None,
+    },
+    "runStartedAt": started,
+    "runFinishedAt": finished,
+    "steps": steps,
+}
+with open(path, "w", encoding="utf-8") as f:
+    json.dump(doc, f, indent=2)
+PY
+}
+
+finish() {
+  local exit_code=$?
+  local overall
+  if [[ "${HALTED}" -eq 1 ]]; then
+    overall="ERROR"
+  else
+    overall="COMPLETED"
+    exit_code=0
+  fi
+  printf "\n===== SUMMARY =====\n"
+  printf "Overall:   %s\n" "${overall}"
+  printf "Exit code: %s\n" "${exit_code}"
+  if [[ "${overall}" == "COMPLETED" ]]; then
+    printf "Status:    %s\n" "${STATUS_PATH}"
+    printf "Report:    %s\n" "${REPORT_PATH}"
+  else
+    printf "ERROR: See full logs in %s\n" "${LOGS_DIR}"
+  fi
+  exit "${exit_code}"
+}
+trap finish EXIT
+trap 'handle_signal INT' INT
+trap 'handle_signal TERM' TERM
+
+run_xcodebuild() {
+  local rc
+  ACTIVE_CHILD_LABEL="$*"
+  "$@" &
+  ACTIVE_CHILD_PID=$!
+  set +e
+  wait "${ACTIVE_CHILD_PID}"
+  rc=$?
+  set -e
+  ACTIVE_CHILD_PID=""
+  ACTIVE_CHILD_LABEL=""
+  return "$rc"
+}
+
+run_step() {
+  local name="$1"; shift
+  local log="${LOGS_DIR}/${name}.log"
+  local start end duration rc tail
+  printf "\n===== [%s] step %s: %s =====\n" "$(date +%H:%M:%S)" "$name" "$*"
+  step_status_set "$name" running
+  start="$(now_ms)"
+  CURRENT_STEP_NAME="$name"
+  CURRENT_STEP_START_MS="$start"
+  if "$@" >"$log" 2>&1 </dev/null; then
+    CURRENT_STEP_NAME=""
+    CURRENT_STEP_START_MS=""
+    end="$(now_ms)"; duration=$((end - start))
+    record_step_ok "$name" "$duration"
+    return 0
+  else
+    rc=$?
+    CURRENT_STEP_NAME=""
+    CURRENT_STEP_START_MS=""
+    end="$(now_ms)"; duration=$((end - start))
+    tail="$(tail -n 5 "$log" | head -n 1 || true)"
+    record_step_fail "$name" "$duration" "$rc" "${tail:-step failed}"
+    return "$rc"
+  fi
+}
+
+generate_report_impl() {
+  python3 "${SCRIPT_DIR}/generate_benchmark_report.py" \
+    --accuracy-summary "${ACCURACY_SUMMARY_PATH}" \
+    --on-device-performance-json "${OUT_DIR}/on-device-performance.json" \
+    --run-status "${STATUS_PATH}" \
+    --xctest-metrics-json "${XCTEST_METRICS_PATH}" \
+    --output "${REPORT_PATH}"
+}
+
+# ---------- Step: preflight ----------
+preflight_impl() {
+  [[ -f "${IOS_DEMO_ROOT}/PaddleOCRDemo/Models/det/inference.yml" ]] \
+    || { echo "Models/det/inference.yml missing; run ./scripts/fetch_ios_demo_models.sh" >&2; return 1; }
+  [[ -f "${IOS_DEMO_ROOT}/PaddleOCRDemo/Models/rec/inference.yml" ]] \
+    || { echo "Models/rec/inference.yml missing; run ./scripts/fetch_ios_demo_models.sh" >&2; return 1; }
+  command -v xcodebuild >/dev/null || { echo "xcodebuild not in PATH" >&2; return 1; }
+  command -v xcrun >/dev/null || { echo "xcrun not in PATH" >&2; return 1; }
+  command -v python3 >/dev/null || { echo "python3 not in PATH" >&2; return 1; }
+  [[ -d "${FIXTURES_DIR}" ]] || { echo "Fixtures/ directory missing" >&2; return 1; }
+  collect_model_metadata
+}
+run_step preflight preflight_impl || { mark_remaining_skipped "preflight failed: ${HALT_REASON}"; exit 1; }
+
+# ---------- Step: resolve-image ----------
+resolve_image_impl() {
+  if [[ -n "$IMAGE" ]]; then
+    [[ -f "$IMAGE" ]] || { echo "--image file not found: $IMAGE" >&2; return 1; }
+    local base sanitized
+    base="$(basename "$IMAGE")"
+    sanitized="$(echo "$base" | tr -c 'A-Za-z0-9._-' '_')"
+    IMAGE_NAME="local-${sanitized}"
+    IMAGE_SRC="$IMAGE"
+    IMAGE_SOURCE="override"
+    cp -f "$IMAGE" "${FIXTURES_DIR}/${IMAGE_NAME}"
+    echo "Copied override image to ${FIXTURES_DIR}/${IMAGE_NAME}"
+    return 0
+  fi
+  if [[ -n "${EXPLICIT_FIXTURE}" ]]; then
+    local name cand
+    name="${EXPLICIT_FIXTURE}"
+    cand=""
+    if [[ -f "${FIXTURES_DIR}/${name}" ]]; then
+      cand="${FIXTURES_DIR}/${name}"
+    else
+      # stem without extension: try common image extensions
+      if [[ "$name" != *.* ]]; then
+        for ext in jpg jpeg png heic webp; do
+          if [[ -f "${FIXTURES_DIR}/${name}.${ext}" ]]; then
+            cand="${FIXTURES_DIR}/${name}.${ext}"
+            break
+          fi
+        done
+      fi
+    fi
+    [[ -n "$cand" ]] || { echo "Fixture not found for --fixture / PADDLEOCR_BENCHMARK_IMAGE_NAME: ${name}" >&2; return 1; }
+    IMAGE_NAME="$(basename "$cand")"
+    IMAGE_SRC="$cand"
+    IMAGE_SOURCE="fixture"
+    echo "Using fixture (explicit): ${IMAGE_NAME}"
+    return 0
+  fi
+  local found=()
+  while IFS= read -r -d '' f; do
+    local b; b="$(basename "$f")"
+    case "$b" in
+      local-*|.gitignore|.*) continue ;;
+      *) found+=("$b") ;;
+    esac
+  done < <(find "${FIXTURES_DIR}" -mindepth 1 -maxdepth 1 -type f -print0)
+  if [[ ${#found[@]} -eq 0 ]]; then
+    echo "No fixture in ${FIXTURES_DIR}/ and no --image provided." >&2
+    echo "Put exactly one fixture image in PaddleOCRDemoTests/Fixtures/, or pass --image <path>." >&2
+    return 1
+  fi
+  if [[ ${#found[@]} -gt 1 ]]; then
+    echo "Multiple fixtures in ${FIXTURES_DIR}/: ${found[*]}" >&2
+    echo "Pass --image <path> to select one explicitly." >&2
+    return 1
+  fi
+  IMAGE_NAME="${found[0]}"
+  IMAGE_SRC="${FIXTURES_DIR}/${IMAGE_NAME}"
+  IMAGE_SOURCE="fixture"
+  echo "Using fixture: ${IMAGE_NAME}"
+}
+run_step resolve-image resolve_image_impl || { mark_remaining_skipped "resolve-image failed: ${HALT_REASON}"; exit 1; }
+
+# ---------- Step: resolve-destination ----------
+resolve_destination_impl() {
+  if [[ -n "${UDID}" ]]; then
+    DEST="platform=iOS,id=${UDID}"
+    IS_SIMULATOR=0
+  elif [[ -n "${SIMULATOR}" ]]; then
+    DEST="platform=iOS Simulator,name=${SIMULATOR}"
+    IS_SIMULATOR=1
+  else
+    DEST="platform=iOS Simulator,name=${DEFAULT_SIMULATOR}"
+    IS_SIMULATOR=1
+  fi
+  echo "Destination: ${DEST}"
+}
+run_step resolve-destination resolve_destination_impl || { mark_remaining_skipped "resolve-destination failed: ${HALT_REASON}"; exit 1; }
+
+# ---------- Step: accuracy-precheck (skipped unless --accuracy-check is set) ----------
+accuracy_precheck_impl() {
+  ACCURACY_STATUS="ERROR"
+  ACCURACY_REASON=""
+  ACCURACY_SUMMARY_USED="${ACCURACY_SUMMARY_PATH}"
+  rm -rf "${ACCURACY_RESULT_PATH}"
+  rm -f "${ACCURACY_IOS_EXPORT_PATH}" "${ACCURACY_SUMMARY_PATH}"
+
+  if [[ -n "${ACCURACY_REFERENCE_JSON}" ]]; then
+    ACCURACY_REFERENCE_USED="${ACCURACY_REFERENCE_JSON}"
+  else
+    ACCURACY_REFERENCE_USED="${ACCURACY_REFERENCE_OUT}"
+    if ! python3 "${SCRIPT_DIR}/ocr_reference_run.py" \
+      --image "${IMAGE_SRC}" \
+      --output "${ACCURACY_REFERENCE_OUT}" \
+      --device cpu \
+      --align-ios-defaults; then
+      ACCURACY_STATUS="ERROR"
+      ACCURACY_REASON="reference generation failed"
+      echo "Accuracy precheck: ${ACCURACY_STATUS} (${ACCURACY_REASON})" >&2
+      return 1
+    fi
+  fi
+
+  local runenv=(
+    env "TEST_RUNNER_PADDLEOCR_BENCHMARK_IMAGE_NAME=${IMAGE_NAME}"
+    "TEST_RUNNER_PADDLEOCR_BENCHMARK_BUILD_CONFIGURATION=${BUILD_CONFIGURATION}"
+  )
+  if [[ -n "${ORT_EP_CANON}" ]]; then
+    runenv+=("TEST_RUNNER_PADDLEOCR_BENCHMARK_ORT_EXECUTION_PROVIDER=${ORT_EP_CANON}")
+  fi
+
+  if ! run_xcodebuild "${runenv[@]}" xcodebuild test \
+    -workspace "${IOS_DEMO_ROOT}/PaddleOCRDemo.xcworkspace" \
+    -scheme PaddleOCRDemo \
+    -configuration "${BUILD_CONFIGURATION}" \
+    -destination "${DEST}" \
+    -resultBundlePath "${ACCURACY_RESULT_PATH}" \
+    "${XCODE_TEST_SINGLE_RUN_FLAGS[@]}" \
+    -only-testing:PaddleOCRDemoTests/OCRBenchmarkTests/testOCRExportJSONSchema; then
+    ACCURACY_STATUS="ERROR"
+    ACCURACY_REASON="accuracy XCTest export failed"
+    echo "Accuracy precheck: ${ACCURACY_STATUS} (${ACCURACY_REASON})" >&2
+    return 1
+  fi
+
+  if ! python3 "${SCRIPT_DIR}/extract_xcresult_attachments.py" \
+    --result "${ACCURACY_RESULT_PATH}" \
+    --output-dir "${OUT_DIR}" \
+    --name ios-ocr-export.json; then
+    ACCURACY_STATUS="ERROR"
+    ACCURACY_REASON="accuracy attachment extraction failed"
+    echo "Accuracy precheck: ${ACCURACY_STATUS} (${ACCURACY_REASON})" >&2
+    return 1
+  fi
+
+  set +e
+  python3 "${SCRIPT_DIR}/compare_ocr_json.py" \
+    "${ACCURACY_REFERENCE_USED}" \
+    "${ACCURACY_IOS_EXPORT_PATH}" \
+    --json-summary-out "${ACCURACY_SUMMARY_PATH}"
+  local compare_rc=$?
+  set -e
+  if [[ "${compare_rc}" -eq 0 ]]; then
+    ACCURACY_STATUS="PASS"
+  elif [[ -f "${ACCURACY_SUMMARY_PATH}" ]]; then
+    ACCURACY_STATUS="FAIL"
+    ACCURACY_REASON="accuracy thresholds were not met"
+  else
+    ACCURACY_STATUS="ERROR"
+    ACCURACY_REASON="accuracy comparison failed"
+  fi
+  echo "Accuracy precheck: ${ACCURACY_STATUS}${ACCURACY_REASON:+ (${ACCURACY_REASON})}"
+  [[ "${ACCURACY_STATUS}" != "ERROR" ]]
+}
+
+if [[ "${ACCURACY_CHECK}" -eq 1 ]]; then
+  if ! run_step accuracy-precheck accuracy_precheck_impl; then
+    HALT_REASON="accuracy precheck error: ${ACCURACY_REASON:-${HALT_REASON}}"
+    step_reason_set accuracy-precheck "${HALT_REASON}"
+    mark_remaining_skipped "${HALT_REASON}"
+    write_status_json "ERROR" 1
+    generate_report_impl || true
+    exit 1
+  fi
+  if [[ "${ACCURACY_STATUS}" == "FAIL" ]]; then
+    printf "Accuracy precheck: %s (%s)\n" "${ACCURACY_STATUS}" "${ACCURACY_REASON:-accuracy thresholds were not met}"
+  fi
+  if [[ "${ACCURACY_STOP_ON_FAILURE}" -eq 1 && "${ACCURACY_STATUS}" == "FAIL" ]]; then
+    HALTED=1
+    HALT_REASON="accuracy precheck did not pass: ${ACCURACY_STATUS}"
+    step_status_set accuracy-precheck fail
+    step_exit_set accuracy-precheck 1
+    step_reason_set accuracy-precheck "${HALT_REASON}"
+    mark_remaining_skipped "${HALT_REASON}"
+    write_status_json "ERROR" 1
+    generate_report_impl || true
+    exit 1
+  fi
+else
+  step_status_set accuracy-precheck skipped
+  step_reason_set accuracy-precheck "accuracy precheck not requested"
+  ACCURACY_STATUS="SKIPPED"
+fi
+
+# ---------- Step: xcodebuild-test ----------
+xcodebuild_test_impl() {
+  rm -rf "${LATENCY_RESULT_PATH}"
+  rm -rf "${MEMORY_RESULT_PATH}"
+  rm -f "${XCTEST_METRICS_PATH}"
+  echo "xcodebuild: configuration=${BUILD_CONFIGURATION} benchmark image name=${IMAGE_NAME} warmup=${WARMUP_MERGED:-} measured=${MEASURED_MERGED:-} ort_ep=${ORT_EP_CANON:-default} ort_profiling=${ORT_PROFILING_MERGED:+on} scope=${TESTING_SCOPE}"
+  local runenv=(
+    env "TEST_RUNNER_PADDLEOCR_BENCHMARK_IMAGE_NAME=${IMAGE_NAME}"
+    "TEST_RUNNER_PADDLEOCR_BENCHMARK_BUILD_CONFIGURATION=${BUILD_CONFIGURATION}"
+  )
+  [[ -n "${WARMUP_MERGED}" ]] && runenv+=("TEST_RUNNER_PADDLEOCR_BENCHMARK_WARMUP_ITERATIONS=${WARMUP_MERGED}")
+  [[ -n "${MEASURED_MERGED}" ]] && runenv+=("TEST_RUNNER_PADDLEOCR_BENCHMARK_MEASURED_ITERATIONS=${MEASURED_MERGED}")
+  if [[ -n "${ORT_EP_CANON}" ]]; then
+    runenv+=("TEST_RUNNER_PADDLEOCR_BENCHMARK_ORT_EXECUTION_PROVIDER=${ORT_EP_CANON}")
+  fi
+  [[ -n "${ORT_PROFILING_MERGED:-}" ]] && runenv+=("TEST_RUNNER_PADDLEOCR_BENCHMARK_ORT_PROFILING=${ORT_PROFILING_MERGED}")
+  [[ -n "${THREADS_MERGED}" ]] && runenv+=("TEST_RUNNER_PADDLEOCR_BENCHMARK_INTRA_OP_THREADS=${THREADS_MERGED}")
+  [[ -n "${XNNPACK_THREADS_MERGED}" ]] && runenv+=("TEST_RUNNER_PADDLEOCR_BENCHMARK_XNNPACK_THREADS=${XNNPACK_THREADS_MERGED}")
+  [[ -n "${REC_BATCH_SIZE_MERGED}" ]] && runenv+=("TEST_RUNNER_PADDLEOCR_BENCHMARK_REC_BATCH_SIZE=${REC_BATCH_SIZE_MERGED}")
+  [[ -n "${COREML_OPTIONS_MERGED}" ]] && runenv+=("TEST_RUNNER_PADDLEOCR_BENCHMARK_COREML_OPTIONS=${COREML_OPTIONS_MERGED}")
+  local scopes=() scope result_path
+  IFS=',' read -r -a scopes <<< "${TESTING_SCOPE}"
+  for scope in "${scopes[@]}"; do
+    scope="$(printf '%s' "$scope" | sed 's/^[[:space:]]*//;s/[[:space:]]*$//')"
+    [[ -n "$scope" ]] || continue
+    result_path="${LATENCY_RESULT_PATH}"
+    if [[ "${scope}" == *"/testOnDeviceMemoryBenchmark" ]]; then
+      result_path="${MEMORY_RESULT_PATH}"
+    fi
+    rm -rf "${result_path}"
+    echo "xcodebuild: running scope=${scope} result=${result_path}"
+    if ! run_xcodebuild "${runenv[@]}" xcodebuild test \
+      -workspace "${IOS_DEMO_ROOT}/PaddleOCRDemo.xcworkspace" \
+      -scheme PaddleOCRDemo \
+      -configuration "${BUILD_CONFIGURATION}" \
+      -destination "${DEST}" \
+      -resultBundlePath "${result_path}" \
+      "${XCODE_TEST_SINGLE_RUN_FLAGS[@]}" \
+      -only-testing:"${scope}"; then
+      return 1
+    fi
+  done
+  collect_app_binary_size
+}
+run_step xcodebuild-test xcodebuild_test_impl || { mark_remaining_skipped "xcodebuild-test failed: ${HALT_REASON}"; exit 1; }
+
+# ---------- Step: extract-artifacts ----------
+extract_artifacts_impl() {
+  local metrics_result="${MEMORY_RESULT_PATH}"
+  [[ -d "${metrics_result}" ]] || metrics_result="${LATENCY_RESULT_PATH}"
+  python3 "${SCRIPT_DIR}/extract_xcresult_attachments.py" \
+    --result "${LATENCY_RESULT_PATH}" \
+    --output-dir "${OUT_DIR}" \
+    --name on-device-performance.json \
+    --optional-name ios-ocr-export.json \
+    --optional-name ort_profile_detection \
+    --optional-name ort_profile_recognition
+  python3 "${SCRIPT_DIR}/extract_xctest_metrics.py" \
+    --result "${metrics_result}" \
+    --output "${XCTEST_METRICS_PATH}"
+}
+run_step extract-artifacts extract_artifacts_impl || { mark_remaining_skipped "extract-artifacts failed: ${HALT_REASON}"; exit 1; }
+
+# ---------- Step: report ----------
+# First pass captures report step timing; write_status_json persists final step state;
+# second pass refreshes the Markdown so **Overall** and the step table match this run.
+report_impl() {
+  generate_report_impl
+}
+report_refresh() {
+  generate_report_impl
+}
+run_step report report_impl || { write_status_json "ERROR" 1; exit 1; }
+write_status_json "COMPLETED" 0
+report_refresh || true
+
+exit 0
diff --git a/deploy/ios_demo/scripts/utils.py b/deploy/ios_demo/scripts/utils.py
new file mode 100644
index 0000000..41a2956
--- /dev/null
+++ b/deploy/ios_demo/scripts/utils.py
@@ -0,0 +1,89 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Shared helpers (calibration I/O) for ``quantize_onnx_model`` / ``debug_onnx_qdq`` / ``build_onnx_calib_npy``.
+
+Entry scripts prepend this file's directory to ``sys.path`` before ``from utils import ...`` so
+``import utils`` does not depend on the process working directory. The top-level name ``utils`` is
+generic; the scripts resolve ``utils.py`` in this same ``scripts/`` directory first.
+"""
+
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+from typing import Any
+
+__all__ = [
+    "die",
+    "user_input_names",
+    "build_npy_dir_reader",
+]
+
+
+def die(msg: str) -> None:
+    print(f"error: {msg}", file=sys.stderr)
+    raise SystemExit(1)
+
+
+def user_input_names(model_path: Path) -> list[str]:
+    """Return non-initializer graph input names (user-visible inputs)."""
+    import onnx
+
+    m = onnx.load(str(model_path), load_external_data=True)
+    init = {t.name for t in m.graph.initializer}
+    return [i.name for i in m.graph.input if i.name not in init]
+
+
+def build_npy_dir_reader(
+    model_path: Path, data_dir: Path, *, max_samples: int | None = None
+) -> Any:
+    """ONNX Runtime ``CalibrationDataReader``: one float32 ``(NCHW)`` ``.npy`` per sample; sorted by filename.
+
+    *model_path* is used to discover the single graph input name (``user_input_names``); it may be
+    the original or an augmented model as long as the first user input is unchanged.
+    """
+    from onnxruntime.quantization import CalibrationDataReader
+
+    class NpyDirectoryDataReader(CalibrationDataReader):
+        def __init__(self) -> None:
+            names = user_input_names(model_path)
+            if len(names) != 1:
+                die(
+                    "expected exactly one non-initializer graph input; "
+                    f"found {len(names)}: {names}"
+                )
+            self._input_name = names[0]
+            files = sorted(data_dir.glob("*.npy"))
+            if not files:
+                die(f"no .npy files under {data_dir}")
+            if max_samples is not None:
+                files = files[: max(0, max_samples)]
+            self._it = iter(files)
+
+        def get_next(self) -> Any:
+            import numpy as np
+
+            try:
+                path = next(self._it)
+            except StopIteration:
+                return None
+            arr = np.load(str(path))
+            if not isinstance(arr, np.ndarray):
+                die(f"expected ndarray in {path}, got {type(arr)}")
+            if arr.dtype != np.float32:
+                arr = arr.astype(np.float32, copy=False)
+            return {self._input_name: arr}
+
+    return NpyDirectoryDataReader()
diff --git a/deploy/lite/Makefile b/deploy/lite/Makefile
new file mode 100644
index 0000000..9aed91a
--- /dev/null
+++ b/deploy/lite/Makefile
@@ -0,0 +1,80 @@
+ARM_ABI = arm8
+export ARM_ABI
+
+include ../Makefile.def
+
+LITE_ROOT=../../../
+
+THIRD_PARTY_DIR=${LITE_ROOT}/third_party
+
+OPENCV_VERSION=opencv4.1.0
+
+OPENCV_LIBS = ../../../third_party/${OPENCV_VERSION}/arm64-v8a/libs/libopencv_imgcodecs.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/libs/libopencv_imgproc.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/libs/libopencv_core.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/libtegra_hal.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/liblibjpeg-turbo.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/liblibwebp.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/liblibpng.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/liblibjasper.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/liblibtiff.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/libIlmImf.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/libtbb.a \
+              ../../../third_party/${OPENCV_VERSION}/arm64-v8a/3rdparty/libs/libcpufeatures.a
+
+OPENCV_INCLUDE = -I../../../third_party/${OPENCV_VERSION}/arm64-v8a/include
+
+CXX_INCLUDES = $(INCLUDES) ${OPENCV_INCLUDE} -I$(LITE_ROOT)/cxx/include
+
+CXX_LIBS = ${OPENCV_LIBS} -L$(LITE_ROOT)/cxx/lib/ -lpaddle_light_api_shared $(SYSTEM_LIBS)
+
+###############################################################
+# How to use one of static libaray:                           #
+#  `libpaddle_api_full_bundled.a`                             #
+#  `libpaddle_api_light_bundled.a`                            #
+###############################################################
+# Note: default use lite's shared library.                    #
+###############################################################
+# 1. Comment above line using `libpaddle_light_api_shared.so`
+# 2. Undo comment below line using `libpaddle_api_light_bundled.a`
+
+#CXX_LIBS = $(LITE_ROOT)/cxx/lib/libpaddle_api_light_bundled.a $(SYSTEM_LIBS)
+
+ocr_db_crnn: fetch_clipper fetch_opencv ocr_db_crnn.o crnn_process.o db_post_process.o clipper.o cls_process.o
+	$(CC) $(SYSROOT_LINK) $(CXXFLAGS_LINK) ocr_db_crnn.o crnn_process.o db_post_process.o clipper.o cls_process.o -o ocr_db_crnn  $(CXX_LIBS) $(LDFLAGS)
+
+ocr_db_crnn.o: ocr_db_crnn.cc
+	$(CC) $(SYSROOT_COMPLILE) $(CXX_DEFINES) $(CXX_INCLUDES) $(CXX_FLAGS) -o ocr_db_crnn.o -c ocr_db_crnn.cc
+
+crnn_process.o: fetch_opencv crnn_process.cc
+	$(CC) $(SYSROOT_COMPLILE) $(CXX_DEFINES) $(CXX_INCLUDES) $(CXX_FLAGS) -o crnn_process.o -c crnn_process.cc
+
+cls_process.o: fetch_opencv cls_process.cc
+	$(CC) $(SYSROOT_COMPLILE) $(CXX_DEFINES) $(CXX_INCLUDES) $(CXX_FLAGS) -o cls_process.o -c cls_process.cc
+
+db_post_process.o: fetch_clipper fetch_opencv db_post_process.cc
+	$(CC) $(SYSROOT_COMPLILE) $(CXX_DEFINES) $(CXX_INCLUDES) $(CXX_FLAGS) -o db_post_process.o -c db_post_process.cc
+
+clipper.o: fetch_clipper
+	$(CC) $(SYSROOT_COMPLILE) $(CXX_DEFINES) $(CXX_INCLUDES) $(CXX_FLAGS) -o clipper.o -c clipper.cpp
+
+fetch_clipper:
+	@test -e clipper.hpp || \
+        ( echo "Fetch clipper " && \
+        wget -c https://paddle-inference-dist.cdn.bcebos.com/PaddleLite/Clipper/clipper.hpp)
+	@ test -e clipper.cpp || \
+        wget -c https://paddle-inference-dist.cdn.bcebos.com/PaddleLite/Clipper/clipper.cpp
+
+fetch_opencv:
+	@ test -d ${THIRD_PARTY_DIR} ||  mkdir ${THIRD_PARTY_DIR}
+	@ test -e ${THIRD_PARTY_DIR}/${OPENCV_VERSION}.tar.gz || \
+      (echo "fetch opencv libs" && \
+      wget -P ${THIRD_PARTY_DIR} https://paddle-inference-dist.bj.bcebos.com/${OPENCV_VERSION}.tar.gz)
+	@ test -d ${THIRD_PARTY_DIR}/${OPENCV_VERSION} || \
+      tar -zxvf ${THIRD_PARTY_DIR}/${OPENCV_VERSION}.tar.gz -C ${THIRD_PARTY_DIR}
+
+
+.PHONY: clean
+clean:
+	rm -f ocr_db_crnn.o clipper.o db_post_process.o crnn_process.o cls_process.o
+	rm -f ocr_db_crnn
diff --git a/deploy/lite/cls_process.cc b/deploy/lite/cls_process.cc
new file mode 100644
index 0000000..591347f
--- /dev/null
+++ b/deploy/lite/cls_process.cc
@@ -0,0 +1,43 @@
+// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "cls_process.h" //NOLINT
+#include <algorithm>
+#include <memory>
+#include <string>
+
+const std::vector<int> rec_image_shape{3, 48, 192};
+
+cv::Mat ClsResizeImg(cv::Mat img) {
+  int imgC, imgH, imgW;
+  imgC = rec_image_shape[0];
+  imgH = rec_image_shape[1];
+  imgW = rec_image_shape[2];
+
+  float ratio = static_cast<float>(img.cols) / static_cast<float>(img.rows);
+
+  int resize_w, resize_h;
+  if (ceilf(imgH * ratio) > imgW)
+    resize_w = imgW;
+  else
+    resize_w = int(ceilf(imgH * ratio));
+  cv::Mat resize_img;
+  cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f,
+             cv::INTER_LINEAR);
+  if (resize_w < imgW) {
+    cv::copyMakeBorder(resize_img, resize_img, 0, 0, 0, imgW - resize_w,
+                       cv::BORDER_CONSTANT, cv::Scalar(0, 0, 0));
+  }
+  return resize_img;
+}
diff --git a/deploy/lite/cls_process.h b/deploy/lite/cls_process.h
new file mode 100644
index 0000000..7fb8ab3
--- /dev/null
+++ b/deploy/lite/cls_process.h
@@ -0,0 +1,29 @@
+// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <cstring>
+#include <fstream>
+#include <iostream>
+#include <memory>
+#include <string>
+#include <vector>
+
+#include "math.h" //NOLINT
+#include "opencv2/core.hpp"
+#include "opencv2/imgcodecs.hpp"
+#include "opencv2/imgproc.hpp"
+
+cv::Mat ClsResizeImg(cv::Mat img);
diff --git a/deploy/lite/config.txt b/deploy/lite/config.txt
new file mode 100644
index 0000000..ecbf7ad
--- /dev/null
+++ b/deploy/lite/config.txt
@@ -0,0 +1,8 @@
+max_side_len  960
+det_db_thresh  0.3
+det_db_box_thresh  0.5
+det_db_unclip_ratio  1.6
+det_db_use_dilate 0
+det_use_polygon_score 1
+use_direction_classify  1
+rec_image_height  48
diff --git a/deploy/lite/crnn_process.cc b/deploy/lite/crnn_process.cc
new file mode 100644
index 0000000..7cd95cd
--- /dev/null
+++ b/deploy/lite/crnn_process.cc
@@ -0,0 +1,118 @@
+// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "crnn_process.h" //NOLINT
+#include <algorithm>
+#include <memory>
+#include <string>
+
+const std::vector<int> rec_image_shape{3, 32, 320};
+
+cv::Mat CrnnResizeImg(cv::Mat img, float wh_ratio, int rec_image_height) {
+  int imgC, imgH, imgW;
+  imgC = rec_image_shape[0];
+  imgH = rec_image_height;
+  imgW = rec_image_shape[2];
+
+  imgW = int(imgH * wh_ratio);
+
+  float ratio = float(img.cols) / float(img.rows);
+  int resize_w, resize_h;
+
+  if (ceilf(imgH * ratio) > imgW)
+    resize_w = imgW;
+  else
+    resize_w = int(ceilf(imgH * ratio));
+  cv::Mat resize_img;
+  cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f,
+             cv::INTER_LINEAR);
+  cv::copyMakeBorder(resize_img, resize_img, 0, 0, 0,
+                     int(imgW - resize_img.cols), cv::BORDER_CONSTANT,
+                     {127, 127, 127});
+  return resize_img;
+}
+
+std::vector<std::string> ReadDict(std::string path) {
+  std::ifstream in(path);
+  std::string filename;
+  std::string line;
+  std::vector<std::string> m_vec;
+  if (in) {
+    while (getline(in, line)) {
+      m_vec.push_back(line);
+    }
+  } else {
+    std::cout << "no such file" << std::endl;
+  }
+  return m_vec;
+}
+
+cv::Mat GetRotateCropImage(cv::Mat srcimage,
+                           std::vector<std::vector<int>> box) {
+  cv::Mat image;
+  srcimage.copyTo(image);
+  std::vector<std::vector<int>> points = box;
+
+  int x_collect[4] = {box[0][0], box[1][0], box[2][0], box[3][0]};
+  int y_collect[4] = {box[0][1], box[1][1], box[2][1], box[3][1]};
+  int left = int(*std::min_element(x_collect, x_collect + 4));
+  int right = int(*std::max_element(x_collect, x_collect + 4));
+  int top = int(*std::min_element(y_collect, y_collect + 4));
+  int bottom = int(*std::max_element(y_collect, y_collect + 4));
+
+  cv::Mat img_crop;
+  image(cv::Rect(left, top, right - left, bottom - top)).copyTo(img_crop);
+
+  for (int i = 0; i < points.size(); i++) {
+    points[i][0] -= left;
+    points[i][1] -= top;
+  }
+
+  int img_crop_width =
+      static_cast<int>(sqrt(pow(points[0][0] - points[1][0], 2) +
+                            pow(points[0][1] - points[1][1], 2)));
+  int img_crop_height =
+      static_cast<int>(sqrt(pow(points[0][0] - points[3][0], 2) +
+                            pow(points[0][1] - points[3][1], 2)));
+
+  cv::Point2f pts_std[4];
+  pts_std[0] = cv::Point2f(0., 0.);
+  pts_std[1] = cv::Point2f(img_crop_width, 0.);
+  pts_std[2] = cv::Point2f(img_crop_width, img_crop_height);
+  pts_std[3] = cv::Point2f(0.f, img_crop_height);
+
+  cv::Point2f pointsf[4];
+  pointsf[0] = cv::Point2f(points[0][0], points[0][1]);
+  pointsf[1] = cv::Point2f(points[1][0], points[1][1]);
+  pointsf[2] = cv::Point2f(points[2][0], points[2][1]);
+  pointsf[3] = cv::Point2f(points[3][0], points[3][1]);
+
+  cv::Mat M = cv::getPerspectiveTransform(pointsf, pts_std);
+
+  cv::Mat dst_img;
+  cv::warpPerspective(img_crop, dst_img, M,
+                      cv::Size(img_crop_width, img_crop_height),
+                      cv::BORDER_REPLICATE);
+
+  const float ratio = 1.5;
+  if (static_cast<float>(dst_img.rows) >=
+      static_cast<float>(dst_img.cols) * ratio) {
+    cv::Mat srcCopy = cv::Mat(dst_img.rows, dst_img.cols, dst_img.depth());
+    cv::transpose(dst_img, srcCopy);
+    cv::flip(srcCopy, srcCopy, 0);
+    return srcCopy;
+  } else {
+    return dst_img;
+  }
+}
diff --git a/deploy/lite/crnn_process.h b/deploy/lite/crnn_process.h
new file mode 100644
index 0000000..ed7a316
--- /dev/null
+++ b/deploy/lite/crnn_process.h
@@ -0,0 +1,38 @@
+// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <cstring>
+#include <fstream>
+#include <iostream>
+#include <memory>
+#include <string>
+#include <vector>
+
+#include "math.h" //NOLINT
+#include "opencv2/core.hpp"
+#include "opencv2/imgcodecs.hpp"
+#include "opencv2/imgproc.hpp"
+
+cv::Mat CrnnResizeImg(cv::Mat img, float wh_ratio, int rec_image_height);
+
+std::vector<std::string> ReadDict(std::string path);
+
+cv::Mat GetRotateCropImage(cv::Mat srcimage, std::vector<std::vector<int>> box);
+
+template <class ForwardIterator>
+inline size_t Argmax(ForwardIterator first, ForwardIterator last) {
+  return std::distance(first, std::max_element(first, last));
+}
diff --git a/deploy/lite/db_post_process.cc b/deploy/lite/db_post_process.cc
new file mode 100644
index 0000000..77919fb
--- /dev/null
+++ b/deploy/lite/db_post_process.cc
@@ -0,0 +1,351 @@
+// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "db_post_process.h" // NOLINT
+#include <algorithm>
+#include <utility>
+
+void GetContourArea(std::vector<std::vector<float>> box, float unclip_ratio,
+                    float &distance) {
+  int pts_num = 4;
+  float area = 0.0f;
+  float dist = 0.0f;
+  for (int i = 0; i < pts_num; i++) {
+    area += box[i][0] * box[(i + 1) % pts_num][1] -
+            box[i][1] * box[(i + 1) % pts_num][0];
+    dist += sqrtf((box[i][0] - box[(i + 1) % pts_num][0]) *
+                      (box[i][0] - box[(i + 1) % pts_num][0]) +
+                  (box[i][1] - box[(i + 1) % pts_num][1]) *
+                      (box[i][1] - box[(i + 1) % pts_num][1]));
+  }
+  area = fabs(float(area / 2.0));
+
+  distance = area * unclip_ratio / dist;
+}
+
+cv::RotatedRect Unclip(std::vector<std::vector<float>> box,
+                       float unclip_ratio) {
+  float distance = 1.0;
+
+  GetContourArea(box, unclip_ratio, distance);
+
+  ClipperLib::ClipperOffset offset;
+  ClipperLib::Path p;
+  p << ClipperLib::IntPoint(static_cast<int>(box[0][0]),
+                            static_cast<int>(box[0][1]))
+    << ClipperLib::IntPoint(static_cast<int>(box[1][0]),
+                            static_cast<int>(box[1][1]))
+    << ClipperLib::IntPoint(static_cast<int>(box[2][0]),
+                            static_cast<int>(box[2][1]))
+    << ClipperLib::IntPoint(static_cast<int>(box[3][0]),
+                            static_cast<int>(box[3][1]));
+  offset.AddPath(p, ClipperLib::jtRound, ClipperLib::etClosedPolygon);
+
+  ClipperLib::Paths soln;
+  offset.Execute(soln, distance);
+  std::vector<cv::Point2f> points;
+
+  for (int j = 0; j < soln.size(); j++) {
+    for (int i = 0; i < soln[soln.size() - 1].size(); i++) {
+      points.emplace_back(soln[j][i].X, soln[j][i].Y);
+    }
+  }
+  cv::RotatedRect res = cv::minAreaRect(points);
+
+  return res;
+}
+
+std::vector<std::vector<float>> Mat2Vector(cv::Mat mat) {
+  std::vector<std::vector<float>> img_vec;
+  std::vector<float> tmp;
+
+  for (int i = 0; i < mat.rows; ++i) {
+    tmp.clear();
+    for (int j = 0; j < mat.cols; ++j) {
+      tmp.push_back(mat.at<float>(i, j));
+    }
+    img_vec.push_back(tmp);
+  }
+  return img_vec;
+}
+
+bool XsortFp32(std::vector<float> a, std::vector<float> b) {
+  if (a[0] != b[0])
+    return a[0] < b[0];
+  return false;
+}
+
+bool XsortInt(std::vector<int> a, std::vector<int> b) {
+  if (a[0] != b[0])
+    return a[0] < b[0];
+  return false;
+}
+
+std::vector<std::vector<int>>
+OrderPointsClockwise(std::vector<std::vector<int>> pts) {
+  std::vector<std::vector<int>> box = pts;
+  std::sort(box.begin(), box.end(), XsortInt);
+
+  std::vector<std::vector<int>> leftmost = {box[0], box[1]};
+  std::vector<std::vector<int>> rightmost = {box[2], box[3]};
+
+  if (leftmost[0][1] > leftmost[1][1])
+    std::swap(leftmost[0], leftmost[1]);
+
+  if (rightmost[0][1] > rightmost[1][1])
+    std::swap(rightmost[0], rightmost[1]);
+
+  std::vector<std::vector<int>> rect = {leftmost[0], rightmost[0], rightmost[1],
+                                        leftmost[1]};
+  return rect;
+}
+
+std::vector<std::vector<float>> GetMiniBoxes(cv::RotatedRect box, float &ssid) {
+  ssid = std::min(box.size.width, box.size.height);
+
+  cv::Mat points;
+  cv::boxPoints(box, points);
+
+  auto array = Mat2Vector(points);
+  std::sort(array.begin(), array.end(), XsortFp32);
+
+  std::vector<float> idx1 = array[0], idx2 = array[1], idx3 = array[2],
+                     idx4 = array[3];
+  if (array[3][1] <= array[2][1]) {
+    idx2 = array[3];
+    idx3 = array[2];
+  } else {
+    idx2 = array[2];
+    idx3 = array[3];
+  }
+  if (array[1][1] <= array[0][1]) {
+    idx1 = array[1];
+    idx4 = array[0];
+  } else {
+    idx1 = array[0];
+    idx4 = array[1];
+  }
+
+  array[0] = idx1;
+  array[1] = idx2;
+  array[2] = idx3;
+  array[3] = idx4;
+
+  return array;
+}
+
+float BoxScoreFast(std::vector<std::vector<float>> box_array, cv::Mat pred) {
+  auto array = box_array;
+  int width = pred.cols;
+  int height = pred.rows;
+
+  float box_x[4] = {array[0][0], array[1][0], array[2][0], array[3][0]};
+  float box_y[4] = {array[0][1], array[1][1], array[2][1], array[3][1]};
+
+  int xmin = clamp(
+      static_cast<int>(std::floorf(*(std::min_element(box_x, box_x + 4)))), 0,
+      width - 1);
+  int xmax =
+      clamp(static_cast<int>(std::ceilf(*(std::max_element(box_x, box_x + 4)))),
+            0, width - 1);
+  int ymin = clamp(
+      static_cast<int>(std::floorf(*(std::min_element(box_y, box_y + 4)))), 0,
+      height - 1);
+  int ymax =
+      clamp(static_cast<int>(std::ceilf(*(std::max_element(box_y, box_y + 4)))),
+            0, height - 1);
+
+  cv::Mat mask;
+  mask = cv::Mat::zeros(ymax - ymin + 1, xmax - xmin + 1, CV_8UC1);
+
+  cv::Point root_point[4];
+  root_point[0] = cv::Point(static_cast<int>(array[0][0]) - xmin,
+                            static_cast<int>(array[0][1]) - ymin);
+  root_point[1] = cv::Point(static_cast<int>(array[1][0]) - xmin,
+                            static_cast<int>(array[1][1]) - ymin);
+  root_point[2] = cv::Point(static_cast<int>(array[2][0]) - xmin,
+                            static_cast<int>(array[2][1]) - ymin);
+  root_point[3] = cv::Point(static_cast<int>(array[3][0]) - xmin,
+                            static_cast<int>(array[3][1]) - ymin);
+  const cv::Point *ppt[1] = {root_point};
+  int npt[] = {4};
+  cv::fillPoly(mask, ppt, npt, 1, cv::Scalar(1));
+
+  cv::Mat croppedImg;
+  pred(cv::Rect(xmin, ymin, xmax - xmin + 1, ymax - ymin + 1))
+      .copyTo(croppedImg);
+
+  auto score = cv::mean(croppedImg, mask)[0];
+  return score;
+}
+
+float PolygonScoreAcc(std::vector<cv::Point> contour, cv::Mat pred) {
+  int width = pred.cols;
+  int height = pred.rows;
+  std::vector<float> box_x;
+  std::vector<float> box_y;
+  for (int i = 0; i < contour.size(); ++i) {
+    box_x.push_back(contour[i].x);
+    box_y.push_back(contour[i].y);
+  }
+
+  int xmin =
+      clamp(int(std::floor(*(std::min_element(box_x.begin(), box_x.end())))), 0,
+            width - 1);
+  int xmax =
+      clamp(int(std::ceil(*(std::max_element(box_x.begin(), box_x.end())))), 0,
+            width - 1);
+  int ymin =
+      clamp(int(std::floor(*(std::min_element(box_y.begin(), box_y.end())))), 0,
+            height - 1);
+  int ymax =
+      clamp(int(std::ceil(*(std::max_element(box_y.begin(), box_y.end())))), 0,
+            height - 1);
+
+  cv::Mat mask;
+  mask = cv::Mat::zeros(ymax - ymin + 1, xmax - xmin + 1, CV_8UC1);
+
+  cv::Point *rook_point = new cv::Point[contour.size()];
+
+  for (int i = 0; i < contour.size(); ++i) {
+    rook_point[i] = cv::Point(int(box_x[i]) - xmin, int(box_y[i]) - ymin);
+  }
+  const cv::Point *ppt[1] = {rook_point};
+  int npt[] = {int(contour.size())};
+
+  cv::fillPoly(mask, ppt, npt, 1, cv::Scalar(1));
+
+  cv::Mat croppedImg;
+  pred(cv::Rect(xmin, ymin, xmax - xmin + 1, ymax - ymin + 1))
+      .copyTo(croppedImg);
+  float score = cv::mean(croppedImg, mask)[0];
+
+  delete[] rook_point;
+  return score;
+}
+
+std::vector<std::vector<std::vector<int>>>
+BoxesFromBitmap(const cv::Mat pred, const cv::Mat bitmap,
+                std::map<std::string, double> Config) {
+  const int min_size = 3;
+  const int max_candidates = 1000;
+  const float box_thresh = static_cast<float>(Config["det_db_box_thresh"]);
+  const float unclip_ratio = static_cast<float>(Config["det_db_unclip_ratio"]);
+  const int det_use_polygon_score = int(Config["det_use_polygon_score"]);
+
+  int width = bitmap.cols;
+  int height = bitmap.rows;
+
+  std::vector<std::vector<cv::Point>> contours;
+  std::vector<cv::Vec4i> hierarchy;
+
+  cv::findContours(bitmap, contours, hierarchy, cv::RETR_LIST,
+                   cv::CHAIN_APPROX_SIMPLE);
+
+  int num_contours =
+      contours.size() >= max_candidates ? max_candidates : contours.size();
+
+  std::vector<std::vector<std::vector<int>>> boxes;
+
+  for (int i = 0; i < num_contours; i++) {
+    float ssid;
+    if (contours[i].size() <= 2)
+      continue;
+
+    cv::RotatedRect box = cv::minAreaRect(contours[i]);
+    auto array = GetMiniBoxes(box, ssid);
+
+    auto box_for_unclip = array;
+    // end get_mini_box
+
+    if (ssid < min_size) {
+      continue;
+    }
+
+    float score;
+    if (det_use_polygon_score) {
+      score = PolygonScoreAcc(contours[i], pred);
+    } else {
+      score = BoxScoreFast(array, pred);
+    }
+    // end box_score_fast
+    if (score < box_thresh)
+      continue;
+
+    // start for unclip
+    cv::RotatedRect points = Unclip(box_for_unclip, unclip_ratio);
+    if (points.size.height < 1.001 && points.size.width < 1.001)
+      continue;
+    // end for unclip
+
+    cv::RotatedRect clipbox = points;
+    auto cliparray = GetMiniBoxes(clipbox, ssid);
+
+    if (ssid < min_size + 2)
+      continue;
+
+    int dest_width = pred.cols;
+    int dest_height = pred.rows;
+    std::vector<std::vector<int>> intcliparray;
+
+    for (int num_pt = 0; num_pt < 4; num_pt++) {
+      std::vector<int> a{
+          static_cast<int>(clamp(
+              roundf(cliparray[num_pt][0] / float(width) * float(dest_width)),
+              float(0), float(dest_width))),
+          static_cast<int>(clamp(
+              roundf(cliparray[num_pt][1] / float(height) * float(dest_height)),
+              float(0), float(dest_height)))};
+      intcliparray.push_back(a);
+    }
+    boxes.push_back(intcliparray);
+
+  } // end for
+  return boxes;
+}
+
+std::vector<std::vector<std::vector<int>>>
+FilterTagDetRes(std::vector<std::vector<std::vector<int>>> boxes, float ratio_h,
+                float ratio_w, cv::Mat srcimg) {
+  int oriimg_h = srcimg.rows;
+  int oriimg_w = srcimg.cols;
+
+  std::vector<std::vector<std::vector<int>>> root_points;
+  for (int n = 0; n < static_cast<int>(boxes.size()); n++) {
+    boxes[n] = OrderPointsClockwise(boxes[n]);
+    for (int m = 0; m < static_cast<int>(boxes[0].size()); m++) {
+      boxes[n][m][0] /= ratio_w;
+      boxes[n][m][1] /= ratio_h;
+
+      boxes[n][m][0] =
+          static_cast<int>(std::min(std::max(boxes[n][m][0], 0), oriimg_w - 1));
+      boxes[n][m][1] =
+          static_cast<int>(std::min(std::max(boxes[n][m][1], 0), oriimg_h - 1));
+    }
+  }
+
+  for (int n = 0; n < boxes.size(); n++) {
+    int rect_width, rect_height;
+    rect_width =
+        static_cast<int>(sqrt(pow(boxes[n][0][0] - boxes[n][1][0], 2) +
+                              pow(boxes[n][0][1] - boxes[n][1][1], 2)));
+    rect_height =
+        static_cast<int>(sqrt(pow(boxes[n][0][0] - boxes[n][3][0], 2) +
+                              pow(boxes[n][0][1] - boxes[n][3][1], 2)));
+    if (rect_width <= 4 || rect_height <= 4)
+      continue;
+    root_points.push_back(boxes[n]);
+  }
+  return root_points;
+}
diff --git a/deploy/lite/db_post_process.h b/deploy/lite/db_post_process.h
new file mode 100644
index 0000000..06dbcb2
--- /dev/null
+++ b/deploy/lite/db_post_process.h
@@ -0,0 +1,62 @@
+// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <math.h>
+
+#include <iostream>
+#include <map>
+#include <vector>
+
+#include "clipper.hpp"
+#include "opencv2/core.hpp"
+#include "opencv2/imgcodecs.hpp"
+#include "opencv2/imgproc.hpp"
+
+template <class T> T clamp(T x, T min, T max) {
+  if (x > max)
+    return max;
+  if (x < min)
+    return min;
+  return x;
+}
+
+std::vector<std::vector<float>> Mat2Vector(cv::Mat mat);
+
+void GetContourArea(std::vector<std::vector<float>> box, float unclip_ratio,
+                    float &distance);
+
+cv::RotatedRect Unclip(std::vector<std::vector<float>> box, float unclip_ratio);
+
+std::vector<std::vector<float>> Mat2Vector(cv::Mat mat);
+
+bool XsortFp32(std::vector<float> a, std::vector<float> b);
+
+bool XsortInt(std::vector<int> a, std::vector<int> b);
+
+std::vector<std::vector<int>>
+OrderPointsClockwise(std::vector<std::vector<int>> pts);
+
+std::vector<std::vector<float>> GetMiniBoxes(cv::RotatedRect box, float &ssid);
+
+float BoxScoreFast(std::vector<std::vector<float>> box_array, cv::Mat pred);
+
+std::vector<std::vector<std::vector<int>>>
+BoxesFromBitmap(const cv::Mat pred, const cv::Mat bitmap,
+                std::map<std::string, double> Config);
+
+std::vector<std::vector<std::vector<int>>>
+FilterTagDetRes(std::vector<std::vector<std::vector<int>>> boxes, float ratio_h,
+                float ratio_w, cv::Mat srcimg);
diff --git a/deploy/lite/imgs/lite_demo.png b/deploy/lite/imgs/lite_demo.png
new file mode 100644
index 0000000..c9daf1b
Binary files /dev/null and b/deploy/lite/imgs/lite_demo.png differ
diff --git a/deploy/lite/ocr_db_crnn.cc b/deploy/lite/ocr_db_crnn.cc
new file mode 100644
index 0000000..46ff22d
--- /dev/null
+++ b/deploy/lite/ocr_db_crnn.cc
@@ -0,0 +1,679 @@
+// Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "paddle_api.h" // NOLINT
+#include "paddle_place.h"
+#include <chrono>
+
+#include "AutoLog/auto_log/lite_autolog.h"
+#include "cls_process.h"
+#include "crnn_process.h"
+#include "db_post_process.h"
+
+using namespace paddle::lite_api; // NOLINT
+using namespace std;
+
+// fill tensor with mean and scale and trans layout: nhwc -> nchw, neon speed up
+void NeonMeanScale(const float *din, float *dout, int size,
+                   const std::vector<float> mean,
+                   const std::vector<float> scale) {
+  if (mean.size() != 3 || scale.size() != 3) {
+    std::cerr << "[ERROR] mean or scale size must equal to 3" << std::endl;
+    exit(1);
+  }
+  float32x4_t vmean0 = vdupq_n_f32(mean[0]);
+  float32x4_t vmean1 = vdupq_n_f32(mean[1]);
+  float32x4_t vmean2 = vdupq_n_f32(mean[2]);
+  float32x4_t vscale0 = vdupq_n_f32(scale[0]);
+  float32x4_t vscale1 = vdupq_n_f32(scale[1]);
+  float32x4_t vscale2 = vdupq_n_f32(scale[2]);
+
+  float *dout_c0 = dout;
+  float *dout_c1 = dout + size;
+  float *dout_c2 = dout + size * 2;
+
+  int i = 0;
+  for (; i < size - 3; i += 4) {
+    float32x4x3_t vin3 = vld3q_f32(din);
+    float32x4_t vsub0 = vsubq_f32(vin3.val[0], vmean0);
+    float32x4_t vsub1 = vsubq_f32(vin3.val[1], vmean1);
+    float32x4_t vsub2 = vsubq_f32(vin3.val[2], vmean2);
+    float32x4_t vs0 = vmulq_f32(vsub0, vscale0);
+    float32x4_t vs1 = vmulq_f32(vsub1, vscale1);
+    float32x4_t vs2 = vmulq_f32(vsub2, vscale2);
+    vst1q_f32(dout_c0, vs0);
+    vst1q_f32(dout_c1, vs1);
+    vst1q_f32(dout_c2, vs2);
+
+    din += 12;
+    dout_c0 += 4;
+    dout_c1 += 4;
+    dout_c2 += 4;
+  }
+  for (; i < size; i++) {
+    *(dout_c0++) = (*(din++) - mean[0]) * scale[0];
+    *(dout_c1++) = (*(din++) - mean[1]) * scale[1];
+    *(dout_c2++) = (*(din++) - mean[2]) * scale[2];
+  }
+}
+
+// resize image to a size multiple of 32 which is required by the network
+cv::Mat DetResizeImg(const cv::Mat img, int max_size_len,
+                     std::vector<float> &ratio_hw) {
+  int w = img.cols;
+  int h = img.rows;
+
+  float ratio = 1.f;
+  int max_wh = w >= h ? w : h;
+  if (max_wh > max_size_len) {
+    if (h > w) {
+      ratio = static_cast<float>(max_size_len) / static_cast<float>(h);
+    } else {
+      ratio = static_cast<float>(max_size_len) / static_cast<float>(w);
+    }
+  }
+
+  int resize_h = static_cast<int>(float(h) * ratio);
+  int resize_w = static_cast<int>(float(w) * ratio);
+  if (resize_h % 32 == 0)
+    resize_h = resize_h;
+  else if (resize_h / 32 < 1 + 1e-5)
+    resize_h = 32;
+  else
+    resize_h = (resize_h / 32 - 1) * 32;
+
+  if (resize_w % 32 == 0)
+    resize_w = resize_w;
+  else if (resize_w / 32 < 1 + 1e-5)
+    resize_w = 32;
+  else
+    resize_w = (resize_w / 32 - 1) * 32;
+
+  cv::Mat resize_img;
+  cv::resize(img, resize_img, cv::Size(resize_w, resize_h));
+
+  ratio_hw.push_back(static_cast<float>(resize_h) / static_cast<float>(h));
+  ratio_hw.push_back(static_cast<float>(resize_w) / static_cast<float>(w));
+  return resize_img;
+}
+
+cv::Mat RunClsModel(cv::Mat img, std::shared_ptr<PaddlePredictor> predictor_cls,
+                    const float thresh = 0.9) {
+  std::vector<float> mean = {0.5f, 0.5f, 0.5f};
+  std::vector<float> scale = {1 / 0.5f, 1 / 0.5f, 1 / 0.5f};
+
+  cv::Mat srcimg;
+  img.copyTo(srcimg);
+  cv::Mat crop_img;
+  img.copyTo(crop_img);
+  cv::Mat resize_img;
+
+  int index = 0;
+  float wh_ratio =
+      static_cast<float>(crop_img.cols) / static_cast<float>(crop_img.rows);
+
+  resize_img = ClsResizeImg(crop_img);
+  resize_img.convertTo(resize_img, CV_32FC3, 1 / 255.f);
+
+  const float *dimg = reinterpret_cast<const float *>(resize_img.data);
+
+  std::unique_ptr<Tensor> input_tensor0(std::move(predictor_cls->GetInput(0)));
+  input_tensor0->Resize({1, 3, resize_img.rows, resize_img.cols});
+  auto *data0 = input_tensor0->mutable_data<float>();
+
+  NeonMeanScale(dimg, data0, resize_img.rows * resize_img.cols, mean, scale);
+  // Run CLS predictor
+  predictor_cls->Run();
+
+  // Get output and run postprocess
+  std::unique_ptr<const Tensor> softmax_out(
+      std::move(predictor_cls->GetOutput(0)));
+  auto *softmax_scores = softmax_out->mutable_data<float>();
+  auto softmax_out_shape = softmax_out->shape();
+  float score = 0;
+  int label = 0;
+  for (int i = 0; i < softmax_out_shape[1]; i++) {
+    if (softmax_scores[i] > score) {
+      score = softmax_scores[i];
+      label = i;
+    }
+  }
+  if (label % 2 == 1 && score > thresh) {
+    cv::rotate(srcimg, srcimg, 1);
+  }
+  return srcimg;
+}
+
+void RunRecModel(std::vector<std::vector<std::vector<int>>> boxes, cv::Mat img,
+                 std::shared_ptr<PaddlePredictor> predictor_crnn,
+                 std::vector<std::string> &rec_text,
+                 std::vector<float> &rec_text_score,
+                 std::vector<std::string> charactor_dict,
+                 std::shared_ptr<PaddlePredictor> predictor_cls,
+                 int use_direction_classify, std::vector<double> *times,
+                 int rec_image_height) {
+  std::vector<float> mean = {0.5f, 0.5f, 0.5f};
+  std::vector<float> scale = {1 / 0.5f, 1 / 0.5f, 1 / 0.5f};
+
+  cv::Mat srcimg;
+  img.copyTo(srcimg);
+  cv::Mat crop_img;
+  cv::Mat resize_img;
+
+  int index = 0;
+
+  std::vector<double> time_info = {0, 0, 0};
+  for (int i = boxes.size() - 1; i >= 0; i--) {
+    auto preprocess_start = std::chrono::steady_clock::now();
+    crop_img = GetRotateCropImage(srcimg, boxes[i]);
+    if (use_direction_classify >= 1) {
+      crop_img = RunClsModel(crop_img, predictor_cls);
+    }
+    float wh_ratio =
+        static_cast<float>(crop_img.cols) / static_cast<float>(crop_img.rows);
+
+    resize_img = CrnnResizeImg(crop_img, wh_ratio, rec_image_height);
+    resize_img.convertTo(resize_img, CV_32FC3, 1 / 255.f);
+
+    const float *dimg = reinterpret_cast<const float *>(resize_img.data);
+
+    std::unique_ptr<Tensor> input_tensor0(
+        std::move(predictor_crnn->GetInput(0)));
+    input_tensor0->Resize({1, 3, resize_img.rows, resize_img.cols});
+    auto *data0 = input_tensor0->mutable_data<float>();
+
+    NeonMeanScale(dimg, data0, resize_img.rows * resize_img.cols, mean, scale);
+    auto preprocess_end = std::chrono::steady_clock::now();
+    //// Run CRNN predictor
+    auto inference_start = std::chrono::steady_clock::now();
+    predictor_crnn->Run();
+
+    // Get output and run postprocess
+    std::unique_ptr<const Tensor> output_tensor0(
+        std::move(predictor_crnn->GetOutput(0)));
+    auto *predict_batch = output_tensor0->data<float>();
+    auto predict_shape = output_tensor0->shape();
+    auto inference_end = std::chrono::steady_clock::now();
+
+    // ctc decode
+    auto postprocess_start = std::chrono::steady_clock::now();
+    std::string str_res;
+    int argmax_idx;
+    int last_index = 0;
+    float score = 0.f;
+    int count = 0;
+    float max_value = 0.0f;
+
+    for (int n = 0; n < predict_shape[1]; n++) {
+      argmax_idx = int(Argmax(&predict_batch[n * predict_shape[2]],
+                              &predict_batch[(n + 1) * predict_shape[2]]));
+      max_value =
+          float(*std::max_element(&predict_batch[n * predict_shape[2]],
+                                  &predict_batch[(n + 1) * predict_shape[2]]));
+      if (argmax_idx > 0 && (!(n > 0 && argmax_idx == last_index))) {
+        score += max_value;
+        count += 1;
+        str_res += charactor_dict[argmax_idx];
+      }
+      last_index = argmax_idx;
+    }
+    score /= count;
+    rec_text.push_back(str_res);
+    rec_text_score.push_back(score);
+    auto postprocess_end = std::chrono::steady_clock::now();
+
+    std::chrono::duration<float> preprocess_diff =
+        preprocess_end - preprocess_start;
+    time_info[0] += double(preprocess_diff.count() * 1000);
+    std::chrono::duration<float> inference_diff =
+        inference_end - inference_start;
+    time_info[1] += double(inference_diff.count() * 1000);
+    std::chrono::duration<float> postprocess_diff =
+        postprocess_end - postprocess_start;
+    time_info[2] += double(postprocess_diff.count() * 1000);
+  }
+
+  times->push_back(time_info[0]);
+  times->push_back(time_info[1]);
+  times->push_back(time_info[2]);
+}
+
+std::vector<std::vector<std::vector<int>>>
+RunDetModel(std::shared_ptr<PaddlePredictor> predictor, cv::Mat img,
+            std::map<std::string, double> Config, std::vector<double> *times) {
+  // Read img
+  int max_side_len = int(Config["max_side_len"]);
+  int det_db_use_dilate = int(Config["det_db_use_dilate"]);
+
+  cv::Mat srcimg;
+  img.copyTo(srcimg);
+
+  auto preprocess_start = std::chrono::steady_clock::now();
+  std::vector<float> ratio_hw;
+  img = DetResizeImg(img, max_side_len, ratio_hw);
+  cv::Mat img_fp;
+  img.convertTo(img_fp, CV_32FC3, 1.0 / 255.f);
+
+  // Prepare input data from image
+  std::unique_ptr<Tensor> input_tensor0(std::move(predictor->GetInput(0)));
+  input_tensor0->Resize({1, 3, img_fp.rows, img_fp.cols});
+  auto *data0 = input_tensor0->mutable_data<float>();
+
+  std::vector<float> mean = {0.485f, 0.456f, 0.406f};
+  std::vector<float> scale = {1 / 0.229f, 1 / 0.224f, 1 / 0.225f};
+  const float *dimg = reinterpret_cast<const float *>(img_fp.data);
+  NeonMeanScale(dimg, data0, img_fp.rows * img_fp.cols, mean, scale);
+  auto preprocess_end = std::chrono::steady_clock::now();
+
+  // Run predictor
+  auto inference_start = std::chrono::steady_clock::now();
+  predictor->Run();
+
+  // Get output and post process
+  std::unique_ptr<const Tensor> output_tensor(
+      std::move(predictor->GetOutput(0)));
+  auto *outptr = output_tensor->data<float>();
+  auto shape_out = output_tensor->shape();
+  auto inference_end = std::chrono::steady_clock::now();
+
+  // Save output
+  auto postprocess_start = std::chrono::steady_clock::now();
+  float pred[shape_out[2] * shape_out[3]];
+  unsigned char cbuf[shape_out[2] * shape_out[3]];
+
+  for (int i = 0; i < int(shape_out[2] * shape_out[3]); i++) {
+    pred[i] = static_cast<float>(outptr[i]);
+    cbuf[i] = static_cast<unsigned char>((outptr[i]) * 255);
+  }
+
+  cv::Mat cbuf_map(shape_out[2], shape_out[3], CV_8UC1,
+                   reinterpret_cast<unsigned char *>(cbuf));
+  cv::Mat pred_map(shape_out[2], shape_out[3], CV_32F,
+                   reinterpret_cast<float *>(pred));
+
+  const double threshold = double(Config["det_db_thresh"]) * 255;
+  const double max_value = 255;
+  cv::Mat bit_map;
+  cv::threshold(cbuf_map, bit_map, threshold, max_value, cv::THRESH_BINARY);
+  if (det_db_use_dilate == 1) {
+    cv::Mat dilation_map;
+    cv::Mat dila_ele =
+        cv::getStructuringElement(cv::MORPH_RECT, cv::Size(2, 2));
+    cv::dilate(bit_map, dilation_map, dila_ele);
+    bit_map = dilation_map;
+  }
+  auto boxes = BoxesFromBitmap(pred_map, bit_map, Config);
+
+  std::vector<std::vector<std::vector<int>>> filter_boxes =
+      FilterTagDetRes(boxes, ratio_hw[0], ratio_hw[1], srcimg);
+  auto postprocess_end = std::chrono::steady_clock::now();
+
+  std::chrono::duration<float> preprocess_diff =
+      preprocess_end - preprocess_start;
+  times->push_back(double(preprocess_diff.count() * 1000));
+  std::chrono::duration<float> inference_diff = inference_end - inference_start;
+  times->push_back(double(inference_diff.count() * 1000));
+  std::chrono::duration<float> postprocess_diff =
+      postprocess_end - postprocess_start;
+  times->push_back(double(postprocess_diff.count() * 1000));
+
+  return filter_boxes;
+}
+
+std::shared_ptr<PaddlePredictor> loadModel(std::string model_file,
+                                           int num_threads) {
+  MobileConfig config;
+  config.set_model_from_file(model_file);
+
+  config.set_threads(num_threads);
+  std::shared_ptr<PaddlePredictor> predictor =
+      CreatePaddlePredictor<MobileConfig>(config);
+  return predictor;
+}
+
+cv::Mat Visualization(cv::Mat srcimg,
+                      std::vector<std::vector<std::vector<int>>> boxes) {
+  cv::Point rook_points[boxes.size()][4];
+  for (int n = 0; n < boxes.size(); n++) {
+    for (int m = 0; m < boxes[0].size(); m++) {
+      rook_points[n][m] = cv::Point(static_cast<int>(boxes[n][m][0]),
+                                    static_cast<int>(boxes[n][m][1]));
+    }
+  }
+  cv::Mat img_vis;
+  srcimg.copyTo(img_vis);
+  for (int n = 0; n < boxes.size(); n++) {
+    const cv::Point *ppt[1] = {rook_points[n]};
+    int npt[] = {4};
+    cv::polylines(img_vis, ppt, npt, 1, 1, CV_RGB(0, 255, 0), 2, 8, 0);
+  }
+
+  cv::imwrite("./vis.jpg", img_vis);
+  std::cout << "The detection visualized image saved in ./vis.jpg" << std::endl;
+  return img_vis;
+}
+
+std::vector<std::string> split(const std::string &str,
+                               const std::string &delim) {
+  std::vector<std::string> res;
+  if ("" == str)
+    return res;
+  char *strs = new char[str.length() + 1];
+  std::strcpy(strs, str.c_str());
+
+  char *d = new char[delim.length() + 1];
+  std::strcpy(d, delim.c_str());
+
+  char *p = std::strtok(strs, d);
+  while (p) {
+    string s = p;
+    res.push_back(s);
+    p = std::strtok(NULL, d);
+  }
+
+  return res;
+}
+
+std::map<std::string, double> LoadConfigTxt(std::string config_path) {
+  auto config = ReadDict(config_path);
+
+  std::map<std::string, double> dict;
+  for (int i = 0; i < config.size(); i++) {
+    std::vector<std::string> res = split(config[i], " ");
+    dict[res[0]] = stod(res[1]);
+  }
+  return dict;
+}
+
+void check_params(int argc, char **argv) {
+  if (argc <= 1 ||
+      (strcmp(argv[1], "det") != 0 && strcmp(argv[1], "rec") != 0 &&
+       strcmp(argv[1], "system") != 0)) {
+    std::cerr << "Please choose one mode of [det, rec, system] !" << std::endl;
+    exit(1);
+  }
+  if (strcmp(argv[1], "det") == 0) {
+    if (argc < 9) {
+      std::cerr << "[ERROR] usage:" << argv[0]
+                << " det det_model runtime_device num_threads batchsize "
+                   "img_dir det_config lite_benchmark_value"
+                << std::endl;
+      exit(1);
+    }
+  }
+
+  if (strcmp(argv[1], "rec") == 0) {
+    if (argc < 9) {
+      std::cerr << "[ERROR] usage:" << argv[0]
+                << " rec rec_model runtime_device num_threads batchsize "
+                   "img_dir key_txt lite_benchmark_value"
+                << std::endl;
+      exit(1);
+    }
+  }
+
+  if (strcmp(argv[1], "system") == 0) {
+    if (argc < 12) {
+      std::cerr << "[ERROR] usage:" << argv[0]
+                << " system det_model rec_model clas_model runtime_device "
+                   "num_threads batchsize img_dir det_config key_txt "
+                   "lite_benchmark_value"
+                << std::endl;
+      exit(1);
+    }
+  }
+}
+
+void system(char **argv) {
+  std::string det_model_file = argv[2];
+  std::string rec_model_file = argv[3];
+  std::string cls_model_file = argv[4];
+  std::string runtime_device = argv[5];
+  std::string precision = argv[6];
+  std::string num_threads = argv[7];
+  std::string batchsize = argv[8];
+  std::string img_dir = argv[9];
+  std::string det_config_path = argv[10];
+  std::string dict_path = argv[11];
+
+  if (strcmp(argv[6], "FP32") != 0 && strcmp(argv[6], "INT8") != 0) {
+    std::cerr << "Only support FP32 or INT8." << std::endl;
+    exit(1);
+  }
+
+  std::vector<cv::String> cv_all_img_names;
+  cv::glob(img_dir, cv_all_img_names);
+
+  //// load config from txt file
+  auto Config = LoadConfigTxt(det_config_path);
+  int use_direction_classify = int(Config["use_direction_classify"]);
+  int rec_image_height = int(Config["rec_image_height"]);
+  auto charactor_dict = ReadDict(dict_path);
+  charactor_dict.insert(charactor_dict.begin(), "#"); // blank char for ctc
+  charactor_dict.push_back(" ");
+
+  auto det_predictor = loadModel(det_model_file, std::stoi(num_threads));
+  auto rec_predictor = loadModel(rec_model_file, std::stoi(num_threads));
+  auto cls_predictor = loadModel(cls_model_file, std::stoi(num_threads));
+
+  std::vector<double> det_time_info = {0, 0, 0};
+  std::vector<double> rec_time_info = {0, 0, 0};
+
+  for (int i = 0; i < cv_all_img_names.size(); ++i) {
+    std::cout << "The predict img: " << cv_all_img_names[i] << std::endl;
+    cv::Mat srcimg = cv::imread(cv_all_img_names[i], cv::IMREAD_COLOR);
+
+    if (!srcimg.data) {
+      std::cerr << "[ERROR] image read failed! image path: "
+                << cv_all_img_names[i] << std::endl;
+      exit(1);
+    }
+
+    std::vector<double> det_times;
+    auto boxes = RunDetModel(det_predictor, srcimg, Config, &det_times);
+
+    std::vector<std::string> rec_text;
+    std::vector<float> rec_text_score;
+
+    std::vector<double> rec_times;
+    RunRecModel(boxes, srcimg, rec_predictor, rec_text, rec_text_score,
+                charactor_dict, cls_predictor, use_direction_classify,
+                &rec_times, rec_image_height);
+
+    //// visualization
+    auto img_vis = Visualization(srcimg, boxes);
+
+    //// print recognized text
+    for (int i = 0; i < rec_text.size(); i++) {
+      std::cout << i << "\t" << rec_text[i] << "\t" << rec_text_score[i]
+                << std::endl;
+    }
+
+    det_time_info[0] += det_times[0];
+    det_time_info[1] += det_times[1];
+    det_time_info[2] += det_times[2];
+    rec_time_info[0] += rec_times[0];
+    rec_time_info[1] += rec_times[1];
+    rec_time_info[2] += rec_times[2];
+  }
+  if (strcmp(argv[12], "True") == 0) {
+    AutoLogger autolog_det(det_model_file, runtime_device,
+                           std::stoi(num_threads), std::stoi(batchsize),
+                           "dynamic", precision, det_time_info,
+                           cv_all_img_names.size());
+    AutoLogger autolog_rec(rec_model_file, runtime_device,
+                           std::stoi(num_threads), std::stoi(batchsize),
+                           "dynamic", precision, rec_time_info,
+                           cv_all_img_names.size());
+
+    autolog_det.report();
+    std::cout << std::endl;
+    autolog_rec.report();
+  }
+}
+
+void det(int argc, char **argv) {
+  std::string det_model_file = argv[2];
+  std::string runtime_device = argv[3];
+  std::string precision = argv[4];
+  std::string num_threads = argv[5];
+  std::string batchsize = argv[6];
+  std::string img_dir = argv[7];
+  std::string det_config_path = argv[8];
+
+  if (strcmp(argv[4], "FP32") != 0 && strcmp(argv[4], "INT8") != 0) {
+    std::cerr << "Only support FP32 or INT8." << std::endl;
+    exit(1);
+  }
+
+  std::vector<cv::String> cv_all_img_names;
+  cv::glob(img_dir, cv_all_img_names);
+
+  //// load config from txt file
+  auto Config = LoadConfigTxt(det_config_path);
+
+  auto det_predictor = loadModel(det_model_file, std::stoi(num_threads));
+
+  std::vector<double> time_info = {0, 0, 0};
+  for (int i = 0; i < cv_all_img_names.size(); ++i) {
+    std::cout << "The predict img: " << cv_all_img_names[i] << std::endl;
+    cv::Mat srcimg = cv::imread(cv_all_img_names[i], cv::IMREAD_COLOR);
+
+    if (!srcimg.data) {
+      std::cerr << "[ERROR] image read failed! image path: "
+                << cv_all_img_names[i] << std::endl;
+      exit(1);
+    }
+
+    std::vector<double> times;
+    auto boxes = RunDetModel(det_predictor, srcimg, Config, &times);
+
+    //// visualization
+    auto img_vis = Visualization(srcimg, boxes);
+    std::cout << boxes.size() << " bboxes have detected:" << std::endl;
+
+    for (int i = 0; i < boxes.size(); i++) {
+      std::cout << "The " << i << " box:" << std::endl;
+      for (int j = 0; j < 4; j++) {
+        for (int k = 0; k < 2; k++) {
+          std::cout << boxes[i][j][k] << "\t";
+        }
+      }
+      std::cout << std::endl;
+    }
+    time_info[0] += times[0];
+    time_info[1] += times[1];
+    time_info[2] += times[2];
+  }
+
+  if (strcmp(argv[9], "True") == 0) {
+    AutoLogger autolog(det_model_file, runtime_device, std::stoi(num_threads),
+                       std::stoi(batchsize), "dynamic", precision, time_info,
+                       cv_all_img_names.size());
+    autolog.report();
+  }
+}
+
+void rec(int argc, char **argv) {
+  std::string rec_model_file = argv[2];
+  std::string runtime_device = argv[3];
+  std::string precision = argv[4];
+  std::string num_threads = argv[5];
+  std::string batchsize = argv[6];
+  std::string img_dir = argv[7];
+  std::string dict_path = argv[8];
+  std::string config_path = argv[9];
+
+  if (strcmp(argv[4], "FP32") != 0 && strcmp(argv[4], "INT8") != 0) {
+    std::cerr << "Only support FP32 or INT8." << std::endl;
+    exit(1);
+  }
+
+  auto Config = LoadConfigTxt(config_path);
+  int rec_image_height = int(Config["rec_image_height"]);
+
+  std::vector<cv::String> cv_all_img_names;
+  cv::glob(img_dir, cv_all_img_names);
+
+  auto charactor_dict = ReadDict(dict_path);
+  charactor_dict.insert(charactor_dict.begin(), "#"); // blank char for ctc
+  charactor_dict.push_back(" ");
+
+  auto rec_predictor = loadModel(rec_model_file, std::stoi(num_threads));
+
+  std::shared_ptr<PaddlePredictor> cls_predictor;
+
+  std::vector<double> time_info = {0, 0, 0};
+  for (int i = 0; i < cv_all_img_names.size(); ++i) {
+    std::cout << "The predict img: " << cv_all_img_names[i] << std::endl;
+    cv::Mat srcimg = cv::imread(cv_all_img_names[i], cv::IMREAD_COLOR);
+
+    if (!srcimg.data) {
+      std::cerr << "[ERROR] image read failed! image path: "
+                << cv_all_img_names[i] << std::endl;
+      exit(1);
+    }
+
+    int width = srcimg.cols;
+    int height = srcimg.rows;
+    std::vector<int> upper_left = {0, 0};
+    std::vector<int> upper_right = {width, 0};
+    std::vector<int> lower_right = {width, height};
+    std::vector<int> lower_left = {0, height};
+    std::vector<std::vector<int>> box = {upper_left, upper_right, lower_right,
+                                         lower_left};
+    std::vector<std::vector<std::vector<int>>> boxes = {box};
+
+    std::vector<std::string> rec_text;
+    std::vector<float> rec_text_score;
+    std::vector<double> times;
+    RunRecModel(boxes, srcimg, rec_predictor, rec_text, rec_text_score,
+                charactor_dict, cls_predictor, 0, &times, rec_image_height);
+
+    //// print recognized text
+    for (int i = 0; i < rec_text.size(); i++) {
+      std::cout << i << "\t" << rec_text[i] << "\t" << rec_text_score[i]
+                << std::endl;
+    }
+    time_info[0] += times[0];
+    time_info[1] += times[1];
+    time_info[2] += times[2];
+  }
+  // TODO: support autolog
+  if (strcmp(argv[9], "True") == 0) {
+    AutoLogger autolog(rec_model_file, runtime_device, std::stoi(num_threads),
+                       std::stoi(batchsize), "dynamic", precision, time_info,
+                       cv_all_img_names.size());
+    autolog.report();
+  }
+}
+
+int main(int argc, char **argv) {
+  check_params(argc, argv);
+  std::cout << "mode: " << argv[1] << endl;
+
+  if (strcmp(argv[1], "system") == 0) {
+    system(argv);
+  }
+
+  if (strcmp(argv[1], "det") == 0) {
+    det(argc, argv);
+  }
+
+  if (strcmp(argv[1], "rec") == 0) {
+    rec(argc, argv);
+  }
+
+  return 0;
+}
diff --git a/deploy/lite/prepare.sh b/deploy/lite/prepare.sh
new file mode 100644
index 0000000..daaa30c
--- /dev/null
+++ b/deploy/lite/prepare.sh
@@ -0,0 +1,9 @@
+#!/bin/bash
+
+mkdir -p  $1/demo/cxx/ocr/debug/
+cp  ../../ppocr/utils/ppocr_keys_v1.txt  $1/demo/cxx/ocr/debug/
+cp -r  ./*   $1/demo/cxx/ocr/
+cp ./config.txt  $1/demo/cxx/ocr/debug/
+cp ../../doc/imgs/11.jpg  $1/demo/cxx/ocr/debug/
+
+echo "Prepare Done"
diff --git a/deploy/lite/readme.md b/deploy/lite/readme.md
new file mode 100644
index 0000000..3d05c38
--- /dev/null
+++ b/deploy/lite/readme.md
@@ -0,0 +1,298 @@
+# Mobile deployment based on Paddle-Lite
+
+- [1. Preparation](#1-preparation)
+    - [Preparation environment](#preparation-environment)
+    - [1.1 Prepare the cross-compilation environment](#11-prepare-the-cross-compilation-environment)
+    - [1.2 Prepare Paddle-Lite library](#12-prepare-paddle-lite-library)
+- [2. Run](#2-run)
+    - [2.1 Inference Model Optimization](#21-inference-model-optimization)
+    - [2.2 Run optimized model on Phone](#22-run-optimized-model-on-phone)
+- [FAQ](#faq)
+
+
+This tutorial will introduce how to use [Paddle-Lite](https://github.com/PaddlePaddle/Paddle-Lite) to deploy PaddleOCR ultra-lightweight Chinese and English detection models on mobile phones.
+
+Paddle-Lite is a lightweight inference engine for PaddlePaddle. It provides efficient inference capabilities for mobile phones and IoT, and extensively integrates cross-platform hardware to provide lightweight deployment solutions for end-side deployment issues.
+
+## 1. Preparation
+
+### Preparation environment
+
+- Computer (for Compiling Paddle Lite)
+- Mobile phone (arm7 or arm8)
+
+### 1.1 Prepare the cross-compilation environment
+The cross-compilation environment is used to compile C++ demos of Paddle Lite and PaddleOCR.
+Supports multiple development environments.
+
+For the compilation process of different development environments, please refer to the corresponding documents.
+
+1. [Docker](https://paddle-lite-pjc.readthedocs.io/zh/latest/user_guides/source_compile.html#docker)
+2. [Linux](https://paddle-lite-pjc.readthedocs.io/zh/latest/user_guides/source_compile.html#linux)
+3. [MAC OS](https://paddle-lite-pjc.readthedocs.io/zh/latest/user_guides/source_compile.html#mac-os)
+
+### 1.2 Prepare Paddle-Lite library
+
+There are two ways to obtain the Paddle-Lite library：
+- 1. [Recommended] Download directly, the download link of the Paddle-Lite library is as follows：
+
+      | Platform | Paddle-Lite library download link |
+      |---|---|
+      |Android|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.android.armv7.gcc.c++_shared.with_extra.with_cv.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.android.armv8.gcc.c++_shared.with_extra.with_cv.tar.gz)|
+      |IOS|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.ios.armv7.with_cv.with_extra.with_log.tiny_publish.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.ios.armv8.with_cv.with_extra.with_log.tiny_publish.tar.gz)|
+
+      Note: 1. The above Paddle-Lite library is compiled from the Paddle-Lite 2.10 branch. For more information about Paddle-Lite 2.10, please refer to [link](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.10).
+
+    **Note: It is recommended to use paddlelite>=2.10 version of the prediction library, other prediction library versions [download link](https://github.com/PaddlePaddle/Paddle-Lite/tags)**
+
+- 2. Compile Paddle-Lite to get the prediction library. The compilation method of Paddle-Lite is as follows：
+```
+git clone https://github.com/PaddlePaddle/Paddle-Lite.git
+cd Paddle-Lite
+# Switch to Paddle-Lite release/v2.10 stable branch
+git checkout release/v2.10
+./lite/tools/build_android.sh  --arch=armv8  --with_cv=ON --with_extra=ON
+```
+
+Note: When compiling Paddle-Lite to obtain the Paddle-Lite library, you need to turn on the two options `--with_cv=ON --with_extra=ON`, `--arch` means the `arm` version, here is designated as armv8,
+
+More compilation commands refer to the introduction [link](https://paddle-lite-pjc.readthedocs.io/zh/latest/user_guides/source_compile.html#paddlelite) 。
+
+After directly downloading the Paddle-Lite library and decompressing it, you can get the `inference_lite_lib.android.armv8/` folder, and the Paddle-Lite library obtained by compiling Paddle-Lite is located
+`Paddle-Lite/build.lite.android.armv8.gcc/inference_lite_lib.android.armv8/` folder.
+
+The structure of the prediction library is as follows:
+```
+inference_lite_lib.android.armv8/
+|-- cxx                                        C++ prebuild library
+|   |-- include                                C++
+|   |   |-- paddle_api.h
+|   |   |-- paddle_image_preprocess.h
+|   |   |-- paddle_lite_factory_helper.h
+|   |   |-- paddle_place.h
+|   |   |-- paddle_use_kernels.h
+|   |   |-- paddle_use_ops.h
+|   |   `-- paddle_use_passes.h
+|   `-- lib                                           C++ library
+|       |-- libpaddle_api_light_bundled.a             C++ static library
+|       `-- libpaddle_light_api_shared.so             C++ dynamic library
+|-- java                                     Java library
+|   |-- jar
+|   |   `-- PaddlePredictor.jar
+|   |-- so
+|   |   `-- libpaddle_lite_jni.so
+|   `-- src
+|-- demo                                     C++ and Java demo
+|   |-- cxx                                  C++ demo
+|   `-- java                                 Java demo
+```
+
+## 2 Run
+
+### 2.1 Inference Model Optimization
+
+Paddle Lite provides a variety of strategies to automatically optimize the original training model, including quantization, sub-graph fusion, hybrid scheduling, Kernel optimization and so on. In order to make the optimization process more convenient and easy to use, Paddle Lite provide opt tools to automatically complete the optimization steps and output a lightweight, optimal executable model.
+
+If you have prepared the model file ending in .nb, you can skip this step.
+
+The following table also provides a series of models that can be deployed on mobile phones to recognize Chinese. You can directly download the optimized model.
+
+|Version|Introduction|Model size|Detection model|Text Direction model|Recognition model|Paddle-Lite branch|
+|---|---|---|---|---|---|---|
+|PP-OCRv3|extra-lightweight chinese OCR optimized model|16.2M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/PP-OCRv3_mobile_det_infer.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_infer_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_infer.nb)|v2.10|
+|PP-OCRv3(slim)|extra-lightweight chinese OCR optimized model|5.9M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_slim_infer.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_slim_infer.nb)|v2.10|
+|PP-OCRv2|extra-lightweight chinese OCR optimized model|11M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_infer_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_infer_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_infer_opt.nb)|v2.10|
+|PP-OCRv2(slim)|extra-lightweight chinese OCR optimized model|4.6M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_slim_opt.nb)|v2.10|
+
+If you directly use the model in the above table for deployment, you can skip the following steps and directly read [Section 2.2](#2.2-Run-optimized-model-on-Phone).
+
+If the model to be deployed is not in the above table, you need to follow the steps below to obtain the optimized model.
+
+- Step 1: Refer to [document](https://www.paddlepaddle.org.cn/lite/v2.10/user_guides/opt/opt_python.html) to install paddlelite, which is used to convert paddle inference model to paddlelite required for running nb model
+```
+pip install paddlelite==2.10 # The paddlelite version should be the same as the prediction library version
+```
+After installation, the following commands can view the help information
+```
+paddle_lite_opt
+```
+
+Introduction to paddle_lite_opt parameters:
+
+|Options|Description|
+|---|---|
+|--model_dir|The path of the PaddlePaddle model to be optimized (non-combined form)|
+|--model_file|The network structure file path of the PaddlePaddle model (combined form) to be optimized|
+|--param_file|The weight file path of the PaddlePaddle model (combined form) to be optimized|
+|--optimize_out_type|Output model type, currently supports two types: protobuf and naive_buffer, among which naive_buffer is a more lightweight serialization/deserialization implementation. If you need to perform model prediction on the mobile side, please set this option to naive_buffer. The default is protobuf|
+|--optimize_out|The output path of the optimized model|
+|--valid_targets|The executable backend of the model, the default is arm. Currently it supports x86, arm, opencl, npu, xpu, multiple backends can be specified at the same time (separated by spaces), and Model Optimize Tool will automatically select the best method. If you need to support Huawei NPU (DaVinci architecture NPU equipped with Kirin 810/990 Soc), it should be set to npu, arm|
+|--record_tailoring_info|When using the function of cutting library files according to the model, set this option to true to record the kernel and OP information contained in the optimized model. The default is false|
+
+`--model_dir` is suitable for the non-combined mode of the model to be optimized, and the inference model of PaddleOCR is the combined mode, that is, the model structure and model parameters are stored in a single file.
+
+- Step 2: Use paddle_lite_opt to convert the inference model to the mobile model format.
+
+The following takes the ultra-lightweight Chinese model of PaddleOCR as an example to introduce the use of the compiled opt file to complete the conversion of the inference model to the Paddle-Lite optimized model
+
+```
+# 【[Recommendation] Download the Chinese and English inference model of PP-OCRv3
+wget  https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_slim_infer.tar && tar xf  ch_PP-OCRv3_det_slim_infer.tar
+wget  https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_slim_infer.tar && tar xf  ch_PP-OCRv2_rec_slim_quant_infer.tar
+wget  https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_cls_slim_infer.tar && tar xf  ch_ppocr_mobile_v2.0_cls_slim_infer.tar
+# Convert detection model
+paddle_lite_opt --model_file=./ch_PP-OCRv3_det_slim_infer/inference.pdmodel  --param_file=./ch_PP-OCRv3_det_slim_infer/inference.pdiparams  --optimize_out=./ch_PP-OCRv3_det_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
+# Convert recognition model
+paddle_lite_opt --model_file=./ch_PP-OCRv3_rec_slim_infer/inference.pdmodel  --param_file=./ch_PP-OCRv3_rec_slim_infer/inference.pdiparams  --optimize_out=./ch_PP-OCRv3_rec_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
+# Convert angle classifier model
+paddle_lite_opt --model_file=./ch_ppocr_mobile_v2.0_cls_slim_infer/inference.pdmodel  --param_file=./ch_ppocr_mobile_v2.0_cls_slim_infer/inference.pdiparams  --optimize_out=./ch_ppocr_mobile_v2.0_cls_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
+```
+
+After the conversion is successful, there will be more files ending with `.nb` in the inference model directory, which is the successfully converted model file.
+
+<a name="2.2-Run-optimized-model-on-Phone"></a>
+### 2.2 Run optimized model on Phone
+
+Some preparatory work is required first.
+ 1. Prepare an Android phone with arm8. If the compiled prediction library and opt file are armv7, you need an arm7 phone and modify ARM_ABI = arm7 in the Makefile.
+ 2. Make sure the phone is connected to the computer, open the USB debugging option of the phone, and select the file transfer mode.
+ 3. Install the adb tool on the computer.
+
+    3.1. Install ADB for MAC:
+    ```
+    brew cask install android-platform-tools
+    ```
+    3.2. Install ADB for Linux
+    ```
+    sudo apt update
+    sudo apt install -y wget adb
+    ```
+    3.3. Install ADB for windows
+
+    To install on win, you need to go to Google's Android platform to download the adb package for installation：[link](https://developer.android.com/studio)
+
+    Verify whether adb is installed successfully
+     ```
+    adb devices
+    ```
+    If there is device output, it means the installation is successful。
+    ```
+       List of devices attached
+       744be294    device
+    ```
+
+ 4. Prepare optimized models, prediction library files, test images and dictionary files used.
+ ```
+ git clone https://github.com/PaddlePaddle/PaddleOCR.git
+ cd PaddleOCR/deploy/lite/
+ # run prepare.sh
+ sh prepare.sh /{lite prediction library path}/inference_lite_lib.android.armv8
+
+ #
+ cd /{lite prediction library path}/inference_lite_lib.android.armv8/
+ cd demo/cxx/ocr/
+ # copy paddle-lite C++ .so file to debug/ directory
+ cp ../../../cxx/lib/libpaddle_light_api_shared.so ./debug/
+
+ cd inference_lite_lib.android.armv8/demo/cxx/ocr/
+ cp ../../../cxx/lib/libpaddle_light_api_shared.so ./debug/
+ ```
+
+Prepare the test image, taking PaddleOCR/doc/imgs/11.jpg as an example, copy the image file to the demo/cxx/ocr/debug/ folder. Prepare the model files optimized by the lite opt tool, ch_PP-OCRv3_det_slim_opt.nb , ch_PP-OCRv3_rec_slim_opt.nb , and place them under the demo/cxx/ocr/debug/ folder.
+
+The structure of the OCR demo is as follows after the above command is executed:
+
+```
+demo/cxx/ocr/
+|-- debug/
+|   |--ch_PP-OCRv3_det_slim_opt.nb           Detection model
+|   |--ch_PP-OCRv3_rec_slim_opt.nb           Recognition model
+|   |--ch_ppocr_mobile_v2.0_cls_slim_opt.nb           Text direction classification model
+|   |--11.jpg                           Image for OCR
+|   |--ppocr_keys_v1.txt                Dictionary file
+|   |--libpaddle_light_api_shared.so    C++ .so file
+|   |--config.txt                       Config file
+|-- config.txt                  Config file
+|-- cls_process.cc              Pre-processing and post-processing files for the angle classifier
+|-- cls_process.h
+|-- crnn_process.cc             Pre-processing and post-processing files for the CRNN model
+|-- crnn_process.h
+|-- db_post_process.cc          Pre-processing and post-processing files for the DB model
+|-- db_post_process.h
+|-- Makefile
+|-- ocr_db_crnn.cc              C++ main code
+```
+
+**Note**:
+1. `ppocr_keys_v1.txt` is a Chinese dictionary file. If the nb model is used for English recognition or other language recognition, dictionary file should be replaced with a dictionary of the corresponding language. PaddleOCR provides a variety of dictionaries under ppocr/utils/, including:
+```
+dict/french_dict.txt     # french
+dict/german_dict.txt     # german
+ic15_dict.txt       # english
+dict/japan_dict.txt      # japan
+dict/korean_dict.txt     # korean
+ppocr_keys_v1.txt   # chinese
+```
+
+2.  `config.txt` of the detector and classifier, as shown below:
+```
+max_side_len  960         #  Limit the maximum image height and width to 960
+det_db_thresh  0.3        # Used to filter the binarized image of DB prediction, setting 0.-0.3 has no obvious effect on the result
+det_db_box_thresh  0.5    # DDB post-processing filter box threshold, if there is a missing box detected, it can be reduced as appropriate
+det_db_unclip_ratio  1.6  # Indicates the compactness of the text box, the smaller the value, the closer the text box to the text
+use_direction_classify  0  # Whether to use the direction classifier, 0 means not to use, 1 means to use
+rec_image_height  48      # The height of the input image of the recognition model, the PP-OCRv3 model needs to be set to 48, and the PP-OCRv2 model needs to be set to 32
+```
+
+ 5. Run Model on phone
+
+After the above steps are completed, you can use adb to push the file to the phone to run, the steps are as follows:
+
+ ```
+ # Execute the compilation and get the executable file ocr_db_crnn
+ # The first execution of this command will download dependent libraries such as opencv. After the download is complete, you need to execute it again
+ make -j
+ # Move the compiled executable file to the debug folder
+ mv ocr_db_crnn ./debug/
+ # Push the debug folder to the phone
+ adb push debug /data/local/tmp/
+ adb shell
+ cd /data/local/tmp/debug
+ export LD_LIBRARY_PATH=${PWD}:$LD_LIBRARY_PATH
+ # The use of ocr_db_crnn is:
+ # ./ocr_db_crnn Mode Detection model file Orientation classifier model file Recognition model file  Hardware  Precision  Threads Batchsize  Test image path Dictionary file path
+ ./ocr_db_crnn system ch_PP-OCRv3_det_slim_opt.nb  ch_PP-OCRv3_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_slim_opt.nb  arm8 INT8 10 1  ./11.jpg  config.txt  ppocr_keys_v1.txt  True
+# precision can be INT8 for quantitative model or FP32 for normal model.
+
+# Only using detection model
+./ocr_db_crnn  det ch_PP-OCRv3_det_slim_opt.nb arm8 INT8 10 1 ./11.jpg  config.txt
+
+# Only using recognition model
+./ocr_db_crnn  rec ch_PP-OCRv3_rec_slim_opt.nb arm8 INT8 10 1 word_1.jpg ppocr_keys_v1.txt config.txt
+ ```
+
+If you modify the code, you need to recompile and push to the phone.
+
+The outputs are as follows:
+
+<div align="center">
+    <img src="imgs/lite_demo.png" width="600">
+</div>
+
+## FAQ
+
+Q1: What if I want to change the model, do I need to run it again according to the process?
+
+A1: If you have performed the above steps, you only need to replace the .nb model file to complete the model replacement.
+
+Q2: How to test with another picture?
+
+A2: Replace the .jpg test image under ./debug with the image you want to test, and run adb push to push new image to the phone.
+
+Q3: How to package it into the mobile APP?
+
+A3: This demo aims to provide the core algorithm part that can run OCR on mobile phones. Further, PaddleOCR/deploy/android_demo is an example of encapsulating this demo into a mobile app for reference.
+
+Q4: When running the demo, an error is reported `Error: This model is not supported, because kernel for 'io_copy' is not supported by Paddle-Lite.`
+
+A4: The problem is that the installed paddlelite version does not match the downloaded prediction library version. Make sure that the paddleliteopt tool matches your prediction library version, and try to switch to the nb model again.
diff --git a/deploy/lite/readme_ch.md b/deploy/lite/readme_ch.md
new file mode 100644
index 0000000..7e07b8b
--- /dev/null
+++ b/deploy/lite/readme_ch.md
@@ -0,0 +1,297 @@
+# 端侧部署
+
+- [1. 准备环境](#1-准备环境)
+    - [运行准备](#运行准备)
+    - [1.1 准备交叉编译环境](#11-准备交叉编译环境)
+    - [1.2 准备预测库](#12-准备预测库)
+- [2 开始运行](#2-开始运行)
+    - [2.1 模型优化](#21-模型优化)
+    - [2.2 与手机联调](#22-与手机联调)
+- [FAQ](#faq)
+
+
+本教程将介绍基于[Paddle Lite](https://github.com/PaddlePaddle/Paddle-Lite) 在移动端部署PaddleOCR超轻量中文检测、识别模型的详细步骤。
+
+Paddle Lite是飞桨轻量化推理引擎，为手机、IOT端提供高效推理能力，并广泛整合跨平台硬件，为端侧部署及应用落地问题提供轻量化的部署方案。
+
+
+## 1. 准备环境
+
+### 运行准备
+- 电脑（编译Paddle Lite）
+- 安卓手机（armv7或armv8）
+
+### 1.1 准备交叉编译环境
+交叉编译环境用于编译 Paddle Lite 和 PaddleOCR 的C++ demo。
+支持多种开发环境，不同开发环境的编译流程请参考对应文档。
+
+1. [Docker](https://paddle-lite-pjc.readthedocs.io/zh/latest/user_guides/source_compile.html#docker)
+2. [Linux](https://paddle-lite-pjc.readthedocs.io/zh/latest/user_guides/source_compile.html#linux)
+3. [MAC OS](https://paddle-lite-pjc.readthedocs.io/zh/latest/user_guides/source_compile.html#mac-os)
+
+### 1.2 准备预测库
+
+预测库有两种获取方式：
+- 1. [推荐]直接下载，预测库下载链接如下：
+
+      | 平台 | 预测库下载链接 |
+      |---|---|
+      |Android|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.android.armv7.gcc.c++_shared.with_extra.with_cv.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.android.armv8.gcc.c++_shared.with_extra.with_cv.tar.gz)|
+      |IOS|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.ios.armv7.with_cv.with_extra.with_log.tiny_publish.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.ios.armv8.with_cv.with_extra.with_log.tiny_publish.tar.gz)|
+
+      注：1. 上述预测库为PaddleLite 2.10分支编译得到，有关PaddleLite 2.10 详细信息可参考 [链接](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.10) 。
+
+**注：建议使用paddlelite>=2.10版本的预测库，其他预测库版本[下载链接](https://github.com/PaddlePaddle/Paddle-Lite/tags)**
+
+- 2. 编译Paddle-Lite得到预测库，Paddle-Lite的编译方式如下：
+```
+git clone https://github.com/PaddlePaddle/Paddle-Lite.git
+cd Paddle-Lite
+# 切换到Paddle-Lite release/v2.10 稳定分支
+git checkout release/v2.10
+./lite/tools/build_android.sh  --arch=armv8  --with_cv=ON --with_extra=ON
+```
+
+注意：编译Paddle-Lite获得预测库时，需要打开`--with_cv=ON --with_extra=ON`两个选项，`--arch`表示`arm`版本，这里指定为armv8，
+更多编译命令
+介绍请参考 [链接](https://paddle-lite-pjc.readthedocs.io/zh/latest/user_guides/source_compile.html#paddlelite) 。
+
+直接下载预测库并解压后，可以得到`inference_lite_lib.android.armv8/`文件夹，通过编译Paddle-Lite得到的预测库位于
+`Paddle-Lite/build.lite.android.armv8.gcc/inference_lite_lib.android.armv8/`文件夹下。
+预测库的文件目录如下：
+```
+inference_lite_lib.android.armv8/
+|-- cxx                                        C++ 预测库和头文件
+|   |-- include                                C++ 头文件
+|   |   |-- paddle_api.h
+|   |   |-- paddle_image_preprocess.h
+|   |   |-- paddle_lite_factory_helper.h
+|   |   |-- paddle_place.h
+|   |   |-- paddle_use_kernels.h
+|   |   |-- paddle_use_ops.h
+|   |   `-- paddle_use_passes.h
+|   `-- lib                                           C++预测库
+|       |-- libpaddle_api_light_bundled.a             C++静态库
+|       `-- libpaddle_light_api_shared.so             C++动态库
+|-- java                                     Java预测库
+|   |-- jar
+|   |   `-- PaddlePredictor.jar
+|   |-- so
+|   |   `-- libpaddle_lite_jni.so
+|   `-- src
+|-- demo                                     C++和Java示例代码
+|   |-- cxx                                  C++  预测库demo
+|   `-- java                                 Java 预测库demo
+```
+
+## 2 开始运行
+
+### 2.1 模型优化
+
+Paddle-Lite 提供了多种策略来自动优化原始的模型，其中包括量化、子图融合、混合调度、Kernel优选等方法，使用Paddle-lite的opt工具可以自动
+对inference模型进行优化，优化后的模型更轻量，模型运行速度更快。
+
+如果已经准备好了 `.nb` 结尾的模型文件，可以跳过此步骤。
+
+下述表格中也提供了一系列中文移动端模型：
+
+|模型版本|模型简介|模型大小|检测模型|文本方向分类模型|识别模型|Paddle-Lite版本|
+|---|---|---|---|---|---|---|
+|PP-OCRv3|蒸馏版超轻量中文OCR移动端模型|16.2M|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/PP-OCRv3_mobile_det_infer.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_infer_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_infer.nb)|v2.10|
+|PP-OCRv3(slim)|蒸馏版超轻量中文OCR移动端模型|5.9M|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_slim_infer.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_slim_infer.nb)|v2.10|
+|PP-OCRv2|蒸馏版超轻量中文OCR移动端模型|11M|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_infer_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_infer_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_infer_opt.nb)|v2.10|
+|PP-OCRv2(slim)|蒸馏版超轻量中文OCR移动端模型|4.6M|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_slim_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_slim_opt.nb)|v2.10|
+
+如果直接使用上述表格中的模型进行部署，可略过下述步骤，直接阅读 [2.2节](#2.2与手机联调)。
+
+如果要部署的模型不在上述表格中，则需要按照如下步骤获得优化后的模型。
+
+- 步骤1：参考[文档](https://www.paddlepaddle.org.cn/lite/v2.10/user_guides/opt/opt_python.html)安装paddlelite，用于转换paddle inference model为paddlelite运行所需的nb模型
+```
+pip install paddlelite==2.10  # paddlelite版本要与预测库版本一致
+```
+安装完后，如下指令可以查看帮助信息
+```
+paddle_lite_opt
+```
+
+paddle_lite_opt 参数介绍：
+|选项|说明|
+|---|---|
+|--model_dir|待优化的PaddlePaddle模型（非combined形式）的路径|
+|--model_file|待优化的PaddlePaddle模型（combined形式）的网络结构文件路径|
+|--param_file|待优化的PaddlePaddle模型（combined形式）的权重文件路径|
+|--optimize_out_type|输出模型类型，目前支持两种类型：protobuf和naive_buffer，其中naive_buffer是一种更轻量级的序列化/反序列化实现。若您需要在mobile端执行模型预测，请将此选项设置为naive_buffer。默认为protobuf|
+|--optimize_out|优化模型的输出路径|
+|--valid_targets|指定模型可执行的backend，默认为arm。目前可支持x86、arm、opencl、npu、xpu，可以同时指定多个backend(以空格分隔)，Model Optimize Tool将会自动选择最佳方式。如果需要支持华为NPU（Kirin 810/990 Soc搭载的达芬奇架构NPU），应当设置为npu, arm|
+|--record_tailoring_info|当使用 根据模型裁剪库文件 功能时，则设置该选项为true，以记录优化后模型含有的kernel和OP信息，默认为false|
+
+`--model_dir`适用于待优化的模型是非combined方式，PaddleOCR的inference模型是combined方式，即模型结构和模型参数使用单独一个文件存储。
+
+- 步骤2：使用paddle_lite_opt将inference模型转换成移动端模型格式。
+
+下面以PaddleOCR的超轻量中文模型为例，介绍使用编译好的opt文件完成inference模型到Paddle-Lite优化模型的转换。
+
+```
+# 【推荐】 下载 PP-OCRv3版本的中英文 inference模型
+wget  https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_slim_infer.tar && tar xf  ch_PP-OCRv3_det_slim_infer.tar
+wget  https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_slim_infer.tar && tar xf  ch_PP-OCRv2_rec_slim_quant_infer.tar
+wget  https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_cls_slim_infer.tar && tar xf  ch_ppocr_mobile_v2.0_cls_slim_infer.tar
+# 转换检测模型
+paddle_lite_opt --model_file=./ch_PP-OCRv3_det_slim_infer/inference.pdmodel  --param_file=./ch_PP-OCRv3_det_slim_infer/inference.pdiparams  --optimize_out=./ch_PP-OCRv3_det_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
+# 转换识别模型
+paddle_lite_opt --model_file=./ch_PP-OCRv3_rec_slim_infer/inference.pdmodel  --param_file=./ch_PP-OCRv3_rec_slim_infer/inference.pdiparams  --optimize_out=./ch_PP-OCRv3_rec_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
+# 转换方向分类器模型
+paddle_lite_opt --model_file=./ch_ppocr_mobile_v2.0_cls_slim_infer/inference.pdmodel  --param_file=./ch_ppocr_mobile_v2.0_cls_slim_infer/inference.pdiparams  --optimize_out=./ch_ppocr_mobile_v2.0_cls_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
+
+```
+
+转换成功后，inference模型目录下会多出`.nb`结尾的文件，即是转换成功的模型文件。
+
+注意：使用paddle-lite部署时，需要使用opt工具优化后的模型。 opt工具的输入模型是paddle保存的inference模型
+
+<a name="2.2与手机联调"></a>
+### 2.2 与手机联调
+
+首先需要进行一些准备工作。
+ 1. 准备一台arm8的安卓手机，如果编译的预测库和opt文件是armv7，则需要arm7的手机，并修改Makefile中`ARM_ABI = arm7`。
+ 2. 打开手机的USB调试选项，选择文件传输模式，连接电脑。
+ 3. 电脑上安装adb工具，用于调试。 adb安装方式如下：
+
+    3.1. MAC电脑安装ADB:
+    ```
+    brew cask install android-platform-tools
+    ```
+    3.2. Linux安装ADB
+    ```
+    sudo apt update
+    sudo apt install -y wget adb
+    ```
+    3.3. Window安装ADB
+
+    win上安装需要去谷歌的安卓平台下载adb软件包进行安装：[链接](https://developer.android.com/studio)
+
+    打开终端，手机连接电脑，在终端中输入
+    ```
+    adb devices
+    ```
+    如果有device输出，则表示安装成功。
+    ```
+       List of devices attached
+       744be294    device
+    ```
+
+ 4. 准备优化后的模型、预测库文件、测试图像和使用的字典文件。
+ ```
+ git clone https://github.com/PaddlePaddle/PaddleOCR.git
+ cd PaddleOCR/deploy/lite/
+ # 运行prepare.sh，准备预测库文件、测试图像和使用的字典文件，并放置在预测库中的demo/cxx/ocr文件夹下
+ sh prepare.sh /{lite prediction library path}/inference_lite_lib.android.armv8
+
+ # 进入OCR demo的工作目录
+ cd /{lite prediction library path}/inference_lite_lib.android.armv8/
+ cd demo/cxx/ocr/
+ # 将C++预测动态库so文件复制到debug文件夹中
+ cp ../../../cxx/lib/libpaddle_light_api_shared.so ./debug/
+ ```
+
+ 准备测试图像，以`PaddleOCR/doc/imgs/11.jpg`为例，将测试的图像复制到`demo/cxx/ocr/debug/`文件夹下。
+ 准备lite opt工具优化后的模型文件，比如使用`ch_PP-OCRv3_det_slim_opt.ch_PP-OCRv3_rec_slim_rec.nb, ch_ppocr_mobile_v2.0_cls_slim_opt.nb`，模型文件放置在`demo/cxx/ocr/debug/`文件夹下。
+
+ 执行完成后，ocr文件夹下将有如下文件格式：
+
+```
+demo/cxx/ocr/
+|-- debug/
+|   |--ch_PP-OCRv3_det_slim_opt.nb           优化后的检测模型文件
+|   |--ch_PP-OCRv3_rec_slim_opt.nb           优化后的识别模型文件
+|   |--ch_ppocr_mobile_v2.0_cls_slim_opt.nb           优化后的文字方向分类器模型文件
+|   |--11.jpg                           待测试图像
+|   |--ppocr_keys_v1.txt                中文字典文件
+|   |--libpaddle_light_api_shared.so    C++预测库文件
+|   |--config.txt                       超参数配置
+|-- config.txt                  超参数配置
+|-- cls_process.cc              方向分类器的预处理和后处理文件
+|-- cls_process.h
+|-- crnn_process.cc             识别模型CRNN的预处理和后处理文件
+|-- crnn_process.h
+|-- db_post_process.cc          检测模型DB的后处理文件
+|-- db_post_process.h
+|-- Makefile                    编译文件
+|-- ocr_db_crnn.cc              C++预测源文件
+```
+
+#### 注意：
+1. ppocr_keys_v1.txt是中文字典文件，如果使用的 nb 模型是英文数字或其他语言的模型，需要更换为对应语言的字典。
+PaddleOCR 在ppocr/utils/下存放了多种字典，包括：
+```
+dict/french_dict.txt     # 法语字典
+dict/german_dict.txt     # 德语字典
+ic15_dict.txt       # 英文字典
+dict/japan_dict.txt      # 日语字典
+dict/korean_dict.txt     # 韩语字典
+ppocr_keys_v1.txt   # 中文字典
+...
+```
+
+2.  `config.txt` 包含了检测器、分类器、识别器的超参数，如下：
+```
+max_side_len  960         # 输入图像长宽大于960时，等比例缩放图像，使得图像最长边为960
+det_db_thresh  0.3        # 用于过滤DB预测的二值化图像，设置为0.-0.3对结果影响不明显
+det_db_box_thresh  0.5    # 检测器后处理过滤box的阈值，如果检测存在漏框情况，可酌情减小
+det_db_unclip_ratio  1.6  # 表示文本框的紧致程度，越小则文本框更靠近文本
+use_direction_classify  0  # 是否使用方向分类器，0表示不使用，1表示使用
+rec_image_height  48      # 识别模型输入图像的高度，PP-OCRv3模型设置为48，PP-OCRv2模型需要设置为32
+```
+
+ 5. 启动调试
+
+ 上述步骤完成后就可以使用adb将文件push到手机上运行，步骤如下：
+
+ ```
+ # 执行编译，得到可执行文件ocr_db_crnn, 第一次执行此命令会下载opencv等依赖库，下载完成后，需要再执行一次
+ make -j
+
+ # 将编译的可执行文件移动到debug文件夹中
+ mv ocr_db_crnn ./debug/
+ # 将debug文件夹push到手机上
+ adb push debug /data/local/tmp/
+ adb shell
+ cd /data/local/tmp/debug
+ export LD_LIBRARY_PATH=${PWD}:$LD_LIBRARY_PATH
+ # 开始使用，ocr_db_crnn可执行文件的使用方式为:
+ # ./ocr_db_crnn 预测模式  检测模型文件 方向分类器模型文件  识别模型文件 运行硬件 运行精度 线程数  batchsize  测试图像路径  参数配置路径  字典文件路径 是否使用benchmark参数
+ ./ocr_db_crnn system  ch_PP-OCRv3_det_slim_opt.nb  ch_PP-OCRv3_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_slim_opt.nb  arm8 INT8 10 1  ./11.jpg  config.txt  ppocr_keys_v1.txt  True
+
+# 仅使用文本检测模型，使用方式如下：
+./ocr_db_crnn  det ch_PP-OCRv3_det_slim_opt.nb arm8 INT8 10 1 ./11.jpg  config.txt
+
+# 仅使用文本识别模型，使用方式如下：
+./ocr_db_crnn  rec ch_PP-OCRv3_rec_slim_opt.nb arm8 INT8 10 1 word_1.jpg ppocr_keys_v1.txt config.txt
+ ```
+
+ 如果对代码做了修改，则需要重新编译并push到手机上。
+
+ 运行效果如下：
+
+<div align="center">
+    <img src="imgs/lite_demo.png" width="600">
+</div>
+
+
+## FAQ
+Q1：如果想更换模型怎么办，需要重新按照流程走一遍吗？
+
+A1：如果已经走通了上述步骤，更换模型只需要替换 .nb 模型文件即可，同时要注意更新字典
+
+Q2：换一个图测试怎么做？
+
+A2：替换debug下的.jpg测试图像为你想要测试的图像，adb push 到手机上即可
+
+Q3：如何封装到手机APP中？
+
+A3：此demo旨在提供能在手机上运行OCR的核心算法部分，PaddleOCR/deploy/android_demo是将这个demo封装到手机app的示例，供参考
+
+Q4：运行demo时遇到报错`Error: This model is not supported, because kernel for 'io_copy' is not supported by Paddle-Lite.`
+
+A4：问题是安装的paddlelite版本和下载的预测库版本不匹配，确保paddleliteopt工具和你的预测库版本匹配，重新转nb模型试试。
diff --git a/deploy/paddle2onnx/images/lite_demo_onnx.png b/deploy/paddle2onnx/images/lite_demo_onnx.png
new file mode 100644
index 0000000..b096f6e
Binary files /dev/null and b/deploy/paddle2onnx/images/lite_demo_onnx.png differ
diff --git a/deploy/paddle2onnx/images/lite_demo_paddle.png b/deploy/paddle2onnx/images/lite_demo_paddle.png
new file mode 100644
index 0000000..b096f6e
Binary files /dev/null and b/deploy/paddle2onnx/images/lite_demo_paddle.png differ
diff --git a/deploy/paddle2onnx/readme.md b/deploy/paddle2onnx/readme.md
new file mode 100644
index 0000000..785db66
--- /dev/null
+++ b/deploy/paddle2onnx/readme.md
@@ -0,0 +1,174 @@
+# Paddle2ONNX model transformation and prediction
+
+This chapter describes how the PaddleOCR model is converted into an ONNX model and predicted based on the ONNXRuntime engine.
+
+## 1. Environment preparation
+
+Need to prepare PaddleOCR, Paddle2ONNX model conversion environment, and ONNXRuntime prediction environment
+
+###  PaddleOCR
+
+Clone the PaddleOCR repository, use the main branch, and install it.
+
+```
+git clone  -b main https://github.com/PaddlePaddle/PaddleOCR.git
+cd PaddleOCR && python3 pip install -e .
+```
+
+###  Paddle2ONNX
+
+Paddle2ONNX supports converting the PaddlePaddle model format to the ONNX model format. The operator currently supports exporting ONNX Opset 9~11 stably, and some Paddle operators support lower ONNX Opset conversion.
+For more details, please refer to [Paddle2ONNX](https://github.com/PaddlePaddle/Paddle2ONNX/blob/develop/README_en.md)
+
+
+- install Paddle2ONNX
+```
+python3 -m pip install paddle2onnx
+```
+
+- install ONNXRuntime
+```
+python3 -m pip install onnxruntime
+```
+
+## 2. Model conversion
+
+
+- Paddle model download
+
+There are two ways to obtain the Paddle model: Download the prediction model provided by PaddleOCR in [model_list](../../doc/doc_en/models_list_en.md);
+Refer to [Model Export Instructions](../../doc/doc_en/inference_en.md#1-convert-training-model-to-inference-model) to convert the trained weights to inference_model.
+
+Take the PP-OCRv3 detection, recognition, and classification model as an example:
+
+```
+wget -nc -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar
+cd ./inference && tar xf en_PP-OCRv3_det_infer.tar && cd ..
+
+wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_infer.tar
+cd ./inference && tar xf en_PP-OCRv3_rec_infer.tar && cd ..
+
+wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar
+cd ./inference && tar xf ch_ppocr_mobile_v2.0_cls_infer.tar && cd ..
+```
+
+- convert model
+
+Convert Paddle inference model to ONNX model format using Paddle2ONNX:
+
+```
+paddle2onnx --model_dir ./inference/en_PP-OCRv3_det_infer \
+--model_filename inference.pdmodel \
+--params_filename inference.pdiparams \
+--save_file ./inference/det_onnx/model.onnx \
+--opset_version 11 \
+--enable_onnx_checker True
+
+paddle2onnx --model_dir ./inference/en_PP-OCRv3_rec_infer \
+--model_filename inference.pdmodel \
+--params_filename inference.pdiparams \
+--save_file ./inference/rec_onnx/model.onnx \
+--opset_version 11 \
+--enable_onnx_checker True
+
+paddle2onnx --model_dir ./inference/ch_ppocr_mobile_v2.0_cls_infer \
+--model_filename inference.pdmodel \
+--params_filename inference.pdiparams \
+--save_file ./inference/cls_onnx/model.onnx \
+--opset_version 11 \
+--enable_onnx_checker True
+```
+After execution, the ONNX model will be saved in `./inference/det_onnx/`, `./inference/rec_onnx/`, `./inference/cls_onnx/` paths respectively
+
+* Note: For the OCR model, the conversion process must be in the form of dynamic shape, otherwise the prediction result may be the same as Predicting directly with Paddle is slightly different.
+  In addition, the following models do not currently support conversion to ONNX models:
+  NRTR, SAR, RARE, SRN
+
+* Note: The current Paddle2ONNX version (v1.2.3) now supports dynamic shapes by default, i.e., float32[p2o.DynamicDimension.0,3,p2o.DynamicDimension.1,p2o.DynamicDimension.2]. The `--input_shape_dict` option has been deprecated. If you need to adjust the shape, you can use the following command to adjust the input shape of the Paddle model.
+
+  ```
+  python3 -m paddle2onnx.optimize --input_model inference/det_onnx/model.onnx \
+    --output_model inference/det_onnx/model.onnx \
+    --input_shape_dict "{'x': [-1,3,-1,-1]}"
+  ```
+
+## 3. prediction
+
+Take the English OCR model as an example, use **ONNXRuntime** to predict and execute the following commands:
+
+```
+python3 tools/infer/predict_system.py --use_gpu=False --use_onnx=True \
+--det_model_dir=./inference/det_onnx/model.onnx  \
+--rec_model_dir=./inference/rec_onnx/model.onnx  \
+--cls_model_dir=./inference/cls_onnx/model.onnx  \
+--image_dir=doc/imgs_en/img_12.jpg \
+--rec_char_dict_path=ppocr/utils/en_dict.txt
+```
+
+Taking the English OCR model as an example, use **Paddle Inference** to predict and execute the following commands:
+
+```
+python3 tools/infer/predict_system.py --use_gpu=False \
+--cls_model_dir=./inference/ch_ppocr_mobile_v2.0_cls_infer \
+--rec_model_dir=./inference/en_PP-OCRv3_rec_infer \
+--det_model_dir=./inference/en_PP-OCRv3_det_infer \
+--image_dir=doc/imgs_en/img_12.jpg \
+--rec_char_dict_path=ppocr/utils/en_dict.txt
+```
+
+
+After executing the command, the predicted identification information will be printed out in the terminal, and the visualization results will be saved under `./inference_results/`.
+
+ONNXRuntime result：
+
+<div align="center">
+    <img src="../../docs/version2.x/ppocr/images/multi_lang/img_12.jpg" width=800">
+</div>
+
+Paddle Inference result：
+
+<div align="center">
+    <img src="../../docs/version2.x/ppocr/images/multi_lang/img_12.jpg" width=800">
+</div>
+
+
+Using ONNXRuntime to predict, terminal output:
+```
+[2022/10/10 12:06:28] ppocr DEBUG: dt_boxes num : 11, elapse : 0.3568880558013916
+[2022/10/10 12:06:31] ppocr DEBUG: rec_res num  : 11, elapse : 2.6445000171661377
+[2022/10/10 12:06:31] ppocr DEBUG: 0  Predict time of doc/imgs_en/img_12.jpg: 3.021s
+[2022/10/10 12:06:31] ppocr DEBUG: ACKNOWLEDGEMENTS, 0.997
+[2022/10/10 12:06:31] ppocr DEBUG: We would like to thank all the designers and, 0.976
+[2022/10/10 12:06:31] ppocr DEBUG: contributors who have been involved in the, 0.979
+[2022/10/10 12:06:31] ppocr DEBUG: production of this book; their contributions, 0.989
+[2022/10/10 12:06:31] ppocr DEBUG: have been indispensable to its creation. We, 0.956
+[2022/10/10 12:06:31] ppocr DEBUG: would also like to express our gratitude to all, 0.991
+[2022/10/10 12:06:31] ppocr DEBUG: the producers for their invaluable opinions, 0.978
+[2022/10/10 12:06:31] ppocr DEBUG: and assistance throughout this project. And to, 0.988
+[2022/10/10 12:06:31] ppocr DEBUG: the many others whose names are not credited, 0.958
+[2022/10/10 12:06:31] ppocr DEBUG: but have made specific input in this book, we, 0.970
+[2022/10/10 12:06:31] ppocr DEBUG: thank you for your continuous support., 0.998
+[2022/10/10 12:06:31] ppocr DEBUG: The visualized image saved in ./inference_results/img_12.jpg
+[2022/10/10 12:06:31] ppocr INFO: The predict total time is 3.2482550144195557
+```
+
+Using Paddle Inference to predict, terminal output:
+
+```
+[2022/10/10 12:06:28] ppocr DEBUG: dt_boxes num : 11, elapse : 0.3568880558013916
+[2022/10/10 12:06:31] ppocr DEBUG: rec_res num  : 11, elapse : 2.6445000171661377
+[2022/10/10 12:06:31] ppocr DEBUG: 0  Predict time of doc/imgs_en/img_12.jpg: 3.021s
+[2022/10/10 12:06:31] ppocr DEBUG: ACKNOWLEDGEMENTS, 0.997
+[2022/10/10 12:06:31] ppocr DEBUG: We would like to thank all the designers and, 0.976
+[2022/10/10 12:06:31] ppocr DEBUG: contributors who have been involved in the, 0.979
+[2022/10/10 12:06:31] ppocr DEBUG: production of this book; their contributions, 0.989
+[2022/10/10 12:06:31] ppocr DEBUG: have been indispensable to its creation. We, 0.956
+[2022/10/10 12:06:31] ppocr DEBUG: would also like to express our gratitude to all, 0.991
+[2022/10/10 12:06:31] ppocr DEBUG: the producers for their invaluable opinions, 0.978
+[2022/10/10 12:06:31] ppocr DEBUG: and assistance throughout this project. And to, 0.988
+[2022/10/10 12:06:31] ppocr DEBUG: the many others whose names are not credited, 0.958
+[2022/10/10 12:06:31] ppocr DEBUG: but have made specific input in this book, we, 0.970
+[2022/10/10 12:06:31] ppocr DEBUG: thank you for your continuous support., 0.998
+[2022/10/10 12:06:31] ppocr DEBUG: The visualized image saved in ./inference_results/img_12.jpg
+[2022/10/10 12:06:31] ppocr INFO: The predict total time is 3.2482550144195557
+```
diff --git a/deploy/paddle2onnx/readme_ch.md b/deploy/paddle2onnx/readme_ch.md
new file mode 100644
index 0000000..51a71b5
--- /dev/null
+++ b/deploy/paddle2onnx/readme_ch.md
@@ -0,0 +1,220 @@
+# Paddle2ONNX模型转化与预测
+
+本章节介绍 PaddleOCR 模型如何转化为 ONNX 模型，并基于 ONNXRuntime 引擎预测。
+
+## 1. 环境准备
+
+需要准备 PaddleOCR、Paddle2ONNX 模型转化环境，和 ONNXRuntime 预测环境
+
+###  PaddleOCR
+
+克隆PaddleOCR的仓库，使用 main 分支，并进行安装，由于 PaddleOCR 仓库比较大，git clone 速度比较慢，所以本教程已下载
+
+```
+git clone  -b main https://github.com/PaddlePaddle/PaddleOCR.git
+cd PaddleOCR && python3 -m pip install -e .
+```
+
+###  Paddle2ONNX
+
+Paddle2ONNX 支持将 PaddlePaddle 模型格式转化到 ONNX 模型格式，算子目前稳定支持导出 ONNX Opset 9~18，部分Paddle算子支持更低的ONNX Opset转换。
+更多细节可参考 [Paddle2ONNX](https://github.com/PaddlePaddle/Paddle2ONNX/blob/develop/README_zh.md)
+
+- 安装 Paddle2ONNX
+```
+python3 -m pip install paddle2onnx
+```
+
+- 安装 ONNXRuntime
+```
+python3 -m pip install onnxruntime
+```
+
+## 2. 模型转换
+
+
+- Paddle 模型下载
+
+有两种方式获取Paddle静态图模型：在 [model_list](../../doc/doc_ch/models_list.md) 中下载PaddleOCR提供的预测模型；
+参考[模型导出说明](../../doc/doc_ch/inference.md#训练模型转inference模型)把训练好的权重转为 inference_model。
+
+以 PP-OCRv3 中文检测、识别、分类模型为例：
+
+```
+wget -nc  -P ./inference https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar
+cd ./inference && tar xf PP-OCRv3_mobile_det_infer.tar && cd ..
+
+wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_infer.tar
+cd ./inference && tar xf ch_PP-OCRv3_rec_infer.tar && cd ..
+
+wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar
+cd ./inference && tar xf ch_ppocr_mobile_v2.0_cls_infer.tar && cd ..
+```
+
+- 模型转换
+
+使用 Paddle2ONNX 将Paddle静态图模型转换为ONNX模型格式：
+
+```
+paddle2onnx --model_dir ./inference/PP-OCRv3_mobile_det_infer \
+--model_filename inference.pdmodel \
+--params_filename inference.pdiparams \
+--save_file ./inference/det_onnx/model.onnx \
+--opset_version 11 \
+--enable_onnx_checker True
+
+paddle2onnx --model_dir ./inference/ch_PP-OCRv3_rec_infer \
+--model_filename inference.pdmodel \
+--params_filename inference.pdiparams \
+--save_file ./inference/rec_onnx/model.onnx \
+--opset_version 11 \
+--enable_onnx_checker True
+
+paddle2onnx --model_dir ./inference/ch_ppocr_mobile_v2.0_cls_infer \
+--model_filename inference.pdmodel \
+--params_filename inference.pdiparams \
+--save_file ./inference/cls_onnx/model.onnx \
+--opset_version 11 \
+--enable_onnx_checker True
+```
+
+执行完毕后，ONNX 模型会被分别保存在 `./inference/det_onnx/`，`./inference/rec_onnx/`，`./inference/cls_onnx/`路径下
+
+* 注意：对于OCR模型，转化过程中必须采用动态shape的形式，否则预测结果可能与直接使用Paddle预测有细微不同。
+  另外，以下几个模型暂不支持转换为 ONNX 模型：
+  NRTR、SAR、RARE、SRN
+
+* 注意：[当前Paddle2ONNX版本(v1.2.3)](https://github.com/PaddlePaddle/Paddle2ONNX/releases/tag/v1.2.3)现已默认支持动态shape，即 `float32[p2o.DynamicDimension.0,3,p2o.DynamicDimension.1,p2o.DynamicDimension.2]`，选项 `--input_shape_dict` 已废弃。如果有shape调整需求可使用如下命令进行Paddle模型输入shape调整。
+
+  ```
+  python3 -m paddle2onnx.optimize --input_model inference/det_onnx/model.onnx \
+    --output_model inference/det_onnx/model.onnx \
+    --input_shape_dict "{'x': [-1,3,-1,-1]}"
+  ```
+
+## 3. 推理预测
+
+以中文OCR模型为例，使用 ONNXRuntime 预测可执行如下命令：
+
+```
+python3 tools/infer/predict_system.py --use_gpu=False --use_onnx=True \
+--det_model_dir=./inference/det_onnx/model.onnx  \
+--rec_model_dir=./inference/rec_onnx/model.onnx  \
+--cls_model_dir=./inference/cls_onnx/model.onnx  \
+--image_dir=./deploy/lite/imgs/lite_demo.png
+```
+
+以中文OCR模型为例，使用 Paddle Inference 预测可执行如下命令：
+
+```
+python3 tools/infer/predict_system.py --use_gpu=False \
+--cls_model_dir=./inference/ch_ppocr_mobile_v2.0_cls_infer \
+--rec_model_dir=./inference/ch_PP-OCRv3_rec_infer \
+--det_model_dir=./inference/PP-OCRv3_mobile_det_infer \
+--image_dir=./deploy/lite/imgs/lite_demo.png
+```
+
+
+执行命令后在终端会打印出预测的识别信息，并在 `./inference_results/` 下保存可视化结果。
+
+ONNXRuntime 执行效果：
+
+<div align="center">
+    <img src="./images/lite_demo_onnx.png" width=800">
+</div>
+
+Paddle Inference 执行效果：
+
+<div align="center">
+    <img src="./images/lite_demo_paddle.png" width=800">
+</div>
+
+
+使用 ONNXRuntime 预测，终端输出：
+```
+[2022/02/22 17:48:27] root DEBUG: dt_boxes num : 38, elapse : 0.043187856674194336
+[2022/02/22 17:48:27] root DEBUG: rec_res num  : 38, elapse : 0.592170000076294
+[2022/02/22 17:48:27] root DEBUG: 0  Predict time of ./deploy/lite/imgs/lite_demo.png: 0.642s
+[2022/02/22 17:48:27] root DEBUG: The, 0.984
+[2022/02/22 17:48:27] root DEBUG: visualized, 0.882
+[2022/02/22 17:48:27] root DEBUG: etect18片, 0.720
+[2022/02/22 17:48:27] root DEBUG: image saved in./vis.jpg, 0.947
+[2022/02/22 17:48:27] root DEBUG: 纯臻营养护发素0.993604, 0.996
+[2022/02/22 17:48:27] root DEBUG: 产品信息/参数, 0.922
+[2022/02/22 17:48:27] root DEBUG: 0.992728, 0.914
+[2022/02/22 17:48:27] root DEBUG: （45元／每公斤，100公斤起订）, 0.926
+[2022/02/22 17:48:27] root DEBUG: 0.97417, 0.977
+[2022/02/22 17:48:27] root DEBUG: 每瓶22元，1000瓶起订）0.993976, 0.962
+[2022/02/22 17:48:27] root DEBUG: 【品牌】：代加工方式/0EMODM, 0.945
+[2022/02/22 17:48:27] root DEBUG: 0.985133, 0.980
+[2022/02/22 17:48:27] root DEBUG: 【品名】：纯臻营养护发素, 0.921
+[2022/02/22 17:48:27] root DEBUG: 0.995007, 0.883
+[2022/02/22 17:48:27] root DEBUG: 【产品编号】：YM-X-30110.96899, 0.955
+[2022/02/22 17:48:27] root DEBUG: 【净含量】：220ml, 0.943
+[2022/02/22 17:48:27] root DEBUG: Q.996577, 0.932
+[2022/02/22 17:48:27] root DEBUG: 【适用人群】：适合所有肤质, 0.913
+[2022/02/22 17:48:27] root DEBUG: 0.995842, 0.969
+[2022/02/22 17:48:27] root DEBUG: 【主要成分】：鲸蜡硬脂醇、燕麦B-葡聚, 0.883
+[2022/02/22 17:48:27] root DEBUG: 0.961928, 0.964
+[2022/02/22 17:48:27] root DEBUG: 10, 0.812
+[2022/02/22 17:48:27] root DEBUG: 糖、椰油酰胺丙基甜菜碱、泛醒, 0.866
+[2022/02/22 17:48:27] root DEBUG: 0.925898, 0.943
+[2022/02/22 17:48:27] root DEBUG: （成品包材）, 0.974
+[2022/02/22 17:48:27] root DEBUG: 0.972573, 0.961
+[2022/02/22 17:48:27] root DEBUG: 【主要功能】：可紧致头发磷层，从而达到, 0.936
+[2022/02/22 17:48:27] root DEBUG: 0.994448, 0.952
+[2022/02/22 17:48:27] root DEBUG: 13, 0.998
+[2022/02/22 17:48:27] root DEBUG: 即时持久改善头发光泽的效果，给干燥的头, 0.994
+[2022/02/22 17:48:27] root DEBUG: 0.990198, 0.975
+[2022/02/22 17:48:27] root DEBUG: 14, 0.977
+[2022/02/22 17:48:27] root DEBUG: 发足够的滋养, 0.991
+[2022/02/22 17:48:27] root DEBUG: 0.997668, 0.918
+[2022/02/22 17:48:27] root DEBUG: 花费了0.457335秒, 0.901
+[2022/02/22 17:48:27] root DEBUG: The visualized image saved in ./inference_results/lite_demo.png
+[2022/02/22 17:48:27] root INFO: The predict total time is 0.7003889083862305
+```
+
+使用 Paddle Inference 预测，终端输出：
+
+```
+[2022/02/22 17:47:25] root DEBUG: dt_boxes num : 38, elapse : 0.11791276931762695
+[2022/02/22 17:47:27] root DEBUG: rec_res num  : 38, elapse : 2.6206860542297363
+[2022/02/22 17:47:27] root DEBUG: 0  Predict time of ./deploy/lite/imgs/lite_demo.png: 2.746s
+[2022/02/22 17:47:27] root DEBUG: The, 0.984
+[2022/02/22 17:47:27] root DEBUG: visualized, 0.882
+[2022/02/22 17:47:27] root DEBUG: etect18片, 0.720
+[2022/02/22 17:47:27] root DEBUG: image saved in./vis.jpg, 0.947
+[2022/02/22 17:47:27] root DEBUG: 纯臻营养护发素0.993604, 0.996
+[2022/02/22 17:47:27] root DEBUG: 产品信息/参数, 0.922
+[2022/02/22 17:47:27] root DEBUG: 0.992728, 0.914
+[2022/02/22 17:47:27] root DEBUG: （45元／每公斤，100公斤起订）, 0.926
+[2022/02/22 17:47:27] root DEBUG: 0.97417, 0.977
+[2022/02/22 17:47:27] root DEBUG: 每瓶22元，1000瓶起订）0.993976, 0.962
+[2022/02/22 17:47:27] root DEBUG: 【品牌】：代加工方式/0EMODM, 0.945
+[2022/02/22 17:47:27] root DEBUG: 0.985133, 0.980
+[2022/02/22 17:47:27] root DEBUG: 【品名】：纯臻营养护发素, 0.921
+[2022/02/22 17:47:27] root DEBUG: 0.995007, 0.883
+[2022/02/22 17:47:27] root DEBUG: 【产品编号】：YM-X-30110.96899, 0.955
+[2022/02/22 17:47:27] root DEBUG: 【净含量】：220ml, 0.943
+[2022/02/22 17:47:27] root DEBUG: Q.996577, 0.932
+[2022/02/22 17:47:27] root DEBUG: 【适用人群】：适合所有肤质, 0.913
+[2022/02/22 17:47:27] root DEBUG: 0.995842, 0.969
+[2022/02/22 17:47:27] root DEBUG: 【主要成分】：鲸蜡硬脂醇、燕麦B-葡聚, 0.883
+[2022/02/22 17:47:27] root DEBUG: 0.961928, 0.964
+[2022/02/22 17:47:27] root DEBUG: 10, 0.812
+[2022/02/22 17:47:27] root DEBUG: 糖、椰油酰胺丙基甜菜碱、泛醒, 0.866
+[2022/02/22 17:47:27] root DEBUG: 0.925898, 0.943
+[2022/02/22 17:47:27] root DEBUG: （成品包材）, 0.974
+[2022/02/22 17:47:27] root DEBUG: 0.972573, 0.961
+[2022/02/22 17:47:27] root DEBUG: 【主要功能】：可紧致头发磷层，从而达到, 0.936
+[2022/02/22 17:47:27] root DEBUG: 0.994448, 0.952
+[2022/02/22 17:47:27] root DEBUG: 13, 0.998
+[2022/02/22 17:47:27] root DEBUG: 即时持久改善头发光泽的效果，给干燥的头, 0.994
+[2022/02/22 17:47:27] root DEBUG: 0.990198, 0.975
+[2022/02/22 17:47:27] root DEBUG: 14, 0.977
+[2022/02/22 17:47:27] root DEBUG: 发足够的滋养, 0.991
+[2022/02/22 17:47:27] root DEBUG: 0.997668, 0.918
+[2022/02/22 17:47:27] root DEBUG: 花费了0.457335秒, 0.901
+[2022/02/22 17:47:27] root DEBUG: The visualized image saved in ./inference_results/lite_demo.png
+[2022/02/22 17:47:27] root INFO: The predict total time is 2.8338775634765625
+```
diff --git a/deploy/paddlecloud/README.md b/deploy/paddlecloud/README.md
new file mode 100644
index 0000000..7ff58ce
--- /dev/null
+++ b/deploy/paddlecloud/README.md
@@ -0,0 +1,339 @@
+# 云上飞桨部署工具
+
+[云上飞桨（PaddleCloud）](https://github.com/PaddlePaddle/PaddleCloud) 是面向飞桨框架及其模型套件的部署工具，
+为用户提供了模型套件Docker化部署和Kubernetes集群部署两种方式，可以满足不同场景与环境的部署需求。
+本章节我们将使用PaddleCloud提供的OCR标准镜像以及云原生组件来训练和部署PP-OCRv3识别模型。
+
+## 云上飞桨部署工具的优势
+
+<div align="center">
+  <img src="./images/paddlecloud.png" title="architecture" width="80%" height="80%" alt="">
+</div>
+
+- **模型套件Docker镜像大礼包。**
+
+  PaddleCloud为用户提供了飞桨模型套件Docker镜像大礼包，这些镜像中包含运行模型套件案例的所有依赖并能持续更新，支持异构硬件环境和常见CUDA版本、开箱即用。
+
+- **具有丰富的云上飞桨组件。**
+
+  云上飞桨具有丰富的云原生功能组件，包括样本数据缓存组件、分布式训练组件、推理服务组件等，使用这些组件用户可以快速地在Kubernetes集群上进行训练和部署工作。
+
+- **功能强大的自运维能力。**
+
+  云上飞桨组件基于Kubernetes的Operator机制提供了功能强大的自运维能力，如训练组件支持多种架构模式并具有分布式容错与弹性训练的能力，推理服务组件支持自动扩缩容与蓝绿发版等。
+
+- **针对飞桨框架的定制优化。**
+
+  除了部署便捷与自运维的优势，PaddleCloud还针对飞桨框架进行了正对性优化，如通过缓存样本数据来加速云上飞桨分布式训练作业、基于飞桨框架和调度器的协同设计来优化集群GPU利用率等。
+
+
+## 1. PP-OCRv3 Docker化部署
+
+PaddleCloud基于 [Tekton](https://github.com/tektoncd/pipeline) 为OCR模型套件提供了镜像持续构建的能力，并支持CPU、GPU以及常见CUDA版本的镜像。
+您可以查看 [PaddleOCR 镜像仓库](https://hub.docker.com/repository/docker/paddlecloud/paddleocr) 来获取所有的镜像列表。
+同时我们也将PP-OCRv3识别模型的训练与推理实战案例放置到了AI Studio平台上，您可以点击 [PP-OCRv3识别训推一体项目实战](https://aistudio.baidu.com/aistudio/projectdetail/3916206?channelType=0&channel=0) 在平台上快速体验。
+
+> **适用场景**：本地测试开发环境、单机部署环境。
+
+### 1.1 安装Docker
+
+如果您所使用的机器上还没有安装 Docker，您可以参考 [Docker 官方文档](https://docs.docker.com/get-docker/) 来进行安装。
+如果您需要使用支持 GPU 版本的镜像，则还需安装好NVIDIA相关驱动和 [nvidia-docker](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html#docker) 。
+
+**注意**：如果您使用的是Windows系统，需要开启 [WSL2（Linux子系统功能）功能](https://docs.microsoft.com/en-us/windows/wsl/install)。
+
+### 1.2 启动容器
+
+**使用CPU版本的Docker镜像**
+
+```bash
+# 这是加上参数 --shm-size=32g 是为了防止容器里内存不足
+docker run --name ppocr -v $PWD:/mnt -p 8888:8888 -it --shm-size=32g paddlecloud/paddleocr:2.5-cpu-efbb0a /bin/bash
+```
+
+**使用GPU版本的Docker镜像**
+
+```bash
+docker run --name ppocr --runtime=nvidia -v $PWD:/mnt -p 8888:8888 -it --shm-size=32g paddlecloud/paddleocr:2.5-gpu-cuda10.2-cudnn7-efbb0a /bin/bash
+```
+
+进入容器内，则可进行 PP-OCRv3 模型的训练和部署工作。
+
+### 1.3 准备训练数据
+
+本教程以HierText数据集为例，HierText是第一个具有自然场景和文档中文本分层注释的数据集。
+该数据集包含从 Open Images 数据集中选择的 11639 张图像，提供高质量的单词 (~1.2M)、行和段落级别的注释。
+我们已经将数据集上传到百度云对象存储（BOS），您可以通过运行如下指令，完成数据集的下载和解压操作：
+
+```bash
+# 下载数据集
+$ wget -P /mnt https://paddleflow-public.hkg.bcebos.com/ppocr/hiertext1.tar
+
+# 解压数据集
+$ tar xf /mnt/hiertext1.tar -C /mnt && mv /mnt/hiertext1 /mnt/hiertext
+```
+
+运行上述命令后，在 `/mnt` 目录下包含以下文件：
+
+```
+/mnt/hiertext
+  └─ train/     HierText训练集数据
+  └─ validation/     HierText验证集数据
+  └─ label_hiertext_train.txt  HierText训练集的行标注
+  └─ label_hiertext_val.txt    HierText验证集的行标注
+```
+
+### 1.4 修改配置文件
+
+PP-OCRv3模型配置文件位于`/home/PaddleOCR/configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml`，需要修改的配置如下：
+
+- 修改训练数据配置：
+
+```yaml
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+```
+
+修改为：
+
+```yaml
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: /mnt/
+    label_file_list:
+      - /mnt/hiertext/label_hiertext_train.txt
+```
+
+- 修改验证数据配置：
+
+```yaml
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+```
+
+修改为：
+
+```yaml
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: /mnt/
+    label_file_list:
+      - /mnt/hiertext/label_hiertext_val.txt
+```
+
+### 1.5 启动训练
+
+下载PP-OCRv3的蒸馏预训练模型并进行训练的方式如下
+
+```bash
+# 下载预训练模型到/home/PaddleOCR/pre_train文件夹下
+$ mkdir /home/PaddleOCR/pre_train
+
+$ wget -P /home/PaddleOCR/pre_train https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
+
+$ tar xf /home/PaddleOCR/pre_train/ch_PP-OCRv3_det_distill_train.tar -C /home/PaddleOCR/pre_train/
+```
+
+启动训练，训练模型默认保存在`output`目录下，加载PP-OCRv3检测预训练模型。
+
+```bash
+# 这里以 GPU 训练为例，使用 CPU 进行训练的话，需要指定参数 Global.use_gpu=false
+python3 tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Global.save_model_dir=./output/ Global.pretrained_model=./pre_train/ch_PP-OCRv3_det_distill_train/best_accuracy
+```
+
+如果要使用多GPU分布式训练，请使用如下命令：
+
+```bash
+# 启动训练，训练模型默认保存在output目录下，--gpus '0,1,2,3'表示使用0，1，2，3号GPU训练
+python3 -m paddle.distributed.launch --log_dir=./debug/ --gpus '0,1,2,3' tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Global.save_model_dir=./output/ Global.pretrained_model=./pre_train/ch_PP-OCRv3_det_distill_train/best_accuracy
+```
+
+### 1.6 模型评估
+
+训练过程中保存的模型在output目录下，包含以下文件：
+
+```
+best_accuracy.states
+best_accuracy.pdparams  # 默认保存最优精度的模型参数
+best_accuracy.pdopt     # 默认保存最优精度的优化器相关参数
+latest.states
+latest.pdparams  # 默认保存的最新模型参数
+latest.pdopt     # 默认保存的最新模型的优化器相关参数
+```
+
+其中，best_accuracy是保存的最优模型，可以直接使用该模型评估
+
+```bash
+# 进行模型评估
+cd /home/PaddleOCR/
+
+python3 tools/eval.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Global.checkpoints=./output/best_accuracy
+```
+
+## 2. PP-OCRv3云端部署
+
+PaddleCloud基于Kubernetes的Operator机制为您提供了多个功能强大的云原生组件，如样本数据缓存组件、分布式训练组件、 以及模型推理服务组件，
+使用这些组件您可以快速地在云上进行分布式训练和模型服务化部署。更多关于PaddleCloud云原生组件的内容，请参考文档 [PaddleCloud架构概览](https://github.com/PaddlePaddle/PaddleCloud/blob/main/docs/zh_CN/paddlecloud-overview.md) 。
+
+> **适用场景**：基于Kubernetes的多机部署环境。
+
+### 2.1 安装云上飞桨组件
+
+**环境要求**
+
+- [Kubernetes v1.16+](https://kubernetes.io/zh/)
+- [kubectl](https://kubernetes.io/docs/tasks/tools/)
+- [Helm](https://helm.sh/zh/docs/intro/install/)
+
+如果您没有Kubernetes环境，可以使用MicroK8S在本地搭建环境，更多详情请参考 [MicroK8S官方文档](https://microk8s.io/docs/getting-started)。
+
+使用Helm一键安装所有组件和所有依赖
+
+```bash
+# 添加PaddleCloud Chart仓库
+$ helm repo add paddlecloud https://paddleflow-public.hkg.bcebos.com/charts
+$ helm repo update
+
+# 安装云上飞桨组件
+$ helm install pdc paddlecloud/paddlecloud --set tags.all-dep=true --namespace paddlecloud --create-namespace
+
+# 检查所有云上飞桨组件是否成功启动，命名空间下的所有Pod都为Runing状态则安装成功。
+$ kubectl get pods -n paddlecloud
+NAME                                                 READY   STATUS    RESTARTS   AGE
+pdc-hostpath-5b6bd6787d-bxvxg                        1/1     Running   0          10h
+juicefs-csi-node-pkldt                               3/3     Running   0          10h
+juicefs-csi-controller-0                             3/3     Running   0          10h
+pdc-paddlecloud-sampleset-767bdf6947-pb6zm           1/1     Running   0          10h
+pdc-paddlecloud-paddlejob-7cc8b7bfc6-7gqnh           1/1     Running   0          10h
+pdc-minio-7cc967669d-824q5                           1/1     Running   0          10h
+pdc-redis-master-0                                   1/1     Running   0          10h
+```
+
+更多安装参数请参考[PaddleCloud安装指南](https://github.com/PaddlePaddle/PaddleCloud/blob/main/docs/zh_CN/installation.md)
+
+### 2.2 云原生组件介绍
+
+<div align="center">
+  <img src="./images/architecture.jpeg" title="architecture" width="60%" height="60%" alt="">
+</div>
+
+- **数据缓存组件。** 数据缓存组件使用JuiceFS作为缓存引擎，能够将远程样本数据缓存到训练集群本地，大幅加速云上飞桨分布式训练作业。
+- **分布式训练组件。** 分布式训练组件支持参数服务器（PS）与集合通信（Collective）两种架构模式，方便用户在云上快速运行飞桨分布式训练作业。
+
+以下内容我们将使用这两个云原生组件来在Kubernetes集群中部署PP-OCRv3识别模型的训练作业。
+
+### 2.3 准备hiertext数据集
+
+使用数据缓存组件来准备数据集，编写SampleSet Yaml文件如下：
+
+```yaml
+# hiertext.yaml
+apiVersion: batch.paddlepaddle.org/v1alpha1
+kind: SampleSet
+metadata:
+  name: hiertext
+  namespace: paddlecloud
+spec:
+  partitions: 1
+  source:
+    uri: bos://paddleflow-public.hkg.bcebos.com/ppocr/hiertext
+    secretRef:
+      name: none
+  secretRef:
+    name: data-center
+```
+
+然后在命令行中，使用kubectl执行如下命令。
+
+```bash
+# 创建hiertext数据集
+$ kubectl apply -f hiertext.yaml
+sampleset.batch.paddlepaddle.org/hiertext created
+
+# 查看数据集的状态
+$ kubectl get sampleset hiertext -n paddlecloud
+NAME       TOTAL SIZE   CACHED SIZE   AVAIL SPACE   RUNTIME   PHASE   AGE
+hiertext   3.3 GiB       3.2 GiB      12 GiB        1/1       Ready   11m
+```
+
+### 2.4 训练PP-OCRv3模型
+
+使用训练组件在Kubernetes集群上训练PP-OCRv3模型，编写PaddleJob Yaml文件如下：
+
+```yaml
+# ppocrv3.yaml
+apiVersion: batch.paddlepaddle.org/v1
+kind: PaddleJob
+metadata:
+  name: ppocrv3
+  namespace: paddlecloud
+spec:
+  cleanPodPolicy: OnCompletion
+  sampleSetRef:
+    name: hiertext
+    namespace: paddlecloud
+    mountPath: /mnt/hiertext
+  worker:
+    replicas: 1
+    template:
+      spec:
+        containers:
+          - name: ppocrv3
+            image: paddlecloud/paddleocr:2.5-gpu-cuda10.2-cudnn7-efbb0a
+            command:
+              - /bin/bash
+            args:
+              - "-c"
+              - >
+                mkdir /home/PaddleOCR/pre_train &&
+                wget -P ./pre_train https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar &&
+                tar xf ./pre_train/ch_PP-OCRv3_det_distill_train.tar -C ./pre_train/ &&
+                python tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o
+                Train.dataset.data_dir=/mnt/
+                Train.dataset.label_file_list=[\"/mnt/hiertext/label_hiertext_train.txt\"]
+                Eval.dataset.data_dir=/mnt/
+                Eval.dataset.label_file_list=[\"/mnt/hiertext/label_hiertext_val.txt\"]
+                Global.save_model_dir=./output/
+                Global.pretrained_model=./pre_train/ch_PP-OCRv3_det_distill_train/best_accuracy
+            resources:
+              limits:
+                nvidia.com/gpu: 1
+            volumeMounts:  # 添加 shared memory 挂载以防止缓存出错
+              - mountPath: /dev/shm
+                name: dshm
+        volumes:
+          - name: dshm
+            emptyDir:
+              medium: Memory
+```
+
+本案例采用GPU进行训练，如果您只有CPU机器，则可以将镜像替换成CPU版本 `paddlecloud/paddleocr:2.5-cpu-efbb0a`，并在args中加上参数`Global.use_gpu=false`。
+
+```bash
+# 创建PaddleJob训练模型
+$ kubectl apply -f ppocrv3.yaml
+paddlejob.batch.paddlepaddle.org/ppocrv3 created
+
+# 查看PaddleJob状态
+$ kubectl get pods -n paddlecloud -l paddle-res-name=ppocrv3-worker-0
+NAME               READY   STATUS    RESTARTS   AGE
+ppocrv3-worker-0   1/1     Running   0          4s
+
+# 查看训练日志
+$ kubectl logs -f ppocrv3-worker-0 -n paddlecloud
+```
+
+## 更多资源
+
+欢迎关注[云上飞桨项目PaddleCloud](https://github.com/PaddlePaddle/PaddleCloud)，我们为您提供了飞桨模型套件标准镜像以及全栈的云原生模型套件部署组件，如您有任何关于飞桨模型套件的部署问题，请联系我们。
+如果你发现任何PaddleCloud存在的问题或者是建议, 欢迎通过[GitHub Issues](https://github.com/PaddlePaddle/PaddleCloud/issues)给我们提issues。
diff --git a/deploy/paddlecloud/images/architecture.jpeg b/deploy/paddlecloud/images/architecture.jpeg
new file mode 100644
index 0000000..7d9250e
Binary files /dev/null and b/deploy/paddlecloud/images/architecture.jpeg differ
diff --git a/deploy/paddlecloud/images/paddlecloud.png b/deploy/paddlecloud/images/paddlecloud.png
new file mode 100644
index 0000000..bbd02a7
Binary files /dev/null and b/deploy/paddlecloud/images/paddlecloud.png differ
diff --git a/deploy/paddleocr_vl_docker/accelerators/amd-gpu/.env b/deploy/paddleocr_vl_docker/accelerators/amd-gpu/.env
new file mode 100644
index 0000000..61a0793
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/amd-gpu/.env
@@ -0,0 +1,3 @@
+API_IMAGE_TAG_SUFFIX=latest-amd-gpu-offline
+VLM_BACKEND=vllm
+VLM_IMAGE_TAG_SUFFIX=latest-amd-gpu-offline
diff --git a/deploy/paddleocr_vl_docker/accelerators/amd-gpu/compose.yaml b/deploy/paddleocr_vl_docker/accelerators/amd-gpu/compose.yaml
new file mode 100644
index 0000000..2b06bfe
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/amd-gpu/compose.yaml
@@ -0,0 +1,47 @@
+services:
+  paddleocr-vl-api:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:${API_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vl-api
+    ports:
+      - 8080:8080
+    depends_on:
+      paddleocr-vlm-server:
+        condition: service_healthy
+    devices:
+      - /dev/kfd
+      - /dev/dri
+    group_add:
+      - video
+    cap_add:
+      - SYS_PTRACE
+    security_opt:
+      - seccomp=unconfined
+    shm_size: 64g
+    # TODO: Allow using a regular user
+    user: root
+    restart: unless-stopped
+    environment:
+      - VLM_BACKEND=${VLM_BACKEND:-vllm}
+    command: /bin/bash -c "paddlex --serve --pipeline /home/paddleocr/pipeline_config_${VLM_BACKEND}.yaml"
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+
+  paddleocr-vlm-server:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-${VLM_BACKEND}-server:${VLM_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vlm-server
+    devices:
+      - /dev/kfd
+      - /dev/dri
+    group_add:
+      - video
+    cap_add:
+      - SYS_PTRACE
+    security_opt:
+      - seccomp=unconfined
+    shm_size: 64g
+    # TODO: Allow using a regular user
+    user: root
+    restart: unless-stopped
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+      start_period: 300s
diff --git a/deploy/paddleocr_vl_docker/accelerators/amd-gpu/pipeline.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/amd-gpu/pipeline.Dockerfile
new file mode 100644
index 0000000..e40b6dc
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/amd-gpu/pipeline.Dockerfile
@@ -0,0 +1,61 @@
+FROM vivienfanghua/vllm_paddle:base
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+ENV PYTHONUNBUFFERED=1
+ENV PYTHONDONTWRITEBYTECODE=1
+
+ENV PIP_BREAK_SYSTEM_PACKAGES=1
+RUN ln -s /usr/bin/python3 /usr/bin/python
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends miopen-hip-dev rocrand-dev rocblas-dev \
+    && apt-get install -y --no-install-recommends \
+        fontconfig \
+        fonts-dejavu-core \
+        fonts-liberation \
+        fonts-noto-cjk \
+        fonts-wqy-microhei \
+        fonts-freefont-ttf \
+    && fc-cache -fv \
+    && rm -rf /var/lib/apt/lists/*
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install "paddleocr[doc-parser]${PADDLEOCR_VERSION}" "paddlex[serving]${PADDLEX_VERSION}"
+
+RUN groupadd -g 1001 paddleocr \
+    && useradd -m -s /bin/bash -u 1001 -g 1001 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayoutV3_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf UVDoc_infer.tar \
+        && mv UVDoc_infer UVDoc \
+        && tar -xf PP-LCNet_x1_0_doc_ori_infer.tar \
+        && mv PP-LCNet_x1_0_doc_ori_infer PP-LCNet_x1_0_doc_ori \
+        && tar -xf PP-DocLayoutV3_infer.tar \
+        && mv PP-DocLayoutV3_infer PP-DocLayoutV3 \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f UVDoc_infer.tar PP-LCNet_x1_0_doc_ori_infer.tar PP-DocLayoutV3_infer.tar PaddleOCR-VL-1.6_infer.tar \
+        && mkdir -p "${HOME}/.paddlex/fonts" \
+        && wget -P "${HOME}/.paddlex/fonts" https://paddle-model-ecology.bj.bcebos.com/paddlex/PaddleX3.0/fonts/PingFang-SC-Regular.ttf; \
+    fi
+
+COPY --chown=paddleocr:paddleocr pipeline_config_vllm.yaml /home/paddleocr
+COPY --chown=paddleocr:paddleocr pipeline_config_fastdeploy.yaml /home/paddleocr
+
+EXPOSE 8080
+
+CMD ["paddlex", "--serve", "--pipeline", "/home/paddleocr/pipeline_config_vllm.yaml"]
diff --git a/deploy/paddleocr_vl_docker/accelerators/amd-gpu/vlm.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/amd-gpu/vlm.Dockerfile
new file mode 100644
index 0000000..43368fa
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/amd-gpu/vlm.Dockerfile
@@ -0,0 +1,33 @@
+ARG BACKEND=vllm
+
+
+FROM rocm/vllm-dev:nightly_main_20260125 AS base-vllm
+
+
+FROM base-${BACKEND}
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install "paddleocr${PADDLEOCR_VERSION}" "paddlex${PADDLEX_VERSION}"
+
+RUN groupadd -g 1001 paddleocr \
+    && useradd -m -s /bin/bash -u 1001 -g 1001 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f PaddleOCR-VL-1.6_infer.tar; \
+    fi
+
+ARG BACKEND
+ENV BACKEND=${BACKEND}
+CMD ["/bin/bash", "-c", "paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8080 --backend ${BACKEND}"]
diff --git a/deploy/paddleocr_vl_docker/accelerators/huawei-npu/.env b/deploy/paddleocr_vl_docker/accelerators/huawei-npu/.env
new file mode 100644
index 0000000..0f20c77
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/huawei-npu/.env
@@ -0,0 +1,3 @@
+API_IMAGE_TAG_SUFFIX=latest-huawei-npu-offline
+VLM_BACKEND=vllm
+VLM_IMAGE_TAG_SUFFIX=latest-huawei-npu-offline
diff --git a/deploy/paddleocr_vl_docker/accelerators/huawei-npu/compose.yaml b/deploy/paddleocr_vl_docker/accelerators/huawei-npu/compose.yaml
new file mode 100644
index 0000000..aca178f
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/huawei-npu/compose.yaml
@@ -0,0 +1,37 @@
+services:
+  paddleocr-vl-api:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:${API_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vl-api
+    ports:
+      - 8080:8080
+    depends_on:
+      paddleocr-vlm-server:
+        condition: service_healthy
+    user: root
+    restart: unless-stopped
+    environment:
+      - VLM_BACKEND=${VLM_BACKEND:-vllm}
+    command: /bin/bash -c "paddlex --serve --pipeline /home/paddleocr/pipeline_config_${VLM_BACKEND}.yaml --device npu"
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+    volumes:
+      - /usr/local/Ascend/driver:/usr/local/Ascend/driver
+      - /usr/local/bin/npu-smi:/usr/local/bin/npu-smi
+      - /usr/local/dcmi:/usr/local/dcmi
+    privileged: true
+    shm_size: 64g
+
+  paddleocr-vlm-server:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-${VLM_BACKEND}-server:${VLM_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vlm-server
+    user: root
+    restart: unless-stopped
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+      start_period: 300s
+    volumes:
+      - /usr/local/Ascend/driver:/usr/local/Ascend/driver
+      - /usr/local/bin/npu-smi:/usr/local/bin/npu-smi
+      - /usr/local/dcmi:/usr/local/dcmi
+    privileged: true
+    shm_size: 64g
diff --git a/deploy/paddleocr_vl_docker/accelerators/huawei-npu/pipeline.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/huawei-npu/pipeline.Dockerfile
new file mode 100644
index 0000000..54835b3
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/huawei-npu/pipeline.Dockerfile
@@ -0,0 +1,124 @@
+# TODO: Allow regular users
+
+ARG TARGETARCH
+
+
+FROM ccr-2vdh3abv-pub.cnc.bj.baidubce.com/device/paddle-npu:cann800-ubuntu20-npu-910b-base-x86_64-gcc84 AS base-amd64
+
+ENV ASCEND_TOOLKIT_HOME=/usr/local/Ascend/ascend-toolkit/latest
+ENV LD_LIBRARY_PATH=${ASCEND_TOOLKIT_HOME}/lib64:${ASCEND_TOOLKIT_HOME}/lib64/plugin/opskernel:${ASCEND_TOOLKIT_HOME}/lib64/plugin/nnengine:${ASCEND_TOOLKIT_HOME}/opp/built-in/op_impl/ai_core/tbe/op_tiling/lib/linux/x86_64:$LD_LIBRARY_PATH
+
+
+FROM ccr-2vdh3abv-pub.cnc.bj.baidubce.com/device/paddle-npu:cann800-ubuntu20-npu-910b-base-aarch64-gcc84 AS base-arm64
+
+ENV ASCEND_TOOLKIT_HOME=/usr/local/Ascend/ascend-toolkit/latest
+ENV LD_LIBRARY_PATH=${ASCEND_TOOLKIT_HOME}/lib64:${ASCEND_TOOLKIT_HOME}/lib64/plugin/opskernel:${ASCEND_TOOLKIT_HOME}/lib64/plugin/nnengine:${ASCEND_TOOLKIT_HOME}/opp/built-in/op_impl/ai_core/tbe/op_tiling/lib/linux/aarch_64:$LD_LIBRARY_PATH
+
+
+FROM base-${TARGETARCH} AS base
+
+ARG TARGETARCH
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+ENV PYTHONUNBUFFERED=1
+ENV PYTHONDONTWRITEBYTECODE=1
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends \
+        fontconfig \
+        fonts-dejavu-core \
+        fonts-liberation \
+        fonts-noto-cjk \
+        fonts-wqy-microhei \
+        fonts-freefont-ttf \
+    && fc-cache -fv \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN --mount=type=cache,id=pip-${TARGETARCH},target=/root/.cache/pip \
+    python -m pip install paddlepaddle==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/ \
+    && python -m pip install paddle-custom-npu==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/npu/
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,id=pip-${TARGETARCH},target=/root/.cache/pip \
+    python -m pip install "paddleocr[doc-parser]${PADDLEOCR_VERSION}" "paddlex[serving]${PADDLEX_VERSION}"
+
+RUN --mount=type=cache,id=pip-${TARGETARCH},target=/root/.cache/pip \
+    python -m pip install numpy==1.26.4 opencv-contrib-python==4.11.0.86
+
+RUN groupadd -g 1001 paddleocr \
+    && useradd -m -s /bin/bash -u 1001 -g 1001 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+ENV LD_LIBRARY_PATH=${ASCEND_TOOLKIT_HOME}/tools/aml/lib64:${ASCEND_TOOLKIT_HOME}/tools/aml/lib64/plugin:$LD_LIBRARY_PATH
+ENV PYTHONPATH=${ASCEND_TOOLKIT_HOME}/python/site-packages:${ASCEND_TOOLKIT_HOME}/opp/built-in/op_impl/ai_core/tbe
+ENV PATH=${ASCEND_TOOLKIT_HOME}/bin:${ASCEND_TOOLKIT_HOME}/compiler/ccec_compiler/bin:${ASCEND_TOOLKIT_HOME}/tools/ccec_compiler/bin:$PATH
+ENV ASCEND_AICPU_PATH=${ASCEND_TOOLKIT_HOME}
+ENV ASCEND_OPP_PATH=${ASCEND_TOOLKIT_HOME}/opp
+ENV TOOLCHAIN_HOME=${ASCEND_TOOLKIT_HOME}/toolkit
+ENV ASCEND_HOME_PATH=${ASCEND_TOOLKIT_HOME}
+
+ENV ATB_HOME_PATH=/usr/local/Ascend/nnal/atb/latest/atb/cxx_abi_1
+ENV LD_LIBRARY_PATH=${ATB_HOME_PATH}/lib:${ATB_HOME_PATH}/examples:${ATB_HOME_PATH}/tests/atbopstest:${LD_LIBRARY_PATH}
+ENV PATH=${ATB_HOME_PATH}/bin:${PATH}
+
+ENV ATB_STREAM_SYNC_EVERY_KERNEL_ENABLE=0
+ENV ATB_STREAM_SYNC_EVERY_RUNNER_ENABLE=0
+ENV ATB_STREAM_SYNC_EVERY_OPERATION_ENABLE=0
+ENV ATB_OPSRUNNER_SETUP_CACHE_ENABLE=1
+ENV ATB_OPSRUNNER_KERNEL_CACHE_TYPE=3
+ENV ATB_OPSRUNNER_KERNEL_CACHE_LOCAL_COUNT=1
+ENV ATB_OPSRUNNER_KERNEL_CACHE_GLOABL_COUNT=5
+ENV ATB_OPSRUNNER_KERNEL_CACHE_TILING_SIZE=10240
+ENV ATB_WORKSPACE_MEM_ALLOC_ALG_TYPE=1
+ENV ATB_WORKSPACE_MEM_ALLOC_GLOBAL=0
+ENV ATB_COMPARE_TILING_EVERY_KERNEL=0
+ENV ATB_HOST_TILING_BUFFER_BLOCK_NUM=128
+ENV ATB_DEVICE_TILING_BUFFER_BLOCK_NUM=32
+ENV ATB_SHARE_MEMORY_NAME_SUFFIX=""
+ENV ATB_LAUNCH_KERNEL_WITH_TILING=1
+ENV ATB_MATMUL_SHUFFLE_K_ENABLE=1
+ENV ATB_RUNNER_POOL_SIZE=64
+
+ENV ASDOPS_HOME_PATH=${ATB_HOME_PATH}
+ENV ASDOPS_MATMUL_PP_FLAG=1
+ENV ASDOPS_LOG_LEVEL=ERROR
+ENV ASDOPS_LOG_TO_STDOUT=0
+ENV ASDOPS_LOG_TO_FILE=1
+ENV ASDOPS_LOG_TO_FILE_FLUSH=0
+ENV ASDOPS_LOG_TO_BOOST_TYPE=atb
+ENV ASDOPS_LOG_PATH=${HOME}
+ENV ASDOPS_TILING_PARSE_CACHE_DISABLE=0
+ENV LCCL_DETERMINISTIC=0
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayoutV3_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf UVDoc_infer.tar \
+        && mv UVDoc_infer UVDoc \
+        && tar -xf PP-LCNet_x1_0_doc_ori_infer.tar \
+        && mv PP-LCNet_x1_0_doc_ori_infer PP-LCNet_x1_0_doc_ori \
+        && tar -xf PP-DocLayoutV3_infer.tar \
+        && mv PP-DocLayoutV3_infer PP-DocLayoutV3 \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f UVDoc_infer.tar PP-LCNet_x1_0_doc_ori_infer.tar PP-DocLayoutV3_infer.tar PaddleOCR-VL-1.6_infer.tar \
+        && mkdir -p "${HOME}/.paddlex/fonts" \
+        && wget -P "${HOME}/.paddlex/fonts" https://paddle-model-ecology.bj.bcebos.com/paddlex/PaddleX3.0/fonts/PingFang-SC-Regular.ttf; \
+    fi
+
+COPY --chown=paddleocr:paddleocr pipeline_config_vllm.yaml /home/paddleocr
+COPY --chown=paddleocr:paddleocr pipeline_config_fastdeploy.yaml /home/paddleocr
+
+EXPOSE 8080
+
+CMD ["paddlex", "--serve", "--pipeline", "/home/paddleocr/pipeline_config_vllm.yaml", "--device", "npu"]
diff --git a/deploy/paddleocr_vl_docker/accelerators/huawei-npu/vlm.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/huawei-npu/vlm.Dockerfile
new file mode 100644
index 0000000..8c8a510
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/huawei-npu/vlm.Dockerfile
@@ -0,0 +1,39 @@
+# TODO: Allow regular users
+
+ARG BACKEND=vllm
+
+
+FROM ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddlex-vllm-huawei-npu:0.12.0rc1 AS base-vllm
+
+
+FROM base-${BACKEND}
+
+RUN apt-get update \
+    && apt-get install -y libgl1 \
+    && rm -rf /var/lib/apt/lists/*
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install "paddleocr${PADDLEOCR_VERSION}" "paddlex${PADDLEX_VERSION}"
+
+RUN groupadd -g 1000 paddleocr \
+    && useradd -m -s /bin/bash -u 1000 -g 1000 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f PaddleOCR-VL-1.6_infer.tar; \
+    fi
+
+ARG BACKEND
+ENV BACKEND=${BACKEND}
+CMD ["/bin/bash", "-c", "paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8080 --backend ${BACKEND}"]
diff --git a/deploy/paddleocr_vl_docker/accelerators/hygon-dcu/.env b/deploy/paddleocr_vl_docker/accelerators/hygon-dcu/.env
new file mode 100644
index 0000000..53cf0d6
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/hygon-dcu/.env
@@ -0,0 +1,3 @@
+API_IMAGE_TAG_SUFFIX=latest-hygon-dcu-offline
+VLM_BACKEND=vllm
+VLM_IMAGE_TAG_SUFFIX=latest-hygon-dcu-offline
diff --git a/deploy/paddleocr_vl_docker/accelerators/hygon-dcu/compose.yaml b/deploy/paddleocr_vl_docker/accelerators/hygon-dcu/compose.yaml
new file mode 100644
index 0000000..aab2510
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/hygon-dcu/compose.yaml
@@ -0,0 +1,53 @@
+services:
+  paddleocr-vl-api:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:${API_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vl-api
+    ports:
+      - 8080:8080
+    depends_on:
+      paddleocr-vlm-server:
+        condition: service_healthy
+    user: root
+    privileged: true
+    devices:
+      - /dev/kfd
+      - /dev/dri
+      - /dev/mkfd
+    group_add:
+      - video
+    cap_add:
+      - SYS_PTRACE
+    security_opt:
+      - seccomp=unconfined
+    restart: unless-stopped
+    environment:
+      - VLM_BACKEND=${VLM_BACKEND:-vllm}
+    command: paddlex --serve --pipeline /home/paddleocr/pipeline_config_${VLM_BACKEND}.yaml --device dcu
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+    volumes:
+      - /opt/hyhal/:/opt/hyhal/:ro
+    shm_size: 64g
+
+  paddleocr-vlm-server:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-${VLM_BACKEND}-server:${VLM_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vlm-server
+    user: root
+    privileged: true
+    devices:
+      - /dev/kfd
+      - /dev/dri
+      - /dev/mkfd
+    group_add:
+      - video
+    cap_add:
+      - SYS_PTRACE
+    security_opt:
+      - seccomp=unconfined
+    restart: unless-stopped
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+      start_period: 300s
+    volumes:
+      - /opt/hyhal/:/opt/hyhal/:ro
+    shm_size: 64g
diff --git a/deploy/paddleocr_vl_docker/accelerators/hygon-dcu/pipeline.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/hygon-dcu/pipeline.Dockerfile
new file mode 100644
index 0000000..c710a54
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/hygon-dcu/pipeline.Dockerfile
@@ -0,0 +1,82 @@
+# TODO: Allow regular users
+
+FROM ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle-dcu:dtk24.04.1-kylinv10-gcc82
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+ENV PYTHONUNBUFFERED=1
+ENV PYTHONDONTWRITEBYTECODE=1
+
+RUN yum install -y \
+        fontconfig \
+        dejavu-sans-fonts \
+        dejavu-serif-fonts \
+        liberation-fonts \
+        liberation-mono-fonts \
+        liberation-sans-fonts \
+        liberation-serif-fonts \
+        google-noto-cjk-fonts \
+        wqy-microhei-fonts \
+        gnu-free-fonts-common \
+        gnu-free-mono-fonts \
+        gnu-free-sans-fonts \
+        gnu-free-serif-fonts \
+    && fc-cache -fv \
+    && yum clean all
+
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install safetensors==0.7.0 \
+    && python -m pip install paddlepaddle-dcu==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/dcu/
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install "paddleocr[doc-parser]${PADDLEOCR_VERSION}" "paddlex[serving]${PADDLEX_VERSION}"
+
+RUN groupadd -g 1000 paddleocr \
+    && useradd -m -s /bin/bash -u 1000 -g 1000 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+ENV HYHAL_PATH=/opt/hyhal
+ENV DTKROOT=/opt/
+ENV AMDGPU_TARGETS="gfx906;gfx926;gfx928"
+ENV ROCM_PATH=/opt/dtk-24.04.1
+ENV HIP_PATH=/opt/dtk-24.04.1/hip
+ENV MIOPEN_FIND_MODE=3
+
+ENV PATH="${ROCM_PATH}/bin:${ROCM_PATH}/llvm/bin:${ROCM_PATH}/hip/bin:${ROCM_PATH}/hip/bin/hipify:${HYHAL_PATH}/bin:${PATH}"
+ENV LD_LIBRARY_PATH="${ROCM_PATH}/lib:${ROCM_PATH}/lib64:${HYHAL_PATH}/lib:${HYHAL_PATH}/lib64:${LD_LIBRARY_PATH}"
+ENV LD_LIBRARY_PATH="${ROCM_PATH}/hip/lib:${ROCM_PATH}/llvm/lib:${LD_LIBRARY_PATH}"
+ENV C_INCLUDE_PATH="${ROCM_PATH}/include:${HYHAL_PATH}/include:${ROCM_PATH}/llvm/include"
+ENV CPLUS_INCLUDE_PATH="${ROCM_PATH}/include:${HYHAL_PATH}/include:${ROCM_PATH}/llvm/include"
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayoutV3_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf UVDoc_infer.tar \
+        && mv UVDoc_infer UVDoc \
+        && tar -xf PP-LCNet_x1_0_doc_ori_infer.tar \
+        && mv PP-LCNet_x1_0_doc_ori_infer PP-LCNet_x1_0_doc_ori \
+        && tar -xf PP-DocLayoutV3_infer.tar \
+        && mv PP-DocLayoutV3_infer PP-DocLayoutV3 \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f UVDoc_infer.tar PP-LCNet_x1_0_doc_ori_infer.tar PP-DocLayoutV3_infer.tar PaddleOCR-VL-1.6_infer.tar \
+        && mkdir -p "${HOME}/.paddlex/fonts" \
+        && wget -P "${HOME}/.paddlex/fonts" https://paddle-model-ecology.bj.bcebos.com/paddlex/PaddleX3.0/fonts/PingFang-SC-Regular.ttf; \
+    fi
+
+COPY --chown=paddleocr:paddleocr pipeline_config_vllm.yaml /home/paddleocr
+COPY --chown=paddleocr:paddleocr pipeline_config_fastdeploy.yaml /home/paddleocr
+
+EXPOSE 8080
+
+CMD ["paddlex", "--serve", "--pipeline", "/home/paddleocr/pipeline_config_vllm.yaml", "--device", "dcu"]
diff --git a/deploy/paddleocr_vl_docker/accelerators/hygon-dcu/vlm.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/hygon-dcu/vlm.Dockerfile
new file mode 100644
index 0000000..0520779
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/hygon-dcu/vlm.Dockerfile
@@ -0,0 +1,35 @@
+# TODO: Allow regular users
+
+ARG BACKEND=vllm
+
+
+FROM image.sourcefind.cn:5000/dcu/admin/base/vllm:0.9.2-ubuntu22.04-dtk25.04.2-py3.10 AS base-vllm
+
+
+FROM base-${BACKEND}
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install "paddleocr${PADDLEOCR_VERSION}" "paddlex${PADDLEX_VERSION}"
+
+RUN groupadd -g 1000 paddleocr \
+    && useradd -m -s /bin/bash -u 1000 -g 1000 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f PaddleOCR-VL-1.6_infer.tar; \
+    fi
+
+ARG BACKEND
+ENV BACKEND=${BACKEND}
+CMD ["/bin/bash", "-c", "paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8080 --backend ${BACKEND}"]
diff --git a/deploy/paddleocr_vl_docker/accelerators/iluvatar-gpu/.env b/deploy/paddleocr_vl_docker/accelerators/iluvatar-gpu/.env
new file mode 100644
index 0000000..8f19a59
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/iluvatar-gpu/.env
@@ -0,0 +1,3 @@
+API_IMAGE_TAG_SUFFIX=latest-iluvatar-gpu-offline
+VLM_BACKEND=fastdeploy
+VLM_IMAGE_TAG_SUFFIX=latest-iluvatar-gpu-offline
diff --git a/deploy/paddleocr_vl_docker/accelerators/iluvatar-gpu/compose.yaml b/deploy/paddleocr_vl_docker/accelerators/iluvatar-gpu/compose.yaml
new file mode 100644
index 0000000..bd17ba1
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/iluvatar-gpu/compose.yaml
@@ -0,0 +1,43 @@
+services:
+  paddleocr-vl-api:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:${API_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vl-api
+    ports:
+      - 8080:8080
+    privileged: true
+    cap_add:
+      - SYS_PTRACE
+    pid: host
+    shm_size: 64g
+    depends_on:
+      paddleocr-vlm-server:
+        condition: service_healthy
+    user: root
+    restart: unless-stopped
+    environment:
+      - VLM_BACKEND=${VLM_BACKEND:-vllm}
+    command: /bin/bash -c "paddlex --serve --pipeline /home/paddleocr/pipeline_config_${VLM_BACKEND}.yaml --device iluvatar_gpu"
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+    volumes:
+      - /usr/src:/usr/src
+      - /lib/modules:/lib/modules
+      - /dev:/dev
+
+  paddleocr-vlm-server:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-${VLM_BACKEND}-server:${VLM_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vlm-server
+    privileged: true
+    cap_add:
+      - SYS_PTRACE
+    pid: host
+    shm_size: 64g
+    user: root
+    restart: unless-stopped
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+      start_period: 300s
+    volumes:
+      - /usr/src:/usr/src
+      - /lib/modules:/lib/modules
+      - /dev:/dev
diff --git a/deploy/paddleocr_vl_docker/accelerators/iluvatar-gpu/pipeline.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/iluvatar-gpu/pipeline.Dockerfile
new file mode 100644
index 0000000..649da97
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/iluvatar-gpu/pipeline.Dockerfile
@@ -0,0 +1,67 @@
+# TODO: Allow regular users
+
+FROM ccr-2vdh3abv-pub.cnc.bj.baidubce.com/device/paddle-ixuca:3.3.0
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+ENV PYTHONUNBUFFERED=1
+ENV PYTHONDONTWRITEBYTECODE=1
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends \
+        fontconfig \
+        fonts-dejavu-core \
+        fonts-liberation \
+        fonts-noto-cjk \
+        fonts-wqy-microhei \
+        fonts-freefont-ttf \
+    && fc-cache -fv \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install paddlepaddle==3.3.0.dev20251219 -i https://www.paddlepaddle.org.cn/packages/nightly/cpu/ \
+    && python -m pip install paddle-iluvatar-gpu==3.0.0.dev20251223 -i https://www.paddlepaddle.org.cn/packages/nightly/ixuca/
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install "paddleocr[doc-parser]${PADDLEOCR_VERSION}" "paddlex[serving]${PADDLEX_VERSION}"
+
+RUN groupadd -g 1000 paddleocr \
+    && useradd -m -s /bin/bash -u 1000 -g 1000 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+ENV PADDLE_XCCL_BACKEND=iluvatar_gpu
+ENV FD_SAMPLING_CLASS=rejection
+ENV LD_PRELOAD=/usr/local/corex/lib64/libcuda.so.1
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayoutV3_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf UVDoc_infer.tar \
+        && mv UVDoc_infer UVDoc \
+        && tar -xf PP-LCNet_x1_0_doc_ori_infer.tar \
+        && mv PP-LCNet_x1_0_doc_ori_infer PP-LCNet_x1_0_doc_ori \
+        && tar -xf PP-DocLayoutV3_infer.tar \
+        && mv PP-DocLayoutV3_infer PP-DocLayoutV3 \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f UVDoc_infer.tar PP-LCNet_x1_0_doc_ori_infer.tar PP-DocLayoutV3_infer.tar PaddleOCR-VL-1.6_infer.tar \
+        && mkdir -p "${HOME}/.paddlex/fonts" \
+        && wget -P "${HOME}/.paddlex/fonts" https://paddle-model-ecology.bj.bcebos.com/paddlex/PaddleX3.0/fonts/PingFang-SC-Regular.ttf; \
+    fi
+
+COPY --chown=paddleocr:paddleocr pipeline_config_vllm.yaml /home/paddleocr
+COPY --chown=paddleocr:paddleocr pipeline_config_fastdeploy.yaml /home/paddleocr
+
+EXPOSE 8080
+
+CMD ["paddlex", "--serve", "--pipeline", "/home/paddleocr/pipeline_config_vllm.yaml", "--device", "iluvatar_gpu"]
diff --git a/deploy/paddleocr_vl_docker/accelerators/iluvatar-gpu/vlm.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/iluvatar-gpu/vlm.Dockerfile
new file mode 100644
index 0000000..645fdca
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/iluvatar-gpu/vlm.Dockerfile
@@ -0,0 +1,43 @@
+# TODO: Allow regular users
+
+ARG BACKEND=fastdeploy
+
+
+FROM ccr-2vdh3abv-pub.cnc.bj.baidubce.com/device/paddle-ixuca:paddle-ocr-vl-1107 AS base-fastdeploy
+
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install fastdeploy_iluvatar_gpu==2.4.0.dev0 -i https://www.paddlepaddle.org.cn/packages/stable/ixuca/
+
+
+FROM base-${BACKEND}
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install "paddleocr${PADDLEOCR_VERSION}" "paddlex${PADDLEX_VERSION}"
+
+RUN groupadd -g 1000 paddleocr \
+    && useradd -m -s /bin/bash -u 1000 -g 1000 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+# TODO: Set these env vars only in FastDeploy image
+ENV PADDLE_XCCL_BACKEND=iluvatar_gpu
+ENV FD_SAMPLING_CLASS=rejection
+ENV LD_PRELOAD=/usr/local/corex/lib64/libcuda.so.1
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f PaddleOCR-VL-1.6_infer.tar; \
+    fi
+
+ARG BACKEND
+ENV BACKEND=${BACKEND}
+CMD ["/bin/bash", "-c", "paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8080 --backend ${BACKEND}"]
diff --git a/deploy/paddleocr_vl_docker/accelerators/intel-gpu/.env b/deploy/paddleocr_vl_docker/accelerators/intel-gpu/.env
new file mode 100644
index 0000000..d677792
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/intel-gpu/.env
@@ -0,0 +1,3 @@
+API_IMAGE_TAG_SUFFIX=latest-intel-gpu-offline
+VLM_BACKEND=vllm
+VLM_IMAGE_TAG_SUFFIX=latest-intel-gpu-offline
diff --git a/deploy/paddleocr_vl_docker/accelerators/intel-gpu/compose.yaml b/deploy/paddleocr_vl_docker/accelerators/intel-gpu/compose.yaml
new file mode 100644
index 0000000..fca0e61
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/intel-gpu/compose.yaml
@@ -0,0 +1,33 @@
+services:
+  paddleocr-vl-api:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:${API_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vl-api
+    ports:
+      - 8080:8080
+    depends_on:
+      paddleocr-vlm-server:
+        condition: service_healthy
+    devices:
+      - /dev:/dev
+    shm_size: 64g
+    # TODO: Allow using a regular user
+    user: root
+    restart: unless-stopped
+    environment:
+      - VLM_BACKEND=${VLM_BACKEND:-vllm}
+    command: /bin/bash -c "paddlex --serve --pipeline /home/paddleocr/pipeline_config_${VLM_BACKEND}.yaml"
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+
+  paddleocr-vlm-server:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-${VLM_BACKEND}-server:${VLM_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vlm-server
+    devices:
+      - /dev:/dev
+    shm_size: 64g
+    # TODO: Allow using a regular user
+    user: root
+    restart: unless-stopped
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+      start_period: 300s
diff --git a/deploy/paddleocr_vl_docker/accelerators/intel-gpu/pipeline.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/intel-gpu/pipeline.Dockerfile
new file mode 100644
index 0000000..841eb96
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/intel-gpu/pipeline.Dockerfile
@@ -0,0 +1,60 @@
+FROM intel/llm-scaler-vllm:0.11.1-b7
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+ENV PYTHONUNBUFFERED=1
+ENV PYTHONDONTWRITEBYTECODE=1
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends \
+        fontconfig \
+        fonts-dejavu-core \
+        fonts-liberation \
+        fonts-noto-cjk \
+        fonts-wqy-microhei \
+        fonts-freefont-ttf \
+    && fc-cache -fv \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install paddlepaddle==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install "paddleocr[doc-parser]${PADDLEOCR_VERSION}" "paddlex[serving]${PADDLEX_VERSION}"
+
+RUN groupadd -g 1001 paddleocr \
+    && useradd -m -s /bin/bash -u 1001 -g 1001 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayoutV3_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf UVDoc_infer.tar \
+        && mv UVDoc_infer UVDoc \
+        && tar -xf PP-LCNet_x1_0_doc_ori_infer.tar \
+        && mv PP-LCNet_x1_0_doc_ori_infer PP-LCNet_x1_0_doc_ori \
+        && tar -xf PP-DocLayoutV3_infer.tar \
+        && mv PP-DocLayoutV3_infer PP-DocLayoutV3 \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f UVDoc_infer.tar PP-LCNet_x1_0_doc_ori_infer.tar PP-DocLayoutV3_infer.tar PaddleOCR-VL-1.6_infer.tar \
+        && mkdir -p "${HOME}/.paddlex/fonts" \
+        && wget -P "${HOME}/.paddlex/fonts" https://paddle-model-ecology.bj.bcebos.com/paddlex/PaddleX3.0/fonts/PingFang-SC-Regular.ttf; \
+    fi
+
+COPY --chown=paddleocr:paddleocr pipeline_config_vllm.yaml /home/paddleocr
+COPY --chown=paddleocr:paddleocr pipeline_config_fastdeploy.yaml /home/paddleocr
+
+EXPOSE 8080
+
+CMD ["paddlex", "--serve", "--pipeline", "/home/paddleocr/pipeline_config_vllm.yaml"]
diff --git a/deploy/paddleocr_vl_docker/accelerators/intel-gpu/vlm.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/intel-gpu/vlm.Dockerfile
new file mode 100644
index 0000000..e175100
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/intel-gpu/vlm.Dockerfile
@@ -0,0 +1,33 @@
+ARG BACKEND=vllm
+
+
+FROM intel/llm-scaler-vllm:0.11.1-b7 AS base-vllm
+
+
+FROM base-${BACKEND}
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install "paddleocr${PADDLEOCR_VERSION}" "paddlex${PADDLEX_VERSION}"
+
+RUN groupadd -g 1001 paddleocr \
+    && useradd -m -s /bin/bash -u 1001 -g 1001 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f PaddleOCR-VL-1.6_infer.tar; \
+    fi
+
+ARG BACKEND
+ENV BACKEND=${BACKEND}
+CMD ["/bin/bash", "-c", "paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8080 --backend ${BACKEND}"]
diff --git a/deploy/paddleocr_vl_docker/accelerators/kunlunxin-xpu/.env b/deploy/paddleocr_vl_docker/accelerators/kunlunxin-xpu/.env
new file mode 100644
index 0000000..ebb2edd
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/kunlunxin-xpu/.env
@@ -0,0 +1,3 @@
+API_IMAGE_TAG_SUFFIX=latest-kunlunxin-xpu-offline
+VLM_BACKEND=fastdeploy
+VLM_IMAGE_TAG_SUFFIX=latest-kunlunxin-xpu-offline
diff --git a/deploy/paddleocr_vl_docker/accelerators/kunlunxin-xpu/compose.yaml b/deploy/paddleocr_vl_docker/accelerators/kunlunxin-xpu/compose.yaml
new file mode 100644
index 0000000..a591ead
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/kunlunxin-xpu/compose.yaml
@@ -0,0 +1,29 @@
+services:
+  paddleocr-vl-api:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:${API_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vl-api
+    ports:
+      - 8080:8080
+    depends_on:
+      paddleocr-vlm-server:
+        condition: service_healthy
+    user: root
+    restart: unless-stopped
+    environment:
+      - VLM_BACKEND=${VLM_BACKEND:-vllm}
+    command: /bin/bash -c "paddlex --serve --pipeline /home/paddleocr/pipeline_config_${VLM_BACKEND}.yaml --device xpu"
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+    privileged: true
+    shm_size: 64g
+
+  paddleocr-vlm-server:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-${VLM_BACKEND}-server:${VLM_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vlm-server
+    user: root
+    restart: unless-stopped
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+      start_period: 300s
+    privileged: true
+    shm_size: 64g
diff --git a/deploy/paddleocr_vl_docker/accelerators/kunlunxin-xpu/pipeline.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/kunlunxin-xpu/pipeline.Dockerfile
new file mode 100644
index 0000000..cedfc14
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/kunlunxin-xpu/pipeline.Dockerfile
@@ -0,0 +1,62 @@
+# TODO: Allow regular users
+
+FROM ccr-2vdh3abv-pub.cnc.bj.baidubce.com/device/paddle-xpu:ubuntu20-x86_64-gcc84-py310
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+ENV PYTHONUNBUFFERED=1
+ENV PYTHONDONTWRITEBYTECODE=1
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends \
+        fontconfig \
+        fonts-dejavu-core \
+        fonts-liberation \
+        fonts-noto-cjk \
+        fonts-wqy-microhei \
+        fonts-freefont-ttf \
+    && fc-cache -fv \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install paddlepaddle-xpu==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/xpu-p800/
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install "paddleocr[doc-parser]${PADDLEOCR_VERSION}" "paddlex[serving]${PADDLEX_VERSION}"
+
+RUN groupadd -g 1000 paddleocr \
+    && useradd -m -s /bin/bash -u 1000 -g 1000 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayoutV3_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf UVDoc_infer.tar \
+        && mv UVDoc_infer UVDoc \
+        && tar -xf PP-LCNet_x1_0_doc_ori_infer.tar \
+        && mv PP-LCNet_x1_0_doc_ori_infer PP-LCNet_x1_0_doc_ori \
+        && tar -xf PP-DocLayoutV3_infer.tar \
+        && mv PP-DocLayoutV3_infer PP-DocLayoutV3 \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f UVDoc_infer.tar PP-LCNet_x1_0_doc_ori_infer.tar PP-DocLayoutV3_infer.tar PaddleOCR-VL-1.6_infer.tar \
+        && mkdir -p "${HOME}/.paddlex/fonts" \
+        && wget -P "${HOME}/.paddlex/fonts" https://paddle-model-ecology.bj.bcebos.com/paddlex/PaddleX3.0/fonts/PingFang-SC-Regular.ttf; \
+    fi
+
+COPY --chown=paddleocr:paddleocr pipeline_config_vllm.yaml /home/paddleocr
+COPY --chown=paddleocr:paddleocr pipeline_config_fastdeploy.yaml /home/paddleocr
+
+EXPOSE 8080
+
+CMD ["paddlex", "--serve", "--pipeline", "/home/paddleocr/pipeline_config_vllm.yaml", "--device", "xpu"]
diff --git a/deploy/paddleocr_vl_docker/accelerators/kunlunxin-xpu/vlm.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/kunlunxin-xpu/vlm.Dockerfile
new file mode 100644
index 0000000..9a337b0
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/kunlunxin-xpu/vlm.Dockerfile
@@ -0,0 +1,39 @@
+# TODO: Allow regular users
+
+ARG BACKEND=fastdeploy
+
+
+FROM ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/fastdeploy-xpu:2.3.0 AS base-fastdeploy
+
+
+FROM base-${BACKEND}
+
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install https://paddle-model-ecology.bj.bcebos.com/paddlex/PaddleX3.0/deploy/hardware/whl/fastdeploy_xpu-2.3.0.dev0-py3-none-any.whl \
+    && python -m pip install https://paddle-model-ecology.bj.bcebos.com/paddlex/PaddleX3.0/deploy/hardware/whl/paddlepaddle_xpu-0.0.0-cp310-cp310-linux_x86_64.whl
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install "paddleocr${PADDLEOCR_VERSION}" "paddlex${PADDLEX_VERSION}"
+
+RUN groupadd -g 1000 paddleocr \
+    && useradd -m -s /bin/bash -u 1000 -g 1000 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f PaddleOCR-VL-1.6_infer.tar; \
+    fi
+
+ARG BACKEND
+ENV BACKEND=${BACKEND}
+CMD ["/bin/bash", "-c", "paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8080 --backend ${BACKEND}"]
diff --git a/deploy/paddleocr_vl_docker/accelerators/metax-gpu/.env b/deploy/paddleocr_vl_docker/accelerators/metax-gpu/.env
new file mode 100644
index 0000000..1a548b1
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/metax-gpu/.env
@@ -0,0 +1,3 @@
+API_IMAGE_TAG_SUFFIX=latest-metax-gpu-offline
+VLM_BACKEND=fastdeploy
+VLM_IMAGE_TAG_SUFFIX=latest-metax-gpu-offline
diff --git a/deploy/paddleocr_vl_docker/accelerators/metax-gpu/compose.yaml b/deploy/paddleocr_vl_docker/accelerators/metax-gpu/compose.yaml
new file mode 100644
index 0000000..a04f19c
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/metax-gpu/compose.yaml
@@ -0,0 +1,45 @@
+services:
+  paddleocr-vl-api:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:${API_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vl-api
+    ports:
+      - 8080:8080
+    privileged: true
+    devices:
+      - /dev/dri:/dev/dri
+      - /dev/mxcd:/dev/mxcd
+    security_opt:
+      - seccomp=unconfined
+      - apparmor=unconfined
+    shm_size: 64g
+    ulimits:
+      memlock: -1
+    depends_on:
+      paddleocr-vlm-server:
+        condition: service_healthy
+    user: root
+    restart: unless-stopped
+    environment:
+      - VLM_BACKEND=${VLM_BACKEND:-vllm}
+    command: /bin/bash -c "paddlex --serve --pipeline /home/paddleocr/pipeline_config_${VLM_BACKEND}.yaml --device metax_gpu"
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+
+  paddleocr-vlm-server:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-${VLM_BACKEND}-server:${VLM_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vlm-server
+    privileged: true
+    devices:
+      - /dev/dri:/dev/dri
+      - /dev/mxcd:/dev/mxcd
+    security_opt:
+      - seccomp=unconfined
+      - apparmor=unconfined
+    shm_size: 64g
+    ulimits:
+      memlock: -1
+    user: root
+    restart: unless-stopped
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+      start_period: 300s
diff --git a/deploy/paddleocr_vl_docker/accelerators/metax-gpu/pipeline.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/metax-gpu/pipeline.Dockerfile
new file mode 100644
index 0000000..30ab4b8
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/metax-gpu/pipeline.Dockerfile
@@ -0,0 +1,73 @@
+# TODO: Allow regular users
+
+FROM ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddlex-paddle-metax-gpu:3.3.0
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+ENV PYTHONUNBUFFERED=1
+ENV PYTHONDONTWRITEBYTECODE=1
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends \
+        fontconfig \
+        fonts-dejavu-core \
+        fonts-liberation \
+        fonts-noto-cjk \
+        fonts-wqy-microhei \
+        fonts-freefont-ttf \
+    && fc-cache -fv \
+    && rm -rf /var/lib/apt/lists/*
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install "paddleocr[doc-parser]${PADDLEOCR_VERSION}" "paddlex[serving]${PADDLEX_VERSION}"
+
+RUN groupadd -g 1000 paddleocr \
+    && useradd -m -s /bin/bash -u 1000 -g 1000 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+ENV MACA_PATH=/opt/maca
+
+RUN "${MACA_PATH}/tools/cu-bridge/tools/pre_make"
+
+ENV CUDA_PATH="${HOME}/cu-bridge/CUDA_DIR"
+
+ENV LD_LIBRARY_PATH="${CUDA_PATH}/lib64:${MACA_PATH}/lib:${MACA_PATH}/mxgpu_llvm/lib:${LD_LIBRARY_PATH}"
+
+ENV MACA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7"
+ENV PADDLE_XCCL_BACKEND="metax_gpu"
+ENV FLAGS_weight_only_linear_arch=80
+ENV FD_MOE_BACKEND="cutlass"
+ENV FD_ENC_DEC_BLOCK_NUM=2
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayoutV3_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf UVDoc_infer.tar \
+        && mv UVDoc_infer UVDoc \
+        && tar -xf PP-LCNet_x1_0_doc_ori_infer.tar \
+        && mv PP-LCNet_x1_0_doc_ori_infer PP-LCNet_x1_0_doc_ori \
+        && tar -xf PP-DocLayoutV3_infer.tar \
+        && mv PP-DocLayoutV3_infer PP-DocLayoutV3 \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f UVDoc_infer.tar PP-LCNet_x1_0_doc_ori_infer.tar PP-DocLayoutV3_infer.tar PaddleOCR-VL-1.6_infer.tar \
+        && mkdir -p "${HOME}/.paddlex/fonts" \
+        && wget -P "${HOME}/.paddlex/fonts" https://paddle-model-ecology.bj.bcebos.com/paddlex/PaddleX3.0/fonts/PingFang-SC-Regular.ttf; \
+    fi
+
+COPY --chown=paddleocr:paddleocr pipeline_config_vllm.yaml /home/paddleocr
+COPY --chown=paddleocr:paddleocr pipeline_config_fastdeploy.yaml /home/paddleocr
+
+EXPOSE 8080
+
+CMD ["paddlex", "--serve", "--pipeline", "/home/paddleocr/pipeline_config_vllm.yaml", "--device", "metax_gpu"]
diff --git a/deploy/paddleocr_vl_docker/accelerators/metax-gpu/vlm.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/metax-gpu/vlm.Dockerfile
new file mode 100644
index 0000000..f86f0e1
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/metax-gpu/vlm.Dockerfile
@@ -0,0 +1,50 @@
+# TODO: Allow regular users
+
+ARG BACKEND=fastdeploy
+
+
+FROM ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddlex-fastdeploy-metax-gpu:2.3.0 AS base-fastdeploy
+
+
+FROM base-${BACKEND}
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install "paddleocr${PADDLEOCR_VERSION}" "paddlex${PADDLEX_VERSION}"
+
+RUN groupadd -g 1000 paddleocr \
+    && useradd -m -s /bin/bash -u 1000 -g 1000 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+ENV MACA_PATH=/opt/maca
+
+RUN "${MACA_PATH}/tools/cu-bridge/tools/pre_make"
+
+ENV CUDA_PATH="${HOME}/cu-bridge/CUDA_DIR"
+
+ENV LD_LIBRARY_PATH="${CUDA_PATH}/lib64:${MACA_PATH}/lib:${MACA_PATH}/mxgpu_llvm/lib:${LD_LIBRARY_PATH}"
+
+ENV MACA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7"
+# TODO: Set these env vars only in FastDeploy image
+ENV PADDLE_XCCL_BACKEND="metax_gpu"
+ENV FLAGS_weight_only_linear_arch=80
+ENV FD_MOE_BACKEND="cutlass"
+ENV FD_ENC_DEC_BLOCK_NUM=2
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f PaddleOCR-VL-1.6_infer.tar; \
+    fi
+
+ARG BACKEND
+ENV BACKEND=${BACKEND}
+CMD ["/bin/bash", "-c", "paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8080 --backend ${BACKEND}"]
diff --git a/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu-sm120/.env b/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu-sm120/.env
new file mode 100644
index 0000000..982b991
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu-sm120/.env
@@ -0,0 +1,3 @@
+API_IMAGE_TAG_SUFFIX=latest-nvidia-gpu-sm120-offline
+VLM_BACKEND=vllm
+VLM_IMAGE_TAG_SUFFIX=latest-nvidia-gpu-sm120-offline
diff --git a/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu-sm120/compose.yaml b/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu-sm120/compose.yaml
new file mode 100644
index 0000000..045a287
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu-sm120/compose.yaml
@@ -0,0 +1,43 @@
+services:
+  paddleocr-vl-api:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:${API_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vl-api
+    ports:
+      - 8080:8080
+    depends_on:
+      paddleocr-vlm-server:
+        condition: service_healthy
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              device_ids: ["0"]
+              capabilities: [gpu]
+    shm_size: 64g
+    # TODO: Allow using a regular user
+    user: root
+    restart: unless-stopped
+    environment:
+      - VLM_BACKEND=${VLM_BACKEND:-vllm}
+    command: /bin/bash -c "paddlex --serve --pipeline /home/paddleocr/pipeline_config_${VLM_BACKEND}.yaml"
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+
+  paddleocr-vlm-server:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-${VLM_BACKEND}-server:${VLM_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vlm-server
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              device_ids: ["0"]
+              capabilities: [gpu]
+    shm_size: 64g
+    # TODO: Allow using a regular user
+    user: root
+    restart: unless-stopped
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+      start_period: 300s
diff --git a/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu-sm120/pipeline.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu-sm120/pipeline.Dockerfile
new file mode 100644
index 0000000..8b88f1a
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu-sm120/pipeline.Dockerfile
@@ -0,0 +1,61 @@
+FROM python:3.10
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+ENV PYTHONUNBUFFERED=1
+ENV PYTHONDONTWRITEBYTECODE=1
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends libgl1 \
+    && apt-get install -y --no-install-recommends \
+        fontconfig \
+        fonts-dejavu-core \
+        fonts-liberation \
+        fonts-noto-cjk \
+        fonts-wqy-microhei \
+        fonts-freefont-ttf \
+    && fc-cache -fv \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install paddlepaddle-gpu==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/cu129/
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install "paddleocr[doc-parser]${PADDLEOCR_VERSION}" "paddlex[serving]${PADDLEX_VERSION}"
+
+RUN groupadd -g 1000 paddleocr \
+    && useradd -m -s /bin/bash -u 1000 -g 1000 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayoutV3_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf UVDoc_infer.tar \
+        && mv UVDoc_infer UVDoc \
+        && tar -xf PP-LCNet_x1_0_doc_ori_infer.tar \
+        && mv PP-LCNet_x1_0_doc_ori_infer PP-LCNet_x1_0_doc_ori \
+        && tar -xf PP-DocLayoutV3_infer.tar \
+        && mv PP-DocLayoutV3_infer PP-DocLayoutV3 \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f UVDoc_infer.tar PP-LCNet_x1_0_doc_ori_infer.tar PP-DocLayoutV3_infer.tar PaddleOCR-VL-1.6_infer.tar \
+        && mkdir -p "${HOME}/.paddlex/fonts" \
+        && wget -P "${HOME}/.paddlex/fonts" https://paddle-model-ecology.bj.bcebos.com/paddlex/PaddleX3.0/fonts/PingFang-SC-Regular.ttf; \
+    fi
+
+COPY --chown=paddleocr:paddleocr pipeline_config_vllm.yaml /home/paddleocr
+COPY --chown=paddleocr:paddleocr pipeline_config_fastdeploy.yaml /home/paddleocr
+
+EXPOSE 8080
+
+CMD ["paddlex", "--serve", "--pipeline", "/home/paddleocr/pipeline_config_vllm.yaml"]
diff --git a/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu-sm120/vlm.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu-sm120/vlm.Dockerfile
new file mode 100644
index 0000000..10a9857
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu-sm120/vlm.Dockerfile
@@ -0,0 +1,33 @@
+ARG BACKEND=vllm
+
+
+FROM ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddlex-genai-vllm-server:latest-sm120 AS base-vllm
+
+
+FROM base-${BACKEND}
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install "paddleocr${PADDLEOCR_VERSION}" "paddlex${PADDLEX_VERSION}"
+
+RUN groupadd -g 1000 paddleocr \
+    && useradd -m -s /bin/bash -u 1000 -g 1000 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f PaddleOCR-VL-1.6_infer.tar; \
+    fi
+
+ARG BACKEND
+ENV BACKEND=${BACKEND}
+CMD ["/bin/bash", "-c", "paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8080 --backend ${BACKEND}"]
diff --git a/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu/.env b/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu/.env
new file mode 100644
index 0000000..78f8f47
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu/.env
@@ -0,0 +1,3 @@
+API_IMAGE_TAG_SUFFIX=latest-nvidia-gpu-offline
+VLM_BACKEND=vllm
+VLM_IMAGE_TAG_SUFFIX=latest-nvidia-gpu-offline
diff --git a/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu/compose.yaml b/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu/compose.yaml
new file mode 100644
index 0000000..045a287
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu/compose.yaml
@@ -0,0 +1,43 @@
+services:
+  paddleocr-vl-api:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:${API_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vl-api
+    ports:
+      - 8080:8080
+    depends_on:
+      paddleocr-vlm-server:
+        condition: service_healthy
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              device_ids: ["0"]
+              capabilities: [gpu]
+    shm_size: 64g
+    # TODO: Allow using a regular user
+    user: root
+    restart: unless-stopped
+    environment:
+      - VLM_BACKEND=${VLM_BACKEND:-vllm}
+    command: /bin/bash -c "paddlex --serve --pipeline /home/paddleocr/pipeline_config_${VLM_BACKEND}.yaml"
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+
+  paddleocr-vlm-server:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-${VLM_BACKEND}-server:${VLM_IMAGE_TAG_SUFFIX}
+    container_name: paddleocr-vlm-server
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              device_ids: ["0"]
+              capabilities: [gpu]
+    shm_size: 64g
+    # TODO: Allow using a regular user
+    user: root
+    restart: unless-stopped
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+      start_period: 300s
diff --git a/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu/pipeline.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu/pipeline.Dockerfile
new file mode 100644
index 0000000..23a378a
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu/pipeline.Dockerfile
@@ -0,0 +1,61 @@
+FROM python:3.10
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+ENV PYTHONUNBUFFERED=1
+ENV PYTHONDONTWRITEBYTECODE=1
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends libgl1 \
+    && apt-get install -y --no-install-recommends \
+        fontconfig \
+        fonts-dejavu-core \
+        fonts-liberation \
+        fonts-noto-cjk \
+        fonts-wqy-microhei \
+        fonts-freefont-ttf \
+    && fc-cache -fv \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install paddlepaddle-gpu==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/cu126/
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install "paddleocr[doc-parser]${PADDLEOCR_VERSION}" "paddlex[serving]${PADDLEX_VERSION}"
+
+RUN groupadd -g 1000 paddleocr \
+    && useradd -m -s /bin/bash -u 1000 -g 1000 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayoutV3_infer.tar \
+            https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf UVDoc_infer.tar \
+        && mv UVDoc_infer UVDoc \
+        && tar -xf PP-LCNet_x1_0_doc_ori_infer.tar \
+        && mv PP-LCNet_x1_0_doc_ori_infer PP-LCNet_x1_0_doc_ori \
+        && tar -xf PP-DocLayoutV3_infer.tar \
+        && mv PP-DocLayoutV3_infer PP-DocLayoutV3 \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f UVDoc_infer.tar PP-LCNet_x1_0_doc_ori_infer.tar PP-DocLayoutV3_infer.tar PaddleOCR-VL-1.6_infer.tar \
+        && mkdir -p "${HOME}/.paddlex/fonts" \
+        && wget -P "${HOME}/.paddlex/fonts" https://paddle-model-ecology.bj.bcebos.com/paddlex/PaddleX3.0/fonts/PingFang-SC-Regular.ttf; \
+    fi
+
+COPY --chown=paddleocr:paddleocr pipeline_config_vllm.yaml /home/paddleocr
+COPY --chown=paddleocr:paddleocr pipeline_config_fastdeploy.yaml /home/paddleocr
+
+EXPOSE 8080
+
+CMD ["paddlex", "--serve", "--pipeline", "/home/paddleocr/pipeline_config_vllm.yaml"]
diff --git a/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu/vlm.Dockerfile b/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu/vlm.Dockerfile
new file mode 100644
index 0000000..53c19ca
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu/vlm.Dockerfile
@@ -0,0 +1,36 @@
+ARG BACKEND=vllm
+
+
+FROM ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddlex-genai-vllm-server:latest AS base-vllm
+
+
+FROM ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/fastdeploy-cuda-12.6:2.3.0 AS base-fastdeploy
+
+
+FROM base-${BACKEND}
+
+ARG PADDLEOCR_VERSION=">=3.4.0,<3.5"
+ARG PADDLEX_VERSION=">=3.4.0,<3.5"
+RUN --mount=type=cache,target=/root/.cache/pip \
+    python -m pip install "paddleocr${PADDLEOCR_VERSION}" "paddlex${PADDLEX_VERSION}"
+
+RUN groupadd -g 1000 paddleocr \
+    && useradd -m -s /bin/bash -u 1000 -g 1000 paddleocr
+ENV HOME=/home/paddleocr
+WORKDIR /home/paddleocr
+
+USER paddleocr
+
+ARG BUILD_FOR_OFFLINE=false
+RUN if [ "${BUILD_FOR_OFFLINE}" = 'true' ]; then \
+        mkdir -p "${HOME}/.paddlex/official_models" \
+        && cd "${HOME}/.paddlex/official_models" \
+        && wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PaddleOCR-VL-1.6_infer.tar \
+        && tar -xf PaddleOCR-VL-1.6_infer.tar \
+        && mv PaddleOCR-VL-1.6_infer PaddleOCR-VL-1.6 \
+        && rm -f PaddleOCR-VL-1.6_infer.tar; \
+    fi
+
+ARG BACKEND
+ENV BACKEND=${BACKEND}
+CMD ["/bin/bash", "-c", "paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8080 --backend ${BACKEND}"]
diff --git a/deploy/paddleocr_vl_docker/build_pipeline.sh b/deploy/paddleocr_vl_docker/build_pipeline.sh
new file mode 100755
index 0000000..9daf8ec
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/build_pipeline.sh
@@ -0,0 +1,266 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+source "${SCRIPT_DIR}/pip_version_arg.sh"
+
+device_type='nvidia-gpu'
+build_for_offline='false'
+paddleocr_version='3.4.0'
+paddlex_version='3.4.0'
+platform='linux/amd64'
+action='load'
+registry='ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle'
+builder=''
+
+show_usage() {
+    cat << EOF
+Usage: $(basename "$0") [OPTIONS]
+
+Options:
+  --device-type <type>      Device type (nvidia-gpu|nvidia-gpu-sm120|hygon-dcu|kunlunxin-xpu|metax-gpu|iluvatar-gpu|huawei-npu|intel-gpu|amd-gpu) [default: ${device_type}]
+  --offline                 Build offline version
+  --ppocr-version <spec>    PaddleOCR version or URL [default: ${paddleocr_version}]
+  --pdx-version <spec>      PaddleX version or URL [default: ${paddlex_version}]
+  --platform <platform>     Build platform [default: ${platform}]
+  --action <action>         Post-build action: load|push|tar|none [default: ${action}]
+                            load: Load to local Docker
+                            push: Push to image registry
+                            tar: Export as tar file
+                            none: Build only, no output
+  --registry <registry>     Custom image registry [default: ${registry}]
+  --builder <name>          Buildx builder name (override default)
+  -h, --help               Show this help message
+
+Examples:
+  $0 --device-type nvidia-gpu --action push
+  $0 --device-type nvidia-gpu --platform linux/amd64,linux/arm64 --action push
+  $0 --device-type nvidia-gpu --action tar --platform linux/amd64
+  $0 --pdx-version git+https://github.com/PaddlePaddle/PaddleX.git@main
+EOF
+}
+
+while [[ $# -gt 0 ]]; do
+    case $1 in
+        --device-type)
+            [ -z "$2" ] && {
+                echo "Error: '--device-type' requires a value" >&2
+                exit 2
+            }
+            device_type="$2"
+            shift
+            shift
+            case "${device_type}" in
+                nvidia-gpu|nvidia-gpu-sm120|hygon-dcu|kunlunxin-xpu|metax-gpu|iluvatar-gpu|huawei-npu|intel-gpu|amd-gpu)
+                    ;;
+                *)
+                    echo "Error: Unknown device type: ${device_type}" >&2
+                    exit 2
+                    ;;
+            esac
+            ;;
+        --offline)
+            build_for_offline='true'
+            shift
+            ;;
+        --ppocr-version)
+            [ -z "$2" ] && {
+                echo "Error: '--ppocr-version' requires a value" >&2
+                exit 2
+            }
+            paddleocr_version="$2"
+            shift
+            shift
+            ;;
+        --pdx-version)
+            [ -z "$2" ] && {
+                echo "Error: '--pdx-version' requires a value" >&2
+                exit 2
+            }
+            paddlex_version="$2"
+            shift
+            shift
+            ;;
+        --platform)
+            [ -z "$2" ] && {
+                echo "Error: '--platform' requires a value" >&2
+                exit 2
+            }
+            platform="$2"
+            shift
+            shift
+            ;;
+        --action)
+            [ -z "$2" ] && {
+                echo "Error: '--action' requires a value" >&2
+                exit 2
+            }
+            action="$2"
+            shift
+            shift
+            case "${action}" in
+                load|push|tar|none)
+                    ;;
+                *)
+                    echo "Error: Unknown action: ${action}. Use load|push|tar|none" >&2
+                    exit 2
+                    ;;
+            esac
+            ;;
+        --registry)
+            [ -z "$2" ] && {
+                echo "Error: '--registry' requires a value" >&2
+                exit 2
+            }
+            registry="$2"
+            shift
+            shift
+            ;;
+        --builder)
+            [ -z "${2-}" ] && {
+                echo "Error: '--builder' requires a value" >&2
+                exit 2
+            }
+            builder="$2"
+            shift 2
+            ;;
+        -h|--help)
+            show_usage
+            exit 0
+            ;;
+        *)
+            echo "Error: Unknown option: $1" >&2
+            echo "Use $0 --help for usage information" >&2
+            exit 2
+            ;;
+    esac
+done
+
+paddleocr_pip_suffix="$(pip_build_arg_suffix "${paddleocr_version}")"
+paddlex_pip_suffix="$(pip_build_arg_suffix "${paddlex_version}")"
+
+# Validate platform compatibility for load action
+if [[ "${action}" == 'load' ]] && [[ "${platform}" == *','* ]]; then
+    echo "Error: Cannot use --action load with multiple platforms" >&2
+    echo "Platform: ${platform}" >&2
+    echo "Please specify a single platform or use other actions" >&2
+    exit 2
+fi
+
+# Build tag suffix
+tag_suffix="latest-${device_type}"
+
+if [ "${build_for_offline}" = 'true' ]; then
+    tag_suffix="${tag_suffix}-offline"
+fi
+
+dockerfile="accelerators/${device_type}/pipeline.Dockerfile"
+if [ ! -f "${dockerfile}" ]; then
+    echo "Error: Dockerfile not found: ${dockerfile}" >&2
+    exit 1
+fi
+
+revision="$(git rev-parse --short HEAD)"
+image_version="${revision}-ppocr$(version_tag_label "${paddleocr_version}")-pdx$(version_tag_label "${paddlex_version}")"
+
+# Image name
+base_image_name='paddleocr-vl'
+
+# Main tags
+main_tag="${registry}/${base_image_name}:${tag_suffix}"
+version_tag="${registry}/${base_image_name}:${tag_suffix/latest/${image_version}}"
+paddleocr_version_tag="${registry}/${base_image_name}:${tag_suffix/latest/paddleocr$(version_tag_label "${paddleocr_version}")}"
+
+# Build arguments array
+build_args=(
+    '--platform' "${platform}"
+    '-f' "${dockerfile}"
+    '-t' "${main_tag}"
+    '-t' "${version_tag}"
+    '-t' "${paddleocr_version_tag}"
+    '--build-arg' "BUILD_FOR_OFFLINE=${build_for_offline}"
+    '--build-arg' "PADDLEOCR_VERSION=${paddleocr_pip_suffix}"
+    '--build-arg' "PADDLEX_VERSION=${paddlex_pip_suffix}"
+    '--build-arg' "http_proxy=${http_proxy:-}"
+    '--build-arg' "https_proxy=${https_proxy:-}"
+    '--build-arg' "no_proxy=${no_proxy:-}"
+    '--label' "org.opencontainers.image.version=${image_version}"
+    '--label' "org.opencontainers.image.revision=${revision}"
+    '.'
+)
+
+if [[ -n "${builder}" ]]; then
+    build_args=('--builder' "${builder}" "${build_args[@]}")
+fi
+
+echo "========================================="
+echo "Build Configuration:"
+echo "  Device Type:     ${device_type}"
+echo "  Offline Mode:    ${build_for_offline}"
+echo "  PaddleOCR:       ${paddleocr_version}"
+echo "  PaddleX:         ${paddlex_version}"
+echo "  Platform:        ${platform}"
+echo "  Action:          ${action}"
+echo "  Registry:        ${registry}"
+echo "  Builder:         ${builder:-<default>}"
+echo "  Dockerfile:      ${dockerfile}"
+echo "  Tags:"
+echo "    - ${main_tag}"
+echo "    - ${version_tag}"
+echo "========================================="
+
+# Add action-specific build options
+case "${action}" in
+    load)
+        echo "Building and loading to local Docker..."
+        build_args+=('--load')
+        ;;
+    push)
+        echo "Building and pushing to registry..."
+        build_args+=('--push')
+        if ! docker info 2>/dev/null | grep -q "Username"; then
+            echo "Warning: Docker login not detected. Push may fail."
+            echo "Tip: Run 'docker login ${registry%%/*}' to login"
+        fi
+        ;;
+    tar)
+        echo "Building and exporting as tar file..."
+        output_dir=./build-output
+        mkdir -p "${output_dir}"
+        tar_file="${output_dir}/${base_image_name}-${tag_suffix}-${platform//\//_}.tar"
+        build_args+=("--output" "type=docker,dest=${tar_file}")
+        echo "  Output file: ${tar_file}"
+        ;;
+    none)
+        echo "Building only (image stays in BuildKit cache)..."
+        echo "Note: Image will be lost if BuildKit cache is cleared"
+        ;;
+esac
+
+# Execute build command
+echo "Executing: docker buildx build ${build_args[*]}"
+docker buildx build "${build_args[@]}"
+
+# Post-build actions
+case "${action}" in
+    load)
+        echo "✅ Image loaded to local Docker"
+        echo "   View: docker images | grep ${base_image_name}"
+        ;;
+    push)
+        echo "✅ Image pushed to registry"
+        echo "   Pull: docker pull ${main_tag}"
+        echo "   Inspect: docker buildx imagetools inspect ${main_tag}"
+        ;;
+    tar)
+        echo "✅ Image exported as tar file"
+        echo "   File size: $(du -h "${tar_file}" | cut -f1)"
+        echo "   Load to Docker: docker load -i ${tar_file}"
+        ;;
+    none)
+        echo "ℹ️  Image saved in BuildKit cache only"
+        echo "   Use --action load to rebuild and load quickly"
+        echo "   Or use --action push to push directly"
+        ;;
+esac
diff --git a/deploy/paddleocr_vl_docker/build_vlm.sh b/deploy/paddleocr_vl_docker/build_vlm.sh
new file mode 100755
index 0000000..554ed88
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/build_vlm.sh
@@ -0,0 +1,286 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+source "${SCRIPT_DIR}/pip_version_arg.sh"
+
+device_type='nvidia-gpu'
+backend='vllm'
+build_for_offline='false'
+paddleocr_version='3.4.0'
+paddlex_version='3.4.0'
+platform='linux/amd64'
+action='load'
+registry='ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle'
+builder=''
+
+show_usage() {
+    cat << EOF
+Usage: $(basename "$0") [OPTIONS]
+
+Options:
+  --device-type <type>      Device type (nvidia-gpu|nvidia-gpu-sm120|hygon-dcu|kunlunxin-xpu|metax-gpu|iluvatar-gpu|huawei-npu|intel-gpu|amd-gpu) [default: ${device_type}]
+  --backend <backend>       Backend type (vllm|fastdeploy) [default: ${backend}]
+  --offline                 Build offline version
+  --ppocr-version <spec>    PaddleOCR version or URL [default: ${paddleocr_version}]
+  --pdx-version <spec>      PaddleX version or URL [default: ${paddlex_version}]
+  --platform <platform>     Build platform [default: ${platform}]
+  --action <action>         Post-build action: load|push|tar|none [default: ${action}]
+                            load: Load to local Docker
+                            push: Push to image registry
+                            tar: Export as tar file
+                            none: Build only, no output
+  --registry <registry>     Custom image registry [default: ${registry}]
+  --builder <name>          Buildx builder name (override default)
+  -h, --help               Show this help message
+
+Examples:
+  $0 --device-type nvidia-gpu --backend vllm --action push
+  $0 --platform linux/amd64,linux/arm64 --action push
+  $0 --action tar --platform linux/amd64
+  $0 --ppocr-version https://github.com/PaddlePaddle/PaddleOCR/archive/main.zip
+EOF
+}
+
+while [[ $# -gt 0 ]]; do
+    case $1 in
+        --device-type)
+            [ -z "$2" ] && {
+                echo "Error: '--device-type' requires a value" >&2
+                exit 2
+            }
+            device_type="$2"
+            shift
+            shift
+            case "${device_type}" in
+                nvidia-gpu|nvidia-gpu-sm120|hygon-dcu|kunlunxin-xpu|metax-gpu|iluvatar-gpu|huawei-npu|intel-gpu|amd-gpu)
+                    ;;
+                *)
+                    echo "Error: Unknown device type: ${device_type}" >&2
+                    exit 2
+                    ;;
+            esac
+            ;;
+        --backend)
+            [ -z "$2" ] && {
+                echo "Error: '--backend' requires a value" >&2
+                exit 2
+            }
+            backend="$2"
+            shift 2
+            case "${backend}" in
+                vllm|fastdeploy)
+                    ;;
+                *)
+                    echo "Error: Unknown backend: ${backend}" >&2
+                    exit 2
+                    ;;
+            esac
+            ;;
+        --offline)
+            build_for_offline='true'
+            shift
+            ;;
+        --ppocr-version)
+            [ -z "$2" ] && {
+                echo "Error: '--ppocr-version' requires a value" >&2
+                exit 2
+            }
+            paddleocr_version="$2"
+            shift
+            shift
+            ;;
+        --pdx-version)
+            [ -z "$2" ] && {
+                echo "Error: '--pdx-version' requires a value" >&2
+                exit 2
+            }
+            paddlex_version="$2"
+            shift
+            shift
+            ;;
+        --platform)
+            [ -z "$2" ] && {
+                echo "Error: '--platform' requires a value" >&2
+                exit 2
+            }
+            platform="$2"
+            shift
+            shift
+            ;;
+        --action)
+            [ -z "$2" ] && {
+                echo "Error: '--action' requires a value" >&2
+                exit 2
+            }
+            action="$2"
+            shift
+            shift
+            case "${action}" in
+                load|push|tar|none)
+                    ;;
+                *)
+                    echo "Error: Unknown action: ${action}. Use load|push|tar|none" >&2
+                    exit 2
+                    ;;
+            esac
+            ;;
+        --registry)
+            [ -z "$2" ] && {
+                echo "Error: '--registry' requires a value" >&2
+                exit 2
+            }
+            registry="$2"
+            shift
+            shift
+            ;;
+        --builder)
+            [ -z "${2-}" ] && {
+                echo "Error: '--builder' requires a value" >&2
+                exit 2
+            }
+            builder="$2"
+            shift 2
+            ;;
+        -h|--help)
+            show_usage
+            exit 0
+            ;;
+        *)
+            echo "Error: Unknown option: $1" >&2
+            echo "Use $0 --help for usage information" >&2
+            exit 2
+            ;;
+    esac
+done
+
+paddleocr_pip_suffix="$(pip_build_arg_suffix "${paddleocr_version}")"
+paddlex_pip_suffix="$(pip_build_arg_suffix "${paddlex_version}")"
+
+# Validate platform compatibility for load action
+if [[ "${action}" == 'load' ]] && [[ "${platform}" == *','* ]]; then
+    echo "Error: Cannot use --action load with multiple platforms" >&2
+    echo "Platform: ${platform}" >&2
+    echo "Please specify a single platform or use other actions" >&2
+    exit 2
+fi
+
+# Build tag suffix
+tag_suffix="latest-${device_type}"
+
+if [ "${build_for_offline}" = 'true' ]; then
+    tag_suffix="${tag_suffix}-offline"
+fi
+
+dockerfile="accelerators/${device_type}/vlm.Dockerfile"
+if [ ! -f "${dockerfile}" ]; then
+    echo "Error: Dockerfile not found: ${dockerfile}" >&2
+    exit 1
+fi
+
+revision="$(git rev-parse --short HEAD)"
+image_version="${revision}-ppocr$(version_tag_label "${paddleocr_version}")-pdx$(version_tag_label "${paddlex_version}")"
+
+# Image name
+base_image_name="paddleocr-genai-${backend}-server"
+
+# Main tags
+main_tag="${registry}/${base_image_name}:${tag_suffix}"
+version_tag="${registry}/${base_image_name}:${tag_suffix/latest/${image_version}}"
+paddleocr_version_tag="${registry}/${base_image_name}:${tag_suffix/latest/paddleocr$(version_tag_label "${paddleocr_version}")}"
+
+# Build arguments array
+build_args=(
+    '--platform' "${platform}"
+    '-f' "${dockerfile}"
+    '-t' "${main_tag}"
+    '-t' "${version_tag}"
+    '-t' "${paddleocr_version_tag}"
+    '--build-arg' "BUILD_FOR_OFFLINE=${build_for_offline}"
+    '--build-arg' "PADDLEOCR_VERSION=${paddleocr_pip_suffix}"
+    '--build-arg' "PADDLEX_VERSION=${paddlex_pip_suffix}"
+    '--build-arg' "BACKEND=${backend}"
+    '--build-arg' "http_proxy=${http_proxy:-}"
+    '--build-arg' "https_proxy=${https_proxy:-}"
+    '--build-arg' "no_proxy=${no_proxy:-}"
+    '--label' "org.opencontainers.image.version=${image_version}"
+    '--label' "org.opencontainers.image.revision=${revision}"
+    '.'
+)
+
+if [[ -n "${builder}" ]]; then
+    build_args=('--builder' "${builder}" "${build_args[@]}")
+fi
+
+echo "========================================="
+echo "Build Configuration:"
+echo "  Device Type:     ${device_type}"
+echo "  Backend:         ${backend}"
+echo "  Offline Mode:    ${build_for_offline}"
+echo "  PaddleOCR:       ${paddleocr_version}"
+echo "  PaddleX:         ${paddlex_version}"
+echo "  Platform:        ${platform}"
+echo "  Action:          ${action}"
+echo "  Registry:        ${registry}"
+echo "  Builder:         ${builder:-<default>}"
+echo "  Dockerfile:      ${dockerfile}"
+echo "  Tags:"
+echo "    - ${main_tag}"
+echo "    - ${version_tag}"
+echo "========================================="
+
+# Add action-specific build options
+case "${action}" in
+    load)
+        echo "Building and loading to local Docker..."
+        build_args+=('--load')
+        ;;
+    push)
+        echo "Building and pushing to registry..."
+        build_args+=('--push')
+        if ! docker info 2>/dev/null | grep -q "Username"; then
+            echo "Warning: Docker login not detected. Push may fail."
+            echo "Tip: Run 'docker login ${registry%%/*}' to login"
+        fi
+        ;;
+    tar)
+        echo "Building and exporting as tar file..."
+        output_dir=./build-output
+        mkdir -p "${output_dir}"
+        tar_file="${output_dir}/${base_image_name}-${tag_suffix}-${platform//\//_}.tar"
+        build_args+=('--output' "type=docker,dest=${tar_file}")
+        echo "  Output file: ${tar_file}"
+        ;;
+    none)
+        echo "Building only (image stays in BuildKit cache)..."
+        echo "Note: Image will be lost if BuildKit cache is cleared"
+        ;;
+esac
+
+# Execute build command
+echo "Executing: docker buildx build ${build_args[*]}"
+docker buildx build "${build_args[@]}"
+
+# Post-build actions
+case "${action}" in
+    load)
+        echo "✅ Image loaded to local Docker"
+        echo "   View: docker images | grep ${base_image_name}"
+        ;;
+    push)
+        echo "✅ Image pushed to registry"
+        echo "   Pull: docker pull ${main_tag}"
+        echo "   Inspect: docker buildx imagetools inspect ${main_tag}"
+        ;;
+    tar)
+        echo "✅ Image exported as tar file"
+        echo "   File size: $(du -h "${tar_file}" | cut -f1)"
+        echo "   Load to Docker: docker load -i ${tar_file}"
+        ;;
+    none)
+        echo "ℹ️  Image saved in BuildKit cache only"
+        echo "   Use --action load to rebuild and load quickly"
+        echo "   Or use --action push to push directly"
+        ;;
+esac
diff --git a/deploy/paddleocr_vl_docker/hps/.env.example b/deploy/paddleocr_vl_docker/hps/.env.example
new file mode 100644
index 0000000..e1b5cff
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/hps/.env.example
@@ -0,0 +1,58 @@
+# PaddleOCR-VL High-Performance Serving Configuration
+# Copy this file to .env and modify as needed
+
+# =============================================================================
+# Pipeline and SDK Configuration
+# =============================================================================
+
+# Pipeline name. Applies to the whole PaddleOCR-VL series, e.g.:
+#   PaddleOCR-VL, PaddleOCR-VL-1.5, PaddleOCR-VL-1.6
+HPS_PIPELINE_NAME=PaddleOCR-VL-1.6
+
+# PaddleX version (major.minor only, e.g. 3.6). Keeps the pipeline base image and
+# the high-stability serving SDK in sync:
+#   3.6 -> base image .../hps:paddlex3.6-gpu  +  SDK release dir v3.6
+HPS_PADDLEX_VERSION=3.6
+
+# Derived SDK directory name. Must match `paddlex_hps_${HPS_PIPELINE_NAME}_sdk`.
+HPS_SDK_DIR=paddlex_hps_PaddleOCR-VL-1.6_sdk
+
+# =============================================================================
+# Gateway Configuration
+# =============================================================================
+
+# Maximum concurrent inference requests (layout-parsing / infer)
+# Increase for higher throughput, decrease if device is overloaded
+HPS_MAX_CONCURRENT_INFERENCE_REQUESTS=16
+
+# Maximum concurrent non-inference requests (restructure-pages)
+# Can be set higher since non-inference operations are lighter
+HPS_MAX_CONCURRENT_NON_INFERENCE_REQUESTS=64
+
+# Inference timeout in seconds
+# Increase for complex documents, decrease for faster failure detection
+HPS_INFERENCE_TIMEOUT=600
+
+# Health check timeout in seconds
+HPS_HEALTH_CHECK_TIMEOUT=5
+
+# VLM server URL
+# HPS_VLM_URL=http://paddleocr-vlm-server:8080
+
+# Log level: DEBUG, INFO, WARNING, ERROR
+HPS_LOG_LEVEL=INFO
+
+# Filter health check endpoints from access logs (true/false)
+# Set to false to see all access logs including health checks
+HPS_FILTER_HEALTH_ACCESS_LOG=true
+
+# Number of Uvicorn worker processes
+# Recommended: 2-4 workers per CPU core
+HPS_UVICORN_WORKERS=4
+
+# =============================================================================
+# Device Configuration
+# =============================================================================
+
+# Inference device ID to use (0, 1, 2, etc.)
+HPS_DEVICE_ID=0
diff --git a/deploy/paddleocr_vl_docker/hps/README.md b/deploy/paddleocr_vl_docker/hps/README.md
new file mode 100644
index 0000000..a03899c
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/hps/README.md
@@ -0,0 +1,234 @@
+# PaddleOCR-VL 高性能服务化部署
+
+[English](README_en.md)
+
+本目录提供一套支持并发请求处理的 **PaddleOCR-VL 系列**高性能服务化部署方案，适用于 `PaddleOCR-VL`、`PaddleOCR-VL-1.5`、`PaddleOCR-VL-1.6` 等产线版本。
+
+> 本方案目前暂时只支持 NVIDIA GPU，对其他推理设备的支持仍在完善中。
+
+## 架构
+
+```
+客户端 → FastAPI 网关 → Triton 服务器 → vLLM 服务器
+```
+
+| 组件　　　　　| 说明　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　 |
+| ---------------| ----------------------------------------------------------------------------------|
+| FastAPI 网关　| 统一访问入口、简化客户端调用、并发控制　　　　　　　　　　　　　　　　　　　　　 |
+| Triton 服务器 | 版面分析模型（如 PP-DocLayoutV3）及产线串联逻辑，负责模型管理、动态批处理、推理调度 |
+| vLLM 服务器　 | VLM，连续批处理推理　　　　　　　　　　　　　　　　　　　　|
+
+**Triton 模型：**
+
+| 模型 | 设备 | 说明 |
+|------|------|------|
+| `layout-parsing` | 推理设备（如 GPU） | 版面解析推理 |
+| `restructure-pages` | CPU | 多页结果后处理（跨页表格合并、标题层级重分配） |
+
+## 环境要求
+
+- x64 CPU
+- NVIDIA GPU，Compute Capability >= 8.0 且 < 10.0
+- NVIDIA 驱动支持 CUDA 12.6
+- Docker >= 19.03
+- Docker Compose >= 2.0
+
+## 快速开始
+
+1. 拉取 PaddleOCR 源码并切换到当前目录：
+
+```bash
+git clone https://github.com/PaddlePaddle/PaddleOCR.git
+cd PaddleOCR/deploy/paddleocr_vl_docker/hps
+```
+
+2. 准备必要文件：
+
+```bash
+cp .env.example .env
+# 按需修改 .env 中的 HPS_PIPELINE_NAME
+bash prepare.sh
+```
+
+`prepare.sh` 会根据 `.env` 下载对应 PaddleOCR-VL 版本的高稳定性服务化部署 SDK，并写入 Triton 产线配置。
+
+3. 启动服务：
+
+```bash
+docker compose up
+```
+
+上述命令将依次启动 3 个容器：
+
+| 服务 | 说明 | 端口 |
+|------|------|------|
+| `paddleocr-vl-api` | FastAPI 网关（对外入口） | 8080 |
+| `paddleocr-vl-pipeline` | 运行产线的 Triton 推理服务器 | 8000（内部） |
+| `paddleocr-vlm-server` | 基于 vLLM 的 VLM 推理服务 | 8080（内部） |
+
+> 首次启动会自动下载并构建镜像，耗时较长；从第二次启动起将直接使用本地镜像，启动速度更快。
+
+## 配置说明
+
+### 环境变量
+
+复制 `.env.example` 到 `.env` 并根据需要修改。
+
+```bash
+cp .env.example .env
+```
+
+除了通过 `.env` 文件设置，也可以直接设置环境变量，如：
+
+```bash
+export HPS_MAX_CONCURRENT_INFERENCE_REQUESTS=8
+```
+
+#### 产线与 SDK 配置
+
+通过以下变量选择 PaddleOCR-VL 系列中的具体版本（修改后需重新执行 `prepare.sh` 并重建镜像）：
+
+本方案复用 PaddleX 的[高稳定性服务化部署](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/serving.html#2) SDK 作为 Triton 服务的基础模型仓库与客户端依赖，并在其基础上增加 PaddleOCR-VL 专用的 FastAPI 网关和 vLLM 服务编排。
+
+| 变量 | 默认值 | 说明 |
+|------|--------|------|
+| `HPS_PIPELINE_NAME` | `PaddleOCR-VL-1.6` | 产线名称 |
+| `HPS_PADDLEX_VERSION` | `3.6` | PaddleX 版本（仅填 major.minor，如 `3.6`），同时决定 Triton 基础镜像标签（`paddlex${HPS_PADDLEX_VERSION}-gpu`）和 SDK 发布目录（`v${HPS_PADDLEX_VERSION}`），二者保持一致 |
+| `HPS_SDK_DIR` | `paddlex_hps_PaddleOCR-VL-1.6_sdk` | 解压后的 SDK 目录，遵循 `paddlex_hps_${HPS_PIPELINE_NAME}_sdk` |
+
+常见配置示例：
+
+| 目标版本 | `HPS_PIPELINE_NAME` | `HPS_SDK_DIR` |
+|----------|-----------------|---------------|
+| PaddleOCR-VL-1.6 | `PaddleOCR-VL-1.6` | `paddlex_hps_PaddleOCR-VL-1.6_sdk` |
+| PaddleOCR-VL-1.5 | `PaddleOCR-VL-1.5` | `paddlex_hps_PaddleOCR-VL-1.5_sdk` |
+| PaddleOCR-VL (v1) | `PaddleOCR-VL` | `paddlex_hps_PaddleOCR-VL_sdk` |
+
+#### 网关与设备
+
+| 变量　　　　　　　　　　　　　　　　　　　　| 默认值　　　　　　　　　　　　　 | 说明　　　　　　　　　　　　　　　　　　|
+| ---------------------------------------------| ----------------------------------| -----------------------------------------|
+| `HPS_MAX_CONCURRENT_INFERENCE_REQUESTS`　　 | 16　　　　　　　　　　　　　　　 | 推理操作（版面解析）最大并发请求数　　　|
+| `HPS_MAX_CONCURRENT_NON_INFERENCE_REQUESTS` | 64　　　　　　　　　　　　　　　 | 非推理操作（多页重组）最大并发请求数　　|
+| `HPS_INFERENCE_TIMEOUT`　　　　　　　　　　 | 600　　　　　　　　　　　　　　　| 请求超时时间（秒）　　　　　　　　　　　|
+| `HPS_HEALTH_CHECK_TIMEOUT`　　　　　　　　　| 5　　　　　　　　　　　　　　　　| 健康检查超时时间（秒）　　　　　　　　　|
+| `HPS_VLM_URL`　　　　　　　　　　　　　　　 | http://paddleocr-vlm-server:8080 | VLM 服务器地址 |
+| `HPS_LOG_LEVEL`　　　　　　　　　　　　　　 | INFO　　　　　　　　　　　　　　 | 日志级别（DEBUG, INFO, WARNING, ERROR） |
+| `HPS_FILTER_HEALTH_ACCESS_LOG`　　　　　　　| true　　　　　　　　　　　　　　 | 是否过滤健康检查的访问日志　　　　　　　|
+| `HPS_UVICORN_WORKERS`　　　　　　　　　　　 | 4　　　　　　　　　　　　　　　　| 网关 Worker 进程数　　　　　　　　　　　|
+| `HPS_DEVICE_ID`　　　　　　　　　　　　　　 | 0　　　　　　　　　　　　　　　　| 使用的推理设备 ID　　　　　　　　　　　 |
+
+### 产线配置调整
+
+如需调整产线相关配置（如模型路径、批处理大小、部署设备等），请参考 [PaddleOCR-VL 使用教程](https://github.com/PaddlePaddle/PaddleOCR/blob/main/docs/version3.x/pipeline_usage/PaddleOCR-VL.md) 中的产线配置调整说明章节。
+
+## API 使用
+
+### 文档解析
+
+请参考 [PaddleOCR-VL 使用教程](https://github.com/PaddlePaddle/PaddleOCR/blob/main/docs/version3.x/pipeline_usage/PaddleOCR-VL.md) 中的客户端调用相关章节。
+
+服务支持 PDF 或图像文件（含 TIFF，多页时按页处理）；多页 TIFF 请使用 `fileType=1`。
+
+### 健康检查
+
+```bash
+# 存活检查
+curl http://localhost:8080/health
+
+# 就绪检查（验证 Triton 和 VLM 服务是否已准备好处理请求）
+curl http://localhost:8080/health/ready
+```
+
+## 性能调优
+
+### 并发设置
+
+网关对推理操作和非推理操作各自独立地进行并发控制：
+
+- **`HPS_MAX_CONCURRENT_INFERENCE_REQUESTS`**（默认 16）：控制 `layout-parsing`（版面解析）等推理操作的并发数
+  - 过低（4）：推理设备利用率不足，请求不必要地排队
+  - 过高（64）：可能导致 Triton 过载，出现 OOM 或超时
+  - 默认值 16 允许在当前批次处理时有足够请求排队形成下一批次
+  - 如推理设备资源有限，建议适当降低此值
+- **`HPS_MAX_CONCURRENT_NON_INFERENCE_REQUESTS`**（默认 64）：控制 `restructure-pages`（多页重组）等非推理操作的并发数
+  - 非推理操作不占用推理设备资源，可以设置更高的并发数
+  - 可根据 CPU 核数和内存情况调整
+
+**高吞吐配置示例：**
+
+```bash
+# .env
+HPS_MAX_CONCURRENT_INFERENCE_REQUESTS=32
+HPS_MAX_CONCURRENT_NON_INFERENCE_REQUESTS=128
+HPS_UVICORN_WORKERS=8
+```
+
+**低延迟配置示例：**
+
+```bash
+# .env
+HPS_MAX_CONCURRENT_INFERENCE_REQUESTS=8
+HPS_MAX_CONCURRENT_NON_INFERENCE_REQUESTS=32
+HPS_INFERENCE_TIMEOUT=300
+HPS_UVICORN_WORKERS=2
+```
+
+### Worker 进程数
+
+每个 Uvicorn Worker 是独立的进程，有自己的事件循环：
+
+- **1 个 Worker**：简单，但受限于单进程
+- **4 个 Worker**：适合大多数场景
+- **8+ 个 Worker**：适用于高并发、大量小请求的场景
+
+### Triton 动态批处理
+
+Triton 自动将请求批处理以提高推理设备利用率。最大批处理大小通过模型配置文件中的 `max_batch_size` 参数控制（默认：8），配置文件位于模型仓库目录下的 `config.pbtxt`（如 `model_repo/layout-parsing/config.pbtxt`）。
+
+### Triton 实例数
+
+每个 Triton 模型的并行推理实例数通过 `config.pbtxt` 中的 `instance_group` 配置（默认：1）。增加实例数可以提高并行处理能力，但会占用更多设备资源。
+
+```
+# model_repo/layout-parsing/config.pbtxt
+instance_group [
+  {
+      count: 1       # 实例数，增大可提高并行度
+      kind: KIND_GPU
+      gpus: [ 0 ]
+  }
+]
+```
+
+实例数与动态批处理之间存在权衡：
+
+- **单实例（`count: 1`）**：动态批处理会将多个请求合并为一个批次并行执行，但同批次的请求需等待最慢的那个完成后才能一起返回，可能导致部分请求的时延升高。同时，单实例同一时刻只能处理一个批次，当前批次未完成时后续请求只能排队等待。适合显存有限或请求耗时较均匀的场景
+- **多实例（`count: 2+`）**：多个实例可以同时各自处理不同的批次，能够同时处理更多请求，减少排队等待时间，单个请求的时延也会有所改善。但需注意，同一实例内的批次仍然遵循动态批处理的行为（批内请求一起开始、一起结束）。每增加一个实例会额外占用一份版面分析模型的显存，同时也会增加对 VLM 推理服务的负载以及内存和 CPU 的使用，需根据推理设备的资源情况酌情设置
+
+非推理模型（如 `restructure-pages`）运行在 CPU 上，可根据 CPU 核数适当增加实例数。
+
+## 故障排查与解决
+
+### 服务无法启动
+
+查看各服务的日志以定位问题：
+
+```bash
+docker compose logs paddleocr-vl-api
+docker compose logs paddleocr-vl-pipeline
+docker compose logs paddleocr-vlm-server
+```
+
+常见原因包括端口被占用、推理设备不可用或镜像拉取失败。
+
+### 超时错误
+
+- 增加 `HPS_INFERENCE_TIMEOUT`（针对复杂文档）
+- 如果推理设备过载，减少 `HPS_MAX_CONCURRENT_INFERENCE_REQUESTS`
+
+### 内存/显存不足
+
+- 减少 `HPS_MAX_CONCURRENT_INFERENCE_REQUESTS`
+- 确保每个推理设备只运行一个服务
+- 检查 compose.yaml 中的 `shm_size`（默认：4GB）
diff --git a/deploy/paddleocr_vl_docker/hps/README_en.md b/deploy/paddleocr_vl_docker/hps/README_en.md
new file mode 100644
index 0000000..f628830
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/hps/README_en.md
@@ -0,0 +1,234 @@
+# PaddleOCR-VL High-Performance Serving
+
+[简体中文](README.md)
+
+This directory provides a high-performance service deployment solution for the **PaddleOCR-VL series** with concurrent request processing support. It applies to VL pipeline releases such as `PaddleOCR-VL`, `PaddleOCR-VL-1.5`, and `PaddleOCR-VL-1.6`.
+
+> This solution currently only supports NVIDIA GPUs. Support for other inference devices is still being developed.
+
+## Architecture
+
+```
+Client → FastAPI Gateway → Triton Server → vLLM Server
+```
+
+| Component       | Description                                                                                                                 |
+| -----------------| -----------------------------------------------------------------------------------------------------------------------------|
+| FastAPI Gateway | Unified access point, simplified client calls, concurrency control                                                          |
+| Triton Server   | Layout analysis model (such as PP-DocLayoutV3) and pipeline orchestration; model management, dynamic batching, inference scheduling |
+| vLLM Server     | VLM, continuous batching inference                                                                                            |
+
+**Triton Models:**
+
+| Model | Device | Description |
+|-------|--------|-------------|
+| `layout-parsing` | Inference device (e.g., GPU) | Layout parsing inference |
+| `restructure-pages` | CPU | Multi-page result post-processing (cross-page table merging, title level reassignment) |
+
+## Requirements
+
+- x64 CPU
+- NVIDIA GPU, Compute Capability >= 8.0 and < 10.0
+- NVIDIA driver supporting CUDA 12.6
+- Docker >= 19.03
+- Docker Compose >= 2.0
+
+## Quick Start
+
+1. Clone the PaddleOCR repository and navigate to this directory:
+
+```bash
+git clone https://github.com/PaddlePaddle/PaddleOCR.git
+cd PaddleOCR/deploy/paddleocr_vl_docker/hps
+```
+
+2. Prepare necessary files:
+
+```bash
+cp .env.example .env
+# Edit HPS_PIPELINE_NAME in .env if needed
+bash prepare.sh
+```
+
+`prepare.sh` downloads the high-stability serving SDK for the selected PaddleOCR-VL release and writes the Triton pipeline config.
+
+3. Start the services:
+
+```bash
+docker compose up
+```
+
+The above command will start 3 containers in sequence:
+
+| Service | Description | Port |
+|---------|-------------|------|
+| `paddleocr-vl-api` | FastAPI gateway (external entry point) | 8080 |
+| `paddleocr-vl-pipeline` | Triton inference server running the pipeline | 8000 (internal) |
+| `paddleocr-vlm-server` | vLLM-based VLM inference service | 8080 (internal) |
+
+> The first startup will automatically download and build images, which takes longer. Subsequent startups will use local images and start faster.
+
+## Configuration
+
+### Environment Variables
+
+Copy `.env.example` to `.env` and modify as needed.
+
+```bash
+cp .env.example .env
+```
+
+You can also set these as environment variables directly instead of using the `.env` file, e.g.:
+
+```bash
+export HPS_MAX_CONCURRENT_INFERENCE_REQUESTS=8
+```
+
+#### Pipeline and SDK Configuration
+
+Use the following variables to choose a release from the PaddleOCR-VL series. After changing them, rerun `prepare.sh` and rebuild the images:
+
+This solution reuses the PaddleX [High-Stability Serving](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_deploy/serving.html#2-high-stability-serving) SDK as the base Triton model repository and client dependency, and adds a PaddleOCR-VL-specific FastAPI gateway and vLLM service orchestration on top.
+
+| Variable | Default | Description |
+|----------|---------|-------------|
+| `HPS_PIPELINE_NAME` | `PaddleOCR-VL-1.6` | Pipeline name |
+| `HPS_PADDLEX_VERSION` | `3.6` | PaddleX version (major.minor only, e.g. `3.6`). Drives both the Triton base image tag (`paddlex${HPS_PADDLEX_VERSION}-gpu`) and the SDK release directory (`v${HPS_PADDLEX_VERSION}`), keeping them in sync |
+| `HPS_SDK_DIR` | `paddlex_hps_PaddleOCR-VL-1.6_sdk` | Extracted SDK directory, following `paddlex_hps_${HPS_PIPELINE_NAME}_sdk` |
+
+Common examples:
+
+| Target release | `HPS_PIPELINE_NAME` | `HPS_SDK_DIR` |
+|----------------|-----------------|---------------|
+| PaddleOCR-VL-1.6 | `PaddleOCR-VL-1.6` | `paddlex_hps_PaddleOCR-VL-1.6_sdk` |
+| PaddleOCR-VL-1.5 | `PaddleOCR-VL-1.5` | `paddlex_hps_PaddleOCR-VL-1.5_sdk` |
+| PaddleOCR-VL (v1) | `PaddleOCR-VL` | `paddlex_hps_PaddleOCR-VL_sdk` |
+
+#### Gateway and Device
+
+| Variable | Default | Description |
+|----------|---------|-------------|
+| `HPS_MAX_CONCURRENT_INFERENCE_REQUESTS` | 16 | Max concurrent inference requests (layout parsing) |
+| `HPS_MAX_CONCURRENT_NON_INFERENCE_REQUESTS` | 64 | Max concurrent non-inference requests (page restructuring) |
+| `HPS_INFERENCE_TIMEOUT` | 600 | Request timeout in seconds |
+| `HPS_HEALTH_CHECK_TIMEOUT` | 5 | Health check timeout in seconds |
+| `HPS_VLM_URL` | http://paddleocr-vlm-server:8080 | VLM server URL |
+| `HPS_LOG_LEVEL` | INFO | Log level (DEBUG, INFO, WARNING, ERROR) |
+| `HPS_FILTER_HEALTH_ACCESS_LOG` | true | Whether to filter health check access logs |
+| `HPS_UVICORN_WORKERS` | 4 | Number of gateway worker processes |
+| `HPS_DEVICE_ID` | 0 | Inference device ID to use |
+
+### Pipeline Configuration
+
+To adjust pipeline configurations (such as model path, batch size, deployment device, etc.), please refer to the Pipeline Configuration section in the [PaddleOCR-VL Usage Tutorial](https://github.com/PaddlePaddle/PaddleOCR/blob/main/docs/version3.x/pipeline_usage/PaddleOCR-VL.en.md).
+
+## API Usage
+
+### Document Parsing
+
+Please refer to the Client-Side Invocation section in the [PaddleOCR-VL Usage Tutorial](https://github.com/PaddlePaddle/PaddleOCR/blob/main/docs/version3.x/pipeline_usage/PaddleOCR-VL.en.md).
+
+The service accepts PDF or image files (including TIFF; multi-page TIFF is processed page by page—use `fileType=1`).
+
+### Health Checks
+
+```bash
+# Liveness check
+curl http://localhost:8080/health
+
+# Readiness check (verifies Triton and VLM services are ready to process requests)
+curl http://localhost:8080/health/ready
+```
+
+## Performance Tuning
+
+### Concurrency Settings
+
+The gateway controls concurrency for inference and non-inference operations independently:
+
+- **`HPS_MAX_CONCURRENT_INFERENCE_REQUESTS`** (default 16): Controls concurrency for inference operations such as `layout-parsing` (layout parsing)
+  - Too low (4): Underutilized inference device, requests queue unnecessarily
+  - Too high (64): May overload Triton, causing OOM or timeouts
+  - Default value of 16 allows enough requests to queue for the next batch while the current batch is being processed
+  - If inference device resources are limited, consider lowering this value
+- **`HPS_MAX_CONCURRENT_NON_INFERENCE_REQUESTS`** (default 64): Controls concurrency for non-inference operations such as `restructure-pages` (page restructuring)
+  - Non-inference operations do not consume inference device resources and can be set to a higher concurrency level
+  - Adjust based on CPU cores and available memory
+
+**High-throughput configuration example:**
+
+```bash
+# .env
+HPS_MAX_CONCURRENT_INFERENCE_REQUESTS=32
+HPS_MAX_CONCURRENT_NON_INFERENCE_REQUESTS=128
+HPS_UVICORN_WORKERS=8
+```
+
+**Low-latency configuration example:**
+
+```bash
+# .env
+HPS_MAX_CONCURRENT_INFERENCE_REQUESTS=8
+HPS_MAX_CONCURRENT_NON_INFERENCE_REQUESTS=32
+HPS_INFERENCE_TIMEOUT=300
+HPS_UVICORN_WORKERS=2
+```
+
+### Worker Processes
+
+Each Uvicorn worker is an independent process with its own event loop:
+
+- **1 worker**: Simple, but limited to a single process
+- **4 workers**: Suitable for most scenarios
+- **8+ workers**: Suitable for high-concurrency scenarios with many small requests
+
+### Triton Dynamic Batching
+
+Triton automatically batches requests to improve inference device utilization. The maximum batch size is controlled by the `max_batch_size` parameter in the model configuration file (default: 8), located at `config.pbtxt` under each model directory in the model repository (e.g., `model_repo/layout-parsing/config.pbtxt`).
+
+### Triton Instance Count
+
+The number of parallel inference instances for each Triton model is configured via the `instance_group` section in `config.pbtxt` (default: 1). Increasing the instance count improves parallelism but consumes more device resources.
+
+```
+# model_repo/layout-parsing/config.pbtxt
+instance_group [
+  {
+      count: 1       # Number of instances; increase for higher parallelism
+      kind: KIND_GPU
+      gpus: [ 0 ]
+  }
+]
+```
+
+There is a trade-off between instance count and dynamic batching:
+
+- **Single instance (`count: 1`)**: Dynamic batching combines multiple requests into one batch for parallel execution, but all requests in the same batch must wait for the slowest one to finish before results are returned, which may increase latency for faster requests. Additionally, a single instance can only process one batch at a time — subsequent requests must queue until the current batch completes. Best suited for scenarios with limited GPU memory or uniform request processing times
+- **Multiple instances (`count: 2+`)**: Multiple instances can process different batches simultaneously, allowing more requests to be handled concurrently. This reduces queuing time and improves latency for individual requests. Note that within each instance, dynamic batching behavior still applies (requests in the same batch start and finish together). Each additional instance consumes an extra copy of the layout analysis model's GPU memory, increases the load on the VLM inference service, and uses more CPU and system memory. Adjust based on the available resources of your inference device
+
+Non-inference models (e.g., `restructure-pages`) run on CPU and can have their instance count increased based on available CPU cores.
+
+## Troubleshooting and Resolution
+
+### Service Fails to Start
+
+Check the logs for each service to identify the issue:
+
+```bash
+docker compose logs paddleocr-vl-api
+docker compose logs paddleocr-vl-pipeline
+docker compose logs paddleocr-vlm-server
+```
+
+Common causes include port conflicts, unavailable inference devices, or image pull failures.
+
+### Timeout Errors
+
+- Increase `HPS_INFERENCE_TIMEOUT` (for complex documents)
+- If the inference device is overloaded, reduce `HPS_MAX_CONCURRENT_INFERENCE_REQUESTS`
+
+### Out of Memory
+
+- Reduce `HPS_MAX_CONCURRENT_INFERENCE_REQUESTS`
+- Ensure only one service runs per inference device
+- Check `shm_size` in compose.yaml (default: 4GB)
diff --git a/deploy/paddleocr_vl_docker/hps/compose.yaml b/deploy/paddleocr_vl_docker/hps/compose.yaml
new file mode 100644
index 0000000..8f6ed5f
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/hps/compose.yaml
@@ -0,0 +1,77 @@
+services:
+  paddleocr-vl-api:
+    build:
+      context: .
+      dockerfile: gateway.Dockerfile
+      args:
+        HPS_SDK_DIR: ${HPS_SDK_DIR:-paddlex_hps_PaddleOCR-VL-1.6_sdk}
+    container_name: paddleocr-vl-api
+    ports:
+      - 8080:8080
+    depends_on:
+      paddleocr-vl-pipeline:
+        condition: service_healthy
+    environment:
+      - HPS_TRITON_URL=paddleocr-vl-pipeline:8001
+      - HPS_VLM_URL=${HPS_VLM_URL:-http://paddleocr-vlm-server:8080}
+      - HPS_MAX_CONCURRENT_INFERENCE_REQUESTS=${HPS_MAX_CONCURRENT_INFERENCE_REQUESTS:-16}
+      - HPS_MAX_CONCURRENT_NON_INFERENCE_REQUESTS=${HPS_MAX_CONCURRENT_NON_INFERENCE_REQUESTS:-64}
+      - HPS_INFERENCE_TIMEOUT=${HPS_INFERENCE_TIMEOUT:-600}
+      - HPS_LOG_LEVEL=${HPS_LOG_LEVEL:-INFO}
+      - HPS_UVICORN_WORKERS=${HPS_UVICORN_WORKERS:-4}
+    restart: unless-stopped
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+      interval: 10s
+      timeout: 5s
+      retries: 3
+
+  paddleocr-vl-pipeline:
+    build:
+      context: .
+      dockerfile: pipeline.Dockerfile
+      args:
+        BASE_IMAGE: ${HPS_TRITON_BASE_IMAGE:-ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlex/hps:paddlex${HPS_PADDLEX_VERSION:-3.6}-gpu}
+        DEVICE_TYPE: ${HPS_DEVICE_TYPE:-gpu}
+        HPS_SDK_DIR: ${HPS_SDK_DIR:-paddlex_hps_PaddleOCR-VL-1.6_sdk}
+    container_name: paddleocr-vl-pipeline
+    depends_on:
+      paddleocr-vlm-server:
+        condition: service_healthy
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              device_ids: ["${HPS_DEVICE_ID:-0}"]
+              capabilities: [gpu]
+    shm_size: 4gb
+    restart: unless-stopped
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8000/v2/health/ready || exit 1"]
+      interval: 10s
+      timeout: 5s
+      retries: 3
+      start_period: 60s
+
+  paddleocr-vlm-server:
+    image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu
+    container_name: paddleocr-vlm-server
+    volumes:
+      - ./${HPS_SDK_DIR:-paddlex_hps_PaddleOCR-VL-1.6_sdk}/server/pipeline_config.yaml:/config/pipeline_config.yaml:ro
+      - ./genai_server_entrypoint.sh:/entrypoint.sh:ro
+    command: ["/bin/sh", "/entrypoint.sh"]
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              device_ids: ["${HPS_DEVICE_ID:-0}"]
+              capabilities: [gpu]
+    restart: unless-stopped
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8080/health || exit 1"]
+      interval: 10s
+      timeout: 5s
+      retries: 3
+      start_period: 300s
diff --git a/deploy/paddleocr_vl_docker/hps/gateway.Dockerfile b/deploy/paddleocr_vl_docker/hps/gateway.Dockerfile
new file mode 100644
index 0000000..a80baad
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/hps/gateway.Dockerfile
@@ -0,0 +1,33 @@
+FROM python:3.10-slim
+
+# Install system dependencies
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends curl libgl1 libglib2.0-0 \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /app
+
+# Copy gateway application
+COPY gateway .
+
+ARG HPS_SDK_DIR=paddlex_hps_PaddleOCR-VL-1.6_sdk
+
+# Install Python dependencies
+RUN --mount=type=bind,source=${HPS_SDK_DIR}/client,target=/tmp/sdk \
+    python -m pip install --no-cache-dir -r requirements.txt \
+    && python -m pip install --no-cache-dir -r /tmp/sdk/requirements.txt \
+    && python -m pip install --no-cache-dir /tmp/sdk/paddlex_hps_client-*.whl
+
+# Configuration via environment variables
+ENV HPS_TRITON_URL=paddleocr-vl-pipeline:8001
+ENV HPS_MAX_CONCURRENT_INFERENCE_REQUESTS=16
+ENV HPS_MAX_CONCURRENT_NON_INFERENCE_REQUESTS=64
+ENV HPS_INFERENCE_TIMEOUT=600
+ENV HPS_LOG_LEVEL=INFO
+ENV HPS_UVICORN_WORKERS=4
+
+EXPOSE 8080
+
+# Run with multiple workers for better concurrency
+# Note: Using shell form to allow environment variable expansion
+CMD uvicorn --host 0.0.0.0 --port 8080 --workers ${HPS_UVICORN_WORKERS} app:app
diff --git a/deploy/paddleocr_vl_docker/hps/gateway/app.py b/deploy/paddleocr_vl_docker/hps/gateway/app.py
new file mode 100755
index 0000000..eb834ae
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/hps/gateway/app.py
@@ -0,0 +1,418 @@
+#!/usr/bin/env python
+
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import asyncio
+import json
+import logging
+import os
+import urllib.request
+from contextlib import asynccontextmanager
+from typing import Optional
+
+import fastapi
+from fastapi import Request
+from fastapi.exceptions import RequestValidationError
+from fastapi.responses import JSONResponse
+from paddlex.inference.serving.infra.models import AIStudioNoResultResponse
+from paddlex.inference.serving.infra.utils import generate_log_id
+from paddlex_hps_client import triton_request_async
+from tritonclient.grpc import aio as triton_grpc_aio
+
+TRITON_URL = os.getenv("HPS_TRITON_URL", "paddleocr-vl-pipeline:8001")
+MAX_CONCURRENT_INFERENCE_REQUESTS = int(
+    os.getenv("HPS_MAX_CONCURRENT_INFERENCE_REQUESTS", "16")
+)
+MAX_CONCURRENT_NON_INFERENCE_REQUESTS = int(
+    os.getenv("HPS_MAX_CONCURRENT_NON_INFERENCE_REQUESTS", "64")
+)
+INFERENCE_TIMEOUT = int(os.getenv("HPS_INFERENCE_TIMEOUT", "600"))
+LOG_LEVEL = os.getenv("HPS_LOG_LEVEL", "INFO")
+HEALTH_CHECK_TIMEOUT = int(os.getenv("HPS_HEALTH_CHECK_TIMEOUT", "5"))
+FILTER_HEALTH_ACCESS_LOG = os.getenv(
+    "HPS_FILTER_HEALTH_ACCESS_LOG", "true"
+).lower() in (
+    "true",
+    "1",
+    "yes",
+)
+
+VLM_URL = os.getenv("HPS_VLM_URL", "http://paddleocr-vlm-server:8080")
+
+TRITON_MODEL_LAYOUT_PARSING = "layout-parsing"
+TRITON_MODEL_RESTRUCTURE_PAGES = "restructure-pages"
+TRITON_MODELS = (TRITON_MODEL_LAYOUT_PARSING, TRITON_MODEL_RESTRUCTURE_PAGES)
+
+
+logger = logging.getLogger(__name__)
+
+
+def _configure_logger(logger: logging.Logger) -> None:
+    level = getattr(logging, LOG_LEVEL.upper(), logging.INFO)
+    logger.setLevel(level)
+    handler = logging.StreamHandler()
+    handler.setLevel(level)
+    formatter = logging.Formatter(
+        "%(asctime)s - %(name)s - %(funcName)s - %(levelname)s - %(message)s"
+    )
+    handler.setFormatter(formatter)
+    logger.addHandler(handler)
+
+
+_configure_logger(logger)
+
+
+def _create_aistudio_output_without_result(
+    error_code: int, error_msg: str, *, log_id: Optional[str] = None
+) -> dict:
+    """Create a standardized error response in AIStudio format."""
+    resp = AIStudioNoResultResponse(
+        logId=log_id if log_id is not None else generate_log_id(),
+        errorCode=error_code,
+        errorMsg=error_msg,
+    )
+    return resp.model_dump()
+
+
+@asynccontextmanager
+async def _lifespan(app: fastapi.FastAPI):
+    """
+    Manage application lifecycle:
+    - Initialize Triton client and semaphores on startup
+    - Clean up resources on shutdown
+    """
+    logger.info("Initializing gateway...")
+    logger.info("Triton URL: %s", TRITON_URL)
+    logger.info(
+        "Max concurrent inference requests: %d", MAX_CONCURRENT_INFERENCE_REQUESTS
+    )
+    logger.info(
+        "Max concurrent non-inference requests: %d",
+        MAX_CONCURRENT_NON_INFERENCE_REQUESTS,
+    )
+    logger.info("Inference timeout: %ds", INFERENCE_TIMEOUT)
+
+    # Initialize async Triton client
+    app.state.triton_client = triton_grpc_aio.InferenceServerClient(
+        url=TRITON_URL,
+        keepalive_options=triton_grpc_aio.KeepAliveOptions(
+            keepalive_timeout_ms=INFERENCE_TIMEOUT * 1000,
+        ),
+    )
+
+    # Separate semaphores for inference and non-inference operations
+    app.state.inference_semaphore = asyncio.Semaphore(MAX_CONCURRENT_INFERENCE_REQUESTS)
+    app.state.non_inference_semaphore = asyncio.Semaphore(
+        MAX_CONCURRENT_NON_INFERENCE_REQUESTS
+    )
+
+    logger.info("Gateway initialized successfully")
+
+    yield
+
+    # Cleanup
+    logger.info("Shutting down gateway...")
+    await app.state.triton_client.close()
+    logger.info("Gateway shutdown complete")
+
+
+app = fastapi.FastAPI(
+    title="PaddleOCR-VL HPS Gateway",
+    description="High Performance Server Gateway for PaddleOCR-VL",
+    version="1.0.0",
+    lifespan=_lifespan,
+)
+
+
+@app.get("/health", operation_id="checkHealth")
+async def health():
+    """Liveness check - returns healthy if the gateway process is running."""
+    return _create_aistudio_output_without_result(0, "Healthy")
+
+
+async def _check_vlm_ready() -> bool:
+    """Check if the VLM server is ready by querying its health endpoint."""
+
+    def _do_check():
+        req = urllib.request.Request(f"{VLM_URL}/health")
+        try:
+            with urllib.request.urlopen(req, timeout=HEALTH_CHECK_TIMEOUT) as resp:
+                return resp.status == 200
+        except Exception:
+            return False
+
+    return await asyncio.to_thread(_do_check)
+
+
+@app.get("/health/ready", operation_id="checkReady")
+async def ready(request: Request):
+    """Readiness check - verifies Triton server, models, and VLM server."""
+    try:
+        client = request.app.state.triton_client
+
+        # Check Triton server readiness with timeout
+        is_server_ready = await asyncio.wait_for(
+            client.is_server_ready(),
+            timeout=HEALTH_CHECK_TIMEOUT,
+        )
+        if not is_server_ready:
+            return JSONResponse(
+                status_code=503,
+                content=_create_aistudio_output_without_result(
+                    503, "Triton server not ready"
+                ),
+            )
+
+        # Check if required models are ready
+        for model_name in TRITON_MODELS:
+            is_model_ready = await asyncio.wait_for(
+                client.is_model_ready(model_name),
+                timeout=HEALTH_CHECK_TIMEOUT,
+            )
+            if not is_model_ready:
+                return JSONResponse(
+                    status_code=503,
+                    content=_create_aistudio_output_without_result(
+                        503, f"Model '{model_name}' not ready"
+                    ),
+                )
+
+        # Check VLM server readiness
+        vlm_ready = await _check_vlm_ready()
+        if not vlm_ready:
+            return JSONResponse(
+                status_code=503,
+                content=_create_aistudio_output_without_result(
+                    503, "VLM server not ready"
+                ),
+            )
+
+        return _create_aistudio_output_without_result(0, "Ready")
+    except asyncio.TimeoutError:
+        logger.error("Health check timed out after %ds", HEALTH_CHECK_TIMEOUT)
+        return JSONResponse(
+            status_code=503,
+            content=_create_aistudio_output_without_result(
+                503, "Health check timed out"
+            ),
+        )
+    except Exception as e:
+        logger.error("Health check failed: %s", e)
+        return JSONResponse(
+            status_code=503,
+            content=_create_aistudio_output_without_result(
+                503, f"Service unavailable: {e}"
+            ),
+        )
+
+
+async def _process_triton_request(
+    request: Request,
+    body: dict,
+    model_name: str,
+    semaphore: asyncio.Semaphore,
+) -> JSONResponse:
+    """Process a request through Triton inference server."""
+    request_log_id = body.get("logId", generate_log_id())
+    logger.info(
+        "Processing %r request %s",
+        model_name,
+        request_log_id,
+    )
+
+    if "logId" in body:
+        logger.debug(
+            "Using external logId for %r request: %s",
+            model_name,
+            request_log_id,
+        )
+    body["logId"] = request_log_id
+
+    client = request.app.state.triton_client
+
+    try:
+        async with semaphore:
+            output = await triton_request_async(
+                client,
+                model_name,
+                body,
+                timeout=INFERENCE_TIMEOUT,
+            )
+    except asyncio.TimeoutError:
+        logger.warning(
+            "Timeout processing %r request %s",
+            model_name,
+            request_log_id,
+        )
+        return JSONResponse(
+            status_code=504,
+            content=_create_aistudio_output_without_result(
+                504, "Gateway timeout", log_id=request_log_id
+            ),
+        )
+    except triton_grpc_aio.InferenceServerException as e:
+        if "Deadline Exceeded" in str(e):
+            logger.warning(
+                "Triton timeout for %r request %s",
+                model_name,
+                request_log_id,
+            )
+            return JSONResponse(
+                status_code=504,
+                content=_create_aistudio_output_without_result(
+                    504, "Gateway timeout", log_id=request_log_id
+                ),
+            )
+        logger.error(
+            "Triton error for %r request %s: %s",
+            model_name,
+            request_log_id,
+            e,
+        )
+        return JSONResponse(
+            status_code=500,
+            content=_create_aistudio_output_without_result(
+                500, "Internal server error", log_id=request_log_id
+            ),
+        )
+    except Exception:
+        logger.exception(
+            "Unexpected error for %r request %s",
+            model_name,
+            request_log_id,
+        )
+        return JSONResponse(
+            status_code=500,
+            content=_create_aistudio_output_without_result(
+                500, "Internal server error", log_id=request_log_id
+            ),
+        )
+
+    if output.get("errorCode", 0) != 0:
+        error_code = output.get("errorCode", 500)
+        error_msg = output.get("errorMsg", "Unknown error")
+        logger.warning(
+            "Triton returned error for %r request %s: %s",
+            model_name,
+            request_log_id,
+            error_msg,
+        )
+        return JSONResponse(
+            status_code=error_code,
+            content=_create_aistudio_output_without_result(
+                error_code, error_msg, log_id=request_log_id
+            ),
+        )
+
+    logger.info(
+        "Completed %r request %s",
+        model_name,
+        request_log_id,
+    )
+    return JSONResponse(status_code=200, content=output)
+
+
+@app.post(
+    "/layout-parsing",
+    operation_id="infer",
+    summary=f"Invoke {TRITON_MODEL_LAYOUT_PARSING} model",
+    response_class=JSONResponse,
+)
+async def _handle_infer(request: Request, body: dict):
+    """Handle layout-parsing inference request."""
+    return await _process_triton_request(
+        request,
+        body,
+        TRITON_MODEL_LAYOUT_PARSING,
+        request.app.state.inference_semaphore,
+    )
+
+
+@app.post(
+    "/restructure-pages",
+    operation_id="restructurePages",
+    summary=f"Invoke {TRITON_MODEL_RESTRUCTURE_PAGES} model",
+    response_class=JSONResponse,
+)
+async def _handle_restructure_pages(request: Request, body: dict):
+    """Handle restructure-pages request (non-inference)."""
+    return await _process_triton_request(
+        request,
+        body,
+        TRITON_MODEL_RESTRUCTURE_PAGES,
+        request.app.state.non_inference_semaphore,
+    )
+
+
+@app.exception_handler(json.JSONDecodeError)
+async def _json_decode_exception_handler(request: Request, exc: json.JSONDecodeError):
+    """Handle invalid JSON in request body."""
+    logger.warning("Invalid JSON for %s: %s", request.url.path, exc.msg)
+    return JSONResponse(
+        status_code=400,
+        content=_create_aistudio_output_without_result(400, f"Invalid JSON: {exc.msg}"),
+    )
+
+
+@app.exception_handler(RequestValidationError)
+async def _validation_exception_handler(request: Request, exc: RequestValidationError):
+    """Handle request validation errors."""
+    error_details = exc.errors()
+    # Format error messages for readability
+    error_messages = []
+    for error in error_details:
+        loc = ".".join(str(x) for x in error.get("loc", []))
+        msg = error.get("msg", "Unknown error")
+        error_messages.append(f"{loc}: {msg}" if loc else msg)
+    error_msg = "; ".join(error_messages)
+
+    logger.warning("Validation error for %s: %s", request.url.path, error_msg)
+    return JSONResponse(
+        status_code=422,
+        content=_create_aistudio_output_without_result(422, error_msg),
+    )
+
+
+@app.exception_handler(asyncio.TimeoutError)
+async def _timeout_exception_handler(request: Request, exc: asyncio.TimeoutError):
+    """Handle timeout errors."""
+    logger.warning("Request timed out: %s", request.url.path)
+    return JSONResponse(
+        status_code=504,
+        content=_create_aistudio_output_without_result(504, "Gateway timeout"),
+    )
+
+
+@app.exception_handler(Exception)
+async def _general_exception_handler(request: Request, exc: Exception):
+    """Handle unexpected errors."""
+    logger.exception("Unhandled exception for %s", request.url.path)
+    return JSONResponse(
+        status_code=500,
+        content=_create_aistudio_output_without_result(500, "Internal server error"),
+    )
+
+
+class _HealthEndpointFilter(logging.Filter):
+    """Filter out health check endpoints from access logs."""
+
+    def filter(self, record: logging.LogRecord) -> bool:
+        message = record.getMessage()
+        return "/health" not in message
+
+
+# Apply filter to reduce log noise from health checks
+if FILTER_HEALTH_ACCESS_LOG:
+    logging.getLogger("uvicorn.access").addFilter(_HealthEndpointFilter())
diff --git a/deploy/paddleocr_vl_docker/hps/gateway/requirements.txt b/deploy/paddleocr_vl_docker/hps/gateway/requirements.txt
new file mode 100644
index 0000000..b6cb284
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/hps/gateway/requirements.txt
@@ -0,0 +1,3 @@
+fastapi == 0.123.6
+paddlex[serving] >= 3.4.0
+uvicorn == 0.35.0
diff --git a/deploy/paddleocr_vl_docker/hps/genai_server_entrypoint.sh b/deploy/paddleocr_vl_docker/hps/genai_server_entrypoint.sh
new file mode 100755
index 0000000..ba8f287
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/hps/genai_server_entrypoint.sh
@@ -0,0 +1,23 @@
+#!/usr/bin/env sh
+
+set -eu
+
+CONFIG="${PIPELINE_CONFIG:-/config/pipeline_config.yaml}"
+
+VLM_NAME=$(
+    grep -A5 'module_name: vl_recognition' "$CONFIG" \
+        | grep 'model_name:' \
+        | head -1 \
+        | awk '{print $2}'
+)
+
+if [ -z "$VLM_NAME" ]; then
+    echo "Failed to read VLM name from ${CONFIG}" >&2
+    exit 1
+fi
+
+exec paddleocr genai_server \
+    --model_name "$VLM_NAME" \
+    --host 0.0.0.0 \
+    --port 8080 \
+    --backend vllm
diff --git a/deploy/paddleocr_vl_docker/hps/pipeline.Dockerfile b/deploy/paddleocr_vl_docker/hps/pipeline.Dockerfile
new file mode 100644
index 0000000..7a1dad3
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/hps/pipeline.Dockerfile
@@ -0,0 +1,15 @@
+ARG BASE_IMAGE=ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlex/hps:paddlex3.6-gpu
+ARG DEVICE_TYPE=gpu
+
+FROM ${BASE_IMAGE}
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends curl \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /app
+ARG HPS_SDK_DIR=paddlex_hps_PaddleOCR-VL-1.6_sdk
+COPY ${HPS_SDK_DIR}/server .
+
+ARG DEVICE_TYPE
+ENV PADDLEX_HPS_DEVICE_TYPE=${DEVICE_TYPE}
+CMD ["/bin/bash", "server.sh"]
diff --git a/deploy/paddleocr_vl_docker/hps/prepare.sh b/deploy/paddleocr_vl_docker/hps/prepare.sh
new file mode 100755
index 0000000..ec0e2cb
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/hps/prepare.sh
@@ -0,0 +1,68 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+if [ -f .env ]; then
+    set -a
+    # shellcheck disable=SC1091
+    source .env
+    set +a
+fi
+
+HPS_PIPELINE_NAME="${HPS_PIPELINE_NAME:-PaddleOCR-VL-1.6}"
+HPS_PADDLEX_VERSION="${HPS_PADDLEX_VERSION:-3.6}"
+HPS_SDK_VERSION="${HPS_SDK_VERSION:-v${HPS_PADDLEX_VERSION}}"
+HPS_SDK_DIR="${HPS_SDK_DIR:-paddlex_hps_${HPS_PIPELINE_NAME}_sdk}"
+HPS_VLM_URL="${HPS_VLM_URL:-http://paddleocr-vlm-server:8080}"
+
+SDK_ARCHIVE="${HPS_SDK_DIR}.tar.gz"
+SDK_URL="https://paddle-model-ecology.bj.bcebos.com/paddlex/PaddleX3.0/deploy/paddlex_hps/public/sdks/${HPS_SDK_VERSION}/${SDK_ARCHIVE}"
+PIPELINE_CONFIG="${HPS_SDK_DIR}/server/pipeline_config.yaml"
+
+_extract_vlm_name() {
+    grep -A5 'module_name: vl_recognition' "$1" \
+        | grep 'model_name:' \
+        | head -1 \
+        | awk '{print $2}'
+}
+
+_set_env_value() {
+    local key="$1"
+    local value="$2"
+    local file=".env"
+
+    if [ ! -f "$file" ]; then
+        return
+    fi
+
+    if grep -q "^${key}=" "$file"; then
+        sed -i "s|^${key}=.*|${key}=${value}|" "$file"
+    else
+        echo "${key}=${value}" >>"$file"
+    fi
+}
+
+echo "Preparing high-stability serving SDK for ${HPS_PIPELINE_NAME}"
+echo "  SDK archive: ${SDK_ARCHIVE}"
+
+wget "${SDK_URL}"
+tar -xf "${SDK_ARCHIVE}"
+
+if grep -qE '^\s*backend: native\s*$' "${PIPELINE_CONFIG}"; then
+    sed -i \
+        -e "s|^\( *\)backend: native$|\1backend: vllm-server\n\1server_url: ${HPS_VLM_URL%/}/v1|" \
+        "${PIPELINE_CONFIG}"
+fi
+
+VLM_NAME="$(_extract_vlm_name "${PIPELINE_CONFIG}")"
+if [ -z "${VLM_NAME}" ]; then
+    echo "Failed to read VLM name from ${PIPELINE_CONFIG}" >&2
+    exit 1
+fi
+echo "  VLM name: ${VLM_NAME}"
+
+_set_env_value "HPS_PIPELINE_NAME" "${HPS_PIPELINE_NAME}"
+_set_env_value "HPS_SDK_DIR" "${HPS_SDK_DIR}"
+_set_env_value "HPS_VLM_URL" "${HPS_VLM_URL}"
+
+echo "High-stability serving SDK prepared at ${HPS_SDK_DIR}/"
diff --git a/deploy/paddleocr_vl_docker/pip_version_arg.sh b/deploy/paddleocr_vl_docker/pip_version_arg.sh
new file mode 100644
index 0000000..6aea7e7
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/pip_version_arg.sh
@@ -0,0 +1,47 @@
+#!/usr/bin/env bash
+
+pip_build_arg_suffix() {
+    local v="$1"
+    [[ -z "$v" ]] && {
+        echo ""
+        return
+    }
+    if [[ "$v" == '=='* ]]; then
+        echo "$v"
+        return
+    fi
+    if [[ "$v" == '>='* ]] || [[ "$v" == '<='* ]] || [[ "$v" == '~='* ]] || [[ "$v" == '!='* ]]; then
+        echo "$v"
+        return
+    fi
+    if [[ "$v" == '<'* ]] || [[ "$v" == '>'* ]]; then
+        echo "$v"
+        return
+    fi
+    if [[ "$v" =~ ^(https?://|git\+|file://) ]]; then
+        echo " @${v}"
+        return
+    fi
+    if [[ "$v" == @* ]]; then
+        echo " @${v#@}"
+        return
+    fi
+    echo "==${v}"
+}
+
+version_tag_label() {
+    local v="$1"
+    if [[ "$v" =~ ^(https?://|git\+|file://) ]] || [[ "$v" == @* ]]; then
+        if command -v shasum >/dev/null 2>&1; then
+            printf 'u%s' "$(printf '%s' "$v" | shasum -a 256 | cut -c1-8)"
+        elif command -v sha256sum >/dev/null 2>&1; then
+            printf 'u%s' "$(printf '%s' "$v" | sha256sum | cut -c1-8)"
+        else
+            printf 'u%s' "$(printf '%s' "$v" | cksum | cut -c1-8)"
+        fi
+    else
+        local s="$v"
+        [[ "$s" == '=='* ]] && s="${s#==}"
+        echo "${s%.*}"
+    fi
+}
diff --git a/deploy/paddleocr_vl_docker/pipeline_config_fastdeploy.yaml b/deploy/paddleocr_vl_docker/pipeline_config_fastdeploy.yaml
new file mode 100644
index 0000000..d959211
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/pipeline_config_fastdeploy.yaml
@@ -0,0 +1,86 @@
+
+pipeline_name: PaddleOCR-VL-1.6
+
+batch_size: 64
+
+use_queues: True
+
+use_doc_preprocessor: False
+use_layout_detection: True
+use_chart_recognition: False
+use_seal_recognition: False
+format_block_content: False
+merge_layout_blocks: True
+markdown_ignore_labels:
+  - number
+  - footnote
+  - header
+  - header_image
+  - footer
+  - footer_image
+  - aside_text
+
+SubModules:
+  LayoutDetection:
+    module_name: layout_detection
+    model_name: PP-DocLayoutV3
+    model_dir: null
+    batch_size: 8
+    threshold: 0.3
+    layout_nms: True
+    layout_unclip_ratio: [1.0, 1.0] 
+    layout_merge_bboxes_mode: 
+      0: "union" # abstract
+      1: "union" # algorithm
+      2: "union" # aside_text
+      3: "large" # chart
+      4: "union" # content
+      5: "large" # display_formula
+      6: "large" # doc_title
+      7: "union" # figure_title
+      8: "union" # footer
+      9: "union" # footer
+      10: "union" # footnote
+      11: "union" # formula_number
+      12: "union" # header
+      13: "union" # header
+      14: "union" # image
+      15: "large" # inline_formula
+      16: "union" # number
+      17: "large" # paragraph_title
+      18: "union" # reference
+      19: "union" # reference_content
+      20: "union" # seal
+      21: "union" # table
+      22: "union" # text
+      23: "union" # text
+      24: "union" # vision_footnote
+  VLRecognition:
+    module_name: vl_recognition
+    model_name: PaddleOCR-VL-1.6-0.9B
+    model_dir: null
+    batch_size: 4096
+    genai_config:
+      backend: fastdeploy-server
+      server_url: http://paddleocr-vlm-server:8080/v1
+
+SubPipelines:
+  DocPreprocessor:
+    pipeline_name: doc_preprocessor
+    batch_size: 8
+    use_doc_orientation_classify: True
+    use_doc_unwarping: True
+    SubModules:
+      DocOrientationClassify:
+        module_name: doc_text_orientation
+        model_name: PP-LCNet_x1_0_doc_ori
+        model_dir: null
+        batch_size: 8
+      DocUnwarping:
+        module_name: image_unwarping
+        model_name: UVDoc
+        model_dir: null
+
+Serving:
+  extra:
+    max_num_input_imgs: null
diff --git a/deploy/paddleocr_vl_docker/pipeline_config_vllm.yaml b/deploy/paddleocr_vl_docker/pipeline_config_vllm.yaml
new file mode 100644
index 0000000..7306e0c
--- /dev/null
+++ b/deploy/paddleocr_vl_docker/pipeline_config_vllm.yaml
@@ -0,0 +1,86 @@
+
+pipeline_name: PaddleOCR-VL-1.6
+
+batch_size: 64
+
+use_queues: True
+
+use_doc_preprocessor: False
+use_layout_detection: True
+use_chart_recognition: False
+use_seal_recognition: False
+format_block_content: False
+merge_layout_blocks: True
+markdown_ignore_labels:
+  - number
+  - footnote
+  - header
+  - header_image
+  - footer
+  - footer_image
+  - aside_text
+
+SubModules:
+  LayoutDetection:
+    module_name: layout_detection
+    model_name: PP-DocLayoutV3
+    model_dir: null
+    batch_size: 8
+    threshold: 0.3
+    layout_nms: True
+    layout_unclip_ratio: [1.0, 1.0] 
+    layout_merge_bboxes_mode: 
+      0: "union" # abstract
+      1: "union" # algorithm
+      2: "union" # aside_text
+      3: "large" # chart
+      4: "union" # content
+      5: "large" # display_formula
+      6: "large" # doc_title
+      7: "union" # figure_title
+      8: "union" # footer
+      9: "union" # footer
+      10: "union" # footnote
+      11: "union" # formula_number
+      12: "union" # header
+      13: "union" # header
+      14: "union" # image
+      15: "large" # inline_formula
+      16: "union" # number
+      17: "large" # paragraph_title
+      18: "union" # reference
+      19: "union" # reference_content
+      20: "union" # seal
+      21: "union" # table
+      22: "union" # text
+      23: "union" # text
+      24: "union" # vision_footnote
+  VLRecognition:
+    module_name: vl_recognition
+    model_name: PaddleOCR-VL-1.6-0.9B
+    model_dir: null
+    batch_size: 4096
+    genai_config:
+      backend: vllm-server
+      server_url: http://paddleocr-vlm-server:8080/v1
+
+SubPipelines:
+  DocPreprocessor:
+    pipeline_name: doc_preprocessor
+    batch_size: 8
+    use_doc_orientation_classify: True
+    use_doc_unwarping: True
+    SubModules:
+      DocOrientationClassify:
+        module_name: doc_text_orientation
+        model_name: PP-LCNet_x1_0_doc_ori
+        model_dir: null
+        batch_size: 8
+      DocUnwarping:
+        module_name: image_unwarping
+        model_name: UVDoc
+        model_dir: null
+
+Serving:
+  extra:
+    max_num_input_imgs: null
diff --git a/deploy/ppocr-android/.gitignore b/deploy/ppocr-android/.gitignore
new file mode 100644
index 0000000..62d8c74
--- /dev/null
+++ b/deploy/ppocr-android/.gitignore
@@ -0,0 +1,93 @@
+# OS generated files
+.DS_Store
+.DS_Store?
+._*
+.Spotlight-V100
+.Trashes
+ehthumbs.db
+Thumbs.db
+
+# Built application files
+*.apk
+*.aar
+*.ap_
+*.aab
+
+# Files for the ART/Dalvik VM
+*.dex
+
+# Java class files
+*.class
+
+# Generated files
+bin/
+gen/
+out/
+release/
+
+# Gradle files
+.gradle/
+build/
+
+# Local configuration file (sdk path, etc)
+local.properties
+
+# Proguard folder generated by Eclipse
+proguard/
+
+# Log Files
+*.log
+
+# Android Studio Navigation editor temp files
+.navigation/
+
+# Android Studio captures folder
+captures/
+
+# IntelliJ
+*.iml
+.idea/
+
+# Keystore files
+# Uncomment the following lines if you do not want to check your keystore files in.
+#*.jks
+#*.keystore
+
+# External native build folder generated in Android Studio 2.2 and later
+.externalNativeBuild
+.cxx/
+
+# Google Services (e.g. APIs or Firebase)
+google-services.json
+
+# Freeline
+freeline.py
+freeline/
+freeline_project_description.json
+
+# fastlane
+fastlane/report.xml
+fastlane/Preview.html
+fastlane/screenshots
+fastlane/test_output
+fastlane/readme.md
+
+# Version control
+vcs.xml
+
+# lint
+lint/intermediates/
+lint/generated/
+lint/outputs/
+lint/tmp/
+
+# Secrets
+*.env
+secrets.properties
+
+# Kotlin
+.kotlin/
+
+# Model
+*.onnx
+*.yml
diff --git a/deploy/ppocr-android/README.md b/deploy/ppocr-android/README.md
new file mode 100644
index 0000000..254c048
--- /dev/null
+++ b/deploy/ppocr-android/README.md
@@ -0,0 +1,5 @@
+# PaddleOCR Android Demo
+
+中文 | [English](./README_en.md)
+
+请查看 [文档](../../docs/version3.x/inference_deployment/cross_platform/android_deployment.md)。
diff --git a/deploy/ppocr-android/README_en.md b/deploy/ppocr-android/README_en.md
new file mode 100644
index 0000000..6e669ff
--- /dev/null
+++ b/deploy/ppocr-android/README_en.md
@@ -0,0 +1,5 @@
+# PaddleOCR Android Demo
+
+[中文](./README.md) | English
+
+Please refer to the [documentation](../../docs/version3.x/inference_deployment/cross_platform/android_deployment.en.md).
diff --git a/deploy/ppocr-android/app/build.gradle.kts b/deploy/ppocr-android/app/build.gradle.kts
new file mode 100644
index 0000000..4d5b991
--- /dev/null
+++ b/deploy/ppocr-android/app/build.gradle.kts
@@ -0,0 +1,54 @@
+plugins {
+    alias(libs.plugins.android.application)
+    alias(libs.plugins.kotlin.android)
+    alias(libs.plugins.kotlin.compose)
+}
+
+android {
+    namespace = "com.paddle.ocr.demo"
+    compileSdk = 35
+
+    defaultConfig {
+        applicationId = "com.paddle.ocr.demo"
+        minSdk = 26
+        targetSdk = 35
+        versionCode = 1
+        versionName = "1.0.0"
+    }
+
+    buildTypes {
+        release {
+            isMinifyEnabled = true
+            proguardFiles(
+                getDefaultProguardFile("proguard-android-optimize.txt"),
+                "proguard-rules.pro"
+            )
+        }
+    }
+    compileOptions {
+        sourceCompatibility = JavaVersion.VERSION_17
+        targetCompatibility = JavaVersion.VERSION_17
+    }
+    kotlin {
+        compilerOptions {
+            jvmTarget.set(org.jetbrains.kotlin.gradle.dsl.JvmTarget.JVM_17)
+        }
+    }
+}
+
+dependencies {
+    implementation(project(":ppocr-sdk"))
+
+    implementation(platform(libs.compose.bom))
+    implementation(libs.compose.ui)
+    implementation(libs.compose.ui.graphics)
+    implementation(libs.compose.ui.tooling.preview)
+    implementation(libs.compose.material3)
+    implementation(libs.androidx.activity.compose)
+    implementation(libs.androidx.lifecycle.runtime.ktx)
+    implementation(libs.androidx.lifecycle.viewmodel.compose)
+    implementation(libs.androidx.core.ktx)
+    implementation(libs.coil.compose)
+
+    debugImplementation(libs.compose.ui.tooling)
+}
diff --git a/deploy/ppocr-android/app/src/main/AndroidManifest.xml b/deploy/ppocr-android/app/src/main/AndroidManifest.xml
new file mode 100644
index 0000000..fc8c983
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/AndroidManifest.xml
@@ -0,0 +1,27 @@
+<?xml version="1.0" encoding="utf-8"?>
+<manifest xmlns:android="http://schemas.android.com/apk/res/android"
+    xmlns:tools="http://schemas.android.com/tools">
+
+    <uses-permission android:name="android.permission.READ_MEDIA_IMAGES" />
+    <uses-permission android:name="android.permission.READ_EXTERNAL_STORAGE"
+        android:maxSdkVersion="32" />
+
+    <application
+        android:name=".OCRApplication"
+        android:allowBackup="true"
+        android:icon="@mipmap/ic_launcher"
+        android:label="PP-OCRv6 Demo"
+        android:supportsRtl="true"
+        android:theme="@style/Theme.PaddleOCR"
+        tools:targetApi="35">
+        <activity
+            android:name=".MainActivity"
+            android:exported="true"
+            android:theme="@style/Theme.PaddleOCR">
+            <intent-filter>
+                <action android:name="android.intent.action.MAIN" />
+                <category android:name="android.intent.category.LAUNCHER" />
+            </intent-filter>
+        </activity>
+    </application>
+</manifest>
diff --git a/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/MainActivity.kt b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/MainActivity.kt
new file mode 100644
index 0000000..c82dfbc
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/MainActivity.kt
@@ -0,0 +1,34 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.demo
+
+import android.os.Bundle
+import androidx.activity.ComponentActivity
+import androidx.activity.compose.setContent
+import androidx.activity.enableEdgeToEdge
+import com.paddle.ocr.demo.ui.screen.HomeScreen
+import com.paddle.ocr.demo.ui.theme.PPOCRTheme
+
+class MainActivity : ComponentActivity() {
+    override fun onCreate(savedInstanceState: Bundle?) {
+        super.onCreate(savedInstanceState)
+        enableEdgeToEdge()
+        setContent {
+            PPOCRTheme {
+                HomeScreen()
+            }
+        }
+    }
+}
diff --git a/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/OCRApplication.kt b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/OCRApplication.kt
new file mode 100644
index 0000000..22446f9
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/OCRApplication.kt
@@ -0,0 +1,109 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.demo
+
+import android.app.Application
+import android.util.Log
+import com.paddle.ocr.PaddleOCR
+import com.paddle.ocr.PaddleOCRConfig
+import com.paddle.ocr.util.OpenCVUtils
+import kotlinx.coroutines.CancellationException
+import kotlinx.coroutines.CoroutineScope
+import kotlinx.coroutines.Dispatchers
+import kotlinx.coroutines.SupervisorJob
+import kotlinx.coroutines.flow.MutableStateFlow
+import kotlinx.coroutines.flow.StateFlow
+import kotlinx.coroutines.flow.asStateFlow
+import kotlinx.coroutines.launch
+
+class OCRApplication : Application() {
+
+    sealed class ModelState {
+        data object Loading : ModelState()
+        data class Ready(val ocr: PaddleOCR) : ModelState()
+        data class Error(val message: String) : ModelState()
+    }
+
+    private val applicationScope = CoroutineScope(SupervisorJob() + Dispatchers.IO)
+    private val _modelState = MutableStateFlow<ModelState>(ModelState.Loading)
+
+    val modelState: StateFlow<ModelState> = _modelState.asStateFlow()
+
+    val ocr: PaddleOCR?
+        get() = (_modelState.value as? ModelState.Ready)?.ocr
+
+    val isModelLoaded: Boolean
+        get() = _modelState.value is ModelState.Ready
+
+    override fun onCreate() {
+        super.onCreate()
+        instance = this
+        loadModels()
+    }
+
+    fun retryLoadModels() {
+        if (_modelState.value is ModelState.Loading) return
+        loadModels()
+    }
+
+    private fun loadModels() {
+        _modelState.value = ModelState.Loading
+        applicationScope.launch {
+            try {
+                if (!OpenCVUtils.init(this@OCRApplication)) {
+                    throw IllegalStateException("Failed to initialize OpenCV native library")
+                }
+                val loadedOcr = PaddleOCR.create(
+                    context = this@OCRApplication,
+                    config = PaddleOCRConfig(
+                        recScoreThresh = 0.0f,
+                        recBatchSize = 1,
+                    ),
+                )
+                _modelState.value = ModelState.Ready(loadedOcr)
+            } catch (t: Throwable) {
+                if (t is CancellationException) throw t
+                val message = modelLoadErrorMessage(t)
+                Log.e(TAG, message, t)
+                _modelState.value = ModelState.Error(message)
+            }
+        }
+    }
+
+    private fun modelLoadErrorMessage(error: Throwable): String {
+        val details = listOfNotNull(error.message, error.cause?.message)
+            .distinct()
+            .joinToString(": ")
+        return if (details.isBlank()) {
+            "Failed to load OCR models"
+        } else {
+            "Failed to load OCR models: $details"
+        }
+    }
+
+    override fun onTerminate() {
+        super.onTerminate()
+        applicationScope.launch {
+            ocr?.release()
+        }
+    }
+
+    companion object {
+        private const val TAG = "OCRApplication"
+
+        lateinit var instance: OCRApplication
+            private set
+    }
+}
diff --git a/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/ErrorDialog.kt b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/ErrorDialog.kt
new file mode 100644
index 0000000..e7429fb
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/ErrorDialog.kt
@@ -0,0 +1,58 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.demo.ui.component
+
+import androidx.compose.foundation.layout.Column
+import androidx.compose.foundation.layout.Spacer
+import androidx.compose.foundation.layout.fillMaxWidth
+import androidx.compose.foundation.layout.height
+import androidx.compose.material3.AlertDialog
+import androidx.compose.material3.MaterialTheme
+import androidx.compose.material3.Text
+import androidx.compose.material3.TextButton
+import androidx.compose.runtime.Composable
+import androidx.compose.ui.Modifier
+import androidx.compose.ui.unit.dp
+
+@Composable
+fun ErrorDialog(
+    message: String,
+    onRetry: () -> Unit,
+    onDismiss: () -> Unit,
+) {
+    AlertDialog(
+        onDismissRequest = onDismiss,
+        title = { Text("Error") },
+        text = {
+            Column(modifier = Modifier.fillMaxWidth()) {
+                Text(
+                    text = message,
+                    style = MaterialTheme.typography.bodyMedium,
+                )
+                Spacer(modifier = Modifier.height(16.dp))
+            }
+        },
+        confirmButton = {
+            TextButton(onClick = onRetry) {
+                Text("Retry")
+            }
+        },
+        dismissButton = {
+            TextButton(onClick = onDismiss) {
+                Text("Dismiss")
+            }
+        },
+    )
+}
diff --git a/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/ImagePicker.kt b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/ImagePicker.kt
new file mode 100644
index 0000000..5058815
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/ImagePicker.kt
@@ -0,0 +1,83 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.demo.ui.component
+
+import androidx.compose.foundation.Image
+import androidx.compose.foundation.clickable
+import androidx.compose.foundation.layout.Arrangement
+import androidx.compose.foundation.layout.Column
+import androidx.compose.foundation.layout.Row
+import androidx.compose.foundation.layout.Spacer
+import androidx.compose.foundation.layout.fillMaxWidth
+import androidx.compose.foundation.layout.height
+import androidx.compose.foundation.layout.padding
+import androidx.compose.foundation.layout.size
+import androidx.compose.foundation.shape.RoundedCornerShape
+import androidx.compose.material.icons.Icons
+import androidx.compose.material.icons.filled.Add
+import androidx.compose.material3.Button
+import androidx.compose.material3.Icon
+import androidx.compose.material3.MaterialTheme
+import androidx.compose.material3.Text
+import androidx.compose.runtime.Composable
+import androidx.compose.ui.Alignment
+import androidx.compose.ui.Modifier
+import androidx.compose.ui.draw.clip
+import androidx.compose.ui.layout.ContentScale
+import androidx.compose.ui.res.painterResource
+import androidx.compose.ui.unit.dp
+
+@Composable
+fun ImagePicker(
+    onGalleryClick: () -> Unit,
+    onSampleClick: (Int) -> Unit,
+    sampleImages: List<Int>,
+    modifier: Modifier = Modifier,
+) {
+    Column(
+        modifier = modifier
+            .fillMaxWidth()
+            .padding(16.dp),
+        horizontalAlignment = Alignment.CenterHorizontally,
+    ) {
+        Button(onClick = onGalleryClick) {
+            Icon(Icons.Default.Add, contentDescription = null)
+            Text("  Select from Gallery", modifier = Modifier.padding(start = 8.dp))
+        }
+
+        if (sampleImages.isNotEmpty()) {
+            Spacer(modifier = Modifier.height(16.dp))
+            Text("Or try a sample:", style = MaterialTheme.typography.bodySmall)
+            Spacer(modifier = Modifier.height(8.dp))
+            Row(
+                modifier = Modifier.fillMaxWidth(),
+                horizontalArrangement = Arrangement.Center,
+            ) {
+                sampleImages.forEach { resId ->
+                    Image(
+                        painter = painterResource(resId),
+                        contentDescription = "Sample image",
+                        modifier = Modifier
+                            .size(64.dp)
+                            .padding(4.dp)
+                            .clip(RoundedCornerShape(8.dp))
+                            .clickable { onSampleClick(resId) },
+                        contentScale = ContentScale.Crop,
+                    )
+                }
+            }
+        }
+    }
+}
diff --git a/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/ImagePreview.kt b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/ImagePreview.kt
new file mode 100644
index 0000000..327426c
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/ImagePreview.kt
@@ -0,0 +1,135 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.demo.ui.component
+
+import android.graphics.Bitmap
+import android.graphics.Paint
+import androidx.compose.foundation.Canvas
+import androidx.compose.foundation.Image
+import androidx.compose.foundation.layout.Box
+import androidx.compose.foundation.layout.aspectRatio
+import androidx.compose.foundation.layout.fillMaxWidth
+import androidx.compose.runtime.Composable
+import androidx.compose.runtime.remember
+import androidx.compose.ui.Modifier
+import androidx.compose.ui.geometry.Offset
+import androidx.compose.ui.graphics.Color
+import androidx.compose.ui.graphics.ImageBitmap
+import androidx.compose.ui.graphics.Path
+import androidx.compose.ui.graphics.asImageBitmap
+import androidx.compose.ui.graphics.drawscope.Stroke
+import androidx.compose.ui.graphics.nativeCanvas
+import androidx.compose.ui.layout.ContentScale
+import com.paddle.ocr.model.OCRResult
+import kotlin.math.max
+
+private val BOX_COLORS = listOf(
+    Color(0xFFE53935),
+    Color(0xFF1E88E5),
+    Color(0xFF43A047),
+    Color(0xFFFB8C00),
+    Color(0xFF8E24AA),
+    Color(0xFF00ACC1),
+    Color(0xFFD81B60),
+    Color(0xFF6D4C41),
+)
+
+@Composable
+fun ImagePreview(
+    bitmap: Bitmap,
+    results: List<OCRResult>,
+    modifier: Modifier = Modifier,
+) {
+    val imageBitmap: ImageBitmap = bitmap.asImageBitmap()
+    val aspect = bitmap.width.toFloat() / bitmap.height.toFloat()
+
+    // Paints for index badges
+    val numberPaint = remember {
+        Paint().apply {
+            color = android.graphics.Color.WHITE
+            textSize = 36f
+            isAntiAlias = true
+            isFakeBoldText = true
+        }
+    }
+    val badgePaint = remember {
+        Paint().apply {
+            color = 0xCC000000.toInt()  // semi-transparent black
+            isAntiAlias = true
+            style = Paint.Style.FILL
+        }
+    }
+
+    Box(modifier = modifier) {
+        Image(
+            bitmap = imageBitmap,
+            contentDescription = "OCR Input",
+            contentScale = ContentScale.Fit,
+            modifier = Modifier
+                .fillMaxWidth()
+                .aspectRatio(aspect),
+        )
+        if (results.isNotEmpty()) {
+            Canvas(modifier = Modifier.matchParentSize()) {
+                // Canvas matches Image exactly (both have same aspectRatio),
+                // and ContentScale.Fit fills the Image composable with the bitmap
+                // (same aspect ratio → no letterboxing), so coordinates map 1:1.
+                val scaleX = size.width / bitmap.width
+                val scaleY = size.height / bitmap.height
+
+                results.forEachIndexed { index, result ->
+                    val color = BOX_COLORS[index % BOX_COLORS.size]
+                    val path = Path().apply {
+                        result.box.points.forEachIndexed { i, pt ->
+                            val x = pt.x * scaleX
+                            val y = pt.y * scaleY
+                            if (i == 0) moveTo(x, y) else lineTo(x, y)
+                        }
+                        close()
+                    }
+                    // Box fill
+                    drawPath(path, color = color.copy(alpha = 0.15f))
+                    // Box outline
+                    drawPath(path, color = color, style = Stroke(width = 3f))
+
+                    // Number badge: rounded rect background + white text
+                    val topLeft = result.box.points.minByOrNull { it.y }?.let { pt ->
+                        Offset(pt.x * scaleX, pt.y * scaleY)
+                    } ?: Offset.Zero
+
+                    val label = "${index + 1}"
+                    val textWidth = numberPaint.measureText(label)
+                    val badgeH = 44f
+                    val badgeW = textWidth + 20f
+                    val badgeX = topLeft.x
+                    val badgeY = max(0f, topLeft.y - badgeH)
+
+                    // Draw badge background
+                    drawContext.canvas.nativeCanvas.drawRoundRect(
+                        badgeX, badgeY, badgeX + badgeW, badgeY + badgeH,
+                        8f, 8f, badgePaint,
+                    )
+                    // Draw number centered in badge
+                    drawContext.canvas.nativeCanvas.drawText(
+                        label,
+                        badgeX + (badgeW - textWidth) / 2f,
+                        badgeY + badgeH / 2f + 12f,  // center vertically
+                        numberPaint,
+                    )
+                }
+            }
+        }
+    }
+}
diff --git a/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/LoadingOverlay.kt b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/LoadingOverlay.kt
new file mode 100644
index 0000000..a02c3aa
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/LoadingOverlay.kt
@@ -0,0 +1,44 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.demo.ui.component
+
+import androidx.compose.foundation.background
+import androidx.compose.foundation.layout.Box
+import androidx.compose.foundation.layout.Column
+import androidx.compose.foundation.layout.Spacer
+import androidx.compose.foundation.layout.fillMaxSize
+import androidx.compose.foundation.layout.height
+import androidx.compose.foundation.layout.size
+import androidx.compose.material3.CircularProgressIndicator
+import androidx.compose.material3.MaterialTheme
+import androidx.compose.material3.Text
+import androidx.compose.runtime.Composable
+import androidx.compose.ui.Alignment
+import androidx.compose.ui.Modifier
+import androidx.compose.ui.unit.dp
+
+@Composable
+fun LoadingOverlay(message: String = "Loading models...", modifier: Modifier = Modifier) {
+    Box(
+        modifier = modifier.fillMaxSize().background(MaterialTheme.colorScheme.background),
+        contentAlignment = Alignment.Center,
+    ) {
+        Column(horizontalAlignment = Alignment.CenterHorizontally) {
+            CircularProgressIndicator(modifier = Modifier.size(48.dp))
+            Spacer(modifier = Modifier.height(16.dp))
+            Text(text = message, style = MaterialTheme.typography.bodyLarge)
+        }
+    }
+}
diff --git a/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/ResultList.kt b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/ResultList.kt
new file mode 100644
index 0000000..4d032bc
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/ResultList.kt
@@ -0,0 +1,82 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.demo.ui.component
+
+import androidx.compose.foundation.layout.Column
+import androidx.compose.foundation.layout.Row
+import androidx.compose.foundation.layout.Spacer
+import androidx.compose.foundation.layout.fillMaxWidth
+import androidx.compose.foundation.layout.padding
+import androidx.compose.foundation.layout.width
+import androidx.compose.material3.HorizontalDivider
+import androidx.compose.material3.MaterialTheme
+import androidx.compose.material3.Text
+import androidx.compose.material3.TextButton
+import androidx.compose.runtime.Composable
+import androidx.compose.ui.Alignment
+import androidx.compose.ui.Modifier
+import androidx.compose.ui.unit.dp
+import com.paddle.ocr.model.OCRResult
+
+@Composable
+fun ResultList(
+    results: List<OCRResult>,
+    onCopyAll: () -> Unit,
+    modifier: Modifier = Modifier,
+) {
+    Column(modifier = modifier.padding(horizontal = 16.dp)) {
+        Row(
+            modifier = Modifier.fillMaxWidth(),
+            verticalAlignment = Alignment.CenterVertically,
+        ) {
+            Text(
+                text = "Results (${results.size})",
+                style = MaterialTheme.typography.titleMedium,
+                modifier = Modifier.weight(1f),
+            )
+            TextButton(onClick = onCopyAll) {
+                Text("Copy All")
+            }
+        }
+        results.forEachIndexed { index, result ->
+            Row(
+                modifier = Modifier
+                    .fillMaxWidth()
+                    .padding(vertical = 8.dp),
+                verticalAlignment = Alignment.CenterVertically,
+            ) {
+                Text(
+                    text = "${index + 1}.",
+                    style = MaterialTheme.typography.bodySmall,
+                    modifier = Modifier.width(24.dp),
+                )
+                Text(
+                    text = result.text,
+                    style = MaterialTheme.typography.bodyLarge,
+                    modifier = Modifier.weight(1f),
+                )
+                Spacer(modifier = Modifier.width(8.dp))
+                Text(
+                    text = "${"%.2f".format(result.confidence * 100)}%",
+                    style = MaterialTheme.typography.labelSmall,
+                    color = MaterialTheme.colorScheme.secondary,
+                )
+            }
+            if (index < results.lastIndex) {
+                HorizontalDivider()
+            }
+        }
+    }
+}
diff --git a/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/TimingBar.kt b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/TimingBar.kt
new file mode 100644
index 0000000..09a14b6
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/component/TimingBar.kt
@@ -0,0 +1,60 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.demo.ui.component
+
+import androidx.compose.foundation.layout.Arrangement
+import androidx.compose.foundation.layout.Row
+import androidx.compose.foundation.layout.fillMaxWidth
+import androidx.compose.foundation.layout.padding
+import androidx.compose.material3.AssistChip
+import androidx.compose.material3.MaterialTheme
+import androidx.compose.material3.Text
+import androidx.compose.runtime.Composable
+import androidx.compose.ui.Modifier
+import androidx.compose.ui.unit.dp
+
+@Composable
+fun TimingBar(
+    detectionMs: Long,
+    recognitionMs: Long,
+    totalMs: Long,
+    modifier: Modifier = Modifier,
+) {
+    Row(
+        modifier = modifier
+            .fillMaxWidth()
+            .padding(horizontal = 16.dp, vertical = 8.dp),
+        horizontalArrangement = Arrangement.spacedBy(8.dp),
+    ) {
+        AssistChip(
+            onClick = {},
+            label = {
+                Text("Det: ${detectionMs}ms", style = MaterialTheme.typography.labelSmall)
+            },
+        )
+        AssistChip(
+            onClick = {},
+            label = {
+                Text("Rec: ${recognitionMs}ms", style = MaterialTheme.typography.labelSmall)
+            },
+        )
+        AssistChip(
+            onClick = {},
+            label = {
+                Text("Total: ${totalMs}ms", style = MaterialTheme.typography.labelSmall)
+            },
+        )
+    }
+}
diff --git a/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/screen/HomeScreen.kt b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/screen/HomeScreen.kt
new file mode 100644
index 0000000..b1d0223
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/screen/HomeScreen.kt
@@ -0,0 +1,106 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.demo.ui.screen
+
+import androidx.activity.compose.rememberLauncherForActivityResult
+import androidx.activity.result.PickVisualMediaRequest
+import androidx.activity.result.contract.ActivityResultContracts
+import androidx.compose.foundation.layout.Column
+import androidx.compose.foundation.layout.Spacer
+import androidx.compose.foundation.layout.fillMaxSize
+import androidx.compose.foundation.layout.fillMaxWidth
+import androidx.compose.foundation.layout.height
+import androidx.compose.foundation.rememberScrollState
+import androidx.compose.foundation.verticalScroll
+import androidx.compose.runtime.Composable
+import androidx.compose.runtime.collectAsState
+import androidx.compose.runtime.getValue
+import androidx.compose.ui.Modifier
+import androidx.compose.ui.unit.dp
+import androidx.lifecycle.viewmodel.compose.viewModel
+import com.paddle.ocr.demo.ui.component.*
+import com.paddle.ocr.demo.ui.viewmodel.OCRViewModel
+
+@Composable
+fun HomeScreen(viewModel: OCRViewModel = viewModel()) {
+    val state by viewModel.uiState.collectAsState()
+    val timing by viewModel.timing.collectAsState()
+
+    val galleryLauncher = rememberLauncherForActivityResult(
+        contract = ActivityResultContracts.PickVisualMedia()
+    ) { uri -> uri?.let { viewModel.onImageSelected(it) } }
+
+    Column(modifier = Modifier.fillMaxSize()) {
+        when (val s = state) {
+            is OCRViewModel.UIState.Loading -> {
+                LoadingOverlay("Loading OCR models...")
+            }
+            is OCRViewModel.UIState.Ready -> {
+                Column(modifier = Modifier.verticalScroll(rememberScrollState())) {
+                    Spacer(modifier = Modifier.height(32.dp))
+                    ImagePicker(
+                        onGalleryClick = { galleryLauncher.launch(PickVisualMediaRequest(ActivityResultContracts.PickVisualMedia.ImageOnly)) },
+                        onSampleClick = { viewModel.onSampleImageClicked(it) },
+                        sampleImages = emptyList(),
+                    )
+                }
+            }
+            is OCRViewModel.UIState.Processing -> {
+                Column(modifier = Modifier.verticalScroll(rememberScrollState())) {
+                    Spacer(modifier = Modifier.height(16.dp))
+                    ImagePreview(bitmap = s.bitmap, results = emptyList())
+                    Spacer(modifier = Modifier.height(16.dp))
+                    LoadingOverlay("Processing...")
+                }
+            }
+            is OCRViewModel.UIState.Result -> {
+                Column(modifier = Modifier.verticalScroll(rememberScrollState())) {
+                    ImagePreview(
+                        bitmap = s.bitmap,
+                        results = s.result.results,
+                        modifier = Modifier.fillMaxWidth(),
+                    )
+                    if (timing != null) {
+                        TimingBar(
+                            detectionMs = timing!!.detectionMs,
+                            recognitionMs = timing!!.recognitionMs,
+                            totalMs = timing!!.totalMs,
+                        )
+                    }
+                    Spacer(modifier = Modifier.height(4.dp))
+                    ResultList(
+                        results = s.result.results,
+                        onCopyAll = { viewModel.copyAllResults(s.result.results) },
+                    )
+                    Spacer(modifier = Modifier.height(8.dp))
+                    ImagePicker(
+                        onGalleryClick = { galleryLauncher.launch(PickVisualMediaRequest(ActivityResultContracts.PickVisualMedia.ImageOnly)) },
+                        onSampleClick = { viewModel.onSampleImageClicked(it) },
+                        sampleImages = emptyList(),
+                    )
+                    Spacer(modifier = Modifier.height(16.dp))
+                }
+            }
+            is OCRViewModel.UIState.Error -> {
+                LoadingOverlay("Error occurred")
+                ErrorDialog(
+                    message = s.message,
+                    onRetry = { viewModel.retry() },
+                    onDismiss = { viewModel.retry() },
+                )
+            }
+        }
+    }
+}
diff --git a/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/theme/Theme.kt b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/theme/Theme.kt
new file mode 100644
index 0000000..84a4b19
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/theme/Theme.kt
@@ -0,0 +1,45 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.demo.ui.theme
+
+import androidx.compose.material3.MaterialTheme
+import androidx.compose.material3.lightColorScheme
+import androidx.compose.runtime.Composable
+import androidx.compose.ui.graphics.Color
+
+private val LightColorScheme = lightColorScheme(
+    primary = Color(0xFF1565C0),
+    onPrimary = Color.White,
+    primaryContainer = Color(0xFFD1E4FF),
+    onPrimaryContainer = Color(0xFF001D36),
+    secondary = Color(0xFF006874),
+    onSecondary = Color.White,
+    secondaryContainer = Color(0xFF9EEFF7),
+    onSecondaryContainer = Color(0xFF001F24),
+    surface = Color(0xFFFDFBFF),
+    onSurface = Color(0xFF1A1C1E),
+    background = Color(0xFFFDFBFF),
+    onBackground = Color(0xFF1A1C1E),
+    error = Color(0xFFBA1A1A),
+    onError = Color.White,
+)
+
+@Composable
+fun PPOCRTheme(content: @Composable () -> Unit) {
+    MaterialTheme(
+        colorScheme = LightColorScheme,
+        content = content,
+    )
+}
diff --git a/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/viewmodel/OCRViewModel.kt b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/viewmodel/OCRViewModel.kt
new file mode 100644
index 0000000..e3845ce
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/java/com/paddle/ocr/demo/ui/viewmodel/OCRViewModel.kt
@@ -0,0 +1,178 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.demo.ui.viewmodel
+
+import android.graphics.Bitmap
+import android.graphics.BitmapFactory
+import android.net.Uri
+import androidx.lifecycle.ViewModel
+import androidx.lifecycle.viewModelScope
+import com.paddle.ocr.demo.OCRApplication
+import com.paddle.ocr.model.OCRResult
+import com.paddle.ocr.model.OCRRunResult
+import kotlinx.coroutines.Dispatchers
+import kotlinx.coroutines.flow.MutableStateFlow
+import kotlinx.coroutines.flow.StateFlow
+import kotlinx.coroutines.flow.asStateFlow
+import kotlinx.coroutines.flow.collect
+import kotlinx.coroutines.launch
+import kotlinx.coroutines.withContext
+
+class OCRViewModel : ViewModel() {
+
+    sealed class UIState {
+        data object Loading : UIState()
+        data object Ready : UIState()
+        data class Processing(val bitmap: Bitmap) : UIState()
+        data class Result(val bitmap: Bitmap, val result: OCRRunResult) : UIState()
+        data class Error(val message: String) : UIState()
+    }
+
+    data class TimingInfo(
+        val detectionMs: Long,
+        val recognitionMs: Long,
+        val totalMs: Long,
+    )
+
+    private val _uiState = MutableStateFlow<UIState>(
+        uiStateForModelState(OCRApplication.instance.modelState.value)
+    )
+    val uiState: StateFlow<UIState> = _uiState.asStateFlow()
+
+    private val _timing = MutableStateFlow<TimingInfo?>(null)
+    val timing: StateFlow<TimingInfo?> = _timing.asStateFlow()
+
+    private fun uiStateForModelState(modelState: OCRApplication.ModelState): UIState {
+        return when (modelState) {
+            is OCRApplication.ModelState.Loading -> UIState.Loading
+            is OCRApplication.ModelState.Ready -> UIState.Ready
+            is OCRApplication.ModelState.Error -> UIState.Error(modelState.message)
+        }
+    }
+
+    init {
+        viewModelScope.launch {
+            OCRApplication.instance.modelState.collect { modelState ->
+                when (modelState) {
+                    is OCRApplication.ModelState.Loading -> _uiState.value = UIState.Loading
+                    is OCRApplication.ModelState.Ready -> {
+                        if (_uiState.value is UIState.Loading || _uiState.value is UIState.Error) {
+                            _uiState.value = UIState.Ready
+                        }
+                    }
+                    is OCRApplication.ModelState.Error -> _uiState.value = UIState.Error(modelState.message)
+                }
+            }
+        }
+    }
+
+    fun onImageSelected(uri: Uri) {
+        viewModelScope.launch {
+            try {
+                val bytes = withContext(Dispatchers.IO) {
+                    OCRApplication.instance.contentResolver.openInputStream(uri)?.use { it.readBytes() }
+                }
+                if (bytes != null) {
+                    val bitmap = withContext(Dispatchers.IO) {
+                        decodeSampledBitmap(bytes, maxWidth = 2048, maxHeight = 2048)
+                    }
+                    if (bitmap != null) {
+                        processImageBytes(bytes, bitmap)
+                    } else {
+                        _uiState.value = UIState.Error("Failed to load image")
+                    }
+                } else {
+                    _uiState.value = UIState.Error("Failed to load image")
+                }
+            } catch (e: Exception) {
+                _uiState.value = UIState.Error(e.message ?: "Unknown error")
+            }
+        }
+    }
+
+    fun onSampleImageClicked(resId: Int) {
+        viewModelScope.launch {
+            val bytes = withContext(Dispatchers.IO) {
+                OCRApplication.instance.resources.openRawResource(resId).use { it.readBytes() }
+            }
+            val bitmap = withContext(Dispatchers.IO) {
+                BitmapFactory.decodeByteArray(bytes, 0, bytes.size)
+            }
+            if (bitmap != null) {
+                processImageBytes(bytes, bitmap)
+            } else {
+                _uiState.value = UIState.Error("Failed to load image")
+            }
+        }
+    }
+
+    private suspend fun processImageBytes(bytes: ByteArray, bitmap: Bitmap) {
+        _uiState.value = UIState.Processing(bitmap)
+
+        try {
+            val ocr = OCRApplication.instance.ocr
+                ?: throw IllegalStateException("OCR engine not initialized")
+
+            val result = ocr.recognize(bytes)
+            _uiState.value = UIState.Result(bitmap, result)
+            _timing.value = TimingInfo(
+                detectionMs = result.detectionTimeMs,
+                recognitionMs = result.recognitionTimeMs,
+                totalMs = result.totalTimeMs,
+            )
+        } catch (e: Exception) {
+            _uiState.value = UIState.Error(e.message ?: "OCR failed")
+        }
+    }
+
+    fun retry() {
+        val app = OCRApplication.instance
+        if (app.isModelLoaded) {
+            _uiState.value = UIState.Ready
+        } else {
+            app.retryLoadModels()
+        }
+    }
+
+    fun copyAllResults(results: List<OCRResult>) {
+        val text = results.joinToString("\n") { it.text }
+        val clipboard = OCRApplication.instance.getSystemService(android.content.Context.CLIPBOARD_SERVICE)
+                as android.content.ClipboardManager
+        clipboard.setPrimaryClip(android.content.ClipData.newPlainText("OCR Results", text))
+    }
+
+    private fun decodeSampledBitmap(
+        bytes: ByteArray,
+        maxWidth: Int,
+        maxHeight: Int,
+    ): Bitmap? {
+        // First pass: decode bounds only
+        val opts = BitmapFactory.Options().apply { inJustDecodeBounds = true }
+        BitmapFactory.decodeByteArray(bytes, 0, bytes.size, opts)
+        val (origW, origH) = opts.outWidth to opts.outHeight
+        if (origW <= 0 || origH <= 0) return null
+
+        // Calculate sample size (power of 2)
+        var sampleSize = 1
+        while (sampleSize * 2 <= maxOf(origW / maxWidth, origH / maxHeight)) {
+            sampleSize *= 2
+        }
+
+        // Second pass: decode with sampling
+        return BitmapFactory.Options().apply {
+            inSampleSize = sampleSize
+        }.let { BitmapFactory.decodeByteArray(bytes, 0, bytes.size, it) }
+    }
+}
diff --git a/deploy/ppocr-android/app/src/main/res/drawable/ic_launcher_foreground.xml b/deploy/ppocr-android/app/src/main/res/drawable/ic_launcher_foreground.xml
new file mode 100644
index 0000000..f9a506c
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/res/drawable/ic_launcher_foreground.xml
@@ -0,0 +1,16 @@
+<?xml version="1.0" encoding="utf-8"?>
+<vector xmlns:android="http://schemas.android.com/apk/res/android"
+    android:width="108dp"
+    android:height="108dp"
+    android:viewportWidth="108"
+    android:viewportHeight="108">
+    <path
+        android:fillColor="#FFFFFF"
+        android:pathData="M54,28 L54,28 C68.36,28 80,39.64 80,54 L80,54 C80,68.36 68.36,80 54,80 L54,80 C39.64,80 28,68.36 28,54 L28,54 C28,39.64 39.64,28 54,28 Z" />
+    <path
+        android:fillColor="#1A73E8"
+        android:pathData="M46,42 L62,42 L62,46 L50,46 L50,52 L60,52 L60,56 L50,56 L50,62 L62,62 L62,66 L46,66 Z" />
+    <path
+        android:fillColor="#1A73E8"
+        android:pathData="M42,50 L42,46 L38,46 L38,62 L42,62 L42,58 C43.5,60.5 46,62 49,62 C53.97,62 58,57.97 58,54 C58,50.03 53.97,46 49,46 C46,46 43.5,47.5 42,50 Z M49,58 C46.79,58 45,56.21 45,54 C45,51.79 46.79,50 49,50 C51.21,50 53,51.79 53,54 C53,56.21 51.21,58 49,58 Z" />
+</vector>
diff --git a/deploy/ppocr-android/app/src/main/res/mipmap-anydpi-v26/ic_launcher.xml b/deploy/ppocr-android/app/src/main/res/mipmap-anydpi-v26/ic_launcher.xml
new file mode 100644
index 0000000..5ed0a2d
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/res/mipmap-anydpi-v26/ic_launcher.xml
@@ -0,0 +1,5 @@
+<?xml version="1.0" encoding="utf-8"?>
+<adaptive-icon xmlns:android="http://schemas.android.com/apk/res/android">
+    <background android:drawable="@color/ic_launcher_background"/>
+    <foreground android:drawable="@drawable/ic_launcher_foreground"/>
+</adaptive-icon>
diff --git a/deploy/ppocr-android/app/src/main/res/mipmap-anydpi-v26/ic_launcher_round.xml b/deploy/ppocr-android/app/src/main/res/mipmap-anydpi-v26/ic_launcher_round.xml
new file mode 100644
index 0000000..5ed0a2d
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/res/mipmap-anydpi-v26/ic_launcher_round.xml
@@ -0,0 +1,5 @@
+<?xml version="1.0" encoding="utf-8"?>
+<adaptive-icon xmlns:android="http://schemas.android.com/apk/res/android">
+    <background android:drawable="@color/ic_launcher_background"/>
+    <foreground android:drawable="@drawable/ic_launcher_foreground"/>
+</adaptive-icon>
diff --git a/deploy/ppocr-android/app/src/main/res/values/colors.xml b/deploy/ppocr-android/app/src/main/res/values/colors.xml
new file mode 100644
index 0000000..f67db2f
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/res/values/colors.xml
@@ -0,0 +1,4 @@
+<?xml version="1.0" encoding="utf-8"?>
+<resources>
+    <color name="ic_launcher_background">#1A73E8</color>
+</resources>
diff --git a/deploy/ppocr-android/app/src/main/res/values/strings.xml b/deploy/ppocr-android/app/src/main/res/values/strings.xml
new file mode 100644
index 0000000..d344644
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/res/values/strings.xml
@@ -0,0 +1,4 @@
+<?xml version="1.0" encoding="utf-8"?>
+<resources>
+    <string name="app_name">PP-OCRv6 Demo</string>
+</resources>
diff --git a/deploy/ppocr-android/app/src/main/res/values/themes.xml b/deploy/ppocr-android/app/src/main/res/values/themes.xml
new file mode 100644
index 0000000..e6d6a9d
--- /dev/null
+++ b/deploy/ppocr-android/app/src/main/res/values/themes.xml
@@ -0,0 +1,4 @@
+<?xml version="1.0" encoding="utf-8"?>
+<resources>
+    <style name="Theme.PaddleOCR" parent="android:Theme.Material.Light.NoActionBar" />
+</resources>
diff --git a/deploy/ppocr-android/build.gradle.kts b/deploy/ppocr-android/build.gradle.kts
new file mode 100644
index 0000000..81f6a88
--- /dev/null
+++ b/deploy/ppocr-android/build.gradle.kts
@@ -0,0 +1,6 @@
+plugins {
+    alias(libs.plugins.android.application) apply false
+    alias(libs.plugins.android.library) apply false
+    alias(libs.plugins.kotlin.android) apply false
+    alias(libs.plugins.kotlin.compose) apply false
+}
diff --git a/deploy/ppocr-android/gradle.properties b/deploy/ppocr-android/gradle.properties
new file mode 100644
index 0000000..f0a2e55
--- /dev/null
+++ b/deploy/ppocr-android/gradle.properties
@@ -0,0 +1,4 @@
+org.gradle.jvmargs=-Xmx2048m -Dfile.encoding=UTF-8
+android.useAndroidX=true
+kotlin.code.style=official
+android.nonTransitiveRClass=true
diff --git a/deploy/ppocr-android/gradle/libs.versions.toml b/deploy/ppocr-android/gradle/libs.versions.toml
new file mode 100644
index 0000000..b13db7c
--- /dev/null
+++ b/deploy/ppocr-android/gradle/libs.versions.toml
@@ -0,0 +1,41 @@
+[versions]
+agp = "8.7.3"
+kotlin = "2.1.0"
+coreKtx = "1.15.0"
+lifecycleRuntime = "2.8.7"
+activityCompose = "1.9.3"
+composeBom = "2024.12.01"
+onnxruntime = "1.21.1"
+opencv = "4.5.3"
+coroutines = "1.9.0"
+coil = "2.7.0"
+test-runner = "1.2.0"
+test-ext-junit-ktx = "1.2.1"
+test-monitor = "1.7.2"
+junit = "4.13.2"
+
+[libraries]
+androidx-core-ktx = { group = "androidx.core", name = "core-ktx", version.ref = "coreKtx" }
+androidx-lifecycle-runtime-ktx = { group = "androidx.lifecycle", name = "lifecycle-runtime-ktx", version.ref = "lifecycleRuntime" }
+androidx-lifecycle-viewmodel-compose = { group = "androidx.lifecycle", name = "lifecycle-viewmodel-compose", version.ref = "lifecycleRuntime" }
+androidx-activity-compose = { group = "androidx.activity", name = "activity-compose", version.ref = "activityCompose" }
+compose-bom = { group = "androidx.compose", name = "compose-bom", version.ref = "composeBom" }
+compose-ui = { group = "androidx.compose.ui", name = "ui" }
+compose-ui-graphics = { group = "androidx.compose.ui", name = "ui-graphics" }
+compose-ui-tooling-preview = { group = "androidx.compose.ui", name = "ui-tooling-preview" }
+compose-ui-tooling = { group = "androidx.compose.ui", name = "ui-tooling" }
+compose-material3 = { group = "androidx.compose.material3", name = "material3" }
+onnxruntime-android = { group = "com.microsoft.onnxruntime", name = "onnxruntime-android", version.ref = "onnxruntime" }
+opencv-android = { group = "com.quickbirdstudios", name = "opencv", version.ref = "opencv" }
+kotlinx-coroutines-android = { group = "org.jetbrains.kotlinx", name = "kotlinx-coroutines-android", version.ref = "coroutines" }
+coil-compose = { group = "io.coil-kt", name = "coil-compose", version.ref = "coil" }
+androidx-test-runner = { group = "androidx.test", name = "runner", version.ref = "test-runner" }
+androidx-test-ext-junit-ktx = { group = "androidx.test.ext", name = "junit-ktx", version.ref = "test-ext-junit-ktx" }
+androidx-test-monitor = { group = "androidx.test", name = "monitor", version.ref = "test-monitor" }
+junit = { group = "junit", name = "junit", version.ref = "junit" }
+
+[plugins]
+android-application = { id = "com.android.application", version.ref = "agp" }
+android-library = { id = "com.android.library", version.ref = "agp" }
+kotlin-android = { id = "org.jetbrains.kotlin.android", version.ref = "kotlin" }
+kotlin-compose = { id = "org.jetbrains.kotlin.plugin.compose", version.ref = "kotlin" }
diff --git a/deploy/ppocr-android/gradle/wrapper/gradle-wrapper.jar b/deploy/ppocr-android/gradle/wrapper/gradle-wrapper.jar
new file mode 100644
index 0000000..2c35211
Binary files /dev/null and b/deploy/ppocr-android/gradle/wrapper/gradle-wrapper.jar differ
diff --git a/deploy/ppocr-android/gradle/wrapper/gradle-wrapper.properties b/deploy/ppocr-android/gradle/wrapper/gradle-wrapper.properties
new file mode 100644
index 0000000..09523c0
--- /dev/null
+++ b/deploy/ppocr-android/gradle/wrapper/gradle-wrapper.properties
@@ -0,0 +1,7 @@
+distributionBase=GRADLE_USER_HOME
+distributionPath=wrapper/dists
+distributionUrl=https\://services.gradle.org/distributions/gradle-8.9-bin.zip
+networkTimeout=10000
+validateDistributionUrl=true
+zipStoreBase=GRADLE_USER_HOME
+zipStorePath=wrapper/dists
diff --git a/deploy/ppocr-android/gradlew b/deploy/ppocr-android/gradlew
new file mode 100644
index 0000000..6696a4d
--- /dev/null
+++ b/deploy/ppocr-android/gradlew
@@ -0,0 +1,72 @@
+#!/bin/sh
+
+#
+# Gradle start up script for POSIX generated by Gradle.
+#
+
+# Attempt to set APP_HOME
+PRG="$0"
+while [ -h "$PRG" ] ; do
+    ls=`ls -ld "$PRG"`
+    link=`expr "$ls" : '.*-> \(.*\)$'`
+    if expr "$link" : '/.*' > /dev/null; then
+        PRG="$link"
+    else
+        PRG=`dirname "$PRG"`"/$link"
+    fi
+done
+SAVED="`pwd`"
+cd "`dirname \"$PRG\"`/" >/dev/null
+APP_HOME="`pwd -P`"
+cd "$SAVED" >/dev/null
+
+APP_NAME="Gradle"
+APP_BASE_NAME=`basename "$0"`
+
+DEFAULT_JVM_OPTS='-Xmx64m -Xms64m'
+
+MAX_FD="maximum"
+
+warn () {
+    echo "$*"
+} >&2
+
+die () {
+    echo
+    echo "$*"
+    echo
+    exit 1
+} >&2
+
+OS_NAME="`uname`"
+case "$OS_NAME" in
+    CYGWIN* | MINGW* | MSYS* )
+        IS_WINDOWS=true
+        ;;
+esac
+
+CLASSPATH=$APP_HOME/gradle/wrapper/gradle-wrapper.jar
+
+if [ -n "$JAVA_HOME" ] ; then
+    JAVACMD="$JAVA_HOME/bin/java"
+else
+    JAVACMD="java"
+fi
+
+if [ ! -x "$JAVACMD" ] ; then
+    die "ERROR: JAVA_HOME is not set and no 'java' command could be found in your PATH.
+
+Please set the JAVA_HOME variable in your environment to match the
+location of your Java installation."
+fi
+
+GRADLE_OPTS="${GRADLE_OPTS} -Dorg.gradle.appname=${APP_BASE_NAME}"
+
+exec "$JAVACMD" \
+    $DEFAULT_JVM_OPTS \
+    $JAVA_OPTS \
+    $GRADLE_OPTS \
+    "-Dorg.gradle.appname=$APP_BASE_NAME" \
+    -classpath "$CLASSPATH" \
+    org.gradle.wrapper.GradleWrapperMain \
+    "$@"
diff --git a/deploy/ppocr-android/ppocr-sdk/build.gradle.kts b/deploy/ppocr-android/ppocr-sdk/build.gradle.kts
new file mode 100644
index 0000000..1a220e3
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/build.gradle.kts
@@ -0,0 +1,42 @@
+plugins {
+    alias(libs.plugins.android.library)
+    alias(libs.plugins.kotlin.android)
+}
+
+android {
+    namespace = "com.paddle.ocr"
+    compileSdk = 35
+
+    defaultConfig {
+        minSdk = 26
+        testInstrumentationRunner = "androidx.test.runner.AndroidJUnitRunner"
+    }
+
+    buildTypes {
+        release {
+            isMinifyEnabled = false
+            consumerProguardFiles("proguard-rules.pro")
+        }
+    }
+    compileOptions {
+        sourceCompatibility = JavaVersion.VERSION_17
+        targetCompatibility = JavaVersion.VERSION_17
+    }
+    kotlin {
+        compilerOptions {
+            jvmTarget.set(org.jetbrains.kotlin.gradle.dsl.JvmTarget.JVM_17)
+        }
+    }
+}
+
+dependencies {
+    implementation(libs.onnxruntime.android)
+    implementation(libs.opencv.android)
+    implementation(libs.kotlinx.coroutines.android)
+    implementation(libs.androidx.core.ktx)
+
+    androidTestImplementation(libs.androidx.test.runner)
+    androidTestImplementation(libs.androidx.test.ext.junit.ktx)
+    androidTestImplementation(libs.androidx.test.monitor)
+    androidTestImplementation(libs.junit)
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/proguard-rules.pro b/deploy/ppocr-android/ppocr-sdk/proguard-rules.pro
new file mode 100644
index 0000000..5c912dc
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/proguard-rules.pro
@@ -0,0 +1,2 @@
+-keep class com.paddle.ocr.** { *; }
+-keep class ai.onnxruntime.** { *; }
diff --git a/deploy/ppocr-android/ppocr-sdk/src/androidTest/java/com/paddle/ocr/benchmark/BenchmarkFixtures.kt b/deploy/ppocr-android/ppocr-sdk/src/androidTest/java/com/paddle/ocr/benchmark/BenchmarkFixtures.kt
new file mode 100644
index 0000000..c5fbdea
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/androidTest/java/com/paddle/ocr/benchmark/BenchmarkFixtures.kt
@@ -0,0 +1,19 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.benchmark
+
+object BenchmarkFixtures {
+    const val defaultReferenceImageName = "android_ocr_benchmark_reference.jpg"
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/androidTest/java/com/paddle/ocr/benchmark/OCRBenchmarkTest.kt b/deploy/ppocr-android/ppocr-sdk/src/androidTest/java/com/paddle/ocr/benchmark/OCRBenchmarkTest.kt
new file mode 100644
index 0000000..b815302
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/androidTest/java/com/paddle/ocr/benchmark/OCRBenchmarkTest.kt
@@ -0,0 +1,374 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.benchmark
+
+import android.os.Build
+import android.os.Debug
+import android.util.Log
+import androidx.test.ext.junit.runners.AndroidJUnit4
+import androidx.test.platform.app.InstrumentationRegistry
+import com.paddle.ocr.EngineConfig
+import com.paddle.ocr.PaddleOCR
+import com.paddle.ocr.PaddleOCRConfig
+import com.paddle.ocr.model.OCRRunResult
+import com.paddle.ocr.util.OpenCVUtils
+import kotlinx.coroutines.runBlocking
+import org.json.JSONArray
+import org.json.JSONObject
+import org.junit.Assert.assertTrue
+import org.junit.BeforeClass
+import org.junit.Test
+import org.junit.runner.RunWith
+import java.io.File
+import kotlin.math.pow
+import kotlin.math.sqrt
+
+@RunWith(AndroidJUnit4::class)
+class OCRBenchmarkTest {
+
+    companion object {
+        private const val TAG = "OCRBenchmark"
+
+        // Instrumentation arguments (passed via -Pandroid.testInstrumentationRunnerArguments.*)
+        private const val ARG_WARMUP = "warmup"
+        private const val ARG_ITERATIONS = "iterations"
+        private const val ARG_REC_BATCH_SIZE = "rec_batch_size"
+
+        private const val DEFAULT_WARMUP = 3
+        private const val DEFAULT_ITERATIONS = 10
+
+        private lateinit var ocr: PaddleOCR
+        private lateinit var outputDir: File
+
+        @JvmStatic
+        @BeforeClass
+        fun setUp() {
+            val context = InstrumentationRegistry.getInstrumentation().targetContext
+            OpenCVUtils.init(context)
+
+            outputDir = File(context.getExternalFilesDir(null), "ocr_benchmark")
+            outputDir.mkdirs()
+            println("[OCRBenchmark] Output dir: ${outputDir.absolutePath}")
+
+            runBlocking {
+                val batchSize = getInstrumentationArgument(ARG_REC_BATCH_SIZE)?.toIntOrNull() ?: 1
+                val config = PaddleOCRConfig(recBatchSize = batchSize)
+                ocr = PaddleOCR.create(context, config, EngineConfig())
+            }
+            println("[OCRBenchmark] Engine ready, coldLoadTimeMs=${ocr.coldLoadTimeMs}")
+        }
+
+        private fun getInstrumentationArgument(key: String): String? {
+            val args = InstrumentationRegistry.getArguments()
+            return args[key] as? String
+        }
+
+        private fun saveAndPrint(filename: String, json: JSONObject) {
+            val file = File(outputDir, filename)
+            file.writeText(json.toString(2))
+            println("[OCRBenchmark] Saved: ${file.absolutePath}")
+            println("[OCRBenchmark] ---- $filename ----")
+            json.toString(2).lines().forEach { println("[OCRBenchmark] $it") }
+            println("[OCRBenchmark] ---- end ----")
+        }
+    }
+
+    // ========================================================================
+    // Accuracy
+    // ========================================================================
+
+    @Test
+    fun testOCRExportJSON() {
+        val imageBytes = loadBenchmarkImageBytes()
+        val result = runBlocking { ocr.recognize(imageBytes) }
+        val payload = buildExportPayload(result)
+        saveAndPrint("accuracy_export.json", payload)
+
+        assertTrue("Should have results", payload.getJSONArray("items").length() > 0)
+        assertTrue("Should detect text", result.lineCount > 0)
+    }
+
+    // ========================================================================
+    // Latency
+    // ========================================================================
+
+    @Test
+    fun testLatencyBenchmark() {
+        val imageBytes = loadBenchmarkImageBytes()
+        val warmup = getInstrumentationArgument(ARG_WARMUP)?.toIntOrNull() ?: DEFAULT_WARMUP
+        val iterations = getInstrumentationArgument(ARG_ITERATIONS)?.toIntOrNull() ?: DEFAULT_ITERATIONS
+
+        println("[OCRBenchmark] Warming up $warmup iterations...")
+        runBlocking { repeat(warmup) { ocr.recognize(imageBytes) } }
+
+        println("[OCRBenchmark] Measuring $iterations iterations...")
+        val totals = mutableListOf<Long>()
+        val dets = mutableListOf<Long>()
+        val detPre = mutableListOf<Long>()
+        val detInf = mutableListOf<Long>()
+        val detPost = mutableListOf<Long>()
+        val recs = mutableListOf<Long>()
+        val recPre = mutableListOf<Long>()
+        val recInf = mutableListOf<Long>()
+        val recPost = mutableListOf<Long>()
+        val overheads = mutableListOf<Long>()
+        val perLineMs = mutableListOf<Long>()
+        val detShapes = mutableListOf<List<Int>>()
+        val recShapes = mutableListOf<List<Int>>()
+        var firstLineCount = 0
+
+        runBlocking {
+            repeat(iterations) { i ->
+                val r = ocr.recognize(imageBytes)
+                if (i == 0) firstLineCount = r.lineCount
+                totals.add(r.totalTimeMs)
+                dets.add(r.detectionTimeMs)
+                detPre.add(r.detPreprocessMs)
+                detInf.add(r.detInferenceMs)
+                detPost.add(r.detPostprocessMs)
+                recs.add(r.recognitionTimeMs)
+                recPre.add(r.recPreprocessMs)
+                recInf.add(r.recInferenceMs)
+                recPost.add(r.recPostprocessMs)
+                overheads.add(r.pipelineOverheadMs)
+                detShapes.add(r.detInputShape)
+                recShapes.addAll(r.recInputShapes)
+                perLineMs.addAll(r.perLineRecMs)
+                if ((i + 1) % 5 == 0) println("[OCRBenchmark]   $i/${iterations}...")
+            }
+        }
+
+        val memInfo = Debug.MemoryInfo()
+        Debug.getMemoryInfo(memInfo)
+
+        val payload = JSONObject().apply {
+            put("schemaVersion", 1)
+            put("deviceModel", Build.MODEL)
+            put("osVersion", Build.VERSION.RELEASE)
+            put("ortExecutionProvider", "CPU")
+            put("inputImage", BenchmarkFixtures.defaultReferenceImageName)
+            put("inputBytes", imageBytes.size)
+            put("warmupIterations", warmup)
+            put("measuredIterations", iterations)
+            put("firstMeasuredLineCount", firstLineCount)
+            put("coldLoadTimeMs", ocr.coldLoadTimeMs)
+            if (totals.isNotEmpty()) {
+                put("firstMeasuredRun", JSONObject().apply {
+                    put("totalTimeMs", totals.first())
+                    put("detectionTimeMs", dets.first())
+                    put("recognitionTimeMs", recs.first())
+                    put("pipelineOverheadTimeMs", overheads.first())
+                })
+            }
+            put("totalTimeMs", summarizeMs(totals).toJson())
+            put("detectionTimeMs", summarizeMs(dets).toJson())
+            put("detectionPreprocessTimeMs", summarizeMs(detPre).toJson())
+            put("detectionInferenceTimeMs", summarizeMs(detInf).toJson())
+            put("detectionPostprocessTimeMs", summarizeMs(detPost).toJson())
+            put("recognitionTimeMs", summarizeMs(recs).toJson())
+            put("recognitionPreprocessTimeMs", summarizeMs(recPre).toJson())
+            put("recognitionInferenceTimeMs", summarizeMs(recInf).toJson())
+            put("recognitionPostprocessTimeMs", summarizeMs(recPost).toJson())
+            put("pipelineOverheadTimeMs", summarizeMs(overheads).toJson())
+            put("memoryPssKb", memInfo.totalPss)
+            put("inputShapeDistribution", JSONObject().apply {
+                put("detection", summarizeShapes(detShapes))
+                put("recognition", summarizeShapes(recShapes))
+            })
+            if (perLineMs.isNotEmpty()) {
+                put("recognitionPerLine", JSONObject().apply {
+                    put("pooled", JSONObject().apply {
+                        put("count", perLineMs.size)
+                        put("totalMs", summarizeMs(perLineMs).toJson())
+                    })
+                })
+            }
+        }
+
+        saveAndPrint("latency_benchmark.json", payload)
+        printSpeedSummary(warmup, iterations, totals, dets, detPre, detInf, detPost, recs, recPre, recInf, recPost, overheads, firstLineCount)
+
+        assertTrue("Should complete all iterations", totals.size == iterations)
+        assertTrue("Should detect text", firstLineCount > 0)
+    }
+
+    private fun printSpeedSummary(
+        warmup: Int,
+        iterations: Int,
+        totals: List<Long>,
+        dets: List<Long>,
+        detPre: List<Long>,
+        detInf: List<Long>,
+        detPost: List<Long>,
+        recs: List<Long>,
+        recPre: List<Long>,
+        recInf: List<Long>,
+        recPost: List<Long>,
+        overheads: List<Long>,
+        lineCount: Int,
+    ) {
+        val bar = "+-----------------------------+----------+----------+----------+----------+"
+        val header = "| Stage                       |  Mean ms |    Stdev |       P90 |    Min ms |"
+        fun row(label: String, samples: List<Long>): String {
+            if (samples.isEmpty()) return "| %-27s | %8s | %8s | %8s | %8s |".format(label, "-", "-", "-", "N/A")
+            val mean = samples.map { it.toDouble() }.average()
+            val variance = if (samples.size > 1) samples.map { (it - mean).pow(2) }.average() else 0.0
+            val stdev = sqrt(variance)
+            val sorted = samples.sorted()
+            val p90 = sorted[(0.9 * (sorted.size - 1)).toInt()]
+            val min = sorted.first()
+            return "| %-27s | %8.2f | %8.2f | %8d | %8d |".format(label, mean, stdev, p90, min)
+        }
+
+        println("[OCRBenchmark] ")
+        println("[OCRBenchmark] ╔════════════════════════════════════════════════════════════════════════════════╗")
+        println("[OCRBenchmark] ║  PP-OCRv6 Speed Benchmark Results                                      ║")
+        println("[OCRBenchmark] ╠════════════════════════════════════════════════════════════════════════════════╣")
+        println("[OCRBenchmark] ║  Device: ${Build.MODEL}  |  OS: Android ${Build.VERSION.RELEASE}  |  Lines: $lineCount")
+        println("[OCRBenchmark] ║  Cold load: ${ocr.coldLoadTimeMs}ms  |  Warmup: $warmup  |  Measured: $iterations")
+        println("[OCRBenchmark] ╠════════════════════════════════════════════════════════════════════════════════╣")
+        println("[OCRBenchmark] $bar")
+        println("[OCRBenchmark] $header")
+        println("[OCRBenchmark] $bar")
+        println("[OCRBenchmark] ${row("Total pipeline", totals)}")
+        println("[OCRBenchmark] $bar")
+        println("[OCRBenchmark] ${row("  Detection (total)", dets)}")
+        println("[OCRBenchmark] ${row("    Preprocess", detPre)}")
+        println("[OCRBenchmark] ${row("    Inference", detInf)}")
+        println("[OCRBenchmark] ${row("    Postprocess", detPost)}")
+        println("[OCRBenchmark] ${row("  Recognition (total)", recs)}")
+        println("[OCRBenchmark] ${row("    Preprocess", recPre)}")
+        println("[OCRBenchmark] ${row("    Inference", recInf)}")
+        println("[OCRBenchmark] ${row("    Postprocess", recPost)}")
+        println("[OCRBenchmark] ${row("  Pipeline overhead", overheads)}")
+        println("[OCRBenchmark] $bar")
+        println("[OCRBenchmark] ╚════════════════════════════════════════════════════════════════════════════════╝")
+        println("[OCRBenchmark] ")
+    }
+
+    // ========================================================================
+    // Memory
+    // ========================================================================
+
+    @Test
+    fun testMemoryBenchmark() {
+        val imageBytes = loadBenchmarkImageBytes()
+        val warmup = getInstrumentationArgument(ARG_WARMUP)?.toIntOrNull() ?: DEFAULT_WARMUP
+        val iterations = getInstrumentationArgument(ARG_ITERATIONS)?.toIntOrNull() ?: DEFAULT_ITERATIONS
+
+        runBlocking { repeat(warmup) { ocr.recognize(imageBytes) } }
+        Runtime.getRuntime().gc()
+        Thread.sleep(500)
+
+        val before = Debug.MemoryInfo()
+        Debug.getMemoryInfo(before)
+
+        runBlocking { repeat(iterations) { ocr.recognize(imageBytes) } }
+
+        val after = Debug.MemoryInfo()
+        Debug.getMemoryInfo(after)
+
+        val payload = JSONObject().apply {
+            put("schemaVersion", 1)
+            put("deviceModel", Build.MODEL)
+            put("iterations", iterations)
+            put("memoryBeforePssKb", before.totalPss)
+            put("memoryAfterPssKb", after.totalPss)
+            put("memoryDeltaPssKb", after.totalPss - before.totalPss)
+            put("nativeBeforeKb", before.nativePss)
+            put("nativeAfterKb", after.nativePss)
+            put("dalvikBeforeKb", before.dalvikPss)
+            put("dalvikAfterKb", after.dalvikPss)
+        }
+
+        saveAndPrint("memory_benchmark.json", payload)
+        assertTrue("Memory should be measurable", after.totalPss > 0)
+    }
+
+    // ========================================================================
+    // Helpers
+    // ========================================================================
+
+    private fun loadBenchmarkImageBytes(): ByteArray {
+        val context = InstrumentationRegistry.getInstrumentation().context
+        val stem = BenchmarkFixtures.defaultReferenceImageName.substringBeforeLast(".")
+        val resId = context.resources.getIdentifier(stem, "raw", context.packageName)
+        if (resId == 0) {
+            throw IllegalStateException(
+                "Missing ${BenchmarkFixtures.defaultReferenceImageName} " +
+                        "(package=${context.packageName})"
+            )
+        }
+        return context.resources.openRawResource(resId).use { it.readBytes() }
+    }
+
+    data class TimingSummary(val mean: Double, val stdev: Double, val p90: Double) {
+        fun toJson() = JSONObject().apply {
+            put("mean", mean)
+            put("stdev", stdev)
+            put("p90", p90)
+        }
+    }
+
+    private fun summarizeMs(samples: List<Long>): TimingSummary {
+        if (samples.isEmpty()) return TimingSummary(0.0, 0.0, 0.0)
+        val doubles = samples.map { it.toDouble() }
+        val mean = doubles.average()
+        val variance = if (doubles.size > 1)
+            doubles.map { (it - mean).pow(2) }.average()
+        else 0.0
+        val stdev = sqrt(variance)
+        val sorted = doubles.sorted()
+        val p90Idx = (0.9 * (sorted.size - 1)).toInt().coerceIn(0, sorted.size - 1)
+        return TimingSummary(mean, stdev, sorted[p90Idx])
+    }
+
+    private fun summarizeShapes(shapes: List<List<Int>>): JSONArray {
+        val counts = mutableMapOf<String, Int>()
+        shapes.forEach { shape ->
+            val key = shape.joinToString(",")
+            counts[key] = (counts[key] ?: 0) + 1
+        }
+        return JSONArray().apply {
+            counts.entries.sortedByDescending { it.value }.forEach { (shape, count) ->
+                put(JSONObject().apply {
+                    put("shape", JSONArray(shape.split(",").map { it.toInt() }))
+                    put("count", count)
+                })
+            }
+        }
+    }
+
+    private fun buildExportPayload(result: OCRRunResult): JSONObject {
+        val items = JSONArray().apply {
+            result.results.forEach { r ->
+                put(JSONObject().apply {
+                    put("polygon", JSONArray().apply {
+                        r.box.points.forEach { pt ->
+                            put(JSONArray(listOf(pt.x.toInt(), pt.y.toInt())))
+                        }
+                    })
+                    put("text", r.text)
+                    put("score", r.confidence.toDouble())
+                })
+            }
+        }
+        return JSONObject().apply {
+            put("schemaVersion", 1)
+            put("source", "android_ocr_demo")
+            put("items", items)
+        }
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/androidTest/res/raw/android_ocr_benchmark_reference.png b/deploy/ppocr-android/ppocr-sdk/src/androidTest/res/raw/android_ocr_benchmark_reference.png
new file mode 100644
index 0000000..3cd2e79
Binary files /dev/null and b/deploy/ppocr-android/ppocr-sdk/src/androidTest/res/raw/android_ocr_benchmark_reference.png differ
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/AndroidManifest.xml b/deploy/ppocr-android/ppocr-sdk/src/main/AndroidManifest.xml
new file mode 100644
index 0000000..9a40236
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/AndroidManifest.xml
@@ -0,0 +1,3 @@
+<?xml version="1.0" encoding="utf-8"?>
+<manifest xmlns:android="http://schemas.android.com/apk/res/android">
+</manifest>
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/EngineConfig.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/EngineConfig.kt
new file mode 100644
index 0000000..54a62d5
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/EngineConfig.kt
@@ -0,0 +1,19 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr
+
+data class EngineConfig(
+    val numThreads: Int = 4,
+)
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/PaddleOCR.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/PaddleOCR.kt
new file mode 100644
index 0000000..6d93490
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/PaddleOCR.kt
@@ -0,0 +1,118 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr
+
+import android.content.Context
+import android.graphics.Bitmap
+import com.paddle.ocr.engine.OCREngine
+import com.paddle.ocr.engine.OCREngineResult
+import com.paddle.ocr.model.OCRRunResult
+import com.paddle.ocr.model.OCRError
+import kotlinx.coroutines.Dispatchers
+import kotlinx.coroutines.withContext
+
+class PaddleOCR private constructor(
+    private val engine: OCREngine,
+) {
+    /** Time spent loading ONNX models (milliseconds). */
+    val coldLoadTimeMs: Long get() = engine.coldLoadTimeMs
+
+    companion object {
+
+        suspend fun create(context: Context): PaddleOCR {
+            val appContext = context.applicationContext
+            return withContext(Dispatchers.IO) {
+                val engine = OCREngine(appContext, PaddleOCRConfig(), EngineConfig())
+                PaddleOCR(engine)
+            }
+        }
+
+        suspend fun create(
+            context: Context,
+            config: PaddleOCRConfig,
+            engineConfig: EngineConfig = EngineConfig(),
+        ): PaddleOCR {
+            val appContext = context.applicationContext
+            return withContext(Dispatchers.IO) {
+                val engine = OCREngine(appContext, config, engineConfig)
+                PaddleOCR(engine)
+            }
+        }
+
+        suspend fun create(
+            context: Context,
+            config: PaddleOCRConfig,
+            engineConfig: EngineConfig,
+            detModelAssetPath: String,
+            recModelAssetPath: String,
+            recConfigAssetPath: String,
+        ): PaddleOCR {
+            val appContext = context.applicationContext
+            return withContext(Dispatchers.IO) {
+                val engine = OCREngine(
+                    appContext, config, engineConfig,
+                    detModelAsset = detModelAssetPath,
+                    recModelAsset = recModelAssetPath,
+                    recConfigAsset = recConfigAssetPath,
+                )
+                PaddleOCR(engine)
+            }
+        }
+    }
+
+    suspend fun recognize(bitmap: Bitmap): OCRRunResult {
+        if (bitmap.width == 0 || bitmap.height == 0) {
+            throw OCRError.InvalidImage()
+        }
+        return recognizeResult { engine.run(bitmap) }
+    }
+
+    suspend fun recognize(imageBytes: ByteArray): OCRRunResult {
+        if (imageBytes.isEmpty()) {
+            throw OCRError.InvalidImage()
+        }
+        return recognizeResult { engine.run(imageBytes) }
+    }
+
+    private suspend fun recognizeResult(runEngine: () -> OCREngineResult): OCRRunResult {
+        return withContext(Dispatchers.IO) {
+            val result = runEngine()
+            OCRRunResult(
+                results = result.results,
+                detectionTimeMs = result.detectionTimeMs,
+                recognitionTimeMs = result.recognitionTimeMs,
+                totalTimeMs = result.totalTimeMs,
+                lineCount = result.lineCount,
+                detPreprocessMs = result.detPreprocessMs,
+                detInferenceMs = result.detInferenceMs,
+                detPostprocessMs = result.detPostprocessMs,
+                recPreprocessMs = result.recPreprocessMs,
+                recInferenceMs = result.recInferenceMs,
+                recPostprocessMs = result.recPostprocessMs,
+                pipelineOverheadMs = result.pipelineOverheadMs,
+                coldLoadTimeMs = result.coldLoadTimeMs,
+                detInputShape = result.detInputShape,
+                recInputShapes = result.recInputShapes,
+                perLineRecMs = result.perLineRecMs,
+            )
+        }
+    }
+
+    suspend fun release() {
+        withContext(Dispatchers.IO) {
+            engine.release()
+        }
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/PaddleOCRConfig.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/PaddleOCRConfig.kt
new file mode 100644
index 0000000..95a63ae
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/PaddleOCRConfig.kt
@@ -0,0 +1,31 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr
+
+data class PaddleOCRConfig(
+    val detImgMode: String = "BGR",
+    val detLimitSideLen: Int = 64,
+    val detLimitType: String = "min",
+    val detMaxSideLimit: Int = 4000,
+    val detThresh: Float = 0.3f,
+    val detBoxThresh: Float = 0.6f,
+    val detUnclipRatio: Float = 1.5f,
+    val detMaxCandidates: Int = 3000,
+    val detUseDilation: Boolean = false,
+    val detScoreMode: String = "fast",
+    val detBoxType: String = "quad",
+    val recScoreThresh: Float = 0.0f,
+    val recBatchSize: Int = 1,
+)
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/engine/DetectionEngine.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/engine/DetectionEngine.kt
new file mode 100644
index 0000000..ab7f2ce
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/engine/DetectionEngine.kt
@@ -0,0 +1,96 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.engine
+
+import android.graphics.Bitmap
+import com.paddle.ocr.PaddleOCRConfig
+import com.paddle.ocr.model.OCRBox
+import com.paddle.ocr.postprocess.DBPostProcessor
+import com.paddle.ocr.preprocess.DetPreprocessResult
+import com.paddle.ocr.preprocess.DetPreprocessor
+import org.opencv.core.Mat
+
+class DetectionEngine(
+    private val ortManager: ORTSessionManager,
+    private val config: PaddleOCRConfig,
+) {
+    data class DetectionResult(
+        val boxes: List<OCRBox>,
+        val preprocessMs: Long,
+        val inferenceMs: Long,
+        val postprocessMs: Long,
+        val timeMs: Long,
+        val inputShape: List<Int>,
+    )
+
+    fun detect(bitmap: Bitmap): DetectionResult {
+        return detect {
+            DetPreprocessor.preprocess(
+                bitmap,
+                config.detLimitSideLen,
+                config.detLimitType,
+                config.detMaxSideLimit,
+                config.detImgMode,
+            )
+        }
+    }
+
+    fun detect(src: Mat): DetectionResult {
+        return detect {
+            DetPreprocessor.preprocess(
+                src,
+                config.detLimitSideLen,
+                config.detLimitType,
+                config.detMaxSideLimit,
+                config.detImgMode,
+            )
+        }
+    }
+
+    private fun detect(preprocessor: () -> DetPreprocessResult): DetectionResult {
+        val preStart = System.currentTimeMillis()
+        val preResult = preprocessor()
+        val preprocessMs = System.currentTimeMillis() - preStart
+
+        val infStart = System.currentTimeMillis()
+        val (outputData, outputShape) = ortManager.runDetection(preResult.tensorData, preResult.shape)
+        val inferenceMs = System.currentTimeMillis() - infStart
+
+        val postStart = System.currentTimeMillis()
+        val boxes = DBPostProcessor.process(
+            pred = outputData,
+            predShape = outputShape,
+            thresh = config.detThresh,
+            boxThresh = config.detBoxThresh,
+            unclipRatio = config.detUnclipRatio,
+            maxCandidates = config.detMaxCandidates,
+            useDilation = config.detUseDilation,
+            scoreMode = config.detScoreMode,
+            boxType = config.detBoxType,
+            originalH = preResult.originalH,
+            originalW = preResult.originalW,
+        )
+        val postprocessMs = System.currentTimeMillis() - postStart
+
+        return DetectionResult(
+            boxes = boxes,
+            preprocessMs = preprocessMs,
+            inferenceMs = inferenceMs,
+            postprocessMs = postprocessMs,
+            timeMs = preprocessMs + inferenceMs + postprocessMs,
+            inputShape = listOf(1, 3, preResult.shape[2].toInt(), preResult.shape[3].toInt()),
+        )
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/engine/OCREngine.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/engine/OCREngine.kt
new file mode 100644
index 0000000..77fd8fa
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/engine/OCREngine.kt
@@ -0,0 +1,184 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.engine
+
+import android.content.Context
+import android.graphics.Bitmap
+import com.paddle.ocr.EngineConfig
+import com.paddle.ocr.PaddleOCRConfig
+import com.paddle.ocr.model.ModelConfig
+import com.paddle.ocr.model.OCRError
+import com.paddle.ocr.model.OCRResult
+import com.paddle.ocr.postprocess.BoxSorter
+import com.paddle.ocr.postprocess.QuadTextCrop
+import com.paddle.ocr.util.BitmapUtils
+
+class OCREngine(
+    context: Context,
+    private val config: PaddleOCRConfig,
+    engineConfig: EngineConfig,
+    detModelAsset: String = "models/det/inference.onnx",
+    recModelAsset: String = "models/rec/inference.onnx",
+    recConfigAsset: String = "models/rec/inference.yml",
+) {
+    private val ortManager = ORTSessionManager(context, engineConfig)
+    private val detectionEngine: DetectionEngine
+    private val recognitionEngine: RecognitionEngine
+    val coldLoadTimeMs: Long get() = ortManager.coldLoadTimeMs
+
+    init {
+        val configured = try {
+            ortManager.loadModels(detModelAsset, recModelAsset)
+            val recModelConfig = ModelConfig.parse(context, recConfigAsset)
+            recModelConfig
+        } catch (t: Throwable) {
+            ortManager.release()
+            throw t
+        }
+        detectionEngine = DetectionEngine(ortManager, config)
+        recognitionEngine = RecognitionEngine(ortManager, configured.characterList)
+    }
+
+    fun run(bitmap: Bitmap): OCREngineResult {
+        val srcMat = BitmapUtils.bitmapToBGRMat(bitmap)
+        return runWithOwnedMat(srcMat)
+    }
+
+    fun run(imageBytes: ByteArray): OCREngineResult {
+        val srcMat = BitmapUtils.imdecodeBGR(imageBytes)
+        if (srcMat.empty()) {
+            srcMat.release()
+            throw OCRError.InvalidImage()
+        }
+        return runWithOwnedMat(srcMat)
+    }
+
+    private fun runWithOwnedMat(srcMat: org.opencv.core.Mat): OCREngineResult {
+        return try {
+            run(srcMat)
+        } finally {
+            srcMat.release()
+        }
+    }
+
+    private fun run(srcMat: org.opencv.core.Mat): OCREngineResult {
+        val totalStart = System.currentTimeMillis()
+        val detResult = detectionEngine.detect(srcMat)
+        val boxes = detResult.boxes
+
+        if (boxes.isEmpty()) {
+            val elapsed = System.currentTimeMillis() - totalStart
+            return OCREngineResult(
+                results = emptyList(),
+                detectionTimeMs = detResult.timeMs,
+                recognitionTimeMs = 0,
+                totalTimeMs = elapsed,
+                lineCount = 0,
+                detPreprocessMs = detResult.preprocessMs,
+                detInferenceMs = detResult.inferenceMs,
+                detPostprocessMs = detResult.postprocessMs,
+                detInputShape = detResult.inputShape,
+                coldLoadTimeMs = ortManager.coldLoadTimeMs,
+            )
+        }
+
+        // 2. Sort boxes
+        val sortedBoxes = BoxSorter.sortInReadingOrder(boxes)
+
+        // 3. Crop and recognize text regions
+        var totalRecPreMs = 0L
+        var totalRecInfMs = 0L
+        var totalRecPostMs = 0L
+        var totalRecMs = 0L
+        val allResults = mutableListOf<OCRResult>()
+        val recInputShapes = mutableListOf<List<Int>>()
+        val perLineRecMs = mutableListOf<Long>()
+        val batchSize = config.recBatchSize.coerceAtLeast(1)
+
+        var i = 0
+        while (i < sortedBoxes.size) {
+            val batchCrops = mutableListOf<org.opencv.core.Mat>()
+            val batchBoxIndices = mutableListOf<Int>()
+            var next = i
+            while (next < sortedBoxes.size && batchCrops.size < batchSize) {
+                val crop = QuadTextCrop.crop(srcMat, sortedBoxes[next])
+                if (crop.rows() > 0 && crop.cols() > 0) {
+                    batchCrops.add(crop)
+                    batchBoxIndices.add(next)
+                } else {
+                    crop.release()
+                }
+                next++
+            }
+
+            try {
+                if (batchCrops.isNotEmpty()) {
+                    val batchResult = recognitionEngine.recognize(batchCrops)
+                    totalRecPreMs += batchResult.preprocessMs
+                    totalRecInfMs += batchResult.inferenceMs
+                    totalRecPostMs += batchResult.postprocessMs
+                    totalRecMs += batchResult.timeMs
+                    recInputShapes.add(batchResult.inputShape)
+                    if (batchSize == 1) {
+                        perLineRecMs.add(batchResult.timeMs)
+                    }
+
+                    for (j in batchResult.texts.indices) {
+                        val boxIdx = batchBoxIndices[j]
+                        val (text, confidence) = batchResult.texts[j]
+                        if (confidence >= config.recScoreThresh) {
+                            allResults.add(
+                                OCRResult(
+                                    box = sortedBoxes[boxIdx],
+                                    text = text,
+                                    confidence = confidence,
+                                )
+                            )
+                        }
+                    }
+                }
+            } finally {
+                batchCrops.forEach { it.release() }
+            }
+            i = next
+        }
+
+        val totalElapsed = System.currentTimeMillis() - totalStart
+        val pipelineOverhead = totalElapsed - detResult.timeMs - totalRecMs
+
+        return OCREngineResult(
+            results = allResults,
+            detectionTimeMs = detResult.timeMs,
+            recognitionTimeMs = totalRecMs,
+            totalTimeMs = totalElapsed,
+            lineCount = allResults.size,
+            detPreprocessMs = detResult.preprocessMs,
+            detInferenceMs = detResult.inferenceMs,
+            detPostprocessMs = detResult.postprocessMs,
+            recPreprocessMs = totalRecPreMs,
+            recInferenceMs = totalRecInfMs,
+            recPostprocessMs = totalRecPostMs,
+            pipelineOverheadMs = pipelineOverhead,
+            coldLoadTimeMs = ortManager.coldLoadTimeMs,
+            detInputShape = detResult.inputShape,
+            recInputShapes = recInputShapes,
+            perLineRecMs = perLineRecMs,
+        )
+    }
+
+    fun release() {
+        ortManager.release()
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/engine/OCREngineResult.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/engine/OCREngineResult.kt
new file mode 100644
index 0000000..e8a240e
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/engine/OCREngineResult.kt
@@ -0,0 +1,39 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.engine
+
+import com.paddle.ocr.model.OCRResult
+
+data class OCREngineResult(
+    val results: List<OCRResult>,
+    val detectionTimeMs: Long,
+    val recognitionTimeMs: Long,
+    val totalTimeMs: Long,
+    val lineCount: Int,
+    // Detailed per-stage timing
+    val detPreprocessMs: Long = 0,
+    val detInferenceMs: Long = 0,
+    val detPostprocessMs: Long = 0,
+    val recPreprocessMs: Long = 0,
+    val recInferenceMs: Long = 0,
+    val recPostprocessMs: Long = 0,
+    val pipelineOverheadMs: Long = 0,
+    val coldLoadTimeMs: Long = 0,
+    // Input tensor shapes
+    val detInputShape: List<Int> = emptyList(),
+    val recInputShapes: List<List<Int>> = emptyList(),
+    // Per-line timing (only populated when recBatchSize == 1)
+    val perLineRecMs: List<Long> = emptyList(),
+)
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/engine/ORTSessionManager.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/engine/ORTSessionManager.kt
new file mode 100644
index 0000000..f053104
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/engine/ORTSessionManager.kt
@@ -0,0 +1,161 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.engine
+
+import ai.onnxruntime.OnnxTensor
+import ai.onnxruntime.OrtEnvironment
+import ai.onnxruntime.OrtSession
+import android.content.Context
+import com.paddle.ocr.EngineConfig
+import com.paddle.ocr.model.OCRError
+import java.nio.FloatBuffer
+
+class ORTSessionManager(
+    private val context: Context,
+    private val config: EngineConfig,
+) {
+    private var env: OrtEnvironment? = null
+    private var detSession: OrtSession? = null
+    private var recSession: OrtSession? = null
+    private var detInputName: String = "x"
+    private var recInputName: String = "x"
+    var coldLoadTimeMs: Long = 0
+        private set
+
+    fun loadModels(detAssetPath: String, recAssetPath: String) {
+        val loadStart = System.currentTimeMillis()
+        env = OrtEnvironment.getEnvironment()
+        val opts = OrtSession.SessionOptions().apply {
+            setOptimizationLevel(OrtSession.SessionOptions.OptLevel.ALL_OPT)
+            setIntraOpNumThreads(config.numThreads)
+        }
+        try {
+            val ortEnv = env ?: throw OCRError.ModelLoadFailed("OCR", Exception("Environment not initialized"))
+            val detBytes = readModelAsset(detAssetPath)
+            val recBytes = readModelAsset(recAssetPath)
+            try {
+                detSession = ortEnv.createSession(detBytes, opts)
+            } catch (t: Throwable) {
+                throw OCRError.ModelLoadFailed("detection", t)
+            }
+            try {
+                recSession = ortEnv.createSession(recBytes, opts)
+            } catch (t: Throwable) {
+                detSession?.close()
+                detSession = null
+                throw OCRError.ModelLoadFailed("recognition", t)
+            }
+
+            detInputName = try {
+                detSession!!.inputNames.iterator().next()
+            } catch (t: Throwable) {
+                throw OCRError.ModelLoadFailed("detection", t)
+            }
+            recInputName = try {
+                recSession!!.inputNames.iterator().next()
+            } catch (t: Throwable) {
+                throw OCRError.ModelLoadFailed("recognition", t)
+            }
+            coldLoadTimeMs = System.currentTimeMillis() - loadStart
+        } finally {
+            opts.close()
+        }
+    }
+
+    fun runDetection(input: FloatArray, shape: LongArray): Pair<FloatArray, LongArray> {
+        val session = detSession
+            ?: throw OCRError.ModelLoadFailed("detection", Exception("Session not initialized"))
+        val ortEnv = env
+            ?: throw OCRError.ModelLoadFailed("detection", Exception("Environment not initialized"))
+        return runSession(ortEnv, session, detInputName, input, shape, "detection")
+    }
+
+    fun runRecognition(input: FloatArray, shape: LongArray): Pair<FloatArray, LongArray> {
+        val session = recSession
+            ?: throw OCRError.ModelLoadFailed("recognition", Exception("Session not initialized"))
+        val ortEnv = env
+            ?: throw OCRError.ModelLoadFailed("recognition", Exception("Environment not initialized"))
+        return runSession(ortEnv, session, recInputName, input, shape, "recognition")
+    }
+
+    fun release() {
+        try {
+            detSession?.close()
+        } finally {
+            detSession = null
+            try {
+                recSession?.close()
+            } finally {
+                recSession = null
+                env = null
+            }
+        }
+    }
+
+    private fun readModelAsset(assetPath: String): ByteArray {
+        return try {
+            context.assets.open(assetPath).use { it.readBytes() }
+        } catch (t: Throwable) {
+            throw OCRError.ModelNotFound(assetPath, t)
+        }
+    }
+
+    private fun runSession(
+        ortEnv: OrtEnvironment,
+        session: OrtSession,
+        inputName: String,
+        input: FloatArray,
+        shape: LongArray,
+        modelName: String,
+    ): Pair<FloatArray, LongArray> {
+        val tensor = try {
+            OnnxTensor.createTensor(ortEnv, FloatBuffer.wrap(input), shape)
+        } catch (t: Throwable) {
+            throw OCRError.InferenceFailed(modelName, t)
+        }
+        val result = try {
+            try {
+                session.run(mapOf(inputName to tensor))
+            } catch (t: Throwable) {
+                throw OCRError.InferenceFailed(modelName, t)
+            }
+        } finally {
+            tensor.close()
+        }
+
+        return try {
+            try {
+                val outputName = session.outputNames.iterator().next()
+                val ortValue = result.get(outputName)
+                    .orElseThrow { Exception("No output tensor found") }
+                val outputTensor = ortValue as? OnnxTensor
+                    ?: throw Exception("Output is not an ONNX tensor")
+                Pair(copyFloatBuffer(outputTensor.floatBuffer), outputTensor.info.shape)
+            } catch (t: Throwable) {
+                throw OCRError.InferenceFailed(modelName, t)
+            }
+        } finally {
+            result.close()
+        }
+    }
+
+    private fun copyFloatBuffer(buffer: FloatBuffer): FloatArray {
+        val duplicate = buffer.duplicate()
+        duplicate.rewind()
+        val output = FloatArray(duplicate.remaining())
+        duplicate.get(output)
+        return output
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/engine/RecognitionEngine.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/engine/RecognitionEngine.kt
new file mode 100644
index 0000000..4e6e0f3
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/engine/RecognitionEngine.kt
@@ -0,0 +1,61 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.engine
+
+import com.paddle.ocr.postprocess.CTCDecoder
+import com.paddle.ocr.preprocess.RecPreprocessor
+import org.opencv.core.Mat
+
+class RecognitionEngine(
+    private val ortManager: ORTSessionManager,
+    private val characterList: List<String>,
+) {
+    data class RecognitionResult(
+        val texts: List<Pair<String, Float>>,
+        val preprocessMs: Long,
+        val inferenceMs: Long,
+        val postprocessMs: Long,
+        val timeMs: Long,
+        val inputShape: List<Int>,
+    )
+
+    fun recognize(crops: List<Mat>): RecognitionResult {
+        // Preprocess
+        val preStart = System.currentTimeMillis()
+        val preResult = RecPreprocessor.preprocessBatch(crops)
+        val preprocessMs = System.currentTimeMillis() - preStart
+
+        // Inference
+        val infStart = System.currentTimeMillis()
+        val (outputData, outputShape) = ortManager.runRecognition(preResult.tensorData, preResult.shape)
+        val inferenceMs = System.currentTimeMillis() - infStart
+
+        // Postprocess (CTC decode)
+        val postStart = System.currentTimeMillis()
+        val decoded = CTCDecoder.decode(outputData, outputShape, characterList)
+        val postprocessMs = System.currentTimeMillis() - postStart
+
+        val inputShape = preResult.shape.map { it.toInt() }
+        val timeMs = preprocessMs + inferenceMs + postprocessMs
+        return RecognitionResult(
+            texts = decoded,
+            preprocessMs = preprocessMs,
+            inferenceMs = inferenceMs,
+            postprocessMs = postprocessMs,
+            timeMs = timeMs,
+            inputShape = inputShape,
+        )
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/model/ModelConfig.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/model/ModelConfig.kt
new file mode 100644
index 0000000..111770f
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/model/ModelConfig.kt
@@ -0,0 +1,123 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.model
+
+import android.content.Context
+import com.paddle.ocr.util.YamlUtils
+
+data class ModelConfig(
+    val characterList: List<String>,
+) {
+    companion object {
+        fun parse(context: Context, assetPath: String): ModelConfig {
+            val content = try {
+                context.assets.open(assetPath).bufferedReader().use { it.readText() }
+            } catch (t: Throwable) {
+                throw OCRError.ConfigParseFailed(assetPath, t)
+            }
+            val characterDict = try {
+                extractCharacterDict(content, assetPath)
+            } catch (e: OCRError.ConfigParseFailed) {
+                throw e
+            } catch (t: Throwable) {
+                throw OCRError.ConfigParseFailed(assetPath, t)
+            }
+            val charListWithSpace = characterDict.toMutableList().apply {
+                if (lastOrNull() != " ") add(" ")
+            }
+
+            return ModelConfig(characterList = charListWithSpace)
+        }
+
+        private fun extractCharacterDict(content: String, assetPath: String): List<String> {
+            val lines = content.replace("\r\n", "\n").replace('\r', '\n').lines()
+            val postProcessLine = lines.indexOfFirst { it.trim() == "PostProcess:" }
+            if (postProcessLine < 0) {
+                throw OCRError.ConfigParseFailed("Missing PostProcess in $assetPath")
+            }
+
+            val postProcessIndent = YamlUtils.leadingSpaces(lines[postProcessLine])
+            val characterDictLine = findCharacterDictLine(lines, postProcessLine + 1, postProcessIndent)
+            if (characterDictLine < 0) {
+                throw OCRError.ConfigParseFailed("Missing character_dict in $assetPath")
+            }
+
+            val characterDictIndent = YamlUtils.leadingSpaces(lines[characterDictLine])
+            val characters = mutableListOf<String>()
+            var lineIndex = characterDictLine + 1
+            while (lineIndex < lines.size) {
+                val line = lines[lineIndex]
+                val indent = YamlUtils.leadingSpaces(line)
+                val content = line.substring(indent)
+                val keyLikeContent = content.trim()
+                if (keyLikeContent.isEmpty() || keyLikeContent.startsWith("#")) {
+                    lineIndex++
+                    continue
+                }
+
+                if (!content.startsWith("-")) {
+                    if (indent <= characterDictIndent) break
+                    lineIndex++
+                    continue
+                }
+
+                characters.add(parseYamlListScalar(content.substring(1)))
+                lineIndex++
+            }
+
+            if (characters.isEmpty()) {
+                throw OCRError.ConfigParseFailed("Empty character_dict in $assetPath")
+            }
+            return characters
+        }
+
+        private fun findCharacterDictLine(
+            lines: List<String>,
+            startLine: Int,
+            postProcessIndent: Int,
+        ): Int {
+            var lineIndex = startLine
+            while (lineIndex < lines.size) {
+                val line = lines[lineIndex]
+                val trimmed = line.trim()
+                if (trimmed.isEmpty() || trimmed.startsWith("#")) {
+                    lineIndex++
+                    continue
+                }
+                val indent = YamlUtils.leadingSpaces(line)
+                if (indent <= postProcessIndent) break
+                if (trimmed == "character_dict:") return lineIndex
+                lineIndex++
+            }
+            return -1
+        }
+
+        private fun parseYamlListScalar(rawValue: String): String {
+            val value = rawValue.dropWhile { it == ' ' }
+            if (value.length >= 2 && value.first() == '\'' && value.last() == '\'') {
+                return value.substring(1, value.length - 1).replace("''", "'")
+            }
+            if (value.length >= 2 && value.first() == '"' && value.last() == '"') {
+                return value.substring(1, value.length - 1)
+                    .replace("\\\"", "\"")
+                    .replace("\\\\", "\\")
+                    .replace("\\n", "\n")
+                    .replace("\\r", "\r")
+                    .replace("\\t", "\t")
+            }
+            return value
+        }
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/model/OCRBox.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/model/OCRBox.kt
new file mode 100644
index 0000000..bf4885c
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/model/OCRBox.kt
@@ -0,0 +1,25 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.model
+
+import android.graphics.PointF
+
+data class OCRBox(
+    val points: List<PointF>,
+) {
+    init {
+        require(points.size == 4) { "OCRBox must have exactly 4 points, got ${points.size}" }
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/model/OCRError.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/model/OCRError.kt
new file mode 100644
index 0000000..e505ca8
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/model/OCRError.kt
@@ -0,0 +1,24 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.model
+
+sealed class OCRError(message: String, cause: Throwable? = null) : Exception(message, cause) {
+    class ModelNotFound(modelPath: String, cause: Throwable? = null) : OCRError("Model not found: $modelPath", cause)
+    class ModelLoadFailed(modelName: String, cause: Throwable) : OCRError("Failed to load $modelName model", cause)
+    class ConfigParseFailed(path: String, cause: Throwable? = null) : OCRError("Failed to parse config: $path", cause)
+    class InvalidImage : OCRError("Input image is empty or invalid")
+    class InferenceFailed(stage: String, cause: Throwable) : OCRError("Inference failed at stage '$stage'", cause)
+    class DecodeError(message: String, cause: Throwable? = null) : OCRError(message, cause)
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/model/OCRResult.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/model/OCRResult.kt
new file mode 100644
index 0000000..159204f
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/model/OCRResult.kt
@@ -0,0 +1,22 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.model
+
+data class OCRResult(
+    val box: OCRBox,
+    val text: String,
+    val confidence: Float,
+    val wordBoxes: List<OCRBox>? = null,
+)
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/model/OCRRunResult.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/model/OCRRunResult.kt
new file mode 100644
index 0000000..7b19596
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/model/OCRRunResult.kt
@@ -0,0 +1,37 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.model
+
+data class OCRRunResult(
+    val results: List<OCRResult>,
+    val detectionTimeMs: Long,
+    val recognitionTimeMs: Long,
+    val totalTimeMs: Long,
+    val lineCount: Int,
+    // Detailed per-stage timing
+    val detPreprocessMs: Long = 0,
+    val detInferenceMs: Long = 0,
+    val detPostprocessMs: Long = 0,
+    val recPreprocessMs: Long = 0,
+    val recInferenceMs: Long = 0,
+    val recPostprocessMs: Long = 0,
+    val pipelineOverheadMs: Long = 0,
+    val coldLoadTimeMs: Long = 0,
+    // Input tensor shapes
+    val detInputShape: List<Int> = emptyList(),
+    val recInputShapes: List<List<Int>> = emptyList(),
+    // Per-line timing (only populated when recBatchSize == 1)
+    val perLineRecMs: List<Long> = emptyList(),
+)
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/BoxSorter.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/BoxSorter.kt
new file mode 100644
index 0000000..72d2bd3
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/BoxSorter.kt
@@ -0,0 +1,47 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.postprocess
+
+import com.paddle.ocr.model.OCRBox
+
+object BoxSorter {
+    private const val ROW_THRESHOLD_Y = 10f
+
+    fun sortInReadingOrder(boxes: List<OCRBox>): List<OCRBox> {
+        if (boxes.size <= 1) return boxes
+
+        val list = boxes.toMutableList()
+        list.sortWith(compareBy({ it.points[0].y }, { it.points[0].x }))
+
+        // Align with PaddleX SortQuadBoxes: bubble left-to-right inside a 10px row band.
+        for (i in 0 until list.size - 1) {
+            var j = i
+            while (j >= 0) {
+                val next = list[j + 1]
+                val curr = list[j]
+                if (kotlin.math.abs(next.points[0].y - curr.points[0].y) < ROW_THRESHOLD_Y &&
+                    next.points[0].x < curr.points[0].x
+                ) {
+                    list[j] = next
+                    list[j + 1] = curr
+                    j--
+                } else {
+                    break
+                }
+            }
+        }
+        return list
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/CTCDecoder.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/CTCDecoder.kt
new file mode 100644
index 0000000..feb56d0
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/CTCDecoder.kt
@@ -0,0 +1,66 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.postprocess
+
+object CTCDecoder {
+    private const val BLANK_IDX = 0
+
+    fun decode(output: FloatArray, shape: LongArray, characterList: List<String>): List<Pair<String, Float>> {
+        val batchSize = shape[0].toInt()
+        val timeSteps = shape[1].toInt()
+        val numClasses = shape[2].toInt()
+
+        val results = mutableListOf<Pair<String, Float>>()
+        for (b in 0 until batchSize) {
+            val baseOffset = b * timeSteps * numClasses
+
+            val indices = IntArray(timeSteps)
+            val probs = FloatArray(timeSteps)
+            for (t in 0 until timeSteps) {
+                val offset = baseOffset + t * numClasses
+                var maxIdx = 0
+                var maxVal = output[offset]
+                for (c in 1 until numClasses) {
+                    val v = output[offset + c]
+                    if (v > maxVal) {
+                        maxVal = v
+                        maxIdx = c
+                    }
+                }
+                indices[t] = maxIdx
+                probs[t] = maxVal
+            }
+
+            val keptProbs = mutableListOf<Float>()
+            val sb = StringBuilder()
+            var prevIdx = -1
+            for (t in 0 until timeSteps) {
+                val idx = indices[t]
+                if (idx != BLANK_IDX && idx != prevIdx) {
+                    val charIdx = idx - 1
+                    if (charIdx >= 0 && charIdx < characterList.size) {
+                        sb.append(characterList[charIdx])
+                        keptProbs.add(probs[t])
+                    }
+                }
+                prevIdx = idx
+            }
+
+            val confidence = if (keptProbs.isNotEmpty()) keptProbs.average().toFloat() else 0f
+            results.add(Pair(sb.toString(), confidence))
+        }
+        return results
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/DBPostProcessor.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/DBPostProcessor.kt
new file mode 100644
index 0000000..78fbb2e
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/DBPostProcessor.kt
@@ -0,0 +1,189 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.postprocess
+
+import android.graphics.PointF
+import com.paddle.ocr.model.OCRBox
+import com.paddle.ocr.util.MathUtils
+import org.opencv.core.Core
+import org.opencv.core.CvType
+import org.opencv.core.Mat
+import org.opencv.core.MatOfPoint
+import org.opencv.core.MatOfPoint2f
+import org.opencv.core.Point
+import org.opencv.core.Scalar
+import org.opencv.imgproc.Imgproc
+
+object DBPostProcessor {
+    private const val MIN_SIZE_BEFORE_UNCLIP = 3f
+    private const val MIN_SIZE_AFTER_UNCLIP = 5f
+
+    fun process(
+        pred: FloatArray,
+        predShape: LongArray,
+        thresh: Float,
+        boxThresh: Float,
+        unclipRatio: Float,
+        maxCandidates: Int,
+        useDilation: Boolean,
+        scoreMode: String,
+        boxType: String,
+        originalH: Int,
+        originalW: Int,
+    ): List<OCRBox> {
+        require(boxType == "quad") { "Only DBPostProcess box_type=quad is supported" }
+
+        val pH = predShape[2].toInt()
+        val pW = predShape[3].toInt()
+        val scaleX = originalW.toDouble() / pW
+        val scaleY = originalH.toDouble() / pH
+        val normalizedScoreMode = scoreMode.lowercase()
+
+        val rawProb = Mat(pH, pW, CvType.CV_32FC1)
+        val mask = Mat(pH, pW, CvType.CV_8UC1)
+        val contours = mutableListOf<MatOfPoint>()
+        val hierarchy = Mat()
+        var contourMask = mask
+
+        return try {
+            rawProb.put(0, 0, pred)
+            val threshMat = Mat()
+            try {
+                Imgproc.threshold(rawProb, threshMat, thresh.toDouble(), 255.0, Imgproc.THRESH_BINARY)
+                threshMat.convertTo(mask, CvType.CV_8UC1)
+            } finally {
+                threshMat.release()
+            }
+
+            contourMask = if (useDilation) {
+                val kernel = Mat.ones(2, 2, CvType.CV_8UC1)
+                try {
+                    Mat().also { Imgproc.dilate(mask, it, kernel) }
+                } finally {
+                    kernel.release()
+                }
+            } else {
+                mask
+            }
+
+            Imgproc.findContours(contourMask, contours, hierarchy, Imgproc.RETR_LIST, Imgproc.CHAIN_APPROX_SIMPLE)
+            val boxes = mutableListOf<OCRBox>()
+            val numContours = minOf(contours.size, maxCandidates)
+            for (index in 0 until numContours) {
+                val contour = contours[index]
+                val contour2f = MatOfPoint2f(*contour.toArray())
+                val rect = try {
+                    Imgproc.minAreaRect(contour2f)
+                } finally {
+                    contour2f.release()
+                }
+                val sside = minOf(rect.size.width, rect.size.height)
+                if (sside < MIN_SIZE_BEFORE_UNCLIP) continue
+
+                val rectPoints = Array(4) { Point() }
+                rect.points(rectPoints)
+                val orderedPoints = QuadGeometry.orderMinAreaRectPoints(rectPoints)
+                val score = if (normalizedScoreMode == "slow") {
+                    computeBoxScore(rawProb, contour.toList())
+                } else {
+                    computeBoxScore(rawProb, orderedPoints)
+                }
+                if (score < boxThresh) continue
+
+                val expandedPts = PolygonUnclip.unclip(orderedPoints, unclipRatio)
+                val expandedRect = MatOfPoint2f().let { expanded2f ->
+                    try {
+                        expanded2f.fromList(expandedPts)
+                        Imgproc.minAreaRect(expanded2f)
+                    } finally {
+                        expanded2f.release()
+                    }
+                }
+                val sside2 = minOf(expandedRect.size.width, expandedRect.size.height)
+                if (sside2 < MIN_SIZE_AFTER_UNCLIP) continue
+
+                val expandedBox = Array(4) { Point() }
+                expandedRect.points(expandedBox)
+                val ePts = QuadGeometry.orderMinAreaRectPoints(expandedBox)
+                val scaled = listOf(
+                    scalePoint(ePts[0], scaleX, scaleY, originalW, originalH),
+                    scalePoint(ePts[1], scaleX, scaleY, originalW, originalH),
+                    scalePoint(ePts[2], scaleX, scaleY, originalW, originalH),
+                    scalePoint(ePts[3], scaleX, scaleY, originalW, originalH),
+                )
+
+                val boxW = kotlin.math.hypot(
+                    scaled[1].x - scaled[0].x,
+                    scaled[1].y - scaled[0].y,
+                )
+                val boxH = kotlin.math.hypot(
+                    scaled[3].x - scaled[0].x,
+                    scaled[3].y - scaled[0].y,
+                )
+                if (boxW <= 3 || boxH <= 3) continue
+
+                boxes.add(OCRBox(points = scaled))
+            }
+            boxes
+        } finally {
+            hierarchy.release()
+            if (contourMask !== mask) {
+                contourMask.release()
+            }
+            contours.forEach { it.release() }
+            mask.release()
+            rawProb.release()
+        }
+    }
+
+    private fun scalePoint(
+        point: Point,
+        scaleX: Double,
+        scaleY: Double,
+        originalW: Int,
+        originalH: Int,
+    ): PointF {
+        return PointF(
+            MathUtils.roundHalfToEven(point.x * scaleX).coerceIn(0, originalW).toFloat(),
+            MathUtils.roundHalfToEven(point.y * scaleY).coerceIn(0, originalH).toFloat(),
+        )
+    }
+
+    private fun computeBoxScore(probMap: Mat, points: List<Point>): Float {
+        if (points.isEmpty()) return 0f
+
+        val h = probMap.rows()
+        val w = probMap.cols()
+        val xmin = points.minOf { kotlin.math.floor(it.x).toInt() }.coerceIn(0, w - 1)
+        val xmax = points.maxOf { kotlin.math.ceil(it.x).toInt() }.coerceIn(0, w - 1)
+        val ymin = points.minOf { kotlin.math.floor(it.y).toInt() }.coerceIn(0, h - 1)
+        val ymax = points.maxOf { kotlin.math.ceil(it.y).toInt() }.coerceIn(0, h - 1)
+
+        val mask = Mat(ymax - ymin + 1, xmax - xmin + 1, CvType.CV_8UC1, Scalar(0.0))
+        val pts = MatOfPoint().apply {
+            fromList(
+                points.map { Point((it.x - xmin).toInt().toDouble(), (it.y - ymin).toInt().toDouble()) }
+            )
+        }
+        Imgproc.fillPoly(mask, mutableListOf(pts), Scalar(1.0))
+
+        val roi = probMap.submat(ymin, ymax + 1, xmin, xmax + 1)
+        val mean = Core.mean(roi, mask)
+        roi.release()
+        mask.release()
+        pts.release()
+        return mean.`val`[0].toFloat()
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/PolygonUnclip.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/PolygonUnclip.kt
new file mode 100644
index 0000000..d4d0b4b
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/PolygonUnclip.kt
@@ -0,0 +1,138 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.postprocess
+
+import org.opencv.core.Point
+import kotlin.math.PI
+import kotlin.math.abs
+import kotlin.math.acos
+import kotlin.math.atan2
+import kotlin.math.ceil
+import kotlin.math.cos
+import kotlin.math.hypot
+import kotlin.math.sin
+
+internal object PolygonUnclip {
+    private const val EPS = 1e-6
+    private const val ARC_TOLERANCE = 0.25
+
+    fun unclip(points: List<Point>, unclipRatio: Float): List<Point> {
+        if (points.size < 3) return points
+
+        val signedArea = signedArea(points)
+        val area = abs(signedArea)
+        val perimeter = perimeter(points)
+        if (!area.isFinite() || !perimeter.isFinite() || area <= EPS || perimeter <= EPS) {
+            return points
+        }
+
+        val distance = area * unclipRatio / perimeter
+        if (!distance.isFinite() || distance <= EPS) return points
+
+        val clockwiseInImageCoords = signedArea > 0.0
+        val normals = mutableListOf<Point>()
+        for (i in points.indices) {
+            val start = points[i]
+            val end = points[(i + 1) % points.size]
+            val dx = end.x - start.x
+            val dy = end.y - start.y
+            val len = hypot(dx, dy)
+            if (!len.isFinite() || len <= EPS) return points
+
+            val normal = if (clockwiseInImageCoords) {
+                Point(dy / len, -dx / len)
+            } else {
+                Point(-dy / len, dx / len)
+            }
+            normals.add(normal)
+        }
+
+        val expanded = mutableListOf<Point>()
+        for (i in points.indices) {
+            appendRoundJoin(
+                output = expanded,
+                center = points[i],
+                fromNormal = normals[(i - 1 + normals.size) % normals.size],
+                toNormal = normals[i],
+                distance = distance,
+                clockwiseInImageCoords = clockwiseInImageCoords,
+            )
+        }
+        return if (expanded.size >= 3) expanded else points
+    }
+
+    private fun appendRoundJoin(
+        output: MutableList<Point>,
+        center: Point,
+        fromNormal: Point,
+        toNormal: Point,
+        distance: Double,
+        clockwiseInImageCoords: Boolean,
+    ) {
+        var startAngle = atan2(fromNormal.y, fromNormal.x)
+        var endAngle = atan2(toNormal.y, toNormal.x)
+        if (clockwiseInImageCoords) {
+            while (endAngle < startAngle) endAngle += 2.0 * PI
+        } else {
+            while (endAngle > startAngle) endAngle -= 2.0 * PI
+        }
+
+        val sweep = endAngle - startAngle
+        val steps = ceil(abs(sweep) / arcStepAngle(distance)).toInt().coerceAtLeast(1)
+        for (step in 0..steps) {
+            val angle = startAngle + sweep * step.toDouble() / steps.toDouble()
+            appendPoint(
+                output,
+                Point(
+                    center.x + cos(angle) * distance,
+                    center.y + sin(angle) * distance,
+                )
+            )
+        }
+    }
+
+    private fun arcStepAngle(distance: Double): Double {
+        val ratio = (1.0 - ARC_TOLERANCE / distance).coerceIn(-1.0, 1.0)
+        val step = 2.0 * acos(ratio)
+        return if (step.isFinite() && step > EPS) step else PI / 8.0
+    }
+
+    private fun appendPoint(output: MutableList<Point>, point: Point) {
+        val last = output.lastOrNull()
+        if (last == null || hypot(point.x - last.x, point.y - last.y) > EPS) {
+            output.add(point)
+        }
+    }
+
+    private fun signedArea(points: List<Point>): Double {
+        var sum = 0.0
+        for (i in points.indices) {
+            val p1 = points[i]
+            val p2 = points[(i + 1) % points.size]
+            sum += p1.x * p2.y - p2.x * p1.y
+        }
+        return sum / 2.0
+    }
+
+    private fun perimeter(points: List<Point>): Double {
+        var length = 0.0
+        for (i in points.indices) {
+            val p1 = points[i]
+            val p2 = points[(i + 1) % points.size]
+            length += hypot(p2.x - p1.x, p2.y - p1.y)
+        }
+        return length
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/QuadGeometry.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/QuadGeometry.kt
new file mode 100644
index 0000000..1dee6ca
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/QuadGeometry.kt
@@ -0,0 +1,43 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.postprocess
+
+import org.opencv.core.Point
+
+internal object QuadGeometry {
+    fun orderMinAreaRectPoints(points: Array<Point>): List<Point> {
+        val sorted = points.sortedBy { it.x }
+        val topLeft: Point
+        val bottomLeft: Point
+        if (sorted[1].y > sorted[0].y) {
+            topLeft = sorted[0]
+            bottomLeft = sorted[1]
+        } else {
+            topLeft = sorted[1]
+            bottomLeft = sorted[0]
+        }
+
+        val topRight: Point
+        val bottomRight: Point
+        if (sorted[3].y > sorted[2].y) {
+            topRight = sorted[2]
+            bottomRight = sorted[3]
+        } else {
+            topRight = sorted[3]
+            bottomRight = sorted[2]
+        }
+        return listOf(topLeft, topRight, bottomRight, bottomLeft)
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/QuadTextCrop.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/QuadTextCrop.kt
new file mode 100644
index 0000000..e8b5bb2
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/postprocess/QuadTextCrop.kt
@@ -0,0 +1,86 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.postprocess
+
+import com.paddle.ocr.model.OCRBox
+import org.opencv.core.Core
+import org.opencv.core.CvType
+import org.opencv.core.Mat
+import org.opencv.core.MatOfPoint2f
+import org.opencv.core.Point
+import org.opencv.core.Size
+import org.opencv.imgproc.Imgproc
+import kotlin.math.hypot
+import kotlin.math.max
+
+object QuadTextCrop {
+    private const val VERTICAL_CROP_RATIO = 1.5
+
+    fun crop(src: Mat, box: OCRBox): Mat {
+        // Align with PaddleX CropByPolys.get_minarea_rect_crop: recompute minAreaRect
+        // from the detected quad before perspective transform.
+        val rectPoints = box.points.map { Point(it.x.toDouble(), it.y.toDouble()) }
+        val rectInput = MatOfPoint2f()
+        rectInput.fromList(rectPoints)
+        val boundingBox = Imgproc.minAreaRect(rectInput)
+        rectInput.release()
+
+        val boxPoints = Array(4) { Point() }
+        boundingBox.points(boxPoints)
+        val ordered = QuadGeometry.orderMinAreaRectPoints(boxPoints)
+
+        val widthTop = hypot(ordered[0].x - ordered[1].x, ordered[0].y - ordered[1].y)
+        val widthBottom = hypot(ordered[2].x - ordered[3].x, ordered[2].y - ordered[3].y)
+        val heightLeft = hypot(ordered[0].x - ordered[3].x, ordered[0].y - ordered[3].y)
+        val heightRight = hypot(ordered[1].x - ordered[2].x, ordered[1].y - ordered[2].y)
+
+        val dstW = max(widthTop, widthBottom).toInt().coerceAtLeast(1)
+        val dstH = max(heightLeft, heightRight).toInt().coerceAtLeast(1)
+
+        val srcPts = MatOfPoint2f()
+        srcPts.fromList(ordered)
+        val dstPts = MatOfPoint2f()
+        dstPts.fromList(
+            listOf(
+                Point(0.0, 0.0),
+                Point(dstW.toDouble(), 0.0),
+                Point(dstW.toDouble(), dstH.toDouble()),
+                Point(0.0, dstH.toDouble()),
+            )
+        )
+        val m = Imgproc.getPerspectiveTransform(srcPts, dstPts)
+        srcPts.release()
+        dstPts.release()
+
+        val dst = Mat(dstH, dstW, CvType.CV_8UC3)
+        Imgproc.warpPerspective(
+            src,
+            dst,
+            m,
+            Size(dstW.toDouble(), dstH.toDouble()),
+            Imgproc.INTER_CUBIC,
+            Core.BORDER_REPLICATE,
+        )
+        m.release()
+
+        if (dst.rows().toDouble() / dst.cols() >= VERTICAL_CROP_RATIO) {
+            val rotated = Mat()
+            Core.rotate(dst, rotated, Core.ROTATE_90_COUNTERCLOCKWISE)
+            dst.release()
+            return rotated
+        }
+        return dst
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/preprocess/DetPreprocessor.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/preprocess/DetPreprocessor.kt
new file mode 100644
index 0000000..7fd9d24
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/preprocess/DetPreprocessor.kt
@@ -0,0 +1,103 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.preprocess
+
+import android.graphics.Bitmap
+import com.paddle.ocr.util.BitmapUtils
+import com.paddle.ocr.util.ImageUtils
+import org.opencv.core.Core
+import org.opencv.core.CvType
+import org.opencv.core.Mat
+import org.opencv.core.Scalar
+import org.opencv.imgproc.Imgproc
+
+data class DetPreprocessResult(
+    val tensorData: FloatArray,
+    val shape: LongArray,
+    val originalH: Int,
+    val originalW: Int,
+)
+
+object DetPreprocessor {
+    private val mean = doubleArrayOf(0.485, 0.456, 0.406)
+    private val std = doubleArrayOf(0.229, 0.224, 0.225)
+    private const val scale = 1.0 / 255.0
+
+    fun preprocess(
+        bitmap: Bitmap,
+        limitSideLen: Int,
+        limitType: String,
+        maxSideLimit: Int,
+        imgMode: String,
+    ): DetPreprocessResult {
+        val src = BitmapUtils.bitmapToBGRMat(bitmap)
+        return try {
+            preprocess(src, limitSideLen, limitType, maxSideLimit, imgMode)
+        } finally {
+            src.release()
+        }
+    }
+
+    fun preprocess(
+        src: Mat,
+        limitSideLen: Int,
+        limitType: String,
+        maxSideLimit: Int,
+        imgMode: String,
+    ): DetPreprocessResult {
+        val originalH = src.rows()
+        val originalW = src.cols()
+        val input = if (imgMode.uppercase() == "RGB") {
+            Mat().also { Imgproc.cvtColor(src, it, Imgproc.COLOR_BGR2RGB) }
+        } else {
+            src
+        }
+
+        val resized = ImageUtils.resizeToMultipleOf32(input, limitSideLen, limitType, maxSideLimit)
+        if (input !== src) input.release()
+
+        val h = resized.rows()
+        val w = resized.cols()
+        val floatMat = Mat(h, w, CvType.CV_32FC3)
+        resized.convertTo(floatMat, CvType.CV_32F)
+
+        val channels = mutableListOf<Mat>()
+        Core.split(floatMat, channels)
+        for (c in 0..2) {
+            Core.multiply(channels[c], Scalar(scale), channels[c])
+            Core.subtract(channels[c], Scalar(mean[c]), channels[c])
+            Core.divide(channels[c], Scalar(std[c]), channels[c])
+        }
+        resized.release()
+        floatMat.release()
+
+        val channelSize = h * w
+        val tensorData = FloatArray(3 * channelSize)
+        for (c in 0..2) {
+            val chan = channels[c]
+            val buf = FloatArray(channelSize)
+            chan.get(0, 0, buf)
+            System.arraycopy(buf, 0, tensorData, c * channelSize, channelSize)
+            chan.release()
+        }
+
+        return DetPreprocessResult(
+            tensorData = tensorData,
+            shape = longArrayOf(1, 3, h.toLong(), w.toLong()),
+            originalH = originalH,
+            originalW = originalW,
+        )
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/preprocess/RecPreprocessor.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/preprocess/RecPreprocessor.kt
new file mode 100644
index 0000000..648be60
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/preprocess/RecPreprocessor.kt
@@ -0,0 +1,105 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.preprocess
+
+import org.opencv.core.Core
+import org.opencv.core.CvType
+import org.opencv.core.Mat
+import org.opencv.core.Size
+import org.opencv.imgproc.Imgproc
+import kotlin.math.ceil
+
+data class RecPreprocessResult(
+    val tensorData: FloatArray,
+    val shape: LongArray,
+)
+
+object RecPreprocessor {
+    private const val FIXED_HEIGHT = 48
+    private const val MAX_IMG_W = 3200
+
+    fun preprocessBatch(crops: List<Mat>): RecPreprocessResult {
+        // Convert BGR to RGB and resize to fixed height while preserving aspect ratio
+        val resizedMats = mutableListOf<Mat>()
+        for (crop in crops) {
+            // Convert BGR to RGB (model expects RGB input)
+            val rgb = Mat()
+            Imgproc.cvtColor(crop, rgb, Imgproc.COLOR_BGR2RGB)
+            val h = rgb.rows()
+            val w = rgb.cols()
+            val aspectRatio = if (h > 0) w.toDouble() / h else 1.0
+            val newW = ceil(FIXED_HEIGHT * aspectRatio).toInt().coerceAtMost(MAX_IMG_W)
+            val dst = Mat()
+            Imgproc.resize(rgb, dst, Size(newW.toDouble(), FIXED_HEIGHT.toDouble()), 0.0, 0.0, Imgproc.INTER_LINEAR)
+            rgb.release()
+            resizedMats.add(dst)
+        }
+
+        // Convert to float and normalize: (x / 255 - 0.5) / 0.5 = x / 127.5 - 1
+        val floatMats = mutableListOf<Mat>()
+        for (mat in resizedMats) {
+            val floatMat = Mat(mat.rows(), mat.cols(), CvType.CV_32FC3)
+            mat.convertTo(floatMat, CvType.CV_32F)
+            // Use Scalar with all 3 channels set — single-value Scalar only sets val[0]!
+            Core.divide(floatMat, org.opencv.core.Scalar(127.5, 127.5, 127.5), floatMat)
+            Core.subtract(floatMat, org.opencv.core.Scalar(1.0, 1.0, 1.0), floatMat)
+
+            floatMats.add(floatMat)
+            mat.release()  // Release resized mat
+        }
+        resizedMats.clear()
+
+        val maxW = floatMats.maxOf { it.cols() }
+        val n = floatMats.size
+
+        // Pad to max width
+        val paddedMats = mutableListOf<Mat>()
+        for (mat in floatMats) {
+            if (mat.cols() == maxW) {
+                paddedMats.add(mat)
+            } else {
+                val padded = Mat(FIXED_HEIGHT, maxW, CvType.CV_32FC3, org.opencv.core.Scalar(0.0))
+                val roi = padded.submat(0, FIXED_HEIGHT, 0, mat.cols())
+                mat.copyTo(roi)
+                roi.release()
+                mat.release()
+                paddedMats.add(padded)
+            }
+        }
+        floatMats.clear()
+
+        // Build tensor data
+        val channelSize = FIXED_HEIGHT * maxW
+        val tensorData = FloatArray(n * 3 * channelSize)
+        for (b in 0 until n) {
+            val mat = paddedMats[b]
+            val channels = mutableListOf<Mat>()
+            Core.split(mat, channels)
+            for (c in 0..2) {
+                val buf = FloatArray(channelSize)
+                channels[c].get(0, 0, buf)
+                System.arraycopy(buf, 0, tensorData, (b * 3 + c) * channelSize, channelSize)
+                channels[c].release()
+            }
+            mat.release()
+        }
+        paddedMats.clear()
+
+        return RecPreprocessResult(
+            tensorData = tensorData,
+            shape = longArrayOf(n.toLong(), 3, FIXED_HEIGHT.toLong(), maxW.toLong()),
+        )
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/util/BitmapUtils.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/util/BitmapUtils.kt
new file mode 100644
index 0000000..a504e70
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/util/BitmapUtils.kt
@@ -0,0 +1,72 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.util
+
+import android.graphics.Bitmap
+import org.opencv.android.Utils
+import org.opencv.core.CvType
+import org.opencv.core.Mat
+import org.opencv.core.MatOfByte
+import org.opencv.imgcodecs.Imgcodecs
+import org.opencv.imgproc.Imgproc
+
+object BitmapUtils {
+
+    fun imdecodeBGR(imageBytes: ByteArray): Mat {
+        val encoded = MatOfByte(*imageBytes)
+        return try {
+            Imgcodecs.imdecode(encoded, Imgcodecs.IMREAD_COLOR)
+        } finally {
+            encoded.release()
+        }
+    }
+
+    fun bitmapToBGRMat(bitmap: Bitmap): Mat {
+        return bitmapToMat(bitmap, Imgproc.COLOR_RGBA2BGR)
+    }
+
+    fun bitmapToRGBMat(bitmap: Bitmap): Mat {
+        return bitmapToMat(bitmap, Imgproc.COLOR_RGBA2RGB)
+    }
+
+    fun bgrMatToBitmap(mat: Mat): Bitmap {
+        val rgba = Mat()
+        return try {
+            Imgproc.cvtColor(mat, rgba, Imgproc.COLOR_BGR2RGBA)
+            Bitmap.createBitmap(rgba.cols(), rgba.rows(), Bitmap.Config.ARGB_8888).also { bitmap ->
+                Utils.matToBitmap(rgba, bitmap)
+            }
+        } finally {
+            rgba.release()
+        }
+    }
+
+    private fun bitmapToMat(bitmap: Bitmap, colorConversionCode: Int): Mat {
+        val bmp = bitmap.copy(Bitmap.Config.ARGB_8888, false)
+        val rgba = Mat(bmp.height, bmp.width, CvType.CV_8UC4)
+        val dst = Mat()
+        return try {
+            Utils.bitmapToMat(bmp, rgba)
+            Imgproc.cvtColor(rgba, dst, colorConversionCode)
+            dst
+        } catch (t: Throwable) {
+            dst.release()
+            throw t
+        } finally {
+            bmp.recycle()
+            rgba.release()
+        }
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/util/ImageUtils.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/util/ImageUtils.kt
new file mode 100644
index 0000000..c672a64
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/util/ImageUtils.kt
@@ -0,0 +1,52 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.util
+
+import org.opencv.core.Mat
+import org.opencv.core.Size
+import org.opencv.imgproc.Imgproc
+
+object ImageUtils {
+
+    fun resizeToMultipleOf32(
+        src: Mat,
+        limitSideLen: Int,
+        limitType: String,
+        maxSideLimit: Int,
+    ): Mat {
+        val h = src.rows()
+        val w = src.cols()
+        var ratio = when (limitType.lowercase()) {
+            "max" -> if (maxOf(h, w) > limitSideLen) limitSideLen.toDouble() / maxOf(h, w) else 1.0
+            "min" -> if (minOf(h, w) < limitSideLen) limitSideLen.toDouble() / minOf(h, w) else 1.0
+            "resize_long" -> limitSideLen.toDouble() / maxOf(h, w)
+            else -> throw IllegalArgumentException("Unsupported det limit type: $limitType")
+        }
+
+        var newH = (h * ratio).toInt()
+        var newW = (w * ratio).toInt()
+        if (maxOf(newH, newW) > maxSideLimit) {
+            ratio = maxSideLimit.toDouble() / maxOf(newH, newW)
+            newH = (newH * ratio).toInt()
+            newW = (newW * ratio).toInt()
+        }
+
+        newH = maxOf(MathUtils.roundHalfToEven(newH / 32.0) * 32, 32)
+        newW = maxOf(MathUtils.roundHalfToEven(newW / 32.0) * 32, 32)
+        val dst = Mat()
+        Imgproc.resize(src, dst, Size(newW.toDouble(), newH.toDouble()), 0.0, 0.0, Imgproc.INTER_LINEAR)
+        return dst
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/util/MathUtils.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/util/MathUtils.kt
new file mode 100644
index 0000000..bbee6ab
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/util/MathUtils.kt
@@ -0,0 +1,31 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.util
+
+import kotlin.math.floor
+
+object MathUtils {
+
+    fun roundHalfToEven(value: Double): Int {
+        val floored = floor(value)
+        val diff = value - floored
+        return when {
+            diff < 0.5 -> floored.toInt()
+            diff > 0.5 -> floored.toInt() + 1
+            floored.toInt() % 2 == 0 -> floored.toInt()
+            else -> floored.toInt() + 1
+        }
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/util/OpenCVUtils.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/util/OpenCVUtils.kt
new file mode 100644
index 0000000..77aa455
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/util/OpenCVUtils.kt
@@ -0,0 +1,34 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.util
+
+import android.content.Context
+import android.util.Log
+
+object OpenCVUtils {
+
+    private var initialized = false
+
+    fun init(context: Context): Boolean {
+        if (initialized) return true
+        try {
+            System.loadLibrary("opencv_java4")
+            initialized = true
+        } catch (e: UnsatisfiedLinkError) {
+            Log.e("OpenCVUtils", "Failed to initialize OpenCV: ${e.message}")
+        }
+        return initialized
+    }
+}
diff --git a/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/util/YamlUtils.kt b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/util/YamlUtils.kt
new file mode 100644
index 0000000..8f066e6
--- /dev/null
+++ b/deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/util/YamlUtils.kt
@@ -0,0 +1,22 @@
+// Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package com.paddle.ocr.util
+
+object YamlUtils {
+
+    fun leadingSpaces(line: String): Int {
+        return line.indexOfFirst { it != ' ' }.let { if (it < 0) line.length else it }
+    }
+}
diff --git a/deploy/ppocr-android/run_benchmark.sh b/deploy/ppocr-android/run_benchmark.sh
new file mode 100644
index 0000000..0b54393
--- /dev/null
+++ b/deploy/ppocr-android/run_benchmark.sh
@@ -0,0 +1,39 @@
+#!/bin/bash
+# PP-OCRv6 Android Speed Benchmark
+# Usage: JAVA_HOME=/opt/android-jdk ./run_benchmark.sh [iterations] [warmup]
+#   e.g. JAVA_HOME=/opt/android-jdk ./run_benchmark.sh 20 5
+#   defaults: 10 iterations, 3 warmup
+
+set -e
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+cd "$SCRIPT_DIR"
+
+ITERATIONS="${1:-50}"
+WARMUP="${2:-30}"
+
+echo "========================================"
+echo "  PP-OCRv6 Android Speed Benchmark"
+echo "========================================"
+echo "  Warmup:   ${WARMUP}"
+echo "  Iterations: ${ITERATIONS}"
+echo "========================================"
+echo ""
+
+adb logcat -c 2>/dev/null || true
+
+echo "[1/2] Running speed benchmark (warmup + ${ITERATIONS} measured iterations)..."
+echo "----------------------------------------"
+./gradlew :ppocr-sdk:connectedAndroidTest \
+    -Pandroid.testInstrumentationRunnerArguments.class=com.paddle.ocr.benchmark.OCRBenchmarkTest#testLatencyBenchmark \
+    -Pandroid.testInstrumentationRunnerArguments.warmup="${WARMUP}" \
+    -Pandroid.testInstrumentationRunnerArguments.iterations="${ITERATIONS}" \
+    2>&1
+echo "----------------------------------------"
+echo ""
+
+echo "[2/2] Speed benchmark results:"
+echo "========================================"
+adb logcat -d -s System.out:I 2>/dev/null | grep "OCRBenchmark" | sed 's/^.*System.out/I/'
+echo "========================================"
+echo ""
+echo "  Done"
diff --git a/deploy/ppocr-android/settings.gradle.kts b/deploy/ppocr-android/settings.gradle.kts
new file mode 100644
index 0000000..7210b45
--- /dev/null
+++ b/deploy/ppocr-android/settings.gradle.kts
@@ -0,0 +1,29 @@
+pluginManagement {
+    repositories {
+        maven { url = uri("https://maven.aliyun.com/repository/google") }
+        maven { url = uri("https://maven.aliyun.com/repository/central") }
+        maven { url = uri("https://maven.aliyun.com/repository/gradle-plugin") }
+        google {
+            content {
+                includeGroupByRegex("com\\.android.*")
+                includeGroupByRegex("com\\.google.*")
+                includeGroupByRegex("androidx.*")
+            }
+        }
+        mavenCentral()
+        gradlePluginPortal()
+    }
+}
+dependencyResolutionManagement {
+    repositoriesMode.set(RepositoriesMode.FAIL_ON_PROJECT_REPOS)
+    repositories {
+        maven { url = uri("https://maven.aliyun.com/repository/google") }
+        maven { url = uri("https://maven.aliyun.com/repository/central") }
+        google()
+        mavenCentral()
+    }
+}
+
+rootProject.name = "ppocr-android"
+include(":app")
+include(":ppocr-sdk")
diff --git a/deploy/slim/auto_compression/README.md b/deploy/slim/auto_compression/README.md
new file mode 100644
index 0000000..70532eb
--- /dev/null
+++ b/deploy/slim/auto_compression/README.md
@@ -0,0 +1,303 @@
+# OCR模型自动压缩示例
+
+目录：
+- [OCR模型自动压缩示例](#ocr模型自动压缩示例)
+  - [1. 简介](#1-简介)
+  - [2. Benchmark](#2-benchmark)
+    - [PPOCRV4\_det](#ppocrv4_det)
+    - [PPOCRV4\_rec](#ppocrv4_rec)
+  - [3. 自动压缩流程](#3-自动压缩流程)
+    - [3.1 准备环境](#31-准备环境)
+    - [3.2 准备数据集](#32-准备数据集)
+      - [3.2.1 PPOCRV4\_det\_server数据集预处理](#321-ppocrv4_det_server数据集预处理)
+    - [3.3 准备预测模型](#33-准备预测模型)
+  - [4.预测部署](#4预测部署)
+      - [4.1 Paddle Inference 验证性能](#41-paddle-inference-验证性能)
+        - [4.1.1 使用测试脚本进行批量测试：](#411-使用测试脚本进行批量测试)
+        - [4.1.2 基于压缩模型进行基于GPU的批量测试：](#412-基于压缩模型进行基于gpu的批量测试)
+        - [4.1.3 基于压缩前模型进行基于GPU的批量测试：](#413-基于压缩前模型进行基于gpu的批量测试)
+        - [4.1.4 基于压缩模型进行基于CPU的批量测试：](#414-基于压缩模型进行基于cpu的批量测试)
+    - [4.2 PaddleLite端侧部署](#42-paddlelite端侧部署)
+  - [5.FAQ](#5faq)
+    - [5.1 报错找不到模型文件或者数据集文件](#51-报错找不到模型文件或者数据集文件)
+    - [5.2 软件环境一致，硬件不同导致精度差异很大？](#52-软件环境一致硬件不同导致精度差异很大)
+
+
+## 1. 简介
+本示例将以图像分类模型PPOCRV3为例，介绍如何使用PaddleOCR中Inference部署模型进行自动压缩。本示例使用的自动压缩策略为量化训练和蒸馏。
+
+## 2. Benchmark
+
+### PPOCRV4_det
+| 模型 | 策略 | Metric(hmean) | GPU 耗时(ms) | ARM CPU 耗时(ms) | 配置文件 | Inference模型 |
+|:------:|:------:|:------:|:------:|:------:|:------:|:------:|
+| PP-OCRv4_mobile_det | Baseline | 72.71 | 5.7 | 92.0 | - | [Model](https://paddle-ocr-models.bj.bcebos.com/ppocrv4_qat/det/ch_PP-OCRv4_det_infer.tar) |
+| PP-OCRv4_mobile_det | 量化+蒸馏 | 71.10 | 2.3 | 94.1 | [Config](./configs/ppocrv4/ppocrv4_det_qat_dist.yaml) | [Model](https://paddle-ocr-models.bj.bcebos.com/ppocrv4_qat/det/det_mobile_qat_3090.zip) |
+| PP-OCRv4_server_det | Baseline | 79.82 | 32.6 | 844.7 | - | [Model](https://paddle-ocr-models.bj.bcebos.com/ppocrv4_qat/det/ch_PP-OCRv4_det_server_infer.tar) |
+| PP-OCRv4_server_det | 量化+蒸馏 | 79.27 | 12.3 | 635.0 | [Config](./configs/ppocrv4/ppocrv4_rec_server_qat_dist.yaml) | [Model](https://paddle-ocr-models.bj.bcebos.com/ppocrv4_qat/det/det_server_qat_3090.zip) |
+> - GPU测试环境：RTX 3090, cuda11.7+tensorrt8.4.2.4+paddle2.5
+> - CPU测试环境：Intel(R) Xeon(R) Gold 6226R，使用12线程测试
+> - PP-OCRv4_server_det在不完整的数据集上测试，数据处理流程参考[ppocrv4_det_server数据集预处理](#321-ppocrv4_det_server数据集预处理)，仅为了展示自动压缩效果，指标并不具有参考性，模型真实表现请参考[PPOCRV4介绍](../../../doc/doc_ch/PP-OCRv4_introduction.md)
+
+| 模型 | 策略 | Metric(hmean) | GPU 耗时(ms) | ARM CPU 耗时(ms) | 配置文件 | Inference模型 |
+|:------:|:------:|:------:|:------:|:------:|:------:|:------:|
+| PP-OCRv4_mobile_det | Baseline | 72.71 | 4.7 | 198.4 | - | [Model](https://paddle-ocr-models.bj.bcebos.com/ppocrv4_qat/det/ch_PP-OCRv4_det_infer.tar) |
+| PP-OCRv4_mobile_det | 量化+蒸馏 | 71.38 | 3.3 | 205.2 | [Config](./configs/ppocrv4/ppocrv4_det_qat_dist.yaml) | [Model](https://paddle-ocr-models.bj.bcebos.com/ppocrv4_qat/det/det_server_qat_v100.zip) |
+| PP-OCRv4_server_det | Baseline | 79.77 | 50.0 | 2159.4 | - | [Model](https://paddle-ocr-models.bj.bcebos.com/ppocrv4_qat/det/ch_PP-OCRv4_det_server_infer.tar) |
+| PP-OCRv4_server_det | 量化+蒸馏 | 79.81 | 42.4 | 1834.8 | [Config](./configs/ppocrv4/ppocrv4_rec_server_qat_dist.yaml) | [Model](https://paddle-ocr-models.bj.bcebos.com/ppocrv4_qat/det/det_server_qat_v100.zip) |
+> - GPU测试环境：Tesla V100, cuda11.7+tensorrt8.4.2.4+paddle2.5.2
+> - CPU测试环境：Intel(R) Xeon(R) Gold 6271C，使用12线程测试
+> - PP-OCRv4_server_det在不完整的数据集上测试，数据处理流程参考[ppocrv4_det_server数据集预处理](#321-ppocrv4_det_server数据集预处理)，仅为了展示自动压缩效果，指标并不具有参考性，模型真实表现请参考[PPOCRV4介绍](../../../doc/doc_ch/PP-OCRv4_introduction.md)
+
+### PPOCRV4_rec
+| 模型 | 策略 | Metric(accuracy) | GPU 耗时(ms) | ARM CPU 耗时(ms) | 配置文件 | Inference模型 |
+|:------:|:------:|:------:|:------:|:------:|:------:|:------:|
+| 中文PPOCRV4-rec_mobile | Baseline | 78.92 | 1.7 | 33.3 | - | [Model](https://paddle-ocr-models.bj.bcebos.com/ppocrv4_qat/rec/ch_PP-OCRv4_rec_infer.tar.gz) |
+| 中文PPOCRV4-rec_mobile | 量化+蒸馏 | 78.41 | 1.4 | 34.0 | [Config](./configs/ppocrv4/ppocrv4_rec_qat_dist.yaml) | [Model](https://paddle-ocr-models.bj.bcebos.com/ppocrv4_qat/rec/rec_mobile_qat.tar.gz) |
+| 中文PPOCRV4-rec_server | Baseline | 81.62 | 4.0 | 62.5 | - | [Model](https://paddle-ocr-models.bj.bcebos.com/ppocrv4_qat/rec/ch_PP-OCRv4_rec_server_infer.tar.gz) |
+| 中文PPOCRV4-rec_server | 量化+蒸馏 | 81.03 | 2.0 | 64.4 | [Config](./configs/ppocrv4/ppocrv4_rec_server_qat_dist.yaml) | [Model](https://paddle-ocr-models.bj.bcebos.com/ppocrv4_qat/rec/rec_server_qat.tar.gz) |
+> - GPU测试环境：Tesla V100, cuda11.2+tensorrt8.0.3.4+paddle2.5
+> - CPU测试环境：Intel(R) Xeon(R) Gold 6271C，使用12线程测试
+
+
+## 3. 自动压缩流程
+
+### 3.1 准备环境
+
+- PaddlePaddle == 2.5 （可从[Paddle官网](https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/install/pip/linux-pip.html)下载安装）
+- PaddleSlim == 2.5
+- PaddleOCR == develop
+
+安装paddlepaddle：
+```shell
+# CPU
+python -m pip install paddlepaddle==2.5.1 -i https://pypi.tuna.tsinghua.edu.cn/simple
+# GPU 以Ubuntu、CUDA 10.2为例
+python -m pip install paddlepaddle-gpu==2.5.1.post102 -f https://www.paddlepaddle.org.cn/whl/linux/mkl/avx/stable.html
+```
+
+安装paddleslim 2.5：
+```shell
+pip install paddleslim@git+https://gitee.com/paddlepaddle/PaddleSlim.git@release/2.5
+```
+
+安装其他依赖：
+```shell
+pip install scikit-image imgaug
+```
+
+
+下载PaddleOCR:
+```shell
+git clone -b release/2.7 https://github.com/PaddlePaddle/PaddleOCR.git
+cd PaddleOCR/
+pip install -r requirements.txt
+```
+
+### 3.2 准备数据集
+公开数据集可参考[OCR数据集](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.7/doc/doc_ch/dataset/ocr_datasets.md)，然后根据程序运行过程中提示放置到对应位置。
+
+#### 3.2.1 PPOCRV4_det_server数据集预处理
+PPOCRV4_det_server在使用原始数据集推理时，默认将输入图像的最小边缩放到736，然而原始数据集中存在一些长宽比很大的图像，比如13:1，此时再进行缩放就会导致长边的尺寸非常大，在实验过程中发现最大的长边尺寸有10000+，这导致在构建TensorRT子图的时候显存不足。
+
+为了能顺利跑通自动压缩的流程，展示自动压缩的效果，因此需要对原始数据集进行预处理，将长宽比过大的图像进行剔除，处理脚本可见[ppocrv4_det_server_dataset_process.py](./ppocrv4_det_server_dataset_process.py)。
+
+
+> 注意：使用不同的数据集需要修改配置文件中`dataset`中数据路径和数据处理部分。
+
+### 3.3 准备预测模型
+预测模型的格式为：`model.pdmodel` 和 `model.pdiparams`两个，带`pdmodel`的是模型文件，带`pdiparams`后缀的是权重文件。
+
+> 注：其他像`__model__`和`__params__`分别对应`model.pdmodel` 和 `model.pdiparams`文件。
+
+可在[PaddleOCR模型库](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.7/doc/doc_ch/models_list.md)中直接获取Inference模型，具体可参考下方获取中文PPOCRV4模型示例：
+
+```shell
+https://paddleocr.bj.bcebos.com/PP-OCRv4/chinese/ch_PP-OCRv4_rec_infer.tar
+tar -xf ch_PP-OCRv4_rec_infer.tar
+```
+
+```shell
+wget https://paddleocr.bj.bcebos.com/PP-OCRv4/chinese/ch_PP-OCRv4_det_infer.tar
+tar -xf ch_PP-OCRv4_det_infer.tar
+```
+
+蒸馏量化自动压缩示例通过run.py脚本启动，会使用接口 ```paddleslim.auto_compression.AutoCompression``` 对模型进行量化训练和蒸馏。配置config文件中模型路径、数据集路径、蒸馏、量化和训练等部分的参数，配置完成后便可开始自动压缩。
+
+**单卡启动**
+
+```shell
+export CUDA_VISIBLE_DEVICES=0
+python run.py --save_dir='./save_quant_ppocrv4_det/' --config_path='./configs/ppocrv4/ppocrv4_det_qat_dist.yaml'
+```
+
+**多卡启动**
+
+若训练任务中包含大量训练数据，如果使用单卡训练，会非常耗时，使用分布式训练可以达到几乎线性的加速比。
+
+```shell
+export CUDA_VISIBLE_DEVICES=0,1,2,3
+python -m paddle.distributed.launch run.py --save_dir='./save_quant_ppocrv4_det/' --config_path='./configs/ppocrv4/ppocrv4_det_qat_dist.yaml'
+```
+多卡训练指的是将训练任务按照一定方法拆分到多个训练节点完成数据读取、前向计算、反向梯度计算等过程，并将计算出的梯度上传至服务节点。服务节点在收到所有训练节点传来的梯度后，会将梯度聚合并更新参数。最后将参数发送给训练节点，开始新一轮的训练。多卡训练一轮训练能训练```batch size * num gpus```的数据，比如单卡的```batch size```为32，单轮训练的数据量即32，而四卡训练的```batch size```为32，单轮训练的数据量为128。
+
+注意 ```learning rate``` 与 ```batch size``` 呈线性关系，这里单卡 ```batch size``` 8，对应的 ```learning rate``` 为0.00005，那么如果 ```batch size``` 增大4倍改为32，```learning rate``` 也需乘以4；多卡时 ```batch size``` 为8，```learning rate``` 需乘上卡数。所以改变 ```batch size``` 或改变训练卡数都需要对应修改 ```learning rate```。
+
+
+**验证精度**
+
+根据训练log可以看到模型验证的精度，若需再次验证精度，修改配置文件```./configs/ppocrv3_det_qat_dist.yaml```中所需验证模型的文件夹路径及模型和参数名称```model_dir, model_filename, params_filename```，然后使用以下命令进行验证：
+
+```shell
+export CUDA_VISIBLE_DEVICES=0
+python eval.py --config_path='./configs/ppocrv3_det_qat_dist.yaml'
+```
+
+## 4.预测部署
+
+#### 4.1 Paddle Inference 验证性能
+
+输出的量化模型也是静态图模型，静态图模型在GPU上可以使用TensorRT进行加速，在CPU上可以使用MKLDNN进行加速。
+
+TensorRT预测环境配置：
+1. 如果使用 TesorRT 预测引擎，需安装 ```WITH_TRT=ON``` 的Paddle，上述paddle下载的2.5满足打开TensorRT编译的要求。
+2. 使用TensorRT预测需要进一步安装TensorRT，安装TensorRT的方式参考[TensorRT安装说明](../../../docs/deployment/installtrt.md)。
+
+以下字段用于配置预测参数：
+
+| 参数名 | 含义 |
+|:------:|:------:|
+| model_path | inference 模型文件所在目录，该目录下需要有文件 .pdmodel 和 .pdiparams 两个文件 |
+| model_filename | inference_model_dir文件夹下的模型文件名称 |
+| params_filename | inference_model_dir文件夹下的参数文件名称 |
+| dataset_config | 数据集配置的config  |
+| image_file | 待测试单张图片的路径，如果设置image_file，则dataset_config将无效。   |
+| device | 预测时的设备，可选：`CPU`, `GPU`。  |
+| use_trt | 是否使用 TesorRT 预测引擎，在device为```GPU```时生效。   |
+| use_mkldnn | 是否启用```MKL-DNN```加速库，注意```use_mkldnn```，在device为```CPU```时生效。  |
+| cpu_threads | CPU预测时，使用CPU线程数量，默认10  |
+| precision | 预测时精度，可选：`fp32`, `fp16`, `int8`。 |
+
+
+准备好预测模型，并且修改dataset_config中数据集路径为正确的路径后，启动测试：
+
+##### 4.1.1 使用测试脚本进行批量测试：
+
+我们提供两个脚本文件用于测试模型自动化压缩的效果，分别是[test_ocr_det.sh](./test_ocr_det.sh)和[test_ocr_rec.sh](./test_ocr_rec.sh)，这两个脚本都接收一个`model_type`参数，用于区分是测试mobile模型还是server模型，可选参数为`mobile`和`server`，使用示例：
+
+  ```shell
+  # 测试mobile模型
+  bash test_ocr_det.sh mobile
+  bash test_ocr_rec.sh mobile
+  # 测试server模型
+  bash test_ocr_det.sh server
+  bash test_ocr_rec.sh server
+  ```
+
+##### 4.1.2 基于压缩模型进行基于GPU的批量测试：
+
+```shell
+cd deploy/slim/auto_compression
+python test_ocr.py \
+      --model_path save_quant_ppocrv4_det \
+      --config_path configs/ppocrv4/ppocrv4_det_qat_dist.yaml \
+      --device GPU  \
+      --use_trt True  \
+      --precision int8
+```
+
+
+##### 4.1.3 基于压缩前模型进行基于GPU的批量测试：
+
+```shell
+cd deploy/slim/auto_compression
+python test_ocr.py \
+      --model_path ch_PP-OCRv4_det_infer \
+      --config_path configs/ppocrv4/ppocrv4_rec_det_dist.yaml \
+      --device GPU  \
+      --use_trt True  \
+      --precision int8
+```
+
+
+##### 4.1.4 基于压缩模型进行基于CPU的批量测试：
+
+- MKLDNN预测：
+
+```shell
+cd deploy/slim/auto_compression
+python test_ocr.py \
+      --model_path save_quant_ppocrv4_det \
+      --config_path configs/ppocrv4/ppocrv4_det_qat_dist.yaml \
+      --device GPU  \
+      --use_trt True  \
+      --use_mkldnn=True \
+      --precision=int8 \
+      --cpu_threads=10
+```
+
+### 4.2 PaddleLite端侧部署
+PaddleLite端侧部署可参考：
+- [Paddle Lite部署](https://github.com/PaddlePaddle/PaddleOCR/tree/9cdab61d909eb595af849db885c257ca8c74cb57/deploy/lite)
+
+## 5.FAQ
+
+### 5.1 报错找不到模型文件或者数据集文件
+
+如果在推理或者跑ACT时报错找不到模型文件或者数据集文件，可以检查一下配置文件中的路径是否正确，以det_mobile为例，配置文件中的指定模型路径的配置信息如下：
+
+```yaml
+Global:
+  model_dir: ./models/ch_PP-OCRv4_det_infer
+  model_filename: inference.pdmodel
+  params_filename: inference.pdiparams
+```
+指定训练集验证集路径的配置信息如下：
+
+```yaml
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: datasets/chinese
+    label_file_list:
+        - datasets/chinese/zhongce_training_fix_1.6k.txt
+        - datasets/chinese/label_train_all_f4_part2.txt
+        - datasets/chinese/label_train_all_f4_part3.txt
+        - datasets/chinese/label_train_all_f4_part4.txt
+        - datasets/chinese/label_train_all_f4_part5.txt
+        - datasets/chinese/synth_en_my_clip.txt
+        - datasets/chinese/synth_ch_my_clip.txt
+        - datasets/chinese/synth_en_my_largeword_clip.txt
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: datasets/v4_4_test_dataset
+    label_file_list:
+      - datasets/v4_4_test_dataset/label.txt
+```
+
+### 5.2 软件环境一致，硬件不同导致精度差异很大？
+
+这种情况是正常的，TensorRT针对不同的硬件设备有着不同的优化方法，同一种优化策略在不同硬件上可能有着截然不同的表现，以本实验的ppocrv4_det_server为举例。截取[test_ocr.py](./test_ocr.py)中的一部分代码如下所示：
+```python
+if args.precision == 'int8' and "ppocrv4_det_server_qat_dist.yaml" in args.config_path:
+    # Use the following settings only when the hardware is a Tesla V100. If you are using
+    # a RTX 3090, use the settings in the else branch.
+    pred_cfg.enable_tensorrt_engine(
+        workspace_size=1 << 30,
+        max_batch_size=1,
+        min_subgraph_size=30,
+        precision_mode=precision_map[args.precision],
+        use_static=True,
+        use_calib_mode=False, )
+    pred_cfg.exp_disable_tensorrt_ops(["elementwise_add"])
+else:
+    pred_cfg.enable_tensorrt_engine(
+    workspace_size=1 << 30,
+    max_batch_size=1,
+    min_subgraph_size=4,
+    precision_mode=precision_map[args.precision],
+    use_static=True,
+    use_calib_mode=False, )
+```
+当硬件为RTX 3090的时候，使用else分支中的策略即可得到正常的结果，但是当硬件是Tesla V100的时候，必须使用if分支中的策略才能保证量化后精度不下降，具体结果参考[benchmark](#2-benchmark)。
diff --git a/deploy/slim/auto_compression/configs/ppocrv4/ppocrv4_det_qat_dist.yaml b/deploy/slim/auto_compression/configs/ppocrv4/ppocrv4_det_qat_dist.yaml
new file mode 100644
index 0000000..4667422
--- /dev/null
+++ b/deploy/slim/auto_compression/configs/ppocrv4/ppocrv4_det_qat_dist.yaml
@@ -0,0 +1,163 @@
+Global:
+  model_type: det
+  model_dir: ./models/ch_PP-OCRv4_det_infer
+  model_filename: inference.pdmodel
+  params_filename: inference.pdiparams
+  algorithm: DB
+  
+Distillation:
+  alpha: 1.0
+  loss: l2
+
+QuantAware:
+  use_pact: false
+  activation_bits: 8
+  is_full_quantize: false
+  onnx_format: false
+  activation_quantize_type: moving_average_abs_max
+  weight_quantize_type: channel_wise_abs_max
+  not_quant_pattern:
+  - skip_quant
+  quantize_op_types:
+  - conv2d
+  weight_bits: 8
+
+TrainConfig:
+  epochs: 2
+  eval_iter: 200
+  learning_rate: 
+    type: CosineAnnealingDecay 
+    learning_rate: 0.000005
+  optimizer_builder:
+    optimizer:
+      type: Adam
+    weight_decay: 5.0e-05
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+  
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+  
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: datasets/chinese
+    label_file_list:
+        - datasets/chinese/zhongce_training_fix_1.6k.txt
+        - datasets/chinese/label_train_all_f4_part2.txt
+        - datasets/chinese/label_train_all_f4_part3.txt
+        - datasets/chinese/label_train_all_f4_part4.txt
+        - datasets/chinese/label_train_all_f4_part5.txt
+        - datasets/chinese/synth_en_my_clip.txt
+        - datasets/chinese/synth_ch_my_clip.txt
+        - datasets/chinese/synth_en_my_largeword_clip.txt
+    ratio_list:
+        - 0.3
+        - 0.2
+        - 0.1
+        - 0.2
+        - 0.2
+        - 0.1
+        - 0.2
+        - 0.2
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            rotate:
+            - -10
+            - 10
+        - type: Resize
+          args:
+            size:
+            - 0.5
+            - 3
+    - EastRandomCropData:
+        size:
+        - 960
+        - 960
+        max_tries: 50
+        keep_ratio: true
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 4
+    num_workers: 4
+    
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: datasets/v4_4_test_dataset
+    label_file_list:
+      - datasets/v4_4_test_dataset/label.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest:
+        limit_side_len: 960
+        limit_type: max
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 10
diff --git a/deploy/slim/auto_compression/configs/ppocrv4/ppocrv4_det_server_qat_dist.yaml b/deploy/slim/auto_compression/configs/ppocrv4/ppocrv4_det_server_qat_dist.yaml
new file mode 100644
index 0000000..febcd0d
--- /dev/null
+++ b/deploy/slim/auto_compression/configs/ppocrv4/ppocrv4_det_server_qat_dist.yaml
@@ -0,0 +1,161 @@
+Global:
+  model_type: det
+  model_dir: ./models/ch_PP-OCRv4_det_server_infer
+  model_filename: inference.pdmodel
+  params_filename: inference.pdiparams
+  algorithm: DB
+  
+Distillation:
+  alpha: 1.0
+  loss: l2
+
+QuantAware:
+  use_pact: false
+  activation_bits: 8
+  is_full_quantize: false
+  onnx_format: false
+  activation_quantize_type: moving_average_abs_max
+  weight_quantize_type: channel_wise_abs_max
+  not_quant_pattern:
+  - skip_quant
+  quantize_op_types:
+  - conv2d
+  weight_bits: 8
+
+TrainConfig:
+  epochs: 1
+  eval_iter: 200
+  learning_rate: 
+    type: CosineAnnealingDecay 
+    learning_rate: 0.000005
+  optimizer_builder:
+    optimizer:
+      type: Adam
+    weight_decay: 5.0e-05
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+  
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+  
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: datasets/chinese
+    label_file_list:
+        - datasets/chinese/zhongce_training_fix_1.6k.txt
+        - datasets/chinese/label_train_all_f4_part2.txt
+        - datasets/chinese/label_train_all_f4_part3.txt
+        - datasets/chinese/label_train_all_f4_part4.txt
+        - datasets/chinese/label_train_all_f4_part5.txt
+        - datasets/chinese/synth_en_my_clip.txt
+        - datasets/chinese/synth_ch_my_clip.txt
+        - datasets/chinese/synth_en_my_largeword_clip.txt
+    ratio_list:
+        - 0.3
+        - 0.2
+        - 0.1
+        - 0.2
+        - 0.2
+        - 0.1
+        - 0.2
+        - 0.2
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - IaaAugment:
+        augmenter_args:
+        - type: Fliplr
+          args:
+            p: 0.5
+        - type: Affine
+          args:
+            rotate:
+            - -10
+            - 10
+        - type: Resize
+          args:
+            size:
+            - 0.5
+            - 3
+    - EastRandomCropData:
+        size:
+        - 960
+        - 960
+        max_tries: 50
+        keep_ratio: true
+    - MakeBorderMap:
+        shrink_ratio: 0.4
+        thresh_min: 0.3
+        thresh_max: 0.7
+    - MakeShrinkMap:
+        shrink_ratio: 0.4
+        min_text_size: 8
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - threshold_map
+        - threshold_mask
+        - shrink_map
+        - shrink_mask
+  loader:
+    shuffle: true
+    drop_last: false
+    batch_size_per_card: 2
+    num_workers: 8
+    
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: datasets/v4_4_test_dataset_small
+    label_file_list:
+      - datasets/v4_4_test_dataset_small/label.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - DetLabelEncode: null
+    - DetResizeForTest: null
+    - NormalizeImage:
+        scale: 1./255.
+        mean:
+        - 0.485
+        - 0.456
+        - 0.406
+        std:
+        - 0.229
+        - 0.224
+        - 0.225
+        order: hwc
+    - ToCHWImage: null
+    - KeepKeys:
+        keep_keys:
+        - image
+        - shape
+        - polys
+        - ignore_tags
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 2
diff --git a/deploy/slim/auto_compression/configs/ppocrv4/ppocrv4_rec_qat_dist.yaml b/deploy/slim/auto_compression/configs/ppocrv4/ppocrv4_rec_qat_dist.yaml
new file mode 100644
index 0000000..8da639c
--- /dev/null
+++ b/deploy/slim/auto_compression/configs/ppocrv4/ppocrv4_rec_qat_dist.yaml
@@ -0,0 +1,115 @@
+Global:
+  model_dir: ./models/ch_PP-OCRv4_rec_infer
+  model_filename: inference.pdmodel
+  params_filename: inference.pdiparams
+  model_type: rec
+  algorithm: SVTR
+  character_dict_path: ./ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  use_space_char: true
+
+Distillation:
+  alpha: [1.0, 1.0]
+  loss: ['skd', 'l2']
+  node:
+   - ['softmax_11.tmp_0']
+   - ['linear_170.tmp_1']
+
+QuantAware:
+  use_pact: false
+  activation_bits: 8
+  is_full_quantize: false
+  onnx_format: false
+  activation_quantize_type: moving_average_abs_max
+  weight_quantize_type: channel_wise_abs_max
+  not_quant_pattern:
+  - skip_quant
+  quantize_op_types:
+  - conv2d
+  weight_bits: 8
+
+TrainConfig:
+  epochs: 1
+  eval_iter: 1000
+  logging_iter: 100
+  learning_rate: 
+    type: CosineAnnealingDecay 
+    learning_rate: 0.00001
+  optimizer_builder:
+    optimizer:
+      type: Adam
+    weight_decay: 5.0e-05
+
+PostProcess:
+  name: CTCLabelDecode 
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False 
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: datasets/real_data/
+    label_file_list:
+      - datasets/real_data/train_list.txt
+    ext_op_transform_idx: 1
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 64
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: datasets/real_data/
+    label_file_list:
+      - datasets/real_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 4
diff --git a/deploy/slim/auto_compression/configs/ppocrv4/ppocrv4_rec_server_qat_dist.yaml b/deploy/slim/auto_compression/configs/ppocrv4/ppocrv4_rec_server_qat_dist.yaml
new file mode 100644
index 0000000..09c1bc6
--- /dev/null
+++ b/deploy/slim/auto_compression/configs/ppocrv4/ppocrv4_rec_server_qat_dist.yaml
@@ -0,0 +1,113 @@
+Global:
+  model_dir: ./models/ch_PP-OCRv4_rec_server_infer
+  model_filename: inference.pdmodel
+  params_filename: inference.pdiparams
+  model_type: rec
+  algorithm: SVTR
+  character_dict_path: ./ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  use_space_char: true
+
+Distillation:
+  alpha: 1.0
+  loss: 'l2'
+
+QuantAware:
+  use_pact: false
+  activation_bits: 8
+  is_full_quantize: false
+  onnx_format: false
+  activation_quantize_type: moving_average_abs_max
+  weight_quantize_type: channel_wise_abs_max
+  not_quant_pattern:
+  - skip_quant
+  quantize_op_types:
+  - conv2d
+  weight_bits: 8
+
+TrainConfig:
+  epochs: 1
+  eval_iter: 1000
+  logging_iter: 100
+  learning_rate: 
+    type: CosineAnnealingDecay 
+    learning_rate: 0.00001
+  optimizer_builder:
+    optimizer:
+      type: Adam
+    weight_decay: 5.0e-05
+
+PostProcess:
+  name: CTCLabelDecode 
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  ignore_space: False 
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: datasets/real_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - datasets/real_data/train_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+        max_text_length: *max_text_length
+    - RecAug:
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  sampler:
+    name: MultiScaleSampler
+    scales: [[320, 32], [320, 48], [320, 64]]
+    first_bs: &bs 64
+    fix_bs: false
+    divided_factor: [8, 16] # w, h
+    is_training: True
+  loader:
+    shuffle: true
+    batch_size_per_card: *bs
+    drop_last: true
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: datasets/real_data/
+    label_file_list:
+    - datasets/real_data/val_list.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+        gtc_encode: NRTRLabelEncode
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_gtc
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 4
diff --git a/deploy/slim/auto_compression/ppocr_keys_v1.txt b/deploy/slim/auto_compression/ppocr_keys_v1.txt
new file mode 100644
index 0000000..b75af21
--- /dev/null
+++ b/deploy/slim/auto_compression/ppocr_keys_v1.txt
@@ -0,0 +1,6623 @@
+'
+疗
+绚
+诚
+娇
+溜
+题
+贿
+者
+廖
+更
+纳
+加
+奉
+公
+一
+就
+汴
+计
+与
+路
+房
+原
+妇
+2
+0
+8
+-
+7
+其
+>
+:
+]
+,
+，
+骑
+刈
+全
+消
+昏
+傈
+安
+久
+钟
+嗅
+不
+影
+处
+驽
+蜿
+资
+关
+椤
+地
+瘸
+专
+问
+忖
+票
+嫉
+炎
+韵
+要
+月
+田
+节
+陂
+鄙
+捌
+备
+拳
+伺
+眼
+网
+盎
+大
+傍
+心
+东
+愉
+汇
+蹿
+科
+每
+业
+里
+航
+晏
+字
+平
+录
+先
+1
+3
+彤
+鲶
+产
+稍
+督
+腴
+有
+象
+岳
+注
+绍
+在
+泺
+文
+定
+核
+名
+水
+过
+理
+让
+偷
+率
+等
+这
+发
+”
+为
+含
+肥
+酉
+相
+鄱
+七
+编
+猥
+锛
+日
+镀
+蒂
+掰
+倒
+辆
+栾
+栗
+综
+涩
+州
+雌
+滑
+馀
+了
+机
+块
+司
+宰
+甙
+兴
+矽
+抚
+保
+用
+沧
+秩
+如
+收
+息
+滥
+页
+疑
+埠
+!
+！
+姥
+异
+橹
+钇
+向
+下
+跄
+的
+椴
+沫
+国
+绥
+獠
+报
+开
+民
+蜇
+何
+分
+凇
+长
+讥
+藏
+掏
+施
+羽
+中
+讲
+派
+嘟
+人
+提
+浼
+间
+世
+而
+古
+多
+倪
+唇
+饯
+控
+庚
+首
+赛
+蜓
+味
+断
+制
+觉
+技
+替
+艰
+溢
+潮
+夕
+钺
+外
+摘
+枋
+动
+双
+单
+啮
+户
+枇
+确
+锦
+曜
+杜
+或
+能
+效
+霜
+盒
+然
+侗
+电
+晁
+放
+步
+鹃
+新
+杖
+蜂
+吒
+濂
+瞬
+评
+总
+隍
+对
+独
+合
+也
+是
+府
+青
+天
+诲
+墙
+组
+滴
+级
+邀
+帘
+示
+已
+时
+骸
+仄
+泅
+和
+遨
+店
+雇
+疫
+持
+巍
+踮
+境
+只
+亨
+目
+鉴
+崤
+闲
+体
+泄
+杂
+作
+般
+轰
+化
+解
+迂
+诿
+蛭
+璀
+腾
+告
+版
+服
+省
+师
+小
+规
+程
+线
+海
+办
+引
+二
+桧
+牌
+砺
+洄
+裴
+修
+图
+痫
+胡
+许
+犊
+事
+郛
+基
+柴
+呼
+食
+研
+奶
+律
+蛋
+因
+葆
+察
+戏
+褒
+戒
+再
+李
+骁
+工
+貂
+油
+鹅
+章
+啄
+休
+场
+给
+睡
+纷
+豆
+器
+捎
+说
+敏
+学
+会
+浒
+设
+诊
+格
+廓
+查
+来
+霓
+室
+溆
+￠
+诡
+寥
+焕
+舜
+柒
+狐
+回
+戟
+砾
+厄
+实
+翩
+尿
+五
+入
+径
+惭
+喹
+股
+宇
+篝
+|
+;
+美
+期
+云
+九
+祺
+扮
+靠
+锝
+槌
+系
+企
+酰
+阊
+暂
+蚕
+忻
+豁
+本
+羹
+执
+条
+钦
+H
+獒
+限
+进
+季
+楦
+于
+芘
+玖
+铋
+茯
+未
+答
+粘
+括
+样
+精
+欠
+矢
+甥
+帷
+嵩
+扣
+令
+仔
+风
+皈
+行
+支
+部
+蓉
+刮
+站
+蜡
+救
+钊
+汗
+松
+嫌
+成
+可
+.
+鹤
+院
+从
+交
+政
+怕
+活
+调
+球
+局
+验
+髌
+第
+韫
+谗
+串
+到
+圆
+年
+米
+/
+*
+友
+忿
+检
+区
+看
+自
+敢
+刃
+个
+兹
+弄
+流
+留
+同
+没
+齿
+星
+聆
+轼
+湖
+什
+三
+建
+蛔
+儿
+椋
+汕
+震
+颧
+鲤
+跟
+力
+情
+璺
+铨
+陪
+务
+指
+族
+训
+滦
+鄣
+濮
+扒
+商
+箱
+十
+召
+慷
+辗
+所
+莞
+管
+护
+臭
+横
+硒
+嗓
+接
+侦
+六
+露
+党
+馋
+驾
+剖
+高
+侬
+妪
+幂
+猗
+绺
+骐
+央
+酐
+孝
+筝
+课
+徇
+缰
+门
+男
+西
+项
+句
+谙
+瞒
+秃
+篇
+教
+碲
+罚
+声
+呐
+景
+前
+富
+嘴
+鳌
+稀
+免
+朋
+啬
+睐
+去
+赈
+鱼
+住
+肩
+愕
+速
+旁
+波
+厅
+健
+茼
+厥
+鲟
+谅
+投
+攸
+炔
+数
+方
+击
+呋
+谈
+绩
+别
+愫
+僚
+躬
+鹧
+胪
+炳
+招
+喇
+膨
+泵
+蹦
+毛
+结
+5
+4
+谱
+识
+陕
+粽
+婚
+拟
+构
+且
+搜
+任
+潘
+比
+郢
+妨
+醪
+陀
+桔
+碘
+扎
+选
+哈
+骷
+楷
+亿
+明
+缆
+脯
+监
+睫
+逻
+婵
+共
+赴
+淝
+凡
+惦
+及
+达
+揖
+谩
+澹
+减
+焰
+蛹
+番
+祁
+柏
+员
+禄
+怡
+峤
+龙
+白
+叽
+生
+闯
+起
+细
+装
+谕
+竟
+聚
+钙
+上
+导
+渊
+按
+艾
+辘
+挡
+耒
+盹
+饪
+臀
+记
+邮
+蕙
+受
+各
+医
+搂
+普
+滇
+朗
+茸
+带
+翻
+酚
+(
+光
+堤
+墟
+蔷
+万
+幻
+〓
+瑙
+辈
+昧
+盏
+亘
+蛀
+吉
+铰
+请
+子
+假
+闻
+税
+井
+诩
+哨
+嫂
+好
+面
+琐
+校
+馊
+鬣
+缂
+营
+访
+炖
+占
+农
+缀
+否
+经
+钚
+棵
+趟
+张
+亟
+吏
+茶
+谨
+捻
+论
+迸
+堂
+玉
+信
+吧
+瞠
+乡
+姬
+寺
+咬
+溏
+苄
+皿
+意
+赉
+宝
+尔
+钰
+艺
+特
+唳
+踉
+都
+荣
+倚
+登
+荐
+丧
+奇
+涵
+批
+炭
+近
+符
+傩
+感
+道
+着
+菊
+虹
+仲
+众
+懈
+濯
+颞
+眺
+南
+释
+北
+缝
+标
+既
+茗
+整
+撼
+迤
+贲
+挎
+耱
+拒
+某
+妍
+卫
+哇
+英
+矶
+藩
+治
+他
+元
+领
+膜
+遮
+穗
+蛾
+飞
+荒
+棺
+劫
+么
+市
+火
+温
+拈
+棚
+洼
+转
+果
+奕
+卸
+迪
+伸
+泳
+斗
+邡
+侄
+涨
+屯
+萋
+胭
+氡
+崮
+枞
+惧
+冒
+彩
+斜
+手
+豚
+随
+旭
+淑
+妞
+形
+菌
+吲
+沱
+争
+驯
+歹
+挟
+兆
+柱
+传
+至
+包
+内
+响
+临
+红
+功
+弩
+衡
+寂
+禁
+老
+棍
+耆
+渍
+织
+害
+氵
+渑
+布
+载
+靥
+嗬
+虽
+苹
+咨
+娄
+库
+雉
+榜
+帜
+嘲
+套
+瑚
+亲
+簸
+欧
+边
+6
+腿
+旮
+抛
+吹
+瞳
+得
+镓
+梗
+厨
+继
+漾
+愣
+憨
+士
+策
+窑
+抑
+躯
+襟
+脏
+参
+贸
+言
+干
+绸
+鳄
+穷
+藜
+音
+折
+详
+)
+举
+悍
+甸
+癌
+黎
+谴
+死
+罩
+迁
+寒
+驷
+袖
+媒
+蒋
+掘
+模
+纠
+恣
+观
+祖
+蛆
+碍
+位
+稿
+主
+澧
+跌
+筏
+京
+锏
+帝
+贴
+证
+糠
+才
+黄
+鲸
+略
+炯
+饱
+四
+出
+园
+犀
+牧
+容
+汉
+杆
+浈
+汰
+瑷
+造
+虫
+瘩
+怪
+驴
+济
+应
+花
+沣
+谔
+夙
+旅
+价
+矿
+以
+考
+s
+u
+呦
+晒
+巡
+茅
+准
+肟
+瓴
+詹
+仟
+褂
+译
+桌
+混
+宁
+怦
+郑
+抿
+些
+余
+鄂
+饴
+攒
+珑
+群
+阖
+岔
+琨
+藓
+预
+环
+洮
+岌
+宀
+杲
+瀵
+最
+常
+囡
+周
+踊
+女
+鼓
+袭
+喉
+简
+范
+薯
+遐
+疏
+粱
+黜
+禧
+法
+箔
+斤
+遥
+汝
+奥
+直
+贞
+撑
+置
+绱
+集
+她
+馅
+逗
+钧
+橱
+魉
+[
+恙
+躁
+唤
+9
+旺
+膘
+待
+脾
+惫
+购
+吗
+依
+盲
+度
+瘿
+蠖
+俾
+之
+镗
+拇
+鲵
+厝
+簧
+续
+款
+展
+啃
+表
+剔
+品
+钻
+腭
+损
+清
+锶
+统
+涌
+寸
+滨
+贪
+链
+吠
+冈
+伎
+迥
+咏
+吁
+览
+防
+迅
+失
+汾
+阔
+逵
+绀
+蔑
+列
+川
+凭
+努
+熨
+揪
+利
+俱
+绉
+抢
+鸨
+我
+即
+责
+膦
+易
+毓
+鹊
+刹
+玷
+岿
+空
+嘞
+绊
+排
+术
+估
+锷
+违
+们
+苟
+铜
+播
+肘
+件
+烫
+审
+鲂
+广
+像
+铌
+惰
+铟
+巳
+胍
+鲍
+康
+憧
+色
+恢
+想
+拷
+尤
+疳
+知
+S
+Y
+F
+D
+A
+峄
+裕
+帮
+握
+搔
+氐
+氘
+难
+墒
+沮
+雨
+叁
+缥
+悴
+藐
+湫
+娟
+苑
+稠
+颛
+簇
+后
+阕
+闭
+蕤
+缚
+怎
+佞
+码
+嘤
+蔡
+痊
+舱
+螯
+帕
+赫
+昵
+升
+烬
+岫
+、
+疵
+蜻
+髁
+蕨
+隶
+烛
+械
+丑
+盂
+梁
+强
+鲛
+由
+拘
+揉
+劭
+龟
+撤
+钩
+呕
+孛
+费
+妻
+漂
+求
+阑
+崖
+秤
+甘
+通
+深
+补
+赃
+坎
+床
+啪
+承
+吼
+量
+暇
+钼
+烨
+阂
+擎
+脱
+逮
+称
+P
+神
+属
+矗
+华
+届
+狍
+葑
+汹
+育
+患
+窒
+蛰
+佼
+静
+槎
+运
+鳗
+庆
+逝
+曼
+疱
+克
+代
+官
+此
+麸
+耧
+蚌
+晟
+例
+础
+榛
+副
+测
+唰
+缢
+迹
+灬
+霁
+身
+岁
+赭
+扛
+又
+菡
+乜
+雾
+板
+读
+陷
+徉
+贯
+郁
+虑
+变
+钓
+菜
+圾
+现
+琢
+式
+乐
+维
+渔
+浜
+左
+吾
+脑
+钡
+警
+T
+啵
+拴
+偌
+漱
+湿
+硕
+止
+骼
+魄
+积
+燥
+联
+踢
+玛
+则
+窿
+见
+振
+畿
+送
+班
+钽
+您
+赵
+刨
+印
+讨
+踝
+籍
+谡
+舌
+崧
+汽
+蔽
+沪
+酥
+绒
+怖
+财
+帖
+肱
+私
+莎
+勋
+羔
+霸
+励
+哼
+帐
+将
+帅
+渠
+纪
+婴
+娩
+岭
+厘
+滕
+吻
+伤
+坝
+冠
+戊
+隆
+瘁
+介
+涧
+物
+黍
+并
+姗
+奢
+蹑
+掣
+垸
+锴
+命
+箍
+捉
+病
+辖
+琰
+眭
+迩
+艘
+绌
+繁
+寅
+若
+毋
+思
+诉
+类
+诈
+燮
+轲
+酮
+狂
+重
+反
+职
+筱
+县
+委
+磕
+绣
+奖
+晋
+濉
+志
+徽
+肠
+呈
+獐
+坻
+口
+片
+碰
+几
+村
+柿
+劳
+料
+获
+亩
+惕
+晕
+厌
+号
+罢
+池
+正
+鏖
+煨
+家
+棕
+复
+尝
+懋
+蜥
+锅
+岛
+扰
+队
+坠
+瘾
+钬
+@
+卧
+疣
+镇
+譬
+冰
+彷
+频
+黯
+据
+垄
+采
+八
+缪
+瘫
+型
+熹
+砰
+楠
+襁
+箐
+但
+嘶
+绳
+啤
+拍
+盥
+穆
+傲
+洗
+盯
+塘
+怔
+筛
+丿
+台
+恒
+喂
+葛
+永
+￥
+烟
+酒
+桦
+书
+砂
+蚝
+缉
+态
+瀚
+袄
+圳
+轻
+蛛
+超
+榧
+遛
+姒
+奘
+铮
+右
+荽
+望
+偻
+卡
+丶
+氰
+附
+做
+革
+索
+戚
+坨
+桷
+唁
+垅
+榻
+岐
+偎
+坛
+莨
+山
+殊
+微
+骇
+陈
+爨
+推
+嗝
+驹
+澡
+藁
+呤
+卤
+嘻
+糅
+逛
+侵
+郓
+酌
+德
+摇
+※
+鬃
+被
+慨
+殡
+羸
+昌
+泡
+戛
+鞋
+河
+宪
+沿
+玲
+鲨
+翅
+哽
+源
+铅
+语
+照
+邯
+址
+荃
+佬
+顺
+鸳
+町
+霭
+睾
+瓢
+夸
+椁
+晓
+酿
+痈
+咔
+侏
+券
+噎
+湍
+签
+嚷
+离
+午
+尚
+社
+锤
+背
+孟
+使
+浪
+缦
+潍
+鞅
+军
+姹
+驶
+笑
+鳟
+鲁
+》
+孽
+钜
+绿
+洱
+礴
+焯
+椰
+颖
+囔
+乌
+孔
+巴
+互
+性
+椽
+哞
+聘
+昨
+早
+暮
+胶
+炀
+隧
+低
+彗
+昝
+铁
+呓
+氽
+藉
+喔
+癖
+瑗
+姨
+权
+胱
+韦
+堑
+蜜
+酋
+楝
+砝
+毁
+靓
+歙
+锲
+究
+屋
+喳
+骨
+辨
+碑
+武
+鸠
+宫
+辜
+烊
+适
+坡
+殃
+培
+佩
+供
+走
+蜈
+迟
+翼
+况
+姣
+凛
+浔
+吃
+飘
+债
+犟
+金
+促
+苛
+崇
+坂
+莳
+畔
+绂
+兵
+蠕
+斋
+根
+砍
+亢
+欢
+恬
+崔
+剁
+餐
+榫
+快
+扶
+‖
+濒
+缠
+鳜
+当
+彭
+驭
+浦
+篮
+昀
+锆
+秸
+钳
+弋
+娣
+瞑
+夷
+龛
+苫
+拱
+致
+%
+嵊
+障
+隐
+弑
+初
+娓
+抉
+汩
+累
+蓖
+"
+唬
+助
+苓
+昙
+押
+毙
+破
+城
+郧
+逢
+嚏
+獭
+瞻
+溱
+婿
+赊
+跨
+恼
+璧
+萃
+姻
+貉
+灵
+炉
+密
+氛
+陶
+砸
+谬
+衔
+点
+琛
+沛
+枳
+层
+岱
+诺
+脍
+榈
+埂
+征
+冷
+裁
+打
+蹴
+素
+瘘
+逞
+蛐
+聊
+激
+腱
+萘
+踵
+飒
+蓟
+吆
+取
+咙
+簋
+涓
+矩
+曝
+挺
+揣
+座
+你
+史
+舵
+焱
+尘
+苏
+笈
+脚
+溉
+榨
+诵
+樊
+邓
+焊
+义
+庶
+儋
+蟋
+蒲
+赦
+呷
+杞
+诠
+豪
+还
+试
+颓
+茉
+太
+除
+紫
+逃
+痴
+草
+充
+鳕
+珉
+祗
+墨
+渭
+烩
+蘸
+慕
+璇
+镶
+穴
+嵘
+恶
+骂
+险
+绋
+幕
+碉
+肺
+戳
+刘
+潞
+秣
+纾
+潜
+銮
+洛
+须
+罘
+销
+瘪
+汞
+兮
+屉
+r
+林
+厕
+质
+探
+划
+狸
+殚
+善
+煊
+烹
+〒
+锈
+逯
+宸
+辍
+泱
+柚
+袍
+远
+蹋
+嶙
+绝
+峥
+娥
+缍
+雀
+徵
+认
+镱
+谷
+=
+贩
+勉
+撩
+鄯
+斐
+洋
+非
+祚
+泾
+诒
+饿
+撬
+威
+晷
+搭
+芍
+锥
+笺
+蓦
+候
+琊
+档
+礁
+沼
+卵
+荠
+忑
+朝
+凹
+瑞
+头
+仪
+弧
+孵
+畏
+铆
+突
+衲
+车
+浩
+气
+茂
+悖
+厢
+枕
+酝
+戴
+湾
+邹
+飚
+攘
+锂
+写
+宵
+翁
+岷
+无
+喜
+丈
+挑
+嗟
+绛
+殉
+议
+槽
+具
+醇
+淞
+笃
+郴
+阅
+饼
+底
+壕
+砚
+弈
+询
+缕
+庹
+翟
+零
+筷
+暨
+舟
+闺
+甯
+撞
+麂
+茌
+蔼
+很
+珲
+捕
+棠
+角
+阉
+媛
+娲
+诽
+剿
+尉
+爵
+睬
+韩
+诰
+匣
+危
+糍
+镯
+立
+浏
+阳
+少
+盆
+舔
+擘
+匪
+申
+尬
+铣
+旯
+抖
+赘
+瓯
+居
+ˇ
+哮
+游
+锭
+茏
+歌
+坏
+甚
+秒
+舞
+沙
+仗
+劲
+潺
+阿
+燧
+郭
+嗖
+霏
+忠
+材
+奂
+耐
+跺
+砀
+输
+岖
+媳
+氟
+极
+摆
+灿
+今
+扔
+腻
+枝
+奎
+药
+熄
+吨
+话
+q
+额
+慑
+嘌
+协
+喀
+壳
+埭
+视
+著
+於
+愧
+陲
+翌
+峁
+颅
+佛
+腹
+聋
+侯
+咎
+叟
+秀
+颇
+存
+较
+罪
+哄
+岗
+扫
+栏
+钾
+羌
+己
+璨
+枭
+霉
+煌
+涸
+衿
+键
+镝
+益
+岢
+奏
+连
+夯
+睿
+冥
+均
+糖
+狞
+蹊
+稻
+爸
+刿
+胥
+煜
+丽
+肿
+璃
+掸
+跚
+灾
+垂
+樾
+濑
+乎
+莲
+窄
+犹
+撮
+战
+馄
+软
+络
+显
+鸢
+胸
+宾
+妲
+恕
+埔
+蝌
+份
+遇
+巧
+瞟
+粒
+恰
+剥
+桡
+博
+讯
+凯
+堇
+阶
+滤
+卖
+斌
+骚
+彬
+兑
+磺
+樱
+舷
+两
+娱
+福
+仃
+差
+找
+桁
+÷
+净
+把
+阴
+污
+戬
+雷
+碓
+蕲
+楚
+罡
+焖
+抽
+妫
+咒
+仑
+闱
+尽
+邑
+菁
+爱
+贷
+沥
+鞑
+牡
+嗉
+崴
+骤
+塌
+嗦
+订
+拮
+滓
+捡
+锻
+次
+坪
+杩
+臃
+箬
+融
+珂
+鹗
+宗
+枚
+降
+鸬
+妯
+阄
+堰
+盐
+毅
+必
+杨
+崃
+俺
+甬
+状
+莘
+货
+耸
+菱
+腼
+铸
+唏
+痤
+孚
+澳
+懒
+溅
+翘
+疙
+杷
+淼
+缙
+骰
+喊
+悉
+砻
+坷
+艇
+赁
+界
+谤
+纣
+宴
+晃
+茹
+归
+饭
+梢
+铡
+街
+抄
+肼
+鬟
+苯
+颂
+撷
+戈
+炒
+咆
+茭
+瘙
+负
+仰
+客
+琉
+铢
+封
+卑
+珥
+椿
+镧
+窨
+鬲
+寿
+御
+袤
+铃
+萎
+砖
+餮
+脒
+裳
+肪
+孕
+嫣
+馗
+嵇
+恳
+氯
+江
+石
+褶
+冢
+祸
+阻
+狈
+羞
+银
+靳
+透
+咳
+叼
+敷
+芷
+啥
+它
+瓤
+兰
+痘
+懊
+逑
+肌
+往
+捺
+坊
+甩
+呻
+〃
+沦
+忘
+膻
+祟
+菅
+剧
+崆
+智
+坯
+臧
+霍
+墅
+攻
+眯
+倘
+拢
+骠
+铐
+庭
+岙
+瓠
+′
+缺
+泥
+迢
+捶
+?
+？
+郏
+喙
+掷
+沌
+纯
+秘
+种
+听
+绘
+固
+螨
+团
+香
+盗
+妒
+埚
+蓝
+拖
+旱
+荞
+铀
+血
+遏
+汲
+辰
+叩
+拽
+幅
+硬
+惶
+桀
+漠
+措
+泼
+唑
+齐
+肾
+念
+酱
+虚
+屁
+耶
+旗
+砦
+闵
+婉
+馆
+拭
+绅
+韧
+忏
+窝
+醋
+葺
+顾
+辞
+倜
+堆
+辋
+逆
+玟
+贱
+疾
+董
+惘
+倌
+锕
+淘
+嘀
+莽
+俭
+笏
+绑
+鲷
+杈
+择
+蟀
+粥
+嗯
+驰
+逾
+案
+谪
+褓
+胫
+哩
+昕
+颚
+鲢
+绠
+躺
+鹄
+崂
+儒
+俨
+丝
+尕
+泌
+啊
+萸
+彰
+幺
+吟
+骄
+苣
+弦
+脊
+瑰
+〈
+诛
+镁
+析
+闪
+剪
+侧
+哟
+框
+螃
+守
+嬗
+燕
+狭
+铈
+缮
+概
+迳
+痧
+鲲
+俯
+售
+笼
+痣
+扉
+挖
+满
+咋
+援
+邱
+扇
+歪
+便
+玑
+绦
+峡
+蛇
+叨
+〖
+泽
+胃
+斓
+喋
+怂
+坟
+猪
+该
+蚬
+炕
+弥
+赞
+棣
+晔
+娠
+挲
+狡
+创
+疖
+铕
+镭
+稷
+挫
+弭
+啾
+翔
+粉
+履
+苘
+哦
+楼
+秕
+铂
+土
+锣
+瘟
+挣
+栉
+习
+享
+桢
+袅
+磨
+桂
+谦
+延
+坚
+蔚
+噗
+署
+谟
+猬
+钎
+恐
+嬉
+雒
+倦
+衅
+亏
+璩
+睹
+刻
+殿
+王
+算
+雕
+麻
+丘
+柯
+骆
+丸
+塍
+谚
+添
+鲈
+垓
+桎
+蚯
+芥
+予
+飕
+镦
+谌
+窗
+醚
+菀
+亮
+搪
+莺
+蒿
+羁
+足
+J
+真
+轶
+悬
+衷
+靛
+翊
+掩
+哒
+炅
+掐
+冼
+妮
+l
+谐
+稚
+荆
+擒
+犯
+陵
+虏
+浓
+崽
+刍
+陌
+傻
+孜
+千
+靖
+演
+矜
+钕
+煽
+杰
+酗
+渗
+伞
+栋
+俗
+泫
+戍
+罕
+沾
+疽
+灏
+煦
+芬
+磴
+叱
+阱
+榉
+湃
+蜀
+叉
+醒
+彪
+租
+郡
+篷
+屎
+良
+垢
+隗
+弱
+陨
+峪
+砷
+掴
+颁
+胎
+雯
+绵
+贬
+沐
+撵
+隘
+篙
+暖
+曹
+陡
+栓
+填
+臼
+彦
+瓶
+琪
+潼
+哪
+鸡
+摩
+啦
+俟
+锋
+域
+耻
+蔫
+疯
+纹
+撇
+毒
+绶
+痛
+酯
+忍
+爪
+赳
+歆
+嘹
+辕
+烈
+册
+朴
+钱
+吮
+毯
+癜
+娃
+谀
+邵
+厮
+炽
+璞
+邃
+丐
+追
+词
+瓒
+忆
+轧
+芫
+谯
+喷
+弟
+半
+冕
+裙
+掖
+墉
+绮
+寝
+苔
+势
+顷
+褥
+切
+衮
+君
+佳
+嫒
+蚩
+霞
+佚
+洙
+逊
+镖
+暹
+唛
+&
+殒
+顶
+碗
+獗
+轭
+铺
+蛊
+废
+恹
+汨
+崩
+珍
+那
+杵
+曲
+纺
+夏
+薰
+傀
+闳
+淬
+姘
+舀
+拧
+卷
+楂
+恍
+讪
+厩
+寮
+篪
+赓
+乘
+灭
+盅
+鞣
+沟
+慎
+挂
+饺
+鼾
+杳
+树
+缨
+丛
+絮
+娌
+臻
+嗳
+篡
+侩
+述
+衰
+矛
+圈
+蚜
+匕
+筹
+匿
+濞
+晨
+叶
+骋
+郝
+挚
+蚴
+滞
+增
+侍
+描
+瓣
+吖
+嫦
+蟒
+匾
+圣
+赌
+毡
+癞
+恺
+百
+曳
+需
+篓
+肮
+庖
+帏
+卿
+驿
+遗
+蹬
+鬓
+骡
+歉
+芎
+胳
+屐
+禽
+烦
+晌
+寄
+媾
+狄
+翡
+苒
+船
+廉
+终
+痞
+殇
+々
+畦
+饶
+改
+拆
+悻
+萄
+￡
+瓿
+乃
+訾
+桅
+匮
+溧
+拥
+纱
+铍
+骗
+蕃
+龋
+缬
+父
+佐
+疚
+栎
+醍
+掳
+蓄
+x
+惆
+颜
+鲆
+榆
+〔
+猎
+敌
+暴
+谥
+鲫
+贾
+罗
+玻
+缄
+扦
+芪
+癣
+落
+徒
+臾
+恿
+猩
+托
+邴
+肄
+牵
+春
+陛
+耀
+刊
+拓
+蓓
+邳
+堕
+寇
+枉
+淌
+啡
+湄
+兽
+酷
+萼
+碚
+濠
+萤
+夹
+旬
+戮
+梭
+琥
+椭
+昔
+勺
+蜊
+绐
+晚
+孺
+僵
+宣
+摄
+冽
+旨
+萌
+忙
+蚤
+眉
+噼
+蟑
+付
+契
+瓜
+悼
+颡
+壁
+曾
+窕
+颢
+澎
+仿
+俑
+浑
+嵌
+浣
+乍
+碌
+褪
+乱
+蔟
+隙
+玩
+剐
+葫
+箫
+纲
+围
+伐
+决
+伙
+漩
+瑟
+刑
+肓
+镳
+缓
+蹭
+氨
+皓
+典
+畲
+坍
+铑
+檐
+塑
+洞
+倬
+储
+胴
+淳
+戾
+吐
+灼
+惺
+妙
+毕
+珐
+缈
+虱
+盖
+羰
+鸿
+磅
+谓
+髅
+娴
+苴
+唷
+蚣
+霹
+抨
+贤
+唠
+犬
+誓
+逍
+庠
+逼
+麓
+籼
+釉
+呜
+碧
+秧
+氩
+摔
+霄
+穸
+纨
+辟
+妈
+映
+完
+牛
+缴
+嗷
+炊
+恩
+荔
+茆
+掉
+紊
+慌
+莓
+羟
+阙
+萁
+磐
+另
+蕹
+辱
+鳐
+湮
+吡
+吩
+唐
+睦
+垠
+舒
+圜
+冗
+瞿
+溺
+芾
+囱
+匠
+僳
+汐
+菩
+饬
+漓
+黑
+霰
+浸
+濡
+窥
+毂
+蒡
+兢
+驻
+鹉
+芮
+诙
+迫
+雳
+厂
+忐
+臆
+猴
+鸣
+蚪
+栈
+箕
+羡
+渐
+莆
+捍
+眈
+哓
+趴
+蹼
+埕
+嚣
+骛
+宏
+淄
+斑
+噜
+严
+瑛
+垃
+椎
+诱
+压
+庾
+绞
+焘
+廿
+抡
+迄
+棘
+夫
+纬
+锹
+眨
+瞌
+侠
+脐
+竞
+瀑
+孳
+骧
+遁
+姜
+颦
+荪
+滚
+萦
+伪
+逸
+粳
+爬
+锁
+矣
+役
+趣
+洒
+颔
+诏
+逐
+奸
+甭
+惠
+攀
+蹄
+泛
+尼
+拼
+阮
+鹰
+亚
+颈
+惑
+勒
+〉
+际
+肛
+爷
+刚
+钨
+丰
+养
+冶
+鲽
+辉
+蔻
+画
+覆
+皴
+妊
+麦
+返
+醉
+皂
+擀
+〗
+酶
+凑
+粹
+悟
+诀
+硖
+港
+卜
+z
+杀
+涕
+±
+舍
+铠
+抵
+弛
+段
+敝
+镐
+奠
+拂
+轴
+跛
+袱
+e
+t
+沉
+菇
+俎
+薪
+峦
+秭
+蟹
+历
+盟
+菠
+寡
+液
+肢
+喻
+染
+裱
+悱
+抱
+氙
+赤
+捅
+猛
+跑
+氮
+谣
+仁
+尺
+辊
+窍
+烙
+衍
+架
+擦
+倏
+璐
+瑁
+币
+楞
+胖
+夔
+趸
+邛
+惴
+饕
+虔
+蝎
+§
+哉
+贝
+宽
+辫
+炮
+扩
+饲
+籽
+魏
+菟
+锰
+伍
+猝
+末
+琳
+哚
+蛎
+邂
+呀
+姿
+鄞
+却
+歧
+仙
+恸
+椐
+森
+牒
+寤
+袒
+婆
+虢
+雅
+钉
+朵
+贼
+欲
+苞
+寰
+故
+龚
+坭
+嘘
+咫
+礼
+硷
+兀
+睢
+汶
+’
+铲
+烧
+绕
+诃
+浃
+钿
+哺
+柜
+讼
+颊
+璁
+腔
+洽
+咐
+脲
+簌
+筠
+镣
+玮
+鞠
+谁
+兼
+姆
+挥
+梯
+蝴
+谘
+漕
+刷
+躏
+宦
+弼
+b
+垌
+劈
+麟
+莉
+揭
+笙
+渎
+仕
+嗤
+仓
+配
+怏
+抬
+错
+泯
+镊
+孰
+猿
+邪
+仍
+秋
+鼬
+壹
+歇
+吵
+炼
+<
+尧
+射
+柬
+廷
+胧
+霾
+凳
+隋
+肚
+浮
+梦
+祥
+株
+堵
+退
+L
+鹫
+跎
+凶
+毽
+荟
+炫
+栩
+玳
+甜
+沂
+鹿
+顽
+伯
+爹
+赔
+蛴
+徐
+匡
+欣
+狰
+缸
+雹
+蟆
+疤
+默
+沤
+啜
+痂
+衣
+禅
+w
+i
+h
+辽
+葳
+黝
+钗
+停
+沽
+棒
+馨
+颌
+肉
+吴
+硫
+悯
+劾
+娈
+马
+啧
+吊
+悌
+镑
+峭
+帆
+瀣
+涉
+咸
+疸
+滋
+泣
+翦
+拙
+癸
+钥
+蜒
++
+尾
+庄
+凝
+泉
+婢
+渴
+谊
+乞
+陆
+锉
+糊
+鸦
+淮
+I
+B
+N
+晦
+弗
+乔
+庥
+葡
+尻
+席
+橡
+傣
+渣
+拿
+惩
+麋
+斛
+缃
+矮
+蛏
+岘
+鸽
+姐
+膏
+催
+奔
+镒
+喱
+蠡
+摧
+钯
+胤
+柠
+拐
+璋
+鸥
+卢
+荡
+倾
+^
+_
+珀
+逄
+萧
+塾
+掇
+贮
+笆
+聂
+圃
+冲
+嵬
+M
+滔
+笕
+值
+炙
+偶
+蜱
+搐
+梆
+汪
+蔬
+腑
+鸯
+蹇
+敞
+绯
+仨
+祯
+谆
+梧
+糗
+鑫
+啸
+豺
+囹
+猾
+巢
+柄
+瀛
+筑
+踌
+沭
+暗
+苁
+鱿
+蹉
+脂
+蘖
+牢
+热
+木
+吸
+溃
+宠
+序
+泞
+偿
+拜
+檩
+厚
+朐
+毗
+螳
+吞
+媚
+朽
+担
+蝗
+橘
+畴
+祈
+糟
+盱
+隼
+郜
+惜
+珠
+裨
+铵
+焙
+琚
+唯
+咚
+噪
+骊
+丫
+滢
+勤
+棉
+呸
+咣
+淀
+隔
+蕾
+窈
+饨
+挨
+煅
+短
+匙
+粕
+镜
+赣
+撕
+墩
+酬
+馁
+豌
+颐
+抗
+酣
+氓
+佑
+搁
+哭
+递
+耷
+涡
+桃
+贻
+碣
+截
+瘦
+昭
+镌
+蔓
+氚
+甲
+猕
+蕴
+蓬
+散
+拾
+纛
+狼
+猷
+铎
+埋
+旖
+矾
+讳
+囊
+糜
+迈
+粟
+蚂
+紧
+鲳
+瘢
+栽
+稼
+羊
+锄
+斟
+睁
+桥
+瓮
+蹙
+祉
+醺
+鼻
+昱
+剃
+跳
+篱
+跷
+蒜
+翎
+宅
+晖
+嗑
+壑
+峻
+癫
+屏
+狠
+陋
+袜
+途
+憎
+祀
+莹
+滟
+佶
+溥
+臣
+约
+盛
+峰
+磁
+慵
+婪
+拦
+莅
+朕
+鹦
+粲
+裤
+哎
+疡
+嫖
+琵
+窟
+堪
+谛
+嘉
+儡
+鳝
+斩
+郾
+驸
+酊
+妄
+胜
+贺
+徙
+傅
+噌
+钢
+栅
+庇
+恋
+匝
+巯
+邈
+尸
+锚
+粗
+佟
+蛟
+薹
+纵
+蚊
+郅
+绢
+锐
+苗
+俞
+篆
+淆
+膀
+鲜
+煎
+诶
+秽
+寻
+涮
+刺
+怀
+噶
+巨
+褰
+魅
+灶
+灌
+桉
+藕
+谜
+舸
+薄
+搀
+恽
+借
+牯
+痉
+渥
+愿
+亓
+耘
+杠
+柩
+锔
+蚶
+钣
+珈
+喘
+蹒
+幽
+赐
+稗
+晤
+莱
+泔
+扯
+肯
+菪
+裆
+腩
+豉
+疆
+骜
+腐
+倭
+珏
+唔
+粮
+亡
+润
+慰
+伽
+橄
+玄
+誉
+醐
+胆
+龊
+粼
+塬
+陇
+彼
+削
+嗣
+绾
+芽
+妗
+垭
+瘴
+爽
+薏
+寨
+龈
+泠
+弹
+赢
+漪
+猫
+嘧
+涂
+恤
+圭
+茧
+烽
+屑
+痕
+巾
+赖
+荸
+凰
+腮
+畈
+亵
+蹲
+偃
+苇
+澜
+艮
+换
+骺
+烘
+苕
+梓
+颉
+肇
+哗
+悄
+氤
+涠
+葬
+屠
+鹭
+植
+竺
+佯
+诣
+鲇
+瘀
+鲅
+邦
+移
+滁
+冯
+耕
+癔
+戌
+茬
+沁
+巩
+悠
+湘
+洪
+痹
+锟
+循
+谋
+腕
+鳃
+钠
+捞
+焉
+迎
+碱
+伫
+急
+榷
+奈
+邝
+卯
+辄
+皲
+卟
+醛
+畹
+忧
+稳
+雄
+昼
+缩
+阈
+睑
+扌
+耗
+曦
+涅
+捏
+瞧
+邕
+淖
+漉
+铝
+耦
+禹
+湛
+喽
+莼
+琅
+诸
+苎
+纂
+硅
+始
+嗨
+傥
+燃
+臂
+赅
+嘈
+呆
+贵
+屹
+壮
+肋
+亍
+蚀
+卅
+豹
+腆
+邬
+迭
+浊
+}
+童
+螂
+捐
+圩
+勐
+触
+寞
+汊
+壤
+荫
+膺
+渌
+芳
+懿
+遴
+螈
+泰
+蓼
+蛤
+茜
+舅
+枫
+朔
+膝
+眙
+避
+梅
+判
+鹜
+璜
+牍
+缅
+垫
+藻
+黔
+侥
+惚
+懂
+踩
+腰
+腈
+札
+丞
+唾
+慈
+顿
+摹
+荻
+琬
+~
+斧
+沈
+滂
+胁
+胀
+幄
+莜
+Z
+匀
+鄄
+掌
+绰
+茎
+焚
+赋
+萱
+谑
+汁
+铒
+瞎
+夺
+蜗
+野
+娆
+冀
+弯
+篁
+懵
+灞
+隽
+芡
+脘
+俐
+辩
+芯
+掺
+喏
+膈
+蝈
+觐
+悚
+踹
+蔗
+熠
+鼠
+呵
+抓
+橼
+峨
+畜
+缔
+禾
+崭
+弃
+熊
+摒
+凸
+拗
+穹
+蒙
+抒
+祛
+劝
+闫
+扳
+阵
+醌
+踪
+喵
+侣
+搬
+仅
+荧
+赎
+蝾
+琦
+买
+婧
+瞄
+寓
+皎
+冻
+赝
+箩
+莫
+瞰
+郊
+笫
+姝
+筒
+枪
+遣
+煸
+袋
+舆
+痱
+涛
+母
+〇
+启
+践
+耙
+绲
+盘
+遂
+昊
+搞
+槿
+诬
+纰
+泓
+惨
+檬
+亻
+越
+C
+o
+憩
+熵
+祷
+钒
+暧
+塔
+阗
+胰
+咄
+娶
+魔
+琶
+钞
+邻
+扬
+杉
+殴
+咽
+弓
+〆
+髻
+】
+吭
+揽
+霆
+拄
+殖
+脆
+彻
+岩
+芝
+勃
+辣
+剌
+钝
+嘎
+甄
+佘
+皖
+伦
+授
+徕
+憔
+挪
+皇
+庞
+稔
+芜
+踏
+溴
+兖
+卒
+擢
+饥
+鳞
+煲
+‰
+账
+颗
+叻
+斯
+捧
+鳍
+琮
+讹
+蛙
+纽
+谭
+酸
+兔
+莒
+睇
+伟
+觑
+羲
+嗜
+宜
+褐
+旎
+辛
+卦
+诘
+筋
+鎏
+溪
+挛
+熔
+阜
+晰
+鳅
+丢
+奚
+灸
+呱
+献
+陉
+黛
+鸪
+甾
+萨
+疮
+拯
+洲
+疹
+辑
+叙
+恻
+谒
+允
+柔
+烂
+氏
+逅
+漆
+拎
+惋
+扈
+湟
+纭
+啕
+掬
+擞
+哥
+忽
+涤
+鸵
+靡
+郗
+瓷
+扁
+廊
+怨
+雏
+钮
+敦
+E
+懦
+憋
+汀
+拚
+啉
+腌
+岸
+f
+痼
+瞅
+尊
+咀
+眩
+飙
+忌
+仝
+迦
+熬
+毫
+胯
+篑
+茄
+腺
+凄
+舛
+碴
+锵
+诧
+羯
+後
+漏
+汤
+宓
+仞
+蚁
+壶
+谰
+皑
+铄
+棰
+罔
+辅
+晶
+苦
+牟
+闽
+\
+烃
+饮
+聿
+丙
+蛳
+朱
+煤
+涔
+鳖
+犁
+罐
+荼
+砒
+淦
+妤
+黏
+戎
+孑
+婕
+瑾
+戢
+钵
+枣
+捋
+砥
+衩
+狙
+桠
+稣
+阎
+肃
+梏
+诫
+孪
+昶
+婊
+衫
+嗔
+侃
+塞
+蜃
+樵
+峒
+貌
+屿
+欺
+缫
+阐
+栖
+诟
+珞
+荭
+吝
+萍
+嗽
+恂
+啻
+蜴
+磬
+峋
+俸
+豫
+谎
+徊
+镍
+韬
+魇
+晴
+U
+囟
+猜
+蛮
+坐
+囿
+伴
+亭
+肝
+佗
+蝠
+妃
+胞
+滩
+榴
+氖
+垩
+苋
+砣
+扪
+馏
+姓
+轩
+厉
+夥
+侈
+禀
+垒
+岑
+赏
+钛
+辐
+痔
+披
+纸
+碳
+“
+坞
+蠓
+挤
+荥
+沅
+悔
+铧
+帼
+蒌
+蝇
+a
+p
+y
+n
+g
+哀
+浆
+瑶
+凿
+桶
+馈
+皮
+奴
+苜
+佤
+伶
+晗
+铱
+炬
+优
+弊
+氢
+恃
+甫
+攥
+端
+锌
+灰
+稹
+炝
+曙
+邋
+亥
+眶
+碾
+拉
+萝
+绔
+捷
+浍
+腋
+姑
+菖
+凌
+涞
+麽
+锢
+桨
+潢
+绎
+镰
+殆
+锑
+渝
+铬
+困
+绽
+觎
+匈
+糙
+暑
+裹
+鸟
+盔
+肽
+迷
+綦
+『
+亳
+佝
+俘
+钴
+觇
+骥
+仆
+疝
+跪
+婶
+郯
+瀹
+唉
+脖
+踞
+针
+晾
+忒
+扼
+瞩
+叛
+椒
+疟
+嗡
+邗
+肆
+跆
+玫
+忡
+捣
+咧
+唆
+艄
+蘑
+潦
+笛
+阚
+沸
+泻
+掊
+菽
+贫
+斥
+髂
+孢
+镂
+赂
+麝
+鸾
+屡
+衬
+苷
+恪
+叠
+希
+粤
+爻
+喝
+茫
+惬
+郸
+绻
+庸
+撅
+碟
+宄
+妹
+膛
+叮
+饵
+崛
+嗲
+椅
+冤
+搅
+咕
+敛
+尹
+垦
+闷
+蝉
+霎
+勰
+败
+蓑
+泸
+肤
+鹌
+幌
+焦
+浠
+鞍
+刁
+舰
+乙
+竿
+裔
+。
+茵
+函
+伊
+兄
+丨
+娜
+匍
+謇
+莪
+宥
+似
+蝽
+翳
+酪
+翠
+粑
+薇
+祢
+骏
+赠
+叫
+Q
+噤
+噻
+竖
+芗
+莠
+潭
+俊
+羿
+耜
+O
+郫
+趁
+嗪
+囚
+蹶
+芒
+洁
+笋
+鹑
+敲
+硝
+啶
+堡
+渲
+揩
+』
+携
+宿
+遒
+颍
+扭
+棱
+割
+萜
+蔸
+葵
+琴
+捂
+饰
+衙
+耿
+掠
+募
+岂
+窖
+涟
+蔺
+瘤
+柞
+瞪
+怜
+匹
+距
+楔
+炜
+哆
+秦
+缎
+幼
+茁
+绪
+痨
+恨
+楸
+娅
+瓦
+桩
+雪
+嬴
+伏
+榔
+妥
+铿
+拌
+眠
+雍
+缇
+‘
+卓
+搓
+哌
+觞
+噩
+屈
+哧
+髓
+咦
+巅
+娑
+侑
+淫
+膳
+祝
+勾
+姊
+莴
+胄
+疃
+薛
+蜷
+胛
+巷
+芙
+芋
+熙
+闰
+勿
+窃
+狱
+剩
+钏
+幢
+陟
+铛
+慧
+靴
+耍
+k
+浙
+浇
+飨
+惟
+绗
+祜
+澈
+啼
+咪
+磷
+摞
+诅
+郦
+抹
+跃
+壬
+吕
+肖
+琏
+颤
+尴
+剡
+抠
+凋
+赚
+泊
+津
+宕
+殷
+倔
+氲
+漫
+邺
+涎
+怠
+$
+垮
+荬
+遵
+俏
+叹
+噢
+饽
+蜘
+孙
+筵
+疼
+鞭
+羧
+牦
+箭
+潴
+c
+眸
+祭
+髯
+啖
+坳
+愁
+芩
+驮
+倡
+巽
+穰
+沃
+胚
+怒
+凤
+槛
+剂
+趵
+嫁
+v
+邢
+灯
+鄢
+桐
+睽
+檗
+锯
+槟
+婷
+嵋
+圻
+诗
+蕈
+颠
+遭
+痢
+芸
+怯
+馥
+竭
+锗
+徜
+恭
+遍
+籁
+剑
+嘱
+苡
+龄
+僧
+桑
+潸
+弘
+澶
+楹
+悲
+讫
+愤
+腥
+悸
+谍
+椹
+呢
+桓
+葭
+攫
+阀
+翰
+躲
+敖
+柑
+郎
+笨
+橇
+呃
+魁
+燎
+脓
+葩
+磋
+垛
+玺
+狮
+沓
+砜
+蕊
+锺
+罹
+蕉
+翱
+虐
+闾
+巫
+旦
+茱
+嬷
+枯
+鹏
+贡
+芹
+汛
+矫
+绁
+拣
+禺
+佃
+讣
+舫
+惯
+乳
+趋
+疲
+挽
+岚
+虾
+衾
+蠹
+蹂
+飓
+氦
+铖
+孩
+稞
+瑜
+壅
+掀
+勘
+妓
+畅
+髋
+W
+庐
+牲
+蓿
+榕
+练
+垣
+唱
+邸
+菲
+昆
+婺
+穿
+绡
+麒
+蚱
+掂
+愚
+泷
+涪
+漳
+妩
+娉
+榄
+讷
+觅
+旧
+藤
+煮
+呛
+柳
+腓
+叭
+庵
+烷
+阡
+罂
+蜕
+擂
+猖
+咿
+媲
+脉
+【
+沏
+貅
+黠
+熏
+哲
+烁
+坦
+酵
+兜
+×
+潇
+撒
+剽
+珩
+圹
+乾
+摸
+樟
+帽
+嗒
+襄
+魂
+轿
+憬
+锡
+〕
+喃
+皆
+咖
+隅
+脸
+残
+泮
+袂
+鹂
+珊
+囤
+捆
+咤
+误
+徨
+闹
+淙
+芊
+淋
+怆
+囗
+拨
+梳
+渤
+R
+G
+绨
+蚓
+婀
+幡
+狩
+麾
+谢
+唢
+裸
+旌
+伉
+纶
+裂
+驳
+砼
+咛
+澄
+樨
+蹈
+宙
+澍
+倍
+貔
+操
+勇
+蟠
+摈
+砧
+虬
+够
+缁
+悦
+藿
+撸
+艹
+摁
+淹
+豇
+虎
+榭
+ˉ
+吱
+d
+°
+喧
+荀
+踱
+侮
+奋
+偕
+饷
+犍
+惮
+坑
+璎
+徘
+宛
+妆
+袈
+倩
+窦
+昂
+荏
+乖
+K
+怅
+撰
+鳙
+牙
+袁
+酞
+X
+痿
+琼
+闸
+雁
+趾
+荚
+虻
+涝
+《
+杏
+韭
+偈
+烤
+绫
+鞘
+卉
+症
+遢
+蓥
+诋
+杭
+荨
+匆
+竣
+簪
+辙
+敕
+虞
+丹
+缭
+咩
+黟
+m
+淤
+瑕
+咂
+铉
+硼
+茨
+嶂
+痒
+畸
+敬
+涿
+粪
+窘
+熟
+叔
+嫔
+盾
+忱
+裘
+憾
+梵
+赡
+珙
+咯
+娘
+庙
+溯
+胺
+葱
+痪
+摊
+荷
+卞
+乒
+髦
+寐
+铭
+坩
+胗
+枷
+爆
+溟
+嚼
+羚
+砬
+轨
+惊
+挠
+罄
+竽
+菏
+氧
+浅
+楣
+盼
+枢
+炸
+阆
+杯
+谏
+噬
+淇
+渺
+俪
+秆
+墓
+泪
+跻
+砌
+痰
+垡
+渡
+耽
+釜
+讶
+鳎
+煞
+呗
+韶
+舶
+绷
+鹳
+缜
+旷
+铊
+皱
+龌
+檀
+霖
+奄
+槐
+艳
+蝶
+旋
+哝
+赶
+骞
+蚧
+腊
+盈
+丁
+`
+蜚
+矸
+蝙
+睨
+嚓
+僻
+鬼
+醴
+夜
+彝
+磊
+笔
+拔
+栀
+糕
+厦
+邰
+纫
+逭
+纤
+眦
+膊
+馍
+躇
+烯
+蘼
+冬
+诤
+暄
+骶
+哑
+瘠
+」
+臊
+丕
+愈
+咱
+螺
+擅
+跋
+搏
+硪
+谄
+笠
+淡
+嘿
+骅
+谧
+鼎
+皋
+姚
+歼
+蠢
+驼
+耳
+胬
+挝
+涯
+狗
+蒽
+孓
+犷
+凉
+芦
+箴
+铤
+孤
+嘛
+坤
+V
+茴
+朦
+挞
+尖
+橙
+诞
+搴
+碇
+洵
+浚
+帚
+蜍
+漯
+柘
+嚎
+讽
+芭
+荤
+咻
+祠
+秉
+跖
+埃
+吓
+糯
+眷
+馒
+惹
+娼
+鲑
+嫩
+讴
+轮
+瞥
+靶
+褚
+乏
+缤
+宋
+帧
+删
+驱
+碎
+扑
+俩
+俄
+偏
+涣
+竹
+噱
+皙
+佰
+渚
+唧
+斡
+#
+镉
+刀
+崎
+筐
+佣
+夭
+贰
+肴
+峙
+哔
+艿
+匐
+牺
+镛
+缘
+仡
+嫡
+劣
+枸
+堀
+梨
+簿
+鸭
+蒸
+亦
+稽
+浴
+{
+衢
+束
+槲
+j
+阁
+揍
+疥
+棋
+潋
+聪
+窜
+乓
+睛
+插
+冉
+阪
+苍
+搽
+「
+蟾
+螟
+幸
+仇
+樽
+撂
+慢
+跤
+幔
+俚
+淅
+覃
+觊
+溶
+妖
+帛
+侨
+曰
+妾
+泗
+·
+：
+瀘
+風
+Ë
+（
+）
+∶
+紅
+紗
+瑭
+雲
+頭
+鶏
+財
+許
+•
+¥
+樂
+焗
+麗
+—
+；
+滙
+東
+榮
+繪
+興
+…
+門
+業
+π
+楊
+國
+顧
+é
+盤
+寳
+Λ
+龍
+鳳
+島
+誌
+緣
+結
+銭
+萬
+勝
+祎
+璟
+優
+歡
+臨
+時
+購
+＝
+★
+藍
+昇
+鐵
+觀
+勅
+農
+聲
+畫
+兿
+術
+發
+劉
+記
+專
+耑
+園
+書
+壴
+種
+Ο
+●
+褀
+號
+銀
+匯
+敟
+锘
+葉
+橪
+廣
+進
+蒄
+鑽
+阝
+祙
+貢
+鍋
+豊
+夬
+喆
+團
+閣
+開
+燁
+賓
+館
+酡
+沔
+順
+＋
+硚
+劵
+饸
+陽
+車
+湓
+復
+萊
+氣
+軒
+華
+堃
+迮
+纟
+戶
+馬
+學
+裡
+電
+嶽
+獨
+マ
+シ
+サ
+ジ
+燘
+袪
+環
+❤
+臺
+灣
+専
+賣
+孖
+聖
+攝
+線
+▪
+α
+傢
+俬
+夢
+達
+莊
+喬
+貝
+薩
+劍
+羅
+壓
+棛
+饦
+尃
+璈
+囍
+醫
+Ｇ
+Ｉ
+Ａ
+＃
+Ｎ
+鷄
+髙
+嬰
+啓
+約
+隹
+潔
+賴
+藝
+～
+寶
+籣
+麺
+　
+嶺
+√
+義
+網
+峩
+長
+∧
+魚
+機
+構
+②
+鳯
+偉
+Ｌ
+Ｂ
+㙟
+畵
+鴿
+＇
+詩
+溝
+嚞
+屌
+藔
+佧
+玥
+蘭
+織
+１
+３
+９
+０
+７
+點
+砭
+鴨
+鋪
+銘
+廳
+弍
+‧
+創
+湯
+坶
+℃
+卩
+骝
+＆
+烜
+荘
+當
+潤
+扞
+係
+懷
+碶
+钅
+蚨
+讠
+☆
+叢
+爲
+埗
+涫
+塗
+→
+楽
+現
+鯨
+愛
+瑪
+鈺
+忄
+悶
+藥
+飾
+樓
+視
+孬
+ㆍ
+燚
+苪
+師
+①
+丼
+锽
+│
+韓
+標
+è
+兒
+閏
+匋
+張
+漢
+Ü
+髪
+會
+閑
+檔
+習
+裝
+の
+峯
+菘
+輝
+И
+雞
+釣
+億
+浐
+Ｋ
+Ｏ
+Ｒ
+８
+Ｈ
+Ｅ
+Ｐ
+Ｔ
+Ｗ
+Ｄ
+Ｓ
+Ｃ
+Ｍ
+Ｆ
+姌
+饹
+»
+晞
+廰
+ä
+嵯
+鷹
+負
+飲
+絲
+冚
+楗
+澤
+綫
+區
+❋
+←
+質
+靑
+揚
+③
+滬
+統
+産
+協
+﹑
+乸
+畐
+經
+運
+際
+洺
+岽
+為
+粵
+諾
+崋
+豐
+碁
+ɔ
+Ｖ
+２
+６
+齋
+誠
+訂
+´
+勑
+雙
+陳
+無
+í
+泩
+媄
+夌
+刂
+ｉ
+ｃ
+ｔ
+ｏ
+ｒ
+ａ
+嘢
+耄
+燴
+暃
+壽
+媽
+靈
+抻
+體
+唻
+É
+冮
+甹
+鎮
+錦
+ʌ
+蜛
+蠄
+尓
+駕
+戀
+飬
+逹
+倫
+貴
+極
+Я
+Й
+寬
+磚
+嶪
+郎
+職
+｜
+間
+ｎ
+ｄ
+剎
+伈
+課
+飛
+橋
+瘊
+№
+譜
+骓
+圗
+滘
+縣
+粿
+咅
+養
+濤
+彳
+®
+％
+Ⅱ
+啰
+㴪
+見
+矞
+薬
+糁
+邨
+鲮
+顔
+罱
+З
+選
+話
+贏
+氪
+俵
+競
+瑩
+繡
+枱
+β
+綉
+á
+獅
+爾
+™
+麵
+戋
+淩
+徳
+個
+劇
+場
+務
+簡
+寵
+ｈ
+實
+膠
+轱
+圖
+築
+嘣
+樹
+㸃
+營
+耵
+孫
+饃
+鄺
+飯
+麯
+遠
+輸
+坫
+孃
+乚
+閃
+鏢
+㎡
+題
+廠
+關
+↑
+爺
+將
+軍
+連
+篦
+覌
+參
+箸
+－
+窠
+棽
+寕
+夀
+爰
+歐
+呙
+閥
+頡
+熱
+雎
+垟
+裟
+凬
+勁
+帑
+馕
+夆
+疌
+枼
+馮
+貨
+蒤
+樸
+彧
+旸
+靜
+龢
+暢
+㐱
+鳥
+珺
+鏡
+灡
+爭
+堷
+廚
+Ó
+騰
+診
+┅
+蘇
+褔
+凱
+頂
+豕
+亞
+帥
+嘬
+⊥
+仺
+桖
+複
+饣
+絡
+穂
+顏
+棟
+納
+▏
+濟
+親
+設
+計
+攵
+埌
+烺
+ò
+頤
+燦
+蓮
+撻
+節
+講
+濱
+濃
+娽
+洳
+朿
+燈
+鈴
+護
+膚
+铔
+過
+補
+Ｚ
+Ｕ
+５
+４
+坋
+闿
+䖝
+餘
+缐
+铞
+貿
+铪
+桼
+趙
+鍊
+［
+㐂
+垚
+菓
+揸
+捲
+鐘
+滏
+𣇉
+爍
+輪
+燜
+鴻
+鮮
+動
+鹞
+鷗
+丄
+慶
+鉌
+翥
+飮
+腸
+⇋
+漁
+覺
+來
+熘
+昴
+翏
+鲱
+圧
+鄉
+萭
+頔
+爐
+嫚
+г
+貭
+類
+聯
+幛
+輕
+訓
+鑒
+夋
+锨
+芃
+珣
+䝉
+扙
+嵐
+銷
+處
+ㄱ
+語
+誘
+苝
+歸
+儀
+燒
+楿
+內
+粢
+葒
+奧
+麥
+礻
+滿
+蠔
+穵
+瞭
+態
+鱬
+榞
+硂
+鄭
+黃
+煙
+祐
+奓
+逺
+＊
+瑄
+獲
+聞
+薦
+讀
+這
+樣
+決
+問
+啟
+們
+執
+説
+轉
+單
+隨
+唘
+帶
+倉
+庫
+還
+贈
+尙
+皺
+■
+餅
+產
+○
+∈
+報
+狀
+楓
+賠
+琯
+嗮
+禮
+｀
+傳
+＞
+≤
+嗞
+Φ
+≥
+換
+咭
+∣
+↓
+曬
+ε
+応
+寫
+″
+終
+様
+純
+費
+療
+聨
+凍
+壐
+郵
+ü
+黒
+∫
+製
+塊
+調
+軽
+確
+撃
+級
+馴
+Ⅲ
+涇
+繹
+數
+碼
+證
+狒
+処
+劑
+＜
+晧
+賀
+衆
+］
+櫥
+兩
+陰
+絶
+對
+鯉
+憶
+◎
+ｐ
+ｅ
+Ｙ
+蕒
+煖
+頓
+測
+試
+鼽
+僑
+碩
+妝
+帯
+≈
+鐡
+舖
+權
+喫
+倆
+ˋ
+該
+悅
+ā
+俫
+．
+ｆ
+ｓ
+ｂ
+ｍ
+ｋ
+ｇ
+ｕ
+ｊ
+貼
+淨
+濕
+針
+適
+備
+ｌ
+／
+給
+謢
+強
+觸
+衛
+與
+⊙
+＄
+緯
+變
+⑴
+⑵
+⑶
+㎏
+殺
+∩
+幚
+─
+價
+▲
+離
+ú
+ó
+飄
+烏
+関
+閟
+﹝
+﹞
+邏
+輯
+鍵
+驗
+訣
+導
+歷
+屆
+層
+▼
+儱
+錄
+熳
+ē
+艦
+吋
+錶
+辧
+飼
+顯
+④
+禦
+販
+気
+対
+枰
+閩
+紀
+幹
+瞓
+貊
+淚
+△
+眞
+墊
+Ω
+獻
+褲
+縫
+緑
+亜
+鉅
+餠
+｛
+｝
+◆
+蘆
+薈
+█
+◇
+溫
+彈
+晳
+粧
+犸
+穩
+訊
+崬
+凖
+熥
+П
+舊
+條
+紋
+圍
+Ⅳ
+筆
+尷
+難
+雜
+錯
+綁
+識
+頰
+鎖
+艶
+□
+殁
+殼
+⑧
+├
+▕
+鵬
+ǐ
+ō
+ǒ
+糝
+綱
+▎
+μ
+盜
+饅
+醬
+籤
+蓋
+釀
+鹽
+據
+à
+ɡ
+辦
+◥
+彐
+┌
+婦
+獸
+鲩
+伱
+ī
+蒟
+蒻
+齊
+袆
+腦
+寧
+凈
+妳
+煥
+詢
+偽
+謹
+啫
+鯽
+騷
+鱸
+損
+傷
+鎻
+髮
+買
+冏
+儥
+両
+﹢
+∞
+載
+喰
+ｚ
+羙
+悵
+燙
+曉
+員
+組
+徹
+艷
+痠
+鋼
+鼙
+縮
+細
+嚒
+爯
+≠
+維
+＂
+鱻
+壇
+厍
+帰
+浥
+犇
+薡
+軎
+²
+應
+醜
+刪
+緻
+鶴
+賜
+噁
+軌
+尨
+镔
+鷺
+槗
+彌
+葚
+濛
+請
+溇
+緹
+賢
+訪
+獴
+瑅
+資
+縤
+陣
+蕟
+栢
+韻
+祼
+恁
+伢
+謝
+劃
+涑
+總
+衖
+踺
+砋
+凉
+籃
+駿
+苼
+瘋
+昽
+紡
+驊
+腎
+﹗
+響
+杋
+剛
+嚴
+禪
+歓
+槍
+傘
+檸
+檫
+炣
+勢
+鏜
+鎢
+銑
+尐
+減
+奪
+惡
+θ
+僮
+婭
+臘
+ū
+ì
+殻
+鉄
+∑
+蛲
+焼
+緖
+續
+紹
+懮
diff --git a/deploy/slim/auto_compression/ppocrv4_det_server_dataset_process.py b/deploy/slim/auto_compression/ppocrv4_det_server_dataset_process.py
new file mode 100644
index 0000000..f99e901
--- /dev/null
+++ b/deploy/slim/auto_compression/ppocrv4_det_server_dataset_process.py
@@ -0,0 +1,33 @@
+import os
+import cv2
+
+dataset_path = "datasets/v4_4_test_dataset"
+annotation_file = "datasets/v4_4_test_dataset/label.txt"
+
+small_images_path = "datasets/v4_4_test_dataset_small"
+new_annotation_file = "datasets/v4_4_test_dataset_small/label.txt"
+
+os.makedirs(small_images_path, exist_ok=True)
+
+with open(annotation_file, "r") as f:
+    lines = f.readlines()
+
+for i, line in enumerate(lines):
+    image_name = line.split("   ")[0]
+
+    image_path = os.path.join(dataset_path, image_name)
+
+    try:
+        image = cv2.imread(image_path)
+        height, width, _ = image.shape
+
+        # 如果图像的宽度和高度都小于2000而且长宽比小于2，将其复制到新的文件夹，并保存其标注信息
+        if height < 2000 and width < 2000:
+            if max(height, width) / min(height, width) < 2:
+                print(i, height, width, image_path)
+                small_image_path = os.path.join(small_images_path, image_name)
+                cv2.imwrite(small_image_path, image)
+                with open(new_annotation_file, "a") as f:
+                    f.write(f"{line}")
+    except:
+        continue
diff --git a/deploy/slim/auto_compression/run.py b/deploy/slim/auto_compression/run.py
new file mode 100644
index 0000000..32bdede
--- /dev/null
+++ b/deploy/slim/auto_compression/run.py
@@ -0,0 +1,171 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+from tqdm import tqdm
+import numpy as np
+import argparse
+import paddle
+from paddleslim.common import load_config as load_slim_config
+from paddleslim.common import get_logger
+from paddleslim.auto_compression import AutoCompression
+from paddleslim.common.dataloader import get_feed_vars
+
+import sys
+
+sys.path.append("../../../")
+from ppocr.data import build_dataloader
+from ppocr.postprocess import build_post_process
+from ppocr.metrics import build_metric
+
+logger = get_logger(__name__, level=logging.INFO)
+
+
+def argsparser():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument(
+        "--config_path",
+        type=str,
+        default=None,
+        help="path of compression strategy config.",
+        required=True,
+    )
+    parser.add_argument(
+        "--save_dir",
+        type=str,
+        default="output",
+        help="directory to save compressed model.",
+    )
+    parser.add_argument(
+        "--devices", type=str, default="gpu", help="which device used to compress."
+    )
+    return parser
+
+
+def reader_wrapper(reader, input_name):
+    if isinstance(input_name, list) and len(input_name) == 1:
+        input_name = input_name[0]
+
+    def gen():  # 形成一个字典输入
+        for i, batch in enumerate(reader()):
+            yield {input_name: batch[0]}
+
+    return gen
+
+
+def eval_function(exe, compiled_test_program, test_feed_names, test_fetch_list):
+    post_process_class = build_post_process(all_config["PostProcess"], global_config)
+    eval_class = build_metric(all_config["Metric"])
+    model_type = global_config["model_type"]
+
+    with tqdm(
+        total=len(val_loader),
+        bar_format="Evaluation stage, Run batch:|{bar}| {n_fmt}/{total_fmt}",
+        ncols=80,
+    ) as t:
+        for batch_id, batch in enumerate(val_loader):
+            images = batch[0]
+
+            try:
+                (preds,) = exe.run(
+                    compiled_test_program,
+                    feed={test_feed_names[0]: images},
+                    fetch_list=test_fetch_list,
+                )
+            except:
+                preds, _ = exe.run(
+                    compiled_test_program,
+                    feed={test_feed_names[0]: images},
+                    fetch_list=test_fetch_list,
+                )
+
+            batch_numpy = []
+            for item in batch:
+                batch_numpy.append(np.array(item))
+
+            if model_type == "det":
+                preds_map = {"maps": preds}
+                post_result = post_process_class(preds_map, batch_numpy[1])
+                eval_class(post_result, batch_numpy)
+            elif model_type == "rec":
+                post_result = post_process_class(preds, batch_numpy[1])
+                eval_class(post_result, batch_numpy)
+            t.update()
+        metric = eval_class.get_metric()
+    logger.info("metric eval ***************")
+    for k, v in metric.items():
+        logger.info("{}:{}".format(k, v))
+
+    if model_type == "det":
+        return metric["hmean"]
+    elif model_type == "rec":
+        return metric["acc"]
+    return metric
+
+
+def main():
+    rank_id = paddle.distributed.get_rank()
+    if args.devices == "gpu":
+        place = paddle.CUDAPlace(rank_id)
+        paddle.set_device("gpu")
+    else:
+        place = paddle.CPUPlace()
+        paddle.set_device("cpu")
+
+    global all_config, global_config
+    all_config = load_slim_config(args.config_path)
+
+    if "Global" not in all_config:
+        raise KeyError(f"Key 'Global' not found in config file. \n{all_config}")
+    global_config = all_config["Global"]
+
+    gpu_num = paddle.distributed.get_world_size()
+
+    train_dataloader = build_dataloader(all_config, "Train", args.devices, logger)
+
+    global val_loader
+    val_loader = build_dataloader(all_config, "Eval", args.devices, logger)
+
+    if (
+        isinstance(all_config["TrainConfig"]["learning_rate"], dict)
+        and all_config["TrainConfig"]["learning_rate"]["type"] == "CosineAnnealingDecay"
+    ):
+        steps = len(train_dataloader) * all_config["TrainConfig"]["epochs"]
+        all_config["TrainConfig"]["learning_rate"]["T_max"] = steps
+        print("total training steps:", steps)
+
+    global_config["input_name"] = get_feed_vars(
+        global_config["model_dir"],
+        global_config["model_filename"],
+        global_config["params_filename"],
+    )
+
+    ac = AutoCompression(
+        model_dir=global_config["model_dir"],
+        model_filename=global_config["model_filename"],
+        params_filename=global_config["params_filename"],
+        save_dir=args.save_dir,
+        config=all_config,
+        train_dataloader=reader_wrapper(train_dataloader, global_config["input_name"]),
+        eval_callback=eval_function if rank_id == 0 else None,
+        eval_dataloader=reader_wrapper(val_loader, global_config["input_name"]),
+    )
+    ac.compress()
+
+
+if __name__ == "__main__":
+    paddle.enable_static()
+    parser = argsparser()
+    args = parser.parse_args()
+    main()
diff --git a/deploy/slim/auto_compression/test_ocr.py b/deploy/slim/auto_compression/test_ocr.py
new file mode 100644
index 0000000..65868f6
--- /dev/null
+++ b/deploy/slim/auto_compression/test_ocr.py
@@ -0,0 +1,292 @@
+# Copyright (c) 2023 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import argparse
+import time
+import os
+import sys
+import cv2
+import numpy as np
+import paddle
+import logging
+import numpy as np
+import argparse
+from tqdm import tqdm
+import paddle
+from paddleslim.common import load_config as load_slim_config
+from paddleslim.common import get_logger
+
+import sys
+
+sys.path.append("../../../")
+from ppocr.data import build_dataloader
+from ppocr.postprocess import build_post_process
+from ppocr.metrics import build_metric
+
+from paddle.inference import create_predictor, PrecisionType
+from paddle.inference import Config as PredictConfig
+
+logger = get_logger(__name__, level=logging.INFO)
+
+
+def find_images_with_bounding_size(dataset: paddle.io.Dataset):
+    max_length_index = -1
+    max_width_index = -1
+    min_length_index = -1
+    min_width_index = -1
+
+    max_length = float("-inf")
+    max_width = float("-inf")
+    min_length = float("inf")
+    min_width = float("inf")
+    for idx, data in enumerate(dataset):
+        image = np.array(data[0])
+        h, w = image.shape[-2:]
+        if h > max_length:
+            max_length = h
+            max_length_index = idx
+        if w > max_width:
+            max_width = w
+            max_width_index = idx
+        if h < min_length:
+            min_length = h
+            min_length_index = idx
+        if w < min_width:
+            min_width = w
+            min_width_index = idx
+    print(f"Found max image length: {max_length}, index: {max_length_index}")
+    print(f"Found max image width: {max_width}, index: {max_width_index}")
+    print(f"Found min image length: {min_length}, index: {min_length_index}")
+    print(f"Found min image width: {min_width}, index: {min_width_index}")
+    return paddle.io.Subset(
+        dataset, [max_width_index, max_length_index, min_width_index, min_length_index]
+    )
+
+
+def load_predictor(args):
+    """
+    load predictor func
+    """
+    rerun_flag = False
+    model_file = os.path.join(args.model_path, args.model_filename)
+    params_file = os.path.join(args.model_path, args.params_filename)
+    pred_cfg = PredictConfig(model_file, params_file)
+    pred_cfg.enable_memory_optim()
+    pred_cfg.switch_ir_optim(True)
+    if args.device == "GPU":
+        pred_cfg.enable_use_gpu(100, 0)
+    else:
+        pred_cfg.disable_gpu()
+        pred_cfg.set_cpu_math_library_num_threads(args.cpu_threads)
+        if args.use_mkldnn:
+            pred_cfg.enable_mkldnn()
+            if args.precision == "int8":
+                pred_cfg.enable_mkldnn_int8({"conv2d"})
+
+            if global_config["model_type"] == "rec":
+                # delete pass which influence the accuracy, please refer to https://github.com/PaddlePaddle/Paddle/issues/55290
+                pred_cfg.delete_pass("fc_mkldnn_pass")
+                pred_cfg.delete_pass("fc_act_mkldnn_fuse_pass")
+
+    if args.use_trt:
+        # To collect the dynamic shapes of inputs for TensorRT engine
+        dynamic_shape_file = os.path.join(args.model_path, "dynamic_shape.txt")
+        if os.path.exists(dynamic_shape_file):
+            pred_cfg.enable_tuned_tensorrt_dynamic_shape(dynamic_shape_file, True)
+            print("trt set dynamic shape done!")
+            precision_map = {
+                "fp16": PrecisionType.Half,
+                "fp32": PrecisionType.Float32,
+                "int8": PrecisionType.Int8,
+            }
+            if (
+                args.precision == "int8"
+                and "ppocrv4_det_server_qat_dist.yaml" in args.config_path
+            ):
+                # Use the following settings only when the hardware is a Tesla V100. If you are using
+                # a RTX 3090, use the settings in the else branch.
+                pred_cfg.enable_tensorrt_engine(
+                    workspace_size=1 << 30,
+                    max_batch_size=1,
+                    min_subgraph_size=30,
+                    precision_mode=precision_map[args.precision],
+                    use_static=True,
+                    use_calib_mode=False,
+                )
+                pred_cfg.exp_disable_tensorrt_ops(["elementwise_add"])
+            else:
+                pred_cfg.enable_tensorrt_engine(
+                    workspace_size=1 << 30,
+                    max_batch_size=1,
+                    min_subgraph_size=4,
+                    precision_mode=precision_map[args.precision],
+                    use_static=True,
+                    use_calib_mode=False,
+                )
+        else:
+            # pred_cfg.disable_gpu()
+            # pred_cfg.set_cpu_math_library_num_threads(24)
+            pred_cfg.collect_shape_range_info(dynamic_shape_file)
+            print("Start collect dynamic shape...")
+            rerun_flag = True
+
+    predictor = create_predictor(pred_cfg)
+    return predictor, rerun_flag
+
+
+def eval(args):
+    """
+    eval mIoU func
+    """
+    # DataLoader need run on cpu
+    paddle.set_device("cpu")
+    devices = paddle.device.get_device().split(":")[0]
+
+    val_loader = build_dataloader(all_config, "Eval", devices, logger)
+    post_process_class = build_post_process(all_config["PostProcess"], global_config)
+    eval_class = build_metric(all_config["Metric"])
+    model_type = global_config["model_type"]
+
+    predictor, rerun_flag = load_predictor(args)
+
+    if rerun_flag:
+        eval_dataset = find_images_with_bounding_size(val_loader.dataset)
+        batch_sampler = paddle.io.BatchSampler(
+            eval_dataset, batch_size=1, shuffle=False, drop_last=False
+        )
+        val_loader = paddle.io.DataLoader(
+            eval_dataset, batch_sampler=batch_sampler, num_workers=4, return_list=True
+        )
+
+    input_names = predictor.get_input_names()
+    input_handle = predictor.get_input_handle(input_names[0])
+    output_names = predictor.get_output_names()
+    output_handle = predictor.get_output_handle(output_names[0])
+    sample_nums = len(val_loader)
+    predict_time = 0.0
+    time_min = float("inf")
+    time_max = float("-inf")
+    print("Start evaluating ( total_iters: {}).".format(sample_nums))
+
+    for batch_id, batch in enumerate(val_loader):
+        images = np.array(batch[0])
+
+        batch_numpy = []
+        for item in batch:
+            batch_numpy.append(np.array(item))
+
+        # ori_shape = np.array(batch_numpy).shape[-2:]
+        input_handle.reshape(images.shape)
+        input_handle.copy_from_cpu(images)
+        start_time = time.time()
+
+        predictor.run()
+        preds = output_handle.copy_to_cpu()
+
+        end_time = time.time()
+        timed = end_time - start_time
+        time_min = min(time_min, timed)
+        time_max = max(time_max, timed)
+        predict_time += timed
+
+        if model_type == "det":
+            preds_map = {"maps": preds}
+            post_result = post_process_class(preds_map, batch_numpy[1])
+            eval_class(post_result, batch_numpy)
+        elif model_type == "rec":
+            post_result = post_process_class(preds, batch_numpy[1])
+            eval_class(post_result, batch_numpy)
+
+        if rerun_flag:
+            if batch_id == 3:
+                print(
+                    "***** Collect dynamic shape done, Please rerun the program to get correct results. *****"
+                )
+                return
+        if batch_id % 100 == 0:
+            print("Eval iter:", batch_id)
+            sys.stdout.flush()
+
+    metric = eval_class.get_metric()
+
+    time_avg = predict_time / sample_nums
+    print(
+        "[Benchmark] Inference time(ms): min={}, max={}, avg={}".format(
+            round(time_min * 1000, 2),
+            round(time_max * 1000, 1),
+            round(time_avg * 1000, 1),
+        )
+    )
+    for k, v in metric.items():
+        print("{}:{}".format(k, v))
+    sys.stdout.flush()
+
+
+def main():
+    global all_config, global_config
+    all_config = load_slim_config(args.config_path)
+    global_config = all_config["Global"]
+    eval(args)
+
+
+if __name__ == "__main__":
+    paddle.enable_static()
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--model_path", type=str, help="inference model filepath")
+    parser.add_argument(
+        "--config_path",
+        type=str,
+        default="./configs/ppocrv3_det_qat_dist.yaml",
+        help="path of compression strategy config.",
+    )
+    parser.add_argument(
+        "--model_filename",
+        type=str,
+        default="inference.pdmodel",
+        help="model file name",
+    )
+    parser.add_argument(
+        "--params_filename",
+        type=str,
+        default="inference.pdiparams",
+        help="params file name",
+    )
+    parser.add_argument(
+        "--device",
+        type=str,
+        default="GPU",
+        choices=["CPU", "GPU"],
+        help="Choose the device you want to run, it can be: CPU/GPU, default is GPU",
+    )
+    parser.add_argument(
+        "--precision",
+        type=str,
+        default="fp32",
+        choices=["fp32", "fp16", "int8"],
+        help="The precision of inference. It can be 'fp32', 'fp16' or 'int8'. Default is 'fp16'.",
+    )
+    parser.add_argument(
+        "--use_trt",
+        type=bool,
+        default=False,
+        help="Whether to use tensorrt engine or not.",
+    )
+    parser.add_argument(
+        "--use_mkldnn", type=bool, default=False, help="Whether use mkldnn or not."
+    )
+    parser.add_argument(
+        "--cpu_threads", type=int, default=10, help="Num of cpu threads."
+    )
+    args = parser.parse_args()
+    main()
diff --git a/deploy/slim/auto_compression/test_ocr_det.sh b/deploy/slim/auto_compression/test_ocr_det.sh
new file mode 100644
index 0000000..db73041
--- /dev/null
+++ b/deploy/slim/auto_compression/test_ocr_det.sh
@@ -0,0 +1,88 @@
+#!/bin/bash
+
+# 本脚本用于测试PPOCRV4_det系列模型的自动压缩功能
+## 运行脚本前，请确保处于以下环境：
+## CUDA11.7+TensorRT8.4.2.4+Paddle2.5.2
+
+model_type="$1"
+
+if [ "$model_type" = "mobile" ]; then
+    echo "test ppocrv4_det_mobile model......"
+    ## 启动自动化压缩训练
+    CUDA_VISIBLE_DEVICES=0 python run.py --save_dir ./models/det_mobile_qat --config_path configs/ppocrv4/ppocrv4_det_qat_dist.yaml
+
+    ## GPU指标测试
+    ### 量化前，预期指标：hmean:72.71%;time:4.7ms
+    python test_ocr.py --model_path ./models/ch_PP-OCRv4_det_infer --config ./configs/ppocrv4/ppocrv4_det_qat_dist.yaml --precision fp32 --use_trt True
+    ### 量化后，预期指标：hmean:71.38%;time:3.3ms
+    python test_ocr.py --model_path ./models/det_mobile_qat --config ./configs/ppocrv4/ppocrv4_det_qat_dist.yaml --precision int8 --use_trt True
+
+    ## CPU指标测试
+    ### 量化前，预期指标：hmean:72.71%;time:198.4ms
+    python test_ocr.py --model_path ./models/ch_PP-OCRv4_det_infer --config ./configs/ppocrv4/ppocrv4_det_qat_dist.yaml --precision fp32 --use_mkldnn True --device CPU --cpu_threads 12
+    ### 量化后，预期指标：hmean:72.30%;time:205.2ms
+    python test_ocr.py --model_path ./models/det_mobile_qat --config ./configs/ppocrv4/ppocrv4_det_qat_dist.yaml --precision int8 --use_mkldnn True --device CPU --cpu_threads 12
+
+    # 量化前模型推理
+    # GPU
+    python tools/infer/predict_det.py --det_model_dir deploy/slim/auto_compression/models/ch_PP-OCRv4_det_infer \
+        --benchmark True --image_dir deploy/slim/auto_compression/datasets/v4_4_test_dataset --use_gpu True \
+        --use_tensorrt True --warmup True --precision fp32
+
+    # CPU
+    python tools/infer/predict_det.py --det_model_dir deploy/slim/auto_compression/models/ch_PP-OCRv4_det_infer \
+        --benchmark True --image_dir deploy/slim/auto_compression/datasets/v4_4_test_dataset --use_gpu False \
+        --enable_mkldnn True --warmup True --precision fp32
+
+    # 量化后模型推理
+    # GPU
+    python tools/infer/predict_det.py --det_model_dir deploy/slim/auto_compression/models/det_mobile_qat \
+        --benchmark True --image_dir deploy/slim/auto_compression/datasets/v4_4_test_dataset --use_gpu True \
+        --use_tensorrt True --warmup True --precision int8
+
+    # CPU
+    python tools/infer/predict_det.py --det_model_dir deploy/slim/auto_compression/models/det_mobile_qat \
+        --benchmark True --image_dir deploy/slim/auto_compression/datasets/v4_4_test_dataset --use_gpu False \
+        --enable_mkldnn True --warmup True --precision int8
+
+elif [ "$model_type" = "server" ]; then
+    echo "test ppocrv4_det_server model......"
+    ## 启动自动化压缩训练
+    CUDA_VISIBLE_DEVICES=0 python run.py --save_dir ./models/det_server_qat --config_path configs/ppocrv4/ppocrv4_det_server_qat_dist.yaml
+
+    ## GPU指标测试
+    ### 量化前，预期指标：hmean:79.77%;time:50.0ms
+    python test_ocr.py --model_path ./models/ch_PP-OCRv4_det_server_infer --config ./configs/ppocrv4/ppocrv4_det_server_qat_dist.yaml --precision fp32 --use_trt True
+    ### 量化后，预期指标：hmean:79.81%;time:42.4ms
+    python test_ocr.py --model_path ./models/det_server_qat --config ./configs/ppocrv4/ppocrv4_det_server_qat_dist.yaml --precision int8 --use_trt True
+
+    ## CPU指标测试
+    ### 量化前，预期指标：hmean:79.77%;time:2159.4ms
+    python test_ocr.py --model_path ./models/ch_PP-OCRv4_det_server_infer --config ./configs/ppocrv4/ppocrv4_det_server_qat_dist.yaml --precision fp32 --use_mkldnn True --device CPU --cpu_threads 12
+    ### 量化后，预期指标：hmean:79.69%;time:1834.8ms
+    python test_ocr.py --model_path ./models/det_server_qat --config ./configs/ppocrv4/ppocrv4_det_server_qat_dist.yaml --precision int8 --use_mkldnn True --device CPU --cpu_threads 12
+
+    ## 量化前模型推理
+    ### GPU
+    python tools/infer/predict_det.py --det_model_dir deploy/slim/auto_compression/models/ch_PP-OCRv4_det_server_infer \
+        --benchmark True --image_dir deploy/slim/auto_compression/datasets/v4_4_test_dataset --use_gpu True \
+        --use_tensorrt True --warmup True --precision fp32
+
+    ### CPU
+    python tools/infer/predict_det.py --det_model_dir deploy/slim/auto_compression/models/ch_PP-OCRv4_det_server_infer \
+        --benchmark True --image_dir deploy/slim/auto_compression/datasets/v4_4_test_dataset --use_gpu False \
+        --enable_mkldnn True --warmup True --precision fp32
+
+    ## 量化后模型推理
+    ### GPU
+    python tools/infer/predict_det.py --det_model_dir deploy/slim/auto_compression/models/det_server_qat \
+        --benchmark True --image_dir deploy/slim/auto_compression/datasets/v4_4_test_dataset --use_gpu True \
+        --use_tensorrt True --warmup True --precision int8
+
+    ### CPU
+    python tools/infer/predict_det.py --det_model_dir deploy/slim/auto_compression/models/det_server_qat \
+        --benchmark True --image_dir deploy/slim/auto_compression/datasets/v4_4_test_dataset --use_gpu False \
+        --enable_mkldnn True --warmup True --precision int8
+else
+    echo "unrecgnized model_type"
+fi
diff --git a/deploy/slim/auto_compression/test_ocr_rec.sh b/deploy/slim/auto_compression/test_ocr_rec.sh
new file mode 100644
index 0000000..1a1e1f7
--- /dev/null
+++ b/deploy/slim/auto_compression/test_ocr_rec.sh
@@ -0,0 +1,88 @@
+#!/bin/bash
+
+# 本脚本用于测试PPOCRV4_rec系列模型的自动压缩功能
+## 运行脚本前，请确保处于以下环境：
+## CUDA11.2+TensorRT8.0.3.4+Paddle2.5.2
+
+model_type="$1"
+
+if [ "$model_type" = "mobile" ]; then
+    echo "test ppocrv4_rec_mobile model......"
+    ## 启动自动化压缩训练
+    CUDA_VISIBLE_DEVICES=0 python run.py --save_dir ./models/rec_mobile_qat --config_path configs/ppocrv4/ppocrv4_rec_qat_dist.yaml
+
+    ## GPU指标测试
+    ### 量化前，预期指标：accuracy:78.92%;time:1.7ms
+    python test_ocr.py --model_path ./models/ch_PP-OCRv4_rec_infer --config ./configs/ppocrv4/ppocrv4_rec_qat_dist.yaml --precision fp32 --use_trt True
+    ### 量化后，预期指标：accuracy:78.41%;time:1.4ms
+    python test_ocr.py --model_path ./models/rec_mobile_qat --config ./configs/ppocrv4/ppocrv4_rec_qat_dist.yaml --precision int8 --use_trt True
+
+    ## CPU指标测试
+    ### 量化前，预期指标：accuracy:78.92%;time:33.3ms
+    python test_ocr.py --model_path ./models/ch_PP-OCRv4_rec_infer --config ./configs/ppocrv4/ppocrv4_rec_qat_dist.yaml --precision fp32 --use_mkldnn True --device CPU --cpu_threads 12
+    ### 量化后，预期指标：accuracy:78.44%;time:34.0ms
+    python test_ocr.py --model_path ./models/rec_mobile_qat --config ./configs/ppocrv4/ppocrv4_rec_qat_dist.yaml --precision int8 --use_mkldnn True --device CPU --cpu_threads 12
+
+    # 量化前模型推理
+    # GPU
+    python tools/infer/predict_det.py --rec_model_dir deploy/slim/auto_compression/models/ch_PP-OCRv4_rec_infer \
+        --benchmark True --image_dir deploy/slim/auto_compression/datasets/v4_4_test_dataset --use_gpu True \
+        --use_tensorrt True --warmup True --precision fp32
+
+    # CPU
+    python tools/infer/predict_det.py --rec_model_dir deploy/slim/auto_compression/models/ch_PP-OCRv4_rec_infer \
+        --benchmark True --image_dir deploy/slim/auto_compression/datasets/v4_4_test_dataset --use_gpu False \
+        --enable_mkldnn True --warmup True --precision fp32
+
+    # 量化后模型推理
+    # GPU
+    python tools/infer/predict_det.py --rec_model_dir deploy/slim/auto_compression/models/rec_mobile_qat \
+        --benchmark True --image_dir deploy/slim/auto_compression/datasets/v4_4_test_dataset --use_gpu True \
+        --use_tensorrt True --warmup True --precision int8
+
+    # CPU
+    python tools/infer/predict_det.py --rec_model_dir deploy/slim/auto_compression/models/rec_mobile_qat \
+        --benchmark True --image_dir deploy/slim/auto_compression/datasets/v4_4_test_dataset --use_gpu False \
+        --enable_mkldnn True --warmup True --precision int8
+
+elif [ "$model_type" = "server" ]; then
+    echo "test ppocrv4_rec_server model......"
+    ## 启动自动化压缩训练
+    CUDA_VISIBLE_DEVICES=0 python run.py --save_dir ./models/rec_server_qat --config_path configs/ppocrv4/ppocrv4_rec_server_qat_dist.yaml
+
+    ## GPU指标测试
+    ### 量化前，预期指标：accuracy:81.62%;time:4.0ms
+    python test_ocr.py --model_path ./models/ch_PP-OCRv4_rec_server_infer --config ./configs/ppocrv4/ppocrv4_rec_server_qat_dist.yaml --precision fp32 --use_trt True
+    ### 量化后，预期指标：accuracy:81.03%;time:2.0ms
+    python test_ocr.py --model_path ./models/rec_server_qat --config ./configs/ppocrv4/ppocrv4_rec_server_qat_dist.yaml --precision int8 --use_trt True
+
+    ## CPU指标测试
+    ### 量化前，预期指标：accuracy:81.62%;time:62.5ms
+    python test_ocr.py --model_path ./models/ch_PP-OCRv4_rec_server_infer --config ./configs/ppocrv4/ppocrv4_rec_server_qat_dist.yaml --precision fp32 --use_mkldnn True --device CPU --cpu_threads 12
+    ### 量化后，预期指标：accuracy:81.00%;time:64.4ms
+    python test_ocr.py --model_path ./models/rec_server_qat --config ./configs/ppocrv4/ppocrv4_rec_server_qat_dist.yaml --precision int8 --use_mkldnn True --device CPU --cpu_threads 12
+
+    ## 量化前模型推理
+    ### GPU
+    python tools/infer/predict_det.py --rec_model_dir deploy/slim/auto_compression/models/ch_PP-OCRv4_rec_server_infer \
+        --benchmark True --image_dir deploy/slim/auto_compression/datasets/v4_4_test_dataset --use_gpu True \
+        --use_tensorrt True --warmup True --precision fp32
+
+    ### CPU
+    python tools/infer/predict_det.py --rec_model_dir deploy/slim/auto_compression/models/ch_PP-OCRv4_rec_server_infer \
+        --benchmark True --image_dir deploy/slim/auto_compression/datasets/v4_4_test_dataset --use_gpu False \
+        --enable_mkldnn True --warmup True --precision fp32
+
+    ## 量化后模型推理
+    ### GPU
+    python tools/infer/predict_det.py --rec_model_dir deploy/slim/auto_compression/models/rec_server_qat \
+        --benchmark True --image_dir deploy/slim/auto_compression/datasets/v4_4_test_dataset --use_gpu True \
+        --use_tensorrt True --warmup True --precision int8
+
+    ### CPU
+    python tools/infer/predict_det.py --rec_model_dir deploy/slim/auto_compression/models/rec_server_qat \
+        --benchmark True --image_dir deploy/slim/auto_compression/datasets/v4_4_test_dataset --use_gpu False \
+        --enable_mkldnn True --warmup True --precision int8
+else
+    echo "unrecgnized model_type"
+fi
diff --git a/deploy/slim/prune/README.md b/deploy/slim/prune/README.md
new file mode 100644
index 0000000..eba4b66
--- /dev/null
+++ b/deploy/slim/prune/README.md
@@ -0,0 +1,67 @@
+
+# PP-OCR模型裁剪
+
+复杂的模型有利于提高模型的性能，但也导致模型中存在一定冗余，模型裁剪通过移出网络模型中的子模型来减少这种冗余，达到减少模型计算复杂度，提高模型推理性能的目的。
+本教程将介绍如何使用飞桨模型压缩库PaddleSlim做PaddleOCR模型的压缩。
+[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim)集成了模型剪枝、量化（包括量化训练和离线量化）、蒸馏和神经网络搜索等多种业界常用且领先的模型压缩功能，如果您感兴趣，可以关注并了解。
+
+
+在开始本教程之前，建议先了解：
+1. [PaddleOCR模型的训练方法](../../../doc/doc_ch/training.md)
+2. [模型裁剪教程](https://github.com/PaddlePaddle/PaddleSlim/blob/release%2F2.0.0/docs/zh_cn/tutorials/pruning/dygraph/filter_pruning.md)
+
+## 快速开始
+
+模型裁剪主要包括四个步骤：
+
+1. 安装 PaddleSlim
+2. 准备训练好的模型
+3. 敏感度分析、裁剪训练
+4. 导出模型、预测部署
+
+### 1. 安装PaddleSlim
+
+```bash
+git clone https://github.com/PaddlePaddle/PaddleSlim.git
+cd PaddleSlim
+git checkout develop
+python3 setup.py install
+```
+
+### 2. 获取预训练模型
+模型裁剪需要加载事先训练好的模型，PaddleOCR也提供了一系列[模型](../../../doc/doc_ch/models_list.md)，开发者可根据需要自行选择模型或使用自己的模型。
+
+### 3. 敏感度分析训练
+
+加载预训练模型后，通过对现有模型的每个网络层进行敏感度分析，得到敏感度文件：sen.pickle，可以通过PaddleSlim提供的[接口](https://github.com/PaddlePaddle/PaddleSlim/blob/9b01b195f0c4bc34a1ab434751cb260e13d64d9e/paddleslim/dygraph/prune/filter_pruner.py#L75)加载文件，获得各网络层在不同裁剪比例下的精度损失。从而了解各网络层冗余度，决定每个网络层的裁剪比例。
+敏感度文件内容格式：
+```
+sen.pickle(Dict){
+            'layer_weight_name_0': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+            'layer_weight_name_1': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+        }
+
+例子：
+        {
+            'conv10_expand_weights': {0.1: 0.006509952684312718, 0.2: 0.01827734339798862, 0.3: 0.014528405644659832, 0.6: 0.06536008804270439, 0.8: 0.11798612250664964, 0.7: 0.12391408417493704, 0.4: 0.030615754498018757, 0.5: 0.047105205602406594}
+            'conv10_linear_weights': {0.1: 0.05113190831455035, 0.2: 0.07705573833558801, 0.3: 0.12096721757739311, 0.6: 0.5135061352930738, 0.8: 0.7908166677143281, 0.7: 0.7272187676899062, 0.4: 0.1819252083008504, 0.5: 0.3728054727792405}
+        }
+```
+
+加载敏感度文件后会返回一个字典，字典中的keys为网络模型参数模型的名字，values为一个字典，里面保存了相应网络层的裁剪敏感度信息。例如在例子中，conv10_expand_weights所对应的网络层在裁掉10%的卷积核后模型性能相较原模型会下降0.65%，详细信息可见[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/algo/algo.md#2-%E5%8D%B7%E7%A7%AF%E6%A0%B8%E5%89%AA%E8%A3%81%E5%8E%9F%E7%90%86)
+
+进入PaddleOCR根目录，通过以下命令对模型进行敏感度分析训练：
+```bash
+python3 deploy/slim/prune/sensitivity_anal.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model="your trained model" Global.save_model_dir=./output/prune_model/
+```
+
+### 4. 导出模型、预测部署
+
+在得到裁剪训练保存的模型后，我们可以将其导出为inference_model：
+```bash
+pytho3 deploy/slim/prune/export_prune_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model=./output/det_db/best_accuracy  Global.save_inference_dir=./prune/prune_inference_model
+```
+
+inference model的预测和部署参考：
+1. [inference model python端预测](../../../doc/doc_ch/inference.md)
+2. [inference model C++预测](../../cpp_infer/readme.md)
diff --git a/deploy/slim/prune/README_en.md b/deploy/slim/prune/README_en.md
new file mode 100644
index 0000000..c23e626
--- /dev/null
+++ b/deploy/slim/prune/README_en.md
@@ -0,0 +1,73 @@
+
+# PP-OCR Models Pruning
+
+Generally, a more complex model would achieve better performance in the task, but it also leads to some redundancy in the model. Model Pruning is a technique that reduces this redundancy by removing the sub-models in the neural network model, so as to reduce model calculation complexity and improve model inference performance.
+
+This example uses PaddleSlim provided[APIs of Pruning](https://github.com/PaddlePaddle/PaddleSlim/tree/develop/docs/zh_cn/api_cn/dygraph/pruners) to compress the OCR model.
+[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim), an open source library which integrates model pruning, quantization (including quantization training and offline quantization), distillation, neural network architecture search, and many other commonly used and leading model compression technique in the industry.
+
+It is recommended that you could understand following pages before reading this example：
+1. [PaddleOCR training methods](../../../doc/doc_ch/quickstart.md)
+2. [The demo of prune](https://github.com/PaddlePaddle/PaddleSlim/blob/release%2F2.0.0/docs/zh_cn/tutorials/pruning/dygraph/filter_pruning.md)
+
+## Quick start
+
+Five steps for OCR model prune:
+1. Install PaddleSlim
+2. Prepare the trained model
+3. Sensitivity analysis and tailoring training
+4. Export model, predict deployment
+
+### 1. Install PaddleSlim
+
+```bash
+git clone https://github.com/PaddlePaddle/PaddleSlim.git
+cd PaddleSlim
+git checkout develop
+python3 setup.py install
+```
+
+
+### 2. Download Pre-trained Model
+Model prune needs to load pre-trained models.
+PaddleOCR also provides a series of [models](../../../doc/doc_en/models_list_en.md). Developers can choose their own models or use their own models according to their needs.
+
+
+### 3. Pruning sensitivity analysis
+
+  After the pre-trained model is loaded, sensitivity analysis is performed on each network layer of the model to understand the redundancy of each network layer, and save a sensitivity file which named: sen.pickle.  After that, user could load the sensitivity file via the [methods provided by PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/paddleslim/prune/sensitive.py#L221) and determining the pruning ratio of each network layer automatically. For specific details of sensitivity analysis, see：[Sensitivity analysis](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/en/tutorials/image_classification_sensitivity_analysis_tutorial_en.md)
+  The data format of sensitivity file：
+
+```
+sen.pickle(Dict){
+              'layer_weight_name_0': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+              'layer_weight_name_1': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+          }
+example：
+          {
+              'conv10_expand_weights': {0.1: 0.006509952684312718, 0.2: 0.01827734339798862, 0.3: 0.014528405644659832, 0.6: 0.06536008804270439, 0.8: 0.11798612250664964, 0.7: 0.12391408417493704, 0.4: 0.030615754498018757, 0.5: 0.047105205602406594}
+              'conv10_linear_weights': {0.1: 0.05113190831455035, 0.2: 0.07705573833558801, 0.3: 0.12096721757739311, 0.6: 0.5135061352930738, 0.8: 0.7908166677143281, 0.7: 0.7272187676899062, 0.4: 0.1819252083008504, 0.5: 0.3728054727792405}
+          }
+  The function would return a dict after loading the sensitivity file. The keys of the dict are name of parameters in each layer. And the value of key is the information about pruning sensitivity of corresponding layer. In example, pruning 10% filter of the layer corresponding to conv10_expand_weights would lead to 0.65% degradation of model performance. The details could be seen at: [Sensitivity analysis](https://github.com/PaddlePaddle/PaddleSlim/blob/release/2.0-alpha/docs/zh_cn/algo/algo.md)
+```
+
+  The function would return a dict after loading the sensitivity file. The keys of the dict are name of parameters in each layer. And the value of key is the information about pruning sensitivity of corresponding layer. In example, pruning 10% filter of the layer corresponding to conv10_expand_weights would lead to 0.65% degradation of model performance. The details could be seen at: [Sensitivity analysis](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/algo/algo.md#2-%E5%8D%B7%E7%A7%AF%E6%A0%B8%E5%89%AA%E8%A3%81%E5%8E%9F%E7%90%86)
+
+
+Enter the PaddleOCR root directory，perform sensitivity analysis on the model with the following command：
+
+```bash
+python3 deploy/slim/prune/sensitivity_anal.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model="your trained model"  Global.save_model_dir=./output/prune_model/
+```
+
+
+### 5.  Export inference model and deploy it
+
+We can export the pruned model as inference_model for deployment:
+```bash
+python deploy/slim/prune/export_prune_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml  -o Global.pretrained_model=./output/det_db/best_accuracy  Global.save_inference_dir=./prune/prune_inference_model
+```
+
+Reference for prediction and deployment of inference model:
+1. [inference model python prediction](../../../doc/doc_en/inference_en.md)
+2. [inference model C++ prediction](../../cpp_infer/readme_en.md)
diff --git a/deploy/slim/prune/export_prune_model.py b/deploy/slim/prune/export_prune_model.py
new file mode 100644
index 0000000..6d19194
--- /dev/null
+++ b/deploy/slim/prune/export_prune_model.py
@@ -0,0 +1,137 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+__dir__ = os.path.dirname(__file__)
+sys.path.append(__dir__)
+sys.path.append(os.path.join(__dir__, "..", "..", ".."))
+sys.path.append(os.path.join(__dir__, "..", "..", "..", "tools"))
+
+import paddle
+from ppocr.data import build_dataloader, set_signal_handlers
+from ppocr.modeling.architectures import build_model
+
+from ppocr.postprocess import build_post_process
+from ppocr.metrics import build_metric
+from ppocr.utils.save_load import load_model
+import tools.program as program
+
+
+def main(config, device, logger, vdl_writer):
+    global_config = config["Global"]
+
+    # build dataloader
+    set_signal_handlers()
+    valid_dataloader = build_dataloader(config, "Eval", device, logger)
+
+    # build post process
+    post_process_class = build_post_process(config["PostProcess"], global_config)
+
+    # build model
+    # for rec algorithm
+    if hasattr(post_process_class, "character"):
+        char_num = len(getattr(post_process_class, "character"))
+        config["Architecture"]["Head"]["out_channels"] = char_num
+    model = build_model(config["Architecture"])
+
+    if config["Architecture"]["model_type"] == "det":
+        input_shape = [1, 3, 640, 640]
+    elif config["Architecture"]["model_type"] == "rec":
+        input_shape = [1, 3, 32, 320]
+
+    flops = paddle.flops(model, input_shape)
+    logger.info("FLOPs before pruning: {}".format(flops))
+
+    from paddleslim.dygraph import FPGMFilterPruner
+
+    model.train()
+    pruner = FPGMFilterPruner(model, input_shape)
+
+    # build metric
+    eval_class = build_metric(config["Metric"])
+
+    def eval_fn():
+        metric = program.eval(model, valid_dataloader, post_process_class, eval_class)
+        if config["Architecture"]["model_type"] == "det":
+            main_indicator = "hmean"
+        else:
+            main_indicator = "acc"
+        logger.info("metric[{}]: {}".format(main_indicator, metric[main_indicator]))
+        return metric[main_indicator]
+
+    params_sensitive = pruner.sensitive(
+        eval_func=eval_fn,
+        sen_file="./sen.pickle",
+        skip_vars=["conv2d_57.w_0", "conv2d_transpose_2.w_0", "conv2d_transpose_3.w_0"],
+    )
+
+    logger.info(
+        "The sensitivity analysis results of model parameters saved in sen.pickle"
+    )
+    # calculate pruned params's ratio
+    params_sensitive = pruner._get_ratios_by_loss(params_sensitive, loss=0.02)
+    for key in params_sensitive.keys():
+        logger.info("{}, {}".format(key, params_sensitive[key]))
+
+    plan = pruner.prune_vars(params_sensitive, [0])
+
+    flops = paddle.flops(model, input_shape)
+    logger.info("FLOPs after pruning: {}".format(flops))
+
+    # load pretrain model
+    load_model(config, model)
+    metric = program.eval(model, valid_dataloader, post_process_class, eval_class)
+    if config["Architecture"]["model_type"] == "det":
+        main_indicator = "hmean"
+    else:
+        main_indicator = "acc"
+    logger.info("metric['']: {}".format(main_indicator, metric[main_indicator]))
+
+    # start export model
+    from paddle.jit import to_static
+
+    infer_shape = [3, -1, -1]
+    if config["Architecture"]["model_type"] == "rec":
+        infer_shape = [3, 32, -1]  # for rec model, H must be 32
+
+        if (
+            "Transform" in config["Architecture"]
+            and config["Architecture"]["Transform"] is not None
+            and config["Architecture"]["Transform"]["name"] == "TPS"
+        ):
+            logger.info(
+                "When there is tps in the network, variable length input is not supported, and the input size needs to be the same as during training"
+            )
+            infer_shape[-1] = 100
+    model = to_static(
+        model,
+        input_spec=[
+            paddle.static.InputSpec(shape=[None] + infer_shape, dtype="float32")
+        ],
+    )
+
+    save_path = "{}/inference".format(config["Global"]["save_inference_dir"])
+    paddle.jit.save(model, save_path)
+    logger.info("inference model is saved to {}".format(save_path))
+
+
+if __name__ == "__main__":
+    config, device, logger, vdl_writer = program.preprocess(is_train=True)
+    main(config, device, logger, vdl_writer)
diff --git a/deploy/slim/prune/sensitivity_anal.py b/deploy/slim/prune/sensitivity_anal.py
new file mode 100644
index 0000000..1399925
--- /dev/null
+++ b/deploy/slim/prune/sensitivity_anal.py
@@ -0,0 +1,200 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+__dir__ = os.path.dirname(__file__)
+sys.path.append(__dir__)
+sys.path.append(os.path.join(__dir__, "..", "..", ".."))
+sys.path.append(os.path.join(__dir__, "..", "..", "..", "tools"))
+
+import paddle
+import paddle.distributed as dist
+from ppocr.data import build_dataloader, set_signal_handlers
+from ppocr.modeling.architectures import build_model
+from ppocr.losses import build_loss
+from ppocr.optimizer import build_optimizer
+from ppocr.postprocess import build_post_process
+from ppocr.metrics import build_metric
+from ppocr.utils.save_load import load_model
+import tools.program as program
+
+dist.get_world_size()
+
+
+def get_pruned_params(parameters):
+    params = []
+
+    for param in parameters:
+        if (
+            len(param.shape) == 4
+            and "depthwise" not in param.name
+            and "transpose" not in param.name
+            and "conv2d_57" not in param.name
+            and "conv2d_56" not in param.name
+        ):
+            params.append(param.name)
+    return params
+
+
+def main(config, device, logger, vdl_writer):
+    # init dist environment
+    if config["Global"]["distributed"]:
+        dist.init_parallel_env()
+
+    global_config = config["Global"]
+
+    # build dataloader
+    set_signal_handlers()
+    train_dataloader = build_dataloader(config, "Train", device, logger)
+    if config["Eval"]:
+        valid_dataloader = build_dataloader(config, "Eval", device, logger)
+    else:
+        valid_dataloader = None
+
+    # build post process
+    post_process_class = build_post_process(config["PostProcess"], global_config)
+
+    # build model
+    # for rec algorithm
+    if hasattr(post_process_class, "character"):
+        char_num = len(getattr(post_process_class, "character"))
+        config["Architecture"]["Head"]["out_channels"] = char_num
+    model = build_model(config["Architecture"])
+    if config["Architecture"]["model_type"] == "det":
+        input_shape = [1, 3, 640, 640]
+    elif config["Architecture"]["model_type"] == "rec":
+        input_shape = [1, 3, 32, 320]
+    flops = paddle.flops(model, input_shape)
+
+    logger.info("FLOPs before pruning: {}".format(flops))
+
+    from paddleslim.dygraph import FPGMFilterPruner
+
+    model.train()
+
+    pruner = FPGMFilterPruner(model, input_shape)
+
+    # build loss
+    loss_class = build_loss(config["Loss"])
+
+    # build optim
+    optimizer, lr_scheduler, _wd_scheduler = build_optimizer(
+        config["Optimizer"],
+        epochs=config["Global"]["epoch_num"],
+        step_each_epoch=len(train_dataloader),
+        model=model,
+    )
+
+    # build metric
+    eval_class = build_metric(config["Metric"])
+    # load pretrain model
+    pre_best_model_dict = load_model(config, model, optimizer)
+
+    logger.info(
+        "train dataloader has {} iters, valid dataloader has {} iters".format(
+            len(train_dataloader), len(valid_dataloader)
+        )
+    )
+    # build metric
+    eval_class = build_metric(config["Metric"])
+
+    logger.info(
+        "train dataloader has {} iters, valid dataloader has {} iters".format(
+            len(train_dataloader), len(valid_dataloader)
+        )
+    )
+
+    def eval_fn():
+        metric = program.eval(
+            model, valid_dataloader, post_process_class, eval_class, False
+        )
+        if config["Architecture"]["model_type"] == "det":
+            main_indicator = "hmean"
+        else:
+            main_indicator = "acc"
+
+        logger.info("metric[{}]: {}".format(main_indicator, metric[main_indicator]))
+        return metric[main_indicator]
+
+    run_sensitive_analysis = False
+    """
+    run_sensitive_analysis=True:
+        Automatically compute the sensitivities of convolutions in a model.
+        The sensitivity of a convolution is the losses of accuracy on test dataset in
+        different pruned ratios. The sensitivities can be used to get a group of best
+        ratios with some condition.
+
+    run_sensitive_analysis=False:
+        Set prune trim ratio to a fixed value, such as 10%. The larger the value,
+        the more convolution weights will be cropped.
+
+    """
+
+    if run_sensitive_analysis:
+        params_sensitive = pruner.sensitive(
+            eval_func=eval_fn,
+            sen_file="./deploy/slim/prune/sen.pickle",
+            skip_vars=[
+                "conv2d_57.w_0",
+                "conv2d_transpose_2.w_0",
+                "conv2d_transpose_3.w_0",
+            ],
+        )
+        logger.info(
+            "The sensitivity analysis results of model parameters saved in sen.pickle"
+        )
+        # calculate pruned params's ratio
+        params_sensitive = pruner._get_ratios_by_loss(params_sensitive, loss=0.02)
+        for key in params_sensitive.keys():
+            logger.info("{}, {}".format(key, params_sensitive[key]))
+    else:
+        params_sensitive = {}
+        for param in model.parameters():
+            if "transpose" not in param.name and "linear" not in param.name:
+                # set prune ratio as 10%. The larger the value, the more convolution weights will be cropped
+                params_sensitive[param.name] = 0.1
+
+    plan = pruner.prune_vars(params_sensitive, [0])
+
+    flops = paddle.flops(model, input_shape)
+    logger.info("FLOPs after pruning: {}".format(flops))
+
+    # start train
+
+    program.train(
+        config,
+        train_dataloader,
+        valid_dataloader,
+        device,
+        model,
+        loss_class,
+        optimizer,
+        lr_scheduler,
+        post_process_class,
+        eval_class,
+        pre_best_model_dict,
+        logger,
+        vdl_writer,
+    )
+
+
+if __name__ == "__main__":
+    config, device, logger, vdl_writer = program.preprocess(is_train=True)
+    main(config, device, logger, vdl_writer)
diff --git a/deploy/slim/quantization/README.md b/deploy/slim/quantization/README.md
new file mode 100644
index 0000000..d401d3b
--- /dev/null
+++ b/deploy/slim/quantization/README.md
@@ -0,0 +1,60 @@
+
+# PP-OCR模型量化
+复杂的模型有利于提高模型的性能，但也导致模型中存在一定冗余，模型量化将全精度缩减到定点数减少这种冗余，达到减少模型计算复杂度，提高模型推理性能的目的。
+模型量化可以在基本不损失模型的精度的情况下，将FP32精度的模型参数转换为Int8精度，减小模型参数大小并加速计算，使用量化后的模型在移动端等部署时更具备速度优势。
+
+本教程将介绍如何使用飞桨模型压缩库PaddleSlim做PaddleOCR模型的压缩。
+[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim) 集成了模型剪枝、量化（包括量化训练和离线量化）、蒸馏和神经网络搜索等多种业界常用且领先的模型压缩功能，如果您感兴趣，可以关注并了解。
+
+在开始本教程之前，建议先了解[PaddleOCR模型的训练方法](../../../doc/doc_ch/training.md)以及[PaddleSlim](https://paddleslim.readthedocs.io/zh_CN/latest/index.html)
+
+
+## 快速开始
+量化多适用于轻量模型在移动端的部署，当训练出一个模型后，如果希望进一步的压缩模型大小并加速预测，可使用量化的方法压缩模型。
+
+模型量化主要包括五个步骤：
+1. 安装 PaddleSlim
+2. 准备训练好的模型
+3. 量化训练
+4. 导出量化推理模型
+5. 量化模型预测部署
+
+### 1. 安装PaddleSlim
+
+```bash
+pip3 install paddleslim==2.3.2
+```
+
+### 2. 准备训练好的模型
+
+PaddleOCR提供了一系列训练好的[模型](../../../doc/doc_ch/models_list.md)，如果待量化的模型不在列表中，需要按照[常规训练](../../../doc/doc_ch/quickstart.md)方法得到训练好的模型。
+
+### 3. 量化训练
+量化训练包括离线量化训练和在线量化训练，在线量化训练效果更好，需加载预训练模型，在定义好量化策略后即可对模型进行量化。
+
+
+量化训练的代码位于slim/quantization/quant.py 中，比如训练检测模型，以PPOCRv3检测模型为例，训练指令如下：
+```
+# 下载检测预训练模型：
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
+tar xf ch_PP-OCRv3_det_distill_train.tar
+
+python deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Global.pretrained_model='./ch_PP-OCRv3_det_distill_train/best_accuracy'   Global.save_model_dir=./output/quant_model_distill/
+```
+如果要训练识别模型的量化，修改配置文件和加载的模型参数即可。
+
+### 4. 导出模型
+
+在得到量化训练保存的模型后，我们可以将其导出为inference_model，用于预测部署：
+
+```bash
+python deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Global.checkpoints=output/quant_model/best_accuracy Global.save_inference_dir=./output/quant_inference_model
+```
+
+### 5. 量化模型部署
+
+上述步骤导出的量化模型，参数精度仍然是FP32，但是参数的数值范围是int8，导出的模型可以通过PaddleLite的opt模型转换工具完成模型转换。
+
+量化模型移动端部署的可参考 [移动端模型部署](../../lite/readme.md)
+
+备注：量化训练后的模型参数是float32类型，转inference model预测时相对不量化无加速效果，原因是量化后模型结构之间存在量化和反量化算子，如果要使用量化模型部署，建议使用TensorRT并设置precision为INT8加速量化模型的预测时间。
diff --git a/deploy/slim/quantization/README_en.md b/deploy/slim/quantization/README_en.md
new file mode 100644
index 0000000..f82c3d8
--- /dev/null
+++ b/deploy/slim/quantization/README_en.md
@@ -0,0 +1,65 @@
+
+# PP-OCR Models Quantization
+
+Generally, a more complex model would achieve better performance in the task, but it also leads to some redundancy in the model.
+Quantization is a technique that reduces this redundancy by reducing the full precision data to a fixed number,
+so as to reduce model calculation complexity and improve model inference performance.
+
+This example uses PaddleSlim provided [APIs of Quantization](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/api_cn/dygraph/quanter/qat.rst) to compress the OCR model.
+
+It is recommended that you could understand following pages before reading this example：
+- [The training strategy of OCR model](../../../doc/doc_en/quickstart_en.md)
+- [PaddleSlim Document](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/api_cn/dygraph/quanter/qat.rst)
+
+## Quick Start
+Quantization is mostly suitable for the deployment of lightweight models on mobile terminals.
+After training, if you want to further compress the model size and accelerate the prediction, you can use quantization methods to compress the model according to the following steps.
+
+1. Install PaddleSlim
+2. Prepare trained model
+3. Quantization-Aware Training
+4. Export inference model
+5. Deploy quantization inference model
+
+
+### 1. Install PaddleSlim
+
+```bash
+pip3 install paddleslim==2.3.2
+```
+
+
+### 2. Download Pre-trained Model
+PaddleOCR provides a series of pre-trained [models](../../../doc/doc_en/models_list_en.md).
+If the model to be quantified is not in the list, you need to follow the [Regular Training](../../../doc/doc_en/quickstart_en.md) method to get the trained model.
+
+
+### 3. Quant-Aware Training
+Quantization training includes offline quantization training and online quantization training.
+Online quantization training is more effective. It is necessary to load the pre-trained model.
+After the quantization strategy is defined, the model can be quantified.
+
+The code for quantization training is located in `slim/quantization/quant.py`. For example, the training instructions of slim PPOCRv3 detection model are as follows:
+```
+# download provided model
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
+tar xf ch_PP-OCRv3_det_distill_train.tar
+
+python deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Global.pretrained_model='./ch_PP-OCRv3_det_distill_train/best_accuracy'   Global.save_model_dir=./output/quant_model_distill/
+```
+
+If you want to quantify the text recognition model, you can modify the configuration file and loaded model parameters.
+
+### 4. Export inference model
+
+Once we got the model after pruning and fine-tuning, we can export it as an inference model for the deployment of predictive tasks:
+
+```bash
+python deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Global.checkpoints=output/quant_model/best_accuracy Global.save_inference_dir=./output/quant_inference_model
+```
+
+### 5. Deploy
+The numerical range of the quantized model parameters derived from the above steps is still FP32, but the numerical range of the parameters is int8.
+The derived model can be converted through the `opt tool` of PaddleLite.
+
+For quantitative model deployment, please refer to [Mobile terminal model deployment](../../lite/readme.md)
diff --git a/deploy/slim/quantization/export_model.py b/deploy/slim/quantization/export_model.py
new file mode 100755
index 0000000..11aa4a1
--- /dev/null
+++ b/deploy/slim/quantization/export_model.py
@@ -0,0 +1,189 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "..", "..", "..")))
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "..", "..", "..", "tools")))
+
+import argparse
+
+import paddle
+from paddle.jit import to_static
+
+from ppocr.modeling.architectures import build_model
+from ppocr.postprocess import build_post_process
+from ppocr.utils.save_load import load_model
+from ppocr.utils.logging import get_logger
+from tools.program import load_config, merge_config, ArgsParser
+from ppocr.metrics import build_metric
+import tools.program as program
+from paddleslim.dygraph.quant import QAT
+from ppocr.data import build_dataloader, set_signal_handlers
+from ppocr.utils.export_model import export_single_model
+
+
+def main():
+    ############################################################################################################
+    # 1. quantization configs
+    ############################################################################################################
+    quant_config = {
+        # weight preprocess type, default is None and no preprocessing is performed.
+        "weight_preprocess_type": None,
+        # activation preprocess type, default is None and no preprocessing is performed.
+        "activation_preprocess_type": None,
+        # weight quantize type, default is 'channel_wise_abs_max'
+        "weight_quantize_type": "channel_wise_abs_max",
+        # activation quantize type, default is 'moving_average_abs_max'
+        "activation_quantize_type": "moving_average_abs_max",
+        # weight quantize bit num, default is 8
+        "weight_bits": 8,
+        # activation quantize bit num, default is 8
+        "activation_bits": 8,
+        # data type after quantization, such as 'uint8', 'int8', etc. default is 'int8'
+        "dtype": "int8",
+        # window size for 'range_abs_max' quantization. default is 10000
+        "window_size": 10000,
+        # The decay coefficient of moving average, default is 0.9
+        "moving_rate": 0.9,
+        # for dygraph quantization, layers of type in quantizable_layer_type will be quantized
+        "quantizable_layer_type": ["Conv2D", "Linear"],
+    }
+    FLAGS = ArgsParser().parse_args()
+    config = load_config(FLAGS.config)
+    config = merge_config(config, FLAGS.opt)
+    logger = get_logger()
+    # build post process
+
+    post_process_class = build_post_process(config["PostProcess"], config["Global"])
+
+    # build model
+    if hasattr(post_process_class, "character"):
+        char_num = len(getattr(post_process_class, "character"))
+        if config["Architecture"]["algorithm"] in [
+            "Distillation",
+        ]:  # distillation model
+            for key in config["Architecture"]["Models"]:
+                if (
+                    config["Architecture"]["Models"][key]["Head"]["name"] == "MultiHead"
+                ):  # for multi head
+                    if config["PostProcess"]["name"] == "DistillationSARLabelDecode":
+                        char_num = char_num - 2
+                    # update SARLoss params
+                    assert (
+                        list(config["Loss"]["loss_config_list"][-1].keys())[0]
+                        == "DistillationSARLoss"
+                    )
+                    config["Loss"]["loss_config_list"][-1]["DistillationSARLoss"][
+                        "ignore_index"
+                    ] = (char_num + 1)
+                    out_channels_list = {}
+                    out_channels_list["CTCLabelDecode"] = char_num
+                    out_channels_list["SARLabelDecode"] = char_num + 2
+                    config["Architecture"]["Models"][key]["Head"][
+                        "out_channels_list"
+                    ] = out_channels_list
+                else:
+                    config["Architecture"]["Models"][key]["Head"][
+                        "out_channels"
+                    ] = char_num
+        elif config["Architecture"]["Head"]["name"] == "MultiHead":  # for multi head
+            if config["PostProcess"]["name"] == "SARLabelDecode":
+                char_num = char_num - 2
+            # update SARLoss params
+            assert list(config["Loss"]["loss_config_list"][1].keys())[0] == "SARLoss"
+            if config["Loss"]["loss_config_list"][1]["SARLoss"] is None:
+                config["Loss"]["loss_config_list"][1]["SARLoss"] = {
+                    "ignore_index": char_num + 1
+                }
+            else:
+                config["Loss"]["loss_config_list"][1]["SARLoss"]["ignore_index"] = (
+                    char_num + 1
+                )
+            out_channels_list = {}
+            out_channels_list["CTCLabelDecode"] = char_num
+            out_channels_list["SARLabelDecode"] = char_num + 2
+            config["Architecture"]["Head"]["out_channels_list"] = out_channels_list
+        else:  # base rec model
+            config["Architecture"]["Head"]["out_channels"] = char_num
+
+        if config["PostProcess"]["name"] == "SARLabelDecode":  # for SAR model
+            config["Loss"]["ignore_index"] = char_num - 1
+
+    model = build_model(config["Architecture"])
+
+    # get QAT model
+    quanter = QAT(config=quant_config)
+    quanter.quantize(model)
+
+    load_model(config, model)
+
+    # build metric
+    eval_class = build_metric(config["Metric"])
+
+    # build dataloader
+    set_signal_handlers()
+    valid_dataloader = build_dataloader(config, "Eval", device, logger)
+
+    use_srn = config["Architecture"]["algorithm"] == "SRN"
+    model_type = config["Architecture"].get("model_type", None)
+    # start eval
+    metric = program.eval(
+        model, valid_dataloader, post_process_class, eval_class, model_type, use_srn
+    )
+    model.eval()
+
+    logger.info("metric eval ***************")
+    for k, v in metric.items():
+        logger.info("{}:{}".format(k, v))
+
+    save_path = config["Global"]["save_inference_dir"]
+
+    arch_config = config["Architecture"]
+
+    if (
+        arch_config["algorithm"] == "SVTR"
+        and arch_config["Head"]["name"] != "MultiHead"
+    ):
+        input_shape = config["Eval"]["dataset"]["transforms"][-2]["SVTRRecResizeImg"][
+            "image_shape"
+        ]
+    else:
+        input_shape = None
+
+    if arch_config["algorithm"] in [
+        "Distillation",
+    ]:  # distillation model
+        archs = list(arch_config["Models"].values())
+        for idx, name in enumerate(model.model_name_list):
+            sub_model_save_path = os.path.join(save_path, name, "inference")
+            export_single_model(
+                model.model_list[idx],
+                archs[idx],
+                sub_model_save_path,
+                logger,
+                input_shape,
+                quanter,
+            )
+    else:
+        save_path = os.path.join(save_path, "inference")
+        export_single_model(model, arch_config, save_path, logger, input_shape, quanter)
+
+
+if __name__ == "__main__":
+    config, device, logger, vdl_writer = program.preprocess()
+    main()
diff --git a/deploy/slim/quantization/quant.py b/deploy/slim/quantization/quant.py
new file mode 100755
index 0000000..d112a40
--- /dev/null
+++ b/deploy/slim/quantization/quant.py
@@ -0,0 +1,226 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.append(os.path.abspath(os.path.join(__dir__, "..", "..", "..")))
+sys.path.append(os.path.abspath(os.path.join(__dir__, "..", "..", "..", "tools")))
+
+import yaml
+import paddle
+import paddle.distributed as dist
+
+paddle.seed(2)
+
+from ppocr.data import build_dataloader, set_signal_handlers
+from ppocr.modeling.architectures import build_model
+from ppocr.losses import build_loss
+from ppocr.optimizer import build_optimizer
+from ppocr.postprocess import build_post_process
+from ppocr.metrics import build_metric
+from ppocr.utils.save_load import load_model
+import tools.program as program
+from paddleslim.dygraph.quant import QAT
+
+dist.get_world_size()
+
+
+class PACT(paddle.nn.Layer):
+    def __init__(self):
+        super(PACT, self).__init__()
+        alpha_attr = paddle.ParamAttr(
+            name=self.full_name() + ".pact",
+            initializer=paddle.nn.initializer.Constant(value=20),
+            learning_rate=1.0,
+            regularizer=paddle.regularizer.L2Decay(2e-5),
+        )
+
+        self.alpha = self.create_parameter(shape=[1], attr=alpha_attr, dtype="float32")
+
+    def forward(self, x):
+        out_left = paddle.nn.functional.relu(x - self.alpha)
+        out_right = paddle.nn.functional.relu(-self.alpha - x)
+        x = x - out_left + out_right
+        return x
+
+
+quant_config = {
+    # weight preprocess type, default is None and no preprocessing is performed.
+    "weight_preprocess_type": None,
+    # activation preprocess type, default is None and no preprocessing is performed.
+    "activation_preprocess_type": None,
+    # weight quantize type, default is 'channel_wise_abs_max'
+    "weight_quantize_type": "channel_wise_abs_max",
+    # activation quantize type, default is 'moving_average_abs_max'
+    "activation_quantize_type": "moving_average_abs_max",
+    # weight quantize bit num, default is 8
+    "weight_bits": 8,
+    # activation quantize bit num, default is 8
+    "activation_bits": 8,
+    # data type after quantization, such as 'uint8', 'int8', etc. default is 'int8'
+    "dtype": "int8",
+    # window size for 'range_abs_max' quantization. default is 10000
+    "window_size": 10000,
+    # The decay coefficient of moving average, default is 0.9
+    "moving_rate": 0.9,
+    # for dygraph quantization, layers of type in quantizable_layer_type will be quantized
+    "quantizable_layer_type": ["Conv2D", "Linear"],
+}
+
+
+def main(config, device, logger, vdl_writer):
+    # init dist environment
+    if config["Global"]["distributed"]:
+        dist.init_parallel_env()
+
+    global_config = config["Global"]
+
+    # build dataloader
+    set_signal_handlers()
+    train_dataloader = build_dataloader(config, "Train", device, logger)
+    if config["Eval"]:
+        valid_dataloader = build_dataloader(config, "Eval", device, logger)
+    else:
+        valid_dataloader = None
+
+    # build post process
+    post_process_class = build_post_process(config["PostProcess"], global_config)
+
+    # build model
+    # for rec algorithm
+    if hasattr(post_process_class, "character"):
+        char_num = len(getattr(post_process_class, "character"))
+        if config["Architecture"]["algorithm"] in [
+            "Distillation",
+        ]:  # distillation model
+            for key in config["Architecture"]["Models"]:
+                if (
+                    config["Architecture"]["Models"][key]["Head"]["name"] == "MultiHead"
+                ):  # for multi head
+                    if config["PostProcess"]["name"] == "DistillationSARLabelDecode":
+                        char_num = char_num - 2
+                    # update SARLoss params
+                    assert (
+                        list(config["Loss"]["loss_config_list"][-1].keys())[0]
+                        == "DistillationSARLoss"
+                    )
+                    config["Loss"]["loss_config_list"][-1]["DistillationSARLoss"][
+                        "ignore_index"
+                    ] = (char_num + 1)
+                    out_channels_list = {}
+                    out_channels_list["CTCLabelDecode"] = char_num
+                    out_channels_list["SARLabelDecode"] = char_num + 2
+                    config["Architecture"]["Models"][key]["Head"][
+                        "out_channels_list"
+                    ] = out_channels_list
+                else:
+                    config["Architecture"]["Models"][key]["Head"][
+                        "out_channels"
+                    ] = char_num
+        elif config["Architecture"]["Head"]["name"] == "MultiHead":  # for multi head
+            if config["PostProcess"]["name"] == "SARLabelDecode":
+                char_num = char_num - 2
+            # update SARLoss params
+            assert list(config["Loss"]["loss_config_list"][1].keys())[0] == "SARLoss"
+            if config["Loss"]["loss_config_list"][1]["SARLoss"] is None:
+                config["Loss"]["loss_config_list"][1]["SARLoss"] = {
+                    "ignore_index": char_num + 1
+                }
+            else:
+                config["Loss"]["loss_config_list"][1]["SARLoss"]["ignore_index"] = (
+                    char_num + 1
+                )
+            out_channels_list = {}
+            out_channels_list["CTCLabelDecode"] = char_num
+            out_channels_list["SARLabelDecode"] = char_num + 2
+            config["Architecture"]["Head"]["out_channels_list"] = out_channels_list
+        else:  # base rec model
+            config["Architecture"]["Head"]["out_channels"] = char_num
+
+        if config["PostProcess"]["name"] == "SARLabelDecode":  # for SAR model
+            config["Loss"]["ignore_index"] = char_num - 1
+    model = build_model(config["Architecture"])
+
+    pre_best_model_dict = dict()
+    # load fp32 model to begin quantization
+    pre_best_model_dict = load_model(
+        config, model, None, config["Architecture"]["model_type"]
+    )
+
+    freeze_params = False
+    if config["Architecture"]["algorithm"] in ["Distillation"]:
+        for key in config["Architecture"]["Models"]:
+            freeze_params = freeze_params or config["Architecture"]["Models"][key].get(
+                "freeze_params", False
+            )
+    act = None if freeze_params else PACT
+    quanter = QAT(config=quant_config, act_preprocess=act)
+    quanter.quantize(model)
+
+    if config["Global"]["distributed"]:
+        model = paddle.DataParallel(model)
+
+    # build loss
+    loss_class = build_loss(config["Loss"])
+
+    # build optim
+    optimizer, lr_scheduler, _wd_scheduler = build_optimizer(
+        config["Optimizer"],
+        epochs=config["Global"]["epoch_num"],
+        step_each_epoch=len(train_dataloader),
+        model=model,
+    )
+
+    # resume PACT training process
+    pre_best_model_dict = load_model(
+        config, model, optimizer, config["Architecture"]["model_type"]
+    )
+
+    # build metric
+    eval_class = build_metric(config["Metric"])
+
+    logger.info(
+        "train dataloader has {} iters, valid dataloader has {} iters".format(
+            len(train_dataloader), len(valid_dataloader)
+        )
+    )
+
+    # start train
+    program.train(
+        config,
+        train_dataloader,
+        valid_dataloader,
+        device,
+        model,
+        loss_class,
+        optimizer,
+        lr_scheduler,
+        post_process_class,
+        eval_class,
+        pre_best_model_dict,
+        logger,
+        vdl_writer,
+    )
+
+
+if __name__ == "__main__":
+    config, device, logger, vdl_writer = program.preprocess(is_train=True)
+    main(config, device, logger, vdl_writer)
diff --git a/deploy/slim/quantization/quant_kl.py b/deploy/slim/quantization/quant_kl.py
new file mode 100755
index 0000000..deaa9e9
--- /dev/null
+++ b/deploy/slim/quantization/quant_kl.py
@@ -0,0 +1,172 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.append(os.path.abspath(os.path.join(__dir__, "..", "..", "..")))
+sys.path.append(os.path.abspath(os.path.join(__dir__, "..", "..", "..", "tools")))
+
+import yaml
+import paddle
+import paddle.distributed as dist
+
+paddle.seed(2)
+
+from ppocr.data import build_dataloader, set_signal_handlers
+from ppocr.modeling.architectures import build_model
+from ppocr.losses import build_loss
+from ppocr.optimizer import build_optimizer
+from ppocr.postprocess import build_post_process
+from ppocr.metrics import build_metric
+from ppocr.utils.save_load import load_model
+import tools.program as program
+import paddleslim
+from paddleslim.dygraph.quant import QAT
+import numpy as np
+
+dist.get_world_size()
+
+
+class PACT(paddle.nn.Layer):
+    def __init__(self):
+        super(PACT, self).__init__()
+        alpha_attr = paddle.ParamAttr(
+            name=self.full_name() + ".pact",
+            initializer=paddle.nn.initializer.Constant(value=20),
+            learning_rate=1.0,
+            regularizer=paddle.regularizer.L2Decay(2e-5),
+        )
+
+        self.alpha = self.create_parameter(shape=[1], attr=alpha_attr, dtype="float32")
+
+    def forward(self, x):
+        out_left = paddle.nn.functional.relu(x - self.alpha)
+        out_right = paddle.nn.functional.relu(-self.alpha - x)
+        x = x - out_left + out_right
+        return x
+
+
+quant_config = {
+    # weight preprocess type, default is None and no preprocessing is performed.
+    "weight_preprocess_type": None,
+    # activation preprocess type, default is None and no preprocessing is performed.
+    "activation_preprocess_type": None,
+    # weight quantize type, default is 'channel_wise_abs_max'
+    "weight_quantize_type": "channel_wise_abs_max",
+    # activation quantize type, default is 'moving_average_abs_max'
+    "activation_quantize_type": "moving_average_abs_max",
+    # weight quantize bit num, default is 8
+    "weight_bits": 8,
+    # activation quantize bit num, default is 8
+    "activation_bits": 8,
+    # data type after quantization, such as 'uint8', 'int8', etc. default is 'int8'
+    "dtype": "int8",
+    # window size for 'range_abs_max' quantization. default is 10000
+    "window_size": 10000,
+    # The decay coefficient of moving average, default is 0.9
+    "moving_rate": 0.9,
+    # for dygraph quantization, layers of type in quantizable_layer_type will be quantized
+    "quantizable_layer_type": ["Conv2D", "Linear"],
+}
+
+
+def sample_generator(loader):
+    def __reader__():
+        for _, data in enumerate(loader):
+            images = np.array(data[0])
+            yield images
+
+    return __reader__
+
+
+def sample_generator_layoutxlm_ser(loader):
+    def __reader__():
+        for _, data in enumerate(loader):
+            input_ids = np.array(data[0])
+            bbox = np.array(data[1])
+            attention_mask = np.array(data[2])
+            token_type_ids = np.array(data[3])
+            images = np.array(data[4])
+            yield [input_ids, bbox, attention_mask, token_type_ids, images]
+
+    return __reader__
+
+
+def main(config, device, logger, vdl_writer):
+    # init dist environment
+    if config["Global"]["distributed"]:
+        dist.init_parallel_env()
+
+    global_config = config["Global"]
+
+    # build dataloader
+    set_signal_handlers()
+    config["Train"]["loader"]["num_workers"] = 0
+    is_layoutxlm_ser = (
+        config["Architecture"]["model_type"] == "kie"
+        and config["Architecture"]["Backbone"]["name"] == "LayoutXLMForSer"
+    )
+    train_dataloader = build_dataloader(config, "Train", device, logger)
+    if config["Eval"]:
+        config["Eval"]["loader"]["num_workers"] = 0
+        valid_dataloader = build_dataloader(config, "Eval", device, logger)
+        if is_layoutxlm_ser:
+            train_dataloader = valid_dataloader
+    else:
+        valid_dataloader = None
+
+    paddle.enable_static()
+    exe = paddle.static.Executor(device)
+
+    if "inference_model" in global_config.keys():  # , 'inference_model'):
+        inference_model_dir = global_config["inference_model"]
+    else:
+        inference_model_dir = os.path.dirname(global_config["pretrained_model"])
+        if not (
+            os.path.exists(os.path.join(inference_model_dir, "inference.pdmodel"))
+            and os.path.exists(os.path.join(inference_model_dir, "inference.pdiparams"))
+        ):
+            raise ValueError(
+                "Please set inference model dir in Global.inference_model or Global.pretrained_model for post-quantization"
+            )
+
+    if is_layoutxlm_ser:
+        generator = sample_generator_layoutxlm_ser(train_dataloader)
+    else:
+        generator = sample_generator(train_dataloader)
+
+    paddleslim.quant.quant_post_static(
+        executor=exe,
+        model_dir=inference_model_dir,
+        model_filename="inference.pdmodel",
+        params_filename="inference.pdiparams",
+        quantize_model_path=global_config["save_inference_dir"],
+        sample_generator=generator,
+        save_model_filename="inference.pdmodel",
+        save_params_filename="inference.pdiparams",
+        batch_size=1,
+        batch_nums=None,
+    )
+
+
+if __name__ == "__main__":
+    config, device, logger, vdl_writer = program.preprocess(is_train=True)
+    main(config, device, logger, vdl_writer)
diff --git a/doc/fonts/arabic.ttf b/doc/fonts/arabic.ttf
new file mode 100644
index 0000000..064b604
Binary files /dev/null and b/doc/fonts/arabic.ttf differ
diff --git a/doc/fonts/chinese_cht.ttf b/doc/fonts/chinese_cht.ttf
new file mode 100644
index 0000000..3416754
Binary files /dev/null and b/doc/fonts/chinese_cht.ttf differ
diff --git a/doc/fonts/cyrillic.ttf b/doc/fonts/cyrillic.ttf
new file mode 100644
index 0000000..be4bf66
Binary files /dev/null and b/doc/fonts/cyrillic.ttf differ
diff --git a/doc/fonts/french.ttf b/doc/fonts/french.ttf
new file mode 100644
index 0000000..ab68fb1
Binary files /dev/null and b/doc/fonts/french.ttf differ
diff --git a/doc/fonts/german.ttf b/doc/fonts/german.ttf
new file mode 100644
index 0000000..ab68fb1
Binary files /dev/null and b/doc/fonts/german.ttf differ
diff --git a/doc/fonts/hindi.ttf b/doc/fonts/hindi.ttf
new file mode 100644
index 0000000..8b0c36f
Binary files /dev/null and b/doc/fonts/hindi.ttf differ
diff --git a/doc/fonts/japan.ttc b/doc/fonts/japan.ttc
new file mode 100644
index 0000000..ad68243
Binary files /dev/null and b/doc/fonts/japan.ttc differ
diff --git a/doc/fonts/kannada.ttf b/doc/fonts/kannada.ttf
new file mode 100644
index 0000000..43b60d4
Binary files /dev/null and b/doc/fonts/kannada.ttf differ
diff --git a/doc/fonts/korean.ttf b/doc/fonts/korean.ttf
new file mode 100644
index 0000000..e638ce3
Binary files /dev/null and b/doc/fonts/korean.ttf differ
diff --git a/doc/fonts/latin.ttf b/doc/fonts/latin.ttf
new file mode 100644
index 0000000..e392413
Binary files /dev/null and b/doc/fonts/latin.ttf differ
diff --git a/doc/fonts/marathi.ttf b/doc/fonts/marathi.ttf
new file mode 100644
index 0000000..a796d3e
Binary files /dev/null and b/doc/fonts/marathi.ttf differ
diff --git a/doc/fonts/nepali.ttf b/doc/fonts/nepali.ttf
new file mode 100644
index 0000000..8b0c36f
Binary files /dev/null and b/doc/fonts/nepali.ttf differ
diff --git a/doc/fonts/persian.ttf b/doc/fonts/persian.ttf
new file mode 100644
index 0000000..bdb1c8d
Binary files /dev/null and b/doc/fonts/persian.ttf differ
diff --git a/doc/fonts/simfang.ttf b/doc/fonts/simfang.ttf
new file mode 100644
index 0000000..2b59eae
Binary files /dev/null and b/doc/fonts/simfang.ttf differ
diff --git a/doc/fonts/spanish.ttf b/doc/fonts/spanish.ttf
new file mode 100644
index 0000000..532353d
Binary files /dev/null and b/doc/fonts/spanish.ttf differ
diff --git a/doc/fonts/tamil.ttf b/doc/fonts/tamil.ttf
new file mode 100644
index 0000000..2e9998e
Binary files /dev/null and b/doc/fonts/tamil.ttf differ
diff --git a/doc/fonts/telugu.ttf b/doc/fonts/telugu.ttf
new file mode 100644
index 0000000..12c91e4
Binary files /dev/null and b/doc/fonts/telugu.ttf differ
diff --git a/doc/fonts/urdu.ttf b/doc/fonts/urdu.ttf
new file mode 100644
index 0000000..625feee
Binary files /dev/null and b/doc/fonts/urdu.ttf differ
diff --git a/doc/fonts/uyghur.ttf b/doc/fonts/uyghur.ttf
new file mode 100644
index 0000000..625feee
Binary files /dev/null and b/doc/fonts/uyghur.ttf differ
diff --git a/docs/FAQ.en.md b/docs/FAQ.en.md
new file mode 100644
index 0000000..965488f
--- /dev/null
+++ b/docs/FAQ.en.md
@@ -0,0 +1,197 @@
+---
+comments: true
+hide:
+  - navigation
+  - toc
+---
+
+> 🎉 **Welcome to PaddleOCR FAQ!**  
+> This document compiles common issues and solutions from GitHub Issues and Discussions, providing reference for OCR developers.
+
+## 1. Installation and Environment Setup
+
+### 1.1 Basic Installation Issues
+
+#### Q: PaddleOCR installation failed with dependency conflicts
+
+**A**: This is a common issue that can be resolved by:
+(1) Create a new virtual environment: `conda create -n paddleocr python=3.8`, then activate and install
+(2) Install with specific versions: `pip install paddleocr==3.2.0 --no-deps`, then install dependencies separately
+(3) If using conda, try: `conda install -c conda-forge paddleocr`
+
+#### Q: GPU environment configuration issues, CUDA version mismatch
+
+**A**: First check CUDA version: `nvidia-smi`, then install corresponding PaddlePaddle version:
+- CUDA 11.8: `pip install paddlepaddle-gpu==3.0.0 -f https://www.paddlepaddle.org.cn/whl/linux/mkl/avx/stable.html`
+- CUDA 12.0: `pip install paddlepaddle-gpu==3.0.0 -f https://www.paddlepaddle.org.cn/whl/linux/mkl/avx/stable.html`
+Verify GPU availability: `python -c "import paddle; print(paddle.is_compiled_with_cuda())"`
+
+#### Q: Model download failed or slow download
+
+**A**: Can be resolved by:
+(1) Set model download source: `os.environ['PADDLE_PDX_MODEL_SOURCE'] = 'BOS'` (use Baidu Cloud Storage)
+(2) Manual model download: directly access model links to download and extract to local directory
+(3) Use local models: specify `model_dir` parameter pointing to local model path during initialization
+
+#### Q: How to run on Windows or Mac?
+
+**A**: PaddleOCR has completed adaptation to Windows and MAC systems. Two points should be noted during operation:
+    1. In [Quick installation](./version3.x/installation.en.md), if you do not want to install docker, you can skip the first step and start with the second step.
+    2. When downloading the inference model, if wget is not installed, you can directly click the model link or copy the link address to the browser to download, then extract and place it in the corresponding directory.
+
+## 2. Model Usage and Configuration
+
+### 2.1 Model Selection
+
+#### Q: How to choose the right model?
+
+**A**: Choose based on application scenario:
+- Server high accuracy: Use `PP-OCRv5_server` series, highest accuracy
+- Mobile deployment: Use `PP-OCRv5_mobile` series, small model fast speed
+- Real-time processing: Use `PP-OCRv5_mobile` series, fast inference speed
+- Batch processing: Use `PP-OCRv5_server` series, high accuracy
+- Multi-language recognition: Use `PP-OCRv5_multi_languages`, supports 37 languages
+
+#### Q: Local model path configuration, how to use in isolated network environment?
+
+**A**: Can be configured by:
+(1) Use local model path: specify `model_dir` parameter during initialization
+(2) Set model download source: `os.environ['PADDLE_PDX_MODEL_SOURCE'] = 'BOS'`
+(3) Manual model download: download models from official links and extract locally
+(4) Example code:
+```python
+ocr = PaddleOCR(
+    det_model_dir='./models/PP-OCRv5_server_det_infer/',
+    rec_model_dir='./models/PP-OCRv5_server_rec_infer/',
+    cls_model_dir='./models/PP-OCRv5_cls_infer/',
+    use_angle_cls=True,
+    lang='ch'
+)
+```
+
+## 3. Performance Optimization
+
+### 3.1 GPU Optimization
+
+#### Q: Slow GPU inference speed, how to optimize performance?
+
+**A**: Can be optimized by:
+(1) Enable high-performance inference: set `enable_hpi=True`, automatically select optimal acceleration strategy
+(2) Enable TensorRT acceleration: set `use_tensorrt=True`, requires CUDA 11.8+ and TensorRT 8.6+
+(3) Use half precision: set `precision="fp16"`, can significantly improve speed
+(4) Adjust batch size: set appropriate `batch_size` based on GPU memory
+(5) Use mobile models: use `PP-OCRv5_mobile` series when accuracy requirements are not high
+
+#### Q: GPU memory insufficient (CUDA out of memory) what to do?
+
+**A**: Can be resolved by:
+(1) Reduce batch size: set `batch_size` to 1
+(2) Reduce image size: set `det_limit_side_len=640`
+(3) Enable memory optimization: set `enable_memory_optim=True`
+(4) Limit GPU memory usage: set `gpu_mem=200`
+(5) Use mobile models: switch to `PP-OCRv5_mobile` series models
+
+## 4. Deployment Issues
+
+### 4.1 Service Deployment
+
+#### Q: How to deploy PaddleOCR as a web service?
+
+**A**: Can be deployed by:
+(1) Use Flask deployment: create simple Web API service
+(2) Use gunicorn deployment: `gunicorn -w 4 -b 0.0.0.0:5000 app:app`
+(3) Use asynchronous processing: combine asyncio and ThreadPoolExecutor
+(4) Example code:
+```python
+from flask import Flask, request, jsonify
+from paddleocr import PaddleOCR
+
+app = Flask(__name__)
+ocr = PaddleOCR(use_angle_cls=True, lang='ch')
+
+@app.route('/ocr', methods=['POST'])
+def ocr_api():
+    file = request.files['image']
+    result = ocr.ocr(file, cls=True)
+    return jsonify({'result': result})
+```
+
+#### Q: C++ deployment common issues and solutions?
+
+**A**: Common issues and solutions:
+(1) Cannot find dynamic library: set `export LD_LIBRARY_PATH=/path/to/paddle/lib:$LD_LIBRARY_PATH`
+(2) OpenCV version mismatch: ensure OpenCV version matches compilation time
+(3) Model format issues: ensure correct inference model format
+(4) Compilation issues: ensure CMake configuration is correct, use `cmake .. -DCMAKE_BUILD_TYPE=Release`
+
+#### Q: Service performance optimization suggestions?
+
+**A**: Can be optimized by:
+(1) Enable high-performance inference: set `enable_hpi=True`
+(2) Use batch processing: set appropriate `batch_size`
+(3) Enable TensorRT: set `use_tensorrt=True`
+(4) Use asynchronous processing: avoid blocking requests
+(5) Load balancing: use multiple service instances
+
+## 5. Legacy Issues (for reference)
+
+1. **Prediction error: got an unexpected keyword argument 'gradient_clip'**
+The installed version of paddle is incorrect. Currently, this project only supports Paddle 1.7, which will be adapted to 1.8 in the near future.
+
+2. **Error when converting attention recognition model: KeyError: 'predict'**
+Solved. Please update to the latest version of the code.
+
+3. **About inference speed**
+When there are many words in the picture, the prediction time will increase. You can use `--rec_batch_num` to set a smaller prediction batch num. The default value is 30, which can be changed to 10 or other values.
+
+4. **Service deployment and mobile deployment**
+It is expected that the service deployment based on Serving and the mobile deployment based on Paddle Lite will be released successively in mid-to-late June. Stay tuned for more updates.
+
+5. **Release time of self-developed algorithm**
+Baidu Self-developed algorithms such as SAST, SRN and end2end PSL will be released in June or July. Please be patient.
+
+6. **How to run on Windows or Mac?**
+PaddleOCR has completed the adaptation to Windows and MAC systems. Two points should be noted during operation:
+    1. In [Quick installation](./version3.x/installation.en.md), if you do not want to install docker, you can skip the first step and start with the second step.
+    2. When downloading the inference model, if wget is not installed, you can directly click the model link or copy the link address to the browser to download, then extract and place it in the corresponding directory.
+
+7. **The difference between ultra-lightweight model and General OCR model**
+At present, PaddleOCR has opensourced two Chinese models, namely 8.6M ultra-lightweight Chinese model and general Chinese OCR model. The comparison information between the two is as follows:
+    - Similarities: Both use the same **algorithm** and **training data**；
+    - Differences: The difference lies in **backbone network** and **channel parameters**, the ultra-lightweight model uses MobileNetV3 as the backbone network, the general model uses Resnet50_vd as the detection model backbone, and Resnet34_vd as the recognition model backbone. You can compare the two model training configuration files to see the differences in parameters.
+
+|Model|Backbone|Detection configuration file|Recognition configuration file|
+|-|-|-|-|
+|8.6M ultra-lightweight Chinese OCR model|MobileNetV3+MobileNetV3|det_mv3_db.yml|rec_chinese_lite_train.yml|
+|General Chinese OCR model|Resnet50_vd+Resnet34_vd|det_r50_vd_db.yml|rec_chinese_common_train.yml|
+
+8. **Is there a plan to opensource a model that only recognizes numbers or only English + numbers?**
+It is not planned to opensource numbers only, numbers + English only, or other vertical text models. PaddleOCR has opensourced a variety of detection and recognition algorithms for customized training. The two Chinese models are also based on the training output of the open-source algorithm library. You can prepare the data according to the tutorial, choose the appropriate configuration file, train yourselves, and we believe that you can get good result. If you have any questions during the training, you are welcome to open issues or ask in the communication group. We will answer them in time.
+
+9. **What is the training data used by the open-source model? Can it be opensourced?**
+At present, the open source model, dataset and magnitude are as follows:
+    - Detection:
+    English dataset: ICDAR2015
+    Chinese dataset: LSVT street view dataset with 3w pictures
+    - Recognition:
+    English dataset: MJSynth and SynthText synthetic dataset, the amount of data is tens of millions.
+    Chinese dataset: LSVT street view dataset with cropped text area, a total of 30w images. In addition, the synthesized data based on LSVT corpus is 500w.
+
+    Among them, the public datasets are opensourced, users can search and download by themselves, or refer to [Chinese data set](./datasets/datasets.en.md), synthetic data is not opensourced, users can use open-source synthesis tools to synthesize data themselves. Current available synthesis tools include [text_renderer](https://github.com/Sanster/text_renderer), [SynthText](https://github.com/ankush-me/SynthText), [TextRecognitionDataGenerator](https://github.com/Belval/TextRecognitionDataGenerator), etc.
+
+10. **Error in using the model with TPS module for prediction**
+Error message: Input(X) dims[3] and Input(Grid) dims[2] should be equal, but received X dimension[3]\(108) != Grid dimension[2]\(100)
+Solution: TPS does not support variable shape. Please set --rec_image_shape='3,32,100' and --rec_char_type='en'
+
+11. **Custom dictionary used during training, the recognition results show that words do not appear in the dictionary**
+The used custom dictionary path is not set when making prediction. The solution is setting parameter `rec_char_dict_path` to the corresponding dictionary file.
+
+12. **Results of cpp_infer and python_inference are very different**
+Versions of exported inference model and inference library should be same. For example, on Windows platform, version of the inference library that PaddlePaddle provides is 1.8, but version of the inference model that PaddleOCR provides is 1.7, you should export model yourself(`tools/export_model.py`) on PaddlePaddle 1.8 and then use the exported model for inference.
+
+13. **How to identify artistic fonts in signs or advertising images**
+Recognizing artistic fonts in signs or advertising images is a very challenging task because the variation in individual characters is much greater compared to standard fonts. If the artistic font to be identified is within a dictionary list, each word in the dictionary can be treated as a template for recognition using a general image retrieval system. You can try using PaddleClas image recognition system.
+
+14. **How to change the font when visualizing the OCR prediction results?**
+
+**A**: You can specify the local font file path by using the environment variable `PADDLE_PDX_LOCAL_FONT_FILE_PATH`, such as `PADDLE_PDX_LOCAL_FONT_FILE_PATH=/root/fonts/simfang.ttf`.
diff --git a/docs/FAQ.md b/docs/FAQ.md
new file mode 100644
index 0000000..1c4850d
--- /dev/null
+++ b/docs/FAQ.md
@@ -0,0 +1,885 @@
+---
+comments: true
+hide:
+  - navigation
+  - toc
+---
+
+> 恭喜你发现宝藏！
+
+PaddleOCR收集整理了自从开源以来在issues和用户群中的常见问题并且给出了简要解答，旨在为OCR的开发者提供一些参考，也希望帮助大家少走一些弯路。
+
+其中[通用问题](#1)一般是初次接触OCR相关算法时用户会提出的问题，在[1.5 垂类场景实现思路](#15)中总结了如何在一些具体的场景中确定技术路线进行优化。[PaddleOCR常见问题](#2-paddleocr)是开发者在使用PaddleOCR之后可能会遇到的问题也是PaddleOCR实践过程中的避坑指南。
+
+同时PaddleOCR也会在review issue的过程中添加 `good issue`、 `good first issue` 标签，但这些问题可能不会被立刻补充在FAQ文档里，开发者也可对应查看。我们也非常希望开发者能够帮助我们将这些内容补充在FAQ中。
+
+OCR领域大佬众多，本文档回答主要依赖有限的项目实践，难免挂一漏万，如有遗漏和不足，也**希望有识之士帮忙补充和修正**，万分感谢。
+
+## 1. 通用问题
+
+### 1.1 检测
+
+#### Q: 基于深度学习的文字检测方法有哪几种？各有什么优缺点？
+
+**A**：常用的基于深度学习的文字检测方法一般可以分为基于回归的、基于分割的两大类，当然还有一些将两者进行结合的方法。
+
+（1）基于回归的方法分为box回归和像素值回归。a. 采用box回归的方法主要有CTPN、Textbox系列和EAST，这类算法对规则形状文本检测效果较好，但无法准确检测不规则形状文本。 b. 像素值回归的方法主要有CRAFT和SA-Text，这类算法能够检测弯曲文本且对小文本效果优秀但是实时性能不够。
+
+（2）基于分割的算法，如PSENet，这类算法不受文本形状的限制，对各种形状的文本都能取得较好的效果，但是往往后处理比较复杂，导致耗时严重。目前也有一些算法专门针对这个问题进行改进，如DB，将二值化进行近似，使其可导，融入训练，从而获取更准确的边界，大大降低了后处理的耗时。
+
+### 1.2 识别
+
+#### Q: PaddleOCR提供的文本识别算法包括哪些？
+
+**A**：PaddleOCR主要提供五种文本识别算法，包括CRNN\StarNet\RARE\Rosetta和SRN, 其中CRNN\StarNet和Rosetta是基于ctc的文字识别算法，RARE是基于attention的文字识别算法；SRN为百度自研的文本识别算法，引入了语义信息，显著提升了准确率。 详情可参照如下页面: 文本识别算法
+
+#### Q: 文本识别方法CRNN关键技术有哪些？
+
+**A**：CRNN 关键技术包括三部分。（1）CNN提取图像卷积特征。（2）深层双向LSTM网络，在卷积特征的基础上继续提取文字序列特征。（3）Connectionist Temporal Classification(CTC)，解决训练时字符无法对齐的问题。
+
+#### Q: 对于中文行文本识别，CTC和Attention哪种更优？
+
+**A**：（1）从效果上来看，通用OCR场景CTC的识别效果优于Attention，因为带识别的字典中的字符比较多，常用中文汉字三千字以上，如果训练样本不足的情况下，对于这些字符的序列关系挖掘比较困难。中文场景下Attention模型的优势无法体现。而且Attention适合短语句识别，对长句子识别比较差。
+
+（2）从训练和预测速度上，Attention的串行解码结构限制了预测速度，而CTC网络结构更高效，预测速度上更有优势。
+
+#### Q: 弯曲形变的文字识别需要怎么处理？TPS应用场景是什么，是否好用？
+
+**A**：（1）在大多数情况下，如果遇到的场景弯曲形变不是太严重，检测4个顶点，然后直接通过仿射变换转正识别就足够了。
+
+（2）如果不能满足需求，可以尝试使用TPS（Thin Plate Spline），即薄板样条插值。TPS是一种插值算法，经常用于图像变形等，通过少量的控制点就可以驱动图像进行变化。一般用在有弯曲形变的文本识别中，当检测到不规则的/弯曲的（如，使用基于分割的方法检测算法）文本区域，往往先使用TPS算法对文本区域矫正成矩形再进行识别，如，STAR-Net、RARE等识别算法中引入了TPS模块。
+
+> **Warning**：TPS看起来美好，在实际应用时经常发现并不够鲁棒，并且会增加耗时，需要谨慎使用。
+
+### 1.3 端到端
+
+#### Q: 请问端到端的pgnet相比于DB+CRNN在准确率上有优势吗？或者是pgnet最擅长的场景是什么场景呢？
+
+**A**：pgnet是端到端算法，检测识别一步到位，不用分开训练2个模型，也支持弯曲文本的识别，但是在中文上的效果还没有充分验证；db+crnn的验证更充分，应用相对成熟，常规非弯曲的文本都能解的不错。
+
+#### Q: 目前OCR普遍是二阶段，端到端的方案在业界落地情况如何？
+
+**A**：端到端在文字分布密集的业务场景，效率会比较有保证，精度的话看自己业务数据积累情况，如果行级别的识别数据积累比较多的话two-stage会比较好。百度的落地场景，比如工业仪表识别、车牌识别都用到端到端解决方案。
+
+#### Q: 二阶段的端到端的场景文本识别方法的不足有哪些？
+
+**A**：这类方法一般需要设计针对ROI提取特征的方法，而ROI操作一般比较耗时。
+
+#### Q: AAAI 2021最新的端到端场景文本识别PGNet算法有什么特点？
+
+**A**：PGNet不需要字符级别的标注，NMS操作以及ROI操作。同时提出预测文本行内的阅读顺序模块和基于图的修正模块来提升文本识别效果。该算法是百度自研，近期会在PaddleOCR开源。
+
+### 1.4 评估方法
+
+#### Q: OCR领域常用的评估指标是什么？
+
+**A**：对于两阶段的可以分开来看，分别是检测和识别阶段
+
+（1）检测阶段：先按照检测框和标注框的IOU评估，IOU大于某个阈值判断为检测准确。这里检测框和标注框不同于一般的通用目标检测框，是采用多边形进行表示。检测准确率：正确的检测框个数在全部检测框的占比，主要是判断检测指标。检测召回率：正确的检测框个数在全部标注框的占比，主要是判断漏检的指标。
+
+（2）识别阶段：
+字符识别准确率，即正确识别的文本行占标注的文本行数量的比例，只有整行文本识别对才算正确识别。
+
+（3）端到端统计：
+端对端召回率：准确检测并正确识别文本行在全部标注文本行的占比；
+端到端准确率：准确检测并正确识别文本行在 检测到的文本行数量 的占比；
+准确检测的标准是检测框与标注框的IOU大于某个阈值，正确识别的检测框中的文本与标注的文本相同。
+
+### 1.5 垂类场景实现思路
+
+#### Q：背景干扰的文字（如印章盖到落款上，需要识别落款或者印章中的文字），如何识别？
+
+**A**：（1）在人眼确认可识别的条件下，对于背景有干扰的文字，首先要保证检测框足够准确，如果检测框不准确，需要考虑是否可以通过过滤颜色等方式对图像预处理并且增加更多相关的训练数据；在识别的部分，注意在训练数据中加入背景干扰类的扩增图像。
+
+（2）如果MobileNet模型不能满足需求，可以尝试ResNet系列大模型来获得更好的效果。
+
+#### Q：请问对于图片中的密集文字，有什么好的处理办法吗？
+
+**A**：可以先试用预训练模型测试一下，例如DB+CRNN，判断下密集文字图片中是检测还是识别的问题，然后针对性的改善。还有一种是如果图象中密集文字较小，可以尝试增大图像分辨率，对图像进行一定范围内的拉伸，将文字稀疏化，提高识别效果。
+
+#### Q： 文本行较紧密的情况下如何准确检测？
+
+**A**：使用基于分割的方法，如DB，检测密集文本行时，最好收集一批数据进行训练，并且在训练时，并将生成二值图像的shrink_ratio参数调小一些。
+
+#### Q：对于一些在识别时稍微模糊的文本，有没有一些图像增强的方式？
+
+**A**：在人类肉眼可以识别的前提下，可以考虑图像处理中的均值滤波、中值滤波或者高斯滤波等模糊算子尝试。也可以尝试从数据扩增扰动来强化模型鲁棒性，另外新的思路有对抗性训练和超分SR思路，可以尝试借鉴。但目前业界尚无普遍认可的最优方案，建议优先在数据采集阶段增加一些限制提升图片质量。
+
+#### Q：低像素文字或者字号比较小的文字有什么超分辨率方法吗
+
+**A**：超分辨率方法分为传统方法和基于深度学习的方法。基于深度学习的方法中，比较经典的有SRCNN，另外CVPR2020也有一篇超分辨率的工作可以参考文章：Unpaired Image Super-Resolution using Pseudo-Supervision，但是没有充分的实践验证过，需要看实际场景下的效果。
+
+#### Q：对于一些尺寸较大的文档类图片，在检测时会有较多的漏检，怎么避免这种漏检的问题呢？
+
+**A**：PaddleOCR中在图像最长边大于960时，将图像等比例缩放为长边960的图像再进行预测，对于这种图像，可以通过修改det_limit_side_len，增大检测的最长边：tools/infer/utility.py#L42
+
+#### Q：文档场景中，使用DB模型会出现整行漏检的情况应该怎么解决？
+
+**A**：可以在预测时调小 det_db_box_thresh 阈值，默认为0.5, 可调小至0.3观察效果。
+
+#### Q: 弯曲文本（如略微形变的文档图像）漏检问题
+
+**A**: db后处理中计算文本框平均得分时，是求rectangle区域的平均分数，容易造成弯曲文本漏检，已新增求polygon区域的平均分数，会更准确，但速度有所降低，可按需选择，在相关pr中可查看[可视化对比效果](https://github.com/PaddlePaddle/PaddleOCR/pull/2604)。该功能通过参数 [det_db_score_mode](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.1/tools/infer/utility.py#L51)进行选择，参数值可选[`fast`(默认)、`slow`]，`fast`对应原始的rectangle方式，`slow`对应polygon方式。感谢用户[buptlihang](https://github.com/buptlihang)提[pr](https://github.com/PaddlePaddle/PaddleOCR/pull/2574)帮助解决该问题🌹。
+
+#### Q：如何识别文字比较长的文本？
+
+**A**：在中文识别模型训练时，并不是采用直接将训练样本缩放到[3,32,320]进行训练，而是先等比例缩放图像，保证图像高度为32，宽度不足320的部分补0，宽高比大于10的样本直接丢弃。预测时，如果是单张图像预测，则按上述操作直接对图像缩放，不做宽度320的限制。如果是多张图预测，则采用batch方式预测，每个batch的宽度动态变换，采用这个batch中最长宽度。
+
+#### Q：如何识别带空格的英文行文本图像？
+
+**A**：空格识别可以考虑以下两种方案：
+
+(1)优化文本检测算法。检测结果在空格处将文本断开。这种方案在检测数据标注时，需要将含有空格的文本行分成好多段。
+
+(2)优化文本识别算法。在识别字典里面引入空格字符，然后在识别的训练数据中，如果用空行，进行标注。此外，合成数据时，通过拼接训练数据，生成含有空格的文本。
+
+#### Q：弯曲文本有试过opencv的TPS进行弯曲校正吗？
+
+**A**：opencv的tps需要标出上下边界对应的点，这个点很难通过传统方法或者深度学习方法获取。PaddleOCR里StarNet网络中的tps模块实现了自动学点，自动校正，可以直接尝试这个。
+
+#### Q: 如何识别招牌或者广告图中的艺术字？
+
+**A**: 招牌或者广告图中的艺术字是文本识别一个非常有挑战性的难题，因为艺术字中的单字和印刷体相比，变化非常大。如果需要识别的艺术字是在一个词典列表内，可以将改每个词典认为是一个待识别图像模板，通过通用图像检索识别系统解决识别问题。可以尝试使用PaddleClas的图像识别系统。
+
+#### Q: 印章如何识别
+
+**A**：1. 使用带tps的识别网络或abcnet,2.使用极坐标变换将图片拉平之后使用crnn
+
+#### Q: 使用预训练模型进行预测，对于特定字符识别识别效果较差，怎么解决？
+
+**A**: 由于我们所提供的识别模型是基于通用大规模数据集进行训练的，部分字符可能在训练集中包含较少，因此您可以构建特定场景的数据集，基于我们提供的预训练模型进行微调。建议用于微调的数据集中，每个字符出现的样本数量不低于300，但同时需要注意不同字符的数量均衡。具体可以参考：微调。
+
+#### Q: 在使用训练好的识别模型进行预测的时候，发现有很多重复的字，这个怎么解决呢？
+
+**A**：可以看下训练的尺度和预测的尺度是否相同，如果训练的尺度为[3, 32, 320]，预测的尺度为[3, 64, 640]，则会有比较多的重复识别现象。
+
+#### Q: 图像正常识别出来的文字是OK的，旋转90度后识别出来的结果就比较差，有什么方法可以优化？
+
+**A**:  整图旋转90之后效果变差是有可能的，因为目前PPOCR默认输入的图片是正向的； 可以自己训练一个整图的方向分类器，放在预测的最前端（可以参照现有方向分类器的方式），或者可以基于规则做一些预处理，比如判断长宽等等。
+
+#### Q: 如何识别竹简上的古文？
+
+**A**：对于字符都是普通的汉字字符的情况，只要标注足够的数据，finetune模型就可以了。如果数据量不足，您可以尝试[StyleText](https://github.com/PFCCLab/StyleText)工具。
+而如果使用的字符是特殊的古文字、甲骨文、象形文字等，那么首先需要构建一个古文字的字典，之后再进行训练。
+
+#### Q: 只想要识别票据中的部分片段，重新训练它的话，只需要训练文本检测模型就可以了吗？问文本识别，方向分类还是用原来的模型这样可以吗？
+
+**A**：可以的。PaddleOCR的检测、识别、方向分类器三个模型是独立的，在实际使用中可以优化和替换其中任何一个模型。
+
+#### Q: 如何用PaddleOCR识别视频中的文字？
+
+**A**: 目前PaddleOCR主要针对图像做处理，如果需要视频识别，可以先对视频抽帧，然后用PPOCR识别。
+
+#### Q: 相机采集的图像为四通道，应该如何处理？
+
+**A**: 有两种方式处理：
+
+- 如果没有其他需要，可以在解码数据的时候指定模式为三通道，例如如果使用opencv，可以使用cv::imread(img_path, cv::IMREAD_COLOR)。
+- 如果其他模块需要处理四通道的图像，那也可以在输入PaddleOCR模块之前进行转换，例如使用cvCvtColor(&img,img3chan,CV_RGBA2RGB)。
+
+#### Q: 遇到中英文识别模型不支持的字符，该如何对模型做微调？
+
+**A**：如果希望识别中英文识别模型中不支持的字符，需要更新识别的字典，并完成微调过程。比如说如果希望模型能够进一步识别罗马数字，可以按照以下步骤完成模型微调过程。
+
+1. 准备中英文识别数据以及罗马数字的识别数据，用于训练，同时保证罗马数字和中英文识别数字的效果；
+2. 修改默认的字典文件，在后面添加罗马数字的字符；
+3. 下载PaddleOCR提供的预训练模型，配置预训练模型和数据的路径，开始训练。
+
+#### Q：特殊字符（例如一些标点符号）识别效果不好怎么办？
+
+**A**：首先请您确认要识别的特殊字符是否在字典中。
+如果字符在已经字典中但效果依然不好，可能是由于识别数据较少导致的，您可以增加相应数据finetune模型。
+
+---
+
+#### Q：单张图上多语种并存识别（如单张图印刷体和手写文字并存），应该如何处理？
+
+**A**：单张图像中存在多种类型文本的情况很常见，典型的以学生的试卷为代表，一张图像同时存在手写体和印刷体两种文本，这类情况下，可以尝试”1个检测模型+1个N分类模型+N个识别模型”的解决方案。
+其中不同类型文本共用同一个检测模型，N分类模型指额外训练一个分类器，将检测到的文本进行分类，如手写+印刷的情况就是二分类，N种语言就是N分类，在识别的部分，针对每个类型的文本单独训练一个识别模型，如手写+印刷的场景，就需要训练一个手写体识别模型，一个印刷体识别模型，如果一个文本框的分类结果是手写体，那么就传给手写体识别模型进行识别，其他情况同理。
+
+#### Q: 多语言的字典里是混合了不同的语种，这个是有什么讲究吗？统一到一个字典里会对精度造成多大的损失？
+
+**A**：统一到一个字典里，会造成最后一层FC过大，增加模型大小。如果有特殊需求的话，可以把需要的几种语言合并字典训练模型，合并字典之后如果引入过多的形近字，可能会造成精度损失，字符平衡的问题可能也需要考虑一下。在PaddleOCR里暂时将语言字典分开。
+
+#### Q：类似泰语这样的小语种，部分字会占用两个字符甚至三个字符，请问如何制作字典
+
+**A**：处理字符的时候，把多字符的当作一个字就行，字典中每行是一个字。
+
+---
+
+#### Q: 想把简历上的文字识别出来后，能够把关系一一对应起来，比如姓名和它后面的名字组成一对，籍贯、邮箱、学历等等都和各自的内容关联起来，这个应该如何处理，PPOCR目前支持吗？
+
+**A**:  这样的需求在企业应用中确实比较常见，但往往都是个性化的需求，没有非常规整统一的处理方式。常见的处理方式有如下两种：
+
+1. 对于单一版式、或者版式差异不大的应用场景，可以基于识别场景的一些先验信息，将识别内容进行配对； 比如运用表单结构信息：常见表单"姓名"关键字的后面，往往紧跟的就是名字信息
+2. 对于版式多样，或者无固定版式的场景， 需要借助于NLP中的NER技术，给识别内容中的某些字段，赋予key值
+
+由于这部分需求和业务场景强相关，难以用一个统一的模型去处理，目前PPOCR暂不支持。 如果需要用到NER技术，可以参照Paddle团队的另一个开源套件: [PaddlePaddle/ERNIE](https://github.com/PaddlePaddle/ERNIE)， 其提供的预训练模型ERNIE,  可以帮助提升NER任务的准确率。
+
+### 1.6 训练过程与模型调优
+
+#### Q: 增大batch_size模型训练速度没有明显提升
+
+**A**：如果batch_size打得太大，加速效果不明显的话，可以试一下增大初始化内存的值，运行代码前设置环境变量：
+export FLAGS_initial_cpu_memory_in_mb=2000  # 设置初始化内存约2G左右
+
+#### Q: 预测时提示图像过大，显存、内存溢出了，应该如何处理？
+
+**A**：可以按照这个PR的修改来缓解显存、内存占用 #2230
+
+#### Q: 识别训练时，训练集精度已经到达90了，但验证集精度一直在70，涨不上去怎么办？
+
+**A**：训练集精度90，测试集70多的话，应该是过拟合了，有两个可尝试的方法：（1）加入更多的增广方式或者调大增广prob的概率，默认为0.4。（2）调大系统的l2 decay值
+
+### 1.7 补充资料
+
+#### Q: 对于小白如何快速入门中文OCR项目实践？
+
+**A**：建议可以先了解OCR方向的基础知识，大概了解基础的检测和识别模型算法。然后在Github上可以查看OCR方向相关的repo。目前来看，从内容的完备性来看，PaddleOCR的中英文双语教程文档是有明显优势的，在数据集、模型训练、预测部署文档详实，可以快速入手。而且还有微信用户群答疑，非常适合学习实践。项目地址：PaddleOCR AI 快车道课程：<https://aistudio.baidu.com/aistudio/course/introduce/1519>
+
+## 2. PaddleOCR实战问题
+
+### 2.1 PaddleOCR repo
+
+#### Q: PaddleOCR develop分支和dygraph分支的区别？
+
+**A**：目前PaddleOCR有四个分支，分别是：
+
+- develop：基于Paddle静态图开发的分支，推荐使用paddle1.8 或者2.0版本，该分支具备完善的模型训练、预测、推理部署、量化裁剪等功能，领先于release/1.1分支。
+- release/1.1：PaddleOCR 发布的第一个稳定版本，基于静态图开发，具备完善的训练、预测、推理部署、量化裁剪等功能。
+- dygraph：基于Paddle动态图开发的分支，目前仍在开发中，未来将作为主要开发分支，运行要求使用Paddle2.0.0版本。
+- release/2.0-rc1-0：PaddleOCR发布的第二个稳定版本，基于动态图和paddle2.0版本开发，动态图开发的工程更易于调试，目前支，支持模型训练、预测，暂不支持移动端部署。
+
+如果您已经上手过PaddleOCR，并且希望在各种环境上部署PaddleOCR，目前建议使用静态图分支，develop或者release/1.1分支。如果您是初学者，想快速训练，调试PaddleOCR中的算法，建议尝鲜PaddleOCR dygraph分支。
+
+**注意**：develop和dygraph分支要求的Paddle版本、本地环境有差别，请注意不同分支环境安装部分的差异。
+
+#### Q：PaddleOCR与百度的其他OCR产品有什么区别？
+
+**A**：PaddleOCR主要聚焦通用ocr，如果有垂类需求，您可以用PaddleOCR+垂类数据自己训练；
+如果缺少带标注的数据，或者不想投入研发成本，建议直接调用开放的API，开放的API覆盖了目前比较常见的一些垂类。
+
+### 2.2 安装环境
+
+#### Q：OSError： [WinError 126] 找不到指定的模块。mac pro python 3.4 shapely import 问题
+
+**A**：这个问题是因为shapely库安装有误，可以参考 #212 这个issue重新安装一下
+
+#### Q：PaddlePaddle怎么指定GPU运行 os.environ["CUDA_VISIBLE_DEVICES"]这种不生效
+
+**A**：通过设置 export CUDA_VISIBLE_DEVICES='0'环境变量
+
+#### Q：PaddleOCR是否支持在Windows或Mac系统上运行？
+
+**A**：PaddleOCR已完成Windows和Mac系统适配，运行时注意两点：
+（1）在快速安装时，如果不想安装docker，可跳过第一步，直接从第二步安装paddle开始。
+（2）inference模型下载时，如果没有安装wget，可直接点击模型链接或将链接地址复制到浏览器进行下载，并解压放置到相应目录。
+
+#### Q：PaddleOCR安装失败，提示依赖冲突怎么办？
+
+**A**：这是常见问题，可以通过以下方式解决：
+（1）创建新的虚拟环境：`conda create -n paddleocr python=3.8`，然后激活环境安装
+（2）使用指定版本安装：`pip install paddleocr==3.2.0 --no-deps`，然后单独安装依赖
+（3）如果使用conda，可以尝试：`conda install -c conda-forge paddleocr`
+
+#### Q：GPU环境配置问题，CUDA版本不匹配怎么办？
+
+**A**：首先检查CUDA版本：`nvidia-smi`，然后安装对应版本的PaddlePaddle：
+- CUDA 11.8：`pip install paddlepaddle-gpu==3.0.0 -f https://www.paddlepaddle.org.cn/whl/linux/mkl/avx/stable.html`
+- CUDA 12.0：`pip install paddlepaddle-gpu==3.0.0 -f https://www.paddlepaddle.org.cn/whl/linux/mkl/avx/stable.html`
+验证GPU是否可用：`python -c "import paddle; print(paddle.is_compiled_with_cuda())"`
+
+#### Q：模型下载失败或下载缓慢怎么办？
+
+**A**：可以通过以下方式解决：
+（1）设置模型下载源：`os.environ['PADDLE_PDX_MODEL_SOURCE'] = 'BOS'`（使用百度云存储）
+（2）手动下载模型：直接访问模型链接下载并解压到本地目录
+（3）使用本地模型：在初始化时指定`model_dir`参数指向本地模型路径
+
+### 2.3 数据量说明
+
+#### Q：简单的对于精度要求不高的OCR任务，数据集需要准备多少张呢？
+
+**A**：（1）训练数据的数量和需要解决问题的复杂度有关系。难度越大，精度要求越高，则数据集需求越大，而且一般情况实际中的训练数据越多效果越好。
+
+（2）对于精度要求不高的场景，检测任务和识别任务需要的数据量是不一样的。对于检测任务，500张图像可以保证基本的检测效果。对于识别任务，需要保证识别字典中每个字符出现在不同场景的行文本图像数目需要大于200张（举例，如果有字典中有5个字，每个字都需要出现在200张图片以上，那么最少要求的图像数量应该在200-1000张之间），这样可以保证基本的识别效果。
+
+#### Q：请问PaddleOCR项目中的中文超轻量和通用模型用了哪些数据集？训练多少样本，gpu什么配置，跑了多少个epoch，大概跑了多久？
+
+**A**：
+（1）检测的话，LSVT街景数据集共3W张图像，超轻量模型，150epoch左右，2卡V100 跑了不到2天；通用模型：2卡V100 150epoch 不到4天。
+（2）识别的话，520W左右的数据集（真实数据26W+合成数据500W）训练，超轻量模型：4卡V100，总共训练了5天左右。通用模型：4卡V100，共训练6天。
+
+超轻量模型训练分为2个阶段：
+(1)全量数据训练50epoch，耗时3天
+(2)合成数据+真实数据按照1:1数据采样，进行finetune训练200epoch，耗时2天
+
+通用模型训练：
+真实数据+合成数据，动态采样(1：1)训练，200epoch，耗时 6天左右。
+
+#### Q：训练文字识别模型，真实数据有30w，合成数据有500w，需要做样本均衡吗？
+
+**A**：需要，一般需要保证一个batch中真实数据样本和合成数据样本的比例是5：1~10：1左右效果比较理想。如果合成数据过大，会过拟合到合成数据，预测效果往往不佳。还有一种启发性的尝试是可以先用大量合成数据训练一个base模型，然后再用真实数据微调，在一些简单场景效果也是会有提升的。
+
+#### Q: 当训练数据量少时，如何获取更多的数据？
+
+**A**：当训练数据量少时，可以尝试以下三种方式获取更多的数据：（1）人工采集更多的训练数据，最直接也是最有效的方式。（2）基于PIL和opencv基本图像处理或者变换。例如PIL中ImageFont, Image, ImageDraw三个模块将文字写到背景中，opencv的旋转仿射变换，高斯滤波等。（3）利用数据生成算法合成数据，例如pix2pix等算法。
+
+### 2.4 数据标注与生成
+
+> [!NOTE]
+> StyleText 已经移动到 [PFCCLab/StyleText](https://github.com/PFCCLab/StyleText)
+
+#### Q: Style-Text 如何不文字风格迁移，就像普通文本生成程序一样默认字体直接输出到分割的背景图？
+
+**A**：使用image_synth模式会输出fake_bg.jpg，即为背景图。如果想要批量提取背景，可以稍微修改一下代码，将fake_bg保存下来即可。要修改的位置：
+<https://github.com/PaddlePaddle/PaddleOCR/blob/de3e2e7cd3b8b65ee02d7a41e570fa5b511a3c1d/StyleText/engine/synthesisers.py#L68>
+
+#### Q: 能否修改StyleText配置文件中的分辨率？
+
+**A**：StyleText目前的训练数据主要是高度32的图片，建议不要改变高度。未来我们会支持更丰富的分辨率。
+
+#### Q: StyleText是否可以更换字体文件？
+
+**A**：StyleText项目中的字体文件为标准字体，主要用作模型的输入部分，不能够修改。
+StyleText的用途主要是：提取style_image中的字体、背景等style信息，根据语料生成同样style的图片。
+
+#### Q: StyleText批量生成图片为什么没有输出？
+
+**A**：需要检查以下您配置文件中的路径是否都存在。尤其要注意的是[label_file配置](https://github.com/PFCCLab/StyleText/blob/main/README_ch.md#%E4%B8%89%E5%BF%AB%E9%80%9F%E4%B8%8A%E6%89%8B)。
+如果您使用的style_image输入没有label信息，您依然需要提供一个图片文件列表。
+
+#### Q：使用StyleText进行数据合成时，文本(TextInput)的长度远超StyleInput的长度，该怎么处理与合成呢？
+
+**A**：在使用StyleText进行数据合成的时候，建议StyleInput的长度长于TextInput的长度。有2种方法可以处理上述问题：
+
+1. 将StyleInput按列的方向进行复制与扩充，直到其超过TextInput的长度。
+2. 将TextInput进行裁剪，保证每段TextInput都稍短于StyleInput，分别合成之后，再拼接在一起。
+
+实际使用中发现，使用第2种方法的效果在长文本合成的场景中的合成效果更好，StyleText中提供的也是第2种数据合成的逻辑。
+
+#### Q: StyleText 合成数据效果不好？
+
+**A**：StyleText模型生成的数据主要用于OCR识别模型的训练。PaddleOCR目前识别模型的输入为32 x N，因此当前版本模型主要适用高度为32的数据。
+建议要合成的数据尺寸设置为32 x N。尺寸相差不多的数据也可以生成，尺寸很大或很小的数据效果确实不佳。
+
+### 2.5 预训练模型与微调
+
+#### Q：如何更换文本检测/识别的backbone？
+
+**A**：无论是文字检测，还是文字识别，骨干网络的选择是预测效果和预测效率的权衡。一般，选择更大规模的骨干网络，例如ResNet101_vd，则检测或识别更准确，但预测耗时相应也会增加。而选择更小规模的骨干网络，例如MobileNetV3_small_x0_35，则预测更快，但检测或识别的准确率会大打折扣。幸运的是不同骨干网络的检测或识别效果与在ImageNet数据集图像1000分类任务效果正相关。飞桨图像分类套件PaddleClas汇总了ResNet_vd、Res2Net、HRNet、MobileNetV3、GhostNet等23种系列的分类网络结构，在上述图像分类任务的top1识别准确率，GPU(V100和T4)和CPU(骁龙855)的预测耗时以及相应的117个预训练模型下载地址。
+
+（1）文字检测骨干网络的替换，主要是确定类似于ResNet的4个stages，以方便集成后续的类似FPN的检测头。此外，对于文字检测问题，使用ImageNet训练的分类预训练模型，可以加速收敛和效果提升。
+
+（2）文字识别的骨干网络的替换，需要注意网络宽高stride的下降位置。由于文本识别一般宽高比例很大，因此高度下降频率少一些，宽度下降频率多一些。可以参考PaddleOCR中MobileNetV3骨干网络的改动。
+
+#### Q: 参照文档做实际项目时，是重新训练还是在官方训练的基础上进行训练？具体如何操作？
+
+**A**： 基于官方提供的模型，进行finetune的话，收敛会更快一些。 具体操作上，以识别模型训练为例：如果修改了字符文件，可以设置pretraind_model为官方提供的预训练模型
+
+#### Q: 下载的识别模型解压后缺失文件，没有期望的inference.pdiparams, inference.pdmodel等文件
+
+**A**：用解压软件解压可能会出现这个问题，建议二次解压下或者用命令行解压tar xf
+
+#### Q: 为什么在checkpoints中load下载的预训练模型会报错？
+
+**A**：这里有两个不同的概念：
+
+pretrained_model：指预训练模型，是已经训练完成的模型。这时会load预训练模型的参数，但并不会load学习率、优化器以及训练状态等。如果需要finetune，应该使用pretrained。
+checkpoints：指之前训练的中间结果，例如前一次训练到了100个epoch，想接着训练。这时会load尝试所有信息，包括模型的参数，之前的状态等。
+
+#### Q: 如何对检测模型finetune，比如冻结前面的层或某些层使用小的学习率学习？
+
+**A**：如果是冻结某些层，可以将变量的stop_gradient属性设置为True，这样计算这个变量之前的所有参数都不会更新了，参考：<https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/faq/train_cn.html#id4>
+
+如果对某些层使用更小的学习率学习，静态图里还不是很方便，一个方法是在参数初始化的时候，给权重的属性设置固定的学习率，参考：<https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/api/paddle/fluid/param_attr/ParamAttr_cn.html#paramattr>
+
+实际上我们实验发现，直接加载模型去fine-tune，不设置某些层不同学习率，效果也都不错
+
+### 2.6 模型超参调整
+
+#### Q: DB检测训练输入尺寸640，可以改大一些吗？
+
+**A**：不建议改大。检测模型训练输入尺寸是预处理中random crop后的尺寸，并非直接将原图进行resize，多数场景下这个尺寸并不小了，改大后可能反而并不合适，而且训练会变慢。另外，代码里可能有的地方参数按照预设输入尺寸适配的，改大后可能有隐藏风险。
+
+#### Q: 预处理部分，图片的长和宽为什么要处理成32的倍数？
+
+**A**：以检测中的resnet骨干网络为例，图像输入网络之后，需要经过5次2倍降采样，共32倍，因此建议输入的图像尺寸为32的倍数。
+
+#### Q: 在识别模型中，为什么降采样残差结构的stride为(2, 1)？
+
+**A**: stride为(2, 1)，表示在图像y方向（高度方向）上stride为2，x方向（宽度方向）上为1。由于待识别的文本图像通常为长方形，这样只在高度方向做下采样，尽量保留宽度方向的序列信息，避免宽度方向下采样后丢失过多的文字信息。
+
+#### Q：训练识别时，如何选择合适的网络输入shape？
+
+**A**：一般高度采用32，最长宽度的选择，有两种方法：
+
+（1）统计训练样本图像的宽高比分布。最大宽高比的选取考虑满足80%的训练样本。
+
+（2）统计训练样本文字数目。最长字符数目的选取考虑满足80%的训练样本。然后中文字符长宽比近似认为是1，英文认为3：1，预估一个最长宽度。
+
+#### Q：识别模型框出来的位置太紧凑，会丢失边缘的文字信息，导致识别错误
+
+**A**：可以在命令中加入 --det_db_unclip_ratio ，参数定义位置，这个参数是检测后处理时控制文本框大小的，默认1.6，可以尝试改成2.5或者更大，反之，如果觉得文本框不够紧凑，也可以把该参数调小。
+
+### 2.7 模型结构
+
+#### Q：文本识别训练不加LSTM是否可以收敛？
+
+**A**：理论上是可以收敛的，加上LSTM模块主要是为了挖掘文字之间的序列关系，提升识别效果。对于有明显上下文语义的场景效果会比较明显。
+
+#### Q：文本识别中LSTM和GRU如何选择？
+
+**A**：从项目实践经验来看，序列模块采用LSTM的识别效果优于GRU，但是LSTM的计算量比GRU大一些，可以根据自己实际情况选择。
+
+#### Q：对于CRNN模型，backbone采用DenseNet和ResNet_vd，哪种网络结构更好？
+
+**A**：Backbone的识别效果在CRNN模型上的效果，与Imagenet 1000 图像分类任务上识别效果和效率一致。在图像分类任务上ResnNet_vd（79%+）的识别精度明显优于DenseNet（77%+），此外对于GPU，Nvidia针对ResNet系列模型做了优化，预测效率更高，所以相对而言，resnet_vd是较好选择。如果是移动端，可以优先考虑MobileNetV3系列。
+
+#### Q:  如何根据不同的硬件平台选用不同的backbone？
+
+**A**：在不同的硬件上，不同的backbone的速度优势不同，可以根据不同平台的速度-精度图来确定backbone，这里可以参考[PaddleClas模型速度-精度图](https://github.com/PaddlePaddle/PaddleClas/tree/release/2.0/docs/zh_CN/models)。
+
+### 2.8 PP-OCR系统
+
+#### Q: 在PP-OCR系统中，文本检测的骨干网络为什么没有使用SE模块？
+
+**A**：SE模块是MobileNetV3网络一个重要模块，目的是估计特征图每个特征通道重要性，给特征图每个特征分配权重，提高网络的表达能力。但是，对于文本检测，输入网络的分辨率比较大，一般是640\*640，利用SE模块估计特征图每个特征通道重要性比较困难，网络提升能力有限，但是该模块又比较耗时，因此在PP-OCR系统中，文本检测的骨干网络没有使用SE模块。实验也表明，当去掉SE模块，超轻量模型大小可以减小40%，文本检测效果基本不受影响。详细可以参考PP-OCR技术文章，<https://arxiv.org/abs/2009.09941>.
+
+#### Q: PP-OCR系统中，文本检测的结果有置信度吗？
+
+**A**：文本检测的结果有置信度，由于推理过程中没有使用，所以没有显示的返回到最终结果中。如果需要文本检测结果的置信度，可以在[文本检测DB的后处理代码](../ppocr/postprocess/db_postprocess.py)的155行，添加scores信息。这样，在[检测预测代码](../tools/infer/predict_det.py)的197行，就可以拿到文本检测的scores信息。
+
+#### Q: DB文本检测，特征提取网络金字塔构建的部分代码在哪儿？
+
+**A**：特征提取网络金字塔构建的部分:[代码位置](../ppocr/modeling/necks/db_fpn.py)。ppocr/modeling文件夹里面是组网相关的代码，其中architectures是文本检测或者文本识别整体流程代码；backbones是骨干网络相关代码；necks是类似与FPN的颈函数代码；heads是提取文本检测或者文本识别预测结果相关的头函数；transforms是类似于TPS特征预处理模块。
+
+#### Q：PaddleOCR如何做到横排和竖排同时支持的？
+
+**A**：合成了一批竖排文字，逆时针旋转90度后加入训练集与横排一起训练。预测时根据图片长宽比判断是否为竖排，若为竖排则将crop出的文本逆时针旋转90度后送入识别网络。
+
+#### Q: 目前知识蒸馏有哪些主要的实践思路？
+
+**A**：知识蒸馏即利用教师模型指导学生模型的训练，目前有3种主要的蒸馏思路：
+
+1. 基于输出结果的蒸馏，即让学生模型学习教师模型的软标签（分类或者OCR识别等任务中）或者概率热度图（分割等任务中）。
+2. 基于特征图的蒸馏，即让学生模型学习教师模型中间层的特征图，拟合中间层的一些特征。
+3. 基于关系的蒸馏，针对不同的样本（假设个数为N），教师模型会有不同的输出，那么可以基于不同样本的输出，计算一个NxN的相关性矩阵，可以让学生模型去学习教师模型关于不同样本的相关性矩阵。
+
+当然，知识蒸馏方法日新月异，也欢迎大家提出更多的总结与建议。
+
+#### Q: 文字识别模型模型的输出矩阵需要进行解码才能得到识别的文本。代码中实现为preds_idx = preds.argmax(axis=2)，也就是最佳路径解码法。这是一种贪心算法，是每一个时间步只将最大概率的字符作为当前时间步的预测输出，但得到的结果不一定是最好的。为什么不使用beam search这种方式进行解码呢？
+
+**A**：实验发现，使用贪心的方法去做解码，识别精度影响不大，但是速度方面的优势比较明显，因此PaddleOCR中使用贪心算法去做识别的解码。
+
+### 2.9 端到端
+
+#### Q: 端到端算法PGNet是否支持中文识别，速度会很慢嘛？
+
+**A**：目前开源的PGNet算法模型主要是用于检测英文数字，对于中文的识别需要自己训练，大家可以使用开源的端到端中文数据集，而对于复杂文本（弯曲文本）的识别，也可以自己构造一批数据集针对进行训练，对于推理速度，可以先将模型转换为inference再进行预测，速度应该会相当可观。
+
+#### Q: 端到端算法PGNet提供了两种后处理方式，两者之间有什么区别呢？
+
+**A**: 两种后处理的区别主要在于速度的推理，config中PostProcess有fast/slow两种模式，slow模式的后处理速度慢，精度相对较高，fast模式的后处理速度快，精度也在可接受的范围之内。建议使用速度快的后处理方式。
+
+#### Q: 使用PGNet进行eval报错？
+
+**A**: 需要注意，我们目前在release/2.1更新了评测代码，目前支持A，B两种评测模式：
+
+- A模式：该模式主要为了方便用户使用，与训练集一样的标注文件就可以正常进行eval操作, 代码中默认是A模式。
+- B模式：该模式主要为了保证我们的评测代码可以和Total Text官方的评测方式对齐，该模式下直接加载官方提供的mat文件进行eval。
+
+#### Q: PGNet有中文预训练模型吗？
+
+**A**: 目前我们尚未提供针对中文的预训练模型，如有需要，可以尝试自己训练。具体需要修改的地方有：
+
+  1. [config文件中](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.1/configs/e2e/e2e_r50_vd_pg.yml#L23-L24)，字典文件路径及语种设置；
+  1. [网络结构中](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.1/ppocr/modeling/heads/e2e_pg_head.py#L181)，`out_channels`修改为字典中的字符数目+1（考虑到空格）；
+  1. [loss中](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.1/ppocr/losses/e2e_pg_loss.py#L93)，修改`37`为字典中的字符数目+1（考虑到空格）；
+
+#### Q: 用于PGNet的训练集，文本框的标注有要求吗？
+
+**A**: PGNet支持多点标注，比如4点、8点、14点等。但需要注意的是，标注点尽可能分布均匀（相邻标注点间隔距离均匀一致），且label文件中的标注点需要从标注框的左上角开始，按标注点顺时针顺序依次编写，以上问题都可能对训练精度造成影响。
+我们提供的，基于Total Text数据集的PGNet预训练模型使用了14点标注方式。
+
+#### Q: 用PGNet做进行端到端训练时，数据集标注的点的个数必须都是统一一样的吗? 能不能随意标点数，只要能够按顺时针从左上角开始标这样?
+
+**A**: 目前代码要求标注为统一的点数。
+
+### 2.10 模型效果与效果不一致
+
+#### Q: PP-OCR检测效果不好，该如何优化？
+
+**A**： 具体问题具体分析:
+如果在你的场景上检测效果不可用，首选是在你的数据上做finetune训练；
+如果图像过大，文字过于密集，建议不要过度压缩图像，可以尝试修改检测预处理的resize逻辑，防止图像被过度压缩；
+检测框大小过于紧贴文字或检测框过大，可以调整db_unclip_ratio这个参数，加大参数可以扩大检测框，减小参数可以减小检测框大小；
+检测框存在很多漏检问题，可以减小DB检测后处理的阈值参数det_db_box_thresh，防止一些检测框被过滤掉，也可以尝试设置det_db_score_mode为'slow';
+其他方法可以选择use_dilation为True，对检测输出的feature map做膨胀处理，一般情况下，会有效果改善；
+
+#### Q: GPU推理速度慢，如何优化性能？
+
+**A**：可以通过以下方式优化：
+（1）启用高性能推理：设置`enable_hpi=True`，自动选择最优加速策略
+（2）启用TensorRT加速：设置`use_tensorrt=True`，需要CUDA 11.8+和TensorRT 8.6+
+（3）使用半精度：设置`precision="fp16"`，可以显著提升速度
+（4）调整批处理大小：根据显存大小设置合适的`batch_size`
+（5）使用移动端模型：在精度要求不高时使用`PP-OCRv5_mobile`系列模型
+
+#### Q: GPU内存不足（CUDA out of memory）怎么办？
+
+**A**：可以通过以下方式解决：
+（1）减小批处理大小：将`batch_size`设置为1
+（2）减小图像尺寸：设置`det_limit_side_len=640`
+（3）启用内存优化：设置`enable_memory_optim=True`
+（4）限制GPU内存使用：设置`gpu_mem=200`
+（5）使用移动端模型：切换到`PP-OCRv5_mobile`系列模型
+
+#### Q: 如何选择合适的模型？
+
+**A**：根据应用场景选择：
+- 服务器高精度场景：使用`PP-OCRv5_server`系列，精度最高
+- 移动端部署：使用`PP-OCRv5_mobile`系列，模型小速度快
+- 实时处理：使用`PP-OCRv5_mobile`系列，推理速度快
+- 批量处理：使用`PP-OCRv5_server`系列，精度高
+- 多语言识别：使用`PP-OCRv5_multi_languages`，支持37种语言
+
+#### Q: 本地模型路径配置问题，如何在隔离网络环境下使用？
+
+**A**：可以通过以下方式配置：
+（1）使用本地模型路径：在初始化时指定`model_dir`参数
+（2）设置模型下载源：`os.environ['PADDLE_PDX_MODEL_SOURCE'] = 'BOS'`
+（3）手动下载模型：从官方链接下载模型并解压到本地
+（4）示例代码：
+```python
+ocr = PaddleOCR(
+    det_model_dir='./models/PP-OCRv5_server_det_infer/',
+    rec_model_dir='./models/PP-OCRv5_server_rec_infer/',
+    cls_model_dir='./models/PP-OCRv5_cls_infer/',
+    use_angle_cls=True,
+    lang='ch'
+)
+```
+
+#### Q：同一张图通用检测出21个条目，轻量级检测出26个 ，难道不是轻量级的好吗？
+
+**A**：可以主要参考可视化效果，通用模型更倾向于检测一整行文字，轻量级可能会有一行文字被分成两段检测的情况，不是数量越多，效果就越好。
+
+#### Q: DB有些框太贴文本了反而去掉了一些文本的边角影响识别，这个问题有什么办法可以缓解吗？
+
+**A**：可以把后处理的参数unclip_ratio适当调大一点。
+
+#### Q: 使用合成数据精调小模型后，效果可以，但是还没开源的小infer模型效果好，这是为什么呢？
+
+**A**：（1）要保证使用的配置文件和pretrain weights是对应的；
+
+（2）在微调时，一般都需要真实数据，如果使用合成数据，效果反而可能会有下降，PaddleOCR中放出的识别inference模型也是基于预训练模型在真实数据上微调得到的，效果提升比较明显；
+
+（3）在训练的时候，文本长度超过25的训练图像都会被丢弃，因此需要看下真正参与训练的图像有多少，太少的话也容易过拟合。
+
+#### Q: 表格识别中，如何提高单字的识别结果？
+
+**A**: 首先需要确认一下检测模型有没有有效的检测出单个字符，如果没有的话，需要在训练集当中添加相应的单字数据集。
+
+#### Q: 动态图分支(dygraph,release/2.0)，训练模型和推理模型效果不一致
+
+**A**：当前问题表现为：使用训练完的模型直接测试结果较好，但是转换为inference model后，预测结果不一致；出现这个问题一般是两个原因：
+
+1. 预处理函数设置的不一致
+2. 后处理参数不一致 repo中config.yml文件的前后处理参数和inference预测默认的超参数有不一致的地方，建议排查下训练模型预测和inference预测的前后处理， 参考issue。
+
+#### Q: 自己训练的det模型，在同一张图片上，inference模型与eval模型结果差别很大，为什么？
+
+**A**：这是由于图片预处理不同造成的。如果训练的det模型图片输入并不是默认的shape[600, 600]，eval的程序中图片预处理方式与train时一致 （由xxx_reader.yml中的test_image_shape参数决定缩放大小，但predict_eval.py中的图片预处理方式由程序里的preprocess_params决定， 最好不要传入max_side_len，而是传入和训练时一样大小的test_image_shape。
+
+#### Q: 训练模型和测试模型的检测结果差距较大
+
+**A**：1. 检查两个模型使用的后处理参数是否是一样的，训练的后处理参数在配置文件中的PostProcess部分，测试模型的后处理参数在tools/infer/utility.py中，最新代码中两个后处理参数已保持一致。
+
+#### Q: PaddleOCR模型Python端预测和C++预测结果不一致？
+
+**A**：正常来说，python端预测和C++预测文本是一致的，如果预测结果差异较大， 建议首先排查diff出现在检测模型还是识别模型，或者尝试换其他模型是否有类似的问题。 其次，检查python端和C++端数据处理部分是否存在差异，建议保存环境，更新PaddleOCR代码再试下。 如果更新代码或者更新代码都没能解决，建议在PaddleOCR微信群里或者issue中抛出您的问题。
+
+用户总结的排查步骤：<https://github.com/PaddlePaddle/PaddleOCR/issues/2470>
+
+### 2.11 训练调试与配置文件
+
+#### Q: 某个类别的样本比较少，通过增加训练的迭代次数或者是epoch，变相增加小样本的数目，这样能缓解这个问题么？
+
+**A**： 尽量保证类别均衡， 某些类别样本少，可以通过补充合成数据的方式处理；实验证明训练集中出现频次较少的字符，识别效果会比较差，增加迭代次数不能改变样本量少的问题。
+
+#### Q：文本检测换成自己的数据没法训练，有一些”###”是什么意思？
+
+**A**：数据格式有问题，”###” 表示要被忽略的文本区域，所以你的数据都被跳过了，可以换成其他任意字符或者就写个空的。
+
+#### Q：如何调试数据读取程序？
+
+**A**：tools/train.py中有一个test_reader()函数用于调试数据读取。
+
+#### Q：中文文本检测、文本识别构建训练集的话，大概需要多少数据量
+
+**A**：检测需要的数据相对较少，在PaddleOCR模型的基础上进行Fine-tune，一般需要500张可达到不错的效果。 识别分英文和中文，一般英文场景需要几十万数据可达到不错的效果，中文则需要几百万甚至更多。
+
+#### Q: config yml文件中的ratio_list参数的作用是什么？
+
+**A**: 在动态图中，ratio_list在有多个数据源的情况下使用，ratio_list中的每个值是每个epoch从对应数据源采样数据的比例。如ratio_list=[0.3,0.2]，label_file_list=['data1','data2'],代表每个epoch的训练数据包含data1 30%的数据，和data2里 20%的数据，ratio_list中数值的和不需要等于1。ratio_list和label_file_list的长度必须一致。
+
+静态图检测数据采样的逻辑与动态图不同，但基本不影响训练精度。
+
+在静态图中，使用 检测 dataloader读取数据时，会先设置每个epoch的数据量，比如这里设置为1000，ratio_list中的值表示在1000中的占比，比如ratio_list是[0.3, 0.7]，则表示使用两个数据源，每个epoch从第一个数据源采样1000*0.3=300张图，从第二个数据源采样700张图。ratio_list的值的和也不需要等于1。
+
+#### Q: iaa里面添加的数据增强方式，是每张图像训练都会做增强还是随机的？如何添加一个数据增强方法？
+
+**A**：iaa增强的训练配置参考：[这里](https://github.com/PaddlePaddle/PaddleOCR/blob/0ccc1720c252beb277b9e522a1b228eb6abffb8a/configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml#L82)。其中{ 'type': Fliplr, 'args': { 'p': 0.5 } } p是概率。新增数据增强，可以参考[这个方法](https://github.com/PaddlePaddle/PaddleOCR/blob/release%2F2.1/doc/doc_ch/add_new_algorithm.md#%E6%95%B0%E6%8D%AE%E5%8A%A0%E8%BD%BD%E5%92%8C%E5%A4%84%E7%90%86)
+
+#### Q: 怎么加速训练过程呢？
+
+**A**：OCR模型训练过程中一般包含大量的数据增广，这些数据增广是比较耗时的，因此可以离线生成大量增广后的图像，直接送入网络进行训练，机器资源充足的情况下，也可以使用分布式训练的方法，可以参考[分布式训练教程文档](./version2.x/ppocr/blog/distributed_training.md)。
+
+#### Q: 一些特殊场景的数据识别效果差，但是数据量很少，不够用来finetune怎么办？
+
+**A**：您可以合成一些接近使用场景的数据用于训练。
+我们计划推出基于特定场景的文本数据合成工具，请您持续关注PaddleOCR的近期更新。
+
+#### Q: PaddleOCR可以识别灰度图吗？
+
+**A**：PaddleOCR的模型均为三通道输入。如果您想使用灰度图作为输入，建议直接用3通道的模式读入灰度图，
+或者将单通道图像转换为三通道图像再识别。例如，opencv的cvtColor函数就可以将灰度图转换为RGB三通道模式。
+
+#### Q: 如何合成手写中文数据集？
+
+**A**: 手写数据集可以通过手写单字数据集合成得到。随机选取一定数量的单字图片和对应的label，将图片高度resize为随机的统一高度后拼接在一起，即可得到合成数据集。对于需要添加文字背景的情况，建议使用阈值化将单字图片的白色背景处理为透明背景，再与真实背景图进行合成。具体可以参考文档[手写数据集](https://github.com/PaddlePaddle/PaddleOCR/blob/a72d6f23be9979e0c103d911a9dca3e4613e8ccf/doc/doc_ch/handwritten_datasets.md)。
+
+#### Q：PaddleOCR默认不是200个step保存一次模型吗？为啥文件夹下面都没有生成
+
+**A**：因为默认保存的起始点不是0，而是4000，将eval_batch_step [4000, 5000]改为[0, 2000] 就是从第0次迭代开始，每2000迭代保存一次模型
+
+#### Q: PaddleOCR在训练的时候一直使用cosine_decay的学习率下降策略，这是为什么呢？
+
+**A**：cosine_decay表示在训练的过程中，学习率按照cosine的变化趋势逐渐下降至0，在迭代轮数更长的情况下，比常量的学习率变化策略会有更好的收敛效果，因此在实际训练的时候，均采用了cosine_decay，来获得精度更高的模型。
+
+#### Q: Cosine学习率的更新策略是怎样的？训练过程中为什么会在一个值上停很久？
+
+**A**: Cosine学习率的说明可以参考[这里](https://www.paddlepaddle.org.cn/documentation/docs/zh/api/paddle/optimizer/lr/CosineAnnealingDecay_cn.html#cosineannealingdecay)
+
+在PaddleOCR中，为了让学习率更加平缓，我们将其中的epoch调整成了iter。
+学习率的更新会和总的iter数量有关。当iter比较大时，会经过较多iter才能看出学习率的值有变化。
+
+#### Q: 之前的CosineWarmup方法为什么不见了？
+
+**A**: 我们对代码结构进行了调整，目前的Cosine可以覆盖原有的CosineWarmup的功能，只需要在配置文件中增加相应配置即可。
+例如下面的代码，可以设置warmup为2个epoch：
+
+```
+lr:
+  name: Cosine
+  learning_rate: 0.001
+  warmup_epoch: 2
+```
+
+#### Q: 训练识别和检测时学习率要加上warmup，目的是什么？
+
+**A**: Warmup机制先使学习率从一个较小的值逐步升到一个较大的值，而不是直接就使用较大的学习率，这样有助于模型的稳定收敛。在OCR检测和OCR识别中，一般会带来精度~0.5%的提升。
+
+#### Q: 关于dygraph分支中，文本识别模型训练，要使用数据增强应该如何设置？
+
+**A**：可以参考[配置文件](../configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml)在`Train['dataset']['transforms']`添加RecAug字段，使数据增强生效。可以通过添加对aug_prob设置，表示每种数据增强采用的概率。aug_prob默认是0.4。详细设置可以参考[ISSUE 1744](https://github.com/PaddlePaddle/PaddleOCR/issues/1744)。
+
+#### Q: 训练过程中，训练程序意外退出/挂起，应该如何解决？
+
+**A**: 考虑内存，显存（使用GPU训练的话）是否不足，可在配置文件中，将训练和评估的batch size调小一些。需要注意，训练batch size调小时，学习率learning rate也要调小，一般可按等比例调整。
+
+#### Q: 训练程序启动后直到结束，看不到训练过程log？
+
+**A**: 可以从以下三方面考虑：
+    1. 检查训练进程是否正常退出、显存占用是否释放、是否有残留进程，如果确定是训练程序卡死，可以检查环境配置，遇到环境问题建议使用docker，可以参考说明文档[安装](./version3.x/installation.md)。
+    2. 检查数据集的数据量是否太小，可调小batch size从而增加一个epoch中的训练step数量，或在训练config文件中，将参数print_batch_step改为1，即每一个step打印一次log信息。
+    3. 如果使用私有数据集训练，可先用PaddleOCR提供/推荐的数据集进行训练，排查私有数据集是否存在问题。
+
+#### Q: 配置文件中的参数num workers是什么意思，应该如何设置？
+
+**A**: 训练数据的读取需要硬盘IO，而硬盘IO速度远小于GPU运算速度，为了避免数据读取成为训练速度瓶颈，可以使用多进程读取数据，num workers表示数据读取的进程数量，0表示不使用多进程读取。在Linux系统下，多进程读取数据时，进程间通信需要基于共享内存，因此使用多进程读取数据时，建议设置共享内存不低于2GB，最好可以达到8GB，此时，num workers可以设置为CPU核心数。如果机器硬件配置较低，或训练进程卡死、dataloader报错，可以将num workers设置为0，即不使用多进程读取数据。
+
+### 2.12 预测
+
+#### Q: 为什么PaddleOCR检测预测是只支持一张图片测试？即test_batch_size_per_card=1
+
+**A**：测试的时候，对图像等比例缩放，最长边960，不同图像等比例缩放后长宽不一致，无法组成batch，所以设置为test_batch_size为1。
+
+#### Q: PaddleOCR支持tensorrt推理吗？
+
+**A**：支持的，需要在编译的时候将CMakeLists.txt文件当中，将相关代码option(WITH_TENSORRT "Compile demo with TensorRT." OFF)的OFF改成ON。关于服务器端部署的更多设置，可以参考飞桨官网
+
+#### Q: 如何使用TensorRT加速PaddleOCR预测？
+
+**A**： 目前paddle的dygraph分支已经支持了python和C++ TensorRT预测的代码，python端inference预测时把参数[--use_tensorrt=True](https://github.com/PaddlePaddle/PaddleOCR/blob/3ec57e8df9263de6fa897e33d2d91bc5d0849ef3/tools/infer/utility.py#L37)即可，
+C++TensorRT预测需要使用支持TRT的预测库并在编译时打开[-DWITH_TENSORRT=ON](https://github.com/PaddlePaddle/PaddleOCR/blob/3ec57e8df9263de6fa897e33d2d91bc5d0849ef3/deploy/cpp_infer/tools/build.sh#L15)。
+如果想修改其他分支代码支持TensorRT预测，可以参考[PR](https://github.com/PaddlePaddle/PaddleOCR/pull/2921)。
+
+注：建议使用TensorRT大于等于6.1.0.5以上的版本。
+
+#### Q: 为什么识别模型做预测的时候，预测图片的数量数量还会影响预测的精度
+
+**A**： 推理时识别模型默认的batch_size=6, 如预测图片长度变化大，可能影响预测效果。如果出现上述问题可在推理的时候设置识别bs=1，命令如下：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/ch/word_4.jpg" --rec_model_dir="./PP-OCRv3_mobile_rec_infer/" --rec_batch_num=1
+```
+
+### 2.13 推理部署
+
+#### Q：PaddleOCR模型推理方式有几种？各自的优缺点是什么
+
+**A**：目前推理方式支持基于训练引擎推理和基于预测引擎推理。
+
+（1）基于训练引擎推理不需要转换模型，但是需要先组网再load参数，语言只支持python，不适合系统集成。
+
+（2）基于预测引擎的推理需要先转换模型为inference格式，然后可以进行不需要组网的推理，语言支持c++和python，适合系统集成。
+
+#### Q：PaddleOCR中，对于模型预测加速，CPU加速的途径有哪些？基于TenorRT加速GPU对输入有什么要求？
+
+**A**：（1）CPU可以使用mkldnn进行加速；对于python inference的话，可以把`enable_mkldnn`改为`true`，[参考代码](https://github.com/PaddlePaddle/PaddleOCR/blob/ee9d22b4524eab44170568c98e73d742ce997ff2/tools/infer/utility.py#L132)，对于cpp inference的话，可参考[文档](./version3.x/inference_deployment/local_inference/cpp/cpp_local_deployment.md)
+
+（2）GPU需要注意变长输入问题等，TRT6 之后才支持变长输入
+
+#### Q：hubserving、pdserving这两种部署方式区别是什么？
+
+**A**：hubserving原本是paddlehub的配套服务部署工具，可以很方便的将paddlehub内置的模型部署为服务，paddleocr使用了这个功能，并将模型路径等参数暴露出来方便用户自定义修改。paddle serving是面向所有paddle模型的部署工具，文档中可以看到我们提供了快速版和标准版，其中快速版和hubserving的本质是一样的，而标准版基于rpc，更稳定，更适合分布式部署。
+
+#### Q: 目前paddle hub serving 只支持 imgpath，如果我想用imgurl 去哪里改呢？
+
+**A**：图片是在[这里](https://github.com/PaddlePaddle/PaddleOCR/blob/67ef25d593c4eabfaaceb22daade4577f53bed81/deploy/hubserving/ocr_system/module.py#L55)读取的， 可以参考下面的写法，将url path转化为np array
+
+```
+response = request.urlopen('http://i1.whymtj.com/uploads/tu/201902/9999/52491ae4ba.jpg')
+img_array = np.array(bytearray(response.read()), dtype=np.uint8)
+img = cv.imdecode(img_array, -1)
+```
+
+#### Q: C++ 端侧部署可以只对OCR的检测部署吗？
+
+**A**：可以的，识别和检测模块是解耦的。如果想对检测部署，需要自己修改一下main函数， 只保留检测相关就可以: [参考](https://github.com/PaddlePaddle/PaddleOCR/blob/de3e2e7cd3b8b65ee02d7a41e570fa5b511a3c1d/deploy/cpp_infer/src/main.cpp#L72)
+
+#### Q：服务部署可以只发布文本识别，而不带文本检测模型么？
+
+**A**：可以的。默认的服务部署是检测和识别串联预测的。也支持单独发布文本检测或文本识别模型，比如使用PaddleHUBPaddleOCR 模型时，deploy下有三个文件夹，分别是
+ocr_det：检测预测
+ocr_rec: 识别预测
+ocr_system: 检测识别串联预测
+
+#### Q: lite预测库和nb模型版本不匹配，该如何解决？
+
+**A**： 如果可以正常预测就不用管，如果这个问题导致无法正常预测，可以尝试使用同一个commit的Paddle Lite代码编译预测库和opt文件，可以参考[移动端部署教程](https://github.com/PaddlePaddle/PaddleOCR/blob/release%2F2.1/deploy/lite/readme.md)。
+
+#### Q：如何将PaddleOCR预测模型封装成SDK
+
+**A**：如果是Python的话，可以使用tools/infer/predict_system.py中的TextSystem进行sdk封装，如果是c++的话，可以使用deploy/cpp_infer/src下面的DBDetector和CRNNRecognizer完成封装
+
+#### Q：为什么PaddleOCR检测预测是只支持一张图片测试？即test_batch_size_per_card=1
+
+**A**：测试的时候，对图像等比例缩放，最长边960，不同图像等比例缩放后长宽不一致，无法组成batch，所以设置为test_batch_size为1。
+
+#### Q：为什么第一张张图预测时间很长，第二张之后预测时间会降低？
+
+**A**：第一张图需要显存资源初始化，耗时较多。完成模型加载后，之后的预测时间会明显缩短。
+
+#### Q: 采用Paddle-Lite进行端侧部署，出现问题，环境没问题
+
+**A**：如果你的预测库是自己编译的，那么你的nb文件也要自己编译，用同一个lite版本。不能直接用下载的nb文件，因为版本不同。
+
+#### Q: 如何多进程运行paddleocr？
+
+**A**：实例化多个paddleocr服务，然后将服务注册到注册中心，之后通过注册中心统一调度即可，关于注册中心，可以搜索eureka了解一下具体使用，其他的注册中心也行。
+
+#### Q: 如何多进程预测？
+
+**A**: 近期PaddleOCR新增了[多进程预测控制参数](https://github.com/PaddlePaddle/PaddleOCR/blob/a312647be716776c1aac33ff939ae358a39e8188/tools/infer/utility.py#L103)，`use_mp`表示是否使用多进程，`total_process_num`表示在使用多进程时的进程数。具体使用方式请参考[文档](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.1/doc/doc_ch/inference.md#1-%E8%B6%85%E8%BD%BB%E9%87%8F%E4%B8%AD%E6%96%87ocr%E6%A8%A1%E5%9E%8B%E6%8E%A8%E7%90%86)。
+
+#### Q: 如何部署PaddleOCR为Web服务？
+
+**A**：可以通过以下方式部署：
+（1）使用Flask部署：创建简单的Web API服务
+（2）使用gunicorn部署：`gunicorn -w 4 -b 0.0.0.0:5000 app:app`
+（3）使用异步处理：结合asyncio和ThreadPoolExecutor
+（4）示例代码：
+```python
+from flask import Flask, request, jsonify
+from paddleocr import PaddleOCR
+
+app = Flask(__name__)
+ocr = PaddleOCR(use_angle_cls=True, lang='ch')
+
+@app.route('/ocr', methods=['POST'])
+def ocr_api():
+    file = request.files['image']
+    result = ocr.ocr(file, cls=True)
+    return jsonify({'result': result})
+```
+
+#### Q: C++部署常见问题及解决方案？
+
+**A**：常见问题及解决方案：
+（1）找不到动态库：设置`export LD_LIBRARY_PATH=/path/to/paddle/lib:$LD_LIBRARY_PATH`
+（2）OpenCV版本不匹配：确保OpenCV版本与编译时一致
+（3）模型格式问题：确保使用正确的推理模型格式
+（4）编译问题：确保CMake配置正确，使用`cmake .. -DCMAKE_BUILD_TYPE=Release`
+
+#### Q: 服务性能优化建议？
+
+**A**：可以通过以下方式优化：
+（1）启用高性能推理：设置`enable_hpi=True`
+（2）使用批处理：合理设置`batch_size`
+（3）启用TensorRT：设置`use_tensorrt=True`
+（4）使用异步处理：避免阻塞请求
+（5）负载均衡：使用多个服务实例
+
+#### Q: 怎么解决paddleOCR在T4卡上有越预测越慢的情况？
+
+**A**：
+
+1. T4 GPU没有主动散热，因此在测试的时候需要在每次infer之后需要sleep 30ms，否则机器容易因为过热而降频(inference速度会变慢)，温度过高也有可能会导致宕机。
+2. T4在不使用的时候，也有可能会降频，因此在做benchmark的时候需要锁频，下面这两条命令可以进行锁频。
+
+```
+nvidia-smi -i 0 -pm ENABLED
+nvidia-smi --lock-gpu-clocks=1590 -i 0
+```
+
+#### Q: 在windows上进行cpp inference的部署时，总是提示找不到`paddle_fluid.dll`和`opencv_world346.dll`
+
+**A**：有2种方法可以解决这个问题：
+
+1. 将paddle预测库和opencv库的地址添加到系统环境变量中。
+2. 将提示缺失的dll文件拷贝到编译产出的`ocr_system.exe`文件夹中。
+
+#### Q: win下C++部署中文识别乱码的解决方法
+
+**A**: win下编码格式不是utf8,而ppocr_keys_v1.txt的编码格式的utf8，将ppocr_keys_v1.txt 的编码从utf-8修改为 Ansi 编码格式就行了。
+
+#### Q: windows 3060显卡GPU模式启动 加载模型慢
+
+**A**: 30系列的显卡需要使用cuda11。
+
+#### Q：想在Mac上部署，从哪里下载预测库呢？
+
+**A**：Mac上的Paddle预测库可以从这里下载：[https://paddle-inference-lib.bj.bcebos.com/mac/2.0.0/cpu_avx_openblas/paddle_inference.tgz](https://paddle-inference-lib.bj.bcebos.com/mac/2.0.0/cpu_avx_openblas/paddle_inference.tgz)
+
+#### Q：内网环境如何进行服务化部署呢？
+
+**A**：仍然可以使用PaddleServing或者HubServing进行服务化部署，保证内网地址可以访问即可。
+
+#### Q: 使用hub_serving部署，延时较高，可能的原因是什么呀？
+
+**A**: 首先，测试的时候第一张图延时较高，可以多测试几张然后观察后几张图的速度；其次，如果是在cpu端部署serving端模型（如backbone为ResNet34），耗时较慢，建议在cpu端部署mobile（如backbone为MobileNetV3）模型。
+
+#### Q: 在使用PaddleLite进行预测部署时，启动预测后卡死/手机死机？
+
+**A**: 请检查模型转换时所用PaddleLite的版本，和预测库的版本是否对齐。即PaddleLite版本为2.8，则预测库版本也要为2.8。
+
+#### Q: 预测时显存爆炸、内存泄漏问题？
+
+**A**: 打开显存/内存优化开关`enable_memory_optim`可以解决该问题，相关代码已合入，[查看详情](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.1/tools/infer/utility.py#L153)。
+
+#### Q: 在可视化预测结果时，能否使用其他字体？
+
+**A**: 可以！通过环境变量`PADDLE_PDX_LOCAL_FONT_FILE_PATH`指定本地字体文件路径即可，如`PADDLE_PDX_LOCAL_FONT_FILE_PATH=/root/fonts/simfang.ttf`。
diff --git a/docs/community/code_and_doc.en.md b/docs/community/code_and_doc.en.md
new file mode 100644
index 0000000..d02ba9d
--- /dev/null
+++ b/docs/community/code_and_doc.en.md
@@ -0,0 +1,336 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+# Appendix
+
+  This appendix contains python, document specifications and Pull Request process.
+
+## Appendix 1：Python Code Specification
+
+The Python code of PaddleOCR follows [PEP8 Specification]( https://www.python.org/dev/peps/pep-0008/ ), some of the key concerns include the following
+
+- Space
+
+    - Spaces should be added after commas, semicolons, colons, not before them
+
+      ```python linenums="1"
+      # true:
+      print(x, y)
+
+      # false:
+      print(x , y)
+      ```
+
+    - When specifying a keyword parameter or default parameter value in a function, do not use spaces on both sides of it
+
+      ```python linenums="1"
+      # true:
+      def complex(real, imag=0.0)
+      # false:
+      def complex(real, imag = 0.0)
+      ```
+
+- comment
+
+    - Inline comments: inline comments are indicated by the` # `sign. Two spaces should be left between code and` # `, and one space should be left between` # `and comments, for example
+
+      ```python linenums="1"
+      x = x + 1  # Compensate for border
+      ```
+
+    - Functions and methods: The definition of each function should include the following:
+
+        - Function description: Utility, input and output of function
+        - Args: Name and description of each parameter
+        - Returns: The meaning and type of the return value
+
+        ```python linenums="1"
+        def fetch_bigtable_rows(big_table, keys, other_silly_variable=None):
+            """Fetches rows from a Bigtable.
+
+            Retrieves rows pertaining to the given keys from the Table instance
+            represented by big_table.  Silly things may happen if
+            other_silly_variable is not None.
+
+            Args:
+                big_table: An open Bigtable Table instance.
+                keys: A sequence of strings representing the key of each table row
+                    to fetch.
+                other_silly_variable: Another optional variable, that has a much
+                    longer name than the other args, and which does nothing.
+
+            Returns:
+                A dict mapping keys to the corresponding table row data
+                fetched. Each row is represented as a tuple of strings. For
+                example:
+
+                {'Serak': ('Rigel VII', 'Preparer'),
+                'Zim': ('Irk', 'Invader'),
+                'Lrrr': ('Omicron Persei 8', 'Emperor')}
+
+                If a key from the keys argument is missing from the dictionary,
+                then that row was not found in the table.
+            """
+            pass
+        ```
+
+## Appendix 2: Document Specification
+
+### 2.1 Overall Description
+
+- Document Location: If you add new features to your original Markdown file, please **Do not re-create** a new file. If you don't know where to add it, you can first PR the code and then ask the official in commit.
+
+- New Markdown Document Name: Describe the content of the document in English, typically a combination of lowercase letters and underscores, such as `add_New_Algorithm.md`
+
+- New Markdown Document Format: Catalog - Body - FAQ
+
+  > The directory generation method can use [this site](https://ecotrust-canada.github.io/markdown-toc/ ) Automatically extract directories after copying MD contents, and then add `
+
+- English and Chinese: Any changes or additions to the document need to be made in both Chinese and English documents.
+
+### 2.2 Format Specification
+
+- Title format: The document title format follows the format of: Arabic decimal point combination-space-title (for example, `2.1 XXXX`, `2.XXXX`)
+
+- Code block: Displays code in code block format that needs to be run, describing the meaning of command parameters before the code block. for example:
+
+  > Pipeline of detection + direction Classify + recognition: Vertical text can be recognized after set direction classifier parameters`--use_angle_cls true`.
+  >
+  > ```bash linenums="1"
+  > paddleocr --image_dir ./imgs/11.jpg --use_angle_cls true
+  > ```
+
+- Variable References: If code variables or command parameters are referenced in line, they need to be represented in line code, for example, above `--use_angle_cls true` with one space in front and one space in back
+
+- Uniform naming: e.g. PP-OCRv2, PP-OCR mobile, `paddleocr` whl package, PPOCRLabel, Paddle Lite, etc.
+
+- Supplementary notes: Supplementary notes by reference format `>`.
+
+- Picture: If a picture is added to the description document, specify the naming of the picture (describing its content) and add the picture under `doc/`.
+
+- Title: Capitalize the first letter of each word in the title.
+
+## Appendix 3: Pull Request Description
+
+### 3.1 PaddleOCR Branch Description
+
+PaddleOCR will maintain two branches in the future, one for each:
+
+- release/x.x family branch: stable release version branch, also the default branch. PaddleOCR releases a new release branch based on feature updates and adapts to the release version of Paddle. As versions iterate, more and more release/x.x family branches are maintained by default with the latest version of the release branch.
+- dygraph branch: For the development branch, adapts the dygraph version of the Paddle dynamic graph to primarily develop new functionality. If you need to redevelop, choose the dygraph branch. To ensure that the dygraph branch pulls out the release/x.x branch when needed, the code for the dygraph branch can only use the valid API in the latest release branch of Paddle. That is, if a new API has been developed in the Paddle dygraph branch but has not yet appeared in the release branch code, do not use it in Paddle OCR. In addition, performance optimization, parameter tuning, policy updates that do not involve API can be developed normally.
+
+The historical branch of PaddleOCR will no longer be maintained in the future. These branches will continue to be maintained, considering that some of you may still be using them:
+
+Develop branch: This branch was used for the development and testing of static diagrams and is currently compatible with version >=1.7. If you have special needs, you can also use this branch to accommodate older versions of Paddle, but you won't update your code until you fix the bug.
+
+PaddleOCR welcomes you to actively contribute code to repo. Here are some basic processes for contributing code.
+
+### 3.2 PaddleOCR Code Submission Process And Specification
+
+If you are familiar with Git use, you can jump directly to [Some Conventions For Submitting Code in 3.2.10](#3210-some-conventions-for-submitting-code)
+
+#### 3.2.1 Create Your `Remote Repo`
+
+In PaddleOCR [GitHub Home]( https://github.com/PaddlePaddle/PaddleOCR ) Click the `Fork` button in the upper left corner to create a `remote repo`in your personal directory, such as `https://github.com/ {your_name}/PaddleOCR`.
+
+  ![banner](./images/banner.png)
+
+Clone `Remote repo`
+
+  ```bash linenums="1"
+  # pull code of develop branch
+  git clone https://github.com/{your_name}/PaddleOCR.git -b dygraph
+  cd PaddleOCR
+  ```
+
+Clone failures are mostly due to network reasons, try again later or configure the proxy
+
+#### 3.2.2 Login And Connect Using Token
+
+Start by viewing the information for the current `remote repo`.
+
+```bash linenums="1"
+git remote -v
+# origin    https://github.com/{your_name}/PaddleOCR.git (fetch)
+# origin    https://github.com/{your_name}/PaddleOCR.git (push)
+```
+
+Only the information of the clone `remote repo`, i.e. the PaddleOCR under your username, is available. Due to the change in Github's login method, you need to reconfigure the `remote repo` address by means of a Token. The token is generated as follows:
+
+1. Find Personal Access Tokens: Click on your avatar in the upper right corner of the Github page and choose Settings --> Developer settings --> Personal access tokens,
+
+2. Click Generate new token: Fill in the token name in Note, such as 'paddle'. In Select scopes, select repo (required), admin:repo_hook, delete_repo, etc. You can check them according to your needs. Then click Generate token to generate the token, and finally copy the generated token.
+
+Delete the original origin configuration
+
+```bash linenums="1"
+git remote rm origin
+```
+
+Change the remote branch to `https://oauth2:{token}@github.com/{your_name}/PaddleOCR.git`. For example, if the token value is 12345 and your user name is PPOCR, run the following command
+
+```bash linenums="1"
+git remote add origin https://oauth2:12345@github.com/PPOCR/PaddleOCR.git
+```
+
+This establishes a connection to our own `remote repo`. Next we create a remote host of the original PaddleOCR repo, named upstream.
+
+```bash linenums="1"
+git remote add upstream https://github.com/PaddlePaddle/PaddleOCR.git
+```
+
+Use `git remote -v` to view current `remote warehouse` information, output as follows, found to include two origin and two upstream of `remote repo` .
+
+```bash linenums="1"
+origin    https://github.com/{your_name}/PaddleOCR.git (fetch)
+origin    https://github.com/{your_name}/PaddleOCR.git (push)
+upstream    https://github.com/PaddlePaddle/PaddleOCR.git (fetch)
+upstream    https://github.com/PaddlePaddle/PaddleOCR.git (push)
+```
+
+This is mainly to keep the local repository up to date when subsequent pull request (PR) submissions are made.
+
+#### 3.2.3 Create Local Branch
+
+First get the latest code of upstream, then create a new_branch branch based on the dygraph of the upstream repo (upstream).
+
+```bash linenums="1"
+git fetch upstream
+git checkout -b new_branch upstream/dygraph
+```
+
+> If for a newly forked PaddleOCR project, the user's remote repo (origin) has the same branch updates as the upstream repository (upstream), you can also create a new local branch based on the default branch of the origin repo or a specified branch with the following command
+>
+> ```bash linenums="1"
+> # Create new_branch branch on user remote repo (origin) based on develop branch
+> git checkout -b new_branch origin/develop
+> # Create new_branch branch based on upstream remote repo develop branch
+> # If you need to create a new branch from upstream,
+> # you need to first use git fetch upstream to get upstream code
+> git checkout -b new_branch upstream/develop
+> ```
+
+The final switch to the new branch is displayed with the following output information.
+
+Branch new_branch set up to track remote branch develop from upstream.
+Switched to a new branch 'new_branch'
+
+After switching branches, file changes can be made on this branch
+
+#### 3.2.4 Use Pre-Commit Hook
+
+Paddle developers use the pre-commit tool to manage Git pre-submit hooks. It helps us format the source code (C++, Python) and automatically check for basic things (such as having only one EOL per file, not adding large files to Git) before committing it.
+
+The pre-commit test is part of the unit test in Travis-CI. PR that does not satisfy the hook cannot be submitted to PaddleOCR. Install it first and run it in the current directory：
+
+```bash linenums="1"
+pip install pre-commit
+pre-commit install
+```
+
+> 1. Paddle uses clang-format to adjust the C/C++ source code format. Make sure the `clang-format` version is above 3.8.
+>
+> 2. Yapf installed through pip install pre-commit is slightly different from conda install-c conda-forge pre-commit, and PaddleOCR developers use `pip install pre-commit`.
+
+#### 3.2.5 Modify And Submit Code
+
+If you make some changes on `README.Md` on PaddleOCR, you can view the changed file through `git status`, and then add the changed file using `git add`。
+
+```bash linenums="1"
+git status # View change files
+git add README.md
+pre-commit
+```
+
+Repeat these steps until the pre-commit format check does not error. As shown below.
+
+![img](./images/precommit_pass.png)
+
+Use the following command to complete the submission.
+
+```bash linenums="1"
+git commit -m "your commit info"
+```
+
+#### 3.2.6 Keep Local Repo Up To Date
+
+Get the latest code for upstream and update the current branch. Here the upstream comes from section 2.2, `Connecting to a remote repo`.
+
+```bash linenums="1"
+git fetch upstream
+# If you want to commit to another branch, you need to pull code from another branch of upstream, here is develop
+git pull upstream develop
+```
+
+#### 3.2.7 Push To Remote Repo
+
+```bash linenums="1"
+git push origin new_branch
+```
+
+#### 3.2.7 Submit Pull Request
+
+Click the new pull request to select the local branch and the target branch, as shown in the following figure. In the description of PR, fill in the functions completed by the PR. Next, wait for review, and if you need to modify something, update the corresponding branch in origin with the steps above.
+
+![banner](./images/pr.png)
+
+#### 3.2.8 Sign CLA Agreement And Pass Unit Tests
+
+Signing the CLA When submitting a Pull Request to PaddlePaddle for the first time, you need to sign a CLA (Contributor License Agreement) agreement to ensure that your code can be incorporated as follows:
+
+1. Please check the Check section in PR, find the license/cla, and click on the right detail to enter the CLA website
+
+2. Click Sign in with GitHub to agree on the CLA website and when clicked, it will jump back to your Pull Request page
+
+#### 3.2.9 Delete Branch
+
+- Remove remote branch
+
+  After PR is merged into the main repo, we can delete the branch of the remote repofrom the PR page.
+  You can also use `git push origin:branch name` to delete remote branches, such as:
+
+  ```bash linenums="1"
+  git push origin :new_branch
+  ```
+
+- Delete local branch
+
+  ```bash linenums="1"
+  # Switch to the development branch, otherwise the current branch cannot be deleted
+  git checkout develop
+
+  # Delete new_ Branch Branch
+  git branch -D new_branch
+  ```
+
+#### 3.2.10 Some Conventions For Submitting Code
+
+In order for official maintainers to better focus on the code itself when reviewing it, please follow the following conventions each time you submit your code:
+
+1）Please ensure that the unit tests in Travis-CI pass smoothly. If not, indicate that there is a problem with the submitted code, and the official maintainer generally does not review it.
+
+2）Before submitting a Pull Request.
+
+- Note the number of commits.
+
+  Reason: If you only modify one file and submit more than a dozen commits, each commit will only make a few modifications, which can be very confusing to the reviewer. The reviewer needs to look at each commit individually to see what changes have been made, and does not exclude the fact that changes between commits overlap each other.
+
+  Suggestion: Keep as few commits as possible each time you submit, and supplement your last commit with git commit --amend. For multiple commits that have been Push to a remote warehouse, you can refer to [squash commits after push](https://stackoverflow.com/questions/5667884/how-to-squash-commits-in-git-after-they-have-been-pushed ).
+
+- Note the name of each commit: it should reflect the content of the current commit, not be too arbitrary.
+
+3） If you have solved a problem, add in the first comment box of the Pull Request:fix #issue_number，This will automatically close the corresponding Issue when the Pull Request is merged. Key words include:close, closes, closed, fix, fixes, fixed, resolve, resolves, resolved,please choose the right vocabulary. Detailed reference [Closing issues via commit messages](https://help.github.com/articles/closing-issues-via-commit-messages).
+
+In addition, in response to the reviewer's comments, you are requested to abide by the following conventions:
+
+1） Each review comment from an official maintainer would like a response, which would better enhance the contribution of the open source community.
+
+- If you agree to the review opinion and modify it accordingly, give a simple Done.
+- If you disagree with the review, please give your own reasons for refuting.
+
+2）If there are many reviews:
+
+- Please give an overview of the changes.
+- Please reply with `start a review', not directly. The reason is that each reply sends an e-mail message, which can cause a mail disaster.
diff --git a/docs/community/code_and_doc.md b/docs/community/code_and_doc.md
new file mode 100644
index 0000000..88bedcc
--- /dev/null
+++ b/docs/community/code_and_doc.md
@@ -0,0 +1,327 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+# 附录
+
+本附录包含了Python、文档规范以及Pull Request流程，请各位开发者遵循相关内容
+
+## 附录1：Python代码规范
+
+PaddleOCR的Python代码遵循 [PEP8规范](https://www.python.org/dev/peps/pep-0008/)，其中一些关注的重点包括如下内容
+
+- 空格
+
+    - 空格应该加在逗号、分号、冒号后，而非他们的前面
+
+      ```python linenums="1"
+      # 正确：
+      print(x, y)
+
+      # 错误：
+      print(x , y)
+      ```
+
+    - 在函数中指定关键字参数或默认参数值时, 不要在其两侧使用空格
+
+      ```python linenums="1"
+      # 正确：
+      def complex(real, imag=0.0)
+      # 错误：
+      def complex(real, imag = 0.0)
+      ```
+
+- 注释
+
+    - 行内注释：行内注释使用 `#` 号表示，在代码与 `#` 之间需要空两个空格， `#`  与注释之间应当空一个空格，例如
+
+      ```python linenums="1"
+      x = x + 1  # Compensate for border
+      ```
+
+    - 函数和方法：每个函数的定义后的描述应该包括以下内容：
+
+        - 函数描述：函数的作用，输入输出的
+        - Args：每个参数的名字以及对该参数的描述
+        - Returns：返回值的含义和类型
+
+        ```python linenums="1"
+        def fetch_bigtable_rows(big_table, keys, other_silly_variable=None):
+            """Fetches rows from a Bigtable.
+
+            Retrieves rows pertaining to the given keys from the Table instance
+            represented by big_table.  Silly things may happen if
+            other_silly_variable is not None.
+
+            Args:
+                big_table: An open Bigtable Table instance.
+                keys: A sequence of strings representing the key of each table row
+                    to fetch.
+                other_silly_variable: Another optional variable, that has a much
+                    longer name than the other args, and which does nothing.
+
+            Returns:
+                A dict mapping keys to the corresponding table row data
+                fetched. Each row is represented as a tuple of strings. For
+                example:
+
+                {'Serak': ('Rigel VII', 'Preparer'),
+                'Zim': ('Irk', 'Invader'),
+                'Lrrr': ('Omicron Persei 8', 'Emperor')}
+
+                If a key from the keys argument is missing from the dictionary,
+                then that row was not found in the table.
+            """
+            pass
+        ```
+
+## 附录2：文档规范
+
+### 2.1 总体说明
+
+- 文档位置：如果您增加的新功能可以补充在原有的Markdown文件中，请**不要重新新建**一个文件。如果您对添加的位置不清楚，可以先PR代码，然后在commit中询问官方人员。
+
+- 新增Markdown文档名称：使用英文描述文档内容，一般由小写字母与下划线组合而成，例如  `add_new_algorithm.md`
+
+- 新增Markdown文档格式：目录 - 正文 - FAQ
+
+  > 目录生成方法可以使用 [此网站](https://ecotrust-canada.github.io/markdown-toc/) 将md内容复制之后自动提取目录，然后在md文件的每个标题前添加
+
+- 中英双语：任何对文档的改动或新增都需要分别在中文和英文文档上进行。
+
+### 2.2 格式规范
+
+- 标题格式：文档标题格式按照：阿拉伯数字小数点组合 - 空格 - 标题的格式（例如 `2.1 XXXX` ， `2. XXXX`）
+
+- 代码块：通过代码块格式展示需要运行的代码，在代码块前描述命令参数的含义。例如：
+
+  > 检测+方向分类器+识别全流程：设置方向分类器参数 `--use_angle_cls true` 后可对竖排文本进行识别。
+  >
+  > ```bash linenums="1"
+  > paddleocr --image_dir ./imgs/11.jpg --use_angle_cls true
+  > ```
+
+- 变量引用：如果在行内引用到代码变量或命令参数，需要用行内代码表示，例如上方  `--use_angle_cls true` ，并在前后各空一格
+
+- 统一命名：如PP-OCRv2、PP-OCR mobile、`paddleocr` whl包、PPOCRLabel、Paddle Lite等
+
+- 补充说明：通过引用格式 `>` 补充说明，或对注意事项进行说明
+
+- 图片：如果在说明文档中增加了图片，请规范图片的命名形式（描述图片内容），并将图片添加在 `doc/` 下
+
+## 附录3：Pull Request说明
+
+### 3.1 PaddleOCR分支说明
+
+PaddleOCR未来将维护2种分支，分别为：
+
+- release/x.x系列分支：为稳定的发行版本分支，也是默认分支。PaddleOCR会根据功能更新情况发布新的release分支，同时适配Paddle的release版本。随着版本迭代，release/x.x系列分支会越来越多，默认维护最新版本的release分支。
+- dygraph分支：为开发分支，适配Paddle动态图的dygraph版本，主要用于开发新功能。如果有同学需要进行二次开发，请选择dygraph分支。为了保证dygraph分支能在需要的时候拉出release/x.x分支，dygraph分支的代码只能使用Paddle最新release分支中有效的api。也就是说，如果Paddle dygraph分支中开发了新的api，但尚未出现在release分支代码中，那么请不要在PaddleOCR中使用。除此之外，对于不涉及api的性能优化、参数调整、策略更新等，都可以正常进行开发。
+
+PaddleOCR的历史分支，未来将不再维护。考虑到一些同学可能仍在使用，这些分支还会继续保留：
+
+- develop分支：这个分支曾用于静态图的开发与测试，目前兼容>=1.7版本的Paddle。如果有特殊需求，要适配旧版本的Paddle，那还可以使用这个分支，但除了修复bug外不再更新代码。
+
+PaddleOCR欢迎大家向repo中积极贡献代码，下面给出一些贡献代码的基本流程。
+
+### 3.2 PaddleOCR代码提交流程与规范
+
+> 如果你熟悉Git使用，可以直接跳转到 [3.2.10 提交代码的一些约定](#3210)
+
+#### 3.2.1 创建你的 `远程仓库`
+
+- 在PaddleOCR的 [GitHub首页](https://github.com/PaddlePaddle/PaddleOCR)，点击左上角 `Fork`  按钮，在你的个人目录下创建 `远程仓库`，比如`https://github.com/{your_name}/PaddleOCR`。
+
+	![banner](./images/banner.png)
+
+- 将 `远程仓库` Clone到本地
+
+  ```bash linenums="1"
+  # 拉取dygraph分支的代码
+  git clone https://github.com/{your_name}/PaddleOCR.git -b dygraph
+  cd PaddleOCR
+  ```
+
+> 多数情况下clone失败是由于网络原因，请稍后重试或配置代理
+
+#### 3.2.2 通过Token方式登录与建立连接
+
+首先查看当前 `远程仓库` 的信息。
+
+```bash linenums="1"
+git remote -v
+# origin    https://github.com/{your_name}/PaddleOCR.git (fetch)
+# origin    https://github.com/{your_name}/PaddleOCR.git (push)
+```
+
+只有clone的 `远程仓库` 的信息，也就是自己用户名下的 PaddleOCR。由于Github的登录方式变化，需要通过Token的方式重新配置 `远程仓库` 的地址。生成Token的方式如下：
+
+1. 找到个人访问令牌（token）：在Github页面右上角点击自己的头像，然后依次选择 Settings --> Developer settings --> Personal access tokens
+2. 点击 Generate new token：在Note中填入token名称，例如’paddle‘。在Select scopes选择repo（必选）、admin:repo_hook、delete_repo等，可根据自身需要勾选。然后点击Generate token生成token。最后复制生成的token。
+
+删除原始的origin配置
+
+```bash linenums="1"
+git remote rm origin
+```
+
+将remote分支改成 `https://oauth2:{token}@github.com/{your_name}/PaddleOCR.git`。例如：如果token值为12345，你的用户名为PPOCR，则运行下方命令
+
+```bash linenums="1"
+git remote add origin https://oauth2:12345@github.com/PPOCR/PaddleOCR.git
+```
+
+这样我们就与自己的 `远程仓库` 建立了连接。接下来我们创建一个原始 PaddleOCR 仓库的远程主机，命名为 upstream。
+
+```bash linenums="1"
+git remote add upstream https://github.com/PaddlePaddle/PaddleOCR.git
+```
+
+使用 `git remote -v` 查看当前 `远程仓库` 的信息，输出如下，发现包括了origin和upstream 2个 `远程仓库` 。
+
+```bash linenums="1"
+origin    https://oauth2:{token}@github.com/{your_name}/PaddleOCR.git (fetch)
+origin    https://oauth2:{token}@github.com/{your_name}/PaddleOCR.git (push)
+upstream    https://github.com/PaddlePaddle/PaddleOCR.git (fetch)
+upstream    https://github.com/PaddlePaddle/PaddleOCR.git (push)
+```
+
+这主要是为了后续在提交pull request(PR)时，始终保持本地仓库最新。
+
+#### 3.2.3 创建本地分支
+
+首先获取 upstream 的最新代码，然后基于上游仓库 (upstream)的dygraph创建new_branch分支。
+
+```bash linenums="1"
+git fetch upstream
+git checkout -b new_branch upstream/dygraph
+```
+
+> 如果对于新Fork的PaddleOCR项目，用户远程仓库(origin)与上游(upstream)仓库的分支更新情况相同，也可以基于origin仓库的默认分支或指定分支创建新的本地分支，命令如下。
+>
+> ```bash linenums="1"
+> # 基于用户远程仓库(origin)的dygraph创建new_branch分支
+> git checkout -b new_branch origin/dygraph
+>
+> # 基于用户远程仓库(origin)的默认分支创建new_branch分支
+> git checkout -b new_branch
+> ```
+
+最终会显示切换到新的分支，输出信息如下
+
+```bash linenums="1"
+Branch new_branch set up to track remote branch develop from upstream.
+Switched to a new branch 'new_branch'
+```
+
+切换分支之后即可在此分支上进行文件改动
+
+#### 3.2.4 使用pre-commit勾子
+
+Paddle 开发人员使用 pre-commit 工具来管理 Git 预提交钩子。 它可以帮助我们格式化源代码（C++，Python），在提交（commit）前自动检查一些基本事宜（如每个文件只有一个 EOL，Git 中不要添加大文件等）。
+
+pre-commit测试是 Travis-CI 中单元测试的一部分，不满足钩子的 PR 不能被提交到 PaddleOCR，首先安装并在当前目录运行它：
+
+```bash linenums="1"
+pip install pre-commit
+pre-commit install
+```
+
+ > 1. Paddle 使用 clang-format 来调整 C/C++ 源代码格式，请确保 `clang-format` 版本在 3.8 以上。
+ >
+ > 2. 通过pip install pre-commit和conda install -c conda-forge pre-commit安装的yapf稍有不同的，PaddleOCR 开发人员使用的是 `pip install pre-commit`。
+
+#### 3.2.5 修改与提交代码
+
+ 假设对PaddleOCR的 `README.md` 做了一些修改，可以通过 `git status` 查看改动的文件，然后使用 `git add` 添加改动文件。
+
+```bash linenums="1"
+git status # 查看改动文件
+git add README.md
+pre-commit
+```
+
+重复上述步骤，直到pre-commit格式检查不报错。如下所示。
+
+![img](./images/precommit_pass.png)
+
+提交修改，并写明修改内容（"your commit info"）
+
+```bash linenums="1"
+git commit -m "your commit info"
+```
+
+#### 3.2.6 Push到远程仓库
+
+使用push命令将修改的commit提交到 `远程仓库`
+
+```bash linenums="1"
+git push origin new_branch
+```
+
+#### 3.2.7 提交Pull Request
+
+打开自己的远程仓库界面，选择提交的分支。点击new pull request或contribute进入PR界面。选择本地分支和目标分支，如下图所示。在PR的描述说明中，填写该PR所完成的功能。接下来等待review，如果有需要修改的地方，参照上述步骤更新 origin 中的对应分支即可。
+
+![img](./images/pr.png)
+
+#### 3.2.8 签署CLA协议和通过单元测试
+
+- 签署CLA 在首次向PaddlePaddle提交Pull Request时，您需要您签署一次CLA(Contributor License Agreement)协议，以保证您的代码可以被合入，具体签署方式如下：
+
+  1. 请您查看PR中的Check部分，找到license/cla，并点击右侧detail，进入CLA网站
+
+  2. 点击CLA网站中的“Sign in with GitHub to agree”,点击完成后将会跳转回您的Pull Request页面
+
+#### 3.2.9 删除分支
+
+- 删除远程分支
+
+  在 PR 被 merge 进主仓库后，我们可以在 PR 的页面删除远程仓库的分支。
+
+  也可以使用 `git push origin :分支名` 删除远程分支，如：
+
+  ```bash linenums="1"
+  git push origin :new_branch
+  ```
+
+- 删除本地分支
+
+  ```bash linenums="1"
+  # 切换到dygraph分支，否则无法删除当前分支
+  git checkout dygraph
+
+  # 删除new_branch分支
+  git branch -D new_branch
+  ```
+
+#### 3.2.10 提交代码的一些约定
+
+为了使官方维护人员在评审代码时更好地专注于代码本身，请您每次提交代码时，遵守以下约定：
+
+1）请保证Travis-CI 中单元测试能顺利通过。如果没过，说明提交的代码存在问题，官方维护人员一般不做评审。
+
+2）提交Pull Request前：
+
+- 请注意commit的数量。
+
+  原因：如果仅仅修改一个文件但提交了十几个commit，每个commit只做了少量的修改，这会给评审人带来很大困扰。评审人需要逐一查看每个commit才能知道做了哪些修改，且不排除commit之间的修改存在相互覆盖的情况。
+
+  建议：每次提交时，保持尽量少的commit，可以通过git commit --amend补充上次的commit。对已经Push到远程仓库的多个commit，可以参考[squash commits after push](https://stackoverflow.com/questions/5667884/how-to-squash-commits-in-git-after-they-have-been-pushed)。
+
+- 请注意每个commit的名称：应能反映当前commit的内容，不能太随意。
+
+3）如果解决了某个Issue的问题，请在该Pull Request的第一个评论框中加上：fix #issue_number，这样当该Pull Request被合并后，会自动关闭对应的Issue。关键词包括：close, closes, closed, fix, fixes, fixed, resolve, resolves, resolved，请选择合适的词汇。详细可参考[Closing issues via commit messages](https://help.github.com/articles/closing-issues-via-commit-messages)。
+
+此外，在回复评审人意见时，请您遵守以下约定：
+
+1）官方维护人员的每一个review意见都希望得到回复，这样会更好地提升开源社区的贡献。
+
+- 对评审意见同意且按其修改完的，给个简单的Done即可；
+- 对评审意见不同意的，请给出您自己的反驳理由。
+
+2）如果评审意见比较多:
+
+- 请给出总体的修改情况。
+- 请采用`start a review`进行回复，而非直接回复的方式。原因是每个回复都会发送一封邮件，会造成邮件灾难。
diff --git a/docs/community/community_contribution.en.md b/docs/community/community_contribution.en.md
new file mode 100644
index 0000000..d115da0
--- /dev/null
+++ b/docs/community/community_contribution.en.md
@@ -0,0 +1,88 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+# Community Contribution
+
+Thank you for your long-term support and attention to PaddleOCR. Building a professional, harmonious, and mutually helpful open-source community with developers is the goal of PaddleOCR. This document showcases existing community contributions, explains various types of contributions, and describes new opportunities and processes, hoping to make the contribution process more efficient and the path clearer.
+
+PaddleOCR hopes to help every developer with a dream realize their ideas through the power of AI and enjoy the pleasure of creating value.
+
+<a href="https://github.com/PaddlePaddle/PaddleOCR/graphs/contributors">
+  <img src="https://contrib.rocks/image?repo=PaddlePaddle/PaddleOCR&max=400&columns=20" />
+</a>
+
+---
+
+## 1. Community Contributions
+
+### 1.1 New Features for PaddleOCR
+
+- Many thanks to [authorfu](https://github.com/authorfu) for contributing the Android demo ([#340](https://github.com/PaddlePaddle/PaddleOCR/pull/340)) and [xxlyu-2046](https://github.com/xxlyu-2046) for contributing the iOS demo code ([#325](https://github.com/PaddlePaddle/PaddleOCR/pull/325)).
+- Many thanks to [tangmq](https://gitee.com/tangmq) for adding Docker-based deployment services to PaddleOCR, supporting quick publishing of callable RESTful API services ([#507](https://github.com/PaddlePaddle/PaddleOCR/pull/507)).
+- Many thanks to [lijinhan](https://github.com/lijinhan) for adding Java SpringBoot integration with the OCR HubServing interface for OCR service deployment ([#1027](https://github.com/PaddlePaddle/PaddleOCR/pull/1027)).
+- Many thanks to [Evezerest](https://github.com/Evezerest), [ninetailskim](https://github.com/ninetailskim), [edencfc](https://github.com/edencfc), [BeyondYourself](https://github.com/BeyondYourself), and [1084667371](https://github.com/1084667371) for contributing the complete code for [PPOCRLabel](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.3/PPOCRLabel/README_ch.md).
+- Many thanks to [bupt906](https://github.com/bupt906) for contributing the MicroNet architecture code ([#5251](https://github.com/PaddlePaddle/PaddleOCR/pull/5251)) and the OneCycle learning rate strategy code ([#5252](https://github.com/PaddlePaddle/PaddleOCR/pull/5252)).
+
+### 1.2 Bug Fixes
+
+- Many thanks to [zhangxin](https://github.com/ZhangXinNan) ([Blog](https://blog.csdn.net/sdlypyzq)) for contributing a new visualization method, adding .gitignore, and fixing the issue of manually setting the PYTHONPATH environment variable ([#210](https://github.com/PaddlePaddle/PaddleOCR/pull/210)).
+- Many thanks to [lyl120117](https://github.com/lyl120117) for contributing the code for printing network structures ([#304](https://github.com/PaddlePaddle/PaddleOCR/pull/304)).
+- Many thanks to [BeyondYourself](https://github.com/BeyondYourself) for providing many excellent suggestions and simplifying some of PaddleOCR's code style ([so many commits](https://github.com/PaddlePaddle/PaddleOCR/commits?author=BeyondYourself)).
+
+### 1.3 Documentation Improvements and Translations
+
+- Many thanks to **[RangeKing](https://github.com/RangeKing), [HustBestCat](https://github.com/HustBestCat), [v3fc](https://github.com/v3fc), [1084667371](https://github.com/1084667371)** for contributing the translation of the [English version of "Erta OCR" notebook e-book](https://github.com/PaddleOCR-Community/Dive-into-OCR/tree/main/notebook_en).
+- Many thanks to [thunderstudying](https://github.com/thunderstudying), [RangeKing](https://github.com/RangeKing), [livingbody](https://github.com/livingbody), [WZMIAOMIAO](https://github.com/WZMIAOMIAO), and [haigang1975](https://github.com/haigang1975) for supplementing multiple English markdown documents.
+- Many thanks to **[fanruinet](https://github.com/fanruinet)** for polishing and fixing 35 English documents ([#5205](https://github.com/PaddlePaddle/PaddleOCR/pull/5205)).
+- Many thanks to [Khanh Tran](https://github.com/xxxpsyduck) and [Karl Horky](https://github.com/karlhorky) for contributing modifications to English documentation.
+
+### 1.4 Multilingual Corpora
+
+- Many thanks to [xiangyubo](https://github.com/xiangyubo) for contributing a handwritten Chinese OCR dataset ([#321](https://github.com/PaddlePaddle/PaddleOCR/pull/321)).
+- Many thanks to [Mejans](https://github.com/Mejans) for adding a new Occitan language dictionary and corpus to PaddleOCR ([#954](https://github.com/PaddlePaddle/PaddleOCR/pull/954)).
+
+## 2. Contribution Guidelines
+
+### 2.1 New Features
+
+PaddleOCR warmly welcomes community contributions of various services, deployment examples, and software applications built around PaddleOCR. Certified community contributions will be added to the community contribution table above, providing increased visibility for developers and bringing honor to PaddleOCR:
+
+- Project format: Officially certified community project code should follow good standards and structure, and should also include a detailed README.md explaining how to use the project. Adding `paddleocr` to your requirements.txt file will automatically include your project in PaddleOCR's "used by" list.
+
+- Merge process: If it is an update or upgrade to an existing PaddleOCR tool, it will be merged into the main repo. If it extends new functionality for PaddleOCR, please contact the official team first to confirm whether the project should be merged into the main repo. *Even if the new functionality is not merged into the main repo, we will still increase visibility for your personal project as a community contribution.*
+
+### 2.2 Code Optimization
+
+If you encounter code bugs or unexpected behavior while using PaddleOCR, you can contribute your fixes:
+
+- For Python code standards, please refer to [Appendix 1: Python Code Specification](./code_and_doc.md#appendix-1python-code-specification).
+
+- Please double-check before submitting code that no new bugs are introduced, and describe the optimization points in the PR. If the PR resolves a specific issue, please link to that issue in the PR. All PRs should follow the conventions in [3.2.10 Some Conventions For Submitting Code](./code_and_doc.md#3210-some-conventions-for-submitting-code).
+
+- Please refer to [Appendix 3: Pull Request Description](./code_and_doc.md#appendix-3-pull-request-description) before submitting. If you are not familiar with the git submission process, you can also refer to section 3.2 of Appendix 3.
+
+### 2.3 Documentation Improvements
+
+If you encounter unclear descriptions, missing content, or broken links while using PaddleOCR, you can contribute your fixes. For documentation writing standards, please refer to [Appendix 2: Document Specification](./code_and_doc.md#appendix-2-document-specification).
+
+## 3. More Contribution Opportunities
+
+We strongly encourage developers to use PaddleOCR to implement their own ideas. We also list some valuable extension directions that have been analyzed and are collected in the community project regular competition.
+
+## 4. Contact Us
+
+We warmly welcome developers to contact us before contributing code, documentation, corpora, or other content to PaddleOCR. This can greatly reduce communication costs during the PR process. Additionally, if you find certain ideas difficult to implement individually, we can recruit like-minded developers through SIG (Special Interest Group) to collaborate on the project. Projects contributed through the SIG channel will receive in-depth R&D support and operational resources (such as public account promotion, live courses, etc.).
+
+Our recommended contribution process is:
+
+- Add the `[third-party]` tag to the GitHub issue title, describe the problem encountered (and your proposed solution) or the functionality you want to extend, and wait for a response from the duty personnel. For example: `[third-party] Contributing an iOS example for PaddleOCR`.
+- After confirming the technical solution or verifying the bug/optimization through communication with us, proceed with adding the new functionality or making the relevant changes. Code and documentation should follow the relevant standards.
+- Link the PR to the above issue and wait for review.
+
+## 5. Acknowledgments and Follow-up
+
+- After the code is merged, information will be updated in the first section of this document. By default, links point to the GitHub username and homepage. If you need to change the homepage, you can also contact us.
+- For major new features, announcements will be made in user groups to share the honor of open-source community contributions.
+- **If you have a project based on PaddleOCR that is not listed above, please follow the steps in "4. Contact Us" to reach out to us.**
diff --git a/docs/community/community_contribution.md b/docs/community/community_contribution.md
new file mode 100644
index 0000000..522eb69
--- /dev/null
+++ b/docs/community/community_contribution.md
@@ -0,0 +1,88 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+# 社区贡献
+
+感谢大家长久以来对PaddleOCR的支持和关注，与广大开发者共同构建一个专业、和谐、相互帮助的开源社区是PaddleOCR的目标。本文档展示了已有的社区贡献、对于各类贡献说明、新的机会与流程，希望贡献流程更加高效、路径更加清晰。
+
+PaddleOCR希望可以通过AI的力量助力任何一位有梦想的开发者实现自己的想法，享受创造价值带来的愉悦。
+
+<a href="https://github.com/PaddlePaddle/PaddleOCR/graphs/contributors">
+  <img src="https://contrib.rocks/image?repo=PaddlePaddle/PaddleOCR&max=400&columns=20" />
+</a>
+
+---
+
+## 1. 社区贡献
+
+### 1.1 为PaddleOCR新增功能
+
+- 非常感谢 [authorfu](https://github.com/authorfu) 贡献Android([#340](https://github.com/PaddlePaddle/PaddleOCR/pull/340))和[xxlyu-2046](https://github.com/xxlyu-2046) 贡献IOS的demo代码([#325](https://github.com/PaddlePaddle/PaddleOCR/pull/325))
+- 非常感谢 [tangmq](https://gitee.com/tangmq) 给PaddleOCR增加Docker化部署服务，支持快速发布可调用的Restful API服务([#507](https://github.com/PaddlePaddle/PaddleOCR/pull/507))。
+- 非常感谢 [lijinhan](https://github.com/lijinhan) 给PaddleOCR增加java SpringBoot 调用OCR Hubserving接口完成对OCR服务化部署的使用([#1027](https://github.com/PaddlePaddle/PaddleOCR/pull/1027))。
+- 非常感谢 [Evezerest](https://github.com/Evezerest)， [ninetailskim](https://github.com/ninetailskim)， [edencfc](https://github.com/edencfc)， [BeyondYourself](https://github.com/BeyondYourself)， [1084667371](https://github.com/1084667371) 贡献了[PPOCRLabel](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.3/PPOCRLabel/README_ch.md) 的完整代码。
+- 非常感谢 [bupt906](https://github.com/bupt906) 贡献MicroNet结构代码([#5251](https://github.com/PaddlePaddle/PaddleOCR/pull/5251))和贡献OneCycle学习率策略代码([#5252](https://github.com/PaddlePaddle/PaddleOCR/pull/5252))
+
+### 1.2 代码修复
+
+- 非常感谢 [zhangxin](https://github.com/ZhangXinNan)([Blog](https://blog.csdn.net/sdlypyzq)) 贡献新的可视化方式、添加.gitgnore、处理手动设置PYTHONPATH环境变量的问题([#210](https://github.com/PaddlePaddle/PaddleOCR/pull/210))。
+- 非常感谢 [lyl120117](https://github.com/lyl120117) 贡献打印网络结构的代码([#304](https://github.com/PaddlePaddle/PaddleOCR/pull/304))。
+- 非常感谢 [BeyondYourself](https://github.com/BeyondYourself) 给PaddleOCR提了很多非常棒的建议，并简化了PaddleOCR的部分代码风格([so many commits)](https://github.com/PaddlePaddle/PaddleOCR/commits?author=BeyondYourself)。
+
+### 1.3 文档优化与翻译
+
+- 非常感谢 **[RangeKing](https://github.com/RangeKing)，[HustBestCat](https://github.com/HustBestCat)，[v3fc](https://github.com/v3fc)，[1084667371](https://github.com/1084667371)** 贡献翻译《动手学OCR》notebook[电子书英文版](https://github.com/PaddleOCR-Community/Dive-into-OCR/tree/main/notebook_en)。
+- 非常感谢 [thunderstudying](https://github.com/thunderstudying)，[RangeKing](https://github.com/RangeKing)，[livingbody](https://github.com/livingbody)， [WZMIAOMIAO](https://github.com/WZMIAOMIAO)，[haigang1975](https://github.com/haigang1975) 补充多个英文markdown文档。
+- 非常感谢 **[fanruinet](https://github.com/fanruinet)** 润色和修复35篇英文文档([#5205](https://github.com/PaddlePaddle/PaddleOCR/pull/5205))。
+- 非常感谢 [Khanh Tran](https://github.com/xxxpsyduck) 和 [Karl Horky](https://github.com/karlhorky) 贡献修改英文文档。
+
+### 1.4 多语言语料
+
+- 非常感谢 [xiangyubo](https://github.com/xiangyubo) 贡献手写中文OCR数据集([#321](https://github.com/PaddlePaddle/PaddleOCR/pull/321))。
+- 非常感谢 [Mejans](https://github.com/Mejans) 给PaddleOCR增加新语言奥克西坦语Occitan的字典和语料([#954](https://github.com/PaddlePaddle/PaddleOCR/pull/954))。
+
+## 2. 贡献说明
+
+### 2.1 新增功能类
+
+PaddleOCR非常欢迎社区贡献以PaddleOCR为核心的各种服务、部署实例与软件应用，经过认证的社区贡献会被添加在上述社区贡献表中，为广大开发者增加曝光，也是PaddleOCR的荣耀，其中：
+
+- 项目形式：官方社区认证的项目代码应有良好的规范和结构，同时，还应配备一个详细的README.md，说明项目的使用方法。通过在requirements.txt文件中增加一行 `paddleocr` 可以自动收录到PaddleOCR的usedby中。
+
+- 合入方式：如果是对PaddleOCR现有工具的更新升级，则会合入主repo。如果为PaddleOCR拓展了新功能，请先与官方人员联系，确认项目是否合入主repo，*即使新功能未合入主repo，我们同样也会以社区贡献的方式为您的个人项目增加曝光。*
+
+### 2.2 代码优化
+
+如果您在使用PaddleOCR时遇到了代码bug、功能不符合预期等问题，可以为PaddleOCR贡献您的修改，其中：
+
+- Python代码规范可参考[附录1：Python代码规范](./code_and_doc.md#1python)。
+
+- 提交代码前请再三确认不会引入新的bug，并在PR中描述优化点。如果该PR解决了某个issue，请在PR中连接到该issue。所有的PR都应该遵守附录3中的[3.2.10 提交代码的一些约定。](./code_and_doc.md#3pull-request)
+
+- 请在提交之前参考下方的[附录3：Pull Request说明](./code_and_doc.md#3pull-request)。如果您对git的提交流程不熟悉，同样可以参考附录3的3.2节。
+
+### 2.3 文档优化
+
+如果您在使用PaddleOCR时遇到了文档表述不清楚、描述缺失、链接失效等问题，可以为PaddleOCR贡献您的修改。文档书写规范请参考[附录2：文档规范](./code_and_doc.md#2)。
+
+## 3. 更多贡献机会
+
+我们非常鼓励开发者使用PaddleOCR实现自己的想法，同时我们也列出一些经过分析后认为有价值的拓展方向，整体收集在社区项目常规赛中。
+
+## 4. 联系我们
+
+我们非常欢迎广大开发者在有意向为PaddleOCR贡献代码、文档、语料等内容前与我们联系，这样可以大大降低PR过程中的沟通成本。同时，如果您觉得某些想法个人难以实现，我们也可以通过SIG的形式定向为项目招募志同道合的开发者一起共建。通过SIG渠道贡献的项目将会获得深层次的研发支持与运营资源（如公众号宣传、直播课等）。
+
+我们推荐的贡献流程是：
+
+- 通过在github issue的题目中增加  `【third-party】` 标记，说明遇到的问题（以及解决的思路）或想拓展的功能，等待值班人员回复。例如 `【third-party】为PaddleOCR贡献IOS示例`
+- 与我们沟通确认技术方案或bug、优化点准确无误后进行功能新增或相应的修改，代码与文档遵循相关规范。
+- PR链接到上述issue，等待review。
+
+## 5. 致谢与后续
+
+- 合入代码之后会在本文档第一节中更新信息，默认链接为github名字及主页，如果有需要更换主页，也可以联系我们。
+- 新增重要功能类，会在用户群广而告之，享受开源社区荣誉时刻。
+- **如果您有基于PaddleOCR的项目，但未出现在上述列表中，请按照 `4. 联系我们` 的步骤与我们联系。**
diff --git a/docs/community/images/banner.png b/docs/community/images/banner.png
new file mode 100644
index 0000000..d72b997
Binary files /dev/null and b/docs/community/images/banner.png differ
diff --git a/docs/community/images/pr.png b/docs/community/images/pr.png
new file mode 100644
index 0000000..3d0d15f
Binary files /dev/null and b/docs/community/images/pr.png differ
diff --git a/docs/community/images/precommit_pass.png b/docs/community/images/precommit_pass.png
new file mode 100644
index 0000000..067fb75
Binary files /dev/null and b/docs/community/images/precommit_pass.png differ
diff --git a/docs/data_anno_synth/data_annotation.en.md b/docs/data_anno_synth/data_annotation.en.md
new file mode 100644
index 0000000..f820793
--- /dev/null
+++ b/docs/data_anno_synth/data_annotation.en.md
@@ -0,0 +1,36 @@
+---
+comments: true
+---
+
+# Other Data Annotation Tools
+
+There are the commonly used data annotation tools, which will be continuously updated. Welcome to contribute tools~
+
+## 1. PPOCRLabel
+
+- Tool description: PaddleOCR semi-automatic annotation tool
+- Tool address: <https://github.com/PFCCLab/PPOCRLabel/blob/main/README.md>
+
+## 2. labelImg
+
+- Tool description: Rectangular label
+- Tool address:  <https://github.com/tzutalin/labelImg>
+- Sketch diagram:
+
+    ![labelimg](./images/labelimg.jpg)
+
+## 3. roLabelImg
+
+- Tool description: Label tool rewritten based on labelImg, supporting rotating rectangular label
+- Tool address:   <https://github.com/cgvict/roLabelImg>
+- Sketch diagram:
+
+    ![roLabelImg](./images/roLabelImg.png)
+
+## 4. labelme
+
+- Tool description: Support four points, polygons, circles and other labels
+- Tool address:   <https://github.com/wkentaro/labelme>
+- Sketch diagram:
+
+    ![labelme](./images/labelme.jpg)
diff --git a/docs/data_anno_synth/data_annotation.md b/docs/data_anno_synth/data_annotation.md
new file mode 100644
index 0000000..6207c52
--- /dev/null
+++ b/docs/data_anno_synth/data_annotation.md
@@ -0,0 +1,40 @@
+---
+comments: true
+---
+
+# 其它数据标注工具
+
+这里整理了常用的数据标注工具，持续更新中，欢迎各位小伙伴贡献工具～
+
+## 1. PPOCRLabel
+
+- 工具描述：PaddleOCR 半自动标注工具
+- 工具地址：<https://github.com/PFCCLab/PPOCRLabel/blob/main/README_ch.md>
+
+## 2. labelImg
+
+- 工具描述：矩形标注
+- 工具地址：<https://github.com/tzutalin/labelImg>
+- 示意图：
+    ![](./images/labelimg.jpg)
+
+## 3. roLabelImg
+
+- 工具描述：基于labelImg重写的标注工具，支持旋转矩形标注
+- 工具地址：<https://github.com/cgvict/roLabelImg>
+- 示意图：
+    ![](./images/roLabelImg.png)
+
+## 4. labelme
+
+- 工具描述：支持四点、多边形、圆形等多种标注
+- 工具地址：<https://github.com/wkentaro/labelme>
+- 示意图：
+    ![](./images/labelme.jpg)
+
+## 5. Vott
+
+- 工具描述：支持矩形,多边形等图片标注.支持视频标注.方便使用的快捷键以及比较好看的界面.同时支持导出多种标签格式.
+- 工具地址：<https://github.com/microsoft/VoTT>
+- 示意图：
+    ![](./images/VoTT.jpg)
diff --git a/docs/data_anno_synth/data_synthesis.en.md b/docs/data_anno_synth/data_synthesis.en.md
new file mode 100644
index 0000000..f329d4d
--- /dev/null
+++ b/docs/data_anno_synth/data_synthesis.en.md
@@ -0,0 +1,19 @@
+---
+comments: true
+---
+
+
+# Data Synthesis Tools
+
+In addition to open source data, users can also use synthesis tools to synthesize data.
+There are the commonly used data synthesis tools, which will be continuously updated. Welcome to contribute tools~
+
+* [Style-Text](https://github.com/PFCCLab/StyleText/blob/main/README.md)
+* [text_renderer (old version)](https://github.com/Sanster/text_renderer)
+* [text_renderer (new version)](https://github.com/oh-my-ocr/text_renderer)
+* [SynthText](https://github.com/ankush-me/SynthText)
+* [SynthText_Chinese_version](https://github.com/Snowty/SynthText_Chinese)
+* [TextRecognitionDataGenerator](https://github.com/Belval/TextRecognitionDataGenerator)
+* [SynthText3D](https://github.com/MhLiao/SynthText3D)
+* [UnrealText](https://github.com/Jyouhou/UnrealText/)
+* [SynthTIGER](https://github.com/clovaai/synthtiger)
diff --git a/docs/data_anno_synth/data_synthesis.md b/docs/data_anno_synth/data_synthesis.md
new file mode 100644
index 0000000..39563e2
--- /dev/null
+++ b/docs/data_anno_synth/data_synthesis.md
@@ -0,0 +1,17 @@
+---
+comments: true
+---
+
+# 数据合成工具
+
+除了开源数据，用户还可使用合成工具自行合成。这里整理了常用的数据合成工具，持续更新中，欢迎各位小伙伴贡献工具～
+
+- [Style-Text](https://github.com/PFCCLab/StyleText/blob/main/README_ch.md)
+- [text_renderer (old version)](https://github.com/Sanster/text_renderer)
+- [text_renderer (new version)](https://github.com/oh-my-ocr/text_renderer)
+- [SynthText](https://github.com/ankush-me/SynthText)
+- [SynthText_Chinese_version](https://github.com/Snowty/SynthText_Chinese)
+- [TextRecognitionDataGenerator](https://github.com/Belval/TextRecognitionDataGenerator)
+- [SynthText3D](https://github.com/MhLiao/SynthText3D)
+- [UnrealText](https://github.com/Jyouhou/UnrealText/)
+- [SynthTIGER](https://github.com/clovaai/synthtiger)
diff --git a/docs/data_anno_synth/images/VoTT.jpg b/docs/data_anno_synth/images/VoTT.jpg
new file mode 100644
index 0000000..7c5c27b
Binary files /dev/null and b/docs/data_anno_synth/images/VoTT.jpg differ
diff --git a/docs/data_anno_synth/images/labelimg.jpg b/docs/data_anno_synth/images/labelimg.jpg
new file mode 100644
index 0000000..8d58a44
Binary files /dev/null and b/docs/data_anno_synth/images/labelimg.jpg differ
diff --git a/docs/data_anno_synth/images/labelme.jpg b/docs/data_anno_synth/images/labelme.jpg
new file mode 100644
index 0000000..ce44e50
Binary files /dev/null and b/docs/data_anno_synth/images/labelme.jpg differ
diff --git a/docs/data_anno_synth/images/roLabelImg.png b/docs/data_anno_synth/images/roLabelImg.png
new file mode 100644
index 0000000..3b02a32
Binary files /dev/null and b/docs/data_anno_synth/images/roLabelImg.png differ
diff --git a/docs/data_anno_synth/x_anylabeling.en.md b/docs/data_anno_synth/x_anylabeling.en.md
new file mode 100644
index 0000000..3dedf54
--- /dev/null
+++ b/docs/data_anno_synth/x_anylabeling.en.md
@@ -0,0 +1,164 @@
+---
+comments: true
+---
+
+# Document Parsing with X-AnyLabeling
+
+## Introduction
+
+The document parsing capabilities of the **PaddleOCR-VL** series are now integrated with the [X-AnyLabeling](https://github.com/CVHub520/X-AnyLabeling) annotation platform.
+
+X-AnyLabeling is an industrial-grade all-in-one intelligent annotation platform from CVHub that unifies training, inference, and annotation. Its `PaddleOCR` panel lets developers run layout parsing, text recognition, formula recognition, table recognition, and seal recognition on images and PDFs with PaddleOCR-VL series models, then review, edit, copy, and export the results.
+
+PaddleOCR-VL is exposed in X-AnyLabeling through two integration modes:
+
+- **Official API (recommended):** call the PaddleOCR official API directly. Best for quickly validating the model, lightweight experimentation, and low-overhead development — no inference service to deploy.
+- **Local deployment:** run inference yourself via [X-AnyLabeling-Server](https://github.com/CVHub520/X-AnyLabeling-Server). Best for private deployments, sensitive data, and sustained annotation workloads.
+
+## 1. Install X-AnyLabeling
+
+Download the prebuilt binary for your platform from the official release page:
+
+- Releases: <https://github.com/CVHub520/X-AnyLabeling/releases>
+
+Alternatively, install via pip:
+
+```bash
+pip install x-anylabeling
+```
+
+After launch, open the PaddleOCR panel from the `PaddleOCR` icon in the left toolbar, or with the shortcut `Ctrl+4`.
+
+![Selecting PaddleOCR from the left toolbar](https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/x_anylabeling/open_panel.png)
+
+Once opened, the initial PaddleOCR panel looks like this:
+
+![Initial PaddleOCR panel](https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/x_anylabeling/panel.png)
+
+## 2. Configure the PaddleOCR Official API
+
+The X-AnyLabeling client supports the PaddleOCR official API out of the box. The first time you open the PaddleOCR panel without API credentials configured, a `PPOCR API Settings` dialog appears automatically. You can reopen it later via the gear button at the top of the right-hand result panel.
+
+![PPOCR API Settings dialog](https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/x_anylabeling/api_settings.png)
+
+To obtain the `API_KEY`:
+
+1. Visit the PaddleOCR website: <https://www.paddleocr.com>
+2. Click **API** in the top-right corner and select **PaddleOCR-VL-1.5**.
+3. Open the example code and copy the `TOKEN` (access token).
+4. Return to X-AnyLabeling, paste it into the `API_KEY` field of `PPOCR API Settings`, and confirm.
+
+The `API_KEY` is used for request authentication, and the official service offers a free quota of tens of thousands of pages per day. The configuration is persisted locally:
+
+```text
+${workspace}/xanylabeling_data/paddleocr/api_settings.json
+```
+
+By default, `${workspace}` is the user home directory `~`; if X-AnyLabeling is launched with `--work-dir`, that directory is used instead.
+
+The right-hand `Parsing Model` dropdown currently supports these official API options:
+
+- `PaddleOCR-VL-1.5 (API)`
+- `PaddleOCR-VL (API)`
+
+## 3. Import Documents to Parse
+
+In the PaddleOCR panel, click `+ New Parsing` at the top of the left panel to import files. Imported files are copied into the local PaddleOCR working directory and added to the parsing queue automatically.
+
+Supported file types:
+
+| Type | Extensions |
+| :--- | :--- |
+| PDF documents | `.pdf` |
+| Images | `.bmp`, `.cif`, `.gif`, `.jpeg`, `.jpg`, `.png`, `.tif`, `.tiff`, `.webp` |
+
+Typical real-world inputs include textbook pages, paper screenshots, receipts, scanned contracts, tabular documents, and government/enterprise materials.
+
+## 4. Choose a Parsing Model
+
+The `Parsing Model` dropdown on the right lets you switch between versions of the PaddleOCR-VL series:
+
+- `PaddleOCR-VL-1.5 (API)` (recommended): achieves 94.5% accuracy on OmniDocBench v1.5, supports polygon localization for greater robustness in scanned, tilted, curved, screen-captured, and complex-lighting scenarios, and adds seal recognition plus text detection and recognition capabilities.
+- `PaddleOCR-VL (API)`: the initial version.
+
+![Parsing Model dropdown switching between PaddleOCR-VL and PaddleOCR-VL-1.5](https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/x_anylabeling/model_switch.png)
+
+## 5. Run Document Parsing
+
+Once a model is selected, X-AnyLabeling starts parsing automatically. The model recognizes and structurizes text, formulas, tables, charts, seals, and other elements.
+
+The colored dot at the lower-left corner of each left-side file item indicates parsing status:
+
+- Blue: queued or parsing
+- Green: finished
+- Red: failed
+
+## 6. Review Results Against the Source
+
+When parsing finishes, X-AnyLabeling shows a side-by-side view: the original document on the left, the model's parsed result on the right. You can review the text, formulas, table structure, chart information, and seal results block by block:
+
+![Document Parsing side-by-side view](https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/x_anylabeling/document_parsing.png)
+
+- Click any block on either side to highlight the matching block on the other side.
+- Double-click a block on the right, or click its `Correct` button, to enter edit mode.
+- Hover over a block in the source preview to reveal a `Copy` button for that block.
+- Manual corrections are recorded in the JSON as edited blocks; use the re-parse button to refetch model results.
+
+Dedicated editors are provided for different content types:
+
+| Editor | Used for |
+| :--- | :--- |
+| Rich text editor | Plain text, titles, footers, seals, and other non-table/non-formula blocks |
+| LaTeX formula editor | `display_formula`, `formula`, `formula_number`, `algorithm` — edits source with live preview |
+| Table editor | `table` blocks and content recognized as table structure — supports cell editing, row/column add/remove |
+
+## 7. Switch to the JSON View
+
+In addition to `Document Parsing`, X-AnyLabeling also offers a `JSON` view:
+
+- `Document Parsing` view: card-based display of layout blocks, text, formulas, tables, and images — best for human review.
+- `JSON` view: the full structured result — best for data processing, training-sample construction, evaluation aggregation, and integration with downstream systems.
+
+![JSON view](https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/x_anylabeling/json_view.png)
+
+## 8. Export Results
+
+After reviewing and correcting the results, you can export the annotations for downstream use:
+
+- Building OCR training datasets
+- Running model evaluation and error analysis
+- Curating internal document parsing data
+- Feeding business systems for automation
+
+In the `Document Parsing` view, the download button on the right toolbar exports a ZIP containing Markdown and related resources; in the `JSON` view, it exports the full JSON.
+
+## Local Data Layout
+
+X-AnyLabeling's PaddleOCR panel stores imported files and parsing results in a local working directory:
+
+```text
+${workspace}/xanylabeling_data/paddleocr/
+├── api_settings.json
+├── ui_state.json
+├── files/
+│   ├── example.pdf
+│   ├── image.png
+│   ├── __PDF_example/
+│   │   ├── page_001.png
+│   │   └── page_002.png
+│   └── __BLOCK_IMAGES_image.png/
+│       └── page_001_block_0001.png
+└── jsons/
+    ├── example.pdf.json
+    └── image.png.json
+```
+
+Deleting a file from the left-hand list also removes its source file, local JSON, PDF preview pages, and block crops.
+
+## References
+
+- X-AnyLabeling repository: <https://github.com/CVHub520/X-AnyLabeling>
+- X-AnyLabeling PaddleOCR panel documentation: <https://github.com/CVHub520/X-AnyLabeling/blob/main/docs/zh_cn/paddle_ocr.md>
+- X-AnyLabeling-Server (local inference): <https://github.com/CVHub520/X-AnyLabeling-Server>
+- PaddleOCR website: <https://www.paddleocr.com>
+- PaddleOCR-VL tutorial: [PaddleOCR-VL Tutorial](../version3.x/pipeline_usage/PaddleOCR-VL.md)
diff --git a/docs/data_anno_synth/x_anylabeling.md b/docs/data_anno_synth/x_anylabeling.md
new file mode 100644
index 0000000..aa2bff2
--- /dev/null
+++ b/docs/data_anno_synth/x_anylabeling.md
@@ -0,0 +1,164 @@
+---
+comments: true
+---
+
+# X-AnyLabeling 文档解析
+
+## 简介
+
+**PaddleOCR-VL** 系列模型的文档解析能力已集成至 [X-AnyLabeling](https://github.com/CVHub520/X-AnyLabeling) 标注平台。
+
+X-AnyLabeling 是由 CVHub 推出的工业级一体化智能标注平台，打通训练、推理、标注全链路。借助其 `PaddleOCR` 面板，开发者可以基于 PaddleOCR-VL 系列模型对图片和 PDF 进行版面解析、文字识别、公式识别、表格识别和印章识别，并在识别完成后对结果进行复核、编辑、复制和导出。
+
+PaddleOCR-VL 在 X-AnyLabeling 中提供两种接入方式：
+
+- **官方 API（推荐）**：直接调用 PaddleOCR 官方 API，适合快速验证模型效果、低成本体验和轻量开发场景，无需额外部署推理服务。
+- **本地部署**：通过 [X-AnyLabeling-Server](https://github.com/CVHub520/X-AnyLabeling-Server) 自行部署推理服务，适合私有化部署、敏感数据处理和持续性标注任务。
+
+## 1. 安装 X-AnyLabeling
+
+推荐从官方 Release 下载对应平台的预编译包：
+
+- Release 地址：<https://github.com/CVHub520/X-AnyLabeling/releases>
+
+也可以通过 pip 安装：
+
+```bash
+pip install x-anylabeling
+```
+
+启动后可以通过左侧工具栏中的 `PaddleOCR` 图标，或使用快捷键 `Ctrl+4` 打开 PaddleOCR 面板。
+
+![在左侧工具栏选择 PaddleOCR](https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/x_anylabeling/open_panel.png)
+
+打开后即可看到 PaddleOCR 面板的初始界面：
+
+![PaddleOCR 面板初始界面](https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/x_anylabeling/panel.png)
+
+## 2. 配置 PaddleOCR 官方 API
+
+X-AnyLabeling 客户端默认支持 PaddleOCR 官方 API。首次打开 PaddleOCR 面板且尚未配置 API 信息时，界面会自动弹出 `PPOCR API Settings` 配置窗口；后续如需修改配置，也可以点击右侧结果面板顶部的齿轮按钮重新打开。
+
+![PPOCR API Settings 配置窗口](https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/x_anylabeling/api_settings.png)
+
+获取 `API_KEY` 的方式：
+
+1. 访问 PaddleOCR 官方网站：<https://www.paddleocr.com>
+2. 点击右上角 **API**，选择 **PaddleOCR-VL-1.5**。
+3. 打开示例代码，复制其中的 `TOKEN`（访问令牌）。
+4. 回到 X-AnyLabeling 的 `PPOCR API Settings`，将其粘贴到 `API_KEY` 字段并确认即可。
+
+`API_KEY` 用于接口鉴权，支持申请每天免费解析数万文档页数。配置会保存在本地：
+
+```text
+${workspace}/xanylabeling_data/paddleocr/api_settings.json
+```
+
+默认情况下，`${workspace}` 为用户主目录 `~`；如果启动 X-AnyLabeling 时传入了 `--work-dir`，则以该目录为准。
+
+右侧 `Parsing Model` 下拉框当前支持以下官方 API 选项：
+
+- `PaddleOCR-VL-1.5 (API)`
+- `PaddleOCR-VL (API)`
+
+## 3. 导入待解析文档
+
+打开 PaddleOCR 面板后，点击左侧面板顶部的 `+ New Parsing` 导入文件。导入后文件会自动复制到本地 PaddleOCR 工作目录，并自动加入解析队列。
+
+当前 PaddleOCR 面板支持的文件类型：
+
+| 类型 | 后缀 |
+| :--- | :--- |
+| PDF 文档 | `.pdf` |
+| 图片 | `.bmp`, `.cif`, `.gif`, `.jpeg`, `.jpg`, `.png`, `.tif`, `.tiff`, `.webp` |
+
+适用的真实场景包括教材页面、论文截图、票据图片、合同扫描件、表格文档、政企材料等。
+
+## 4. 选择解析模型
+
+在右侧 `Parsing Model` 下拉框中可在 PaddleOCR-VL 系列的不同版本之间切换：
+
+- `PaddleOCR-VL-1.5 (API)`（推荐）：在 OmniDocBench v1.5 上达到 94.5% 精度，支持异形框定位，对扫描、倾斜、弯折、屏幕拍摄及复杂光照等场景具备更强鲁棒性，并新增印章识别与文本检测识别能力。
+- `PaddleOCR-VL (API)`：初代版本。
+
+![Parsing Model 切换 PaddleOCR-VL 与 PaddleOCR-VL-1.5](https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/x_anylabeling/model_switch.png)
+
+## 5. 执行 Document Parsing
+
+选择模型后，X-AnyLabeling 会自动开始解析。模型会对文档中的文本、公式、表格、图表、印章等内容进行识别和结构化处理。
+
+左侧文件项左下角的彩色点表示解析状态：
+
+- 蓝色：待解析或正在解析
+- 绿色：解析完成
+- 红色：解析失败
+
+## 6. 对照原文档复核解析结果
+
+解析完成后，X-AnyLabeling 会呈现左右对照视图：左侧为原始文档，右侧为模型的解析结果。开发者可以逐段检查文字内容、公式表达、表格结构、图表信息和印章结果是否准确，并通过以下操作进行复核：
+
+![Document Parsing 左右对照视图](https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/x_anylabeling/document_parsing.png)
+
+- 单击左侧预览区或右侧结果区中的任意块，可以在两侧快速匹配并高亮对应内容。
+- 双击右侧识别结果中的某个块，或点击该块的 `纠正` 按钮，进入编辑状态。
+- 鼠标悬停在源文件预览区的块上时，可以直接点击浮出的 `复制` 按钮复制该块内容。
+- 对识别结果做人工修正后，JSON 中会记录已编辑块；如需重新获取模型结果，可使用右侧重解析按钮。
+
+针对不同类型的内容，X-AnyLabeling 提供专用编辑器：
+
+| 编辑器 | 适用场景 |
+| :--- | :--- |
+| 富文本编辑器 | 普通文本、标题、页脚、印章等非表格/公式内容 |
+| LaTeX 公式编辑器 | `display_formula`, `formula`, `formula_number`, `algorithm`，支持源码编辑与实时预览 |
+| 表格编辑器 | `table` 或被识别为表格结构的内容，支持单元格编辑、增删行列等 |
+
+## 7. 切换 JSON 视图查看结果
+
+除了 `Document Parsing` 视图外，X-AnyLabeling 还支持 `JSON` 视图：
+
+- `Document Parsing` 视图：以卡片形式展示版面块、文本、公式、表格、图片，适合人工阅读和结果复核。
+- `JSON` 视图：查看完整的结构化结果，适合开发者进行数据处理、训练样本构建、评测结果整理和业务系统集成。
+
+![JSON 视图](https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/x_anylabeling/json_view.png)
+
+## 8. 导出结果
+
+完成复核和修正后，可以将标注结果导出，用于后续任务，典型用途包括：
+
+- 构建 OCR 训练数据集
+- 进行模型评测和误差分析
+- 沉淀企业内部文档解析数据
+- 接入业务系统进行自动化处理
+
+在 `Document Parsing` 视图下，右侧工具栏的下载按钮会下载包含 Markdown 与相关资源的 ZIP；在 `JSON` 视图下则下载完整 JSON。
+
+## 数据目录结构
+
+X-AnyLabeling PaddleOCR 面板会将导入文件和解析结果保存在本地工作目录中：
+
+```text
+${workspace}/xanylabeling_data/paddleocr/
+├── api_settings.json
+├── ui_state.json
+├── files/
+│   ├── example.pdf
+│   ├── image.png
+│   ├── __PDF_example/
+│   │   ├── page_001.png
+│   │   └── page_002.png
+│   └── __BLOCK_IMAGES_image.png/
+│       └── page_001_block_0001.png
+└── jsons/
+    ├── example.pdf.json
+    └── image.png.json
+```
+
+删除左侧文件项时，会同时删除源文件、本地 JSON、PDF 预览页和 block 裁剪图。
+
+## 参考链接
+
+- X-AnyLabeling 项目主页：<https://github.com/CVHub520/X-AnyLabeling>
+- X-AnyLabeling PaddleOCR 面板文档：<https://github.com/CVHub520/X-AnyLabeling/blob/main/docs/zh_cn/paddle_ocr.md>
+- X-AnyLabeling-Server（本地推理服务）：<https://github.com/CVHub520/X-AnyLabeling-Server>
+- PaddleOCR 官方网站：<https://www.paddleocr.com>
+- PaddleOCR-VL 使用教程：[PaddleOCR-VL 使用教程](../version3.x/pipeline_usage/PaddleOCR-VL.md)
diff --git a/docs/datasets/datasets.en.md b/docs/datasets/datasets.en.md
new file mode 100644
index 0000000..90701eb
--- /dev/null
+++ b/docs/datasets/datasets.en.md
@@ -0,0 +1,93 @@
+---
+comments: true
+---
+
+This is a collection of commonly used Chinese datasets, which is being updated continuously. You are welcome to contribute to this list～
+
+In addition to opensource data, users can also use synthesis tools to synthesize data themselves. Current available synthesis tools include [text_renderer](https://github.com/Sanster/text_renderer), [SynthText](https://github.com/ankush-me/SynthText), [TextRecognitionDataGenerator](https://github.com/Belval/TextRecognitionDataGenerator), etc.
+
+#### 1. ICDAR2019-LSVT
+
+- **Data sources**：<https://aistudio.baidu.com/datasetdetail/177210>
+- **Introduction**：A total of 45w Chinese street view images, including 5w (2w test + 3w training) fully labeled data (text coordinates + text content), 40w weakly labeled data (text content only), as shown in the following figure:
+    ![](./images/LSVT_1.jpg)
+
+    (a) Fully labeled data
+
+    ![](./images/LSVT_2.jpg)
+
+    (b) Weakly labeled data
+- **Download link**：<https://ai.baidu.com/broad/download?dataset=lsvt>
+
+#### 2. ICDAR2017-RCTW-17
+
+- **Data sources**：<https://rctw.vlrlab.net/>
+- **Introduction**：It contains 12000 + images, most of them are collected in the wild through mobile camera. Some are screenshots. These images show a variety of scenes, including street views, posters, menus, indoor scenes and screenshots of mobile applications.
+    ![](./images/rctw.jpg)
+- **Download link**：<https://rctw.vlrlab.net/dataset/>
+
+#### 3. Chinese Street View Text Recognition
+
+- **Data sources**：<https://aistudio.baidu.com/aistudio/competition/detail/8>
+- **Introduction**：A total of 290000 pictures are included, of which 210000 are used as training sets (with labels) and 80000 are used as test sets (without labels). The dataset is collected from the Chinese street view, and is formed by by cutting out the text line area (such as shop signs, landmarks, etc.) in the street view picture. All the images are preprocessed: by using affine transform, the text area is proportionally mapped to a picture with a height of 48 pixels, as shown in the figure:
+
+    ![](./images/ch_street_rec_1.png)
+
+    (a) Label: 魅派集成吊顶
+
+    ![](./images/ch_street_rec_2.png)
+    (b) Label: 母婴用品连锁
+- **Download link**
+<https://aistudio.baidu.com/aistudio/datasetdetail/8429>
+
+#### 4. Chinese Document Text Recognition
+
+- **Data sources**：<https://github.com/YCG09/chinese_ocr>
+- **Introduction**：
+    - A total of 3.64 million pictures are divided into training set and validation set according to 99:1.
+    - Using Chinese corpus (news + classical Chinese), the data is randomly generated through changes in font, size, grayscale, blur, perspective, stretching, etc.
+    - 5990 characters including Chinese characters, English letters, numbers and punctuation（Characters set: <https://github.com/YCG09/chinese_ocr/blob/master/train/char_std_5990.txt> ）
+    - Each sample is fixed with 10 characters, and the characters are randomly intercepted from the sentences in the corpus
+    - Image resolution is 280x32
+
+        ![](./images/ch_doc1.jpg)
+
+        ![](./images/ch_doc3.jpg)
+- **Download link**：<https://pan.baidu.com/s/1QkI7kjah8SPHwOQ40rS1Pw> (Password: lu7m)
+
+#### 5、ICDAR2019-ArT
+
+- **Data source**：<https://aistudio.baidu.com/datasetdetail/177206>
+- **Introduction**：It includes 10166 images, 5603 in training sets and 4563 in test sets. It is composed of three parts: total text, scut-ctw1500 and Baidu curved scene text, including text with various shapes such as horizontal, multi-directional and curved.
+    ![](./images/ArT.jpg)
+- **Download link**：<https://ai.baidu.com/broad/download?dataset=art>
+
+#### 6. Electronic seal dataset
+
+- **Data source**: <https://aistudio.baidu.com/aistudio/datasetdetail/154271/0>
+- **Data introduction**: Contains 10,000 images in total, 8,000 images in the training set, and 2,000 images in the test set. The dataset is synthesized by a program and does not involve privacy security. It is mainly used for the training and detection of seal curved text. Contributed by developer [jingsongliujing](https://github.com/jingsongliujing)
+- **Download address**: <https://aistudio.baidu.com/aistudio/datasetdetail/154271/0>
+
+## References
+
+**ICDAR 2019-LSVT Challenge**
+
+```bibtex
+@article{sun2019icdar,
+  title={ICDAR 2019 Competition on Large-scale Street View Text with Partial Labeling--RRC-LSVT},
+  author={Sun, Yipeng and Ni, Zihan and Chng, Chee-Kheng and Liu, Yuliang and Luo, Canjie and Ng, Chun Chet and Han, Junyu and Ding, Errui and Liu, Jingtuo and Karatzas, Dimosthenis and others},
+  journal={arXiv preprint arXiv:1909.07741},
+  year={2019}
+}
+```
+
+**ICDAR 2019-ArT Challenge**
+
+```bibtex
+@article{chng2019icdar2019,
+  title={ICDAR2019 Robust Reading Challenge on Arbitrary-Shaped Text (RRC-ArT)},
+  author={Chng, Chee-Kheng and Liu, Yuliang and Sun, Yipeng and Ng, Chun Chet and Luo, Canjie and Ni, Zihan and Fang, ChuanMing and Zhang, Shuaitao and Han, Junyu and Ding, Errui and others},
+  journal={arXiv preprint arXiv:1909.07145},
+  year={2019}
+}
+```
diff --git a/docs/datasets/datasets.md b/docs/datasets/datasets.md
new file mode 100644
index 0000000..210bc71
--- /dev/null
+++ b/docs/datasets/datasets.md
@@ -0,0 +1,93 @@
+---
+comments: true
+---
+
+这里整理了常用中文数据集，持续更新中，欢迎各位小伙伴贡献数据集～
+
+除了开源数据，用户还可使用合成工具自行合成，可参考[数据合成工具](../data_anno_synth/data_synthesis.md)；
+
+如果需要标注自己的数据，可参考[数据标注工具](../data_anno_synth/data_annotation.md)。
+
+#### 1、ICDAR2019-LSVT
+
+- **数据来源**：<https://aistudio.baidu.com/datasetdetail/177210>
+- **数据简介**：共45w中文街景图像，包含5w（2w测试+3w训练）全标注数据（文本坐标+文本内容），40w弱标注数据（仅文本内容），如下图所示：
+    ![](./images/LSVT_1.jpg)
+    (a) 全标注数据
+    ![](./images/LSVT_2.jpg)
+    (b) 弱标注数据
+- **下载地址**：<https://ai.baidu.com/broad/download?dataset=lsvt>
+- **说明**：其中，test数据集的label目前没有开源，如要评估结果，可以去官网提交：<https://rrc.cvc.uab.es/?ch=16>
+
+#### 2、ICDAR2017-RCTW-17
+
+- **数据来源**：<https://rctw.vlrlab.net/>
+- **数据简介**：共包含12,000+图像，大部分图片是通过手机摄像头在野外采集的。有些是截图。这些图片展示了各种各样的场景，包括街景、海报、菜单、室内场景和手机应用程序的截图。
+    ![](./images/rctw.jpg)
+- **下载地址**：<https://rctw.vlrlab.net/dataset/>
+
+#### 3、中文街景文字识别
+
+- **数据来源**：<https://aistudio.baidu.com/aistudio/competition/detail/8>
+- **数据简介**：ICDAR2019-LSVT行识别任务，共包括29万张图片，其中21万张图片作为训练集（带标注），8万张作为测试集（无标注）。数据集采自中国街景，并由街景图片中的文字行区域（例如店铺标牌、地标等等）截取出来而形成。所有图像都经过一些预处理，将文字区域利用仿射变化，等比映射为一张高为48像素的图片，如图所示：
+
+    ![](./images/ch_street_rec_1.png)
+
+    (a) 标注：魅派集成吊顶
+
+    ![](./images/ch_street_rec_2.png)
+    (b) 标注：母婴用品连锁
+- **下载地址**
+<https://aistudio.baidu.com/aistudio/datasetdetail/8429>
+
+#### 4、中文文档文字识别
+
+- **数据来源**：<https://github.com/YCG09/chinese_ocr>
+- **数据简介**：
+    - 共约364万张图片，按照99:1划分成训练集和验证集。
+    - 数据利用中文语料库（新闻 + 文言文），通过字体、大小、灰度、模糊、透视、拉伸等变化随机生成
+    - 包含汉字、英文字母、数字和标点共5990个字符（字符集合：<https://github.com/YCG09/chinese_ocr/blob/master/train/char_std_5990.txt> ）
+    - 每个样本固定10个字符，字符随机截取自语料库中的句子
+    - 图片分辨率统一为280x32
+
+      ![](./images/ch_doc1.jpg)
+
+      ![](./images/ch_doc3.jpg)
+- **下载地址**：<https://pan.baidu.com/s/1QkI7kjah8SPHwOQ40rS1Pw> (密码：lu7m)
+
+#### 5、ICDAR2019-ArT
+
+- **数据来源**：<https://aistudio.baidu.com/datasetdetail/177206>
+- **数据简介**：共包含10,166张图像，训练集5603图，测试集4563图。由Total-Text、SCUT-CTW1500、Baidu Curved Scene Text (ICDAR2019-LSVT部分弯曲数据) 三部分组成，包含水平、多方向和弯曲等多种形状的文本。
+    ![](./images/ArT.jpg)
+- **下载地址**：<https://ai.baidu.com/broad/download?dataset=art>
+
+#### 6、电子印章数据集
+
+- **数据来源**：<https://aistudio.baidu.com/aistudio/datasetdetail/154271/0>
+- **数据简介**：共包含10000张图像，训练集8000图，测试集2000图。数据集是用程序合成的，并不涉及隐私安全，主要用于印章弯曲文本的训练与检测。由开发者[jingsongliujing](https://github.com/jingsongliujing)贡献
+- **下载地址**：<https://aistudio.baidu.com/aistudio/datasetdetail/154271/0>
+
+## 参考文献
+
+**ICDAR 2019-LSVT Challenge**
+
+```bibtex
+@article{sun2019icdar,
+  title={ICDAR 2019 Competition on Large-scale Street View Text with Partial Labeling--RRC-LSVT},
+  author={Sun, Yipeng and Ni, Zihan and Chng, Chee-Kheng and Liu, Yuliang and Luo, Canjie and Ng, Chun Chet and Han, Junyu and Ding, Errui and Liu, Jingtuo and Karatzas, Dimosthenis and others},
+  journal={arXiv preprint arXiv:1909.07741},
+  year={2019}
+}
+```
+
+**ICDAR 2019-ArT Challenge**
+
+```bibtex
+@article{chng2019icdar2019,
+  title={ICDAR2019 Robust Reading Challenge on Arbitrary-Shaped Text (RRC-ArT)},
+  author={Chng, Chee-Kheng and Liu, Yuliang and Sun, Yipeng and Ng, Chun Chet and Luo, Canjie and Ni, Zihan and Fang, ChuanMing and Zhang, Shuaitao and Han, Junyu and Ding, Errui and others},
+  journal={arXiv preprint arXiv:1909.07145},
+  year={2019}
+}
+```
diff --git a/docs/datasets/handwritten_datasets.en.md b/docs/datasets/handwritten_datasets.en.md
new file mode 100644
index 0000000..2877458
--- /dev/null
+++ b/docs/datasets/handwritten_datasets.en.md
@@ -0,0 +1,31 @@
+---
+comments: true
+---
+
+
+# Handwritten OCR dataset
+
+Here we have sorted out the commonly used handwritten OCR dataset datasets, which are being updated continuously. We welcome you to contribute datasets ~
+
+- [Institute of automation, Chinese Academy of Sciences - handwritten Chinese dataset](#institute-of-automation-chinese-academy-of-sciences-handwritten-chinese-dataset)
+- [NIST handwritten single character dataset - English](#nist-handwritten-single-character-dataset-englishnist-handprinted-forms-and-characters-database)
+
+## Institute of automation, Chinese Academy of Sciences - handwritten Chinese dataset
+
+- **Data source**: <http://www.nlpr.ia.ac.cn/databases/handwriting/Download.html>
+- **Data introduction**:
+    - It includes online and offline handwritten data,`HWDB1.0~1.2` has totally 3895135 handwritten single character samples, which belong to 7356 categories (7185 Chinese characters and 171 English letters, numbers and symbols);`HWDB2.0~2.2` has totally 5091 pages of images, which are divided into 52230 text lines and 1349414 words. All text and text samples are stored as grayscale images. Some sample words are shown below.
+
+      ![](./images/CASIA_0.jpg)
+
+- **Download address**:<http://www.nlpr.ia.ac.cn/databases/handwriting/Download.html>
+- **使用建议**:Data for single character, white background, can form a large number of text lines for training. White background can be processed into transparent state, which is convenient to add various backgrounds. For the case of semantic needs, it is suggested to extract single character from real corpus to form text lines.
+
+## NIST handwritten single character dataset - English(NIST Handprinted Forms and Characters Database)
+
+- **Data source**: [https://www.nist.gov/srd/nist-special-database-19](https://www.nist.gov/srd/nist-special-database-19)
+- **Data introduction**: NIST19 dataset is suitable for handwritten document and character recognition model training. It is extracted from the handwritten sample form of 3600 authors and contains 810000 character images in total. Nine of them are shown below.
+
+    ![](./images/nist_demo.png)
+
+- **Download address**: [https://www.nist.gov/srd/nist-special-database-19](https://www.nist.gov/srd/nist-special-database-19)
diff --git a/docs/datasets/handwritten_datasets.md b/docs/datasets/handwritten_datasets.md
new file mode 100644
index 0000000..f8a0d0e
--- /dev/null
+++ b/docs/datasets/handwritten_datasets.md
@@ -0,0 +1,28 @@
+---
+comments: true
+---
+
+
+# 手写OCR数据集
+
+这里整理了常用手写数据集，持续更新中，欢迎各位小伙伴贡献数据集～
+
+## 中科院自动化研究所-手写中文数据集
+
+- **数据来源**：<http://www.nlpr.ia.ac.cn/databases/handwriting/Download.html>
+- **数据简介**：
+    - 包含在线和离线两类手写数据，`HWDB1.0~1.2`总共有3895135个手写单字样本，分属7356类（7185个汉字和171个英文字母、数字、符号）；`HWDB2.0~2.2`总共有5091页图像，分割为52230个文本行和1349414个文字。所有文字和文本样本均存为灰度图像。部分单字样本图片如下所示。
+
+    ![](./images/CASIA_0.jpg)
+
+- **下载地址**：<http://www.nlpr.ia.ac.cn/databases/handwriting/Download.html>
+- **使用建议**：数据为单字，白色背景，可以大量合成文字行进行训练。白色背景可以处理成透明状态，方便添加各种背景。对于需要语义的情况，建议从真实语料出发，抽取单字组成文字行
+
+## NIST手写单字数据集-英文(NIST Handprinted Forms and Characters Database)
+
+- **数据来源**: [https://www.nist.gov/srd/nist-special-database-19](https://www.nist.gov/srd/nist-special-database-19)
+- **数据简介**: NIST19数据集适用于手写文档和字符识别的模型训练，从3600位作者的手写样本表格中提取得到，总共包含81万张字符图片。其中9张图片示例如下：
+
+    ![](./images/nist_demo.png)
+
+- **下载地址**: [https://www.nist.gov/srd/nist-special-database-19](https://www.nist.gov/srd/nist-special-database-19)
diff --git a/docs/datasets/images/20210816_210413.gif b/docs/datasets/images/20210816_210413.gif
new file mode 100644
index 0000000..288f615
Binary files /dev/null and b/docs/datasets/images/20210816_210413.gif differ
diff --git a/docs/datasets/images/ArT.jpg b/docs/datasets/images/ArT.jpg
new file mode 100644
index 0000000..7855ba4
Binary files /dev/null and b/docs/datasets/images/ArT.jpg differ
diff --git a/docs/datasets/images/CASIA_0.jpg b/docs/datasets/images/CASIA_0.jpg
new file mode 100644
index 0000000..d65924b
Binary files /dev/null and b/docs/datasets/images/CASIA_0.jpg differ
diff --git a/docs/datasets/images/CDLA_demo/val_0633.jpg b/docs/datasets/images/CDLA_demo/val_0633.jpg
new file mode 100644
index 0000000..8348485
Binary files /dev/null and b/docs/datasets/images/CDLA_demo/val_0633.jpg differ
diff --git a/docs/datasets/images/CDLA_demo/val_0941.jpg b/docs/datasets/images/CDLA_demo/val_0941.jpg
new file mode 100644
index 0000000..f7d548e
Binary files /dev/null and b/docs/datasets/images/CDLA_demo/val_0941.jpg differ
diff --git a/docs/datasets/images/LSVT_1.jpg b/docs/datasets/images/LSVT_1.jpg
new file mode 100644
index 0000000..ea11a7d
Binary files /dev/null and b/docs/datasets/images/LSVT_1.jpg differ
diff --git a/docs/datasets/images/LSVT_2.jpg b/docs/datasets/images/LSVT_2.jpg
new file mode 100644
index 0000000..67bfbe5
Binary files /dev/null and b/docs/datasets/images/LSVT_2.jpg differ
diff --git a/docs/datasets/images/captcha_demo.png b/docs/datasets/images/captcha_demo.png
new file mode 100644
index 0000000..047a726
Binary files /dev/null and b/docs/datasets/images/captcha_demo.png differ
diff --git a/docs/datasets/images/ccpd_demo.png b/docs/datasets/images/ccpd_demo.png
new file mode 100644
index 0000000..a750d05
Binary files /dev/null and b/docs/datasets/images/ccpd_demo.png differ
diff --git a/docs/datasets/images/ch_doc1.jpg b/docs/datasets/images/ch_doc1.jpg
new file mode 100644
index 0000000..5353440
Binary files /dev/null and b/docs/datasets/images/ch_doc1.jpg differ
diff --git a/docs/datasets/images/ch_doc3.jpg b/docs/datasets/images/ch_doc3.jpg
new file mode 100644
index 0000000..c0c2053
Binary files /dev/null and b/docs/datasets/images/ch_doc3.jpg differ
diff --git a/docs/datasets/images/ch_street_rec_1.png b/docs/datasets/images/ch_street_rec_1.png
new file mode 100644
index 0000000..a0e158c
Binary files /dev/null and b/docs/datasets/images/ch_street_rec_1.png differ
diff --git a/docs/datasets/images/ch_street_rec_2.png b/docs/datasets/images/ch_street_rec_2.png
new file mode 100644
index 0000000..bfa0fd0
Binary files /dev/null and b/docs/datasets/images/ch_street_rec_2.png differ
diff --git a/docs/datasets/images/cmb_demo.jpg b/docs/datasets/images/cmb_demo.jpg
new file mode 100644
index 0000000..8299149
Binary files /dev/null and b/docs/datasets/images/cmb_demo.jpg differ
diff --git a/docs/datasets/images/crohme_demo/hme_00.jpg b/docs/datasets/images/crohme_demo/hme_00.jpg
new file mode 100644
index 0000000..66ff27d
Binary files /dev/null and b/docs/datasets/images/crohme_demo/hme_00.jpg differ
diff --git a/docs/datasets/images/crohme_demo/hme_01.jpg b/docs/datasets/images/crohme_demo/hme_01.jpg
new file mode 100644
index 0000000..68b7f09
Binary files /dev/null and b/docs/datasets/images/crohme_demo/hme_01.jpg differ
diff --git a/docs/datasets/images/crohme_demo/hme_02.jpg b/docs/datasets/images/crohme_demo/hme_02.jpg
new file mode 100644
index 0000000..ecc760f
Binary files /dev/null and b/docs/datasets/images/crohme_demo/hme_02.jpg differ
diff --git a/docs/datasets/images/funsd_demo/gt_train_00040534.jpg b/docs/datasets/images/funsd_demo/gt_train_00040534.jpg
new file mode 100644
index 0000000..9f7cf4d
Binary files /dev/null and b/docs/datasets/images/funsd_demo/gt_train_00040534.jpg differ
diff --git a/docs/datasets/images/funsd_demo/gt_train_00070353.jpg b/docs/datasets/images/funsd_demo/gt_train_00070353.jpg
new file mode 100644
index 0000000..36d3345
Binary files /dev/null and b/docs/datasets/images/funsd_demo/gt_train_00070353.jpg differ
diff --git a/docs/datasets/images/ic15_location_download.png b/docs/datasets/images/ic15_location_download.png
new file mode 100644
index 0000000..7cb8540
Binary files /dev/null and b/docs/datasets/images/ic15_location_download.png differ
diff --git a/docs/datasets/images/icdar_rec.png b/docs/datasets/images/icdar_rec.png
new file mode 100644
index 0000000..a840d6a
Binary files /dev/null and b/docs/datasets/images/icdar_rec.png differ
diff --git a/docs/datasets/images/nist_demo.png b/docs/datasets/images/nist_demo.png
new file mode 100644
index 0000000..4c2ce11
Binary files /dev/null and b/docs/datasets/images/nist_demo.png differ
diff --git a/docs/datasets/images/pme_demo/0000013.png b/docs/datasets/images/pme_demo/0000013.png
new file mode 100644
index 0000000..9f8b11b
Binary files /dev/null and b/docs/datasets/images/pme_demo/0000013.png differ
diff --git a/docs/datasets/images/pme_demo/0000099.png b/docs/datasets/images/pme_demo/0000099.png
new file mode 100644
index 0000000..a179886
Binary files /dev/null and b/docs/datasets/images/pme_demo/0000099.png differ
diff --git a/docs/datasets/images/pme_demo/0000295.png b/docs/datasets/images/pme_demo/0000295.png
new file mode 100644
index 0000000..26a271a
Binary files /dev/null and b/docs/datasets/images/pme_demo/0000295.png differ
diff --git a/docs/datasets/images/pme_demo/0000562.png b/docs/datasets/images/pme_demo/0000562.png
new file mode 100644
index 0000000..121eda8
Binary files /dev/null and b/docs/datasets/images/pme_demo/0000562.png differ
diff --git a/docs/datasets/images/publaynet_demo/gt_PMC3724501_00006.jpg b/docs/datasets/images/publaynet_demo/gt_PMC3724501_00006.jpg
new file mode 100644
index 0000000..3b7ee89
Binary files /dev/null and b/docs/datasets/images/publaynet_demo/gt_PMC3724501_00006.jpg differ
diff --git a/docs/datasets/images/publaynet_demo/gt_PMC5086060_00002.jpg b/docs/datasets/images/publaynet_demo/gt_PMC5086060_00002.jpg
new file mode 100644
index 0000000..cad8f30
Binary files /dev/null and b/docs/datasets/images/publaynet_demo/gt_PMC5086060_00002.jpg differ
diff --git a/docs/datasets/images/rctw.jpg b/docs/datasets/images/rctw.jpg
new file mode 100644
index 0000000..1e1f945
Binary files /dev/null and b/docs/datasets/images/rctw.jpg differ
diff --git a/docs/datasets/images/table_PubTabNet_demo/PMC524509_007_00.png b/docs/datasets/images/table_PubTabNet_demo/PMC524509_007_00.png
new file mode 100755
index 0000000..5b9d631
Binary files /dev/null and b/docs/datasets/images/table_PubTabNet_demo/PMC524509_007_00.png differ
diff --git a/docs/datasets/images/table_PubTabNet_demo/PMC535543_007_01.png b/docs/datasets/images/table_PubTabNet_demo/PMC535543_007_01.png
new file mode 100755
index 0000000..e808de7
Binary files /dev/null and b/docs/datasets/images/table_PubTabNet_demo/PMC535543_007_01.png differ
diff --git a/docs/datasets/images/table_tal_demo/1.jpg b/docs/datasets/images/table_tal_demo/1.jpg
new file mode 100644
index 0000000..e7ddd6d
Binary files /dev/null and b/docs/datasets/images/table_tal_demo/1.jpg differ
diff --git a/docs/datasets/images/table_tal_demo/2.jpg b/docs/datasets/images/table_tal_demo/2.jpg
new file mode 100644
index 0000000..e7ddd6d
Binary files /dev/null and b/docs/datasets/images/table_tal_demo/2.jpg differ
diff --git a/docs/datasets/images/tablebank_demo/004.png b/docs/datasets/images/tablebank_demo/004.png
new file mode 100644
index 0000000..c1a2d36
Binary files /dev/null and b/docs/datasets/images/tablebank_demo/004.png differ
diff --git a/docs/datasets/images/tablebank_demo/005.png b/docs/datasets/images/tablebank_demo/005.png
new file mode 100644
index 0000000..0d4d6ab
Binary files /dev/null and b/docs/datasets/images/tablebank_demo/005.png differ
diff --git a/docs/datasets/images/wildreceipt_demo/1bbe854b8817dedb8585e0732089fd1f752d2cec.jpeg b/docs/datasets/images/wildreceipt_demo/1bbe854b8817dedb8585e0732089fd1f752d2cec.jpeg
new file mode 100644
index 0000000..dfed3a0
Binary files /dev/null and b/docs/datasets/images/wildreceipt_demo/1bbe854b8817dedb8585e0732089fd1f752d2cec.jpeg differ
diff --git a/docs/datasets/images/wildreceipt_demo/2769.jpeg b/docs/datasets/images/wildreceipt_demo/2769.jpeg
new file mode 100644
index 0000000..d5a2876
Binary files /dev/null and b/docs/datasets/images/wildreceipt_demo/2769.jpeg differ
diff --git a/docs/datasets/images/xfund_demo/gt_zh_train_0.jpg b/docs/datasets/images/xfund_demo/gt_zh_train_0.jpg
new file mode 100644
index 0000000..95e0cf8
Binary files /dev/null and b/docs/datasets/images/xfund_demo/gt_zh_train_0.jpg differ
diff --git a/docs/datasets/images/xfund_demo/gt_zh_train_1.jpg b/docs/datasets/images/xfund_demo/gt_zh_train_1.jpg
new file mode 100644
index 0000000..6a1e53a
Binary files /dev/null and b/docs/datasets/images/xfund_demo/gt_zh_train_1.jpg differ
diff --git a/docs/datasets/kie_datasets.en.md b/docs/datasets/kie_datasets.en.md
new file mode 100644
index 0000000..ca912b2
--- /dev/null
+++ b/docs/datasets/kie_datasets.en.md
@@ -0,0 +1,47 @@
+---
+comments: true
+---
+
+
+## Key Information Extraction dataset
+
+Here are the common datasets key information extraction, which are being updated continuously. Welcome to contribute datasets.
+
+### 1. FUNSD dataset
+
+- **Data source**: <https://guillaumejaume.github.io/FUNSD/>
+- **Data Introduction**: The FUNSD dataset is a dataset for form comprehension. It contains 199 real, fully annotated scanned images, including market reports, advertisements, and academic reports, etc., and is divided into 149 training set and 50 test set. The FUNSD dataset is suitable for many types of DocVQA tasks, such as field-level entity classification, field-level entity connection, etc. Part of the image and the annotation box visualization are shown below:
+
+    ![](./images/funsd_demo/gt_train_00040534.jpg)
+
+    ![](./images/funsd_demo/gt_train_00070353.jpg)
+
+    In the figure, the orange area represents `header`, the light blue area represents `question`, the green area represents `answer`, and the pink area represents `other`.
+
+- **Download address**: <https://guillaumejaume.github.io/FUNSD/download/>
+
+### 2. XFUND dataset
+
+- **Data source**: <https://github.com/doc-analysis/XFUND>
+- **Data introduction**: XFUND is a multilingual form comprehension dataset, which contains form data in 7 different languages, and all are manually annotated in the form of key-value pairs. The data for each language contains 199 form data, which are divided into 149 training sets and 50 test sets. Part of the image and the annotation box visualization are shown below.
+
+    ![](./images/xfund_demo/gt_zh_train_0.jpg)
+
+    ![](./images/xfund_demo/gt_zh_train_1.jpg)
+
+- **Download address**: <https://github.com/doc-analysis/XFUND/releases/tag/v1.0>
+
+### 3. wildreceipt dataset
+
+- **Data source**: <https://arxiv.org/abs/2103.14470>
+- **Data introduction**: wildreceipt is an English receipt dataset, which contains 26 different categories. There are 1267 training images and 472 evaluation images, in which 50,000 textlines and boxes are annotated. Part of the image and the annotation box visualization are shown below.
+
+    ![](./images/wildreceipt_demo/2769.jpeg)
+
+    ![](./images/wildreceipt_demo/1bbe854b8817dedb8585e0732089fd1f752d2cec.jpeg)
+
+**Note：** Boxes with category `Ignore` or `Others` are not visualized here.
+
+- **Download address**：
+    - Official dataset: [link](https://download.openmmlab.com/mmocr/data/wildreceipt.tar)
+    - Dataset converted for PaddleOCR training process: [link](https://paddleocr.bj.bcebos.com/ppstructure/dataset/wildreceipt.tar)
diff --git a/docs/datasets/kie_datasets.md b/docs/datasets/kie_datasets.md
new file mode 100644
index 0000000..6dbe82d
--- /dev/null
+++ b/docs/datasets/kie_datasets.md
@@ -0,0 +1,49 @@
+---
+comments: true
+---
+
+
+# 关键信息抽取数据集
+
+这里整理了常见的关键信息抽取数据集，持续更新中，欢迎各位小伙伴贡献数据集～
+
+## 1. FUNSD数据集
+
+- **数据来源**：<https://guillaumejaume.github.io/FUNSD/>
+- **数据简介**：FUNSD数据集是一个用于表单理解的数据集，它包含199张真实的、完全标注的扫描版图片，类型包括市场报告、广告以及学术报告等，并分为149张训练集以及50张测试集。FUNSD数据集适用于多种类型的DocVQA任务，如字段级实体分类、字段级实体连接等。部分图像以及标注框可视化如下所示:
+
+<div align="center">
+    <img src="./images/funsd_demo/gt_train_00040534.jpg" width="500">
+    <img src="./images/funsd_demo/gt_train_00070353.jpg" width="500">
+</div>
+    图中，橙色区域代表`header`，淡蓝色区域代表`question`, 绿色区域表`answer`，粉红色代区域表`other`。
+
+- **下载地址**：<https://guillaumejaume.github.io/FUNSD/download/>
+
+## 2. XFUND数据集
+
+- **数据来源**：<https://github.com/doc-analysis/XFUND>
+- **数据简介**：XFUND是一个多语种表单理解数据集，它包含7种不同语种的表单数据，并且全部用人工进行了键-值对形式的标注。其中每个语种的数据都包含了199张表单数据，并分为149张训练集以及50张测试集。部分图像以及标注框可视化如下所示:
+
+<div align="center">
+    <img src="./images/xfund_demo/gt_zh_train_0.jpg" width="500">
+    <img src="./images/xfund_demo/gt_zh_train_1.jpg" width="500">
+</div>
+
+- **下载地址**：<https://github.com/doc-analysis/XFUND/releases/tag/v1.0>
+
+## 3. wildreceipt数据集
+
+- **数据来源**：<https://arxiv.org/abs/2103.14470>
+- **数据简介**：wildreceipt数据集是英文发票数据集，包含26个类别（此处类别体系包含`Ignore`类别），共标注了50000个文本框。其中训练集包含1267张图片，测试集包含472张图片。部分图像以及标注框可视化如下所示:
+
+<div align="center">
+    <img src="./images/wildreceipt_demo/2769.jpeg" width="500">
+    <img src="./images/wildreceipt_demo/1bbe854b8817dedb8585e0732089fd1f752d2cec.jpeg" width="500">
+</div>
+
+**注：** 这里对于类别为`Ignore`或者`Others`的文本，没有进行可视化。
+
+- **下载地址**：
+    - 原始数据下载地址：[链接](https://download.openmmlab.com/mmocr/data/wildreceipt.tar)
+    - 数据格式转换后适配于PaddleOCR训练的数据下载地址：[链接](https://paddleocr.bj.bcebos.com/ppstructure/dataset/wildreceipt.tar)
diff --git a/docs/datasets/layout_datasets.en.md b/docs/datasets/layout_datasets.en.md
new file mode 100644
index 0000000..2e9a074
--- /dev/null
+++ b/docs/datasets/layout_datasets.en.md
@@ -0,0 +1,46 @@
+---
+comments: true
+---
+
+
+## Layout Analysis Dataset
+
+Here are the common datasets of layout analysis, which are being updated continuously. Welcome to contribute datasets.
+
+Most of the layout analysis datasets are object detection datasets. In addition to open source datasets, you can also label or synthesize datasets using tools such as [labelme](https://github.com/wkentaro/labelme) and so on.
+
+### 1. PubLayNet dataset
+
+- **Data source**: <https://github.com/ibm-aur-nlp/PubLayNet>
+- **Data introduction**: The PubLayNet dataset contains 350000 training images and 11000 validation images. There are 5 categories in total, namely: `text, title, list, table, figure`. Some images and their annotations as shown below.
+
+    ![](./images/publaynet_demo/gt_PMC3724501_00006.jpg)
+
+    ![](./images/publaynet_demo/gt_PMC5086060_00002.jpg)
+
+- **Download address**: <https://developer.ibm.com/exchanges/data/all/publaynet/>
+- **Note**: When using this dataset, you need to follow [CDLA-Permissive](https://cdla.io/permissive-1-0/) license.
+
+### 2、CDLA dataset
+
+- **Data source**: <https://github.com/buptlihang/CDLA>
+- **Data introduction**: CDLA dataset contains 5000 training images and 1000 validation images with 10 categories, which are `Text, Title, Figure, Figure caption, Table, Table caption, Header, Footer, Reference, Equation`. Some images and their annotations as shown below.
+
+    ![](./images/CDLA_demo/val_0633.jpg)
+
+    ![](./images/CDLA_demo/val_0941.jpg)
+
+- **Download address**: <https://github.com/buptlihang/CDLA>
+- **Note**: When you train detection model on CDLA dataset using [PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection/tree/develop), you need to remove the label `__ignore__` and `_background_`.
+
+### 3、TableBank dataset
+
+- **Data source**: <https://doc-analysis.github.io/tablebank-page/index.html>
+- **Data introduction**: TableBank dataset contains 2 types of document: Latex (187199 training images, 7265 validation images and 5719 testing images) and Word (73383 training images 2735 validation images and 2281 testing images). Some images and their annotations as shown below.
+
+    ![](./images/tablebank_demo/004.png)
+
+    ![](./images/tablebank_demo/005.png)
+
+- **Data source**: <https://doc-analysis.github.io/tablebank-page/index.html>
+- **Note**: When using this dataset, you need to follow [Apache-2.0](https://github.com/doc-analysis/TableBank/blob/master/LICENSE) license.
diff --git a/docs/datasets/layout_datasets.md b/docs/datasets/layout_datasets.md
new file mode 100644
index 0000000..7396048
--- /dev/null
+++ b/docs/datasets/layout_datasets.md
@@ -0,0 +1,49 @@
+---
+comments: true
+---
+
+
+## 版面分析数据集
+
+这里整理了常用版面分析数据集，持续更新中，欢迎各位小伙伴贡献数据集～
+
+版面分析数据集多为目标检测数据集，除了开源数据，用户还可使用合成工具自行合成，如[labelme](https://github.com/wkentaro/labelme)等。
+
+### 1、publaynet数据集
+
+- **数据来源**：<https://github.com/ibm-aur-nlp/PubLayNet>
+- **数据简介**：publaynet数据集的训练集合中包含35万张图像，验证集合中包含1.1万张图像。总共包含5个类别，分别是： `text, title, list, table, figure`。部分图像以及标注框可视化如下所示。
+
+<div align="center">
+    <img src="./images/publaynet_demo/gt_PMC3724501_00006.jpg" width="500">
+    <img src="./images/publaynet_demo/gt_PMC5086060_00002.jpg" width="500">
+</div>
+
+- **下载地址**：<https://developer.ibm.com/exchanges/data/all/publaynet/>
+- **说明**：使用该数据集时，需要遵守[CDLA-Permissive](https://cdla.io/permissive-1-0/)协议。
+
+### 2、CDLA数据集
+
+- **数据来源**：<https://github.com/buptlihang/CDLA>
+- **数据简介**：CDLA据集的训练集合中包含5000张图像，验证集合中包含1000张图像。总共包含10个类别，分别是： `Text, Title, Figure, Figure caption, Table, Table caption, Header, Footer, Reference, Equation`。部分图像以及标注框可视化如下所示。
+
+<div align="center">
+    <img src="./images/CDLA_demo/val_0633.jpg" width="500">
+    <img src="./images/CDLA_demo/val_0941.jpg" width="500">
+</div>
+
+- **下载地址**：<https://github.com/buptlihang/CDLA>
+- **说明**：基于[PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection/tree/develop)套件，在该数据集上训练目标检测模型时，在转换label时，需要将`label.txt`中的`__ignore__`与`_background_`去除。
+
+### 3、TableBank数据集
+
+- **数据来源**：<https://doc-analysis.github.io/tablebank-page/index.html>
+- **数据简介**：TableBank数据集包含Latex（训练集187199张，验证集7265张，测试集5719张）与Word（训练集73383张，验证集2735张，测试集2281张）两种类别的文档。仅包含`Table` 1个类别。部分图像以及标注框可视化如下所示。
+
+<div align="center">
+    <img src="./images/tablebank_demo/004.png" height="700">
+    <img src="./images/tablebank_demo/005.png" height="700">
+</div>
+
+- **下载地址**：<https://doc-analysis.github.io/tablebank-page/index.html>
+- **说明**：使用该数据集时，需要遵守[Apache-2.0](https://github.com/doc-analysis/TableBank/blob/master/LICENSE)协议。
diff --git a/docs/datasets/ocr_datasets.en.md b/docs/datasets/ocr_datasets.en.md
new file mode 100644
index 0000000..e4000e9
--- /dev/null
+++ b/docs/datasets/ocr_datasets.en.md
@@ -0,0 +1,154 @@
+---
+comments: true
+---
+
+
+# OCR datasets
+
+Here is a list of public datasets commonly used in OCR, which are being continuously updated. Welcome to contribute datasets!
+
+## 1. Text detection
+
+### 1.1 PaddleOCR text detection format annotation
+
+The annotation file formats supported by the PaddleOCR text detection algorithm are as follows, separated by "\t":
+
+```text linenums="1"
+"Image file name             Image annotation information encoded by json.dumps"
+ch4_test_images/img_61.jpg    [{"transcription": "MASA", "points": [[310, 104], [416, 141], [418, 216], [312, 179]]}, {...}]
+```
+
+The image annotation after **json.dumps()** encoding is a list containing multiple dictionaries.
+
+The `points` in the dictionary represent the coordinates (x, y) of the four points of the text box, arranged clockwise from the point at the upper left corner.
+
+`transcription` represents the text of the current text box. **When its content is "###" it means that the text box is invalid and will be skipped during training.**
+
+If you want to train PaddleOCR on other datasets, please build the annotation file according to the above format.
+
+### 1.2 Public dataset
+
+| dataset | Image download link | PaddleOCR format annotation download link |
+|---|---|---|
+| ICDAR 2015 | <https://rrc.cvc.uab.es/?ch=4&com=downloads>            | [train](https://paddleocr.bj.bcebos.com/dataset/train_icdar2015_label.txt) / [test](https://paddleocr.bj.bcebos.com/dataset/test_icdar2015_label.txt) |
+| ctw1500 | <https://paddleocr.bj.bcebos.com/dataset/ctw1500.zip>   | Included in the downloaded image zip                                                                                                           |
+| total text | <https://paddleocr.bj.bcebos.com/dataset/total_text.tar> |  Included in the downloaded image zip  |
+
+#### 1.2.1 ICDAR 2015
+
+The icdar2015 dataset contains train set which has 1000 images obtained with wearable cameras and test set which has 500 images obtained with wearable cameras. The icdar2015 dataset can be downloaded from the link in the table above. Registration is required for downloading.
+
+After registering and logging in, download the part marked in the red box in the figure below. And, the content downloaded by `Training Set Images` should be saved as the folder `icdar_c4_train_imgs`, and the content downloaded by `Test Set Images` is saved as the folder `ch4_test_images`
+
+![](./images/ic15_location_download.png)
+
+Decompress the downloaded dataset to the working directory, assuming it is decompressed under PaddleOCR/train_data/. Then download the PaddleOCR format annotation file from the table above.
+
+PaddleOCR also provides a data format conversion script, which can convert the official website label to the PaddleOCR format. The data conversion tool is in `ppocr/utils/gen_label.py`, here is the training set as an example:
+
+```bash linenums="1"
+# Convert the label file downloaded from the official website to train_icdar2015_label.txt
+python gen_label.py --mode="det" --root_path="/path/to/icdar_c4_train_imgs/"  \
+                    --input_path="/path/to/ch4_training_localization_transcription_gt" \
+                    --output_label="/path/to/train_icdar2015_label.txt"
+```
+
+After decompressing the data set and downloading the annotation file, PaddleOCR/train_data/ has two folders and two files, which are:
+
+```text linenums="1"
+/PaddleOCR/train_data/icdar2015/text_localization/
+  └─ icdar_c4_train_imgs/         Training data of icdar dataset
+  └─ ch4_test_images/             Testing data of icdar dataset
+  └─ train_icdar2015_label.txt    Training annotation of icdar dataset
+  └─ test_icdar2015_label.txt     Test annotation of icdar dataset
+```
+
+## 2. Text recognition
+
+### 2.1 PaddleOCR text recognition format annotation
+
+The text recognition algorithm in PaddleOCR supports two data formats:
+
+- `lmdb` is used to train data sets stored in lmdb format, use [lmdb_dataset.py](https://github.com/PaddlePaddle/PaddleOCR/blob/ee9d22b4524eab44170568c98e73d742ce997ff2/ppocr/data/lmdb_dataset.py) to load;
+- `common dataset` is used to train data sets stored in text files, use [simple_dataset.py](https://github.com/PaddlePaddle/PaddleOCR/blob/ee9d22b4524eab44170568c98e73d742ce997ff2/ppocr/data/simple_dataset.py) to load.
+
+If you want to use your own data for training, please refer to the following to organize your data.
+
+#### Training set
+
+It is recommended to put the training images in the same folder, and use a txt file (rec_gt_train.txt) to store the image path and label. The contents of the txt file are as follows:
+
+- Note: by default, the image path and image label are split with \t, if you use other methods to split, it will cause training error
+
+```text linenums="1"
+" Image file name           Image annotation "
+
+train_data/rec/train/word_001.jpg   简单可依赖
+train_data/rec/train/word_002.jpg   用科技让复杂的世界更简单
+...
+```
+
+The final training set should have the following file structure:
+
+```text linenums="1"
+|-train_data
+  |-rec
+    |- rec_gt_train.txt
+    |- train
+        |- word_001.png
+        |- word_002.jpg
+        |- word_003.jpg
+        | ...
+```
+
+#### Test set
+
+Similar to the training set, the test set also needs to be provided a folder containing all images (test) and a rec_gt_test.txt. The structure of the test set is as follows:
+
+```text linenums="1"
+|-train_data
+  |-rec
+    |-ic15_data
+        |- rec_gt_test.txt
+        |- test
+            |- word_001.jpg
+            |- word_002.jpg
+            |- word_003.jpg
+            | ...
+```
+
+### 2.2 Public dataset
+
+| dataset | Image download link | PaddleOCR format annotation download link |
+|---|---|---|
+| en benchmark(MJ, SJ, IIIT, SVT, IC03, IC13, IC15, SVTP, and CUTE.) | [DTRB](https://github.com/clovaai/deep-text-recognition-benchmark#download-lmdb-dataset-for-traininig-and-evaluation-from-here) | LMDB format, which can be loaded directly with [lmdb_dataset.py](https://github.com/PaddlePaddle/PaddleOCR/blob/ee9d22b4524eab44170568c98e73d742ce997ff2/ppocr/data/lmdb_dataset.py) |
+|ICDAR 2015| <http://rrc.cvc.uab.es/?ch=4&com=downloads> | [train](https://paddleocr.bj.bcebos.com/dataset/rec_gt_train.txt)/ [test](https://paddleocr.bj.bcebos.com/dataset/rec_gt_test.txt) |
+| Multilingual datasets |[Baidu network disk](https://pan.baidu.com/s/1bS_u207Rm7YbY33wOECKDA) Extraction code: frgi | Included in the downloaded image zip |
+
+#### 2.1 ICDAR 2015
+
+The ICDAR 2015 dataset can be downloaded from the link in the table above for quick validation. The lmdb format dataset required by en benchmark can also be downloaded from the table above.
+
+Then download the PaddleOCR format annotation file from the table above.
+
+PaddleOCR also provides a data format conversion script, which can convert the ICDAR official website label to the data format supported by PaddleOCR. The data conversion tool is in `ppocr/utils/gen_label.py`, here is the training set as an example:
+
+```bash linenums="1"
+# Convert the label file downloaded from the official website to rec_gt_label.txt
+python gen_label.py --mode="rec" --input_path="{path/of/origin/label}" --output_label="rec_gt_label.txt"
+```
+
+The data format is as follows, (a) is the original picture, (b) is the Ground Truth text file corresponding to each picture:
+
+![](./images/icdar_rec.png)
+
+## 3. Data storage path
+
+The default storage path for PaddleOCR training data is `PaddleOCR/train_data`, if you already have a dataset on your disk, just create a soft link to the dataset directory:
+
+```bash linenums="1"
+# linux and mac os
+ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/dataset
+# windows
+mklink /d <path/to/paddle_ocr>/train_data/dataset <path/to/dataset>
+```
diff --git a/docs/datasets/ocr_datasets.md b/docs/datasets/ocr_datasets.md
new file mode 100644
index 0000000..be6722b
--- /dev/null
+++ b/docs/datasets/ocr_datasets.md
@@ -0,0 +1,160 @@
+---
+comments: true
+---
+
+
+# OCR数据集
+
+这里整理了OCR中常用的公开数据集，持续更新中，欢迎各位小伙伴贡献数据集～
+
+## 1. 文本检测
+
+### 1.1 PaddleOCR 文字检测数据格式
+
+PaddleOCR 中的文本检测算法支持的标注文件格式如下，中间用"\t"分隔：
+
+```text linenums="1"
+" 图像文件名                    json.dumps编码的图像标注信息"
+ch4_test_images/img_61.jpg    [{"transcription": "MASA", "points": [[310, 104], [416, 141], [418, 216], [312, 179]]}, {...}]
+```
+
+json.dumps编码前的图像标注信息是包含多个字典的list，字典中的 `points` 表示文本框的四个点的坐标(x, y)，从左上角的点开始顺时针排列。
+`transcription` 表示当前文本框的文字，**当其内容为“###”时，表示该文本框无效，在训练时会跳过。**
+
+如果您想在我们未提供的数据集上训练，可以按照上述形式构建标注文件。
+
+### 1.2 公开数据集
+
+| 数据集名称 |图片下载地址| PaddleOCR 标注下载地址 |
+|---|---|---|
+| ICDAR 2015 |<https://rrc.cvc.uab.es/?ch=4&com=downloads>| [train](https://paddleocr.bj.bcebos.com/dataset/train_icdar2015_label.txt) / [test](https://paddleocr.bj.bcebos.com/dataset/test_icdar2015_label.txt) |
+| ctw1500 |<https://paddleocr.bj.bcebos.com/dataset/ctw1500.zip>| 图片下载地址中已包含 |
+| total text |<https://paddleocr.bj.bcebos.com/dataset/total_text.tar>| 图片下载地址中已包含 |
+| td tr |<https://paddleocr.bj.bcebos.com/dataset/TD_TR.tar>| 图片下载地址中已包含 |
+
+#### 1.2.1 ICDAR 2015
+
+ICDAR 2015 数据集包含1000张训练图像和500张测试图像。ICDAR 2015 数据集可以从上表中链接下载，首次下载需注册。
+注册完成登陆后，下载下图中红色框标出的部分，其中， `Training Set Images`下载的内容保存在`icdar_c4_train_imgs`文件夹下，`Test Set Images` 下载的内容保存早`ch4_test_images`文件夹下
+
+ <img src="./images/ic15_location_download.png" align="middle" width = "700"/>
+
+将下载到的数据集解压到工作目录下，假设解压在 PaddleOCR/train_data/下。然后从上表中下载转换好的标注文件。
+
+PaddleOCR 也提供了数据格式转换脚本，可以将官网 label 转换支持的数据格式。 数据转换工具在 `ppocr/utils/gen_label.py`, 这里以训练集为例：
+
+```bash linenums="1"
+# 将官网下载的标签文件转换为 train_icdar2015_label.txt
+python gen_label.py --mode="det" --root_path="/path/to/icdar_c4_train_imgs/"  \
+                    --input_path="/path/to/ch4_training_localization_transcription_gt" \
+                    --output_label="/path/to/train_icdar2015_label.txt"
+```
+
+解压数据集和下载标注文件后，PaddleOCR/train_data/ 有两个文件夹和两个文件，按照如下方式组织icdar2015数据集：
+
+```text linenums="1"
+/PaddleOCR/train_data/icdar2015/text_localization/
+  └─ icdar_c4_train_imgs/         icdar 2015 数据集的训练数据
+  └─ ch4_test_images/             icdar 2015 数据集的测试数据
+  └─ train_icdar2015_label.txt    icdar 2015 数据集的训练标注
+  └─ test_icdar2015_label.txt     icdar 2015 数据集的测试标注
+```
+
+## 2. 文本识别
+
+### 2.1 PaddleOCR 文字识别数据格式
+
+PaddleOCR 中的文字识别算法支持两种数据格式:
+
+- `lmdb` 用于训练以lmdb格式存储的数据集，使用 [lmdb_dataset.py](https://github.com/PaddlePaddle/PaddleOCR/blob/ee9d22b4524eab44170568c98e73d742ce997ff2/ppocr/data/lmdb_dataset.py) 进行读取;
+- `通用数据` 用于训练以文本文件存储的数据集，使用 [simple_dataset.py](https://github.com/PaddlePaddle/PaddleOCR/blob/ee9d22b4524eab44170568c98e73d742ce997ff2/ppocr/data/simple_dataset.py)进行读取。
+
+下面以通用数据集为例， 介绍如何准备数据集：
+
+#### 训练集
+
+建议将训练图片放入同一个文件夹，并用一个txt文件（rec_gt_train.txt）记录图片路径和标签，txt文件里的内容如下:
+
+**注意：** txt文件中默认请将图片路径和图片标签用 \t 分割，如用其他方式分割将造成训练报错。
+
+```text linenums="1"
+" 图像文件名                 图像标注信息 "
+
+train_data/rec/train/word_001.jpg   简单可依赖
+train_data/rec/train/word_002.jpg   用科技让复杂的世界更简单
+...
+```
+
+最终训练集应有如下文件结构：
+
+```text linenums="1"
+|-train_data
+  |-rec
+    |- rec_gt_train.txt
+    |- train
+        |- word_001.png
+        |- word_002.jpg
+        |- word_003.jpg
+        | ...
+```
+
+除上述单张图像为一行格式之外，PaddleOCR也支持对离线增广后的数据进行训练，为了防止相同样本在同一个batch中被多次采样，我们可以将相同标签对应的图片路径写在一行中，以列表的形式给出，在训练中，PaddleOCR会随机选择列表中的一张图片进行训练。对应地，标注文件的格式如下：
+
+```text linenums="1"
+["11.jpg", "12.jpg"]   简单可依赖
+["21.jpg", "22.jpg", "23.jpg"]   用科技让复杂的世界更简单
+3.jpg   ocr
+```
+
+上述示例标注文件中，"11.jpg"和"12.jpg"的标签相同，都是`简单可依赖`，在训练的时候，对于该行标注，会随机选择其中的一张图片进行训练。
+
+#### 验证集
+
+同训练集类似，验证集也需要提供一个包含所有图片的文件夹（test）和一个rec_gt_test.txt，验证集的结构如下所示：
+
+```text linenums="1"
+|-train_data
+  |-rec
+    |- rec_gt_test.txt
+    |- test
+        |- word_001.jpg
+        |- word_002.jpg
+        |- word_003.jpg
+        | ...
+```
+
+### 2.2 公开数据集
+
+| 数据集名称 | 图片下载地址 | PaddleOCR 标注下载地址                                                         |
+|---|---|---------------------------------------------------------------------|
+| en benchmark(MJ, SJ, IIIT, SVT, IC03, IC13, IC15, SVTP, and CUTE.) | [DTRB](https://github.com/clovaai/deep-text-recognition-benchmark#download-lmdb-dataset-for-traininig-and-evaluation-from-here) | LMDB格式，可直接用[lmdb_dataset.py](../../ppocr/data/lmdb_dataset.py)加载 |
+|ICDAR 2015| <http://rrc.cvc.uab.es/?ch=4&com=downloads> | [train](https://paddleocr.bj.bcebos.com/dataset/rec_gt_train.txt)/ [test](https://paddleocr.bj.bcebos.com/dataset/rec_gt_test.txt) |
+| 多语言数据集 |[百度网盘](https://pan.baidu.com/s/1bS_u207Rm7YbY33wOECKDA) 提取码：frgi | 图片下载地址中已包含 |
+
+#### 2.1 ICDAR 2015
+
+ICDAR 2015 数据集可以在上表中链接下载，用于快速验证。也可以从上表中下载 en benchmark 所需的lmdb格式数据集。
+
+下载完图片后从上表中下载转换好的标注文件。
+
+PaddleOCR 也提供了数据格式转换脚本，可以将ICDAR官网 label 转换为PaddleOCR支持的数据格式。 数据转换工具在 `ppocr/utils/gen_label.py`, 这里以训练集为例：
+
+```bash linenums="1"
+# 将官网下载的标签文件转换为 rec_gt_label.txt
+python gen_label.py --mode="rec" --input_path="{path/of/origin/label}" --output_label="rec_gt_label.txt"
+```
+
+数据样式格式如下，(a)为原始图片,(b)为每张图片对应的 Ground Truth 文本文件：
+
+![](./images/icdar_rec.png)
+
+## 3. 数据存放路径
+
+PaddleOCR训练数据的默认存储路径是 `PaddleOCR/train_data`,如果您的磁盘上已有数据集，只需创建软链接至数据集目录：
+
+```bash linenums="1"
+# linux and mac os
+ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/dataset
+# windows
+mklink /d <path/to/paddle_ocr>/train_data/dataset <path/to/dataset>
+```
diff --git a/docs/datasets/table_datasets.en.md b/docs/datasets/table_datasets.en.md
new file mode 100644
index 0000000..d568c21
--- /dev/null
+++ b/docs/datasets/table_datasets.en.md
@@ -0,0 +1,40 @@
+---
+comments: true
+---
+
+
+# Table Recognition Datasets
+
+Here are the commonly used table recognition datasets, which are being updated continuously. Welcome to contribute datasets~
+
+## Dataset Summary
+
+| dataset | Image download link | PPOCR format annotation download link |
+|---|---|---|
+| PubTabNet |<https://github.com/ibm-aur-nlp/PubTabNet>| jsonl format, which can be loaded directly with [pubtab_dataset.py](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/ppocr/data/pubtab_dataset.py) |
+| TAL Table Recognition Competition Dataset |<https://ai.100tal.com/dataset>| jsonl format, which can be loaded directly with [pubtab_dataset.py](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/ppocr/data/pubtab_dataset.py) |
+| WTW Chinese scene table dataset |<https://github.com/wangwen-whu/WTW-Dataset>| Conversion is required to load with [pubtab_dataset.py](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/ppocr/data/pubtab_dataset.py)|
+
+## 1. PubTabNet
+
+- **Data Introduction**：The training set of the PubTabNet dataset contains 500,000 images and the validation set contains 9000 images. Part of the image visualization is shown below.
+
+    ![](./images/table_PubTabNet_demo/PMC524509_007_00.png)
+
+    ![](./images/table_PubTabNet_demo/PMC535543_007_01.png)
+
+- **illustrate**：When using this dataset, the [CDLA-Permissive](https://cdla.io/permissive-1-0/) protocol is required.
+
+## 2. TAL Table Recognition Competition Dataset
+
+- **Data Introduction**：The training set of the TAL table recognition competition dataset contains 16,000 images. The validation set does not give trainable annotations.
+
+    ![](./images/table_tal_demo/1.jpg)
+
+    ![](./images/table_tal_demo/2.jpg)
+
+## 3. WTW Chinese scene table dataset
+
+- **Data Introduction**：The WTW Chinese scene table dataset consists of two parts: table detection and table data. The dataset contains images of two scenes, scanned and photographed.
+
+    ![img](./images/20210816_210413.gif)
diff --git a/docs/datasets/table_datasets.md b/docs/datasets/table_datasets.md
new file mode 100644
index 0000000..b45137d
--- /dev/null
+++ b/docs/datasets/table_datasets.md
@@ -0,0 +1,43 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+
+# 表格识别数据集
+
+这里整理了常用表格识别数据集，持续更新中，欢迎各位小伙伴贡献数据集～
+
+## 数据集汇总
+
+| 数据集名称 |图片下载地址| PPOCR标注下载地址 |
+|---|---|---|
+| PubTabNet |<https://github.com/ibm-aur-nlp/PubTabNet>| jsonl格式，可直接用[pubtab_dataset.py](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/ppocr/data/pubtab_dataset.py)加载 |
+| 好未来表格识别竞赛数据集 |<https://ai.100tal.com/dataset>| jsonl格式，可直接用[pubtab_dataset.py](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/ppocr/data/pubtab_dataset.py)加载 |
+| WTW中文场景表格数据集 |<https://github.com/wangwen-whu/WTW-Dataset>| 需要进行转换后才能用[pubtab_dataset.py](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/ppocr/data/pubtab_dataset.py)加载 |
+
+## 1. PubTabNet数据集
+
+- **数据简介**：PubTabNet数据集的训练集合中包含50万张图像，验证集合中包含0.9万张图像。部分图像可视化如下所示。
+
+<div align="center">
+    <img src="./images/table_PubTabNet_demo/PMC524509_007_00.png" width="500">
+    <img src="./images/table_PubTabNet_demo/PMC535543_007_01.png" width="500">
+</div>
+
+- **说明**：使用该数据集时，需要遵守[CDLA-Permissive](https://cdla.io/permissive-1-0/)协议。
+
+## 2. 好未来表格识别竞赛数据集
+
+- **数据简介**：好未来表格识别竞赛数据集的训练集合中包含1.6万张图像。验证集未给出可训练的标注。
+
+<div align="center">
+    <img src="./images/table_tal_demo/1.jpg" width="500">
+    <img src="./images/table_tal_demo/2.jpg" width="500">
+</div>
+
+## 3. WTW中文场景表格数据集
+
+- **数据简介**：WTW中文场景表格数据集包含表格检测和表格数据两部分数据，数据集中同时包含扫描和拍照两张场景的图像。
+
+  ![img](./images/20210816_210413.gif)
diff --git a/docs/datasets/vertical_and_multilingual_datasets.en.md b/docs/datasets/vertical_and_multilingual_datasets.en.md
new file mode 100644
index 0000000..e6fcb2c
--- /dev/null
+++ b/docs/datasets/vertical_and_multilingual_datasets.en.md
@@ -0,0 +1,72 @@
+---
+comments: true
+---
+
+
+# Vertical multi-language OCR dataset
+
+Here we have sorted out the commonly used vertical multi-language OCR dataset datasets, which are being updated continuously. We welcome you to contribute datasets ～
+
+- [Chinese urban license plate dataset](#chinese-urban-license-plate-dataset)
+- [Bank credit card dataset](#bank-credit-card-dataset)
+- [Captcha dataset-Captcha](#captcha-dataset-captcha)
+- [multi-language dataset](#multi-language-datasetmulti-lingual-scene-text-detection-and-recognition)
+
+## Chinese urban license plate dataset
+
+- **Data source**：[CCPD](https://github.com/detectRecog/CCPD)
+
+- **Data introduction**: It contains more than 250000 vehicle license plate images and vehicle license plate detection and recognition information labeling. It contains the following license plate image information in different scenes.
+
+    - CCPD-Base: General license plate picture
+    - CCPD-DB: The brightness of license plate area is bright, dark or uneven
+    - CCPD-FN: The license plate is farther or closer to the camera location
+    - CCPD-Rotate: License plate includes rotation (horizontal 20\~50 degrees, vertical-10\~10 degrees)
+    - CCPD-Tilt: License plate includes rotation (horizontal 15\~45 degrees, vertical 15\~45 degrees)
+    - CCPD-Blur: The license plate contains blurring due to camera lens jitter
+    - CCPD-Weather: The license plate is photographed on rainy, snowy or foggy days
+    - CCPD-Challenge: So far, some of the most challenging images in license plate detection and recognition tasks
+    - CCPD-NP: Pictures of new cars without license plates.
+
+      ![](./images/ccpd_demo.png)
+
+- **Download address**
+    - Baidu cloud download address (extracted code is hm0U): [https://pan.baidu.com/s/1i5AOjAbtkwb17Zy-NQGqkw](https://pan.baidu.com/s/1i5AOjAbtkwb17Zy-NQGqkw)
+    - Google drive download address:[https://drive.google.com/file/d/1rdEsCUcIUaYOVRkx5IMTRNA7PcGMmSgc/view](https://drive.google.com/file/d/1rdEsCUcIUaYOVRkx5IMTRNA7PcGMmSgc/view)
+
+## Bank credit card dataset
+
+- **Data source**: [source](https://www.heywhale.com/mw/dataset/5954cf1372ead054a5e25870)
+
+- **Data introduction**: There are three types of training data
+    - 1.Sample card data of China Merchants Bank: including card image data and annotation data, a total of 618 pictures
+    - 2.Single character data: including pictures and annotation data, 37 pictures in total.
+    - 3.There are only other bank cards, no more detailed information, a total of 50 pictures.
+
+    - The demo image is shown as follows. The annotation information is stored in excel, and the demo image below is marked as
+        - Top 8 card number: 62257583
+        - Card type: card of our bank
+        - End of validity: 07/41
+        - Chinese phonetic alphabet of card users: MICHAEL
+
+          ![](./images/cmb_demo.jpg)
+
+- **Download address**: [cmb2017-2.zip](https://cdn.kesci.com/cmb2017-2.zip)
+
+## Captcha dataset-Captcha
+
+- **Data source**: [captcha](https://github.com/lepture/captcha)
+- **Data introduction**: This is a toolkit for data synthesis. You can output captcha images according to the input text. Use the toolkit to generate several demo images as follows.
+
+    ![](./images/captcha_demo.png)
+
+- **Download address**: The dataset is generated and has no download address.
+
+## multi-language dataset(Multi-lingual scene text detection and recognition)
+
+- **Data source**: [source](https://rrc.cvc.uab.es/?ch=15&com=downloads)
+- **Data introduction**: Multi language detection dataset MLT contains both language recognition and detection tasks.
+    - In the detection task, the training set contains 10000 images in 10 languages, and each language contains 1000 training images. The test set contains 10000 images.
+    - In the recognition task, the training set contains 111998 samples.
+- **Download address**: The training set is large and can be downloaded in two parts. It can only be downloaded after registering on the website:
+[source](https://rrc.cvc.uab.es/?ch=15&com=downloads)
diff --git a/docs/datasets/vertical_and_multilingual_datasets.md b/docs/datasets/vertical_and_multilingual_datasets.md
new file mode 100644
index 0000000..181d2bb
--- /dev/null
+++ b/docs/datasets/vertical_and_multilingual_datasets.md
@@ -0,0 +1,65 @@
+---
+comments: true
+---
+
+
+# 垂类多语言OCR数据集
+
+这里整理了常用垂类和多语言OCR数据集，持续更新中，欢迎各位小伙伴贡献数据集～
+
+## 中国城市车牌数据集
+
+- **数据来源**：[CCPD](https://github.com/detectRecog/CCPD)
+- **数据简介**: 包含超过25万张中国城市车牌图片及车牌检测、识别信息的标注。包含以下几种不同场景中的车牌图片信息。
+    - CCPD-Base: 通用车牌图片
+    - CCPD-DB: 车牌区域亮度较亮、较暗或者不均匀
+    - CCPD-FN: 车牌离摄像头拍摄位置相对更远或者更近
+    - CCPD-Rotate: 车牌包含旋转（水平20\~50度，竖直-10\~10度）
+    - CCPD-Tilt: 车牌包含旋转（水平15\~45度，竖直15\~45度）
+    - CCPD-Blur: 车牌包含由于摄像机镜头抖动导致的模糊情况
+    - CCPD-Weather: 车牌在雨天、雪天或者雾天拍摄得到
+    - CCPD-Challenge: 至今在车牌检测识别任务中最有挑战性的一些图片
+    - CCPD-NP: 没有安装车牌的新车图片。
+
+      ![](./images/ccpd_demo.png)
+
+- **下载地址**
+    - 百度云下载地址(提取码是hm0U): [link](https://pan.baidu.com/s/1i5AOjAbtkwb17Zy-NQGqkw)
+    - Google drive下载地址：[link](https://drive.google.com/file/d/1rdEsCUcIUaYOVRkx5IMTRNA7PcGMmSgc/view)
+
+## 银行信用卡数据集
+
+- **数据来源**: [source](https://www.heywhale.com/mw/dataset/5954cf1372ead054a5e25870)
+
+- **数据简介**: 训练数据共提供了三类数据
+    - 1.招行样卡数据： 包括卡面图片数据及标注数据，总共618张图片
+    - 2.单字符数据： 包括图片及标注数据，总共37张图片。
+    - 3.仅包含其他银行卡面，不具有更细致的信息，总共50张图片。
+
+    - demo图片展示如下，标注信息存储在excel表格中，下面的demo图片标注为
+        - 前8位卡号：62257583
+        - 卡片种类：本行卡
+        - 有效期结束：07/41
+        - 卡用户拼音：MICHAEL
+
+    ![](./images/cmb_demo.jpg)
+
+- **下载地址**: [cmb2017-2.zip](https://cdn.kesci.com/cmb2017-2.zip)
+
+## 验证码数据集-Captcha
+
+- **数据来源**: [captcha](https://github.com/lepture/captcha)
+- **数据简介**: 这是一个数据合成的工具包，可以根据输入的文本，输出验证码图片，使用该工具包生成几张demo图片如下：
+
+    ![](./images/captcha_demo.png)
+
+- **下载地址**: 该数据集是生成得到，无下载地址。
+
+## 多语言数据集(Multi-lingual scene text detection and recognition)
+
+- **数据来源**: [source](https://rrc.cvc.uab.es/?ch=15&com=downloads)
+- **数据简介**: 多语言检测数据集MLT同时包含了语种识别和检测任务。
+    - 在检测任务中，训练集包含10000张图片，共有10种语言，每种语言包含1000张训练图片。测试集包含10000张图片。
+    - 在识别任务中，训练集包含111998个样本。
+- **下载地址**: 训练集较大，分2部分下载，需要在网站上注册之后才能下载：
+[link](https://rrc.cvc.uab.es/?ch=15&com=downloads)
diff --git a/docs/images/00006737.jpg b/docs/images/00006737.jpg
new file mode 100644
index 0000000..d7762d2
Binary files /dev/null and b/docs/images/00006737.jpg differ
diff --git a/docs/images/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808-20240708082238739.jpg b/docs/images/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808-20240708082238739.jpg
new file mode 100644
index 0000000..6a5fd84
Binary files /dev/null and b/docs/images/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808-20240708082238739.jpg differ
diff --git a/docs/images/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb-20240708082310650.jpg b/docs/images/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb-20240708082310650.jpg
new file mode 100644
index 0000000..d406a52
Binary files /dev/null and b/docs/images/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb-20240708082310650.jpg differ
diff --git a/docs/images/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a-20240708082247529.png b/docs/images/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a-20240708082247529.png
new file mode 100644
index 0000000..4293434
Binary files /dev/null and b/docs/images/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a-20240708082247529.png differ
diff --git a/docs/images/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f-20240708082316558.jpg b/docs/images/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f-20240708082316558.jpg
new file mode 100644
index 0000000..bdef2ed
Binary files /dev/null and b/docs/images/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f-20240708082316558.jpg differ
diff --git a/docs/images/186094813-3a8e16cc-42e5-4982-b9f4-0134dfb5688d-20240708082323916.png b/docs/images/186094813-3a8e16cc-42e5-4982-b9f4-0134dfb5688d-20240708082323916.png
new file mode 100644
index 0000000..44845fb
Binary files /dev/null and b/docs/images/186094813-3a8e16cc-42e5-4982-b9f4-0134dfb5688d-20240708082323916.png differ
diff --git a/docs/images/197464552-69de557f-edff-4c7f-acbf-069df1ba097f-20240708082253634.png b/docs/images/197464552-69de557f-edff-4c7f-acbf-069df1ba097f-20240708082253634.png
new file mode 100644
index 0000000..7b4c271
Binary files /dev/null and b/docs/images/197464552-69de557f-edff-4c7f-acbf-069df1ba097f-20240708082253634.png differ
diff --git a/docs/images/Banner.png b/docs/images/Banner.png
new file mode 100644
index 0000000..456c928
Binary files /dev/null and b/docs/images/Banner.png differ
diff --git a/docs/images/Banner_cn.png b/docs/images/Banner_cn.png
new file mode 100644
index 0000000..3d372d7
Binary files /dev/null and b/docs/images/Banner_cn.png differ
diff --git a/docs/images/Banner_ja.png b/docs/images/Banner_ja.png
new file mode 100644
index 0000000..5f3047a
Binary files /dev/null and b/docs/images/Banner_ja.png differ
diff --git a/docs/images/PP-OCRv3-pic001.jpg b/docs/images/PP-OCRv3-pic001.jpg
new file mode 100644
index 0000000..c35936c
Binary files /dev/null and b/docs/images/PP-OCRv3-pic001.jpg differ
diff --git a/docs/images/PP-OCRv3-pic002.jpg b/docs/images/PP-OCRv3-pic002.jpg
new file mode 100644
index 0000000..e5ad6a4
Binary files /dev/null and b/docs/images/PP-OCRv3-pic002.jpg differ
diff --git a/docs/images/PP-OCRv3-pic003.jpg b/docs/images/PP-OCRv3-pic003.jpg
new file mode 100644
index 0000000..dc02429
Binary files /dev/null and b/docs/images/PP-OCRv3-pic003.jpg differ
diff --git a/docs/images/en_1.png b/docs/images/en_1.png
new file mode 100644
index 0000000..3624561
Binary files /dev/null and b/docs/images/en_1.png differ
diff --git a/docs/images/en_2.png b/docs/images/en_2.png
new file mode 100644
index 0000000..d2df855
Binary files /dev/null and b/docs/images/en_2.png differ
diff --git a/docs/images/en_3-0398013.png b/docs/images/en_3-0398013.png
new file mode 100644
index 0000000..baf146c
Binary files /dev/null and b/docs/images/en_3-0398013.png differ
diff --git a/docs/images/en_3.png b/docs/images/en_3.png
new file mode 100644
index 0000000..baf146c
Binary files /dev/null and b/docs/images/en_3.png differ
diff --git a/docs/images/japan_2.jpg b/docs/images/japan_2.jpg
new file mode 100644
index 0000000..076ced9
Binary files /dev/null and b/docs/images/japan_2.jpg differ
diff --git a/docs/images/korean_1.jpg b/docs/images/korean_1.jpg
new file mode 100644
index 0000000..f93de40
Binary files /dev/null and b/docs/images/korean_1.jpg differ
diff --git a/docs/images/ppocrv4.png b/docs/images/ppocrv4.png
new file mode 100644
index 0000000..6be449c
Binary files /dev/null and b/docs/images/ppocrv4.png differ
diff --git a/docs/images/ppocrv4_en.jpg b/docs/images/ppocrv4_en.jpg
new file mode 100644
index 0000000..3e3fde4
Binary files /dev/null and b/docs/images/ppocrv4_en.jpg differ
diff --git a/docs/images/ppstructure-20240708082235651.gif b/docs/images/ppstructure-20240708082235651.gif
new file mode 100644
index 0000000..bff836e
Binary files /dev/null and b/docs/images/ppstructure-20240708082235651.gif differ
diff --git a/docs/images/test_add_91.jpg b/docs/images/test_add_91.jpg
new file mode 100644
index 0000000..b5ded6e
Binary files /dev/null and b/docs/images/test_add_91.jpg differ
diff --git a/docs/index.en.md b/docs/index.en.md
new file mode 100644
index 0000000..24564ff
--- /dev/null
+++ b/docs/index.en.md
@@ -0,0 +1,119 @@
+---
+comments: true
+hide:
+  - navigation
+  - toc
+---
+
+<div align="center">
+  <p>
+    <a href="https://paddlepaddle.github.io/PaddleOCR/latest/index.html" target="_blank">
+      <img width="100%" src="https://raw.githubusercontent.com/PaddlePaddle/PaddleOCR/main/docs/images/Banner.png" alt="PaddleOCR Banner">
+    </a>
+  </p>
+</div>
+
+Since its initial release, PaddleOCR has gained widespread acclaim across academia, industry, and research communities, thanks to its cutting-edge algorithms and proven performance in real-world applications. It’s already powering popular open-source projects like Umi-OCR, OmniParser, MinerU, and RAGFlow, making it the go-to OCR toolkit for developers worldwide.
+
+On January 29, 2026, PaddleOCR open-sourced the advanced and efficient document parsing model PaddleOCR-VL-1.5. PaddleOCR-VL-1.5 is a new iterative version of the PaddleOCR-VL series. Based on comprehensive optimization of the core capabilities of version 1.0, the model achieves 94.5% accuracy on the authoritative document parsing benchmark OmniDocBench v1.5, surpassing top global general-purpose large models and document parsing–specific models. PaddleOCR-VL-1.5 innovatively supports irregular-shaped bounding box localization of document elements, enabling excellent performance in real-world application scenarios such as scanning, skew, warping, screen-photography, and complex illumination, achieving comprehensive SOTA performance. In addition, the model further integrates seal recognition and spotting tasks, with key metrics continuing to lead mainstream models.
+
+**On May 28, 2026, PaddleOCR open-sourced PaddleOCR-VL-1.6, the latest version in the PaddleOCR-VL series.** With an accuracy of 96.3%, PaddleOCR-VL-1.6 once again set a new benchmark on OmniDocBench v1.6, while also achieving new state-of-the-art (SOTA) results on OmniDocBench v1.5 and Real5-OmniDocBench. It delivers industry-leading performance in text, formula, and table recognition across both open-source and proprietary solutions. In addition, the model shows substantial improvements in ancient document and rare character recognition, as well as significantly enhanced capabilities in multiple scenarios such as seal recognition, spotting, and chart understanding. PaddleOCR-VL-1.6 further upgrades the VLM component (PaddleOCR-VL-1.6-0.9B) on top of PaddleOCR-VL-1.5 and continues to use PP-DocLayoutV3 for layout analysis. The model architecture remains fully consistent with PaddleOCR-VL-1.5, enabling seamless migration at zero cost.
+
+**On June 11, 2026, PaddleOCR released PP-OCRv6, the next-generation universal OCR system.** Built on the newly designed PPLCNetV4 unified backbone, PP-OCRv6 offers tiny/small/medium tiers (1.5M–34.5M parameters). In terms of accuracy, the medium tier achieves +5.1% recognition and +4.6% detection over PP-OCRv5_server, surpassing mainstream Vision-Language Models with only 34.5M parameters. For multilingual support, a single model covers 50 languages (Chinese, English, Japanese, and 46 Latin-script languages) without switching models. PP-OCRv6 significantly improves recognition in specialized scenarios such as digital displays, dot-matrix characters, tire prints, and industrial text that are traditionally underserved by VLMs. In terms of speed, PP-OCRv6_medium achieves 5.2× speedup over PP-OCRv5_server on Intel Xeon CPU with OpenVINO (1.40s vs 7.30s), the tiny tier reaches 6.1× speedup on Apple M4 (0.96s vs 5.82s), and only 0.13s on A100 GPU. See [PP-OCRv6 Technical Documentation](./version3.x/algorithm/PP-OCRv6/PP-OCRv6.en.md) for details.
+
+You can use it online on the [PaddleOCR official website](https://www.paddleocr.com) or call the model API.
+
+
+**Major Features in PaddleOCR 3.x:**
+
+- **PaddleOCR-VL - Multilingual Document Parsing via a 0.9B VLM**  
+  **The SOTA and resource-efficient model tailored for document parsing**, that supports 109 languages and excels in recognizing complex elements (e.g., text, tables, formulas, and charts), while maintaining minimal resource consumption.
+
+- **PP-OCRv6 — Universal Multi-Language Text Recognition**
+  **Single model supports 50 languages** (Chinese, English, Japanese, and 46 Latin-script languages), with **+4.6% detection** and **+5.1% recognition** accuracy over PP-OCRv5, surpassing mainstream Vision-Language Models. Medium tier achieves 5.2× CPU speedup end-to-end, covering specialized scenarios like digital displays, dot-matrix, and industrial characters. Three tiers (tiny/small/medium, 1.5M–34.5M params) for edge-to-server deployment.
+
+- **PP-StructureV3 — Complex Document Parsing**  
+  Intelligently converts complex PDFs and document images into **Markdown and JSON files that preserve original structure**. **Outperforms** numerous commercial solutions in public benchmarks. **Perfectly maintains document layout and hierarchical structure**.
+
+- **PP-ChatOCRv4 — Intelligent Information Extraction**  
+  Natively integrates ERNIE 4.5 to **precisely extract key information** from massive documents, with 15% accuracy improvement over previous generation. Makes documents "**understand**" your questions and provide accurate answers.
+
+> 💡 Tips
+> 
+> PaddleOCR's free API now supports up to **20,000** pages of document parsing per day, enabling large-scale PDF file processing, along with MCP and Skills services. For more details, please visit [PaddleOCR Official Website](https://www.paddleocr.com).
+
+In addition to its strong model library, PaddleOCR 3.0 also provides easy-to-use tools covering model training, inference, and serving, helping developers bring AI applications into production more efficiently.
+
+**In addition, PaddleOCR provides official [Agent Skills](./version3.x/integrations/skills.en.md) for invoking text recognition, document parsing, and related capabilities in Skills-enabled AI apps.**
+
+<div align="center">
+  <p>
+    <a href="https://paddlepaddle.github.io/PaddleOCR/latest/index.html" target="_blank">
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Arch.jpg" alt="PaddleOCR Architecture"></a>
+  </p>
+</div>
+
+You can [Quick Start](./quick_start.en.md) directly, find comprehensive documentation in the [PaddleOCR Docs](https://paddlepaddle.github.io/PaddleOCR/main/index.html), get support via [Github Issues](https://github.com/PaddlePaddle/PaddleOCR/issues), and explore our OCR courses on [OCR courses on AIStudio](https://aistudio.baidu.com/course/introduce/25207).
+
+**Special Note**: PaddleOCR 3.x introduces several significant interface changes. **Old code written based on PaddleOCR 2.x is likely incompatible with PaddleOCR 3.x**. Please ensure that the documentation you are reading matches the version of PaddleOCR you are using. [This document](./update/upgrade_notes.en.md) explains the reasons for the upgrade and the major changes from PaddleOCR 2.x to 3.x.
+
+## 🔄 Quick Overview of Execution Results
+
+
+### PP-OCRv5
+
+<div align="center">
+  <p>
+       <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-OCRv5_demo.gif" alt="PP-OCRv5 Demo">
+  </p>
+</div>
+
+
+
+### PP-StructureV3
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-StructureV3_demo.gif" alt="PP-StructureV3 Demo">
+  </p>
+</div>
+
+### PaddleOCR-VL
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PaddleOCR-VL_demo.gif" alt="PP-StructureV3 Demo">
+  </p>
+</div>
+
+## 👩‍👩‍👧‍👦 PaddleOCR OCEAN Ecosystem Alliance
+
+The lead in single-point technology is just the beginning — the prosperity of the ecosystem is where long-term value truly lies. To better serve global developers and industry scenarios with OCR and document intelligence technologies, we are officially launching the **PaddleOCR OCEAN Ecosystem Alliance**.
+
+The alliance name **OCEAN** embodies five core pillars:
+
+* **O**pen Source – Open source as the foundation
+* **C**ommunity – Community-driven
+* **E**cosystem – Shared ecosystem success
+* **A**pplication – Real-world application
+* **N**etwork – Networked collaboration
+
+**Positioning**: An ecosystem alliance centered on open-source co-building, open to global upstream and downstream partners in OCR and document intelligence. The alliance involves no commercial exclusivity and does not interfere with partners' independent business decisions. It focuses on technical collaboration, community engagement, and mutual influence expansion. Guided by the core principles of openness, symbiosis, and shared success, it brings together developers, platform providers, and application builders to jointly advance the full-chain application of OCR technology and ecosystem prosperity. The alliance is committed to achieving dual growth in full-chain application scale and the number of derivative projects, enabling developers and users worldwide to share in the dividends of OCR technology advancement.
+
+**Join Us: Walk into the Deep End with Like-Minded Partners**
+
+The PaddleOCR OCEAN Ecosystem Alliance is open to global partners across the OCR and document intelligence value chain. We firmly believe: **the value of an ecosystem lies not in quantity, but in quality.**
+
+We look forward to welcoming partners who:
+
+* **Genuinely embrace the open-source spirit** and are willing to co-build and share with an open mindset
+* **Have the willingness and capability to contribute consistently** — whether through code, use case scenarios, or platform integration
+* **Are committed to growing together with the alliance** — not chasing short-term traffic, but cultivating long-term value
+
+**The alliance is not a hall of fame — it is a rallying call for those who take action.**
+
+We will carefully evaluate every application, giving priority to partners who have already taken action within the PaddleOCR ecosystem or have a clear co-building plan in place. We do not pursue being "large and all-encompassing." Instead, we seek to work hand in hand with truly like-minded organizations and individuals, diving deep into the frontiers of OCR together.
+
+If you resonate with the above vision, we welcome you to reach out through the following channel:
+
+* Send an email to paddleocr@baidu.com with a brief introduction of your collaboration with PaddleOCR or your co-building plan.
diff --git a/docs/index.md b/docs/index.md
new file mode 100644
index 0000000..d339d89
--- /dev/null
+++ b/docs/index.md
@@ -0,0 +1,117 @@
+---
+comments: true
+hide:
+  - navigation
+  - toc
+---
+
+<div align="center">
+  <p>
+    <a href="https://paddlepaddle.github.io/PaddleOCR/latest/index.html" target="_blank">
+      <img width="100%" src="./images/Banner_cn.png" alt="PaddleOCR Banner"></a>
+  </p>
+</div>
+
+
+PaddleOCR自发布以来凭借学术前沿算法和产业落地实践，受到了产学研各方的喜爱，并被广泛应用于众多知名开源项目，例如：Umi-OCR、OmniParser、MinerU、RAGFlow等，已成为广大开发者心中的开源OCR领域的首选工具。2025年5月20日，飞桨团队发布**PaddleOCR 3.0**，全面适配[飞桨框架3.0](https://github.com/PaddlePaddle/Paddle)正式版，进一步**提升文字识别精度**，支持**多文字类型识别**和**手写体识别**，满足大模型应用对**复杂文档高精度解析**的旺盛需求，结合**文心大模型4.5**显著提升关键信息抽取精度，并新增**对昆仑芯、昇腾等国产硬件**的支持。
+
+2026 年 1 月 29 日，PaddleOCR 开源了先进、高效的文档解析模型 PaddleOCR-VL-1.5。PaddleOCR-VL-1.5 是 PaddleOCR-VL 系列的全新迭代版本。在全面优化 1.0 版本核心能力的基础上，该模型在文档解析权威评测集 OmniDocBench v1.5 上斩获了 94.5% 的高精度，超越了全球的顶尖通用大模型及文档解析专用模型。PaddleOCR-VL-1.5 创新性地支持了文档元素的异形框定位，使得 PaddleOCR-VL-1.5 在扫描、倾斜、弯折、屏幕拍摄及复杂光照等真实落地场景中均表现卓越，实现了全面的 SOTA。此外，模型进一步集成了印章识别与文本检测识别任务，关键指标持续领跑主流模型。
+
+**2026 年 5 月 28 日，PaddleOCR 开源了 PaddleOCR-VL-1.6，作为 PaddleOCR-VL 系列的最新版本。** PaddleOCR-VL-1.6 以 96.3% 精度再次刷新评测集 OmniDocBench v1.6，并在 OmniDocBench v1.5、Real5-OmniDocBench 上同步达到全新 SOTA，文本、公式、表格识别全面领先开源与闭源方案。此外，模型在古籍、生僻字识别上大幅提升，印章、spotting、图表识别等多场景能力也显著增强。PaddleOCR-VL-1.6 在 PaddleOCR-VL-1.5 的基础上进一步升级了 VLM 组件（PaddleOCR-VL-1.6-0.9B），并继续采用 PP-DocLayoutV3 作为版面分析模型；模型结构与 PaddleOCR-VL-1.5 完全一致，支持零成本无缝迁移。
+
+**2026 年 6 月 11 日，PaddleOCR 发布 PP-OCRv6，作为 PP-OCR 系列的最新版本。** PP-OCRv6 基于全新设计的 PPLCNetV4 统一骨干网络，提供 tiny/small/medium 三档模型（1.5M~34.5M 参数）。在精度方面，medium 档相比 PP-OCRv5_server 识别精度提升 5.1%、检测精度提升 4.6%，以仅 34.5M 参数的规模超越 Qwen3-VL-235B、GPT-5.5 等大型视觉语言模型。在多语种方面，单一模型统一支持中文、英文、日文及 46 种拉丁语系共 50 种语言，无需为不同语种切换模型。在场景覆盖方面，PP-OCRv6 大幅提升了数码管、点阵字符、轮胎印字、工业字符等传统 VLM 难以覆盖的专业场景识别能力。在速度方面，PP-OCRv6_medium 在 Intel Xeon CPU OpenVINO 下相比 PP-OCRv5_server 加速 5.2×（1.40s vs 7.30s），tiny 档在纯前端浏览器环境的极低算力下，直接轰出了单图 97ms 的惊人成绩。
+
+您可以在 [PaddleOCR官网](https://www.paddleocr.com) 在线使用或者调用该模型的API。
+
+**PaddleOCR 3.x 核心特色能力：**
+
+
+- **PaddleOCR-VL - 通过 0.9B 超紧凑视觉语言模型增强多语种文档解析**  
+  **面向文档解析的 SOTA 且资源高效的模型**, 支持 109 种语言，在复杂元素（如文本、表格、公式和图表）识别方面表现出色，同时资源消耗极低。
+
+- **PP-OCRv6 — 全场景多语言文字识别**
+  **单模型支持 50 种语言**（中、英、日及 46 种拉丁语系），检测精度提升 **4.6%**、识别精度提升 **5.1%**，超越主流视觉语言大模型。medium 档位模型端到端 CPU 推理加速 5.2×，覆盖数码屏/点阵/工业字符等专业场景。三档模型（tiny/small/medium，1.5M~34.5M 参数）满足端侧到服务端全场景部署。
+
+- **PP-StructureV3 — 复杂文档解析**  
+  将复杂PDF和文档图像智能转换为保留**原始结构的Markdown文件和JSON**文件，在公开评测中**领先**众多商业方案。**完美保持文档版式和层次结构**。
+
+- **PP-ChatOCRv4 — 智能信息抽取**  
+  原生集成ERNIE 4.5，从海量文档中**精准提取关键信息**，精度较上一代提升15个百分点。让文档"**听懂**"您的问题并给出准确答案。
+
+> 💡 Tips
+> 
+> PaddleOCR 官网免费 API 调用现已将每日文档解析上限提升至 20,000 页，支持大批量 PDF 文件解析，同时提供 MCP 及 Skills 服务。更多详情请参见 [PaddleOCR 官网](https://www.paddleocr.com)。
+
+PaddleOCR 3.0 除了提供优秀的模型库外，还提供覆盖模型训练、推理和服务化部署等环节的易用工具，方便开发者快速落地 AI 应用。
+
+**此外，PaddleOCR 还提供官方 [Agent Skills](./version3.x/integrations/skills.md)，便于在支持 Skills 的 AI 应用中调用文字识别、文档解析等能力。**
+
+<div align="center">
+  <p>
+    <a href="https://paddlepaddle.github.io/PaddleOCR/latest/index.html" target="_blank">
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Arch_cn.jpg" alt="PaddleOCR Architecture"></a>
+  </p>
+</div>
+
+您可直接[快速开始](./quick_start.md)，或查阅完整的 [PaddleOCR 文档](https://paddlepaddle.github.io/PaddleOCR/main/index.html)，或通过 [Github Issues](https://github.com/PaddlePaddle/PaddleOCR/issues) 获取支持，或在 [AIStudio 课程平台](https://aistudio.baidu.com/course/introduce/25207) 探索我们的 OCR 课程。
+
+**特别说明**：PaddleOCR 3.x 引入了多项重要的接口变动，**基于 PaddleOCR 2.x 编写的旧代码很可能无法使用 PaddleOCR 3.x 运行**。请确保您阅读的文档与实际使用的 PaddleOCR 版本匹配。[此文档](./update/upgrade_notes.md) 阐述了升级原因及 PaddleOCR 2.x 到 PaddleOCR 3.x 的主要变更。
+
+## 🔄 快速一览运行效果
+
+### PP-OCRv5
+
+<div align="center">
+  <p>
+       <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-OCRv5_demo.gif" alt="PP-OCRv5 Demo">
+  </p>
+</div>
+
+
+
+### PP-StructureV3
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-StructureV3_demo.gif" alt="PP-StructureV3 Demo">
+  </p>
+</div>
+
+### PaddleOCR-VL
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PaddleOCR-VL_demo.gif" alt="PP-StructureV3 Demo">
+  </p>
+</div>
+
+
+## 👩‍👩‍👧‍👦 PaddleOCR OCEAN 生态联盟
+
+单点技术的领先只是开始，生态的繁荣才是长期价值所在。为了让 OCR 及文档智能技术更好地服务于全球开发者和产业场景，我们正式发起 PaddleOCR OCEAN 生态联盟。
+
+联盟名称 OCEAN 蕴含五大核心：
+
+  * **O**pen Source – 开源为本   
+  * **C**ommunity – 社区驱动
+  * **E**cosystem – 生态共赢
+  * **A**pplication – 应用落地
+  * **N**etwork – 网络互联
+
+**定位**：以开源共建为核心的生态联盟，面向全球OCR及文档智能上下游伙伴，不涉及商业排他、不干预伙伴独立商业选择，聚焦技术共建、社区联动与影响力互换。以开放、共生、共赢为核心理念，汇聚开发者、平台方、应用方，共同推动 OCR 技术的全链条应用与生态繁荣。联盟致力于实现生态全链条应用规模与衍生项目数量的双重提升，让全球开发者与用户共享 OCR 技术发展的红利。
+
+**加入我们：与志同道合者，共赴深水区**
+
+PaddleOCR OCEAN生态联盟面向全球OCR及文档智能上下游伙伴开放。我们深知：**生态的价值不在于数量，而在于质量**。
+我们期待这样的伙伴加入：
+
+  * **真心认同开源精神**，愿意以开放的心态共建、共享
+  * **具备持续贡献的意愿与能力**，无论是代码、场景案例还是平台集成
+  * **愿意与联盟共同成长**，不追求短期流量，而是深耕长期价值
+
+**联盟不是荣誉墙，而是行动者的集结号。**
+
+我们将对每一份申请进行审慎评估，优先邀请那些已经在PaddleOCR生态中有所行动、或具备明确共建规划的伙伴。我们不追求“大而全”，而是希望与真正志同道合的机构和个人，在OCR深水区携手深耕。
+如果您符合以上理念，欢迎通过以下方式与我们联系：
+
+  * 发送邮件至 [paddleocr@baidu.com](mailto:paddleocr@baidu.com)，简要介绍您与PaddleOCR的合作情况或共建计划
diff --git a/docs/index/index.en.md b/docs/index/index.en.md
new file mode 100644
index 0000000..23c8ad3
--- /dev/null
+++ b/docs/index/index.en.md
@@ -0,0 +1,178 @@
+---
+comments: true
+hide:
+  - navigation
+  - toc
+---
+
+<div align="center">
+ <img src="https://github.com/PaddlePaddle/PaddleOCR/releases/download/v2.9.1/PaddleOCR_log.png" align="middle" width = "600"/>
+  <p align="center">
+      <a href="https://discord.gg/z9xaRVjdbD"><img src="https://img.shields.io/badge/Chat-on%20discord-7289da.svg?sanitize=true" alt="Chat"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/LICENSE"><img src="https://img.shields.io/badge/license-Apache%202-dfd.svg"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/releases"><img src="https://img.shields.io/github/v/release/PaddlePaddle/PaddleOCR?color=ffa"></a>
+      <a href=""><img src="https://img.shields.io/badge/python-3.7+-aff.svg"></a>
+      <a href=""><img src="https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-pink.svg"></a>
+      <a href="https://pypi.org/project/PaddleOCR/"><img src="https://img.shields.io/pypi/dm/PaddleOCR?color=9cf"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/stargazers"><img src="https://img.shields.io/github/stars/PaddlePaddle/PaddleOCR?color=ccf"></a>
+  </p>
+</div>
+
+## Introduction
+
+PaddleOCR aims to create multilingual, awesome, leading, and practical OCR tools that help users train better models and apply them into practice.
+
+## 🚀 Community
+
+PaddleOCR is being oversight by a [PMC](https://github.com/PaddlePaddle/PaddleOCR/issues/12122). Issues and PRs will be reviewed on a best-effort basis. For a complete overview of PaddlePaddle community, please visit [community](https://github.com/PaddlePaddle/community).
+
+⚠️ Note: The [Issues](https://github.com/PaddlePaddle/PaddleOCR/issues) module is only for reporting program 🐞 bugs, for the rest of the questions, please move to the [Discussions](https://github.com/PaddlePaddle/PaddleOCR/discussions). Please note that if the Issue mentioned is not a bug, it will be moved to the Discussions module.
+
+## 📣 Recent updates
+
+- **🔥🔥2025.3.7 release PaddleOCR v2.10, including**:
+
+  - **12 new self-developed single models:**
+    - **[Layout Detection](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/layout_detection.html)** series with 3 models: PP-DocLayout-L, PP-DocLayout-M, PP-DocLayout-S, supporting prediction of 23 common layout categories. High-quality layout detection for various document types such as papers, reports, exams, books, magazines, contracts, newspapers in both English and Chinese. **mAP@0.5 reaches up to 90.4%, lightweight models can process over 100 pages of document images per second end-to-end.**
+    - **[Formula Recognition](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/formula_recognition.html)** series with 2 models: PP-FormulaNet-L, PP-FormulaNet-S, supporting 50,000 common LaTeX vocabulary, capable of recognizing complex printed and handwritten formulas. **PP-FormulaNet-L has 6 percentage points higher accuracy than models of the same level, and PP-FormulaNet-S is 16 times faster than models with similar accuracy.**
+    - **[Table Structure Recognition](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/table_structure_recognition.html)** series with 2 models: SLANeXt_wired, SLANeXt_wireless. A newly developed table structure recognition model, supporting structured prediction for both wired and wireless tables. Compared to SLANet_plus, SLANeXt shows significant improvement in table structure, **with 6 percentage points higher accuracy on internal high-difficulty table recognition evaluation sets.**
+    - **[Table Classification](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/table_classification.html)** series with 1 model: PP-LCNet_x1_0_table_cls, an ultra-lightweight classification model for both wired and wireless tables.
+    - **[Table Cell Detection](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/table_cells_detection.html)** series with 2 models: RT-DETR-L_wired_table_cell_det, RT-DETR-L_wireless_table_cell_det, supporting cell detection in both wired and wireless tables. These can be combined with SLANeXt_wired, SLANeXt_wireless, text detection, and text recognition modules for end-to-end table prediction. (See the newly added Table Recognition v2 pipeline)
+    - **[Text Recognition](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/text_recognition.html)** series with 1 model: PP-OCRv4_server_rec_doc, **supports over 15,000 characters, with a broader text recognition range, additionally improving the recognition accuracy of certain texts. The accuracy is more than 3 percentage points higher than PP-OCRv4_server_rec on internal datasets.**
+    - **[Text Line Orientation Classification](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/text_recognition.html)** series with 1 model: PP-LCNet_x0_25_textline_ori, **an ultra-lightweight text line orientation classification model with only 0.3M storage.**
+
+   - **4 high-value multi-model combination solutions:**
+     - **[Document Image Preprocessing Pipeline](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/doc_preprocessor.html)**: Achieve correction of distortion and orientation in document images through the combination of ultra-lightweight models.
+     - **[Layout Parsing v2 Pipeline](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/layout_parsing.html)**: Combines multiple self-developed different types of OCR models to optimize complex layout reading order, achieving end-to-end conversion of various complex PDF files to Markdown and JSON files. The conversion effect is better than other open-source solutions in multiple document scenarios. It can provide high-quality data production capabilities for large model training and application.
+     - **[Table Recognition v2 Pipeline](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/table_recognition_v2.html)**: **Provides better table recognition capabilities.** By combining table classification module, table cell detection module, table structure recognition module, text detection module, text recognition module, etc., it achieves prediction of various styles of tables. Users can customize and finetune any module to improve the effect of vertical tables.
+     - **[PP-ChatOCRv4-doc Pipeline](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction_v4.html)**: Based on PP-ChatOCRv3-doc, **integrating multi-modal large models, optimizing Prompt and multi-model combination post-processing logic. It effectively addresses common complex document information extraction challenges such as layout analysis, rare characters, multi-page PDFs, tables, and seal recognition, achieving 15 percentage points higher accuracy than PP-ChatOCRv3-doc. The large model upgrades local deployment capabilities, providing a standard OpenAI interface, supporting calls to locally deployed large models like DeepSeek-R1.**
+
+- **🔥 2024.10.18 release PaddleOCR v2.9, including**:
+    - PaddleX, an All-in-One development tool based on PaddleOCR's advanced technology, supports low-code full-process development capabilities in the OCR field:
+        - 🎨 [**Rich Model One-Click Call**](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.10/docs/paddlex/quick_start.en.md): Integrates **17 models** related to text image intelligent analysis, general OCR, general layout parsing, table recognition, formula recognition, and seal recognition into 6 pipelines, which can be quickly experienced through a simple **Python API one-click call**. In addition, the same set of APIs also supports a total of **200+ models** in image classification, object detection, image segmentation, and time series forecasting, forming 20+ single-function modules, making it convenient for developers to use **model combinations**.
+
+        - 🚀 [**High Efficiency and Low barrier of entry**](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.10/docs/paddlex/overview.en.md): Provides two methods based on **unified commands** and **GUI** to achieve simple and efficient use, combination, and customization of models. Supports multiple deployment methods such as **high-performance inference, service-oriented deployment, and on-device deployment**. Additionally, for various mainstream hardware such as **NVIDIA GPU, Kunlunxin XPU, Ascend NPU, Cambricon MLU, and Haiguang DCU**, models can be developed with **seamless switching**.
+
+    - Supports [PP-ChatOCRv3-doc](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction_en.md), [high-precision layout detection model based on RT-DETR](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/layout_detection_en.md) and [high-efficiency layout area detection model based on PicoDet](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/layout_detection_en.md), [high-precision table structure recognition model](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/table_structure_recognition_en.md), text image unwarping model [UVDoc](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/text_image_unwarping_en.md), formula recognition model [LatexOCR](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/formula_recognition_en.md), and [document image orientation classification model based on PP-LCNet](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/doc_img_orientation_classification_en.md).
+
+- **🔥2023.8.7 Release PaddleOCR[release/2.7](https://github.com/PaddlePaddle/PaddleOCR/tree/release/2.7)**
+
+    - Release [PP-OCRv4](../version2.x/ppocr/blog/PP-OCRv4_introduction.md), support mobile version and server version
+
+        - PP-OCRv4-mobile：When the speed is comparable, the effect of the Chinese scene is improved by 4.5% compared with PP-OCRv3, the English scene is improved by 10%, and the average recognition accuracy of the 80-language multilingual model is increased by more than 8%.
+        - PP-OCRv4-server：Release the OCR model with the highest accuracy at present, the detection model accuracy increased by 4.9% in the Chinese and English scenes, and the recognition model accuracy increased by 2%
+        refer [quickstart](../version2.x/ppocr/quick_start.en.md) quick use by one line command, At the same time, the whole process of model training, reasoning, and high-performance deployment can also be completed with few code in the [General OCR Industry Solution](https://aistudio.baidu.com/aistudio/modelsdetail?modelId=286) in PaddleX.
+
+    - Release[PP-ChatOCR](https://aistudio.baidu.com/aistudio/modelsdetail?modelId=332), a new scheme for extracting key information of general scenes using PP-OCR model and ERNIE LLM.
+
+- 🔨**2022.11 Add implementation of [4 cutting-edge algorithms](../version2.x/algorithm/overview.en.md)**：Text Detection [DRRG](../version2.x/algorithm/text_detection/algorithm_det_drrg.en.md),  Text Recognition [RFL](../version2.x/algorithm/text_recognition/algorithm_rec_rfl.en.md), Image Super-Resolution [Text Telescope](../version2.x/algorithm/super_resolution/algorithm_sr_telescope.en.md)，Handwritten Mathematical Expression Recognition [CAN](../version2.x/algorithm/formula_recognition/algorithm_rec_can.en.md)
+
+- **2022.10 release [optimized JS version PP-OCRv3 model](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.10/deploy/paddlejs/README.md)** with 4.3M model size, 8x faster inference time, and a ready-to-use web demo
+
+    - 💥 **Live Playback: Introduction to PP-StructureV2 optimization strategy**. Scan [the QR code below](#community) using WeChat, follow the PaddlePaddle official account and fill out the questionnaire to join the WeChat group, get the live link and 20G OCR learning materials (including PDF2Word application, 10 models in vertical scenarios, etc.)
+
+- **🔥2022.8.24 Release PaddleOCR [release/2.6](https://github.com/PaddlePaddle/PaddleOCR/tree/release/2.6)**
+
+    - Release [PP-StructureV2](../version2.x/ppstructure/)，with functions and performance fully upgraded, adapted to Chinese scenes, and new support for [Layout Recovery](../version2.x/ppstructure/model_train/recovery_to_doc.en.md) and **one line command to convert PDF to Word**;
+    - [Layout Analysis](../version2.x/ppstructure/model_train/train_layout.en.md) optimization: model storage reduced by 95%, while speed increased by 11 times, and the average CPU time-cost is only 41ms;
+    - [Table Recognition](../version2.x/ppstructure/model_train/train_table.en.md) optimization: 3 optimization strategies are designed, and the model accuracy is improved by 6% under comparable time consumption;
+    - [Key Information Extraction](../version2.x/ppstructure/model_train/train_kie.en.md) optimization：a visual-independent model structure is designed, the accuracy of semantic entity recognition is increased by 2.8%, and the accuracy of relation extraction is increased by 9.1%.
+
+- **🔥2022.8 Release [OCR scene application collection](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.8/applications/README_en.md)**
+
+    - Release **9 vertical models** such as digital tube, LCD screen, license plate, handwriting recognition model, high-precision SVTR model, etc, covering the main OCR vertical applications in general, manufacturing, finance, and transportation industries.
+
+- **2022.8 Add implementation of [8 cutting-edge algorithms](../version2.x/algorithm/overview.en.md)**
+
+    - Text Detection: [FCENet](../version2.x/algorithm/text_detection/algorithm_det_fcenet.en.md), [DB++](../version2.x/algorithm/text_detection/algorithm_det_db.en.md)
+    - Text Recognition: [ViTSTR](../version2.x/algorithm/text_recognition/algorithm_rec_vitstr.en.md), [ABINet](../version2.x/algorithm/text_recognition/algorithm_rec_abinet.en.md), [VisionLAN](../version2.x/algorithm/text_recognition/algorithm_rec_visionlan.en.md), [SPIN](../version2.x/algorithm/text_recognition/algorithm_rec_spin.en.md), [RobustScanner](../version2.x/algorithm/text_recognition/algorithm_rec_robustscanner.en.md)
+    - Table Recognition: [TableMaster](../version2.x/algorithm/table_recognition/algorithm_table_master.en.md)
+- **2022.5.9 Release PaddleOCR [release/2.5](https://github.com/PaddlePaddle/PaddleOCR/tree/release/2.5)**
+
+    - Release [PP-OCRv3](../version2.x/ppocr/overview.en.md#pp-ocrv3): With comparable speed, the effect of Chinese scene is further improved by 5% compared with PP-OCRv2, the effect of English scene is improved by 11%, and the average recognition accuracy of 80 language multilingual models is improved by more than 5%.
+    - Release [PPOCRLabelv2](https://github.com/PFCCLab/PPOCRLabel): Add the annotation function for table recognition task, key information extraction task and irregular text image.
+    - Release interactive e-book [*"Dive into OCR"*](../version2.x/ppocr/blog/ocr_book.en.md), covers the cutting-edge theory and code practice of OCR full stack technology.
+- [more](../update/update.en.md)
+
+## 🌟 Features
+
+PaddleOCR support a variety of cutting-edge algorithms related to OCR, and developed industrial featured models/solution [PP-OCR](../version2.x/ppocr/overview.md)、[PP-Structure](../../ppstructure/README.md) and [PP-ChatOCR](https://aistudio.baidu.com/aistudio/projectdetail/6488689) on this basis, and get through the whole process of data production, model training, compression, inference and deployment.
+
+![img](../images/ppocrv4_en.jpg)
+
+> It is recommended to start with the “quick experience” in the document tutorial
+
+## 📖 Technical exchange and cooperation
+
+- PaddleX provides a one-stop full-process high-efficiency development platform for flying paddle ecological model training, pressure, and push. Its mission is to help AI technology quickly land, and its vision is to make everyone an AI Developer!
+
+    - PaddleX currently covers areas such as image classification, object detection, image segmentation, 3D, OCR, and time series prediction, and has built-in 36 basic single models, such as RP-DETR, PP-YOLOE, PP-HGNet, PP-LCNet, PP- LiteSeg, etc.; integrated 12 practical industrial solutions, such as PP-OCRv4, PP-ChatOCR, PP-ShiTu, PP-TS, vehicle-mounted road waste detection, identification of prohibited wildlife products, etc.
+    - PaddleX provides two AI development modes: "Toolbox" and "Developer". The toolbox mode can tune key hyperparameters without code, and the developer mode can perform single-model training, push and multi-model serial inference with low code, and supports both cloud and local terminals.
+    - PaddleX also supports joint innovation and development, profit sharing! At present, PaddleX is rapidly iterating, and welcomes the participation of individual developers and enterprise developers to create a prosperous AI technology ecosystem!
+
+## 🇺🇳 Guideline for New Language Requests
+
+If you want to request a new language support, a PR with 1 following files are needed：
+
+- In folder [ppocr/utils/dict](../../ppocr/utils/dict),
+it is necessary to submit the dict text to this path and name it with `{language}_dict.txt` that contains a list of all characters. Please see the format example from other files in that folder.
+
+If your language has unique elements, please tell me in advance within any way, such as useful links, wikipedia and so on.
+
+More details, please refer to [Multilingual OCR Development Plan](https://github.com/PaddlePaddle/PaddleOCR/issues/1048).
+
+## Visualization
+
+## PP-OCRv3
+
+### PP-OCRv3 Chinese model
+
+![img](../images/test_add_91.jpg)
+
+![img](../images/00006737.jpg)
+
+![](../images/PP-OCRv3-pic001.jpg)
+
+![](../images/PP-OCRv3-pic002.jpg)
+
+![](../images/PP-OCRv3-pic003.jpg)
+
+### PP-OCRv3 English model
+
+![](../images/en_1.png)
+
+![](../images/en_2.png)
+
+![](../images/en_3-0398013.png)
+
+### PP-OCRv3 Multilingual model
+
+![img](../images/japan_2.jpg)
+
+![img](../images/korean_1.jpg)
+
+#### PP-StructureV2
+
+- layout analysis + table recognition
+
+  ![img](../images/ppstructure-20240708082235651.gif)
+
+- SER (Semantic entity recognition)
+
+  ![img](../images/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808-20240708082238739.jpg)
+
+  ![img](../images/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a-20240708082247529.png)
+
+  ![img](../images/197464552-69de557f-edff-4c7f-acbf-069df1ba097f-20240708082253634.png)
+
+- RE (Relation Extraction)
+
+  ![img](../images/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb-20240708082310650.jpg)
+
+  ![img](../images/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f-20240708082316558.jpg)
+
+  ![img](../images/186094813-3a8e16cc-42e5-4982-b9f4-0134dfb5688d-20240708082323916.png)
+
+## 📄 License
+
+This project is released under <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/LICENSE">Apache 2.0 license</a>
diff --git a/docs/index/index.hi.md b/docs/index/index.hi.md
new file mode 100644
index 0000000..eb3790b
--- /dev/null
+++ b/docs/index/index.hi.md
@@ -0,0 +1,211 @@
+---
+comments: true
+hide:
+  - navigation
+  - toc
+---
+
+<div align="center">
+ <img src="https://github.com/PaddlePaddle/PaddleOCR/releases/download/v2.9.1/PaddleOCR_log.png" align="middle" width = "600"/>
+  <p align="center">
+      <a href="https://discord.gg/z9xaRVjdbD"><img src="https://img.shields.io/badge/Chat-on%20discord-7289da.svg?sanitize=true" alt="Chat"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/LICENSE"><img src="https://img.shields.io/badge/license-Apache%202-dfd.svg"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/releases"><img src="https://img.shields.io/github/v/release/PaddlePaddle/PaddleOCR?color=ffa"></a>
+      <a href=""><img src="https://img.shields.io/badge/python-3.7+-aff.svg"></a>
+      <a href=""><img src="https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-pink.svg"></a>
+      <a href="https://pypi.org/project/PaddleOCR/"><img src="https://img.shields.io/pypi/dm/PaddleOCR?color=9cf"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/stargazers"><img src="https://img.shields.io/github/stars/PaddlePaddle/PaddleOCR?color=ccf"></a>
+  </p>
+</div>
+
+## प्रस्तावना
+
+पैडलओसीआर का उद्देश्य बहुभाषी,शानदार , ओसीआर और व्यावहारिक ओसीआरउपकरण बनाना है जो यूजर्स को बेहतर मॉडलों  के लिए प्रशिक्षित करने और उन्हें व्यवहार में लागू करने में मदद करते हैं।
+
+<div align="center">
+    <img src="https://github.com/PaddlePaddle/PaddleOCR/releases/download/v2.8.0/demo.gif" width="800">
+</div>
+
+## 📣 हाल के अद्यतन
+
+- **🔥2022.8.24 रिलीज Paddleओसीआर [रिलीज/2.6](https://github.com/PaddlePaddle/PaddleOCR/tree/release/2.6)**
+    - रिलीज [PP-Structurev2](../../ppstructure)，फंक्शन और परफॉरमेंस के साथ पूरी तरह से उन्नत, चायनीज शीन्स के अनुकूल, और मदद के लिए [लेआउट रिकवरी](../../ppstructure/recovery) और **पीडीएफ को वर्ड में बदलने के लिए वन लाइन कमांड**;
+    - [लेआउट एनालाइस](../../ppstructure/layout) ऑप्टिमाइजेशन: मॉडल स्टोरेज में 95% की कमी, जबकि स्पीड में 11 गुना वृद्धि , और एवरेज CPU स टाइम-कॉस्ट केवल 41ms है;
+    - [टेबल रिकोगनाइजेशन](../../ppstructure/table) ऑप्टिमाइजेशन: 3 ऑप्टिमाइज़ेशन के तरीके डिजाइन किए गए हैं, और तुलनात्मक समय की खपत के तहत मॉडल सटीकता में 6% का सुधार हुआ है;
+    - [की इंफॉर्मेशन एक्स्ट्रेक्शन](../../ppstructure/kie) ऑप्टिमाइजेशन : एक बिजुवल-स्वतंत्र मॉडल संरचना डिजाइन की गई है, सिमेंटिक एन्टाइटी रिकग्निशन की सटीकता में 2.8% की वृद्धि हुई है, और रिलेशन एक्सट्रैक्शन की सटीकता में 9.1% की वृद्धि हुई है।
+
+- **🔥2022.7 रिलीज [ओसीआर दृश्य आवेदन संग्रह](https://github.com/PaddlePaddle/PaddleOCR/tree/release/2.8/applications/README_en.md)**
+    - रिलीज **9 वर्टिकल मॉडल** जैसे कि डिजिटल ट्यूब, एलसीडी स्क्रीन, लाइसेंस प्लेट, हस्तलेखन पहचान मॉडल, उच्च-सटीक एसवीटीआर मॉडल, आदि, जो सामान्य रूप से मुख्य ओसीआर वर्टिकल अनुप्रयोगों, विनिर्माण, वित्त और परिवहन उद्योगों को कवर करते हैं।
+
+- **🔥2022.5.9 रिलीज Paddleओसीआर [रिलीज/2.5](https://github.com/PaddlePaddle/PaddleOCR/tree/release/2.5)**
+    - रिलीज [PP-OCRv3](../version2.x/ppocr/overview.en.md#pp-ocrv3):      तुलनात्मक स्पीड के साथ, चाइनीज शीन्स का प्रभाव PP-ओसीआर v2 की तुलना में 5% की और वृद्धि हुयी है इंगलिस शीन्स के प्रभाव में 11% का  सुधार हुआ है, और 80 भाषाओं के बहुभाषी मॉडलों की औसत पहचान सटीकता में 5% से अधिक सुधार हुआ है।
+    - रिलीज़ [PPOCRLabelv2](https://github.com/PFCCLab/PPOCRLabel/blob/main/README.md): टेबल टेबल रिकोगनाइजेशन टास्क की इंफॉर्मेशन एक्स्ट्रेक्शन टास्क और अनियमित  टेक्सट इमेज के लिए एनोटेशन फ़ंक्शन एड करे।
+
+    - इंटरएक्टिव ई-बुक जारी करें [*"ओसीआर में गोता लगाएँ"*](../version2.x/ppocr/blog/ocr_book.en.md), ओसीआर पूर्ण स्टैक तकनीक के अत्याधुनिक सिद्धांत और कोड प्रेक्टिस को कवर करता है।
+
+- [और अधिक](../update/update.en.md)
+
+## 🌟 विशेषताएँ
+
+Paddleओसीआर से संबंधित विभिन्न प्रकार के अत्याधुनिक एल्गोरिथ्म को सपोर्ट करता है, और विकसित औद्योगिक विशेष रुप से प्रदर्शित मॉडल/समाधान [PP- OCR](../version2.x/ppocr/overview.en.md) और [PP-Structure](../../ppstructure/README.md) इस आधार पर और डेटा प्रोडक्शन की पूरी प्रोसेस के माध्यम से प्राप्त करें, मॉडल ट्रेनिंग, दबाव, अनुमान और  तैनाती।
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/50011306/196920323-9d386ab0-1233-4415-8508-99d459d256bb.png">
+</div>
+
+## ⚡ शीघ्र अनुभव
+
+```bash
+pip3 install paddlepaddle # for gpu user please install paddlepaddle-gpu
+pip3 install paddleocr
+paddleocr --image_dir /your/test/image.jpg --lang=hi
+```
+
+> यदि आपके पास पायथन एनवायरनमेंट नहीं है, कृपया फॉलो कीजिए [एनवायरनमेंट प्रिपेरेशन](../version2.x/ppocr/environment.en.md).    हम अनुशंसा करते हैं कि आप इसके साथ शुरुआत करें [ट्यूटोरियल](#Tutorials).
+
+## 📚 ई-बुक: *ओसीआर में गोता लगाएँ*
+
+- [ओसीआर में गोता लगाएँ](../version2.x/ppocr/blog/ocr_book.en.md)
+
+## 👫 समुदाय
+
+अंतरराष्ट्रीय डेवलपर्स के लिए, हम सम्मान करते हैं [पैडलओसीआर चर्चाएँ] (<https://github.com/PaddlePaddle/PaddleOCR/discussions>) हमारे अंतरराष्ट्रीय कम्युनिटी मंच के रूप में। यहां सभी विचारों और प्रश्नों पर अंग्रेजी में चर्चा की जा सकती है।
+
+## 🛠️ PP-ओसीआर श्रृंखला मॉडल सूची
+
+|     मॉडल   प्रस्तावना                                                                                    |          मॉडल    नाम                 | रिकमेंडिड  सीन    |   डिटेक्शन मॉडल                                                                                 | रिकोगनाइजेशन मॉडल                                                                        |
+| ------------------------------------------------------------ | ---------------------------- | ----------------- | ------------------------------------------------------------ | ------------------------------------------------------------ |
+| हिन्दी：हिन्दी अल्ट्रा-लाइटवेट PP-OCRv3 सिस्टम (9.9M) | devanagari_PP-OCRv3_xx | मोबाइल और सर्वर |[इन्फरन्स मॉडल](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_infer.tar) / [प्रशिक्षितमॉडल](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_distill_train.tar) | [इन्फरन्समॉडल](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/devanagari_PP-OCRv3_rec_infer.tar) / [प्रशिक्षित मॉडल](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/devanagari_PP-OCRv3_rec_train.tar) |
+| इंग्लिश अल्ट्रा- लाइट वेट PP-OCRv3 मॉडल （13.4M） | en_PP-OCRv3_xx  | मोबाइल और  सर्वर  | [इन्फरन्स मॉडल](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar) / [प्रशिक्षितमॉडल](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_distill_train.tar)| [इन्फरन्समॉडल](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv3_mobile_rec_infer.tar) / [प्रशिक्षित मॉडल](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_train.tar) |
+| चाइनीस और इंग्लिश अल्ट्रा- लाइट वेट PP-OCRv3 मॉडल（16.2M）     | ch_PP-OCRv3_xx          | मोबाइल और सर्वर | [इन्फरन्स मॉडल](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar) / [प्रशिक्षित मॉडल](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams) | [प्रशिक्षित मॉडल](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar) / [प्रशिक्षित मॉडल](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar) |
+
+- अधिक मॉडल डाउनलोड (एकाधिक भाषाओं सहित) के लिए, कृपया [PP-ओसीआर सीरीज मॉडल डाउनलोड](../version2.x/ppocr/model_list.en.md) देखें।
+- एक नए भाषा अनुरोध के लिए, कृपया [नई भाषा अनुरोधों के लिए दिशानिर्देश](#language_requests).
+- स्ट्रक्चर मॉडल डोकोमेंट एनालाइज के लिए, कृपया देखें [PP-Structure models](../version2.x/ppstructure/models_list.en.md).
+
+## 📖 ट्यूटोरियल
+
+- [एनवायरनमेंट प्रिपरेशन](../version2.x/ppocr/environment.en.md)
+- [PP-OCR 🔥](../version2.x/ppocr/overview.en.md)
+    - [क्विक स्टार्ट](../version2.x/ppocr/quick_start.en.md)
+    - [मॉडल जू](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.9/doc/doc_en/models_en.md)
+    - [मॉडल ट्रेनिंग](../version2.x/ppocr/model_train/training.en.md)
+        - [टेक्सट डिटेक्शन](../version2.x/ppocr/model_train/detection.en.md)
+        - [टेक्सट रिकोगनीशन](../version2.x/ppocr/model_train/recognition.en.md)
+        - [टेक्सट डायरेक्शन क्लासिफिकेशन](../version2.x/ppocr/model_train/angle_class.en.md)
+    - मॉडल कम्प्रेशन
+        - [मॉडल परिमाणीकरण](../../deploy/slim/quantization/README_en.md)
+        - [मॉडल प्रूनिंग](../../deploy/slim/prune/README_en.md)
+        - [ज्ञान आसवन](../version2.x/ppocr/model_compress/knowledge_distillation.en.md)
+    - [इन्फरन्स और डिप्लोमेन्ट](../../deploy/README.md)
+        - [Python इन्फरन्स](../version2.x/legacy/python_infer.en.md)
+        - [C++ इन्फरन्स](../version2.x/legacy/cpp_infer.en.md)
+        - [सरविंग](../version2.x/legacy/paddle_server.en.md)
+        - [मोबाइल](../../deploy/lite/readme.md)
+        - [Paddle2ONNX](../../deploy/paddle2onnx/readme.md)
+        - [पैडल क्लाउड](../../deploy/paddlecloud/README.md)
+        - [Benchmark](../version2.x/legacy/benchmark.en.md)
+- [PP-Structure 🔥](../../ppstructure/README.md)
+    - [क्विक स्टार्ट](../version2.x/ppstructure/quick_start.en.md)
+    - [मॉडल जू](../version2.x/ppstructure/models_list.en.md)
+    - [मॉडल ट्रेनिंग](../version2.x/ppocr/model_train/training.en.md)
+        - [लेआउट एनालाइस](../../ppstructure/layout/README.md)
+        - [टेबल रिकोगनाइजेशन](../../ppstructure/table/README.md)
+        - [की इंफॉर्मेशन एक्स्ट्रेक्शन](../../ppstructure/kie/README.md)
+    - [इन्फरन्स और डिप्लोमेन्ट](../../deploy/README.md)
+        - [Python इन्फरन्स](../version2.x/ppstructure/infer_deploy/python_infer.en.md)
+        - [C++ इन्फरन्स](../version2.x/legacy/cpp_infer.en.md)
+        - [सरविंग](../../deploy/hubserving/readme_en.md)
+- [एकेडमिक एल्गोरिथम](../version2.x/algorithm/overview.en.md)
+    - [टेक्स्ट डिनेक्शन](../version2.x/algorithm/overview.en.md)
+    - [टेक्स्ट रिकोगनाइजेशन](../version2.x/algorithm/overview.en.md)
+    - [एंड-टू-एंड ओसीआर](../version2.x/algorithm/overview.en.md)
+    - [टेबल रिकोगनाइजेशन](../version2.x/algorithm/overview.en.md)
+    - [की इंफॉर्मेशन एक्स्ट्रेक्शन](../version2.x/algorithm/overview.en.md)
+    - [पैडलओसीआर में नए एल्गोरिदम जोड़ें](../version2.x/algorithm/add_new_algorithm.en.md)
+- डेटा एनोटेशन और सिंथेसिस
+    - [सेमी-ऑटोमैटिक एनोटेशन टूल: PPओसीआरलेबल](https://github.com/PFCCLab/PPOCRLabel/blob/main/README.md)
+    - [डेटा सिंथेसिस टूल: स्टाइल-टेक्सट](https://github.com/PFCCLab/StyleText/blob/main/README.md)
+    - [अन्य डेटा एनोटेशन टूल](../data_anno_synth/data_annotation.en.md)
+    - [अन्य डेटा सिंथेसिस टूल](../data_anno_synth/data_synthesis.en.md)
+- डेटा सेट
+    - [सामान्य ओसीआर डेटासेट (चीनी/अंग्रेज़ी)](../datasets/datasets.en.md)
+    - [हस्तलिखित_ओसीआर_डेटासेट (चीनी)](../datasets/handwritten_datasets.en.md)
+    - [विभिन्न ओसीआर
+    डेटासेट (बहुभाषी)](../datasets/vertical_and_multilingual_datasets.en.md)
+    - [लेआउट एनालाइस](../datasets/layout_datasets.en.md)
+    - [टेबल रिकोगनाइजेशन](../datasets/table_datasets.en.md)
+    - [की इंफॉर्मेशन एक्स्ट्रेक्शन](../datasets/kie_datasets.en.md)
+- [कोड संरचना](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.9/doc/doc_en/tree_en.md)
+- [विसुमलाइजेशन](#Visualization)
+- [कम्युनिटी](#Community)
+- [नई भाषा के लिए अनुरोध](#language_requests)
+- [सामान्य प्रश्न](../FAQ.en.md)
+- [रेफरेन्सेस](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.9/doc/doc_en/reference_en.md)
+- [लाइसेंस](#LICENSE)
+
+## 🇺🇳 नई भाषा अनुरोधों के लिए संयुक्त राष्ट्र दिशानिर्देश
+
+अगर आप **एक नए भाषा मॉडल का अनुरोध करना चाहते हैं**, तो कृपया [बहुभाषी मॉडल अपग्रेड के लिए वोट करें](https://github.com/PaddlePaddle/PaddleOCR/discussions/7253) में वोट करें। हम नियमित रूप से परिणाम के अनुसार मॉडल को अपग्रेड करेंगे। **अपने दोस्तों को एक साथ वोट करने के लिए आमंत्रित करें!**
+
+यदि आपको **एक नए भाषा मॉडल को प्रशिक्षित करने** अपने परिदृश्य के आधार पर, तो यह [बहुभाषी  मॉडल  ट्रेनिंग  प्रोजेक्ट  ट्रेनिंग](https://github.com/PaddlePaddle/PaddleOCR/discussions/7252) ट्यूटोरियल आपको डेटासेट तैयार करने में मदद करेगा और आपको स्टेप बाए स्टेप पूरा प्रोसेस दिखाएगा
+
+मूल [बहुभाषी ओसीआर विकास योजना](https://github.com/PaddlePaddle/PaddleOCR/issues/1048) अभी भी आपको बहुत सारे उपयोगी संग्रह और शब्दकोश दिखाता है
+
+## 👀 विज़ुअलाइज़ेशन [अधिक](../version2.x/ppocr/visualization.en.md)
+
+<details open>
+<summary>PP-OCRv3 बहुभाषी मॉडल</summary>
+<div align="center">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/multi_lang/japan_2.jpg" width="800">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/multi_lang/korean_1.jpg" width="800">
+</div>
+</details>
+
+<details open>
+<summary>PP-OCRv3 अंग्रेजी मॉडल</summary>
+<div align="center">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/en/en_1.png" width="800">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/en/en_2.png" width="800">
+</div>
+</details>
+<details open>
+<summary>PP-OCRv3 चीनी मॉडल</summary>
+<div align="center">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic001.jpg" width="800">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic002.jpg" width="800">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic003.jpg" width="800">
+</div>
+</details>
+
+<details open>
+<summary>PP-Structurev2</summary>
+1. लेआउट एनालाइस + टेबल रिकोगनाइजेशन
+<div align="center">
+    <img src="../version2.x/ppstructure/images/ppstructure.gif" width="800">
+</div>
+2. SER (सिमेंटिक एंटिटी रिकोगनाइजेशन)
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/25809855/186094456-01a1dd11-1433-4437-9ab2-6480ac94ec0a.png" width="600">
+</div>
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808.jpg" width="600">
+</div>
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a.png" width="600">
+</div>
+3. RE (रिलेशन एक्सट्रैक्शन)
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/25809855/186094813-3a8e16cc-42e5-4982-b9f4-0134dfb5688d.png" width="600">
+</div>
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb.jpg" width="600">
+</div>
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f.jpg" width="600">
+</div>
+</details>
+
+## 📄 लाइसेंस
+
+इस प्रोजेक्ट को इन परियोजना के तहत जारी किया गया है <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/LICENSE">Apache 2.0 license</a>
diff --git a/docs/index/index.ja.md b/docs/index/index.ja.md
new file mode 100644
index 0000000..0c3fb6e
--- /dev/null
+++ b/docs/index/index.ja.md
@@ -0,0 +1,223 @@
+---
+comments: true
+hide:
+  - navigation
+  - toc
+---
+
+<div align="center">
+ <img src="https://github.com/PaddlePaddle/PaddleOCR/releases/download/v2.9.1/PaddleOCR_log.png" align="middle" width = "600"/>
+  <p align="center">
+      <a href="https://discord.gg/z9xaRVjdbD"><img src="https://img.shields.io/badge/Chat-on%20discord-7289da.svg?sanitize=true" alt="Chat"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/LICENSE"><img src="https://img.shields.io/badge/license-Apache%202-dfd.svg"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/releases"><img src="https://img.shields.io/github/v/release/PaddlePaddle/PaddleOCR?color=ffa"></a>
+      <a href=""><img src="https://img.shields.io/badge/python-3.7+-aff.svg"></a>
+      <a href=""><img src="https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-pink.svg"></a>
+      <a href="https://pypi.org/project/PaddleOCR/"><img src="https://img.shields.io/pypi/dm/PaddleOCR?color=9cf"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/stargazers"><img src="https://img.shields.io/github/stars/PaddlePaddle/PaddleOCR?color=ccf"></a>
+  </p>
+</div>
+
+## 紹介
+
+PaddleOCR は、さまざまな言語で、優れた最先端かつ実用的な OCR ツールを作成することを目的とし、ユーザーがより優れたモデルをトレーニングし、実践的に対応できるようになるために役立つAIOCRです。
+
+<div align="center">
+    <img src="https://github.com/PaddlePaddle/PaddleOCR/releases/download/v2.8.0/demo.gif" width="800">
+</div>
+
+## 📣 最新アップデート
+
+- **🔥2022.8.24 リリース PaddleOCR [release/2.6](https://github.com/PaddlePaddle/PaddleOCR/tree/release/2.6)**
+    - [PP-Structurev2](../../ppstructure/)がリリース。機能と使いやすさがアップグレード、中国語のさまざまな文字に適応、 [レイアウトの復旧](../../ppstructure/recovery)  さらに**1 行のコマンドをPDFへ転換、そして Word**に変換可能。
+    - [レイアウト分析](../../ppstructure/layout) の最適化：モデルのストレージが 95% 削減、速度が 11 倍向上、平均 CPU 時間コストはわずか 41 ミリ秒です。
+    - [表認識](../../ppstructure/table) 最適化：3つの最適化戦略設計、モデルの精度が従来より同時間比が 6% 向上。
+    - [キー情報抽出](../../ppstructure/kie) 最適化:視覚に依存しないモデル構造設計、語彙の実態識別精度が 2.8% 向上、関係抽出の精度が 9.1% 向上。
+
+- **🔥2022.7 リリース [OCR scene application collection](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.8/applications/README_en.md)**
+    - デジタルチューブ、液晶画面、ナンバー プレート、手書き認識モデル、高精度 SVTR モデルなど、**9つの垂直モデル**をリリース、一般、製造、金融、運輸業界の主要な OCR 垂直アプリケーションをカバー。
+
+- **🔥2022.5.9 リリース PaddleOCR [release/2.5](https://github.com/PaddlePaddle/PaddleOCR/tree/release/2.5)**
+    - [PP-OCRv3](../version2.x/ppocr/overview.en.md#pp-ocrv3)リリース: 同等の速度で、中国語の識別効果は PP-OCRv2 より 5% 向上、英語の識別効果は 11% 向上し、80 言語の多言語モデルの平均認識精度は 5% 以上向上。
+    - [PPOCRLabelv2](https://github.com/PFCCLab/PPOCRLabel)リリース: 表認識タスク、キー情報抽出タスク、イレギュラーテキスト画像のアノテーション機能を追加。
+    - インタラクティブな電子書籍 [*"OCR に没入"*](../version2.x/ppocr/blog/ocr_book.en.md)、 をリリース。 OCRフルスタック技術の最先端の理論とコードの実践をカバー。
+
+- [もっと](../update/update.en.md)
+
+## 🌟 PaddleOCRとは？
+
+PaddleOCRは、OCRに関連するさまざまな最先端のアルゴリズムに対応する、産業用の機能モデル/ソリューション [PP-OCR](../version2.x/ppocr/overview.en.md) や [PP-Structure](../../ppstructure/README.md) を開発。これに基づき、データの生成、モデルのトレーニング、圧縮、推論、展開の全プロセスを実行可能。
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/50011306/195771471-fad5eb1d-190d-4a7b-8b0c-0433fb32445f.png">
+</div>
+
+## ⚡ 今すぐトライアル
+
+```bash
+pip3 install paddlepaddle # for gpu user please install paddlepaddle-gpu
+pip3 install paddleocr
+paddleocr --image_dir /your/test/image.jpg --lang=japan # change for i18n abbr
+```
+
+>Python環境がない場合は [環境の準備](../version2.x/ppocr/environment.en.md)に従ってください。[チュートリアル](#Tutorials) から始めることをお勧めします。
+
+<a name="本"></a>
+
+## 📚 電子書籍：*OCRに入る*
+
+- [OCRに没入](../version2.x/ppocr/blog/ocr_book.en.md)
+
+<a name="コミュニティ"></a>
+
+## 👫コミュニティー
+
+他国の開発者の方は [PaddleOCR Discussions](https://github.com/PaddlePaddle/PaddleOCR/discussions) を国際的なコミュニティ プラットフォームとして使用します。みなさんのアイデアや質問がある場合、ここで英語で話し合うことができます。
+
+<a name="対応中国機種一覧"></a>
+
+## 🛠️ シリーズ モデル式一覧
+
+| モデル紹介                                           | モデル名                   | 推奨のシーン | 検出モデル                                             | 認識モデル                                           |
+| ------------------------------------------------------------ | ---------------------------- | ----------------- | ------------------------------------------------------------ | ------------------------------------------------------------ |
+| 日本語超軽量 PP-OCRv3 モデル(14.8M) | japan_PP-OCRv3_xx | モバイル & サーバー |[推論モデル](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_infer.tar)/[トレーニングモデル](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_distill_train.tar) |[推論モデル](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/japan_PP-OCRv3_rec_infer.tar)/[トレーニングモデル](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/japan_PP-OCRv3_rec_train.tar) |
+| 英語超軽量PP-OCRv3モデル（13.4M） | en_PP-OCRv3_xx | モバイル & サーバー | [推論モデル](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar) / [トレーニングモデル](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_distill_train.tar) | [推論モデル](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv3_mobile_rec_infer.tar) / [トレーニングモデル](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_train.tar) |
+| 中国語と英語の超軽量 PP-OCRv3 モデル（16.2M）    | ch_PP-OCRv3_xx          | モバイル & サーバー | [推論モデル](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar) / [トレーニングモデル](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams) | [推論モデル](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar) / [トレーニングモデル](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar) |
+
+- その他のモデルのダウンロード (多言語を含む) については、[PP-OCR シリーズ モデルのダウンロード](../version2.x/ppocr/model_list.en.md)をご参照ください。
+- 新しい言語のリクエストについては、 [新しい言語_リクエストのガイドライン](#language_requests)を参照してください。
+- 構造文書分析モデルについては、[PP-Structure models](../version2.x/ppstructure/models_list.en.md)をご参照ください。
+
+<a name="チュートリアル"></a>
+
+## 📖 チュートリアル
+
+- [環境の準備](../version2.x/ppocr/environment.en.md)
+- [PP-OCR 🔥](../version2.x/ppocr/overview.en.md)
+    - [クイックスタート](../version2.x/ppocr/quick_start.en.md)
+    - [Model Zoo](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.9/doc/doc_en/models_en.md)
+    - [トレーニング モデル](../version2.x/ppocr/model_train/training.en.md)
+        - [テキスト検出](../version2.x/ppocr/model_train/detection.en.md)
+        - [テキスト認識](../version2.x/ppocr/model_train/recognition.en.md)
+        - [テキスト方向の分類](../version2.x/ppocr/model_train/angle_class.en.md)
+    - モデル圧縮
+        - [モデルの量子化](../../deploy/slim/quantization/README_en.md)
+        - [モデルの剪裁](../../deploy/slim/prune/README_en.md)
+        - [知識の蒸留](../version2.x/ppocr/model_compress/knowledge_distillation.en.md)
+    - [推論と展開](../../deploy/README.md)
+        - [Python 推論](../version2.x/legacy/python_infer.en.md)
+        - [C++ 推論](../version2.x/legacy/cpp_infer.en.md)
+        - [サービング](../version2.x/legacy/paddle_server.en.md)
+        - [モバイル](../../deploy/lite/readme.md)
+        - [Paddle2ONNX](../../deploy/paddle2onnx/readme.md)
+        - [PaddleCloud](../../deploy/paddlecloud/README.md)
+        - [Benchmark](../version2.x/legacy/benchmark.en.md)
+- [PP-Structure 🔥](../../ppstructure/README.md)
+    - [クイックスタート](../version2.x/ppstructure/quick_start.en.md)
+    - [Model Zoo](../version2.x/ppstructure/models_list.en.md)
+    - [トレーニング モデル](../version2.x/ppocr/model_train/training.en.md)
+        - [レイアウト分析](../../ppstructure/layout/README.md)
+        - [表認識](../../ppstructure/table/README.md)
+        - [キー情報抽出](../../ppstructure/kie/README.md)
+    - [推論と展開](../../deploy/README.md)
+        - [Python 推論](../version2.x/ppstructure/infer_deploy/python_infer.en.md)
+        - [C++ 推論](../version2.x/legacy/cpp_infer.en.md)
+        - [サービング](../../deploy/hubserving/readme_en.md)
+- [アカデミックアリゴリズム](../version2.x/algorithm/overview.en.md)
+    - [テキスト検出](../version2.x/algorithm/overview.en.md)
+    - [テキスト認識](../version2.x/algorithm/overview.en.md)
+    - [エンド・ツー・エンド OCR](../version2.x/algorithm/overview.en.md)
+    - [表認識](../version2.x/algorithm/overview.en.md)
+    - [キー情報抽出](../version2.x/algorithm/overview.en.md)
+    - [PaddleOCR に新しいアルゴリズムを追加する](../version2.x/algorithm/add_new_algorithm.en.md)
+- データの注釈と合成
+    - [半自動注釈ツール: PPOCRLabel](https://github.com/PFCCLab/PPOCRLabel/blob/main/README.md)
+    - [データ合成ツール: Style-Text](https://github.com/PFCCLab/StyleText/blob/main/README.md)
+    - [その他のデータ注釈ツール](../data_anno_synth/data_annotation.en.md)
+    - [その他のデータ合成ツール](../data_anno_synth/data_synthesis.en.md)
+- データセット
+    - [一般OCRデータセット(中国語/英語)](../datasets/datasets.en.md)
+    - [HandWritten_OCR_Datasets(中国語)](../datasets/handwritten_datasets.en.md)
+    - [各種OCRデータセット(多言語対応)](../datasets/vertical_and_multilingual_datasets.en.md)
+    - [レイアウト分析](../datasets/layout_datasets.en.md)
+    - [表認識](../datasets/table_datasets.en.md)
+    - [キー情報抽出](../datasets/kie_datasets.en.md)
+- [コード構造](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.9/doc/doc_en/tree_en.md)
+- [視覚化](#Visualization)
+- [コミュニティ](#Community)
+- [新言語のリクエスト](#language_requests)
+- [よくある質問](../FAQ.en.md)
+- [参考文献](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.9/doc/doc_en/reference_en.md)
+- [ライセンス](#LICENSE)
+
+<a name="language_requests"></a>
+
+## 🇺🇳 新しい言語リクエストのガイドライン
+
+**新言語モデルをリクエスト**したい場合、[多言語モデルのアップグレードへの投票](https://github.com/PaddlePaddle/PaddleOCR/discussions/7253)で投票してください。投票結果に応じて定期的にモデルがアップグレードされます。**友達を招待して一緒に投票しましょう!**
+
+シナリオに基づいて**新しい言語モデルをトレーニング** する必要がある場合は、[多言語モデル トレーニング プロジェクト](https://github.com/PaddlePaddle/PaddleOCR/discussions/7252) のチュートリアルがデータセットの準備にご利用でき、 プロセス全体を段階的に表示することができます。
+
+元の[多言語 OCR 開発計画](https://github.com/PaddlePaddle/PaddleOCR/issues/1048) には、まだ多くの有用なコーパスと辞書が表示されています
+
+<a name="ビジュアリゼーション"></a>
+
+## 👀 ビジュアリゼーション [more](../version2.x/ppocr/visualization.en.md)
+
+<details open>
+<summary>PP-OCRv3 多言語モデル</summary>
+<div align="center">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/multi_lang/japan_2.jpg" width="800">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/multi_lang/korean_1.jpg" width="800">
+</div>
+</details>
+
+<details open>
+<summary>PP-OCRv3 英語 モデル</summary>
+<div align="center">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/en/en_1.png" width="800">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/en/en_2.png" width="800">
+</div>
+</details>
+<details open>
+<summary>PP-OCRv3 中国語 モデル</summary>
+<div align="center">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic001.jpg" width="800">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic002.jpg" width="800">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic003.jpg" width="800">
+</div>
+</details>
+
+<details open>
+<summary>PP-Structurev2</summary>
+1. レイアウト分析＋テーブル認識
+<div align="center">
+    <img src="../version2.x/ppstructure/images/ppstructure.gif" width="800">
+</div>
+2. SER (セマンティックエンティティ認識)
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/25809855/186094456-01a1dd11-1433-4437-9ab2-6480ac94ec0a.png" width="600">
+</div>
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808.jpg" width="600">
+</div>
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a.png" width="600">
+</div>
+3. RE (関係抽出)
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/25809855/186094813-3a8e16cc-42e5-4982-b9f4-0134dfb5688d.png" width="600">
+</div>
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb.jpg" width="600">
+</div>
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f.jpg" width="600">
+</div>
+</details>
+
+<a name="ライセンス"></a>
+
+## 📄 ライセンス
+
+このプロジェクトは以下の場所でリリースされています <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/LICENSE">Apache 2.0 license</a>
diff --git a/docs/index/index.ko.md b/docs/index/index.ko.md
new file mode 100644
index 0000000..87ab2b0
--- /dev/null
+++ b/docs/index/index.ko.md
@@ -0,0 +1,220 @@
+---
+comments: true
+hide:
+  - navigation
+  - toc
+---
+
+<div align="center">
+ <img src="https://github.com/PaddlePaddle/PaddleOCR/releases/download/v2.9.1/PaddleOCR_log.png" align="middle" width = "600"/>
+  <p align="center">
+      <a href="https://discord.gg/z9xaRVjdbD"><img src="https://img.shields.io/badge/Chat-on%20discord-7289da.svg?sanitize=true" alt="Chat"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/LICENSE"><img src="https://img.shields.io/badge/license-Apache%202-dfd.svg"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/releases"><img src="https://img.shields.io/github/v/release/PaddlePaddle/PaddleOCR?color=ffa"></a>
+      <a href=""><img src="https://img.shields.io/badge/python-3.7+-aff.svg"></a>
+      <a href=""><img src="https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-pink.svg"></a>
+      <a href="https://pypi.org/project/PaddleOCR/"><img src="https://img.shields.io/pypi/dm/PaddleOCR?color=9cf"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/stargazers"><img src="https://img.shields.io/github/stars/PaddlePaddle/PaddleOCR?color=ccf"></a>
+  </p>
+</div>
+
+## 소개
+
+PaddleOCR은 사용자들이 보다 나은 모델을 훈련하여 실전에 투입하는데 도움을 주는 다중 언어로 된 엄청나게 멋지고 주도적이며 실용적인 OCR 툴을 만드는데 목표를 두고 있습니다.
+
+<div align="center">
+    <img src="https://github.com/PaddlePaddle/PaddleOCR/releases/download/v2.8.0/demo.gif" width="800">
+</div>
+
+## 📣최근 업데이트
+
+- **🔥2022년 8월 24일에 패들 OCR 출시 [출시/2.6](https://github.com/PaddlePaddle/PaddleOCR/tree/release/2.6)**
+  차이니즈 씬에 맞춘 완전 업그레이드 된 기능과 성능을 갖춘 ; [PP-Structurev2](../../ppstructure/) 출시, 그리고 레이아웃 [리커버리](../../ppstructure/recovery) 신규 지원 및 PDF 를 워드로 전환하는 원 라인 명령
+    - [레이아웃 분석](../../ppstructure/layout)  최적화: 95% 감소된 모델 저장, while 반면 속도는 11배 증가하고, 평균 CPU 시간 비용은 41ms에 불과함;
+    - [표 인식](../../ppstructure/table) 최적화: 3 최적화 전략이 디자인되고 모델 정확도는 비교 가능한 시간 소비 하에 6% 개선됨;
+    - [핵심 정보 추출](../../ppstructure/kie)  최적화： 시각에 의존하지 않는 모델 구조가 디자인되고, 의미체 인식 정확도가 2.8% 증가되며 관계 추출 정확도는 9.1% 증가됨.
+
+- **🔥2022년 7월 출시[OCR 씬 애플리케이션 컬렉션](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.8/applications/README_en.md)**
+    디지털 튜브, LCD 스크린, 라이선스 플레이트, 수기 인식 모델, 고정밀 SVTR 모델 등등과 같은 “9수직 모델” 출시로, 일반적으로 주된 OCR 수직 애플리케이션, 제조, 금융 및 수송 산업 커버
+
+- **🔥2022년 5월 9일에 패들 OCR 출시 [출시/2.5](https://github.com/PaddlePaddle/PaddleOCR/tree/release/2.5)**
+    - [PP-OCRv3](../version2.x/ppocr/overview.en.md#pp-ocrv3)출시: 5%.비교 가능한 속도로, 차이니즈 씬의 효과는 PP-OCRv2와 비교해 볼 때 추가로 5% 정도 더 개선되고, 잉글리쉬 씬 효과는 11% 개선되었으며, 80개 언어 다중 언어 모델 평균 인식 정확도는 5% 이상 개선됨.
+    - [PPOCRLabelv2](https://github.com/PFCCLab/PPOCRLabel)출시: 표 인식 업무, 핵심 정보 추출 업무 및 불규칙한 텍스트 이미지주석 기능 추가.
+    - 쌍방향e-북 출시 [*"OCR 뛰어들기"*](../version2.x/ppocr/blog/ocr_book.en.md), 첨단 이론 및 OCR 정식 스택 기술 코드 연습 포함.
+
+- [추가](../update/update.en.md)
+
+## 🌟특징
+
+패들OCR은 OCR 관련 다양한 첨단 알고리즘 지원  [PP-OCR](../version2.x/ppocr/overview.en.md) 및 [PP-Structure](../../ppstructure/README.md)  이를 기반으로, 그리고 전체 데이터 생산 처리, 모델 훈련, 압축, 추론 및 배치를 통해 획득.
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/50011306/196963392-6cd1b251-109b-49c3-9b3d-ccf203dcec49.png">
+</div>
+
+## ⚡ 신속한 경험
+
+```bash
+pip3 install paddlepaddle # for gpu user please install paddlepaddle-gpu
+pip3 install paddleocr
+paddleocr --image_dir /your/test/image.jpg --lang=korean
+```
+
+>만일 당신이 파이톤 환경이 없다면 [환경 준비를 따르기 바람](../version2.x/ppocr/environment.en.md). 우리는 당신이[사용지침 프로그램]으로 시작할 것을 권장합니다.(#Tutorials).
+
+<a name="북"></a>
+
+## 📚 E-북: *OCR로 뛰어들기*
+
+- [OCR로 뛰어들기](../version2.x/ppocr/blog/ocr_book.en.md)
+
+<a name="커뮤니티"></a>
+
+## 👫 커뮤니티로
+
+국제 개발자들을 위해 우리는 [PaddleOCR 논의하기](https://github.com/PaddlePaddle/PaddleOCR/discussions) 를 우리의 국제 커뮤니티로 간주. 모든 아이디어와 질문은 여기서 영어로 논의 가능.
+
+<a name="지원됨 – 차이니즈-모델-목록- "></a>
+
+## PP-OCR 시리즈 모델 목록
+
+| 모델 소개                                          |모델 명                   | 권장 씬| 감지 모델                                              |인식 모델                                            |
+| ------------------------------------------------------------ | ---------------------------- | ----------------- | ------------------------------------------------------------ | ------------------------------------------------------------ |
+| 한국어 초경량 PP-OCRv3 모델(14.8M) | korean_PP-OCRv3_xx | 모바일 & 서버 | [추론 모델](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_infer.tar) / [훈련 모델](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_distill_train.tar) | [추론 모델](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/korean_PP-OCRv3_rec_infer.tar) / [훈련 모델](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/korean_PP-OCRv3_rec_train.tar) |
+| 영어 초경량 PP-OCRv3 모델（13.4M） | en_PP-OCRv3_xx | 모바일 & 서버 | [추론 모델](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar) / [훈련 모델](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_distill_train.tar) | [추론 모델](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv3_mobile_rec_infer.tar) / [훈련 모델](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_train.tar) |
+| 중국어 및 영어 초경량 PP-OCRv3 model（16.2M）     | ch_PP-OCRv3_xx          | 모바일 & 서버 | [추론 모델](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar) / [훈련 모델](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams) | [추론 모델](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar) / [훈련 모델](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar) |
+
+- (다중 언어를 포함하여)더 많은 모델을 다운로드 하려면, [PP-OCR 시리즈 모델 다운로드](../version2.x/ppocr/model_list.en.md)를 참조할 것.
+- 신규 언어 요청에 대해서는, [신규 언어 요청 지침](#language_requests)을 참조할 것.
+- 구조적 문서 분석 모델에 대해서는, [PP-Structure models](../version2.x/ppstructure/models_list.en.md).을 참조할 것.
+
+<a name="사용 지침 프로그램"></a>
+
+## 📖 사용 지침 프로그램
+
+- [환경 준비](../version2.x/ppocr/environment.en.md)
+- [PP-OCR 🔥](../version2.x/ppocr/overview.en.md)
+    - [신속한 시작](../version2.x/ppocr/quick_start.en.md)
+    - [동물원 모델](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.9/doc/doc_en/models_en.md)
+    - [모델 훈련](../version2.x/ppocr/model_train/training.en.md)
+        - [텍스트 감지](../version2.x/ppocr/model_train/detection.en.md)
+        - [텍스트 인식](../version2.x/ppocr/model_train/recognition.en.md)
+        - [텍스트 방향 분류](../version2.x/ppocr/model_train/angle_class.en.md)
+    - 모델 압축
+        - [모델 계량화](../../deploy/slim/quantization/README_en.md)
+        - [모델 전지작업](../../deploy/slim/prune/README_en.md)
+        - [지식 정제](../version2.x/ppocr/model_compress/knowledge_distillation.en.md)
+    - [추론 및 배치](../../deploy/README.md)
+        - [파이톤 추론](../version2.x/legacy/python_infer.en.md)
+        - [C++ 추론](../version2.x/legacy/cpp_infer.en.md)
+        - [서빙](../version2.x/legacy/paddle_server.en.md)
+        - [모바일](../../deploy/lite/readme.md)
+        - [Paddle2ONNX](../../deploy/paddle2onnx/readme.md)
+        - [패들 클라우드](../../deploy/paddlecloud/README.md)
+        - [Benchmark](../version2.x/legacy/benchmark.en.md)
+- [PP-Structure 🔥](../../ppstructure/README.md)
+    - [신속한 시작](../version2.x/ppstructure/quick_start.en.md)
+    - [동물원 모델](../version2.x/ppstructure/models_list.en.md)
+    - [모델 훈련](../version2.x/ppocr/model_train/training.en.md)
+        - [레이아웃 분석](../../ppstructure/layout/README.md)
+        - [표 인식](../../ppstructure/table/README.md)
+        - [핵심 정보 추출](../../ppstructure/kie/README.md)
+    - [추론 및 배치](../../deploy/README.md)
+        - [파이톤 추론](../version2.x/ppstructure/infer_deploy/python_infer.en.md)
+        - [C++ 추론](../version2.x/legacy/cpp_infer.en.md)
+        - [서빙](../../deploy/hubserving/readme_en.md)
+- [학문적 알고리즘](../version2.x/algorithm/overview.en.md)
+    - [텍스트 감지](../version2.x/algorithm/overview.en.md)
+    - [텍스트 인식](../version2.x/algorithm/overview.en.md)
+    - [종단종OCR](../version2.x/algorithm/overview.en.md)
+    - [표 인식](../version2.x/algorithm/overview.en.md)
+    - [핵심 정보 추출](../version2.x/algorithm/overview.en.md)
+    - [PaddleOCR에 신규 알고리즘 추가](../version2.x/algorithm/add_new_algorithm.en.md)
+- 데이터 주석 및 합성
+    - [반-자동 주석 툴: PPOCRLabel](https://github.com/PFCCLab/PPOCRLabel/blob/main/README.md)
+    - [데이터 합성 툴: 스타일-텍스트](https://github.com/PFCCLab/StyleText/blob/main/README.md)
+    - [기타 데이터 주석 툴](../data_anno_synth/data_annotation.en.md)
+    - [기타 데이터 합성 툴](../data_anno_synth/data_synthesis.en.md)
+- 데이터세트
+    - [일반 OCR 데이터세트(중국어/영어)](../datasets/datasets.en.md)
+    - [수기_OCR_데이터세트(중국어)](../datasets/handwritten_datasets.en.md)
+    - [다양한 OCR 데이터세트(다중언어)](../datasets/vertical_and_multilingual_datasets.en.md)
+    - [레이아웃 분석](../datasets/layout_datasets.en.md)
+    - [표 인식](../datasets/table_datasets.en.md)
+    - [핵심 정보 추출](../datasets/kie_datasets.en.md)
+- [코드 구조](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.9/doc/doc_en/tree_en.md)
+- [시각화](#Visualization)
+- [커뮤니티](#Community)
+- [신규 언어 요청](#language_requests)
+- [자주 묻는 질문](../FAQ.en.md)
+- [추론](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.9/doc/doc_en/reference_en.md)
+- [라이선스](#LICENSE)
+
+<a name="language_requests"></a>
+
+## 신규 언어 요청에 대한 유엔 가이드라인
+
+만일 신규 언어 모델을 요청하고자 한다면**, [다중 언어 모델 업그레이드 투표하기](https://github.com/PaddlePaddle/PaddleOCR/discussions/7253)에서 투표하기 바람. 우리는 결과에 따라 규칙적으로 모델을 업그레이드 시킬 것임**함께 투표하고자 당신의 친구들을 초대할 것!**
+만일 당신이 시나리오 기반 “신규 언어 모델”을 훈련하고자 한다면, [다중 언어 모델 훈련 프로젝트](https://github.com/PaddlePaddle/PaddleOCR/discussions/7252) 를 통해 당신의 데이터세트를 작성하는데 도움이 되고 단계별로 전체 절차를 보여줄 것입니다.
+원본[다중 언어 OCR 개발 계획](https://github.com/PaddlePaddle/PaddleOCR/issues/1048)은 여전히 수많은 유용한 말뭉치와 사전을 보여줍니다.
+
+<a name="시각화"></a>
+
+## 👀 시각화[추가](../version2.x/ppocr/visualization.en.md)
+
+<details open>
+<summary>PP-OCRv3 다중 언어 모델</summary>
+<div align="center">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/multi_lang/japan_2.jpg" width="800">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/multi_lang/korean_1.jpg" width="800">
+</div>
+</details>
+
+<details open>
+<summary>PP-OCRv3 영어 모델</summary>
+<div align="center">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/en/en_1.png" width="800">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/en/en_2.png" width="800">
+</div>
+</details>
+<details open>
+<summary>PP-OCRv3 중국어 모델</summary>
+<div align="center">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic001.jpg" width="800">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic002.jpg" width="800">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic003.jpg" width="800">
+</div>
+</details>
+
+<details open>
+<summary>PP-Structurev2</summary>
+1.  레이아웃 분석 + 표 인식
+<div align="center">
+    <img src="../version2.x/ppstructure/images/ppstructure.gif" width="800">
+</div>
+2. SER (의미체 인식)
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/25809855/186094456-01a1dd11-1433-4437-9ab2-6480ac94ec0a.png" width="600">
+</div>
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808.jpg" width="600">
+</div>
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a.png" width="600">
+</div>
+3. RE (관계 추출)
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/25809855/186094813-3a8e16cc-42e5-4982-b9f4-0134dfb5688d.png" width="600">
+</div>
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb.jpg" width="600">
+</div>
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f.jpg" width="600">
+</div>
+</details>
+<a name="라이선스"></a>
+
+## 📄 라이선스
+
+본 프로젝트는 <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/LICENSE">Apache 2.0 license</a> 하에 출시됨.
diff --git a/docs/index/index.md b/docs/index/index.md
new file mode 100644
index 0000000..13c01b4
--- /dev/null
+++ b/docs/index/index.md
@@ -0,0 +1,123 @@
+---
+comments: true
+hide:
+  - navigation
+  - toc
+---
+
+<div align="center">
+ <img src="https://github.com/PaddlePaddle/PaddleOCR/releases/download/v2.9.1/PaddleOCR_log.png" align="middle" width = "600"/>
+  <p align="center">
+      <a href="https://discord.gg/z9xaRVjdbD"><img src="https://img.shields.io/badge/Chat-on%20discord-7289da.svg?sanitize=true" alt="Chat"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/LICENSE"><img src="https://img.shields.io/badge/license-Apache%202-dfd.svg"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/releases"><img src="https://img.shields.io/github/v/release/PaddlePaddle/PaddleOCR?color=ffa"></a>
+      <a href=""><img src="https://img.shields.io/badge/python-3.7+-aff.svg"></a>
+      <a href=""><img src="https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-pink.svg"></a>
+      <a href="https://pypi.org/project/PaddleOCR/"><img src="https://img.shields.io/pypi/dm/PaddleOCR?color=9cf"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/stargazers"><img src="https://img.shields.io/github/stars/PaddlePaddle/PaddleOCR?color=ccf"></a>
+  </p>
+</div>
+
+## 简介
+
+PaddleOCR 旨在打造一套丰富、领先、且实用的 OCR 工具库，助力开发者训练出更好的模型，并应用落地。
+
+## 🚀 社区
+
+PaddleOCR 由 [PMC](https://github.com/PaddlePaddle/PaddleOCR/issues/12122) 监督。Issues 和 PRs 将在尽力的基础上进行审查。欲了解 PaddlePaddle 社区的完整概况，请访问 [community](https://github.com/PaddlePaddle/community)。
+
+⚠️注意：[Issues](https://github.com/PaddlePaddle/PaddleOCR/issues)模块仅用来报告程序🐞Bug，其余提问请移步[Discussions](https://github.com/PaddlePaddle/PaddleOCR/discussions)模块提问。如所提Issue不是Bug，会被移到Discussions模块，敬请谅解。
+
+## 📣 近期更新
+
+- **🔥🔥2025.3.7 PaddleOCR 2.10 版本，主要包含如下内容**：
+
+  - **重磅新增 OCR 领域 12 个自研单模型：**
+    - **[版面区域检测](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/layout_detection.html)** 系列 3 个模型：PP-DocLayout-L、PP-DocLayout-M、PP-DocLayout-S，支持预测 23 个常见版面类别，中英论文、研报、试卷、书籍、杂志、合同、报纸等丰富类型的文档实现高质量版面检测，**mAP@0.5 最高达 90.4%，轻量模型端到端每秒处理超百页文档图像。**
+    - **[公式识别](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/formula_recognition.html)** 系列 2 个模型：PP-FormulaNet-L、PP-FormulaNet-S，支持 5 万种 LaTeX 常见词汇，支持识别高难度印刷公式和手写公式，其中 **PP-FormulaNet-L 较开源同等量级模型精度高 6 个百分点，PP-FormulaNet-S 较同等精度模型速度快 16 倍。**
+    - **[表格结构识别](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/table_structure_recognition.html)** 系列 2 个模型：SLANeXt_wired、SLANeXt_wireless。飞桨自研新一代表格结构识别模型，分别支持有线表格和无线表格的结构预测。相比于SLANet_plus，SLANeXt在表格结构方面有较大提升，**在内部高难度表格识别评测集上精度高 6 个百分点。**
+    - **[表格分类](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/table_classification.html)** 系列 1 个模型：PP-LCNet_x1_0_table_cls，超轻量级有线表格和无线表格的分类模型。
+    - **[表格单元格检测](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/table_cells_detection.html)** 系列 2 个模型：RT-DETR-L_wired_table_cell_det、RT-DETR-L_wireless_table_cell_det，分别支持有线表格和无线表格的单元格检测，可配合SLANeXt_wired、SLANeXt_wireless、文本检测、文本识别模块完成对表格的端到端预测。（参见本次新增的表格识别v2产线）
+    - **[文本识别](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/text_recognition.html)** 系列 1 个模型： PP-OCRv4_server_rec_doc，**支持1.5万+字典，文字识别范围更广，与此同时提升了部分文字的识别精准度，在内部数据集上，精度较 PP-OCRv4_server_rec 高 3 个百分点以上。**
+    - **[文本行方向分类](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/text_recognition.html)** 系列 1 个模型：PP-LCNet_x0_25_textline_ori，**存储只有 0.3M** 的超轻量级文本行方向分类模型。
+
+   - **重磅推出 4 条高价值多模型组合方案：** 
+     - **[文档图像预处理产线](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/doc_preprocessor.html)**：通过超轻量级模型组合使用，实现对文档图像的扭曲和方向的矫正。
+     - **[版面解析v2产线](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/layout_parsing.html)**：组合多个自研的不同类型的 OCR 类模型，优化复杂版面阅读顺序，实现多种复杂 PDF 文件端到端转换 Markdown 文件和 JSON 文件。在多个文档场景下，转换效果较其他开源方案更好。可以为大模型训练和应用提供高质量的数据生产能力。
+     - **[表格识别v2产线](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/table_recognition_v2.html)**：**提供更好的表格端到端识别能力。** 通过将表格分类模块、表格单元格检测模块、表格结构识别模块、文本检测模块、文本识别模块等组合使用，实现对多种样式的表格预测，用户可自定义微调其中任意模块以提升垂类表格的效果。
+     - **[PP-ChatOCRv4-doc产线](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction_v4.html)**：在 PP-ChatOCRv3-doc 的基础上，**融合了多模态大模型，优化了 Prompt 和多模型组合后处理逻辑，更好地解决了版面分析、生僻字、多页 pdf、表格、印章识别等常见的复杂文档信息抽取难点问题，准确率较 PP-ChatOCRv3-doc 高 15 个百分点。其中，大模型升级了本地部署的能力，提供了标准的 OpenAI 调用接口，支持对本地大模型如 DeepSeek-R1 部署的调用。**
+
+- **🔥2024.10.1 添加OCR领域低代码全流程开发能力**:
+    - 飞桨低代码开发工具PaddleX，依托于PaddleOCR的先进技术，支持了OCR领域的低代码全流程开发能力：
+        - 🎨 [**模型丰富一键调用**](https://www.paddleocr.ai/v2.10.0/paddlex/quick_start.html)：将文本图像智能分析、通用OCR、通用版面解析、通用表格识别、公式识别、印章文本识别涉及的**17个模型**整合为6条模型产线，通过极简的**Python API一键调用**，快速体验模型效果。此外，同一套API，也支持图像分类、目标检测、图像分割、时序预测等共计**200+模型**，形成20+单功能模块，方便开发者进行**模型组合**使用。
+        - 🚀[**提高效率降低门槛**](https://www.paddleocr.ai/v2.10.0/paddlex/overview.html)：提供基于**统一命令**和**图形界面**两种方式，实现模型简洁高效的使用、组合与定制。支持**高性能推理、服务化部署和端侧部署**等多种部署方式。此外，对于各种主流硬件如**英伟达GPU、昆仑芯、昇腾、寒武纪和海光**等，进行模型开发时，都可以**无缝切换**。
+
+    - 支持文档场景信息抽取v3[PP-ChatOCRv3-doc](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction.md)、基于RT-DETR的[高精度版面区域检测模型](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/layout_detection.md)和PicoDet的[高效率版面区域检测模型](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/layout_detection.md)、高精度表格结构识别模型[SLANet_Plus](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/table_structure_recognition.md)、文本图像矫正模型[UVDoc](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/text_image_unwarping.md)、公式识别模型[LatexOCR](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/formula_recognition.md)、基于PP-LCNet的[文档图像方向分类模型](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.md)
+
+- **🔥2024.7 添加 PaddleOCR 算法模型挑战赛冠军方案**：
+    - 赛题一：OCR 端到端识别任务冠军方案——[场景文本识别算法-SVTRv2](https://www.paddleocr.ai/v2.10.0/algorithm/text_recognition/algorithm_rec_svtrv2.html)；
+    - 赛题二：通用表格识别任务冠军方案——[表格识别算法-SLANet-LCNetV2](https://www.paddleocr.ai/v2.10.0/algorithm/table_recognition/algorithm_table_slanet.html)。
+
+> [更多](../update/update.md)
+
+## 🌟 特性
+
+支持多种 OCR 相关前沿算法，在此基础上打造产业级特色模型PP-OCR、PP-Structure和PP-ChatOCR，并打通数据生产、模型训练、压缩、预测部署全流程。
+
+<img src="../images/ppocrv4.png" width="600" />
+
+## 效果展示
+
+### 超轻量PP-OCRv3效果展示
+
+#### PP-OCRv3中文模型
+
+![img](../images/test_add_91.jpg)
+
+<img src="../images/00006737.jpg" width="600" />
+
+<img src="../images/PP-OCRv3-pic001.jpg" width="600" />
+
+<img src="../images/PP-OCRv3-pic002.jpg" width="600" />
+
+<img src="../images/PP-OCRv3-pic003.jpg" width="600" />
+
+#### PP-OCRv3英文数字模型
+
+<img src="../images/en_1.png" width="600" />
+
+<img src="../images/en_2.png" width="600" />
+
+<img src="../images/en_3-0398013.png" width="600" />
+
+#### PP-OCRv3多语言模型
+
+<img src="../images/japan_2.jpg" width="600" />
+
+<img src="../images/korean_1.jpg" width="600" />
+
+#### PP-Structure 文档分析
+
+- 版面分析+表格识别
+
+  <img src="../images/ppstructure-20240708082235651.gif" width="600" />
+
+- SER（语义实体识别）
+
+  <img src="../images/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808-20240708082238739.jpg" width="600" />
+
+  <img src="../images/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a-20240708082247529.png" width="600" />
+
+  <img src="../images/197464552-69de557f-edff-4c7f-acbf-069df1ba097f-20240708082253634.png" width="600" />
+
+- RE（关系提取）
+
+  <img src="../images/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb-20240708082310650.jpg" width="600" />
+
+  <img src="../images/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f-20240708082316558.jpg" width="600" />
+
+  <img src="../images/186094813-3a8e16cc-42e5-4982-b9f4-0134dfb5688d-20240708082323916.png" width="600" />
+
+## 许可证书
+
+本项目的发布受<a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/LICENSE">Apache 2.0 license</a>许可认证。
diff --git a/docs/index/index.ru.md b/docs/index/index.ru.md
new file mode 100644
index 0000000..48146ad
--- /dev/null
+++ b/docs/index/index.ru.md
@@ -0,0 +1,222 @@
+---
+comments: true
+hide:
+  - navigation
+  - toc
+---
+
+<div align="center">
+ <img src="https://github.com/PaddlePaddle/PaddleOCR/releases/download/v2.9.1/PaddleOCR_log.png" align="middle" width = "600"/>
+  <p align="center">
+      <a href="https://discord.gg/z9xaRVjdbD"><img src="https://img.shields.io/badge/Chat-on%20discord-7289da.svg?sanitize=true" alt="Chat"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/LICENSE"><img src="https://img.shields.io/badge/license-Apache%202-dfd.svg"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/releases"><img src="https://img.shields.io/github/v/release/PaddlePaddle/PaddleOCR?color=ffa"></a>
+      <a href=""><img src="https://img.shields.io/badge/python-3.7+-aff.svg"></a>
+      <a href=""><img src="https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-pink.svg"></a>
+      <a href="https://pypi.org/project/PaddleOCR/"><img src="https://img.shields.io/pypi/dm/PaddleOCR?color=9cf"></a>
+      <a href="https://github.com/PaddlePaddle/PaddleOCR/stargazers"><img src="https://img.shields.io/github/stars/PaddlePaddle/PaddleOCR?color=ccf"></a>
+  </p>
+</div>
+
+## Введение
+
+PaddleOCR стремится создавать многоязычные, потрясающие, передовые и практичные инструменты OCR, которые помогают пользователям обучать лучшие модели и применять их на практике
+
+<div align="center">
+    <img src="https://github.com/PaddlePaddle/PaddleOCR/releases/download/v2.8.0/demo.gif" width="800">
+</div>
+
+## 📣 Последние обновления
+
+- **🔥2022.8.24 Выпуск PaddleOCR [Выпуск /2.6](https://github.com/PaddlePaddle/PaddleOCR/tree/release/2.6)**
+    - Выпускать [PP-Structurev2](../../ppstructure/)，с полностью обновленными функциями и производительностью, адаптированными для китайских сцен и новой поддержкой pаспознавание таблиц
+     [Восстановление макета](../../ppstructure/recovery) и **однострочная команда для преобразования PDF в Word**;
+    - [Анализ макета](../../ppstructure/layout) оптимизация: память модели уменьшена на 95%, а скорость увеличена в 11 раз, а среднее время процессорного времени составляет всего 41 мс;
+    - [Распознавание таблиц](../../ppstructure/table) оптимизация: разработано 3 стратегии оптимизации, а точность модели улучшена на 6% при сопоставимых затратах времени;
+    - [Извлечение ключевой информации](../../ppstructure/kie) оптимизация: разработана визуально независимая структура модели, точность распознавания семантической сущности увеличена на 2,8%, а точность извлечения отношения увеличена на 9,1%.
+- **🔥2022.7 Выпуск [Коллекция приложений сцены OCR](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.8/applications/README_en.md)**
+- Выпуск **9 вертикальных моделей**, таких как цифровая трубка, ЖК-экран, номерной знак, модель распознавания рукописного ввода, высокоточная модель SVTR и т. д., охватывающих основные вертикальные приложения OCR в целом, производственной, финансовой и транспортной отраслях.
+- **🔥2022.5.9 Выпуск PaddleOCR [Выпуск /2.5](https://github.com/PaddlePaddle/PaddleOCR/tree/release/2.5)**
+- Выпускать [PP-OCRv3](../version2.x/ppocr/overview.en.md#pp-ocrv3): При сопоставимой скорости эффект китайской сцены улучшен на 5% по сравнению с ПП-OCRRv2, эффект английской сцены улучшен на 11%, а средняя точность распознавания 80 языковых многоязычных моделей улучшена более чем на 5%.
+- Выпускать [PPOCRLabelv2](https://github.com/PFCCLab/PPOCRLabel/blob/main/README.md): Добавьте функцию аннотации для задачи распознавания таблиц, задачи извлечения ключевой информации и изображения неправильного текста.
+    - Выпустить интерактивную электронную книгу [*"Погружение в OCR"*](../version2.x/ppocr/blog/ocr_book.en.md), охватывает передовую теорию и практику кодирования технологии полного стека OCR.
+- [подробнее](../update/update.en.md)
+
+## 🌟 Функции
+
+PaddleOCR поддерживает множество передовых алгоритмов, связанных с распознаванием текста, и разработала промышленные модели/решения. [PP-OCR](../version2.x/ppocr/overview.en.md) и [PP-Structure](../../ppstructure/README.md) на этой основе и пройти весь процесс производства данных, обучения модели, сжатия, логического вывода и развертывания.
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/50011306/196963669-f53b0ee5-3cb4-481c-b73c-97c4b3e2efb8.png">
+</div>
+
+## ⚡ Быстрый опыт
+
+```bash
+pip3 install paddlepaddle # for gpu user please install paddlepaddle-gpu
+pip3 install paddleocr
+paddleocr --image_dir /your/test/image.jpg --lang=ru
+```
+
+> Если у вас нет среды Python, выполните [Подготовка среды](../version2.x/ppocr/environment.en.md). Мы рекомендуем вам начать с [Учебники](#Tutorials).
+
+<a name="книга"></a>
+
+## 📚 Электронная книга: *Погружение в OCR*
+
+- [Погружение в распознавание символов](../version2.x/ppocr/blog/ocr_book.en.md)
+
+<a name="Сообщество"></a>
+
+## 👫 Сообщество
+
+Что касается международных разработчиков, мы рассматриваем [Обсуждения PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR/discussions) как нашу платформу для международного сообщества. Все идеи и вOCRосы можно обсудить здесь на английском языке.
+
+<a name="Список-поддерживаемых-китайских-моделей"></a>
+
+## 🛠️ Список моделей серии ПП -OCR
+
+| Введение модели | Название модели | Рекомендуемая сцена | Модель обнаружения | Модель распознавания |
+| ------------------------------------------------------------ | ---------------------------- | ----------------- | ------------------------------------------------------------ | ------------------------------------------------------------ |
+| Ру́сский язы́к：Ру́сский язы́к Сверхлегкая модель PP-OCRv3 (13.4M) | cyrillic_PP-OCRv3_xx | Мобильный и сервер |[модель вывода](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_infer.tar)/[обученный модель](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_distill_train.tar) | [модель вывода](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/cyrillic_PP-OCRv3_rec_infer.tar)/[обученный модель](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/cyrillic_PP-OCRv3_rec_train.tar)  |
+| Английский сверхлегкая модель PP-OCRv3 (13,4 Мб) | en_PP-OCRv3_xx |Мобильный и сервер | [модель вывода](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar) / [обученный модель](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_distill_train.tar) | [вывод модель](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv3_mobile_rec_infer.tar) / [обученный модель](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_train.tar) |
+| Сверхлегкая китайская и английская модель PP-OCRv3 (16,2M) | ch_PP-OCRv3_xx | Мобильный и сервер | [вывод модель](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar) / [обученный модель](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams) | [вывод модель](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar) / [обученный модель](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar) |
+
+- Для получения дополнительных загрузок моделей (включая несколько языков) см. [Загрузки моделей серии ПП-OCR](../version2.x/ppocr/model_list.en.md).
+- Для запроса нового языка см [Руководство для новых языковых_запросов](#language_requests).
+- Модели структурного анализа документов см [PP-Structure модельs](../version2.x/ppstructure/models_list.en.md).
+
+<a name=" Учебники "></a>
+
+## 📖 Учебники
+
+- [Подготовка окружающей среды](../version2.x/ppocr/environment.en.md)
+- [PP-OCR 🔥](../version2.x/ppocr/overview.en.md)
+
+    - [Быстрый старт](../version2.x/ppocr/quick_start.en.md)
+    - [Модель Zoo](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.9/doc/doc_en/models_en.md)
+    - [Модель тренировки](../version2.x/ppocr/model_train/training.en.md)
+    - [Обнаружение текста](../version2.x/ppocr/model_train/detection.en.md)
+        - [Распознавание текста](../version2.x/ppocr/model_train/recognition.en.md)
+        - [Классификация направления текста](../version2.x/ppocr/model_train/angle_class.en.md)
+    - Модель Сжатие
+        - [Модель квантования](../../deploy/slim/quantization/README_en.md)
+        - [Модель Обрезка](../../deploy/slim/prune/README_en.md)
+        - [Дистилляция знаний](../version2.x/ppocr/model_compress/knowledge_distillation.en.md)
+    - [Вывод и развертывание](../../deploy/README.md)
+        - [Python Вывод](../version2.x/legacy/python_infer.en.md)
+        - [Вывод C++](../version2.x/legacy/cpp_infer.en.md)
+        - [Подача](../version2.x/legacy/paddle_server.en.md)
+        - [Мобильный](../../deploy/lite/readme.md)
+        - [Paddle2ONNX](../../deploy/paddle2onnx/readme.md)
+        - [ВеслоОблако](../../deploy/paddlecloud/README.md)
+        - [Benchmark](../version2.x/legacy/benchmark.en.md)
+- [PP-Structure 🔥](../../ppstructure/README.md)
+    - [Быстрый старт](../version2.x/ppstructure/quick_start.en.md)
+        - [Модель Zoo](../version2.x/ppstructure/models_list.en.md)
+        - [Модель тренировки](../version2.x/ppocr/model_train/training.en.md)
+    - [Анализ макета](../../ppstructure/layout/README.md)
+        - [Распознавание таблиц](../../ppstructure/table/README.md)
+        - [Извлечение ключевой информации](../../ppstructure/kie/README.md)
+    - [Вывод и развертывание](../../deploy/README.md)
+        - [Вывод Python](../version2.x/ppstructure/infer_deploy/python_infer.en.md)
+        - [Вывод С++](../version2.x/legacy/cpp_infer.en.md)
+        - [Обслуживание](../../deploy/hubserving/readme_en.md)
+- [Академические алгоритмы](../version2.x/algorithm/overview.en.md)
+    - [Обнаружение текста](../version2.x/algorithm/overview.en.md)
+- [Распознавание текста](../version2.x/algorithm/overview.en.md)
+    - [Непрерывной цепью OCR](../version2.x/algorithm/overview.en.md)
+    - [Распознавание таблиц](../version2.x/algorithm/overview.en.md)
+    - [Извлечение ключевой информации](../version2.x/algorithm/overview.en.md)
+    - [Добавьте новые алгоритмы в PaddleOCR](../version2.x/algorithm/add_new_algorithm.en.md)
+- Аннотации и синтез данных
+    - [Полуавтоматический инструмент аннотации данных: метка ППOCRR](https://github.com/PFCCLab/PPOCRLabel/blob/main/README.md)
+    - [Инструмент синтеза данных: Стиль-текст](https://github.com/PFCCLab/StyleText/blob/main/README.md)
+    - [Другие инструменты аннотирования данных](../data_anno_synth/data_annotation.en.md)
+    - [Другие инструменты синтеза данных](../data_anno_synth/data_synthesis.en.md)
+- Наборы данных
+    - [Общие наборы данных OCR (китайский/английский)](../datasets/datasets.en.md)
+    - [Наборы данных Рукописный/*OCR* наборы данных (китайский)](../datasets/handwritten_datasets.en.md)
+    - [Различные наборы данных OCR (многоязычные)](../datasets/vertical_and_multilingual_datasets.en.md)
+    - [Анализ макета](../datasets/layout_datasets.en.md)
+    - [Распознавание таблиц](../datasets/table_datasets.en.md)
+    - [Извлечение ключевой информации](../datasets/kie_datasets.en.md)
+- [Структура кода](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.9/doc/doc_en/tree_en.md)
+- [Визуализация](#Visualization)
+- [Сообщество](#Community)
+- [Новые языковые запросы](#language_requests)
+- [ЧАСТО ЗАДАВАЕМЫЕ ВOCRОСЫ](../FAQ.en.md)
+- [Использованная литература](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.9/doc/doc_en/reference_en.md)
+- [ЛИЦЕНЗИЯ](#LICENSE)
+
+<a name="language_requests"></a>
+
+## 🇺🇳 Руководство по запросам на новый язык
+
+Если вы хотите **запросить новую языковую модель**, проголосуйте в [Голосуйте за обновление многоязычной модели](https://github.com/PaddlePaddle/PaddleOCR/discussions/7253). Мы будем регулярно обновлять модель по результату. **Пригласите друзей проголосовать вместе!**
+
+Если вам нужно **обучить новую языковую модель** на основе вашего сценария, учебное пособие в [Проекте обучения многоязычной модели](https://github.com/PaddlePaddle/PaddleOCR/discussions/7252) поможет вам подготовить набор данных и показать вам весь процесс шаг за шагом.
+
+Оригинальный [Многоязычный план разработки OCR](https://github.com/PaddlePaddle/PaddleOCR/issues/1048) по-прежнему показывает вам много полезных корпусов и словарей.
+
+<a name=" Визуализация "></a>
+
+## 👀 Визуализация [больше](../version2.x/ppocr/visualization.en.md)
+
+<details open>
+<summary>PP-OCRv3 Многоязычная модель </summary>
+<div align="center">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/multi_lang/japan_2.jpg" width="800">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/multi_lang/korean_1.jpg" width="800">
+</div>
+</details>
+
+<details open>
+<summary>PP-OCRv3 Aнглийская модель </summary>
+<div align="center">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/en/en_1.png" width="800">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/en/en_2.png" width="800">
+</div>
+</details>
+<details open>
+<summary>PP-OCRv3 Kитайская модель </summary>
+<div align="center">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic001.jpg" width="800">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic002.jpg" width="800">
+    <img src="../version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic003.jpg" width="800">
+</div>
+</details>
+
+<details open>
+<summary>PP-Structurev2</summary>
+1. анализ макета + распознавание таблиц
+<div align="center">
+    <img src="../version2.x/ppstructure/images/ppstructure.gif" width="800">
+</div>
+2. SER (Семантическое распознавание объектов)
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/25809855/186094456-01a1dd11-1433-4437-9ab2-6480ac94ec0a.png" width="600">
+</div>
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808.jpg" width="600">
+</div>
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a.png" width="600">
+</div>
+3. RE (Извлечение отношений)
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/25809855/186094813-3a8e16cc-42e5-4982-b9f4-0134dfb5688d.png" width="600">
+</div>
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb.jpg" width="600">
+</div>
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f.jpg" width="600">
+</div>
+</details>
+
+<a name="ЛИЦЕНЗИЯ"></a>
+
+## 📄 Лицензия
+
+Этот проект выпущен под <a href="https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/LICENSE">Apache 2.0 license</a>
diff --git a/docs/javascripts/katex.js b/docs/javascripts/katex.js
new file mode 100644
index 0000000..f7fd704
--- /dev/null
+++ b/docs/javascripts/katex.js
@@ -0,0 +1,10 @@
+document$.subscribe(({ body }) => {
+  renderMathInElement(body, {
+    delimiters: [
+      { left: "$$",  right: "$$",  display: true },
+      { left: "$",   right: "$",   display: false },
+      { left: "\\(", right: "\\)", display: false },
+      { left: "\\[", right: "\\]", display: true }
+    ],
+  })
+})
diff --git a/docs/quick_start.en.md b/docs/quick_start.en.md
new file mode 100644
index 0000000..686bf12
--- /dev/null
+++ b/docs/quick_start.en.md
@@ -0,0 +1,308 @@
+---
+comments: true
+hide:
+  - navigation
+---
+
+### Installation
+
+#### 1. Install the inference engine
+
+PaddleOCR supports unified inference-engine configuration, allowing you to choose the underlying runtime. PaddleOCR currently supports PaddlePaddle and Transformers.
+
+=== "PaddlePaddle"
+
+    Installation for CPU:
+
+    ```bash
+    python -m pip install paddlepaddle==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+    ```
+
+    Installation for GPU:
+
+    Since GPU installation requires specific CUDA versions, the following example is for installing NVIDIA GPU on the Linux platform with CUDA 11.8. For other platforms, please refer to the instructions in the [PaddlePaddle official installation documentation](https://www.paddlepaddle.org.cn/install/quick).
+
+    ```bash
+    python -m pip install paddlepaddle-gpu==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cu118/
+    ```
+
+    **Please note that when using PaddlePaddle for inference, PaddleOCR 3.x depends on PaddlePaddle version `3.0` or above.**
+
+=== "Transformers"
+
+    When using Transformers for inference, install Hugging Face Transformers:
+
+    ```bash
+    python -m pip install "transformers>=5.8.0"
+    ```
+
+    In many cases, you also need to install the underlying inference framework used by Transformers. For details, see the [Transformers official installation documentation](https://huggingface.co/docs/transformers/installation).
+
+#### 2. Install `paddleocr`
+
+Install the full functionality of PaddleOCR by running the following command:
+
+```bash
+python -m pip install "paddleocr[all]"
+```
+
+PaddleOCR also supports installing specific features as needed. For details, please refer to the [PaddleOCR installation documentation](version3.x/installation.en.md).
+
+### Command Line Usage
+
+=== "PP-OCRv6"
+
+    ```bash linenums="1"
+    # Use PaddlePaddle for inference
+    paddleocr ocr -i ./general_ocr_002.png \
+        --use_doc_orientation_classify False \
+        --use_doc_unwarping False \
+        --use_textline_orientation False \
+        --engine paddle
+    ```
+
+    ```bash linenums="1"
+    # Use Transformers for inference
+    paddleocr ocr -i ./general_ocr_002.png \
+        --use_doc_orientation_classify False \
+        --use_doc_unwarping False \
+        --use_textline_orientation False \
+        --engine transformers
+    ```
+
+=== "PP-OCRv6 Text Detection Module"
+
+    ```bash linenums="1"
+    # Use PaddlePaddle for inference
+    paddleocr text_detection -i ./general_ocr_001.png --engine paddle
+    ```
+
+    ```bash linenums="1"
+    # Use Transformers for inference
+    paddleocr text_detection -i ./general_ocr_001.png --engine transformers
+    ```
+
+=== "PP-OCRv6 Text Recognition Module"
+
+    ```bash linenums="1"
+    # Use PaddlePaddle for inference
+    paddleocr text_recognition -i ./general_ocr_rec_001.png --engine paddle
+    ```
+
+    ```bash linenums="1"
+    # Use Transformers for inference
+    paddleocr text_recognition -i ./general_ocr_rec_001.png --engine transformers
+    ```
+
+=== "PP-StructureV3"
+
+    ```bash linenums="1"
+    # Use PaddlePaddle for inference
+    paddleocr pp_structurev3 -i ./pp_structure_v3_demo.png \
+        --use_doc_orientation_classify False \
+        --use_doc_unwarping False \
+        --engine paddle
+    ```
+
+    ```bash linenums="1"
+    # Use Transformers for inference
+    # Currently, some models are still being supported, so formula recognition must be disabled and the wireless table structure recognition model must be replaced.
+    paddleocr pp_structurev3 -i ./pp_structure_v3_demo.png \
+        --engine transformers \
+        --use_formula_recognition False \
+        --wireless_table_structure_recognition_model_name SLANeXt_wireless
+    ```
+
+### Python Script Usage
+
+=== "PP-OCRv6"
+
+    ```python linenums="1"
+    # Use PaddlePaddle for inference
+    from paddleocr import PaddleOCR
+
+    ocr = PaddleOCR(
+        use_doc_orientation_classify=False,
+        use_doc_unwarping=False,
+        use_textline_orientation=False,
+        engine="paddle",
+    )
+    result = ocr.predict("./general_ocr_002.png")
+    for res in result:
+        res.print()
+        res.save_to_img("output")
+        res.save_to_json("output")
+    ```
+
+    ```python linenums="1"
+    # Use Transformers for inference
+    from paddleocr import PaddleOCR
+
+    ocr = PaddleOCR(
+        use_doc_orientation_classify=False,
+        use_doc_unwarping=False,
+        use_textline_orientation=False,
+        engine="transformers",
+    )
+    result = ocr.predict("./general_ocr_002.png")
+    for res in result:
+        res.print()
+        res.save_to_img("output")
+        res.save_to_json("output")
+    ```
+
+    Example output:
+
+    ```bash
+    {'res': {'input_path': './general_ocr_002.png', 'page_index': None, 'model_settings': {'use_doc_preprocessor': True, 'use_textline_orientation': False}, 'doc_preprocessor_res': {'input_path': None, 'page_index': None, 'model_settings': {'use_doc_orientation_classify': False, 'use_doc_unwarping': False}, 'angle': -1}, 'dt_polys': array([[[  1,   4],
+        ...,
+        [  1,  33]],
+
+       ...,
+
+       [[ 99, 455],
+        ...,
+        [ 99, 480]]], dtype=int16), 'text_det_params': {'limit_side_len': 960, 'limit_type': 'max', 'thresh': 0.3, 'max_side_limit': 4000, 'box_thresh': 0.6, 'unclip_ratio': 1.5}, 'text_type': 'general', 'textline_orientation_angles': array([-1, ..., -1]), 'text_rec_score_thresh': 0.0, 'rec_texts': ['www.997788.com', '登机牌', 'BOARDING PASS', '舱位CLASS', '序号 SERIAL NO.', '座位号', 'SEAT NO', '航班FLIGHT', '日期', 'DATE', 'MU 2379', '03DEC', 'W', '035', '', '始发地', 'FROM', '登机口', 'GATE', '登机时间BDT', '目的地TO', '福州', 'TAIYUAN', 'G11', 'FUZHOU', '身份识别IDNO.', '姓名NAME', 'ZHANGQIWEI', '票号TKTNO.', '张祺伟', '票价FARE', 'ETKT7813699238489/1', '登机口于起飞前10分钟关闭 GATESCL0SE10MINUTESBEFOREDEPARTURETIME'], 'rec_scores': array([0.99684608, ..., 0.97179604]), 'rec_polys': array([[[  1,   4],
+        ...,
+        [  1,  33]],
+
+       ...,
+
+       [[ 99, 455],
+        ...,
+        [ 99, 480]]], dtype=int16), 'rec_boxes': array([[  1, ...,  33],
+       ...,
+       [ 99, ..., 480]], dtype=int16)}}
+    ```
+
+=== "PP-OCRv6 Text Detection Module"
+
+    ```python linenums="1"
+    # Use PaddlePaddle for inference
+    from paddleocr import TextDetection
+
+    model = TextDetection(engine="paddle")
+    output = model.predict("general_ocr_001.png")
+    for res in output:
+        res.print()
+        res.save_to_img(save_path="./output/")
+        res.save_to_json(save_path="./output/res.json")
+    ```
+
+    ```python linenums="1"
+    # Use Transformers for inference
+    from paddleocr import TextDetection
+
+    model = TextDetection(engine="transformers")
+    output = model.predict("general_ocr_001.png")
+    for res in output:
+        res.print()
+        res.save_to_img(save_path="./output/")
+        res.save_to_json(save_path="./output/res.json")
+    ```
+
+    ```bash
+    {'res': {'input_path': 'general_ocr_001.png', 'page_index': None, 'dt_polys': array([[[ 77, 551],
+        ...,
+        [ 78, 587]],
+
+       ...,
+
+       [[ 34, 408],
+        ...,
+        [ 36, 456]]], dtype=int16), 'dt_scores': [0.8562385635646694, 0.8818259002228059, 0.8406072284043453, 0.8855339313157491]}}
+    ```
+
+=== "PP-OCRv6 Text Recognition Module"
+
+    ```python linenums="1"
+    # Use PaddlePaddle for inference
+    from paddleocr import TextRecognition
+
+    model = TextRecognition(engine="paddle")
+    output = model.predict(input="general_ocr_rec_001.png")
+    for res in output:
+        res.print()
+        res.save_to_img(save_path="./output/")
+        res.save_to_json(save_path="./output/res.json")
+    ```
+
+    ```python linenums="1"
+    # Use Transformers for inference
+    from paddleocr import TextRecognition
+
+    model = TextRecognition(engine="transformers")
+    output = model.predict(input="general_ocr_rec_001.png")
+    for res in output:
+        res.print()
+        res.save_to_img(save_path="./output/")
+        res.save_to_json(save_path="./output/res.json")
+    ```
+
+    Example output:
+    
+    ```bash
+    {'res': {'input_path': 'general_ocr_rec_001.png', 'page_index': None, 'rec_text': '绿洲仕格维花园公寓', 'rec_score': 0.990813672542572}}
+    ```
+
+=== "PP-StructureV3"
+
+    ```python linenums="1"
+    # Use PaddlePaddle for inference
+    from paddleocr import PPStructureV3
+
+    pipeline = PPStructureV3(
+        use_doc_orientation_classify=False,
+        use_doc_unwarping=False,
+        engine="paddle",
+    )
+    output = pipeline.predict(
+        input="./pp_structure_v3_demo.png")
+    for res in output:
+        res.print()
+        res.save_to_json(save_path="output")
+        res.save_to_markdown(save_path="output")
+    ```
+
+    ```python linenums="1"
+    # Use Transformers for inference
+    # Currently, some models are still being supported, so formula recognition must be disabled and the wireless table structure recognition model must be replaced.
+    from paddleocr import PPStructureV3
+
+    pipeline = PPStructureV3(
+        use_doc_orientation_classify=False,
+        use_doc_unwarping=False,
+        use_formula_recognition=False,
+        wireless_table_structure_recognition_model_name="SLANeXt_wireless",
+        engine="transformers",
+    )
+    output = pipeline.predict(input="./pp_structure_v3_demo.png")
+    for res in output:
+        res.print()
+        res.save_to_json(save_path="output")
+        res.save_to_markdown(save_path="output")
+    ```
+
+    PaddlePaddle inference example output:
+
+    ```bash
+    {'res': {'input_path': './pp_structure_v3_demo.png', 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_seal_recognition': True, 'use_table_recognition': True, 'use_formula_recognition': True, 'use_chart_recognition': False, 'use_region_detection': True}, 'layout_det_res': {'input_path': None, 'page_index': None, 'boxes': [{'cls_id': 1, 'label': 'image', 'score': 0.9864752888679504, 'coordinate': [774.821, 201.05177, 1502.1008, 685.7733]}, {'cls_id': 2, 'label': 'text', 'score': 0.9859225749969482, 'coordinate': [769.8655, 776.2446, 1121.5986, 1058.417]}, {'cls_id': 2, 'label': 'text', 'score': 0.9857110381126404, 'coordinate': [1151.98, 1112.5356, 1502.7852, 1346.3569]}, {'cls_id': 2, 'label': 'text', 'score': 0.9847239255905151, 'coordinate': [389.0322, 1136.3547, 740.2322, 1345.928]}, {'cls_id': 2, 'label': 'text', 'score': 0.9842492938041687, 'coordinate': [1152.1504, 800.1625, 1502.1265, 986.1522]}, {'cls_id': 2, 'label': 'text', 'score': 0.9840831160545349, 'coordinate': [9.158066, 848.8696, 358.5725, 1057.832]}, {'cls_id': 2, 'label': 'text', 'score': 0.9802583456039429, 'coordinate': [9.335953, 201.10046, 358.31543, 338.78876]}, {'cls_id': 2, 'label': 'text', 'score': 0.9801402688026428, 'coordinate': [389.1556, 297.4113, 740.07556, 435.41647]}, {'cls_id': 2, 'label': 'text', 'score': 0.9793564081192017, 'coordinate': [389.18976, 752.0959, 740.0832, 889.88043]}, {'cls_id': 2, 'label': 'text', 'score': 0.9793409109115601, 'coordinate': [389.02496, 896.34143, 740.7431, 1033.9465]}, {'cls_id': 2, 'label': 'text', 'score': 0.9776486754417419, 'coordinate': [8.950775, 1184.7842, 358.75067, 1297.8755]}, {'cls_id': 2, 'label': 'text', 'score': 0.9773538708686829, 'coordinate': [770.7178, 1064.5714, 1121.2249, 1177.9928]}, {'cls_id': 2, 'label': 'text', 'score': 0.9773064255714417, 'coordinate': [389.38086, 609.7071, 740.0553, 745.3206]}, {'cls_id': 2, 'label': 'text', 'score': 0.9765821099281311, 'coordinate': [1152.0112, 992.296, 1502.4927, 1106.1166]}, {'cls_id': 2, 'label': 'text', 'score': 0.9761461019515991, 'coordinate': [9.46727, 536.993, 358.2047, 651.32025]}, {'cls_id': 2, 'label': 'text', 'score': 0.975399911403656, 'coordinate': [9.353531, 1064.3059, 358.45312, 1177.8347]}, {'cls_id': 2, 'label': 'text', 'score': 0.9730532169342041, 'coordinate': [9.932312, 345.36237, 358.03476, 435.1646]}, {'cls_id': 2, 'label': 'text', 'score': 0.9722575545310974, 'coordinate': [388.91736, 200.93637, 740.00793, 290.80692]}, {'cls_id': 2, 'label': 'text', 'score': 0.9710633158683777, 'coordinate': [389.39496, 1040.3186, 740.0091, 1129.7168]}, {'cls_id': 2, 'label': 'text', 'score': 0.9696939587593079, 'coordinate': [9.6145935, 658.1123, 359.06088, 770.0288]}, {'cls_id': 2, 'label': 'text', 'score': 0.9664146900177002, 'coordinate': [770.235, 1280.4562, 1122.0927, 1346.4742]}, {'cls_id': 2, 'label': 'text', 'score': 0.9597565531730652, 'coordinate': [389.66678, 537.5609, 740.06274, 603.17725]}, {'cls_id': 2, 'label': 'text', 'score': 0.9594324827194214, 'coordinate': [10.162949, 776.86414, 359.08307, 842.1771]}, {'cls_id': 2, 'label': 'text', 'score': 0.9484634399414062, 'coordinate': [10.402863, 1304.7743, 358.9441, 1346.3749]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.9476125240325928, 'coordinate': [28.159409, 456.7627, 339.5631, 514.9665]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.9427680969238281, 'coordinate': [790.6992, 1200.3663, 1102.3799, 1259.1647]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.9424256682395935, 'coordinate': [409.02832, 456.6831, 718.8154, 515.5757]}, {'cls_id': 10, 'label': 'doc_title', 'score': 0.9376171827316284, 'coordinate': [133.77905, 36.8844, 1379.6667, 123.46869]}, {'cls_id': 2, 'label': 'text', 'score': 0.9020252823829651, 'coordinate': [584.9165, 159.1416, 927.22876, 179.01605]}, {'cls_id': 2, 'label': 'text', 'score': 0.895164430141449, 'coordinate': [1154.3364, 776.74646, 1331.8564, 794.2301]}, {'cls_id': 6, 'label': 'figure_title', 'score': 0.7892374396324158, 'coordinate': [808.9641, 704.2555, 1484.0623, 747.2296]}]}, 'overall_ocr_res': {'input_path': None, 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_textline_orientation': False}, 'dt_polys': array([[[ 129,   42],
+            ...,
+            [ 129,  140]],
+
+        ...,
+
+        [[1156, 1330],
+            ...,
+            [1156, 1351]]], dtype=int16), 'text_det_params': {'limit_side_len': 736, 'limit_type': 'min', 'thresh': 0.3, 'max_side_limit': 4000, 'box_thresh': 0.6, 'unclip_ratio': 1.5}, 'text_type': 'general', 'textline_orientation_angles': array([-1, ..., -1]), 'text_rec_score_thresh': 0.0, 'rec_texts': ['助力双方交往', '搭建友谊桥梁', '本报记者沈小晓', '任', '彦', '黄培昭', '身着中国传统民族服装的厄立特里亚青', '厄立特里亚高等教育与研究院合作建立，开', '年依次登台表演中国民族舞、现代舞、扇子舞', '设了中国语言课程和中国文化课程，注册学', '等，曼妙的舞姿赢得现场观众阵阵掌声。这', '生2万余人次。10余年来，厄特孔院已成为', '是日前危立特里亚高等教育与研究院孔子学', '当地民众了解中国的一扇窗口。', '院(以下简称“厄特孔院")举办“喜迎新年"中国', '黄鸣飞表示，随着来学习中文的人日益', '歌舞比赛的场景。', '增多，阿斯马拉大学教学点已难以满足教学', '中国和厄立特里亚传统友谊深厚。近年', '需要。2024年4月，由中企蜀道集团所属四', '来，在高质量共建“一带一路”框架下，中厄两', '川路桥承建的孔院教学楼项目在阿斯马拉开', '国人文交流不断深化，互利合作的民意基础', '工建设，预计今年上半年竣工，建成后将为危', '日益深厚。', '特孔院提供全新的办学场地。', '“学好中文，我们的', '“在中国学习的经历', '未来不是梦”', '让我看到更广阔的世界”', '“鲜花曾告诉我你怎样走过，大地知道你', '多年来，厄立特里亚广大赴华留学生和', '心中的每一个角落……"厄立特里亚阿斯马拉', '培训人员积极投身国家建设，成为助力该国', '大学综合楼二层，一阵优美的歌声在走廊里回', '发展的人才和厄中友好的见证者和推动者。', '响。循着熟悉的旋律轻轻推开一间教室的门，', '在厄立特里亚全国妇女联盟工作的约翰', '学生们正跟着老师学唱中文歌曲《同一首歌》。', '娜·特韦尔德·凯莱塔就是其中一位。她曾在', '这是厄特孔院阿斯马拉大学教学点的一', '中华女子学院攻读硕士学位，研究方向是女', '节中文歌曲课。为了让学生们更好地理解歌', '性领导力与社会发展。其间，她实地走访中国', '词大意，老师尤斯拉·穆罕默德萨尔·侯赛因逐', '多个地区，获得了观察中国社会发展的第一', '在厄立特里亚不久前举办的第六届中国风筝文化节上，当地小学生体验风筝制作。', '字翻译和解释歌词。随着伴奏声响起，学生们', '手资料。', '中国驻厄立特里亚大使馆供图', '边唱边随着节拍摇动身体，现场气氛热烈。', '谈起在中国求学的经历，约翰娜记忆犹', '“这是中文歌曲初级班，共有32人。学', '新：“中国的发展在当今世界是独一无二的。', '“不管远近都是客人，请不用客气；相约', '瓦的北红海省博物馆。', '生大部分来自首都阿斯马拉的中小学，年龄', '沿着中国特色社会主义道路坚定前行，中国', '好了在一起，我们欢迎你……”在一场中厄青', '博物馆二层陈列着一个发掘自阿杜利', '最小的仅有6岁。”尤斯拉告诉记者。', '创造了发展奇迹，这一切都离不开中国共产党', '年联谊活动上，四川路桥中方员工同当地大', '斯古城的中国古代陶制酒器，罐身上写着', '尤斯拉今年23岁，是厄立特里亚一所公立', '的领导。中国的发展经验值得许多国家学习', '学生合唱《北京欢迎你》。厄立特里亚技术学', '“万”“和”“禅”“山”等汉字。“这件文物证', '学校的艺术老师。她12岁开始在厄特孔院学', '借鉴。”', '院计算机科学与工程专业学生鲁夫塔·谢拉', '明，很早以前我们就通过海上丝绸之路进行', '习中文，在2017年第十届“汉语桥"世界中学生', '正在西南大学学习的厄立特里亚博士生', '是其中一名演唱者，她很早便在孔院学习中', '贸易往来与文化交流。这也是厄立特里亚', '中文比赛中获得厄立特里亚赛区第一名，并和', '穆卢盖塔·泽穆伊对中国怀有深厚感情。8', '文，一直在为去中国留学作准备。“这句歌词', '与中国友好交往历史的有力证明。”北红海', '同伴代表厄立特里亚前往中国参加决赛，获得', '年前，在北京师范大学获得硕士学位后，穆卢', '是我们两国人民友谊的生动写照。无论是投', '省博物馆研究与文献部负责人伊萨亚斯·特', '团体优胜奖。2022年起，尤斯拉开始在厄特孔', '盖塔在社交媒体上写下这样一段话：“这是我', '身于厄立特里亚基础设施建设的中企员工，', '斯法兹吉说。', '院兼职教授中文歌曲，每周末两个课时。“中国', '人生的重要一步，自此我拥有了一双坚固的', '还是在中国留学的厄立特里亚学子，两国人', '厄立特里亚国家博物馆考古学和人类学', '文化博大精深，我希望我的学生们能够通过中', '鞋子，赋予我穿越荆棘的力量。”', '民携手努力，必将推动两国关系不断向前发', '研究员菲尔蒙·特韦尔德十分喜爱中国文', '文歌曲更好地理解中国文化。”她说。', '穆卢盖塔密切关注中国在经济、科技、教', '展。”鲁夫塔说。', '化。他表示：“学习彼此的语言和文化，将帮', '“姐姐，你想去中国吗?”“非常想！我想', '育等领域的发展，“中国在科研等方面的实力', '厄立特里亚高等教育委员会主任助理萨', '助厄中两国人民更好地理解彼此，助力双方', '去看故宫、爬长城。”尤斯拉的学生中有一对', '与日俱增。在中国学习的经历让我看到更广', '马瑞表示：“每年我们都会组织学生到中国访', '交往，搭建友谊桥梁。”', '能歌善舞的姐妹，姐姐露娅今年15岁，妹妹', '阔的世界，从中受益匪浅。”', '问学习，自前有超过5000名厄立特里亚学生', '厄立特里亚国家博物馆馆长塔吉丁·努', '莉娅14岁，两人都已在厄特孔院学习多年，', '23岁的莉迪亚·埃斯蒂法诺斯已在厄特', '在中国留学。学习中国的教育经验，有助于', '重达姆·优素福曾多次访问中国，对中华文明', '中文说得格外流利。', '孔院学习3年，在中国书法、中国画等方面表', '提升厄立特里亚的教育水平。”', '的传承与创新、现代化博物馆的建设与发展', '露娅对记者说：“这些年来，怀着对中文', '现干分优秀，在2024年厄立特里亚赛区的', '印象深刻。“中国博物馆不仅有许多保存完好', '“共同向世界展示非', '和中国文化的热爱，我们姐妹俩始终相互鼓', '“汉语桥”比赛中获得一等奖。莉迪亚说：“学', '的文物，还充分运用先进科技手段进行展示，', '励，一起学习。我们的中文一天比一天好，还', '习中国书法让我的内心变得安宁和纯粹。我', '洲和亚洲的灿烂文明”', '帮助人们更好理解中华文明。”塔吉丁说，“危', '学会了中文歌和中国舞。我们一定要到中国', '也喜欢中国的服饰，希望未来能去中国学习，', '立特里亚与中国都拥有悠久的文明，始终相', '去。学好中文，我们的未来不是梦！”', '把中国不同民族元素融入服装设计中，创作', '从阿斯马拉出发，沿着蜿蜒曲折的盘山', '互理解、相互尊重。我希望未来与中国同行', '据厄特孔院中方院长黄鸣飞介绍，这所', '出更多精美作品，也把厄特文化分享给更多', '公路一路向东寻找丝路印迹。驱车两个小', '加强合作，共同向世界展示非洲和亚洲的灿', '孔院成立于2013年3月，由贵州财经大学和', '的中国朋友。”', '时，记者来到位于厄立特里亚港口城市马萨', '烂文明。”'], 'rec_scores': array([0.99113536, ..., 0.95110035]), 'rec_polys': array([[[ 129,   42],
+            ...,
+            [ 129,  140]],
+
+        ...,
+
+        [[1156, 1330],
+            ...,
+            [1156, 1351]]], dtype=int16), 'rec_boxes': array([[ 129, ...,  140],
+        ...,
+        [1156, ..., 1351]], dtype=int16)}}}
+    ```
diff --git a/docs/quick_start.md b/docs/quick_start.md
new file mode 100644
index 0000000..854782e
--- /dev/null
+++ b/docs/quick_start.md
@@ -0,0 +1,307 @@
+---
+comments: true
+hide:
+  - navigation
+---
+
+### 安装
+
+#### 1. 安装推理引擎
+
+PaddleOCR 支持通过统一的推理引擎配置选择底层运行时，目前可使用 PaddlePaddle 或 Transformers。
+
+=== "PaddlePaddle"
+
+    CPU端安装：
+
+    ```bash
+    python -m pip install paddlepaddle==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+    ```
+
+    GPU端安装，由于GPU端需要根据具体CUDA版本来对应安装使用，以下仅以Linux平台，pip安装英伟达GPU， CUDA 11.8为例，其他平台，请参考[飞桨官网安装文档](https://www.paddlepaddle.org.cn/install/quick)中的说明进行操作。
+
+    ```bash
+    python -m pip install paddlepaddle-gpu==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cu118/
+    ```
+
+    **请注意，使用 PaddlePaddle 推理时，PaddleOCR 3.x 版本依赖于 `3.0` 及以上版本的 PaddlePaddle。**
+
+=== "Transformers"
+
+    使用 Transformers 推理时，需要安装 Hugging Face Transformers：
+
+    ```bash
+    python -m pip install "transformers>=5.8.0"
+    ```
+
+    通常还需要安装 Transformers 所使用的底层推理框架，详情可参考 [Transformers 官方安装文档](https://huggingface.co/docs/transformers/installation)。
+
+#### 2. 安装`paddleocr`
+
+执行如下命令安装 PaddleOCR 的完整功能：
+
+```bash
+python -m pip install "paddleocr[all]"
+```
+
+PaddleOCR 也支持根据需要安装部分功能，详情请参考 [PaddleOCR 安装文档](version3.x/installation.md)。
+
+### 命令行使用
+
+=== "PP-OCRv6"
+
+    ```bash linenums="1"
+    # 使用 PaddlePaddle 进行推理
+    paddleocr ocr -i ./general_ocr_002.png \
+        --use_doc_orientation_classify False \
+        --use_doc_unwarping False \
+        --use_textline_orientation False \
+        --engine paddle
+    ```
+
+    ```bash linenums="1"
+    # 使用 Transformers 进行推理
+    paddleocr ocr -i ./general_ocr_002.png \
+        --use_doc_orientation_classify False \
+        --use_doc_unwarping False \
+        --use_textline_orientation False \
+        --engine transformers
+    ```
+
+=== "PP-OCRv6文本检测模块"
+
+    ```bash linenums="1"
+    # 使用 PaddlePaddle 进行推理
+    paddleocr text_detection -i ./general_ocr_001.png --engine paddle
+    ```
+
+    ```bash linenums="1"
+    # 使用 Transformers 进行推理
+    paddleocr text_detection -i ./general_ocr_001.png --engine transformers
+    ```
+
+=== "PP-OCRv6文本识别模块"
+
+    ```bash linenums="1"
+    # 使用 PaddlePaddle 进行推理
+    paddleocr text_recognition -i ./general_ocr_rec_001.png --engine paddle
+    ```
+
+    ```bash linenums="1"
+    # 使用 Transformers 进行推理
+    paddleocr text_recognition -i ./general_ocr_rec_001.png --engine transformers
+    ```
+
+=== "PP-StructureV3"
+
+    ```bash linenums="1"
+    # 使用 PaddlePaddle 进行推理
+    paddleocr pp_structurev3 -i ./pp_structure_v3_demo.png \
+        --use_doc_orientation_classify False \
+        --use_doc_unwarping False \
+        --engine paddle
+    ```
+
+    ```bash linenums="1"
+    # 使用 Transformers 进行推理
+    # 目前部分模型尚在支持中，需关闭公式识别功能并更换无线表格结构识别模型
+    paddleocr pp_structurev3 -i ./pp_structure_v3_demo.png \
+        --engine transformers \
+        --use_formula_recognition False \
+        --wireless_table_structure_recognition_model_name SLANeXt_wireless
+    ```
+
+### Python脚本使用
+
+=== "PP-OCRv6"
+
+    ```python linenums="1"
+    # 使用 PaddlePaddle 进行推理
+    from paddleocr import PaddleOCR
+
+    ocr = PaddleOCR(
+        use_doc_orientation_classify=False,
+        use_doc_unwarping=False,
+        use_textline_orientation=False,
+        engine="paddle",
+    )
+    result = ocr.predict("./general_ocr_002.png")
+    for res in result:
+        res.print()
+        res.save_to_img("output")
+        res.save_to_json("output")
+    ```
+
+    ```python linenums="1"
+    # 使用 Transformers 进行推理
+    from paddleocr import PaddleOCR
+
+    ocr = PaddleOCR(
+        use_doc_orientation_classify=False,
+        use_doc_unwarping=False,
+        use_textline_orientation=False,
+        engine="transformers",
+    )
+    result = ocr.predict("./general_ocr_002.png")
+    for res in result:
+        res.print()
+        res.save_to_img("output")
+        res.save_to_json("output")
+    ```
+
+    输出示例：
+
+    ```bash
+    {'res': {'input_path': '/root/.paddlex/predict_input/general_ocr_002.png', 'page_index': None, 'model_settings': {'use_doc_preprocessor': True, 'use_textline_orientation': False}, 'doc_preprocessor_res': {'input_path': None, 'page_index': None, 'model_settings': {'use_doc_orientation_classify': False, 'use_doc_unwarping': False}, 'angle': -1}, 'dt_polys': array([[[  3,  10],
+            ...,
+            [  4,  30]],
+
+        ...,
+
+        [[ 99, 456],
+            ...,
+            [ 99, 479]]], dtype=int16), 'text_det_params': {'limit_side_len': 736, 'limit_type': 'min', 'thresh': 0.3, 'max_side_limit': 4000, 'box_thresh': 0.6, 'unclip_ratio': 1.5}, 'text_type': 'general', 'textline_orientation_angles': array([-1, ..., -1]), 'text_rec_score_thresh': 0.0, 'rec_texts': ['www.997700', '', 'Cm', '登机牌', 'BOARDING', 'PASS', 'CLASS', '序号SERIAL NO.', '座位号', 'SEAT NO.', '航班FLIGHT', '日期DATE', '舱位', '', 'W', '035', '12F', 'MU2379', '03DEc', '始发地', 'FROM', '登机口', 'GATE', '登机时间BDT', '目的地TO', '福州', 'TAIYUAN', 'G11', 'FUZHOU', '身份识别IDNO.', '姓名NAME', 'ZHANGQIWEI', '票号TKT NO.', '张祺伟', '票价FARE', 'ETKT7813699238489/1', '登机口于起飞前10分钟关闭 GATESCL0SE10MINUTESBEFOREDEPARTURETIME'], 'rec_scores': array([0.67582953, ..., 0.97418666]), 'rec_polys': array([[[  3,  10],
+            ...,
+            [  4,  30]],
+
+        ...,
+
+        [[ 99, 456],
+            ...,
+            [ 99, 479]]], dtype=int16), 'rec_boxes': array([[  3, ...,  30],
+        ...,
+        [ 99, ..., 479]], dtype=int16)}}
+    ```
+
+=== "PP-OCRv6文本检测模块"
+
+    ```python linenums="1"
+    # 使用 PaddlePaddle 进行推理
+    from paddleocr import TextDetection
+
+    model = TextDetection(engine="paddle")
+    output = model.predict("general_ocr_001.png")
+    for res in output:
+        res.print()
+        res.save_to_img(save_path="./output/")
+        res.save_to_json(save_path="./output/res.json")
+    ```
+
+    ```python linenums="1"
+    # 使用 Transformers 进行推理
+    from paddleocr import TextDetection
+
+    model = TextDetection(engine="transformers")
+    output = model.predict("general_ocr_001.png")
+    for res in output:
+        res.print()
+        res.save_to_img(save_path="./output/")
+        res.save_to_json(save_path="./output/res.json")
+    ```
+
+    ```bash
+    {'res': {'input_path': 'general_ocr_001.png', 'page_index': None, 'dt_polys': array([[[ 77, 551],
+        ...,
+        [ 78, 587]],
+
+       ...,
+
+       [[ 34, 408],
+        ...,
+        [ 36, 456]]], dtype=int16), 'dt_scores': [0.8562385635646694, 0.8818259002228059, 0.8406072284043453, 0.8855339313157491]}}
+    ```
+
+=== "PP-OCRv6文本识别模块"
+
+    ```python linenums="1"
+    # 使用 PaddlePaddle 进行推理
+    from paddleocr import TextRecognition
+
+    model = TextRecognition(engine="paddle")
+    output = model.predict(input="general_ocr_rec_001.png")
+    for res in output:
+        res.print()
+        res.save_to_img(save_path="./output/")
+        res.save_to_json(save_path="./output/res.json")
+    ```
+
+    ```python linenums="1"
+    # 使用 Transformers 进行推理
+    from paddleocr import TextRecognition
+
+    model = TextRecognition(engine="transformers")
+    output = model.predict(input="general_ocr_rec_001.png")
+    for res in output:
+        res.print()
+        res.save_to_img(save_path="./output/")
+        res.save_to_json(save_path="./output/res.json")
+    ```
+
+    输出示例：
+    
+    ```bash
+    {'res': {'input_path': 'general_ocr_rec_001.png', 'page_index': None, 'rec_text': '绿洲仕格维花园公寓', 'rec_score': 0.990813672542572}}
+    ```
+
+=== "PP-StructureV3"
+
+    ```python linenums="1"
+    # 使用 PaddlePaddle 进行推理
+    from paddleocr import PPStructureV3
+
+    pipeline = PPStructureV3(
+        use_doc_orientation_classify=False,
+        use_doc_unwarping=False,
+        engine="paddle",
+    )
+    output = pipeline.predict(
+        input="./pp_structure_v3_demo.png",
+    )
+    for res in output:
+        res.print()
+        res.save_to_json(save_path="output")
+        res.save_to_markdown(save_path="output")
+    ```
+
+    ```python linenums="1"
+    # 使用 Transformers 进行推理
+    # 目前部分模型尚在支持中，需关闭公式识别功能并更换无线表格结构识别模型
+    from paddleocr import PPStructureV3
+
+    pipeline = PPStructureV3(
+        use_doc_orientation_classify=False,
+        use_doc_unwarping=False,
+        use_formula_recognition=False,
+        wireless_table_structure_recognition_model_name="SLANeXt_wireless",
+        engine="transformers",
+    )
+    output = pipeline.predict(input="./pp_structure_v3_demo.png")
+    for res in output:
+        res.print()
+        res.save_to_json(save_path="output")
+        res.save_to_markdown(save_path="output")
+    ```
+
+    PaddlePaddle 推理输出示例：
+
+    ```bash
+    {'res': {'input_path': './pp_structure_v3_demo.png', 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_seal_recognition': True, 'use_table_recognition': True, 'use_formula_recognition': True, 'use_chart_recognition': False, 'use_region_detection': True}, 'layout_det_res': {'input_path': None, 'page_index': None, 'boxes': [{'cls_id': 1, 'label': 'image', 'score': 0.9864752888679504, 'coordinate': [774.821, 201.05177, 1502.1008, 685.7733]}, {'cls_id': 2, 'label': 'text', 'score': 0.9859225749969482, 'coordinate': [769.8655, 776.2446, 1121.5986, 1058.417]}, {'cls_id': 2, 'label': 'text', 'score': 0.9857110381126404, 'coordinate': [1151.98, 1112.5356, 1502.7852, 1346.3569]}, {'cls_id': 2, 'label': 'text', 'score': 0.9847239255905151, 'coordinate': [389.0322, 1136.3547, 740.2322, 1345.928]}, {'cls_id': 2, 'label': 'text', 'score': 0.9842492938041687, 'coordinate': [1152.1504, 800.1625, 1502.1265, 986.1522]}, {'cls_id': 2, 'label': 'text', 'score': 0.9840831160545349, 'coordinate': [9.158066, 848.8696, 358.5725, 1057.832]}, {'cls_id': 2, 'label': 'text', 'score': 0.9802583456039429, 'coordinate': [9.335953, 201.10046, 358.31543, 338.78876]}, {'cls_id': 2, 'label': 'text', 'score': 0.9801402688026428, 'coordinate': [389.1556, 297.4113, 740.07556, 435.41647]}, {'cls_id': 2, 'label': 'text', 'score': 0.9793564081192017, 'coordinate': [389.18976, 752.0959, 740.0832, 889.88043]}, {'cls_id': 2, 'label': 'text', 'score': 0.9793409109115601, 'coordinate': [389.02496, 896.34143, 740.7431, 1033.9465]}, {'cls_id': 2, 'label': 'text', 'score': 0.9776486754417419, 'coordinate': [8.950775, 1184.7842, 358.75067, 1297.8755]}, {'cls_id': 2, 'label': 'text', 'score': 0.9773538708686829, 'coordinate': [770.7178, 1064.5714, 1121.2249, 1177.9928]}, {'cls_id': 2, 'label': 'text', 'score': 0.9773064255714417, 'coordinate': [389.38086, 609.7071, 740.0553, 745.3206]}, {'cls_id': 2, 'label': 'text', 'score': 0.9765821099281311, 'coordinate': [1152.0112, 992.296, 1502.4927, 1106.1166]}, {'cls_id': 2, 'label': 'text', 'score': 0.9761461019515991, 'coordinate': [9.46727, 536.993, 358.2047, 651.32025]}, {'cls_id': 2, 'label': 'text', 'score': 0.975399911403656, 'coordinate': [9.353531, 1064.3059, 358.45312, 1177.8347]}, {'cls_id': 2, 'label': 'text', 'score': 0.9730532169342041, 'coordinate': [9.932312, 345.36237, 358.03476, 435.1646]}, {'cls_id': 2, 'label': 'text', 'score': 0.9722575545310974, 'coordinate': [388.91736, 200.93637, 740.00793, 290.80692]}, {'cls_id': 2, 'label': 'text', 'score': 0.9710633158683777, 'coordinate': [389.39496, 1040.3186, 740.0091, 1129.7168]}, {'cls_id': 2, 'label': 'text', 'score': 0.9696939587593079, 'coordinate': [9.6145935, 658.1123, 359.06088, 770.0288]}, {'cls_id': 2, 'label': 'text', 'score': 0.9664146900177002, 'coordinate': [770.235, 1280.4562, 1122.0927, 1346.4742]}, {'cls_id': 2, 'label': 'text', 'score': 0.9597565531730652, 'coordinate': [389.66678, 537.5609, 740.06274, 603.17725]}, {'cls_id': 2, 'label': 'text', 'score': 0.9594324827194214, 'coordinate': [10.162949, 776.86414, 359.08307, 842.1771]}, {'cls_id': 2, 'label': 'text', 'score': 0.9484634399414062, 'coordinate': [10.402863, 1304.7743, 358.9441, 1346.3749]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.9476125240325928, 'coordinate': [28.159409, 456.7627, 339.5631, 514.9665]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.9427680969238281, 'coordinate': [790.6992, 1200.3663, 1102.3799, 1259.1647]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.9424256682395935, 'coordinate': [409.02832, 456.6831, 718.8154, 515.5757]}, {'cls_id': 10, 'label': 'doc_title', 'score': 0.9376171827316284, 'coordinate': [133.77905, 36.8844, 1379.6667, 123.46869]}, {'cls_id': 2, 'label': 'text', 'score': 0.9020252823829651, 'coordinate': [584.9165, 159.1416, 927.22876, 179.01605]}, {'cls_id': 2, 'label': 'text', 'score': 0.895164430141449, 'coordinate': [1154.3364, 776.74646, 1331.8564, 794.2301]}, {'cls_id': 6, 'label': 'figure_title', 'score': 0.7892374396324158, 'coordinate': [808.9641, 704.2555, 1484.0623, 747.2296]}]}, 'overall_ocr_res': {'input_path': None, 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_textline_orientation': False}, 'dt_polys': array([[[ 129,   42],
+            ...,
+            [ 129,  140]],
+
+        ...,
+
+        [[1156, 1330],
+            ...,
+            [1156, 1351]]], dtype=int16), 'text_det_params': {'limit_side_len': 736, 'limit_type': 'min', 'thresh': 0.3, 'max_side_limit': 4000, 'box_thresh': 0.6, 'unclip_ratio': 1.5}, 'text_type': 'general', 'textline_orientation_angles': array([-1, ..., -1]), 'text_rec_score_thresh': 0.0, 'rec_texts': ['助力双方交往', '搭建友谊桥梁', '本报记者沈小晓', '任', '彦', '黄培昭', '身着中国传统民族服装的厄立特里亚青', '厄立特里亚高等教育与研究院合作建立，开', '年依次登台表演中国民族舞、现代舞、扇子舞', '设了中国语言课程和中国文化课程，注册学', '等，曼妙的舞姿赢得现场观众阵阵掌声。这', '生2万余人次。10余年来，厄特孔院已成为', '是日前危立特里亚高等教育与研究院孔子学', '当地民众了解中国的一扇窗口。', '院(以下简称“厄特孔院")举办“喜迎新年"中国', '黄鸣飞表示，随着来学习中文的人日益', '歌舞比赛的场景。', '增多，阿斯马拉大学教学点已难以满足教学', '中国和厄立特里亚传统友谊深厚。近年', '需要。2024年4月，由中企蜀道集团所属四', '来，在高质量共建“一带一路”框架下，中厄两', '川路桥承建的孔院教学楼项目在阿斯马拉开', '国人文交流不断深化，互利合作的民意基础', '工建设，预计今年上半年竣工，建成后将为危', '日益深厚。', '特孔院提供全新的办学场地。', '“学好中文，我们的', '“在中国学习的经历', '未来不是梦”', '让我看到更广阔的世界”', '“鲜花曾告诉我你怎样走过，大地知道你', '多年来，厄立特里亚广大赴华留学生和', '心中的每一个角落……"厄立特里亚阿斯马拉', '培训人员积极投身国家建设，成为助力该国', '大学综合楼二层，一阵优美的歌声在走廊里回', '发展的人才和厄中友好的见证者和推动者。', '响。循着熟悉的旋律轻轻推开一间教室的门，', '在厄立特里亚全国妇女联盟工作的约翰', '学生们正跟着老师学唱中文歌曲《同一首歌》。', '娜·特韦尔德·凯莱塔就是其中一位。她曾在', '这是厄特孔院阿斯马拉大学教学点的一', '中华女子学院攻读硕士学位，研究方向是女', '节中文歌曲课。为了让学生们更好地理解歌', '性领导力与社会发展。其间，她实地走访中国', '词大意，老师尤斯拉·穆罕默德萨尔·侯赛因逐', '多个地区，获得了观察中国社会发展的第一', '在厄立特里亚不久前举办的第六届中国风筝文化节上，当地小学生体验风筝制作。', '字翻译和解释歌词。随着伴奏声响起，学生们', '手资料。', '中国驻厄立特里亚大使馆供图', '边唱边随着节拍摇动身体，现场气氛热烈。', '谈起在中国求学的经历，约翰娜记忆犹', '“这是中文歌曲初级班，共有32人。学', '新：“中国的发展在当今世界是独一无二的。', '“不管远近都是客人，请不用客气；相约', '瓦的北红海省博物馆。', '生大部分来自首都阿斯马拉的中小学，年龄', '沿着中国特色社会主义道路坚定前行，中国', '好了在一起，我们欢迎你……”在一场中厄青', '博物馆二层陈列着一个发掘自阿杜利', '最小的仅有6岁。”尤斯拉告诉记者。', '创造了发展奇迹，这一切都离不开中国共产党', '年联谊活动上，四川路桥中方员工同当地大', '斯古城的中国古代陶制酒器，罐身上写着', '尤斯拉今年23岁，是厄立特里亚一所公立', '的领导。中国的发展经验值得许多国家学习', '学生合唱《北京欢迎你》。厄立特里亚技术学', '“万”“和”“禅”“山”等汉字。“这件文物证', '学校的艺术老师。她12岁开始在厄特孔院学', '借鉴。”', '院计算机科学与工程专业学生鲁夫塔·谢拉', '明，很早以前我们就通过海上丝绸之路进行', '习中文，在2017年第十届“汉语桥"世界中学生', '正在西南大学学习的厄立特里亚博士生', '是其中一名演唱者，她很早便在孔院学习中', '贸易往来与文化交流。这也是厄立特里亚', '中文比赛中获得厄立特里亚赛区第一名，并和', '穆卢盖塔·泽穆伊对中国怀有深厚感情。8', '文，一直在为去中国留学作准备。“这句歌词', '与中国友好交往历史的有力证明。”北红海', '同伴代表厄立特里亚前往中国参加决赛，获得', '年前，在北京师范大学获得硕士学位后，穆卢', '是我们两国人民友谊的生动写照。无论是投', '省博物馆研究与文献部负责人伊萨亚斯·特', '团体优胜奖。2022年起，尤斯拉开始在厄特孔', '盖塔在社交媒体上写下这样一段话：“这是我', '身于厄立特里亚基础设施建设的中企员工，', '斯法兹吉说。', '院兼职教授中文歌曲，每周末两个课时。“中国', '人生的重要一步，自此我拥有了一双坚固的', '还是在中国留学的厄立特里亚学子，两国人', '厄立特里亚国家博物馆考古学和人类学', '文化博大精深，我希望我的学生们能够通过中', '鞋子，赋予我穿越荆棘的力量。”', '民携手努力，必将推动两国关系不断向前发', '研究员菲尔蒙·特韦尔德十分喜爱中国文', '文歌曲更好地理解中国文化。”她说。', '穆卢盖塔密切关注中国在经济、科技、教', '展。”鲁夫塔说。', '化。他表示：“学习彼此的语言和文化，将帮', '“姐姐，你想去中国吗?”“非常想！我想', '育等领域的发展，“中国在科研等方面的实力', '厄立特里亚高等教育委员会主任助理萨', '助厄中两国人民更好地理解彼此，助力双方', '去看故宫、爬长城。”尤斯拉的学生中有一对', '与日俱增。在中国学习的经历让我看到更广', '马瑞表示：“每年我们都会组织学生到中国访', '交往，搭建友谊桥梁。”', '能歌善舞的姐妹，姐姐露娅今年15岁，妹妹', '阔的世界，从中受益匪浅。”', '问学习，自前有超过5000名厄立特里亚学生', '厄立特里亚国家博物馆馆长塔吉丁·努', '莉娅14岁，两人都已在厄特孔院学习多年，', '23岁的莉迪亚·埃斯蒂法诺斯已在厄特', '在中国留学。学习中国的教育经验，有助于', '重达姆·优素福曾多次访问中国，对中华文明', '中文说得格外流利。', '孔院学习3年，在中国书法、中国画等方面表', '提升厄立特里亚的教育水平。”', '的传承与创新、现代化博物馆的建设与发展', '露娅对记者说：“这些年来，怀着对中文', '现干分优秀，在2024年厄立特里亚赛区的', '印象深刻。“中国博物馆不仅有许多保存完好', '“共同向世界展示非', '和中国文化的热爱，我们姐妹俩始终相互鼓', '“汉语桥”比赛中获得一等奖。莉迪亚说：“学', '的文物，还充分运用先进科技手段进行展示，', '励，一起学习。我们的中文一天比一天好，还', '习中国书法让我的内心变得安宁和纯粹。我', '洲和亚洲的灿烂文明”', '帮助人们更好理解中华文明。”塔吉丁说，“危', '学会了中文歌和中国舞。我们一定要到中国', '也喜欢中国的服饰，希望未来能去中国学习，', '立特里亚与中国都拥有悠久的文明，始终相', '去。学好中文，我们的未来不是梦！”', '把中国不同民族元素融入服装设计中，创作', '从阿斯马拉出发，沿着蜿蜒曲折的盘山', '互理解、相互尊重。我希望未来与中国同行', '据厄特孔院中方院长黄鸣飞介绍，这所', '出更多精美作品，也把厄特文化分享给更多', '公路一路向东寻找丝路印迹。驱车两个小', '加强合作，共同向世界展示非洲和亚洲的灿', '孔院成立于2013年3月，由贵州财经大学和', '的中国朋友。”', '时，记者来到位于厄立特里亚港口城市马萨', '烂文明。”'], 'rec_scores': array([0.99113536, ..., 0.95110035]), 'rec_polys': array([[[ 129,   42],
+            ...,
+            [ 129,  140]],
+
+        ...,
+
+        [[1156, 1330],
+            ...,
+            [1156, 1351]]], dtype=int16), 'rec_boxes': array([[ 129, ...,  140],
+        ...,
+        [1156, ..., 1351]], dtype=int16)}}}
+    ```
diff --git a/docs/update/update.en.md b/docs/update/update.en.md
new file mode 100644
index 0000000..0c9d932
--- /dev/null
+++ b/docs/update/update.en.md
@@ -0,0 +1,244 @@
+---
+comments: true
+hide:
+  - navigation
+  - toc
+---
+
+### Recently Update
+
+#### 🔥🔥**2025.08.21: Release of PaddleOCR 3.2.0**, includes
+
+- **Significant Model Additions:**
+    - Introduced training, inference, and deployment for PP-OCRv5 recognition models in English, Thai, and Greek. **The PP-OCRv5 English model delivers an 11% improvement in English scenarios compared to the main PP-OCRv5 model, with the Thai and Greek recognition models achieving accuracies of 82.68% and 89.28%, respectively.**
+
+- **Deployment Capability Upgrades:**
+    - **Full support for PaddlePaddle framework versions 3.1.0 and 3.1.1.**
+    - **Comprehensive upgrade of the PP-OCRv5 C++ local deployment solution, now supporting both Linux and Windows, with feature parity and identical accuracy to the Python implementation.**
+    - **High-performance inference now supports CUDA 12, and inference can be performed using either the Paddle Inference or ONNX Runtime backends.**
+    - **The high-stability service-oriented deployment solution is now fully open-sourced, allowing users to customize Docker images and SDKs as required.**
+    - The high-stability service-oriented deployment solution also supports invocation via manually constructed HTTP requests, enabling client-side code development in any programming language.
+
+- **Benchmark Support:**
+    - **All production lines now support fine-grained benchmarking, enabling measurement of end-to-end inference time as well as per-layer and per-module latency data to assist with performance analysis.**
+    - **Documentation has been updated to include key metrics for commonly used configurations on mainstream hardware, such as inference latency and memory usage, providing deployment references for users.**
+
+- **Bug Fixes:**
+    - Resolved the issue of failed log saving during model training.
+    - Upgraded the data augmentation component for formula models for compatibility with newer versions of the albumentations dependency, and fixed deadlock warnings when using the tokenizers package in multi-process scenarios.
+    - Fixed inconsistencies in switch behaviors (e.g., `use_chart_parsing`) in the PP-StructureV3 configuration files compared to other pipelines.
+
+- **Other Enhancements:**
+    - **Separated core and optional dependencies. Only minimal core dependencies are required for basic text recognition; additional dependencies for document parsing and information extraction can be installed as needed.**
+    - **Enabled support for NVIDIA RTX 50 series graphics cards on Windows; users can refer to the [installation guide](../version3.x/installation.en.md) for the corresponding PaddlePaddle framework versions.**
+    - **PP-OCR series models now support returning single-character coordinates.**
+    - Added AIStudio, ModelScope, and other model download sources, allowing users to specify the source for model downloads.
+    - Added support for chart-to-table conversion via the PP-Chart2Table module.
+    - Optimized documentation descriptions to improve usability.
+
+#### **2025.08.15: Release of PaddleOCR 3.1.1**, includes
+
+- **Bug Fixes:**
+    - Added the missing methods `save_vector`, `save_visual_info_list`, `load_vector`, and `load_visual_info_list` in the `PP-ChatOCRv4` class.
+    - Added the missing parameters `glossary` and `llm_request_interval` to the `translate` method in the `PPDocTranslation` class.
+
+- **Documentation Improvements:**
+    - Added a demo to the MCP documentation.
+    - Added information about the PaddlePaddle and PaddleOCR version used for performance metrics testing in the documentation.
+    - Fixed errors and omissions in the production line document translation.
+
+- **Others:**
+    - Changed the MCP server dependency to use the pure Python library `puremagic` instead of `python-magic` to reduce installation issues.
+    - Retested PP-OCRv5 performance metrics with PaddleOCR version 3.1.0 and updated the documentation.
+
+#### **2025.06.29: Release of PaddleOCR 3.1.0**, includes
+
+- **Key Models and Pipelines:**
+    - **Added PP-OCRv5 Multilingual Text Recognition Model**, which supports the training and inference process for text recognition models in 37 languages, including French, Spanish, Portuguese, Russian, Korean, etc. **Average accuracy improved by over 30%.** [Details](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/algorithm/PP-OCRv5/PP-OCRv5_multi_languages.html)
+    - Upgraded the **PP-Chart2Table model** in PP-StructureV3, further enhancing the capability of converting charts to tables. On internal custom evaluation sets, the metric (RMS-F1) **increased by 9.36 percentage points (71.24% -> 80.60%).**
+    - Newly launched **document translation pipeline, PP-DocTranslation, based on PP-StructureV3 and ERNIE 4.5**, which supports the translation of Markdown format documents, various complex-layout PDF documents, and document images, with the results saved as Markdown format documents. [Details](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/pipeline_usage/PP-DocTranslation.html)
+
+- **New MCP server:** [Details](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/integrations/mcp_server.html)
+    - **Supports both OCR and PP-StructureV3 pipelines.**
+    - Supports three working modes: local Python library, AIStudio Community Cloud Service, and self-hosted service.
+    - Supports invoking local services via stdio and remote services via Streamable HTTP.
+
+- **Documentation Optimization:** Improved the descriptions in some user guides for a smoother reading experience.
+
+#### **2025.06.26: Release of PaddleOCR 3.0.3**, includes
+
+- Bug Fix: Resolved the issue where the `enable_mkldnn` parameter was not effective, restoring the default behavior of using MKL-DNN for CPU inference.
+
+#### **2025.06.19: Release of PaddleOCR v3.0.2, which includes:**
+
+- **New Features:**
+
+    - The default download source has been changed from `BOS` to `HuggingFace`. Users can also change the environment variable `PADDLE_PDX_MODEL_SOURCE` to `BOS` to set the model download source back to Baidu Object Storage (BOS).
+    - Added service invocation examples for six languages—C++, Java, Go, C#, Node.js, and PHP—for pipelines like PP-OCRv5, PP-StructureV3, and PP-ChatOCRv4.
+    - Improved the layout partition sorting algorithm in the PP-StructureV3 pipeline, enhancing the sorting logic for complex vertical layouts to deliver better results.
+    - Enhanced model selection logic: when a language is specified but a model version is not, the system will automatically select the latest model version supporting that language.
+    - Set a default upper limit for MKL-DNN cache size to prevent unlimited growth, while also allowing users to configure cache capacity.
+    - Updated default configurations for high-performance inference to support Paddle MKL-DNN acceleration and optimized the logic for automatic configuration selection for smarter choices.
+    - Adjusted the logic for obtaining the default device to consider the actual support for computing devices by the installed Paddle framework, making program behavior more intuitive.
+    - Added Android example for PP-OCRv5. [Details](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/cross_platform/on_device_deployment.html).
+
+- **Bug Fixes:**
+
+    - Fixed an issue with some CLI parameters in PP-StructureV3 not taking effect.
+    - Resolved an issue where `export_paddlex_config_to_yaml` would not function correctly in certain cases.
+    - Corrected the discrepancy between the actual behavior of `save_path` and its documentation description.
+    - Fixed potential multithreading errors when using MKL-DNN in basic service deployment.
+    - Corrected channel order errors in image preprocessing for the Latex-OCR model.
+    - Fixed channel order errors in saving visualized images within the text recognition module.
+    - Resolved channel order errors in visualized table results within PP-StructureV3 pipeline.
+    - Fixed an overflow issue in the calculation of `overlap_ratio` under extremely special circumstances in the PP-StructureV3 pipeline.
+
+- **Documentation Improvements:**
+
+    - Updated the description of the `enable_mkldnn` parameter in the documentation to accurately reflect the program's actual behavior.
+    - Fixed errors in the documentation regarding the `lang` and `ocr_version` parameters.
+    - Added instructions for exporting production line configuration files via CLI.
+    - Fixed missing columns in the performance data table for PP-OCRv5.
+    - Refined benchmark metrics for PP-StructureV3 pipeline across different configurations.
+
+- **Others:**
+
+    - Relaxed version restrictions on dependencies like numpy and pandas, restoring support for Python 3.12.
+
+#### **2025.06.05: Release of PaddleOCR v3.0.1, which includes:**
+
+- **Optimisation of certain models and model configurations:**
+    - Updated the default model configuration for PP-OCRv5, changing both detection and recognition from mobile to server models. To improve default performance in most scenarios, the parameter `limit_side_len` in the configuration has been changed from 736 to 64.
+    - Added a new text line orientation classification model `PP-LCNet_x1_0_textline_ori` with an accuracy of 99.42%. The default text line orientation classifier for OCR, PP-StructureV3, and PP-ChatOCRv4 pipelines has been updated to this model.
+    - Optimised the text line orientation classification model `PP-LCNet_x0_25_textline_ori`, improving accuracy by 3.3 percentage points to a current accuracy of 98.85%.
+
+- **Optimisation of issues present in version 3.0.0:**
+    - **Improved CLI usage experience:** When using the PaddleOCR CLI without passing any parameters, a usage prompt is now provided.
+    - **New parameters added:** PP-ChatOCRv3 and PP-StructureV3 now support the `use_textline_orientation` parameter.
+    - **CPU inference speed optimisation:** All pipeline CPU inferences now enable MKL-DNN by default.
+    - **Support for C++ inference:** The detection and recognition concatenation part of PP-OCRv5 now supports C++ inference.
+
+- **Fixes for issues present in version 3.0.0:**
+    - Fixed an issue where PP-StructureV3 encountered CPU inference errors due to the inability to use MKL-DNN with formula and table recognition models.
+    - Fixed an issue where GPU environments encountered the error `FatalError: Process abort signal is detected by the operating system` during inference.
+    - Fixed type hint issues in some Python 3.8 environments.
+    - Fixed the issue where the method `PPStructureV3.concatenate_markdown_pages` was missing.
+    - Fixed an issue where specifying both `lang` and `model_name` when instantiating `paddleocr.PaddleOCR` resulted in `model_name` being ineffective.
+
+#### **2025.05.20: PaddleOCR 3.0 Official Release Highlights**
+
+- **PP-OCRv5: All-Scene Text Recognition Model**
+    - Supports five text types and complex handwriting in a single model.
+    - Achieves a 13% accuracy improvement over the previous generation.
+
+- **PP-StructureV3: General Document Parsing Solution**
+    - Offers high-precision parsing for multi-scene, multi-layout PDFs.
+    - Outperforms numerous open and closed-source solutions in public benchmarks.
+
+- **PP-ChatOCRv4: Intelligent Document Understanding Solution**
+    - Natively supports ERNIE 4.5.
+    - Delivers a 15% accuracy boost over the previous version.
+
+- **Rebuilt Deployment Capabilities with Unified Inference Interface:**
+    - Integrates PaddleX3.0's core features for a comprehensive upgrade of the inference and deployment modules.
+    - Optimizes the design from version 2.x and unifies the Python API and CLI.
+    - Supports high-performance inference, serving, and on-device deployment scenarios.
+
+- **Optimized Training with PaddlePaddle Framework 3.0:**
+    - Compatible with the latest features such as the CINN compiler.
+    - Inference model files now use `xxx.json` instead of `xxx.pdmodel`.
+
+- **Unified Model Naming:**
+    - Updated naming conventions for models supported by PaddleOCR 3.0 for consistency and easier maintenance.
+
+- For more details, check out the [Upgrade Notes from 2.x to 3.x](./upgrade_notes.en.md).
+
+#### **2025.3.7 release PaddleOCR v2.10, including**
+
+- **12 new self-developed single models:**
+    - **[Layout Detection](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/layout_detection.html)** series with 3 models: PP-DocLayout-L, PP-DocLayout-M, PP-DocLayout-S, supporting prediction of 23 common layout categories. High-quality layout detection for various document types such as papers, reports, exams, books, magazines, contracts, newspapers in both English and Chinese. **mAP@0.5 reaches up to 90.4%, lightweight models can process over 100 pages of document images per second end-to-end.**
+    - **[Formula Recognition](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/formula_recognition.html)** series with 2 models: PP-FormulaNet-L, PP-FormulaNet-S, supporting 50,000 common LaTeX vocabulary, capable of recognizing complex printed and handwritten formulas. **PP-FormulaNet-L has 6 percentage points higher accuracy than models of the same level, and PP-FormulaNet-S is 16 times faster than models with similar accuracy.**
+    - **[Table Structure Recognition](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/table_structure_recognition.html)** series with 2 models: SLANeXt_wired, SLANeXt_wireless. A newly developed table structure recognition model, supporting structured prediction for both wired and wireless tables. Compared to SLANet_plus, SLANeXt shows significant improvement in table structure, **with 6 percentage points higher accuracy on internal high-difficulty table recognition evaluation sets.**
+    - **[Table Classification](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/table_classification.html)** series with 1 model: PP-LCNet_x1_0_table_cls, an ultra-lightweight classification model for both wired and wireless tables.
+    - **[Table Cell Detection](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/table_cells_detection.html)** series with 2 models: RT-DETR-L_wired_table_cell_det, RT-DETR-L_wireless_table_cell_det, supporting cell detection in both wired and wireless tables. These can be combined with SLANeXt_wired, SLANeXt_wireless, text detection, and text recognition modules for end-to-end table prediction. (See the newly added Table Recognition v2 pipeline)
+    - **[Text Recognition](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/text_recognition.html)** series with 1 model: PP-OCRv4_server_rec_doc, **supports over 15,000 characters, with a broader text recognition range, additionally improving the recognition accuracy of certain texts. The accuracy is more than 3 percentage points higher than PP-OCRv4_server_rec on internal datasets.**
+    - **[Text Line Orientation Classification](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/text_recognition.html)** series with 1 model: PP-LCNet_x0_25_textline_ori, **an ultra-lightweight text line orientation classification model with only 0.3M storage.**
+
+- **4 high-value multi-model combination solutions:**
+    - **[Document Image Preprocessing Pipeline](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/doc_preprocessor.html)**: Achieve correction of distortion and orientation in document images through the combination of ultra-lightweight models.
+    - **[Layout Parsing v2 Pipeline](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/layout_parsing.html)**: Combines multiple self-developed different types of OCR models to optimize complex layout reading order, achieving end-to-end conversion of various complex PDF files to Markdown and JSON files. The conversion effect is better than other open-source solutions in multiple document scenarios. It can provide high-quality data production capabilities for large model training and application.
+    - **[Table Recognition v2 Pipeline](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/table_recognition_v2.html)**: **Provides better table recognition capabilities.** By combining table classification module, table cell detection module, table structure recognition module, text detection module, text recognition module, etc., it achieves prediction of various styles of tables. Users can customize and finetune any module to improve the effect of vertical tables.
+    - **[PP-ChatOCRv4-doc Pipeline](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction_v4.html)**: Based on PP-ChatOCRv3-doc, **integrating multi-modal large models, optimizing Prompt and multi-model combination post-processing logic. It effectively addresses common complex document information extraction challenges such as layout analysis, rare characters, multi-page PDFs, tables, and seal recognition, achieving 15 percentage points higher accuracy than PP-ChatOCRv3-doc. The large model upgrades local deployment capabilities, providing a standard OpenAI interface, supporting calls to locally deployed large models like DeepSeek-R1.**
+
+#### **2024.10.18 release PaddleOCR v2.9, including**
+
+- PaddleX, an All-in-One development tool based on PaddleOCR's advanced technology, supports low-code full-process development capabilities in the OCR field:
+    - 🎨 [**Rich Model One-Click Call**](https://www.paddleocr.ai/v2.10.0/en/paddlex/quick_start.html): Integrates **17 models** related to text image intelligent analysis, general OCR, general layout parsing, table recognition, formula recognition, and seal recognition into 6 pipelines, which can be quickly experienced through a simple **Python API one-click call**. In addition, the same set of APIs also supports a total of **200+ models** in image classification, object detection, image segmentation, and time series forecasting, forming 20+ single-function modules, making it convenient for developers to use **model combinations**.
+
+    - 🚀 [**High Efficiency and Low barrier of entry**](https://www.paddleocr.ai/v2.10.0/en/paddlex/overview.html): Provides two methods based on **unified commands** and **GUI** to achieve simple and efficient use, combination, and customization of models. Supports multiple deployment methods such as **high-performance inference, service-oriented deployment, and on-device deployment**. Additionally, for various mainstream hardware such as **NVIDIA GPU, Kunlunxin XPU, Ascend NPU, Cambricon MLU, and Haiguang DCU**, models can be developed with **seamless switching**.
+
+- Supports [PP-ChatOCRv3-doc](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction_en.md), [high-precision layout detection model based on RT-DETR](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/layout_detection_en.md) and [high-efficiency layout area detection model based on PicoDet](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/layout_detection_en.md), [high-precision table structure recognition model](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/table_structure_recognition_en.md), text image unwarping model [UVDoc](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/text_image_unwarping_en.md), formula recognition model [LatexOCR](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/formula_recognition_en.md), and [document image orientation classification model based on PP-LCNet](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/doc_img_orientation_classification_en.md).
+
+#### 2022.5.9 release PaddleOCR v2.5, including
+
+- [PP-OCRv3](../version2.x/ppocr/overview.en.md#pp-ocrv3): With comparable speed, the effect of Chinese scene is further improved by 5% compared with PP-OCRv2, the effect of English scene is improved by 11%, and the average recognition accuracy of 80 language multilingual models is improved by more than 5%.
+- [PPOCRLabelv2](https://github.com/PFCCLab/PPOCRLabel/blob/main/README.md): Add the annotation function for table recognition task, key information extraction task and irregular text image.
+- Interactive e-book [*"Dive into OCR"*](../version2.x/ppocr/blog//ocr_book.en.md), covers the cutting-edge theory and code practice of OCR full stack technology.
+
+#### 2022.5.7 Add support for metric and model logging during training to [Weights & Biases](https://docs.wandb.ai/)
+
+#### 2021.12.21 OCR open source online course starts. The lesson starts at 8:30 every night and lasts for ten days. Free registration: <https://aistudio.baidu.com/aistudio/course/introduce/25207>
+
+#### 2021.12.21 release PaddleOCR v2.4, release 1 text detection algorithm (PSENet), 3 text recognition algorithms (NRTR、SEED、SAR), 1 key information extraction algorithm (SDMGR) and 3 DocVQA algorithms (LayoutLM、LayoutLMv2，LayoutXLM)
+
+#### 2021.9.7 release PaddleOCR v2.3, PP-OCRv2 is proposed. The CPU inference speed of PP-OCRv2 is 220% higher than that of PP-OCR server. The F-score of PP-OCRv2 is 7% higher than that of PP-OCR mobile
+
+#### 2021.8.3 released PaddleOCR v2.2, add a new structured documents analysis toolkit, i.e., [PP-Structure](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.2/ppstructure/README.md), support layout analysis and table recognition (One-key to export chart images to Excel files)
+
+#### 2021.4.8 release end-to-end text recognition algorithm [PGNet](https://cdn.aaai.org/ojs/16383/16383-13-19877-1-2-20210518.pdf) which is published in AAAI 2021. Find tutorial [here](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.1/doc/doc_en/pgnet_en.md)；release multi language recognition [models](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.1/doc/doc_en/multi_languages_en.md), support more than 80 languages recognition; especially, the performance of [English recognition model](../version3.x/model_list.md) is Optimized
+
+#### 2021.1.21 update more than 25+ multilingual recognition models [models list](../version2.x/ppocr/model_list.en.md), including：English, Chinese, German, French, Japanese，Spanish，Portuguese Russia Arabic and so on.  Models for more languages will continue to be updated [Develop Plan](https://github.com/PaddlePaddle/PaddleOCR/issues/1048)
+
+#### 2020.12.15 update Data synthesis tool, i.e., [Style-Text](https://github.com/PFCCLab/StyleText/blob/main/README.md)，easy to synthesize a large number of images which are similar to the target scene image
+
+#### 2020.11.25 Update a new data annotation tool, i.e., [PPOCRLabel](https://github.com/PFCCLab/PPOCRLabel/blob/main/README.md), which is helpful to improve the labeling efficiency. Moreover, the labeling results can be used in training of the PP-OCR system directly
+
+#### 2020.9.22 Update the PP-OCR technical article, <https://arxiv.org/abs/2009.09941>
+
+#### 2020.9.19 Update the ultra lightweight compressed ppocr_mobile_slim series models, the overall model size is 3.5M, suitable for mobile deployment
+
+#### 2020.9.17 update English recognition model and Multilingual recognition model, `English`, `Chinese`, `German`, `French`, `Japanese` and `Korean` have been supported. Models for more languages will continue to be updated
+
+#### 2020.8.24 Support the use of PaddleOCR through whl package installation，please refer  [PaddleOCR Package](https://github.com/PaddlePaddle/PaddleOCR/blob/develop/doc/doc_en/whl_en.md)
+
+#### 2020.8.16 Release text detection algorithm [SAST](https://arxiv.org/abs/1908.05498) and text recognition algorithm [SRN](https://arxiv.org/abs/2003.12294)
+
+#### 2020.7.23, Release the playback and PPT of live class on BiliBili station, PaddleOCR Introduction, [address](https://aistudio.baidu.com/aistudio/course/introduce/1519)
+
+#### 2020.7.15, Add mobile App demo , support both iOS and  Android  (based on easyedge and Paddle Lite)
+
+#### 2020.7.15, Improve the  deployment ability, add the C + +  inference , serving deployment. In addition, the benchmarks of the ultra-lightweight Chinese OCR model are provided
+
+#### 2020.7.15, Add several related datasets, data annotation and synthesis tools
+
+#### 2020.7.9 Add a new model to support recognize the  character "space"
+
+#### 2020.7.9 Add the data argument and learning rate decay strategies during training
+
+#### 2020.6.8 Add [datasets](../datasets/datasets.en.md) and keep updating
+
+#### 2020.6.5 Support exporting `attention` model to `inference_model`
+
+#### 2020.6.5 Support separate prediction and recognition, output result score
+
+#### 2020.5.30 Provide Lightweight Chinese OCR online experience
+
+#### 2020.5.30 Model prediction and training support on Windows system
+
+#### 2020.5.30 Open source general Chinese OCR model
+
+#### 2020.5.14 Release [PaddleOCR Open Class](https://www.bilibili.com/video/BV1nf4y1U7RX?p=4)
+
+#### 2020.5.14 Release [PaddleOCR Practice Notebook](https://aistudio.baidu.com/aistudio/projectdetail/467229)
+
+#### 2020.5.14 Open source 8.6M lightweight Chinese OCR model
diff --git a/docs/update/update.md b/docs/update/update.md
new file mode 100644
index 0000000..8d183d6
--- /dev/null
+++ b/docs/update/update.md
@@ -0,0 +1,271 @@
+---
+comments: true
+hide:
+  - navigation
+  - toc
+---
+
+### 更新
+
+#### 2025.08.21: **PaddleOCR 3.2.0** 发布，新增能力如下：
+
+- **重要模型新增：**
+    - 新增 PP-OCRv5 英文、泰文、希腊文识别模型的训练、推理、部署。**其中 PP-OCRv5 英文模型较 PP-OCRv5 主模型在英文场景提升 11%，泰文识别模型精度 82.68%，希腊文识别模型精度 89.28%。**
+
+- **部署能力升级：**
+    - **全面支持飞桨框架 3.1.0 和 3.1.1 版本。**
+    - **全面升级 PP-OCRv5 C++ 本地部署方案，支持 Linux、Windows，功能及精度效果与 Python 方案保持一致。**
+    - **高性能推理支持 CUDA 12，可使用 Paddle Inference、ONNX Runtime 后端推理。**
+    - **高稳定性服务化部署方案全面开源，支持用户根据需求对 Docker 镜像和 SDK 进行定制化修改。**
+    - 高稳定性服务化部署方案支持通过手动构造HTTP请求的方式调用，该方式允许客户端代码使用任意编程语言编写。
+
+- **Benchmark支持**：
+    - **全部产线支持产线细粒度 benchmark，能够测量产线端到端推理时间以及逐层、逐模块的耗时数据，可用于辅助产线性能分析。**
+    - **文档中补充各产线常用配置在主流硬件上的关键指标，包括推理耗时和内存占用等，为用户部署提供参考。**
+
+- **Bug修复：**
+    - 修复模型训练时训练日志保存失败的问题。
+    - 对公式模型的数据增强部分进行了版本兼容性升级，以适应新版本的 albumentations 依赖，并修复了在多进程使用 tokenizers 依赖包时出现的死锁警告。
+    - 修复 PP-StructureV3 配置文件中的 `use_chart_parsing` 等开关行为与其他产线不统一的问题。
+
+- **其他升级：**
+    - **分离必要依赖与可选依赖。使用基础文字识别功能时，仅需安装少量核心依赖；若需文档解析、信息抽取等功能，用户可按需选择安装额外依赖。**
+    - **支持 Windows 用户使用英伟达 50 系显卡，可根据 [安装文档](../version3.x/installation.md) 安装对应版本的 paddle 框架。**
+    - **PP-OCR 系列模型支持返回单文字坐标。**
+    - 模型新增 AIStudio、ModelScope 等下载源。可指定相关下载源下载对应的模型。
+    - 支持图表转表PP-Chart2Table单功能模块推理能力。
+    - 优化部分使用文档中的描述，提升易用性。
+
+#### 2025.08.15: **PaddleOCR 3.1.1** 发布，新增能力如下：
+
+- **bug修复：**
+  - 补充 `PP-ChatOCRv4` 类缺失的`save_vector`、`save_visual_info_list`、`load_vector、load_visual_info_list` 方法。
+  - 补充 `PPDocTranslation` 类的 `translate` 方法缺失的 `glossary 和 `llm_request_interval 参数。
+
+- **文档优化：**
+  - 补充 MCP 文档中的 demo。
+  - 补充文档中测试性能指标使用的飞桨框架与 PaddleOCR 版本。
+  - 修复文档翻译产线文档中的错漏。
+
+- **其他：**
+  - 修改 MCP 服务器依赖，使用纯 Python 库 `puremagic` 代替 `python-magic`，减少安装问题。
+  - 使用 3.1.0 版本 PaddleOCR 重新测试 PP-OCRv5 性能指标，更新文档。
+
+#### 2025.06.29: **PaddleOCR 3.1.0** 发布，新增能力如下：
+
+- **重要模型和产线：**
+  - **新增 PP-OCRv5 多语种文本识别模型**，支持法语、西班牙语、葡萄牙语、俄语、韩语等 37 种语言的文字识别模型的训推流程。**平均精度涨幅超30%。**[详情](https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/algorithm/PP-OCRv5/PP-OCRv5_multi_languages.html)
+  - 升级 PP-StructureV3 中的 **PP-Chart2Table 模型**，图表转表能力进一步升级，在内部自建测评集合上指标（RMS-F1）**提升 9.36 个百分点（71.24% -> 80.60%）。**
+  - 新增基于 PP-StructureV3 和 ERNIE 4.5 的**文档翻译产线 PP-DocTranslation，支持翻译 Markdown 格式文档、各种复杂版式的 PDF 文档和文档图像，结果保存为 Markdown 格式文档。**[详情](https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/pipeline_usage/PP-DocTranslation.html)
+
+- **新增MCP server：**[详情](https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/integrations/mcp_server.html)
+  - **支持 OCR 和 PP-StructureV3 两种工具；**
+  - 支持本地Python库、星河社区云服务、自托管服务三种工作模式；
+  - 支持通过 stdio 调用本地服务，通过 Streamable HTTP 调用远程服务。
+
+- **文档优化：** 优化了部分使用文档描述，提升阅读体验。
+
+#### 2025.06.26: **PaddleOCR 3.0.3** 发布，包含：
+- Bug修复：修复`enable_mkldnn`参数不生效的问题，恢复CPU默认使用MKL-DNN推理的行为。
+
+
+#### **2025.06.19: PaddleOCR v3.0.2 版本发布，包含：**
+
+- **功能新增：**
+    - 模型默认下载源从`BOS`改为`HuggingFace`，同时也支持用户通过更改环境变量`PADDLE_PDX_MODEL_SOURCE`为`BOS`，将模型下载源设置为百度云对象存储BOS。
+    - PP-OCRv5、PP-StructureV3、PP-ChatOCRv4等pipeline新增C++、Java、Go、C#、Node.js、PHP 6种语言的服务调用示例。
+    - 优化PP-StructureV3产线中版面分区排序算法，对复杂竖版版面排序逻辑进行完善，进一步提升了复杂版面排序效果。
+    - 优化模型选择逻辑，当指定语言、未指定模型版本时，自动选择支持该语言的最新版本的模型。 @timminator
+    -  为MKL-DNN缓存大小设置默认上界，防止缓存无限增长。同时，支持用户配置缓存容量。@timminator 
+    - 更新高性能推理默认配置，支持Paddle MKL-DNN加速。优化高性能推理自动配置逻辑，支持更智能的配置选择。
+    - 调整默认设备获取逻辑，考虑环境中安装的Paddle框架对计算设备的实际支持情况，使程序行为更符合直觉。
+    - 新增PP-OCRv5的Android端示例，[详情](https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/inference_deployment/cross_platform/on_device_deployment.html)。
+
+- **Bug修复：**
+    - 修复PP-StructureV3部分CLI参数不生效的问题。
+    - 修复部分情况下`export_paddlex_config_to_yaml`无法正常工作的问题。
+    - 修复save_path实际行为与文档描述不符的问题。
+    - 修复基础服务化部署在使用MKL-DNN时可能出现的多线程错误。
+    - 修复Latex-OCR模型的图像预处理的通道顺序错误。
+    - 修复文本识别模块保存可视化图像的通道顺序错误。
+    - 修复PP-StructureV3中表格可视化结果通道顺序错误。
+    - 修复PP-StructureV3产线中极特殊的情况下，计算overlap_ratio时，变量溢出问题。
+
+- **文档优化：**
+    - 更新文档中对`enable_mkldnn`参数的说明，使其更准确地描述程序的实际行为。
+    - 修复文档中对`lang`和`ocr_version`参数描述的错误。
+    - 补充通过CLI导出产线配置文件的说明。
+    - 修复PP-OCRv5性能数据表格中的列缺失问题。
+    - 润色PP-StructureV3在不同配置下的benchmark指标。
+
+- **其他：**
+    - 放松numpy、pandas等依赖的版本限制，恢复对Python 3.12的支持。
+
+#### **2025.06.05: PaddleOCR v3.0.1 版本发布，包含：**
+- **优化部分模型和模型配置：**
+    - 更新 PP-OCRv5默认模型配置，检测和识别均由mobile改为server模型。为了改善大多数的场景默认效果，配置中的参数`limit_side_len`由736改为64
+    - 新增文本行方向分类`PP-LCNet_x1_0_textline_ori`模型，精度99.42%，OCR、PP-StructureV3、PP-ChatOCRv4产线的默认文本行方向分类器改为该模型
+    - 优化文本行方向分类`PP-LCNet_x0_25_textline_ori`模型，精度提升3.3个百分点，当前精度98.85%
+- **优化3.0.0版本部分存在的问题**
+    - **优化CLI使用体验：** 当使用PaddleOCR CLI不传入任何参数时，给出用法提示。
+    - **新增参数：** PP-ChatOCRv3、PP-StructureV3支持`use_textline_orientation`参数。
+    - **CPU推理速度优化：** 所有产线CPU推理默认开启MKL-DNN。
+    - **C++推理支持：** PP-OCRv5的检测和识别串联部分支持C++推理
+- **修复3.0.0版本部分存在的问题**
+    - 修复由于公式识别、表格识别模型无法使用MKL-DNN导致PP-StructureV3在部分cpu推理报错的问题
+    - 修复在部分GPU环境中推理报`FatalError: Process abort signal is detected by the operating system`错误的问题
+    - 修复部分Python3.8环境的type hint的问题
+    - 修复`PPStructureV3.concatenate_markdown_pages`方法不存在的问题。
+    - 修复实例化`paddleocr.PaddleOCR`时同时指定`lang`和`model_name`时`model_name`不生效的问题。
+
+#### **2025.05.20: PaddleOCR 3.0 正式发布，包含：**
+
+- **发布全场景文字识别模型PP-OCRv5：** 单模型支持五种文字类型和复杂手写体识别；整体识别精度相比上一代提升13个百分点。
+ 
+- **发布通用文档解析方案PP-StructureV3：** 支持多场景、多版式 PDF 高精度解析，在公开评测集中领先众多开源和闭源方案。
+
+- **发布智能文档理解方案PP-ChatOCRv4：** 原生支持文心大模型4.5，精度相比上一代提升15个百分点。
+
+- **重构部署能力，统一推理接口：** PaddleOCR 3.0 融合了飞桨 PaddleX3.0 工具的底层能力，全面升级推理、部署模块，优化 2.x 版本的设计，统一并优化了 Python API 和命令行接口（CLI）。部署能力现覆盖高性能推理、服务化部署及端侧部署三大场景。
+
+- **适配飞桨框架 3.0，优化训练流程：** 新版本已兼容飞桨 3.0 的 CINN 编译器等最新特性，静态图模型存储文件名由 `xxx.pdmodel` 改为 `xxx.json`。
+
+- **统一模型名称：** 对PaddleOCR3.0支持的模型命名体系进行了更新，采用更规范、统一的命名规则，为后续迭代与维护奠定基础。
+
+- [2.x升级3.x其他说明](./upgrade_notes.md)。
+
+#### **2025.3.7 PaddleOCR 2.10 版本，主要包含如下内容**
+
+- **重磅新增 OCR 领域 12 个自研单模型：**
+    - **[版面区域检测](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/layout_detection.html)** 系列 3 个模型：PP-DocLayout-L、PP-DocLayout-M、PP-DocLayout-S，支持预测 23 个常见版面类别，中英论文、研报、试卷、书籍、杂志、合同、报纸等丰富类型的文档实现高质量版面检测，**mAP@0.5 最高达 90.4%，轻量模型端到端每秒处理超百页文档图像。**
+    - **[公式识别](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/formula_recognition.html)** 系列 2 个模型：PP-FormulaNet-L、PP-FormulaNet-S，支持 5 万种 LaTeX 常见词汇，支持识别高难度印刷公式和手写公式，其中 **PP-FormulaNet-L 较开源同等量级模型精度高 6 个百分点，PP-FormulaNet-S 较同等精度模型速度快 16 倍。**
+    - **[表格结构识别](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/table_structure_recognition.html)** 系列 2 个模型：SLANeXt_wired、SLANeXt_wireless。飞桨自研新一代表格结构识别模型，分别支持有线表格和无线表格的结构预测。相比于SLANet_plus，SLANeXt在表格结构方面有较大提升，**在内部高难度表格识别评测集上精度高 6 个百分点。**
+    - **[表格分类](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/table_classification.html)** 系列 1 个模型：PP-LCNet_x1_0_table_cls，超轻量级有线表格和无线表格的分类模型。
+    - **[表格单元格检测](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/table_cells_detection.html)** 系列 2 个模型：RT-DETR-L_wired_table_cell_det、RT-DETR-L_wireless_table_cell_det，分别支持有线表格和无线表格的单元格检测，可配合SLANeXt_wired、SLANeXt_wireless、文本检测、文本识别模块完成对表格的端到端预测。（参见本次新增的表格识别v2产线）
+    - **[文本识别](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/text_recognition.html)** 系列 1 个模型： PP-OCRv4_server_rec_doc，**支持1.5万+字典，文字识别范围更广，与此同时提升了部分文字的识别精准度，在内部数据集上，精度较 PP-OCRv4_server_rec 高 3 个百分点以上。**
+    - **[文本行方向分类](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/text_recognition.html)** 系列 1 个模型：PP-LCNet_x0_25_textline_ori，**存储只有 0.3M** 的超轻量级文本行方向分类模型。
+
+- **重磅推出 4 条高价值多模型组合方案：**
+    - **[文档图像预处理产线](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/doc_preprocessor.html)**：通过超轻量级模型组合使用，实现对文档图像的扭曲和方向的矫正。
+    - **[版面解析v2产线](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/layout_parsing.html)**：组合多个自研的不同类型的 OCR 类模型，优化复杂版面阅读顺序，实现多种复杂 PDF 文件端到端转换 Markdown 文件和 JSON 文件。在多个文档场景下，转换效果较其他开源方案更好。可以为大模型训练和应用提供高质量的数据生产能力。
+    - **[表格识别v2产线](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/table_recognition_v2.html)**：**提供更好的表格端到端识别能力。** 通过将表格分类模块、表格单元格检测模块、表格结构识别模块、文本检测模块、文本识别模块等组合使用，实现对多种样式的表格预测，用户可自定义微调其中任意模块以提升垂类表格的效果。
+    - **[PP-ChatOCRv4-doc产线](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction_v4.html)**：在 PP-ChatOCRv3-doc 的基础上，**融合了多模态大模型，优化了 Prompt 和多模型组合后处理逻辑，更好地解决了版面分析、生僻字、多页 pdf、表格、印章识别等常见的复杂文档信息抽取难点问题，准确率较 PP-ChatOCRv3-doc 高 15 个百分点。其中，大模型升级了本地部署的能力，提供了标准的 OpenAI 调用接口，支持对本地大模型如 DeepSeek-R1 部署的调用。**
+
+#### **2024.10.1 添加OCR领域低代码全流程开发能力**
+
+- 飞桨低代码开发工具PaddleX，依托于PaddleOCR的先进技术，支持了OCR领域的低代码全流程开发能力：
+    - 🎨 [**模型丰富一键调用**](https://www.paddleocr.ai/v2.10.0/paddlex/quick_start.html)：将文本图像智能分析、通用OCR、通用版面解析、通用表格识别、公式识别、印章文本识别涉及的**17个模型**整合为6条模型产线，通过极简的**Python API一键调用**，快速体验模型效果。此外，同一套API，也支持图像分类、目标检测、图像分割、时序预测等共计**200+模型**，形成20+单功能模块，方便开发者进行**模型组合**使用。
+    - 🚀[**提高效率降低门槛**](https://www.paddleocr.ai/v2.10.0/paddlex/overview.html)：提供基于**统一命令**和**图形界面**两种方式，实现模型简洁高效的使用、组合与定制。支持**高性能推理、服务化部署和端侧部署**等多种部署方式。此外，对于各种主流硬件如**英伟达GPU、昆仑芯、昇腾、寒武纪和海光**等，进行模型开发时，都可以**无缝切换**。
+
+- 支持文档场景信息抽取v3[PP-ChatOCRv3-doc](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction.md)、基于RT-DETR的[高精度版面区域检测模型](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/layout_detection.md)和PicoDet的[高效率版面区域检测模型](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/layout_detection.md)、高精度表格结构识别模型[SLANet_Plus](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/table_structure_recognition.md)、文本图像矫正模型[UVDoc](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/text_image_unwarping.md)、公式识别模型[LatexOCR](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/formula_recognition.md)、基于PP-LCNet的[文档图像方向分类模型](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.md)
+
+#### 2024.7 添加 PaddleOCR 算法模型挑战赛冠军方案
+
+- 赛题一：OCR 端到端识别任务冠军方案——[场景文本识别算法-SVTRv2](https://www.paddleocr.ai/v2.10.0/algorithm/text_recognition/algorithm_rec_svtrv2.html)；
+- 赛题二：通用表格识别任务冠军方案——[表格识别算法-SLANet-LCNetV2](https://www.paddleocr.ai/v2.10.0/algorithm/table_recognition/algorithm_table_slanet.html)。
+
+#### **2024.5.10 上线星河零代码产线(OCR 相关)**
+
+全面覆盖了以下四大 OCR 核心任务，提供极便捷的 Badcase 分析和实用的在线体验
+
+- [通用 OCR](https://aistudio.baidu.com/community/app/91660) (PP-OCRv4)。
+- [通用表格识别](https://aistudio.baidu.com/community/app/91661) (SLANet)。
+- [通用图像信息抽取](https://aistudio.baidu.com/community/app/91662) (PP-ChatOCRv2-common)。
+- [文档场景信息抽取](https://aistudio.baidu.com/community/app/70303) (PP-ChatOCRv2-doc)。
+
+  同时采用了 **[全新的场景任务开发范式](https://aistudio.baidu.com/pipeline/mine)** ,将模型统一汇聚，实现训练部署的零代码开发，并支持在线服务化部署和导出离线服务化部署包。
+
+#### 2023.8.7 发布 PaddleOCR [release/2.7](https://github.com/PaddlePaddle/PaddleOCR/tree/release/2.7)
+
+- 发布[PP-OCRv4](../version2.x/ppocr/blog/PP-OCRv4_introduction.md)，提供 mobile 和 server 两种模型
+    - PP-OCRv4-mobile：速度可比情况下，中文场景效果相比于 PP-OCRv3 再提升 4.5%，英文场景提升 10%，80 语种多语言模型平均识别准确率提升 8%以上
+    - PP-OCRv4-server：发布了目前精度最高的 OCR 模型，中英文场景上检测模型精度提升 4.9%， 识别模型精度提升 2%
+可参考[快速开始](../version2.x/ppocr/quick_start.md) 一行命令快速使用，同时也可在飞桨 AI 套件(PaddleX)中的[通用 OCR 产业方案](https://aistudio.baidu.com/aistudio/modelsdetail?modelId=286)中低代码完成模型训练、推理、高性能部署全流程
+
+#### 🔨**2022.11 新增实现[4 种前沿算法](../version2.x/algorithm/overview.md)**：文本检测 [DRRG](../version2.x/algorithm/text_detection/algorithm_det_drrg.md),  文本识别 [RFL](../version2.x/algorithm/text_recognition/algorithm_rec_rfl.md), 文本超分[Text Telescope](../version2.x/algorithm/super_resolution/algorithm_sr_telescope.md)，公式识别[CAN](../version2.x/algorithm/formula_recognition/algorithm_rec_can.md)
+
+#### **2022.10 优化[JS 版 PP-OCRv3 模型](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.10/deploy/paddlejs/README_ch.md)**：模型大小仅 4.3M，预测速度提升 8 倍，配套 web demo 开箱即用
+
+- **💥 直播回放：PaddleOCR 研发团队详解 PP-StructureV2 优化策略**。微信扫描下方二维码，关注公众号并填写问卷后进入官方交流群，获取直播回放链接与 20G 重磅 OCR 学习大礼包（内含 PDF 转 Word 应用程序、10 种垂类模型、《动手学 OCR》电子书等）
+
+#### **2022.8.24 发布 PaddleOCR [release/2.6](https://github.com/PaddlePaddle/PaddleOCR/tree/release/2.6)**
+
+- 发布[PP-StructureV2](../version2.x/ppstructure/overview.md)，系统功能性能全面升级，适配中文场景，新增支持[版面复原](../../ppstructure/recovery/README_ch.md)，支持**一行命令完成 PDF 转 Word**；
+- [版面分析](../../ppstructure/layout/README_ch.md)模型优化：模型存储减少 95%，速度提升 11 倍，平均 CPU 耗时仅需 41ms；
+- [表格识别](../../ppstructure/table/README_ch.md)模型优化：设计 3 大优化策略，预测耗时不变情况下，模型精度提升 6%；
+- [关键信息抽取](../../ppstructure/kie/README_ch.md)模型优化：设计视觉无关模型结构，语义实体识别精度提升 2.8%，关系抽取精度提升 9.1%。
+
+#### **2022.8 发布 [OCR 场景应用集合](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.8/applications/README_en.md)**：包含数码管、液晶屏、车牌、高精度 SVTR 模型、手写体识别等**9 个垂类模型**，覆盖通用，制造、金融、交通行业的主要 OCR 垂类应用
+
+#### 2022.5.9 发布PaddleOCR v2.5。发布内容包括
+
+- [PP-OCRv3](../version2.x/ppocr/overview.md#pp-ocrv3)，速度可比情况下，中文场景效果相比于PP-OCRv2再提升5%，英文场景提升11%，80语种多语言模型平均识别准确率提升5%以上；
+- 半自动标注工具[PPOCRLabelv2](https://github.com/PFCCLab/PPOCRLabel)：新增表格文字图像、图像关键信息抽取任务和不规则文字图像的标注功能；
+- OCR产业落地工具集：打通22种训练部署软硬件环境与方式，覆盖企业90%的训练部署环境需求
+- 交互式OCR开源电子书[《动手学OCR》](../version2.x/ppocr/blog/ocr_book.md)，覆盖OCR全栈技术的前沿理论与代码实践，并配套教学视频。
+
+#### 2022.5.7 添加对[Weights & Biases](https://docs.wandb.ai/)训练日志记录工具的支持
+
+#### 2021.12.21 《OCR十讲》课程开讲，12月21日起每晚八点半线上授课！ 【免费】报名地址：<https://aistudio.baidu.com/aistudio/course/introduce/25207>
+
+#### 2021.12.21 发布PaddleOCR v2.4。OCR算法新增1种文本检测算法（PSENet），3种文本识别算法（NRTR、SEED、SAR）；文档结构化算法新增1种关键信息提取算法（SDMGR），3种DocVQA算法（LayoutLM、LayoutLMv2，LayoutXLM）
+
+#### 2021.9.7 发布PaddleOCR v2.3，发布PP-OCRv2，CPU推理速度相比于PP-OCR server提升220%；效果相比于PP-OCR mobile 提升7%
+
+#### 2021.8.3 发布PaddleOCR v2.2，新增文档结构分析[PP-Structure](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.2/ppstructure/README_ch.md)工具包，支持版面分析与表格识别（含Excel导出）
+
+#### 2021.6.29 [FAQ](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.2/doc/doc_ch/FAQ.md)新增5个高频问题，总数248个，每周一都会更新，欢迎大家持续关注
+
+#### 2021.4.8 release 2.1版本，新增AAAI 2021论文[端到端识别算法PGNet](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.2/doc/doc_ch/pgnet.md)开源，[多语言模型](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.2/doc/doc_ch/multi_languages.md)支持种类增加到80+
+
+#### 2020.12.15 更新数据合成工具[Style-Text](https://github.com/PFCCLab/StyleText/blob/main/README_ch.md)，可以批量合成大量与目标场景类似的图像，在多个场景验证，效果明显提升
+
+#### 2020.12.07 [FAQ](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.1/doc/doc_ch/FAQ.md)新增5个高频问题，总数124个，并且计划以后每周一都会更新，欢迎大家持续关注
+
+#### 2020.11.25 更新半自动标注工具[PPOCRLabel](https://github.com/PFCCLab/PPOCRLabel/blob/main/README_ch.md)，辅助开发者高效完成标注任务，输出格式与PP-OCR训练任务完美衔接
+
+#### 2020.9.22 更新PP-OCR技术文章，<https://arxiv.org/abs/2009.09941>
+
+#### 2020.9.19 更新超轻量压缩ppocr_mobile_slim系列模型，整体模型3.5M(详见PP-OCR Pipeline)，适合在移动端部署使用
+
+#### 2020.9.17 更新超轻量ppocr_mobile系列和通用ppocr_server系列中英文ocr模型，媲美商业效果
+
+#### 2020.9.17 更新[英文识别模型](../version2.x/ppocr/model_list.md#22)和[多语种识别模型](../version2.x/ppocr/model_list.md#22)，已支持`德语、法语、日语、韩语`，更多语种识别模型将持续更新
+
+#### 2020.8.26 更新OCR相关的84个常见问题及解答，具体参考[FAQ](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.1/doc/doc_ch/FAQ.md)
+
+#### 2020.8.24 支持通过whl包安装使用PaddleOCR，具体参考[Paddleocr Package使用说明](https://github.com/PaddlePaddle/PaddleOCR/blob/develop/doc/doc_ch/whl.md)
+
+#### 2020.8.21 更新8月18日B站直播课回放和PPT，课节2，易学易用的OCR工具大礼包，[获取地址](https://aistudio.baidu.com/aistudio/education/group/info/1519)
+
+#### 2020.8.16 开源文本检测算法[SAST](https://arxiv.org/abs/1908.05498)和文本识别算法[SRN](https://arxiv.org/abs/2003.12294)
+
+#### 2020.7.23 发布7月21日B站直播课回放和PPT，课节1，PaddleOCR开源大礼包全面解读，[获取地址](https://aistudio.baidu.com/aistudio/course/introduce/1519)
+
+#### 2020.7.15 添加基于EasyEdge和Paddle-Lite的移动端DEMO，支持iOS和Android系统
+
+#### 2020.7.15 完善预测部署，添加基于C++预测引擎推理、服务化部署和端侧部署方案，以及超轻量级中文OCR模型预测耗时Benchmark
+
+#### 2020.7.15 整理OCR相关数据集、常用数据标注以及合成工具
+
+#### 2020.7.9 添加支持空格的识别模型，识别效果，预测及训练方式请参考快速开始和文本识别训练相关文档
+
+#### 2020.7.9 添加数据增强、学习率衰减策略,具体参考[配置文件](../version2.x/ppocr/blog/config.md)
+
+#### 2020.6.8 添加[数据集](../datasets/datasets.md)，并保持持续更新
+
+#### 2020.6.5 支持 `attetnion` 模型导出 `inference_model`
+
+#### 2020.6.5 支持单独预测识别时，输出结果得分
+
+#### 2020.5.30 提供超轻量级中文OCR在线体验
+
+#### 2020.5.30 模型预测、训练支持Windows系统
+
+#### 2020.5.30 开源通用中文OCR模型
+
+#### 2020.5.14 发布[PaddleOCR公开课](https://www.bilibili.com/video/BV1nf4y1U7RX?p=4)
+
+#### 2020.5.14 发布[PaddleOCR实战练习](https://aistudio.baidu.com/aistudio/projectdetail/467229)
+
+#### 2020.5.14 开源8.6M超轻量级中文OCR模型
diff --git a/docs/update/upgrade_notes.en.md b/docs/update/upgrade_notes.en.md
new file mode 100644
index 0000000..c9d9c85
--- /dev/null
+++ b/docs/update/upgrade_notes.en.md
@@ -0,0 +1,83 @@
+# PaddleOCR 3.x Upgrade Notes
+
+## 1. Why Upgrade from PaddleOCR 2.x to 3.x?
+
+Since the release of PaddleOCR 2.0 in February 2021, the community has experienced over four years of rapid growth. The number of GitHub stars, community users and contributors, as well as issues and PRs, have all increased exponentially. With emerging needs such as multilingual recognition and layout analysis, PaddleOCR continued to expand its capabilities in the 2.x series. However, the original lightweight-centric architecture has struggled to accommodate the growing complexity and rising maintenance costs brought by the feature boom.
+
+As more module branches and "bridging" layers were added to the codebase, issues such as code duplication and inconsistent interfaces became increasingly prominent. Testing became more difficult, and development efficiency was severely constrained. In addition, legacy dependencies became incompatible with newer versions of PaddlePaddle, limiting access to its latest features and slowing down training and inference. Under such circumstances, continuing to patch the existing architecture would only increase technical debt and system fragility.
+
+Meanwhile, Transformer-based vision-language models are injecting new momentum into advanced scenarios such as document understanding, image-text summarization, and intelligent proofreading. The community is eager to go beyond traditional OCR recognition and fully harness the powerful contextual understanding and reasoning capabilities of these models. At the same time, lightweight OCR models can still work in tandem with large models—both supporting the input needs of large models in document parsing and achieving complementary strengths to further enhance overall system performance.
+
+Moreover, the official release of PaddlePaddle 3.0 in April 2025 brought groundbreaking upgrades in unified training/inference and domestic hardware adaptation. This calls for a significant update to PaddleOCR in both its training and inference components.
+
+Given this background, we’ve decided to implement a major, non-backward-compatible upgrade—transitioning from 2.x to 3.x. The new version introduces a modular and plugin-based architecture. While retaining familiar usage patterns for users as much as possible, it integrates large model capabilities, offers richer features, and leverages the latest advancements of PaddlePaddle 3.0. The result is reduced maintenance cost, improved performance, and a solid foundation for future feature expansion.
+
+## 2. Key Upgrades from PaddleOCR 2.x to 3.x
+
+The 3.x upgrade consists of three major enhancements:
+
+1. **New Model Pipelines**: Introduced several new pipelines such as PP-OCRv5, PP-StructureV3, and PP-ChatOCR v4, covering a wide range of base models. These significantly enhance recognition capabilities for various text types, including handwriting, to meet the growing demand for high-precision parsing in complex documents. All models are ready-to-use out of the box, improving development efficiency.
+2. **Refactored Deployment and Unified Inference Interface**: The deployment module in PaddleOCR 3.x is rebuilt using [PaddleX](../version3.x/paddleocr_and_paddlex.en.md)’s underlying capabilities, fixing design flaws from 2.x and unifying both Python APIs and CLI interfaces. The deployment now supports three main scenarios: high-performance inference, service-oriented deployment, and on-device deployment.
+3. **PaddlePaddle 3.0 Compatibility and Optimized Training**: The new version is fully compatible with PaddlePaddle 3.0, including features like the CINN compiler. It also introduces a standardized model naming system to streamline future updates and maintenance.
+
+Some legacy features from PaddleOCR 2.x remain partially supported in 3.x. For more information, refer to [Legacy Features](../version2.x/legacy/index.en.md).
+
+## 3. Migrating Inference Code from PaddleOCR 2.x to 3.x
+
+For OCR tasks, PaddleOCR 3.x still supports a usage pattern similar to 2.x. Here’s an example using the Python API in 2.x:
+
+```python
+from paddleocr import PaddleOCR
+
+ocr = PaddleOCR(lang="en")
+result = ocr.ocr("img.png")
+for res in result:
+    for line in res:
+        print(line)
+
+# Visualization
+from PIL import Image
+from paddleocr import draw_ocr
+result = result[0]
+image = Image.open(img_path).convert("RGB")
+boxes = [line[0] for line in result]
+txts = [line[1][0] for line in result]
+scores = [line[1][1] for line in result]
+im_show = draw_ocr(image, boxes, txts, scores, font_path="simfang.ttf")
+im_show = Image.fromarray(im_show)
+im_show.save("result.jpg")
+```
+
+In PaddleOCR 3.x, this workflow is further simplified:
+
+```python
+from paddleocr import PaddleOCR
+
+# Basic initialization parameters remain the same
+ocr = PaddleOCR(lang="en")
+result = ocr.ocr("img.png")
+# Or use the new unified interface
+# result = ocr.predict("img.png")
+for res in result:
+    # Directly print recognition results, no nested loops required
+    res.print()
+
+# Visualization and saving results are simpler
+res.save_to_img("result")
+```
+
+In PaddleOCR 3.x, the `PaddleOCR.ocr` method no longer accepts parameters such as `det` and `rec`. Instead, users can perform module inference using clearer interfaces like `TextDetection` and `TextRecognition`. Additionally, in PaddleOCR 2.x, users could control log output by passing the `show_log` parameter when constructing a `PaddleOCR` object. However, this design had a limitation: since all `PaddleOCR` instances shared a single logger, configuring logging behavior in one instance would affect others—clearly not the desired behavior. To address this issue, PaddleOCR 3.x introduces an entirely new logging system. For details, please refer to the [Logging](../version3.x/logging.en.md) documentation.
+
+In PaddleOCR 3.x, the original `use_onnx` parameter has been replaced by a more powerful high-performance inference feature. For details, refer to [High-Performance Inference](../version3.x/inference_deployment/local_inference/high_performance_inference.en.md).
+
+The `PPStructure` provided in PaddleOCR 2.x has been removed in PaddleOCR 3.x. It is recommended to use the more feature-rich and better-performing `PPStructureV3` instead. Please consult the relevant documentation to learn how to use the new interface.
+
+## 4. Known Issues in PaddleOCR 3.0
+
+PaddleOCR 3.0 is still under active development. Current known limitations include:
+
+1. Incomplete support for native C++ deployment.
+2. High-performance service-oriented deployment is not yet on par with PaddleServing in 2.x.
+3. On-Device Deployment currently supports only a subset of key models, with broader support pending.
+
+If you encounter any issues during use, feel free to submit feedback via GitHub issues. We also warmly welcome more community members to contribute to PaddleOCR's future. Thank you for your continued support and interest!
diff --git a/docs/update/upgrade_notes.md b/docs/update/upgrade_notes.md
new file mode 100644
index 0000000..10a4d96
--- /dev/null
+++ b/docs/update/upgrade_notes.md
@@ -0,0 +1,83 @@
+# PaddleOCR 3.x 升级说明
+
+## 1. PaddleOCR 为什么要从 2.x 升级到 3.x？
+
+自 2021 年 2 月发布 2.0 版本以来，PaddleOCR 社区已走过四年多的快速发展期，GitHub star 数量、社区用户和贡献者、issue 与 PR 数量等均有指数级的增长。在多语种识别、版面分析等新需求的推动下，PaddleOCR 在 2.x 系列中不断增添功能，但最初以轻量化为核心的架构已难以应对功能繁荣带来的复杂性与维护成本。
+
+随着代码中模块分支与“桥接”层频繁增加，重复实现、接口不统一的问题愈发突出，测试难度也不断增加，开发效率严重受限；而旧版依赖与最新 PaddlePaddle 更新的不兼容，限制了对飞桨新特性的使用，进一步拖慢了训练与推理速度。这种状况下，继续在现有架构基础上打补丁，只会带来更多技术债与系统脆弱性。
+
+另一方面，基于 Transformer 的视觉语言大模型正在为文档理解、图文摘要、智能校对等高级应用场景注入新动能。社区迫切期待，这类模型能够突破传统 OCR 识别的局限，直接发挥其更强的上下文理解与推理能力。同时，传统的 OCR 小模型亦可与大模型协同工作，既能满足大模型在文档解析等方面的输入需求，又能通过大小模型协同，实现优势互补，进一步提升系统整体性能。
+
+此外，飞桨框架于 2025 年 4 月发布的 3.0 正式版，在训推一体化、国产硬件适配等方面实现了颠覆性升级，这也对 PaddleOCR 在训练和推理层面提出了新的改造需求。
+
+综合以上背景，我们决定对 PaddleOCR 进行一次重大、非兼容性升级——从 2.x 跳至 3.x。新版本将在架构层面实现模块化、插件化设计，在尽可能不改变用户使用习惯的同时，结合大模型，提供更加丰富的功能，并充分利用飞桨 3.0 的新特性，既清理冗余、降低维护成本，又为性能与功能扩展提供更坚实的基础。
+
+## 2. PaddleOCR 2.x 到 3.x 主要升级内容
+
+本次升级内容主要可分为三个部分：
+
+1. **新增多条模型产线**：推出 PP-OCRv5、PP-StructureV3、PP-ChatOCR v4 等多条模型产线，并补充覆盖多种方向的基础模型，重点增强了多文字类型识别、手写体识别等能力，满足大模型应用对复杂文档高精度解析的旺盛需求。用户可直接开箱使用，提升开发效率。
+2. **重构部署能力，统一推理接口**：PaddleOCR 3.x 融合了飞桨 [PaddleX](../version3.x/paddleocr_and_paddlex.md) 工具的底层能力，全面升级推理、部署模块，修正 2.x 版本中的设计错误，统一并优化了 Python API 和命令行接口（CLI）。部署能力现覆盖高性能推理、服务化部署及端侧部署三大场景。
+3. **适配飞桨 3.0，优化训练流程**：新版本已兼容飞桨 3.0 的 CINN 编译器等最新特性，并对模型命名体系进行了更新，采用更规范、统一的命名规则，为后续迭代与维护奠定基础。
+
+对于 PaddleOCR 2.x 中的部分历史遗留功能，PaddleOCR 3.x 目前仍提供了一定程度的兼容支持。详情请参阅 [历史遗留功能](../version2.x/legacy/index.md)。
+
+## 3. 将 PaddleOCR 2.x 的推理代码移到 PaddleOCR 3.x
+
+对于 OCR 任务，PaddleOCR 3.x 仍然支持与 PaddleOCR 2.x 类似的用法。以 Python API 为例，以下是 PaddleOCR 2.x 的常见使用方式：
+
+```python
+from paddleocr import PaddleOCR
+
+ocr = PaddleOCR(lang="en")
+result = ocr.ocr("img.png")
+for res in result:
+    for line in res:
+        print(line)
+        
+# 可视化
+from PIL import Image
+from paddleocr import draw_ocr
+result = result[0]
+image = Image.open(img_path).convert("RGB")
+boxes = [line[0] for line in result]
+txts = [line[1][0] for line in result]
+scores = [line[1][1] for line in result]
+im_show = draw_ocr(image, boxes, txts, scores, font_path="simfang.ttf")
+im_show = Image.fromarray(im_show)
+im_show.save("result.jpg")
+```
+
+在 PaddleOCR 3.x 中，以上流程得到了进一步简化，示例如下：
+
+```python
+from paddleocr import PaddleOCR
+
+# 基础的初始化参数保持一致
+ocr = PaddleOCR(lang="en")
+result = ocr.ocr("img.png")
+# 也可以使用新的统一接口
+# result = ocr.predict("img.png")
+for res in result:
+    # 可直接调用方法打印识别结果，无需嵌套循环
+    res.print()
+
+# 可视化及结果保存更为简洁
+res.save_to_img("result")
+```
+
+在 PaddleOCR 3.x 中，`PaddleOCR.ocr` 方法不再接受 `det`、`rec` 等参数。作为替代，用户可以通过使用更加清晰的 `TextDetection`、`TextRecognition` 等接口实现单功能模块的推理。此外，在 PaddleOCR 2.x 中，可以通过在构造 `PaddleOCR` 对象时传入 `show_log` 参数来控制日志输出。然而，这种设计存在局限：由于所有 `PaddleOCR` 实例共享一个日志器，当一个实例设置了日志行为后，其它实例也会受到影响，这显然不符合预期。为了解决这一问题，PaddleOCR 3.x 引入了全新的日志系统。详细内容请参阅 [日志](../version3.x/logging.md)。
+
+在 PaddleOCR 3.x 中，原有的 `use_onnx` 参数被更加强大的高性能推理功能代替，详情可参考 [高性能推理](../version3.x/inference_deployment/local_inference/high_performance_inference.md)。
+
+PaddleOCR 2.x 提供的 `PPStructure` 在 PaddleOCR 3.x 中已被移除。建议使用功能更丰富、解析效果更好的 `PPStructureV3` 替代，并参考相关文档了解新接口的用法。
+
+## 4. PaddleOCR 3.0 已知问题
+
+PaddleOCR 3.x 仍在持续迭代与优化中，目前已知存在以下尚待完善之处：
+
+1. 对 C++ 本地部署的支持尚不完整。
+2. 暂未提供性能与 PaddleOCR 2.x 中 PaddleServing 部署方案对齐的高性能服务化部署方案。
+3. 端侧部署目前仅支持部分重点模型，其余模型尚未开放支持。
+
+如果你在使用过程中遇到问题，欢迎随时在 issue 区提交反馈。我们也诚挚邀请更多社区用户参与到 PaddleOCR 的建设中来，感谢大家一直以来的关注与支持！
diff --git a/docs/version2.x/README.en.md b/docs/version2.x/README.en.md
new file mode 100644
index 0000000..3c062c7
--- /dev/null
+++ b/docs/version2.x/README.en.md
@@ -0,0 +1,3 @@
+# Features of the 2.x Branch
+
+Due to the upgrade of the 3.x branch, the wheel package has undergone refactoring, resulting in some models and features no longer being compatible with the older branch. To ensure that users relying on the 2.x branch's features can continue using them, we have retained the code related to the 2.x branch in this directory.
diff --git a/docs/version2.x/README.md b/docs/version2.x/README.md
new file mode 100644
index 0000000..32d261a
--- /dev/null
+++ b/docs/version2.x/README.md
@@ -0,0 +1,3 @@
+# 关于 2.x 分支的功能
+
+由于 3.x 分支的升级，wheel 包经历了重构，导致部分模型和功能与旧版分支不再兼容。为了使得依赖于 2.x 分支功能的用户能够继续使用这些功能，我们在此目录中保留了与 2.x 分支相关的代码。
diff --git a/docs/version2.x/algorithm/add_new_algorithm.en.md b/docs/version2.x/algorithm/add_new_algorithm.en.md
new file mode 100644
index 0000000..e52790a
--- /dev/null
+++ b/docs/version2.x/algorithm/add_new_algorithm.en.md
@@ -0,0 +1,307 @@
+---
+comments: true
+---
+
+# Add New Algorithm
+
+PaddleOCR decomposes an algorithm into the following parts, and modularizes each part to make it more convenient to develop new algorithms.
+
+* Data loading and processing
+* Network
+* Post-processing
+* Loss
+* Metric
+* Optimizer
+
+The following will introduce each part separately, and introduce how to add the modules required for the new algorithm.
+
+## Data loading and processing
+
+Data loading and processing are composed of different modules, which complete the image reading, data augment and label production. This part is under [ppocr/data](../../../ppocr/data). The explanation of each file and folder are as follows:
+
+```bash linenums="1"
+ppocr/data/
+├── imaug             # Scripts for image reading, data augment and label production
+│   ├── label_ops.py  # Modules that transform the label
+│   ├── operators.py  # Modules that transform the image
+│   ├──.....
+├── __init__.py
+├── lmdb_dataset.py   # The dataset that reads the lmdb
+└── simple_dataset.py # Read the dataset saved in the form of `image_path\tgt`
+```
+
+PaddleOCR has a large number of built-in image operation related modules. For modules that are not built-in, you can add them through the following steps:
+
+1. Create a new file under the [ppocr/data/imaug](../../../ppocr/data/imaug) folder, such as my_module.py.
+2. Add code in the my_module.py file, the sample code is as follows:
+
+    ```python linenums="1"
+    class MyModule:
+        def __init__(self, *args, **kwargs):
+            # your init code
+            pass
+
+        def __call__(self, data):
+            img = data['image']
+            label = data['label']
+            # your process code
+
+            data['image'] = img
+            data['label'] = label
+            return data
+    ```
+
+3. Import the added module in the [ppocr/data/imaug/\__init\__.py](../../../ppocr/data/imaug/__init__.py) file.
+
+All different modules of data processing are executed by sequence, combined and executed in the form of a list in the config file. Such as:
+
+```yaml linenums="1"
+# angle class data process
+transforms:
+  - DecodeImage: # load image
+      img_mode: BGR
+      channel_first: False
+  - MyModule:
+      args1: args1
+      args2: args2
+  - KeepKeys:
+      keep_keys: [ 'image', 'label' ] # dataloader will return list in this order
+```
+
+## Network
+
+The network part completes the construction of the network, and PaddleOCR divides the network into four parts, which are under [ppocr/modeling](../../../ppocr/modeling). The data entering the network will pass through these four parts in sequence(transforms->backbones->
+necks->heads).
+
+```bash linenums="1"
+├── architectures # Code for building network
+├── transforms    # Image Transformation Module
+├── backbones     # Feature extraction module
+├── necks         # Feature enhancement module
+└── heads         # Output module
+```
+
+PaddleOCR has built-in commonly used modules related to algorithms such as DB, EAST, SAST, CRNN and Attention. For modules that do not have built-in, you can add them through the following steps, the four parts are added in the same steps, take backbones as an example:
+
+1. Create a new file under the [ppocr/modeling/backbones](../../../ppocr/modeling/backbones) folder, such as my_backbone.py.
+2. Add code in the my_backbone.py file, the sample code is as follows:
+
+    ```python linenums="1"
+    import paddle
+    import paddle.nn as nn
+    import paddle.nn.functional as F
+
+
+    class MyBackbone(nn.Layer):
+        def __init__(self, *args, **kwargs):
+            super(MyBackbone, self).__init__()
+            # your init code
+            self.conv = nn.xxxx
+
+        def forward(self, inputs):
+            # your network forward
+            y = self.conv(inputs)
+            return y
+    ```
+
+3. Import the added module in the [ppocr/modeling/backbones/\__init\__.py](../../../ppocr/modeling/backbones/__init__.py) file.
+
+After adding the four-part modules of the network, you only need to configure them in the configuration file to use, such as:
+
+```yaml linenums="1"
+Architecture:
+model_type: rec
+algorithm: CRNN
+Transform:
+    name: MyTransform
+    args1: args1
+    args2: args2
+Backbone:
+    name: MyBackbone
+    args1: args1
+Neck:
+    name: MyNeck
+    args1: args1
+Head:
+    name: MyHead
+    args1: args1
+```
+
+## Post-processing
+
+Post-processing realizes decoding network output to obtain text box or recognized text. This part is under [ppocr/postprocess](../../../ppocr/postprocess).
+PaddleOCR has built-in post-processing modules related to algorithms such as DB, EAST, SAST, CRNN and Attention. For components that are not built-in, they can be added through the following steps:
+
+1. Create a new file under the [ppocr/postprocess](../../../ppocr/postprocess) folder, such as my_postprocess.py.
+2. Add code in the my_postprocess.py file, the sample code is as follows:
+
+    ```python linenums="1"
+    import paddle
+
+
+    class MyPostProcess:
+        def __init__(self, *args, **kwargs):
+            # your init code
+            pass
+
+        def __call__(self, preds, label=None, *args, **kwargs):
+            if isinstance(preds, paddle.Tensor):
+                preds = preds.numpy()
+            # you preds decode code
+            preds = self.decode_preds(preds)
+            if label is None:
+                return preds
+            # you label decode code
+            label = self.decode_label(label)
+            return preds, label
+
+        def decode_preds(self, preds):
+            # you preds decode code
+            pass
+
+        def decode_label(self, preds):
+            # you label decode code
+            pass
+    ```
+
+3. Import the added module in the [ppocr/postprocess/\__init\__.py](../../../ppocr/postprocess/__init__.py) file.
+
+After the post-processing module is added, you only need to configure it in the configuration file to use, such as:
+
+```yaml linenums="1"
+PostProcess:
+name: MyPostProcess
+args1: args1
+args2: args2
+```
+
+## Loss
+
+The loss function is used to calculate the distance between the network output and the label. This part is under [ppocr/losses](../../../ppocr/losses).
+PaddleOCR has built-in loss function modules related to algorithms such as DB, EAST, SAST, CRNN and Attention. For modules that do not have built-in modules, you can add them through the following steps:
+
+1. Create a new file in the [ppocr/losses](../../../ppocr/losses) folder, such as my_loss.py.
+2. Add code in the my_loss.py file, the sample code is as follows:
+
+    ```python linenums="1"
+    import paddle
+    from paddle import nn
+
+
+    class MyLoss(nn.Layer):
+        def __init__(self, **kwargs):
+            super(MyLoss, self).__init__()
+            # you init code
+            pass
+
+        def __call__(self, predicts, batch):
+            label = batch[1]
+            # your loss code
+            loss = self.loss(input=predicts, label=label)
+            return {'loss': loss}
+    ```
+
+3. Import the added module in the [ppocr/losses/\__init\__.py](../../../ppocr/losses/__init__.py) file.
+
+After the loss function module is added, you only need to configure it in the configuration file to use it, such as:
+
+```yaml linenums="1"
+Loss:
+  name: MyLoss
+  args1: args1
+  args2: args2
+```
+
+## Metric
+
+Metric is used to calculate the performance of the network on the current batch. This part is under [ppocr/metrics](../../../ppocr/metrics). PaddleOCR has built-in evaluation modules related to algorithms such as detection, classification and recognition. For modules that do not have built-in modules, you can add them through the following steps:
+
+1. Create a new file under the [ppocr/metrics](../../../ppocr/metrics) folder, such as my_metric.py.
+2. Add code in the my_metric.py file, the sample code is as follows:
+
+    ```python linenums="1"
+
+    class MyMetric(object):
+        def __init__(self, main_indicator='acc', **kwargs):
+            # main_indicator is used for select best model
+            self.main_indicator = main_indicator
+            self.reset()
+
+        def __call__(self, preds, batch, *args, **kwargs):
+            # preds is out of postprocess
+            # batch is out of dataloader
+            labels = batch[1]
+            cur_correct_num = 0
+            cur_all_num = 0
+            # you metric code
+            self.correct_num += cur_correct_num
+            self.all_num += cur_all_num
+            return {'acc': cur_correct_num / cur_all_num, }
+
+        def get_metric(self):
+            """
+            return metrics {
+                    'acc': 0,
+                    'norm_edit_dis': 0,
+                }
+            """
+            acc = self.correct_num / self.all_num
+            self.reset()
+            return {'acc': acc}
+
+        def reset(self):
+            # reset metric
+            self.correct_num = 0
+            self.all_num = 0
+
+    ```
+
+3. Import the added module in the [ppocr/metrics/\__init\__.py](../../../ppocr/metrics/__init__.py) file.
+
+After the metric module is added, you only need to configure it in the configuration file to use it, such as:
+
+```yaml linenums="1"
+Metric:
+  name: MyMetric
+  main_indicator: acc
+```
+
+## Optimizer
+
+The optimizer is used to train the network. The optimizer also contains network regularization and learning rate decay modules. This part is under [ppocr/optimizer](../../../ppocr/optimizer). PaddleOCR has built-in
+Commonly used optimizer modules such as `Momentum`, `Adam` and `RMSProp`, common regularization modules such as `Linear`, `Cosine`, `Step` and `Piecewise`, and common learning rate decay modules such as `L1Decay` and `L2Decay`.
+Modules without built-in can be added through the following steps, take `optimizer` as an example:
+
+1. Create your own optimizer in the [ppocr/optimizer/optimizer.py](../../../ppocr/optimizer/optimizer.py) file, the sample code is as follows:
+
+    ```python linenums="1"
+    from paddle import optimizer as optim
+
+
+    class MyOptim(object):
+        def __init__(self, learning_rate=0.001, *args, **kwargs):
+            self.learning_rate = learning_rate
+
+        def __call__(self, parameters):
+            # It is recommended to wrap the built-in optimizer of paddle
+            opt = optim.XXX(
+                learning_rate=self.learning_rate,
+                parameters=parameters)
+            return opt
+
+    ```
+
+After the optimizer module is added, you only need to configure it in the configuration file to use, such as:
+
+```yaml linenums="1"
+Optimizer:
+  name: MyOptim
+  args1: args1
+  args2: args2
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0
+```
diff --git a/docs/version2.x/algorithm/add_new_algorithm.md b/docs/version2.x/algorithm/add_new_algorithm.md
new file mode 100644
index 0000000..8b98725
--- /dev/null
+++ b/docs/version2.x/algorithm/add_new_algorithm.md
@@ -0,0 +1,300 @@
+---
+comments: true
+---
+
+# 添加新算法
+
+PaddleOCR将一个算法分解为以下几个部分，并对各部分进行模块化处理，方便快速组合出新的算法。
+
+下面将分别对每个部分进行介绍，并介绍如何在该部分里添加新算法所需模块。
+
+## 1. 数据加载和处理
+
+数据加载和处理由不同的模块(module)组成，其完成了图片的读取、数据增强和label的制作。这一部分在[ppocr/data](../../../ppocr/data)下。 各个文件及文件夹作用说明如下:
+
+```bash linenums="1"
+ppocr/data/
+├── imaug             # 图片的读取、数据增强和label制作相关的文件
+│   ├── label_ops.py  # 对label进行变换的modules
+│   ├── operators.py  # 对image进行变换的modules
+│   ├──.....
+├── __init__.py
+├── lmdb_dataset.py   # 读取lmdb的数据集的dataset
+└── simple_dataset.py # 读取以`image_path\tgt`形式保存的数据集的dataset
+```
+
+PaddleOCR内置了大量图像操作相关模块，对于没有没有内置的模块可通过如下步骤添加:
+
+1. 在 [ppocr/data/imaug](../../../ppocr/data/imaug) 文件夹下新建文件，如my_module.py。
+2. 在 my_module.py 文件内添加相关代码，示例代码如下:
+
+    ```python linenums="1"
+    class MyModule:
+        def __init__(self, *args, **kwargs):
+            # your init code
+            pass
+
+        def __call__(self, data):
+            img = data['image']
+            label = data['label']
+            # your process code
+
+            data['image'] = img
+            data['label'] = label
+            return data
+    ```
+
+3. 在 [ppocr/data/imaug/\__init\__.py](../../../ppocr/data/imaug/__init__.py) 文件内导入添加的模块。
+
+数据处理的所有处理步骤由不同的模块顺序执行而成，在config文件中按照列表的形式组合并执行。如:
+
+```yaml linenums="1"
+# angle class data process
+transforms:
+  - DecodeImage: # load image
+      img_mode: BGR
+      channel_first: False
+  - MyModule:
+      args1: args1
+      args2: args2
+  - KeepKeys:
+      keep_keys: [ 'image', 'label' ] # dataloader will return list in this order
+```
+
+## 2. 网络
+
+网络部分完成了网络的组网操作，PaddleOCR将网络划分为四部分，这一部分在[ppocr/modeling](../../../ppocr/modeling)下。 进入网络的数据将按照顺序(transforms->backbones->
+necks->heads)依次通过这四个部分。
+
+```bash linenums="1"
+├── architectures # 网络的组网代码
+├── transforms    # 网络的图像变换模块
+├── backbones     # 网络的特征提取模块
+├── necks         # 网络的特征增强模块
+└── heads         # 网络的输出模块
+```
+
+PaddleOCR内置了DB,EAST,SAST,CRNN和Attention等算法相关的常用模块，对于没有内置的模块可通过如下步骤添加，四个部分添加步骤一致，以backbones为例:
+
+1. 在 [ppocr/modeling/backbones](../../../ppocr/modeling/backbones) 文件夹下新建文件，如my_backbone.py。
+2. 在 my_backbone.py 文件内添加相关代码，示例代码如下:
+
+    ```python linenums="1"
+    import paddle
+    import paddle.nn as nn
+    import paddle.nn.functional as F
+
+
+    class MyBackbone(nn.Layer):
+        def __init__(self, *args, **kwargs):
+            super(MyBackbone, self).__init__()
+            # your init code
+            self.conv = nn.xxxx
+
+        def forward(self, inputs):
+            # your network forward
+            y = self.conv(inputs)
+            return y
+    ```
+
+3. 在 [ppocr/modeling/backbones/\__init\__.py](../../../ppocr/modeling/backbones/__init__.py)文件内导入添加的模块。
+
+在完成网络的四部分模块添加之后，只需要配置文件中进行配置即可使用，如:
+
+```yaml linenums="1"
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+    name: MyTransform
+    args1: args1
+    args2: args2
+  Backbone:
+    name: MyBackbone
+    args1: args1
+  Neck:
+    name: MyNeck
+    args1: args1
+  Head:
+    name: MyHead
+    args1: args1
+```
+
+## 3. 后处理
+
+后处理实现解码网络输出获得文本框或者识别到的文字。这一部分在[ppocr/postprocess](../../../ppocr/postprocess)下。
+PaddleOCR内置了DB,EAST,SAST,CRNN和Attention等算法相关的后处理模块，对于没有内置的组件可通过如下步骤添加:
+
+1. 在 [ppocr/postprocess](../../../ppocr/postprocess) 文件夹下新建文件，如 my_postprocess.py。
+2. 在 my_postprocess.py 文件内添加相关代码，示例代码如下:
+
+    ```python linenums="1"
+    import paddle
+
+
+    class MyPostProcess:
+        def __init__(self, *args, **kwargs):
+            # your init code
+            pass
+
+        def __call__(self, preds, label=None, *args, **kwargs):
+            if isinstance(preds, paddle.Tensor):
+                preds = preds.numpy()
+            # you preds decode code
+            preds = self.decode_preds(preds)
+            if label is None:
+                return preds
+            # you label decode code
+            label = self.decode_label(label)
+            return preds, label
+
+        def decode_preds(self, preds):
+            # you preds decode code
+            pass
+
+        def decode_label(self, preds):
+            # you label decode code
+            pass
+    ```
+
+3. 在 [ppocr/postprocess/\__init\__.py](../../../ppocr/postprocess/__init__.py)文件内导入添加的模块。
+
+在后处理模块添加之后，只需要配置文件中进行配置即可使用，如:
+
+```yaml linenums="1"
+PostProcess:
+  name: MyPostProcess
+  args1: args1
+  args2: args2
+```
+
+## 4. 损失函数
+
+损失函数用于计算网络输出和label之间的距离。这一部分在[ppocr/losses](../../../ppocr/losses)下。
+PaddleOCR内置了DB,EAST,SAST,CRNN和Attention等算法相关的损失函数模块，对于没有内置的模块可通过如下步骤添加:
+
+1. 在 [ppocr/losses](../../../ppocr/losses) 文件夹下新建文件，如 my_loss.py。
+2. 在 my_loss.py 文件内添加相关代码，示例代码如下:
+
+    ```python linenums="1"
+    import paddle
+    from paddle import nn
+
+
+    class MyLoss(nn.Layer):
+        def __init__(self, **kwargs):
+            super(MyLoss, self).__init__()
+            # you init code
+            pass
+
+        def __call__(self, predicts, batch):
+            label = batch[1]
+            # your loss code
+            loss = self.loss(input=predicts, label=label)
+            return {'loss': loss}
+    ```
+
+3. 在 [ppocr/losses/\__init\__.py](../../../ppocr/losses/__init__.py)文件内导入添加的模块。
+
+在损失函数添加之后，只需要配置文件中进行配置即可使用，如:
+
+```yaml linenums="1"
+Loss:
+  name: MyLoss
+  args1: args1
+  args2: args2
+```
+
+## 5. 指标评估
+
+指标评估用于计算网络在当前batch上的性能。这一部分在[ppocr/metrics](../../../ppocr/metrics)下。 PaddleOCR内置了检测，分类和识别等算法相关的指标评估模块，对于没有内置的模块可通过如下步骤添加:
+
+1. 在 [ppocr/metrics](../../../ppocr/metrics) 文件夹下新建文件，如my_metric.py。
+2. 在 my_metric.py 文件内添加相关代码，示例代码如下:
+
+    ```python linenums="1"
+
+    class MyMetric(object):
+        def __init__(self, main_indicator='acc', **kwargs):
+            # main_indicator is used for select best model
+            self.main_indicator = main_indicator
+            self.reset()
+
+        def __call__(self, preds, batch, *args, **kwargs):
+            # preds is out of postprocess
+            # batch is out of dataloader
+            labels = batch[1]
+            cur_correct_num = 0
+            cur_all_num = 0
+            # you metric code
+            self.correct_num += cur_correct_num
+            self.all_num += cur_all_num
+            return {'acc': cur_correct_num / cur_all_num, }
+
+        def get_metric(self):
+            """
+            return metrics {
+                    'acc': 0,
+                    'norm_edit_dis': 0,
+                }
+            """
+            acc = self.correct_num / self.all_num
+            self.reset()
+            return {'acc': acc}
+
+        def reset(self):
+            # reset metric
+            self.correct_num = 0
+            self.all_num = 0
+
+    ```
+
+3. 在 [ppocr/metrics/\__init\__.py](../../../ppocr/metrics/__init__.py)文件内导入添加的模块。
+
+在指标评估模块添加之后，只需要配置文件中进行配置即可使用，如:
+
+```yaml linenums="1"
+Metric:
+  name: MyMetric
+  main_indicator: acc
+```
+
+## 6. 优化器
+
+优化器用于训练网络。优化器内部还包含了网络正则化和学习率衰减模块。 这一部分在[ppocr/optimizer](../../../ppocr/optimizer)下。 PaddleOCR内置了`Momentum`,`Adam`
+和`RMSProp`等常用的优化器模块，`Linear`,`Cosine`,`Step`和`Piecewise`等常用的正则化模块与`L1Decay`和`L2Decay`等常用的学习率衰减模块。
+对于没有内置的模块可通过如下步骤添加，以`optimizer`为例:
+
+1. 在 [ppocr/optimizer/optimizer.py](../../../ppocr/optimizer/optimizer.py) 文件内创建自己的优化器，示例代码如下:
+
+    ```python linenums="1"
+    from paddle import optimizer as optim
+
+
+    class MyOptim(object):
+        def __init__(self, learning_rate=0.001, *args, **kwargs):
+            self.learning_rate = learning_rate
+
+        def __call__(self, parameters):
+            # It is recommended to wrap the built-in optimizer of paddle
+            opt = optim.XXX(
+                learning_rate=self.learning_rate,
+                parameters=parameters)
+            return opt
+
+    ```
+
+在优化器模块添加之后，只需要配置文件中进行配置即可使用，如:
+
+```yaml linenums="1"
+Optimizer:
+  name: MyOptim
+  args1: args1
+  args2: args2
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0
+```
diff --git a/docs/version2.x/algorithm/end_to_end/algorithm_e2e_pgnet.en.md b/docs/version2.x/algorithm/end_to_end/algorithm_e2e_pgnet.en.md
new file mode 100644
index 0000000..1fc45c9
--- /dev/null
+++ b/docs/version2.x/algorithm/end_to_end/algorithm_e2e_pgnet.en.md
@@ -0,0 +1,224 @@
+---
+comments: true
+---
+
+## 1. Brief Introduction
+
+OCR algorithms can be divided into two categories: two-stage algorithm and end-to-end algorithm. The two-stage OCR algorithm is generally divided into two parts, text detection and text recognition algorithm. The text detection algorithm locates the box of the text line from the image, and then the recognition algorithm identifies the content of the text box. The end-to-end OCR algorithm combines text detection and recognition in one algorithm. Its basic idea is to design a model with both detection unit and recognition module, share the CNN features of both and train them together. Because one algorithm can complete character recognition, the end-to-end model is smaller and faster.
+
+### Introduction Of PGNet Algorithm
+
+During the recent years, the end-to-end OCR algorithm has been well developed, including MaskTextSpotter series, TextSnake, TextDragon, PGNet series and so on. Among these algorithms, PGNet algorithm has some advantages over the other algorithms.
+
+- PGNet loss is designed to guide training, and no character-level annotations is needed.
+- NMS and ROI related operations are not needed. It can accelerate the prediction
+- The reading order prediction module is proposed
+- A graph based modification module (GRM) is proposed to further improve the performance of model recognition
+- Higher accuracy and faster prediction speed
+
+For details of PGNet algorithm, please refer to [paper](https://cdn.aaai.org/ojs/16383/16383-13-19877-1-2-20210518.pdf). The schematic diagram of the algorithm is as follows:
+
+![](./images/pgnet_framework.png)
+
+After feature extraction, the input image is sent to four branches: TBO module for text edge offset prediction, TCL module for text center-line prediction, TDO module for text direction offset prediction, and TCC module for text character classification graph prediction.
+The output of TBO and TCL can get text detection results after post-processing, and TCL, TDO and TCC are responsible for text recognition.
+
+The results of detection and recognition are as follows:
+
+![](./images/e2e_res_img293_pgnet.png)
+
+![](./images/e2e_res_img295_pgnet.png)
+
+### Performance
+
+#### Test set: Total Text
+
+#### Test environment: NVIDIA Tesla V100-SXM2-16GB
+
+|PGNetA|det_precision|det_recall|det_f_score|e2e_precision|e2e_recall|e2e_f_score|FPS|download|
+| --- | --- | --- | --- | --- | --- | --- | --- | --- |
+|Paper|85.30|86.80|86.10|-|-|61.70|38.20 (size=640)|-|
+|Ours|87.03|82.48|84.69|61.71|58.43|60.03|48.73 (size=768)|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/en_server_pgnetA.tar)|
+
+*note:PGNet in PaddleOCR optimizes the prediction speed, and can significantly improve the end-to-end prediction speed within the acceptable range of accuracy reduction*
+
+## 2. Environment Configuration
+
+Please refer to [Operation Environment Preparation](../../ppocr/environment.en.md) to configure PaddleOCR operating environment first, refer to [Project Clone](../../ppocr/blog/clone.en.md) to clone the project
+
+## 3. Quick Use
+
+### Inference model download
+
+This section takes the trained end-to-end model as an example to quickly use the model prediction. First, download the trained end-to-end inference model [download address](https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/e2e_server_pgnetA_infer.tar)
+
+```bash linenums="1"
+mkdir inference && cd inference
+# Download the English end-to-end model and unzip it
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/e2e_server_pgnetA_infer.tar && tar xf e2e_server_pgnetA_infer.tar
+```
+
+- In Windows environment, if 'wget' is not installed, the link can be copied to the browser when downloading the model, and decompressed and placed in the corresponding directory
+
+After decompression, there should be the following file structure:
+
+```text linenums="1"
+├── e2e_server_pgnetA_infer
+│   ├── inference.pdiparams
+│   ├── inference.pdiparams.info
+│   └── inference.pdmodel
+```
+
+### Single image or image set prediction
+
+```bash linenums="1"
+# Prediction single image specified by image_dir
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img623.jpg" --e2e_model_dir="./inference/e2e_server_pgnetA_infer/" --e2e_pgnet_valid_set="totaltext"
+
+# Prediction the collection of images specified by image_dir
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/" --e2e_model_dir="./inference/e2e_server_pgnetA_infer/" --e2e_pgnet_valid_set="totaltext"
+
+# If you want to use CPU for prediction, you need to set use_gpu parameter is false
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img623.jpg" --e2e_model_dir="./inference/e2e_server_pgnetA_infer/" --use_gpu=False --e2e_pgnet_valid_set="totaltext"
+```
+
+### Visualization results
+
+The visualized end-to-end results are saved to the `./inference_results` folder by default, and the name of the result file is prefixed with 'e2e_res'. Examples of results are as follows:
+
+![](./images//e2e_res_img623_pgnet.jpg)
+
+## 4. Model Training,Evaluation And Inference
+
+This section takes the totaltext dataset as an example to introduce the training, evaluation and testing of the end-to-end model in PaddleOCR.
+
+### Data Preparation
+
+Download and unzip [totaltext](https://paddleocr.bj.bcebos.com/dataset/total_text.tar) dataset to PaddleOCR/train_data/, dataset organization structure is as follow:
+
+```text linenums="1"
+/PaddleOCR/train_data/total_text/train/
+  |- rgb/            # total_text training data of dataset
+      |- img11.png
+      | ...
+  |- train.txt       # total_text training annotation of dataset
+```
+
+total_text.txt: the format of dimension file is as follows，the file name and annotation information are separated by "\t":
+
+```text linenums="1"
+" Image file name             Image annotation information encoded by json.dumps"
+rgb/img11.jpg    [{"transcription": "ASRAMA", "points": [[214.0, 325.0], [235.0, 308.0], [259.0, 296.0], [286.0, 291.0], [313.0, 295.0], [338.0, 305.0], [362.0, 320.0], [349.0, 347.0], [330.0, 337.0], [310.0, 329.0], [290.0, 324.0], [269.0, 328.0], [249.0, 336.0], [231.0, 346.0]]}, {...}]
+```
+
+The image annotation after **json.dumps()** encoding is a list containing multiple dictionaries.
+
+The `points` in the dictionary represent the coordinates (x, y) of the four points of the text box, arranged clockwise from the point at the upper left corner.
+
+`transcription` represents the text of the current text box. **When its content is "###" it means that the text box is invalid and will be skipped during training.**
+
+If you want to train PaddleOCR on other datasets, please build the annotation file according to the above format.
+
+### Start Training
+
+PGNet training is divided into two steps: Step 1: training on the synthetic data to get the pretrain_model, and the accuracy of the model is still low; step 2: loading the pretrain_model and training on the totaltext data set; for fast training, we directly provide the pre training model of step 1[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/train_step1.tar).
+
+```bash linenums="1"
+cd PaddleOCR/
+
+# download step1 pretrain_models
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/train_step1.tar
+
+# You can get the following file format
+./pretrain_models/train_step1/
+  └─ best_accuracy.pdopt
+  └─ best_accuracy.states
+  └─ best_accuracy.pdparams
+```
+
+*If CPU version installed, please set the parameter `use_gpu` to `false` in the configuration.*
+
+```bash linenums="1"
+# single GPU training
+python3 tools/train.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.pretrained_model=./pretrain_models/train_step1/best_accuracy Global.load_static_weights=False
+# multi-GPU training
+# Set the GPU ID used by the '--gpus' parameter.
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.pretrained_model=./pretrain_models/train_step1/best_accuracy  Global.load_static_weights=False
+```
+
+In the above instruction, use `-c` to select the training to use the `configs/e2e/e2e_r50_vd_pg.yml` configuration file.
+For a detailed explanation of the configuration file, please refer to [config](../../ppocr/blog/config.en.md).
+
+You can also use `-o` to change the training parameters without modifying the yml file. For example, adjust the training learning rate to 0.0001
+
+```bash linenums="1"
+python3 tools/train.py -c configs/e2e/e2e_r50_vd_pg.yml -o Optimizer.base_lr=0.0001
+```
+
+#### Load trained model and continue training
+
+If you would like to load trained model and continue the training again, you can specify the parameter `Global.checkpoints` as the model path to be loaded.
+
+```bash linenums="1"
+python3 tools/train.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.checkpoints=./your/trained/model
+```
+
+**Note**: The priority of `Global.checkpoints` is higher than that of `Global.pretrain_weights`, that is, when two parameters are specified at the same time, the model specified by `Global.checkpoints` will be loaded first. If the model path specified by `Global.checkpoints` is wrong, the one specified by `Global.pretrain_weights` will be loaded.
+
+PaddleOCR calculates three indicators for evaluating performance of OCR end-to-end task: Precision, Recall, and Hmean.
+
+Run the following code to calculate the evaluation indicators. The result will be saved in the test result file specified by `save_res_path` in the configuration file `e2e_r50_vd_pg.yml`
+When evaluating, set post-processing parameters `max_side_len=768`. If you use different datasets, different models for training.
+The model parameters during training are saved in the `Global.save_model_dir` directory by default. When evaluating indicators, you need to set `Global.checkpoints` to point to the saved parameter file.
+
+```bash linenums="1"
+python3 tools/eval.py -c configs/e2e/e2e_r50_vd_pg.yml  -o Global.checkpoints="{path/to/weights}/best_accuracy"
+```
+
+### Model Test
+
+Test the end-to-end result on a single image:
+
+```bash linenums="1"
+python3 tools/infer_e2e.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.infer_img="./doc/imgs_en/img_10.jpg" Global.pretrained_model="./output/e2e_pgnet/best_accuracy" Global.load_static_weights=false
+```
+
+Test the end-to-end result on all images in the folder:
+
+```bash linenums="1"
+python3 tools/infer_e2e.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.infer_img="./doc/imgs_en/" Global.pretrained_model="./output/e2e_pgnet/best_accuracy" Global.load_static_weights=false
+```
+
+### Model inference
+
+#### (1).Quadrangle text detection model (ICDAR2015)
+
+First, convert the model saved in the PGNet end-to-end training process into an inference model. In the first stage of training based on composite dataset, the model of English data set training is taken as an example[model download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/en_server_pgnetA.tar), you can use the following command to convert:
+
+```bash linenums="1"
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/en_server_pgnetA.tar && tar xf en_server_pgnetA.tar
+python3 tools/export_model.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.pretrained_model=./en_server_pgnetA/best_accuracy Global.load_static_weights=False Global.save_inference_dir=./inference/e2e
+```
+
+**For PGNet quadrangle end-to-end model inference, you need to set the parameter `--e2e_algorithm="PGNet"` and `--e2e_pgnet_valid_set="partvgg"`**, run the following command:
+
+```bash linenums="1"
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img_10.jpg" --e2e_model_dir="./inference/e2e/" --e2e_pgnet_valid_set="partvgg"
+```
+
+The visualized text detection results are saved to the `./inference_results` folder by default, and the name of the result file is prefixed with 'e2e_res'. Examples of results are as follows:
+
+![](./images//e2e_res_img_10_pgnet.jpg)
+
+#### (2). Curved text detection model (Total-Text)
+
+For the curved text example, we use the same model as the quadrilateral
+**For PGNet end-to-end curved text detection model inference, you need to set the parameter `--e2e_algorithm="PGNet"` and `--e2e_pgnet_valid_set="totaltext"`**, run the following command:
+
+```bash linenums="1"
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img623.jpg" --e2e_model_dir="./inference/e2e/" --e2e_pgnet_valid_set="totaltext"
+```
+
+The visualized text detection results are saved to the `./inference_results` folder by default, and the name of the result file is prefixed with 'e2e_res'. Examples of results are as follows:
+
+![](./images//e2e_res_img623_pgnet.jpg)
diff --git a/docs/version2.x/algorithm/end_to_end/algorithm_e2e_pgnet.md b/docs/version2.x/algorithm/end_to_end/algorithm_e2e_pgnet.md
new file mode 100644
index 0000000..f6fd70d
--- /dev/null
+++ b/docs/version2.x/algorithm/end_to_end/algorithm_e2e_pgnet.md
@@ -0,0 +1,220 @@
+---
+comments: true
+---
+
+## 一、简介
+
+OCR算法可以分为两阶段算法和端对端的算法。二阶段OCR算法一般分为两个部分，文本检测和文本识别算法，文件检测算法从图像中得到文本行的检测框，然后识别算法去识别文本框中的内容。而端对端OCR算法可以在一个算法中完成文字检测和文字识别，其基本思想是设计一个同时具有检测单元和识别模块的模型，共享其中两者的CNN特征，并联合训练。由于一个算法即可完成文字识别，端对端模型更小，速度更快。
+
+### PGNet算法介绍
+
+近些年来，端对端OCR算法得到了良好的发展，包括MaskTextSpotter系列、TextSnake、TextDragon、PGNet系列等算法。在这些算法中，PGNet算法具备其他算法不具备的优势，包括：
+
+- 设计PGNet loss指导训练，不需要字符级别的标注
+- 不需要NMS和ROI相关操作，加速预测
+- 提出预测文本行内的阅读顺序模块；
+- 提出基于图的修正模块（GRM）来进一步提高模型识别性能
+- 精度更高，预测速度更快
+
+PGNet算法细节详见[论文](https://cdn.aaai.org/ojs/16383/16383-13-19877-1-2-20210518.pdf) ,算法原理图如下所示：
+
+![](./images/pgnet_framework.png)
+
+输入图像经过特征提取送入四个分支，分别是：文本边缘偏移量预测TBO模块，文本中心线预测TCL模块，文本方向偏移量预测TDO模块，以及文本字符分类图预测TCC模块。
+其中TBO以及TCL的输出经过后处理后可以得到文本的检测结果，TCL、TDO、TCC负责文本识别。
+
+其检测识别效果图如下：
+
+![](./images//e2e_res_img293_pgnet.png)
+
+![](./images//e2e_res_img295_pgnet.png)
+
+### 性能指标
+
+#### 测试集: Total Text
+
+#### 测试环境: NVIDIA Tesla V100-SXM2-16GB
+
+|PGNetA|det_precision|det_recall|det_f_score|e2e_precision|e2e_recall|e2e_f_score|FPS|下载|
+| --- | --- | --- | --- | --- | --- | --- | --- | --- |
+|Paper|85.30|86.80|86.10|-|-|61.70|38.20 (size=640)|-|
+|Ours|87.03|82.48|84.69|61.71|58.43|60.03|48.73 (size=768)|[下载链接](https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/en_server_pgnetA.tar)|
+
+*note：PaddleOCR里的PGNet实现针对预测速度做了优化，在精度下降可接受范围内，可以显著提升端对端预测速度*
+
+## 二、环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目
+
+## 三、快速使用
+
+### inference模型下载
+
+本节以训练好的端到端模型为例，快速使用模型预测，首先下载训练好的端到端inference模型[下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/e2e_server_pgnetA_infer.tar)
+
+```bash linenums="1"
+mkdir inference && cd inference
+# 下载英文端到端模型并解压
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/e2e_server_pgnetA_infer.tar && tar xf e2e_server_pgnetA_infer.tar
+```
+
+- windows 环境下如果没有安装wget,下载模型时可将链接复制到浏览器中下载，并解压放置在相应目录下
+
+解压完毕后应有如下文件结构：
+
+```text linenums="1"
+├── e2e_server_pgnetA_infer
+│   ├── inference.pdiparams
+│   ├── inference.pdiparams.info
+│   └── inference.pdmodel
+```
+
+### 单张图像或者图像集合预测
+
+```bash linenums="1"
+# 预测image_dir指定的单张图像
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img623.jpg" --e2e_model_dir="./inference/e2e_server_pgnetA_infer/" --e2e_pgnet_valid_set="totaltext"
+
+# 预测image_dir指定的图像集合
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/" --e2e_model_dir="./inference/e2e_server_pgnetA_infer/" --e2e_pgnet_valid_set="totaltext"
+
+# 如果想使用CPU进行预测，需设置use_gpu参数为False
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img623.jpg" --e2e_model_dir="./inference/e2e_server_pgnetA_infer/" --e2e_pgnet_valid_set="totaltext" --use_gpu=False
+```
+
+### 可视化结果
+
+可视化文本检测结果默认保存到./inference_results文件夹里面，结果文件的名称前缀为'e2e_res'。结果示例如下：
+
+![](./images//e2e_res_img623_pgnet.jpg)
+
+## 四、模型训练、评估、推理
+
+本节以totaltext数据集为例，介绍PaddleOCR中端到端模型的训练、评估与测试。
+
+### 准备数据
+
+下载解压[totaltext](https://paddleocr.bj.bcebos.com/dataset/total_text.tar) 数据集到PaddleOCR/train_data/目录，数据集组织结构：
+
+```text linenums="1"
+/PaddleOCR/train_data/total_text/train/
+  |- rgb/            # total_text数据集的训练数据
+      |- img11.jpg
+      | ...
+  |- train.txt       # total_text数据集的训练标注
+```
+
+train.txt标注文件格式如下，文件名和标注信息中间用"\t"分隔：
+
+```text linenums="1"
+" 图像文件名                    json.dumps编码的图像标注信息"
+rgb/img11.jpg    [{"transcription": "ASRAMA", "points": [[214.0, 325.0], [235.0, 308.0], [259.0, 296.0], [286.0, 291.0], [313.0, 295.0], [338.0, 305.0], [362.0, 320.0], [349.0, 347.0], [330.0, 337.0], [310.0, 329.0], [290.0, 324.0], [269.0, 328.0], [249.0, 336.0], [231.0, 346.0]]}, {...}]
+```
+
+json.dumps编码前的图像标注信息是包含多个字典的list，字典中的 `points` 表示文本框的四个点的坐标(x, y)，从左上角的点开始顺时针排列。
+`transcription` 表示当前文本框的文字，**当其内容为“###”时，表示该文本框无效，在训练时会跳过。**
+如果您想在其他数据集上训练，可以按照上述形式构建标注文件。
+
+### 启动训练
+
+PGNet训练分为两个步骤：step1: 在合成数据上训练，得到预训练模型，此时模型精度依然较低；step2: 加载预训练模型，在totaltext数据集上训练；为快速训练，我们直接提供了step1的预训练模型。
+
+```bash linenums="1"
+cd PaddleOCR/
+# 下载step1 预训练模型
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/train_step1.tar
+
+# 可以得到以下的文件格式
+./pretrain_models/train_step1/
+  └─ best_accuracy.pdopt
+  └─ best_accuracy.states
+  └─ best_accuracy.pdparams
+```
+
+*如果您安装的是cpu版本，请将配置文件中的 `use_gpu` 字段修改为false*
+
+```bash linenums="1"
+# 单机单卡训练 e2e 模型
+python3 tools/train.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.pretrained_model=./pretrain_models/train_step1/best_accuracy Global.load_static_weights=False
+# 单机多卡训练，通过 --gpus 参数设置使用的GPU ID
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.pretrained_model=./pretrain_models/train_step1/best_accuracy  Global.load_static_weights=False
+```
+
+上述指令中，通过-c 选择训练使用configs/e2e/e2e_r50_vd_pg.yml配置文件。
+有关配置文件的详细解释，请参考[链接](../../ppocr/blog/config.md)。
+
+您也可以通过-o参数在不需要修改yml文件的情况下，改变训练的参数，比如，调整训练的学习率为0.0001
+
+```bash linenums="1"
+python3 tools/train.py -c configs/e2e/e2e_r50_vd_pg.yml -o Optimizer.base_lr=0.0001
+```
+
+#### 断点训练
+
+如果训练程序中断，如果希望加载训练中断的模型从而恢复训练，可以通过指定Global.checkpoints指定要加载的模型路径：
+
+```bash linenums="1"
+python3 tools/train.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.checkpoints=./your/trained/model
+```
+
+**注意**：`Global.checkpoints`的优先级高于`Global.pretrain_weights`的优先级，即同时指定两个参数时，优先加载`Global.checkpoints`指定的模型，如果`Global.checkpoints`指定的模型路径有误，会加载`Global.pretrain_weights`指定的模型。
+
+PaddleOCR计算三个OCR端到端相关的指标，分别是：Precision、Recall、Hmean。
+
+运行如下代码，根据配置文件`e2e_r50_vd_pg.yml`中`save_res_path`指定的测试集检测结果文件，计算评估指标。
+
+评估时设置后处理参数`max_side_len=768`，使用不同数据集、不同模型训练，可调整参数进行优化
+训练中模型参数默认保存在`Global.save_model_dir`目录下。在评估指标时，需要设置`Global.checkpoints`指向保存的参数文件。
+
+```bash linenums="1"
+python3 tools/eval.py -c configs/e2e/e2e_r50_vd_pg.yml  -o Global.checkpoints="{path/to/weights}/best_accuracy"
+```
+
+### 模型预测
+
+测试单张图像的端到端识别效果
+
+```bash linenums="1"
+python3 tools/infer_e2e.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.infer_img="./doc/imgs_en/img_10.jpg" Global.pretrained_model="./output/e2e_pgnet/best_accuracy" Global.load_static_weights=false
+```
+
+测试文件夹下所有图像的端到端识别效果
+
+```bash linenums="1"
+python3 tools/infer_e2e.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.infer_img="./doc/imgs_en/" Global.pretrained_model="./output/e2e_pgnet/best_accuracy" Global.load_static_weights=false
+```
+
+### 预测推理
+
+#### (1). 四边形文本检测模型（ICDAR2015）
+
+首先将PGNet端到端训练过程中保存的模型，转换成inference model。以基于Resnet50_vd骨干网络，以英文数据集训练的模型为例[模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/en_server_pgnetA.tar) ，可以使用如下命令进行转换：
+
+```bash linenums="1"
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/en_server_pgnetA.tar && tar xf en_server_pgnetA.tar
+python3 tools/export_model.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.pretrained_model=./en_server_pgnetA/best_accuracy Global.load_static_weights=False Global.save_inference_dir=./inference/e2e
+```
+
+**PGNet端到端模型推理，需要设置参数`--e2e_algorithm="PGNet"` and `--e2e_pgnet_valid_set="partvgg"`**，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img_10.jpg" --e2e_model_dir="./inference/e2e/"  --e2e_pgnet_valid_set="partvgg" --e2e_pgnet_valid_set="totaltext"
+```
+
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为'e2e_res'。结果示例如下：
+
+![](./images//e2e_res_img_10_pgnet.jpg)
+
+#### (2). 弯曲文本检测模型（Total-Text）
+
+对于弯曲文本样例
+
+**PGNet端到端模型推理，需要设置参数`--e2e_algorithm="PGNet"`，同时，还需要增加参数`--e2e_pgnet_valid_set="totaltext"`，**可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_e2e.py --e2e_algorithm="PGNet" --image_dir="./doc/imgs_en/img623.jpg" --e2e_model_dir="./inference/e2e/" --e2e_pgnet_valid_set="totaltext"
+```
+
+可视化文本端到端结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为'e2e_res'。结果示例如下：
+
+![](./images//e2e_res_img623_pgnet.jpg)
diff --git a/docs/version2.x/algorithm/end_to_end/images/e2e_res_img293_pgnet.png b/docs/version2.x/algorithm/end_to_end/images/e2e_res_img293_pgnet.png
new file mode 100644
index 0000000..232f829
Binary files /dev/null and b/docs/version2.x/algorithm/end_to_end/images/e2e_res_img293_pgnet.png differ
diff --git a/docs/version2.x/algorithm/end_to_end/images/e2e_res_img295_pgnet.png b/docs/version2.x/algorithm/end_to_end/images/e2e_res_img295_pgnet.png
new file mode 100644
index 0000000..69337e3
Binary files /dev/null and b/docs/version2.x/algorithm/end_to_end/images/e2e_res_img295_pgnet.png differ
diff --git a/docs/version2.x/algorithm/end_to_end/images/e2e_res_img623_pgnet.jpg b/docs/version2.x/algorithm/end_to_end/images/e2e_res_img623_pgnet.jpg
new file mode 100644
index 0000000..b45dc05
Binary files /dev/null and b/docs/version2.x/algorithm/end_to_end/images/e2e_res_img623_pgnet.jpg differ
diff --git a/docs/version2.x/algorithm/end_to_end/images/e2e_res_img_10_pgnet.jpg b/docs/version2.x/algorithm/end_to_end/images/e2e_res_img_10_pgnet.jpg
new file mode 100644
index 0000000..a096299
Binary files /dev/null and b/docs/version2.x/algorithm/end_to_end/images/e2e_res_img_10_pgnet.jpg differ
diff --git a/docs/version2.x/algorithm/end_to_end/images/pgnet_framework.png b/docs/version2.x/algorithm/end_to_end/images/pgnet_framework.png
new file mode 100644
index 0000000..88fbca3
Binary files /dev/null and b/docs/version2.x/algorithm/end_to_end/images/pgnet_framework.png differ
diff --git a/docs/version2.x/algorithm/formula_recognition/algorithm_rec_can.en.md b/docs/version2.x/algorithm/formula_recognition/algorithm_rec_can.en.md
new file mode 100644
index 0000000..552084e
--- /dev/null
+++ b/docs/version2.x/algorithm/formula_recognition/algorithm_rec_can.en.md
@@ -0,0 +1,99 @@
+---
+comments: true
+---
+
+## 1. Introduction
+
+Paper:
+> [When Counting Meets HMER: Counting-Aware Network for Handwritten Mathematical Expression Recognition](https://arxiv.org/abs/2207.11463)
+> Bohan Li, Ye Yuan, Dingkang Liang, Xiao Liu, Zhilong Ji, Jinfeng Bai, Wenyu Liu, Xiang Bai
+> ECCV, 2022
+
+Using CROHME handwrittem mathematical expression recognition datasets for training, and evaluating on its test sets, the algorithm reproduction effect is as follows:
+
+|Model|Backbone|config|exprate|Download link|
+| --- | --- | --- | --- | --- |
+|CAN|DenseNet|[rec_d28_can.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_d28_can.yml)|51.72%|[trained model](https://paddleocr.bj.bcebos.com/contribution/rec_d28_can_train.tar)|
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_d28_can.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_d28_can.yml
+```
+
+### Evaluation
+
+```bash linenums="1"
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_d28_can.yml -o Global.pretrained_model=./rec_d28_can_train/best_accuracy.pdparams
+```
+
+Prediction:
+
+```bash linenums="1"
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_d28_can.yml -o Architecture.Head.attdecoder.is_train=False Global.infer_img='./doc/crohme_demo/hme_00.jpg' Global.pretrained_model=./rec_d28_can_train/best_accuracy.pdparams
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the CAN handwritten mathematical expression recognition training process is converted into an inference model. you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_d28_can.yml -o Global.pretrained_model=./rec_d28_can_train/best_accuracy.pdparams Global.save_inference_dir=./inference/rec_d28_can/ Architecture.Head.attdecoder.is_train=False
+
+# The default output max length of the model is 36. If you need to predict a longer sequence, please specify its output sequence as an appropriate value when exporting the model, as: Architecture.Head.max_ text_ length=72
+```
+
+For CAN handwritten mathematical expression recognition model inference, the following commands can be executed:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/datasets/crohme_demo/hme_00.jpg" --rec_algorithm="CAN" --rec_batch_num=1 --rec_model_dir="./inference/rec_d28_can/" --rec_char_dict_path="./ppocr/utils/dict/latex_symbol_dict.txt"
+
+# If you need to predict on a picture with black characters on a white background, please set: -- rec_ image_ inverse=False
+```
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@misc{https://doi.org/10.48550/arxiv.2207.11463,
+  doi = {10.48550/ARXIV.2207.11463},
+  url = {https://arxiv.org/abs/2207.11463},
+  author = {Li, Bohan and Yuan, Ye and Liang, Dingkang and Liu, Xiao and Ji, Zhilong and Bai, Jinfeng and Liu, Wenyu and Bai, Xiang},
+  keywords = {Computer Vision and Pattern Recognition (cs.CV), Artificial Intelligence (cs.AI), FOS: Computer and information sciences, FOS: Computer and information sciences},
+  title = {When Counting Meets HMER: Counting-Aware Network for Handwritten Mathematical Expression Recognition},
+  publisher = {arXiv},
+  year = {2022},
+  copyright = {arXiv.org perpetual, non-exclusive license}
+}
+```
diff --git a/docs/version2.x/algorithm/formula_recognition/algorithm_rec_can.md b/docs/version2.x/algorithm/formula_recognition/algorithm_rec_can.md
new file mode 100644
index 0000000..1a07964
--- /dev/null
+++ b/docs/version2.x/algorithm/formula_recognition/algorithm_rec_can.md
@@ -0,0 +1,155 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# 手写数学公式识别算法-CAN
+
+## 1. 算法简介
+
+论文信息：
+> [When Counting Meets HMER: Counting-Aware Network for Handwritten Mathematical Expression Recognition](https://arxiv.org/abs/2207.11463)
+> Bohan Li, Ye Yuan, Dingkang Liang, Xiao Liu, Zhilong Ji, Jinfeng Bai, Wenyu Liu, Xiang Bai
+> ECCV, 2022
+
+`CAN`使用CROHME手写公式数据集进行训练，在对应测试集上的精度如下：
+
+|模型    |骨干网络|配置文件|ExpRate|下载链接|
+| ----- | ----- | ----- | ----- | ----- |
+|CAN|DenseNet|[rec_d28_can.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_d28_can.yml)|51.72%|[训练模型](https://paddleocr.bj.bcebos.com/contribution/rec_d28_can_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+### 3.1 模型训练
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练`CAN`识别模型时需要**更换配置文件**为`CAN`的[配置文件](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_d28_can.yml)。
+
+#### 启动训练
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_d28_can.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_d28_can.yml
+```
+
+**注意：**
+
+- 我们提供的数据集，即[`CROHME数据集`](https://paddleocr.bj.bcebos.com/dataset/CROHME.tar)将手写公式存储为黑底白字的格式，若您自行准备的数据集与之相反，即以白底黑字模式存储，请在训练时做出如下修改
+
+  ```bash linenums="1"
+  python3 tools/train.py -c configs/rec/rec_d28_can.yml -o Train.dataset.transforms.GrayImageChannelFormat.inverse=False
+  ```
+
+- 默认每训练1个epoch（1105次iteration）进行1次评估，若您更改训练的batch_size，或更换数据集，请在训练时作出如下修改
+
+  ```bash linenums="1"
+  python3 tools/train.py -c configs/rec/rec_d28_can.yml -o Global.eval_batch_step=[0, {length_of_dataset//batch_size}]
+  ```
+
+### 3.2 评估
+
+可下载已训练完成的[模型文件](https://paddleocr.bj.bcebos.com/contribution/rec_d28_can_train.tar)，使用如下命令进行评估：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。若使用自行训练保存的模型，请注意修改路径和文件名为{path/to/weights}/{model_name}。
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_d28_can.yml -o Global.pretrained_model=./rec_d28_can_train/best_accuracy.pdparams
+```
+
+### 3.3 预测
+
+使用如下命令进行单张图片预测：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/infer_rec.py -c configs/rec/rec_d28_can.yml -o Architecture.Head.attdecoder.is_train=False Global.infer_img='./doc/datasets/crohme_demo/hme_00.jpg' Global.pretrained_model=./rec_d28_can_train/best_accuracy.pdparams
+
+# 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/datasets/crohme_demo/'。
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将训练得到best模型，转换成inference model。这里以训练完成的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/contribution/rec_d28_can_train.tar) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/export_model.py -c configs/rec/rec_d28_can.yml -o Global.pretrained_model=./rec_d28_can_train/best_accuracy.pdparams Global.save_inference_dir=./inference/rec_d28_can/ Architecture.Head.attdecoder.is_train=False
+
+# 目前的静态图模型默认的输出长度最大为36，如果您需要预测更长的序列，请在导出模型时指定其输出序列为合适的值，例如 Architecture.Head.max_text_length=72
+```
+
+**注意：**
+如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否是所需要的字典文件。
+
+转换成功后，在目录下有三个文件：
+
+```text linenums="1"
+/inference/rec_d28_can/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+执行如下命令进行模型推理：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/datasets/crohme_demo/hme_00.jpg" --rec_algorithm="CAN" --rec_batch_num=1 --rec_model_dir="./inference/rec_d28_can/" --rec_char_dict_path="./ppocr/utils/dict/latex_symbol_dict.txt"
+
+# 预测文件夹下所有图像时，可修改image_dir为文件夹，如 --image_dir='./doc/datasets/crohme_demo/'。
+
+# 如果您需要在白底黑字的图片上进行预测，请设置 --rec_image_inverse=False
+```
+
+![测试图片样例](./images/hme_00.jpg)
+
+执行命令后，上面图像的预测结果（识别的文本）会打印到屏幕上，示例如下：
+
+```bash linenums="1"
+Predicts of ./doc/imgs_hme/hme_00.jpg:['x _ { k } x x _ { k } + y _ { k } y x _ { k }', []]
+```
+
+**注意**：
+
+- 需要注意预测图像为**黑底白字**，即手写公式部分为白色，背景为黑色的图片。
+- 在推理时需要设置参数`rec_char_dict_path`指定字典，如果您修改了字典，请修改该参数为您的字典文件。
+- 如果您修改了预处理方法，需修改`tools/infer/predict_rec.py`中CAN的预处理为您的预处理方法。
+
+### 4.2 C++推理部署
+
+由于C++预处理后处理还未支持CAN，所以暂未支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+1. CROHME数据集来自于[CAN源repo](https://github.com/LBH1024/CAN) 。
+
+## 引用
+
+```bibtex
+@misc{https://doi.org/10.48550/arxiv.2207.11463,
+  doi = {10.48550/ARXIV.2207.11463},
+  url = {https://arxiv.org/abs/2207.11463},
+  author = {Li, Bohan and Yuan, Ye and Liang, Dingkang and Liu, Xiao and Ji, Zhilong and Bai, Jinfeng and Liu, Wenyu and Bai, Xiang},
+  keywords = {Computer Vision and Pattern Recognition (cs.CV), Artificial Intelligence (cs.AI), FOS: Computer and information sciences, FOS: Computer and information sciences},
+  title = {When Counting Meets HMER: Counting-Aware Network for Handwritten Mathematical Expression Recognition},
+  publisher = {arXiv},
+  year = {2022},
+  copyright = {arXiv.org perpetual, non-exclusive license}
+}
+```
diff --git a/docs/version2.x/algorithm/formula_recognition/algorithm_rec_latex_ocr.en.md b/docs/version2.x/algorithm/formula_recognition/algorithm_rec_latex_ocr.en.md
new file mode 100644
index 0000000..eb35d0d
--- /dev/null
+++ b/docs/version2.x/algorithm/formula_recognition/algorithm_rec_latex_ocr.en.md
@@ -0,0 +1,105 @@
+# LaTeX-OCR
+
+## 1. Introduction
+
+Original Project:
+> [https://github.com/lukas-blecher/LaTeX-OCR](https://github.com/lukas-blecher/LaTeX-OCR)
+
+
+Using LaTeX-OCR printed mathematical expression recognition datasets for training, and evaluating on its test sets, the algorithm reproduction effect is as follows:
+
+| Model       | Backbone |config| BLEU score  | normed edit distance  |  ExpRate  |Download link|
+|-----------|----------| ---- |:-----------:|:---------------------:|:---------:| ----- |
+| LaTeX-OCR | Hybrid ViT |[LaTeX_OCR_rec.yaml](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/configs/rec/LaTeX_OCR_rec.yaml)|   0.8821    |        0.0823         |  40.01%   |[trained model](https://paddleocr.bj.bcebos.com/contribution/rec_latex_ocr_train.tar)|
+
+## 2. Environment
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md) to clone the project code.
+
+Furthermore, additional dependencies need to be installed:
+```shell
+pip install -r docs/algorithm/formula_recognition/requirements.txt
+```
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+Pickle File Generation:
+
+Download formulae.zip and math.txt in [Google Drive](https://drive.google.com/drive/folders/13CA4vAmOmD_I_dSbvLp-Lf0s6KiaNfuO), and then use the following command to generate the pickle file.
+
+```shell
+# Create a LaTeX-OCR dataset directory
+mkdir -p train_data/LaTeXOCR
+# Unzip formulae.zip and copy math.txt
+unzip -d train_data/LaTeXOCR path/formulae.zip
+cp path/math.txt train_data/LaTeXOCR
+# Convert the original .txt file to a .pkl file to group images of different scales
+# Training set conversion
+python ppocr/utils/formula_utils/math_txt2pkl.py --image_dir=train_data/LaTeXOCR/train --mathtxt_path=train_data/LaTeXOCR/math.txt --output_dir=train_data/LaTeXOCR/
+# Validation set conversion
+python ppocr/utils/formula_utils/math_txt2pkl.py --image_dir=train_data/LaTeXOCR/val --mathtxt_path=train_data/LaTeXOCR/math.txt --output_dir=train_data/LaTeXOCR/
+# Test set conversion
+python ppocr/utils/formula_utils/math_txt2pkl.py --image_dir=train_data/LaTeXOCR/test --mathtxt_path=train_data/LaTeXOCR/math.txt --output_dir=train_data/LaTeXOCR/
+```
+
+
+Training:
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```
+#Single GPU training (Default training method)
+python3 tools/train.py -c configs/rec/LaTeX_OCR_rec.yaml
+
+#Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/LaTeX_OCR_rec.yaml
+```
+
+Evaluation:
+
+```
+# GPU evaluation
+# Validation set evaluation
+python3 tools/eval.py -c configs/rec/LaTeX_OCR_rec.yaml -o Global.pretrained_model=./rec_latex_ocr_train/best_accuracy.pdparams
+# Test set evaluation
+python3 tools/eval.py -c configs/rec/LaTeX_OCR_rec.yaml -o Global.pretrained_model=./rec_latex_ocr_train/best_accuracy.pdparams Eval.dataset.data_dir=./train_data/LaTeXOCR/test Eval.dataset.data=./train_data/LaTeXOCR/latexocr_test.pkl
+```
+
+Prediction:
+
+```
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/LaTeX_OCR_rec.yaml  -o  Global.infer_img='./docs/datasets/images/pme_demo/0000013.png' Global.pretrained_model=./rec_latex_ocr_train/best_accuracy.pdparams
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+First, the model saved during the LaTeX-OCR printed mathematical expression recognition training process is converted into an inference model. you can use the following command to convert:
+
+```
+python3 tools/export_model.py -c configs/rec/LaTeX_OCR_rec.yaml -o Global.pretrained_model=./rec_latex_ocr_train/best_accuracy.pdparams Global.save_inference_dir=./inference/rec_latex_ocr_infer/ 
+
+# The default output max length of the model is 512.
+```
+
+For LaTeX-OCR printed mathematical expression recognition model inference, the following commands can be executed:
+
+```
+python3 tools/infer/predict_rec.py --image_dir='./docs/datasets/images/pme_demo/0000295.png' --rec_algorithm="LaTeXOCR" --rec_batch_num=1 --rec_model_dir="./inference/rec_latex_ocr_infer/"  --rec_char_dict_path="./ppocr/utils/dict/latex_ocr_tokenizer.json"
+```
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
diff --git a/docs/version2.x/algorithm/formula_recognition/algorithm_rec_latex_ocr.md b/docs/version2.x/algorithm/formula_recognition/algorithm_rec_latex_ocr.md
new file mode 100644
index 0000000..4bad426
--- /dev/null
+++ b/docs/version2.x/algorithm/formula_recognition/algorithm_rec_latex_ocr.md
@@ -0,0 +1,147 @@
+# 印刷数学公式识别算法-LaTeX-OCR
+
+## 1. 算法简介
+
+原始项目：
+> [https://github.com/lukas-blecher/LaTeX-OCR](https://github.com/lukas-blecher/LaTeX-OCR)
+
+
+`LaTeX-OCR`使用[`LaTeX-OCR印刷公式数据集`](https://drive.google.com/drive/folders/13CA4vAmOmD_I_dSbvLp-Lf0s6KiaNfuO)进行训练，在对应测试集上的精度如下：
+
+| 模型        | 骨干网络       |配置文件 | BLEU score  | normed edit distance  |  ExpRate  |下载链接|
+|-----------|------------| ----- |:-----------:|:---------------------:|:---------:| ----- |
+| LaTeX-OCR | Hybrid ViT |[LaTeX_OCR_rec.yaml](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/configs/rec/LaTeX_OCR_rec.yaml)|   0.8821    |        0.0823         |  40.01%   |[训练模型](https://paddleocr.bj.bcebos.com/contribution/rec_latex_ocr_train.tar)|
+
+## 2. 环境配置
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+此外，需要安装额外的依赖：
+```shell
+pip install -r docs/algorithm/formula_recognition/requirements.txt
+```
+
+## 3. 模型训练、评估、预测
+
+### 3.1 pickle 标签文件生成
+从[谷歌云盘](https://drive.google.com/drive/folders/13CA4vAmOmD_I_dSbvLp-Lf0s6KiaNfuO)中下载 formulae.zip 和 math.txt，之后，使用如下命令，生成 pickle 标签文件。
+
+```shell
+# 创建 LaTeX-OCR 数据集目录
+mkdir -p train_data/LaTeXOCR
+# 解压formulae.zip ，并拷贝math.txt
+unzip -d train_data/LaTeXOCR path/formulae.zip
+cp path/math.txt train_data/LaTeXOCR
+# 将原始的 .txt 文件转换为 .pkl 文件，从而对不同尺度的图像进行分组
+# 训练集转换
+python ppocr/utils/formula_utils/math_txt2pkl.py --image_dir=train_data/LaTeXOCR/train --mathtxt_path=train_data/LaTeXOCR/math.txt --output_dir=train_data/LaTeXOCR/
+# 验证集转换
+python ppocr/utils/formula_utils/math_txt2pkl.py --image_dir=train_data/LaTeXOCR/val --mathtxt_path=train_data/LaTeXOCR/math.txt --output_dir=train_data/LaTeXOCR/
+# 测试集转换
+python ppocr/utils/formula_utils/math_txt2pkl.py --image_dir=train_data/LaTeXOCR/test --mathtxt_path=train_data/LaTeXOCR/math.txt --output_dir=train_data/LaTeXOCR/
+```
+
+### 3.2 模型训练
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练`LaTeX-OCR`识别模型时需要**更换配置文件**为`LaTeX-OCR`的[配置文件](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/configs/rec/LaTeX_OCR_rec.yaml)。
+
+#### 启动训练
+
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+```shell
+#单卡训练 (默认训练方式)
+python3 tools/train.py -c configs/rec/LaTeX_OCR_rec.yaml
+#多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/LaTeX_OCR_rec.yaml
+```
+
+**注意：**
+
+- 默认每训练22个epoch（60000次iteration）进行1次评估，若您更改训练的batch_size，或更换数据集，请在训练时作出如下修改
+```
+python3 tools/train.py -c configs/rec/LaTeX_OCR_rec.yaml -o Global.eval_batch_step=[0,{length_of_dataset//batch_size*22}]
+```
+
+### 3.3 评估
+
+可下载已训练完成的[模型文件](https://paddleocr.bj.bcebos.com/contribution/rec_latex_ocr_train.tar)，使用如下命令进行评估：
+
+```shell
+# 注意将pretrained_model的路径设置为本地路径。若使用自行训练保存的模型，请注意修改路径和文件名为{path/to/weights}/{model_name}。
+# 验证集评估
+python3 tools/eval.py -c configs/rec/LaTeX_OCR_rec.yaml -o Global.pretrained_model=./rec_latex_ocr_train/best_accuracy.pdparams
+# 测试集评估
+python3 tools/eval.py -c configs/rec/LaTeX_OCR_rec.yaml -o Global.pretrained_model=./rec_latex_ocr_train/best_accuracy.pdparams Eval.dataset.data_dir=./train_data/LaTeXOCR/test Eval.dataset.data=./train_data/LaTeXOCR/latexocr_test.pkl
+```
+
+### 3.4 预测
+
+使用如下命令进行单张图片预测：
+```shell
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/infer_rec.py -c configs/rec/LaTeX_OCR_rec.yaml  -o  Global.infer_img='./docs/datasets/images/pme_demo/0000013.png' Global.pretrained_model=./rec_latex_ocr_train/best_accuracy.pdparams
+# 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/datasets/pme_demo/'。
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+首先将训练得到best模型，转换成inference model。这里以训练完成的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/contribution/rec_latex_ocr_train.tar)），可以使用如下命令进行转换：
+
+```shell
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/export_model.py -c configs/rec/LaTeX_OCR_rec.yaml -o Global.pretrained_model=./rec_latex_ocr_train/best_accuracy.pdparams Global.save_inference_dir=./inference/rec_latex_ocr_infer/ 
+
+# 目前的静态图模型支持的最大输出长度为512
+```
+**注意：**
+- 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请检查配置文件中的`rec_char_dict_path`是否为所需要的字典文件。
+- [转换后模型下载地址](https://paddleocr.bj.bcebos.com/contribution/rec_latex_ocr_infer.tar)
+
+转换成功后，在目录下有三个文件：
+```
+/inference/rec_latex_ocr_infer/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+执行如下命令进行模型推理：
+
+```shell
+python3 tools/infer/predict_rec.py --image_dir='./docs/datasets/images/pme_demo/0000295.png' --rec_algorithm="LaTeXOCR" --rec_batch_num=1 --rec_model_dir="./inference/rec_latex_ocr_infer/"  --rec_char_dict_path="./ppocr/utils/dict/latex_ocr_tokenizer.json"
+
+# 预测文件夹下所有图像时，可修改image_dir为文件夹，如 --image_dir='./doc/datasets/pme_demo/'。
+```
+&nbsp;
+
+![测试图片样例](../../../datasets/images/pme_demo/0000295.png)
+
+执行命令后，上面图像的预测结果（识别的公式）会打印到屏幕上，示例如下：
+```shell
+Predicts of ./doc/datasets/pme_demo/0000295.png:\zeta_{0}(\nu)=-{\frac{\nu\varrho^{-2\nu}}{\pi}}\int_{\mu}^{\infty}d\omega\int_{C_{+}}d z{\frac{2z^{2}}{(z^{2}+\omega^{2})^{\nu+1}}}{\tilde{\Psi}}(\omega;z)e^{i\epsilon z}~~~,
+```
+
+
+**注意**：
+
+- 需要注意预测图像为**白底黑字**，即手写公式部分为黑色，背景为白色的图片。
+- 在推理时需要设置参数`rec_char_dict_path`指定字典，如果您修改了字典，请修改该参数为您的字典文件。
+- 如果您修改了预处理方法，需修改`tools/infer/predict_rec.py`中 LaTeX-OCR 的预处理为您的预处理方法。
+
+
+### 4.2 C++推理部署
+
+由于C++预处理后处理还未支持 LaTeX-OCR，所以暂未支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+1. LaTeX-OCR 数据集来自于[LaTeXOCR源repo](https://github.com/lukas-blecher/LaTeX-OCR) 。
diff --git a/docs/version2.x/algorithm/formula_recognition/algorithm_rec_ppformulanet.md b/docs/version2.x/algorithm/formula_recognition/algorithm_rec_ppformulanet.md
new file mode 100644
index 0000000..ee8fa6a
--- /dev/null
+++ b/docs/version2.x/algorithm/formula_recognition/algorithm_rec_ppformulanet.md
@@ -0,0 +1,95 @@
+# 印刷数学公式识别算法-PP-FormulaNet
+
+## 1. 算法简介
+
+`PP-FormulaNet` 是由百度飞桨视觉团队开发的一款先进的公式识别模型，支持识别 5 万个常见 LaTeX 源码词汇。其中，PP-FormulaNet-S 版本采用 PP-HGNetV2-B4 作为骨干网络，通过并行掩码和模型蒸馏等技术大幅提升推理速度并保持高识别精度，适用于简单印刷公式和跨行简单印刷公式等场景；而 PP-FormulaNet-L 版本基于 Vary_VIT_B 并经过大规模公式数据集的深入训练，在复杂公式识别方面表现显著提升，适用于简单印刷、复杂印刷和手写公式。
+
+上述模型在对应测试集上的精度如下：
+
+| 模型          | 骨干网络       | 配置文件                   | En-BLEU↑  |GPU推理耗时（ms）| 下载链接 |
+|-----------|--------|----------------------------------------|:-----------------:|:--------------:|:--------------:|
+| UniMERNet | Donut Swin | [UniMERNet.yaml](../../../../configs/rec/UniMERNet.yaml) |     85.91   | 2266.96 | [训练模型](https://paddleocr.bj.bcebos.com/contribution/rec_unimernet_train.tar)|
+| PP-FormulaNet-S | PPHGNetV2_B4 | [PP-FormulaNet-S.yaml](../../../../configs/rec/PP-FormuaNet/PP-FormulaNet-S.yaml) |   87.00 | 202.25 |[训练模型](https://paddleocr.bj.bcebos.com/contribution/rec_ppformulanet_s_train.tar)|
+| PP-FormulaNet-L | Vary_VIT_B | [PP-FormulaNet-L.yaml](../../../../configs/rec/PP-FormuaNet/PP-FormulaNet-L.yaml) |    90.36    | 1976.52  |[训练模型](https://paddleocr.bj.bcebos.com/contribution/rec_ppformulanet_l_train.tar )|
+| LaTeX-OCR | Hybrid ViT |[LaTeX_OCR_rec.yaml](../../../../configs/rec/LaTeX_OCR_rec.yaml)|   74.55   | 	1244.61   |[训练模型](https://paddleocr.bj.bcebos.com/contribution/rec_latex_ocr_train.tar)|
+
+这里，英文公式评估集包含UniMERNet的简单和复杂公式，以及PaddleX内部自建的简单、中等和复杂公式。
+
+## 2. 环境配置
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+此外，需要安装额外的依赖：
+```shell
+sudo apt-get update
+sudo apt-get install libmagickwand-dev
+pip install -r docs/algorithm/formula_recognition/requirements.txt
+```
+
+## 3. 模型训练、评估、预测
+
+### 3.1 准备数据集
+
+```shell
+# 下载 PaddleX 官方示例数据集
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_rec_latexocr_dataset_example.tar
+tar -xf ocr_rec_latexocr_dataset_example.tar
+```
+
+
+### 3.2 下载预训练模型
+
+```shell
+# 下载 PP-FormulaNet-S 预训练模型
+wget https://paddleocr.bj.bcebos.com/contribution/rec_ppformulanet_s_train.tar 
+tar -xf rec_ppformulanet_s_train.tar
+```
+
+
+### 3.3 模型训练
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练 `PP-FormulaNet-S` 识别模型时需要**更换配置文件**为 `PP-FormulaNet-S` 的[配置文件](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/configs/rec/PP-FormuaNet/PP-FormulaNet-S.yaml)。
+
+#### 启动训练
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+```shell
+#单卡训练 (默认训练方式)
+python3 tools/train.py -c configs/rec/PP-FormuaNet/PP-FormulaNet-S.yaml \
+   -o Global.pretrained_model=./rec_ppformulanet_s_train/best_accuracy.pdparams
+#多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' --ips=127.0.0.1   tools/train.py -c configs/rec/PP-FormuaNet/PP-FormulaNet-S.yaml \
+        -o Global.pretrained_model=./rec_ppformulanet_s_train/best_accuracy.pdparams
+```
+
+**注意：**
+
+- 默认每训练 1个epoch（179 次iteration）进行1次评估，若您更改训练的batch_size，或更换数据集，请在训练时作出如下修改
+```
+python3  -m paddle.distributed.launch --gpus '0,1,2,3' --ips=127.0.0.1   tools/train.py -c configs/rec/PP-FormuaNet/PP-FormulaNet-S.yaml \
+  -o Global.eval_batch_step=[0,{length_of_dataset//batch_size//4}] \
+   Global.pretrained_model=./rec_ppformulanet_s_train/best_accuracy.pdparams
+```
+
+### 3.4 评估
+
+可下载已训练完成的[模型文件](https://paddleocr.bj.bcebos.com/contribution/rec_ppformulanet_s_train.tar )，使用如下命令进行评估：
+
+```shell
+# 注意将pretrained_model的路径设置为本地路径。若使用自行训练保存的模型，请注意修改路径和文件名为{path/to/weights}/{model_name}。
+ # demo 测试集评估
+ python3 tools/eval.py -c configs/rec/PP-FormuaNet/PP-FormulaNet-S.yaml -o \
+ Global.pretrained_model=./rec_ppformulanet_s_train/best_accuracy.pdparams
+```
+
+### 3.5 预测
+
+使用如下命令进行单张图片预测：
+```shell
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/infer_rec.py -c configs/rec/PP-FormuaNet/PP-FormulaNet-S.yaml \
+  -o  Global.infer_img='./docs/datasets/images/pme_demo/0000295.png'\
+   Global.pretrained_model=./rec_ppformulanet_s_train/best_accuracy.pdparams
+# 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/datasets/pme_demo/'。
+```
+
+## 4. FAQ
diff --git a/docs/version2.x/algorithm/formula_recognition/algorithm_rec_ppformulanet_en.md b/docs/version2.x/algorithm/formula_recognition/algorithm_rec_ppformulanet_en.md
new file mode 100644
index 0000000..ad16560
--- /dev/null
+++ b/docs/version2.x/algorithm/formula_recognition/algorithm_rec_ppformulanet_en.md
@@ -0,0 +1,82 @@
+# PP-FormulaNet
+
+## 1. Introduction
+
+
+`PP-FormulaNet` is an advanced formula recognition model developed by Baidu’s PaddlePaddle Vision Team, supporting the recognition of 50,000 common LaTeX source terms. The PP-FormulaNet-S version uses PP-HGNetV2-B4 as its backbone network, leveraging techniques like parallel masking and model distillation to significantly enhance inference speed while maintaining high recognition accuracy, making it suitable for scenarios involving simple printed formulas and cross-line simple printed formulas. On the other hand, the PP-FormulaNet-L version is based on Vary_VIT_B and has undergone extensive training on a large-scale formula dataset, showing significant improvement in recognizing complex formulas, and is applicable to simple printed, complex printed, and handwritten formulas.
+
+The accuracy of the above models on the corresponding test sets is as follows:
+
+| Model           | Backbone       | config                  | En-BLEU↑  | GPU Inference Time (ms)| Download link |
+|-----------|--------|----------------------------------------|:-----------------:|:--------------:|:--------------:|
+| UniMERNet | Donut Swin | [UniMERNet.yaml](../../../../configs/rec/UniMERNet.yaml) |     85.91  | 2266.96 | [trained model](https://paddleocr.bj.bcebos.com/contribution/rec_unimernet_train.tar)|
+| PP-FormulaNet-S | PPHGNetV2_B4 | [PP-FormulaNet-S.yaml](../../../../configs/rec/PP-FormuaNet/PP-FormulaNet-S.yaml) |   87.00   | 202.25 |[trained model](https://paddleocr.bj.bcebos.com/contribution/rec_ppformulanet_s_train.tar)|
+| PP-FormulaNet-L | Vary_VIT_B | [PP-FormulaNet-L.yaml](../../../../configs/rec/PP-FormuaNet/PP-FormulaNet-L.yaml) |    90.36   | 1976.52  |[trained model](https://paddleocr.bj.bcebos.com/contribution/rec_ppformulanet_l_train.tar )|
+| LaTeX-OCR | Hybrid ViT |[LaTeX_OCR_rec.yaml](../../../../configs/rec/LaTeX_OCR_rec.yaml)|   74.55   | 	1244.61   |[trained model](https://paddleocr.bj.bcebos.com/contribution/rec_latex_ocr_train.tar)|
+
+
+The English formula evaluation set here contains both simple and complex formulas from UniMERNet, as well as simple, medium, and complex formulas independently created by PaddleX.
+
+
+## 2. Environment
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md) to clone the project code.
+
+Furthermore, additional dependencies need to be installed:
+```shell
+sudo apt-get update
+sudo apt-get install libmagickwand-dev
+pip install -r docs/algorithm/formula_recognition/requirements.txt
+```
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+
+Dataset Preparation:
+
+```shell
+# download PaddleX official example dataset
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_rec_latexocr_dataset_example.tar
+tar -xf ocr_rec_latexocr_dataset_example.tar
+```
+
+Download the Pre-trained Model:
+
+```shell
+# download the PP-FormulaNet-S pretrained model
+wget https://paddleocr.bj.bcebos.com/contribution/rec_ppformulanet_s_train.tar 
+tar -xf rec_ppformulanet_s_train.tar
+```
+
+Training:
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```shell
+#Single GPU training 
+python3 tools/train.py -c configs/rec/PP-FormuaNet/PP-FormulaNet-S.yaml \
+   -o Global.pretrained_model=./rec_ppformulanet_s_train/best_accuracy.pdparams
+#Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' --ips=127.0.0.1   tools/train.py -c configs/rec/PP-FormuaNet/PP-FormulaNet-S.yaml \
+        -o Global.pretrained_model=./rec_ppformulanet_s_train/best_accuracy.pdparams
+```
+
+Evaluation:
+
+```shell
+# GPU evaluation
+ python3 tools/eval.py -c configs/rec/PP-FormuaNet/PP-FormulaNet-S.yaml -o \
+ Global.pretrained_model=./rec_ppformulanet_s_train/best_accuracy.pdparams
+```
+
+Prediction:
+
+```shell
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/PP-FormuaNet/PP-FormulaNet-S.yaml \
+  -o  Global.infer_img='./docs/datasets/images/pme_demo/0000295.png'\
+   Global.pretrained_model=./rec_ppformulanet_s_train/best_accuracy.pdparams
+```
+
+## 4. FAQ
diff --git a/docs/version2.x/algorithm/formula_recognition/algorithm_rec_unimernet.md b/docs/version2.x/algorithm/formula_recognition/algorithm_rec_unimernet.md
new file mode 100644
index 0000000..ff206c9
--- /dev/null
+++ b/docs/version2.x/algorithm/formula_recognition/algorithm_rec_unimernet.md
@@ -0,0 +1,154 @@
+# 通用数学公式识别算法-UniMERNet
+
+## 1. 算法简介
+
+原始项目：
+> [https://github.com/opendatalab/UniMERNet](https://github.com/opendatalab/UniMERNet)
+
+
+`UniMERNet`使用[`UniMERNet通用公式识别数据集`](https://opendatalab.com/OpenDataLab/UniMER-Dataset)进行训练，在对应测试集上的精度如下：
+
+| 模型        | 骨干网络       | 配置文件                                                  | SPE-<br/>BLEU↑ | SPE-<br/>EditDis↓ | CPE-<br/>BLEU↑  |CPE-<br/>EditDis↓ | SCE-<br/>BLEU↑ | SCE-<br/>EditDis↓ | HWE-<br/>BLEU↑ | HWE-<br/>EditDis↓ | 下载链接 |
+|-----------|------------|-------------------------------------------------------|:--------------:|:-----------------:|:----------:|:----------------:|:---------:|:-----------------:|:--------------:|:-----------------:|-------|
+| UniMERNet | Donut Swin | [UniMERNet.yaml](../../../../configs/rec/UniMERNet.yaml) |     0.9187     |      0.0584       |  0.9252    |      0.0596      | 0.6068 |     0.2297        |   0.9157|     0.0546           |[训练模型](https://paddleocr.bj.bcebos.com/contribution/rec_unimernet_train.tar)|
+
+其中，SPE表示简单公式，CPE表示复杂公式，SCE表示扫描捕捉公式，HWE表示手写公式。每种类型的公式示例图如下：
+![unimernet_dataset](https://github.com/user-attachments/assets/fb801a36-5614-4031-8585-700bd9f8fb2e)
+
+## 2. 环境配置
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+此外，需要安装额外的依赖：
+```shell
+sudo apt-get update
+sudo apt-get install libmagickwand-dev
+pip install -r docs/algorithm/formula_recognition/requirements.txt
+```
+
+## 3. 模型训练、评估、预测
+
+### 3.1 准备数据集
+
+从 [OpenDataLab](https://opendatalab.com/OpenDataLab/UniMER-Dataset) 上下载 UniMER-1M.zip 和 UniMER-Test.zip。
+从 [好未来平台](https://ai.100tal.com/dataset) 下载 HME100K 数据集。之后， 使用如下命令创建数据集目录，并对数据集进行转换。
+
+```shell
+# 创建 UniMERNet 数据集目录
+mkdir -p train_data/UniMERNet
+# 解压 UniMERNet 、 UniMER-Test.zip 和 HME100K.zip
+unzip -d train_data/UniMERNet path/UniMER-1M.zip
+unzip -d train_data/UniMERNet path/UniMER-Test.zip
+unzip -d train_data/UniMERNet/HME100K train_data/UniMERNet/HME100K/train.zip
+unzip -d train_data/UniMERNet/HME100K train_data/UniMERNet/HME100K/test.zip
+# 训练集转换   
+python ppocr/utils/formula_utils/unimernet_data_convert.py \
+     --image_dir=train_data/UniMERNet \
+     --datatype=unimernet_train \
+     --unimernet_txt_path=train_data/UniMERNet/UniMER-1M/train.txt \
+     --hme100k_txt_path=train_data/UniMERNet/HME100K/train_labels.txt \
+     --output_path=train_data/UniMERNet/train_unimernet_1M.txt
+# 测试集转换
+# SPE
+python ppocr/utils/formula_utils/unimernet_data_convert.py \
+     --image_dir=train_data/UniMERNet/UniMER-Test/spe \
+     --datatype=unimernet_test \
+     --unimernet_txt_path=train_data/UniMERNet/UniMER-Test/spe.txt \
+     --output_path=train_data/UniMERNet/test_unimernet_spe.txt
+# CPE
+python ppocr/utils/formula_utils/unimernet_data_convert.py \
+     --image_dir=train_data/UniMERNet/UniMER-Test/cpe \
+     --datatype=unimernet_test \
+     --unimernet_txt_path=train_data/UniMERNet/UniMER-Test/cpe.txt \
+     --output_path=train_data/UniMERNet/test_unimernet_cpe.txt
+# SCE
+python ppocr/utils/formula_utils/unimernet_data_convert.py \
+     --image_dir=train_data/UniMERNet/UniMER-Test/sce \
+     --datatype=unimernet_test \
+     --unimernet_txt_path=train_data/UniMERNet/UniMER-Test/sce.txt \
+     --output_path=train_data/UniMERNet/test_unimernet_sce.txt
+# HWE
+python ppocr/utils/formula_utils/unimernet_data_convert.py \
+     --image_dir=train_data/UniMERNet/UniMER-Test/hwe \
+     --datatype=unimernet_test \
+     --unimernet_txt_path=train_data/UniMERNet/UniMER-Test/hwe.txt \
+     --output_path=train_data/UniMERNet/test_unimernet_hwe.txt
+```
+
+
+### 3.2 下载预训练模型
+
+```shell
+# 下载 Texify 预训练模型
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/texify.pdparams
+```
+
+
+### 3.3 模型训练
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练`UniMERNet`识别模型时需要**更换配置文件**为`UniMERNet`的[配置文件](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/configs/rec/UniMERNet.yaml)。
+
+#### 启动训练
+
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+```shell
+#单卡训练 (默认训练方式)
+python3 tools/train.py -c configs/rec/UniMERNet.yaml \
+   -o Global.pretrained_model=./pretrain_models/texify.pdparams
+#多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' --ips=127.0.0.1   tools/train.py -c configs/rec/UniMERNet.yaml \
+        -o Global.pretrained_model=./pretrain_models/texify.pdparams
+```
+
+**注意：**
+
+- 默认每训练 1个epoch（37880 次iteration）进行1次评估，若您更改训练的batch_size，或更换数据集，请在训练时作出如下修改
+```
+python3  -m paddle.distributed.launch --gpus '0,1,2,3' --ips=127.0.0.1   tools/train.py -c configs/rec/UniMERNet.yaml \
+  -o Global.eval_batch_step=[0,{length_of_dataset//batch_size//4}] \
+   Global.pretrained_model=./pretrain_models/texify.pdparams
+```
+
+### 3.4 评估
+
+可下载已训练完成的[模型文件](https://paddleocr.bj.bcebos.com/contribution/rec_unimernet_train.tar)，使用如下命令进行评估：
+
+```shell
+# 注意将pretrained_model的路径设置为本地路径。若使用自行训练保存的模型，请注意修改路径和文件名为{path/to/weights}/{model_name}。
+ # SPE 测试集评估
+ python3 tools/eval.py -c configs/rec/UniMERNet.yaml -o \
+  Eval.dataset.data_dir=./train_data/UniMERNet/UniMER-Test/spe \
+  Eval.dataset.label_file_list=["./train_data/UniMERNet/test_unimernet_spe.txt"] \
+ Global.pretrained_model=./rec_unimernet_train/best_accuracy.pdparams
+ # CPE 测试集评估
+ python3 tools/eval.py -c configs/rec/UniMERNet.yaml -o \
+  Eval.dataset.data_dir=./train_data/UniMERNet/UniMER-Test/cpe \
+  Eval.dataset.label_file_list=["./train_data/UniMERNet/test_unimernet_cpe.txt"] \
+ Global.pretrained_model=./rec_unimernet_train/best_accuracy.pdparams
+ # SCE 测试集评估
+  python3 tools/eval.py -c configs/rec/UniMERNet.yaml -o \
+  Eval.dataset.data_dir=./train_data/UniMERNet/UniMER-Test/sce \
+  Eval.dataset.label_file_list=["./train_data/UniMERNet/test_unimernet_sce.txt"] \
+ Global.pretrained_model=./rec_unimernet_train/best_accuracy.pdparams
+ # HWE 测试集评估
+ python3 tools/eval.py -c configs/rec/UniMERNet.yaml -o \
+  Eval.dataset.data_dir=./train_data/UniMERNet/UniMER-Test/hwe \
+  Eval.dataset.label_file_list=["./train_data/UniMERNet/test_unimernet_hwe.txt"] \
+ Global.pretrained_model=./rec_unimernet_train/best_accuracy.pdparams
+
+```
+
+### 3.5 预测
+
+使用如下命令进行单张图片预测：
+```shell
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/infer_rec.py -c configs/rec/UniMERNet.yaml \
+  -o  Global.infer_img='./docs/datasets/images/pme_demo/0000099.png'\
+   Global.pretrained_model=./rec_unimernet_train/best_accuracy.pdparams
+# 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/datasets/pme_demo/'。
+```
+
+## 4. FAQ
+
+1. UniMERNet 数据集来自于[UniMERNet源repo](https://github.com/opendatalab/UniMERNet) 。
diff --git a/docs/version2.x/algorithm/formula_recognition/algorithm_rec_unimernet_en.md b/docs/version2.x/algorithm/formula_recognition/algorithm_rec_unimernet_en.md
new file mode 100644
index 0000000..f571f37
--- /dev/null
+++ b/docs/version2.x/algorithm/formula_recognition/algorithm_rec_unimernet_en.md
@@ -0,0 +1,136 @@
+# UniMERNet
+
+## 1. Introduction
+
+Original Project:
+> [https://github.com/opendatalab/UniMERNet](https://github.com/opendatalab/UniMERNet)
+
+
+Using UniMERNet general mathematical expression recognition datasets for training, and evaluating on its test sets, the algorithm reproduction effect is as follows:
+
+| Model           | Backbone       | config                                                  | SPE-<br/>BLEU↑ | SPE-<br/>EditDis↓ | CPE-<br/>BLEU↑  |CPE-<br/>EditDis↓ | SCE-<br/>BLEU↑ | SCE-<br/>EditDis↓ | HWE-<br/>BLEU↑ | HWE-<br/>EditDis↓ | Download link |
+|-----------|--------|---------------------------------------------------|:--------------:|:-----------------:|:----------:|:----------------:|:---------:|:-----------------:|:--------------:|:-----------------:|---|
+| UniMERNet | Donut Swin | [UniMERNet.yaml](../../../../configs/rec/UniMERNet.yaml) |     0.9187     |      0.0584       |  0.9252    |      0.0596      | 0.6068 |     0.2297        |   0.9157|     0.0546           |[trained model](https://paddleocr.bj.bcebos.com/contribution/rec_unimernet_train.tar)|
+
+SPE represents simple formulas, CPE represents complex formulas, SCE represents scanned captured formulas, and HWE represents handwritten formulas. Example images of each type of formula are shown below:
+
+![unimernet_dataset](https://github.com/user-attachments/assets/fb801a36-5614-4031-8585-700bd9f8fb2e)
+
+## 2. Environment
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md) to clone the project code.
+
+Furthermore, additional dependencies need to be installed:
+```shell
+sudo apt-get update
+sudo apt-get install libmagickwand-dev
+pip install -r docs/algorithm/formula_recognition/requirements.txt
+```
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+
+Dataset Preparation:
+
+Download UniMER-1M.zip and UniMER-Test.zip from [OpenDataLab](https://opendatalab.com/OpenDataLab/UniMER-Dataset). Download the HME100K dataset from the [TAL AI Platform](https://ai.100tal.com/dataset). After that, use the following command to create a dataset directory and convert the dataset.
+
+```shell
+# create the UniMERNet dataset directory
+mkdir -p train_data/UniMERNet
+# unzip UniMERNet 、 UniMER-Test.zip and HME100K.zip
+unzip -d train_data/UniMERNet path/UniMER-1M.zip
+unzip -d train_data/UniMERNet path/UniMER-Test.zip
+unzip -d train_data/UniMERNet/HME100K train_data/UniMERNet/HME100K/train.zip
+unzip -d train_data/UniMERNet/HME100K train_data/UniMERNet/HME100K/test.zip
+# convert the training set 
+python ppocr/utils/formula_utils/unimernet_data_convert.py \
+     --image_dir=train_data/UniMERNet \
+     --datatype=unimernet_train \
+     --unimernet_txt_path=train_data/UniMERNet/UniMER-1M/train.txt \
+     --hme100k_txt_path=train_data/UniMERNet/HME100K/train_labels.txt \
+     --output_path=train_data/UniMERNet/train_unimernet_1M.txt
+# convert the test set
+# SPE
+python ppocr/utils/formula_utils/unimernet_data_convert.py \
+     --image_dir=train_data/UniMERNet/UniMER-Test/spe \
+     --datatype=unimernet_test \
+     --unimernet_txt_path=train_data/UniMERNet/UniMER-Test/spe.txt \
+     --output_path=train_data/UniMERNet/test_unimernet_spe.txt
+# CPE
+python ppocr/utils/formula_utils/unimernet_data_convert.py \
+     --image_dir=train_data/UniMERNet/UniMER-Test/cpe \
+     --datatype=unimernet_test \
+     --unimernet_txt_path=train_data/UniMERNet/UniMER-Test/cpe.txt \
+     --output_path=train_data/UniMERNet/test_unimernet_cpe.txt
+# SCE
+python ppocr/utils/formula_utils/unimernet_data_convert.py \
+     --image_dir=train_data/UniMERNet/UniMER-Test/sce \
+     --datatype=unimernet_test \
+     --unimernet_txt_path=train_data/UniMERNet/UniMER-Test/sce.txt \
+     --output_path=train_data/UniMERNet/test_unimernet_sce.txt
+# HWE
+python ppocr/utils/formula_utils/unimernet_data_convert.py \
+     --image_dir=train_data/UniMERNet/UniMER-Test/hwe \
+     --datatype=unimernet_test \
+     --unimernet_txt_path=train_data/UniMERNet/UniMER-Test/hwe.txt \
+     --output_path=train_data/UniMERNet/test_unimernet_hwe.txt
+```
+
+Download the Pre-trained Model:
+
+```shell
+# download the Texify pre-trained model
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/texify.pdparams
+```
+
+Training:
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```shell
+#Single GPU training 
+python3 tools/train.py -c configs/rec/UniMERNet.yaml \
+   -o Global.pretrained_model=./pretrain_models/texify.pdparams
+#Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' --ips=127.0.0.1   tools/train.py -c configs/rec/UniMERNet.yaml \
+        -o Global.pretrained_model=./pretrain_models/texify.pdparams
+```
+
+Evaluation:
+
+```shell
+# GPU evaluation
+ # SPE test set evaluation
+ python3 tools/eval.py -c configs/rec/UniMERNet.yaml -o \
+  Eval.dataset.data_dir=./train_data/UniMERNet/UniMER-Test/spe \
+  Eval.dataset.label_file_list=["./train_data/UniMERNet/test_unimernet_spe.txt"] \
+ Global.pretrained_model=./rec_unimernet_train/best_accuracy.pdparams
+ # CPE test set evaluation
+ python3 tools/eval.py -c configs/rec/UniMERNet.yaml -o \
+  Eval.dataset.data_dir=./train_data/UniMERNet/UniMER-Test/cpe \
+  Eval.dataset.label_file_list=["./train_data/UniMERNet/test_unimernet_cpe.txt"] \
+ Global.pretrained_model=./rec_unimernet_train/best_accuracy.pdparams
+ # SCE test set evaluation
+  python3 tools/eval.py -c configs/rec/UniMERNet.yaml -o \
+  Eval.dataset.data_dir=./train_data/UniMERNet/UniMER-Test/sce \
+  Eval.dataset.label_file_list=["./train_data/UniMERNet/test_unimernet_sce.txt"] \
+ Global.pretrained_model=./rec_unimernet_train/best_accuracy.pdparams
+ # HWE test set evaluation
+ python3 tools/eval.py -c configs/rec/UniMERNet.yaml -o \
+  Eval.dataset.data_dir=./train_data/UniMERNet/UniMER-Test/hwe \
+  Eval.dataset.label_file_list=["./train_data/UniMERNet/test_unimernet_hwe.txt"] \
+ Global.pretrained_model=./rec_unimernet_train/best_accuracy.pdparams
+
+```
+
+Prediction:
+
+```shell
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/UniMERNet.yaml \
+  -o  Global.infer_img='./docs/datasets/images/pme_demo/0000099.png'\
+   Global.pretrained_model=./rec_unimernet_train/best_accuracy.pdparams
+```
+
+## 4. FAQ
diff --git a/docs/version2.x/algorithm/formula_recognition/images/hme_00.jpg b/docs/version2.x/algorithm/formula_recognition/images/hme_00.jpg
new file mode 100644
index 0000000..66ff27d
Binary files /dev/null and b/docs/version2.x/algorithm/formula_recognition/images/hme_00.jpg differ
diff --git a/docs/version2.x/algorithm/formula_recognition/requirements.txt b/docs/version2.x/algorithm/formula_recognition/requirements.txt
new file mode 100644
index 0000000..54809ff
--- /dev/null
+++ b/docs/version2.x/algorithm/formula_recognition/requirements.txt
@@ -0,0 +1,4 @@
+tokenizers==0.19.1
+imagesize
+ftfy
+Wand
diff --git a/docs/version2.x/algorithm/kie/algorithm_kie_layoutxlm.en.md b/docs/version2.x/algorithm/kie/algorithm_kie_layoutxlm.en.md
new file mode 100644
index 0000000..24afe92
--- /dev/null
+++ b/docs/version2.x/algorithm/kie/algorithm_kie_layoutxlm.en.md
@@ -0,0 +1,171 @@
+---
+comments: true
+---
+
+# KIE Algorithm - LayoutXLM
+
+## 1. Introduction
+
+Paper:
+
+> [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836)
+>
+> Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei
+>
+> 2021
+
+On XFUND_zh dataset, the algorithm reproduction Hmean is as follows.
+
+|Model|Backbone|Task |Cnnfig|Hmean|Download link|
+| --- | --- |--|--- | --- | --- |
+|LayoutXLM|LayoutXLM-base|SER |[ser_layoutxlm_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/ser_layoutxlm_xfund_zh.yml)|90.38%|[trained model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar)/[inference model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh_infer.tar)|
+|LayoutXLM|LayoutXLM-base|RE | [re_layoutxlm_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/re_layoutxlm_xfund_zh.yml)|74.83%|[trained model](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar)/[inference model](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh_infer.tar)|
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [KIE tutorial](../../ppocr/model_train/kie.en.md)。PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different models.
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+#### SER
+
+First, we need to export the trained model into inference model. Take LayoutXLM model trained on XFUND_zh as an example ([trained model download link](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar)). Use the following command to export.
+
+``` bash
+wget https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar
+tar -xf ser_LayoutXLM_xfun_zh.tar
+python3 tools/export_model.py -c configs/kie/layoutlm_series/ser_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./ser_LayoutXLM_xfun_zh Global.save_inference_dir=./inference/ser_layoutxlm_infer
+```
+
+Use the following command to infer using LayoutXLM SER model:
+
+```bash linenums="1"
+cd ppstructure
+python3 kie/predict_kie_token_ser.py \
+  --kie_algorithm=LayoutXLM \
+  --ser_model_dir=../inference/ser_layoutxlm_infer \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf
+```
+
+The SER visualization results are saved in the `./output` directory by default. The results are as follows.
+
+![](./images/zh_val_42_ser.jpg)
+
+#### RE
+
+First, we need to export the trained model into inference model. Take LayoutXLM model trained on XFUND_zh as an example ([trained model download link](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar)). Use the following command to export.
+
+``` bash
+wget https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar
+tar -xf re_LayoutXLM_xfun_zh.tar
+python3 tools/export_model.py -c configs/kie/layoutlm_series/re_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./re_LayoutXLM_xfun_zh Global.save_inference_dir=./inference/re_layoutxlm_infer
+```
+
+Use the following command to infer using LayoutXLM RE model:
+
+```bash linenums="1"
+cd ppstructure
+python3 kie/predict_kie_token_ser_re.py \
+  --kie_algorithm=LayoutXLM \
+  --re_model_dir=../inference/re_layoutxlm_infer \
+  --ser_model_dir=../inference/ser_layoutxlm_infer \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf
+```
+
+The RE visualization results are saved in the `./output` directory by default. The results are as follows.
+
+![](./images/zh_val_42_re.jpg)
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@article{DBLP:journals/corr/abs-2104-08836,
+  author    = {Yiheng Xu and
+               Tengchao Lv and
+               Lei Cui and
+               Guoxin Wang and
+               Yijuan Lu and
+               Dinei Flor{\^{e}}ncio and
+               Cha Zhang and
+               Furu Wei},
+  title     = {LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich
+               Document Understanding},
+  journal   = {CoRR},
+  volume    = {abs/2104.08836},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2104.08836},
+  eprinttype = {arXiv},
+  eprint    = {2104.08836},
+  timestamp = {Thu, 14 Oct 2021 09:17:23 +0200},
+  biburl    = {https://dblp.org/rec/journals/corr/abs-2104-08836.bib},
+  bibsource = {dblp computer science bibliography, https://dblp.org}
+}
+
+@article{DBLP:journals/corr/abs-1912-13318,
+  author    = {Yiheng Xu and
+               Minghao Li and
+               Lei Cui and
+               Shaohan Huang and
+               Furu Wei and
+               Ming Zhou},
+  title     = {LayoutLM: Pre-training of Text and Layout for Document Image Understanding},
+  journal   = {CoRR},
+  volume    = {abs/1912.13318},
+  year      = {2019},
+  url       = {http://arxiv.org/abs/1912.13318},
+  eprinttype = {arXiv},
+  eprint    = {1912.13318},
+  timestamp = {Mon, 01 Jun 2020 16:20:46 +0200},
+  biburl    = {https://dblp.org/rec/journals/corr/abs-1912-13318.bib},
+  bibsource = {dblp computer science bibliography, https://dblp.org}
+}
+
+@article{DBLP:journals/corr/abs-2012-14740,
+  author    = {Yang Xu and
+               Yiheng Xu and
+               Tengchao Lv and
+               Lei Cui and
+               Furu Wei and
+               Guoxin Wang and
+               Yijuan Lu and
+               Dinei A. F. Flor{\^{e}}ncio and
+               Cha Zhang and
+               Wanxiang Che and
+               Min Zhang and
+               Lidong Zhou},
+  title     = {LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding},
+  journal   = {CoRR},
+  volume    = {abs/2012.14740},
+  year      = {2020},
+  url       = {https://arxiv.org/abs/2012.14740},
+  eprinttype = {arXiv},
+  eprint    = {2012.14740},
+  timestamp = {Tue, 27 Jul 2021 09:53:52 +0200},
+  biburl    = {https://dblp.org/rec/journals/corr/abs-2012-14740.bib},
+  bibsource = {dblp computer science bibliography, https://dblp.org}
+}
+```
diff --git a/docs/version2.x/algorithm/kie/algorithm_kie_layoutxlm.md b/docs/version2.x/algorithm/kie/algorithm_kie_layoutxlm.md
new file mode 100644
index 0000000..1485ff9
--- /dev/null
+++ b/docs/version2.x/algorithm/kie/algorithm_kie_layoutxlm.md
@@ -0,0 +1,171 @@
+---
+comments: true
+---
+
+# 关键信息抽取算法-LayoutXLM
+
+## 1. 算法简介
+
+论文信息：
+
+> [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836)
+>
+> Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei
+>
+> 2021
+
+在XFUND_zh数据集上，算法复现效果如下：
+
+|模型|骨干网络|任务|配置文件|hmean|下载链接|
+| --- | --- |--|--- | --- | --- |
+|LayoutXLM|LayoutXLM-base|SER |[ser_layoutxlm_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/ser_layoutxlm_xfund_zh.yml)|90.38%|[训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar)/[推理模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh_infer.tar)|
+|LayoutXLM|LayoutXLM-base|RE | [re_layoutxlm_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/re_layoutxlm_xfund_zh.yml)|74.83%|[训练模型](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar)/[推理模型](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh_infer.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+请参考[关键信息抽取教程](../../ppocr/model_train/kie.md)。PaddleOCR对代码进行了模块化，训练不同的关键信息抽取模型只需要**更换配置文件**即可。
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+#### SER
+
+首先将训练得到的模型转换成inference model。LayoutXLM模型在XFUND_zh数据集上训练的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar)），可以使用下面的命令进行转换。
+
+``` bash
+wget https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar
+tar -xf ser_LayoutXLM_xfun_zh.tar
+python3 tools/export_model.py -c configs/kie/layoutlm_series/ser_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./ser_LayoutXLM_xfun_zh Global.save_inference_dir=./inference/ser_layoutxlm_infer
+```
+
+LayoutXLM模型基于SER任务进行推理，可以执行如下命令：
+
+```bash linenums="1"
+cd ppstructure
+python3 kie/predict_kie_token_ser.py \
+  --kie_algorithm=LayoutXLM \
+  --ser_model_dir=../inference/ser_layoutxlm_infer \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf
+```
+
+SER可视化结果默认保存到`./output`文件夹里面，结果示例如下：
+
+![](./images/zh_val_42_ser.jpg)
+
+#### RE
+
+首先将训练得到的模型转换成inference model。LayoutXLM模型在XFUND_zh数据集上训练的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar)），可以使用下面的命令进行转换。
+
+``` bash
+wget https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar
+tar -xf re_LayoutXLM_xfun_zh.tar
+python3 tools/export_model.py -c configs/kie/layoutlm_series/re_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./re_LayoutXLM_xfun_zh Global.save_inference_dir=./inference/ser_layoutxlm_infer
+```
+
+LayoutXLM模型基于RE任务进行推理，可以执行如下命令：
+
+```bash linenums="1"
+cd ppstructure
+python3 kie/predict_kie_token_ser_re.py \
+  --kie_algorithm=LayoutXLM \
+  --re_model_dir=../inference/re_layoutxlm_infer \
+  --ser_model_dir=../inference/ser_layoutxlm_infer \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf
+```
+
+RE可视化结果默认保存到`./output`文件夹里面，结果示例如下：
+
+![](./images/zh_val_42_re.jpg)
+
+### 4.2 C++推理部署
+
+暂不支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@article{DBLP:journals/corr/abs-2104-08836,
+  author    = {Yiheng Xu and
+               Tengchao Lv and
+               Lei Cui and
+               Guoxin Wang and
+               Yijuan Lu and
+               Dinei Flor{\^{e}}ncio and
+               Cha Zhang and
+               Furu Wei},
+  title     = {LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich
+               Document Understanding},
+  journal   = {CoRR},
+  volume    = {abs/2104.08836},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2104.08836},
+  eprinttype = {arXiv},
+  eprint    = {2104.08836},
+  timestamp = {Thu, 14 Oct 2021 09:17:23 +0200},
+  biburl    = {https://dblp.org/rec/journals/corr/abs-2104-08836.bib},
+  bibsource = {dblp computer science bibliography, https://dblp.org}
+}
+
+@article{DBLP:journals/corr/abs-1912-13318,
+  author    = {Yiheng Xu and
+               Minghao Li and
+               Lei Cui and
+               Shaohan Huang and
+               Furu Wei and
+               Ming Zhou},
+  title     = {LayoutLM: Pre-training of Text and Layout for Document Image Understanding},
+  journal   = {CoRR},
+  volume    = {abs/1912.13318},
+  year      = {2019},
+  url       = {http://arxiv.org/abs/1912.13318},
+  eprinttype = {arXiv},
+  eprint    = {1912.13318},
+  timestamp = {Mon, 01 Jun 2020 16:20:46 +0200},
+  biburl    = {https://dblp.org/rec/journals/corr/abs-1912-13318.bib},
+  bibsource = {dblp computer science bibliography, https://dblp.org}
+}
+
+@article{DBLP:journals/corr/abs-2012-14740,
+  author    = {Yang Xu and
+               Yiheng Xu and
+               Tengchao Lv and
+               Lei Cui and
+               Furu Wei and
+               Guoxin Wang and
+               Yijuan Lu and
+               Dinei A. F. Flor{\^{e}}ncio and
+               Cha Zhang and
+               Wanxiang Che and
+               Min Zhang and
+               Lidong Zhou},
+  title     = {LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding},
+  journal   = {CoRR},
+  volume    = {abs/2012.14740},
+  year      = {2020},
+  url       = {https://arxiv.org/abs/2012.14740},
+  eprinttype = {arXiv},
+  eprint    = {2012.14740},
+  timestamp = {Tue, 27 Jul 2021 09:53:52 +0200},
+  biburl    = {https://dblp.org/rec/journals/corr/abs-2012-14740.bib},
+  bibsource = {dblp computer science bibliography, https://dblp.org}
+}
+```
diff --git a/docs/version2.x/algorithm/kie/algorithm_kie_sdmgr.en.md b/docs/version2.x/algorithm/kie/algorithm_kie_sdmgr.en.md
new file mode 100644
index 0000000..4981739
--- /dev/null
+++ b/docs/version2.x/algorithm/kie/algorithm_kie_sdmgr.en.md
@@ -0,0 +1,113 @@
+---
+comments: true
+---
+
+# KIE Algorithm - SDMGR
+
+## 1. Introduction
+
+Paper:
+
+> [Spatial Dual-Modality Graph Reasoning for Key Information Extraction](https://arxiv.org/abs/2103.14470)
+>
+> Hongbin Sun and Zhanghui Kuang and Xiaoyu Yue and Chenhao Lin and Wayne Zhang
+>
+> 2021
+
+On wildreceipt dataset, the algorithm reproduction Hmean is as follows.
+
+|Model|Backbone |Cnnfig|Hmean|Download link|
+| --- | --- | --- | --- | --- |
+|SDMGR|VGG6|[configs/kie/sdmgr/kie_unet_sdmgr.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/sdmgr/kie_unet_sdmgr.yml)|86.70%|[trained model]( https://paddleocr.bj.bcebos.com/dygraph_v2.1/kie/kie_vgg16.tar)/[inference model(coming soon)]()|
+
+## 2. 环境配置
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+SDMGR is a key information extraction algorithm that classifies each detected textline into predefined categories, such as order ID, invoice number, amount, etc.
+
+The training and test data are collected in the wildreceipt dataset, use following command to downloaded the dataset.
+
+```bash linenums="1"
+wget https://paddleocr.bj.bcebos.com/ppstructure/dataset/wildreceipt.tar && tar xf wildreceipt.tar
+```
+
+Create dataset soft link to `PaddleOCR/train_data` directory.
+
+```bash linenums="1"
+cd PaddleOCR/ && mkdir train_data && cd train_data
+ln -s ../../wildreceipt ./
+```
+
+### 3.1 Model training
+
+The config file is `configs/kie/sdmgr/kie_unet_sdmgr.yml`， the default dataset path is `train_data/wildreceipt`.
+
+Use the following command to train the model.
+
+```bash linenums="1"
+python3 tools/train.py -c configs/kie/sdmgr/kie_unet_sdmgr.yml -o Global.save_model_dir=./output/kie/
+```
+
+### 3.2 Model evaluation
+
+Use the following command to evaluate the model:
+
+```bash linenums="1"
+python3 tools/eval.py -c configs/kie/sdmgr/kie_unet_sdmgr.yml -o Global.checkpoints=./output/kie/best_accuracy
+```
+
+An example of output information is shown below.
+
+```bash linenums="1"
+[2022/08/10 05:22:23] ppocr INFO: metric eval ***************
+[2022/08/10 05:22:23] ppocr INFO: hmean:0.8670120239257812
+[2022/08/10 05:22:23] ppocr INFO: fps:10.18816520530961
+```
+
+### 3.3 Model prediction
+
+Use the following command to load the model and predict. During the prediction, the text file storing the image path and OCR information needs to be loaded in advance. Use `Global.infer_img` to assign.
+
+```bash linenums="1"
+python3 tools/infer_kie.py -c configs/kie/kie_unet_sdmgr.yml -o Global.checkpoints=kie_vgg16/best_accuracy  Global.infer_img=./train_data/wildreceipt/1.txt
+```
+
+The visualization results and texts are saved in the `./output/sdmgr_kie/` directory by default. The results are as follows.
+
+![img](./images/sdmgr_result.png)
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+Not supported
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@misc{sun2021spatial,
+      title={Spatial Dual-Modality Graph Reasoning for Key Information Extraction},
+      author={Hongbin Sun and Zhanghui Kuang and Xiaoyu Yue and Chenhao Lin and Wayne Zhang},
+      year={2021},
+      eprint={2103.14470},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV}
+}
+```
diff --git a/docs/version2.x/algorithm/kie/algorithm_kie_sdmgr.md b/docs/version2.x/algorithm/kie/algorithm_kie_sdmgr.md
new file mode 100644
index 0000000..417072f
--- /dev/null
+++ b/docs/version2.x/algorithm/kie/algorithm_kie_sdmgr.md
@@ -0,0 +1,114 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# 关键信息抽取算法-SDMGR
+
+## 1. 算法简介
+
+论文信息：
+
+> [Spatial Dual-Modality Graph Reasoning for Key Information Extraction](https://arxiv.org/abs/2103.14470)
+>
+> Hongbin Sun and Zhanghui Kuang and Xiaoyu Yue and Chenhao Lin and Wayne Zhang
+>
+> 2021
+
+在wildreceipt发票公开数据集上，算法复现效果如下：
+
+|模型|骨干网络|配置文件|hmean|下载链接|
+| --- | --- | --- | --- | --- |
+|SDMGR|VGG6|[configs/kie/sdmgr/kie_unet_sdmgr.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/sdmgr/kie_unet_sdmgr.yml)|86.70%|[训练模型]( https://paddleocr.bj.bcebos.com/dygraph_v2.1/kie/kie_vgg16.tar)/[推理模型(coming soon)]()|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+SDMGR是一个关键信息提取算法，将每个检测到的文本区域分类为预定义的类别，如订单ID、发票号码，金额等。
+
+训练和测试的数据采用wildreceipt数据集，通过如下指令下载数据集：
+
+```bash linenums="1"
+wget https://paddleocr.bj.bcebos.com/ppstructure/dataset/wildreceipt.tar && tar xf wildreceipt.tar
+```
+
+创建数据集软链到PaddleOCR/train_data目录下：
+
+```bash linenums="1"
+cd PaddleOCR/ && mkdir train_data && cd train_data
+ln -s ../../wildreceipt ./
+```
+
+### 3.1 模型训练
+
+训练采用的配置文件是`configs/kie/sdmgr/kie_unet_sdmgr.yml`，配置文件中默认训练数据路径是`train_data/wildreceipt`，准备好数据后，可以通过如下指令执行训练：
+
+```bash linenums="1"
+python3 tools/train.py -c configs/kie/sdmgr/kie_unet_sdmgr.yml -o Global.save_model_dir=./output/kie/
+```
+
+### 3.2 模型评估
+
+执行下面的命令进行模型评估
+
+```bash linenums="1"
+python3 tools/eval.py -c configs/kie/sdmgr/kie_unet_sdmgr.yml -o Global.checkpoints=./output/kie/best_accuracy
+```
+
+输出信息示例如下所示:
+
+```bash linenums="1"
+[2022/08/10 05:22:23] ppocr INFO: metric eval ***************
+[2022/08/10 05:22:23] ppocr INFO: hmean:0.8670120239257812
+[2022/08/10 05:22:23] ppocr INFO: fps:10.18816520530961
+```
+
+### 3.3 模型预测
+
+执行下面的命令进行模型预测，预测的时候需要预先加载存储图片路径以及OCR信息的文本文件，使用`Global.infer_img`进行指定。
+
+```bash linenums="1"
+python3 tools/infer_kie.py -c configs/kie/kie_unet_sdmgr.yml -o Global.checkpoints=kie_vgg16/best_accuracy  Global.infer_img=./train_data/wildreceipt/1.txt
+```
+
+执行预测后的结果保存在`./output/sdmgr_kie/predicts_kie.txt`文件中，可视化结果保存在`/output/sdmgr_kie/kie_results/`目录下。
+
+可视化结果如下图所示：
+
+![img](./images/sdmgr_result.png)
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+暂不支持
+
+### 4.2 C++推理部署
+
+暂不支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@misc{sun2021spatial,
+      title={Spatial Dual-Modality Graph Reasoning for Key Information Extraction},
+      author={Hongbin Sun and Zhanghui Kuang and Xiaoyu Yue and Chenhao Lin and Wayne Zhang},
+      year={2021},
+      eprint={2103.14470},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV}
+}
+```
diff --git a/docs/version2.x/algorithm/kie/algorithm_kie_vi_layoutxlm.en.md b/docs/version2.x/algorithm/kie/algorithm_kie_vi_layoutxlm.en.md
new file mode 100644
index 0000000..969206c
--- /dev/null
+++ b/docs/version2.x/algorithm/kie/algorithm_kie_vi_layoutxlm.en.md
@@ -0,0 +1,168 @@
+---
+comments: true
+---
+
+# KIE Algorithm - VI-LayoutXLM
+
+## 1. Introduction
+
+VI-LayoutXLM is improved based on LayoutXLM. In the process of downstream finetuning, the visual backbone network module is removed, and the model infernce speed is further improved on the basis of almost lossless accuracy.
+
+On XFUND_zh dataset, the algorithm reproduction Hmean is as follows.
+
+|Model|Backbone|Task |Config|Hmean|Download link|
+| --- | --- |---| --- | --- | --- |
+|VI-LayoutXLM |VI-LayoutXLM-base | SER |[ser_vi_layoutxlm_xfund_zh_udml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh_udml.yml)|93.19%|[trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar)/[inference model](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_infer.tar)|
+|VI-LayoutXLM |VI-LayoutXLM-base |RE | [re_vi_layoutxlm_xfund_zh_udml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh_udml.yml)|83.92%|[trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar)/[inference model](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_infer.tar)|
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [KIE tutorial](../../ppocr/model_train/kie.en.md). PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different models.
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+#### SER
+
+First, we need to export the trained model into inference model. Take VI-LayoutXLM model trained on XFUND_zh as an example ([trained model download link](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar)). Use the following command to export.
+
+``` bash
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar
+tar -xf ser_vi_layoutxlm_xfund_pretrained.tar
+python3 tools/export_model.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./ser_vi_layoutxlm_xfund_pretrained/best_accuracy Global.save_inference_dir=./inference/ser_vi_layoutxlm_infer
+```
+
+Use the following command to infer using VI-LayoutXLM SER model.
+
+```bash linenums="1"
+cd ppstructure
+python3 kie/predict_kie_token_ser.py \
+  --kie_algorithm=LayoutXLM \
+  --ser_model_dir=../inference/ser_vi_layoutxlm_infer \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx"
+```
+
+The SER visualization results are saved in the `./output` folder by default. The results are as follows.
+
+![](./images/zh_val_42_ser.jpg)
+
+#### RE
+
+First, we need to export the trained model into inference model. Take VI-LayoutXLM model trained on XFUND_zh as an example ([trained model download link](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar)). Use the following command to export.
+
+``` bash
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar
+tar -xf re_vi_layoutxlm_xfund_pretrained.tar
+python3 tools/export_model.py -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./re_vi_layoutxlm_xfund_pretrained/best_accuracy Global.save_inference_dir=./inference/re_vi_layoutxlm_infer
+```
+
+Use the following command to infer using VI-LayoutXLM RE model.
+
+```bash linenums="1"
+cd ppstructure
+python3 kie/predict_kie_token_ser_re.py \
+  --kie_algorithm=LayoutXLM \
+  --re_model_dir=../inference/re_vi_layoutxlm_infer \
+  --ser_model_dir=../inference/ser_vi_layoutxlm_infer \
+  --use_visual_backbone=False \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx"
+```
+
+The RE visualization results are saved in the `./output` folder by default. The results are as follows.
+
+![](./images/zh_val_42_re.jpg)
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@article{DBLP:journals/corr/abs-2104-08836,
+  author    = {Yiheng Xu and
+               Tengchao Lv and
+               Lei Cui and
+               Guoxin Wang and
+               Yijuan Lu and
+               Dinei Flor{\^{e}}ncio and
+               Cha Zhang and
+               Furu Wei},
+  title     = {LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich
+               Document Understanding},
+  journal   = {CoRR},
+  volume    = {abs/2104.08836},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2104.08836},
+  eprinttype = {arXiv},
+  eprint    = {2104.08836},
+  timestamp = {Thu, 14 Oct 2021 09:17:23 +0200},
+  biburl    = {https://dblp.org/rec/journals/corr/abs-2104-08836.bib},
+  bibsource = {dblp computer science bibliography, https://dblp.org}
+}
+
+@article{DBLP:journals/corr/abs-1912-13318,
+  author    = {Yiheng Xu and
+               Minghao Li and
+               Lei Cui and
+               Shaohan Huang and
+               Furu Wei and
+               Ming Zhou},
+  title     = {LayoutLM: Pre-training of Text and Layout for Document Image Understanding},
+  journal   = {CoRR},
+  volume    = {abs/1912.13318},
+  year      = {2019},
+  url       = {http://arxiv.org/abs/1912.13318},
+  eprinttype = {arXiv},
+  eprint    = {1912.13318},
+  timestamp = {Mon, 01 Jun 2020 16:20:46 +0200},
+  biburl    = {https://dblp.org/rec/journals/corr/abs-1912-13318.bib},
+  bibsource = {dblp computer science bibliography, https://dblp.org}
+}
+
+@article{DBLP:journals/corr/abs-2012-14740,
+  author    = {Yang Xu and
+               Yiheng Xu and
+               Tengchao Lv and
+               Lei Cui and
+               Furu Wei and
+               Guoxin Wang and
+               Yijuan Lu and
+               Dinei A. F. Flor{\^{e}}ncio and
+               Cha Zhang and
+               Wanxiang Che and
+               Min Zhang and
+               Lidong Zhou},
+  title     = {LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding},
+  journal   = {CoRR},
+  volume    = {abs/2012.14740},
+  year      = {2020},
+  url       = {https://arxiv.org/abs/2012.14740},
+  eprinttype = {arXiv},
+  eprint    = {2012.14740},
+  timestamp = {Tue, 27 Jul 2021 09:53:52 +0200},
+  biburl    = {https://dblp.org/rec/journals/corr/abs-2012-14740.bib},
+  bibsource = {dblp computer science bibliography, https://dblp.org}
+}
+```
diff --git a/docs/version2.x/algorithm/kie/algorithm_kie_vi_layoutxlm.md b/docs/version2.x/algorithm/kie/algorithm_kie_vi_layoutxlm.md
new file mode 100644
index 0000000..5d37b55
--- /dev/null
+++ b/docs/version2.x/algorithm/kie/algorithm_kie_vi_layoutxlm.md
@@ -0,0 +1,169 @@
+---
+comments: true
+---
+
+
+# 关键信息抽取算法-VI-LayoutXLM
+
+## 1. 算法简介
+
+VI-LayoutXLM基于LayoutXLM进行改进，在下游任务训练过程中，去除视觉骨干网络模块，最终精度基本无损的情况下，模型推理速度进一步提升。
+
+在XFUND_zh数据集上，算法复现效果如下：
+
+|模型|骨干网络|任务|配置文件|hmean|下载链接|
+| --- | --- |---| --- | --- | --- |
+|VI-LayoutXLM |VI-LayoutXLM-base | SER |[ser_vi_layoutxlm_xfund_zh_udml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh_udml.yml)|93.19%|[训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar)/[推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_infer.tar)|
+|VI-LayoutXLM |VI-LayoutXLM-base |RE | [re_vi_layoutxlm_xfund_zh_udml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh_udml.yml)|83.92%|[训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar)/[推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_infer.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+请参考[关键信息抽取教程](../../ppocr/model_train/kie.md)。PaddleOCR对代码进行了模块化，训练不同的关键信息抽取模型只需要**更换配置文件**即可。
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+#### SER
+
+首先将训练得到的模型转换成inference model。以VI-LayoutXLM模型在XFUND_zh数据集上训练的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar)），可以使用下面的命令进行转换。
+
+``` bash
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar
+tar -xf ser_vi_layoutxlm_xfund_pretrained.tar
+python3 tools/export_model.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./ser_vi_layoutxlm_xfund_pretrained/best_accuracy Global.save_inference_dir=./inference/ser_vi_layoutxlm_infer
+```
+
+VI-LayoutXLM模型基于SER任务进行推理，可以执行如下命令：
+
+```bash linenums="1"
+cd ppstructure
+python3 kie/predict_kie_token_ser.py \
+  --kie_algorithm=LayoutXLM \
+  --ser_model_dir=../inference/ser_vi_layoutxlm_infer \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx"
+```
+
+SER可视化结果默认保存到`./output`文件夹里面，结果示例如下：
+
+![](./images/zh_val_42_ser.jpg)
+
+#### RE
+
+首先将训练得到的模型转换成inference model。以VI-LayoutXLM模型在XFUND_zh数据集上训练的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar)），可以使用下面的命令进行转换。
+
+``` bash
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar
+tar -xf re_vi_layoutxlm_xfund_pretrained.tar
+python3 tools/export_model.py -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./re_vi_layoutxlm_xfund_pretrained/best_accuracy Global.save_inference_dir=./inference/re_vi_layoutxlm_infer
+```
+
+VI-LayoutXLM模型基于RE任务进行推理，可以执行如下命令：
+
+```bash linenums="1"
+cd ppstructure
+python3 kie/predict_kie_token_ser_re.py \
+  --kie_algorithm=LayoutXLM \
+  --re_model_dir=../inference/re_vi_layoutxlm_infer \
+  --ser_model_dir=../inference/ser_vi_layoutxlm_infer \
+  --use_visual_backbone=False \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx"
+```
+
+RE可视化结果默认保存到`./output`文件夹里面，结果示例如下：
+
+![](images/zh_val_42_re.jpg)
+
+### 4.2 C++推理部署
+
+暂不支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@article{DBLP:journals/corr/abs-2104-08836,
+  author    = {Yiheng Xu and
+               Tengchao Lv and
+               Lei Cui and
+               Guoxin Wang and
+               Yijuan Lu and
+               Dinei Flor{\^{e}}ncio and
+               Cha Zhang and
+               Furu Wei},
+  title     = {LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich
+               Document Understanding},
+  journal   = {CoRR},
+  volume    = {abs/2104.08836},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2104.08836},
+  eprinttype = {arXiv},
+  eprint    = {2104.08836},
+  timestamp = {Thu, 14 Oct 2021 09:17:23 +0200},
+  biburl    = {https://dblp.org/rec/journals/corr/abs-2104-08836.bib},
+  bibsource = {dblp computer science bibliography, https://dblp.org}
+}
+
+@article{DBLP:journals/corr/abs-1912-13318,
+  author    = {Yiheng Xu and
+               Minghao Li and
+               Lei Cui and
+               Shaohan Huang and
+               Furu Wei and
+               Ming Zhou},
+  title     = {LayoutLM: Pre-training of Text and Layout for Document Image Understanding},
+  journal   = {CoRR},
+  volume    = {abs/1912.13318},
+  year      = {2019},
+  url       = {http://arxiv.org/abs/1912.13318},
+  eprinttype = {arXiv},
+  eprint    = {1912.13318},
+  timestamp = {Mon, 01 Jun 2020 16:20:46 +0200},
+  biburl    = {https://dblp.org/rec/journals/corr/abs-1912-13318.bib},
+  bibsource = {dblp computer science bibliography, https://dblp.org}
+}
+
+@article{DBLP:journals/corr/abs-2012-14740,
+  author    = {Yang Xu and
+               Yiheng Xu and
+               Tengchao Lv and
+               Lei Cui and
+               Furu Wei and
+               Guoxin Wang and
+               Yijuan Lu and
+               Dinei A. F. Flor{\^{e}}ncio and
+               Cha Zhang and
+               Wanxiang Che and
+               Min Zhang and
+               Lidong Zhou},
+  title     = {LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding},
+  journal   = {CoRR},
+  volume    = {abs/2012.14740},
+  year      = {2020},
+  url       = {https://arxiv.org/abs/2012.14740},
+  eprinttype = {arXiv},
+  eprint    = {2012.14740},
+  timestamp = {Tue, 27 Jul 2021 09:53:52 +0200},
+  biburl    = {https://dblp.org/rec/journals/corr/abs-2012-14740.bib},
+  bibsource = {dblp computer science bibliography, https://dblp.org}
+}
+```
diff --git a/docs/version2.x/algorithm/kie/images/sdmgr_result.png b/docs/version2.x/algorithm/kie/images/sdmgr_result.png
new file mode 100644
index 0000000..6fa4fe8
Binary files /dev/null and b/docs/version2.x/algorithm/kie/images/sdmgr_result.png differ
diff --git a/docs/version2.x/algorithm/kie/images/zh_val_42_re.jpg b/docs/version2.x/algorithm/kie/images/zh_val_42_re.jpg
new file mode 100644
index 0000000..49a0fad
Binary files /dev/null and b/docs/version2.x/algorithm/kie/images/zh_val_42_re.jpg differ
diff --git a/docs/version2.x/algorithm/kie/images/zh_val_42_ser.jpg b/docs/version2.x/algorithm/kie/images/zh_val_42_ser.jpg
new file mode 100644
index 0000000..d69d835
Binary files /dev/null and b/docs/version2.x/algorithm/kie/images/zh_val_42_ser.jpg differ
diff --git a/docs/version2.x/algorithm/overview.en.md b/docs/version2.x/algorithm/overview.en.md
new file mode 100755
index 0000000..deddc3d
--- /dev/null
+++ b/docs/version2.x/algorithm/overview.en.md
@@ -0,0 +1,179 @@
+---
+comments: true
+---
+
+# Algorithms
+
+This tutorial lists the OCR algorithms supported by PaddleOCR, as well as the models and metrics of each algorithm on **English public datasets**. It is mainly used for algorithm introduction and algorithm performance comparison. For more models on other datasets including Chinese, please refer to [PP-OCRv3 models list](../ppocr/model_list.en.md).
+
+>
+Developers are welcome to contribute more algorithms! Please refer to [add new algorithm](./add_new_algorithm.en.md) guideline.
+
+## 1. Two-stage OCR Algorithms
+
+### 1.1 Text Detection Algorithms
+
+Supported text detection algorithms (Click the link to get the tutorial):
+
+- [x]  [DB && DB++](./text_detection/algorithm_det_db.en.md)
+- [x]  [EAST](./text_detection/algorithm_det_east.en.md)
+- [x]  [SAST](./text_detection/algorithm_det_sast.en.md)
+- [x]  [PSENet](./text_detection/algorithm_det_psenet.en.md)
+- [x]  [FCENet](./text_detection/algorithm_det_fcenet.en.md)
+- [x]  [DRRG](./text_detection/algorithm_det_drrg.en.md)
+- [x]  [CT](./text_detection/algorithm_det_ct.en.md)
+
+On the ICDAR2015 dataset, the text detection result is as follows:
+
+|Model|Backbone|Precision|Recall|Hmean|Download link|
+| --- | --- | --- | --- | --- | --- |
+|EAST|ResNet50_vd|88.71%|81.36%|84.88%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar)|
+|EAST|MobileNetV3|78.20%|79.10%|78.65%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_east_v2.0_train.tar)|
+|DB|ResNet50_vd|86.41%|78.72%|82.38%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)|
+|DB|MobileNetV3|77.29%|73.08%|75.12%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)|
+|SAST|ResNet50_vd|91.39%|83.77%|87.42%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar)|
+|PSE|ResNet50_vd|85.81%|79.53%|82.55%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_vd_pse_v2.0_train.tar)|
+|PSE|MobileNetV3|82.20%|70.48%|75.89%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_mv3_pse_v2.0_train.tar)|
+|DB++|ResNet50|90.89%|82.66%|86.58%|[pretrained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/ResNet50_dcn_asf_synthtext_pretrained.pdparams)/[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_db%2B%2B_icdar15_train.tar)|
+
+On Total-Text dataset, the text detection result is as follows:
+
+|Model|Backbone|Precision|Recall|Hmean|Download link|
+| --- | --- | --- | --- | --- | --- |
+|SAST|ResNet50_vd|89.63%|78.44%|83.66%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_totaltext_v2.0_train.tar)|
+|CT|ResNet18_vd|88.68%|81.70%|85.05%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r18_ct_train.tar)|
+
+On CTW1500 dataset, the text detection result is as follows:
+
+|Model|Backbone|Precision|Recall|Hmean| Download link|
+| --- | --- | --- | --- | --- |---|
+|FCE|ResNet50_dcn|88.39%|82.18%|85.27%| [trained model](https://paddleocr.bj.bcebos.com/contribution/det_r50_dcn_fce_ctw_v2.0_train.tar) |
+|DRRG|ResNet50_vd|89.92%|80.91%|85.18%|[trained model](https://paddleocr.bj.bcebos.com/contribution/det_r50_drrg_ctw_train.tar)|
+
+**Note：** Additional data, like icdar2013, icdar2017, COCO-Text, ArT, was added to the model training of SAST. Download English public dataset in organized format used by PaddleOCR from:
+
+- [Baidu Drive](https://pan.baidu.com/s/12cPnZcVuV1zn5DOd4mqjVw) (download code: 2bpi).
+- [Google Drive](https://drive.google.com/drive/folders/1ll2-XEVyCQLpJjawLDiRlvo_i4BqHCJe?usp=sharing)
+
+### 1.2 Text Recognition Algorithms
+
+Supported text recognition algorithms (Click the link to get the tutorial):
+
+- [x]  [CRNN](./text_recognition/algorithm_rec_crnn.en.md)
+- [x]  [Rosetta](./text_recognition/algorithm_rec_rosetta.en.md)
+- [x]  [STAR-Net](./text_recognition/algorithm_rec_starnet.en.md)
+- [x]  [RARE](./text_recognition/algorithm_rec_rare.en.md)
+- [x]  [SRN](./text_recognition/algorithm_rec_srn.en.md)
+- [x]  [NRTR](./text_recognition/algorithm_rec_nrtr.en.md)
+- [x]  [SAR](./text_recognition/algorithm_rec_sar.en.md)
+- [x]  [SEED](./text_recognition/algorithm_rec_seed.en.md)
+- [x]  [SVTR](./text_recognition/algorithm_rec_svtr.en.md)
+- [x]  [ViTSTR](./text_recognition/algorithm_rec_vitstr.en.md)
+- [x]  [ABINet](./text_recognition/algorithm_rec_abinet.en.md)
+- [x]  [VisionLAN](./text_recognition/algorithm_rec_visionlan.en.md)
+- [x]  [SPIN](./text_recognition/algorithm_rec_spin.en.md)
+- [x]  [RobustScanner](./text_recognition/algorithm_rec_robustscanner.en.md)
+- [x]  [RFL](./text_recognition/algorithm_rec_rfl.en.md)
+- [x]  [ParseQ](./text_recognition/algorithm_rec_parseq.md)
+- [x]  [CPPD](./text_recognition/algorithm_rec_cppd.en.md)
+- [x]  [SATRN](./text_recognition/algorithm_rec_satrn.en.md)
+
+Refer to [DTRB](https://arxiv.org/abs/1904.01906), the training and evaluation result of these above text recognition (using MJSynth and SynthText for training, evaluate on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE) is as follow:
+
+|Model|Backbone|Avg Accuracy|Module combination|Download link|
+|---|---|---|---|---|
+|Rosetta|Resnet34_vd|79.11%|rec_r34_vd_none_none_ctc|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_none_ctc_v2.0_train.tar)|
+|Rosetta|MobileNetV3|75.80%|rec_mv3_none_none_ctc|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_none_ctc_v2.0_train.tar)|
+|CRNN|Resnet34_vd|81.04%|rec_r34_vd_none_bilstm_ctc|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_bilstm_ctc_v2.0_train.tar)|
+|CRNN|MobileNetV3|77.95%|rec_mv3_none_bilstm_ctc|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_bilstm_ctc_v2.0_train.tar)|
+|StarNet|Resnet34_vd|82.85%|rec_r34_vd_tps_bilstm_ctc|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_ctc_v2.0_train.tar)|
+|StarNet|MobileNetV3|79.28%|rec_mv3_tps_bilstm_ctc|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_tps_bilstm_ctc_v2.0_train.tar)|
+|RARE|Resnet34_vd|83.98%|rec_r34_vd_tps_bilstm_att |[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_att_v2.0_train.tar)|
+|RARE|MobileNetV3|81.76%|rec_mv3_tps_bilstm_att |[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_tps_bilstm_att_v2.0_train.tar)|
+|SRN|Resnet50_vd_fpn| 86.31% | rec_r50fpn_vd_none_srn |[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r50_vd_srn_train.tar)|
+|NRTR|NRTR_MTB| 84.21% | rec_mtb_nrtr | [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mtb_nrtr_train.tar) |
+|SAR|Resnet31| 87.20% | rec_r31_sar | [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_sar_train.tar) |
+|SEED|Aster_Resnet| 85.35% | rec_resnet_stn_bilstm_att | [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_resnet_stn_bilstm_att.tar) |
+|SVTR|SVTR-Tiny| 89.25% | rec_svtr_tiny_none_ctc_en | [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar) |
+|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ce | [trained model](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar) |
+|ABINet|Resnet45| 90.75% | rec_r45_abinet | [trained model](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar) |
+|VisionLAN|Resnet45| 90.30% | rec_r45_visionlan | [trained model](https://paddleocr.bj.bcebos.com/VisionLAN/rec_r45_visionlan_train.tar) |
+|SPIN|ResNet32| 90.00% | rec_r32_gaspin_bilstm_att | [trained model](https://paddleocr.bj.bcebos.com/contribution/rec_r32_gaspin_bilstm_att.tar) |
+|RobustScanner|ResNet31| 87.77% | rec_r31_robustscanner | [trained model](https://paddleocr.bj.bcebos.com/contribution/rec_r31_robustscanner.tar)|
+|RFL|ResNetRFL| 88.63% | rec_resnet_rfl_att | [trained model](https://paddleocr.bj.bcebos.com/contribution/rec_resnet_rfl_att_train.tar) |
+|ParseQ|VIT| 91.24% | rec_vit_parseq_synth | [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/parseq/rec_vit_parseq_synth.tgz) |
+|CPPD|SVTR-Base| 93.8% | rec_svtrnet_cppd_base_en | [trained model](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_en_train.tar) |
+|SATRN|ShallowCNN| 88.05% | rec_satrn | [trained model](https://pan.baidu.com/s/10J-Bsd881bimKaclKszlaQ?pwd=lk8a) |
+
+### 1.3 Text Super-Resolution Algorithms
+
+Supported text super-resolution algorithms (Click the link to get the tutorial):
+
+- [x]  [Text Gestalt](./super_resolution/algorithm_sr_gestalt.en.md)
+- [x]  [Text Telescope](./super_resolution/algorithm_sr_telescope.en.md)
+
+On the TextZoom public dataset, the effect of the algorithm is as follows:
+
+|Model|Backbone|PSNR_Avg|SSIM_Avg|Config|Download link|
+|---|---|---|---|---|---|
+|Text Gestalt|tsrn|19.28|0.6560| [configs/sr/sr_tsrn_transformer_strock.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/sr/sr_tsrn_transformer_strock.yml)|[trained model](https://paddleocr.bj.bcebos.com/sr_tsrn_transformer_strock_train.tar)|
+|Text Telescope|tbsrn|21.56|0.7411| [configs/sr/sr_telescope.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/sr/sr_telescope.yml)|[trained model](https://paddleocr.bj.bcebos.com/contribution/sr_telescope_train.tar)|
+
+### 1.4 Formula Recognition Algorithm
+
+Supported formula recognition algorithms (Click the link to get the tutorial):
+
+- [x]  [CAN](./formula_recognition/algorithm_rec_can.en.md)
+- [x]  [LaTeX-OCR](./formula_recognition/algorithm_rec_latex_ocr.en.md)
+
+On the CROHME handwritten formula dataset, the effect of the algorithm is as follows:
+
+|Model    |Backbone|Config|ExpRate|Download link|
+| ----- | ----- | ----- | ----- | ----- |
+|CAN|DenseNet|[rec_d28_can.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_d28_can.yml)|51.72%|[trained model](https://paddleocr.bj.bcebos.com/contribution/rec_d28_can_train.tar)|
+
+## 2. End-to-end OCR Algorithms
+
+Supported end-to-end algorithms (Click the link to get the tutorial):
+
+- [x]  [PGNet](./end_to_end/algorithm_e2e_pgnet.en.md)
+
+## 3. Table Recognition Algorithms
+
+Supported table recognition algorithms (Click the link to get the tutorial):
+
+- [x]  [TableMaster](./table_recognition/algorithm_table_master.en.md)
+
+On the PubTabNet dataset, the algorithm result is as follows:
+
+|Model|Backbone|Config|Acc|Download link|
+|---|---|---|---|---|
+|TableMaster|TableResNetExtra|[configs/table/table_master.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/table/table_master.yml)|77.47%|[trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar) / [inference model](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_infer.tar)|
+
+## 4. Key Information Extraction Algorithms
+
+Supported KIE algorithms (Click the link to get the tutorial):
+
+- [x]  [VI-LayoutXLM](./kie/algorithm_kie_vi_layoutxlm.en.md)
+- [x]  [LayoutLM](./kie/algorithm_kie_layoutxlm.en.md)
+- [x]  [LayoutLMv2](./kie/algorithm_kie_layoutxlm.en.md)
+- [x]  [LayoutXLM](./kie/algorithm_kie_layoutxlm.en.md)
+- [x]  [SDMGR](./kie/algorithm_kie_sdmgr.en.md)
+
+On wildreceipt dataset, the algorithm result is as follows:
+
+|Model|Backbone|Config|Hmean|Download link|
+| --- | --- | --- | --- | --- |
+|SDMGR|VGG6|[configs/kie/sdmgr/kie_unet_sdmgr.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/sdmgr/kie_unet_sdmgr.yml)|86.70%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/kie/kie_vgg16.tar)|
+
+On XFUND_zh dataset, the algorithm result is as follows:
+
+|Model|Backbone|Task|Config|Hmean|Download link|
+| --- | --- |  --- | --- | --- | --- |
+|VI-LayoutXLM| VI-LayoutXLM-base | SER | [ser_vi_layoutxlm_xfund_zh_udml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh_udml.yml)|**93.19%**|[trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar)|
+|LayoutXLM| LayoutXLM-base | SER | [ser_layoutxlm_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/ser_layoutxlm_xfund_zh.yml)|90.38%|[trained model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar)|
+|LayoutLM| LayoutLM-base | SER | [ser_layoutlm_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/ser_layoutlm_xfund_zh.yml)|77.31%|[trained model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLM_xfun_zh.tar)|
+|LayoutLMv2| LayoutLMv2-base | SER | [ser_layoutlmv2_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/ser_layoutlmv2_xfund_zh.yml)|85.44%|[trained model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLMv2_xfun_zh.tar)|
+|VI-LayoutXLM| VI-LayoutXLM-base | RE | [re_vi_layoutxlm_xfund_zh_udml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh_udml.yml)|**83.92%**|[trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar)|
+|LayoutXLM| LayoutXLM-base | RE | [re_layoutxlm_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/re_layoutxlm_xfund_zh.yml)|74.83%|[trained model](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar)|
+|LayoutLMv2| LayoutLMv2-base | RE | [re_layoutlmv2_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/re_layoutlmv2_xfund_zh.yml)|67.77%|[trained model](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutLMv2_xfun_zh.tar)|
diff --git a/docs/version2.x/algorithm/overview.md b/docs/version2.x/algorithm/overview.md
new file mode 100755
index 0000000..40ca55b
--- /dev/null
+++ b/docs/version2.x/algorithm/overview.md
@@ -0,0 +1,180 @@
+---
+comments: true
+---
+
+# 前沿算法与模型
+
+本文给出了PaddleOCR已支持的OCR算法列表，以及每个算法在**英文公开数据集**上的模型和指标，主要用于算法简介和算法性能对比，更多包括中文在内的其他数据集上的模型请参考[PP-OCRv3 系列模型下载](../ppocr/model_list.md)。
+
+PaddleOCR将**持续新增**支持OCR领域前沿算法与模型，**欢迎广大开发者合作共建，贡献更多算法。**
+
+新增算法可参考教程：[使用PaddleOCR架构添加新算法](./add_new_algorithm.md)
+
+## 1. 两阶段算法
+
+### 1.1 文本检测算法
+
+已支持的文本检测算法列表（戳链接获取使用教程）：
+
+- [x]  [DB与DB++](./text_detection/algorithm_det_db.md)
+- [x]  [EAST](./text_detection/algorithm_det_east.md)
+- [x]  [SAST](./text_detection/algorithm_det_sast.md)
+- [x]  [PSENet](./text_detection/algorithm_det_psenet.md)
+- [x]  [FCENet](./text_detection/algorithm_det_fcenet.md)
+- [x]  [DRRG](./text_detection/algorithm_det_drrg.md)
+- [x]  [CT](./text_detection/algorithm_det_ct.md)
+
+在ICDAR2015文本检测公开数据集上，算法效果如下：
+
+|模型|骨干网络|precision|recall|Hmean|下载链接|
+| --- | --- | --- | --- | --- | --- |
+|EAST|ResNet50_vd|88.71%|81.36%|84.88%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar)|
+|EAST|MobileNetV3|78.20%|79.10%|78.65%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_east_v2.0_train.tar)|
+|DB|ResNet50_vd|86.41%|78.72%|82.38%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)|
+|DB|MobileNetV3|77.29%|73.08%|75.12%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)|
+|SAST|ResNet50_vd|91.39%|83.77%|87.42%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar)|
+|PSE|ResNet50_vd|85.81%|79.53%|82.55%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_vd_pse_v2.0_train.tar)|
+|PSE|MobileNetV3|82.20%|70.48%|75.89%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_mv3_pse_v2.0_train.tar)|
+|DB++|ResNet50|90.89%|82.66%|86.58%|[合成数据预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/ResNet50_dcn_asf_synthtext_pretrained.pdparams)/[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_db%2B%2B_icdar15_train.tar)|
+
+在Total-text文本检测公开数据集上，算法效果如下：
+
+|模型|骨干网络|precision|recall|Hmean|下载链接|
+| --- | --- | --- | --- | --- | --- |
+|SAST|ResNet50_vd|89.63%|78.44%|83.66%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_totaltext_v2.0_train.tar)|
+|CT|ResNet18_vd|88.68%|81.70%|85.05%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r18_ct_train.tar)|
+
+在CTW1500文本检测公开数据集上，算法效果如下：
+
+|模型|骨干网络|precision|recall|Hmean|下载链接|
+| --- | --- | --- | --- | --- | --- |
+|FCE|ResNet50_dcn|88.39%|82.18%|85.27%|[训练模型](https://paddleocr.bj.bcebos.com/contribution/det_r50_dcn_fce_ctw_v2.0_train.tar)|
+|DRRG|ResNet50_vd|89.92%|80.91%|85.18%|[训练模型](https://paddleocr.bj.bcebos.com/contribution/det_r50_drrg_ctw_train.tar)|
+
+**说明：** SAST模型训练额外加入了icdar2013、icdar2017、COCO-Text、ArT等公开数据集进行调优。PaddleOCR用到的经过整理格式的英文公开数据集下载：
+
+- [百度云地址](https://pan.baidu.com/s/12cPnZcVuV1zn5DOd4mqjVw) (提取码: 2bpi)
+- [Google Drive下载地址](https://drive.google.com/drive/folders/1ll2-XEVyCQLpJjawLDiRlvo_i4BqHCJe?usp=sharing)
+
+### 1.2 文本识别算法
+
+已支持的文本识别算法列表（戳链接获取使用教程）：
+
+- [x]  [CRNN](./text_recognition/algorithm_rec_crnn.md)
+- [x]  [Rosetta](./text_recognition/algorithm_rec_rosetta.md)
+- [x]  [STAR-Net](./text_recognition/algorithm_rec_starnet.md)
+- [x]  [RARE](./text_recognition/algorithm_rec_rare.md)
+- [x]  [SRN](./text_recognition/algorithm_rec_srn.md)
+- [x]  [NRTR](./text_recognition/algorithm_rec_nrtr.md)
+- [x]  [SAR](./text_recognition/algorithm_rec_sar.md)
+- [x]  [SEED](./text_recognition/algorithm_rec_seed.md)
+- [x]  [SVTR](./text_recognition/algorithm_rec_svtr.md)
+- [x]  [ViTSTR](./text_recognition/algorithm_rec_vitstr.md)
+- [x]  [ABINet](./text_recognition/algorithm_rec_abinet.md)
+- [x]  [VisionLAN](./text_recognition/algorithm_rec_visionlan.md)
+- [x]  [SPIN](./text_recognition/algorithm_rec_spin.md)
+- [x]  [RobustScanner](./text_recognition/algorithm_rec_robustscanner.md)
+- [x]  [RFL](./text_recognition/algorithm_rec_rfl.md)
+- [x]  [ParseQ](./text_recognition/algorithm_rec_parseq.md)
+- [x]  [CPPD](./text_recognition/algorithm_rec_cppd.md)
+- [x]  [SATRN](./text_recognition/algorithm_rec_satrn.md)
+
+参考[DTRB](https://arxiv.org/abs/1904.01906) (3)文字识别训练和评估流程，使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法效果如下：
+
+|模型|骨干网络|Avg Accuracy|模型存储命名|下载链接|
+|---|---|---|---|---|
+|Rosetta|Resnet34_vd|79.11%|rec_r34_vd_none_none_ctc|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_none_ctc_v2.0_train.tar)|
+|Rosetta|MobileNetV3|75.80%|rec_mv3_none_none_ctc|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_none_ctc_v2.0_train.tar)|
+|CRNN|Resnet34_vd|81.04%|rec_r34_vd_none_bilstm_ctc|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_bilstm_ctc_v2.0_train.tar)|
+|CRNN|MobileNetV3|77.95%|rec_mv3_none_bilstm_ctc|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_bilstm_ctc_v2.0_train.tar)|
+|StarNet|Resnet34_vd|82.85%|rec_r34_vd_tps_bilstm_ctc|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_ctc_v2.0_train.tar)|
+|StarNet|MobileNetV3|79.28%|rec_mv3_tps_bilstm_ctc|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_tps_bilstm_ctc_v2.0_train.tar)|
+|RARE|Resnet34_vd|83.98%|rec_r34_vd_tps_bilstm_att |[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_att_v2.0_train.tar)|
+|RARE|MobileNetV3|81.76%|rec_mv3_tps_bilstm_att |[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_tps_bilstm_att_v2.0_train.tar)|
+|SRN|Resnet50_vd_fpn| 86.31% | rec_r50fpn_vd_none_srn | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r50_vd_srn_train.tar) |
+|NRTR|NRTR_MTB| 84.21% | rec_mtb_nrtr | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mtb_nrtr_train.tar) |
+|SAR|Resnet31| 87.20% | rec_r31_sar | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_sar_train.tar) |
+|SEED|Aster_Resnet| 85.35% | rec_resnet_stn_bilstm_att | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_resnet_stn_bilstm_att.tar) |
+|SVTR|SVTR-Tiny| 89.25% | rec_svtr_tiny_none_ctc_en | [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar) |
+|ViTSTR|ViTSTR| 79.82% | rec_vitstr_none_ce | [训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar) |
+|ABINet|Resnet45| 90.75% | rec_r45_abinet | [训练模型](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar) |
+|VisionLAN|Resnet45| 90.30% | rec_r45_visionlan | [训练模型](https://paddleocr.bj.bcebos.com/VisionLAN/rec_r45_visionlan_train.tar) |
+|SPIN|ResNet32| 90.00% | rec_r32_gaspin_bilstm_att | [训练模型](https://paddleocr.bj.bcebos.com/contribution/rec_r32_gaspin_bilstm_att.tar) |
+|RobustScanner|ResNet31| 87.77% | rec_r31_robustscanner | [训练模型](https://paddleocr.bj.bcebos.com/contribution/rec_r31_robustscanner.tar)|
+|RFL|ResNetRFL| 88.63% | rec_resnet_rfl_att | [训练模型](https://paddleocr.bj.bcebos.com/contribution/rec_resnet_rfl_att_train.tar) |
+|ParseQ|VIT| 91.24% | rec_vit_parseq_synth | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/parseq/rec_vit_parseq_synth.tgz) |
+|CPPD|SVTR-Base| 93.8% | rec_svtrnet_cppd_base_en | [训练模型](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_en_train.tar) |
+|SATRN|ShallowCNN| 88.05% | rec_satrn | [训练模型](https://pan.baidu.com/s/10J-Bsd881bimKaclKszlaQ?pwd=lk8a) |
+
+### 1.3 文本超分辨率算法
+
+已支持的文本超分辨率算法列表（戳链接获取使用教程）：
+
+- [x]  [Text Gestalt](./super_resolution/algorithm_sr_gestalt.md)
+- [x]  [Text Telescope](./super_resolution/algorithm_sr_telescope.md)
+
+在TextZoom公开数据集上，算法效果如下：
+
+|模型|骨干网络|PSNR_Avg|SSIM_Avg|配置文件|下载链接|
+|---|---|---|---|---|---|
+|Text Gestalt|tsrn|19.28|0.6560| [configs/sr/sr_tsrn_transformer_strock.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/sr/sr_tsrn_transformer_strock.yml)|[训练模型](https://paddleocr.bj.bcebos.com/sr_tsrn_transformer_strock_train.tar)|
+|Text Telescope|tbsrn|21.56|0.7411| [configs/sr/sr_telescope.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/sr/sr_telescope.yml)|[训练模型](https://paddleocr.bj.bcebos.com/contribution/sr_telescope_train.tar)|
+
+### 1.4 公式识别算法
+
+已支持的公式识别算法列表（戳链接获取使用教程）：
+
+- [x]  [CAN](./formula_recognition/algorithm_rec_can.md)
+- [x]  [LaTeX-OCR](./formula_recognition/algorithm_rec_latex_ocr.md)
+
+在CROHME手写公式数据集上，算法效果如下：
+
+|模型    |骨干网络|配置文件|ExpRate|下载链接|
+| ----- | ----- | ----- | ----- | ----- |
+|CAN|DenseNet|[rec_d28_can.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_d28_can.yml)|51.72%|[训练模型](https://paddleocr.bj.bcebos.com/contribution/rec_d28_can_train.tar)|
+
+## 2. 端到端算法
+
+已支持的端到端OCR算法列表（戳链接获取使用教程）：
+
+- [x]  [PGNet](./end_to_end/algorithm_e2e_pgnet.md)
+
+## 3. 表格识别算法
+
+已支持的表格识别算法列表（戳链接获取使用教程）：
+
+- [x]  [TableMaster](./table_recognition/algorithm_table_master.md)
+
+在PubTabNet表格识别公开数据集上，算法效果如下：
+
+|模型|骨干网络|配置文件|acc|下载链接|
+|---|---|---|---|---|
+|TableMaster|TableResNetExtra|[configs/table/table_master.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/table/table_master.yml)|77.47%|[训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar) / [推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_infer.tar)|
+
+## 4. 关键信息抽取算法
+
+已支持的关键信息抽取算法列表（戳链接获取使用教程）：
+
+- [x]  [VI-LayoutXLM](./kie/algorithm_kie_vi_layoutxlm.md)
+- [x]  [LayoutLM](./kie/algorithm_kie_layoutxlm.md)
+- [x]  [LayoutLMv2](./kie/algorithm_kie_layoutxlm.md)
+- [x]  [LayoutXLM](./kie/algorithm_kie_layoutxlm.md)
+- [x]  [SDMGR](./kie/algorithm_kie_sdmgr.md)
+
+在wildreceipt发票公开数据集上，算法复现效果如下：
+
+|模型|骨干网络|配置文件|hmean|下载链接|
+| --- | --- | --- | --- | --- |
+|SDMGR|VGG6|[configs/kie/sdmgr/kie_unet_sdmgr.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/sdmgr/kie_unet_sdmgr.yml)|86.70%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/kie/kie_vgg16.tar)|
+
+在XFUND_zh公开数据集上，算法效果如下：
+
+|模型|骨干网络|任务|配置文件|hmean|下载链接|
+| --- | --- |  --- | --- | --- | --- |
+|VI-LayoutXLM| VI-LayoutXLM-base | SER | [ser_vi_layoutxlm_xfund_zh_udml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh_udml.yml)|**93.19%**|[训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar)|
+|LayoutXLM| LayoutXLM-base | SER | [ser_layoutxlm_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/ser_layoutxlm_xfund_zh.yml)|90.38%|[训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar)|
+|LayoutLM| LayoutLM-base | SER | [ser_layoutlm_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/ser_layoutlm_xfund_zh.yml)|77.31%|[训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLM_xfun_zh.tar)|
+|LayoutLMv2| LayoutLMv2-base | SER | [ser_layoutlmv2_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/ser_layoutlmv2_xfund_zh.yml)|85.44%|[训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLMv2_xfun_zh.tar)|
+|VI-LayoutXLM| VI-LayoutXLM-base | RE | [re_vi_layoutxlm_xfund_zh_udml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh_udml.yml)|**83.92%**|[训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar)|
+|LayoutXLM| LayoutXLM-base | RE | [re_layoutxlm_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/re_layoutxlm_xfund_zh.yml)|74.83%|[训练模型](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar)|
+|LayoutLMv2| LayoutLMv2-base | RE | [re_layoutlmv2_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/re_layoutlmv2_xfund_zh.yml)|67.77%|[训练模型](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutLMv2_xfun_zh.tar)|
diff --git a/docs/version2.x/algorithm/super_resolution/algorithm_sr_gestalt.en.md b/docs/version2.x/algorithm/super_resolution/algorithm_sr_gestalt.en.md
new file mode 100644
index 0000000..faa58fc
--- /dev/null
+++ b/docs/version2.x/algorithm/super_resolution/algorithm_sr_gestalt.en.md
@@ -0,0 +1,107 @@
+---
+comments: true
+---
+
+# Text Gestalt
+
+## 1. Introduction
+
+Paper:
+> [Text Gestalt: Stroke-Aware Scene Text Image Super-Resolution](https://arxiv.org/pdf/2112.08171.pdf)
+> Chen, Jingye and Yu, Haiyang and Ma, Jianqi and Li, Bin and Xue, Xiangyang
+> AAAI, 2022
+
+Referring to the [FudanOCR](https://github.com/FudanVI/FudanOCR/tree/main/text-gestalt) data download instructions, the effect of the super-score algorithm on the TextZoom test set is as follows:
+
+|Model | Backbone|config|Acc|Download link|
+|---|---|---|---|---|
+|Text Gestalt|tsrn|19.28|0.6560| [configs/sr/sr_tsrn_transformer_strock.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/sr/sr_tsrn_transformer_strock.yml)|[train model](https://paddleocr.bj.bcebos.com/sr_tsrn_transformer_strock_train.tar)|
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/sr/sr_tsrn_transformer_strock.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/sr/sr_tsrn_transformer_strock.yml
+```
+
+### Evaluation
+
+```bash linenums="1"
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/sr/sr_tsrn_transformer_strock.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### Prediction
+
+```bash linenums="1"
+# The configuration file used for prediction must match the training
+
+python3 tools/infer_sr.py -c configs/sr/sr_tsrn_transformer_strock.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words_en/word_52.png
+```
+
+![img](./images/word_52.png)
+
+After executing the command, the super-resolution result of the above image is as follows:
+
+![img](./images/sr_word_52.png)
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/sr_tsrn_transformer_strock_train.tar) ), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/sr/sr_tsrn_transformer_strock.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.save_inference_dir=./inference/sr_out
+```
+
+For Text-Gestalt super-resolution model inference, the following commands can be executed:
+
+```bash linenums="1"
+python3 tools/infer/predict_sr.py --sr_model_dir=./inference/sr_out --image_dir=doc/imgs_words_en/word_52.png --sr_image_shape=3,32,128
+```
+
+After executing the command, the super-resolution result of the above image is as follows:
+
+![img](./images/sr_word_52-20240704093810101.png)
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@inproceedings{chen2022text,
+  title={Text gestalt: Stroke-aware scene text image super-resolution},
+  author={Chen, Jingye and Yu, Haiyang and Ma, Jianqi and Li, Bin and Xue, Xiangyang},
+  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
+  volume={36},
+  number={1},
+  pages={285--293},
+  year={2022}
+}
+```
diff --git a/docs/version2.x/algorithm/super_resolution/algorithm_sr_gestalt.md b/docs/version2.x/algorithm/super_resolution/algorithm_sr_gestalt.md
new file mode 100644
index 0000000..41b3ea4
--- /dev/null
+++ b/docs/version2.x/algorithm/super_resolution/algorithm_sr_gestalt.md
@@ -0,0 +1,107 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# Text Gestalt
+
+## 1. 算法简介
+
+论文信息：
+> [Text Gestalt: Stroke-Aware Scene Text Image Super-Resolution](https://arxiv.org/pdf/2112.08171.pdf)
+> Chen, Jingye and Yu, Haiyang and Ma, Jianqi and Li, Bin and Xue, Xiangyang
+> AAAI, 2022
+
+参考[FudanOCR](https://github.com/FudanVI/FudanOCR/tree/main/text-gestalt) 数据下载说明，在TextZoom测试集合上超分算法效果如下：
+
+|模型|骨干网络|PSNR_Avg|SSIM_Avg|配置文件|下载链接|
+|---|---|---|---|---|---|
+|Text Gestalt|tsrn|19.28|0.6560| [configs/sr/sr_tsrn_transformer_strock.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/sr/sr_tsrn_transformer_strock.yml)|[训练模型](https://paddleocr.bj.bcebos.com/sr_tsrn_transformer_strock_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练不同的识别模型只需要**更换配置文件**即可。
+
+### 训练
+
+在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/sr/sr_tsrn_transformer_strock.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/sr/sr_tsrn_transformer_strock.yml
+```
+
+### 评估
+
+```bash linenums="1"
+# GPU 评估， Global.pretrained_model 为待测权重
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/sr/sr_tsrn_transformer_strock.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### 预测
+
+```bash linenums="1"
+# 预测使用的配置文件必须与训练一致
+python3 tools/infer_sr.py -c configs/sr/sr_tsrn_transformer_strock.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words_en/word_52.png
+```
+
+![img](./images/word_52.png)
+
+执行命令后，上面图像的超分结果如下：
+
+![img](./images/sr_word_52.png)
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将文本超分训练过程中保存的模型，转换成inference model。以 Text-Gestalt 训练的[模型](https://paddleocr.bj.bcebos.com/sr_tsrn_transformer_strock_train.tar) 为例，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/sr/sr_tsrn_transformer_strock.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.save_inference_dir=./inference/sr_out
+```
+
+Text-Gestalt 文本超分模型推理，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_sr.py --sr_model_dir=./inference/sr_out --image_dir=doc/imgs_words_en/word_52.png --sr_image_shape=3,32,128
+```
+
+执行命令后，图像的超分结果如下：
+
+![img](./images/sr_word_52-20240704093810101.png)
+
+### 4.2 C++推理
+
+暂未支持
+
+### 4.3 Serving服务化部署
+
+暂未支持
+
+### 4.4 更多推理部署
+
+暂未支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@inproceedings{chen2022text,
+  title={Text gestalt: Stroke-aware scene text image super-resolution},
+  author={Chen, Jingye and Yu, Haiyang and Ma, Jianqi and Li, Bin and Xue, Xiangyang},
+  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
+  volume={36},
+  number={1},
+  pages={285--293},
+  year={2022}
+}
+```
diff --git a/docs/version2.x/algorithm/super_resolution/algorithm_sr_telescope.en.md b/docs/version2.x/algorithm/super_resolution/algorithm_sr_telescope.en.md
new file mode 100644
index 0000000..b89847a
--- /dev/null
+++ b/docs/version2.x/algorithm/super_resolution/algorithm_sr_telescope.en.md
@@ -0,0 +1,110 @@
+---
+comments: true
+---
+
+# Text Gestalt
+
+## 1. Introduction
+
+Paper:
+> [Scene Text Telescope: Text-Focused Scene Image Super-Resolution](https://openaccess.thecvf.com/content/CVPR2021/papers/Chen_Scene_Text_Telescope_Text-Focused_Scene_Image_Super-Resolution_CVPR_2021_paper.pdf)
+> Chen, Jingye, Bin Li, and Xiangyang Xue
+> CVPR, 2021
+
+Referring to the [FudanOCR](https://github.com/FudanVI/FudanOCR/tree/main/scene-text-telescope) data download instructions, the effect of the super-score algorithm on the TextZoom test set is as follows:
+
+|Model|Backbone|config|Acc|Download link|
+|---|---|---|---|---|
+|Text Gestalt|tsrn|21.56|0.7411| [configs/sr/sr_telescope.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/sr/sr_telescope.yml)|[train model](https://paddleocr.bj.bcebos.com/contribution/sr_telescope_train.tar)|
+
+The [TextZoom dataset](https://paddleocr.bj.bcebos.com/dataset/TextZoom.tar) comes from two superfraction data sets, RealSR and SR-RAW, both of which contain LR-HR pairs. TextZoom has 17367 pairs of training data and 4373 pairs of test data.
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/sr/sr_telescope.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/sr/sr_telescope.yml
+```
+
+### Evaluation
+
+```bash linenums="1"
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/sr/sr_telescope.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### Prediction
+
+```bash linenums="1"
+# The configuration file used for prediction must match the training
+
+python3 tools/infer_sr.py -c configs/sr/sr_telescope.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words_en/word_52.png
+```
+
+![img](./images/word_52-20240704094304807.png)
+
+After executing the command, the super-resolution result of the above image is as follows:
+
+![img](./images/sr_word_52-20240704094309205.png)
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/contribution/Telescope_train.tar.gz) ), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/sr/sr_telescope.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.save_inference_dir=./inference/sr_out
+```
+
+For Text-Telescope super-resolution model inference, the following commands can be executed:
+
+```bash linenums="1"
+python3 tools/infer/predict_sr.py --sr_model_dir=./inference/sr_out --image_dir=doc/imgs_words_en/word_52.png --sr_image_shape=3,32,128
+
+```
+
+After executing the command, the super-resolution result of the above image is as follows:
+
+![img](./images/sr_word_52-20240704094309205.png)
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@INPROCEEDINGS{9578891,
+  author={Chen, Jingye and Li, Bin and Xue, Xiangyang},
+  booktitle={2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
+  title={Scene Text Telescope: Text-Focused Scene Image Super-Resolution},
+  year={2021},
+  volume={},
+  number={},
+  pages={12021-12030},
+  doi={10.1109/CVPR46437.2021.01185}}
+```
diff --git a/docs/version2.x/algorithm/super_resolution/algorithm_sr_telescope.md b/docs/version2.x/algorithm/super_resolution/algorithm_sr_telescope.md
new file mode 100644
index 0000000..4f2ff80
--- /dev/null
+++ b/docs/version2.x/algorithm/super_resolution/algorithm_sr_telescope.md
@@ -0,0 +1,109 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# Text Telescope
+
+## 1. 算法简介
+
+论文信息：
+> [Scene Text Telescope: Text-Focused Scene Image Super-Resolution](https://openaccess.thecvf.com/content/CVPR2021/papers/Chen_Scene_Text_Telescope_Text-Focused_Scene_Image_Super-Resolution_CVPR_2021_paper.pdf)
+> Chen, Jingye, Bin Li, and Xiangyang Xue
+> CVPR, 2021
+
+参考[FudanOCR](https://github.com/FudanVI/FudanOCR/tree/main/scene-text-telescope) 数据下载说明，在TextZoom测试集合上超分算法效果如下：
+
+|模型|骨干网络|PSNR_Avg|SSIM_Avg|配置文件|下载链接|
+|---|---|---|---|---|---|
+|Text Telescope|tbsrn|21.56|0.7411| [configs/sr/sr_telescope.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/sr/sr_telescope.yml)|[训练模型](https://paddleocr.bj.bcebos.com/contribution/sr_telescope_train.tar)|
+
+[TextZoom数据集](https://paddleocr.bj.bcebos.com/dataset/TextZoom.tar) 来自两个超分数据集RealSR和SR-RAW，两个数据集都包含LR-HR对，TextZoom有17367对训数据和4373对测试数据。
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练不同的识别模型只需要**更换配置文件**即可。
+
+### 训练
+
+在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/sr/sr_telescope.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/sr/sr_telescope.yml
+```
+
+### 评估
+
+```bash linenums="1"
+# GPU 评估， Global.pretrained_model 为待测权重
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/sr/sr_telescope.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### 预测
+
+```bash linenums="1"
+# 预测使用的配置文件必须与训练一致
+python3 tools/infer_sr.py -c configs/sr/sr_telescope.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words_en/word_52.png
+```
+
+![img](./images/word_52-20240704094304807.png)
+
+执行命令后，上面图像的超分结果如下：
+
+![img](./images/sr_word_52-20240704094309205.png)
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将文本超分训练过程中保存的模型，转换成inference model。以 Text-Telescope 训练的[模型](https://paddleocr.bj.bcebos.com/contribution/Telescope_train.tar.gz) 为例，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/sr/sr_telescope.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.save_inference_dir=./inference/sr_out
+```
+
+Text-Telescope 文本超分模型推理，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_sr.py --sr_model_dir=./inference/sr_out --image_dir=doc/imgs_words_en/word_52.png --sr_image_shape=3,32,128
+```
+
+执行命令后，图像的超分结果如下：
+
+![img](./images/sr_word_52-20240704094309205.png)
+
+### 4.2 C++推理
+
+暂未支持
+
+### 4.3 Serving服务化部署
+
+暂未支持
+
+### 4.4 更多推理部署
+
+暂未支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@INPROCEEDINGS{9578891,
+  author={Chen, Jingye and Li, Bin and Xue, Xiangyang},
+  booktitle={2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
+  title={Scene Text Telescope: Text-Focused Scene Image Super-Resolution},
+  year={2021},
+  volume={},
+  number={},
+  pages={12021-12030},
+  doi={10.1109/CVPR46437.2021.01185}}
+```
diff --git a/docs/version2.x/algorithm/super_resolution/images/sr_word_52-20240704093810101.png b/docs/version2.x/algorithm/super_resolution/images/sr_word_52-20240704093810101.png
new file mode 100644
index 0000000..c983e9a
Binary files /dev/null and b/docs/version2.x/algorithm/super_resolution/images/sr_word_52-20240704093810101.png differ
diff --git a/docs/version2.x/algorithm/super_resolution/images/sr_word_52-20240704094309205.png b/docs/version2.x/algorithm/super_resolution/images/sr_word_52-20240704094309205.png
new file mode 100644
index 0000000..c983e9a
Binary files /dev/null and b/docs/version2.x/algorithm/super_resolution/images/sr_word_52-20240704094309205.png differ
diff --git a/docs/version2.x/algorithm/super_resolution/images/sr_word_52.png b/docs/version2.x/algorithm/super_resolution/images/sr_word_52.png
new file mode 100644
index 0000000..c983e9a
Binary files /dev/null and b/docs/version2.x/algorithm/super_resolution/images/sr_word_52.png differ
diff --git a/docs/version2.x/algorithm/super_resolution/images/word_52-20240704094304807.png b/docs/version2.x/algorithm/super_resolution/images/word_52-20240704094304807.png
new file mode 100644
index 0000000..493c590
Binary files /dev/null and b/docs/version2.x/algorithm/super_resolution/images/word_52-20240704094304807.png differ
diff --git a/docs/version2.x/algorithm/super_resolution/images/word_52.png b/docs/version2.x/algorithm/super_resolution/images/word_52.png
new file mode 100644
index 0000000..493c590
Binary files /dev/null and b/docs/version2.x/algorithm/super_resolution/images/word_52.png differ
diff --git a/docs/version2.x/algorithm/table_recognition/algorithm_table_master.en.md b/docs/version2.x/algorithm/table_recognition/algorithm_table_master.en.md
new file mode 100644
index 0000000..7fe44ab
--- /dev/null
+++ b/docs/version2.x/algorithm/table_recognition/algorithm_table_master.en.md
@@ -0,0 +1,93 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+
+# Table Recognition Algorithm-TableMASTER
+
+## 1. Introduction
+
+Paper:
+> [TableMaster: PINGAN-VCGROUP’S SOLUTION FOR ICDAR 2021 COMPETITION ON SCIENTIFIC LITERATURE PARSING TASK B: TABLE RECOGNITION TO HTML](https://arxiv.org/pdf/2105.01848.pdf)
+> Ye, Jiaquan and Qi, Xianbiao and He, Yelin and Chen, Yihao and Gu, Dengyi and Gao, Peng and Xiao, Rong
+> 2021
+
+On the PubTabNet table recognition public data set, the algorithm reproduction acc is as follows:
+
+|Model|Backbone|Cnnfig|Acc|Download link|
+| --- | --- | --- | --- | --- |
+|TableMaster|TableResNetExtra|[configs/table/table_master.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/table/table_master.yml)|77.47%|[trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar)/[inference model](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_infer.tar)|
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+The above TableMaster model is trained using the PubTabNet table recognition public dataset. For the download of the dataset, please refer to [table_datasets](../../../datasets/table_datasets.en.md).
+
+After the data download is complete, please refer to [Text Recognition Training Tutorial](../../ppocr/model_train/recognition.en.md) for training. PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different models.
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, convert the model saved in the TableMaster table recognition training process into an inference model. Taking the model based on the TableResNetExtra backbone network and trained on the PubTabNet dataset as example ([model download link](https://paddleocr.bj.bcebos.com/contribution/table_master.tar)), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/table/table_master.yml -o Global.pretrained_model=output/table_master/best_accuracy Global.save_inference_dir=./inference/table_master
+```
+
+**Note:**
+
+- If you trained the model on your own dataset and adjusted the dictionary file, please pay attention to whether the `character_dict_path` in the modified configuration file is the correct dictionary file
+
+Execute the following command for model inference:
+
+```bash linenums="1"
+cd ppstructure/
+# When predicting all images in a folder, you can modify image_dir to a folder, such as --image_dir='docs/table'.
+python3 table/predict_structure.py --table_model_dir=../output/table_master/table_structure_tablemaster_infer/ --table_algorithm=TableMaster --table_char_dict_path=../ppocr/utils/dict/table_master_structure_dict.txt --table_max_len=480 --image_dir=docs/table/table.jpg
+```
+
+After executing the command, the prediction results of the above image (structural information and the coordinates of each cell in the table) are printed to the screen, and the visualization of the cell coordinates is also saved. An example is as follows:
+
+result：
+
+```bash linenums="1"
+[2022/06/16 13:06:54] ppocr INFO: result: ['<html>', '<body>', '<table>', '<thead>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '</thead>', '<tbody>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '</tbody>', '</table>', '</body>', '</html>'], [[72.17591094970703, 10.759100914001465, 60.29658508300781, 16.6805362701416], [161.85562133789062, 10.884308815002441, 14.9495210647583, 16.727018356323242], [277.79876708984375, 29.54340362548828, 31.490320205688477, 18.143272399902344],
+...
+[336.11724853515625, 280.3601989746094, 39.456939697265625, 18.121286392211914]]
+[2022/06/16 13:06:54] ppocr INFO: save vis result to ./output/table.jpg
+[2022/06/16 13:06:54] ppocr INFO: Predict time of docs/table/table.jpg: 17.36806297302246
+```
+
+**Note**:
+
+- TableMaster is relatively slow during inference, and it is recommended to use GPU for use.
+
+### 4.2 C++ Inference
+
+Since the post-processing is not written in CPP, the TableMaster does not support CPP inference.
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@article{ye2021pingan,
+  title={PingAn-VCGroup's Solution for ICDAR 2021 Competition on Scientific Literature Parsing Task B: Table Recognition to HTML},
+  author={Ye, Jiaquan and Qi, Xianbiao and He, Yelin and Chen, Yihao and Gu, Dengyi and Gao, Peng and Xiao, Rong},
+  journal={arXiv preprint arXiv:2105.01848},
+  year={2021}
+}
+```
diff --git a/docs/version2.x/algorithm/table_recognition/algorithm_table_master.md b/docs/version2.x/algorithm/table_recognition/algorithm_table_master.md
new file mode 100644
index 0000000..e95aaba
--- /dev/null
+++ b/docs/version2.x/algorithm/table_recognition/algorithm_table_master.md
@@ -0,0 +1,100 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+
+# 表格识别算法-TableMASTER
+
+## 1. 算法简介
+
+论文信息：
+> [TableMaster: PINGAN-VCGROUP’S SOLUTION FOR ICDAR 2021 COMPETITION ON SCIENTIFIC LITERATURE PARSING TASK B: TABLE RECOGNITION TO HTML](https://arxiv.org/pdf/2105.01848.pdf)
+> Ye, Jiaquan and Qi, Xianbiao and He, Yelin and Chen, Yihao and Gu, Dengyi and Gao, Peng and Xiao, Rong
+> 2021
+
+在PubTabNet表格识别公开数据集上，算法复现效果如下：
+
+|模型|骨干网络|配置文件|acc|下载链接|
+| --- | --- | --- | --- | --- |
+|TableMaster|TableResNetExtra|[configs/table/table_master.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/table/table_master.yml)|77.47%|[训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar)/[推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_infer.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+上述TableMaster模型使用PubTabNet表格识别公开数据集训练得到，数据集下载可参考 [table_datasets](../../../datasets/table_datasets.md)。
+
+数据下载完成后，请参考[文本识别教程](../../ppocr/model_train/recognition.md)进行训练。PaddleOCR对代码进行了模块化，训练不同的模型只需要**更换配置文件**即可。
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将训练得到best模型，转换成inference model。以基于TableResNetExtra骨干网络，在PubTabNet数据集训练的模型为例([模型下载地址](https://paddleocr.bj.bcebos.com/contribution/table_master.tar))，可以使用如下命令进行转换：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/export_model.py -c configs/table/table_master.yml -o Global.pretrained_model=output/table_master/best_accuracy Global.save_inference_dir=./inference/table_master
+```
+
+**注意：** 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否为所正确的字典文件。
+
+转换成功后，在目录下有三个文件：
+
+```text linenums="1"
+./inference/table_master/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+执行如下命令进行模型推理：
+
+```bash linenums="1"
+cd ppstructure/
+python3 table/predict_structure.py --table_model_dir=../output/table_master/table_structure_tablemaster_infer/ --table_algorithm=TableMaster --table_char_dict_path=../ppocr/utils/dict/table_master_structure_dict.txt --table_max_len=480 --image_dir=docs/table/table.jpg
+# 预测文件夹下所有图像时，可修改image_dir为文件夹，如 --image_dir='docs/table'。
+```
+
+执行命令后，上面图像的预测结果（结构信息和表格中每个单元格的坐标）会打印到屏幕上，同时会保存单元格坐标的可视化结果。示例如下：
+结果如下：
+
+```bash linenums="1"
+[2022/06/16 13:06:54] ppocr INFO: result: ['<html>', '<body>', '<table>', '<thead>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '</thead>', '<tbody>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '</tbody>', '</table>', '</body>', '</html>'], [[72.17591094970703, 10.759100914001465, 60.29658508300781, 16.6805362701416], [161.85562133789062, 10.884308815002441, 14.9495210647583, 16.727018356323242], [277.79876708984375, 29.54340362548828, 31.490320205688477, 18.143272399902344],
+...
+[336.11724853515625, 280.3601989746094, 39.456939697265625, 18.121286392211914]]
+[2022/06/16 13:06:54] ppocr INFO: save vis result to ./output/table.jpg
+[2022/06/16 13:06:54] ppocr INFO: Predict time of docs/table/table.jpg: 17.36806297302246
+```
+
+**注意**：
+
+- TableMaster在推理时比较慢，建议使用GPU进行使用。
+
+### 4.2 C++推理部署
+
+由于C++预处理后处理还未支持TableMaster，所以暂未支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@article{ye2021pingan,
+  title={PingAn-VCGroup's Solution for ICDAR 2021 Competition on Scientific Literature Parsing Task B: Table Recognition to HTML},
+  author={Ye, Jiaquan and Qi, Xianbiao and He, Yelin and Chen, Yihao and Gu, Dengyi and Gao, Peng and Xiao, Rong},
+  journal={arXiv preprint arXiv:2105.01848},
+  year={2021}
+}
+```
diff --git a/docs/version2.x/algorithm/table_recognition/algorithm_table_slanet.md b/docs/version2.x/algorithm/table_recognition/algorithm_table_slanet.md
new file mode 100644
index 0000000..ecca71a
--- /dev/null
+++ b/docs/version2.x/algorithm/table_recognition/algorithm_table_slanet.md
@@ -0,0 +1,97 @@
+---
+comments: true
+---
+
+# 表格识别算法-SLANet-LCNetV2
+
+## 1. 算法简介
+
+该算法由来自北京交通大学机器学习与认识计算研究团队的ocr识别队研发，其在PaddleOCR算法模型挑战赛 - 赛题二：通用表格识别任务中排行榜荣获一等奖，排行榜精度相比PP-Structure表格识别模型提升0.8%，推理速度提升3倍。优化思路如下：
+
+1. 改善推理过程，至EOS停止，速度提升3倍；
+2. 升级Backbone为LCNetV2（SSLD版本）；
+3. 行列特征增强模块；
+4. 提升分辨率488至512；
+5. 三阶段训练策略。
+
+在PubTabNet表格识别公开数据集上，算法复现效果如下：
+
+|模型|骨干网络|配置文件|acc|下载链接|
+| --- | --- | --- | --- | --- |
+|SLANet|LCNetV2|[configs/table/SLANet_lcnetv2.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/table/SLANet_lcnetv2.yml)|76.67%| [训练模型](https://paddleocr.bj.bcebos.com/openatom/ch_ppstructure_openatom_SLANetv2_train.tar) /[推理模型](https://paddleocr.bj.bcebos.com/openatom/ch_ppstructure_openatom_SLANetv2_infer.tar) |
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+上述SLANet_LCNetv2模型使用PubTabNet表格识别公开数据集训练得到，数据集下载可参考 [table_datasets](../../../datasets/table_datasets.md)。
+
+### 启动训练
+
+数据下载完成后，请参考[文本识别教程](../../ppocr/model_train/recognition.md)进行训练。PaddleOCR对代码进行了模块化，训练不同的模型只需要**更换配置文件**即可。
+
+训练命令如下：
+
+```bash linenums="1"
+# stage1
+python3 -m paddle.distributed.launch --gpus '0,1,2,3,4,5,6,7'  tools/train.py -c configs/table/SLANet_lcnetv2.yml
+# stage2 加载stage1的best model作为预训练模型，学习率调整为0.0001;
+# stage3 加载stage2的best model作为预训练模型，不调整学习率，将配置文件中所有的488修改为512.
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+将训练得到best模型，转换成inference model，可以使用如下命令进行转换：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/export_model.py -c configs/table/SLANet_lcnetv2.yml -o Global.pretrained_model=path/best_accuracy Global.save_inference_dir=./inference/slanet_lcnetv2_infer
+```
+
+**注意：** 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否为所正确的字典文件。
+
+转换成功后，在目录下有三个文件：
+
+```text linenums="1"
+./inference/slanet_lcnetv2_infer/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+执行如下命令进行模型推理：
+
+```bash linenums="1"
+cd ppstructure/
+python table/predict_structure.py --table_model_dir=../inference/slanet_lcnetv2_infer/ --table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt --image_dir=docs/table/table.jpg --output=../output/table_slanet_lcnetv2 --use_gpu=False --benchmark=True --enable_mkldnn=True --table_max_len=512
+# 预测文件夹下所有图像时，可修改image_dir为文件夹，如 --image_dir='docs/table'。
+```
+
+执行命令后，上面图像的预测结果（结构信息和表格中每个单元格的坐标）会打印到屏幕上，同时会保存单元格坐标的可视化结果。示例如下：
+结果如下：
+
+```bash linenums="1"
+[2022/06/16 13:06:54] ppocr INFO: result: ['<html>', '<body>', '<table>', '<thead>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '</thead>', '<tbody>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '</tbody>', '</table>', '</body>', '</html>'], [[72.17591094970703, 10.759100914001465, 60.29658508300781, 16.6805362701416], [161.85562133789062, 10.884308815002441, 14.9495210647583, 16.727018356323242], [277.79876708984375, 29.54340362548828, 31.490320205688477, 18.143272399902344],
+...
+[336.11724853515625, 280.3601989746094, 39.456939697265625, 18.121286392211914]]
+[2022/06/16 13:06:54] ppocr INFO: save vis result to ./output/table.jpg
+[2022/06/16 13:06:54] ppocr INFO: Predict time of docs/table/table.jpg: 17.36806297302246
+```
+
+### 4.2 C++推理部署
+
+由于C++预处理后处理还未支持SLANet
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
diff --git a/docs/version2.x/algorithm/text_detection/algorithm_det_ct.en.md b/docs/version2.x/algorithm/text_detection/algorithm_det_ct.en.md
new file mode 100644
index 0000000..d37e37b
--- /dev/null
+++ b/docs/version2.x/algorithm/text_detection/algorithm_det_ct.en.md
@@ -0,0 +1,74 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# CT
+
+## 1. Introduction
+
+Paper:
+> [CentripetalText: An Efficient Text Instance Representation for Scene Text Detection](https://arxiv.org/abs/2107.05945)
+> Tao Sheng, Jie Chen, Zhouhui Lian
+> NeurIPS, 2021
+
+On the Total-Text dataset, the text detection result is as follows:
+
+|Model|Backbone|Configuration|Precision|Recall|Hmean|Download|
+| --- | --- | --- | --- | --- | --- | --- |
+|CT|ResNet18_vd|[configs/det/det_r18_vd_ct.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r18_vd_ct.yml)|88.68%|81.70%|85.05%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r18_ct_train.tar)|
+
+## 2. Environment
+
+Please prepare your environment referring to [prepare the environment](../../ppocr/environment.en.md) and [clone the repo](../../ppocr/blog/clone.en.md).
+
+## 3. Model Training / Evaluation / Prediction
+
+The above CT model is trained using the Total-Text text detection public dataset. For the download of the dataset, please refer to [Total-Text-Dataset](https://github.com/cs-chan/Total-Text-Dataset/tree/master/Dataset). PaddleOCR format annotation download link [train.txt](https://paddleocr.bj.bcebos.com/dataset/ct_tipc/train.txt), [test.txt](https://paddleocr.bj.bcebos.com/dataset/ct_tipc/test.txt).
+
+Please refer to [text detection training tutorial](../../ppocr/model_train/detection.en.md). PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different detection models.
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, convert the model saved in the CT text detection training process into an inference model. Taking the model based on the Resnet18_vd backbone network and trained on the Total Text English dataset as example ([model download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r18_ct_train.tar)), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r18_vd_ct.yml -o Global.pretrained_model=./det_r18_ct_train/best_accuracy  Global.save_inference_dir=./inference/det_ct
+```
+
+CT text detection model inference, you can execute the following command:
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img623.jpg" --det_model_dir="./inference/det_ct/" --det_algorithm="CT"
+```
+
+The visualized text detection results are saved to the `./inference_results` folder by default, and the name of the result file is prefixed with `det_res`. Examples of results are as follows:
+
+![img](./images/det_res_img623_ct.jpg)
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@inproceedings{sheng2021centripetaltext,
+    title={CentripetalText: An Efficient Text Instance Representation for Scene Text Detection},
+    author={Tao Sheng and Jie Chen and Zhouhui Lian},
+    booktitle={Thirty-Fifth Conference on Neural Information Processing Systems},
+    year={2021}
+}
+```
diff --git a/docs/version2.x/algorithm/text_detection/algorithm_det_ct.md b/docs/version2.x/algorithm/text_detection/algorithm_det_ct.md
new file mode 100644
index 0000000..3c6b7a1
--- /dev/null
+++ b/docs/version2.x/algorithm/text_detection/algorithm_det_ct.md
@@ -0,0 +1,74 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# CT
+
+## 1. 算法简介
+
+论文信息：
+> [CentripetalText: An Efficient Text Instance Representation for Scene Text Detection](https://arxiv.org/abs/2107.05945)
+> Tao Sheng, Jie Chen, Zhouhui Lian
+> NeurIPS, 2021
+
+在Total-Text文本检测公开数据集上，算法复现效果如下：
+
+|模型|骨干网络|配置文件|precision|recall|Hmean|下载链接|
+| --- | --- | --- | --- | --- | --- | --- |
+|CT|ResNet18_vd|[configs/det/det_r18_vd_ct.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r18_vd_ct.yml)|88.68%|81.70%|85.05%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r18_ct_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+CT模型使用Total-Text文本检测公开数据集训练得到，数据集下载可参考 [Total-Text-Dataset](https://github.com/cs-chan/Total-Text-Dataset/tree/master/Dataset), 我们将标签文件转成了paddleocr格式，转换好的标签文件下载参考[train.txt](https://paddleocr.bj.bcebos.com/dataset/ct_tipc/train.txt), [text.txt](https://paddleocr.bj.bcebos.com/dataset/ct_tipc/test.txt)。
+
+请参考[文本检测训练教程](../../ppocr/model_train/detection.md)。PaddleOCR对代码进行了模块化，训练不同的检测模型只需要**更换配置文件**即可。
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将CT文本检测训练过程中保存的模型，转换成inference model。以基于Resnet18_vd骨干网络，在Total-Text英文数据集训练的模型为例（ [模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r18_ct_train.tar) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r18_vd_ct.yml -o Global.pretrained_model=./det_r18_ct_train/best_accuracy  Global.save_inference_dir=./inference/det_ct
+```
+
+CT文本检测模型推理，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img623.jpg" --det_model_dir="./inference/det_ct/" --det_algorithm="CT"
+```
+
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为`det_res`。结果示例如下：
+
+![img](./images/det_res_img623_ct.jpg)
+
+### 4.2 C++推理
+
+暂不支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@inproceedings{sheng2021centripetaltext,
+    title={CentripetalText: An Efficient Text Instance Representation for Scene Text Detection},
+    author={Tao Sheng and Jie Chen and Zhouhui Lian},
+    booktitle={Thirty-Fifth Conference on Neural Information Processing Systems},
+    year={2021}
+}
+```
diff --git a/docs/version2.x/algorithm/text_detection/algorithm_det_db.en.md b/docs/version2.x/algorithm/text_detection/algorithm_det_db.en.md
new file mode 100644
index 0000000..80c042f
--- /dev/null
+++ b/docs/version2.x/algorithm/text_detection/algorithm_det_db.en.md
@@ -0,0 +1,98 @@
+---
+comments: true
+---
+
+# DB && DB++
+
+## 1. Introduction
+
+Paper:
+> [Real-time Scene Text Detection with Differentiable Binarization](https://arxiv.org/abs/1911.08947)
+> Liao, Minghui and Wan, Zhaoyi and Yao, Cong and Chen, Kai and Bai, Xiang
+> AAAI, 2020
+
+> [Real-Time Scene Text Detection with Differentiable Binarization and Adaptive Scale Fusion](https://arxiv.org/abs/2202.10304)
+> Liao, Minghui and Zou, Zhisheng and Wan, Zhaoyi and Yao, Cong and Bai, Xiang
+> TPAMI, 2022
+
+On the ICDAR2015 dataset, the text detection result is as follows:
+
+|Model|Backbone|Configuration|Precision|Recall|Hmean|Download|
+| --- | --- | --- | --- | --- | --- | --- |
+|DB|ResNet50_vd|[configs/det/det_r50_vd_db.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_vd_db.yml)|86.41%|78.72%|82.38%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)|
+|DB|MobileNetV3|[configs/det/det_mv3_db.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_mv3_db.yml)|77.29%|73.08%|75.12%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)|
+|DB++|ResNet50|[configs/det/det_r50_db++_icdar15.yml](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_db++_icdar15.yml)|90.89%|82.66%|86.58%|[pretrained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/ResNet50_dcn_asf_synthtext_pretrained.pdparams)/[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_db%2B%2B_icdar15_train.tar)|
+
+On the TD_TR dataset, the text detection result is as follows:
+
+|Model|Backbone|Configuration|Precision|Recall|Hmean|Download|
+| --- | --- | --- | --- | --- | --- | --- |
+|DB++|ResNet50|[configs/det/det_r50_db++_td_tr.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_db++_td_tr.yml)|92.92%|86.48%|89.58%|[pretrained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/ResNet50_dcn_asf_synthtext_pretrained.pdparams)/[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_db%2B%2B_td_tr_train.tar)|
+
+## 2. Environment
+
+Please prepare your environment referring to [prepare the environment](../../ppocr/environment.en.md) and [clone the repo](../../ppocr/blog/clone.en.md).
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [text detection training tutorial](../../ppocr/model_train/detection.en.md). PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different detection models.
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, convert the model saved in the DB text detection training process into an inference model. Taking the model based on the Resnet50_vd backbone network and trained on the ICDAR2015 English dataset as example ([model download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r50_vd_db.yml -o Global.pretrained_model=./det_r50_vd_db_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/det_db
+```
+
+DB text detection model inference, you can execute the following command:
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_db/"
+```
+
+The visualized text detection results are saved to the `./inference_results` folder by default, and the name of the result file is prefixed with `det_res`. Examples of results are as follows:
+
+![img](./images/det_res_img_10_db.jpg)
+
+**Note**: Since the ICDAR2015 dataset has only 1,000 training images, mainly for English scenes, the above model has very poor detection result on Chinese text images.
+
+### 4.2 C++ Inference
+
+With the inference model prepared, refer to the [cpp infer](../../../version2.x/legacy/cpp_infer.en.md) tutorial for C++ inference.
+
+### 4.3 Serving
+
+With the inference model prepared, refer to the [pdserving](../../../version2.x/legacy/paddle_server.en.md) tutorial for service deployment by Paddle Serving.
+
+### 4.4 More
+
+More deployment schemes supported for DB:
+
+- Paddle2ONNX: with the inference model prepared, please refer to the [paddle2onnx](../../../version2.x/legacy/paddle2onnx.en.md) tutorial.
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@inproceedings{liao2020real,
+  title={Real-time scene text detection with differentiable binarization},
+  author={Liao, Minghui and Wan, Zhaoyi and Yao, Cong and Chen, Kai and Bai, Xiang},
+  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
+  volume={34},
+  number={07},
+  pages={11474--11481},
+  year={2020}
+}
+
+@article{liao2022real,
+  title={Real-Time Scene Text Detection with Differentiable Binarization and Adaptive Scale Fusion},
+  author={Liao, Minghui and Zou, Zhisheng and Wan, Zhaoyi and Yao, Cong and Bai, Xiang},
+  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
+  year={2022},
+  publisher={IEEE}
+}
+```
diff --git a/docs/version2.x/algorithm/text_detection/algorithm_det_db.md b/docs/version2.x/algorithm/text_detection/algorithm_det_db.md
new file mode 100644
index 0000000..0c24204
--- /dev/null
+++ b/docs/version2.x/algorithm/text_detection/algorithm_det_db.md
@@ -0,0 +1,99 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# DB与DB++
+
+## 1. 算法简介
+
+论文信息：
+> [Real-time Scene Text Detection with Differentiable Binarization](https://arxiv.org/abs/1911.08947)
+> Liao, Minghui and Wan, Zhaoyi and Yao, Cong and Chen, Kai and Bai, Xiang
+> AAAI, 2020
+
+> [Real-Time Scene Text Detection with Differentiable Binarization and Adaptive Scale Fusion](https://arxiv.org/abs/2202.10304)
+> Liao, Minghui and Zou, Zhisheng and Wan, Zhaoyi and Yao, Cong and Bai, Xiang
+> TPAMI, 2022
+
+在ICDAR2015文本检测公开数据集上，算法复现效果如下：
+
+|模型|骨干网络|配置文件|precision|recall|Hmean|下载链接|
+| --- | --- | --- | --- | --- | --- | --- |
+|DB|ResNet50_vd|[configs/det/det_r50_vd_db.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_vd_db.yml)|86.41%|78.72%|82.38%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar)|
+|DB|MobileNetV3|[configs/det/det_mv3_db.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_mv3_db.yml)|77.29%|73.08%|75.12%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar)|
+|DB++|ResNet50|[configs/det/det_r50_db++_icdar15.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_db++_icdar15.yml)|90.89%|82.66%|86.58%|[合成数据预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/ResNet50_dcn_asf_synthtext_pretrained.pdparams)/[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_db%2B%2B_icdar15_train.tar)|
+
+在TD_TR文本检测公开数据集上，算法复现效果如下：
+
+|模型|骨干网络|配置文件|precision|recall|Hmean|下载链接|
+| --- | --- | --- | --- | --- | --- | --- |
+|DB++|ResNet50|[configs/det/det_r50_db++_td_tr.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_db++_td_tr.yml)|92.92%|86.48%|89.58%|[合成数据预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/ResNet50_dcn_asf_synthtext_pretrained.pdparams)/[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_db%2B%2B_td_tr_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+请参考[文本检测训练教程](../../ppocr/model_train/detection.md)。PaddleOCR对代码进行了模块化，训练不同的检测模型只需要**更换配置文件**即可。
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将DB文本检测训练过程中保存的模型，转换成inference model。以基于Resnet50_vd骨干网络，在ICDAR2015英文数据集训练的模型为例（ [模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r50_vd_db.yml -o Global.pretrained_model=./det_r50_vd_db_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/det_db
+```
+
+DB文本检测模型推理，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_db/" --det_algorithm="DB"
+```
+
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为`det_res`。结果示例如下：
+
+![img](./images/det_res_img_10_db.jpg)
+
+**注意**：由于ICDAR2015数据集只有1000张训练图像，且主要针对英文场景，所以上述模型对中文文本图像检测效果会比较差。
+
+### 4.2 C++推理
+
+准备好推理模型后，参考[cpp infer](../../../version2.x/legacy/cpp_infer.md)教程进行操作即可。
+
+### 4.3 Serving服务化部署
+
+准备好推理模型后，参考[pdserving](../../../version2.x/legacy/paddle_server.md)教程进行Serving服务化部署，包括Python Serving和C++ Serving两种模式。
+
+### 4.4 更多推理部署
+
+DB模型还支持以下推理部署方式：
+
+- Paddle2ONNX推理：准备好推理模型后，参考[paddle2onnx](../../../version2.x/legacy/paddle2onnx.md)教程操作。
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@inproceedings{liao2020real,
+  title={Real-time scene text detection with differentiable binarization},
+  author={Liao, Minghui and Wan, Zhaoyi and Yao, Cong and Chen, Kai and Bai, Xiang},
+  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
+  volume={34},
+  number={07},
+  pages={11474--11481},
+  year={2020}
+}
+
+@article{liao2022real,
+  title={Real-Time Scene Text Detection with Differentiable Binarization and Adaptive Scale Fusion},
+  author={Liao, Minghui and Zou, Zhisheng and Wan, Zhaoyi and Yao, Cong and Bai, Xiang},
+  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
+  year={2022},
+  publisher={IEEE}
+}
+```
diff --git a/docs/version2.x/algorithm/text_detection/algorithm_det_drrg.en.md b/docs/version2.x/algorithm/text_detection/algorithm_det_drrg.en.md
new file mode 100644
index 0000000..d6def85
--- /dev/null
+++ b/docs/version2.x/algorithm/text_detection/algorithm_det_drrg.en.md
@@ -0,0 +1,61 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# DRRG
+
+## 1. Introduction
+
+Paper:
+> [Deep Relational Reasoning Graph Network for Arbitrary Shape Text Detection](https://arxiv.org/abs/2003.07493)
+> Zhang, Shi-Xue and Zhu, Xiaobin and Hou, Jie-Bo and Liu, Chang and Yang, Chun and Wang, Hongfa and Yin, Xu-Cheng
+> CVPR, 2020
+
+On the CTW1500 dataset, the text detection result is as follows:
+
+|Model|Backbone|Configuration|Precision|Recall|Hmean|Download|
+| --- | --- | --- | --- | --- | --- | --- |
+| DRRG | ResNet50_vd | [configs/det/det_r50_drrg_ctw.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_drrg_ctw.yml)| 89.92%|80.91%|85.18%|[trained model](https://paddleocr.bj.bcebos.com/contribution/det_r50_drrg_ctw_train.tar)|
+
+## 2. Environment
+
+Please prepare your environment referring to [prepare the environment](../../ppocr/environment.en.md) and [clone the repo](../../ppocr/blog/clone.en.md).
+
+## 3. Model Training / Evaluation / Prediction
+
+The above DRRG model is trained using the CTW1500 text detection public dataset. For the download of the dataset, please refer to [ocr_datasets](../../../datasets/ocr_datasets.en.md).
+
+After the data download is complete, please refer to [Text Detection Training Tutorial](../../ppocr/model_train/detection.en.md) for training. PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different detection models.
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+Since the model needs to be converted to Numpy data for many times in the forward, DRRG dynamic graph to static graph is not supported.
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@inproceedings{zhang2020deep,
+  title={Deep relational reasoning graph network for arbitrary shape text detection},
+  author={Zhang, Shi-Xue and Zhu, Xiaobin and Hou, Jie-Bo and Liu, Chang and Yang, Chun and Wang, Hongfa and Yin, Xu-Cheng},
+  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
+  pages={9699--9708},
+  year={2020}
+}
+```
diff --git a/docs/version2.x/algorithm/text_detection/algorithm_det_drrg.md b/docs/version2.x/algorithm/text_detection/algorithm_det_drrg.md
new file mode 100644
index 0000000..3603db6
--- /dev/null
+++ b/docs/version2.x/algorithm/text_detection/algorithm_det_drrg.md
@@ -0,0 +1,61 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# DRRG
+
+## 1. 算法简介
+
+论文信息：
+> [Deep Relational Reasoning Graph Network for Arbitrary Shape Text Detection](https://arxiv.org/abs/2003.07493)
+> Zhang, Shi-Xue and Zhu, Xiaobin and Hou, Jie-Bo and Liu, Chang and Yang, Chun and Wang, Hongfa and Yin, Xu-Cheng
+> CVPR, 2020
+
+在CTW1500文本检测公开数据集上，算法复现效果如下：
+
+| 模型  |骨干网络|配置文件|precision|recall|Hmean|下载链接|
+|-----| --- | --- | --- | --- | --- | --- |
+| DRRG | ResNet50_vd | [configs/det/det_r50_drrg_ctw.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_drrg_ctw.yml)| 89.92%|80.91%|85.18%|[训练模型](https://paddleocr.bj.bcebos.com/contribution/det_r50_drrg_ctw_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+上述DRRG模型使用CTW1500文本检测公开数据集训练得到，数据集下载可参考 [ocr_datasets](../../../datasets/ocr_datasets.md)。
+
+数据下载完成后，请参考[文本检测训练教程](../../ppocr/model_train/detection.md)进行训练。PaddleOCR对代码进行了模块化，训练不同的检测模型只需要**更换配置文件**即可。
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+由于模型前向运行时需要多次转换为Numpy数据进行运算，因此DRRG的动态图转静态图暂未支持。
+
+### 4.2 C++推理
+
+暂未支持
+
+### 4.3 Serving服务化部署
+
+暂未支持
+
+### 4.4 更多推理部署
+
+暂未支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@inproceedings{zhang2020deep,
+  title={Deep relational reasoning graph network for arbitrary shape text detection},
+  author={Zhang, Shi-Xue and Zhu, Xiaobin and Hou, Jie-Bo and Liu, Chang and Yang, Chun and Wang, Hongfa and Yin, Xu-Cheng},
+  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
+  pages={9699--9708},
+  year={2020}
+}
+```
diff --git a/docs/version2.x/algorithm/text_detection/algorithm_det_east.en.md b/docs/version2.x/algorithm/text_detection/algorithm_det_east.en.md
new file mode 100644
index 0000000..b4f40c2
--- /dev/null
+++ b/docs/version2.x/algorithm/text_detection/algorithm_det_east.en.md
@@ -0,0 +1,76 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# EAST
+
+## 1. Introduction
+
+Paper:
+> [EAST: An Efficient and Accurate Scene Text Detector](https://arxiv.org/abs/1704.03155)
+> Xinyu Zhou, Cong Yao, He Wen, Yuzhi Wang, Shuchang Zhou, Weiran He, Jiajun Liang
+> CVPR, 2017
+
+On the ICDAR2015 dataset, the text detection result is as follows:
+
+|Model|Backbone|Configuration|Precision|Recall|Hmean|Download|
+| --- | --- | --- | --- | --- | --- | --- |
+|EAST|ResNet50_vd| [det_r50_vd_east.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_vd_east.yml)|88.71%|    81.36%|    84.88%|    [model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar)|
+|EAST|MobileNetV3|[det_mv3_east.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_mv3_east.yml) | 78.20%|    79.10%|    78.65%|    [model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_east_v2.0_train.tar)|
+
+## 2. Environment
+
+Please prepare your environment referring to [prepare the environment](../../ppocr/environment.en.md) and [clone the repo](../../ppocr/blog/clone.en.md).
+
+## 3. Model Training / Evaluation / Prediction
+
+The above EAST model is trained using the ICDAR2015 text detection public dataset. For the download of the dataset, please refer to [ocr_datasets](../../../datasets/ocr_datasets.en.md).
+
+After the data download is complete, please refer to [Text Detection Training Tutorial](../../ppocr/model_train/detection.en.md) for training. PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different detection models.
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, convert the model saved in the EAST text detection training process into an inference model. Taking the model based on the Resnet50_vd backbone network and trained on the ICDAR2015 English dataset as example ([model download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar)), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r50_vd_east.yml -o Global.pretrained_model=./det_r50_vd_east_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/det_r50_east/
+```
+
+For EAST text detection model inference, you need to set the parameter --det_algorithm="EAST", run the following command:
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_r50_east/" --det_algorithm="EAST"
+```
+
+The visualized text detection results are saved to the `./inference_results` folder by default, and the name of the result file is prefixed with `det_res`.
+
+![img](./images/det_res_img_10_east.jpg)
+
+### 4.2 C++ Inference
+
+Since the post-processing is not written in CPP, the EAST text detection model does not support CPP inference.
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@inproceedings{zhou2017east,
+  title={East: an efficient and accurate scene text detector},
+  author={Zhou, Xinyu and Yao, Cong and Wen, He and Wang, Yuzhi and Zhou, Shuchang and He, Weiran and Liang, Jiajun},
+  booktitle={Proceedings of the IEEE conference on Computer Vision and Pattern Recognition},
+  pages={5551--5560},
+  year={2017}
+}
+```
diff --git a/docs/version2.x/algorithm/text_detection/algorithm_det_east.md b/docs/version2.x/algorithm/text_detection/algorithm_det_east.md
new file mode 100644
index 0000000..7a4d612
--- /dev/null
+++ b/docs/version2.x/algorithm/text_detection/algorithm_det_east.md
@@ -0,0 +1,76 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# EAST
+
+## 1. 算法简介
+
+论文信息：
+> [EAST: An Efficient and Accurate Scene Text Detector](https://arxiv.org/abs/1704.03155)
+> Xinyu Zhou, Cong Yao, He Wen, Yuzhi Wang, Shuchang Zhou, Weiran He, Jiajun Liang
+> CVPR, 2017
+
+在ICDAR2015文本检测公开数据集上，算法复现效果如下：
+
+|模型|骨干网络|配置文件|precision|recall|Hmean|下载链接|
+| --- | --- | --- | --- | --- | --- | --- |
+|EAST|ResNet50_vd| [det_r50_vd_east.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_vd_east.yml)|88.71%|    81.36%|    84.88%|    [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar)|
+|EAST|MobileNetV3|[det_mv3_east.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_mv3_east.yml) | 78.20%|    79.10%|    78.65%|    [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_east_v2.0_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+上表中的EAST训练模型使用ICDAR2015文本检测公开数据集训练得到，数据集下载可参考 [ocr_datasets](../../../datasets/ocr_datasets.md)。
+
+数据下载完成后，请参考[文本检测训练教程](../../ppocr/model_train/detection.md)进行训练。PaddleOCR对代码进行了模块化，训练不同的检测模型只需要**更换配置文件**即可。
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将EAST文本检测训练过程中保存的模型，转换成inference model。以基于Resnet50_vd骨干网络，在ICDAR2015英文数据集训练的模型为例（[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar))，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r50_vd_east.yml -o Global.pretrained_model=./det_r50_vd_east_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/det_r50_east/
+```
+
+EAST文本检测模型推理，需要设置参数--det_algorithm="EAST"，执行预测：
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_r50_east/" --det_algorithm="EAST"
+```
+
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为`det_res`。
+
+![img](./images/det_res_img_10_east.jpg)
+
+### 4.2 C++推理
+
+由于后处理暂未使用CPP编写，EAST文本检测模型暂不支持CPP推理。
+
+### 4.3 Serving服务化部署
+
+暂未支持
+
+### 4.4 更多推理部署
+
+暂未支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@inproceedings{zhou2017east,
+  title={East: an efficient and accurate scene text detector},
+  author={Zhou, Xinyu and Yao, Cong and Wen, He and Wang, Yuzhi and Zhou, Shuchang and He, Weiran and Liang, Jiajun},
+  booktitle={Proceedings of the IEEE conference on Computer Vision and Pattern Recognition},
+  pages={5551--5560},
+  year={2017}
+}
+```
diff --git a/docs/version2.x/algorithm/text_detection/algorithm_det_fcenet.en.md b/docs/version2.x/algorithm/text_detection/algorithm_det_fcenet.en.md
new file mode 100644
index 0000000..dfa92f9
--- /dev/null
+++ b/docs/version2.x/algorithm/text_detection/algorithm_det_fcenet.en.md
@@ -0,0 +1,86 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# FCENet
+
+## 1. Introduction
+
+Paper:
+> [Fourier Contour Embedding for Arbitrary-Shaped Text Detection](https://arxiv.org/abs/2104.10442)
+> Yiqin Zhu and Jianyong Chen and Lingyu Liang and Zhanghui Kuang and Lianwen Jin and Wayne Zhang
+> CVPR, 2021
+
+On the CTW1500 dataset, the text detection result is as follows:
+
+|Model|Backbone|Configuration|Precision|Recall|Hmean|Download|
+| --- | --- | --- | --- | --- | --- | --- |
+| FCE | ResNet50_dcn | [configs/det/det_r50_vd_dcn_fce_ctw.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_vd_dcn_fce_ctw.yml)| 88.39%|82.18%|85.27%|[trained model](https://paddleocr.bj.bcebos.com/contribution/det_r50_dcn_fce_ctw_v2.0_train.tar)|
+
+## 2. Environment
+
+Please prepare your environment referring to [prepare the environment](../../ppocr/environment.en.md) and [clone the repo](../../ppocr/blog/clone.en.md).
+
+## 3. Model Training / Evaluation / Prediction
+
+The above FCE model is trained using the CTW1500 text detection public dataset. For the download of the dataset, please refer to [ocr_datasets](../../../datasets/ocr_datasets.en.md).
+
+After the data download is complete, please refer to [Text Detection Training Tutorial](../../ppocr/model_train/detection.en.md) for training. PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different detection models.
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, convert the model saved in the FCE text detection training process into an inference model. Taking the model based on the Resnet50_vd_dcn backbone network and trained on the CTW1500 English dataset as example ([model download link](https://paddleocr.bj.bcebos.com/contribution/det_r50_dcn_fce_ctw_v2.0_train.tar)), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r50_vd_dcn_fce_ctw.yml -o Global.pretrained_model=./det_r50_dcn_fce_ctw_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/det_fce
+```
+
+FCE text detection model inference, to perform non-curved text detection, you can run the following commands:
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_fce/" --det_algorithm="FCE" --det_fce_box_type=quad
+```
+
+The visualized text detection results are saved to the `./inference_results` folder by default, and the name of the result file is prefixed with 'det_res'. Examples of results are as follows:
+
+![img](./images/det_res_img_10_fce.jpg)
+
+If you want to perform curved text detection, you can execute the following command:
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img623.jpg" --det_model_dir="./inference/det_fce/" --det_algorithm="FCE" --det_fce_box_type=poly
+```
+
+The visualized text detection results are saved to the `./inference_results` folder by default, and the name of the result file is prefixed with 'det_res'. Examples of results are as follows:
+
+![img](./images/det_res_img623_fce.jpg)
+
+**Note**: Since the CTW1500 dataset has only 1,000 training images, mainly for English scenes, the above model has very poor detection result on Chinese or curved text images.
+
+### 4.2 C++ Inference
+
+Since the post-processing is not written in CPP, the FCE text detection model does not support CPP inference.
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@InProceedings{zhu2021fourier,
+  title={Fourier Contour Embedding for Arbitrary-Shaped Text Detection},
+  author={Yiqin Zhu and Jianyong Chen and Lingyu Liang and Zhanghui Kuang and Lianwen Jin and Wayne Zhang},
+  year={2021},
+  booktitle = {CVPR}
+}
+```
diff --git a/docs/version2.x/algorithm/text_detection/algorithm_det_fcenet.md b/docs/version2.x/algorithm/text_detection/algorithm_det_fcenet.md
new file mode 100644
index 0000000..86b6a7c
--- /dev/null
+++ b/docs/version2.x/algorithm/text_detection/algorithm_det_fcenet.md
@@ -0,0 +1,86 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# FCENet
+
+## 1. 算法简介
+
+论文信息：
+> [Fourier Contour Embedding for Arbitrary-Shaped Text Detection](https://arxiv.org/abs/2104.10442)
+> Yiqin Zhu and Jianyong Chen and Lingyu Liang and Zhanghui Kuang and Lianwen Jin and Wayne Zhang
+> CVPR, 2021
+
+在CTW1500文本检测公开数据集上，算法复现效果如下：
+
+| 模型  |骨干网络|配置文件|precision|recall|Hmean|下载链接|
+|-----| --- | --- | --- | --- | --- | --- |
+| FCE | ResNet50_dcn | [configs/det/det_r50_vd_dcn_fce_ctw.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_vd_dcn_fce_ctw.yml)| 88.39%|82.18%|85.27%|[训练模型](https://paddleocr.bj.bcebos.com/contribution/det_r50_dcn_fce_ctw_v2.0_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+上述FCE模型使用CTW1500文本检测公开数据集训练得到，数据集下载可参考 [ocr_datasets](../../../datasets/ocr_datasets.md)。
+
+数据下载完成后，请参考[文本检测训练教程](../../ppocr/model_train/detection.md)进行训练。PaddleOCR对代码进行了模块化，训练不同的检测模型只需要**更换配置文件**即可。
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将FCE文本检测训练过程中保存的模型，转换成inference model。以基于Resnet50_vd_dcn骨干网络，在CTW1500英文数据集训练的模型为例（ [模型下载地址](https://paddleocr.bj.bcebos.com/contribution/det_r50_dcn_fce_ctw_v2.0_train.tar) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r50_vd_dcn_fce_ctw.yml -o Global.pretrained_model=./det_r50_dcn_fce_ctw_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/det_fce
+```
+
+FCE文本检测模型推理，执行非弯曲文本检测，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_fce/" --det_algorithm="FCE" --det_fce_box_type=quad
+```
+
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为'det_res'。结果示例如下：
+
+![img](./images/det_res_img_10_fce.jpg)
+
+如果想执行弯曲文本检测，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img623.jpg" --det_model_dir="./inference/det_fce/" --det_algorithm="FCE" --det_fce_box_type=poly
+```
+
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为'det_res'。结果示例如下：
+
+![img](./images/det_res_img623_fce.jpg)
+
+**注意**：由于CTW1500数据集只有1000张训练图像，且主要针对英文场景，所以上述模型对中文文本图像检测效果会比较差。
+
+### 4.2 C++推理
+
+由于后处理暂未使用CPP编写，FCE文本检测模型暂不支持CPP推理。
+
+### 4.3 Serving服务化部署
+
+暂未支持
+
+### 4.4 更多推理部署
+
+暂未支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@InProceedings{zhu2021fourier,
+  title={Fourier Contour Embedding for Arbitrary-Shaped Text Detection},
+  author={Yiqin Zhu and Jianyong Chen and Lingyu Liang and Zhanghui Kuang and Lianwen Jin and Wayne Zhang},
+  year={2021},
+  booktitle = {CVPR}
+}
+```
diff --git a/docs/version2.x/algorithm/text_detection/algorithm_det_psenet.en.md b/docs/version2.x/algorithm/text_detection/algorithm_det_psenet.en.md
new file mode 100644
index 0000000..3de8561
--- /dev/null
+++ b/docs/version2.x/algorithm/text_detection/algorithm_det_psenet.en.md
@@ -0,0 +1,89 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+
+# PSENet
+
+## 1. Introduction
+
+Paper:
+> [Shape robust text detection with progressive scale expansion network](https://arxiv.org/abs/1903.12473)
+> Wang, Wenhai and Xie, Enze and Li, Xiang and Hou, Wenbo and Lu, Tong and Yu, Gang and Shao, Shuai
+> CVPR, 2019
+
+On the ICDAR2015 dataset, the text detection result is as follows:
+
+|Model|Backbone|Configuration|Precision|Recall|Hmean|Download|
+| --- | --- | --- | --- | --- | --- | --- |
+|PSE| ResNet50_vd | [configs/det/det_r50_vd_pse.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_vd_pse.yml)| 85.81%    |79.53%|82.55%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_vd_pse_v2.0_train.tar)|
+|PSE| MobileNetV3| [configs/det/det_mv3_pse.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_mv3_pse.yml) | 82.20%    |70.48%|75.89%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_mv3_pse_v2.0_train.tar)|
+
+## 2. Environment
+
+Please prepare your environment referring to [prepare the environment](../../ppocr/environment.en.md) and [clone the repo](../../ppocr/blog/clone.en.md).
+
+## 3. Model Training / Evaluation / Prediction
+
+The above PSE model is trained using the ICDAR2015 text detection public dataset. For the download of the dataset, please refer to [ocr_datasets](../../../datasets/ocr_datasets.en.md).
+
+After the data download is complete, please refer to [Text Detection Training Tutorial](../../ppocr/model_train/detection.en.md) for training. PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different detection models.
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, convert the model saved in the PSE text detection training process into an inference model. Taking the model based on the Resnet50_vd backbone network and trained on the ICDAR2015 English dataset as example ([model download link](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_vd_pse_v2.0_train.tar)), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r50_vd_pse.yml -o Global.pretrained_model=./det_r50_vd_pse_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/det_pse
+```
+
+PSE text detection model inference, to perform non-curved text detection, you can run the following commands:
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_pse/" --det_algorithm="PSE" --det_pse_box_type=quad
+```
+
+The visualized text detection results are saved to the `./inference_results` folder by default, and the name of the result file is prefixed with 'det_res'. Examples of results are as follows:
+
+![img](./images/det_res_img_10_pse.jpg)
+
+If you want to perform curved text detection, you can execute the following command:
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_pse/" --det_algorithm="PSE" --det_pse_box_type=poly
+```
+
+The visualized text detection results are saved to the `./inference_results` folder by default, and the name of the result file is prefixed with 'det_res'. Examples of results are as follows:
+
+![](./images/det_res_img_10_pse_poly.jpg)
+
+**Note**: Since the ICDAR2015 dataset has only 1,000 training images, mainly for English scenes, the above model has very poor detection result on Chinese or curved text images.
+
+### 4.2 C++ Inference
+
+Since the post-processing is not written in CPP, the PSE text detection model does not support CPP inference.
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@inproceedings{wang2019shape,
+  title={Shape robust text detection with progressive scale expansion network},
+  author={Wang, Wenhai and Xie, Enze and Li, Xiang and Hou, Wenbo and Lu, Tong and Yu, Gang and Shao, Shuai},
+  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
+  pages={9336--9345},
+  year={2019}
+}
+```
diff --git a/docs/version2.x/algorithm/text_detection/algorithm_det_psenet.md b/docs/version2.x/algorithm/text_detection/algorithm_det_psenet.md
new file mode 100644
index 0000000..b3082ba
--- /dev/null
+++ b/docs/version2.x/algorithm/text_detection/algorithm_det_psenet.md
@@ -0,0 +1,88 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# PSENet
+
+## 1. 算法简介
+
+论文信息：
+> [Shape robust text detection with progressive scale expansion network](https://arxiv.org/abs/1903.12473)
+> Wang, Wenhai and Xie, Enze and Li, Xiang and Hou, Wenbo and Lu, Tong and Yu, Gang and Shao, Shuai
+> CVPR, 2019
+
+在ICDAR2015文本检测公开数据集上，算法复现效果如下：
+
+|模型|骨干网络|配置文件|precision|recall|Hmean|下载链接|
+| --- | --- | --- | --- | --- | --- | --- |
+|PSE| ResNet50_vd | [configs/det/det_r50_vd_pse.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_vd_pse.yml)| 85.81%    |79.53%|82.55%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_vd_pse_v2.0_train.tar)|
+|PSE| MobileNetV3| [configs/det/det_mv3_pse.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_mv3_pse.yml) | 82.20%    |70.48%|75.89%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_mv3_pse_v2.0_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+上述PSE模型使用ICDAR2015文本检测公开数据集训练得到，数据集下载可参考 [ocr_datasets](../../../datasets/ocr_datasets.md)。
+
+数据下载完成后，请参考[文本检测训练教程](../../ppocr/model_train/detection.md)进行训练。PaddleOCR对代码进行了模块化，训练不同的检测模型只需要**更换配置文件**即可。
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将PSE文本检测训练过程中保存的模型，转换成inference model。以基于Resnet50_vd骨干网络，在ICDAR2015英文数据集训练的模型为例（ [模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_vd_pse_v2.0_train.tar) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r50_vd_pse.yml -o Global.pretrained_model=./det_r50_vd_pse_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/det_pse
+```
+
+PSE文本检测模型推理，执行非弯曲文本检测，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_pse/" --det_algorithm="PSE" --det_pse_box_type=quad
+```
+
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为'det_res'。结果示例如下：
+
+![img](./images/det_res_img_10_pse.jpg)
+
+如果想执行弯曲文本检测，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_pse/" --det_algorithm="PSE" --det_pse_box_type=poly
+```
+
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为'det_res'。结果示例如下：
+
+![img](./images/det_res_img_10_pse_poly.jpg)
+
+**注意**：由于ICDAR2015数据集只有1000张训练图像，且主要针对英文场景，所以上述模型对中文或弯曲文本图像检测效果会比较差。
+
+### 4.2 C++推理
+
+由于后处理暂未使用CPP编写，PSE文本检测模型暂不支持CPP推理。
+
+### 4.3 Serving服务化部署
+
+暂未支持
+
+### 4.4 更多推理部署
+
+暂未支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@inproceedings{wang2019shape,
+  title={Shape robust text detection with progressive scale expansion network},
+  author={Wang, Wenhai and Xie, Enze and Li, Xiang and Hou, Wenbo and Lu, Tong and Yu, Gang and Shao, Shuai},
+  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
+  pages={9336--9345},
+  year={2019}
+}
+```
diff --git a/docs/version2.x/algorithm/text_detection/algorithm_det_sast.en.md b/docs/version2.x/algorithm/text_detection/algorithm_det_sast.en.md
new file mode 100644
index 0000000..afd0eb5
--- /dev/null
+++ b/docs/version2.x/algorithm/text_detection/algorithm_det_sast.en.md
@@ -0,0 +1,101 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# SAST
+
+## 1. Introduction
+
+Paper:
+> [A Single-Shot Arbitrarily-Shaped Text Detector based on Context Attended Multi-Task Learning](https://arxiv.org/abs/1908.05498)
+> Wang, Pengfei and Zhang, Chengquan and Qi, Fei and Huang, Zuming and En, Mengyi and Han, Junyu and Liu, Jingtuo and Ding, Errui and Shi, Guangming
+> ACM MM, 2019
+
+On the ICDAR2015 dataset, the text detection result is as follows:
+
+|Model|Backbone|Configuration|Precision|Recall|Hmean|Download|
+| --- | --- | --- | --- | --- | --- | --- |
+|SAST|ResNet50_vd|[configs/det/det_r50_vd_sast_icdar15.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_vd_sast_icdar15.yml)|91.39%|83.77%|87.42%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar)|
+
+On the Total-text dataset, the text detection result is as follows:
+
+|Model|Backbone|Configuration|Precision|Recall|Hmean|Download|
+| --- | --- | --- | --- | --- | --- | --- |
+|SAST|ResNet50_vd|[configs/det/det_r50_vd_sast_totaltext.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_vd_sast_totaltext.yml)|89.63%|78.44%|83.66%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_totaltext_v2.0_train.tar)|
+
+## 2. Environment
+
+Please prepare your environment referring to [prepare the environment](../../ppocr/environment.en.md) and [clone the repo](../../ppocr/blog/clone.en.md).
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [text detection training tutorial](../../ppocr/model_train/detection.en.md). PaddleOCR has modularized the code structure, so that you only need to **replace the configuration file** to train different detection models.
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+#### (1). Quadrangle text detection model (ICDAR2015)
+
+First, convert the model saved in the SAST text detection training process into an inference model. Taking the model based on the Resnet50_vd backbone network and trained on the ICDAR2015 English dataset as an example ([model download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar)), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r50_vd_sast_icdar15.yml -o Global.pretrained_model=./det_r50_vd_sast_icdar15_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/det_sast_ic15
+```
+
+**For SAST quadrangle text detection model inference, you need to set the parameter `--det_algorithm="SAST"`**, run the following command:
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --det_algorithm="SAST" --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_sast_ic15/"
+```
+
+The visualized text detection results are saved to the `./inference_results` folder by default, and the name of the result file is prefixed with 'det_res'. Examples of results are as follows:
+
+![img](./images/det_res_img_10_sast.jpg)
+
+#### (2). Curved text detection model (Total-Text)
+
+First, convert the model saved in the SAST text detection training process into an inference model. Taking the model based on the Resnet50_vd backbone network and trained on the Total-Text English dataset as an example ([model download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_totaltext_v2.0_train.tar)), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r50_vd_sast_totaltext.yml -o Global.pretrained_model=./det_r50_vd_sast_totaltext_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/det_sast_tt
+```
+
+For SAST curved text detection model inference, you need to set the parameter `--det_algorithm="SAST"` and `--det_box_type=poly`, run the following command:
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --det_algorithm="SAST" --image_dir="./doc/imgs_en/img623.jpg" --det_model_dir="./inference/det_sast_tt/" --det_box_type='poly'
+```
+
+The visualized text detection results are saved to the `./inference_results` folder by default, and the name of the result file is prefixed with 'det_res'. Examples of results are as follows:
+
+![img](./images/det_res_img623_sast.jpg)
+
+**Note**: SAST post-processing locality aware NMS has two versions: Python and C++. The speed of C++ version is obviously faster than that of Python version. Due to the compilation version problem of NMS of C++ version, C++ version NMS will be called only in Python 3.5 environment, and python version NMS will be called in other cases.
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@inproceedings{wang2019single,
+  title={A Single-Shot Arbitrarily-Shaped Text Detector based on Context Attended Multi-Task Learning},
+  author={Wang, Pengfei and Zhang, Chengquan and Qi, Fei and Huang, Zuming and En, Mengyi and Han, Junyu and Liu, Jingtuo and Ding, Errui and Shi, Guangming},
+  booktitle={Proceedings of the 27th ACM International Conference on Multimedia},
+  pages={1277--1285},
+  year={2019}
+}
+```
diff --git a/docs/version2.x/algorithm/text_detection/algorithm_det_sast.md b/docs/version2.x/algorithm/text_detection/algorithm_det_sast.md
new file mode 100644
index 0000000..02ac017
--- /dev/null
+++ b/docs/version2.x/algorithm/text_detection/algorithm_det_sast.md
@@ -0,0 +1,101 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# SAST
+
+## 1. 算法简介
+
+论文信息：
+> [A Single-Shot Arbitrarily-Shaped Text Detector based on Context Attended Multi-Task Learning](https://arxiv.org/abs/1908.05498)
+> Wang, Pengfei and Zhang, Chengquan and Qi, Fei and Huang, Zuming and En, Mengyi and Han, Junyu and Liu, Jingtuo and Ding, Errui and Shi, Guangming
+> ACM MM, 2019
+
+在ICDAR2015文本检测公开数据集上，算法复现效果如下：
+
+|模型|骨干网络|配置文件|precision|recall|Hmean|下载链接|
+| --- | --- | --- | --- | --- | --- | --- |
+|SAST|ResNet50_vd|[configs/det/det_r50_vd_sast_icdar15.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_vd_sast_icdar15.yml)|91.39%|83.77%|87.42%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar)|
+
+在Total-text文本检测公开数据集上，算法复现效果如下：
+
+|模型|骨干网络|配置文件|precision|recall|Hmean|下载链接|
+| --- | --- | --- | --- | --- | --- | --- |
+|SAST|ResNet50_vd|[configs/det/det_r50_vd_sast_totaltext.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/det_r50_vd_sast_totaltext.yml)|89.63%|78.44%|83.66%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_totaltext_v2.0_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+请参考[文本检测训练教程](../../ppocr/model_train/detection.md)。PaddleOCR对代码进行了模块化，训练不同的检测模型只需要**更换配置文件**即可。
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+#### (1). 四边形文本检测模型（ICDAR2015）
+
+首先将SAST文本检测训练过程中保存的模型，转换成inference model。以基于Resnet50_vd骨干网络，在ICDAR2015英文数据集训练的模型为例([模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar))，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r50_vd_sast_icdar15.yml -o Global.pretrained_model=./det_r50_vd_sast_icdar15_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/det_sast_ic15
+```
+
+**SAST文本检测模型推理，需要设置参数`--det_algorithm="SAST"`**，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --det_algorithm="SAST" --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_sast_ic15/"
+```
+
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为'det_res'。结果示例如下：
+
+![img](./images/det_res_img_10_sast.jpg)
+
+#### (2). 弯曲文本检测模型（Total-Text）
+
+首先将SAST文本检测训练过程中保存的模型，转换成inference model。以基于Resnet50_vd骨干网络，在Total-Text英文数据集训练的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_totaltext_v2.0_train.tar))，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r50_vd_sast_totaltext.yml -o Global.pretrained_model=./det_r50_vd_sast_totaltext_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/det_sast_tt
+```
+
+SAST文本检测模型推理，需要设置参数`--det_algorithm="SAST"`，同时，还需要增加参数`--det_box_type=poly`，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --det_algorithm="SAST" --image_dir="./doc/imgs_en/img623.jpg" --det_model_dir="./inference/det_sast_tt/" --det_box_type='poly'
+```
+
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为'det_res'。结果示例如下：
+
+![img](./images/det_res_img623_sast.jpg)
+
+**注意**：本代码库中，SAST后处理Locality-Aware NMS有python和c++两种版本，c++版速度明显快于python版。由于c++版本nms编译版本问题，只有python3.5环境下会调用c++版nms，其他情况将调用python版nms。
+
+### 4.2 C++推理
+
+暂未支持
+
+### 4.3 Serving服务化部署
+
+暂未支持
+
+### 4.4 更多推理部署
+
+暂未支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@inproceedings{wang2019single,
+  title={A Single-Shot Arbitrarily-Shaped Text Detector based on Context Attended Multi-Task Learning},
+  author={Wang, Pengfei and Zhang, Chengquan and Qi, Fei and Huang, Zuming and En, Mengyi and Han, Junyu and Liu, Jingtuo and Ding, Errui and Shi, Guangming},
+  booktitle={Proceedings of the 27th ACM International Conference on Multimedia},
+  pages={1277--1285},
+  year={2019}
+}
+```
diff --git a/docs/version2.x/algorithm/text_detection/images/det_res_img623_ct.jpg b/docs/version2.x/algorithm/text_detection/images/det_res_img623_ct.jpg
new file mode 100644
index 0000000..2c5f57d
Binary files /dev/null and b/docs/version2.x/algorithm/text_detection/images/det_res_img623_ct.jpg differ
diff --git a/docs/version2.x/algorithm/text_detection/images/det_res_img623_fce.jpg b/docs/version2.x/algorithm/text_detection/images/det_res_img623_fce.jpg
new file mode 100644
index 0000000..938ae4c
Binary files /dev/null and b/docs/version2.x/algorithm/text_detection/images/det_res_img623_fce.jpg differ
diff --git a/docs/version2.x/algorithm/text_detection/images/det_res_img623_sast.jpg b/docs/version2.x/algorithm/text_detection/images/det_res_img623_sast.jpg
new file mode 100644
index 0000000..af5e2d6
Binary files /dev/null and b/docs/version2.x/algorithm/text_detection/images/det_res_img623_sast.jpg differ
diff --git a/docs/version2.x/algorithm/text_detection/images/det_res_img_10_db.jpg b/docs/version2.x/algorithm/text_detection/images/det_res_img_10_db.jpg
new file mode 100644
index 0000000..6af89f6
Binary files /dev/null and b/docs/version2.x/algorithm/text_detection/images/det_res_img_10_db.jpg differ
diff --git a/docs/version2.x/algorithm/text_detection/images/det_res_img_10_east.jpg b/docs/version2.x/algorithm/text_detection/images/det_res_img_10_east.jpg
new file mode 100644
index 0000000..908d077
Binary files /dev/null and b/docs/version2.x/algorithm/text_detection/images/det_res_img_10_east.jpg differ
diff --git a/docs/version2.x/algorithm/text_detection/images/det_res_img_10_fce.jpg b/docs/version2.x/algorithm/text_detection/images/det_res_img_10_fce.jpg
new file mode 100644
index 0000000..fb32950
Binary files /dev/null and b/docs/version2.x/algorithm/text_detection/images/det_res_img_10_fce.jpg differ
diff --git a/docs/version2.x/algorithm/text_detection/images/det_res_img_10_pse.jpg b/docs/version2.x/algorithm/text_detection/images/det_res_img_10_pse.jpg
new file mode 100644
index 0000000..cdb7625
Binary files /dev/null and b/docs/version2.x/algorithm/text_detection/images/det_res_img_10_pse.jpg differ
diff --git a/docs/version2.x/algorithm/text_detection/images/det_res_img_10_pse_poly.jpg b/docs/version2.x/algorithm/text_detection/images/det_res_img_10_pse_poly.jpg
new file mode 100644
index 0000000..9c06a17
Binary files /dev/null and b/docs/version2.x/algorithm/text_detection/images/det_res_img_10_pse_poly.jpg differ
diff --git a/docs/version2.x/algorithm/text_detection/images/det_res_img_10_sast.jpg b/docs/version2.x/algorithm/text_detection/images/det_res_img_10_sast.jpg
new file mode 100644
index 0000000..702f773
Binary files /dev/null and b/docs/version2.x/algorithm/text_detection/images/det_res_img_10_sast.jpg differ
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_abinet.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_abinet.en.md
new file mode 100644
index 0000000..0436031
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_abinet.en.md
@@ -0,0 +1,121 @@
+---
+comments: true
+---
+
+# ABINet
+
+## 1. Introduction
+
+Paper:
+> [ABINet: Read Like Humans: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Recognition](https://openaccess.thecvf.com/content/CVPR2021/papers/Fang_Read_Like_Humans_Autonomous_Bidirectional_and_Iterative_Language_Modeling_for_CVPR_2021_paper.pdf)
+> Shancheng Fang and Hongtao Xie and Yuxin Wang and Zhendong Mao and Yongdong Zhang
+> CVPR, 2021
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Model|Backbone|config|Acc|Download link|
+| --- | --- | --- | --- | --- |
+|ABINet|ResNet45|[rec_r45_abinet.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r45_abinet.yml)|90.75%|[pretrained & trained model](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar)|
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_r45_abinet.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r45_abinet.yml
+```
+
+### Evaluation
+
+```bash linenums="1"
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r45_abinet.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### Prediction
+
+```bash linenums="1"
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_r45_abinet.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_r45_abinet_train/best_accuracy
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the ABINet text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar)) ), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r45_abinet.yml -o Global.pretrained_model=./rec_r45_abinet_train/best_accuracy  Global.save_inference_dir=./inference/rec_r45_abinet
+```
+
+**Note:**
+
+- If you are training the model on your own dataset and have modified the dictionary file, please pay attention to modify the `character_dict_path` in the configuration file to the modified dictionary file.
+- If you modified the input size during training, please modify the `infer_shape` corresponding to ABINet in the `tools/export_model.py` file.
+
+After the conversion is successful, there are three files in the directory:
+
+```text linenums="1"
+/inference/rec_r45_abinet/
+    ├── inference.pdiparams
+    ├── inference.pdiparams.info
+    └── inference.pdmodel
+```
+
+For ABINet text recognition model inference, the following commands can be executed:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_r45_abinet/' --rec_algorithm='ABINet' --rec_image_shape='3,32,128' --rec_char_dict_path='./ppocr/utils/ic15_dict.txt'
+```
+
+![img](./images/word_10.png)
+
+After executing the command, the prediction result (recognized text and score) of the image above is printed to the screen, an example is as follows:
+The result is as follows:
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9999995231628418)
+```
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+1. Note that the MJSynth and SynthText datasets come from [ABINet repo](https://github.com/FangShancheng/ABINet).
+2. We use the pre-trained model provided by the ABINet authors for finetune training.
+
+## Citation
+
+```bibtex
+@article{Fang2021ABINet,
+  title     = {ABINet: Read Like Humans: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Recognition},
+  author    = {Shancheng Fang and Hongtao Xie and Yuxin Wang and Zhendong Mao and Yongdong Zhang},
+  booktitle = {CVPR},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2103.06495},
+  pages     = {7098-7107}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_abinet.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_abinet.md
new file mode 100644
index 0000000..a9e9bfa
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_abinet.md
@@ -0,0 +1,136 @@
+---
+comments: true
+---
+
+# 场景文本识别算法-ABINet
+
+## 1. 算法简介
+
+论文信息：
+> [ABINet: Read Like Humans: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Recognition](https://openaccess.thecvf.com/content/CVPR2021/papers/Fang_Read_Like_Humans_Autonomous_Bidirectional_and_Iterative_Language_Modeling_for_CVPR_2021_paper.pdf)
+> Shancheng Fang and Hongtao Xie and Yuxin Wang and Zhendong Mao and Yongdong Zhang
+> CVPR, 2021
+
+`ABINet`使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法复现效果如下：
+
+|模型|骨干网络|配置文件|Acc|下载链接|
+| --- | --- | --- | --- | --- |
+|ABINet|ResNet45|[rec_r45_abinet.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r45_abinet.yml)|90.75%|[预训练、训练模型](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+### 3.1 模型训练
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练`ABINet`识别模型时需要**更换配置文件**为`ABINet`的[配置文件](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r45_abinet.yml)。
+
+#### 启动训练
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_r45_abinet.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r45_abinet.yml
+```
+
+### 3.2 评估
+
+可下载已训练完成的模型文件，使用如下命令进行评估：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r45_abinet.yml -o Global.pretrained_model=./rec_r45_abinet_train/best_accuracy
+```
+
+### 3.3 预测
+
+使用如下命令进行单张图片预测：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/infer_rec.py -c configs/rec/rec_r45_abinet.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_r45_abinet_train/best_accuracy
+# 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/imgs_words_en/'。
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将训练得到best模型，转换成inference model。这里以训练完成的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/rec_r45_abinet_train.tar) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/export_model.py -c configs/rec/rec_r45_abinet.yml -o Global.pretrained_model=./rec_r45_abinet_train/best_accuracy Global.save_inference_dir=./inference/rec_r45_abinet/
+```
+
+**注意：**
+
+- 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否是所需要的字典文件。
+- 如果您修改了训练时的输入大小，请修改`tools/export_model.py`文件中的对应ABINet的`infer_shape`。
+
+转换成功后，在目录下有三个文件：
+
+```text linenums="1"
+/inference/rec_r45_abinet/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+执行如下命令进行模型推理：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_r45_abinet/' --rec_algorithm='ABINet' --rec_image_shape='3,32,128' --rec_char_dict_path='./ppocr/utils/ic15_dict.txt'
+# 预测文件夹下所有图像时，可修改image_dir为文件夹，如 --image_dir='./doc/imgs_words_en/'。
+```
+
+![img](./images/word_10.png)
+
+执行命令后，上面图像的预测结果（识别的文本和得分）会打印到屏幕上，示例如下：
+结果如下：
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9999995231628418)
+```
+
+**注意**：
+
+- 训练上述模型采用的图像分辨率是[3，32，128]，需要通过参数`rec_image_shape`设置为您训练时的识别图像形状。
+- 在推理时需要设置参数`rec_char_dict_path`指定字典，如果您修改了字典，请修改该参数为您的字典文件。
+- 如果您修改了预处理方法，需修改`tools/infer/predict_rec.py`中ABINet的预处理为您的预处理方法。
+
+### 4.2 C++推理部署
+
+由于C++预处理后处理还未支持ABINet，所以暂未支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+1. MJSynth和SynthText两种数据集来自于[ABINet源repo](https://github.com/FangShancheng/ABINet) 。
+2. 我们使用ABINet作者提供的预训练模型进行finetune训练。
+
+## 引用
+
+```bibtex
+@article{Fang2021ABINet,
+  title     = {ABINet: Read Like Humans: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Recognition},
+  author    = {Shancheng Fang and Hongtao Xie and Yuxin Wang and Zhendong Mao and Yongdong Zhang},
+  booktitle = {CVPR},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2103.06495},
+  pages     = {7098-7107}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_aster.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_aster.en.md
new file mode 100644
index 0000000..1226994
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_aster.en.md
@@ -0,0 +1,99 @@
+---
+comments: true
+---
+
+# STAR-Net
+
+## 1. Introduction
+
+Paper:
+> [STAR-Net: a spatial attention residue network for scene text recognition.](https://www.bmva-archive.org.uk/bmvc/2016/papers/paper043/paper043.pdf)
+> Wei Liu, Chaofeng Chen, Kwan-Yee K. Wong, Zhizhong Su and Junyu Han.
+> BMVC, pages 43.1-43.13, 2016
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Model|Backbone|ACC|config|Download link|
+| --- | --- | --- | --- | --- |
+|---|---|---|---|---|
+|StarNet|Resnet34_vd|84.44%|[configs/rec/rec_r34_vd_tps_bilstm_ctc.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r34_vd_tps_bilstm_ctc.yml)|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_ctc_v2.0_train.tar)|
+|StarNet|MobileNetV3|81.42%|[configs/rec/rec_mv3_tps_bilstm_ctc.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_mv3_tps_bilstm_ctc.yml)|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_tps_bilstm_ctc_v2.0_train.tar)|
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_r34_vd_tps_bilstm_ctc.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c rec_r34_vd_tps_bilstm_ctc.yml
+```
+
+### Evaluation
+
+```bash linenums="1"
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r34_vd_tps_bilstm_ctc.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### Prediction
+
+```bash linenums="1"
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_r34_vd_tps_bilstm_ctc.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the STAR-Net text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_STAR-Net_train.tar) ), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r34_vd_tps_bilstm_ctc.yml -o Global.pretrained_model=./rec_r34_vd_tps_bilstm_ctc_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/rec_starnet
+```
+
+For STAR-Net text recognition model inference, the following commands can be executed:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_336.png" --rec_model_dir="./inference/rec_starnet/" --rec_image_shape="3, 32, 100" --rec_char_dict_path="./ppocr/utils/ic15_dict.txt"
+```
+
+### 4.2 C++ Inference
+
+With the inference model prepared, refer to the [cpp infer](../../../version2.x/legacy/cpp_infer.en.md) tutorial for C++ inference.
+
+### 4.3 Serving
+
+With the inference model prepared, refer to the [pdserving](../../../version2.x/legacy/paddle_server.en.md) tutorial for service deployment by Paddle Serving.
+
+### 4.4 More
+
+More deployment schemes supported for STAR-Net:
+
+- Paddle2ONNX: with the inference model prepared, please refer to the [paddle2onnx](../../../version2.x/legacy/paddle2onnx.en.md) tutorial.
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@inproceedings{liu2016star,
+  title={STAR-Net: a spatial attention residue network for scene text recognition.},
+  author={Liu, Wei and Chen, Chaofeng and Wong, Kwan-Yee K and Su, Zhizhong and Han, Junyu},
+  booktitle={BMVC},
+  volume={2},
+  pages={7},
+  year={2016}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_cppd.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_cppd.en.md
new file mode 100644
index 0000000..f82cbfa
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_cppd.en.md
@@ -0,0 +1,150 @@
+---
+comments: true
+---
+
+# CPPD
+
+## 1. Introduction
+
+Paper:
+> [Context Perception Parallel Decoder for Scene Text Recognition](https://arxiv.org/abs/2307.12270)
+> Yongkun Du and Zhineng Chen and Caiyan Jia and Xiaoting Yin and Chenxia Li and Yuning Du and Yu-Gang Jiang
+
+Scene text recognition models based on deep learning typically follow an Encoder-Decoder structure, where the decoder can be categorized into two types: (1) CTC and (2) Attention-based. Currently, most state-of-the-art (SOTA) models use an Attention-based decoder, which can be further divided into AR and PD types. In general, AR decoders achieve higher recognition accuracy than PD, while PD decoders are faster than AR. CPPD, with carefully designed CO and CC modules, achieves a balance between the accuracy of AR and the speed of PD.
+
+The accuracy (%) and model files of CPPD on the public dataset of scene text recognition are as follows:：
+
+* English dataset from [PARSeq](https://github.com/baudm/parseq).
+
+|    Model      |IC13<br/>857 |  SVT  |IIIT5k<br/>3000 |IC15<br/>1811| SVTP  |CUTE80 | Avg |      Download       |
+|:----------:|:------:|:-----:|:---------:|:------:|:-----:|:-----:|:-----:|:-------:|
+| CPPD Tiny  | 97.1  | 94.4 |   96.6   | 86.6  | 88.5 | 90.3 | 92.25 | [en](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_tiny_en_train.tar) |
+| CPPD Base | 98.2  | 95.5 |   97.6   | 87.9  | 90.0 | 92.7 | 93.80 | [en](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_en_train.tar)|
+| CPPD Base 48*160  | 97.5  | 95.5 |   97.7   | 87.7  | 92.4 | 93.7 | 94.10 | [en](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_48_160_en_train.tar) |
+
+* Trained on Synth dataset(MJ+ST), Test on Union14M-L benchmark from [U14m](https://github.com/Mountchicken/Union14M/).
+
+|    Model      |Curve |  Multi-<br/>Oriented  |Artistic |Contextless| Salient  | Multi-<br/>word | General | Avg |     Download       |
+|:----------:|:------:|:-----:|:---------:|:------:|:-----:|:-----:|:-----:|:-------:|:-------:|
+| CPPD Tiny  | 52.4  | 12.3 |   48.2   | 54.4  | 61.5 | 53.4 | 61.4 | 49.10 | Same as the table above. |
+| CPPD Base | 65.5  | 18.6 |   56.0   | 61.9  | 71.0 | 57.5 | 65.8 | 56.63 | Same as the table above. |
+| CPPD Base 48*160  | 71.9  | 22.1 |   60.5   | 67.9  | 78.3 | 63.9 | 67.1 | 61.69 | Same as the table above. |
+
+* Trained on Union14M-L training dataset.
+
+|    Model      |IC13<br/>857 |  SVT  |IIIT5k<br/>3000 |IC15<br/>1811| SVTP  |CUTE80 | Avg |      Download       |
+|:----------:|:------:|:-----:|:---------:|:------:|:-----:|:-----:|:-----:|:-------:|
+| CPPD Base 32*128  | 98.7  | 98.5 |   99.4   | 91.7  | 96.7 | 99.7 | 97.44 | [en](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_u14m_train.tar) |
+
+|    Model      |Curve |  Multi-<br/>Oriented  |Artistic |Contextless| Salient  | Multi-<br/>word | General | Avg |     Download       |
+|:----------:|:------:|:-----:|:---------:|:------:|:-----:|:-----:|:-----:|:-------:|:-------:|
+| CPPD Base 32*128  | 87.5  | 70.7 |   78.2   | 82.9  | 85.5 | 85.4 | 84.3 | 82.08 | Same as the table above. |
+
+* Chinese dataset from [Chinese Benchmark](https://github.com/FudanVI/benchmarking-chinese-text-recognition).
+
+|    Model      | Scene | Web | Document | Handwriting | Avg |      Download       |
+|:----------:|:------:|:-----:|:---------:|:------:|:-----:|:-----:|
+| CPPD Base  | 74.4  | 76.1 |   98.6   | 55.3  | 76.10 | [ch](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_ch_train.tar)  |
+| CPPD Base + STN | 78.4  | 79.3 |   98.9   | 57.6  | 78.55 | [ch](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_stn_ch_train.tar) |
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+### Dataset Preparation
+
+[English dataset download](https://github.com/baudm/parseq)
+[Union14M-Benchmark download](https://github.com/Mountchicken/Union14M)
+[Chinese dataset download](https://github.com/fudanvi/benchmarking-chinese-text-recognition#download)
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_svtrnet_cppd_base_en.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_svtrnet_cppd_base_en.yml
+```
+
+### Evaluation
+
+You can download the model files and configuration files provided by `CPPD`: [download link](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_en_train.tar), take `CPPD-B` as an example, using the following command to evaluate:
+
+```bash linenums="1"
+# Download the tar archive containing the model files and configuration files of CPPD-B and extract it
+wget https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_en_train.tar && tar xf rec_svtr_cppd_base_en_train.tar
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c ./rec_svtr_cppd_base_en_train/rec_svtrnet_cppd_base_en.yml -o Global.pretrained_model=./rec_svtr_cppd_base_en_train/best_model
+```
+
+### Prediction
+
+```bash linenums="1"
+python3 tools/infer_rec.py -c ./rec_svtr_cppd_base_en_train/rec_svtrnet_cppd_base_en.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_svtr_cppd_base_en_train/best_model
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the CPPD text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_en_train.tar) ), you can use the following command to convert:
+
+```bash linenums="1"
+# export model
+# en
+python3 tools/export_model.py -c configs/rec/rec_svtrnet_cppd_base_en.yml -o Global.pretrained_model=./rec_svtr_cppd_base_en_train/best_model.pdparams Global.save_inference_dir=./rec_svtr_cppd_base_en_infer
+# ch
+python3 tools/export_model.py -c configs/rec/rec_svtrnet_cppd_base_ch.yml -o Global.pretrained_model=./rec_svtr_cppd_base_ch_train/best_model.pdparams Global.save_inference_dir=./rec_svtr_cppd_base_ch_infer
+
+# speed test
+# docker image https://hub.docker.com/r/paddlepaddle/paddle/tags/: sudo docker pull paddlepaddle/paddle:2.4.2-gpu-cuda11.2-cudnn8.2-trt8.0
+# install auto_log: pip install https://paddleocr.bj.bcebos.com/libs/auto_log-1.2.0-py3-none-any.whl
+# en
+python3 tools/infer/predict_rec.py --image_dir='../iiik' --rec_model_dir='./rec_svtr_cppd_base_en_infer/' --rec_algorithm='CPPD' --rec_image_shape='3,32,100' --rec_char_dict_path='./ppocr/utils/ic15_dict.txt' --warmup=True --benchmark=True --rec_batch_num=1 --use_tensorrt=True
+# ch
+python3 tools/infer/predict_rec.py --image_dir='../iiik' --rec_model_dir='./rec_svtr_cppd_base_ch_infer/' --rec_algorithm='CPPDPadding' --rec_image_shape='3,32,256' --warmup=True --benchmark=True --rec_batch_num=1 --use_tensorrt=True
+# stn_ch
+python3 tools/infer/predict_rec.py --image_dir='../iiik' --rec_model_dir='./rec_svtr_cppd_base_stn_ch_infer/' --rec_algorithm='CPPD' --rec_image_shape='3,64,256' --warmup=True --benchmark=True --rec_batch_num=1 --use_tensorrt=True
+```
+
+**Note:** If you are training the model on your own dataset and have modified the dictionary file, please pay attention to modify the `character_dict_path` in the configuration file to the modified dictionary file.
+
+After the conversion is successful, there are three files in the directory:
+
+```text linenums="1"
+/inference/rec_svtr_cppd_base_en_infer/
+    ├── inference.pdiparams
+    ├── inference.pdiparams.info
+    └── inference.pdmodel
+```
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## Citation
+
+```bibtex
+@article{Du2023CPPD,
+  title     = {Context Perception Parallel Decoder for Scene Text Recognition},
+  author    = {Du, Yongkun and Chen, Zhineng and Jia, Caiyan and Yin, Xiaoting and Li, Chenxia and Du, Yuning and Jiang, Yu-Gang},
+  booktitle = {Arxiv},
+  year      = {2023},
+  url       = {https://arxiv.org/abs/2307.12270}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_cppd.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_cppd.md
new file mode 100644
index 0000000..958699a
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_cppd.md
@@ -0,0 +1,175 @@
+---
+comments: true
+---
+
+# 场景文本识别算法-CPPD
+
+## 1. 算法简介
+
+论文信息：
+> [Context Perception Parallel Decoder for Scene Text Recognition](https://arxiv.org/abs/2307.12270)
+> Yongkun Du and Zhineng Chen and Caiyan Jia and Xiaoting Yin and Chenxia Li and Yuning Du and Yu-Gang Jiang
+
+### CPPD算法简介
+
+基于深度学习的场景文本识别模型通常是Encoder-Decoder结构，其中decoder可以分为两种：(1)CTC，(2)Attention-based。目前SOTA模型大多使用Attention-based的decoder，而attention-based可以分为AR和PD两种，一般来说，AR解码器识别精度优于PD，而PD解码速度快于AR，CPPD通过精心设计的CO和CC模块，达到了“AR的精度，PD的速度”的效果。
+
+CPPD在场景文本识别公开数据集上的精度(%)和模型文件如下：
+
+* 英文训练集和测试集来自于[PARSeq](https://github.com/baudm/parseq)。
+
+|    模型      |IC13<br/>857 |  SVT  |IIIT5k<br/>3000 |IC15<br/>1811| SVTP  |CUTE80 | Avg |      下载链接       |
+|:----------:|:------:|:-----:|:---------:|:------:|:-----:|:-----:|:-----:|:-------:|
+| CPPD Tiny  | 97.1  | 94.4 |   96.6   | 86.6  | 88.5 | 90.3 | 92.25 | [英文](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_tiny_en_train.tar) |
+| CPPD Base | 98.2  | 95.5 |   97.6   | 87.9  | 90.0 | 92.7 | 93.80 | [英文](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_en_train.tar)|
+| CPPD Base 48*160  | 97.5  | 95.5 |   97.7   | 87.7  | 92.4 | 93.7 | 94.10 | [英文](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_48_160_en_train.tar) |
+
+* 英文合成数据集(MJ+ST)训练，英文Union14M-L benchmark测试结果[U14m](https://github.com/Mountchicken/Union14M/)。
+
+|    模型      |Curve |  Multi-<br/>Oriented  |Artistic |Contextless| Salient  | Multi-<br/>word | General | Avg |     下载链接       |
+|:----------:|:------:|:-----:|:---------:|:------:|:-----:|:-----:|:-----:|:-------:|:-------:|
+| CPPD Tiny  | 52.4  | 12.3 |   48.2   | 54.4  | 61.5 | 53.4 | 61.4 | 49.10 | 同上表 |
+| CPPD Base | 65.5  | 18.6 |   56.0   | 61.9  | 71.0 | 57.5 | 65.8 | 56.63 | 同上表 |
+| CPPD Base 48*160  | 71.9  | 22.1 |   60.5   | 67.9  | 78.3 | 63.9 | 67.1 | 61.69 | 同上表 |
+
+* Union14M-L 训练集From scratch训练，英文测试结果。
+
+|    模型      |IC13<br/>857 |  SVT  |IIIT5k<br/>3000 |IC15<br/>1811| SVTP  |CUTE80 | Avg |      下载链接       |
+|:----------:|:------:|:-----:|:---------:|:------:|:-----:|:-----:|:-----:|:-------:|
+| CPPD Base 32*128  | 98.5  | 97.7 |   99.2   | 90.3  | 94.6 | 98.3 | 96.42 | Coming soon |
+
+|    模型      |Curve |  Multi-<br/>Oriented  |Artistic |Contextless| Salient  | Multi-<br/>word | General | Avg |     下载链接       |
+|:----------:|:------:|:-----:|:---------:|:------:|:-----:|:-----:|:-----:|:-------:|:-------:|
+| CPPD Base 32*128  | 83.0  | 71.2 |   75.1   | 80.9  | 79.4 | 82.6 | 83.7 | 79.41 | Coming soon |
+
+* 加载合成数据集预训练模型，Union14M-L 训练集微调训练，英文测试结果。
+
+|    模型      |IC13<br/>857 |  SVT  |IIIT5k<br/>3000 |IC15<br/>1811| SVTP  |CUTE80 | Avg |      下载链接       |
+|:----------:|:------:|:-----:|:---------:|:------:|:-----:|:-----:|:-----:|:-------:|
+| CPPD Base 32*128  | 98.7  | 98.5 |   99.4   | 91.7  | 96.7 | 99.7 | 97.44 | [英文](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_u14m_train.tar) |
+
+|    模型      |Curve |  Multi-<br/>Oriented  |Artistic |Contextless| Salient  | Multi-<br/>word | General | Avg |     下载链接       |
+|:----------:|:------:|:-----:|:---------:|:------:|:-----:|:-----:|:-----:|:-------:|:-------:|
+| CPPD Base 32*128  | 87.5  | 70.7 |   78.2   | 82.9  | 85.5 | 85.4 | 84.3 | 82.08 | 同上表 |
+
+* 中文训练集和测试集来自于[Chinese Benchmark](https://github.com/FudanVI/benchmarking-chinese-text-recognition)。
+
+|    模型      | Scene | Web | Document | Handwriting | Avg |      下载链接       |
+|:----------:|:------:|:-----:|:---------:|:------:|:-----:|:-----:|
+| CPPD Base  | 74.4  | 76.1 |   98.6   | 55.3  | 76.10 | [中文](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_ch_train.tar)  |
+| CPPD Base + STN | 78.4  | 79.3 |   98.9   | 57.6  | 78.55 | [中文](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_stn_ch_train.tar) |
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+### 3.1 模型训练
+
+#### 数据集准备
+
+[英文数据集下载](https://github.com/baudm/parseq)
+
+[Union14M-L 下载](https://github.com/Mountchicken/Union14M)
+
+[中文数据集下载](https://github.com/fudanvi/benchmarking-chinese-text-recognition#download)
+
+#### 启动训练
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练`CPPD`识别模型时需要**更换配置文件**为`CPPD`的[配置文件](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_svtrnet_cppd_base_en.yml)。
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_svtrnet_cppd_base_en.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_svtrnet_cppd_base_en.yml
+```
+
+### 3.2 评估
+
+可下载`CPPD`提供的模型文件和配置文件：[下载地址](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_en_train.tar) ，以`CPPD-B`为例，使用如下命令进行评估：
+
+```bash linenums="1"
+# 下载包含CPPD-B的模型文件和配置文件的tar压缩包并解压
+wget https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_en_train.tar && tar xf rec_svtr_cppd_base_en_train.tar
+# 注意将pretrained_model的路径设置为本地路径。
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c ./rec_svtr_cppd_base_en_train/rec_svtrnet_cppd_base_en.yml -o Global.pretrained_model=./rec_svtr_cppd_base_en_train/best_model
+```
+
+### 3.3 预测
+
+使用如下命令进行单张图片预测：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/infer_rec.py -c ./rec_svtr_cppd_base_en_train/rec_svtrnet_cppd_base_en.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_svtr_cppd_base_en_train/best_model
+# 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/imgs_words_en/'。
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将训练得到best模型，转换成inference model。下面以基于`CPPD-B`，在英文数据集训练的模型为例（[模型和配置文件下载地址](https://paddleocr.bj.bcebos.com/CCPD/rec_svtr_cppd_base_en_train.tar)，可以使用如下命令进行转换：
+
+**注意：**
+
+* 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否为所正确的字典文件。
+
+执行如下命令进行模型导出和推理：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+# export model
+# en
+python3 tools/export_model.py -c configs/rec/rec_svtrnet_cppd_base_en.yml -o Global.pretrained_model=./rec_svtr_cppd_base_en_train/best_model.pdparams Global.save_inference_dir=./rec_svtr_cppd_base_en_infer
+# ch
+python3 tools/export_model.py -c configs/rec/rec_svtrnet_cppd_base_ch.yml -o Global.pretrained_model=./rec_svtr_cppd_base_ch_train/best_model.pdparams Global.save_inference_dir=./rec_svtr_cppd_base_ch_infer
+
+# speed test
+# docker image https://hub.docker.com/r/paddlepaddle/paddle/tags/: sudo docker pull paddlepaddle/paddle:2.4.2-gpu-cuda11.2-cudnn8.2-trt8.0
+# install auto_log: pip install https://paddleocr.bj.bcebos.com/libs/auto_log-1.2.0-py3-none-any.whl
+# en
+python3 tools/infer/predict_rec.py --image_dir='../iiik' --rec_model_dir='./rec_svtr_cppd_base_en_infer/' --rec_algorithm='CPPD' --rec_image_shape='3,32,100' --rec_char_dict_path='./ppocr/utils/ic15_dict.txt' --warmup=True --benchmark=True --rec_batch_num=1 --use_tensorrt=True
+# ch
+python3 tools/infer/predict_rec.py --image_dir='../iiik' --rec_model_dir='./rec_svtr_cppd_base_ch_infer/' --rec_algorithm='CPPDPadding' --rec_image_shape='3,32,256' --warmup=True --benchmark=True --rec_batch_num=1 --use_tensorrt=True
+# stn_ch
+python3 tools/infer/predict_rec.py --image_dir='../iiik' --rec_model_dir='./rec_svtr_cppd_base_stn_ch_infer/' --rec_algorithm='CPPD' --rec_image_shape='3,64,256' --warmup=True --benchmark=True --rec_batch_num=1 --use_tensorrt=True
+```
+
+导出成功后，在目录下有三个文件：
+
+```
+/inference/rec_svtr_cppd_base_en_infer/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+### 4.2 C++推理部署
+
+由于C++预处理后处理还未支持CPPD，所以暂未支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 引用
+
+```bibtex
+@article{Du2023CPPD,
+  title     = {Context Perception Parallel Decoder for Scene Text Recognition},
+  author    = {Du, Yongkun and Chen, Zhineng and Jia, Caiyan and Yin, Xiaoting and Li, Chenxia and Du, Yuning and Jiang, Yu-Gang},
+  booktitle = {Arxiv},
+  year      = {2023},
+  url       = {https://arxiv.org/abs/2307.12270}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_crnn.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_crnn.en.md
new file mode 100644
index 0000000..ebabba8
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_crnn.en.md
@@ -0,0 +1,102 @@
+---
+comments: true
+---
+
+# CRNN
+
+## 1. Introduction
+
+Paper:
+> [An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition](https://arxiv.org/abs/1507.05717)
+
+> Baoguang Shi, Xiang Bai, Cong Yao
+
+> IEEE, 2015
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Model|Backbone|ACC|config|Download link|
+| --- | --- | --- | --- | --- |
+|---|---|---|---|---|
+|CRNN|Resnet34_vd|81.04%|[configs/rec/rec_r34_vd_none_bilstm_ctc.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r34_vd_none_bilstm_ctc.yml)|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_bilstm_ctc_v2.0_train.tar)|
+|CRNN|MobileNetV3|77.95%|[configs/rec/rec_mv3_none_bilstm_ctc.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_mv3_none_bilstm_ctc.yml)|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_bilstm_ctc_v2.0_train.tar)|
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_r34_vd_none_bilstm_ctc.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r34_vd_none_bilstm_ctc.yml
+```
+
+### Evaluation
+
+```bash linenums="1"
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r34_vd_none_bilstm_ctc.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### Prediction
+
+```bash linenums="1"
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_r34_vd_none_bilstm_ctc.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the CRNN text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_CRNN_train.tar) ), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r34_vd_none_bilstm_ctc.yml -o Global.pretrained_model=./rec_r34_vd_none_bilstm_ctc_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/rec_crnn
+```
+
+For CRNN text recognition model inference, the following commands can be executed:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_336.png" --rec_model_dir="./inference/rec_crnn/" --rec_image_shape="3, 32, 100" --rec_char_dict_path="./ppocr/utils/ic15_dict.txt"
+```
+
+### 4.2 C++ Inference
+
+With the inference model prepared, refer to the [cpp infer](../../../version2.x/legacy/cpp_infer.en.md) tutorial for C++ inference.
+
+### 4.3 Serving
+
+With the inference model prepared, refer to the [pdserving](../../../version2.x/legacy/paddle_server.en.md) tutorial for service deployment by Paddle Serving.
+
+### 4.4 More
+
+More deployment schemes supported for CRNN:
+
+- Paddle2ONNX: with the inference model prepared, please refer to the [paddle2onnx](../../../version2.x/legacy/paddle_server.en.md) tutorial.
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@ARTICLE{7801919,
+  author={Shi, Baoguang and Bai, Xiang and Yao, Cong},
+  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
+  title={An End-to-End Trainable Neural Network for Image-Based Sequence Recognition and Its Application to Scene Text Recognition},
+  year={2017},
+  volume={39},
+  number={11},
+  pages={2298-2304},
+  doi={10.1109/TPAMI.2016.2646371}}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_crnn.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_crnn.md
new file mode 100644
index 0000000..50640dd
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_crnn.md
@@ -0,0 +1,118 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# CRNN
+
+## 1. 算法简介
+
+论文信息：
+> [An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition](https://arxiv.org/abs/1507.05717)
+> Baoguang Shi, Xiang Bai, Cong Yao
+> IEEE, 2015
+
+参考[DTRB](https://arxiv.org/abs/1904.01906) 文字识别训练和评估流程，使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法效果如下：
+
+|模型|骨干网络|Avg Accuracy|配置文件|下载链接|
+|---|---|---|---|---|
+|CRNN|Resnet34_vd|81.04%|[configs/rec/rec_r34_vd_none_bilstm_ctc.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r34_vd_none_bilstm_ctc.yml)|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_bilstm_ctc_v2.0_train.tar)|
+|CRNN|MobileNetV3|77.95%|[configs/rec/rec_mv3_none_bilstm_ctc.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_mv3_none_bilstm_ctc.yml)|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_bilstm_ctc_v2.0_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练不同的识别模型只需要**更换配置文件**即可。
+
+### 训练
+
+在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+# 单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_r34_vd_none_bilstm_ctc.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c rec_r34_vd_none_bilstm_ctc.yml
+```
+
+### 评估
+
+```bash linenums="1"
+# GPU 评估， Global.pretrained_model 为待测权重
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r34_vd_none_bilstm_ctc.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### 预测
+
+```bash linenums="1"
+# 预测使用的配置文件必须与训练一致
+python3 tools/infer_rec.py -c configs/rec/rec_r34_vd_none_bilstm_ctc.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将 CRNN 文本识别训练过程中保存的模型，转换成inference model。以基于Resnet34_vd骨干网络，使用MJSynth和SynthText两个英文文本识别合成数据集训练的[模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_bilstm_ctc_v2.0_train.tar) 为例，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r34_vd_none_bilstm_ctc.yml -o Global.pretrained_model=./rec_r34_vd_none_bilstm_ctc_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/rec_crnn
+```
+
+CRNN 文本识别模型推理，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_336.png" --rec_model_dir="./inference/rec_crnn/" --rec_image_shape="3, 32, 100" --rec_char_dict_path="./ppocr/utils/ic15_dict.txt"
+```
+
+![img](./images/word_336-20240705082445918.png)
+
+执行命令后，上面图像的识别结果如下：
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words_en/word_336.png:('super', 0.9999073)
+```
+
+**注意**：由于上述模型是参考[DTRB](https://arxiv.org/abs/1904.01906)文本识别训练和评估流程，与超轻量级中文识别模型训练有两方面不同：
+
+- 训练时采用的图像分辨率不同，训练上述模型采用的图像分辨率是[3，32，100]，而中文模型训练时，为了保证长文本的识别效果，训练时采用的图像分辨率是[3, 32, 320]。预测推理程序默认的形状参数是训练中文采用的图像分辨率，即[3, 32, 320]。因此，这里推理上述英文模型时，需要通过参数rec_image_shape设置识别图像的形状。
+- 字符列表，DTRB论文中实验只是针对26个小写英文本母和10个数字进行实验，总共36个字符。所有大小字符都转成了小写字符，不在上面列表的字符都忽略，认为是空格。因此这里没有输入字符字典，而是通过如下命令生成字典.因此在推理时需要设置参数rec_char_dict_path，指定为英文字典"./ppocr/utils/ic15_dict.txt"。
+
+  ```python linenums="1"
+  self.character_str = "0123456789abcdefghijklmnopqrstuvwxyz"
+  dict_character = list(self.character_str)
+  ```
+
+### 4.2 C++推理
+
+准备好推理模型后，参考[cpp infer](../../../version2.x/legacy/cpp_infer.md)教程进行操作即可。
+
+### 4.3 Serving服务化部署
+
+准备好推理模型后，参考[pdserving](../../../version2.x/legacy/paddle_server.md)教程进行Serving服务化部署，包括Python Serving和C++ Serving两种模式。
+
+### 4.4 更多推理部署
+
+CRNN模型还支持以下推理部署方式：
+
+- Paddle2ONNX推理：准备好推理模型后，参考[paddle2onnx](../../../version2.x/legacy/paddle2onnx.md)教程操作。
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@ARTICLE{7801919,
+  author={Shi, Baoguang and Bai, Xiang and Yao, Cong},
+  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
+  title={An End-to-End Trainable Neural Network for Image-Based Sequence Recognition and Its Application to Scene Text Recognition},
+  year={2017},
+  volume={39},
+  number={11},
+  pages={2298-2304},
+  doi={10.1109/TPAMI.2016.2646371}}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_nrtr.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_nrtr.en.md
new file mode 100644
index 0000000..5b97e16
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_nrtr.en.md
@@ -0,0 +1,256 @@
+---
+comments: true
+---
+
+# NRTR
+
+## 1. Introduction
+
+Paper:
+> [NRTR: A No-Recurrence Sequence-to-Sequence Model For Scene Text Recognition](https://arxiv.org/abs/1806.00926)
+> Fenfen Sheng and Zhineng Chen and Bo Xu
+> ICDAR, 2019
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Model|Backbone|config|Acc|Download link|
+| --- | --- | --- | --- | --- |
+|NRTR|MTB|[rec_mtb_nrtr.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_mtb_nrtr.yml)|84.21%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mtb_nrtr_train.tar)|
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_mtb_nrtr.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_mtb_nrtr.yml
+```
+
+### Evaluation
+
+```bash linenums="1"
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_mtb_nrtr.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### Prediction
+
+```bash linenums="1"
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_mtb_nrtr.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_mtb_nrtr_train/best_accuracy
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the NRTR text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mtb_nrtr_train.tar)) ), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_mtb_nrtr.yml -o Global.pretrained_model=./rec_mtb_nrtr_train/best_accuracy  Global.save_inference_dir=./inference/rec_mtb_nrtr
+```
+
+**Note:**
+
+- If you are training the model on your own dataset and have modified the dictionary file, please pay attention to modify the `character_dict_path` in the configuration file to the modified dictionary file.
+- If you modified the input size during training, please modify the `infer_shape` corresponding to NRTR in the `tools/export_model.py` file.
+
+After the conversion is successful, there are three files in the directory:
+
+```text linenums="1"
+/inference/rec_mtb_nrtr/
+    ├── inference.pdiparams
+    ├── inference.pdiparams.info
+    └── inference.pdmodel
+```
+
+For NRTR text recognition model inference, the following commands can be executed:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_mtb_nrtr/' --rec_algorithm='NRTR' --rec_image_shape='1,32,100' --rec_char_dict_path='./ppocr/utils/EN_symbol_dict.txt'
+```
+
+![img](./images/word_10.png)
+
+After executing the command, the prediction result (recognized text and score) of the image above is printed to the screen, an example is as follows:
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9465042352676392)
+```
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+1. In the `NRTR` paper, Beam search is used to decode characters, but the speed is slow. Beam search is not used by default here, and greedy search is used to decode characters.
+
+## 6. Release Note
+
+1. The release/2.6 version updates the NRTR code structure. The new version of NRTR can load the model parameters of the old version (release/2.5 and before), and you may use the following code to convert the old version model parameters to the new version model parameters:
+
+    <details>
+    <summary>Click to expand</summary>
+
+    ```python linenums="1"
+    params = paddle.load('path/' + '.pdparams') # the old version parameters
+    state_dict = model.state_dict() # the new version model parameters
+    new_state_dict = {}
+
+    for k1, v1 in state_dict.items():
+
+        k = k1
+        if 'encoder' in k and 'self_attn' in k and 'qkv' in k and 'weight' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')].transpose((1, 0, 2, 3))
+            k = params[k_para.replace('qkv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('qkv', 'conv3')].transpose((1, 0, 2, 3))
+
+            new_state_dict[k1] = np.concatenate([q[:, :, 0, 0], k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'encoder' in k and 'self_attn' in k and 'qkv' in k and 'bias' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')]
+            k = params[k_para.replace('qkv', 'conv2')]
+            v = params[k_para.replace('qkv', 'conv3')]
+
+            new_state_dict[k1] = np.concatenate([q, k, v], -1)
+
+        elif 'encoder' in k and 'self_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+        elif 'encoder' in k and 'norm3' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para.replace('norm3', 'norm2')]
+
+        elif 'encoder' in k and 'norm1' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+
+        elif 'decoder' in k and 'self_attn' in k and 'qkv' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')].transpose((1, 0, 2, 3))
+            k = params[k_para.replace('qkv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('qkv', 'conv3')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = np.concatenate([q[:, :, 0, 0], k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'decoder' in k and 'self_attn' in k and 'qkv' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')]
+            k = params[k_para.replace('qkv', 'conv2')]
+            v = params[k_para.replace('qkv', 'conv3')]
+            new_state_dict[k1] = np.concatenate([q, k, v], -1)
+
+        elif 'decoder' in k and 'self_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+        elif 'decoder' in k and 'cross_attn' in k and 'q' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            q = params[k_para.replace('q', 'conv1')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = q[:, :, 0, 0]
+
+        elif 'decoder' in k and 'cross_attn' in k and 'q' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            q = params[k_para.replace('q', 'conv1')]
+            new_state_dict[k1] = q
+
+        elif 'decoder' in k and 'cross_attn' in k and 'kv' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            k = params[k_para.replace('kv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('kv', 'conv3')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = np.concatenate([k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'decoder' in k and 'cross_attn' in k and 'kv' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            k = params[k_para.replace('kv', 'conv2')]
+            v = params[k_para.replace('kv', 'conv3')]
+            new_state_dict[k1] = np.concatenate([k, v], -1)
+
+        elif 'decoder' in k and 'cross_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            new_state_dict[k1] = params[k_para]
+        elif 'decoder' in k and 'norm' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+        elif 'mlp' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('fc', 'conv')
+            k_para = k_para.replace('mlp.', '')
+            w = params[k_para].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = w[:, :, 0, 0]
+        elif 'mlp' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('fc', 'conv')
+            k_para = k_para.replace('mlp.', '')
+            w = params[k_para]
+            new_state_dict[k1] = w
+
+        else:
+            new_state_dict[k1] = params[k1]
+
+        if list(new_state_dict[k1].shape) != list(v1.shape):
+            print(k1)
+
+
+    for k, v1 in state_dict.items():
+        if k not in new_state_dict.keys():
+            print(1, k)
+        elif list(new_state_dict[k].shape) != list(v1.shape):
+            print(2, k)
+
+
+
+    model.set_state_dict(new_state_dict)
+    paddle.save(model.state_dict(), 'nrtrnew_from_old_params.pdparams')
+
+    ```
+
+    </details>
+
+2. The new version has a clean code structure and improved inference speed compared with the old version.
+
+## Citation
+
+```bibtex
+@article{Sheng2019NRTR,
+  title     = {NRTR: A No-Recurrence Sequence-to-Sequence Model For Scene Text Recognition},
+  author    = {Fenfen Sheng and Zhineng Chen and Bo Xu},
+  booktitle = {ICDAR},
+  year      = {2019},
+  url       = {http://arxiv.org/abs/1806.00926},
+  pages     = {781-786}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_nrtr.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_nrtr.md
new file mode 100644
index 0000000..8029bea
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_nrtr.md
@@ -0,0 +1,272 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# 场景文本识别算法-NRTR
+
+## 1. 算法简介
+
+论文信息：
+> [NRTR: A No-Recurrence Sequence-to-Sequence Model For Scene Text Recognition](https://arxiv.org/abs/1806.00926)
+> Fenfen Sheng and Zhineng Chen and Bo Xu
+> ICDAR, 2019
+
+`NRTR`使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法复现效果如下：
+
+|模型|骨干网络|配置文件|Acc|下载链接|
+| --- | --- | --- | --- | --- |
+|NRTR|MTB|[rec_mtb_nrtr.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_mtb_nrtr.yml)|84.21%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mtb_nrtr_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+### 3.1 模型训练
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练`NRTR`识别模型时需要**更换配置文件**为`NRTR`的[配置文件](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_mtb_nrtr.yml)。
+
+#### 启动训练
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_mtb_nrtr.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_mtb_nrtr.yml
+```
+
+### 3.2 评估
+
+可下载已训练完成的模型文件，使用如下命令进行评估：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_mtb_nrtr.yml -o Global.pretrained_model=./rec_mtb_nrtr_train/best_accuracy
+```
+
+### 3.3 预测
+
+使用如下命令进行单张图片预测：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/infer_rec.py -c configs/rec/rec_mtb_nrtr.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_mtb_nrtr_train/best_accuracy
+# 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/imgs_words_en/'。
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将训练得到best模型，转换成inference model。这里以训练完成的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mtb_nrtr_train.tar) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/export_model.py -c configs/rec/rec_mtb_nrtr.yml -o Global.pretrained_model=./rec_mtb_nrtr_train/best_accuracy Global.save_inference_dir=./inference/rec_mtb_nrtr/
+```
+
+**注意：**
+
+- 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否是所需要的字典文件。
+- 如果您修改了训练时的输入大小，请修改`tools/export_model.py`文件中的对应NRTR的`infer_shape`。
+
+转换成功后，在目录下有三个文件：
+
+```text linenums="1"
+/inference/rec_mtb_nrtr/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+执行如下命令进行模型推理：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_mtb_nrtr/' --rec_algorithm='NRTR' --rec_image_shape='1,32,100' --rec_char_dict_path='./ppocr/utils/EN_symbol_dict.txt'
+# 预测文件夹下所有图像时，可修改image_dir为文件夹，如 --image_dir='./doc/imgs_words_en/'。
+```
+
+![img](./images/word_10.png)
+
+执行命令后，上面图像的预测结果（识别的文本和得分）会打印到屏幕上，示例如下：
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9465042352676392)
+```
+
+**注意**：
+
+- 训练上述模型采用的图像分辨率是[1，32，100]，需要通过参数`rec_image_shape`设置为您训练时的识别图像形状。
+- 在推理时需要设置参数`rec_char_dict_path`指定字典，如果您修改了字典，请修改该参数为您的字典文件。
+- 如果您修改了预处理方法，需修改`tools/infer/predict_rec.py`中NRTR的预处理为您的预处理方法。
+
+### 4.2 C++推理部署
+
+由于C++预处理后处理还未支持NRTR，所以暂未支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+1. `NRTR`论文中使用Beam搜索进行解码字符，但是速度较慢，这里默认未使用Beam搜索，以贪婪搜索进行解码字符。
+
+## 6. 发行公告
+
+1. release/2.6更新NRTR代码结构，新版NRTR可加载旧版(release/2.5及之前)模型参数，使用下面示例代码将旧版模型参数转换为新版模型参数：
+
+    <details>
+    <summary>详情</summary>
+
+    ```python linenums="1"
+    params = paddle.load('path/' + '.pdparams') # 旧版本参数
+    state_dict = model.state_dict() # 新版模型参数
+    new_state_dict = {}
+
+    for k1, v1 in state_dict.items():
+
+        k = k1
+        if 'encoder' in k and 'self_attn' in k and 'qkv' in k and 'weight' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')].transpose((1, 0, 2, 3))
+            k = params[k_para.replace('qkv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('qkv', 'conv3')].transpose((1, 0, 2, 3))
+
+            new_state_dict[k1] = np.concatenate([q[:, :, 0, 0], k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'encoder' in k and 'self_attn' in k and 'qkv' in k and 'bias' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')]
+            k = params[k_para.replace('qkv', 'conv2')]
+            v = params[k_para.replace('qkv', 'conv3')]
+
+            new_state_dict[k1] = np.concatenate([q, k, v], -1)
+
+        elif 'encoder' in k and 'self_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+        elif 'encoder' in k and 'norm3' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para.replace('norm3', 'norm2')]
+
+        elif 'encoder' in k and 'norm1' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+
+        elif 'decoder' in k and 'self_attn' in k and 'qkv' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')].transpose((1, 0, 2, 3))
+            k = params[k_para.replace('qkv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('qkv', 'conv3')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = np.concatenate([q[:, :, 0, 0], k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'decoder' in k and 'self_attn' in k and 'qkv' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            q = params[k_para.replace('qkv', 'conv1')]
+            k = params[k_para.replace('qkv', 'conv2')]
+            v = params[k_para.replace('qkv', 'conv3')]
+            new_state_dict[k1] = np.concatenate([q, k, v], -1)
+
+        elif 'decoder' in k and 'self_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+
+        elif 'decoder' in k and 'cross_attn' in k and 'q' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            q = params[k_para.replace('q', 'conv1')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = q[:, :, 0, 0]
+
+        elif 'decoder' in k and 'cross_attn' in k and 'q' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            q = params[k_para.replace('q', 'conv1')]
+            new_state_dict[k1] = q
+
+        elif 'decoder' in k and 'cross_attn' in k and 'kv' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            k = params[k_para.replace('kv', 'conv2')].transpose((1, 0, 2, 3))
+            v = params[k_para.replace('kv', 'conv3')].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = np.concatenate([k[:, :, 0, 0], v[:, :, 0, 0]], -1)
+
+        elif 'decoder' in k and 'cross_attn' in k and 'kv' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            k = params[k_para.replace('kv', 'conv2')]
+            v = params[k_para.replace('kv', 'conv3')]
+            new_state_dict[k1] = np.concatenate([k, v], -1)
+
+        elif 'decoder' in k and 'cross_attn' in k and 'out_proj' in k:
+
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('cross_attn', 'multihead_attn')
+            new_state_dict[k1] = params[k_para]
+        elif 'decoder' in k and 'norm' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            new_state_dict[k1] = params[k_para]
+        elif 'mlp' in k and 'weight' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('fc', 'conv')
+            k_para = k_para.replace('mlp.', '')
+            w = params[k_para].transpose((1, 0, 2, 3))
+            new_state_dict[k1] = w[:, :, 0, 0]
+        elif 'mlp' in k and 'bias' in k:
+            k_para = k[:13] + 'layers.' + k[13:]
+            k_para = k_para.replace('fc', 'conv')
+            k_para = k_para.replace('mlp.', '')
+            w = params[k_para]
+            new_state_dict[k1] = w
+
+        else:
+            new_state_dict[k1] = params[k1]
+
+        if list(new_state_dict[k1].shape) != list(v1.shape):
+            print(k1)
+
+
+    for k, v1 in state_dict.items():
+        if k not in new_state_dict.keys():
+            print(1, k)
+        elif list(new_state_dict[k].shape) != list(v1.shape):
+            print(2, k)
+
+
+
+    model.set_state_dict(new_state_dict)
+    paddle.save(model.state_dict(), 'nrtrnew_from_old_params.pdparams')
+
+    ```
+
+    </details>
+
+2. 新版相比与旧版，代码结构简洁，推理速度有所提高。
+
+## 引用
+
+```bibtex
+@article{Sheng2019NRTR,
+  title     = {NRTR: A No-Recurrence Sequence-to-Sequence Model For Scene Text Recognition},
+  author    = {Fenfen Sheng and Zhineng Chen and Bo Xu},
+  booktitle = {ICDAR},
+  year      = {2019},
+  url       = {http://arxiv.org/abs/1806.00926},
+  pages     = {781-786}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_parseq.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_parseq.en.md
new file mode 100644
index 0000000..d191973
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_parseq.en.md
@@ -0,0 +1,105 @@
+---
+comments: true
+---
+
+# PasreQ
+
+## 1. Introduction
+
+Paper:
+> [Scene Text Recognition with Permuted Autoregressive Sequence Models](https://arxiv.org/abs/2207.06966)
+> Darwin Bautista, Rowel Atienza
+> ECCV, 2021
+
+Using real datasets (real) and synthetic datasets (synth) for training respectively，and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets.
+
+- The real datasets include COCO-Text, RCTW17, Uber-Text, ArT, LSVT, MLT19, ReCTS, TextOCR and OpenVINO datasets.
+- The synthesis datasets include MJSynth and SynthText datasets.
+
+the algorithm reproduction effect is as follows:
+
+|Training Dataset|Model|Backbone|config|Acc|Download link|
+| --- | --- | --- | --- | --- | --- |
+|Synth|ParseQ|VIT|[rec_vit_parseq.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_vit_parseq.yml)|91.24%|[train model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/parseq/rec_vit_parseq_synth.tgz)|
+|Real|ParseQ|VIT|[rec_vit_parseq.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_vit_parseq.yml)|94.74%|[train model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/parseq/rec_vit_parseq_real.tgz)|
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_vit_parseq.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_vit_parseq.yml
+```
+
+### Evaluation
+
+```bash linenums="1"
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_vit_parseq.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### Prediction
+
+```bash linenums="1"
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_vit_parseq.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the SAR text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/dygraph_v2.1/parseq/rec_vit_parseq_real.tgz) ), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_vit_parseq.yml -o Global.pretrained_model=./rec_vit_parseq_real/best_accuracy Global.save_inference_dir=./inference/rec_parseq
+```
+
+For SAR text recognition model inference, the following commands can be executed:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/en/word_1.png" --rec_model_dir="./inference/rec_parseq/" --rec_image_shape="3, 32, 128" --rec_algorithm="ParseQ" --rec_char_dict_path="ppocr/utils/dict/parseq_dict.txt" --max_text_length=25 --use_space_char=False
+```
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@InProceedings{bautista2022parseq,
+  title={Scene Text Recognition with Permuted Autoregressive Sequence Models},
+  author={Bautista, Darwin and Atienza, Rowel},
+  booktitle={European Conference on Computer Vision},
+  pages={178--196},
+  month={10},
+  year={2022},
+  publisher={Springer Nature Switzerland},
+  address={Cham},
+  doi={10.1007/978-3-031-19815-1_11},
+  url={https://doi.org/10.1007/978-3-031-19815-1_11}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_parseq.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_parseq.md
new file mode 100644
index 0000000..788bc11
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_parseq.md
@@ -0,0 +1,105 @@
+---
+comments: true
+---
+
+# ParseQ
+
+## 1. 算法简介
+
+论文信息：
+> [Scene Text Recognition with Permuted Autoregressive Sequence Models](https://arxiv.org/abs/2207.06966)
+> Darwin Bautista, Rowel Atienza
+> ECCV, 2021
+
+原论文分别使用真实文本识别数据集(Real)和合成文本识别数据集(Synth)进行训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估。其中：
+
+- 真实文本识别数据集(Real)包含COCO-Text, RCTW17, Uber-Text, ArT, LSVT, MLT19, ReCTS, TextOCR, OpenVINO数据集
+- 合成文本识别数据集(Synth)包含MJSynth和SynthText数据集
+
+在不同数据集上训练的算法的复现效果如下：
+
+|数据集|模型|骨干网络|配置文件|Acc|下载链接|
+| --- | --- | --- | --- | --- | --- |
+|Synth|ParseQ|VIT|[rec_vit_parseq.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_vit_parseq.yml)|91.24%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/parseq/rec_vit_parseq_synth.tgz)|
+|Real|ParseQ|VIT|[rec_vit_parseq.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_vit_parseq.yml)|94.74%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/parseq/rec_vit_parseq_real.tgz)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+请参考[文本识别教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练不同的识别模型只需要**更换配置文件**即可。
+
+### 训练
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+# 单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_vit_parseq.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_vit_parseq.yml
+```
+
+### 评估
+
+```bash linenums="1"
+# GPU 评估， Global.pretrained_model 为待测权重
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_vit_parseq.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### 预测
+
+```bash linenums="1"
+# 预测使用的配置文件必须与训练一致
+python3 tools/infer_rec.py -c configs/rec/rec_vit_parseq.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将ParseQ文本识别训练过程中保存的模型，转换成inference model。（ [模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.1/parseq/rec_vit_parseq_real.tgz) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_vit_parseq.yml -o Global.pretrained_model=./rec_vit_parseq_real/best_accuracy Global.save_inference_dir=./inference/rec_parseq
+```
+
+ParseQ文本识别模型推理，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/en/word_1.png" --rec_model_dir="./inference/rec_parseq/" --rec_image_shape="3, 32, 128" --rec_algorithm="ParseQ" --rec_char_dict_path="ppocr/utils/dict/parseq_dict.txt" --max_text_length=25 --use_space_char=False
+```
+
+### 4.2 C++推理
+
+由于C++预处理后处理还未支持ParseQ，所以暂未支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@InProceedings{bautista2022parseq,
+  title={Scene Text Recognition with Permuted Autoregressive Sequence Models},
+  author={Bautista, Darwin and Atienza, Rowel},
+  booktitle={European Conference on Computer Vision},
+  pages={178--196},
+  month={10},
+  year={2022},
+  publisher={Springer Nature Switzerland},
+  address={Cham},
+  doi={10.1007/978-3-031-19815-1_11},
+  url={https://doi.org/10.1007/978-3-031-19815-1_11}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_rare.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_rare.en.md
new file mode 100644
index 0000000..264fe1a
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_rare.en.md
@@ -0,0 +1,100 @@
+---
+comments: true
+---
+
+# RARE
+
+## 1. Introduction
+
+Paper information:
+> [Robust Scene Text Recognition with Automatic Rectification](https://arxiv.org/abs/1603.03915v2)
+> Baoguang Shi, Xinggang Wang, Pengyuan Lyu, Cong Yao, Xiang Bai∗
+> CVPR, 2016
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Models|Backbone Networks|Configuration Files|Avg Accuracy|Download Links|
+| --- | --- | --- | --- | --- |
+|RARE|Resnet34_vd|[configs/rec/rec_r34_vd_tps_bilstm_att.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r34_vd_tps_bilstm_att.yml)|83.60%|[training model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_att_v2.0_train.tar)|
+|RARE|MobileNetV3|[configs/rec/rec_mv3_tps_bilstm_att.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_mv3_tps_bilstm_att.yml)|82.50%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_tps_bilstm_att_v2.0_train.tar)|
+
+## 2. Environment
+
+Please refer to [Operating Environment Preparation](../../ppocr/environment.en.md) to configure the PaddleOCR operating environment, and refer to [Project Clone](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Training Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**. Take the backbone network based on Resnet34_vd as an example:
+
+### 3.1 Training
+
+````bash linenums="1"
+#  Single card training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_r34_vd_tps_bilstm_att.yml
+# Multi-card training, specify the card number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/rec/rec_r34_vd_tps_bilstm_att.yml
+````
+
+### 3.2 Evaluation
+
+````bash linenums="1"
+# GPU evaluation, Global.pretrained_model is the model to be evaluated
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r34_vd_tps_bilstm_att.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+````
+
+### 3.3 Prediction
+
+````bash linenums="1"
+python3 tools/infer_rec.py -c configs/rec/rec_r34_vd_tps_bilstm_att.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+````
+
+## 4. Inference
+
+### 4.1 Python Inference
+
+First, convert the model saved during the RARE text recognition training process into an inference model. Take the model trained on the MJSynth and SynthText text recognition datasets based on the Resnet34_vd backbone network as an example ([Model download address](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_att_v2.0_train.tar) ), which can be converted using the following command:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r34_vd_tps_bilstm_att.yml -o Global.pretrained_model=./rec_r34_vd_tps_bilstm_att_v2.0_train/best_accuracy Global.save_inference_dir=./inference/rec_rare
+````
+
+RARE text recognition model inference, you can execute the following commands:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="doc/imgs_words/en/word_1.png" --rec_model_dir="./inference/rec_rare/" --rec_image_shape="3, 32, 100" --rec_char_dict_path= "./ppocr/utils/ic15_dict.txt"
+````
+
+The inference results are as follows:
+
+![img](./images/word_1-20240704184113913.png)
+
+````text linenums="1"
+Predicts of doc/imgs_words/en/word_1.png:('joint ', 0.9999969601631165)
+````
+
+### 4.2 C++ Inference
+
+Not currently supported
+
+### 4.3 Serving
+
+Not currently supported
+
+### 4.4 More
+
+The RARE model also supports the following inference deployment methods:
+
+- Paddle2ONNX Inference: After preparing the inference model, refer to the [paddle2onnx](../../../version2.x/legacy/paddle2onnx.en.md) tutorial.
+
+## 5. FAQ
+
+## Citation
+
+````bibtex
+@inproceedings{2016Robust,
+  title={Robust Scene Text Recognition with Automatic Rectification},
+  author={ Shi, B. and Wang, X. and Lyu, P. and Cong, Y. and Xiang, B. },
+  booktitle={2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
+  year={2016},
+}
+````
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_rare.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_rare.md
new file mode 100644
index 0000000..9bb5aa2
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_rare.md
@@ -0,0 +1,100 @@
+---
+comments: true
+---
+
+# RARE
+
+## 1. 算法简介
+
+论文信息：
+> [Robust Scene Text Recognition with Automatic Rectification](https://arxiv.org/abs/1603.03915v2)
+> Baoguang Shi, Xinggang Wang, Pengyuan Lyu, Cong Yao, Xiang Bai∗
+> CVPR, 2016
+
+使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法复现效果如下：
+
+|模型|骨干网络|配置文件|Avg Accuracy|下载链接|
+| --- | --- | --- | --- | --- |
+|RARE|Resnet34_vd|[configs/rec/rec_r34_vd_tps_bilstm_att.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r34_vd_tps_bilstm_att.yml)|83.60%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_att_v2.0_train.tar)|
+|RARE|MobileNetV3|[configs/rec/rec_mv3_tps_bilstm_att.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_mv3_tps_bilstm_att.yml)|82.50%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_tps_bilstm_att_v2.0_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练不同的识别模型只需要**更换配置文件**即可。以基于Resnet34_vd骨干网络为例:
+
+### 3.1 训练
+
+```bash linenums="1"
+# 单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_r34_vd_tps_bilstm_att.yml
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r34_vd_tps_bilstm_att.yml
+```
+
+### 3.2 评估
+
+```bash linenums="1"
+# GPU评估, Global.pretrained_model为待评估模型
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r34_vd_tps_bilstm_att.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### 3.3 预测
+
+```bash linenums="1"
+python3 tools/infer_rec.py -c configs/rec/rec_r34_vd_tps_bilstm_att.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将RARE文本识别训练过程中保存的模型，转换成inference model。以基于Resnet34_vd骨干网络，在MJSynth和SynthText两个文字识别数据集训练得到的模型为例（ [模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_att_v2.0_train.tar) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r34_vd_tps_bilstm_att.yml -o Global.pretrained_model=./rec_r34_vd_tps_bilstm_att_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/rec_rare
+```
+
+RARE文本识别模型推理，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="doc/imgs_words/en/word_1.png" --rec_model_dir="./inference/rec_rare/" --rec_image_shape="3, 32, 100" --rec_char_dict_path="./ppocr/utils/ic15_dict.txt"
+```
+
+推理结果如下所示：
+
+![img](./images/word_1-20240704184113913.png)
+
+```text linenums="1"
+Predicts of doc/imgs_words/en/word_1.png:('joint ', 0.9999969601631165)
+```
+
+### 4.2 C++推理
+
+暂不支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+RARE模型还支持以下推理部署方式：
+
+- Paddle2ONNX推理：准备好推理模型后，参考[paddle2onnx](../../../version2.x/legacy/paddle2onnx.md)教程操作。
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@inproceedings{2016Robust,
+  title={Robust Scene Text Recognition with Automatic Rectification},
+  author={ Shi, B.  and  Wang, X.  and  Lyu, P.  and  Cong, Y.  and  Xiang, B. },
+  booktitle={2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
+  year={2016},
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_rfl.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_rfl.en.md
new file mode 100644
index 0000000..0cf4f9a
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_rfl.en.md
@@ -0,0 +1,126 @@
+---
+comments: true
+---
+
+# RFL
+
+## 1. Introduction
+
+Paper:
+> [Reciprocal Feature Learning via Explicit and Implicit Tasks in Scene Text Recognition](https://arxiv.org/abs/2105.06229.pdf)
+> Hui Jiang, Yunlu Xu, Zhanzhan Cheng, Shiliang Pu, Yi Niu, Wenqi Ren, Fei Wu, and Wenming Tan
+> ICDAR, 2021
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Model|Backbone|config|Acc|Download link|
+| --- | --- | --- | --- | --- |
+|RFL-CNT|ResNetRFL|[rec_resnet_rfl_visual.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_resnet_rfl_visual.yml)|93.40%|[训练模型](https://paddleocr.bj.bcebos.com/contribution/rec_resnet_rfl_visual_train.tar)|
+|RFL-Att|ResNetRFL|[rec_resnet_rfl_att.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_resnet_rfl_att.yml)|88.63%|[训练模型](https://paddleocr.bj.bcebos.com/contribution/rec_resnet_rfl_att_train.tar)|
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+#step1:train the CNT branch
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_resnet_rfl_visual.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_resnet_rfl_visual.yml
+
+#step2:joint training of CNT and Att branches
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_resnet_rfl_att.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_resnet_rfl_att.yml  -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### Evaluation
+
+```bash linenums="1"
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_resnet_rfl_att.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### Prediction
+
+```bash linenums="1"
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_resnet_rfl_att.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the RFL text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/contribution/rec_resnet_rfl.tar)) ), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_resnet_rfl_att.yml -o Global.pretrained_model={path/to/weights}/best_accuracy  Global.save_inference_dir=./inference/rec_resnet_rfl_att
+```
+
+**Note:**
+
+- If you are training the model on your own dataset and have modified the dictionary file, please pay attention to modify the `character_dict_path` in the configuration file to the modified dictionary file.
+- If you modified the input size during training, please modify the `infer_shape` corresponding to NRTR in the `tools/export_model.py` file.
+
+After the conversion is successful, there are three files in the directory:
+
+```text linenums="1"
+/inference/rec_resnet_rfl_att/
+    ├── inference.pdiparams
+    ├── inference.pdiparams.info
+    └── inference.pdmodel
+```
+
+For RFL text recognition model inference, the following commands can be executed:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_resnet_rfl_att/' --rec_algorithm='RFL' --rec_image_shape='1,32,100'
+```
+
+![img](./images/word_10.png)
+
+After executing the command, the prediction result (recognized text and score) of the image above is printed to the screen, an example is as follows:
+The result is as follows:
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9999927282333374)
+```
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@article{2021Reciprocal,
+  title     = {Reciprocal Feature Learning via Explicit and Implicit Tasks in Scene Text Recognition},
+  author    = {Jiang, H.  and  Xu, Y.  and  Cheng, Z.  and  Pu, S.  and  Niu, Y.  and  Ren, W.  and  Wu, F.  and  Tan, W. },
+  booktitle = {ICDAR},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2105.06229}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_rfl.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_rfl.md
new file mode 100644
index 0000000..5e8ad87
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_rfl.md
@@ -0,0 +1,140 @@
+---
+comments: true
+---
+
+# 场景文本识别算法-RFL
+
+## 1. 算法简介
+
+论文信息：
+> [Reciprocal Feature Learning via Explicit and Implicit Tasks in Scene Text Recognition](https://arxiv.org/abs/2105.06229.pdf)
+> Hui Jiang, Yunlu Xu, Zhanzhan Cheng, Shiliang Pu, Yi Niu, Wenqi Ren, Fei Wu, and Wenming Tan
+> ICDAR, 2021
+
+`RFL`使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法复现效果如下：
+
+|模型|骨干网络|配置文件|Acc|下载链接|
+| --- | --- | --- | --- | --- |
+|RFL-CNT|ResNetRFL|[rec_resnet_rfl_visual.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_resnet_rfl_visual.yml)|93.40%|[训练模型](https://paddleocr.bj.bcebos.com/contribution/rec_resnet_rfl_visual_train.tar)|
+|RFL-Att|ResNetRFL|[rec_resnet_rfl_att.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_resnet_rfl_att.yml)|88.63%|[训练模型](https://paddleocr.bj.bcebos.com/contribution/rec_resnet_rfl_att_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+### 3.1 模型训练
+
+PaddleOCR对代码进行了模块化，训练`RFL`识别模型时需要**更换配置文件**为`RFL`的[配置文件](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_resnet_rfl_att.yml)。
+
+#### 启动训练
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+#step1:训练CNT分支
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_resnet_rfl_visual.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_resnet_rfl_visual.yml
+
+#step2:联合训练CNT和Att分支,注意将pretrained_model的路径设置为本地路径。
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_resnet_rfl_att.yml -o Global.pretrained_model=./output/rec/rec_resnet_rfl_visual/best_accuracy
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_resnet_rfl_att.yml  -o Global.pretrained_model=./output/rec/rec_resnet_rfl_visual/best_accuracy
+```
+
+### 3.2 评估
+
+可下载已训练完成的[模型文件](https://paddleocr.bj.bcebos.com/contribution/rec_resnet_rfl.tar)，使用如下命令进行评估：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_resnet_rfl_att.yml -o Global.pretrained_model=./output/rec/rec_resnet_rfl_att/best_accuracy
+```
+
+### 3.3 预测
+
+使用如下命令进行单张图片预测：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/infer_rec.py -c configs/rec/rec_resnet_rfl_att.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./output/rec/rec_resnet_rfl_att/best_accuracy
+# 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/imgs_words_en/'。
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将训练得到best模型，转换成inference model。这里以训练完成的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/contribution/rec_resnet_rfl.tar) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/export_model.py -c configs/rec/rec_resnet_rfl_att.yml -o Global.pretrained_model=./output/rec/rec_resnet_rfl_att/best_accuracy Global.save_inference_dir=./inference/rec_resnet_rfl_att/
+```
+
+**注意：** 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否是所需要的字典文件。
+
+- 如果您修改了训练时的输入大小，请修改`tools/export_model.py`文件中的对应RFL的`infer_shape`。
+
+转换成功后，在目录下有三个文件：
+
+```text linenums="1"
+/inference/rec_resnet_rfl_att/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+执行如下命令进行模型推理：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_resnet_rfl_att/' --rec_algorithm='RFL' --rec_image_shape='1,32,100'
+# 预测文件夹下所有图像时，可修改image_dir为文件夹，如 --image_dir='./doc/imgs_words_en/'。
+```
+
+![img](./images/word_10.png)
+
+执行命令后，上面图像的预测结果（识别的文本和得分）会打印到屏幕上，示例如下：
+结果如下：
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9999927282333374)
+```
+
+**注意**：
+
+- 训练上述模型采用的图像分辨率是[1，32，100]，需要通过参数`rec_image_shape`设置为您训练时的识别图像形状。
+- 在推理时需要设置参数`rec_char_dict_path`指定字典，如果您修改了字典，请修改该参数为您的字典文件。
+- 如果您修改了预处理方法，需修改`tools/infer/predict_rec.py`中RFL的预处理为您的预处理方法。
+
+### 4.2 C++推理部署
+
+由于C++预处理后处理还未支持RFL，所以暂未支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@article{2021Reciprocal,
+  title     = {Reciprocal Feature Learning via Explicit and Implicit Tasks in Scene Text Recognition},
+  author    = {Jiang, H.  and  Xu, Y.  and  Cheng, Z.  and  Pu, S.  and  Niu, Y.  and  Ren, W.  and  Wu, F.  and  Tan, W. },
+  booktitle = {ICDAR},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2105.06229}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_robustscanner.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_robustscanner.en.md
new file mode 100644
index 0000000..4e7c1db
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_robustscanner.en.md
@@ -0,0 +1,96 @@
+---
+comments: true
+---
+
+# RobustScanner
+
+## 1. Introduction
+
+Paper:
+> [RobustScanner: Dynamically Enhancing Positional Clues for Robust Text Recognition](https://arxiv.org/pdf/2007.07542.pdf)
+> Xiaoyu Yue, Zhanghui Kuang, Chenhao Lin, Hongbin Sun, Wayne
+Zhang
+> ECCV, 2020
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Model|Backbone|config|Acc|Download link|
+| --- | --- | --- | --- | --- |
+|RobustScanner|ResNet31|[rec_r31_robustscanner.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r31_robustscanner.yml)|87.77%|[trained model](https://paddleocr.bj.bcebos.com/contribution/rec_r31_robustscanner.tar)|
+
+Note:In addition to using the two text recognition datasets MJSynth and SynthText, [SynthAdd](https://pan.baidu.com/share/init?surl=uV0LtoNmcxbO-0YA7Ch4dg) data (extraction code: 627x), and some real data are used in training, the specific data details can refer to the paper.
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_r31_robustscanner.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r31_robustscanner.yml
+```
+
+### Evaluation
+
+```bash linenums="1"
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r31_robustscanner.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### Prediction
+
+```bash linenums="1"
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_r31_robustscanner.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the RobustScanner text recognition training process is converted into an inference model. you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r31_robustscanner.yml -o Global.pretrained_model={path/to/weights}/best_accuracy  Global.save_inference_dir=./inference/rec_r31_robustscanner
+```
+
+For RobustScanner text recognition model inference, the following commands can be executed:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/en/word_1.png" --rec_model_dir="./inference/rec_r31_robustscanner/" --rec_image_shape="3, 48, 48, 160" --rec_algorithm="RobustScanner" --rec_char_dict_path="ppocr/utils/dict90.txt" --use_space_char=False
+```
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@article{2020RobustScanner,
+  title={RobustScanner: Dynamically Enhancing Positional Clues for Robust Text Recognition},
+  author={Xiaoyu Yue and Zhanghui Kuang and Chenhao Lin and Hongbin Sun and Wayne Zhang},
+  journal={ECCV2020},
+  year={2020},
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_robustscanner.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_robustscanner.md
new file mode 100644
index 0000000..9574284
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_robustscanner.md
@@ -0,0 +1,97 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# RobustScanner
+
+## 1. 算法简介
+
+论文信息：
+> [RobustScanner: Dynamically Enhancing Positional Clues for Robust Text Recognition](https://arxiv.org/pdf/2007.07542.pdf)
+> Xiaoyu Yue, Zhanghui Kuang, Chenhao Lin, Hongbin Sun, Wayne
+Zhang
+> ECCV, 2020
+
+使用MJSynth和SynthText两个合成文字识别数据集训练，在IIIT, SVT, IC13, IC15, SVTP, CUTE数据集上进行评估，算法复现效果如下：
+
+|模型|骨干网络|配置文件|Acc|下载链接|
+| --- | --- | --- | --- | --- |
+|RobustScanner|ResNet31|[rec_r31_robustscanner.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r31_robustscanner.yml)|87.77%|[训练模型](https://paddleocr.bj.bcebos.com/contribution/rec_r31_robustscanner.tar)|
+
+注：除了使用MJSynth和SynthText两个文字识别数据集外，还加入了[SynthAdd](https://pan.baidu.com/share/init?surl=uV0LtoNmcxbO-0YA7Ch4dg)数据（提取码：627x），和部分真实数据，具体数据细节可以参考论文。
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+请参考[文本识别教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练不同的识别模型只需要**更换配置文件**即可。
+
+### 训练
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_r31_robustscanner.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r31_robustscanner.yml
+```
+
+### 评估
+
+```bash linenums="1"
+# GPU 评估， Global.pretrained_model 为待测权重
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r31_robustscanner.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### 预测
+
+```bash linenums="1"
+# 预测使用的配置文件必须与训练一致
+python3 tools/infer_rec.py -c configs/rec/rec_r31_robustscanner.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将RobustScanner文本识别训练过程中保存的模型，转换成inference model。可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r31_robustscanner.yml -o Global.pretrained_model={path/to/weights}/best_accuracy  Global.save_inference_dir=./inference/rec_r31_robustscanner
+```
+
+RobustScanner文本识别模型推理，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/en/word_1.png" --rec_model_dir="./inference/rec_r31_robustscanner/" --rec_image_shape="3, 48, 48, 160" --rec_algorithm="RobustScanner" --rec_char_dict_path="ppocr/utils/dict90.txt" --use_space_char=False
+```
+
+### 4.2 C++推理
+
+由于C++预处理后处理还未支持RobustScanner，所以暂未支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@article{2020RobustScanner,
+  title={RobustScanner: Dynamically Enhancing Positional Clues for Robust Text Recognition},
+  author={Xiaoyu Yue and Zhanghui Kuang and Chenhao Lin and Hongbin Sun and Wayne Zhang},
+  journal={ECCV2020},
+  year={2020},
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_rosetta.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_rosetta.en.md
new file mode 100644
index 0000000..fba5c6c
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_rosetta.en.md
@@ -0,0 +1,100 @@
+---
+comments: true
+---
+
+# Rosetta
+
+## 1. Introduction
+
+Paper information:
+> [Rosetta: Large Scale System for Text Detection and Recognition in Images](https://arxiv.org/abs/1910.05085)
+> Borisyuk F , Gordo A , V Sivakumar
+> KDD, 2018
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Models|Backbone Networks|Configuration Files|Avg Accuracy|Download Links|
+| --- | --- | --- | --- | --- |
+|Rosetta|Resnet34_vd|[configs/rec/rec_r34_vd_none_none_ctc.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r34_vd_none_none_ctc.yml)|79.11%|[training model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_none_ctc_v2.0_train.tar)|
+|Rosetta|MobileNetV3|[configs/rec/rec_mv3_none_none_ctc.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_mv3_none_none_ctc.yml)|75.80%|[training model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_none_ctc_v2.0_train.tar)|
+
+## 2. Environment
+
+Please refer to [Operating Environment Preparation](../../ppocr/environment.en.md) to configure the PaddleOCR operating environment, and refer to [Project Clone](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Training Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**. Take the backbone network based on Resnet34_vd as an example:
+
+### 3.1 Training
+
+```bash linenums="1"
+# Single card training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_r34_vd_none_none_ctc.yml
+# Multi-card training, specify the card number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/rec/rec_r34_vd_none_none_ctc.yml
+```
+
+### 3.2 Evaluation
+
+```bash linenums="1"
+# GPU evaluation, Global.pretrained_model is the model to be evaluated
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r34_vd_none_none_ctc.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### 3.3 Prediction
+
+```bash linenums="1"
+python3 tools/infer_rec.py -c configs/rec/rec_r34_vd_none_none_ctc.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, convert the model saved during the Rosetta text recognition training process into an inference model. Take the model trained on the MJSynth and SynthText text recognition datasets based on the Resnet34_vd backbone network as an example ( [Model download address](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_none_ctc_v2.0_train.tar) ), which can be converted using the following command:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r34_vd_none_none_ctc.yml -o Global.pretrained_model=./rec_r34_vd_none_none_ctc_v2.0_train/best_accuracy Global.save_inference_dir=./inference/rec_rosetta
+```
+
+Rosetta text recognition model inference, you can execute the following commands:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="doc/imgs_words/en/word_1.png" --rec_model_dir="./inference/rec_rosetta/" --rec_image_shape="3, 32, 100" --rec_char_dict_path= "./ppocr/utils/ic15_dict.txt"
+```
+
+The inference results are as follows:
+
+![img](./images/word_1-20240704183926496.png)
+
+```bash linenums="1"
+Predicts of doc/imgs_words/en/word_1.png:('joint', 0.9999982714653015)
+```
+
+### 4.2 C++ Inference
+
+Not currently supported
+
+### 4.3 Serving
+
+Not currently supported
+
+### 4.4 More
+
+The Rosetta model also supports the following inference deployment methods:
+
+- Paddle2ONNX Inference: After preparing the inference model, refer to the [paddle2onnx](../../../version2.x/legacy/paddle2onnx.en.md) tutorial.
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@inproceedings{2018Rosetta,
+  title={Rosetta: Large Scale System for Text Detection and Recognition in Images},
+  author={ Borisyuk, Fedor and Gordo, Albert and Sivakumar, Viswanath },
+  booktitle={the 24th ACM SIGKDD International Conference},
+  year={2018},
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_rosetta.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_rosetta.md
new file mode 100644
index 0000000..a82e5cf
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_rosetta.md
@@ -0,0 +1,101 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# Rosetta
+
+## 1. 算法简介
+
+论文信息：
+> [Rosetta: Large Scale System for Text Detection and Recognition in Images](https://arxiv.org/abs/1910.05085)
+> Borisyuk F ,  Gordo A ,  V  Sivakumar
+> KDD, 2018
+
+使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估， 算法复现效果如下：
+
+|模型|骨干网络|配置文件|Avg Accuracy|下载链接|
+| --- | --- | --- | --- | --- |
+|Rosetta|Resnet34_vd|[configs/rec/rec_r34_vd_none_none_ctc.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r34_vd_none_none_ctc.yml)|79.11%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_none_ctc_v2.0_train.tar)|
+|Rosetta|MobileNetV3|[configs/rec/rec_mv3_none_none_ctc.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_mv3_none_none_ctc.yml)|75.80%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_none_ctc_v2.0_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练不同的识别模型只需要**更换配置文件**即可。 以基于Resnet34_vd骨干网络为例:
+
+### 3.1 训练
+
+```bash linenums="1"
+# 单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_r34_vd_none_none_ctc.yml
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r34_vd_none_none_ctc.yml
+```
+
+### 3.2 评估
+
+```bash linenums="1"
+# GPU评估, Global.pretrained_model为待评估模型
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r34_vd_none_none_ctc.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### 3.3 预测
+
+```bash linenums="1"
+python3 tools/infer_rec.py -c configs/rec/rec_r34_vd_none_none_ctc.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将Rosetta文本识别训练过程中保存的模型，转换成inference model。以基于Resnet34_vd骨干网络，在MJSynth和SynthText两个文字识别数据集训练得到的模型为例（ [模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_none_ctc_v2.0_train.tar) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r34_vd_none_none_ctc.yml -o Global.pretrained_model=./rec_r34_vd_none_none_ctc_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/rec_rosetta
+```
+
+Rosetta文本识别模型推理，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="doc/imgs_words/en/word_1.png" --rec_model_dir="./inference/rec_rosetta/" --rec_image_shape="3, 32, 100" --rec_char_dict_path="./ppocr/utils/ic15_dict.txt"
+```
+
+推理结果如下所示：
+
+![img](./images/word_1-20240704184113913.png)
+
+```bash linenums="1"
+Predicts of doc/imgs_words/en/word_1.png:('joint', 0.9999982714653015)
+```
+
+### 4.2 C++推理
+
+暂不支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+Rosetta模型还支持以下推理部署方式：
+
+- Paddle2ONNX推理：准备好推理模型后，参考[paddle2onnx](../../../version2.x/legacy/paddle2onnx.md)教程操作。
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@inproceedings{2018Rosetta,
+  title={Rosetta: Large Scale System for Text Detection and Recognition in Images},
+  author={ Borisyuk, Fedor  and  Gordo, Albert  and  Sivakumar, Viswanath },
+  booktitle={the 24th ACM SIGKDD International Conference},
+  year={2018},
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_sar.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_sar.en.md
new file mode 100644
index 0000000..df8c75e
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_sar.en.md
@@ -0,0 +1,96 @@
+---
+comments: true
+---
+
+# SAR
+
+## 1. Introduction
+
+Paper:
+> [Show, Attend and Read: A Simple and Strong Baseline for Irregular Text Recognition](https://arxiv.org/abs/1811.00751)
+> Hui Li, Peng Wang, Chunhua Shen, Guyu Zhang
+> AAAI, 2019
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Model|Backbone|config|Acc|Download link|
+| --- | --- | --- | --- | --- |
+|SAR|ResNet31|[rec_r31_sar.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r31_sar.yml)|87.20%|[train model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_sar_train.tar)|
+
+Note:In addition to using the two text recognition datasets MJSynth and SynthText, [SynthAdd](https://pan.baidu.com/share/init?surl=uV0LtoNmcxbO-0YA7Ch4dg) data (extraction code: 627x), and some real data are used in training, the specific data details can refer to the paper.
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_r31_sar.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r31_sar.yml
+```
+
+### Evaluation
+
+```bash linenums="1"
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r31_sar.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### Prediction
+
+```bash linenums="1"
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_r31_sar.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the SAR text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_sar_train.tar) ), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r31_sar.yml -o Global.pretrained_model=./rec_r31_sar_train/best_accuracy  Global.save_inference_dir=./inference/rec_sar
+```
+
+For SAR text recognition model inference, the following commands can be executed:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/en/word_1.png" --rec_model_dir="./inference/rec_sar/" --rec_image_shape="3, 48, 48, 160" --rec_algorithm="SAR" --rec_char_dict_path="ppocr/utils/dict90.txt" --max_text_length=30 --use_space_char=False
+```
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@article{Li2019ShowAA,
+  title={Show, Attend and Read: A Simple and Strong Baseline for Irregular Text Recognition},
+  author={Hui Li and Peng Wang and Chunhua Shen and Guyu Zhang},
+  journal={ArXiv},
+  year={2019},
+  volume={abs/1811.00751}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_sar.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_sar.md
new file mode 100644
index 0000000..d74a213
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_sar.md
@@ -0,0 +1,96 @@
+---
+comments: true
+---
+
+# SAR
+
+## 1. 算法简介
+
+论文信息：
+> [Show, Attend and Read: A Simple and Strong Baseline for Irregular Text Recognition](https://arxiv.org/abs/1811.00751)
+> Hui Li, Peng Wang, Chunhua Shen, Guyu Zhang
+> AAAI, 2019
+
+使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法复现效果如下：
+
+|模型|骨干网络|配置文件|Acc|下载链接|
+| --- | --- | --- | --- | --- |
+|SAR|ResNet31|[rec_r31_sar.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r31_sar.yml)|87.20%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_sar_train.tar)|
+
+注：除了使用MJSynth和SynthText两个文字识别数据集外，还加入了[SynthAdd](https://pan.baidu.com/share/init?surl=uV0LtoNmcxbO-0YA7Ch4dg)数据（提取码：627x），和部分真实数据，具体数据细节可以参考论文。
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+请参考[文本识别教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练不同的识别模型只需要**更换配置文件**即可。
+
+### 训练
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_r31_sar.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r31_sar.yml
+```
+
+### 评估
+
+```bash linenums="1"
+# GPU 评估， Global.pretrained_model 为待测权重
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r31_sar.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### 预测
+
+```bash linenums="1"
+# 预测使用的配置文件必须与训练一致
+python3 tools/infer_rec.py -c configs/rec/rec_r31_sar.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将SAR文本识别训练过程中保存的模型，转换成inference model。（ [模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_sar_train.tar) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r31_sar.yml -o Global.pretrained_model=./rec_r31_sar_train/best_accuracy  Global.save_inference_dir=./inference/rec_sar
+```
+
+SAR文本识别模型推理，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/en/word_1.png" --rec_model_dir="./inference/rec_sar/" --rec_image_shape="3, 48, 48, 160" --rec_algorithm="SAR" --rec_char_dict_path="ppocr/utils/dict90.txt" --max_text_length=30 --use_space_char=False
+```
+
+### 4.2 C++推理
+
+由于C++预处理后处理还未支持SAR，所以暂未支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@article{Li2019ShowAA,
+  title={Show, Attend and Read: A Simple and Strong Baseline for Irregular Text Recognition},
+  author={Hui Li and Peng Wang and Chunhua Shen and Guyu Zhang},
+  journal={ArXiv},
+  year={2019},
+  volume={abs/1811.00751}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_satrn.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_satrn.en.md
new file mode 100644
index 0000000..25ec13a
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_satrn.en.md
@@ -0,0 +1,93 @@
+---
+comments: true
+---
+
+# SATRN
+
+## 1. Introduction
+
+论文信息：
+> [On Recognizing Texts of Arbitrary Shapes with 2D Self-Attention](https://arxiv.org/abs/1910.04396)
+> Junyeop Lee, Sungrae Park, Jeonghun Baek, Seong Joon Oh, Seonghyeon Kim, Hwalsuk Lee
+> CVPR, 2020
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Model|Backbone|config|Acc|Download link|
+| --- | --- | --- | --- | --- |
+|SATRN|ShallowCNN|88.05%|[configs/rec/rec_satrn.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_satrn.yml)|[训练模型](https://pan.baidu.com/s/10J-Bsd881bimKaclKszlaQ?pwd=lk8a)|
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_satrn.yml
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_satrn.yml
+```
+
+### Evaluation
+
+```bash linenums="1"
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_satrn.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### Prediction
+
+```bash linenums="1"
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_satrn.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the SATRN text recognition training process is converted into an inference model. ( [Model download link](https://pan.baidu.com/s/10J-Bsd881bimKaclKszlaQ?pwd=lk8a) ), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_satrn.yml -o Global.pretrained_model=./rec_satrn_train/best_accuracy  Global.save_inference_dir=./inference/rec_satrn
+```
+
+For SATRN text recognition model inference, the following commands can be executed:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/en/word_1.png" --rec_model_dir="./inference/rec_satrn/" --rec_image_shape="3, 48, 48, 160" --rec_algorithm="SATRN" --rec_char_dict_path="ppocr/utils/dict90.txt" --max_text_length=30 --use_space_char=False
+```
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@article{lee2019recognizing,
+      title={On Recognizing Texts of Arbitrary Shapes with 2D Self-Attention},
+      author={Junyeop Lee and Sungrae Park and Jeonghun Baek and Seong Joon Oh and Seonghyeon Kim and Hwalsuk Lee},
+      year={2019},
+      eprint={1910.04396},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_satrn.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_satrn.md
new file mode 100644
index 0000000..ab848eb
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_satrn.md
@@ -0,0 +1,93 @@
+---
+comments: true
+---
+
+# SATRN
+
+## 1. 算法简介
+
+论文信息：
+> [On Recognizing Texts of Arbitrary Shapes with 2D Self-Attention](https://arxiv.org/abs/1910.04396)
+> Junyeop Lee, Sungrae Park, Jeonghun Baek, Seong Joon Oh, Seonghyeon Kim, Hwalsuk Lee
+> CVPR, 2020
+参考[DTRB](https://arxiv.org/abs/1904.01906) 文字识别训练和评估流程，使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法效果如下：
+
+|模型|骨干网络|Avg Accuracy|配置文件|下载链接|
+|---|---|---|---|---|
+|SATRN|ShallowCNN|88.05%|[configs/rec/rec_satrn.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_satrn.yml)|[训练模型](https://pan.baidu.com/s/10J-Bsd881bimKaclKszlaQ?pwd=lk8a)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练不同的识别模型只需要**更换配置文件**即可。
+
+### 训练
+
+在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+# 单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_satrn.yml
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c rec_satrn.yml
+```
+
+### 评估
+
+```bash linenums="1"
+# GPU 评估， Global.pretrained_model 为待测权重
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_satrn.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### 预测
+
+```bash linenums="1"
+# 预测使用的配置文件必须与训练一致
+python3 tools/infer_rec.py -c configs/rec/rec_satrn.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将SATRN文本识别训练过程中保存的模型，转换成inference model。（ [模型下载地址](https://pan.baidu.com/s/10J-Bsd881bimKaclKszlaQ?pwd=lk8a) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_satrn.yml -o Global.pretrained_model=./rec_satrn/best_accuracy  Global.save_inference_dir=./inference/rec_satrn
+```
+
+SATRN文本识别模型推理，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/en/word_1.png" --rec_model_dir="./inference/rec_satrn/" --rec_image_shape="3, 48, 48, 160" --rec_algorithm="SATRN" --rec_char_dict_path="ppocr/utils/dict90.txt" --max_text_length=30 --use_space_char=False
+```
+
+### 4.2 C++推理
+
+由于C++预处理后处理还未支持SATRN，所以暂未支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@article{lee2019recognizing,
+      title={On Recognizing Texts of Arbitrary Shapes with 2D Self-Attention},
+      author={Junyeop Lee and Sungrae Park and Jeonghun Baek and Seong Joon Oh and Seonghyeon Kim and Hwalsuk Lee},
+      year={2019},
+      eprint={1910.04396},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_seed.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_seed.en.md
new file mode 100644
index 0000000..30a3511
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_seed.en.md
@@ -0,0 +1,92 @@
+---
+comments: true
+---
+
+# SEED
+
+## 1. Introduction
+
+Paper:
+> [SEED: Semantics Enhanced Encoder-Decoder Framework for Scene Text Recognition](https://arxiv.org/pdf/2005.10977.pdf)
+
+> Qiao, Zhi and Zhou, Yu and Yang, Dongbao and Zhou, Yucan and Wang, Weiping
+
+> CVPR, 2020
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Model|Backbone|ACC|config|Download link|
+| --- | --- | --- | --- | --- |
+|SEED|Aster_Resnet| 85.20% | [configs/rec/rec_resnet_stn_bilstm_att.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_resnet_stn_bilstm_att.yml) | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_resnet_stn_bilstm_att.tar) |
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+### Training
+
+The SEED model needs to additionally load the [language model](https://dl.fbaipublicfiles.com/fasttext/vectors-crawl/cc.en.300.bin.gz) trained by FastText, and install the fasttext dependencies:
+
+```bash linenums="1"
+python3 -m pip install fasttext==0.9.1
+```
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_resnet_stn_bilstm_att.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c rec_resnet_stn_bilstm_att.yml
+```
+
+### Evaluation
+
+```bash linenums="1"
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_resnet_stn_bilstm_att.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### Prediction
+
+```bash linenums="1"
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_resnet_stn_bilstm_att.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+Not support
+
+### 4.2 C++ Inference
+
+Not support
+
+### 4.3 Serving
+
+Not support
+
+### 4.4 More
+
+Not support
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@inproceedings{qiao2020seed,
+  title={Seed: Semantics enhanced encoder-decoder framework for scene text recognition},
+  author={Qiao, Zhi and Zhou, Yu and Yang, Dongbao and Zhou, Yucan and Wang, Weiping},
+  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
+  pages={13528--13537},
+  year={2020}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_seed.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_seed.md
new file mode 100644
index 0000000..af45f6c
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_seed.md
@@ -0,0 +1,91 @@
+---
+comments: true
+---
+
+# SEED
+
+## 1. 算法简介
+
+论文信息：
+> [SEED: Semantics Enhanced Encoder-Decoder Framework for Scene Text Recognition](https://arxiv.org/pdf/2005.10977.pdf)
+> Qiao, Zhi and Zhou, Yu and Yang, Dongbao and Zhou, Yucan and Wang, Weiping
+> CVPR, 2020
+
+参考[DTRB](https://arxiv.org/abs/1904.01906) 文字识别训练和评估流程，使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法效果如下：
+
+|模型|骨干网络|Avg Accuracy|配置文件|下载链接|
+|---|---|---|---|---|
+|SEED|Aster_Resnet| 85.20% | [configs/rec/rec_resnet_stn_bilstm_att.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_resnet_stn_bilstm_att.yml) | [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_resnet_stn_bilstm_att.tar) |
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练不同的识别模型只需要**更换配置文件**即可。
+
+### 训练
+
+SEED模型需要额外加载FastText训练好的[语言模型](https://dl.fbaipublicfiles.com/fasttext/vectors-crawl/cc.en.300.bin.gz) ,并且安装 fasttext 依赖：
+
+```bash linenums="1"
+python3 -m pip install fasttext==0.9.1
+```
+
+然后，在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+# 单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_resnet_stn_bilstm_att.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c rec_resnet_stn_bilstm_att.yml
+
+```
+
+### 评估
+
+```bash linenums="1"
+# GPU 评估， Global.pretrained_model 为待测权重
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_resnet_stn_bilstm_att.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### 预测
+
+```bash linenums="1"
+# 预测使用的配置文件必须与训练一致
+python3 tools/infer_rec.py -c configs/rec/rec_resnet_stn_bilstm_att.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+coming soon
+
+### 4.2 C++推理
+
+coming soon
+
+### 4.3 Serving服务化部署
+
+coming soon
+
+### 4.4 更多推理部署
+
+coming soon
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@inproceedings{qiao2020seed,
+  title={Seed: Semantics enhanced encoder-decoder framework for scene text recognition},
+  author={Qiao, Zhi and Zhou, Yu and Yang, Dongbao and Zhou, Yucan and Wang, Weiping},
+  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
+  pages={13528--13537},
+  year={2020}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_spin.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_spin.en.md
new file mode 100644
index 0000000..7e01f70
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_spin.en.md
@@ -0,0 +1,93 @@
+---
+comments: true
+---
+
+# SPIN: Structure-Preserving Inner Offset Network for Scene Text Recognition
+
+## 1. Introduction
+
+Paper:
+> [SPIN: Structure-Preserving Inner Offset Network for Scene Text Recognition](https://arxiv.org/abs/2005.13117)
+> Chengwei Zhang, Yunlu Xu, Zhanzhan Cheng, Shiliang Pu, Yi Niu, Fei Wu, Futai Zou
+> AAAI, 2020
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets. The algorithm reproduction effect is as follows:
+
+|Model|Backbone|config|Acc|Download link|
+| --- | --- | --- | --- | --- |
+|SPIN|ResNet32|[rec_r32_gaspin_bilstm_att.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r32_gaspin_bilstm_att.yml)|90.00%|[trained model](https://paddleocr.bj.bcebos.com/contribution/rec_r32_gaspin_bilstm_att.tar) |
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_r32_gaspin_bilstm_att.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r32_gaspin_bilstm_att.yml
+```
+
+### Evaluation
+
+```bash linenums="1"
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r32_gaspin_bilstm_att.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### Prediction
+
+```bash linenums="1"
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_r32_gaspin_bilstm_att.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the SPIN text recognition training process is converted into an inference model. you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r32_gaspin_bilstm_att.yml -o Global.pretrained_model={path/to/weights}/best_accuracy  Global.save_inference_dir=./inference/rec_r32_gaspin_bilstm_att
+```
+
+For SPIN text recognition model inference, the following commands can be executed:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/en/word_1.png" --rec_model_dir="./inference/rec_r32_gaspin_bilstm_att/" --rec_image_shape="3, 32, 100" --rec_algorithm="SPIN" --rec_char_dict_path="/ppocr/utils/dict/spin_dict.txt" --use_space_char=False
+```
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@article{2020SPIN,
+  title={SPIN: Structure-Preserving Inner Offset Network for Scene Text Recognition},
+  author={Chengwei Zhang and Yunlu Xu and Zhanzhan Cheng and Shiliang Pu and Yi Niu and Fei Wu and Futai Zou},
+  journal={AAAI2020},
+  year={2020},
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_spin.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_spin.md
new file mode 100644
index 0000000..3d109f2
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_spin.md
@@ -0,0 +1,94 @@
+---
+comments: true
+---
+
+# SPIN: Structure-Preserving Inner Offset Network for Scene Text Recognition
+
+## 1. 算法简介
+
+论文信息：
+> [SPIN: Structure-Preserving Inner Offset Network for Scene Text Recognition](https://arxiv.org/abs/2005.13117)
+> Chengwei Zhang, Yunlu Xu, Zhanzhan Cheng, Shiliang Pu, Yi Niu, Fei Wu, Futai Zou
+> AAAI, 2020
+
+SPIN收录于AAAI2020。主要用于OCR识别任务。在任意形状文本识别中，矫正网络是一种较为常见的前置处理模块，但诸如RARE\ASTER\ESIR等只考虑了空间变换，并没有考虑色度变换。本文提出了一种结构Structure-Preserving Inner Offset Network (SPIN)，可以在色彩空间上进行变换。该模块是可微分的，可以加入到任意识别器中。
+使用MJSynth和SynthText两个合成文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法复现效果如下：
+
+|模型|骨干网络|配置文件|Acc|下载链接|
+| --- | --- | --- | --- | --- |
+|SPIN|ResNet32|[rec_r32_gaspin_bilstm_att.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r32_gaspin_bilstm_att.yml)|90.00%|[训练模型](https://paddleocr.bj.bcebos.com/contribution/rec_r32_gaspin_bilstm_att.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+请参考[文本识别教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练不同的识别模型只需要**更换配置文件**即可。
+
+### 训练
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_r32_gaspin_bilstm_att.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r32_gaspin_bilstm_att.yml
+```
+
+### 评估
+
+```bash linenums="1"
+# GPU 评估， Global.pretrained_model 为待测权重
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r32_gaspin_bilstm_att.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### 预测
+
+```bash linenums="1"
+# 预测使用的配置文件必须与训练一致
+python3 tools/infer_rec.py -c configs/rec/rec_r32_gaspin_bilstm_att.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将SPIN文本识别训练过程中保存的模型，转换成inference model。可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r32_gaspin_bilstm_att.yml -o Global.pretrained_model={path/to/weights}/best_accuracy  Global.save_inference_dir=./inference/rec_r32_gaspin_bilstm_att
+```
+
+SPIN文本识别模型推理，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/en/word_1.png" --rec_model_dir="./inference/rec_r32_gaspin_bilstm_att/" --rec_image_shape="3, 32, 100" --rec_algorithm="SPIN" --rec_char_dict_path="/ppocr/utils/dict/spin_dict.txt" --use_space_char=Falsee
+```
+
+### 4.2 C++推理
+
+由于C++预处理后处理还未支持SPIN，所以暂未支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@article{2020SPIN,
+  title={SPIN: Structure-Preserving Inner Offset Network for Scene Text Recognition},
+  author={Chengwei Zhang and Yunlu Xu and Zhanzhan Cheng and Shiliang Pu and Yi Niu and Fei Wu and Futai Zou},
+  journal={AAAI2020},
+  year={2020},
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_srn.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_srn.en.md
new file mode 100644
index 0000000..ce1b0d1
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_srn.en.md
@@ -0,0 +1,94 @@
+---
+comments: true
+---
+
+# SRN
+
+## 1. Introduction
+
+Paper:
+> [Towards Accurate Scene Text Recognition with Semantic Reasoning Networks](https://arxiv.org/abs/2003.12294#)
+> Deli Yu, Xuan Li, Chengquan Zhang, Junyu Han, Jingtuo Liu, Errui Ding
+> CVPR,2020
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Model|Backbone|config|Acc|Download link|
+| --- | --- | --- | --- | --- |
+|SRN|Resnet50_vd_fpn|[rec_r50_fpn_srn.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r50_fpn_srn.yml)|86.31%|[train model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r50_vd_srn_train.tar)|
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_r50_fpn_srn.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r50_fpn_srn.yml
+```
+
+### Evaluation
+
+```bash linenums="1"
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r50_fpn_srn.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### Prediction
+
+```bash linenums="1"
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_r50_fpn_srn.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the SRN text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r50_vd_srn_train.tar) ), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r50_fpn_srn.yml -o Global.pretrained_model=./rec_r50_vd_srn_train/best_accuracy  Global.save_inference_dir=./inference/rec_srn
+```
+
+For SRN text recognition model inference, the following commands can be executed:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/en/word_1.png" --rec_model_dir="./inference/rec_srn/" --rec_image_shape="1,64,256" --rec_char_type="ch" --rec_algorithm="SRN" --rec_char_dict_path="ppocr/utils/ic15_dict.txt" --use_space_char=False
+```
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@article{Yu2020TowardsAS,
+  title={Towards Accurate Scene Text Recognition With Semantic Reasoning Networks},
+  author={Deli Yu and Xuan Li and Chengquan Zhang and Junyu Han and Jingtuo Liu and Errui Ding},
+  journal={2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
+  year={2020},
+  pages={12110-12119}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_srn.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_srn.md
new file mode 100644
index 0000000..f5876df
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_srn.md
@@ -0,0 +1,94 @@
+---
+comments: true
+---
+
+# SRN
+
+## 1. 算法简介
+
+论文信息：
+> [Towards Accurate Scene Text Recognition with Semantic Reasoning Networks](https://arxiv.org/abs/2003.12294#)
+> Deli Yu, Xuan Li, Chengquan Zhang, Junyu Han, Jingtuo Liu, Errui Ding
+> CVPR,2020
+
+使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法复现效果如下：
+
+|模型|骨干网络|配置文件|Acc|下载链接|
+| --- | --- | --- | --- | --- |
+|SRN|Resnet50_vd_fpn|[rec_r50_fpn_srn.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r50_fpn_srn.yml)|86.31%|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r50_vd_srn_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+请参考[文本识别教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练不同的识别模型只需要**更换配置文件**即可。
+
+### 训练
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_r50_fpn_srn.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r50_fpn_srn.yml
+```
+
+### 评估
+
+```bash linenums="1"
+# GPU 评估， Global.pretrained_model 为待测权重
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r50_fpn_srn.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### 预测
+
+```bash linenums="1"
+# 预测使用的配置文件必须与训练一致
+python3 tools/infer_rec.py -c configs/rec/rec_r50_fpn_srn.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将SRN文本识别训练过程中保存的模型，转换成inference model。（ [模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r50_vd_srn_train.tar) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r50_fpn_srn.yml -o Global.pretrained_model=./rec_r50_vd_srn_train/best_accuracy  Global.save_inference_dir=./inference/rec_srn
+```
+
+SRN文本识别模型推理，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/en/word_1.png" --rec_model_dir="./inference/rec_srn/" --rec_image_shape="1,64,256"  --rec_algorithm="SRN" --rec_char_dict_path=./ppocr/utils/ic15_dict.txt  --use_space_char=False
+```
+
+### 4.2 C++推理
+
+由于C++预处理后处理还未支持SRN，所以暂未支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@article{Yu2020TowardsAS,
+  title={Towards Accurate Scene Text Recognition With Semantic Reasoning Networks},
+  author={Deli Yu and Xuan Li and Chengquan Zhang and Junyu Han and Jingtuo Liu and Errui Ding},
+  journal={2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
+  year={2020},
+  pages={12110-12119}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_starnet.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_starnet.en.md
new file mode 100644
index 0000000..cfe2886
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_starnet.en.md
@@ -0,0 +1,115 @@
+---
+comments: true
+---
+
+# STAR-Net
+
+## 1. Introduction
+
+Paper information:
+> [STAR-Net: a spatial attention residue network for scene text recognition.](https://www.bmva-archive.org.uk/bmvc/2016/papers/paper043/paper043.pdf)
+> Wei Liu, Chaofeng Chen, Kwan-Yee K. Wong, Zhizhong Su and Junyu Han.
+> BMVC, pages 43.1-43.13, 2016
+
+Refer to [DTRB](https://arxiv.org/abs/1904.01906) text Recognition Training and Evaluation Process . Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Models|Backbone Networks|Avg Accuracy|Configuration Files|Download Links|
+| --- | --- | --- | --- | --- |
+|StarNet|Resnet34_vd|84.44%|[configs/rec/rec_r34_vd_tps_bilstm_ctc.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r34_vd_tps_bilstm_ctc.yml)|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_ctc_v2.0_train.tar)|
+|StarNet|MobileNetV3|81.42%|[configs/rec/rec_mv3_tps_bilstm_ctc.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_mv3_tps_bilstm_ctc.yml)|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_tps_bilstm_ctc_v2.0_train.tar)|
+
+## 2. Environment
+
+Please refer to [Operating Environment Preparation](../../ppocr/environment.en.md) to configure the PaddleOCR operating environment, and refer to [Project Clone](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Training Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**. Take the backbone network based on Resnet34_vd as an example:
+
+### 3.1 Training
+
+After the data preparation is complete, the training can be started. The training command is as follows:
+
+````bash linenums="1"
+#  Single card training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_r34_vd_tps_bilstm_ctc.yml # Multi-card training, specify the card number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c rec_r34_vd_tps_bilstm_ctc.yml
+````
+
+### 3.2 Evaluation
+
+````bash linenums="1"
+# GPU evaluation, Global.pretrained_model is the model to be evaluated
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r34_vd_tps_bilstm_ctc.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+````
+
+### 3.3 Prediction
+
+````bash linenums="1"
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_r34_vd_tps_bilstm_ctc.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+````
+
+## 4. Inference
+
+### 4.1 Python Inference
+
+First, convert the model saved during the STAR-Net text recognition training process into an inference model. Take the model trained on the MJSynth and SynthText text recognition datasets based on the Resnet34_vd backbone network as an example [Model download address]( https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_bilstm_ctc_v2.0_train.tar) , which can be converted using the following command:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r34_vd_tps_bilstm_ctc.yml -o Global.pretrained_model=./rec_r34_vd_tps_bilstm_ctc_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/rec_starnet
+```
+
+STAR-Net text recognition model inference, you can execute the following commands:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_336.png" --rec_model_dir="./inference/rec_starnet/" --rec_image_shape="3, 32, 100" --rec_char_dict_path="./ppocr/utils/ic15_dict.txt"
+```
+
+![img](./images/word_336.png)
+
+The inference results are as follows:
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words_en/word_336.png:('super', 0.9999073)
+```
+
+**Attention** Since the above model refers to the [DTRB](https://arxiv.org/abs/1904.01906) text recognition training and evaluation process, it is different from the ultra-lightweight Chinese recognition model training in two aspects:
+
+- The image resolutions used during training are different. The image resolutions used for training the above models are [3, 32, 100], while for Chinese model training, in order to ensure the recognition effect of long texts, the image resolutions used during training are [ 3, 32, 320]. The default shape parameter of the predictive inference program is the image resolution used for training Chinese, i.e. [3, 32, 320]. Therefore, when inferring the above English model here, it is necessary to set the shape of the recognized image through the parameter rec_image_shape.
+
+- Character list, the experiment in the DTRB paper is only for 26 lowercase English letters and 10 numbers, a total of 36 characters. All uppercase and lowercase characters are converted to lowercase characters, and characters not listed above are ignored and considered spaces. Therefore, there is no input character dictionary here, but a dictionary is generated by the following command. Therefore, the parameter rec_char_dict_path needs to be set during inference, which is specified as an English dictionary "./ppocr/utils/ic15_dict.txt".
+
+```python linenums="1"
+self.character_str = "0123456789abcdefghijklmnopqrstuvwxyz"
+dict_character = list(self.character_str)
+```
+
+### 4.2 C++ Inference
+
+After preparing the inference model, refer to the [cpp infer](../../../version2.x/legacy/cpp_infer.en.md) tutorial to operate.
+
+### 4.3 Serving
+
+After preparing the inference model, refer to the [pdserving](../../../version2.x/legacy/paddle_server.en.md) tutorial for Serving deployment, including two modes: Python Serving and C++ Serving.
+
+### 4.4 More
+
+The STAR-Net model also supports the following inference deployment methods:
+
+- Paddle2ONNX Inference: After preparing the inference model, refer to the [paddle2onnx](../../../version2.x/legacy/paddle2onnx.en.md) tutorial.
+
+## 5. FAQ
+
+## Citation
+
+```bibtex
+@inproceedings{liu2016star,
+  title={STAR-Net: a spatial attention residue network for scene text recognition.},
+  author={Liu, Wei and Chen, Chaofeng and Wong, Kwan-Yee K and Su, Zhizhong and Han, Junyu},
+  booktitle={BMVC},
+  volume={2},
+  pages={7},
+  year={2016}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_starnet.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_starnet.md
new file mode 100644
index 0000000..97a65be
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_starnet.md
@@ -0,0 +1,118 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# STAR-Net
+
+## 1. 算法简介
+
+论文信息：
+> [STAR-Net: a spatial attention residue network for scene text recognition.](https://www.bmva-archive.org.uk/bmvc/2016/papers/paper043/paper043.pdf)
+> Wei Liu, Chaofeng Chen, Kwan-Yee K. Wong, Zhizhong Su and Junyu Han.
+> BMVC, pages 43.1-43.13, 2016
+
+参考[DTRB](https://arxiv.org/abs/1904.01906) 文字识别训练和评估流程，使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法效果如下：
+
+|模型|骨干网络|Avg Accuracy|配置文件|下载链接|
+|---|---|---|---|---|
+|StarNet|Resnet34_vd|84.44%|[configs/rec/rec_r34_vd_tps_bilstm_ctc.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r34_vd_tps_bilstm_ctc.yml)|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_ctc_v2.0_train.tar)|
+|StarNet|MobileNetV3|81.42%|[configs/rec/rec_mv3_tps_bilstm_ctc.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_mv3_tps_bilstm_ctc.yml)|[训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_tps_bilstm_ctc_v2.0_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练不同的识别模型只需要**更换配置文件**即可。
+
+### 训练
+
+在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+# 单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_r34_vd_tps_bilstm_ctc.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c rec_r34_vd_tps_bilstm_ctc.yml
+```
+
+### 评估
+
+```bash linenums="1"
+# GPU 评估， Global.pretrained_model 为待测权重
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_r34_vd_tps_bilstm_ctc.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### 预测
+
+```bash linenums="1"
+# 预测使用的配置文件必须与训练一致
+python3 tools/infer_rec.py -c configs/rec/rec_r34_vd_tps_bilstm_ctc.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将 STAR-Net 文本识别训练过程中保存的模型，转换成inference model。以基于Resnet34_vd骨干网络，使用MJSynth和SynthText两个英文文本识别合成数据集训练的[模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_bilstm_ctc_v2.0_train.tar) 为例，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r34_vd_tps_bilstm_ctc.yml -o Global.pretrained_model=./rec_r34_vd_tps_bilstm_ctc_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/rec_starnet
+```
+
+STAR-Net 文本识别模型推理，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_336.png" --rec_model_dir="./inference/rec_starnet/" --rec_image_shape="3, 32, 100" --rec_char_dict_path="./ppocr/utils/ic15_dict.txt"
+```
+
+![img](./images/word_336.png)
+
+执行命令后，上面图像的识别结果如下：
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words_en/word_336.png:('super', 0.9999073)
+```
+
+**注意**：由于上述模型是参考[DTRB](https://arxiv.org/abs/1904.01906)文本识别训练和评估流程，与超轻量级中文识别模型训练有两方面不同：
+
+- 训练时采用的图像分辨率不同，训练上述模型采用的图像分辨率是[3，32，100]，而中文模型训练时，为了保证长文本的识别效果，训练时采用的图像分辨率是[3, 32, 320]。预测推理程序默认的形状参数是训练中文采用的图像分辨率，即[3, 32, 320]。因此，这里推理上述英文模型时，需要通过参数rec_image_shape设置识别图像的形状。
+
+- 字符列表，DTRB论文中实验只是针对26个小写英文本母和10个数字进行实验，总共36个字符。所有大小字符都转成了小写字符，不在上面列表的字符都忽略，认为是空格。因此这里没有输入字符字典，而是通过如下命令生成字典.因此在推理时需要设置参数rec_char_dict_path，指定为英文字典"./ppocr/utils/ic15_dict.txt"。
+
+```python linenums="1"
+self.character_str = "0123456789abcdefghijklmnopqrstuvwxyz"
+dict_character = list(self.character_str)
+```
+
+### 4.2 C++推理
+
+准备好推理模型后，参考[cpp infer](../../../version2.x/legacy/cpp_infer.md)教程进行操作即可。
+
+### 4.3 Serving服务化部署
+
+准备好推理模型后，参考[pdserving](../../../version2.x/legacy/paddle_server.md)教程进行Serving服务化部署，包括Python Serving和C++ Serving两种模式。
+
+### 4.4 更多推理部署
+
+STAR-Net模型还支持以下推理部署方式：
+
+- Paddle2ONNX推理：准备好推理模型后，参考[paddle2onnx](../../../version2.x/legacy/paddle2onnx.md)教程操作。
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@inproceedings{liu2016star,
+  title={STAR-Net: a spatial attention residue network for scene text recognition.},
+  author={Liu, Wei and Chen, Chaofeng and Wong, Kwan-Yee K and Su, Zhizhong and Han, Junyu},
+  booktitle={BMVC},
+  volume={2},
+  pages={7},
+  year={2016}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_svtr.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_svtr.en.md
new file mode 100644
index 0000000..ef39eca
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_svtr.en.md
@@ -0,0 +1,144 @@
+---
+comments: true
+---
+
+# SVTR
+
+## 1. Introduction
+
+Paper:
+> [SVTR: Scene Text Recognition with a Single Visual Model](https://arxiv.org/abs/2205.00159)
+> Yongkun Du and Zhineng Chen and Caiyan Jia Xiaoting Yin and Tianlun Zheng and Chenxia Li and Yuning Du and Yu-Gang Jiang
+> IJCAI, 2022
+
+The accuracy (%) and model files of SVTR on the public dataset of scene text recognition are as follows:
+
+* Chinese dataset from [Chinese Benchmark](https://arxiv.org/abs/2112.15093) , and the Chinese training evaluation strategy of SVTR follows the paper.
+
+|   Model    |IC13<br/>857 |  SVT  |IIIT5k<br/>3000 |IC15<br/>1811| SVTP  |CUTE80 | Avg_6 |IC15<br/>2077 |IC13<br/>1015 |IC03<br/>867|IC03<br/>860|Avg_10 | Chinese<br/>scene_test|                                                                                            Download link                                                                                            |
+|:----------:|:------:|:-----:|:---------:|:------:|:-----:|:-----:|:-----:|:-------:|:-------:|:-----:|:-----:|:---------------------------------------------:|:-----:|:------:|
+| SVTR Tiny  | 96.85  | 91.34 |   94.53   | 83.99  | 85.43 | 89.24 | 90.87 |  80.55  |  95.37  | 95.27 | 95.70 | 90.13 | 67.90 | [English](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar)  / [Chinese](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_ch_train.tar)  |
+| SVTR Small | 95.92  | 93.04 |   95.03   | 84.70  | 87.91 | 92.01 | 91.63 |  82.72  |  94.88  | 96.08 | 96.28 | 91.02 | 69.00 | [English](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_small_none_ctc_en_train.tar) / [Chinese](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_small_none_ctc_ch_train.tar) |
+| SVTR Base  | 97.08  | 91.50 |   96.03   | 85.20  | 89.92 | 91.67 | 92.33 |  83.73  |  95.66  | 95.62 | 95.81 | 91.61 | 71.40 |                          [English](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_base_none_ctc_en_train.tar)  /                                              -                          |
+| SVTR Large | 97.20  | 91.65 |   96.30   | 86.58  | 88.37 | 95.14 | 92.82 |  84.54  |  96.35  | 96.54 | 96.74 | 92.24 | 72.10 | [English](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_large_none_ctc_en_train.tar) / [Chinese](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_large_none_ctc_ch_train.tar) |
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+### Dataset Preparation
+
+[English dataset download](https://github.com/clovaai/deep-text-recognition-benchmark#download-lmdb-dataset-for-traininig-and-evaluation-from-here)
+[Chinese dataset download](https://github.com/fudanvi/benchmarking-chinese-text-recognition#download)
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_svtrnet.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_svtrnet.yml
+```
+
+### Evaluation
+
+You can download the model files and configuration files provided by `SVTR`: [download link](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar), take `SVTR-T` as an example, using the following command to evaluate:
+
+```bash linenums="1"
+# Download the tar archive containing the model files and configuration files of SVTR-T and extract it
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar && tar xf rec_svtr_tiny_none_ctc_en_train.tar
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c ./rec_svtr_tiny_none_ctc_en_train/rec_svtr_tiny_6local_6global_stn_en.yml -o Global.pretrained_model=./rec_svtr_tiny_none_ctc_en_train/best_accuracy
+```
+
+### Prediction
+
+```bash linenums="1"
+python3 tools/infer_rec.py -c ./rec_svtr_tiny_none_ctc_en_train/rec_svtr_tiny_6local_6global_stn_en.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_svtr_tiny_none_ctc_en_train/best_accuracy
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the SVTR text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar) ), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_svtrnet.yml -o Global.pretrained_model=./rec_svtr_tiny_none_ctc_en_train/best_accuracy  Global.save_inference_dir=./inference/rec_svtr_tiny_stn_en
+```
+
+**Note:** If you are training the model on your own dataset and have modified the dictionary file, please pay attention to modify the `character_dict_path` in the configuration file to the modified dictionary file.
+
+After the conversion is successful, there are three files in the directory:
+
+```text linenums="1"
+/inference/rec_svtr_tiny_stn_en/
+    ├── inference.pdiparams
+    ├── inference.pdiparams.info
+    └── inference.pdmodel
+```
+
+For SVTR text recognition model inference, the following commands can be executed:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_svtr_tiny_stn_en/' --rec_algorithm='SVTR' --rec_image_shape='3,64,256' --rec_char_dict_path='./ppocr/utils/ic15_dict.txt'
+```
+
+![](./images/word_10.png)
+
+After executing the command, the prediction result (recognized text and score) of the image above is printed to the screen, an example is as follows:
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9999998807907104)
+```
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+* 1. Speed situation on CPU and GPU
+  * Since most of the operators used by `SVTR` are matrix multiplication, in the GPU environment, the speed has an advantage, but in the environment where mkldnn is enabled on the CPU, `SVTR` has no advantage over the optimized convolutional network.
+* 2. SVTR model convert to ONNX failed
+  * Ensure `paddle2onnx` and `onnxruntime` versions are up to date, refer to [SVTR model to onnx step-by-step example](https://github.com/PaddlePaddle/PaddleOCR/issues/7821#issuecomment-) for the convert onnx command. 1271214273).
+* 3. SVTR model convert to ONNX is successful but the inference result is incorrect
+  * The possible reason is that the model parameter `out_char_num` is not set correctly, it should be set to W//4, W//8 or W//12, please refer to [Section 3.3.3 of SVTR, a high-precision Chinese scene text recognition model](https://aistudio.baidu.com/aistudio/) projectdetail/5073182?contributionType=1).
+* 4. Optimization of long text recognition
+  * Refer to [Section 3.3 of SVTR, a high-precision Chinese scene text recognition model](https://aistudio.baidu.com/aistudio/projectdetail/5073182?contributionType=1).
+* 5. Notes on the reproduction of the paper results
+  * Dataset using provided by [ABINet](https://github.com/FangShancheng/ABINet).
+  * By default, 4 cards of GPUs are used for training, the default Batchsize of a single card is 512, and the total Batchsize is 2048, corresponding to a learning rate of 0.0005. When modifying the Batchsize or changing the number of GPU cards, the learning rate should be modified in equal proportion.
+* 6. Exploration Directions for further optimization
+  * Learning rate adjustment: adjusting to twice the default to keep Batchsize unchanged; or reducing Batchsize to 1/2 the default to keep the learning rate unchanged.
+  * Data augmentation strategies: optionally `RecConAug` and `RecAug`.
+  * If STN is not used, `Local` of `mixer` can be replaced by `Conv` and `local_mixer` can all be modified to `[5, 5]`.
+  * Grid search for optimal `embed_dim`, `depth`, `num_heads` configurations.
+  * Use the `Post-Normalization strategy`, which is to modify the model configuration `prenorm` to `True`.
+
+## Citation
+
+```bibtex
+@article{Du2022SVTR,
+  title     = {SVTR: Scene Text Recognition with a Single Visual Model},
+  author    = {Du, Yongkun and Chen, Zhineng and Jia, Caiyan and Yin, Xiaoting and Zheng, Tianlun and Li, Chenxia and Du, Yuning and Jiang, Yu-Gang},
+  booktitle = {IJCAI},
+  year      = {2022},
+  url       = {https://arxiv.org/abs/2205.00159}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_svtr.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_svtr.md
new file mode 100644
index 0000000..c52dffd
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_svtr.md
@@ -0,0 +1,170 @@
+---
+comments: true
+---
+
+# 场景文本识别算法-SVTR
+
+## 1. 算法简介
+
+论文信息：
+> [SVTR: Scene Text Recognition with a Single Visual Model](https://arxiv.org/abs/2205.00159)
+> Yongkun Du and Zhineng Chen and Caiyan Jia and Xiaoting Yin and Tianlun Zheng and Chenxia Li and Yuning Du and Yu-Gang Jiang
+> IJCAI, 2022
+
+场景文本识别旨在将自然图像中的文本转录为数字字符序列，从而传达对场景理解至关重要的高级语义。这项任务由于文本变形、字体、遮挡、杂乱背景等方面的变化具有一定的挑战性。先前的方法为提高识别精度做出了许多工作。然而文本识别器除了准确度外，还因为实际需求需要考虑推理速度等因素。
+
+### SVTR算法简介
+
+主流的场景文本识别模型通常包含两个模块：用于特征提取的视觉模型和用于文本转录的序列模型。这种架构虽然准确，但复杂且效率较低，限制了在实际场景中的应用。SVTR提出了一种用于场景文本识别的单视觉模型，该模型在patch-wise image tokenization框架内，完全摒弃了序列建模，在精度具有竞争力的前提下，模型参数量更少，速度更快，主要有以下几点贡献：
+
+1. 首次发现单视觉模型可以达到与视觉语言模型相媲美甚至更高的准确率，并且其具有效率高和适应多语言的优点，在实际应用中很有前景。
+2. SVTR从字符组件的角度出发，逐渐的合并字符组件，自下而上地完成字符的识别。
+3. SVTR引入了局部和全局Mixing，分别用于提取字符组件特征和字符间依赖关系，与多尺度的特征一起，形成多粒度特征描述。
+
+SVTR在场景文本识别公开数据集上的精度(%)和模型文件如下：
+
+* 中文数据集来自于[Chinese Benchmark](https://arxiv.org/abs/2112.15093) ，SVTR的中文训练评估策略遵循该论文。
+
+|    模型      |IC13<br/>857 |  SVT  |IIIT5k<br/>3000 |IC15<br/>1811| SVTP  |CUTE80 | Avg_6 |IC15<br/>2077 |IC13<br/>1015 |IC03<br/>867|IC03<br/>860|Avg_10 | Chinese<br/>scene_test|      下载链接       |
+|:----------:|:------:|:-----:|:---------:|:------:|:-----:|:-----:|:-----:|:-------:|:-------:|:-----:|:-----:|:------:|:-----:|:-----:|
+| SVTR Tiny  | 96.85  | 91.34 |   94.53   | 83.99  | 85.43 | 89.24 | 90.87 |  80.55  |  95.37  | 95.27 | 95.70 | 90.13 | 67.90 | [英文](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar)  / [中文](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_ch_train.tar)  |
+| SVTR Small | 95.92  | 93.04 |   95.03   | 84.70  | 87.91 | 92.01 | 91.63 |  82.72  |  94.88  | 96.08 | 96.28 | 91.02 | 69.00 | [英文](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_small_none_ctc_en_train.tar) / [中文](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_small_none_ctc_ch_train.tar) |
+| SVTR Base  | 97.08  | 91.50 |   96.03   | 85.20  | 89.92 | 91.67 | 92.33 |  83.73  |  95.66  | 95.62 | 95.81 | 91.61 | 71.40 | [英文](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_base_none_ctc_en_train.tar)  /                                              -                                             |
+| SVTR Large | 97.20  | 91.65 |   96.30   | 86.58  | 88.37 | 95.14 | 92.82 |  84.54  |  96.35  | 96.54 | 96.74 | 92.24 | 72.10 | [英文](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_large_none_ctc_en_train.tar) / [中文](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_large_none_ctc_ch_train.tar) |
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+### 3.1 模型训练
+
+#### 数据集准备
+
+[英文数据集下载](https://github.com/clovaai/deep-text-recognition-benchmark#download-lmdb-dataset-for-traininig-and-evaluation-from-here)
+[中文数据集下载](https://github.com/fudanvi/benchmarking-chinese-text-recognition#download)
+
+#### 启动训练
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练`SVTR`识别模型时需要**更换配置文件**为`SVTR`的[配置文件](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_svtrnet.yml)。
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_svtrnet.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_svtrnet.yml
+```
+
+### 3.2 评估
+
+可下载`SVTR`提供的模型文件和配置文件：[下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar) ，以`SVTR-T`为例，使用如下命令进行评估：
+
+```bash linenums="1"
+# 下载包含SVTR-T的模型文件和配置文件的tar压缩包并解压
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar && tar xf rec_svtr_tiny_none_ctc_en_train.tar
+# 注意将pretrained_model的路径设置为本地路径。
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c ./rec_svtr_tiny_none_ctc_en_train/rec_svtr_tiny_6local_6global_stn_en.yml -o Global.pretrained_model=./rec_svtr_tiny_none_ctc_en_train/best_accuracy
+```
+
+### 3.3 预测
+
+使用如下命令进行单张图片预测：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/infer_rec.py -c ./rec_svtr_tiny_none_ctc_en_train/rec_svtr_tiny_6local_6global_stn_en.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_svtr_tiny_none_ctc_en_train/best_accuracy
+# 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/imgs_words_en/'。
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将训练得到best模型，转换成inference model。下面以`SVTR-T`在英文数据集训练的模型为例（[模型和配置文件下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/rec_svtr_tiny_none_ctc_en_train.tar) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/export_model.py -c ./rec_svtr_tiny_none_ctc_en_train/rec_svtr_tiny_6local_6global_stn_en.yml -o Global.pretrained_model=./rec_svtr_tiny_none_ctc_en_train/best_accuracy Global.save_inference_dir=./inference/rec_svtr_tiny_stn_en
+```
+
+**注意：** 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否为所正确的字典文件。
+
+转换成功后，在目录下有三个文件：
+
+```text linenums="1"
+/inference/rec_svtr_tiny_stn_en/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+执行如下命令进行模型推理：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_svtr_tiny_stn_en/' --rec_algorithm='SVTR' --rec_image_shape='3,64,256' --rec_char_dict_path='./ppocr/utils/ic15_dict.txt'
+# 预测文件夹下所有图像时，可修改image_dir为文件夹，如 --image_dir='./doc/imgs_words_en/'。
+```
+
+![](./images/word_10.png)
+
+执行命令后，上面图像的预测结果（识别的文本和得分）会打印到屏幕上，示例如下：
+结果如下：
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9999998807907104)
+```
+
+**注意**：
+
+* 如果您调整了训练时的输入分辨率，需要通过参数`rec_image_shape`设置为您需要的识别图像形状。
+* 在推理时需要设置参数`rec_char_dict_path`指定字典，如果您修改了字典，请修改该参数为您的字典文件。
+* 如果您修改了预处理方法，需修改`tools/infer/predict_rec.py`中SVTR的预处理为您的预处理方法。
+
+### 4.2 C++推理部署
+
+由于C++预处理后处理还未支持SVTR，所以暂未支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+* 1. GPU和CPU速度对比
+  * 由于`SVTR`使用的算子大多为矩阵相乘，在GPU环境下，速度具有优势，但在CPU开启mkldnn加速环境下，`SVTR`相比于被优化的卷积网络没有优势。
+
+* 2. SVTR模型转ONNX失败
+  * 保证`paddle2onnx`和`onnxruntime`版本最新，转onnx命令参考[SVTR模型转onnx步骤实例](https://github.com/PaddlePaddle/PaddleOCR/issues/7821#issuecomment-1271214273)。
+* 3. SVTR转ONNX成功但是推理结果不正确
+  * 可能的原因模型参数`out_char_num`设置不正确，应设置为W//4、W//8或者W//12，可以参考[高精度中文场景文本识别模型SVTR的3.3.3章节](https://aistudio.baidu.com/aistudio/projectdetail/5073182?contributionType=1)。
+* 4. 长文本识别优化
+  * 参考[高精度中文场景文本识别模型SVTR的3.3章节](https://aistudio.baidu.com/aistudio/projectdetail/5073182?contributionType=1)。
+* 5. 论文结果复现注意事项
+  * 数据集使用[ABINet](https://github.com/FangShancheng/ABINet)提供的数据集；
+  * 默认使用4卡GPU训练，单卡Batchsize默认为512，总Batchsize为2048，对应的学习率为0.0005，当修改Batchsize或者改变GPU卡数，学习率应等比例修改。
+* 6. 进一步优化的探索点
+  * 学习率调整：可以调整为默认的两倍保持Batchsize不变；或者将Batchsize减小为默认的1/2，保持学习率不变；
+  * 数据增强策略：可选`RecConAug`和`RecAug`；
+  * 如果不使用STN时，可以将`mixer`的`Local`替换为`Conv`、`local_mixer`全部修改为`[5, 5]`；
+  * 网格搜索最优的`embed_dim`、`depth`、`num_heads`配置；
+  * 使用`后Normalization策略`，即是将模型配置`prenorm`修改为`True`。
+
+## 引用
+
+```bibtex
+@article{Du2022SVTR,
+  title     = {SVTR: Scene Text Recognition with a Single Visual Model},
+  author    = {Du, Yongkun and Chen, Zhineng and Jia, Caiyan and Yin, Xiaoting and Zheng, Tianlun and Li, Chenxia and Du, Yuning and Jiang, Yu-Gang},
+  booktitle = {IJCAI},
+  year      = {2022},
+  url       = {https://arxiv.org/abs/2205.00159}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_svtrv2.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_svtrv2.md
new file mode 100644
index 0000000..754d5e3
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_svtrv2.md
@@ -0,0 +1,129 @@
+---
+comments: true
+---
+
+# 场景文本识别算法-SVTRv2
+
+## 1. 算法简介
+
+### SVTRv2算法简介
+
+🔥 该算法由来自复旦大学视觉与学习实验室([FVL](https://fvl.fudan.edu.cn))的[OpenOCR](https://github.com/Topdu/OpenOCR)团队研发，其在[PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务](https://aistudio.baidu.com/competition/detail/1131/0/introduction)中荣获一等奖，B榜端到端识别精度相比PP-OCRv4提升2.5%，推理速度持平。主要思路：1、检测和识别模型的Backbone升级为RepSVTR；2、识别教师模型升级为SVTRv2，可识别长文本。
+
+|模型|配置文件|端到端|下载链接|
+| --- | --- | --- | --- |
+|PP-OCRv4| |A榜 62.77% <br> B榜 62.51%| [Model List](../../ppocr/model_list.md) |
+|SVTRv2(Rec Sever)|[configs/rec/SVTRv2/ch_SVTRv2_rec.yml](../../../../configs/rec/SVTRv2/ch_SVTRv2_rec.yml)|A榜 68.81% (使用PP-OCRv4检测模型)| [训练模型](https://paddleocr.bj.bcebos.com/openatom/openatom_rec_svtrv2_ch_train.tar) / [推理模型](https://paddleocr.bj.bcebos.com/openatom/openatom_rec_svtrv2_ch_infer.tar) |
+|RepSVTR(Mobile)|[识别](../../../../configs/rec/SVTRv2/ch_RepSVTR_rec.yml) <br> [识别蒸馏](../../../../configs/rec/SVTRv2/ch_SVTRv2_rec_distillation.yml) <br> [检测](../../../../configs/det/det_repsvtr_db.yml)|B榜 65.07%| 识别: [训练模型](https://paddleocr.bj.bcebos.com/openatom/openatom_rec_repsvtr_ch_train.tar) / [推理模型](https://paddleocr.bj.bcebos.com/openatom/openatom_rec_repsvtr_ch_infer.tar) <br> 识别蒸馏: [训练模型](https://paddleocr.bj.bcebos.com/openatom/openatom_rec_svtrv2_distill_ch_train.tar) / [推理模型](https://paddleocr.bj.bcebos.com/openatom/openatom_rec_svtrv2_distill_ch_infer.tar) <br> 检测: [训练模型](https://paddleocr.bj.bcebos.com/openatom/openatom_det_repsvtr_ch_train.tar) / [推理模型](https://paddleocr.bj.bcebos.com/openatom/openatom_det_repsvtr_ch_infer.tar) |
+
+🚀 快速使用：参考PP-OCR推理[说明文档](../../legacy/python_infer.md)，将检测和识别模型替换为上表中对应的RepSVTR或SVTRv2推理模型即可使用。
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+### 3.1 模型训练
+
+训练命令：
+
+```bash linenums="1"
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/SVTRv2/ch_RepSVTR_rec_gtc.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+# Rec 学生模型
+python -m paddle.distributed.launch --gpus '0,1,2,3,4,5,6,7'  tools/train.py -c configs/rec/SVTRv2/ch_RepSVTR_rec_gtc.yml
+# Rec 教师模型
+python -m paddle.distributed.launch --gpus '0,1,2,3,4,5,6,7'  tools/train.py -c configs/rec/SVTRv2/ch_SVTRv2_rec_gtc.yml
+# Rec 蒸馏训练
+python -m paddle.distributed.launch --gpus '0,1,2,3,4,5,6,7'  tools/train.py -c configs/rec/SVTRv2/ch_SVTRv2_rec_gtc_distill.yml
+```
+
+### 3.2 评估
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/SVTRv2/ch_RepSVTR_rec_gtc.yml -o Global.pretrained_model=output/ch_RepSVTR_rec_gtc/best_accuracy
+```
+
+### 3.3 预测
+
+使用如下命令进行单张图片预测：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/infer_rec.py -c tools/eval.py -c configs/rec/SVTRv2/ch_RepSVTR_rec_gtc.yml -o Global.pretrained_model=output/ch_RepSVTR_rec_gtc/best_accuracy Global.infer_img='./doc/imgs_words_en/word_10.png'
+# 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/imgs_words_en/'。
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将训练得到best模型，转换成inference model，以RepSVTR为例，可以使用如下命令进行转换：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/export_model.py -c configs/rec/SVTRv2/ch_RepSVTR_rec_gtc.yml -o Global.pretrained_model=output/ch_RepSVTR_rec_gtc/best_accuracy Global.save_inference_dir=./inference/rec_repsvtr_infer
+```
+
+**注意：** 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否为所正确的字典文件。
+
+转换成功后，在目录下有三个文件：
+
+```text linenums="1"
+./inference/rec_repsvtr_infer/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+执行如下命令进行模型推理：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_repsvtr_infer/'
+# 预测文件夹下所有图像时，可修改image_dir为文件夹，如 --image_dir='./doc/imgs_words_en/'。
+```
+
+![](./images/word_10.png)
+
+执行命令后，上面图像的预测结果（识别的文本和得分）会打印到屏幕上，示例如下：
+结果如下：
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9999998807907104)
+```
+
+**注意**：
+
+- 如果您调整了训练时的输入分辨率，需要通过参数`rec_image_shape`设置为您需要的识别图像形状。
+- 在推理时需要设置参数`rec_char_dict_path`指定字典，如果您修改了字典，请修改该参数为您的字典文件。
+- 如果您修改了预处理方法，需修改`tools/infer/predict_rec.py`中SVTR的预处理为您的预处理方法。
+
+### 4.2 C++推理部署
+
+准备好推理模型后，参考[cpp infer](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/deploy/cpp_infer)教程进行操作即可。
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+- Paddle2ONNX推理：准备好推理模型后，参考[paddle2onnx](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/deploy/paddle2onnx)教程操作。
+
+## 5. FAQ
+
+## 引用
+
+```bibtex
+@article{Du2022SVTR,
+  title     = {SVTR: Scene Text Recognition with a Single Visual Model},
+  author    = {Du, Yongkun and Chen, Zhineng and Jia, Caiyan and Yin, Xiaoting and Zheng, Tianlun and Li, Chenxia and Du, Yuning and Jiang, Yu-Gang},
+  booktitle = {IJCAI},
+  year      = {2022},
+  url       = {https://arxiv.org/abs/2205.00159}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_visionlan.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_visionlan.en.md
new file mode 100644
index 0000000..cf4e2ed
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_visionlan.en.md
@@ -0,0 +1,120 @@
+---
+comments: true
+---
+
+# VisionLAN
+
+## 1. Introduction
+
+Paper:
+> [From Two to One: A New Scene Text Recognizer with Visual Language Modeling Network](https://arxiv.org/abs/2108.09661)
+> Yuxin Wang, Hongtao Xie, Shancheng Fang, Jing Wang, Shenggao Zhu, Yongdong Zhang
+> ICCV, 2021
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Model|Backbone|config|Acc|Download link|
+| --- | --- | --- | --- | --- |
+|VisionLAN|ResNet45|[rec_r45_visionlan.yml](../../../../configs/rec/rec_r45_visionlan.yml)|90.30%|[model link](https://paddleocr.bj.bcebos.com/VisionLAN/rec_r45_visionlan_train.tar)|
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_r45_visionlan.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r45_visionlan.yml
+```
+
+### Evaluation
+
+```bash linenums="1"
+# GPU evaluation
+python3 tools/eval.py -c configs/rec/rec_r45_visionlan.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### Prediction
+
+```bash linenums="1"
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_r45_visionlan.yml -o Global.infer_img='./doc/imgs_words/en/word_2.png' Global.pretrained_model=./rec_r45_visionlan_train/best_accuracy
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the VisionLAN text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/VisionLAN/rec_r45_visionlan_train.tar)) ), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r45_visionlan.yml -o Global.pretrained_model=./rec_r45_visionlan_train/best_accuracy Global.save_inference_dir=./inference/rec_r45_visionlan/
+```
+
+**Note:**
+
+- If you are training the model on your own dataset and have modified the dictionary file, please pay attention to modify the `character_dict_path` in the configuration file to the modified dictionary file.
+- If you modified the input size during training, please modify the `infer_shape` corresponding to VisionLAN in the `tools/export_model.py` file.
+
+After the conversion is successful, there are three files in the directory:
+
+```text linenums="1"
+./inference/rec_r45_visionlan/
+    ├── inference.pdiparams
+    ├── inference.pdiparams.info
+    └── inference.pdmodel
+```
+
+For VisionLAN text recognition model inference, the following commands can be executed:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words/en/word_2.png' --rec_model_dir='./inference/rec_r45_visionlan/' --rec_algorithm='VisionLAN' --rec_image_shape='3,64,256' --rec_char_dict_path='./ppocr/utils/ic15_dict.txt' --use_space_char=False
+```
+
+![img](./images/word_10.png)
+
+After executing the command, the prediction result (recognized text and score) of the image above is printed to the screen, an example is as follows:
+The result is as follows:
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words/en/word_2.png:('yourself', 0.9999493)
+```
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+1. Note that the MJSynth and SynthText datasets come from [VisionLAN repo](https://github.com/wangyuxin87/VisionLAN).
+2. We use the pre-trained model provided by the VisionLAN authors for finetune training. The dictionary for the pre-trained model is 'ppocr/utils/ic15_dict.txt'.
+
+## Citation
+
+```bibtex
+@inproceedings{wang2021two,
+  title={From Two to One: A New Scene Text Recognizer with Visual Language Modeling Network},
+  author={Wang, Yuxin and Xie, Hongtao and Fang, Shancheng and Wang, Jing and Zhu, Shenggao and Zhang, Yongdong},
+  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
+  pages={14194--14203},
+  year={2021}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_visionlan.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_visionlan.md
new file mode 100644
index 0000000..be58cd6
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_visionlan.md
@@ -0,0 +1,135 @@
+---
+comments: true
+---
+
+# 场景文本识别算法-VisionLAN
+
+## 1. 算法简介
+
+论文信息：
+> [From Two to One: A New Scene Text Recognizer with Visual Language Modeling Network](https://arxiv.org/abs/2108.09661)
+> Yuxin Wang, Hongtao Xie, Shancheng Fang, Jing Wang, Shenggao Zhu, Yongdong Zhang
+> ICCV, 2021
+
+`VisionLAN`使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC13, IC15, SVTP, CUTE数据集上进行评估，算法复现效果如下：
+
+|模型|骨干网络|配置文件|Acc|下载链接|
+| --- | --- | --- | --- | --- |
+|VisionLAN|ResNet45|[rec_r45_visionlan.yml](../../../../configs/rec/rec_r45_visionlan.yml)|90.30%|[预训练、训练模型](https://paddleocr.bj.bcebos.com/VisionLAN/rec_r45_visionlan_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+### 3.1 模型训练
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练`VisionLAN`识别模型时需要**更换配置文件**为`VisionLAN`的[配置文件](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_r45_visionlan.yml)。
+
+#### 启动训练
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_r45_visionlan.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_r45_visionlan.yml
+```
+
+### 3.2 评估
+
+可下载已训练完成的模型文件，使用如下命令进行评估：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/eval.py -c configs/rec/rec_r45_visionlan.yml -o Global.pretrained_model=./rec_r45_visionlan_train/best_accuracy
+```
+
+### 3.3 预测
+
+使用如下命令进行单张图片预测：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/infer_rec.py -c configs/rec/rec_r45_visionlan.yml -o Global.infer_img='./doc/imgs_words/en/word_2.png' Global.pretrained_model=./rec_r45_visionlan_train/best_accuracy
+# 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/imgs_words_en/'。
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将训练得到best模型，转换成inference model。这里以训练完成的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/VisionLAN/rec_r45_visionlan_train.tar))，可以使用如下命令进行转换：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/export_model.py -c configs/rec/rec_r45_visionlan.yml -o Global.pretrained_model=./rec_r45_visionlan_train/best_accuracy Global.save_inference_dir=./inference/rec_r45_visionlan/
+```
+
+**注意：**
+
+- 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否是所需要的字典文件。
+- 如果您修改了训练时的输入大小，请修改`tools/export_model.py`文件中的对应VisionLAN的`infer_shape`。
+
+转换成功后，在目录下有三个文件：
+
+```text linenums="1"
+./inference/rec_r45_visionlan/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+执行如下命令进行模型推理：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words/en/word_2.png' --rec_model_dir='./inference/rec_r45_visionlan/' --rec_algorithm='VisionLAN' --rec_image_shape='3,64,256' --rec_char_dict_path='./ppocr/utils/ic15_dict.txt' --use_space_char=False
+# 预测文件夹下所有图像时，可修改image_dir为文件夹，如 --image_dir='./doc/imgs_words_en/'。
+```
+
+![img](./images/word_10.png)
+
+执行命令后，上面图像的预测结果（识别的文本和得分）会打印到屏幕上，示例如下：
+结果如下：
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words/en/word_2.png:('yourself', 0.9999493)
+```
+
+**注意**：
+
+- 训练上述模型采用的图像分辨率是[3，64，256]，需要通过参数`rec_image_shape`设置为您训练时的识别图像形状。
+- 在推理时需要设置参数`rec_char_dict_path`指定字典，如果您修改了字典，请修改该参数为您的字典文件。
+- 如果您修改了预处理方法，需修改`tools/infer/predict_rec.py`中VisionLAN的预处理为您的预处理方法。
+
+### 4.2 C++推理部署
+
+由于C++预处理后处理还未支持VisionLAN，所以暂未支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+1. MJSynth和SynthText两种数据集来自于[VisionLAN源repo](https://github.com/wangyuxin87/VisionLAN) 。
+2. 我们使用VisionLAN作者提供的预训练模型进行finetune训练，预训练模型配套字典为'ppocr/utils/ic15_dict.txt'。
+
+## 引用
+
+```bibtex
+@inproceedings{wang2021two,
+  title={From Two to One: A New Scene Text Recognizer with Visual Language Modeling Network},
+  author={Wang, Yuxin and Xie, Hongtao and Fang, Shancheng and Wang, Jing and Zhu, Shenggao and Zhang, Yongdong},
+  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
+  pages={14194--14203},
+  year={2021}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_vitstr.en.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_vitstr.en.md
new file mode 100644
index 0000000..18fefc6
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_vitstr.en.md
@@ -0,0 +1,119 @@
+---
+comments: true
+---
+
+# ViTSTR
+
+## 1. Introduction
+
+Paper:
+> [Vision Transformer for Fast and Efficient Scene Text Recognition](https://arxiv.org/abs/2105.08582)
+> Rowel Atienza
+> ICDAR, 2021
+
+Using MJSynth and SynthText two text recognition datasets for training, and evaluating on IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE datasets, the algorithm reproduction effect is as follows:
+
+|Model|Backbone|config|Acc|Download link|
+| --- | --- | --- | --- | --- |
+|ViTSTR|ViTSTR|[rec_vitstr_none_ce.yml](../../../../configs/rec/rec_vitstr_none_ce.yml)|79.82%|[trained model](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar)|
+
+## 2. Environment
+
+Please refer to ["Environment Preparation"](../../ppocr/environment.en.md) to configure the PaddleOCR environment, and refer to ["Project Clone"](../../ppocr/blog/clone.en.md)to clone the project code.
+
+## 3. Model Training / Evaluation / Prediction
+
+Please refer to [Text Recognition Tutorial](../../ppocr/model_train/recognition.en.md). PaddleOCR modularizes the code, and training different recognition models only requires **changing the configuration file**.
+
+### Training
+
+Specifically, after the data preparation is completed, the training can be started. The training command is as follows:
+
+```bash linenums="1"
+# Single GPU training (long training period, not recommended)
+python3 tools/train.py -c configs/rec/rec_vitstr_none_ce.yml
+
+# Multi GPU training, specify the gpu number through the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_vitstr_none_ce.yml
+```
+
+### Evaluation
+
+```bash linenums="1"
+# GPU evaluation
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.pretrained_model={path/to/weights}/best_accuracy
+```
+
+### Prediction
+
+```bash linenums="1"
+# The configuration file used for prediction must match the training
+python3 tools/infer_rec.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy
+```
+
+## 4. Inference and Deployment
+
+### 4.1 Python Inference
+
+First, the model saved during the ViTSTR text recognition training process is converted into an inference model. ( [Model download link](https://paddleocr.bj.bcebos.com/rec_vitstr_none_none_train.tar)) ), you can use the following command to convert:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy  Global.save_inference_dir=./inference/rec_vitstr
+```
+
+**Note:**
+
+- If you are training the model on your own dataset and have modified the dictionary file, please pay attention to modify the `character_dict_path` in the configuration file to the modified dictionary file.
+- If you modified the input size during training, please modify the `infer_shape` corresponding to ViTSTR in the `tools/export_model.py` file.
+
+After the conversion is successful, there are three files in the directory:
+
+```text linenums="1"
+/inference/rec_vitstr/
+    ├── inference.pdiparams
+    ├── inference.pdiparams.info
+    └── inference.pdmodel
+```
+
+For ViTSTR text recognition model inference, the following commands can be executed:
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_vitstr/' --rec_algorithm='ViTSTR' --rec_image_shape='1,224,224' --rec_char_dict_path='./ppocr/utils/EN_symbol_dict.txt'
+```
+
+![img](./images/word_10.png)
+
+After executing the command, the prediction result (recognized text and score) of the image above is printed to the screen, an example is as follows:
+The result is as follows:
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9998350143432617)
+```
+
+### 4.2 C++ Inference
+
+Not supported
+
+### 4.3 Serving
+
+Not supported
+
+### 4.4 More
+
+Not supported
+
+## 5. FAQ
+
+1. In the `ViTSTR` paper, using pre-trained weights on ImageNet1k for initial training, we did not use pre-trained weights in training, and the final accuracy did not change or even improved.
+
+## Citation
+
+```bibtex
+@article{Atienza2021ViTSTR,
+  title     = {Vision Transformer for Fast and Efficient Scene Text Recognition},
+  author    = {Rowel Atienza},
+  booktitle = {ICDAR},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2105.08582}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/algorithm_rec_vitstr.md b/docs/version2.x/algorithm/text_recognition/algorithm_rec_vitstr.md
new file mode 100644
index 0000000..4880623
--- /dev/null
+++ b/docs/version2.x/algorithm/text_recognition/algorithm_rec_vitstr.md
@@ -0,0 +1,135 @@
+---
+comments: true
+---
+
+# 场景文本识别算法-ViTSTR
+
+## 1. 算法简介
+
+论文信息：
+> [Vision Transformer for Fast and Efficient Scene Text Recognition](https://arxiv.org/abs/2105.08582)
+> Rowel Atienza
+> ICDAR, 2021
+
+`ViTSTR`使用MJSynth和SynthText两个文字识别数据集训练，在IIIT, SVT, IC03, IC13, IC15, SVTP, CUTE数据集上进行评估，算法复现效果如下：
+
+|模型|骨干网络|配置文件|Acc|下载链接|
+| --- | --- | --- | --- | --- |
+|ViTSTR|ViTSTR|[rec_vitstr_none_ce.yml](../../../../configs/rec/rec_vitstr_none_ce.yml)|79.82%|[训练模型](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar)|
+
+## 2. 环境配置
+
+请先参考[《运行环境准备》](../../ppocr/environment.md)配置PaddleOCR运行环境，参考[《项目克隆》](../../ppocr/blog/clone.md)克隆项目代码。
+
+## 3. 模型训练、评估、预测
+
+### 3.1 模型训练
+
+请参考[文本识别训练教程](../../ppocr/model_train/recognition.md)。PaddleOCR对代码进行了模块化，训练`ViTSTR`识别模型时需要**更换配置文件**为`ViTSTR`的[配置文件](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/rec_vitstr_none_ce.yml)。
+
+#### 启动训练
+
+具体地，在完成数据准备后，便可以启动训练，训练命令如下：
+
+```bash linenums="1"
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/rec_vitstr_none_ce.yml
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_vitstr_none_ce.yml
+```
+
+### 3.2 评估
+
+可下载已训练完成的模型文件，使用如下命令进行评估：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy
+```
+
+### 3.3 预测
+
+使用如下命令进行单张图片预测：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/infer_rec.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.infer_img='./doc/imgs_words_en/word_10.png' Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy
+# 预测文件夹下所有图像时，可修改infer_img为文件夹，如 Global.infer_img='./doc/imgs_words_en/'。
+```
+
+## 4. 推理部署
+
+### 4.1 Python推理
+
+首先将训练得到best模型，转换成inference model。这里以训练完成的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/rec_vitstr_none_ce_train.tar) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+# 注意将pretrained_model的路径设置为本地路径。
+python3 tools/export_model.py -c configs/rec/rec_vitstr_none_ce.yml -o Global.pretrained_model=./rec_vitstr_none_ce_train/best_accuracy Global.save_inference_dir=./inference/rec_vitstr/
+```
+
+**注意：**
+
+- 如果您是在自己的数据集上训练的模型，并且调整了字典文件，请注意修改配置文件中的`character_dict_path`是否是所需要的字典文件。
+- 如果您修改了训练时的输入大小，请修改`tools/export_model.py`文件中的对应ViTSTR的`infer_shape`。
+
+转换成功后，在目录下有三个文件：
+
+```text linenums="1"
+/inference/rec_vitstr/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+执行如下命令进行模型推理：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir='./doc/imgs_words_en/word_10.png' --rec_model_dir='./inference/rec_vitstr/' --rec_algorithm='ViTSTR' --rec_image_shape='1,224,224' --rec_char_dict_path='./ppocr/utils/EN_symbol_dict.txt'
+# 预测文件夹下所有图像时，可修改image_dir为文件夹，如 --image_dir='./doc/imgs_words_en/'。
+```
+
+![img](./images/word_10.png)
+
+执行命令后，上面图像的预测结果（识别的文本和得分）会打印到屏幕上，示例如下：
+结果如下：
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words_en/word_10.png:('pain', 0.9998350143432617)
+```
+
+**注意**：
+
+- 训练上述模型采用的图像分辨率是[1，224，224]，需要通过参数`rec_image_shape`设置为您训练时的识别图像形状。
+- 在推理时需要设置参数`rec_char_dict_path`指定字典，如果您修改了字典，请修改该参数为您的字典文件。
+- 如果您修改了预处理方法，需修改`tools/infer/predict_rec.py`中ViTSTR的预处理为您的预处理方法。
+
+### 4.2 C++推理部署
+
+由于C++预处理后处理还未支持ViTSTR，所以暂未支持
+
+### 4.3 Serving服务化部署
+
+暂不支持
+
+### 4.4 更多推理部署
+
+暂不支持
+
+## 5. FAQ
+
+1. 在`ViTSTR`论文中，使用在ImageNet1k上的预训练权重进行初始化训练，我们在训练未采用预训练权重，最终精度没有变化甚至有所提高。
+2. 我们仅仅复现了`ViTSTR`中的tiny版本，如果需要使用small、base版本，可将[ViTSTR源repo](https://github.com/roatienza/deep-text-recognition-benchmark) 中的预训练权重转为Paddle权重使用。
+
+## 引用
+
+```bibtex
+@article{Atienza2021ViTSTR,
+  title     = {Vision Transformer for Fast and Efficient Scene Text Recognition},
+  author    = {Rowel Atienza},
+  booktitle = {ICDAR},
+  year      = {2021},
+  url       = {https://arxiv.org/abs/2105.08582}
+}
+```
diff --git a/docs/version2.x/algorithm/text_recognition/images/word_1-20240704183926496.png b/docs/version2.x/algorithm/text_recognition/images/word_1-20240704183926496.png
new file mode 100644
index 0000000..7b915fd
Binary files /dev/null and b/docs/version2.x/algorithm/text_recognition/images/word_1-20240704183926496.png differ
diff --git a/docs/version2.x/algorithm/text_recognition/images/word_1-20240704184113913.png b/docs/version2.x/algorithm/text_recognition/images/word_1-20240704184113913.png
new file mode 100644
index 0000000..7b915fd
Binary files /dev/null and b/docs/version2.x/algorithm/text_recognition/images/word_1-20240704184113913.png differ
diff --git a/docs/version2.x/algorithm/text_recognition/images/word_10.png b/docs/version2.x/algorithm/text_recognition/images/word_10.png
new file mode 100644
index 0000000..07370f7
Binary files /dev/null and b/docs/version2.x/algorithm/text_recognition/images/word_10.png differ
diff --git a/docs/version2.x/algorithm/text_recognition/images/word_336-20240705082445918.png b/docs/version2.x/algorithm/text_recognition/images/word_336-20240705082445918.png
new file mode 100644
index 0000000..3bddd29
Binary files /dev/null and b/docs/version2.x/algorithm/text_recognition/images/word_336-20240705082445918.png differ
diff --git a/docs/version2.x/algorithm/text_recognition/images/word_336.png b/docs/version2.x/algorithm/text_recognition/images/word_336.png
new file mode 100644
index 0000000..3bddd29
Binary files /dev/null and b/docs/version2.x/algorithm/text_recognition/images/word_336.png differ
diff --git a/docs/version2.x/javascripts/katex.min.js b/docs/version2.x/javascripts/katex.min.js
new file mode 100644
index 0000000..fee59b9
--- /dev/null
+++ b/docs/version2.x/javascripts/katex.min.js
@@ -0,0 +1,12 @@
+document$.subscribe(({ body }) => {
+    renderMathInElement(body, {
+        delimiters: [
+            { left: "$$", right: "$$", display: true },
+            { left: "$", right: "$", display: false },
+            { left: "\\(", right: "\\)", display: false },
+            { left: "\\[", right: "\\]", display: true }
+        ],
+    })
+})
+
+!function (e, t) { "object" == typeof exports && "object" == typeof module ? module.exports = t() : "function" == typeof define && define.amd ? define([], t) : "object" == typeof exports ? exports.katex = t() : e.katex = t() }("undefined" != typeof self ? self : this, (function () { return function () { "use strict"; var e = { d: function (t, r) { for (var n in r) e.o(r, n) && !e.o(t, n) && Object.defineProperty(t, n, { enumerable: !0, get: r[n] }) }, o: function (e, t) { return Object.prototype.hasOwnProperty.call(e, t) } }, t = {}; e.d(t, { default: function () { return Yn } }); class r { constructor(e, t) { this.name = void 0, this.position = void 0, this.length = void 0, this.rawMessage = void 0; let n, o, s = "KaTeX parse error: " + e; const i = t && t.loc; if (i && i.start <= i.end) { const e = i.lexer.input; n = i.start, o = i.end, n === e.length ? s += " at end of input: " : s += " at position " + (n + 1) + ": "; const t = e.slice(n, o).replace(/[^]/g, "$&\u0332"); let r, a; r = n > 15 ? "\u2026" + e.slice(n - 15, n) : e.slice(0, n), a = o + 15 < e.length ? e.slice(o, o + 15) + "\u2026" : e.slice(o), s += r + t + a } const a = new Error(s); return a.name = "ParseError", a.__proto__ = r.prototype, a.position = n, null != n && null != o && (a.length = o - n), a.rawMessage = e, a } } r.prototype.__proto__ = Error.prototype; var n = r; const o = /([A-Z])/g, s = { "&": "&amp;", ">": "&gt;", "<": "&lt;", '"': "&quot;", "'": "&#x27;" }, i = /[&><"']/g; const a = function (e) { return "ordgroup" === e.type || "color" === e.type ? 1 === e.body.length ? a(e.body[0]) : e : "font" === e.type ? a(e.body) : e }; var l = { contains: function (e, t) { return -1 !== e.indexOf(t) }, deflt: function (e, t) { return void 0 === e ? t : e }, escape: function (e) { return String(e).replace(i, (e => s[e])) }, hyphenate: function (e) { return e.replace(o, "-$1").toLowerCase() }, getBaseElem: a, isCharacterBox: function (e) { const t = a(e); return "mathord" === t.type || "textord" === t.type || "atom" === t.type }, protocolFromUrl: function (e) { const t = /^[\x00-\x20]*([^\\/#?]*?)(:|&#0*58|&#x0*3a|&colon)/i.exec(e); return t ? ":" !== t[2] ? null : /^[a-zA-Z][a-zA-Z0-9+\-.]*$/.test(t[1]) ? t[1].toLowerCase() : null : "_relative" } }; const h = { displayMode: { type: "boolean", description: "Render math in display mode, which puts the math in display style (so \\int and \\sum are large, for example), and centers the math on the page on its own line.", cli: "-d, --display-mode" }, output: { type: { enum: ["htmlAndMathml", "html", "mathml"] }, description: "Determines the markup language of the output.", cli: "-F, --format <type>" }, leqno: { type: "boolean", description: "Render display math in leqno style (left-justified tags)." }, fleqn: { type: "boolean", description: "Render display math flush left." }, throwOnError: { type: "boolean", default: !0, cli: "-t, --no-throw-on-error", cliDescription: "Render errors (in the color given by --error-color) instead of throwing a ParseError exception when encountering an error." }, errorColor: { type: "string", default: "#cc0000", cli: "-c, --error-color <color>", cliDescription: "A color string given in the format 'rgb' or 'rrggbb' (no #). This option determines the color of errors rendered by the -t option.", cliProcessor: e => "#" + e }, macros: { type: "object", cli: "-m, --macro <def>", cliDescription: "Define custom macro of the form '\\foo:expansion' (use multiple -m arguments for multiple macros).", cliDefault: [], cliProcessor: (e, t) => (t.push(e), t) }, minRuleThickness: { type: "number", description: "Specifies a minimum thickness, in ems, for fraction lines, `\\sqrt` top lines, `{array}` vertical lines, `\\hline`, `\\hdashline`, `\\underline`, `\\overline`, and the borders of `\\fbox`, `\\boxed`, and `\\fcolorbox`.", processor: e => Math.max(0, e), cli: "--min-rule-thickness <size>", cliProcessor: parseFloat }, colorIsTextColor: { type: "boolean", description: "Makes \\color behave like LaTeX's 2-argument \\textcolor, instead of LaTeX's one-argument \\color mode change.", cli: "-b, --color-is-text-color" }, strict: { type: [{ enum: ["warn", "ignore", "error"] }, "boolean", "function"], description: "Turn on strict / LaTeX faithfulness mode, which throws an error if the input uses features that are not supported by LaTeX.", cli: "-S, --strict", cliDefault: !1 }, trust: { type: ["boolean", "function"], description: "Trust the input, enabling all HTML features such as \\url.", cli: "-T, --trust" }, maxSize: { type: "number", default: 1 / 0, description: "If non-zero, all user-specified sizes, e.g. in \\rule{500em}{500em}, will be capped to maxSize ems. Otherwise, elements and spaces can be arbitrarily large", processor: e => Math.max(0, e), cli: "-s, --max-size <n>", cliProcessor: parseInt }, maxExpand: { type: "number", default: 1e3, description: "Limit the number of macro expansions to the specified number, to prevent e.g. infinite macro loops. If set to Infinity, the macro expander will try to fully expand as in LaTeX.", processor: e => Math.max(0, e), cli: "-e, --max-expand <n>", cliProcessor: e => "Infinity" === e ? 1 / 0 : parseInt(e) }, globalGroup: { type: "boolean", cli: !1 } }; function c(e) { if (e.default) return e.default; const t = e.type, r = Array.isArray(t) ? t[0] : t; if ("string" != typeof r) return r.enum[0]; switch (r) { case "boolean": return !1; case "string": return ""; case "number": return 0; case "object": return {} } } class m { constructor(e) { this.displayMode = void 0, this.output = void 0, this.leqno = void 0, this.fleqn = void 0, this.throwOnError = void 0, this.errorColor = void 0, this.macros = void 0, this.minRuleThickness = void 0, this.colorIsTextColor = void 0, this.strict = void 0, this.trust = void 0, this.maxSize = void 0, this.maxExpand = void 0, this.globalGroup = void 0, e = e || {}; for (const t in h) if (h.hasOwnProperty(t)) { const r = h[t]; this[t] = void 0 !== e[t] ? r.processor ? r.processor(e[t]) : e[t] : c(r) } } reportNonstrict(e, t, r) { let o = this.strict; if ("function" == typeof o && (o = o(e, t, r)), o && "ignore" !== o) { if (!0 === o || "error" === o) throw new n("LaTeX-incompatible input and strict mode is set to 'error': " + t + " [" + e + "]", r); "warn" === o ? "undefined" != typeof console && console.warn("LaTeX-incompatible input and strict mode is set to 'warn': " + t + " [" + e + "]") : "undefined" != typeof console && console.warn("LaTeX-incompatible input and strict mode is set to unrecognized '" + o + "': " + t + " [" + e + "]") } } useStrictBehavior(e, t, r) { let n = this.strict; if ("function" == typeof n) try { n = n(e, t, r) } catch (e) { n = "error" } return !(!n || "ignore" === n) && (!0 === n || "error" === n || ("warn" === n ? ("undefined" != typeof console && console.warn("LaTeX-incompatible input and strict mode is set to 'warn': " + t + " [" + e + "]"), !1) : ("undefined" != typeof console && console.warn("LaTeX-incompatible input and strict mode is set to unrecognized '" + n + "': " + t + " [" + e + "]"), !1))) } isTrusted(e) { if (e.url && !e.protocol) { const t = l.protocolFromUrl(e.url); if (null == t) return !1; e.protocol = t } const t = "function" == typeof this.trust ? this.trust(e) : this.trust; return Boolean(t) } } class p { constructor(e, t, r) { this.id = void 0, this.size = void 0, this.cramped = void 0, this.id = e, this.size = t, this.cramped = r } sup() { return u[d[this.id]] } sub() { return u[g[this.id]] } fracNum() { return u[f[this.id]] } fracDen() { return u[b[this.id]] } cramp() { return u[y[this.id]] } text() { return u[x[this.id]] } isTight() { return this.size >= 2 } } const u = [new p(0, 0, !1), new p(1, 0, !0), new p(2, 1, !1), new p(3, 1, !0), new p(4, 2, !1), new p(5, 2, !0), new p(6, 3, !1), new p(7, 3, !0)], d = [4, 5, 4, 5, 6, 7, 6, 7], g = [5, 5, 5, 5, 7, 7, 7, 7], f = [2, 3, 4, 5, 6, 7, 6, 7], b = [3, 3, 5, 5, 7, 7, 7, 7], y = [1, 1, 3, 3, 5, 5, 7, 7], x = [0, 1, 2, 3, 2, 3, 2, 3]; var w = { DISPLAY: u[0], TEXT: u[2], SCRIPT: u[4], SCRIPTSCRIPT: u[6] }; const v = [{ name: "latin", blocks: [[256, 591], [768, 879]] }, { name: "cyrillic", blocks: [[1024, 1279]] }, { name: "armenian", blocks: [[1328, 1423]] }, { name: "brahmic", blocks: [[2304, 4255]] }, { name: "georgian", blocks: [[4256, 4351]] }, { name: "cjk", blocks: [[12288, 12543], [19968, 40879], [65280, 65376]] }, { name: "hangul", blocks: [[44032, 55215]] }]; const k = []; function S(e) { for (let t = 0; t < k.length; t += 2)if (e >= k[t] && e <= k[t + 1]) return !0; return !1 } v.forEach((e => e.blocks.forEach((e => k.push(...e))))); const M = 80, z = { doubleleftarrow: "M262 157\nl10-10c34-36 62.7-77 86-123 3.3-8 5-13.3 5-16 0-5.3-6.7-8-20-8-7.3\n 0-12.2.5-14.5 1.5-2.3 1-4.8 4.5-7.5 10.5-49.3 97.3-121.7 169.3-217 216-28\n 14-57.3 25-88 33-6.7 2-11 3.8-13 5.5-2 1.7-3 4.2-3 7.5s1 5.8 3 7.5\nc2 1.7 6.3 3.5 13 5.5 68 17.3 128.2 47.8 180.5 91.5 52.3 43.7 93.8 96.2 124.5\n 157.5 9.3 8 15.3 12.3 18 13h6c12-.7 18-4 18-10 0-2-1.7-7-5-15-23.3-46-52-87\n-86-123l-10-10h399738v-40H218c328 0 0 0 0 0l-10-8c-26.7-20-65.7-43-117-69 2.7\n-2 6-3.7 10-5 36.7-16 72.3-37.3 107-64l10-8h399782v-40z\nm8 0v40h399730v-40zm0 194v40h399730v-40z", doublerightarrow: "M399738 392l\n-10 10c-34 36-62.7 77-86 123-3.3 8-5 13.3-5 16 0 5.3 6.7 8 20 8 7.3 0 12.2-.5\n 14.5-1.5 2.3-1 4.8-4.5 7.5-10.5 49.3-97.3 121.7-169.3 217-216 28-14 57.3-25 88\n-33 6.7-2 11-3.8 13-5.5 2-1.7 3-4.2 3-7.5s-1-5.8-3-7.5c-2-1.7-6.3-3.5-13-5.5-68\n-17.3-128.2-47.8-180.5-91.5-52.3-43.7-93.8-96.2-124.5-157.5-9.3-8-15.3-12.3-18\n-13h-6c-12 .7-18 4-18 10 0 2 1.7 7 5 15 23.3 46 52 87 86 123l10 10H0v40h399782\nc-328 0 0 0 0 0l10 8c26.7 20 65.7 43 117 69-2.7 2-6 3.7-10 5-36.7 16-72.3 37.3\n-107 64l-10 8H0v40zM0 157v40h399730v-40zm0 194v40h399730v-40z", leftarrow: "M400000 241H110l3-3c68.7-52.7 113.7-120\n 135-202 4-14.7 6-23 6-25 0-7.3-7-11-21-11-8 0-13.2.8-15.5 2.5-2.3 1.7-4.2 5.8\n-5.5 12.5-1.3 4.7-2.7 10.3-4 17-12 48.7-34.8 92-68.5 130S65.3 228.3 18 247\nc-10 4-16 7.7-18 11 0 8.7 6 14.3 18 17 47.3 18.7 87.8 47 121.5 85S196 441.3 208\n 490c.7 2 1.3 5 2 9s1.2 6.7 1.5 8c.3 1.3 1 3.3 2 6s2.2 4.5 3.5 5.5c1.3 1 3.3\n 1.8 6 2.5s6 1 10 1c14 0 21-3.7 21-11 0-2-2-10.3-6-25-20-79.3-65-146.7-135-202\n l-3-3h399890zM100 241v40h399900v-40z", leftbrace: "M6 548l-6-6v-35l6-11c56-104 135.3-181.3 238-232 57.3-28.7 117\n-45 179-50h399577v120H403c-43.3 7-81 15-113 26-100.7 33-179.7 91-237 174-2.7\n 5-6 9-10 13-.7 1-7.3 1-20 1H6z", leftbraceunder: "M0 6l6-6h17c12.688 0 19.313.3 20 1 4 4 7.313 8.3 10 13\n 35.313 51.3 80.813 93.8 136.5 127.5 55.688 33.7 117.188 55.8 184.5 66.5.688\n 0 2 .3 4 1 18.688 2.7 76 4.3 172 5h399450v120H429l-6-1c-124.688-8-235-61.7\n-331-161C60.687 138.7 32.312 99.3 7 54L0 41V6z", leftgroup: "M400000 80\nH435C64 80 168.3 229.4 21 260c-5.9 1.2-18 0-18 0-2 0-3-1-3-3v-38C76 61 257 0\n 435 0h399565z", leftgroupunder: "M400000 262\nH435C64 262 168.3 112.6 21 82c-5.9-1.2-18 0-18 0-2 0-3 1-3 3v38c76 158 257 219\n 435 219h399565z", leftharpoon: "M0 267c.7 5.3 3 10 7 14h399993v-40H93c3.3\n-3.3 10.2-9.5 20.5-18.5s17.8-15.8 22.5-20.5c50.7-52 88-110.3 112-175 4-11.3 5\n-18.3 3-21-1.3-4-7.3-6-18-6-8 0-13 .7-15 2s-4.7 6.7-8 16c-42 98.7-107.3 174.7\n-196 228-6.7 4.7-10.7 8-12 10-1.3 2-2 5.7-2 11zm100-26v40h399900v-40z", leftharpoonplus: "M0 267c.7 5.3 3 10 7 14h399993v-40H93c3.3-3.3 10.2-9.5\n 20.5-18.5s17.8-15.8 22.5-20.5c50.7-52 88-110.3 112-175 4-11.3 5-18.3 3-21-1.3\n-4-7.3-6-18-6-8 0-13 .7-15 2s-4.7 6.7-8 16c-42 98.7-107.3 174.7-196 228-6.7 4.7\n-10.7 8-12 10-1.3 2-2 5.7-2 11zm100-26v40h399900v-40zM0 435v40h400000v-40z\nm0 0v40h400000v-40z", leftharpoondown: "M7 241c-4 4-6.333 8.667-7 14 0 5.333.667 9 2 11s5.333\n 5.333 12 10c90.667 54 156 130 196 228 3.333 10.667 6.333 16.333 9 17 2 .667 5\n 1 9 1h5c10.667 0 16.667-2 18-6 2-2.667 1-9.667-3-21-32-87.333-82.667-157.667\n-152-211l-3-3h399907v-40zM93 281 H400000 v-40L7 241z", leftharpoondownplus: "M7 435c-4 4-6.3 8.7-7 14 0 5.3.7 9 2 11s5.3 5.3 12\n 10c90.7 54 156 130 196 228 3.3 10.7 6.3 16.3 9 17 2 .7 5 1 9 1h5c10.7 0 16.7\n-2 18-6 2-2.7 1-9.7-3-21-32-87.3-82.7-157.7-152-211l-3-3h399907v-40H7zm93 0\nv40h399900v-40zM0 241v40h399900v-40zm0 0v40h399900v-40z", lefthook: "M400000 281 H103s-33-11.2-61-33.5S0 197.3 0 164s14.2-61.2 42.5\n-83.5C70.8 58.2 104 47 142 47 c16.7 0 25 6.7 25 20 0 12-8.7 18.7-26 20-40 3.3\n-68.7 15.7-86 37-10 12-15 25.3-15 40 0 22.7 9.8 40.7 29.5 54 19.7 13.3 43.5 21\n 71.5 23h399859zM103 281v-40h399897v40z", leftlinesegment: "M40 281 V428 H0 V94 H40 V241 H400000 v40z\nM40 281 V428 H0 V94 H40 V241 H400000 v40z", leftmapsto: "M40 281 V448H0V74H40V241H400000v40z\nM40 281 V448H0V74H40V241H400000v40z", leftToFrom: "M0 147h400000v40H0zm0 214c68 40 115.7 95.7 143 167h22c15.3 0 23\n-.3 23-1 0-1.3-5.3-13.7-16-37-18-35.3-41.3-69-70-101l-7-8h399905v-40H95l7-8\nc28.7-32 52-65.7 70-101 10.7-23.3 16-35.7 16-37 0-.7-7.7-1-23-1h-22C115.7 265.3\n 68 321 0 361zm0-174v-40h399900v40zm100 154v40h399900v-40z", longequal: "M0 50 h400000 v40H0z m0 194h40000v40H0z\nM0 50 h400000 v40H0z m0 194h40000v40H0z", midbrace: "M200428 334\nc-100.7-8.3-195.3-44-280-108-55.3-42-101.7-93-139-153l-9-14c-2.7 4-5.7 8.7-9 14\n-53.3 86.7-123.7 153-211 199-66.7 36-137.3 56.3-212 62H0V214h199568c178.3-11.7\n 311.7-78.3 403-201 6-8 9.7-12 11-12 .7-.7 6.7-1 18-1s17.3.3 18 1c1.3 0 5 4 11\n 12 44.7 59.3 101.3 106.3 170 141s145.3 54.3 229 60h199572v120z", midbraceunder: "M199572 214\nc100.7 8.3 195.3 44 280 108 55.3 42 101.7 93 139 153l9 14c2.7-4 5.7-8.7 9-14\n 53.3-86.7 123.7-153 211-199 66.7-36 137.3-56.3 212-62h199568v120H200432c-178.3\n 11.7-311.7 78.3-403 201-6 8-9.7 12-11 12-.7.7-6.7 1-18 1s-17.3-.3-18-1c-1.3 0\n-5-4-11-12-44.7-59.3-101.3-106.3-170-141s-145.3-54.3-229-60H0V214z", oiintSize1: "M512.6 71.6c272.6 0 320.3 106.8 320.3 178.2 0 70.8-47.7 177.6\n-320.3 177.6S193.1 320.6 193.1 249.8c0-71.4 46.9-178.2 319.5-178.2z\nm368.1 178.2c0-86.4-60.9-215.4-368.1-215.4-306.4 0-367.3 129-367.3 215.4 0 85.8\n60.9 214.8 367.3 214.8 307.2 0 368.1-129 368.1-214.8z", oiintSize2: "M757.8 100.1c384.7 0 451.1 137.6 451.1 230 0 91.3-66.4 228.8\n-451.1 228.8-386.3 0-452.7-137.5-452.7-228.8 0-92.4 66.4-230 452.7-230z\nm502.4 230c0-111.2-82.4-277.2-502.4-277.2s-504 166-504 277.2\nc0 110 84 276 504 276s502.4-166 502.4-276z", oiiintSize1: "M681.4 71.6c408.9 0 480.5 106.8 480.5 178.2 0 70.8-71.6 177.6\n-480.5 177.6S202.1 320.6 202.1 249.8c0-71.4 70.5-178.2 479.3-178.2z\nm525.8 178.2c0-86.4-86.8-215.4-525.7-215.4-437.9 0-524.7 129-524.7 215.4 0\n85.8 86.8 214.8 524.7 214.8 438.9 0 525.7-129 525.7-214.8z", oiiintSize2: "M1021.2 53c603.6 0 707.8 165.8 707.8 277.2 0 110-104.2 275.8\n-707.8 275.8-606 0-710.2-165.8-710.2-275.8C311 218.8 415.2 53 1021.2 53z\nm770.4 277.1c0-131.2-126.4-327.6-770.5-327.6S248.4 198.9 248.4 330.1\nc0 130 128.8 326.4 772.7 326.4s770.5-196.4 770.5-326.4z", rightarrow: "M0 241v40h399891c-47.3 35.3-84 78-110 128\n-16.7 32-27.7 63.7-33 95 0 1.3-.2 2.7-.5 4-.3 1.3-.5 2.3-.5 3 0 7.3 6.7 11 20\n 11 8 0 13.2-.8 15.5-2.5 2.3-1.7 4.2-5.5 5.5-11.5 2-13.3 5.7-27 11-41 14.7-44.7\n 39-84.5 73-119.5s73.7-60.2 119-75.5c6-2 9-5.7 9-11s-3-9-9-11c-45.3-15.3-85\n-40.5-119-75.5s-58.3-74.8-73-119.5c-4.7-14-8.3-27.3-11-40-1.3-6.7-3.2-10.8-5.5\n-12.5-2.3-1.7-7.5-2.5-15.5-2.5-14 0-21 3.7-21 11 0 2 2 10.3 6 25 20.7 83.3 67\n 151.7 139 205zm0 0v40h399900v-40z", rightbrace: "M400000 542l\n-6 6h-17c-12.7 0-19.3-.3-20-1-4-4-7.3-8.3-10-13-35.3-51.3-80.8-93.8-136.5-127.5\ns-117.2-55.8-184.5-66.5c-.7 0-2-.3-4-1-18.7-2.7-76-4.3-172-5H0V214h399571l6 1\nc124.7 8 235 61.7 331 161 31.3 33.3 59.7 72.7 85 118l7 13v35z", rightbraceunder: "M399994 0l6 6v35l-6 11c-56 104-135.3 181.3-238 232-57.3\n 28.7-117 45-179 50H-300V214h399897c43.3-7 81-15 113-26 100.7-33 179.7-91 237\n-174 2.7-5 6-9 10-13 .7-1 7.3-1 20-1h17z", rightgroup: "M0 80h399565c371 0 266.7 149.4 414 180 5.9 1.2 18 0 18 0 2 0\n 3-1 3-3v-38c-76-158-257-219-435-219H0z", rightgroupunder: "M0 262h399565c371 0 266.7-149.4 414-180 5.9-1.2 18 0 18\n 0 2 0 3 1 3 3v38c-76 158-257 219-435 219H0z", rightharpoon: "M0 241v40h399993c4.7-4.7 7-9.3 7-14 0-9.3\n-3.7-15.3-11-18-92.7-56.7-159-133.7-199-231-3.3-9.3-6-14.7-8-16-2-1.3-7-2-15-2\n-10.7 0-16.7 2-18 6-2 2.7-1 9.7 3 21 15.3 42 36.7 81.8 64 119.5 27.3 37.7 58\n 69.2 92 94.5zm0 0v40h399900v-40z", rightharpoonplus: "M0 241v40h399993c4.7-4.7 7-9.3 7-14 0-9.3-3.7-15.3-11\n-18-92.7-56.7-159-133.7-199-231-3.3-9.3-6-14.7-8-16-2-1.3-7-2-15-2-10.7 0-16.7\n 2-18 6-2 2.7-1 9.7 3 21 15.3 42 36.7 81.8 64 119.5 27.3 37.7 58 69.2 92 94.5z\nm0 0v40h399900v-40z m100 194v40h399900v-40zm0 0v40h399900v-40z", rightharpoondown: "M399747 511c0 7.3 6.7 11 20 11 8 0 13-.8 15-2.5s4.7-6.8\n 8-15.5c40-94 99.3-166.3 178-217 13.3-8 20.3-12.3 21-13 5.3-3.3 8.5-5.8 9.5\n-7.5 1-1.7 1.5-5.2 1.5-10.5s-2.3-10.3-7-15H0v40h399908c-34 25.3-64.7 57-92 95\n-27.3 38-48.7 77.7-64 119-3.3 8.7-5 14-5 16zM0 241v40h399900v-40z", rightharpoondownplus: "M399747 705c0 7.3 6.7 11 20 11 8 0 13-.8\n 15-2.5s4.7-6.8 8-15.5c40-94 99.3-166.3 178-217 13.3-8 20.3-12.3 21-13 5.3-3.3\n 8.5-5.8 9.5-7.5 1-1.7 1.5-5.2 1.5-10.5s-2.3-10.3-7-15H0v40h399908c-34 25.3\n-64.7 57-92 95-27.3 38-48.7 77.7-64 119-3.3 8.7-5 14-5 16zM0 435v40h399900v-40z\nm0-194v40h400000v-40zm0 0v40h400000v-40z", righthook: "M399859 241c-764 0 0 0 0 0 40-3.3 68.7-15.7 86-37 10-12 15-25.3\n 15-40 0-22.7-9.8-40.7-29.5-54-19.7-13.3-43.5-21-71.5-23-17.3-1.3-26-8-26-20 0\n-13.3 8.7-20 26-20 38 0 71 11.2 99 33.5 0 0 7 5.6 21 16.7 14 11.2 21 33.5 21\n 66.8s-14 61.2-42 83.5c-28 22.3-61 33.5-99 33.5L0 241z M0 281v-40h399859v40z", rightlinesegment: "M399960 241 V94 h40 V428 h-40 V281 H0 v-40z\nM399960 241 V94 h40 V428 h-40 V281 H0 v-40z", rightToFrom: "M400000 167c-70.7-42-118-97.7-142-167h-23c-15.3 0-23 .3-23\n 1 0 1.3 5.3 13.7 16 37 18 35.3 41.3 69 70 101l7 8H0v40h399905l-7 8c-28.7 32\n-52 65.7-70 101-10.7 23.3-16 35.7-16 37 0 .7 7.7 1 23 1h23c24-69.3 71.3-125 142\n-167z M100 147v40h399900v-40zM0 341v40h399900v-40z", twoheadleftarrow: "M0 167c68 40\n 115.7 95.7 143 167h22c15.3 0 23-.3 23-1 0-1.3-5.3-13.7-16-37-18-35.3-41.3-69\n-70-101l-7-8h125l9 7c50.7 39.3 85 86 103 140h46c0-4.7-6.3-18.7-19-42-18-35.3\n-40-67.3-66-96l-9-9h399716v-40H284l9-9c26-28.7 48-60.7 66-96 12.7-23.333 19\n-37.333 19-42h-46c-18 54-52.3 100.7-103 140l-9 7H95l7-8c28.7-32 52-65.7 70-101\n 10.7-23.333 16-35.7 16-37 0-.7-7.7-1-23-1h-22C115.7 71.3 68 127 0 167z", twoheadrightarrow: "M400000 167\nc-68-40-115.7-95.7-143-167h-22c-15.3 0-23 .3-23 1 0 1.3 5.3 13.7 16 37 18 35.3\n 41.3 69 70 101l7 8h-125l-9-7c-50.7-39.3-85-86-103-140h-46c0 4.7 6.3 18.7 19 42\n 18 35.3 40 67.3 66 96l9 9H0v40h399716l-9 9c-26 28.7-48 60.7-66 96-12.7 23.333\n-19 37.333-19 42h46c18-54 52.3-100.7 103-140l9-7h125l-7 8c-28.7 32-52 65.7-70\n 101-10.7 23.333-16 35.7-16 37 0 .7 7.7 1 23 1h22c27.3-71.3 75-127 143-167z", tilde1: "M200 55.538c-77 0-168 73.953-177 73.953-3 0-7\n-2.175-9-5.437L2 97c-1-2-2-4-2-6 0-4 2-7 5-9l20-12C116 12 171 0 207 0c86 0\n 114 68 191 68 78 0 168-68 177-68 4 0 7 2 9 5l12 19c1 2.175 2 4.35 2 6.525 0\n 4.35-2 7.613-5 9.788l-19 13.05c-92 63.077-116.937 75.308-183 76.128\n-68.267.847-113-73.952-191-73.952z", tilde2: "M344 55.266c-142 0-300.638 81.316-311.5 86.418\n-8.01 3.762-22.5 10.91-23.5 5.562L1 120c-1-2-1-3-1-4 0-5 3-9 8-10l18.4-9C160.9\n 31.9 283 0 358 0c148 0 188 122 331 122s314-97 326-97c4 0 8 2 10 7l7 21.114\nc1 2.14 1 3.21 1 4.28 0 5.347-3 9.626-7 10.696l-22.3 12.622C852.6 158.372 751\n 181.476 676 181.476c-149 0-189-126.21-332-126.21z", tilde3: "M786 59C457 59 32 175.242 13 175.242c-6 0-10-3.457\n-11-10.37L.15 138c-1-7 3-12 10-13l19.2-6.4C378.4 40.7 634.3 0 804.3 0c337 0\n 411.8 157 746.8 157 328 0 754-112 773-112 5 0 10 3 11 9l1 14.075c1 8.066-.697\n 16.595-6.697 17.492l-21.052 7.31c-367.9 98.146-609.15 122.696-778.15 122.696\n -338 0-409-156.573-744-156.573z", tilde4: "M786 58C457 58 32 177.487 13 177.487c-6 0-10-3.345\n-11-10.035L.15 143c-1-7 3-12 10-13l22-6.7C381.2 35 637.15 0 807.15 0c337 0 409\n 177 744 177 328 0 754-127 773-127 5 0 10 3 11 9l1 14.794c1 7.805-3 13.38-9\n 14.495l-20.7 5.574c-366.85 99.79-607.3 139.372-776.3 139.372-338 0-409\n -175.236-744-175.236z", vec: "M377 20c0-5.333 1.833-10 5.5-14S391 0 397 0c4.667 0 8.667 1.667 12 5\n3.333 2.667 6.667 9 10 19 6.667 24.667 20.333 43.667 41 57 7.333 4.667 11\n10.667 11 18 0 6-1 10-3 12s-6.667 5-14 9c-28.667 14.667-53.667 35.667-75 63\n-1.333 1.333-3.167 3.5-5.5 6.5s-4 4.833-5 5.5c-1 .667-2.5 1.333-4.5 2s-4.333 1\n-7 1c-4.667 0-9.167-1.833-13.5-5.5S337 184 337 178c0-12.667 15.667-32.333 47-59\nH213l-171-1c-8.667-6-13-12.333-13-19 0-4.667 4.333-11.333 13-20h359\nc-16-25.333-24-45-24-59z", widehat1: "M529 0h5l519 115c5 1 9 5 9 10 0 1-1 2-1 3l-4 22\nc-1 5-5 9-11 9h-2L532 67 19 159h-2c-5 0-9-4-11-9l-5-22c-1-6 2-12 8-13z", widehat2: "M1181 0h2l1171 176c6 0 10 5 10 11l-2 23c-1 6-5 10\n-11 10h-1L1182 67 15 220h-1c-6 0-10-4-11-10l-2-23c-1-6 4-11 10-11z", widehat3: "M1181 0h2l1171 236c6 0 10 5 10 11l-2 23c-1 6-5 10\n-11 10h-1L1182 67 15 280h-1c-6 0-10-4-11-10l-2-23c-1-6 4-11 10-11z", widehat4: "M1181 0h2l1171 296c6 0 10 5 10 11l-2 23c-1 6-5 10\n-11 10h-1L1182 67 15 340h-1c-6 0-10-4-11-10l-2-23c-1-6 4-11 10-11z", widecheck1: "M529,159h5l519,-115c5,-1,9,-5,9,-10c0,-1,-1,-2,-1,-3l-4,-22c-1,\n-5,-5,-9,-11,-9h-2l-512,92l-513,-92h-2c-5,0,-9,4,-11,9l-5,22c-1,6,2,12,8,13z", widecheck2: "M1181,220h2l1171,-176c6,0,10,-5,10,-11l-2,-23c-1,-6,-5,-10,\n-11,-10h-1l-1168,153l-1167,-153h-1c-6,0,-10,4,-11,10l-2,23c-1,6,4,11,10,11z", widecheck3: "M1181,280h2l1171,-236c6,0,10,-5,10,-11l-2,-23c-1,-6,-5,-10,\n-11,-10h-1l-1168,213l-1167,-213h-1c-6,0,-10,4,-11,10l-2,23c-1,6,4,11,10,11z", widecheck4: "M1181,340h2l1171,-296c6,0,10,-5,10,-11l-2,-23c-1,-6,-5,-10,\n-11,-10h-1l-1168,273l-1167,-273h-1c-6,0,-10,4,-11,10l-2,23c-1,6,4,11,10,11z", baraboveleftarrow: "M400000 620h-399890l3 -3c68.7 -52.7 113.7 -120 135 -202\nc4 -14.7 6 -23 6 -25c0 -7.3 -7 -11 -21 -11c-8 0 -13.2 0.8 -15.5 2.5\nc-2.3 1.7 -4.2 5.8 -5.5 12.5c-1.3 4.7 -2.7 10.3 -4 17c-12 48.7 -34.8 92 -68.5 130\ns-74.2 66.3 -121.5 85c-10 4 -16 7.7 -18 11c0 8.7 6 14.3 18 17c47.3 18.7 87.8 47\n121.5 85s56.5 81.3 68.5 130c0.7 2 1.3 5 2 9s1.2 6.7 1.5 8c0.3 1.3 1 3.3 2 6\ns2.2 4.5 3.5 5.5c1.3 1 3.3 1.8 6 2.5s6 1 10 1c14 0 21 -3.7 21 -11\nc0 -2 -2 -10.3 -6 -25c-20 -79.3 -65 -146.7 -135 -202l-3 -3h399890z\nM100 620v40h399900v-40z M0 241v40h399900v-40zM0 241v40h399900v-40z", rightarrowabovebar: "M0 241v40h399891c-47.3 35.3-84 78-110 128-16.7 32\n-27.7 63.7-33 95 0 1.3-.2 2.7-.5 4-.3 1.3-.5 2.3-.5 3 0 7.3 6.7 11 20 11 8 0\n13.2-.8 15.5-2.5 2.3-1.7 4.2-5.5 5.5-11.5 2-13.3 5.7-27 11-41 14.7-44.7 39\n-84.5 73-119.5s73.7-60.2 119-75.5c6-2 9-5.7 9-11s-3-9-9-11c-45.3-15.3-85-40.5\n-119-75.5s-58.3-74.8-73-119.5c-4.7-14-8.3-27.3-11-40-1.3-6.7-3.2-10.8-5.5\n-12.5-2.3-1.7-7.5-2.5-15.5-2.5-14 0-21 3.7-21 11 0 2 2 10.3 6 25 20.7 83.3 67\n151.7 139 205zm96 379h399894v40H0zm0 0h399904v40H0z", baraboveshortleftharpoon: "M507,435c-4,4,-6.3,8.7,-7,14c0,5.3,0.7,9,2,11\nc1.3,2,5.3,5.3,12,10c90.7,54,156,130,196,228c3.3,10.7,6.3,16.3,9,17\nc2,0.7,5,1,9,1c0,0,5,0,5,0c10.7,0,16.7,-2,18,-6c2,-2.7,1,-9.7,-3,-21\nc-32,-87.3,-82.7,-157.7,-152,-211c0,0,-3,-3,-3,-3l399351,0l0,-40\nc-398570,0,-399437,0,-399437,0z M593 435 v40 H399500 v-40z\nM0 281 v-40 H399908 v40z M0 281 v-40 H399908 v40z", rightharpoonaboveshortbar: "M0,241 l0,40c399126,0,399993,0,399993,0\nc4.7,-4.7,7,-9.3,7,-14c0,-9.3,-3.7,-15.3,-11,-18c-92.7,-56.7,-159,-133.7,-199,\n-231c-3.3,-9.3,-6,-14.7,-8,-16c-2,-1.3,-7,-2,-15,-2c-10.7,0,-16.7,2,-18,6\nc-2,2.7,-1,9.7,3,21c15.3,42,36.7,81.8,64,119.5c27.3,37.7,58,69.2,92,94.5z\nM0 241 v40 H399908 v-40z M0 475 v-40 H399500 v40z M0 475 v-40 H399500 v40z", shortbaraboveleftharpoon: "M7,435c-4,4,-6.3,8.7,-7,14c0,5.3,0.7,9,2,11\nc1.3,2,5.3,5.3,12,10c90.7,54,156,130,196,228c3.3,10.7,6.3,16.3,9,17c2,0.7,5,1,9,\n1c0,0,5,0,5,0c10.7,0,16.7,-2,18,-6c2,-2.7,1,-9.7,-3,-21c-32,-87.3,-82.7,-157.7,\n-152,-211c0,0,-3,-3,-3,-3l399907,0l0,-40c-399126,0,-399993,0,-399993,0z\nM93 435 v40 H400000 v-40z M500 241 v40 H400000 v-40z M500 241 v40 H400000 v-40z", shortrightharpoonabovebar: "M53,241l0,40c398570,0,399437,0,399437,0\nc4.7,-4.7,7,-9.3,7,-14c0,-9.3,-3.7,-15.3,-11,-18c-92.7,-56.7,-159,-133.7,-199,\n-231c-3.3,-9.3,-6,-14.7,-8,-16c-2,-1.3,-7,-2,-15,-2c-10.7,0,-16.7,2,-18,6\nc-2,2.7,-1,9.7,3,21c15.3,42,36.7,81.8,64,119.5c27.3,37.7,58,69.2,92,94.5z\nM500 241 v40 H399408 v-40z M500 435 v40 H400000 v-40z" }; class A { constructor(e) { this.children = void 0, this.classes = void 0, this.height = void 0, this.depth = void 0, this.maxFontSize = void 0, this.style = void 0, this.children = e, this.classes = [], this.height = 0, this.depth = 0, this.maxFontSize = 0, this.style = {} } hasClass(e) { return l.contains(this.classes, e) } toNode() { const e = document.createDocumentFragment(); for (let t = 0; t < this.children.length; t++)e.appendChild(this.children[t].toNode()); return e } toMarkup() { let e = ""; for (let t = 0; t < this.children.length; t++)e += this.children[t].toMarkup(); return e } toText() { return this.children.map((e => e.toText())).join("") } } var T = { "AMS-Regular": { 32: [0, 0, 0, 0, .25], 65: [0, .68889, 0, 0, .72222], 66: [0, .68889, 0, 0, .66667], 67: [0, .68889, 0, 0, .72222], 68: [0, .68889, 0, 0, .72222], 69: [0, .68889, 0, 0, .66667], 70: [0, .68889, 0, 0, .61111], 71: [0, .68889, 0, 0, .77778], 72: [0, .68889, 0, 0, .77778], 73: [0, .68889, 0, 0, .38889], 74: [.16667, .68889, 0, 0, .5], 75: [0, .68889, 0, 0, .77778], 76: [0, .68889, 0, 0, .66667], 77: [0, .68889, 0, 0, .94445], 78: [0, .68889, 0, 0, .72222], 79: [.16667, .68889, 0, 0, .77778], 80: [0, .68889, 0, 0, .61111], 81: [.16667, .68889, 0, 0, .77778], 82: [0, .68889, 0, 0, .72222], 83: [0, .68889, 0, 0, .55556], 84: [0, .68889, 0, 0, .66667], 85: [0, .68889, 0, 0, .72222], 86: [0, .68889, 0, 0, .72222], 87: [0, .68889, 0, 0, 1], 88: [0, .68889, 0, 0, .72222], 89: [0, .68889, 0, 0, .72222], 90: [0, .68889, 0, 0, .66667], 107: [0, .68889, 0, 0, .55556], 160: [0, 0, 0, 0, .25], 165: [0, .675, .025, 0, .75], 174: [.15559, .69224, 0, 0, .94666], 240: [0, .68889, 0, 0, .55556], 295: [0, .68889, 0, 0, .54028], 710: [0, .825, 0, 0, 2.33334], 732: [0, .9, 0, 0, 2.33334], 770: [0, .825, 0, 0, 2.33334], 771: [0, .9, 0, 0, 2.33334], 989: [.08167, .58167, 0, 0, .77778], 1008: [0, .43056, .04028, 0, .66667], 8245: [0, .54986, 0, 0, .275], 8463: [0, .68889, 0, 0, .54028], 8487: [0, .68889, 0, 0, .72222], 8498: [0, .68889, 0, 0, .55556], 8502: [0, .68889, 0, 0, .66667], 8503: [0, .68889, 0, 0, .44445], 8504: [0, .68889, 0, 0, .66667], 8513: [0, .68889, 0, 0, .63889], 8592: [-.03598, .46402, 0, 0, .5], 8594: [-.03598, .46402, 0, 0, .5], 8602: [-.13313, .36687, 0, 0, 1], 8603: [-.13313, .36687, 0, 0, 1], 8606: [.01354, .52239, 0, 0, 1], 8608: [.01354, .52239, 0, 0, 1], 8610: [.01354, .52239, 0, 0, 1.11111], 8611: [.01354, .52239, 0, 0, 1.11111], 8619: [0, .54986, 0, 0, 1], 8620: [0, .54986, 0, 0, 1], 8621: [-.13313, .37788, 0, 0, 1.38889], 8622: [-.13313, .36687, 0, 0, 1], 8624: [0, .69224, 0, 0, .5], 8625: [0, .69224, 0, 0, .5], 8630: [0, .43056, 0, 0, 1], 8631: [0, .43056, 0, 0, 1], 8634: [.08198, .58198, 0, 0, .77778], 8635: [.08198, .58198, 0, 0, .77778], 8638: [.19444, .69224, 0, 0, .41667], 8639: [.19444, .69224, 0, 0, .41667], 8642: [.19444, .69224, 0, 0, .41667], 8643: [.19444, .69224, 0, 0, .41667], 8644: [.1808, .675, 0, 0, 1], 8646: [.1808, .675, 0, 0, 1], 8647: [.1808, .675, 0, 0, 1], 8648: [.19444, .69224, 0, 0, .83334], 8649: [.1808, .675, 0, 0, 1], 8650: [.19444, .69224, 0, 0, .83334], 8651: [.01354, .52239, 0, 0, 1], 8652: [.01354, .52239, 0, 0, 1], 8653: [-.13313, .36687, 0, 0, 1], 8654: [-.13313, .36687, 0, 0, 1], 8655: [-.13313, .36687, 0, 0, 1], 8666: [.13667, .63667, 0, 0, 1], 8667: [.13667, .63667, 0, 0, 1], 8669: [-.13313, .37788, 0, 0, 1], 8672: [-.064, .437, 0, 0, 1.334], 8674: [-.064, .437, 0, 0, 1.334], 8705: [0, .825, 0, 0, .5], 8708: [0, .68889, 0, 0, .55556], 8709: [.08167, .58167, 0, 0, .77778], 8717: [0, .43056, 0, 0, .42917], 8722: [-.03598, .46402, 0, 0, .5], 8724: [.08198, .69224, 0, 0, .77778], 8726: [.08167, .58167, 0, 0, .77778], 8733: [0, .69224, 0, 0, .77778], 8736: [0, .69224, 0, 0, .72222], 8737: [0, .69224, 0, 0, .72222], 8738: [.03517, .52239, 0, 0, .72222], 8739: [.08167, .58167, 0, 0, .22222], 8740: [.25142, .74111, 0, 0, .27778], 8741: [.08167, .58167, 0, 0, .38889], 8742: [.25142, .74111, 0, 0, .5], 8756: [0, .69224, 0, 0, .66667], 8757: [0, .69224, 0, 0, .66667], 8764: [-.13313, .36687, 0, 0, .77778], 8765: [-.13313, .37788, 0, 0, .77778], 8769: [-.13313, .36687, 0, 0, .77778], 8770: [-.03625, .46375, 0, 0, .77778], 8774: [.30274, .79383, 0, 0, .77778], 8776: [-.01688, .48312, 0, 0, .77778], 8778: [.08167, .58167, 0, 0, .77778], 8782: [.06062, .54986, 0, 0, .77778], 8783: [.06062, .54986, 0, 0, .77778], 8785: [.08198, .58198, 0, 0, .77778], 8786: [.08198, .58198, 0, 0, .77778], 8787: [.08198, .58198, 0, 0, .77778], 8790: [0, .69224, 0, 0, .77778], 8791: [.22958, .72958, 0, 0, .77778], 8796: [.08198, .91667, 0, 0, .77778], 8806: [.25583, .75583, 0, 0, .77778], 8807: [.25583, .75583, 0, 0, .77778], 8808: [.25142, .75726, 0, 0, .77778], 8809: [.25142, .75726, 0, 0, .77778], 8812: [.25583, .75583, 0, 0, .5], 8814: [.20576, .70576, 0, 0, .77778], 8815: [.20576, .70576, 0, 0, .77778], 8816: [.30274, .79383, 0, 0, .77778], 8817: [.30274, .79383, 0, 0, .77778], 8818: [.22958, .72958, 0, 0, .77778], 8819: [.22958, .72958, 0, 0, .77778], 8822: [.1808, .675, 0, 0, .77778], 8823: [.1808, .675, 0, 0, .77778], 8828: [.13667, .63667, 0, 0, .77778], 8829: [.13667, .63667, 0, 0, .77778], 8830: [.22958, .72958, 0, 0, .77778], 8831: [.22958, .72958, 0, 0, .77778], 8832: [.20576, .70576, 0, 0, .77778], 8833: [.20576, .70576, 0, 0, .77778], 8840: [.30274, .79383, 0, 0, .77778], 8841: [.30274, .79383, 0, 0, .77778], 8842: [.13597, .63597, 0, 0, .77778], 8843: [.13597, .63597, 0, 0, .77778], 8847: [.03517, .54986, 0, 0, .77778], 8848: [.03517, .54986, 0, 0, .77778], 8858: [.08198, .58198, 0, 0, .77778], 8859: [.08198, .58198, 0, 0, .77778], 8861: [.08198, .58198, 0, 0, .77778], 8862: [0, .675, 0, 0, .77778], 8863: [0, .675, 0, 0, .77778], 8864: [0, .675, 0, 0, .77778], 8865: [0, .675, 0, 0, .77778], 8872: [0, .69224, 0, 0, .61111], 8873: [0, .69224, 0, 0, .72222], 8874: [0, .69224, 0, 0, .88889], 8876: [0, .68889, 0, 0, .61111], 8877: [0, .68889, 0, 0, .61111], 8878: [0, .68889, 0, 0, .72222], 8879: [0, .68889, 0, 0, .72222], 8882: [.03517, .54986, 0, 0, .77778], 8883: [.03517, .54986, 0, 0, .77778], 8884: [.13667, .63667, 0, 0, .77778], 8885: [.13667, .63667, 0, 0, .77778], 8888: [0, .54986, 0, 0, 1.11111], 8890: [.19444, .43056, 0, 0, .55556], 8891: [.19444, .69224, 0, 0, .61111], 8892: [.19444, .69224, 0, 0, .61111], 8901: [0, .54986, 0, 0, .27778], 8903: [.08167, .58167, 0, 0, .77778], 8905: [.08167, .58167, 0, 0, .77778], 8906: [.08167, .58167, 0, 0, .77778], 8907: [0, .69224, 0, 0, .77778], 8908: [0, .69224, 0, 0, .77778], 8909: [-.03598, .46402, 0, 0, .77778], 8910: [0, .54986, 0, 0, .76042], 8911: [0, .54986, 0, 0, .76042], 8912: [.03517, .54986, 0, 0, .77778], 8913: [.03517, .54986, 0, 0, .77778], 8914: [0, .54986, 0, 0, .66667], 8915: [0, .54986, 0, 0, .66667], 8916: [0, .69224, 0, 0, .66667], 8918: [.0391, .5391, 0, 0, .77778], 8919: [.0391, .5391, 0, 0, .77778], 8920: [.03517, .54986, 0, 0, 1.33334], 8921: [.03517, .54986, 0, 0, 1.33334], 8922: [.38569, .88569, 0, 0, .77778], 8923: [.38569, .88569, 0, 0, .77778], 8926: [.13667, .63667, 0, 0, .77778], 8927: [.13667, .63667, 0, 0, .77778], 8928: [.30274, .79383, 0, 0, .77778], 8929: [.30274, .79383, 0, 0, .77778], 8934: [.23222, .74111, 0, 0, .77778], 8935: [.23222, .74111, 0, 0, .77778], 8936: [.23222, .74111, 0, 0, .77778], 8937: [.23222, .74111, 0, 0, .77778], 8938: [.20576, .70576, 0, 0, .77778], 8939: [.20576, .70576, 0, 0, .77778], 8940: [.30274, .79383, 0, 0, .77778], 8941: [.30274, .79383, 0, 0, .77778], 8994: [.19444, .69224, 0, 0, .77778], 8995: [.19444, .69224, 0, 0, .77778], 9416: [.15559, .69224, 0, 0, .90222], 9484: [0, .69224, 0, 0, .5], 9488: [0, .69224, 0, 0, .5], 9492: [0, .37788, 0, 0, .5], 9496: [0, .37788, 0, 0, .5], 9585: [.19444, .68889, 0, 0, .88889], 9586: [.19444, .74111, 0, 0, .88889], 9632: [0, .675, 0, 0, .77778], 9633: [0, .675, 0, 0, .77778], 9650: [0, .54986, 0, 0, .72222], 9651: [0, .54986, 0, 0, .72222], 9654: [.03517, .54986, 0, 0, .77778], 9660: [0, .54986, 0, 0, .72222], 9661: [0, .54986, 0, 0, .72222], 9664: [.03517, .54986, 0, 0, .77778], 9674: [.11111, .69224, 0, 0, .66667], 9733: [.19444, .69224, 0, 0, .94445], 10003: [0, .69224, 0, 0, .83334], 10016: [0, .69224, 0, 0, .83334], 10731: [.11111, .69224, 0, 0, .66667], 10846: [.19444, .75583, 0, 0, .61111], 10877: [.13667, .63667, 0, 0, .77778], 10878: [.13667, .63667, 0, 0, .77778], 10885: [.25583, .75583, 0, 0, .77778], 10886: [.25583, .75583, 0, 0, .77778], 10887: [.13597, .63597, 0, 0, .77778], 10888: [.13597, .63597, 0, 0, .77778], 10889: [.26167, .75726, 0, 0, .77778], 10890: [.26167, .75726, 0, 0, .77778], 10891: [.48256, .98256, 0, 0, .77778], 10892: [.48256, .98256, 0, 0, .77778], 10901: [.13667, .63667, 0, 0, .77778], 10902: [.13667, .63667, 0, 0, .77778], 10933: [.25142, .75726, 0, 0, .77778], 10934: [.25142, .75726, 0, 0, .77778], 10935: [.26167, .75726, 0, 0, .77778], 10936: [.26167, .75726, 0, 0, .77778], 10937: [.26167, .75726, 0, 0, .77778], 10938: [.26167, .75726, 0, 0, .77778], 10949: [.25583, .75583, 0, 0, .77778], 10950: [.25583, .75583, 0, 0, .77778], 10955: [.28481, .79383, 0, 0, .77778], 10956: [.28481, .79383, 0, 0, .77778], 57350: [.08167, .58167, 0, 0, .22222], 57351: [.08167, .58167, 0, 0, .38889], 57352: [.08167, .58167, 0, 0, .77778], 57353: [0, .43056, .04028, 0, .66667], 57356: [.25142, .75726, 0, 0, .77778], 57357: [.25142, .75726, 0, 0, .77778], 57358: [.41951, .91951, 0, 0, .77778], 57359: [.30274, .79383, 0, 0, .77778], 57360: [.30274, .79383, 0, 0, .77778], 57361: [.41951, .91951, 0, 0, .77778], 57366: [.25142, .75726, 0, 0, .77778], 57367: [.25142, .75726, 0, 0, .77778], 57368: [.25142, .75726, 0, 0, .77778], 57369: [.25142, .75726, 0, 0, .77778], 57370: [.13597, .63597, 0, 0, .77778], 57371: [.13597, .63597, 0, 0, .77778] }, "Caligraphic-Regular": { 32: [0, 0, 0, 0, .25], 65: [0, .68333, 0, .19445, .79847], 66: [0, .68333, .03041, .13889, .65681], 67: [0, .68333, .05834, .13889, .52653], 68: [0, .68333, .02778, .08334, .77139], 69: [0, .68333, .08944, .11111, .52778], 70: [0, .68333, .09931, .11111, .71875], 71: [.09722, .68333, .0593, .11111, .59487], 72: [0, .68333, .00965, .11111, .84452], 73: [0, .68333, .07382, 0, .54452], 74: [.09722, .68333, .18472, .16667, .67778], 75: [0, .68333, .01445, .05556, .76195], 76: [0, .68333, 0, .13889, .68972], 77: [0, .68333, 0, .13889, 1.2009], 78: [0, .68333, .14736, .08334, .82049], 79: [0, .68333, .02778, .11111, .79611], 80: [0, .68333, .08222, .08334, .69556], 81: [.09722, .68333, 0, .11111, .81667], 82: [0, .68333, 0, .08334, .8475], 83: [0, .68333, .075, .13889, .60556], 84: [0, .68333, .25417, 0, .54464], 85: [0, .68333, .09931, .08334, .62583], 86: [0, .68333, .08222, 0, .61278], 87: [0, .68333, .08222, .08334, .98778], 88: [0, .68333, .14643, .13889, .7133], 89: [.09722, .68333, .08222, .08334, .66834], 90: [0, .68333, .07944, .13889, .72473], 160: [0, 0, 0, 0, .25] }, "Fraktur-Regular": { 32: [0, 0, 0, 0, .25], 33: [0, .69141, 0, 0, .29574], 34: [0, .69141, 0, 0, .21471], 38: [0, .69141, 0, 0, .73786], 39: [0, .69141, 0, 0, .21201], 40: [.24982, .74947, 0, 0, .38865], 41: [.24982, .74947, 0, 0, .38865], 42: [0, .62119, 0, 0, .27764], 43: [.08319, .58283, 0, 0, .75623], 44: [0, .10803, 0, 0, .27764], 45: [.08319, .58283, 0, 0, .75623], 46: [0, .10803, 0, 0, .27764], 47: [.24982, .74947, 0, 0, .50181], 48: [0, .47534, 0, 0, .50181], 49: [0, .47534, 0, 0, .50181], 50: [0, .47534, 0, 0, .50181], 51: [.18906, .47534, 0, 0, .50181], 52: [.18906, .47534, 0, 0, .50181], 53: [.18906, .47534, 0, 0, .50181], 54: [0, .69141, 0, 0, .50181], 55: [.18906, .47534, 0, 0, .50181], 56: [0, .69141, 0, 0, .50181], 57: [.18906, .47534, 0, 0, .50181], 58: [0, .47534, 0, 0, .21606], 59: [.12604, .47534, 0, 0, .21606], 61: [-.13099, .36866, 0, 0, .75623], 63: [0, .69141, 0, 0, .36245], 65: [0, .69141, 0, 0, .7176], 66: [0, .69141, 0, 0, .88397], 67: [0, .69141, 0, 0, .61254], 68: [0, .69141, 0, 0, .83158], 69: [0, .69141, 0, 0, .66278], 70: [.12604, .69141, 0, 0, .61119], 71: [0, .69141, 0, 0, .78539], 72: [.06302, .69141, 0, 0, .7203], 73: [0, .69141, 0, 0, .55448], 74: [.12604, .69141, 0, 0, .55231], 75: [0, .69141, 0, 0, .66845], 76: [0, .69141, 0, 0, .66602], 77: [0, .69141, 0, 0, 1.04953], 78: [0, .69141, 0, 0, .83212], 79: [0, .69141, 0, 0, .82699], 80: [.18906, .69141, 0, 0, .82753], 81: [.03781, .69141, 0, 0, .82699], 82: [0, .69141, 0, 0, .82807], 83: [0, .69141, 0, 0, .82861], 84: [0, .69141, 0, 0, .66899], 85: [0, .69141, 0, 0, .64576], 86: [0, .69141, 0, 0, .83131], 87: [0, .69141, 0, 0, 1.04602], 88: [0, .69141, 0, 0, .71922], 89: [.18906, .69141, 0, 0, .83293], 90: [.12604, .69141, 0, 0, .60201], 91: [.24982, .74947, 0, 0, .27764], 93: [.24982, .74947, 0, 0, .27764], 94: [0, .69141, 0, 0, .49965], 97: [0, .47534, 0, 0, .50046], 98: [0, .69141, 0, 0, .51315], 99: [0, .47534, 0, 0, .38946], 100: [0, .62119, 0, 0, .49857], 101: [0, .47534, 0, 0, .40053], 102: [.18906, .69141, 0, 0, .32626], 103: [.18906, .47534, 0, 0, .5037], 104: [.18906, .69141, 0, 0, .52126], 105: [0, .69141, 0, 0, .27899], 106: [0, .69141, 0, 0, .28088], 107: [0, .69141, 0, 0, .38946], 108: [0, .69141, 0, 0, .27953], 109: [0, .47534, 0, 0, .76676], 110: [0, .47534, 0, 0, .52666], 111: [0, .47534, 0, 0, .48885], 112: [.18906, .52396, 0, 0, .50046], 113: [.18906, .47534, 0, 0, .48912], 114: [0, .47534, 0, 0, .38919], 115: [0, .47534, 0, 0, .44266], 116: [0, .62119, 0, 0, .33301], 117: [0, .47534, 0, 0, .5172], 118: [0, .52396, 0, 0, .5118], 119: [0, .52396, 0, 0, .77351], 120: [.18906, .47534, 0, 0, .38865], 121: [.18906, .47534, 0, 0, .49884], 122: [.18906, .47534, 0, 0, .39054], 160: [0, 0, 0, 0, .25], 8216: [0, .69141, 0, 0, .21471], 8217: [0, .69141, 0, 0, .21471], 58112: [0, .62119, 0, 0, .49749], 58113: [0, .62119, 0, 0, .4983], 58114: [.18906, .69141, 0, 0, .33328], 58115: [.18906, .69141, 0, 0, .32923], 58116: [.18906, .47534, 0, 0, .50343], 58117: [0, .69141, 0, 0, .33301], 58118: [0, .62119, 0, 0, .33409], 58119: [0, .47534, 0, 0, .50073] }, "Main-Bold": { 32: [0, 0, 0, 0, .25], 33: [0, .69444, 0, 0, .35], 34: [0, .69444, 0, 0, .60278], 35: [.19444, .69444, 0, 0, .95833], 36: [.05556, .75, 0, 0, .575], 37: [.05556, .75, 0, 0, .95833], 38: [0, .69444, 0, 0, .89444], 39: [0, .69444, 0, 0, .31944], 40: [.25, .75, 0, 0, .44722], 41: [.25, .75, 0, 0, .44722], 42: [0, .75, 0, 0, .575], 43: [.13333, .63333, 0, 0, .89444], 44: [.19444, .15556, 0, 0, .31944], 45: [0, .44444, 0, 0, .38333], 46: [0, .15556, 0, 0, .31944], 47: [.25, .75, 0, 0, .575], 48: [0, .64444, 0, 0, .575], 49: [0, .64444, 0, 0, .575], 50: [0, .64444, 0, 0, .575], 51: [0, .64444, 0, 0, .575], 52: [0, .64444, 0, 0, .575], 53: [0, .64444, 0, 0, .575], 54: [0, .64444, 0, 0, .575], 55: [0, .64444, 0, 0, .575], 56: [0, .64444, 0, 0, .575], 57: [0, .64444, 0, 0, .575], 58: [0, .44444, 0, 0, .31944], 59: [.19444, .44444, 0, 0, .31944], 60: [.08556, .58556, 0, 0, .89444], 61: [-.10889, .39111, 0, 0, .89444], 62: [.08556, .58556, 0, 0, .89444], 63: [0, .69444, 0, 0, .54305], 64: [0, .69444, 0, 0, .89444], 65: [0, .68611, 0, 0, .86944], 66: [0, .68611, 0, 0, .81805], 67: [0, .68611, 0, 0, .83055], 68: [0, .68611, 0, 0, .88194], 69: [0, .68611, 0, 0, .75555], 70: [0, .68611, 0, 0, .72361], 71: [0, .68611, 0, 0, .90416], 72: [0, .68611, 0, 0, .9], 73: [0, .68611, 0, 0, .43611], 74: [0, .68611, 0, 0, .59444], 75: [0, .68611, 0, 0, .90138], 76: [0, .68611, 0, 0, .69166], 77: [0, .68611, 0, 0, 1.09166], 78: [0, .68611, 0, 0, .9], 79: [0, .68611, 0, 0, .86388], 80: [0, .68611, 0, 0, .78611], 81: [.19444, .68611, 0, 0, .86388], 82: [0, .68611, 0, 0, .8625], 83: [0, .68611, 0, 0, .63889], 84: [0, .68611, 0, 0, .8], 85: [0, .68611, 0, 0, .88472], 86: [0, .68611, .01597, 0, .86944], 87: [0, .68611, .01597, 0, 1.18888], 88: [0, .68611, 0, 0, .86944], 89: [0, .68611, .02875, 0, .86944], 90: [0, .68611, 0, 0, .70277], 91: [.25, .75, 0, 0, .31944], 92: [.25, .75, 0, 0, .575], 93: [.25, .75, 0, 0, .31944], 94: [0, .69444, 0, 0, .575], 95: [.31, .13444, .03194, 0, .575], 97: [0, .44444, 0, 0, .55902], 98: [0, .69444, 0, 0, .63889], 99: [0, .44444, 0, 0, .51111], 100: [0, .69444, 0, 0, .63889], 101: [0, .44444, 0, 0, .52708], 102: [0, .69444, .10903, 0, .35139], 103: [.19444, .44444, .01597, 0, .575], 104: [0, .69444, 0, 0, .63889], 105: [0, .69444, 0, 0, .31944], 106: [.19444, .69444, 0, 0, .35139], 107: [0, .69444, 0, 0, .60694], 108: [0, .69444, 0, 0, .31944], 109: [0, .44444, 0, 0, .95833], 110: [0, .44444, 0, 0, .63889], 111: [0, .44444, 0, 0, .575], 112: [.19444, .44444, 0, 0, .63889], 113: [.19444, .44444, 0, 0, .60694], 114: [0, .44444, 0, 0, .47361], 115: [0, .44444, 0, 0, .45361], 116: [0, .63492, 0, 0, .44722], 117: [0, .44444, 0, 0, .63889], 118: [0, .44444, .01597, 0, .60694], 119: [0, .44444, .01597, 0, .83055], 120: [0, .44444, 0, 0, .60694], 121: [.19444, .44444, .01597, 0, .60694], 122: [0, .44444, 0, 0, .51111], 123: [.25, .75, 0, 0, .575], 124: [.25, .75, 0, 0, .31944], 125: [.25, .75, 0, 0, .575], 126: [.35, .34444, 0, 0, .575], 160: [0, 0, 0, 0, .25], 163: [0, .69444, 0, 0, .86853], 168: [0, .69444, 0, 0, .575], 172: [0, .44444, 0, 0, .76666], 176: [0, .69444, 0, 0, .86944], 177: [.13333, .63333, 0, 0, .89444], 184: [.17014, 0, 0, 0, .51111], 198: [0, .68611, 0, 0, 1.04166], 215: [.13333, .63333, 0, 0, .89444], 216: [.04861, .73472, 0, 0, .89444], 223: [0, .69444, 0, 0, .59722], 230: [0, .44444, 0, 0, .83055], 247: [.13333, .63333, 0, 0, .89444], 248: [.09722, .54167, 0, 0, .575], 305: [0, .44444, 0, 0, .31944], 338: [0, .68611, 0, 0, 1.16944], 339: [0, .44444, 0, 0, .89444], 567: [.19444, .44444, 0, 0, .35139], 710: [0, .69444, 0, 0, .575], 711: [0, .63194, 0, 0, .575], 713: [0, .59611, 0, 0, .575], 714: [0, .69444, 0, 0, .575], 715: [0, .69444, 0, 0, .575], 728: [0, .69444, 0, 0, .575], 729: [0, .69444, 0, 0, .31944], 730: [0, .69444, 0, 0, .86944], 732: [0, .69444, 0, 0, .575], 733: [0, .69444, 0, 0, .575], 915: [0, .68611, 0, 0, .69166], 916: [0, .68611, 0, 0, .95833], 920: [0, .68611, 0, 0, .89444], 923: [0, .68611, 0, 0, .80555], 926: [0, .68611, 0, 0, .76666], 928: [0, .68611, 0, 0, .9], 931: [0, .68611, 0, 0, .83055], 933: [0, .68611, 0, 0, .89444], 934: [0, .68611, 0, 0, .83055], 936: [0, .68611, 0, 0, .89444], 937: [0, .68611, 0, 0, .83055], 8211: [0, .44444, .03194, 0, .575], 8212: [0, .44444, .03194, 0, 1.14999], 8216: [0, .69444, 0, 0, .31944], 8217: [0, .69444, 0, 0, .31944], 8220: [0, .69444, 0, 0, .60278], 8221: [0, .69444, 0, 0, .60278], 8224: [.19444, .69444, 0, 0, .51111], 8225: [.19444, .69444, 0, 0, .51111], 8242: [0, .55556, 0, 0, .34444], 8407: [0, .72444, .15486, 0, .575], 8463: [0, .69444, 0, 0, .66759], 8465: [0, .69444, 0, 0, .83055], 8467: [0, .69444, 0, 0, .47361], 8472: [.19444, .44444, 0, 0, .74027], 8476: [0, .69444, 0, 0, .83055], 8501: [0, .69444, 0, 0, .70277], 8592: [-.10889, .39111, 0, 0, 1.14999], 8593: [.19444, .69444, 0, 0, .575], 8594: [-.10889, .39111, 0, 0, 1.14999], 8595: [.19444, .69444, 0, 0, .575], 8596: [-.10889, .39111, 0, 0, 1.14999], 8597: [.25, .75, 0, 0, .575], 8598: [.19444, .69444, 0, 0, 1.14999], 8599: [.19444, .69444, 0, 0, 1.14999], 8600: [.19444, .69444, 0, 0, 1.14999], 8601: [.19444, .69444, 0, 0, 1.14999], 8636: [-.10889, .39111, 0, 0, 1.14999], 8637: [-.10889, .39111, 0, 0, 1.14999], 8640: [-.10889, .39111, 0, 0, 1.14999], 8641: [-.10889, .39111, 0, 0, 1.14999], 8656: [-.10889, .39111, 0, 0, 1.14999], 8657: [.19444, .69444, 0, 0, .70277], 8658: [-.10889, .39111, 0, 0, 1.14999], 8659: [.19444, .69444, 0, 0, .70277], 8660: [-.10889, .39111, 0, 0, 1.14999], 8661: [.25, .75, 0, 0, .70277], 8704: [0, .69444, 0, 0, .63889], 8706: [0, .69444, .06389, 0, .62847], 8707: [0, .69444, 0, 0, .63889], 8709: [.05556, .75, 0, 0, .575], 8711: [0, .68611, 0, 0, .95833], 8712: [.08556, .58556, 0, 0, .76666], 8715: [.08556, .58556, 0, 0, .76666], 8722: [.13333, .63333, 0, 0, .89444], 8723: [.13333, .63333, 0, 0, .89444], 8725: [.25, .75, 0, 0, .575], 8726: [.25, .75, 0, 0, .575], 8727: [-.02778, .47222, 0, 0, .575], 8728: [-.02639, .47361, 0, 0, .575], 8729: [-.02639, .47361, 0, 0, .575], 8730: [.18, .82, 0, 0, .95833], 8733: [0, .44444, 0, 0, .89444], 8734: [0, .44444, 0, 0, 1.14999], 8736: [0, .69224, 0, 0, .72222], 8739: [.25, .75, 0, 0, .31944], 8741: [.25, .75, 0, 0, .575], 8743: [0, .55556, 0, 0, .76666], 8744: [0, .55556, 0, 0, .76666], 8745: [0, .55556, 0, 0, .76666], 8746: [0, .55556, 0, 0, .76666], 8747: [.19444, .69444, .12778, 0, .56875], 8764: [-.10889, .39111, 0, 0, .89444], 8768: [.19444, .69444, 0, 0, .31944], 8771: [.00222, .50222, 0, 0, .89444], 8773: [.027, .638, 0, 0, .894], 8776: [.02444, .52444, 0, 0, .89444], 8781: [.00222, .50222, 0, 0, .89444], 8801: [.00222, .50222, 0, 0, .89444], 8804: [.19667, .69667, 0, 0, .89444], 8805: [.19667, .69667, 0, 0, .89444], 8810: [.08556, .58556, 0, 0, 1.14999], 8811: [.08556, .58556, 0, 0, 1.14999], 8826: [.08556, .58556, 0, 0, .89444], 8827: [.08556, .58556, 0, 0, .89444], 8834: [.08556, .58556, 0, 0, .89444], 8835: [.08556, .58556, 0, 0, .89444], 8838: [.19667, .69667, 0, 0, .89444], 8839: [.19667, .69667, 0, 0, .89444], 8846: [0, .55556, 0, 0, .76666], 8849: [.19667, .69667, 0, 0, .89444], 8850: [.19667, .69667, 0, 0, .89444], 8851: [0, .55556, 0, 0, .76666], 8852: [0, .55556, 0, 0, .76666], 8853: [.13333, .63333, 0, 0, .89444], 8854: [.13333, .63333, 0, 0, .89444], 8855: [.13333, .63333, 0, 0, .89444], 8856: [.13333, .63333, 0, 0, .89444], 8857: [.13333, .63333, 0, 0, .89444], 8866: [0, .69444, 0, 0, .70277], 8867: [0, .69444, 0, 0, .70277], 8868: [0, .69444, 0, 0, .89444], 8869: [0, .69444, 0, 0, .89444], 8900: [-.02639, .47361, 0, 0, .575], 8901: [-.02639, .47361, 0, 0, .31944], 8902: [-.02778, .47222, 0, 0, .575], 8968: [.25, .75, 0, 0, .51111], 8969: [.25, .75, 0, 0, .51111], 8970: [.25, .75, 0, 0, .51111], 8971: [.25, .75, 0, 0, .51111], 8994: [-.13889, .36111, 0, 0, 1.14999], 8995: [-.13889, .36111, 0, 0, 1.14999], 9651: [.19444, .69444, 0, 0, 1.02222], 9657: [-.02778, .47222, 0, 0, .575], 9661: [.19444, .69444, 0, 0, 1.02222], 9667: [-.02778, .47222, 0, 0, .575], 9711: [.19444, .69444, 0, 0, 1.14999], 9824: [.12963, .69444, 0, 0, .89444], 9825: [.12963, .69444, 0, 0, .89444], 9826: [.12963, .69444, 0, 0, .89444], 9827: [.12963, .69444, 0, 0, .89444], 9837: [0, .75, 0, 0, .44722], 9838: [.19444, .69444, 0, 0, .44722], 9839: [.19444, .69444, 0, 0, .44722], 10216: [.25, .75, 0, 0, .44722], 10217: [.25, .75, 0, 0, .44722], 10815: [0, .68611, 0, 0, .9], 10927: [.19667, .69667, 0, 0, .89444], 10928: [.19667, .69667, 0, 0, .89444], 57376: [.19444, .69444, 0, 0, 0] }, "Main-BoldItalic": { 32: [0, 0, 0, 0, .25], 33: [0, .69444, .11417, 0, .38611], 34: [0, .69444, .07939, 0, .62055], 35: [.19444, .69444, .06833, 0, .94444], 37: [.05556, .75, .12861, 0, .94444], 38: [0, .69444, .08528, 0, .88555], 39: [0, .69444, .12945, 0, .35555], 40: [.25, .75, .15806, 0, .47333], 41: [.25, .75, .03306, 0, .47333], 42: [0, .75, .14333, 0, .59111], 43: [.10333, .60333, .03306, 0, .88555], 44: [.19444, .14722, 0, 0, .35555], 45: [0, .44444, .02611, 0, .41444], 46: [0, .14722, 0, 0, .35555], 47: [.25, .75, .15806, 0, .59111], 48: [0, .64444, .13167, 0, .59111], 49: [0, .64444, .13167, 0, .59111], 50: [0, .64444, .13167, 0, .59111], 51: [0, .64444, .13167, 0, .59111], 52: [.19444, .64444, .13167, 0, .59111], 53: [0, .64444, .13167, 0, .59111], 54: [0, .64444, .13167, 0, .59111], 55: [.19444, .64444, .13167, 0, .59111], 56: [0, .64444, .13167, 0, .59111], 57: [0, .64444, .13167, 0, .59111], 58: [0, .44444, .06695, 0, .35555], 59: [.19444, .44444, .06695, 0, .35555], 61: [-.10889, .39111, .06833, 0, .88555], 63: [0, .69444, .11472, 0, .59111], 64: [0, .69444, .09208, 0, .88555], 65: [0, .68611, 0, 0, .86555], 66: [0, .68611, .0992, 0, .81666], 67: [0, .68611, .14208, 0, .82666], 68: [0, .68611, .09062, 0, .87555], 69: [0, .68611, .11431, 0, .75666], 70: [0, .68611, .12903, 0, .72722], 71: [0, .68611, .07347, 0, .89527], 72: [0, .68611, .17208, 0, .8961], 73: [0, .68611, .15681, 0, .47166], 74: [0, .68611, .145, 0, .61055], 75: [0, .68611, .14208, 0, .89499], 76: [0, .68611, 0, 0, .69777], 77: [0, .68611, .17208, 0, 1.07277], 78: [0, .68611, .17208, 0, .8961], 79: [0, .68611, .09062, 0, .85499], 80: [0, .68611, .0992, 0, .78721], 81: [.19444, .68611, .09062, 0, .85499], 82: [0, .68611, .02559, 0, .85944], 83: [0, .68611, .11264, 0, .64999], 84: [0, .68611, .12903, 0, .7961], 85: [0, .68611, .17208, 0, .88083], 86: [0, .68611, .18625, 0, .86555], 87: [0, .68611, .18625, 0, 1.15999], 88: [0, .68611, .15681, 0, .86555], 89: [0, .68611, .19803, 0, .86555], 90: [0, .68611, .14208, 0, .70888], 91: [.25, .75, .1875, 0, .35611], 93: [.25, .75, .09972, 0, .35611], 94: [0, .69444, .06709, 0, .59111], 95: [.31, .13444, .09811, 0, .59111], 97: [0, .44444, .09426, 0, .59111], 98: [0, .69444, .07861, 0, .53222], 99: [0, .44444, .05222, 0, .53222], 100: [0, .69444, .10861, 0, .59111], 101: [0, .44444, .085, 0, .53222], 102: [.19444, .69444, .21778, 0, .4], 103: [.19444, .44444, .105, 0, .53222], 104: [0, .69444, .09426, 0, .59111], 105: [0, .69326, .11387, 0, .35555], 106: [.19444, .69326, .1672, 0, .35555], 107: [0, .69444, .11111, 0, .53222], 108: [0, .69444, .10861, 0, .29666], 109: [0, .44444, .09426, 0, .94444], 110: [0, .44444, .09426, 0, .64999], 111: [0, .44444, .07861, 0, .59111], 112: [.19444, .44444, .07861, 0, .59111], 113: [.19444, .44444, .105, 0, .53222], 114: [0, .44444, .11111, 0, .50167], 115: [0, .44444, .08167, 0, .48694], 116: [0, .63492, .09639, 0, .385], 117: [0, .44444, .09426, 0, .62055], 118: [0, .44444, .11111, 0, .53222], 119: [0, .44444, .11111, 0, .76777], 120: [0, .44444, .12583, 0, .56055], 121: [.19444, .44444, .105, 0, .56166], 122: [0, .44444, .13889, 0, .49055], 126: [.35, .34444, .11472, 0, .59111], 160: [0, 0, 0, 0, .25], 168: [0, .69444, .11473, 0, .59111], 176: [0, .69444, 0, 0, .94888], 184: [.17014, 0, 0, 0, .53222], 198: [0, .68611, .11431, 0, 1.02277], 216: [.04861, .73472, .09062, 0, .88555], 223: [.19444, .69444, .09736, 0, .665], 230: [0, .44444, .085, 0, .82666], 248: [.09722, .54167, .09458, 0, .59111], 305: [0, .44444, .09426, 0, .35555], 338: [0, .68611, .11431, 0, 1.14054], 339: [0, .44444, .085, 0, .82666], 567: [.19444, .44444, .04611, 0, .385], 710: [0, .69444, .06709, 0, .59111], 711: [0, .63194, .08271, 0, .59111], 713: [0, .59444, .10444, 0, .59111], 714: [0, .69444, .08528, 0, .59111], 715: [0, .69444, 0, 0, .59111], 728: [0, .69444, .10333, 0, .59111], 729: [0, .69444, .12945, 0, .35555], 730: [0, .69444, 0, 0, .94888], 732: [0, .69444, .11472, 0, .59111], 733: [0, .69444, .11472, 0, .59111], 915: [0, .68611, .12903, 0, .69777], 916: [0, .68611, 0, 0, .94444], 920: [0, .68611, .09062, 0, .88555], 923: [0, .68611, 0, 0, .80666], 926: [0, .68611, .15092, 0, .76777], 928: [0, .68611, .17208, 0, .8961], 931: [0, .68611, .11431, 0, .82666], 933: [0, .68611, .10778, 0, .88555], 934: [0, .68611, .05632, 0, .82666], 936: [0, .68611, .10778, 0, .88555], 937: [0, .68611, .0992, 0, .82666], 8211: [0, .44444, .09811, 0, .59111], 8212: [0, .44444, .09811, 0, 1.18221], 8216: [0, .69444, .12945, 0, .35555], 8217: [0, .69444, .12945, 0, .35555], 8220: [0, .69444, .16772, 0, .62055], 8221: [0, .69444, .07939, 0, .62055] }, "Main-Italic": { 32: [0, 0, 0, 0, .25], 33: [0, .69444, .12417, 0, .30667], 34: [0, .69444, .06961, 0, .51444], 35: [.19444, .69444, .06616, 0, .81777], 37: [.05556, .75, .13639, 0, .81777], 38: [0, .69444, .09694, 0, .76666], 39: [0, .69444, .12417, 0, .30667], 40: [.25, .75, .16194, 0, .40889], 41: [.25, .75, .03694, 0, .40889], 42: [0, .75, .14917, 0, .51111], 43: [.05667, .56167, .03694, 0, .76666], 44: [.19444, .10556, 0, 0, .30667], 45: [0, .43056, .02826, 0, .35778], 46: [0, .10556, 0, 0, .30667], 47: [.25, .75, .16194, 0, .51111], 48: [0, .64444, .13556, 0, .51111], 49: [0, .64444, .13556, 0, .51111], 50: [0, .64444, .13556, 0, .51111], 51: [0, .64444, .13556, 0, .51111], 52: [.19444, .64444, .13556, 0, .51111], 53: [0, .64444, .13556, 0, .51111], 54: [0, .64444, .13556, 0, .51111], 55: [.19444, .64444, .13556, 0, .51111], 56: [0, .64444, .13556, 0, .51111], 57: [0, .64444, .13556, 0, .51111], 58: [0, .43056, .0582, 0, .30667], 59: [.19444, .43056, .0582, 0, .30667], 61: [-.13313, .36687, .06616, 0, .76666], 63: [0, .69444, .1225, 0, .51111], 64: [0, .69444, .09597, 0, .76666], 65: [0, .68333, 0, 0, .74333], 66: [0, .68333, .10257, 0, .70389], 67: [0, .68333, .14528, 0, .71555], 68: [0, .68333, .09403, 0, .755], 69: [0, .68333, .12028, 0, .67833], 70: [0, .68333, .13305, 0, .65277], 71: [0, .68333, .08722, 0, .77361], 72: [0, .68333, .16389, 0, .74333], 73: [0, .68333, .15806, 0, .38555], 74: [0, .68333, .14028, 0, .525], 75: [0, .68333, .14528, 0, .76888], 76: [0, .68333, 0, 0, .62722], 77: [0, .68333, .16389, 0, .89666], 78: [0, .68333, .16389, 0, .74333], 79: [0, .68333, .09403, 0, .76666], 80: [0, .68333, .10257, 0, .67833], 81: [.19444, .68333, .09403, 0, .76666], 82: [0, .68333, .03868, 0, .72944], 83: [0, .68333, .11972, 0, .56222], 84: [0, .68333, .13305, 0, .71555], 85: [0, .68333, .16389, 0, .74333], 86: [0, .68333, .18361, 0, .74333], 87: [0, .68333, .18361, 0, .99888], 88: [0, .68333, .15806, 0, .74333], 89: [0, .68333, .19383, 0, .74333], 90: [0, .68333, .14528, 0, .61333], 91: [.25, .75, .1875, 0, .30667], 93: [.25, .75, .10528, 0, .30667], 94: [0, .69444, .06646, 0, .51111], 95: [.31, .12056, .09208, 0, .51111], 97: [0, .43056, .07671, 0, .51111], 98: [0, .69444, .06312, 0, .46], 99: [0, .43056, .05653, 0, .46], 100: [0, .69444, .10333, 0, .51111], 101: [0, .43056, .07514, 0, .46], 102: [.19444, .69444, .21194, 0, .30667], 103: [.19444, .43056, .08847, 0, .46], 104: [0, .69444, .07671, 0, .51111], 105: [0, .65536, .1019, 0, .30667], 106: [.19444, .65536, .14467, 0, .30667], 107: [0, .69444, .10764, 0, .46], 108: [0, .69444, .10333, 0, .25555], 109: [0, .43056, .07671, 0, .81777], 110: [0, .43056, .07671, 0, .56222], 111: [0, .43056, .06312, 0, .51111], 112: [.19444, .43056, .06312, 0, .51111], 113: [.19444, .43056, .08847, 0, .46], 114: [0, .43056, .10764, 0, .42166], 115: [0, .43056, .08208, 0, .40889], 116: [0, .61508, .09486, 0, .33222], 117: [0, .43056, .07671, 0, .53666], 118: [0, .43056, .10764, 0, .46], 119: [0, .43056, .10764, 0, .66444], 120: [0, .43056, .12042, 0, .46389], 121: [.19444, .43056, .08847, 0, .48555], 122: [0, .43056, .12292, 0, .40889], 126: [.35, .31786, .11585, 0, .51111], 160: [0, 0, 0, 0, .25], 168: [0, .66786, .10474, 0, .51111], 176: [0, .69444, 0, 0, .83129], 184: [.17014, 0, 0, 0, .46], 198: [0, .68333, .12028, 0, .88277], 216: [.04861, .73194, .09403, 0, .76666], 223: [.19444, .69444, .10514, 0, .53666], 230: [0, .43056, .07514, 0, .71555], 248: [.09722, .52778, .09194, 0, .51111], 338: [0, .68333, .12028, 0, .98499], 339: [0, .43056, .07514, 0, .71555], 710: [0, .69444, .06646, 0, .51111], 711: [0, .62847, .08295, 0, .51111], 713: [0, .56167, .10333, 0, .51111], 714: [0, .69444, .09694, 0, .51111], 715: [0, .69444, 0, 0, .51111], 728: [0, .69444, .10806, 0, .51111], 729: [0, .66786, .11752, 0, .30667], 730: [0, .69444, 0, 0, .83129], 732: [0, .66786, .11585, 0, .51111], 733: [0, .69444, .1225, 0, .51111], 915: [0, .68333, .13305, 0, .62722], 916: [0, .68333, 0, 0, .81777], 920: [0, .68333, .09403, 0, .76666], 923: [0, .68333, 0, 0, .69222], 926: [0, .68333, .15294, 0, .66444], 928: [0, .68333, .16389, 0, .74333], 931: [0, .68333, .12028, 0, .71555], 933: [0, .68333, .11111, 0, .76666], 934: [0, .68333, .05986, 0, .71555], 936: [0, .68333, .11111, 0, .76666], 937: [0, .68333, .10257, 0, .71555], 8211: [0, .43056, .09208, 0, .51111], 8212: [0, .43056, .09208, 0, 1.02222], 8216: [0, .69444, .12417, 0, .30667], 8217: [0, .69444, .12417, 0, .30667], 8220: [0, .69444, .1685, 0, .51444], 8221: [0, .69444, .06961, 0, .51444], 8463: [0, .68889, 0, 0, .54028] }, "Main-Regular": { 32: [0, 0, 0, 0, .25], 33: [0, .69444, 0, 0, .27778], 34: [0, .69444, 0, 0, .5], 35: [.19444, .69444, 0, 0, .83334], 36: [.05556, .75, 0, 0, .5], 37: [.05556, .75, 0, 0, .83334], 38: [0, .69444, 0, 0, .77778], 39: [0, .69444, 0, 0, .27778], 40: [.25, .75, 0, 0, .38889], 41: [.25, .75, 0, 0, .38889], 42: [0, .75, 0, 0, .5], 43: [.08333, .58333, 0, 0, .77778], 44: [.19444, .10556, 0, 0, .27778], 45: [0, .43056, 0, 0, .33333], 46: [0, .10556, 0, 0, .27778], 47: [.25, .75, 0, 0, .5], 48: [0, .64444, 0, 0, .5], 49: [0, .64444, 0, 0, .5], 50: [0, .64444, 0, 0, .5], 51: [0, .64444, 0, 0, .5], 52: [0, .64444, 0, 0, .5], 53: [0, .64444, 0, 0, .5], 54: [0, .64444, 0, 0, .5], 55: [0, .64444, 0, 0, .5], 56: [0, .64444, 0, 0, .5], 57: [0, .64444, 0, 0, .5], 58: [0, .43056, 0, 0, .27778], 59: [.19444, .43056, 0, 0, .27778], 60: [.0391, .5391, 0, 0, .77778], 61: [-.13313, .36687, 0, 0, .77778], 62: [.0391, .5391, 0, 0, .77778], 63: [0, .69444, 0, 0, .47222], 64: [0, .69444, 0, 0, .77778], 65: [0, .68333, 0, 0, .75], 66: [0, .68333, 0, 0, .70834], 67: [0, .68333, 0, 0, .72222], 68: [0, .68333, 0, 0, .76389], 69: [0, .68333, 0, 0, .68056], 70: [0, .68333, 0, 0, .65278], 71: [0, .68333, 0, 0, .78472], 72: [0, .68333, 0, 0, .75], 73: [0, .68333, 0, 0, .36111], 74: [0, .68333, 0, 0, .51389], 75: [0, .68333, 0, 0, .77778], 76: [0, .68333, 0, 0, .625], 77: [0, .68333, 0, 0, .91667], 78: [0, .68333, 0, 0, .75], 79: [0, .68333, 0, 0, .77778], 80: [0, .68333, 0, 0, .68056], 81: [.19444, .68333, 0, 0, .77778], 82: [0, .68333, 0, 0, .73611], 83: [0, .68333, 0, 0, .55556], 84: [0, .68333, 0, 0, .72222], 85: [0, .68333, 0, 0, .75], 86: [0, .68333, .01389, 0, .75], 87: [0, .68333, .01389, 0, 1.02778], 88: [0, .68333, 0, 0, .75], 89: [0, .68333, .025, 0, .75], 90: [0, .68333, 0, 0, .61111], 91: [.25, .75, 0, 0, .27778], 92: [.25, .75, 0, 0, .5], 93: [.25, .75, 0, 0, .27778], 94: [0, .69444, 0, 0, .5], 95: [.31, .12056, .02778, 0, .5], 97: [0, .43056, 0, 0, .5], 98: [0, .69444, 0, 0, .55556], 99: [0, .43056, 0, 0, .44445], 100: [0, .69444, 0, 0, .55556], 101: [0, .43056, 0, 0, .44445], 102: [0, .69444, .07778, 0, .30556], 103: [.19444, .43056, .01389, 0, .5], 104: [0, .69444, 0, 0, .55556], 105: [0, .66786, 0, 0, .27778], 106: [.19444, .66786, 0, 0, .30556], 107: [0, .69444, 0, 0, .52778], 108: [0, .69444, 0, 0, .27778], 109: [0, .43056, 0, 0, .83334], 110: [0, .43056, 0, 0, .55556], 111: [0, .43056, 0, 0, .5], 112: [.19444, .43056, 0, 0, .55556], 113: [.19444, .43056, 0, 0, .52778], 114: [0, .43056, 0, 0, .39167], 115: [0, .43056, 0, 0, .39445], 116: [0, .61508, 0, 0, .38889], 117: [0, .43056, 0, 0, .55556], 118: [0, .43056, .01389, 0, .52778], 119: [0, .43056, .01389, 0, .72222], 120: [0, .43056, 0, 0, .52778], 121: [.19444, .43056, .01389, 0, .52778], 122: [0, .43056, 0, 0, .44445], 123: [.25, .75, 0, 0, .5], 124: [.25, .75, 0, 0, .27778], 125: [.25, .75, 0, 0, .5], 126: [.35, .31786, 0, 0, .5], 160: [0, 0, 0, 0, .25], 163: [0, .69444, 0, 0, .76909], 167: [.19444, .69444, 0, 0, .44445], 168: [0, .66786, 0, 0, .5], 172: [0, .43056, 0, 0, .66667], 176: [0, .69444, 0, 0, .75], 177: [.08333, .58333, 0, 0, .77778], 182: [.19444, .69444, 0, 0, .61111], 184: [.17014, 0, 0, 0, .44445], 198: [0, .68333, 0, 0, .90278], 215: [.08333, .58333, 0, 0, .77778], 216: [.04861, .73194, 0, 0, .77778], 223: [0, .69444, 0, 0, .5], 230: [0, .43056, 0, 0, .72222], 247: [.08333, .58333, 0, 0, .77778], 248: [.09722, .52778, 0, 0, .5], 305: [0, .43056, 0, 0, .27778], 338: [0, .68333, 0, 0, 1.01389], 339: [0, .43056, 0, 0, .77778], 567: [.19444, .43056, 0, 0, .30556], 710: [0, .69444, 0, 0, .5], 711: [0, .62847, 0, 0, .5], 713: [0, .56778, 0, 0, .5], 714: [0, .69444, 0, 0, .5], 715: [0, .69444, 0, 0, .5], 728: [0, .69444, 0, 0, .5], 729: [0, .66786, 0, 0, .27778], 730: [0, .69444, 0, 0, .75], 732: [0, .66786, 0, 0, .5], 733: [0, .69444, 0, 0, .5], 915: [0, .68333, 0, 0, .625], 916: [0, .68333, 0, 0, .83334], 920: [0, .68333, 0, 0, .77778], 923: [0, .68333, 0, 0, .69445], 926: [0, .68333, 0, 0, .66667], 928: [0, .68333, 0, 0, .75], 931: [0, .68333, 0, 0, .72222], 933: [0, .68333, 0, 0, .77778], 934: [0, .68333, 0, 0, .72222], 936: [0, .68333, 0, 0, .77778], 937: [0, .68333, 0, 0, .72222], 8211: [0, .43056, .02778, 0, .5], 8212: [0, .43056, .02778, 0, 1], 8216: [0, .69444, 0, 0, .27778], 8217: [0, .69444, 0, 0, .27778], 8220: [0, .69444, 0, 0, .5], 8221: [0, .69444, 0, 0, .5], 8224: [.19444, .69444, 0, 0, .44445], 8225: [.19444, .69444, 0, 0, .44445], 8230: [0, .123, 0, 0, 1.172], 8242: [0, .55556, 0, 0, .275], 8407: [0, .71444, .15382, 0, .5], 8463: [0, .68889, 0, 0, .54028], 8465: [0, .69444, 0, 0, .72222], 8467: [0, .69444, 0, .11111, .41667], 8472: [.19444, .43056, 0, .11111, .63646], 8476: [0, .69444, 0, 0, .72222], 8501: [0, .69444, 0, 0, .61111], 8592: [-.13313, .36687, 0, 0, 1], 8593: [.19444, .69444, 0, 0, .5], 8594: [-.13313, .36687, 0, 0, 1], 8595: [.19444, .69444, 0, 0, .5], 8596: [-.13313, .36687, 0, 0, 1], 8597: [.25, .75, 0, 0, .5], 8598: [.19444, .69444, 0, 0, 1], 8599: [.19444, .69444, 0, 0, 1], 8600: [.19444, .69444, 0, 0, 1], 8601: [.19444, .69444, 0, 0, 1], 8614: [.011, .511, 0, 0, 1], 8617: [.011, .511, 0, 0, 1.126], 8618: [.011, .511, 0, 0, 1.126], 8636: [-.13313, .36687, 0, 0, 1], 8637: [-.13313, .36687, 0, 0, 1], 8640: [-.13313, .36687, 0, 0, 1], 8641: [-.13313, .36687, 0, 0, 1], 8652: [.011, .671, 0, 0, 1], 8656: [-.13313, .36687, 0, 0, 1], 8657: [.19444, .69444, 0, 0, .61111], 8658: [-.13313, .36687, 0, 0, 1], 8659: [.19444, .69444, 0, 0, .61111], 8660: [-.13313, .36687, 0, 0, 1], 8661: [.25, .75, 0, 0, .61111], 8704: [0, .69444, 0, 0, .55556], 8706: [0, .69444, .05556, .08334, .5309], 8707: [0, .69444, 0, 0, .55556], 8709: [.05556, .75, 0, 0, .5], 8711: [0, .68333, 0, 0, .83334], 8712: [.0391, .5391, 0, 0, .66667], 8715: [.0391, .5391, 0, 0, .66667], 8722: [.08333, .58333, 0, 0, .77778], 8723: [.08333, .58333, 0, 0, .77778], 8725: [.25, .75, 0, 0, .5], 8726: [.25, .75, 0, 0, .5], 8727: [-.03472, .46528, 0, 0, .5], 8728: [-.05555, .44445, 0, 0, .5], 8729: [-.05555, .44445, 0, 0, .5], 8730: [.2, .8, 0, 0, .83334], 8733: [0, .43056, 0, 0, .77778], 8734: [0, .43056, 0, 0, 1], 8736: [0, .69224, 0, 0, .72222], 8739: [.25, .75, 0, 0, .27778], 8741: [.25, .75, 0, 0, .5], 8743: [0, .55556, 0, 0, .66667], 8744: [0, .55556, 0, 0, .66667], 8745: [0, .55556, 0, 0, .66667], 8746: [0, .55556, 0, 0, .66667], 8747: [.19444, .69444, .11111, 0, .41667], 8764: [-.13313, .36687, 0, 0, .77778], 8768: [.19444, .69444, 0, 0, .27778], 8771: [-.03625, .46375, 0, 0, .77778], 8773: [-.022, .589, 0, 0, .778], 8776: [-.01688, .48312, 0, 0, .77778], 8781: [-.03625, .46375, 0, 0, .77778], 8784: [-.133, .673, 0, 0, .778], 8801: [-.03625, .46375, 0, 0, .77778], 8804: [.13597, .63597, 0, 0, .77778], 8805: [.13597, .63597, 0, 0, .77778], 8810: [.0391, .5391, 0, 0, 1], 8811: [.0391, .5391, 0, 0, 1], 8826: [.0391, .5391, 0, 0, .77778], 8827: [.0391, .5391, 0, 0, .77778], 8834: [.0391, .5391, 0, 0, .77778], 8835: [.0391, .5391, 0, 0, .77778], 8838: [.13597, .63597, 0, 0, .77778], 8839: [.13597, .63597, 0, 0, .77778], 8846: [0, .55556, 0, 0, .66667], 8849: [.13597, .63597, 0, 0, .77778], 8850: [.13597, .63597, 0, 0, .77778], 8851: [0, .55556, 0, 0, .66667], 8852: [0, .55556, 0, 0, .66667], 8853: [.08333, .58333, 0, 0, .77778], 8854: [.08333, .58333, 0, 0, .77778], 8855: [.08333, .58333, 0, 0, .77778], 8856: [.08333, .58333, 0, 0, .77778], 8857: [.08333, .58333, 0, 0, .77778], 8866: [0, .69444, 0, 0, .61111], 8867: [0, .69444, 0, 0, .61111], 8868: [0, .69444, 0, 0, .77778], 8869: [0, .69444, 0, 0, .77778], 8872: [.249, .75, 0, 0, .867], 8900: [-.05555, .44445, 0, 0, .5], 8901: [-.05555, .44445, 0, 0, .27778], 8902: [-.03472, .46528, 0, 0, .5], 8904: [.005, .505, 0, 0, .9], 8942: [.03, .903, 0, 0, .278], 8943: [-.19, .313, 0, 0, 1.172], 8945: [-.1, .823, 0, 0, 1.282], 8968: [.25, .75, 0, 0, .44445], 8969: [.25, .75, 0, 0, .44445], 8970: [.25, .75, 0, 0, .44445], 8971: [.25, .75, 0, 0, .44445], 8994: [-.14236, .35764, 0, 0, 1], 8995: [-.14236, .35764, 0, 0, 1], 9136: [.244, .744, 0, 0, .412], 9137: [.244, .745, 0, 0, .412], 9651: [.19444, .69444, 0, 0, .88889], 9657: [-.03472, .46528, 0, 0, .5], 9661: [.19444, .69444, 0, 0, .88889], 9667: [-.03472, .46528, 0, 0, .5], 9711: [.19444, .69444, 0, 0, 1], 9824: [.12963, .69444, 0, 0, .77778], 9825: [.12963, .69444, 0, 0, .77778], 9826: [.12963, .69444, 0, 0, .77778], 9827: [.12963, .69444, 0, 0, .77778], 9837: [0, .75, 0, 0, .38889], 9838: [.19444, .69444, 0, 0, .38889], 9839: [.19444, .69444, 0, 0, .38889], 10216: [.25, .75, 0, 0, .38889], 10217: [.25, .75, 0, 0, .38889], 10222: [.244, .744, 0, 0, .412], 10223: [.244, .745, 0, 0, .412], 10229: [.011, .511, 0, 0, 1.609], 10230: [.011, .511, 0, 0, 1.638], 10231: [.011, .511, 0, 0, 1.859], 10232: [.024, .525, 0, 0, 1.609], 10233: [.024, .525, 0, 0, 1.638], 10234: [.024, .525, 0, 0, 1.858], 10236: [.011, .511, 0, 0, 1.638], 10815: [0, .68333, 0, 0, .75], 10927: [.13597, .63597, 0, 0, .77778], 10928: [.13597, .63597, 0, 0, .77778], 57376: [.19444, .69444, 0, 0, 0] }, "Math-BoldItalic": { 32: [0, 0, 0, 0, .25], 48: [0, .44444, 0, 0, .575], 49: [0, .44444, 0, 0, .575], 50: [0, .44444, 0, 0, .575], 51: [.19444, .44444, 0, 0, .575], 52: [.19444, .44444, 0, 0, .575], 53: [.19444, .44444, 0, 0, .575], 54: [0, .64444, 0, 0, .575], 55: [.19444, .44444, 0, 0, .575], 56: [0, .64444, 0, 0, .575], 57: [.19444, .44444, 0, 0, .575], 65: [0, .68611, 0, 0, .86944], 66: [0, .68611, .04835, 0, .8664], 67: [0, .68611, .06979, 0, .81694], 68: [0, .68611, .03194, 0, .93812], 69: [0, .68611, .05451, 0, .81007], 70: [0, .68611, .15972, 0, .68889], 71: [0, .68611, 0, 0, .88673], 72: [0, .68611, .08229, 0, .98229], 73: [0, .68611, .07778, 0, .51111], 74: [0, .68611, .10069, 0, .63125], 75: [0, .68611, .06979, 0, .97118], 76: [0, .68611, 0, 0, .75555], 77: [0, .68611, .11424, 0, 1.14201], 78: [0, .68611, .11424, 0, .95034], 79: [0, .68611, .03194, 0, .83666], 80: [0, .68611, .15972, 0, .72309], 81: [.19444, .68611, 0, 0, .86861], 82: [0, .68611, .00421, 0, .87235], 83: [0, .68611, .05382, 0, .69271], 84: [0, .68611, .15972, 0, .63663], 85: [0, .68611, .11424, 0, .80027], 86: [0, .68611, .25555, 0, .67778], 87: [0, .68611, .15972, 0, 1.09305], 88: [0, .68611, .07778, 0, .94722], 89: [0, .68611, .25555, 0, .67458], 90: [0, .68611, .06979, 0, .77257], 97: [0, .44444, 0, 0, .63287], 98: [0, .69444, 0, 0, .52083], 99: [0, .44444, 0, 0, .51342], 100: [0, .69444, 0, 0, .60972], 101: [0, .44444, 0, 0, .55361], 102: [.19444, .69444, .11042, 0, .56806], 103: [.19444, .44444, .03704, 0, .5449], 104: [0, .69444, 0, 0, .66759], 105: [0, .69326, 0, 0, .4048], 106: [.19444, .69326, .0622, 0, .47083], 107: [0, .69444, .01852, 0, .6037], 108: [0, .69444, .0088, 0, .34815], 109: [0, .44444, 0, 0, 1.0324], 110: [0, .44444, 0, 0, .71296], 111: [0, .44444, 0, 0, .58472], 112: [.19444, .44444, 0, 0, .60092], 113: [.19444, .44444, .03704, 0, .54213], 114: [0, .44444, .03194, 0, .5287], 115: [0, .44444, 0, 0, .53125], 116: [0, .63492, 0, 0, .41528], 117: [0, .44444, 0, 0, .68102], 118: [0, .44444, .03704, 0, .56666], 119: [0, .44444, .02778, 0, .83148], 120: [0, .44444, 0, 0, .65903], 121: [.19444, .44444, .03704, 0, .59028], 122: [0, .44444, .04213, 0, .55509], 160: [0, 0, 0, 0, .25], 915: [0, .68611, .15972, 0, .65694], 916: [0, .68611, 0, 0, .95833], 920: [0, .68611, .03194, 0, .86722], 923: [0, .68611, 0, 0, .80555], 926: [0, .68611, .07458, 0, .84125], 928: [0, .68611, .08229, 0, .98229], 931: [0, .68611, .05451, 0, .88507], 933: [0, .68611, .15972, 0, .67083], 934: [0, .68611, 0, 0, .76666], 936: [0, .68611, .11653, 0, .71402], 937: [0, .68611, .04835, 0, .8789], 945: [0, .44444, 0, 0, .76064], 946: [.19444, .69444, .03403, 0, .65972], 947: [.19444, .44444, .06389, 0, .59003], 948: [0, .69444, .03819, 0, .52222], 949: [0, .44444, 0, 0, .52882], 950: [.19444, .69444, .06215, 0, .50833], 951: [.19444, .44444, .03704, 0, .6], 952: [0, .69444, .03194, 0, .5618], 953: [0, .44444, 0, 0, .41204], 954: [0, .44444, 0, 0, .66759], 955: [0, .69444, 0, 0, .67083], 956: [.19444, .44444, 0, 0, .70787], 957: [0, .44444, .06898, 0, .57685], 958: [.19444, .69444, .03021, 0, .50833], 959: [0, .44444, 0, 0, .58472], 960: [0, .44444, .03704, 0, .68241], 961: [.19444, .44444, 0, 0, .6118], 962: [.09722, .44444, .07917, 0, .42361], 963: [0, .44444, .03704, 0, .68588], 964: [0, .44444, .13472, 0, .52083], 965: [0, .44444, .03704, 0, .63055], 966: [.19444, .44444, 0, 0, .74722], 967: [.19444, .44444, 0, 0, .71805], 968: [.19444, .69444, .03704, 0, .75833], 969: [0, .44444, .03704, 0, .71782], 977: [0, .69444, 0, 0, .69155], 981: [.19444, .69444, 0, 0, .7125], 982: [0, .44444, .03194, 0, .975], 1009: [.19444, .44444, 0, 0, .6118], 1013: [0, .44444, 0, 0, .48333], 57649: [0, .44444, 0, 0, .39352], 57911: [.19444, .44444, 0, 0, .43889] }, "Math-Italic": { 32: [0, 0, 0, 0, .25], 48: [0, .43056, 0, 0, .5], 49: [0, .43056, 0, 0, .5], 50: [0, .43056, 0, 0, .5], 51: [.19444, .43056, 0, 0, .5], 52: [.19444, .43056, 0, 0, .5], 53: [.19444, .43056, 0, 0, .5], 54: [0, .64444, 0, 0, .5], 55: [.19444, .43056, 0, 0, .5], 56: [0, .64444, 0, 0, .5], 57: [.19444, .43056, 0, 0, .5], 65: [0, .68333, 0, .13889, .75], 66: [0, .68333, .05017, .08334, .75851], 67: [0, .68333, .07153, .08334, .71472], 68: [0, .68333, .02778, .05556, .82792], 69: [0, .68333, .05764, .08334, .7382], 70: [0, .68333, .13889, .08334, .64306], 71: [0, .68333, 0, .08334, .78625], 72: [0, .68333, .08125, .05556, .83125], 73: [0, .68333, .07847, .11111, .43958], 74: [0, .68333, .09618, .16667, .55451], 75: [0, .68333, .07153, .05556, .84931], 76: [0, .68333, 0, .02778, .68056], 77: [0, .68333, .10903, .08334, .97014], 78: [0, .68333, .10903, .08334, .80347], 79: [0, .68333, .02778, .08334, .76278], 80: [0, .68333, .13889, .08334, .64201], 81: [.19444, .68333, 0, .08334, .79056], 82: [0, .68333, .00773, .08334, .75929], 83: [0, .68333, .05764, .08334, .6132], 84: [0, .68333, .13889, .08334, .58438], 85: [0, .68333, .10903, .02778, .68278], 86: [0, .68333, .22222, 0, .58333], 87: [0, .68333, .13889, 0, .94445], 88: [0, .68333, .07847, .08334, .82847], 89: [0, .68333, .22222, 0, .58056], 90: [0, .68333, .07153, .08334, .68264], 97: [0, .43056, 0, 0, .52859], 98: [0, .69444, 0, 0, .42917], 99: [0, .43056, 0, .05556, .43276], 100: [0, .69444, 0, .16667, .52049], 101: [0, .43056, 0, .05556, .46563], 102: [.19444, .69444, .10764, .16667, .48959], 103: [.19444, .43056, .03588, .02778, .47697], 104: [0, .69444, 0, 0, .57616], 105: [0, .65952, 0, 0, .34451], 106: [.19444, .65952, .05724, 0, .41181], 107: [0, .69444, .03148, 0, .5206], 108: [0, .69444, .01968, .08334, .29838], 109: [0, .43056, 0, 0, .87801], 110: [0, .43056, 0, 0, .60023], 111: [0, .43056, 0, .05556, .48472], 112: [.19444, .43056, 0, .08334, .50313], 113: [.19444, .43056, .03588, .08334, .44641], 114: [0, .43056, .02778, .05556, .45116], 115: [0, .43056, 0, .05556, .46875], 116: [0, .61508, 0, .08334, .36111], 117: [0, .43056, 0, .02778, .57246], 118: [0, .43056, .03588, .02778, .48472], 119: [0, .43056, .02691, .08334, .71592], 120: [0, .43056, 0, .02778, .57153], 121: [.19444, .43056, .03588, .05556, .49028], 122: [0, .43056, .04398, .05556, .46505], 160: [0, 0, 0, 0, .25], 915: [0, .68333, .13889, .08334, .61528], 916: [0, .68333, 0, .16667, .83334], 920: [0, .68333, .02778, .08334, .76278], 923: [0, .68333, 0, .16667, .69445], 926: [0, .68333, .07569, .08334, .74236], 928: [0, .68333, .08125, .05556, .83125], 931: [0, .68333, .05764, .08334, .77986], 933: [0, .68333, .13889, .05556, .58333], 934: [0, .68333, 0, .08334, .66667], 936: [0, .68333, .11, .05556, .61222], 937: [0, .68333, .05017, .08334, .7724], 945: [0, .43056, .0037, .02778, .6397], 946: [.19444, .69444, .05278, .08334, .56563], 947: [.19444, .43056, .05556, 0, .51773], 948: [0, .69444, .03785, .05556, .44444], 949: [0, .43056, 0, .08334, .46632], 950: [.19444, .69444, .07378, .08334, .4375], 951: [.19444, .43056, .03588, .05556, .49653], 952: [0, .69444, .02778, .08334, .46944], 953: [0, .43056, 0, .05556, .35394], 954: [0, .43056, 0, 0, .57616], 955: [0, .69444, 0, 0, .58334], 956: [.19444, .43056, 0, .02778, .60255], 957: [0, .43056, .06366, .02778, .49398], 958: [.19444, .69444, .04601, .11111, .4375], 959: [0, .43056, 0, .05556, .48472], 960: [0, .43056, .03588, 0, .57003], 961: [.19444, .43056, 0, .08334, .51702], 962: [.09722, .43056, .07986, .08334, .36285], 963: [0, .43056, .03588, 0, .57141], 964: [0, .43056, .1132, .02778, .43715], 965: [0, .43056, .03588, .02778, .54028], 966: [.19444, .43056, 0, .08334, .65417], 967: [.19444, .43056, 0, .05556, .62569], 968: [.19444, .69444, .03588, .11111, .65139], 969: [0, .43056, .03588, 0, .62245], 977: [0, .69444, 0, .08334, .59144], 981: [.19444, .69444, 0, .08334, .59583], 982: [0, .43056, .02778, 0, .82813], 1009: [.19444, .43056, 0, .08334, .51702], 1013: [0, .43056, 0, .05556, .4059], 57649: [0, .43056, 0, .02778, .32246], 57911: [.19444, .43056, 0, .08334, .38403] }, "SansSerif-Bold": { 32: [0, 0, 0, 0, .25], 33: [0, .69444, 0, 0, .36667], 34: [0, .69444, 0, 0, .55834], 35: [.19444, .69444, 0, 0, .91667], 36: [.05556, .75, 0, 0, .55], 37: [.05556, .75, 0, 0, 1.02912], 38: [0, .69444, 0, 0, .83056], 39: [0, .69444, 0, 0, .30556], 40: [.25, .75, 0, 0, .42778], 41: [.25, .75, 0, 0, .42778], 42: [0, .75, 0, 0, .55], 43: [.11667, .61667, 0, 0, .85556], 44: [.10556, .13056, 0, 0, .30556], 45: [0, .45833, 0, 0, .36667], 46: [0, .13056, 0, 0, .30556], 47: [.25, .75, 0, 0, .55], 48: [0, .69444, 0, 0, .55], 49: [0, .69444, 0, 0, .55], 50: [0, .69444, 0, 0, .55], 51: [0, .69444, 0, 0, .55], 52: [0, .69444, 0, 0, .55], 53: [0, .69444, 0, 0, .55], 54: [0, .69444, 0, 0, .55], 55: [0, .69444, 0, 0, .55], 56: [0, .69444, 0, 0, .55], 57: [0, .69444, 0, 0, .55], 58: [0, .45833, 0, 0, .30556], 59: [.10556, .45833, 0, 0, .30556], 61: [-.09375, .40625, 0, 0, .85556], 63: [0, .69444, 0, 0, .51945], 64: [0, .69444, 0, 0, .73334], 65: [0, .69444, 0, 0, .73334], 66: [0, .69444, 0, 0, .73334], 67: [0, .69444, 0, 0, .70278], 68: [0, .69444, 0, 0, .79445], 69: [0, .69444, 0, 0, .64167], 70: [0, .69444, 0, 0, .61111], 71: [0, .69444, 0, 0, .73334], 72: [0, .69444, 0, 0, .79445], 73: [0, .69444, 0, 0, .33056], 74: [0, .69444, 0, 0, .51945], 75: [0, .69444, 0, 0, .76389], 76: [0, .69444, 0, 0, .58056], 77: [0, .69444, 0, 0, .97778], 78: [0, .69444, 0, 0, .79445], 79: [0, .69444, 0, 0, .79445], 80: [0, .69444, 0, 0, .70278], 81: [.10556, .69444, 0, 0, .79445], 82: [0, .69444, 0, 0, .70278], 83: [0, .69444, 0, 0, .61111], 84: [0, .69444, 0, 0, .73334], 85: [0, .69444, 0, 0, .76389], 86: [0, .69444, .01528, 0, .73334], 87: [0, .69444, .01528, 0, 1.03889], 88: [0, .69444, 0, 0, .73334], 89: [0, .69444, .0275, 0, .73334], 90: [0, .69444, 0, 0, .67223], 91: [.25, .75, 0, 0, .34306], 93: [.25, .75, 0, 0, .34306], 94: [0, .69444, 0, 0, .55], 95: [.35, .10833, .03056, 0, .55], 97: [0, .45833, 0, 0, .525], 98: [0, .69444, 0, 0, .56111], 99: [0, .45833, 0, 0, .48889], 100: [0, .69444, 0, 0, .56111], 101: [0, .45833, 0, 0, .51111], 102: [0, .69444, .07639, 0, .33611], 103: [.19444, .45833, .01528, 0, .55], 104: [0, .69444, 0, 0, .56111], 105: [0, .69444, 0, 0, .25556], 106: [.19444, .69444, 0, 0, .28611], 107: [0, .69444, 0, 0, .53056], 108: [0, .69444, 0, 0, .25556], 109: [0, .45833, 0, 0, .86667], 110: [0, .45833, 0, 0, .56111], 111: [0, .45833, 0, 0, .55], 112: [.19444, .45833, 0, 0, .56111], 113: [.19444, .45833, 0, 0, .56111], 114: [0, .45833, .01528, 0, .37222], 115: [0, .45833, 0, 0, .42167], 116: [0, .58929, 0, 0, .40417], 117: [0, .45833, 0, 0, .56111], 118: [0, .45833, .01528, 0, .5], 119: [0, .45833, .01528, 0, .74445], 120: [0, .45833, 0, 0, .5], 121: [.19444, .45833, .01528, 0, .5], 122: [0, .45833, 0, 0, .47639], 126: [.35, .34444, 0, 0, .55], 160: [0, 0, 0, 0, .25], 168: [0, .69444, 0, 0, .55], 176: [0, .69444, 0, 0, .73334], 180: [0, .69444, 0, 0, .55], 184: [.17014, 0, 0, 0, .48889], 305: [0, .45833, 0, 0, .25556], 567: [.19444, .45833, 0, 0, .28611], 710: [0, .69444, 0, 0, .55], 711: [0, .63542, 0, 0, .55], 713: [0, .63778, 0, 0, .55], 728: [0, .69444, 0, 0, .55], 729: [0, .69444, 0, 0, .30556], 730: [0, .69444, 0, 0, .73334], 732: [0, .69444, 0, 0, .55], 733: [0, .69444, 0, 0, .55], 915: [0, .69444, 0, 0, .58056], 916: [0, .69444, 0, 0, .91667], 920: [0, .69444, 0, 0, .85556], 923: [0, .69444, 0, 0, .67223], 926: [0, .69444, 0, 0, .73334], 928: [0, .69444, 0, 0, .79445], 931: [0, .69444, 0, 0, .79445], 933: [0, .69444, 0, 0, .85556], 934: [0, .69444, 0, 0, .79445], 936: [0, .69444, 0, 0, .85556], 937: [0, .69444, 0, 0, .79445], 8211: [0, .45833, .03056, 0, .55], 8212: [0, .45833, .03056, 0, 1.10001], 8216: [0, .69444, 0, 0, .30556], 8217: [0, .69444, 0, 0, .30556], 8220: [0, .69444, 0, 0, .55834], 8221: [0, .69444, 0, 0, .55834] }, "SansSerif-Italic": { 32: [0, 0, 0, 0, .25], 33: [0, .69444, .05733, 0, .31945], 34: [0, .69444, .00316, 0, .5], 35: [.19444, .69444, .05087, 0, .83334], 36: [.05556, .75, .11156, 0, .5], 37: [.05556, .75, .03126, 0, .83334], 38: [0, .69444, .03058, 0, .75834], 39: [0, .69444, .07816, 0, .27778], 40: [.25, .75, .13164, 0, .38889], 41: [.25, .75, .02536, 0, .38889], 42: [0, .75, .11775, 0, .5], 43: [.08333, .58333, .02536, 0, .77778], 44: [.125, .08333, 0, 0, .27778], 45: [0, .44444, .01946, 0, .33333], 46: [0, .08333, 0, 0, .27778], 47: [.25, .75, .13164, 0, .5], 48: [0, .65556, .11156, 0, .5], 49: [0, .65556, .11156, 0, .5], 50: [0, .65556, .11156, 0, .5], 51: [0, .65556, .11156, 0, .5], 52: [0, .65556, .11156, 0, .5], 53: [0, .65556, .11156, 0, .5], 54: [0, .65556, .11156, 0, .5], 55: [0, .65556, .11156, 0, .5], 56: [0, .65556, .11156, 0, .5], 57: [0, .65556, .11156, 0, .5], 58: [0, .44444, .02502, 0, .27778], 59: [.125, .44444, .02502, 0, .27778], 61: [-.13, .37, .05087, 0, .77778], 63: [0, .69444, .11809, 0, .47222], 64: [0, .69444, .07555, 0, .66667], 65: [0, .69444, 0, 0, .66667], 66: [0, .69444, .08293, 0, .66667], 67: [0, .69444, .11983, 0, .63889], 68: [0, .69444, .07555, 0, .72223], 69: [0, .69444, .11983, 0, .59722], 70: [0, .69444, .13372, 0, .56945], 71: [0, .69444, .11983, 0, .66667], 72: [0, .69444, .08094, 0, .70834], 73: [0, .69444, .13372, 0, .27778], 74: [0, .69444, .08094, 0, .47222], 75: [0, .69444, .11983, 0, .69445], 76: [0, .69444, 0, 0, .54167], 77: [0, .69444, .08094, 0, .875], 78: [0, .69444, .08094, 0, .70834], 79: [0, .69444, .07555, 0, .73611], 80: [0, .69444, .08293, 0, .63889], 81: [.125, .69444, .07555, 0, .73611], 82: [0, .69444, .08293, 0, .64584], 83: [0, .69444, .09205, 0, .55556], 84: [0, .69444, .13372, 0, .68056], 85: [0, .69444, .08094, 0, .6875], 86: [0, .69444, .1615, 0, .66667], 87: [0, .69444, .1615, 0, .94445], 88: [0, .69444, .13372, 0, .66667], 89: [0, .69444, .17261, 0, .66667], 90: [0, .69444, .11983, 0, .61111], 91: [.25, .75, .15942, 0, .28889], 93: [.25, .75, .08719, 0, .28889], 94: [0, .69444, .0799, 0, .5], 95: [.35, .09444, .08616, 0, .5], 97: [0, .44444, .00981, 0, .48056], 98: [0, .69444, .03057, 0, .51667], 99: [0, .44444, .08336, 0, .44445], 100: [0, .69444, .09483, 0, .51667], 101: [0, .44444, .06778, 0, .44445], 102: [0, .69444, .21705, 0, .30556], 103: [.19444, .44444, .10836, 0, .5], 104: [0, .69444, .01778, 0, .51667], 105: [0, .67937, .09718, 0, .23889], 106: [.19444, .67937, .09162, 0, .26667], 107: [0, .69444, .08336, 0, .48889], 108: [0, .69444, .09483, 0, .23889], 109: [0, .44444, .01778, 0, .79445], 110: [0, .44444, .01778, 0, .51667], 111: [0, .44444, .06613, 0, .5], 112: [.19444, .44444, .0389, 0, .51667], 113: [.19444, .44444, .04169, 0, .51667], 114: [0, .44444, .10836, 0, .34167], 115: [0, .44444, .0778, 0, .38333], 116: [0, .57143, .07225, 0, .36111], 117: [0, .44444, .04169, 0, .51667], 118: [0, .44444, .10836, 0, .46111], 119: [0, .44444, .10836, 0, .68334], 120: [0, .44444, .09169, 0, .46111], 121: [.19444, .44444, .10836, 0, .46111], 122: [0, .44444, .08752, 0, .43472], 126: [.35, .32659, .08826, 0, .5], 160: [0, 0, 0, 0, .25], 168: [0, .67937, .06385, 0, .5], 176: [0, .69444, 0, 0, .73752], 184: [.17014, 0, 0, 0, .44445], 305: [0, .44444, .04169, 0, .23889], 567: [.19444, .44444, .04169, 0, .26667], 710: [0, .69444, .0799, 0, .5], 711: [0, .63194, .08432, 0, .5], 713: [0, .60889, .08776, 0, .5], 714: [0, .69444, .09205, 0, .5], 715: [0, .69444, 0, 0, .5], 728: [0, .69444, .09483, 0, .5], 729: [0, .67937, .07774, 0, .27778], 730: [0, .69444, 0, 0, .73752], 732: [0, .67659, .08826, 0, .5], 733: [0, .69444, .09205, 0, .5], 915: [0, .69444, .13372, 0, .54167], 916: [0, .69444, 0, 0, .83334], 920: [0, .69444, .07555, 0, .77778], 923: [0, .69444, 0, 0, .61111], 926: [0, .69444, .12816, 0, .66667], 928: [0, .69444, .08094, 0, .70834], 931: [0, .69444, .11983, 0, .72222], 933: [0, .69444, .09031, 0, .77778], 934: [0, .69444, .04603, 0, .72222], 936: [0, .69444, .09031, 0, .77778], 937: [0, .69444, .08293, 0, .72222], 8211: [0, .44444, .08616, 0, .5], 8212: [0, .44444, .08616, 0, 1], 8216: [0, .69444, .07816, 0, .27778], 8217: [0, .69444, .07816, 0, .27778], 8220: [0, .69444, .14205, 0, .5], 8221: [0, .69444, .00316, 0, .5] }, "SansSerif-Regular": { 32: [0, 0, 0, 0, .25], 33: [0, .69444, 0, 0, .31945], 34: [0, .69444, 0, 0, .5], 35: [.19444, .69444, 0, 0, .83334], 36: [.05556, .75, 0, 0, .5], 37: [.05556, .75, 0, 0, .83334], 38: [0, .69444, 0, 0, .75834], 39: [0, .69444, 0, 0, .27778], 40: [.25, .75, 0, 0, .38889], 41: [.25, .75, 0, 0, .38889], 42: [0, .75, 0, 0, .5], 43: [.08333, .58333, 0, 0, .77778], 44: [.125, .08333, 0, 0, .27778], 45: [0, .44444, 0, 0, .33333], 46: [0, .08333, 0, 0, .27778], 47: [.25, .75, 0, 0, .5], 48: [0, .65556, 0, 0, .5], 49: [0, .65556, 0, 0, .5], 50: [0, .65556, 0, 0, .5], 51: [0, .65556, 0, 0, .5], 52: [0, .65556, 0, 0, .5], 53: [0, .65556, 0, 0, .5], 54: [0, .65556, 0, 0, .5], 55: [0, .65556, 0, 0, .5], 56: [0, .65556, 0, 0, .5], 57: [0, .65556, 0, 0, .5], 58: [0, .44444, 0, 0, .27778], 59: [.125, .44444, 0, 0, .27778], 61: [-.13, .37, 0, 0, .77778], 63: [0, .69444, 0, 0, .47222], 64: [0, .69444, 0, 0, .66667], 65: [0, .69444, 0, 0, .66667], 66: [0, .69444, 0, 0, .66667], 67: [0, .69444, 0, 0, .63889], 68: [0, .69444, 0, 0, .72223], 69: [0, .69444, 0, 0, .59722], 70: [0, .69444, 0, 0, .56945], 71: [0, .69444, 0, 0, .66667], 72: [0, .69444, 0, 0, .70834], 73: [0, .69444, 0, 0, .27778], 74: [0, .69444, 0, 0, .47222], 75: [0, .69444, 0, 0, .69445], 76: [0, .69444, 0, 0, .54167], 77: [0, .69444, 0, 0, .875], 78: [0, .69444, 0, 0, .70834], 79: [0, .69444, 0, 0, .73611], 80: [0, .69444, 0, 0, .63889], 81: [.125, .69444, 0, 0, .73611], 82: [0, .69444, 0, 0, .64584], 83: [0, .69444, 0, 0, .55556], 84: [0, .69444, 0, 0, .68056], 85: [0, .69444, 0, 0, .6875], 86: [0, .69444, .01389, 0, .66667], 87: [0, .69444, .01389, 0, .94445], 88: [0, .69444, 0, 0, .66667], 89: [0, .69444, .025, 0, .66667], 90: [0, .69444, 0, 0, .61111], 91: [.25, .75, 0, 0, .28889], 93: [.25, .75, 0, 0, .28889], 94: [0, .69444, 0, 0, .5], 95: [.35, .09444, .02778, 0, .5], 97: [0, .44444, 0, 0, .48056], 98: [0, .69444, 0, 0, .51667], 99: [0, .44444, 0, 0, .44445], 100: [0, .69444, 0, 0, .51667], 101: [0, .44444, 0, 0, .44445], 102: [0, .69444, .06944, 0, .30556], 103: [.19444, .44444, .01389, 0, .5], 104: [0, .69444, 0, 0, .51667], 105: [0, .67937, 0, 0, .23889], 106: [.19444, .67937, 0, 0, .26667], 107: [0, .69444, 0, 0, .48889], 108: [0, .69444, 0, 0, .23889], 109: [0, .44444, 0, 0, .79445], 110: [0, .44444, 0, 0, .51667], 111: [0, .44444, 0, 0, .5], 112: [.19444, .44444, 0, 0, .51667], 113: [.19444, .44444, 0, 0, .51667], 114: [0, .44444, .01389, 0, .34167], 115: [0, .44444, 0, 0, .38333], 116: [0, .57143, 0, 0, .36111], 117: [0, .44444, 0, 0, .51667], 118: [0, .44444, .01389, 0, .46111], 119: [0, .44444, .01389, 0, .68334], 120: [0, .44444, 0, 0, .46111], 121: [.19444, .44444, .01389, 0, .46111], 122: [0, .44444, 0, 0, .43472], 126: [.35, .32659, 0, 0, .5], 160: [0, 0, 0, 0, .25], 168: [0, .67937, 0, 0, .5], 176: [0, .69444, 0, 0, .66667], 184: [.17014, 0, 0, 0, .44445], 305: [0, .44444, 0, 0, .23889], 567: [.19444, .44444, 0, 0, .26667], 710: [0, .69444, 0, 0, .5], 711: [0, .63194, 0, 0, .5], 713: [0, .60889, 0, 0, .5], 714: [0, .69444, 0, 0, .5], 715: [0, .69444, 0, 0, .5], 728: [0, .69444, 0, 0, .5], 729: [0, .67937, 0, 0, .27778], 730: [0, .69444, 0, 0, .66667], 732: [0, .67659, 0, 0, .5], 733: [0, .69444, 0, 0, .5], 915: [0, .69444, 0, 0, .54167], 916: [0, .69444, 0, 0, .83334], 920: [0, .69444, 0, 0, .77778], 923: [0, .69444, 0, 0, .61111], 926: [0, .69444, 0, 0, .66667], 928: [0, .69444, 0, 0, .70834], 931: [0, .69444, 0, 0, .72222], 933: [0, .69444, 0, 0, .77778], 934: [0, .69444, 0, 0, .72222], 936: [0, .69444, 0, 0, .77778], 937: [0, .69444, 0, 0, .72222], 8211: [0, .44444, .02778, 0, .5], 8212: [0, .44444, .02778, 0, 1], 8216: [0, .69444, 0, 0, .27778], 8217: [0, .69444, 0, 0, .27778], 8220: [0, .69444, 0, 0, .5], 8221: [0, .69444, 0, 0, .5] }, "Script-Regular": { 32: [0, 0, 0, 0, .25], 65: [0, .7, .22925, 0, .80253], 66: [0, .7, .04087, 0, .90757], 67: [0, .7, .1689, 0, .66619], 68: [0, .7, .09371, 0, .77443], 69: [0, .7, .18583, 0, .56162], 70: [0, .7, .13634, 0, .89544], 71: [0, .7, .17322, 0, .60961], 72: [0, .7, .29694, 0, .96919], 73: [0, .7, .19189, 0, .80907], 74: [.27778, .7, .19189, 0, 1.05159], 75: [0, .7, .31259, 0, .91364], 76: [0, .7, .19189, 0, .87373], 77: [0, .7, .15981, 0, 1.08031], 78: [0, .7, .3525, 0, .9015], 79: [0, .7, .08078, 0, .73787], 80: [0, .7, .08078, 0, 1.01262], 81: [0, .7, .03305, 0, .88282], 82: [0, .7, .06259, 0, .85], 83: [0, .7, .19189, 0, .86767], 84: [0, .7, .29087, 0, .74697], 85: [0, .7, .25815, 0, .79996], 86: [0, .7, .27523, 0, .62204], 87: [0, .7, .27523, 0, .80532], 88: [0, .7, .26006, 0, .94445], 89: [0, .7, .2939, 0, .70961], 90: [0, .7, .24037, 0, .8212], 160: [0, 0, 0, 0, .25] }, "Size1-Regular": { 32: [0, 0, 0, 0, .25], 40: [.35001, .85, 0, 0, .45834], 41: [.35001, .85, 0, 0, .45834], 47: [.35001, .85, 0, 0, .57778], 91: [.35001, .85, 0, 0, .41667], 92: [.35001, .85, 0, 0, .57778], 93: [.35001, .85, 0, 0, .41667], 123: [.35001, .85, 0, 0, .58334], 125: [.35001, .85, 0, 0, .58334], 160: [0, 0, 0, 0, .25], 710: [0, .72222, 0, 0, .55556], 732: [0, .72222, 0, 0, .55556], 770: [0, .72222, 0, 0, .55556], 771: [0, .72222, 0, 0, .55556], 8214: [-99e-5, .601, 0, 0, .77778], 8593: [1e-5, .6, 0, 0, .66667], 8595: [1e-5, .6, 0, 0, .66667], 8657: [1e-5, .6, 0, 0, .77778], 8659: [1e-5, .6, 0, 0, .77778], 8719: [.25001, .75, 0, 0, .94445], 8720: [.25001, .75, 0, 0, .94445], 8721: [.25001, .75, 0, 0, 1.05556], 8730: [.35001, .85, 0, 0, 1], 8739: [-.00599, .606, 0, 0, .33333], 8741: [-.00599, .606, 0, 0, .55556], 8747: [.30612, .805, .19445, 0, .47222], 8748: [.306, .805, .19445, 0, .47222], 8749: [.306, .805, .19445, 0, .47222], 8750: [.30612, .805, .19445, 0, .47222], 8896: [.25001, .75, 0, 0, .83334], 8897: [.25001, .75, 0, 0, .83334], 8898: [.25001, .75, 0, 0, .83334], 8899: [.25001, .75, 0, 0, .83334], 8968: [.35001, .85, 0, 0, .47222], 8969: [.35001, .85, 0, 0, .47222], 8970: [.35001, .85, 0, 0, .47222], 8971: [.35001, .85, 0, 0, .47222], 9168: [-99e-5, .601, 0, 0, .66667], 10216: [.35001, .85, 0, 0, .47222], 10217: [.35001, .85, 0, 0, .47222], 10752: [.25001, .75, 0, 0, 1.11111], 10753: [.25001, .75, 0, 0, 1.11111], 10754: [.25001, .75, 0, 0, 1.11111], 10756: [.25001, .75, 0, 0, .83334], 10758: [.25001, .75, 0, 0, .83334] }, "Size2-Regular": { 32: [0, 0, 0, 0, .25], 40: [.65002, 1.15, 0, 0, .59722], 41: [.65002, 1.15, 0, 0, .59722], 47: [.65002, 1.15, 0, 0, .81111], 91: [.65002, 1.15, 0, 0, .47222], 92: [.65002, 1.15, 0, 0, .81111], 93: [.65002, 1.15, 0, 0, .47222], 123: [.65002, 1.15, 0, 0, .66667], 125: [.65002, 1.15, 0, 0, .66667], 160: [0, 0, 0, 0, .25], 710: [0, .75, 0, 0, 1], 732: [0, .75, 0, 0, 1], 770: [0, .75, 0, 0, 1], 771: [0, .75, 0, 0, 1], 8719: [.55001, 1.05, 0, 0, 1.27778], 8720: [.55001, 1.05, 0, 0, 1.27778], 8721: [.55001, 1.05, 0, 0, 1.44445], 8730: [.65002, 1.15, 0, 0, 1], 8747: [.86225, 1.36, .44445, 0, .55556], 8748: [.862, 1.36, .44445, 0, .55556], 8749: [.862, 1.36, .44445, 0, .55556], 8750: [.86225, 1.36, .44445, 0, .55556], 8896: [.55001, 1.05, 0, 0, 1.11111], 8897: [.55001, 1.05, 0, 0, 1.11111], 8898: [.55001, 1.05, 0, 0, 1.11111], 8899: [.55001, 1.05, 0, 0, 1.11111], 8968: [.65002, 1.15, 0, 0, .52778], 8969: [.65002, 1.15, 0, 0, .52778], 8970: [.65002, 1.15, 0, 0, .52778], 8971: [.65002, 1.15, 0, 0, .52778], 10216: [.65002, 1.15, 0, 0, .61111], 10217: [.65002, 1.15, 0, 0, .61111], 10752: [.55001, 1.05, 0, 0, 1.51112], 10753: [.55001, 1.05, 0, 0, 1.51112], 10754: [.55001, 1.05, 0, 0, 1.51112], 10756: [.55001, 1.05, 0, 0, 1.11111], 10758: [.55001, 1.05, 0, 0, 1.11111] }, "Size3-Regular": { 32: [0, 0, 0, 0, .25], 40: [.95003, 1.45, 0, 0, .73611], 41: [.95003, 1.45, 0, 0, .73611], 47: [.95003, 1.45, 0, 0, 1.04445], 91: [.95003, 1.45, 0, 0, .52778], 92: [.95003, 1.45, 0, 0, 1.04445], 93: [.95003, 1.45, 0, 0, .52778], 123: [.95003, 1.45, 0, 0, .75], 125: [.95003, 1.45, 0, 0, .75], 160: [0, 0, 0, 0, .25], 710: [0, .75, 0, 0, 1.44445], 732: [0, .75, 0, 0, 1.44445], 770: [0, .75, 0, 0, 1.44445], 771: [0, .75, 0, 0, 1.44445], 8730: [.95003, 1.45, 0, 0, 1], 8968: [.95003, 1.45, 0, 0, .58334], 8969: [.95003, 1.45, 0, 0, .58334], 8970: [.95003, 1.45, 0, 0, .58334], 8971: [.95003, 1.45, 0, 0, .58334], 10216: [.95003, 1.45, 0, 0, .75], 10217: [.95003, 1.45, 0, 0, .75] }, "Size4-Regular": { 32: [0, 0, 0, 0, .25], 40: [1.25003, 1.75, 0, 0, .79167], 41: [1.25003, 1.75, 0, 0, .79167], 47: [1.25003, 1.75, 0, 0, 1.27778], 91: [1.25003, 1.75, 0, 0, .58334], 92: [1.25003, 1.75, 0, 0, 1.27778], 93: [1.25003, 1.75, 0, 0, .58334], 123: [1.25003, 1.75, 0, 0, .80556], 125: [1.25003, 1.75, 0, 0, .80556], 160: [0, 0, 0, 0, .25], 710: [0, .825, 0, 0, 1.8889], 732: [0, .825, 0, 0, 1.8889], 770: [0, .825, 0, 0, 1.8889], 771: [0, .825, 0, 0, 1.8889], 8730: [1.25003, 1.75, 0, 0, 1], 8968: [1.25003, 1.75, 0, 0, .63889], 8969: [1.25003, 1.75, 0, 0, .63889], 8970: [1.25003, 1.75, 0, 0, .63889], 8971: [1.25003, 1.75, 0, 0, .63889], 9115: [.64502, 1.155, 0, 0, .875], 9116: [1e-5, .6, 0, 0, .875], 9117: [.64502, 1.155, 0, 0, .875], 9118: [.64502, 1.155, 0, 0, .875], 9119: [1e-5, .6, 0, 0, .875], 9120: [.64502, 1.155, 0, 0, .875], 9121: [.64502, 1.155, 0, 0, .66667], 9122: [-99e-5, .601, 0, 0, .66667], 9123: [.64502, 1.155, 0, 0, .66667], 9124: [.64502, 1.155, 0, 0, .66667], 9125: [-99e-5, .601, 0, 0, .66667], 9126: [.64502, 1.155, 0, 0, .66667], 9127: [1e-5, .9, 0, 0, .88889], 9128: [.65002, 1.15, 0, 0, .88889], 9129: [.90001, 0, 0, 0, .88889], 9130: [0, .3, 0, 0, .88889], 9131: [1e-5, .9, 0, 0, .88889], 9132: [.65002, 1.15, 0, 0, .88889], 9133: [.90001, 0, 0, 0, .88889], 9143: [.88502, .915, 0, 0, 1.05556], 10216: [1.25003, 1.75, 0, 0, .80556], 10217: [1.25003, 1.75, 0, 0, .80556], 57344: [-.00499, .605, 0, 0, 1.05556], 57345: [-.00499, .605, 0, 0, 1.05556], 57680: [0, .12, 0, 0, .45], 57681: [0, .12, 0, 0, .45], 57682: [0, .12, 0, 0, .45], 57683: [0, .12, 0, 0, .45] }, "Typewriter-Regular": { 32: [0, 0, 0, 0, .525], 33: [0, .61111, 0, 0, .525], 34: [0, .61111, 0, 0, .525], 35: [0, .61111, 0, 0, .525], 36: [.08333, .69444, 0, 0, .525], 37: [.08333, .69444, 0, 0, .525], 38: [0, .61111, 0, 0, .525], 39: [0, .61111, 0, 0, .525], 40: [.08333, .69444, 0, 0, .525], 41: [.08333, .69444, 0, 0, .525], 42: [0, .52083, 0, 0, .525], 43: [-.08056, .53055, 0, 0, .525], 44: [.13889, .125, 0, 0, .525], 45: [-.08056, .53055, 0, 0, .525], 46: [0, .125, 0, 0, .525], 47: [.08333, .69444, 0, 0, .525], 48: [0, .61111, 0, 0, .525], 49: [0, .61111, 0, 0, .525], 50: [0, .61111, 0, 0, .525], 51: [0, .61111, 0, 0, .525], 52: [0, .61111, 0, 0, .525], 53: [0, .61111, 0, 0, .525], 54: [0, .61111, 0, 0, .525], 55: [0, .61111, 0, 0, .525], 56: [0, .61111, 0, 0, .525], 57: [0, .61111, 0, 0, .525], 58: [0, .43056, 0, 0, .525], 59: [.13889, .43056, 0, 0, .525], 60: [-.05556, .55556, 0, 0, .525], 61: [-.19549, .41562, 0, 0, .525], 62: [-.05556, .55556, 0, 0, .525], 63: [0, .61111, 0, 0, .525], 64: [0, .61111, 0, 0, .525], 65: [0, .61111, 0, 0, .525], 66: [0, .61111, 0, 0, .525], 67: [0, .61111, 0, 0, .525], 68: [0, .61111, 0, 0, .525], 69: [0, .61111, 0, 0, .525], 70: [0, .61111, 0, 0, .525], 71: [0, .61111, 0, 0, .525], 72: [0, .61111, 0, 0, .525], 73: [0, .61111, 0, 0, .525], 74: [0, .61111, 0, 0, .525], 75: [0, .61111, 0, 0, .525], 76: [0, .61111, 0, 0, .525], 77: [0, .61111, 0, 0, .525], 78: [0, .61111, 0, 0, .525], 79: [0, .61111, 0, 0, .525], 80: [0, .61111, 0, 0, .525], 81: [.13889, .61111, 0, 0, .525], 82: [0, .61111, 0, 0, .525], 83: [0, .61111, 0, 0, .525], 84: [0, .61111, 0, 0, .525], 85: [0, .61111, 0, 0, .525], 86: [0, .61111, 0, 0, .525], 87: [0, .61111, 0, 0, .525], 88: [0, .61111, 0, 0, .525], 89: [0, .61111, 0, 0, .525], 90: [0, .61111, 0, 0, .525], 91: [.08333, .69444, 0, 0, .525], 92: [.08333, .69444, 0, 0, .525], 93: [.08333, .69444, 0, 0, .525], 94: [0, .61111, 0, 0, .525], 95: [.09514, 0, 0, 0, .525], 96: [0, .61111, 0, 0, .525], 97: [0, .43056, 0, 0, .525], 98: [0, .61111, 0, 0, .525], 99: [0, .43056, 0, 0, .525], 100: [0, .61111, 0, 0, .525], 101: [0, .43056, 0, 0, .525], 102: [0, .61111, 0, 0, .525], 103: [.22222, .43056, 0, 0, .525], 104: [0, .61111, 0, 0, .525], 105: [0, .61111, 0, 0, .525], 106: [.22222, .61111, 0, 0, .525], 107: [0, .61111, 0, 0, .525], 108: [0, .61111, 0, 0, .525], 109: [0, .43056, 0, 0, .525], 110: [0, .43056, 0, 0, .525], 111: [0, .43056, 0, 0, .525], 112: [.22222, .43056, 0, 0, .525], 113: [.22222, .43056, 0, 0, .525], 114: [0, .43056, 0, 0, .525], 115: [0, .43056, 0, 0, .525], 116: [0, .55358, 0, 0, .525], 117: [0, .43056, 0, 0, .525], 118: [0, .43056, 0, 0, .525], 119: [0, .43056, 0, 0, .525], 120: [0, .43056, 0, 0, .525], 121: [.22222, .43056, 0, 0, .525], 122: [0, .43056, 0, 0, .525], 123: [.08333, .69444, 0, 0, .525], 124: [.08333, .69444, 0, 0, .525], 125: [.08333, .69444, 0, 0, .525], 126: [0, .61111, 0, 0, .525], 127: [0, .61111, 0, 0, .525], 160: [0, 0, 0, 0, .525], 176: [0, .61111, 0, 0, .525], 184: [.19445, 0, 0, 0, .525], 305: [0, .43056, 0, 0, .525], 567: [.22222, .43056, 0, 0, .525], 711: [0, .56597, 0, 0, .525], 713: [0, .56555, 0, 0, .525], 714: [0, .61111, 0, 0, .525], 715: [0, .61111, 0, 0, .525], 728: [0, .61111, 0, 0, .525], 730: [0, .61111, 0, 0, .525], 770: [0, .61111, 0, 0, .525], 771: [0, .61111, 0, 0, .525], 776: [0, .61111, 0, 0, .525], 915: [0, .61111, 0, 0, .525], 916: [0, .61111, 0, 0, .525], 920: [0, .61111, 0, 0, .525], 923: [0, .61111, 0, 0, .525], 926: [0, .61111, 0, 0, .525], 928: [0, .61111, 0, 0, .525], 931: [0, .61111, 0, 0, .525], 933: [0, .61111, 0, 0, .525], 934: [0, .61111, 0, 0, .525], 936: [0, .61111, 0, 0, .525], 937: [0, .61111, 0, 0, .525], 8216: [0, .61111, 0, 0, .525], 8217: [0, .61111, 0, 0, .525], 8242: [0, .61111, 0, 0, .525], 9251: [.11111, .21944, 0, 0, .525] } }; const B = { slant: [.25, .25, .25], space: [0, 0, 0], stretch: [0, 0, 0], shrink: [0, 0, 0], xHeight: [.431, .431, .431], quad: [1, 1.171, 1.472], extraSpace: [0, 0, 0], num1: [.677, .732, .925], num2: [.394, .384, .387], num3: [.444, .471, .504], denom1: [.686, .752, 1.025], denom2: [.345, .344, .532], sup1: [.413, .503, .504], sup2: [.363, .431, .404], sup3: [.289, .286, .294], sub1: [.15, .143, .2], sub2: [.247, .286, .4], supDrop: [.386, .353, .494], subDrop: [.05, .071, .1], delim1: [2.39, 1.7, 1.98], delim2: [1.01, 1.157, 1.42], axisHeight: [.25, .25, .25], defaultRuleThickness: [.04, .049, .049], bigOpSpacing1: [.111, .111, .111], bigOpSpacing2: [.166, .166, .166], bigOpSpacing3: [.2, .2, .2], bigOpSpacing4: [.6, .611, .611], bigOpSpacing5: [.1, .143, .143], sqrtRuleThickness: [.04, .04, .04], ptPerEm: [10, 10, 10], doubleRuleSep: [.2, .2, .2], arrayRuleWidth: [.04, .04, .04], fboxsep: [.3, .3, .3], fboxrule: [.04, .04, .04] }, C = { "\xc5": "A", "\xd0": "D", "\xde": "o", "\xe5": "a", "\xf0": "d", "\xfe": "o", "\u0410": "A", "\u0411": "B", "\u0412": "B", "\u0413": "F", "\u0414": "A", "\u0415": "E", "\u0416": "K", "\u0417": "3", "\u0418": "N", "\u0419": "N", "\u041a": "K", "\u041b": "N", "\u041c": "M", "\u041d": "H", "\u041e": "O", "\u041f": "N", "\u0420": "P", "\u0421": "C", "\u0422": "T", "\u0423": "y", "\u0424": "O", "\u0425": "X", "\u0426": "U", "\u0427": "h", "\u0428": "W", "\u0429": "W", "\u042a": "B", "\u042b": "X", "\u042c": "B", "\u042d": "3", "\u042e": "X", "\u042f": "R", "\u0430": "a", "\u0431": "b", "\u0432": "a", "\u0433": "r", "\u0434": "y", "\u0435": "e", "\u0436": "m", "\u0437": "e", "\u0438": "n", "\u0439": "n", "\u043a": "n", "\u043b": "n", "\u043c": "m", "\u043d": "n", "\u043e": "o", "\u043f": "n", "\u0440": "p", "\u0441": "c", "\u0442": "o", "\u0443": "y", "\u0444": "b", "\u0445": "x", "\u0446": "n", "\u0447": "n", "\u0448": "w", "\u0449": "w", "\u044a": "a", "\u044b": "m", "\u044c": "a", "\u044d": "e", "\u044e": "m", "\u044f": "r" }; function N(e, t, r) { if (!T[t]) throw new Error("Font metrics not found for font: " + t + "."); let n = e.charCodeAt(0), o = T[t][n]; if (!o && e[0] in C && (n = C[e[0]].charCodeAt(0), o = T[t][n]), o || "text" !== r || S(n) && (o = T[t][77]), o) return { depth: o[0], height: o[1], italic: o[2], skew: o[3], width: o[4] } } const q = {}; const I = [[1, 1, 1], [2, 1, 1], [3, 1, 1], [4, 2, 1], [5, 2, 1], [6, 3, 1], [7, 4, 2], [8, 6, 3], [9, 7, 6], [10, 8, 7], [11, 10, 9]], R = [.5, .6, .7, .8, .9, 1, 1.2, 1.44, 1.728, 2.074, 2.488], H = function (e, t) { return t.size < 2 ? e : I[e - 1][t.size - 1] }; class O { constructor(e) { this.style = void 0, this.color = void 0, this.size = void 0, this.textSize = void 0, this.phantom = void 0, this.font = void 0, this.fontFamily = void 0, this.fontWeight = void 0, this.fontShape = void 0, this.sizeMultiplier = void 0, this.maxSize = void 0, this.minRuleThickness = void 0, this._fontMetrics = void 0, this.style = e.style, this.color = e.color, this.size = e.size || O.BASESIZE, this.textSize = e.textSize || this.size, this.phantom = !!e.phantom, this.font = e.font || "", this.fontFamily = e.fontFamily || "", this.fontWeight = e.fontWeight || "", this.fontShape = e.fontShape || "", this.sizeMultiplier = R[this.size - 1], this.maxSize = e.maxSize, this.minRuleThickness = e.minRuleThickness, this._fontMetrics = void 0 } extend(e) { const t = { style: this.style, size: this.size, textSize: this.textSize, color: this.color, phantom: this.phantom, font: this.font, fontFamily: this.fontFamily, fontWeight: this.fontWeight, fontShape: this.fontShape, maxSize: this.maxSize, minRuleThickness: this.minRuleThickness }; for (const r in e) e.hasOwnProperty(r) && (t[r] = e[r]); return new O(t) } havingStyle(e) { return this.style === e ? this : this.extend({ style: e, size: H(this.textSize, e) }) } havingCrampedStyle() { return this.havingStyle(this.style.cramp()) } havingSize(e) { return this.size === e && this.textSize === e ? this : this.extend({ style: this.style.text(), size: e, textSize: e, sizeMultiplier: R[e - 1] }) } havingBaseStyle(e) { e = e || this.style.text(); const t = H(O.BASESIZE, e); return this.size === t && this.textSize === O.BASESIZE && this.style === e ? this : this.extend({ style: e, size: t }) } havingBaseSizing() { let e; switch (this.style.id) { case 4: case 5: e = 3; break; case 6: case 7: e = 1; break; default: e = 6 }return this.extend({ style: this.style.text(), size: e }) } withColor(e) { return this.extend({ color: e }) } withPhantom() { return this.extend({ phantom: !0 }) } withFont(e) { return this.extend({ font: e }) } withTextFontFamily(e) { return this.extend({ fontFamily: e, font: "" }) } withTextFontWeight(e) { return this.extend({ fontWeight: e, font: "" }) } withTextFontShape(e) { return this.extend({ fontShape: e, font: "" }) } sizingClasses(e) { return e.size !== this.size ? ["sizing", "reset-size" + e.size, "size" + this.size] : [] } baseSizingClasses() { return this.size !== O.BASESIZE ? ["sizing", "reset-size" + this.size, "size" + O.BASESIZE] : [] } fontMetrics() { return this._fontMetrics || (this._fontMetrics = function (e) { let t; if (t = e >= 5 ? 0 : e >= 3 ? 1 : 2, !q[t]) { const e = q[t] = { cssEmPerMu: B.quad[t] / 18 }; for (const r in B) B.hasOwnProperty(r) && (e[r] = B[r][t]) } return q[t] }(this.size)), this._fontMetrics } getColor() { return this.phantom ? "transparent" : this.color } } O.BASESIZE = 6; var E = O; const L = { pt: 1, mm: 7227 / 2540, cm: 7227 / 254, in: 72.27, bp: 1.00375, pc: 12, dd: 1238 / 1157, cc: 14856 / 1157, nd: 685 / 642, nc: 1370 / 107, sp: 1 / 65536, px: 1.00375 }, D = { ex: !0, em: !0, mu: !0 }, V = function (e) { return "string" != typeof e && (e = e.unit), e in L || e in D || "ex" === e }, P = function (e, t) { let r; if (e.unit in L) r = L[e.unit] / t.fontMetrics().ptPerEm / t.sizeMultiplier; else if ("mu" === e.unit) r = t.fontMetrics().cssEmPerMu; else { let o; if (o = t.style.isTight() ? t.havingStyle(t.style.text()) : t, "ex" === e.unit) r = o.fontMetrics().xHeight; else { if ("em" !== e.unit) throw new n("Invalid unit: '" + e.unit + "'"); r = o.fontMetrics().quad } o !== t && (r *= o.sizeMultiplier / t.sizeMultiplier) } return Math.min(e.number * r, t.maxSize) }, F = function (e) { return +e.toFixed(4) + "em" }, G = function (e) { return e.filter((e => e)).join(" ") }, U = function (e, t, r) { if (this.classes = e || [], this.attributes = {}, this.height = 0, this.depth = 0, this.maxFontSize = 0, this.style = r || {}, t) { t.style.isTight() && this.classes.push("mtight"); const e = t.getColor(); e && (this.style.color = e) } }, Y = function (e) { const t = document.createElement(e); t.className = G(this.classes); for (const e in this.style) this.style.hasOwnProperty(e) && (t.style[e] = this.style[e]); for (const e in this.attributes) this.attributes.hasOwnProperty(e) && t.setAttribute(e, this.attributes[e]); for (let e = 0; e < this.children.length; e++)t.appendChild(this.children[e].toNode()); return t }, X = function (e) { let t = "<" + e; this.classes.length && (t += ' class="' + l.escape(G(this.classes)) + '"'); let r = ""; for (const e in this.style) this.style.hasOwnProperty(e) && (r += l.hyphenate(e) + ":" + this.style[e] + ";"); r && (t += ' style="' + l.escape(r) + '"'); for (const e in this.attributes) this.attributes.hasOwnProperty(e) && (t += " " + e + '="' + l.escape(this.attributes[e]) + '"'); t += ">"; for (let e = 0; e < this.children.length; e++)t += this.children[e].toMarkup(); return t += "</" + e + ">", t }; class W { constructor(e, t, r, n) { this.children = void 0, this.attributes = void 0, this.classes = void 0, this.height = void 0, this.depth = void 0, this.width = void 0, this.maxFontSize = void 0, this.style = void 0, U.call(this, e, r, n), this.children = t || [] } setAttribute(e, t) { this.attributes[e] = t } hasClass(e) { return l.contains(this.classes, e) } toNode() { return Y.call(this, "span") } toMarkup() { return X.call(this, "span") } } class _ { constructor(e, t, r, n) { this.children = void 0, this.attributes = void 0, this.classes = void 0, this.height = void 0, this.depth = void 0, this.maxFontSize = void 0, this.style = void 0, U.call(this, t, n), this.children = r || [], this.setAttribute("href", e) } setAttribute(e, t) { this.attributes[e] = t } hasClass(e) { return l.contains(this.classes, e) } toNode() { return Y.call(this, "a") } toMarkup() { return X.call(this, "a") } } class j { constructor(e, t, r) { this.src = void 0, this.alt = void 0, this.classes = void 0, this.height = void 0, this.depth = void 0, this.maxFontSize = void 0, this.style = void 0, this.alt = t, this.src = e, this.classes = ["mord"], this.style = r } hasClass(e) { return l.contains(this.classes, e) } toNode() { const e = document.createElement("img"); e.src = this.src, e.alt = this.alt, e.className = "mord"; for (const t in this.style) this.style.hasOwnProperty(t) && (e.style[t] = this.style[t]); return e } toMarkup() { let e = '<img src="' + l.escape(this.src) + '" alt="' + l.escape(this.alt) + '"', t = ""; for (const e in this.style) this.style.hasOwnProperty(e) && (t += l.hyphenate(e) + ":" + this.style[e] + ";"); return t && (e += ' style="' + l.escape(t) + '"'), e += "'/>", e } } const $ = { "\xee": "\u0131\u0302", "\xef": "\u0131\u0308", "\xed": "\u0131\u0301", "\xec": "\u0131\u0300" }; class Z { constructor(e, t, r, n, o, s, i, a) { this.text = void 0, this.height = void 0, this.depth = void 0, this.italic = void 0, this.skew = void 0, this.width = void 0, this.maxFontSize = void 0, this.classes = void 0, this.style = void 0, this.text = e, this.height = t || 0, this.depth = r || 0, this.italic = n || 0, this.skew = o || 0, this.width = s || 0, this.classes = i || [], this.style = a || {}, this.maxFontSize = 0; const l = function (e) { for (let t = 0; t < v.length; t++) { const r = v[t]; for (let t = 0; t < r.blocks.length; t++) { const n = r.blocks[t]; if (e >= n[0] && e <= n[1]) return r.name } } return null }(this.text.charCodeAt(0)); l && this.classes.push(l + "_fallback"), /[\xee\xef\xed\xec]/.test(this.text) && (this.text = $[this.text]) } hasClass(e) { return l.contains(this.classes, e) } toNode() { const e = document.createTextNode(this.text); let t = null; this.italic > 0 && (t = document.createElement("span"), t.style.marginRight = F(this.italic)), this.classes.length > 0 && (t = t || document.createElement("span"), t.className = G(this.classes)); for (const e in this.style) this.style.hasOwnProperty(e) && (t = t || document.createElement("span"), t.style[e] = this.style[e]); return t ? (t.appendChild(e), t) : e } toMarkup() { let e = !1, t = "<span"; this.classes.length && (e = !0, t += ' class="', t += l.escape(G(this.classes)), t += '"'); let r = ""; this.italic > 0 && (r += "margin-right:" + this.italic + "em;"); for (const e in this.style) this.style.hasOwnProperty(e) && (r += l.hyphenate(e) + ":" + this.style[e] + ";"); r && (e = !0, t += ' style="' + l.escape(r) + '"'); const n = l.escape(this.text); return e ? (t += ">", t += n, t += "</span>", t) : n } } class K { constructor(e, t) { this.children = void 0, this.attributes = void 0, this.children = e || [], this.attributes = t || {} } toNode() { const e = document.createElementNS("http://www.w3.org/2000/svg", "svg"); for (const t in this.attributes) Object.prototype.hasOwnProperty.call(this.attributes, t) && e.setAttribute(t, this.attributes[t]); for (let t = 0; t < this.children.length; t++)e.appendChild(this.children[t].toNode()); return e } toMarkup() { let e = '<svg xmlns="http://www.w3.org/2000/svg"'; for (const t in this.attributes) Object.prototype.hasOwnProperty.call(this.attributes, t) && (e += " " + t + '="' + l.escape(this.attributes[t]) + '"'); e += ">"; for (let t = 0; t < this.children.length; t++)e += this.children[t].toMarkup(); return e += "</svg>", e } } class J { constructor(e, t) { this.pathName = void 0, this.alternate = void 0, this.pathName = e, this.alternate = t } toNode() { const e = document.createElementNS("http://www.w3.org/2000/svg", "path"); return this.alternate ? e.setAttribute("d", this.alternate) : e.setAttribute("d", z[this.pathName]), e } toMarkup() { return this.alternate ? '<path d="' + l.escape(this.alternate) + '"/>' : '<path d="' + l.escape(z[this.pathName]) + '"/>' } } class Q { constructor(e) { this.attributes = void 0, this.attributes = e || {} } toNode() { const e = document.createElementNS("http://www.w3.org/2000/svg", "line"); for (const t in this.attributes) Object.prototype.hasOwnProperty.call(this.attributes, t) && e.setAttribute(t, this.attributes[t]); return e } toMarkup() { let e = "<line"; for (const t in this.attributes) Object.prototype.hasOwnProperty.call(this.attributes, t) && (e += " " + t + '="' + l.escape(this.attributes[t]) + '"'); return e += "/>", e } } function ee(e) { if (e instanceof Z) return e; throw new Error("Expected symbolNode but got " + String(e) + ".") } const te = { bin: 1, close: 1, inner: 1, open: 1, punct: 1, rel: 1 }, re = { "accent-token": 1, mathord: 1, "op-token": 1, spacing: 1, textord: 1 }, ne = { math: {}, text: {} }; var oe = ne; function se(e, t, r, n, o, s) { ne[e][o] = { font: t, group: r, replace: n }, s && n && (ne[e][n] = ne[e][o]) } const ie = "math", ae = "text", le = "main", he = "ams", ce = "accent-token", me = "bin", pe = "close", ue = "inner", de = "mathord", ge = "op-token", fe = "open", be = "punct", ye = "rel", xe = "spacing", we = "textord"; se(ie, le, ye, "\u2261", "\\equiv", !0), se(ie, le, ye, "\u227a", "\\prec", !0), se(ie, le, ye, "\u227b", "\\succ", !0), se(ie, le, ye, "\u223c", "\\sim", !0), se(ie, le, ye, "\u22a5", "\\perp"), se(ie, le, ye, "\u2aaf", "\\preceq", !0), se(ie, le, ye, "\u2ab0", "\\succeq", !0), se(ie, le, ye, "\u2243", "\\simeq", !0), se(ie, le, ye, "\u2223", "\\mid", !0), se(ie, le, ye, "\u226a", "\\ll", !0), se(ie, le, ye, "\u226b", "\\gg", !0), se(ie, le, ye, "\u224d", "\\asymp", !0), se(ie, le, ye, "\u2225", "\\parallel"), se(ie, le, ye, "\u22c8", "\\bowtie", !0), se(ie, le, ye, "\u2323", "\\smile", !0), se(ie, le, ye, "\u2291", "\\sqsubseteq", !0), se(ie, le, ye, "\u2292", "\\sqsupseteq", !0), se(ie, le, ye, "\u2250", "\\doteq", !0), se(ie, le, ye, "\u2322", "\\frown", !0), se(ie, le, ye, "\u220b", "\\ni", !0), se(ie, le, ye, "\u221d", "\\propto", !0), se(ie, le, ye, "\u22a2", "\\vdash", !0), se(ie, le, ye, "\u22a3", "\\dashv", !0), se(ie, le, ye, "\u220b", "\\owns"), se(ie, le, be, ".", "\\ldotp"), se(ie, le, be, "\u22c5", "\\cdotp"), se(ie, le, we, "#", "\\#"), se(ae, le, we, "#", "\\#"), se(ie, le, we, "&", "\\&"), se(ae, le, we, "&", "\\&"), se(ie, le, we, "\u2135", "\\aleph", !0), se(ie, le, we, "\u2200", "\\forall", !0), se(ie, le, we, "\u210f", "\\hbar", !0), se(ie, le, we, "\u2203", "\\exists", !0), se(ie, le, we, "\u2207", "\\nabla", !0), se(ie, le, we, "\u266d", "\\flat", !0), se(ie, le, we, "\u2113", "\\ell", !0), se(ie, le, we, "\u266e", "\\natural", !0), se(ie, le, we, "\u2663", "\\clubsuit", !0), se(ie, le, we, "\u2118", "\\wp", !0), se(ie, le, we, "\u266f", "\\sharp", !0), se(ie, le, we, "\u2662", "\\diamondsuit", !0), se(ie, le, we, "\u211c", "\\Re", !0), se(ie, le, we, "\u2661", "\\heartsuit", !0), se(ie, le, we, "\u2111", "\\Im", !0), se(ie, le, we, "\u2660", "\\spadesuit", !0), se(ie, le, we, "\xa7", "\\S", !0), se(ae, le, we, "\xa7", "\\S"), se(ie, le, we, "\xb6", "\\P", !0), se(ae, le, we, "\xb6", "\\P"), se(ie, le, we, "\u2020", "\\dag"), se(ae, le, we, "\u2020", "\\dag"), se(ae, le, we, "\u2020", "\\textdagger"), se(ie, le, we, "\u2021", "\\ddag"), se(ae, le, we, "\u2021", "\\ddag"), se(ae, le, we, "\u2021", "\\textdaggerdbl"), se(ie, le, pe, "\u23b1", "\\rmoustache", !0), se(ie, le, fe, "\u23b0", "\\lmoustache", !0), se(ie, le, pe, "\u27ef", "\\rgroup", !0), se(ie, le, fe, "\u27ee", "\\lgroup", !0), se(ie, le, me, "\u2213", "\\mp", !0), se(ie, le, me, "\u2296", "\\ominus", !0), se(ie, le, me, "\u228e", "\\uplus", !0), se(ie, le, me, "\u2293", "\\sqcap", !0), se(ie, le, me, "\u2217", "\\ast"), se(ie, le, me, "\u2294", "\\sqcup", !0), se(ie, le, me, "\u25ef", "\\bigcirc", !0), se(ie, le, me, "\u2219", "\\bullet", !0), se(ie, le, me, "\u2021", "\\ddagger"), se(ie, le, me, "\u2240", "\\wr", !0), se(ie, le, me, "\u2a3f", "\\amalg"), se(ie, le, me, "&", "\\And"), se(ie, le, ye, "\u27f5", "\\longleftarrow", !0), se(ie, le, ye, "\u21d0", "\\Leftarrow", !0), se(ie, le, ye, "\u27f8", "\\Longleftarrow", !0), se(ie, le, ye, "\u27f6", "\\longrightarrow", !0), se(ie, le, ye, "\u21d2", "\\Rightarrow", !0), se(ie, le, ye, "\u27f9", "\\Longrightarrow", !0), se(ie, le, ye, "\u2194", "\\leftrightarrow", !0), se(ie, le, ye, "\u27f7", "\\longleftrightarrow", !0), se(ie, le, ye, "\u21d4", "\\Leftrightarrow", !0), se(ie, le, ye, "\u27fa", "\\Longleftrightarrow", !0), se(ie, le, ye, "\u21a6", "\\mapsto", !0), se(ie, le, ye, "\u27fc", "\\longmapsto", !0), se(ie, le, ye, "\u2197", "\\nearrow", !0), se(ie, le, ye, "\u21a9", "\\hookleftarrow", !0), se(ie, le, ye, "\u21aa", "\\hookrightarrow", !0), se(ie, le, ye, "\u2198", "\\searrow", !0), se(ie, le, ye, "\u21bc", "\\leftharpoonup", !0), se(ie, le, ye, "\u21c0", "\\rightharpoonup", !0), se(ie, le, ye, "\u2199", "\\swarrow", !0), se(ie, le, ye, "\u21bd", "\\leftharpoondown", !0), se(ie, le, ye, "\u21c1", "\\rightharpoondown", !0), se(ie, le, ye, "\u2196", "\\nwarrow", !0), se(ie, le, ye, "\u21cc", "\\rightleftharpoons", !0), se(ie, he, ye, "\u226e", "\\nless", !0), se(ie, he, ye, "\ue010", "\\@nleqslant"), se(ie, he, ye, "\ue011", "\\@nleqq"), se(ie, he, ye, "\u2a87", "\\lneq", !0), se(ie, he, ye, "\u2268", "\\lneqq", !0), se(ie, he, ye, "\ue00c", "\\@lvertneqq"), se(ie, he, ye, "\u22e6", "\\lnsim", !0), se(ie, he, ye, "\u2a89", "\\lnapprox", !0), se(ie, he, ye, "\u2280", "\\nprec", !0), se(ie, he, ye, "\u22e0", "\\npreceq", !0), se(ie, he, ye, "\u22e8", "\\precnsim", !0), se(ie, he, ye, "\u2ab9", "\\precnapprox", !0), se(ie, he, ye, "\u2241", "\\nsim", !0), se(ie, he, ye, "\ue006", "\\@nshortmid"), se(ie, he, ye, "\u2224", "\\nmid", !0), se(ie, he, ye, "\u22ac", "\\nvdash", !0), se(ie, he, ye, "\u22ad", "\\nvDash", !0), se(ie, he, ye, "\u22ea", "\\ntriangleleft"), se(ie, he, ye, "\u22ec", "\\ntrianglelefteq", !0), se(ie, he, ye, "\u228a", "\\subsetneq", !0), se(ie, he, ye, "\ue01a", "\\@varsubsetneq"), se(ie, he, ye, "\u2acb", "\\subsetneqq", !0), se(ie, he, ye, "\ue017", "\\@varsubsetneqq"), se(ie, he, ye, "\u226f", "\\ngtr", !0), se(ie, he, ye, "\ue00f", "\\@ngeqslant"), se(ie, he, ye, "\ue00e", "\\@ngeqq"), se(ie, he, ye, "\u2a88", "\\gneq", !0), se(ie, he, ye, "\u2269", "\\gneqq", !0), se(ie, he, ye, "\ue00d", "\\@gvertneqq"), se(ie, he, ye, "\u22e7", "\\gnsim", !0), se(ie, he, ye, "\u2a8a", "\\gnapprox", !0), se(ie, he, ye, "\u2281", "\\nsucc", !0), se(ie, he, ye, "\u22e1", "\\nsucceq", !0), se(ie, he, ye, "\u22e9", "\\succnsim", !0), se(ie, he, ye, "\u2aba", "\\succnapprox", !0), se(ie, he, ye, "\u2246", "\\ncong", !0), se(ie, he, ye, "\ue007", "\\@nshortparallel"), se(ie, he, ye, "\u2226", "\\nparallel", !0), se(ie, he, ye, "\u22af", "\\nVDash", !0), se(ie, he, ye, "\u22eb", "\\ntriangleright"), se(ie, he, ye, "\u22ed", "\\ntrianglerighteq", !0), se(ie, he, ye, "\ue018", "\\@nsupseteqq"), se(ie, he, ye, "\u228b", "\\supsetneq", !0), se(ie, he, ye, "\ue01b", "\\@varsupsetneq"), se(ie, he, ye, "\u2acc", "\\supsetneqq", !0), se(ie, he, ye, "\ue019", "\\@varsupsetneqq"), se(ie, he, ye, "\u22ae", "\\nVdash", !0), se(ie, he, ye, "\u2ab5", "\\precneqq", !0), se(ie, he, ye, "\u2ab6", "\\succneqq", !0), se(ie, he, ye, "\ue016", "\\@nsubseteqq"), se(ie, he, me, "\u22b4", "\\unlhd"), se(ie, he, me, "\u22b5", "\\unrhd"), se(ie, he, ye, "\u219a", "\\nleftarrow", !0), se(ie, he, ye, "\u219b", "\\nrightarrow", !0), se(ie, he, ye, "\u21cd", "\\nLeftarrow", !0), se(ie, he, ye, "\u21cf", "\\nRightarrow", !0), se(ie, he, ye, "\u21ae", "\\nleftrightarrow", !0), se(ie, he, ye, "\u21ce", "\\nLeftrightarrow", !0), se(ie, he, ye, "\u25b3", "\\vartriangle"), se(ie, he, we, "\u210f", "\\hslash"), se(ie, he, we, "\u25bd", "\\triangledown"), se(ie, he, we, "\u25ca", "\\lozenge"), se(ie, he, we, "\u24c8", "\\circledS"), se(ie, he, we, "\xae", "\\circledR"), se(ae, he, we, "\xae", "\\circledR"), se(ie, he, we, "\u2221", "\\measuredangle", !0), se(ie, he, we, "\u2204", "\\nexists"), se(ie, he, we, "\u2127", "\\mho"), se(ie, he, we, "\u2132", "\\Finv", !0), se(ie, he, we, "\u2141", "\\Game", !0), se(ie, he, we, "\u2035", "\\backprime"), se(ie, he, we, "\u25b2", "\\blacktriangle"), se(ie, he, we, "\u25bc", "\\blacktriangledown"), se(ie, he, we, "\u25a0", "\\blacksquare"), se(ie, he, we, "\u29eb", "\\blacklozenge"), se(ie, he, we, "\u2605", "\\bigstar"), se(ie, he, we, "\u2222", "\\sphericalangle", !0), se(ie, he, we, "\u2201", "\\complement", !0), se(ie, he, we, "\xf0", "\\eth", !0), se(ae, le, we, "\xf0", "\xf0"), se(ie, he, we, "\u2571", "\\diagup"), se(ie, he, we, "\u2572", "\\diagdown"), se(ie, he, we, "\u25a1", "\\square"), se(ie, he, we, "\u25a1", "\\Box"), se(ie, he, we, "\u25ca", "\\Diamond"), se(ie, he, we, "\xa5", "\\yen", !0), se(ae, he, we, "\xa5", "\\yen", !0), se(ie, he, we, "\u2713", "\\checkmark", !0), se(ae, he, we, "\u2713", "\\checkmark"), se(ie, he, we, "\u2136", "\\beth", !0), se(ie, he, we, "\u2138", "\\daleth", !0), se(ie, he, we, "\u2137", "\\gimel", !0), se(ie, he, we, "\u03dd", "\\digamma", !0), se(ie, he, we, "\u03f0", "\\varkappa"), se(ie, he, fe, "\u250c", "\\@ulcorner", !0), se(ie, he, pe, "\u2510", "\\@urcorner", !0), se(ie, he, fe, "\u2514", "\\@llcorner", !0), se(ie, he, pe, "\u2518", "\\@lrcorner", !0), se(ie, he, ye, "\u2266", "\\leqq", !0), se(ie, he, ye, "\u2a7d", "\\leqslant", !0), se(ie, he, ye, "\u2a95", "\\eqslantless", !0), se(ie, he, ye, "\u2272", "\\lesssim", !0), se(ie, he, ye, "\u2a85", "\\lessapprox", !0), se(ie, he, ye, "\u224a", "\\approxeq", !0), se(ie, he, me, "\u22d6", "\\lessdot"), se(ie, he, ye, "\u22d8", "\\lll", !0), se(ie, he, ye, "\u2276", "\\lessgtr", !0), se(ie, he, ye, "\u22da", "\\lesseqgtr", !0), se(ie, he, ye, "\u2a8b", "\\lesseqqgtr", !0), se(ie, he, ye, "\u2251", "\\doteqdot"), se(ie, he, ye, "\u2253", "\\risingdotseq", !0), se(ie, he, ye, "\u2252", "\\fallingdotseq", !0), se(ie, he, ye, "\u223d", "\\backsim", !0), se(ie, he, ye, "\u22cd", "\\backsimeq", !0), se(ie, he, ye, "\u2ac5", "\\subseteqq", !0), se(ie, he, ye, "\u22d0", "\\Subset", !0), se(ie, he, ye, "\u228f", "\\sqsubset", !0), se(ie, he, ye, "\u227c", "\\preccurlyeq", !0), se(ie, he, ye, "\u22de", "\\curlyeqprec", !0), se(ie, he, ye, "\u227e", "\\precsim", !0), se(ie, he, ye, "\u2ab7", "\\precapprox", !0), se(ie, he, ye, "\u22b2", "\\vartriangleleft"), se(ie, he, ye, "\u22b4", "\\trianglelefteq"), se(ie, he, ye, "\u22a8", "\\vDash", !0), se(ie, he, ye, "\u22aa", "\\Vvdash", !0), se(ie, he, ye, "\u2323", "\\smallsmile"), se(ie, he, ye, "\u2322", "\\smallfrown"), se(ie, he, ye, "\u224f", "\\bumpeq", !0), se(ie, he, ye, "\u224e", "\\Bumpeq", !0), se(ie, he, ye, "\u2267", "\\geqq", !0), se(ie, he, ye, "\u2a7e", "\\geqslant", !0), se(ie, he, ye, "\u2a96", "\\eqslantgtr", !0), se(ie, he, ye, "\u2273", "\\gtrsim", !0), se(ie, he, ye, "\u2a86", "\\gtrapprox", !0), se(ie, he, me, "\u22d7", "\\gtrdot"), se(ie, he, ye, "\u22d9", "\\ggg", !0), se(ie, he, ye, "\u2277", "\\gtrless", !0), se(ie, he, ye, "\u22db", "\\gtreqless", !0), se(ie, he, ye, "\u2a8c", "\\gtreqqless", !0), se(ie, he, ye, "\u2256", "\\eqcirc", !0), se(ie, he, ye, "\u2257", "\\circeq", !0), se(ie, he, ye, "\u225c", "\\triangleq", !0), se(ie, he, ye, "\u223c", "\\thicksim"), se(ie, he, ye, "\u2248", "\\thickapprox"), se(ie, he, ye, "\u2ac6", "\\supseteqq", !0), se(ie, he, ye, "\u22d1", "\\Supset", !0), se(ie, he, ye, "\u2290", "\\sqsupset", !0), se(ie, he, ye, "\u227d", "\\succcurlyeq", !0), se(ie, he, ye, "\u22df", "\\curlyeqsucc", !0), se(ie, he, ye, "\u227f", "\\succsim", !0), se(ie, he, ye, "\u2ab8", "\\succapprox", !0), se(ie, he, ye, "\u22b3", "\\vartriangleright"), se(ie, he, ye, "\u22b5", "\\trianglerighteq"), se(ie, he, ye, "\u22a9", "\\Vdash", !0), se(ie, he, ye, "\u2223", "\\shortmid"), se(ie, he, ye, "\u2225", "\\shortparallel"), se(ie, he, ye, "\u226c", "\\between", !0), se(ie, he, ye, "\u22d4", "\\pitchfork", !0), se(ie, he, ye, "\u221d", "\\varpropto"), se(ie, he, ye, "\u25c0", "\\blacktriangleleft"), se(ie, he, ye, "\u2234", "\\therefore", !0), se(ie, he, ye, "\u220d", "\\backepsilon"), se(ie, he, ye, "\u25b6", "\\blacktriangleright"), se(ie, he, ye, "\u2235", "\\because", !0), se(ie, he, ye, "\u22d8", "\\llless"), se(ie, he, ye, "\u22d9", "\\gggtr"), se(ie, he, me, "\u22b2", "\\lhd"), se(ie, he, me, "\u22b3", "\\rhd"), se(ie, he, ye, "\u2242", "\\eqsim", !0), se(ie, le, ye, "\u22c8", "\\Join"), se(ie, he, ye, "\u2251", "\\Doteq", !0), se(ie, he, me, "\u2214", "\\dotplus", !0), se(ie, he, me, "\u2216", "\\smallsetminus"), se(ie, he, me, "\u22d2", "\\Cap", !0), se(ie, he, me, "\u22d3", "\\Cup", !0), se(ie, he, me, "\u2a5e", "\\doublebarwedge", !0), se(ie, he, me, "\u229f", "\\boxminus", !0), se(ie, he, me, "\u229e", "\\boxplus", !0), se(ie, he, me, "\u22c7", "\\divideontimes", !0), se(ie, he, me, "\u22c9", "\\ltimes", !0), se(ie, he, me, "\u22ca", "\\rtimes", !0), se(ie, he, me, "\u22cb", "\\leftthreetimes", !0), se(ie, he, me, "\u22cc", "\\rightthreetimes", !0), se(ie, he, me, "\u22cf", "\\curlywedge", !0), se(ie, he, me, "\u22ce", "\\curlyvee", !0), se(ie, he, me, "\u229d", "\\circleddash", !0), se(ie, he, me, "\u229b", "\\circledast", !0), se(ie, he, me, "\u22c5", "\\centerdot"), se(ie, he, me, "\u22ba", "\\intercal", !0), se(ie, he, me, "\u22d2", "\\doublecap"), se(ie, he, me, "\u22d3", "\\doublecup"), se(ie, he, me, "\u22a0", "\\boxtimes", !0), se(ie, he, ye, "\u21e2", "\\dashrightarrow", !0), se(ie, he, ye, "\u21e0", "\\dashleftarrow", !0), se(ie, he, ye, "\u21c7", "\\leftleftarrows", !0), se(ie, he, ye, "\u21c6", "\\leftrightarrows", !0), se(ie, he, ye, "\u21da", "\\Lleftarrow", !0), se(ie, he, ye, "\u219e", "\\twoheadleftarrow", !0), se(ie, he, ye, "\u21a2", "\\leftarrowtail", !0), se(ie, he, ye, "\u21ab", "\\looparrowleft", !0), se(ie, he, ye, "\u21cb", "\\leftrightharpoons", !0), se(ie, he, ye, "\u21b6", "\\curvearrowleft", !0), se(ie, he, ye, "\u21ba", "\\circlearrowleft", !0), se(ie, he, ye, "\u21b0", "\\Lsh", !0), se(ie, he, ye, "\u21c8", "\\upuparrows", !0), se(ie, he, ye, "\u21bf", "\\upharpoonleft", !0), se(ie, he, ye, "\u21c3", "\\downharpoonleft", !0), se(ie, le, ye, "\u22b6", "\\origof", !0), se(ie, le, ye, "\u22b7", "\\imageof", !0), se(ie, he, ye, "\u22b8", "\\multimap", !0), se(ie, he, ye, "\u21ad", "\\leftrightsquigarrow", !0), se(ie, he, ye, "\u21c9", "\\rightrightarrows", !0), se(ie, he, ye, "\u21c4", "\\rightleftarrows", !0), se(ie, he, ye, "\u21a0", "\\twoheadrightarrow", !0), se(ie, he, ye, "\u21a3", "\\rightarrowtail", !0), se(ie, he, ye, "\u21ac", "\\looparrowright", !0), se(ie, he, ye, "\u21b7", "\\curvearrowright", !0), se(ie, he, ye, "\u21bb", "\\circlearrowright", !0), se(ie, he, ye, "\u21b1", "\\Rsh", !0), se(ie, he, ye, "\u21ca", "\\downdownarrows", !0), se(ie, he, ye, "\u21be", "\\upharpoonright", !0), se(ie, he, ye, "\u21c2", "\\downharpoonright", !0), se(ie, he, ye, "\u21dd", "\\rightsquigarrow", !0), se(ie, he, ye, "\u21dd", "\\leadsto"), se(ie, he, ye, "\u21db", "\\Rrightarrow", !0), se(ie, he, ye, "\u21be", "\\restriction"), se(ie, le, we, "\u2018", "`"), se(ie, le, we, "$", "\\$"), se(ae, le, we, "$", "\\$"), se(ae, le, we, "$", "\\textdollar"), se(ie, le, we, "%", "\\%"), se(ae, le, we, "%", "\\%"), se(ie, le, we, "_", "\\_"), se(ae, le, we, "_", "\\_"), se(ae, le, we, "_", "\\textunderscore"), se(ie, le, we, "\u2220", "\\angle", !0), se(ie, le, we, "\u221e", "\\infty", !0), se(ie, le, we, "\u2032", "\\prime"), se(ie, le, we, "\u25b3", "\\triangle"), se(ie, le, we, "\u0393", "\\Gamma", !0), se(ie, le, we, "\u0394", "\\Delta", !0), se(ie, le, we, "\u0398", "\\Theta", !0), se(ie, le, we, "\u039b", "\\Lambda", !0), se(ie, le, we, "\u039e", "\\Xi", !0), se(ie, le, we, "\u03a0", "\\Pi", !0), se(ie, le, we, "\u03a3", "\\Sigma", !0), se(ie, le, we, "\u03a5", "\\Upsilon", !0), se(ie, le, we, "\u03a6", "\\Phi", !0), se(ie, le, we, "\u03a8", "\\Psi", !0), se(ie, le, we, "\u03a9", "\\Omega", !0), se(ie, le, we, "A", "\u0391"), se(ie, le, we, "B", "\u0392"), se(ie, le, we, "E", "\u0395"), se(ie, le, we, "Z", "\u0396"), se(ie, le, we, "H", "\u0397"), se(ie, le, we, "I", "\u0399"), se(ie, le, we, "K", "\u039a"), se(ie, le, we, "M", "\u039c"), se(ie, le, we, "N", "\u039d"), se(ie, le, we, "O", "\u039f"), se(ie, le, we, "P", "\u03a1"), se(ie, le, we, "T", "\u03a4"), se(ie, le, we, "X", "\u03a7"), se(ie, le, we, "\xac", "\\neg", !0), se(ie, le, we, "\xac", "\\lnot"), se(ie, le, we, "\u22a4", "\\top"), se(ie, le, we, "\u22a5", "\\bot"), se(ie, le, we, "\u2205", "\\emptyset"), se(ie, he, we, "\u2205", "\\varnothing"), se(ie, le, de, "\u03b1", "\\alpha", !0), se(ie, le, de, "\u03b2", "\\beta", !0), se(ie, le, de, "\u03b3", "\\gamma", !0), se(ie, le, de, "\u03b4", "\\delta", !0), se(ie, le, de, "\u03f5", "\\epsilon", !0), se(ie, le, de, "\u03b6", "\\zeta", !0), se(ie, le, de, "\u03b7", "\\eta", !0), se(ie, le, de, "\u03b8", "\\theta", !0), se(ie, le, de, "\u03b9", "\\iota", !0), se(ie, le, de, "\u03ba", "\\kappa", !0), se(ie, le, de, "\u03bb", "\\lambda", !0), se(ie, le, de, "\u03bc", "\\mu", !0), se(ie, le, de, "\u03bd", "\\nu", !0), se(ie, le, de, "\u03be", "\\xi", !0), se(ie, le, de, "\u03bf", "\\omicron", !0), se(ie, le, de, "\u03c0", "\\pi", !0), se(ie, le, de, "\u03c1", "\\rho", !0), se(ie, le, de, "\u03c3", "\\sigma", !0), se(ie, le, de, "\u03c4", "\\tau", !0), se(ie, le, de, "\u03c5", "\\upsilon", !0), se(ie, le, de, "\u03d5", "\\phi", !0), se(ie, le, de, "\u03c7", "\\chi", !0), se(ie, le, de, "\u03c8", "\\psi", !0), se(ie, le, de, "\u03c9", "\\omega", !0), se(ie, le, de, "\u03b5", "\\varepsilon", !0), se(ie, le, de, "\u03d1", "\\vartheta", !0), se(ie, le, de, "\u03d6", "\\varpi", !0), se(ie, le, de, "\u03f1", "\\varrho", !0), se(ie, le, de, "\u03c2", "\\varsigma", !0), se(ie, le, de, "\u03c6", "\\varphi", !0), se(ie, le, me, "\u2217", "*", !0), se(ie, le, me, "+", "+"), se(ie, le, me, "\u2212", "-", !0), se(ie, le, me, "\u22c5", "\\cdot", !0), se(ie, le, me, "\u2218", "\\circ", !0), se(ie, le, me, "\xf7", "\\div", !0), se(ie, le, me, "\xb1", "\\pm", !0), se(ie, le, me, "\xd7", "\\times", !0), se(ie, le, me, "\u2229", "\\cap", !0), se(ie, le, me, "\u222a", "\\cup", !0), se(ie, le, me, "\u2216", "\\setminus", !0), se(ie, le, me, "\u2227", "\\land"), se(ie, le, me, "\u2228", "\\lor"), se(ie, le, me, "\u2227", "\\wedge", !0), se(ie, le, me, "\u2228", "\\vee", !0), se(ie, le, we, "\u221a", "\\surd"), se(ie, le, fe, "\u27e8", "\\langle", !0), se(ie, le, fe, "\u2223", "\\lvert"), se(ie, le, fe, "\u2225", "\\lVert"), se(ie, le, pe, "?", "?"), se(ie, le, pe, "!", "!"), se(ie, le, pe, "\u27e9", "\\rangle", !0), se(ie, le, pe, "\u2223", "\\rvert"), se(ie, le, pe, "\u2225", "\\rVert"), se(ie, le, ye, "=", "="), se(ie, le, ye, ":", ":"), se(ie, le, ye, "\u2248", "\\approx", !0), se(ie, le, ye, "\u2245", "\\cong", !0), se(ie, le, ye, "\u2265", "\\ge"), se(ie, le, ye, "\u2265", "\\geq", !0), se(ie, le, ye, "\u2190", "\\gets"), se(ie, le, ye, ">", "\\gt", !0), se(ie, le, ye, "\u2208", "\\in", !0), se(ie, le, ye, "\ue020", "\\@not"), se(ie, le, ye, "\u2282", "\\subset", !0), se(ie, le, ye, "\u2283", "\\supset", !0), se(ie, le, ye, "\u2286", "\\subseteq", !0), se(ie, le, ye, "\u2287", "\\supseteq", !0), se(ie, he, ye, "\u2288", "\\nsubseteq", !0), se(ie, he, ye, "\u2289", "\\nsupseteq", !0), se(ie, le, ye, "\u22a8", "\\models"), se(ie, le, ye, "\u2190", "\\leftarrow", !0), se(ie, le, ye, "\u2264", "\\le"), se(ie, le, ye, "\u2264", "\\leq", !0), se(ie, le, ye, "<", "\\lt", !0), se(ie, le, ye, "\u2192", "\\rightarrow", !0), se(ie, le, ye, "\u2192", "\\to"), se(ie, he, ye, "\u2271", "\\ngeq", !0), se(ie, he, ye, "\u2270", "\\nleq", !0), se(ie, le, xe, "\xa0", "\\ "), se(ie, le, xe, "\xa0", "\\space"), se(ie, le, xe, "\xa0", "\\nobreakspace"), se(ae, le, xe, "\xa0", "\\ "), se(ae, le, xe, "\xa0", " "), se(ae, le, xe, "\xa0", "\\space"), se(ae, le, xe, "\xa0", "\\nobreakspace"), se(ie, le, xe, null, "\\nobreak"), se(ie, le, xe, null, "\\allowbreak"), se(ie, le, be, ",", ","), se(ie, le, be, ";", ";"), se(ie, he, me, "\u22bc", "\\barwedge", !0), se(ie, he, me, "\u22bb", "\\veebar", !0), se(ie, le, me, "\u2299", "\\odot", !0), se(ie, le, me, "\u2295", "\\oplus", !0), se(ie, le, me, "\u2297", "\\otimes", !0), se(ie, le, we, "\u2202", "\\partial", !0), se(ie, le, me, "\u2298", "\\oslash", !0), se(ie, he, me, "\u229a", "\\circledcirc", !0), se(ie, he, me, "\u22a1", "\\boxdot", !0), se(ie, le, me, "\u25b3", "\\bigtriangleup"), se(ie, le, me, "\u25bd", "\\bigtriangledown"), se(ie, le, me, "\u2020", "\\dagger"), se(ie, le, me, "\u22c4", "\\diamond"), se(ie, le, me, "\u22c6", "\\star"), se(ie, le, me, "\u25c3", "\\triangleleft"), se(ie, le, me, "\u25b9", "\\triangleright"), se(ie, le, fe, "{", "\\{"), se(ae, le, we, "{", "\\{"), se(ae, le, we, "{", "\\textbraceleft"), se(ie, le, pe, "}", "\\}"), se(ae, le, we, "}", "\\}"), se(ae, le, we, "}", "\\textbraceright"), se(ie, le, fe, "{", "\\lbrace"), se(ie, le, pe, "}", "\\rbrace"), se(ie, le, fe, "[", "\\lbrack", !0), se(ae, le, we, "[", "\\lbrack", !0), se(ie, le, pe, "]", "\\rbrack", !0), se(ae, le, we, "]", "\\rbrack", !0), se(ie, le, fe, "(", "\\lparen", !0), se(ie, le, pe, ")", "\\rparen", !0), se(ae, le, we, "<", "\\textless", !0), se(ae, le, we, ">", "\\textgreater", !0), se(ie, le, fe, "\u230a", "\\lfloor", !0), se(ie, le, pe, "\u230b", "\\rfloor", !0), se(ie, le, fe, "\u2308", "\\lceil", !0), se(ie, le, pe, "\u2309", "\\rceil", !0), se(ie, le, we, "\\", "\\backslash"), se(ie, le, we, "\u2223", "|"), se(ie, le, we, "\u2223", "\\vert"), se(ae, le, we, "|", "\\textbar", !0), se(ie, le, we, "\u2225", "\\|"), se(ie, le, we, "\u2225", "\\Vert"), se(ae, le, we, "\u2225", "\\textbardbl"), se(ae, le, we, "~", "\\textasciitilde"), se(ae, le, we, "\\", "\\textbackslash"), se(ae, le, we, "^", "\\textasciicircum"), se(ie, le, ye, "\u2191", "\\uparrow", !0), se(ie, le, ye, "\u21d1", "\\Uparrow", !0), se(ie, le, ye, "\u2193", "\\downarrow", !0), se(ie, le, ye, "\u21d3", "\\Downarrow", !0), se(ie, le, ye, "\u2195", "\\updownarrow", !0), se(ie, le, ye, "\u21d5", "\\Updownarrow", !0), se(ie, le, ge, "\u2210", "\\coprod"), se(ie, le, ge, "\u22c1", "\\bigvee"), se(ie, le, ge, "\u22c0", "\\bigwedge"), se(ie, le, ge, "\u2a04", "\\biguplus"), se(ie, le, ge, "\u22c2", "\\bigcap"), se(ie, le, ge, "\u22c3", "\\bigcup"), se(ie, le, ge, "\u222b", "\\int"), se(ie, le, ge, "\u222b", "\\intop"), se(ie, le, ge, "\u222c", "\\iint"), se(ie, le, ge, "\u222d", "\\iiint"), se(ie, le, ge, "\u220f", "\\prod"), se(ie, le, ge, "\u2211", "\\sum"), se(ie, le, ge, "\u2a02", "\\bigotimes"), se(ie, le, ge, "\u2a01", "\\bigoplus"), se(ie, le, ge, "\u2a00", "\\bigodot"), se(ie, le, ge, "\u222e", "\\oint"), se(ie, le, ge, "\u222f", "\\oiint"), se(ie, le, ge, "\u2230", "\\oiiint"), se(ie, le, ge, "\u2a06", "\\bigsqcup"), se(ie, le, ge, "\u222b", "\\smallint"), se(ae, le, ue, "\u2026", "\\textellipsis"), se(ie, le, ue, "\u2026", "\\mathellipsis"), se(ae, le, ue, "\u2026", "\\ldots", !0), se(ie, le, ue, "\u2026", "\\ldots", !0), se(ie, le, ue, "\u22ef", "\\@cdots", !0), se(ie, le, ue, "\u22f1", "\\ddots", !0), se(ie, le, we, "\u22ee", "\\varvdots"), se(ie, le, ce, "\u02ca", "\\acute"), se(ie, le, ce, "\u02cb", "\\grave"), se(ie, le, ce, "\xa8", "\\ddot"), se(ie, le, ce, "~", "\\tilde"), se(ie, le, ce, "\u02c9", "\\bar"), se(ie, le, ce, "\u02d8", "\\breve"), se(ie, le, ce, "\u02c7", "\\check"), se(ie, le, ce, "^", "\\hat"), se(ie, le, ce, "\u20d7", "\\vec"), se(ie, le, ce, "\u02d9", "\\dot"), se(ie, le, ce, "\u02da", "\\mathring"), se(ie, le, de, "\ue131", "\\@imath"), se(ie, le, de, "\ue237", "\\@jmath"), se(ie, le, we, "\u0131", "\u0131"), se(ie, le, we, "\u0237", "\u0237"), se(ae, le, we, "\u0131", "\\i", !0), se(ae, le, we, "\u0237", "\\j", !0), se(ae, le, we, "\xdf", "\\ss", !0), se(ae, le, we, "\xe6", "\\ae", !0), se(ae, le, we, "\u0153", "\\oe", !0), se(ae, le, we, "\xf8", "\\o", !0), se(ae, le, we, "\xc6", "\\AE", !0), se(ae, le, we, "\u0152", "\\OE", !0), se(ae, le, we, "\xd8", "\\O", !0), se(ae, le, ce, "\u02ca", "\\'"), se(ae, le, ce, "\u02cb", "\\`"), se(ae, le, ce, "\u02c6", "\\^"), se(ae, le, ce, "\u02dc", "\\~"), se(ae, le, ce, "\u02c9", "\\="), se(ae, le, ce, "\u02d8", "\\u"), se(ae, le, ce, "\u02d9", "\\."), se(ae, le, ce, "\xb8", "\\c"), se(ae, le, ce, "\u02da", "\\r"), se(ae, le, ce, "\u02c7", "\\v"), se(ae, le, ce, "\xa8", '\\"'), se(ae, le, ce, "\u02dd", "\\H"), se(ae, le, ce, "\u25ef", "\\textcircled"); const ve = { "--": !0, "---": !0, "``": !0, "''": !0 }; se(ae, le, we, "\u2013", "--", !0), se(ae, le, we, "\u2013", "\\textendash"), se(ae, le, we, "\u2014", "---", !0), se(ae, le, we, "\u2014", "\\textemdash"), se(ae, le, we, "\u2018", "`", !0), se(ae, le, we, "\u2018", "\\textquoteleft"), se(ae, le, we, "\u2019", "'", !0), se(ae, le, we, "\u2019", "\\textquoteright"), se(ae, le, we, "\u201c", "``", !0), se(ae, le, we, "\u201c", "\\textquotedblleft"), se(ae, le, we, "\u201d", "''", !0), se(ae, le, we, "\u201d", "\\textquotedblright"), se(ie, le, we, "\xb0", "\\degree", !0), se(ae, le, we, "\xb0", "\\degree"), se(ae, le, we, "\xb0", "\\textdegree", !0), se(ie, le, we, "\xa3", "\\pounds"), se(ie, le, we, "\xa3", "\\mathsterling", !0), se(ae, le, we, "\xa3", "\\pounds"), se(ae, le, we, "\xa3", "\\textsterling", !0), se(ie, he, we, "\u2720", "\\maltese"), se(ae, he, we, "\u2720", "\\maltese"); const ke = '0123456789/@."'; for (let e = 0; e < ke.length; e++) { const t = ke.charAt(e); se(ie, le, we, t, t) } const Se = '0123456789!@*()-=+";:?/.,'; for (let e = 0; e < Se.length; e++) { const t = Se.charAt(e); se(ae, le, we, t, t) } const Me = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"; for (let e = 0; e < Me.length; e++) { const t = Me.charAt(e); se(ie, le, de, t, t), se(ae, le, we, t, t) } se(ie, he, we, "C", "\u2102"), se(ae, he, we, "C", "\u2102"), se(ie, he, we, "H", "\u210d"), se(ae, he, we, "H", "\u210d"), se(ie, he, we, "N", "\u2115"), se(ae, he, we, "N", "\u2115"), se(ie, he, we, "P", "\u2119"), se(ae, he, we, "P", "\u2119"), se(ie, he, we, "Q", "\u211a"), se(ae, he, we, "Q", "\u211a"), se(ie, he, we, "R", "\u211d"), se(ae, he, we, "R", "\u211d"), se(ie, he, we, "Z", "\u2124"), se(ae, he, we, "Z", "\u2124"), se(ie, le, de, "h", "\u210e"), se(ae, le, de, "h", "\u210e"); let ze = ""; for (let e = 0; e < Me.length; e++) { const t = Me.charAt(e); ze = String.fromCharCode(55349, 56320 + e), se(ie, le, de, t, ze), se(ae, le, we, t, ze), ze = String.fromCharCode(55349, 56372 + e), se(ie, le, de, t, ze), se(ae, le, we, t, ze), ze = String.fromCharCode(55349, 56424 + e), se(ie, le, de, t, ze), se(ae, le, we, t, ze), ze = String.fromCharCode(55349, 56580 + e), se(ie, le, de, t, ze), se(ae, le, we, t, ze), ze = String.fromCharCode(55349, 56684 + e), se(ie, le, de, t, ze), se(ae, le, we, t, ze), ze = String.fromCharCode(55349, 56736 + e), se(ie, le, de, t, ze), se(ae, le, we, t, ze), ze = String.fromCharCode(55349, 56788 + e), se(ie, le, de, t, ze), se(ae, le, we, t, ze), ze = String.fromCharCode(55349, 56840 + e), se(ie, le, de, t, ze), se(ae, le, we, t, ze), ze = String.fromCharCode(55349, 56944 + e), se(ie, le, de, t, ze), se(ae, le, we, t, ze), e < 26 && (ze = String.fromCharCode(55349, 56632 + e), se(ie, le, de, t, ze), se(ae, le, we, t, ze), ze = String.fromCharCode(55349, 56476 + e), se(ie, le, de, t, ze), se(ae, le, we, t, ze)) } ze = String.fromCharCode(55349, 56668), se(ie, le, de, "k", ze), se(ae, le, we, "k", ze); for (let e = 0; e < 10; e++) { const t = e.toString(); ze = String.fromCharCode(55349, 57294 + e), se(ie, le, de, t, ze), se(ae, le, we, t, ze), ze = String.fromCharCode(55349, 57314 + e), se(ie, le, de, t, ze), se(ae, le, we, t, ze), ze = String.fromCharCode(55349, 57324 + e), se(ie, le, de, t, ze), se(ae, le, we, t, ze), ze = String.fromCharCode(55349, 57334 + e), se(ie, le, de, t, ze), se(ae, le, we, t, ze) } const Ae = "\xd0\xde\xfe"; for (let e = 0; e < Ae.length; e++) { const t = Ae.charAt(e); se(ie, le, de, t, t), se(ae, le, we, t, t) } const Te = [["mathbf", "textbf", "Main-Bold"], ["mathbf", "textbf", "Main-Bold"], ["mathnormal", "textit", "Math-Italic"], ["mathnormal", "textit", "Math-Italic"], ["boldsymbol", "boldsymbol", "Main-BoldItalic"], ["boldsymbol", "boldsymbol", "Main-BoldItalic"], ["mathscr", "textscr", "Script-Regular"], ["", "", ""], ["", "", ""], ["", "", ""], ["mathfrak", "textfrak", "Fraktur-Regular"], ["mathfrak", "textfrak", "Fraktur-Regular"], ["mathbb", "textbb", "AMS-Regular"], ["mathbb", "textbb", "AMS-Regular"], ["mathboldfrak", "textboldfrak", "Fraktur-Regular"], ["mathboldfrak", "textboldfrak", "Fraktur-Regular"], ["mathsf", "textsf", "SansSerif-Regular"], ["mathsf", "textsf", "SansSerif-Regular"], ["mathboldsf", "textboldsf", "SansSerif-Bold"], ["mathboldsf", "textboldsf", "SansSerif-Bold"], ["mathitsf", "textitsf", "SansSerif-Italic"], ["mathitsf", "textitsf", "SansSerif-Italic"], ["", "", ""], ["", "", ""], ["mathtt", "texttt", "Typewriter-Regular"], ["mathtt", "texttt", "Typewriter-Regular"]], Be = [["mathbf", "textbf", "Main-Bold"], ["", "", ""], ["mathsf", "textsf", "SansSerif-Regular"], ["mathboldsf", "textboldsf", "SansSerif-Bold"], ["mathtt", "texttt", "Typewriter-Regular"]], Ce = function (e, t, r) { return oe[r][e] && oe[r][e].replace && (e = oe[r][e].replace), { value: e, metrics: N(e, t, r) } }, Ne = function (e, t, r, n, o) { const s = Ce(e, t, r), i = s.metrics; let a; if (e = s.value, i) { let t = i.italic; ("text" === r || n && "mathit" === n.font) && (t = 0), a = new Z(e, i.height, i.depth, t, i.skew, i.width, o) } else "undefined" != typeof console && console.warn("No character metrics for '" + e + "' in style '" + t + "' and mode '" + r + "'"), a = new Z(e, 0, 0, 0, 0, 0, o); if (n) { a.maxFontSize = n.sizeMultiplier, n.style.isTight() && a.classes.push("mtight"); const e = n.getColor(); e && (a.style.color = e) } return a }, qe = (e, t) => { if (G(e.classes) !== G(t.classes) || e.skew !== t.skew || e.maxFontSize !== t.maxFontSize) return !1; if (1 === e.classes.length) { const t = e.classes[0]; if ("mbin" === t || "mord" === t) return !1 } for (const r in e.style) if (e.style.hasOwnProperty(r) && e.style[r] !== t.style[r]) return !1; for (const r in t.style) if (t.style.hasOwnProperty(r) && e.style[r] !== t.style[r]) return !1; return !0 }, Ie = function (e) { let t = 0, r = 0, n = 0; for (let o = 0; o < e.children.length; o++) { const s = e.children[o]; s.height > t && (t = s.height), s.depth > r && (r = s.depth), s.maxFontSize > n && (n = s.maxFontSize) } e.height = t, e.depth = r, e.maxFontSize = n }, Re = function (e, t, r, n) { const o = new W(e, t, r, n); return Ie(o), o }, He = (e, t, r, n) => new W(e, t, r, n), Oe = function (e) { const t = new A(e); return Ie(t), t }, Ee = function (e, t, r) { let n, o = ""; switch (e) { case "amsrm": o = "AMS"; break; case "textrm": o = "Main"; break; case "textsf": o = "SansSerif"; break; case "texttt": o = "Typewriter"; break; default: o = e }return n = "textbf" === t && "textit" === r ? "BoldItalic" : "textbf" === t ? "Bold" : "textit" === t ? "Italic" : "Regular", o + "-" + n }, Le = { mathbf: { variant: "bold", fontName: "Main-Bold" }, mathrm: { variant: "normal", fontName: "Main-Regular" }, textit: { variant: "italic", fontName: "Main-Italic" }, mathit: { variant: "italic", fontName: "Main-Italic" }, mathnormal: { variant: "italic", fontName: "Math-Italic" }, mathbb: { variant: "double-struck", fontName: "AMS-Regular" }, mathcal: { variant: "script", fontName: "Caligraphic-Regular" }, mathfrak: { variant: "fraktur", fontName: "Fraktur-Regular" }, mathscr: { variant: "script", fontName: "Script-Regular" }, mathsf: { variant: "sans-serif", fontName: "SansSerif-Regular" }, mathtt: { variant: "monospace", fontName: "Typewriter-Regular" } }, De = { vec: ["vec", .471, .714], oiintSize1: ["oiintSize1", .957, .499], oiintSize2: ["oiintSize2", 1.472, .659], oiiintSize1: ["oiiintSize1", 1.304, .499], oiiintSize2: ["oiiintSize2", 1.98, .659] }; var Ve = { fontMap: Le, makeSymbol: Ne, mathsym: function (e, t, r, n) { return void 0 === n && (n = []), "boldsymbol" === r.font && Ce(e, "Main-Bold", t).metrics ? Ne(e, "Main-Bold", t, r, n.concat(["mathbf"])) : "\\" === e || "main" === oe[t][e].font ? Ne(e, "Main-Regular", t, r, n) : Ne(e, "AMS-Regular", t, r, n.concat(["amsrm"])) }, makeSpan: Re, makeSvgSpan: He, makeLineSpan: function (e, t, r) { const n = Re([e], [], t); return n.height = Math.max(r || t.fontMetrics().defaultRuleThickness, t.minRuleThickness), n.style.borderBottomWidth = F(n.height), n.maxFontSize = 1, n }, makeAnchor: function (e, t, r, n) { const o = new _(e, t, r, n); return Ie(o), o }, makeFragment: Oe, wrapFragment: function (e, t) { return e instanceof A ? Re([], [e], t) : e }, makeVList: function (e, t) { const { children: r, depth: n } = function (e) { if ("individualShift" === e.positionType) { const t = e.children, r = [t[0]], n = -t[0].shift - t[0].elem.depth; let o = n; for (let e = 1; e < t.length; e++) { const n = -t[e].shift - o - t[e].elem.depth, s = n - (t[e - 1].elem.height + t[e - 1].elem.depth); o += n, r.push({ type: "kern", size: s }), r.push(t[e]) } return { children: r, depth: n } } let t; if ("top" === e.positionType) { let r = e.positionData; for (let t = 0; t < e.children.length; t++) { const n = e.children[t]; r -= "kern" === n.type ? n.size : n.elem.height + n.elem.depth } t = r } else if ("bottom" === e.positionType) t = -e.positionData; else { const r = e.children[0]; if ("elem" !== r.type) throw new Error('First child must have type "elem".'); if ("shift" === e.positionType) t = -r.elem.depth - e.positionData; else { if ("firstBaseline" !== e.positionType) throw new Error("Invalid positionType " + e.positionType + "."); t = -r.elem.depth } } return { children: e.children, depth: t } }(e); let o = 0; for (let e = 0; e < r.length; e++) { const t = r[e]; if ("elem" === t.type) { const e = t.elem; o = Math.max(o, e.maxFontSize, e.height) } } o += 2; const s = Re(["pstrut"], []); s.style.height = F(o); const i = []; let a = n, l = n, h = n; for (let e = 0; e < r.length; e++) { const t = r[e]; if ("kern" === t.type) h += t.size; else { const e = t.elem, r = t.wrapperClasses || [], n = t.wrapperStyle || {}, a = Re(r, [s, e], void 0, n); a.style.top = F(-o - h - e.depth), t.marginLeft && (a.style.marginLeft = t.marginLeft), t.marginRight && (a.style.marginRight = t.marginRight), i.push(a), h += e.height + e.depth } a = Math.min(a, h), l = Math.max(l, h) } const c = Re(["vlist"], i); let m; if (c.style.height = F(l), a < 0) { const e = Re([], []), t = Re(["vlist"], [e]); t.style.height = F(-a); const r = Re(["vlist-s"], [new Z("\u200b")]); m = [Re(["vlist-r"], [c, r]), Re(["vlist-r"], [t])] } else m = [Re(["vlist-r"], [c])]; const p = Re(["vlist-t"], m); return 2 === m.length && p.classes.push("vlist-t2"), p.height = l, p.depth = -a, p }, makeOrd: function (e, t, r) { const o = e.mode, s = e.text, i = ["mord"], a = "math" === o || "text" === o && t.font, l = a ? t.font : t.fontFamily; let h = "", c = ""; if (55349 === s.charCodeAt(0) && ([h, c] = function (e, t) { const r = 1024 * (e.charCodeAt(0) - 55296) + (e.charCodeAt(1) - 56320) + 65536, o = "math" === t ? 0 : 1; if (119808 <= r && r < 120484) { const e = Math.floor((r - 119808) / 26); return [Te[e][2], Te[e][o]] } if (120782 <= r && r <= 120831) { const e = Math.floor((r - 120782) / 10); return [Be[e][2], Be[e][o]] } if (120485 === r || 120486 === r) return [Te[0][2], Te[0][o]]; if (120486 < r && r < 120782) return ["", ""]; throw new n("Unsupported character: " + e) }(s, o)), h.length > 0) return Ne(s, h, o, t, i.concat(c)); if (l) { let e, n; if ("boldsymbol" === l) { const t = function (e, t, r, n, o) { return "textord" !== o && Ce(e, "Math-BoldItalic", t).metrics ? { fontName: "Math-BoldItalic", fontClass: "boldsymbol" } : { fontName: "Main-Bold", fontClass: "mathbf" } }(s, o, 0, 0, r); e = t.fontName, n = [t.fontClass] } else a ? (e = Le[l].fontName, n = [l]) : (e = Ee(l, t.fontWeight, t.fontShape), n = [l, t.fontWeight, t.fontShape]); if (Ce(s, e, o).metrics) return Ne(s, e, o, t, i.concat(n)); if (ve.hasOwnProperty(s) && "Typewriter" === e.slice(0, 10)) { const r = []; for (let a = 0; a < s.length; a++)r.push(Ne(s[a], e, o, t, i.concat(n))); return Oe(r) } } if ("mathord" === r) return Ne(s, "Math-Italic", o, t, i.concat(["mathnormal"])); if ("textord" === r) { const e = oe[o][s] && oe[o][s].font; if ("ams" === e) { const e = Ee("amsrm", t.fontWeight, t.fontShape); return Ne(s, e, o, t, i.concat("amsrm", t.fontWeight, t.fontShape)) } if ("main" !== e && e) { const r = Ee(e, t.fontWeight, t.fontShape); return Ne(s, r, o, t, i.concat(r, t.fontWeight, t.fontShape)) } { const e = Ee("textrm", t.fontWeight, t.fontShape); return Ne(s, e, o, t, i.concat(t.fontWeight, t.fontShape)) } } throw new Error("unexpected type: " + r + " in makeOrd") }, makeGlue: (e, t) => { const r = Re(["mspace"], [], t), n = P(e, t); return r.style.marginRight = F(n), r }, staticSvg: function (e, t) { const [r, n, o] = De[e], s = new J(r), i = new K([s], { width: F(n), height: F(o), style: "width:" + F(n), viewBox: "0 0 " + 1e3 * n + " " + 1e3 * o, preserveAspectRatio: "xMinYMin" }), a = He(["overlay"], [i], t); return a.height = o, a.style.height = F(o), a.style.width = F(n), a }, svgData: De, tryCombineChars: e => { for (let t = 0; t < e.length - 1; t++) { const r = e[t], n = e[t + 1]; r instanceof Z && n instanceof Z && qe(r, n) && (r.text += n.text, r.height = Math.max(r.height, n.height), r.depth = Math.max(r.depth, n.depth), r.italic = n.italic, e.splice(t + 1, 1), t--) } return e } }; const Pe = { number: 3, unit: "mu" }, Fe = { number: 4, unit: "mu" }, Ge = { number: 5, unit: "mu" }, Ue = { mord: { mop: Pe, mbin: Fe, mrel: Ge, minner: Pe }, mop: { mord: Pe, mop: Pe, mrel: Ge, minner: Pe }, mbin: { mord: Fe, mop: Fe, mopen: Fe, minner: Fe }, mrel: { mord: Ge, mop: Ge, mopen: Ge, minner: Ge }, mopen: {}, mclose: { mop: Pe, mbin: Fe, mrel: Ge, minner: Pe }, mpunct: { mord: Pe, mop: Pe, mrel: Ge, mopen: Pe, mclose: Pe, mpunct: Pe, minner: Pe }, minner: { mord: Pe, mop: Pe, mbin: Fe, mrel: Ge, mopen: Pe, mpunct: Pe, minner: Pe } }, Ye = { mord: { mop: Pe }, mop: { mord: Pe, mop: Pe }, mbin: {}, mrel: {}, mopen: {}, mclose: { mop: Pe }, mpunct: {}, minner: { mop: Pe } }, Xe = {}, We = {}, _e = {}; function je(e) { let { type: t, names: r, props: n, handler: o, htmlBuilder: s, mathmlBuilder: i } = e; const a = { type: t, numArgs: n.numArgs, argTypes: n.argTypes, allowedInArgument: !!n.allowedInArgument, allowedInText: !!n.allowedInText, allowedInMath: void 0 === n.allowedInMath || n.allowedInMath, numOptionalArgs: n.numOptionalArgs || 0, infix: !!n.infix, primitive: !!n.primitive, handler: o }; for (let e = 0; e < r.length; ++e)Xe[r[e]] = a; t && (s && (We[t] = s), i && (_e[t] = i)) } function $e(e) { let { type: t, htmlBuilder: r, mathmlBuilder: n } = e; je({ type: t, names: [], props: { numArgs: 0 }, handler() { throw new Error("Should never be called.") }, htmlBuilder: r, mathmlBuilder: n }) } const Ze = function (e) { return "ordgroup" === e.type && 1 === e.body.length ? e.body[0] : e }, Ke = function (e) { return "ordgroup" === e.type ? e.body : [e] }, Je = Ve.makeSpan, Qe = ["leftmost", "mbin", "mopen", "mrel", "mop", "mpunct"], et = ["rightmost", "mrel", "mclose", "mpunct"], tt = { display: w.DISPLAY, text: w.TEXT, script: w.SCRIPT, scriptscript: w.SCRIPTSCRIPT }, rt = { mord: "mord", mop: "mop", mbin: "mbin", mrel: "mrel", mopen: "mopen", mclose: "mclose", mpunct: "mpunct", minner: "minner" }, nt = function (e, t, r, n) { void 0 === n && (n = [null, null]); const o = []; for (let r = 0; r < e.length; r++) { const n = ht(e[r], t); if (n instanceof A) { const e = n.children; o.push(...e) } else o.push(n) } if (Ve.tryCombineChars(o), !r) return o; let s = t; if (1 === e.length) { const r = e[0]; "sizing" === r.type ? s = t.havingSize(r.size) : "styling" === r.type && (s = t.havingStyle(tt[r.style])) } const i = Je([n[0] || "leftmost"], [], t), a = Je([n[1] || "rightmost"], [], t), h = "root" === r; return ot(o, ((e, t) => { const r = t.classes[0], n = e.classes[0]; "mbin" === r && l.contains(et, n) ? t.classes[0] = "mord" : "mbin" === n && l.contains(Qe, r) && (e.classes[0] = "mord") }), { node: i }, a, h), ot(o, ((e, t) => { const r = at(t), n = at(e), o = r && n ? e.hasClass("mtight") ? Ye[r][n] : Ue[r][n] : null; if (o) return Ve.makeGlue(o, s) }), { node: i }, a, h), o }, ot = function (e, t, r, n, o) { n && e.push(n); let s = 0; for (; s < e.length; s++) { const n = e[s], i = st(n); if (i) { ot(i.children, t, r, null, o); continue } const a = !n.hasClass("mspace"); if (a) { const o = t(n, r.node); o && (r.insertAfter ? r.insertAfter(o) : (e.unshift(o), s++)) } a ? r.node = n : o && n.hasClass("newline") && (r.node = Je(["leftmost"])), r.insertAfter = (t => r => { e.splice(t + 1, 0, r), s++ })(s) } n && e.pop() }, st = function (e) { return e instanceof A || e instanceof _ || e instanceof W && e.hasClass("enclosing") ? e : null }, it = function (e, t) { const r = st(e); if (r) { const e = r.children; if (e.length) { if ("right" === t) return it(e[e.length - 1], "right"); if ("left" === t) return it(e[0], "left") } } return e }, at = function (e, t) { return e ? (t && (e = it(e, t)), rt[e.classes[0]] || null) : null }, lt = function (e, t) { const r = ["nulldelimiter"].concat(e.baseSizingClasses()); return Je(t.concat(r)) }, ht = function (e, t, r) { if (!e) return Je(); if (We[e.type]) { let n = We[e.type](e, t); if (r && t.size !== r.size) { n = Je(t.sizingClasses(r), [n], t); const e = t.sizeMultiplier / r.sizeMultiplier; n.height *= e, n.depth *= e } return n } throw new n("Got group of unknown type: '" + e.type + "'") }; function ct(e, t) { const r = Je(["base"], e, t), n = Je(["strut"]); return n.style.height = F(r.height + r.depth), r.depth && (n.style.verticalAlign = F(-r.depth)), r.children.unshift(n), r } function mt(e, t) { let r = null; 1 === e.length && "tag" === e[0].type && (r = e[0].tag, e = e[0].body); const n = nt(e, t, "root"); let o; 2 === n.length && n[1].hasClass("tag") && (o = n.pop()); const s = []; let i, a = []; for (let e = 0; e < n.length; e++)if (a.push(n[e]), n[e].hasClass("mbin") || n[e].hasClass("mrel") || n[e].hasClass("allowbreak")) { let r = !1; for (; e < n.length - 1 && n[e + 1].hasClass("mspace") && !n[e + 1].hasClass("newline");)e++, a.push(n[e]), n[e].hasClass("nobreak") && (r = !0); r || (s.push(ct(a, t)), a = []) } else n[e].hasClass("newline") && (a.pop(), a.length > 0 && (s.push(ct(a, t)), a = []), s.push(n[e])); a.length > 0 && s.push(ct(a, t)), r ? (i = ct(nt(r, t, !0)), i.classes = ["tag"], s.push(i)) : o && s.push(o); const l = Je(["katex-html"], s); if (l.setAttribute("aria-hidden", "true"), i) { const e = i.children[0]; e.style.height = F(l.height + l.depth), l.depth && (e.style.verticalAlign = F(-l.depth)) } return l } function pt(e) { return new A(e) } class ut { constructor(e, t, r) { this.type = void 0, this.attributes = void 0, this.children = void 0, this.classes = void 0, this.type = e, this.attributes = {}, this.children = t || [], this.classes = r || [] } setAttribute(e, t) { this.attributes[e] = t } getAttribute(e) { return this.attributes[e] } toNode() { const e = document.createElementNS("http://www.w3.org/1998/Math/MathML", this.type); for (const t in this.attributes) Object.prototype.hasOwnProperty.call(this.attributes, t) && e.setAttribute(t, this.attributes[t]); this.classes.length > 0 && (e.className = G(this.classes)); for (let t = 0; t < this.children.length; t++)e.appendChild(this.children[t].toNode()); return e } toMarkup() { let e = "<" + this.type; for (const t in this.attributes) Object.prototype.hasOwnProperty.call(this.attributes, t) && (e += " " + t + '="', e += l.escape(this.attributes[t]), e += '"'); this.classes.length > 0 && (e += ' class ="' + l.escape(G(this.classes)) + '"'), e += ">"; for (let t = 0; t < this.children.length; t++)e += this.children[t].toMarkup(); return e += "</" + this.type + ">", e } toText() { return this.children.map((e => e.toText())).join("") } } class dt { constructor(e) { this.text = void 0, this.text = e } toNode() { return document.createTextNode(this.text) } toMarkup() { return l.escape(this.toText()) } toText() { return this.text } } var gt = { MathNode: ut, TextNode: dt, SpaceNode: class { constructor(e) { this.width = void 0, this.character = void 0, this.width = e, this.character = e >= .05555 && e <= .05556 ? "\u200a" : e >= .1666 && e <= .1667 ? "\u2009" : e >= .2222 && e <= .2223 ? "\u2005" : e >= .2777 && e <= .2778 ? "\u2005\u200a" : e >= -.05556 && e <= -.05555 ? "\u200a\u2063" : e >= -.1667 && e <= -.1666 ? "\u2009\u2063" : e >= -.2223 && e <= -.2222 ? "\u205f\u2063" : e >= -.2778 && e <= -.2777 ? "\u2005\u2063" : null } toNode() { if (this.character) return document.createTextNode(this.character); { const e = document.createElementNS("http://www.w3.org/1998/Math/MathML", "mspace"); return e.setAttribute("width", F(this.width)), e } } toMarkup() { return this.character ? "<mtext>" + this.character + "</mtext>" : '<mspace width="' + F(this.width) + '"/>' } toText() { return this.character ? this.character : " " } }, newDocumentFragment: pt }; const ft = function (e, t, r) { return !oe[t][e] || !oe[t][e].replace || 55349 === e.charCodeAt(0) || ve.hasOwnProperty(e) && r && (r.fontFamily && "tt" === r.fontFamily.slice(4, 6) || r.font && "tt" === r.font.slice(4, 6)) || (e = oe[t][e].replace), new gt.TextNode(e) }, bt = function (e) { return 1 === e.length ? e[0] : new gt.MathNode("mrow", e) }, yt = function (e, t) { if ("texttt" === t.fontFamily) return "monospace"; if ("textsf" === t.fontFamily) return "textit" === t.fontShape && "textbf" === t.fontWeight ? "sans-serif-bold-italic" : "textit" === t.fontShape ? "sans-serif-italic" : "textbf" === t.fontWeight ? "bold-sans-serif" : "sans-serif"; if ("textit" === t.fontShape && "textbf" === t.fontWeight) return "bold-italic"; if ("textit" === t.fontShape) return "italic"; if ("textbf" === t.fontWeight) return "bold"; const r = t.font; if (!r || "mathnormal" === r) return null; const n = e.mode; if ("mathit" === r) return "italic"; if ("boldsymbol" === r) return "textord" === e.type ? "bold" : "bold-italic"; if ("mathbf" === r) return "bold"; if ("mathbb" === r) return "double-struck"; if ("mathfrak" === r) return "fraktur"; if ("mathscr" === r || "mathcal" === r) return "script"; if ("mathsf" === r) return "sans-serif"; if ("mathtt" === r) return "monospace"; let o = e.text; if (l.contains(["\\imath", "\\jmath"], o)) return null; oe[n][o] && oe[n][o].replace && (o = oe[n][o].replace); return N(o, Ve.fontMap[r].fontName, n) ? Ve.fontMap[r].variant : null }, xt = function (e, t, r) { if (1 === e.length) { const n = vt(e[0], t); return r && n instanceof ut && "mo" === n.type && (n.setAttribute("lspace", "0em"), n.setAttribute("rspace", "0em")), [n] } const n = []; let o; for (let r = 0; r < e.length; r++) { const s = vt(e[r], t); if (s instanceof ut && o instanceof ut) { if ("mtext" === s.type && "mtext" === o.type && s.getAttribute("mathvariant") === o.getAttribute("mathvariant")) { o.children.push(...s.children); continue } if ("mn" === s.type && "mn" === o.type) { o.children.push(...s.children); continue } if ("mi" === s.type && 1 === s.children.length && "mn" === o.type) { const e = s.children[0]; if (e instanceof dt && "." === e.text) { o.children.push(...s.children); continue } } else if ("mi" === o.type && 1 === o.children.length) { const e = o.children[0]; if (e instanceof dt && "\u0338" === e.text && ("mo" === s.type || "mi" === s.type || "mn" === s.type)) { const e = s.children[0]; e instanceof dt && e.text.length > 0 && (e.text = e.text.slice(0, 1) + "\u0338" + e.text.slice(1), n.pop()) } } } n.push(s), o = s } return n }, wt = function (e, t, r) { return bt(xt(e, t, r)) }, vt = function (e, t) { if (!e) return new gt.MathNode("mrow"); if (_e[e.type]) { return _e[e.type](e, t) } throw new n("Got group of unknown type: '" + e.type + "'") }; function kt(e, t, r, n, o) { const s = xt(e, r); let i; i = 1 === s.length && s[0] instanceof ut && l.contains(["mrow", "mtable"], s[0].type) ? s[0] : new gt.MathNode("mrow", s); const a = new gt.MathNode("annotation", [new gt.TextNode(t)]); a.setAttribute("encoding", "application/x-tex"); const h = new gt.MathNode("semantics", [i, a]), c = new gt.MathNode("math", [h]); c.setAttribute("xmlns", "http://www.w3.org/1998/Math/MathML"), n && c.setAttribute("display", "block"); const m = o ? "katex" : "katex-mathml"; return Ve.makeSpan([m], [c]) } const St = function (e) { return new E({ style: e.displayMode ? w.DISPLAY : w.TEXT, maxSize: e.maxSize, minRuleThickness: e.minRuleThickness }) }, Mt = function (e, t) { if (t.displayMode) { const r = ["katex-display"]; t.leqno && r.push("leqno"), t.fleqn && r.push("fleqn"), e = Ve.makeSpan(r, [e]) } return e }, zt = function (e, t, r) { const n = St(r); let o; if ("mathml" === r.output) return kt(e, t, n, r.displayMode, !0); if ("html" === r.output) { const t = mt(e, n); o = Ve.makeSpan(["katex"], [t]) } else { const s = kt(e, t, n, r.displayMode, !1), i = mt(e, n); o = Ve.makeSpan(["katex"], [s, i]) } return Mt(o, r) }; const At = { widehat: "^", widecheck: "\u02c7", widetilde: "~", utilde: "~", overleftarrow: "\u2190", underleftarrow: "\u2190", xleftarrow: "\u2190", overrightarrow: "\u2192", underrightarrow: "\u2192", xrightarrow: "\u2192", underbrace: "\u23df", overbrace: "\u23de", overgroup: "\u23e0", undergroup: "\u23e1", overleftrightarrow: "\u2194", underleftrightarrow: "\u2194", xleftrightarrow: "\u2194", Overrightarrow: "\u21d2", xRightarrow: "\u21d2", overleftharpoon: "\u21bc", xleftharpoonup: "\u21bc", overrightharpoon: "\u21c0", xrightharpoonup: "\u21c0", xLeftarrow: "\u21d0", xLeftrightarrow: "\u21d4", xhookleftarrow: "\u21a9", xhookrightarrow: "\u21aa", xmapsto: "\u21a6", xrightharpoondown: "\u21c1", xleftharpoondown: "\u21bd", xrightleftharpoons: "\u21cc", xleftrightharpoons: "\u21cb", xtwoheadleftarrow: "\u219e", xtwoheadrightarrow: "\u21a0", xlongequal: "=", xtofrom: "\u21c4", xrightleftarrows: "\u21c4", xrightequilibrium: "\u21cc", xleftequilibrium: "\u21cb", "\\cdrightarrow": "\u2192", "\\cdleftarrow": "\u2190", "\\cdlongequal": "=" }, Tt = { overrightarrow: [["rightarrow"], .888, 522, "xMaxYMin"], overleftarrow: [["leftarrow"], .888, 522, "xMinYMin"], underrightarrow: [["rightarrow"], .888, 522, "xMaxYMin"], underleftarrow: [["leftarrow"], .888, 522, "xMinYMin"], xrightarrow: [["rightarrow"], 1.469, 522, "xMaxYMin"], "\\cdrightarrow": [["rightarrow"], 3, 522, "xMaxYMin"], xleftarrow: [["leftarrow"], 1.469, 522, "xMinYMin"], "\\cdleftarrow": [["leftarrow"], 3, 522, "xMinYMin"], Overrightarrow: [["doublerightarrow"], .888, 560, "xMaxYMin"], xRightarrow: [["doublerightarrow"], 1.526, 560, "xMaxYMin"], xLeftarrow: [["doubleleftarrow"], 1.526, 560, "xMinYMin"], overleftharpoon: [["leftharpoon"], .888, 522, "xMinYMin"], xleftharpoonup: [["leftharpoon"], .888, 522, "xMinYMin"], xleftharpoondown: [["leftharpoondown"], .888, 522, "xMinYMin"], overrightharpoon: [["rightharpoon"], .888, 522, "xMaxYMin"], xrightharpoonup: [["rightharpoon"], .888, 522, "xMaxYMin"], xrightharpoondown: [["rightharpoondown"], .888, 522, "xMaxYMin"], xlongequal: [["longequal"], .888, 334, "xMinYMin"], "\\cdlongequal": [["longequal"], 3, 334, "xMinYMin"], xtwoheadleftarrow: [["twoheadleftarrow"], .888, 334, "xMinYMin"], xtwoheadrightarrow: [["twoheadrightarrow"], .888, 334, "xMaxYMin"], overleftrightarrow: [["leftarrow", "rightarrow"], .888, 522], overbrace: [["leftbrace", "midbrace", "rightbrace"], 1.6, 548], underbrace: [["leftbraceunder", "midbraceunder", "rightbraceunder"], 1.6, 548], underleftrightarrow: [["leftarrow", "rightarrow"], .888, 522], xleftrightarrow: [["leftarrow", "rightarrow"], 1.75, 522], xLeftrightarrow: [["doubleleftarrow", "doublerightarrow"], 1.75, 560], xrightleftharpoons: [["leftharpoondownplus", "rightharpoonplus"], 1.75, 716], xleftrightharpoons: [["leftharpoonplus", "rightharpoondownplus"], 1.75, 716], xhookleftarrow: [["leftarrow", "righthook"], 1.08, 522], xhookrightarrow: [["lefthook", "rightarrow"], 1.08, 522], overlinesegment: [["leftlinesegment", "rightlinesegment"], .888, 522], underlinesegment: [["leftlinesegment", "rightlinesegment"], .888, 522], overgroup: [["leftgroup", "rightgroup"], .888, 342], undergroup: [["leftgroupunder", "rightgroupunder"], .888, 342], xmapsto: [["leftmapsto", "rightarrow"], 1.5, 522], xtofrom: [["leftToFrom", "rightToFrom"], 1.75, 528], xrightleftarrows: [["baraboveleftarrow", "rightarrowabovebar"], 1.75, 901], xrightequilibrium: [["baraboveshortleftharpoon", "rightharpoonaboveshortbar"], 1.75, 716], xleftequilibrium: [["shortbaraboveleftharpoon", "shortrightharpoonabovebar"], 1.75, 716] }; var Bt = function (e, t, r, n, o) { let s; const i = e.height + e.depth + r + n; if (/fbox|color|angl/.test(t)) { if (s = Ve.makeSpan(["stretchy", t], [], o), "fbox" === t) { const e = o.color && o.getColor(); e && (s.style.borderColor = e) } } else { const e = []; /^[bx]cancel$/.test(t) && e.push(new Q({ x1: "0", y1: "0", x2: "100%", y2: "100%", "stroke-width": "0.046em" })), /^x?cancel$/.test(t) && e.push(new Q({ x1: "0", y1: "100%", x2: "100%", y2: "0", "stroke-width": "0.046em" })); const r = new K(e, { width: "100%", height: F(i) }); s = Ve.makeSvgSpan([], [r], o) } return s.height = i, s.style.height = F(i), s }, Ct = function (e) { const t = new gt.MathNode("mo", [new gt.TextNode(At[e.replace(/^\\/, "")])]); return t.setAttribute("stretchy", "true"), t }, Nt = function (e, t) { const { span: r, minWidth: n, height: o } = function () { let r = 4e5; const n = e.label.slice(1); if (l.contains(["widehat", "widecheck", "widetilde", "utilde"], n)) { const s = "ordgroup" === (o = e.base).type ? o.body.length : 1; let i, a, l; if (s > 5) "widehat" === n || "widecheck" === n ? (i = 420, r = 2364, l = .42, a = n + "4") : (i = 312, r = 2340, l = .34, a = "tilde4"); else { const e = [1, 1, 2, 2, 3, 3][s]; "widehat" === n || "widecheck" === n ? (r = [0, 1062, 2364, 2364, 2364][e], i = [0, 239, 300, 360, 420][e], l = [0, .24, .3, .3, .36, .42][e], a = n + e) : (r = [0, 600, 1033, 2339, 2340][e], i = [0, 260, 286, 306, 312][e], l = [0, .26, .286, .3, .306, .34][e], a = "tilde" + e) } const h = new J(a), c = new K([h], { width: "100%", height: F(l), viewBox: "0 0 " + r + " " + i, preserveAspectRatio: "none" }); return { span: Ve.makeSvgSpan([], [c], t), minWidth: 0, height: l } } { const e = [], o = Tt[n], [s, i, a] = o, l = a / 1e3, h = s.length; let c, m; if (1 === h) { c = ["hide-tail"], m = [o[3]] } else if (2 === h) c = ["halfarrow-left", "halfarrow-right"], m = ["xMinYMin", "xMaxYMin"]; else { if (3 !== h) throw new Error("Correct katexImagesData or update code here to support\n                    " + h + " children."); c = ["brace-left", "brace-center", "brace-right"], m = ["xMinYMin", "xMidYMin", "xMaxYMin"] } for (let n = 0; n < h; n++) { const o = new J(s[n]), p = new K([o], { width: "400em", height: F(l), viewBox: "0 0 " + r + " " + a, preserveAspectRatio: m[n] + " slice" }), u = Ve.makeSvgSpan([c[n]], [p], t); if (1 === h) return { span: u, minWidth: i, height: l }; u.style.height = F(l), e.push(u) } return { span: Ve.makeSpan(["stretchy"], e, t), minWidth: i, height: l } } var o }(); return r.height = o, r.style.height = F(o), n > 0 && (r.style.minWidth = F(n)), r }; function qt(e, t) { if (!e || e.type !== t) throw new Error("Expected node of type " + t + ", but got " + (e ? "node of type " + e.type : String(e))); return e } function It(e) { const t = Rt(e); if (!t) throw new Error("Expected node of symbol group type, but got " + (e ? "node of type " + e.type : String(e))); return t } function Rt(e) { return e && ("atom" === e.type || re.hasOwnProperty(e.type)) ? e : null } const Ht = (e, t) => { let r, n, o; e && "supsub" === e.type ? (n = qt(e.base, "accent"), r = n.base, e.base = r, o = function (e) { if (e instanceof W) return e; throw new Error("Expected span<HtmlDomNode> but got " + String(e) + ".") }(ht(e, t)), e.base = n) : (n = qt(e, "accent"), r = n.base); const s = ht(r, t.havingCrampedStyle()); let i = 0; if (n.isShifty && l.isCharacterBox(r)) { const e = l.getBaseElem(r); i = ee(ht(e, t.havingCrampedStyle())).skew } const a = "\\c" === n.label; let h, c = a ? s.height + s.depth : Math.min(s.height, t.fontMetrics().xHeight); if (n.isStretchy) h = Nt(n, t), h = Ve.makeVList({ positionType: "firstBaseline", children: [{ type: "elem", elem: s }, { type: "elem", elem: h, wrapperClasses: ["svg-align"], wrapperStyle: i > 0 ? { width: "calc(100% - " + F(2 * i) + ")", marginLeft: F(2 * i) } : void 0 }] }, t); else { let e, r; "\\vec" === n.label ? (e = Ve.staticSvg("vec", t), r = Ve.svgData.vec[1]) : (e = Ve.makeOrd({ mode: n.mode, text: n.label }, t, "textord"), e = ee(e), e.italic = 0, r = e.width, a && (c += e.depth)), h = Ve.makeSpan(["accent-body"], [e]); const o = "\\textcircled" === n.label; o && (h.classes.push("accent-full"), c = s.height); let l = i; o || (l -= r / 2), h.style.left = F(l), "\\textcircled" === n.label && (h.style.top = ".2em"), h = Ve.makeVList({ positionType: "firstBaseline", children: [{ type: "elem", elem: s }, { type: "kern", size: -c }, { type: "elem", elem: h }] }, t) } const m = Ve.makeSpan(["mord", "accent"], [h], t); return o ? (o.children[0] = m, o.height = Math.max(m.height, o.height), o.classes[0] = "mord", o) : m }, Ot = (e, t) => { const r = e.isStretchy ? Ct(e.label) : new gt.MathNode("mo", [ft(e.label, e.mode)]), n = new gt.MathNode("mover", [vt(e.base, t), r]); return n.setAttribute("accent", "true"), n }, Et = new RegExp(["\\acute", "\\grave", "\\ddot", "\\tilde", "\\bar", "\\breve", "\\check", "\\hat", "\\vec", "\\dot", "\\mathring"].map((e => "\\" + e)).join("|")); je({ type: "accent", names: ["\\acute", "\\grave", "\\ddot", "\\tilde", "\\bar", "\\breve", "\\check", "\\hat", "\\vec", "\\dot", "\\mathring", "\\widecheck", "\\widehat", "\\widetilde", "\\overrightarrow", "\\overleftarrow", "\\Overrightarrow", "\\overleftrightarrow", "\\overgroup", "\\overlinesegment", "\\overleftharpoon", "\\overrightharpoon"], props: { numArgs: 1 }, handler: (e, t) => { const r = Ze(t[0]), n = !Et.test(e.funcName), o = !n || "\\widehat" === e.funcName || "\\widetilde" === e.funcName || "\\widecheck" === e.funcName; return { type: "accent", mode: e.parser.mode, label: e.funcName, isStretchy: n, isShifty: o, base: r } }, htmlBuilder: Ht, mathmlBuilder: Ot }), je({ type: "accent", names: ["\\'", "\\`", "\\^", "\\~", "\\=", "\\u", "\\.", '\\"', "\\c", "\\r", "\\H", "\\v", "\\textcircled"], props: { numArgs: 1, allowedInText: !0, allowedInMath: !0, argTypes: ["primitive"] }, handler: (e, t) => { const r = t[0]; let n = e.parser.mode; return "math" === n && (e.parser.settings.reportNonstrict("mathVsTextAccents", "LaTeX's accent " + e.funcName + " works only in text mode"), n = "text"), { type: "accent", mode: n, label: e.funcName, isStretchy: !1, isShifty: !0, base: r } }, htmlBuilder: Ht, mathmlBuilder: Ot }), je({ type: "accentUnder", names: ["\\underleftarrow", "\\underrightarrow", "\\underleftrightarrow", "\\undergroup", "\\underlinesegment", "\\utilde"], props: { numArgs: 1 }, handler: (e, t) => { let { parser: r, funcName: n } = e; const o = t[0]; return { type: "accentUnder", mode: r.mode, label: n, base: o } }, htmlBuilder: (e, t) => { const r = ht(e.base, t), n = Nt(e, t), o = "\\utilde" === e.label ? .12 : 0, s = Ve.makeVList({ positionType: "top", positionData: r.height, children: [{ type: "elem", elem: n, wrapperClasses: ["svg-align"] }, { type: "kern", size: o }, { type: "elem", elem: r }] }, t); return Ve.makeSpan(["mord", "accentunder"], [s], t) }, mathmlBuilder: (e, t) => { const r = Ct(e.label), n = new gt.MathNode("munder", [vt(e.base, t), r]); return n.setAttribute("accentunder", "true"), n } }); const Lt = e => { const t = new gt.MathNode("mpadded", e ? [e] : []); return t.setAttribute("width", "+0.6em"), t.setAttribute("lspace", "0.3em"), t }; je({ type: "xArrow", names: ["\\xleftarrow", "\\xrightarrow", "\\xLeftarrow", "\\xRightarrow", "\\xleftrightarrow", "\\xLeftrightarrow", "\\xhookleftarrow", "\\xhookrightarrow", "\\xmapsto", "\\xrightharpoondown", "\\xrightharpoonup", "\\xleftharpoondown", "\\xleftharpoonup", "\\xrightleftharpoons", "\\xleftrightharpoons", "\\xlongequal", "\\xtwoheadrightarrow", "\\xtwoheadleftarrow", "\\xtofrom", "\\xrightleftarrows", "\\xrightequilibrium", "\\xleftequilibrium", "\\\\cdrightarrow", "\\\\cdleftarrow", "\\\\cdlongequal"], props: { numArgs: 1, numOptionalArgs: 1 }, handler(e, t, r) { let { parser: n, funcName: o } = e; return { type: "xArrow", mode: n.mode, label: o, body: t[0], below: r[0] } }, htmlBuilder(e, t) { const r = t.style; let n = t.havingStyle(r.sup()); const o = Ve.wrapFragment(ht(e.body, n, t), t), s = "\\x" === e.label.slice(0, 2) ? "x" : "cd"; let i; o.classes.push(s + "-arrow-pad"), e.below && (n = t.havingStyle(r.sub()), i = Ve.wrapFragment(ht(e.below, n, t), t), i.classes.push(s + "-arrow-pad")); const a = Nt(e, t), l = -t.fontMetrics().axisHeight + .5 * a.height; let h, c = -t.fontMetrics().axisHeight - .5 * a.height - .111; if ((o.depth > .25 || "\\xleftequilibrium" === e.label) && (c -= o.depth), i) { const e = -t.fontMetrics().axisHeight + i.height + .5 * a.height + .111; h = Ve.makeVList({ positionType: "individualShift", children: [{ type: "elem", elem: o, shift: c }, { type: "elem", elem: a, shift: l }, { type: "elem", elem: i, shift: e }] }, t) } else h = Ve.makeVList({ positionType: "individualShift", children: [{ type: "elem", elem: o, shift: c }, { type: "elem", elem: a, shift: l }] }, t); return h.children[0].children[0].children[1].classes.push("svg-align"), Ve.makeSpan(["mrel", "x-arrow"], [h], t) }, mathmlBuilder(e, t) { const r = Ct(e.label); let n; if (r.setAttribute("minsize", "x" === e.label.charAt(0) ? "1.75em" : "3.0em"), e.body) { const o = Lt(vt(e.body, t)); if (e.below) { const s = Lt(vt(e.below, t)); n = new gt.MathNode("munderover", [r, s, o]) } else n = new gt.MathNode("mover", [r, o]) } else if (e.below) { const o = Lt(vt(e.below, t)); n = new gt.MathNode("munder", [r, o]) } else n = Lt(), n = new gt.MathNode("mover", [r, n]); return n } }); const Dt = Ve.makeSpan; function Vt(e, t) { const r = nt(e.body, t, !0); return Dt([e.mclass], r, t) } function Pt(e, t) { let r; const n = xt(e.body, t); return "minner" === e.mclass ? r = new gt.MathNode("mpadded", n) : "mord" === e.mclass ? e.isCharacterBox ? (r = n[0], r.type = "mi") : r = new gt.MathNode("mi", n) : (e.isCharacterBox ? (r = n[0], r.type = "mo") : r = new gt.MathNode("mo", n), "mbin" === e.mclass ? (r.attributes.lspace = "0.22em", r.attributes.rspace = "0.22em") : "mpunct" === e.mclass ? (r.attributes.lspace = "0em", r.attributes.rspace = "0.17em") : "mopen" === e.mclass || "mclose" === e.mclass ? (r.attributes.lspace = "0em", r.attributes.rspace = "0em") : "minner" === e.mclass && (r.attributes.lspace = "0.0556em", r.attributes.width = "+0.1111em")), r } je({ type: "mclass", names: ["\\mathord", "\\mathbin", "\\mathrel", "\\mathopen", "\\mathclose", "\\mathpunct", "\\mathinner"], props: { numArgs: 1, primitive: !0 }, handler(e, t) { let { parser: r, funcName: n } = e; const o = t[0]; return { type: "mclass", mode: r.mode, mclass: "m" + n.slice(5), body: Ke(o), isCharacterBox: l.isCharacterBox(o) } }, htmlBuilder: Vt, mathmlBuilder: Pt }); const Ft = e => { const t = "ordgroup" === e.type && e.body.length ? e.body[0] : e; return "atom" !== t.type || "bin" !== t.family && "rel" !== t.family ? "mord" : "m" + t.family }; je({ type: "mclass", names: ["\\@binrel"], props: { numArgs: 2 }, handler(e, t) { let { parser: r } = e; return { type: "mclass", mode: r.mode, mclass: Ft(t[0]), body: Ke(t[1]), isCharacterBox: l.isCharacterBox(t[1]) } } }), je({ type: "mclass", names: ["\\stackrel", "\\overset", "\\underset"], props: { numArgs: 2 }, handler(e, t) { let { parser: r, funcName: n } = e; const o = t[1], s = t[0]; let i; i = "\\stackrel" !== n ? Ft(o) : "mrel"; const a = { type: "op", mode: o.mode, limits: !0, alwaysHandleSupSub: !0, parentIsSupSub: !1, symbol: !1, suppressBaseShift: "\\stackrel" !== n, body: Ke(o) }, h = { type: "supsub", mode: s.mode, base: a, sup: "\\underset" === n ? null : s, sub: "\\underset" === n ? s : null }; return { type: "mclass", mode: r.mode, mclass: i, body: [h], isCharacterBox: l.isCharacterBox(h) } }, htmlBuilder: Vt, mathmlBuilder: Pt }), je({ type: "pmb", names: ["\\pmb"], props: { numArgs: 1, allowedInText: !0 }, handler(e, t) { let { parser: r } = e; return { type: "pmb", mode: r.mode, mclass: Ft(t[0]), body: Ke(t[0]) } }, htmlBuilder(e, t) { const r = nt(e.body, t, !0), n = Ve.makeSpan([e.mclass], r, t); return n.style.textShadow = "0.02em 0.01em 0.04px", n }, mathmlBuilder(e, t) { const r = xt(e.body, t), n = new gt.MathNode("mstyle", r); return n.setAttribute("style", "text-shadow: 0.02em 0.01em 0.04px"), n } }); const Gt = { ">": "\\\\cdrightarrow", "<": "\\\\cdleftarrow", "=": "\\\\cdlongequal", A: "\\uparrow", V: "\\downarrow", "|": "\\Vert", ".": "no arrow" }, Ut = e => "textord" === e.type && "@" === e.text; function Yt(e, t, r) { const n = Gt[e]; switch (n) { case "\\\\cdrightarrow": case "\\\\cdleftarrow": return r.callFunction(n, [t[0]], [t[1]]); case "\\uparrow": case "\\downarrow": { const e = { type: "atom", text: n, mode: "math", family: "rel" }, o = { type: "ordgroup", mode: "math", body: [r.callFunction("\\\\cdleft", [t[0]], []), r.callFunction("\\Big", [e], []), r.callFunction("\\\\cdright", [t[1]], [])] }; return r.callFunction("\\\\cdparent", [o], []) } case "\\\\cdlongequal": return r.callFunction("\\\\cdlongequal", [], []); case "\\Vert": { const e = { type: "textord", text: "\\Vert", mode: "math" }; return r.callFunction("\\Big", [e], []) } default: return { type: "textord", text: " ", mode: "math" } } } je({ type: "cdlabel", names: ["\\\\cdleft", "\\\\cdright"], props: { numArgs: 1 }, handler(e, t) { let { parser: r, funcName: n } = e; return { type: "cdlabel", mode: r.mode, side: n.slice(4), label: t[0] } }, htmlBuilder(e, t) { const r = t.havingStyle(t.style.sup()), n = Ve.wrapFragment(ht(e.label, r, t), t); return n.classes.push("cd-label-" + e.side), n.style.bottom = F(.8 - n.depth), n.height = 0, n.depth = 0, n }, mathmlBuilder(e, t) { let r = new gt.MathNode("mrow", [vt(e.label, t)]); return r = new gt.MathNode("mpadded", [r]), r.setAttribute("width", "0"), "left" === e.side && r.setAttribute("lspace", "-1width"), r.setAttribute("voffset", "0.7em"), r = new gt.MathNode("mstyle", [r]), r.setAttribute("displaystyle", "false"), r.setAttribute("scriptlevel", "1"), r } }), je({ type: "cdlabelparent", names: ["\\\\cdparent"], props: { numArgs: 1 }, handler(e, t) { let { parser: r } = e; return { type: "cdlabelparent", mode: r.mode, fragment: t[0] } }, htmlBuilder(e, t) { const r = Ve.wrapFragment(ht(e.fragment, t), t); return r.classes.push("cd-vert-arrow"), r }, mathmlBuilder(e, t) { return new gt.MathNode("mrow", [vt(e.fragment, t)]) } }), je({ type: "textord", names: ["\\@char"], props: { numArgs: 1, allowedInText: !0 }, handler(e, t) { let { parser: r } = e; const o = qt(t[0], "ordgroup").body; let s = ""; for (let e = 0; e < o.length; e++) { s += qt(o[e], "textord").text } let i, a = parseInt(s); if (isNaN(a)) throw new n("\\@char has non-numeric argument " + s); if (a < 0 || a >= 1114111) throw new n("\\@char with invalid code point " + s); return a <= 65535 ? i = String.fromCharCode(a) : (a -= 65536, i = String.fromCharCode(55296 + (a >> 10), 56320 + (1023 & a))), { type: "textord", mode: r.mode, text: i } } }); const Xt = (e, t) => { const r = nt(e.body, t.withColor(e.color), !1); return Ve.makeFragment(r) }, Wt = (e, t) => { const r = xt(e.body, t.withColor(e.color)), n = new gt.MathNode("mstyle", r); return n.setAttribute("mathcolor", e.color), n }; je({ type: "color", names: ["\\textcolor"], props: { numArgs: 2, allowedInText: !0, argTypes: ["color", "original"] }, handler(e, t) { let { parser: r } = e; const n = qt(t[0], "color-token").color, o = t[1]; return { type: "color", mode: r.mode, color: n, body: Ke(o) } }, htmlBuilder: Xt, mathmlBuilder: Wt }), je({ type: "color", names: ["\\color"], props: { numArgs: 1, allowedInText: !0, argTypes: ["color"] }, handler(e, t) { let { parser: r, breakOnTokenText: n } = e; const o = qt(t[0], "color-token").color; r.gullet.macros.set("\\current@color", o); const s = r.parseExpression(!0, n); return { type: "color", mode: r.mode, color: o, body: s } }, htmlBuilder: Xt, mathmlBuilder: Wt }), je({ type: "cr", names: ["\\\\"], props: { numArgs: 0, numOptionalArgs: 0, allowedInText: !0 }, handler(e, t, r) { let { parser: n } = e; const o = "[" === n.gullet.future().text ? n.parseSizeGroup(!0) : null, s = !n.settings.displayMode || !n.settings.useStrictBehavior("newLineInDisplayMode", "In LaTeX, \\\\ or \\newline does nothing in display mode"); return { type: "cr", mode: n.mode, newLine: s, size: o && qt(o, "size").value } }, htmlBuilder(e, t) { const r = Ve.makeSpan(["mspace"], [], t); return e.newLine && (r.classes.push("newline"), e.size && (r.style.marginTop = F(P(e.size, t)))), r }, mathmlBuilder(e, t) { const r = new gt.MathNode("mspace"); return e.newLine && (r.setAttribute("linebreak", "newline"), e.size && r.setAttribute("height", F(P(e.size, t)))), r } }); const _t = { "\\global": "\\global", "\\long": "\\\\globallong", "\\\\globallong": "\\\\globallong", "\\def": "\\gdef", "\\gdef": "\\gdef", "\\edef": "\\xdef", "\\xdef": "\\xdef", "\\let": "\\\\globallet", "\\futurelet": "\\\\globalfuture" }, jt = e => { const t = e.text; if (/^(?:[\\{}$&#^_]|EOF)$/.test(t)) throw new n("Expected a control sequence", e); return t }, $t = (e, t, r, n) => { let o = e.gullet.macros.get(r.text); null == o && (r.noexpand = !0, o = { tokens: [r], numArgs: 0, unexpandable: !e.gullet.isExpandable(r.text) }), e.gullet.macros.set(t, o, n) }; je({ type: "internal", names: ["\\global", "\\long", "\\\\globallong"], props: { numArgs: 0, allowedInText: !0 }, handler(e) { let { parser: t, funcName: r } = e; t.consumeSpaces(); const o = t.fetch(); if (_t[o.text]) return "\\global" !== r && "\\\\globallong" !== r || (o.text = _t[o.text]), qt(t.parseFunction(), "internal"); throw new n("Invalid token after macro prefix", o) } }), je({ type: "internal", names: ["\\def", "\\gdef", "\\edef", "\\xdef"], props: { numArgs: 0, allowedInText: !0, primitive: !0 }, handler(e) { let { parser: t, funcName: r } = e, o = t.gullet.popToken(); const s = o.text; if (/^(?:[\\{}$&#^_]|EOF)$/.test(s)) throw new n("Expected a control sequence", o); let i, a = 0; const l = [[]]; for (; "{" !== t.gullet.future().text;)if (o = t.gullet.popToken(), "#" === o.text) { if ("{" === t.gullet.future().text) { i = t.gullet.future(), l[a].push("{"); break } if (o = t.gullet.popToken(), !/^[1-9]$/.test(o.text)) throw new n('Invalid argument number "' + o.text + '"'); if (parseInt(o.text) !== a + 1) throw new n('Argument number "' + o.text + '" out of order'); a++, l.push([]) } else { if ("EOF" === o.text) throw new n("Expected a macro definition"); l[a].push(o.text) } let { tokens: h } = t.gullet.consumeArg(); return i && h.unshift(i), "\\edef" !== r && "\\xdef" !== r || (h = t.gullet.expandTokens(h), h.reverse()), t.gullet.macros.set(s, { tokens: h, numArgs: a, delimiters: l }, r === _t[r]), { type: "internal", mode: t.mode } } }), je({ type: "internal", names: ["\\let", "\\\\globallet"], props: { numArgs: 0, allowedInText: !0, primitive: !0 }, handler(e) { let { parser: t, funcName: r } = e; const n = jt(t.gullet.popToken()); t.gullet.consumeSpaces(); const o = (e => { let t = e.gullet.popToken(); return "=" === t.text && (t = e.gullet.popToken(), " " === t.text && (t = e.gullet.popToken())), t })(t); return $t(t, n, o, "\\\\globallet" === r), { type: "internal", mode: t.mode } } }), je({ type: "internal", names: ["\\futurelet", "\\\\globalfuture"], props: { numArgs: 0, allowedInText: !0, primitive: !0 }, handler(e) { let { parser: t, funcName: r } = e; const n = jt(t.gullet.popToken()), o = t.gullet.popToken(), s = t.gullet.popToken(); return $t(t, n, s, "\\\\globalfuture" === r), t.gullet.pushToken(s), t.gullet.pushToken(o), { type: "internal", mode: t.mode } } }); const Zt = function (e, t, r) { const n = N(oe.math[e] && oe.math[e].replace || e, t, r); if (!n) throw new Error("Unsupported symbol " + e + " and font size " + t + "."); return n }, Kt = function (e, t, r, n) { const o = r.havingBaseStyle(t), s = Ve.makeSpan(n.concat(o.sizingClasses(r)), [e], r), i = o.sizeMultiplier / r.sizeMultiplier; return s.height *= i, s.depth *= i, s.maxFontSize = o.sizeMultiplier, s }, Jt = function (e, t, r) { const n = t.havingBaseStyle(r), o = (1 - t.sizeMultiplier / n.sizeMultiplier) * t.fontMetrics().axisHeight; e.classes.push("delimcenter"), e.style.top = F(o), e.height -= o, e.depth += o }, Qt = function (e, t, r, n, o, s) { const i = function (e, t, r, n) { return Ve.makeSymbol(e, "Size" + t + "-Regular", r, n) }(e, t, o, n), a = Kt(Ve.makeSpan(["delimsizing", "size" + t], [i], n), w.TEXT, n, s); return r && Jt(a, n, w.TEXT), a }, er = function (e, t, r) { let n; n = "Size1-Regular" === t ? "delim-size1" : "delim-size4"; return { type: "elem", elem: Ve.makeSpan(["delimsizinginner", n], [Ve.makeSpan([], [Ve.makeSymbol(e, t, r)])]) } }, tr = function (e, t, r) { const n = T["Size4-Regular"][e.charCodeAt(0)] ? T["Size4-Regular"][e.charCodeAt(0)][4] : T["Size1-Regular"][e.charCodeAt(0)][4], o = new J("inner", function (e, t) { switch (e) { case "\u239c": return "M291 0 H417 V" + t + " H291z M291 0 H417 V" + t + " H291z"; case "\u2223": return "M145 0 H188 V" + t + " H145z M145 0 H188 V" + t + " H145z"; case "\u2225": return "M145 0 H188 V" + t + " H145z M145 0 H188 V" + t + " H145zM367 0 H410 V" + t + " H367z M367 0 H410 V" + t + " H367z"; case "\u239f": return "M457 0 H583 V" + t + " H457z M457 0 H583 V" + t + " H457z"; case "\u23a2": return "M319 0 H403 V" + t + " H319z M319 0 H403 V" + t + " H319z"; case "\u23a5": return "M263 0 H347 V" + t + " H263z M263 0 H347 V" + t + " H263z"; case "\u23aa": return "M384 0 H504 V" + t + " H384z M384 0 H504 V" + t + " H384z"; case "\u23d0": return "M312 0 H355 V" + t + " H312z M312 0 H355 V" + t + " H312z"; case "\u2016": return "M257 0 H300 V" + t + " H257z M257 0 H300 V" + t + " H257zM478 0 H521 V" + t + " H478z M478 0 H521 V" + t + " H478z"; default: return "" } }(e, Math.round(1e3 * t))), s = new K([o], { width: F(n), height: F(t), style: "width:" + F(n), viewBox: "0 0 " + 1e3 * n + " " + Math.round(1e3 * t), preserveAspectRatio: "xMinYMin" }), i = Ve.makeSvgSpan([], [s], r); return i.height = t, i.style.height = F(t), i.style.width = F(n), { type: "elem", elem: i } }, rr = { type: "kern", size: -.008 }, nr = ["|", "\\lvert", "\\rvert", "\\vert"], or = ["\\|", "\\lVert", "\\rVert", "\\Vert"], sr = function (e, t, r, n, o, s) { let i, a, h, c, m = "", p = 0; i = h = c = e, a = null; let u = "Size1-Regular"; "\\uparrow" === e ? h = c = "\u23d0" : "\\Uparrow" === e ? h = c = "\u2016" : "\\downarrow" === e ? i = h = "\u23d0" : "\\Downarrow" === e ? i = h = "\u2016" : "\\updownarrow" === e ? (i = "\\uparrow", h = "\u23d0", c = "\\downarrow") : "\\Updownarrow" === e ? (i = "\\Uparrow", h = "\u2016", c = "\\Downarrow") : l.contains(nr, e) ? (h = "\u2223", m = "vert", p = 333) : l.contains(or, e) ? (h = "\u2225", m = "doublevert", p = 556) : "[" === e || "\\lbrack" === e ? (i = "\u23a1", h = "\u23a2", c = "\u23a3", u = "Size4-Regular", m = "lbrack", p = 667) : "]" === e || "\\rbrack" === e ? (i = "\u23a4", h = "\u23a5", c = "\u23a6", u = "Size4-Regular", m = "rbrack", p = 667) : "\\lfloor" === e || "\u230a" === e ? (h = i = "\u23a2", c = "\u23a3", u = "Size4-Regular", m = "lfloor", p = 667) : "\\lceil" === e || "\u2308" === e ? (i = "\u23a1", h = c = "\u23a2", u = "Size4-Regular", m = "lceil", p = 667) : "\\rfloor" === e || "\u230b" === e ? (h = i = "\u23a5", c = "\u23a6", u = "Size4-Regular", m = "rfloor", p = 667) : "\\rceil" === e || "\u2309" === e ? (i = "\u23a4", h = c = "\u23a5", u = "Size4-Regular", m = "rceil", p = 667) : "(" === e || "\\lparen" === e ? (i = "\u239b", h = "\u239c", c = "\u239d", u = "Size4-Regular", m = "lparen", p = 875) : ")" === e || "\\rparen" === e ? (i = "\u239e", h = "\u239f", c = "\u23a0", u = "Size4-Regular", m = "rparen", p = 875) : "\\{" === e || "\\lbrace" === e ? (i = "\u23a7", a = "\u23a8", c = "\u23a9", h = "\u23aa", u = "Size4-Regular") : "\\}" === e || "\\rbrace" === e ? (i = "\u23ab", a = "\u23ac", c = "\u23ad", h = "\u23aa", u = "Size4-Regular") : "\\lgroup" === e || "\u27ee" === e ? (i = "\u23a7", c = "\u23a9", h = "\u23aa", u = "Size4-Regular") : "\\rgroup" === e || "\u27ef" === e ? (i = "\u23ab", c = "\u23ad", h = "\u23aa", u = "Size4-Regular") : "\\lmoustache" === e || "\u23b0" === e ? (i = "\u23a7", c = "\u23ad", h = "\u23aa", u = "Size4-Regular") : "\\rmoustache" !== e && "\u23b1" !== e || (i = "\u23ab", c = "\u23a9", h = "\u23aa", u = "Size4-Regular"); const d = Zt(i, u, o), g = d.height + d.depth, f = Zt(h, u, o), b = f.height + f.depth, y = Zt(c, u, o), x = y.height + y.depth; let v = 0, k = 1; if (null !== a) { const e = Zt(a, u, o); v = e.height + e.depth, k = 2 } const S = g + x + v, M = S + Math.max(0, Math.ceil((t - S) / (k * b))) * k * b; let z = n.fontMetrics().axisHeight; r && (z *= n.sizeMultiplier); const A = M / 2 - z, T = []; if (m.length > 0) { const e = M - g - x, t = Math.round(1e3 * M), r = function (e, t) { switch (e) { case "lbrack": return "M403 1759 V84 H666 V0 H319 V1759 v" + t + " v1759 h347 v-84\nH403z M403 1759 V0 H319 V1759 v" + t + " v1759 h84z"; case "rbrack": return "M347 1759 V0 H0 V84 H263 V1759 v" + t + " v1759 H0 v84 H347z\nM347 1759 V0 H263 V1759 v" + t + " v1759 h84z"; case "vert": return "M145 15 v585 v" + t + " v585 c2.667,10,9.667,15,21,15\nc10,0,16.667,-5,20,-15 v-585 v" + -t + " v-585 c-2.667,-10,-9.667,-15,-21,-15\nc-10,0,-16.667,5,-20,15z M188 15 H145 v585 v" + t + " v585 h43z"; case "doublevert": return "M145 15 v585 v" + t + " v585 c2.667,10,9.667,15,21,15\nc10,0,16.667,-5,20,-15 v-585 v" + -t + " v-585 c-2.667,-10,-9.667,-15,-21,-15\nc-10,0,-16.667,5,-20,15z M188 15 H145 v585 v" + t + " v585 h43z\nM367 15 v585 v" + t + " v585 c2.667,10,9.667,15,21,15\nc10,0,16.667,-5,20,-15 v-585 v" + -t + " v-585 c-2.667,-10,-9.667,-15,-21,-15\nc-10,0,-16.667,5,-20,15z M410 15 H367 v585 v" + t + " v585 h43z"; case "lfloor": return "M319 602 V0 H403 V602 v" + t + " v1715 h263 v84 H319z\nMM319 602 V0 H403 V602 v" + t + " v1715 H319z"; case "rfloor": return "M319 602 V0 H403 V602 v" + t + " v1799 H0 v-84 H319z\nMM319 602 V0 H403 V602 v" + t + " v1715 H319z"; case "lceil": return "M403 1759 V84 H666 V0 H319 V1759 v" + t + " v602 h84z\nM403 1759 V0 H319 V1759 v" + t + " v602 h84z"; case "rceil": return "M347 1759 V0 H0 V84 H263 V1759 v" + t + " v602 h84z\nM347 1759 V0 h-84 V1759 v" + t + " v602 h84z"; case "lparen": return "M863,9c0,-2,-2,-5,-6,-9c0,0,-17,0,-17,0c-12.7,0,-19.3,0.3,-20,1\nc-5.3,5.3,-10.3,11,-15,17c-242.7,294.7,-395.3,682,-458,1162c-21.3,163.3,-33.3,349,\n-36,557 l0," + (t + 84) + "c0.2,6,0,26,0,60c2,159.3,10,310.7,24,454c53.3,528,210,\n949.7,470,1265c4.7,6,9.7,11.7,15,17c0.7,0.7,7,1,19,1c0,0,18,0,18,0c4,-4,6,-7,6,-9\nc0,-2.7,-3.3,-8.7,-10,-18c-135.3,-192.7,-235.5,-414.3,-300.5,-665c-65,-250.7,-102.5,\n-544.7,-112.5,-882c-2,-104,-3,-167,-3,-189\nl0,-" + (t + 92) + "c0,-162.7,5.7,-314,17,-454c20.7,-272,63.7,-513,129,-723c65.3,\n-210,155.3,-396.3,270,-559c6.7,-9.3,10,-15.3,10,-18z"; case "rparen": return "M76,0c-16.7,0,-25,3,-25,9c0,2,2,6.3,6,13c21.3,28.7,42.3,60.3,\n63,95c96.7,156.7,172.8,332.5,228.5,527.5c55.7,195,92.8,416.5,111.5,664.5\nc11.3,139.3,17,290.7,17,454c0,28,1.7,43,3.3,45l0," + (t + 9) + "\nc-3,4,-3.3,16.7,-3.3,38c0,162,-5.7,313.7,-17,455c-18.7,248,-55.8,469.3,-111.5,664\nc-55.7,194.7,-131.8,370.3,-228.5,527c-20.7,34.7,-41.7,66.3,-63,95c-2,3.3,-4,7,-6,11\nc0,7.3,5.7,11,17,11c0,0,11,0,11,0c9.3,0,14.3,-0.3,15,-1c5.3,-5.3,10.3,-11,15,-17\nc242.7,-294.7,395.3,-681.7,458,-1161c21.3,-164.7,33.3,-350.7,36,-558\nl0,-" + (t + 144) + "c-2,-159.3,-10,-310.7,-24,-454c-53.3,-528,-210,-949.7,\n-470,-1265c-4.7,-6,-9.7,-11.7,-15,-17c-0.7,-0.7,-6.7,-1,-18,-1z"; default: throw new Error("Unknown stretchy delimiter.") } }(m, Math.round(1e3 * e)), o = new J(m, r), s = (p / 1e3).toFixed(3) + "em", i = (t / 1e3).toFixed(3) + "em", a = new K([o], { width: s, height: i, viewBox: "0 0 " + p + " " + t }), l = Ve.makeSvgSpan([], [a], n); l.height = t / 1e3, l.style.width = s, l.style.height = i, T.push({ type: "elem", elem: l }) } else { if (T.push(er(c, u, o)), T.push(rr), null === a) { const e = M - g - x + .016; T.push(tr(h, e, n)) } else { const e = (M - g - x - v) / 2 + .016; T.push(tr(h, e, n)), T.push(rr), T.push(er(a, u, o)), T.push(rr), T.push(tr(h, e, n)) } T.push(rr), T.push(er(i, u, o)) } const B = n.havingBaseStyle(w.TEXT), C = Ve.makeVList({ positionType: "bottom", positionData: A, children: T }, B); return Kt(Ve.makeSpan(["delimsizing", "mult"], [C], B), w.TEXT, n, s) }, ir = .08, ar = function (e, t, r, n, o) { const s = function (e, t, r) { t *= 1e3; let n = ""; switch (e) { case "sqrtMain": n = function (e, t) { return "M95," + (622 + e + t) + "\nc-2.7,0,-7.17,-2.7,-13.5,-8c-5.8,-5.3,-9.5,-10,-9.5,-14\nc0,-2,0.3,-3.3,1,-4c1.3,-2.7,23.83,-20.7,67.5,-54\nc44.2,-33.3,65.8,-50.3,66.5,-51c1.3,-1.3,3,-2,5,-2c4.7,0,8.7,3.3,12,10\ns173,378,173,378c0.7,0,35.3,-71,104,-213c68.7,-142,137.5,-285,206.5,-429\nc69,-144,104.5,-217.7,106.5,-221\nl" + e / 2.075 + " -" + e + "\nc5.3,-9.3,12,-14,20,-14\nH400000v" + (40 + e) + "H845.2724\ns-225.272,467,-225.272,467s-235,486,-235,486c-2.7,4.7,-9,7,-19,7\nc-6,0,-10,-1,-12,-3s-194,-422,-194,-422s-65,47,-65,47z\nM" + (834 + e) + " " + t + "h400000v" + (40 + e) + "h-400000z" }(t, M); break; case "sqrtSize1": n = function (e, t) { return "M263," + (601 + e + t) + "c0.7,0,18,39.7,52,119\nc34,79.3,68.167,158.7,102.5,238c34.3,79.3,51.8,119.3,52.5,120\nc340,-704.7,510.7,-1060.3,512,-1067\nl" + e / 2.084 + " -" + e + "\nc4.7,-7.3,11,-11,19,-11\nH40000v" + (40 + e) + "H1012.3\ns-271.3,567,-271.3,567c-38.7,80.7,-84,175,-136,283c-52,108,-89.167,185.3,-111.5,232\nc-22.3,46.7,-33.8,70.3,-34.5,71c-4.7,4.7,-12.3,7,-23,7s-12,-1,-12,-1\ns-109,-253,-109,-253c-72.7,-168,-109.3,-252,-110,-252c-10.7,8,-22,16.7,-34,26\nc-22,17.3,-33.3,26,-34,26s-26,-26,-26,-26s76,-59,76,-59s76,-60,76,-60z\nM" + (1001 + e) + " " + t + "h400000v" + (40 + e) + "h-400000z" }(t, M); break; case "sqrtSize2": n = function (e, t) { return "M983 " + (10 + e + t) + "\nl" + e / 3.13 + " -" + e + "\nc4,-6.7,10,-10,18,-10 H400000v" + (40 + e) + "\nH1013.1s-83.4,268,-264.1,840c-180.7,572,-277,876.3,-289,913c-4.7,4.7,-12.7,7,-24,7\ns-12,0,-12,0c-1.3,-3.3,-3.7,-11.7,-7,-25c-35.3,-125.3,-106.7,-373.3,-214,-744\nc-10,12,-21,25,-33,39s-32,39,-32,39c-6,-5.3,-15,-14,-27,-26s25,-30,25,-30\nc26.7,-32.7,52,-63,76,-91s52,-60,52,-60s208,722,208,722\nc56,-175.3,126.3,-397.3,211,-666c84.7,-268.7,153.8,-488.2,207.5,-658.5\nc53.7,-170.3,84.5,-266.8,92.5,-289.5z\nM" + (1001 + e) + " " + t + "h400000v" + (40 + e) + "h-400000z" }(t, M); break; case "sqrtSize3": n = function (e, t) { return "M424," + (2398 + e + t) + "\nc-1.3,-0.7,-38.5,-172,-111.5,-514c-73,-342,-109.8,-513.3,-110.5,-514\nc0,-2,-10.7,14.3,-32,49c-4.7,7.3,-9.8,15.7,-15.5,25c-5.7,9.3,-9.8,16,-12.5,20\ns-5,7,-5,7c-4,-3.3,-8.3,-7.7,-13,-13s-13,-13,-13,-13s76,-122,76,-122s77,-121,77,-121\ns209,968,209,968c0,-2,84.7,-361.7,254,-1079c169.3,-717.3,254.7,-1077.7,256,-1081\nl" + e / 4.223 + " -" + e + "c4,-6.7,10,-10,18,-10 H400000\nv" + (40 + e) + "H1014.6\ns-87.3,378.7,-272.6,1166c-185.3,787.3,-279.3,1182.3,-282,1185\nc-2,6,-10,9,-24,9\nc-8,0,-12,-0.7,-12,-2z M" + (1001 + e) + " " + t + "\nh400000v" + (40 + e) + "h-400000z" }(t, M); break; case "sqrtSize4": n = function (e, t) { return "M473," + (2713 + e + t) + "\nc339.3,-1799.3,509.3,-2700,510,-2702 l" + e / 5.298 + " -" + e + "\nc3.3,-7.3,9.3,-11,18,-11 H400000v" + (40 + e) + "H1017.7\ns-90.5,478,-276.2,1466c-185.7,988,-279.5,1483,-281.5,1485c-2,6,-10,9,-24,9\nc-8,0,-12,-0.7,-12,-2c0,-1.3,-5.3,-32,-16,-92c-50.7,-293.3,-119.7,-693.3,-207,-1200\nc0,-1.3,-5.3,8.7,-16,30c-10.7,21.3,-21.3,42.7,-32,64s-16,33,-16,33s-26,-26,-26,-26\ns76,-153,76,-153s77,-151,77,-151c0.7,0.7,35.7,202,105,604c67.3,400.7,102,602.7,104,\n606zM" + (1001 + e) + " " + t + "h400000v" + (40 + e) + "H1017.7z" }(t, M); break; case "sqrtTall": n = function (e, t, r) { return "M702 " + (e + t) + "H400000" + (40 + e) + "\nH742v" + (r - 54 - t - e) + "l-4 4-4 4c-.667.7 -2 1.5-4 2.5s-4.167 1.833-6.5 2.5-5.5 1-9.5 1\nh-12l-28-84c-16.667-52-96.667 -294.333-240-727l-212 -643 -85 170\nc-4-3.333-8.333-7.667-13 -13l-13-13l77-155 77-156c66 199.333 139 419.667\n219 661 l218 661zM702 " + t + "H400000v" + (40 + e) + "H742z" }(t, M, r) }return n }(e, n, r), i = new J(e, s), a = new K([i], { width: "400em", height: F(t), viewBox: "0 0 400000 " + r, preserveAspectRatio: "xMinYMin slice" }); return Ve.makeSvgSpan(["hide-tail"], [a], o) }, lr = ["(", "\\lparen", ")", "\\rparen", "[", "\\lbrack", "]", "\\rbrack", "\\{", "\\lbrace", "\\}", "\\rbrace", "\\lfloor", "\\rfloor", "\u230a", "\u230b", "\\lceil", "\\rceil", "\u2308", "\u2309", "\\surd"], hr = ["\\uparrow", "\\downarrow", "\\updownarrow", "\\Uparrow", "\\Downarrow", "\\Updownarrow", "|", "\\|", "\\vert", "\\Vert", "\\lvert", "\\rvert", "\\lVert", "\\rVert", "\\lgroup", "\\rgroup", "\u27ee", "\u27ef", "\\lmoustache", "\\rmoustache", "\u23b0", "\u23b1"], cr = ["<", ">", "\\langle", "\\rangle", "/", "\\backslash", "\\lt", "\\gt"], mr = [0, 1.2, 1.8, 2.4, 3], pr = [{ type: "small", style: w.SCRIPTSCRIPT }, { type: "small", style: w.SCRIPT }, { type: "small", style: w.TEXT }, { type: "large", size: 1 }, { type: "large", size: 2 }, { type: "large", size: 3 }, { type: "large", size: 4 }], ur = [{ type: "small", style: w.SCRIPTSCRIPT }, { type: "small", style: w.SCRIPT }, { type: "small", style: w.TEXT }, { type: "stack" }], dr = [{ type: "small", style: w.SCRIPTSCRIPT }, { type: "small", style: w.SCRIPT }, { type: "small", style: w.TEXT }, { type: "large", size: 1 }, { type: "large", size: 2 }, { type: "large", size: 3 }, { type: "large", size: 4 }, { type: "stack" }], gr = function (e) { if ("small" === e.type) return "Main-Regular"; if ("large" === e.type) return "Size" + e.size + "-Regular"; if ("stack" === e.type) return "Size4-Regular"; throw new Error("Add support for delim type '" + e.type + "' here.") }, fr = function (e, t, r, n) { for (let o = Math.min(2, 3 - n.style.size); o < r.length && "stack" !== r[o].type; o++) { const s = Zt(e, gr(r[o]), "math"); let i = s.height + s.depth; if ("small" === r[o].type) { i *= n.havingBaseStyle(r[o].style).sizeMultiplier } if (i > t) return r[o] } return r[r.length - 1] }, br = function (e, t, r, n, o, s) { let i; "<" === e || "\\lt" === e || "\u27e8" === e ? e = "\\langle" : ">" !== e && "\\gt" !== e && "\u27e9" !== e || (e = "\\rangle"), i = l.contains(cr, e) ? pr : l.contains(lr, e) ? dr : ur; const a = fr(e, t, i, n); return "small" === a.type ? function (e, t, r, n, o, s) { const i = Ve.makeSymbol(e, "Main-Regular", o, n), a = Kt(i, t, n, s); return r && Jt(a, n, t), a }(e, a.style, r, n, o, s) : "large" === a.type ? Qt(e, a.size, r, n, o, s) : sr(e, t, r, n, o, s) }; var yr = { sqrtImage: function (e, t) { const r = t.havingBaseSizing(), n = fr("\\surd", e * r.sizeMultiplier, dr, r); let o = r.sizeMultiplier; const s = Math.max(0, t.minRuleThickness - t.fontMetrics().sqrtRuleThickness); let i, a, l = 0, h = 0, c = 0; return "small" === n.type ? (c = 1e3 + 1e3 * s + 80, e < 1 ? o = 1 : e < 1.4 && (o = .7), l = (1 + s + ir) / o, h = (1 + s) / o, i = ar("sqrtMain", l, c, s, t), i.style.minWidth = "0.853em", a = .833 / o) : "large" === n.type ? (c = 1080 * mr[n.size], h = (mr[n.size] + s) / o, l = (mr[n.size] + s + ir) / o, i = ar("sqrtSize" + n.size, l, c, s, t), i.style.minWidth = "1.02em", a = 1 / o) : (l = e + s + ir, h = e + s, c = Math.floor(1e3 * e + s) + 80, i = ar("sqrtTall", l, c, s, t), i.style.minWidth = "0.742em", a = 1.056), i.height = h, i.style.height = F(l), { span: i, advanceWidth: a, ruleWidth: (t.fontMetrics().sqrtRuleThickness + s) * o } }, sizedDelim: function (e, t, r, o, s) { if ("<" === e || "\\lt" === e || "\u27e8" === e ? e = "\\langle" : ">" !== e && "\\gt" !== e && "\u27e9" !== e || (e = "\\rangle"), l.contains(lr, e) || l.contains(cr, e)) return Qt(e, t, !1, r, o, s); if (l.contains(hr, e)) return sr(e, mr[t], !1, r, o, s); throw new n("Illegal delimiter: '" + e + "'") }, sizeToMaxHeight: mr, customSizedDelim: br, leftRightDelim: function (e, t, r, n, o, s) { const i = n.fontMetrics().axisHeight * n.sizeMultiplier, a = 5 / n.fontMetrics().ptPerEm, l = Math.max(t - i, r + i), h = Math.max(l / 500 * 901, 2 * l - a); return br(e, h, !0, n, o, s) } }; const xr = { "\\bigl": { mclass: "mopen", size: 1 }, "\\Bigl": { mclass: "mopen", size: 2 }, "\\biggl": { mclass: "mopen", size: 3 }, "\\Biggl": { mclass: "mopen", size: 4 }, "\\bigr": { mclass: "mclose", size: 1 }, "\\Bigr": { mclass: "mclose", size: 2 }, "\\biggr": { mclass: "mclose", size: 3 }, "\\Biggr": { mclass: "mclose", size: 4 }, "\\bigm": { mclass: "mrel", size: 1 }, "\\Bigm": { mclass: "mrel", size: 2 }, "\\biggm": { mclass: "mrel", size: 3 }, "\\Biggm": { mclass: "mrel", size: 4 }, "\\big": { mclass: "mord", size: 1 }, "\\Big": { mclass: "mord", size: 2 }, "\\bigg": { mclass: "mord", size: 3 }, "\\Bigg": { mclass: "mord", size: 4 } }, wr = ["(", "\\lparen", ")", "\\rparen", "[", "\\lbrack", "]", "\\rbrack", "\\{", "\\lbrace", "\\}", "\\rbrace", "\\lfloor", "\\rfloor", "\u230a", "\u230b", "\\lceil", "\\rceil", "\u2308", "\u2309", "<", ">", "\\langle", "\u27e8", "\\rangle", "\u27e9", "\\lt", "\\gt", "\\lvert", "\\rvert", "\\lVert", "\\rVert", "\\lgroup", "\\rgroup", "\u27ee", "\u27ef", "\\lmoustache", "\\rmoustache", "\u23b0", "\u23b1", "/", "\\backslash", "|", "\\vert", "\\|", "\\Vert", "\\uparrow", "\\Uparrow", "\\downarrow", "\\Downarrow", "\\updownarrow", "\\Updownarrow", "."]; function vr(e, t) { const r = Rt(e); if (r && l.contains(wr, r.text)) return r; throw new n(r ? "Invalid delimiter '" + r.text + "' after '" + t.funcName + "'" : "Invalid delimiter type '" + e.type + "'", e) } function kr(e) { if (!e.body) throw new Error("Bug: The leftright ParseNode wasn't fully parsed.") } je({ type: "delimsizing", names: ["\\bigl", "\\Bigl", "\\biggl", "\\Biggl", "\\bigr", "\\Bigr", "\\biggr", "\\Biggr", "\\bigm", "\\Bigm", "\\biggm", "\\Biggm", "\\big", "\\Big", "\\bigg", "\\Bigg"], props: { numArgs: 1, argTypes: ["primitive"] }, handler: (e, t) => { const r = vr(t[0], e); return { type: "delimsizing", mode: e.parser.mode, size: xr[e.funcName].size, mclass: xr[e.funcName].mclass, delim: r.text } }, htmlBuilder: (e, t) => "." === e.delim ? Ve.makeSpan([e.mclass]) : yr.sizedDelim(e.delim, e.size, t, e.mode, [e.mclass]), mathmlBuilder: e => { const t = []; "." !== e.delim && t.push(ft(e.delim, e.mode)); const r = new gt.MathNode("mo", t); "mopen" === e.mclass || "mclose" === e.mclass ? r.setAttribute("fence", "true") : r.setAttribute("fence", "false"), r.setAttribute("stretchy", "true"); const n = F(yr.sizeToMaxHeight[e.size]); return r.setAttribute("minsize", n), r.setAttribute("maxsize", n), r } }), je({ type: "leftright-right", names: ["\\right"], props: { numArgs: 1, primitive: !0 }, handler: (e, t) => { const r = e.parser.gullet.macros.get("\\current@color"); if (r && "string" != typeof r) throw new n("\\current@color set to non-string in \\right"); return { type: "leftright-right", mode: e.parser.mode, delim: vr(t[0], e).text, color: r } } }), je({ type: "leftright", names: ["\\left"], props: { numArgs: 1, primitive: !0 }, handler: (e, t) => { const r = vr(t[0], e), n = e.parser; ++n.leftrightDepth; const o = n.parseExpression(!1); --n.leftrightDepth, n.expect("\\right", !1); const s = qt(n.parseFunction(), "leftright-right"); return { type: "leftright", mode: n.mode, body: o, left: r.text, right: s.delim, rightColor: s.color } }, htmlBuilder: (e, t) => { kr(e); const r = nt(e.body, t, !0, ["mopen", "mclose"]); let n, o, s = 0, i = 0, a = !1; for (let e = 0; e < r.length; e++)r[e].isMiddle ? a = !0 : (s = Math.max(r[e].height, s), i = Math.max(r[e].depth, i)); if (s *= t.sizeMultiplier, i *= t.sizeMultiplier, n = "." === e.left ? lt(t, ["mopen"]) : yr.leftRightDelim(e.left, s, i, t, e.mode, ["mopen"]), r.unshift(n), a) for (let t = 1; t < r.length; t++) { const n = r[t].isMiddle; n && (r[t] = yr.leftRightDelim(n.delim, s, i, n.options, e.mode, [])) } if ("." === e.right) o = lt(t, ["mclose"]); else { const r = e.rightColor ? t.withColor(e.rightColor) : t; o = yr.leftRightDelim(e.right, s, i, r, e.mode, ["mclose"]) } return r.push(o), Ve.makeSpan(["minner"], r, t) }, mathmlBuilder: (e, t) => { kr(e); const r = xt(e.body, t); if ("." !== e.left) { const t = new gt.MathNode("mo", [ft(e.left, e.mode)]); t.setAttribute("fence", "true"), r.unshift(t) } if ("." !== e.right) { const t = new gt.MathNode("mo", [ft(e.right, e.mode)]); t.setAttribute("fence", "true"), e.rightColor && t.setAttribute("mathcolor", e.rightColor), r.push(t) } return bt(r) } }), je({ type: "middle", names: ["\\middle"], props: { numArgs: 1, primitive: !0 }, handler: (e, t) => { const r = vr(t[0], e); if (!e.parser.leftrightDepth) throw new n("\\middle without preceding \\left", r); return { type: "middle", mode: e.parser.mode, delim: r.text } }, htmlBuilder: (e, t) => { let r; if ("." === e.delim) r = lt(t, []); else { r = yr.sizedDelim(e.delim, 1, t, e.mode, []); const n = { delim: e.delim, options: t }; r.isMiddle = n } return r }, mathmlBuilder: (e, t) => { const r = "\\vert" === e.delim || "|" === e.delim ? ft("|", "text") : ft(e.delim, e.mode), n = new gt.MathNode("mo", [r]); return n.setAttribute("fence", "true"), n.setAttribute("lspace", "0.05em"), n.setAttribute("rspace", "0.05em"), n } }); const Sr = (e, t) => { const r = Ve.wrapFragment(ht(e.body, t), t), n = e.label.slice(1); let o, s = t.sizeMultiplier, i = 0; const a = l.isCharacterBox(e.body); if ("sout" === n) o = Ve.makeSpan(["stretchy", "sout"]), o.height = t.fontMetrics().defaultRuleThickness / s, i = -.5 * t.fontMetrics().xHeight; else if ("phase" === n) { const e = P({ number: .6, unit: "pt" }, t), n = P({ number: .35, unit: "ex" }, t); s /= t.havingBaseSizing().sizeMultiplier; const a = r.height + r.depth + e + n; r.style.paddingLeft = F(a / 2 + e); const l = Math.floor(1e3 * a * s), c = "M400000 " + (h = l) + " H0 L" + h / 2 + " 0 l65 45 L145 " + (h - 80) + " H400000z", m = new K([new J("phase", c)], { width: "400em", height: F(l / 1e3), viewBox: "0 0 400000 " + l, preserveAspectRatio: "xMinYMin slice" }); o = Ve.makeSvgSpan(["hide-tail"], [m], t), o.style.height = F(a), i = r.depth + e + n } else { /cancel/.test(n) ? a || r.classes.push("cancel-pad") : "angl" === n ? r.classes.push("anglpad") : r.classes.push("boxpad"); let s = 0, l = 0, h = 0; /box/.test(n) ? (h = Math.max(t.fontMetrics().fboxrule, t.minRuleThickness), s = t.fontMetrics().fboxsep + ("colorbox" === n ? 0 : h), l = s) : "angl" === n ? (h = Math.max(t.fontMetrics().defaultRuleThickness, t.minRuleThickness), s = 4 * h, l = Math.max(0, .25 - r.depth)) : (s = a ? .2 : 0, l = s), o = Bt(r, n, s, l, t), /fbox|boxed|fcolorbox/.test(n) ? (o.style.borderStyle = "solid", o.style.borderWidth = F(h)) : "angl" === n && .049 !== h && (o.style.borderTopWidth = F(h), o.style.borderRightWidth = F(h)), i = r.depth + l, e.backgroundColor && (o.style.backgroundColor = e.backgroundColor, e.borderColor && (o.style.borderColor = e.borderColor)) } var h; let c; if (e.backgroundColor) c = Ve.makeVList({ positionType: "individualShift", children: [{ type: "elem", elem: o, shift: i }, { type: "elem", elem: r, shift: 0 }] }, t); else { const e = /cancel|phase/.test(n) ? ["svg-align"] : []; c = Ve.makeVList({ positionType: "individualShift", children: [{ type: "elem", elem: r, shift: 0 }, { type: "elem", elem: o, shift: i, wrapperClasses: e }] }, t) } return /cancel/.test(n) && (c.height = r.height, c.depth = r.depth), /cancel/.test(n) && !a ? Ve.makeSpan(["mord", "cancel-lap"], [c], t) : Ve.makeSpan(["mord"], [c], t) }, Mr = (e, t) => { let r = 0; const n = new gt.MathNode(e.label.indexOf("colorbox") > -1 ? "mpadded" : "menclose", [vt(e.body, t)]); switch (e.label) { case "\\cancel": n.setAttribute("notation", "updiagonalstrike"); break; case "\\bcancel": n.setAttribute("notation", "downdiagonalstrike"); break; case "\\phase": n.setAttribute("notation", "phasorangle"); break; case "\\sout": n.setAttribute("notation", "horizontalstrike"); break; case "\\fbox": n.setAttribute("notation", "box"); break; case "\\angl": n.setAttribute("notation", "actuarial"); break; case "\\fcolorbox": case "\\colorbox": if (r = t.fontMetrics().fboxsep * t.fontMetrics().ptPerEm, n.setAttribute("width", "+" + 2 * r + "pt"), n.setAttribute("height", "+" + 2 * r + "pt"), n.setAttribute("lspace", r + "pt"), n.setAttribute("voffset", r + "pt"), "\\fcolorbox" === e.label) { const r = Math.max(t.fontMetrics().fboxrule, t.minRuleThickness); n.setAttribute("style", "border: " + r + "em solid " + String(e.borderColor)) } break; case "\\xcancel": n.setAttribute("notation", "updiagonalstrike downdiagonalstrike") }return e.backgroundColor && n.setAttribute("mathbackground", e.backgroundColor), n }; je({ type: "enclose", names: ["\\colorbox"], props: { numArgs: 2, allowedInText: !0, argTypes: ["color", "text"] }, handler(e, t, r) { let { parser: n, funcName: o } = e; const s = qt(t[0], "color-token").color, i = t[1]; return { type: "enclose", mode: n.mode, label: o, backgroundColor: s, body: i } }, htmlBuilder: Sr, mathmlBuilder: Mr }), je({ type: "enclose", names: ["\\fcolorbox"], props: { numArgs: 3, allowedInText: !0, argTypes: ["color", "color", "text"] }, handler(e, t, r) { let { parser: n, funcName: o } = e; const s = qt(t[0], "color-token").color, i = qt(t[1], "color-token").color, a = t[2]; return { type: "enclose", mode: n.mode, label: o, backgroundColor: i, borderColor: s, body: a } }, htmlBuilder: Sr, mathmlBuilder: Mr }), je({ type: "enclose", names: ["\\fbox"], props: { numArgs: 1, argTypes: ["hbox"], allowedInText: !0 }, handler(e, t) { let { parser: r } = e; return { type: "enclose", mode: r.mode, label: "\\fbox", body: t[0] } } }), je({ type: "enclose", names: ["\\cancel", "\\bcancel", "\\xcancel", "\\sout", "\\phase"], props: { numArgs: 1 }, handler(e, t) { let { parser: r, funcName: n } = e; const o = t[0]; return { type: "enclose", mode: r.mode, label: n, body: o } }, htmlBuilder: Sr, mathmlBuilder: Mr }), je({ type: "enclose", names: ["\\angl"], props: { numArgs: 1, argTypes: ["hbox"], allowedInText: !1 }, handler(e, t) { let { parser: r } = e; return { type: "enclose", mode: r.mode, label: "\\angl", body: t[0] } } }); const zr = {}; function Ar(e) { let { type: t, names: r, props: n, handler: o, htmlBuilder: s, mathmlBuilder: i } = e; const a = { type: t, numArgs: n.numArgs || 0, allowedInText: !1, numOptionalArgs: 0, handler: o }; for (let e = 0; e < r.length; ++e)zr[r[e]] = a; s && (We[t] = s), i && (_e[t] = i) } const Tr = {}; function Br(e, t) { Tr[e] = t } class Cr { constructor(e, t, r) { this.lexer = void 0, this.start = void 0, this.end = void 0, this.lexer = e, this.start = t, this.end = r } static range(e, t) { return t ? e && e.loc && t.loc && e.loc.lexer === t.loc.lexer ? new Cr(e.loc.lexer, e.loc.start, t.loc.end) : null : e && e.loc } } class Nr { constructor(e, t) { this.text = void 0, this.loc = void 0, this.noexpand = void 0, this.treatAsRelax = void 0, this.text = e, this.loc = t } range(e, t) { return new Nr(t, Cr.range(this, e)) } } function qr(e) { const t = []; e.consumeSpaces(); let r = e.fetch().text; for ("\\relax" === r && (e.consume(), e.consumeSpaces(), r = e.fetch().text); "\\hline" === r || "\\hdashline" === r;)e.consume(), t.push("\\hdashline" === r), e.consumeSpaces(), r = e.fetch().text; return t } const Ir = e => { if (!e.parser.settings.displayMode) throw new n("{" + e.envName + "} can be used only in display mode.") }; function Rr(e) { if (-1 === e.indexOf("ed")) return -1 === e.indexOf("*") } function Hr(e, t, r) { let { hskipBeforeAndAfter: o, addJot: s, cols: i, arraystretch: a, colSeparationType: l, autoTag: h, singleRow: c, emptySingleRow: m, maxNumCols: p, leqno: u } = t; if (e.gullet.beginGroup(), c || e.gullet.macros.set("\\cr", "\\\\\\relax"), !a) { const t = e.gullet.expandMacroAsText("\\arraystretch"); if (null == t) a = 1; else if (a = parseFloat(t), !a || a < 0) throw new n("Invalid \\arraystretch: " + t) } e.gullet.beginGroup(); let d = []; const g = [d], f = [], b = [], y = null != h ? [] : void 0; function x() { h && e.gullet.macros.set("\\@eqnsw", "1", !0) } function w() { y && (e.gullet.macros.get("\\df@tag") ? (y.push(e.subparse([new Nr("\\df@tag")])), e.gullet.macros.set("\\df@tag", void 0, !0)) : y.push(Boolean(h) && "1" === e.gullet.macros.get("\\@eqnsw"))) } for (x(), b.push(qr(e)); ;) { let t = e.parseExpression(!1, c ? "\\end" : "\\\\"); e.gullet.endGroup(), e.gullet.beginGroup(), t = { type: "ordgroup", mode: e.mode, body: t }, r && (t = { type: "styling", mode: e.mode, style: r, body: [t] }), d.push(t); const o = e.fetch().text; if ("&" === o) { if (p && d.length === p) { if (c || l) throw new n("Too many tab characters: &", e.nextToken); e.settings.reportNonstrict("textEnv", "Too few columns specified in the {array} column argument.") } e.consume() } else { if ("\\end" === o) { w(), 1 === d.length && "styling" === t.type && 0 === t.body[0].body.length && (g.length > 1 || !m) && g.pop(), b.length < g.length + 1 && b.push([]); break } if ("\\\\" !== o) throw new n("Expected & or \\\\ or \\cr or \\end", e.nextToken); { let t; e.consume(), " " !== e.gullet.future().text && (t = e.parseSizeGroup(!0)), f.push(t ? t.value : null), w(), b.push(qr(e)), d = [], g.push(d), x() } } } return e.gullet.endGroup(), e.gullet.endGroup(), { type: "array", mode: e.mode, addJot: s, arraystretch: a, body: g, cols: i, rowGaps: f, hskipBeforeAndAfter: o, hLinesBeforeRow: b, colSeparationType: l, tags: y, leqno: u } } function Or(e) { return "d" === e.slice(0, 1) ? "display" : "text" } const Er = function (e, t) { let r, o; const s = e.body.length, i = e.hLinesBeforeRow; let a = 0, h = new Array(s); const c = [], m = Math.max(t.fontMetrics().arrayRuleWidth, t.minRuleThickness), p = 1 / t.fontMetrics().ptPerEm; let u = 5 * p; if (e.colSeparationType && "small" === e.colSeparationType) { u = t.havingStyle(w.SCRIPT).sizeMultiplier / t.sizeMultiplier * .2778 } const d = "CD" === e.colSeparationType ? P({ number: 3, unit: "ex" }, t) : 12 * p, g = 3 * p, f = e.arraystretch * d, b = .7 * f, y = .3 * f; let x = 0; function v(e) { for (let t = 0; t < e.length; ++t)t > 0 && (x += .25), c.push({ pos: x, isDashed: e[t] }) } for (v(i[0]), r = 0; r < e.body.length; ++r) { const n = e.body[r]; let s = b, l = y; a < n.length && (a = n.length); const c = new Array(n.length); for (o = 0; o < n.length; ++o) { const e = ht(n[o], t); l < e.depth && (l = e.depth), s < e.height && (s = e.height), c[o] = e } const m = e.rowGaps[r]; let p = 0; m && (p = P(m, t), p > 0 && (p += y, l < p && (l = p), p = 0)), e.addJot && (l += g), c.height = s, c.depth = l, x += s, c.pos = x, x += l + p, h[r] = c, v(i[r + 1]) } const k = x / 2 + t.fontMetrics().axisHeight, S = e.cols || [], M = []; let z, A; const T = []; if (e.tags && e.tags.some((e => e))) for (r = 0; r < s; ++r) { const n = h[r], o = n.pos - k, s = e.tags[r]; let i; i = !0 === s ? Ve.makeSpan(["eqn-num"], [], t) : !1 === s ? Ve.makeSpan([], [], t) : Ve.makeSpan([], nt(s, t, !0), t), i.depth = n.depth, i.height = n.height, T.push({ type: "elem", elem: i, shift: o }) } for (o = 0, A = 0; o < a || A < S.length; ++o, ++A) { let i, c = S[A] || {}, p = !0; for (; "separator" === c.type;) { if (p || (z = Ve.makeSpan(["arraycolsep"], []), z.style.width = F(t.fontMetrics().doubleRuleSep), M.push(z)), "|" !== c.separator && ":" !== c.separator) throw new n("Invalid separator type: " + c.separator); { const e = "|" === c.separator ? "solid" : "dashed", r = Ve.makeSpan(["vertical-separator"], [], t); r.style.height = F(x), r.style.borderRightWidth = F(m), r.style.borderRightStyle = e, r.style.margin = "0 " + F(-m / 2); const n = x - k; n && (r.style.verticalAlign = F(-n)), M.push(r) } A++, c = S[A] || {}, p = !1 } if (o >= a) continue; (o > 0 || e.hskipBeforeAndAfter) && (i = l.deflt(c.pregap, u), 0 !== i && (z = Ve.makeSpan(["arraycolsep"], []), z.style.width = F(i), M.push(z))); let d = []; for (r = 0; r < s; ++r) { const e = h[r], t = e[o]; if (!t) continue; const n = e.pos - k; t.depth = e.depth, t.height = e.height, d.push({ type: "elem", elem: t, shift: n }) } d = Ve.makeVList({ positionType: "individualShift", children: d }, t), d = Ve.makeSpan(["col-align-" + (c.align || "c")], [d]), M.push(d), (o < a - 1 || e.hskipBeforeAndAfter) && (i = l.deflt(c.postgap, u), 0 !== i && (z = Ve.makeSpan(["arraycolsep"], []), z.style.width = F(i), M.push(z))) } if (h = Ve.makeSpan(["mtable"], M), c.length > 0) { const e = Ve.makeLineSpan("hline", t, m), r = Ve.makeLineSpan("hdashline", t, m), n = [{ type: "elem", elem: h, shift: 0 }]; for (; c.length > 0;) { const t = c.pop(), o = t.pos - k; t.isDashed ? n.push({ type: "elem", elem: r, shift: o }) : n.push({ type: "elem", elem: e, shift: o }) } h = Ve.makeVList({ positionType: "individualShift", children: n }, t) } if (0 === T.length) return Ve.makeSpan(["mord"], [h], t); { let e = Ve.makeVList({ positionType: "individualShift", children: T }, t); return e = Ve.makeSpan(["tag"], [e], t), Ve.makeFragment([h, e]) } }, Lr = { c: "center ", l: "left ", r: "right " }, Dr = function (e, t) { const r = [], n = new gt.MathNode("mtd", [], ["mtr-glue"]), o = new gt.MathNode("mtd", [], ["mml-eqn-num"]); for (let s = 0; s < e.body.length; s++) { const i = e.body[s], a = []; for (let e = 0; e < i.length; e++)a.push(new gt.MathNode("mtd", [vt(i[e], t)])); e.tags && e.tags[s] && (a.unshift(n), a.push(n), e.leqno ? a.unshift(o) : a.push(o)), r.push(new gt.MathNode("mtr", a)) } let s = new gt.MathNode("mtable", r); const i = .5 === e.arraystretch ? .1 : .16 + e.arraystretch - 1 + (e.addJot ? .09 : 0); s.setAttribute("rowspacing", F(i)); let a = "", l = ""; if (e.cols && e.cols.length > 0) { const t = e.cols; let r = "", n = !1, o = 0, i = t.length; "separator" === t[0].type && (a += "top ", o = 1), "separator" === t[t.length - 1].type && (a += "bottom ", i -= 1); for (let e = o; e < i; e++)"align" === t[e].type ? (l += Lr[t[e].align], n && (r += "none "), n = !0) : "separator" === t[e].type && n && (r += "|" === t[e].separator ? "solid " : "dashed ", n = !1); s.setAttribute("columnalign", l.trim()), /[sd]/.test(r) && s.setAttribute("columnlines", r.trim()) } if ("align" === e.colSeparationType) { const t = e.cols || []; let r = ""; for (let e = 1; e < t.length; e++)r += e % 2 ? "0em " : "1em "; s.setAttribute("columnspacing", r.trim()) } else "alignat" === e.colSeparationType || "gather" === e.colSeparationType ? s.setAttribute("columnspacing", "0em") : "small" === e.colSeparationType ? s.setAttribute("columnspacing", "0.2778em") : "CD" === e.colSeparationType ? s.setAttribute("columnspacing", "0.5em") : s.setAttribute("columnspacing", "1em"); let h = ""; const c = e.hLinesBeforeRow; a += c[0].length > 0 ? "left " : "", a += c[c.length - 1].length > 0 ? "right " : ""; for (let e = 1; e < c.length - 1; e++)h += 0 === c[e].length ? "none " : c[e][0] ? "dashed " : "solid "; return /[sd]/.test(h) && s.setAttribute("rowlines", h.trim()), "" !== a && (s = new gt.MathNode("menclose", [s]), s.setAttribute("notation", a.trim())), e.arraystretch && e.arraystretch < 1 && (s = new gt.MathNode("mstyle", [s]), s.setAttribute("scriptlevel", "1")), s }, Vr = function (e, t) { -1 === e.envName.indexOf("ed") && Ir(e); const r = [], o = e.envName.indexOf("at") > -1 ? "alignat" : "align", s = "split" === e.envName, i = Hr(e.parser, { cols: r, addJot: !0, autoTag: s ? void 0 : Rr(e.envName), emptySingleRow: !0, colSeparationType: o, maxNumCols: s ? 2 : void 0, leqno: e.parser.settings.leqno }, "display"); let a, l = 0; const h = { type: "ordgroup", mode: e.mode, body: [] }; if (t[0] && "ordgroup" === t[0].type) { let e = ""; for (let r = 0; r < t[0].body.length; r++) { e += qt(t[0].body[r], "textord").text } a = Number(e), l = 2 * a } const c = !l; i.body.forEach((function (e) { for (let t = 1; t < e.length; t += 2) { const r = qt(e[t], "styling"); qt(r.body[0], "ordgroup").body.unshift(h) } if (c) l < e.length && (l = e.length); else { const t = e.length / 2; if (a < t) throw new n("Too many math in a row: expected " + a + ", but got " + t, e[0]) } })); for (let e = 0; e < l; ++e) { let t = "r", n = 0; e % 2 == 1 ? t = "l" : e > 0 && c && (n = 1), r[e] = { type: "align", align: t, pregap: n, postgap: 0 } } return i.colSeparationType = c ? "align" : "alignat", i }; Ar({ type: "array", names: ["array", "darray"], props: { numArgs: 1 }, handler(e, t) { const r = (Rt(t[0]) ? [t[0]] : qt(t[0], "ordgroup").body).map((function (e) { const t = It(e).text; if (-1 !== "lcr".indexOf(t)) return { type: "align", align: t }; if ("|" === t) return { type: "separator", separator: "|" }; if (":" === t) return { type: "separator", separator: ":" }; throw new n("Unknown column alignment: " + t, e) })), o = { cols: r, hskipBeforeAndAfter: !0, maxNumCols: r.length }; return Hr(e.parser, o, Or(e.envName)) }, htmlBuilder: Er, mathmlBuilder: Dr }), Ar({ type: "array", names: ["matrix", "pmatrix", "bmatrix", "Bmatrix", "vmatrix", "Vmatrix", "matrix*", "pmatrix*", "bmatrix*", "Bmatrix*", "vmatrix*", "Vmatrix*"], props: { numArgs: 0 }, handler(e) { const t = { matrix: null, pmatrix: ["(", ")"], bmatrix: ["[", "]"], Bmatrix: ["\\{", "\\}"], vmatrix: ["|", "|"], Vmatrix: ["\\Vert", "\\Vert"] }[e.envName.replace("*", "")]; let r = "c"; const o = { hskipBeforeAndAfter: !1, cols: [{ type: "align", align: r }] }; if ("*" === e.envName.charAt(e.envName.length - 1)) { const t = e.parser; if (t.consumeSpaces(), "[" === t.fetch().text) { if (t.consume(), t.consumeSpaces(), r = t.fetch().text, -1 === "lcr".indexOf(r)) throw new n("Expected l or c or r", t.nextToken); t.consume(), t.consumeSpaces(), t.expect("]"), t.consume(), o.cols = [{ type: "align", align: r }] } } const s = Hr(e.parser, o, Or(e.envName)), i = Math.max(0, ...s.body.map((e => e.length))); return s.cols = new Array(i).fill({ type: "align", align: r }), t ? { type: "leftright", mode: e.mode, body: [s], left: t[0], right: t[1], rightColor: void 0 } : s }, htmlBuilder: Er, mathmlBuilder: Dr }), Ar({ type: "array", names: ["smallmatrix"], props: { numArgs: 0 }, handler(e) { const t = Hr(e.parser, { arraystretch: .5 }, "script"); return t.colSeparationType = "small", t }, htmlBuilder: Er, mathmlBuilder: Dr }), Ar({ type: "array", names: ["subarray"], props: { numArgs: 1 }, handler(e, t) { const r = (Rt(t[0]) ? [t[0]] : qt(t[0], "ordgroup").body).map((function (e) { const t = It(e).text; if (-1 !== "lc".indexOf(t)) return { type: "align", align: t }; throw new n("Unknown column alignment: " + t, e) })); if (r.length > 1) throw new n("{subarray} can contain only one column"); let o = { cols: r, hskipBeforeAndAfter: !1, arraystretch: .5 }; if (o = Hr(e.parser, o, "script"), o.body.length > 0 && o.body[0].length > 1) throw new n("{subarray} can contain only one column"); return o }, htmlBuilder: Er, mathmlBuilder: Dr }), Ar({ type: "array", names: ["cases", "dcases", "rcases", "drcases"], props: { numArgs: 0 }, handler(e) { const t = Hr(e.parser, { arraystretch: 1.2, cols: [{ type: "align", align: "l", pregap: 0, postgap: 1 }, { type: "align", align: "l", pregap: 0, postgap: 0 }] }, Or(e.envName)); return { type: "leftright", mode: e.mode, body: [t], left: e.envName.indexOf("r") > -1 ? "." : "\\{", right: e.envName.indexOf("r") > -1 ? "\\}" : ".", rightColor: void 0 } }, htmlBuilder: Er, mathmlBuilder: Dr }), Ar({ type: "array", names: ["align", "align*", "aligned", "split"], props: { numArgs: 0 }, handler: Vr, htmlBuilder: Er, mathmlBuilder: Dr }), Ar({ type: "array", names: ["gathered", "gather", "gather*"], props: { numArgs: 0 }, handler(e) { l.contains(["gather", "gather*"], e.envName) && Ir(e); const t = { cols: [{ type: "align", align: "c" }], addJot: !0, colSeparationType: "gather", autoTag: Rr(e.envName), emptySingleRow: !0, leqno: e.parser.settings.leqno }; return Hr(e.parser, t, "display") }, htmlBuilder: Er, mathmlBuilder: Dr }), Ar({ type: "array", names: ["alignat", "alignat*", "alignedat"], props: { numArgs: 1 }, handler: Vr, htmlBuilder: Er, mathmlBuilder: Dr }), Ar({ type: "array", names: ["equation", "equation*"], props: { numArgs: 0 }, handler(e) { Ir(e); const t = { autoTag: Rr(e.envName), emptySingleRow: !0, singleRow: !0, maxNumCols: 1, leqno: e.parser.settings.leqno }; return Hr(e.parser, t, "display") }, htmlBuilder: Er, mathmlBuilder: Dr }), Ar({ type: "array", names: ["CD"], props: { numArgs: 0 }, handler(e) { return Ir(e), function (e) { const t = []; for (e.gullet.beginGroup(), e.gullet.macros.set("\\cr", "\\\\\\relax"), e.gullet.beginGroup(); ;) { t.push(e.parseExpression(!1, "\\\\")), e.gullet.endGroup(), e.gullet.beginGroup(); const r = e.fetch().text; if ("&" !== r && "\\\\" !== r) { if ("\\end" === r) { 0 === t[t.length - 1].length && t.pop(); break } throw new n("Expected \\\\ or \\cr or \\end", e.nextToken) } e.consume() } let r = []; const o = [r]; for (let a = 0; a < t.length; a++) { const l = t[a]; let h = { type: "styling", body: [], mode: "math", style: "display" }; for (let t = 0; t < l.length; t++)if (Ut(l[t])) { r.push(h), t += 1; const o = It(l[t]).text, a = new Array(2); if (a[0] = { type: "ordgroup", mode: "math", body: [] }, a[1] = { type: "ordgroup", mode: "math", body: [] }, "=|.".indexOf(o) > -1); else { if (!("<>AV".indexOf(o) > -1)) throw new n('Expected one of "<>AV=|." after @', l[t]); for (let e = 0; e < 2; e++) { let r = !0; for (let h = t + 1; h < l.length; h++) { if (i = o, ("mathord" === (s = l[h]).type || "atom" === s.type) && s.text === i) { r = !1, t = h; break } if (Ut(l[h])) throw new n("Missing a " + o + " character to complete a CD arrow.", l[h]); a[e].body.push(l[h]) } if (r) throw new n("Missing a " + o + " character to complete a CD arrow.", l[t]) } } const c = { type: "styling", body: [Yt(o, a, e)], mode: "math", style: "display" }; r.push(c), h = { type: "styling", body: [], mode: "math", style: "display" } } else h.body.push(l[t]); a % 2 == 0 ? r.push(h) : r.shift(), r = [], o.push(r) } var s, i; return e.gullet.endGroup(), e.gullet.endGroup(), { type: "array", mode: "math", body: o, arraystretch: 1, addJot: !0, rowGaps: [null], cols: new Array(o[0].length).fill({ type: "align", align: "c", pregap: .25, postgap: .25 }), colSeparationType: "CD", hLinesBeforeRow: new Array(o.length + 1).fill([]) } }(e.parser) }, htmlBuilder: Er, mathmlBuilder: Dr }), Br("\\nonumber", "\\gdef\\@eqnsw{0}"), Br("\\notag", "\\nonumber"), je({ type: "text", names: ["\\hline", "\\hdashline"], props: { numArgs: 0, allowedInText: !0, allowedInMath: !0 }, handler(e, t) { throw new n(e.funcName + " valid only within array environment") } }); var Pr = zr; je({ type: "environment", names: ["\\begin", "\\end"], props: { numArgs: 1, argTypes: ["text"] }, handler(e, t) { let { parser: r, funcName: o } = e; const s = t[0]; if ("ordgroup" !== s.type) throw new n("Invalid environment name", s); let i = ""; for (let e = 0; e < s.body.length; ++e)i += qt(s.body[e], "textord").text; if ("\\begin" === o) { if (!Pr.hasOwnProperty(i)) throw new n("No such environment: " + i, s); const e = Pr[i], { args: t, optArgs: o } = r.parseArguments("\\begin{" + i + "}", e), a = { mode: r.mode, envName: i, parser: r }, l = e.handler(a, t, o); r.expect("\\end", !1); const h = r.nextToken, c = qt(r.parseFunction(), "environment"); if (c.name !== i) throw new n("Mismatch: \\begin{" + i + "} matched by \\end{" + c.name + "}", h); return l } return { type: "environment", mode: r.mode, name: i, nameGroup: s } } }); const Fr = (e, t) => { const r = e.font, n = t.withFont(r); return ht(e.body, n) }, Gr = (e, t) => { const r = e.font, n = t.withFont(r); return vt(e.body, n) }, Ur = { "\\Bbb": "\\mathbb", "\\bold": "\\mathbf", "\\frak": "\\mathfrak", "\\bm": "\\boldsymbol" }; je({ type: "font", names: ["\\mathrm", "\\mathit", "\\mathbf", "\\mathnormal", "\\mathbb", "\\mathcal", "\\mathfrak", "\\mathscr", "\\mathsf", "\\mathtt", "\\Bbb", "\\bold", "\\frak"], props: { numArgs: 1, allowedInArgument: !0 }, handler: (e, t) => { let { parser: r, funcName: n } = e; const o = Ze(t[0]); let s = n; return s in Ur && (s = Ur[s]), { type: "font", mode: r.mode, font: s.slice(1), body: o } }, htmlBuilder: Fr, mathmlBuilder: Gr }), je({ type: "mclass", names: ["\\boldsymbol", "\\bm"], props: { numArgs: 1 }, handler: (e, t) => { let { parser: r } = e; const n = t[0], o = l.isCharacterBox(n); return { type: "mclass", mode: r.mode, mclass: Ft(n), body: [{ type: "font", mode: r.mode, font: "boldsymbol", body: n }], isCharacterBox: o } } }), je({ type: "font", names: ["\\rm", "\\sf", "\\tt", "\\bf", "\\it", "\\cal"], props: { numArgs: 0, allowedInText: !0 }, handler: (e, t) => { let { parser: r, funcName: n, breakOnTokenText: o } = e; const { mode: s } = r, i = r.parseExpression(!0, o); return { type: "font", mode: s, font: "math" + n.slice(1), body: { type: "ordgroup", mode: r.mode, body: i } } }, htmlBuilder: Fr, mathmlBuilder: Gr }); const Yr = (e, t) => { let r = t; return "display" === e ? r = r.id >= w.SCRIPT.id ? r.text() : w.DISPLAY : "text" === e && r.size === w.DISPLAY.size ? r = w.TEXT : "script" === e ? r = w.SCRIPT : "scriptscript" === e && (r = w.SCRIPTSCRIPT), r }, Xr = (e, t) => { const r = Yr(e.size, t.style), n = r.fracNum(), o = r.fracDen(); let s; s = t.havingStyle(n); const i = ht(e.numer, s, t); if (e.continued) { const e = 8.5 / t.fontMetrics().ptPerEm, r = 3.5 / t.fontMetrics().ptPerEm; i.height = i.height < e ? e : i.height, i.depth = i.depth < r ? r : i.depth } s = t.havingStyle(o); const a = ht(e.denom, s, t); let l, h, c, m, p, u, d, g, f, b; if (e.hasBarLine ? (e.barSize ? (h = P(e.barSize, t), l = Ve.makeLineSpan("frac-line", t, h)) : l = Ve.makeLineSpan("frac-line", t), h = l.height, c = l.height) : (l = null, h = 0, c = t.fontMetrics().defaultRuleThickness), r.size === w.DISPLAY.size || "display" === e.size ? (m = t.fontMetrics().num1, p = h > 0 ? 3 * c : 7 * c, u = t.fontMetrics().denom1) : (h > 0 ? (m = t.fontMetrics().num2, p = c) : (m = t.fontMetrics().num3, p = 3 * c), u = t.fontMetrics().denom2), l) { const e = t.fontMetrics().axisHeight; m - i.depth - (e + .5 * h) < p && (m += p - (m - i.depth - (e + .5 * h))), e - .5 * h - (a.height - u) < p && (u += p - (e - .5 * h - (a.height - u))); const r = -(e - .5 * h); d = Ve.makeVList({ positionType: "individualShift", children: [{ type: "elem", elem: a, shift: u }, { type: "elem", elem: l, shift: r }, { type: "elem", elem: i, shift: -m }] }, t) } else { const e = m - i.depth - (a.height - u); e < p && (m += .5 * (p - e), u += .5 * (p - e)), d = Ve.makeVList({ positionType: "individualShift", children: [{ type: "elem", elem: a, shift: u }, { type: "elem", elem: i, shift: -m }] }, t) } return s = t.havingStyle(r), d.height *= s.sizeMultiplier / t.sizeMultiplier, d.depth *= s.sizeMultiplier / t.sizeMultiplier, g = r.size === w.DISPLAY.size ? t.fontMetrics().delim1 : r.size === w.SCRIPTSCRIPT.size ? t.havingStyle(w.SCRIPT).fontMetrics().delim2 : t.fontMetrics().delim2, f = null == e.leftDelim ? lt(t, ["mopen"]) : yr.customSizedDelim(e.leftDelim, g, !0, t.havingStyle(r), e.mode, ["mopen"]), b = e.continued ? Ve.makeSpan([]) : null == e.rightDelim ? lt(t, ["mclose"]) : yr.customSizedDelim(e.rightDelim, g, !0, t.havingStyle(r), e.mode, ["mclose"]), Ve.makeSpan(["mord"].concat(s.sizingClasses(t)), [f, Ve.makeSpan(["mfrac"], [d]), b], t) }, Wr = (e, t) => { let r = new gt.MathNode("mfrac", [vt(e.numer, t), vt(e.denom, t)]); if (e.hasBarLine) { if (e.barSize) { const n = P(e.barSize, t); r.setAttribute("linethickness", F(n)) } } else r.setAttribute("linethickness", "0px"); const n = Yr(e.size, t.style); if (n.size !== t.style.size) { r = new gt.MathNode("mstyle", [r]); const e = n.size === w.DISPLAY.size ? "true" : "false"; r.setAttribute("displaystyle", e), r.setAttribute("scriptlevel", "0") } if (null != e.leftDelim || null != e.rightDelim) { const t = []; if (null != e.leftDelim) { const r = new gt.MathNode("mo", [new gt.TextNode(e.leftDelim.replace("\\", ""))]); r.setAttribute("fence", "true"), t.push(r) } if (t.push(r), null != e.rightDelim) { const r = new gt.MathNode("mo", [new gt.TextNode(e.rightDelim.replace("\\", ""))]); r.setAttribute("fence", "true"), t.push(r) } return bt(t) } return r }; je({ type: "genfrac", names: ["\\dfrac", "\\frac", "\\tfrac", "\\dbinom", "\\binom", "\\tbinom", "\\\\atopfrac", "\\\\bracefrac", "\\\\brackfrac"], props: { numArgs: 2, allowedInArgument: !0 }, handler: (e, t) => { let { parser: r, funcName: n } = e; const o = t[0], s = t[1]; let i, a = null, l = null, h = "auto"; switch (n) { case "\\dfrac": case "\\frac": case "\\tfrac": i = !0; break; case "\\\\atopfrac": i = !1; break; case "\\dbinom": case "\\binom": case "\\tbinom": i = !1, a = "(", l = ")"; break; case "\\\\bracefrac": i = !1, a = "\\{", l = "\\}"; break; case "\\\\brackfrac": i = !1, a = "[", l = "]"; break; default: throw new Error("Unrecognized genfrac command") }switch (n) { case "\\dfrac": case "\\dbinom": h = "display"; break; case "\\tfrac": case "\\tbinom": h = "text" }return { type: "genfrac", mode: r.mode, continued: !1, numer: o, denom: s, hasBarLine: i, leftDelim: a, rightDelim: l, size: h, barSize: null } }, htmlBuilder: Xr, mathmlBuilder: Wr }), je({ type: "genfrac", names: ["\\cfrac"], props: { numArgs: 2 }, handler: (e, t) => { let { parser: r, funcName: n } = e; const o = t[0], s = t[1]; return { type: "genfrac", mode: r.mode, continued: !0, numer: o, denom: s, hasBarLine: !0, leftDelim: null, rightDelim: null, size: "display", barSize: null } } }), je({ type: "infix", names: ["\\over", "\\choose", "\\atop", "\\brace", "\\brack"], props: { numArgs: 0, infix: !0 }, handler(e) { let t, { parser: r, funcName: n, token: o } = e; switch (n) { case "\\over": t = "\\frac"; break; case "\\choose": t = "\\binom"; break; case "\\atop": t = "\\\\atopfrac"; break; case "\\brace": t = "\\\\bracefrac"; break; case "\\brack": t = "\\\\brackfrac"; break; default: throw new Error("Unrecognized infix genfrac command") }return { type: "infix", mode: r.mode, replaceWith: t, token: o } } }); const _r = ["display", "text", "script", "scriptscript"], jr = function (e) { let t = null; return e.length > 0 && (t = e, t = "." === t ? null : t), t }; je({ type: "genfrac", names: ["\\genfrac"], props: { numArgs: 6, allowedInArgument: !0, argTypes: ["math", "math", "size", "text", "math", "math"] }, handler(e, t) { let { parser: r } = e; const n = t[4], o = t[5], s = Ze(t[0]), i = "atom" === s.type && "open" === s.family ? jr(s.text) : null, a = Ze(t[1]), l = "atom" === a.type && "close" === a.family ? jr(a.text) : null, h = qt(t[2], "size"); let c, m = null; h.isBlank ? c = !0 : (m = h.value, c = m.number > 0); let p = "auto", u = t[3]; if ("ordgroup" === u.type) { if (u.body.length > 0) { const e = qt(u.body[0], "textord"); p = _r[Number(e.text)] } } else u = qt(u, "textord"), p = _r[Number(u.text)]; return { type: "genfrac", mode: r.mode, numer: n, denom: o, continued: !1, hasBarLine: c, barSize: m, leftDelim: i, rightDelim: l, size: p } }, htmlBuilder: Xr, mathmlBuilder: Wr }), je({ type: "infix", names: ["\\above"], props: { numArgs: 1, argTypes: ["size"], infix: !0 }, handler(e, t) { let { parser: r, funcName: n, token: o } = e; return { type: "infix", mode: r.mode, replaceWith: "\\\\abovefrac", size: qt(t[0], "size").value, token: o } } }), je({ type: "genfrac", names: ["\\\\abovefrac"], props: { numArgs: 3, argTypes: ["math", "size", "math"] }, handler: (e, t) => { let { parser: r, funcName: n } = e; const o = t[0], s = function (e) { if (!e) throw new Error("Expected non-null, but got " + String(e)); return e }(qt(t[1], "infix").size), i = t[2], a = s.number > 0; return { type: "genfrac", mode: r.mode, numer: o, denom: i, continued: !1, hasBarLine: a, barSize: s, leftDelim: null, rightDelim: null, size: "auto" } }, htmlBuilder: Xr, mathmlBuilder: Wr }); const $r = (e, t) => { const r = t.style; let n, o; "supsub" === e.type ? (n = e.sup ? ht(e.sup, t.havingStyle(r.sup()), t) : ht(e.sub, t.havingStyle(r.sub()), t), o = qt(e.base, "horizBrace")) : o = qt(e, "horizBrace"); const s = ht(o.base, t.havingBaseStyle(w.DISPLAY)), i = Nt(o, t); let a; if (o.isOver ? (a = Ve.makeVList({ positionType: "firstBaseline", children: [{ type: "elem", elem: s }, { type: "kern", size: .1 }, { type: "elem", elem: i }] }, t), a.children[0].children[0].children[1].classes.push("svg-align")) : (a = Ve.makeVList({ positionType: "bottom", positionData: s.depth + .1 + i.height, children: [{ type: "elem", elem: i }, { type: "kern", size: .1 }, { type: "elem", elem: s }] }, t), a.children[0].children[0].children[0].classes.push("svg-align")), n) { const e = Ve.makeSpan(["mord", o.isOver ? "mover" : "munder"], [a], t); a = o.isOver ? Ve.makeVList({ positionType: "firstBaseline", children: [{ type: "elem", elem: e }, { type: "kern", size: .2 }, { type: "elem", elem: n }] }, t) : Ve.makeVList({ positionType: "bottom", positionData: e.depth + .2 + n.height + n.depth, children: [{ type: "elem", elem: n }, { type: "kern", size: .2 }, { type: "elem", elem: e }] }, t) } return Ve.makeSpan(["mord", o.isOver ? "mover" : "munder"], [a], t) }; je({ type: "horizBrace", names: ["\\overbrace", "\\underbrace"], props: { numArgs: 1 }, handler(e, t) { let { parser: r, funcName: n } = e; return { type: "horizBrace", mode: r.mode, label: n, isOver: /^\\over/.test(n), base: t[0] } }, htmlBuilder: $r, mathmlBuilder: (e, t) => { const r = Ct(e.label); return new gt.MathNode(e.isOver ? "mover" : "munder", [vt(e.base, t), r]) } }), je({ type: "href", names: ["\\href"], props: { numArgs: 2, argTypes: ["url", "original"], allowedInText: !0 }, handler: (e, t) => { let { parser: r } = e; const n = t[1], o = qt(t[0], "url").url; return r.settings.isTrusted({ command: "\\href", url: o }) ? { type: "href", mode: r.mode, href: o, body: Ke(n) } : r.formatUnsupportedCmd("\\href") }, htmlBuilder: (e, t) => { const r = nt(e.body, t, !1); return Ve.makeAnchor(e.href, [], r, t) }, mathmlBuilder: (e, t) => { let r = wt(e.body, t); return r instanceof ut || (r = new ut("mrow", [r])), r.setAttribute("href", e.href), r } }), je({ type: "href", names: ["\\url"], props: { numArgs: 1, argTypes: ["url"], allowedInText: !0 }, handler: (e, t) => { let { parser: r } = e; const n = qt(t[0], "url").url; if (!r.settings.isTrusted({ command: "\\url", url: n })) return r.formatUnsupportedCmd("\\url"); const o = []; for (let e = 0; e < n.length; e++) { let t = n[e]; "~" === t && (t = "\\textasciitilde"), o.push({ type: "textord", mode: "text", text: t }) } const s = { type: "text", mode: r.mode, font: "\\texttt", body: o }; return { type: "href", mode: r.mode, href: n, body: Ke(s) } } }), je({ type: "hbox", names: ["\\hbox"], props: { numArgs: 1, argTypes: ["text"], allowedInText: !0, primitive: !0 }, handler(e, t) { let { parser: r } = e; return { type: "hbox", mode: r.mode, body: Ke(t[0]) } }, htmlBuilder(e, t) { const r = nt(e.body, t, !1); return Ve.makeFragment(r) }, mathmlBuilder(e, t) { return new gt.MathNode("mrow", xt(e.body, t)) } }), je({ type: "html", names: ["\\htmlClass", "\\htmlId", "\\htmlStyle", "\\htmlData"], props: { numArgs: 2, argTypes: ["raw", "original"], allowedInText: !0 }, handler: (e, t) => { let { parser: r, funcName: o, token: s } = e; const i = qt(t[0], "raw").string, a = t[1]; let l; r.settings.strict && r.settings.reportNonstrict("htmlExtension", "HTML extension is disabled on strict mode"); const h = {}; switch (o) { case "\\htmlClass": h.class = i, l = { command: "\\htmlClass", class: i }; break; case "\\htmlId": h.id = i, l = { command: "\\htmlId", id: i }; break; case "\\htmlStyle": h.style = i, l = { command: "\\htmlStyle", style: i }; break; case "\\htmlData": { const e = i.split(","); for (let t = 0; t < e.length; t++) { const r = e[t].split("="); if (2 !== r.length) throw new n("Error parsing key-value for \\htmlData"); h["data-" + r[0].trim()] = r[1].trim() } l = { command: "\\htmlData", attributes: h }; break } default: throw new Error("Unrecognized html command") }return r.settings.isTrusted(l) ? { type: "html", mode: r.mode, attributes: h, body: Ke(a) } : r.formatUnsupportedCmd(o) }, htmlBuilder: (e, t) => { const r = nt(e.body, t, !1), n = ["enclosing"]; e.attributes.class && n.push(...e.attributes.class.trim().split(/\s+/)); const o = Ve.makeSpan(n, r, t); for (const t in e.attributes) "class" !== t && e.attributes.hasOwnProperty(t) && o.setAttribute(t, e.attributes[t]); return o }, mathmlBuilder: (e, t) => wt(e.body, t) }), je({ type: "htmlmathml", names: ["\\html@mathml"], props: { numArgs: 2, allowedInText: !0 }, handler: (e, t) => { let { parser: r } = e; return { type: "htmlmathml", mode: r.mode, html: Ke(t[0]), mathml: Ke(t[1]) } }, htmlBuilder: (e, t) => { const r = nt(e.html, t, !1); return Ve.makeFragment(r) }, mathmlBuilder: (e, t) => wt(e.mathml, t) }); const Zr = function (e) { if (/^[-+]? *(\d+(\.\d*)?|\.\d+)$/.test(e)) return { number: +e, unit: "bp" }; { const t = /([-+]?) *(\d+(?:\.\d*)?|\.\d+) *([a-z]{2})/.exec(e); if (!t) throw new n("Invalid size: '" + e + "' in \\includegraphics"); const r = { number: +(t[1] + t[2]), unit: t[3] }; if (!V(r)) throw new n("Invalid unit: '" + r.unit + "' in \\includegraphics."); return r } }; je({ type: "includegraphics", names: ["\\includegraphics"], props: { numArgs: 1, numOptionalArgs: 1, argTypes: ["raw", "url"], allowedInText: !1 }, handler: (e, t, r) => { let { parser: o } = e, s = { number: 0, unit: "em" }, i = { number: .9, unit: "em" }, a = { number: 0, unit: "em" }, l = ""; if (r[0]) { const e = qt(r[0], "raw").string.split(","); for (let t = 0; t < e.length; t++) { const r = e[t].split("="); if (2 === r.length) { const e = r[1].trim(); switch (r[0].trim()) { case "alt": l = e; break; case "width": s = Zr(e); break; case "height": i = Zr(e); break; case "totalheight": a = Zr(e); break; default: throw new n("Invalid key: '" + r[0] + "' in \\includegraphics.") } } } } const h = qt(t[0], "url").url; return "" === l && (l = h, l = l.replace(/^.*[\\/]/, ""), l = l.substring(0, l.lastIndexOf("."))), o.settings.isTrusted({ command: "\\includegraphics", url: h }) ? { type: "includegraphics", mode: o.mode, alt: l, width: s, height: i, totalheight: a, src: h } : o.formatUnsupportedCmd("\\includegraphics") }, htmlBuilder: (e, t) => { const r = P(e.height, t); let n = 0; e.totalheight.number > 0 && (n = P(e.totalheight, t) - r); let o = 0; e.width.number > 0 && (o = P(e.width, t)); const s = { height: F(r + n) }; o > 0 && (s.width = F(o)), n > 0 && (s.verticalAlign = F(-n)); const i = new j(e.src, e.alt, s); return i.height = r, i.depth = n, i }, mathmlBuilder: (e, t) => { const r = new gt.MathNode("mglyph", []); r.setAttribute("alt", e.alt); const n = P(e.height, t); let o = 0; if (e.totalheight.number > 0 && (o = P(e.totalheight, t) - n, r.setAttribute("valign", F(-o))), r.setAttribute("height", F(n + o)), e.width.number > 0) { const n = P(e.width, t); r.setAttribute("width", F(n)) } return r.setAttribute("src", e.src), r } }), je({ type: "kern", names: ["\\kern", "\\mkern", "\\hskip", "\\mskip"], props: { numArgs: 1, argTypes: ["size"], primitive: !0, allowedInText: !0 }, handler(e, t) { let { parser: r, funcName: n } = e; const o = qt(t[0], "size"); if (r.settings.strict) { const e = "m" === n[1], t = "mu" === o.value.unit; e ? (t || r.settings.reportNonstrict("mathVsTextUnits", "LaTeX's " + n + " supports only mu units, not " + o.value.unit + " units"), "math" !== r.mode && r.settings.reportNonstrict("mathVsTextUnits", "LaTeX's " + n + " works only in math mode")) : t && r.settings.reportNonstrict("mathVsTextUnits", "LaTeX's " + n + " doesn't support mu units") } return { type: "kern", mode: r.mode, dimension: o.value } }, htmlBuilder(e, t) { return Ve.makeGlue(e.dimension, t) }, mathmlBuilder(e, t) { const r = P(e.dimension, t); return new gt.SpaceNode(r) } }), je({ type: "lap", names: ["\\mathllap", "\\mathrlap", "\\mathclap"], props: { numArgs: 1, allowedInText: !0 }, handler: (e, t) => { let { parser: r, funcName: n } = e; const o = t[0]; return { type: "lap", mode: r.mode, alignment: n.slice(5), body: o } }, htmlBuilder: (e, t) => { let r; "clap" === e.alignment ? (r = Ve.makeSpan([], [ht(e.body, t)]), r = Ve.makeSpan(["inner"], [r], t)) : r = Ve.makeSpan(["inner"], [ht(e.body, t)]); const n = Ve.makeSpan(["fix"], []); let o = Ve.makeSpan([e.alignment], [r, n], t); const s = Ve.makeSpan(["strut"]); return s.style.height = F(o.height + o.depth), o.depth && (s.style.verticalAlign = F(-o.depth)), o.children.unshift(s), o = Ve.makeSpan(["thinbox"], [o], t), Ve.makeSpan(["mord", "vbox"], [o], t) }, mathmlBuilder: (e, t) => { const r = new gt.MathNode("mpadded", [vt(e.body, t)]); if ("rlap" !== e.alignment) { const t = "llap" === e.alignment ? "-1" : "-0.5"; r.setAttribute("lspace", t + "width") } return r.setAttribute("width", "0px"), r } }), je({ type: "styling", names: ["\\(", "$"], props: { numArgs: 0, allowedInText: !0, allowedInMath: !1 }, handler(e, t) { let { funcName: r, parser: n } = e; const o = n.mode; n.switchMode("math"); const s = "\\(" === r ? "\\)" : "$", i = n.parseExpression(!1, s); return n.expect(s), n.switchMode(o), { type: "styling", mode: n.mode, style: "text", body: i } } }), je({ type: "text", names: ["\\)", "\\]"], props: { numArgs: 0, allowedInText: !0, allowedInMath: !1 }, handler(e, t) { throw new n("Mismatched " + e.funcName) } }); const Kr = (e, t) => { switch (t.style.size) { case w.DISPLAY.size: return e.display; case w.TEXT.size: return e.text; case w.SCRIPT.size: return e.script; case w.SCRIPTSCRIPT.size: return e.scriptscript; default: return e.text } }; je({ type: "mathchoice", names: ["\\mathchoice"], props: { numArgs: 4, primitive: !0 }, handler: (e, t) => { let { parser: r } = e; return { type: "mathchoice", mode: r.mode, display: Ke(t[0]), text: Ke(t[1]), script: Ke(t[2]), scriptscript: Ke(t[3]) } }, htmlBuilder: (e, t) => { const r = Kr(e, t), n = nt(r, t, !1); return Ve.makeFragment(n) }, mathmlBuilder: (e, t) => { const r = Kr(e, t); return wt(r, t) } }); const Jr = (e, t, r, n, o, s, i) => { e = Ve.makeSpan([], [e]); const a = r && l.isCharacterBox(r); let h, c, m; if (t) { const e = ht(t, n.havingStyle(o.sup()), n); c = { elem: e, kern: Math.max(n.fontMetrics().bigOpSpacing1, n.fontMetrics().bigOpSpacing3 - e.depth) } } if (r) { const e = ht(r, n.havingStyle(o.sub()), n); h = { elem: e, kern: Math.max(n.fontMetrics().bigOpSpacing2, n.fontMetrics().bigOpSpacing4 - e.height) } } if (c && h) { const t = n.fontMetrics().bigOpSpacing5 + h.elem.height + h.elem.depth + h.kern + e.depth + i; m = Ve.makeVList({ positionType: "bottom", positionData: t, children: [{ type: "kern", size: n.fontMetrics().bigOpSpacing5 }, { type: "elem", elem: h.elem, marginLeft: F(-s) }, { type: "kern", size: h.kern }, { type: "elem", elem: e }, { type: "kern", size: c.kern }, { type: "elem", elem: c.elem, marginLeft: F(s) }, { type: "kern", size: n.fontMetrics().bigOpSpacing5 }] }, n) } else if (h) { const t = e.height - i; m = Ve.makeVList({ positionType: "top", positionData: t, children: [{ type: "kern", size: n.fontMetrics().bigOpSpacing5 }, { type: "elem", elem: h.elem, marginLeft: F(-s) }, { type: "kern", size: h.kern }, { type: "elem", elem: e }] }, n) } else { if (!c) return e; { const t = e.depth + i; m = Ve.makeVList({ positionType: "bottom", positionData: t, children: [{ type: "elem", elem: e }, { type: "kern", size: c.kern }, { type: "elem", elem: c.elem, marginLeft: F(s) }, { type: "kern", size: n.fontMetrics().bigOpSpacing5 }] }, n) } } const p = [m]; if (h && 0 !== s && !a) { const e = Ve.makeSpan(["mspace"], [], n); e.style.marginRight = F(s), p.unshift(e) } return Ve.makeSpan(["mop", "op-limits"], p, n) }, Qr = ["\\smallint"], en = (e, t) => { let r, n, o, s = !1; "supsub" === e.type ? (r = e.sup, n = e.sub, o = qt(e.base, "op"), s = !0) : o = qt(e, "op"); const i = t.style; let a, h = !1; if (i.size === w.DISPLAY.size && o.symbol && !l.contains(Qr, o.name) && (h = !0), o.symbol) { const e = h ? "Size2-Regular" : "Size1-Regular"; let r = ""; if ("\\oiint" !== o.name && "\\oiiint" !== o.name || (r = o.name.slice(1), o.name = "oiint" === r ? "\\iint" : "\\iiint"), a = Ve.makeSymbol(o.name, e, "math", t, ["mop", "op-symbol", h ? "large-op" : "small-op"]), r.length > 0) { const e = a.italic, n = Ve.staticSvg(r + "Size" + (h ? "2" : "1"), t); a = Ve.makeVList({ positionType: "individualShift", children: [{ type: "elem", elem: a, shift: 0 }, { type: "elem", elem: n, shift: h ? .08 : 0 }] }, t), o.name = "\\" + r, a.classes.unshift("mop"), a.italic = e } } else if (o.body) { const e = nt(o.body, t, !0); 1 === e.length && e[0] instanceof Z ? (a = e[0], a.classes[0] = "mop") : a = Ve.makeSpan(["mop"], e, t) } else { const e = []; for (let r = 1; r < o.name.length; r++)e.push(Ve.mathsym(o.name[r], o.mode, t)); a = Ve.makeSpan(["mop"], e, t) } let c = 0, m = 0; return (a instanceof Z || "\\oiint" === o.name || "\\oiiint" === o.name) && !o.suppressBaseShift && (c = (a.height - a.depth) / 2 - t.fontMetrics().axisHeight, m = a.italic), s ? Jr(a, r, n, t, i, m, c) : (c && (a.style.position = "relative", a.style.top = F(c)), a) }, tn = (e, t) => { let r; if (e.symbol) r = new ut("mo", [ft(e.name, e.mode)]), l.contains(Qr, e.name) && r.setAttribute("largeop", "false"); else if (e.body) r = new ut("mo", xt(e.body, t)); else { r = new ut("mi", [new dt(e.name.slice(1))]); const t = new ut("mo", [ft("\u2061", "text")]); r = e.parentIsSupSub ? new ut("mrow", [r, t]) : pt([r, t]) } return r }, rn = { "\u220f": "\\prod", "\u2210": "\\coprod", "\u2211": "\\sum", "\u22c0": "\\bigwedge", "\u22c1": "\\bigvee", "\u22c2": "\\bigcap", "\u22c3": "\\bigcup", "\u2a00": "\\bigodot", "\u2a01": "\\bigoplus", "\u2a02": "\\bigotimes", "\u2a04": "\\biguplus", "\u2a06": "\\bigsqcup" }; je({ type: "op", names: ["\\coprod", "\\bigvee", "\\bigwedge", "\\biguplus", "\\bigcap", "\\bigcup", "\\intop", "\\prod", "\\sum", "\\bigotimes", "\\bigoplus", "\\bigodot", "\\bigsqcup", "\\smallint", "\u220f", "\u2210", "\u2211", "\u22c0", "\u22c1", "\u22c2", "\u22c3", "\u2a00", "\u2a01", "\u2a02", "\u2a04", "\u2a06"], props: { numArgs: 0 }, handler: (e, t) => { let { parser: r, funcName: n } = e, o = n; return 1 === o.length && (o = rn[o]), { type: "op", mode: r.mode, limits: !0, parentIsSupSub: !1, symbol: !0, name: o } }, htmlBuilder: en, mathmlBuilder: tn }), je({ type: "op", names: ["\\mathop"], props: { numArgs: 1, primitive: !0 }, handler: (e, t) => { let { parser: r } = e; const n = t[0]; return { type: "op", mode: r.mode, limits: !1, parentIsSupSub: !1, symbol: !1, body: Ke(n) } }, htmlBuilder: en, mathmlBuilder: tn }); const nn = { "\u222b": "\\int", "\u222c": "\\iint", "\u222d": "\\iiint", "\u222e": "\\oint", "\u222f": "\\oiint", "\u2230": "\\oiiint" }; je({ type: "op", names: ["\\arcsin", "\\arccos", "\\arctan", "\\arctg", "\\arcctg", "\\arg", "\\ch", "\\cos", "\\cosec", "\\cosh", "\\cot", "\\cotg", "\\coth", "\\csc", "\\ctg", "\\cth", "\\deg", "\\dim", "\\exp", "\\hom", "\\ker", "\\lg", "\\ln", "\\log", "\\sec", "\\sin", "\\sinh", "\\sh", "\\tan", "\\tanh", "\\tg", "\\th"], props: { numArgs: 0 }, handler(e) { let { parser: t, funcName: r } = e; return { type: "op", mode: t.mode, limits: !1, parentIsSupSub: !1, symbol: !1, name: r } }, htmlBuilder: en, mathmlBuilder: tn }), je({ type: "op", names: ["\\det", "\\gcd", "\\inf", "\\lim", "\\max", "\\min", "\\Pr", "\\sup"], props: { numArgs: 0 }, handler(e) { let { parser: t, funcName: r } = e; return { type: "op", mode: t.mode, limits: !0, parentIsSupSub: !1, symbol: !1, name: r } }, htmlBuilder: en, mathmlBuilder: tn }), je({ type: "op", names: ["\\int", "\\iint", "\\iiint", "\\oint", "\\oiint", "\\oiiint", "\u222b", "\u222c", "\u222d", "\u222e", "\u222f", "\u2230"], props: { numArgs: 0 }, handler(e) { let { parser: t, funcName: r } = e, n = r; return 1 === n.length && (n = nn[n]), { type: "op", mode: t.mode, limits: !1, parentIsSupSub: !1, symbol: !0, name: n } }, htmlBuilder: en, mathmlBuilder: tn }); const on = (e, t) => { let r, n, o, s, i = !1; if ("supsub" === e.type ? (r = e.sup, n = e.sub, o = qt(e.base, "operatorname"), i = !0) : o = qt(e, "operatorname"), o.body.length > 0) { const e = o.body.map((e => { const t = e.text; return "string" == typeof t ? { type: "textord", mode: e.mode, text: t } : e })), r = nt(e, t.withFont("mathrm"), !0); for (let e = 0; e < r.length; e++) { const t = r[e]; t instanceof Z && (t.text = t.text.replace(/\u2212/, "-").replace(/\u2217/, "*")) } s = Ve.makeSpan(["mop"], r, t) } else s = Ve.makeSpan(["mop"], [], t); return i ? Jr(s, r, n, t, t.style, 0, 0) : s }; function sn(e, t, r) { const n = nt(e, t, !1), o = t.sizeMultiplier / r.sizeMultiplier; for (let e = 0; e < n.length; e++) { const s = n[e].classes.indexOf("sizing"); s < 0 ? Array.prototype.push.apply(n[e].classes, t.sizingClasses(r)) : n[e].classes[s + 1] === "reset-size" + t.size && (n[e].classes[s + 1] = "reset-size" + r.size), n[e].height *= o, n[e].depth *= o } return Ve.makeFragment(n) } je({ type: "operatorname", names: ["\\operatorname@", "\\operatornamewithlimits"], props: { numArgs: 1 }, handler: (e, t) => { let { parser: r, funcName: n } = e; const o = t[0]; return { type: "operatorname", mode: r.mode, body: Ke(o), alwaysHandleSupSub: "\\operatornamewithlimits" === n, limits: !1, parentIsSupSub: !1 } }, htmlBuilder: on, mathmlBuilder: (e, t) => { let r = xt(e.body, t.withFont("mathrm")), n = !0; for (let e = 0; e < r.length; e++) { const t = r[e]; if (t instanceof gt.SpaceNode); else if (t instanceof gt.MathNode) switch (t.type) { case "mi": case "mn": case "ms": case "mspace": case "mtext": break; case "mo": { const e = t.children[0]; 1 === t.children.length && e instanceof gt.TextNode ? e.text = e.text.replace(/\u2212/, "-").replace(/\u2217/, "*") : n = !1; break } default: n = !1 } else n = !1 } if (n) { const e = r.map((e => e.toText())).join(""); r = [new gt.TextNode(e)] } const o = new gt.MathNode("mi", r); o.setAttribute("mathvariant", "normal"); const s = new gt.MathNode("mo", [ft("\u2061", "text")]); return e.parentIsSupSub ? new gt.MathNode("mrow", [o, s]) : gt.newDocumentFragment([o, s]) } }), Br("\\operatorname", "\\@ifstar\\operatornamewithlimits\\operatorname@"), $e({ type: "ordgroup", htmlBuilder(e, t) { return e.semisimple ? Ve.makeFragment(nt(e.body, t, !1)) : Ve.makeSpan(["mord"], nt(e.body, t, !0), t) }, mathmlBuilder(e, t) { return wt(e.body, t, !0) } }), je({ type: "overline", names: ["\\overline"], props: { numArgs: 1 }, handler(e, t) { let { parser: r } = e; const n = t[0]; return { type: "overline", mode: r.mode, body: n } }, htmlBuilder(e, t) { const r = ht(e.body, t.havingCrampedStyle()), n = Ve.makeLineSpan("overline-line", t), o = t.fontMetrics().defaultRuleThickness, s = Ve.makeVList({ positionType: "firstBaseline", children: [{ type: "elem", elem: r }, { type: "kern", size: 3 * o }, { type: "elem", elem: n }, { type: "kern", size: o }] }, t); return Ve.makeSpan(["mord", "overline"], [s], t) }, mathmlBuilder(e, t) { const r = new gt.MathNode("mo", [new gt.TextNode("\u203e")]); r.setAttribute("stretchy", "true"); const n = new gt.MathNode("mover", [vt(e.body, t), r]); return n.setAttribute("accent", "true"), n } }), je({ type: "phantom", names: ["\\phantom"], props: { numArgs: 1, allowedInText: !0 }, handler: (e, t) => { let { parser: r } = e; const n = t[0]; return { type: "phantom", mode: r.mode, body: Ke(n) } }, htmlBuilder: (e, t) => { const r = nt(e.body, t.withPhantom(), !1); return Ve.makeFragment(r) }, mathmlBuilder: (e, t) => { const r = xt(e.body, t); return new gt.MathNode("mphantom", r) } }), je({ type: "hphantom", names: ["\\hphantom"], props: { numArgs: 1, allowedInText: !0 }, handler: (e, t) => { let { parser: r } = e; const n = t[0]; return { type: "hphantom", mode: r.mode, body: n } }, htmlBuilder: (e, t) => { let r = Ve.makeSpan([], [ht(e.body, t.withPhantom())]); if (r.height = 0, r.depth = 0, r.children) for (let e = 0; e < r.children.length; e++)r.children[e].height = 0, r.children[e].depth = 0; return r = Ve.makeVList({ positionType: "firstBaseline", children: [{ type: "elem", elem: r }] }, t), Ve.makeSpan(["mord"], [r], t) }, mathmlBuilder: (e, t) => { const r = xt(Ke(e.body), t), n = new gt.MathNode("mphantom", r), o = new gt.MathNode("mpadded", [n]); return o.setAttribute("height", "0px"), o.setAttribute("depth", "0px"), o } }), je({ type: "vphantom", names: ["\\vphantom"], props: { numArgs: 1, allowedInText: !0 }, handler: (e, t) => { let { parser: r } = e; const n = t[0]; return { type: "vphantom", mode: r.mode, body: n } }, htmlBuilder: (e, t) => { const r = Ve.makeSpan(["inner"], [ht(e.body, t.withPhantom())]), n = Ve.makeSpan(["fix"], []); return Ve.makeSpan(["mord", "rlap"], [r, n], t) }, mathmlBuilder: (e, t) => { const r = xt(Ke(e.body), t), n = new gt.MathNode("mphantom", r), o = new gt.MathNode("mpadded", [n]); return o.setAttribute("width", "0px"), o } }), je({ type: "raisebox", names: ["\\raisebox"], props: { numArgs: 2, argTypes: ["size", "hbox"], allowedInText: !0 }, handler(e, t) { let { parser: r } = e; const n = qt(t[0], "size").value, o = t[1]; return { type: "raisebox", mode: r.mode, dy: n, body: o } }, htmlBuilder(e, t) { const r = ht(e.body, t), n = P(e.dy, t); return Ve.makeVList({ positionType: "shift", positionData: -n, children: [{ type: "elem", elem: r }] }, t) }, mathmlBuilder(e, t) { const r = new gt.MathNode("mpadded", [vt(e.body, t)]), n = e.dy.number + e.dy.unit; return r.setAttribute("voffset", n), r } }), je({ type: "internal", names: ["\\relax"], props: { numArgs: 0, allowedInText: !0 }, handler(e) { let { parser: t } = e; return { type: "internal", mode: t.mode } } }), je({ type: "rule", names: ["\\rule"], props: { numArgs: 2, numOptionalArgs: 1, argTypes: ["size", "size", "size"] }, handler(e, t, r) { let { parser: n } = e; const o = r[0], s = qt(t[0], "size"), i = qt(t[1], "size"); return { type: "rule", mode: n.mode, shift: o && qt(o, "size").value, width: s.value, height: i.value } }, htmlBuilder(e, t) { const r = Ve.makeSpan(["mord", "rule"], [], t), n = P(e.width, t), o = P(e.height, t), s = e.shift ? P(e.shift, t) : 0; return r.style.borderRightWidth = F(n), r.style.borderTopWidth = F(o), r.style.bottom = F(s), r.width = n, r.height = o + s, r.depth = -s, r.maxFontSize = 1.125 * o * t.sizeMultiplier, r }, mathmlBuilder(e, t) { const r = P(e.width, t), n = P(e.height, t), o = e.shift ? P(e.shift, t) : 0, s = t.color && t.getColor() || "black", i = new gt.MathNode("mspace"); i.setAttribute("mathbackground", s), i.setAttribute("width", F(r)), i.setAttribute("height", F(n)); const a = new gt.MathNode("mpadded", [i]); return o >= 0 ? a.setAttribute("height", F(o)) : (a.setAttribute("height", F(o)), a.setAttribute("depth", F(-o))), a.setAttribute("voffset", F(o)), a } }); const an = ["\\tiny", "\\sixptsize", "\\scriptsize", "\\footnotesize", "\\small", "\\normalsize", "\\large", "\\Large", "\\LARGE", "\\huge", "\\Huge"]; je({ type: "sizing", names: an, props: { numArgs: 0, allowedInText: !0 }, handler: (e, t) => { let { breakOnTokenText: r, funcName: n, parser: o } = e; const s = o.parseExpression(!1, r); return { type: "sizing", mode: o.mode, size: an.indexOf(n) + 1, body: s } }, htmlBuilder: (e, t) => { const r = t.havingSize(e.size); return sn(e.body, r, t) }, mathmlBuilder: (e, t) => { const r = t.havingSize(e.size), n = xt(e.body, r), o = new gt.MathNode("mstyle", n); return o.setAttribute("mathsize", F(r.sizeMultiplier)), o } }), je({ type: "smash", names: ["\\smash"], props: { numArgs: 1, numOptionalArgs: 1, allowedInText: !0 }, handler: (e, t, r) => { let { parser: n } = e, o = !1, s = !1; const i = r[0] && qt(r[0], "ordgroup"); if (i) { let e = ""; for (let t = 0; t < i.body.length; ++t) { if (e = i.body[t].text, "t" === e) o = !0; else { if ("b" !== e) { o = !1, s = !1; break } s = !0 } } } else o = !0, s = !0; const a = t[0]; return { type: "smash", mode: n.mode, body: a, smashHeight: o, smashDepth: s } }, htmlBuilder: (e, t) => { const r = Ve.makeSpan([], [ht(e.body, t)]); if (!e.smashHeight && !e.smashDepth) return r; if (e.smashHeight && (r.height = 0, r.children)) for (let e = 0; e < r.children.length; e++)r.children[e].height = 0; if (e.smashDepth && (r.depth = 0, r.children)) for (let e = 0; e < r.children.length; e++)r.children[e].depth = 0; const n = Ve.makeVList({ positionType: "firstBaseline", children: [{ type: "elem", elem: r }] }, t); return Ve.makeSpan(["mord"], [n], t) }, mathmlBuilder: (e, t) => { const r = new gt.MathNode("mpadded", [vt(e.body, t)]); return e.smashHeight && r.setAttribute("height", "0px"), e.smashDepth && r.setAttribute("depth", "0px"), r } }), je({ type: "sqrt", names: ["\\sqrt"], props: { numArgs: 1, numOptionalArgs: 1 }, handler(e, t, r) { let { parser: n } = e; const o = r[0], s = t[0]; return { type: "sqrt", mode: n.mode, body: s, index: o } }, htmlBuilder(e, t) { let r = ht(e.body, t.havingCrampedStyle()); 0 === r.height && (r.height = t.fontMetrics().xHeight), r = Ve.wrapFragment(r, t); const n = t.fontMetrics().defaultRuleThickness; let o = n; t.style.id < w.TEXT.id && (o = t.fontMetrics().xHeight); let s = n + o / 4; const i = r.height + r.depth + s + n, { span: a, ruleWidth: l, advanceWidth: h } = yr.sqrtImage(i, t), c = a.height - l; c > r.height + r.depth + s && (s = (s + c - r.height - r.depth) / 2); const m = a.height - r.height - s - l; r.style.paddingLeft = F(h); const p = Ve.makeVList({ positionType: "firstBaseline", children: [{ type: "elem", elem: r, wrapperClasses: ["svg-align"] }, { type: "kern", size: -(r.height + m) }, { type: "elem", elem: a }, { type: "kern", size: l }] }, t); if (e.index) { const r = t.havingStyle(w.SCRIPTSCRIPT), n = ht(e.index, r, t), o = .6 * (p.height - p.depth), s = Ve.makeVList({ positionType: "shift", positionData: -o, children: [{ type: "elem", elem: n }] }, t), i = Ve.makeSpan(["root"], [s]); return Ve.makeSpan(["mord", "sqrt"], [i, p], t) } return Ve.makeSpan(["mord", "sqrt"], [p], t) }, mathmlBuilder(e, t) { const { body: r, index: n } = e; return n ? new gt.MathNode("mroot", [vt(r, t), vt(n, t)]) : new gt.MathNode("msqrt", [vt(r, t)]) } }); const ln = { display: w.DISPLAY, text: w.TEXT, script: w.SCRIPT, scriptscript: w.SCRIPTSCRIPT }; je({ type: "styling", names: ["\\displaystyle", "\\textstyle", "\\scriptstyle", "\\scriptscriptstyle"], props: { numArgs: 0, allowedInText: !0, primitive: !0 }, handler(e, t) { let { breakOnTokenText: r, funcName: n, parser: o } = e; const s = o.parseExpression(!0, r), i = n.slice(1, n.length - 5); return { type: "styling", mode: o.mode, style: i, body: s } }, htmlBuilder(e, t) { const r = ln[e.style], n = t.havingStyle(r).withFont(""); return sn(e.body, n, t) }, mathmlBuilder(e, t) { const r = ln[e.style], n = t.havingStyle(r), o = xt(e.body, n), s = new gt.MathNode("mstyle", o), i = { display: ["0", "true"], text: ["0", "false"], script: ["1", "false"], scriptscript: ["2", "false"] }[e.style]; return s.setAttribute("scriptlevel", i[0]), s.setAttribute("displaystyle", i[1]), s } }); $e({ type: "supsub", htmlBuilder(e, t) { const r = function (e, t) { const r = e.base; if (r) return "op" === r.type ? r.limits && (t.style.size === w.DISPLAY.size || r.alwaysHandleSupSub) ? en : null : "operatorname" === r.type ? r.alwaysHandleSupSub && (t.style.size === w.DISPLAY.size || r.limits) ? on : null : "accent" === r.type ? l.isCharacterBox(r.base) ? Ht : null : "horizBrace" === r.type && !e.sub === r.isOver ? $r : null; return null }(e, t); if (r) return r(e, t); const { base: n, sup: o, sub: s } = e, i = ht(n, t); let a, h; const c = t.fontMetrics(); let m = 0, p = 0; const u = n && l.isCharacterBox(n); if (o) { const e = t.havingStyle(t.style.sup()); a = ht(o, e, t), u || (m = i.height - e.fontMetrics().supDrop * e.sizeMultiplier / t.sizeMultiplier) } if (s) { const e = t.havingStyle(t.style.sub()); h = ht(s, e, t), u || (p = i.depth + e.fontMetrics().subDrop * e.sizeMultiplier / t.sizeMultiplier) } let d; d = t.style === w.DISPLAY ? c.sup1 : t.style.cramped ? c.sup3 : c.sup2; const g = t.sizeMultiplier, f = F(.5 / c.ptPerEm / g); let b, y = null; if (h) { const t = e.base && "op" === e.base.type && e.base.name && ("\\oiint" === e.base.name || "\\oiiint" === e.base.name); (i instanceof Z || t) && (y = F(-i.italic)) } if (a && h) { m = Math.max(m, d, a.depth + .25 * c.xHeight), p = Math.max(p, c.sub2); const e = 4 * c.defaultRuleThickness; if (m - a.depth - (h.height - p) < e) { p = e - (m - a.depth) + h.height; const t = .8 * c.xHeight - (m - a.depth); t > 0 && (m += t, p -= t) } const r = [{ type: "elem", elem: h, shift: p, marginRight: f, marginLeft: y }, { type: "elem", elem: a, shift: -m, marginRight: f }]; b = Ve.makeVList({ positionType: "individualShift", children: r }, t) } else if (h) { p = Math.max(p, c.sub1, h.height - .8 * c.xHeight); const e = [{ type: "elem", elem: h, marginLeft: y, marginRight: f }]; b = Ve.makeVList({ positionType: "shift", positionData: p, children: e }, t) } else { if (!a) throw new Error("supsub must have either sup or sub."); m = Math.max(m, d, a.depth + .25 * c.xHeight), b = Ve.makeVList({ positionType: "shift", positionData: -m, children: [{ type: "elem", elem: a, marginRight: f }] }, t) } const x = at(i, "right") || "mord"; return Ve.makeSpan([x], [i, Ve.makeSpan(["msupsub"], [b])], t) }, mathmlBuilder(e, t) { let r, n, o = !1; e.base && "horizBrace" === e.base.type && (n = !!e.sup, n === e.base.isOver && (o = !0, r = e.base.isOver)), !e.base || "op" !== e.base.type && "operatorname" !== e.base.type || (e.base.parentIsSupSub = !0); const s = [vt(e.base, t)]; let i; if (e.sub && s.push(vt(e.sub, t)), e.sup && s.push(vt(e.sup, t)), o) i = r ? "mover" : "munder"; else if (e.sub) if (e.sup) { const r = e.base; i = r && "op" === r.type && r.limits && t.style === w.DISPLAY || r && "operatorname" === r.type && r.alwaysHandleSupSub && (t.style === w.DISPLAY || r.limits) ? "munderover" : "msubsup" } else { const r = e.base; i = r && "op" === r.type && r.limits && (t.style === w.DISPLAY || r.alwaysHandleSupSub) || r && "operatorname" === r.type && r.alwaysHandleSupSub && (r.limits || t.style === w.DISPLAY) ? "munder" : "msub" } else { const r = e.base; i = r && "op" === r.type && r.limits && (t.style === w.DISPLAY || r.alwaysHandleSupSub) || r && "operatorname" === r.type && r.alwaysHandleSupSub && (r.limits || t.style === w.DISPLAY) ? "mover" : "msup" } return new gt.MathNode(i, s) } }), $e({ type: "atom", htmlBuilder(e, t) { return Ve.mathsym(e.text, e.mode, t, ["m" + e.family]) }, mathmlBuilder(e, t) { const r = new gt.MathNode("mo", [ft(e.text, e.mode)]); if ("bin" === e.family) { const n = yt(e, t); "bold-italic" === n && r.setAttribute("mathvariant", n) } else "punct" === e.family ? r.setAttribute("separator", "true") : "open" !== e.family && "close" !== e.family || r.setAttribute("stretchy", "false"); return r } }); const hn = { mi: "italic", mn: "normal", mtext: "normal" }; $e({ type: "mathord", htmlBuilder(e, t) { return Ve.makeOrd(e, t, "mathord") }, mathmlBuilder(e, t) { const r = new gt.MathNode("mi", [ft(e.text, e.mode, t)]), n = yt(e, t) || "italic"; return n !== hn[r.type] && r.setAttribute("mathvariant", n), r } }), $e({ type: "textord", htmlBuilder(e, t) { return Ve.makeOrd(e, t, "textord") }, mathmlBuilder(e, t) { const r = ft(e.text, e.mode, t), n = yt(e, t) || "normal"; let o; return o = "text" === e.mode ? new gt.MathNode("mtext", [r]) : /[0-9]/.test(e.text) ? new gt.MathNode("mn", [r]) : "\\prime" === e.text ? new gt.MathNode("mo", [r]) : new gt.MathNode("mi", [r]), n !== hn[o.type] && o.setAttribute("mathvariant", n), o } }); const cn = { "\\nobreak": "nobreak", "\\allowbreak": "allowbreak" }, mn = { " ": {}, "\\ ": {}, "~": { className: "nobreak" }, "\\space": {}, "\\nobreakspace": { className: "nobreak" } }; $e({ type: "spacing", htmlBuilder(e, t) { if (mn.hasOwnProperty(e.text)) { const r = mn[e.text].className || ""; if ("text" === e.mode) { const n = Ve.makeOrd(e, t, "textord"); return n.classes.push(r), n } return Ve.makeSpan(["mspace", r], [Ve.mathsym(e.text, e.mode, t)], t) } if (cn.hasOwnProperty(e.text)) return Ve.makeSpan(["mspace", cn[e.text]], [], t); throw new n('Unknown type of space "' + e.text + '"') }, mathmlBuilder(e, t) { let r; if (!mn.hasOwnProperty(e.text)) { if (cn.hasOwnProperty(e.text)) return new gt.MathNode("mspace"); throw new n('Unknown type of space "' + e.text + '"') } return r = new gt.MathNode("mtext", [new gt.TextNode("\xa0")]), r } }); const pn = () => { const e = new gt.MathNode("mtd", []); return e.setAttribute("width", "50%"), e }; $e({ type: "tag", mathmlBuilder(e, t) { const r = new gt.MathNode("mtable", [new gt.MathNode("mtr", [pn(), new gt.MathNode("mtd", [wt(e.body, t)]), pn(), new gt.MathNode("mtd", [wt(e.tag, t)])])]); return r.setAttribute("width", "100%"), r } }); const un = { "\\text": void 0, "\\textrm": "textrm", "\\textsf": "textsf", "\\texttt": "texttt", "\\textnormal": "textrm" }, dn = { "\\textbf": "textbf", "\\textmd": "textmd" }, gn = { "\\textit": "textit", "\\textup": "textup" }, fn = (e, t) => { const r = e.font; return r ? un[r] ? t.withTextFontFamily(un[r]) : dn[r] ? t.withTextFontWeight(dn[r]) : "\\emph" === r ? "textit" === t.fontShape ? t.withTextFontShape("textup") : t.withTextFontShape("textit") : t.withTextFontShape(gn[r]) : t }; je({ type: "text", names: ["\\text", "\\textrm", "\\textsf", "\\texttt", "\\textnormal", "\\textbf", "\\textmd", "\\textit", "\\textup", "\\emph"], props: { numArgs: 1, argTypes: ["text"], allowedInArgument: !0, allowedInText: !0 }, handler(e, t) { let { parser: r, funcName: n } = e; const o = t[0]; return { type: "text", mode: r.mode, body: Ke(o), font: n } }, htmlBuilder(e, t) { const r = fn(e, t), n = nt(e.body, r, !0); return Ve.makeSpan(["mord", "text"], n, r) }, mathmlBuilder(e, t) { const r = fn(e, t); return wt(e.body, r) } }), je({ type: "underline", names: ["\\underline"], props: { numArgs: 1, allowedInText: !0 }, handler(e, t) { let { parser: r } = e; return { type: "underline", mode: r.mode, body: t[0] } }, htmlBuilder(e, t) { const r = ht(e.body, t), n = Ve.makeLineSpan("underline-line", t), o = t.fontMetrics().defaultRuleThickness, s = Ve.makeVList({ positionType: "top", positionData: r.height, children: [{ type: "kern", size: o }, { type: "elem", elem: n }, { type: "kern", size: 3 * o }, { type: "elem", elem: r }] }, t); return Ve.makeSpan(["mord", "underline"], [s], t) }, mathmlBuilder(e, t) { const r = new gt.MathNode("mo", [new gt.TextNode("\u203e")]); r.setAttribute("stretchy", "true"); const n = new gt.MathNode("munder", [vt(e.body, t), r]); return n.setAttribute("accentunder", "true"), n } }), je({ type: "vcenter", names: ["\\vcenter"], props: { numArgs: 1, argTypes: ["original"], allowedInText: !1 }, handler(e, t) { let { parser: r } = e; return { type: "vcenter", mode: r.mode, body: t[0] } }, htmlBuilder(e, t) { const r = ht(e.body, t), n = t.fontMetrics().axisHeight, o = .5 * (r.height - n - (r.depth + n)); return Ve.makeVList({ positionType: "shift", positionData: o, children: [{ type: "elem", elem: r }] }, t) }, mathmlBuilder(e, t) { return new gt.MathNode("mpadded", [vt(e.body, t)], ["vcenter"]) } }), je({ type: "verb", names: ["\\verb"], props: { numArgs: 0, allowedInText: !0 }, handler(e, t, r) { throw new n("\\verb ended by end of line instead of matching delimiter") }, htmlBuilder(e, t) { const r = bn(e), n = [], o = t.havingStyle(t.style.text()); for (let t = 0; t < r.length; t++) { let s = r[t]; "~" === s && (s = "\\textasciitilde"), n.push(Ve.makeSymbol(s, "Typewriter-Regular", e.mode, o, ["mord", "texttt"])) } return Ve.makeSpan(["mord", "text"].concat(o.sizingClasses(t)), Ve.tryCombineChars(n), o) }, mathmlBuilder(e, t) { const r = new gt.TextNode(bn(e)), n = new gt.MathNode("mtext", [r]); return n.setAttribute("mathvariant", "monospace"), n } }); const bn = e => e.body.replace(/ /g, e.star ? "\u2423" : "\xa0"); var yn = Xe; const xn = "[ \r\n\t]", wn = "(\\\\[a-zA-Z@]+)" + xn + "*", vn = "[\u0300-\u036f]", kn = new RegExp(vn + "+$"), Sn = "(" + xn + "+)|\\\\(\n|[ \r\t]+\n?)[ \r\t]*|([!-\\[\\]-\u2027\u202a-\ud7ff\uf900-\uffff]" + vn + "*|[\ud800-\udbff][\udc00-\udfff]" + vn + "*|\\\\verb\\*([^]).*?\\4|\\\\verb([^*a-zA-Z]).*?\\5|" + wn + "|\\\\[^\ud800-\udfff])"; class Mn { constructor(e, t) { this.input = void 0, this.settings = void 0, this.tokenRegex = void 0, this.catcodes = void 0, this.input = e, this.settings = t, this.tokenRegex = new RegExp(Sn, "g"), this.catcodes = { "%": 14, "~": 13 } } setCatcode(e, t) { this.catcodes[e] = t } lex() { const e = this.input, t = this.tokenRegex.lastIndex; if (t === e.length) return new Nr("EOF", new Cr(this, t, t)); const r = this.tokenRegex.exec(e); if (null === r || r.index !== t) throw new n("Unexpected character: '" + e[t] + "'", new Nr(e[t], new Cr(this, t, t + 1))); const o = r[6] || r[3] || (r[2] ? "\\ " : " "); if (14 === this.catcodes[o]) { const t = e.indexOf("\n", this.tokenRegex.lastIndex); return -1 === t ? (this.tokenRegex.lastIndex = e.length, this.settings.reportNonstrict("commentAtEnd", "% comment has no terminating newline; LaTeX would fail because of commenting the end of math mode (e.g. $)")) : this.tokenRegex.lastIndex = t + 1, this.lex() } return new Nr(o, new Cr(this, t, this.tokenRegex.lastIndex)) } } class zn { constructor(e, t) { void 0 === e && (e = {}), void 0 === t && (t = {}), this.current = void 0, this.builtins = void 0, this.undefStack = void 0, this.current = t, this.builtins = e, this.undefStack = [] } beginGroup() { this.undefStack.push({}) } endGroup() { if (0 === this.undefStack.length) throw new n("Unbalanced namespace destruction: attempt to pop global namespace; please report this as a bug"); const e = this.undefStack.pop(); for (const t in e) e.hasOwnProperty(t) && (null == e[t] ? delete this.current[t] : this.current[t] = e[t]) } endGroups() { for (; this.undefStack.length > 0;)this.endGroup() } has(e) { return this.current.hasOwnProperty(e) || this.builtins.hasOwnProperty(e) } get(e) { return this.current.hasOwnProperty(e) ? this.current[e] : this.builtins[e] } set(e, t, r) { if (void 0 === r && (r = !1), r) { for (let t = 0; t < this.undefStack.length; t++)delete this.undefStack[t][e]; this.undefStack.length > 0 && (this.undefStack[this.undefStack.length - 1][e] = t) } else { const t = this.undefStack[this.undefStack.length - 1]; t && !t.hasOwnProperty(e) && (t[e] = this.current[e]) } null == t ? delete this.current[e] : this.current[e] = t } } var An = Tr; Br("\\noexpand", (function (e) { const t = e.popToken(); return e.isExpandable(t.text) && (t.noexpand = !0, t.treatAsRelax = !0), { tokens: [t], numArgs: 0 } })), Br("\\expandafter", (function (e) { const t = e.popToken(); return e.expandOnce(!0), { tokens: [t], numArgs: 0 } })), Br("\\@firstoftwo", (function (e) { return { tokens: e.consumeArgs(2)[0], numArgs: 0 } })), Br("\\@secondoftwo", (function (e) { return { tokens: e.consumeArgs(2)[1], numArgs: 0 } })), Br("\\@ifnextchar", (function (e) { const t = e.consumeArgs(3); e.consumeSpaces(); const r = e.future(); return 1 === t[0].length && t[0][0].text === r.text ? { tokens: t[1], numArgs: 0 } : { tokens: t[2], numArgs: 0 } })), Br("\\@ifstar", "\\@ifnextchar *{\\@firstoftwo{#1}}"), Br("\\TextOrMath", (function (e) { const t = e.consumeArgs(2); return "text" === e.mode ? { tokens: t[0], numArgs: 0 } : { tokens: t[1], numArgs: 0 } })); const Tn = { 0: 0, 1: 1, 2: 2, 3: 3, 4: 4, 5: 5, 6: 6, 7: 7, 8: 8, 9: 9, a: 10, A: 10, b: 11, B: 11, c: 12, C: 12, d: 13, D: 13, e: 14, E: 14, f: 15, F: 15 }; Br("\\char", (function (e) { let t, r = e.popToken(), o = ""; if ("'" === r.text) t = 8, r = e.popToken(); else if ('"' === r.text) t = 16, r = e.popToken(); else if ("`" === r.text) if (r = e.popToken(), "\\" === r.text[0]) o = r.text.charCodeAt(1); else { if ("EOF" === r.text) throw new n("\\char` missing argument"); o = r.text.charCodeAt(0) } else t = 10; if (t) { if (o = Tn[r.text], null == o || o >= t) throw new n("Invalid base-" + t + " digit " + r.text); let s; for (; null != (s = Tn[e.future().text]) && s < t;)o *= t, o += s, e.popToken() } return "\\@char{" + o + "}" })); const Bn = (e, t, r) => { let o = e.consumeArg().tokens; if (1 !== o.length) throw new n("\\newcommand's first argument must be a macro name"); const s = o[0].text, i = e.isDefined(s); if (i && !t) throw new n("\\newcommand{" + s + "} attempting to redefine " + s + "; use \\renewcommand"); if (!i && !r) throw new n("\\renewcommand{" + s + "} when command " + s + " does not yet exist; use \\newcommand"); let a = 0; if (o = e.consumeArg().tokens, 1 === o.length && "[" === o[0].text) { let t = "", r = e.expandNextToken(); for (; "]" !== r.text && "EOF" !== r.text;)t += r.text, r = e.expandNextToken(); if (!t.match(/^\s*[0-9]+\s*$/)) throw new n("Invalid number of arguments: " + t); a = parseInt(t), o = e.consumeArg().tokens } return e.macros.set(s, { tokens: o, numArgs: a }), "" }; Br("\\newcommand", (e => Bn(e, !1, !0))), Br("\\renewcommand", (e => Bn(e, !0, !1))), Br("\\providecommand", (e => Bn(e, !0, !0))), Br("\\message", (e => { const t = e.consumeArgs(1)[0]; return console.log(t.reverse().map((e => e.text)).join("")), "" })), Br("\\errmessage", (e => { const t = e.consumeArgs(1)[0]; return console.error(t.reverse().map((e => e.text)).join("")), "" })), Br("\\show", (e => { const t = e.popToken(), r = t.text; return console.log(t, e.macros.get(r), yn[r], oe.math[r], oe.text[r]), "" })), Br("\\bgroup", "{"), Br("\\egroup", "}"), Br("~", "\\nobreakspace"), Br("\\lq", "`"), Br("\\rq", "'"), Br("\\aa", "\\r a"), Br("\\AA", "\\r A"), Br("\\textcopyright", "\\html@mathml{\\textcircled{c}}{\\char`\xa9}"), Br("\\copyright", "\\TextOrMath{\\textcopyright}{\\text{\\textcopyright}}"), Br("\\textregistered", "\\html@mathml{\\textcircled{\\scriptsize R}}{\\char`\xae}"), Br("\u212c", "\\mathscr{B}"), Br("\u2130", "\\mathscr{E}"), Br("\u2131", "\\mathscr{F}"), Br("\u210b", "\\mathscr{H}"), Br("\u2110", "\\mathscr{I}"), Br("\u2112", "\\mathscr{L}"), Br("\u2133", "\\mathscr{M}"), Br("\u211b", "\\mathscr{R}"), Br("\u212d", "\\mathfrak{C}"), Br("\u210c", "\\mathfrak{H}"), Br("\u2128", "\\mathfrak{Z}"), Br("\\Bbbk", "\\Bbb{k}"), Br("\xb7", "\\cdotp"), Br("\\llap", "\\mathllap{\\textrm{#1}}"), Br("\\rlap", "\\mathrlap{\\textrm{#1}}"), Br("\\clap", "\\mathclap{\\textrm{#1}}"), Br("\\mathstrut", "\\vphantom{(}"), Br("\\underbar", "\\underline{\\text{#1}}"), Br("\\not", '\\html@mathml{\\mathrel{\\mathrlap\\@not}}{\\char"338}'), Br("\\neq", "\\html@mathml{\\mathrel{\\not=}}{\\mathrel{\\char`\u2260}}"), Br("\\ne", "\\neq"), Br("\u2260", "\\neq"), Br("\\notin", "\\html@mathml{\\mathrel{{\\in}\\mathllap{/\\mskip1mu}}}{\\mathrel{\\char`\u2209}}"), Br("\u2209", "\\notin"), Br("\u2258", "\\html@mathml{\\mathrel{=\\kern{-1em}\\raisebox{0.4em}{$\\scriptsize\\frown$}}}{\\mathrel{\\char`\u2258}}"), Br("\u2259", "\\html@mathml{\\stackrel{\\tiny\\wedge}{=}}{\\mathrel{\\char`\u2258}}"), Br("\u225a", "\\html@mathml{\\stackrel{\\tiny\\vee}{=}}{\\mathrel{\\char`\u225a}}"), Br("\u225b", "\\html@mathml{\\stackrel{\\scriptsize\\star}{=}}{\\mathrel{\\char`\u225b}}"), Br("\u225d", "\\html@mathml{\\stackrel{\\tiny\\mathrm{def}}{=}}{\\mathrel{\\char`\u225d}}"), Br("\u225e", "\\html@mathml{\\stackrel{\\tiny\\mathrm{m}}{=}}{\\mathrel{\\char`\u225e}}"), Br("\u225f", "\\html@mathml{\\stackrel{\\tiny?}{=}}{\\mathrel{\\char`\u225f}}"), Br("\u27c2", "\\perp"), Br("\u203c", "\\mathclose{!\\mkern-0.8mu!}"), Br("\u220c", "\\notni"), Br("\u231c", "\\ulcorner"), Br("\u231d", "\\urcorner"), Br("\u231e", "\\llcorner"), Br("\u231f", "\\lrcorner"), Br("\xa9", "\\copyright"), Br("\xae", "\\textregistered"), Br("\ufe0f", "\\textregistered"), Br("\\ulcorner", '\\html@mathml{\\@ulcorner}{\\mathop{\\char"231c}}'), Br("\\urcorner", '\\html@mathml{\\@urcorner}{\\mathop{\\char"231d}}'), Br("\\llcorner", '\\html@mathml{\\@llcorner}{\\mathop{\\char"231e}}'), Br("\\lrcorner", '\\html@mathml{\\@lrcorner}{\\mathop{\\char"231f}}'), Br("\\vdots", "\\mathord{\\varvdots\\rule{0pt}{15pt}}"), Br("\u22ee", "\\vdots"), Br("\\varGamma", "\\mathit{\\Gamma}"), Br("\\varDelta", "\\mathit{\\Delta}"), Br("\\varTheta", "\\mathit{\\Theta}"), Br("\\varLambda", "\\mathit{\\Lambda}"), Br("\\varXi", "\\mathit{\\Xi}"), Br("\\varPi", "\\mathit{\\Pi}"), Br("\\varSigma", "\\mathit{\\Sigma}"), Br("\\varUpsilon", "\\mathit{\\Upsilon}"), Br("\\varPhi", "\\mathit{\\Phi}"), Br("\\varPsi", "\\mathit{\\Psi}"), Br("\\varOmega", "\\mathit{\\Omega}"), Br("\\substack", "\\begin{subarray}{c}#1\\end{subarray}"), Br("\\colon", "\\nobreak\\mskip2mu\\mathpunct{}\\mathchoice{\\mkern-3mu}{\\mkern-3mu}{}{}{:}\\mskip6mu\\relax"), Br("\\boxed", "\\fbox{$\\displaystyle{#1}$}"), Br("\\iff", "\\DOTSB\\;\\Longleftrightarrow\\;"), Br("\\implies", "\\DOTSB\\;\\Longrightarrow\\;"), Br("\\impliedby", "\\DOTSB\\;\\Longleftarrow\\;"); const Cn = { ",": "\\dotsc", "\\not": "\\dotsb", "+": "\\dotsb", "=": "\\dotsb", "<": "\\dotsb", ">": "\\dotsb", "-": "\\dotsb", "*": "\\dotsb", ":": "\\dotsb", "\\DOTSB": "\\dotsb", "\\coprod": "\\dotsb", "\\bigvee": "\\dotsb", "\\bigwedge": "\\dotsb", "\\biguplus": "\\dotsb", "\\bigcap": "\\dotsb", "\\bigcup": "\\dotsb", "\\prod": "\\dotsb", "\\sum": "\\dotsb", "\\bigotimes": "\\dotsb", "\\bigoplus": "\\dotsb", "\\bigodot": "\\dotsb", "\\bigsqcup": "\\dotsb", "\\And": "\\dotsb", "\\longrightarrow": "\\dotsb", "\\Longrightarrow": "\\dotsb", "\\longleftarrow": "\\dotsb", "\\Longleftarrow": "\\dotsb", "\\longleftrightarrow": "\\dotsb", "\\Longleftrightarrow": "\\dotsb", "\\mapsto": "\\dotsb", "\\longmapsto": "\\dotsb", "\\hookrightarrow": "\\dotsb", "\\doteq": "\\dotsb", "\\mathbin": "\\dotsb", "\\mathrel": "\\dotsb", "\\relbar": "\\dotsb", "\\Relbar": "\\dotsb", "\\xrightarrow": "\\dotsb", "\\xleftarrow": "\\dotsb", "\\DOTSI": "\\dotsi", "\\int": "\\dotsi", "\\oint": "\\dotsi", "\\iint": "\\dotsi", "\\iiint": "\\dotsi", "\\iiiint": "\\dotsi", "\\idotsint": "\\dotsi", "\\DOTSX": "\\dotsx" }; Br("\\dots", (function (e) { let t = "\\dotso"; const r = e.expandAfterFuture().text; return r in Cn ? t = Cn[r] : ("\\not" === r.slice(0, 4) || r in oe.math && l.contains(["bin", "rel"], oe.math[r].group)) && (t = "\\dotsb"), t })); const Nn = { ")": !0, "]": !0, "\\rbrack": !0, "\\}": !0, "\\rbrace": !0, "\\rangle": !0, "\\rceil": !0, "\\rfloor": !0, "\\rgroup": !0, "\\rmoustache": !0, "\\right": !0, "\\bigr": !0, "\\biggr": !0, "\\Bigr": !0, "\\Biggr": !0, $: !0, ";": !0, ".": !0, ",": !0 }; Br("\\dotso", (function (e) { return e.future().text in Nn ? "\\ldots\\," : "\\ldots" })), Br("\\dotsc", (function (e) { const t = e.future().text; return t in Nn && "," !== t ? "\\ldots\\," : "\\ldots" })), Br("\\cdots", (function (e) { return e.future().text in Nn ? "\\@cdots\\," : "\\@cdots" })), Br("\\dotsb", "\\cdots"), Br("\\dotsm", "\\cdots"), Br("\\dotsi", "\\!\\cdots"), Br("\\dotsx", "\\ldots\\,"), Br("\\DOTSI", "\\relax"), Br("\\DOTSB", "\\relax"), Br("\\DOTSX", "\\relax"), Br("\\tmspace", "\\TextOrMath{\\kern#1#3}{\\mskip#1#2}\\relax"), Br("\\,", "\\tmspace+{3mu}{.1667em}"), Br("\\thinspace", "\\,"), Br("\\>", "\\mskip{4mu}"), Br("\\:", "\\tmspace+{4mu}{.2222em}"), Br("\\medspace", "\\:"), Br("\\;", "\\tmspace+{5mu}{.2777em}"), Br("\\thickspace", "\\;"), Br("\\!", "\\tmspace-{3mu}{.1667em}"), Br("\\negthinspace", "\\!"), Br("\\negmedspace", "\\tmspace-{4mu}{.2222em}"), Br("\\negthickspace", "\\tmspace-{5mu}{.277em}"), Br("\\enspace", "\\kern.5em "), Br("\\enskip", "\\hskip.5em\\relax"), Br("\\quad", "\\hskip1em\\relax"), Br("\\qquad", "\\hskip2em\\relax"), Br("\\tag", "\\@ifstar\\tag@literal\\tag@paren"), Br("\\tag@paren", "\\tag@literal{({#1})}"), Br("\\tag@literal", (e => { if (e.macros.get("\\df@tag")) throw new n("Multiple \\tag"); return "\\gdef\\df@tag{\\text{#1}}" })), Br("\\bmod", "\\mathchoice{\\mskip1mu}{\\mskip1mu}{\\mskip5mu}{\\mskip5mu}\\mathbin{\\rm mod}\\mathchoice{\\mskip1mu}{\\mskip1mu}{\\mskip5mu}{\\mskip5mu}"), Br("\\pod", "\\allowbreak\\mathchoice{\\mkern18mu}{\\mkern8mu}{\\mkern8mu}{\\mkern8mu}(#1)"), Br("\\pmod", "\\pod{{\\rm mod}\\mkern6mu#1}"), Br("\\mod", "\\allowbreak\\mathchoice{\\mkern18mu}{\\mkern12mu}{\\mkern12mu}{\\mkern12mu}{\\rm mod}\\,\\,#1"), Br("\\newline", "\\\\\\relax"), Br("\\TeX", "\\textrm{\\html@mathml{T\\kern-.1667em\\raisebox{-.5ex}{E}\\kern-.125emX}{TeX}}"); const qn = F(T["Main-Regular"]["T".charCodeAt(0)][1] - .7 * T["Main-Regular"]["A".charCodeAt(0)][1]); Br("\\LaTeX", "\\textrm{\\html@mathml{L\\kern-.36em\\raisebox{" + qn + "}{\\scriptstyle A}\\kern-.15em\\TeX}{LaTeX}}"), Br("\\KaTeX", "\\textrm{\\html@mathml{K\\kern-.17em\\raisebox{" + qn + "}{\\scriptstyle A}\\kern-.15em\\TeX}{KaTeX}}"), Br("\\hspace", "\\@ifstar\\@hspacer\\@hspace"), Br("\\@hspace", "\\hskip #1\\relax"), Br("\\@hspacer", "\\rule{0pt}{0pt}\\hskip #1\\relax"), Br("\\ordinarycolon", ":"), Br("\\vcentcolon", "\\mathrel{\\mathop\\ordinarycolon}"), Br("\\dblcolon", '\\html@mathml{\\mathrel{\\vcentcolon\\mathrel{\\mkern-.9mu}\\vcentcolon}}{\\mathop{\\char"2237}}'), Br("\\coloneqq", '\\html@mathml{\\mathrel{\\vcentcolon\\mathrel{\\mkern-1.2mu}=}}{\\mathop{\\char"2254}}'), Br("\\Coloneqq", '\\html@mathml{\\mathrel{\\dblcolon\\mathrel{\\mkern-1.2mu}=}}{\\mathop{\\char"2237\\char"3d}}'), Br("\\coloneq", '\\html@mathml{\\mathrel{\\vcentcolon\\mathrel{\\mkern-1.2mu}\\mathrel{-}}}{\\mathop{\\char"3a\\char"2212}}'), Br("\\Coloneq", '\\html@mathml{\\mathrel{\\dblcolon\\mathrel{\\mkern-1.2mu}\\mathrel{-}}}{\\mathop{\\char"2237\\char"2212}}'), Br("\\eqqcolon", '\\html@mathml{\\mathrel{=\\mathrel{\\mkern-1.2mu}\\vcentcolon}}{\\mathop{\\char"2255}}'), Br("\\Eqqcolon", '\\html@mathml{\\mathrel{=\\mathrel{\\mkern-1.2mu}\\dblcolon}}{\\mathop{\\char"3d\\char"2237}}'), Br("\\eqcolon", '\\html@mathml{\\mathrel{\\mathrel{-}\\mathrel{\\mkern-1.2mu}\\vcentcolon}}{\\mathop{\\char"2239}}'), Br("\\Eqcolon", '\\html@mathml{\\mathrel{\\mathrel{-}\\mathrel{\\mkern-1.2mu}\\dblcolon}}{\\mathop{\\char"2212\\char"2237}}'), Br("\\colonapprox", '\\html@mathml{\\mathrel{\\vcentcolon\\mathrel{\\mkern-1.2mu}\\approx}}{\\mathop{\\char"3a\\char"2248}}'), Br("\\Colonapprox", '\\html@mathml{\\mathrel{\\dblcolon\\mathrel{\\mkern-1.2mu}\\approx}}{\\mathop{\\char"2237\\char"2248}}'), Br("\\colonsim", '\\html@mathml{\\mathrel{\\vcentcolon\\mathrel{\\mkern-1.2mu}\\sim}}{\\mathop{\\char"3a\\char"223c}}'), Br("\\Colonsim", '\\html@mathml{\\mathrel{\\dblcolon\\mathrel{\\mkern-1.2mu}\\sim}}{\\mathop{\\char"2237\\char"223c}}'), Br("\u2237", "\\dblcolon"), Br("\u2239", "\\eqcolon"), Br("\u2254", "\\coloneqq"), Br("\u2255", "\\eqqcolon"), Br("\u2a74", "\\Coloneqq"), Br("\\ratio", "\\vcentcolon"), Br("\\coloncolon", "\\dblcolon"), Br("\\colonequals", "\\coloneqq"), Br("\\coloncolonequals", "\\Coloneqq"), Br("\\equalscolon", "\\eqqcolon"), Br("\\equalscoloncolon", "\\Eqqcolon"), Br("\\colonminus", "\\coloneq"), Br("\\coloncolonminus", "\\Coloneq"), Br("\\minuscolon", "\\eqcolon"), Br("\\minuscoloncolon", "\\Eqcolon"), Br("\\coloncolonapprox", "\\Colonapprox"), Br("\\coloncolonsim", "\\Colonsim"), Br("\\simcolon", "\\mathrel{\\sim\\mathrel{\\mkern-1.2mu}\\vcentcolon}"), Br("\\simcoloncolon", "\\mathrel{\\sim\\mathrel{\\mkern-1.2mu}\\dblcolon}"), Br("\\approxcolon", "\\mathrel{\\approx\\mathrel{\\mkern-1.2mu}\\vcentcolon}"), Br("\\approxcoloncolon", "\\mathrel{\\approx\\mathrel{\\mkern-1.2mu}\\dblcolon}"), Br("\\notni", "\\html@mathml{\\not\\ni}{\\mathrel{\\char`\u220c}}"), Br("\\limsup", "\\DOTSB\\operatorname*{lim\\,sup}"), Br("\\liminf", "\\DOTSB\\operatorname*{lim\\,inf}"), Br("\\injlim", "\\DOTSB\\operatorname*{inj\\,lim}"), Br("\\projlim", "\\DOTSB\\operatorname*{proj\\,lim}"), Br("\\varlimsup", "\\DOTSB\\operatorname*{\\overline{lim}}"), Br("\\varliminf", "\\DOTSB\\operatorname*{\\underline{lim}}"), Br("\\varinjlim", "\\DOTSB\\operatorname*{\\underrightarrow{lim}}"), Br("\\varprojlim", "\\DOTSB\\operatorname*{\\underleftarrow{lim}}"), Br("\\gvertneqq", "\\html@mathml{\\@gvertneqq}{\u2269}"), Br("\\lvertneqq", "\\html@mathml{\\@lvertneqq}{\u2268}"), Br("\\ngeqq", "\\html@mathml{\\@ngeqq}{\u2271}"), Br("\\ngeqslant", "\\html@mathml{\\@ngeqslant}{\u2271}"), Br("\\nleqq", "\\html@mathml{\\@nleqq}{\u2270}"), Br("\\nleqslant", "\\html@mathml{\\@nleqslant}{\u2270}"), Br("\\nshortmid", "\\html@mathml{\\@nshortmid}{\u2224}"), Br("\\nshortparallel", "\\html@mathml{\\@nshortparallel}{\u2226}"), Br("\\nsubseteqq", "\\html@mathml{\\@nsubseteqq}{\u2288}"), Br("\\nsupseteqq", "\\html@mathml{\\@nsupseteqq}{\u2289}"), Br("\\varsubsetneq", "\\html@mathml{\\@varsubsetneq}{\u228a}"), Br("\\varsubsetneqq", "\\html@mathml{\\@varsubsetneqq}{\u2acb}"), Br("\\varsupsetneq", "\\html@mathml{\\@varsupsetneq}{\u228b}"), Br("\\varsupsetneqq", "\\html@mathml{\\@varsupsetneqq}{\u2acc}"), Br("\\imath", "\\html@mathml{\\@imath}{\u0131}"), Br("\\jmath", "\\html@mathml{\\@jmath}{\u0237}"), Br("\\llbracket", "\\html@mathml{\\mathopen{[\\mkern-3.2mu[}}{\\mathopen{\\char`\u27e6}}"), Br("\\rrbracket", "\\html@mathml{\\mathclose{]\\mkern-3.2mu]}}{\\mathclose{\\char`\u27e7}}"), Br("\u27e6", "\\llbracket"), Br("\u27e7", "\\rrbracket"), Br("\\lBrace", "\\html@mathml{\\mathopen{\\{\\mkern-3.2mu[}}{\\mathopen{\\char`\u2983}}"), Br("\\rBrace", "\\html@mathml{\\mathclose{]\\mkern-3.2mu\\}}}{\\mathclose{\\char`\u2984}}"), Br("\u2983", "\\lBrace"), Br("\u2984", "\\rBrace"), Br("\\minuso", "\\mathbin{\\html@mathml{{\\mathrlap{\\mathchoice{\\kern{0.145em}}{\\kern{0.145em}}{\\kern{0.1015em}}{\\kern{0.0725em}}\\circ}{-}}}{\\char`\u29b5}}"), Br("\u29b5", "\\minuso"), Br("\\darr", "\\downarrow"), Br("\\dArr", "\\Downarrow"), Br("\\Darr", "\\Downarrow"), Br("\\lang", "\\langle"), Br("\\rang", "\\rangle"), Br("\\uarr", "\\uparrow"), Br("\\uArr", "\\Uparrow"), Br("\\Uarr", "\\Uparrow"), Br("\\N", "\\mathbb{N}"), Br("\\R", "\\mathbb{R}"), Br("\\Z", "\\mathbb{Z}"), Br("\\alef", "\\aleph"), Br("\\alefsym", "\\aleph"), Br("\\Alpha", "\\mathrm{A}"), Br("\\Beta", "\\mathrm{B}"), Br("\\bull", "\\bullet"), Br("\\Chi", "\\mathrm{X}"), Br("\\clubs", "\\clubsuit"), Br("\\cnums", "\\mathbb{C}"), Br("\\Complex", "\\mathbb{C}"), Br("\\Dagger", "\\ddagger"), Br("\\diamonds", "\\diamondsuit"), Br("\\empty", "\\emptyset"), Br("\\Epsilon", "\\mathrm{E}"), Br("\\Eta", "\\mathrm{H}"), Br("\\exist", "\\exists"), Br("\\harr", "\\leftrightarrow"), Br("\\hArr", "\\Leftrightarrow"), Br("\\Harr", "\\Leftrightarrow"), Br("\\hearts", "\\heartsuit"), Br("\\image", "\\Im"), Br("\\infin", "\\infty"), Br("\\Iota", "\\mathrm{I}"), Br("\\isin", "\\in"), Br("\\Kappa", "\\mathrm{K}"), Br("\\larr", "\\leftarrow"), Br("\\lArr", "\\Leftarrow"), Br("\\Larr", "\\Leftarrow"), Br("\\lrarr", "\\leftrightarrow"), Br("\\lrArr", "\\Leftrightarrow"), Br("\\Lrarr", "\\Leftrightarrow"), Br("\\Mu", "\\mathrm{M}"), Br("\\natnums", "\\mathbb{N}"), Br("\\Nu", "\\mathrm{N}"), Br("\\Omicron", "\\mathrm{O}"), Br("\\plusmn", "\\pm"), Br("\\rarr", "\\rightarrow"), Br("\\rArr", "\\Rightarrow"), Br("\\Rarr", "\\Rightarrow"), Br("\\real", "\\Re"), Br("\\reals", "\\mathbb{R}"), Br("\\Reals", "\\mathbb{R}"), Br("\\Rho", "\\mathrm{P}"), Br("\\sdot", "\\cdot"), Br("\\sect", "\\S"), Br("\\spades", "\\spadesuit"), Br("\\sub", "\\subset"), Br("\\sube", "\\subseteq"), Br("\\supe", "\\supseteq"), Br("\\Tau", "\\mathrm{T}"), Br("\\thetasym", "\\vartheta"), Br("\\weierp", "\\wp"), Br("\\Zeta", "\\mathrm{Z}"), Br("\\argmin", "\\DOTSB\\operatorname*{arg\\,min}"), Br("\\argmax", "\\DOTSB\\operatorname*{arg\\,max}"), Br("\\plim", "\\DOTSB\\mathop{\\operatorname{plim}}\\limits"), Br("\\bra", "\\mathinner{\\langle{#1}|}"), Br("\\ket", "\\mathinner{|{#1}\\rangle}"), Br("\\braket", "\\mathinner{\\langle{#1}\\rangle}"), Br("\\Bra", "\\left\\langle#1\\right|"), Br("\\Ket", "\\left|#1\\right\\rangle"); const In = e => t => { const r = t.consumeArg().tokens, n = t.consumeArg().tokens, o = t.consumeArg().tokens, s = t.consumeArg().tokens, i = t.macros.get("|"), a = t.macros.get("\\|"); t.macros.beginGroup(); const l = t => r => { e && (r.macros.set("|", i), o.length && r.macros.set("\\|", a)); let s = t; if (!t && o.length) { "|" === r.future().text && (r.popToken(), s = !0) } return { tokens: s ? o : n, numArgs: 0 } }; t.macros.set("|", l(!1)), o.length && t.macros.set("\\|", l(!0)); const h = t.consumeArg().tokens, c = t.expandTokens([...s, ...h, ...r]); return t.macros.endGroup(), { tokens: c.reverse(), numArgs: 0 } }; Br("\\bra@ket", In(!1)), Br("\\bra@set", In(!0)), Br("\\Braket", "\\bra@ket{\\left\\langle}{\\,\\middle\\vert\\,}{\\,\\middle\\vert\\,}{\\right\\rangle}"), Br("\\Set", "\\bra@set{\\left\\{\\:}{\\;\\middle\\vert\\;}{\\;\\middle\\Vert\\;}{\\:\\right\\}}"), Br("\\set", "\\bra@set{\\{\\,}{\\mid}{}{\\,\\}}"), Br("\\angln", "{\\angl n}"), Br("\\blue", "\\textcolor{##6495ed}{#1}"), Br("\\orange", "\\textcolor{##ffa500}{#1}"), Br("\\pink", "\\textcolor{##ff00af}{#1}"), Br("\\red", "\\textcolor{##df0030}{#1}"), Br("\\green", "\\textcolor{##28ae7b}{#1}"), Br("\\gray", "\\textcolor{gray}{#1}"), Br("\\purple", "\\textcolor{##9d38bd}{#1}"), Br("\\blueA", "\\textcolor{##ccfaff}{#1}"), Br("\\blueB", "\\textcolor{##80f6ff}{#1}"), Br("\\blueC", "\\textcolor{##63d9ea}{#1}"), Br("\\blueD", "\\textcolor{##11accd}{#1}"), Br("\\blueE", "\\textcolor{##0c7f99}{#1}"), Br("\\tealA", "\\textcolor{##94fff5}{#1}"), Br("\\tealB", "\\textcolor{##26edd5}{#1}"), Br("\\tealC", "\\textcolor{##01d1c1}{#1}"), Br("\\tealD", "\\textcolor{##01a995}{#1}"), Br("\\tealE", "\\textcolor{##208170}{#1}"), Br("\\greenA", "\\textcolor{##b6ffb0}{#1}"), Br("\\greenB", "\\textcolor{##8af281}{#1}"), Br("\\greenC", "\\textcolor{##74cf70}{#1}"), Br("\\greenD", "\\textcolor{##1fab54}{#1}"), Br("\\greenE", "\\textcolor{##0d923f}{#1}"), Br("\\goldA", "\\textcolor{##ffd0a9}{#1}"), Br("\\goldB", "\\textcolor{##ffbb71}{#1}"), Br("\\goldC", "\\textcolor{##ff9c39}{#1}"), Br("\\goldD", "\\textcolor{##e07d10}{#1}"), Br("\\goldE", "\\textcolor{##a75a05}{#1}"), Br("\\redA", "\\textcolor{##fca9a9}{#1}"), Br("\\redB", "\\textcolor{##ff8482}{#1}"), Br("\\redC", "\\textcolor{##f9685d}{#1}"), Br("\\redD", "\\textcolor{##e84d39}{#1}"), Br("\\redE", "\\textcolor{##bc2612}{#1}"), Br("\\maroonA", "\\textcolor{##ffbde0}{#1}"), Br("\\maroonB", "\\textcolor{##ff92c6}{#1}"), Br("\\maroonC", "\\textcolor{##ed5fa6}{#1}"), Br("\\maroonD", "\\textcolor{##ca337c}{#1}"), Br("\\maroonE", "\\textcolor{##9e034e}{#1}"), Br("\\purpleA", "\\textcolor{##ddd7ff}{#1}"), Br("\\purpleB", "\\textcolor{##c6b9fc}{#1}"), Br("\\purpleC", "\\textcolor{##aa87ff}{#1}"), Br("\\purpleD", "\\textcolor{##7854ab}{#1}"), Br("\\purpleE", "\\textcolor{##543b78}{#1}"), Br("\\mintA", "\\textcolor{##f5f9e8}{#1}"), Br("\\mintB", "\\textcolor{##edf2df}{#1}"), Br("\\mintC", "\\textcolor{##e0e5cc}{#1}"), Br("\\grayA", "\\textcolor{##f6f7f7}{#1}"), Br("\\grayB", "\\textcolor{##f0f1f2}{#1}"), Br("\\grayC", "\\textcolor{##e3e5e6}{#1}"), Br("\\grayD", "\\textcolor{##d6d8da}{#1}"), Br("\\grayE", "\\textcolor{##babec2}{#1}"), Br("\\grayF", "\\textcolor{##888d93}{#1}"), Br("\\grayG", "\\textcolor{##626569}{#1}"), Br("\\grayH", "\\textcolor{##3b3e40}{#1}"), Br("\\grayI", "\\textcolor{##21242c}{#1}"), Br("\\kaBlue", "\\textcolor{##314453}{#1}"), Br("\\kaGreen", "\\textcolor{##71B307}{#1}"); const Rn = { "^": !0, _: !0, "\\limits": !0, "\\nolimits": !0 }; class Hn { constructor(e, t, r) { this.settings = void 0, this.expansionCount = void 0, this.lexer = void 0, this.macros = void 0, this.stack = void 0, this.mode = void 0, this.settings = t, this.expansionCount = 0, this.feed(e), this.macros = new zn(An, t.macros), this.mode = r, this.stack = [] } feed(e) { this.lexer = new Mn(e, this.settings) } switchMode(e) { this.mode = e } beginGroup() { this.macros.beginGroup() } endGroup() { this.macros.endGroup() } endGroups() { this.macros.endGroups() } future() { return 0 === this.stack.length && this.pushToken(this.lexer.lex()), this.stack[this.stack.length - 1] } popToken() { return this.future(), this.stack.pop() } pushToken(e) { this.stack.push(e) } pushTokens(e) { this.stack.push(...e) } scanArgument(e) { let t, r, n; if (e) { if (this.consumeSpaces(), "[" !== this.future().text) return null; t = this.popToken(), ({ tokens: n, end: r } = this.consumeArg(["]"])) } else ({ tokens: n, start: t, end: r } = this.consumeArg()); return this.pushToken(new Nr("EOF", r.loc)), this.pushTokens(n), t.range(r, "") } consumeSpaces() { for (; ;) { if (" " !== this.future().text) break; this.stack.pop() } } consumeArg(e) { const t = [], r = e && e.length > 0; r || this.consumeSpaces(); const o = this.future(); let s, i = 0, a = 0; do { if (s = this.popToken(), t.push(s), "{" === s.text) ++i; else if ("}" === s.text) { if (--i, -1 === i) throw new n("Extra }", s) } else if ("EOF" === s.text) throw new n("Unexpected end of input in a macro argument, expected '" + (e && r ? e[a] : "}") + "'", s); if (e && r) if ((0 === i || 1 === i && "{" === e[a]) && s.text === e[a]) { if (++a, a === e.length) { t.splice(-a, a); break } } else a = 0 } while (0 !== i || r); return "{" === o.text && "}" === t[t.length - 1].text && (t.pop(), t.shift()), t.reverse(), { tokens: t, start: o, end: s } } consumeArgs(e, t) { if (t) { if (t.length !== e + 1) throw new n("The length of delimiters doesn't match the number of args!"); const r = t[0]; for (let e = 0; e < r.length; e++) { const t = this.popToken(); if (r[e] !== t.text) throw new n("Use of the macro doesn't match its definition", t) } } const r = []; for (let n = 0; n < e; n++)r.push(this.consumeArg(t && t[n + 1]).tokens); return r } countExpansion(e) { if (this.expansionCount += e, this.expansionCount > this.settings.maxExpand) throw new n("Too many expansions: infinite loop or need to increase maxExpand setting") } expandOnce(e) { const t = this.popToken(), r = t.text, o = t.noexpand ? null : this._getExpansion(r); if (null == o || e && o.unexpandable) { if (e && null == o && "\\" === r[0] && !this.isDefined(r)) throw new n("Undefined control sequence: " + r); return this.pushToken(t), !1 } this.countExpansion(1); let s = o.tokens; const i = this.consumeArgs(o.numArgs, o.delimiters); if (o.numArgs) { s = s.slice(); for (let e = s.length - 1; e >= 0; --e) { let t = s[e]; if ("#" === t.text) { if (0 === e) throw new n("Incomplete placeholder at end of macro body", t); if (t = s[--e], "#" === t.text) s.splice(e + 1, 1); else { if (!/^[1-9]$/.test(t.text)) throw new n("Not a valid argument number", t); s.splice(e, 2, ...i[+t.text - 1]) } } } } return this.pushTokens(s), s.length } expandAfterFuture() { return this.expandOnce(), this.future() } expandNextToken() { for (; ;)if (!1 === this.expandOnce()) { const e = this.stack.pop(); return e.treatAsRelax && (e.text = "\\relax"), e } throw new Error } expandMacro(e) { return this.macros.has(e) ? this.expandTokens([new Nr(e)]) : void 0 } expandTokens(e) { const t = [], r = this.stack.length; for (this.pushTokens(e); this.stack.length > r;)if (!1 === this.expandOnce(!0)) { const e = this.stack.pop(); e.treatAsRelax && (e.noexpand = !1, e.treatAsRelax = !1), t.push(e) } return this.countExpansion(t.length), t } expandMacroAsText(e) { const t = this.expandMacro(e); return t ? t.map((e => e.text)).join("") : t } _getExpansion(e) { const t = this.macros.get(e); if (null == t) return t; if (1 === e.length) { const t = this.lexer.catcodes[e]; if (null != t && 13 !== t) return } const r = "function" == typeof t ? t(this) : t; if ("string" == typeof r) { let e = 0; if (-1 !== r.indexOf("#")) { const t = r.replace(/##/g, ""); for (; -1 !== t.indexOf("#" + (e + 1));)++e } const t = new Mn(r, this.settings), n = []; let o = t.lex(); for (; "EOF" !== o.text;)n.push(o), o = t.lex(); n.reverse(); return { tokens: n, numArgs: e } } return r } isDefined(e) { return this.macros.has(e) || yn.hasOwnProperty(e) || oe.math.hasOwnProperty(e) || oe.text.hasOwnProperty(e) || Rn.hasOwnProperty(e) } isExpandable(e) { const t = this.macros.get(e); return null != t ? "string" == typeof t || "function" == typeof t || !t.unexpandable : yn.hasOwnProperty(e) && !yn[e].primitive } } const On = /^[\u208a\u208b\u208c\u208d\u208e\u2080\u2081\u2082\u2083\u2084\u2085\u2086\u2087\u2088\u2089\u2090\u2091\u2095\u1d62\u2c7c\u2096\u2097\u2098\u2099\u2092\u209a\u1d63\u209b\u209c\u1d64\u1d65\u2093\u1d66\u1d67\u1d68\u1d69\u1d6a]/, En = Object.freeze({ "\u208a": "+", "\u208b": "-", "\u208c": "=", "\u208d": "(", "\u208e": ")", "\u2080": "0", "\u2081": "1", "\u2082": "2", "\u2083": "3", "\u2084": "4", "\u2085": "5", "\u2086": "6", "\u2087": "7", "\u2088": "8", "\u2089": "9", "\u2090": "a", "\u2091": "e", "\u2095": "h", "\u1d62": "i", "\u2c7c": "j", "\u2096": "k", "\u2097": "l", "\u2098": "m", "\u2099": "n", "\u2092": "o", "\u209a": "p", "\u1d63": "r", "\u209b": "s", "\u209c": "t", "\u1d64": "u", "\u1d65": "v", "\u2093": "x", "\u1d66": "\u03b2", "\u1d67": "\u03b3", "\u1d68": "\u03c1", "\u1d69": "\u03d5", "\u1d6a": "\u03c7", "\u207a": "+", "\u207b": "-", "\u207c": "=", "\u207d": "(", "\u207e": ")", "\u2070": "0", "\xb9": "1", "\xb2": "2", "\xb3": "3", "\u2074": "4", "\u2075": "5", "\u2076": "6", "\u2077": "7", "\u2078": "8", "\u2079": "9", "\u1d2c": "A", "\u1d2e": "B", "\u1d30": "D", "\u1d31": "E", "\u1d33": "G", "\u1d34": "H", "\u1d35": "I", "\u1d36": "J", "\u1d37": "K", "\u1d38": "L", "\u1d39": "M", "\u1d3a": "N", "\u1d3c": "O", "\u1d3e": "P", "\u1d3f": "R", "\u1d40": "T", "\u1d41": "U", "\u2c7d": "V", "\u1d42": "W", "\u1d43": "a", "\u1d47": "b", "\u1d9c": "c", "\u1d48": "d", "\u1d49": "e", "\u1da0": "f", "\u1d4d": "g", "\u02b0": "h", "\u2071": "i", "\u02b2": "j", "\u1d4f": "k", "\u02e1": "l", "\u1d50": "m", "\u207f": "n", "\u1d52": "o", "\u1d56": "p", "\u02b3": "r", "\u02e2": "s", "\u1d57": "t", "\u1d58": "u", "\u1d5b": "v", "\u02b7": "w", "\u02e3": "x", "\u02b8": "y", "\u1dbb": "z", "\u1d5d": "\u03b2", "\u1d5e": "\u03b3", "\u1d5f": "\u03b4", "\u1d60": "\u03d5", "\u1d61": "\u03c7", "\u1dbf": "\u03b8" }), Ln = { "\u0301": { text: "\\'", math: "\\acute" }, "\u0300": { text: "\\`", math: "\\grave" }, "\u0308": { text: '\\"', math: "\\ddot" }, "\u0303": { text: "\\~", math: "\\tilde" }, "\u0304": { text: "\\=", math: "\\bar" }, "\u0306": { text: "\\u", math: "\\breve" }, "\u030c": { text: "\\v", math: "\\check" }, "\u0302": { text: "\\^", math: "\\hat" }, "\u0307": { text: "\\.", math: "\\dot" }, "\u030a": { text: "\\r", math: "\\mathring" }, "\u030b": { text: "\\H" }, "\u0327": { text: "\\c" } }, Dn = { "\xe1": "a\u0301", "\xe0": "a\u0300", "\xe4": "a\u0308", "\u01df": "a\u0308\u0304", "\xe3": "a\u0303", "\u0101": "a\u0304", "\u0103": "a\u0306", "\u1eaf": "a\u0306\u0301", "\u1eb1": "a\u0306\u0300", "\u1eb5": "a\u0306\u0303", "\u01ce": "a\u030c", "\xe2": "a\u0302", "\u1ea5": "a\u0302\u0301", "\u1ea7": "a\u0302\u0300", "\u1eab": "a\u0302\u0303", "\u0227": "a\u0307", "\u01e1": "a\u0307\u0304", "\xe5": "a\u030a", "\u01fb": "a\u030a\u0301", "\u1e03": "b\u0307", "\u0107": "c\u0301", "\u1e09": "c\u0327\u0301", "\u010d": "c\u030c", "\u0109": "c\u0302", "\u010b": "c\u0307", "\xe7": "c\u0327", "\u010f": "d\u030c", "\u1e0b": "d\u0307", "\u1e11": "d\u0327", "\xe9": "e\u0301", "\xe8": "e\u0300", "\xeb": "e\u0308", "\u1ebd": "e\u0303", "\u0113": "e\u0304", "\u1e17": "e\u0304\u0301", "\u1e15": "e\u0304\u0300", "\u0115": "e\u0306", "\u1e1d": "e\u0327\u0306", "\u011b": "e\u030c", "\xea": "e\u0302", "\u1ebf": "e\u0302\u0301", "\u1ec1": "e\u0302\u0300", "\u1ec5": "e\u0302\u0303", "\u0117": "e\u0307", "\u0229": "e\u0327", "\u1e1f": "f\u0307", "\u01f5": "g\u0301", "\u1e21": "g\u0304", "\u011f": "g\u0306", "\u01e7": "g\u030c", "\u011d": "g\u0302", "\u0121": "g\u0307", "\u0123": "g\u0327", "\u1e27": "h\u0308", "\u021f": "h\u030c", "\u0125": "h\u0302", "\u1e23": "h\u0307", "\u1e29": "h\u0327", "\xed": "i\u0301", "\xec": "i\u0300", "\xef": "i\u0308", "\u1e2f": "i\u0308\u0301", "\u0129": "i\u0303", "\u012b": "i\u0304", "\u012d": "i\u0306", "\u01d0": "i\u030c", "\xee": "i\u0302", "\u01f0": "j\u030c", "\u0135": "j\u0302", "\u1e31": "k\u0301", "\u01e9": "k\u030c", "\u0137": "k\u0327", "\u013a": "l\u0301", "\u013e": "l\u030c", "\u013c": "l\u0327", "\u1e3f": "m\u0301", "\u1e41": "m\u0307", "\u0144": "n\u0301", "\u01f9": "n\u0300", "\xf1": "n\u0303", "\u0148": "n\u030c", "\u1e45": "n\u0307", "\u0146": "n\u0327", "\xf3": "o\u0301", "\xf2": "o\u0300", "\xf6": "o\u0308", "\u022b": "o\u0308\u0304", "\xf5": "o\u0303", "\u1e4d": "o\u0303\u0301", "\u1e4f": "o\u0303\u0308", "\u022d": "o\u0303\u0304", "\u014d": "o\u0304", "\u1e53": "o\u0304\u0301", "\u1e51": "o\u0304\u0300", "\u014f": "o\u0306", "\u01d2": "o\u030c", "\xf4": "o\u0302", "\u1ed1": "o\u0302\u0301", "\u1ed3": "o\u0302\u0300", "\u1ed7": "o\u0302\u0303", "\u022f": "o\u0307", "\u0231": "o\u0307\u0304", "\u0151": "o\u030b", "\u1e55": "p\u0301", "\u1e57": "p\u0307", "\u0155": "r\u0301", "\u0159": "r\u030c", "\u1e59": "r\u0307", "\u0157": "r\u0327", "\u015b": "s\u0301", "\u1e65": "s\u0301\u0307", "\u0161": "s\u030c", "\u1e67": "s\u030c\u0307", "\u015d": "s\u0302", "\u1e61": "s\u0307", "\u015f": "s\u0327", "\u1e97": "t\u0308", "\u0165": "t\u030c", "\u1e6b": "t\u0307", "\u0163": "t\u0327", "\xfa": "u\u0301", "\xf9": "u\u0300", "\xfc": "u\u0308", "\u01d8": "u\u0308\u0301", "\u01dc": "u\u0308\u0300", "\u01d6": "u\u0308\u0304", "\u01da": "u\u0308\u030c", "\u0169": "u\u0303", "\u1e79": "u\u0303\u0301", "\u016b": "u\u0304", "\u1e7b": "u\u0304\u0308", "\u016d": "u\u0306", "\u01d4": "u\u030c", "\xfb": "u\u0302", "\u016f": "u\u030a", "\u0171": "u\u030b", "\u1e7d": "v\u0303", "\u1e83": "w\u0301", "\u1e81": "w\u0300", "\u1e85": "w\u0308", "\u0175": "w\u0302", "\u1e87": "w\u0307", "\u1e98": "w\u030a", "\u1e8d": "x\u0308", "\u1e8b": "x\u0307", "\xfd": "y\u0301", "\u1ef3": "y\u0300", "\xff": "y\u0308", "\u1ef9": "y\u0303", "\u0233": "y\u0304", "\u0177": "y\u0302", "\u1e8f": "y\u0307", "\u1e99": "y\u030a", "\u017a": "z\u0301", "\u017e": "z\u030c", "\u1e91": "z\u0302", "\u017c": "z\u0307", "\xc1": "A\u0301", "\xc0": "A\u0300", "\xc4": "A\u0308", "\u01de": "A\u0308\u0304", "\xc3": "A\u0303", "\u0100": "A\u0304", "\u0102": "A\u0306", "\u1eae": "A\u0306\u0301", "\u1eb0": "A\u0306\u0300", "\u1eb4": "A\u0306\u0303", "\u01cd": "A\u030c", "\xc2": "A\u0302", "\u1ea4": "A\u0302\u0301", "\u1ea6": "A\u0302\u0300", "\u1eaa": "A\u0302\u0303", "\u0226": "A\u0307", "\u01e0": "A\u0307\u0304", "\xc5": "A\u030a", "\u01fa": "A\u030a\u0301", "\u1e02": "B\u0307", "\u0106": "C\u0301", "\u1e08": "C\u0327\u0301", "\u010c": "C\u030c", "\u0108": "C\u0302", "\u010a": "C\u0307", "\xc7": "C\u0327", "\u010e": "D\u030c", "\u1e0a": "D\u0307", "\u1e10": "D\u0327", "\xc9": "E\u0301", "\xc8": "E\u0300", "\xcb": "E\u0308", "\u1ebc": "E\u0303", "\u0112": "E\u0304", "\u1e16": "E\u0304\u0301", "\u1e14": "E\u0304\u0300", "\u0114": "E\u0306", "\u1e1c": "E\u0327\u0306", "\u011a": "E\u030c", "\xca": "E\u0302", "\u1ebe": "E\u0302\u0301", "\u1ec0": "E\u0302\u0300", "\u1ec4": "E\u0302\u0303", "\u0116": "E\u0307", "\u0228": "E\u0327", "\u1e1e": "F\u0307", "\u01f4": "G\u0301", "\u1e20": "G\u0304", "\u011e": "G\u0306", "\u01e6": "G\u030c", "\u011c": "G\u0302", "\u0120": "G\u0307", "\u0122": "G\u0327", "\u1e26": "H\u0308", "\u021e": "H\u030c", "\u0124": "H\u0302", "\u1e22": "H\u0307", "\u1e28": "H\u0327", "\xcd": "I\u0301", "\xcc": "I\u0300", "\xcf": "I\u0308", "\u1e2e": "I\u0308\u0301", "\u0128": "I\u0303", "\u012a": "I\u0304", "\u012c": "I\u0306", "\u01cf": "I\u030c", "\xce": "I\u0302", "\u0130": "I\u0307", "\u0134": "J\u0302", "\u1e30": "K\u0301", "\u01e8": "K\u030c", "\u0136": "K\u0327", "\u0139": "L\u0301", "\u013d": "L\u030c", "\u013b": "L\u0327", "\u1e3e": "M\u0301", "\u1e40": "M\u0307", "\u0143": "N\u0301", "\u01f8": "N\u0300", "\xd1": "N\u0303", "\u0147": "N\u030c", "\u1e44": "N\u0307", "\u0145": "N\u0327", "\xd3": "O\u0301", "\xd2": "O\u0300", "\xd6": "O\u0308", "\u022a": "O\u0308\u0304", "\xd5": "O\u0303", "\u1e4c": "O\u0303\u0301", "\u1e4e": "O\u0303\u0308", "\u022c": "O\u0303\u0304", "\u014c": "O\u0304", "\u1e52": "O\u0304\u0301", "\u1e50": "O\u0304\u0300", "\u014e": "O\u0306", "\u01d1": "O\u030c", "\xd4": "O\u0302", "\u1ed0": "O\u0302\u0301", "\u1ed2": "O\u0302\u0300", "\u1ed6": "O\u0302\u0303", "\u022e": "O\u0307", "\u0230": "O\u0307\u0304", "\u0150": "O\u030b", "\u1e54": "P\u0301", "\u1e56": "P\u0307", "\u0154": "R\u0301", "\u0158": "R\u030c", "\u1e58": "R\u0307", "\u0156": "R\u0327", "\u015a": "S\u0301", "\u1e64": "S\u0301\u0307", "\u0160": "S\u030c", "\u1e66": "S\u030c\u0307", "\u015c": "S\u0302", "\u1e60": "S\u0307", "\u015e": "S\u0327", "\u0164": "T\u030c", "\u1e6a": "T\u0307", "\u0162": "T\u0327", "\xda": "U\u0301", "\xd9": "U\u0300", "\xdc": "U\u0308", "\u01d7": "U\u0308\u0301", "\u01db": "U\u0308\u0300", "\u01d5": "U\u0308\u0304", "\u01d9": "U\u0308\u030c", "\u0168": "U\u0303", "\u1e78": "U\u0303\u0301", "\u016a": "U\u0304", "\u1e7a": "U\u0304\u0308", "\u016c": "U\u0306", "\u01d3": "U\u030c", "\xdb": "U\u0302", "\u016e": "U\u030a", "\u0170": "U\u030b", "\u1e7c": "V\u0303", "\u1e82": "W\u0301", "\u1e80": "W\u0300", "\u1e84": "W\u0308", "\u0174": "W\u0302", "\u1e86": "W\u0307", "\u1e8c": "X\u0308", "\u1e8a": "X\u0307", "\xdd": "Y\u0301", "\u1ef2": "Y\u0300", "\u0178": "Y\u0308", "\u1ef8": "Y\u0303", "\u0232": "Y\u0304", "\u0176": "Y\u0302", "\u1e8e": "Y\u0307", "\u0179": "Z\u0301", "\u017d": "Z\u030c", "\u1e90": "Z\u0302", "\u017b": "Z\u0307", "\u03ac": "\u03b1\u0301", "\u1f70": "\u03b1\u0300", "\u1fb1": "\u03b1\u0304", "\u1fb0": "\u03b1\u0306", "\u03ad": "\u03b5\u0301", "\u1f72": "\u03b5\u0300", "\u03ae": "\u03b7\u0301", "\u1f74": "\u03b7\u0300", "\u03af": "\u03b9\u0301", "\u1f76": "\u03b9\u0300", "\u03ca": "\u03b9\u0308", "\u0390": "\u03b9\u0308\u0301", "\u1fd2": "\u03b9\u0308\u0300", "\u1fd1": "\u03b9\u0304", "\u1fd0": "\u03b9\u0306", "\u03cc": "\u03bf\u0301", "\u1f78": "\u03bf\u0300", "\u03cd": "\u03c5\u0301", "\u1f7a": "\u03c5\u0300", "\u03cb": "\u03c5\u0308", "\u03b0": "\u03c5\u0308\u0301", "\u1fe2": "\u03c5\u0308\u0300", "\u1fe1": "\u03c5\u0304", "\u1fe0": "\u03c5\u0306", "\u03ce": "\u03c9\u0301", "\u1f7c": "\u03c9\u0300", "\u038e": "\u03a5\u0301", "\u1fea": "\u03a5\u0300", "\u03ab": "\u03a5\u0308", "\u1fe9": "\u03a5\u0304", "\u1fe8": "\u03a5\u0306", "\u038f": "\u03a9\u0301", "\u1ffa": "\u03a9\u0300" }; class Vn { constructor(e, t) { this.mode = void 0, this.gullet = void 0, this.settings = void 0, this.leftrightDepth = void 0, this.nextToken = void 0, this.mode = "math", this.gullet = new Hn(e, t, this.mode), this.settings = t, this.leftrightDepth = 0 } expect(e, t) { if (void 0 === t && (t = !0), this.fetch().text !== e) throw new n("Expected '" + e + "', got '" + this.fetch().text + "'", this.fetch()); t && this.consume() } consume() { this.nextToken = null } fetch() { return null == this.nextToken && (this.nextToken = this.gullet.expandNextToken()), this.nextToken } switchMode(e) { this.mode = e, this.gullet.switchMode(e) } parse() { this.settings.globalGroup || this.gullet.beginGroup(), this.settings.colorIsTextColor && this.gullet.macros.set("\\color", "\\textcolor"); try { const e = this.parseExpression(!1); return this.expect("EOF"), this.settings.globalGroup || this.gullet.endGroup(), e } finally { this.gullet.endGroups() } } subparse(e) { const t = this.nextToken; this.consume(), this.gullet.pushToken(new Nr("}")), this.gullet.pushTokens(e); const r = this.parseExpression(!1); return this.expect("}"), this.nextToken = t, r } parseExpression(e, t) { const r = []; for (; ;) { "math" === this.mode && this.consumeSpaces(); const n = this.fetch(); if (-1 !== Vn.endOfExpression.indexOf(n.text)) break; if (t && n.text === t) break; if (e && yn[n.text] && yn[n.text].infix) break; const o = this.parseAtom(t); if (!o) break; "internal" !== o.type && r.push(o) } return "text" === this.mode && this.formLigatures(r), this.handleInfixNodes(r) } handleInfixNodes(e) { let t, r = -1; for (let o = 0; o < e.length; o++)if ("infix" === e[o].type) { if (-1 !== r) throw new n("only one infix operator per group", e[o].token); r = o, t = e[o].replaceWith } if (-1 !== r && t) { let n, o; const s = e.slice(0, r), i = e.slice(r + 1); let a; return n = 1 === s.length && "ordgroup" === s[0].type ? s[0] : { type: "ordgroup", mode: this.mode, body: s }, o = 1 === i.length && "ordgroup" === i[0].type ? i[0] : { type: "ordgroup", mode: this.mode, body: i }, a = "\\\\abovefrac" === t ? this.callFunction(t, [n, e[r], o], []) : this.callFunction(t, [n, o], []), [a] } return e } handleSupSubscript(e) { const t = this.fetch(), r = t.text; this.consume(), this.consumeSpaces(); const o = this.parseGroup(e); if (!o) throw new n("Expected group after '" + r + "'", t); return o } formatUnsupportedCmd(e) { const t = []; for (let r = 0; r < e.length; r++)t.push({ type: "textord", mode: "text", text: e[r] }); const r = { type: "text", mode: this.mode, body: t }; return { type: "color", mode: this.mode, color: this.settings.errorColor, body: [r] } } parseAtom(e) { const t = this.parseGroup("atom", e); if ("text" === this.mode) return t; let r, o; for (; ;) { this.consumeSpaces(); const e = this.fetch(); if ("\\limits" === e.text || "\\nolimits" === e.text) { if (t && "op" === t.type) { const r = "\\limits" === e.text; t.limits = r, t.alwaysHandleSupSub = !0 } else { if (!t || "operatorname" !== t.type) throw new n("Limit controls must follow a math operator", e); t.alwaysHandleSupSub && (t.limits = "\\limits" === e.text) } this.consume() } else if ("^" === e.text) { if (r) throw new n("Double superscript", e); r = this.handleSupSubscript("superscript") } else if ("_" === e.text) { if (o) throw new n("Double subscript", e); o = this.handleSupSubscript("subscript") } else if ("'" === e.text) { if (r) throw new n("Double superscript", e); const t = { type: "textord", mode: this.mode, text: "\\prime" }, o = [t]; for (this.consume(); "'" === this.fetch().text;)o.push(t), this.consume(); "^" === this.fetch().text && o.push(this.handleSupSubscript("superscript")), r = { type: "ordgroup", mode: this.mode, body: o } } else { if (!En[e.text]) break; { const t = On.test(e.text), n = []; for (n.push(new Nr(En[e.text])), this.consume(); ;) { const e = this.fetch().text; if (!En[e]) break; if (On.test(e) !== t) break; n.unshift(new Nr(En[e])), this.consume() } const s = this.subparse(n); t ? o = { type: "ordgroup", mode: "math", body: s } : r = { type: "ordgroup", mode: "math", body: s } } } } return r || o ? { type: "supsub", mode: this.mode, base: t, sup: r, sub: o } : t } parseFunction(e, t) { const r = this.fetch(), o = r.text, s = yn[o]; if (!s) return null; if (this.consume(), t && "atom" !== t && !s.allowedInArgument) throw new n("Got function '" + o + "' with no arguments" + (t ? " as " + t : ""), r); if ("text" === this.mode && !s.allowedInText) throw new n("Can't use function '" + o + "' in text mode", r); if ("math" === this.mode && !1 === s.allowedInMath) throw new n("Can't use function '" + o + "' in math mode", r); const { args: i, optArgs: a } = this.parseArguments(o, s); return this.callFunction(o, i, a, r, e) } callFunction(e, t, r, o, s) { const i = { funcName: e, parser: this, token: o, breakOnTokenText: s }, a = yn[e]; if (a && a.handler) return a.handler(i, t, r); throw new n("No function handler for " + e) } parseArguments(e, t) { const r = t.numArgs + t.numOptionalArgs; if (0 === r) return { args: [], optArgs: [] }; const o = [], s = []; for (let i = 0; i < r; i++) { let r = t.argTypes && t.argTypes[i]; const a = i < t.numOptionalArgs; (t.primitive && null == r || "sqrt" === t.type && 1 === i && null == s[0]) && (r = "primitive"); const l = this.parseGroupOfType("argument to '" + e + "'", r, a); if (a) s.push(l); else { if (null == l) throw new n("Null argument, please report this as a bug"); o.push(l) } } return { args: o, optArgs: s } } parseGroupOfType(e, t, r) { switch (t) { case "color": return this.parseColorGroup(r); case "size": return this.parseSizeGroup(r); case "url": return this.parseUrlGroup(r); case "math": case "text": return this.parseArgumentGroup(r, t); case "hbox": { const e = this.parseArgumentGroup(r, "text"); return null != e ? { type: "styling", mode: e.mode, body: [e], style: "text" } : null } case "raw": { const e = this.parseStringGroup("raw", r); return null != e ? { type: "raw", mode: "text", string: e.text } : null } case "primitive": { if (r) throw new n("A primitive argument cannot be optional"); const t = this.parseGroup(e); if (null == t) throw new n("Expected group as " + e, this.fetch()); return t } case "original": case null: case void 0: return this.parseArgumentGroup(r); default: throw new n("Unknown group type as " + e, this.fetch()) } } consumeSpaces() { for (; " " === this.fetch().text;)this.consume() } parseStringGroup(e, t) { const r = this.gullet.scanArgument(t); if (null == r) return null; let n, o = ""; for (; "EOF" !== (n = this.fetch()).text;)o += n.text, this.consume(); return this.consume(), r.text = o, r } parseRegexGroup(e, t) { const r = this.fetch(); let o, s = r, i = ""; for (; "EOF" !== (o = this.fetch()).text && e.test(i + o.text);)s = o, i += s.text, this.consume(); if ("" === i) throw new n("Invalid " + t + ": '" + r.text + "'", r); return r.range(s, i) } parseColorGroup(e) { const t = this.parseStringGroup("color", e); if (null == t) return null; const r = /^(#[a-f0-9]{3}|#?[a-f0-9]{6}|[a-z]+)$/i.exec(t.text); if (!r) throw new n("Invalid color: '" + t.text + "'", t); let o = r[0]; return /^[0-9a-f]{6}$/i.test(o) && (o = "#" + o), { type: "color-token", mode: this.mode, color: o } } parseSizeGroup(e) { let t, r = !1; if (this.gullet.consumeSpaces(), t = e || "{" === this.gullet.future().text ? this.parseStringGroup("size", e) : this.parseRegexGroup(/^[-+]? *(?:$|\d+|\d+\.\d*|\.\d*) *[a-z]{0,2} *$/, "size"), !t) return null; e || 0 !== t.text.length || (t.text = "0pt", r = !0); const o = /([-+]?) *(\d+(?:\.\d*)?|\.\d+) *([a-z]{2})/.exec(t.text); if (!o) throw new n("Invalid size: '" + t.text + "'", t); const s = { number: +(o[1] + o[2]), unit: o[3] }; if (!V(s)) throw new n("Invalid unit: '" + s.unit + "'", t); return { type: "size", mode: this.mode, value: s, isBlank: r } } parseUrlGroup(e) { this.gullet.lexer.setCatcode("%", 13), this.gullet.lexer.setCatcode("~", 12); const t = this.parseStringGroup("url", e); if (this.gullet.lexer.setCatcode("%", 14), this.gullet.lexer.setCatcode("~", 13), null == t) return null; const r = t.text.replace(/\\([#$%&~_^{}])/g, "$1"); return { type: "url", mode: this.mode, url: r } } parseArgumentGroup(e, t) { const r = this.gullet.scanArgument(e); if (null == r) return null; const n = this.mode; t && this.switchMode(t), this.gullet.beginGroup(); const o = this.parseExpression(!1, "EOF"); this.expect("EOF"), this.gullet.endGroup(); const s = { type: "ordgroup", mode: this.mode, loc: r.loc, body: o }; return t && this.switchMode(n), s } parseGroup(e, t) { const r = this.fetch(), o = r.text; let s; if ("{" === o || "\\begingroup" === o) { this.consume(); const e = "{" === o ? "}" : "\\endgroup"; this.gullet.beginGroup(); const t = this.parseExpression(!1, e), n = this.fetch(); this.expect(e), this.gullet.endGroup(), s = { type: "ordgroup", mode: this.mode, loc: Cr.range(r, n), body: t, semisimple: "\\begingroup" === o || void 0 } } else if (s = this.parseFunction(t, e) || this.parseSymbol(), null == s && "\\" === o[0] && !Rn.hasOwnProperty(o)) { if (this.settings.throwOnError) throw new n("Undefined control sequence: " + o, r); s = this.formatUnsupportedCmd(o), this.consume() } return s } formLigatures(e) { let t = e.length - 1; for (let r = 0; r < t; ++r) { const n = e[r], o = n.text; "-" === o && "-" === e[r + 1].text && (r + 1 < t && "-" === e[r + 2].text ? (e.splice(r, 3, { type: "textord", mode: "text", loc: Cr.range(n, e[r + 2]), text: "---" }), t -= 2) : (e.splice(r, 2, { type: "textord", mode: "text", loc: Cr.range(n, e[r + 1]), text: "--" }), t -= 1)), "'" !== o && "`" !== o || e[r + 1].text !== o || (e.splice(r, 2, { type: "textord", mode: "text", loc: Cr.range(n, e[r + 1]), text: o + o }), t -= 1) } } parseSymbol() { const e = this.fetch(); let t = e.text; if (/^\\verb[^a-zA-Z]/.test(t)) { this.consume(); let e = t.slice(5); const r = "*" === e.charAt(0); if (r && (e = e.slice(1)), e.length < 2 || e.charAt(0) !== e.slice(-1)) throw new n("\\verb assertion failed --\n                    please report what input caused this bug"); return e = e.slice(1, -1), { type: "verb", mode: "text", body: e, star: r } } Dn.hasOwnProperty(t[0]) && !oe[this.mode][t[0]] && (this.settings.strict && "math" === this.mode && this.settings.reportNonstrict("unicodeTextInMathMode", 'Accented Unicode text character "' + t[0] + '" used in math mode', e), t = Dn[t[0]] + t.slice(1)); const r = kn.exec(t); let o; if (r && (t = t.substring(0, r.index), "i" === t ? t = "\u0131" : "j" === t && (t = "\u0237")), oe[this.mode][t]) { this.settings.strict && "math" === this.mode && Ae.indexOf(t) >= 0 && this.settings.reportNonstrict("unicodeTextInMathMode", 'Latin-1/Unicode text character "' + t[0] + '" used in math mode', e); const r = oe[this.mode][t].group, n = Cr.range(e); let s; if (te.hasOwnProperty(r)) { const e = r; s = { type: "atom", mode: this.mode, family: e, loc: n, text: t } } else s = { type: r, mode: this.mode, loc: n, text: t }; o = s } else { if (!(t.charCodeAt(0) >= 128)) return null; this.settings.strict && (S(t.charCodeAt(0)) ? "math" === this.mode && this.settings.reportNonstrict("unicodeTextInMathMode", 'Unicode text character "' + t[0] + '" used in math mode', e) : this.settings.reportNonstrict("unknownSymbol", 'Unrecognized Unicode character "' + t[0] + '" (' + t.charCodeAt(0) + ")", e)), o = { type: "textord", mode: "text", loc: Cr.range(e), text: t } } if (this.consume(), r) for (let t = 0; t < r[0].length; t++) { const s = r[0][t]; if (!Ln[s]) throw new n("Unknown accent ' " + s + "'", e); const i = Ln[s][this.mode] || Ln[s].text; if (!i) throw new n("Accent " + s + " unsupported in " + this.mode + " mode", e); o = { type: "accent", mode: this.mode, loc: Cr.range(e), label: i, isStretchy: !1, isShifty: !0, base: o } } return o } } Vn.endOfExpression = ["}", "\\endgroup", "\\end", "\\right", "&"]; var Pn = function (e, t) { if (!("string" == typeof e || e instanceof String)) throw new TypeError("KaTeX can only parse string typed expression"); const r = new Vn(e, t); delete r.gullet.macros.current["\\df@tag"]; let o = r.parse(); if (delete r.gullet.macros.current["\\current@color"], delete r.gullet.macros.current["\\color"], r.gullet.macros.get("\\df@tag")) { if (!t.displayMode) throw new n("\\tag works only in display equations"); o = [{ type: "tag", mode: "text", body: o, tag: r.subparse([new Nr("\\df@tag")]) }] } return o }; let Fn = function (e, t, r) { t.textContent = ""; const n = Un(e, r).toNode(); t.appendChild(n) }; "undefined" != typeof document && "CSS1Compat" !== document.compatMode && ("undefined" != typeof console && console.warn("Warning: KaTeX doesn't work in quirks mode. Make sure your website has a suitable doctype."), Fn = function () { throw new n("KaTeX doesn't work in quirks mode.") }); const Gn = function (e, t, r) { if (r.throwOnError || !(e instanceof n)) throw e; const o = Ve.makeSpan(["katex-error"], [new Z(t)]); return o.setAttribute("title", e.toString()), o.setAttribute("style", "color:" + r.errorColor), o }, Un = function (e, t) { const r = new m(t); try { const t = Pn(e, r); return zt(t, e, r) } catch (t) { return Gn(t, e, r) } }; var Yn = { version: "0.16.11", render: Fn, renderToString: function (e, t) { return Un(e, t).toMarkup() }, ParseError: n, SETTINGS_SCHEMA: h, __parse: function (e, t) { const r = new m(t); return Pn(e, r) }, __renderToDomTree: Un, __renderToHTMLTree: function (e, t) { const r = new m(t); try { return function (e, t, r) { const n = mt(e, St(r)), o = Ve.makeSpan(["katex"], [n]); return Mt(o, r) }(Pn(e, r), 0, r) } catch (t) { return Gn(t, e, r) } }, __setFontMetrics: function (e, t) { T[e] = t }, __defineSymbol: se, __defineFunction: je, __defineMacro: Br, __domTree: { Span: W, Anchor: _, SymbolNode: Z, SvgNode: K, PathNode: J, LineNode: Q } }; return t = t.default }() }));
diff --git a/docs/version2.x/legacy/android_demo.en.md b/docs/version2.x/legacy/android_demo.en.md
new file mode 100644
index 0000000..f6f6f3e
--- /dev/null
+++ b/docs/version2.x/legacy/android_demo.en.md
@@ -0,0 +1,114 @@
+---
+typora-copy-images-to: images
+comments: true
+hide:
+  - toc
+---
+
+# Android Demo
+
+## 1. Introduction
+
+This document describes the [Android Demo](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/deploy/android_demo) for PaddleOCR. The demo currently supports text detection, text direction classification, and text recognition models. It is developed using [PaddleLite v2.10](https://github.com/PaddlePaddle/Paddle-Lite/tree/release/v2.10).
+
+## 2. Recent Updates
+
+- **February 27, 2022**
+  - Updated the prediction library to PaddleLite v2.10.
+  - Added support for six operation modes:
+    - Detection + Classification + Recognition
+    - Detection + Recognition
+    - Classification + Recognition
+    - Detection
+    - Recognition
+    - Classification
+
+## 3. Quick Start
+
+### 3.1 Environment Setup
+
+1. **Install Android Studio:**  
+   Download and install Android Studio on your development machine. For detailed instructions, please refer to the [Android Studio official website](https://developer.android.com/studio).
+
+2. **Prepare Your Android Device:**  
+   Ensure that you have an Android phone and enable USB debugging. To do so, navigate to:  
+   **Settings → Developer Options → Enable Developer Options and USB Debugging.**
+
+   **Note:** If your Android Studio installation does not have the NDK configured, please follow the [Install and Configure NDK and CMake](https://developer.android.com/studio/projects/install-ndk) guide. You may choose the latest NDK version or use the same version as the PaddleLite prediction library.
+
+### 3.2 Importing the Project
+
+1. In Android Studio, select **File → New → Import Project…** and follow the wizard to import the project.
+2. Upon successful import, your project interface should resemble the following:  
+   [](./images/import_demo.jpg)
+
+### 3.3 Running the Demo
+
+1. Connect your Android device to your computer.
+2. Click the **Run** button in the Android Studio toolbar.  
+   - Your device may prompt you with "Allow installation of software from USB." Click **Allow**.
+3. Once the installation is complete, the app icon will appear on the last page of your home screen.
+4. Tap the app icon to launch the application. The home page will display as shown below:  
+   ![App Home Screen](./images/68747470733a.png)
+
+On the home screen, you will find the following elements:
+
+- **Run Model:** Executes the model(s) using the selected operation mode.
+- **Take Photo for Recognition:** Activates the device camera to capture an image. After taking a photo, tap **Run Model** to perform recognition.
+- **Select Image:** Opens the photo gallery to select an image. After selection, tap **Run Model** to perform recognition.
+- **Clear Drawing:** Clears any drawn text boxes from the currently displayed image.
+- **Dropdown List:** Allows you to choose one of the six supported operation modes. The default mode is **Detection + Classification + Recognition**.
+- **Menu Button:** Opens the settings interface for configurations related to the models and built-in images.
+
+When you tap **Run Model**, the demo executes the corresponding model(s) in your selected mode. For example, in **Detection + Classification + Recognition** mode, the output appears as follows:
+  
+<img src="./images/run_det_cls_rec.jpg" width="400">
+
+The status display area shows the current model status (e.g., `run model succeeded`), indicating that the model ran successfully. The recognition results are formatted as follows:
+
+```text
+Serial Number: Det: (x1,y1)(x2,y2)(x3,y3)(x4,y4) Rec: Recognized Text, Confidence Score Cls: Classification Label, Classification Score
+```
+
+### 3.4 Operation Modes
+
+The PaddleOCR demo supports six operation modes, shown in the illustration below:
+
+![Operation Modes Overview](./images/68747470733adfd.jpeg)
+
+The expected output for each mode is as follows:
+
+| **Detection + Classification + Recognition**                                   | **Detection + Recognition**                            | **Classification + Recognition**                        |
+|----------------------------------------------------------------------------------|--------------------------------------------------------|--------------------------------------------------------|
+| <img src="./images/run_det_cls_rec-20240806100407184.jpg" width="400">            | <img src="./images/run_det_rec.jpg" width="400">        | <img src="./images/run_cls_rec.jpg" width="400">        |
+
+| **Detection**                                                                  | **Recognition**                                        | **Classification**                                     |
+|--------------------------------------------------------------------------------|--------------------------------------------------------|--------------------------------------------------------|
+| <img src="./images/run_det.jpg" width="400">                                    | <img src="./images/run_rec.jpg" width="400">            | <img src="./images/run_cls.jpg" width="400">            |
+
+### 3.5 Settings
+
+The settings interface allows you to customize the following parameters:
+
+1. **General Settings**
+   - **Enable Custom Settings:** Check this box to allow modifications.
+   - **Model Path:** The file path to the model (use the default value unless necessary).
+   - **Label Path:** The dictionary file for the text recognition model.
+   - **Image Path:** The filename of the built-in image used for recognition.
+
+2. **Model Runtime Settings:** (Changes here trigger an automatic model reload when returning to the main interface.)
+   - **CPU Thread Number:** The number of CPU cores allocated for model inference.
+   - **CPU Power Mode:** Defines the model’s execution strategy, corresponding to different core configurations.
+
+3. **Input Settings**
+   - **det long size:** The target value for the longer side of the image during DB model preprocessing. If the image's long side exceeds this value, it will be resized proportionally; if it is shorter, no resizing is applied.
+
+4. **Output Settings**
+   - **Score Threshold:** The minimum score threshold for detection boxes in DB model post-processing. Boxes scoring below this threshold are filtered out and not displayed.
+
+## 4. Further Support
+
+For additional support and updates:
+
+1. Refer to the [Paddle-Lite Demo for OCR](https://github.com/PaddlePaddle/Paddle-Lite-Demo/tree/develop/ocr/android/app/cxx/ppocr_demo) for real-time recognition and prediction library updates.
+2. Visit the [PaddleLite repository](https://github.com/PaddlePaddle/Paddle-Lite) for more development support and community discussions.
diff --git a/docs/version2.x/legacy/android_demo.md b/docs/version2.x/legacy/android_demo.md
new file mode 100644
index 0000000..47477d3
--- /dev/null
+++ b/docs/version2.x/legacy/android_demo.md
@@ -0,0 +1,114 @@
+---
+typora-copy-images-to: images
+comments: true
+hide:
+  - toc
+---
+
+# Android Demo
+
+## 1. 简介
+
+此为PaddleOCR的[Android Demo](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/deploy/android_demo)，目前支持文本检测，文本方向分类器和文本识别模型的使用。使用 [PaddleLite v2.10](https://github.com/PaddlePaddle/Paddle-Lite/tree/release/v2.10) 进行开发。
+
+## 2. 近期更新
+
+- 2022.02.27
+    - 预测库更新到PaddleLite v2.10
+    - 支持6种运行模式：
+        - 检测+分类+识别
+        - 检测+识别
+        - 分类+识别
+        - 检测
+        - 识别
+        - 分类
+
+## 3. 快速使用
+
+### 3.1 环境准备
+
+1. 在本地环境安装好 Android Studio 工具，详细安装方法请见[Android Studio 官网](https://developer.android.com/studio)。
+2. 准备一部 Android 手机，并开启 USB 调试模式。开启方法: `手机设置 -> 查找开发者选项 -> 打开开发者选项和 USB 调试模式`
+
+**注意**：如果您的 Android Studio 尚未配置 NDK ，请根据 Android Studio 用户指南中的[安装及配置 NDK 和 CMake](https://developer.android.com/studio/projects/install-ndk)内容，预先配置好 NDK 。您可以选择最新的 NDK 版本，或者使用 Paddle Lite 预测库版本一样的 NDK
+
+### 3.2 导入项目
+
+点击 File->New->Import Project...， 然后跟着Android Studio的引导导入
+导入完成后呈现如下界面
+![](./images/import_demo.jpg)
+
+### 3.3 运行demo
+
+将手机连接上电脑后，点击Android Studio工具栏中的运行按钮即可运行demo。在此过程中，手机会弹出"允许从 USB 安装软件权限"的弹窗，点击允许即可。
+
+软件安转到手机上后会在手机主屏最后一页看到如下app
+
+![img](./images/68747470733a.jpeg)
+
+点击app图标即可启动app，启动后app主页如下
+
+![img](./images/68747470733a.png)
+
+app主页中有四个按钮，一个下拉列表和一个菜单按钮，他们的功能分别为
+
+- 运行模型：按照已选择的模式，运行对应的模型组合
+- 拍照识别：唤起手机相机拍照并获取拍照的图像，拍照完成后需要点击运行模型进行识别
+- 选取图片：唤起手机相册拍照选择图像，选择完成后需要点击运行模型进行识别
+- 清空绘图：清空当前显示图像上绘制的文本框，以便进行下一次识别(每次识别使用的图像都是当前显示的图像)
+- 下拉列表：进行运行模式的选择，目前包含6种运行模式，默认模式为**检测+分类+识别**详细说明见下一节。
+- 菜单按钮：点击后会进入菜单界面，进行模型和内置图像有关设置
+
+点击运行模型后，会按照所选择的模式运行对应的模型，**检测+分类+识别**模式下运行的模型结果如下所示：
+
+<img src="./images/run_det_cls_rec.jpg" width="400">
+
+模型运行完成后，模型和运行状态显示区`STATUS`字段显示了当前模型的运行状态，这里显示为`run model succeeded`表明模型运行成功。
+
+模型的运行结果显示在运行结果显示区，显示格式为
+
+```text
+序号：Det：(x1,y1)(x2,y2)(x3,y3)(x4,y4) Rec: 识别文本,识别置信度 Cls：分类类别,分类分时
+```
+
+### 3.4 运行模式
+
+PaddleOCR demo共提供了6种运行模式，如下图
+
+![img](./images/68747470733adfd.jpeg)
+
+每种模式的运行结果如下表所示
+
+| 检测+分类+识别                                               | 检测+识别                                        | 分类+识别                                        |
+| ------------------------------------------------------------ | ------------------------------------------------ | ------------------------------------------------ |
+| <img src="./images/run_det_cls_rec-20240806100407184.jpg" width="400"> | <img src="./images/run_det_rec.jpg" width="400"> | <img src="./images/run_cls_rec.jpg" width="400"> |
+
+| 检测                                         | 识别                                         | 分类                                         |
+| -------------------------------------------- | -------------------------------------------- | -------------------------------------------- |
+| <img src="./images/run_det.jpg" width="400"> | <img src="./images/run_rec.jpg" width="400"> | <img src="./images/run_cls.jpg" width="400"> |
+
+### 3.5 设置
+
+设置界面如下
+
+![img](./images/68747470733a2f2f7061.jpeg)
+
+在设置界面可以进行如下几项设定：
+
+1. 普通设置
+   - Enable custom settings: 选中状态下才能更改设置
+   - Model Path: 所运行的模型地址，使用默认值就好
+   - Label Path: 识别模型的字典
+   - Image Path: 进行识别的内置图像名
+2. 模型运行态设置，此项设置更改后返回主界面时，会自动重新加载模型
+   - CPU Thread Num: 模型运行使用的CPU核心数量
+   - CPU Power Mode: 模型运行模式，大小核设定
+3. 输入设置
+   - det long size: DB模型预处理时图像的长边长度，超过此长度resize到该值，短边进行等比例缩放，小于此长度不进行处理。
+4. 输出设置
+   - Score Threshold: DB模型后处理box的阈值，低于此阈值的box进行过滤，不显示。
+
+## 4 更多支持
+
+1. 实时识别，更新预测库可参考 <https://github.com/PaddlePaddle/Paddle-Lite-Demo/tree/develop/ocr/android/app/cxx/ppocr_demo>
+2. 更多Paddle-Lite相关问题可前往[Paddle-Lite](https://github.com/PaddlePaddle/Paddle-Lite) ，获得更多开发支持
diff --git a/docs/version2.x/legacy/benchmark.en.md b/docs/version2.x/legacy/benchmark.en.md
new file mode 100755
index 0000000..59b29d4
--- /dev/null
+++ b/docs/version2.x/legacy/benchmark.en.md
@@ -0,0 +1,43 @@
+---
+comments: true
+hide:
+  - toc
+---
+
+# Benchmark
+
+This document gives the performance of the series models for Chinese and English recognition.
+
+## Test Data
+
+We collected 300 images for different real application scenarios to evaluate the overall OCR system, including contract samples, license plates, nameplates, train tickets, test sheets, forms, certificates, street view images, business cards, digital meter, etc. The following figure shows some images of the test set.
+
+![img](./images/doc.jpg)
+
+## Measurement
+
+Explanation:
+
+- The long size of the input for the text detector is 960.
+
+- The evaluation time-consuming stage is the complete stage from image input to result output, including image pre-processing and post-processing.
+
+- `Intel Xeon 6148` is the server-side CPU model. Intel MKL-DNN is used in the test to accelerate the CPU prediction speed.
+
+- `Snapdragon 855` is a mobile processing platform model.
+
+Compares the model size and F-score:
+
+| Model Name                    | Model Size <br> of the <br> Whole System\(M\) | Model Size <br>of the Text <br> Detector\(M\) | Model Size <br> of the Direction <br> Classifier\(M\) | Model Size<br>of the Text <br> Recognizer \(M\) | F\-score |
+|:-:|:-:|:-:|:-:|:-:|:-:|
+| PP-OCRv2                 | 11\.6        | 3\.0        | 0\.9           | 8\.6        | 0\.5224      |
+| PP-OCR mobile            |   8\.1       | 2\.6        | 0\.9           | 4\.6        | 0\.503       |
+| PP-OCR server            | 155\.1       | 47\.2       | 0\.9           | 107         | 0\.570       |
+
+Compares the time-consuming on CPU and T4 GPU (ms):
+
+| Model Name    | CPU  | T4 GPU |
+|:-:|:-:|:-:|
+| PP-OCRv2      | 330  | 111 |
+| PP-OCR mobile | 356  | 116|
+| PP-OCR server | 1056 | 200 |
diff --git a/docs/version2.x/legacy/benchmark.md b/docs/version2.x/legacy/benchmark.md
new file mode 100644
index 0000000..49af1c2
--- /dev/null
+++ b/docs/version2.x/legacy/benchmark.md
@@ -0,0 +1,41 @@
+---
+typora-copy-images-to: images
+comments: true
+hide:
+  - toc
+---
+
+# Benchmark
+
+本文给出了中英文OCR系列模型精度指标和在各平台预测耗时的benchmark。
+
+## 测试数据
+
+针对OCR实际应用场景，包括合同，车牌，铭牌，火车票，化验单，表格，证书，街景文字，名片，数码显示屏等，收集的300张图像，每张图平均有17个文本框，下图给出了一些图像示例。
+
+![img](./images/doc.jpg)
+
+## 评估指标
+
+说明：
+
+- 检测输入图像的长边尺寸是960。
+- 评估耗时阶段为图像预测耗时，不包括图像的预处理和后处理。
+- `Intel至强6148`为服务器端CPU型号，测试中使用Intel MKL-DNN 加速。
+- `骁龙855`为移动端处理平台型号。
+
+预测模型大小和整体识别精度对比
+
+|   模型名称    | 整体模型<br>大小\(M\) | 检测模型<br>大小\(M\) | 方向分类器<br>模型大小\(M\) | 识别模型<br>大小\(M\) | 整体识别<br>F\-score |
+| :-----------: | :-------------------: | :-------------------: | :-------------------------: | :-------------------: | :------------------: |
+|   PP-OCRv2    |         11\.6         |         3\.0          |            0\.9             |         8\.6          |       0\.5224        |
+| PP-OCR mobile |         8\.1          |         2\.6          |            0\.9             |         4\.6          |        0\.503        |
+| PP-OCR server |        155\.1         |         47\.2         |            0\.9             |          107          |        0\.570        |
+
+预测模型在CPU和GPU上的速度对比，单位ms
+
+|   模型名称    |  CPU  | T4 GPU |
+| :-----------: | :---: | :----: |
+|   PP-OCRv2    |  330  |  111   |
+| PP-OCR mobile |  356  |  11 6  |
+| PP-OCR server | 1056  |  200   |
diff --git a/docs/version2.x/legacy/cpp_infer.en.md b/docs/version2.x/legacy/cpp_infer.en.md
new file mode 100644
index 0000000..f27151b
--- /dev/null
+++ b/docs/version2.x/legacy/cpp_infer.en.md
@@ -0,0 +1,435 @@
+---
+comments: true
+hide:
+  - toc
+---
+
+# Server-side C++ Inference
+
+This chapter introduces the C++ deployment steps of the PaddleOCR model. C++ is better than Python in terms of performance. Therefore, in CPU and GPU deployment scenarios, C++ deployment is mostly used.
+This section will introduce how to configure the C++ environment and deploy PaddleOCR in Linux (CPU\GPU) environment. For Windows deployment please refer to [Windows](./windows_vs2019_build.en.md) compilation guidelines.
+
+## 1. Prepare the Environment
+
+### 1.1 Environment
+
+- Linux, docker is recommended.
+- Windows.
+
+### 1.2 Compile OpenCV
+
+- First of all, you need to download the source code compiled package in the Linux environment from the OpenCV official website. Taking OpenCV 3.4.7 as an example, the download command is as follows.
+
+```bash linenums="1"
+cd deploy/cpp_infer
+wget https://paddleocr.bj.bcebos.com/libs/opencv/opencv-3.4.7.tar.gz
+tar -xf opencv-3.4.7.tar.gz
+```
+
+Finally, you will see the folder of `opencv-3.4.7/` in the current directory.
+
+- Compile OpenCV, the OpenCV source path (`root_path`) and installation path (`install_path`) should be set by yourself. Enter the OpenCV source code path and compile it in the following way.
+
+```bash linenums="1"
+root_path=your_opencv_root_path
+install_path=${root_path}/opencv3
+
+rm -rf build
+mkdir build
+cd build
+
+cmake .. \
+    -DCMAKE_INSTALL_PREFIX=${install_path} \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DBUILD_SHARED_LIBS=OFF \
+    -DWITH_IPP=OFF \
+    -DBUILD_IPP_IW=OFF \
+    -DWITH_LAPACK=OFF \
+    -DWITH_EIGEN=OFF \
+    -DCMAKE_INSTALL_LIBDIR=lib64 \
+    -DWITH_ZLIB=ON \
+    -DBUILD_ZLIB=ON \
+    -DWITH_JPEG=ON \
+    -DBUILD_JPEG=ON \
+    -DWITH_PNG=ON \
+    -DBUILD_PNG=ON \
+    -DWITH_TIFF=ON \
+    -DBUILD_TIFF=ON
+
+make -j
+make install
+```
+
+In the above commands, `root_path` is the downloaded OpenCV source code path, and `install_path` is the installation path of OpenCV. After `make install` is completed, the OpenCV header file and library file will be generated in this folder for later OCR source code compilation.
+
+The final file structure under the OpenCV installation path is as follows.
+
+```text linenums="1"
+opencv3/
+|-- bin
+|-- include
+|-- lib
+|-- lib64
+|-- share
+```
+
+### 1.3 Compile or Download or the Paddle Inference Library
+
+- There are 2 ways to obtain the Paddle inference library, described in detail below.
+
+#### 1.3.1 Direct download and installation
+
+[Paddle inference library official website](https://www.paddlepaddle.org.cn/inference/master/guides/install/download_lib.html#linux). You can review and select the appropriate version of the inference library on the official website.
+
+- After downloading, use the following command to extract files.
+
+```bash linenums="1"
+tar -xf paddle_inference.tgz
+```
+
+Finally you will see the folder of `paddle_inference/` in the current path.
+
+#### 1.3.2 Compile the inference source code
+
+- If you want to get the latest Paddle inference library features, you can download the latest code from Paddle GitHub repository and compile the inference library from the source code. It is recommended to download the inference library with paddle version greater than or equal to 2.0.1.
+
+- You can refer to [Paddle inference library](https://www.paddlepaddle.org.cn/documentation/docs/en/advanced_guide/inference_deployment/inference/build_and_install_lib_en.html) to get the Paddle source code from GitHub, and then compile To generate the latest inference library. The method of using git to access the code is as follows.
+
+```bash linenums="1"
+git clone https://github.com/PaddlePaddle/Paddle.git
+git checkout develop
+```
+
+- Enter the Paddle directory and run the following commands to compile the paddle inference library.
+
+```bash linenums="1"
+rm -rf build
+mkdir build
+cd build
+
+cmake  .. \
+    -DWITH_CONTRIB=OFF \
+    -DWITH_MKL=ON \
+    -DWITH_MKLDNN=ON  \
+    -DWITH_TESTING=OFF \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DWITH_INFERENCE_API_TEST=OFF \
+    -DON_INFER=ON \
+    -DWITH_PYTHON=ON
+make -j
+make inference_lib_dist
+```
+
+For more compilation parameter options, please refer to the [document](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0/guides/05_inference_deployment/inference/build_and_install_lib_cn.html#congyuanmabianyi).
+
+- After the compilation process, you can see the following files in the folder of `build/paddle_inference_install_dir/`.
+
+```text linenums="1"
+build/paddle_inference_install_dir/
+|-- CMakeCache.txt
+|-- paddle
+|-- third_party
+|-- version.txt
+```
+
+`paddle` is the Paddle library required for C++ prediction later, and `version.txt` contains the version information of the current inference library.
+
+## 2. Compile and Run the Demo
+
+### 2.1 Export the inference model
+
+- You can refer to [Model inference](./python_infer.en.md) and export the inference model. After the model is exported, assuming it is placed in the `inference` directory, the directory structure is as follows.
+
+```text linenums="1"
+inference/
+|-- det_db
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+|-- rec_rcnn
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+|-- cls
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+|-- table
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+|-- layout
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+```
+
+### 2.2 Compile PaddleOCR C++ inference demo
+
+- The compilation commands are as follows. The addresses of Paddle C++ inference library, opencv and other Dependencies need to be replaced with the actual addresses on your own machines.
+
+```bash linenums="1"
+sh tools/build.sh
+```
+
+Specifically, you should modify the paths in `tools/build.sh`. The related content is as follows.
+
+```bash linenums="1"
+OPENCV_DIR=your_opencv_dir
+LIB_DIR=your_paddle_inference_dir
+CUDA_LIB_DIR=your_cuda_lib_dir
+CUDNN_LIB_DIR=your_cudnn_lib_dir
+```
+
+`OPENCV_DIR` is the OpenCV installation path; `LIB_DIR` is the download (`paddle_inference` folder)
+or the generated Paddle inference library path (`build/paddle_inference_install_dir` folder);
+`CUDA_LIB_DIR` is the CUDA library file path, in docker; it is `/usr/local/cuda/lib64`; `CUDNN_LIB_DIR` is the cuDNN library file path, in docker it is `/usr/lib/x86_64-linux-gnu/`.
+
+- After the compilation is completed, an executable file named `ppocr` will be generated in the `build` folder.
+
+### 2.3 Run the demo
+
+Execute the built executable file:
+
+```bash linenums="1"
+./build/ppocr [--param1] [--param2] [...]
+```
+
+**Note**:ppocr uses the `PP-OCRv3` model by default, and the input shape used by the recognition model is `3, 48, 320`, if you want to use the old version model, you should add the parameter `--rec_img_h=32`.
+
+Specifically,
+
+#### 1. det+cls+rec
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --rec_model_dir=inference/rec_rcnn \
+    --cls_model_dir=inference/cls \
+    --image_dir=../../doc/imgs/12.jpg \
+    --use_angle_cls=true \
+    --det=true \
+    --rec=true \
+    --cls=true \
+```
+
+#### 2. det+rec
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --rec_model_dir=inference/rec_rcnn \
+    --image_dir=../../doc/imgs/12.jpg \
+    --use_angle_cls=false \
+    --det=true \
+    --rec=true \
+    --cls=false \
+```
+
+#### 3. det
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --image_dir=../../doc/imgs/12.jpg \
+    --det=true \
+    --rec=false
+```
+
+#### 4. cls+rec
+
+```bash linenums="1"
+./build/ppocr --rec_model_dir=inference/rec_rcnn \
+    --cls_model_dir=inference/cls \
+    --image_dir=../../doc/imgs_words/ch/word_1.jpg \
+    --use_angle_cls=true \
+    --det=false \
+    --rec=true \
+    --cls=true \
+```
+
+#### 5. rec
+
+```bash linenums="1"
+./build/ppocr --rec_model_dir=inference/rec_rcnn \
+    --image_dir=../../doc/imgs_words/ch/word_1.jpg \
+    --use_angle_cls=false \
+    --det=false \
+    --rec=true \
+    --cls=false \
+```
+
+#### 6. cls
+
+```bash linenums="1"
+./build/ppocr --cls_model_dir=inference/cls \
+    --cls_model_dir=inference/cls \
+    --image_dir=../../doc/imgs_words/ch/word_1.jpg \
+    --use_angle_cls=true \
+    --det=false \
+    --rec=false \
+    --cls=true \
+```
+
+#### 7. layout+table
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --rec_model_dir=inference/rec_rcnn \
+    --table_model_dir=inference/table \
+    --image_dir=../../ppstructure/docs/table/table.jpg \
+    --layout_model_dir=inference/layout \
+    --type=structure \
+    --table=true \
+    --layout=true
+```
+
+#### 8. layout
+
+```bash linenums="1"
+./build/ppocr --layout_model_dir=inference/layout \
+    --image_dir=../../ppstructure/docs/table/1.png \
+    --type=structure \
+    --table=false \
+    --layout=true \
+    --det=false \
+    --rec=false
+```
+
+#### 9. table
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --rec_model_dir=inference/rec_rcnn \
+    --table_model_dir=inference/table \
+    --image_dir=../../ppstructure/docs/table/table.jpg \
+    --type=structure \
+    --table=true
+```
+
+More parameters are as follows,
+
+Common parameters
+
+|parameter|data type|default|meaning|
+| --- | --- | --- | --- |
+|use_gpu|bool|false|Whether to use GPU|
+|gpu_id|int|0|GPU id when use_gpu is true|
+|gpu_mem|int|4000|GPU memory requested|
+|cpu_math_library_num_threads|int|10|Number of threads when using CPU inference. When machine cores is enough, the large the value, the faster the inference speed|
+|enable_mkldnn|bool|true|Whether to use mkdlnn library|
+|output|str|./output|Path where visualization results are saved|
+
+forward
+
+|parameter|data type|default|meaning|
+| :---: | :---: | :---: | :---: |
+|det|bool|true|Whether to perform text detection in the forward direction|
+|rec|bool|true|Whether to perform text recognition in the forward direction|
+|cls|bool|false|Whether to perform text direction classification in the forward direction|
+
+Detection related parameters
+
+|parameter|data type|default|meaning|
+| --- | --- | --- | --- |
+|det_model_dir|string|-|Address of detection inference model|
+|max_side_len|int|960|Limit the maximum image height and width to 960|
+|det_db_thresh|float|0.3|Used to filter the binarized image of DB prediction, setting 0.-0.3 has no obvious effect on the result|
+|det_db_box_thresh|float|0.5|DB post-processing filter box threshold, if there is a missing box detected, it can be reduced as appropriate|
+|det_db_unclip_ratio|float|1.6|Indicates the compactness of the text box, the smaller the value, the closer the text box to the text|
+|det_db_score_mode|string|slow| slow: use polygon box to calculate bbox score, fast: use rectangle box to calculate. Use rectangular box to calculate faster, and polygonal box more accurate for curved text area.|
+|visualize|bool|true|Whether to visualize the results，when it is set as true, the prediction results will be saved in the folder specified by the `output` field on an image with the same name as the input image.|
+
+Classifier related parameters
+
+|parameter|data type|default|meaning|
+| --- | --- | --- | --- |
+|use_angle_cls|bool|false|Whether to use the direction classifier|
+|cls_model_dir|string|-|Address of direction classifier inference model|
+|cls_thresh|float|0.9|Score threshold of the  direction classifier|
+|cls_batch_num|int|1|batch size of classifier|
+
+Recognition related parameters
+
+|parameter|data type|default|meaning|
+| --- | --- | --- | --- |
+|rec_model_dir|string|-|Address of recognition inference model|
+|rec_char_dict_path|string|../../ppocr/utils/ppocr_keys_v1.txt|dictionary file|
+|rec_batch_num|int|6|batch size of recognition|
+|rec_img_h|int|48|image height of recognition|
+|rec_img_w|int|320|image width of recognition|
+
+Layout related parameters
+
+|parameter|data type|default|meaning|
+| :---: | :---: | :---: | :---: |
+|layout_model_dir|string|-| Address of layout inference model|
+|layout_dict_path|string|../../ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt|dictionary file|
+|layout_score_threshold|float|0.5|Threshold of score.|
+|layout_nms_threshold|float|0.5|Threshold of nms.|
+
+Table recognition related parameters
+
+|parameter|data type|default|meaning|
+| :---: | :---: | :---: | :---: |
+|table_model_dir|string|-|Address of table recognition inference model|
+|table_char_dict_path|string|../../ppocr/utils/dict/table_structure_dict.txt|dictionary file|
+|table_max_len|int|488|The size of the long side of the input image of the table recognition model, the final input image size of the network is（table_max_len，table_max_len）|
+|merge_no_span_structure|bool|true|Whether to merge <td> and </td> to <td></td|
+
+Multi-language inference is also supported in PaddleOCR, you can refer to [recognition tutorial](../ppocr/model_train/recognition.en.md) for more supported languages and models in PaddleOCR. Specifically, if you want to infer using multi-language models, you just need to modify values of `rec_char_dict_path` and `rec_model_dir`.
+
+The detection results will be shown on the screen, which is as follows.
+
+```bash linenums="1"
+predict img: ../../doc/imgs/12.jpg
+../../doc/imgs/12.jpg
+0       det boxes: [[74,553],[427,542],[428,571],[75,582]] rec text: 打浦路252935号 rec score: 0.947724
+1       det boxes: [[23,507],[513,488],[515,529],[24,548]] rec text: 绿洲仕格维花园公寓 rec score: 0.993728
+2       det boxes: [[187,456],[399,448],[400,480],[188,488]] rec text: 打浦路15号 rec score: 0.964994
+3       det boxes: [[42,413],[483,391],[484,428],[43,450]] rec text: 上海斯格威铂尔大酒店 rec score: 0.980086
+The detection visualized image saved in ./output//12.jpg
+```
+
+- layout+table
+
+```bash linenums="1"
+predict img: ../../ppstructure/docs/table/1.png
+0       type: text, region: [12,729,410,848], score: 0.781044, res: count of ocr result is : 7
+********** print ocr result **********
+0       det boxes: [[4,1],[79,1],[79,12],[4,12]] rec text: CTW1500. rec score: 0.769472
+...
+6       det boxes: [[4,99],[391,99],[391,112],[4,112]] rec text: sate-of-the-artmethods[12.34.36l.ourapproachachieves rec score: 0.90414
+********** end print ocr result **********
+1       type: text, region: [69,342,342,359], score: 0.703666, res: count of ocr result is : 1
+********** print ocr result **********
+0       det boxes: [[8,2],[269,2],[269,13],[8,13]] rec text: Table6.Experimentalresults on CTW-1500 rec score: 0.890454
+********** end print ocr result **********
+2       type: text, region: [70,316,706,332], score: 0.659738, res: count of ocr result is : 2
+********** print ocr result **********
+0       det boxes: [[373,2],[630,2],[630,11],[373,11]] rec text: oroposals.andthegreencontoursarefinal rec score: 0.919729
+1       det boxes: [[8,3],[357,3],[357,11],[8,11]] rec text: Visualexperimentalresultshebluecontoursareboundar rec score: 0.915963
+********** end print ocr result **********
+3       type: text, region: [489,342,789,359], score: 0.630538, res: count of ocr result is : 1
+********** print ocr result **********
+0       det boxes: [[8,2],[294,2],[294,14],[8,14]] rec text: Table7.Experimentalresults onMSRA-TD500 rec score: 0.942251
+********** end print ocr result **********
+4       type: text, region: [444,751,841,848], score: 0.607345, res: count of ocr result is : 5
+********** print ocr result **********
+0       det boxes: [[19,3],[389,3],[389,17],[19,17]] rec text: Inthispaper,weproposeanovel adaptivebound rec score: 0.941031
+1       det boxes: [[4,22],[390,22],[390,36],[4,36]] rec text: aryproposalnetworkforarbitraryshapetextdetection rec score: 0.960172
+2       det boxes: [[4,42],[392,42],[392,56],[4,56]] rec text: whichadoptanboundaryproposalmodeltogeneratecoarse rec score: 0.934647
+3       det boxes: [[4,61],[389,61],[389,75],[4,75]] rec text: ooundaryproposals,andthenadoptanadaptiveboundary rec score: 0.946296
+4       det boxes: [[5,80],[387,80],[387,93],[5,93]] rec text: leformationmodelcombinedwithGCNandRNNtoper rec score: 0.952401
+********** end print ocr result **********
+5       type: title, region: [444,705,564,724], score: 0.785429, res: count of ocr result is : 1
+********** print ocr result **********
+0       det boxes: [[6,2],[113,2],[113,14],[6,14]] rec text: 5.Conclusion rec score: 0.856903
+********** end print ocr result **********
+6       type: table, region: [14,360,402,711], score: 0.963643, res: <html><body><table><thead><tr><td>Methods</td><td>Ext</td><td>R</td><td>P</td><td>F</td><td>FPS</td></tr></thead><tbody><tr><td>TextSnake [18]</td><td>Syn</td><td>85.3</td><td>67.9</td><td>75.6</td><td></td></tr><tr><td>CSE [17]</td><td>MiLT</td><td>76.1</td><td>78.7</td><td>77.4</td><td>0.38</td></tr><tr><td>LOMO[40]</td><td>Syn</td><td>76.5</td><td>85.7</td><td>80.8</td><td>4.4</td></tr><tr><td>ATRR[35]</td><td>Sy-</td><td>80.2</td><td>80.1</td><td>80.1</td><td>-</td></tr><tr><td>SegLink++ [28]</td><td>Syn</td><td>79.8</td><td>82.8</td><td>81.3</td><td>-</td></tr><tr><td>TextField [37]</td><td>Syn</td><td>79.8</td><td>83.0</td><td>81.4</td><td>6.0</td></tr><tr><td>MSR[38]</td><td>Syn</td><td>79.0</td><td>84.1</td><td>81.5</td><td>4.3</td></tr><tr><td>PSENet-1s [33]</td><td>MLT</td><td>79.7</td><td>84.8</td><td>82.2</td><td>3.9</td></tr><tr><td>DB [12]</td><td>Syn</td><td>80.2</td><td>86.9</td><td>83.4</td><td>22.0</td></tr><tr><td>CRAFT [2]</td><td>Syn</td><td>81.1</td><td>86.0</td><td>83.5</td><td>-</td></tr><tr><td>TextDragon [5]</td><td>MLT+</td><td>82.8</td><td>84.5</td><td>83.6</td><td></td></tr><tr><td>PAN [34]</td><td>Syn</td><td>81.2</td><td>86.4</td><td>83.7</td><td>39.8</td></tr><tr><td>ContourNet [36]</td><td></td><td>84.1</td><td>83.7</td><td>83.9</td><td>4.5</td></tr><tr><td>DRRG [41]</td><td>MLT</td><td>83.02</td><td>85.93</td><td>84.45</td><td>-</td></tr><tr><td>TextPerception[23]</td><td>Syn</td><td>81.9</td><td>87.5</td><td>84.6</td><td></td></tr><tr><td>Ours</td><td> Syn</td><td>80.57</td><td>87.66</td><td>83.97</td><td>12.08</td></tr><tr><td>Ours</td><td></td><td>81.45</td><td>87.81</td><td>84.51</td><td>12.15</td></tr><tr><td>Ours</td><td>MLT</td><td>83.60</td><td>86.45</td><td>85.00</td><td>12.21</td></tr></tbody></table></body></html>
+The table visualized image saved in ./output//6_1.png
+7       type: table, region: [462,359,820,657], score: 0.953917, res: <html><body><table><thead><tr><td>Methods</td><td>R</td><td>P</td><td>F</td><td>FPS</td></tr></thead><tbody><tr><td>SegLink [26]</td><td>70.0</td><td>86.0</td><td>77.0</td><td>8.9</td></tr><tr><td>PixelLink [4]</td><td>73.2</td><td>83.0</td><td>77.8</td><td>-</td></tr><tr><td>TextSnake [18]</td><td>73.9</td><td>83.2</td><td>78.3</td><td>1.1</td></tr><tr><td>TextField [37]</td><td>75.9</td><td>87.4</td><td>81.3</td><td>5.2 </td></tr><tr><td>MSR[38]</td><td>76.7</td><td>87.4</td><td>81.7</td><td>-</td></tr><tr><td>FTSN[3]</td><td>77.1</td><td>87.6</td><td>82.0</td><td>:</td></tr><tr><td>LSE[30]</td><td>81.7</td><td>84.2</td><td>82.9</td><td></td></tr><tr><td>CRAFT [2]</td><td>78.2</td><td>88.2</td><td>82.9</td><td>8.6</td></tr><tr><td>MCN [16]</td><td>79</td><td>88</td><td>83</td><td>-</td></tr><tr><td>ATRR[35]</td><td>82.1</td><td>85.2</td><td>83.6</td><td>-</td></tr><tr><td>PAN [34]</td><td>83.8</td><td>84.4</td><td>84.1</td><td>30.2</td></tr><tr><td>DB[12]</td><td>79.2</td><td>91.5</td><td>84.9</td><td>32.0</td></tr><tr><td>DRRG [41]</td><td>82.30</td><td>88.05</td><td>85.08</td><td>-</td></tr><tr><td>Ours (SynText)</td><td>80.68</td><td>85.40</td><td>82.97</td><td>12.68</td></tr><tr><td>Ours (MLT-17)</td><td>84.54</td><td>86.62</td><td>85.57</td><td>12.31</td></tr></tbody></table></body></html>
+The table visualized image saved in ./output//7_1.png
+8       type: figure, region: [14,3,836,310], score: 0.969443, res: count of ocr result is : 26
+********** print ocr result **********
+0       det boxes: [[506,14],[539,15],[539,22],[506,21]] rec text: E rec score: 0.318073
+...
+25      det boxes: [[680,290],[759,288],[759,303],[680,305]] rec text: (d) CTW1500 rec score: 0.95911
+********** end print ocr result **********
+```
+
+## 3. FAQ
+
+ 1. Encountered the error `unable to access 'https://github.com/LDOUBLEV/AutoLog.git/': gnutls_handshake() failed: The TLS connection was non-properly terminated.`, change the github address in `deploy/cpp_infer/external-cmake/auto-log.cmake` to the <https://gitee.com/Double_V/AutoLog> address.
diff --git a/docs/version2.x/legacy/cpp_infer.md b/docs/version2.x/legacy/cpp_infer.md
new file mode 100644
index 0000000..b326c43
--- /dev/null
+++ b/docs/version2.x/legacy/cpp_infer.md
@@ -0,0 +1,445 @@
+---
+comments: true
+hide:
+  - toc
+---
+
+# 服务器端C++预测
+
+本章节介绍PaddleOCR 模型的C++部署方法。C++在性能计算上优于Python，因此，在大多数CPU、GPU部署场景，多采用C++的部署方式，本节将介绍如何在Linux\Windows (CPU\GPU)环境下配置C++环境并完成PaddleOCR模型部署。
+
+## 1. 准备环境
+
+### 1.1 运行准备
+
+- Linux环境，推荐使用docker。
+- Windows环境。
+
+- 该文档主要介绍基于Linux环境的PaddleOCR C++预测流程，如果需要在Windows下基于预测库进行C++预测，具体编译方法请参考[Windows下编译教程](./windows_vs2019_build.md)
+
+### 1.2 编译opencv库
+
+- 首先需要从opencv官网上下载在Linux环境下源码编译的包，以opencv3.4.7为例，下载命令如下：
+
+```bash linenums="1"
+cd deploy/cpp_infer
+wget https://paddleocr.bj.bcebos.com/libs/opencv/opencv-3.4.7.tar.gz
+tar -xf opencv-3.4.7.tar.gz
+```
+
+最终可以在当前目录下看到`opencv-3.4.7/`的文件夹。
+
+- 编译opencv，设置opencv源码路径(`root_path`)以及安装路径(`install_path`)。进入opencv源码路径下，按照下面的方式进行编译。
+
+```bash linenums="1"
+root_path="your_opencv_root_path"
+install_path=${root_path}/opencv3
+build_dir=${root_path}/build
+
+rm -rf ${build_dir}
+mkdir ${build_dir}
+cd ${build_dir}
+
+cmake .. \
+    -DCMAKE_INSTALL_PREFIX=${install_path} \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DBUILD_SHARED_LIBS=OFF \
+    -DWITH_IPP=OFF \
+    -DBUILD_IPP_IW=OFF \
+    -DWITH_LAPACK=OFF \
+    -DWITH_EIGEN=OFF \
+    -DCMAKE_INSTALL_LIBDIR=lib64 \
+    -DWITH_ZLIB=ON \
+    -DBUILD_ZLIB=ON \
+    -DWITH_JPEG=ON \
+    -DBUILD_JPEG=ON \
+    -DWITH_PNG=ON \
+    -DBUILD_PNG=ON \
+    -DWITH_TIFF=ON \
+    -DBUILD_TIFF=ON
+
+make -j
+make install
+```
+
+也可以直接修改`tools/build_opencv.sh`的内容，然后直接运行下面的命令进行编译。
+
+```bash linenums="1"
+sh tools/build_opencv.sh
+```
+
+其中`root_path`为下载的opencv源码路径，`install_path`为opencv的安装路径，`make install`完成之后，会在该文件夹下生成opencv头文件和库文件，用于后面的OCR代码编译。
+
+最终在安装路径下的文件结构如下所示。
+
+```
+opencv3/
+|-- bin
+|-- include
+|-- lib
+|-- lib64
+|-- share
+```
+
+### 1.3 下载或者编译Paddle预测库
+
+可以选择直接下载安装或者从源码编译，下文分别进行具体说明。
+
+#### 1.3.1 直接下载安装
+
+[Paddle预测库官网](https://www.paddlepaddle.org.cn/inference/master/guides/install/download_lib.html#linux) 上提供了不同cuda版本的Linux预测库，可以在官网查看并选择合适的预测库版本（*建议选择paddle版本>=2.0.1版本的预测库* ）。
+
+下载之后解压:
+
+```bash linenums="1"
+tar -xf paddle_inference.tgz
+```
+
+最终会在当前的文件夹中生成`paddle_inference/`的子文件夹。
+
+#### 1.3.2 预测库源码编译
+
+如果希望获取最新预测库特性，可以从github上克隆最新Paddle代码进行编译，生成最新的预测库。
+
+使用git获取代码:
+
+```bash linenums="1"
+git clone https://github.com/PaddlePaddle/Paddle.git
+git checkout develop
+```
+
+进入Paddle目录，进行编译:
+
+```bash linenums="1"
+rm -rf build
+mkdir build
+cd build
+
+cmake  .. \
+    -DWITH_CONTRIB=OFF \
+    -DWITH_MKL=ON \
+    -DWITH_MKLDNN=ON  \
+    -DWITH_TESTING=OFF \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DWITH_INFERENCE_API_TEST=OFF \
+    -DON_INFER=ON \
+    -DWITH_PYTHON=ON
+make -j
+make inference_lib_dist
+```
+
+更多编译参数选项介绍可以参考[Paddle预测库编译文档](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0/guides/05_inference_deployment/inference/build_and_install_lib_cn.html#congyuanmabianyi)。
+
+编译完成之后，可以在`build/paddle_inference_install_dir/`文件下看到生成了以下文件及文件夹。
+
+```
+build/paddle_inference_install_dir/
+|-- CMakeCache.txt
+|-- paddle
+|-- third_party
+|-- version.txt
+```
+
+其中`paddle`就是C++预测所需的Paddle库，`version.txt`中包含当前预测库的版本信息。
+
+## 2. 开始运行
+
+### 2.1 准备模型
+
+直接下载PaddleOCR提供的推理模型，或者参考[模型预测章节](./python_infer.md)，将训练好的模型导出为推理模型。模型导出之后，假设放在`inference`目录下，则目录结构如下：
+
+```text linenums="1"
+inference/
+|-- det_db
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+|-- rec_rcnn
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+|-- cls
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+|-- table
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+|-- layout
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+```
+
+### 2.2 编译PaddleOCR C++预测demo
+
+编译命令如下，其中Paddle C++预测库、opencv等其他依赖库的地址需要换成自己机器上的实际地址。
+
+```bash linenums="1"
+sh tools/build.sh
+```
+
+具体的，需要修改`tools/build.sh`中环境路径，相关内容如下：
+
+```bash linenums="1"
+OPENCV_DIR=your_opencv_dir
+LIB_DIR=your_paddle_inference_dir
+CUDA_LIB_DIR=your_cuda_lib_dir
+CUDNN_LIB_DIR=/your_cudnn_lib_dir
+```
+
+其中，`OPENCV_DIR`为opencv编译安装的地址；`LIB_DIR`为下载(`paddle_inference`文件夹)或者编译生成的Paddle预测库地址(`build/paddle_inference_install_dir`文件夹)；`CUDA_LIB_DIR`为cuda库文件地址，在docker中为`/usr/local/cuda/lib64`；`CUDNN_LIB_DIR`为cudnn库文件地址，在docker中为`/usr/lib/x86_64-linux-gnu/`。**注意：以上路径都写绝对路径，不要写相对路径。**
+
+编译完成之后，会在`build`文件夹下生成一个名为`ppocr`的可执行文件。
+
+### 2.3 运行demo
+
+本demo支持系统串联调用，也支持单个功能的调用，如，只使用检测或识别功能。
+
+**注意** ppocr默认使用`PP-OCRv3`模型，识别模型使用的输入shape为`3,48,320`, 如需使用旧版本的PP-OCR模型，则需要设置参数`--rec_img_h=32`。
+
+运行方式：
+
+```bash linenums="1"
+./build/ppocr [--param1] [--param2] [...]
+```
+
+具体命令如下：
+
+##### 1. 检测+分类+识别
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --rec_model_dir=inference/rec_rcnn \
+    --cls_model_dir=inference/cls \
+    --image_dir=../../doc/imgs/12.jpg \
+    --use_angle_cls=true \
+    --det=true \
+    --rec=true \
+    --cls=true \
+```
+
+##### 2. 检测+识别
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --rec_model_dir=inference/rec_rcnn \
+    --image_dir=../../doc/imgs/12.jpg \
+    --use_angle_cls=false \
+    --det=true \
+    --rec=true \
+    --cls=false \
+```
+
+##### 3. 检测
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --image_dir=../../doc/imgs/12.jpg \
+    --det=true \
+    --rec=false
+```
+
+##### 4. 分类+识别
+
+```bash linenums="1"
+./build/ppocr --rec_model_dir=inference/rec_rcnn \
+    --cls_model_dir=inference/cls \
+    --image_dir=../../doc/imgs_words/ch/word_1.jpg \
+    --use_angle_cls=true \
+    --det=false \
+    --rec=true \
+    --cls=true \
+```
+
+##### 5. 识别
+
+```bash linenums="1"
+./build/ppocr --rec_model_dir=inference/rec_rcnn \
+    --image_dir=../../doc/imgs_words/ch/word_1.jpg \
+    --use_angle_cls=false \
+    --det=false \
+    --rec=true \
+    --cls=false \
+```
+
+##### 6. 分类
+
+```bash linenums="1"
+./build/ppocr --cls_model_dir=inference/cls \
+    --cls_model_dir=inference/cls \
+    --image_dir=../../doc/imgs_words/ch/word_1.jpg \
+    --use_angle_cls=true \
+    --det=false \
+    --rec=false \
+    --cls=true \
+```
+
+##### 7. 版面分析+表格识别
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --rec_model_dir=inference/rec_rcnn \
+    --table_model_dir=inference/table \
+    --image_dir=../../ppstructure/docs/table/table.jpg \
+    --layout_model_dir=inference/layout \
+    --type=structure \
+    --table=true \
+    --layout=true
+```
+
+##### 8. 版面分析
+
+```bash linenums="1"
+./build/ppocr --layout_model_dir=inference/layout \
+    --image_dir=../../ppstructure/docs/table/1.png \
+    --type=structure \
+    --table=false \
+    --layout=true \
+    --det=false \
+    --rec=false
+```
+
+##### 9. 表格识别
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --rec_model_dir=inference/rec_rcnn \
+    --table_model_dir=inference/table \
+    --image_dir=../../ppstructure/docs/table/table.jpg \
+    --type=structure \
+    --table=true
+```
+
+更多支持的可调节参数解释如下：
+
+通用参数
+
+|           参数名称           | 类型  | 默认参数 |                               意义                                |
+| :--------------------------: | :---: | :------: | :---------------------------------------------------------------: |
+|           use_gpu            | bool  |  false   |                            是否使用GPU                            |
+|            gpu_id            |  int  |    0     |                       GPU id，使用GPU时有效                       |
+|           gpu_mem            |  int  |   4000   |                           申请的GPU内存                           |
+| cpu_math_library_num_threads |  int  |    10    | CPU预测时的线程数，在机器核数充足的情况下，该值越大，预测速度越快 |
+|        enable_mkldnn         | bool  |   true   |                         是否使用mkldnn库                          |
+|            output            |  str  | ./output |                       可视化结果保存的路径                        |
+
+前向相关
+
+| 参数名称 | 类型  | 默认参数 |           意义           |
+| :------: | :---: | :------: | :----------------------: |
+|   det    | bool  |   true   |   前向是否执行文字检测   |
+|   rec    | bool  |   true   |   前向是否执行文字识别   |
+|   cls    | bool  |  false   | 前向是否执行文字方向分类 |
+
+检测模型相关
+
+|      参数名称       |  类型  | 默认参数 |                                                     意义                                                     |
+| :-----------------: | :----: | :------: | :----------------------------------------------------------------------------------------------------------: |
+|    det_model_dir    | string |    -     |                                         检测模型inference model地址                                          |
+|    max_side_len     |  int   |   960    |                          输入图像长宽大于960时，等比例缩放图像，使得图像最长边为960                          |
+|    det_db_thresh    | float  |   0.3    |                           用于过滤DB预测的二值化图像，设置为0.-0.3对结果影响不明显                           |
+|  det_db_box_thresh  | float  |   0.5    |                           DB后处理过滤box的阈值，如果检测存在漏框情况，可酌情减小                            |
+| det_db_unclip_ratio | float  |   1.6    |                                 表示文本框的紧致程度，越小则文本框更靠近文本                                 |
+|  det_db_score_mode  | string |   slow   | slow:使用多边形框计算bbox score，fast:使用矩形框计算。矩形框计算速度更快，多边形框对弯曲文本区域计算更准确。 |
+|      visualize      |  bool  |   true   |       是否对结果进行可视化，为1时，预测结果会保存在`output`字段指定的文件夹下和输入图像同名的图像上。        |
+
+方向分类器相关
+
+|   参数名称    |  类型  | 默认参数 |             意义              |
+| :-----------: | :----: | :------: | :---------------------------: |
+| use_angle_cls |  bool  |  false   |      是否使用方向分类器       |
+| cls_model_dir | string |    -     | 方向分类器inference model地址 |
+|  cls_thresh   | float  |   0.9    |     方向分类器的得分阈值      |
+| cls_batch_num |  int   |    1     |      方向分类器batchsize      |
+
+文字识别模型相关
+
+|      参数名称      |  类型  |              默认参数               |              意义               |
+| :----------------: | :----: | :---------------------------------: | :-----------------------------: |
+|   rec_model_dir    | string |                  -                  | 文字识别模型inference model地址 |
+| rec_char_dict_path | string | ../../ppocr/utils/ppocr_keys_v1.txt |            字典文件             |
+|   rec_batch_num    |  int   |                  6                  |      文字识别模型batchsize      |
+|     rec_img_h      |  int   |                 48                  |    文字识别模型输入图像高度     |
+|     rec_img_w      |  int   |                 320                 |    文字识别模型输入图像宽度     |
+
+版面分析模型相关
+
+|        参数名称        |  类型  |                           默认参数                           |              意义               |
+| :--------------------: | :----: | :----------------------------------------------------------: | :-----------------------------: |
+|    layout_model_dir    | string |                              -                               | 版面分析模型inference model地址 |
+|    layout_dict_path    | string | ../../ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt |            字典文件             |
+| layout_score_threshold | float  |                             0.5                              |        检测框的分数阈值         |
+|  layout_nms_threshold  | float  |                             0.5                              |            nms的阈值            |
+
+表格识别模型相关
+
+|        参数名称         |  类型  |                      默认参数                      |                                         意义                                         |
+| :---------------------: | :----: | :------------------------------------------------: | :----------------------------------------------------------------------------------: |
+|     table_model_dir     | string |                         -                          |                           表格识别模型inference model地址                            |
+|  table_char_dict_path   | string | ../../ppocr/utils/dict/table_structure_dict_ch.txt |                                       字典文件                                       |
+|      table_max_len      |  int   |                        488                         | 表格识别模型输入图像长边大小，最终网络输入图像大小为（table_max_len，table_max_len） |
+| merge_no_span_structure |  bool  |                        true                        |                          是否合并<td> 和 </td> 为<td></td>                           |
+
+- PaddleOCR也支持多语言的预测，更多支持的语言和模型可以参考[识别文档](../ppocr/model_train/recognition.md)中的多语言字典与模型部分，如果希望进行多语言预测，只需将修改`rec_char_dict_path`（字典文件路径）以及`rec_model_dir`（inference模型路径）字段即可。
+
+最终屏幕上会输出检测结果如下：
+
+- ocr
+
+```bash linenums="1"
+predict img: ../../doc/imgs/12.jpg
+../../doc/imgs/12.jpg
+0       det boxes: [[74,553],[427,542],[428,571],[75,582]] rec text: 打浦路252935号 rec score: 0.947724
+1       det boxes: [[23,507],[513,488],[515,529],[24,548]] rec text: 绿洲仕格维花园公寓 rec score: 0.993728
+2       det boxes: [[187,456],[399,448],[400,480],[188,488]] rec text: 打浦路15号 rec score: 0.964994
+3       det boxes: [[42,413],[483,391],[484,428],[43,450]] rec text: 上海斯格威铂尔大酒店 rec score: 0.980086
+The detection visualized image saved in ./output//12.jpg
+```
+
+- layout+table
+
+```bash linenums="1"
+predict img: ../../ppstructure/docs/table/1.png
+0       type: text, region: [12,729,410,848], score: 0.781044, res: count of ocr result is : 7
+********** print ocr result **********
+0       det boxes: [[4,1],[79,1],[79,12],[4,12]] rec text: CTW1500. rec score: 0.769472
+...
+6       det boxes: [[4,99],[391,99],[391,112],[4,112]] rec text: sate-of-the-artmethods[12.34.36l.ourapproachachieves rec score: 0.90414
+********** end print ocr result **********
+1       type: text, region: [69,342,342,359], score: 0.703666, res: count of ocr result is : 1
+********** print ocr result **********
+0       det boxes: [[8,2],[269,2],[269,13],[8,13]] rec text: Table6.Experimentalresults on CTW-1500 rec score: 0.890454
+********** end print ocr result **********
+2       type: text, region: [70,316,706,332], score: 0.659738, res: count of ocr result is : 2
+********** print ocr result **********
+0       det boxes: [[373,2],[630,2],[630,11],[373,11]] rec text: oroposals.andthegreencontoursarefinal rec score: 0.919729
+1       det boxes: [[8,3],[357,3],[357,11],[8,11]] rec text: Visualexperimentalresultshebluecontoursareboundar rec score: 0.915963
+********** end print ocr result **********
+3       type: text, region: [489,342,789,359], score: 0.630538, res: count of ocr result is : 1
+********** print ocr result **********
+0       det boxes: [[8,2],[294,2],[294,14],[8,14]] rec text: Table7.Experimentalresults onMSRA-TD500 rec score: 0.942251
+********** end print ocr result **********
+4       type: text, region: [444,751,841,848], score: 0.607345, res: count of ocr result is : 5
+********** print ocr result **********
+0       det boxes: [[19,3],[389,3],[389,17],[19,17]] rec text: Inthispaper,weproposeanovel adaptivebound rec score: 0.941031
+1       det boxes: [[4,22],[390,22],[390,36],[4,36]] rec text: aryproposalnetworkforarbitraryshapetextdetection rec score: 0.960172
+2       det boxes: [[4,42],[392,42],[392,56],[4,56]] rec text: whichadoptanboundaryproposalmodeltogeneratecoarse rec score: 0.934647
+3       det boxes: [[4,61],[389,61],[389,75],[4,75]] rec text: ooundaryproposals,andthenadoptanadaptiveboundary rec score: 0.946296
+4       det boxes: [[5,80],[387,80],[387,93],[5,93]] rec text: leformationmodelcombinedwithGCNandRNNtoper rec score: 0.952401
+********** end print ocr result **********
+5       type: title, region: [444,705,564,724], score: 0.785429, res: count of ocr result is : 1
+********** print ocr result **********
+0       det boxes: [[6,2],[113,2],[113,14],[6,14]] rec text: 5.Conclusion rec score: 0.856903
+********** end print ocr result **********
+6       type: table, region: [14,360,402,711], score: 0.963643, res: <html><body><table><thead><tr><td>Methods</td><td>Ext</td><td>R</td><td>P</td><td>F</td><td>FPS</td></tr></thead><tbody><tr><td>TextSnake [18]</td><td>Syn</td><td>85.3</td><td>67.9</td><td>75.6</td><td></td></tr><tr><td>CSE [17]</td><td>MiLT</td><td>76.1</td><td>78.7</td><td>77.4</td><td>0.38</td></tr><tr><td>LOMO[40]</td><td>Syn</td><td>76.5</td><td>85.7</td><td>80.8</td><td>4.4</td></tr><tr><td>ATRR[35]</td><td>Sy-</td><td>80.2</td><td>80.1</td><td>80.1</td><td>-</td></tr><tr><td>SegLink++ [28]</td><td>Syn</td><td>79.8</td><td>82.8</td><td>81.3</td><td>-</td></tr><tr><td>TextField [37]</td><td>Syn</td><td>79.8</td><td>83.0</td><td>81.4</td><td>6.0</td></tr><tr><td>MSR[38]</td><td>Syn</td><td>79.0</td><td>84.1</td><td>81.5</td><td>4.3</td></tr><tr><td>PSENet-1s [33]</td><td>MLT</td><td>79.7</td><td>84.8</td><td>82.2</td><td>3.9</td></tr><tr><td>DB [12]</td><td>Syn</td><td>80.2</td><td>86.9</td><td>83.4</td><td>22.0</td></tr><tr><td>CRAFT [2]</td><td>Syn</td><td>81.1</td><td>86.0</td><td>83.5</td><td>-</td></tr><tr><td>TextDragon [5]</td><td>MLT+</td><td>82.8</td><td>84.5</td><td>83.6</td><td></td></tr><tr><td>PAN [34]</td><td>Syn</td><td>81.2</td><td>86.4</td><td>83.7</td><td>39.8</td></tr><tr><td>ContourNet [36]</td><td></td><td>84.1</td><td>83.7</td><td>83.9</td><td>4.5</td></tr><tr><td>DRRG [41]</td><td>MLT</td><td>83.02</td><td>85.93</td><td>84.45</td><td>-</td></tr><tr><td>TextPerception[23]</td><td>Syn</td><td>81.9</td><td>87.5</td><td>84.6</td><td></td></tr><tr><td>Ours</td><td> Syn</td><td>80.57</td><td>87.66</td><td>83.97</td><td>12.08</td></tr><tr><td>Ours</td><td></td><td>81.45</td><td>87.81</td><td>84.51</td><td>12.15</td></tr><tr><td>Ours</td><td>MLT</td><td>83.60</td><td>86.45</td><td>85.00</td><td>12.21</td></tr></tbody></table></body></html>
+The table visualized image saved in ./output//6_1.png
+7       type: table, region: [462,359,820,657], score: 0.953917, res: <html><body><table><thead><tr><td>Methods</td><td>R</td><td>P</td><td>F</td><td>FPS</td></tr></thead><tbody><tr><td>SegLink [26]</td><td>70.0</td><td>86.0</td><td>77.0</td><td>8.9</td></tr><tr><td>PixelLink [4]</td><td>73.2</td><td>83.0</td><td>77.8</td><td>-</td></tr><tr><td>TextSnake [18]</td><td>73.9</td><td>83.2</td><td>78.3</td><td>1.1</td></tr><tr><td>TextField [37]</td><td>75.9</td><td>87.4</td><td>81.3</td><td>5.2 </td></tr><tr><td>MSR[38]</td><td>76.7</td><td>87.4</td><td>81.7</td><td>-</td></tr><tr><td>FTSN[3]</td><td>77.1</td><td>87.6</td><td>82.0</td><td>:</td></tr><tr><td>LSE[30]</td><td>81.7</td><td>84.2</td><td>82.9</td><td></td></tr><tr><td>CRAFT [2]</td><td>78.2</td><td>88.2</td><td>82.9</td><td>8.6</td></tr><tr><td>MCN [16]</td><td>79</td><td>88</td><td>83</td><td>-</td></tr><tr><td>ATRR[35]</td><td>82.1</td><td>85.2</td><td>83.6</td><td>-</td></tr><tr><td>PAN [34]</td><td>83.8</td><td>84.4</td><td>84.1</td><td>30.2</td></tr><tr><td>DB[12]</td><td>79.2</td><td>91.5</td><td>84.9</td><td>32.0</td></tr><tr><td>DRRG [41]</td><td>82.30</td><td>88.05</td><td>85.08</td><td>-</td></tr><tr><td>Ours (SynText)</td><td>80.68</td><td>85.40</td><td>82.97</td><td>12.68</td></tr><tr><td>Ours (MLT-17)</td><td>84.54</td><td>86.62</td><td>85.57</td><td>12.31</td></tr></tbody></table></body></html>
+The table visualized image saved in ./output//7_1.png
+8       type: figure, region: [14,3,836,310], score: 0.969443, res: count of ocr result is : 26
+********** print ocr result **********
+0       det boxes: [[506,14],[539,15],[539,22],[506,21]] rec text: E rec score: 0.318073
+...
+25      det boxes: [[680,290],[759,288],[759,303],[680,305]] rec text: (d) CTW1500 rec score: 0.95911
+********** end print ocr result **********
+```
+
+## 3. FAQ
+
+ 1. 遇到报错 `unable to access 'https://github.com/LDOUBLEV/AutoLog.git/': gnutls_handshake() failed: The TLS connection was non-properly terminated.`， 将 `deploy/cpp_infer/external-cmake/auto-log.cmake` 中的github地址改为 <https://gitee.com/Double_V/AutoLog> 地址即可。
diff --git a/docs/version2.x/legacy/images/00057937.jpg b/docs/version2.x/legacy/images/00057937.jpg
new file mode 100644
index 0000000..a35896e
Binary files /dev/null and b/docs/version2.x/legacy/images/00057937.jpg differ
diff --git a/docs/version2.x/legacy/images/1.jpg b/docs/version2.x/legacy/images/1.jpg
new file mode 100644
index 0000000..48a8938
Binary files /dev/null and b/docs/version2.x/legacy/images/1.jpg differ
diff --git a/docs/version2.x/legacy/images/196874354-1b5eecb0-f273-403c-aa6c-4463bf6d78db.png b/docs/version2.x/legacy/images/196874354-1b5eecb0-f273-403c-aa6c-4463bf6d78db.png
new file mode 100644
index 0000000..708871f
Binary files /dev/null and b/docs/version2.x/legacy/images/196874354-1b5eecb0-f273-403c-aa6c-4463bf6d78db.png differ
diff --git a/docs/version2.x/legacy/images/197918203-c7d46f8a-75d4-47f9-9687-405ee0d6727e.gif b/docs/version2.x/legacy/images/197918203-c7d46f8a-75d4-47f9-9687-405ee0d6727e.gif
new file mode 100644
index 0000000..fdd910a
Binary files /dev/null and b/docs/version2.x/legacy/images/197918203-c7d46f8a-75d4-47f9-9687-405ee0d6727e.gif differ
diff --git a/docs/version2.x/legacy/images/68747470733a.jpeg b/docs/version2.x/legacy/images/68747470733a.jpeg
new file mode 100644
index 0000000..ca3d966
Binary files /dev/null and b/docs/version2.x/legacy/images/68747470733a.jpeg differ
diff --git a/docs/version2.x/legacy/images/68747470733a.png b/docs/version2.x/legacy/images/68747470733a.png
new file mode 100644
index 0000000..04818fb
Binary files /dev/null and b/docs/version2.x/legacy/images/68747470733a.png differ
diff --git a/docs/version2.x/legacy/images/68747470733a2f2f7061.jpeg b/docs/version2.x/legacy/images/68747470733a2f2f7061.jpeg
new file mode 100644
index 0000000..090c84f
Binary files /dev/null and b/docs/version2.x/legacy/images/68747470733a2f2f7061.jpeg differ
diff --git a/docs/version2.x/legacy/images/68747470733adfd.jpeg b/docs/version2.x/legacy/images/68747470733adfd.jpeg
new file mode 100644
index 0000000..f75803f
Binary files /dev/null and b/docs/version2.x/legacy/images/68747470733adfd.jpeg differ
diff --git a/docs/version2.x/legacy/images/cmake_step1.png b/docs/version2.x/legacy/images/cmake_step1.png
new file mode 100644
index 0000000..3b9a762
Binary files /dev/null and b/docs/version2.x/legacy/images/cmake_step1.png differ
diff --git a/docs/version2.x/legacy/images/cmake_step2.jpg b/docs/version2.x/legacy/images/cmake_step2.jpg
new file mode 100644
index 0000000..3438157
Binary files /dev/null and b/docs/version2.x/legacy/images/cmake_step2.jpg differ
diff --git a/docs/version2.x/legacy/images/cmake_step3.jpg b/docs/version2.x/legacy/images/cmake_step3.jpg
new file mode 100644
index 0000000..a9483d2
Binary files /dev/null and b/docs/version2.x/legacy/images/cmake_step3.jpg differ
diff --git a/docs/version2.x/legacy/images/cmake_step4.jpg b/docs/version2.x/legacy/images/cmake_step4.jpg
new file mode 100644
index 0000000..d6eb71c
Binary files /dev/null and b/docs/version2.x/legacy/images/cmake_step4.jpg differ
diff --git a/docs/version2.x/legacy/images/deployment.jpg b/docs/version2.x/legacy/images/deployment.jpg
new file mode 100644
index 0000000..c20897e
Binary files /dev/null and b/docs/version2.x/legacy/images/deployment.jpg differ
diff --git a/docs/version2.x/legacy/images/deployment_en.jpg b/docs/version2.x/legacy/images/deployment_en.jpg
new file mode 100644
index 0000000..9920bcc
Binary files /dev/null and b/docs/version2.x/legacy/images/deployment_en.jpg differ
diff --git a/docs/version2.x/legacy/images/det_res_00018069.jpg b/docs/version2.x/legacy/images/det_res_00018069.jpg
new file mode 100644
index 0000000..02f35de
Binary files /dev/null and b/docs/version2.x/legacy/images/det_res_00018069.jpg differ
diff --git a/docs/version2.x/legacy/images/det_res_french_0.jpg b/docs/version2.x/legacy/images/det_res_french_0.jpg
new file mode 100644
index 0000000..5f0e488
Binary files /dev/null and b/docs/version2.x/legacy/images/det_res_french_0.jpg differ
diff --git a/docs/version2.x/legacy/images/det_res_img623_sast.jpg b/docs/version2.x/legacy/images/det_res_img623_sast.jpg
new file mode 100644
index 0000000..af5e2d6
Binary files /dev/null and b/docs/version2.x/legacy/images/det_res_img623_sast.jpg differ
diff --git a/docs/version2.x/legacy/images/det_res_img_10_db.jpg b/docs/version2.x/legacy/images/det_res_img_10_db.jpg
new file mode 100644
index 0000000..6af89f6
Binary files /dev/null and b/docs/version2.x/legacy/images/det_res_img_10_db.jpg differ
diff --git a/docs/version2.x/legacy/images/det_res_img_10_east.jpg b/docs/version2.x/legacy/images/det_res_img_10_east.jpg
new file mode 100644
index 0000000..908d077
Binary files /dev/null and b/docs/version2.x/legacy/images/det_res_img_10_east.jpg differ
diff --git a/docs/version2.x/legacy/images/det_res_img_10_sast.jpg b/docs/version2.x/legacy/images/det_res_img_10_sast.jpg
new file mode 100644
index 0000000..702f773
Binary files /dev/null and b/docs/version2.x/legacy/images/det_res_img_10_sast.jpg differ
diff --git a/docs/version2.x/legacy/images/doc.jpg b/docs/version2.x/legacy/images/doc.jpg
new file mode 100644
index 0000000..f57e62a
Binary files /dev/null and b/docs/version2.x/legacy/images/doc.jpg differ
diff --git a/docs/version2.x/legacy/images/img_10_east_starnet.jpg b/docs/version2.x/legacy/images/img_10_east_starnet.jpg
new file mode 100644
index 0000000..fd8c039
Binary files /dev/null and b/docs/version2.x/legacy/images/img_10_east_starnet.jpg differ
diff --git a/docs/version2.x/legacy/images/img_12.jpg b/docs/version2.x/legacy/images/img_12.jpg
new file mode 100644
index 0000000..b0d2895
Binary files /dev/null and b/docs/version2.x/legacy/images/img_12.jpg differ
diff --git a/docs/version2.x/legacy/images/img_12_result.jpg b/docs/version2.x/legacy/images/img_12_result.jpg
new file mode 100644
index 0000000..822d562
Binary files /dev/null and b/docs/version2.x/legacy/images/img_12_result.jpg differ
diff --git a/docs/version2.x/legacy/images/import_demo.jpg b/docs/version2.x/legacy/images/import_demo.jpg
new file mode 100644
index 0000000..5bb244a
Binary files /dev/null and b/docs/version2.x/legacy/images/import_demo.jpg differ
diff --git a/docs/version2.x/legacy/images/lite_demo.png b/docs/version2.x/legacy/images/lite_demo.png
new file mode 100644
index 0000000..c9daf1b
Binary files /dev/null and b/docs/version2.x/legacy/images/lite_demo.png differ
diff --git a/docs/version2.x/legacy/images/lite_demo_onnx.png b/docs/version2.x/legacy/images/lite_demo_onnx.png
new file mode 100644
index 0000000..b096f6e
Binary files /dev/null and b/docs/version2.x/legacy/images/lite_demo_onnx.png differ
diff --git a/docs/version2.x/legacy/images/lite_demo_paddle.png b/docs/version2.x/legacy/images/lite_demo_paddle.png
new file mode 100644
index 0000000..b096f6e
Binary files /dev/null and b/docs/version2.x/legacy/images/lite_demo_paddle.png differ
diff --git a/docs/version2.x/legacy/images/pipeline_result.png b/docs/version2.x/legacy/images/pipeline_result.png
new file mode 100644
index 0000000..ba7f24a
Binary files /dev/null and b/docs/version2.x/legacy/images/pipeline_result.png differ
diff --git a/docs/version2.x/legacy/images/result.jpg b/docs/version2.x/legacy/images/result.jpg
new file mode 100644
index 0000000..f15955b
Binary files /dev/null and b/docs/version2.x/legacy/images/result.jpg differ
diff --git a/docs/version2.x/legacy/images/results.png b/docs/version2.x/legacy/images/results.png
new file mode 100644
index 0000000..35322bf
Binary files /dev/null and b/docs/version2.x/legacy/images/results.png differ
diff --git a/docs/version2.x/legacy/images/run_cls.jpg b/docs/version2.x/legacy/images/run_cls.jpg
new file mode 100644
index 0000000..602c400
Binary files /dev/null and b/docs/version2.x/legacy/images/run_cls.jpg differ
diff --git a/docs/version2.x/legacy/images/run_cls_rec.jpg b/docs/version2.x/legacy/images/run_cls_rec.jpg
new file mode 100644
index 0000000..a93fb8e
Binary files /dev/null and b/docs/version2.x/legacy/images/run_cls_rec.jpg differ
diff --git a/docs/version2.x/legacy/images/run_det.jpg b/docs/version2.x/legacy/images/run_det.jpg
new file mode 100644
index 0000000..e911978
Binary files /dev/null and b/docs/version2.x/legacy/images/run_det.jpg differ
diff --git a/docs/version2.x/legacy/images/run_det_cls_rec-20240806100407184.jpg b/docs/version2.x/legacy/images/run_det_cls_rec-20240806100407184.jpg
new file mode 100644
index 0000000..9c862ec
Binary files /dev/null and b/docs/version2.x/legacy/images/run_det_cls_rec-20240806100407184.jpg differ
diff --git a/docs/version2.x/legacy/images/run_det_cls_rec.jpg b/docs/version2.x/legacy/images/run_det_cls_rec.jpg
new file mode 100644
index 0000000..9c862ec
Binary files /dev/null and b/docs/version2.x/legacy/images/run_det_cls_rec.jpg differ
diff --git a/docs/version2.x/legacy/images/run_det_rec.jpg b/docs/version2.x/legacy/images/run_det_rec.jpg
new file mode 100644
index 0000000..842c5df
Binary files /dev/null and b/docs/version2.x/legacy/images/run_det_rec.jpg differ
diff --git a/docs/version2.x/legacy/images/run_rec.jpg b/docs/version2.x/legacy/images/run_rec.jpg
new file mode 100644
index 0000000..45ec327
Binary files /dev/null and b/docs/version2.x/legacy/images/run_rec.jpg differ
diff --git a/docs/version2.x/legacy/images/start_server.png b/docs/version2.x/legacy/images/start_server.png
new file mode 100644
index 0000000..60e19cc
Binary files /dev/null and b/docs/version2.x/legacy/images/start_server.png differ
diff --git a/docs/version2.x/legacy/images/system_res_00018069_v3.jpg b/docs/version2.x/legacy/images/system_res_00018069_v3.jpg
new file mode 100644
index 0000000..51808ca
Binary files /dev/null and b/docs/version2.x/legacy/images/system_res_00018069_v3.jpg differ
diff --git a/docs/version2.x/legacy/images/vs_step1.jpg b/docs/version2.x/legacy/images/vs_step1.jpg
new file mode 100644
index 0000000..69ab06c
Binary files /dev/null and b/docs/version2.x/legacy/images/vs_step1.jpg differ
diff --git a/docs/version2.x/legacy/images/word_1.jpg b/docs/version2.x/legacy/images/word_1.jpg
new file mode 100644
index 0000000..cb5451e
Binary files /dev/null and b/docs/version2.x/legacy/images/word_1.jpg differ
diff --git a/docs/version2.x/legacy/images/word_1.png b/docs/version2.x/legacy/images/word_1.png
new file mode 100644
index 0000000..7b915fd
Binary files /dev/null and b/docs/version2.x/legacy/images/word_1.png differ
diff --git a/docs/version2.x/legacy/images/word_10.png b/docs/version2.x/legacy/images/word_10.png
new file mode 100644
index 0000000..07370f7
Binary files /dev/null and b/docs/version2.x/legacy/images/word_10.png differ
diff --git a/docs/version2.x/legacy/images/word_336.png b/docs/version2.x/legacy/images/word_336.png
new file mode 100644
index 0000000..3bddd29
Binary files /dev/null and b/docs/version2.x/legacy/images/word_336.png differ
diff --git a/docs/version2.x/legacy/images/word_4.jpg b/docs/version2.x/legacy/images/word_4.jpg
new file mode 100644
index 0000000..2c34cd3
Binary files /dev/null and b/docs/version2.x/legacy/images/word_4.jpg differ
diff --git a/docs/version2.x/legacy/index.en.md b/docs/version2.x/legacy/index.en.md
new file mode 100644
index 0000000..032ae84
--- /dev/null
+++ b/docs/version2.x/legacy/index.en.md
@@ -0,0 +1,30 @@
+---
+typora-copy-images-to: images
+comments: true
+hide:
+  - toc
+---
+
+# Legacy Features
+
+## Overview
+
+This section introduces the features and models related to the PaddleOCR 2.x branch. Due to the upgrades in the 3.x branch, some models and features are no longer compatible with the older branch. Therefore, users who need to use or refer to the features of the older branch can refer to this part of the documentation.
+
+## Models Supported by PaddleOCR 2.x Branch:
+
+* [Model List](../ppocr/model_list.en.md)
+
+## Features Supported by PaddleOCR 2.x Branch:
+
+* [Inference with Python Prediction Engine](python_infer.en.md)
+* [Inference with C++ Prediction Engine](cpp_infer.en.md)
+* [Compilation Guide for Visual Studio 2019 Community CMake](windows_vs2019_build.en.md)
+* [Service-Oriented Deployment](paddle_server.en.md)
+* [Android Deployment](android_demo.en.md)
+* [Jetson Deployment](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.10/docs/infer_deploy/Jetson_infer.en.md)
+* [Edge Device Deployment](lite.en.md)
+* [Web Frontend Deployment](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.10/docs/infer_deploy/paddle_js.en.md)
+* [Paddle2ONNX Model Conversion and Prediction](paddle2onnx.en.md)
+* [PaddlePaddle Cloud Deployment Tool](paddle_cloud.md)
+* [Benchmark](benchmark.en.md)
diff --git a/docs/version2.x/legacy/index.md b/docs/version2.x/legacy/index.md
new file mode 100644
index 0000000..b5cc8dd
--- /dev/null
+++ b/docs/version2.x/legacy/index.md
@@ -0,0 +1,30 @@
+---
+typora-copy-images-to: images
+comments: true
+hide:
+  - toc
+---
+
+# 历史遗留功能
+
+## 概述
+
+本节介绍了 PaddleOCR 2.x 分支的相关功能和模型。由于 3.x 分支的升级，部分模型和功能与旧分支不再兼容。因此，需要使用或参考旧分支特性的用户可以参考这部分文档。
+
+## PaddleOCR 2.x 分支支持的模型：
+
+* [模型列表](../ppocr/model_list.md)
+
+## PaddleOCR 2.x 分支支持的功能：
+
+* [基于Python预测引擎推理](python_infer.md)
+* [基于C++预测引擎推理](cpp_infer.md)
+* [Visual Studio 2019 Community CMake 编译指南](windows_vs2019_build.md)
+* [服务化部署](paddle_server.md)
+* [Android 部署](android_demo.md)
+* [Jetson 部署](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.10/docs/infer_deploy/Jetson_infer.md)
+* [端侧部署](lite.md)
+* [网页前端部署](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.10/docs/infer_deploy/paddle_js.md)
+* [Paddle2ONNX模型转化与预测](paddle2onnx.md)
+* [云上飞桨部署工具](paddle_cloud.md)
+* [Benchmark](benchmark.md)
diff --git a/docs/version2.x/legacy/lite.en.md b/docs/version2.x/legacy/lite.en.md
new file mode 100644
index 0000000..5c4e227
--- /dev/null
+++ b/docs/version2.x/legacy/lite.en.md
@@ -0,0 +1,309 @@
+---
+comments: true
+hide:
+  - toc
+---
+
+# Mobile deployment based on Paddle-Lite
+
+This tutorial will introduce how to use [Paddle-Lite](https://github.com/PaddlePaddle/Paddle-Lite) to deploy PaddleOCR ultra-lightweight Chinese and English detection models on mobile phones.
+
+Paddle-Lite is a lightweight inference engine for PaddlePaddle. It provides efficient inference capabilities for mobile phones and IoT, and extensively integrates cross-platform hardware to provide lightweight deployment solutions for end-side deployment issues.
+
+## 1. Preparation
+
+### Preparation environment
+
+- Computer (for Compiling Paddle Lite)
+- Mobile phone (arm7 or arm8)
+
+### 1.1 Prepare the cross-compilation environment
+
+The cross-compilation environment is used to compile C++ demos of Paddle Lite and PaddleOCR.
+Supports multiple development environments.
+
+For the compilation process of different development environments, please refer to the corresponding documents.
+
+1. [Docker](https://paddle-lite-pjc.readthedocs.io/zh/latest/user_guides/source_compile.html#docker)
+2. [Linux](https://paddle-lite-pjc.readthedocs.io/zh/latest/user_guides/source_compile.html#linux)
+3. [MAC OS](https://paddle-lite-pjc.readthedocs.io/zh/latest/user_guides/source_compile.html#mac-os)
+
+### 1.2 Prepare Paddle-Lite library
+
+There are two ways to obtain the Paddle-Lite library:
+
+1. [Recommended] Download directly, the download link of the Paddle-Lite library is as follows:
+
+      | Platform | Paddle-Lite library download link |
+      | --- | --- |
+      |Android|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.android.armv7.gcc.c++_shared.with_extra.with_cv.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.android.armv8.gcc.c++_shared.with_extra.with_cv.tar.gz)|
+      |IOS|[arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.ios.armv7.with_cv.with_extra.with_log.tiny_publish.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.ios.armv8.with_cv.with_extra.with_log.tiny_publish.tar.gz)|
+
+      Note: 1. The above Paddle-Lite library is compiled from the Paddle-Lite 2.10 branch. For more information about Paddle-Lite 2.10, please refer to [link](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.10).
+
+    **Note: It is recommended to use paddlelite>=2.10 version of the prediction library, other prediction library versions [download link](https://github.com/PaddlePaddle/Paddle-Lite/tags)**
+
+2. Compile Paddle-Lite to get the prediction library. The compilation method of Paddle-Lite is as follows:
+
+   ```bash linenums="1"
+   git clone https://github.com/PaddlePaddle/Paddle-Lite.git
+   cd Paddle-Lite
+   # Switch to Paddle-Lite release/v2.10 stable branch
+   git checkout release/v2.10
+   ./lite/tools/build_android.sh  --arch=armv8  --with_cv=ON --with_extra=ON
+   ```
+
+Note: When compiling Paddle-Lite to obtain the Paddle-Lite library, you need to turn on the two options `--with_cv=ON --with_extra=ON`, `--arch` means the `arm` version, here is designated as armv8,
+
+More compilation commands refer to the introduction [link](https://paddle-lite-pjc.readthedocs.io/zh/latest/user_guides/source_compile.html#paddlelite) 。
+
+After directly downloading the Paddle-Lite library and decompressing it, you can get the `inference_lite_lib.android.armv8/` folder, and the Paddle-Lite library obtained by compiling Paddle-Lite is located
+`Paddle-Lite/build.lite.android.armv8.gcc/inference_lite_lib.android.armv8/` folder.
+
+The structure of the prediction library is as follows:
+
+```text linenums="1"
+inference_lite_lib.android.armv8/
+|-- cxx                                        C++ prebuild library
+|   |-- include                                C++
+|   |   |-- paddle_api.h
+|   |   |-- paddle_image_preprocess.h
+|   |   |-- paddle_lite_factory_helper.h
+|   |   |-- paddle_place.h
+|   |   |-- paddle_use_kernels.h
+|   |   |-- paddle_use_ops.h
+|   |   `-- paddle_use_passes.h
+|   `-- lib                                           C++ library
+|       |-- libpaddle_api_light_bundled.a             C++ static library
+|       `-- libpaddle_light_api_shared.so             C++ dynamic library
+|-- java                                     Java library
+|   |-- jar
+|   |   `-- PaddlePredictor.jar
+|   |-- so
+|   |   `-- libpaddle_lite_jni.so
+|   `-- src
+|-- demo                                     C++ and Java demo
+|   |-- cxx                                  C++ demo
+|   `-- java                                 Java demo
+```
+
+## 2 Run
+
+### 2.1 Inference Model Optimization
+
+Paddle Lite provides a variety of strategies to automatically optimize the original training model, including quantization, sub-graph fusion, hybrid scheduling, Kernel optimization and so on. In order to make the optimization process more convenient and easy to use, Paddle Lite provide opt tools to automatically complete the optimization steps and output a lightweight, optimal executable model.
+
+If you have prepared the model file ending in .nb, you can skip this step.
+
+The following table also provides a series of models that can be deployed on mobile phones to recognize Chinese. You can directly download the optimized model.
+
+|Version|Introduction|Model size|Detection model|Text Direction model|Recognition model|Paddle-Lite branch|
+|---|---|---|---|---|---|---|
+|PP-OCRv3|extra-lightweight chinese OCR optimized model|16.2M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/PP-OCRv3_mobile_det_infer.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_infer_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/PP-OCRv3_mobile_rec_infer.nb)|v2.10|
+|PP-OCRv3(slim)|extra-lightweight chinese OCR optimized model|5.9M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_slim_infer.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_slim_infer.nb)|v2.10|
+
+If you directly use the model in the above table for deployment, you can skip the following steps and directly read [Section 2.2](#22-run-optimized-model-on-phone).
+
+If the model to be deployed is not in the above table, you need to follow the steps below to obtain the optimized model.
+
+Step 1: Refer to [document](https://www.paddlepaddle.org.cn/lite/v2.10/user_guides/opt/opt_python.html) to install paddlelite, which is used to convert paddle inference model to paddlelite required for running nb model
+
+```bash linenums="1"
+pip install paddlelite==2.10 # The paddlelite version should be the same as the prediction library version
+```
+
+After installation, the following commands can view the help information
+
+```bash linenums="1"
+paddle_lite_opt
+```
+
+Introduction to paddle_lite_opt parameters:
+
+|Options|Description|
+|---|---|
+|--model_dir|The path of the PaddlePaddle model to be optimized (non-combined form)|
+|--model_file|The network structure file path of the PaddlePaddle model (combined form) to be optimized|
+|--param_file|The weight file path of the PaddlePaddle model (combined form) to be optimized|
+|--optimize_out_type|Output model type, currently supports two types: protobuf and naive_buffer, among which naive_buffer is a more lightweight serialization/deserialization implementation. If you need to perform model prediction on the mobile side, please set this option to naive_buffer. The default is protobuf|
+|--optimize_out|The output path of the optimized model|
+|--valid_targets|The executable backend of the model, the default is arm. Currently it supports x86, arm, opencl, npu, xpu, multiple backends can be specified at the same time (separated by spaces), and Model Optimize Tool will automatically select the best method. If you need to support Huawei NPU (DaVinci architecture NPU equipped with Kirin 810/990 Soc), it should be set to npu, arm|
+|--record_tailoring_info|When using the function of cutting library files according to the model, set this option to true to record the kernel and OP information contained in the optimized model. The default is false|
+
+`--model_dir` is suitable for the non-combined mode of the model to be optimized, and the inference model of PaddleOCR is the combined mode, that is, the model structure and model parameters are stored in a single file.
+
+Step 2: Use paddle_lite_opt to convert the inference model to the mobile model format.
+
+The following takes the ultra-lightweight Chinese model of PaddleOCR as an example to introduce the use of the compiled opt file to complete the conversion of the inference model to the Paddle-Lite optimized model
+
+```bash linenums="1"
+# 【[Recommendation] Download the Chinese and English inference model of PP-OCRv3
+wget  https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_slim_infer.tar && tar xf  ch_PP-OCRv3_det_slim_infer.tar
+wget  https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_slim_infer.tar && tar xf  ch_PP-OCRv2_rec_slim_quant_infer.tar
+wget  https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_cls_slim_infer.tar && tar xf  ch_ppocr_mobile_v2.0_cls_slim_infer.tar
+# Convert detection model
+paddle_lite_opt --model_file=./ch_PP-OCRv3_det_slim_infer/inference.pdmodel  --param_file=./ch_PP-OCRv3_det_slim_infer/inference.pdiparams  --optimize_out=./ch_PP-OCRv3_det_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
+# Convert recognition model
+paddle_lite_opt --model_file=./ch_PP-OCRv3_rec_slim_infer/inference.pdmodel  --param_file=./ch_PP-OCRv3_rec_slim_infer/inference.pdiparams  --optimize_out=./ch_PP-OCRv3_rec_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
+# Convert angle classifier model
+paddle_lite_opt --model_file=./ch_ppocr_mobile_v2.0_cls_slim_infer/inference.pdmodel  --param_file=./ch_ppocr_mobile_v2.0_cls_slim_infer/inference.pdiparams  --optimize_out=./ch_ppocr_mobile_v2.0_cls_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
+```
+
+After the conversion is successful, there will be more files ending with `.nb` in the inference model directory, which is the successfully converted model file.
+
+### 2.2 Run optimized model on Phone
+
+Some preparatory work is required first.
+
+1. Prepare an Android phone with arm8. If the compiled prediction library and opt file are armv7, you need an arm7 phone and modify ARM_ABI = arm7 in the Makefile.
+
+2. Make sure the phone is connected to the computer, open the USB debugging option of the phone, and select the file transfer mode.
+3. Install the adb tool on the computer.
+
+    3.1. Install ADB for MAC:
+
+    ```bash linenums="1"
+    brew cask install android-platform-tools
+    ```
+
+    3.2. Install ADB for Linux
+
+    ```bash linenums="1"
+    sudo apt update
+    sudo apt install -y wget adb
+    ```
+
+    3.3. Install ADB for windows
+
+    To install on win, you need to go to Google's Android platform to download the adb package for installation:[link](https://developer.android.com/studio)
+
+    Verify whether adb is installed successfully
+
+    ```bash linenums="1"
+    adb devices
+    ```
+
+    If there is device output, it means the installation is successful。
+
+    ```bash linenums="1"
+    List of devices attached
+    744be294    device
+    ```
+
+4. Prepare optimized models, prediction library files, test images and dictionary files used.
+
+   ```bash linenums="1"
+   git clone https://github.com/PaddlePaddle/PaddleOCR.git
+   cd PaddleOCR/deploy/lite/
+   # run prepare.sh
+   sh prepare.sh /{lite prediction library path}/inference_lite_lib.android.armv8
+
+   #
+   cd /{lite prediction library path}/inference_lite_lib.android.armv8/
+   cd demo/cxx/ocr/
+   # copy paddle-lite C++ .so file to debug/ directory
+   cp ../../../cxx/lib/libpaddle_light_api_shared.so ./debug/
+
+   cd inference_lite_lib.android.armv8/demo/cxx/ocr/
+   cp ../../../cxx/lib/libpaddle_light_api_shared.so ./debug/
+   ```
+
+Prepare the test image, taking PaddleOCR/doc/imgs/11.jpg as an example, copy the image file to the demo/cxx/ocr/debug/ folder. Prepare the model files optimized by the lite opt tool, ch_PP-OCRv3_det_slim_opt.nb , ch_PP-OCRv3_rec_slim_opt.nb , and place them under the demo/cxx/ocr/debug/ folder.
+
+The structure of the OCR demo is as follows after the above command is executed:
+
+```text linenums="1"
+demo/cxx/ocr/
+|-- debug/
+|   |--ch_PP-OCRv3_det_slim_opt.nb           Detection model
+|   |--ch_PP-OCRv3_rec_slim_opt.nb           Recognition model
+|   |--ch_ppocr_mobile_v2.0_cls_slim_opt.nb           Text direction classification model
+|   |--11.jpg                           Image for OCR
+|   |--ppocr_keys_v1.txt                Dictionary file
+|   |--libpaddle_light_api_shared.so    C++ .so file
+|   |--config.txt                       Config file
+|-- config.txt                  Config file
+|-- cls_process.cc              Pre-processing and post-processing files for the angle classifier
+|-- cls_process.h
+|-- crnn_process.cc             Pre-processing and post-processing files for the CRNN model
+|-- crnn_process.h
+|-- db_post_process.cc          Pre-processing and post-processing files for the DB model
+|-- db_post_process.h
+|-- Makefile
+|-- ocr_db_crnn.cc              C++ main code
+```
+
+**Note**:
+
+1. `ppocr_keys_v1.txt` is a Chinese dictionary file. If the nb model is used for English recognition or other language recognition, dictionary file should be replaced with a dictionary of the corresponding language. PaddleOCR provides a variety of dictionaries under ppocr/utils/, including:
+
+   ```python linenums="1"
+   dict/french_dict.txt     # french
+   dict/german_dict.txt     # german
+   ic15_dict.txt       # english
+   dict/japan_dict.txt      # japan
+   dict/korean_dict.txt     # korean
+   ppocr_keys_v1.txt   # chinese
+   ```
+
+2. `config.txt` of the detector and classifier, as shown below:
+
+    ```python linenums="1"
+    max_side_len  960         #  Limit the maximum image height and width to  960
+    det_db_thresh  0.3        # Used to filter the binarized image of DB  prediction, setting 0.-0.3 has no obvious effect on the result
+    det_db_box_thresh  0.5    # DDB post-processing filter box threshold, if  there is a missing box detected, it can be reduced as appropriate
+    det_db_unclip_ratio  1.6  # Indicates the compactness of the text box,  the smaller the value, the closer the text box to the text
+    use_direction_classify  0  # Whether to use the direction classifier, 0  means not to use, 1 means to use
+    rec_image_height  48      # The height of the input image of the  recognition model, the PP-OCRv3 model needs to be set to 48, and the  PP-OCRv2 model needs to be set to 32
+    ```
+
+3. Run Model on phone
+
+   After the above steps are completed, you can use adb to push the file to the phone to run, the steps are as follows:
+
+   ```bash linenums="1"
+   # Execute the compilation and get the executable file ocr_db_crnn
+   # The first execution of this command will download dependent libraries such as opencv. After the download is complete, you need to execute it again
+   make -j
+   # Move the compiled executable file to the debug folder
+   mv ocr_db_crnn ./debug/
+   # Push the debug folder to the phone
+   adb push debug /data/local/tmp/
+   adb shell
+   cd /data/local/tmp/debug
+   export LD_LIBRARY_PATH=${PWD}:$LD_LIBRARY_PATH
+   # The use of ocr_db_crnn is:
+   # ./ocr_db_crnn Mode Detection model file Orientation classifier model file Recognition model file  Hardware  Precision  Threads Batchsize  Test image path Dictionary file path
+   ./ocr_db_crnn system ch_PP-OCRv3_det_slim_opt.nb  ch_PP-OCRv3_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_slim_opt.nb  arm8 INT8 10 1  ./11.jpg  config.txt  ppocr_keys_v1.txt  True
+   # precision can be INT8 for quantitative model or FP32 for normal model.
+
+   # Only using detection model
+   ./ocr_db_crnn  det ch_PP-OCRv3_det_slim_opt.nb arm8 INT8 10 1 ./11.jpg  config.txt
+
+   # Only using recognition model
+   ./ocr_db_crnn  rec ch_PP-OCRv3_rec_slim_opt.nb arm8 INT8 10 1 word_1.jpg ppocr_keys_v1.txt config.txt
+   ```
+
+If you modify the code, you need to recompile and push to the phone.
+
+The outputs are as follows:
+
+![img](./images/lite_demo.png)
+
+## FAQ
+
+Q1: What if I want to change the model, do I need to run it again according to the process?
+
+A1: If you have performed the above steps, you only need to replace the .nb model file to complete the model replacement.
+
+Q2: How to test with another picture?
+
+A2: Replace the .jpg test image under ./debug with the image you want to test, and run adb push to push new image to the phone.
+
+Q3: How to package it into the mobile APP?
+
+A3: This demo aims to provide the core algorithm part that can run OCR on mobile phones. Further, PaddleOCR/deploy/android_demo is an example of encapsulating this demo into a mobile app for reference.
+
+Q4: When running the demo, an error is reported `Error: This model is not supported, because kernel for 'io_copy' is not supported by Paddle-Lite.`
+
+A4: The problem is that the installed paddlelite version does not match the downloaded prediction library version. Make sure that the paddleliteopt tool matches your prediction library version, and try to switch to the nb model again.
diff --git a/docs/version2.x/legacy/lite.md b/docs/version2.x/legacy/lite.md
new file mode 100644
index 0000000..14c0801
--- /dev/null
+++ b/docs/version2.x/legacy/lite.md
@@ -0,0 +1,309 @@
+---
+typora-copy-images-to: images
+comments: true
+hide:
+  - toc
+---
+
+# 端侧部署
+
+本教程将介绍基于[Paddle Lite](https://github.com/PaddlePaddle/Paddle-Lite) 在移动端部署PaddleOCR超轻量中文检测、识别模型的详细步骤。
+
+Paddle Lite是飞桨轻量化推理引擎，为手机、IOT端提供高效推理能力，并广泛整合跨平台硬件，为端侧部署及应用落地问题提供轻量化的部署方案。
+
+## 1. 准备环境
+
+### 运行准备
+
+- 电脑（编译Paddle Lite）
+- 安卓手机（armv7或armv8）
+
+### 1.1 准备交叉编译环境
+
+交叉编译环境用于编译 Paddle Lite 和 PaddleOCR 的C++ demo。
+支持多种开发环境，不同开发环境的编译流程请参考对应文档。
+
+1. [Docker](https://paddle-lite-pjc.readthedocs.io/zh/latest/user_guides/source_compile.html#docker)
+2. [Linux](https://paddle-lite-pjc.readthedocs.io/zh/latest/user_guides/source_compile.html#linux)
+3. [MAC OS](https://paddle-lite-pjc.readthedocs.io/zh/latest/user_guides/source_compile.html#mac-os)
+
+### 1.2 准备预测库
+
+预测库有两种获取方式：
+
+1. [推荐]直接下载，预测库下载链接如下：
+
+   | 平台 | 预测库下载链接 |
+   | ---- | ---- |
+   | Android | [arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.android.armv7.gcc.c++_shared.with_extra.with_cv.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.android.armv8.gcc.c++_shared.with_extra.with_cv.tar.gz)       |
+   | IOS     | [arm7](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.ios.armv7.with_cv.with_extra.with_log.tiny_publish.tar.gz) / [arm8](https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10/inference_lite_lib.ios.armv8.with_cv.with_extra.with_log.tiny_publish.tar.gz) |
+
+   注：1. 上述预测库为PaddleLite 2.10分支编译得到，有关PaddleLite 2.10 详细信息可参考 [链接](https://github.com/PaddlePaddle/Paddle-Lite/releases/tag/v2.10) 。
+
+   **注：建议使用paddlelite>=2.10版本的预测库，其他预测库版本[下载链接](https://github.com/PaddlePaddle/Paddle-Lite/tags)**
+
+2. 编译Paddle-Lite得到预测库，Paddle-Lite的编译方式如下：
+
+   ```bash linenums="1"
+   git clone https://github.com/PaddlePaddle/Paddle-Lite.git
+   cd Paddle-Lite
+   # 切换到Paddle-Lite release/v2.10 稳定分支
+   git checkout release/v2.10
+   ./lite/tools/build_android.sh  --arch=armv8  --with_cv=ON --with_extra=ON
+   ```
+
+注意：编译Paddle-Lite获得预测库时，需要打开`--with_cv=ON --with_extra=ON`两个选项，`--arch`表示`arm`版本，这里指定为armv8，
+更多编译命令
+介绍请参考 [链接](https://paddle-lite-pjc.readthedocs.io/zh/latest/user_guides/source_compile.html#paddlelite) 。
+
+直接下载预测库并解压后，可以得到`inference_lite_lib.android.armv8/`文件夹，通过编译Paddle-Lite得到的预测库位于
+`Paddle-Lite/build.lite.android.armv8.gcc/inference_lite_lib.android.armv8/`文件夹下。
+预测库的文件目录如下：
+
+```text linenums="1"
+inference_lite_lib.android.armv8/
+|-- cxx                                        C++ 预测库和头文件
+|   |-- include                                C++ 头文件
+|   |   |-- paddle_api.h
+|   |   |-- paddle_image_preprocess.h
+|   |   |-- paddle_lite_factory_helper.h
+|   |   |-- paddle_place.h
+|   |   |-- paddle_use_kernels.h
+|   |   |-- paddle_use_ops.h
+|   |   `-- paddle_use_passes.h
+|   `-- lib                                           C++预测库
+|       |-- libpaddle_api_light_bundled.a             C++静态库
+|       `-- libpaddle_light_api_shared.so             C++动态库
+|-- java                                     Java预测库
+|   |-- jar
+|   |   `-- PaddlePredictor.jar
+|   |-- so
+|   |   `-- libpaddle_lite_jni.so
+|   `-- src
+|-- demo                                     C++和Java示例代码
+|   |-- cxx                                  C++  预测库demo
+|   `-- java                                 Java 预测库demo
+```
+
+## 2 开始运行
+
+### 2.1 模型优化
+
+Paddle-Lite 提供了多种策略来自动优化原始的模型，其中包括量化、子图融合、混合调度、Kernel优选等方法，使用Paddle-lite的opt工具可以自动
+对inference模型进行优化，优化后的模型更轻量，模型运行速度更快。
+
+如果已经准备好了 `.nb` 结尾的模型文件，可以跳过此步骤。
+
+下述表格中也提供了一系列中文移动端模型：
+
+| 模型版本       | 模型简介                      | 模型大小 | 检测模型                                                                                   | 文本方向分类模型                                                                                | 识别模型                                                                                   | Paddle-Lite版本 |
+| -------------- | ----------------------------- | -------- | ------------------------------------------------------------------------------------------ | ----------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------ | --------------- |
+| PP-OCRv3       | 蒸馏版超轻量中文OCR移动端模型 | 16.2M    | [下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/PP-OCRv3_mobile_det_infer.nb)      | [下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_infer_opt.nb) | [下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_infer.nb)      | v2.10           |
+| PP-OCRv3(slim) | 蒸馏版超轻量中文OCR移动端模型 | 5.9M     | [下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_slim_infer.nb) | [下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)  | [下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_slim_infer.nb) | v2.10           |
+
+
+如果直接使用上述表格中的模型进行部署，可略过下述步骤，直接阅读 [2.2节](#22)。
+
+如果要部署的模型不在上述表格中，则需要按照如下步骤获得优化后的模型。
+
+步骤1：参考[文档](https://www.paddlepaddle.org.cn/lite/v2.10/user_guides/opt/opt_python.html)安装paddlelite，用于转换paddle inference model为paddlelite运行所需的nb模型
+
+```bash linenums="1"
+pip install paddlelite==2.10  # paddlelite版本要与预测库版本一致
+```
+
+安装完后，如下指令可以查看帮助信息
+
+```bash linenums="1"
+paddle_lite_opt
+```
+
+paddle_lite_opt 参数介绍：
+
+| 选项   | 说明  |
+| ---| --- |
+| --model_dir             | 待优化的PaddlePaddle模型（非combined形式）的路径                                                                                                                                                                                        |
+| --model_file            | 待优化的PaddlePaddle模型（combined形式）的网络结构文件路径                                                                                                                                                                              |
+| --param_file            | 待优化的PaddlePaddle模型（combined形式）的权重文件路径                                                                                                                                                                                  |
+| --optimize_out_type     | 输出模型类型，目前支持两种类型：protobuf和naive_buffer，其中naive_buffer是一种更轻量级的序列化/反序列化实现。若您需要在mobile端执行模型预测，请将此选项设置为naive_buffer。默认为protobuf                                               |
+| --optimize_out          | 优化模型的输出路径                                                                                                                                                                                                                      |
+| --valid_targets         | 指定模型可执行的backend，默认为arm。目前可支持x86、arm、opencl、npu、xpu，可以同时指定多个backend(以空格分隔)，Model Optimize Tool将会自动选择最佳方式。如果需要支持华为NPU（Kirin 810/990 Soc搭载的达芬奇架构NPU），应当设置为npu, arm |
+| --record_tailoring_info | 当使用 根据模型裁剪库文件 功能时，则设置该选项为true，以记录优化后模型含有的kernel和OP信息，默认为false                                                                                                                                 |
+
+`--model_dir`适用于待优化的模型是非combined方式，PaddleOCR的inference模型是combined方式，即模型结构和模型参数使用单独一个文件存储。
+
+步骤2：使用paddle_lite_opt将inference模型转换成移动端模型格式。
+
+下面以PaddleOCR的超轻量中文模型为例，介绍使用编译好的opt文件完成inference模型到Paddle-Lite优化模型的转换。
+
+```bash linenums="1"
+# 【推荐】 下载 PP-OCRv3版本的中英文 inference模型
+wget  https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_slim_infer.tar && tar xf  ch_PP-OCRv3_det_slim_infer.tar
+wget  https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_slim_infer.tar && tar xf  ch_PP-OCRv2_rec_slim_quant_infer.tar
+wget  https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_cls_slim_infer.tar && tar xf  ch_ppocr_mobile_v2.0_cls_slim_infer.tar
+# 转换检测模型
+paddle_lite_opt --model_file=./ch_PP-OCRv3_det_slim_infer/inference.pdmodel  --param_file=./ch_PP-OCRv3_det_slim_infer/inference.pdiparams  --optimize_out=./ch_PP-OCRv3_det_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
+# 转换识别模型
+paddle_lite_opt --model_file=./ch_PP-OCRv3_rec_slim_infer/inference.pdmodel  --param_file=./ch_PP-OCRv3_rec_slim_infer/inference.pdiparams  --optimize_out=./ch_PP-OCRv3_rec_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
+# 转换方向分类器模型
+paddle_lite_opt --model_file=./ch_ppocr_mobile_v2.0_cls_slim_infer/inference.pdmodel  --param_file=./ch_ppocr_mobile_v2.0_cls_slim_infer/inference.pdiparams  --optimize_out=./ch_ppocr_mobile_v2.0_cls_slim_opt --valid_targets=arm  --optimize_out_type=naive_buffer
+```
+
+转换成功后，inference模型目录下会多出`.nb`结尾的文件，即是转换成功的模型文件。
+
+注意：使用paddle-lite部署时，需要使用opt工具优化后的模型。 opt工具的输入模型是paddle保存的inference模型
+
+### 2.2 与手机联调
+
+首先需要进行一些准备工作：
+
+1. 准备一台arm8的安卓手机，如果编译的预测库和opt文件是armv7，则需要arm7的手机，并修改Makefile中`ARM_ABI = arm7`。
+
+2. 打开手机的USB调试选项，选择文件传输模式，连接电脑。
+
+3. 电脑上安装adb工具，用于调试。 adb安装方式如下：
+
+   3.1. MAC电脑安装ADB:
+
+   ```bash linenums="1"
+   brew cask install android-platform-tools
+   ```
+
+   3.2. Linux安装ADB
+
+   ```bash linenums="1"
+   sudo apt update
+   sudo apt install -y wget adb
+   ```
+
+   3.3. Window安装ADB
+   win上安装需要去谷歌的安卓平台下载adb软件包进行安装：[链接](https://developer.android.com/studio)
+
+   打开终端，手机连接电脑，在终端中输入
+
+   ```bash linenums="1"
+   adb devices
+   ```
+
+   如果有device输出，则表示安装成功。
+
+   ```bash linenums="1"
+   List of devices attached
+   744be294    device
+   ```
+
+4. 准备优化后的模型、预测库文件、测试图像和使用的字典文件。
+
+   ```bash linenums="1"
+   git clone https://github.com/PaddlePaddle/PaddleOCR.git
+   cd PaddleOCR/deploy/lite/
+   # 运行prepare.sh，准备预测库文件、测试图像和使用的字典文件，并放置在预测库中的demo/cxx/ocr文件夹下
+   sh prepare.sh /{lite prediction library path}/inference_lite_lib.android.armv8
+
+   # 进入OCR demo的工作目录
+   cd /{lite prediction library path}/inference_lite_lib.android.armv8/
+   cd demo/cxx/ocr/
+   # 将C++预测动态库so文件复制到debug文件夹中
+   cp ../../../cxx/lib/libpaddle_light_api_shared.so ./debug/
+   ```
+
+ 准备测试图像，以`PaddleOCR/doc/imgs/11.jpg`为例，将测试的图像复制到`demo/cxx/ocr/debug/`文件夹下。
+ 准备lite opt工具优化后的模型文件，比如使用`ch_PP-OCRv3_det_slim_opt.ch_PP-OCRv3_rec_slim_rec.nb, ch_ppocr_mobile_v2.0_cls_slim_opt.nb`，模型文件放置在`demo/cxx/ocr/debug/`文件夹下。
+
+ 执行完成后，ocr文件夹下将有如下文件格式：
+
+   ```text linenums="1"
+   demo/cxx/ocr/
+   |-- debug/
+   |   |--ch_PP-OCRv3_det_slim_opt.nb           优化后的检测模型文件
+   |   |--ch_PP-OCRv3_rec_slim_opt.nb           优化后的识别模型文件
+   |   |--ch_ppocr_mobile_v2.0_cls_slim_opt.nb           优化后的文字方向分类器模型文件
+   |   |--11.jpg                           待测试图像
+   |   |--ppocr_keys_v1.txt                中文字典文件
+   |   |--libpaddle_light_api_shared.so    C++预测库文件
+   |   |--config.txt                       超参数配置
+   |-- config.txt                  超参数配置
+   |-- cls_process.cc              方向分类器的预处理和后处理文件
+   |-- cls_process.h
+   |-- crnn_process.cc             识别模型CRNN的预处理和后处理文件
+   |-- crnn_process.h
+   |-- db_post_process.cc          检测模型DB的后处理文件
+   |-- db_post_process.h
+   |-- Makefile                    编译文件
+   |-- ocr_db_crnn.cc              C++预测源文件
+   ```
+
+#### 注意
+
+1. ppocr_keys_v1.txt是中文字典文件，如果使用的 nb 模型是英文数字或其他语言的模型，需要更换为对应语言的字典。PaddleOCR 在ppocr/utils/下存放了多种字典，包括：
+
+   ```text linenums="1"
+   dict/french_dict.txt     # 法语字典
+   dict/german_dict.txt     # 德语字典
+   ic15_dict.txt       # 英文字典
+   dict/japan_dict.txt      # 日语字典
+   dict/korean_dict.txt     # 韩语字典
+   ppocr_keys_v1.txt   # 中文字典
+   ...
+   ```
+
+2. `config.txt` 包含了检测器、分类器、识别器的超参数，如下：
+
+    ```python linenums="1"
+    max_side_len  960         # 输入图像长宽大于960时，等比例缩放图像，使得图像最长边 为960
+    det_db_thresh  0.3        # 用于过滤DB预测的二值化图像，设置为0.-0.3对结果影响不 明显
+    det_db_box_thresh  0.5    # 检测器后处理过滤box的阈值，如果检测存在漏框情况，可酌 情减小
+    det_db_unclip_ratio  1.6  # 表示文本框的紧致程度，越小则文本框更靠近文本
+    use_direction_classify  0  # 是否使用方向分类器，0表示不使用，1表示使用
+    rec_image_height  48      # 识别模型输入图像的高度，PP-OCRv3模型设置为48， PP-OCRv2模型需要设置为32
+    ```
+
+3. 启动调试
+
+上述步骤完成后就可以使用adb将文件push到手机上运行，步骤如下：
+
+```bash linenums="1"
+# 执行编译，得到可执行文件ocr_db_crnn, 第一次执行此命令会下载opencv等依赖库，下载完成后，需要再执行一次
+make -j
+
+# 将编译的可执行文件移动到debug文件夹中
+mv ocr_db_crnn ./debug/
+# 将debug文件夹push到手机上
+adb push debug /data/local/tmp/
+adb shell
+cd /data/local/tmp/debug
+export LD_LIBRARY_PATH=${PWD}:$LD_LIBRARY_PATH
+# 开始使用，ocr_db_crnn可执行文件的使用方式为:
+# ./ocr_db_crnn 预测模式  检测模型文件 方向分类器模型文件  识别模型文件 运行硬件 运行精度 线程数  batchsize  测试图像路径  参数配置路径  字典文件路径 是否使用benchmark参数
+./ocr_db_crnn system  ch_PP-OCRv3_det_slim_opt.nb  ch_PP-OCRv3_rec_slim_opt.nb  ch_ppocr_mobile_v2.0_cls_slim_opt.nb  arm8 INT8 10 1  ./11.jpg  config.txt  ppocr_keys_v1.txt  True
+
+# 仅使用文本检测模型，使用方式如下：
+./ocr_db_crnn  det ch_PP-OCRv3_det_slim_opt.nb arm8 INT8 10 1 ./11.jpg  config.txt
+
+# 仅使用文本识别模型，使用方式如下：
+./ocr_db_crnn  rec ch_PP-OCRv3_rec_slim_opt.nb arm8 INT8 10 1 word_1.jpg ppocr_keys_v1.txt config.txt
+```
+
+ 如果对代码做了修改，则需要重新编译并push到手机上。
+
+ 运行效果如下：
+
+![img](./images/lite_demo.png)
+
+## FAQ
+
+Q1：如果想更换模型怎么办，需要重新按照流程走一遍吗？
+
+A1：如果已经走通了上述步骤，更换模型只需要替换 .nb 模型文件即可，同时要注意更新字典
+
+Q2：换一个图测试怎么做？
+
+A2：替换debug下的.jpg测试图像为你想要测试的图像，adb push 到手机上即可
+
+Q3：如何封装到手机APP中？
+
+A3：此demo旨在提供能在手机上运行OCR的核心算法部分，PaddleOCR/deploy/android_demo是将这个demo封装到手机app的示例，供参考
+
+Q4：运行demo时遇到报错`Error: This model is not supported, because kernel for 'io_copy' is not supported by Paddle-Lite.`
+
+A4：问题是安装的paddlelite版本和下载的预测库版本不匹配，确保paddleliteopt工具和你的预测库版本匹配，重新转nb模型试试。
diff --git a/docs/version2.x/legacy/model_list_2.x.en.md b/docs/version2.x/legacy/model_list_2.x.en.md
new file mode 100644
index 0000000..7c45587
--- /dev/null
+++ b/docs/version2.x/legacy/model_list_2.x.en.md
@@ -0,0 +1,106 @@
+# Models Supported by PaddleOCR 2.x and Earlier Versions
+
+Due to differences in the concatenation logic and configurations used during model training and inference, the PP-OCRv4 and PP-OCRv3 series models from the PaddleOCR 2.x branch cannot be used interchangeably with those from the PaddleOCR 3.0 and later branches.
+
+## Detection Models
+
+### Chinese Detection Models
+
+* ch_PP-OCRv4_det
+* ch_PP-OCRv4_server_det
+* ch_PP-OCRv3_det_slim
+* ch_PP-OCRv3_det
+* ch_PP-OCRv2_det_slim
+* ch_PP-OCRv2_det
+* ch_ppocr_mobile_slim_v2.0_det
+* ch_ppocr_mobile_v2.0_det
+* ch_ppocr_server_v2.0_det
+
+### English Detection Models
+
+* en_PP-OCRv3_det_slim
+* en_PP-OCRv3_det
+
+### Multilingual Detection Models
+
+* ml_PP-OCRv3_det_slim
+* ml_PP-OCRv3_det
+
+## Recognition Models
+
+### Chinese Recognition Models
+
+* ch_PP-OCRv4_rec
+* ch_PP-OCRv4_server_rec
+* ch_PP-OCRv4_server_rec_doc
+* ch_PP-OCRv3_rec_slim
+* ch_PP-OCRv3_rec
+* ch_PP-OCRv2_rec_slim
+* ch_PP-OCRv2_rec
+* ch_ppocr_mobile_slim_v2.0_rec
+* ch_ppocr_mobile_v2.0_rec
+* ch_ppocr_server_v2.0_rec
+* SVTRv2(Rec Sever)
+* RepSVTR(Mobile)
+
+### English Recognition Models
+
+* en_PP-OCRv4_rec
+* en_PP-OCRv3_rec_slim
+* en_PP-OCRv3_rec
+* en_number_mobile_slim_v2.0_rec
+* en_number_mobile_v2.0_rec
+
+### Multilingual Recognition Models
+
+* korean_PP-OCRv3_rec
+* japan_PP-OCRv3_rec
+* chinese_cht_PP-OCRv3_rec
+* te_PP-OCRv3_rec
+* ka_PP-OCRv3_rec
+* ta_PP-OCRv3_rec
+* latin_PP-OCRv3_rec
+* arabic_PP-OCRv3_rec
+* cyrillic_PP-OCRv3_rec
+* devanagari_PP-OCRv3_rec
+
+## End-to-End OCR Models
+
+* PGNet
+
+## Text Direction Classification Models
+
+* ch_ppocr_mobile_slim_v2.0_cls
+* ch_ppocr_mobile_v2.0_cls
+
+## Formula Recognition Models
+
+* CAN
+* UniMERNet
+* LaTeX-OCR
+* PP-FormulaNet-S
+* PP-FormulaNet-L
+
+## Table Recognition Models
+
+* TableMaster
+* SLANet
+* SLANeXt_wired
+* SLANeXt_wireless
+* en_ppocr_mobile_v2.0_table_structure
+* en_ppstructure_mobile_v2.0_SLANet
+* ch_ppstructure_mobile_v2.0_SLANet
+
+## Table OCR Models
+
+* en_ppocr_mobile_v2.0_table_det
+* en_ppocr_mobile_v2.0_table_rec
+
+## Layout Detection Models
+
+* picodet_lcnet_x1_0_fgd_layout
+* ppyolov2_r50vd_dcn_365e_publaynet
+* picodet_lcnet_x1_0_fgd_layout_cdla
+* picodet_lcnet_x1_0_fgd_layout_table
+* ppyolov2_r50vd_dcn_365e_tableBank_word
+* ppyolov2_r50vd_dcn_365e_tableBank_latex
diff --git a/docs/version2.x/legacy/model_list_2.x.md b/docs/version2.x/legacy/model_list_2.x.md
new file mode 100644
index 0000000..0324f10
--- /dev/null
+++ b/docs/version2.x/legacy/model_list_2.x.md
@@ -0,0 +1,106 @@
+# PaddleOCR 2.x及更低版本支持的模型
+
+由于串联逻辑和模型训练、推理时用到的配置不同，PaddleOCR 2.x 分支的 PP-OCRv4、PP-OCRv3 系列模型与 PaddleOCR 3.0 及以上分支的 PP-OCRv4、PP-OCRv3 系列模型不能互相使用。
+
+## 检测模型
+
+### 中文检测模型
+
+* ch_PP-OCRv4_det
+* ch_PP-OCRv4_server_det
+* ch_PP-OCRv3_det_slim
+* ch_PP-OCRv3_det
+* ch_PP-OCRv2_det_slim
+* ch_PP-OCRv2_det
+* ch_ppocr_mobile_slim_v2.0_det
+* ch_ppocr_mobile_v2.0_det
+* ch_ppocr_server_v2.0_det
+
+### 英文检测模型
+
+* en_PP-OCRv3_det_slim
+* en_PP-OCRv3_det
+
+### 多语言检测模型
+
+* ml_PP-OCRv3_det_slim
+* ml_PP-OCRv3_det
+
+## 识别模型
+
+### 中文识别模型
+
+* ch_PP-OCRv4_rec
+* ch_PP-OCRv4_server_rec
+* ch_PP-OCRv4_server_rec_doc
+* ch_PP-OCRv3_rec_slim
+* ch_PP-OCRv3_rec
+* ch_PP-OCRv2_rec_slim
+* ch_PP-OCRv2_rec
+* ch_ppocr_mobile_slim_v2.0_rec
+* ch_ppocr_mobile_v2.0_rec
+* ch_ppocr_server_v2.0_rec
+* SVTRv2(Rec Sever)
+* RepSVTR(Mobile)
+
+### 英文识别模型
+
+* en_PP-OCRv4_rec
+* en_PP-OCRv3_rec_slim
+* en_PP-OCRv3_rec
+* en_number_mobile_slim_v2.0_rec
+* en_number_mobile_v2.0_rec
+
+### 多语言识别模型
+
+* korean_PP-OCRv3_rec
+* japan_PP-OCRv3_rec
+* chinese_cht_PP-OCRv3_rec
+* te_PP-OCRv3_rec
+* ka_PP-OCRv3_rec
+* ta_PP-OCRv3_rec
+* latin_PP-OCRv3_rec
+* arabic_PP-OCRv3_rec
+* cyrillic_PP-OCRv3_rec
+* devanagari_PP-OCRv3_rec
+
+## 端到端OCR模型
+
+* PGNet
+
+## 文本方向分类模型
+
+* ch_ppocr_mobile_slim_v2.0_cls
+* ch_ppocr_mobile_v2.0_cls
+
+## 公式识别模型
+
+* CAN
+* UniMERNet
+* LaTeX-OCR
+* PP-FormulaNet-S
+* PP-FormulaNet-L
+
+## 表格识别模型
+
+* TableMaster
+* SLANet
+* SLANeXt_wired
+* SLANeXt_wireless
+* en_ppocr_mobile_v2.0_table_structure
+* en_ppstructure_mobile_v2.0_SLANet
+* ch_ppstructure_mobile_v2.0_SLANet
+
+## 表格OCR模型
+
+* en_ppocr_mobile_v2.0_table_det
+* en_ppocr_mobile_v2.0_table_rec
+
+## 版面检测模型
+
+* picodet_lcnet_x1_0_fgd_layout
+* ppyolov2_r50vd_dcn_365e_publaynet
+* picodet_lcnet_x1_0_fgd_layout_cdla
+* picodet_lcnet_x1_0_fgd_layout_table
+* ppyolov2_r50vd_dcn_365e_tableBank_word
+* ppyolov2_r50vd_dcn_365e_tableBank_latex
diff --git a/docs/version2.x/legacy/paddle2onnx.en.md b/docs/version2.x/legacy/paddle2onnx.en.md
new file mode 100644
index 0000000..c2f5812
--- /dev/null
+++ b/docs/version2.x/legacy/paddle2onnx.en.md
@@ -0,0 +1,267 @@
+---
+typora-copy-images-to: images
+comments: true
+hide:
+  - toc
+---
+
+# Paddle2ONNX Model Conversion and Prediction
+
+This chapter introduces how to convert PaddleOCR models to ONNX models and perform predictions based on the ONNXRuntime engine. We also prepared an [online AI Studio project](https://aistudio.baidu.com/projectdetail/8808858) for your convenience.
+
+## 1. Environment Setup
+
+You need to prepare the environments for PaddleOCR, Paddle2ONNX model conversion, and ONNXRuntime prediction.
+
+### PaddleOCR
+
+Clone the PaddleOCR repository, use the main branch, and install it. Since the PaddleOCR repository is relatively large and cloning via `git clone` can be slow, this tutorial has already downloaded it.
+
+```bash linenums="1"
+git clone -b main https://github.com/PaddlePaddle/PaddleOCR.git
+cd PaddleOCR && python3 -m pip install -e .
+```
+
+### Paddle2ONNX
+
+Paddle2ONNX supports converting models in the PaddlePaddle format to the ONNX format. Operators currently stably support exporting ONNX Opset versions 7~19, and some Paddle operators support conversion to lower ONNX Opsets. For more details, please refer to [Paddle2ONNX](https://github.com/PaddlePaddle/Paddle2ONNX.git).
+
+- Install Paddle2ONNX
+
+  ```bash linenums="1"
+  python3 -m pip install paddle2onnx
+  ```
+
+- Install ONNXRuntime
+
+  ```bash linenums="1"
+  python3 -m pip install onnxruntime
+  ```
+
+## 2. Model Conversion
+
+### Get Paddle Static Graph Models
+
+There are two ways to obtain Paddle static graph models: download the prediction models provided by PaddleOCR in the [model list](../ppocr/model_list.en.md); or refer to the [Model Export Instructions](./python_infer.en.md) to convert trained weights into inference models.
+
+Using the PP-OCR series English detection, recognition, and classification models as examples:
+
+#### Download Provided Inference Models
+
+=== "PP-OCRv3"
+
+    ```bash linenums="1"
+    wget -nc -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar
+    cd ./inference && tar xf en_PP-OCRv3_det_infer.tar && cd ..
+
+    wget -nc  -P ./inference https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv3_mobile_rec_infer.tar
+    cd ./inference && tar xf en_PP-OCRv3_mobile_rec_infer.tar && cd ..
+
+    wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar
+    cd ./inference && tar xf ch_ppocr_mobile_v2.0_cls_infer.tar && cd ..
+    ```
+
+=== "PP-OCRv4"
+
+    ```bash linenums="1"
+    wget -nc -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar
+    cd ./inference && tar xf en_PP-OCRv3_det_infer.tar && cd ..
+
+    wget -nc  -P ./inference https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv4_mobile_rec_infer.tar
+    cd ./inference && tar xf en_PP-OCRv4_mobile_rec_infer.tar && cd ..
+
+    wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar
+    cd ./inference && tar xf ch_ppocr_mobile_v2.0_cls_infer.tar && cd ..
+    ```
+
+#### Convert Dynamic Graph to Static Graph (Optional)
+
+Download dynamic graph models:
+
+```bash linenums="1"
+wget -nc -P pretrained https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams
+
+wget -nc -P pretrained https://paddleocr.bj.bcebos.com/PP-OCRv4/chinese/ch_PP-OCRv4_rec_train.tar
+cd pretrained && tar xf ch_PP-OCRv4_rec_train.tar && cd ..
+
+wget -nc -P pretrained https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar
+cd pretrained && tar xf ch_ppocr_mobile_v2.0_cls_train.tar && cd ..
+```
+
+Convert to static graph models:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/PP-OCRv4/PP-OCRv4_mobile_det.yml \
+-o Global.pretrained_model=./pretrained/PP-OCRv4_mobile_det_pretrained \
+Global.save_inference_dir=./inference/PP-OCRv4_mobile_det_infer/
+
+python3 tools/export_model.py -c configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec.yml \
+-o Global.pretrained_model=./pretrained/ch_PP-OCRv4_rec_train/student \
+Global.save_inference_dir=./inference/PP-OCRv4_mobile_rec_infer/
+
+python3 tools/export_model.py -c configs/cls/cls_mv3.yml \
+-o Global.pretrained_model=./pretrained/ch_ppocr_mobile_v2.0_cls_train/best_accuracy \
+Global.save_inference_dir=./inference/ch_ppocr_mobile_v2.0_cls_infer/
+```
+
+### Model Conversion
+
+Use Paddle2ONNX to convert Paddle static graph models to the ONNX model format:
+
+=== "PP-OCRv3"
+
+    ```bash linenums="1"
+    paddle2onnx --model_dir ./inference/en_PP-OCRv3_det_infer \
+    --model_filename inference.pdmodel \
+    --params_filename inference.pdiparams \
+    --save_file ./inference/det_onnx/model.onnx \
+    --opset_version 11 \
+    --enable_onnx_checker True
+
+    paddle2onnx --model_dir ./inference/en_PP-OCRv3_mobile_rec_infer \
+    --model_filename inference.pdmodel \
+    --params_filename inference.pdiparams \
+    --save_file ./inference/rec_onnx/model.onnx \
+    --opset_version 11 \
+    --enable_onnx_checker True
+
+    paddle2onnx --model_dir ./inference/ch_ppocr_mobile_v2.0_cls_infer \
+    --model_filename inference.pdmodel \
+    --params_filename inference.pdiparams \
+    --save_file ./inference/cls_onnx/model.onnx \
+    --opset_version 11 \
+    --enable_onnx_checker True
+    ```
+
+=== "PP-OCRv4"
+
+    ```bash linenums="1"
+    paddle2onnx --model_dir ./inference/en_PP-OCRv3_det_infer \
+    --model_filename inference.pdmodel \
+    --params_filename inference.pdiparams \
+    --save_file ./inference/det_onnx/model.onnx \
+    --opset_version 11 \
+    --enable_onnx_checker True
+
+    paddle2onnx --model_dir ./inference/en_PP-OCRv4_mobile_rec_infer \
+    --model_filename inference.pdmodel \
+    --params_filename inference.pdiparams \
+    --save_file ./inference/rec_onnx/model.onnx \
+    --opset_version 11 \
+    --enable_onnx_checker True
+
+    paddle2onnx --model_dir ./inference/ch_ppocr_mobile_v2.0_cls_infer \
+    --model_filename inference.pdmodel \
+    --params_filename inference.pdiparams \
+    --save_file ./inference/cls_onnx/model.onnx \
+    --opset_version 11 \
+    --enable_onnx_checker True
+    ```
+
+After execution, the ONNX models will be saved respectively under `./inference/det_onnx/`, `./inference/rec_onnx/`, and `./inference/cls_onnx/`.
+
+- **Note**: For OCR models, dynamic shapes must be used during conversion; otherwise, the prediction results may slightly differ from directly using Paddle for prediction. Additionally, the following models currently do not support conversion to ONNX models: NRTR, SAR, RARE, SRN.
+
+- **Note**: After [Paddle2ONNX version v1.2.3](https://github.com/PaddlePaddle/Paddle2ONNX/releases/tag/v1.2.3), dynamic shapes are supported by default, i.e., `float32[p2o.DynamicDimension.0,3,p2o.DynamicDimension.1,p2o.DynamicDimension.2]`. The option `--input_shape_dict` has been deprecated. If you need to adjust shapes, you can use the following command to adjust the input shape of the Paddle model.
+
+  ```bash linenums="1"
+  python3 -m paddle2onnx.optimize --input_model inference/det_onnx/model.onnx \
+    --output_model inference/det_onnx/model.onnx \
+    --input_shape_dict "{'x': [-1,3,-1,-1]}"
+  ```
+
+If you have optimization requirements for the exported ONNX model, it is recommended to use [OnnxSlim](https://github.com/inisis/OnnxSlim) to optimize the model:
+
+```bash linenums="1"
+pip install onnxslim
+onnxslim model.onnx slim.onnx
+```
+
+## 3. Inference and Prediction
+
+Using the Chinese OCR model as an example, you can perform prediction using ONNXRuntime by executing the following command:
+
+```bash linenums="1"
+python3 tools/infer/predict_system.py --use_gpu=False --use_onnx=True \
+--det_model_dir=./inference/det_onnx/model.onnx  \
+--rec_model_dir=./inference/rec_onnx/model.onnx  \
+--cls_model_dir=./inference/cls_onnx/model.onnx  \
+--image_dir=./docs/infer_deploy/images/img_12.jpg \
+--rec_char_dict_path=ppocr/utils/en_dict.txt
+```
+
+Taking the English OCR model as an example, you can perform prediction using Paddle Inference by executing the following command:
+
+=== "PP-OCRv3"
+
+    ```bash linenums="1"
+    python3 tools/infer/predict_system.py --use_gpu=False \
+    --cls_model_dir=./inference/ch_ppocr_mobile_v2.0_cls_infer \
+    --rec_model_dir=./inference/en_PP-OCRv3_mobile_rec_infer \
+    --det_model_dir=./inference/en_PP-OCRv3_det_infer \
+    --image_dir=./docs/infer_deploy/images/img_12.jpg\
+    --rec_char_dict_path=ppocr/utils/en_dict.txt
+    ```
+
+=== "PP-OCRv4"
+
+    ```bash linenums="1"
+    python3 tools/infer/predict_system.py --use_gpu=False \
+    --cls_model_dir=./inference/ch_ppocr_mobile_v2.0_cls_infer \
+    --rec_model_dir=./inference/en_PP-OCRv4_mobile_rec_infer \
+    --det_model_dir=./inference/en_PP-OCRv3_det_infer \
+    --image_dir=./docs/infer_deploy/images/img_12.jpg \
+    --rec_char_dict_path=ppocr/utils/en_dict.txt
+    ```
+
+After executing the command, the terminal will print out the predicted recognition information, and the visualization results will be saved under `./inference_results/`.
+
+**ONNXRuntime Execution Result:**
+
+![](./images/img_12_result.jpg)
+
+**Paddle Inference Execution Result:**
+
+![](./images/img_12_result.jpg)
+
+Using ONNXRuntime for prediction, terminal output:
+
+```bash linenums="1"
+[2022/10/10 12:06:28] ppocr DEBUG: dt_boxes num : 11, elapse : 0.3568880558013916
+[2022/10/10 12:06:31] ppocr DEBUG: rec_res num  : 11, elapse : 2.6445000171661377
+[2022/10/10 12:06:31] ppocr DEBUG: 0  Predict time of doc/imgs_en/img_12.jpg: 3.021s
+[2022/10/10 12:06:31] ppocr DEBUG: ACKNOWLEDGEMENTS, 0.997
+[2022/10/10 12:06:31] ppocr DEBUG: We would like to thank all the designers and, 0.976
+[2022/10/10 12:06:31] ppocr DEBUG: contributors who have been involved in the, 0.979
+[2022/10/10 12:06:31] ppocr DEBUG: production of this book; their contributions, 0.989
+[2022/10/10 12:06:31] ppocr DEBUG: have been indispensable to its creation. We, 0.956
+[2022/10/10 12:06:31] ppocr DEBUG: would also like to express our gratitude to all, 0.991
+[2022/10/10 12:06:31] ppocr DEBUG: the producers for their invaluable opinions, 0.978
+[2022/10/10 12:06:31] ppocr DEBUG: and assistance throughout this project. And to, 0.988
+[2022/10/10 12:06:31] ppocr DEBUG: the many others whose names are not credited, 0.958
+[2022/10/10 12:06:31] ppocr DEBUG: but have made specific input in this book, we, 0.970
+[2022/10/10 12:06:31] ppocr DEBUG: thank you for your continuous support., 0.998
+[2022/10/10 12:06:31] ppocr DEBUG: The visualized image saved in ./inference_results/img_12.jpg
+[2022/10/10 12:06:31] ppocr INFO: The predict total time is 3.2482550144195557
+```
+
+Using Paddle Inference for prediction, terminal output:
+
+```bash linenums="1"
+[2022/10/10 12:06:28] ppocr DEBUG: dt_boxes num : 11, elapse : 0.3568880558013916
+[2022/10/10 12:06:31] ppocr DEBUG: rec_res num  : 11, elapse : 2.6445000171661377
+[2022/10/10 12:06:31] ppocr DEBUG: 0  Predict time of doc/imgs_en/img_12.jpg: 3.021s
+[2022/10/10 12:06:31] ppocr DEBUG: ACKNOWLEDGEMENTS, 0.997
+[2022/10/10 12:06:31] ppocr DEBUG: We would like to thank all the designers and, 0.976
+[2022/10/10 12:06:31] ppocr DEBUG: contributors who have been involved in the, 0.979
+[2022/10/10 12:06:31] ppocr DEBUG: production of this book; their contributions, 0.989
+[2022/10/10 12:06:31] ppocr DEBUG: have been indispensable to its creation. We, 0.956
+[2022/10/10 12:06:31] ppocr DEBUG: would also like to express our gratitude to all, 0.991
+[2022/10/10 12:06:31] ppocr DEBUG: the producers for their invaluable opinions, 0.978
+[2022/10/10 12:06:31] ppocr DEBUG: and assistance throughout this project. And to, 0.988
+[2022/10/10 12:06:31] ppocr DEBUG: the many others whose names are not credited, 0.958
+[2022/10/10 12:06:31] ppocr DEBUG: but have made specific input in this book, we, 0.970
+[2022/10/10 12:06:31] ppocr DEBUG: thank you for your continuous support., 0.998
+[2022/10/10 12:06:31] ppocr DEBUG: The visualized image saved in ./inference_results/img_12.jpg
+[2022/10/10 12:06:31] ppocr INFO: The predict total time is 3.2482550144195557
+```
diff --git a/docs/version2.x/legacy/paddle2onnx.md b/docs/version2.x/legacy/paddle2onnx.md
new file mode 100644
index 0000000..23ee15a
--- /dev/null
+++ b/docs/version2.x/legacy/paddle2onnx.md
@@ -0,0 +1,315 @@
+---
+typora-copy-images-to: images
+comments: true
+hide:
+  - toc
+---
+
+# Paddle2ONNX模型转化与预测
+
+本章节介绍 PaddleOCR 模型如何转化为 ONNX 模型，并基于 ONNXRuntime 引擎预测。同时我们准备了一个[在线 AI Studio 项目](https://aistudio.baidu.com/projectdetail/8808858)，可以方便用户进行测试。
+
+## 1. 环境准备
+
+需要准备 PaddleOCR、Paddle2ONNX 模型转化环境，和 ONNXRuntime 预测环境。
+
+### PaddleOCR
+
+克隆PaddleOCR的仓库，使用 main 分支，并进行安装，由于 PaddleOCR 仓库比较大，git clone 速度比较慢，所以本教程已下载。
+
+```bash linenums="1"
+git clone  -b main https://github.com/PaddlePaddle/PaddleOCR.git
+cd PaddleOCR && python3 -m pip install -e .
+```
+
+### Paddle2ONNX
+
+Paddle2ONNX 支持将 PaddlePaddle 模型格式转化到 ONNX 模型格式，算子目前稳定支持导出 ONNX Opset 7~19，部分Paddle算子支持更低的ONNX Opset转换。
+更多细节可参考 [Paddle2ONNX](https://github.com/PaddlePaddle/Paddle2ONNX.git)
+
+- 安装 Paddle2ONNX
+
+  ```bash linenums="1"
+  python3 -m pip install paddle2onnx
+  ```
+
+- 安装 ONNXRuntime
+
+  ```bash linenums="1"
+  python3 -m pip install onnxruntime
+  ```
+
+## 2. 模型转换
+
+### 获取 Paddle 静态图模型
+
+有两种方式获取Paddle静态图模型：在 [model_list](../ppocr/model_list.md) 中下载PaddleOCR提供的预测模型；参考[模型导出说明](./python_infer.md#inference)把训练好的权重转为推理模型。
+
+#### 下载准备好的静态图模型
+
+以 PP-OCR 系列中文检测、识别、分类模型为例：
+
+=== "PP-OCRv3"
+
+    ```bash linenums="1"
+    wget -nc  -P ./inference https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar
+    cd ./inference && tar xf PP-OCRv3_mobile_det_infer.tar && cd ..
+
+    wget -nc  -P ./inference https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar
+    cd ./inference && tar xf PP-OCRv3_mobile_rec_infer.tar && cd ..
+
+    wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar
+    cd ./inference && tar xf ch_ppocr_mobile_v2.0_cls_infer.tar && cd ..
+    ```
+
+=== "PP-OCRv4"
+
+    ```bash linenums="1"
+    wget -nc  -P ./inference https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar
+    cd ./inference && tar xf PP-OCRv4_mobile_det_infer.tar && cd ..
+
+    wget -nc  -P ./inference https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar
+    cd ./inference && tar xf PP-OCRv4_mobile_rec_infer.tar && cd ..
+
+    wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar
+    cd ./inference && tar xf ch_ppocr_mobile_v2.0_cls_infer.tar && cd ..
+    ```
+
+#### 将动态图导出为静态图模型 (可选)
+
+下载动态图模型：
+
+```bash linenums="1"
+wget -nc -P pretrained https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams
+
+wget -nc -P pretrained https://paddleocr.bj.bcebos.com/PP-OCRv4/chinese/ch_PP-OCRv4_rec_train.tar
+cd pretrained && tar xf ch_PP-OCRv4_rec_train.tar && cd ..
+
+wget -nc -P pretrained https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar
+cd pretrained && tar xf ch_ppocr_mobile_v2.0_cls_train.tar && cd ..
+```
+
+转换为静态图模型：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/PP-OCRv4/PP-OCRv4_mobile_det.yml \
+-o Global.pretrained_model=./pretrained/PP-OCRv4_mobile_det_pretrained \
+Global.save_inference_dir=./inference/PP-OCRv4_mobile_det_infer/
+
+python3 tools/export_model.py -c configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec.yml \
+-o Global.pretrained_model=./pretrained/ch_PP-OCRv4_rec_train/student \
+Global.save_inference_dir=./inference/PP-OCRv4_mobile_rec_infer/
+
+python3 tools/export_model.py -c configs/cls/cls_mv3.yml \
+-o Global.pretrained_model=./pretrained/ch_ppocr_mobile_v2.0_cls_train/best_accuracy \
+Global.save_inference_dir=./inference/ch_ppocr_mobile_v2.0_cls_infer/
+```
+
+### 模型转换
+
+使用 Paddle2ONNX 将Paddle静态图模型转换为ONNX模型格式：
+
+=== "PP-OCRv3"
+
+    ```bash linenums="1"
+    paddle2onnx --model_dir ./inference/PP-OCRv3_mobile_det_infer \
+    --model_filename inference.pdmodel \
+    --params_filename inference.pdiparams \
+    --save_file ./inference/det_onnx/model.onnx \
+    --opset_version 11 \
+    --enable_onnx_checker True
+
+    paddle2onnx --model_dir ./inference/PP-OCRv3_mobile_rec_infer \
+    --model_filename inference.pdmodel \
+    --params_filename inference.pdiparams \
+    --save_file ./inference/rec_onnx/model.onnx \
+    --opset_version 11 \
+    --enable_onnx_checker True
+
+    paddle2onnx --model_dir ./inference/ch_ppocr_mobile_v2.0_cls_infer \
+    --model_filename inference.pdmodel \
+    --params_filename inference.pdiparams \
+    --save_file ./inference/cls_onnx/model.onnx \
+    --opset_version 11 \
+    --enable_onnx_checker True
+    ```
+
+=== "PP-OCRv4"
+
+    ```bash linenums="1"
+    paddle2onnx --model_dir ./inference/PP-OCRv4_mobile_det_infer \
+    --model_filename inference.pdmodel \
+    --params_filename inference.pdiparams \
+    --save_file ./inference/det_onnx/model.onnx \
+    --opset_version 11 \
+    --enable_onnx_checker True
+
+    paddle2onnx --model_dir ./inference/PP-OCRv4_mobile_rec_infer \
+    --model_filename inference.pdmodel \
+    --params_filename inference.pdiparams \
+    --save_file ./inference/rec_onnx/model.onnx \
+    --opset_version 11 \
+    --enable_onnx_checker True
+
+    paddle2onnx --model_dir ./inference/ch_ppocr_mobile_v2.0_cls_infer \
+    --model_filename inference.pdmodel \
+    --params_filename inference.pdiparams \
+    --save_file ./inference/cls_onnx/model.onnx \
+    --opset_version 11 \
+    --enable_onnx_checker True
+    ```
+
+执行完毕后，ONNX 模型会被分别保存在 `./inference/det_onnx/`，`./inference/rec_onnx/`，`./inference/cls_onnx/`路径下
+
+- 注意：对于OCR模型，转化过程中必须采用动态shape的形式，否则预测结果可能与直接使用Paddle预测有细微不同。
+  另外，以下几个模型暂不支持转换为 ONNX 模型：
+  NRTR、SAR、RARE、SRN
+
+- 注意：[Paddle2ONNX 版本 v1.2.3](https://github.com/PaddlePaddle/Paddle2ONNX/releases/tag/v1.2.3)后已默认支持动态shape，即 `float32[p2o.DynamicDimension.0,3,p2o.DynamicDimension.1,p2o.DynamicDimension.2]`，选项 `--input_shape_dict` 已废弃。如果有shape调整需求可使用如下命令进行Paddle模型输入shape调整。
+
+  ```bash linenums="1"
+  python3 -m paddle2onnx.optimize --input_model inference/det_onnx/model.onnx \
+    --output_model inference/det_onnx/model.onnx \
+    --input_shape_dict "{'x': [-1,3,-1,-1]}"
+  ```
+
+如你对导出的 ONNX 模型有优化的需求，推荐使用 [onnxslim](https://github.com/inisis/OnnxSlim) 对模型进行优化:
+
+```bash linenums="1"
+pip install onnxslim
+onnxslim model.onnx slim.onnx
+```
+
+## 3. 推理预测
+
+以中文OCR模型为例，使用 ONNXRuntime 预测可执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_system.py --use_gpu=False --use_onnx=True \
+--det_model_dir=./inference/det_onnx/model.onnx  \
+--rec_model_dir=./inference/rec_onnx/model.onnx  \
+--cls_model_dir=./inference/cls_onnx/model.onnx  \
+--image_dir=./docs/infer_deploy/images/lite_demo.png
+```
+
+以中文OCR模型为例，使用 Paddle Inference 预测可执行如下命令：
+
+=== "PP-OCRv3"
+
+    ```bash linenums="1"
+    python3 tools/infer/predict_system.py --use_gpu=False \
+    --cls_model_dir=./inference/ch_ppocr_mobile_v2.0_cls_infer \
+    --rec_model_dir=./inference/PP-OCRv3_mobile_rec_infer \
+    --det_model_dir=./inference/PP-OCRv3_mobile_det_infer \
+    --image_dir=./docs/infer_deploy/images/lite_demo.png
+    ```
+
+=== "PP-OCRv4"
+
+    ```bash linenums="1"
+    python3 tools/infer/predict_system.py --use_gpu=False \
+    --cls_model_dir=./inference/ch_ppocr_mobile_v2.0_cls_infer \
+    --rec_model_dir=./inference/PP-OCRv4_mobile_rec_infer \
+    --det_model_dir=./inference/PP-OCRv4_mobile_det_infer \
+    --image_dir=./docs/infer_deploy/images/lite_demo.png
+    ```
+
+执行命令后在终端会打印出预测的识别信息，并在 `./inference_results/` 下保存可视化结果。
+
+ONNXRuntime 执行效果：
+
+![](./images/lite_demo_onnx.png)
+
+Paddle Inference 执行效果：
+
+![](./images/lite_demo_paddle.png)
+
+使用 ONNXRuntime 预测，终端输出：
+
+```bash linenums="1"
+[2022/02/22 17:48:27] root DEBUG: dt_boxes num : 38, elapse : 0.043187856674194336
+[2022/02/22 17:48:27] root DEBUG: rec_res num  : 38, elapse : 0.592170000076294
+[2022/02/22 17:48:27] root DEBUG: 0  Predict time of ./deploy/lite/imgs/lite_demo.png: 0.642s
+[2022/02/22 17:48:27] root DEBUG: The, 0.984
+[2022/02/22 17:48:27] root DEBUG: visualized, 0.882
+[2022/02/22 17:48:27] root DEBUG: etect18片, 0.720
+[2022/02/22 17:48:27] root DEBUG: image saved in./vis.jpg, 0.947
+[2022/02/22 17:48:27] root DEBUG: 纯臻营养护发素0.993604, 0.996
+[2022/02/22 17:48:27] root DEBUG: 产品信息/参数, 0.922
+[2022/02/22 17:48:27] root DEBUG: 0.992728, 0.914
+[2022/02/22 17:48:27] root DEBUG: （45元／每公斤，100公斤起订）, 0.926
+[2022/02/22 17:48:27] root DEBUG: 0.97417, 0.977
+[2022/02/22 17:48:27] root DEBUG: 每瓶22元，1000瓶起订）0.993976, 0.962
+[2022/02/22 17:48:27] root DEBUG: 【品牌】：代加工方式/0EMODM, 0.945
+[2022/02/22 17:48:27] root DEBUG: 0.985133, 0.980
+[2022/02/22 17:48:27] root DEBUG: 【品名】：纯臻营养护发素, 0.921
+[2022/02/22 17:48:27] root DEBUG: 0.995007, 0.883
+[2022/02/22 17:48:27] root DEBUG: 【产品编号】：YM-X-30110.96899, 0.955
+[2022/02/22 17:48:27] root DEBUG: 【净含量】：220ml, 0.943
+[2022/02/22 17:48:27] root DEBUG: Q.996577, 0.932
+[2022/02/22 17:48:27] root DEBUG: 【适用人群】：适合所有肤质, 0.913
+[2022/02/22 17:48:27] root DEBUG: 0.995842, 0.969
+[2022/02/22 17:48:27] root DEBUG: 【主要成分】：鲸蜡硬脂醇、燕麦B-葡聚, 0.883
+[2022/02/22 17:48:27] root DEBUG: 0.961928, 0.964
+[2022/02/22 17:48:27] root DEBUG: 10, 0.812
+[2022/02/22 17:48:27] root DEBUG: 糖、椰油酰胺丙基甜菜碱、泛醒, 0.866
+[2022/02/22 17:48:27] root DEBUG: 0.925898, 0.943
+[2022/02/22 17:48:27] root DEBUG: （成品包材）, 0.974
+[2022/02/22 17:48:27] root DEBUG: 0.972573, 0.961
+[2022/02/22 17:48:27] root DEBUG: 【主要功能】：可紧致头发磷层，从而达到, 0.936
+[2022/02/22 17:48:27] root DEBUG: 0.994448, 0.952
+[2022/02/22 17:48:27] root DEBUG: 13, 0.998
+[2022/02/22 17:48:27] root DEBUG: 即时持久改善头发光泽的效果，给干燥的头, 0.994
+[2022/02/22 17:48:27] root DEBUG: 0.990198, 0.975
+[2022/02/22 17:48:27] root DEBUG: 14, 0.977
+[2022/02/22 17:48:27] root DEBUG: 发足够的滋养, 0.991
+[2022/02/22 17:48:27] root DEBUG: 0.997668, 0.918
+[2022/02/22 17:48:27] root DEBUG: 花费了0.457335秒, 0.901
+[2022/02/22 17:48:27] root DEBUG: The visualized image saved in ./inference_results/lite_demo.png
+[2022/02/22 17:48:27] root INFO: The predict total time is 0.7003889083862305
+```
+
+使用 Paddle Inference 预测，终端输出：
+
+```bash linenums="1"
+[2022/02/22 17:47:25] root DEBUG: dt_boxes num : 38, elapse : 0.11791276931762695
+[2022/02/22 17:47:27] root DEBUG: rec_res num  : 38, elapse : 2.6206860542297363
+[2022/02/22 17:47:27] root DEBUG: 0  Predict time of ./deploy/lite/imgs/lite_demo.png: 2.746s
+[2022/02/22 17:47:27] root DEBUG: The, 0.984
+[2022/02/22 17:47:27] root DEBUG: visualized, 0.882
+[2022/02/22 17:47:27] root DEBUG: etect18片, 0.720
+[2022/02/22 17:47:27] root DEBUG: image saved in./vis.jpg, 0.947
+[2022/02/22 17:47:27] root DEBUG: 纯臻营养护发素0.993604, 0.996
+[2022/02/22 17:47:27] root DEBUG: 产品信息/参数, 0.922
+[2022/02/22 17:47:27] root DEBUG: 0.992728, 0.914
+[2022/02/22 17:47:27] root DEBUG: （45元／每公斤，100公斤起订）, 0.926
+[2022/02/22 17:47:27] root DEBUG: 0.97417, 0.977
+[2022/02/22 17:47:27] root DEBUG: 每瓶22元，1000瓶起订）0.993976, 0.962
+[2022/02/22 17:47:27] root DEBUG: 【品牌】：代加工方式/0EMODM, 0.945
+[2022/02/22 17:47:27] root DEBUG: 0.985133, 0.980
+[2022/02/22 17:47:27] root DEBUG: 【品名】：纯臻营养护发素, 0.921
+[2022/02/22 17:47:27] root DEBUG: 0.995007, 0.883
+[2022/02/22 17:47:27] root DEBUG: 【产品编号】：YM-X-30110.96899, 0.955
+[2022/02/22 17:47:27] root DEBUG: 【净含量】：220ml, 0.943
+[2022/02/22 17:47:27] root DEBUG: Q.996577, 0.932
+[2022/02/22 17:47:27] root DEBUG: 【适用人群】：适合所有肤质, 0.913
+[2022/02/22 17:47:27] root DEBUG: 0.995842, 0.969
+[2022/02/22 17:47:27] root DEBUG: 【主要成分】：鲸蜡硬脂醇、燕麦B-葡聚, 0.883
+[2022/02/22 17:47:27] root DEBUG: 0.961928, 0.964
+[2022/02/22 17:47:27] root DEBUG: 10, 0.812
+[2022/02/22 17:47:27] root DEBUG: 糖、椰油酰胺丙基甜菜碱、泛醒, 0.866
+[2022/02/22 17:47:27] root DEBUG: 0.925898, 0.943
+[2022/02/22 17:47:27] root DEBUG: （成品包材）, 0.974
+[2022/02/22 17:47:27] root DEBUG: 0.972573, 0.961
+[2022/02/22 17:47:27] root DEBUG: 【主要功能】：可紧致头发磷层，从而达到, 0.936
+[2022/02/22 17:47:27] root DEBUG: 0.994448, 0.952
+[2022/02/22 17:47:27] root DEBUG: 13, 0.998
+[2022/02/22 17:47:27] root DEBUG: 即时持久改善头发光泽的效果，给干燥的头, 0.994
+[2022/02/22 17:47:27] root DEBUG: 0.990198, 0.975
+[2022/02/22 17:47:27] root DEBUG: 14, 0.977
+[2022/02/22 17:47:27] root DEBUG: 发足够的滋养, 0.991
+[2022/02/22 17:47:27] root DEBUG: 0.997668, 0.918
+[2022/02/22 17:47:27] root DEBUG: 花费了0.457335秒, 0.901
+[2022/02/22 17:47:27] root DEBUG: The visualized image saved in ./inference_results/lite_demo.png
+[2022/02/22 17:47:27] root INFO: The predict total time is 2.8338775634765625
+```
diff --git a/docs/version2.x/legacy/paddle_cloud.md b/docs/version2.x/legacy/paddle_cloud.md
new file mode 100644
index 0000000..165ec3d
--- /dev/null
+++ b/docs/version2.x/legacy/paddle_cloud.md
@@ -0,0 +1,343 @@
+---
+comments: true
+hide:
+  - toc
+---
+
+# 云上飞桨部署工具
+
+[云上飞桨（PaddleCloud）](https://github.com/PaddlePaddle/PaddleCloud) 是面向飞桨框架及其模型套件的部署工具，
+为用户提供了模型套件Docker化部署和Kubernetes集群部署两种方式，可以满足不同场景与环境的部署需求。
+本章节我们将使用PaddleCloud提供的OCR标准镜像以及云原生组件来训练和部署PP-OCRv3识别模型。
+
+## 云上飞桨部署工具的优势
+
+<div align="center">
+  <img src="../../../deploy/paddlecloud/images/paddlecloud.png" title="architecture" width="80%" height="80%" alt="">
+</div>
+
+- **模型套件Docker镜像大礼包。**
+
+  PaddleCloud为用户提供了飞桨模型套件Docker镜像大礼包，这些镜像中包含运行模型套件案例的所有依赖并能持续更新，支持异构硬件环境和常见CUDA版本、开箱即用。
+
+- **具有丰富的云上飞桨组件。**
+
+  云上飞桨具有丰富的云原生功能组件，包括样本数据缓存组件、分布式训练组件、推理服务组件等，使用这些组件用户可以快速地在Kubernetes集群上进行训练和部署工作。
+
+- **功能强大的自运维能力。**
+
+  云上飞桨组件基于Kubernetes的Operator机制提供了功能强大的自运维能力，如训练组件支持多种架构模式并具有分布式容错与弹性训练的能力，推理服务组件支持自动扩缩容与蓝绿发版等。
+
+- **针对飞桨框架的定制优化。**
+
+  除了部署便捷与自运维的优势，PaddleCloud还针对飞桨框架进行了正对性优化，如通过缓存样本数据来加速云上飞桨分布式训练作业、基于飞桨框架和调度器的协同设计来优化集群GPU利用率等。
+
+## 1. PP-OCRv3 Docker化部署
+
+PaddleCloud基于 [Tekton](https://github.com/tektoncd/pipeline) 为OCR模型套件提供了镜像持续构建的能力，并支持CPU、GPU以及常见CUDA版本的镜像。
+您可以查看 PaddleOCR 镜像仓库 来获取所有的镜像列表。
+同时我们也将PP-OCRv3识别模型的训练与推理实战案例放置到了AI Studio平台上，您可以点击 [PP-OCRv3识别训推一体项目实战](https://aistudio.baidu.com/aistudio/projectdetail/3916206?channelType=0&channel=0) 在平台上快速体验。
+
+> **适用场景**：本地测试开发环境、单机部署环境。
+
+### 1.1 安装Docker
+
+如果您所使用的机器上还没有安装 Docker，您可以参考 [Docker 官方文档](https://docs.docker.com/get-docker/) 来进行安装。
+如果您需要使用支持 GPU 版本的镜像，则还需安装好NVIDIA相关驱动和 [nvidia-docker](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html#docker) 。
+
+**注意**：如果您使用的是Windows系统，需要开启 [WSL2（Linux子系统功能）功能](https://docs.microsoft.com/en-us/windows/wsl/install)。
+
+### 1.2 启动容器
+
+**使用CPU版本的Docker镜像**
+
+```bash linenums="1"
+# 这是加上参数 --shm-size=32g 是为了防止容器里内存不足
+docker run --name ppocr -v $PWD:/mnt -p 8888:8888 -it --shm-size=32g paddlecloud/paddleocr:2.5-cpu-efbb0a /bin/bash
+```
+
+**使用GPU版本的Docker镜像**
+
+```bash linenums="1"
+docker run --name ppocr --runtime=nvidia -v $PWD:/mnt -p 8888:8888 -it --shm-size=32g paddlecloud/paddleocr:2.5-gpu-cuda10.2-cudnn7-efbb0a /bin/bash
+```
+
+进入容器内，则可进行 PP-OCRv3 模型的训练和部署工作。
+
+### 1.3 准备训练数据
+
+本教程以HierText数据集为例，HierText是第一个具有自然场景和文档中文本分层注释的数据集。
+该数据集包含从 Open Images 数据集中选择的 11639 张图像，提供高质量的单词 (~1.2M)、行和段落级别的注释。
+我们已经将数据集上传到百度云对象存储（BOS），您可以通过运行如下指令，完成数据集的下载和解压操作：
+
+```bash linenums="1"
+# 下载数据集
+$ wget -P /mnt https://paddleflow-public.hkg.bcebos.com/ppocr/hiertext1.tar
+
+# 解压数据集
+$ tar xf /mnt/hiertext1.tar -C /mnt && mv /mnt/hiertext1 /mnt/hiertext
+```
+
+运行上述命令后，在 `/mnt` 目录下包含以下文件：
+
+```
+/mnt/hiertext
+  └─ train/     HierText训练集数据
+  └─ validation/     HierText验证集数据
+  └─ label_hiertext_train.txt  HierText训练集的行标注
+  └─ label_hiertext_val.txt    HierText验证集的行标注
+```
+
+### 1.4 修改配置文件
+
+PP-OCRv3模型配置文件位于`/home/PaddleOCR/configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml`，需要修改的配置如下：
+
+- 修改训练数据配置：
+
+```yaml linenums="1"
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+```
+
+修改为：
+
+```yaml linenums="1"
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: /mnt/
+    label_file_list:
+      - /mnt/hiertext/label_hiertext_train.txt
+```
+
+- 修改验证数据配置：
+
+```yaml linenums="1"
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+```
+
+修改为：
+
+```yaml linenums="1"
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: /mnt/
+    label_file_list:
+      - /mnt/hiertext/label_hiertext_val.txt
+```
+
+### 1.5 启动训练
+
+下载PP-OCRv3的蒸馏预训练模型并进行训练的方式如下
+
+```bash linenums="1"
+# 下载预训练模型到/home/PaddleOCR/pre_train文件夹下
+$ mkdir /home/PaddleOCR/pre_train
+
+$ wget -P /home/PaddleOCR/pre_train https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams
+
+```
+
+启动训练，训练模型默认保存在`output`目录下，加载PP-OCRv3检测预训练模型。
+
+```bash linenums="1"
+# 这里以 GPU 训练为例，使用 CPU 进行训练的话，需要指定参数 Global.use_gpu=false
+python3 tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml -o Global.save_model_dir=./output/ Global.pretrained_model=./pre_train/ch_PP-OCRv3_det_distill_train/best_accuracy
+```
+
+如果要使用多GPU分布式训练，请使用如下命令：
+
+```bash linenums="1"
+# 启动训练，训练模型默认保存在output目录下，--gpus '0,1,2,3'表示使用0，1，2，3号GPU训练
+python3 -m paddle.distributed.launch --log_dir=./debug/ --gpus '0,1,2,3' tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml -o Global.save_model_dir=./output/ Global.pretrained_model=./pre_train/ch_PP-OCRv3_det_distill_train/best_accuracy
+```
+
+### 1.6 模型评估
+
+训练过程中保存的模型在output目录下，包含以下文件：
+
+```
+best_accuracy.states
+best_accuracy.pdparams  # 默认保存最优精度的模型参数
+best_accuracy.pdopt     # 默认保存最优精度的优化器相关参数
+latest.states
+latest.pdparams  # 默认保存的最新模型参数
+latest.pdopt     # 默认保存的最新模型的优化器相关参数
+```
+
+其中，best_accuracy是保存的最优模型，可以直接使用该模型评估
+
+```bash linenums="1"
+# 进行模型评估
+cd /home/PaddleOCR/
+
+python3 tools/eval.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml -o Global.checkpoints=./output/best_accuracy
+```
+
+## 2. PP-OCRv3云端部署
+
+PaddleCloud基于Kubernetes的Operator机制为您提供了多个功能强大的云原生组件，如样本数据缓存组件、分布式训练组件、 以及模型推理服务组件，
+使用这些组件您可以快速地在云上进行分布式训练和模型服务化部署。更多关于PaddleCloud云原生组件的内容，请参考文档 [PaddleCloud架构概览](https://github.com/PaddlePaddle/PaddleCloud/blob/main/docs/zh_CN/paddlecloud-overview.md) 。
+
+> **适用场景**：基于Kubernetes的多机部署环境。
+
+### 2.1 安装云上飞桨组件
+
+#### 环境要求
+
+- [Kubernetes v1.16+](https://kubernetes.io/zh/)
+- [kubectl](https://kubernetes.io/docs/tasks/tools/)
+- [Helm](https://helm.sh/zh/docs/intro/install/)
+
+如果您没有Kubernetes环境，可以使用MicroK8S在本地搭建环境，更多详情请参考 [MicroK8S官方文档](https://microk8s.io/docs/getting-started)。
+
+使用Helm一键安装所有组件和所有依赖
+
+```bash linenums="1"
+# 添加PaddleCloud Chart仓库
+$ helm repo add paddlecloud https://paddleflow-public.hkg.bcebos.com/charts
+$ helm repo update
+
+# 安装云上飞桨组件
+$ helm install pdc paddlecloud/paddlecloud --set tags.all-dep=true --namespace paddlecloud --create-namespace
+
+# 检查所有云上飞桨组件是否成功启动，命名空间下的所有Pod都为Running状态则安装成功。
+$ kubectl get pods -n paddlecloud
+NAME                                                 READY   STATUS    RESTARTS   AGE
+pdc-hostpath-5b6bd6787d-bxvxg                        1/1     Running   0          10h
+juicefs-csi-node-pkldt                               3/3     Running   0          10h
+juicefs-csi-controller-0                             3/3     Running   0          10h
+pdc-paddlecloud-sampleset-767bdf6947-pb6zm           1/1     Running   0          10h
+pdc-paddlecloud-paddlejob-7cc8b7bfc6-7gqnh           1/1     Running   0          10h
+pdc-minio-7cc967669d-824q5                           1/1     Running   0          10h
+pdc-redis-master-0                                   1/1     Running   0          10h
+```
+
+更多安装参数请参考[PaddleCloud安装指南](https://github.com/PaddlePaddle/PaddleCloud/blob/main/docs/zh_CN/installation.md)
+
+### 2.2 云原生组件介绍
+
+<div align="center">
+  <img src="../../../deploy/paddlecloud/images/architecture.jpeg" title="architecture" width="60%" height="60%" alt="">
+</div>
+
+- **数据缓存组件。** 数据缓存组件使用JuiceFS作为缓存引擎，能够将远程样本数据缓存到训练集群本地，大幅加速云上飞桨分布式训练作业。
+- **分布式训练组件。** 分布式训练组件支持参数服务器（PS）与集合通信（Collective）两种架构模式，方便用户在云上快速运行飞桨分布式训练作业。
+
+以下内容我们将使用这两个云原生组件来在Kubernetes集群中部署PP-OCRv3识别模型的训练作业。
+
+### 2.3 准备hiertext数据集
+
+使用数据缓存组件来准备数据集，编写SampleSet Yaml文件如下：
+
+```yaml linenums="1"
+# hiertext.yaml
+apiVersion: batch.paddlepaddle.org/v1alpha1
+kind: SampleSet
+metadata:
+  name: hiertext
+  namespace: paddlecloud
+spec:
+  partitions: 1
+  source:
+    uri: bos://paddleflow-public.hkg.bcebos.com/ppocr/hiertext
+    secretRef:
+      name: none
+  secretRef:
+    name: data-center
+```
+
+然后在命令行中，使用kubectl执行如下命令。
+
+```bash linenums="1"
+# 创建hiertext数据集
+$ kubectl apply -f hiertext.yaml
+sampleset.batch.paddlepaddle.org/hiertext created
+
+# 查看数据集的状态
+$ kubectl get sampleset hiertext -n paddlecloud
+NAME       TOTAL SIZE   CACHED SIZE   AVAIL SPACE   RUNTIME   PHASE   AGE
+hiertext   3.3 GiB       3.2 GiB      12 GiB        1/1       Ready   11m
+```
+
+### 2.4 训练PP-OCRv3模型
+
+使用训练组件在Kubernetes集群上训练PP-OCRv3模型，编写PaddleJob Yaml文件如下：
+
+```yaml linenums="1"
+# ppocrv3.yaml
+apiVersion: batch.paddlepaddle.org/v1
+kind: PaddleJob
+metadata:
+  name: ppocrv3
+  namespace: paddlecloud
+spec:
+  cleanPodPolicy: OnCompletion
+  sampleSetRef:
+    name: hiertext
+    namespace: paddlecloud
+    mountPath: /mnt/hiertext
+  worker:
+    replicas: 1
+    template:
+      spec:
+        containers:
+          - name: ppocrv3
+            image: paddlecloud/paddleocr:2.5-gpu-cuda10.2-cudnn7-efbb0a
+            command:
+              - /bin/bash
+            args:
+              - "-c"
+              - >
+                mkdir /home/PaddleOCR/pre_train &&
+                wget -P ./pre_train https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar &&
+                tar xf ./pre_train/ch_PP-OCRv3_det_distill_train.tar -C ./pre_train/ &&
+                python tools/train.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o
+                Train.dataset.data_dir=/mnt/
+                Train.dataset.label_file_list=[\"/mnt/hiertext/label_hiertext_train.txt\"]
+                Eval.dataset.data_dir=/mnt/
+                Eval.dataset.label_file_list=[\"/mnt/hiertext/label_hiertext_val.txt\"]
+                Global.save_model_dir=./output/
+                Global.pretrained_model=./pre_train/ch_PP-OCRv3_det_distill_train/best_accuracy
+            resources:
+              limits:
+                nvidia.com/gpu: 1
+            volumeMounts:  # 添加 shared memory 挂载以防止缓存出错
+              - mountPath: /dev/shm
+                name: dshm
+        volumes:
+          - name: dshm
+            emptyDir:
+              medium: Memory
+```
+
+本案例采用GPU进行训练，如果您只有CPU机器，则可以将镜像替换成CPU版本 `paddlecloud/paddleocr:2.5-cpu-efbb0a`，并在args中加上参数`Global.use_gpu=false`。
+
+```bash linenums="1"
+# 创建PaddleJob训练模型
+$ kubectl apply -f ppocrv3.yaml
+paddlejob.batch.paddlepaddle.org/ppocrv3 created
+
+# 查看PaddleJob状态
+$ kubectl get pods -n paddlecloud -l paddle-res-name=ppocrv3-worker-0
+NAME               READY   STATUS    RESTARTS   AGE
+ppocrv3-worker-0   1/1     Running   0          4s
+
+# 查看训练日志
+$ kubectl logs -f ppocrv3-worker-0 -n paddlecloud
+```
+
+## 更多资源
+
+欢迎关注[云上飞桨项目PaddleCloud](https://github.com/PaddlePaddle/PaddleCloud)，我们为您提供了飞桨模型套件标准镜像以及全栈的云原生模型套件部署组件，如您有任何关于飞桨模型套件的部署问题，请联系我们。
+如果你发现任何PaddleCloud存在的问题或者是建议, 欢迎通过[GitHub Issues](https://github.com/PaddlePaddle/PaddleCloud/issues)给我们提issues。
diff --git a/docs/version2.x/legacy/paddle_server.en.md b/docs/version2.x/legacy/paddle_server.en.md
new file mode 100644
index 0000000..0777858
--- /dev/null
+++ b/docs/version2.x/legacy/paddle_server.en.md
@@ -0,0 +1,288 @@
+---
+comments: true
+hide:
+  - toc
+---
+
+## OCR Pipeline WebService
+
+PaddleOCR provides two service deployment methods:
+
+- Based on **PaddleHub Serving**: Code path is "`./deploy/hubserving`". Please refer to the [tutorial](https://github.com/PaddlePaddle/PaddleOCR/blob/db0ad17cf631fafc01650c177e00ce76413af97f/deploy/hubserving/readme_en.md)
+- Based on **PaddleServing**: Code path is "`./deploy/pdserving`". Please follow this tutorial.
+
+### Service deployment based on PaddleServing
+
+This document will introduce how to use the [PaddleServing](https://github.com/PaddlePaddle/Serving/blob/develop/README.md) to deploy the PPOCR dynamic graph model as a pipeline online service.
+
+Some Key Features of Paddle Serving:
+
+- Integrate with Paddle training pipeline seamlessly, most paddle models can be deployed with one line command.
+- Industrial serving features supported, such as models management, online loading, online A/B testing etc.
+- Highly concurrent and efficient communication between clients and servers supported.
+
+PaddleServing supports deployment in multiple languages. In this example, two deployment methods, python pipeline and C++, are provided. The comparison between the two is as follows:
+
+| Language | Speed | Secondary development | Do you need to compile |
+|-----|-----|---------|------------|
+| C++ | fast | Slightly difficult | Single model prediction does not need to be compiled, multi-model concatenation needs to be compiled |
+| python | general | easy | single-model/multi-model no compilation required |
+
+The introduction and tutorial of Paddle Serving service deployment framework reference [document](https://github.com/PaddlePaddle/Serving/blob/develop/README.md).
+
+### Environmental preparation
+
+PaddleOCR operating environment and Paddle Serving operating environment are needed.
+
+1. Please prepare PaddleOCR operating environment reference [link](../ppocr/environment.en.md).
+   Download the corresponding paddlepaddle whl package according to the environment, it is recommended to install version 2.2.2.
+
+2. The steps of PaddleServing operating environment prepare are as follows:
+
+    ```bash linenums="1"
+    # Install serving which used to start the service
+    wget https://paddle-serving.bj.bcebos.com/test-dev/whl/paddle_serving_server_gpu-0.8.3.post102-py3-none-any.whl
+    pip3 install paddle_serving_server_gpu-0.8.3.post102-py3-none-any.whl
+
+    # Install paddle-serving-server for cuda10.1
+    # wget https://paddle-serving.bj.bcebos.com/test-dev/whl/paddle_serving_server_gpu-0.8.3.post101-py3-none-any.whl
+    # pip3 install paddle_serving_server_gpu-0.8.3.post101-py3-none-any.whl
+
+    # Install serving which used to start the service
+    wget https://paddle-serving.bj.bcebos.com/test-dev/whl/paddle_serving_client-0.8.3-cp37-none-any.whl
+    pip3 install paddle_serving_client-0.8.3-cp37-none-any.whl
+
+    # Install serving-app
+    wget https://paddle-serving.bj.bcebos.com/test-dev/whl/paddle_serving_app-0.8.3-py3-none-any.whl
+    pip3 install paddle_serving_app-0.8.3-py3-none-any.whl
+    ```
+
+   **note:** If you want to install the latest version of PaddleServing, refer to [link](https://github.com/PaddlePaddle/Serving/blob/v0.8.3/doc/Latest_Packages_CN.md).
+
+### Model conversion
+
+When using PaddleServing for service deployment, you need to convert the saved inference model into a serving model that is easy to deploy.
+
+Firstly, download the [inference model](../ppocr/model_list.en.md) of PPOCR
+
+```bash linenums="1"
+# Download and unzip the OCR text detection model
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar -O PP-OCRv3_mobile_det_infer.tar && tar -xf PP-OCRv3_mobile_det_infer.tar
+# Download and unzip the OCR text recognition model
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar -O PP-OCRv3_mobile_rec_infer.tar &&  tar -xf PP-OCRv3_mobile_rec_infer.tar
+```
+
+Then, you can use installed paddle_serving_client tool to convert inference model to mobile model.
+
+```bash linenums="1"
+#  Detection model conversion
+python3 -m paddle_serving_client.convert --dirname ./PP-OCRv3_mobile_det_infer/ \
+                                         --model_filename inference.pdmodel          \
+                                         --params_filename inference.pdiparams       \
+                                         --serving_server ./ppocr_det_v3_serving/ \
+                                         --serving_client ./ppocr_det_v3_client/
+
+#  Recognition model conversion
+python3 -m paddle_serving_client.convert --dirname ./PP-OCRv3_mobile_rec_infer/ \
+                                         --model_filename inference.pdmodel          \
+                                         --params_filename inference.pdiparams       \
+                                         --serving_server ./ppocr_rec_v3_serving/  \
+                                         --serving_client ./ppocr_rec_v3_client/
+
+```
+
+After the detection model is converted, there will be additional folders of `ppocr_det_v3_serving` and `ppocr_det_v3_client` in the current folder, with the following format:
+
+```text linenums="1"
+|- ppocr_det_v3_serving/
+  |- __model__
+  |- __params__
+  |- serving_server_conf.prototxt
+  |- serving_server_conf.stream.prototxt
+
+|- ppocr_det_v3_client
+  |- serving_client_conf.prototxt
+  |- serving_client_conf.stream.prototxt
+
+```
+
+The recognition model is the same.
+
+### Paddle Serving pipeline deployment
+
+1. Download the PaddleOCR code, if you have already downloaded it, you can skip this step.
+
+    ```bash linenums="1"
+    git clone https://github.com/PaddlePaddle/PaddleOCR
+
+    # Enter the working directory
+    cd PaddleOCR/deploy/pdserving/
+    ```
+
+    The pdserver directory contains the code to start the pipeline service and send prediction requests, including:
+
+    ```bash linenums="1"
+    __init__.py
+    config.yml # Start the service configuration file
+    ocr_reader.py # OCR model pre-processing and post-processing code implementation
+    pipeline_http_client.py # Script to send pipeline prediction request
+    web_service.py # Start the script of the pipeline server
+    ```
+
+2. Run the following command to start the service.
+
+    ```bash linenums="1"
+    # Start the service and save the running log in log.txt
+    python3 web_service.py --config=config.yml &>log.txt &
+    ```
+
+    After the service is successfully started, a log similar to the following will be printed in log.txt
+
+    ![](./images/start_server.png)
+
+3. Send service request
+
+    ```bash linenums="1"
+    python3 pipeline_http_client.py
+    ```
+
+    After successfully running, the predicted result of the model will be printed in the cmd window. An example of the result is:
+
+    ![](./images/results.png)
+
+    Adjust the number of concurrency in config.yml to get the largest QPS. Generally, the number of concurrent detection and recognition is 2:1
+
+    ```yaml linenums="1"
+    det:
+        concurrency: 8
+        ...
+    rec:
+        concurrency: 4
+        ...
+    ```
+
+    Multiple service requests can be sent at the same time if necessary.
+
+    The predicted performance data will be automatically written into the `PipelineServingLogs/pipeline.tracer` file.
+
+    Tested on 200 real pictures, and limited the detection long side to 960. The average QPS on T4 GPU can reach around 23:
+
+    ```bash linenums="1"
+    2021-05-13 03:42:36,895 ==================== TRACER ======================
+    2021-05-13 03:42:36,975 Op(rec):
+    2021-05-13 03:42:36,976         in[14.472382882882883 ms]
+    2021-05-13 03:42:36,976         prep[9.556855855855856 ms]
+    2021-05-13 03:42:36,976         midp[59.921905405405404 ms]
+    2021-05-13 03:42:36,976         postp[15.345945945945946 ms]
+    2021-05-13 03:42:36,976         out[1.9921216216216215 ms]
+    2021-05-13 03:42:36,976         idle[0.16254943864471572]
+    2021-05-13 03:42:36,976 Op(det):
+    2021-05-13 03:42:36,976         in[315.4468035714286 ms]
+    2021-05-13 03:42:36,976         prep[69.5980625 ms]
+    2021-05-13 03:42:36,976         midp[18.989535714285715 ms]
+    2021-05-13 03:42:36,976         postp[18.857803571428573 ms]
+    2021-05-13 03:42:36,977         out[3.1337544642857145 ms]
+    2021-05-13 03:42:36,977         idle[0.7477961159203756]
+    2021-05-13 03:42:36,977 DAGExecutor:
+    2021-05-13 03:42:36,977         Query count[224]
+    2021-05-13 03:42:36,977         QPS[22.4 q/s]
+    2021-05-13 03:42:36,977         Succ[0.9910714285714286]
+    2021-05-13 03:42:36,977         Error req[169, 170]
+    2021-05-13 03:42:36,977         Latency:
+    2021-05-13 03:42:36,977                 ave[535.1678348214285 ms]
+    2021-05-13 03:42:36,977                 .50[172.651 ms]
+    2021-05-13 03:42:36,977                 .60[187.904 ms]
+    2021-05-13 03:42:36,977                 .70[245.675 ms]
+    2021-05-13 03:42:36,977                 .80[526.684 ms]
+    2021-05-13 03:42:36,977                 .90[854.596 ms]
+    2021-05-13 03:42:36,977                 .95[1722.728 ms]
+    2021-05-13 03:42:36,977                 .99[3990.292 ms]
+    2021-05-13 03:42:36,978 Channel (server worker num[10]):
+    2021-05-13 03:42:36,978         chl0(In: ['@DAGExecutor'], Out: ['det']) size[0/0]
+    2021-05-13 03:42:36,979         chl1(In: ['det'], Out: ['rec']) size[6/0]
+    2021-05-13 03:42:36,979         chl2(In: ['rec'], Out: ['@DAGExecutor']) size[0/0]
+    ```
+
+### C++ Serving
+
+Service deployment based on python obviously has the advantage of convenient secondary development. However, the real application often needs to pursue better performance. PaddleServing also provides a more performant C++ deployment version.
+
+The C++ service deployment is the same as python in the environment setup and data preparation stages, the difference is when the service is started and the client sends requests.
+
+1. Compile Serving
+
+   To improve predictive performance, C++ services also provide multiple model concatenation services. Unlike Python Pipeline services, multiple model concatenation requires the pre - and post-model processing code to be written on the server side, so local recompilation is required to generate serving. Specific may refer to the official document: [how to compile Serving](https://github.com/PaddlePaddle/Serving/blob/v0.8.3/doc/Compile_EN.md)
+
+2. Run the following command to start the service.
+
+    ```bash linenums="1"
+    # Start the service and save the running log in log.txt
+    python3 -m paddle_serving_server.serve --model ppocr_det_v3_serving ppocr_rec_v3_serving --op GeneralDetectionOp GeneralInferOp --port 8181 &>log.txt &
+    ```
+
+    After the service is successfully started, a log similar to the following will be printed in log.txt
+    ![](./images/start_server.png)
+
+3. Send service request
+
+   Due to the need for pre and post-processing in the C++Server part, in order to speed up the input to the C++Server is only the base64 encoded string of the picture, it needs to be manually modified
+   Change the feed_type field and shape field in ppocr_det_v3_client/serving_client_conf.prototxt to the following:
+
+   ```bash linenums="1"
+    feed_var {
+    name: "x"
+    alias_name: "x"
+    is_lod_tensor: false
+    feed_type: 20
+    shape: 1
+    }
+   ```
+
+   start the client:
+
+    ```bash linenums="1"
+    python3 ocr_cpp_client.py ppocr_det_v3_client ppocr_rec_v3_client
+    ```
+
+    After successfully running, the predicted result of the model will be printed in the cmd window. An example of the result is:
+
+    ![](./images/results.png)
+
+### WINDOWS Users
+
+Windows does not support Pipeline Serving, if we want to launch paddle serving on Windows, we should use Web Service, for more information please refer to [Paddle Serving for Windows Users](https://github.com/PaddlePaddle/Serving/blob/develop/doc/Windows_Tutorial_EN.md)
+
+**WINDOWS user can only use version 0.5.0 CPU Mode**
+
+**Prepare Stage:**
+
+```bash linenums="1"
+pip3 install paddle-serving-server==0.5.0
+pip3 install paddle-serving-app==0.3.1
+```
+
+1. Start Server
+
+    ```bash linenums="1"
+    cd win
+    python3 ocr_web_server.py gpu(for gpu user)
+    or
+    python3 ocr_web_server.py cpu(for cpu user)
+    ```
+
+2. Client Send Requests
+
+    ```bash linenums="1"
+    python3 ocr_web_client.py
+    ```
+
+### FAQ
+
+**Q1**: No result return after sending the request.
+
+**A1**: Do not set the proxy when starting the service and sending the request. You can close the proxy before starting the service and before sending the request. The command to close the proxy is:
+
+```bash linenums="1"
+unset https_proxy
+unset http_proxy
+```
diff --git a/docs/version2.x/legacy/paddle_server.md b/docs/version2.x/legacy/paddle_server.md
new file mode 100644
index 0000000..52b1783
--- /dev/null
+++ b/docs/version2.x/legacy/paddle_server.md
@@ -0,0 +1,307 @@
+---
+typora-copy-images-to: images
+comments: true
+hide:
+  - toc
+---
+
+## PPOCR 服务化部署
+
+PaddleOCR提供2种服务部署方式：
+
+- 基于PaddleHub Serving的部署：代码路径为"`./deploy/hubserving`"，使用方法参考[文档](https://github.com/PaddlePaddle/PaddleOCR/blob/db0ad17cf631fafc01650c177e00ce76413af97f/deploy/hubserving/readme.md)；
+- 基于PaddleServing的部署：代码路径为"`./deploy/pdserving`"，按照本教程使用。
+
+### 基于PaddleServing的服务部署
+
+本文档将介绍如何使用[PaddleServing](https://github.com/PaddlePaddle/Serving/blob/develop/README_CN.md) 工具部署PP-OCR动态图模型的pipeline在线服务。
+
+相比较于hubserving部署，PaddleServing具备以下优点：
+
+- 支持客户端和服务端之间高并发和高效通信
+- 支持 工业级的服务能力 例如模型管理，在线加载，在线A/B测试等
+- 支持 多种编程语言 开发客户端，例如C++, Python和Java
+
+PaddleServing 支持多种语言部署，本例中提供了python pipeline 和 C++ 两种部署方式，两者的对比如下：
+
+| 语言   | 速度 | 二次开发 | 是否需要编译                           |
+| ------ | ---- | -------- | -------------------------------------- |
+| C++    | 很快 | 略有难度 | 单模型预测无需编译，多模型串联需要编译 |
+| python | 一般 | 容易     | 单模型/多模型 均无需编译               |
+
+更多有关PaddleServing服务化部署框架介绍和使用教程参考[文档](https://github.com/PaddlePaddle/Serving/blob/develop/README_CN.md)。
+
+AIStudio演示案例可参考 [基于PaddleServing的OCR服务化部署实战](https://aistudio.baidu.com/aistudio/projectdetail/3630726)。
+
+#### 环境准备
+
+需要准备PaddleOCR的运行环境和Paddle Serving的运行环境。
+
+- 准备PaddleOCR的运行环境[链接](../ppocr/environment.md)
+
+    ```bash linenums="1"
+    git clone https://github.com/PaddlePaddle/PaddleOCR
+
+    # 进入到工作目录
+    cd PaddleOCR/deploy/pdserving/
+    ```
+
+- 准备PaddleServing的运行环境，步骤如下
+
+```bash linenums="1"
+# 安装serving，用于启动服务
+wget https://paddle-serving.bj.bcebos.com/test-dev/whl/paddle_serving_server_gpu-0.8.3.post102-py3-none-any.whl
+pip3 install paddle_serving_server_gpu-0.8.3.post102-py3-none-any.whl
+# 如果是cuda10.1环境，可以使用下面的命令安装paddle-serving-server
+# wget https://paddle-serving.bj.bcebos.com/test-dev/whl/paddle_serving_server_gpu-0.8.3.post101-py3-none-any.whl
+# pip3 install paddle_serving_server_gpu-0.8.3.post101-py3-none-any.whl
+
+# 安装client，用于向服务发送请求
+wget https://paddle-serving.bj.bcebos.com/test-dev/whl/paddle_serving_client-0.8.3-cp37-none-any.whl
+pip3 install paddle_serving_client-0.8.3-cp37-none-any.whl
+
+# 安装serving-app
+wget https://paddle-serving.bj.bcebos.com/test-dev/whl/paddle_serving_app-0.8.3-py3-none-any.whl
+pip3 install paddle_serving_app-0.8.3-py3-none-any.whl
+```
+
+**Note:** 如果要安装最新版本的PaddleServing参考[链接](https://github.com/PaddlePaddle/Serving/blob/v0.9.0/doc/Latest_Packages_CN.md)。
+
+### 模型转换
+
+使用PaddleServing做服务化部署时，需要将保存的inference模型转换为serving易于部署的模型。
+
+首先，下载PP-OCR的[inference模型](https://github.com/PaddlePaddle/PaddleOCR#pp-ocr-series-model-listupdate-on-september-8th)
+
+```bash linenums="1"
+# 下载并解压 OCR 文本检测模型
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar -O PP-OCRv3_mobile_det_infer.tar && tar -xf PP-OCRv3_mobile_det_infer.tar
+# 下载并解压 OCR 文本识别模型
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar -O PP-OCRv3_mobile_rec_infer.tar &&  tar -xf PP-OCRv3_mobile_rec_infer.tar
+```
+
+接下来，用安装的paddle_serving_client把下载的inference模型转换成易于server部署的模型格式。
+
+```bash linenums="1"
+# 转换检测模型
+python3 -m paddle_serving_client.convert --dirname ./PP-OCRv3_mobile_det_infer/ \
+                                         --model_filename inference.pdmodel          \
+                                         --params_filename inference.pdiparams       \
+                                         --serving_server ./ppocr_det_v3_serving/ \
+                                         --serving_client ./ppocr_det_v3_client/
+
+# 转换识别模型
+python3 -m paddle_serving_client.convert --dirname ./PP-OCRv3_mobile_rec_infer/ \
+                                         --model_filename inference.pdmodel          \
+                                         --params_filename inference.pdiparams       \
+                                         --serving_server ./ppocr_rec_v3_serving/  \
+                                         --serving_client ./ppocr_rec_v3_client/
+```
+
+检测模型转换完成后，会在当前文件夹多出`ppocr_det_v3_serving` 和`ppocr_det_v3_client`的文件夹，具备如下格式：
+
+```text linenums="1"
+|- ppocr_det_v3_serving/
+  |- __model__
+  |- __params__
+  |- serving_server_conf.prototxt
+  |- serving_server_conf.stream.prototxt
+
+|- ppocr_det_v3_client
+  |- serving_client_conf.prototxt
+  |- serving_client_conf.stream.prototxt
+
+```
+
+识别模型同理。
+
+### Paddle Serving pipeline部署
+
+#### 1. 确认工作目录下文件结构
+
+pdserver目录包含启动pipeline服务和发送预测请求的代码，包括：
+
+```bash linenums="1"
+__init__.py
+config.yml            # 启动服务的配置文件
+ocr_reader.py         # OCR模型预处理和后处理的代码实现
+pipeline_http_client.py   # 发送pipeline预测请求的脚本
+web_service.py        # 启动pipeline服务端的脚本
+```
+
+#### 2. 启动服务
+
+```bash linenums="1"
+# 启动服务，运行日志保存在log.txt
+python3 web_service.py --config=config.yml &>log.txt &
+```
+
+成功启动服务后，log.txt中会打印类似如下日志
+
+![img](./images/start_server.png)
+
+#### 3. 发送服务请求
+
+```bash linenums="1"
+python3 pipeline_http_client.py
+```
+
+成功运行后，模型预测的结果会打印在cmd窗口中，结果示例为：
+
+![img](./images/pipeline_result.png)
+
+调整 config.yml 中的并发个数获得最大的QPS, 一般检测和识别的并发数为2：1
+
+```yaml linenums="1"
+det:
+    #并发数，is_thread_op=True时，为线程并发；否则为进程并发
+    concurrency: 8
+    ...
+rec:
+    #并发数，is_thread_op=True时，为线程并发；否则为进程并发
+    concurrency: 4
+    ...
+```
+
+有需要的话可以同时发送多个服务请求
+
+预测性能数据会被自动写入 `PipelineServingLogs/pipeline.tracer` 文件中。
+
+在200张真实图片上测试，把检测长边限制为960。T4 GPU 上 QPS 均值可达到23左右：
+
+```bash linenums="1"
+2021-05-13 03:42:36,895 ==================== TRACER ======================
+2021-05-13 03:42:36,975 Op(rec):
+2021-05-13 03:42:36,976         in[14.472382882882883 ms]
+2021-05-13 03:42:36,976         prep[9.556855855855856 ms]
+2021-05-13 03:42:36,976         midp[59.921905405405404 ms]
+2021-05-13 03:42:36,976         postp[15.345945945945946 ms]
+2021-05-13 03:42:36,976         out[1.9921216216216215 ms]
+2021-05-13 03:42:36,976         idle[0.16254943864471572]
+2021-05-13 03:42:36,976 Op(det):
+2021-05-13 03:42:36,976         in[315.4468035714286 ms]
+2021-05-13 03:42:36,976         prep[69.5980625 ms]
+2021-05-13 03:42:36,976         midp[18.989535714285715 ms]
+2021-05-13 03:42:36,976         postp[18.857803571428573 ms]
+2021-05-13 03:42:36,977         out[3.1337544642857145 ms]
+2021-05-13 03:42:36,977         idle[0.7477961159203756]
+2021-05-13 03:42:36,977 DAGExecutor:
+2021-05-13 03:42:36,977         Query count[224]
+2021-05-13 03:42:36,977         QPS[22.4 q/s]
+2021-05-13 03:42:36,977         Succ[0.9910714285714286]
+2021-05-13 03:42:36,977         Error req[169, 170]
+2021-05-13 03:42:36,977         Latency:
+2021-05-13 03:42:36,977                 ave[535.1678348214285 ms]
+2021-05-13 03:42:36,977                 .50[172.651 ms]
+2021-05-13 03:42:36,977                 .60[187.904 ms]
+2021-05-13 03:42:36,977                 .70[245.675 ms]
+2021-05-13 03:42:36,977                 .80[526.684 ms]
+2021-05-13 03:42:36,977                 .90[854.596 ms]
+2021-05-13 03:42:36,977                 .95[1722.728 ms]
+2021-05-13 03:42:36,977                 .99[3990.292 ms]
+2021-05-13 03:42:36,978 Channel (server worker num[10]):
+2021-05-13 03:42:36,978         chl0(In: ['@DAGExecutor'], Out: ['det']) size[0/0]
+2021-05-13 03:42:36,979         chl1(In: ['det'], Out: ['rec']) size[6/0]
+2021-05-13 03:42:36,979         chl2(In: ['rec'], Out: ['@DAGExecutor']) size[0/0]
+```
+
+### Paddle Serving C++ 部署
+
+基于python的服务部署，显然具有二次开发便捷的优势，然而真正落地应用，往往需要追求更优的性能。PaddleServing 也提供了性能更优的C++部署版本。
+
+C++ 服务部署在环境搭建和数据准备阶段与 python 相同，区别在于启动服务和客户端发送请求时不同。
+
+#### 1. 准备 Serving 环境
+
+为了提高预测性能，C++ 服务同样提供了多模型串联服务。与python pipeline服务不同，多模型串联的过程中需要将模型前后处理代码写在服务端，因此需要在本地重新编译生成serving。
+
+首先需要下载Serving代码库, 把OCR文本检测预处理相关代码替换到Serving库中
+
+```bash linenums="1"
+git clone https://github.com/PaddlePaddle/Serving
+cp -rf general_detection_op.cpp Serving/core/general-server/op
+```
+
+具体可参考官方文档：[如何编译Serving](https://github.com/PaddlePaddle/Serving/blob/v0.8.3/doc/Compile_CN.md)，注意需要开启 WITH_OPENCV 选项。
+
+完成编译后，注意要安装编译出的三个whl包，并设置SERVING_BIN环境变量。
+
+#### 2. 启动服务
+
+一个服务启动两个模型串联，只需要在--model后依次按顺序传入模型文件夹的相对路径，且需要在--op后依次传入自定义C++OP类名称：
+
+```bash linenums="1"
+# 启动服务，运行日志保存在log.txt
+python3 -m paddle_serving_server.serve --model ppocr_det_v3_serving ppocr_rec_v3_serving --op GeneralDetectionOp GeneralInferOp --port 8181 &>log.txt &
+```
+
+成功启动服务后，log.txt中会打印类似如下日志
+![](./images/start_server.png)
+
+#### 3. 发送服务请求
+
+由于需要在C++Server部分进行前后处理，为了加速传入C++Server的仅仅是图片的base64编码的字符串，故需要手动修改
+ppocr_det_v3_client/serving_client_conf.prototxt 中 feed_type 字段 和 shape 字段，修改成如下内容：
+
+```bash linenums="1"
+feed_var {
+name: "x"
+alias_name: "x"
+is_lod_tensor: false
+feed_type: 20
+shape: 1
+}
+```
+
+启动客户端
+
+```bash linenums="1"
+python3 ocr_cpp_client.py ppocr_det_v3_client ppocr_rec_v3_client
+```
+
+成功运行后，模型预测的结果会打印在cmd窗口中，结果示例为：
+
+![img](./images/results.png)
+
+在浏览器中输入服务器 ip:端口号，可以看到当前服务的实时QPS。(端口号范围需要是8000-9000)
+
+在200张真实图片上测试，把检测长边限制为960。T4 GPU 上 QPS 峰值可达到51左右,约为pipeline的 2.12 倍。
+
+### Windows用户
+
+Windows用户不能使用上述的启动方式，需要使用Web Service，详情参见[Windows平台使用Paddle Serving指导](https://github.com/PaddlePaddle/Serving/blob/develop/doc/Windows_Tutorial_CN.md)
+
+**WINDOWS只能使用0.5.0版本的CPU模式**
+
+准备阶段：
+
+```bash linenums="1"
+pip3 install paddle-serving-server==0.5.0
+pip3 install paddle-serving-app==0.3.1
+```
+
+#### 1. 启动服务端程序
+
+```bash linenums="1"
+cd win
+python3 ocr_web_server.py gpu(使用gpu方式)
+或者
+python3 ocr_web_server.py cpu(使用cpu方式)
+```
+
+#### 2. 发送服务请求
+
+```bash linenums="1"
+python3 ocr_web_client.py
+```
+
+### FAQ
+
+**Q1**： 发送请求后没有结果返回或者提示输出解码报错
+
+**A1**： 启动服务和发送请求时不要设置代理，可以在启动服务前和发送请求前关闭代理，关闭代理的命令是：
+
+```bash linenums="1"
+unset https_proxy
+unset http_proxy
+```
diff --git a/docs/version2.x/legacy/python_infer.en.md b/docs/version2.x/legacy/python_infer.en.md
new file mode 100755
index 0000000..7cf3048
--- /dev/null
+++ b/docs/version2.x/legacy/python_infer.en.md
@@ -0,0 +1,178 @@
+---
+typora-copy-images-to: images
+comments: true
+hide:
+  - toc
+---
+
+# Python Inference for PP-OCR Model Zoo
+
+This article introduces the use of the Python inference engine for the PP-OCR model library. The content is in order of text detection, text recognition, direction classifier and the prediction method of the three in series on the CPU and GPU.
+
+## Text Detection Model Inference
+
+The default configuration is based on the inference setting of the DB text detection model. For lightweight Chinese detection model inference, you can execute the following commands:
+
+```bash linenums="1"
+# download DB text detection inference model
+wget  https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar
+tar xf PP-OCRv3_mobile_det_infer.tar
+# run inference
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs/00018069.jpg" --det_model_dir="./PP-OCRv3_mobile_det_infer/"
+```
+
+The visual text detection results are saved to the ./inference_results folder by default, and the name of the result file is prefixed with 'det_res'. Examples of results are as follows:
+
+![img](./images/det_res_00018069.jpg)
+
+You can use the parameters `limit_type` and `det_limit_side_len` to limit the size of the input image,
+The optional parameters of `limit_type` are [`max`, `min`], and
+`det_limit_size_len` is a positive integer, generally set to a multiple of 32, such as 960.
+
+The default setting of the parameters is `limit_type='max', det_limit_side_len=960`. Indicates that the longest side of the network input image cannot exceed 960,
+If this value is exceeded, the image will be resized with the same width ratio to ensure that the longest side is `det_limit_side_len`.
+Set as `limit_type='min', det_limit_side_len=960`, it means that the shortest side of the image is limited to 960.
+
+If the resolution of the input picture is relatively large and you want to use a larger resolution prediction, you can set det_limit_side_len to the desired value, such as 1216:
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs/1.jpg" --det_model_dir="./PP-OCRv3_mobile_det_infer/" --det_limit_type=max --det_limit_side_len=1216
+```
+
+If you want to use the CPU for prediction, execute the command as follows
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs/1.jpg" --det_model_dir="./PP-OCRv3_mobile_det_infer/"  --use_gpu=False
+```
+
+## Text Recognition Model Inference
+
+### 1. Lightweight Chinese Recognition Model Inference
+
+**Note**: The input shape used by the recognition model of `PP-OCRv3` is `3, 48, 320`. If you use other recognition models, you need to set the parameter `--rec_image_shape` according to the model. In addition, the `rec_algorithm` used by the recognition model of `PP-OCRv3` is `SVTR_LCNet` by default. Note the difference from the original `SVTR`.
+
+For lightweight Chinese recognition model inference, you can execute the following commands:
+
+```bash linenums="1"
+# download CRNN text recognition inference model
+wget  https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar
+tar xf PP-OCRv3_mobile_rec_infer.tar
+# run inference
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_10.png" --rec_model_dir="./PP-OCRv3_mobile_rec_infer/" --rec_image_shape=3,48,320
+```
+
+![img](./images/word_10.png)
+
+After executing the command, the prediction results (recognized text and score) of the above image will be printed on the screen.
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words_en/word_10.png:('PAIN', 0.988671)
+```
+
+### 2. English Recognition Model Inference
+
+For English recognition model inference, you can execute the following commands,you need to specify the dictionary path used by `--rec_char_dict_path`:
+
+```bash linenums="1"
+# download en model：
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv3_mobile_rec_infer.tar
+tar xf en_PP-OCRv3_mobile_rec_infer.tar
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/en/word_1.png" --rec_model_dir="./en_PP-OCRv3_mobile_rec_infer/" --rec_char_dict_path="ppocr/utils/en_dict.txt"
+```
+
+![img](./images/word_1.png)
+
+After executing the command, the prediction result of the above figure is:
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words/en/word_1.png: ('JOINT', 0.998160719871521)
+```
+
+### 3. Multilingual Model Inference
+
+If you need to predict [other language models](../ppocr/model_list.en.md), when using inference model prediction, you need to specify the dictionary path used by `--rec_char_dict_path`. At the same time, in order to get the correct visualization results,
+You need to specify the visual font path through `--vis_font_path`. There are small language fonts provided by default under the `doc/fonts` path, such as Korean recognition:
+
+```bash linenums="1"
+wget wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/korean_mobile_v2.0_rec_infer.tar
+
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/korean/1.jpg" --rec_model_dir="./your inference model" --rec_char_dict_path="ppocr/utils/dict/korean_dict.txt" --vis_font_path="doc/fonts/korean.ttf"
+```
+
+![img](./images/1.jpg)
+
+After executing the command, the prediction result of the above figure is:
+
+```text linenums="1"
+Predicts of ./doc/imgs_words/korean/1.jpg:('바탕으로', 0.9948904)
+```
+
+## Angle Classification Model Inference
+
+For angle classification model inference, you can execute the following commands:
+
+```bash linenums="1"
+# download text angle class inference model：
+wget  https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar
+tar xf ch_ppocr_mobile_v2.0_cls_infer.tar
+python3 tools/infer/predict_cls.py --image_dir="./doc/imgs_words_en/word_10.png" --cls_model_dir="ch_ppocr_mobile_v2.0_cls_infer"
+```
+
+![img](./images/word_10.png)
+
+After executing the command, the prediction results (classification angle and score) of the above image will be printed on the screen.
+
+```text linenums="1"
+ Predicts of ./doc/imgs_words_en/word_10.png:['0', 0.9999995]
+```
+
+## Text Detection Angle Classification and Recognition Inference Concatenation
+
+**Note**: The input shape used by the recognition model of `PP-OCRv3` is `3, 48, 320`. If you use other recognition models, you need to set the parameter `--rec_image_shape` according to the model. In addition, the `rec_algorithm` used by the recognition model of `PP-OCRv3` is `SVTR_LCNet` by default. Note the difference from the original `SVTR`.
+
+When performing prediction, you need to specify the path of a single image or a folder of images through the parameter `image_dir`, pdf file is also supported, the parameter `det_model_dir` specifies the path to detect the inference model, the parameter `cls_model_dir` specifies the path to angle classification inference model and the parameter `rec_model_dir` specifies the path to identify the inference model. The parameter `use_angle_cls` is used to control whether to enable the angle classification model. The parameter `use_mp` specifies whether to use multi-process to infer `total_process_num` specifies process number when using multi-process. The parameter . The visualized recognition results are saved to the `./inference_results` folder by default.
+
+```bash linenums="1"
+# use direction classifier
+python3 tools/infer/predict_system.py --image_dir="./doc/imgs/00018069.jpg" --det_model_dir="./PP-OCRv3_mobile_det_infer/" --cls_model_dir="./cls/" --rec_model_dir="./PP-OCRv3_mobile_rec_infer/" --use_angle_cls=true
+# not use use direction classifier
+python3 tools/infer/predict_system.py --image_dir="./doc/imgs/00018069.jpg" --det_model_dir="./PP-OCRv3_mobile_det_infer/" --rec_model_dir="./PP-OCRv3_mobile_rec_infer/" --use_angle_cls=false
+# use multi-process
+python3 tools/infer/predict_system.py --image_dir="./doc/imgs/00018069.jpg" --det_model_dir="./PP-OCRv3_mobile_det_infer/" --rec_model_dir="./PP-OCRv3_mobile_rec_infer/" --use_angle_cls=false --use_mp=True --total_process_num=6
+# use PDF files, you can infer the first few pages by using the `page_num` parameter, the default is 0, which means infer all pages
+python3 tools/infer/predict_system.py --image_dir="./xxx.pdf" --det_model_dir="./PP-OCRv3_mobile_det_infer/" --cls_model_dir="./cls/" --rec_model_dir="./PP-OCRv3_mobile_rec_infer/" --use_angle_cls=true --page_num=2
+```
+
+After executing the command, the recognition result image is as follows:
+
+![](./images/system_res_00018069_v3.jpg)
+
+For more configuration and explanation of inference parameters, please refer to：[Model Inference Parameters Explained Tutorial](../ppocr/blog/inference_args.en.md)。
+
+## TensorRT Inference
+
+Paddle Inference ensembles TensorRT using subgraph mode. For GPU deployment scenarios, TensorRT can optimize some subgraphs, including horizontal and vertical integration of OPs, filter redundant OPs, and automatically select the optimal OP kernels for to speed up inference.
+
+You need to do the following 2 steps for inference using TRT.
+
+* (1) Collect the dynamic shape information of the model about a specific dataset and store it in a file.
+* (2) Load the dynamic shape information file for TRT inference.
+
+Taking the text detection model as an example. Firstly, you can use the following command to generate a dynamic shape file, which will eventually be named as `det_trt_dynamic_shape.txt` and stored in the `PP-OCRv3_mobile_det_infer` folder.
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs/1.jpg" --det_model_dir="./PP-OCRv3_mobile_det_infer/" --use_tensorrt=True
+```
+
+The above command is only used to collect dynamic shape information, and TRT is not used during inference.
+
+Then, you can use the following command to perform TRT inference.
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs/1.jpg" --det_model_dir="./PP-OCRv3_mobile_det_infer/" --use_tensorrt=True
+```
+
+**Note:**
+
+* In the first step, if the dynamic shape information file already exists, it does not need to be collected again. If you want to regenerate the dynamic shape information file, you need to delete the dynamic shape information file in the model folder firstly, and then regenerate it.
+* In general, dynamic shape information file only needs to be generated once. In the actual deployment process, it is recommended that the dynamic shape information file can be generated on offline validation set or test set, and then the file can be directly loaded for online TRT inference.
diff --git a/docs/version2.x/legacy/python_infer.md b/docs/version2.x/legacy/python_infer.md
new file mode 100644
index 0000000..892f0f1
--- /dev/null
+++ b/docs/version2.x/legacy/python_infer.md
@@ -0,0 +1,504 @@
+---
+typora-copy-images-to: images
+comments: true
+hide:
+  - toc
+---
+
+# 基于Python预测引擎推理
+
+inference 模型（`paddle.jit.save`保存的模型）
+一般是模型训练，把模型结构和模型参数保存在文件中的固化模型，多用于预测部署场景。
+训练过程中保存的模型是checkpoints模型，保存的只有模型的参数，多用于恢复训练等。
+与checkpoints模型相比，inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成。
+
+本文介绍针对PP-OCR模型库的Python推理引擎使用方法，内容依次为文本检测、文本识别、方向分类器以及三者串联在CPU、GPU上的预测方法。
+
+## 一、训练模型转inference模型
+
+### 检测模型转inference模型
+
+下载超轻量级中文检测模型：
+
+```bash linenums="1"
+wget -P ./ch_lite/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_train.tar && tar xf ./ch_lite/ch_ppocr_mobile_v2.0_det_train.tar -C ./ch_lite/
+```
+
+上述模型是以MobileNetV3为backbone训练的DB算法，将训练好的模型转换成inference模型只需要运行如下命令：
+
+```bash linenums="1"
+# -c 后面设置训练算法的yml配置文件
+# -o 配置可选参数
+# Global.pretrained_model 参数设置待转换的训练模型地址，不用添加文件后缀 .pdmodel，.pdopt或.pdparams。
+# Global.save_inference_dir参数设置转换的模型将保存的地址。
+
+python3 tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model=./ch_lite/ch_ppocr_mobile_v2.0_det_train/best_accuracy Global.save_inference_dir=./inference/det_db/
+```
+
+转inference模型时，使用的配置文件和训练时使用的配置文件相同。另外，还需要设置配置文件中的`Global.pretrained_model`参数，其指向训练中保存的模型参数文件。
+转换成功后，在模型保存目录下有三个文件：
+
+```text linenums="1"
+inference/det_db/
+    ├── inference.pdiparams         # 检测inference模型的参数文件
+    ├── inference.pdiparams.info    # 检测inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 检测inference模型的program文件
+```
+
+### 识别模型转inference模型
+
+下载超轻量中文识别模型：
+
+```bash linenums="1"
+wget -P ./ch_lite/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_train.tar && tar xf ./ch_lite/ch_ppocr_mobile_v2.0_rec_train.tar -C ./ch_lite/
+```
+
+识别模型转inference模型与检测的方式相同，如下：
+
+```bash linenums="1"
+# -c 后面设置训练算法的yml配置文件
+# -o 配置可选参数
+# Global.pretrained_model 参数设置待转换的训练模型地址，不用添加文件后缀 .pdmodel，.pdopt或.pdparams。
+# Global.save_inference_dir参数设置转换的模型将保存的地址。
+
+python3 tools/export_model.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml -o Global.pretrained_model=./ch_lite/ch_ppocr_mobile_v2.0_rec_train/best_accuracy  Global.save_inference_dir=./inference/rec_crnn/
+```
+
+**注意：** 如果您是在自己的数据集上训练的模型，并且调整了中文字符的字典文件，请注意修改配置文件中的`character_dict_path`是否是所需要的字典文件。
+
+转换成功后，在目录下有三个文件：
+
+```bash linenums="1"
+/inference/rec_crnn/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+### 方向分类模型转inference模型
+
+下载方向分类模型：
+
+```bash linenums="1"
+wget -P ./ch_lite/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar && tar xf ./ch_lite/ch_ppocr_mobile_v2.0_cls_train.tar -C ./ch_lite/
+```
+
+方向分类模型转inference模型与检测的方式相同，如下：
+
+```bash linenums="1"
+# -c 后面设置训练算法的yml配置文件
+# -o 配置可选参数
+# Global.pretrained_model 参数设置待转换的训练模型地址，不用添加文件后缀 .pdmodel，.pdopt或.pdparams。
+# Global.save_inference_dir参数设置转换的模型将保存的地址。
+
+python3 tools/export_model.py -c configs/cls/cls_mv3.yml -o Global.pretrained_model=./ch_lite/ch_ppocr_mobile_v2.0_cls_train/best_accuracy  Global.save_inference_dir=./inference/cls/
+```
+
+转换成功后，在目录下有三个文件：
+
+```text linenums="1"
+/inference/cls/
+    ├── inference.pdiparams         # 分类inference模型的参数文件
+    ├── inference.pdiparams.info    # 分类inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 分类inference模型的program文件
+```
+
+## 二、文本检测模型推理
+
+文本检测模型推理，默认使用DB模型的配置参数。当不使用DB模型时，在推理时，需要通过传入相应的参数进行算法适配，细节参考下文。
+
+### 1. 超轻量中文检测模型推理
+
+超轻量中文检测模型推理，可以执行如下命令：
+
+```bash linenums="1"
+# 下载超轻量中文检测模型：
+wget  https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar
+tar xf ch_ppocr_mobile_v2.0_det_infer.tar
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs/00018069.jpg" --det_model_dir="./ch_ppocr_mobile_v2.0_det_infer/"
+```
+
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为`det_res`。结果示例如下：
+
+![](./images/det_res_00018069.jpg)
+
+通过参数`limit_type`和`det_limit_side_len`来对图片的尺寸进行限制，
+`limit_type`可选参数为[`max`, `min`]，
+`det_limit_size_len` 为正整数，一般设置为32 的倍数，比如960。
+
+参数默认设置为`limit_type='max', det_limit_side_len=960`。表示网络输入图像的最长边不能超过960，
+如果超过这个值，会对图像做等宽比的resize操作，确保最长边为`det_limit_side_len`。
+设置为`limit_type='min', det_limit_side_len=960` 则表示限制图像的最短边为960。
+
+如果输入图片的分辨率比较大，而且想使用更大的分辨率预测，可以设置det_limit_side_len 为想要的值，比如1216：
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs/1.jpg" --det_model_dir="./inference/det_db/" --det_limit_type=max --det_limit_side_len=1216
+```
+
+如果想使用CPU进行预测，执行命令如下
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs/1.jpg" --det_model_dir="./inference/det_db/"  --use_gpu=False
+```
+
+### 2. DB文本检测模型推理
+
+首先将DB文本检测训练过程中保存的模型，转换成inference model。以基于Resnet50_vd骨干网络，在ICDAR2015英文数据集训练的模型为例（ [模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r50_vd_db.yml -o Global.pretrained_model=./det_r50_vd_db_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/det_db
+```
+
+DB文本检测模型推理，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_db/"
+```
+
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为`det_res`。结果示例如下：
+
+![img](./images/det_res_img_10_db.jpg)
+
+**注意**：由于ICDAR2015数据集只有1000张训练图像，且主要针对英文场景，所以上述模型对中文文本图像检测效果会比较差。
+
+### 3. EAST文本检测模型推理
+
+首先将EAST文本检测训练过程中保存的模型，转换成inference model。以基于Resnet50_vd骨干网络，在ICDAR2015英文数据集训练的模型为例（ [模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r50_vd_east.yml -o Global.pretrained_model=./det_r50_vd_east_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/det_east
+```
+
+**EAST文本检测模型推理，需要设置参数`--det_algorithm="EAST"`**，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --det_algorithm="EAST" --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_east/"
+```
+
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为`det_res`。结果示例如下：
+
+![img](./images/det_res_img_10_east.jpg)
+
+**注意**：本代码库中，EAST后处理Locality-Aware NMS有python和c++两种版本，c++版速度明显快于python版。由于c++版本nms编译版本问题，只有python3.5环境下会调用c++版nms，其他情况将调用python版nms。
+
+### 4. SAST文本检测模型推理
+
+#### (1). 四边形文本检测模型（ICDAR2015）
+
+首先将SAST文本检测训练过程中保存的模型，转换成inference model。以基于Resnet50_vd骨干网络，在ICDAR2015英文数据集训练的模型为例([模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar))，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r50_vd_sast_icdar15.yml -o Global.pretrained_model=./det_r50_vd_sast_icdar15_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/det_sast_ic15
+
+```
+
+**SAST文本检测模型推理，需要设置参数`--det_algorithm="SAST"`**，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --det_algorithm="SAST" --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_sast_ic15/"
+```
+
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为`det_res`。结果示例如下：
+
+![](./images/det_res_img_10_sast.jpg)
+
+#### (2). 弯曲文本检测模型（Total-Text）
+
+首先将SAST文本检测训练过程中保存的模型，转换成inference model。以基于Resnet50_vd骨干网络，在Total-Text英文数据集训练的模型为例（[模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_totaltext_v2.0_train.tar))，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_r50_vd_sast_totaltext.yml -o Global.pretrained_model=./det_r50_vd_sast_totaltext_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/det_sast_tt
+
+```
+
+SAST文本检测模型推理，需要设置参数`--det_algorithm="SAST"`，同时，还需要增加参数`--det_sast_polygon=True`，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --det_algorithm="SAST" --image_dir="./doc/imgs_en/img623.jpg" --det_model_dir="./inference/det_sast_tt/" --det_sast_polygon=True
+```
+
+可视化文本检测结果默认保存到`./inference_results`文件夹里面，结果文件的名称前缀为`det_res`。结果示例如下：
+
+![img](./images/det_res_img623_sast.jpg)
+
+**注意**：本代码库中，SAST后处理Locality-Aware NMS有python和c++两种版本，c++版速度明显快于python版。由于c++版本nms编译版本问题，只有python3.5环境下会调用c++版nms，其他情况将调用python版nms。
+
+## 三、文本识别模型推理
+
+下面将介绍超轻量中文识别模型推理、基于CTC损失的识别模型推理和基于Attention损失的识别模型推理。对于中文文本识别，建议优先选择基于CTC损失的识别模型，实践中也发现基于Attention损失的效果不如基于CTC损失的识别模型。此外，如果训练时修改了文本的字典，请参考下面的自定义文本识别字典的推理。
+
+### 1. 超轻量中文识别模型推理
+
+超轻量中文识别模型推理，可以执行如下命令：
+
+```bash linenums="1"
+# 下载超轻量中文识别模型：
+wget  https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar
+tar xf ch_ppocr_mobile_v2.0_rec_infer.tar
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/ch/word_4.jpg" --rec_model_dir="ch_ppocr_mobile_v2.0_rec_infer"
+```
+
+![img](./images/word_4.jpg)
+
+执行命令后，上面图像的预测结果（识别的文本和得分）会打印到屏幕上，示例如下：
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words/ch/word_4.jpg:('实力活力', 0.98458153)
+```
+
+### 2. 基于CTC损失的识别模型推理
+
+我们以 CRNN 为例，介绍基于CTC损失的识别模型推理。 Rosetta 使用方式类似，不用设置识别算法参数rec_algorithm。
+
+首先将 CRNN 文本识别训练过程中保存的模型，转换成inference model。以基于Resnet34_vd骨干网络，使用MJSynth和SynthText两个英文文本识别合成数据集训练
+的模型为例（ [模型下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_bilstm_ctc_v2.0_train.tar) )，可以使用如下命令进行转换：
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/rec/rec_r34_vd_none_bilstm_ctc.yml -o Global.pretrained_model=./rec_r34_vd_none_bilstm_ctc_v2.0_train/best_accuracy  Global.save_inference_dir=./inference/rec_crnn
+```
+
+CRNN 文本识别模型推理，可以执行如下命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_336.png" --rec_model_dir="./inference/rec_crnn/" --rec_image_shape="3, 32, 100" --rec_char_dict_path="./ppocr/utils/ic15_dict.txt"
+```
+
+![img](./images/word_336.png)
+
+执行命令后，上面图像的识别结果如下：
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words_en/word_336.png:('super', 0.9999073)
+```
+
+**注意**：由于上述模型是参考[DTRB](https://arxiv.org/abs/1904.01906)文本识别训练和评估流程，与超轻量级中文识别模型训练有两方面不同：
+
+- 训练时采用的图像分辨率不同，训练上述模型采用的图像分辨率是[3，32，100]，而中文模型训练时，为了保证长文本的识别效果，训练时采用的图像分辨率是[3, 32, 320]。预测推理程序默认的形状参数是训练中文采用的图像分辨率，即[3, 32, 320]。因此，这里推理上述英文模型时，需要通过参数rec_image_shape设置识别图像的形状。
+
+- 字符列表，DTRB论文中实验只是针对26个小写英文本母和10个数字进行实验，总共36个字符。所有大小字符都转成了小写字符，不在上面列表的字符都忽略，认为是空格。因此这里没有输入字符字典，而是通过如下命令生成字典.因此在推理时需要设置参数rec_char_dict_path，指定为英文字典"./ppocr/utils/ic15_dict.txt"。
+
+```python linenums="1"
+self.character_str = "0123456789abcdefghijklmnopqrstuvwxyz"
+dict_character = list(self.character_str)
+```
+
+### 3. 基于SRN损失的识别模型推理
+
+基于SRN损失的识别模型，需要额外设置识别算法参数 --rec_algorithm="SRN"。
+同时需要保证预测shape与训练时一致，如： --rec_image_shape="1, 64, 256"
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_336.png" \
+                                   --rec_model_dir="./inference/srn/" \
+                                   --rec_image_shape="1, 64, 256" \
+                                   --rec_char_dict_path="./ppocr/utils/ic15_dict.txt" \
+                                   --rec_algorithm="SRN"
+```
+
+### 4. 自定义文本识别字典的推理
+
+如果训练时修改了文本的字典，在使用inference模型预测时，需要通过`--rec_char_dict_path`指定使用的字典路径
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_336.png" --rec_model_dir="./your inference model" --rec_image_shape="3, 32, 100"  --rec_char_dict_path="your text dict path"
+```
+
+### 5. 多语言模型的推理
+
+如果您需要预测的是其他语言模型，在使用inference模型预测时，需要通过`--rec_char_dict_path`指定使用的字典路径, 同时为了得到正确的可视化结果，
+需要通过 `--vis_font_path` 指定可视化的字体路径，`doc/fonts/` 路径下有默认提供的小语种字体，例如韩文识别：
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words/korean/1.jpg" --rec_model_dir="./your inference model"  --rec_char_dict_path="ppocr/utils/dict/korean_dict.txt" --vis_font_path="doc/fonts/korean.ttf"
+```
+
+![img](./images/1.jpg)
+
+执行命令后，上图的预测结果为：
+
+``` text
+Predicts of ./doc/imgs_words/korean/1.jpg:('바탕으로', 0.9948904)
+```
+
+## 四、方向分类模型推理
+
+下面将介绍方向分类模型推理。
+
+### 1. 方向分类模型推理
+
+方向分类模型推理，可以执行如下命令：
+
+```bash linenums="1"
+# 下载超轻量中文方向分类器模型：
+wget  https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar
+tar xf ch_ppocr_mobile_v2.0_cls_infer.tar
+python3 tools/infer/predict_cls.py --image_dir="./doc/imgs_words/ch/word_4.jpg" --cls_model_dir="ch_ppocr_mobile_v2.0_cls_infer"
+```
+
+![img](./images/word_1.jpg)
+
+执行命令后，上面图像的预测结果（分类的方向和得分）会打印到屏幕上，示例如下：
+
+```bash linenums="1"
+Predicts of ./doc/imgs_words/ch/word_4.jpg:['0', 0.9999982]
+```
+
+## 五、文本检测、方向分类和文字识别串联推理
+
+### 1. 超轻量中文OCR模型推理
+
+在执行预测时，需要通过参数`image_dir`指定单张图像或者图像集合的路径、参数`det_model_dir`,`cls_model_dir`和`rec_model_dir`分别指定检测，方向分类和识别的inference模型路径。参数`use_angle_cls`用于控制是否启用方向分类模型。`use_mp`表示是否使用多进程。`total_process_num`表示在使用多进程时的进程数。可视化识别结果默认保存到 ./inference_results 文件夹里面。
+
+```bash linenums="1"
+# 使用方向分类器
+python3 tools/infer/predict_system.py --image_dir="./doc/imgs/00018069.jpg" --det_model_dir="./inference/det_db/" --cls_model_dir="./inference/cls/" --rec_model_dir="./inference/rec_crnn/" --use_angle_cls=true
+
+# 不使用方向分类器
+python3 tools/infer/predict_system.py --image_dir="./doc/imgs/00018069.jpg" --det_model_dir="./inference/det_db/" --rec_model_dir="./inference/rec_crnn/" --use_angle_cls=false
+
+# 使用多进程
+python3 tools/infer/predict_system.py --image_dir="./doc/imgs/00018069.jpg" --det_model_dir="./inference/det_db/" --rec_model_dir="./inference/rec_crnn/" --use_angle_cls=false --use_mp=True --total_process_num=6
+```
+
+执行命令后，识别结果图像如下：
+
+![](./images/system_res_00018069_v3.jpg)
+
+### 2. 其他模型推理
+
+如果想尝试使用其他检测算法或者识别算法，请参考上述文本检测模型推理和文本识别模型推理，更新相应配置和模型。
+
+**注意：由于检测框矫正逻辑的局限性，暂不支持使用SAST弯曲文本检测模型（即，使用参数`--det_sast_polygon=True`时）进行模型串联。**
+
+下面给出基于EAST文本检测和STAR-Net文本识别执行命令：
+
+```bash linenums="1"
+python3 tools/infer/predict_system.py --image_dir="./doc/imgs_en/img_10.jpg" --det_model_dir="./inference/det_east/" --det_algorithm="EAST" --rec_model_dir="./inference/starnet/" --rec_image_shape="3, 32, 100" --rec_char_dict_path="./ppocr/utils/ic15_dict.txt"
+```
+
+执行命令后，识别结果图像如下：
+
+![img](./images/img_10_east_starnet.jpg)
+
+## 六、参数解释
+
+更多关于预测过程的参数解释如下所示。
+
+- 全局信息
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  image_dir | str | 无，必须显式指定 | 图像或者文件夹路径 |
+|  vis_font_path | str | "./doc/fonts/simfang.ttf" | 用于可视化的字体路径 |
+|  drop_score | float | 0.5 | 识别得分小于该值的结果会被丢弃，不会作为返回结果 |
+|  use_pdserving | bool | False | 是否使用Paddle Serving进行预测 |
+|  warmup | bool | False | 是否开启warmup，在统计预测耗时的时候，可以使用这种方法 |
+|  draw_img_save_dir | str | "./inference_results" | 系统串联预测OCR结果的保存文件夹 |
+|  save_crop_res | bool | False  | 是否保存OCR的识别文本图像 |
+|  crop_res_save_dir | str | "./output" | 保存OCR识别出来的文本图像路径 |
+|  use_mp | bool | False | 是否开启多进程预测  |
+|  total_process_num | int | 6 | 开启的进城数，`use_mp`为`True`时生效  |
+|  process_id | int | 0 | 当前进程的id号，无需自己修改  |
+|  benchmark | bool | False | 是否开启benchmark，对预测速度、显存占用等进行统计  |
+|  save_log_path | str | "./log_output/" | 开启`benchmark`时，日志结果的保存文件夹 |
+|  show_log | bool | True | 是否显示预测中的日志信息  |
+|  use_onnx | bool | False | 是否开启onnx预测 |
+
+- 预测引擎相关
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  use_gpu | bool | True | 是否使用GPU进行预测 |
+|  ir_optim | bool | True | 是否对计算图进行分析与优化，开启后可以加速预测过程 |
+|  use_tensorrt | bool | False | 是否开启tensorrt |
+|  min_subgraph_size | int | 15 | tensorrt中最小子图size，当子图的size大于该值时，才会尝试对该子图使用trt engine计算 |
+|  precision | str | fp32 | 预测的精度，支持`fp32`, `fp16`, `int8` 3种输入 |
+|  enable_mkldnn | bool | True | 是否开启mkldnn |
+|  cpu_threads | int | 10 | 开启mkldnn时，cpu预测的线程数 |
+
+- 文本检测模型相关
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  det_algorithm | str | "DB" | 文本检测算法名称，目前支持`DB`, `EAST`, `SAST`, `PSE`  |
+|  det_model_dir | str | xx | 检测inference模型路径 |
+|  det_limit_side_len | int | 960 | 检测的图像边长限制 |
+|  det_limit_type | str | "max" | 检测的变成限制类型，目前支持`min`, `max`，`min`表示保证图像最短边不小于`det_limit_side_len`，`max`表示保证图像最长边不大于`det_limit_side_len` |
+
+其中，DB算法相关参数如下
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  det_db_thresh | float | 0.3 | DB输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点 |
+|  det_db_box_thresh | float | 0.6 | 检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域 |
+|  det_db_unclip_ratio | float | 1.5 | `Vatti clipping`算法的扩张系数，使用该方法对文字区域进行扩张 |
+|  max_batch_size | int | 10 | 预测的batch size |
+|  use_dilation | bool | False | 是否对分割结果进行膨胀以获取更优检测效果 |
+|  det_db_score_mode | str | "fast" | DB的检测结果得分计算方法，支持`fast`和`slow`，`fast`是根据polygon的外接矩形边框内的所有像素计算平均得分，`slow`是根据原始polygon内的所有像素计算平均得分，计算速度相对较慢一些，但是更加准确一些。 |
+
+EAST算法相关参数如下
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  det_east_score_thresh | float | 0.8 | EAST后处理中score map的阈值 |
+|  det_east_cover_thresh | float | 0.1 | EAST后处理中文本框的平均得分阈值 |
+|  det_east_nms_thresh | float | 0.2 | EAST后处理中nms的阈值 |
+
+SAST算法相关参数如下
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  det_sast_score_thresh | float | 0.5 | SAST后处理中的得分阈值 |
+|  det_sast_nms_thresh | float | 0.5 | SAST后处理中nms的阈值 |
+|  det_sast_polygon | bool | False | 是否多边形检测，弯曲文本场景（如Total-Text）设置为True |
+
+PSE算法相关参数如下
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  det_pse_thresh | float | 0.0 | 对输出图做二值化的阈值 |
+|  det_pse_box_thresh | float | 0.85 | 对box进行过滤的阈值，低于此阈值的丢弃 |
+|  det_pse_min_area | float | 16 | box的最小面积，低于此阈值的丢弃 |
+|  det_pse_box_type | str | "box" | 返回框的类型，box:四点坐标，poly: 弯曲文本的所有点坐标 |
+|  det_pse_scale | int | 1 | 输入图像相对于进后处理的图的比例，如`640*640`的图像，网络输出为`160*160`，scale为2的情况下，进后处理的图片shape为`320*320`。这个值调大可以加快后处理速度，但是会带来精度的下降 |
+
+- 文本识别模型相关
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  rec_algorithm | str | "CRNN" | 文本识别算法名称，目前支持`CRNN`, `SRN`, `RARE`, `NETR`, `SAR` |
+|  rec_model_dir | str | 无，如果使用识别模型，该项是必填项 | 识别inference模型路径 |
+|  rec_image_shape | list | [3, 32, 320] | 识别时的图像尺寸， |
+|  rec_batch_num | int | 6 | 识别的batch size |
+|  max_text_length | int | 25 | 识别结果最大长度，在`SRN`中有效 |
+|  rec_char_dict_path | str | "./ppocr/utils/ppocr_keys_v1.txt" | 识别的字符字典文件 |
+|  use_space_char | bool | True | 是否包含空格，如果为`True`，则会在最后字符字典中补充`空格`字符 |
+
+- 端到端文本检测与识别模型相关
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  e2e_algorithm | str | "PGNet" | 端到端算法名称，目前支持`PGNet` |
+|  e2e_model_dir | str | 无，如果使用端到端模型，该项是必填项 | 端到端模型inference模型路径 |
+|  e2e_limit_side_len | int | 768 | 端到端的输入图像边长限制 |
+|  e2e_limit_type | str | "max" | 端到端的边长限制类型，目前支持`min`, `max`，`min`表示保证图像最短边不小于`e2e_limit_side_len`，`max`表示保证图像最长边不大于`e2e_limit_side_len` |
+|  e2e_pgnet_score_thresh | float | 0.5 | 端到端得分阈值，小于该阈值的结果会被丢弃 |
+|  e2e_char_dict_path | str | "./ppocr/utils/ic15_dict.txt" | 识别的字典文件路径 |
+|  e2e_pgnet_valid_set | str | "totaltext" | 验证集名称，目前支持`totaltext`, `partvgg`，不同数据集对应的后处理方式不同，与训练过程保持一致即可 |
+|  e2e_pgnet_mode | str | "fast" | PGNet的检测结果得分计算方法，支持`fast`和`slow`，`fast`是根据polygon的外接矩形边框内的所有像素计算平均得分，`slow`是根据原始polygon内的所有像素计算平均得分，计算速度相对较慢一些，但是更加准确一些。 |
+
+- 方向分类器模型相关
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  use_angle_cls | bool | False | 是否使用方向分类器 |
+|  cls_model_dir | str | 无，如果需要使用，则必须显式指定路径 | 方向分类器inference模型路径 |
+|  cls_image_shape | list | [3, 48, 192] | 预测尺度 |
+|  label_list | list | ['0', '180'] | class id对应的角度值 |
+|  cls_batch_num | int | 6 | 方向分类器预测的batch size |
+|  cls_thresh | float | 0.9 | 预测阈值，模型预测结果为180度，且得分大于该阈值时，认为最终预测结果为180度，需要翻转 |
+
+## 七、FAQ
+
+- 如果是使用paddle2.0之前版本的代码导出的`inference模型`，则其文件名为`model`与`params`，分别对应paddle2.0或者之后版本导出的`inference.pdmodel`与`inference.pdiparams`；不过目前PaddleOCR的release分支已经不支持paddle2.0之前版本导出的inference 模型，如果希望使用，需要使用develop分支（静态图分支）的代码与文档。
diff --git a/docs/version2.x/legacy/windows_vs2019_build.en.md b/docs/version2.x/legacy/windows_vs2019_build.en.md
new file mode 100644
index 0000000..c801768
--- /dev/null
+++ b/docs/version2.x/legacy/windows_vs2019_build.en.md
@@ -0,0 +1,147 @@
+---
+comments: true
+hide:
+  - toc
+---
+
+# Visual Studio 2022 Community CMake Compilation Guide
+
+PaddleOCR has been tested on Windows using `Visual Studio 2022 Community`. Microsoft started supporting direct `CMake` project management from `Visual Studio 2017`, but it wasn't fully stable and reliable until `2019`. If you want to use CMake for project management and compilation, we recommend using `Visual Studio 2022`.
+
+**All examples below assume the working directory is **`D:\projects\cpp`**.**
+
+## 1. Environment Preparation
+
+### 1.1 Install Required Dependencies
+
+- Visual Studio 2019 or newer
+- CUDA 10.2, cuDNN 7+ (only required for the GPU version of the prediction library). Additionally, the NVIDIA Computing Toolkit must be installed, and the NVIDIA cuDNN library must be downloaded.
+- CMake 3.22+
+
+Ensure that the above dependencies are installed before proceeding. In this tutorial the Community Edition of `VS2022` was used.
+
+### 1.2 Download PaddlePaddle C++ Prediction Library and OpenCV
+
+#### 1.2.1 Download PaddlePaddle C++ Prediction Library
+
+PaddlePaddle C++ prediction libraries offer different precompiled versions for various `CPU` and `CUDA` configurations. Download the appropriate version from: [C++ Prediction Library Download List](https://www.paddlepaddle.org.cn/inference/master/guides/install/download_lib.html#windows)
+
+After extraction, the `D:\projects\paddle_inference` directory should contain:
+
+```
+paddle_inference
+├── paddle # Core Paddle library and header files
+|
+├── third_party # Third-party dependencies and headers
+|
+└── version.txt # Version and compilation information
+```
+
+#### 1.2.2 Install and Configure OpenCV
+
+1. Download OpenCV for Windows from the [official release page](https://github.com/opencv/opencv/releases).
+2. Run the downloaded executable and extract OpenCV to a specified directory, e.g., `D:\projects\cpp\opencv`.
+
+#### 1.2.3 Download PaddleOCR Code
+
+```bash
+git clone https://github.com/PaddlePaddle/Paddle.git
+git checkout develop
+```
+
+## 2. Running the Project
+
+### Step 1: Create a Visual Studio Project
+
+Once CMake is installed, open the `cmake-gui` application. Specify the source code directory in the first input box and the build output directory in the second input box.
+
+![step1](./images/cmake_step1.png)
+
+### Step 2: Run CMake Configuration
+
+Click the `Configure` button at the bottom of the interface. The first time you run it, a prompt will appear asking for the Visual Studio configuration. Select your `Visual Studio` version and set the target platform to `x64`. Click `Finish` to start the configuration process.
+
+![step2](./images/cmake_step2.jpg)
+
+The first run will result in errors, which is expected. You now need to configure OpenCV and the prediction library.
+
+- **For CPU version**, configure the following variables:
+
+  - `OPENCV_DIR`: Path to the OpenCV `lib` folder
+  - `OpenCV_DIR`: Same as `OPENCV_DIR`
+  - `PADDLE_LIB`: Path to the `paddle_inference` folder
+
+- **For GPU version**, configure additional variables:
+
+  - `CUDA_LIB`: CUDA path, e.g., `C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\lib\x64`
+  - `CUDNN_LIB`: Path to extracted CuDNN library, e.g., `D:\CuDNN-8.9.7.29`
+  - `TENSORRT_DIR`: Path to extracted TensorRT, e.g., `D:\TensorRT-8.0.1.6`
+  - `WITH_GPU`: Check this option
+  - `WITH_TENSORRT`: Check this option
+
+Example configuration:
+
+![step3](./images/cmake_step3.jpg)
+
+Once configured, click `Configure` again.
+
+**Note:**
+
+1. If using `openblas`, uncheck `WITH_MKL`.
+2. If you encounter the error `unable to access 'https://github.com/LDOUBLEV/AutoLog.git/': gnutls_handshake() failed`, update `deploy/cpp_infer/external-cmake/auto-log.cmake` to use `https://gitee.com/Double_V/AutoLog`.
+
+### Step 3: Generate Visual Studio Project
+
+Click `Generate` to create the `.sln` file for the Visual Studio project.
+![step4](./images/cmake_step4.jpg)
+
+Click `Open Project` to launch the project in Visual Studio. The interface should look like this:
+![step5](./images/vs_step1.jpg)
+
+Before building the solution, perform the following steps:
+
+1. Change `Debug` to `Release` mode.
+2. Download [dirent.h](https://paddleocr.bj.bcebos.com/deploy/cpp_infer/cpp_files/dirent.h) and copy it to the Visual Studio include directory, e.g., `C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Tools\MSVC\include`.
+
+Click `Build -> Build Solution`. Once completed, the `ppocr.exe` file should appear in the `build/Release/` folder.
+
+Before running, copy the following files to `build/Release/`:
+
+1. `paddle_inference/paddle/lib/paddle_inference.dll`
+2. `paddle_inference/paddle/lib/common.dll`
+3. `paddle_inference/third_party/install/mklml/lib/mklml.dll`
+4. `paddle_inference/third_party/install/mklml/lib/libiomp5md.dll`
+5. `paddle_inference/third_party/install/onednn/lib/mkldnn.dll`
+6. `opencv/build/x64/vc15/bin/opencv_world455.dll`
+7. If using the `openblas` version, also copy `paddle_inference/third_party/install/openblas/lib/openblas.dll`.
+
+### Step 4: Run the Prediction
+
+The compiled executable is located in the `build/Release/` directory. Open `cmd` and navigate to `D:\projects\cpp\PaddleOCR\deploy\cpp_infer\`:
+
+```
+cd /d D:\projects\cpp\PaddleOCR\deploy\cpp_infer
+```
+
+Run the prediction using `ppocr.exe`. For more usage details, refer to the to the [Instructions](./cpp_infer.en.md) section of running the demo.
+
+```shell
+# Switch terminal encoding to UTF-8
+CHCP 65001
+
+# If using PowerShell, run this command before execution to fix character encoding issues:
+$OutputEncoding = [console]::InputEncoding = [console]::OutputEncoding = New-Object System.Text.UTF8Encoding
+
+# Execute prediction
+.\build\Release\ppocr.exe system --det_model_dir=D:\projects\cpp\ch_PP-OCRv2_det_slim_quant_infer --rec_model_dir=D:\projects\cpp\ch_PP-OCRv2_rec_slim_quant_infer --image_dir=D:\projects\cpp\PaddleOCR\doc\imgs\11.jpg
+```
+
+<br>
+
+## Sample result:
+![result](./images/result.jpg)
+
+## FAQ
+
+- **Issue:** Application fails to start with error `(0xc0000142)` and `cmd` output shows `You are using Paddle compiled with TensorRT, but TensorRT dynamic library is not found.`
+- **Solution:** Copy all `.dll` files from the `TensorRT` directory's `lib` folder into the `release` directory and try running it again.
diff --git a/docs/version2.x/legacy/windows_vs2019_build.md b/docs/version2.x/legacy/windows_vs2019_build.md
new file mode 100644
index 0000000..de120a9
--- /dev/null
+++ b/docs/version2.x/legacy/windows_vs2019_build.md
@@ -0,0 +1,139 @@
+---
+comments: true
+hide:
+  - toc
+---
+
+# Visual Studio 2019 Community CMake 编译指南
+
+PaddleOCR在Windows 平台下基于`Visual Studio 2019 Community` 进行了测试。微软从`Visual Studio 2017`开始即支持直接管理`CMake`跨平台编译项目，但是直到`2019`才提供了稳定和完全的支持，所以如果你想使用CMake管理项目编译构建，我们推荐你使用`Visual Studio 2019`环境下构建。
+
+**下面所有示例以工作目录为 `D:\projects\cpp`演示**。
+
+## 1. 环境准备
+
+### 1.1 安装必须环境
+
+- Visual Studio 2019
+- CUDA 10.2，cudnn 7+ （仅在使用GPU版本的预测库时需要）
+- CMake 3.22+
+
+请确保系统已经安装好上述基本软件，我们使用的是`VS2019`的社区版。
+
+### 1.2 下载 PaddlePaddle C++ 预测库和 Opencv
+
+#### 1.2.1 下载 PaddlePaddle C++ 预测库
+
+PaddlePaddle C++ 预测库针对不同的`CPU`和`CUDA`版本提供了不同的预编译版本，请根据实际情况下载:  [C++预测库下载列表](https://www.paddlepaddle.org.cn/inference/master/guides/install/download_lib.html#windows)
+
+解压后`D:\projects\paddle_inference`目录包含内容为：
+
+```
+paddle_inference
+├── paddle # paddle核心库和头文件
+|
+├── third_party # 第三方依赖库和头文件
+|
+└── version.txt # 版本和编译信息
+```
+
+#### 1.2.2 安装配置OpenCV
+
+1. 在OpenCV官网下载适用于Windows平台的Opencv， [下载地址](https://github.com/opencv/opencv/releases)
+2. 运行下载的可执行文件，将OpenCV解压至指定目录，如`D:\projects\cpp\opencv`
+
+#### 1.2.3 下载PaddleOCR代码
+
+```bash linenums="1"
+git clone -b dygraph https://github.com/PaddlePaddle/PaddleOCR
+```
+
+## 2. 开始运行
+
+### Step1: 构建Visual Studio项目
+
+cmake安装完后后系统里会有一个cmake-gui程序，打开cmake-gui，在第一个输入框处填写源代码路径，第二个输入框处填写编译输出路径
+
+![step1](./images/cmake_step1.png)
+
+### Step2: 执行cmake配置
+
+点击界面下方的`Configure`按钮，第一次点击会弹出提示框进行Visual Studio配置，如下图，选择你的Visual Studio版本即可，目标平台选择x64。然后点击`finish`按钮即开始自动执行配置。
+
+![step2](./images/cmake_step2.jpg)
+
+第一次执行会报错，这是正常现象，接下来进行Opencv和预测库的配置
+
+- cpu版本，仅需考虑OPENCV_DIR、OpenCV_DIR、PADDLE_LIB三个参数
+
+    - OPENCV_DIR：填写opencv lib文件夹所在位置
+    - OpenCV_DIR：同填写opencv lib文件夹所在位置
+    - PADDLE_LIB：paddle_inference文件夹所在位置
+
+- GPU版本，在cpu版本的基础上，还需填写以下变量
+CUDA_LIB、CUDNN_LIB、TENSORRT_DIR、WITH_GPU、WITH_TENSORRT
+
+- CUDA_LIB: CUDA地址，如 `C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\lib\x64`
+- CUDNN_LIB: 和CUDA_LIB一致
+- TENSORRT_DIR：TRT下载后解压缩的位置，如 `D:\TensorRT-8.0.1.6`
+- WITH_GPU: 打钩
+- WITH_TENSORRT：打勾
+
+配置好的截图如下
+
+![step3](./images/cmake_step3.jpg)
+
+配置完成后，再次点击`Configure`按钮。
+
+**注意：**
+
+  1. 如果使用的是`openblas`版本，请把`WITH_MKL`勾去掉
+  2. 遇到报错 `unable to access 'https://github.com/LDOUBLEV/AutoLog.git/': gnutls_handshake() failed: The TLS connection was non-properly terminated.`， 将 `deploy/cpp_infer/external-cmake/auto-log.cmake` 中的github地址改为 <https://gitee.com/Double_V/AutoLog> 地址即可。
+
+### Step3: 生成Visual Studio 项目
+
+点击`Generate`按钮即可生成Visual Studio 项目的sln文件。
+![step4](./images/cmake_step4.jpg)
+
+点击`Open Project`按钮即可在Visual Studio 中打开项目。打开后截图如下
+
+![step5](./images/vs_step1.jpg)
+
+在开始生成解决方案之前，执行下面步骤：
+
+1. 将`Debug`改为`Release`
+2. 下载[dirent.h](https://paddleocr.bj.bcebos.com/deploy/cpp_infer/cpp_files/dirent.h)，并拷贝到 Visual Studio 的 include 文件夹下，如`C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Auxiliary\VS\include`。
+
+点击`生成->生成解决方案`，即可在`build/Release/`文件夹下看见`ppocr.exe`文件。
+
+运行之前，将下面文件拷贝到`build/Release/`文件夹下
+
+1. `paddle_inference/paddle/lib/paddle_inference.dll`
+2. `paddle_inference/third_party/install/onnxruntime/lib/onnxruntime.dll`
+3. `paddle_inference/third_party/install/paddle2onnx/lib/paddle2onnx.dll`
+4. `opencv/build/x64/vc15/bin/opencv_world455.dll`
+5. 如果使用openblas版本的预测库还需要拷贝 `paddle_inference/third_party/install/openblas/lib/openblas.dll`
+
+### Step4: 预测
+
+上述`Visual Studio 2019`编译产出的可执行文件在`build/Release/`目录下，打开`cmd`，并切换到`D:\projects\cpp\PaddleOCR\deploy\cpp_infer\`：
+
+```bash linenums="1"
+cd /d D:\projects\cpp\PaddleOCR\deploy\cpp_infer
+```
+
+可执行文件`ppocr.exe`即为样例的预测程序，其主要使用方法如下，更多使用方法可以参考[说明文档](./cpp_infer.md)`运行demo`部分。
+
+```bash linenums="1"
+# 切换终端编码为utf8
+CHCP 65001
+# 执行预测
+.\build\Release\ppocr.exe system --det_model_dir=D:\projects\cpp\ch_PP-OCRv2_det_slim_quant_infer --rec_model_dir=D:\projects\cpp\ch_PP-OCRv2_rec_slim_quant_infer --image_dir=D:\projects\cpp\PaddleOCR\doc\imgs\11.jpg
+```
+
+识别结果如下
+![result](./images/result.jpg)
+
+## FAQ
+
+- 运行时，弹窗报错提示`应用程序无法正常启动(0xc0000142)`，并且`cmd`窗口内提示`You are using Paddle compiled with TensorRT, but TensorRT dynamic library is not found.`，把tensort目录下的lib里面的所有dll文件复制到release目录下，再次运行即可。
diff --git a/docs/version2.x/model/hardware/install_other_devices.md b/docs/version2.x/model/hardware/install_other_devices.md
new file mode 100644
index 0000000..f5fbb4b
--- /dev/null
+++ b/docs/version2.x/model/hardware/install_other_devices.md
@@ -0,0 +1,66 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+# 多硬件安装飞桨
+
+本文档主要针对昇腾 NPU 硬件平台，介绍如何安装飞桨。
+
+## 1. 昇腾 NPU 飞桨安装
+
+### 1.1 环境准备
+
+当前 PaddleOCR 支持昇腾 910B 芯片，昇腾驱动版本为 23.0.3。考虑到环境差异性，我们推荐使用飞桨官方提供的标准镜像完成环境准备。
+
+#### 拉取镜像
+
+此镜像仅为开发环境，镜像中不包含预编译的飞桨安装包，镜像中已经默认安装了昇腾算子库 CANN-8.0.RC1。
+
+```bash linenums="1"
+# 适用于 X86 架构，暂时不提供 Arch64 架构镜像
+docker pull registry.baidubce.com/device/paddle-npu:cann80RC1-ubuntu20-x86_64-gcc84-py39
+```
+
+#### 启动容器
+
+ASCEND_RT_VISIBLE_DEVICES 指定可见的 NPU 卡号
+
+```bash linenums="1"
+docker run -it --name paddle-npu-dev -v $(pwd):/work \
+    --privileged --network=host --shm-size=128G -w=/work \
+    -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
+    -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
+    -v /usr/local/dcmi:/usr/local/dcmi \
+    -e ASCEND_RT_VISIBLE_DEVICES="0,1,2,3,4,5,6,7" \
+    registry.baidubce.com/device/paddle-npu:cann80RC1-ubuntu20-x86_64-gcc84-py39 /bin/bash
+```
+
+### 1.2 安装 paddle 包
+
+当前提供 Python3.9 的 wheel 安装包。如有其他 Python 版本需求，可以参考[飞桨官方文档](https://www.paddlepaddle.org.cn/install/quick)自行编译安装。
+
+#### 1. 下载安装 Python3.9 的 wheel 安装包
+
+```bash linenums="1"
+# 注意需要先安装飞桨 cpu 版本
+pip install https://paddle-model-ecology.bj.bcebos.com/paddlex/whl/paddle-device/npu/paddlepaddle-0.0.0-cp39-cp39-linux_x86_64.whl
+pip install https://paddle-model-ecology.bj.bcebos.com/paddlex/whl/paddle-device/npu/paddle_custom_npu-0.0.0-cp39-cp39-linux_x86_64.whl
+```
+
+#### 2. 验证安装包
+
+安装完成之后，运行如下命令。
+
+```bash linenums="1"
+python -c "import paddle; paddle.utils.run_check()"
+```
+
+预期得到如下输出结果
+
+```bash linenums="1"
+Running verify PaddlePaddle program ...
+PaddlePaddle works well on 1 npu.
+PaddlePaddle works well on 8 npus.
+PaddlePaddle is installed successfully! Let's start deep learning with PaddlePaddle now.
+```
diff --git a/docs/version2.x/model/hardware/supported_models.md b/docs/version2.x/model/hardware/supported_models.md
new file mode 100644
index 0000000..9ca4bd5
--- /dev/null
+++ b/docs/version2.x/model/hardware/supported_models.md
@@ -0,0 +1,12 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+# PaddleOCR模型列表
+
+*多硬件安装方式请参考[多硬件安装文档](./install_other_devices.md)*
+
+| 模型名称 | 昇腾NPU |
+| ---------------- | -------- |
+| PP-OCRv4 | √ |
diff --git a/docs/version2.x/model/index.md b/docs/version2.x/model/index.md
new file mode 100644
index 0000000..f9183bb
--- /dev/null
+++ b/docs/version2.x/model/index.md
@@ -0,0 +1,25 @@
+---
+comments: true
+hide:
+    - toc
+---
+
+## PP-OCR 系列模型列表（更新中）
+
+| 模型简介 |  模型名称 | 推荐场景  | 检测模型| 方向分类器   | 识别模型|
+| ----- | -------------- | --------------- | ------- | ------- | ---------- |
+| 中英文超轻量 PP-OCRv4 模型（15.8M） | PP-OCRv4_mobile_det | 移动端&服务器端 | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar) / [训练模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams)         | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv4/chinese/ch_PP-OCRv4_rec_train.tar) |
+| 中英文超轻量 PP-OCRv3 模型（16.2M） | PP-OCRv3_mobile_det  | 移动端&服务器端 | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar) / [训练模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams ) | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar) |
+| 英文超轻量 PP-OCRv3 模型（13.4M）   | en_PP-OCRv3_xx | 移动端&服务器端 | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_distill_train.tar) | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv3_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_train.tar) |
+
+- 超轻量 OCR 系列更多模型下载（包括多语言），可以参考[PP-OCR 系列模型下载](../ppocr/model_list.md)，文档分析相关模型参考[PP-Structure 系列模型下载](../ppstructure/models_list.md)
+
+### PaddleOCR 场景应用模型
+
+| 行业 | 类别 | 亮点| 文档说明| 模型下载 |
+| ---- | ---- | -------- | ---- | ----- |
+| 制造 | 数码管识别   | 数码管数据合成、漏识别调优         | [光功率计数码管字符识别](https://www.paddleocr.ai/v2.10.0/applications/光功率计数码管字符识别.html) | [下载链接](https://www.paddleocr.ai/v2.10.0/applications/overview.html) |
+| 金融 | 通用表单识别 | 多模态通用表单结构化提取           | [多模态表单识别](https://www.paddleocr.ai/v2.10.0/applications/多模态表单识别.html)                                        | [下载链接](https://www.paddleocr.ai/v2.10.0/applications/overview.html) |
+| 交通 | 车牌识别     | 多角度图像处理、轻量模型、端侧部署 | [轻量级车牌识别](https://www.paddleocr.ai/v2.10.0/applications/轻量级车牌识别.html)                                        | [下载链接](https://www.paddleocr.ai/v2.10.0/applications/overview.html) |
+
+- 更多制造、金融、交通行业的主要 OCR 垂类应用模型（如电表、液晶屏、高精度 SVTR 模型等），可参考[场景应用模型下载](https://www.paddleocr.ai/v2.10.0/applications/overview.html)
diff --git a/docs/version2.x/ppocr/blog/PP-OCRv3_introduction.md b/docs/version2.x/ppocr/blog/PP-OCRv3_introduction.md
new file mode 100644
index 0000000..8add9a5
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/PP-OCRv3_introduction.md
@@ -0,0 +1,183 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# PP-OCRv3
+
+## 1. 简介
+
+PP-OCRv3在PP-OCRv2的基础上进一步升级。整体的框架图保持了与PP-OCRv2相同的pipeline，针对检测模型和识别模型进行了优化。其中，检测模块仍基于DB算法优化，而识别模块不再采用CRNN，换成了IJCAI 2022最新收录的文本识别算法[SVTR](https://arxiv.org/abs/2205.00159)，并对其进行产业适配。PP-OCRv3系统框图如下所示（粉色框中为PP-OCRv3新增策略）：
+
+![img](./images/ppocrv3_framework-0052468.png)
+
+从算法改进思路上看，分别针对检测和识别模型，进行了共9个方面的改进：
+
+- 检测模块：
+    - LK-PAN：大感受野的PAN结构；
+    - DML：教师模型互学习策略；
+    - RSE-FPN：残差注意力机制的FPN结构；
+
+- 识别模块：
+    - SVTR_LCNet：轻量级文本识别网络；
+    - GTC：Attention指导CTC训练策略；
+    - TextConAug：挖掘文字上下文信息的数据增广策略；
+    - TextRotNet：自监督的预训练模型；
+    - UDML：联合互学习策略；
+    - UIM：无标注数据挖掘方案。
+
+从效果上看，速度可比情况下，多种场景精度均有大幅提升：
+
+- 中文场景，相对于PP-OCRv2中文模型提升超5%；
+- 英文数字场景，相比于PP-OCRv2英文模型提升11%；
+- 多语言场景，优化80+语种识别效果，平均准确率提升超5%。
+
+## 2. 检测优化
+
+PP-OCRv3检测模型是对PP-OCRv2中的[CML](https://arxiv.org/pdf/2109.03144.pdf)（Collaborative Mutual Learning) 协同互学习文本检测蒸馏策略进行了升级。如下图所示，CML的核心思想结合了①传统的Teacher指导Student的标准蒸馏与 ②Students网络之间的DML互学习，可以让Students网络互学习的同时，Teacher网络予以指导。PP-OCRv3分别针对教师模型和学生模型进行进一步效果优化。其中，在对教师模型优化时，提出了大感受野的PAN结构LK-PAN和引入了DML（Deep Mutual Learning）蒸馏策略；在对学生模型优化时，提出了残差注意力机制的FPN结构RSE-FPN。
+
+![img](./images/ppocrv3_det_cml.png)
+
+消融实验如下：
+
+|序号|策略|模型大小|hmean|速度（cpu + mkldnn)|
+|-|-|-|-|-|
+|baseline teacher|PP-OCR server|49.0M|83.20%|171ms|
+|teacher1|DB-R50-LK-PAN|124.0M|85.00%|396ms|
+|teacher2|DB-R50-LK-PAN-DML|124.0M|86.00%|396ms|
+|baseline student|PP-OCRv2|3.0M|83.20%|117ms|
+|student0|DB-MV3-RSE-FPN|3.6M|84.50%|124ms|
+|student1|DB-MV3-CML（teacher2）|3.0M|84.30%|117ms|
+|student2|DB-MV3-RSE-FPN-CML（teacher2）|3.60M|85.40%|124ms|
+
+测试环境： Intel Gold 6148 CPU，预测时开启MKLDNN加速。
+
+### （1）LK-PAN：大感受野的PAN结构
+
+LK-PAN (Large Kernel PAN) 是一个具有更大感受野的轻量级[PAN](https://arxiv.org/pdf/1803.01534.pdf)结构，核心是将PAN结构的path augmentation中卷积核从`3*3`改为`9*9`。通过增大卷积核，提升特征图每个位置覆盖的感受野，更容易检测大字体的文字以及极端长宽比的文字。使用LK-PAN结构，可以将教师模型的hmean从83.2%提升到85.0%。
+
+![img](./images/LKPAN.png)
+
+### （2）DML：教师模型互学习策略
+
+[DML](https://arxiv.org/abs/1706.00384) （Deep Mutual Learning）互学习蒸馏方法，如下图所示，通过两个结构相同的模型互相学习，可以有效提升文本检测模型的精度。教师模型采用DML策略，hmean从85%提升到86%。将PP-OCRv2中CML的教师模型更新为上述更高精度的教师模型，学生模型的hmean可以进一步从83.2%提升到84.3%。
+
+![img](./images/teacher_dml.png)
+
+### （3）RSE-FPN：残差注意力机制的FPN结构
+
+RSE-FPN（Residual Squeeze-and-Excitation FPN）如下图所示，引入残差结构和通道注意力结构，将FPN中的卷积层更换为通道注意力结构的RSEConv层，进一步提升特征图的表征能力。考虑到PP-OCRv2的检测模型中FPN通道数非常小，仅为96，如果直接用SEblock代替FPN中卷积会导致某些通道的特征被抑制，精度会下降。RSEConv引入残差结构会缓解上述问题，提升文本检测效果。进一步将PP-OCRv2中CML的学生模型的FPN结构更新为RSE-FPN，学生模型的hmean可以进一步从84.3%提升到85.4%。
+
+![img](./images/RSEFPN.png)
+
+## 3. 识别优化
+
+PP-OCRv3的识别模块是基于文本识别算法[SVTR](https://arxiv.org/abs/2205.00159)优化。SVTR不再采用RNN结构，通过引入Transformers结构更加有效地挖掘文本行图像的上下文信息，从而提升文本识别能力。直接将PP-OCRv2的识别模型，替换成SVTR_Tiny，识别准确率从74.8%提升到80.1%（+5.3%），但是预测速度慢了将近11倍，CPU上预测一条文本行，将近100ms。因此，如下图所示，PP-OCRv3采用如下6个优化策略进行识别模型加速。
+
+![img](./images/v3_rec_pipeline.png)
+
+基于上述策略，PP-OCRv3识别模型相比PP-OCRv2，在速度可比的情况下，精度进一步提升4.6%。 具体消融实验如下所示：
+
+| ID | 策略 |  模型大小 | 精度 | 预测耗时（CPU + MKLDNN)|
+|-----|-----|--------|----| --- |
+| 01 | PP-OCRv2 | 8.0M | 74.80% | 8.54ms |
+| 02 | SVTR_Tiny | 21.0M | 80.10% | 97.00ms |
+| 03 | SVTR_LCNet(h32) | 12.0M | 71.90% | 6.60ms |
+| 04 | SVTR_LCNet(h48) | 12.0M | 73.98% | 7.60ms |
+| 05 | + GTC | 12.0M | 75.80% | 7.60ms |
+| 06 | + TextConAug | 12.0M | 76.30% | 7.60ms |
+| 07 | + TextRotNet | 12.0M | 76.90% | 7.60ms |
+| 08 | + UDML | 12.0M | 78.40% | 7.60ms |
+| 09 | + UIM | 12.0M | 79.40% | 7.60ms |
+
+注： 测试速度时，实验01-03输入图片尺寸均为(3,32,320)，04-08输入图片尺寸均为(3,48,320)。在实际预测时，图像为变长输入，速度会有所变化。测试环境： Intel Gold 6148 CPU，预测时开启MKLDNN加速。
+
+### （1）SVTR_LCNet：轻量级文本识别网络
+
+SVTR_LCNet是针对文本识别任务，将基于Transformer的[SVTR](https://arxiv.org/abs/2205.00159)网络和轻量级CNN网络[PP-LCNet](https://arxiv.org/abs/2109.15099) 融合的一种轻量级文本识别网络。使用该网络，预测速度优于PP-OCRv2的识别模型20%，但是由于没有采用蒸馏策略，该识别模型效果略差。此外，进一步将输入图片规范化高度从32提升到48，预测速度稍微变慢，但是模型效果大幅提升，识别准确率达到73.98%（+2.08%），接近PP-OCRv2采用蒸馏策略的识别模型效果。
+
+SVTR_Tiny 网络结构如下所示：
+
+![img](./images/svtr_tiny.png)
+
+由于 MKLDNN 加速库支持的模型结构有限，SVTR 在 CPU+MKLDNN 上相比 PP-OCRv2 慢了10倍。PP-OCRv3 期望在提升模型精度的同时，不带来额外的推理耗时。通过分析发现，SVTR_Tiny 结构的主要耗时模块为 Mixing Block，因此我们对 SVTR_Tiny 的结构进行了一系列优化（详细速度数据请参考下方消融实验表格）:
+
+1. 将 SVTR 网络前半部分替换为 PP-LCNet 的前三个stage，保留4个 Global Mixing Block ，精度为76%，加速69%，网络结构如下所示：
+
+    ![img](./images/svtr_g4.png)
+
+2. 将4个 Global Mixing Block 减小到2个，精度为72.9%，加速69%，网络结构如下所示：
+
+    ![img](./images/svtr_g2.png)
+
+3. 实验发现 Global Mixing Block 的预测速度与输入其特征的shape有关，因此后移 Global Mixing Block 的位置到池化层之后，精度下降为71.9%，速度超越基于CNN结构的PP-OCRv2-baseline 22%，网络结构如下所示：
+
+    ![img](./images/LCNet_SVTR.png)
+
+具体消融实验如下所示：
+
+| ID | 策略 |  模型大小 | 精度 | 速度（CPU + MKLDNN)|
+|-----|-----|--------|----| --- |
+| 01 | PP-OCRv2-baseline | 8.0M | 69.30%  | 8.54ms |
+| 02 | SVTR_Tiny | 21.0M | 80.10% | 97.00ms |
+| 03 | SVTR_LCNet(G4) | 9.2M | 76.00% | 30.00ms |
+| 04 | SVTR_LCNet(G2) | 13.0M | 72.98% | 9.37ms |
+| 05 | SVTR_LCNet(h32) | 12.0M | 71.90% | 6.60ms |
+| 06 | SVTR_LCNet(h48)  | 12.0M | 73.98% | 7.60ms |
+
+注： 测试速度时，01-05输入图片尺寸均为(3,32,320)； PP-OCRv2-baseline 代表没有借助蒸馏方法训练得到的模型
+
+### （2）GTC：Attention指导CTC训练策略
+
+[GTC](https://arxiv.org/pdf/2002.01276.pdf)（Guided Training of CTC），利用Attention模块CTC训练，融合多种文本特征的表达，是一种有效的提升文本识别的策略。使用该策略，预测时完全去除 Attention 模块，在推理阶段不增加任何耗时，识别模型的准确率进一步提升到75.8%（+1.82%）。训练流程如下所示：
+
+![img](./images/GTC.png)
+
+### （3）TextConAug：挖掘文字上下文信息的数据增广策略
+
+TextConAug是一种挖掘文字上下文信息的数据增广策略，主要思想来源于论文[ConCLR](https://www.cse.cuhk.edu.hk/~byu/papers/C139-AAAI2022-ConCLR.pdf)，作者提出ConAug数据增广，在一个batch内对2张不同的图像进行联结，组成新的图像并进行自监督对比学习。PP-OCRv3将此方法应用到有监督的学习任务中，设计了TextConAug数据增强方法，可以丰富训练数据上下文信息，提升训练数据多样性。使用该策略，识别模型的准确率进一步提升到76.3%（+0.5%）。TextConAug示意图如下所示：
+
+![img](./images/recconaug.png)
+
+### （4）TextRotNet：自监督的预训练模型
+
+TextRotNet是使用大量无标注的文本行数据，通过自监督方式训练的预训练模型，参考于论文[STR-Fewer-Labels](https://github.com/ku21fan/STR-Fewer-Labels)。该模型可以初始化SVTR_LCNet的初始权重，从而帮助文本识别模型收敛到更佳位置。使用该策略，识别模型的准确率进一步提升到76.9%（+0.6%）。TextRotNet训练流程如下图所示：
+
+<img src="./images/SSL.png" alt="img" style="zoom:67%;" />
+
+### （5）UDML：联合互学习策略
+
+UDML（Unified-Deep Mutual Learning）联合互学习是PP-OCRv2中就采用的对于文本识别非常有效的提升模型效果的策略。在PP-OCRv3中，针对两个不同的SVTR_LCNet和Attention结构，对他们之间的PP-LCNet的特征图、SVTR模块的输出和Attention模块的输出同时进行监督训练。使用该策略，识别模型的准确率进一步提升到78.4%（+1.5%）。
+
+### （6）UIM：无标注数据挖掘方案
+
+UIM（Unlabeled Images Mining）是一种非常简单的无标注数据挖掘方案。核心思想是利用高精度的文本识别大模型对无标注数据进行预测，获取伪标签，并且选择预测置信度高的样本作为训练数据，用于训练小模型。使用该策略，识别模型的准确率进一步提升到79.4%（+1%）。实际操作中，我们使用全量数据集训练高精度SVTR-Tiny模型（acc=82.5%）进行数据挖掘，点击获取[模型下载地址和使用教程](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.8/applications/%E9%AB%98%E7%B2%BE%E5%BA%A6%E4%B8%AD%E6%96%87%E8%AF%86%E5%88%AB%E6%A8%A1%E5%9E%8B.md)。
+
+<img src="./images/UIM.png" alt="img" style="zoom:67%;" />
+
+## 4. 端到端评估
+
+经过以上优化，最终PP-OCRv3在速度可比情况下，中文场景端到端Hmean指标相比于PP-OCRv2提升5%，效果大幅提升。具体指标如下表所示：
+
+| Model | Hmean |  Model Size (M) | Time Cost (CPU, ms) | Time Cost (T4 GPU, ms) |
+|-----|-----|--------|----| --- |
+| PP-OCR mobile | 50.30% | 8.1 | 356.00  | 116.00 |
+| PP-OCR server | 57.00% | 155.1 | 1056.00 | 200.00 |
+| PP-OCRv2 | 57.60% | 11.6 | 330.00 | 111.00 |
+| PP-OCRv3 | 62.90% | 15.6 | 331.00 | 86.64 |
+
+测试环境：CPU型号为Intel Gold 6148，CPU预测时开启MKLDNN加速。
+
+除了更新中文模型，本次升级也同步优化了英文数字模型，端到端效果提升11%，如下表所示：
+
+| Model | Recall |  Precision | Hmean |
+|-----|-----|--------|----|
+| PP-OCR_en | 38.99% | 45.91% | 42.17%  |
+| PP-OCRv3_en | 50.95% | 55.53% | 53.14% |
+
+同时，也对已支持的80余种语言识别模型进行了升级更新，在有评估集的四种语系识别准确率平均提升5%以上，如下表所示：
+
+| Model | 拉丁语系 |  阿拉伯语系 | 日语 | 韩语 |
+|-----|-----|--------|----| --- |
+| PP-OCR_mul | 69.60% | 40.50% | 38.50%  | 55.40% |
+| PP-OCRv3_mul | 75.20%| 45.37% | 45.80% | 60.10% |
diff --git a/docs/version2.x/ppocr/blog/PP-OCRv4_introduction.md b/docs/version2.x/ppocr/blog/PP-OCRv4_introduction.md
new file mode 100644
index 0000000..bc8457e
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/PP-OCRv4_introduction.md
@@ -0,0 +1,153 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# PP-OCRv4
+
+## 1. 简介
+
+PP-OCRv4在PP-OCRv3的基础上进一步升级。整体的框架图保持了与PP-OCRv3相同的pipeline，针对检测模型和识别模型进行了数据、网络结构、训练策略等多个模块的优化。 PP-OCRv4系统框图如下所示：
+
+![img](./images/ppocrv4_framework.png)
+
+从算法改进思路上看，分别针对检测和识别模型，进行了共10个方面的改进：
+
+* 检测模块：
+    * LCNetV3：精度更高的骨干网络
+    * PFHead：并行head分支融合结构
+    * DSR: 训练中动态增加shrink ratio
+    * CML：添加Student和Teacher网络输出的KL div loss
+
+* 识别模块：
+    * SVTR_LCNetV3：精度更高的骨干网络
+    * Lite-Neck：精简的Neck结构
+    * GTC-NRTR：稳定的Attention指导分支
+    * Multi-Scale：多尺度训练策略
+    * DF: 数据挖掘方案
+    * DKD ：DKD蒸馏策略
+
+从效果上看，速度可比情况下，多种场景精度均有大幅提升：
+
+* 中文场景，相对于PP-OCRv3中文模型提升超4%；
+* 英文数字场景，相比于PP-OCRv3英文模型提升6%；
+* 多语言场景，优化80个语种识别效果，平均准确率提升超8%。
+
+## 2. 检测优化
+
+PP-OCRv4检测模型在PP-OCRv3检测模型的基础上，在网络结构，训练策略，蒸馏策略三个方面做了优化。首先，PP-OCRv4检测模型使用PP-LCNetV3替换MobileNetv3，并提出并行分支融合的PFhead结构；其次，训练时动态调整shrink ratio的比例；最后，PP-OCRv4对CML的蒸馏loss进行优化，进一步提升文字检测效果。
+
+消融实验如下：
+
+|序号|策略|模型大小|hmean|速度（cpu + mkldnn)|
+|-|-|-|-|-|
+|baseline|PP-OCRv3|3.4M|78.84%|69ms|
+|baseline student|PP-OCRv3 student|3.4M|76.22%|69ms|
+|01|+PFHead|3.6M|76.97%|96ms|
+|02|+Dynamic Shrink Ratio|3.6M|78.24%|96ms|
+|03|+PP-LCNetv3|4.8M|79.08%|94ms|
+|03|+CML|4.8M|79.87%|67ms|
+
+测试环境： Intel Gold 6148 CPU，预测引擎使用openvino。
+
+### （1）PFhead：多分支融合Head结构
+
+PFhead结构如下图所示，PFHead在经过第一个转置卷积后，分别进行上采样和转置卷积，上采样的输出通过3x3卷积得到输出结果，然后和转置卷积的分支的结果级联并经过1x1卷积层，最后1x1卷积的结果和转置卷积的结果相加得到最后输出的概率图。PP-OCRv4学生检测模型使用PFhead，hmean从76.22%增加到76.97%。
+
+  ![img](./images/PFHead.png)
+
+### （2）DSR: 收缩比例动态调整策略
+
+动态shrink ratio(dynamic shrink ratio): 在训练中，shrink ratio由固定值调整为动态变化，随着训练epoch的增加，shrink ratio从0.4线性增加到0.6。该策略在PP-OCRv4学生检测模型上，hmean从76.97%提升到78.24%。
+
+### (3) PP-LCNetV3：精度更高的骨干网络
+
+PP-LCNetV3系列模型是PP-LCNet系列模型的延续，覆盖了更大的精度范围，能够适应不同下游任务的需要。PP-LCNetV3系列模型从多个方面进行了优化，提出了可学习仿射变换模块，对重参数化策略、激活函数进行了改进，同时调整了网络深度与宽度。最终，PP-LCNetV3系列模型能够在性能与效率之间达到最佳的平衡，在不同精度范围内取得极致的推理速度。使用PP-LCNetV3替换MobileNetv3 backbone，PP-OCRv4学生检测模型hmean从78.24%提升到79.08%。
+
+### （4）CML: 融合KD的互学习策略
+
+PP-OCRv4检测模型对PP-OCRv3中的CML（Collaborative Mutual Learning) 协同互学习文本检测蒸馏策略进行了优化。如下图所示，在计算Student Model和Teacher Model的distill Loss时，额外添加KL div loss，让两者输出的response maps分布接近，由此进一步提升Student网络的精度，检测Hmean从79.08%增加到79.56%，端到端指标从61.31%增加到61.87%。
+
+  ![img](./images/ppocrv4_det_cml.png)
+
+## 3. 识别优化
+
+PP-OCRv4识别模型在PP-OCRv3的基础上进一步升级。如下图所示，整体的框架图保持了与PP-OCRv3识别模型相同的pipeline，分别进行了数据、网络结构、训练策略等方面的优化。
+
+![img](./images/v4_rec_pipeline.png)
+
+经过如图所示的策略优化，PP-OCRv4识别模型相比PP-OCRv3，在速度可比的情况下，精度进一步提升4%。 具体消融实验如下所示：
+
+| ID | 策略 |  模型大小 | 精度 | 预测耗时（CPU openvino)|
+|-----|-----|--------|----| --- |
+| 01 | PP-OCRv3 | 12M | 71.50% | 8.54ms |
+| 02 | +DF | 12M | 72.70% | 8.54ms |
+| 03 | + LiteNeck + GTC | 9.6M | 73.21% | 9.09ms |
+| 04 | + PP-LCNetV3 | 11M | 74.18% | 9.8ms |
+| 05 | + multi-scale | 11M | 74.20% | 9.8ms |
+| 06 | + TextConAug | 11M | 74.72% | 9.8ms |
+| 08 | + UDML | 11M | 75.45% | 9.8ms |
+
+注： 测试速度时，输入图片尺寸均为(3,48,320)。在实际预测时，图像为变长输入，速度会有所变化。测试环境： Intel Gold 6148 CPU，预测时使用Openvino预测引擎。
+
+### （1）DF：数据挖掘方案
+
+DF(Data Filter) 是一种简单有效的数据挖掘方案。核心思想是利用已有模型预测训练数据，通过置信度和预测结果等信息，对全量的训练数据进行筛选。具体的：首先使用少量数据快速训练得到一个低精度模型，使用该低精度模型对千万级的数据进行预测，去除置信度大于0.95的样本，该部分被认为是对提升模型精度无效的冗余样本。其次使用PP-OCRv3作为高精度模型，对剩余数据进行预测，去除置信度小于0.15的样本，该部分被认为是难以识别或质量很差的样本。
+使用该策略，千万级别训练数据被精简至百万级，模型训练时间从2周减少到5天，显著提升了训练效率，同时精度提升至72.7%(+1.2%)。
+
+![img](./images/DF.png)
+
+### （2）PP-LCNetV3：精度更优的骨干网络
+
+PP-LCNetV3系列模型是PP-LCNet系列模型的延续，覆盖了更大的精度范围，能够适应不同下游任务的需要。PP-LCNetV3系列模型从多个方面进行了优化，提出了可学习仿射变换模块，对重参数化策略、激活函数进行了改进，同时调整了网络深度与宽度。最终，PP-LCNetV3系列模型能够在性能与效率之间达到最佳的平衡，在不同精度范围内取得极致的推理速度。
+
+### （3）Lite-Neck：精简参数的Neck结构
+
+Lite-Neck整体结构沿用PP-OCRv3版本的结构，在参数上稍作精简，识别模型整体的模型大小可从12M降低到8.5M，而精度不变；在CTCHead中，将Neck输出特征的维度从64提升到120，此时模型大小从8.5M提升到9.6M。
+
+### （4）GTC-NRTR：Attention指导CTC训练策略
+
+GTC（Guided Training of CTC），是PP-OCRv3识别模型的最有效的策略之一，融合多种文本特征的表达，有效的提升文本识别精度。在PP-OCRv4中使用训练更稳定的Transformer模型NRTR作为指导分支，相比V3版本中的SAR基于循环神经网络的结构，NRTR基于Transformer实现解码过程泛化能力更强，能有效指导CTC分支学习，解决简单场景下快速过拟合的问题。使用Lite-Neck和GTC-NRTR两个策略，识别精度提升至73.21%(+0.5%)。
+
+![img](./images/ppocrv4_gtc.png)
+
+### （5）Multi-Scale：多尺度训练策略
+
+动态尺度训练策略，是在训练过程中随机resize输入图片的高度，以增强识别模型在端到端串联使用时的鲁棒性。在训练时，每个iter从（32，48，64）三种高度中随机选择一种高度进行resize。实验证明，使用该策略，尽管在识别测试集上准确率没有提升，但在端到端串联评估时，指标提升0.5%。
+
+![img](./images/multi_scale.png)
+
+### （6）DKD：蒸馏策略
+
+识别模型的蒸馏包含两个部分，NRTRhead蒸馏和CTCHead蒸馏;
+
+对于NRTR head，使用了DKD loss蒸馏，拉近学生模型和教师模型的NRTR head logits。最终NRTR head的loss是学生与教师间的DKD loss和与ground truth的cross entropy loss的加权和，用于监督学生模型的backbone训练。通过实验，我们发现加入DKD loss后，计算与ground truth的cross entropy loss时去除label smoothing可以进一步提高精度，因此我们在这里使用的是不带label smoothing的cross entropy loss。
+
+对于CTCHead，由于CTC的输出中存在Blank位，即使教师模型和学生模型的预测结果一样，二者的输出的logits分布也会存在差异，影响教师模型向学生模型的知识传递。PP-OCRv4识别模型蒸馏策略中，将CTC输出logits沿着文本长度维度计算均值，将多字符识别问题转换为多字符分类问题，用于监督CTC Head的训练。使用该策略融合NRTRhead DKD蒸馏策略，指标从74.72%提升到75.45%。
+
+## 4. 端到端评估
+
+经过以上优化，最终PP-OCRv4在速度可比情况下，中文场景端到端Hmean指标相比于PP-OCRv3提升4.5%，效果大幅提升。具体指标如下表所示：
+
+| Model | Hmean |  Model Size (M) | Time Cost (CPU, ms) |
+|-----|-----|--------|----|
+| PP-OCRv3 | 57.99% | 15.6 | 78 |
+| PP-OCRv4 | 62.24% | 15.8 | 76 |
+
+测试环境：CPU型号为Intel Gold 6148，CPU预测时使用openvino。
+
+除了更新中文模型，本次升级也优化了英文数字模型，在自有评估集上文本识别准确率提升6%，如下表所示：
+
+| Model | ACC |
+|-----|-----|
+| PP-OCR_en | 54.38% |
+| PP-OCRv3_en | 64.04% |
+| PP-OCRv4_en | 70.1% |
+
+同时，对已支持的80余种语言识别模型进行了升级更新，在有评估集的四种语系识别准确率平均提升8%以上，如下表所示：
+
+| Model | 拉丁语系 |  阿拉伯语系 | 日语 | 韩语 |
+|-----|-----|--------|----| --- |
+| PP-OCR_mul | 69.60% | 40.50% | 38.50%  | 55.40% |
+| PP-OCRv3_mul | 71.57%| 72.90% | 45.85% | 77.23% |
+| PP-OCRv4_mul | 80.00%| 75.48% | 56.50% | 83.25% |
diff --git a/docs/version2.x/ppocr/blog/clone.en.md b/docs/version2.x/ppocr/blog/clone.en.md
new file mode 100644
index 0000000..f72ce04
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/clone.en.md
@@ -0,0 +1,31 @@
+---
+comments: true
+---
+
+# Project Clone
+
+## 1. Clone PaddleOCR
+
+```bash linenums="1"
+# Recommend
+git clone https://github.com/PaddlePaddle/PaddleOCR
+
+# If you cannot pull successfully due to network problems, you can switch to the mirror hosted on Gitee:
+
+git clone https://gitee.com/paddlepaddle/PaddleOCR
+
+# Note: The mirror on Gitee may not keep in synchronization with the latest project on GitHub. There might be a delay of 3-5 days. Please try GitHub at first.
+```
+
+## 2. Install third-party libraries
+
+```bash linenums="1"
+cd PaddleOCR
+pip3 install -r requirements.txt
+```
+
+If you getting this error `OSError: [WinError 126] The specified module could not be found` when you install shapely on windows.
+
+Please try to download Shapely whl file from [http://www.lfd.uci.edu/~gohlke/pythonlibs/#shapely](http://www.lfd.uci.edu/~gohlke/pythonlibs/#shapely).
+
+Reference: [Solve shapely installation on windows](https://stackoverflow.com/questions/44398265/install-shapely-oserror-winerror-126-the-specified-module-could-not-be-found)
diff --git a/docs/version2.x/ppocr/blog/clone.md b/docs/version2.x/ppocr/blog/clone.md
new file mode 100644
index 0000000..efc4b65
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/clone.md
@@ -0,0 +1,26 @@
+---
+comments: true
+---
+
+# 项目克隆
+
+## 1. 克隆PaddleOCR repo代码
+
+```bash linenums="1"
+git clone https://github.com/PaddlePaddle/PaddleOCR
+```
+
+如果因为网络问题无法pull成功，也可选择使用码云上的托管：
+
+```bash linenums="1"
+git clone https://gitee.com/paddlepaddle/PaddleOCR
+```
+
+注：码云托管代码可能无法实时同步本github项目更新，存在3~5天延时，请优先使用推荐方式。
+
+## 2. 安装第三方库
+
+```bash linenums="1"
+cd PaddleOCR
+pip3 install -r requirements.txt
+```
diff --git a/docs/version2.x/ppocr/blog/config.en.md b/docs/version2.x/ppocr/blog/config.en.md
new file mode 100644
index 0000000..f000b5a
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/config.en.md
@@ -0,0 +1,244 @@
+---
+comments: true
+---
+
+# Configuration
+
+## 1. Optional Parameter List
+
+The following list can be viewed through `--help`
+
+|         FLAG             |     Supported script    |        Use        |      Defaults       |         Note         |
+| :----------------------: | :------------: | :---------------: | :--------------: | :-----------------: |
+|          -c              |      ALL       |  Specify configuration file to use  |  None  |  **Please refer to the parameter introduction for configuration file usage** |
+|          -o              |      ALL       |  set configuration options  |  None  |  Configuration using -o has higher priority than the configuration file selected with -c. E.g: -o Global.use_gpu=false |
+
+## 2. Introduction to Global Parameters of Configuration File
+
+Take rec_chinese_lite_train_v2.0.yml as an example
+
+### Global
+
+|         Parameter             |            Use                |      Defaults       |            Note            |
+| :----------------------: |  :---------------------:   | :--------------:  |   :--------------------:   |
+|      use_gpu             |    Set using GPU or not           |       true        |                \                 |
+|      epoch_num           |    Maximum training epoch number             |       500        |                \                 |
+|      log_smooth_window   |    Log queue length, the median value in the queue each time will be printed           |       20          |                \                 |
+|      print_batch_step    |    Set print log interval         |       10          |                \                 |
+|      save_model_dir      |    Set model save path        |  output/{算法名称}  |                \                 |
+|      save_epoch_step     |    Set model save interval        |       3           |                \                 |
+|      eval_batch_step     |    Set the model evaluation interval        | 2000 or [1000, 2000]        | running evaluation every 2000 iters or evaluation is run every 2000 iterations after the 1000th iteration   |
+|      cal_metric_during_train     |    Set whether to evaluate the metric during the training process. At this time, the metric of the model under the current batch is evaluated        |       true         |                \                 |
+|      load_static_weights     |   Set whether the pre-training model is saved in static graph mode (currently only required by the detection algorithm)        |       true         |                \                 |
+|      pretrained_model    |    Set the path of the pre-trained model      |  ./pretrain_models/CRNN/best_accuracy  |  \          |
+|      checkpoints         |    set model parameter path            |       None        |   Used to load parameters after interruption to continue training|
+|      use_visualdl  |    Set whether to enable visualdl for visual log display |          False        |    [Tutorial](https://www.paddlepaddle.org.cn/paddle/visualdl) |
+|      use_wandb     |    Set whether to enable W&B for visual log display      | False | [Documentation](https://docs.wandb.ai/)
+|      infer_img            |    Set inference image path or folder path     |       ./infer_img | \ |
+|      character_dict_path |    Set dictionary path            |  ./ppocr/utils/ppocr_keys_v1.txt  | If the character_dict_path is None, model can only recognize number and lower letters |
+|      max_text_length     |    Set the maximum length of text        |       25          |                \                 |
+|      use_space_char     |    Set whether to recognize spaces             |        True      |          \               |
+|      label_list          |    Set the angle supported by the direction classifier       |    ['0','180']    |     Only valid in angle classifier model |
+|      save_res_path          |    Set the save address of the test model results       |    ./output/det_db/predicts_db.txt    |     Only valid in the text detection model |
+
+### Optimizer ([ppocr/optimizer](../../../../ppocr/optimizer))
+
+|         Parameter             |            Use            |      Defaults        |            Note             |
+| :---------------------: |  :---------------------:   | :--------------:  |   :--------------------:   |
+|      name        |         Optimizer class name          |  Adam  |  Currently supports `Momentum`,`Adam`,`RMSProp`, see [ppocr/optimizer/optimizer.py](../../../../ppocr/optimizer/optimizer.py)  |
+|      beta1           |    Set the exponential decay rate for the 1st moment estimates  |       0.9         |               \             |
+|      beta2           |    Set the exponential decay rate for the 2nd moment estimates  |     0.999         |               \             |
+|      clip_norm           |    The maximum norm value  |    -         |               \             |
+|      **lr**                |         Set the learning rate decay method       |   -    |       \  |
+|        name    |      Learning rate decay class name   |         Cosine       | Currently supports`Linear`,`Cosine`,`Step`,`Piecewise`, see [ppocr/optimizer/learning_rate.py](../../../../ppocr/optimizer/learning_rate.py) |
+|        learning_rate      |    Set the base learning rate        |       0.001      |  \        |
+|      **regularizer**      |  Set network regularization method        |       -      | \        |
+|        name      |    Regularizer class name      |       L2     |  Currently support `L1`,`L2`, see [ppocr/optimizer/regularizer.py](../../../../ppocr/optimizer/regularizer.py)        |
+|        factor      |    Regularizer coefficient       |       0.00001     |  \        |
+
+### Architecture ([ppocr/modeling](../../../../ppocr/modeling))
+
+In PaddleOCR, the network is divided into four stages: Transform, Backbone, Neck and Head
+
+|         Parameter             |            Use            |      Defaults        |            Note             |
+| :---------------------: |  :---------------------:   | :--------------:  |   :--------------------:   |
+|      model_type        |         Network Type          |  rec  |  Currently support`rec`,`det`,`cls`  |
+|      algorithm           |    Model name  |       CRNN         |               See [algorithm_overview](../../algorithm/overview.en.md) for the support list             |
+|      **Transform**           |    Set the transformation method  |       -       |               Currently only recognition algorithms are supported, see [ppocr/modeling/transform](../../../../ppocr/modeling/transforms) for details            |
+|        name    |      Transformation class name   |         TPS       | Currently supports `TPS` |
+|        num_fiducial      |   Number of TPS control points        |       20      |  Ten on the top and bottom       |
+|        loc_lr      |    Localization network learning rate        |       0.1      |  \      |
+|        model_name      |    Localization network size        |       small      |  Currently support`small`,`large`       |
+|      **Backbone**      |  Set the network backbone class name        |       -      | see [ppocr/modeling/backbones](../../../../ppocr/modeling/backbones)        |
+|        name      |    backbone class name       |       ResNet     | Currently support`MobileNetV3`,`ResNet`        |
+|        layers      |    resnet layers       |       34     |  Currently support18,34,50,101,152,200       |
+|        model_name      |    MobileNetV3 network size       |       small     |  Currently support`small`,`large`       |
+|      **Neck**      |  Set network neck        |       -      | see [ppocr/modeling/necks](../../../../ppocr/modeling/necks)        |
+|        name      |    neck class name       |       SequenceEncoder     | Currently support`SequenceEncoder`,`DBFPN`        |
+|        encoder_type      |    SequenceEncoder encoder type       |       rnn     |  Currently support`reshape`,`fc`,`rnn`       |
+|        hidden_size      |   rnn number of internal units       |       48     |  \      |
+|        out_channels      |   Number of DBFPN output channels       |       256     |  \      |
+|      **Head**      |  Set the network head        |       -      | see [ppocr/modeling/heads](../../../../ppocr/modeling/heads)        |
+|        name      |    head class name       |       CTCHead     | Currently support`CTCHead`,`DBHead`,`ClsHead`        |
+|        fc_decay      |    CTCHead regularization coefficient       |       0.0004     |  \      |
+|        k      |   DBHead binarization coefficient       |       50     |  \      |
+|        class_dim      |   ClsHead output category number       |       2     |  \      |
+
+### Loss ([ppocr/losses](../../../../ppocr/losses))
+
+|         Parameter             |            Use            |      Defaults        |            Note             |
+| :---------------------: |  :---------------------:   | :--------------:  |   :--------------------:   |
+|      name        |         loss class name          |  CTCLoss  |  Currently support`CTCLoss`,`DBLoss`,`ClsLoss`  |
+|      balance_loss        |        Whether to balance the number of positive and negative samples in DBLossloss (using OHEM)         |  True  |  \  |
+|      ohem_ratio        |        The negative and positive sample ratio of OHEM in DBLossloss         |  3  |  \  |
+|      main_loss_type        |        The loss used by shrink_map in DBLossloss        |  DiceLoss  |  Currently support`DiceLoss`,`BCELoss`  |
+|      alpha        |        The coefficient of shrink_map_loss in DBLossloss       |  5  |  \  |
+|      beta        |        The coefficient of threshold_map_loss in DBLossloss       |  10  |  \  |
+
+### PostProcess ([ppocr/postprocess](../../../../ppocr/postprocess))
+
+|         Parameter             |            Use            |      Defaults        |            Note             |
+| :---------------------: |  :---------------------:   | :--------------:  |   :--------------------:   |
+|      name        |         Post-processing class name          |  CTCLabelDecode  |  Currently support`CTCLoss`,`AttnLabelDecode`,`DBPostProcess`,`ClsPostProcess`  |
+|      thresh        |        The threshold for binarization of the segmentation map in DBPostProcess         |  0.3  |  \  |
+|      box_thresh        |        The threshold for filtering output boxes in DBPostProcess. Boxes below this threshold will not be output         |  0.7  |  \  |
+|      max_candidates        |        The maximum number of text boxes output in DBPostProcess        |  1000  |   |
+|      unclip_ratio        |        The unclip ratio of the text box in DBPostProcess       |  2.0  |  \  |
+
+### Metric ([ppocr/metrics](../../../../ppocr/metrics))
+
+|         Parameter             |            Use            |      Defaults        |            Note             |
+| :---------------------: |  :---------------------:   | :--------------:  |   :--------------------:   |
+|      name        |         Metric method name          |  CTCLabelDecode  |  Currently support`DetMetric`,`RecMetric`,`ClsMetric`  |
+|      main_indicator        |        Main indicators, used to select the best model        |  acc |  For the detection method is hmean, the recognition and classification method is acc  |
+
+### Dataset  ([ppocr/data](../../../../ppocr/data))
+
+|         Parameter             |            Use            |      Defaults        |            Note             |
+| :---------------------: |  :---------------------:   | :--------------:  |   :--------------------:   |
+|      **dataset**        |         Return one sample per iteration          |  -  |  -  |
+|      name        |        dataset class name         |  SimpleDataSet |   Currently support`SimpleDataSet`,`LMDBDataSet`  |
+|      data_dir        |        Image folder path        |  ./train_data |  \  |
+|      label_file_list        |        Groundtruth file path         |  ["./train_data/train_list.txt"] | This parameter is not required when dataset is LMDBDataSet   |
+|      ratio_list        |        Ratio of data set         |  [1.0] | If there are two train_lists in label_file_list and ratio_list is [0.4,0.6], 40% will be sampled from train_list1, and 60% will be sampled from train_list2 to combine the entire dataset   |
+|      transforms        |        List of methods to transform images and labels         |  [DecodeImage,CTCLabelEncode,RecResizeImg,KeepKeys] |   see [ppocr/data/imaug](../../../../ppocr/data/imaug)  |
+|      **loader**        |        dataloader related         |  - |   |
+|      shuffle        |        Does each epoch disrupt the order of the data set         |  True | \  |
+|      batch_size_per_card        |        Single card batch size during training         |  256 | \  |
+|      drop_last        |        Whether to discard the last incomplete mini-batch because the number of samples in the data set cannot be divisible by batch_size        |  True | \  |
+|      num_workers        |        The number of sub-processes used to load data, if it is 0, the sub-process is not started, and the data is loaded in the main process       |  8 | \  |
+
+### Weights & Biases ([W&B](../../../../ppocr/utils/loggers/wandb_logger.py))
+
+|         Parameter             |            Use            |      Defaults        |            Note             |
+| :---------------------: |  :---------------------:   | :--------------:  |   :--------------------:   |
+|          project              |     Project to which the run is to be logged | uncategorized | \
+|          name                 |     Alias/Name of the run | Randomly generated by wandb | \
+|          id                   |     ID of the run    | Randomly generated by wandb     | \
+|          entity               | User or team to which the run is being logged         | The logged in user | \
+|          save_dir             | local directory in which all the models and other data is saved | wandb | \
+|          config               | model configuration | None | \
+
+## 3. Multilingual Config File Generation
+
+PaddleOCR currently supports recognition for 80 languages (besides Chinese). A multi-language configuration file template is
+provided under the path `configs/rec/multi_languages`: [rec_multi_language_lite_train.yml](../../../../configs/rec/multi_language/rec_multi_language_lite_train.yml).
+
+There are two ways to create the required configuration file:
+
+1. Automatically generated by script
+
+Script [generate_multi_language_configs.py](../../../../configs/rec/multi_language/generate_multi_language_configs.py) can help you generate configuration files for multi-language models.
+
+- Take Italian as an example, if your data is prepared in the following format:
+
+    ```text linenums="1"
+    |-train_data
+        |- it_train.txt # train_set label
+        |- it_val.txt # val_set label
+        |- data
+            |- word_001.jpg
+            |- word_002.jpg
+            |- word_003.jpg
+            | ...
+    ```
+
+    You can use the default parameters to generate a configuration file:
+
+    ```bash linenums="1"
+    # The code needs to be run in the specified directory
+    cd PaddleOCR/configs/rec/multi_language/
+    # Set the configuration file of the language to be generated through the -l or --language parameter.
+    # This command will write the default parameters into the configuration file
+    python3 generate_multi_language_configs.py -l it
+    ```
+
+- If your data is placed in another location, or you want to use your own dictionary, you can generate the configuration file by specifying the relevant parameters:
+
+    ```bash linenums="1"
+    # -l or --language field is required
+    # --train to modify the training set
+    # --val to modify the validation set
+    # --data_dir to modify the data set directory
+    # --dict to modify the dict path
+    # -o to modify the corresponding default parameters
+    cd PaddleOCR/configs/rec/multi_language/
+    python3 generate_multi_language_configs.py -l it \  # language
+    --train {path/of/train_label.txt} \ # path of train_label
+    --val {path/of/val_label.txt} \     # path of val_label
+    --data_dir {train_data/path} \      # root directory of training data
+    --dict {path/of/dict} \             # path of dict
+    -o Global.use_gpu=False             # whether to use gpu
+    ...
+
+    ```
+
+Italian is made up of Latin letters, so after executing the command, you will get the rec_latin_lite_train.yml.
+
+2. Manually modify the configuration file
+
+   You can also manually modify the following fields in the template:
+
+   ```yaml linenums="1"
+    Global:
+      use_gpu: True
+      epoch_num: 500
+      ...
+      character_dict_path:  {path/of/dict} # path of dict
+
+   Train:
+      dataset:
+        name: SimpleDataSet
+        data_dir: train_data/ # root directory of training data
+        label_file_list: ["./train_data/train_list.txt"] # train label path
+      ...
+
+   Eval:
+      dataset:
+        name: SimpleDataSet
+        data_dir: train_data/ # root directory of val data
+        label_file_list: ["./train_data/val_list.txt"] # val label path
+      ...
+
+   ```
+
+Currently, the multi-language algorithms supported by PaddleOCR are:
+
+| Configuration file |  Algorithm name |   backbone |   trans   |   seq      |     pred     |  language |
+| :--------: |  :-------:   | :-------:  |   :-------:   |   :-----:   |  :-----:   | :-----:  |
+| rec_chinese_cht_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | chinese traditional  |
+| rec_en_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | English(Case sensitive)   |
+| rec_french_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | French |
+| rec_ger_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | German   |
+| rec_japan_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | Japanese |
+| rec_korean_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | Korean  |
+| rec_latin_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | Latin  |
+| rec_arabic_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | arabic |
+| rec_cyrillic_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | cyrillic   |
+| rec_devanagari_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | devanagari  |
+
+For more supported languages, please refer to: [Multi-language model](./multi_languages.en.md)
+
+The multi-language model training method is the same as the Chinese model. The training data set is 100w synthetic data. A small amount of fonts and test data can be downloaded using the following two methods.
+
+- [Baidu Netdisk](https://pan.baidu.com/s/1bS_u207Rm7YbY33wOECKDA), Extraction code:frgi.
diff --git a/docs/version2.x/ppocr/blog/config.md b/docs/version2.x/ppocr/blog/config.md
new file mode 100644
index 0000000..9a2dc76
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/config.md
@@ -0,0 +1,222 @@
+---
+comments: true
+---
+
+# 配置文件内容与生成
+
+## 1. 可选参数列表
+
+以下列表可以通过`--help`查看
+
+|         FLAG             |     支持脚本    |        用途        |      默认值       |         备注         |
+| :----------------------: | :------------: | :---------------: | :--------------: | :-----------------: |
+|          -c              |      ALL       |  指定配置文件  |  None  |  **配置模块说明请参考 参数介绍** |
+|          -o              |      ALL       |  设置配置文件里的参数内容  |  None  |  使用-o配置相较于-c选择的配置文件具有更高的优先级。例如：`-o Global.use_gpu=false`  |
+
+## 2. 配置文件参数介绍
+
+以 `rec_chinese_lite_train_v2.0.yml` 为例
+
+### Global
+
+|         字段             |            用途                |      默认值       |            备注            |
+| :----------------------: |  :---------------------:   | :--------------:  |   :--------------------:   |
+|      use_gpu             |    设置代码是否在gpu运行           |       true        |                \                 |
+|      epoch_num           |    最大训练epoch数             |       500        |                \                 |
+|      log_smooth_window   |    log队列长度，每次打印输出队列里的中间值            |       20          |                \                 |
+|      print_batch_step    |    设置打印log间隔         |       10          |                \                 |
+|      save_model_dir      |    设置模型保存路径        |  output/{算法名称}  |                \                 |
+|      save_epoch_step     |    设置模型保存间隔        |       3           |                \                 |
+|      eval_batch_step     |    设置模型评估间隔        | 2000 或 [1000, 2000]        | 2000 表示每2000次迭代评估一次，[1000， 2000]表示从1000次迭代开始，每2000次评估一次   |
+|      cal_metric_during_train     |    设置是否在训练过程中评估指标，此时评估的是模型在当前batch下的指标        |       true         |                \                 |
+|      load_static_weights     |   设置预训练模型是否是静态图模式保存(目前仅检测算法需要)        |       true         |                \                 |
+|      pretrained_model    |    设置加载预训练模型路径      |  ./pretrain_models/CRNN/best_accuracy  |  \          |
+|      checkpoints         |    加载模型参数路径            |       None        |    用于中断后加载参数继续训练 |
+|      use_visualdl  |    设置是否启用visualdl进行可视化log展示 |          False        |    [教程地址](https://www.paddlepaddle.org.cn/paddle/visualdl) |
+|      infer_img            |    设置预测图像路径或文件夹路径     |       ./infer_img | \||
+|      character_dict_path |    设置字典路径            |  ./ppocr/utils/ppocr_keys_v1.txt  | 如果为空，则默认使用小写字母+数字作为字典 |
+|      max_text_length     |    设置文本最大长度        |       25          |                \                 |
+|      use_space_char     |    设置是否识别空格             |        True      |          \|               |
+|      label_list          |    设置方向分类器支持的角度       |    ['0','180']    |     仅在方向分类器中生效 |
+|      save_res_path          |    设置检测模型的结果保存地址       |    ./output/det_db/predicts_db.txt    |     仅在检测模型中生效 |
+
+### Optimizer ([ppocr/optimizer](../../../../ppocr/optimizer))
+
+|         字段             |      用途       |      默认值      |            备注             |
+| :---------------------: |:-------------:|:-------------:|   :--------------------:   |
+|      name        |     优化器类名     |     Adam      |  目前支持`Momentum`,`Adam`,`RMSProp`, 见[ppocr/optimizer/optimizer.py](../../../../ppocr/optimizer/optimizer.py)  |
+|      beta1           | 设置一阶矩估计的指数衰减率 |      0.9      |               \             |
+|      beta2           | 设置二阶矩估计的指数衰减率 |     0.999     |               \             |
+|      clip_norm           |  所允许的二范数最大值   |               |               \             |
+|      **lr**                | 设置学习率decay方式  |       -       |       \  |
+|        name    |  学习率decay类名   |    Cosine     | 目前支持`Linear`,`Cosine`,`Step`,`Piecewise`, 见[ppocr/optimizer/learning_rate.py](../../../../ppocr/optimizer/learning_rate.py) |
+|        learning_rate      |     基础学习率     |     0.001     |  \        |
+|      **regularizer**      |   设置网络正则化方式   |       -       | \        |
+|        name      |     正则化类名     |      L2       | 目前支持`L1`,`L2`, 见[ppocr/optimizer/regularizer.py](../../../../ppocr/optimizer/regularizer.py)        |
+|        factor      |     正则化系数     |       0.00001        |  \        |
+
+### Architecture ([ppocr/modeling](../../../../ppocr/modeling))
+
+在PaddleOCR中，网络被划分为Transform,Backbone,Neck和Head四个阶段
+
+|         字段             |            用途            |      默认值        |            备注             |
+| :---------------------: |  :---------------------:   | :--------------:  |   :--------------------:   |
+|      model_type        |         网络类型          |  rec  |  目前支持`rec`,`det`,`cls`  |
+|      algorithm           |    模型名称  |       CRNN         |               支持列表见[algorithm_overview](../../algorithm/overview.en.md)             |
+|      **Transform**           |    设置变换方式  |       -       |               目前仅rec类型的算法支持, 具体见[ppocr/modeling/transforms](../../../../ppocr/modeling/transforms)              |
+|        name    |      变换方式类名   |         TPS       | 目前支持`TPS` |
+|        num_fiducial      |    TPS控制点数        |       20      |  上下边各十个       |
+|        loc_lr      |    定位网络学习率        |       0.1      |  \      |
+|        model_name      |    定位网络大小        |       small      |  目前支持`small`,`large`       |
+|      **Backbone**      |  设置网络backbone类名        |       -      | 具体见[ppocr/modeling/backbones](../../../../ppocr/modeling/backbones)        |
+|        name      |    backbone类名       |       ResNet     | 目前支持`MobileNetV3`,`ResNet`        |
+|        layers      |    resnet层数       |       34     |  支持18,34,50,101,152,200       |
+|        model_name      |    MobileNetV3 网络大小       |       small     |  支持`small`,`large`       |
+|      **Neck**      |  设置网络neck        |       -      | 具体见[ppocr/modeling/necks](../../../../ppocr/modeling/necks)        |
+|        name      |    neck类名       |       SequenceEncoder     | 目前支持`SequenceEncoder`,`DBFPN`        |
+|        encoder_type      |    SequenceEncoder编码器类型       |       rnn     |  支持`reshape`,`fc`,`rnn`       |
+|        hidden_size      |   rnn内部单元数       |       48     |  \      |
+|        out_channels      |   DBFPN输出通道数       |       256     |  \      |
+|      **Head**      |  设置网络Head        |       -      | 具体见[ppocr/modeling/heads](../../../../ppocr/modeling/heads)        |
+|        name      |    head类名       |       CTCHead     | 目前支持`CTCHead`,`DBHead`,`ClsHead`        |
+|        fc_decay      |    CTCHead正则化系数       |       0.0004     |  \      |
+|        k      |   DBHead二值化系数       |       50     |  \      |
+|        class_dim      |   ClsHead输出分类数       |       2     |  \      |
+
+### Loss ([ppocr/losses](../../../../ppocr/losses))
+
+|         字段             |            用途            |      默认值        |            备注             |
+| :---------------------: |  :---------------------:   | :--------------:  |   :--------------------:   |
+|      name        |         网络loss类名          |  CTCLoss  |  目前支持`CTCLoss`,`DBLoss`,`ClsLoss`  |
+|      balance_loss        |        DBLossloss中是否对正负样本数量进行均衡(使用OHEM)         |  True  |  \  |
+|      ohem_ratio        |        DBLossloss中的OHEM的负正样本比例         |  3  |  \  |
+|      main_loss_type        |        DBLossloss中shrink_map所采用的loss        |  DiceLoss  |  支持`DiceLoss`,`BCELoss`  |
+|      alpha        |        DBLossloss中shrink_map_loss的系数       |  5  |  \  |
+|      beta        |        DBLossloss中threshold_map_loss的系数       |  10  |  \  |
+
+### PostProcess ([ppocr/postprocess](../../../../ppocr/postprocess))
+
+|         字段             |            用途            |      默认值        |            备注             |
+| :---------------------: |  :---------------------:   | :--------------:  |   :--------------------:   |
+|      name        |         后处理类名          |  CTCLabelDecode  |  目前支持`CTCLoss`,`AttnLabelDecode`,`DBPostProcess`,`ClsPostProcess`  |
+|      thresh        |        DBPostProcess中分割图进行二值化的阈值         |  0.3  |  \  |
+|      box_thresh        |        DBPostProcess中对输出框进行过滤的阈值，低于此阈值的框不会输出         |  0.7  |  \  |
+|      max_candidates        |        DBPostProcess中输出的最大文本框数量        |  1000  |   |
+|      unclip_ratio        |        DBPostProcess中对文本框进行放大的比例       |  2.0  |  \  |
+
+### Metric ([ppocr/metrics](../../../../ppocr/metrics))
+
+|         字段             |            用途            |      默认值        |            备注             |
+| :---------------------: |  :---------------------:   | :--------------:  |   :--------------------:   |
+|      name        |         指标评估方法名称          |  CTCLabelDecode  |  目前支持`DetMetric`,`RecMetric`,`ClsMetric`  |
+|      main_indicator        |        主要指标,用于选取最优模型         |  acc |  对于检测方法为hmean，识别和分类方法为acc  |
+
+### Dataset  ([ppocr/data](../../../../ppocr/data))
+
+|         字段             |            用途            |      默认值        |            备注             |
+| :---------------------: |  :---------------------:   | :--------------:  |   :--------------------:   |
+|      **dataset**        |         每次迭代返回一个样本          |  -  |  -  |
+|      name        |        dataset类名         |  SimpleDataSet |  目前支持`SimpleDataSet`和`LMDBDataSet`  |
+|      data_dir        |        数据集图片存放路径         |  ./train_data |  \  |
+|      label_file_list        |        数据标签路径         |  ["./train_data/train_list.txt"] | dataset为LMDBDataSet时不需要此参数   |
+|      ratio_list        |        数据集的比例         |  [1.0] | 若label_file_list中有两个train_list，且ratio_list为[0.4,0.6]，则从train_list1中采样40%，从train_list2中采样60%组合整个dataset   |
+|      transforms        |        对图片和标签进行变换的方法列表         |  [DecodeImage,CTCLabelEncode,RecResizeImg,KeepKeys] |   见[ppocr/data/imaug](../../../../ppocr/data/imaug)  |
+|      **loader**        |        dataloader相关         |  - |   |
+|      shuffle        |        每个epoch是否将数据集顺序打乱         |  True | \  |
+|      batch_size_per_card        |        训练时单卡batch size         |  256 | \  |
+|      drop_last        |        是否丢弃因数据集样本数不能被 batch_size 整除而产生的最后一个不完整的mini-batch        |  True | \  |
+|      num_workers        |        用于加载数据的子进程个数，若为0即为不开启子进程，在主进程中进行数据加载        |  8 | \  |
+
+## 3. 多语言配置文件生成
+
+PaddleOCR目前已支持80种（除中文外）语种识别，`configs/rec/multi_languages` 路径下提供了一个多语言的配置文件模版: [rec_multi_language_lite_train.yml](../../../../configs/rec/multi_language/rec_multi_language_lite_train.yml)。
+
+您有两种方式创建所需的配置文件：
+
+1. 通过脚本自动生成
+
+[generate_multi_language_configs.py](../../../../configs/rec/multi_language/generate_multi_language_configs.py) 可以帮助您生成多语言模型的配置文件
+
+* 以意大利语为例，如果您的数据是按如下格式准备的：
+
+    ```text linenums="1"
+    |-train_data
+        |- it_train.txt # 训练集标签
+        |- it_val.txt # 验证集标签
+        |- data
+            |- word_001.jpg
+            |- word_002.jpg
+            |- word_003.jpg
+            | ...
+    ```
+
+    可以使用默认参数，生成配置文件：
+
+    ```bash linenums="1"
+    # 该代码需要在指定目录运行
+    cd PaddleOCR/configs/rec/multi_language/
+    # 通过-l或者--language参数设置需要生成的语种的配置文件，该命令会将默认参数写入配置文件
+    python3 generate_multi_language_configs.py -l it
+    ```
+
+* 如果您的数据放置在其他位置，或希望使用自己的字典，可以通过指定相关参数来生成配置文件:
+
+    ```bash linenums="1"
+    # -l或者--language字段是必须的
+    # --train修改训练集，--val修改验证集，--data_dir修改数据集目录，--dict修改字典路径， -o修改对应默认参数
+    cd PaddleOCR/configs/rec/multi_language/
+    python3 generate_multi_language_configs.py -l it \  # 语种
+    --train {path/of/train_label.txt} \ # 训练标签文件的路径
+    --val {path/of/val_label.txt} \     # 验证集标签文件的路径
+    --data_dir {train_data/path} \      # 训练数据的根目录
+    --dict {path/of/dict} \             # 字典文件路径
+    -o Global.use_gpu=False             # 是否使用gpu
+    ...
+
+    ```
+
+意大利文由拉丁字母组成，因此执行完命令后会得到名为 rec_latin_lite_train.yml 的配置文件。
+
+2. 手动修改配置文件
+
+   您也可以手动修改模版中的以下几个字段得到配置文件:
+
+   ```yaml linenums="1"
+    Global:
+      use_gpu: True
+      epoch_num: 500
+      ...
+      character_dict_path:  {path/of/dict} # 字典文件所在路径
+
+   Train:
+      dataset:
+        name: SimpleDataSet
+        data_dir: train_data/ # 数据存放根目录
+        label_file_list: ["./train_data/train_list.txt"] # 训练集label路径
+      ...
+
+   Eval:
+      dataset:
+        name: SimpleDataSet
+        data_dir: train_data/ # 数据存放根目录
+        label_file_list: ["./train_data/val_list.txt"] # 验证集label路径
+      ...
+
+   ```
+
+目前PaddleOCR支持的多语言算法有：
+
+| 配置文件 |  算法名称 |   backbone |   trans   |   seq      |     pred     |  language |
+| :--------: |  :-------:   | :-------:  |   :-------:   |   :-----:   |  :-----:   | :-----:  |
+| rec_chinese_cht_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 中文繁体  |
+| rec_en_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 英语（区分大小写）   |
+| rec_french_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 法语 |
+| rec_ger_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 德语   |
+| rec_japan_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 日语  |
+| rec_korean_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 韩语  |
+| rec_latin_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 拉丁字母  |
+| rec_arabic_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 阿拉伯字母 |
+| rec_cyrillic_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 斯拉夫字母  |
+| rec_devanagari_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 梵文字母  |
+
+更多支持语种请参考: [多语言模型](./multi_languages.md)
diff --git a/docs/version2.x/ppocr/blog/customize.en.md b/docs/version2.x/ppocr/blog/customize.en.md
new file mode 100644
index 0000000..f46d21c
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/customize.en.md
@@ -0,0 +1,39 @@
+---
+comments: true
+---
+
+# HOW TO MAKE YOUR OWN LIGHTWEIGHT OCR MODEL?
+
+The process of making a customized ultra-lightweight OCR models can be divided into three steps: training text detection model, training text recognition model, and concatenate the predictions from previous steps.
+
+## STEP1: TRAIN TEXT DETECTION MODEL
+
+PaddleOCR provides two text detection algorithms: EAST and DB. Both support MobileNetV3 and ResNet50_vd backbone networks, select the corresponding configuration file as needed and start training. For example, to train with MobileNetV3 as the backbone network for DB detection model :
+
+```bash linenums="1"
+python3 tools/train.py -c configs/det/det_mv3_db.yml 2>&1 | tee det_db.log
+```
+
+For more details about data preparation and training tutorials, refer to the documentation [Text detection model training/evaluation/prediction](../model_train/detection.en.md)
+
+## STEP2: TRAIN TEXT RECOGNITION MODEL
+
+PaddleOCR provides four text recognition algorithms: CRNN, Rosetta, STAR-Net, and RARE. They all support two backbone networks: MobileNetV3 and ResNet34_vd, select the corresponding configuration files as needed to start training. For example, to train a CRNN recognition model that uses MobileNetV3 as the backbone network:
+
+```bash linenums="1"
+python3 tools/train.py -c configs/rec/rec_chinese_lite_train.yml 2>&1 | tee rec_ch_lite.log
+```
+
+For more details about data preparation and training tutorials, refer to the documentation [Text recognition model training/evaluation/prediction](../model_train/recognition.en.md)
+
+## STEP3: CONCATENATE PREDICTIONS
+
+PaddleOCR provides a concatenation tool for detection and recognition models, which can connect any trained detection model and any recognition model into a two-stage text recognition system. The input image goes through four main stages: text detection, text rectification, text recognition, and score filtering to output the text position and recognition results, and at the same time, you can choose to visualize the results.
+
+When performing prediction, you need to specify the path of a single image or a image folder through the parameter `image_dir`, the parameter `det_model_dir` specifies the path of detection model, and the parameter `rec_model_dir` specifies the path of recognition model. The visualized results are saved to the `./inference_results` folder by default.
+
+```bash linenums="1"
+python3 tools/infer/predict_system.py --image_dir="./doc/imgs/11.jpg" --det_model_dir="./inference/det/"  --rec_model_dir="./inference/rec/"
+```
+
+For more details about text detection and recognition concatenation, please refer to the document [Inference](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.9/doc/doc_en/inference_en.md)
diff --git a/docs/version2.x/ppocr/blog/customize.md b/docs/version2.x/ppocr/blog/customize.md
new file mode 100644
index 0000000..51a4e31
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/customize.md
@@ -0,0 +1,39 @@
+---
+comments: true
+---
+
+# 如何生产自定义超轻量模型？
+
+生产自定义的超轻量模型可分为三步：训练文本检测模型、训练文本识别模型、模型串联预测。
+
+## step1：训练文本检测模型
+
+PaddleOCR提供了EAST、DB两种文本检测算法，均支持MobileNetV3、ResNet50_vd两种骨干网络，根据需要选择相应的配置文件，启动训练。例如，训练使用MobileNetV3作为骨干网络的DB检测模型（即超轻量模型使用的配置）：
+
+```bash linenums="1"
+python3 tools/train.py -c configs/det/det_mv3_db.yml 2>&1 | tee det_db.log
+```
+
+更详细的数据准备和训练教程参考文档教程中[文本检测模型训练/评估/预测](../model_train/detection.md)。
+
+## step2：训练文本识别模型
+
+PaddleOCR提供了CRNN、Rosetta、STAR-Net、RARE四种文本识别算法，均支持MobileNetV3、ResNet34_vd两种骨干网络，根据需要选择相应的配置文件，启动训练。例如，训练使用MobileNetV3作为骨干网络的CRNN识别模型（即超轻量模型使用的配置）：
+
+```bash linenums="1"
+python3 tools/train.py -c configs/rec/rec_chinese_lite_train.yml 2>&1 | tee rec_ch_lite.log
+```
+
+更详细的数据准备和训练教程参考文档教程中[文本识别模型训练/评估/预测](../model_train/recognition.md)。
+
+## step3：模型串联预测
+
+PaddleOCR提供了检测和识别模型的串联工具，可以将训练好的任一检测模型和任一识别模型串联成两阶段的文本识别系统。输入图像经过文本检测、检测框矫正、文本识别、得分过滤四个主要阶段输出文本位置和识别结果，同时可选择对结果进行可视化。
+
+在执行预测时，需要通过参数image_dir指定单张图像或者图像集合的路径、参数det_model_dir指定检测inference模型的路径和参数rec_model_dir指定识别inference模型的路径。可视化识别结果默认保存到 ./inference_results 文件夹里面。
+
+```bash linenums="1"
+python3 tools/infer/predict_system.py --image_dir="./doc/imgs/11.jpg" --det_model_dir="./inference/det/"  --rec_model_dir="./inference/rec/"
+```
+
+更多的文本检测、识别串联推理使用方式请参考文档教程中的[基于预测引擎推理](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.9/doc/doc_ch/inference.md)。
diff --git a/docs/version2.x/ppocr/blog/distributed_training.en.md b/docs/version2.x/ppocr/blog/distributed_training.en.md
new file mode 100644
index 0000000..19e84bb
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/distributed_training.en.md
@@ -0,0 +1,65 @@
+---
+comments: true
+---
+
+# Distributed training
+
+## Introduction
+
+The high performance of distributed training is one of the core advantages of PaddlePaddle. In the classification task, distributed training can achieve almost linear speedup ratio. Generally, OCR training task need massive training data. Such as recognition, PP-OCR v2.0 model is trained based on 1800W dataset, which is very time-consuming if using single machine. Therefore, the distributed training is used in PaddleOCR to speedup the training task. For more information about distributed training, please refer to [distributed training quick start tutorial](https://fleet-x.readthedocs.io/en/latest/paddle_fleet_rst/parameter_server/ps_quick_start.html).
+
+## Quick Start
+
+### Training with single machine
+
+Take recognition as an example. After the data is prepared locally, start the training task with the interface of `paddle.distributed.launch`. The start command as follows:
+
+```bash linenums="1"
+python3 -m paddle.distributed.launch \
+    --log_dir=./log/ \
+    --gpus "0,1,2,3,4,5,6,7" \
+    tools/train.py \
+    -c configs/rec/rec_mv3_none_bilstm_ctc.yml
+```
+
+### Training with multi machine
+
+Compared with single machine, training with multi machine only needs to add the parameter `--ips` to start command, which represents the IP list of machines used for distributed training, and the IP of different machines are separated by commas. The start command as follows:
+
+```bash linenums="1"
+ip_list="192.168.0.1,192.168.0.2"
+python3 -m paddle.distributed.launch \
+    --log_dir=./log/ \
+    --ips="${ip_list}" \
+    --gpus="0,1,2,3,4,5,6,7" \
+    tools/train.py \
+    -c configs/rec/rec_mv3_none_bilstm_ctc.yml
+```
+
+**Notice:**
+
+* The IP addresses of different machines need to be separated by commas, which can be queried through `ifconfig` or `ipconfig`.
+* Different machines need to be set to be secret free and can `ping` success with others directly, otherwise communication cannot establish between them.
+* The code, data and start command between different machines must be completely consistent and then all machines need to run start command. The first machine in the `ip_list` is set to `trainer0`, and so on.
+
+## Performance comparison
+
+We conducted model training on 2x8 P40 GPUs. Accuracy, training time, and multi machine acceleration ratio of different models are shown below.
+
+| Model    | Configuration | Configuration   | 8 GPU training time / Accuracy | 3x8 GPU training time / Accuracy | Acceleration ratio  |
+|:------:|:-----:|:--------:|:--------:|:--------:|:-----:|
+| CRNN | [rec_chinese_lite_train_v2.0.yml](../../../../configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml) |  260k Chinese dataset | 2.50d/66.70%   | 1.67d/67.00%  | **1.5** |
+
+We conducted model training on 3x8 V100 GPUs. Accuracy, training time, and multi machine acceleration ratio of different models are shown below.
+
+| Model    | Configuration | Configuration   | 8 GPU training time / Accuracy | 3x8 GPU training time / Accuracy | Acceleration ratio  |
+|:------:|:-----:|:--------:|:--------:|:--------:|:-----:|
+| SLANet | [SLANet.yml](../../../../configs/table/SLANet.yml) |  PubTabNet | 49.80h/76.20%   | 19.75h/74.77%  | **2.52** |
+
+Note: when training with 3x8 GPUs, the single card batch size is unchanged compared with the 1x8 GPUs' training process, and the learning rate is multiplied by 2 (if it is multiplied by 3 by default, the accuracy is only 73.42%).
+
+We conducted model training on 4x8 V100 GPUs. Accuracy, training time, and multi machine acceleration ratio of different models are shown below.
+
+| Model    | Configuration | Configuration   | 8 GPU training time / Accuracy | 4x8 GPU training time / Accuracy | Acceleration ratio  |
+|:------:|:-----:|:--------:|:--------:|:--------:|:-----:|
+| SVTR | [PP-OCRv3_mobile_rec_distillation.yml](../../../../configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec_distillation.yml) |  PP-OCRv3_rec data | 10d/-   | 2.84d/74.00%  | **3.5** |
diff --git a/docs/version2.x/ppocr/blog/distributed_training.md b/docs/version2.x/ppocr/blog/distributed_training.md
new file mode 100644
index 0000000..461c70c
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/distributed_training.md
@@ -0,0 +1,67 @@
+---
+comments: true
+---
+
+# 分布式训练
+
+## 简介
+
+* 分布式训练的高性能，是飞桨的核心优势技术之一，在分类任务上，分布式训练可以达到几乎线性的加速比。OCR训练任务中往往包含大量训练数据，以识别为例，ppocrv2.0模型在训练时使用了1800W数据，如果使用单机训练，会非常耗时。因此，PaddleOCR中使用分布式训练接口完成训练任务，同时支持单机训练与多机训练。更多关于分布式训练的方法与文档可以参考：[分布式训练快速开始教程](https://fleet-x.readthedocs.io/en/latest/paddle_fleet_rst/parameter_server/ps_quick_start.html)。
+
+## 使用方法
+
+### 单机训练
+
+* 以识别为例，本地准备好数据之后，使用`paddle.distributed.launch`的接口启动训练任务即可。下面为运行代码示例。
+
+```bash linenums="1"
+python3 -m paddle.distributed.launch \
+    --log_dir=./log/ \
+    --gpus "0,1,2,3,4,5,6,7" \
+    tools/train.py \
+    -c configs/rec/rec_mv3_none_bilstm_ctc.yml
+```
+
+### 多机训练
+
+* 相比单机训练，多机训练时，只需要添加`--ips`的参数，该参数表示需要参与分布式训练的机器的ip列表，不同机器的ip用逗号隔开。下面为运行代码示例。
+
+```bash linenums="1"
+ip_list="192.168.0.1,192.168.0.2"
+python3 -m paddle.distributed.launch \
+    --log_dir=./log/ \
+    --ips="${ip_list}" \
+    --gpus="0,1,2,3,4,5,6,7" \
+    tools/train.py \
+    -c configs/rec/rec_mv3_none_bilstm_ctc.yml
+```
+
+**注：**
+
+* 不同机器的ip信息需要用逗号隔开，可以通过`ifconfig`或者`ipconfig`查看。
+* 不同机器之间需要做免密设置，且可以直接ping通，否则无法完成通信。
+* 不同机器之间的代码、数据与运行命令或脚本需要保持一致，且所有的机器上都需要运行设置好的训练命令或者脚本。最终`ip_list`中的第一台机器的第一块设备是trainer0，以此类推。
+
+## 性能效果测试
+
+在2机8卡P40的机器上进行模型训练，不同模型的精度、训练耗时、多机加速比情况如下所示。
+
+| 模型   | 配置  | 数据集   | 单机8卡耗时/精度 | 2机8卡耗时/精度 | 加速比 |
+|:------:|:-----:|:--------:|:--------:|:--------:|:-----:|
+| CRNN | [rec_chinese_lite_train_v2.0.yml](../../../../configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml) |  26W中文数据集 | 2.50d/66.7%   | 1.67d/67.0%  | **1.5** |
+
+在3机8卡V100的机器上进行模型训练，不同模型的精度、训练耗时、多机加速比情况如下所示。
+
+| 模型   | 配置  | 数据集   | 单机8卡耗时/精度 | 3机8卡耗时/精度 | 加速比 |
+|:------:|:-----:|:--------:|:--------:|:--------:|:-----:|
+| SLANet | [SLANet.yml](../../../../configs/table/SLANet.yml) |  PubTabNet | 49.8h/76.2%   | 19.75h/74.77%  | **2.52** |
+
+注意：这里3机8卡训练时，单卡batch size相比于单机8卡不变，学习率乘以2 (默认乘以3的话，精度仅有73.42%)
+
+在4机8卡V100的机器上进行模型训练，不同模型的精度、训练耗时、多机加速比情况如下所示。
+
+| 模型   | 配置  | 数据集   | 单机8卡耗时/精度 | 4机8卡耗时/精度 | 加速比 |
+|:------:|:-----:|:--------:|:--------:|:--------:|:-----:|
+| SVTR | [PP-OCRv3_mobile_rec_distillation.yml](../../../../configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec_distillation.yml) |  PP-OCRv3_rec data | 10d/-   | 2.84d/74.0%  | **3.5** |
+
+**注意**: 在训练的GPU卡数过多时，精度会稍微有所损失（1%左右），此时可以尝试通过添加warmup或者适当增加迭代轮数来弥补精度损失。
diff --git a/docs/version2.x/ppocr/blog/enhanced_ctc_loss.en.md b/docs/version2.x/ppocr/blog/enhanced_ctc_loss.en.md
new file mode 100644
index 0000000..48f277d
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/enhanced_ctc_loss.en.md
@@ -0,0 +1,100 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+# Enhanced CTC Loss
+
+In OCR recognition, CRNN is a text recognition algorithm widely applied in the industry. In the training phase, it uses CTCLoss to calculate the network loss. In the inference phase, it uses CTCDecode to obtain the decoding result. Although the CRNN algorithm has been proven to achieve reliable recognition results in actual business, users have endless requirements for recognition accuracy. So how to improve the accuracy of text recognition? Taking CTCLoss as the starting point, this paper explores the improved fusion scheme of CTCLoss from three different perspectives: Hard Example Mining, Multi-task Learning, and Metric Learning. Based on the exploration, we propose EnhancedCTCLoss, which includes the following 3 components: Focal-CTC Loss, A-CTC Loss, C-CTC Loss.
+
+## 1. Focal-CTC Loss
+
+Focal Loss was proposed by the paper, "[Focal Loss for Dense Object Detection](https://arxiv.org/abs/1708.02002)". When the loss was first proposed, it was mainly to solve the problem of a serious imbalance in the ratio of positive and negative samples in one-stage target detection. This loss function reduces the weight of a large number of simple negative samples in training and also can be understood as a kind of difficult sample mining.
+The form of the loss function is as follows:
+
+$$
+\begin{equation}
+L_{fl}=\left\{
+\begin{array}{cl}
+-\alpha(1 - y^{'})^{\gamma}logy^{'} ,&  y = 1 \\
+-(1 - \alpha)y^{'\gamma}log(1 - y^{'}), &  y = 0 \\
+\end{array} \right.
+\end{equation}
+$$
+
+Among them, y' is the output of the activation function, and the value is between 0-1. It adds a modulation factor (1-y’)^&gamma; and a balance factor &alpha; on the basis of the original cross-entropy loss. When &alpha; = 1, y = 1, the comparison between the loss function and the cross-entropy loss is shown in the following figure:
+
+![img](./images/focal_loss_image.png)
+
+As can be seen from the above figure, when &gamma; > 0, the adjustment coefficient (1-y’)^&gamma; gives smaller weight to the easy-to-classify sample loss, making the network pay more attention to the difficult and misclassified samples. The adjustment factor &gamma; is used to adjust the rate at which the weight of simple samples decreases. When &gamma; = 0, it is the cross-entropy loss function. When &gamma; increases, the influence of the adjustment factor will also increase. Experiments revealed that 2 is the optimal value of &gamma;. The balance factor &alpha; is used to balance the uneven proportions of the positive and negative samples. In the text, &alpha; is taken as 0.25.
+
+For the classic CTC algorithm, suppose a certain feature sequence (f<sub>1</sub>, f<sub>2</sub>, ......f<sub>t</sub>), after CTC decoding, the probability that the result is equal to label is y', then the probability that the CTC decoding result is not equal to label is (1-y'); it is not difficult to find that the CTCLoss value and y' have the following relationship:
+
+$$
+L_{CTC} = -log(y^{'})
+$$
+
+Combining the idea of Focal Loss, assigning larger weights to difficult samples and smaller weights to simple samples can make the network focus more on the mining of difficult samples and further improve the accuracy of recognition. Therefore, we propose Focal-CTC Loss. Its definition is as follows:
+
+$$
+L_{Focal\_CTC} = \alpha * (1 - y^{'})^{\gamma} * L_{CTC}
+$$
+
+In the experiment, the value of &gamma; is 2, &alpha; = 1, see this for specific implementation: [rec_ctc_loss.py](../../../../ppocr/losses/rec_ctc_loss.py)
+
+## 2. A-CTC Loss
+A-CTC Loss is short for CTC Loss + ACE Loss. Among them, ACE Loss was proposed by the paper, “[Aggregation Cross-Entropy for Sequence Recognition](https://arxiv.org/abs/1904.08364)”. Compared with CTCLoss, ACE Loss has the following two advantages:
++ ACE Loss can solve the recognition problem of 2-D text, while CTCLoss can only process 1-D text
++ ACE Loss is better than CTC loss in time complexity and space complexity
+
+The advantages and disadvantages of the OCR recognition algorithm summarized by the predecessors are shown in the following figure:
+
+![img](./images/rec_algo_compare.png)
+
+Although ACELoss does handle 2D predictions, as shown in the figure above, and has advantages in memory usage and inference speed, in practice, we found that using ACELoss alone, the recognition effect is not as good as CTCLoss. Consequently, we tried to combine CTCLoss and ACELoss, and CTCLoss is the mainstay while ACELoss acts as an auxiliary supervision loss. This attempt has achieved better results. On our internal experimental data set, compared to using CTCLoss alone, the recognition accuracy can be improved by about 1%.
+A_CTC Loss is defined as follows:
+
+$$
+L_{A-CTC} = L_{CTC} + \lambda * L_{ACE}
+$$
+
+In the experiment, λ = 0.1. See the ACE loss implementation code: [ace_loss.py](../../../../ppocr/losses/ace_loss.py)
+
+## 3. C-CTC Loss
+C-CTC Loss is short for CTC Loss + Center Loss. Among them, Center Loss was proposed by the paper, “[A Discriminative Feature Learning Approach for Deep Face Recognition](https://link.springer.com/chapter/10.1007/978-3-319-46478-7_31)“. It was first used in face recognition tasks to increase the distance between classes and reduce the distance within classes. It is an earlier and also widely used algorithm.
+
+In the task of Chinese OCR recognition, through the analysis of bad cases, we found that a major difficulty in Chinese recognition is that there are many similar characters, which are easy to misunderstand. From this, we thought about whether we can learn from the idea of n to increase the class spacing of similar characters, to improve recognition accuracy. However, Metric Learning is mainly used in the field of image recognition, and the label of the training data is a fixed value; for OCR recognition, it is a sequence recognition task essentially, and there is no explicit alignment between features and labels. Therefore, how to combine the two is still a direction worth exploring.
+
+By trying Arcmargin, Cosmargin and other methods, we finally found that Centerloss can help further improve the accuracy of recognition. C_CTC Loss is defined as follows:
+
+$$
+L_{C-CTC} = L_{CTC} + \lambda * L_{center}
+$$
+
+In the experiment, we set λ=0.25. See the center_loss implementation code: [center_loss.py](../../../../ppocr/losses/center_loss.py)
+
+It is worth mentioning that in C-CTC Loss, choosing to initialize the Center randomly does not bring significant improvement. Our Center initialization method is as follows:
++ Based on the original CTCLoss, a network N is obtained by training
++ Select the training set, identify the completely correct part, and form the set G
++ Send each sample in G to the network, perform forward calculation, and extract the correspondence between the input of the last FC layer (ie feature) and the result of argmax calculation (ie index)
++ Aggregate features with the same index, calculate the average, and get the initial center of each character.
+
+Taking the configuration file `configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml` as an example, the center extraction command is as follows:
+
+```bash linenums="1"
+python tools/export_center.py -c configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml -o Global.pretrained_model="./output/rec_mobile_pp-OCRv2/best_accuracy"
+```
+
+After running, `train_center.pkl` will be generated in the main directory of PaddleOCR.
+
+## 4. Experiment
+
+For the above three solutions, we conducted training and evaluation based on Baidu's internal data set. The experimental conditions are shown in the following table:
+
+| algorithm | Focal_CTC | A_CTC | C-CTC |
+| :-------- | :-------- | ----: | :---: |
+| gain      | +0.3%     | +0.7% | +1.7% |
+
+Based on the above experimental conclusions, we adopted the C-CTC strategy in PP-OCRv2. It is worth mentioning that, because PP-OCRv2 deals with the recognition task of 6625 Chinese characters, the character set is relatively large and there are many similar characters, so the C-CTC solution brings a significant improvement on this task. But if you switch to other OCR recognition tasks, the conclusion may be different. You can try Focal-CTC, A-CTC, C-CTC, and the combined solution EnhancedCTC. We believe it will bring different degrees of improvement.
+
+The unified combined plan is shown in the following file: [rec_enhanced_ctc_loss.py](../../../../ppocr/losses/rec_enhanced_ctc_loss.py)
diff --git a/docs/version2.x/ppocr/blog/enhanced_ctc_loss.md b/docs/version2.x/ppocr/blog/enhanced_ctc_loss.md
new file mode 100644
index 0000000..a9c1144
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/enhanced_ctc_loss.md
@@ -0,0 +1,101 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+# Enhanced CTC Loss
+
+在OCR识别中， CRNN是一种在工业界广泛使用的文字识别算法。 在训练阶段，其采用CTCLoss来计算网络损失； 在推理阶段，其采用CTCDecode来获得解码结果。虽然CRNN算法在实际业务中被证明能够获得很好的识别效果， 然而用户对识别准确率的要求却是无止境的，如何进一步提升文字识别的准确率呢？ 本文以CTCLoss为切人点，分别从难例挖掘、 多任务学习、 Metric Learning 3个不同的角度探索了CTCLoss的改进融合方案，提出了EnhancedCTCLoss，其包括如下3个组成部分： Focal-CTC Loss，A-CTC Loss， C-CTC Loss。
+
+## 1. Focal-CTC Loss
+
+Focal Loss 出自论文《Focal Loss for Dense Object Detection》, 该loss最先提出的时候主要是为了解决one-stage目标检测中正负样本比例严重失衡的问题。该损失函数降低了大量简单负样本在训练中所占的权重，也可理解为一种困难样本挖掘。
+其损失函数形式如下：
+
+$$
+\begin{equation}
+L_{fl}=\left\{
+\begin{array}{cl}
+-\alpha(1 - y^{'})^{\gamma}logy^{'} ,&  y = 1 \\
+-(1 - \alpha)y^{'\gamma}log(1 - y^{'}), &  y = 0 \\
+\end{array} \right.
+\end{equation}
+$$
+
+其中,  y' 是经过激活函数的输出，取值在0-1之间。其在原始的交叉熵损失的基础上加了一个调制系数（1 – y’)^ &gamma;和平衡因子&alpha;。 当&alpha; = 1，y=1时，其损失函数与交叉熵损失的对比如下图所示:
+
+![img](./images/focal_loss_image.png)
+
+从上图可以看到, 当&gamma;> 0时，调整系数（1-y’）^&gamma; 赋予易分类样本损失一个更小的权重，使得网络更关注于困难的、错分的样本。 调整因子&gamma;用于调节简单样本权重降低的速率，当&gamma;为0时即为交叉熵损失函数，当&gamma;增加时，调整因子的影响也会随之增大。实验发现&gamma;为2是最优。平衡因子&alpha;用来平衡正负样本本身的比例不均，文中&alpha;取0.25。
+
+对于经典的CTC算法，假设某个特征序列（f<sub>1</sub>, f<sub>2</sub>, ......f<sub>t</sub>), 经过CTC解码之后结果等于label的概率为y’, 则CTC解码结果不为label的概率即为（1-y’)；不难发现, CTCLoss值和y’有如下关系：
+
+$$
+L_{CTC} = -log(y^{'})
+$$
+
+结合Focal Loss的思想，赋予困难样本较大的权重，简单样本较小的权重，可以使网络更加聚焦于对困难样本的挖掘，进一步提升识别的准确率，由此我们提出了Focal-CTC Loss； 其定义如下所示：
+
+$$
+L_{Focal\_CTC} = \alpha * (1 - y^{'})^{\gamma} * L_{CTC}
+$$
+
+实验中，&gamma;取值为2, &alpha;= 1, 具体实现见:  [rec_ctc_loss.py](../../../../ppocr/losses/rec_ctc_loss.py)
+
+## 2. A-CTC Loss
+
+A-CTC Loss是CTC Loss + ACE Loss的简称。 其中ACE Loss出自论文< Aggregation Cross-Entropy for Sequence Recognition>.  ACE Loss相比于CTCLoss，主要有如下两点优势:
+
++ ACE Loss能够解决2-D文本的识别问题;  CTCLoss只能够处理1-D文本
++ ACE Loss 在时间复杂度和空间复杂度上优于CTC loss
+
+前人总结的OCR识别算法的优劣如下图所示：
+
+![img](./images/rec_algo_compare.png)
+
+虽然ACELoss确实如上图所说，可以处理2D预测，在内存占用及推理速度方面具备优势，但在实践过程中，我们发现单独使用ACE Loss,  识别效果并不如CTCLoss.  因此，我们尝试将CTCLoss和ACELoss进行结合，同时以CTCLoss为主，将ACELoss 定位为一个辅助监督loss。 这一尝试收到了效果，在我们内部的实验数据集上，相比单独使用CTCLoss，识别准确率可以提升1%左右。
+A_CTC Loss定义如下:
+
+$$
+L_{A-CTC} = L_{CTC} + \lambda * L_{ACE}
+$$
+
+实验中，λ = 0.1.  ACE loss实现代码见:  [ace_loss.py](../../../../ppocr/losses/ace_loss.py)
+
+## 3. C-CTC Loss
+
+C-CTC Loss是CTC Loss + Center Loss的简称。 其中Center Loss出自论文 < A Discriminative Feature Learning Approach for Deep Face Recognition>.  最早用于人脸识别任务，用于增大类间距离，减小类内距离,  是Metric Learning领域一种较早的、也比较常用的一种算法。
+在中文OCR识别任务中，通过对badcase分析， 我们发现中文识别的一大难点是相似字符多，容易误识。 由此我们想到是否可以借鉴Metric Learing的想法， 增大相似字符的类间距，从而提高识别准确率。然而，MetricLearning主要用于图像识别领域，训练数据的标签为一个固定的值；而对于OCR识别来说，其本质上是一个序列识别任务，特征和label之间并不具有显式的对齐关系，因此两者如何结合依然是一个值得探索的方向。
+通过尝试Arcmargin, Cosmargin等方法， 我们最终发现Centerloss 有助于进一步提升识别的准确率。C_CTC Loss定义如下：
+
+$$
+L_{C-CTC} = L_{CTC} + \lambda * L_{center}
+$$
+
+实验中，我们设置λ=0.25. center_loss实现代码见:  [center_loss.py](../../../../ppocr/losses/center_loss.py)
+
+值得一提的是， 在C-CTC Loss中，选择随机初始化Center并不能够带来明显的提升. 我们的Center初始化方法如下：
+
++ 基于原始的CTCLoss， 训练得到一个网络N
++ 挑选出训练集中，识别完全正确的部分, 组成集合G
++ 将G中的每个样本送入网络，进行前向计算， 提取最后一个FC层的输入（即feature）及其经过argmax计算的结果（即index）之间的对应关系
++ 将相同index的feature进行聚合，计算平均值，得到各自字符的初始center.
+
+以配置文件`configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml`为例， center提取命令如下所示:
+
+```bash linenums="1"
+python tools/export_center.py -c configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml -o Global.pretrained_model="./output/rec_mobile_pp-OCRv2/best_accuracy"
+```
+
+运行完后，会在PaddleOCR主目录下生成`train_center.pkl`.
+
+## 4. 实验
+
+对于上述的三种方案，我们基于百度内部数据集进行了训练、评测，实验情况如下表所示：
+
+|algorithm| Focal_CTC | A_CTC | C-CTC |
+|:------| :------| ------: | :------: |
+|gain| +0.3% | +0.7% | +1.7% |
+
+基于上述实验结论，我们在PP-OCRv2中，采用了C-CTC的策略。 值得一提的是，由于PP-OCRv2 处理的是6625个中文字符的识别任务，字符集比较大，形似字较多，所以在该任务上C-CTC 方案带来的提升较大。 但如果换做其他OCR识别任务，结论可能会有所不同。大家可以尝试Focal-CTC，A-CTC, C-CTC以及组合方案EnhancedCTC，相信会带来不同程度的提升效果。
+统一的融合方案见如下文件：  [rec_enhanced_ctc_loss.py](../../../../ppocr/losses/rec_enhanced_ctc_loss.py)
diff --git a/docs/version2.x/ppocr/blog/images/11_det.jpg b/docs/version2.x/ppocr/blog/images/11_det.jpg
new file mode 100644
index 0000000..fe0cd23
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/11_det.jpg differ
diff --git a/docs/version2.x/ppocr/blog/images/11_det_rec.jpg b/docs/version2.x/ppocr/blog/images/11_det_rec.jpg
new file mode 100644
index 0000000..31c5664
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/11_det_rec.jpg differ
diff --git a/docs/version2.x/ppocr/blog/images/12_det.jpg b/docs/version2.x/ppocr/blog/images/12_det.jpg
new file mode 100644
index 0000000..71627f0
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/12_det.jpg differ
diff --git a/docs/version2.x/ppocr/blog/images/12_det_rec.jpg b/docs/version2.x/ppocr/blog/images/12_det_rec.jpg
new file mode 100644
index 0000000..9db8b57
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/12_det_rec.jpg differ
diff --git a/docs/version2.x/ppocr/blog/images/187578511-9f3c351e-b68c-4359-a6e5-475810993c61.png b/docs/version2.x/ppocr/blog/images/187578511-9f3c351e-b68c-4359-a6e5-475810993c61.png
new file mode 100644
index 0000000..86c0626
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/187578511-9f3c351e-b68c-4359-a6e5-475810993c61.png differ
diff --git a/docs/version2.x/ppocr/blog/images/254-20240709081442260.jpg b/docs/version2.x/ppocr/blog/images/254-20240709081442260.jpg
new file mode 100644
index 0000000..c871fb0
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/254-20240709081442260.jpg differ
diff --git a/docs/version2.x/ppocr/blog/images/5e612.png b/docs/version2.x/ppocr/blog/images/5e612.png
new file mode 100644
index 0000000..79e3d0e
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/5e612.png differ
diff --git a/docs/version2.x/ppocr/blog/images/DF.png b/docs/version2.x/ppocr/blog/images/DF.png
new file mode 100644
index 0000000..f14953d
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/DF.png differ
diff --git a/docs/version2.x/ppocr/blog/images/GTC.png b/docs/version2.x/ppocr/blog/images/GTC.png
new file mode 100644
index 0000000..30a9cdd
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/GTC.png differ
diff --git a/docs/version2.x/ppocr/blog/images/LCNet_SVTR.png b/docs/version2.x/ppocr/blog/images/LCNet_SVTR.png
new file mode 100644
index 0000000..7f0d701
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/LCNet_SVTR.png differ
diff --git a/docs/version2.x/ppocr/blog/images/LKPAN.png b/docs/version2.x/ppocr/blog/images/LKPAN.png
new file mode 100644
index 0000000..6b16053
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/LKPAN.png differ
diff --git a/docs/version2.x/ppocr/blog/images/PFHead.png b/docs/version2.x/ppocr/blog/images/PFHead.png
new file mode 100644
index 0000000..3728dc4
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/PFHead.png differ
diff --git a/docs/version2.x/ppocr/blog/images/RSEFPN.png b/docs/version2.x/ppocr/blog/images/RSEFPN.png
new file mode 100644
index 0000000..ddf7c52
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/RSEFPN.png differ
diff --git a/docs/version2.x/ppocr/blog/images/SSL.png b/docs/version2.x/ppocr/blog/images/SSL.png
new file mode 100644
index 0000000..1344a2a
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/SSL.png differ
diff --git a/docs/version2.x/ppocr/blog/images/UIM.png b/docs/version2.x/ppocr/blog/images/UIM.png
new file mode 100644
index 0000000..7479bdf
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/UIM.png differ
diff --git a/docs/version2.x/ppocr/blog/images/arabic_0.jpg b/docs/version2.x/ppocr/blog/images/arabic_0.jpg
new file mode 100644
index 0000000..9941b90
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/arabic_0.jpg differ
diff --git a/docs/version2.x/ppocr/blog/images/focal_loss_image.png b/docs/version2.x/ppocr/blog/images/focal_loss_image.png
new file mode 100644
index 0000000..430550a
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/focal_loss_image.png differ
diff --git a/docs/version2.x/ppocr/blog/images/french_0.jpg b/docs/version2.x/ppocr/blog/images/french_0.jpg
new file mode 100644
index 0000000..3c2abe6
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/french_0.jpg differ
diff --git a/docs/version2.x/ppocr/blog/images/img_02.jpg b/docs/version2.x/ppocr/blog/images/img_02.jpg
new file mode 100644
index 0000000..3e139c7
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/img_02.jpg differ
diff --git a/docs/version2.x/ppocr/blog/images/img_12.jpg b/docs/version2.x/ppocr/blog/images/img_12.jpg
new file mode 100644
index 0000000..822d562
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/img_12.jpg differ
diff --git a/docs/version2.x/ppocr/blog/images/japan_2-20240709081138234.jpg b/docs/version2.x/ppocr/blog/images/japan_2-20240709081138234.jpg
new file mode 100644
index 0000000..7038ba2
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/japan_2-20240709081138234.jpg differ
diff --git a/docs/version2.x/ppocr/blog/images/korean.jpg b/docs/version2.x/ppocr/blog/images/korean.jpg
new file mode 100644
index 0000000..e5d863c
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/korean.jpg differ
diff --git a/docs/version2.x/ppocr/blog/images/korean_0.jpg b/docs/version2.x/ppocr/blog/images/korean_0.jpg
new file mode 100644
index 0000000..3fe6305
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/korean_0.jpg differ
diff --git a/docs/version2.x/ppocr/blog/images/multi_scale.png b/docs/version2.x/ppocr/blog/images/multi_scale.png
new file mode 100644
index 0000000..673d306
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/multi_scale.png differ
diff --git a/docs/version2.x/ppocr/blog/images/ppocrv3_det_cml.png b/docs/version2.x/ppocr/blog/images/ppocrv3_det_cml.png
new file mode 100644
index 0000000..ccb5c8b
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/ppocrv3_det_cml.png differ
diff --git a/docs/version2.x/ppocr/blog/images/ppocrv3_framework-0052468.png b/docs/version2.x/ppocr/blog/images/ppocrv3_framework-0052468.png
new file mode 100644
index 0000000..e05279f
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/ppocrv3_framework-0052468.png differ
diff --git a/docs/version2.x/ppocr/blog/images/ppocrv4_det_cml.png b/docs/version2.x/ppocr/blog/images/ppocrv4_det_cml.png
new file mode 100644
index 0000000..9132c0a
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/ppocrv4_det_cml.png differ
diff --git a/docs/version2.x/ppocr/blog/images/ppocrv4_framework.png b/docs/version2.x/ppocr/blog/images/ppocrv4_framework.png
new file mode 100644
index 0000000..fa31f4c
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/ppocrv4_framework.png differ
diff --git a/docs/version2.x/ppocr/blog/images/ppocrv4_gtc.png b/docs/version2.x/ppocr/blog/images/ppocrv4_gtc.png
new file mode 100644
index 0000000..7e6a3f5
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/ppocrv4_gtc.png differ
diff --git a/docs/version2.x/ppocr/blog/images/rec_algo_compare.png b/docs/version2.x/ppocr/blog/images/rec_algo_compare.png
new file mode 100644
index 0000000..2dde496
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/rec_algo_compare.png differ
diff --git a/docs/version2.x/ppocr/blog/images/recconaug.png b/docs/version2.x/ppocr/blog/images/recconaug.png
new file mode 100644
index 0000000..899bc43
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/recconaug.png differ
diff --git a/docs/version2.x/ppocr/blog/images/svtr_g2.png b/docs/version2.x/ppocr/blog/images/svtr_g2.png
new file mode 100644
index 0000000..2573afa
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/svtr_g2.png differ
diff --git a/docs/version2.x/ppocr/blog/images/svtr_g4.png b/docs/version2.x/ppocr/blog/images/svtr_g4.png
new file mode 100644
index 0000000..f85d66d
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/svtr_g4.png differ
diff --git a/docs/version2.x/ppocr/blog/images/svtr_tiny.png b/docs/version2.x/ppocr/blog/images/svtr_tiny.png
new file mode 100644
index 0000000..01e22e7
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/svtr_tiny.png differ
diff --git a/docs/version2.x/ppocr/blog/images/teacher_dml.png b/docs/version2.x/ppocr/blog/images/teacher_dml.png
new file mode 100644
index 0000000..ea09cac
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/teacher_dml.png differ
diff --git a/docs/version2.x/ppocr/blog/images/v3_rec_pipeline.png b/docs/version2.x/ppocr/blog/images/v3_rec_pipeline.png
new file mode 100644
index 0000000..aa61cc4
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/v3_rec_pipeline.png differ
diff --git a/docs/version2.x/ppocr/blog/images/v4_rec_pipeline.png b/docs/version2.x/ppocr/blog/images/v4_rec_pipeline.png
new file mode 100644
index 0000000..b1ec7a9
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/v4_rec_pipeline.png differ
diff --git a/docs/version2.x/ppocr/blog/images/word_308.png b/docs/version2.x/ppocr/blog/images/word_308.png
new file mode 100644
index 0000000..a8d094f
Binary files /dev/null and b/docs/version2.x/ppocr/blog/images/word_308.png differ
diff --git a/docs/version2.x/ppocr/blog/inference_args.en.md b/docs/version2.x/ppocr/blog/inference_args.en.md
new file mode 100644
index 0000000..1a79fef
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/inference_args.en.md
@@ -0,0 +1,131 @@
+---
+comments: true
+---
+
+
+# PaddleOCR Model Inference Parameter Explanation
+
+When using PaddleOCR for model inference, you can customize the modification parameters to modify the model, data, preprocessing, postprocessing, etc. (parameter file: [utility.py](../../../../tools/infer/utility.py))，The detailed parameter explanation is as follows:
+
+* Global parameters
+
+| parameters | type | default | implication |
+| :--: | :--: | :--: | :--: |
+|  image_dir | str | None, must be specified explicitly | Image or folder path |
+|  page_num | int | 0 | Valid when the input type is pdf file, specify to predict the previous page_num pages, all pages are predicted by default |
+|  vis_font_path | str | "./doc/fonts/simfang.ttf" | font path for visualization |
+|  drop_score | float | 0.5 | Results with a recognition score less than this value will be discarded and will not be returned as results |
+|  use_pdserving | bool | False | Whether to use Paddle Serving for prediction |
+|  warmup | bool | False | Whether to enable warmup, this method can be used when statistical prediction time |
+|  draw_img_save_dir | str | "./inference_results" | The saving folder of the system's tandem prediction OCR results |
+|  save_crop_res | bool | False  | Whether to save the recognized text image for OCR |
+|  crop_res_save_dir | str | "./output" | Save the text image path recognized by OCR |
+|  use_mp | bool | False | Whether to enable multi-process prediction  |
+|  total_process_num | int | 6 | The number of processes, which takes effect when `use_mp` is `True` |
+|  process_id | int | 0 | The id number of the current process, no need to modify it yourself |
+|  benchmark | bool | False | Whether to enable benchmark, and make statistics on prediction speed, memory usage, etc. |
+|  save_log_path | str | "./log_output/" | Folder where log results are saved when `benchmark` is enabled |
+|  show_log | bool | True | Whether to show the log information in the inference |
+|  use_onnx | bool | False | Whether to enable onnx prediction |
+
+* Prediction engine related parameters
+
+| parameters | type | default | implication |
+| :--: | :--: | :--: | :--: |
+|  use_gpu | bool | True | Whether to use GPU for prediction |
+|  ir_optim | bool | True | Whether to analyze and optimize the calculation graph. The prediction process can be accelerated when `ir_optim` is enabled |
+|  use_tensorrt | bool | False | Whether to enable tensorrt |
+|  min_subgraph_size | int | 15 | The minimum subgraph size in tensorrt. When the size of the subgraph is greater than this value, it will try to use the trt engine to calculate the subgraph. |
+|  precision | str | fp32 | The precision of prediction, supports `fp32`, `fp16`, `int8` |
+|  enable_mkldnn | bool | True | Whether to enable mkldnn |
+|  cpu_threads | int | 10 | When mkldnn is enabled, the number of threads predicted by the cpu |
+
+* Text detection model related parameters
+
+| parameters | type | default | implication |
+| :--: | :--: | :--: | :--: |
+|  det_algorithm | str | "DB" | Text detection algorithm name, currently supports `DB`, `EAST`, `SAST`, `PSE`, `DB++`, `FCE` |
+|  det_model_dir | str | xx | Detection inference model paths |
+|  det_limit_side_len | int | 960 | image side length limit |
+|  det_limit_type | str | "max" | The side length limit type, currently supports `min`and `max`. `min` means to ensure that the shortest side of the image is not less than `det_limit_side_len`, `max` means to ensure that the longest side of the image is not greater than `det_limit_side_len` |
+
+The relevant parameters of the DB algorithm are as follows
+
+| parameters | type | default | implication |
+| :--: | :--: | :--: | :--: |
+|  det_db_thresh | float | 0.3 | In the probability map output by DB, only pixels with a score greater than this threshold will be considered as text pixels |
+|  det_db_box_thresh | float | 0.6 | Within the detection box, when the average score of all pixels is greater than the threshold, the result will be considered as a text area |
+|  det_db_unclip_ratio | float | 1.5 | The expansion factor of the `Vatti clipping` algorithm, which is used to expand the text area |
+|  max_batch_size | int | 10 | max batch size |
+|  use_dilation | bool | False | Whether to inflate the segmentation results to obtain better detection results |
+|  det_db_score_mode | str | "fast" | DB detection result score calculation method, supports `fast` and `slow`, `fast` calculates the average score according to all pixels within the bounding rectangle of the polygon, `slow` calculates the average score according to all pixels within the original polygon, The calculation speed is relatively slower, but more accurate. |
+
+The relevant parameters of the EAST algorithm are as follows
+
+| parameters | type | default | implication |
+| :--: | :--: | :--: | :--: |
+|  det_east_score_thresh | float | 0.8 | Threshold for score map in EAST postprocess |
+|  det_east_cover_thresh | float | 0.1 | Average score threshold for text boxes in EAST postprocess |
+|  det_east_nms_thresh | float | 0.2 | Threshold of nms in EAST postprocess |
+
+The relevant parameters of the SAST algorithm are as follows
+
+| parameters | type | default | implication |
+| :--: | :--: | :--: | :--: |
+|  det_sast_score_thresh | float | 0.5 | Score thresholds in SAST postprocess |
+|  det_sast_nms_thresh | float | 0.5 | Thresholding of nms in SAST postprocess |
+|  det_box_type | str | 'quad' | Whether polygon detection, curved text scene (such as Total-Text) is set to 'poly' |
+
+The relevant parameters of the PSE algorithm are as follows
+
+| parameters | type | default | implication |
+| :--: | :--: | :--: | :--: |
+|  det_pse_thresh | float | 0.0 | Threshold for binarizing the output image |
+|  det_pse_box_thresh | float | 0.85 | Threshold for filtering boxes, below this threshold is discarded |
+|  det_pse_min_area | float | 16 | The minimum area of the box, below this threshold is discarded |
+|  det_box_type | str | "quad" | The type of the returned box, quad: four point coordinates, poly: all point coordinates of the curved text |
+|  det_pse_scale | int | 1 | The ratio of the input image relative to the post-processed image, such as an image of `640*640`, the network output is `160*160`, and when the scale is 2, the shape of the post-processed image is `320*320`. Increasing this value can speed up the post-processing speed, but it will bring about a decrease in accuracy |
+
+* Text recognition model related parameters
+
+| parameters | type | default | implication |
+| :--: | :--: | :--: | :--: |
+|  rec_algorithm | str | "CRNN" | Text recognition algorithm name, currently supports `CRNN`, `SRN`, `RARE`, `NETR`, `SAR`, `ViTSTR`, `ABINet`, `VisionLAN`, `SPIN`, `RobustScanner`, `SVTR`, `SVTR_LCNet` |
+|  rec_model_dir | str | None, it is required if using the recognition model | recognition inference model paths |
+|  rec_image_shape | str | "3,48,320" ] | Image size at the time of recognition |
+|  rec_batch_num | int | 6 | batch size |
+|  max_text_length | int | 25 | The maximum length of the recognition result, valid in `SRN` |
+|  rec_char_dict_path | str | "./ppocr/utils/ppocr_keys_v1.txt" | character dictionary file |
+|  use_space_char | bool | True | Whether to include spaces, if `True`, the `space` character will be added at the end of the character dictionary |
+
+* End-to-end text detection and recognition model related parameters
+
+| parameters | type | default | implication |
+| :--: | :--: | :--: | :--: |
+|  e2e_algorithm | str | "PGNet" | End-to-end algorithm name, currently supports `PGNet` |
+|  e2e_model_dir | str | None, it is required if using the end-to-end model | end-to-end model inference model path |
+|  e2e_limit_side_len | int | 768 | End-to-end input image side length limit |
+|  e2e_limit_type | str | "max" | End-to-end side length limit type, currently supports `min` and `max`. `min` means to ensure that the shortest side of the image is not less than `e2e_limit_side_len`, `max` means to ensure that the longest side of the image is not greater than `e2e_limit_side_len` |
+|  e2e_pgnet_score_thresh | float | 0.5 | End-to-end score threshold, results below this threshold are discarded |
+|  e2e_char_dict_path | str | "./ppocr/utils/ic15_dict.txt" | Recognition dictionary file path |
+|  e2e_pgnet_valid_set | str | "totaltext" | The name of the validation set, currently supports `totaltext`, `partvgg`, the post-processing methods corresponding to different data sets are different, and it can be consistent with the training process |
+|  e2e_pgnet_mode | str | "fast" | PGNet's detection result score calculation method, supports `fast` and `slow`, `fast` calculates the average score according to all pixels within the bounding rectangle of the polygon, `slow` calculates the average score according to all pixels within the original polygon, The calculation speed is relatively slower, but more accurate. |
+
+* Angle classifier model related parameters
+
+| parameters | type | default | implication |
+| :--: | :--: | :--: | :--: |
+|  use_angle_cls | bool | False | whether to use an angle classifier |
+|  cls_model_dir | str | None, if you need to use, you must specify the path explicitly | angle classifier inference model path |
+|  cls_image_shape | str | "3,48,192" | prediction shape |
+|  label_list | list | ['0', '180'] | The angle value corresponding to the class id |
+|  cls_batch_num | int | 6 | batch size |
+|  cls_thresh | float | 0.9 | Prediction threshold, when the model prediction result is 180 degrees, and the score is greater than the threshold, the final prediction result is considered to be 180 degrees and needs to be flipped |
+
+* OCR image preprocessing parameters
+
+| parameters | type | default | implication |
+| :--: | :--: | :--: | :--: |
+|  invert | bool | False | whether to invert image before processing |
+|  binarize | bool | False | whether to threshold binarize image before processing |
+|  alphacolor | tuple | "255,255,255" | Replacement color for the alpha channel, if the latter is present; R,G,B integers |
diff --git a/docs/version2.x/ppocr/blog/inference_args.md b/docs/version2.x/ppocr/blog/inference_args.md
new file mode 100644
index 0000000..6304c8e
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/inference_args.md
@@ -0,0 +1,118 @@
+# PaddleOCR模型推理参数解释
+
+在使用PaddleOCR进行模型推理时，可以自定义修改参数，来修改模型、数据、预处理、后处理等内容（参数文件：[utility.py](../../../../tools/infer/utility.py)），详细的参数解释如下所示。
+
+* 全局信息
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  image_dir | str | 无，必须显式指定 | 图像或者文件夹路径 |
+|  page_num | int | 0 | 当输入类型为pdf文件时有效，指定预测前面page_num页，默认预测所有页 |
+|  vis_font_path | str | "./doc/fonts/simfang.ttf" | 用于可视化的字体路径 |
+|  drop_score | float | 0.5 | 识别得分小于该值的结果会被丢弃，不会作为返回结果 |
+|  use_pdserving | bool | False | 是否使用Paddle Serving进行预测 |
+|  warmup | bool | False | 是否开启warmup，在统计预测耗时的时候，可以使用这种方法 |
+|  draw_img_save_dir | str | "./inference_results" | 系统串联预测OCR结果的保存文件夹 |
+|  save_crop_res | bool | False  | 是否保存OCR的识别文本图像 |
+|  crop_res_save_dir | str | "./output" | 保存OCR识别出来的文本图像路径 |
+|  use_mp | bool | False | 是否开启多进程预测  |
+|  total_process_num | int | 6 | 开启的进程数，`use_mp`为`True`时生效  |
+|  process_id | int | 0 | 当前进程的id号，无需自己修改  |
+|  benchmark | bool | False | 是否开启benchmark，对预测速度、显存占用等进行统计  |
+|  save_log_path | str | "./log_output/" | 开启`benchmark`时，日志结果的保存文件夹 |
+|  show_log | bool | True | 是否显示预测中的日志信息  |
+|  use_onnx | bool | False | 是否开启onnx预测 |
+
+* 预测引擎相关
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  use_gpu | bool | True | 是否使用GPU进行预测 |
+|  ir_optim | bool | True | 是否对计算图进行分析与优化，开启后可以加速预测过程 |
+|  use_tensorrt | bool | False | 是否开启tensorrt |
+|  min_subgraph_size | int | 15 | tensorrt中最小子图size，当子图的size大于该值时，才会尝试对该子图使用trt engine计算 |
+|  precision | str | fp32 | 预测的精度，支持`fp32`, `fp16`, `int8` 3种输入 |
+|  enable_mkldnn | bool | True | 是否开启mkldnn |
+|  cpu_threads | int | 10 | 开启mkldnn时，cpu预测的线程数 |
+
+* 文本检测模型相关
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  det_algorithm | str | "DB" | 文本检测算法名称，目前支持`DB`, `EAST`, `SAST`, `PSE`, `DB++`, `FCE`  |
+|  det_model_dir | str | xx | 检测inference模型路径 |
+|  det_limit_side_len | int | 960 | 检测的图像边长限制 |
+|  det_limit_type | str | "max" | 检测的边长限制类型，目前支持`min`和`max`，`min`表示保证图像最短边不小于`det_limit_side_len`，`max`表示保证图像最长边不大于`det_limit_side_len` |
+
+其中，DB算法相关参数如下
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  det_db_thresh | float | 0.3 | DB输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点 |
+|  det_db_box_thresh | float | 0.6 | 检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域 |
+|  det_db_unclip_ratio | float | 1.5 | `Vatti clipping`算法的扩张系数，使用该方法对文字区域进行扩张 |
+|  max_batch_size | int | 10 | 预测的batch size |
+|  use_dilation | bool | False | 是否对分割结果进行膨胀以获取更优检测效果 |
+|  det_db_score_mode | str | "fast" | DB的检测结果得分计算方法，支持`fast`和`slow`，`fast`是根据polygon的外接矩形边框内的所有像素计算平均得分，`slow`是根据原始polygon内的所有像素计算平均得分，计算速度相对较慢一些，但是更加准确一些。 |
+
+EAST算法相关参数如下
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  det_east_score_thresh | float | 0.8 | EAST后处理中score map的阈值 |
+|  det_east_cover_thresh | float | 0.1 | EAST后处理中文本框的平均得分阈值 |
+|  det_east_nms_thresh | float | 0.2 | EAST后处理中nms的阈值 |
+
+SAST算法相关参数如下
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  det_sast_score_thresh | float | 0.5 | SAST后处理中的得分阈值 |
+|  det_sast_nms_thresh | float | 0.5 | SAST后处理中nms的阈值 |
+|  det_box_type | str | quad | 是否多边形检测，弯曲文本场景（如Total-Text）设置为'poly' |
+
+PSE算法相关参数如下
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  det_pse_thresh | float | 0.0 | 对输出图做二值化的阈值 |
+|  det_pse_box_thresh | float | 0.85 | 对box进行过滤的阈值，低于此阈值的丢弃 |
+|  det_pse_min_area | float | 16 | box的最小面积，低于此阈值的丢弃 |
+|  det_box_type | str | "quad" | 返回框的类型，quad:四点坐标，poly: 弯曲文本的所有点坐标 |
+|  det_pse_scale | int | 1 | 输入图像相对于进后处理的图的比例，如`640*640`的图像，网络输出为`160*160`，scale为2的情况下，进后处理的图片shape为`320*320`。这个值调大可以加快后处理速度，但是会带来精度的下降 |
+
+* 文本识别模型相关
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  rec_algorithm | str | "CRNN" | 文本识别算法名称，目前支持`CRNN`, `SRN`, `RARE`, `NETR`, `SAR`, `ViTSTR`, `ABINet`, `VisionLAN`, `SPIN`, `RobustScanner`, `SVTR`, `SVTR_LCNet` |
+|  rec_model_dir | str | 无，如果使用识别模型，该项是必填项 | 识别inference模型路径 |
+|  rec_image_shape | str | "3,48,320" | 识别时的图像尺寸 |
+|  rec_batch_num | int | 6 | 识别的batch size |
+|  max_text_length | int | 25 | 识别结果最大长度，在`SRN`中有效 |
+|  rec_char_dict_path | str | "./ppocr/utils/ppocr_keys_v1.txt" | 识别的字符字典文件 |
+|  use_space_char | bool | True | 是否包含空格，如果为`True`，则会在最后字符字典中补充`空格`字符 |
+
+* 端到端文本检测与识别模型相关
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  e2e_algorithm | str | "PGNet" | 端到端算法名称，目前支持`PGNet` |
+|  e2e_model_dir | str | 无，如果使用端到端模型，该项是必填项 | 端到端模型inference模型路径 |
+|  e2e_limit_side_len | int | 768 | 端到端的输入图像边长限制 |
+|  e2e_limit_type | str | "max" | 端到端的边长限制类型，目前支持`min`, `max`，`min`表示保证图像最短边不小于`e2e_limit_side_len`，`max`表示保证图像最长边不大于`e2e_limit_side_len` |
+|  e2e_pgnet_score_thresh | float | 0.5 | 端到端得分阈值，小于该阈值的结果会被丢弃 |
+|  e2e_char_dict_path | str | "./ppocr/utils/ic15_dict.txt" | 识别的字典文件路径 |
+|  e2e_pgnet_valid_set | str | "totaltext" | 验证集名称，目前支持`totaltext`, `partvgg`，不同数据集对应的后处理方式不同，与训练过程保持一致即可 |
+|  e2e_pgnet_mode | str | "fast" | PGNet的检测结果得分计算方法，支持`fast`和`slow`，`fast`是根据polygon的外接矩形边框内的所有像素计算平均得分，`slow`是根据原始polygon内的所有像素计算平均得分，计算速度相对较慢一些，但是更加准确一些。 |
+
+* 方向分类器模型相关
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  use_angle_cls | bool | False | 是否使用方向分类器 |
+|  cls_model_dir | str | 无，如果需要使用，则必须显式指定路径 | 方向分类器inference模型路径 |
+|  cls_image_shape | str | "3,48,192" | 预测尺度 |
+|  label_list | list | ['0', '180'] | class id对应的角度值 |
+|  cls_batch_num | int | 6 | 方向分类器预测的batch size |
+|  cls_thresh | float | 0.9 | 预测阈值，模型预测结果为180度，且得分大于该阈值时，认为最终预测结果为180度，需要翻转 |
diff --git a/docs/version2.x/ppocr/blog/multi_languages.en.md b/docs/version2.x/ppocr/blog/multi_languages.en.md
new file mode 100644
index 0000000..6deee73
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/multi_languages.en.md
@@ -0,0 +1,225 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+# Multi-language model
+
+**Recent Update**
+
+- 2022.5.8 update the `PP-OCRv3` version of the multi-language detection and recognition model, and the average recognition accuracy has increased by more than 5%.
+- 2021.4.9 supports the detection and recognition of 80 languages
+- 2021.4.9 supports **lightweight high-precision** English model detection and recognition
+
+PaddleOCR aims to create a rich, leading, and practical OCR tool library, which not only provides
+Chinese and English models in general scenarios, but also provides models specifically trained
+in English scenarios. And multilingual models covering [80 languages](#5-support-languages-and-abbreviations).
+
+Among them, the English model supports the detection and recognition of uppercase and lowercase
+letters and common punctuation, and the recognition of space characters is optimized:
+
+![img](./images/img_12.jpg)
+
+The multilingual models cover Latin, Arabic, Traditional Chinese, Korean, Japanese, etc.:
+
+![img](./images/japan_2-20240709081138234.jpg)
+
+![img](./images/french_0.jpg)
+
+![img](./images/korean_0.jpg)
+
+![img](./images/arabic_0.jpg)
+
+This document will briefly introduce how to use the multilingual model.
+
+## 1 Installation
+
+### 1.1 Paddle installation
+
+```bash linenums="1"
+# cpu
+pip install "paddlepaddle<=2.6"
+
+# gpu
+pip install "paddlepaddle-gpu<=2.6"
+```
+
+### 1.2 PaddleOCR package installation
+
+```bash linenums="1"
+pip install "paddleocr<3.0"
+```
+
+Build and install locally
+
+```bash linenums="1"
+python3 -m build
+pip3 install dist/paddleocr-x.x.x-py3-none-any.whl # x.x.x is the version number of paddleocr
+```
+
+## 2 Quick use
+
+### 2.1 Command line operation
+
+View help information
+
+```bash linenums="1"
+paddleocr -h
+```
+
+- Whole image prediction (detection + recognition)
+
+PaddleOCR currently supports 80 languages, which can be specified by the --lang parameter.
+The supported languages are listed in the [table](#5-support-languages-and-abbreviations).
+
+``` bash
+paddleocr --image_dir doc/imgs_en/254.jpg --lang=en
+```
+
+![](./images/254-20240709081442260.jpg)
+
+![img](./images/img_02.jpg)
+
+The result is a list. Each item contains a text box, text and recognition confidence
+
+```text linenums="1"
+[('PHO CAPITAL', 0.95723116), [[66.0, 50.0], [327.0, 44.0], [327.0, 76.0], [67.0, 82.0]]]
+[('107 State Street', 0.96311164), [[72.0, 90.0], [451.0, 84.0], [452.0, 116.0], [73.0, 121.0]]]
+[('Montpelier Vermont', 0.97389287), [[69.0, 132.0], [501.0, 126.0], [501.0, 158.0], [70.0, 164.0]]]
+[('8022256183', 0.99810505), [[71.0, 175.0], [363.0, 170.0], [364.0, 202.0], [72.0, 207.0]]]
+[('REG 07-24-201706:59 PM', 0.93537045), [[73.0, 299.0], [653.0, 281.0], [654.0, 318.0], [74.0, 336.0]]]
+[('045555', 0.99346405), [[509.0, 331.0], [651.0, 325.0], [652.0, 356.0], [511.0, 362.0]]]
+[('CT1', 0.9988654), [[535.0, 367.0], [654.0, 367.0], [654.0, 406.0], [535.0, 406.0]]]
+......
+```
+
+- Recognition
+
+```bash linenums="1"
+paddleocr --image_dir doc/imgs_words_en/word_308.png --det false --lang=en
+```
+
+![img](./images/word_308.png)
+
+The result is a 2-tuple, which contains the recognition result and recognition confidence
+
+```text linenums="1"
+(0.99879867, 'LITTLE')
+```
+
+- Detection
+
+```bash linenums="1"
+paddleocr --image_dir PaddleOCR/doc/imgs/11.jpg --rec false
+```
+
+The result is a list. Each item represents the coordinates of a text box.
+
+```bash linenums="1"
+[[26.0, 457.0], [137.0, 457.0], [137.0, 477.0], [26.0, 477.0]]
+[[25.0, 425.0], [372.0, 425.0], [372.0, 448.0], [25.0, 448.0]]
+[[128.0, 397.0], [273.0, 397.0], [273.0, 414.0], [128.0, 414.0]]
+......
+```
+
+### 2.2 Run with Python script
+
+PPOCR is able to run with Python scripts for easy integration with your own code:
+
+- Whole image prediction (detection + recognition)
+
+```python linenums="1"
+from paddleocr import PaddleOCR, draw_ocr
+
+# Also switch the language by modifying the lang parameter
+ocr = PaddleOCR(lang="korean") # The model file will be downloaded automatically when executed for the first time
+img_path ='doc/imgs/korean_1.jpg'
+result = ocr.ocr(img_path)
+# Recognition and detection can be performed separately through parameter control
+# result = ocr.ocr(img_path, det=False)  Only perform recognition
+# result = ocr.ocr(img_path, rec=False)  Only perform detection
+# Print detection frame and recognition result
+for line in result:
+    print(line)
+
+# Visualization
+from PIL import Image
+image = Image.open(img_path).convert('RGB')
+boxes = [line[0] for line in result]
+txts = [line[1][0] for line in result]
+scores = [line[1][1] for line in result]
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/korean.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+Visualization of results:
+
+![img](./images/korean.jpg)
+
+PPOCR also supports direction classification. For more detailed usage, please refer to: [whl package instructions](whl.en.md).
+
+## 3 Custom training
+
+PPOCR supports using your own data for custom training or fine-tune, where the recognition model can refer to [French configuration file](../../../../configs/rec/multi_language/rec_french_lite_train.yml)
+Modify the training data path, dictionary and other parameters.
+
+For specific data preparation and training process, please refer to: [Text Detection](../../ppocr/model_train/detection.en.md), [Text Recognition](../../ppocr/model_train/recognition.en.md), more functions such as predictive deployment,
+For functions such as data annotation, you can read the complete [Document Tutorial](../../../index/index.en.md).
+
+## 4 Inference and Deployment
+
+In addition to installing the whl package for quick forecasting,
+PPOCR also provides a variety of forecasting deployment methods.
+If necessary, you can read related documents:
+
+- [Python Inference](../../legacy/python_infer.en.md)
+- [C++ Inference](../../legacy/cpp_infer.en.md)
+- [Serving](../../../../deploy/hubserving/readme_en.md)
+- [Mobile](../../../../deploy/lite/readme.md)
+- [Benchmark](../../legacy/benchmark.en.md)
+
+## 5 Support languages and abbreviations
+
+| Language  | Abbreviation | | Language  | Abbreviation |
+| ---  | --- | --- | ---  | --- |
+|Chinese & English|ch| |Arabic|ar|
+|English|en| |Hindi|hi|
+|French|fr| |Uyghur|ug|
+|German|german| |Persian|fa|
+|Japanese|japan| |Urdu|ur|
+|Korean|korean| | Serbian(latin) |rs_latin|
+|Chinese Traditional |chinese_cht| |Occitan |oc|
+| Italian |it| |Marathi|mr|
+|Spanish |es| |Nepali|ne|
+| Portuguese|pt| |Serbian(cyrillic)|rs_cyrillic|
+|Russian|ru||Bulgarian |bg|
+|Ukranian|uk| |Estonian |et|
+|Belarusian|be| |Irish |ga|
+|Telugu |te| |Croatian |hr|
+|Sanskrit|sa| |Hungarian |hu|
+|Tamil |ta| |Indonesian|id|
+|Afrikaans |af| |Icelandic|is|
+|Azerbaijani  |az||Kurdish|ku|
+|Bosnian|bs| |Lithuanian |lt|
+|Czech|cs| |Latvian |lv|
+|Welsh |cy| |Maori|mi|
+|Danish|da| |Malay|ms|
+|Maltese |mt| |Adyghe |ady|
+|Dutch |nl| |Kabardian |kbd|
+|Norwegian |no| |Avar |ava|
+|Polish |pl| |Dargwa |dar|
+|Romanian |ro| |Ingush |inh|
+|Slovak |sk| |Lak |lbe|
+|Slovenian |sl| |Lezghian |lez|
+|Albanian |sq| |Tabassaran |tab|
+|Swedish |sv| |Bihari |bh|
+|Swahili |sw| |Maithili |mai|
+|Tagalog |tl| |Angika |ang|
+|Turkish |tr| |Bhojpuri |bho|
+|Uzbek |uz| |Magahi |mah|
+|Vietnamese |vi| |Nagpur |sck|
+|Mongolian |mn| |Newari |new|
+|Abaza |abq| |Goan Konkani|gom|
+|Chechen |che| |Pali |pi|
+|Haryanvi |bgc| |Latin |la|
diff --git a/docs/version2.x/ppocr/blog/multi_languages.md b/docs/version2.x/ppocr/blog/multi_languages.md
new file mode 100644
index 0000000..f3cccc3
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/multi_languages.md
@@ -0,0 +1,275 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+# 多语言模型
+
+**近期更新**
+
+- 2022.5.8 更新`PP-OCRv3`版 多语言检测和识别模型，平均识别准确率提升5%以上。
+- 2021.4.9 支持**80种**语言的检测和识别
+- 2021.4.9 支持**轻量高精度**英文模型检测识别
+
+PaddleOCR 旨在打造一套丰富、领先、且实用的OCR工具库，不仅提供了通用场景下的中英文模型，也提供了专门在英文场景下训练的模型，
+和覆盖[80个语言](#5)的小语种模型。
+
+其中英文模型支持，大小写字母和常见标点的检测识别，并优化了空格字符的识别：
+
+![img](./images/img_12.jpg)
+
+小语种模型覆盖了拉丁语系、阿拉伯语系、中文繁体、韩语、日语等等：
+
+![img](./images/japan_2-20240709081138234.jpg)
+
+![img](./images/french_0.jpg)
+
+![img](./images/korean_0.jpg)
+
+![img](./images/arabic_0.jpg)
+
+本文档将简要介绍小语种模型的使用方法。
+
+## 1 安装
+
+### 1.1 paddle 安装
+
+```bash linenums="1"
+# cpu
+pip install "paddlepaddle<=2.6"
+
+# gpu
+pip install "paddlepaddle-gpu<=2.6"
+```
+
+### 1.2 paddleocr package 安装
+
+pip 安装
+
+```bash linenums="1"
+pip install "paddleocr<3.0"
+```
+
+本地构建并安装
+
+```bash linenums="1"
+python3 -m build
+pip3 install dist/paddleocr-x.x.x-py3-none-any.whl # x.x.x是paddleocr的版本号
+```
+
+## 2 快速使用
+
+### 2.1 命令行运行
+
+查看帮助信息
+
+```bash linenums="1"
+paddleocr -h
+```
+
+- 整图预测（检测+识别）
+
+Paddleocr目前支持80个语种，可以通过修改--lang参数进行切换，具体支持的[语种](#5)可查看表格。
+
+``` bash
+paddleocr --image_dir doc/imgs_en/254.jpg --lang=en
+```
+
+![](./images/254-20240709081442260.jpg)
+
+![img](./images/img_02.jpg)
+
+结果是一个list，每个item包含了文本框，文字和识别置信度
+
+```text linenums="1"
+[('PHO CAPITAL', 0.95723116), [[66.0, 50.0], [327.0, 44.0], [327.0, 76.0], [67.0, 82.0]]]
+[('107 State Street', 0.96311164), [[72.0, 90.0], [451.0, 84.0], [452.0, 116.0], [73.0, 121.0]]]
+[('Montpelier Vermont', 0.97389287), [[69.0, 132.0], [501.0, 126.0], [501.0, 158.0], [70.0, 164.0]]]
+[('8022256183', 0.99810505), [[71.0, 175.0], [363.0, 170.0], [364.0, 202.0], [72.0, 207.0]]]
+[('REG 07-24-201706:59 PM', 0.93537045), [[73.0, 299.0], [653.0, 281.0], [654.0, 318.0], [74.0, 336.0]]]
+[('045555', 0.99346405), [[509.0, 331.0], [651.0, 325.0], [652.0, 356.0], [511.0, 362.0]]]
+[('CT1', 0.9988654), [[535.0, 367.0], [654.0, 367.0], [654.0, 406.0], [535.0, 406.0]]]
+......
+```
+
+- 识别预测
+
+```bash linenums="1"
+paddleocr --image_dir doc/imgs_words_en/word_308.png --det false --lang=en
+```
+
+![img](./images/word_308.png)
+
+结果是一个tuple，返回识别结果和识别置信度
+
+```text linenums="1"
+(0.99879867, 'LITTLE')
+```
+
+- 检测预测
+
+```bash linenums="1"
+paddleocr --image_dir PaddleOCR/doc/imgs/11.jpg --rec false
+```
+
+结果是一个list，每个item只包含文本框
+
+```bash linenums="1"
+[[26.0, 457.0], [137.0, 457.0], [137.0, 477.0], [26.0, 477.0]]
+[[25.0, 425.0], [372.0, 425.0], [372.0, 448.0], [25.0, 448.0]]
+[[128.0, 397.0], [273.0, 397.0], [273.0, 414.0], [128.0, 414.0]]
+......
+```
+
+### 2.2 python 脚本运行
+
+ppocr 也支持在python脚本中运行，便于嵌入到您自己的代码中 ：
+
+- 整图预测（检测+识别）
+
+```python linenums="1"
+from paddleocr import PaddleOCR, draw_ocr
+
+# 同样也是通过修改 lang 参数切换语种
+ocr = PaddleOCR(lang="korean") # 首次执行会自动下载模型文件
+img_path = 'doc/imgs/korean_1.jpg '
+result = ocr.ocr(img_path)
+# 可通过参数控制单独执行识别、检测
+# result = ocr.ocr(img_path, det=False) 只执行识别
+# result = ocr.ocr(img_path, rec=False) 只执行检测
+# 打印检测框和识别结果
+for line in result:
+    print(line)
+
+# 可视化
+from PIL import Image
+image = Image.open(img_path).convert('RGB')
+boxes = [line[0] for line in result]
+txts = [line[1][0] for line in result]
+scores = [line[1][1] for line in result]
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/korean.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+结果可视化:
+
+![img](./images/korean.jpg)
+
+ppocr 还支持方向分类， 更多使用方式请参考：[whl包使用说明](whl.md)
+
+## 3 自定义训练
+
+ppocr 支持使用自己的数据进行自定义训练或finetune, 其中识别模型可以参考[法语配置文件](../../../../configs/rec/multi_language/rec_french_lite_train.yml)
+修改训练数据路径、字典等参数。
+
+详细数据准备、训练过程可参考：[文本识别](../../ppocr/model_train/recognition.md)、[文本检测](../../ppocr/model_train/detection.md)。
+
+假设已经准备好了训练数据，可根据以下步骤快速启动训练：
+
+- 修改配置文件
+
+以 `rec_french_lite_train.yml` 为例：
+
+```yaml linenums="1"
+Global:
+  ...
+  # 添加自定义字典，如修改字典请将路径指向新字典
+  character_dict_path: ./ppocr/utils/dict/french_dict.txt
+  ...
+  # 识别空格
+  use_space_char: True
+
+...
+
+Train:
+  dataset:
+    # 数据集格式，支持LMDBDataSet以及SimpleDataSet
+    name: SimpleDataSet
+    # 数据集路径
+    data_dir: ./train_data/
+    # 训练集标签文件
+    label_file_list: ["./train_data/french_train.txt"]
+    ...
+
+Eval:
+  dataset:
+    # 数据集格式，支持LMDBDataSet以及SimpleDataSet
+    name: SimpleDataSet
+    # 数据集路径
+    data_dir: ./train_data
+    # 验证集标签文件
+    label_file_list: ["./train_data/french_val.txt"]
+    ...
+```
+
+- 启动训练：
+
+```bash linenums="1"
+# 下载预训练模型
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/french_mobile_v2.0_rec_train.tar
+tar -xf french_mobile_v2.0_rec_train.tar
+
+#加载预训练模型 单卡训练
+python3 tools/train.py -c configs/rec/rec_french_lite_train.yml -o Global.pretrained_model=french_mobile_v2.0_rec_train/best_accuracy
+
+#加载预训练模型  多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/rec_french_lite_train.yml -o Global.pretrained_model=french_mobile_v2.0_rec_train/best_accuracy
+```
+
+更多功能如预测部署、数据标注等功能可以阅读完整的[文档教程](../../../index/index.md)。
+
+## 4 预测部署
+
+除了安装whl包进行快速预测，ppocr 也提供了多种预测部署方式，如有需求可阅读相关文档：
+
+- [基于Python脚本预测引擎推理](../../legacy/python_infer.md)
+- [基于C++预测引擎推理](../../legacy/cpp_infer.md)
+- [服务化部署](../../../../deploy/hubserving/readme.md)
+- [端侧部署](../../../../deploy/lite/readme_ch.md)
+- [Benchmark](../../legacy/benchmark.md)
+
+## 5 支持语种及缩写
+
+| 语种 | 描述 | 缩写 | | 语种 | 描述 | 缩写 |
+| --- | --- | --- | ---|--- | --- | --- |
+|中文|Chinese & English|ch| |保加利亚文|Bulgarian |bg|
+|英文|English|en| |乌克兰文|Ukranian|uk|
+|法文|French|fr| |白俄罗斯文|Belarusian|be|
+|德文|German|german| |泰卢固文|Telugu |te|
+|日文|Japanese|japan| | 阿巴扎文 |Abaza | abq |
+|韩文|Korean|korean| |泰米尔文|Tamil |ta|
+|中文繁体|Chinese Traditional |chinese_cht| |南非荷兰文 |Afrikaans |af|
+|意大利文| Italian |it| |阿塞拜疆文 |Azerbaijani    |az|
+|西班牙文|Spanish |es| |波斯尼亚文|Bosnian|bs|
+|葡萄牙文| Portuguese|pt| |捷克文|Czech|cs|
+|俄罗斯文|Russian|ru| |威尔士文 |Welsh |cy|
+|阿拉伯文|Arabic|ar| |丹麦文 |Danish|da|
+|印地文|Hindi|hi| |爱沙尼亚文 |Estonian |et|
+|维吾尔|Uyghur|ug| |爱尔兰文 |Irish |ga|
+|波斯文|Persian|fa| |克罗地亚文|Croatian |hr|
+|乌尔都文|Urdu|ur| |匈牙利文|Hungarian |hu|
+|塞尔维亚文（latin)| Serbian(latin) |rs_latin| |印尼文|Indonesian|id|
+|欧西坦文|Occitan |oc| |冰岛文 |Icelandic|is|
+|马拉地文|Marathi|mr| |库尔德文 |Kurdish|ku|
+|尼泊尔文|Nepali|ne| |立陶宛文|Lithuanian |lt|
+|塞尔维亚文（cyrillic)|Serbian(cyrillic)|rs_cyrillic| |拉脱维亚文 |Latvian |lv|
+|毛利文|Maori|mi| | 达尔瓦文|Dargwa |dar|
+|马来文 |Malay|ms| | 因古什文|Ingush |inh|
+|马耳他文 |Maltese |mt| | 拉克文|Lak |lbe|
+|荷兰文 |Dutch |nl| | 莱兹甘文|Lezghian |lez|
+|挪威文 |Norwegian |no| |塔巴萨兰文 |Tabassaran |tab|
+|波兰文|Polish |pl| | 比尔哈文|Bihari |bh|
+| 罗马尼亚文|Romanian |ro| | 迈蒂利文|Maithili |mai|
+| 斯洛伐克文|Slovak |sk| | 昂加文|Angika |ang|
+| 斯洛文尼亚文|Slovenian |sl| | 孟加拉文|Bhojpuri |bho|
+| 阿尔巴尼亚文|Albanian |sq| | 摩揭陀文 |Magahi |mah|
+| 瑞典文|Swedish |sv| | 那格浦尔文|Nagpur |sck|
+| 西瓦希里文|Swahili |sw| | 尼瓦尔文|Newari |new|
+| 塔加洛文|Tagalog |tl| | 保加利亚文 |Goan Konkani|gom|
+| 土耳其文|Turkish |tr| | 梵文|Sanskrit|sa|
+| 乌兹别克文|Uzbek |uz| | 阿瓦尔文|Avar |ava|
+| 越南文|Vietnamese |vi| | 阿瓦尔文|Avar |ava|
+| 蒙古文|Mongolian |mn| | 阿迪赫文|Adyghe |ady|
+| 车臣文 |Chechen |che| | 巴利文 |Pali |pi|
+| 哈里亚纳语 |Haryanvi |bgc| | 拉丁文 |Latin |la|
diff --git a/docs/version2.x/ppocr/blog/ocr_book.en.md b/docs/version2.x/ppocr/blog/ocr_book.en.md
new file mode 100644
index 0000000..3228eb9
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/ocr_book.en.md
@@ -0,0 +1,28 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+# E-book: *Dive Into OCR*
+
+"Dive Into OCR" is a textbook that combines OCR theory and practice, written by the PaddleOCR community. The main features are as follows:
+
+- OCR full-stack technology covering text detection, recognition and document analysis
+- Closely integrate theory and practice, cross the code implementation gap, and supporting instructional videos
+- Jupyter Notebook textbook, flexibly modifying code for instant results
+
+## Structure
+
+![img](./images/187578511-9f3c351e-b68c-4359-a6e5-475810993c61.png)
+
+- The first part is the preliminary knowledge of the book, including the knowledge index and resource links needed in the process of positioning and using the book content of the book
+
+- The second part is chapters 4-8 of the book, which introduce the concepts, applications, and industry practices related to the detection and identification capabilities of the OCR engine. In the "Introduction to OCR Technology", the application scenarios and challenges of OCR, the basic concepts of technology, and the pain points in industrial applications are comprehensively explained. Then, in the two chapters of "Text Detection" and "Text Recognition", the two basic tasks of OCR are introduced. In each chapter, an algorithm is accompanied by a detailed explanation of the code and practical exercises. Chapters 6 and 7 are a detailed introduction to the PP-OCR series model, PP-OCR is a set of OCR systems for industrial applications, on the basis of the basic detection and identification model, after a series of optimization strategies to achieve the general field of industrial SOTA model, while opening up a variety of predictive deployment solutions, enabling enterprises to quickly land OCR applications.
+
+- The third part is chapter 9-12 of the book, which introduces applications other than the two-stage OCR engine, including data synthesis, preprocessing algorithm, and end-to-end model, focusing on OCR's layout analysis, table recognition, visual document question and answer capabilities in the document scene, and also through the combination of algorithm and code, so that readers can deeply understand and apply.
+
+## Address
+
+- [E-book: *Dive Into OCR* (PDF)](https://paddleocr.bj.bcebos.com/ebook/Dive_into_OCR.pdf)
+- [Notebook (.ipynb)](https://github.com/PaddleOCR-Community/Dive-into-OCR)
+- [Videos (Chinese only)](https://aistudio.baidu.com/aistudio/education/group/info/25207)
diff --git a/docs/version2.x/ppocr/blog/ocr_book.md b/docs/version2.x/ppocr/blog/ocr_book.md
new file mode 100644
index 0000000..f333b19
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/ocr_book.md
@@ -0,0 +1,29 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+# 《动手学OCR》电子书
+
+《动手学OCR》是PaddleOCR团队携手华中科技大学博导/教授，IAPR Fellow 白翔、复旦大学青年研究员陈智能、中国移动研究院视觉领域资深专家黄文辉、中国工商银行大数据人工智能实验室研究员等产学研同仁，以及OCR开发者共同打造的结合OCR前沿理论与代码实践的教材。主要特色如下：
+
+- 覆盖从文本检测识别到文档分析的OCR全栈技术
+- 紧密结合理论实践，跨越代码实现鸿沟，并配套教学视频
+- Notebook交互式学习，灵活修改代码，即刻获得结果
+
+## 本书结构
+
+![](./images/5e612.png)
+
+- 第一部分是本书的推荐序、序言与预备知识，包含本书的定位与使用书籍内容的过程中需要用到的知识索引、资源链接等
+- 第二部分是本书的4-8章，介绍与OCR核心的检测、识别能力相关的概念、应用与产业实践。在“OCR技术导论”中总括性的解释OCR的应用场景和挑战、技术基本概念以及在产业应用中的痛点问题。然后在
+“文本检测”与“文本识别”两章中介绍OCR的两个基本任务，并在每章中配套一个算法展开代码详解与实战练习。第6、7章是关于PP-OCR系列模型的详细介绍，PP-OCR是一套面向产业应用的OCR系统，在
+基础检测和识别模型的基础之上经过一系列优化策略达到通用领域的产业级SOTA模型，同时打通多种预测部署方案，赋能企业快速落地OCR应用。
+- 第三部分是本书的9-12章，介绍两阶段OCR引擎之外的应用，包括数据合成、预处理算法、端到端模型，重点展开了OCR在文档场景下的版面分析、表格识别、视觉文档问答的能力，同样通过算法与代码结
+合的方式使得读者能够深入理解并应用。
+
+## 资料地址
+
+-  🎁《动手学OCR》、《OCR产业范例20讲》电子书、OCR垂类模型、PDF2Word软件以及其他学习大礼包领取链接：[百度网盘 PaddleOCR 开源大礼包](https://pan.baidu.com/s/1h__OYbbQ_rASB-m8ZC1bCA)，提取码：4232
+- [notebook教程](https://github.com/PaddleOCR-Community/Dive-into-OCR)
+- [教学视频](https://aistudio.baidu.com/aistudio/education/group/info/25207)
diff --git a/docs/version2.x/ppocr/blog/slice.en.md b/docs/version2.x/ppocr/blog/slice.en.md
new file mode 100644
index 0000000..0b8b95d
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/slice.en.md
@@ -0,0 +1,22 @@
+---
+comments: true
+---
+
+
+# Slice Operator
+
+If you have a very large image/document that you would like to run PaddleOCR (detection and recognition) on, you can use the slice operation as follows:
+
+`ocr_inst  =  PaddleOCR(**ocr_settings)`
+`results  =  ocr_inst.ocr(img, det=True,rec=True, slice=slice, cls=False,bin=False,inv=False,alpha_color=False)`
+
+where
+`slice  = {'horizontal_stride': h_stride, 'vertical_stride':v_stride, 'merge_x_thres':x_thres, 'merge_y_thres': y_thres}`
+
+Here, `h_stride`, `v_stride`, `x_thres`, and `y_thres` are user-configurable values and need to be set manually. The way the `slice` operator works is that it runs a sliding window across the large input image, creating slices of it and runs the OCR algorithms on it.
+
+The fragmented slice-level results are then merged together to output image-level detection and recognition results. The horizontal and vertical strides cannot be lower than a certain limit (as too low values would create so many slices it would be very computationally expensive to get results for each of them). However, as an example the recommended values for an image with dimensions 6616x14886 would be as follows.
+
+`slice = {'horizontal_stride': 300, 'vertical_stride':500, 'merge_x_thres':50, 'merge_y_thres': 35}`
+
+All slice-level detections with bounding boxes as close as `merge_x_thres` and `merge_y_thres` will be merged together.
diff --git a/docs/version2.x/ppocr/blog/slice.md b/docs/version2.x/ppocr/blog/slice.md
new file mode 100644
index 0000000..eae91ad
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/slice.md
@@ -0,0 +1,25 @@
+---
+comments: true
+---
+
+# 切片操作
+
+如果希望运行 PaddleOCR 处理一张非常大的图像或文档，对其进行检测和识别，可以使用切片操作，如下所示：
+
+```python linenums="1"
+ocr_inst = PaddleOCR(**ocr_settings)
+results = ocr_inst.ocr(img, det=True, rec=True, slice=slice, cls=False, bin=False, inv=False, alpha_color=False)
+```
+
+其中，
+`slice = {'horizontal_stride': h_stride, 'vertical_stride': v_stride, 'merge_x_thres': x_thres, 'merge_y_thres': y_thres}`
+
+这里的 `h_stride`、`v_stride`、`x_thres` 和 `y_thres` 是用户可配置的参数，需要手动设置。切片操作符的工作原理是，在大图像上运行一个滑动窗口，创建图像的切片，并在这些切片上运行 OCR 算法。
+
+然后将这些切片级别的零散结果合并，生成图像级别的检测和识别结果。水平和垂直步幅不能低于一定限度，因为过低的值会产生太多切片，导致计算结果非常耗时。例如，对于尺寸为 6616x14886 的图像，推荐使用以下参数：
+
+```python linenums="1"
+slice = {'horizontal_stride': 300, 'vertical_stride': 500, 'merge_x_thres': 50, 'merge_y_thres': 35}
+```
+
+所有边界框接近 `merge_x_thres` 和 `merge_y_thres` 的切片级检测结果将被合并在一起。
diff --git a/docs/version2.x/ppocr/blog/whl.en.md b/docs/version2.x/ppocr/blog/whl.en.md
new file mode 100644
index 0000000..f3ba202
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/whl.en.md
@@ -0,0 +1,488 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# Paddleocr Package
+
+## 1 Get started quickly
+
+### 1.1 Install package
+
+install by pypi
+
+```bash linenums="1"
+pip install "paddleocr>=2.0.1" # Recommend to use version 2.0.1+
+```
+
+build own whl package and install
+
+```bash linenums="1"
+python3 -m build
+pip3 install dist/paddleocr-x.x.x-py3-none-any.whl # x.x.x is the version of paddleocr
+```
+
+## 2 Use
+
+### 2.1 Use by code
+
+The paddleocr whl package will automatically download the ppocr lightweight model as the default model, which can be customized and replaced according to the section 3 **Custom Model**.
+
+* detection angle classification and recognition
+
+```python linenums="1"
+from paddleocr import PaddleOCR,draw_ocr
+# Paddleocr supports Chinese, English, French, German, Korean and Japanese.
+# You can set the parameter `lang` as `ch`, `en`, `french`, `german`, `korean`, `japan`
+# to switch the language model in order.
+ocr = PaddleOCR(use_angle_cls=True, lang='en') # need to run only once to download and load model into memory
+img_path = 'PaddleOCR/doc/imgs_en/img_12.jpg'
+result = ocr.ocr(img_path, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+
+# draw result
+from PIL import Image
+result = result[0]
+image = Image.open(img_path).convert('RGB')
+boxes = [line[0] for line in result]
+txts = [line[1][0] for line in result]
+scores = [line[1][1] for line in result]
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+Output will be a list, each item contains bounding box, text and recognition confidence
+
+```bash linenums="1"
+[[[442.0, 173.0], [1169.0, 173.0], [1169.0, 225.0], [442.0, 225.0]], ['ACKNOWLEDGEMENTS', 0.99283075]]
+[[[393.0, 340.0], [1207.0, 342.0], [1207.0, 389.0], [393.0, 387.0]], ['We would like to thank all the designers and', 0.9357758]]
+[[[399.0, 398.0], [1204.0, 398.0], [1204.0, 433.0], [399.0, 433.0]], ['contributors whohave been involved in the', 0.9592447]]
+......
+```
+
+Visualization of results
+
+![img](./images/12_det_rec.jpg)
+
+* detection and recognition
+
+```python linenums="1"
+from paddleocr import PaddleOCR,draw_ocr
+ocr = PaddleOCR(lang='en') # need to run only once to download and load model into memory
+img_path = 'PaddleOCR/doc/imgs_en/img_12.jpg'
+result = ocr.ocr(img_path, cls=False)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+
+# draw result
+from PIL import Image
+result = result[0]
+image = Image.open(img_path).convert('RGB')
+boxes = [line[0] for line in result]
+txts = [line[1][0] for line in result]
+scores = [line[1][1] for line in result]
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+Output will be a list, each item contains bounding box, text and recognition confidence
+
+```bash linenums="1"
+[[[442.0, 173.0], [1169.0, 173.0], [1169.0, 225.0], [442.0, 225.0]], ['ACKNOWLEDGEMENTS', 0.99283075]]
+[[[393.0, 340.0], [1207.0, 342.0], [1207.0, 389.0], [393.0, 387.0]], ['We would like to thank all the designers and', 0.9357758]]
+[[[399.0, 398.0], [1204.0, 398.0], [1204.0, 433.0], [399.0, 433.0]], ['contributors whohave been involved in the', 0.9592447]]
+......
+```
+
+Visualization of results
+
+![img](./images/12_det_rec.jpg)
+
+* classification and recognition
+
+```python linenums="1"
+from paddleocr import PaddleOCR
+ocr = PaddleOCR(use_angle_cls=True, lang='en') # need to run only once to load model into memory
+img_path = 'PaddleOCR/doc/imgs_words_en/word_10.png'
+result = ocr.ocr(img_path, det=False, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+```
+
+Output will be a list, each item contains recognition text and confidence
+
+```bash linenums="1"
+['PAIN', 0.990372]
+```
+
+* only detection
+
+```python linenums="1"
+from paddleocr import PaddleOCR,draw_ocr
+ocr = PaddleOCR() # need to run only once to download and load model into memory
+img_path = 'PaddleOCR/doc/imgs_en/img_12.jpg'
+result = ocr.ocr(img_path,rec=False)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+
+# draw result
+from PIL import Image
+result = result[0]
+image = Image.open(img_path).convert('RGB')
+im_show = draw_ocr(image, result, txts=None, scores=None, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+Output will be a list, each item only contains bounding box
+
+```bash linenums="1"
+[[756.0, 812.0], [805.0, 812.0], [805.0, 830.0], [756.0, 830.0]]
+[[820.0, 803.0], [1085.0, 801.0], [1085.0, 836.0], [820.0, 838.0]]
+[[393.0, 801.0], [715.0, 805.0], [715.0, 839.0], [393.0, 836.0]]
+......
+```
+
+Visualization of results
+
+![img](./images/12_det.jpg)
+
+* only recognition
+
+```python linenums="1"
+from paddleocr import PaddleOCR
+ocr = PaddleOCR(lang='en') # need to run only once to load model into memory
+img_path = 'PaddleOCR/doc/imgs_words_en/word_10.png'
+result = ocr.ocr(img_path, det=False, cls=False)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+```
+
+Output will be a list, each item contains recognition text and confidence
+
+```bash linenums="1"
+['PAIN', 0.990372]
+```
+
+* only classification
+
+```python linenums="1"
+from paddleocr import PaddleOCR
+ocr = PaddleOCR(use_angle_cls=True) # need to run only once to load model into memory
+img_path = 'PaddleOCR/doc/imgs_words_en/word_10.png'
+result = ocr.ocr(img_path, det=False, rec=False, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+```
+
+Output will be a list, each item contains classification result and confidence
+
+```bash linenums="1"
+['0', 0.99999964]
+```
+
+### 2.2 Use by command line
+
+show help information
+
+```bash linenums="1"
+paddleocr -h
+```
+
+* detection classification and recognition
+
+```bash linenums="1"
+paddleocr --image_dir PaddleOCR/doc/imgs_en/img_12.jpg --use_angle_cls true --lang en
+```
+
+Output will be a list, each item contains bounding box, text and recognition confidence
+
+```bash linenums="1"
+[[[441.0, 174.0], [1166.0, 176.0], [1165.0, 222.0], [441.0, 221.0]], ('ACKNOWLEDGEMENTS', 0.9971134662628174)]
+[[[403.0, 346.0], [1204.0, 348.0], [1204.0, 384.0], [402.0, 383.0]], ('We would like to thank all the designers and', 0.9761400818824768)]
+[[[403.0, 396.0], [1204.0, 398.0], [1204.0, 434.0], [402.0, 433.0]], ('contributors who have been involved in the', 0.9791957139968872)]
+......
+```
+
+pdf file is also supported, you can infer the first few pages by using the `page_num` parameter, the default is 0, which means infer all pages
+
+```bash linenums="1"
+paddleocr --image_dir ./xxx.pdf --use_angle_cls true --use_gpu false --page_num 2
+```
+
+* detection and recognition
+
+```bash linenums="1"
+paddleocr --image_dir PaddleOCR/doc/imgs_en/img_12.jpg --lang en
+```
+
+Output will be a list, each item contains bounding box, text and recognition confidence
+
+```bash linenums="1"
+[[[441.0, 174.0], [1166.0, 176.0], [1165.0, 222.0], [441.0, 221.0]], ('ACKNOWLEDGEMENTS', 0.9971134662628174)]
+[[[403.0, 346.0], [1204.0, 348.0], [1204.0, 384.0], [402.0, 383.0]], ('We would like to thank all the designers and', 0.9761400818824768)]
+[[[403.0, 396.0], [1204.0, 398.0], [1204.0, 434.0], [402.0, 433.0]], ('contributors who have been involved in the', 0.9791957139968872)]
+......
+```
+
+* classification and recognition
+
+```bash linenums="1"
+paddleocr --image_dir PaddleOCR/doc/imgs_words_en/word_10.png --use_angle_cls true --det false --lang en
+```
+
+Output will be a list, each item contains text and recognition confidence
+
+```bash linenums="1"
+['PAIN', 0.9934559464454651]
+```
+
+* only detection
+
+```bash linenums="1"
+paddleocr --image_dir PaddleOCR/doc/imgs_en/img_12.jpg --rec false
+```
+
+Output will be a list, each item only contains bounding box
+
+```bash linenums="1"
+[[397.0, 802.0], [1092.0, 802.0], [1092.0, 841.0], [397.0, 841.0]]
+[[397.0, 750.0], [1211.0, 750.0], [1211.0, 789.0], [397.0, 789.0]]
+[[397.0, 702.0], [1209.0, 698.0], [1209.0, 734.0], [397.0, 738.0]]
+......
+```
+
+* only recognition
+
+```bash linenums="1"
+paddleocr --image_dir PaddleOCR/doc/imgs_words_en/word_10.png --det false --lang en
+```
+
+Output will be a list, each item contains text and recognition confidence
+
+```bash linenums="1"
+['PAIN', 0.9934559464454651]
+```
+
+* only classification
+
+```bash linenums="1"
+paddleocr --image_dir PaddleOCR/doc/imgs_words_en/word_10.png --use_angle_cls true --det false --rec false
+```
+
+Output will be a list, each item contains classification result and confidence
+
+```bash linenums="1"
+['0', 0.99999964]
+```
+
+## 3 Use custom model
+
+When the built-in model cannot meet the needs, you need to use your own trained model.
+First, refer to [export](../model_train/detection.en.md#4-inference) doc to convert your det and rec model to inference model, and then use it as follows
+
+### 3.1 Use by code
+
+```python linenums="1"
+from paddleocr import PaddleOCR,draw_ocr
+# The path of detection and recognition model must contain model and params files
+ocr = PaddleOCR(det_model_dir='{your_det_model_dir}', rec_model_dir='{your_rec_model_dir}', rec_char_dict_path='{your_rec_char_dict_path}', cls_model_dir='{your_cls_model_dir}', use_angle_cls=True)
+img_path = 'PaddleOCR/doc/imgs_en/img_12.jpg'
+result = ocr.ocr(img_path, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+
+# draw result
+from PIL import Image
+result = result[0]
+image = Image.open(img_path).convert('RGB')
+boxes = [line[0] for line in result]
+txts = [line[1][0] for line in result]
+scores = [line[1][1] for line in result]
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+### 3.2 Use by command line
+
+```bash linenums="1"
+paddleocr --image_dir PaddleOCR/doc/imgs/11.jpg --det_model_dir {your_det_model_dir} --rec_model_dir {your_rec_model_dir} --rec_char_dict_path {your_rec_char_dict_path} --cls_model_dir {your_cls_model_dir} --use_angle_cls true
+```
+
+## 4 Use web images or numpy array as input
+
+### 4.1 Web image
+
+* Use by code
+
+```python linenums="1"
+from paddleocr import PaddleOCR, draw_ocr
+ocr = PaddleOCR(use_angle_cls=True, lang="ch") # need to run only once to download and load model into memory
+img_path = 'http://n.sinaimg.cn/ent/transform/w630h933/20171222/o111-fypvuqf1838418.jpg'
+result = ocr.ocr(img_path, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+
+# show result
+from PIL import Image
+result = result[0]
+image = Image.open(img_path).convert('RGB')
+boxes = [line[0] for line in result]
+txts = [line[1][0] for line in result]
+scores = [line[1][1] for line in result]
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+* Use by command line
+
+```bash linenums="1"
+paddleocr --image_dir http://n.sinaimg.cn/ent/transform/w630h933/20171222/o111-fypvuqf1838418.jpg --use_angle_cls=true
+```
+
+### 4.2 Numpy array
+
+Support numpy array as input only when used by code
+
+```python linenums="1"
+import cv2
+from paddleocr import PaddleOCR, draw_ocr, download_with_progressbar
+ocr = PaddleOCR(use_angle_cls=True, lang="ch") # need to run only once to download and load model into memory
+img_path = 'PaddleOCR/doc/imgs/11.jpg'
+img = cv2.imread(img_path)
+# img = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY), If your own training model supports grayscale images, you can uncomment this line
+result = ocr.ocr(img, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+
+# show result
+from PIL import Image
+result = result[0]
+download_with_progressbar(img_path, 'tmp.jpg')
+image = Image.open('tmp.jpg').convert('RGB')
+boxes = [line[0] for line in result]
+txts = [line[1][0] for line in result]
+scores = [line[1][1] for line in result]
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+## 5 PDF file
+
+* Use by command line
+
+you can infer the first few pages by using the `page_num` parameter, the default is 0, which means infer all pages
+
+```bash linenums="1"
+paddleocr --image_dir ./xxx.pdf --use_angle_cls true --use_gpu false --page_num 2
+```
+
+* Use by code
+
+```python linenums="1"
+from paddleocr import PaddleOCR, draw_ocr
+
+# Paddleocr supports Chinese, English, French, German, Korean and Japanese.
+# You can set the parameter `lang` as `ch`, `en`, `fr`, `german`, `korean`, `japan`
+# to switch the language model in order.
+ocr = PaddleOCR(use_angle_cls=True, lang="ch"， page_num=2)  # need to run only once to download and load model into memory
+img_path = './xxx.pdf'
+result = ocr.ocr(img_path, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+
+# draw result
+import fitz
+from PIL import Image
+import cv2
+import numpy as np
+imgs = []
+with fitz.open(img_path) as pdf:
+    for pg in range(0, pdf.pageCount):
+        page = pdf[pg]
+        mat = fitz.Matrix(2, 2)
+        pm = page.getPixmap(matrix=mat, alpha=False)
+        # if width or height > 2000 pixels, don't enlarge the image
+        if pm.width > 2000 or pm.height > 2000:
+            pm = page.getPixmap(matrix=fitz.Matrix(1, 1), alpha=False)
+
+        img = Image.frombytes("RGB", [pm.width, pm.height], pm.samples)
+        img = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
+        imgs.append(img)
+for idx in range(len(result)):
+    res = result[idx]
+    image = imgs[idx]
+    boxes = [line[0] for line in res]
+    txts = [line[1][0] for line in res]
+    scores = [line[1][1] for line in res]
+    im_show = draw_ocr(image, boxes, txts, scores, font_path='doc/fonts/simfang.ttf')
+    im_show = Image.fromarray(im_show)
+    im_show.save('result_page_{}.jpg'.format(idx))
+```
+
+## 6 Parameter Description
+
+| Parameter                    | Description                                                                                                                                                                                                                 | Default value                  |
+|-------------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-------------------------|
+| use_gpu                 | use GPU or not                                                                                                                                                                                                          | TRUE                    |
+| gpu_mem                 | GPU memory size used for initialization                                                                                                                                                                                              | 8000M                   |
+| image_dir               | The images path or folder path for predicting when used by the command line                                                                                                                                                                           |                         |
+| page_num               | Valid when the input type is pdf file, specify to predict the previous page_num pages, all pages are predicted by default                                                                                                                                                                           |          0               |
+| det_algorithm           | Type of detection algorithm selected                                                                                                                                                                                                   | DB                      |
+| det_model_dir           | the text detection inference model folder. There are two ways to transfer parameters, 1. None: Automatically download the built-in model to `~/.paddleocr/det`; 2. The path of the inference model converted by yourself, the model and params files must be included in the model path | None           |
+| det_max_side_len        | The maximum size of the long side of the image. When the long side exceeds this value, the long side will be resized to this size, and the short side will be scaled proportionally                                                                                                                         | 960                     |
+| det_db_thresh           | Binarization threshold value of DB output map                                                                                                                                                                                        | 0.3                     |
+| det_db_box_thresh       | The threshold value of the DB output box. Boxes score lower than this value will be discarded                                                                                                                                                                         | 0.5                     |
+| det_db_unclip_ratio     | The expanded ratio of DB output box                                                                                                                                                                                             | 2                       |
+| det_db_score_mode |  The parameter that control how the score of the detection frame is calculated. There are 'fast' and 'slow' options. If the text to be detected is curved, it is recommended to use 'slow'  | 'fast' |
+| det_east_score_thresh   | Binarization threshold value of EAST output map                                                                                                                                                                                       | 0.8                     |
+| det_east_cover_thresh   | The threshold value of the EAST output box. Boxes score lower than this value will be discarded                                                                                                                                                                         | 0.1                     |
+| det_east_nms_thresh     | The NMS threshold value of EAST model output box                                                                                                                                                                                              | 0.2                     |
+| rec_algorithm           | Type of recognition algorithm selected                                                                                                                                                                                                | CRNN                    |
+| rec_model_dir           | the text recognition inference model folder. There are two ways to transfer parameters, 1. None: Automatically download the built-in model to `~/.paddleocr/rec`; 2. The path of the inference model converted by yourself, the model and params files must be included in the model path | None |
+| rec_image_shape         | image shape of recognition algorithm                                                                                                                                                                                            | "3,32,320"              |
+| rec_batch_num           | When performing recognition, the batchsize of forward images                                                                                                                                                                                         | 30                      |
+| max_text_length         | The maximum text length that the recognition algorithm can recognize                                                                                                                                                                                         | 25                      |
+| rec_char_dict_path      | the alphabet path which needs to be modified to your own path when `rec_model_Name` use mode 2                                                                                                                                              | ./ppocr/utils/ppocr_keys_v1.txt                        |
+| use_space_char          | Whether to recognize spaces                                                                                                                                                                                                         | TRUE                    |
+| drop_score          | Filter the output by score (from the recognition model), and those below this score will not be returned                                                                                                                                                                                                        | 0.5                    |
+| use_angle_cls          | Whether to load classification model                                                                                                                                                                                                       | FALSE                    |
+| cls_model_dir           | the classification inference model folder. There are two ways to transfer parameters, 1. None: Automatically download the built-in model to `~/.paddleocr/cls`; 2. The path of the inference model converted by yourself, the model and params files must be included in the model path | None |
+| cls_image_shape         | image shape of classification algorithm                                                                                                                                                                                            | "3,48,192"              |
+| label_list         | label list of classification algorithm                                                                                                                                                                                            | ['0','180']           |
+| cls_batch_num           | When performing classification, the batchsize of forward images                                                                                                                                                                                         | 30                      |
+| enable_mkldnn           | Whether to enable mkldnn                                                                                                                                                                                                       | FALSE                   |
+| use_zero_copy_run           | Whether to forward by zero_copy_run                                                                                                                                                                               | FALSE                   |
+| lang                     | The support language, now only Chinese(ch)、English(en)、French(french)、German(german)、Korean(korean)、Japanese(japan) are supported                                                                                                                                                                                                  | ch                    |
+| det                     | Enable detection when `ppocr.ocr` func exec                                                                                                                                                                                                   | TRUE                    |
+| rec                     | Enable recognition when `ppocr.ocr` func exec                                                                                                                                                                                                   | TRUE                    |
+| cls                     | Enable classification when `ppocr.ocr` func exec((Use use_angle_cls in command line mode to control whether to start classification in the forward direction)                                                                                                                                                                                                   | FALSE                    |
+| show_log                     | Whether to print log| FALSE                    |
+| type                     | Perform ocr or table structuring, the value is selected in ['ocr','structure']                                                                                                                                                                                             | ocr                    |
+| ocr_version                     | OCR Model version number, the current model support list is as follows: PP-OCRv3 supports Chinese and English detection, recognition, multilingual recognition, direction classifier models, PP-OCRv2 support Chinese detection and recognition model, PP-OCR support Chinese detection, recognition and direction classifier, multilingual recognition model | PP-OCRv3                 |
diff --git a/docs/version2.x/ppocr/blog/whl.md b/docs/version2.x/ppocr/blog/whl.md
new file mode 100644
index 0000000..87b9bb2
--- /dev/null
+++ b/docs/version2.x/ppocr/blog/whl.md
@@ -0,0 +1,501 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# paddleocr package使用说明
+
+## 1 快速上手
+
+### 1.1 安装whl包
+
+pip安装
+
+```bash linenums="1"
+pip install "paddleocr<3.0"
+```
+
+本地构建并安装
+
+```bash linenums="1"
+python3 -m build
+pip3 install dist/paddleocr-x.x.x-py3-none-any.whl # x.x.x是paddleocr的版本号
+```
+
+## 2 使用
+
+### 2.1 代码使用
+
+paddleocr whl包会自动下载ppocr轻量级模型作为默认模型，可以根据第3节**自定义模型**进行自定义更换。
+
+* 检测+方向分类器+识别全流程
+
+```python linenums="1"
+from paddleocr import PaddleOCR, draw_ocr
+
+# Paddleocr目前支持中英文、英文、法语、德语、韩语、日语，可以通过修改lang参数进行切换
+# 参数依次为`ch`, `en`, `french`, `german`, `korean`, `japan`。
+ocr = PaddleOCR(use_angle_cls=True, lang="ch")  # need to run only once to download and load model into memory
+img_path = 'PaddleOCR/doc/imgs/11.jpg'
+result = ocr.ocr(img_path, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+
+# 显示结果
+from PIL import Image
+result = result[0]
+image = Image.open(img_path).convert('RGB')
+boxes = [line[0] for line in result]
+txts = [line[1][0] for line in result]
+scores = [line[1][1] for line in result]
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+结果是一个list，每个item包含了文本框，文字和识别置信度
+
+```bash linenums="1"
+[[[24.0, 36.0], [304.0, 34.0], [304.0, 72.0], [24.0, 74.0]], ['纯臻营养护发素', 0.964739]]
+[[[24.0, 80.0], [172.0, 80.0], [172.0, 104.0], [24.0, 104.0]], ['产品信息/参数', 0.98069626]]
+[[[24.0, 109.0], [333.0, 109.0], [333.0, 136.0], [24.0, 136.0]], ['（45元/每公斤，100公斤起订）', 0.9676722]]
+......
+```
+
+结果可视化
+
+<div align="center">
+    <img src="./images/11_det_rec.jpg" width="800">
+</div>
+
+* 检测+识别
+
+```python linenums="1"
+from paddleocr import PaddleOCR, draw_ocr
+
+ocr = PaddleOCR()  # need to run only once to download and load model into memory
+img_path = 'PaddleOCR/doc/imgs/11.jpg'
+result = ocr.ocr(img_path, cls=False)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+
+# 显示结果
+from PIL import Image
+result = result[0]
+image = Image.open(img_path).convert('RGB')
+boxes = [line[0] for line in result]
+txts = [line[1][0] for line in result]
+scores = [line[1][1] for line in result]
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+结果是一个list，每个item包含了文本框，文字和识别置信度
+
+```bash linenums="1"
+[[[24.0, 36.0], [304.0, 34.0], [304.0, 72.0], [24.0, 74.0]], ['纯臻营养护发素', 0.964739]]
+[[[24.0, 80.0], [172.0, 80.0], [172.0, 104.0], [24.0, 104.0]], ['产品信息/参数', 0.98069626]]
+[[[24.0, 109.0], [333.0, 109.0], [333.0, 136.0], [24.0, 136.0]], ['（45元/每公斤，100公斤起订）', 0.9676722]]
+......
+```
+
+结果可视化
+
+<div align="center">
+    <img src="./images/11_det_rec.jpg" width="800">
+</div>
+
+* 方向分类器+识别
+
+```python linenums="1"
+from paddleocr import PaddleOCR
+
+ocr = PaddleOCR(use_angle_cls=True)  # need to run only once to download and load model into memory
+img_path = 'PaddleOCR/doc/imgs_words/ch/word_1.jpg'
+result = ocr.ocr(img_path, det=False, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+```
+
+结果是一个list，每个item只包含识别结果和识别置信度
+
+```bash linenums="1"
+['韩国小馆', 0.9907421]
+```
+
+* 单独执行检测
+
+```python linenums="1"
+from paddleocr import PaddleOCR, draw_ocr
+
+ocr = PaddleOCR()  # need to run only once to download and load model into memory
+img_path = 'PaddleOCR/doc/imgs/11.jpg'
+result = ocr.ocr(img_path, rec=False)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+
+# 显示结果
+from PIL import Image
+result = result[0]
+image = Image.open(img_path).convert('RGB')
+im_show = draw_ocr(image, result, txts=None, scores=None, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+结果是一个list，每个item只包含文本框
+
+```bash linenums="1"
+[[26.0, 457.0], [137.0, 457.0], [137.0, 477.0], [26.0, 477.0]]
+[[25.0, 425.0], [372.0, 425.0], [372.0, 448.0], [25.0, 448.0]]
+[[128.0, 397.0], [273.0, 397.0], [273.0, 414.0], [128.0, 414.0]]
+......
+```
+
+结果可视化
+
+<div align="center">
+    <img src="./images/11_det.jpg" width="800">
+</div>
+
+* 单独执行识别
+
+```python linenums="1"
+from paddleocr import PaddleOCR
+
+ocr = PaddleOCR()  # need to run only once to download and load model into memory
+img_path = 'PaddleOCR/doc/imgs_words/ch/word_1.jpg'
+result = ocr.ocr(img_path, det=False)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+```
+
+结果是一个list，每个item只包含识别结果和识别置信度
+
+```bash linenums="1"
+['韩国小馆', 0.9907421]
+```
+
+* 单独执行方向分类器
+
+```python linenums="1"
+from paddleocr import PaddleOCR
+
+ocr = PaddleOCR(use_angle_cls=True)  # need to run only once to download and load model into memory
+img_path = 'PaddleOCR/doc/imgs_words/ch/word_1.jpg'
+result = ocr.ocr(img_path, det=False, rec=False, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+```
+
+结果是一个list，每个item只包含分类结果和分类置信度
+
+```bash linenums="1"
+['0', 0.9999924]
+```
+
+### 2.2 通过命令行使用
+
+查看帮助信息
+
+```bash linenums="1"
+paddleocr -h
+```
+
+* 检测+方向分类器+识别全流程
+
+```bash linenums="1"
+paddleocr --image_dir PaddleOCR/doc/imgs/11.jpg --use_angle_cls true
+```
+
+结果是一个list，每个item包含了文本框，文字和识别置信度
+
+```bash linenums="1"
+[[[28.0, 37.0], [302.0, 39.0], [302.0, 72.0], [27.0, 70.0]], ('纯臻营养护发素', 0.9658738374710083)]
+......
+```
+
+此外，paddleocr也支持输入pdf文件，并且可以通过指定参数`page_num`来控制推理前面几页，默认为0，表示推理所有页。
+
+```bash linenums="1"
+paddleocr --image_dir ./xxx.pdf --use_angle_cls true --use_gpu false --page_num 2
+```
+
+* 检测+识别
+
+```bash linenums="1"
+paddleocr --image_dir PaddleOCR/doc/imgs/11.jpg
+```
+
+结果是一个list，每个item包含了文本框，文字和识别置信度
+
+```bash linenums="1"
+[[[28.0, 37.0], [302.0, 39.0], [302.0, 72.0], [27.0, 70.0]], ('纯臻营养护发素', 0.9658738374710083)]
+......
+```
+
+* 方向分类器+识别
+
+```bash linenums="1"
+paddleocr --image_dir PaddleOCR/doc/imgs_words/ch/word_1.jpg --use_angle_cls true --det false
+```
+
+结果是一个list，每个item只包含识别结果和识别置信度
+
+```bash linenums="1"
+['韩国小馆', 0.994467]
+```
+
+* 单独执行检测
+
+```bash linenums="1"
+paddleocr --image_dir PaddleOCR/doc/imgs/11.jpg --rec false
+```
+
+结果是一个list，每个item只包含文本框
+
+```bash linenums="1"
+[[27.0, 459.0], [136.0, 459.0], [136.0, 479.0], [27.0, 479.0]]
+[[28.0, 429.0], [372.0, 429.0], [372.0, 445.0], [28.0, 445.0]]
+......
+```
+
+* 单独执行识别
+
+```bash linenums="1"
+paddleocr --image_dir PaddleOCR/doc/imgs_words/ch/word_1.jpg --det false
+```
+
+结果是一个list，每个item只包含识别结果和识别置信度
+
+```bash linenums="1"
+['韩国小馆', 0.994467]
+```
+
+* 单独执行方向分类器
+
+```bash linenums="1"
+paddleocr --image_dir PaddleOCR/doc/imgs_words/ch/word_1.jpg --use_angle_cls true --det false --rec false
+```
+
+结果是一个list，每个item只包含分类结果和分类置信度
+
+```bash linenums="1"
+['0', 0.9999924]
+```
+
+## 3 自定义模型
+
+当内置模型无法满足需求时，需要使用到自己训练的模型。 首先，参照[模型导出](../model_train/detection.md#4)将检测、分类和识别模型转换为inference模型，然后按照如下方式使用
+
+### 3.1 代码使用
+
+```python linenums="1"
+from paddleocr import PaddleOCR, draw_ocr
+
+# 模型路径下必须含有model和params文件
+ocr = PaddleOCR(det_model_dir='{your_det_model_dir}', rec_model_dir='{your_rec_model_dir}',
+                rec_char_dict_path='{your_rec_char_dict_path}', cls_model_dir='{your_cls_model_dir}',
+                use_angle_cls=True)
+img_path = 'PaddleOCR/doc/imgs/11.jpg'
+result = ocr.ocr(img_path, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+
+# 显示结果
+from PIL import Image
+result = result[0]
+image = Image.open(img_path).convert('RGB')
+boxes = [line[0] for line in result]
+txts = [line[1][0] for line in result]
+scores = [line[1][1] for line in result]
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+### 3.2 通过命令行使用
+
+```bash linenums="1"
+paddleocr --image_dir PaddleOCR/doc/imgs/11.jpg --det_model_dir {your_det_model_dir} --rec_model_dir {your_rec_model_dir} --rec_char_dict_path {your_rec_char_dict_path} --cls_model_dir {your_cls_model_dir} --use_angle_cls true
+```
+
+## 4 使用网络图片或者numpy数组作为输入
+
+### 4.1 网络图片
+
+* 代码使用
+
+```python linenums="1"
+from paddleocr import PaddleOCR, draw_ocr, download_with_progressbar
+
+# Paddleocr目前支持中英文、英文、法语、德语、韩语、日语，可以通过修改lang参数进行切换
+# 参数依次为`ch`, `en`, `french`, `german`, `korean`, `japan`。
+ocr = PaddleOCR(use_angle_cls=True, lang="ch")  # need to run only once to download and load model into memory
+img_path = 'http://n.sinaimg.cn/ent/transform/w630h933/20171222/o111-fypvuqf1838418.jpg'
+result = ocr.ocr(img_path, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+
+# 显示结果
+from PIL import Image
+result = result[0]
+download_with_progressbar(img_path, 'tmp.jpg')
+image = Image.open('tmp.jpg').convert('RGB')
+boxes = [line[0] for line in result]
+txts = [line[1][0] for line in result]
+scores = [line[1][1] for line in result]
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+* 命令行模式
+
+```bash linenums="1"
+paddleocr --image_dir http://n.sinaimg.cn/ent/transform/w630h933/20171222/o111-fypvuqf1838418.jpg --use_angle_cls=true
+```
+
+### 4.2 numpy数组
+
+仅通过代码使用时支持numpy数组作为输入
+
+```python linenums="1"
+import cv2
+from paddleocr import PaddleOCR, draw_ocr
+
+# Paddleocr目前支持中英文、英文、法语、德语、韩语、日语，可以通过修改lang参数进行切换
+# 参数依次为`ch`, `en`, `french`, `german`, `korean`, `japan`。
+ocr = PaddleOCR(use_angle_cls=True, lang="ch")  # need to run only once to download and load model into memory
+img_path = 'PaddleOCR/doc/imgs/11.jpg'
+img = cv2.imread(img_path)
+# img = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY), 如果你自己训练的模型支持灰度图，可以将这句话的注释取消
+result = ocr.ocr(img, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+
+# 显示结果
+from PIL import Image
+result = result[0]
+image = Image.open(img_path).convert('RGB')
+boxes = [line[0] for line in result]
+txts = [line[1][0] for line in result]
+scores = [line[1][1] for line in result]
+im_show = draw_ocr(image, boxes, txts, scores, font_path='/path/to/PaddleOCR/doc/fonts/simfang.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+## 5 PDF文件作为输入
+
+* 命令行模式
+
+可以通过指定参数`page_num`来控制推理前面几页，默认为0，表示推理所有页。
+
+```bash linenums="1"
+paddleocr --image_dir ./xxx.pdf --use_angle_cls true --use_gpu false --page_num 2
+```
+
+* 代码使用
+
+```python linenums="1"
+from paddleocr import PaddleOCR, draw_ocr
+
+# Paddleocr目前支持的多语言语种可以通过修改lang参数进行切换
+# 例如`ch`, `en`, `fr`, `german`, `korean`, `japan`
+ocr = PaddleOCR(use_angle_cls=True, lang="ch"， page_num=2)  # need to run only once to download and load model into memory
+img_path = './xxx.pdf'
+result = ocr.ocr(img_path, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+
+# 显示结果
+import fitz
+from PIL import Image
+import cv2
+import numpy as np
+imgs = []
+with fitz.open(img_path) as pdf:
+    for pg in range(0, pdf.pageCount):
+        page = pdf[pg]
+        mat = fitz.Matrix(2, 2)
+        pm = page.getPixmap(matrix=mat, alpha=False)
+        # if width or height > 2000 pixels, don't enlarge the image
+        if pm.width > 2000 or pm.height > 2000:
+            pm = page.getPixmap(matrix=fitz.Matrix(1, 1), alpha=False)
+
+        img = Image.frombytes("RGB", [pm.width, pm.height], pm.samples)
+        img = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
+        imgs.append(img)
+for idx in range(len(result)):
+    res = result[idx]
+    image = imgs[idx]
+    boxes = [line[0] for line in res]
+    txts = [line[1][0] for line in res]
+    scores = [line[1][1] for line in res]
+    im_show = draw_ocr(image, boxes, txts, scores, font_path='doc/fonts/simfang.ttf')
+    im_show = Image.fromarray(im_show)
+    im_show.save('result_page_{}.jpg'.format(idx))
+```
+
+## 6 参数说明
+
+| 字段   | 说明| 默认值 |
+|-------------------------|----------|-------------------------|
+| use_gpu| 是否使用GPU          | TRUE   |
+| gpu_mem| 初始化占用的GPU内存大小               | 8000M  |
+| image_dir               | 通过命令行调用时执行预测的图片或文件夹路径   |
+| page_num               | 当输入类型为pdf文件时有效，指定预测前面page_num页，默认预测所有页    |        0|
+| det_algorithm           | 使用的检测算法类型   | DB     |
+| det_model_dir          |  检测模型所在文件夹。传参方式有两种，1. None: 自动下载内置模型到 `~/.paddleocr/det`；2.自己转换好的inference模型路径，模型路径下必须包含model和params文件 |   None        |
+| det_max_side_len        | 检测算法前向时图片长边的最大尺寸，当长边超出这个值时会将长边resize到这个大小，短边等比例缩放   | 960    |
+| det_db_thresh           | DB模型输出预测图的二值化阈值          | 0.3    |
+| det_db_box_thresh       | DB模型输出框的阈值，低于此值的预测框会被丢弃   | 0.5    |
+| det_db_unclip_ratio     | DB模型输出框扩大的比例                | 2      |
+| det_db_score_mode |  计算检测框score的方式，有'fast'和'slow'，如果要检测的文字有弯曲，建议用'slow'，'slow'模式计算的box的score偏大，box不容易被过滤掉  | 'fast' |
+| det_east_score_thresh   | EAST模型输出预测图的二值化阈值        | 0.8    |
+| det_east_cover_thresh   | EAST模型输出框的阈值，低于此值的预测框会被丢弃 | 0.1    |
+| det_east_nms_thresh     | EAST模型输出框NMS的阈值               | 0.2    |
+| rec_algorithm           | 使用的识别算法类型   | CRNN   |
+| rec_model_dir          | 识别模型所在文件夹。传参方式有两种，1. None: 自动下载内置模型到 `~/.paddleocr/rec`；2.自己转换好的inference模型路径，模型路径下必须包含model和params文件 | None |
+| rec_image_shape         | 识别算法的输入图片尺寸              | "3,32,320"              |
+| rec_batch_num           | 进行识别时，同时前向的图片数          | 30     |
+| max_text_length         | 识别算法能识别的最大文字长度          | 25     |
+| rec_char_dict_path      | 识别模型字典路径，当rec_model_dir使用方式2传参时需要修改为自己的字典路径 | ./ppocr/utils/ppocr_keys_v1.txt       |
+| use_space_char          | 是否识别空格         | TRUE   |
+| drop_score          | 对输出按照分数(来自于识别模型)进行过滤，低于此分数的不返回         | 0.5   |
+| use_angle_cls          | 是否加载分类模型         | FALSE   |
+| cls_model_dir          | 分类模型所在文件夹。传参方式有两种，1. None: 自动下载内置模型到 `~/.paddleocr/cls`；2.自己转换好的inference模型路径，模型路径下必须包含model和params文件     | None   |
+| cls_image_shape          | 分类算法的输入图片尺寸                | "3, 48, 192"   |
+| label_list          | 分类算法的标签列表                | ['0', '180'] |
+| cls_batch_num          | 进行分类时，同时前向的图片数               |30|
+| enable_mkldnn           | 是否启用mkldnn       | FALSE  |
+| use_zero_copy_run           | 是否通过zero_copy_run的方式进行前向| FALSE  |
+| lang    | 模型语言类型,目前支持 目前支持中英文(ch)、英文(en)、法语(french)、德语(german)、韩语(korean)、日语(japan)                | ch   |
+| det    | 前向时使用启动检测   | TRUE   |
+| rec    | 前向时是否启动识别   | TRUE   |
+| cls    | 前向时是否启动分类 (命令行模式下使用use_angle_cls控制前向是否启动分类)| FALSE   |
+| show_log    | 是否打印logger信息| FALSE   |
+| type    | 执行ocr或者表格结构化, 值可选['ocr','structure']              | ocr   |
+| ocr_version    | OCR模型版本，可选PP-OCRv3, PP-OCRv2, PP-OCR。PP-OCRv3 支持中、英文的检测、识别、多语种识别，方向分类器等模型；PP-OCRv2 目前仅支持中文的检测和识别模型；PP-OCR支持中文的检测，识别，多语种识别，方向分类器等模型          | PP-OCRv3  |
diff --git a/docs/version2.x/ppocr/environment.en.md b/docs/version2.x/ppocr/environment.en.md
new file mode 100644
index 0000000..27abedb
--- /dev/null
+++ b/docs/version2.x/ppocr/environment.en.md
@@ -0,0 +1,313 @@
+---
+comments: true
+---
+
+# Environment Preparation
+
+Windows and Mac users are recommended to use Anaconda to build a Python environment, and Linux users are recommended to use docker to build a Python environment.
+
+Recommended working environment:
+
+- PaddlePaddle >= 2.1.2
+- Python 3
+- CUDA 10.1 / CUDA 10.2
+- cuDNN 7.6
+
+> If you already have a Python environment installed, you can skip to [PaddleOCR Quick Start](../ppocr/quick_start.en.md).
+
+## 1. Python Environment Setup
+
+### 1.1 Windows
+
+#### 1.1.1 Install Anaconda
+
+- Note: To use PaddlePaddle you need to install python environment first, here we choose python integrated environment Anaconda toolkit
+
+  - Anaconda is a common python package manager
+  - After installing Anaconda, you can install the python environment, as well as numpy and other required toolkit environment.
+
+- Anaconda download.
+
+  - Address: <https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/?C=M&O=D>
+  - Most Win10 computers are 64-bit operating systems, choose x86_64 version; if the computer is a 32-bit operating system, choose x86.exe
+
+    ![anaconda download](./images/Anaconda_download.png)
+
+  - After the download is complete, double-click the installer to enter the graphical interface
+
+  - The default installation location is C drive, it is recommended to change the installation location to D drive.
+
+    ![](./images/anaconda_install_folder.png)
+
+  - Check Conda to add environment variables and ignore the warning that
+    ![](./images/anaconda_install_env.png)
+
+#### 1.1.2 Opening the terminal and creating the Conda environment
+
+- Open Anaconda Prompt terminal: bottom left Windows Start Menu -> Anaconda3 -> Anaconda Prompt start console
+
+  ![anaconda download](./images/anaconda_prompt.png)
+
+- Create a new Conda environment
+
+  ```bash linenums="1"
+  # Enter the following command at the command line to create an environment named paddle_env
+  # Here to speed up the download, use the Tsinghua source
+  conda create --name paddle_env python=3.8 --channel https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/ # This is a one line command
+  ```
+
+  This command will create an executable environment named paddle_env with python version 3.8, which will take a while depending on the network status
+
+  The command line will then output a prompt, type y and enter to continue the installation
+
+  ![conda create](./images/conda_new_env.png)
+
+- To activate the Conda environment you just created, enter the following command at the command line.
+
+  ```bash linenums="1"
+  # Activate the paddle_env environment
+  conda activate paddle_env
+  # View the current location of python
+  where python
+  ```
+
+  ![create environment](./images/conda_list_env.png)
+
+The above anaconda environment and python environment are installed
+
+### 1.2 Mac
+
+#### 1.2.1 Installing Anaconda
+
+- Note: To use PaddlePaddle you need to install the python environment first, here we choose the python integrated environment Anaconda toolkit
+
+  - Anaconda is a common python package manager
+  - After installing Anaconda, you can install the python environment, as well as numpy and other required toolkit environment
+
+- Anaconda download:.
+
+  - Address: <https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/?C=M&O=D>
+
+  ![anaconda download](./images/anaconda_start.png)
+
+  - Select `Anaconda3-2021.05-MacOSX-x86_64.pkg` at the bottom to download
+
+- After downloading, double click on the .pkg file to enter the graphical interface
+
+  - Just follow the default settings, it will take a while to install
+
+- It is recommended to install a code editor such as VSCode or PyCharm
+
+#### 1.2.2 Open a terminal and create a Conda environment
+
+- Open the terminal
+
+  - Press command and spacebar at the same time, type "terminal" in the focus search, double click to enter terminal
+
+- **Add Conda to the environment variables**
+
+  - Environment variables are added so that the system can recognize the Conda command
+
+  - Open `~/.bash_profile` in the terminal by typing the following command.
+
+    ```bash linenums="1"
+    vim ~/.bash_profile
+    ```
+
+  - Add Conda as an environment variable in `~/.bash_profile`.
+
+    ```bash linenums="1"
+    # Press i first to enter edit mode
+    # In the first line type.
+    export PATH="~/opt/anaconda3/bin:$PATH"
+    # If you customized the installation location during installation, change ~/opt/anaconda3/bin to the bin folder in the customized installation directory
+
+    # The modified ~/.bash_profile file should look like this (where xxx is the username)
+    export PATH="~/opt/anaconda3/bin:$PATH"
+    # >>> conda initialize >>>
+    # !!! Contents within this block are managed by 'conda init' !!!
+    __conda_setup="$('/Users/xxx/opt/anaconda3/bin/conda' 'shell.bash' 'hook' 2> /dev/null)"
+    if [ $? -eq 0 ]; then
+            eval "$__conda_setup"
+    else
+            if [ -f "/Users/xxx/opt/anaconda3/etc/profile.d/conda.sh" ]; then
+                    . "/Users/xxx/opt/anaconda3/etc/profile.d/conda.sh"
+            else
+                    export PATH="/Users/xxx/opt/anaconda3/bin:$PATH"
+            fi
+    fi
+    unset __conda_setup
+    # <<< conda initialize <<<
+    ```
+
+    - When you are done, press `esc` to exit edit mode, then type `:wq!` and enter to save and exit
+
+  - Verify that the Conda command is recognized.
+
+    - Enter `source ~/.bash_profile` in the terminal to update the environment variables
+    - Enter `conda info --envs` in the terminal again, if it shows that there is a base environment, then Conda has been added to the environment variables
+
+- Create a new Conda environment
+
+  ```bash linenums="1"
+  # Enter the following command at the command line to create an environment called paddle_env
+  # Here to speed up the download, use Tsinghua source
+  conda create --name paddle_env python=3.8 --channel https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
+  ```
+
+  - This command will create an executable environment named paddle_env with python version 3.8, which will take a while depending on the network status
+
+  - The command line will then output a prompt, type y and enter to continue the installation
+
+    ![conda_create](./images/conda_create.png)
+
+- To activate the Conda environment you just created, enter the following command at the command line.
+
+  ```bash linenums="1"
+  # Activate the paddle_env environment
+  conda activate paddle_env
+  # View the current location of python
+  where python
+  ```
+
+  ![conda_actviate](./images/conda_activate.png)
+
+The above anaconda environment and python environment are installed
+
+### 1.3 Linux
+
+Linux users can choose to run either Anaconda or Docker. If you are familiar with Docker and need to train the PaddleOCR model, it is recommended to use the Docker environment, where the development process of PaddleOCR is run. If you are not familiar with Docker, you can also use Anaconda to run the project.
+
+#### 1.3.1 Anaconda environment configuration
+
+- Note: To use PaddlePaddle you need to install the python environment first, here we choose the python integrated environment Anaconda toolkit
+
+  - Anaconda is a common python package manager
+  - After installing Anaconda, you can install the python environment, as well as numpy and other required toolkit environment
+
+- **Download Anaconda**.
+
+  - Download at: <https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/?C=M&O=D>
+
+  ![img](./images/anaconda_download-20240704081644684.png)
+
+  - Select the appropriate version for your operating system
+    - Type `uname -m` in the terminal to check the command set used by your system
+
+  - Download method 1: Download locally, then transfer the installation package to the Linux server
+
+  - Download method 2: Directly use Linux command line to download
+
+    ```bash linenums="1"
+    # First install wget
+    sudo apt-get install wget # Ubuntu
+    sudo yum install wget # CentOS
+    ```
+
+    ```bash linenums="1"
+    # Then use wget to download from Tsinghua source
+    # If you want to download Anaconda3-2021.05-Linux-x86_64.sh, the download command is as follows
+    wget https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/Anaconda3-2021.05-Linux-x86_64.sh
+    # If you want to download another version, you need to change the file name after the last 1 / to the version you want to download
+    ```
+
+- To install Anaconda.
+
+  - Type `sh Anaconda3-2021.05-Linux-x86_64.sh` at the command line
+    - If you downloaded a different version, replace the file name of the command with the name of the file you downloaded
+  - Just follow the installation instructions
+    - You can exit by typing q when viewing the license
+
+- **Add conda to the environment variables**
+
+  - If you have already added conda to the environment variable path during the installation, you can skip this step
+
+  - Open `~/.bashrc` in a terminal.
+
+    ```bash linenums="1"
+    # Enter the following command in the terminal.
+    vim ~/.bashrc
+    ```
+
+  - Add conda as an environment variable in `~/.bashrc`.
+
+    ```bash linenums="1"
+    # Press i first to enter edit mode # In the first line enter.
+    export PATH="~/anaconda3/bin:$PATH"
+    # If you customized the installation location during installation, change ~/anaconda3/bin to the bin folder in the customized installation directory
+    ```
+
+    ```bash linenums="1"
+    # The modified ~/.bash_profile file should look like this (where xxx is the username)
+    export PATH="~/opt/anaconda3/bin:$PATH"
+    # >>> conda initialize >>>
+    # !!! Contents within this block are managed by 'conda init' !!!
+    __conda_setup="$('/Users/xxx/opt/anaconda3/bin/conda' 'shell.bash' 'hook' 2> /dev/null)"
+    if [ $? -eq 0 ]; then
+        eval "$__conda_setup"
+    else
+        if [ -f "/Users/xxx/opt/anaconda3/etc/profile.d/conda.sh" ]; then
+            . "/Users/xxx/opt/anaconda3/etc/profile.d/conda.sh"
+        else
+            export PATH="/Users/xxx/opt/anaconda3/bin:$PATH"
+        fi
+    fi
+    unset __conda_setup
+    # <<< conda initialize <<<
+    ```
+
+    - When you are done, press `esc` to exit edit mode, then type `:wq!` and enter to save and exit
+
+  - Verify that the Conda command is recognized.
+
+    - Enter `source ~/.bash_profile` in the terminal to update the environment variables
+    - Enter `conda info --envs` in the terminal again, if it shows that there is a base environment, then Conda has been added to the environment variables
+
+- Create a new Conda environment
+
+  ```bash linenums="1"
+  # Enter the following command at the command line to create an environment called paddle_env
+  # Here to speed up the download, use Tsinghua source
+  conda create --name paddle_env python=3.8 --channel https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
+  ```
+
+  - This command will create an executable environment named paddle_env with python version 3.8, which will take a while depending on the network status
+
+  - The command line will then output a prompt, type y and enter to continue the installation
+
+    ![conda_create](./images/conda_create-20240704081656378.png)
+
+- To activate the Conda environment you just created, enter the following command at the command line.
+
+  ```bash linenums="1"
+  # Activate the paddle_env environment
+  conda activate paddle_env
+  ```
+
+The above anaconda environment and python environment are installed
+
+#### 1.3.2 Docker environment preparation
+
+**The first time you use this docker image, it will be downloaded automatically. Please be patient.**
+
+```bash linenums="1"
+# Switch to the working directory
+cd /home/Projects
+# You need to create a docker container for the first run, and do not need to run the current command when you run it again
+# Create a docker container named ppocr and map the current directory to the /paddle directory of the container
+
+# If using CPU, use docker instead of nvidia-docker to create docker
+sudo docker run --name ppocr -v $PWD:/paddle --network=host -it  registry.baidubce.com/paddlepaddle/paddle:2.1.3-gpu-cuda10.2-cudnn7  /bin/bash
+
+# If using GPU, use nvidia-docker to create docker
+# docker image registry.baidubce.com/paddlepaddle/paddle:2.1.3-gpu-cuda11.2-cudnn8 is recommended for CUDA11.2 + CUDNN8.
+sudo nvidia-docker run --name ppocr -v $PWD:/paddle --shm-size=64G --network=host -it registry.baidubce.com/paddlepaddle/paddle:2.1.3-gpu-cuda10.2-cudnn7 /bin/bash
+
+```
+
+You can also visit [DockerHub](https://hub.docker.com/r/paddlepaddle/paddle/tags/) to get the image that fits your machine.
+
+```bash linenums="1"
+# ctrl+P+Q to exit docker, to re-enter docker using the following command:
+sudo docker container exec -it ppocr /bin/bash
+```
diff --git a/docs/version2.x/ppocr/environment.md b/docs/version2.x/ppocr/environment.md
new file mode 100644
index 0000000..3ed45b0
--- /dev/null
+++ b/docs/version2.x/ppocr/environment.md
@@ -0,0 +1,302 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# 运行环境准备
+
+Windows和Mac用户推荐使用Anaconda搭建Python环境，Linux用户建议使用docker搭建Python环境。
+
+推荐环境：
+
+- PaddlePaddle >= 2.1.2
+- Python 3
+- CUDA10.1 / CUDA10.2
+- CUDNN 7.6
+
+> 如果您已经安装Python环境，可以直接参考[PaddleOCR快速开始](./quick_start.md)
+
+## 1. Python环境搭建
+
+### 1.1 Windows
+
+#### 1.1.1 安装Anaconda
+
+- 说明：使用paddlepaddle需要先安装python环境，这里我们选择python集成环境Anaconda工具包
+  - Anaconda是1个常用的python包管理程序
+  - 安装完Anaconda后，可以安装python环境，以及numpy等所需的工具包环境。
+
+- Anaconda下载：
+  - 地址：<https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/?C=M&O=D>
+  - 大部分win10电脑均为64位操作系统，选择x86_64版本；若电脑为32位操作系统，则选择x86.exe
+
+    ![anaconda download](./images/Anaconda_download.png)
+  - 下载完成后，双击安装程序进入图形界面
+  - 默认安装位置为C盘，建议将安装位置更改到D盘：
+
+    <img src="./images/anaconda_install_folder.png" alt="install config" style="zoom:67%;" />
+
+  - 勾选conda加入环境变量，忽略警告：
+
+    <img src="./images/anaconda_install_env.png" alt="add conda to path" style="zoom: 67%;" />
+
+#### 1.1.2 打开终端并创建conda环境
+
+- 打开Anaconda Prompt终端：左下角Windows Start Menu -> Anaconda3 -> Anaconda Prompt启动控制台
+
+  ![anaconda download](./images/anaconda_prompt.png)
+
+- 创建新的conda环境
+
+  ```bash linenums="1"
+  # 在命令行输入以下命令，创建名为paddle_env的环境
+  # 此处为加速下载，使用清华源
+  conda create --name paddle_env python=3.8 --channel https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/  # 这是一行命令
+  ```
+
+  该命令会创建1个名为paddle_env、python版本为3.8的可执行环境，根据网络状态，需要花费一段时间
+
+  之后命令行中会输出提示信息，输入y并回车继续安装
+
+  ![conda create](./images/conda_new_env.png)
+
+- 激活刚创建的conda环境，在命令行中输入以下命令：
+
+  ```bash linenums="1"
+  # 激活paddle_env环境
+  conda activate paddle_env
+  # 查看当前python的位置
+  where python
+  ```
+
+    ![create environment](./images/conda_list_env.png)
+
+以上anaconda环境和python环境安装完毕
+
+### 1.2 Mac
+
+#### 1.2.1 安装Anaconda
+
+- 说明：使用paddlepaddle需要先安装python环境，这里我们选择python集成环境Anaconda工具包
+  - Anaconda是1个常用的python包管理程序
+  - 安装完Anaconda后，可以安装python环境，以及numpy等所需的工具包环境
+- Anaconda下载：
+  - 地址：<https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/?C=M&O=D>
+
+  ![anaconda download](./images/anaconda_start.png)
+
+  - 选择最下方的`Anaconda3-2021.05-MacOSX-x86_64.pkg`下载
+- 下载完成后，双击.pkg文件进入图形界面
+  - 按默认设置即可，安装需要花费一段时间
+- 建议安装vscode或pycharm等代码编辑器
+
+#### 1.2.2 打开终端并创建conda环境
+
+- 打开终端
+  - 同时按下command键和空格键，在聚焦搜索中输入"终端"，双击进入终端
+
+- **将conda加入环境变量**
+
+  - 加入环境变量是为了让系统能识别conda命令
+
+  - 输入以下命令，在终端中打开`~/.bash_profile`：
+
+     ```bash linenums="1"
+     vim ~/.bash_profile
+     ```
+
+  - 在`~/.bash_profile`中将conda添加为环境变量：
+
+     ```bash linenums="1"
+     # 先按i进入编辑模式
+     # 在第一行输入：
+     export PATH="~/opt/anaconda3/bin:$PATH"
+     # 若安装时自定义了安装位置，则将~/opt/anaconda3/bin改为自定义的安装目录下的bin文件夹
+     ```
+
+    ```bash linenums="1"
+    # 修改后的~/.bash_profile文件应如下（其中xxx为用户名）：
+    export PATH="~/opt/anaconda3/bin:$PATH"
+    # >>> conda initialize >>>
+    # !! Contents within this block are managed by 'conda init' !!
+    __conda_setup="$('/Users/xxx/opt/anaconda3/bin/conda' 'shell.bash' 'hook' 2> /dev/null)"
+    if [ $? -eq 0 ]; then
+            eval "$__conda_setup"
+    else
+            if [ -f "/Users/xxx/opt/anaconda3/etc/profile.d/conda.sh" ]; then
+                    . "/Users/xxx/opt/anaconda3/etc/profile.d/conda.sh"
+            else
+                    export PATH="/Users/xxx/opt/anaconda3/bin:$PATH"
+            fi
+    fi
+    unset __conda_setup
+    # <<< conda initialize <<<
+    ```
+
+    - 修改完成后，先按`esc`键退出编辑模式，再输入`:wq!`并回车，以保存退出
+
+  - 验证是否能识别conda命令：
+
+    - 在终端中输入`source ~/.bash_profile`以更新环境变量
+    - 再在终端输入`conda info --envs`，若能显示当前有base环境，则conda已加入环境变量
+
+- 创建新的conda环境
+
+  ```bash linenums="1"
+  # 在命令行输入以下命令，创建名为paddle_env的环境
+  # 此处为加速下载，使用清华源
+  conda create --name paddle_env python=3.8 --channel https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
+  ```
+
+  - 该命令会创建1个名为paddle_env、python版本为3.8的可执行环境，根据网络状态，需要花费一段时间
+
+  - 之后命令行中会输出提示信息，输入y并回车继续安装
+
+    ![conda_create](./images/conda_create.png)
+
+- 激活刚创建的conda环境，在命令行中输入以下命令：
+
+  ```bash linenums="1"
+  # 激活paddle_env环境
+  conda activate paddle_env
+  # 查看当前python的位置
+  where python
+  ```
+
+  ![conda_actviate](./images/conda_activate.png)
+
+以上anaconda环境和python环境安装完毕
+
+### 1.3 Linux
+
+Linux用户可选择Anaconda或Docker两种方式运行。如果你熟悉Docker且需要训练PaddleOCR模型，推荐使用Docker环境，PaddleOCR的开发流程均在Docker环境下运行。如果你不熟悉Docker，也可以使用Anaconda来运行项目。
+
+#### 1.3.1 Anaconda环境配置
+
+- 说明：使用paddlepaddle需要先安装python环境，这里我们选择python集成环境Anaconda工具包
+  - Anaconda是1个常用的python包管理程序
+  - 安装完Anaconda后，可以安装python环境，以及numpy等所需的工具包环境
+
+- **下载Anaconda**：
+
+  - 下载地址：<https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/?C=M&O=D>
+
+    ![img](./images/anaconda_download-20240704081644684.png)
+
+    - 选择适合您操作系统的版本
+    - 可在终端输入`uname -m`查询系统所用的指令集
+
+- 下载法1：本地下载，再将安装包传到linux服务器上
+
+- 下载法2：直接使用linux命令行下载
+
+  ```bash linenums="1"
+    # 首先安装wget
+  sudo apt-get install wget  # Ubuntu
+    sudo yum install wget  # CentOS
+  ```
+
+  ```bash linenums="1"
+    # 然后使用wget从清华源上下载
+  # 如要下载Anaconda3-2021.05-Linux-x86_64.sh，则下载命令如下：
+    wget https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/Anaconda3-2021.05-Linux-x86_64.sh
+
+  # 若您要下载其他版本，需要将最后1个/后的文件名改成您希望下载的版本
+  ```
+
+- 安装Anaconda：
+
+  - 在命令行输入`sh Anaconda3-2021.05-Linux-x86_64.sh`
+    - 若您下载的是其它版本，则将该命令的文件名替换为您下载的文件名
+  - 按照安装提示安装即可
+    - 查看许可时可输入q来退出
+
+- **将conda加入环境变量**
+
+  - 加入环境变量是为了让系统能识别conda命令，若您在安装时已将conda加入环境变量path，则可跳过本步
+
+  - 在终端中打开`~/.bashrc`：
+
+    ```bash linenums="1"
+    # 在终端中输入以下命令：
+    vim ~/.bashrc
+    ```
+
+  - 在`~/.bashrc`中将conda添加为环境变量：
+
+    ```bash linenums="1"
+    # 先按i进入编辑模式
+    # 在第一行输入：
+    export PATH="~/anaconda3/bin:$PATH"
+    # 若安装时自定义了安装位置，则将~/anaconda3/bin改为自定义的安装目录下的bin文件夹
+    ```
+
+    ```bash linenums="1"
+    # 修改后的~/.bash_profile文件应如下（其中xxx为用户名）：
+    export PATH="~/opt/anaconda3/bin:$PATH"
+    # >>> conda initialize >>>
+    # !! Contents within this block are managed by 'conda init' !!
+    __conda_setup="$('/Users/xxx/opt/anaconda3/bin/conda' 'shell.bash' 'hook' 2> /dev/null)"
+    if [ $? -eq 0 ]; then
+        eval "$__conda_setup"
+    else
+        if [ -f "/Users/xxx/opt/anaconda3/etc/profile.d/conda.sh" ]; then
+            . "/Users/xxx/opt/anaconda3/etc/profile.d/conda.sh"
+        else
+            export PATH="/Users/xxx/opt/anaconda3/bin:$PATH"
+        fi
+    fi
+    unset __conda_setup
+    # <<< conda initialize <<<
+    ```
+
+    - 修改完成后，先按`esc`键退出编辑模式，再输入`:wq!`并回车，以保存退出
+
+  - 验证是否能识别conda命令：
+
+    - 在终端中输入`source ~/.bash_profile`以更新环境变量
+    - 再在终端输入`conda info --envs`，若能显示当前有base环境，则conda已加入环境变量
+
+- 创建新的conda环境
+
+   ```bash linenums="1"
+   # 在命令行输入以下命令，创建名为paddle_env的环境
+   # 此处为加速下载，使用清华源
+   conda create --name paddle_env python=3.8 --channel https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
+   ```
+
+  - 该命令会创建1个名为paddle_env、python版本为3.8的可执行环境，根据网络状态，需要花费一段时间
+
+  - 之后命令行中会输出提示信息，输入y并回车继续安装
+
+    ![conda_create](./images/conda_create-20240704081656378.png)
+
+- 激活刚创建的conda环境，在命令行中输入以下命令：
+
+  ```bash linenums="1"
+  # 激活paddle_env环境
+  conda activate paddle_env
+  ```
+
+以上anaconda环境和python环境安装完毕
+
+#### 1.3.2 Docker环境配置
+
+**注意：第一次使用这个镜像，会自动下载该镜像，请耐心等待。您也可以访问[DockerHub](https://hub.docker.com/r/paddlepaddle/paddle/tags/)获取与您机器适配的镜像。**
+
+```bash linenums="1"
+# 切换到工作目录下
+cd /home/Projects
+# 首次运行需创建一个docker容器，再次运行时不需要运行当前命令
+# 创建一个名字为ppocr的docker容器，并将当前目录映射到容器的/paddle目录下
+
+#如果您希望在CPU环境下使用docker，使用docker而不是nvidia-docker创建docker
+sudo docker run --name ppocr -v $PWD:/paddle --network=host -it registry.baidubce.com/paddlepaddle/paddle:2.1.3-gpu-cuda10.2-cudnn7 /bin/bash
+
+#如果使用CUDA10，请运行以下命令创建容器，设置docker容器共享内存shm-size为64G，建议设置32G以上
+# 如果是CUDA11+CUDNN8，推荐使用镜像registry.baidubce.com/paddlepaddle/paddle:2.1.3-gpu-cuda11.2-cudnn8
+sudo nvidia-docker run --name ppocr -v $PWD:/paddle --shm-size=64G --network=host -it registry.baidubce.com/paddlepaddle/paddle:2.1.3-gpu-cuda10.2-cudnn7 /bin/bash
+
+# ctrl+P+Q可退出docker 容器，重新进入docker 容器使用如下命令
+sudo docker container exec -it ppocr /bin/bash
+```
diff --git a/docs/version2.x/ppocr/images/254.jpg b/docs/version2.x/ppocr/images/254.jpg
new file mode 100644
index 0000000..c871fb0
Binary files /dev/null and b/docs/version2.x/ppocr/images/254.jpg differ
diff --git a/docs/version2.x/ppocr/images/Anaconda_download.png b/docs/version2.x/ppocr/images/Anaconda_download.png
new file mode 100644
index 0000000..83a0341
Binary files /dev/null and b/docs/version2.x/ppocr/images/Anaconda_download.png differ
diff --git a/docs/version2.x/ppocr/images/PP-OCRv2/PP-OCRv2-pic001.jpg b/docs/version2.x/ppocr/images/PP-OCRv2/PP-OCRv2-pic001.jpg
new file mode 100644
index 0000000..45ffdb5
Binary files /dev/null and b/docs/version2.x/ppocr/images/PP-OCRv2/PP-OCRv2-pic001.jpg differ
diff --git a/docs/version2.x/ppocr/images/PP-OCRv2/PP-OCRv2-pic002.jpg b/docs/version2.x/ppocr/images/PP-OCRv2/PP-OCRv2-pic002.jpg
new file mode 100644
index 0000000..7ac153a
Binary files /dev/null and b/docs/version2.x/ppocr/images/PP-OCRv2/PP-OCRv2-pic002.jpg differ
diff --git a/docs/version2.x/ppocr/images/PP-OCRv2/PP-OCRv2-pic003.jpg b/docs/version2.x/ppocr/images/PP-OCRv2/PP-OCRv2-pic003.jpg
new file mode 100644
index 0000000..781aade
Binary files /dev/null and b/docs/version2.x/ppocr/images/PP-OCRv2/PP-OCRv2-pic003.jpg differ
diff --git a/docs/version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic001.jpg b/docs/version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic001.jpg
new file mode 100644
index 0000000..c35936c
Binary files /dev/null and b/docs/version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic001.jpg differ
diff --git a/docs/version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic002.jpg b/docs/version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic002.jpg
new file mode 100644
index 0000000..e5ad6a4
Binary files /dev/null and b/docs/version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic002.jpg differ
diff --git a/docs/version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic003.jpg b/docs/version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic003.jpg
new file mode 100644
index 0000000..dc02429
Binary files /dev/null and b/docs/version2.x/ppocr/images/PP-OCRv3/ch/PP-OCRv3-pic003.jpg differ
diff --git a/docs/version2.x/ppocr/images/PP-OCRv3/en/en_1.png b/docs/version2.x/ppocr/images/PP-OCRv3/en/en_1.png
new file mode 100644
index 0000000..3624561
Binary files /dev/null and b/docs/version2.x/ppocr/images/PP-OCRv3/en/en_1.png differ
diff --git a/docs/version2.x/ppocr/images/PP-OCRv3/en/en_2.png b/docs/version2.x/ppocr/images/PP-OCRv3/en/en_2.png
new file mode 100644
index 0000000..d2df855
Binary files /dev/null and b/docs/version2.x/ppocr/images/PP-OCRv3/en/en_2.png differ
diff --git a/docs/version2.x/ppocr/images/PP-OCRv3/en/en_3.png b/docs/version2.x/ppocr/images/PP-OCRv3/en/en_3.png
new file mode 100644
index 0000000..baf146c
Binary files /dev/null and b/docs/version2.x/ppocr/images/PP-OCRv3/en/en_3.png differ
diff --git a/docs/version2.x/ppocr/images/PP-OCRv3/en/en_4.png b/docs/version2.x/ppocr/images/PP-OCRv3/en/en_4.png
new file mode 100644
index 0000000..f0f19db
Binary files /dev/null and b/docs/version2.x/ppocr/images/PP-OCRv3/en/en_4.png differ
diff --git a/docs/version2.x/ppocr/images/PP-OCRv3/multi_lang/japan_2.jpg b/docs/version2.x/ppocr/images/PP-OCRv3/multi_lang/japan_2.jpg
new file mode 100644
index 0000000..076ced9
Binary files /dev/null and b/docs/version2.x/ppocr/images/PP-OCRv3/multi_lang/japan_2.jpg differ
diff --git a/docs/version2.x/ppocr/images/PP-OCRv3/multi_lang/korean_1.jpg b/docs/version2.x/ppocr/images/PP-OCRv3/multi_lang/korean_1.jpg
new file mode 100644
index 0000000..f93de40
Binary files /dev/null and b/docs/version2.x/ppocr/images/PP-OCRv3/multi_lang/korean_1.jpg differ
diff --git a/docs/version2.x/ppocr/images/anaconda_download-20240704081644684.png b/docs/version2.x/ppocr/images/anaconda_download-20240704081644684.png
new file mode 100644
index 0000000..6ab6db3
Binary files /dev/null and b/docs/version2.x/ppocr/images/anaconda_download-20240704081644684.png differ
diff --git a/docs/version2.x/ppocr/images/anaconda_install_env.png b/docs/version2.x/ppocr/images/anaconda_install_env.png
new file mode 100644
index 0000000..7a22542
Binary files /dev/null and b/docs/version2.x/ppocr/images/anaconda_install_env.png differ
diff --git a/docs/version2.x/ppocr/images/anaconda_install_folder.png b/docs/version2.x/ppocr/images/anaconda_install_folder.png
new file mode 100644
index 0000000..e9fac29
Binary files /dev/null and b/docs/version2.x/ppocr/images/anaconda_install_folder.png differ
diff --git a/docs/version2.x/ppocr/images/anaconda_prompt.png b/docs/version2.x/ppocr/images/anaconda_prompt.png
new file mode 100644
index 0000000..1087610
Binary files /dev/null and b/docs/version2.x/ppocr/images/anaconda_prompt.png differ
diff --git a/docs/version2.x/ppocr/images/anaconda_start.png b/docs/version2.x/ppocr/images/anaconda_start.png
new file mode 100644
index 0000000..a860f5e
Binary files /dev/null and b/docs/version2.x/ppocr/images/anaconda_start.png differ
diff --git a/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/00006737.jpg b/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/00006737.jpg
new file mode 100644
index 0000000..d7762d2
Binary files /dev/null and b/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/00006737.jpg differ
diff --git a/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/00009282.jpg b/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/00009282.jpg
new file mode 100644
index 0000000..0383d44
Binary files /dev/null and b/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/00009282.jpg differ
diff --git a/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/00015504.jpg b/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/00015504.jpg
new file mode 100644
index 0000000..9162cf1
Binary files /dev/null and b/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/00015504.jpg differ
diff --git a/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/00059985.jpg b/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/00059985.jpg
new file mode 100644
index 0000000..03fd197
Binary files /dev/null and b/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/00059985.jpg differ
diff --git a/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/00111002.jpg b/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/00111002.jpg
new file mode 100644
index 0000000..7dae24a
Binary files /dev/null and b/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/00111002.jpg differ
diff --git a/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/img_12.jpg b/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/img_12.jpg
new file mode 100644
index 0000000..11ac4ed
Binary files /dev/null and b/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/img_12.jpg differ
diff --git a/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/rotate_00052204.jpg b/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/rotate_00052204.jpg
new file mode 100644
index 0000000..643b850
Binary files /dev/null and b/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/rotate_00052204.jpg differ
diff --git a/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/test_add_91.jpg b/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/test_add_91.jpg
new file mode 100644
index 0000000..b5ded6e
Binary files /dev/null and b/docs/version2.x/ppocr/images/ch_ppocr_mobile_v2.0/test_add_91.jpg differ
diff --git a/docs/version2.x/ppocr/images/conda_activate.png b/docs/version2.x/ppocr/images/conda_activate.png
new file mode 100644
index 0000000..a2e6074
Binary files /dev/null and b/docs/version2.x/ppocr/images/conda_activate.png differ
diff --git a/docs/version2.x/ppocr/images/conda_create-20240704081656378.png b/docs/version2.x/ppocr/images/conda_create-20240704081656378.png
new file mode 100644
index 0000000..533f592
Binary files /dev/null and b/docs/version2.x/ppocr/images/conda_create-20240704081656378.png differ
diff --git a/docs/version2.x/ppocr/images/conda_create.png b/docs/version2.x/ppocr/images/conda_create.png
new file mode 100644
index 0000000..9ff10c2
Binary files /dev/null and b/docs/version2.x/ppocr/images/conda_create.png differ
diff --git a/docs/version2.x/ppocr/images/conda_list_env.png b/docs/version2.x/ppocr/images/conda_list_env.png
new file mode 100644
index 0000000..5ffa003
Binary files /dev/null and b/docs/version2.x/ppocr/images/conda_list_env.png differ
diff --git a/docs/version2.x/ppocr/images/conda_new_env.png b/docs/version2.x/ppocr/images/conda_new_env.png
new file mode 100644
index 0000000..eed667e
Binary files /dev/null and b/docs/version2.x/ppocr/images/conda_new_env.png differ
diff --git a/docs/version2.x/ppocr/images/model_prod_flow_ch.png b/docs/version2.x/ppocr/images/model_prod_flow_ch.png
new file mode 100644
index 0000000..4906b27
Binary files /dev/null and b/docs/version2.x/ppocr/images/model_prod_flow_ch.png differ
diff --git a/docs/version2.x/ppocr/images/model_prod_flow_en.png b/docs/version2.x/ppocr/images/model_prod_flow_en.png
new file mode 100644
index 0000000..eb5cb9c
Binary files /dev/null and b/docs/version2.x/ppocr/images/model_prod_flow_en.png differ
diff --git a/docs/version2.x/ppocr/images/multi_lang/arabic_0.jpg b/docs/version2.x/ppocr/images/multi_lang/arabic_0.jpg
new file mode 100644
index 0000000..9941b90
Binary files /dev/null and b/docs/version2.x/ppocr/images/multi_lang/arabic_0.jpg differ
diff --git a/docs/version2.x/ppocr/images/multi_lang/en_1.jpg b/docs/version2.x/ppocr/images/multi_lang/en_1.jpg
new file mode 100644
index 0000000..2dc84d3
Binary files /dev/null and b/docs/version2.x/ppocr/images/multi_lang/en_1.jpg differ
diff --git a/docs/version2.x/ppocr/images/multi_lang/en_2.jpg b/docs/version2.x/ppocr/images/multi_lang/en_2.jpg
new file mode 100644
index 0000000..455ec98
Binary files /dev/null and b/docs/version2.x/ppocr/images/multi_lang/en_2.jpg differ
diff --git a/docs/version2.x/ppocr/images/multi_lang/en_3.jpg b/docs/version2.x/ppocr/images/multi_lang/en_3.jpg
new file mode 100644
index 0000000..36eb063
Binary files /dev/null and b/docs/version2.x/ppocr/images/multi_lang/en_3.jpg differ
diff --git a/docs/version2.x/ppocr/images/multi_lang/french_0.jpg b/docs/version2.x/ppocr/images/multi_lang/french_0.jpg
new file mode 100644
index 0000000..3c2abe6
Binary files /dev/null and b/docs/version2.x/ppocr/images/multi_lang/french_0.jpg differ
diff --git a/docs/version2.x/ppocr/images/multi_lang/img_01.jpg b/docs/version2.x/ppocr/images/multi_lang/img_01.jpg
new file mode 100644
index 0000000..ee6ca69
Binary files /dev/null and b/docs/version2.x/ppocr/images/multi_lang/img_01.jpg differ
diff --git a/docs/version2.x/ppocr/images/multi_lang/img_02.jpg b/docs/version2.x/ppocr/images/multi_lang/img_02.jpg
new file mode 100644
index 0000000..3e139c7
Binary files /dev/null and b/docs/version2.x/ppocr/images/multi_lang/img_02.jpg differ
diff --git a/docs/version2.x/ppocr/images/multi_lang/img_12.jpg b/docs/version2.x/ppocr/images/multi_lang/img_12.jpg
new file mode 100644
index 0000000..822d562
Binary files /dev/null and b/docs/version2.x/ppocr/images/multi_lang/img_12.jpg differ
diff --git a/docs/version2.x/ppocr/images/multi_lang/japan_2.jpg b/docs/version2.x/ppocr/images/multi_lang/japan_2.jpg
new file mode 100644
index 0000000..7038ba2
Binary files /dev/null and b/docs/version2.x/ppocr/images/multi_lang/japan_2.jpg differ
diff --git a/docs/version2.x/ppocr/images/multi_lang/korean_0.jpg b/docs/version2.x/ppocr/images/multi_lang/korean_0.jpg
new file mode 100644
index 0000000..3fe6305
Binary files /dev/null and b/docs/version2.x/ppocr/images/multi_lang/korean_0.jpg differ
diff --git a/docs/version2.x/ppocr/images/ppocrv2_framework.jpg b/docs/version2.x/ppocr/images/ppocrv2_framework.jpg
new file mode 100644
index 0000000..e5f1a2e
Binary files /dev/null and b/docs/version2.x/ppocr/images/ppocrv2_framework.jpg differ
diff --git a/docs/version2.x/ppocr/images/ppocrv3_framework.png b/docs/version2.x/ppocr/images/ppocrv3_framework.png
new file mode 100644
index 0000000..8696e67
Binary files /dev/null and b/docs/version2.x/ppocr/images/ppocrv3_framework.png differ
diff --git a/docs/version2.x/ppocr/installation.en.md b/docs/version2.x/ppocr/installation.en.md
new file mode 100644
index 0000000..244e88d
--- /dev/null
+++ b/docs/version2.x/ppocr/installation.en.md
@@ -0,0 +1,85 @@
+---
+comments: true
+---
+
+## Quick Installation
+
+After testing, PaddleOCR can run on glibc 2.23. You can also test other glibc versions or install glibc 2.23 for the best compatibility.
+
+PaddleOCR working environment:
+
+- PaddlePaddle > 2.0.0
+- Python 3
+- glibc 2.23
+
+It is recommended to use the docker provided by us to run PaddleOCR. Please refer to the docker tutorial [link](https://www.runoob.com/docker/docker-tutorial.html/).
+
+*If you want to directly run the prediction code on Mac or Windows, you can start from step 2.*
+
+### 1. (Recommended) Prepare a docker environment
+
+For the first time you use this docker image, it will be downloaded automatically. Please be patient.
+
+```bash linenums="1"
+# Switch to the working directory
+cd /home/Projects
+# You need to create a docker container for the first run, and do not need to run the current command when you run it again
+# Create a docker container named ppocr and map the current directory to the /paddle directory of the container
+
+#If using CPU, use docker instead of nvidia-docker to create docker
+sudo docker run --name ppocr -v $PWD:/paddle --network=host -it  paddlepaddle/paddle:latest-dev-cuda10.1-cudnn7-gcc82  /bin/bash
+```
+
+With CUDA10, please run the following command to create a container.
+It is recommended to set a shared memory greater than or equal to 32G through the --shm-size parameter:
+
+```bash linenums="1"
+sudo nvidia-docker run --name ppocr -v $PWD:/paddle --shm-size=64G --network=host -it paddlepaddle/paddle:latest-dev-cuda10.1-cudnn7-gcc82 /bin/bash
+```
+
+You can also visit [DockerHub](https://hub.docker.com/r/paddlepaddle/paddle/tags/) to get the image that fits your machine.
+
+```bash linenums="1"
+# ctrl+P+Q to exit docker, to re-enter docker using the following command:
+sudo docker container exec -it ppocr /bin/bash
+```
+
+### 2. Install PaddlePaddle 2.0
+
+```bash linenums="1"
+pip3 install --upgrade pip
+
+# If you have cuda9 or cuda10 installed on your machine, please run the following command to install
+python3 -m pip install paddlepaddle-gpu==2.0.0 -i https://mirror.baidu.com/pypi/simple
+
+# If you only have cpu on your machine, please run the following command to install
+python3 -m pip install paddlepaddle==2.0.0 -i https://mirror.baidu.com/pypi/simple
+```
+
+For more software version requirements, please refer to the instructions in [Installation Document](https://www.paddlepaddle.org.cn/install/quick) for operation.
+
+### 3. Clone PaddleOCR repo
+
+```bash linenums="1"
+# Recommend
+git clone https://github.com/PaddlePaddle/PaddleOCR
+
+# If you cannot pull successfully due to network problems, you can switch to the mirror hosted on Gitee:
+
+git clone https://gitee.com/paddlepaddle/PaddleOCR
+
+# Note: The mirror on Gitee may not keep in synchronization with the latest update with the project on GitHub. There might be a delay of 3-5 days. Please try GitHub at first.
+```
+
+### 4. Install third-party libraries
+
+```bash linenums="1"
+cd PaddleOCR
+pip3 install -r requirements.txt
+```
+
+If you getting this error `OSError: [WinError 126] The specified module could not be found` when you install shapely on windows.
+
+Please try to download Shapely whl file from [http://www.lfd.uci.edu/~gohlke/pythonlibs/#shapely](http://www.lfd.uci.edu/~gohlke/pythonlibs/#shapely).
+
+Reference: [Solve shapely installation on windows](https://stackoverflow.com/questions/44398265/install-shapely-oserror-winerror-126-the-specified-module-could-not-be-found)
diff --git a/docs/version2.x/ppocr/installation.md b/docs/version2.x/ppocr/installation.md
new file mode 100644
index 0000000..a5198e4
--- /dev/null
+++ b/docs/version2.x/ppocr/installation.md
@@ -0,0 +1,76 @@
+---
+comments: true
+---
+
+## 快速安装
+
+经测试PaddleOCR可在glibc 2.23上运行，您也可以测试其他glibc版本或安装glic 2.23
+PaddleOCR 工作环境
+
+- PaddlePaddle 2.0.0
+- python3
+- glibc 2.23
+- cuDNN 7.6+ (GPU)
+
+建议使用我们提供的docker运行PaddleOCR，有关docker、nvidia-docker使用请参考[链接](https://www.runoob.com/docker/docker-tutorial.html/)。
+
+*如您希望使用 mac 或 windows直接运行预测代码，可以从第2步开始执行。*
+
+### 1. （建议）准备docker环境
+
+第一次使用这个镜像，会自动下载该镜像，请耐心等待
+
+```bash linenums="1"
+# 切换到工作目录下
+cd /home/Projects
+# 首次运行需创建一个docker容器，再次运行时不需要运行当前命令
+# 创建一个名字为ppocr的docker容器，并将当前目录映射到容器的/paddle目录下
+
+如果您希望在CPU环境下使用docker，使用docker而不是nvidia-docker创建docker
+sudo docker run --name ppocr -v $PWD:/paddle --network=host -it paddlepaddle/paddle:latest-dev-cuda10.1-cudnn7-gcc82 /bin/bash
+
+如果使用CUDA10，请运行以下命令创建容器，设置docker容器共享内存shm-size为64G，建议设置32G以上
+sudo nvidia-docker run --name ppocr -v $PWD:/paddle --shm-size=64G --network=host -it paddlepaddle/paddle:latest-dev-cuda10.1-cudnn7-gcc82 /bin/bash
+
+您也可以访问[DockerHub](https://hub.docker.com/r/paddlepaddle/paddle/tags/)获取与您机器适配的镜像。
+
+# ctrl+P+Q可退出docker 容器，重新进入docker 容器使用如下命令
+sudo docker container exec -it ppocr /bin/bash
+```
+
+### 2. 安装PaddlePaddle 2.0
+
+```bash linenums="1"
+pip3 install --upgrade pip
+
+# 如果您的机器安装的是CUDA9或CUDA10，请运行以下命令安装
+python3 -m pip install paddlepaddle-gpu==2.0.0 -i https://mirror.baidu.com/pypi/simple
+
+# 如果您的机器是CPU，请运行以下命令安装
+python3 -m pip install paddlepaddle==2.0.0 -i https://mirror.baidu.com/pypi/simple
+
+# 更多的版本需求，请参照[安装文档](https://www.paddlepaddle.org.cn/install/quick)中的说明进行操作。
+```
+
+### 3. 克隆PaddleOCR repo代码
+
+```bash linenums="1"
+#【推荐】
+git clone https://github.com/PaddlePaddle/PaddleOCR
+
+# 如果因为网络问题无法pull成功，也可选择使用码云上的托管：
+
+git clone https://gitee.com/paddlepaddle/PaddleOCR
+
+# 注：码云托管代码可能无法实时同步本github项目更新，存在3~5天延时，请优先使用推荐方式。
+```
+
+### 4. 安装第三方库
+
+```bash linenums="1"
+cd PaddleOCR
+pip3 install -r requirements.txt
+```
+
+注意，windows环境下，建议从[这里](https://www.lfd.uci.edu/~gohlke/pythonlibs/#shapely)下载shapely安装包完成安装，
+直接通过pip安装的shapely库可能出现`[winError 126] 找不到指定模块的问题`。
diff --git a/docs/version2.x/ppocr/model_compress/knowledge_distillation.en.md b/docs/version2.x/ppocr/model_compress/knowledge_distillation.en.md
new file mode 100755
index 0000000..eda44a9
--- /dev/null
+++ b/docs/version2.x/ppocr/model_compress/knowledge_distillation.en.md
@@ -0,0 +1,610 @@
+---
+comments: true
+---
+
+# Knowledge Distillation
+
+## 1. Introduction
+
+### 1.1 Introduction to Knowledge Distillation
+
+In recent years, deep neural networks have been proved to be an extremely effective method for solving problems in the fields of computer vision and natural language processing.
+By constructing a suitable neural network and training it, the performance metrics of the final network model will basically exceed the traditional algorithm.
+When the amount of data is large enough, increasing the amount of parameters by constructing a reasonable network model can significantly improve the performance of the model,
+but this brings about the problem of a sharp increase in the complexity of the model. Large models are more expensive to use in actual scenarios.
+Deep neural networks generally have more parameter redundancy. At present, there are several main methods to compress the model and reduce the amount of its parameters.
+Such as pruning, quantification, knowledge distillation, etc., where knowledge distillation refers to the use of teacher models to guide student models to learn specific tasks,
+to ensure that the small model obtains a relatively large performance improvement under the condition of unchanged parameters.
+In addition, in the knowledge distillation task, a mutual learning model training method was also derived.
+The paper [Deep Mutual Learning](https://arxiv.org/abs/1706.00384) pointed out that using two identical models to supervise each other during the training process can achieve better results than a single model training.
+
+### 1.2 Introduction to PaddleOCR Knowledge Distillation
+
+Whether it is a large model distilling a small model, or a small model learning from each other and updating parameters,
+they are essentially the output between different models or mutual supervision between feature maps.
+The only difference is (1) whether the model requires fixed parameters. (2) Whether the model needs to be loaded with a pre-trained model.
+For the case where a large model distills a small model, the large model generally needs to load the pre-trained model and fix the parameters.
+For the situation where small models distill each other, the small models generally do not load the pre-trained model, and the parameters are also in a learnable state.
+
+In the task of knowledge distillation, it is not only the distillation between two models, but also the situation where multiple models learn from each other.
+Therefore, in the knowledge distillation code framework, it is also necessary to support this type of distillation method.
+
+The algorithm of knowledge distillation is integrated in PaddleOCR. Specifically, it has the following main features:
+
+- It supports mutual learning of any network, and does not require the sub-network structure to be completely consistent or to have a pre-trained model. At the same time, there is no limit to the number of sub-networks, just add it in the configuration file.
+- Support arbitrarily configuring the loss function through the configuration file, not only can use a certain loss, but also a combination of multiple losses.
+- Support all model-related environments such as knowledge distillation training, prediction, evaluation, and export, which is convenient for use and deployment.
+
+Through knowledge distillation, in the common Chinese and English text recognition task, without adding any time-consuming prediction,
+the accuracy of the model can be improved by more than 3%. Combining the learning rate adjustment strategy and the model structure fine-tuning strategy,
+the final improvement is more than 5%.
+
+## 2. Configuration File Analysis
+
+In the process of knowledge distillation training, there is no change in data preprocessing, optimizer, learning rate, and some global attributes.
+The configuration files of the model structure, loss function, post-processing, metric calculation and other modules need to be fine-tuned.
+
+The following takes the knowledge distillation configuration file for recognition and detection as an example to analyze the training and configuration of knowledge distillation.
+
+### 2.1 Recognition Model Configuration File Analysis
+
+The configuration file is in [ch_PP-OCRv2_rec_distillation.yml](../../../../configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec_distillation.yml).
+
+#### 2.1.1 Model Structure
+
+In the knowledge distillation task, the model structure configuration is as follows.
+
+```yaml linenums="1"
+Architecture:
+  model_type: &model_type "rec"    # Model category, recognition, detection, etc.
+  name: DistillationModel          # Structure name, in the distillation task, it is DistillationModel
+  algorithm: Distillation          # Algorithm name
+  Models:                          # Model, including the configuration information of the subnet
+    Teacher:                       # The name of the subnet, it must include at least the `pretrained` and `freeze_params` parameters, and the other parameters are the construction parameters of the subnet
+      pretrained:                  # Does this sub-network need to load pre-training weights
+      freeze_params: false         # Do you need fixed parameters
+      return_all_feats: true       # Do you need to return all features, if it is False, only the final output is returned
+      model_type: *model_type      # Model category
+      algorithm: SVTR              # The algorithm name of the sub-network. The remaining parameters of the sub-network are consistent with the general model training configuration
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+        last_conv_stride: [1, 2]
+        last_pool_type: avg
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 64
+                depth: 2
+                hidden_dims: 120
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - SARHead:
+              enc_dim: 512
+              max_text_length: *max_text_length
+    Student:                       # Another sub-network, here is a distillation example of DML, the two sub-networks have the same structure, and both need to learn parameters
+      pretrained:                  # The following parameters are the same as above
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: SVTR
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+        last_conv_stride: [1, 2]
+        last_pool_type: avg
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 64
+                depth: 2
+                hidden_dims: 120
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - SARHead:
+              enc_dim: 512
+              max_text_length: *max_text_length
+```
+
+If you want to add more sub-networks for training, you can also add the corresponding fields in the configuration file according to the way of adding `Student` and `Teacher`.
+For example, if you want 3 models to supervise each other and train together, then `Architecture` can be written in the following format.
+
+```yaml linenums="1"
+Architecture:
+  model_type: &model_type "rec"
+  name: DistillationModel
+  algorithm: Distillation
+  Models:
+    Teacher:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: SVTR
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+        last_conv_stride: [1, 2]
+        last_pool_type: avg
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 64
+                depth: 2
+                hidden_dims: 120
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - SARHead:
+              enc_dim: 512
+              max_text_length: *max_text_length
+    Student:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: SVTR
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+        last_conv_stride: [1, 2]
+        last_pool_type: avg
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 64
+                depth: 2
+                hidden_dims: 120
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - SARHead:
+              enc_dim: 512
+              max_text_length: *max_text_length
+    Student2:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: SVTR
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+        last_conv_stride: [1, 2]
+        last_pool_type: avg
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 64
+                depth: 2
+                hidden_dims: 120
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - SARHead:
+              enc_dim: 512
+              max_text_length: *max_text_length
+```
+
+When the model is finally trained, it contains 3 sub-networks: `Teacher`, `Student`, `Student2`.
+
+The specific implementation code of the `DistillationModel` class can refer to [distillation_model.py](../../ppocr/modeling/architectures/distillation_model.py).
+The final model output is a dictionary, the key is the name of all the sub-networks, for example, here are `Student` and `Teacher`, and the value is the output of the corresponding sub-network,
+which can be `Tensor` (only the last layer of the network is returned) and `dict` (also returns the characteristic information in the middle).
+In the recognition task, in order to add more loss functions and ensure the scalability of the distillation method, the output of each sub-network is saved as a `dict`, which contains the sub-module output.
+Take the recognition model as an example. The output result of each sub-network is `dict`, the key contains `backbone_out`, `neck_out`, `head_out`, and `value` is the tensor of the corresponding module. Finally, for the above configuration file, `DistillationModel` The output format is as follows.
+
+```json
+{
+  "Teacher": {
+    "backbone_out": tensor,
+    "neck_out": tensor,
+    "head_out": tensor,
+  },
+  "Student": {
+    "backbone_out": tensor,
+    "neck_out": tensor,
+    "head_out": tensor,
+  }
+}
+```
+
+#### 2.1.2 Loss Function
+
+In the knowledge distillation task, the loss function configuration is as follows.
+
+```yaml linenums="1"
+Loss:
+  name: CombinedLoss                           # Loss function name
+  loss_config_list:                            # List of loss function configuration files, mandatory functions for CombinedLoss
+  - DistillationCTCLoss:                       # CTC loss function based on distillation, inherited from standard CTC loss
+      weight: 1.0                              # The weight of the loss function. In loss_config_list, each loss function must include this field
+      model_name_list: ["Student", "Teacher"]  # For the prediction results of the distillation model, extract the output of these two sub-networks and calculate the CTC loss with gt
+      key: head_out                            # In the sub-network output dict, take the corresponding tensor
+  - DistillationDMLLoss:                       # DML loss function, inherited from the standard DMLLoss
+      weight: 1.0
+      act: "softmax"                           # Activation function, use it to process the input, can be softmax, sigmoid or None, the default is None
+      model_name_pairs:                        # The subnet name pair used to calculate DML loss. If you want to calculate the DML loss of other subnets, you can continue to add it below the list
+      - ["Student", "Teacher"]
+      key: head_out
+      multi_head: True                         # whether to use mult_head
+      dis_head: ctc                            # assign the head name to calculate loss
+      name: dml_ctc                            # prefix name of the loss
+  - DistillationDMLLoss:                       # DML loss function, inherited from the standard DMLLoss
+      weight: 0.5
+      act: "softmax"                           # Activation function, use it to process the input, can be softmax, sigmoid or None, the default is None
+      model_name_pairs:                        # The subnet name pair used to calculate DML loss. If you want to calculate the DML loss of other subnets, you can continue to add it below the list
+      - ["Student", "Teacher"]
+      key: head_out
+      multi_head: True                         # whether to use mult_head
+      dis_head: sar                            # assign the head name to calculate loss
+      name: dml_sar                            # prefix name of the loss
+  - DistillationDistanceLoss:                  # Distilled distance loss function
+      weight: 1.0
+      mode: "l2"                               # Support l1, l2 or smooth_l1
+      model_name_pairs:                        # Calculate the distance loss of the subnet name pair
+      - ["Student", "Teacher"]
+      key: backbone_out
+  - DistillationSARLoss:                       # SAR loss function based on distillation, inherited from standard SAR loss
+      weight: 1.0                              # The weight of the loss function. In loss_config_list, each loss function must include this field
+      model_name_list: ["Student", "Teacher"]  # For the prediction results of the distillation model, extract the output of these two sub-networks and calculate the SAR loss with gt
+      key: head_out                            # In the sub-network output dict, take the corresponding tensor
+      multi_head: True                         # whether it is multi-head or not, if true, SAR branch is used to calculate the loss
+```
+
+Among the above loss functions, all distillation loss functions are inherited from the standard loss function class.
+The main functions are: Analyze the output of the distillation model, find the intermediate node (tensor) used to calculate the loss,
+and then use the standard loss function class to calculate.
+
+Taking the above configuration as an example, the final distillation training loss function contains the following five parts.
+
+- CTC branch of the final output `head_out` for `Student` and `Teacher` calculates the CTC loss with gt (loss weight equals 1.0). Here, because both sub-networks need to update the parameters, both of them need to calculate the loss with gt.
+- SAR branch of the final output `head_out` for `Student` and `Teacher` calculates the SAR loss with gt (loss weight equals 1.0). Here, because both sub-networks need to update the parameters, both of them need to calculate the loss with gt.
+- DML loss between CTC branch of  `Student` and `Teacher`'s final output `head_out` (loss weight equals 1.0).
+- DML loss between SAR branch of `Student` and `Teacher`'s final output `head_out` (loss weight equals 0.5).
+- L2 loss between `Student` and `Teacher`'s backbone network output `backbone_out` (loss weight equals 1.0).
+
+For more specific implementation of `CombinedLoss`, please refer to: [combined_loss.py](../../../../ppocr/losses/combined_loss.py#L23).
+For more specific implementations of distillation loss functions such as `DistillationCTCLoss`, please refer to [distillation_loss.py](../../../../ppocr/losses/distillation_loss.py)
+
+#### 2.1.3 Post-processing
+
+In the knowledge distillation task, the post-processing configuration is as follows.
+
+```yaml linenums="1"
+PostProcess:
+  name: DistillationCTCLabelDecode       # CTC decoding post-processing of distillation tasks, inherited from the standard CTCLabelDecode class
+  model_name: ["Student", "Teacher"]     # For the prediction results of the distillation model, extract the outputs of these two sub-networks and decode them
+  key: head_out                          # Take the corresponding tensor in the subnet output dict
+  multi_head: True                       # whether it is multi-head or not, if true, CTC branch is used to calculate the loss
+```
+
+Taking the above configuration as an example, the CTC decoding output of the two sub-networks `Student` and `Teacher` will be calculated at the same time.
+Among them, `key` is the name of the subnet, and `value` is the list of subnets.
+
+For more specific implementation of `DistillationCTCLabelDecode`, please refer to: [rec_postprocess.py](../../../../ppocr/postprocess/rec_postprocess.py#L128)
+
+#### 2.1.4 Metric Calculation
+
+In the knowledge distillation task, the metric calculation configuration is as follows.
+
+```yaml linenums="1"
+Metric:
+  name: DistillationMetric         # CTC decoding post-processing of distillation tasks, inherited from the standard CTCLabelDecode class
+  base_metric_name: RecMetric      # The base class of indicator calculation. For the output of the model, the indicator will be calculated based on this class
+  main_indicator: acc              # The name of the indicator
+  key: "Student"                   # Select the main_indicator of this subnet as the criterion for saving the best model
+  ignore_space: False              # whether to ignore space during evaluation
+```
+
+Taking the above configuration as an example, the accuracy metric of the `Student` subnet will be used as the judgment metric for saving the best model.
+At the same time, the accuracy metric of all subnets will be printed out in the log.
+
+For more specific implementation of `DistillationMetric`, please refer to: [distillation_metric.py](../../../../ppocr/metrics/distillation_metric.py#L24).
+
+#### 2.1.5 Fine-tuning Distillation Model
+
+There are two ways to fine-tune the recognition distillation task.
+
+1. Fine-tuning based on knowledge distillation: this situation is relatively simple, download the pre-trained model. Then configure the pre-training model path and your own data path in [PP-OCRv3_mobile_rec_distillation.yml](../../../../configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec_distillation.yml) to perform fine-tuning training of the model.
+2. Do not use knowledge distillation in fine-tuning: In this case, you need to first extract the student model parameters from the pre-training model. The specific steps are as follows.
+
+- First download the pre-trained model and unzip it.
+
+```bash linenums="1"
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar
+tar -xf ch_PP-OCRv3_rec_train.tar
+```
+
+- Then use python to extract the student model parameters
+
+```python linenums="1"
+import paddle
+# Load the pre-trained model
+all_params = paddle.load("ch_PP-OCRv3_rec_train/best_accuracy.pdparams")
+# View the keys of the weight parameter
+print(all_params.keys())
+# Weight extraction of student model
+s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
+# View the keys of the weight parameters of the student model
+print(s_params.keys())
+# Save weight parameters
+paddle.save(s_params, "ch_PP-OCRv3_rec_train/student.pdparams")
+```
+
+After the extraction is complete, use [PP-OCRv3_mobile_rec.yml](../../../../configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec.yml) to modify the path of the pre-trained model (the path of the exported `student.pdparams` model) and your own data path to fine-tune the model.
+
+### 2.2 Detection Model Configuration File Analysis
+
+The configuration file of the detection model distillation is in the ```PaddleOCR/configs/det/ch_PP-OCRv3/``` directory, which contains three distillation configuration files:
+
+- ```PP-OCRv3_det_cml.yml```, Use one large model to distill two small models, and the two small models learn from each other
+- ```PP-OCRv3_det_dml.yml```, Method of mutual distillation of two student models
+
+#### 2.2.1 Model Structure
+
+In the knowledge distillation task, the model structure configuration is as follows:
+
+```yaml linenums="1"
+Architecture:
+  name: DistillationModel          # Structure name, in the distillation task, it is DistillationModel
+  algorithm: Distillation          # Algorithm name
+  Models:                          # Model, including the configuration information of the subnet
+    Student:                       # The name of the subnet, it must include at least the `pretrained` and `freeze_params` parameters, and the other parameters are the construction parameters of the subnet
+      pretrained: ./pretrain_models/MobileNetV3_large_x0_5_pretrained  # Does this sub-network need to load pre-training weights
+      freeze_params: false         # Do you need fixed parameters
+      return_all_feats: false      # Do you need to return all features, if it is False, only the final output is returned
+      model_type: det
+      algorithm: DB
+      Backbone:
+        name: ResNet
+        in_channels: 3
+        layers: 50
+      Neck:
+        name: LKPAN
+        out_channels: 256
+      Head:
+        name: DBHead
+        kernel_list: [7,2,2]
+        k: 50
+    Teacher:                      # Another sub-network, here is a distillation example of a large model distill a small model
+      pretrained: ./pretrain_models/ch_ppocr_server_v2.0_det_train/best_accuracy
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Transform:
+      Backbone:
+        name: ResNet
+        in_channels: 3
+        layers: 50
+      Neck:
+        name: LKPAN
+        out_channels: 256
+      Head:
+        name: DBHead
+        kernel_list: [7,2,2]
+        k: 50
+
+```
+
+If DML is used, that is, the method of two small models learning from each other, the Teacher network structure in the above configuration file needs to be set to the same configuration as the Student model.
+Refer to the configuration file for details. [PP-OCRv3_det_dml.yml](../../../../configs/det/PP-OCRv3/PP-OCRv3_det_dml.yml)
+
+The following describes the configuration file parameters [PP-OCRv3_det_cml.yml](../../../../configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml):
+
+```yaml linenums="1"
+Architecture:
+  name: DistillationModel
+  algorithm: Distillation
+  model_type: det
+  Models:
+    Teacher:                         # Teacher model configuration of CML distillation
+      pretrained: ./pretrain_models/ch_ppocr_server_v2.0_det_train/best_accuracy
+      freeze_params: true            # Teacher does not train
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Transform:
+      Backbone:
+        name: ResNet
+        in_channels: 3
+        layers: 50
+      Neck:
+        name: LKPAN
+        out_channels: 256
+      Head:
+        name: DBHead
+        kernel_list: [7,2,2]
+        k: 50
+    Student:                         # Student model configuration for CML distillation
+      pretrained: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+      freeze_params: false
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Backbone:
+        name: MobileNetV3
+        scale: 0.5
+        model_name: large
+        disable_se: true
+      Neck:
+        name: RSEFPN
+        out_channels: 96
+        shortcut: True
+      Head:
+        name: DBHead
+        k: 50
+    Student2:                          # Student2 model configuration for CML distillation
+      pretrained: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+      freeze_params: false
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Transform:
+      Backbone:
+        name: MobileNetV3
+        scale: 0.5
+        model_name: large
+        disable_se: true
+      Neck:
+        name: RSEFPN
+        out_channels: 96
+        shortcut: True
+      Head:
+        name: DBHead
+        k: 50
+
+```
+
+The specific implementation code of the distillation model `DistillationModel` class can refer to [distillation_model.py](../../../../ppocr/modeling/architectures/distillation_model.py).
+
+The final model output is a dictionary, the key is the name of all the sub-networks, for example, here are `Student` and `Teacher`, and the value is the output of the corresponding sub-network,
+which can be `Tensor` (only the last layer of the network is returned) and `dict` (also returns the characteristic information in the middle).
+
+In the distillation task, in order to facilitate the addition of the distillation loss function, the output of each network is saved as a `dict`, which contains the sub-module output.
+The key contains `backbone_out`, `neck_out`, `head_out`, and `value` is the tensor of the corresponding module. Finally, for the above configuration file, the output format of `DistillationModel` is as follows.
+
+```json
+{
+  "Teacher": {
+    "backbone_out": tensor,
+    "neck_out": tensor,
+    "head_out": tensor,
+  },
+  "Student": {
+    "backbone_out": tensor,
+    "neck_out": tensor,
+    "head_out": tensor,
+  }
+}
+```
+
+#### 2.2.2 Loss Function
+
+The distillation loss function configuration(`PP-OCRv3_det_cml.yml`) is shown below.
+
+```yaml linenums="1"
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationDilaDBLoss:
+      weight: 1.0
+      model_name_pairs:
+      - ["Student", "Teacher"]
+      - ["Student2", "Teacher"]                  # 1. Calculate the loss of two Student and Teacher
+      key: maps
+      balance_loss: true
+      main_loss_type: DiceLoss
+      alpha: 5
+      beta: 10
+      ohem_ratio: 3
+  - DistillationDMLLoss:                         # 2. Add to calculate the loss between two students
+      model_name_pairs:
+      - ["Student", "Student2"]
+      maps_name: "thrink_maps"
+      weight: 1.0
+      # act: None
+      key: maps
+  - DistillationDBLoss:
+      weight: 1.0
+      model_name_list: ["Student", "Student2"]   # 3. Calculate the loss between two students and GT
+      balance_loss: true
+      main_loss_type: DiceLoss
+      alpha: 5
+      beta: 10
+      ohem_ratio: 3
+```
+
+For more specific implementation of `DistillationDilaDBLoss`, please refer to: [distillation_loss.py](../../../../ppocr/losses/distillation_loss.py#L185).
+For more specific implementations of distillation loss functions such as `DistillationDBLoss`, please refer to: [distillation_loss.py](../../../../ppocr/losses/distillation_loss.py#L148)
+
+#### 2.2.3 Post-processing
+
+In the task of detecting knowledge distillation, the post-processing configuration of detecting distillation is as follows.
+
+```yaml linenums="1"
+PostProcess:
+  name: DistillationDBPostProcess                  # The post-processing of the DB detection distillation task, inherited from the standard DBPostProcess class
+  model_name: ["Student", "Student2", "Teacher"]   # Extract the output of multiple sub-networks and decode them. The network that does not require post-processing is not set in model_name
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+```
+
+Taking the above configuration as an example, the output of the three subnets `Student`, `Student2` and `Teacher` will be calculated at the same time for post-processing calculations.
+Since there are multiple inputs, there are also multiple outputs returned by post-processing.
+For a more specific implementation of `DistillationDBPostProcess`, please refer to: [db_postprocess.py](../../../../ppocr/postprocess/db_postprocess.py#L195)
+
+#### 2.2.4 Metric Calculation
+
+In the knowledge distillation task, the metric calculation configuration is as follows.
+
+```yaml linenums="1"
+Metric:
+  name: DistillationMetric
+  base_metric_name: DetMetric
+  main_indicator: hmean
+  key: "Student"
+```
+
+Since distillation needs to include multiple networks, only one network metrics needs to be calculated when calculating the metrics.
+The `key` field is set to `Student`, it means that only the metrics of the `Student` network is calculated.
+Model Structure
+
+#### 2.2.5 Fine-tuning Distillation Model
+
+There are three ways to fine-tune the detection distillation task:
+
+- `ch_PP-OCRv3_det_distill.yml`, The teacher model is set to the model provided by PaddleOCR or the large model you have trained.
+- `PP-OCRv3_det_cml.yml`, Use cml distillation. Similarly, the Teacher model is set to the model provided by PaddleOCR or the large model you have trained.
+- `PP-OCRv3_det_dml.yml`, Distillation using DML. The method of mutual distillation of the two Student models has an accuracy improvement of about 1.7% on the data set used by PaddleOCR.
+
+In fine-tune, you need to set the pre-trained model to be loaded in the `pretrained` parameter of the network structure.
+
+In terms of accuracy improvement, `cml` > `dml` > `distill`. When the amount of data is insufficient or the accuracy of the teacher model is similar to that of the student, this conclusion may change.
+
+In addition, since the distillation pre-training model provided by PaddleOCR contains multiple model parameters, if you want to extract the parameters of the student model, you can refer to the following code:
+
+```sh
+# Download the parameters of the distillation training model
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams
+```
+
+```python linenums="1"
+import paddle
+# Load the pre-trained model
+all_params = paddle.load("ch_PP-OCRv3_det_distill_train/best_accuracy.pdparams")
+# View the keys of the weight parameter
+print(all_params.keys())
+# Extract the weights of the student model
+s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
+# View the keys of the weight parameters of the student model
+print(s_params.keys())
+# Save
+paddle.save(s_params, "ch_PP-OCRv3_det_distill_train/student.pdparams")
+```
+
+Finally, the parameters of the student model will be saved in `ch_PP-OCRv3_det_distill_train/student.pdparams` for the fine-tune of the model.
diff --git a/docs/version2.x/ppocr/model_compress/knowledge_distillation.md b/docs/version2.x/ppocr/model_compress/knowledge_distillation.md
new file mode 100644
index 0000000..9aa5ea7
--- /dev/null
+++ b/docs/version2.x/ppocr/model_compress/knowledge_distillation.md
@@ -0,0 +1,593 @@
+---
+comments: true
+---
+
+# 知识蒸馏
+
+## 1. 简介
+
+### 1.1 知识蒸馏介绍
+
+近年来，深度神经网络在计算机视觉、自然语言处理等领域被验证是一种极其有效的解决问题的方法。通过构建合适的神经网络，加以训练，最终网络模型的性能指标基本上都会超过传统算法。
+
+在数据量足够大的情况下，通过合理构建网络模型的方式增加其参数量，可以显著改善模型性能，但是这又带来了模型复杂度急剧提升的问题。大模型在实际场景中使用的成本较高。
+
+深度神经网络一般有较多的参数冗余，目前有几种主要的方法对模型进行压缩，减小其参数量。如裁剪、量化、知识蒸馏等，其中知识蒸馏是指使用教师模型(teacher model)去指导学生模型(student model)学习特定任务，保证小模型在参数量不变的情况下，得到比较大的性能提升。
+
+此外，在知识蒸馏任务中，也衍生出了互学习的模型训练方法，论文[Deep Mutual Learning](https://arxiv.org/abs/1706.00384)中指出，使用两个完全相同的模型在训练的过程中互相监督，可以达到比单个模型训练更好的效果。
+
+### 1.2 PaddleOCR知识蒸馏简介
+
+无论是大模型蒸馏小模型，还是小模型之间互相学习，更新参数，他们本质上是都是不同模型之间输出或者特征图(feature map)之间的相互监督，区别仅在于 (1) 模型是否需要固定参数。(2) 模型是否需要加载预训练模型。
+
+对于大模型蒸馏小模型的情况，大模型一般需要加载预训练模型并固定参数；对于小模型之间互相蒸馏的情况，小模型一般都不加载预训练模型，参数也都是可学习的状态。
+
+在知识蒸馏任务中，不只有2个模型之间进行蒸馏的情况，多个模型之间互相学习的情况也非常普遍。因此在知识蒸馏代码框架中，也有必要支持该种类别的蒸馏方法。
+
+PaddleOCR中集成了知识蒸馏的算法，具体地，有以下几个主要的特点：
+
+- 支持任意网络的互相学习，不要求子网络结构完全一致或者具有预训练模型；同时子网络数量也没有任何限制，只需要在配置文件中添加即可。
+- 支持loss函数通过配置文件任意配置，不仅可以使用某种loss，也可以使用多种loss的组合
+- 支持知识蒸馏训练、预测、评估与导出等所有模型相关的环境，方便使用与部署。
+
+通过知识蒸馏，在中英文通用文字识别任务中，不增加任何预测耗时的情况下，可以给模型带来3%以上的精度提升，结合学习率调整策略以及模型结构微调策略，最终提升提升超过5%。
+
+## 2. 配置文件解析
+
+在知识蒸馏训练的过程中，数据预处理、优化器、学习率、全局的一些属性没有任何变化。模型结构、损失函数、后处理、指标计算等模块的配置文件需要进行微调。
+
+下面以识别与检测的知识蒸馏配置文件为例，对知识蒸馏的训练与配置进行解析。
+
+### 2.1 识别配置文件解析
+
+配置文件在[PP-OCRv3_mobile_rec_distillation.yml](../../../../configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec_distillation.yml)。
+
+#### 2.1.1 模型结构
+
+知识蒸馏任务中，模型结构配置如下所示。
+
+```yaml linenums="1"
+Architecture:
+  model_type: &model_type "rec"    # 模型类别，rec、det等，每个子网络的模型类别
+  name: DistillationModel          # 结构名称，蒸馏任务中，为DistillationModel，用于构建对应的结构
+  algorithm: Distillation          # 算法名称
+  Models:                          # 模型，包含子网络的配置信息
+    Teacher:                       # 子网络名称，至少需要包含`pretrained`与`freeze_params`信息，其他的参数为子网络的构造参数
+      pretrained:                  # 该子网络是否需要加载预训练模型
+      freeze_params: false         # 是否需要固定参数
+      return_all_feats: true       # 子网络的参数，表示是否需要返回所有的features，如果为False，则只返回最后的输出
+      model_type: *model_type      # 模型类别
+      algorithm: SVTR              # 子网络的算法名称，该子网络其余参数均为构造参数，与普通的模型训练配置一致
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+        last_conv_stride: [1, 2]
+        last_pool_type: avg
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 64
+                depth: 2
+                hidden_dims: 120
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - SARHead:
+              enc_dim: 512
+              max_text_length: *max_text_length
+    Student:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: SVTR
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+        last_conv_stride: [1, 2]
+        last_pool_type: avg
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 64
+                depth: 2
+                hidden_dims: 120
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - SARHead:
+              enc_dim: 512
+              max_text_length: *max_text_length
+```
+
+当然，这里如果希望添加更多的子网络进行训练，也可以按照`Student`与`Teacher`的添加方式，在配置文件中添加相应的字段。比如说如果希望有3个模型互相监督，共同训练，那么`Architecture`可以写为如下格式。
+
+```yaml linenums="1"
+Architecture:
+  model_type: &model_type "rec"
+  name: DistillationModel
+  algorithm: Distillation
+  Models:
+    Teacher:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: SVTR
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+        last_conv_stride: [1, 2]
+        last_pool_type: avg
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 64
+                depth: 2
+                hidden_dims: 120
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - SARHead:
+              enc_dim: 512
+              max_text_length: *max_text_length
+    Student:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: SVTR
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+        last_conv_stride: [1, 2]
+        last_pool_type: avg
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 64
+                depth: 2
+                hidden_dims: 120
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - SARHead:
+              enc_dim: 512
+              max_text_length: *max_text_length
+    Student2:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: SVTR
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+        last_conv_stride: [1, 2]
+        last_pool_type: avg
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 64
+                depth: 2
+                hidden_dims: 120
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - SARHead:
+              enc_dim: 512
+              max_text_length: *max_text_length
+```
+
+最终该模型训练时，包含3个子网络：`Teacher`, `Student`, `Student2`。
+
+蒸馏模型`DistillationModel`类的具体实现代码可以参考[distillation_model.py](../../ppocr/modeling/architectures/distillation_model.py)。
+
+最终模型`forward`输出为一个字典，key为所有的子网络名称，例如这里为`Student`与`Teacher`，value为对应子网络的输出，可以为`Tensor`（只返回该网络的最后一层）和`dict`（也返回了中间的特征信息）。
+
+在识别任务中，为了添加更多损失函数，保证蒸馏方法的可扩展性，将每个子网络的输出保存为`dict`，其中包含子模块输出。以该识别模型为例，每个子网络的输出结果均为`dict`，key包含`backbone_out`,`neck_out`, `head_out`，`value`为对应模块的tensor，最终对于上述配置文件，`DistillationModel`的输出格式如下：
+
+```json
+{
+  "Teacher": {
+    "backbone_out": tensor,
+    "neck_out": tensor,
+    "head_out": tensor,
+  },
+  "Student": {
+    "backbone_out": tensor,
+    "neck_out": tensor,
+    "head_out": tensor,
+  }
+}
+```
+
+#### 2.1.2 损失函数
+
+知识蒸馏任务中，损失函数配置如下所示。
+
+```yaml linenums="1"
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationDMLLoss:                       # 蒸馏的DML损失函数，继承自标准的DMLLoss
+      weight: 1.0                              # 权重
+      act: "softmax"                           # 激活函数，对输入使用激活函数处理，可以为softmax, sigmoid或者为None，默认为None
+      use_log: true                            # 对输入计算log，如果函数已经
+      model_name_pairs:                        # 用于计算DML loss的子网络名称对，如果希望计算其他子网络的DML loss，可以在列表下面继续填充
+      - ["Student", "Teacher"]
+      key: head_out                            # 取子网络输出dict中，该key对应的tensor
+      multi_head: True                         # 是否为多头结构
+      dis_head: ctc                            # 指定用于计算损失函数的head
+      name: dml_ctc                            # 蒸馏loss的前缀名称，避免不同loss之间的命名冲突
+  - DistillationDMLLoss:                       # 蒸馏的DML损失函数，继承自标准的DMLLoss
+      weight: 0.5                              # 权重
+      act: "softmax"                           # 激活函数，对输入使用激活函数处理，可以为softmax, sigmoid或者为None，默认为None
+      use_log: true                            # 对输入计算log，如果函数已经
+      model_name_pairs:                        # 用于计算DML loss的子网络名称对，如果希望计算其他子网络的DML loss，可以在列表下面继续填充
+      - ["Student", "Teacher"]
+      key: head_out                            # 取子网络输出dict中，该key对应的tensor
+      multi_head: True                         # 是否为多头结构
+      dis_head: sar                            # 指定用于计算损失函数的head
+      name: dml_sar                            # 蒸馏loss的前缀名称，避免不同loss之间的命名冲突
+  - DistillationDistanceLoss:                  # 蒸馏的距离损失函数
+      weight: 1.0                              # 权重
+      mode: "l2"                               # 距离计算方法，目前支持l1, l2, smooth_l1
+      model_name_pairs:                        # 用于计算distance loss的子网络名称对
+      - ["Student", "Teacher"]
+      key: backbone_out                        # 取子网络输出dict中，该key对应的tensor
+  - DistillationCTCLoss:                       # 基于蒸馏的CTC损失函数，继承自标准的CTC loss
+      weight: 1.0                              # 损失函数的权重，loss_config_list中，每个损失函数的配置都必须包含该字段
+      model_name_list: ["Student", "Teacher"]  # 对于蒸馏模型的预测结果，提取这两个子网络的输出，与gt计算CTC loss
+      key: head_out                            # 取子网络输出dict中，该key对应的tensor
+  - DistillationSARLoss:                       # 基于蒸馏的SAR损失函数，继承自标准的SARLoss
+      weight: 1.0                              # 损失函数的权重，loss_config_list中，每个损失函数的配置都必须包含该字段
+      model_name_list: ["Student", "Teacher"]  # 对于蒸馏模型的预测结果，提取这两个子网络的输出，与gt计算CTC loss
+      key: head_out                            # 取子网络输出dict中，该key对应的tensor
+      multi_head: True                         # 是否为多头结构，为true时，取出其中的SAR分支计算损失函数
+```
+
+上述损失函数中，所有的蒸馏损失函数均继承自标准的损失函数类，主要功能为: 对蒸馏模型的输出进行解析，找到用于计算损失的中间节点(tensor)，再使用标准的损失函数类去计算。
+
+以上述配置为例，最终蒸馏训练的损失函数包含下面5个部分。
+
+- `Student`和`Teacher`最终输出(`head_out`)的CTC分支与gt的CTC loss，权重为1。在这里因为2个子网络都需要更新参数，因此2者都需要计算与g的loss。
+- `Student`和`Teacher`最终输出(`head_out`)的SAR分支与gt的SAR loss，权重为1.0。在这里因为2个子网络都需要更新参数，因此2者都需要计算与g的loss。
+- `Student`和`Teacher`最终输出(`head_out`)的CTC分支之间的DML loss，权重为1。
+- `Student`和`Teacher`最终输出(`head_out`)的SAR分支之间的DML loss，权重为0.5。
+- `Student`和`Teacher`的骨干网络输出(`backbone_out`)之间的l2 loss，权重为1。
+
+关于`CombinedLoss`更加具体的实现可以参考: [combined_loss.py](../../../../ppocr/losses/combined_loss.py#L23)。关于`DistillationCTCLoss`等蒸馏损失函数更加具体的实现可以参考[distillation_loss.py](../../../../ppocr/losses/distillation_loss.py)。
+
+#### 2.1.3 后处理
+
+知识蒸馏任务中，后处理配置如下所示。
+
+```yaml linenums="1"
+PostProcess:
+  name: DistillationCTCLabelDecode       # 蒸馏任务的CTC解码后处理，继承自标准的CTCLabelDecode类
+  model_name: ["Student", "Teacher"]     # 对于蒸馏模型的预测结果，提取这两个子网络的输出，进行解码
+  key: head_out                          # 取子网络输出dict中，该key对应的tensor
+  multi_head: True                       # 多头结构时，会取出其中的CTC分支进行计算
+```
+
+以上述配置为例，最终会同时计算`Student`和`Teahcer` 2个子网络的CTC解码输出，返回一个`dict`，`key`为用于处理的子网络名称，`value`为用于处理的子网络列表。
+
+关于`DistillationCTCLabelDecode`更加具体的实现可以参考: [rec_postprocess.py](../../../../ppocr/postprocess/rec_postprocess.py#L128)
+
+#### 2.1.4 指标计算
+
+知识蒸馏任务中，指标计算配置如下所示。
+
+```yaml linenums="1"
+Metric:
+  name: DistillationMetric         # 蒸馏任务的CTC解码后处理，继承自标准的CTCLabelDecode类
+  base_metric_name: RecMetric      # 指标计算的基类，对于模型的输出，会基于该类，计算指标
+  main_indicator: acc              # 指标的名称
+  key: "Student"                   # 选取该子网络的 main_indicator 作为作为保存保存best model的判断标准
+  ignore_space: False              # 评估时是否忽略空格的影响
+```
+
+以上述配置为例，最终会使用`Student`子网络的acc指标作为保存best model的判断指标，同时，日志中也会打印出所有子网络的acc指标。
+
+关于`DistillationMetric`更加具体的实现可以参考: [distillation_metric.py](../../../../ppocr/metrics/distillation_metric.py#L24)。
+
+#### 2.1.5 蒸馏模型微调
+
+对蒸馏得到的识别蒸馏进行微调有2种方式。
+
+（1）基于知识蒸馏的微调：这种情况比较简单，下载预训练模型，在[PP-OCRv3_mobile_rec_distillation.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec_distillation.yml)中配置好预训练模型路径以及自己的数据路径，即可进行模型微调训练。
+
+（2）微调时不使用知识蒸馏：这种情况，需要首先将预训练模型中的学生模型参数提取出来，具体步骤如下：
+
+- 首先下载预训练模型并解压。
+
+```bash linenums="1"
+# 下面预训练模型并解压
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar
+tar -xf ch_PP-OCRv3_rec_train.tar
+```
+
+- 然后使用python，对其中的学生模型参数进行提取
+
+```python linenums="1"
+import paddle
+# 加载预训练模型
+all_params = paddle.load("ch_PP-OCRv3_rec_train/best_accuracy.pdparams")
+# 查看权重参数的keys
+print(all_params.keys())
+# 学生模型的权重提取
+s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
+# 查看学生模型权重参数的keys
+print(s_params.keys())
+# 保存
+paddle.save(s_params, "ch_PP-OCRv3_rec_train/student.pdparams")
+```
+
+转化完成之后，使用[PP-OCRv3_mobile_rec.yml](../../../../configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec.yml)，修改预训练模型的路径（为导出的`student.pdparams`模型路径）以及自己的数据路径，即可进行模型微调。
+
+### 2.2 检测配置文件解析
+
+检测模型蒸馏的配置文件在PaddleOCR/configs/det/ch_PP-OCRv3/目录下，包含两个个蒸馏配置文件：
+
+- PP-OCRv3_det_cml.yml，采用cml蒸馏，采用一个大模型蒸馏两个小模型，且两个小模型互相学习的方法
+- PP-OCRv3_det_dml.yml，采用DML的蒸馏，两个Student模型互蒸馏的方法
+
+#### 2.2.1 模型结构
+
+知识蒸馏任务中，模型结构配置如下所示：
+
+```yaml linenums="1"
+Architecture:
+  name: DistillationModel          # 结构名称，蒸馏任务中，为DistillationModel，用于构建对应的结构
+  algorithm: Distillation          # 算法名称
+  Models:                          # 模型，包含子网络的配置信息
+    Student:                       # 子网络名称，至少需要包含`pretrained`与`freeze_params`信息，其他的参数为子网络的构造参数
+      freeze_params: false         # 是否需要固定参数
+      return_all_feats: false      # 子网络的参数，表示是否需要返回所有的features，如果为False，则只返回最后的输出
+      model_type: det
+      algorithm: DB
+      Backbone:
+        name: ResNet
+        in_channels: 3
+        layers: 50
+      Neck:
+        name: LKPAN
+        out_channels: 256
+      Head:
+        name: DBHead
+        kernel_list: [7,2,2]
+        k: 50
+    Teacher:                      # 另外一个子网络，这里给的是DML蒸馏示例，
+      freeze_params: true
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Transform:
+      Backbone:
+        name: ResNet
+        in_channels: 3
+        layers: 50
+      Neck:
+        name: LKPAN
+        out_channels: 256
+      Head:
+        name: DBHead
+        kernel_list: [7,2,2]
+        k: 50
+
+```
+
+如果是采用DML，即两个小模型互相学习的方法，上述配置文件里的Teacher网络结构需要设置为Student模型一样的配置，具体参考配置文件[PP-OCRv3_det_dml.yml](../../../../configs/det/PP-OCRv3/PP-OCRv3_det_dml.yml)。
+
+下面介绍[PP-OCRv3_det_cml.yml](../../../../configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml)的配置文件参数：
+
+```yaml linenums="1"
+Architecture:
+  name: DistillationModel
+  algorithm: Distillation
+  model_type: det
+  Models:
+    Teacher:                         # CML蒸馏的Teacher模型配置
+      pretrained: ./pretrain_models/ch_ppocr_server_v2.0_det_train/best_accuracy
+      freeze_params: true            # Teacher 不训练
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Transform:
+      Backbone:
+        name: ResNet
+        in_channels: 3
+        layers: 50
+      Neck:
+        name: LKPAN
+        out_channels: 256
+      Head:
+        name: DBHead
+        kernel_list: [7,2,2]
+        k: 50
+    Student:                         # CML蒸馏的Student模型配置
+      pretrained: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+      freeze_params: false
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Backbone:
+        name: MobileNetV3
+        scale: 0.5
+        model_name: large
+        disable_se: true
+      Neck:
+        name: RSEFPN
+        out_channels: 96
+        shortcut: True
+      Head:
+        name: DBHead
+        k: 50
+    Student2:                          # CML蒸馏的Student2模型配置
+      pretrained: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+      freeze_params: false
+      return_all_feats: false
+      model_type: det
+      algorithm: DB
+      Transform:
+      Backbone:
+        name: MobileNetV3
+        scale: 0.5
+        model_name: large
+        disable_se: true
+      Neck:
+        name: RSEFPN
+        out_channels: 96
+        shortcut: True
+      Head:
+        name: DBHead
+        k: 50
+
+```
+
+蒸馏模型`DistillationModel`类的具体实现代码可以参考[distillation_model.py](../../../../ppocr/modeling/architectures/distillation_model.py)。
+
+最终模型`forward`输出为一个字典，key为所有的子网络名称，例如这里为`Student`与`Teacher`，value为对应子网络的输出，可以为`Tensor`（只返回该网络的最后一层）和`dict`（也返回了中间的特征信息）。
+
+在蒸馏任务中，为了方便添加蒸馏损失函数，每个网络的输出保存为`dict`，其中包含子模块输出。每个子网络的输出结果均为`dict`，key包含`backbone_out`,`neck_out`, `head_out`，`value`为对应模块的tensor，最终对于上述配置文件，`DistillationModel`的输出格式如下：
+
+```json
+{
+  "Teacher": {
+    "backbone_out": tensor,
+    "neck_out": tensor,
+    "head_out": tensor,
+  },
+  "Student": {
+    "backbone_out": tensor,
+    "neck_out": tensor,
+    "head_out": tensor,
+  }
+}
+```
+
+#### 2.2.2 损失函数
+
+检测PP-OCRv3_det_cml.yml蒸馏损失函数配置如下所示。
+
+```yaml linenums="1"
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationDilaDBLoss:
+      weight: 1.0
+      model_name_pairs:
+      - ["Student", "Teacher"]
+      - ["Student2", "Teacher"]                  # 改动1，计算两个Student和Teacher的损失
+      key: maps
+      balance_loss: true
+      main_loss_type: DiceLoss
+      alpha: 5
+      beta: 10
+      ohem_ratio: 3
+  - DistillationDMLLoss:                         # 改动2，增加计算两个Student之间的损失
+      model_name_pairs:
+      - ["Student", "Student2"]
+      maps_name: "thrink_maps"
+      weight: 1.0
+      # act: None
+      key: maps
+  - DistillationDBLoss:
+      weight: 1.0
+      model_name_list: ["Student", "Student2"]   # 改动3，计算两个Student和GT之间的损失
+      balance_loss: true
+      main_loss_type: DiceLoss
+      alpha: 5
+      beta: 10
+      ohem_ratio: 3
+
+```
+
+关于`DistillationDilaDBLoss`更加具体的实现可以参考: [distillation_loss.py](../../../../ppocr/losses/distillation_loss.py#L185)。关于`DistillationDBLoss`等蒸馏损失函数更加具体的实现可以参考[distillation_loss.py](../../../../ppocr/losses/distillation_loss.py#L148)。
+
+#### 2.2.3 后处理
+
+知识蒸馏任务中，检测蒸馏后处理配置如下所示。
+
+```yaml linenums="1"
+PostProcess:
+  name: DistillationDBPostProcess                  # DB检测蒸馏任务的CTC解码后处理，继承自标准的DBPostProcess类
+  model_name: ["Student", "Student2", "Teacher"]   # 对于蒸馏模型的预测结果，提取多个子网络的输出，进行解码，不需要后处理的网络可以不在model_name中设置
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+```
+
+以上述配置为例，最终会同时计算`Student`，`Student2`和`Teacher` 3个子网络的输出做后处理计算。同时，由于有多个输入，后处理返回的输出也有多个，
+
+关于`DistillationDBPostProcess`更加具体的实现可以参考: [db_postprocess.py](../../../../ppocr/postprocess/db_postprocess.py#L195)
+
+#### 2.2.4 蒸馏指标计算
+
+知识蒸馏任务中，检测蒸馏指标计算配置如下所示。
+
+```yaml linenums="1"
+Metric:
+  name: DistillationMetric
+  base_metric_name: DetMetric
+  main_indicator: hmean
+  key: "Student"
+```
+
+由于蒸馏需要包含多个网络，甚至多个Student网络，在计算指标的时候只需要计算一个Student网络的指标即可，`key`字段设置为`Student`则表示只计算`Student`网络的精度。
+
+#### 2.2.5 检测蒸馏模型finetune
+
+PP-OCRv3检测蒸馏有两种方式：
+
+- 采用PP-OCRv3_det_cml.yml，采用cml蒸馏，同样Teacher模型设置为PaddleOCR提供的模型或者您训练好的大模型
+- 采用PP-OCRv3_det_dml.yml，采用DML的蒸馏，两个Student模型互蒸馏的方法，在PaddleOCR采用的数据集上相比单独训练Student模型有1%-2%的提升。
+
+在具体fine-tune时，需要在网络结构的`pretrained`参数中设置要加载的预训练模型。
+
+在精度提升方面，cml的精度>dml的精度蒸馏方法的精度。当数据量不足或者Teacher模型精度与Student精度相差不大的时候，这个结论或许会改变。
+
+另外，由于PaddleOCR提供的蒸馏预训练模型包含了多个模型的参数，如果您希望提取Student模型的参数，可以参考如下代码：
+
+```bash linenums="1"
+# 下载蒸馏训练模型的参数
+wget https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv3_det_distill_train.tar
+```
+
+```python linenums="1"
+import paddle
+# 加载预训练模型
+all_params = paddle.load("ch_PP-OCRv3_det_distill_train/best_accuracy.pdparams")
+# 查看权重参数的keys
+print(all_params.keys())
+# 学生模型的权重提取
+s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
+# 查看学生模型权重参数的keys
+print(s_params.keys())
+# 保存
+paddle.save(s_params, "ch_PP-OCRv3_det_distill_train/student.pdparams")
+```
+
+最终`Student`模型的参数将会保存在`ch_PP-OCRv3_det_distill_train/student.pdparams`中，用于模型的fine-tune。
diff --git a/docs/version2.x/ppocr/model_compress/prune.en.md b/docs/version2.x/ppocr/model_compress/prune.en.md
new file mode 100644
index 0000000..8b02858
--- /dev/null
+++ b/docs/version2.x/ppocr/model_compress/prune.en.md
@@ -0,0 +1,75 @@
+---
+comments: true
+---
+
+# PP-OCR Models Pruning
+
+Generally, a more complex model would achieve better performance in the task, but it also leads to some redundancy in the model. Model Pruning is a technique that reduces this redundancy by removing the sub-models in the neural network model, so as to reduce model calculation complexity and improve model inference performance.
+
+This example uses PaddleSlim provided [APIs of Pruning](https://github.com/PaddlePaddle/PaddleSlim/tree/develop/docs/zh_cn/api_cn/dygraph/pruners) to compress the OCR model.
+[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim), an open source library which integrates model pruning, quantization (including quantization training and offline quantization), distillation, neural network architecture search, and many other commonly used and leading model compression technique in the industry.
+
+It is recommended that you could understand following pages before reading this example：
+
+1. [PaddleOCR training methods](../model_train/training.en.md)
+2. [The demo of prune](https://github.com/PaddlePaddle/PaddleSlim/blob/release%2F2.0.0/docs/zh_cn/tutorials/pruning/dygraph/filter_pruning.md)
+
+## Quick start
+
+### 1. Install PaddleSlim
+
+```bash linenums="1"
+git clone https://github.com/PaddlePaddle/PaddleSlim.git
+cd PaddleSlim
+git checkout develop
+python3 setup.py install
+```
+
+### 2. Download Pre-trained Model
+
+Model prune needs to load pre-trained models.
+PaddleOCR also provides a series of [models](../model_list.en.md). Developers can choose their own models or use their own models according to their needs.
+
+### 3. Pruning sensitivity analysis
+
+After the pre-trained model is loaded, sensitivity analysis is performed on each network layer of the model to understand the redundancy of each network layer, and save a sensitivity file which named: sen.pickle.  After that, user could load the sensitivity file via the [methods provided by PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/paddleslim/prune/sensitive.py#L221) and determining the pruning ratio of each network layer automatically. For specific details of sensitivity analysis, see：[Sensitivity analysis](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/en/tutorials/image_classification_sensitivity_analysis_tutorial_en.md)
+The data format of sensitivity file：
+
+```python linenums="1"
+sen.pickle(Dict){
+              'layer_weight_name_0': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+              'layer_weight_name_1': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+          }
+```
+
+example：
+
+```python linenums="1"
+{
+    'conv10_expand_weights': {0.1: 0.006509952684312718, 0.2: 0.01827734339798862, 0.3: 0.014528405644659832, 0.6: 0.06536008804270439, 0.8: 0.11798612250664964, 0.7: 0.12391408417493704, 0.4: 0.030615754498018757, 0.5: 0.047105205602406594}
+    'conv10_linear_weights': {0.1: 0.05113190831455035, 0.2: 0.07705573833558801, 0.3: 0.12096721757739311, 0.6: 0.5135061352930738, 0.8: 0.7908166677143281, 0.7: 0.7272187676899062, 0.4: 0.1819252083008504, 0.5: 0.3728054727792405}
+}
+```
+
+The function would return a dict after loading the sensitivity file. The keys of the dict are name of parameters in each layer. And the value of key is the information about pruning sensitivity of corresponding layer. In example, pruning 10% filter of the layer corresponding to conv10_expand_weights would lead to 0.65% degradation of model performance. The details could be seen at: [Sensitivity analysis](https://github.com/PaddlePaddle/PaddleSlim/blob/release/2.0-alpha/docs/zh_cn/algo/algo.md)
+
+The function would return a dict after loading the sensitivity file. The keys of the dict are name of parameters in each layer. And the value of key is the information about pruning sensitivity of corresponding layer. In example, pruning 10% filter of the layer corresponding to conv10_expand_weights would lead to 0.65% degradation of model performance. The details could be seen at: [Sensitivity analysis](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/algo/algo.md#2-%E5%8D%B7%E7%A7%AF%E6%A0%B8%E5%89%AA%E8%A3%81%E5%8E%9F%E7%90%86)
+
+Enter the PaddleOCR root directory，perform sensitivity analysis on the model with the following command：
+
+```bash linenums="1"
+python3 deploy/slim/prune/sensitivity_anal.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model="your trained model"  Global.save_model_dir=./output/prune_model/
+```
+
+### 5. Export inference model and deploy it
+
+We can export the pruned model as inference_model for deployment:
+
+```bash linenums="1"
+python deploy/slim/prune/export_prune_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml  -o Global.pretrained_model=./output/det_db/best_accuracy  Global.save_inference_dir=./prune/prune_inference_model
+```
+
+Reference for prediction and deployment of inference model:
+
+1. [inference model python prediction](../../legacy/python_infer.en.md)
+2. [inference model C++ prediction](../../legacy/cpp_infer.en.md)
diff --git a/docs/version2.x/ppocr/model_compress/prune.md b/docs/version2.x/ppocr/model_compress/prune.md
new file mode 100644
index 0000000..6483713
--- /dev/null
+++ b/docs/version2.x/ppocr/model_compress/prune.md
@@ -0,0 +1,71 @@
+---
+comments: true
+---
+
+# PP-OCR模型裁剪
+
+复杂的模型有利于提高模型的性能，但也导致模型中存在一定冗余，模型裁剪通过移出网络模型中的子模型来减少这种冗余，达到减少模型计算复杂度，提高模型推理性能的目的。
+本教程将介绍如何使用飞桨模型压缩库PaddleSlim做PaddleOCR模型的压缩。
+[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim)集成了模型剪枝、量化（包括量化训练和离线量化）、蒸馏和神经网络搜索等多种业界常用且领先的模型压缩功能，如果您感兴趣，可以关注并了解。
+
+在开始本教程之前，建议先了解：
+
+1. [PaddleOCR模型的训练方法](../model_train/training.md)
+2. [模型裁剪教程](https://github.com/PaddlePaddle/PaddleSlim/blob/release%2F2.0.0/docs/zh_cn/tutorials/pruning/dygraph/filter_pruning.md)
+
+## 快速开始
+
+### 1.安装PaddleSlim
+
+```bash linenums="1"
+git clone https://github.com/PaddlePaddle/PaddleSlim.git
+cd PaddleSlim
+git checkout develop
+python3 setup.py install
+```
+
+### 2.获取预训练模型
+
+模型裁剪需要加载事先训练好的模型，PaddleOCR也提供了一系列[模型](../model_list.md)，开发者可根据需要自行选择模型或使用自己的模型。
+
+### 3.敏感度分析训练
+
+加载预训练模型后，通过对现有模型的每个网络层进行敏感度分析，得到敏感度文件：sen.pickle，可以通过PaddleSlim提供的[接口](https://github.com/PaddlePaddle/PaddleSlim/blob/9b01b195f0c4bc34a1ab434751cb260e13d64d9e/paddleslim/dygraph/prune/filter_pruner.py#L75)加载文件，获得各网络层在不同裁剪比例下的精度损失。从而了解各网络层冗余度，决定每个网络层的裁剪比例。
+敏感度文件内容格式：
+
+```python linenums="1"
+sen.pickle(Dict){
+            'layer_weight_name_0': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+            'layer_weight_name_1': sens_of_each_ratio(Dict){'pruning_ratio_0': acc_loss, 'pruning_ratio_1': acc_loss}
+        }
+```
+
+例子：
+
+```python linenums="1"
+{
+    'conv10_expand_weights': {0.1: 0.006509952684312718, 0.2: 0.01827734339798862, 0.3: 0.014528405644659832, 0.6: 0.06536008804270439, 0.8: 0.11798612250664964, 0.7: 0.12391408417493704, 0.4: 0.030615754498018757, 0.5: 0.047105205602406594}
+    'conv10_linear_weights': {0.1: 0.05113190831455035, 0.2: 0.07705573833558801, 0.3: 0.12096721757739311, 0.6: 0.5135061352930738, 0.8: 0.7908166677143281, 0.7: 0.7272187676899062, 0.4: 0.1819252083008504, 0.5: 0.3728054727792405}
+}
+```
+
+加载敏感度文件后会返回一个字典，字典中的keys为网络模型参数模型的名字，values为一个字典，里面保存了相应网络层的裁剪敏感度信息。例如在例子中，conv10_expand_weights所对应的网络层在裁掉10%的卷积核后模型性能相较原模型会下降0.65%，详细信息可见[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/algo/algo.md#2-%E5%8D%B7%E7%A7%AF%E6%A0%B8%E5%89%AA%E8%A3%81%E5%8E%9F%E7%90%86)
+
+进入PaddleOCR根目录，通过以下命令对模型进行敏感度分析训练：
+
+```bash linenums="1"
+python3 deploy/slim/prune/sensitivity_anal.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model="your trained model" Global.save_model_dir=./output/prune_model/
+```
+
+### 4.导出模型、预测部署
+
+在得到裁剪训练保存的模型后，我们可以将其导出为inference_model：
+
+```bash linenums="1"
+python3.7 deploy/slim/prune/export_prune_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model=./output/det_db/best_accuracy  Global.save_inference_dir=./prune/prune_inference_model
+```
+
+inference model的预测和部署参考：
+
+1. [inference model python端预测](../../legacy/python_infer.md)
+2. [inference model C++预测](../../legacy/cpp_infer.md)
diff --git a/docs/version2.x/ppocr/model_compress/quantization.en.md b/docs/version2.x/ppocr/model_compress/quantization.en.md
new file mode 100644
index 0000000..166aa38
--- /dev/null
+++ b/docs/version2.x/ppocr/model_compress/quantization.en.md
@@ -0,0 +1,71 @@
+---
+comments: true
+---
+
+# PP-OCR Models Quantization
+
+Generally, a more complex model would achieve better performance in the task, but it also leads to some redundancy in the model.
+Quantization is a technique that reduces this redundancy by reducing the full precision data to a fixed number,
+so as to reduce model calculation complexity and improve model inference performance.
+
+This example uses PaddleSlim provided [APIs of Quantization](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/api_cn/dygraph/quanter/qat.rst) to compress the OCR model.
+
+It is recommended that you could understand following pages before reading this example：
+
+- [The training strategy of OCR model](../model_train/training.en.md)
+- [PaddleSlim Document](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/docs/zh_cn/api_cn/dygraph/quanter/qat.rst)
+
+## Quick Start
+
+Quantization is mostly suitable for the deployment of lightweight models on mobile terminals.
+After training, if you want to further compress the model size and accelerate the prediction, you can use quantization methods to compress the model according to the following steps.
+
+1. Install PaddleSlim
+2. Prepare trained model
+3. Quantization-Aware Training
+4. Export inference model
+5. Deploy quantization inference model
+
+### 1. Install PaddleSlim
+
+```bash linenums="1"
+pip3 install paddleslim==2.3.2
+```
+
+### 2. Download Pre-trained Model
+
+PaddleOCR provides a series of pre-trained [models](../model_list.en.md).
+If the model to be quantified is not in the list, you need to follow the [Regular Training](../quick_start.en.md) method to get the trained model.
+
+### 3. Quant-Aware Training
+
+Quantization training includes offline quantization training and online quantization training.
+Online quantization training is more effective. It is necessary to load the pre-trained model.
+After the quantization strategy is defined, the model can be quantified.
+
+The code for quantization training is located in `slim/quantization/quant.py`. For example, the training instructions of slim PPOCRv3 detection model are as follows:
+
+```bash linenums="1"
+# download provided model
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
+tar xf ch_PP-OCRv3_det_distill_train.tar
+
+python deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Global.pretrained_model='./ch_PP-OCRv3_det_distill_train/best_accuracy'   Global.save_model_dir=./output/quant_model_distill/
+```
+
+If you want to quantify the text recognition model, you can modify the configuration file and loaded model parameters.
+
+### 4. Export inference model
+
+Once we got the model after pruning and fine-tuning, we can export it as an inference model for the deployment of predictive tasks:
+
+```bash linenums="1"
+python deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Global.checkpoints=output/quant_model/best_accuracy Global.save_inference_dir=./output/quant_inference_model
+```
+
+### 5. Deploy
+
+The numerical range of the quantized model parameters derived from the above steps is still FP32, but the numerical range of the parameters is int8.
+The derived model can be converted through the `opt tool` of PaddleLite.
+
+For quantitative model deployment, please refer to [Mobile terminal model deployment](../../../../deploy/lite/readme.md)
diff --git a/docs/version2.x/ppocr/model_compress/quantization.md b/docs/version2.x/ppocr/model_compress/quantization.md
new file mode 100644
index 0000000..3e7e6b8
--- /dev/null
+++ b/docs/version2.x/ppocr/model_compress/quantization.md
@@ -0,0 +1,67 @@
+---
+comments: true
+---
+
+# PP-OCR模型量化
+
+复杂的模型有利于提高模型的性能，但也导致模型中存在一定冗余，模型量化将全精度缩减到定点数减少这种冗余，达到减少模型计算复杂度，提高模型推理性能的目的。
+模型量化可以在基本不损失模型的精度的情况下，将FP32精度的模型参数转换为Int8精度，减小模型参数大小并加速计算，使用量化后的模型在移动端等部署时更具备速度优势。
+
+本教程将介绍如何使用飞桨模型压缩库PaddleSlim做PaddleOCR模型的压缩。
+[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim) 集成了模型剪枝、量化（包括量化训练和离线量化）、蒸馏和神经网络搜索等多种业界常用且领先的模型压缩功能，如果您感兴趣，可以关注并了解。
+
+在开始本教程之前，建议先了解[PaddleOCR模型的训练方法](../model_train/training.md)以及[PaddleSlim](https://github.com/PaddlePaddle/PaddleSlim/blob/develop/README.md)
+
+## 快速开始
+
+量化多适用于轻量模型在移动端的部署，当训练出一个模型后，如果希望进一步的压缩模型大小并加速预测，可使用量化的方法压缩模型。
+
+模型量化主要包括五个步骤：
+
+1. 安装 PaddleSlim
+2. 准备训练好的模型
+3. 量化训练
+4. 导出量化推理模型
+5. 量化模型预测部署
+
+### 1. 安装PaddleSlim
+
+```bash linenums="1"
+pip3 install paddleslim==2.3.2
+```
+
+### 2. 准备训练好的模型
+
+PaddleOCR提供了一系列训练好的[模型](../model_list.md)，如果待量化的模型不在列表中，需要按照[常规训练](../quick_start.md)方法得到训练好的模型。
+
+### 3. 量化训练
+
+量化训练包括离线量化训练和在线量化训练，在线量化训练效果更好，需加载预训练模型，在定义好量化策略后即可对模型进行量化。
+
+量化训练的代码位于slim/quantization/quant.py 中，比如训练检测模型，以PPOCRv3检测模型为例，训练指令如下：
+
+```bash linenums="1"
+# 下载检测预训练模型：
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
+tar xf ch_PP-OCRv3_det_distill_train.tar
+
+python deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_cml.yml -o Global.pretrained_model='./ch_PP-OCRv3_det_distill_train/best_accuracy'   Global.save_model_dir=./output/quant_model_distill/
+```
+
+如果要训练识别模型的量化，修改配置文件和加载的模型参数即可。
+
+### 4. 导出模型
+
+在得到量化训练保存的模型后，我们可以将其导出为inference_model，用于预测部署：
+
+```bash linenums="1"
+python deploy/slim/quantization/export_model.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml -o Global.checkpoints=output/quant_model/best_accuracy Global.save_inference_dir=./output/quant_inference_model
+```
+
+### 5. 量化模型部署
+
+上述步骤导出的量化模型，参数精度仍然是FP32，但是参数的数值范围是int8，导出的模型可以通过PaddleLite的opt模型转换工具完成模型转换。
+
+量化模型移动端部署的可参考 [移动端模型部署](../../../../deploy/lite/readme_ch.md)
+
+备注：量化训练后的模型参数是float32类型，转inference model预测时相对不量化无加速效果，原因是量化后模型结构之间存在量化和反量化算子，如果要使用量化模型部署，建议使用TensorRT并设置precision为INT8加速量化模型的预测时间。
diff --git a/docs/version2.x/ppocr/model_list.en.md b/docs/version2.x/ppocr/model_list.en.md
new file mode 100644
index 0000000..3e41ce4
--- /dev/null
+++ b/docs/version2.x/ppocr/model_list.en.md
@@ -0,0 +1,118 @@
+---
+comments: true
+---
+
+# OCR Model List（V3, updated on 2022.4.28）
+>
+> **Note**
+>
+> 1. Compared with model v2, the 3rd version of the detection model has an improvement in accuracy, and the 2.1 version of the recognition model has optimizations in accuracy and speed with CPU.
+> 2. Compared with [models 1.1](https://github.com/PaddlePaddle/PaddleOCR/blob/develop/doc/doc_en/models_list_en.md), which are trained with static graph programming paradigm, models 2.0 or higher are the dynamic graph trained version and achieve close performance.
+> 3. All models in this tutorial are from the PaddleOCR series, for more introduction to algorithms and models based on the public dataset, you can refer to [algorithm overview tutorial](../algorithm/overview.en.md).
+
+The downloadable models provided by PaddleOCR include the `inference model`, `trained model`, `pre-trained model` and `nb model`. The differences between the models are as follows:
+
+|model type|model format|description|
+|--- | --- | --- |
+|inference model|inference.pdmodel、inference.pdiparams|Used for inference based on Paddle inference engine，[detail](../legacy/python_infer.en.md)|
+|trained model, pre-trained model|\*.pdparams、\*.pdopt、\*.states |The checkpoints model saved in the training process, which stores the parameters of the model, is mostly used for model evaluation and continuous training.|
+|nb model|\*.nb| Model optimized by Paddle-Lite, which is suitable for mobile-side deployment scenarios (Paddle-Lite is needed for nb model deployment). |
+
+The relationship of the above models is as follows.
+
+![](./images/model_prod_flow_en.png)
+
+## 1. Text Detection Model
+
+### 1. Chinese Detection Model
+
+|model name|description|config|model size|download|
+| --- | --- | --- | --- | --- |
+| PP-OCRv4_mobile_det              | [New] Original lightweight model, supporting Chinese, English, multilingual text detection              | [PP-OCRv4_mobile_det.yml](../../../configs/det/PP-OCRv4/PP-OCRv4_mobile_det.yml)         | 4.70M        | [inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar) / [trained model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams)                |
+| PP-OCRv4_server_det        | [New] Original high-precision model, supporting Chinese, English, multilingual text detection      | [PP-OCRv4_server_det.yml](../../../configs/det/PP-OCRv4/PP-OCRv4_server_det.yml) | 110M         | [inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_det_infer.tar) / [trained model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_det_pretrained.pdparams)  |
+| PP-OCRv3_mobile_det        | Original lightweight model, supporting Chinese, English, multilingual text detection           | [PP-OCRv3_mobile_det.yml](../../../configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml)         | 1.1M         | [inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar) / [trained model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams) / [nb模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_slim_infer.nb) |
+| PP-OCRv3_server_det               |  Original high-precision model, supporting Chinese, English, multilingual text detection                           | [PP-OCRv3_server_det.yml](../../../configs/det/PP-OCRv3/PP-OCRv3_server_det.yml)         | 3.80M        | [inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_server_det_infer.tar) / [trained model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_server_det_pretrained.pdparams)  
+
+### 1.2 English Detection Model
+
+|model name|description|config|model size|download|
+| --- | --- | --- | --- | --- |
+|en_PP-OCRv3_det_slim | [New] Slim quantization with distillation lightweight detection model, supporting English | [PP-OCRv3_det_cml.yml](../../../configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml) | 1.1M |[inference model](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_slim_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_slim_distill_train.tar) / [nb model](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_slim_infer.nb) |
+|en_PP-OCRv3_det | [New] Original lightweight detection model, supporting English |[PP-OCRv3_det_cml.yml](../../../configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml)| 3.8M | [inference model](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_distill_train.tar) |
+
+* Note: English configuration file is the same as Chinese except for training data, here we only provide one configuration file.
+
+### 1.3 Multilingual Detection Model
+
+|model name|description|config|model size|download|
+| --- | --- | --- | --- | --- |
+| ml_PP-OCRv3_det_slim | [New] Slim quantization with distillation lightweight detection model, supporting English | [PP-OCRv3_det_cml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml) | 1.1M | [inference model](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_slim_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_slim_distill_train.tar) / [nb model](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_slim_infer.nb) |
+| ml_PP-OCRv3_det |[New] Original lightweight detection model, supporting English | [PP-OCRv3_det_cml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml)| 3.8M | [inference model](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_distill_train.tar) |
+
+* Note: English configuration file is the same as Chinese except for training data, here we only provide one configuration file.
+
+## 2. Text Recognition Model
+
+### 2.1 Chinese Recognition Model
+
+|model name|description|config|model size|download|
+| --- | --- | --- | --- | --- |
+|PP-OCRv3_mobile_rec_slim | [New] Slim quantization with distillation lightweight model, supporting Chinese, English text recognition |[PP-OCRv3_mobile_rec_distillation.yml](../../../configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec_distillation.yml)| 4.9M |[inference model](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_slim_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_slim_train.tar) / [nb model](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_slim_infer.nb) |
+|PP-OCRv3_mobile_rec| [New] Original lightweight model, supporting Chinese, English, multilingual text recognition |[PP-OCRv3_mobile_rec_distillation.yml](../../../configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec_distillation.yml)| 12.4M |[inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar) |
+|ch_PP-OCRv2_rec_slim| Slim quantization with distillation lightweight model, supporting Chinese, English text recognition|[ch_PP-OCRv2_rec.yml](../../../configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml)| 9.0M |[inference model](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_slim_quant_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_slim_quant_train.tar) |
+|ch_PP-OCRv2_rec| Original lightweight model, supporting Chinese, English, and multilingual text recognition |[ch_PP-OCRv2_rec_distillation.yml](../../../configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec_distillation.yml)|8.5M|[inference model](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_train.tar) |
+|ch_ppocr_mobile_slim_v2.0_rec|Slim pruned and quantized lightweight model, supporting Chinese, English and number recognition|[rec_chinese_lite_train_v2.0.yml](../../../configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml)| 6.0M | [inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_slim_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_slim_train.tar) |
+|ch_ppocr_mobile_v2.0_rec|Original lightweight model, supporting Chinese, English and number recognition|[rec_chinese_lite_train_v2.0.yml](../../../configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml)|5.2M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_train.tar) / [pre-trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_pre.tar) |
+|ch_ppocr_server_v2.0_rec|General model, supporting Chinese, English and number recognition|[rec_chinese_common_train_v2.0.yml](../../../configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0.yml)|94.8M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_train.tar) / [pre-trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_pre.tar) |
+
+**Note:** The `trained model` is fine-tuned on the `pre-trained model` with real data and synthesized vertical text data, which achieved better performance in the real scene. The `pre-trained model` is directly trained on the full amount of real data and synthesized data, which is more suitable for fine-tuning your dataset.
+
+### 2.2 English Recognition Model
+
+|model name|description|config|model size|download|
+| --- | --- | --- | --- | --- |
+|en_PP-OCRv3_mobile_rec_slim | [New] Slim quantization with distillation lightweight model, supporting English, English text recognition |[en_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml)| 3.2M |[inference model](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_slim_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_slim_train.tar) / [nb model](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_slim_infer.nb) |
+|en_PP-OCRv3_mobile_rec| [New] Original lightweight model, supporting English, English, multilingual text recognition |[en_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml)| 9.6M |[inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv3_mobile_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_train.tar) |
+|en_number_mobile_slim_v2.0_rec|Slim pruned and quantized lightweight model, supporting English and number recognition|[rec_en_number_lite_train.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/multi_language/rec_en_number_lite_train.yml)| 2.7M | [inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/en_number_mobile_v2.0_rec_slim_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/en_number_mobile_v2.0_rec_slim_train.tar) |
+|en_number_mobile_v2.0_rec|Original lightweight model, supporting English and number recognition|[rec_en_number_lite_train.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/multi_language/rec_en_number_lite_train.yml)|2.6M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/en_number_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/en_number_mobile_v2.0_rec_train.tar) |
+
+**Note:** Dictionary file of all English recognition models is `ppocr/utils/en_dict.txt`.
+
+### 2.3 Multilingual Recognition Model（Updating...）
+
+|model name| dict file | description|config|model size|download|
+| --- | --- | --- |--- | --- | --- |
+| korean_PP-OCRv3_mobile_rec | ppocr/utils/dict/korean_dict.txt |Lightweight model for Korean recognition|[korean_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/korean_PP-OCRv3_mobile_rec.yml)|11.0M|[inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/japan_PP-OCRv3_mobile_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/korean_PP-OCRv3_rec_train.tar) |
+| japan_PP-OCRv3_mobile_rec | ppocr/utils/dict/japan_dict.txt |Lightweight model for Japanese recognition|[japan_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/japan_PP-OCRv3_mobile_rec.yml)|11.0M|[inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/japan_PP-OCRv3_mobile_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/japan_PP-OCRv3_rec_train.tar) |
+| chinese_cht_PP-OCRv3_mobile_rec | ppocr/utils/dict/chinese_cht_dict.txt | Lightweight model for chinese cht|[chinese_cht_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/chinese_cht_PP-OCRv3_mobile_rec.yaml)|12.0M|[inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/chinese_cht_PP-OCRv3_mobile_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/chinese_cht_PP-OCRv3_rec_train.tar) |
+| te_PP-OCRv3_mobile_rec | ppocr/utils/dict/te_dict.txt | Lightweight model for Telugu recognition |[te_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/te_PP-OCRv3_mobile_rec.yml)|9.6M|[inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/te_PP-OCRv3_mobile_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/te_PP-OCRv3_rec_train.tar) |
+| ka_PP-OCRv3_mobile_rec | ppocr/utils/dict/ka_dict.txt | Lightweight model for Kannada recognition |[ka_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/ka_PP-OCRv3_mobile_rec.yml)|9.9M|[inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ka_PP-OCRv3_mobile_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/ka_PP-OCRv3_rec_train.tar) |
+| ta_PP-OCRv3_mobile_rec | ppocr/utils/dict/ta_dict.txt |Lightweight model for Tamil recognition|[ta_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/ta_PP-OCRv3_mobile_rec.yml)|9.6M|[inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ta_PP-OCRv3_mobile_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/ta_PP-OCRv3_rec_train.tar) |
+| latin_PP-OCRv3_mobile_rec |  ppocr/utils/dict/latin_dict.txt | Lightweight model for latin recognition |  [latin_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/latin_PP-OCRv3_mobile_rec.yml) |9.7M|[inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/latin_PP-OCRv3_mobile_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/latin_PP-OCRv3_rec_train.tar) |
+| arabic_PP-OCRv3_mobile_rec | ppocr/utils/dict/arabic_dict.txt | Lightweight model for arabic recognition  | [arabic_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/arabic_PP-OCRv3_mobile_rec.yml) |9.6M|[inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/arabic_PP-OCRv3_mobile_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/arabic_PP-OCRv3_rec_train.tar) |
+| cyrillic_PP-OCRv3_mobile_rec | ppocr/utils/dict/cyrillic_dict.txt | Lightweight model for cyrillic recognition  | [cyrillic_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/cyrillic_PP-OCRv3_mobile_rec.yml) |9.6M|[inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/cyrillic_PP-OCRv3_mobile_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/cyrillic_PP-OCRv3_rec_train.tar) |
+| devanagari_PP-OCRv3_mobile_rec | ppocr/utils/dict/devanagari_dict.txt | Lightweight model for devanagari recognition | [devanagari_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/devanagari_PP-OCRv3_mobile_rec.yml) |9.9M|[inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/devanagari_PP-OCRv3_mobile_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/devanagari_PP-OCRv3_rec_train.tar) |
+
+For a complete list of languages ​​and tutorials, please refer to [Multi-language model](./blog/multi_languages.en.md)
+
+## 3. Text Angle Classification Model
+
+|model name|description|config|model size|download|
+| --- | --- | --- | --- | --- |
+|ch_ppocr_mobile_slim_v2.0_cls|Slim quantized model for text angle classification|[cls_mv3.yml](../../../configs/cls/cls_mv3.yml)| 2.1M | [inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_slim_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_slim_train.tar) / [nb model](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_infer_opt.nb) |
+|ch_ppocr_mobile_v2.0_cls|Original model for text angle classification|[cls_mv3.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/cls/cls_mv3.yml)|1.38M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) |
+
+## 4. Paddle-Lite Model
+
+Paddle Lite is an updated version of Paddle-Mobile, an open-open source deep learning framework designed to make it easy to perform inference on mobile, embedded, and IoT devices. It can further optimize the inference model and generate the `nb model` used for edge devices. It's suggested to optimize the quantization model using Paddle-Lite because the `INT8` format is used for the model storage and inference.
+
+This chapter lists OCR nb models with PP-OCRv2 or earlier versions. You can access the latest nb models from the above tables.
+
+|Version|Introduction|Model size|Detection model|Text Direction model|Recognition model|Paddle-Lite branch|
+|---|---|---|---|---|---|---|
+|PP-OCRv2|extra-lightweight chinese OCR optimized model|11.0M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_infer_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_infer_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_infer_opt.nb)|v2.10|
+|PP-OCRv2(slim)|extra-lightweight chinese OCR optimized model|4.6M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_slim_opt.nb)|v2.10|
+|PP-OCRv2|extra-lightweight chinese OCR optimized model|11.0M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer_opt.nb)|v2.9|
+|PP-OCRv2(slim)|extra-lightweight chinese OCR optimized model|4.9M|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_slim_opt.nb)|v2.9|
+|V2.0|ppocr_v2.0 extra-lightweight chinese OCR optimized model|7.8M|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_det_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_rec_opt.nb)|v2.9|
+|V2.0(slim)|ppocr_v2.0 extra-lightweight chinese OCR optimized model|3.3M|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_det_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)|[download link](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_rec_slim_opt.nb)|v2.9|
diff --git a/docs/version2.x/ppocr/model_list.md b/docs/version2.x/ppocr/model_list.md
new file mode 100644
index 0000000..76b2f5e
--- /dev/null
+++ b/docs/version2.x/ppocr/model_list.md
@@ -0,0 +1,127 @@
+---
+comments: true
+---
+
+# PP-OCR系列模型列表（V4，2023年8月1日更新）
+
+> **说明**
+>
+> 1. V4版模型相比V3版模型，在模型精度上有进一步提升
+>
+> 2. V3版模型相比V2版模型，在模型精度上有进一步提升
+>
+> 3. 2.0+版模型和[1.1版模型](https://github.com/PaddlePaddle/PaddleOCR/blob/develop/doc/doc_ch/models_list.md) 的主要区别在于动态图训练vs.静态图训练，模型性能上无明显差距。
+>
+> 4. 本文档提供的是PPOCR自研模型列表，更多基于公开数据集的算法介绍与预训练模型可以参考：[算法概览文档](../algorithm/overview.md)。
+
+PaddleOCR提供的可下载模型包括`推理模型`、`训练模型`、`预训练模型`、`nb模型`，模型区别说明如下：
+
+| 模型类型  | 模型格式    | 简介  |
+| ---- | ----- | ----|
+| 推理模型             | inference.pdmodel、inference.pdiparams | 用于预测引擎推理，[详情](../legacy/python_infer.md)                   |
+| 训练模型、预训练模型 | \*.pdparams、\*.pdopt、\*.states       | 训练过程中保存的模型的参数、优化器状态和训练中间信息，多用于模型指标评估和恢复训练                   |
+| nb模型               | \*.nb        | 经过飞桨Paddle-Lite工具优化后的模型，适用于移动端/IoT端等端侧部署场景（需使用飞桨Paddle Lite部署）。 |
+
+各个模型的关系如下面的示意图所示。
+
+![](./images/model_prod_flow_ch.png)
+
+## 1. 文本检测模型
+
+### 1.1 中文检测模型
+
+| 模型名称     | 模型简介       | 配置文件    | 推理模型大小 | 下载地址   |
+| ------- | --------- | ------- | ------------ | ------- |
+| PP-OCRv4_mobile_det              | 【最新】原始超轻量模型，支持中英文、多语种文本检测                 | [PP-OCRv4_mobile_det.yml](../../../configs/det/PP-OCRv4/PP-OCRv4_mobile_det.yml)         | 4.70M        | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar) / [训练模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams)                |
+| PP-OCRv4_server_det        | 【最新】原始高精度模型，支持中英文、多语种文本检测                 | [PP-OCRv4_server_det.yml](../../../configs/det/PP-OCRv4/PP-OCRv4_server_det.yml) | 110M         | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_det_infer.tar) / [训练模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_det_pretrained.pdparams)  |
+
+| PP-OCRv3_mobile_det        | 原始超轻量模型，支持中英文、多语种文本检测              | [PP-OCRv3_mobile_det.yml](../../../configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml)         | 1.1M         | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar) / [训练模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams) / [nb模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_slim_infer.nb) |
+| PP-OCRv3_server_det               | 原始高精度模型，支持中英文、多语种文本检测                         | [PP-OCRv3_server_det.yml](../../../configs/det/PP-OCRv3/PP-OCRv3_server_det.yml)         | 3.80M        | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_server_det_infer.tar) / [训练模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_server_det_pretrained.pdparams)        |
+
+### 1.2 英文检测模型
+
+| 模型名称    | 模型简介   | 配置文件   | 推理模型大小 | 下载地址|
+| -------- | ----------- | --------- | ------------ | --------- |
+| en_PP-OCRv3_det_slim | 【最新】slim量化版超轻量模型，支持英文、数字检测 | [PP-OCRv3_det_cml.yml](../../../configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml) | 1.1M         | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_slim_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_slim_distill_train.tar) / [nb模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_slim_infer.nb) |
+| en_PP-OCRv3_det      | 【最新】原始超轻量模型，支持英文、数字检测       | [PP-OCRv3_det_cml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml) | 3.8M         | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_distill_train.tar)        |
+
+* 注：英文检测模型与中文检测模型结构完全相同，只有训练数据不同，在此仅提供相同的配置文件。
+
+### 1.3 多语言检测模型
+
+| 模型名称  | 模型简介   | 配置文件  | 推理模型大小 | 下载地址 |
+| ------ | --------| --------- | ------------ | --------- |
+| ml_PP-OCRv3_det_slim | 【最新】slim量化版超轻量模型，支持多语言检测 | [PP-OCRv3_det_cml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml) | 1.1M         | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_slim_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_slim_distill_train.tar) / [nb模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_slim_infer.nb) |
+| ml_PP-OCRv3_det      | 【最新】原始超轻量模型，支持多语言检测       | [PP-OCRv3_det_cml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml) | 3.8M         | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/Multilingual_PP-OCRv3_det_distill_train.tar)                       |
+
+* 注：多语言检测模型与中文检测模型结构完全相同，只有训练数据不同，在此仅提供相同的配置文件。
+
+## 2. 文本识别模型
+
+### 2.1 中文识别模型
+
+| 模型名称 | 模型简介 | 配置文件  | 推理模型大小 | 下载地址 |
+| ----- | --------- | ------ | ------------ | ------- |
+| PP-OCRv4_mobile_rec               | 【最新】超轻量模型，支持中英文、数字识别       | [PP-OCRv4_mobile_rec_distillation.yml](../../../configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec_distillation.yml)                  | 10M          | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv4/chinese/ch_PP-OCRv4_rec_train.tar)                    |
+| PP-OCRv4_server_rec        | 【最新】高精度模型，支持中英文、数字识别       | [PP-OCRv4_server_rec.yml](../../../configs/rec/PP-OCRv4/PP-OCRv4_server_rec.yml)                      | 88M          | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv4/chinese/ch_PP-OCRv4_rec_server_train.tar)      |
+| PP-OCRv4_server_rec_doc        | 【最新】高精度模型，支持中英文、数字识别，支持1.5万+字符和部分生僻字识别       | [PP-OCRv4_server_rec_doc.yml](../../../configs/rec/PP-OCRv4/PP-OCRv4_server_rec_doc.yml)                      | 75M          | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_doc_infer.tar) / [训练模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams)      |
+| PP-OCRv3_mobile_rec_slim          | slim量化版超轻量模型，支持中英文、数字识别     | [PP-OCRv3_mobile_rec_distillation.yml](../../../configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec_distillation.yml)        | 4.9M         | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_slim_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_slim_train.tar) / [nb模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_slim_infer.nb)             |
+| PP-OCRv3_mobile_rec               | 原始超轻量模型，支持中英文、数字识别           | [PP-OCRv3_mobile_rec_distillation.yml](../../../configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec_distillation.yml)        | 12.4M        | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar)                    |
+| ch_PP-OCRv2_rec_slim          | slim量化版超轻量模型，支持中英文、数字识别     | [ch_PP-OCRv2_rec.yml](../../../configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec.yml)     | 9.0M         | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_slim_quant_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_slim_quant_train.tar)                    |
+| ch_PP-OCRv2_rec               | 原始超轻量模型，支持中英文、数字识别           | [ch_PP-OCRv2_rec_distillation.yml](../../../configs/rec/ch_PP-OCRv2/ch_PP-OCRv2_rec_distillation.yml)     | 8.50M        | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_train.tar)                    |
+| ch_ppocr_mobile_slim_v2.0_rec | slim裁剪量化版超轻量模型，支持中英文、数字识别 | [rec_chinese_lite_train_v2.0.yml](../../../configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml)     | 6.0M         | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_slim_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_slim_train.tar)                |
+| ch_ppocr_mobile_v2.0_rec      | 原始超轻量模型，支持中英文、数字识别           | [rec_chinese_lite_train_v2.0.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml)     | 5.20M        | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_train.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_pre.tar) |
+| ch_ppocr_server_v2.0_rec      | 通用模型，支持中英文、数字识别                 | [rec_chinese_common_train_v2.0.yml](../../../configs/rec/ch_ppocr_v2.0/rec_chinese_common_train_v2.0.yml) | 94.8M        | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_train.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_pre.tar) |
+
+**说明：** `训练模型`是基于预训练模型在真实数据与竖排合成文本数据上finetune得到的模型，在真实应用场景中有着更好的表现，`预训练模型`则是直接基于全量真实数据与合成数据训练得到，更适合用于在自己的数据集上finetune。
+
+### 2.2 英文识别模型
+
+| 模型名称   | 模型简介 | 配置文件| 推理模型大小 | 下载地址|
+| ------- | -------- | ------ | ------------ | ----- |
+| en_PP-OCRv4_mobile_rec                | 【最新】原始超轻量模型，支持英文、数字识别   | [en_PP-OCRv4_mobile_rec.yml](../../../configs/rec/PP-OCRv4/en_PP-OCRv4_mobile_rec.yml)                         | 9.7M         | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv4_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv4/english/en_PP-OCRv4_rec_train.tar)        |
+| en_PP-OCRv3_mobile_rec_slim           | slim量化版超轻量模型，支持英文、数字识别     | [en_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml)                         | 3.2M         | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_slim_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_slim_train.tar) / [nb模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_slim_infer.nb) |
+| en_PP-OCRv3_mobile_rec                | 原始超轻量模型，支持英文、数字识别           | [en_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml)                         | 9.6M         | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv3_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_train.tar)        |
+| en_number_mobile_slim_v2.0_rec | slim裁剪量化版超轻量模型，支持英文、数字识别 | [rec_en_number_lite_train.yml](../../../configs/rec/multi_language/rec_en_number_lite_train.yml) | 2.7M         | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/en_number_mobile_v2.0_rec_slim_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/en_number_mobile_v2.0_rec_slim_train.tar)  |
+| en_number_mobile_v2.0_rec      | 原始超轻量模型，支持英文、数字识别           | [rec_en_number_lite_train.yml](../../../configs/rec/multi_language/rec_en_number_lite_train.yml) | 2.6M         | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/en_number_mobile_v2.0_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/multilingual/en_number_mobile_v2.0_rec_train.tar)  |
+
+**注意：** 所有英文识别模型的字典文件均为`ppocr/utils/en_dict.txt`
+
+### 2.3 多语言识别模型（更多语言持续更新中...）
+
+| 模型名称                 | 字典文件    | 模型简介     | 配置文件 | 推理模型大小 | 下载地址  |
+| ----- | ------ | ------------ | ------ | ------------ | ---- |
+| korean_PP-OCRv3_mobile_rec      | ppocr/utils/dict/korean_dict.txt      | 韩文识别     | [korean_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/korean_PP-OCRv3_mobile_rec.yml)           | 11.0M        | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/korean_PP-OCRv3_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/korean_PP-OCRv3_rec_train.tar)           |
+| japan_PP-OCRv3_mobile_rec       | ppocr/utils/dict/japan_dict.txt       | 日文识别     | [japan_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/japan_PP-OCRv3_mobile_rec.yml)             | 11.0M        | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/japan_PP-OCRv3_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/japan_PP-OCRv3_rec_train.tar)             |
+| chinese_cht_PP-OCRv3_mobile_rec | ppocr/utils/dict/chinese_cht_dict.txt | 中文繁体识别 | [chinese_cht_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/chinese_cht_PP-OCRv3_mobile_rec.yaml) | 12.0M        | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/chinese_cht_PP-OCRv3_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/chinese_cht_PP-OCRv3_rec_train.tar) |
+|  te_PP-OCRv3_mobile_rec         | ppocr/utils/dict/te_dict.txt          | 泰卢固文识别 | [te_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/te_PP-OCRv3_mobile_rec.yml)                   | 9.6M         | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/te_PP-OCRv3_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/te_PP-OCRv3_rec_train.tar)                   |
+| ka_PP-OCRv3_mobile_rec          | ppocr/utils/dict/ka_dict.txt          | 卡纳达文识别 | [ka_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/ka_PP-OCRv3_mobile_rec.yml)                   | 9.9M         | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ka_PP-OCRv3_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/ka_PP-OCRv3_rec_train.tar)                   |
+| ta_PP-OCRv3_mobile_rec          | ppocr/utils/dict/ta_dict.txt          | 泰米尔文识别 | [ta_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/ta_PP-OCRv3_mobile_rec.yml)                   | 9.6M         | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ta_PP-OCRv3_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/ta_PP-OCRv3_rec_train.tar)                   |
+| latin_PP-OCRv3_mobile_rec       | ppocr/utils/dict/latin_dict.txt       | 拉丁文识别   | [latin_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/latin_PP-OCRv3_mobile_rec.yml)             | 9.7M         | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/latin_PP-OCRv3_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/latin_PP-OCRv3_rec_train.tar)             |
+| arabic_PP-OCRv3_mobile_rec      | ppocr/utils/dict/arabic_dict.txt      | 阿拉伯字母   | [arabic_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/arabic_PP-OCRv3_mobile_rec.yml)           | 9.6M         | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/arabic_PP-OCRv3_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/arabic_PP-OCRv3_rec_train.tar)           |
+| cyrillic_PP-OCRv3_mobile_rec    | ppocr/utils/dict/cyrillic_dict.txt    | 斯拉夫字母   | [cyrillic_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/cyrillic_PP-OCRv3_mobile_rec.yml)       | 9.6M         | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/cyrillic_PP-OCRv3_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/cyrillic_PP-OCRv3_rec_train.tar)       |
+| devanagari_PP-OCRv3_mobile_rec  | ppocr/utils/dict/devanagari_dict.txt  | 梵文字母     | [devanagari_PP-OCRv3_mobile_rec.yml](../../../configs/rec/PP-OCRv3/multi_language/devanagari_PP-OCRv3_mobile_rec.yml)   | 9.9M         | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/devanagari_PP-OCRv3_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/multilingual/devanagari_PP-OCRv3_rec_train.tar)   |
+
+查看完整语种列表与使用教程请参考: [多语言模型](./blog/multi_languages.md)
+
+## 3. 文本方向分类模型
+
+| 模型名称                      | 模型简介 | 配置文件 | 推理模型大小 | 下载地址  |
+| ---------- | ------------- | ------------- | ------------ | -- |
+| ch_ppocr_mobile_slim_v2.0_cls | slim量化版模型，对检测到的文本行文字角度分类 | [cls_mv3.yml](../../../configs/cls/cls_mv3.yml) | 2.1M         | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_slim_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_slim_infer.tar) / [nb模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_infer_opt.nb) |
+| ch_ppocr_mobile_v2.0_cls      | 原始分类器模型，对检测到的文本行文字角度分类 | [cls_mv3.yml](../../../../configs/cls/cls_mv3.yml) | 1.38M        | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar)             |
+
+## 4. Paddle-Lite 模型
+
+Paddle-Lite 是一个高性能、轻量级、灵活性强且易于扩展的深度学习推理框架，它可以对inference模型进一步优化，得到适用于移动端/IoT端等端侧部署场景的`nb模型`。一般建议基于量化模型进行转换，因为可以将模型以INT8形式进行存储与推理，从而进一步减小模型大小，提升模型速度。
+
+本节主要列出PP-OCRv2以及更早版本的检测与识别nb模型，最新版本的nb模型可以直接从上面的模型列表中获得。
+
+| 模型版本       | 模型简介| 模型大小 | 检测模型                   | 文本方向分类模型           | 识别模型                   | Paddle-Lite版本 |
+| ----- | --- | ---- | --- | --- | --- | --- |
+| PP-OCRv2       | 蒸馏版超轻量中文OCR移动端模型     | 11.0M    | [下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_infer_opt.nb)             | [下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_infer_opt.nb)    | [下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_infer_opt.nb)             | v2.10           |
+| PP-OCRv2(slim) | 蒸馏版超轻量中文OCR移动端模型     | 4.6M     | [下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_det_slim_opt.nb)              | [下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb)     | [下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/lite/ch_PP-OCRv2_rec_slim_opt.nb)              | v2.10           |
+| PP-OCRv2       | 蒸馏版超轻量中文OCR移动端模型     | 11.0M    | [下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer_opt.nb)          | [下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_opt.nb)      | [下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer_opt.nb)          | v2.9            |
+| PP-OCRv2(slim) | 蒸馏版超轻量中文OCR移动端模型     | 4.9M     | [下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_slim_opt.nb)           | [下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb) | [下载地址](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_slim_opt.nb)           | v2.9            |
+| V2.0           | ppocr_v2.0超轻量中文OCR移动端模型 | 7.8M     | [下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_det_opt.nb)      | [下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_opt.nb)      | [下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_rec_opt.nb)      | v2.9            |
+| V2.0(slim)     | ppocr_v2.0超轻量中文OCR移动端模型 | 3.3M     | [下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_det_slim_opt.nb) | [下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_cls_slim_opt.nb) | [下载地址](https://paddleocr.bj.bcebos.com/dygraph_v2.0/lite/ch_ppocr_mobile_v2.0_rec_slim_opt.nb) | v2.9            |
diff --git a/docs/version2.x/ppocr/model_train/PPOCRv3_det_train.en.md b/docs/version2.x/ppocr/model_train/PPOCRv3_det_train.en.md
new file mode 100644
index 0000000..616d3a7
--- /dev/null
+++ b/docs/version2.x/ppocr/model_train/PPOCRv3_det_train.en.md
@@ -0,0 +1,254 @@
+---
+comments: true
+---
+
+# PP-OCRv3 text detection model training
+
+## 1. Introduction
+
+PP-OCRv3 is a further upgrade of PP-OCRv2. This section introduces the training steps of the PP-OCRv3 detection model. For an introduction to the PP-OCRv3 strategy, refer to [document](../blog/PP-OCRv3_introduction.md).
+
+## 2. Detection training
+
+The PP-OCRv3 detection model is an upgrade of the [CML](https://arxiv.org/pdf/2109.03144.pdf) (Collaborative Mutual Learning) collaborative mutual learning text detection distillation strategy in PP-OCRv2. PP-OCRv3 further optimizes the detection teacher model and student model. Among them, when optimizing the teacher model, the PAN structure LK-PAN with a large receptive field and the DML (Deep Mutual Learning) distillation strategy are proposed; when optimizing the student model, the FPN structure RSE-FPN with a residual attention mechanism is proposed.
+
+PP-OCRv3 detection training includes two steps:
+
+- Step 1: Use DML distillation method to train detection teacher model
+
+- Step 2: Use the teacher model obtained in step 1 to train a lightweight student model using CML method
+
+### 2.1 Prepare data and operating environment
+
+The training data uses icdar2015 data. For the steps of preparing the training set, refer to [ocr_dataset](../../../datasets/ocr_datasets.md).
+
+For the preparation of the operating environment, refer to [document](../installation.en.md).
+
+### 2.2 Train the teacher model
+
+The configuration file for teacher model training is [PP-OCRv3_det_dml.yml](../../../../configs/det/PP-OCRv3/PP-OCRv3_det_dml.yml). The Backbone, Neck, and Head of the teacher model structure are Resnet50, LKPAN, and DBHead respectively, and are trained using the DML distillation method. For a detailed introduction to the configuration file, refer to [Document](../model_compress/knowledge_distillation.en.md).
+
+Download ImageNet pre-trained model:
+
+```bash linenums="1"
+# Download ResNet50_vd pre-trained model
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/ResNet50_vd_ssld_pretrained.pdparams
+```
+
+**Start training**
+
+```bash linenums="1"
+# Single card training
+python3 tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_dml.yml \
+-o Architecture.Models.Student.pretrained=./pretrain_models/ResNet50_vd_ssld_pretrained \
+Architecture.Models.Student2.pretrained=./pretrain_models/ResNet50_vd_ssld_pretrained \
+Global.save_model_dir=./output/
+# If you want to use multi-GPU distributed training, please use the following command:
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_dml.yml \
+-o Architecture.Models.Student.pretrained=./pretrain_models/ResNet50_vd_ssld_pretrained \
+Architecture.Models.Student2.pretrained=./pretrain_models/ResNet50_vd_ssld_pretrained \
+Global.save_model_dir=./output/
+```
+
+The model saved during training is in the output directory, which contains the following files:
+
+```bash linenums="1"
+best_accuracy.states
+best_accuracy.pdparams # The model parameters with the best accuracy are saved by default
+best_accuracy.pdopt # The optimizer-related parameters with the best accuracy are saved by default
+latest.states
+latest.pdparams # The latest model parameters saved by default
+latest.pdopt # The optimizer-related parameters of the latest model saved by default
+```
+
+Among them, best_accuracy is the model parameter with the highest accuracy saved, and the model can be directly used for evaluation.
+
+The model evaluation command is as follows:
+
+```bash linenums="1"
+python3 tools/eval.py -c configs/det/PP-OCRv3/PP-OCRv3_det_dml.yml -o Global.checkpoints=./output/best_accuracy
+```
+
+The trained teacher model has a larger structure and higher accuracy, which is used to improve the accuracy of the student model.
+
+**Extract teacher model parameters**
+best_accuracy contains the parameters of two models, corresponding to Student and Student2 in the configuration file. The method to extract the parameters of Student is as follows:
+
+```bash linenums="1"
+import paddle
+# Load pre-trained model
+all_params = paddle.load("output/best_accuracy.pdparams")
+# View the keys of weight parameters
+print(all_params.keys())
+# Model weight extraction
+s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
+# View the keys of model weight parameters
+print(s_params.keys())
+# Save
+paddle.save(s_params, "./pretrain_models/dml_teacher.pdparams")
+```
+
+The extracted model parameters can be used for further fine-tuning or distillation training of the model.
+
+### 2.3 Training the student model
+
+The configuration file for training the student model is [PP-OCRv3_det_cml.yml](../../../../configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml)
+The teacher model trained in the previous section is used as supervision, and the CML method is used to train a lightweight student model.
+
+Download the ImageNet pre-trained model of the student model:
+
+```bash linenums="1"
+# Download the pre-trained model of MobileNetV3
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/MobileNetV3_large_x0_5_pretrained.pdparams
+```
+
+**Start training**
+
+```bash linenums="1"
+# Single card training
+python3 tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml \
+-o Architecture.Models.Student.pretrained=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+Architecture.Models.Student2.pretrained=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+Architecture.Models.Teacher.pretrained=./pretrain_models/dml_teacher \
+Global.save_model_dir=./output/
+# If you want to use multi-GPU distributed training, please use the following command:
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml \
+-o Architecture.Models.Student.pretrained=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+Architecture.Models.Student2.pretrained=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+Architecture.Models.Teacher.pretrained=./pretrain_models/dml_teacher \
+Global.save_model_dir=./output/
+```
+
+The model saved during the training process is in the output directory.
+The model evaluation command is as follows:
+
+```bash linenums="1"
+python3 tools/eval.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml -o Global.checkpoints=./output/best_accuracy
+```
+
+best_accuracy contains the parameters of three models, corresponding to Student, Student2, and Teacher in the configuration file. The method to extract Student parameters is as follows:
+
+```bash linenums="1"
+import paddle
+# Load pre-trained model
+all_params = paddle.load("output/best_accuracy.pdparams")
+# View the keys of weight parameters
+print(all_params.keys())
+# Model weight extraction
+s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
+# View the keys of model weight parameters
+print(s_params.keys())
+# Save
+paddle.save(s_params, "./pretrain_models/cml_student.pdparams")
+```
+
+The extracted Student parameters can be used for model deployment or further fine-tuning training.
+
+## 3. Fine-tune training based on PP-OCRv3 detection
+
+This section describes how to use the PP-OCRv3 detection model for fine-tune training in other scenarios.
+
+Fine-tune training is applicable to three scenarios:
+
+- Fine-tune training based on the CML distillation method is applicable to scenarios where the teacher model has higher accuracy than the PP-OCRv3 detection model in the usage scenario and a lightweight detection model is desired.
+
+- Fine-tune training based on the PP-OCRv3 lightweight detection model does not require the training of the teacher model and is intended to improve the accuracy of the usage scenario based on the PP-OCRv3 detection model.
+
+- Fine-tune training based on the DML distillation method is applicable to scenarios where the DML method is used to further improve accuracy.
+
+**Finetune training based on CML distillation method**
+
+Download PP-OCRv3 training model:
+
+```bash linenums="1"
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
+tar xf ch_PP-OCRv3_det_distill_train.tar
+```
+
+ch_PP-OCRv3_det_distill_train/best_accuracy.pdparams contains the parameters of Student, Student2, and Teacher models in the CML configuration file.
+
+Start training:
+
+```bash linenums="1"
+# Single card training
+python3 tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml \
+-o Global.pretrained_model=./ch_PP-OCRv3_det_distill_train/best_accuracy \
+Global.save_model_dir=./output/
+# If you want to use multi-GPU distributed training, please use the following command:
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml \
+-o Global.pretrained_model=./ch_PP-OCRv3_det_distill_train/best_accuracy \
+Global.save_model_dir=./output/
+```
+
+**Finetune training based on PP-OCRv3 lightweight detection model**
+
+Download PP-OCRv3 training model and extract model parameters of Student structure:
+
+```
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
+tar xf ch_PP-OCRv3_det_distill_train.tar
+```
+
+The method to extract Student parameters is as follows:
+
+```bash linenums="1"
+import paddle
+# Load pre-trained model
+all_params = paddle.load("output/best_accuracy.pdparams")
+# View the keys of weight parameters
+print(all_params.keys())
+# Model weight extraction
+s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
+# View the keys of the model weight parameters
+print(s_params.keys())
+# Save
+paddle.save(s_params, "./student.pdparams")
+```
+
+Train using the configuration file [PP-OCRv3_mobile_det.yml](../../../../configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml).
+
+**Start training**
+
+```bash linenums="1"
+# Single card training
+python3 tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml \
+-o Global.pretrained_model=./student \
+Global.save_model_dir=./output/
+# If you want to use multi-GPU distributed training, please use the following command:
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml \
+-o Global.pretrained_model=./student \
+Global.save_model_dir=./output/
+```
+
+**Finetune training based on DML distillation method**
+
+Take the Teacher model in ch_PP-OCRv3_det_distill_train as an example. First, extract the parameters of the Teacher structure. The method is as follows:
+
+```bash linenums="1"
+import paddle
+# Load pre-trained model
+all_params = paddle.load("ch_PP-OCRv3_det_distill_train/best_accuracy.pdparams")
+# View the keys of weight parameters
+print(all_params.keys())
+# Model weight extraction
+s_params = {key[len("Teacher."):]: all_params[key] for key in all_params if "Teacher." in key}
+# View the keys of model weight parameters
+print(s_params.keys())
+# Save
+paddle.save(s_params, "./teacher.pdparams")
+```
+
+**Start training**
+
+```bash linenums="1"
+# Single card training
+python3 tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_dml.yml \
+-o Architecture.Models.Student.pretrained=./teacher \
+Architecture.Models.Student2.pretrained=./teacher \
+Global.save_model_dir=./output/
+# If you want to use multi-GPU distributed training, please use the following command:
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_dml.yml \
+-o Architecture.Models.Student.pretrained=./teacher \
+Architecture.Models.Student2.pretrained=./teacher \
+Global.save_model_dir=./output/
+```
diff --git a/docs/version2.x/ppocr/model_train/PPOCRv3_det_train.md b/docs/version2.x/ppocr/model_train/PPOCRv3_det_train.md
new file mode 100644
index 0000000..a85d971
--- /dev/null
+++ b/docs/version2.x/ppocr/model_train/PPOCRv3_det_train.md
@@ -0,0 +1,251 @@
+---
+comments: true
+---
+
+# PP-OCRv3 文本检测模型训练
+
+## 1. 简介
+
+PP-OCRv3在PP-OCRv2的基础上进一步升级。本节介绍PP-OCRv3检测模型的训练步骤。有关PP-OCRv3策略介绍参考[文档](../blog/PP-OCRv3_introduction.md)。
+
+## 2. 检测训练
+
+PP-OCRv3检测模型是对PP-OCRv2中的[CML](https://arxiv.org/pdf/2109.03144.pdf)（Collaborative Mutual Learning) 协同互学习文本检测蒸馏策略进行了升级。PP-OCRv3分别针对检测教师模型和学生模型进行进一步效果优化。其中，在对教师模型优化时，提出了大感受野的PAN结构LK-PAN和引入了DML（Deep Mutual Learning）蒸馏策略；在对学生模型优化时，提出了残差注意力机制的FPN结构RSE-FPN。
+
+PP-OCRv3检测训练包括两个步骤：
+
+- 步骤1：采用DML蒸馏方法训练检测教师模型
+- 步骤2：使用步骤1得到的教师模型采用CML方法训练出轻量学生模型
+
+### 2.1 准备数据和运行环境
+
+训练数据采用icdar2015数据，准备训练集步骤参考[ocr_dataset](../../../datasets/ocr_datasets.md).
+
+运行环境准备参考[文档](../installation.md)。
+
+### 2.2 训练教师模型
+
+教师模型训练的配置文件是[PP-OCRv3_det_dml.yml](../../../../configs/det/PP-OCRv3/PP-OCRv3_det_dml.yml)。教师模型模型结构的Backbone、Neck、Head分别为Resnet50, LKPAN, DBHead，采用DML的蒸馏方法训练。有关配置文件的详细介绍参考[文档](../model_compress/knowledge_distillation.md)。
+
+下载ImageNet预训练模型：
+
+```bash linenums="1"
+# 下载ResNet50_vd的预训练模型
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/ResNet50_vd_ssld_pretrained.pdparams
+```
+
+**启动训练**
+
+```bash linenums="1"
+# 单卡训练
+python3 tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_dml.yml \
+    -o Architecture.Models.Student.pretrained=./pretrain_models/ResNet50_vd_ssld_pretrained \
+       Architecture.Models.Student2.pretrained=./pretrain_models/ResNet50_vd_ssld_pretrained \
+       Global.save_model_dir=./output/
+# 如果要使用多GPU分布式训练，请使用如下命令：
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_dml.yml \
+    -o Architecture.Models.Student.pretrained=./pretrain_models/ResNet50_vd_ssld_pretrained \
+       Architecture.Models.Student2.pretrained=./pretrain_models/ResNet50_vd_ssld_pretrained \
+       Global.save_model_dir=./output/
+```
+
+训练过程中保存的模型在output目录下，包含以下文件：
+
+```bash linenums="1"
+best_accuracy.states
+best_accuracy.pdparams  # 默认保存最优精度的模型参数
+best_accuracy.pdopt     # 默认保存最优精度的优化器相关参数
+latest.states
+latest.pdparams  # 默认保存的最新模型参数
+latest.pdopt     # 默认保存的最新模型的优化器相关参数
+```
+
+其中，best_accuracy是保存的精度最高的模型参数，可以直接使用该模型评估。
+
+模型评估命令如下：
+
+```bash linenums="1"
+python3 tools/eval.py -c configs/det/PP-OCRv3/PP-OCRv3_det_dml.yml -o Global.checkpoints=./output/best_accuracy
+```
+
+训练的教师模型结构更大，精度更高，用于提升学生模型的精度。
+
+**提取教师模型参数**
+best_accuracy包含两个模型的参数，分别对应配置文件中的Student，Student2。提取Student的参数方法如下：
+
+```bash linenums="1"
+import paddle
+# 加载预训练模型
+all_params = paddle.load("output/best_accuracy.pdparams")
+# 查看权重参数的keys
+print(all_params.keys())
+# 模型的权重提取
+s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
+# 查看模型权重参数的keys
+print(s_params.keys())
+# 保存
+paddle.save(s_params, "./pretrain_models/dml_teacher.pdparams")
+```
+
+提取出来的模型参数可以用于模型进一步的finetune训练或者蒸馏训练。
+
+### 2.3 训练学生模型
+
+训练学生模型的配置文件是[PP-OCRv3_det_cml.yml](../../../../configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml)
+上一节训练得到的教师模型作为监督，采用CML方式训练得到轻量的学生模型。
+
+下载学生模型的ImageNet预训练模型：
+
+```bash linenums="1"
+# 下载MobileNetV3的预训练模型
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/MobileNetV3_large_x0_5_pretrained.pdparams
+```
+
+**启动训练**
+
+```bash linenums="1"
+# 单卡训练
+python3 tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml \
+    -o Architecture.Models.Student.pretrained=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+       Architecture.Models.Student2.pretrained=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+       Architecture.Models.Teacher.pretrained=./pretrain_models/dml_teacher \
+       Global.save_model_dir=./output/
+# 如果要使用多GPU分布式训练，请使用如下命令：
+python3  -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml \
+    -o Architecture.Models.Student.pretrained=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+       Architecture.Models.Student2.pretrained=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+       Architecture.Models.Teacher.pretrained=./pretrain_models/dml_teacher \
+       Global.save_model_dir=./output/
+```
+
+训练过程中保存的模型在output目录下，
+模型评估命令如下：
+
+```bash linenums="1"
+python3 tools/eval.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml -o Global.checkpoints=./output/best_accuracy
+```
+
+best_accuracy包含三个模型的参数，分别对应配置文件中的Student，Student2，Teacher。提取Student参数的方法如下：
+
+```bash linenums="1"
+import paddle
+# 加载预训练模型
+all_params = paddle.load("output/best_accuracy.pdparams")
+# 查看权重参数的keys
+print(all_params.keys())
+# 模型的权重提取
+s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
+# 查看模型权重参数的keys
+print(s_params.keys())
+# 保存
+paddle.save(s_params, "./pretrain_models/cml_student.pdparams")
+```
+
+提取出来的Student的参数可用于模型部署或者做进一步的finetune训练。
+
+## 3. 基于PP-OCRv3检测finetune训练
+
+本节介绍如何使用PP-OCRv3检测模型在其他场景上的finetune训练。
+
+finetune训练适用于三种场景：
+
+- 基于CML蒸馏方法的finetune训练，适用于教师模型在使用场景上精度高于PP-OCRv3检测模型，且希望得到一个轻量检测模型。
+- 基于PP-OCRv3轻量检测模型的finetune训练，无需训练教师模型，希望在PP-OCRv3检测模型基础上提升使用场景上的精度。
+- 基于DML蒸馏方法的finetune训练，适用于采用DML方法进一步提升精度的场景。
+
+**基于CML蒸馏方法的finetune训练**
+
+下载PP-OCRv3训练模型：
+
+```bash linenums="1"
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
+tar xf ch_PP-OCRv3_det_distill_train.tar
+```
+
+ch_PP-OCRv3_det_distill_train/best_accuracy.pdparams包含CML配置文件中Student、Student2、Teacher模型的参数。
+
+启动训练：
+
+```bash linenums="1"
+# 单卡训练
+python3 tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml \
+    -o Global.pretrained_model=./ch_PP-OCRv3_det_distill_train/best_accuracy \
+       Global.save_model_dir=./output/
+# 如果要使用多GPU分布式训练，请使用如下命令：
+python3  -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml \
+    -o Global.pretrained_model=./ch_PP-OCRv3_det_distill_train/best_accuracy  \
+       Global.save_model_dir=./output/
+```
+
+**基于PP-OCRv3轻量检测模型的finetune训练**
+
+下载PP-OCRv3训练模型，并提取Student结构的模型参数：
+
+```
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar
+tar xf ch_PP-OCRv3_det_distill_train.tar
+```
+
+提取Student参数的方法如下：
+
+```bash linenums="1"
+import paddle
+# 加载预训练模型
+all_params = paddle.load("output/best_accuracy.pdparams")
+# 查看权重参数的keys
+print(all_params.keys())
+# 模型的权重提取
+s_params = {key[len("Student."):]: all_params[key] for key in all_params if "Student." in key}
+# 查看模型权重参数的keys
+print(s_params.keys())
+# 保存
+paddle.save(s_params, "./student.pdparams")
+```
+
+使用配置文件[PP-OCRv3_mobile_det.yml](../../../../configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml)训练。
+
+**启动训练**
+
+```bash linenums="1"
+# 单卡训练
+python3 tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml \
+    -o Global.pretrained_model=./student \
+       Global.save_model_dir=./output/
+# 如果要使用多GPU分布式训练，请使用如下命令：
+python3  -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml \
+    -o Global.pretrained_model=./student \
+       Global.save_model_dir=./output/
+```
+
+**基于DML蒸馏方法的finetune训练**
+
+以ch_PP-OCRv3_det_distill_train中的Teacher模型为例，首先提取Teacher结构的参数，方法如下：
+
+```bash linenums="1"
+import paddle
+# 加载预训练模型
+all_params = paddle.load("ch_PP-OCRv3_det_distill_train/best_accuracy.pdparams")
+# 查看权重参数的keys
+print(all_params.keys())
+# 模型的权重提取
+s_params = {key[len("Teacher."):]: all_params[key] for key in all_params if "Teacher." in key}
+# 查看模型权重参数的keys
+print(s_params.keys())
+# 保存
+paddle.save(s_params, "./teacher.pdparams")
+```
+
+**启动训练**
+
+```bash linenums="1"
+# 单卡训练
+python3 tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_dml.yml \
+    -o Architecture.Models.Student.pretrained=./teacher \
+       Architecture.Models.Student2.pretrained=./teacher \
+       Global.save_model_dir=./output/
+# 如果要使用多GPU分布式训练，请使用如下命令：
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_dml.yml \
+    -o Architecture.Models.Student.pretrained=./teacher \
+       Architecture.Models.Student2.pretrained=./teacher \
+       Global.save_model_dir=./output/
+```
diff --git a/docs/version2.x/ppocr/model_train/angle_class.en.md b/docs/version2.x/ppocr/model_train/angle_class.en.md
new file mode 100644
index 0000000..b6ea146
--- /dev/null
+++ b/docs/version2.x/ppocr/model_train/angle_class.en.md
@@ -0,0 +1,149 @@
+---
+comments: true
+---
+
+# Text Direction Classification
+
+## 1. Method Introduction
+
+The angle classification is used in the scene where the image is not 0 degrees. In this scene, it is necessary to perform a correction operation on the text line detected in the picture. In the PaddleOCR system,
+The text line image obtained after text detection is sent to the recognition model after affine transformation. At this time, only a 0 and 180 degree angle classification of the text is required, so the built-in PaddleOCR text angle classifier **only supports 0 and 180 degree classification**. If you want to support more angles, you can modify the algorithm yourself to support.
+
+Example of 0 and 180 degree data samples：
+
+![img](./images/angle_class_example.jpg)
+
+## 2. Data Preparation
+
+Please organize the dataset as follows:
+
+The default storage path for training data is `PaddleOCR/train_data/cls`, if you already have a dataset on your disk, just create a soft link to the dataset directory:
+
+```bash linenums="1"
+ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/cls/dataset
+```
+
+please refer to the following to organize your data.
+
+### Training set
+
+First put the training images in the same folder (train_images), and use a txt file (cls_gt_train.txt) to store the image path and label.
+
+- Note: by default, the image path and image label are split with `\t`, if you use other methods to split, it will cause training error
+
+0 and 180 indicate that the angle of the image is 0 degrees and 180 degrees, respectively.
+
+```text linenums="1"
+" Image file name           Image annotation "
+
+train/word_001.jpg   0
+train/word_002.jpg   180
+```
+
+The final training set should have the following file structure:
+
+```text linenums="1"
+|-train_data
+    |-cls
+        |- cls_gt_train.txt
+        |- train
+            |- word_001.png
+            |- word_002.jpg
+            |- word_003.jpg
+            | ...
+```
+
+### Test set
+
+Similar to the training set, the test set also needs to be provided a folder
+containing all images (test) and a cls_gt_test.txt. The structure of the test set is as follows:
+
+```text linenums="1"
+|-train_data
+    |-cls
+        |- cls_gt_test.txt
+        |- test
+            |- word_001.jpg
+            |- word_002.jpg
+            |- word_003.jpg
+            | ...
+```
+
+## 3. Training
+
+Write the prepared txt file and image folder path into the configuration file under the `Train/Eval.dataset.label_file_list` and `Train/Eval.dataset.data_dir` fields, the absolute path of the image consists of the `Train/Eval.dataset.data_dir` field and the image name recorded in the txt file.
+
+PaddleOCR provides training scripts, evaluation scripts, and prediction scripts.
+
+### Start training
+
+```bash linenums="1"
+# Set PYTHONPATH path
+export PYTHONPATH=$PYTHONPATH:.
+# GPU training Support single card and multi-card training, specify the card number through --gpus.
+# Start training, the following command has been written into the train.sh file, just modify the configuration file path in the file
+python3 -m paddle.distributed.launch --gpus '0,1,2,3,4,5,6,7'  tools/train.py -c configs/cls/cls_mv3.yml
+```
+
+### Data Augmentation
+
+PaddleOCR provides a variety of data augmentation methods. If you want to add disturbance during training, Please uncomment the `RecAug` and `RandAugment` fields under `Train.dataset.transforms` in the configuration file.
+
+The default perturbation methods are: cvtColor, blur, jitter, Gauss noise, random crop, perspective, color reverse, RandAugment.
+
+Except for RandAugment, each disturbance method is selected with a 50% probability during the training process. For specific code implementation, please refer to:
+[rec_img_aug.py](../../../../ppocr/data/imaug/rec_img_aug.py)
+[randaugment.py](../../../../ppocr/data/imaug/randaugment.py)
+
+### Training
+
+PaddleOCR supports alternating training and evaluation. You can modify `eval_batch_step` in `configs/cls/cls_mv3.yml` to set the evaluation frequency. By default, it is evaluated every 1000 iter. The following content will be saved during training:
+
+```bash linenums="1"
+├── best_accuracy.pdopt # Optimizer parameters for the best model
+├── best_accuracy.pdparams # Parameters of the best model
+├── best_accuracy.states # Metric info and epochs of the best model
+├── config.yml # Configuration file for this experiment
+├── latest.pdopt # Optimizer parameters for the latest model
+├── latest.pdparams # Parameters of the latest model
+├── latest.states # Metric info and epochs of the latest model
+└── train.log # Training log
+```
+
+If the evaluation set is large, the test will be time-consuming. It is recommended to reduce the number of evaluations, or evaluate after training.
+
+**Note that the configuration file for prediction/evaluation must be consistent with the training.**
+
+## 4. Evaluation
+
+The evaluation dataset can be set by modifying the `Eval.dataset.label_file_list` field in the `configs/cls/cls_mv3.yml` file.
+
+```bash linenums="1"
+export CUDA_VISIBLE_DEVICES=0
+# GPU evaluation, Global.checkpoints is the weight to be tested
+python3 tools/eval.py -c configs/cls/cls_mv3.yml -o Global.checkpoints={path/to/weights}/best_accuracy
+```
+
+## 5. Prediction
+
+### Training engine prediction
+
+Using the model trained by paddleocr, you can quickly get prediction through the following script.
+
+Use `Global.infer_img` to specify the path of the predicted picture or folder, and use `Global.checkpoints` to specify the weight:
+
+```bash linenums="1"
+# Predict English results
+python3 tools/infer_cls.py -c configs/cls/cls_mv3.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.load_static_weights=false Global.infer_img=doc/imgs_words_en/word_10.png
+```
+
+Input image:
+
+![](./images/word_1-20240704092713071.jpg)
+
+Get the prediction result of the input image:
+
+```bash linenums="1"
+infer_img: doc/imgs_words_en/word_10.png
+     result: ('0', 0.9999995)
+```
diff --git a/docs/version2.x/ppocr/model_train/angle_class.md b/docs/version2.x/ppocr/model_train/angle_class.md
new file mode 100644
index 0000000..a6db43f
--- /dev/null
+++ b/docs/version2.x/ppocr/model_train/angle_class.md
@@ -0,0 +1,151 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+# 文本方向分类器
+
+## 1. 方法介绍
+
+文本方向分类器主要用于图片非0度的场景下，在这种场景下需要对图片里检测到的文本行进行一个转正的操作。在PaddleOCR系统内，
+文字检测之后得到的文本行图片经过仿射变换之后送入识别模型，此时只需要对文字进行一个0和180度的角度分类，因此PaddleOCR内置的
+文本方向分类器**只支持了0和180度的分类**。如果想支持更多角度，可以自己修改算法进行支持。
+
+0和180度数据样本例子：
+
+![img](./images/angle_class_example.jpg)
+
+## 2. 数据准备
+
+请按如下步骤设置数据集：
+
+训练数据的默认存储路径是 `PaddleOCR/train_data/cls`,如果您的磁盘上已有数据集，只需创建软链接至数据集目录：
+
+```bash linenums="1"
+ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/cls/dataset
+```
+
+请参考下文组织您的数据。
+
+### 训练集
+
+首先建议将训练图片放入同一个文件夹，并用一个txt文件（cls_gt_train.txt）记录图片路径和标签。
+
+**注意：** 默认请将图片路径和图片标签用 `\t` 分割，如用其他方式分割将造成训练报错
+
+0和180分别表示图片的角度为0度和180度
+
+```text linenums="1"
+" 图像文件名                 图像标注信息 "
+train/cls/train/word_001.jpg   0
+train/cls/train/word_002.jpg   180
+```
+
+最终训练集应有如下文件结构：
+
+```text linenums="1"
+|-train_data
+    |-cls
+        |- cls_gt_train.txt
+        |- train
+            |- word_001.png
+            |- word_002.jpg
+            |- word_003.jpg
+            | ...
+```
+
+### 测试集
+
+同训练集类似，测试集也需要提供一个包含所有图片的文件夹（test）和一个cls_gt_test.txt，测试集的结构如下所示：
+
+```text linenums="1"
+|-train_data
+    |-cls
+        |- cls_gt_test.txt
+        |- test
+            |- word_001.jpg
+            |- word_002.jpg
+            |- word_003.jpg
+            | ...
+```
+
+## 3. 启动训练
+
+将准备好的txt文件和图片文件夹路径分别写入配置文件的 `Train/Eval.dataset.label_file_list` 和 `Train/Eval.dataset.data_dir` 字段下，`Train/Eval.dataset.data_dir`字段下的路径和文件里记载的图片名构成了图片的绝对路径。
+
+PaddleOCR提供了训练脚本、评估脚本和预测脚本。
+
+### 开始训练
+
+*如果您安装的是cpu版本，请将配置文件中的 `use_gpu` 字段修改为false*
+
+```bash linenums="1"
+# GPU训练 支持单卡，多卡训练，通过 '--gpus' 指定卡号。
+# 启动训练，下面的命令已经写入train.sh文件中，只需修改文件里的配置文件路径即可
+python3 -m paddle.distributed.launch --gpus '0,1,2,3,4,5,6,7'  tools/train.py -c configs/cls/cls_mv3.yml
+```
+
+### 数据增强
+
+PaddleOCR提供了多种数据增强方式，如果您希望在训练时加入扰动，请在配置文件中取消`Train.dataset.transforms`下的`RecAug`和`RandAugment`字段的注释。
+
+默认的扰动方式有：颜色空间转换(cvtColor)、模糊(blur)、抖动(jitter)、噪声(Gasuss noise)、随机切割(random crop)、透视(perspective)、颜色反转(reverse),随机数据增强(RandAugment)。
+
+训练过程中除随机数据增强外每种扰动方式以50%的概率被选择，具体代码实现请参考：
+[rec_img_aug.py](../../../../ppocr/data/imaug/rec_img_aug.py)
+[randaugment.py](../../../../ppocr/data/imaug/randaugment.py)
+
+*由于OpenCV的兼容性问题，扰动操作暂时只支持linux*
+
+## 4. 训练
+
+PaddleOCR支持训练和评估交替进行, 可以在 `configs/cls/cls_mv3.yml` 中修改 `eval_batch_step` 设置评估频率，默认每1000个iter评估一次。训练过程中将会保存如下内容：
+
+```bash linenums="1"
+├── best_accuracy.pdopt # 最佳模型的优化器参数
+├── best_accuracy.pdparams # 最佳模型的参数
+├── best_accuracy.states # 最佳模型的指标和epoch等信息
+├── config.yml # 本次实验的配置文件
+├── latest.pdopt # 最新模型的优化器参数
+├── latest.pdparams # 最新模型的参数
+├── latest.states # 最新模型的指标和epoch等信息
+└── train.log # 训练日志
+```
+
+如果验证集很大，测试将会比较耗时，建议减少评估次数，或训练完再进行评估。
+
+**注意，预测/评估时的配置文件请务必与训练一致。**
+
+## 5. 评估
+
+评估数据集可以通过修改`configs/cls/cls_mv3.yml`文件里的`Eval.dataset.label_file_list` 字段设置。
+
+```bash linenums="1"
+export CUDA_VISIBLE_DEVICES=0
+# GPU 评估， Global.checkpoints 为待测权重
+python3 tools/eval.py -c configs/cls/cls_mv3.yml -o Global.checkpoints={path/to/weights}/best_accuracy
+```
+
+## 6. 预测
+
+### 训练引擎的预测
+
+使用 PaddleOCR 训练好的模型，可以通过以下脚本进行快速预测。
+
+通过 `Global.infer_img` 指定预测图片或文件夹路径，通过 `Global.checkpoints` 指定权重：
+
+```bash linenums="1"
+# 预测分类结果
+python3 tools/infer_cls.py -c configs/cls/cls_mv3.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.load_static_weights=false Global.infer_img=doc/imgs_words/ch/word_1.jpg
+```
+
+预测图片：
+
+![img](./images/word_1-20240704092713071.jpg)
+
+得到输入图像的预测结果：
+
+```bash linenums="1"
+infer_img: doc/imgs_words/ch/word_1.jpg
+     result: ('0', 0.9998784)
+```
diff --git a/docs/version2.x/ppocr/model_train/detection.en.md b/docs/version2.x/ppocr/model_train/detection.en.md
new file mode 100644
index 0000000..3cf4350
--- /dev/null
+++ b/docs/version2.x/ppocr/model_train/detection.en.md
@@ -0,0 +1,248 @@
+---
+comments: true
+---
+
+# Text Detection
+
+This section uses the icdar2015 dataset as an example to introduce the training, evaluation, and testing of the detection model in PaddleOCR.
+
+## 1. Data and Weights Preparation
+
+### 1.1 Data Preparation
+
+To prepare datasets, refer to [ocr_datasets](../../../datasets/ocr_datasets.en.md).
+
+### 1.2 Download Pre-trained Model
+
+First download the pre-trained model. The detection model of PaddleOCR currently supports 3 backbones, namely MobileNetV3, ResNet18_vd and ResNet50_vd. You can use the model in [PaddleClas](https://github.com/PaddlePaddle/PaddleClas/tree/release/2.0/ppcls/modeling/architectures) to replace backbone according to your needs.
+And the responding download link of backbone pre-trained weights can be found in (<https://github.com/PaddlePaddle/PaddleClas/blob/release%2F2.0/README_cn.md#resnet%E5%8F%8A%E5%85%B6vd%E7%B3%BB%E5%88%97>).
+
+```bash linenums="1"
+cd PaddleOCR/
+# Download the pre-trained model of MobileNetV3
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/MobileNetV3_large_x0_5_pretrained.pdparams
+# or, download the pre-trained model of ResNet18_vd
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/ResNet18_vd_pretrained.pdparams
+# or, download the pre-trained model of ResNet50_vd
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/ResNet50_vd_ssld_pretrained.pdparams
+
+```
+
+## 2. Training
+
+### 2.1 Start Training
+
+*If CPU version installed, please set the parameter `use_gpu` to `false` in the configuration.*
+
+```bash linenums="1"
+python3 tools/train.py -c configs/det/det_mv3_db.yml  \
+         -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained
+```
+
+In the above instruction, use `-c` to select the training to use the `configs/det/det_mv3_db.yml` configuration file.
+For a detailed explanation of the configuration file, please refer to [config](../blog/config.en.md).
+
+You can also use `-o` to change the training parameters without modifying the yml file. For example, adjust the training learning rate to 0.0001.
+
+```bash linenums="1"
+# single GPU training
+python3 tools/train.py -c configs/det/det_mv3_db.yml -o   \
+         Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained  \
+         Optimizer.base_lr=0.0001
+
+# multi-GPU training
+# Set the GPU ID used by the '--gpus' parameter.
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/det/det_mv3_db.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained
+
+# multi-Node, multi-GPU training
+# Set the IPs of your nodes used by the '--ips' parameter. Set the GPU ID used by the '--gpus' parameter.
+python3 -m paddle.distributed.launch --ips="xx.xx.xx.xx,xx.xx.xx.xx" --gpus '0,1,2,3' tools/train.py -c configs/det/det_mv3_db.yml \
+     -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained
+```
+
+**Note:** For multi-Node multi-GPU training, you need to replace the `ips` value in the preceding command with the address of your machine, and the machines must be able to ping each other. In addition, it requires activating commands separately on multiple machines when we start the training. The command for viewing the IP address of the machine is `ifconfig`.
+
+If you want to further speed up the training, you can use [automatic mixed precision training](https://www.paddlepaddle.org.cn/documentation/docs/zh/guides/01_paddle2.0_introduction/basic_concept/amp_en.html). for single card training, the command is as follows:
+
+```bash linenums="1"
+python3 tools/train.py -c configs/det/det_mv3_db.yml \
+     -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+     Global.use_amp=True Global.scale_loss=1024.0 Global.use_dynamic_loss_scaling=True
+```
+
+### 2.2 Load Trained Model and Continue Training
+
+If you expect to load trained model and continue the training again, you can specify the parameter `Global.checkpoints` as the model path to be loaded.
+
+For example:
+
+```bash linenums="1"
+python3 tools/train.py -c configs/det/det_mv3_db.yml -o Global.checkpoints=./your/trained/model
+```
+
+**Note**: The priority of `Global.checkpoints` is higher than that of `Global.pretrained_model`, that is, when two parameters are specified at the same time, the model specified by `Global.checkpoints` will be loaded first. If the model path specified by `Global.checkpoints` is wrong, the one specified by `Global.pretrained_model` will be loaded.
+
+### 2.3 Training with New Backbone
+
+The network part completes the construction of the network, and PaddleOCR divides the network into four parts, which are under [ppocr/modeling](../../../../ppocr/modeling). The data entering the network will pass through these four parts in sequence(transforms->backbones->
+necks->heads).
+
+```bash linenums="1"
+├── architectures # Code for building network
+├── transforms    # Image Transformation Module
+├── backbones     # Feature extraction module
+├── necks         # Feature enhancement module
+└── heads         # Output module
+```
+
+If the Backbone to be replaced has a corresponding implementation in PaddleOCR, you can directly modify the parameters in the `Backbone` part of the configuration yml file.
+
+However, if you want to use a new Backbone, an example of replacing the backbones is as follows:
+
+1. Create a new file under the [ppocr/modeling/backbones](../../../../ppocr/modeling/backbones) folder, such as my_backbone.py.
+2. Add code in the my_backbone.py file, the sample code is as follows:
+
+```python linenums="1"
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+
+class MyBackbone(nn.Layer):
+    def __init__(self, *args, **kwargs):
+        super(MyBackbone, self).__init__()
+        # your init code
+        self.conv = nn.xxxx
+
+    def forward(self, inputs):
+        # your network forward
+        y = self.conv(inputs)
+        return y
+```
+
+3. Import the added module in the [ppocr/modeling/backbones/\_*init\_*.py](../../../../ppocr/modeling/backbones/__init__.py) file.
+
+After adding the four-part modules of the network, you only need to configure them in the configuration file to use, such as:
+
+```yaml linenums="1"
+  Backbone:
+    name: MyBackbone
+    args1: args1
+```
+
+**NOTE**: More details about replace Backbone and other module can be found in [doc](../../algorithm/add_new_algorithm.en.md).
+
+### 2.4 Mixed Precision Training
+
+If you want to speed up your training further, you can use [Auto Mixed Precision Training](https://www.paddlepaddle.org.cn/documentation/docs/zh/guides/01_paddle2.0_introduction/basic_concept/amp_cn.html), taking a single machine and a single gpu as an example, the commands are as follows:
+
+```bash linenums="1"
+python3 tools/train.py -c configs/det/det_mv3_db.yml \
+     -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+     Global.use_amp=True Global.scale_loss=1024.0 Global.use_dynamic_loss_scaling=True
+```
+
+### 2.5 Distributed Training
+
+During multi-machine multi-gpu training, use the `--ips` parameter to set the used machine IP address, and the `--gpus` parameter to set the used GPU ID:
+
+```bash linenums="1"
+python3 -m paddle.distributed.launch --ips="xx.xx.xx.xx,xx.xx.xx.xx" --gpus '0,1,2,3' tools/train.py -c configs/det/det_mv3_db.yml \
+     -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained
+```
+
+**Note:** (1) When using multi-machine and multi-gpu training, you need to replace the ips value in the above command with the address of your machine, and the machines need to be able to ping each other. (2) Training needs to be launched separately on multiple machines. The command to view the ip address of the machine is `ifconfig`. (3) For more details about the distributed training speedup ratio, please refer to [Distributed Training Tutorial](../blog/distributed_training.en.md).
+
+### 2.6 Training with knowledge distillation
+
+Knowledge distillation is supported in PaddleOCR for text detection training process. For more details, please refer to [doc](./../model_compress/knowledge_distillation.en.md).
+
+### 2.7 Training on other platform(Windows/macOS/Linux DCU)
+
+- Windows GPU/CPU
+The Windows platform is slightly different from the Linux platform:
+Windows platform only supports `single gpu` training and inference, specify GPU for training `set CUDA_VISIBLE_DEVICES=0`
+On the Windows platform, DataLoader only supports single-process mode, so you need to set `num_workers` to 0;
+
+- macOS
+GPU mode is not supported, you need to set `use_gpu` to False in the configuration file, and the rest of the training evaluation prediction commands are exactly the same as Linux GPU.
+
+- Linux DCU
+Running on a DCU device requires setting the environment variable `export HIP_VISIBLE_DEVICES=0,1,2,3`, and the rest of the training and evaluation prediction commands are exactly the same as the Linux GPU.
+
+### 2.8 Fine-tuning
+
+In actual use, it is recommended to load the official pre-trained model and fine-tune it in your own data set. For the fine-tuning method of the detection model, please refer to: [Model Fine-tuning Tutorial](./finetune.en.md).
+
+## 3. Evaluation and Test
+
+### 3.1 Evaluation
+
+PaddleOCR calculates three indicators for evaluating performance of OCR detection task: Precision, Recall, and Hmean(F-Score).
+
+Run the following code to calculate the evaluation indicators. The result will be saved in the test result file specified by `save_res_path` in the configuration file `det_db_mv3.yml`
+
+When evaluating, set post-processing parameters `box_thresh=0.6`, `unclip_ratio=1.5`. If you use different datasets, different models for training, these two parameters should be adjusted for better result.
+
+The model parameters during training are saved in the `Global.save_model_dir` directory by default. When evaluating indicators, you need to set `Global.checkpoints` to point to the saved parameter file.
+
+```bash linenums="1"
+python3 tools/eval.py -c configs/det/det_mv3_db.yml  -o Global.checkpoints="{path/to/weights}/best_accuracy" PostProcess.box_thresh=0.6 PostProcess.unclip_ratio=1.5
+```
+
+- Note: `box_thresh` and `unclip_ratio` are parameters required for DB post-processing, and not need to be set when evaluating the EAST and SAST model.
+
+### 3.2 Test
+
+Test the detection result on a single image:
+
+```bash linenums="1"
+python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/img_10.jpg" Global.pretrained_model="./output/det_db/best_accuracy"
+```
+
+When testing the DB model, adjust the post-processing threshold:
+
+```bash linenums="1"
+python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/img_10.jpg" Global.pretrained_model="./output/det_db/best_accuracy"  PostProcess.box_thresh=0.6 PostProcess.unclip_ratio=2.0
+```
+
+Test the detection result on all images in the folder:
+
+```bash linenums="1"
+python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/" Global.pretrained_model="./output/det_db/best_accuracy"
+```
+
+## 4. Inference
+
+The inference model (the model saved by `paddle.jit.save`) is generally a solidified model saved after the model training is completed, and is mostly used to give prediction in deployment.
+
+The model saved during the training process is the checkpoints model, which saves the parameters of the model and is mostly used to resume training.
+
+Compared with the checkpoints model, the inference model will additionally save the structural information of the model. Therefore, it is easier to deploy because the model structure and model parameters are already solidified in the inference model file, and is suitable for integration with actual systems.
+
+Firstly, we can convert DB trained model to inference model:
+
+```bash linenums="1"
+python3 tools/export_model.py -c configs/det/det_mv3_db.yml -o Global.pretrained_model="./output/det_db/best_accuracy" Global.save_inference_dir="./output/det_db_inference/"
+```
+
+The detection inference model prediction：
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --det_algorithm="DB" --det_model_dir="./output/det_db_inference/" --image_dir="./doc/imgs/" --use_gpu=True
+```
+
+If it is other detection algorithms, such as the EAST, the det_algorithm parameter needs to be modified to EAST, and the default is the DB algorithm:
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --det_algorithm="EAST" --det_model_dir="./output/det_db_inference/" --image_dir="./doc/imgs/" --use_gpu=True
+```
+
+## 5. FAQ
+
+Q1: The prediction results of trained model and inference model are inconsistent?
+
+**A**: Most of the problems are caused by the inconsistency of the pre-processing and post-processing parameters during the prediction of the trained model and the pre-processing and post-processing parameters during the prediction of the inference model. Taking the model trained by the det_mv3_db.yml configuration file as an example, the solution to the problem of inconsistent prediction results between the training model and the inference model is as follows:
+
+- Check whether the [trained model preprocessing](../../../../configs/det/det_mv3_db.yml#L116) is consistent with the prediction [preprocessing function of the inference model](../../../../tools/infer/predict_det.py#L42). When the algorithm is evaluated, the input image size will affect the accuracy. In order to be consistent with the paper, the image is resized to [736, 1280] in the training icdar15 configuration file, but there is only a set of default parameters when the inference model predicts, which will be considered. To predict the speed problem, the longest side of the image is limited to 960 for resize by default. The preprocessing function of the training model preprocessing and the inference model is located in [ppocr/data/imaug/operators.py](../../../../ppocr/data/imaug/operators.py#L147).
+- Check whether the [post-processing of the trained model](../../../../configs/det/det_mv3_db.yml#L51) is consistent with the [post-processing parameters of the inference](../../../../tools/infer/utility.py#L50).
diff --git a/docs/version2.x/ppocr/model_train/detection.md b/docs/version2.x/ppocr/model_train/detection.md
new file mode 100644
index 0000000..e747176
--- /dev/null
+++ b/docs/version2.x/ppocr/model_train/detection.md
@@ -0,0 +1,236 @@
+---
+comments: true
+---
+
+
+# 文字检测
+
+本节以icdar2015数据集为例，介绍PaddleOCR中检测模型训练、评估、测试的使用方式。
+
+## 1. 准备数据和模型
+
+### 1.1 准备数据集
+
+准备数据集可参考 [ocr_datasets](../../../datasets/ocr_datasets.md) 。
+
+### 1.2 下载预训练模型
+
+首先下载模型backbone的pretrain model，PaddleOCR的检测模型目前支持两种backbone，分别是MobileNetV3、ResNet_vd系列，
+您可以根据需求使用[PaddleClas](https://github.com/PaddlePaddle/PaddleClas/tree/release/2.0/ppcls/modeling/architectures)中的模型更换backbone，
+对应的backbone预训练模型可以从[PaddleClas repo 主页中找到下载链接](https://github.com/PaddlePaddle/PaddleClas/blob/release%2F2.0/README_cn.md#resnet%E5%8F%8A%E5%85%B6vd%E7%B3%BB%E5%88%97)。
+
+```bash linenums="1"
+cd PaddleOCR/
+# 根据backbone的不同选择下载对应的预训练模型
+# 下载MobileNetV3的预训练模型
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/MobileNetV3_large_x0_5_pretrained.pdparams
+# 或，下载ResNet18_vd的预训练模型
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/ResNet18_vd_pretrained.pdparams
+# 或，下载ResNet50_vd的预训练模型
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/ResNet50_vd_ssld_pretrained.pdparams
+```
+
+## 2. 开始训练
+
+### 2.1 启动训练
+
+*如果您安装的是cpu版本，请将配置文件中的 `use_gpu` 字段修改为false*
+
+```bash linenums="1"
+# 单机单卡训练 mv3_db 模型
+python3 tools/train.py -c configs/det/det_mv3_db.yml \
+     -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained
+
+# 单机多卡训练，通过 --gpus 参数设置使用的GPU ID
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/det/det_mv3_db.yml \
+     -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained
+
+```
+
+上述指令中，通过-c 选择训练使用configs/det/det_mv3_db.yml配置文件。
+有关配置文件的详细解释，请参考[链接](../blog/config.md)。
+
+您也可以通过-o参数在不需要修改yml文件的情况下，改变训练的参数，比如，调整训练的学习率为0.0001
+
+```bash linenums="1"
+python3 tools/train.py -c configs/det/det_mv3_db.yml -o Optimizer.base_lr=0.0001
+```
+
+### 2.2 断点训练
+
+如果训练程序中断，如果希望加载训练中断的模型从而恢复训练，可以通过指定Global.checkpoints指定要加载的模型路径：
+
+```bash linenums="1"
+python3 tools/train.py -c configs/det/det_mv3_db.yml -o Global.checkpoints=./your/trained/model
+```
+
+**注意**：`Global.checkpoints`的优先级高于`Global.pretrained_model`的优先级，即同时指定两个参数时，优先加载`Global.checkpoints`指定的模型，如果`Global.checkpoints`指定的模型路径有误，会加载`Global.pretrained_model`指定的模型。
+
+### 2.3 更换Backbone 训练
+
+PaddleOCR将网络划分为四部分，分别在[ppocr/modeling](../../../../ppocr/modeling)下。 进入网络的数据将按照顺序(transforms->backbones->
+necks->heads)依次通过这四个部分。
+
+```bash linenums="1"
+├── architectures # 网络的组网代码
+├── transforms    # 网络的图像变换模块
+├── backbones     # 网络的特征提取模块
+├── necks         # 网络的特征增强模块
+└── heads         # 网络的输出模块
+```
+
+如果要更换的Backbone 在PaddleOCR中有对应实现，直接修改配置yml文件中`Backbone`部分的参数即可。
+
+如果要使用新的Backbone，更换backbones的例子如下:
+
+1. 在 [ppocr/modeling/backbones](../../../../ppocr/modeling/backbones) 文件夹下新建文件，如my_backbone.py。
+2. 在 my_backbone.py 文件内添加相关代码，示例代码如下:
+
+```python linenums="1"
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+
+class MyBackbone(nn.Layer):
+    def __init__(self, *args, **kwargs):
+        super(MyBackbone, self).__init__()
+        # your init code
+        self.conv = nn.xxxx
+
+    def forward(self, inputs):
+        # your network forward
+        y = self.conv(inputs)
+        return y
+```
+
+3. 在 [ppocr/modeling/backbones/\_*init\_*.py](../../../../ppocr/modeling/backbones/__init__.py)文件内导入添加的`MyBackbone`模块，然后修改配置文件中Backbone进行配置即可使用，格式如下:
+
+```yaml linenums="1"
+Backbone:
+name: MyBackbone
+args1: args1
+```
+
+**注意**：如果要更换网络的其他模块，可以参考[文档](../../algorithm/add_new_algorithm.md)。
+
+### 2.4 混合精度训练
+
+如果您想进一步加快训练速度，可以使用[自动混合精度训练](https://www.paddlepaddle.org.cn/documentation/docs/zh/guides/01_paddle2.0_introduction/basic_concept/amp_cn.html)， 以单机单卡为例，命令如下：
+
+```bash linenums="1"
+python3 tools/train.py -c configs/det/det_mv3_db.yml \
+     -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained \
+     Global.use_amp=True Global.scale_loss=1024.0 Global.use_dynamic_loss_scaling=True
+```
+
+**注意:** 文本检测模型使用AMP时可能遇到训练不收敛问题，可以参考[discussions](https://github.com/PaddlePaddle/PaddleOCR/discussions/12445)中的临时解决方案进行使用。
+
+### 2.5 分布式训练
+
+多机多卡训练时，通过 `--ips` 参数设置使用的机器IP地址，通过 `--gpus` 参数设置使用的GPU ID：
+
+```bash linenums="1"
+python3 -m paddle.distributed.launch --ips="xx.xx.xx.xx,xx.xx.xx.xx" --gpus '0,1,2,3' tools/train.py -c configs/det/det_mv3_db.yml \
+     -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained
+```
+
+**注意:** （1）采用多机多卡训练时，需要替换上面命令中的ips值为您机器的地址，机器之间需要能够相互ping通；（2）训练时需要在多个机器上分别启动命令。查看机器ip地址的命令为`ifconfig`；（3）更多关于分布式训练的性能优势等信息，请参考：[分布式训练教程](../blog/distributed_training.md)。
+
+### 2.6 知识蒸馏训练
+
+PaddleOCR支持了基于知识蒸馏的检测模型训练过程，更多内容可以参考[知识蒸馏说明文档](../model_compress/knowledge_distillation.md)。
+
+**注意：** 知识蒸馏训练目前只支持PP-OCR使用的`DB`和`CRNN`算法。
+
+### 2.7 其他训练环境
+
+- Windows GPU/CPU
+在Windows平台上与Linux平台略有不同:
+Windows平台只支持`单卡`的训练与预测，指定GPU进行训练`set CUDA_VISIBLE_DEVICES=0`
+在Windows平台，DataLoader只支持单进程模式，因此需要设置 `num_workers` 为0;
+
+- macOS
+不支持GPU模式，需要在配置文件中设置`use_gpu`为False，其余训练评估预测命令与Linux GPU完全相同。
+
+- Linux DCU
+DCU设备上运行需要设置环境变量 `export HIP_VISIBLE_DEVICES=0,1,2,3`，其余训练评估预测命令与Linux GPU完全相同。
+
+### 2.8 模型微调
+
+实际使用过程中，建议加载官方提供的预训练模型，在自己的数据集中进行微调，关于检测模型的微调方法，请参考：[模型微调教程](./finetune.md)。
+
+## 3. 模型评估与预测
+
+### 3.1 指标评估
+
+PaddleOCR计算三个OCR检测相关的指标，分别是：Precision、Recall、Hmean（F-Score）。
+
+训练中模型参数默认保存在`Global.save_model_dir`目录下。在评估指标时，需要设置`Global.checkpoints`指向保存的参数文件。
+
+```bash linenums="1"
+python3 tools/eval.py -c configs/det/det_mv3_db.yml  -o Global.checkpoints="{path/to/weights}/best_accuracy"
+```
+
+### 3.2 测试检测效果
+
+测试单张图像的检测效果：
+
+```bash linenums="1"
+python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/img_10.jpg" Global.pretrained_model="./output/det_db/best_accuracy"
+```
+
+测试DB模型时，调整后处理阈值：
+
+```bash linenums="1"
+python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/img_10.jpg" Global.pretrained_model="./output/det_db/best_accuracy"  PostProcess.box_thresh=0.6 PostProcess.unclip_ratio=2.0
+```
+
+- 注：`box_thresh`、`unclip_ratio`是DB后处理参数，其他检测模型不支持。
+
+测试文件夹下所有图像的检测效果：
+
+```bash linenums="1"
+python3 tools/infer_det.py -c configs/det/det_mv3_db.yml -o Global.infer_img="./doc/imgs_en/" Global.pretrained_model="./output/det_db/best_accuracy"
+```
+
+## 4. 模型导出与预测
+
+inference 模型（`paddle.jit.save`保存的模型）
+一般是模型训练，把模型结构和模型参数保存在文件中的固化模型，多用于预测部署场景。
+训练过程中保存的模型是checkpoints模型，保存的只有模型的参数，多用于恢复训练等。
+与checkpoints模型相比，inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成。
+
+检测模型转inference 模型方式：
+
+```bash linenums="1"
+# 加载配置文件`det_mv3_db.yml`，从`output/det_db`目录下加载`best_accuracy`模型，inference模型保存在`./output/det_db_inference`目录下
+python3 tools/export_model.py -c configs/det/det_mv3_db.yml -o Global.pretrained_model="./output/det_db/best_accuracy" Global.save_inference_dir="./output/det_db_inference/"
+```
+
+DB检测模型inference 模型预测：
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --det_algorithm="DB" --det_model_dir="./output/det_db_inference/" --image_dir="./doc/imgs/" --use_gpu=True
+```
+
+如果是其他检测，比如EAST模型，det_algorithm参数需要修改为EAST，默认为DB算法：
+
+```bash linenums="1"
+python3 tools/infer/predict_det.py --det_algorithm="EAST" --det_model_dir="./output/det_db_inference/" --image_dir="./doc/imgs/" --use_gpu=True
+```
+
+更多关于推理超参数的配置与解释，请参考：[模型推理超参数解释教程](../blog/inference_args.md)。
+
+## 5. FAQ
+
+Q1: 训练模型转inference 模型之后预测效果不一致？
+
+**A**：此类问题出现较多，问题多是trained model预测时候的预处理、后处理参数和inference model预测的时候的预处理、后处理参数不一致导致的。以det_mv3_db.yml配置文件训练的模型为例，训练模型、inference模型预测结果不一致问题解决方式如下：
+
+- 检查[trained model预处理](../../../../configs/det/det_mv3_db.yml#L116)，和[inference model的预测预处理](../../../../tools/infer/predict_det.py#L42)函数是否一致。算法在评估的时候，输入图像大小会影响精度，为了和论文保持一致，训练icdar15配置文件中将图像resize到[736, 1280]，但是在inference model预测的时候只有一套默认参数，会考虑到预测速度问题，默认限制图像最长边为960做resize的。训练模型预处理和inference模型的预处理函数位于[ppocr/data/imaug/operators.py](../../../../ppocr/data/imaug/operators.py#L147)
+- 检查[trained model后处理](../../../../configs/det/det_mv3_db.yml#L51)，和[inference 后处理参数](../../../../tools/infer/utility.py#L50)是否一致。
+
+Q1: 训练EAST模型提示找不到lanms库？
+
+**A**：执行pip3 install lanms-nova 即可。
diff --git a/docs/version2.x/ppocr/model_train/finetune.en.md b/docs/version2.x/ppocr/model_train/finetune.en.md
new file mode 100644
index 0000000..f505c3c
--- /dev/null
+++ b/docs/version2.x/ppocr/model_train/finetune.en.md
@@ -0,0 +1,227 @@
+---
+comments: true
+---
+
+# Fine-tune
+
+## 1. Background and meaning
+
+The PP-OCR series models provided by PaddleOCR have excellent performance in general scenarios and can solve detection and recognition problems in most cases. In vertical scenarios, if you want to obtain better model, you can further improve the accuracy of the PP-OCR series detection and recognition models through fine-tune.
+
+This article mainly introduces some precautions when fine-tuning the text detection and recognition model. Finally, you can obtain a text detection and recognition model with higher accuracy through model fine-tuning in your own scenarios.
+
+The core points of this article are as follows:
+
+1. The pre-trained model provided by PP-OCR has better generalization ability
+2. Adding a small amount of real data (detection:>=500, recognition:>=5000) will greatly improve the detection and recognition effect of vertical scenes
+3. When fine-tuning the model, adding real general scene data can further improve the model accuracy and generalization performance
+4. In the text detection task, increasing the prediction shape of the image can further improve the detection effect of the smaller text area
+5. When fine-tuning the model, it is necessary to properly adjust the hyperparameters (learning rate and batch size are the most important) to obtain a better fine-tuning effect.
+
+For more details, please refer to Chapter 2 and Chapter 3.
+
+## 2. Text detection model fine-tuning
+
+### 2.1 Dataset
+
+* Dataset: It is recommended to prepare at least 500 text detection datasets for model fine-tuning.
+
+* Dataset annotation: single-line text annotation format, it is recommended that the labeled detection frame be consistent with the actual semantic content. For example, in the train ticket scene, the surname and first name may be far apart, but they belong to the same detection field semantically. Here, the entire name also needs to be marked as a detection frame.
+
+### 2.2 Model
+
+It is recommended to choose the PP-OCRv3 model (configuration file: [PP-OCRv3_mobile_det.yml](../../../../configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml)，pre-trained model: [ch_PP-OCRv3_det_distill_train.tar](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar), its accuracy and generalization performance is the best pre-training model currently available.
+
+For more PP-OCR series models, please refer to [PP-OCR Series Model Library](../model_list.en.md).
+
+Note: When using the above pre-trained model, you need to use the `student.pdparams` file in the folder as the pre-trained model, that is, only use the student model.
+
+### 2.3 Training hyperparameter
+
+When fine-tuning the model, the most important hyperparameter is the pre-training model path `pretrained_model`, `learning_rate` and `batch_size`，some hyperparameters are as follows:
+
+```yaml linenums="1"
+Global:
+  pretrained_model: ./ch_PP-OCRv3_det_distill_train/student.pdparams # pre-training model path
+Optimizer:
+  lr:
+    name: Cosine
+    learning_rate: 0.001 # learning_rate
+    warmup_epoch: 2
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Train:
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8  # single gpu batch size
+    num_workers: 4
+```
+
+In the above configuration file, you need to specify the `pretrained_model` field as the `student.pdparams` file path.
+
+The configuration file provided by PaddleOCR is for 8-gpu training (equivalent to a total batch size of `8*8=64`) and no pre-trained model is loaded. Therefore, in your scenario, the learning rate is the same as the total The batch size needs to be adjusted linearly, for example
+
+* If your scenario is single-gpu training, single gpu batch_size=8, then the total batch_size=8, it is recommended to adjust the learning rate to about `1e-4`.
+* If your scenario is for single-gpu training, due to memory limitations, you can only set batch_size=4 for a single gpu, and the total batch_size=4. It is recommended to adjust the learning rate to about `5e-5`.
+
+### 2.4 Prediction hyperparameter
+
+When exporting and inferring the trained model, you can further adjust the predicted image scale to improve the detection effect of small-area text. The following are some hyperparameters during DBNet inference, which can be adjusted appropriately to improve the effect.
+
+| hyperparameter | type | default | meaning |
+| :--: | :--: | :--: | :--: |
+|  det_db_thresh | float | 0.3 | In the probability map output by DB, pixels with a score greater than the threshold will be considered as text pixels |
+|  det_db_box_thresh | float | 0.6 | When the average score of all pixels within the frame of the detection result is greater than the threshold, the result will be considered as a text area |
+|  det_db_unclip_ratio | float | 1.5 | The expansion coefficient of `Vatti clipping`, using this method to expand the text area |
+|  max_batch_size | int | 10 | batch size |
+|  use_dilation | bool | False | Whether to expand the segmentation results to obtain better detection results |
+|  det_db_score_mode | str | "fast" | DB's detection result score calculation method supports `fast` and `slow`. `fast` calculates the average score based on all pixels in the polygon’s circumscribed rectangle border, and `slow` calculates the average score based on all pixels in the original polygon. The calculation speed is relatively slower, but more accurate. |
+
+For more information on inference methods, please refer to [Paddle Inference doc](../../legacy/python_infer.en.md).
+
+## 3. Text recognition model fine-tuning
+
+### 3.1 Dataset
+
+* Dataset：If the dictionary is not changed, it is recommended to prepare at least 5,000 text recognition datasets for model fine-tuning; if the dictionary is changed (not recommended), more quantities are required.
+
+* Data distribution: It is recommended that the distribution be as consistent as possible with the actual measurement scenario. If the actual scene contains a lot of short text, it is recommended to include more short text in the training data. If the actual scene has high requirements for the recognition effect of spaces, it is recommended to include more text content with spaces in the training data.
+
+* Data synthesis: In the case of some character recognition errors, it is recommended to obtain a batch of specific character dataset, add it to the original dataset and use a small learning rate for fine-tuning. The ratio of original dataset to new dataset can be 10:1 to 5:1 to avoid overfitting of the model caused by too much data in a single scene. At the same time, try to balance the word frequency of the corpus to ensure that the frequency of common words will not be too low.
+
+  Specific characters can be generated using the TextRenderer tool, for synthesis examples, please refer to [data synthesis](https://www.paddleocr.ai/v2.10.0/applications/%E5%85%89%E5%8A%9F%E7%8E%87%E8%AE%A1%E6%95%B0%E7%A0%81%E7%AE%A1%E5%AD%97%E7%AC%A6%E8%AF%86%E5%88%AB.html)
+  . The synthetic data corpus should come from real usage scenarios as much as possible, and keep the richness of fonts and backgrounds on the basis of being close to the real scene, which will help improve the model effect.
+
+* Common Chinese and English data: During training, common real data can be added to the training set (for example, in the fine-tuning scenario without changing the dictionary, it is recommended to add real data such as LSVT, RCTW, MTWI) to further improve the generalization performance of the model.
+
+### 3.2 Model
+
+It is recommended to choose the PP-OCRv3 model (configuration file: [PP-OCRv3_mobile_rec_distillation.yml](../../../../configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec_distillation.yml)，pre-trained model: [PP-OCRv3_mobile_rec_train.tar](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar)，its accuracy and generalization performance is the best pre-training model currently available.
+
+For more PP-OCR series models, please refer to [PP-OCR Series Model Library](../model_list.en.md).
+
+The PP-OCRv3 model uses the GTC strategy. The SAR branch has a large number of parameters. When the training data is a simple scene, the model is easy to overfit, resulting in poor fine-tuning effect. It is recommended to remove the GTC strategy. The configuration file of the model structure is modified as follows:
+
+```yaml linenums="1"
+Architecture:
+  model_type: rec
+  algorithm: SVTR
+  Transform:
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+    last_conv_stride: [1, 2]
+    last_pool_type: avg
+  Neck:
+    name: SequenceEncoder
+    encoder_type: svtr
+    dims: 64
+    depth: 2
+    hidden_dims: 120
+    use_guide: False
+  Head:
+    name: CTCHead
+    fc_decay: 0.00001
+Loss:
+  name: CTCLoss
+
+Train:
+  dataset:
+  ......
+    transforms:
+    # remove RecConAug
+    # - RecConAug:
+    #     prob: 0.5
+    #     ext_data_num: 2
+    #     image_shape: [48, 320, 3]
+    #     max_text_length: *max_text_length
+    - RecAug:
+    # modify Encode
+    - CTCLabelEncode:
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+...
+
+Eval:
+  dataset:
+  ...
+    transforms:
+    ...
+    - CTCLabelEncode:
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+...
+
+
+```
+
+### 3.3 Training hyperparameter
+
+Similar to text detection task fine-tuning, when fine-tuning the recognition model, the most important hyperparameters are the pre-trained model path `pretrained_model`, `learning_rate` and `batch_size`, some default configuration files are shown below.
+
+```yaml linenums="1"
+Global:
+  pretrained_model:  # pre-training model path
+Optimizer:
+  lr:
+    name: Piecewise
+    decay_epochs : [700, 800]
+    values : [0.001, 0.0001]  # learning_rate
+    warmup_epoch: 5
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/train_list.txt
+    ratio_list: [1.0] # Sampling ratio, the default value is [1.0]
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 128 # single gpu batch size
+    num_workers: 8
+
+```
+
+In the above configuration file, you first need to specify the `pretrained_model` field as the `ch_PP-OCRv3_rec_train/best_accuracy.pdparams` file path decompressed in Chapter 3.2.
+
+The configuration file provided by PaddleOCR is for 8-gpu training (equivalent to a total batch size of `8*128=1024`) and no pre-trained model is loaded. Therefore, in your scenario, the learning rate is the same as the total The batch size needs to be adjusted linearly, for example:
+
+* If your scenario is single-gpu training, single gpu batch_size=128, then the total batch_size=128, in the case of loading the pre-trained model, it is recommended to adjust the learning rate to about `[1e-4, 2e-5]` (For the piecewise learning rate strategy, two values need to be set, the same below).
+* If your scenario is for single-gpu training, due to memory limitations, you can only set batch_size=64 for a single gpu, and the total batch_size=64. When loading the pre-trained model, it is recommended to adjust the learning rate to `[5e-5 , 1e-5]`about.
+
+If there is general real scene data added, it is recommended that in each epoch, the amount of vertical scene data and real scene data should be kept at about 1:1.
+
+For example: your own vertical scene recognition data volume is 1W, the data label file is `vertical.txt`, the collected general scene recognition data volume is 10W, and the data label file is `general.txt`.
+
+Then, the `label_file_list` and `ratio_list` parameters can be set as shown below. In each epoch, `vertical.txt` will be fully sampled (sampling ratio is 1.0), including 1W pieces of data; `general.txt` will be sampled according to a sampling ratio of 0.1, including `10W*0.1=1W` pieces of data, the final ratio of the two is `1:1`.
+
+```yaml linenums="1"
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - vertical.txt
+    - general.txt
+    ratio_list: [1.0, 0.1]
+```
+
+### 3.4 Training optimization
+
+The training process does not happen overnight. After completing a stage of training evaluation, it is recommended to collect and analyze the badcase of the current model in the real scene, adjust the proportion of training data in a targeted manner, or further add synthetic data. Through multiple iterations of training, the model effect is continuously optimized.
+
+If you modify the custom dictionary during training, since the parameters of the last layer of FC cannot be loaded, it is normal for acc=0 at the beginning of the iteration. Don't worry, loading the pre-trained model can still speed up the model convergence.
diff --git a/docs/version2.x/ppocr/model_train/finetune.md b/docs/version2.x/ppocr/model_train/finetune.md
new file mode 100644
index 0000000..bfbfbef
--- /dev/null
+++ b/docs/version2.x/ppocr/model_train/finetune.md
@@ -0,0 +1,227 @@
+---
+comments: true
+---
+
+# 模型微调
+
+## 1. 模型微调背景与意义
+
+PaddleOCR提供的PP-OCR系列模型在通用场景中性能优异，能够解决绝大多数情况下的检测与识别问题。在垂类场景中，如果希望获取更优的模型效果，可以通过模型微调的方法，进一步提升PP-OCR系列检测与识别模型的精度。
+
+本文主要介绍文本检测与识别模型在模型微调时的一些注意事项，最终希望您在自己的场景中，通过模型微调，可以获取精度更高的文本检测与识别模型。
+
+本文核心要点如下所示。
+
+1. PP-OCR提供的预训练模型有较好的泛化能力
+2. 加入少量真实数据（检测任务>=500张, 识别任务>=5000张），会大幅提升垂类场景的检测与识别效果
+3. 在模型微调时，加入真实通用场景数据，可以进一步提升模型精度与泛化性能
+4. 在图像检测任务中，增大图像的预测尺度，能够进一步提升较小文字区域的检测效果
+5. 在模型微调时，需要适当调整超参数（学习率，batch size最为重要），以获得更优的微调效果。
+
+更多详细内容，请参考第2章与第3章。
+
+## 2. 文本检测模型微调
+
+### 2.1 数据选择
+
+* 数据量：建议至少准备500张的文本检测数据集用于模型微调。
+
+* 数据标注：单行文本标注格式，建议标注的检测框与实际语义内容一致。如在火车票场景中，姓氏与名字可能离得较远，但是它们在语义上属于同一个检测字段，这里也需要将整个姓名标注为1个检测框。
+
+### 2.2 模型选择
+
+建议选择PP-OCRv3模型（配置文件：[PP-OCRv3_mobile_det.yml](../../../../configs/det/PP-OCRv3/PP-OCRv3_mobile_det.yml)，预训练模型：[ch_PP-OCRv3_det_distill_train.tar](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams)）进行微调，其精度与泛化性能是目前提供的最优预训练模型。
+
+更多PP-OCR系列模型，请参考[PP-OCR 系列模型库](../model_list.md)。
+
+注意：在使用上述预训练模型的时候，需要使用文件夹中的`student.pdparams`文件作为预训练模型，即，仅使用学生模型。
+
+### 2.3 训练超参选择
+
+在模型微调的时候，最重要的超参就是预训练模型路径`pretrained_model`, 学习率`learning_rate`与`batch_size`，部分配置文件如下所示。
+
+```yaml linenums="1"
+Global:
+  pretrained_model: ./ch_PP-OCRv3_det_distill_train/student.pdparams # 预训练模型路径
+Optimizer:
+  lr:
+    name: Cosine
+    learning_rate: 0.001 # 学习率
+    warmup_epoch: 2
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Train:
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8  # 单卡batch size
+    num_workers: 4
+```
+
+上述配置文件中，首先需要将`pretrained_model`字段指定为`student.pdparams`文件路径。
+
+PaddleOCR提供的配置文件是在8卡训练（相当于总的batch size是`8*8=64`）、且没有加载预训练模型情况下的配置文件，因此您的场景中，学习率与总的batch size需要对应线性调整，例如
+
+* 如果您的场景中是单卡训练，单卡batch_size=8，则总的batch_size=8，建议将学习率调整为`1e-4`左右。
+* 如果您的场景中是单卡训练，由于显存限制，只能设置单卡batch_size=4，则总的batch_size=4，建议将学习率调整为`5e-5`左右。
+
+### 2.4 预测超参选择
+
+对训练好的模型导出并进行推理时，可以通过进一步调整预测的图像尺度，来提升小面积文本的检测效果，下面是DBNet推理时的一些超参数，可以通过适当调整，提升效果。
+
+| 参数名称 | 类型 | 默认值 | 含义 |
+| :--: | :--: | :--: | :--: |
+|  det_db_thresh | float | 0.3 | DB输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点 |
+|  det_db_box_thresh | float | 0.6 | 检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域 |
+|  det_db_unclip_ratio | float | 1.5 | `Vatti clipping`算法的扩张系数，使用该方法对文字区域进行扩张 |
+|  max_batch_size | int | 10 | 预测的batch size |
+|  use_dilation | bool | False | 是否对分割结果进行膨胀以获取更优检测效果 |
+|  det_db_score_mode | str | "fast" | DB的检测结果得分计算方法，支持`fast`和`slow`，`fast`是根据polygon的外接矩形边框内的所有像素计算平均得分，`slow`是根据原始polygon内的所有像素计算平均得分，计算速度相对较慢一些，但是更加准确一些。 |
+
+更多关于推理方法的介绍可以参考 [Paddle Inference推理教程](../../legacy/python_infer.md)。
+
+## 3. 文本识别模型微调
+
+### 3.1 数据选择
+
+* 数据量：不更换字典的情况下，建议至少准备5000张的文本识别数据集用于模型微调；如果更换了字典（不建议），需要的数量更多。
+
+* 数据分布：建议分布与实测场景尽量一致。如果实测场景包含大量短文本，则训练数据中建议也包含较多短文本，如果实测场景对于空格识别效果要求较高，则训练数据中建议也包含较多带空格的文本内容。
+
+* 数据合成：针对部分字符识别有误的情况，建议获取一批特定字符数据，加入到原数据中使用小学习率微调。其中原始数据与新增数据比例可尝试 10:1 ～ 5：1， 避免单一场景数据过多导致模型过拟合，同时尽量平衡语料词频，确保常用字的出现频率不会过低。
+
+  特定字符生成可以使用 TextRenderer 工具，合成例子可参考 [数码管数据合成](https://www.paddleocr.ai/v2.10.0/applications/%E5%85%89%E5%8A%9F%E7%8E%87%E8%AE%A1%E6%95%B0%E7%A0%81%E7%AE%A1%E5%AD%97%E7%AC%A6%E8%AF%86%E5%88%AB.html)
+  ，合成数据语料尽量来自真实使用场景，在贴近真实场景的基础上保持字体、背景的丰富性，有助于提升模型效果。
+
+* 通用中英文数据：在训练的时候，可以在训练集中添加通用真实数据（如在不更换字典的微调场景中，建议添加LSVT、RCTW、MTWI等真实数据），进一步提升模型的泛化性能。
+
+### 3.2 模型选择
+
+建议选择PP-OCRv3模型（配置文件：[PP-OCRv3_mobile_rec_distillation.yml](../../../../configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec_distillation.yml)，预训练模型：[ch_PP-OCRv3_rec_train.tar](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar)）进行微调，其精度与泛化性能是目前提供的最优预训练模型。
+
+更多PP-OCR系列模型，请参考[PP-OCR 系列模型库](../model_list.md)。
+
+PP-OCRv3 模型使用了GTC策略，其中SAR分支参数量大，当训练数据为简单场景时模型容易过拟合，导致微调效果不佳，建议去除GTC策略，模型结构部分配置文件修改如下：
+
+```yaml linenums="1"
+Architecture:
+  model_type: rec
+  algorithm: SVTR
+  Transform:
+  Backbone:
+    name: MobileNetV1Enhance
+    scale: 0.5
+    last_conv_stride: [1, 2]
+    last_pool_type: avg
+  Neck:
+    name: SequenceEncoder
+    encoder_type: svtr
+    dims: 64
+    depth: 2
+    hidden_dims: 120
+    use_guide: False
+  Head:
+    name: CTCHead
+    fc_decay: 0.00001
+Loss:
+  name: CTCLoss
+
+Train:
+  dataset:
+  ......
+    transforms:
+    # 去除 RecConAug 增广
+    # - RecConAug:
+    #     prob: 0.5
+    #     ext_data_num: 2
+    #     image_shape: [48, 320, 3]
+    #     max_text_length: *max_text_length
+    - RecAug:
+    # 修改 Encode 方式
+    - CTCLabelEncode:
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+...
+
+Eval:
+  dataset:
+  ...
+    transforms:
+    ...
+    - CTCLabelEncode:
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+...
+
+
+```
+
+### 3.3 训练超参选择
+
+与文本检测任务微调相同，在识别模型微调的时候，最重要的超参就是预训练模型路径`pretrained_model`, 学习率`learning_rate`与`batch_size`，部分默认配置文件如下所示。
+
+```yaml linenums="1"
+Global:
+  pretrained_model:  # 预训练模型路径
+Optimizer:
+  lr:
+    name: Piecewise
+    decay_epochs : [700, 800]
+    values : [0.001, 0.0001]  # 学习率
+    warmup_epoch: 5
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/train_list.txt
+    ratio_list: [1.0] # 采样比例，默认值是[1.0]
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 128 # 单卡batch size
+    num_workers: 8
+
+```
+
+上述配置文件中，首先需要将`pretrained_model`字段指定为3.2章节中解压得到的`ch_PP-OCRv3_rec_train/best_accuracy.pdparams`文件路径。
+
+PaddleOCR提供的配置文件是在8卡训练（相当于总的batch size是`8*128=1024`）、且没有加载预训练模型情况下的配置文件，因此您的场景中，学习率与总的batch size需要对应线性调整，例如：
+
+* 如果您的场景中是单卡训练，单卡batch_size=128，则总的batch_size=128，在加载预训练模型的情况下，建议将学习率调整为`[1e-4, 2e-5]`左右（piecewise学习率策略，需设置2个值，下同）。
+* 如果您的场景中是单卡训练，因为显存限制，只能设置单卡batch_size=64，则总的batch_size=64，在加载预训练模型的情况下，建议将学习率调整为`[5e-5, 1e-5]`左右。
+
+如果有通用真实场景数据加进来，建议每个epoch中，垂类场景数据与真实场景的数据量保持在1:1左右。
+
+比如：您自己的垂类场景识别数据量为1W，数据标签文件为`vertical.txt`，收集到的通用场景识别数据量为10W，数据标签文件为`general.txt`，
+
+那么，可以设置`label_file_list`和`ratio_list`参数如下所示。每个epoch中，`vertical.txt`中会进行全采样（采样比例为1.0），包含1W条数据；`general.txt`中会按照0.1的采样比例进行采样，包含`10W*0.1=1W`条数据，最终二者的比例为`1:1`。
+
+```yaml linenums="1"
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - vertical.txt
+    - general.txt
+    ratio_list: [1.0, 0.1]
+```
+
+### 3.4 训练调优
+
+训练过程并非一蹴而就的，完成一个阶段的训练评估后，建议收集分析当前模型在真实场景中的 badcase，有针对性的调整训练数据比例，或者进一步新增合成数据。通过多次迭代训练，不断优化模型效果。
+
+如果在训练时修改了自定义字典，由于无法加载最后一层FC的参数，在迭代初期acc=0是正常的情况，不必担心，加载预训练模型依然可以加快模型收敛。
diff --git a/docs/version2.x/ppocr/model_train/images/angle_class_example.jpg b/docs/version2.x/ppocr/model_train/images/angle_class_example.jpg
new file mode 100644
index 0000000..8e683be
Binary files /dev/null and b/docs/version2.x/ppocr/model_train/images/angle_class_example.jpg differ
diff --git a/docs/version2.x/ppocr/model_train/images/en_paper.jpg b/docs/version2.x/ppocr/model_train/images/en_paper.jpg
new file mode 100644
index 0000000..c051d3f
Binary files /dev/null and b/docs/version2.x/ppocr/model_train/images/en_paper.jpg differ
diff --git a/docs/version2.x/ppocr/model_train/images/icdar_rec.png b/docs/version2.x/ppocr/model_train/images/icdar_rec.png
new file mode 100644
index 0000000..a840d6a
Binary files /dev/null and b/docs/version2.x/ppocr/model_train/images/icdar_rec.png differ
diff --git a/docs/version2.x/ppocr/model_train/images/image-20240710082046188.jpg b/docs/version2.x/ppocr/model_train/images/image-20240710082046188.jpg
new file mode 100644
index 0000000..4f3e945
Binary files /dev/null and b/docs/version2.x/ppocr/model_train/images/image-20240710082046188.jpg differ
diff --git a/docs/version2.x/ppocr/model_train/images/long_text_examples.jpg b/docs/version2.x/ppocr/model_train/images/long_text_examples.jpg
new file mode 100644
index 0000000..4576152
Binary files /dev/null and b/docs/version2.x/ppocr/model_train/images/long_text_examples.jpg differ
diff --git a/docs/version2.x/ppocr/model_train/images/word_1-20240704092705543.png b/docs/version2.x/ppocr/model_train/images/word_1-20240704092705543.png
new file mode 100644
index 0000000..7b915fd
Binary files /dev/null and b/docs/version2.x/ppocr/model_train/images/word_1-20240704092705543.png differ
diff --git a/docs/version2.x/ppocr/model_train/images/word_1-20240704092713071.jpg b/docs/version2.x/ppocr/model_train/images/word_1-20240704092713071.jpg
new file mode 100644
index 0000000..cb5451e
Binary files /dev/null and b/docs/version2.x/ppocr/model_train/images/word_1-20240704092713071.jpg differ
diff --git a/docs/version2.x/ppocr/model_train/kie.en.md b/docs/version2.x/ppocr/model_train/kie.en.md
new file mode 100644
index 0000000..70c1fd9
--- /dev/null
+++ b/docs/version2.x/ppocr/model_train/kie.en.md
@@ -0,0 +1,476 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+# Key Information Extraction
+
+This tutorial provides a guide to the whole process of key information extraction using PaddleOCR, including data preparation, model training, optimization, evaluation, prediction of semantic entity recognition (SER) and relationship extraction (RE) tasks.
+
+## 1. Data Preparation
+
+### 1.1. Prepare for dataset
+
+PaddleOCR supports the following data format when training KIE models.
+
+- `general data` is used to train a dataset whose annotation is stored in a text file (SimpleDataset).
+
+The default storage path of training data is `PaddleOCR/train_data`. If you already have datasets on your disk, you only need to create a soft link to the dataset directory.
+
+```bash linenums="1"
+# linux and mac os
+ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/dataset
+# windows
+mklink /d <path/to/paddle_ocr>/train_data/dataset <path/to/dataset>
+```
+
+### 1.2. Custom Dataset
+
+The training process generally includes the training set and the evaluation set. The data formats of the two sets are same.
+
+#### (1) Training set
+
+It is recommended to put the training images into the same folder, record the path and annotation of images in a text file. The contents of the text file are as follows:
+
+```text linenums="1"
+" image path                 annotation information "
+zh_train_0.jpg   [{"transcription": "汇丰晋信", "label": "other", "points": [[104, 114], [530, 114], [530, 175], [104, 175]], "id": 1, "linking": []}, {"transcription": "受理时间:", "label": "question", "points": [[126, 267], [266, 267], [266, 305], [126, 305]], "id": 7, "linking": [[7, 13]]}, {"transcription": "2020.6.15", "label": "answer", "points": [[321, 239], [537, 239], [537, 285], [321, 285]], "id": 13, "linking": [[7, 13]]}]
+zh_train_1.jpg   [{"transcription": "中国人体器官捐献", "label": "other", "points": [[544, 459], [954, 459], [954, 517], [544, 517]], "id": 1, "linking": []}, {"transcription": ">编号:MC545715483585", "label": "other", "points": [[1462, 470], [2054, 470], [2054, 543], [1462, 543]], "id": 10, "linking": []}, {"transcription": "CHINAORGANDONATION", "label": "other", "points": [[543, 516], [958, 516], [958, 551], [543, 551]], "id": 14, "linking": []}, {"transcription": "中国人体器官捐献志愿登记表", "label": "header", "points": [[635, 793], [1892, 793], [1892, 904], [635, 904]], "id": 18, "linking": []}]
+...
+```
+
+**Note:** In the text file, please split the image path and annotation with `\t`. Otherwise, error will happen when training.
+
+The annotation can be parsed by `json` into a list of sub-annotations. Each element in the list is a dict, which stores the required information of each text line. The required fields are as follows.
+
+- transcription: stores the text content of the text line
+- label: the category of the text line content
+- points: stores the four point position information of the text line
+- id: stores the ID information of the text line for RE model training
+- linking: stores the connection information between text lines for RE model training
+
+#### (2) Evaluation set
+
+The evaluation set is constructed in the same way as the training set.
+
+#### (3) Dictionary file
+
+The textlines in the training set and the evaluation set contain label information. The list of all labels is stored in the dictionary file (such as `class_list.txt`). Each line in the dictionary file is represented as a label name.
+
+For example, FUND_zh data contains four categories. The contents of the dictionary file are as follows.
+
+```text linenums="1"
+OTHER
+QUESTION
+ANSWER
+HEADER
+```
+
+In the annotation file, the annotation information of the `label` field of the text line content of each annotation needs to belong to the dictionary content.
+
+The final dataset shall have the following file structure.
+
+```text linenums="1"
+|-train_data
+  |-data_name
+    |- train.json
+    |- train
+        |- zh_train_0.png
+        |- zh_train_1.jpg
+        | ...
+    |- val.json
+    |- val
+        |- zh_val_0.png
+        |- zh_val_1.jpg
+        | ...
+```
+
+**Note:**
+
+-The category information in the annotation file is not case sensitive. For example, 'HEADER' and 'header' will be seen as the same category ID.
+
+- In the dictionary file, it is recommended to put the `other` category (other textlines that need not be paid attention to can be labeled as `other`) on the first line. When parsing, the category ID of the 'other' category will be resolved to 0, and the textlines predicted as `other` will not be visualized later.
+
+### 1.3. Download data
+
+If you do not have local dataset, you can download the source files of [XFUND](https://github.com/doc-analysis/XFUND) or [FUNSD](https://guillaumejaume.github.io/FUNSD) and use the scripts of [XFUND](../../../../ppstructure/kie/tools/trans_xfun_data.py) or [FUNSD](../../../../ppstructure/kie/tools/trans_funsd_label.py) for transform them into PaddleOCR format. Then you can use the public dataset to quick experience KIE.
+
+For more information about public KIE datasets, please refer to [KIE dataset tutorial](../../../datasets/kie_datasets.en.md).
+
+PaddleOCR also supports the annotation of KIE models. Please refer to [PPOCRLabel tutorial](https://github.com/PFCCLab/PPOCRLabel/blob/main/README.md).
+
+## 2. Training
+
+PaddleOCR provides training scripts, evaluation scripts and inference scripts. We will introduce based on VI-LayoutXLM model in this section.
+This section will take the VI layoutxlm multimodal pre training model as an example to explain.
+
+> If you want to use the SDMGR based KIE algorithm, please refer to: [SDMGR tutorial](../../algorithm/kie/algorithm_kie_sdmgr.en.md).
+
+### 2.1. Start Training
+
+If you do not use a custom dataset, you can use XFUND_zh that has been processed in PaddleOCR dataset for quick experience.
+
+```bash linenums="1"
+mkdir train_data
+cd train_data
+wget https://paddleocr.bj.bcebos.com/ppstructure/dataset/XFUND.tar && tar -xf XFUND.tar
+cd ..
+```
+
+If you don't want to train, and want to directly experience the process of model evaluation, prediction, and inference, you can download the training model provided in PaddleOCR and skip section 2.1.
+
+Use the following command to download the trained model.
+
+```bash linenums="1"
+mkdir pretrained_model
+cd pretrained_model
+# download and uncompress SER model
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar & tar -xf ser_vi_layoutxlm_xfund_pretrained.tar
+
+# download and uncompress RE model
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar & tar -xf re_vi_layoutxlm_xfund_pretrained.tar
+```
+
+Start training:
+
+- If your paddlepaddle version is `CPU`, you need to set `Global.use_gpu=False` in your config file.
+- During training, PaddleOCR will download the VI-LayoutXLM pretraining model by default. There is no need to download it in advance.
+
+```bash linenums="1"
+# GPU training, support single card and multi-cards
+# The training log will be save in "{Global.save_model_dir}/train.log"
+
+# train SER model using single card
+python3 tools/train.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml
+
+# train SER model using multi-cards, you can use --gpus to assign the GPU ids.
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml
+
+# train RE model using single card
+python3 tools/train.py -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml
+```
+
+Take the SER model training as an example. After the training is started, you will see the following log output.
+
+```bash linenums="1"
+[2022/08/08 16:28:28] ppocr INFO: epoch: [1/200], global_step: 10, lr: 0.000006, loss: 1.871535, avg_reader_cost: 0.28200 s, avg_batch_cost: 0.82318 s, avg_samples: 8.0, ips: 9.71838 samples/s, eta: 0:51:59
+[2022/08/08 16:28:33] ppocr INFO: epoch: [1/200], global_step: 19, lr: 0.000018, loss: 1.461939, avg_reader_cost: 0.00042 s, avg_batch_cost: 0.32037 s, avg_samples: 6.9, ips: 21.53773 samples/s, eta: 0:37:55
+[2022/08/08 16:28:39] ppocr INFO: cur metric, precision: 0.11526348939743859, recall: 0.19776657060518732, hmean: 0.14564265817747712, fps: 34.008392345050055
+[2022/08/08 16:28:45] ppocr INFO: save best model is to ./output/ser_vi_layoutxlm_xfund_zh/best_accuracy
+[2022/08/08 16:28:45] ppocr INFO: best metric, hmean: 0.14564265817747712, precision: 0.11526348939743859, recall: 0.19776657060518732, fps: 34.008392345050055, best_epoch: 1
+[2022/08/08 16:28:51] ppocr INFO: save model in ./output/ser_vi_layoutxlm_xfund_zh/latest
+```
+
+The following information will be automatically printed.
+
+|Field | meaning|
+| :----: | :------: |
+|epoch | current iteration round|
+|iter | current iteration times|
+|lr | current learning rate|
+|loss | current loss function|
+| reader_cost | current batch data processing time|
+| batch_ Cost | total current batch time|
+|samples | number of samples in the current batch|
+|ips | number of samples processed per second|
+
+PaddleOCR supports evaluation during training. you can modify `eval_batch_step` in the config file `configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml` (default as 19 iters). Trained model with best hmean will be saved as `output/ser_vi_layoutxlm_xfund_zh/best_accuracy/`.
+
+If the evaluation dataset is very large, it's recommended to enlarge the eval interval or evaluate the model after training.
+
+**Note:** for more KIE models training and configuration files, you can go into `configs/kie/` or refer to [Frontier KIE algorithms](../../algorithm/overview.en.md).
+
+If you want to train model on your own dataset, you need to modify the data path, dictionary file and category number in the configuration file.
+
+Take `configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml` as an example, contents we need to fix is as follows.
+
+```yaml linenums="1"
+Architecture:
+  # ...
+  Backbone:
+    name: LayoutXLMForSer
+    pretrained: True
+    mode: vi
+    # Assuming that n categories are included in the dictionary file (other is included), the the num_classes is set as 2n-1
+    num_classes: &num_classes 7
+
+PostProcess:
+  name: kieSerTokenLayoutLMPostProcess
+  # Modify the dictionary file path for your custom dataset
+  class_path: &class_path train_data/XFUND/class_list_xfun.txt
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    # Modify the data path for your training dataset
+    data_dir: train_data/XFUND/zh_train/image
+    # Modify the data annotation path for your training dataset
+    label_file_list:
+      - train_data/XFUND/zh_train/train.json
+    ...
+  loader:
+    # batch size for single card when training
+    batch_size_per_card: 8
+    ...
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    # Modify the data path for your evaluation dataset
+    data_dir: train_data/XFUND/zh_val/image
+    # Modify the data annotation path for your evaluation dataset
+    label_file_list:
+      - train_data/XFUND/zh_val/val.json
+    ...
+  loader:
+    # batch size for single card when evaluation
+    batch_size_per_card: 8
+```
+
+**Note that the configuration file for prediction/evaluation must be consistent with the training file.**
+
+### 2.2. Resume Training
+
+If the training process is interrupted and you want to load the saved model to resume training, you can specify the path of the model to be loaded by specifying `Architecture.Backbone.checkpoints`.
+
+```bash linenums="1"
+python3 tools/train.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./output/ser_vi_layoutxlm_xfund_zh/best_accuracy
+```
+
+**Note:**
+
+- Priority of `Architecture.Backbone.checkpoints` is higher than`Architecture.Backbone.pretrained`. You need to set `Architecture.Backbone.checkpoints` for model finetuning, resume and evaluation. If you want to train with the NLP pretrained model, you need to set `Architecture.Backbone.pretrained` as `True` and set `Architecture.Backbone.checkpoints` as null (`null`).
+- PaddleNLP pretrained models are used here for LayoutXLM series models, the model loading and saving logic is same as those in PaddleNLP. Therefore we do not need to set `Global.pretrained_model` or `Global.checkpoints` here.
+- If you use knowledge distillation to train the LayoutXLM series models, resuming training is not supported now.
+
+### 2.3. Mixed Precision Training
+
+coming soon!
+
+### 2.4. Distributed Training
+
+During multi-machine multi-gpu training, use the `--ips` parameter to set the used machine IP address, and the `--gpus` parameter to set the used GPU ID:
+
+```bash linenums="1"
+python3 -m paddle.distributed.launch --ips="xx.xx.xx.xx,xx.xx.xx.xx" --gpus '0,1,2,3' tools/train.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml
+```
+
+**Note:** (1) When using multi-machine and multi-gpu training, you need to replace the ips value in the above command with the address of your machine, and the machines need to be able to ping each other. (2) Training needs to be launched separately on multiple machines. The command to view the ip address of the machine is `ifconfig`. (3) For more details about the distributed training speedup ratio, please refer to [Distributed Training Tutorial](../blog/distributed_training.en.md).
+
+### 2.5. Train with Knowledge Distillation
+
+Knowledge distillation is supported in PaddleOCR for KIE model training process. The configuration file is [ser_vi_layoutxlm_xfund_zh_udml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh_udml.yml). For more information, please refer to [doc](../model_compress/knowledge_distillation.en.md).
+
+**Note:** The saving and loading logic of the LayoutXLM series KIE models in PaddleOCR is consistent with PaddleNLP, so only the parameters of the student model are saved in the distillation process. If you want to use the saved model for evaluation, you need to use the configuration of the student model (the student model corresponding to the distillation file above is [ser_vi_layoutxlm_xfund_zh.yml](../../../../configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml).
+
+### 2.6. Training on other platform
+
+- Windows GPU/CPU
+The Windows platform is slightly different from the Linux platform:
+Windows platform only supports `single gpu` training and inference, specify GPU for training `set CUDA_VISIBLE_DEVICES=0`
+On the Windows platform, DataLoader only supports single-process mode, so you need to set `num_workers` to 0;
+
+- macOS
+GPU mode is not supported, you need to set `use_gpu` to False in the configuration file, and the rest of the training evaluation prediction commands are exactly the same as Linux GPU.
+
+- Linux DCU
+Running on a DCU device requires setting the environment variable `export HIP_VISIBLE_DEVICES=0,1,2,3`, and the rest of the training and evaluation prediction commands are exactly the same as the Linux GPU.
+
+## 3. Evaluation and Test
+
+### 3.1. Evaluation
+
+The trained model will be saved in `Global.save_model_dir`. When evaluation, you need to set `Architecture.Backbone.checkpoints` as your model directory. The evaluation dataset can be set by modifying the `Eval.dataset.label_file_list` field in the `configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml` file.
+
+```bash linenums="1"
+# GPU evaluation, Global.checkpoints is the weight to be tested
+python3 tools/eval.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./output/ser_vi_layoutxlm_xfund_zh/best_accuracy
+```
+
+The following information will be printed such as precision, recall, hmean and so on.
+
+```bash linenums="1"
+[2022/08/09 07:59:28] ppocr INFO: metric eval ***************
+[2022/08/09 07:59:28] ppocr INFO: precision:0.697476609016161
+[2022/08/09 07:59:28] ppocr INFO: recall:0.8861671469740634
+[2022/08/09 07:59:28] ppocr INFO: hmean:0.7805806758686339
+[2022/08/09 07:59:28] ppocr INFO: fps:17.367364606899105
+```
+
+### 3.2. Test
+
+Using the model trained by PaddleOCR, we can quickly get prediction through the following script.
+
+The default prediction image is stored in `Global.infer_img`, and the trained model weight is specified via `-o Global.checkpoints`.
+
+According to the `Global.save_model_dir` and `save_epoch_step` fields set in the configuration file, the following parameters will be saved.
+
+```text linenums="1"
+output/ser_vi_layoutxlm_xfund_zh/
+├── best_accuracy
+       ├── metric.states
+       ├── model_config.json
+       ├── model_state.pdparams
+├── best_accuracy.pdopt
+├── config.yml
+├── train.log
+├── latest
+       ├── metric.states
+       ├── model_config.json
+       ├── model_state.pdparams
+├── latest.pdopt
+```
+
+Among them, best_accuracy.*is the best model on the evaluation set; latest.* is the model of the last epoch.
+
+The configuration file for prediction must be consistent with the training file. If you finish the training process using `python3 tools/train.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml`. You can use the following command for prediction.
+
+```bash linenums="1"
+python3 tools/infer_kie_token_ser.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./output/ser_vi_layoutxlm_xfund_zh/best_accuracy Global.infer_img=./ppstructure/docs/kie/input/zh_val_42.jpg
+```
+
+The output image is as follows, which is also saved in `Global.save_res_path`.
+
+![image-20240710082046188](./images/image-20240710082046188.jpg)
+
+During the prediction process, the detection and recognition model of PP-OCRv3 will be loaded by default for information extraction of OCR. If you want to load the OCR results obtained in advance, you can use the following method to predict, and specify `Global.infer_img` as the annotation file, which contains the image path and OCR information, and specifies `Global.infer_mode` as False, indicating that the OCR inference engine is not used at this time.
+
+```bash linenums="1"
+python3 tools/infer_kie_token_ser.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./output/ser_vi_layoutxlm_xfund_zh/best_accuracy Global.infer_img=./train_data/XFUND/zh_val/val.json Global.infer_mode=False
+```
+
+For the above image, if information extraction is performed using the labeled OCR results, the prediction results are as follows.
+
+![image-20240710082059968](./images/image-20240710082046188.jpg)
+
+It can be seen that part of the detection information is more accurate, but the overall information extraction results are basically the same.
+
+In RE model prediction, the SER model result needs to be given first, so the configuration file and model weight of SER need to be loaded at the same time, as shown in the following example.
+
+```bash linenums="1"
+python3 ./tools/infer_kie_token_ser_re.py \
+  -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrain_models/re_vi_layoutxlm_udml_xfund_zh/best_accuracy/ \
+  Global.infer_img=./train_data/XFUND/zh_val/image/ \
+  -c_ser configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o_ser Architecture.Backbone.checkpoints=pretrain_models/ \
+  ser_vi_layoutxlm_udml_xfund_zh/best_accuracy/
+```
+
+The result is as follows.
+
+![image-20240710082109713](./images/image-20240710082046188.jpg)
+
+If you want to load the OCR results obtained in advance, you can use the following method to predict, and specify `Global.infer_img` as the annotation file, which contains the image path and OCR information, and specifies `Global.infer_mode` as False, indicating that the OCR inference engine is not used at this time.
+
+```bash linenums="1"
+python3 ./tools/infer_kie_token_ser_re.py \
+  -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrain_models/re_vi_layoutxlm_udml_xfund_zh/best_accuracy/ \
+  Global.infer_img=./train_data/XFUND/zh_val/val.json \
+  Global.infer_mode=False \
+  -c_ser configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o_ser Architecture.Backbone.checkpoints=pretrain_models/ser_vi_layoutxlm_udml_xfund_zh/best_accuracy/
+```
+
+`c_ser` denotes SER configurations file, `o_ser` denotes the SER model configurations that will override corresponding content in the file.
+
+The result is as follows.
+
+![image-20240710082117146](./images/image-20240710082046188.jpg)
+
+It can be seen that the re prediction results directly using the annotated OCR results are more accurate.
+
+## 4. Model inference
+
+### 4.1 Export the model
+
+The inference model (the model saved by `paddle.jit.save`) is generally a solidified model saved after the model training is completed, and is mostly used to give prediction in deployment.
+
+The model saved during the training process is the checkpoints model, which saves the parameters of the model and is mostly used to resume training.
+
+Compared with the checkpoints model, the inference model will additionally save the structural information of the model. Therefore, it is easier to deploy because the model structure and model parameters are already solidified in the inference model file, and is suitable for integration with actual systems.
+
+The SER model can be converted to the inference model using the following command.
+
+```bash linenums="1"
+# -c Set the training algorithm yml configuration file.
+# -o Set optional parameters.
+# Architecture.Backbone.checkpoints Set the training model address.
+# Global.save_inference_dir Set the address where the converted model will be saved.
+python3 tools/export_model.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./output/ser_vi_layoutxlm_xfund_zh/best_accuracy Global.save_inference_dir=./inference/ser_vi_layoutxlm
+```
+
+After the conversion is successful, there are three files in the model save directory:
+
+```text linenums="1"
+inference/ser_vi_layoutxlm/
+    ├── inference.pdiparams         # The parameter file of recognition inference model
+    ├── inference.pdiparams.info    # The parameter information of recognition inference model, which can be ignored
+    └── inference.pdmodel           # The program file of recognition
+```
+
+The RE model can be converted to the inference model using the following command.
+
+```bash linenums="1"
+# -c Set the training algorithm yml configuration file.
+# -o Set optional parameters.
+# Architecture.Backbone.checkpoints Set the training model address.
+# Global.save_inference_dir Set the address where the converted model will be saved.
+python3 tools/export_model.py -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./output/re_vi_layoutxlm_xfund_zh/best_accuracy Global.save_inference_dir=./inference/re_vi_layoutxlm
+```
+
+After the conversion is successful, there are three files in the model save directory:
+
+```text linenums="1"
+inference/re_vi_layoutxlm/
+    ├── inference.pdiparams         # The parameter file of recognition inference model
+    ├── inference.pdiparams.info    # The parameter information of recognition inference model, which can be ignored
+    └── inference.pdmodel           # The program file of recognition
+```
+
+### 4.2 Model inference
+
+The VI layoutxlm model performs reasoning based on the ser task, and can execute the following commands:
+
+Using the following command to infer the VI-LayoutXLM SER model.
+
+```bash linenums="1"
+cd ppstructure
+python3 kie/predict_kie_token_ser.py \
+  --kie_algorithm=LayoutXLM \
+  --ser_model_dir=../inference/ser_vi_layoutxlm \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx"
+```
+
+The visualized result will be saved in `./output`, which is shown as follows.
+
+![image-20240710082128694](./images/image-20240710082046188.jpg)
+
+Using the following command to infer the VI-LayoutXLM RE model.
+
+```bash linenums="1"
+cd ppstructure
+python3 kie/predict_kie_token_ser_re.py \
+  --kie_algorithm=LayoutXLM \
+  --re_model_dir=../inference/re_vi_layoutxlm \
+  --ser_model_dir=../inference/ser_vi_layoutxlm \
+  --use_visual_backbone=False \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx"
+```
+
+The visualized result will be saved in `./output`, which is shown as follows.
+
+![image-20240710082147184](./images/image-20240710082046188.jpg)
+
+## 5. FAQ
+
+Q1: After the training model is transferred to the inference model, the prediction effect is inconsistent?
+
+**A**：The problems are mostly caused by inconsistent preprocessing and postprocessing parameters when the trained model predicts and the preprocessing and postprocessing parameters when the inference model predicts. You can compare whether there are differences in preprocessing, postprocessing, and prediction in the configuration files used for training.
diff --git a/docs/version2.x/ppocr/model_train/kie.md b/docs/version2.x/ppocr/model_train/kie.md
new file mode 100644
index 0000000..7079fd1
--- /dev/null
+++ b/docs/version2.x/ppocr/model_train/kie.md
@@ -0,0 +1,469 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+# 关键信息抽取
+
+本文提供了PaddleOCR关键信息抽取的全流程指南，包括语义实体识别 (Semantic Entity Recognition) 以及关系抽取 (Relation Extraction, RE) 任务的数据准备、模型训练、调优、评估、预测，各个阶段的详细说明。
+
+## 1. 数据准备
+
+### 1.1. 准备数据集
+
+在训练信息抽取相关模型时，PaddleOCR支持以下数据格式。
+
+- `通用数据` 用于训练以文本文件存储的数据集(SimpleDataSet);
+
+训练数据的默认存储路径是 `PaddleOCR/train_data`,如果您的磁盘上已有数据集，只需创建软链接至数据集目录：
+
+```bash linenums="1"
+# linux and mac os
+ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/dataset
+# windows
+mklink /d <path/to/paddle_ocr>/train_data/dataset <path/to/dataset>
+```
+
+### 1.2. 自定义数据集
+
+训练过程中一般包含训练集与验证集，二者数据格式相同，下面介绍如何自定义数据集。
+
+#### （1）训练集
+
+建议将训练图片放入同一个文件夹，并用一个文本文件记录图片路径和标签，文本文件里的内容如下:
+
+```python linenums="1"
+" 图像文件名                 图像标注信息 "
+zh_train_0.jpg   [{"transcription": "汇丰晋信", "label": "other", "points": [[104, 114], [530, 114], [530, 175], [104, 175]], "id": 1, "linking": []}, {"transcription": "受理时间:", "label": "question", "points": [[126, 267], [266, 267], [266, 305], [126, 305]], "id": 7, "linking": [[7, 13]]}, {"transcription": "2020.6.15", "label": "answer", "points": [[321, 239], [537, 239], [537, 285], [321, 285]], "id": 13, "linking": [[7, 13]]}]
+zh_train_1.jpg   [{"transcription": "中国人体器官捐献", "label": "other", "points": [[544, 459], [954, 459], [954, 517], [544, 517]], "id": 1, "linking": []}, {"transcription": ">编号:MC545715483585", "label": "other", "points": [[1462, 470], [2054, 470], [2054, 543], [1462, 543]], "id": 10, "linking": []}, {"transcription": "CHINAORGANDONATION", "label": "other", "points": [[543, 516], [958, 516], [958, 551], [543, 551]], "id": 14, "linking": []}, {"transcription": "中国人体器官捐献志愿登记表", "label": "header", "points": [[635, 793], [1892, 793], [1892, 904], [635, 904]], "id": 18, "linking": []}]
+...
+```
+
+**注意：** 文本文件中默认请将图片路径和图片标签用 `\t` 分割，如用其他方式分割将造成训练报错。
+
+其中图像标注信息字符串经过json解析之后可以得到一个列表信息，列表中每个元素是一个字典，存储了每个文本行的需要信息，各个字段的含义如下。
+
+- transcription: 存储了文本行的文字内容
+- label: 该文本行内容所属的类别
+- points: 存储文本行的四点位置信息
+- id: 存储文本行的id信息，用于RE任务的训练
+- linking: 存储文本行的之间的连接信息，用于RE任务的训练
+
+#### （2）验证集
+
+验证集构建方式与训练集相同。
+
+#### （3）字典文件
+
+训练集与验证集中的文本行包含标签信息，所有标签的列表存在字典文件中（如`class_list.txt`），字典文件中的每一行表示为一个类别名称。
+
+以XFUND_zh数据为例，共包含4个类别，字典文件内容如下所示。
+
+```text linenums="1"
+OTHER
+QUESTION
+ANSWER
+HEADER
+```
+
+在标注文件中，每个标注的文本行内容的`label`字段标注信息需要属于字典内容。
+
+最终数据集应有如下文件结构：
+
+```text linenums="1"
+|-train_data
+  |-data_name
+    |- train.json
+    |- train
+        |- zh_train_0.png
+        |- zh_train_1.jpg
+        | ...
+    |- val.json
+    |- val
+        |- zh_val_0.png
+        |- zh_val_1.jpg
+        | ...
+```
+
+**注：**
+
+- 标注文件中的类别信息不区分大小写，如`HEADER`与`header`会被解析为相同的类别id，因此在标注的时候，不能使用小写处理后相同的字符串表示不同的类别。
+- 在整理标注文件的时候，建议将other这个类别（其他，无需关注的文本行可以标注为other）放在第一行，在解析的时候，会将`other`类别的类别id解析为0，后续不会对该类进行可视化。
+
+### 1.3. 数据下载
+
+如果你没有本地数据集，可以从[XFUND](https://github.com/doc-analysis/XFUND)或者[FUNSD](https://guillaumejaume.github.io/FUNSD/)官网下载数据，然后使用XFUND与FUNSD的处理脚本([XFUND](../../../../ppstructure/kie/tools/trans_xfun_data.py), [FUNSD](../../../../ppstructure/kie/tools/trans_funsd_label.py))，生成用于PaddleOCR训练的数据格式，并使用公开数据集快速体验关键信息抽取的流程。
+
+更多关于公开数据集的介绍，请参考[关键信息抽取数据集说明文档](../../../datasets/kie_datasets.md)。
+
+PaddleOCR也支持了关键信息抽取模型的标注，具体使用方法请参考：[PPOCRLabel使用文档](https://github.com/PFCCLab/PPOCRLabel/blob/main/README_ch.md)。
+
+## 2. 开始训练
+
+PaddleOCR提供了训练脚本、评估脚本和预测脚本，本节将以 VI-LayoutXLM 多模态预训练模型为例进行讲解。
+
+> 如果希望使用基于SDMGR的关键信息抽取算法，请参考：[SDMGR使用](../../algorithm/kie/algorithm_kie_sdmgr.md)。
+
+### 2.1. 启动训练
+
+如果你没有使用自定义数据集，可以使用PaddleOCR中已经处理好的XFUND_zh数据集进行快速体验。
+
+```bash linenums="1"
+mkdir train_data
+cd train_data
+wget https://paddleocr.bj.bcebos.com/ppstructure/dataset/XFUND.tar && tar -xf XFUND.tar
+cd ..
+```
+
+如果不希望训练，直接体验后面的模型评估、预测、动转静、推理的流程，可以下载PaddleOCR中提供的预训练模型，并跳过2.1部分。
+
+使用下面的方法，下载基于XFUND数据的SER与RE任务预训练模型。
+
+```bash linenums="1"
+mkdir pretrained_model
+cd pretrained_model
+# 下载并解压SER预训练模型
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar & tar -xf ser_vi_layoutxlm_xfund_pretrained.tar
+
+# 下载并解压RE预训练模型
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar & tar -xf re_vi_layoutxlm_xfund_pretrained.tar
+```
+
+开始训练:
+
+- 如果您安装的是cpu版本，请将配置文件中的 `use_gpu` 字段修改为false
+- PaddleOCR在训练时，会默认下载VI-LayoutXLM预训练模型，这里无需预先下载。
+
+```bash linenums="1"
+# GPU训练 支持单卡，多卡训练
+# 训练日志会自动保存到 配置文件中"{Global.save_model_dir}" 下的train.log文件中
+
+# SER单卡训练
+python3 tools/train.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml
+
+# SER多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml
+
+# RE任务单卡训练
+python3 tools/train.py -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml
+```
+
+以SER任务为例，正常启动训练后，会看到以下log输出：
+
+```bash linenums="1"
+[2022/08/08 16:28:28] ppocr INFO: epoch: [1/200], global_step: 10, lr: 0.000006, loss: 1.871535, avg_reader_cost: 0.28200 s, avg_batch_cost: 0.82318 s, avg_samples: 8.0, ips: 9.71838 samples/s, eta: 0:51:59
+[2022/08/08 16:28:33] ppocr INFO: epoch: [1/200], global_step: 19, lr: 0.000018, loss: 1.461939, avg_reader_cost: 0.00042 s, avg_batch_cost: 0.32037 s, avg_samples: 6.9, ips: 21.53773 samples/s, eta: 0:37:55
+[2022/08/08 16:28:39] ppocr INFO: cur metric, precision: 0.11526348939743859, recall: 0.19776657060518732, hmean: 0.14564265817747712, fps: 34.008392345050055
+[2022/08/08 16:28:45] ppocr INFO: save best model is to ./output/ser_vi_layoutxlm_xfund_zh/best_accuracy
+[2022/08/08 16:28:45] ppocr INFO: best metric, hmean: 0.14564265817747712, precision: 0.11526348939743859, recall: 0.19776657060518732, fps: 34.008392345050055, best_epoch: 1
+[2022/08/08 16:28:51] ppocr INFO: save model in ./output/ser_vi_layoutxlm_xfund_zh/latest
+```
+
+log 中自动打印如下信息：
+
+|  字段   |   含义   |
+| :----: | :------: |
+|  epoch | 当前迭代轮次 |
+|  iter  | 当前迭代次数 |
+|  lr    | 当前学习率 |
+|  loss  | 当前损失函数 |
+|  reader_cost | 当前 batch 数据处理耗时 |
+|  batch_cost | 当前 batch 总耗时 |
+|  samples  | 当前 batch 内的样本数 |
+|  ips  | 每秒处理图片的数量 |
+
+PaddleOCR支持训练和评估交替进行, 可以在 `configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml` 中修改 `eval_batch_step` 设置评估频率，默认每19个iter评估一次。评估过程中默认将最佳hmean模型，保存为 `output/ser_vi_layoutxlm_xfund_zh/best_accuracy/` 。
+
+如果验证集很大，测试将会比较耗时，建议减少评估次数，或训练完再进行评估。
+
+**提示：** 可通过 -c 参数选择 `configs/kie/` 路径下的多种模型配置进行训练，PaddleOCR支持的信息抽取算法可以参考[前沿算法列表](../../algorithm/overview.md)。
+
+如果你希望训练自己的数据集，需要修改配置文件中的数据配置、字典文件以及类别数。
+
+以 `configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml` 为例，修改的内容如下所示。
+
+```yaml linenums="1"
+
+Architecture:
+  # ...
+  Backbone:
+    name: LayoutXLMForSer
+    pretrained: True
+    mode: vi
+    # 由于采用BIO标注，假设字典中包含n个字段（包含other）时，则类别数为2n-1; 假设字典中包含n个字段（不含other）时，则类别数为2n+1。否则在train过程会报：IndexError: (OutOfRange) label value should less than the shape of axis dimension 。
+    num_classes: &num_classes 7
+
+PostProcess:
+  name: kieSerTokenLayoutLMPostProcess
+  # 修改字典文件的路径为你自定义的数据集的字典路径
+  class_path: &class_path train_data/XFUND/class_list_xfun.txt
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    # 修改为你自己的训练数据目录
+    data_dir: train_data/XFUND/zh_train/image
+    # 修改为你自己的训练数据标签文件
+    label_file_list:
+      - train_data/XFUND/zh_train/train.json
+    ...
+  loader:
+    # 训练时的单卡batch_size
+    batch_size_per_card: 8
+    ...
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    # 修改为你自己的验证数据目录
+    data_dir: train_data/XFUND/zh_val/image
+    # 修改为你自己的验证数据标签文件
+    label_file_list:
+      - train_data/XFUND/zh_val/val.json
+    ...
+  loader:
+    # 验证时的单卡batch_size
+    batch_size_per_card: 8
+```
+
+**注意，预测/评估时的配置文件请务必与训练一致。**
+
+### 2.2. 断点训练
+
+如果训练程序中断，如果希望加载训练中断的模型从而恢复训练，可以通过指定`Architecture.Backbone.checkpoints`指定要加载的模型路径：
+
+```bash linenums="1"
+python3 tools/train.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./output/ser_vi_layoutxlm_xfund_zh/best_accuracy
+```
+
+**注意**：
+
+- `Architecture.Backbone.checkpoints`的优先级高于`Architecture.Backbone.pretrained`，需要加载之前训练好的训练模型进行模型微调、恢复训练、模型评估时，需要使用`Architecture.Backbone.checkpoints`指定模型参数路径；如果需要使用默认提供的通用预训练模型进行训练，则需要指定`Architecture.Backbone.pretrained`为`True`，同时指定`Architecture.Backbone.checkpoints`为空（`null`）。
+- LayoutXLM系列模型均是调用了PaddleNLP中的预训练模型，模型加载与保存的逻辑与PaddleNLP基本一致，因此在这里不需要指定`Global.pretrained_model`或者`Global.checkpoints`参数；此外，LayoutXLM系列模型的蒸馏训练目前不支持断点训练。
+
+### 2.3. 混合精度训练
+
+coming soon!
+
+### 2.4. 分布式训练
+
+多机多卡训练时，通过 `--ips` 参数设置使用的机器IP地址，通过 `--gpus` 参数设置使用的GPU ID：
+
+```bash linenums="1"
+python3 -m paddle.distributed.launch --ips="xx.xx.xx.xx,xx.xx.xx.xx" --gpus '0,1,2,3' tools/train.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml
+```
+
+**注意:** （1）采用多机多卡训练时，需要替换上面命令中的ips值为您机器的地址，机器之间需要能够相互ping通；（2）训练时需要在多个机器上分别启动命令。查看机器ip地址的命令为`ifconfig`；（3）更多关于分布式训练的性能优势等信息，请参考：[分布式训练教程](../blog/distributed_training.md)。
+
+### 2.5. 知识蒸馏训练
+
+PaddleOCR支持了基于U-DML知识蒸馏的关键信息抽取模型训练过程，配置文件请参考：[ser_vi_layoutxlm_xfund_zh_udml.yml](../../../../configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh_udml.yml)，更多关于知识蒸馏的说明文档请参考：[知识蒸馏说明文档](../model_compress/knowledge_distillation.md)。
+
+**注意**： PaddleOCR中LayoutXLM系列关键信息抽取模型的保存与加载逻辑与PaddleNLP保持一致，因此在蒸馏的过程中仅保存了学生模型的参数，如果希望使用保存的模型进行评估，需要使用学生模型的配置（上面的蒸馏文件对应的学生模型为[ser_vi_layoutxlm_xfund_zh.yml](../../../../configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml)）
+
+### 2.6. 其他训练环境
+
+- Windows GPU/CPU
+在Windows平台上与Linux平台略有不同:
+Windows平台只支持`单卡`的训练与预测，指定GPU进行训练`set CUDA_VISIBLE_DEVICES=0`
+在Windows平台，DataLoader只支持单进程模式，因此需要设置 `num_workers` 为0;
+
+- macOS
+不支持GPU模式，需要在配置文件中设置`use_gpu`为False，其余训练评估预测命令与Linux GPU完全相同。
+
+- Linux DCU
+DCU设备上运行需要设置环境变量 `export HIP_VISIBLE_DEVICES=0,1,2,3`，其余训练评估预测命令与Linux GPU完全相同。
+
+## 3. 模型评估与预测
+
+### 3.1. 指标评估
+
+训练中模型参数默认保存在`Global.save_model_dir`目录下。在评估指标时，需要设置`Architecture.Backbone.checkpoints`指向保存的参数文件。评估数据集可以通过 `configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml`  修改Eval中的 `label_file_path` 设置。
+
+```bash linenums="1"
+# GPU 评估， Global.checkpoints 为待测权重
+python3 tools/eval.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./output/ser_vi_layoutxlm_xfund_zh/best_accuracy
+```
+
+会输出以下信息，打印出precision、recall、hmean等信息。
+
+```bash linenums="1"
+[2022/08/09 07:59:28] ppocr INFO: metric eval ***************
+[2022/08/09 07:59:28] ppocr INFO: precision:0.697476609016161
+[2022/08/09 07:59:28] ppocr INFO: recall:0.8861671469740634
+[2022/08/09 07:59:28] ppocr INFO: hmean:0.7805806758686339
+[2022/08/09 07:59:28] ppocr INFO: fps:17.367364606899105
+```
+
+### 3.2. 测试信息抽取结果
+
+使用 PaddleOCR 训练好的模型，可以通过以下脚本进行快速预测。
+
+默认预测的图片存储在 `infer_img` 里，通过 `-o Architecture.Backbone.checkpoints` 加载训练好的参数文件：
+
+根据配置文件中设置的 `save_model_dir` 和 `save_epoch_step` 字段，会有以下几种参数被保存下来：
+
+```text linenums="1"
+output/ser_vi_layoutxlm_xfund_zh/
+├── best_accuracy
+       ├── metric.states
+       ├── model_config.json
+       ├── model_state.pdparams
+├── best_accuracy.pdopt
+├── config.yml
+├── train.log
+├── latest
+       ├── metric.states
+       ├── model_config.json
+       ├── model_state.pdparams
+├── latest.pdopt
+```
+
+其中 best_accuracy.*是评估集上的最优模型；latest.* 是最新保存的一个模型。
+
+预测使用的配置文件必须与训练一致，如您通过 `python3 tools/train.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml` 完成了模型的训练过程。
+
+您可以使用如下命令进行中文模型预测。
+
+```bash linenums="1"
+python3 tools/infer_kie_token_ser.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./output/ser_vi_layoutxlm_xfund_zh/best_accuracy Global.infer_img=./ppstructure/docs/kie/input/zh_val_42.jpg
+```
+
+预测图片如下所示，图片会存储在`Global.save_res_path`路径中。
+
+![image-20240710082046188](./images/image-20240710082046188.jpg)
+
+预测过程中，默认会加载PP-OCRv3的检测识别模型，用于OCR的信息抽取，如果希望加载预先获取的OCR结果，可以使用下面的方式进行预测，指定`Global.infer_img`为标注文件，其中包含图片路径以及OCR信息，同时指定`Global.infer_mode`为False，表示此时不使用OCR预测引擎。
+
+```bash linenums="1"
+python3 tools/infer_kie_token_ser.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./output/ser_vi_layoutxlm_xfund_zh/best_accuracy Global.infer_img=./train_data/XFUND/zh_val/val.json Global.infer_mode=False
+```
+
+对于上述图片，如果使用标注的OCR结果进行信息抽取，预测结果如下。
+
+![image-20240710082059968](./images/image-20240710082046188.jpg)
+
+可以看出，部分检测框信息更加准确，但是整体信息抽取识别结果基本一致。
+
+在RE任务模型预测时，需要先给出模型SER结果，因此需要同时加载SER的配置文件与模型权重，示例如下。
+
+```bash linenums="1"
+python3 ./tools/infer_kie_token_ser_re.py \
+  -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrain_models/re_vi_layoutxlm_udml_xfund_zh/best_accuracy/ \
+  Global.infer_img=./train_data/XFUND/zh_val/image/ \
+  -c_ser configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o_ser Architecture.Backbone.checkpoints=pretrain_models/ \
+  ser_vi_layoutxlm_udml_xfund_zh/best_accuracy/
+```
+
+预测结果如下所示。
+
+![image-20240710082109713](./images/image-20240710082046188.jpg)
+
+如果希望使用标注或者预先获取的OCR信息进行关键信息抽取，同上，可以指定`Global.infer_mode`为False，指定`Global.infer_img`为标注文件。
+
+```bash linenums="1"
+python3 ./tools/infer_kie_token_ser_re.py -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./pretrain_models/re_vi_layoutxlm_udml_xfund_zh/re_layoutxlm_xfund_zh_v4_udml/best_accuracy/ Global.infer_img=./train_data/XFUND/zh_val/val.json Global.infer_mode=False -c_ser configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o_ser Architecture.Backbone.checkpoints=pretrain_models/ser_vi_layoutxlm_udml_xfund_zh/best_accuracy/
+```
+
+其中`c_ser`表示SER的配置文件，`o_ser` 后面需要加上待修改的SER模型与配置文件，如预训练权重等。
+
+预测结果如下所示。
+
+![image-20240710082117146](./images/image-20240710082046188.jpg)
+
+可以看出，直接使用标注的OCR结果的RE预测结果要更加准确一些。
+
+## 4. 模型导出与预测
+
+### 4.1 模型导出
+
+inference 模型（`paddle.jit.save`保存的模型）
+一般是模型训练，把模型结构和模型参数保存在文件中的固化模型，多用于预测部署场景。
+训练过程中保存的模型是checkpoints模型，保存的只有模型的参数，多用于恢复训练等。
+与checkpoints模型相比，inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成。
+
+信息抽取模型中的SER任务转inference模型步骤如下：
+
+```bash linenums="1"
+# -c 后面设置训练算法的yml配置文件
+# -o 配置可选参数
+# Architecture.Backbone.checkpoints 参数设置待转换的训练模型地址
+# Global.save_inference_dir 参数设置转换的模型将保存的地址
+
+python3 tools/export_model.py -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./output/ser_vi_layoutxlm_xfund_zh/best_accuracy Global.save_inference_dir=./inference/ser_vi_layoutxlm
+```
+
+转换成功后，在目录下有三个文件：
+
+```text linenums="1"
+inference/ser_vi_layoutxlm/
+    ├── inference.pdiparams         # inference模型的参数文件
+    ├── inference.pdiparams.info    # inference模型的参数信息，可忽略
+    └── inference.pdmodel           # inference模型的模型结构文件
+```
+
+信息抽取模型中的RE任务转inference模型步骤如下：
+
+```bash linenums="1"
+# -c 后面设置训练算法的yml配置文件
+# -o 配置可选参数
+# Architecture.Backbone.checkpoints 参数设置待转换的训练模型地址
+# Global.save_inference_dir 参数设置转换的模型将保存的地址
+
+python3 tools/export_model.py -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml -o Architecture.Backbone.checkpoints=./output/re_vi_layoutxlm_xfund_zh/best_accuracy Global.save_inference_dir=./inference/re_vi_layoutxlm
+```
+
+转换成功后，在目录下有三个文件：
+
+```text linenums="1"
+inference/re_vi_layoutxlm/
+    ├── inference.pdiparams         # inference模型的参数文件
+    ├── inference.pdiparams.info    # inference模型的参数信息，可忽略
+    └── inference.pdmodel           # inference模型的模型结构文件
+```
+
+### 4.2 模型推理
+
+VI-LayoutXLM模型基于SER任务进行推理，可以执行如下命令：
+
+```bash linenums="1"
+cd ppstructure
+python3 kie/predict_kie_token_ser.py \
+  --kie_algorithm=LayoutXLM \
+  --ser_model_dir=../inference/ser_vi_layoutxlm \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx"
+```
+
+可视化SER结果结果默认保存到`./output`文件夹里面。结果示例如下：
+
+![image-20240710082128694](./images/image-20240710082046188.jpg)
+
+VI-LayoutXLM模型基于RE任务进行推理，可以执行如下命令：
+
+```bash linenums="1"
+cd ppstructure
+python3 kie/predict_kie_token_ser_re.py \
+  --kie_algorithm=LayoutXLM \
+  --re_model_dir=../inference/re_vi_layoutxlm \
+  --ser_model_dir=../inference/ser_vi_layoutxlm \
+  --use_visual_backbone=False \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx"
+```
+
+RE可视化结果默认保存到`./output`文件夹里面，结果示例如下：
+
+![image-20240710082147184](./images/image-20240710082046188.jpg)
+
+## 5. FAQ
+
+Q1: 训练模型转inference 模型之后预测效果不一致？
+
+**A**：该问题多是trained model预测时候的预处理、后处理参数和inference model预测的时候的预处理、后处理参数不一致导致的。可以对比训练使用的配置文件中的预处理、后处理和预测时是否存在差异。
diff --git a/docs/version2.x/ppocr/model_train/recognition.en.md b/docs/version2.x/ppocr/model_train/recognition.en.md
new file mode 100644
index 0000000..a0c3762
--- /dev/null
+++ b/docs/version2.x/ppocr/model_train/recognition.en.md
@@ -0,0 +1,558 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+# Text Recognition
+
+This article provides a comprehensive guide for the PaddleOCR text recognition task, covering the entire workflow including data preparation, model training, fine-tuning, evaluation, and prediction, with detailed explanations for each phase.
+
+## 1. Data Preparation
+
+### 1.1. Prepare the Dataset
+
+PaddleOCR supports two data formats:
+
+- `lmdb`: Used for training with datasets stored in LMDB format (LMDBDataSet);
+- `General Data`: Used for training with datasets stored in text files (SimpleDataSet);
+
+The default storage path for training data is `PaddleOCR/train_data`. If you already have a dataset on your disk, simply create a symbolic link to the dataset directory:
+
+```bash
+# Linux and macOS
+ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/dataset
+# Windows
+mklink /d <path/to/paddle_ocr>/train_data/dataset <path/to/dataset>
+```
+
+### 1.2. Custom Dataset
+
+Here, we will use a general dataset as an example to explain how to prepare the dataset:
+
+- Training Dataset
+
+It is recommended to place the training images in the same folder and record the image paths and labels in a txt file (`rec_gt_train.txt`). The content of the txt file should be as follows:
+
+**Note:** In the txt file, please use `\t` to separate the image path and the label. Using any other separator will cause errors during training.
+
+```text
+" Image Filename                   Image Label "
+
+train_data/rec/train/word_001.jpg   Simple and reliable
+train_data/rec/train/word_002.jpg   Making the complex world simpler with technology
+...
+```
+
+The final structure of the training dataset should look like this:
+
+```text
+|-train_data
+  |-rec
+    |- rec_gt_train.txt
+    |- train
+        |- word_001.png
+        |- word_002.jpg
+        |- word_003.jpg
+        | ...
+```
+
+In addition to the single-image-per-line format described above, PaddleOCR also supports training on data augmented offline. To avoid sampling the same sample multiple times in the same batch, we can list image paths with the same label on one line. During training, PaddleOCR will randomly select one image from the list. The corresponding format of the label file is as follows:
+
+```text
+["11.jpg", "12.jpg"]   Simple and reliable
+["21.jpg", "22.jpg", "23.jpg"]   Making the complex world simpler with technology
+3.jpg   ocr
+```
+
+In the above example, both "11.jpg" and "12.jpg" have the same label `Simple and reliable`. During training, one of these images will be randomly chosen for training.
+
+- Validation Dataset
+
+Similarly to the training dataset, the validation dataset should also provide a folder containing all the images (test) and a `rec_gt_test.txt` file. The structure of the validation dataset is as follows:
+
+```text
+|-train_data
+  |-rec
+    |- rec_gt_test.txt
+    |- test
+        |- word_001.jpg
+        |- word_002.jpg
+        |- word_003.jpg
+        | ...
+```
+
+### 1.3. Data Download
+
+- ICDAR2015
+
+If you don't have a dataset locally, you can download the [ICDAR2015](http://rrc.cvc.uab.es/?ch=4&com=downloads) dataset from the official website for quick testing. You can also refer to [DTRB](https://github.com/clovaai/deep-text-recognition-benchmark#download-lmdb-dataset-for-traininig-and-evaluation-from-here) to download the LMDB formatted dataset needed for benchmarking.
+
+If you're using the public ICDAR2015 dataset, PaddleOCR provides a label file for training the ICDAR2015 dataset. You can download it as follows:
+
+```bash linenums="1"
+# Training set label
+wget -P ./train_data/ic15_data  https://paddleocr.bj.bcebos.com/dataset/rec_gt_train.txt
+# Test Set Label
+wget -P ./train_data/ic15_data  https://paddleocr.bj.bcebos.com/dataset/rec_gt_test.txt
+```
+
+PaddleOCR also provides a data format conversion script, which can convert ICDAR official website label to a data format
+supported by PaddleOCR. The data conversion tool is in `ppocr/utils/gen_label.py`, here is the training set as an example:
+
+```bash linenums="1"
+# convert the official gt to rec_gt_label.txt
+python gen_label.py --mode="rec" --input_path="{path/of/origin/label}" --output_label="rec_gt_label.txt"
+```
+
+The data format is as follows, (a) is the original picture, (b) is the Ground Truth text file corresponding to each picture:
+
+![img](./images/icdar_rec.png)
+
+- Multilingual Datasets
+
+The multi-language model training method is the same as the Chinese model. The training data set is 100w synthetic data. A small amount of fonts and test data can be downloaded using the following two methods.
+
+- [Baidu Netdisk](https://pan.baidu.com/s/1bS_u207Rm7YbY33wOECKDA) ,Extraction code:frgi.
+
+### 1.4. Dictionary
+
+Finally, a dictionary ({word_dict_name}.txt) needs to be provided so that when the model is trained, all the characters that appear can be mapped to the dictionary index.
+
+Therefore, the dictionary needs to contain all the characters that you want to be recognized correctly. {word_dict_name}.txt needs to be written in the following format and saved in the `utf-8` encoding format:
+
+```text linenums="1"
+l
+d
+a
+d
+r
+n
+```
+
+In `word_dict.txt`, there is a single word in each line, which maps characters and numeric indexes together, e.g "and" will be mapped to [2 5 1]
+
+PaddleOCR includes several built-in dictionaries that can be used as needed:
+
+- `ppocr/utils/ppocr_keys_v1.txt`: A Chinese dictionary containing 6623 characters.
+- `ppocr/utils/ic15_dict.txt`: An English dictionary containing 36 characters.
+- `ppocr/utils/dict/french_dict.txt`: A French dictionary containing 118 characters.
+- `ppocr/utils/dict/japan_dict.txt`: A Japanese dictionary containing 4399 characters.
+- `ppocr/utils/dict/korean_dict.txt`: A Korean dictionary containing 3636 characters.
+- `ppocr/utils/dict/german_dict.txt`: A German dictionary containing 131 characters.
+- `ppocr/utils/en_dict.txt`: An English dictionary containing 96 characters.
+
+Currently, the multilingual models are still in the demo stage, and we are continuously improving the models and adding new languages. **We highly welcome you to provide dictionaries and fonts for other languages**. If you are willing, you can submit your dictionary files to the [dict](../../../../ppocr/utils/dict) directory, and we will credit you in the repo.
+To customize the dict file, please modify the `character_dict_path` field in `configs/rec/rec_icdar15_train.yml`.
+
+- Custom Dictionary
+
+If you need to customize dic file, please add character_dict_path field in configs/rec/rec_icdar15_train.yml to point to your dictionary path. And set character_type to ch.
+
+### 1.5. Add Space Category
+
+To support recognition of the "space" category, set the `use_space_char` field in the YAML file to `True`.
+
+### 1.6. Data Augmentation
+
+PaddleOCR provides a variety of data augmentation methods. All the augmentation methods are enabled by default.
+
+The default perturbation methods are: cvtColor, blur, jitter, Gasuss noise, random crop, perspective, color reverse, TIA augmentation.
+
+Each disturbance method is selected with a 40% probability during the training process. For specific code implementation, please refer to: [rec_img_aug.py](../../../../ppocr/data/imaug/rec_img_aug.py)
+
+## 2. Training
+
+PaddleOCR provides training scripts, evaluation scripts, and prediction scripts. This section will take the PP-OCRv3 English recognition model as an example:
+
+### 2.1. Start Training
+
+First download the pretrain model, you can download the trained model to finetune on the icdar2015 data:
+
+```bash linenums="1"
+cd PaddleOCR/
+# Download the pre-trained model of en_PP-OCRv3
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_train.tar
+# Decompress model parameters
+cd pretrain_models
+tar -xf en_PP-OCRv3_rec_train.tar && rm -rf en_PP-OCRv3_rec_train.tar
+```
+
+Start training:
+
+```bash linenums="1"
+# GPU training Support single card and multi-card training
+# Training icdar15 English data and The training log will be automatically saved as train.log under "{save_model_dir}"
+
+#specify the single card training(Long training time, not recommended)
+python3 tools/train.py -c configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml -o Global.pretrained_model=en_PP-OCRv3_rec_train/best_accuracy
+
+#specify the card number through --gpus
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml -o Global.pretrained_model=en_PP-OCRv3_rec_train/best_accuracy
+```
+
+PaddleOCR supports alternating training and evaluation. You can modify `eval_batch_step` in `configs/rec/rec_icdar15_train.yml` to set the evaluation frequency. By default, it is evaluated every 500 iter and the best acc model is saved under `output/rec_CRNN/best_accuracy` during the evaluation process.
+
+If the evaluation set is large, the test will be time-consuming. It is recommended to reduce the number of evaluations, or evaluate after training.
+
+- Tip: You can use the `-c` parameter to select multiple model configurations under the `configs/rec/` path for training. The recognition algorithms supported at [rec_algorithm](../../algorithm/overview.en.md):
+
+For training Chinese data, it is recommended to use
+[PP-OCRv3_mobile_rec_distillation.yml](../../../../configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec_distillation.yml). If you want to try the result of other algorithms on the Chinese data set, please refer to the following instructions to modify the configuration file:
+
+Take `PP-OCRv3_mobile_rec_distillation.yml` as an example:
+
+```yaml linenums="1"
+Global:
+  ...
+  # Add a custom dictionary, such as modify the dictionary, please point the path to the new dictionary
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  # Modify character type
+  ...
+  # Whether to recognize spaces
+  use_space_char: True
+
+
+Optimizer:
+  ...
+  # Add learning rate decay strategy
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  ...
+
+...
+
+Train:
+  dataset:
+    # Type of dataset，we support LMDBDataSet and SimpleDataSet
+    name: SimpleDataSet
+    # Path of dataset
+    data_dir: ./train_data/
+    # Path of train list
+    label_file_list: ["./train_data/train_list.txt"]
+    transforms:
+      ...
+      - RecResizeImg:
+          # Modify image_shape to fit long text
+          image_shape: [3, 48, 320]
+      ...
+  loader:
+    ...
+    # Train batch_size for Single card
+    batch_size_per_card: 256
+    ...
+
+Eval:
+  dataset:
+    # Type of dataset，we support LMDBDataSet and SimpleDataSet
+    name: SimpleDataSet
+    # Path of dataset
+    data_dir: ./train_data
+    # Path of eval list
+    label_file_list: ["./train_data/val_list.txt"]
+    transforms:
+      ...
+      - RecResizeImg:
+          # Modify image_shape to fit long text
+          image_shape: [3, 48, 320]
+      ...
+  loader:
+    # Eval batch_size for Single card
+    batch_size_per_card: 256
+    ...
+```
+
+**Note that the configuration file for prediction/evaluation must be consistent with the training.**
+
+### 2.2 Load Trained Model and Continue Training
+
+If you expect to load trained model and continue the training again, you can specify the parameter `Global.checkpoints` as the model path to be loaded.
+
+For example:
+
+```bash linenums="1"
+python3 tools/train.py -c configs/rec/rec_icdar15_train.yml -o Global.checkpoints=./your/trained/model
+```
+
+**Note**: The priority of `Global.checkpoints` is higher than that of `Global.pretrained_model`, that is, when two parameters are specified at the same time, the model specified by `Global.checkpoints` will be loaded first. If the model path specified by `Global.checkpoints` is wrong, the one specified by `Global.pretrained_model` will be loaded.
+
+### 2.3 Training with New Backbone
+
+The network part completes the construction of the network, and PaddleOCR divides the network into four parts, which are under [ppocr/modeling](../../../../ppocr/modeling). The data entering the network will pass through these four parts in sequence(transforms->backbones->
+necks->heads).
+
+```bash linenums="1"
+├── architectures # Code for building network
+├── transforms    # Image Transformation Module
+├── backbones     # Feature extraction module
+├── necks         # Feature enhancement module
+└── heads         # Output module
+```
+
+If the Backbone to be replaced has a corresponding implementation in PaddleOCR, you can directly modify the parameters in the `Backbone` part of the configuration yml file.
+
+However, if you want to use a new Backbone, an example of replacing the backbones is as follows:
+
+1. Create a new file under the [ppocr/modeling/backbones](../../../../ppocr/modeling/backbones) folder, such as my_backbone.py.
+2. Add code in the my_backbone.py file, the sample code is as follows:
+
+```python linenums="1"
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+
+class MyBackbone(nn.Layer):
+    def __init__(self, *args, **kwargs):
+        super(MyBackbone, self).__init__()
+        # your init code
+        self.conv = nn.xxxx
+
+    def forward(self, inputs):
+        # your network forward
+        y = self.conv(inputs)
+        return y
+```
+
+3. Import the added module in the [ppocr/modeling/backbones/\__init\__.py](../../../../ppocr/modeling/backbones/__init__.py) file.
+
+After adding the four-part modules of the network, you only need to configure them in the configuration file to use, such as:
+
+```yaml linenums="1"
+  Backbone:
+    name: MyBackbone
+    args1: args1
+```
+
+**NOTE**: More details about replace Backbone and other module can be found in [doc](../../algorithm/add_new_algorithm.en.md).
+
+### 2.4. Mixed Precision Training
+
+If you want to speed up your training further, you can use [Auto Mixed Precision Training](https://www.paddlepaddle.org.cn/documentation/docs/en/guides/performance_improving/amp_en.html), taking a single machine and a single gpu as an example, the commands are as follows:
+
+```bash linenums="1"
+python3 tools/train.py -c configs/rec/rec_icdar15_train.yml \
+     -o Global.pretrained_model=./pretrain_models/rec_mv3_none_bilstm_ctc_v2.0_train \
+     Global.use_amp=True Global.scale_loss=1024.0 Global.use_dynamic_loss_scaling=True
+```
+
+### 2.5. Distributed Training
+
+During multi-machine multi-gpu training, use the `--ips` parameter to set the used machine IP address, and the `--gpus` parameter to set the used GPU ID:
+
+```bash linenums="1"
+python3 -m paddle.distributed.launch --ips="xx.xx.xx.xx,xx.xx.xx.xx" --gpus '0,1,2,3' tools/train.py -c configs/rec/rec_icdar15_train.yml \
+     -o Global.pretrained_model=./pretrain_models/rec_mv3_none_bilstm_ctc_v2.0_train
+```
+
+**Note:**
+1. When using multi-machine and multi-gpu training, you need to replace the ips value in the above command with the address of your machine, and the machines need to be able to ping each other.
+2. Training needs to be launched separately on multiple machines. The command to view the ip address of the machine is `ifconfig`. 
+3. For more details about the distributed training speedup ratio, please refer to [Distributed Training Tutorial](../blog/distributed_training.en.md).
+
+### 2.6. Training with Knowledge Distillation
+
+Knowledge distillation is supported in PaddleOCR for text recognition training process. For more details, please refer to [doc](../model_compress/knowledge_distillation.en.md).
+
+### 2.7. Multi-Language Model Training
+
+Currently, the multi-language algorithms supported by PaddleOCR are:
+
+| Configuration file |  Algorithm name |   backbone |   trans   |   seq      |     pred     |  language |
+| :--------: |  :-------:   | :-------:  |   :-------:   |   :-----:   |  :-----:   | :-----:  |
+| rec_chinese_cht_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | chinese traditional  |
+| rec_en_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | English(Case sensitive)   |
+| rec_french_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | French |
+| rec_ger_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | German   |
+| rec_japan_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | Japanese |
+| rec_korean_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | Korean  |
+| rec_latin_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | Latin  |
+| rec_arabic_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | arabic |
+| rec_cyrillic_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | cyrillic   |
+| rec_devanagari_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | devanagari  |
+
+For more supported languages, please refer to : [Multi-language model](../blog/multi_languages.en.md)
+
+If you want to finetune on the basis of the existing model effect, please refer to the following instructions to modify the configuration file:
+
+Take `rec_french_lite_train` as an example:
+
+```yaml linenums="1"
+Global:
+  ...
+  # Add a custom dictionary, such as modify the dictionary, please point the path to the new dictionary
+  character_dict_path: ./ppocr/utils/dict/french_dict.txt
+  ...
+  # Whether to recognize spaces
+  use_space_char: True
+
+...
+
+Train:
+  dataset:
+    # Type of dataset，we support LMDBDataSet and SimpleDataSet
+    name: SimpleDataSet
+    # Path of dataset
+    data_dir: ./train_data/
+    # Path of train list
+    label_file_list: ["./train_data/french_train.txt"]
+    ...
+
+Eval:
+  dataset:
+    # Type of dataset，we support LMDBDataSet and SimpleDataSet
+    name: SimpleDataSet
+    # Path of dataset
+    data_dir: ./train_data
+    # Path of eval list
+    label_file_list: ["./train_data/french_val.txt"]
+    ...
+```
+
+### 2.8 Training on other platform(Windows/macOS/Linux DCU)
+
+- Windows GPU/CPU
+The Windows platform is slightly different from the Linux platform:
+Windows platform only supports `single gpu` training and inference, specify GPU for training `set CUDA_VISIBLE_DEVICES=0`
+On the Windows platform, DataLoader only supports single-process mode, so you need to set `num_workers` to 0;
+
+- macOS
+GPU mode is not supported, you need to set `use_gpu` to False in the configuration file, and the rest of the training evaluation prediction commands are exactly the same as Linux GPU.
+
+- Linux DCU
+Running on a DCU device requires setting the environment variable `export HIP_VISIBLE_DEVICES=0,1,2,3`, and the rest of the training and evaluation prediction commands are exactly the same as the Linux GPU.
+
+## 2.9 Fine-tuning
+
+In actual use, it is recommended to load the official pre-trained model and fine-tune it in your own data set. For the fine-tuning method of the recognition model, please refer to: [Model Fine-tuning Tutorial](./finetune.en.md).
+
+## 3. Evaluation and Test
+
+### 3.1. Evaluation
+
+The model parameters during training are saved in the `Global.save_model_dir` directory by default. When evaluating indicators, you need to set `Global.checkpoints` to point to the saved parameter file. The evaluation dataset can be set by modifying the `Eval.dataset.label_file_list` field in the `configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml` file.
+
+```bash linenums="1"
+# GPU evaluation, Global.checkpoints is the weight to be tested
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml -o Global.checkpoints={path/to/weights}/best_accuracy
+```
+
+### 3.2 Test
+
+Using the model trained by paddleocr, you can quickly get prediction through the following script.
+
+The default prediction picture is stored in `infer_img`, and the trained weight is specified via `-o Global.checkpoints`:
+
+According to the `save_model_dir` and `save_epoch_step` fields set in the configuration file, the following parameters will be saved:
+
+```text linenums="1"
+output/rec/
+├── best_accuracy.pdopt
+├── best_accuracy.pdparams
+├── best_accuracy.states
+├── config.yml
+├── iter_epoch_3.pdopt
+├── iter_epoch_3.pdparams
+├── iter_epoch_3.states
+├── latest.pdopt
+├── latest.pdparams
+├── latest.states
+└── train.log
+```
+
+Among them, best_accuracy._is the best model on the evaluation set; iter_epoch_x._ is the model saved at intervals of `save_epoch_step`; latest.* is the model of the last epoch.
+
+```bash linenums="1"
+# Predict English results
+python3 tools/infer_rec.py -c configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml -o Global.pretrained_model={path/to/weights}/best_accuracy  Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+Input image:
+
+![img](./images/word_1-20240704092705543.png)
+
+Get the prediction result of the input image:
+
+```bash linenums="1"
+infer_img: doc/imgs_words/en/word_1.png
+        result: ('joint', 0.9998967)
+```
+
+The configuration file used for prediction must be consistent with the training. For example, you completed the training of the Chinese model with `python3 tools/train.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml`, you can use the following command to predict the Chinese model:
+
+```bash linenums="1"
+# Predict Chinese results
+python3 tools/infer_rec.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/ch/word_1.jpg
+```
+
+Input image:
+
+![img](./images/word_1-20240704092713071.jpg)
+
+Get the prediction result of the input image:
+
+```bash linenums="1"
+infer_img: doc/imgs_words/ch/word_1.jpg
+        result: ('韩国小馆', 0.997218)
+```
+
+### 4. Model Export and Prediction
+
+**Inference Model** (saved using `paddle.jit.save`)
+
+The inference model is a "frozen" version of the model, where both the model structure and model parameters are saved in a file. It is typically used for prediction and deployment scenarios.  
+In contrast, the **checkpoint model** only saves the model's parameters and is mostly used for training resumption, etc. Compared to the checkpoint model, the inference model also includes the model structure information, which makes it more efficient for deployment, inference acceleration, and flexible integration with systems.
+
+The process of converting a recognition model to an inference model is similar to the detection model conversion, as shown below:
+
+```bash linenums="1"
+# Enable old IR mode
+export FLAGS_enable_pir_api=0
+
+# -c Set the training algorithm yml configuration file
+# -o Set optional parameters
+# Global.pretrained_model parameter Set the training model address to be converted without adding the file suffix .pdmodel, .pdopt or .pdparams.
+# Global.save_inference_dir Set the address where the converted model will be saved.
+
+python3 tools/export_model.py -c configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml -o Global.pretrained_model=en_PP-OCRv3_rec_train/best_accuracy  Global.save_inference_dir=./inference/en_PP-OCRv3_mobile_rec/
+```
+
+If you have a model trained on your own dataset with a different dictionary file, please make sure that you modify the `character_dict_path` in the configuration file to your dictionary file path.
+
+After the conversion is successful, there are three files in the model save directory:
+
+```text linenums="1"
+inference/en_PP-OCRv3_mobile_rec/
+    ├── inference.pdiparams         # The parameter file of recognition inference model
+    ├── inference.pdiparams.info    # The parameter information of recognition inference model, which can be ignored
+    └── inference.pdmodel           # The program file of recognition model
+```
+
+**Note**: If you need to store the model in the new IR mode (i.e., `.json` format), use the following command to switch to the new IR mode:
+
+```bash
+export FLAGS_enable_pir_api=1
+python3 tools/export_model.py -c configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml -o Global.pretrained_model=./pretrain_models/en_PP-OCRv3_rec_train/best_accuracy Global.save_inference_dir=./inference/en_PP-OCRv3_mobile_rec/
+```
+
+Once successful, you will have two files in the directory:
+
+```text
+inference/en_PP-OCRv3_mobile_rec/
+    ├── inference.pdiparams         # Model parameter file for the inference model
+    └── inference.json              # Program file for the inference model
+```
+
+### Custom Model Inference
+
+If you modified the text dictionary during training, you must specify the path to the custom dictionary when using the inference model for prediction. For more information about configuring and explaining inference hyperparameters, refer to the [Inference Hyperparameters Explanation Tutorial](../blog/inference_args.md).
+
+```bash linenums="1"
+python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_336.png" --rec_model_dir="./your_inference_model" --rec_image_shape="3, 48, 320" --rec_char_dict_path="your_text_dict_path"
+```
+
+### 5. FAQ
+
+**Q1:** Why is the prediction result inconsistent after converting a trained model to an inference model?
+
+**A**: This is a common issue. It typically arises due to differences in the preprocessing and postprocessing parameters used during training and inference. To troubleshoot, check whether the preprocessing, postprocessing, and prediction settings in the configuration file used for training match those used during inference.
diff --git a/docs/version2.x/ppocr/model_train/recognition.md b/docs/version2.x/ppocr/model_train/recognition.md
new file mode 100644
index 0000000..9cfff31
--- /dev/null
+++ b/docs/version2.x/ppocr/model_train/recognition.md
@@ -0,0 +1,585 @@
+---
+comments: true
+typora-copy-images-to: images
+---
+
+# 文字识别
+
+本文提供了PaddleOCR文本识别任务的全流程指南，包括数据准备、模型训练、调优、评估、预测，各个阶段的详细说明：
+
+## 1. 数据准备
+
+### 1.1. 准备数据集
+
+PaddleOCR 支持两种数据格式:
+
+- `lmdb` 用于训练以lmdb格式存储的数据集(LMDBDataSet);
+- `通用数据` 用于训练以文本文件存储的数据集(SimpleDataSet);
+
+训练数据的默认存储路径是 `PaddleOCR/train_data`,如果您的磁盘上已有数据集，只需创建软链接至数据集目录：
+
+```bash linenums="1"
+# linux and mac os
+ln -sf <path/to/dataset> <path/to/paddle_ocr>/train_data/dataset
+# windows
+mklink /d <path/to/paddle_ocr>/train_data/dataset <path/to/dataset>
+```
+
+### 1.2. 自定义数据集
+
+下面以通用数据集为例， 介绍如何准备数据集：
+
+- 训练集
+
+建议将训练图片放入同一个文件夹，并用一个txt文件（rec_gt_train.txt）记录图片路径和标签，txt文件里的内容如下:
+
+**注意：** txt文件中默认请将图片路径和图片标签用 \t 分割，如用其他方式分割将造成训练报错。
+
+```text linenums="1"
+" 图像文件名                 图像标注信息 "
+
+train_data/rec/train/word_001.jpg   简单可依赖
+train_data/rec/train/word_002.jpg   用科技让复杂的世界更简单
+...
+```
+
+最终训练集应有如下文件结构：
+
+```text linenums="1"
+|-train_data
+  |-rec
+    |- rec_gt_train.txt
+    |- train
+        |- word_001.png
+        |- word_002.jpg
+        |- word_003.jpg
+        | ...
+```
+
+除上述单张图像为一行格式之外，PaddleOCR也支持对离线增广后的数据进行训练，为了防止相同样本在同一个batch中被多次采样，我们可以将相同标签对应的图片路径写在一行中，以列表的形式给出，在训练中，PaddleOCR会随机选择列表中的一张图片进行训练。对应地，标注文件的格式如下：
+
+```text linenums="1"
+["11.jpg", "12.jpg"]   简单可依赖
+["21.jpg", "22.jpg", "23.jpg"]   用科技让复杂的世界更简单
+3.jpg   ocr
+```
+
+上述示例标注文件中，"11.jpg"和"12.jpg"的标签相同，都是`简单可依赖`，在训练的时候，对于该行标注，会随机选择其中的一张图片进行训练。
+
+- 验证集
+
+同训练集类似，验证集也需要提供一个包含所有图片的文件夹（test）和一个rec_gt_test.txt，验证集的结构如下所示：
+
+```text linenums="1"
+|-train_data
+  |-rec
+    |- rec_gt_test.txt
+    |- test
+        |- word_001.jpg
+        |- word_002.jpg
+        |- word_003.jpg
+        | ...
+```
+
+### 1.3. 数据下载
+
+- ICDAR2015
+
+若您本地没有数据集，可以在官网下载 [ICDAR2015](http://rrc.cvc.uab.es/?ch=4&com=downloads) 数据，用于快速验证。也可以参考[DTRB](https://github.com/clovaai/deep-text-recognition-benchmark#download-lmdb-dataset-for-traininig-and-evaluation-from-here) ，下载 benchmark 所需的lmdb格式数据集。
+
+如果你使用的是icdar2015的公开数据集，PaddleOCR 提供了一份用于训练 ICDAR2015 数据集的标签文件，通过以下方式下载：
+
+```bash linenums="1"
+# 训练集标签
+wget -P ./train_data/ic15_data  https://paddleocr.bj.bcebos.com/dataset/rec_gt_train.txt
+# 测试集标签
+wget -P ./train_data/ic15_data  https://paddleocr.bj.bcebos.com/dataset/rec_gt_test.txt
+```
+
+PaddleOCR 也提供了数据格式转换脚本，可以将ICDAR官网 label 转换为PaddleOCR支持的数据格式。 数据转换工具在 `ppocr/utils/gen_label.py`, 这里以训练集为例：
+
+```bash linenums="1"
+# 将官网下载的标签文件转换为 rec_gt_label.txt
+python gen_label.py --mode="rec" --input_path="{path/of/origin/label}" --output_label="rec_gt_label.txt"
+```
+
+数据样式格式如下，(a)为原始图片,(b)为每张图片对应的 Ground Truth 文本文件：
+![img](./images/icdar_rec.png)
+
+- 多语言数据集
+
+多语言模型的训练数据集均为100w的合成数据，使用了开源合成工具 [text_renderer](https://github.com/Sanster/text_renderer) ，少量的字体可以通过下面两种方式下载。
+
+- [百度网盘](https://pan.baidu.com/s/1bS_u207Rm7YbY33wOECKDA) 提取码：frgi
+
+### 1.4. 字典
+
+最后需要提供一个字典（{word_dict_name}.txt），使模型在训练时，可以将所有出现的字符映射为字典的索引。
+
+因此字典需要包含所有希望被正确识别的字符，{word_dict_name}.txt需要写成如下格式，并以 `utf-8` 编码格式保存：
+
+```text linenums="1"
+l
+d
+a
+d
+r
+n
+```
+
+word_dict.txt 每行有一个单字，将字符与数字索引映射在一起，“and” 将被映射成 [2 5 1]
+
+- 内置字典
+
+PaddleOCR内置了一部分字典，可以按需使用。
+
+`ppocr/utils/ppocr_keys_v1.txt` 是一个包含6623个字符的中文字典
+
+`ppocr/utils/ic15_dict.txt` 是一个包含36个字符的英文字典
+
+`ppocr/utils/dict/french_dict.txt` 是一个包含118个字符的法文字典
+
+`ppocr/utils/dict/japan_dict.txt` 是一个包含4399个字符的日文字典
+
+`ppocr/utils/dict/korean_dict.txt` 是一个包含3636个字符的韩文字典
+
+`ppocr/utils/dict/german_dict.txt` 是一个包含131个字符的德文字典
+
+`ppocr/utils/en_dict.txt` 是一个包含96个字符的英文字典
+
+目前的多语言模型仍处在demo阶段，会持续优化模型并补充语种，**非常欢迎您为我们提供其他语言的字典和字体**，
+如您愿意可将字典文件提交至 [dict](../../../../ppocr/utils/dict)，我们会在Repo中感谢您。
+
+- 自定义字典
+
+如需自定义dic文件，请在 `configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml` 中添加 `character_dict_path` 字段, 指向您的字典路径。
+
+### 1.5. 添加空格类别
+
+如果希望支持识别"空格"类别, 请将yml文件中的 `use_space_char` 字段设置为 `True`。
+
+### 1.6. 数据增强
+
+PaddleOCR提供了多种数据增强方式，默认配置文件中已经添加了数据增广。
+
+默认的扰动方式有：颜色空间转换(cvtColor)、模糊(blur)、抖动(jitter)、噪声(Gasuss noise)、随机切割(random crop)、透视(perspective)、颜色反转(reverse)、TIA数据增广。
+
+训练过程中每种扰动方式以40%的概率被选择，具体代码实现请参考：[rec_img_aug.py](../../../../ppocr/data/imaug/rec_img_aug.py)
+
+*由于OpenCV的兼容性问题，扰动操作暂时只支持Linux*
+
+## 2. 开始训练
+
+PaddleOCR提供了训练脚本、评估脚本和预测脚本，本节将以 PP-OCRv3 英文识别模型为例：
+
+### 2.1. 启动训练
+
+首先下载pretrain model，您可以下载训练好的模型在 icdar2015 数据上进行finetune
+
+```bash linenums="1"
+cd PaddleOCR/
+# 下载英文PP-OCRv3的预训练模型
+wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_train.tar
+# 解压模型参数
+cd pretrain_models
+tar -xf en_PP-OCRv3_rec_train.tar && rm -rf en_PP-OCRv3_rec_train.tar
+```
+
+开始训练:
+
+*如果您安装的是cpu版本，请将配置文件中的 `use_gpu` 字段修改为false*
+
+```bash linenums="1"
+# GPU训练 支持单卡，多卡训练
+# 训练icdar15英文数据 训练日志会自动保存为 "{save_model_dir}" 下的train.log
+
+#单卡训练（训练周期长，不建议）
+python3 tools/train.py -c configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml -o Global.pretrained_model=./pretrain_models/en_PP-OCRv3_rec_train/best_accuracy
+
+# 多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml -o Global.pretrained_model=./pretrain_models/en_PP-OCRv3_rec_train/best_accuracy
+```
+
+正常启动训练后，会看到以下log输出：
+
+```bash linenums="1"
+[2022/02/22 07:58:05] root INFO: epoch: [1/800], iter: 10, lr: 0.000000, loss: 0.754281, acc: 0.000000, norm_edit_dis: 0.000008, reader_cost: 0.55541 s, batch_cost: 0.91654 s, samples: 1408, ips: 153.62133
+[2022/02/22 07:58:13] root INFO: epoch: [1/800], iter: 20, lr: 0.000001, loss: 0.924677, acc: 0.000000, norm_edit_dis: 0.000008, reader_cost: 0.00236 s, batch_cost: 0.28528 s, samples: 1280, ips: 448.68599
+[2022/02/22 07:58:23] root INFO: epoch: [1/800], iter: 30, lr: 0.000002, loss: 0.967231, acc: 0.000000, norm_edit_dis: 0.000008, reader_cost: 0.14527 s, batch_cost: 0.42714 s, samples: 1280, ips: 299.66507
+[2022/02/22 07:58:31] root INFO: epoch: [1/800], iter: 40, lr: 0.000003, loss: 0.895318, acc: 0.000000, norm_edit_dis: 0.000008, reader_cost: 0.00173 s, batch_cost: 0.27719 s, samples: 1280, ips: 461.77252
+```
+
+log 中自动打印如下信息：
+
+|  字段   |   含义   |
+| :----: | :------: |
+|  epoch | 当前迭代轮次 |
+|  iter  | 当前迭代次数 |
+|  lr    | 当前学习率 |
+|  loss  | 当前损失函数 |
+|  acc   | 当前batch的准确率 |
+|  norm_edit_dis | 当前 batch 的编辑距离 |
+|  reader_cost | 当前 batch 数据处理耗时 |
+|  batch_cost | 当前 batch 总耗时 |
+|  samples  | 当前 batch 内的样本数 |
+|  ips  | 每秒处理图片的数量 |
+
+PaddleOCR支持训练和评估交替进行, 可以在 `configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml` 中修改 `eval_batch_step` 设置评估频率，默认每500个iter评估一次。评估过程中默认将最佳acc模型，保存为 `output/en_PP-OCRv3_mobile_rec/best_accuracy` 。
+
+如果验证集很大，测试将会比较耗时，建议减少评估次数，或训练完再进行评估。
+
+**提示：** 可通过 -c 参数选择 `configs/rec/` 路径下的多种模型配置进行训练，PaddleOCR支持的识别算法可以参考[前沿算法列表](../../algorithm/overview.md)：
+
+训练中文数据，推荐使用[PP-OCRv3_mobile_rec_distillation.yml](../../../../configs/rec/PP-OCRv3/PP-OCRv3_mobile_rec_distillation.yml)，如您希望尝试其他算法在中文数据集上的效果，请参考下列说明修改配置文件：
+
+以 `PP-OCRv3_mobile_rec_distillation.yml` 为例：
+
+```yaml linenums="1"
+Global:
+  ...
+  # 添加自定义字典，如修改字典请将路径指向新字典
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  ...
+  # 识别空格
+  use_space_char: True
+
+
+Optimizer:
+  ...
+  # 添加学习率衰减策略
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  ...
+
+...
+
+Train:
+  dataset:
+    # 数据集格式，支持LMDBDataSet以及SimpleDataSet
+    name: SimpleDataSet
+    # 数据集路径
+    data_dir: ./train_data/
+    # 训练集标签文件
+    label_file_list: ["./train_data/train_list.txt"]
+    transforms:
+      ...
+      - RecResizeImg:
+          # 修改 image_shape 以适应长文本
+          image_shape: [3, 48, 320]
+      ...
+  loader:
+    ...
+    # 单卡训练的batch_size
+    batch_size_per_card: 256
+    ...
+
+Eval:
+  dataset:
+    # 数据集格式，支持LMDBDataSet以及SimpleDataSet
+    name: SimpleDataSet
+    # 数据集路径
+    data_dir: ./train_data
+    # 验证集标签文件
+    label_file_list: ["./train_data/val_list.txt"]
+    transforms:
+      ...
+      - RecResizeImg:
+          # 修改 image_shape 以适应长文本
+          image_shape: [3, 48, 320]
+      ...
+  loader:
+    # 单卡验证的batch_size
+    batch_size_per_card: 256
+    ...
+```
+
+**注意，预测/评估时的配置文件请务必与训练一致。**
+
+### 2.2. 断点训练
+
+如果训练程序中断，如果希望加载训练中断的模型从而恢复训练，可以通过指定Global.checkpoints指定要加载的模型路径：
+
+```bash linenums="1"
+python3 tools/train.py -c configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml -o Global.checkpoints=./your/trained/model
+```
+
+**注意**：`Global.checkpoints`的优先级高于`Global.pretrained_model`的优先级，即同时指定两个参数时，优先加载`Global.checkpoints`指定的模型，如果`Global.checkpoints`指定的模型路径有误，会加载`Global.pretrained_model`指定的模型。
+
+### 2.3. 更换Backbone 训练
+
+PaddleOCR将网络划分为四部分，分别在[ppocr/modeling](../../../../ppocr/modeling)下。 进入网络的数据将按照顺序(transforms->backbones->necks->heads)依次通过这四个部分。
+
+```bash linenums="1"
+├── architectures # 网络的组网代码
+├── transforms    # 网络的图像变换模块
+├── backbones     # 网络的特征提取模块
+├── necks         # 网络的特征增强模块
+└── heads         # 网络的输出模块
+```
+
+如果要更换的Backbone 在PaddleOCR中有对应实现，直接修改配置yml文件中`Backbone`部分的参数即可。
+
+如果要使用新的Backbone，更换backbones的例子如下:
+
+1. 在 [ppocr/modeling/backbones](../../../../ppocr/modeling/backbones) 文件夹下新建文件，如my_backbone.py。
+2. 在 my_backbone.py 文件内添加相关代码，示例代码如下:
+
+```python linenums="1"
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+
+class MyBackbone(nn.Layer):
+    def __init__(self, *args, **kwargs):
+        super(MyBackbone, self).__init__()
+        # your init code
+        self.conv = nn.xxxx
+
+    def forward(self, inputs):
+        # your network forward
+        y = self.conv(inputs)
+        return y
+```
+
+3. 在 [ppocr/modeling/backbones/\_*init\_*.py](../../../../ppocr/modeling/backbones/__init__.py)文件内导入添加的`MyBackbone`模块，然后修改配置文件中Backbone进行配置即可使用，格式如下:
+
+```yaml linenums="1"
+Backbone:
+name: MyBackbone
+args1: args1
+```
+
+**注意**：如果要更换网络的其他模块，可以参考[文档](../../algorithm/add_new_algorithm.md)。
+
+### 2.4. 混合精度训练
+
+如果您想进一步加快训练速度，可以使用[自动混合精度训练](https://www.paddlepaddle.org.cn/documentation/docs/zh/guides/performance_improving/amp_cn.html)， 以单机单卡为例，命令如下：
+
+```bash linenums="1"
+python3 tools/train.py -c configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml \
+     -o Global.pretrained_model=./pretrain_models/en_PP-OCRv3_rec_train/best_accuracy \
+     Global.use_amp=True Global.scale_loss=1024.0 Global.use_dynamic_loss_scaling=True
+```
+
+### 2.5. 分布式训练
+
+多机多卡训练时，通过 `--ips` 参数设置使用的机器IP地址，通过 `--gpus` 参数设置使用的GPU ID：
+
+```bash linenums="1"
+python3 -m paddle.distributed.launch --ips="xx.xx.xx.xx,xx.xx.xx.xx" --gpus '0,1,2,3' tools/train.py -c configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml \
+     -o Global.pretrained_model=./pretrain_models/en_PP-OCRv3_rec_train/best_accuracy
+```
+
+**注意:** （1）采用多机多卡训练时，需要替换上面命令中的ips值为您机器的地址，机器之间需要能够相互ping通；（2）训练时需要在多个机器上分别启动命令。查看机器ip地址的命令为`ifconfig`；（3）更多关于分布式训练的性能优势等信息，请参考：[分布式训练教程](../blog/distributed_training.md)。
+
+### 2.6. 知识蒸馏训练
+
+PaddleOCR支持了基于知识蒸馏的文本识别模型训练过程，更多内容可以参考[知识蒸馏说明文档](../model_compress/knowledge_distillation.md)。
+
+### 2.7. 多语言模型训练
+
+PaddleOCR目前已支持80种（除中文外）语种识别，`configs/rec/multi_languages` 路径下提供了一个多语言的配置文件模版: [rec_multi_language_lite_train.yml](../../../../configs/rec/multi_language/rec_multi_language_lite_train.yml)。
+
+按语系划分，目前PaddleOCR支持的语种有：
+
+| 配置文件 |  算法名称 |   backbone |   trans   |   seq      |     pred     |  language |
+| :--------: |  :-------:   | :-------:  |   :-------:   |   :-----:   |  :-----:   | :-----:  |
+| rec_chinese_cht_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 中文繁体  |
+| rec_en_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 英语（区分大小写）   |
+| rec_french_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 法语 |
+| rec_ger_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 德语   |
+| rec_japan_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 日语  |
+| rec_korean_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 韩语  |
+| rec_latin_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 拉丁字母  |
+| rec_arabic_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 阿拉伯字母 |
+| rec_cyrillic_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 斯拉夫字母  |
+| rec_devanagari_lite_train.yml |  CRNN |   Mobilenet_v3 small 0.5 |  None   |  BiLSTM |  ctc  | 梵文字母  |
+
+更多支持语种请参考: [多语言模型](../blog/multi_languages.md)
+
+如您希望在现有模型效果的基础上调优，请参考下列说明修改配置文件：
+
+以 `rec_french_lite_train` 为例：
+
+```yaml linenums="1"
+Global:
+  ...
+  # 添加自定义字典，如修改字典请将路径指向新字典
+  character_dict_path: ./ppocr/utils/dict/french_dict.txt
+  ...
+  # 识别空格
+  use_space_char: True
+
+...
+
+Train:
+  dataset:
+    # 数据集格式，支持LMDBDataSet以及SimpleDataSet
+    name: SimpleDataSet
+    # 数据集路径
+    data_dir: ./train_data/
+    # 训练集标签文件
+    label_file_list: ["./train_data/french_train.txt"]
+    ...
+
+Eval:
+  dataset:
+    # 数据集格式，支持LMDBDataSet以及SimpleDataSet
+    name: SimpleDataSet
+    # 数据集路径
+    data_dir: ./train_data
+    # 验证集标签文件
+    label_file_list: ["./train_data/french_val.txt"]
+    ...
+```
+
+### 2.8. 其他训练环境
+
+- Windows GPU/CPU
+在Windows平台上与Linux平台略有不同:
+Windows平台只支持`单卡`的训练与预测，指定GPU进行训练`set CUDA_VISIBLE_DEVICES=0`
+在Windows平台，DataLoader只支持单进程模式，因此需要设置 `num_workers` 为0;
+
+- macOS
+不支持GPU模式，需要在配置文件中设置`use_gpu`为False，其余训练评估预测命令与Linux GPU完全相同。
+
+- Linux DCU
+DCU设备上运行需要设置环境变量 `export HIP_VISIBLE_DEVICES=0,1,2,3`，其余训练评估预测命令与Linux GPU完全相同。
+
+### 2.9 模型微调
+
+实际使用过程中，建议加载官方提供的预训练模型，在自己的数据集中进行微调，关于识别模型的微调方法，请参考：[模型微调教程](./finetune.md)。
+
+## 3. 模型评估与预测
+
+### 3.1. 指标评估
+
+训练中模型参数默认保存在`Global.save_model_dir`目录下。在评估指标时，需要设置`Global.checkpoints`指向保存的参数文件。评估数据集可以通过 `configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml`  修改Eval中的 `label_file_path` 设置。
+
+```bash linenums="1"
+# GPU 评估， Global.checkpoints 为待测权重
+python3 -m paddle.distributed.launch --gpus '0' tools/eval.py -c configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml -o Global.checkpoints={path/to/weights}/best_accuracy
+```
+
+### 3.2. 测试识别效果
+
+使用 PaddleOCR 训练好的模型，可以通过以下脚本进行快速预测。
+
+默认预测图片存储在 `infer_img` 里，通过 `-o Global.checkpoints` 加载训练好的参数文件：
+
+根据配置文件中设置的 `save_model_dir` 和 `save_epoch_step` 字段，会有以下几种参数被保存下来：
+
+```text linenums="1"
+output/rec/
+├── best_accuracy.pdopt
+├── best_accuracy.pdparams
+├── best_accuracy.states
+├── config.yml
+├── iter_epoch_3.pdopt
+├── iter_epoch_3.pdparams
+├── iter_epoch_3.states
+├── latest.pdopt
+├── latest.pdparams
+├── latest.states
+└── train.log
+```
+
+其中 best_accuracy.*是评估集上的最优模型；iter_epoch_x.* 是以 `save_epoch_step` 为间隔保存下来的模型；latest.* 是最后一个epoch的模型。
+
+```bash linenums="1"
+# 预测英文结果
+python3 tools/infer_rec.py -c configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml -o Global.pretrained_model={path/to/weights}/best_accuracy  Global.infer_img=doc/imgs_words/en/word_1.png
+```
+
+预测图片：
+
+![img](./images/word_1-20240704092705543.png)
+
+得到输入图像的预测结果：
+
+```bash linenums="1"
+infer_img: doc/imgs_words/en/word_1.png
+        result: ('joint', 0.9998967)
+```
+
+预测使用的配置文件必须与训练一致，如您通过 `python3 tools/train.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml` 完成了中文模型的训练，
+您可以使用如下命令进行中文模型预测。
+
+```bash linenums="1"
+# 预测中文结果
+python3 tools/infer_rec.py -c configs/rec/ch_ppocr_v2.0/rec_chinese_lite_train_v2.0.yml -o Global.pretrained_model={path/to/weights}/best_accuracy Global.infer_img=doc/imgs_words/ch/word_1.jpg
+```
+
+预测图片：
+
+![img](./images/word_1-20240704092713071.jpg)
+
+得到输入图像的预测结果：
+
+```bash linenums="1"
+infer_img: doc/imgs_words/ch/word_1.jpg
+        result: ('韩国小馆', 0.997218)
+```
+
+## 4. 模型导出与预测
+
+inference 模型（`paddle.jit.save`保存的模型）
+一般是模型训练，把模型结构和模型参数保存在文件中的固化模型，多用于预测部署场景。
+训练过程中保存的模型是checkpoints模型，保存的只有模型的参数，多用于恢复训练等。
+与checkpoints模型相比，inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成。
+
+识别模型转inference模型与检测的方式相同，如下：
+
+```bash linenums="1"
+# 开启旧 IR 模式
+export FLAGS_enable_pir_api=0
+
+# -c 后面设置训练算法的yml配置文件
+# -o 配置可选参数
+# Global.pretrained_model 参数设置待转换的训练模型地址，不用添加文件后缀 .pdmodel，.pdopt或.pdparams。
+# Global.save_inference_dir参数设置转换的模型将保存的地址。
+
+python3 tools/export_model.py -c configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml -o Global.pretrained_model=./pretrain_models/en_PP-OCRv3_rec_train/best_accuracy  Global.save_inference_dir=./inference/en_PP-OCRv3_mobile_rec/
+```
+
+**注意：**如果您是在自己的数据集上训练的模型，并且调整了中文字符的字典文件，请注意修改配置文件中的`character_dict_path`为自定义字典文件。
+
+转换成功后，在目录下有三个文件：
+
+```text linenums="1"
+inference/en_PP-OCRv3_mobile_rec/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    ├── inference.pdiparams.info    # 识别inference模型的参数信息，可忽略
+    └── inference.pdmodel           # 识别inference模型的program文件
+```
+
+**注意：** 如果需要以新 IR 模式（`FLAGS_enable_pir_api=1`）存储 `.json` 文件，请执行以下命令切换到新 IR 模式：
+
+```bash linenums="1"
+export FLAGS_enable_pir_api=1
+python3 tools/export_model.py -c configs/rec/PP-OCRv3/en_PP-OCRv3_mobile_rec.yml -o Global.pretrained_model=./pretrain_models/en_PP-OCRv3_rec_train/best_accuracy  Global.save_inference_dir=./inference/en_PP-OCRv3_mobile_rec/
+```
+
+转换成功后，在目录下有三个文件：
+
+```text linenums="1"
+inference/en_PP-OCRv3_mobile_rec/
+    ├── inference.pdiparams         # 识别inference模型的参数文件
+    └── inference.json              # 识别inference模型的program文件
+```
+
+- 自定义模型推理
+
+  如果训练时修改了文本的字典，在使用inference模型预测时，需要通过`--rec_char_dict_path`指定使用的字典路径，更多关于推理超参数的配置与解释，请参考：[模型推理超参数解释教程](../blog/inference_args.md)。
+
+  ```bash linenums="1"
+  python3 tools/infer/predict_rec.py --image_dir="./doc/imgs_words_en/word_336.png" --rec_model_dir="./your inference model" --rec_image_shape="3, 48, 320" --rec_char_dict_path="your text dict path"
+  ```
+
+## 5. FAQ
+
+Q1: 训练模型转inference 模型之后预测效果不一致？
+
+**A**：此类问题出现较多，问题多是trained model预测时候的预处理、后处理参数和inference model预测的时候的预处理、后处理参数不一致导致的。可以对比训练使用的配置文件中的预处理、后处理和预测时是否存在差异。
diff --git a/docs/version2.x/ppocr/model_train/training.en.md b/docs/version2.x/ppocr/model_train/training.en.md
new file mode 100644
index 0000000..ce0de1d
--- /dev/null
+++ b/docs/version2.x/ppocr/model_train/training.en.md
@@ -0,0 +1,132 @@
+---
+comments: true
+---
+
+# Model Training
+
+This article will introduce the basic concepts that is necessary for model training and tuning.
+
+At the same time, it will briefly introduce the structure of the training data and how to prepare the data to fine-tune model in vertical scenes.
+
+## 1. Yml Configuration
+
+The PaddleOCR uses configuration files to control network training and evaluation parameters. In the configuration file, you can set the model, optimizer, loss function, and pre- and post-processing parameters of the model. PaddleOCR reads these parameters from the configuration file, and then builds a complete training process to train the model. Fine-tuning can also be completed by modifying the parameters in the configuration file, which is simple and convenient.
+
+For the complete configuration file description, please refer to [Configuration File](../blog/config.en.md).
+
+## 2. Basic Concepts
+
+During the model training process, some hyper-parameters can be manually specified to obtain the optimal result at the least cost. Different data volumes may require different hyper-parameters. When you want to fine-tune the model based on your own data, there are several parameter adjustment strategies for reference:
+
+### 2.1 Learning Rate
+
+The learning rate is one of the most important hyper-parameters for training neural networks. It represents the step length of the gradient moving towards the optimal solution of the loss function in each iteration.
+A variety of learning rate update strategies are provided by PaddleOCR, which can be specified in configuration files. For example:
+
+```yaml linenums="1"
+Optimizer:
+  ...
+  lr:
+    name: Piecewise
+    decay_epochs : [700, 800]
+    values : [0.001, 0.0001]
+    warmup_epoch: 5
+```
+
+`Piecewise` stands for piece-wise constant attenuation. Different learning rates are specified in different learning stages, and the learning rate stay the same in each stage.
+
+`warmup_epoch` means that in the first 5 epochs, the learning rate will be increased gradually from 0 to base_lr. For all strategies, please refer to the code [learning_rate.py](../../../../ppocr/optimizer/learning_rate.py).
+
+### 2.2 Regularization
+
+Regularization can effectively avoid algorithm over-fitting. PaddleOCR provides L1 and L2 regularization methods.
+L1 and L2 regularization are the most widely used regularization methods.
+L1 regularization adds a regularization term to the objective function to reduce the sum of absolute values of the parameters;
+while in L2 regularization, the purpose of adding a regularization term is to reduce the sum of squared parameters.
+The configuration method is as follows:
+
+```yaml linenums="1"
+Optimizer:
+  ...
+  regularizer:
+    name: L2
+    factor: 2.0e-05
+```
+
+### 2.3 Evaluation Indicators
+
+(1) Detection stage: First, evaluate according to the IOU of the detection frame and the labeled frame. If the IOU is greater than a certain threshold, it is judged that the detection is accurate. Here, the detection frame and the label frame are different from the general target detection frame, and they are represented by polygons. Detection accuracy: the percentage of the correct detection frame number in all detection frames is mainly used to judge the detection index. Detection recall rate: the percentage of correct detection frames in all marked frames, which is mainly an indicator of missed detection.
+
+(2) Recognition stage: Character recognition accuracy, that is, the ratio of correctly recognized text lines to the number of marked text lines. Only the entire line of text recognition pairs can be regarded as correct recognition.
+
+(3) End-to-end statistics: End-to-end recall rate: accurately detect and correctly identify the proportion of text lines in all labeled text lines; End-to-end accuracy rate: accurately detect and correctly identify the number of text lines in the detected text lines. The standard for accurate detection is that the IOU of the detection box and the labeled box is greater than a certain threshold, and the text in the correctly identified detection box is the same as the labeled text.
+
+## 3. Data and Vertical Scenes
+
+### 3.1 Training Data
+
+The current open source models, data sets and magnitudes are as follows:
+
+- Detection:
+  - English data set, ICDAR2015
+  - Chinese data set, LSVT street view data set training data 3w pictures
+
+- Identification:
+  - English data set, MJSynth and SynthText synthetic data, the data volume is tens of millions.
+  - Chinese data set, LSVT street view data set crops the image according to the truth value, and performs position calibration, a total of 30w images. In addition, based on the LSVT corpus, 500w of synthesized data.
+  - Small language data set, using different corpora and fonts, respectively generated 100w synthetic data set, and using ICDAR-MLT as the verification set.
+
+Among them, the public data sets are all open source, users can search and download by themselves, or refer to [Chinese data set](../../../datasets/datasets.en.md), synthetic data is not open source, users can use open source synthesis tools to synthesize by themselves. Synthesis tools include [text_renderer](https://github.com/Sanster/text_renderer), [SynthText](https://github.com/ankush-me/SynthText), [TextRecognitionDataGenerator](https://github.com/Belval/TextRecognitionDataGenerator) etc.
+
+### 3.2 Vertical Scene
+
+PaddleOCR mainly focuses on general OCR. If you have vertical requirements, you can use PaddleOCR + vertical data to train yourself;
+If there is a lack of labeled data, or if you do not want to invest in research and development costs, it is recommended to directly call the open API, which covers some of the more common vertical categories.
+
+### 3.3 Build Your Own Dataset
+
+There are several experiences for reference when constructing the data set:
+
+(1) The amount of data in the training set:
+
+a. The data required for detection is relatively small. For Fine-tune based on the PaddleOCR model, 500 sheets are generally required to achieve good results.
+
+b. Recognition is divided into English and Chinese. Generally, English scenarios require hundreds of thousands of data to achieve good results, while Chinese requires several million or more.
+
+(2) When the amount of training data is small, you can try the following three ways to get more data:
+
+a. Manually collect more training data, the most direct and effective way.
+
+b. Basic image processing or transformation based on PIL and opencv. For example, the three modules of ImageFont, Image, ImageDraw in PIL write text into the background, opencv's rotating affine transformation, Gaussian filtering and so on.
+
+c. Use data generation algorithms to synthesize data, such as algorithms such as pix2pix.
+
+## 4. FAQ
+
+**Q**: How to choose a suitable network input shape when training CRNN recognition?
+
+    A: The general height is 32, the longest width is selected, there are two methods:
+    
+    (1) Calculate the aspect ratio distribution of training sample images. The selection of the maximum aspect ratio considers 80% of the training samples.
+    
+    (2) Count the number of texts in training samples. The selection of the longest number of characters considers the training sample that satisfies 80%. Then the aspect ratio of Chinese characters is approximately considered to be 1, and that of English is 3:1, and the longest width is estimated.
+
+**Q**: During the recognition training, the accuracy of the training set has reached 90, but the accuracy of the verification set has been kept at 70, what should I do?
+
+    A: If the accuracy of the training set is 90 and the test set is more than 70, it should be over-fitting. There are two methods to try:
+    
+    (1) Add more augmentation methods or increase the [probability] of augmented prob (https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppocr/data/imaug/rec_img_aug.py#L341), The default is 0.4.
+    
+    (2) Increase the [l2 dcay value] of the system (https://github.com/PaddlePaddle/PaddleOCR/blob/a501603d54ff5513fc4fc760319472e59da25424/configs/rec/ch_ppocr_v1.1/rec_chinese_lite_train_v1.1.yml#L47)
+
+**Q**: When the recognition model is trained, loss can drop normally, but acc is always 0
+
+    A: It is normal for the acc to be 0 at the beginning of the recognition model training, and the indicator will come up after a longer training period.
+
+***
+
+Click the following links for detailed training tutorial:
+
+- [text detection model training](./detection.en.md)
+- [text recognition model training](./recognition.en.md)
+- [text direction classification model training](./angle_class.en.md)
diff --git a/docs/version2.x/ppocr/model_train/training.md b/docs/version2.x/ppocr/model_train/training.md
new file mode 100644
index 0000000..e188e59
--- /dev/null
+++ b/docs/version2.x/ppocr/model_train/training.md
@@ -0,0 +1,128 @@
+---
+comments: true
+---
+
+# PP-OCR模型训练
+
+本文将介绍模型训练时需掌握的基本概念，和训练时的调优方法。
+
+同时会简单介绍PaddleOCR模型训练数据的组成部分，以及如何在垂类场景中准备数据finetune模型。
+
+## 1. 配置文件说明
+
+PaddleOCR模型使用配置文件管理网络训练、评估的参数。在配置文件中，可以设置组建模型、优化器、损失函数、模型前后处理的参数，PaddleOCR从配置文件中读取到这些参数，进而组建出完整的训练流程，完成模型训练，在需要对模型进行优化的时，可以通过修改配置文件中的参数完成配置，使用简单且方便修改。
+
+完整的配置文件说明可以参考[配置文件](../blog/config.md)
+
+## 2. 基本概念
+
+模型训练过程中需要手动调整一些超参数，帮助模型以最小的代价获得最优指标。不同的数据量可能需要不同的超参，当您希望在自己的数据上finetune或对模型效果调优时，有以下几个参数调整策略可供参考：
+
+### 2.1 学习率
+
+学习率是训练神经网络的重要超参数之一，它代表在每一次迭代中梯度向损失函数最优解移动的步长。
+在PaddleOCR中提供了多种学习率更新策略,可以通过配置文件修改，例如：
+
+```yaml linenums="1"
+Optimizer:
+  ...
+  lr:
+    name: Piecewise
+    decay_epochs : [700, 800]
+    values : [0.001, 0.0001]
+    warmup_epoch: 5
+```
+
+Piecewise 代表分段常数衰减，在不同的学习阶段指定不同的学习率，在每段内学习率相同。
+warmup_epoch 代表在前5个epoch中，学习率将逐渐从0增加到base_lr。全部策略可以参考代码 [learning_rate.py](../../../../ppocr/optimizer/learning_rate.py) 。
+
+### 2.2 正则化
+
+正则化可以有效的避免算法过拟合，PaddleOCR中提供了L1、L2正则方法，L1 和 L2 正则化是最常用的正则化方法。L1 正则化向目标函数添加正则化项，以减少参数的绝对值总和；而 L2 正则化中，添加正则化项的目的在于减少参数平方的总和。配置方法如下：
+
+```yaml linenums="1"
+Optimizer:
+  ...
+  regularizer:
+    name: L2
+    factor: 2.0e-05
+```
+
+### 2.3 评估指标
+
+（1）检测阶段：先按照检测框和标注框的IOU评估，IOU大于某个阈值判断为检测准确。这里检测框和标注框不同于一般的通用目标检测框，是采用多边形进行表示。检测准确率：正确的检测框个数在全部检测框的占比，主要是判断检测指标。检测召回率：正确的检测框个数在全部标注框的占比，主要是判断漏检的指标。
+
+（2）识别阶段： 字符识别准确率，即正确识别的文本行占标注的文本行数量的比例，只有整行文本识别对才算正确识别。
+
+（3）端到端统计： 端对端召回率：准确检测并正确识别文本行在全部标注文本行的占比； 端到端准确率：准确检测并正确识别文本行在 检测到的文本行数量 的占比； 准确检测的标准是检测框与标注框的IOU大于某个阈值，正确识别的检测框中的文本与标注的文本相同。
+
+## 3. 数据与垂类场景
+
+### 3.1 训练数据
+
+目前开源的模型，数据集和量级如下：
+
+- 检测：
+    - 英文数据集，ICDAR2015
+    - 中文数据集，LSVT街景数据集训练数据3w张图片
+
+- 识别：
+    - 英文数据集，MJSynth和SynthText合成数据，数据量上千万。
+    - 中文数据集，LSVT街景数据集根据真值将图crop出来，并进行位置校准，总共30w张图像。此外基于LSVT的语料，合成数据500w。
+    - 小语种数据集，使用不同语料和字体，分别生成了100w合成数据集，并使用ICDAR-MLT作为验证集。
+
+其中，公开数据集都是开源的，用户可自行搜索下载，也可参考[中文数据集](../../../datasets/datasets.md)，合成数据暂不开源，用户可使用开源合成工具自行合成，可参考的合成工具包括[text_renderer](https://github.com/Sanster/text_renderer) 、[SynthText](https://github.com/ankush-me/SynthText) 、[TextRecognitionDataGenerator](https://github.com/Belval/TextRecognitionDataGenerator) 等。
+
+### 3.2 垂类场景
+
+PaddleOCR主要聚焦通用OCR，如果有垂类需求，您可以用PaddleOCR+垂类数据自己训练；
+如果缺少带标注的数据，或者不想投入研发成本，建议直接调用开放的API，开放的API覆盖了目前比较常见的一些垂类。
+
+### 3.3 自己构建数据集
+
+在构建数据集时有几个经验可供参考：
+
+（1） 训练集的数据量：
+
+a. 检测需要的数据相对较少，在PaddleOCR模型的基础上进行Fine-tune，一般需要500张可达到不错的效果。
+
+b. 识别分英文和中文，一般英文场景需要几十万数据可达到不错的效果，中文则需要几百万甚至更多。
+
+（2）当训练数据量少时，可以尝试以下三种方式获取更多的数据：
+
+a. 人工采集更多的训练数据，最直接也是最有效的方式。
+
+b. 基于PIL和opencv基本图像处理或者变换。例如PIL中ImageFont, Image, ImageDraw三个模块将文字写到背景中，opencv的旋转仿射变换，高斯滤波等。
+
+c. 利用数据生成算法合成数据，例如pix2pix或[StyleText](https://github.com/PFCCLab/StyleText)等算法。
+
+## 4. 常见问题
+
+**Q**：训练CRNN识别时，如何选择合适的网络输入shape？
+
+    A：一般高度采用32，最长宽度的选择，有两种方法：
+
+    （1）统计训练样本图像的宽高比分布。最大宽高比的选取考虑满足80%的训练样本。
+
+    （2）统计训练样本文字数目。最长字符数目的选取考虑满足80%的训练样本。然后中文字符长宽比近似认为是1，英文认为3：1，预估一个最长宽度。
+
+**Q**：识别训练时，训练集精度已经到达90了，但验证集精度一直在70，涨不上去怎么办？
+
+    A：训练集精度90，测试集70多的话，应该是过拟合了，有两个可尝试的方法：
+
+    （1）加入更多的增广方式或者调大增广prob的[概率](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppocr/data/imaug/rec_img_aug.py#L341)，默认为0.4。
+
+    （2）调大系统的[l2 dcay值](https://github.com/PaddlePaddle/PaddleOCR/blob/a501603d54ff5513fc4fc760319472e59da25424/configs/rec/ch_ppocr_v1.1/rec_chinese_lite_train_v1.1.yml#L47)
+
+**Q**: 识别模型训练时，loss能正常下降，但acc一直为0
+
+    A：识别模型训练初期acc为0是正常的，多训一段时间指标就上来了。
+
+***
+
+具体的训练教程可点击下方链接跳转：
+
+- [文本检测模型训练](./detection.md)
+- [文本识别模型训练](./recognition.md)
+- [文本方向分类器训练](./angle_class.md)
+- [知识蒸馏](../model_compress/knowledge_distillation.md)
diff --git a/docs/version2.x/ppocr/model_train/tricks.en.md b/docs/version2.x/ppocr/model_train/tricks.en.md
new file mode 100644
index 0000000..4db1b22
--- /dev/null
+++ b/docs/version2.x/ppocr/model_train/tricks.en.md
@@ -0,0 +1,63 @@
+---
+comments: true
+---
+
+Here we have sorted out some Chinese OCR training and prediction tricks, which are being updated continuously. You are welcome to contribute more OCR tricks!
+
+#### 1. Replace Backbone Network
+
+- **Problem Description**
+
+  At present, ResNet_vd series and MobileNetV3 series are the backbone networks used in PaddleOCR, whether replacing the other backbone networks will help to improve the accuracy? What should be paid attention to when replacing?
+
+- **Tips**
+    - Whether text detection or text recognition, the choice of backbone network is a trade-off between prediction effect and prediction efficiency. Generally, a larger backbone network is selected, e.g. ResNet101_vd, then the performance of the detection or recognition is more accurate, but the time cost will increase accordingly. And a smaller backbone network is selected, e.g. MobileNetV3_small_x0_35, the prediction speed is faster, but the accuracy of detection or recognition will be reduced. Fortunately, the detection or recognition effect of different backbone networks is positively correlated with the performance of ImageNet 1000 classification task. [**PaddleClas**](https://github.com/PaddlePaddle/PaddleClas/blob/release/2.3/docs/en/models/models_intro_en.md) have sorted out the 23 series of classification network structures, such as ResNet_vd、Res2Net、HRNet、MobileNetV3、GhostNet. It provides the top1 accuracy of classification, the time cost of GPU(V100 and T4) and CPU(SD 855), and the 117 pretrained models [**download addresses**](https://paddleclas-en.readthedocs.io/en/latest/models/models_intro_en.html).
+
+    - Similar as the 4 stages of ResNet, the replacement of text detection backbone network is to determine those four stages to facilitate the integration of FPN like the object detection heads. In addition, for the text detection problem, the pre trained model in ImageNet1000 can accelerate the convergence and improve the accuracy.
+
+    - In order to replace the backbone network of text recognition, we need to pay attention to the descending position of network width and height stride. Since the ratio between width and height is large in chinese text recognition, the frequency of height decrease is less and the frequency of width decrease is more. You can refer the [modifies of MobileNetV3](../../../../ppocr/modeling/backbones/rec_mobilenet_v3.py) in PaddleOCR.
+
+#### 2. Long Chinese Text Recognition
+
+- **Problem Description**
+  The maximum resolution of Chinese recognition model during training is [3,32,320], if the text image to be recognized is too long, as shown in the figure below, how to adapt?
+
+  ![img](./images/long_text_examples.jpg)
+
+- **Tips**
+
+  During the training, the training samples are not directly resized to [3,32,320]. At first, the height of samples are resized to 32 and keep the ratio between the width and the height. When the width is less than 320, the excess parts are padding 0. Besides, when the ratio between the width and the height of the samples is larger than 10, these samples will be ignored. When the prediction for one image, do as above, but do not limit the max ratio between the width and the height. When the prediction for an images batch, do as training, but the resized target width is the longest width of the images in the batch. [Code as following](../../../../tools/infer/predict_rec.py)：
+
+  ```python linenums="1"
+  def resize_norm_img(self, img, max_wh_ratio):
+      imgC, imgH, imgW = self.rec_image_shape
+      assert imgC == img.shape[2]
+      if self.character_type == "ch":
+          imgW = int((32 * max_wh_ratio))
+      h, w = img.shape[:2]
+      ratio = w / float(h)
+      if math.ceil(imgH * ratio) > imgW:
+          resized_w = imgW
+      else:
+          resized_w = int(math.ceil(imgH * ratio))
+      resized_image = cv2.resize(img, (resized_w, imgH))
+      resized_image = resized_image.astype('float32')
+      resized_image = resized_image.transpose((2, 0, 1)) / 255
+      resized_image -= 0.5
+      resized_image /= 0.5
+      padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
+      padding_im[:, :, 0:resized_w] = resized_image
+      return padding_im
+  ```
+
+#### 3. Space Recognition
+
+- **Problem Description**
+
+  As shown in the figure below, for Chinese and English mixed scenes, in order to facilitate reading and using the recognition results, it is often necessary to recognize the spaces between words. How can this situation be adapted?
+
+  ![img](./images/en_paper.jpg)
+
+- **Tips**
+
+  There are two possible methods for space recognition. (1) Optimize the text detection. For splitting the text at the space in detection results, it needs to divide the text line with space into many segments when label the data for detection. (2) Optimize the text recognition. The space character is introduced into the recognition dictionary. Label the blank line in the training data for text recognition. In addition, we can also concat multiple word lines to synthesize the training data with spaces. PaddleOCR currently uses the second method.
diff --git a/docs/version2.x/ppocr/model_train/tricks.md b/docs/version2.x/ppocr/model_train/tricks.md
new file mode 100644
index 0000000..2b7c964
--- /dev/null
+++ b/docs/version2.x/ppocr/model_train/tricks.md
@@ -0,0 +1,63 @@
+---
+comments: true
+---
+
+这里我们整理了一些中文OCR训练和预测技巧，持续更新中，欢迎大家贡献更多OCR技巧~
+
+#### 1、更换骨干网络
+
+- **问题描述**
+
+    目前PaddleOCR使用的主干网络为ResNet_vd系列和MobileNetV3系列，更换其他主干网络是否有助于提高准确率？更换时需要注意什么？
+
+- **技巧**
+    - 无论是文本检测还是文本识别，主干网络的选择都是预测效果和预测效率的权衡。一般选择较大的主干网络，如ResNet101_vd，则检测或识别的性能更准确，但时间成本也会相应增加。而选择较小的主干网络，如MobileNetV3_small_x0_35，预测速度更快，但检测或识别的准确率会降低。幸运的是，不同骨干网络的检测或识别效果与ImageNet 1000分类任务的性能呈正相关。[**PaddleClas**](https://github.com/PaddlePaddle/PaddleClas/blob/release/2.3/docs/en/models/models_intro_en.md)整理了ResNet_vd、Res2Net、HRNet、MobileNetV3、GhostNet等23个系列的分类网络结构，提供了分类top1准确率、GPU(V100和T4)和CPU(SD 855)的时间成本，以及117个预训练模型[**下载地址**](https://paddleclas-en.readthedocs.io/en/latest/models/models_intro_en.html)。
+
+- 和ResNet的4个阶段类似，文本检测骨干网络的更换就是确定这4个阶段，以便于像物体检测heads一样集成FPN。另外，对于文本检测问题，ImageNet1000中的预训练模型可以加速收敛并提高准确率。
+
+- 更换文本识别骨干网络时，需要注意网络宽度和高度步长的下降位置。由于中文文本识别中宽度和高度的比值较大，因此高度下降的频率较少，宽度下降的频率较多。可以参考PaddleOCR中 [MobileNetV3的修改](../../../../ppocr/modeling/backbones/rec_mobilenet_v3.py)。
+
+#### 2、长中文文本识别
+
+- **问题描述**
+  中文识别模型在训练时的最大分辨率为[3,32,320]，如果待识别的文本图像过长，如下图所示，该如何适配？
+
+  ![img](./images/long_text_examples.jpg)
+
+- **小技巧**
+
+在训练时，不要直接将训练样本resize到[3,32,320]，先将样本的高度resize为32，并保持宽高比，当宽度小于320时，超出部分用0填充。另外，当样本的宽高比大于10时，这些样本将被忽略。对一张图片进行预测时，同上，但不限制最大宽高比。对一批图像进行预测时，按照训练的方式进行，但调整后的目标宽度是该批图像的最长宽度。 [代码如下](../../../../tools/infer/predict_rec.py)：
+
+  ```python linenums="1"
+  def resize_norm_img(self, img, max_wh_ratio):
+      imgC, imgH, imgW = self.rec_image_shape
+      assert imgC == img.shape[2]
+      if self.character_type == "ch":
+          imgW = int((32 * max_wh_ratio))
+      h, w = img.shape[:2]
+      ratio = w / float(h)
+      if math.ceil(imgH * ratio) > imgW:
+          resized_w = imgW
+      else:
+          resized_w = int(math.ceil(imgH * ratio))
+      resized_image = cv2.resize(img, (resized_w, imgH))
+      resized_image = resized_image.astype('float32')
+      resized_image = resized_image.transpose((2, 0, 1)) / 255
+      resized_image -= 0.5
+      resized_image /= 0.5
+      padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
+      padding_im[:, :, 0:resized_w] = resized_image
+      return padding_im
+  ```
+
+#### 3、空格识别
+
+- **问题描述**
+
+如下图所示，对于中英文混合场景，为了方便阅读和使用识别结果，经常需要识别单词之间的空格，这种情况该如何适配？
+
+![img](./images/en_paper.jpg)
+
+- **小技巧**
+
+空格识别有两种可能的方法。（1）优化文本检测。为了将检测结果中的文本分割在空格处，在对数据进行标记时，需要将带有空格的文本行分成许多段
diff --git a/docs/version2.x/ppocr/overview.en.md b/docs/version2.x/ppocr/overview.en.md
new file mode 100644
index 0000000..5ca01e8
--- /dev/null
+++ b/docs/version2.x/ppocr/overview.en.md
@@ -0,0 +1,95 @@
+---
+comments: true
+---
+
+# PP-OCR
+
+## 1. Introduction
+
+PP-OCR is a self-developed practical ultra-lightweight OCR system, which is slimed and optimized based on the reimplemented [academic algorithms](../algorithm/overview.en.md), considering the balance between **accuracy** and **speed**.
+
+### PP-OCR
+
+PP-OCR is a two-stage OCR system, in which the text detection algorithm is [DB](../algorithm/text_detection/algorithm_det_db.en.md), and the text recognition algorithm is [CRNN](../algorithm/text_recognition/algorithm_rec_crnn.en.md). Besides, a [text direction classifier](./model_train/angle_class.en.md) is added between the detection and recognition modules to deal with text in different directions.
+
+PP-OCR pipeline is as follows:
+
+![](./images/ppocrv2_framework.jpg)
+
+PP-OCR system is in continuous optimization. At present, PP-OCR and PP-OCRv2 have been released:
+
+PP-OCR adopts 19 effective strategies from 8 aspects including backbone network selection and adjustment, prediction head design, data augmentation, learning rate transformation strategy, regularization parameter selection, pre-training model use, and automatic model tailoring and quantization to optimize and slim down the models of each module (as shown in the green box above). The final results are an ultra-lightweight Chinese and English OCR model with an overall size of 3.5M and a 2.8M English digital OCR model. For more details, please refer to [PP-OCR technical report](https://arxiv.org/abs/2009.09941).
+
+### PP-OCRv2
+
+On the basis of PP-OCR, PP-OCRv2 is further optimized in five aspects. The detection model adopts CML(Collaborative Mutual Learning) knowledge distillation strategy and CopyPaste data expansion strategy. The recognition model adopts LCNet lightweight backbone network, U-DML knowledge distillation strategy and [enhanced CTC loss](./blog/enhanced_ctc_loss.en.md) function improvement (as shown in the red box above), which further improves the inference speed and prediction effect. For more details, please refer to [PP-OCRv2 technical report](https://arxiv.org/abs/2109.03144).
+
+### PP-OCRv3
+
+PP-OCRv3 upgraded the detection model and recognition model in 9 aspects based on PP-OCRv2:
+
+- PP-OCRv3 detector upgrades the CML(Collaborative Mutual Learning) text detection strategy proposed in PP-OCRv2, and further optimizes the effect of teacher model and student model respectively. In the optimization of teacher model, a PAN module with large receptive field named LK-PAN is proposed and the DML distillation strategy is adopted; In the optimization of student model, a FPN module with residual attention mechanism named RSE-FPN is proposed.
+- PP-OCRv3 recognizer is optimized based on text recognition algorithm [SVTR](https://arxiv.org/abs/2205.00159). SVTR no longer adopts RNN by introducing transformers structure, which can mine the context information of text line image more effectively, so as to improve the ability of text recognition. PP-OCRv3 adopts lightweight text recognition network SVTR_LCNet, guided training of CTC by attention, data augmentation strategy TextConAug, better pre-trained model by self-supervised TextRotNet, UDML(Unified Deep Mutual Learning), and UIM (Unlabeled Images Mining) to accelerate the model and improve the effect.
+
+PP-OCRv3 pipeline is as follows:
+
+![](./images/ppocrv3_framework.png)
+
+For more details, please refer to [PP-OCRv3 technical report](https://arxiv.org/abs/2206.03001v2).
+
+## 2. Features
+
+- Ultra lightweight PP-OCRv3 series models: detection (3.6M) + direction classifier (1.4M) + recognition 12M) = 17.0M
+- Ultra lightweight PP-OCRv2 series models: detection (3.1M) + direction classifier (1.4M) + recognition 8.5M) = 13.0M
+- Ultra lightweight PP-OCR mobile series models: detection (3.0M) + direction classifier (1.4M) + recognition (5.0M) = 9.4M
+- General PP-OCR server series models: detection (47.1M) + direction classifier (1.4M) + recognition (94.9M) = 143.4M
+- Support Chinese, English, and digit recognition, vertical text recognition, and long text recognition
+- Support multi-lingual recognition: about 80 languages like Korean, Japanese, German, French, etc
+
+## 3. Benchmark
+
+For the performance comparison between PP-OCR series models, please check the [benchmark](../legacy/benchmark.en.md) documentation.
+
+## 4. Visualization [more](./visualization.en.md)
+
+### PP-OCRv3 Chinese model
+
+![](./images/PP-OCRv3/ch/PP-OCRv3-pic001.jpg)
+![](./images/PP-OCRv3/ch/PP-OCRv3-pic002.jpg)
+![](./images/PP-OCRv3/ch/PP-OCRv3-pic003.jpg)
+
+### PP-OCRv3 English model
+
+![](./images/PP-OCRv3/en/en_1.png)
+![](./images/PP-OCRv3/en/en_2.png)
+
+### PP-OCRv3 Multilingual model
+
+![](./images/PP-OCRv3/multi_lang/japan_2.jpg)
+![](./images/PP-OCRv3/multi_lang/korean_1.jpg)
+
+## 5. Tutorial
+
+### 5.1 Quick start
+
+- You can also quickly experience the ultra-lightweight OCR : [Online Experience](https://www.paddlepaddle.org.cn/hub/scene/ocr)
+- Mobile DEMO experience (based on EasyEdge and Paddle-Lite, supports iOS and Android systems): [Sign in to the website to obtain the QR code for  installing the App](https://ai.baidu.com/easyedge/app/openSource?from=paddlelite)
+- One line of code quick use: [Quick Start](./quick_start.en.md)
+
+### 5.2 Model training / compression / deployment
+
+For more tutorials, including model training, model compression, deployment, etc., please refer to [tutorials](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6.1/README.md#Tutorials).
+
+## 6. Model zoo
+
+## PP-OCR Series Model List（Update on 2022.04.28）
+
+| Model introduction  | Model name | Recommended scene | Detection model  | Direction classifier | Recognition model  |
+| ----- | --- | --- | ------ | ---- | ------ |
+| Chinese and English ultra-lightweight PP-OCRv3 model（16.2M） | ch_PP-OCRv3_xx          | Mobile & Server   | [inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar) / [trained model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams)         | [inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) | [inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar)                 |
+| English ultra-lightweight PP-OCRv3 model（13.4M）             | en_PP-OCRv3_xx          | Mobile & Server   | [inference model](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_distill_train.tar)         | [inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) | [inference model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv3_mobile_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_train.tar)                 |
+| Chinese and English ultra-lightweight PP-OCRv2 model（11.6M） | ch_PP-OCRv2_xx          | Mobile & Server   | [inference model](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_distill_train.tar)         | [inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) | [inference model](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_train.tar)                 |
+| Chinese and English ultra-lightweight PP-OCR model (9.4M)     | ch_ppocr_mobile_v2.0_xx | Mobile & server   | [inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_train.tar) | [inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) | [inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_train.tar) |
+| Chinese and English general PP-OCR model (143.4M)             | ch_ppocr_server_v2.0_xx | Server            | [inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_train.tar) | [inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) | [inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_train.tar) |
+
+For more model downloads (including multiple languages), please refer to [PP-OCR series model downloads](./model_list.en.md).
diff --git a/docs/version2.x/ppocr/overview.md b/docs/version2.x/ppocr/overview.md
new file mode 100644
index 0000000..e8f9a7a
--- /dev/null
+++ b/docs/version2.x/ppocr/overview.md
@@ -0,0 +1,104 @@
+---
+comments: true
+---
+
+# PP-OCR
+
+## 1. 简介
+
+PP-OCR是PaddleOCR自研的实用的超轻量OCR系统。在实现[前沿算法](../algorithm/overview.md)的基础上，考虑精度与速度的平衡，进行**模型瘦身**和**深度优化**，使其尽可能满足产业落地需求。
+
+### PP-OCR
+
+PP-OCR是一个两阶段的OCR系统，其中文本检测算法选用[DB](../algorithm/text_detection/algorithm_det_db.md)，文本识别算法选用[CRNN](../algorithm/text_recognition/algorithm_rec_crnn.md)，并在检测和识别模块之间添加[文本方向分类器](./model_train/angle_class.md)，以应对不同方向的文本识别。
+
+PP-OCR系统pipeline如下：
+
+![](./images/ppocrv2_framework.jpg)
+
+PP-OCR系统在持续迭代优化，目前已发布PP-OCR和PP-OCRv2两个版本：
+
+PP-OCR从骨干网络选择和调整、预测头部的设计、数据增强、学习率变换策略、正则化参数选择、预训练模型使用以及模型自动裁剪量化8个方面，采用19个有效策略，对各个模块的模型进行效果调优和瘦身(如绿框所示)，最终得到整体大小为3.5M的超轻量中英文OCR和2.8M的英文数字OCR。更多细节请参考[PP-OCR技术报告](https://arxiv.org/abs/2009.09941)。
+
+### PP-OCRv2
+
+PP-OCRv2在PP-OCR的基础上，进一步在5个方面重点优化，检测模型采用CML协同互学习知识蒸馏策略和CopyPaste数据增广策略；识别模型采用LCNet轻量级骨干网络、UDML 改进知识蒸馏策略和[Enhanced CTC loss](./blog/enhanced_ctc_loss.md)损失函数改进（如上图红框所示），进一步在推理速度和预测效果上取得明显提升。更多细节请参考[PP-OCRv2技术报告](https://arxiv.org/abs/2109.03144)。
+
+### PP-OCRv3
+
+PP-OCRv3在PP-OCRv2的基础上，针对检测模型和识别模型，进行了共计9个方面的升级：
+
+- PP-OCRv3检测模型对PP-OCRv2中的CML协同互学习文本检测蒸馏策略进行了升级，分别针对教师模型和学生模型进行进一步效果优化。其中，在对教师模型优化时，提出了大感受野的PAN结构LK-PAN和引入了DML蒸馏策略；在对学生模型优化时，提出了残差注意力机制的FPN结构RSE-FPN。
+- PP-OCRv3的识别模块是基于文本识别算法[SVTR](https://arxiv.org/abs/2205.00159)优化。SVTR不再采用RNN结构，通过引入Transformers结构更加有效地挖掘文本行图像的上下文信息，从而提升文本识别能力。PP-OCRv3通过轻量级文本识别网络SVTR_LCNet、Attention损失指导CTC损失训练策略、挖掘文字上下文信息的数据增广策略TextConAug、TextRotNet自监督预训练模型、UDML联合互学习策略、UIM无标注数据挖掘方案，6个方面进行模型加速和效果提升。
+
+PP-OCRv3系统pipeline如下：
+
+![](./images/ppocrv3_framework.png)
+
+更多细节请参考[PP-OCRv3技术报告](https://arxiv.org/abs/2206.03001v2) 👉[中文简洁版](./blog/PP-OCRv3_introduction.md)
+
+### PP-OCRv4
+
+- PP-OCRv4-mobile：速度可比情况下，中文场景效果相比于 PP-OCRv3 再提升 4.5%，英文场景提升 10%，80 语种多语言模型平均识别准确率提升 8%以上
+- PP-OCRv4-server：发布了目前精度最高的 OCR 模型，中英文场景上检测模型精度提升 4.9%， 识别模型精度提升 2%
+
+更多细节参见：[PP-OCRv4技术报告](./blog/PP-OCRv4_introduction.md)
+
+## 2. 特性
+
+- 超轻量PP-OCRv3系列：检测（3.6M）+ 方向分类器（1.4M）+ 识别（12M）= 17.0M
+- 超轻量PP-OCRv2系列：检测（3.1M）+ 方向分类器（1.4M）+ 识别（8.5M）= 13.0M
+- 超轻量PP-OCR mobile移动端系列：检测（3.0M）+方向分类器（1.4M）+ 识别（5.0M）= 9.4M
+- 通用PP-OCR server系列：检测（47.1M）+方向分类器（1.4M）+ 识别（94.9M）= 143.4M
+- 支持中英文数字组合识别、竖排文本识别、长文本识别
+- 支持多语言识别：韩语、日语、德语、法语等约80种语言
+
+## 3. benchmark
+
+关于PP-OCR系列模型之间的性能对比，请查看[benchmark](../legacy/benchmark.md)文档。
+
+## 4. 效果展示 [more](./visualization.md)
+
+### PP-OCRv3 中文模型
+
+![](./images/PP-OCRv3/ch/PP-OCRv3-pic001.jpg)
+
+![](./images/PP-OCRv3/ch/PP-OCRv3-pic002.jpg)
+
+![](./images/PP-OCRv3/ch/PP-OCRv3-pic003.jpg)
+
+### PP-OCRv3 英文模型
+
+![](./images/PP-OCRv3/en/en_1.png)
+![](./images/PP-OCRv3/en/en_2.png)
+
+### PP-OCRv3 多语言模型
+
+![](./images/PP-OCRv3/multi_lang/japan_2.jpg)
+![](./images/PP-OCRv3/multi_lang/korean_1.jpg)
+
+## 5. 使用教程
+
+### 5.1 快速体验
+
+- 在线网站体验：超轻量PP-OCR mobile模型体验地址：<https://www.paddlepaddle.org.cn/hub/scene/ocr>
+- 移动端demo体验：[安装包DEMO下载地址](https://ai.baidu.com/easyedge/app/openSource?from=paddlelite)(基于EasyEdge和Paddle-Lite, 支持iOS和Android系统)
+- 一行命令快速使用：[快速开始（中英文/多语言）](./quick_start.md)
+
+### 5.2 模型训练、压缩、推理部署
+
+更多教程，包括模型训练、模型压缩、推理部署等，
+
+## 6. 模型库
+
+PP-OCR中英文模型列表如下：
+
+| 模型简介  | 模型名称   | 推荐场景  | 检测模型   | 方向分类器     | 识别模型    |
+| ----- | ----- | --------------- | ---- | -------------- | --- |
+| 中英文超轻量PP-OCRv3模型（16.2M）     | ch_PP-OCRv3_xx          | 移动端&服务器端 | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar) / [训练模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams)           | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar)   | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_train.tar)                 |
+| 英文超轻量PP-OCRv3模型（13.4M）       | en_PP-OCRv3_xx          | 移动端&服务器端 | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_distill_train.tar)           | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar)   | [推理模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv3_mobile_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_train.tar)                 |
+| 中英文超轻量PP-OCRv2模型（13.0M）     | ch_PP-OCRv2_xx          | 移动端&服务器端 | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_distill_train.tar)           | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar)   | [推理模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_train.tar)                 |
+| 中英文超轻量PP-OCR mobile模型（9.4M） | ch_ppocr_mobile_v2.0_xx | 移动端&服务器端 | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_train.tar) | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_pre.tar) |
+| 中英文通用PP-OCR server模型（143.4M） | ch_ppocr_server_v2.0_xx | 服务器端        | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_train.tar) | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_train.tar) | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar) / [预训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_pre.tar) |
+
+更多模型下载（包括英文数字模型、多语言模型、Paddle-Lite模型等），可以参考[PP-OCR 系列模型下载](./model_list.md)。
diff --git a/docs/version2.x/ppocr/quick_start.en.md b/docs/version2.x/ppocr/quick_start.en.md
new file mode 100644
index 0000000..e5e180e
--- /dev/null
+++ b/docs/version2.x/ppocr/quick_start.en.md
@@ -0,0 +1,284 @@
+---
+comments: true
+---
+
+# PaddleOCR Quick Start
+
+**Note:** This tutorial mainly introduces the usage of PP-OCR series models, please refer to [PP-Structure Quick Start](../ppstructure/overview.en.md) for the quick use of document analysis related functions. In addition, the All-in-One development tool PaddleX relies on the advanced technology of PaddleOCR to support low-code full-process development capabilities in the OCR field, significantly reducing development time and complexity. It also integrates the 17 models involved in text image intelligent analysis, OCR, layout parsing, table recognition, formula recognition, and seal text recognition into 6 pipelines, which can be invoked with a simple Python API. For more details, please see [Low-Code Full-Process Development](https://www.paddleocr.ai/v2.10.0/en/paddlex/quick_start.html).
+
+## 1. Installation
+
+### 1.1 Install PaddlePaddle
+
+> If you do not have a Python environment, please refer to [Environment Preparation](./environment.en.md).
+
+- If you have CUDA 11 installed on your machine, please run the following command to install
+
+  ```bash linenums="1"
+  pip install "paddlepaddle-gpu<=2.6"
+  ```
+
+- If you have no available GPU on your machine, please run the following command to install the CPU version
+
+  ```bash linenums="1"
+  python -m pip install "paddlepaddle<=2.6"
+  ```
+
+For more software version requirements, please refer to the instructions in [Installation Document](https://www.paddlepaddle.org.cn/en/install/quick) for operation.
+
+### 1.2 Install PaddleOCR Whl Package
+
+```bash linenums="1"
+pip install "paddleocr>=2.0.1" # Recommend to use version 2.0.1+
+```
+
+- **For windows users:** If you getting this error `OSError: [WinError 126] The specified module could not be found` when you install shapely on windows. Please try to download Shapely whl file [here](http://www.lfd.uci.edu/~gohlke/pythonlibs/#shapely).
+
+  Reference: [Solve shapely installation on windows](https://stackoverflow.com/questions/44398265/install-shapely-oserror-winerror-126-the-specified-module-could-not-be-found)
+
+## 2. Easy-to-Use
+
+### 2.1 Use by Command Line
+
+PaddleOCR provides a series of test images, click [here](https://paddleocr.bj.bcebos.com/dygraph_v2.1/ppocr_img.zip) to download, and then switch to the corresponding directory in the terminal
+
+```bash linenums="1"
+cd /path/to/ppocr_img
+```
+
+If you do not use the provided test image, you can replace the following `--image_dir` parameter with the corresponding test image path
+
+#### 2.1.1 Chinese and English Model
+
+- Detection, direction classification and recognition: set the parameter`--use_gpu false` to disable the gpu device
+
+  ```bash linenums="1"
+  paddleocr --image_dir ./imgs_en/img_12.jpg --use_angle_cls true --lang en --use_gpu false
+  ```
+
+  Output will be a list, each item contains bounding box, text and recognition confidence
+
+  ```bash linenums="1"
+  [[[441.0, 174.0], [1166.0, 176.0], [1165.0, 222.0], [441.0, 221.0]], ('ACKNOWLEDGEMENTS', 0.9971134662628174)]
+  [[[403.0, 346.0], [1204.0, 348.0], [1204.0, 384.0], [402.0, 383.0]], ('We would like to thank all the designers and', 0.9761400818824768)]
+  [[[403.0, 396.0], [1204.0, 398.0], [1204.0, 434.0], [402.0, 433.0]], ('contributors who have been involved in the', 0.9791957139968872)]
+  ......
+  ```
+
+  pdf file is also supported, you can infer the first few pages by using the `page_num` parameter, the default is 0, which means infer all pages
+
+  ```bash linenums="1"
+  paddleocr --image_dir ./xxx.pdf --use_angle_cls true --use_gpu false --page_num 2
+  ```
+
+- Only detection: set `--rec` to `false`
+
+  ```bash linenums="1"
+  paddleocr --image_dir ./imgs_en/img_12.jpg --rec false
+  ```
+
+  Output will be a list, each item only contains bounding box
+
+  ```bash linenums="1"
+  [[397.0, 802.0], [1092.0, 802.0], [1092.0, 841.0], [397.0, 841.0]]
+  [[397.0, 750.0], [1211.0, 750.0], [1211.0, 789.0], [397.0, 789.0]]
+  [[397.0, 702.0], [1209.0, 698.0], [1209.0, 734.0], [397.0, 738.0]]
+  ......
+  ```
+
+- Only recognition: set `--det` to `false`
+
+  ```bash linenums="1"
+  paddleocr --image_dir ./imgs_words_en/word_10.png --det false --lang en
+  ```
+
+  Output will be a list, each item contains text and recognition confidence
+
+  ```bash linenums="1"
+  ['PAIN', 0.9934559464454651]
+  ```
+
+**Version**
+paddleocr uses the PP-OCRv4 model by default(`--ocr_version PP-OCRv4`). If you want to use other versions, you can set the parameter `--ocr_version`, the specific version description is as follows:
+
+|  version name |  description |
+|    ---    |   ---   |
+| PP-OCRv4 | support Chinese and English detection and recognition, direction classifier, support multilingual recognition |
+| PP-OCRv3 | support Chinese and English detection and recognition, direction classifier, support multilingual recognition |
+| PP-OCRv2 | only supports Chinese and English detection and recognition, direction classifier, multilingual model is not updated |
+| PP-OCR   | support Chinese and English detection and recognition, direction classifier, support multilingual recognition |
+
+If you want to add your own trained model, you can add model links and keys in [paddleocr](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.10/paddleocr.py) and recompile.
+
+More whl package usage can be found in [whl package](./blog/whl.en.md)
+
+#### 2.1.2 Multi-language Model
+
+PaddleOCR currently supports 80 languages, which can be switched by modifying the `--lang` parameter.
+
+``` bash
+paddleocr --image_dir ./doc/imgs_en/254.jpg --lang=en
+```
+
+![](./images/254.jpg)
+
+![](./images/multi_lang/img_02.jpg)
+
+The result is a list, each item contains a text box, text and recognition confidence
+
+```text linenums="1"
+[[[67.0, 51.0], [327.0, 46.0], [327.0, 74.0], [68.0, 80.0]], ('PHOCAPITAL', 0.9944712519645691)]
+[[[72.0, 92.0], [453.0, 84.0], [454.0, 114.0], [73.0, 122.0]], ('107 State Street', 0.9744491577148438)]
+[[[69.0, 135.0], [501.0, 125.0], [501.0, 156.0], [70.0, 165.0]], ('Montpelier Vermont', 0.9357033967971802)]
+......
+```
+
+Commonly used multilingual abbreviations include
+
+| Language            | Abbreviation |      | Language | Abbreviation |      | Language | Abbreviation |
+| ------------------- | ------------ | ---- | -------- | ------------ | ---- | -------- | ------------ |
+| Chinese & English   | ch           |      | French   | fr           |      | Japanese | japan        |
+| English             | en           |      | German   | german       |      | Korean   | korean       |
+| Chinese Traditional | chinese_cht  |      | Italian  | it           |      | Russian  | ru           |
+
+A list of all languages and their corresponding abbreviations can be found in [Multi-Language Model Tutorial](./blog/multi_languages.en.md)
+
+### 2.2 Use by Code
+
+#### 2.2.1 Chinese & English Model and Multilingual Model
+
+- detection, angle classification and recognition:
+
+```python linenums="1"
+from paddleocr import PaddleOCR,draw_ocr
+# Paddleocr supports Chinese, English, French, German, Korean and Japanese.
+# You can set the parameter `lang` as `ch`, `en`, `fr`, `german`, `korean`, `japan`
+# to switch the language model in order.
+ocr = PaddleOCR(use_angle_cls=True, lang='en') # need to run only once to download and load model into memory
+img_path = './imgs_en/img_12.jpg'
+result = ocr.ocr(img_path, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+
+
+# draw result
+from PIL import Image
+result = result[0]
+image = Image.open(img_path).convert('RGB')
+boxes = [line[0] for line in result]
+txts = [line[1][0] for line in result]
+scores = [line[1][1] for line in result]
+im_show = draw_ocr(image, boxes, txts, scores, font_path='./fonts/simfang.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+Output will be a list, each item contains bounding box, text and recognition confidence
+
+```bash linenums="1"
+[[[441.0, 174.0], [1166.0, 176.0], [1165.0, 222.0], [441.0, 221.0]], ('ACKNOWLEDGEMENTS', 0.9971134662628174)]
+  [[[403.0, 346.0], [1204.0, 348.0], [1204.0, 384.0], [402.0, 383.0]], ('We would like to thank all the designers and', 0.9761400818824768)]
+  [[[403.0, 396.0], [1204.0, 398.0], [1204.0, 434.0], [402.0, 433.0]], ('contributors who have been involved in the', 0.9791957139968872)]
+  ......
+```
+
+Visualization of results
+
+![](./blog/images/11_det_rec.jpg)
+
+If the input is a PDF file, you can refer to the following code for visualization
+
+```python linenums="1"
+from paddleocr import PaddleOCR, draw_ocr
+
+# Paddleocr supports Chinese, English, French, German, Korean and Japanese.
+# You can set the parameter `lang` as `ch`, `en`, `fr`, `german`, `korean`, `japan`
+# to switch the language model in order.
+PAGE_NUM = 10 # Set the recognition page number
+pdf_path = 'default.pdf'
+ocr = PaddleOCR(use_angle_cls=True, lang="ch", page_num=PAGE_NUM)  # need to run only once to download and load model into memory
+# ocr = PaddleOCR(use_angle_cls=True, lang="ch", page_num=PAGE_NUM,use_gpu=0) # To Use GPU,uncomment this line and comment the above one.
+result = ocr.ocr(pdf_path, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    if res == None: # Skip when empty result detected to avoid TypeError:NoneType
+        print(f"[DEBUG] Empty page {idx+1} detected, skip it.")
+        continue
+    for line in res:
+        print(line)
+
+# draw the result
+import fitz
+from PIL import Image
+import cv2
+import numpy as np
+imgs = []
+with fitz.open(pdf_path) as pdf:
+    for pg in range(0, PAGE_NUM):
+        page = pdf[pg]
+        mat = fitz.Matrix(2, 2)
+        pm = page.get_pixmap(matrix=mat, alpha=False)
+        # if width or height > 2000 pixels, don't enlarge the image
+        if pm.width > 2000 or pm.height > 2000:
+            pm = page.get_pixmap(matrix=fitz.Matrix(1, 1), alpha=False)
+        img = Image.frombytes("RGB", [pm.width, pm.height], pm.samples)
+        img = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
+        imgs.append(img)
+for idx in range(len(result)):
+    res = result[idx]
+    if res == None:
+        continue
+    image = imgs[idx]
+    boxes = [line[0] for line in res]
+    txts = [line[1][0] for line in res]
+    scores = [line[1][1] for line in res]
+    im_show = draw_ocr(image, boxes, txts, scores, font_path='doc/fonts/simfang.ttf')
+    im_show = Image.fromarray(im_show)
+    im_show.save('result_page_{}.jpg'.format(idx))
+```
+
+- Detection and Recognition Using Sliding Windows
+
+To perform OCR using sliding windows, the following code snippet can be employed:
+
+```python linenums="1"
+from paddleocr import PaddleOCR
+from PIL import Image, ImageDraw, ImageFont
+
+# Initialize OCR engine
+ocr = PaddleOCR(use_angle_cls=True, lang="en")
+
+img_path = "./very_large_image.jpg"
+slice = {'horizontal_stride': 300, 'vertical_stride': 500, 'merge_x_thres': 50, 'merge_y_thres': 35}
+results = ocr.ocr(img_path, cls=True, slice=slice)
+
+# Load image
+image = Image.open(img_path).convert("RGB")
+draw = ImageDraw.Draw(image)
+font = ImageFont.truetype("./doc/fonts/simfang.ttf", size=20)  # Adjust size as needed
+
+# Process and draw results
+for res in results:
+    for line in res:
+        box = [tuple(point) for point in line[0]]
+        # Finding the bounding box
+        box = [(min(point[0] for point in box), min(point[1] for point in box)),
+               (max(point[0] for point in box), max(point[1] for point in box))]
+        txt = line[1][0]
+        draw.rectangle(box, outline="red", width=2)  # Draw rectangle
+        draw.text((box[0][0], box[0][1] - 25), txt, fill="blue", font=font)  # Draw text above the box
+
+# Save result
+image.save("result.jpg")
+
+```
+
+This example initializes the PaddleOCR instance with angle classification enabled and sets the language to English. The `ocr` method is then called with several parameters to customize the detection and recognition process, including the `slice` parameter for handling image slices.
+
+For a more comprehensive understanding of the slicing operation, please refer to the [slice operation documentation](./blog/slice.en.md).
+
+## 3. Summary
+
+In this section, you have mastered the use of PaddleOCR whl package.
diff --git a/docs/version2.x/ppocr/quick_start.md b/docs/version2.x/ppocr/quick_start.md
new file mode 100644
index 0000000..5c6c8ab
--- /dev/null
+++ b/docs/version2.x/ppocr/quick_start.md
@@ -0,0 +1,278 @@
+---
+comments: true
+---
+
+
+# PaddleOCR 快速开始
+
+ **说明：** 本文主要介绍PaddleOCR wheel包对PP-OCR系列模型的快速使用。如要体验文档分析相关功能，请参考[PP-Structure快速使用教程](../ppstructure/overview.md)。此外，飞桨低代码开发工具<b>PaddleX</b>依托PaddleOCR的先进技术，支持了OCR领域的<b>低代码全流程开发能力，大幅减少开发时间和难度</b>，同时将<b>文本图像智能分析、通用OCR、通用版面解析、通用表格识别、公式识别、印章文本识别</b>涉及的17个模型整合为6条模型产线，通过极简的Python API一键调用,详情请见[低代码全流程开发](https://www.paddleocr.ai/v2.10.0/paddlex/quick_start.html)。
+
+
+## 1. 安装
+
+### 1.1 安装PaddlePaddle
+
+> 如果您没有基础的Python运行环境，请参考[运行环境准备](./environment.md)。
+
+- 您的机器安装的是CUDA 11，请运行以下命令安装
+
+  ```bash linenums="1"
+  pip install "paddlepaddle-gpu<=2.6"
+  ```
+
+- 您的机器是CPU，请运行以下命令安装
+
+  ```bash linenums="1"
+  pip install "paddlepaddle<=2.6"
+  ```
+
+更多的版本需求，请参照[飞桨官网安装文档](https://www.paddlepaddle.org.cn/install/quick)中的说明进行操作。
+
+### 1.2 安装PaddleOCR whl包
+
+```bash linenums="1"
+pip install "paddleocr<3.0"
+```
+
+- 对于Windows环境用户：直接通过pip安装的shapely库可能出现`[winError 126] 找不到指定模块的问题`。建议从[这里](https://www.lfd.uci.edu/~gohlke/pythonlibs/#shapely)下载shapely安装包完成安装。
+
+## 2. 便捷使用
+
+### 2.1 命令行使用
+
+PaddleOCR提供了一系列测试图片，点击[这里](https://paddleocr.bj.bcebos.com/dygraph_v2.1/ppocr_img.zip)下载并解压，然后在终端中切换到相应目录
+
+```bash linenums="1"
+cd /path/to/ppocr_img
+```
+
+如果不使用提供的测试图片，可以将下方`--image_dir`参数替换为相应的测试图片路径。
+
+#### 2.1.1 中英文模型
+
+- 检测+方向分类器+识别全流程：`--use_angle_cls true`设置使用方向分类器识别180度旋转文字，`--use_gpu false`设置不使用GPU
+
+  ```bash linenums="1"
+  paddleocr --image_dir ./imgs/11.jpg --use_angle_cls true --use_gpu false
+  ```
+
+  结果是一个list，每个item包含了文本框，文字和识别置信度
+
+  ```bash linenums="1"
+  [[[28.0, 37.0], [302.0, 39.0], [302.0, 72.0], [27.0, 70.0]], ('纯臻营养护发素', 0.9658738374710083)]
+  ......
+  ```
+
+  此外，paddleocr也支持输入pdf文件，并且可以通过指定参数`page_num`来控制推理前面几页，默认为0，表示推理所有页。
+
+  ```bash linenums="1"
+  paddleocr --image_dir ./xxx.pdf --use_angle_cls true --use_gpu false --page_num 2
+  ```
+
+- 单独使用检测：设置`--rec`为`false`
+
+  ```bash linenums="1"
+  paddleocr --image_dir ./imgs/11.jpg --rec false
+  ```
+
+  结果是一个list，每个item只包含文本框
+
+  ```bash linenums="1"
+  [[27.0, 459.0], [136.0, 459.0], [136.0, 479.0], [27.0, 479.0]]
+  [[28.0, 429.0], [372.0, 429.0], [372.0, 445.0], [28.0, 445.0]]
+  ......
+  ```
+
+- 单独使用识别：设置`--det`为`false`
+
+  ```bash linenums="1"
+  paddleocr --image_dir ./imgs_words/ch/word_1.jpg --det false
+  ```
+
+  结果是一个list，每个item只包含识别结果和识别置信度
+
+  ```bash linenums="1"
+  ['韩国小馆', 0.994467]
+  ```
+
+**版本说明**
+paddleocr默认使用PP-OCRv4模型(`--ocr_version PP-OCRv4`)，如需使用其他版本可通过设置参数`--ocr_version`，具体版本说明如下：
+
+| 版本名称 | 版本说明                                           |
+| -------- | -------------------------------------------------- |
+| PP-OCRv4 | 支持中、英文检测和识别，方向分类器，支持多语种识别 |
+| PP-OCRv3 | 支持中、英文检测和识别，方向分类器，支持多语种识别 |
+| PP-OCRv2 | 支持中英文的检测和识别，方向分类器，多语言暂未更新 |
+| PP-OCR   | 支持中、英文检测和识别，方向分类器，支持多语种识别 |
+
+如需新增自己训练的模型，可以在[paddleocr](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.10/paddleocr.py)中增加模型链接和字段，重新编译即可。
+
+更多whl包使用可参考[whl包文档](./blog/whl.md)
+
+#### 2.1.2 多语言模型
+
+PaddleOCR目前支持80个语种，可以通过修改`--lang`参数进行切换，对于英文模型，指定`--lang=en`。
+
+``` bash
+paddleocr --image_dir ./imgs_en/254.jpg --lang=en
+```
+
+![](./images/254.jpg)
+
+![](./images/multi_lang/img_02.jpg)
+
+结果是一个list，每个item包含了文本框，文字和识别置信度
+
+```text linenums="1"
+[[[67.0, 51.0], [327.0, 46.0], [327.0, 74.0], [68.0, 80.0]], ('PHOCAPITAL', 0.9944712519645691)]
+[[[72.0, 92.0], [453.0, 84.0], [454.0, 114.0], [73.0, 122.0]], ('107 State Street', 0.9744491577148438)]
+[[[69.0, 135.0], [501.0, 125.0], [501.0, 156.0], [70.0, 165.0]], ('Montpelier Vermont', 0.9357033967971802)]
+......
+```
+
+常用的多语言简写包括
+
+| 语种     | 缩写        |     | 语种     | 缩写   |     | 语种     | 缩写   |
+| -------- | ----------- | --- | -------- | ------ | --- | -------- | ------ |
+| 中文     | ch          |     | 法文     | fr     |     | 日文     | japan  |
+| 英文     | en          |     | 德文     | german |     | 韩文     | korean |
+| 繁体中文 | chinese_cht |     | 意大利文 | it     |     | 俄罗斯文 | ru     |
+
+全部语种及其对应的缩写列表可查看[多语言模型教程](./blog/multi_languages.md)
+
+### 2.2 Python脚本使用
+
+#### 2.2.1 中英文与多语言使用
+
+通过Python脚本使用PaddleOCR whl包，whl包会自动下载ppocr轻量级模型作为默认模型。
+
+- 检测+方向分类器+识别全流程
+
+```python linenums="1"
+from paddleocr import PaddleOCR, draw_ocr
+
+# Paddleocr目前支持的多语言语种可以通过修改lang参数进行切换
+# 例如`ch`, `en`, `fr`, `german`, `korean`, `japan`
+ocr = PaddleOCR(use_angle_cls=True, lang="ch")  # need to run only once to download and load model into memory
+img_path = './imgs/11.jpg'
+result = ocr.ocr(img_path, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    for line in res:
+        print(line)
+
+# 显示结果
+from PIL import Image
+result = result[0]
+image = Image.open(img_path).convert('RGB')
+boxes = [line[0] for line in result]
+txts = [line[1][0] for line in result]
+scores = [line[1][1] for line in result]
+im_show = draw_ocr(image, boxes, txts, scores, font_path='./fonts/simfang.ttf')
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+结果是一个list，每个item包含了文本框，文字和识别置信度
+
+```bash linenums="1"
+[[[28.0, 37.0], [302.0, 39.0], [302.0, 72.0], [27.0, 70.0]], ('纯臻营养护发素', 0.9658738374710083)]
+......
+```
+
+结果可视化
+
+![](./blog/images/11_det_rec.jpg)
+
+如果输入是PDF文件，那么可以参考下面代码进行可视化
+
+```python linenums="1"
+from paddleocr import PaddleOCR, draw_ocr
+
+# Paddleocr目前支持的多语言语种可以通过修改lang参数进行切换
+# 例如`ch`, `en`, `fr`, `german`, `korean`, `japan`
+PAGE_NUM = 10 # 将识别页码前置作为全局，防止后续打开pdf的参数和前文识别参数不一致 / Set the recognition page number
+pdf_path = 'default.pdf'
+ocr = PaddleOCR(use_angle_cls=True, lang="ch", page_num=PAGE_NUM)  # need to run only once to download and load model into memory
+# ocr = PaddleOCR(use_angle_cls=True, lang="ch", page_num=PAGE_NUM,use_gpu=0) # 如果需要使用GPU，请取消此行的注释 并注释上一行 / To Use GPU,uncomment this line and comment the above one.
+result = ocr.ocr(pdf_path, cls=True)
+for idx in range(len(result)):
+    res = result[idx]
+    if res == None: # 识别到空页就跳过，防止程序报错 / Skip when empty result detected to avoid TypeError:NoneType
+        print(f"[DEBUG] Empty page {idx+1} detected, skip it.")
+        continue
+    for line in res:
+        print(line)
+# 显示结果
+import fitz
+from PIL import Image
+import cv2
+import numpy as np
+imgs = []
+with fitz.open(pdf_path) as pdf:
+    for pg in range(0, PAGE_NUM):
+        page = pdf[pg]
+        mat = fitz.Matrix(2, 2)
+        pm = page.get_pixmap(matrix=mat, alpha=False)
+        # if width or height > 2000 pixels, don't enlarge the image
+        if pm.width > 2000 or pm.height > 2000:
+            pm = page.get_pixmap(matrix=fitz.Matrix(1, 1), alpha=False)
+        img = Image.frombytes("RGB", [pm.width, pm.height], pm.samples)
+        img = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
+        imgs.append(img)
+for idx in range(len(result)):
+    res = result[idx]
+    if res == None:
+        continue
+    image = imgs[idx]
+    boxes = [line[0] for line in res]
+    txts = [line[1][0] for line in res]
+    scores = [line[1][1] for line in res]
+    im_show = draw_ocr(image, boxes, txts, scores, font_path='doc/fonts/simfang.ttf')
+    im_show = Image.fromarray(im_show)
+    im_show.save('result_page_{}.jpg'.format(idx))
+```
+
+- 使用滑动窗口进行检测和识别
+
+要使用滑动窗口进行光学字符识别（OCR），可以使用以下代码片段：
+
+```python linenums="1"
+from paddleocr import PaddleOCR
+from PIL import Image, ImageDraw, ImageFont
+
+# 初始化OCR引擎
+ocr = PaddleOCR(use_angle_cls=True, lang="en")
+
+img_path = "./very_large_image.jpg"
+slice = {'horizontal_stride': 300, 'vertical_stride': 500, 'merge_x_thres': 50, 'merge_y_thres': 35}
+results = ocr.ocr(img_path, cls=True, slice=slice)
+
+# 加载图像
+image = Image.open(img_path).convert("RGB")
+draw = ImageDraw.Draw(image)
+font = ImageFont.truetype("./doc/fonts/simfang.ttf", size=20)  # 根据需要调整大小
+
+# 处理并绘制结果
+for res in results:
+    for line in res:
+        box = [tuple(point) for point in line[0]]
+        # 找出边界框
+        box = [(min(point[0] for point in box), min(point[1] for point in box)),
+               (max(point[0] for point in box), max(point[1] for point in box))]
+        txt = line[1][0]
+        draw.rectangle(box, outline="red", width=2)  # 绘制矩形
+        draw.text((box[0][0], box[0][1] - 25), txt, fill="blue", font=font)  # 在矩形上方绘制文本
+
+# 保存结果
+image.save("result.jpg")
+
+```
+
+此示例初始化了启用角度分类的PaddleOCR实例，并将语言设置为英语。然后调用`ocr`方法，并使用多个参数来自定义检测和识别过程，包括处理图像切片的`slice`参数。
+
+要更全面地了解切片操作，请参考[切片操作文档](./blog/slice.md)。
+
+## 3. 小结
+
+通过本节内容，相信您已经熟练掌握PaddleOCR whl包的使用方法并获得了初步效果。
diff --git a/docs/version2.x/ppocr/visualization.en.md b/docs/version2.x/ppocr/visualization.en.md
new file mode 100644
index 0000000..9fd6601
--- /dev/null
+++ b/docs/version2.x/ppocr/visualization.en.md
@@ -0,0 +1,55 @@
+---
+comments: true
+---
+
+# Visualization
+
+## PP-OCRv3
+
+### PP-OCRv3 Chinese model
+
+![](./images/PP-OCRv3/ch/PP-OCRv3-pic001.jpg)
+
+![](./images/PP-OCRv3/ch/PP-OCRv3-pic002.jpg)
+
+![](./images/PP-OCRv3/ch/PP-OCRv3-pic003.jpg)
+
+### PP-OCRv3 English model
+
+![](./images/PP-OCRv3/en/en_1.png)
+
+![](./images/PP-OCRv3/en/en_2.png)
+
+![](./images/PP-OCRv3/en/en_3.png)
+
+### PP-OCRv3 Multilingual model
+
+![](./images/PP-OCRv3/multi_lang/japan_2.jpg)
+
+![](./images/PP-OCRv3/multi_lang/korean_1.jpg)
+
+## PP-OCRv2
+
+![](./images/PP-OCRv2/PP-OCRv2-pic001.jpg)
+
+![](./images/PP-OCRv2/PP-OCRv2-pic002.jpg)
+
+![](./images/PP-OCRv2/PP-OCRv2-pic003.jpg)
+
+## ch_ppocr_server_2.0
+
+![](./images/ch_ppocr_mobile_v2.0/00006737.jpg)
+
+![](./images/ch_ppocr_mobile_v2.0/00009282.jpg)
+
+![](./images/ch_ppocr_mobile_v2.0/00015504.jpg)
+
+## en_ppocr_mobile_2.0
+
+![](./images/ch_ppocr_mobile_v2.0/img_12.jpg)
+
+## (multilingual)_ppocr_mobile_2.0
+
+![](./images/multi_lang/french_0.jpg)
+
+![](./images/multi_lang/korean_0.jpg)
diff --git a/docs/version2.x/ppocr/visualization.md b/docs/version2.x/ppocr/visualization.md
new file mode 100644
index 0000000..2e6180d
--- /dev/null
+++ b/docs/version2.x/ppocr/visualization.md
@@ -0,0 +1,55 @@
+---
+comments: true
+---
+
+# 效果展示
+
+## 超轻量PP-OCRv3效果展示
+
+### PP-OCRv3中文模型
+
+![](./images/PP-OCRv3/ch/PP-OCRv3-pic001.jpg)
+
+![](./images/PP-OCRv3/ch/PP-OCRv3-pic002.jpg)
+
+![](./images/PP-OCRv3/ch/PP-OCRv3-pic003.jpg)
+
+### PP-OCRv3英文数字模型
+
+![](./images/PP-OCRv3/en/en_1.png)
+
+![](./images/PP-OCRv3/en/en_2.png)
+
+![](./images/PP-OCRv3/en/en_3.png)
+
+### PP-OCRv3多语言模型
+
+![](./images/PP-OCRv3/multi_lang/japan_2.jpg)
+
+![](./images/PP-OCRv3/multi_lang/korean_1.jpg)
+
+## 超轻量PP-OCRv2效果展示
+
+![](./images/PP-OCRv2/PP-OCRv2-pic001.jpg)
+
+![](./images/PP-OCRv2/PP-OCRv2-pic002.jpg)
+
+![](./images/PP-OCRv2/PP-OCRv2-pic003.jpg)
+
+## 通用PP-OCR server 效果展示
+
+![](./images/ch_ppocr_mobile_v2.0/00006737.jpg)
+
+![](./images/ch_ppocr_mobile_v2.0/00009282.jpg)
+
+![](./images/ch_ppocr_mobile_v2.0/00015504.jpg)
+
+## 英文识别模型效果展示
+
+![](./images/ch_ppocr_mobile_v2.0/img_12.jpg)
+
+## 多语言识别模型效果展示
+
+![](./images/multi_lang/french_0.jpg)
+
+![](./images/multi_lang/korean_0.jpg)
diff --git a/docs/version2.x/ppstructure/blog/how_to_do_kie.en.md b/docs/version2.x/ppstructure/blog/how_to_do_kie.en.md
new file mode 100644
index 0000000..7a3b732
--- /dev/null
+++ b/docs/version2.x/ppstructure/blog/how_to_do_kie.en.md
@@ -0,0 +1,150 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# Key Information Extraction Pipeline
+
+## 1. Introduction
+
+### 1.1 Background
+
+Key information extraction (KIE) refers to extracting key information from text or images. As the downstream task of OCR, KIE of document image has many practical application scenarios, such as form recognition, ticket information extraction, ID card information extraction, etc. However, it is time-consuming and laborious to extract  key information from these document images by manpower. It's challengable but also valuable to combine multi-modal features (visual, layout, text, etc) together and complete KIE tasks.
+
+For the document images in a specific scene, the position and layout of the key information are relatively fixed. Therefore, in the early stage of the research, there are many methods based on template matching to extract the key information. This method is still widely used in many simple scenarios at present. However, it takes long time to adjust the template for different scenarios.
+
+The KIE in the document image generally contains 2 subtasks, which is as shown follows.
+
+* (1) SER: semantic entity recognition, which classifies each detected textline, such as dividing it into name and ID No. As shown in the red boxes in the following figure.
+
+* (2) RE: relationship extraction, which matches the question and answer based on SER results. As shown in the figure below, the yellow arrows match the question and answer.
+
+![](./images/184588654-d87f54f3-13ab-42c4-afc0-da79bead3f14.png)
+
+### 1.2 Mainstream Deep-learning Solutions
+
+General KIE methods are based on Named Entity Recognition (NER), but such methods only use text information and ignore location and visual feature information, which leads to limited accuracy. In recent years, most scholars have started to combine mutil-modal features to improve the accuracy of KIE model. The main methods are as follows:
+
+* (1) Grid based methods. These methods mainly focus on the fusion of multi-modal information at the image level. Most texts are of character granularity. The text and structure information embedding method is simple, such as the algorithm of chargrid [1].
+
+* (2) Token based methods. These methods refer to the NLP methods such as Bert, which encode the position, vision and other feature information into the multi-modal model, and conduct pre-training on large-scale datasets, so that in downstream tasks, only a small amount of annotation data is required to obtain excellent results. The representative algorithms are layoutlm [2], layoutlmv2 [3], layoutxlm [4], structext [5], etc.
+
+* (3) GCN based methods. These methods try to learn the structural information between images and characters, so as to solve the problem of extracting open set information (templates not seen in the training set), such as GCN [6], SDMGR [7] and other algorithms.
+
+* (4) End to end based methods: these methods put the existing OCR character recognition and KIE information extraction tasks into a unified network for common learning, and strengthen each other in the learning process. Such as TRIE [8].
+
+For more detailed introduction of the algorithms, please refer to Chapter 6 of [Diving into OCR](https://aistudio.baidu.com/aistudio/education/group/info/25207).
+
+## 2. KIE Pipeline
+
+Token based methods such as LayoutXLM are implemented in PaddleOCR. What's more, in PP-StructureV2, we simplify the LayoutXLM model and proposed VI-LayoutXLM, in which the visual feature extraction module is removed for speed-up. The textline sorting strategy conforming to the human reading order and UDML knowledge distillation strategy are utilized for higher model accuracy.
+
+In the non end-to-end KIE method, KIE needs at least **2 steps**. Firstly, the OCR model is used to extract the text and its position. Secondly, the KIE model is used to extract the key information according to the image, text position and text content.
+
+### 2.1 Train OCR Models
+
+#### 2.1.1 Text Detection
+
+##### (1) Data
+
+Most of the models provided in PaddleOCR are general models. In the process of text detection, the detection of adjacent text lines is generally based on the distance of the position. As shown in the figure above, when using PP-OCRv3 general English detection model for text detection, it is easy to detect the two fields representing different properties as one. Therefore, it is suggested to finetune a detection model according to your scenario firstly during the KIE task.
+
+During data annotation, the different key information needs to be separated. Otherwise, it will increase the difficulty of subsequent KIE tasks.
+
+For downstream tasks, generally speaking, `200~300` training images can guarantee the basic training effect. If there is not too much prior knowledge, **`200~300`** images can be labeled firstly for subsequent text detection model training.
+
+##### (2) Model
+
+In terms of model selection, PP-OCRv3 detection model is recommended. For more information about the training methods of the detection model, please refer to: [Text detection tutorial](../../ppocr/model_train/detection.en.md) and [PP-OCRv3 detection model tutorial](../../ppocr/model_train/PPOCRv3_det_train.en.md).
+
+#### 2.1.2 Text recognition
+
+Compared with the natural scene, the text recognition in the document image is generally relatively easier (the background is not too complex), so **it is suggested to** try the PP-OCRv3 general text recognition model provided in PaddleOCR ([PP-OCRv3 model list](../../ppocr/model_list.en.md))
+
+##### (1) Data
+
+However, there are also some challenges in some document scenarios, such as rare words in ID card scenarios and special fonts in invoice and other scenarios. These problems will increase the difficulty of text recognition. At this time, if you want to ensure or further improve the model accuracy, it is recommended to load PP-OCRv3 model based on the text recognition dataset of specific document scenarios for finetuning.
+
+In the process of model finetuning, it is recommended to prepare at least `5000` vertical scene text recognition images to ensure the basic model fine-tuning effect. If you want to improve the accuracy and generalization ability of the model, you can synthesize more text recognition images similar to the scene, collect general real text recognition data from the public data set, and add them to the text recognition training process. In the training process, it is suggested that the ratio of real data, synthetic data and general data of each epoch should be around `1:1:1`, which can be controlled by setting the sampling ratio of different data sources. If there are 3 training text files, including 10k, 20k and 50k pieces of data respectively, the data can be set in the configuration file as follows:
+
+```yaml linenums="1" linenums="1"
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/train_list_10k.txt
+    - ./train_data/train_list_10k.txt
+    - ./train_data/train_list_50k.txt
+    ratio_list: [1.0, 0.5, 0.2]
+    ...
+```
+
+##### (2) Model
+
+In terms of model selection, PP-OCRv3 recognition model is recommended. For more information about the training methods of the recognition model, please refer to: [Text recognition tutorial](../../ppocr/model_train/recognition.en.md) and [PP-OCRv3 model list](../../ppocr/model_list.en.md).
+
+### 2.2 Train KIE Models
+
+There are two main methods to extract the key information from the recognized texts.
+
+(1) Directly use SER model to obtain the key information category. For example, in the ID card scenario, we mark "name" and "Geoff Sample" as "name_key" and "name_value", respectively. The **text field** corresponding to the category "name_value" finally identified is the key information we need.
+
+(2) Joint use SER and RE models. For this case, we firstly use SER model to obtain all questions (keys) and questions (values) for the image text, and then use RE model to match all keys and values to find the relationship, so as to complete the extraction of key information.
+
+#### 2.2.1 SER
+
+Take the ID card scenario as an example. The key information generally includes `name`, `DOB`, etc. We can directly mark the corresponding fields as specific categories, as shown in the following figure.
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185728456-dc396f47-0880-4279-9c7c-c99601bf16a7.png" width="500">
+</div>
+
+**Note:**
+
+* In the labeling process, text content without key information about KIE shall be labeled as`other`, which is equivalent to background information. For example, in the ID card scenario, if we do not pay attention to `DOB` information, we can mark the categories of `DOB` and `Area manager` as `other`.
+* In the annotation process of, it is required to annotate the **textline** position rather than the character.
+
+In terms of data, generally speaking, for relatively fixed scenes, **50** training images can achieve acceptable effects. You can refer to [PPOCRLabel](https://github.com/PFCCLab/PPOCRLabel/blob/main/README.md) for finish the labeling process.
+
+In terms of model, it is recommended to use the VI-layoutXLM model proposed in PP-StructureV2. It is improved based on the LayoutXLM model, removing the visual feature extraction module, and further improving the model inference speed without the significant reduction on model accuracy. For more tutorials, please refer to [VI-LayoutXLM introduction](../../algorithm/kie/algorithm_kie_vi_layoutxlm.en.md) and [KIE tutorial](../../ppocr/model_train/kie.en.md).
+
+#### 2.2.2 SER + RE
+
+The SER model is mainly used to identify all keys and values in the document image, and the RE model is mainly used to match all keys and values.
+
+Taking the ID card scenario as an example, the key information generally includes key information such as `name`, `DOB`, etc. in the SER stage, we need to identify all questions (keys) and answers (values). The demo annotation is as follows. All keys can be annotated as `question`, and all values can be annotated as `answer`.
+
+![](./images/184526785-c3d2d310-cd57-4d31-b933-912716b29856.jpg)
+
+In the RE stage, the ID and connection information of each field need to be marked, as shown in the following figure.
+
+![](./images/184528728-626f77eb-fd9f-4709-a7dc-5411cc417dab.jpg)
+
+For each textline, you need to add 'ID' and 'linking' field information. The 'ID' records the unique identifier of the textline. Different text contents in the same images cannot be repeated. The 'linking' is a list that records the connection information between different texts. If the ID of the field "name" is 0 and the ID of the field "Geoff Sample" is 1, then they all have [[0, 1]] 'linking' marks, indicating that the fields with `id=0` and `id=1` form a key value relationship (the fields such as DOB and Expires are similar, and will not be repeated here).
+
+**Note:**
+
+During annotation, if value is multiple text lines, a key-value pair can be added in linking, such as `[[0, 1], [0, 2]]`.
+
+In terms of data, generally speaking, for relatively fixed scenes, about **50** training images can achieve acceptable effects.
+
+In terms of model, it is recommended to use the VI-layoutXLM model proposed in PP-StructureV2. It is improved based on the LayoutXLM model, removing the visual feature extraction module, and further improving the model inference speed without the significant reduction on model accuracy. For more tutorials, please refer to [VI-LayoutXLM introduction](../../algorithm/kie/algorithm_kie_vi_layoutxlm.en.md) and [KIE tutorial](../../ppocr/model_train/kie.en.md).
+
+## 3. Reference
+
+[1] Katti A R, Reisswig C, Guder C, et al. Chargrid: Towards understanding 2d documents[J]. arXiv preprint arXiv:1809.08799, 2018.
+
+[2] Xu Y, Li M, Cui L, et al. Layoutlm: Pre-training of text and layout for document image understanding[C]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2020: 1192-1200.
+
+[3] Xu Y, Xu Y, Lv T, et al. LayoutLMv2: Multi-modal pre-training for visually-rich document understanding[J]. arXiv preprint arXiv:2012.14740, 2020.
+
+[4]: Xu Y, Lv T, Cui L, et al. Layoutxlm: Multimodal pre-training for multilingual visually-rich document understanding[J]. arXiv preprint arXiv:2104.08836, 2021.
+
+[5] Li Y, Qian Y, Yu Y, et al. StrucTexT: Structured Text Understanding with Multi-Modal Transformers[C]//Proceedings of the 29th ACM International Conference on Multimedia. 2021: 1912-1920.
+
+[6] Liu X, Gao F, Zhang Q, et al. Graph convolution for multimodal information extraction from visually rich documents[J]. arXiv preprint arXiv:1903.11279, 2019.
+
+[7] Sun H, Kuang Z, Yue X, et al. Spatial Dual-Modality Graph Reasoning for Key Information Extraction[J]. arXiv preprint arXiv:2103.14470, 2021.
+
+[8] Zhang P, Xu Y, Cheng Z, et al. Trie: End-to-end text reading and information extraction for document understanding[C]//Proceedings of the 28th ACM International Conference on Multimedia. 2020: 1413-1422.
diff --git a/docs/version2.x/ppstructure/blog/how_to_do_kie.md b/docs/version2.x/ppstructure/blog/how_to_do_kie.md
new file mode 100644
index 0000000..51ff3a7
--- /dev/null
+++ b/docs/version2.x/ppstructure/blog/how_to_do_kie.md
@@ -0,0 +1,146 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# 怎样完成基于图像数据的信息抽取任务
+
+## 1. 简介
+
+### 1.1 背景
+
+关键信息抽取 (Key Information Extraction, KIE)指的是是从文本或者图像中，抽取出关键的信息。针对文档图像的关键信息抽取任务作为OCR的下游任务，存在非常多的实际应用场景，如表单识别、车票信息抽取、身份证信息抽取等。然而，使用人力从这些文档图像中提取或者收集关键信息耗时费力，怎样自动化融合图像中的视觉、布局、文字等特征并完成关键信息抽取是一个价值与挑战并存的问题。
+
+对于特定场景的文档图像，其中的关键信息位置、版式等较为固定，因此在研究早期有很多基于模板匹配的方法进行关键信息的抽取，考虑到其流程较为简单，该方法仍然被广泛应用在目前的很多场景中。但是这种基于模板匹配的方法在应用到不同的场景中时，需要耗费大量精力去调整与适配模板，迁移成本较高。
+
+文档图像中的KIE一般包含2个子任务，示意图如下图所示。
+
+* （1）SER: 语义实体识别 (Semantic Entity Recognition)，对每一个检测到的文本进行分类，如将其分为姓名，身份证。如下图中的黑色框和红色框。
+* （2）RE: 关系抽取 (Relation Extraction)，对每一个检测到的文本进行分类，如将其分为问题 (key) 和答案 (value) 。然后对每一个问题找到对应的答案，相当于完成key-value的匹配过程。如下图中的红色框和黑色框分别代表问题和答案，黄色线代表问题和答案之间的对应关系。
+
+![](./images/184588654-d87f54f3-13ab-42c4-afc0-da79bead3f14.png)
+
+### 1.2 基于深度学习的主流方法
+
+一般的KIE方法基于命名实体识别(Named Entity Recognition,NER)来展开研究，但是此类方法仅使用了文本信息而忽略了位置与视觉特征信息，因此精度受限。近几年大多学者开始融合多个模态的输入信息，进行特征融合，并对多模态信息进行处理，从而提升KIE的精度。主要方法有以下几种
+
+* （1）基于Grid的方法：此类方法主要关注图像层面多模态信息的融合，文本大多大多为字符粒度，对文本与结构结构信息的嵌入方式较为简单，如Chargrid[1]等算法。
+* （2）基于Token的方法：此类方法参考NLP中的BERT等方法，将位置、视觉等特征信息共同编码到多模态模型中，并且在大规模数据集上进行预训练，从而在下游任务中，仅需要少量的标注数据便可以获得很好的效果。如LayoutLM[2], LayoutLMv2[3], LayoutXLM[4], StrucText[5]等算法。
+* （3）基于GCN的方法：此类方法尝试学习图像、文字之间的结构信息，从而可以解决开集信息抽取的问题（训练集中没有见过的模板），如GCN[6]、SDMGR[7]等算法。
+* （4）基于End-to-end的方法：此类方法将现有的OCR文字识别以及KIE信息抽取2个任务放在一个统一的网络中进行共同学习，并在学习过程中相互加强。如Trie[8]等算法。
+
+更多关于该系列算法的详细介绍，请参考“动手学OCR·十讲”课程的课节六部分：[文档分析理论与实践](https://aistudio.baidu.com/aistudio/education/group/info/25207)。
+
+## 2. 关键信息抽取任务流程
+
+PaddleOCR中实现了LayoutXLM等算法（基于Token），同时，在PP-StructureV2中，对LayoutXLM多模态预训练模型的网络结构进行简化，去除了其中的Visual backbone部分，设计了视觉无关的VI-LayoutXLM模型，同时引入符合人类阅读顺序的排序逻辑以及UDML知识蒸馏策略，最终同时提升了关键信息抽取模型的精度与推理速度。
+
+下面介绍怎样基于PaddleOCR完成关键信息抽取任务。
+
+在非End-to-end的KIE方法中，完成关键信息抽取，至少需要**2个步骤**：首先使用OCR模型，完成文字位置与内容的提取，然后使用KIE模型，根据图像、文字位置以及文字内容，提取出其中的关键信息。
+
+### 2.1 训练OCR模型
+
+#### 2.1.1 文本检测
+
+##### （1）数据
+
+PaddleOCR中提供的模型大多数为通用模型，在进行文本检测的过程中，相邻文本行的检测一般是根据位置的远近进行区分，如上图，使用PP-OCRv3通用中英文检测模型进行文本检测时，容易将”民族“与“汉”这2个代表不同的字段检测到一起，从而增加后续KIE任务的难度。因此建议在做KIE任务的过程中，首先训练一个针对该文档数据集的检测模型。
+
+在数据标注时，关键信息的标注需要隔开，比上图中的 “民族汉” 3个字相隔较近，此时需要将”民族“与”汉“标注为2个文本检测框，否则会增加后续KIE任务的难度。
+
+对于下游任务，一般来说，`200~300`张的文本训练数据即可保证基本的训练效果，如果没有太多的先验知识，可以先标注 **`200~300`** 张图片，进行后续文本检测模型的训练。
+
+##### （2）模型
+
+在模型选择方面，推荐使用PP-OCRv3_det，关于更多关于检测模型的训练方法介绍，请参考：[OCR文本检测模型训练教程](../../ppocr/model_train/detection.md)。
+
+#### 2.1.2 文本识别
+
+相对自然场景，文档图像中的文本内容识别难度一般相对较低（背景相对不太复杂），因此**优先建议**尝试PaddleOCR中提供的PP-OCRv3通用文本识别模型([PP-OCRv3模型库链接](../../ppocr/model_list.md))。
+
+##### （1）数据
+
+然而，在部分文档场景中也会存在一些挑战，如身份证场景中存在着罕见字，在发票等场景中的字体比较特殊，这些问题都会增加文本识别的难度，此时如果希望保证或者进一步提升模型的精度，建议基于特定文档场景的文本识别数据集，加载PP-OCRv3模型进行微调。
+
+在模型微调的过程中，建议准备至少`5000`张垂类场景的文本识别图像，可以保证基本的模型微调效果。如果希望提升模型的精度与泛化能力，可以合成更多与该场景类似的文本识别数据，从公开数据集中收集通用真实文本识别数据，一并添加到该场景的文本识别训练任务过程中。在训练过程中，建议每个epoch的真实垂类数据、合成数据、通用数据比例在`1:1:1`左右，这可以通过设置不同数据源的采样比例进行控制。如有3个训练文本文件，分别包含1W、2W、5W条数据，那么可以在配置文件中设置数据如下：
+
+```yaml linenums="1" linenums="1"
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/train_list_1W.txt
+    - ./train_data/train_list_2W.txt
+    - ./train_data/train_list_5W.txt
+    ratio_list: [1.0, 0.5, 0.2]
+    ...
+```
+
+##### （2）模型
+
+在模型选择方面，推荐使用通用中英文文本识别模型PP-OCRv3_rec，关于更多关于文本识别模型的训练方法介绍，请参考：[OCR文本识别模型训练教程](../../ppocr/model_train/recognition.md)。
+
+### 2.2 训练KIE模型
+
+对于识别得到的文字进行关键信息抽取，有2种主要的方法。
+
+（1）直接使用SER，获取关键信息的类别：如身份证场景中，将“姓名“与”张三“分别标记为`name_key`与`name_value`。最终识别得到的类别为`name_value`对应的**文本字段**即为我们所需要的关键信息。
+
+（2）联合SER与RE进行使用：这种方法中，首先使用SER，获取图像文字内容中所有的key与value，然后使用RE方法，对所有的key与value进行配对，找到映射关系，从而完成关键信息的抽取。
+
+#### 2.2.1 SER
+
+以身份证场景为例， 关键信息一般包含`姓名`、`性别`、`民族`等，我们直接将对应的字段标注为特定的类别即可，如下图所示。
+
+![](./images/184526682-8b810397-5a93-4395-93da-37b8b8494c41.png)
+
+**注意：**
+
+* 标注过程中，对于无关于KIE关键信息的文本内容，均需要将其标注为`other`类别，相当于背景信息。如在身份证场景中，如果我们不关注性别信息，那么可以将“性别”与“男”这2个字段的类别均标注为`other`。
+* 标注过程中，需要以**文本行**为单位进行标注，无需标注单个字符的位置信息。
+
+数据量方面，一般来说，对于比较固定的场景，**50张**左右的训练图片即可达到可以接受的效果，可以使用[PPOCRLabel](https://github.com/PFCCLab/PPOCRLabel/blob/main/README_ch.md)完成KIE的标注过程。
+
+模型方面，推荐使用PP-StructureV2中提出的VI-LayoutXLM模型，它基于LayoutXLM模型进行改进，去除其中的视觉特征提取模块，在精度基本无损的情况下，进一步提升了模型推理速度。更多教程请参考：[VI-LayoutXLM算法介绍](../../algorithm/kie/algorithm_kie_layoutxlm.md)与[KIE关键信息抽取使用教程](../model_train/train_kie.md)。
+
+#### 2.2.2 SER + RE
+
+该过程主要包含SER与RE 2个过程。SER阶段主要用于识别出文档图像中的所有key与value，RE阶段主要用于对所有的key与value进行匹配。
+
+以身份证场景为例， 关键信息一般包含`姓名`、`性别`、`民族`等关键信息，在SER阶段，我们需要识别所有的question (key) 与answer (value) 。标注如下所示。每个字段的类别信息（`label`字段）可以是question、answer或者other（与待抽取的关键信息无关的字段）
+
+![](./images/184526785-c3d2d310-cd57-4d31-b933-912716b29856.jpg)
+
+在RE阶段，需要标注每个字段的的id与连接信息，如下图所示。
+
+![](./images/184528728-626f77eb-fd9f-4709-a7dc-5411cc417dab.jpg)
+
+每个文本行字段中，需要添加`id`与`linking`字段信息，`id`记录该文本行的唯一标识，同一张图片中的不同文本内容不能重复，`linking`是一个列表，记录了不同文本之间的连接信息。如字段“出生”的id为0，字段“1996年1月11日”的id为1，那么它们均有[[0, 1]]的`linking`标注，表示该id=0与id=1的字段构成key-value的关系（姓名、性别等字段类似，此处不再一一赘述）。
+
+**注意：**
+
+* 标注过程中，如果value是多个字符，那么linking中可以新增一个key-value对，如`[[0, 1], [0, 2]]`
+
+数据量方面，一般来说，对于比较固定的场景，**50张**左右的训练图片即可达到可以接受的效果，可以使用PPOCRLabel完成KIE的标注过程。
+
+模型方面，推荐使用PP-StructureV2中提出的VI-LayoutXLM模型，它基于LayoutXLM模型进行改进，去除其中的视觉特征提取模块，在精度基本无损的情况下，进一步提升了模型推理速度。更多教程请参考：[VI-LayoutXLM算法介绍](../../algorithm/kie/algorithm_kie_layoutxlm.md)与[KIE关键信息抽取使用教程](../model_train/train_kie.md)。
+
+## 3. 参考文献
+
+[1] Katti A R, Reisswig C, Guder C, et al. Chargrid: Towards understanding 2d documents[J]. arXiv preprint arXiv:1809.08799, 2018.
+
+[2] Xu Y, Li M, Cui L, et al. Layoutlm: Pre-training of text and layout for document image understanding[C]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2020: 1192-1200.
+
+[3] Xu Y, Xu Y, Lv T, et al. LayoutLMv2: Multi-modal pre-training for visually-rich document understanding[J]. arXiv preprint arXiv:2012.14740, 2020.
+
+[4]: Xu Y, Lv T, Cui L, et al. Layoutxlm: Multimodal pre-training for multilingual visually-rich document understanding[J]. arXiv preprint arXiv:2104.08836, 2021.
+
+[5] Li Y, Qian Y, Yu Y, et al. StrucTexT: Structured Text Understanding with Multi-Modal Transformers[C]//Proceedings of the 29th ACM International Conference on Multimedia. 2021: 1912-1920.
+
+[6] Liu X, Gao F, Zhang Q, et al. Graph convolution for multimodal information extraction from visually rich documents[J]. arXiv preprint arXiv:1903.11279, 2019.
+
+[7] Sun H, Kuang Z, Yue X, et al. Spatial Dual-Modality Graph Reasoning for Key Information Extraction[J]. arXiv preprint arXiv:2103.14470, 2021.
+
+[8] Zhang P, Xu Y, Cheng Z, et al. Trie: End-to-end text reading and information extraction for document understanding[C]//Proceedings of the 28th ACM International Conference on Multimedia. 2020: 1413-1422.
diff --git a/docs/version2.x/ppstructure/blog/images/184526682-8b810397-5a93-4395-93da-37b8b8494c41.png b/docs/version2.x/ppstructure/blog/images/184526682-8b810397-5a93-4395-93da-37b8b8494c41.png
new file mode 100644
index 0000000..dfea9ea
Binary files /dev/null and b/docs/version2.x/ppstructure/blog/images/184526682-8b810397-5a93-4395-93da-37b8b8494c41.png differ
diff --git a/docs/version2.x/ppstructure/blog/images/184526785-c3d2d310-cd57-4d31-b933-912716b29856.jpg b/docs/version2.x/ppstructure/blog/images/184526785-c3d2d310-cd57-4d31-b933-912716b29856.jpg
new file mode 100644
index 0000000..c2ef8f5
Binary files /dev/null and b/docs/version2.x/ppstructure/blog/images/184526785-c3d2d310-cd57-4d31-b933-912716b29856.jpg differ
diff --git a/docs/version2.x/ppstructure/blog/images/184528728-626f77eb-fd9f-4709-a7dc-5411cc417dab.jpg b/docs/version2.x/ppstructure/blog/images/184528728-626f77eb-fd9f-4709-a7dc-5411cc417dab.jpg
new file mode 100644
index 0000000..dcd4e52
Binary files /dev/null and b/docs/version2.x/ppstructure/blog/images/184528728-626f77eb-fd9f-4709-a7dc-5411cc417dab.jpg differ
diff --git a/docs/version2.x/ppstructure/blog/images/184588654-d87f54f3-13ab-42c4-afc0-da79bead3f14.png b/docs/version2.x/ppstructure/blog/images/184588654-d87f54f3-13ab-42c4-afc0-da79bead3f14.png
new file mode 100644
index 0000000..845aeb3
Binary files /dev/null and b/docs/version2.x/ppstructure/blog/images/184588654-d87f54f3-13ab-42c4-afc0-da79bead3f14.png differ
diff --git a/docs/version2.x/ppstructure/blog/images/3c200538-f2e6-4d79-847a-4c4587efa9f0.jpeg b/docs/version2.x/ppstructure/blog/images/3c200538-f2e6-4d79-847a-4c4587efa9f0.jpeg
new file mode 100644
index 0000000..e19f991
Binary files /dev/null and b/docs/version2.x/ppstructure/blog/images/3c200538-f2e6-4d79-847a-4c4587efa9f0.jpeg differ
diff --git a/docs/version2.x/ppstructure/blog/images/799450d4-d2c5-4b61-b490-e160dc0f515c.jpeg b/docs/version2.x/ppstructure/blog/images/799450d4-d2c5-4b61-b490-e160dc0f515c.jpeg
new file mode 100644
index 0000000..cf51f42
Binary files /dev/null and b/docs/version2.x/ppstructure/blog/images/799450d4-d2c5-4b61-b490-e160dc0f515c.jpeg differ
diff --git a/docs/version2.x/ppstructure/blog/images/d0858341-a889-483c-8373-5ecaa57f3b20.png b/docs/version2.x/ppstructure/blog/images/d0858341-a889-483c-8373-5ecaa57f3b20.png
new file mode 100644
index 0000000..a1ea4f8
Binary files /dev/null and b/docs/version2.x/ppstructure/blog/images/d0858341-a889-483c-8373-5ecaa57f3b20.png differ
diff --git a/docs/version2.x/ppstructure/blog/return_word_pos.en.md b/docs/version2.x/ppstructure/blog/return_word_pos.en.md
new file mode 100644
index 0000000..bdff053
--- /dev/null
+++ b/docs/version2.x/ppstructure/blog/return_word_pos.en.md
@@ -0,0 +1,100 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# Return recognition position
+
+According to the horizontal document, the recognition model not only returns the recognized content, but also the position of each word.
+
+## English document recovery
+
+### Download the inference model first
+
+```bash linenums="1"
+cd PaddleOCR/ppstructure
+
+## download model
+mkdir inference && cd inference
+## Download the detection model of the ultra-lightweight English PP-OCRv3 model and unzip it
+https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar && tar xf en_PP-OCRv3_det_infer.tar
+## Download the recognition model of the ultra-lightweight English PP-OCRv3 model and unzip it
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv3_mobile_rec_infer.tar && tar xf en_PP-OCRv3_mobile_rec_infer.tar
+## Download the ultra-lightweight English table inch model and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/en_ppstructure_mobile_v2.0_SLANet_infer.tar
+tar xf en_ppstructure_mobile_v2.0_SLANet_infer.tar
+## Download the layout model of publaynet dataset and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar
+tar xf picodet_lcnet_x1_0_fgd_layout_infer.tar
+cd ..
+```
+
+### Then use the following command inference in the /ppstructure/ directory
+
+```bash linenums="1"
+python predict_system.py \
+--image_dir=../docs/ppstructure/images/table_1.png \
+--det_model_dir=inference/en_PP-OCRv3_det_infer \
+--rec_model_dir=inference/en_PP-OCRv3_mobile_rec_infer \
+--rec_char_dict_path=../ppocr/utils/en_dict.txt \
+--table_model_dir=inference/en_ppstructure_mobile_v2.0_SLANet_infer \
+--table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt \
+--layout_model_dir=inference/picodet_lcnet_x1_0_fgd_layout_infer \
+--layout_dict_path=../ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt \
+--vis_font_path=../doc/fonts/simfang.ttf \
+--recovery=True \
+--output=../output/ \
+--return_word_box=True
+```
+
+### View the visualization of the inference results under `../output/structure/table_1/show_0.jpg`, as shown below
+
+![show_0_mdf_v2](./images/799450d4-d2c5-4b61-b490-e160dc0f515c.jpeg)
+
+## Recover Chinese documents
+
+### Download the inference model first
+
+```bash linenums="1"
+cd PaddleOCR/ppstructure
+
+## download model
+cd inference
+## Download the detection model of the ultra-lightweight Chinese PP-OCRv3 model and unzip it
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar && tar xf PP-OCRv3_mobile_det_infer.tar
+## Download the recognition model of the ultra-lightweight Chinese PP-OCRv3 model and unzip it
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar && tar xf PP-OCRv3_mobile_rec_infer.tar
+## Download the ultra-lightweight Chinese table inch model and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/ch_ppstructure_mobile_v2.0_SLANet_infer.tar
+tar xf ch_ppstructure_mobile_v2.0_SLANet_infer.tar
+## Download the layout model of CDLA dataset and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_cdla_infer.tar
+tar xf picodet_lcnet_x1_0_fgd_layout_cdla_infer.tar
+cd ..
+```
+
+### Upload the following test image "2.png" to the directory ./docs/table/
+
+![2](./images/d0858341-a889-483c-8373-5ecaa57f3b20.png)
+
+### Then use the following command inference in the /ppstructure/ directory
+
+```bash linenums="1"
+python predict_system.py \
+--image_dir=./docs/table/2.png \
+--det_model_dir=inference/PP-OCRv3_mobile_det_infer \
+--rec_model_dir=inference/PP-OCRv3_mobile_rec_infer \
+--rec_char_dict_path=../ppocr/utils/ppocr_keys_v1.txt \
+--table_model_dir=inference/ch_ppstructure_mobile_v2.0_SLANet_infer \
+--table_char_dict_path=../ppocr/utils/dict/table_structure_dict_ch.txt \
+--layout_model_dir=inference/picodet_lcnet_x1_0_fgd_layout_cdla_infer \
+--layout_dict_path=../ppocr/utils/dict/layout_dict/layout_cdla_dict.txt \
+--vis_font_path=../doc/fonts/chinese_cht.ttf \
+--recovery=True \
+--output=../output/ \
+--return_word_box=True
+```
+
+### View the visualization of the inference results under `../output/structure/2/show_0.jpg`, as shown below
+
+![show_1_mdf_v2](./images/3c200538-f2e6-4d79-847a-4c4587efa9f0.jpeg)
diff --git a/docs/version2.x/ppstructure/blog/return_word_pos.md b/docs/version2.x/ppstructure/blog/return_word_pos.md
new file mode 100644
index 0000000..f4ac367
--- /dev/null
+++ b/docs/version2.x/ppstructure/blog/return_word_pos.md
@@ -0,0 +1,100 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# 返回识别位置
+
+根据横排的文档，识别模型不仅返回识别的内容，还返回每个文字的位置。
+
+## 英文文档恢复
+
+### 先下载推理模型
+
+```bash linenums="1"
+cd PaddleOCR/ppstructure
+
+## download model
+mkdir inference && cd inference
+## Download the detection model of the ultra-lightweight English PP-OCRv3 model and unzip it
+https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar && tar xf en_PP-OCRv3_det_infer.tar
+## Download the recognition model of the ultra-lightweight English PP-OCRv3 model and unzip it
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv3_mobile_rec_infer.tar && tar xf en_PP-OCRv3_mobile_rec_infer.tar
+## Download the ultra-lightweight English table inch model and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/en_ppstructure_mobile_v2.0_SLANet_infer.tar
+tar xf en_ppstructure_mobile_v2.0_SLANet_infer.tar
+## Download the layout model of publaynet dataset and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar
+tar xf picodet_lcnet_x1_0_fgd_layout_infer.tar
+cd ..
+```
+
+### 然后在/ppstructure/目录下使用下面的指令推理
+
+```bash linenums="1"
+python predict_system.py \
+    --image_dir=./docs/ppstructure/images/table_1.png \
+    --det_model_dir=inference/en_PP-OCRv3_det_infer \
+    --rec_model_dir=inference/en_PP-OCRv3_mobile_rec_infer \
+    --rec_char_dict_path=../ppocr/utils/en_dict.txt \
+    --table_model_dir=inference/en_ppstructure_mobile_v2.0_SLANet_infer \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt \
+    --layout_model_dir=inference/picodet_lcnet_x1_0_fgd_layout_infer \
+    --layout_dict_path=../ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt \
+    --vis_font_path=../doc/fonts/simfang.ttf \
+    --recovery=True \
+    --output=../output/ \
+    --return_word_box=True
+```
+
+### 在`../output/structure/table_1/show_0.jpg`下查看推理结果的可视化，如下图所示
+
+![show_0_mdf_v2](./images/799450d4-d2c5-4b61-b490-e160dc0f515c.jpeg)
+
+## 针对中文文档恢复
+
+### 先下载推理模型
+
+```bash linenums="1"
+cd PaddleOCR/ppstructure
+
+## download model
+cd inference
+## Download the detection model of the ultra-lightweight Chinese PP-OCRv3 model and unzip it
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar && tar xf PP-OCRv3_mobile_det_infer.tar
+## Download the recognition model of the ultra-lightweight Chinese PP-OCRv3 model and unzip it
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar && tar xf PP-OCRv3_mobile_rec_infer.tar
+## Download the ultra-lightweight Chinese table inch model and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/ch_ppstructure_mobile_v2.0_SLANet_infer.tar
+tar xf ch_ppstructure_mobile_v2.0_SLANet_infer.tar
+## Download the layout model of CDLA dataset and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_cdla_infer.tar
+tar xf picodet_lcnet_x1_0_fgd_layout_cdla_infer.tar
+cd ..
+```
+
+### 上传下面的测试图片 "2.png" 至目录 ./docs/table/ 中
+
+![2](./images/d0858341-a889-483c-8373-5ecaa57f3b20.png)
+
+### 然后在/ppstructure/目录下使用下面的指令推理
+
+```bash linenums="1"
+python predict_system.py \
+    --image_dir=./docs/table/2.png \
+    --det_model_dir=inference/PP-OCRv3_mobile_det_infer \
+    --rec_model_dir=inference/PP-OCRv3_mobile_rec_infer \
+    --rec_char_dict_path=../ppocr/utils/ppocr_keys_v1.txt \
+    --table_model_dir=inference/ch_ppstructure_mobile_v2.0_SLANet_infer \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict_ch.txt \
+    --layout_model_dir=inference/picodet_lcnet_x1_0_fgd_layout_cdla_infer \
+    --layout_dict_path=../ppocr/utils/dict/layout_dict/layout_cdla_dict.txt \
+    --vis_font_path=../doc/fonts/chinese_cht.ttf \
+    --recovery=True \
+    --output=../output/ \
+    --return_word_box=True
+```
+
+### 在`../output/structure/2/show_0.jpg`下查看推理结果的可视化，如下图所示
+
+![show_1_mdf_v2](./images/3c200538-f2e6-4d79-847a-4c4587efa9f0.jpeg)
diff --git a/docs/version2.x/ppstructure/images/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808-20240705094001639.jpg b/docs/version2.x/ppstructure/images/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808-20240705094001639.jpg
new file mode 100644
index 0000000..6a5fd84
Binary files /dev/null and b/docs/version2.x/ppstructure/images/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808-20240705094001639.jpg differ
diff --git a/docs/version2.x/ppstructure/images/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb-20240705094037073.jpg b/docs/version2.x/ppstructure/images/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb-20240705094037073.jpg
new file mode 100644
index 0000000..d406a52
Binary files /dev/null and b/docs/version2.x/ppstructure/images/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb-20240705094037073.jpg differ
diff --git a/docs/version2.x/ppstructure/images/185539141-68e71c75-5cf7-4529-b2ca-219d29fa5f68-20240705093932704.jpg b/docs/version2.x/ppstructure/images/185539141-68e71c75-5cf7-4529-b2ca-219d29fa5f68-20240705093932704.jpg
new file mode 100644
index 0000000..42930f0
Binary files /dev/null and b/docs/version2.x/ppstructure/images/185539141-68e71c75-5cf7-4529-b2ca-219d29fa5f68-20240705093932704.jpg differ
diff --git a/docs/version2.x/ppstructure/images/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a-20240705094013236.png b/docs/version2.x/ppstructure/images/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a-20240705094013236.png
new file mode 100644
index 0000000..cd5bc50
Binary files /dev/null and b/docs/version2.x/ppstructure/images/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a-20240705094013236.png differ
diff --git a/docs/version2.x/ppstructure/images/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f-20240705094043151.jpg b/docs/version2.x/ppstructure/images/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f-20240705094043151.jpg
new file mode 100644
index 0000000..bdef2ed
Binary files /dev/null and b/docs/version2.x/ppstructure/images/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f-20240705094043151.jpg differ
diff --git a/docs/version2.x/ppstructure/images/186094813-3a8e16cc-42e5-4982-b9f4-0134dfb5688d.png b/docs/version2.x/ppstructure/images/186094813-3a8e16cc-42e5-4982-b9f4-0134dfb5688d.png
new file mode 100644
index 0000000..44845fb
Binary files /dev/null and b/docs/version2.x/ppstructure/images/186094813-3a8e16cc-42e5-4982-b9f4-0134dfb5688d.png differ
diff --git a/docs/version2.x/ppstructure/images/186095641-5843b4da-34d7-4c1c-943a-b1036a859fe3.png b/docs/version2.x/ppstructure/images/186095641-5843b4da-34d7-4c1c-943a-b1036a859fe3.png
new file mode 100644
index 0000000..e47dbbb
Binary files /dev/null and b/docs/version2.x/ppstructure/images/186095641-5843b4da-34d7-4c1c-943a-b1036a859fe3.png differ
diff --git a/docs/version2.x/ppstructure/images/186095702-9acef674-12af-4d09-97fc-abf4ab32600e.png b/docs/version2.x/ppstructure/images/186095702-9acef674-12af-4d09-97fc-abf4ab32600e.png
new file mode 100644
index 0000000..014ec50
Binary files /dev/null and b/docs/version2.x/ppstructure/images/186095702-9acef674-12af-4d09-97fc-abf4ab32600e.png differ
diff --git a/docs/version2.x/ppstructure/images/195265734-6f4b5a7f-59b1-4fcc-af6d-89afc9bd51e1-20240705140834325.jpg b/docs/version2.x/ppstructure/images/195265734-6f4b5a7f-59b1-4fcc-af6d-89afc9bd51e1-20240705140834325.jpg
new file mode 100644
index 0000000..3bb2ddd
Binary files /dev/null and b/docs/version2.x/ppstructure/images/195265734-6f4b5a7f-59b1-4fcc-af6d-89afc9bd51e1-20240705140834325.jpg differ
diff --git a/docs/version2.x/ppstructure/images/197464552-69de557f-edff-4c7f-acbf-069df1ba097f.png b/docs/version2.x/ppstructure/images/197464552-69de557f-edff-4c7f-acbf-069df1ba097f.png
new file mode 100644
index 0000000..7b4c271
Binary files /dev/null and b/docs/version2.x/ppstructure/images/197464552-69de557f-edff-4c7f-acbf-069df1ba097f.png differ
diff --git a/docs/version2.x/ppstructure/images/799450d4-d2c5-4b61-b490-e160dc0f515c.jpeg b/docs/version2.x/ppstructure/images/799450d4-d2c5-4b61-b490-e160dc0f515c.jpeg
new file mode 100644
index 0000000..0a51342
Binary files /dev/null and b/docs/version2.x/ppstructure/images/799450d4-d2c5-4b61-b490-e160dc0f515c.jpeg differ
diff --git a/docs/version2.x/ppstructure/images/ppstructure.gif b/docs/version2.x/ppstructure/images/ppstructure.gif
new file mode 100644
index 0000000..bff836e
Binary files /dev/null and b/docs/version2.x/ppstructure/images/ppstructure.gif differ
diff --git a/docs/version2.x/ppstructure/images/recovery.jpg b/docs/version2.x/ppstructure/images/recovery.jpg
new file mode 100644
index 0000000..a3817ab
Binary files /dev/null and b/docs/version2.x/ppstructure/images/recovery.jpg differ
diff --git a/docs/version2.x/ppstructure/images/table_1.png b/docs/version2.x/ppstructure/images/table_1.png
new file mode 100644
index 0000000..faff6e3
Binary files /dev/null and b/docs/version2.x/ppstructure/images/table_1.png differ
diff --git a/docs/version2.x/ppstructure/infer_deploy/cpp_infer.en.md b/docs/version2.x/ppstructure/infer_deploy/cpp_infer.en.md
new file mode 100644
index 0000000..539c102
--- /dev/null
+++ b/docs/version2.x/ppstructure/infer_deploy/cpp_infer.en.md
@@ -0,0 +1,433 @@
+---
+comments: true
+---
+
+# Server-side C++ Inference
+
+This chapter introduces the C++ deployment steps of the PaddleOCR model. C++ is better than Python in terms of performance. Therefore, in CPU and GPU deployment scenarios, C++ deployment is mostly used.
+This section will introduce how to configure the C++ environment and deploy PaddleOCR in Linux (CPU\GPU) environment. For Windows deployment please refer to [Windows](../../legacy/windows_vs2019_build.en.md) compilation guidelines.
+
+## 1. Prepare the Environment
+
+### 1.1 Environment
+
+- Linux, docker is recommended.
+- Windows.
+
+### 1.2 Compile OpenCV
+
+- First of all, you need to download the source code compiled package in the Linux environment from the OpenCV official website. Taking OpenCV 3.4.7 as an example, the download command is as follows.
+
+```bash linenums="1"
+cd deploy/cpp_infer
+wget https://paddleocr.bj.bcebos.com/libs/opencv/opencv-3.4.7.tar.gz
+tar -xf opencv-3.4.7.tar.gz
+```
+
+Finally, you will see the folder of `opencv-3.4.7/` in the current directory.
+
+- Compile OpenCV, the OpenCV source path (`root_path`) and installation path (`install_path`) should be set by yourself. Enter the OpenCV source code path and compile it in the following way.
+
+```bash linenums="1"
+root_path=your_opencv_root_path
+install_path=${root_path}/opencv3
+
+rm -rf build
+mkdir build
+cd build
+
+cmake .. \
+    -DCMAKE_INSTALL_PREFIX=${install_path} \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DBUILD_SHARED_LIBS=OFF \
+    -DWITH_IPP=OFF \
+    -DBUILD_IPP_IW=OFF \
+    -DWITH_LAPACK=OFF \
+    -DWITH_EIGEN=OFF \
+    -DCMAKE_INSTALL_LIBDIR=lib64 \
+    -DWITH_ZLIB=ON \
+    -DBUILD_ZLIB=ON \
+    -DWITH_JPEG=ON \
+    -DBUILD_JPEG=ON \
+    -DWITH_PNG=ON \
+    -DBUILD_PNG=ON \
+    -DWITH_TIFF=ON \
+    -DBUILD_TIFF=ON
+
+make -j
+make install
+```
+
+In the above commands, `root_path` is the downloaded OpenCV source code path, and `install_path` is the installation path of OpenCV. After `make install` is completed, the OpenCV header file and library file will be generated in this folder for later OCR source code compilation.
+
+The final file structure under the OpenCV installation path is as follows.
+
+```
+opencv3/
+|-- bin
+|-- include
+|-- lib
+|-- lib64
+|-- share
+```
+
+### 1.3 Compile or Download or the Paddle Inference Library
+
+- There are 2 ways to obtain the Paddle inference library, described in detail below.
+
+#### 1.3.1 Direct download and installation
+
+[Paddle inference library official website](https://www.paddlepaddle.org.cn/inference/master/guides/install/download_lib.html#linux). You can review and select the appropriate version of the inference library on the official website.
+
+- After downloading, use the following command to extract files.
+
+```bash linenums="1"
+tar -xf paddle_inference.tgz
+```
+
+Finally you will see the folder of `paddle_inference/` in the current path.
+
+#### 1.3.2 Compile the inference source code
+
+- If you want to get the latest Paddle inference library features, you can download the latest code from Paddle GitHub repository and compile the inference library from the source code. It is recommended to download the inference library with paddle version greater than or equal to 2.0.1.
+
+- You can refer to [Paddle inference library](https://www.paddlepaddle.org.cn/documentation/docs/en/advanced_guide/inference_deployment/inference/build_and_install_lib_en.html) to get the Paddle source code from GitHub, and then compile To generate the latest inference library. The method of using git to access the code is as follows.
+
+```bash linenums="1"
+git clone https://github.com/PaddlePaddle/Paddle.git
+git checkout develop
+```
+
+- Enter the Paddle directory and run the following commands to compile the paddle inference library.
+
+```bash linenums="1"
+rm -rf build
+mkdir build
+cd build
+
+cmake  .. \
+    -DWITH_CONTRIB=OFF \
+    -DWITH_MKL=ON \
+    -DWITH_MKLDNN=ON  \
+    -DWITH_TESTING=OFF \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DWITH_INFERENCE_API_TEST=OFF \
+    -DON_INFER=ON \
+    -DWITH_PYTHON=ON
+make -j
+make inference_lib_dist
+```
+
+For more compilation parameter options, please refer to the [document](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0/guides/05_inference_deployment/inference/build_and_install_lib_cn.html#congyuanmabianyi).
+
+- After the compilation process, you can see the following files in the folder of `build/paddle_inference_install_dir/`.
+
+```text linenums="1"
+build/paddle_inference_install_dir/
+|-- CMakeCache.txt
+|-- paddle
+|-- third_party
+|-- version.txt
+```
+
+`paddle` is the Paddle library required for C++ prediction later, and `version.txt` contains the version information of the current inference library.
+
+## 2. Compile and Run the Demo
+
+### 2.1 Export the inference model
+
+- You can refer to [Model inference](./python_infer.en.md) and export the inference model. After the model is exported, assuming it is placed in the `inference` directory, the directory structure is as follows.
+
+```text linenums="1"
+inference/
+|-- det_db
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+|-- rec_rcnn
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+|-- cls
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+|-- table
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+|-- layout
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+```
+
+### 2.2 Compile PaddleOCR C++ inference demo
+
+- The compilation commands are as follows. The addresses of Paddle C++ inference library, opencv and other Dependencies need to be replaced with the actual addresses on your own machines.
+
+```bash linenums="1"
+sh tools/build.sh
+```
+
+Specifically, you should modify the paths in `tools/build.sh`. The related content is as follows.
+
+```bash linenums="1"
+OPENCV_DIR=your_opencv_dir
+LIB_DIR=your_paddle_inference_dir
+CUDA_LIB_DIR=your_cuda_lib_dir
+CUDNN_LIB_DIR=your_cudnn_lib_dir
+```
+
+`OPENCV_DIR` is the OpenCV installation path; `LIB_DIR` is the download (`paddle_inference` folder)
+or the generated Paddle inference library path (`build/paddle_inference_install_dir` folder);
+`CUDA_LIB_DIR` is the CUDA library file path, in docker; it is `/usr/local/cuda/lib64`; `CUDNN_LIB_DIR` is the cuDNN library file path, in docker it is `/usr/lib/x86_64-linux-gnu/`.
+
+- After the compilation is completed, an executable file named `ppocr` will be generated in the `build` folder.
+
+### 2.3 Run the demo
+
+Execute the built executable file:
+
+```bash linenums="1"
+./build/ppocr [--param1] [--param2] [...]
+```
+
+**Note**:ppocr uses the `PP-OCRv3` model by default, and the input shape used by the recognition model is `3, 48, 320`, if you want to use the old version model, you should add the parameter `--rec_img_h=32`.
+
+Specifically,
+
+#### 1. det+cls+rec
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --rec_model_dir=inference/rec_rcnn \
+    --cls_model_dir=inference/cls \
+    --image_dir=../../doc/imgs/12.jpg \
+    --use_angle_cls=true \
+    --det=true \
+    --rec=true \
+    --cls=true \
+```
+
+##### 2. det+rec
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --rec_model_dir=inference/rec_rcnn \
+    --image_dir=../../doc/imgs/12.jpg \
+    --use_angle_cls=false \
+    --det=true \
+    --rec=true \
+    --cls=false \
+```
+
+##### 3. det
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --image_dir=../../doc/imgs/12.jpg \
+    --det=true \
+    --rec=false
+```
+
+##### 4. cls+rec
+
+```bash linenums="1"
+./build/ppocr --rec_model_dir=inference/rec_rcnn \
+    --cls_model_dir=inference/cls \
+    --image_dir=../../doc/imgs_words/ch/word_1.jpg \
+    --use_angle_cls=true \
+    --det=false \
+    --rec=true \
+    --cls=true \
+```
+
+##### 5. rec
+
+```bash linenums="1"
+./build/ppocr --rec_model_dir=inference/rec_rcnn \
+    --image_dir=../../doc/imgs_words/ch/word_1.jpg \
+    --use_angle_cls=false \
+    --det=false \
+    --rec=true \
+    --cls=false \
+```
+
+##### 6. cls
+
+```bash linenums="1"
+./build/ppocr --cls_model_dir=inference/cls \
+    --cls_model_dir=inference/cls \
+    --image_dir=../../doc/imgs_words/ch/word_1.jpg \
+    --use_angle_cls=true \
+    --det=false \
+    --rec=false \
+    --cls=true \
+```
+
+##### 7. layout+table
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --rec_model_dir=inference/rec_rcnn \
+    --table_model_dir=inference/table \
+    --image_dir=../../ppstructure/docs/table/table.jpg \
+    --layout_model_dir=inference/layout \
+    --type=structure \
+    --table=true \
+    --layout=true
+```
+
+##### 8. layout
+
+```bash linenums="1"
+./build/ppocr --layout_model_dir=inference/layout \
+    --image_dir=../../ppstructure/docs/table/1.png \
+    --type=structure \
+    --table=false \
+    --layout=true \
+    --det=false \
+    --rec=false
+```
+
+##### 9. table
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --rec_model_dir=inference/rec_rcnn \
+    --table_model_dir=inference/table \
+    --image_dir=../../ppstructure/docs/table/table.jpg \
+    --type=structure \
+    --table=true
+```
+
+More parameters are as follows,
+
+- Common parameters
+
+|parameter|data type|default|meaning|
+| --- | --- | --- | --- |
+|use_gpu|bool|false|Whether to use GPU|
+|gpu_id|int|0|GPU id when use_gpu is true|
+|gpu_mem|int|4000|GPU memory requested|
+|cpu_math_library_num_threads|int|10|Number of threads when using CPU inference. When machine cores is enough, the large the value, the faster the inference speed|
+|enable_mkldnn|bool|true|Whether to use mkdlnn library|
+|output|str|./output|Path where visualization results are saved|
+
+- forward
+
+|parameter|data type|default|meaning|
+| :---: | :---: | :---: | :---: |
+|det|bool|true|Whether to perform text detection in the forward direction|
+|rec|bool|true|Whether to perform text recognition in the forward direction|
+|cls|bool|false|Whether to perform text direction classification in the forward direction|
+
+- Detection related parameters
+
+|parameter|data type|default|meaning|
+| --- | --- | --- | --- |
+|det_model_dir|string|-|Address of detection inference model|
+|max_side_len|int|960|Limit the maximum image height and width to 960|
+|det_db_thresh|float|0.3|Used to filter the binarized image of DB prediction, setting 0.-0.3 has no obvious effect on the result|
+|det_db_box_thresh|float|0.5|DB post-processing filter box threshold, if there is a missing box detected, it can be reduced as appropriate|
+|det_db_unclip_ratio|float|1.6|Indicates the compactness of the text box, the smaller the value, the closer the text box to the text|
+|det_db_score_mode|string|slow| slow: use polygon box to calculate bbox score, fast: use rectangle box to calculate. Use rectangular box to calculate faster, and polygonal box more accurate for curved text area.|
+|visualize|bool|true|Whether to visualize the results，when it is set as true, the prediction results will be saved in the folder specified by the `output` field on an image with the same name as the input image.|
+
+- Classifier related parameters
+
+|parameter|data type|default|meaning|
+| --- | --- | --- | --- |
+|use_angle_cls|bool|false|Whether to use the direction classifier|
+|cls_model_dir|string|-|Address of direction classifier inference model|
+|cls_thresh|float|0.9|Score threshold of the  direction classifier|
+|cls_batch_num|int|1|batch size of classifier|
+
+- Recognition related parameters
+
+|parameter|data type|default|meaning|
+| --- | --- | --- | --- |
+|rec_model_dir|string|-|Address of recognition inference model|
+|rec_char_dict_path|string|../../ppocr/utils/ppocr_keys_v1.txt|dictionary file|
+|rec_batch_num|int|6|batch size of recognition|
+|rec_img_h|int|48|image height of recognition|
+|rec_img_w|int|320|image width of recognition|
+
+- Layout related parameters
+
+|parameter|data type|default|meaning|
+| :---: | :---: | :---: | :---: |
+|layout_model_dir|string|-| Address of layout inference model|
+|layout_dict_path|string|../../ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt|dictionary file|
+|layout_score_threshold|float|0.5|Threshold of score.|
+|layout_nms_threshold|float|0.5|Threshold of nms.|
+
+- Table recognition related parameters
+
+|parameter|data type|default|meaning|
+| :---: | :---: | :---: | :---: |
+|table_model_dir|string|-|Address of table recognition inference model|
+|table_char_dict_path|string|../../ppocr/utils/dict/table_structure_dict.txt|dictionary file|
+|table_max_len|int|488|The size of the long side of the input image of the table recognition model, the final input image size of the network is（table_max_len，table_max_len）|
+|merge_no_span_structure|bool|true|Whether to merge <td> and </td> to <td></td|
+
+- Multi-language inference is also supported in PaddleOCR, you can refer to [recognition tutorial](../../ppocr/blog/multi_languages.en.md) for more supported languages and models in PaddleOCR. Specifically, if you want to infer using multi-language models, you just need to modify values of `rec_char_dict_path` and `rec_model_dir`.
+
+The detection results will be shown on the screen, which is as follows.
+
+```bash linenums="1"
+predict img: ../../doc/imgs/12.jpg
+../../doc/imgs/12.jpg
+0       det boxes: [[74,553],[427,542],[428,571],[75,582]] rec text: 打浦路252935号 rec score: 0.947724
+1       det boxes: [[23,507],[513,488],[515,529],[24,548]] rec text: 绿洲仕格维花园公寓 rec score: 0.993728
+2       det boxes: [[187,456],[399,448],[400,480],[188,488]] rec text: 打浦路15号 rec score: 0.964994
+3       det boxes: [[42,413],[483,391],[484,428],[43,450]] rec text: 上海斯格威铂尔大酒店 rec score: 0.980086
+The detection visualized image saved in ./output//12.jpg
+```
+
+- layout+table
+
+```bash linenums="1"
+predict img: ../../ppstructure/docs/table/1.png
+0       type: text, region: [12,729,410,848], score: 0.781044, res: count of ocr result is : 7
+********** print ocr result **********
+0       det boxes: [[4,1],[79,1],[79,12],[4,12]] rec text: CTW1500. rec score: 0.769472
+...
+6       det boxes: [[4,99],[391,99],[391,112],[4,112]] rec text: sate-of-the-artmethods[12.34.36l.ourapproachachieves rec score: 0.90414
+********** end print ocr result **********
+1       type: text, region: [69,342,342,359], score: 0.703666, res: count of ocr result is : 1
+********** print ocr result **********
+0       det boxes: [[8,2],[269,2],[269,13],[8,13]] rec text: Table6.Experimentalresults on CTW-1500 rec score: 0.890454
+********** end print ocr result **********
+2       type: text, region: [70,316,706,332], score: 0.659738, res: count of ocr result is : 2
+********** print ocr result **********
+0       det boxes: [[373,2],[630,2],[630,11],[373,11]] rec text: oroposals.andthegreencontoursarefinal rec score: 0.919729
+1       det boxes: [[8,3],[357,3],[357,11],[8,11]] rec text: Visualexperimentalresultshebluecontoursareboundar rec score: 0.915963
+********** end print ocr result **********
+3       type: text, region: [489,342,789,359], score: 0.630538, res: count of ocr result is : 1
+********** print ocr result **********
+0       det boxes: [[8,2],[294,2],[294,14],[8,14]] rec text: Table7.Experimentalresults onMSRA-TD500 rec score: 0.942251
+********** end print ocr result **********
+4       type: text, region: [444,751,841,848], score: 0.607345, res: count of ocr result is : 5
+********** print ocr result **********
+0       det boxes: [[19,3],[389,3],[389,17],[19,17]] rec text: Inthispaper,weproposeanovel adaptivebound rec score: 0.941031
+1       det boxes: [[4,22],[390,22],[390,36],[4,36]] rec text: aryproposalnetworkforarbitraryshapetextdetection rec score: 0.960172
+2       det boxes: [[4,42],[392,42],[392,56],[4,56]] rec text: whichadoptanboundaryproposalmodeltogeneratecoarse rec score: 0.934647
+3       det boxes: [[4,61],[389,61],[389,75],[4,75]] rec text: ooundaryproposals,andthenadoptanadaptiveboundary rec score: 0.946296
+4       det boxes: [[5,80],[387,80],[387,93],[5,93]] rec text: leformationmodelcombinedwithGCNandRNNtoper rec score: 0.952401
+********** end print ocr result **********
+5       type: title, region: [444,705,564,724], score: 0.785429, res: count of ocr result is : 1
+********** print ocr result **********
+0       det boxes: [[6,2],[113,2],[113,14],[6,14]] rec text: 5.Conclusion rec score: 0.856903
+********** end print ocr result **********
+6       type: table, region: [14,360,402,711], score: 0.963643, res: <html><body><table><thead><tr><td>Methods</td><td>Ext</td><td>R</td><td>P</td><td>F</td><td>FPS</td></tr></thead><tbody><tr><td>TextSnake [18]</td><td>Syn</td><td>85.3</td><td>67.9</td><td>75.6</td><td></td></tr><tr><td>CSE [17]</td><td>MiLT</td><td>76.1</td><td>78.7</td><td>77.4</td><td>0.38</td></tr><tr><td>LOMO[40]</td><td>Syn</td><td>76.5</td><td>85.7</td><td>80.8</td><td>4.4</td></tr><tr><td>ATRR[35]</td><td>Sy-</td><td>80.2</td><td>80.1</td><td>80.1</td><td>-</td></tr><tr><td>SegLink++ [28]</td><td>Syn</td><td>79.8</td><td>82.8</td><td>81.3</td><td>-</td></tr><tr><td>TextField [37]</td><td>Syn</td><td>79.8</td><td>83.0</td><td>81.4</td><td>6.0</td></tr><tr><td>MSR[38]</td><td>Syn</td><td>79.0</td><td>84.1</td><td>81.5</td><td>4.3</td></tr><tr><td>PSENet-1s [33]</td><td>MLT</td><td>79.7</td><td>84.8</td><td>82.2</td><td>3.9</td></tr><tr><td>DB [12]</td><td>Syn</td><td>80.2</td><td>86.9</td><td>83.4</td><td>22.0</td></tr><tr><td>CRAFT [2]</td><td>Syn</td><td>81.1</td><td>86.0</td><td>83.5</td><td>-</td></tr><tr><td>TextDragon [5]</td><td>MLT+</td><td>82.8</td><td>84.5</td><td>83.6</td><td></td></tr><tr><td>PAN [34]</td><td>Syn</td><td>81.2</td><td>86.4</td><td>83.7</td><td>39.8</td></tr><tr><td>ContourNet [36]</td><td></td><td>84.1</td><td>83.7</td><td>83.9</td><td>4.5</td></tr><tr><td>DRRG [41]</td><td>MLT</td><td>83.02</td><td>85.93</td><td>84.45</td><td>-</td></tr><tr><td>TextPerception[23]</td><td>Syn</td><td>81.9</td><td>87.5</td><td>84.6</td><td></td></tr><tr><td>Ours</td><td> Syn</td><td>80.57</td><td>87.66</td><td>83.97</td><td>12.08</td></tr><tr><td>Ours</td><td></td><td>81.45</td><td>87.81</td><td>84.51</td><td>12.15</td></tr><tr><td>Ours</td><td>MLT</td><td>83.60</td><td>86.45</td><td>85.00</td><td>12.21</td></tr></tbody></table></body></html>
+The table visualized image saved in ./output//6_1.png
+7       type: table, region: [462,359,820,657], score: 0.953917, res: <html><body><table><thead><tr><td>Methods</td><td>R</td><td>P</td><td>F</td><td>FPS</td></tr></thead><tbody><tr><td>SegLink [26]</td><td>70.0</td><td>86.0</td><td>77.0</td><td>8.9</td></tr><tr><td>PixelLink [4]</td><td>73.2</td><td>83.0</td><td>77.8</td><td>-</td></tr><tr><td>TextSnake [18]</td><td>73.9</td><td>83.2</td><td>78.3</td><td>1.1</td></tr><tr><td>TextField [37]</td><td>75.9</td><td>87.4</td><td>81.3</td><td>5.2 </td></tr><tr><td>MSR[38]</td><td>76.7</td><td>87.4</td><td>81.7</td><td>-</td></tr><tr><td>FTSN[3]</td><td>77.1</td><td>87.6</td><td>82.0</td><td>:</td></tr><tr><td>LSE[30]</td><td>81.7</td><td>84.2</td><td>82.9</td><td></td></tr><tr><td>CRAFT [2]</td><td>78.2</td><td>88.2</td><td>82.9</td><td>8.6</td></tr><tr><td>MCN [16]</td><td>79</td><td>88</td><td>83</td><td>-</td></tr><tr><td>ATRR[35]</td><td>82.1</td><td>85.2</td><td>83.6</td><td>-</td></tr><tr><td>PAN [34]</td><td>83.8</td><td>84.4</td><td>84.1</td><td>30.2</td></tr><tr><td>DB[12]</td><td>79.2</td><td>91.5</td><td>84.9</td><td>32.0</td></tr><tr><td>DRRG [41]</td><td>82.30</td><td>88.05</td><td>85.08</td><td>-</td></tr><tr><td>Ours (SynText)</td><td>80.68</td><td>85.40</td><td>82.97</td><td>12.68</td></tr><tr><td>Ours (MLT-17)</td><td>84.54</td><td>86.62</td><td>85.57</td><td>12.31</td></tr></tbody></table></body></html>
+The table visualized image saved in ./output//7_1.png
+8       type: figure, region: [14,3,836,310], score: 0.969443, res: count of ocr result is : 26
+********** print ocr result **********
+0       det boxes: [[506,14],[539,15],[539,22],[506,21]] rec text: E rec score: 0.318073
+...
+25      det boxes: [[680,290],[759,288],[759,303],[680,305]] rec text: (d) CTW1500 rec score: 0.95911
+********** end print ocr result **********
+```
+
+## 3. FAQ
+
+ 1. Encountered the error `unable to access 'https://github.com/LDOUBLEV/AutoLog.git/': gnutls_handshake() failed: The TLS connection was non-properly terminated.`, change the github address in `deploy/cpp_infer/external-cmake/auto-log.cmake` to the <https://gitee.com/Double_V/AutoLog> address.
diff --git a/docs/version2.x/ppstructure/infer_deploy/cpp_infer.md b/docs/version2.x/ppstructure/infer_deploy/cpp_infer.md
new file mode 100644
index 0000000..276c337
--- /dev/null
+++ b/docs/version2.x/ppstructure/infer_deploy/cpp_infer.md
@@ -0,0 +1,443 @@
+---
+comments: true
+---
+
+# 服务器端C++预测
+
+本章节介绍PaddleOCR 模型的C++部署方法。C++在性能计算上优于Python，因此，在大多数CPU、GPU部署场景，多采用C++的部署方式，本节将介绍如何在Linux\Windows (CPU\GPU)环境下配置C++环境并完成PaddleOCR模型部署。
+
+## 1. 准备环境
+
+### 1.1 运行准备
+
+- Linux环境，推荐使用docker。
+- Windows环境。
+
+- 该文档主要介绍基于Linux环境的PaddleOCR C++预测流程，如果需要在Windows下基于预测库进行C++预测，具体编译方法请参考[Windows下编译教程](../../legacy/windows_vs2019_build.md)
+
+### 1.2 编译opencv库
+
+- 首先需要从opencv官网上下载在Linux环境下源码编译的包，以opencv3.4.7为例，下载命令如下：
+
+```bash linenums="1"
+cd deploy/cpp_infer
+wget https://paddleocr.bj.bcebos.com/libs/opencv/opencv-3.4.7.tar.gz
+tar -xf opencv-3.4.7.tar.gz
+```
+
+最终可以在当前目录下看到`opencv-3.4.7/`的文件夹。
+
+- 编译opencv，设置opencv源码路径(`root_path`)以及安装路径(`install_path`)。进入opencv源码路径下，按照下面的方式进行编译。
+
+```bash linenums="1"
+root_path="your_opencv_root_path"
+install_path=${root_path}/opencv3
+build_dir=${root_path}/build
+
+rm -rf ${build_dir}
+mkdir ${build_dir}
+cd ${build_dir}
+
+cmake .. \
+    -DCMAKE_INSTALL_PREFIX=${install_path} \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DBUILD_SHARED_LIBS=OFF \
+    -DWITH_IPP=OFF \
+    -DBUILD_IPP_IW=OFF \
+    -DWITH_LAPACK=OFF \
+    -DWITH_EIGEN=OFF \
+    -DCMAKE_INSTALL_LIBDIR=lib64 \
+    -DWITH_ZLIB=ON \
+    -DBUILD_ZLIB=ON \
+    -DWITH_JPEG=ON \
+    -DBUILD_JPEG=ON \
+    -DWITH_PNG=ON \
+    -DBUILD_PNG=ON \
+    -DWITH_TIFF=ON \
+    -DBUILD_TIFF=ON
+
+make -j
+make install
+```
+
+也可以直接修改`tools/build_opencv.sh`的内容，然后直接运行下面的命令进行编译。
+
+```bash linenums="1"
+sh tools/build_opencv.sh
+```
+
+其中`root_path`为下载的opencv源码路径，`install_path`为opencv的安装路径，`make install`完成之后，会在该文件夹下生成opencv头文件和库文件，用于后面的OCR代码编译。
+
+最终在安装路径下的文件结构如下所示。
+
+```text linenums="1"
+opencv3/
+|-- bin
+|-- include
+|-- lib
+|-- lib64
+|-- share
+```
+
+### 1.3 下载或者编译Paddle预测库
+
+可以选择直接下载安装或者从源码编译，下文分别进行具体说明。
+
+#### 1.3.1 直接下载安装
+
+[Paddle预测库官网](https://www.paddlepaddle.org.cn/inference/master/guides/install/download_lib.html#linux) 上提供了不同cuda版本的Linux预测库，可以在官网查看并选择合适的预测库版本（*建议选择paddle版本>=2.0.1版本的预测库* ）。
+
+下载之后解压:
+
+```bash linenums="1"
+tar -xf paddle_inference.tgz
+```
+
+最终会在当前的文件夹中生成`paddle_inference/`的子文件夹。
+
+#### 1.3.2 预测库源码编译
+
+如果希望获取最新预测库特性，可以从github上克隆最新Paddle代码进行编译，生成最新的预测库。
+
+- 使用git获取代码:
+
+```bash linenums="1"
+git clone https://github.com/PaddlePaddle/Paddle.git
+git checkout develop
+```
+
+- 进入Paddle目录，进行编译:
+
+```bash linenums="1"
+rm -rf build
+mkdir build
+cd build
+
+cmake  .. \
+    -DWITH_CONTRIB=OFF \
+    -DWITH_MKL=ON \
+    -DWITH_MKLDNN=ON  \
+    -DWITH_TESTING=OFF \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DWITH_INFERENCE_API_TEST=OFF \
+    -DON_INFER=ON \
+    -DWITH_PYTHON=ON
+make -j
+make inference_lib_dist
+```
+
+更多编译参数选项介绍可以参考[Paddle预测库编译文档](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0/guides/05_inference_deployment/inference/build_and_install_lib_cn.html#congyuanmabianyi)。
+
+- 编译完成之后，可以在`build/paddle_inference_install_dir/`文件下看到生成了以下文件及文件夹。
+
+```
+build/paddle_inference_install_dir/
+|-- CMakeCache.txt
+|-- paddle
+|-- third_party
+|-- version.txt
+```
+
+其中`paddle`就是C++预测所需的Paddle库，`version.txt`中包含当前预测库的版本信息。
+
+## 2. 开始运行
+
+### 2.1 准备模型
+
+直接下载PaddleOCR提供的推理模型，或者参考[模型预测章节](../../legacy/python_infer.md)，将训练好的模型导出为推理模型。模型导出之后，假设放在`inference`目录下，则目录结构如下：
+
+```text linenums="1"
+inference/
+|-- det_db
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+|-- rec_rcnn
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+|-- cls
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+|-- table
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+|-- layout
+|   |--inference.pdiparams
+|   |--inference.pdmodel
+```
+
+### 2.2 编译PaddleOCR C++预测demo
+
+编译命令如下，其中Paddle C++预测库、opencv等其他依赖库的地址需要换成自己机器上的实际地址。
+
+```bash linenums="1"
+sh tools/build.sh
+```
+
+具体的，需要修改`tools/build.sh`中环境路径，相关内容如下：
+
+```bash linenums="1"
+OPENCV_DIR=your_opencv_dir
+LIB_DIR=your_paddle_inference_dir
+CUDA_LIB_DIR=your_cuda_lib_dir
+CUDNN_LIB_DIR=/your_cudnn_lib_dir
+```
+
+其中，`OPENCV_DIR`为opencv编译安装的地址；`LIB_DIR`为下载(`paddle_inference`文件夹)或者编译生成的Paddle预测库地址(`build/paddle_inference_install_dir`文件夹)；`CUDA_LIB_DIR`为cuda库文件地址，在docker中为`/usr/local/cuda/lib64`；`CUDNN_LIB_DIR`为cudnn库文件地址，在docker中为`/usr/lib/x86_64-linux-gnu/`。**注意：以上路径都写绝对路径，不要写相对路径。**
+
+编译完成之后，会在`build`文件夹下生成一个名为`ppocr`的可执行文件。
+
+### 2.3 运行demo
+
+本demo支持系统串联调用，也支持单个功能的调用，如，只使用检测或识别功能。
+
+**注意** ppocr默认使用`PP-OCRv3`模型，识别模型使用的输入shape为`3,48,320`, 如需使用旧版本的PP-OCR模型，则需要设置参数`--rec_img_h=32`。
+
+运行方式：
+
+```bash linenums="1"
+./build/ppocr [--param1] [--param2] [...]
+```
+
+具体命令如下：
+
+#### 1. 检测+分类+识别
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --rec_model_dir=inference/rec_rcnn \
+    --cls_model_dir=inference/cls \
+    --image_dir=../../doc/imgs/12.jpg \
+    --use_angle_cls=true \
+    --det=true \
+    --rec=true \
+    --cls=true \
+```
+
+##### 2. 检测+识别
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --rec_model_dir=inference/rec_rcnn \
+    --image_dir=../../doc/imgs/12.jpg \
+    --use_angle_cls=false \
+    --det=true \
+    --rec=true \
+    --cls=false \
+```
+
+##### 3. 检测
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --image_dir=../../doc/imgs/12.jpg \
+    --det=true \
+    --rec=false
+```
+
+##### 4. 分类+识别
+
+```bash linenums="1"
+./build/ppocr --rec_model_dir=inference/rec_rcnn \
+    --cls_model_dir=inference/cls \
+    --image_dir=../../doc/imgs_words/ch/word_1.jpg \
+    --use_angle_cls=true \
+    --det=false \
+    --rec=true \
+    --cls=true \
+```
+
+##### 5. 识别
+
+```bash linenums="1"
+./build/ppocr --rec_model_dir=inference/rec_rcnn \
+    --image_dir=../../doc/imgs_words/ch/word_1.jpg \
+    --use_angle_cls=false \
+    --det=false \
+    --rec=true \
+    --cls=false \
+```
+
+##### 6. 分类
+
+```bash linenums="1"
+./build/ppocr --cls_model_dir=inference/cls \
+    --cls_model_dir=inference/cls \
+    --image_dir=../../doc/imgs_words/ch/word_1.jpg \
+    --use_angle_cls=true \
+    --det=false \
+    --rec=false \
+    --cls=true \
+```
+
+##### 7. 版面分析+表格识别
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --rec_model_dir=inference/rec_rcnn \
+    --table_model_dir=inference/table \
+    --image_dir=../../ppstructure/docs/table/table.jpg \
+    --layout_model_dir=inference/layout \
+    --type=structure \
+    --table=true \
+    --layout=true
+```
+
+##### 8. 版面分析
+
+```bash linenums="1"
+./build/ppocr --layout_model_dir=inference/layout \
+    --image_dir=../../ppstructure/docs/table/1.png \
+    --type=structure \
+    --table=false \
+    --layout=true \
+    --det=false \
+    --rec=false
+```
+
+##### 9. 表格识别
+
+```bash linenums="1"
+./build/ppocr --det_model_dir=inference/det_db \
+    --rec_model_dir=inference/rec_rcnn \
+    --table_model_dir=inference/table \
+    --image_dir=../../ppstructure/docs/table/table.jpg \
+    --type=structure \
+    --table=true
+```
+
+更多支持的可调节参数解释如下：
+
+- 通用参数
+
+|           参数名称           | 类型  | 默认参数 |                               意义                                |
+| :--------------------------: | :---: | :------: | :---------------------------------------------------------------: |
+|           use_gpu            | bool  |  false   |                            是否使用GPU                            |
+|            gpu_id            |  int  |    0     |                       GPU id，使用GPU时有效                       |
+|           gpu_mem            |  int  |   4000   |                           申请的GPU内存                           |
+| cpu_math_library_num_threads |  int  |    10    | CPU预测时的线程数，在机器核数充足的情况下，该值越大，预测速度越快 |
+|        enable_mkldnn         | bool  |   true   |                         是否使用mkldnn库                          |
+|            output            |  str  | ./output |                       可视化结果保存的路径                        |
+
+- 前向相关
+
+| 参数名称 | 类型  | 默认参数 |           意义           |
+| :------: | :---: | :------: | :----------------------: |
+|   det    | bool  |   true   |   前向是否执行文字检测   |
+|   rec    | bool  |   true   |   前向是否执行文字识别   |
+|   cls    | bool  |  false   | 前向是否执行文字方向分类 |
+
+- 检测模型相关
+
+|      参数名称       |  类型  | 默认参数 |                                                     意义                                                     |
+| :-----------------: | :----: | :------: | :----------------------------------------------------------------------------------------------------------: |
+|    det_model_dir    | string |    -     |                                         检测模型inference model地址                                          |
+|    max_side_len     |  int   |   960    |                          输入图像长宽大于960时，等比例缩放图像，使得图像最长边为960                          |
+|    det_db_thresh    | float  |   0.3    |                           用于过滤DB预测的二值化图像，设置为0.-0.3对结果影响不明显                           |
+|  det_db_box_thresh  | float  |   0.5    |                           DB后处理过滤box的阈值，如果检测存在漏框情况，可酌情减小                            |
+| det_db_unclip_ratio | float  |   1.6    |                                 表示文本框的紧致程度，越小则文本框更靠近文本                                 |
+|  det_db_score_mode  | string |   slow   | slow:使用多边形框计算bbox score，fast:使用矩形框计算。矩形框计算速度更快，多边形框对弯曲文本区域计算更准确。 |
+|      visualize      |  bool  |   true   |       是否对结果进行可视化，为1时，预测结果会保存在`output`字段指定的文件夹下和输入图像同名的图像上。        |
+
+- 方向分类器相关
+
+|   参数名称    |  类型  | 默认参数 |             意义              |
+| :-----------: | :----: | :------: | :---------------------------: |
+| use_angle_cls |  bool  |  false   |      是否使用方向分类器       |
+| cls_model_dir | string |    -     | 方向分类器inference model地址 |
+|  cls_thresh   | float  |   0.9    |     方向分类器的得分阈值      |
+| cls_batch_num |  int   |    1     |      方向分类器batchsize      |
+
+- 文字识别模型相关
+
+|      参数名称      |  类型  |              默认参数               |              意义               |
+| :----------------: | :----: | :---------------------------------: | :-----------------------------: |
+|   rec_model_dir    | string |                  -                  | 文字识别模型inference model地址 |
+| rec_char_dict_path | string | ../../ppocr/utils/ppocr_keys_v1.txt |            字典文件             |
+|   rec_batch_num    |  int   |                  6                  |      文字识别模型batchsize      |
+|     rec_img_h      |  int   |                 48                  |    文字识别模型输入图像高度     |
+|     rec_img_w      |  int   |                 320                 |    文字识别模型输入图像宽度     |
+
+- 版面分析模型相关
+
+|        参数名称        |  类型  |                           默认参数                           |              意义               |
+| :--------------------: | :----: | :----------------------------------------------------------: | :-----------------------------: |
+|    layout_model_dir    | string |                              -                               | 版面分析模型inference model地址 |
+|    layout_dict_path    | string | ../../ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt |            字典文件             |
+| layout_score_threshold | float  |                             0.5                              |        检测框的分数阈值         |
+|  layout_nms_threshold  | float  |                             0.5                              |            nms的阈值            |
+
+- 表格识别模型相关
+
+|        参数名称         |  类型  |                      默认参数                      |                                         意义                                         |
+| :---------------------: | :----: | :------------------------------------------------: | :----------------------------------------------------------------------------------: |
+|     table_model_dir     | string |                         -                          |                           表格识别模型inference model地址                            |
+|  table_char_dict_path   | string | ../../ppocr/utils/dict/table_structure_dict_ch.txt |                                       字典文件                                       |
+|      table_max_len      |  int   |                        488                         | 表格识别模型输入图像长边大小，最终网络输入图像大小为（table_max_len，table_max_len） |
+| merge_no_span_structure |  bool  |                        true                        |                          是否合并<td> 和 </td> 为<td></td>                           |
+
+- PaddleOCR也支持多语言的预测，更多支持的语言和模型可以参考[识别文档](../../ppocr/blog/multi_languages.md)中的多语言字典与模型部分，如果希望进行多语言预测，只需将修改`rec_char_dict_path`（字典文件路径）以及`rec_model_dir`（inference模型路径）字段即可。
+
+最终屏幕上会输出检测结果如下：
+
+- ocr
+
+```bash linenums="1"
+predict img: ../../doc/imgs/12.jpg
+../../doc/imgs/12.jpg
+0       det boxes: [[74,553],[427,542],[428,571],[75,582]] rec text: 打浦路252935号 rec score: 0.947724
+1       det boxes: [[23,507],[513,488],[515,529],[24,548]] rec text: 绿洲仕格维花园公寓 rec score: 0.993728
+2       det boxes: [[187,456],[399,448],[400,480],[188,488]] rec text: 打浦路15号 rec score: 0.964994
+3       det boxes: [[42,413],[483,391],[484,428],[43,450]] rec text: 上海斯格威铂尔大酒店 rec score: 0.980086
+The detection visualized image saved in ./output//12.jpg
+```
+
+- layout+table
+
+```bash linenums="1"
+predict img: ../../ppstructure/docs/table/1.png
+0       type: text, region: [12,729,410,848], score: 0.781044, res: count of ocr result is : 7
+********** print ocr result **********
+0       det boxes: [[4,1],[79,1],[79,12],[4,12]] rec text: CTW1500. rec score: 0.769472
+...
+6       det boxes: [[4,99],[391,99],[391,112],[4,112]] rec text: sate-of-the-artmethods[12.34.36l.ourapproachachieves rec score: 0.90414
+********** end print ocr result **********
+1       type: text, region: [69,342,342,359], score: 0.703666, res: count of ocr result is : 1
+********** print ocr result **********
+0       det boxes: [[8,2],[269,2],[269,13],[8,13]] rec text: Table6.Experimentalresults on CTW-1500 rec score: 0.890454
+********** end print ocr result **********
+2       type: text, region: [70,316,706,332], score: 0.659738, res: count of ocr result is : 2
+********** print ocr result **********
+0       det boxes: [[373,2],[630,2],[630,11],[373,11]] rec text: oroposals.andthegreencontoursarefinal rec score: 0.919729
+1       det boxes: [[8,3],[357,3],[357,11],[8,11]] rec text: Visualexperimentalresultshebluecontoursareboundar rec score: 0.915963
+********** end print ocr result **********
+3       type: text, region: [489,342,789,359], score: 0.630538, res: count of ocr result is : 1
+********** print ocr result **********
+0       det boxes: [[8,2],[294,2],[294,14],[8,14]] rec text: Table7.Experimentalresults onMSRA-TD500 rec score: 0.942251
+********** end print ocr result **********
+4       type: text, region: [444,751,841,848], score: 0.607345, res: count of ocr result is : 5
+********** print ocr result **********
+0       det boxes: [[19,3],[389,3],[389,17],[19,17]] rec text: Inthispaper,weproposeanovel adaptivebound rec score: 0.941031
+1       det boxes: [[4,22],[390,22],[390,36],[4,36]] rec text: aryproposalnetworkforarbitraryshapetextdetection rec score: 0.960172
+2       det boxes: [[4,42],[392,42],[392,56],[4,56]] rec text: whichadoptanboundaryproposalmodeltogeneratecoarse rec score: 0.934647
+3       det boxes: [[4,61],[389,61],[389,75],[4,75]] rec text: ooundaryproposals,andthenadoptanadaptiveboundary rec score: 0.946296
+4       det boxes: [[5,80],[387,80],[387,93],[5,93]] rec text: leformationmodelcombinedwithGCNandRNNtoper rec score: 0.952401
+********** end print ocr result **********
+5       type: title, region: [444,705,564,724], score: 0.785429, res: count of ocr result is : 1
+********** print ocr result **********
+0       det boxes: [[6,2],[113,2],[113,14],[6,14]] rec text: 5.Conclusion rec score: 0.856903
+********** end print ocr result **********
+6       type: table, region: [14,360,402,711], score: 0.963643, res: <html><body><table><thead><tr><td>Methods</td><td>Ext</td><td>R</td><td>P</td><td>F</td><td>FPS</td></tr></thead><tbody><tr><td>TextSnake [18]</td><td>Syn</td><td>85.3</td><td>67.9</td><td>75.6</td><td></td></tr><tr><td>CSE [17]</td><td>MiLT</td><td>76.1</td><td>78.7</td><td>77.4</td><td>0.38</td></tr><tr><td>LOMO[40]</td><td>Syn</td><td>76.5</td><td>85.7</td><td>80.8</td><td>4.4</td></tr><tr><td>ATRR[35]</td><td>Sy-</td><td>80.2</td><td>80.1</td><td>80.1</td><td>-</td></tr><tr><td>SegLink++ [28]</td><td>Syn</td><td>79.8</td><td>82.8</td><td>81.3</td><td>-</td></tr><tr><td>TextField [37]</td><td>Syn</td><td>79.8</td><td>83.0</td><td>81.4</td><td>6.0</td></tr><tr><td>MSR[38]</td><td>Syn</td><td>79.0</td><td>84.1</td><td>81.5</td><td>4.3</td></tr><tr><td>PSENet-1s [33]</td><td>MLT</td><td>79.7</td><td>84.8</td><td>82.2</td><td>3.9</td></tr><tr><td>DB [12]</td><td>Syn</td><td>80.2</td><td>86.9</td><td>83.4</td><td>22.0</td></tr><tr><td>CRAFT [2]</td><td>Syn</td><td>81.1</td><td>86.0</td><td>83.5</td><td>-</td></tr><tr><td>TextDragon [5]</td><td>MLT+</td><td>82.8</td><td>84.5</td><td>83.6</td><td></td></tr><tr><td>PAN [34]</td><td>Syn</td><td>81.2</td><td>86.4</td><td>83.7</td><td>39.8</td></tr><tr><td>ContourNet [36]</td><td></td><td>84.1</td><td>83.7</td><td>83.9</td><td>4.5</td></tr><tr><td>DRRG [41]</td><td>MLT</td><td>83.02</td><td>85.93</td><td>84.45</td><td>-</td></tr><tr><td>TextPerception[23]</td><td>Syn</td><td>81.9</td><td>87.5</td><td>84.6</td><td></td></tr><tr><td>Ours</td><td> Syn</td><td>80.57</td><td>87.66</td><td>83.97</td><td>12.08</td></tr><tr><td>Ours</td><td></td><td>81.45</td><td>87.81</td><td>84.51</td><td>12.15</td></tr><tr><td>Ours</td><td>MLT</td><td>83.60</td><td>86.45</td><td>85.00</td><td>12.21</td></tr></tbody></table></body></html>
+The table visualized image saved in ./output//6_1.png
+7       type: table, region: [462,359,820,657], score: 0.953917, res: <html><body><table><thead><tr><td>Methods</td><td>R</td><td>P</td><td>F</td><td>FPS</td></tr></thead><tbody><tr><td>SegLink [26]</td><td>70.0</td><td>86.0</td><td>77.0</td><td>8.9</td></tr><tr><td>PixelLink [4]</td><td>73.2</td><td>83.0</td><td>77.8</td><td>-</td></tr><tr><td>TextSnake [18]</td><td>73.9</td><td>83.2</td><td>78.3</td><td>1.1</td></tr><tr><td>TextField [37]</td><td>75.9</td><td>87.4</td><td>81.3</td><td>5.2 </td></tr><tr><td>MSR[38]</td><td>76.7</td><td>87.4</td><td>81.7</td><td>-</td></tr><tr><td>FTSN[3]</td><td>77.1</td><td>87.6</td><td>82.0</td><td>:</td></tr><tr><td>LSE[30]</td><td>81.7</td><td>84.2</td><td>82.9</td><td></td></tr><tr><td>CRAFT [2]</td><td>78.2</td><td>88.2</td><td>82.9</td><td>8.6</td></tr><tr><td>MCN [16]</td><td>79</td><td>88</td><td>83</td><td>-</td></tr><tr><td>ATRR[35]</td><td>82.1</td><td>85.2</td><td>83.6</td><td>-</td></tr><tr><td>PAN [34]</td><td>83.8</td><td>84.4</td><td>84.1</td><td>30.2</td></tr><tr><td>DB[12]</td><td>79.2</td><td>91.5</td><td>84.9</td><td>32.0</td></tr><tr><td>DRRG [41]</td><td>82.30</td><td>88.05</td><td>85.08</td><td>-</td></tr><tr><td>Ours (SynText)</td><td>80.68</td><td>85.40</td><td>82.97</td><td>12.68</td></tr><tr><td>Ours (MLT-17)</td><td>84.54</td><td>86.62</td><td>85.57</td><td>12.31</td></tr></tbody></table></body></html>
+The table visualized image saved in ./output//7_1.png
+8       type: figure, region: [14,3,836,310], score: 0.969443, res: count of ocr result is : 26
+********** print ocr result **********
+0       det boxes: [[506,14],[539,15],[539,22],[506,21]] rec text: E rec score: 0.318073
+...
+25      det boxes: [[680,290],[759,288],[759,303],[680,305]] rec text: (d) CTW1500 rec score: 0.95911
+********** end print ocr result **********
+```
+
+## 3. FAQ
+
+ 1. 遇到报错 `unable to access 'https://github.com/LDOUBLEV/AutoLog.git/': gnutls_handshake() failed: The TLS connection was non-properly terminated.`， 将 `deploy/cpp_infer/external-cmake/auto-log.cmake` 中的github地址改为 <https://gitee.com/Double_V/AutoLog> 地址即可。
diff --git a/docs/version2.x/ppstructure/infer_deploy/images/deployment-20240704135743247.png b/docs/version2.x/ppstructure/infer_deploy/images/deployment-20240704135743247.png
new file mode 100644
index 0000000..afd3cf5
Binary files /dev/null and b/docs/version2.x/ppstructure/infer_deploy/images/deployment-20240704135743247.png differ
diff --git a/docs/version2.x/ppstructure/infer_deploy/index.en.md b/docs/version2.x/ppstructure/infer_deploy/index.en.md
new file mode 100644
index 0000000..3f80b27
--- /dev/null
+++ b/docs/version2.x/ppstructure/infer_deploy/index.en.md
@@ -0,0 +1,21 @@
+---
+comments: true
+---
+
+# PP-OCR Deployment
+
+## Paddle Deployment Introduction
+
+Paddle provides a variety of deployment schemes to meet the deployment requirements of different scenarios. Please choose according to the actual situation:
+
+![img](./images/deployment-20240704135743247.png)
+
+PP-OCR has supported multi deployment schemes. Click the link to get the specific tutorial.
+
+- [Python Inference](./python_infer.en.md)
+- [C++ Inference](./cpp_infer.en.md)
+- [Serving (Python/C++)](./paddle_server.en.md)
+- [Paddle-Lite (ARM CPU/OpenCL ARM GPU)](../..//legacy/lite.en.md)
+- [Paddle2ONNX](../../legacy/paddle2onnx.en.md)
+
+If you need the deployment tutorial of academic algorithm models other than PP-OCR, please directly enter the main page of corresponding algorithms, [entrance](../../algorithm/overview.en.md)。
diff --git a/docs/version2.x/ppstructure/infer_deploy/index.md b/docs/version2.x/ppstructure/infer_deploy/index.md
new file mode 100644
index 0000000..7f936d5
--- /dev/null
+++ b/docs/version2.x/ppstructure/infer_deploy/index.md
@@ -0,0 +1,24 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# PP-OCR 模型推理部署
+
+## Paddle 推理部署方式简介
+
+飞桨提供多种部署方案，以满足不同场景的部署需求，请根据实际情况进行选择：
+
+![img](./images/deployment-20240704135743247.png)
+
+## PP-OCR 推理部署
+
+PP-OCR模型已打通多种场景部署方案，点击链接获取具体的使用教程。
+
+- [Python 推理](./python_infer.md)
+- [C++ 推理](./cpp_infer.md)
+- [Serving 服务化部署（Python/C++）](./paddle_server.md)
+- [Paddle-Lite 端侧部署（ARM CPU/OpenCL ARM GPU）](../../legacy/lite.md)
+- [Paddle2ONNX 推理](../../legacy/paddle2onnx.md)
+
+需要PP-OCR以外的学术算法模型的推理部署，请直接进入相应算法主页面，[入口](../../algorithm/overview.md)。
diff --git a/docs/version2.x/ppstructure/infer_deploy/paddle_server.en.md b/docs/version2.x/ppstructure/infer_deploy/paddle_server.en.md
new file mode 100755
index 0000000..1052c66
--- /dev/null
+++ b/docs/version2.x/ppstructure/infer_deploy/paddle_server.en.md
@@ -0,0 +1,273 @@
+---
+comments: true
+---
+
+PaddleOCR provides 2 service deployment methods:
+
+- Based on **PaddleHub Serving**: Code path is `./deploy/hubserving`. Please follow this tutorial.
+- Based on **PaddleServing**: Code path is `./deploy/pdserving`. Please refer to the [tutorial](../../legacy/paddle_server.en.md) for usage.
+
+# Service deployment based on PaddleHub Serving
+
+The hubserving service deployment directory includes seven service packages: text detection, text angle class, text recognition, text detection+text angle class+text recognition three-stage series connection, layout analysis, table recognition, and PP-Structure. Please select the corresponding service package to install and start the service according to your needs. The directory is as follows:
+
+```text linenums="1"
+deploy/hubserving/
+  └─  ocr_det     text detection module service package
+  └─  ocr_cls     text angle class module service package
+  └─  ocr_rec     text recognition module service package
+  └─  ocr_system  text detection+text angle class+text recognition three-stage series connection service package
+  └─  structure_layout  layout analysis service package
+  └─  structure_table  table recognition service package
+  └─  structure_system  PP-Structure service package
+  └─  kie_ser  KIE(SER) service package
+  └─  kie_ser_re  KIE(SER+RE) service package
+```
+
+Each service pack contains 3 files. Take the 2-stage series connection service package as an example, the directory is as follows:
+
+```text linenums="1"
+deploy/hubserving/ocr_system/
+  └─  __init__.py    Empty file, required
+  └─  config.json    Configuration file, optional, passed in as a parameter when using configuration to start the service
+  └─  module.py      Main module file, required, contains the complete logic of the service
+  └─  params.py      Parameter file, required, including parameters such as model path, pre and post-processing parameters
+```
+
+## 1. Update
+
+- 2022.10.09 add KIE services.
+- 2022.08.23 add layout analysis services.
+- 2022.03.30 add PP-Structure and table recognition services.
+- 2022.05.05 add PP-OCRv3 text detection and recognition services.
+
+## 2. Quick start service
+
+The following steps take the 2-stage series service as an example. If only the detection service or recognition service is needed, replace the corresponding file path.
+
+### 2.1 Install PaddleHub
+
+```bash linenums="1"
+pip3 install paddlehub==2.1.0 --upgrade
+```
+
+### 2.2 Download inference model
+
+Before installing the service module, you need to prepare the inference model and put it in the correct path. By default, the PP-OCRv3 models are used, and the default model path is:
+
+| Model | Path |
+| ------- | - |
+| text detection model | ./inference/PP-OCRv3_mobile_det_infer/ |
+| text recognition model | ./inference/PP-OCRv3_mobile_rec_infer/ |
+| text angle classifier | ./inference/ch_ppocr_mobile_v2.0_cls_infer/ |
+| layout parse model | ./inference/picodet_lcnet_x1_0_fgd_layout_infer/ |
+| tanle recognition | ./inference/ch_ppstructure_mobile_v2.0_SLANet_infer/ |
+| KIE(SER) | ./inference/ser_vi_layoutxlm_xfund_infer/ |
+| KIE(SER+RE) | ./inference/re_vi_layoutxlm_xfund_infer/ |
+
+**The model path can be found and modified in `params.py`.**
+More models provided by PaddleOCR can be obtained from the [model library](../../ppocr/model_list.en.md). You can also use models trained by yourself.
+
+### 2.3 Install Service Module
+
+PaddleOCR provides 5 kinds of service modules, install the required modules according to your needs.
+
+- On the Linux platform(replace `/` with `\` if using Windows), the examples are as the following table:
+
+| Service model | Command |
+| text detection | `hub install deploy/hubserving/ocr_det` |
+| text angle class: | `hub install deploy/hubserving/ocr_cls` |
+| text recognition: | `hub install deploy/hubserving/ocr_rec` |
+| 2-stage series: | `hub install deploy/hubserving/ocr_system` |
+| table recognition | `hub install deploy/hubserving/structure_table` |
+| PP-Structure | `hub install deploy/hubserving/structure_system` |
+| KIE(SER) | `hub install deploy/hubserving/kie_ser` |
+| KIE(SER+RE) | `hub install deploy/hubserving/kie_ser_re` |
+
+### 2.4 Start service
+
+#### 2.4.1 Start with command line parameters (CPU only)
+
+**start command:**
+
+```bash linenums="1"
+hub serving start --modules Module1==Version1, Module2==Version2, ... \
+                  --port 8866 \
+                  --use_multiprocess \
+                  --workers \
+```
+
+**Parameters:**
+|parameters|usage|
+|---|---|
+|`--modules`/`-m`|PaddleHub Serving pre-installed model, listed in the form of multiple Module==Version key-value pairs<br>**When Version is not specified, the latest version is selected by default**|
+|`--port`/`-p`|Service port, default is 8866|
+|`--use_multiprocess`|Enable concurrent mode, by default using the single-process mode, this mode is recommended for multi-core CPU machines<br>**Windows operating system only supports single-process mode**|
+|`--workers`|The number of concurrent tasks specified in concurrent mode, the default is `2*cpu_count-1`, where `cpu_count` is the number of CPU cores|
+
+For example, start the 2-stage series service:
+
+```bash linenums="1"
+hub serving start -m ocr_system
+```
+
+This completes the deployment of a service API, using the default port number 8866.
+
+#### 2.4.2 Start with configuration file（CPU and GPU）
+
+**start command:**
+
+```bash linenums="1"
+hub serving start --config/-c config.json
+```
+
+In which the format of `config.json` is as follows:
+
+```json
+{
+    "modules_info": {
+        "ocr_system": {
+            "init_args": {
+                "version": "1.0.0",
+                "use_gpu": true
+            },
+            "predict_args": {
+            }
+        }
+    },
+    "port": 8868,
+    "use_multiprocess": false,
+    "workers": 2
+}
+```
+
+- The configurable parameters in `init_args` are consistent with the `_initialize` function interface in `module.py`.
+
+  **When `use_gpu` is `true`, it means that the GPU is used to start the service**.
+- The configurable parameters in `predict_args` are consistent with the `predict` function interface in `module.py`.
+
+  **Note:**
+  - When using the configuration file to start the service, other parameters will be ignored.
+  - If you use GPU prediction (that is, `use_gpu` is set to `true`), you need to set the environment variable CUDA_VISIBLE_DEVICES before starting the service, such as:
+
+    ```bash linenums="1"
+    export CUDA_VISIBLE_DEVICES=0
+    ```
+
+  - **`use_gpu` and `use_multiprocess` cannot be `true` at the same time.**
+
+For example, use GPU card No. 3 to start the 2-stage series service:
+
+```bash linenums="1"
+export CUDA_VISIBLE_DEVICES=3
+hub serving start -c deploy/hubserving/ocr_system/config.json
+```
+
+## 3. Send prediction requests
+
+After the service starts, you can use the following command to send a prediction request to obtain the prediction result:
+
+```bash linenums="1"
+python tools/test_hubserving.py --server_url=server_url --image_dir=image_path
+```
+
+Two parameters need to be passed to the script:
+
+- **server_url**:service address, the format of which is
+  `http://[ip_address]:[port]/predict/[module_name]`
+
+  For example, if using the configuration file to start the text angle classification, text detection, text recognition, detection+classification+recognition 3 stages, table recognition and PP-Structure service,
+
+  also modified the port for each service, then the `server_url` to send the request will be:
+
+  ```text linenums="1"
+  http://127.0.0.1:8865/predict/ocr_det
+  http://127.0.0.1:8866/predict/ocr_cls
+  http://127.0.0.1:8867/predict/ocr_rec
+  http://127.0.0.1:8868/predict/ocr_system
+  http://127.0.0.1:8869/predict/structure_table
+  http://127.0.0.1:8870/predict/structure_system
+  http://127.0.0.1:8870/predict/structure_layout
+  http://127.0.0.1:8871/predict/kie_ser
+  http://127.0.0.1:8872/predict/kie_ser_re
+  ```
+
+- **image_dir**:Test image path, which can be a single image path or an image directory path
+- **visualize**:Whether to visualize the results, the default value is False
+- **output**:The folder to save the Visualization result, the default value is `./hubserving_result`
+
+Example:
+
+```bash linenums="1"
+python tools/test_hubserving.py --server_url=http://127.0.0.1:8868/predict/ocr_system --image_dir=./doc/imgs/ --visualize=false`
+```
+
+## 4. Returned result format
+
+The returned result is a list. Each item in the list is a dictionary which may contain three fields. The information is as follows:
+
+|field name|data type|description|
+|----|----|----|
+|angle|str|angle|
+|text|str|text content|
+|confidence|float|text recognition confidence|
+|text_region|list|text location coordinates|
+|html|str|table HTML string|
+|regions|list|The result of layout analysis + table recognition + OCR, each item is a list<br>including `bbox` indicating area coordinates, `type` of area type and `res` of area results|
+|layout|list|The result of layout analysis, each item is a dict, including `bbox` indicating area coordinates, `label` of area type|
+
+The fields returned by different modules are different. For example, the results returned by the text recognition service module do not contain `text_region`, detailed table is as follows:
+
+|field name/module name |ocr_det |ocr_cls |ocr_rec |ocr_system |structure_table |structure_system |structure_layout |kie_ser |kie_re |
+|---                    |---     |---     |---     |---        |---             |---              |---              |---     |---    |
+|angle                  |        |✔       |        |✔          |                |                 |                 |
+|text                   |        |        |✔       |✔          |                |✔                |                 |✔       |✔      |
+|confidence             |        |✔       |✔       |✔          |                |✔                |                 |✔       |✔      |
+|text_region            |✔       |        |        |✔          |                |✔                |                 |✔       |✔      |
+|html                   |        |        |        |           |✔               |✔                |                 |        |       |
+|regions                |        |        |        |           |✔               |✔                |                 |        |       |
+|layout                 |        |        |        |           |                |                 |✔                |        |       |
+|ser_res                |        |        |        |           |                |                 |                 |✔       |       |
+|re_res                 |        |        |        |           |                |                 |                 |        |✔      |
+
+**Note:** If you need to add, delete or modify the returned fields, you can modify the file `module.py` of the corresponding module. For the complete process, refer to the user-defined modification service module in the next section.
+
+## 5. User-defined service module modification
+
+If you need to modify the service logic, the following steps are generally required (take the modification of `deploy/hubserving/ocr_system` for example):
+
+1. Stop service:
+
+```bash linenums="1"
+hub serving stop --port/-p XXXX
+```
+
+2. Modify the code in the corresponding files under `deploy/hubserving/ocr_system`, such as `module.py` and `params.py`, to your actual needs.
+
+   For example, if you need to replace the model used by the deployed service, you need to modify model path parameters `det_model_dir` and `rec_model_dir` in `params.py`. If you want to turn off the text direction classifier, set the parameter `use_angle_cls` to `False`.
+
+   Of course, other related parameters may need to be modified at the same time. Please modify and debug according to the actual situation.
+
+   **It is suggested to run `module.py` directly for debugging after modification before starting the service test.**
+
+   **Note** The image input shape used by the PPOCR-v3 recognition model is `3, 48, 320`, so you need to modify `cfg.rec_image_shape = "3, 48, 320"` in `params.py`, if you do not use the PPOCR-v3 recognition model, then there is no need to modify this parameter.
+3. (Optional) If you want to rename the module, the following lines should be modified:
+   - [`ocr_system` within `from deploy.hubserving.ocr_system.params import read_params`](../../../../deploy/hubserving/ocr_system/module.py#L35)
+   - [`ocr_system` within `name="ocr_system",`](../../../../deploy/hubserving/ocr_system/module.py#L39)
+4. (Optional) It may require you to delete the directory `__pycache__` to force flush build cache of CPython:
+
+   ```bash linenums="1"
+   find deploy/hubserving/ocr_system -name '__pycache__' -exec rm -r {} \;
+   ```
+
+5. Install modified service module:
+
+   ```bash linenums="1"
+   hub install deploy/hubserving/ocr_system/
+   ```
+
+6. Restart service:
+
+   ```bash linenums="1"
+   hub serving start -m ocr_system
+   ```
diff --git a/docs/version2.x/ppstructure/infer_deploy/paddle_server.md b/docs/version2.x/ppstructure/infer_deploy/paddle_server.md
new file mode 100644
index 0000000..843fd91
--- /dev/null
+++ b/docs/version2.x/ppstructure/infer_deploy/paddle_server.md
@@ -0,0 +1,279 @@
+---
+comments: true
+---
+
+PaddleOCR提供2种服务部署方式：
+
+- 基于PaddleHub Serving的部署：代码路径为`./deploy/hubserving`，按照本教程使用；
+- 基于PaddleServing的部署：代码路径为`./deploy/pdserving`，使用方法参考[文档](../../legacy/paddle_server.md)。
+
+# 基于PaddleHub Serving的服务部署
+
+hubserving服务部署目录下包括文本检测、文本方向分类，文本识别、文本检测+文本方向分类+文本识别3阶段串联，版面分析、表格识别和PP-Structure七种服务包，请根据需求选择相应的服务包进行安装和启动。目录结构如下：
+
+```text linenums="1"
+deploy/hubserving/
+  └─  ocr_cls     文本方向分类模块服务包
+  └─  ocr_det     文本检测模块服务包
+  └─  ocr_rec     文本识别模块服务包
+  └─  ocr_system  文本检测+文本方向分类+文本识别串联服务包
+  └─  structure_layout  版面分析服务包
+  └─  structure_table  表格识别服务包
+  └─  structure_system  PP-Structure服务包
+  └─  kie_ser  关键信息抽取-SER服务包
+  └─  kie_ser_re  关键信息抽取-SER+RE服务包
+```
+
+每个服务包下包含3个文件。以2阶段串联服务包为例，目录如下：
+
+```text linenums="1"
+deploy/hubserving/ocr_system/
+  └─  __init__.py    空文件，必选
+  └─  config.json    配置文件，可选，使用配置启动服务时作为参数传入
+  └─  module.py      主模块，必选，包含服务的完整逻辑
+  └─  params.py      参数文件，必选，包含模型路径、前后处理参数等参数
+```
+
+## 1. 近期更新
+
+- 2022.10.09 新增关键信息抽取服务。
+- 2022.08.23 新增版面分析服务。
+- 2022.05.05 新增PP-OCRv3检测和识别模型。
+- 2022.03.30 新增PP-Structure和表格识别两种服务。
+
+## 2. 快速启动服务
+
+以下步骤以检测+识别2阶段串联服务为例，如果只需要检测服务或识别服务，替换相应文件路径即可。
+
+### 2.1 安装PaddleHub
+
+paddlehub 需要 python>3.6.2
+
+```bash linenums="1"
+pip3 install paddlehub==2.1.0 --upgrade -i https://mirror.baidu.com/pypi/simple
+```
+
+### 2.2 下载推理模型
+
+安装服务模块前，需要准备推理模型并放到正确路径。默认使用的是PP-OCRv3模型，默认模型路径为：
+
+| 模型                | 路径                                                   |
+| ------------------- | ------------------------------------------------------ |
+| 检测模型            | `./inference/PP-OCRv3_mobile_det_infer/`                   |
+| 识别模型            | `./inference/PP-OCRv3_mobile_rec_infer/`                   |
+| 方向分类器          | `./inference/ch_ppocr_mobile_v2.0_cls_infer/`          |
+| 版面分析模型        | `./inference/picodet_lcnet_x1_0_fgd_layout_infer/`     |
+| 表格结构识别模型    | `./inference/ch_ppstructure_mobile_v2.0_SLANet_infer/` |
+| 关键信息抽取SER模型 | `./inference/ser_vi_layoutxlm_xfund_infer/`            |
+| 关键信息抽取RE模型  | `./inference/re_vi_layoutxlm_xfund_infer/`             |
+
+**模型路径可在`params.py`中查看和修改。**
+
+更多模型可以从PaddleOCR提供的模型库[PP-OCR](../../ppocr/model_list.md)和[PP-Structure](../models_list.md)下载，也可以替换成自己训练转换好的模型。
+
+### 2.3 安装服务模块
+
+PaddleOCR提供5种服务模块，根据需要安装所需模块。
+
+在Linux环境（Windows环境请将`/`替换为`\`）下，安装模块命令如下表：
+
+| 服务模块           | 命令                                             |
+| ------------------ | ------------------------------------------------ |
+| 检测               | `hub install deploy/hubserving/ocr_det`          |
+| 分类               | `hub install deploy/hubserving/ocr_cls`          |
+| 识别               | `hub install deploy/hubserving/ocr_rec`          |
+| 检测+识别串联      | `hub install deploy/hubserving/ocr_system`       |
+| 表格识别           | `hub install deploy/hubserving/structure_table`  |
+| PP-Structure       | `hub install deploy/hubserving/structure_system` |
+| 版面分析           | `hub install deploy/hubserving/structure_layout` |
+| 关键信息抽取SER    | `hub install deploy/hubserving/kie_ser`          |
+| 关键信息抽取SER+RE | `hub install deploy/hubserving/kie_ser_re`       |
+
+### 2.4 启动服务
+
+#### 2.4.1. 命令行命令启动（仅支持CPU）
+
+**启动命令：**
+
+```bash linenums="1"
+hub serving start --modules Module1==Version1, Module2==Version2, ... \
+                  --port 8866 \
+                  --use_multiprocess \
+                  --workers \
+```
+
+**参数：**
+
+| 参数                 | 用途                                                                                                          |
+| ----- | ---- |
+| `--modules`/`-m`     | PaddleHub Serving预安装模型，以多个Module==Version键值对的形式列出<br>**当不指定Version时，默认选择最新版本** |
+| `--port`/`-p`        | 服务端口，默认为8866                                                                                          |
+| `--use_multiprocess` | 是否启用并发方式，默认为单进程方式，推荐多核CPU机器使用此方式<br>**Windows操作系统只支持单进程方式**          |
+| `--workers`          | 在并发方式下指定的并发任务数，默认为`2*cpu_count-1`，其中`cpu_count`为CPU核数                                 |
+
+如启动串联服务：
+
+```bash linenums="1"
+hub serving start -m ocr_system
+```
+
+这样就完成了一个服务化API的部署，使用默认端口号8866。
+
+#### 2.4.2 配置文件启动（支持CPU、GPU）
+
+**启动命令：**
+
+```bash linenums="1"
+hub serving start -c config.json
+```
+
+其中，`config.json`格式如下：
+
+```json
+{
+    "modules_info": {
+        "ocr_system": {
+            "init_args": {
+                "version": "1.0.0",
+                "use_gpu": true
+            },
+            "predict_args": {
+            }
+        }
+    },
+    "port": 8868,
+    "use_multiprocess": false,
+    "workers": 2
+}
+```
+
+- `init_args`中的可配参数与`module.py`中的`_initialize`函数接口一致。
+
+   **当`use_gpu`为`true`时，表示使用GPU启动服务。**
+- `predict_args`中的可配参数与`module.py`中的`predict`函数接口一致。
+
+**注意：**
+
+- 使用配置文件启动服务时，其他参数会被忽略。
+- 如果使用GPU预测(即，`use_gpu`置为`true`)，则需要在启动服务之前，设置CUDA_VISIBLE_DEVICES环境变量，如：
+
+  ```bash linenums="1"
+  export CUDA_VISIBLE_DEVICES=0
+  ```
+
+- **`use_gpu`不可与`use_multiprocess`同时为`true`**。
+
+如，使用GPU 3号卡启动串联服务：
+
+```bash linenums="1"
+export CUDA_VISIBLE_DEVICES=3
+hub serving start -c deploy/hubserving/ocr_system/config.json
+```
+
+## 3. 发送预测请求
+
+配置好服务端，可使用以下命令发送预测请求，获取预测结果：
+
+```bash linenums="1"
+python tools/test_hubserving.py --server_url=server_url --image_dir=image_path
+```
+
+需要给脚本传递2个参数：
+
+- `server_url`：服务地址，格式为`http://[ip_address]:[port]/predict/[module_name]`
+
+   例如，如果使用配置文件启动分类，检测、识别，检测+分类+识别3阶段，表格识别和PP-Structure服务
+
+   并为每个服务修改了port，那么发送请求的url将分别是：
+
+   ```text linenums="1"
+   http://127.0.0.1:8865/predict/ocr_det
+   http://127.0.0.1:8866/predict/ocr_cls
+   http://127.0.0.1:8867/predict/ocr_rec
+   http://127.0.0.1:8868/predict/ocr_system
+   http://127.0.0.1:8869/predict/structure_table
+   http://127.0.0.1:8870/predict/structure_system
+   http://127.0.0.1:8870/predict/structure_layout
+   http://127.0.0.1:8871/predict/kie_ser
+   http://127.0.0.1:8872/predict/kie_ser_re
+   ```
+
+- `image_dir`：测试图像路径，可以是单张图片路径，也可以是图像集合目录路径
+- `visualize`：是否可视化结果，默认为False
+- `output`：可视化结果保存路径，默认为`./hubserving_result`
+
+访问示例：
+
+```bash linenums="1"
+python tools/test_hubserving.py --server_url=http://127.0.0.1:8868/predict/ocr_system --image_dir=./doc/imgs/ --visualize=false
+```
+
+## 4. 返回结果格式说明
+
+返回结果为列表（list），列表中的每一项为词典（dict），词典一共可能包含3种字段，信息如下：
+
+| 字段名称    | 数据类型 | 意义  |
+| ----------- | -------- | ----- |
+| angle       | str      | 文本角度                                                                                                             |
+| text        | str      | 文本内容                                                                                                             |
+| confidence  | float    | 文本识别置信度或文本角度分类置信度                                                                                   |
+| text_region | list     | 文本位置坐标                                                                                                         |
+| html        | str      | 表格的html字符串                                                                                                     |
+| regions     | list     | 版面分析+表格识别+OCR的结果，每一项为一个list<br>包含表示区域坐标的`bbox`，区域类型的`type`和区域结果的`res`三个字段 |
+| layout      | list     | 版面分析的结果，每一项一个dict，包含版面区域坐标的`bbox`，区域类型的`label`                                          |
+
+不同模块返回的字段不同，如，文本识别服务模块返回结果不含`text_region`字段，具体信息如下：
+
+| 字段名/模块名 | ocr_det | ocr_cls | ocr_rec | ocr_system | structure_table | structure_system | structure_layout | kie_ser | kie_re |
+| ------------- | ------- | ------- | ------- | ---------- | --------------- | ---------------- | ---------------- | ------- | ------ |
+| angle         |         | ✔       |         | ✔          |                 |                  |                  |
+| text          |         |         | ✔       | ✔          |                 | ✔                |                  | ✔       | ✔      |
+| confidence    |         | ✔       | ✔       | ✔          |                 | ✔                |                  | ✔       | ✔      |
+| text_region   | ✔       |         |         | ✔          |                 | ✔                |                  | ✔       | ✔      |
+| html          |         |         |         |            | ✔               | ✔                |                  |         |        |
+| regions       |         |         |         |            | ✔               | ✔                |                  |         |        |
+| layout        |         |         |         |            |                 |                  | ✔                |         |        |
+| ser_res       |         |         |         |            |                 |                  |                  | ✔       |        |
+| re_res        |         |         |         |            |                 |                  |                  |         | ✔      |
+
+**说明：** 如果需要增加、删除、修改返回字段，可在相应模块的`module.py`文件中进行修改，完整流程参考下一节自定义修改服务模块。
+
+## 5. 自定义修改服务模块
+
+如果需要修改服务逻辑，一般需要操作以下步骤（以修改`deploy/hubserving/ocr_system`为例）：
+
+1. 停止服务：
+
+   ```bash linenums="1"
+   hub serving stop --port/-p XXXX
+   ```
+
+2. 到`deploy/hubserving/ocr_system`下的`module.py`和`params.py`等文件中根据实际需求修改代码。
+
+   例如，如果需要替换部署服务所用模型，则需要到`params.py`中修改模型路径参数`det_model_dir`和`rec_model_dir`，如果需要关闭文本方向分类器，则将参数`use_angle_cls`置为`False`
+
+   当然，同时可能还需要修改其他相关参数，请根据实际情况修改调试。
+
+   **强烈建议修改后先直接运行`module.py`调试，能正确运行预测后再启动服务测试。**
+
+   **注意：** PPOCR-v3识别模型使用的图片输入shape为`3,48,320`,因此需要修改`params.py`中的`cfg.rec_image_shape = "3, 48, 320"`，如果不使用PPOCR-v3识别模型，则无需修改该参数。
+3. （可选）如果想要重命名模块需要更改`module.py`文件中的以下行：
+   - [`from deploy.hubserving.ocr_system.params import read_params`中的`ocr_system`](../../../../deploy/hubserving/ocr_system/module.py#L35)
+   - [`name="ocr_system",`中的`ocr_system`](../../../../deploy/hubserving/ocr_system/module.py#L39)
+4. （可选）可能需要删除`__pycache__`目录以强制刷新CPython缓存：
+
+   ```bash linenums="1"
+   find deploy/hubserving/ocr_system -name '__pycache__' -exec rm -r {} \;
+   ```
+
+5. 安装修改后的新服务包：
+
+   ```bash linenums="1"
+   hub install deploy/hubserving/ocr_system
+   ```
+
+6. 重新启动服务：
+
+   ```bash linenums="1"
+   hub serving start -m ocr_system
+   ```
diff --git a/docs/version2.x/ppstructure/infer_deploy/python_infer.en.md b/docs/version2.x/ppstructure/infer_deploy/python_infer.en.md
new file mode 100644
index 0000000..458be69
--- /dev/null
+++ b/docs/version2.x/ppstructure/infer_deploy/python_infer.en.md
@@ -0,0 +1,116 @@
+---
+comments: true
+---
+
+# Python Inference
+
+## 1. Layout Structured Analysis
+
+Go to the `ppstructure` directory
+
+```bash linenums="1"
+cd ppstructure
+
+# download model
+mkdir inference && cd inference
+# Download the PP-StructureV2 layout analysis model and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_layout_infer.tar && tar xf picodet_lcnet_x1_0_layout_infer.tar
+# Download the PP-OCRv3 text detection model and unzip it
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar && tar xf PP-OCRv3_mobile_det_infer.tar
+# Download the PP-OCRv3 text recognition model and unzip it
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar && tar xf PP-OCRv3_mobile_rec_infer.tar
+# Download the PP-StructureV2 form recognition model and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/ch_ppstructure_mobile_v2.0_SLANet_infer.tar && tar xf ch_ppstructure_mobile_v2.0_SLANet_infer.tar
+cd ..
+```
+
+### 1.1 layout analysis + table recognition
+
+```bash linenums="1"
+python3 predict_system.py --det_model_dir=inference/PP-OCRv3_mobile_det_infer \
+                          --rec_model_dir=inference/PP-OCRv3_mobile_rec_infer \
+                          --table_model_dir=inference/ch_ppstructure_mobile_v2.0_SLANet_infer \
+                          --layout_model_dir=inference/picodet_lcnet_x1_0_layout_infer \
+                          --image_dir=./docs/table/1.png \
+                          --rec_char_dict_path=../ppocr/utils/ppocr_keys_v1.txt \
+                          --table_char_dict_path=../ppocr/utils/dict/table_structure_dict_ch.txt \
+                          --output=../output \
+                          --vis_font_path=../doc/fonts/simfang.ttf
+```
+
+After the operation is completed, each image will have a directory with the same name in the `structure` directory under the directory specified by the `output` field. Each table in the image will be stored as an excel, and the picture area will be cropped and saved. The filename of excel and picture is their coordinates in the image. Detailed results are stored in the `res.txt` file.
+
+### 1.2 layout analysis
+
+```bash linenums="1"
+python3 predict_system.py --layout_model_dir=inference/picodet_lcnet_x1_0_layout_infer \
+                          --image_dir=./docs/table/1.png \
+                          --output=../output \
+                          --table=false \
+                          --ocr=false
+```
+
+After the operation is completed, each image will have a directory with the same name in the `structure` directory under the directory specified by the `output` field. Each picture in image will be cropped and saved. The filename of picture area is their coordinates in the image. Layout analysis results will be stored in the `res.txt` file
+
+### 1.3 table recognition
+
+```bash linenums="1"
+python3 predict_system.py --det_model_dir=inference/PP-OCRv3_mobile_det_infer \
+                          --rec_model_dir=inference/PP-OCRv3_mobile_rec_infer \
+                          --table_model_dir=inference/ch_ppstructure_mobile_v2.0_SLANet_infer \
+                          --image_dir=./docs/table/table.jpg \
+                          --rec_char_dict_path=../ppocr/utils/ppocr_keys_v1.txt \
+                          --table_char_dict_path=../ppocr/utils/dict/table_structure_dict_ch.txt \
+                          --output=../output \
+                          --vis_font_path=../doc/fonts/simfang.ttf \
+                          --layout=false
+```
+
+After the operation is completed, each image will have a directory with the same name in the `structure` directory under the directory specified by the `output` field. Each table in the image will be stored as an excel. The filename of excel is their coordinates in the image.
+
+## 2. Key Information Extraction
+
+### 2.1 SER
+
+```bash linenums="1"
+cd ppstructure
+
+mkdir inference && cd inference
+# download model
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_infer.tar && tar -xf ser_vi_layoutxlm_xfund_infer.tar
+cd ..
+python3 predict_system.py \
+  --kie_algorithm=LayoutXLM \
+  --ser_model_dir=./inference/ser_vi_layoutxlm_xfund_infer \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../ppocr/utils/dict/kie_dict/xfund_class_list.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx" \
+  --mode=kie
+```
+
+After the operation is completed, each image will store the visualized image in the `kie` directory under the directory specified by the `output` field, and the image name is the same as the input image name.
+
+### 2.2 RE+SER
+
+```bash linenums="1"
+cd ppstructure
+
+mkdir inference && cd inference
+# download model
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_infer.tar && tar -xf ser_vi_layoutxlm_xfund_infer.tar
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_infer.tar && tar -xf re_vi_layoutxlm_xfund_infer.tar
+cd ..
+
+python3 predict_system.py \
+  --kie_algorithm=LayoutXLM \
+  --re_model_dir=./inference/re_vi_layoutxlm_xfund_infer \
+  --ser_model_dir=./inference/ser_vi_layoutxlm_xfund_infer \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../ppocr/utils/dict/kie_dict/xfund_class_list.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx" \
+  --mode=kie
+```
+
+After the operation is completed, each image will have a directory with the same name in the `kie` directory under the directory specified by the `output` field, where the visual images and prediction results are stored.
diff --git a/docs/version2.x/ppstructure/infer_deploy/python_infer.md b/docs/version2.x/ppstructure/infer_deploy/python_infer.md
new file mode 100644
index 0000000..3fcf3bf
--- /dev/null
+++ b/docs/version2.x/ppstructure/infer_deploy/python_infer.md
@@ -0,0 +1,119 @@
+---
+comments: true
+---
+
+# 基于Python预测引擎推理
+
+## 1. 版面信息抽取
+
+进入`ppstructure`目录
+
+```bash linenums="1"
+cd ppstructure
+```
+
+下载模型
+
+```bash linenums="1"
+mkdir inference && cd inference
+# 下载PP-StructureV2版面分析模型并解压
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_layout_infer.tar && tar xf picodet_lcnet_x1_0_layout_infer.tar
+# 下载PP-OCRv3文本检测模型并解压
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar && tar xf PP-OCRv3_mobile_det_infer.tar
+# 下载PP-OCRv3文本识别模型并解压
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar && tar xf PP-OCRv3_mobile_rec_infer.tar
+# 下载PP-StructureV2表格识别模型并解压
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/ch_ppstructure_mobile_v2.0_SLANet_infer.tar && tar xf ch_ppstructure_mobile_v2.0_SLANet_infer.tar
+cd ..
+```
+
+### 1.1 版面分析+表格识别
+
+```bash linenums="1"
+python3 predict_system.py --det_model_dir=inference/PP-OCRv3_mobile_det_infer \
+                          --rec_model_dir=inference/PP-OCRv3_mobile_rec_infer \
+                          --table_model_dir=inference/ch_ppstructure_mobile_v2.0_SLANet_infer \
+                          --layout_model_dir=inference/picodet_lcnet_x1_0_layout_infer \
+                          --image_dir=./docs/table/1.png \
+                          --rec_char_dict_path=../ppocr/utils/ppocr_keys_v1.txt \
+                          --table_char_dict_path=../ppocr/utils/dict/table_structure_dict_ch.txt \
+                          --output=../output \
+                          --vis_font_path=../doc/fonts/simfang.ttf
+```
+
+运行完成后，每张图片会在`output`字段指定的目录下的`structure`目录下有一个同名目录，图片里的每个表格会存储为一个excel，图片区域会被裁剪之后保存下来，excel文件和图片名为表格在图片里的坐标。详细的结果会存储在`res.txt`文件中。
+
+### 1.2 版面分析
+
+```bash linenums="1"
+python3 predict_system.py --layout_model_dir=inference/picodet_lcnet_x1_0_layout_infer \
+                          --image_dir=./docs/table/1.png \
+                          --output=../output \
+                          --table=false \
+                          --ocr=false
+```
+
+运行完成后，每张图片会在`output`字段指定的目录下的`structure`目录下有一个同名目录，图片区域会被裁剪之后保存下来，图片名为表格在图片里的坐标。版面分析结果会存储在`res.txt`文件中。
+
+### 1.3 表格识别
+
+```bash linenums="1"
+python3 predict_system.py --det_model_dir=inference/PP-OCRv3_mobile_det_infer \
+                          --rec_model_dir=inference/PP-OCRv3_mobile_rec_infer \
+                          --table_model_dir=inference/ch_ppstructure_mobile_v2.0_SLANet_infer \
+                          --image_dir=./docs/table/table.jpg \
+                          --rec_char_dict_path=../ppocr/utils/ppocr_keys_v1.txt \
+                          --table_char_dict_path=../ppocr/utils/dict/table_structure_dict_ch.txt \
+                          --output=../output \
+                          --vis_font_path=../doc/fonts/simfang.ttf \
+                          --layout=false
+```
+
+运行完成后，每张图片会在`output`字段指定的目录下的`structure`目录下有一个同名目录，表格会存储为一个excel，excel文件名为`[0,0,img_h,img_w]`。
+
+## 2. 关键信息抽取
+
+### 2.1 SER
+
+```bash linenums="1"
+cd ppstructure
+
+mkdir inference && cd inference
+# 下载SER XFUND 模型并解压
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_infer.tar && tar -xf ser_vi_layoutxlm_xfund_infer.tar
+cd ..
+python3 predict_system.py \
+  --kie_algorithm=LayoutXLM \
+  --ser_model_dir=./inference/ser_vi_layoutxlm_xfund_infer \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../ppocr/utils/dict/kie_dict/xfund_class_list.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx" \
+  --mode=kie
+```
+
+运行完成后，每张图片会在`output`字段指定的目录下的`kie`目录下存放可视化之后的图片，图片名和输入图片名一致。
+
+### 2.2 RE+SER
+
+```bash linenums="1"
+cd ppstructure
+
+mkdir inference && cd inference
+# 下载RE SER XFUND 模型并解压
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_infer.tar && tar -xf ser_vi_layoutxlm_xfund_infer.tar
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_infer.tar && tar -xf re_vi_layoutxlm_xfund_infer.tar
+cd ..
+
+python3 predict_system.py \
+  --kie_algorithm=LayoutXLM \
+  --re_model_dir=./inference/re_vi_layoutxlm_xfund_infer \
+  --ser_model_dir=./inference/ser_vi_layoutxlm_xfund_infer \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../ppocr/utils/dict/kie_dict/xfund_class_list.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx" \
+  --mode=kie
+```
+
+运行完成后，每张图片会在`output`字段指定的目录下的`kie`目录下有一个同名目录，目录中存放可视化图片和预测结果。
diff --git a/docs/version2.x/ppstructure/model_train/images/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808.jpg b/docs/version2.x/ppstructure/model_train/images/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808.jpg
new file mode 100644
index 0000000..6a5fd84
Binary files /dev/null and b/docs/version2.x/ppstructure/model_train/images/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808.jpg differ
diff --git a/docs/version2.x/ppstructure/model_train/images/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb.jpg b/docs/version2.x/ppstructure/model_train/images/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb.jpg
new file mode 100644
index 0000000..d406a52
Binary files /dev/null and b/docs/version2.x/ppstructure/model_train/images/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb.jpg differ
diff --git a/docs/version2.x/ppstructure/model_train/images/185539141-68e71c75-5cf7-4529-b2ca-219d29fa5f68.jpg b/docs/version2.x/ppstructure/model_train/images/185539141-68e71c75-5cf7-4529-b2ca-219d29fa5f68.jpg
new file mode 100644
index 0000000..42930f0
Binary files /dev/null and b/docs/version2.x/ppstructure/model_train/images/185539141-68e71c75-5cf7-4529-b2ca-219d29fa5f68.jpg differ
diff --git a/docs/version2.x/ppstructure/model_train/images/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a.png b/docs/version2.x/ppstructure/model_train/images/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a.png
new file mode 100644
index 0000000..cd5bc50
Binary files /dev/null and b/docs/version2.x/ppstructure/model_train/images/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a.png differ
diff --git a/docs/version2.x/ppstructure/model_train/images/185539735-37b5c2ef-629d-43fe-9abb-44bb717ef7ee.jpg b/docs/version2.x/ppstructure/model_train/images/185539735-37b5c2ef-629d-43fe-9abb-44bb717ef7ee.jpg
new file mode 100644
index 0000000..4abdc09
Binary files /dev/null and b/docs/version2.x/ppstructure/model_train/images/185539735-37b5c2ef-629d-43fe-9abb-44bb717ef7ee.jpg differ
diff --git a/docs/version2.x/ppstructure/model_train/images/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f.jpg b/docs/version2.x/ppstructure/model_train/images/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f.jpg
new file mode 100644
index 0000000..bdef2ed
Binary files /dev/null and b/docs/version2.x/ppstructure/model_train/images/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f.jpg differ
diff --git a/docs/version2.x/ppstructure/model_train/images/185540291-f64e5daf-6d42-4e7c-bbbb-471e3fac4fcc.png b/docs/version2.x/ppstructure/model_train/images/185540291-f64e5daf-6d42-4e7c-bbbb-471e3fac4fcc.png
new file mode 100644
index 0000000..b5f4577
Binary files /dev/null and b/docs/version2.x/ppstructure/model_train/images/185540291-f64e5daf-6d42-4e7c-bbbb-471e3fac4fcc.png differ
diff --git a/docs/version2.x/ppstructure/model_train/images/195319840-68fc60ec-ea66-4095-b734-0ec115860341.png b/docs/version2.x/ppstructure/model_train/images/195319840-68fc60ec-ea66-4095-b734-0ec115860341.png
new file mode 100644
index 0000000..d5b30f1
Binary files /dev/null and b/docs/version2.x/ppstructure/model_train/images/195319840-68fc60ec-ea66-4095-b734-0ec115860341.png differ
diff --git a/docs/version2.x/ppstructure/model_train/images/layout.jpg b/docs/version2.x/ppstructure/model_train/images/layout.jpg
new file mode 100644
index 0000000..b210054
Binary files /dev/null and b/docs/version2.x/ppstructure/model_train/images/layout.jpg differ
diff --git a/docs/version2.x/ppstructure/model_train/images/layout_res.jpg b/docs/version2.x/ppstructure/model_train/images/layout_res.jpg
new file mode 100644
index 0000000..93b3a8b
Binary files /dev/null and b/docs/version2.x/ppstructure/model_train/images/layout_res.jpg differ
diff --git a/docs/version2.x/ppstructure/model_train/images/recovery-20240708091126891.jpg b/docs/version2.x/ppstructure/model_train/images/recovery-20240708091126891.jpg
new file mode 100644
index 0000000..a3817ab
Binary files /dev/null and b/docs/version2.x/ppstructure/model_train/images/recovery-20240708091126891.jpg differ
diff --git a/docs/version2.x/ppstructure/model_train/images/recovery_ch.jpg b/docs/version2.x/ppstructure/model_train/images/recovery_ch.jpg
new file mode 100644
index 0000000..df5a506
Binary files /dev/null and b/docs/version2.x/ppstructure/model_train/images/recovery_ch.jpg differ
diff --git a/docs/version2.x/ppstructure/model_train/images/table_ch_result1.jpg b/docs/version2.x/ppstructure/model_train/images/table_ch_result1.jpg
new file mode 100644
index 0000000..c75eee4
Binary files /dev/null and b/docs/version2.x/ppstructure/model_train/images/table_ch_result1.jpg differ
diff --git a/docs/version2.x/ppstructure/model_train/images/table_ch_result2.jpg b/docs/version2.x/ppstructure/model_train/images/table_ch_result2.jpg
new file mode 100644
index 0000000..802871a
Binary files /dev/null and b/docs/version2.x/ppstructure/model_train/images/table_ch_result2.jpg differ
diff --git a/docs/version2.x/ppstructure/model_train/images/table_ch_result3.jpg b/docs/version2.x/ppstructure/model_train/images/table_ch_result3.jpg
new file mode 100644
index 0000000..bdd92aa
Binary files /dev/null and b/docs/version2.x/ppstructure/model_train/images/table_ch_result3.jpg differ
diff --git a/docs/version2.x/ppstructure/model_train/images/tableocr_pipeline.jpg b/docs/version2.x/ppstructure/model_train/images/tableocr_pipeline.jpg
new file mode 100644
index 0000000..da86879
Binary files /dev/null and b/docs/version2.x/ppstructure/model_train/images/tableocr_pipeline.jpg differ
diff --git a/docs/version2.x/ppstructure/model_train/images/zh_val_42_ser.jpg b/docs/version2.x/ppstructure/model_train/images/zh_val_42_ser.jpg
new file mode 100644
index 0000000..7b05e84
Binary files /dev/null and b/docs/version2.x/ppstructure/model_train/images/zh_val_42_ser.jpg differ
diff --git a/docs/version2.x/ppstructure/model_train/recovery_to_doc.en.md b/docs/version2.x/ppstructure/model_train/recovery_to_doc.en.md
new file mode 100644
index 0000000..1621714
--- /dev/null
+++ b/docs/version2.x/ppstructure/model_train/recovery_to_doc.en.md
@@ -0,0 +1,184 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# Layout Recovery
+
+## 1. Introduction
+
+The layout recovery module is used to restore the image or pdf to an
+editable Word file consistent with the original image layout.
+
+Two layout recovery methods are provided, you can choose by PDF format:
+
+- **Standard PDF parse(the input is standard PDF)**: Python based PDF to word library [pdf2docx](https://github.com/dothinking/pdf2docx) is optimized, the method extracts data from PDF with PyMuPDF, then parse layout with rule, finally, generate docx with python-docx.
+
+- **Image format PDF parse(the input can be standard PDF or image format PDF)**: Layout recovery combines [layout analysis](./train_layout.en.md)、[table recognition](./train_table.en.md) to better recover images, tables, titles, etc. supports input files in PDF and document image formats in Chinese and English.
+
+The input formats and application scenarios of the two methods are as follows:
+
+|  method   | input formats |                      application scenarios/problem                       |
+| :-----: | :----------: | :----------------------------------------------------------: |
+| Standard PDF parse |     pdf      | Advantages: Better recovery for non-paper documents, each page remains on the same page after restoration<br>Disadvantages: English characters in some Chinese documents are garbled, some contents are still beyond the current page, the whole page content is restored to the table format, and the recovery effect of some pictures is not good |
+| Image format PDF parse( |  pdf、picture   | Advantages: More suitable for paper document content recovery,  OCR recognition effect is more good<br>Disadvantages: Currently, the recovery is based on rules, the effect of content typesetting (spacing, fonts, etc.) need to be further improved, and the effect of layout recovery depends on layout analysis |
+
+The following figure shows the effect of restoring the layout of documents by using PDF parse:
+
+![img](./images/195319840-68fc60ec-ea66-4095-b734-0ec115860341.png)
+
+The following figures show the effect of restoring the layout of English and Chinese documents by using OCR technique:
+
+![img](./images/recovery-20240708091126891.jpg)
+
+![img](./images/recovery_ch.jpg)
+
+## 2. Install
+
+### 2.1 Install PaddlePaddle
+
+```bash linenums="1"
+python3 -m pip install --upgrade pip
+
+# If you have cuda9 or cuda10 installed on your machine, please run the following command to install
+python3 -m pip install "paddlepaddle-gpu" -i https://mirror.baidu.com/pypi/simple
+
+# CPU installation
+python3 -m pip install "paddlepaddle" -i https://mirror.baidu.com/pypi/simple
+````
+
+For more requirements, please refer to the instructions in [Installation Documentation](https://www.paddlepaddle.org.cn/en/install/quick?docurl=/documentation/docs/en/install/pip/macos-pip_en.html).
+
+### 2.2 Install PaddleOCR
+
+- **(1) Download source code**
+
+```bash linenums="1"
+[Recommended] git clone https://github.com/PaddlePaddle/PaddleOCR
+
+# If the pull cannot be successful due to network problems, you can also choose to use the hosting on the code cloud:
+git clone https://gitee.com/paddlepaddle/PaddleOCR
+
+# Note: Code cloud hosting code may not be able to synchronize the update of this github project in real time, there is a delay of 3 to 5 days, please use the recommended method first.
+````
+
+- **(2) Install recovery `requirements`**
+
+The layout restoration is exported as docx files, so python-docx API need to be installed, and PyMuPDF api([requires Python >= 3.7](https://pypi.org/project/PyMuPDF/)) need to be installed to process the input files in pdf format.
+
+Install all the libraries by running the following command:
+
+```bash linenums="1"
+python3 -m pip install -r ppstructure/recovery/requirements.txt
+````
+
+ And if using pdf parse method, we need to install pdf2docx api.
+
+```bash linenums="1"
+wget https://paddleocr.bj.bcebos.com/whl/pdf2docx-0.0.0-py3-none-any.whl
+pip3 install pdf2docx-0.0.0-py3-none-any.whl
+```
+
+## 3. Quick Start using standard PDF parse
+
+`use_pdf2docx_api` use PDF parse for layout recovery, The whl package is also provided  for quick use, follow the above code, for more information please refer to [quickstart](../quick_start.en.md) for details.
+
+```bash linenums="1"
+# install paddleocr
+pip3 install "paddleocr>=2.6"
+paddleocr --image_dir=ppstructure/docs/recovery/UnrealText.pdf --type=structure --recovery=true --use_pdf2docx_api=true
+```
+
+Command line:
+
+```bash linenums="1"
+python3 predict_system.py \
+    --image_dir=ppstructure/docs/recovery/UnrealText.pdf \
+    --recovery=True \
+    --use_pdf2docx_api=True \
+    --output=../output/
+```
+
+## 4. Quick Start using image format PDF parse
+
+Through layout analysis, we divided the image/PDF documents into regions, located the key regions, such as text, table, picture, etc., and recorded the location, category, and regional pixel value information of each region. Different regions are processed separately, where:
+
+- OCR detection and recognition is performed in the text area, and the coordinates of the OCR detection box and the text content information are added on the basis of the previous information
+
+- The table area identifies tables and records html and text information of tables
+- Save the image directly
+
+We can restore the test picture through the layout information, OCR detection and recognition structure, table information, and saved pictures.
+
+The whl package is also provided for quick use, follow the above code, for more information please refer to [quickstart](../quick_start.en.md) for details.
+
+```bash linenums="1"
+paddleocr --image_dir=ppstructure/docs/table/1.png --type=structure --recovery=true --lang='en'
+```
+
+### 4.1 Download models
+
+If input is English document, download English models:
+
+```bash linenums="1"
+cd PaddleOCR/ppstructure
+
+# download model
+mkdir inference && cd inference
+# Download the detection model of the ultra-lightweight English PP-OCRv3 model and unzip it
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar && tar xf en_PP-OCRv3_det_infer.tar
+# Download the recognition model of the ultra-lightweight English PP-OCRv3 model and unzip it
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv3_mobile_rec_infer.tar && tar xf en_PP-OCRv3_mobile_rec_infer.tar
+# Download the ultra-lightweight English table inch model and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/en_ppstructure_mobile_v2.0_SLANet_infer.tar
+tar xf en_ppstructure_mobile_v2.0_SLANet_infer.tar
+# Download the layout model of publaynet dataset and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar
+tar xf picodet_lcnet_x1_0_fgd_layout_infer.tar
+cd ..
+```
+
+If input is Chinese document，download Chinese models:
+[Chinese and English ultra-lightweight PP-OCRv3 model](../../ppocr/model_list.md)
+
+### 4.2 Layout recovery
+
+```bash linenums="1"
+python3 predict_system.py \
+    --image_dir=./docs/table/1.png \
+    --det_model_dir=inference/en_PP-OCRv3_det_infer \
+    --rec_model_dir=inference/en_PP-OCRv3_mobile_rec_infer \
+    --rec_char_dict_path=../ppocr/utils/en_dict.txt \
+    --table_model_dir=inference/en_ppstructure_mobile_v2.0_SLANet_infer \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt \
+    --layout_model_dir=inference/picodet_lcnet_x1_0_fgd_layout_infer \
+    --layout_dict_path=../ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt \
+    --vis_font_path=../doc/fonts/simfang.ttf \
+    --recovery=True \
+    --output=../output/
+```
+
+After running, the docx of each picture will be saved in the directory specified by the output field
+
+Field：
+
+- image_dir：test file， can be picture, picture directory, pdf file, pdf file directory
+- det_model_dir：OCR detection model path
+- rec_model_dir：OCR recognition model path
+- rec_char_dict_path：OCR recognition dict path. If the Chinese model is used, change to "../ppocr/utils/ppocr_keys_v1.txt". And if you trained the model on your own dataset, change to the trained dictionary
+- table_model_dir：table recognition model path
+- table_char_dict_path：table recognition dict path. If the Chinese model is used, no need to change
+- layout_model_dir：layout analysis model path
+- layout_dict_path：layout analysis dict path. If the Chinese model is used, change to "../ppocr/utils/dict/layout_dict/layout_cdla_dict.txt"
+- recovery：whether to enable layout of recovery, default False
+- output：save the recovery result path
+
+## 5. More
+
+For training, evaluation and inference tutorial for text detection models, please refer to [text detection doc](../../ppocr/model_train/detection.en.md).
+
+For training, evaluation and inference tutorial for text recognition models, please refer to [text recognition doc](../../ppocr/model_train/recognition.en.md).
+
+For training, evaluation and inference tutorial for layout analysis models, please refer to [layout analysis doc](./train_layout.en.md)
+
+For training, evaluation and inference tutorial for table recognition models, please refer to [table recognition doc](./train_table.en.md)
diff --git a/docs/version2.x/ppstructure/model_train/recovery_to_doc.md b/docs/version2.x/ppstructure/model_train/recovery_to_doc.md
new file mode 100644
index 0000000..b2bbd24
--- /dev/null
+++ b/docs/version2.x/ppstructure/model_train/recovery_to_doc.md
@@ -0,0 +1,191 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# 版面恢复
+
+## 1. 简介
+
+版面恢复就是将输入的图片、pdf内容仍然像原文档那样排列着，段落不变、顺序不变的输出到word文档中等。
+
+提供了2种版面恢复方法，可根据输入PDF的格式进行选择：
+
+- **标准PDF解析(输入须为标准PDF)**：基于Python的pdf转word库[pdf2docx](https://github.com/dothinking/pdf2docx)进行优化，该方法通过PyMuPDF获取页面元素，然后利用规则解析章节、段落、表格等布局及样式，最后通过python-docx将解析的内容元素重建到word文档中。
+- **图片格式PDF解析(输入可为标准PDF或图片格式PDF)**：结合[版面分析](./train_layout.md)、[表格识别](./train_table.md)技术，从而更好地恢复图片、表格、标题等内容，支持中、英文pdf文档、文档图片格式的输入文件。
+
+2种方法输入格式、适用场景如下：
+
+|      方法       | 支持输入文件 |                      适用场景/存在问题                       |
+| :-------------: | :----------: | :----------------------------------------------------------: |
+|   标准PDF解析   |     pdf      | 优点：非论文文档恢复效果更优、每一页内容恢复后仍在同一页<br>缺点：有些中文文档中的英文乱码、仍存在内容超出当前页面的情况、整页内容恢复为表格格式、部分图片恢复效果不佳 |
+| 图片格式PDF解析 |  pdf、图片   | 优点：更适合论文文档正文内容的恢复、中英文文档OCR识别效果好<br>缺点：目前内容恢复基于规则，内容排版效果(间距、字体等)待进一步提升、版面恢复效果依赖于版面分析效果 |
+
+下图展示了通过PDF解析版面恢复效果：
+
+![img](./images/195319840-68fc60ec-ea66-4095-b734-0ec115860341.png)
+
+下图分别展示了通过OCR技术，英文文档和中文文档版面恢复的效果：
+
+![img](./images/recovery-20240708091126891.jpg)
+
+![img](./images/recovery_ch.jpg)
+
+## 2. 安装
+
+### 2.1 安装PaddlePaddle
+
+```bash linenums="1"
+python3 -m pip install --upgrade pip
+
+# 您的机器安装的是CUDA9或CUDA10，请运行以下命令安装
+python3 -m pip install "paddlepaddle-gpu" -i https://mirror.baidu.com/pypi/simple
+
+# 您的机器是CPU，请运行以下命令安装
+python3 -m pip install "paddlepaddle" -i https://mirror.baidu.com/pypi/simple
+```
+
+更多需求，请参照[安装文档](https://www.paddlepaddle.org.cn/install/quick)中的说明进行操作。
+
+### 2.2 安装PaddleOCR
+
+- **（1）下载版面恢复源码**
+
+```bash linenums="1"
+【推荐】git clone https://github.com/PaddlePaddle/PaddleOCR
+
+# 如果因为网络问题无法pull成功，也可选择使用码云上的托管：
+git clone https://gitee.com/paddlepaddle/PaddleOCR
+
+# 注：码云托管代码可能无法实时同步本github项目更新，存在3~5天延时，请优先使用推荐方式。
+```
+
+- **（2）安装recovery的`requirements`**
+
+版面恢复导出为docx文件，所以需要安装Python处理word文档的python-docx API，同时处理pdf格式的输入文件，需要安装PyMuPDF API([要求Python >= 3.7](https://pypi.org/project/PyMuPDF/))。
+
+通过如下命令安装全部库：
+
+```bash linenums="1"
+python3 -m pip install -r ppstructure/recovery/requirements.txt
+```
+
+使用pdf2docx库解析的方式恢复文档需要安装优化的pdf2docx。
+
+```bash linenums="1"
+wget https://paddleocr.bj.bcebos.com/whl/pdf2docx-0.0.0-py3-none-any.whl
+pip3 install pdf2docx-0.0.0-py3-none-any.whl
+```
+
+## 3.使用标准PDF解析进行版面恢复
+
+`use_pdf2docx_api`表示使用PDF解析的方式进行版面恢复，通过whl包的形式方便快速使用，代码如下，更多信息详见 [quickstart](../quick_start.md)。
+
+```bash linenums="1"
+# 安装 paddleocr，推荐使用2.6版本
+pip3 install "paddleocr>=2.6"
+paddleocr --image_dir=ppstructure/docs/recovery/UnrealText.pdf --type=structure --recovery=true --use_pdf2docx_api=true
+```
+
+通过命令行的方式：
+
+```bash linenums="1"
+python3 predict_system.py \
+    --image_dir=ppstructure/docs/recovery/UnrealText.pdf \
+    --recovery=True \
+    --use_pdf2docx_api=True \
+    --output=../output/
+```
+
+## 4.使用图片格式PDF解析进行版面恢复
+
+我们通过版面分析对图片/pdf形式的文档进行区域划分，定位其中的关键区域，如文字、表格、图片等，记录每个区域的位置、类别、区域像素值信息。对不同的区域分别处理，其中：
+
+- 文字区域直接进行OCR检测和识别，在之前信息基础上增加OCR检测框坐标和文本内容信息
+- 表格区域进行表格识别，记录表格html和文字信息
+- 图片直接保存
+
+我们通过版面信息、OCR检测和识别结构、表格信息、保存的图片，对测试图片进行恢复即可。
+
+提供如下代码实现版面恢复，也提供了whl包的形式方便快速使用，代码如下，更多信息详见 [quickstart](../quick_start.md)。
+
+```bash linenums="1"
+# 安装 paddleocr，推荐使用2.6版本
+pip3 install "paddleocr>=2.6"
+# 中文测试图
+paddleocr --image_dir=ppstructure/docs/table/1.png --type=structure --recovery=true
+# 英文测试图
+paddleocr --image_dir=ppstructure/docs/table/1.png --type=structure --recovery=true --lang='en'
+# pdf测试文件
+paddleocr --image_dir=ppstructure/docs/recovery/UnrealText.pdf --type=structure --recovery=true --lang='en'
+```
+
+### 4.1 下载模型
+
+如果输入为英文文档类型，下载OCR检测和识别、版面分析、表格识别的英文模型
+
+```bash linenums="1"
+cd PaddleOCR/ppstructure
+
+# 下载模型
+mkdir inference && cd inference
+# 下载英文超轻量PP-OCRv3检测模型并解压
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar && tar xf en_PP-OCRv3_det_infer.tar
+# 下载英文超轻量PP-OCRv3识别模型并解压
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv3_mobile_rec_infer.tar && tar xf en_PP-OCRv3_mobile_rec_infer.tar
+# 下载英文表格识别模型并解压
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/en_ppstructure_mobile_v2.0_SLANet_infer.tar
+tar xf en_ppstructure_mobile_v2.0_SLANet_infer.tar
+# 下载英文版面分析模型
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar
+tar xf picodet_lcnet_x1_0_fgd_layout_infer.tar
+cd ..
+```
+
+如果输入为中文文档类型，在下述链接中下载中文模型即可：
+
+[PP-OCRv3中英文超轻量文本检测和识别模型](../../ppocr/model_list.md)
+
+### 4.2 版面恢复
+
+使用下载的模型恢复给定文档的版面，以英文模型为例，执行如下命令：
+
+```bash linenums="1"
+python3 predict_system.py \
+    --image_dir=./docs/table/1.png \
+    --det_model_dir=inference/en_PP-OCRv3_det_infer \
+    --rec_model_dir=inference/en_PP-OCRv3_mobile_rec_infer \
+    --rec_char_dict_path=../ppocr/utils/en_dict.txt \
+    --table_model_dir=inference/en_ppstructure_mobile_v2.0_SLANet_infer \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt \
+    --layout_model_dir=inference/picodet_lcnet_x1_0_fgd_layout_infer \
+    --layout_dict_path=../ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt \
+    --vis_font_path=../doc/fonts/simfang.ttf \
+    --recovery=True \
+    --output=../output/
+```
+
+运行完成后，恢复版面的docx文档会保存到`output`字段指定的目录下
+
+字段含义：
+
+- image_dir：测试文件，可以是图片、图片目录、pdf文件、pdf文件目录
+- det_model_dir：OCR检测模型路径
+- rec_model_dir：OCR识别模型路径
+- rec_char_dict_path：OCR识别字典，如果更换为中文模型，需要更改为"../ppocr/utils/ppocr_keys_v1.txt"，如果您在自己的数据集上训练的模型，则更改为训练的字典的文件
+- table_model_dir：表格识别模型路径
+- table_char_dict_path：表格识别字典，如果更换为中文模型，不需要更换字典
+- layout_model_dir：版面分析模型路径
+- layout_dict_path：版面分析字典，如果更换为中文模型，需要更改为"../ppocr/utils/dict/layout_dict/layout_cdla_dict.txt"
+- recovery：是否进行版面恢复，默认False
+- output：版面恢复结果保存路径
+
+## 5. 更多
+
+关于OCR检测模型的训练评估与推理，请参考：[文本检测教程](../../ppocr/model_train/detection.md)
+
+关于OCR识别模型的训练评估与推理，请参考：[文本识别教程](../../ppocr/model_train/recognition.md)
+
+关于版面分析模型的训练评估与推理，请参考：[版面分析教程](./train_layout.md)
+
+关于表格识别模型的训练评估与推理，请参考：[表格识别教程](./train_table.md)
diff --git a/docs/version2.x/ppstructure/model_train/train_kie.en.md b/docs/version2.x/ppstructure/model_train/train_kie.en.md
new file mode 100644
index 0000000..d265ffd
--- /dev/null
+++ b/docs/version2.x/ppstructure/model_train/train_kie.en.md
@@ -0,0 +1,245 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# Key Information Extraction (KIE)
+
+## 1. Introduction
+
+Key information extraction (KIE) refers to extracting key information from text or images. As downstream task of OCR, the key information extraction task of document image has many practical application scenarios, such as form recognition, ticket information extraction, ID card information extraction, etc.
+
+PP-Structure conducts research based on the LayoutXLM multi-modal, and proposes the VI-LayoutXLM, which gets rid of visual features when finetuning the downstream tasks. An textline sorting method is also utilized to fit in reading order. What's more, UDML knowledge distillation is used for higher accuracy. Finally, the accuracy and inference speed of VI-LayoutXLM surpass those of LayoutXLM.
+
+The main features of the key information extraction module in PP-Structure are as follows.
+
+- Integrate multi-modal methods such as [LayoutXLM](https://arxiv.org/pdf/2104.08836.pdf), VI-LayoutXLM, and PP-OCR inference engine.
+- Supports Semantic Entity Recognition (SER) and Relation Extraction (RE) tasks based on multimodal methods. Based on the SER task, the text recognition and classification in the image can be completed; based on the RE task, the relationship extraction of the text content in the image can be completed, such as judging the problem pair (pair).
+- Supports custom training for SER tasks and RE tasks.
+- Supports end-to-end system prediction and evaluation of OCR+SER.
+- Supports end-to-end system prediction of OCR+SER+RE.
+- Support SER model export and inference using PaddleInference.
+
+## 2. Performance
+
+We evaluate the methods on the Chinese dataset of [XFUND](https://github.com/doc-analysis/XFUND), and the performance is as follows
+
+|Model | Backbone | Task | Config file | Hmean | Inference time (ms) | Download link|
+| --- | --- |  --- | --- | --- | --- | --- |
+|VI-LayoutXLM| VI-LayoutXLM-base | SER | [ser_vi_layoutxlm_xfund_zh_udml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh_udml.yml)|**93.19%**| 15.49|[trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar)|
+|LayoutXLM| LayoutXLM-base | SER | [ser_layoutxlm_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/ser_layoutxlm_xfund_zh.yml)|90.38%| 19.49 | [trained model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar)|
+|VI-LayoutXLM| VI-LayoutXLM-base | RE | [re_vi_layoutxlm_xfund_zh_udml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh_udml.yml)|**83.92%**| 15.49|[trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar)|
+|LayoutXLM| LayoutXLM-base | RE | [re_layoutxlm_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/re_layoutxlm_xfund_zh.yml)|74.83%| 19.49|[trained model](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar)|
+
+- Note：Inference environment：V100 GPU + cuda10.2 + cudnn8.1.1 + TensorRT 7.2.3.4，tested using fp16.
+
+For more KIE models in PaddleOCR, please refer to [KIE model zoo](../../algorithm/overview.en.md).
+
+## 3. Visualization
+
+There are two main solutions to the key information extraction task based on VI-LayoutXLM series model.
+
+(1) Text detection + text recognition + semantic entity recognition (SER)
+
+(2) Text detection + text recognition + semantic entity recognition (SER) + relationship extraction (RE)
+
+The following images are demo results of the SER and RE models. For more detailed introduction to the above solutions, please refer to [KIE Guide](../blog/how_to_do_kie.en.md).
+
+### 3.1 SER
+
+Demo results for SER task are as follows.
+
+![img](./images/185539141-68e71c75-5cf7-4529-b2ca-219d29fa5f68.jpg)
+
+![img](./images/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808.jpg)
+
+![img](./images/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a.png)
+
+![img](./images/185539735-37b5c2ef-629d-43fe-9abb-44bb717ef7ee.jpg)
+
+**Note:** test pictures are from [xfund dataset](https://github.com/doc-analysis/XFUND), [invoice dataset](https://aistudio.baidu.com/aistudio/datasetdetail/165561) and a composite ID card dataset.
+
+Boxes of different colors in the image represent different categories.
+
+The invoice and application form images have three categories: `request`, `answer` and `header`. The `question` and `answer` can be used to extract the relationship.
+
+For the ID card image, the model can directly identify the key information such as `name`, `gender`, `nationality`, so that the subsequent relationship extraction process is not required, and the key information extraction task can be completed using only one model.
+
+### 3.2 RE
+
+Demo results for RE task are as follows.
+
+![img](./images/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb.jpg)
+
+![img](./images/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f.jpg)
+
+![img](./images/185540291-f64e5daf-6d42-4e7c-bbbb-471e3fac4fcc.png)
+
+Red boxes are questions, blue boxes are answers. The green lines means the two connected objects are a pair.
+
+## 4. Usage
+
+### 4.1 Prepare for the environment
+
+Use the following command to install KIE dependencies.
+
+```bash linenums="1"
+git clone https://github.com/PaddlePaddle/PaddleOCR.git
+cd PaddleOCR
+pip install -r requirements.txt
+pip install -r ppstructure/kie/requirements.txt
+# 安装PaddleOCR引擎用于预测
+pip install "paddleocr<3.0"
+```
+
+NOTE: For KIE tasks, it is necessary to downgrade the Paddle framework version (Paddle<2.6) and the PaddleNLP version (PaddleNLP<2.6).
+
+The visualized results of SER are saved in the `./output` folder by default. Examples of results are as follows.
+
+![img](./images/zh_val_42_ser.jpg)
+
+### 4.2 Quick start
+
+Here we use XFUND dataset to quickly experience the SER model and RE model.
+
+#### 4.2.1 Prepare for the dataset
+
+```bash linenums="1"
+mkdir train_data
+cd train_data
+# download and uncompress the dataset
+wget https://paddleocr.bj.bcebos.com/ppstructure/dataset/XFUND.tar && tar -xf XFUND.tar
+cd ..
+```
+
+#### 4.2.2 Predict images using the trained model
+
+Use the following command to download the models.
+
+```bash linenums="1"
+mkdir pretrained_model
+cd pretrained_model
+# download and uncompress the SER trained model
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar && tar -xf ser_vi_layoutxlm_xfund_pretrained.tar
+
+# download and uncompress the RE trained model
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar && tar -xf re_vi_layoutxlm_xfund_pretrained.tar
+```
+
+If you want to use OCR engine to obtain end-to-end prediction results, you can use the following command to predict.
+
+```bash linenums="1"
+# just predict using SER trained model
+python3 tools/infer_kie_token_ser.py \
+  -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrained_model/ser_vi_layoutxlm_xfund_pretrained/best_accuracy \
+  Global.infer_img=./ppstructure/docs/kie/input/zh_val_42.jpg
+
+# predict using SER and RE trained model at the same time
+python3 ./tools/infer_kie_token_ser_re.py \
+  -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrained_model/re_vi_layoutxlm_xfund_pretrained/best_accuracy \
+  Global.infer_img=./train_data/XFUND/zh_val/image/zh_val_42.jpg \
+  -c_ser configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o_ser Architecture.Backbone.checkpoints=./pretrained_model/ser_vi_layoutxlm_xfund_pretrained/best_accuracy
+```
+
+The visual result images and the predicted text file will be saved in the `Global.save_res_path` directory.
+
+If you want to use a custom ocr model, you can set it through the following fields
+
+- `Global.kie_det_model_dir`: the detection inference model path
+- `Global.kie_rec_model_dir`: the recognition inference model path
+
+If you want to load the text detection and recognition results collected before, you can use the following command to predict.
+
+```bash linenums="1"
+# just predict using SER trained model
+python3 tools/infer_kie_token_ser.py \
+  -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrained_model/ser_vi_layoutxlm_xfund_pretrained/best_accuracy \
+  Global.infer_img=./train_data/XFUND/zh_val/val.json \
+  Global.infer_mode=False
+
+# predict using SER and RE trained model at the same time
+python3 ./tools/infer_kie_token_ser_re.py \
+  -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrained_model/re_vi_layoutxlm_xfund_pretrained/best_accuracy \
+  Global.infer_img=./train_data/XFUND/zh_val/val.json \
+  Global.infer_mode=False \
+  -c_ser configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o_ser Architecture.Backbone.checkpoints=./pretrained_model/ser_vi_layoutxlm_xfund_pretrained/best_accuracy
+```
+
+#### 4.2.3 Inference using PaddleInference
+
+Firstly, download the inference SER inference model.
+
+```bash linenums="1"
+mkdir inference
+cd inference
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_infer.tar && tar -xf ser_vi_layoutxlm_xfund_infer.tar
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_infer.tar && tar -xf re_vi_layoutxlm_xfund_infer.tar
+cd ..
+```
+
+- SER
+
+Use the following command for inference.
+
+```bash linenums="1"
+cd ppstructure
+python3 kie/predict_kie_token_ser.py \
+  --kie_algorithm=LayoutXLM \
+  --ser_model_dir=../inference/ser_vi_layoutxlm_xfund_infer \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx"
+```
+
+The visual results and text file will be saved in directory `output`.
+
+- RE
+
+Use the following command for inference.
+
+```bash linenums="1"
+cd ppstructure
+python3 kie/predict_kie_token_ser_re.py \
+  --kie_algorithm=LayoutXLM \
+  --re_model_dir=../inference/re_vi_layoutxlm_xfund_infer \
+  --ser_model_dir=../inference/ser_vi_layoutxlm_xfund_infer \
+  --use_visual_backbone=False \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx"
+```
+
+The visual results and text file will be saved in directory `output`.
+
+If you want to use a custom ocr model, you can set it through the following fields
+
+- `--det_model_dir`: the detection inference model path
+- `--rec_model_dir`: the recognition inference model path
+
+### 4.3 More
+
+For training, evaluation and inference tutorial for KIE models, please refer to [KIE doc](../model_train/train_kie.en.md).
+
+For training, evaluation and inference tutorial for text detection models, please refer to [text detection doc](../../ppocr/model_train/detection.en.md).
+
+For training, evaluation and inference tutorial for text recognition models, please refer to [text recognition doc](../../ppocr/model_train/recognition.en.md).
+
+To complete the key information extraction task in your own scenario from data preparation to model selection, please refer to: [Guide to End-to-end KIE](../blog/how_to_do_kie.en.md)。
+
+## 5. Reference
+
+- LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding, <https://arxiv.org/pdf/2104.08836.pdf>
+- microsoft/unilm/layoutxlm, <https://github.com/microsoft/unilm/tree/master/layoutxlm>
+- XFUND dataset, <https://github.com/doc-analysis/XFUND>
+
+## 6. License
+
+The content of this project itself is licensed under the [Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)](https://creativecommons.org/licenses/by-nc-sa/4.0/)
diff --git a/docs/version2.x/ppstructure/model_train/train_kie.md b/docs/version2.x/ppstructure/model_train/train_kie.md
new file mode 100644
index 0000000..d84e610
--- /dev/null
+++ b/docs/version2.x/ppstructure/model_train/train_kie.md
@@ -0,0 +1,240 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# 关键信息抽取
+
+## 1. 简介
+
+关键信息抽取 (Key Information Extraction, KIE)指的是是从文本或者图像中，抽取出关键的信息。针对文档图像的关键信息抽取任务作为OCR的下游任务，存在非常多的实际应用场景，如表单识别、车票信息抽取、身份证信息抽取等。
+
+PP-Structure 基于 LayoutXLM 文档多模态系列方法进行研究与优化，设计了视觉特征无关的多模态模型结构VI-LayoutXLM，同时引入符合阅读顺序的文本行排序方法以及UDML联合互学习蒸馏方法，最终在精度与速度均超越LayoutXLM。
+
+PP-Structure中关键信息抽取模块的主要特性如下：
+
+- 集成[LayoutXLM](https://arxiv.org/pdf/2104.08836.pdf)、VI-LayoutXLM等多模态模型以及PP-OCR预测引擎。
+- 支持基于多模态方法的语义实体识别 (Semantic Entity Recognition, SER) 以及关系抽取 (Relation Extraction, RE) 任务。基于 SER 任务，可以完成对图像中的文本识别与分类；基于 RE 任务，可以完成对图象中的文本内容的关系提取，如判断问题对(pair)。
+- 支持SER任务和RE任务的自定义训练。
+- 支持OCR+SER的端到端系统预测与评估。
+- 支持OCR+SER+RE的端到端系统预测。
+- 支持SER模型的动转静导出与基于PaddleInfernece的模型推理。
+
+## 2. 精度与性能
+
+我们在 [XFUND](https://github.com/doc-analysis/XFUND) 的中文数据集上对算法进行了评估，SER与RE上的任务性能如下
+
+| 模型         | 骨干网络          | 任务 | 配置文件                                                                                                | hmean      | 预测耗时(ms) | 下载链接                                                                                                          |
+| ------------ | ----------------- | ---- | ------- | ---------- | ------ | ------ |
+| VI-LayoutXLM | VI-LayoutXLM-base | SER  | [ser_vi_layoutxlm_xfund_zh_udml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh_udml.yml) | **93.19%** | 15.49        | [训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar) |
+| LayoutXLM    | LayoutXLM-base    | SER  | [ser_layoutxlm_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/ser_layoutxlm_xfund_zh.yml)              | 90.38%     | 19.49        | [训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar)                                    |
+| VI-LayoutXLM | VI-LayoutXLM-base | RE   | [re_vi_layoutxlm_xfund_zh_udml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh_udml.yml)   | **83.92%** | 15.49        | [训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar)  |
+| LayoutXLM    | LayoutXLM-base    | RE   | [re_layoutxlm_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/re_layoutxlm_xfund_zh.yml)                | 74.83%     | 19.49        | [训练模型](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar)                                     |
+
+- 注：预测耗时测试条件：V100 GPU + cuda10.2 + cudnn8.1.1 + TensorRT 7.2.3.4，使用FP16进行测试。
+
+更多关于PaddleOCR中关键信息抽取模型的介绍，请参考[关键信息抽取模型库](../../algorithm/overview.md)。
+
+## 3. 效果演示
+
+基于多模态模型的关键信息抽取任务有2种主要的解决方案。
+
+（1）文本检测 + 文本识别 + 语义实体识别(SER)
+（2）文本检测 + 文本识别 + 语义实体识别(SER) + 关系抽取(RE)
+
+下面给出SER与RE任务的示例效果，关于上述解决方案的详细介绍，请参考[关键信息抽取全流程指南](../blog/how_to_do_kie.md)。
+
+### 3.1 SER
+
+对于SER任务，效果如下所示。
+
+![img](./images/185539141-68e71c75-5cf7-4529-b2ca-219d29fa5f68.jpg)
+
+![img](./images/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808.jpg)
+
+![img](./images/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a.png)
+
+![img](./images/185539735-37b5c2ef-629d-43fe-9abb-44bb717ef7ee.jpg)
+
+**注意：** 测试图片来源于[XFUND数据集](https://github.com/doc-analysis/XFUND)、[发票数据集](https://aistudio.baidu.com/aistudio/datasetdetail/165561)以及合成的身份证数据集。
+
+图中不同颜色的框表示不同的类别。
+
+图中的发票以及申请表图像，有`QUESTION`, `ANSWER`, `HEADER` 3种类别，识别的`QUESTION`, `ANSWER`可以用于后续的问题与答案的关系抽取。
+
+图中的身份证图像，则直接识别出其中的`姓名`、`性别`、`民族`等关键信息，这样就无需后续的关系抽取过程，一个模型即可完成关键信息抽取。
+
+### 3.2 RE
+
+对于RE任务，效果如下所示。
+
+![img](./images/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb.jpg)
+
+![img](./images/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f.jpg)
+
+![img](./images/185540291-f64e5daf-6d42-4e7c-bbbb-471e3fac4fcc.png)
+
+红色框是问题，蓝色框是答案。绿色线条表示连接的两端为一个key-value的pair。
+
+## 4. 使用
+
+### 4.1 准备环境
+
+使用下面的命令安装运行SER与RE关键信息抽取的依赖。
+
+```bash linenums="1"
+git clone https://github.com/PaddlePaddle/PaddleOCR.git
+cd PaddleOCR
+pip install -r requirements.txt
+pip install -r ppstructure/kie/requirements.txt
+# 安装PaddleOCR引擎用于预测
+pip install "paddleocr<3.0"
+```
+
+NOTE: 对于KIE任务需要降低Paddle框架版本（Paddle<2.6），和PaddleNLP版本（PaddleNLP<2.6）。
+
+### 4.2 快速开始
+
+下面XFUND数据集，快速体验SER模型与RE模型。
+
+#### 4.2.1 准备数据
+
+```bash linenums="1"
+mkdir train_data
+cd train_data
+# 下载与解压数据
+wget https://paddleocr.bj.bcebos.com/ppstructure/dataset/XFUND.tar && tar -xf XFUND.tar
+cd ..
+```
+
+#### 4.2.2 基于动态图的预测
+
+首先下载模型。
+
+```bash linenums="1"
+mkdir pretrained_model
+cd pretrained_model
+# 下载并解压SER预训练模型
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar && tar -xf ser_vi_layoutxlm_xfund_pretrained.tar
+
+# 下载并解压RE预训练模型
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar && tar -xf re_vi_layoutxlm_xfund_pretrained.tar
+```
+
+如果希望使用OCR引擎，获取端到端的预测结果，可以使用下面的命令进行预测。
+
+```bash linenums="1"
+# 仅预测SER模型
+python3 tools/infer_kie_token_ser.py \
+  -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrained_model/ser_vi_layoutxlm_xfund_pretrained/best_accuracy \
+  Global.infer_img=./ppstructure/docs/kie/input/zh_val_42.jpg
+
+# SER + RE模型串联
+python3 ./tools/infer_kie_token_ser_re.py \
+  -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrained_model/re_vi_layoutxlm_xfund_pretrained/best_accuracy \
+  Global.infer_img=./train_data/XFUND/zh_val/image/zh_val_42.jpg \
+  -c_ser configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o_ser Architecture.Backbone.checkpoints=./pretrained_model/ser_vi_layoutxlm_xfund_pretrained/best_accuracy
+```
+
+`Global.save_res_path`目录中会保存可视化的结果图像以及预测的文本文件。
+
+如果想使用自定义OCR模型，可通过如下字段进行设置
+
+- `Global.kie_det_model_dir`: 设置检测inference模型地址
+- `Global.kie_rec_model_dir`: 设置识别inference模型地址
+
+如果希望加载标注好的文本检测与识别结果，仅预测可以使用下面的命令进行预测。
+
+```bash linenums="1"
+# 仅预测SER模型
+python3 tools/infer_kie_token_ser.py \
+  -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrained_model/ser_vi_layoutxlm_xfund_pretrained/best_accuracy \
+  Global.infer_img=./train_data/XFUND/zh_val/val.json \
+  Global.infer_mode=False
+
+# SER + RE模型串联
+python3 ./tools/infer_kie_token_ser_re.py \
+  -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrained_model/re_vi_layoutxlm_xfund_pretrained/best_accuracy \
+  Global.infer_img=./train_data/XFUND/zh_val/val.json \
+  Global.infer_mode=False \
+  -c_ser configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o_ser Architecture.Backbone.checkpoints=./pretrained_model/ser_vi_layoutxlm_xfund_pretrained/best_accuracy
+```
+
+#### 4.2.3 基于PaddleInference的预测
+
+首先下载SER和RE的推理模型。
+
+```bash linenums="1"
+mkdir inference
+cd inference
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_infer.tar && tar -xf ser_vi_layoutxlm_xfund_infer.tar
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_infer.tar && tar -xf re_vi_layoutxlm_xfund_infer.tar
+cd ..
+```
+
+- SER
+
+执行下面的命令进行预测。
+
+```bash linenums="1"
+cd ppstructure
+python3 kie/predict_kie_token_ser.py \
+  --kie_algorithm=LayoutXLM \
+  --ser_model_dir=../inference/ser_vi_layoutxlm_xfund_infer \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx"
+```
+
+可视化结果保存在`output`目录下。
+
+- RE
+
+执行下面的命令进行预测。
+
+```bash linenums="1"
+cd ppstructure
+python3 kie/predict_kie_token_ser_re.py \
+  --kie_algorithm=LayoutXLM \
+  --re_model_dir=../inference/re_vi_layoutxlm_xfund_infer \
+  --ser_model_dir=../inference/ser_vi_layoutxlm_xfund_infer \
+  --use_visual_backbone=False \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx"
+```
+
+可视化结果保存在`output`目录下。
+
+如果想使用自定义OCR模型，可通过如下字段进行设置
+
+- `--det_model_dir`: 设置检测inference模型地址
+- `--rec_model_dir`: 设置识别inference模型地址
+
+### 4.3 更多
+
+关于KIE模型的训练评估与推理，请参考：[关键信息抽取教程](./train_kie.md)。
+
+关于文本检测模型的训练评估与推理，请参考：[文本检测教程](../../ppocr/model_train/detection.md)。
+
+关于文本识别模型的训练评估与推理，请参考：[文本识别教程](../../ppocr/model_train/recognition.md)。
+
+关于怎样在自己的场景中完成关键信息抽取任务，请参考：[关键信息抽取全流程指南](../blog/how_to_do_kie.md)。
+
+## 5. 参考链接
+
+- LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding, <https://arxiv.org/pdf/2104.08836.pdf>
+- microsoft/unilm/layoutxlm, <https://github.com/microsoft/unilm/tree/master/layoutxlm>
+- XFUND dataset, <https://github.com/doc-analysis/XFUND>
+
+## 6. License
+
+The content of this project itself is licensed under the [Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)](https://creativecommons.org/licenses/by-nc-sa/4.0/)
diff --git a/docs/version2.x/ppstructure/model_train/train_layout.en.md b/docs/version2.x/ppstructure/model_train/train_layout.en.md
new file mode 100644
index 0000000..0af0a5b
--- /dev/null
+++ b/docs/version2.x/ppstructure/model_train/train_layout.en.md
@@ -0,0 +1,444 @@
+---
+comments: true
+---
+
+# Layout analysis
+
+## 1. Introduction
+
+Layout analysis refers to the regional division of documents in the form of pictures and the positioning of key areas, such as text, title, table, picture, etc. The layout analysis algorithm is based on the lightweight model PP-picodet of [PaddleDetection]( https://github.com/PaddlePaddle/PaddleDetection ), including English layout analysis, Chinese layout analysis and table layout analysis models.  English layout analysis models can detect document layout elements such as text, title, table, figure, list. Chinese layout analysis models can detect document layout elements such as text, figure, figure caption, table, table caption, header, footer, reference, and equation. Table layout analysis models can detect table regions.
+
+![img](./images/layout.jpg)
+
+## 2. Quick start
+
+PP-Structure currently provides layout analysis models in Chinese, English and table documents. For the model link, see [models_list](../models_list.en.md). The whl package is also provided for quick use, see [quickstart](../quick_start.en.md) for details.
+
+## 3. Install
+
+### 3.1. Install PaddlePaddle
+
+- **（1) Install PaddlePaddle**
+
+```bash linenums="1"
+python3 -m pip install --upgrade pip
+
+# GPU Install
+python3 -m pip install "paddlepaddle-gpu>=2.3" -i https://mirror.baidu.com/pypi/simple
+
+# CPU Install
+python3 -m pip install "paddlepaddle>=2.3" -i https://mirror.baidu.com/pypi/simple
+```
+
+For more requirements, please refer to the instructions in the [Install file](https://www.paddlepaddle.org.cn/install/quick)。
+
+### 3.2. Install PaddleDetection
+
+- **（1）Download PaddleDetection Source code**
+
+```bash linenums="1"
+git clone https://github.com/PaddlePaddle/PaddleDetection.git
+```
+
+- **（2）Install third-party libraries**
+
+```bash linenums="1"
+cd PaddleDetection
+python3 -m pip install -r requirements.txt
+```
+
+## 4. Data preparation
+
+If you want to experience the prediction process directly, you can skip data preparation and download the pre-training model.
+
+### 4.1. English data set
+
+Download document analysis data set [PubLayNet](https://developer.ibm.com/exchanges/data/all/publaynet/)（Dataset 96G），contains 5 classes：`{0: "Text", 1: "Title", 2: "List", 3:"Table", 4:"Figure"}`
+
+```
+# Download data
+wget https://dax-cdn.cdn.appdomain.cloud/dax-publaynet/1.0.0/publaynet.tar.gz
+# Decompress data
+tar -xvf publaynet.tar.gz
+```
+
+Uncompressed **directory structure：**
+
+```
+|-publaynet
+  |- test
+     |- PMC1277013_00004.jpg
+     |- PMC1291385_00002.jpg
+     | ...
+  |- train.json
+  |- train
+     |- PMC1291385_00002.jpg
+     |- PMC1277013_00004.jpg
+     | ...
+  |- val.json
+  |- val
+     |- PMC538274_00004.jpg
+     |- PMC539300_00004.jpg
+     | ...
+```
+
+**data distribution：**
+
+| File or Folder | Description    | num     |
+| :------------- | :------------- | ------- |
+| `train/`       | Training set pictures     | 335,703 |
+| `val/`         | Verification set pictures     | 11,245  |
+| `test/`        | Test set pictures     | 11,405  |
+| `train.json`   | Training set annotation files | -       |
+| `val.json`     | Validation set dimension files | -       |
+
+**Data Annotation**
+
+The JSON file contains the annotations of all images, and the data is stored in a dictionary nested manner.Contains the following keys：
+
+- info，represents the dimension file info。
+
+- licenses，represents the dimension file licenses。
+
+- images，represents the list of image information in the annotation file，each element is the information of an image。The information of one of the images is as follows:
+
+  ```
+  {
+      'file_name': 'PMC4055390_00006.jpg',    # file_name
+      'height': 601,                      # image height
+      'width': 792,                       # image width
+      'id': 341427                        # image id
+  }
+  ```
+
+- annotations， represents the list of annotation information of the target object in the annotation file，each element is the annotation information of a target object。The following is the annotation information of one of the target objects:
+
+  ```
+  {
+
+      'segmentation':             # Segmentation annotation of objects
+      'area': 60518.099043117836, # Area of object
+      'iscrowd': 0,               # iscrowd
+      'image_id': 341427,         # image id
+      'bbox': [50.58, 490.86, 240.15, 252.16], # bbox [x1,y1,w,h]
+      'category_id': 1,           # category_id
+      'id': 3322348               # image id
+  }
+  ```
+
+### 4.2. More datasets
+
+We provide CDLA(Chinese layout analysis), TableBank(Table layout analysis)etc. data set download links，process to the JSON format of the above annotation file，that is, the training can be conducted in the same way。
+
+| dataset                                                      | 简介                                                         |
+| ------------------------------------------------------------ | ------------------------------------------------------------ |
+| [cTDaR2019_cTDaR](https://cndplab-founder.github.io/cTDaR2019/) | For form detection (TRACKA) and form identification (TRACKB).Image types include historical data sets (beginning with cTDaR_t0, such as CTDAR_T00872.jpg) and modern data sets (beginning with cTDaR_t1, CTDAR_T10482.jpg). |
+| [IIIT-AR-13K](http://cvit.iiit.ac.in/usodi/iiitar13k.php)    | Data sets constructed by manually annotating figures or pages from publicly available annual reports, containing 5 categories:table, figure, natural image, logo, and signature. |
+| [TableBank](https://github.com/doc-analysis/TableBank)       | For table detection and recognition of large datasets, including Word and Latex document formats |
+| [CDLA](https://github.com/buptlihang/CDLA)                   | Chinese document layout analysis data set, for Chinese literature (paper) scenarios, including 10 categories:Text, Title, Figure, Figure caption, Table, Table caption, Header, Footer, Reference, Equation |
+| [DocBank](https://github.com/doc-analysis/DocBank)           | Large-scale dataset (500K document pages) constructed using weakly supervised methods for document layout analysis, containing 12 categories:Author, Caption, Date, Equation, Figure, Footer, List, Paragraph, Reference, Section, Table, Title |
+
+## 5. Start training
+
+Training scripts, evaluation scripts, and prediction scripts are provided, and the PubLayNet pre-training model is used as an example in this section.
+
+If you do not want training and directly experience the following process of model evaluation, prediction, motion to static, and inference, you can download the provided pre-trained model (PubLayNet dataset) and skip this part.
+
+```bash linenums="1"
+mkdir pretrained_model
+cd pretrained_model
+# Download PubLayNet pre-training model（Direct experience model evaluates, predicts, and turns static）
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout.pdparams
+# Download the PubLaynet inference model（Direct experience model reasoning）
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar
+```
+
+If the test image is Chinese, the pre-trained model of Chinese CDLA dataset can be downloaded to identify 10 types of document regions：Table, Figure, Figure caption, Table, Table caption, Header, Footer, Reference, Equation，Download the training model and inference model of Model 'picodet_lcnet_x1_0_fgd_layout_cdla' in [layout analysis model](../models_list.en.md)。If only the table area in the image is detected, you can download the pre-trained model of the table dataset, and download the training model and inference model of the 'picodet_LCnet_x1_0_FGd_layout_table' model in [Layout Analysis model](../models_list.en.md)
+
+### 5.1. Train
+
+Start training with the PaddleDetection [layout analysis profile](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/picodet/legacy_model/application/layout_analysis)
+
+- Modify Profile
+
+If you want to train your own data set, you need to modify the data configuration and the number of categories in the configuration file.
+
+Using 'configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml' as an example, the change is as follows:
+
+```yaml linenums="1"
+metric: COCO
+# Number of categories
+num_classes: 5
+
+TrainDataset:
+  !COCODataSet
+    # Modify to your own training data directory
+    image_dir: train
+    # Modify to your own training data label file
+    anno_path: train.json
+    # Modify to your own training data root directory
+    dataset_dir: /root/publaynet/
+    data_fields: ['image', 'gt_bbox', 'gt_class', 'is_crowd']
+
+EvalDataset:
+  !COCODataSet
+    # Modify to your own validation data directory
+    image_dir: val
+    # Modify to your own validation data label file
+    anno_path: val.json
+    # Modify to your own validation data root
+    dataset_dir: /root/publaynet/
+
+TestDataset:
+  !ImageFolder
+    # Modify to your own test data label file
+    anno_path: /root/publaynet/val.json
+```
+
+- Start training. During training, PP picodet pre training model will be downloaded by default. There is no need to download in advance.
+
+```bash linenums="1"
+# GPU training supports single-card and multi-card training
+# The training log is automatically saved to the log directory
+
+# Single card training
+export CUDA_VISIBLE_DEVICES=0
+python3 tools/train.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --eval
+
+# Multi-card training, with the -- GPUS parameter specifying the card number
+export CUDA_VISIBLE_DEVICES=0,1,2,3
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --eval
+```
+
+**Attention：**If the video memory is out during training, adjust Batch_size in TrainReader and base_LR in LearningRate. The published config is obtained by 8-card training. If the number of GPU cards is changed to 1, then the base_LR needs to be reduced by 8 times.
+
+After starting training normally, you will see the following log output:
+
+```
+[08/15 04:02:30] ppdet.utils.checkpoint INFO: Finish loading model weights: /root/.cache/paddle/weights/LCNet_x1_0_pretrained.pdparams
+[08/15 04:02:46] ppdet.engine INFO: Epoch: [0] [   0/1929] learning_rate: 0.040000 loss_vfl: 1.216707 loss_bbox: 1.142163 loss_dfl: 0.544196 loss: 2.903065 eta: 17 days, 13:50:26 batch_cost: 15.7452 data_cost: 2.9112 ips: 1.5243 images/s
+[08/15 04:03:19] ppdet.engine INFO: Epoch: [0] [  20/1929] learning_rate: 0.064000 loss_vfl: 1.180627 loss_bbox: 0.939552 loss_dfl: 0.442436 loss: 2.628206 eta: 2 days, 12:18:53 batch_cost: 1.5770 data_cost: 0.0008 ips: 15.2184 images/s
+[08/15 04:03:47] ppdet.engine INFO: Epoch: [0] [  40/1929] learning_rate: 0.088000 loss_vfl: 0.543321 loss_bbox: 1.071401 loss_dfl: 0.457817 loss: 2.057003 eta: 2 days, 0:07:03 batch_cost: 1.3190 data_cost: 0.0007 ips: 18.1954 images/s
+[08/15 04:04:12] ppdet.engine INFO: Epoch: [0] [  60/1929] learning_rate: 0.112000 loss_vfl: 0.630989 loss_bbox: 0.859183 loss_dfl: 0.384702 loss: 1.883143 eta: 1 day, 19:01:29 batch_cost: 1.2177 data_cost: 0.0006 ips: 19.7087 images/s
+```
+
+- `--eval` indicates that the best model is saved as `output/picodet_lcnet_x1_0_layout/best_accuracy`  by default during the evaluation process 。
+
+**Note that the configuration file for prediction / evaluation must be consistent with the training.**
+
+### 5.2. FGD Distillation Training
+
+PaddleDetection supports FGD-based [Focal and Global Knowledge Distillation for Detectors]( https://arxiv.org/abs/2111.11837v1)  The training process of the target detection model of distillation, FGD distillation is divided into two parts `Focal` and `Global`.     `Focal` Distillation separates the foreground and background of the image, allowing the student model to focus on the key pixels of the foreground and background features of the teacher model respectively;`Global`Distillation section reconstructs the relationships between different pixels and transfers them from the teacher to the student to compensate for the global information lost in `Focal`Distillation.
+
+Change the dataset and modify the data configuration and number of categories in the [TODO] configuration, referring to 4.1. Start training:
+
+```bash linenums="1"
+# Single Card Training
+export CUDA_VISIBLE_DEVICES=0
+python3 tools/train.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \
+    --eval
+```
+
+- `-c`: Specify the model configuration file.
+- `--slim_config`:  Specify the compression policy profile.
+
+## 6. Model evaluation and prediction
+
+### 6.1. Indicator evaluation
+
+ Model parameters in training are saved by default in `output/picodet_ Lcnet_ X1_ 0_ Under the layout` directory. When evaluating indicators, you need to set `weights` to point to the saved parameter file.Assessment datasets can be accessed via `configs/picodet/legacy_ Model/application/layout_ Analysis/picodet_ Lcnet_ X1_ 0_ Layout. Yml` . Modify `EvalDataset`  : `img_dir`,`anno_ Path`and`dataset_dir` setting.
+
+```bash linenums="1"
+# GPU evaluation, weights as weights to be measured
+python3 tools/eval.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    -o weights=./output/picodet_lcnet_x1_0_layout/best_model
+```
+
+The following information will be printed out, such as mAP, AP0.5, etc.
+
+```python linenums="1"
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.935
+ Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.979
+ Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.956
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.404
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.782
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.969
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.539
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.938
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.949
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.495
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.818
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.978
+[08/15 07:07:09] ppdet.engine INFO: Total sample number: 11245, averge FPS: 24.405059207157436
+[08/15 07:07:09] ppdet.engine INFO: Best test bbox ap is 0.935.
+```
+
+If you use the provided pre-training model for evaluation or the FGD distillation training model, replace the `weights` model path and execute the following command for evaluation:
+
+```bash linenums="1"
+python3 tools/eval.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \
+    -o weights=output/picodet_lcnet_x2_5_layout/best_model
+```
+
+- `-c`: Specify the model configuration file.
+- `--slim_config`:  Specify the distillation policy profile.
+- `-o weights`: Specify the model path trained by the distillation algorithm.
+
+### 6.2. Test Layout Analysis Results
+
+The profile predicted to be used must be consistent with the training, for example, if you pass `python3 tools/train'. Py-c configs/picodet/legacy_ Model/application/layout_ Analysis/picodet_ Lcnet_ X1_ 0_ Layout. Yml` completed the training process for the model.
+
+With  trained PaddleDetection model, you can use the following commands to make model predictions.
+
+```bash linenums="1"
+python3 tools/infer.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    -o weights='output/picodet_lcnet_x1_0_layout/best_model.pdparams' \
+    --infer_img='docs/images/layout.jpg' \
+    --output_dir=output_dir/ \
+    --draw_threshold=0.5
+```
+
+- `--infer_img`:  Reasoning for a single picture can also be done via `--infer_ Dir`Inform all pictures in the file.
+- `--output_dir`:  Specify the path to save the visualization results.
+- `--draw_threshold`:Specify the NMS threshold for drawing the result box.
+
+If you use the provided pre-training model for prediction or the FGD distillation training model, change the `weights` model path and execute the following command to make the prediction:
+
+```bash linenums="1"
+python3 tools/infer.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \
+    -o weights='output/picodet_lcnet_x2_5_layout/best_model.pdparams' \
+    --infer_img='docs/images/layout.jpg' \
+    --output_dir=output_dir/ \
+    --draw_threshold=0.5
+```
+
+## 7. Model Export and Inference
+
+### 7.1 Model Export
+
+The inference model (the model saved by `paddle.jit.save`) is generally a solidified model saved after the model training is completed, and is mostly used to give prediction in deployment.
+
+The model saved during the training process is the checkpoints model, which saves the parameters of the model and is mostly used to resume training.
+
+Compared with the checkpoints model, the inference model will additionally save the structural information of the model. Therefore, it is easier to deploy because the model structure and model parameters are already solidified in the inference model file, and is suitable for integration with actual systems.
+
+Layout analysis model to inference model steps are as follows：
+
+```bash linenums="1"
+python3 tools/export_model.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    -o weights=output/picodet_lcnet_x1_0_layout/best_model \
+    --output_dir=output_inference/
+```
+
+- If no post-export processing is required, specify：`-o export.benchmark=True`（If -o already exists, delete -o here）
+- If you do not need to export NMS, specify：`-o export.nms=False`
+
+After successful conversion, there are three files in the directory:
+
+```
+output_inference/picodet_lcnet_x1_0_layout/
+    ├── model.pdiparams         # inference Parameter file for model
+    ├── model.pdiparams.info    # inference Model parameter information, ignorable
+    └── model.pdmodel           # inference Model Structure File for Model
+```
+
+If you change the `weights` model path using the provided pre-training model to the Inference model, or using the FGD distillation training model, the model to inference model steps are as follows:
+
+```bash linenums="1"
+python3 tools/export_model.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \
+    -o weights=./output/picodet_lcnet_x2_5_layout/best_model \
+    --output_dir=output_inference/
+```
+
+### 7.2 Model inference
+
+Replace model_with the provided inference training model for inference or the FGD distillation training `model_dir`Inference model path, execute the following commands for inference:
+
+```bash linenums="1"
+python3 deploy/python/infer.py \
+    --model_dir=output_inference/picodet_lcnet_x1_0_layout/ \
+    --image_file=docs/images/layout.jpg \
+    --device=CPU
+```
+
+- --device：Specify the GPU or CPU device
+
+When model inference is complete, you will see the following log output:
+
+```
+------------------------------------------
+-----------  Model Configuration -----------
+Model Arch: PicoDet
+Transform Order:
+--transform op: Resize
+--transform op: NormalizeImage
+--transform op: Permute
+--transform op: PadStride
+--------------------------------------------
+class_id:0, confidence:0.9921, left_top:[20.18,35.66],right_bottom:[341.58,600.99]
+class_id:0, confidence:0.9914, left_top:[19.77,611.42],right_bottom:[341.48,901.82]
+class_id:0, confidence:0.9904, left_top:[369.36,375.10],right_bottom:[691.29,600.59]
+class_id:0, confidence:0.9835, left_top:[369.60,608.60],right_bottom:[691.38,736.72]
+class_id:0, confidence:0.9830, left_top:[369.58,805.38],right_bottom:[690.97,901.80]
+class_id:0, confidence:0.9716, left_top:[383.68,271.44],right_bottom:[688.93,335.39]
+class_id:0, confidence:0.9452, left_top:[370.82,34.48],right_bottom:[688.10,63.54]
+class_id:1, confidence:0.8712, left_top:[370.84,771.03],right_bottom:[519.30,789.13]
+class_id:3, confidence:0.9856, left_top:[371.28,67.85],right_bottom:[685.73,267.72]
+save result to: output/layout.jpg
+Test iter 0
+------------------ Inference Time Info ----------------------
+total_time(ms): 2196.0, img_num: 1
+average latency time(ms): 2196.00, QPS: 0.455373
+preprocess_time(ms): 2172.50, inference_time(ms): 11.90, postprocess_time(ms): 11.60
+```
+
+- Model：model structure
+- Transform Order：Preprocessing operation
+- class_id, confidence, left_top, right_bottom：Indicates category id, confidence level, upper left coordinate, lower right coordinate, respectively
+- save result to：Save path of visual layout analysis results, default save to ./output folder
+- inference time info：Inference time, where preprocess_time represents the preprocessing time, Inference_time represents the model prediction time, and postprocess_time represents the post-processing time
+
+The result of visualization layout is shown in the following figure
+
+![img](./images/layout_res.jpg)
+
+## Citations
+
+```bibtex
+@inproceedings{zhong2019publaynet,
+  title={PubLayNet: largest dataset ever for document layout analysis},
+  author={Zhong, Xu and Tang, Jianbin and Yepes, Antonio Jimeno},
+  booktitle={2019 International Conference on Document Analysis and Recognition (ICDAR)},
+  year={2019},
+  volume={},
+  number={},
+  pages={1015-1022},
+  doi={10.1109/ICDAR.2019.00166},
+  ISSN={1520-5363},
+  month={Sep.},
+  organization={IEEE}
+}
+
+@inproceedings{yang2022focal,
+  title={Focal and global knowledge distillation for detectors},
+  author={Yang, Zhendong and Li, Zhe and Jiang, Xiaohu and Gong, Yuan and Yuan, Zehuan and Zhao, Danpei and Yuan, Chun},
+  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
+  pages={4643--4652},
+  year={2022}
+}
+```
diff --git a/docs/version2.x/ppstructure/model_train/train_layout.md b/docs/version2.x/ppstructure/model_train/train_layout.md
new file mode 100644
index 0000000..15207db
--- /dev/null
+++ b/docs/version2.x/ppstructure/model_train/train_layout.md
@@ -0,0 +1,441 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# 版面分析
+
+## 1. 简介
+
+版面分析指的是对图片形式的文档进行区域划分，定位其中的关键区域，如文字、标题、表格、图片等。版面分析算法基于[PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection)的轻量模型PP-PicoDet进行开发，包含英文、中文、表格版面分析3类模型。其中，英文模型支持Text、Title、Tale、Figure、List5类区域的检测，中文模型支持Text、Title、Figure、Figure caption、Table、Table caption、Header、Footer、Reference、Equation10类区域的检测，表格版面分析支持Table区域的检测，版面分析效果如下图所示：
+
+![img](./images/layout.jpg)
+
+## 2. 快速开始
+
+PP-Structure目前提供了中文、英文、表格三类文档版面分析模型，模型链接见 [models_list](../models_list.md)。也提供了whl包的形式方便快速使用，详见 [quickstart](../quick_start.md)。
+
+## 3. 安装
+
+### 3.1. 安装PaddlePaddle
+
+- **（1) 安装PaddlePaddle**
+
+```bash linenums="1"
+python3 -m pip install --upgrade pip
+
+# GPU安装
+python3 -m pip install "paddlepaddle-gpu>=2.3" -i https://mirror.baidu.com/pypi/simple
+
+# CPU安装
+python3 -m pip install "paddlepaddle>=2.3" -i https://mirror.baidu.com/pypi/simple
+```
+
+更多需求，请参照[安装文档](https://www.paddlepaddle.org.cn/install/quick)中的说明进行操作。
+
+### 3.2. 安装PaddleDetection
+
+- **（1）下载PaddleDetection源码**
+
+```bash linenums="1"
+git clone https://github.com/PaddlePaddle/PaddleDetection.git
+```
+
+- **（2）安装其他依赖**
+
+```bash linenums="1"
+cd PaddleDetection
+python3 -m pip install -r requirements.txt
+```
+
+## 4. 数据准备
+
+如果希望直接体验预测过程，可以跳过数据准备，下载我们提供的预训练模型。
+
+### 4.1. 英文数据集
+
+下载文档分析数据集[PubLayNet](https://developer.ibm.com/exchanges/data/all/publaynet/)（数据集96G），包含5个类：`{0: "Text", 1: "Title", 2: "List", 3:"Table", 4:"Figure"}`
+
+```
+# 下载数据
+wget https://dax-cdn.cdn.appdomain.cloud/dax-publaynet/1.0.0/publaynet.tar.gz
+# 解压数据
+tar -xvf publaynet.tar.gz
+```
+
+解压之后的**目录结构：**
+
+```
+|-publaynet
+  |- test
+     |- PMC1277013_00004.jpg
+     |- PMC1291385_00002.jpg
+     | ...
+  |- train.json
+  |- train
+     |- PMC1291385_00002.jpg
+     |- PMC1277013_00004.jpg
+     | ...
+  |- val.json
+  |- val
+     |- PMC538274_00004.jpg
+     |- PMC539300_00004.jpg
+     | ...
+```
+
+**数据分布：**
+
+| File or Folder | Description    | num     |
+| :------------- | :------------- | ------- |
+| `train/`       | 训练集图片     | 335,703 |
+| `val/`         | 验证集图片     | 11,245  |
+| `test/`        | 测试集图片     | 11,405  |
+| `train.json`   | 训练集标注文件 | -       |
+| `val.json`     | 验证集标注文件 | -       |
+
+**标注格式：**
+
+json文件包含所有图像的标注，数据以字典嵌套的方式存放，包含以下key：
+
+- info，表示标注文件info。
+
+- licenses，表示标注文件licenses。
+
+- images，表示标注文件中图像信息列表，每个元素是一张图像的信息。如下为其中一张图像的信息：
+
+  ```
+  {
+      'file_name': 'PMC4055390_00006.jpg',    # file_name
+      'height': 601,                      # image height
+      'width': 792,                       # image width
+      'id': 341427                        # image id
+  }
+  ```
+
+- annotations，表示标注文件中目标物体的标注信息列表，每个元素是一个目标物体的标注信息。如下为其中一个目标物体的标注信息：
+
+  ```
+  {
+
+      'segmentation':             # 物体的分割标注
+      'area': 60518.099043117836, # 物体的区域面积
+      'iscrowd': 0,               # iscrowd
+      'image_id': 341427,         # image id
+      'bbox': [50.58, 490.86, 240.15, 252.16], # bbox [x1,y1,w,h]
+      'category_id': 1,           # category_id
+      'id': 3322348               # image id
+  }
+  ```
+
+### 4.2. 更多数据集
+
+我们提供了CDLA(中文版面分析)、TableBank(表格版面分析)等数据集的下连接，处理为上述标注文件json格式，即可以按相同方式进行训练。
+
+| dataset                                                         | 简介                                                                                                                                                                           |
+| --------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| [cTDaR2019_cTDaR](https://cndplab-founder.github.io/cTDaR2019/) | 用于表格检测(TRACKA)和表格识别(TRACKB)。图片类型包含历史数据集(以cTDaR_t0开头，如cTDaR_t00872.jpg)和现代数据集(以cTDaR_t1开头，cTDaR_t10482.jpg)。                             |
+| [IIIT-AR-13K](http://cvit.iiit.ac.in/usodi/iiitar13k.php)       | 手动注释公开的年度报告中的图形或页面而构建的数据集，包含5类：table, figure, natural image, logo, and signature                                                                 |
+| [CDLA](https://github.com/buptlihang/CDLA)                      | 中文文档版面分析数据集，面向中文文献类（论文）场景，包含10类：Text、Title、Figure、Figure caption、Table、Table caption、Header、Footer、Reference、Equation                   |
+| [TableBank](https://github.com/doc-analysis/TableBank)          | 用于表格检测和识别大型数据集，包含Word和Latex2种文档格式                                                                                                                       |
+| [DocBank](https://github.com/doc-analysis/DocBank)              | 使用弱监督方法构建的大规模数据集(500K文档页面)，用于文档布局分析，包含12类：Author、Caption、Date、Equation、Figure、Footer、List、Paragraph、Reference、Section、Table、Title |
+
+## 5. 开始训练
+
+提供了训练脚本、评估脚本和预测脚本，本节将以PubLayNet预训练模型为例进行讲解。
+
+如果不希望训练，直接体验后面的模型评估、预测、动转静、推理的流程，可以下载提供的预训练模型(PubLayNet数据集)，并跳过5.1和5.2。
+
+```bash linenums="1"
+mkdir pretrained_model
+cd pretrained_model
+# 下载PubLayNet预训练模型（直接体验模型评估、预测、动转静）
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout.pdparams
+# 下载PubLaynet推理模型（直接体验模型推理）
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar
+```
+
+如果测试图片为中文，可以下载中文CDLA数据集的预训练模型，识别10类文档区域：Table、Figure、Figure caption、Table、Table caption、Header、Footer、Reference、Equation，在[版面分析模型](../models_list.md)中下载`picodet_lcnet_x1_0_fgd_layout_cdla`模型的训练模型和推理模型。如果只检测图片中的表格区域，可以下载表格数据集的预训练模型，在[版面分析模型](../models_list.md)中下载`picodet_lcnet_x1_0_fgd_layout_table`模型的训练模型和推理模型。
+
+### 5.1. 启动训练
+
+使用PaddleDetection[版面分析配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/picodet/legacy_model/application/layout_analysis)启动训练
+
+- 修改配置文件
+
+如果你希望训练自己的数据集，需要修改配置文件中的数据配置、类别数。
+
+以`configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml` 为例，修改的内容如下所示。
+
+```yaml linenums="1"
+metric: COCO
+# 类别数
+num_classes: 5
+
+TrainDataset:
+  !COCODataSet
+    # 修改为你自己的训练数据目录
+    image_dir: train
+    # 修改为你自己的训练数据标签文件
+    anno_path: train.json
+    # 修改为你自己的训练数据根目录
+    dataset_dir: /root/publaynet/
+    data_fields: ['image', 'gt_bbox', 'gt_class', 'is_crowd']
+
+EvalDataset:
+  !COCODataSet
+    # 修改为你自己的验证数据目录
+    image_dir: val
+    # 修改为你自己的验证数据标签文件
+    anno_path: val.json
+    # 修改为你自己的验证数据根目录
+    dataset_dir: /root/publaynet/
+
+TestDataset:
+  !ImageFolder
+    # 修改为你自己的测试数据标签文件
+    anno_path: /root/publaynet/val.json
+```
+
+- 开始训练，在训练时，会默认下载PP-PicoDet预训练模型，这里无需预先下载。
+
+```bash linenums="1"
+# GPU训练 支持单卡，多卡训练
+# 训练日志会自动保存到 log 目录中
+
+# 单卡训练
+export CUDA_VISIBLE_DEVICES=0
+python3 tools/train.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --eval
+
+# 多卡训练，通过--gpus参数指定卡号
+export CUDA_VISIBLE_DEVICES=0,1,2,3
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --eval
+```
+
+**注意：**如果训练时显存out memory，将TrainReader中batch_size调小，同时LearningRate中base_lr等比例减小。发布的config均由8卡训练得到，如果改变GPU卡数为1，那么base_lr需要减小8倍。
+
+正常启动训练后，会看到以下log输出：
+
+```bash linenums="1"
+[08/15 04:02:30] ppdet.utils.checkpoint INFO: Finish loading model weights: /root/.cache/paddle/weights/LCNet_x1_0_pretrained.pdparams
+[08/15 04:02:46] ppdet.engine INFO: Epoch: [0] [   0/1929] learning_rate: 0.040000 loss_vfl: 1.216707 loss_bbox: 1.142163 loss_dfl: 0.544196 loss: 2.903065 eta: 17 days, 13:50:26 batch_cost: 15.7452 data_cost: 2.9112 ips: 1.5243 images/s
+[08/15 04:03:19] ppdet.engine INFO: Epoch: [0] [  20/1929] learning_rate: 0.064000 loss_vfl: 1.180627 loss_bbox: 0.939552 loss_dfl: 0.442436 loss: 2.628206 eta: 2 days, 12:18:53 batch_cost: 1.5770 data_cost: 0.0008 ips: 15.2184 images/s
+[08/15 04:03:47] ppdet.engine INFO: Epoch: [0] [  40/1929] learning_rate: 0.088000 loss_vfl: 0.543321 loss_bbox: 1.071401 loss_dfl: 0.457817 loss: 2.057003 eta: 2 days, 0:07:03 batch_cost: 1.3190 data_cost: 0.0007 ips: 18.1954 images/s
+[08/15 04:04:12] ppdet.engine INFO: Epoch: [0] [  60/1929] learning_rate: 0.112000 loss_vfl: 0.630989 loss_bbox: 0.859183 loss_dfl: 0.384702 loss: 1.883143 eta: 1 day, 19:01:29 batch_cost: 1.2177 data_cost: 0.0006 ips: 19.7087 images/s
+```
+
+- `--eval`表示训练的同时，进行评估， 评估过程中默认将最佳模型，保存为 `output/picodet_lcnet_x1_0_layout/best_accuracy` 。
+
+**注意，预测/评估时的配置文件请务必与训练一致。**
+
+### 5.2. FGD蒸馏训练
+
+PaddleDetection支持了基于FGD([Focal and Global Knowledge Distillation for Detectors](https://arxiv.org/abs/2111.11837v1))蒸馏的目标检测模型训练过程，FGD蒸馏分为两个部分`Focal`和`Global`。`Focal`蒸馏分离图像的前景和背景，让学生模型分别关注教师模型的前景和背景部分特征的关键像素；`Global`蒸馏部分重建不同像素之间的关系并将其从教师转移到学生，以补偿`Focal`蒸馏中丢失的全局信息。
+
+更换数据集，修改【TODO】配置中的数据配置、类别数，具体可以参考4.1。启动训练：
+
+```bash linenums="1"
+# 单卡训练
+export CUDA_VISIBLE_DEVICES=0
+python3 tools/train.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \
+    --eval
+```
+
+- `-c`: 指定模型配置文件。
+- `--slim_config`: 指定压缩策略配置文件。
+
+## 6. 模型评估与预测
+
+### 6.1. 指标评估
+
+训练中模型参数默认保存在`output/picodet_lcnet_x1_0_layout`目录下。在评估指标时，需要设置`weights`指向保存的参数文件。评估数据集可以通过 `configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml`  修改`EvalDataset`中的 `image_dir`、`anno_path`和`dataset_dir` 设置。
+
+```bash linenums="1"
+# GPU 评估， weights 为待测权重
+python3 tools/eval.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    -o weights=./output/picodet_lcnet_x1_0_layout/best_model
+```
+
+会输出以下信息，打印出mAP、AP0.5等信息。
+
+```python linenums="1"
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.935
+ Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.979
+ Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.956
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.404
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.782
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.969
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.539
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.938
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.949
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.495
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.818
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.978
+[08/15 07:07:09] ppdet.engine INFO: Total sample number: 11245, averge FPS: 24.405059207157436
+[08/15 07:07:09] ppdet.engine INFO: Best test bbox ap is 0.935.
+```
+
+若使用**提供的预训练模型进行评估**，或使用**FGD蒸馏训练的模型**，更换`weights`模型路径，执行如下命令进行评估：
+
+```bash linenums="1"
+python3 tools/eval.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \
+    -o weights=output/picodet_lcnet_x2_5_layout/best_model
+```
+
+- `-c`: 指定模型配置文件。
+- `--slim_config`: 指定蒸馏策略配置文件。
+- `-o weights`: 指定蒸馏算法训好的模型路径。
+
+### 6.2 测试版面分析结果
+
+预测使用的配置文件必须与训练一致，如您通过 `python3 tools/train.py -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml` 完成了模型的训练过程。
+
+使用 PaddleDetection 训练好的模型，您可以使用如下命令进行模型预测。
+
+```bash linenums="1"
+python3 tools/infer.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    -o weights='output/picodet_lcnet_x1_0_layout/best_model.pdparams' \
+    --infer_img='docs/images/layout.jpg' \
+    --output_dir=output_dir/ \
+    --draw_threshold=0.5
+```
+
+- `--infer_img`: 推理单张图片，也可以通过`--infer_dir`推理文件中的所有图片。
+- `--output_dir`: 指定可视化结果保存路径。
+- `--draw_threshold`:指定绘制结果框的NMS阈值。
+
+若使用**提供的预训练模型进行预测**，或使用**FGD蒸馏训练的模型**，更换`weights`模型路径，执行如下命令进行预测：
+
+```bash linenums="1"
+python3 tools/infer.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \
+    -o weights='output/picodet_lcnet_x2_5_layout/best_model.pdparams' \
+    --infer_img='docs/images/layout.jpg' \
+    --output_dir=output_dir/ \
+    --draw_threshold=0.5
+```
+
+## 7. 模型导出与预测
+
+### 7.1 模型导出
+
+inference 模型（`paddle.jit.save`保存的模型） 一般是模型训练，把模型结构和模型参数保存在文件中的固化模型，多用于预测部署场景。 训练过程中保存的模型是checkpoints模型，保存的只有模型的参数，多用于恢复训练等。 与checkpoints模型相比，inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成。
+
+版面分析模型转inference模型步骤如下：
+
+```bash linenums="1"
+python3 tools/export_model.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    -o weights=output/picodet_lcnet_x1_0_layout/best_model \
+    --output_dir=output_inference/
+```
+
+- 如无需导出后处理，请指定：`-o export.benchmark=True`（如果-o已出现过，此处删掉-o）
+- 如无需导出NMS，请指定：`-o export.nms=False`
+
+转换成功后，在目录下有三个文件：
+
+```
+output_inference/picodet_lcnet_x1_0_layout/
+    ├── model.pdiparams         # inference模型的参数文件
+    ├── model.pdiparams.info    # inference模型的参数信息，可忽略
+    └── model.pdmodel           # inference模型的模型结构文件
+```
+
+若使用**提供的预训练模型转Inference模型**，或使用**FGD蒸馏训练的模型**，更换`weights`模型路径，模型转inference模型步骤如下：
+
+```bash linenums="1"
+python3 tools/export_model.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \
+    -o weights=./output/picodet_lcnet_x2_5_layout/best_model \
+    --output_dir=output_inference/
+```
+
+### 7.2 模型推理
+
+若使用**提供的推理训练模型推理**，或使用**FGD蒸馏训练的模型**，更换`model_dir`推理模型路径，执行如下命令进行推理：
+
+```bash linenums="1"
+python3 deploy/python/infer.py \
+    --model_dir=output_inference/picodet_lcnet_x1_0_layout/ \
+    --image_file=docs/images/layout.jpg \
+    --device=CPU
+```
+
+- --device：指定GPU、CPU设备
+
+模型推理完成，会看到以下log输出
+
+```
+------------------------------------------
+-----------  Model Configuration -----------
+Model Arch: PicoDet
+Transform Order:
+--transform op: Resize
+--transform op: NormalizeImage
+--transform op: Permute
+--transform op: PadStride
+--------------------------------------------
+class_id:0, confidence:0.9921, left_top:[20.18,35.66],right_bottom:[341.58,600.99]
+class_id:0, confidence:0.9914, left_top:[19.77,611.42],right_bottom:[341.48,901.82]
+class_id:0, confidence:0.9904, left_top:[369.36,375.10],right_bottom:[691.29,600.59]
+class_id:0, confidence:0.9835, left_top:[369.60,608.60],right_bottom:[691.38,736.72]
+class_id:0, confidence:0.9830, left_top:[369.58,805.38],right_bottom:[690.97,901.80]
+class_id:0, confidence:0.9716, left_top:[383.68,271.44],right_bottom:[688.93,335.39]
+class_id:0, confidence:0.9452, left_top:[370.82,34.48],right_bottom:[688.10,63.54]
+class_id:1, confidence:0.8712, left_top:[370.84,771.03],right_bottom:[519.30,789.13]
+class_id:3, confidence:0.9856, left_top:[371.28,67.85],right_bottom:[685.73,267.72]
+save result to: output/layout.jpg
+Test iter 0
+------------------ Inference Time Info ----------------------
+total_time(ms): 2196.0, img_num: 1
+average latency time(ms): 2196.00, QPS: 0.455373
+preprocess_time(ms): 2172.50, inference_time(ms): 11.90, postprocess_time(ms): 11.60
+```
+
+- Model：模型结构
+- Transform Order：预处理操作
+- class_id、confidence、left_top、right_bottom：分别表示类别id、置信度、左上角坐标、右下角坐标
+- save result to：可视化版面分析结果保存路径，默认保存到`./output`文件夹
+- Inference Time Info：推理时间，其中preprocess_time表示预处理耗时，inference_time表示模型预测耗时，postprocess_time表示后处理耗时
+
+可视化版面结果如下图所示
+
+![img](./images/layout_res.jpg)
+
+## Citations
+
+```bibtex
+@inproceedings{zhong2019publaynet,
+  title={PubLayNet: largest dataset ever for document layout analysis},
+  author={Zhong, Xu and Tang, Jianbin and Yepes, Antonio Jimeno},
+  booktitle={2019 International Conference on Document Analysis and Recognition (ICDAR)},
+  year={2019},
+  volume={},
+  number={},
+  pages={1015-1022},
+  doi={10.1109/ICDAR.2019.00166},
+  ISSN={1520-5363},
+  month={Sep.},
+  organization={IEEE}
+}
+
+@inproceedings{yang2022focal,
+  title={Focal and global knowledge distillation for detectors},
+  author={Yang, Zhendong and Li, Zhe and Jiang, Xiaohu and Gong, Yuan and Yuan, Zehuan and Zhao, Danpei and Yuan, Chun},
+  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
+  pages={4643--4652},
+  year={2022}
+}
+```
diff --git a/docs/version2.x/ppstructure/model_train/train_table.en.md b/docs/version2.x/ppstructure/model_train/train_table.en.md
new file mode 100644
index 0000000..b76822d
--- /dev/null
+++ b/docs/version2.x/ppstructure/model_train/train_table.en.md
@@ -0,0 +1,164 @@
+---
+comments: true
+---
+
+# Table Recognition
+
+## 1. pipeline
+
+The table recognition mainly contains three models
+
+1. Single line text detection-DB
+2. Single line text recognition-CRNN
+3. Table structure and cell coordinate prediction-SLANet
+
+The table recognition flow chart is as follows
+
+![tableocr_pipeline](./images/tableocr_pipeline.jpg)
+
+1. The coordinates of single-line text is detected by DB model, and then sends it to the recognition model to get the recognition result.
+2. The table structure and cell coordinates is predicted by SLANet model.
+3. The recognition result of the cell is combined by the coordinates, recognition result of the single line and the coordinates of the cell.
+4. The cell recognition result and the table structure together construct the html string of the table.
+
+## 2. Performance
+
+We evaluated the algorithm on the PubTabNet<sup>[1]</sup> eval dataset, and the performance is as follows:
+
+|Method|Acc|[TEDS(Tree-Edit-Distance-based Similarity)](https://github.com/ibm-aur-nlp/PubTabNet/tree/master/src)|Speed|
+| --- | --- | --- | ---|
+| EDD<sup>[2]</sup> |x| 88.30% |x|
+| TableRec-RARE(ours) | 71.73%| 93.88% |779ms|
+| SLANet(ours) | 76.31%|    95.89%|766ms|
+
+The performance indicators are explained as follows:
+
+- Acc: The accuracy of the table structure in each image, a wrong token is considered an error.
+- TEDS: The accuracy of the model's restoration of table information. This indicator evaluates not only the table structure, but also the text content in the table.
+- Speed: The inference speed of a single image when the model runs on the CPU machine and MKL is enabled.
+
+## 3. Result
+
+![img](./images/table_ch_result1.jpg)
+
+![img](./images/table_ch_result2.jpg)
+
+![img](./images/table_ch_result3.jpg)
+
+## 4. How to use
+
+### 4.1 Quick start
+
+PP-Structure currently provides table recognition models in both Chinese and English. For the model link, see [models_list](../models_list.en.md). The whl package is also provided for quick use, see [quickstart](../quick_start.en.md) for details.
+
+The following takes the Chinese table recognition model as an example to introduce how to recognize a table.
+
+Use the following commands to quickly complete the identification of a table.
+
+```bash linenums="1"
+cd PaddleOCR/ppstructure
+
+# download model
+mkdir inference && cd inference
+# Download the PP-OCRv3 text detection model and unzip it
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar && tar xf PP-OCRv3_mobile_det_infer.tar
+# Download the PP-OCRv3 text recognition model and unzip it
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar && tar xf PP-OCRv3_mobile_rec_infer.tar
+# Download the PP-StructureV2 form recognition model and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/ch_ppstructure_mobile_v2.0_SLANet_infer.tar && tar xf ch_ppstructure_mobile_v2.0_SLANet_infer.tar
+cd ..
+# run
+python3 table/predict_table.py \
+    --det_model_dir=inference/PP-OCRv3_mobile_det_infer \
+    --rec_model_dir=inference/PP-OCRv3_mobile_rec_infer  \
+    --table_model_dir=inference/ch_ppstructure_mobile_v2.0_SLANet_infer \
+    --rec_char_dict_path=../ppocr/utils/ppocr_keys_v1.txt \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict_ch.txt \
+    --image_dir=docs/table/table.jpg \
+    --output=../output/table
+
+```
+
+After the operation is completed, the excel table of each image will be saved to the directory specified by the output field, and an html file will be produced in the directory to visually view the cell coordinates and the recognized table.
+
+**NOTE**
+
+1. If you want to use the English table recognition model, you need to download the English text detection and recognition model and the English table recognition model in [models_list](../models_list.en.md), and replace `table_structure_dict_ch.txt` with `table_structure_dict.txt`.
+2. To use the TableRec-RARE model, you need to replace `table_structure_dict_ch.txt` with `table_structure_dict.txt`, and add parameter `--merge_no_span_structure=False`
+
+### 4.2 Training, Evaluation and Inference
+
+The training, evaluation and inference process of the text detection model can be referred to [detection](../../ppocr/model_train/detection.en.md)
+
+The training, evaluation and inference process of the text recognition model can be referred to [recognition](../../ppocr/model_train/recognition.en.md)
+
+The training, evaluation and inference process of the table recognition model can be referred to [table_recognition](./train_table.en.md)
+
+### 4.3 Calculate TEDS
+
+The table uses [TEDS(Tree-Edit-Distance-based Similarity)](https://github.com/ibm-aur-nlp/PubTabNet/tree/master/src) as the evaluation metric of the model. Before the model evaluation, the three models in the pipeline need to be exported as inference models (we have provided them), and the gt for evaluation needs to be prepared. Examples of gt are as follows:
+
+```txt
+PMC5755158_010_01.png    <html><body><table><thead><tr><td></td><td><b>Weaning</b></td><td><b>Week 15</b></td><td><b>Off-test</b></td></tr></thead><tbody><tr><td>Weaning</td><td>–</td><td>–</td><td>–</td></tr><tr><td>Week 15</td><td>–</td><td>0.17 ± 0.08</td><td>0.16 ± 0.03</td></tr><tr><td>Off-test</td><td>–</td><td>0.80 ± 0.24</td><td>0.19 ± 0.09</td></tr></tbody></table></body></html>
+```
+
+Each line in gt consists of the file name and the html string of the table. The file name and the html string of the table are separated by `\t`.
+
+You can also use the following command to generate an evaluation gt file from the annotation file:
+
+```bash linenums="1"
+python3 ppstructure/table/convert_label2html.py --ori_gt_path /path/to/your_label_file --save_path /path/to/save_file
+```
+
+Use the following command to evaluate. After the evaluation is completed, the teds indicator will be output.
+
+```bash linenums="1"
+python3 table/eval_table.py \
+    --det_model_dir=path/to/det_model_dir \
+    --rec_model_dir=path/to/rec_model_dir \
+    --table_model_dir=path/to/table_model_dir \
+    --image_dir=docs/table/table.jpg \
+    --rec_char_dict_path=../ppocr/utils/dict/table_dict.txt \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt \
+    --det_limit_side_len=736 \
+    --det_limit_type=min \
+    --gt_path=path/to/gt.txt
+```
+
+Evaluate on the PubLatNet dataset using the English model
+
+```bash linenums="1"
+cd PaddleOCR/ppstructure
+# Download the model
+mkdir inference && cd inference
+# Download the text detection model trained on the PubTabNet dataset and unzip it
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar
+# Download the text recognition model trained on the PubTabNet dataset and unzip it
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar
+# Download the table recognition model trained on the PubTabNet dataset and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/en_ppstructure_mobile_v2.0_SLANet_infer.tar && tar xf en_ppstructure_mobile_v2.0_SLANet_infer.tar
+cd ..
+
+python3 table/eval_table.py \
+    --det_model_dir=inference/en_ppocr_mobile_v2.0_table_det_infer \
+    --rec_model_dir=inference/en_ppocr_mobile_v2.0_table_rec_infer \
+    --table_model_dir=inference/en_ppstructure_mobile_v2.0_SLANet_infer \
+    --image_dir=train_data/table/pubtabnet/val/ \
+    --rec_char_dict_path=../ppocr/utils/dict/table_dict.txt \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt \
+    --det_limit_side_len=736 \
+    --det_limit_type=min \
+    --rec_image_shape=3,32,320 \
+    --gt_path=path/to/gt.txt
+```
+
+output is
+
+```bash linenums="1"
+teds: 95.89
+```
+
+## 5. Reference
+
+1. <https://github.com/ibm-aur-nlp/PubTabNet>
+2. <https://arxiv.org/pdf/1911.10683>
diff --git a/docs/version2.x/ppstructure/model_train/train_table.md b/docs/version2.x/ppstructure/model_train/train_table.md
new file mode 100644
index 0000000..15608bc
--- /dev/null
+++ b/docs/version2.x/ppstructure/model_train/train_table.md
@@ -0,0 +1,167 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# 表格识别
+
+## 1. 表格识别 pipeline
+
+表格识别主要包含三个模型
+
+1. 单行文本检测-DB
+2. 单行文本识别-CRNN
+3. 表格结构和cell坐标预测-SLANet
+
+具体流程图如下
+
+![tableocr_pipeline](./images/tableocr_pipeline.jpg)
+
+流程说明:
+
+1. 图片由单行文字检测模型检测到单行文字的坐标，然后送入识别模型拿到识别结果。
+2. 图片由SLANet模型拿到表格的结构信息和单元格的坐标信息。
+3. 由单行文字的坐标、识别结果和单元格的坐标一起组合出单元格的识别结果。
+4. 单元格的识别结果和表格结构一起构造表格的html字符串。
+
+## 2. 性能
+
+我们在 PubTabNet<sup>[1]</sup> 评估数据集上对算法进行了评估，性能如下
+
+| 算法                | Acc    | [TEDS(Tree-Edit-Distance-based Similarity)](https://github.com/ibm-aur-nlp/PubTabNet/tree/master/src) | Speed |
+| ------------------- | ------ | ----------------------------------------------------------------------------------------------------- | ----- |
+| EDD<sup>[2]</sup>   | x      | 88.30%                                                                                                | x     |
+| TableRec-RARE(ours) | 71.73% | 93.88%                                                                                                | 779ms |
+| SLANet(ours)        | 76.31% | 95.89%                                                                                                | 766ms |
+
+性能指标解释如下：
+
+- Acc: 模型对每张图像里表格结构的识别准确率，错一个token就算错误。
+- TEDS: 模型对表格信息还原的准确度，此指标评价内容不仅包含表格结构，还包含表格内的文字内容。
+- Speed: 模型在CPU机器上，开启MKL的情况下，单张图片的推理速度。
+
+## 3. 效果演示
+
+![img](./images/table_ch_result1.jpg)
+
+![img](./images/table_ch_result2.jpg)
+
+![img](./images/table_ch_result3.jpg)
+
+## 4. 使用
+
+### 4.1 快速开始
+
+PP-Structure目前提供了中英文两种语言的表格识别模型，模型链接见 [models_list](../models_list.md)。也提供了whl包的形式方便快速使用，详见 [quickstart](../quick_start.md)。
+
+下面以中文表格识别模型为例，介绍如何识别一张表格。
+
+使用如下命令即可快速完成一张表格的识别。
+
+```bash linenums="1"
+cd PaddleOCR/ppstructure
+
+# 下载模型
+mkdir inference && cd inference
+# 下载PP-OCRv3文本检测模型并解压
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar && tar xf PP-OCRv3_mobile_det_infer.tar
+# 下载PP-OCRv3文本识别模型并解压
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar && tar xf PP-OCRv3_mobile_rec_infer.tar
+# 下载PP-StructureV2中文表格识别模型并解压
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/ch_ppstructure_mobile_v2.0_SLANet_infer.tar && tar xf ch_ppstructure_mobile_v2.0_SLANet_infer.tar
+cd ..
+# 执行表格识别
+python table/predict_table.py \
+    --det_model_dir=inference/PP-OCRv3_mobile_det_infer \
+    --rec_model_dir=inference/PP-OCRv3_mobile_rec_infer  \
+    --table_model_dir=inference/ch_ppstructure_mobile_v2.0_SLANet_infer \
+    --rec_char_dict_path=../ppocr/utils/ppocr_keys_v1.txt \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict_ch.txt \
+    --image_dir=docs/table/table.jpg \
+    --output=../output/table
+```
+
+运行完成后，每张图片的excel表格会保存到output字段指定的目录下，同时在该目录下回生产一个html文件，用于可视化查看单元格坐标和识别的表格。
+
+**NOTE**
+
+1. 如果想使用英文模型，需要在 [models_list](../../ppocr/model_list.md) 中下载英文文字检测识别模型和英文表格识别模型，同时替换`table_structure_dict_ch.txt`为`table_structure_dict.txt`即可。
+2. 如需使用TableRec-RARE模型，需要替换`table_structure_dict_ch.txt`为`table_structure_dict.txt`，同时参数`--merge_no_span_structure=False`
+
+### 4.2 模型训练、评估与推理
+
+文本检测模型的训练、评估和推理流程可参考 [detection](../../ppocr/model_train/detection.md)
+
+文本识别模型的训练、评估和推理流程可参考 [recognition](../../ppocr/model_train/recognition.md)
+
+表格识别模型的训练、评估和推理流程可参考 [table_recognition](../../ppstructure/model_train/train_table.md)
+
+### 4.3 计算TEDS
+
+表格使用 [TEDS(Tree-Edit-Distance-based Similarity)](https://github.com/ibm-aur-nlp/PubTabNet/tree/master/src) 作为模型的评估指标。在进行模型评估之前，需要将pipeline中的三个模型分别导出为inference模型(我们已经提供好)，还需要准备评估的gt， gt示例如下:
+
+```txt
+PMC5755158_010_01.png    <html><body><table><thead><tr><td></td><td><b>Weaning</b></td><td><b>Week 15</b></td><td><b>Off-test</b></td></tr></thead><tbody><tr><td>Weaning</td><td>–</td><td>–</td><td>–</td></tr><tr><td>Week 15</td><td>–</td><td>0.17 ± 0.08</td><td>0.16 ± 0.03</td></tr><tr><td>Off-test</td><td>–</td><td>0.80 ± 0.24</td><td>0.19 ± 0.09</td></tr></tbody></table></body></html>
+```
+
+gt每一行都由文件名和表格的html字符串组成，文件名和表格的html字符串之间使用`\t`分隔。
+
+也可使用如下命令，由标注文件生成评估的gt文件：
+
+```bash linenums="1"
+python3 ppstructure/table/convert_label2html.py --ori_gt_path /path/to/your_label_file --save_path /path/to/save_file
+```
+
+准备完成后使用如下命令进行评估，评估完成后会输出teds指标。
+
+```bash linenums="1"
+cd PaddleOCR/ppstructure
+python3 table/eval_table.py \
+    --det_model_dir=path/to/det_model_dir \
+    --rec_model_dir=path/to/rec_model_dir \
+    --table_model_dir=path/to/table_model_dir \
+    --image_dir=docs/table/table.jpg \
+    --rec_char_dict_path=../ppocr/utils/dict/table_dict.txt \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt \
+    --det_limit_side_len=736 \
+    --det_limit_type=min \
+    --gt_path=path/to/gt.txt
+```
+
+如使用英文表格识别模型在PubLatNet数据集上进行评估
+
+```bash linenums="1"
+cd PaddleOCR/ppstructure
+# 下载模型
+mkdir inference && cd inference
+# 下载基于PubTabNet数据集训练的文本检测模型并解压
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar
+# 下载基于PubTabNet数据集训练的文本识别模型并解压
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar
+# 下载基于PubTabNet数据集训练的表格识别模型并解压
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/en_ppstructure_mobile_v2.0_SLANet_infer.tar && tar xf en_ppstructure_mobile_v2.0_SLANet_infer.tar
+cd ..
+
+python3 table/eval_table.py \
+    --det_model_dir=inference/en_ppocr_mobile_v2.0_table_det_infer \
+    --rec_model_dir=inference/en_ppocr_mobile_v2.0_table_rec_infer \
+    --table_model_dir=inference/en_ppstructure_mobile_v2.0_SLANet_infer \
+    --image_dir=train_data/table/pubtabnet/val/ \
+    --rec_char_dict_path=../ppocr/utils/dict/table_dict.txt \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt \
+    --det_limit_side_len=736 \
+    --det_limit_type=min \
+    --rec_image_shape=3,32,320 \
+    --gt_path=path/to/gt.txt
+```
+
+将会输出
+
+```bash linenums="1"
+teds: 95.89
+```
+
+## 5. Reference
+
+1. <https://github.com/ibm-aur-nlp/PubTabNet>
+2. <https://arxiv.org/pdf/1911.10683>
diff --git a/docs/version2.x/ppstructure/model_train/training.en.md b/docs/version2.x/ppstructure/model_train/training.en.md
new file mode 100644
index 0000000..c402bb2
--- /dev/null
+++ b/docs/version2.x/ppstructure/model_train/training.en.md
@@ -0,0 +1,132 @@
+---
+comments: true
+---
+
+# Model Training
+
+This article will introduce the basic concepts that is necessary for model training and tuning.
+
+At the same time, it will briefly introduce the structure of the training data and how to prepare the data to fine-tune model in vertical scenes.
+
+## 1. Yml Configuration
+
+The PaddleOCR uses configuration files to control network training and evaluation parameters. In the configuration file, you can set the model, optimizer, loss function, and pre- and post-processing parameters of the model. PaddleOCR reads these parameters from the configuration file, and then builds a complete training process to train the model. Fine-tuning can also be completed by modifying the parameters in the configuration file, which is simple and convenient.
+
+For the complete configuration file description, please refer to [Configuration File](../../ppocr/blog/config.en.md)
+
+## 2. Basic Concepts
+
+During the model training process, some hyper-parameters can be manually specified to obtain the optimal result at the least cost. Different data volumes may require different hyper-parameters. When you want to fine-tune the model based on your own data, there are several parameter adjustment strategies for reference:
+
+### 2.1 Learning Rate
+
+The learning rate is one of the most important hyper-parameters for training neural networks. It represents the step length of the gradient moving towards the optimal solution of the loss function in each iteration.
+A variety of learning rate update strategies are provided by PaddleOCR, which can be specified in configuration files. For example,
+
+```yaml linenums="1"
+Optimizer:
+  ...
+  lr:
+    name: Piecewise
+    decay_epochs : [700, 800]
+    values : [0.001, 0.0001]
+    warmup_epoch: 5
+```
+
+`Piecewise` stands for piece-wise constant attenuation. Different learning rates are specified in different learning stages, and the learning rate stay the same in each stage.
+
+`warmup_epoch` means that in the first 5 epochs, the learning rate will be increased gradually from 0 to base_lr. For all strategies, please refer to the code [learning_rate.py](../../../../ppocr/optimizer/learning_rate.py).
+
+### 2.2 Regularization
+
+Regularization can effectively avoid algorithm over-fitting. PaddleOCR provides L1 and L2 regularization methods.
+L1 and L2 regularization are the most widely used regularization methods.
+L1 regularization adds a regularization term to the objective function to reduce the sum of absolute values of the parameters;
+while in L2 regularization, the purpose of adding a regularization term is to reduce the sum of squared parameters.
+The configuration method is as follows:
+
+```yaml linenums="1"
+Optimizer:
+  ...
+  regularizer:
+    name: L2
+    factor: 2.0e-05
+```
+
+### 2.3 Evaluation Indicators
+
+(1) Detection stage: First, evaluate according to the IOU of the detection frame and the labeled frame. If the IOU is greater than a certain threshold, it is judged that the detection is accurate. Here, the detection frame and the label frame are different from the general general target detection frame, and they are represented by polygons. Detection accuracy: the percentage of the correct detection frame number in all detection frames is mainly used to judge the detection index. Detection recall rate: the percentage of correct detection frames in all marked frames, which is mainly an indicator of missed detection.
+
+(2) Recognition stage: Character recognition accuracy, that is, the ratio of correctly recognized text lines to the number of marked text lines. Only the entire line of text recognition pairs can be regarded as correct recognition.
+
+(3) End-to-end statistics: End-to-end recall rate: accurately detect and correctly identify the proportion of text lines in all labeled text lines; End-to-end accuracy rate: accurately detect and correctly identify the number of text lines in the detected text lines The standard for accurate detection is that the IOU of the detection box and the labeled box is greater than a certain threshold, and the text in the correctly identified detection box is the same as the labeled text.
+
+## 3. Data and Vertical Scenes
+
+### 3.1 Training Data
+
+The current open source models, data sets and magnitudes are as follows:
+
+- Detection:
+  - English data set, ICDAR2015
+  - Chinese data set, LSVT street view data set training data 3w pictures
+
+- Identification:
+  - English data set, MJSynth and SynthText synthetic data, the data volume is tens of millions.
+  - Chinese data set, LSVT street view data set crops the image according to the truth value, and performs position calibration, a total of 30w images. In addition, based on the LSVT corpus, 500w of synthesized data.
+  - Small language data set, using different corpora and fonts, respectively generated 100w synthetic data set, and using ICDAR-MLT as the verification set.
+
+Among them, the public data sets are all open source, users can search and download by themselves, or refer to [Chinese data set](../../../datasets/datasets.en.md), synthetic data is not open source, users can use open source synthesis tools to synthesize by themselves. Synthesis tools include [text_renderer](https://github.com/Sanster/text_renderer), [SynthText](https://github.com/ankush-me/SynthText), [TextRecognitionDataGenerator](https://github.com/Belval/TextRecognitionDataGenerator) etc.
+
+### 3.2 Vertical Scene
+
+PaddleOCR mainly focuses on general OCR. If you have vertical requirements, you can use PaddleOCR + vertical data to train yourself;
+If there is a lack of labeled data, or if you do not want to invest in research and development costs, it is recommended to directly call the open API, which covers some of the more common vertical categories.
+
+### 3.3 Build Your Own Dataset
+
+There are several experiences for reference when constructing the data set:
+
+(1) The amount of data in the training set:
+
+a. The data required for detection is relatively small. For Fine-tune based on the PaddleOCR model, 500 sheets are generally required to achieve good results.
+
+b. Recognition is divided into English and Chinese. Generally, English scenarios require hundreds of thousands of data to achieve good results, while Chinese requires several million or more.
+
+(2) When the amount of training data is small, you can try the following three ways to get more data:
+
+a. Manually collect more training data, the most direct and effective way.
+
+b. Basic image processing or transformation based on PIL and opencv. For example, the three modules of ImageFont, Image, ImageDraw in PIL write text into the background, opencv's rotating affine transformation, Gaussian filtering and so on.
+
+c. Use data generation algorithms to synthesize data, such as algorithms such as pix2pix.
+
+## 4. FAQ
+
+**Q**: How to choose a suitable network input shape when training CRNN recognition?
+
+    A: The general height is 32, the longest width is selected, there are two methods:
+    
+    (1) Calculate the aspect ratio distribution of training sample images. The selection of the maximum aspect ratio considers 80% of the training samples.
+    
+    (2) Count the number of texts in training samples. The selection of the longest number of characters considers the training sample that satisfies 80%. Then the aspect ratio of Chinese characters is approximately considered to be 1, and that of English is 3:1, and the longest width is estimated.
+
+**Q**: During the recognition training, the accuracy of the training set has reached 90, but the accuracy of the verification set has been kept at 70, what should I do?
+
+    A: If the accuracy of the training set is 90 and the test set is more than 70, it should be over-fitting. There are two methods to try:
+    
+    (1) Add more augmentation methods or increase the [probability] of augmented prob (https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppocr/data/imaug/rec_img_aug.py#L341), The default is 0.4.
+    
+    (2) Increase the [l2 dcay value] of the system (https://github.com/PaddlePaddle/PaddleOCR/blob/a501603d54ff5513fc4fc760319472e59da25424/configs/rec/ch_ppocr_v1.1/rec_chinese_lite_train_v1.1.yml#L47)
+
+**Q**: When the recognition model is trained, loss can drop normally, but acc is always 0
+
+    A: It is normal for the acc to be 0 at the beginning of the recognition model training, and the indicator will come up after a longer training period.
+
+***
+
+Click the following links for detailed training tutorial:
+
+- [text detection model training](../../ppocr/model_train/detection.en.md)
+- [text recognition model training](../../ppocr/model_train/recognition.en.md)
+- [text direction classification model training](../../ppocr/model_train/angle_class.en.md)
diff --git a/docs/version2.x/ppstructure/model_train/training.md b/docs/version2.x/ppstructure/model_train/training.md
new file mode 100644
index 0000000..c761ef0
--- /dev/null
+++ b/docs/version2.x/ppstructure/model_train/training.md
@@ -0,0 +1,128 @@
+---
+comments: true
+---
+
+# PP-OCR模型训练
+
+本文将介绍模型训练时需掌握的基本概念，和训练时的调优方法。
+
+同时会简单介绍PaddleOCR模型训练数据的组成部分，以及如何在垂类场景中准备数据finetune模型。
+
+## 1. 配置文件说明
+
+PaddleOCR模型使用配置文件管理网络训练、评估的参数。在配置文件中，可以设置组建模型、优化器、损失函数、模型前后处理的参数，PaddleOCR从配置文件中读取到这些参数，进而组建出完整的训练流程，完成模型训练，在需要对模型进行优化的时，可以通过修改配置文件中的参数完成配置，使用简单且方便修改。
+
+完整的配置文件说明可以参考[配置文件](../../ppocr/blog/config.md)
+
+## 2. 基本概念
+
+模型训练过程中需要手动调整一些超参数，帮助模型以最小的代价获得最优指标。不同的数据量可能需要不同的超参，当您希望在自己的数据上finetune或对模型效果调优时，有以下几个参数调整策略可供参考：
+
+### 2.1 学习率
+
+学习率是训练神经网络的重要超参数之一，它代表在每一次迭代中梯度向损失函数最优解移动的步长。
+在PaddleOCR中提供了多种学习率更新策略,可以通过配置文件修改，例如：
+
+```yaml linenums="1"
+Optimizer:
+  ...
+  lr:
+    name: Piecewise
+    decay_epochs : [700, 800]
+    values : [0.001, 0.0001]
+    warmup_epoch: 5
+```
+
+Piecewise 代表分段常数衰减，在不同的学习阶段指定不同的学习率，在每段内学习率相同。
+warmup_epoch 代表在前5个epoch中，学习率将逐渐从0增加到base_lr。全部策略可以参考代码 [learning_rate.py](../../../../ppocr/optimizer/learning_rate.py) 。
+
+### 2.2 正则化
+
+正则化可以有效的避免算法过拟合，PaddleOCR中提供了L1、L2正则方法，L1 和 L2 正则化是最常用的正则化方法。L1 正则化向目标函数添加正则化项，以减少参数的绝对值总和；而 L2 正则化中，添加正则化项的目的在于减少参数平方的总和。配置方法如下：
+
+```yaml linenums="1"
+Optimizer:
+  ...
+  regularizer:
+    name: L2
+    factor: 2.0e-05
+```
+
+### 2.3 评估指标
+
+（1）检测阶段：先按照检测框和标注框的IOU评估，IOU大于某个阈值判断为检测准确。这里检测框和标注框不同于一般的通用目标检测框，是采用多边形进行表示。检测准确率：正确的检测框个数在全部检测框的占比，主要是判断检测指标。检测召回率：正确的检测框个数在全部标注框的占比，主要是判断漏检的指标。
+
+（2）识别阶段： 字符识别准确率，即正确识别的文本行占标注的文本行数量的比例，只有整行文本识别对才算正确识别。
+
+（3）端到端统计： 端对端召回率：准确检测并正确识别文本行在全部标注文本行的占比； 端到端准确率：准确检测并正确识别文本行在 检测到的文本行数量 的占比； 准确检测的标准是检测框与标注框的IOU大于某个阈值，正确识别的检测框中的文本与标注的文本相同。
+
+## 3. 数据与垂类场景
+
+### 3.1 训练数据
+
+目前开源的模型，数据集和量级如下：
+
+- 检测：
+  - 英文数据集，ICDAR2015
+  - 中文数据集，LSVT街景数据集训练数据3w张图片
+
+- 识别：
+  - 英文数据集，MJSynth和SynthText合成数据，数据量上千万。
+  - 中文数据集，LSVT街景数据集根据真值将图crop出来，并进行位置校准，总共30w张图像。此外基于LSVT的语料，合成数据500w。
+  - 小语种数据集，使用不同语料和字体，分别生成了100w合成数据集，并使用ICDAR-MLT作为验证集。
+
+其中，公开数据集都是开源的，用户可自行搜索下载，也可参考[中文数据集](../../../datasets/datasets.md)，合成数据暂不开源，用户可使用开源合成工具自行合成，可参考的合成工具包括[text_renderer](https://github.com/Sanster/text_renderer) 、[SynthText](https://github.com/ankush-me/SynthText) 、[TextRecognitionDataGenerator](https://github.com/Belval/TextRecognitionDataGenerator) 等。
+
+### 3.2 垂类场景
+
+PaddleOCR主要聚焦通用OCR，如果有垂类需求，您可以用PaddleOCR+垂类数据自己训练；
+如果缺少带标注的数据，或者不想投入研发成本，建议直接调用开放的API，开放的API覆盖了目前比较常见的一些垂类。
+
+### 3.3 自己构建数据集
+
+在构建数据集时有几个经验可供参考：
+
+（1） 训练集的数据量：
+
+a. 检测需要的数据相对较少，在PaddleOCR模型的基础上进行Fine-tune，一般需要500张可达到不错的效果。
+
+b. 识别分英文和中文，一般英文场景需要几十万数据可达到不错的效果，中文则需要几百万甚至更多。
+
+（2）当训练数据量少时，可以尝试以下三种方式获取更多的数据：
+
+a. 人工采集更多的训练数据，最直接也是最有效的方式。
+
+b. 基于PIL和opencv基本图像处理或者变换。例如PIL中ImageFont, Image, ImageDraw三个模块将文字写到背景中，opencv的旋转仿射变换，高斯滤波等。
+
+c. 利用数据生成算法合成数据，例如pix2pix或[StyleText](https://github.com/PFCCLab/StyleText)等算法。
+
+## 4. 常见问题
+
+**Q**：训练CRNN识别时，如何选择合适的网络输入shape？
+
+    A：一般高度采用32，最长宽度的选择，有两种方法：
+    
+    （1）统计训练样本图像的宽高比分布。最大宽高比的选取考虑满足80%的训练样本。
+    
+    （2）统计训练样本文字数目。最长字符数目的选取考虑满足80%的训练样本。然后中文字符长宽比近似认为是1，英文认为3：1，预估一个最长宽度。
+
+**Q**：识别训练时，训练集精度已经到达90了，但验证集精度一直在70，涨不上去怎么办？
+
+    A：训练集精度90，测试集70多的话，应该是过拟合了，有两个可尝试的方法：
+    
+    （1）加入更多的增广方式或者调大增广prob的[概率](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppocr/data/imaug/rec_img_aug.py#L341)，默认为0.4。
+    
+    （2）调大系统的[l2 dcay值](https://github.com/PaddlePaddle/PaddleOCR/blob/a501603d54ff5513fc4fc760319472e59da25424/configs/rec/ch_ppocr_v1.1/rec_chinese_lite_train_v1.1.yml#L47)
+
+**Q**: 识别模型训练时，loss能正常下降，但acc一直为0
+
+    A：识别模型训练初期acc为0是正常的，多训一段时间指标就上来了。
+
+***
+
+具体的训练教程可点击下方链接跳转：
+
+- [文本检测模型训练](../../ppocr/model_train/detection.md)
+- [文本识别模型训练](../../ppocr/model_train/recognition.md)
+- [文本方向分类器训练](../../ppocr/model_train/angle_class.md)
+- [知识蒸馏](../../ppocr/model_compress/knowledge_distillation.md)
diff --git a/docs/version2.x/ppstructure/models_list.en.md b/docs/version2.x/ppstructure/models_list.en.md
new file mode 100644
index 0000000..fa469e6
--- /dev/null
+++ b/docs/version2.x/ppstructure/models_list.en.md
@@ -0,0 +1,57 @@
+---
+comments: true
+---
+
+# PP-Structure Model list
+
+## 1. Layout Analysis
+
+|model name| description                                                                                                                                             | inference model size                                                                                                                         |download|dict path|
+| --- |----| --- | --- | --- |
+| picodet_lcnet_x1_0_fgd_layout | The layout analysis English model trained on the PubLayNet dataset based on PicoDet LCNet_x1_0 and FGD . the model can recognition 5 types of areas such as **Text, Title, Table, Picture and List** | 9.7M | [inference model](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout.pdparams) | [PubLayNet dict](../../../ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt) |
+| ppyolov2_r50vd_dcn_365e_publaynet | The layout analysis English model trained on the PubLayNet dataset based on PP-YOLOv2 | 221.0M | [inference_moel](https://paddle-model-ecology.bj.bcebos.com/model/layout-parser/ppyolov2_r50vd_dcn_365e_publaynet.tar) / [trained model](https://paddle-model-ecology.bj.bcebos.com/model/layout-parser/ppyolov2_r50vd_dcn_365e_publaynet_pretrained.pdparams) | same as above |
+| picodet_lcnet_x1_0_fgd_layout_cdla | The layout analysis Chinese model trained on the CDLA dataset, the model can recognition 10 types of areas such as **Table、Figure、Figure caption、Table、Table caption、Header、Footer、Reference、Equation** | 9.7M | [inference model](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_cdla_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_cdla.pdparams) | [CDLA dict](../../../ppocr/utils/dict/layout_dict/layout_cdla_dict.txt) |
+| picodet_lcnet_x1_0_fgd_layout_table | The layout analysis model trained on the table dataset, the model can detect tables in Chinese and English documents                     | 9.7M                                                  | [inference model](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_table_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_table.pdparams) | [Table dict](../../../ppocr/utils/dict/layout_dict/layout_table_dict.txt) |
+| ppyolov2_r50vd_dcn_365e_tableBank_word | The layout analysis model trained on the TableBank Word dataset based on PP-YOLOv2, the model can detect  tables  in English documents | 221.0M | [inference model](https://paddle-model-ecology.bj.bcebos.com/model/layout-parser/ppyolov2_r50vd_dcn_365e_tableBank_word.tar) | same as above |
+| ppyolov2_r50vd_dcn_365e_tableBank_latex | The layout analysis model trained on the TableBank Latex dataset based on PP-YOLOv2, the model can detect  tables  in English documents | 221.0M                 | [inference model](https://paddle-model-ecology.bj.bcebos.com/model/layout-parser/ppyolov2_r50vd_dcn_365e_tableBank_latex.tar) | same as above |
+
+## 2. OCR and Table Recognition
+
+### 2.1 OCR
+
+|model name| description | inference model size |download|
+| --- |---|---| --- |
+|en_ppocr_mobile_v2.0_table_det| Text detection model of English table scenes trained on PubTabNet dataset | 4.7M                |[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_det_train.tar) |
+|en_ppocr_mobile_v2.0_table_rec| Text recognition model of English table scenes trained on PubTabNet dataset | 6.9M                |[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_rec_train.tar) |
+
+If you need to use other OCR models, you can download the model in [PP-OCR model_list](../ppocr/model_list.en.md) or use the model you trained yourself to configure to `det_model_dir`, `rec_model_dir` field.
+
+### 2.2 Table Recognition
+
+|model| description |inference model size|download|
+| --- |-----| --- | --- |
+|en_ppocr_mobile_v2.0_table_structure| English table recognition model trained on PubTabNet dataset based on TableRec-RARE |6.8M|[inference model](https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_structure_train.tar) |
+|en_ppstructure_mobile_v2.0_SLANet|English table recognition model trained on PubTabNet dataset based on SLANet|9.2M|[inference model](https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/en_ppstructure_mobile_v2.0_SLANet_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/en_ppstructure_mobile_v2.0_SLANet_train.tar) |
+|ch_ppstructure_mobile_v2.0_SLANet|Chinese table recognition model based on SLANet|9.3M|[inference model](https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/ch_ppstructure_mobile_v2.0_SLANet_infer.tar) / [trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/ch_ppstructure_mobile_v2.0_SLANet_train.tar) |
+
+## 3. KIE
+
+On XFUND_zh dataset, Accuracy and time cost of different models on V100 GPU are as follows.
+
+|Model|Backbone|Task|Config|Hmean|Time cost(ms)|Download link|
+| --- | --- |  --- | --- | --- | --- |--- |
+|VI-LayoutXLM| VI-LayoutXLM-base | SER | [ser_vi_layoutxlm_xfund_zh_udml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh_udml.yml)|**93.19%**| 15.49| [trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar)|
+|LayoutXLM| LayoutXLM-base | SER | [ser_layoutxlm_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/ser_layoutxlm_xfund_zh.yml)|90.38%| 19.49 |[trained model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar)|
+|LayoutLM| LayoutLM-base | SER | [ser_layoutlm_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/ser_layoutlm_xfund_zh.yml)|77.31%|-|[trained model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLM_xfun_zh.tar)|
+|LayoutLMv2| LayoutLMv2-base | SER | [ser_layoutlmv2_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/ser_layoutlmv2_xfund_zh.yml)|85.44%|31.46|[trained model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLMv2_xfun_zh.tar)|
+|VI-LayoutXLM| VI-LayoutXLM-base | RE | [re_vi_layoutxlm_xfund_zh_udml.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh_udml.yml)|**83.92%**|15.49|[trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar)|
+|LayoutXLM| LayoutXLM-base | RE | [re_layoutxlm_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/re_layoutxlm_xfund_zh.yml)|74.83%|19.49|[trained model](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar)|
+|LayoutLMv2| LayoutLMv2-base | RE | [re_layoutlmv2_xfund_zh.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/layoutlm_series/re_layoutlmv2_xfund_zh.yml)|67.77%|31.46|[trained model](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutLMv2_xfun_zh.tar)|
+
+* Note: The above time cost information just considers inference time without preprocess or postprocess, test environment: `V100 GPU + CUDA 10.2 + CUDNN 8.1.1 + TRT 7.2.3.4`
+
+On wildreceipt dataset, the algorithm result is as follows:
+
+|Model|Backbone|Config|Hmean|Download link|
+| --- | --- | --- | --- | --- |
+|SDMGR|VGG6|[configs/kie/sdmgr/kie_unet_sdmgr.yml](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/configs/kie/sdmgr/kie_unet_sdmgr.yml)|86.70%|[trained model](https://paddleocr.bj.bcebos.com/dygraph_v2.1/kie/kie_vgg16.tar)|
diff --git a/docs/version2.x/ppstructure/models_list.md b/docs/version2.x/ppstructure/models_list.md
new file mode 100644
index 0000000..b242f00
--- /dev/null
+++ b/docs/version2.x/ppstructure/models_list.md
@@ -0,0 +1,57 @@
+---
+comments: true
+---
+
+# PP-Structure 系列模型列表
+
+## 1. 版面分析模型
+
+|模型名称|模型简介|推理模型大小|下载地址|dict path|
+| --- | --- | --- | --- | --- |
+| picodet_lcnet_x1_0_fgd_layout | 基于PicoDet LCNet_x1_0和FGD蒸馏在PubLayNet 数据集训练的英文版面分析模型，可以划分**文字、标题、表格、图片以及列表**5类区域 | 9.7M | [推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout.pdparams) | [PubLayNet dict](../../../ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt) |
+| ppyolov2_r50vd_dcn_365e_publaynet | 基于PP-YOLOv2在PubLayNet数据集上训练的英文版面分析模型 | 221.0M | [推理模型](https://paddle-model-ecology.bj.bcebos.com/model/layout-parser/ppyolov2_r50vd_dcn_365e_publaynet.tar) / [训练模型](https://paddle-model-ecology.bj.bcebos.com/model/layout-parser/ppyolov2_r50vd_dcn_365e_publaynet_pretrained.pdparams) | 同上 |
+| picodet_lcnet_x1_0_fgd_layout_cdla | CDLA数据集训练的中文版面分析模型，可以划分为**表格、图片、图片标题、表格、表格标题、页眉、脚本、引用、公式**10类区域 | 9.7M | [推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_cdla_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_cdla.pdparams) | [CDLA dict](../../../ppocr/utils/dict/layout_dict/layout_cdla_dict.txt) |
+| picodet_lcnet_x1_0_fgd_layout_table | 表格数据集训练的版面分析模型，支持中英文文档表格区域的检测 | 9.7M | [推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_table_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_table.pdparams) | [Table dict](../../../ppocr/utils/dict/layout_dict/layout_table_dict.txt) |
+| ppyolov2_r50vd_dcn_365e_tableBank_word | 基于PP-YOLOv2在TableBank Word 数据集训练的版面分析模型，支持英文文档表格区域的检测 | 221.0M | [推理模型](https://paddle-model-ecology.bj.bcebos.com/model/layout-parser/ppyolov2_r50vd_dcn_365e_tableBank_word.tar) | 同上 |
+| ppyolov2_r50vd_dcn_365e_tableBank_latex | 基于PP-YOLOv2在TableBank Latex数据集训练的版面分析模型，支持英文文档表格区域的检测 | 221.0M | [推理模型](https://paddle-model-ecology.bj.bcebos.com/model/layout-parser/ppyolov2_r50vd_dcn_365e_tableBank_latex.tar) | 同上 |
+
+## 2. OCR和表格识别模型
+
+### 2.1 OCR
+
+|模型名称|模型简介|推理模型大小|下载地址|
+| --- | --- | --- | --- |
+|en_ppocr_mobile_v2.0_table_det|PubTabNet数据集训练的英文表格场景的文字检测|4.7M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_det_train.tar) |
+|en_ppocr_mobile_v2.0_table_rec|PubTabNet数据集训练的英文表格场景的文字识别|6.9M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_rec_train.tar) |
+
+如需要使用其他OCR模型，可以在 [PP-OCR model_list](../ppocr/model_list.md) 下载模型或者使用自己训练好的模型配置到 `det_model_dir`, `rec_model_dir`两个字段即可。
+
+### 2.2 表格识别模型
+
+|模型名称|模型简介|推理模型大小|下载地址|
+| --- | --- | --- | --- |
+|en_ppocr_mobile_v2.0_table_structure|基于TableRec-RARE在PubTabNet数据集上训练的英文表格识别模型|6.8M|[推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_structure_train.tar) |
+|en_ppstructure_mobile_v2.0_SLANet|基于SLANet在PubTabNet数据集上训练的英文表格识别模型|9.2M|[推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/en_ppstructure_mobile_v2.0_SLANet_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/en_ppstructure_mobile_v2.0_SLANet_train.tar) |
+|ch_ppstructure_mobile_v2.0_SLANet|基于SLANet的中文表格识别模型|9.3M|[推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/ch_ppstructure_mobile_v2.0_SLANet_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/ch_ppstructure_mobile_v2.0_SLANet_train.tar) |
+
+## 3. KIE模型
+
+在XFUND_zh数据集上，不同模型的精度与V100 GPU上速度信息如下所示。
+
+|模型名称|模型简介 | 推理模型大小| 精度(hmean) | 预测耗时(ms) | 下载地址|
+| --- | --- | --- |--- |--- | --- |
+|ser_VI-LayoutXLM_xfund_zh|基于VI-LayoutXLM在xfund中文数据集上训练的SER模型|1.1G| 93.19% | 15.49 | [推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar) |
+|re_VI-LayoutXLM_xfund_zh|基于VI-LayoutXLM在xfund中文数据集上训练的RE模型|1.1G| 83.92% | 15.49 |[推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar) |
+|ser_LayoutXLM_xfund_zh|基于LayoutXLM在xfund中文数据集上训练的SER模型|1.4G| 90.38% | 19.49 |[推理模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar) |
+|re_LayoutXLM_xfund_zh|基于LayoutXLM在xfund中文数据集上训练的RE模型|1.4G| 74.83% | 19.49 |[推理模型](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar) |
+|ser_LayoutLMv2_xfund_zh|基于LayoutLMv2在xfund中文数据集上训练的SER模型|778.0M| 85.44% | 31.46 |[推理模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLMv2_xfun_zh_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLMv2_xfun_zh.tar) |
+|re_LayoutLMv2_xfund_zh|基于LayoutLMv2在xfun中文数据集上训练的RE模型|765.0M| 67.77% | 31.46 |[推理模型 coming soon]() / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutLMv2_xfun_zh.tar) |
+|ser_LayoutLM_xfund_zh|基于LayoutLM在xfund中文数据集上训练的SER模型|430.0M| 77.31% | - |[推理模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLM_xfun_zh_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLM_xfun_zh.tar) |
+
+* 注：上述预测耗时信息仅包含了inference模型的推理耗时，没有统计预处理与后处理耗时，测试环境为`V100 GPU + CUDA 10.2 + CUDNN 8.1.1 + TRT 7.2.3.4`。
+
+在wildreceipt数据集上，SDMGR模型精度与下载地址如下所示。
+
+|模型名称|模型简介|模型大小|精度|下载地址|
+| --- | --- | --- |--- | --- |
+|SDMGR|关键信息提取模型|78.0M| 86.70% | [推理模型 coming soon]() / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/kie/kie_vgg16.tar)|
diff --git a/docs/version2.x/ppstructure/overview.en.md b/docs/version2.x/ppstructure/overview.en.md
new file mode 100644
index 0000000..2e38a94
--- /dev/null
+++ b/docs/version2.x/ppstructure/overview.en.md
@@ -0,0 +1,104 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# PP-Structure
+
+## 1. Introduction
+
+PP-Structure is an intelligent document analysis system developed by the PaddleOCR team, which aims to help developers better complete tasks related to document understanding such as layout analysis and table recognition.
+
+The pipeline of PP-StructureV2 system is shown below. The document image first passes through the image direction correction module to identify the direction of the entire image and complete the direction correction. Then, two tasks of layout information analysis and key information extraction can be completed.
+
+- In the layout analysis task, the image first goes through the layout analysis model to divide the image into different areas such as text, table, and figure, and then analyze these areas separately. For example, the table area is sent to the form recognition module for structured recognition, and the text area is sent to the OCR engine for text recognition. Finally, the layout recovery module restores it to a word or pdf file with the same layout as the original image;
+- In the key information extraction task, the OCR engine is first used to extract the text content, and then the SER(semantic entity recognition) module obtains the semantic entities in the image, and finally the RE(relationship extraction) module obtains the correspondence between the semantic entities, thereby extracting the required key information.
+
+![img](./images/195265734-6f4b5a7f-59b1-4fcc-af6d-89afc9bd51e1-20240705140834325.jpg)
+
+More technical details: 👉 [PP-StructureV2 Technical Report](https://arxiv.org/abs/2210.05391)
+
+PP-StructureV2 supports independent use or flexible collocation of each module. For example, you can use layout analysis alone or table recognition alone. Click the corresponding link below to get the tutorial for each independent module:
+
+- [Layout Analysis](./model_train/train_layout.en.md)
+- [Table Recognition](./model_train/train_table.en.md)
+- [Key Information Extraction](./model_train/train_kie.en.md)
+- [Layout Recovery](./model_train/recovery_to_doc.en.md)
+
+## 2. Features
+
+The main features of PP-StructureV2 are as follows:
+
+- Support layout analysis of documents in the form of images/pdfs, which can be divided into areas such as **text, titles, tables, figures, formulas, etc.**;
+- Support common Chinese and English **table detection** tasks;
+- Support structured table recognition, and output the final result to **Excel file**;
+- Support multimodal-based Key Information Extraction (KIE) tasks - **Semantic Entity Recognition** (SER) and **Relation Extraction** (RE);
+- Support **layout recovery**, that is, restore the document in word or pdf format with the same layout as the original image;
+- Support customized training and multiple inference deployment methods such as python whl package quick start;
+- Connect with the semi-automatic data labeling tool PPOCRLabel, which supports the labeling of layout analysis, table recognition, and SER.
+
+## 3. Results
+
+PP-StructureV2 supports the independent use or flexible collocation of each module. For example, layout analysis can be used alone, or table recognition can be used alone. Only the visualization effects of several representative usage methods are shown here.
+
+### 3.1 Layout analysis and table recognition
+
+The figure shows the pipeline of layout analysis + table recognition. The image is first divided into four areas of image, text, title and table by layout analysis, and then OCR detection and recognition is performed on the three areas of image, text and title, and the table is performed table recognition, where the image will also be stored for use.
+
+![img](./images/ppstructure.gif)
+
+### 3.1.1 Layout recognition returns the coordinates of a single word
+
+The following figure shows the result of layout analysis on single word，please refer to the [doc](./blog/return_word_pos.en.md).
+
+![show_0_mdf_v2](./images/799450d4-d2c5-4b61-b490-e160dc0f515c.jpeg)
+
+### 3.2 Layout recovery
+
+The following figure shows the effect of layout recovery based on the results of layout analysis and table recognition in the previous section.
+
+![img](./images/recovery.jpg)
+
+### 3.3 KIE
+
+- SER
+
+Different colored boxes in the figure represent different categories.
+
+![img](./images/185539141-68e71c75-5cf7-4529-b2ca-219d29fa5f68-20240705093932704.jpg)
+
+![img](./images/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808-20240705094001639.jpg)
+
+![img](./images/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a-20240705094013236.png)
+
+![img](./images/197464552-69de557f-edff-4c7f-acbf-069df1ba097f.png)
+
+![img](./images/186095702-9acef674-12af-4d09-97fc-abf4ab32600e.png)
+
+- RE
+
+In the figure, the red box represents `Question`, the blue box represents `Answer`, and `Question` and `Answer` are connected by green lines.
+
+![img](./images/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb-20240705094037073.jpg)
+
+![img](./images/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f-20240705094043151.jpg)
+
+![img](./images/186094813-3a8e16cc-42e5-4982-b9f4-0134dfb5688d.png)
+
+![img](./images/186095641-5843b4da-34d7-4c1c-943a-b1036a859fe3.png)
+
+## 4. Quick start
+
+Start from [Quick Start](./quick_start.en.md).
+
+## 5. Model List
+
+Some tasks need to use both the structured analysis models and the OCR models. For example, the table recognition task needs to use the table recognition model for structured analysis, and the OCR model to recognize the text in the table. Please select the appropriate models according to your specific needs.
+
+For structural analysis related model downloads, please refer to:
+
+- [PP-Structure Model Zoo](./models_list.en.md)
+
+For OCR related model downloads, please refer to:
+
+- [PP-OCR Model Zoo](../ppocr/model_list.en.md)
diff --git a/docs/version2.x/ppstructure/overview.md b/docs/version2.x/ppstructure/overview.md
new file mode 100644
index 0000000..75768d6
--- /dev/null
+++ b/docs/version2.x/ppstructure/overview.md
@@ -0,0 +1,104 @@
+---
+typora-copy-images-to: images
+comments: true
+---
+
+# PP-Structure 文档分析
+
+## 1. 简介
+
+PP-Structure是PaddleOCR团队自研的智能文档分析系统，旨在帮助开发者更好的完成版面分析、表格识别等文档理解相关任务。
+
+PP-StructureV2系统流程图如下所示，文档图像首先经过图像矫正模块，判断整图方向并完成转正，随后可以完成版面信息分析与关键信息抽取2类任务。
+
+- 版面分析任务中，图像首先经过版面分析模型，将图像划分为文本、表格、图像等不同区域，随后对这些区域分别进行识别，如，将表格区域送入表格识别模块进行结构化识别，将文本区域送入OCR引擎进行文字识别，最后使用版面恢复模块将其恢复为与原始图像布局一致的word或者pdf格式的文件；
+- 关键信息抽取任务中，首先使用OCR引擎提取文本内容，然后由语义实体识别模块获取图像中的语义实体，最后经关系抽取模块获取语义实体之间的对应关系，从而提取需要的关键信息。
+
+![img](./images/195265734-6f4b5a7f-59b1-4fcc-af6d-89afc9bd51e1-20240705140834325.jpg)
+
+更多技术细节：👉 PP-StructureV2技术报告 [中文版](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.9/ppstructure/docs/PP-StructureV2_introduction.md)，[英文版](https://arxiv.org/abs/2210.05391)。
+
+PP-StructureV2支持各个模块独立使用或灵活搭配，如，可以单独使用版面分析，或单独使用表格识别，点击下面相应链接获取各个独立模块的使用教程：
+
+- [版面分析](model_train/train_layout.md)
+- [表格识别](model_train/train_table.md)
+- [关键信息抽取](model_train/train_kie.md)
+- [版面复原](model_train/recovery_to_doc.md)
+
+## 2. 特性
+
+PP-StructureV2的主要特性如下：
+
+- 支持对图片/pdf形式的文档进行版面分析，可以划分**文字、标题、表格、图片、公式等**区域；
+- 支持通用的中英文**表格检测**任务；
+- 支持表格区域进行结构化识别，最终结果输出**Excel文件**；
+- 支持基于多模态的关键信息抽取(Key Information Extraction，KIE)任务-**语义实体识别**(Semantic Entity Recognition，SER)和**关系抽取**(Relation Extraction，RE)；
+- 支持**版面复原**，即恢复为与原始图像布局一致的word或者pdf格式的文件；
+- 支持自定义训练及python whl包调用等多种推理部署方式，简单易用；
+- 与半自动数据标注工具PPOCRLabel打通，支持版面分析、表格识别、SER三种任务的标注。
+
+## 3. 效果展示
+
+PP-StructureV2支持各个模块独立使用或灵活搭配，如，可以单独使用版面分析，或单独使用表格识别，这里仅展示几种代表性使用方式的可视化效果。
+
+### 3.1 版面分析和表格识别
+
+下图展示了版面分析+表格识别的整体流程，图片先有版面分析划分为图像、文本、标题和表格四种区域，然后对图像、文本和标题三种区域进行OCR的检测识别，对表格进行表格识别，其中图像还会被存储下来以便使用。
+
+![img](./images/ppstructure.gif)
+
+### 3.1.1 版面识别返回单字坐标
+
+下图展示了基于上一节版面分析对文字进行定位的效果， 可参考[文档](blog/return_word_pos.md)。
+
+![show_0_mdf_v2](./images/799450d4-d2c5-4b61-b490-e160dc0f515c.jpeg)
+
+### 3.2 版面恢复
+
+下图展示了基于上一节版面分析和表格识别的结果进行版面恢复的效果。
+
+![img](./images/recovery.jpg)
+
+### 3.3 关键信息抽取
+
+- SER
+
+图中不同颜色的框表示不同的类别。
+
+![img](./images/185539141-68e71c75-5cf7-4529-b2ca-219d29fa5f68-20240705093932704.jpg)
+
+![img](./images/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808-20240705094001639.jpg)
+
+![img](./images/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a-20240705094013236.png)
+
+![img](./images/197464552-69de557f-edff-4c7f-acbf-069df1ba097f.png)
+
+![img](./images/186095702-9acef674-12af-4d09-97fc-abf4ab32600e.png)
+
+- RE
+
+图中红色框表示`问题`，蓝色框表示`答案`，`问题`和`答案`之间使用绿色线连接。
+
+![img](./images/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb-20240705094037073.jpg)
+
+![img](./images/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f-20240705094043151.jpg)
+
+![img](./images/186094813-3a8e16cc-42e5-4982-b9f4-0134dfb5688d.png)
+
+![img](./images/186095641-5843b4da-34d7-4c1c-943a-b1036a859fe3.png)
+
+## 4. 快速体验
+
+请参考[快速使用](./quick_start.md)教程。
+
+## 5. 模型库
+
+部分任务需要同时用到结构化分析模型和OCR模型，如表格识别需要使用表格识别模型进行结构化解析，同时也要用到OCR模型对表格内的文字进行识别，请根据具体需求选择合适的模型。
+
+结构化分析相关模型下载可以参考：
+
+- [PP-Structure 模型库](./models_list.md)
+
+OCR相关模型下载可以参考：
+
+- [PP-OCR 模型库](../ppocr/model_list.md)
diff --git a/docs/version2.x/ppstructure/ppstructure_model.md b/docs/version2.x/ppstructure/ppstructure_model.md
new file mode 100644
index 0000000..41fd17d
--- /dev/null
+++ b/docs/version2.x/ppstructure/ppstructure_model.md
@@ -0,0 +1,60 @@
+---
+comments: true
+hide:
+    - toc
+---
+
+
+# PP-Structure 系列模型列表
+
+## 1. 版面分析模型
+
+| 模型名称                                | 模型简介      | 推理模型大小 | 下载地址        | dict path                                                                      |
+| --------------------------------------- | ----- | ------------ | ------------------- | ------ |
+| picodet_lcnet_x1_0_fgd_layout           | 基于PicoDet LCNet_x1_0和FGD蒸馏在PubLayNet 数据集训练的英文版面分析模型，可以划分**文字、标题、表格、图片以及列表**5类区域 | 9.7M         | [推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout.pdparams)                        | [PubLayNet dict](../../../ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt) |
+| ppyolov2_r50vd_dcn_365e_publaynet       | 基于PP-YOLOv2在PubLayNet数据集上训练的英文版面分析模型                                                                     | 221.0M       | [推理模型](https://paddle-model-ecology.bj.bcebos.com/model/layout-parser/ppyolov2_r50vd_dcn_365e_publaynet.tar) / [训练模型](https://paddle-model-ecology.bj.bcebos.com/model/layout-parser/ppyolov2_r50vd_dcn_365e_publaynet_pretrained.pdparams) | 同上                                                                           |
+| picodet_lcnet_x1_0_fgd_layout_cdla      | CDLA数据集训练的中文版面分析模型，可以划分为**表格、图片、图片标题、表格、表格标题、页眉、脚本、引用、公式**10类区域       | 9.7M         | [推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_cdla_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_cdla.pdparams)              | [CDLA dict](../../../ppocr/utils/dict/layout_dict/layout_cdla_dict.txt)           |
+| picodet_lcnet_x1_0_fgd_layout_table     | 表格数据集训练的版面分析模型，支持中英文文档表格区域的检测                                                                 | 9.7M         | [推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_table_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_table.pdparams)            | [Table dict](../../../ppocr/utils/dict/layout_dict/layout_table_dict.txt)         |
+| ppyolov2_r50vd_dcn_365e_tableBank_word  | 基于PP-YOLOv2在TableBank Word 数据集训练的版面分析模型，支持英文文档表格区域的检测                                         | 221.0M       | [推理模型](https://paddle-model-ecology.bj.bcebos.com/model/layout-parser/ppyolov2_r50vd_dcn_365e_tableBank_word.tar)                                                                                                                               | 同上                                                                           |
+| ppyolov2_r50vd_dcn_365e_tableBank_latex | 基于PP-YOLOv2在TableBank Latex数据集训练的版面分析模型，支持英文文档表格区域的检测                                         | 221.0M       | [推理模型](https://paddle-model-ecology.bj.bcebos.com/model/layout-parser/ppyolov2_r50vd_dcn_365e_tableBank_latex.tar)                                                                                                                              | 同上                                                                           |
+
+## 2. OCR和表格识别模型
+
+### 2.1 OCR
+
+| 模型名称                       | 模型简介                                    | 推理模型大小 | 下载地址                                                                                                                                                                                                          |
+| ------------------------------ | ------------------------------------------- | ------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| en_ppocr_mobile_v2.0_table_det | PubTabNet数据集训练的英文表格场景的文字检测 | 4.7M         | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_det_train.tar) |
+| en_ppocr_mobile_v2.0_table_rec | PubTabNet数据集训练的英文表格场景的文字识别 | 6.9M         | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_rec_train.tar) |
+
+如需要使用其他OCR模型，可以在 [PP-OCR model_list](../ppocr/model_list.md) 下载模型或者使用自己训练好的模型配置到 `det_model_dir`, `rec_model_dir`两个字段即可。
+
+### 2.2 表格识别模型
+
+| 模型名称                             | 模型简介                                                   | 推理模型大小 | 下载地址                                                                                                                                                                                                                              |
+| ------------------------------------ | ---------------------------------------------------------- | ------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| en_ppocr_mobile_v2.0_table_structure | 基于TableRec-RARE在PubTabNet数据集上训练的英文表格识别模型 | 6.8M         | [推理模型](https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_structure_train.tar)         |
+| en_ppstructure_mobile_v2.0_SLANet    | 基于SLANet在PubTabNet数据集上训练的英文表格识别模型        | 9.2M         | [推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/en_ppstructure_mobile_v2.0_SLANet_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/en_ppstructure_mobile_v2.0_SLANet_train.tar) |
+| ch_ppstructure_mobile_v2.0_SLANet    | 基于SLANet的中文表格识别模型                               | 9.3M         | [推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/ch_ppstructure_mobile_v2.0_SLANet_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/ch_ppstructure_mobile_v2.0_SLANet_train.tar) |
+
+## 3. KIE模型
+
+在XFUND_zh数据集上，不同模型的精度与V100 GPU上速度信息如下所示。
+
+| 模型名称                  | 模型简介                                         | 推理模型大小 | 精度(hmean) | 预测耗时(ms) | 下载地址                                                                                                                                                                                                                         |
+| ------------------------- | ------------------------------------------------ | ------------ | ----------- | ------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| ser_VI-LayoutXLM_xfund_zh | 基于VI-LayoutXLM在xfund中文数据集上训练的SER模型 | 1.1G         | 93.19%      | 15.49        | [推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar) |
+| re_VI-LayoutXLM_xfund_zh  | 基于VI-LayoutXLM在xfund中文数据集上训练的RE模型  | 1.1G         | 83.92%      | 15.49        | [推理模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar)   |
+| ser_LayoutXLM_xfund_zh    | 基于LayoutXLM在xfund中文数据集上训练的SER模型    | 1.4G         | 90.38%      | 19.49        | [推理模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar)                                                            |
+| re_LayoutXLM_xfund_zh     | 基于LayoutXLM在xfund中文数据集上训练的RE模型     | 1.4G         | 74.83%      | 19.49        | [推理模型](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar)                                                              |
+| ser_LayoutLMv2_xfund_zh   | 基于LayoutLMv2在xfund中文数据集上训练的SER模型   | 778.0M       | 85.44%      | 31.46        | [推理模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLMv2_xfun_zh_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLMv2_xfun_zh.tar)                                                          |
+| re_LayoutLMv2_xfund_zh    | 基于LayoutLMv2在xfun中文数据集上训练的RE模型     | 765.0M       | 67.77%      | 31.46        | [推理模型 coming soon]() / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutLMv2_xfun_zh.tar)                                                                                                                        |
+| ser_LayoutLM_xfund_zh     | 基于LayoutLM在xfund中文数据集上训练的SER模型     | 430.0M       | 77.31%      | -            | [推理模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLM_xfun_zh_infer.tar) / [训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutLM_xfun_zh.tar)                                                              |
+
+* 注：上述预测耗时信息仅包含了inference模型的推理耗时，没有统计预处理与后处理耗时，测试环境为`V100 GPU + CUDA 10.2 + CUDNN 8.1.1 + TRT 7.2.3.4`。
+
+在wildreceipt数据集上，SDMGR模型精度与下载地址如下所示。
+
+| 模型名称 | 模型简介         | 模型大小 | 精度   | 下载地址                                                                                              |
+| -------- | ---------------- | -------- | ------ | ----------------------------------------------------------------------------------------------------- |
+| SDMGR    | 关键信息提取模型 | 78.0M    | 86.70% | [推理模型 coming soon]() / [训练模型](https://paddleocr.bj.bcebos.com/dygraph_v2.1/kie/kie_vgg16.tar) |
diff --git a/docs/version2.x/ppstructure/quick_start.en.md b/docs/version2.x/ppstructure/quick_start.en.md
new file mode 100644
index 0000000..946625c
--- /dev/null
+++ b/docs/version2.x/ppstructure/quick_start.en.md
@@ -0,0 +1,395 @@
+---
+comments: true
+---
+
+# PP-Structure Quick Start
+
+## 1. Environment Preparation
+
+### 1.1 Install PaddlePaddle
+
+> If you do not have a Python environment, please refer to [Environment Preparation](../ppocr/environment.en.md).
+
+- PaddlePaddle with CUDA 11.8
+
+  ```bash linenums="1"
+  python3 -m pip install "paddlepaddle-gpu<=2.6" -i https://www.paddlepaddle.org.cn/packages/stable/cu118/
+  ```
+
+- PaddlePaddle with CUDA 12.3
+
+  ```bash linenums="1"
+  python3 -m pip install "paddlepaddle-gpu<=2.6" -i https://www.paddlepaddle.org.cn/packages/stable/cu123/
+  ```
+
+- If your machine does not have an available GPU, please run the following command to install the CPU version
+
+  ```bash linenums="1"
+  python3 -m pip install "paddlepaddle<=2.6" -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+  ```
+
+For more software version requirements, please refer to the instructions in the [Installation Document](https://www.paddlepaddle.org.cn/en/install/quick).
+
+### 1.2 Install PaddleOCR Whl Package
+
+```bash linenums="1"
+python3 -m pip install "paddleocr<3.0"
+
+# Install the image direction classification dependency package paddleclas (if you do not use the image direction classification, you can skip it)
+python3 -m pip install paddleclas
+```
+
+## 2. Quick Use
+
+### 2.1 Use by command line
+
+#### 2.1.1 image orientation + layout analysis + table recognition
+
+```bash linenums="1"
+# Temporarily disable the new IR feature
+export FLAGS_enable_pir_api=0
+paddleocr --image_dir=ppstructure/docs/table/1.png --type=structure --image_orientation=true
+```
+
+#### 2.1.2 layout analysis + table recognition
+
+```bash linenums="1"
+paddleocr --image_dir=ppstructure/docs/table/1.png --type=structure
+```
+
+#### 2.1.3 layout analysis
+
+```bash linenums="1"
+paddleocr --image_dir=ppstructure/docs/table/1.png --type=structure --table=false --ocr=false
+```
+
+#### 2.1.4 table recognition
+
+```bash linenums="1"
+paddleocr --image_dir=ppstructure/docs/table/table.jpg --type=structure --layout=false
+```
+
+#### 2.1.5 Key Information Extraction
+
+Key information extraction does not currently support use by the whl package. For detailed usage tutorials, please refer to: [Key Information Extraction](../ppocr/model_train/kie.en.md).
+
+#### 2.1.6 layout recovery
+
+Two layout recovery methods are provided, For detailed usage tutorials, please refer to: [Layout Recovery](./model_train/recovery_to_doc.en.md).
+
+- PDF parse
+- OCR
+
+Recovery by using PDF parse (only support pdf as input):
+
+```bash linenums="1"
+paddleocr --image_dir=ppstructure/docs/recovery/UnrealText.pdf --type=structure --recovery=true --use_pdf2docx_api=true
+```
+
+Recovery by using OCR：
+
+```bash linenums="1"
+paddleocr --image_dir=ppstructure/docs/table/1.png --type=structure --recovery=true --lang='en'
+```
+
+#### 2.1.7 layout recovery(PDF to Markdown)
+
+Do not use LaTeXCOR model for formula recognition：
+
+```bash linenums="1"
+paddleocr --image_dir=ppstructure/docs/recovery/UnrealText.pdf --type=structure --recovery=true --recovery_to_markdown=true --lang='en'
+```
+
+Use LaTeXCOR model for formula recognition, where Chinese layout model must be used：
+
+```bash linenums="1"
+paddleocr --image_dir=ppstructure/docs/recovery/UnrealText.pdf --type=structure --recovery=true --formula=true --recovery_to_markdown=true --lang='ch'
+```
+
+### 2.2 Use by python script
+
+#### 2.2.1 image orientation + layout analysis + table recognition
+
+```python linenums="1"
+import os
+import cv2
+from paddleocr import PPStructure,draw_structure_result,save_structure_res
+
+table_engine = PPStructure(show_log=True, image_orientation=True)
+
+save_folder = './output'
+img_path = 'ppstructure/docs/table/1.png'
+img = cv2.imread(img_path)
+result = table_engine(img)
+save_structure_res(result, save_folder,os.path.basename(img_path).split('.')[0])
+
+for line in result:
+    line.pop('img')
+    print(line)
+
+from PIL import Image
+
+font_path = 'doc/fonts/simfang.ttf' # PaddleOCR下提供字体包
+image = Image.open(img_path).convert('RGB')
+im_show = draw_structure_result(image, result,font_path=font_path)
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+#### 2.2.2 layout analysis + table recognition
+
+```python linenums="1"
+import os
+import cv2
+from paddleocr import PPStructure,draw_structure_result,save_structure_res
+
+table_engine = PPStructure(show_log=True)
+
+save_folder = './output'
+img_path = 'ppstructure/docs/table/1.png'
+img = cv2.imread(img_path)
+result = table_engine(img)
+save_structure_res(result, save_folder,os.path.basename(img_path).split('.')[0])
+
+for line in result:
+    line.pop('img')
+    print(line)
+
+from PIL import Image
+
+font_path = 'doc/fonts/simfang.ttf' # font provided in PaddleOCR
+image = Image.open(img_path).convert('RGB')
+im_show = draw_structure_result(image, result,font_path=font_path)
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+#### 2.2.3 layout analysis
+
+```python linenums="1"
+import os
+import cv2
+from paddleocr import PPStructure,save_structure_res
+
+table_engine = PPStructure(table=False, ocr=False, show_log=True)
+
+save_folder = './output'
+img_path = 'ppstructure/docs/table/1.png'
+img = cv2.imread(img_path)
+result = table_engine(img)
+save_structure_res(result, save_folder, os.path.basename(img_path).split('.')[0])
+
+for line in result:
+    line.pop('img')
+    print(line)
+```
+
+```python linenums="1"
+import os
+import cv2
+from paddleocr import PPStructure,save_structure_res
+
+ocr_engine = PPStructure(table=False, ocr=True, show_log=True)
+
+save_folder = './output'
+img_path = 'ppstructure/docs/recovery/UnrealText.pdf'
+result = ocr_engine(img_path)
+for index, res in enumerate(result):
+    save_structure_res(res, save_folder, os.path.basename(img_path).split('.')[0], index)
+
+for res in result:
+    for line in res:
+        line.pop('img')
+        print(line)
+```
+
+```python linenums="1"
+import os
+import cv2
+import numpy as np
+from paddleocr import PPStructure,save_structure_res
+from paddle.utils import try_import
+from PIL import Image
+
+ocr_engine = PPStructure(table=False, ocr=True, show_log=True)
+
+save_folder = './output'
+img_path = 'ppstructure/docs/recovery/UnrealText.pdf'
+
+fitz = try_import("fitz")
+imgs = []
+with fitz.open(img_path) as pdf:
+    for pg in range(0, pdf.page_count):
+        page = pdf[pg]
+        mat = fitz.Matrix(2, 2)
+        pm = page.get_pixmap(matrix=mat, alpha=False)
+
+        # if width or height > 2000 pixels, don't enlarge the image
+        if pm.width > 2000 or pm.height > 2000:
+            pm = page.get_pixmap(matrix=fitz.Matrix(1, 1), alpha=False)
+
+        img = Image.frombytes("RGB", [pm.width, pm.height], pm.samples)
+        img = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
+        imgs.append(img)
+
+for index, img in enumerate(imgs):
+    result = ocr_engine(img)
+    save_structure_res(result, save_folder, os.path.basename(img_path).split('.')[0], index)
+    for line in result:
+        line.pop('img')
+        print(line)
+```
+
+#### 2.2.4 table recognition
+
+```python linenums="1"
+import os
+import cv2
+from paddleocr import PPStructure,save_structure_res
+
+table_engine = PPStructure(layout=False, show_log=True)
+
+save_folder = './output'
+img_path = 'ppstructure/docs/table/table.jpg'
+img = cv2.imread(img_path)
+result = table_engine(img)
+save_structure_res(result, save_folder, os.path.basename(img_path).split('.')[0])
+
+for line in result:
+    line.pop('img')
+    print(line)
+```
+
+#### 2.2.5 Key Information Extraction
+
+Key information extraction does not currently support use by the whl package. For detailed usage tutorials, please refer to: [Inference](../legacy/python_infer.en.md).
+
+#### 2.2.6 layout recovery
+
+```python linenums="1"
+import os
+import cv2
+from paddleocr import PPStructure,save_structure_res
+from paddleocr.ppstructure.recovery.recovery_to_doc import sorted_layout_boxes, convert_info_docx
+
+# Chinese image
+table_engine = PPStructure(recovery=True)
+# English image
+# table_engine = PPStructure(recovery=True, lang='en')
+
+save_folder = './output'
+img_path = 'ppstructure/docs/table/1.png'
+img = cv2.imread(img_path)
+result = table_engine(img)
+save_structure_res(result, save_folder, os.path.basename(img_path).split('.')[0])
+
+for line in result:
+    line.pop('img')
+    print(line)
+
+h, w, _ = img.shape
+res = sorted_layout_boxes(result, w)
+convert_info_docx(img, res, save_folder, os.path.basename(img_path).split('.')[0])
+```
+
+#### 2.2.7 layout recovery(PDF to Markdown)
+
+```python linenums="1"
+import os
+import cv2
+from paddleocr import PPStructure,save_structure_res
+from paddleocr.ppstructure.recovery.recovery_to_doc import sorted_layout_boxes
+from paddleocr.ppstructure.recovery.recovery_to_markdown import convert_info_markdown
+
+# Chinese image
+table_engine = PPStructure(recovery=True)
+# English image
+# table_engine = PPStructure(recovery=True, lang='en')
+
+save_folder = './output'
+img_path = 'ppstructure/docs/table/1.png'
+img = cv2.imread(img_path)
+result = table_engine(img)
+save_structure_res(result, save_folder, os.path.basename(img_path).split('.')[0])
+
+for line in result:
+    line.pop('img')
+    print(line)
+
+h, w, _ = img.shape
+res = sorted_layout_boxes(result, w)
+convert_info_markdown(res, save_folder, os.path.basename(img_path).split('.')[0])
+```
+
+### 2.3 Result description
+
+The return of PP-Structure is a list of dicts, the example is as follows:
+
+#### 2.3.1 layout analysis + table recognition
+
+```bash linenums="1"
+[
+  {   'type': 'Text',
+      'bbox': [34, 432, 345, 462],
+      'res': ([[36.0, 437.0, 341.0, 437.0, 341.0, 446.0, 36.0, 447.0], [41.0, 454.0, 125.0, 453.0, 125.0, 459.0, 41.0, 460.0]],
+                [('Tigure-6. The performance of CNN and IPT models using difforen', 0.90060663), ('Tent  ', 0.465441)])
+  }
+]
+```
+
+Each field in dict is described as follows:
+
+| field | description  |
+| --- |---|
+|type| Type of image area. |
+|bbox| The coordinates of the image area in the original image, respectively [upper left corner x, upper left corner y, lower right corner x, lower right corner y]. |
+|res| OCR or table recognition result of the image area. <br> table: a dict with field descriptions as follows: <br>&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp; `html`: html str of table.<br>&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp; In the code usage mode, set return_ocr_result_in_table=True whrn call can get the detection and recognition results of each text in the table area, corresponding to the following fields: <br>&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp; `boxes`: text detection boxes.<br>&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp; `rec_res`: text recognition results.<br> OCR: A tuple containing the detection boxes and recognition results of each single text. |
+
+After the recognition is completed, each image will have a directory with the same name under the directory specified by the `output` field. Each table in the image will be stored as an excel, and the picture area will be cropped and saved. The filename of  excel and picture is their coordinates in the image.
+
+  ```text linenums="1"
+  /output/table/1/
+    └─ res.txt
+    └─ [454, 360, 824, 658].xlsx        table recognition result
+    └─ [16, 2, 828, 305].jpg            picture in Image
+    └─ [17, 361, 404, 711].xlsx        table recognition result
+  ```
+
+#### 2.3.2 Key Information Extraction
+
+Please refer to: [Key Information Extraction](../ppocr/model_train/kie.en.md) .
+
+### 2.4 Parameter Description
+
+| field                   | description                                                                                                                | default |
+|-------------------------|----------------------------------------------------------------------------------------------------------------------------|---|
+| output                  | result save path                                                                                                           | ./output/table |
+| table_max_len           | long side of the image resize in table structure model                                                                     | 488 |
+| table_model_dir         | Table structure model inference model path                                                                                 | None |
+| table_char_dict_path    | The dictionary path of table structure model                                                                               | ../ppocr/utils/dict/table_structure_dict.txt  |
+| merge_no_span_structure | In the table recognition model, whether to merge '\<td>' and '\</td>'                                                      | False |
+| formula_model_dir       | Formula recognition model inference model path                                                                             | None                                          |
+| formula_char_dict_path  | The dictionary path of formula recognition model                                                                           | ../ppocr/utils/dict/latex_ocr_tokenizer.json |
+| layout_model_dir        | Layout analysis model inference model path                                                                                 | None |
+| layout_dict_path        | The dictionary path of layout analysis model                                                                               | ../ppocr/utils/dict/layout_publaynet_dict.txt |
+| layout_score_threshold  | The box threshold path of layout analysis model                                                                            | 0.5|
+| layout_nms_threshold    | The nms threshold path of layout analysis model                                                                            | 0.5|
+| kie_algorithm           | kie model algorithm                                                                                                        | LayoutXLM|
+| ser_model_dir           | Ser model inference model path                                                                                             | None|
+| ser_dict_path           | The dictionary path of Ser model                                                                                           | ../train_data/XFUND/class_list_xfun.txt|
+| mode                    | structure or kie                                                                                                           | structure   |
+| image_orientation       | Whether to perform image orientation classification in forward                                                             | False   |
+| layout                  | Whether to perform layout analysis in forward                                                                              | True   |
+| table                   | Whether to perform table recognition in forward                                                                            | True   | 
+| formula                 | Whether to perform formula recognition in forward                                                                          | False |
+| ocr                     | Whether to perform ocr for non-table areas in layout analysis. When layout is False, it will be automatically set to False | True |
+| recovery                | Whether to perform layout recovery in forward                                                                              | False |
+| recovery_to_markdown    | Whether to convert the layout recovery results into a markdown file                                                        | False |
+| save_pdf                | Whether to convert docx to pdf when recovery                                                                               | False |
+| structure_version       | Structure version, optional PP-structure and PP-structurev2                                                                | PP-structure |
+
+Most of the parameters are consistent with the PaddleOCR whl package, see [whl package documentation](../ppocr/blog/whl.en.md)
+
+## 3. Summary
+
+Through the content in this section, you can master the use of PP-Structure related functions through PaddleOCR whl package. Please refer to [documentation tutorial](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.7.1/README_en.md) for more detailed usage tutorials including model training, inference and deployment, etc.
diff --git a/docs/version2.x/ppstructure/quick_start.md b/docs/version2.x/ppstructure/quick_start.md
new file mode 100644
index 0000000..282720c
--- /dev/null
+++ b/docs/version2.x/ppstructure/quick_start.md
@@ -0,0 +1,400 @@
+---
+comments: true
+---
+
+# PP-Structure 快速开始
+
+## 1. 准备环境
+
+### 1.1 安装PaddlePaddle
+>
+> 如果您没有基础的Python运行环境，请参考[运行环境准备](../ppocr/environment.md)。
+
+- CUDA11.8 的 PaddlePaddle
+
+  ```bash linenums="1"
+  python3 -m pip install "paddlepaddle-gpu<=2.6" -i https://www.paddlepaddle.org.cn/packages/stable/cu118/
+  ```
+
+-  CUDA12.3 的 PaddlePaddle
+
+  ```bash linenums="1"
+  python3 -m pip install "paddlepaddle-gpu<=2.6" -i https://www.paddlepaddle.org.cn/packages/stable/cu123/
+  ```
+
+- 您的机器是CPU，请运行以下命令安装
+
+  ```bash linenums="1"
+  python3 -m pip install "paddlepaddle<=2.6" -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+  ```
+
+更多的版本需求，请参照[飞桨官网安装文档](https://www.paddlepaddle.org.cn/install/quick)中的说明进行操作。
+
+### 1.2 安装PaddleOCR whl包
+
+```bash linenums="1"
+python3 -m pip install "paddleocr<3.0"
+
+# 安装 图像方向分类依赖包paddleclas（如不需要图像方向分类功能，可跳过）
+python3 -m pip install paddleclas
+```
+
+## 2. 便捷使用
+
+### 2.1 命令行使用
+
+#### 2.1.1 图像方向分类+版面分析+表格识别
+
+```bash linenums="1"
+# 暂时关闭新 IR 功能
+export FLAGS_enable_pir_api=0
+paddleocr --image_dir=ppstructure/docs/table/1.png --type=structure --image_orientation=true
+```
+
+#### 2.1.2 版面分析+表格识别
+
+```bash linenums="1"
+paddleocr --image_dir=ppstructure/docs/table/1.png --type=structure
+```
+
+#### 2.1.3 版面分析
+
+```bash linenums="1"
+paddleocr --image_dir=ppstructure/docs/table/1.png --type=structure --table=false --ocr=false
+```
+
+#### 2.1.4 表格识别
+
+```bash linenums="1"
+paddleocr --image_dir=ppstructure/docs/table/table.jpg --type=structure --layout=false
+```
+
+#### 2.1.5 关键信息抽取
+
+关键信息抽取暂不支持通过whl包调用，详细使用教程请参考：[关键信息抽取教程](../ppocr/model_train/kie.md)。
+
+#### 2.1.6 版面恢复
+
+版面恢复分为2种方法，详细介绍请参考：[版面恢复教程](./model_train/recovery_to_doc.md)：
+
+- PDF解析
+- OCR技术
+
+通过PDF解析(只支持pdf格式的输入)：
+
+```bash linenums="1"
+paddleocr --image_dir=ppstructure/docs/recovery/UnrealText.pdf --type=structure --recovery=true --use_pdf2docx_api=true
+```
+
+通过OCR技术：
+
+```bash linenums="1"
+# 中文测试图
+paddleocr --image_dir=ppstructure/docs/table/1.png --type=structure --recovery=true
+# 英文测试图
+paddleocr --image_dir=ppstructure/docs/table/1.png --type=structure --recovery=true --lang='en'
+# pdf测试文件
+paddleocr --image_dir=ppstructure/docs/recovery/UnrealText.pdf --type=structure --recovery=true --lang='en'
+```
+
+#### 2.1.7 版面恢复+转换为markdown文件
+
+不使用LaTeXOCR模型进行公式识别：
+
+```bash linenums="1"
+paddleocr --image_dir=ppstructure/docs/recovery/UnrealText.pdf --type=structure --recovery=true --recovery_to_markdown=true --lang='en'
+```
+
+使用LaTeXOCR模型进行公式识别，其中必须使用中文layout模型：
+
+```bash linenums="1"
+paddleocr --image_dir=ppstructure/docs/recovery/UnrealText.pdf --type=structure --recovery=true --formula=true --recovery_to_markdown=true --lang='ch'
+```
+
+### 2.2 Python脚本使用
+
+#### 2.2.1 图像方向分类+版面分析+表格识别
+
+```python linenums="1"
+import os
+import cv2
+from paddleocr import PPStructure,draw_structure_result,save_structure_res
+
+table_engine = PPStructure(show_log=True, image_orientation=True)
+
+save_folder = './output'
+img_path = 'ppstructure/docs/table/1.png'
+img = cv2.imread(img_path)
+result = table_engine(img)
+save_structure_res(result, save_folder,os.path.basename(img_path).split('.')[0])
+
+for line in result:
+    line.pop('img')
+    print(line)
+
+from PIL import Image
+
+font_path = 'doc/fonts/simfang.ttf' # PaddleOCR下提供字体包
+image = Image.open(img_path).convert('RGB')
+im_show = draw_structure_result(image, result,font_path=font_path)
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+#### 2.2.2 版面分析+表格识别
+
+```python linenums="1"
+import os
+import cv2
+from paddleocr import PPStructure,draw_structure_result,save_structure_res
+
+table_engine = PPStructure(show_log=True)
+
+save_folder = './output'
+img_path = 'ppstructure/docs/table/1.png'
+img = cv2.imread(img_path)
+result = table_engine(img)
+save_structure_res(result, save_folder,os.path.basename(img_path).split('.')[0])
+
+for line in result:
+    line.pop('img')
+    print(line)
+
+from PIL import Image
+
+font_path = 'doc/fonts/simfang.ttf' # PaddleOCR下提供字体包
+image = Image.open(img_path).convert('RGB')
+im_show = draw_structure_result(image, result,font_path=font_path)
+im_show = Image.fromarray(im_show)
+im_show.save('result.jpg')
+```
+
+#### 2.2.3 版面分析
+
+```python linenums="1"
+import os
+import cv2
+from paddleocr import PPStructure,save_structure_res
+
+table_engine = PPStructure(table=False, ocr=False, show_log=True)
+
+save_folder = './output'
+img_path = 'ppstructure/docs/table/1.png'
+img = cv2.imread(img_path)
+result = table_engine(img)
+save_structure_res(result, save_folder, os.path.basename(img_path).split('.')[0])
+
+for line in result:
+    line.pop('img')
+    print(line)
+```
+
+```python linenums="1"
+import os
+import cv2
+from paddleocr import PPStructure,save_structure_res
+
+ocr_engine = PPStructure(table=False, ocr=True, show_log=True)
+
+save_folder = './output'
+img_path = 'ppstructure/docs/recovery/UnrealText.pdf'
+result = ocr_engine(img_path)
+for index, res in enumerate(result):
+    save_structure_res(res, save_folder, os.path.basename(img_path).split('.')[0], index)
+
+for res in result:
+    for line in res:
+        line.pop('img')
+        print(line)
+```
+
+```python linenums="1"
+import os
+import cv2
+import numpy as np
+from paddleocr import PPStructure,save_structure_res
+from paddle.utils import try_import
+from PIL import Image
+
+ocr_engine = PPStructure(table=False, ocr=True, show_log=True)
+
+save_folder = './output'
+img_path = 'ppstructure/docs/recovery/UnrealText.pdf'
+
+fitz = try_import("fitz")
+imgs = []
+with fitz.open(img_path) as pdf:
+    for pg in range(0, pdf.page_count):
+        page = pdf[pg]
+        mat = fitz.Matrix(2, 2)
+        pm = page.get_pixmap(matrix=mat, alpha=False)
+
+        # if width or height > 2000 pixels, don't enlarge the image
+        if pm.width > 2000 or pm.height > 2000:
+            pm = page.get_pixmap(matrix=fitz.Matrix(1, 1), alpha=False)
+
+        img = Image.frombytes("RGB", [pm.width, pm.height], pm.samples)
+        img = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
+        imgs.append(img)
+
+for index, img in enumerate(imgs):
+    result = ocr_engine(img)
+    save_structure_res(result, save_folder, os.path.basename(img_path).split('.')[0], index)
+    for line in result:
+        line.pop('img')
+        print(line)
+```
+
+#### 2.2.4 表格识别
+
+```python linenums="1"
+import os
+import cv2
+from paddleocr import PPStructure,save_structure_res
+
+table_engine = PPStructure(layout=False, show_log=True)
+
+save_folder = './output'
+img_path = 'ppstructure/docs/table/table.jpg'
+img = cv2.imread(img_path)
+result = table_engine(img)
+save_structure_res(result, save_folder, os.path.basename(img_path).split('.')[0])
+
+for line in result:
+    line.pop('img')
+    print(line)
+```
+
+#### 2.2.5 关键信息抽取
+
+关键信息抽取暂不支持通过whl包调用，详细使用教程请参考：[inference文档](./infer_deploy/python_infer.md)。
+
+#### 2.2.6 版面恢复
+
+```python linenums="1"
+import os
+import cv2
+from paddleocr import PPStructure,save_structure_res
+from paddleocr.ppstructure.recovery.recovery_to_doc import sorted_layout_boxes, convert_info_docx
+
+# 中文测试图
+table_engine = PPStructure(recovery=True)
+# 英文测试图
+# table_engine = PPStructure(recovery=True, lang='en')
+
+save_folder = './output'
+img_path = 'ppstructure/docs/table/1.png'
+img = cv2.imread(img_path)
+result = table_engine(img)
+save_structure_res(result, save_folder, os.path.basename(img_path).split('.')[0])
+
+for line in result:
+    line.pop('img')
+    print(line)
+
+h, w, _ = img.shape
+res = sorted_layout_boxes(result, w)
+convert_info_docx(img, res, save_folder, os.path.basename(img_path).split('.')[0])
+```
+
+#### 2.2.7 版面恢复+转换为markdown文件
+
+```python linenums="1"
+import os
+import cv2
+from paddleocr import PPStructure,save_structure_res
+from paddleocr.ppstructure.recovery.recovery_to_doc import sorted_layout_boxes
+from paddleocr.ppstructure.recovery.recovery_to_markdown import convert_info_markdown
+
+# 中文测试图
+table_engine = PPStructure(recovery=True)
+# 英文测试图
+# table_engine = PPStructure(recovery=True, lang='en')
+
+save_folder = './output'
+img_path = 'ppstructure/docs/table/1.png'
+img = cv2.imread(img_path)
+result = table_engine(img)
+save_structure_res(result, save_folder, os.path.basename(img_path).split('.')[0])
+
+for line in result:
+    line.pop('img')
+    print(line)
+
+h, w, _ = img.shape
+res = sorted_layout_boxes(result, w)
+convert_info_markdown(res, save_folder, os.path.basename(img_path).split('.')[0])
+```
+
+### 2.3 返回结果说明
+
+PP-Structure的返回结果为一个dict组成的list，示例如下：
+
+#### 2.3.1 版面分析+表格识别
+
+```bash linenums="1"
+[
+  {   'type': 'Text',
+      'bbox': [34, 432, 345, 462],
+      'res': ([[36.0, 437.0, 341.0, 437.0, 341.0, 446.0, 36.0, 447.0], [41.0, 454.0, 125.0, 453.0, 125.0, 459.0, 41.0, 460.0]],
+                [('Tigure-6. The performance of CNN and IPT models using difforen', 0.90060663), ('Tent  ', 0.465441)])
+  }
+]
+```
+
+dict 里各个字段说明如下：
+
+| 字段 | 说明                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                    |
+| ---- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| type | 图片区域的类型                                                                                                                                                                                                                                                                                                                                                                                                                                                                                          |
+| bbox | 图片区域的在原图的坐标，分别[左上角x，左上角y，右下角x，右下角y]                                                                                                                                                                                                                                                                                                                                                                                                                                        |
+| res  | 图片区域的OCR或表格识别结果。<br> 表格: 一个dict，字段说明如下<br>&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp; `html`: 表格的HTML字符串<br>&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp; 在代码使用模式下，前向传入return_ocr_result_in_table=True可以拿到表格中每个文本的检测识别结果，对应为如下字段: <br>&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp; `boxes`: 文本检测坐标<br>&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp; `rec_res`: 文本识别结果。<br> OCR: 一个包含各个单行文字的检测坐标和识别结果的元组 |
+
+运行完成后，每张图片会在`output`字段指定的目录下有一个同名目录，图片里的每个表格会存储为一个excel，图片区域会被裁剪之后保存下来，excel文件和图片名为表格在图片里的坐标。
+
+  ```
+  /output/table/1/
+    └─ res.txt
+    └─ [454, 360, 824, 658].xlsx  表格识别结果
+    └─ [16, 2, 828, 305].jpg            被裁剪出的图片区域
+    └─ [17, 361, 404, 711].xlsx        表格识别结果
+  ```
+
+#### 2.3.2 关键信息抽取
+
+请参考：[关键信息抽取教程](../ppocr/model_train/kie.md)。
+
+### 2.4 参数说明
+
+| 字段                      | 说明                                              | 默认值                                        |
+|-------------------------|-------------------------------------------------| ------ |
+| output                  | 结果保存地址                                          | ./output/table                                |
+| table_max_len           | 表格结构模型预测时，图像的长边resize尺度                         | 488                                           |
+| table_model_dir         | 表格结构模型 inference 模型地址                           | None                                          |
+| table_char_dict_path    | 表格结构模型所用字典地址                                    | ../ppocr/utils/dict/table_structure_dict.txt  |
+| merge_no_span_structure | 表格识别模型中，是否对'\<td>'和'\</td>' 进行合并                | False                                         |
+| formula_model_dir       | 公式识别模型 inference 模型地址                           | None                                          |
+| formula_char_dict_path  | 公式识别模型所用字典地址                                    | ../ppocr/utils/dict/latex_ocr_tokenizer.json |
+| layout_model_dir        | 版面分析模型 inference 模型地址                           | None                                          |
+| layout_dict_path        | 版面分析模型字典                                        | ../ppocr/utils/dict/layout_publaynet_dict.txt |
+| layout_score_threshold  | 版面分析模型检测框阈值                                     | 0.5                                           |
+| layout_nms_threshold    | 版面分析模型nms阈值                                     | 0.5                                           |
+| kie_algorithm           | kie模型算法                                         | LayoutXLM                                     |
+| ser_model_dir           | ser模型  inference 模型地址                           | None                                          |
+| ser_dict_path           | ser模型字典                                         | ../train_data/XFUND/class_list_xfun.txt       |
+| mode                    | structure or kie                                | structure                                     |
+| image_orientation       | 前向中是否执行图像方向分类                                   | False                                         |
+| layout                  | 前向中是否执行版面分析                                     | True                                          |
+| table                   | 前向中是否执行表格识别                                     | True                                          |
+| formula                 | 前向中是否执行公式识别                                     | False                                         |
+| ocr                     | 对于版面分析中的非表格区域，是否执行ocr。当layout为False时会被自动设置为False | True                                          |
+| recovery                | 前向中是否执行版面恢复                                     | False                                         |
+| recovery_to_markdown    | 是否将版面恢复结果转换为markdown文件                        | False                                         |
+| save_pdf                | 版面恢复导出docx文件的同时，是否导出pdf文件                       | False                                         |
+| structure_version       | 模型版本，可选 PP-structure和PP-structurev2             | PP-structure                                  |
+
+大部分参数和PaddleOCR whl包保持一致，见 [whl包文档](../ppocr/blog/whl.md)
+
+## 3. 小结
+
+通过本节内容，相信您已经熟练掌握通过PaddleOCR whl包调用PP-Structure相关功能的使用方法，您可以参考[文档教程](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.7.1/README_ch.md)，获取包括模型训练、推理部署等更详细的使用教程。
diff --git a/docs/version2.x/static/images/demo.jpg b/docs/version2.x/static/images/demo.jpg
new file mode 100644
index 0000000..d8b5ca1
Binary files /dev/null and b/docs/version2.x/static/images/demo.jpg differ
diff --git a/docs/version2.x/static/images/favicon.ico b/docs/version2.x/static/images/favicon.ico
new file mode 100644
index 0000000..639c44e
Binary files /dev/null and b/docs/version2.x/static/images/favicon.ico differ
diff --git a/docs/version3.x/algorithm/PP-ChatOCRv4/PP-ChatOCRv4.en.md b/docs/version3.x/algorithm/PP-ChatOCRv4/PP-ChatOCRv4.en.md
new file mode 100644
index 0000000..28f64f5
--- /dev/null
+++ b/docs/version3.x/algorithm/PP-ChatOCRv4/PP-ChatOCRv4.en.md
@@ -0,0 +1,64 @@
+# Introduction to PP-ChatOCRV4
+**PP-ChatOCRv4** is a unique document and image intelligent analysis solution from PaddlePaddle, combining LLM, MLLM, and OCR technologies to address complex document information extraction challenges such as layout analysis, rare characters, multi-page PDFs, tables, and seal recognition. Integrated with ERNIE Bot, it fuses massive data and knowledge, achieving high accuracy and wide applicability. This pipeline also provides flexible service deployment options, supporting deployment on various hardware. Furthermore, it offers custom development capabilities, allowing you to train and fine-tune models on your own datasets, with seamless integration of trained models.
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-ChatOCRv4/algorithm_ppchatocrv4.png" width="600"/>
+</div>
+
+# Key Metrics
+
+<div align="center">
+<table>
+ <thead>
+  <tr > 
+   <th class>Solution</td> 
+   <th class>Avg Recall</td> 
+  </tr> 
+<thead>
+ <tbody>
+  <tr> 
+   <td>GPT-4o</td> 
+   <td>63.47%</td> 
+  </tr>
+  <tr> 
+   <td>PP-ChatOCRv3</td> 
+   <td class>70.08%</td> 
+  </tr> 
+  <tr> 
+   <td>Qwen2.5-VL-72B</td> 
+   <td>80.26%</td> 
+  </tr> 
+  <tr> 
+   <td><b>PP-ChatOCRv4</b></td> 
+   <td><b>85.55%</b></td> 
+  </tr> 
+ </tbody>
+</table>
+</div>
+
+# Demo
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-ChatOCRv4/algorithm_ppchatocrv4_demo1.png" width="350"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-ChatOCRv4/algorithm_ppchatocrv4_demo2.png" width="350"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-ChatOCRv4/algorithm_ppchatocrv4_demo3.png" width="350"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-ChatOCRv4/algorithm_ppchatocrv4_demo4.png" width="350"/>
+</div>
+
+
+# FAQ
+
+1. Does support other multimodal models?
+
+Yes, only set on pipeline configuration.
+
+2. How to reduce latency and improve throughput?
+
+Use the High-performance inference plugin, and deploy multi instances.
+
+3. How to further improve accuracy?
+
+Firstly, it is necessary to check whether the extracted visual information is correct. If the visual information is incorrect, it is necessary to visualize the visual prediction results to determine which model performs poorly, and then fine-tune train the model with more data. If the visual information is correct but cannot extract the correct information, the prompt needs to be adjusted according to the analysing about the question and answer.
diff --git a/docs/version3.x/algorithm/PP-ChatOCRv4/PP-ChatOCRv4.md b/docs/version3.x/algorithm/PP-ChatOCRv4/PP-ChatOCRv4.md
new file mode 100644
index 0000000..cf42b46
--- /dev/null
+++ b/docs/version3.x/algorithm/PP-ChatOCRv4/PP-ChatOCRv4.md
@@ -0,0 +1,65 @@
+# 一、PP-ChatOCRV4简介
+**PP-ChatOCRv4**是飞桨特色的文档图像智能分析解决方案，结合了 LLM、MLLM 和 OCR 等技术，一站式解决版面分析、生僻字识别、多页 PDF 文件批量解析、复杂表格识别、印章识别等常见的复杂文档信息抽取难点问题，结合文心大模型将海量数据和知识相融合，信息抽取准确率高且应用广泛。本产线同时提供了灵活的服务化部署方式，支持在多种硬件上部署。不仅如此，本产线也提供了二次开发的能力，您可以基于本产线在您自己的数据集上训练调优，训练后的模型也可以无缝集成。
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-ChatOCRv4/algorithm_ppchatocrv4.png" width="600"/>
+</div>
+
+# 二、关键指标
+
+<div align="center">
+<table>
+ <thead>
+  <tr > 
+   <th class>Solution</td> 
+   <th class>Avg Recall</td> 
+  </tr> 
+<thead>
+ <tbody>
+  <tr> 
+   <td>GPT-4o</td> 
+   <td>63.47%</td> 
+  </tr>
+  <tr> 
+   <td>PP-ChatOCRv3</td> 
+   <td class>70.08%</td> 
+  </tr> 
+  <tr> 
+   <td>Qwen2.5-VL-72B</td> 
+   <td>80.26%</td> 
+  </tr> 
+  <tr> 
+   <td><b>PP-ChatOCRv4</b></td> 
+   <td><b>85.55%</b></td> 
+  </tr> 
+ </tbody>
+</table>
+</div>
+
+# 三、PP-ChatOCRv4 Demo示例
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-ChatOCRv4/algorithm_ppchatocrv4_demo1.png" width="350"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-ChatOCRv4/algorithm_ppchatocrv4_demo2.png" width="350"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-ChatOCRv4/algorithm_ppchatocrv4_demo3.png" width="350"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-ChatOCRv4/algorithm_ppchatocrv4_demo4.png" width="350"/>
+</div>
+
+# 四、使用方法和常见问题
+
+1. 多模态大模型除了DocBee，是否支持其他多模态模型？
+
+支持，只需在配置文件中进行设置即可。
+
+2. 如何降低时延、提升吞吐？
+
+无论使用哪一种服务化部署方案，都可以通过启用高性能推理插件提升模型推理速度，从而降低处理时延。
+
+此外，对于高稳定性服务化部署方案，通过调整服务配置，设置多个实例，也可以充分利用部署机器的资源，有效提升吞吐。
+
+3. 如何进一步提升精度？
+
+首先需要检查提取的视觉信息是否正确，如果视觉信息有误，则需要通过可视化视觉预测结果，判断哪个模型效果较差，从而针对性地训练微调较差的模型；如果视觉信息无误，但无法抽取正确信息，则需要根据问答的具体情况调整Prompt。
diff --git a/docs/version3.x/algorithm/PP-OCRv5/PP-OCRv5.en.md b/docs/version3.x/algorithm/PP-OCRv5/PP-OCRv5.en.md
new file mode 100644
index 0000000..e6d423a
--- /dev/null
+++ b/docs/version3.x/algorithm/PP-OCRv5/PP-OCRv5.en.md
@@ -0,0 +1,1494 @@
+# Introduction to PP-OCRv5
+
+**PP-OCRv5** is the new generation text recognition solution of PP-OCR, focusing on multi-scenario and multi-text type recognition. In terms of text types, PP-OCRv5 supports 5 major mainstream text types: Simplified Chinese, Chinese Pinyin, Traditional Chinese, English, and Japanese. For scenarios, PP-OCRv5 has upgraded recognition capabilities for challenging scenarios such as complex Chinese and English handwriting, vertical text, and uncommon characters. On internal complex evaluation sets across multiple scenarios, PP-OCRv5 achieved a 13 percentage point end-to-end improvement over PP-OCRv4.
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv5/algorithm_ppocrv5.png" width="600"/>
+</div>
+
+# Key Metrics
+
+### 1. Text Detection Metrics
+<table>
+  <thead>
+    <tr>
+      <th>Model</th>
+      <th>Handwritten Chinese</th>
+      <th>Handwritten English</th>
+      <th>Printed Chinese</th>
+      <th>Printed English</th>
+      <th>Traditional Chinese</th>
+      <th>Ancient Text</th>
+      <th>Japanese</th>
+      <th>General Scenario</th>
+      <th>Pinyin</th>
+      <th>Rotation</th>
+      <th>Distortion</th>
+      <th>Artistic Text</th>
+      <th>Average</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td><b>PP-OCRv5_server_det</b></td>
+      <td><b>0.803</b></td>
+      <td><b>0.841</b></td>
+      <td><b>0.945</b></td>
+      <td><b>0.917</b></td>
+      <td><b>0.815</b></td>
+      <td><b>0.676</b></td>
+      <td><b>0.772</b></td>
+      <td><b>0.797</b></td>
+      <td><b>0.671</b></td>
+      <td><b>0.8</b></td>
+      <td><b>0.876</b></td>
+      <td><b>0.673</b></td>
+      <td><b>0.827</b></td>
+    </tr>
+    <tr>
+      <td>PP-OCRv4_server_det</td>
+      <td>0.706</td>
+      <td>0.249</td>
+      <td>0.888</td>
+      <td>0.690</td>
+      <td>0.759</td>
+      <td>0.473</td>
+      <td>0.685</td>
+      <td>0.715</td>
+      <td>0.542</td>
+      <td>0.366</td>
+      <td>0.775</td>
+      <td>0.583</td>
+      <td>0.662</td>
+    </tr>
+    <tr>
+      <td><b>PP-OCRv5_mobile_det</b></td>
+      <td><b>0.744</b></td>
+      <td><b>0.777</b></td>
+      <td><b>0.905</b></td>
+      <td><b>0.910</b></td>
+      <td><b>0.823</b></td>
+      <td><b>0.581</b></td>
+      <td><b>0.727</b></td>
+      <td><b>0.721</b></td>
+      <td><b>0.575</b></td>
+      <td><b>0.647</b></td>
+      <td><b>0.827</b></td>
+      <td>0.525</td>
+      <td><b>0.770</b></td>
+    </tr>
+    <tr>
+      <td>PP-OCRv4_mobile_det</td>
+      <td>0.583</td>
+      <td>0.369</td>
+      <td>0.872</td>
+      <td>0.773</td>
+      <td>0.663</td>
+      <td>0.231</td>
+      <td>0.634</td>
+      <td>0.710</td>
+      <td>0.430</td>
+      <td>0.299</td>
+      <td>0.715</td>
+      <td><b>0.549</b></td>
+      <td>0.624</td>
+    </tr>
+  </tbody>
+</table>
+
+Compared to PP-OCRv4, PP-OCRv5 shows significant improvement in all detection scenarios, especially in handwriting, ancient texts, and Japanese detection capabilities.
+
+### 2. Text Recognition Metrics
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv5/ocrv5_rec_acc.png" width="600"/>
+</div>
+
+<table>
+  <thead>
+    <tr>
+      <th>Evaluation Set Category</th>
+      <th>Handwritten Chinese</th>
+      <th>Handwritten English</th>
+      <th>Printed Chinese</th>
+      <th>Printed English</th>
+      <th>Traditional Chinese</th>
+      <th>Ancient Text</th>
+      <th>Japanese</th>
+      <th>Confusable Characters</th>
+      <th>General Scenario</th>
+      <th>Pinyin</th>
+      <th>Vertical Text</th>
+      <th>Artistic Text</th>
+      <th>Weighted Average</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td>PP-OCRv5_server_rec</td>
+      <td><b>0.5807</b></td>
+      <td><b>0.5806</b></td>
+      <td><b>0.9013</b></td>
+      <td><b>0.8679</b></td>
+      <td><b>0.7472</b></td>
+      <td><b>0.6039</b></td>
+      <td><b>0.7372</b></td>
+      <td><b>0.5946</b></td>
+      <td><b>0.8384</b></td>
+      <td><b>0.7435</b></td>
+      <td><b>0.9314</b></td>
+      <td><b>0.6397</b></td>
+      <td><b>0.8401</b></td>
+    </tr>
+    <tr>
+      <td>PP-OCRv4_server_rec</td>
+      <td>0.3626</td>
+      <td>0.2661</td>
+      <td>0.8486</td>
+      <td>0.6677</td>
+      <td>0.4097</td>
+      <td>0.3080</td>
+      <td>0.4623</td>
+      <td>0.5028</td>
+      <td>0.8362</td>
+      <td>0.2694</td>
+      <td>0.5455</td>
+      <td>0.5892</td>
+      <td>0.5735</td>
+    </tr>
+    <tr>
+      <td>PP-OCRv5_mobile_rec</td>
+      <td><b>0.4166</b></td>
+      <td><b>0.4944</b></td>
+      <td><b>0.8605</b></td>
+      <td><b>0.8753</b></td>
+      <td><b>0.7199</b></td>
+      <td><b>0.5786</b></td>
+      <td><b>0.7577</b></td>
+      <td><b>0.5570</b></td>
+      <td>0.7703</td>
+      <td><b>0.7248</b></td>
+      <td><b>0.8089</b></td>
+      <td>0.5398</td>
+      <td><b>0.8015</b></td>
+    </tr>
+    <tr>
+      <td>PP-OCRv4_mobile_rec</td>
+      <td>0.2980</td>
+      <td>0.2550</td>
+      <td>0.8398</td>
+      <td>0.6598</td>
+      <td>0.3218</td>
+      <td>0.2593</td>
+      <td>0.4724</td>
+      <td>0.4599</td>
+      <td><b>0.8106</b></td>
+      <td>0.2593</td>
+      <td>0.5924</td>
+      <td><b>0.5555</b></td>
+      <td>0.5301</td>
+    </tr>
+  </tbody>
+</table>
+
+A single model can cover multiple languages and text types, with recognition accuracy significantly ahead of previous generation products and mainstream open-source solutions.
+
+# PP-OCRv5 Demo Examples
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv5/algorithm_ppocrv5_demo1.png" width="600"/>
+</div>
+
+<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/PaddleX3.0/doc_images/PP-OCRv5/algorithm_ppocrv5_demo.pdf">More Demos</a>
+
+## Reference Data for Inference Performance
+
+Test Environment:
+
+- NVIDIA Tesla V100
+- Intel Xeon Gold 6271C
+- PaddlePaddle 3.0.0
+
+Tested on 200 images (including both general and document images). During testing, images are read from disk, so the image reading time and other associated overhead are also included in the total time consumption. If the images are preloaded into memory, the average time per image can be further reduced by approximately 25 ms.
+
+Unless otherwise specified:
+
+- PP-OCRv4_mobile_det and PP-OCRv4_mobile_rec models are used.
+- Document orientation classification, image correction, and text line orientation classification are not used.
+- `text_det_limit_type` is set to `"min"` and `text_det_limit_side_len` to `736`.
+
+### 1. Comparison of Inference Performance Between PP-OCRv5 and PP-OCRv4
+
+| Config | Description                                                  |
+| --------------- | ------------------------------------------------------------ |
+| v5_mobile      | Uses PP-OCRv5_mobile_det and PP-OCRv5_mobile_rec models. |
+| v4_mobile      | Uses PP-OCRv4_mobile_det and PP-OCRv4_mobile_rec models. |
+| v5_server      | Uses PP-OCRv5_server_det and PP-OCRv5_server_rec models. |
+| v4_server      | Uses PP-OCRv4_server_det and PP-OCRv4_server_rec models. |
+
+**GPU**
+
+| Configuration | Avg. Time per Image (s) | Avg. Characters Predicted per Second | Avg. CPU Utilization (%) | Peak RAM Usage (MB) | Avg. RAM Usage (MB) | Avg. GPU Utilization (%) | Peak VRAM Usage (MB) | Avg. VRAM Usage (MB) |
+| --- | --- | --- | --- | --- | --- | --- | --- | --- |
+| v5_mobile | 0.62 | 1054.23 | 106.35 | 1829.36 | 1521.92 | 17.42 | 4190.00  | 3114.02 |
+| v4_mobile | 0.29 | 2062.53 | 112.21 | 1713.10 | 1456.14 | 26.53  | 1304.00 | 1166.68 |
+| v5_server | 0.74 | 878.84 | 105.68 | 1899.80 | 1569.46 | 34.39 | 5402.00 | 4683.93 |
+| v4_server | 0.47 | 1322.06 | 108.06 | 1773.10 | 1518.94 | 55.25 | 6760.67 | 5788.02 |
+
+
+**CPU**
+
+| Configuration | Avg. Time per Image (s) | Avg. Characters Predicted per Second | Avg. CPU Utilization (%) | Peak RAM Usage (MB) | Avg. RAM Usage (MB) |
+| ------------- | ----------------------- | ------------------------------------ | ------------------------ | ------------------- | ------------------- |
+| v5_mobile | 1.75 | 371.82 | 965.89 | 2219.98 | 1830.97 |
+| v4_mobile | 1.37 | 444.27 | 1007.33 | 2090.53 | 1797.76 |
+| v5_server | 4.34 | 149.98 | 990.24 | 4020.85 | 3137.20 |
+| v4_server | 5.42 | 115.20 | 999.03 | 4018.35 | 3105.29 |
+
+> Note: PP-OCRv5 uses a larger dictionary in the recognition model, which increases inference time and causes slower performance compared to PP-OCRv4.
+
+### 2. Impact of Auxiliary Features on PP-OCRv5 Inference Performance
+
+| Config | Description                                                                                               |
+| --------------- | --------------------------------------------------------------------------------------------------------- |
+| base            | No document orientation classification, no image correction, no text line orientation classification.     |
+| with_textline  | Includes text line orientation classification only.                                                       |
+| with_all       | Includes document orientation classification, image correction, and text line orientation classification. |
+
+**GPU**
+
+| Configuration  | Avg. Time per Image (s) | Avg. Characters Predicted per Second | Avg. CPU Utilization (%) | Peak RAM Usage (MB) | Avg. RAM Usage (MB) | Avg. GPU Utilization (%) | Peak VRAM Usage (MB) | Avg. VRAM Usage (MB) |
+| -------------- | ----------------------- | ------------------------------------ | ------------------------ | ------------------- | ------------------- | ------------------------ | -------------------- | -------------------- |
+| base | 0.62 | 1054.23 | 106.35 | 1829.36 | 1521.92 | 17.42 | 4190.00 | 3114.02 |
+| with_textline | 0.64 | 1012.32 | 106.37 | 1867.69 | 1527.42 | 19.16 | 4198.00 | 3115.05 |
+| with_all | 1.09 | 562.99 | 105.67 | 2381.53 | 1792.48 | 10.77 | 2480.00 | 2065.54 |
+
+**CPU**
+
+| Configuration  | Avg. Time per Image (s) | Avg. Characters Predicted per Second | Avg. CPU Utilization (%) | Peak RAM Usage (MB) | Avg. RAM Usage (MB) |
+| -------------- | ----------------------- | ------------------------------------ | ------------------------ | ------------------- | ------------------- |
+| base | 1.75 | 371.82 | 965.89 | 2219.98 | 1830.97 |
+| with_textline | 1.87 | 347.61 | 972.08 | 2232.38 | 1822.13 |
+| with_all | 3.13 | 195.25 | 828.37 | 2751.47 | 2179.70 |
+
+> Note: Auxiliary features such as image unwarping can impact inference accuracy. More features do not necessarily yield better results and may increase resource usage.
+
+### 3. Impact of Input Scaling Strategy in Text Detection Module on PP-OCRv5 Inference Performance
+
+| Config            | Description                                                                            |
+| ----------------- | -------------------------------------------------------------------------------------- |
+| mobile_min_1280 | Uses `min` limit type and `text_det_limit_side_len=1280` with PP-OCRv5_mobile models. |
+| mobile_min_736  | Same as default, `min`, `side_len=736`.                                                |
+| mobile_max_960  | Uses `max` limit type and `side_len=960`.                                              |
+| mobile_max_640  | Uses `max` limit type and `side_len=640`.                                              |
+| server_min_1280 | Uses `min`, `side_len=1280` with PP-OCRv5_server models.                              |
+| server_min_736  | Same as default, `min`, `side_len=736`.                                                |
+| server_max_960  | Uses `max`, `side_len=960`.                                                            |
+| server_max_640  | Uses `max`, `side_len=640`.                                                            |
+
+**GPU**
+
+| Configuration     | Avg. Time per Image (s) | Avg. Characters Predicted per Second | Avg. CPU Utilization (%) | Peak RAM Usage (MB) | Avg. RAM Usage (MB) | Avg. GPU Utilization (%) | Peak VRAM Usage (MB) | Avg. VRAM Usage (MB) |
+| ----------------- | ----------------------- | ------------------------------------ | ------------------------ | ------------------- | ------------------- | ------------------------ | -------------------- | -------------------- |
+| mobile_min_1280 | 0.66 | 985.77 | 109.52 | 1878.74 | 1536.43 | 18.01 | 4050.00 | 3407.33 |
+| mobile_min_736 | 0.62 | 1054.23 | 106.35 | 1829.36 | 1521.92 | 17.42 | 4190.00 | 3114.02 |
+| mobile_max_960 | 0.52 | 1206.68 | 104.01 | 1795.27 | 1484.73 | 18.66 | 2490.00 | 2173.91 |
+| mobile_max_640 | 0.45 | 1353.49 | 103.32 | 1728.91 | 1470.64 | 18.55 | 2378.00 | 1998.62 |
+| server_min_1280 | 0.86 | 759.10 | 107.81 | 1876.31 | 1572.20 | 37.33 | 10368.00 | 8287.41 |
+| server_min_736 | 0.74 | 878.84 | 105.68 | 1899.80 | 1569.46 | 34.39 | 5402.00 | 4683.93 |
+| server_max_960 | 0.64 | 988.85 | 103.61 | 1831.31 | 1544.26 | 30.29 | 2929.33 | 2079.90 |
+| server_max_640 | 0.57 | 1036.90 | 102.89 | 1838.36 | 1532.50 | 28.91 | 3153.33 | 2743.40 |
+
+### 4. Pipeline benchmark data
+
+<details>
+<summary>Click to expand/collapse the table</summary>
+
+<table border="1">
+<tr><th>Pipeline configuration</th><th>Hardware</th><th>Avg. inference time (s)</th><th>Peak CPU utilization (%)</th><th>Avg. CPU utilization (%)</th><th>Peak host memory (MB)</th><th>Avg. host memory (MB)</th><th>Peak GPU utilization (%)</th><th>Avg. GPU utilization (%)</th><th>Peak device memory (MB)</th><th>Avg. device memory (MB)</th></tr>
+<tr>
+<td rowspan="7">OCR-default</td>
+<td>Intel 6271C</td>
+<td>3.97</td>
+<td>1015.40</td>
+<td>917.61</td>
+<td>4381.22</td>
+<td>3457.78</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>3.79</td>
+<td>1022.50</td>
+<td>921.68</td>
+<td>4675.46</td>
+<td>3585.96</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.65</td>
+<td>113.50</td>
+<td>102.48</td>
+<td>2240.15</td>
+<td>1868.44</td>
+<td>47</td>
+<td>19.60</td>
+<td>7612.00</td>
+<td>6634.15</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>1.06</td>
+<td>114.90</td>
+<td>103.05</td>
+<td>2142.66</td>
+<td>1791.43</td>
+<td>72</td>
+<td>20.01</td>
+<td>5516.00</td>
+<td>4812.81</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.65</td>
+<td>108.90</td>
+<td>101.95</td>
+<td>2456.05</td>
+<td>2080.26</td>
+<td>100</td>
+<td>36.52</td>
+<td>6736.00</td>
+<td>6017.05</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.74</td>
+<td>115.90</td>
+<td>102.22</td>
+<td>2352.88</td>
+<td>1993.39</td>
+<td>100</td>
+<td>25.56</td>
+<td>6762.00</td>
+<td>6039.93</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>1.17</td>
+<td>107.10</td>
+<td>101.78</td>
+<td>2361.88</td>
+<td>1986.61</td>
+<td>100</td>
+<td>51.11</td>
+<td>5282.00</td>
+<td>4585.10</td>
+</tr>
+<tr>
+<td rowspan="9">OCR-nopp-mobile</td>
+<td>Intel 6271C</td>
+<td>1.39</td>
+<td>1019.60</td>
+<td>1007.69</td>
+<td>2178.12</td>
+<td>1873.73</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>1.15</td>
+<td>1015.70</td>
+<td>1006.87</td>
+<td>2184.91</td>
+<td>1916.85</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Hygon 7490 + P800</td>
+<td>0.35</td>
+<td>110.80</td>
+<td>103.77</td>
+<td>2022.49</td>
+<td>1808.11</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.27</td>
+<td>110.90</td>
+<td>103.80</td>
+<td>1762.36</td>
+<td>1525.04</td>
+<td>31</td>
+<td>19.30</td>
+<td>4328.00</td>
+<td>3356.30</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.55</td>
+<td>113.80</td>
+<td>103.68</td>
+<td>1728.02</td>
+<td>1470.52</td>
+<td>38</td>
+<td>18.59</td>
+<td>4198.00</td>
+<td>3199.12</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.22</td>
+<td>111.90</td>
+<td>103.99</td>
+<td>2073.88</td>
+<td>1876.14</td>
+<td>32</td>
+<td>20.25</td>
+<td>4386.00</td>
+<td>3435.86</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.31</td>
+<td>119.90</td>
+<td>104.24</td>
+<td>2037.38</td>
+<td>1771.06</td>
+<td>52</td>
+<td>32.74</td>
+<td>3446.00</td>
+<td>2733.21</td>
+</tr>
+<tr>
+<td>M4</td>
+<td>6.51</td>
+<td>147.30</td>
+<td>106.24</td>
+<td>3550.58</td>
+<td>3236.75</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.46</td>
+<td>111.90</td>
+<td>103.11</td>
+<td>2035.38</td>
+<td>1742.39</td>
+<td>65</td>
+<td>46.77</td>
+<td>3968.00</td>
+<td>2991.91</td>
+</tr>
+<tr>
+<td rowspan="7">OCR-nopp-server</td>
+<td>Intel 6271C</td>
+<td>3.00</td>
+<td>1016.00</td>
+<td>1004.87</td>
+<td>4445.46</td>
+<td>3179.86</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>3.23</td>
+<td>1010.70</td>
+<td>1002.63</td>
+<td>4175.39</td>
+<td>3137.58</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.34</td>
+<td>110.90</td>
+<td>103.30</td>
+<td>1904.99</td>
+<td>1591.10</td>
+<td>57</td>
+<td>32.29</td>
+<td>7494.00</td>
+<td>6551.47</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.69</td>
+<td>108.90</td>
+<td>102.95</td>
+<td>1808.30</td>
+<td>1568.64</td>
+<td>72</td>
+<td>35.30</td>
+<td>5410.00</td>
+<td>4741.18</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.38</td>
+<td>109.40</td>
+<td>102.34</td>
+<td>2100.00</td>
+<td>1863.73</td>
+<td>100</td>
+<td>50.18</td>
+<td>6614.00</td>
+<td>5926.51</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.41</td>
+<td>109.00</td>
+<td>103.18</td>
+<td>2055.21</td>
+<td>1845.14</td>
+<td>100</td>
+<td>47.15</td>
+<td>6654.00</td>
+<td>5951.22</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.82</td>
+<td>104.40</td>
+<td>101.73</td>
+<td>1906.88</td>
+<td>1689.69</td>
+<td>100</td>
+<td>76.41</td>
+<td>5178.00</td>
+<td>4502.64</td>
+</tr>
+<tr>
+<td rowspan="9">OCR-nopp-min736-mobile</td>
+<td>Intel 6271C</td>
+<td>1.41</td>
+<td>1020.10</td>
+<td>1008.14</td>
+<td>2184.16</td>
+<td>1911.86</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>1.20</td>
+<td>1015.70</td>
+<td>1007.08</td>
+<td>2254.04</td>
+<td>1935.18</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Hygon 7490 + P800</td>
+<td>0.36</td>
+<td>112.90</td>
+<td>104.29</td>
+<td>2174.58</td>
+<td>1827.67</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.27</td>
+<td>113.90</td>
+<td>104.48</td>
+<td>1717.55</td>
+<td>1529.77</td>
+<td>30</td>
+<td>19.54</td>
+<td>4328.00</td>
+<td>3388.44</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.57</td>
+<td>118.80</td>
+<td>104.45</td>
+<td>1693.10</td>
+<td>1470.74</td>
+<td>40</td>
+<td>19.83</td>
+<td>4198.00</td>
+<td>3206.91</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.22</td>
+<td>113.40</td>
+<td>104.66</td>
+<td>2037.13</td>
+<td>1797.10</td>
+<td>31</td>
+<td>20.64</td>
+<td>4384.00</td>
+<td>3427.91</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.31</td>
+<td>119.30</td>
+<td>106.05</td>
+<td>1879.15</td>
+<td>1732.39</td>
+<td>49</td>
+<td>30.40</td>
+<td>3446.00</td>
+<td>2751.08</td>
+</tr>
+<tr>
+<td>M4</td>
+<td>6.39</td>
+<td>124.90</td>
+<td>107.16</td>
+<td>3578.98</td>
+<td>3209.90</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.47</td>
+<td>109.60</td>
+<td>103.26</td>
+<td>1961.40</td>
+<td>1742.95</td>
+<td>60</td>
+<td>44.26</td>
+<td>3968.00</td>
+<td>3002.81</td>
+</tr>
+<tr>
+<td rowspan="7">OCR-nopp-min736-server</td>
+<td>Intel 6271C</td>
+<td>3.26</td>
+<td>1068.50</td>
+<td>1004.96</td>
+<td>4582.52</td>
+<td>3135.68</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>3.52</td>
+<td>1010.70</td>
+<td>1002.33</td>
+<td>4723.23</td>
+<td>3209.27</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.35</td>
+<td>108.90</td>
+<td>103.94</td>
+<td>1703.65</td>
+<td>1485.50</td>
+<td>60</td>
+<td>35.54</td>
+<td>7492.00</td>
+<td>6576.97</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.71</td>
+<td>110.80</td>
+<td>103.54</td>
+<td>1800.06</td>
+<td>1559.28</td>
+<td>78</td>
+<td>36.65</td>
+<td>5410.00</td>
+<td>4741.55</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.40</td>
+<td>110.20</td>
+<td>102.75</td>
+<td>2012.64</td>
+<td>1843.45</td>
+<td>100</td>
+<td>55.74</td>
+<td>6614.00</td>
+<td>5940.44</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.44</td>
+<td>114.90</td>
+<td>103.87</td>
+<td>2002.72</td>
+<td>1773.17</td>
+<td>100</td>
+<td>49.28</td>
+<td>6654.00</td>
+<td>5980.68</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.89</td>
+<td>105.00</td>
+<td>101.91</td>
+<td>2149.31</td>
+<td>1795.35</td>
+<td>100</td>
+<td>76.39</td>
+<td>5176.00</td>
+<td>4528.77</td>
+</tr>
+<tr>
+<td rowspan="9">OCR-nopp-max640-mobile</td>
+<td>Intel 6271C</td>
+<td>1.00</td>
+<td>1033.70</td>
+<td>1005.95</td>
+<td>2021.88</td>
+<td>1743.27</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>0.88</td>
+<td>1043.60</td>
+<td>1006.77</td>
+<td>1980.82</td>
+<td>1724.51</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Hygon 7490 + P800</td>
+<td>0.28</td>
+<td>125.70</td>
+<td>101.56</td>
+<td>1962.27</td>
+<td>1782.68</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.21</td>
+<td>122.50</td>
+<td>101.87</td>
+<td>1772.39</td>
+<td>1569.55</td>
+<td>29</td>
+<td>18.74</td>
+<td>2360.00</td>
+<td>2039.07</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.43</td>
+<td>133.80</td>
+<td>101.82</td>
+<td>1636.93</td>
+<td>1464.10</td>
+<td>37</td>
+<td>20.94</td>
+<td>2386.00</td>
+<td>2055.30</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.18</td>
+<td>119.90</td>
+<td>102.12</td>
+<td>2119.93</td>
+<td>1889.49</td>
+<td>29</td>
+<td>20.92</td>
+<td>2636.00</td>
+<td>2321.11</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.24</td>
+<td>126.80</td>
+<td>101.78</td>
+<td>1905.14</td>
+<td>1739.93</td>
+<td>48</td>
+<td>30.71</td>
+<td>2232.00</td>
+<td>1911.18</td>
+</tr>
+<tr>
+<td>M4</td>
+<td>7.08</td>
+<td>137.80</td>
+<td>104.83</td>
+<td>2931.08</td>
+<td>2658.25</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.36</td>
+<td>124.80</td>
+<td>101.70</td>
+<td>1983.21</td>
+<td>1729.43</td>
+<td>61</td>
+<td>46.10</td>
+<td>2162.00</td>
+<td>1836.63</td>
+</tr>
+<tr>
+<td rowspan="9">OCR-nopp-max960-mobile</td>
+<td>Intel 6271C</td>
+<td>1.21</td>
+<td>1020.00</td>
+<td>1008.49</td>
+<td>2200.30</td>
+<td>1800.74</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>1.01</td>
+<td>1024.10</td>
+<td>1007.32</td>
+<td>2038.80</td>
+<td>1800.05</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Hygon 7490 + P800</td>
+<td>0.32</td>
+<td>107.50</td>
+<td>102.00</td>
+<td>2001.21</td>
+<td>1799.01</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.23</td>
+<td>107.70</td>
+<td>102.33</td>
+<td>1727.89</td>
+<td>1490.18</td>
+<td>30</td>
+<td>20.19</td>
+<td>2646.00</td>
+<td>2385.40</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.49</td>
+<td>109.90</td>
+<td>102.26</td>
+<td>1726.01</td>
+<td>1504.90</td>
+<td>38</td>
+<td>20.11</td>
+<td>2498.00</td>
+<td>2227.73</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.20</td>
+<td>109.90</td>
+<td>102.52</td>
+<td>1959.46</td>
+<td>1798.35</td>
+<td>28</td>
+<td>19.38</td>
+<td>2712.00</td>
+<td>2450.10</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.27</td>
+<td>102.90</td>
+<td>101.19</td>
+<td>1938.48</td>
+<td>1741.19</td>
+<td>47</td>
+<td>29.27</td>
+<td>3344.00</td>
+<td>2585.02</td>
+</tr>
+<tr>
+<td>M4</td>
+<td>5.44</td>
+<td>122.10</td>
+<td>105.91</td>
+<td>3094.72</td>
+<td>2686.52</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.41</td>
+<td>106.00</td>
+<td>101.81</td>
+<td>1859.88</td>
+<td>1722.62</td>
+<td>68</td>
+<td>47.05</td>
+<td>2264.00</td>
+<td>2001.07</td>
+</tr>
+<tr>
+<td rowspan="9">OCR-nopp-max640-server</td>
+<td>Intel 6271C</td>
+<td>2.16</td>
+<td>1026.30</td>
+<td>1005.10</td>
+<td>3467.93</td>
+<td>3074.06</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>2.30</td>
+<td>1008.70</td>
+<td>1003.32</td>
+<td>3435.54</td>
+<td>3042.62</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Hygon 7490 + P800</td>
+<td>0.35</td>
+<td>104.70</td>
+<td>101.27</td>
+<td>1948.85</td>
+<td>1779.77</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.25</td>
+<td>104.90</td>
+<td>101.42</td>
+<td>1833.93</td>
+<td>1560.71</td>
+<td>41</td>
+<td>27.61</td>
+<td>4480.00</td>
+<td>3955.14</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.56</td>
+<td>106.20</td>
+<td>101.47</td>
+<td>1669.73</td>
+<td>1500.87</td>
+<td>58</td>
+<td>31.78</td>
+<td>3160.00</td>
+<td>2838.78</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.23</td>
+<td>109.40</td>
+<td>101.45</td>
+<td>1968.77</td>
+<td>1800.81</td>
+<td>58</td>
+<td>30.81</td>
+<td>2602.00</td>
+<td>2588.77</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.30</td>
+<td>106.10</td>
+<td>101.55</td>
+<td>2027.13</td>
+<td>1749.07</td>
+<td>69</td>
+<td>39.10</td>
+<td>3318.00</td>
+<td>2795.54</td>
+</tr>
+<tr>
+<td>M4</td>
+<td>7.26</td>
+<td>133.90</td>
+<td>104.48</td>
+<td>5473.38</td>
+<td>3472.28</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.58</td>
+<td>103.90</td>
+<td>100.86</td>
+<td>1884.23</td>
+<td>1714.48</td>
+<td>84</td>
+<td>63.50</td>
+<td>2852.00</td>
+<td>2540.37</td>
+</tr>
+<tr>
+<td rowspan="9">OCR-nopp-max960-server</td>
+<td>Intel 6271C</td>
+<td>2.53</td>
+<td>1014.50</td>
+<td>1005.22</td>
+<td>3625.57</td>
+<td>3151.73</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>2.66</td>
+<td>1010.60</td>
+<td>1003.39</td>
+<td>3580.64</td>
+<td>3197.09</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Hygon 7490 + P800</td>
+<td>0.40</td>
+<td>105.90</td>
+<td>101.76</td>
+<td>2040.65</td>
+<td>1810.97</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.29</td>
+<td>108.90</td>
+<td>102.12</td>
+<td>1821.03</td>
+<td>1620.02</td>
+<td>44</td>
+<td>30.38</td>
+<td>4290.00</td>
+<td>2928.79</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.60</td>
+<td>109.90</td>
+<td>101.98</td>
+<td>1797.75</td>
+<td>1544.96</td>
+<td>61</td>
+<td>32.48</td>
+<td>2936.00</td>
+<td>2117.71</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.28</td>
+<td>108.80</td>
+<td>101.92</td>
+<td>2016.22</td>
+<td>1811.74</td>
+<td>73</td>
+<td>41.82</td>
+<td>2636.00</td>
+<td>2241.23</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.34</td>
+<td>111.00</td>
+<td>102.75</td>
+<td>1964.21</td>
+<td>1750.21</td>
+<td>68</td>
+<td>41.25</td>
+<td>2722.00</td>
+<td>2293.74</td>
+</tr>
+<tr>
+<td>M4</td>
+<td>6.28</td>
+<td>129.10</td>
+<td>103.74</td>
+<td>7780.70</td>
+<td>3571.92</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.67</td>
+<td>116.90</td>
+<td>101.33</td>
+<td>1941.09</td>
+<td>1693.39</td>
+<td>88</td>
+<td>65.48</td>
+<td>2714.00</td>
+<td>1923.06</td>
+</tr>
+<tr>
+<td rowspan="7">OCR-nopp-min1280-server</td>
+<td>Intel 6271C</td>
+<td>4.13</td>
+<td>1043.40</td>
+<td>1005.45</td>
+<td>5993.70</td>
+<td>3454.00</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>4.46</td>
+<td>1011.70</td>
+<td>996.72</td>
+<td>5633.51</td>
+<td>3489.79</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.42</td>
+<td>113.90</td>
+<td>106.08</td>
+<td>1747.88</td>
+<td>1546.18</td>
+<td>85</td>
+<td>43.73</td>
+<td>13558.00</td>
+<td>11297.98</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.82</td>
+<td>116.80</td>
+<td>105.18</td>
+<td>1873.38</td>
+<td>1609.55</td>
+<td>100</td>
+<td>39.57</td>
+<td>10376.00</td>
+<td>8427.30</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.55</td>
+<td>114.80</td>
+<td>103.14</td>
+<td>2036.36</td>
+<td>1864.45</td>
+<td>100</td>
+<td>69.67</td>
+<td>13224.00</td>
+<td>11411.31</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.55</td>
+<td>105.90</td>
+<td>101.86</td>
+<td>1931.35</td>
+<td>1764.44</td>
+<td>100</td>
+<td>56.16</td>
+<td>12418.00</td>
+<td>10510.77</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>1.13</td>
+<td>105.90</td>
+<td>102.35</td>
+<td>2066.73</td>
+<td>1787.78</td>
+<td>100</td>
+<td>83.50</td>
+<td>10142.00</td>
+<td>8338.80</td>
+</tr>
+<tr>
+<td rowspan="9">OCR-nopp-min1280-mobile</td>
+<td>Intel 6271C</td>
+<td>1.59</td>
+<td>1019.90</td>
+<td>1008.39</td>
+<td>2366.86</td>
+<td>1992.03</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>1.29</td>
+<td>1017.70</td>
+<td>1007.28</td>
+<td>2501.24</td>
+<td>2059.99</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Hygon 7490 + P800</td>
+<td>0.43</td>
+<td>120.90</td>
+<td>107.02</td>
+<td>2108.87</td>
+<td>1821.91</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.29</td>
+<td>117.90</td>
+<td>107.19</td>
+<td>1847.97</td>
+<td>1570.89</td>
+<td>31</td>
+<td>18.98</td>
+<td>3746.00</td>
+<td>3321.86</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.61</td>
+<td>122.80</td>
+<td>107.07</td>
+<td>1789.25</td>
+<td>1542.56</td>
+<td>39</td>
+<td>20.52</td>
+<td>4058.00</td>
+<td>3487.46</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.24</td>
+<td>116.80</td>
+<td>106.80</td>
+<td>2092.63</td>
+<td>1882.77</td>
+<td>28</td>
+<td>18.67</td>
+<td>3902.00</td>
+<td>3444.00</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.34</td>
+<td>125.80</td>
+<td>106.79</td>
+<td>1959.45</td>
+<td>1783.97</td>
+<td>49</td>
+<td>32.66</td>
+<td>3532.00</td>
+<td>3094.29</td>
+</tr>
+<tr>
+<td>M4</td>
+<td>6.64</td>
+<td>139.40</td>
+<td>107.63</td>
+<td>4283.97</td>
+<td>3112.59</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.51</td>
+<td>116.90</td>
+<td>105.06</td>
+<td>1927.22</td>
+<td>1675.34</td>
+<td>68</td>
+<td>45.78</td>
+<td>3828.00</td>
+<td>3283.78</td>
+</tr>
+</table>
+
+
+<table border="1">
+<tr><th>Pipeline configuration</th><th>description</th></tr>
+<tr>
+<td>OCR-default</td>
+<td>Default configuration</td>
+</tr>
+<tr>
+<td>OCR-nopp-mobile</td>
+<td>Based on the default configuration, document image preprocessing is disabled and mobile det and rec models are used</td>
+</tr>
+<tr>
+<td>OCR-nopp-server</td>
+<td>Based on the default configuration, document image preprocessing is disabled</td>
+</tr>
+<tr>
+<td>OCR-nopp-min736-mobile</td>
+<td>Based on the default configuration, document image preprocessing is disabled, det model input resizing strategy is set to min+736, and mobile det and rec models are used</td>
+</tr>
+<tr>
+<td>OCR-nopp-min736-server</td>
+<td>Based on the default configuration, document image preprocessing is disabled, and the det model input resizing strategy is set to min+736</td>
+</tr>
+<tr>
+<td>OCR-nopp-max640-mobile</td>
+<td>Based on the default configuration, document image preprocessing is disabled, det model input resizing strategy is set to max+640, and mobile det and rec models are used</td>
+</tr>
+<tr>
+<td>OCR-nopp-max960-mobile</td>
+<td>Based on the default configuration, document image preprocessing is disabled, det model input resizing strategy is set to max+960, and mobile det and rec models are used</td>
+</tr>
+<tr>
+<td>OCR-nopp-max640-server</td>
+<td>Based on the default configuration, document image preprocessing is disabled, and the det model input resizing strategy is set to max+640</td>
+</tr>
+<tr>
+<td>OCR-nopp-max960-server</td>
+<td>Based on the default configuration, document image preprocessing is disabled, and the det model input resizing strategy is set to max+960</td>
+</tr>
+<tr>
+<td>OCR-nopp-min1280-server</td>
+<td>Based on the default configuration, document image preprocessing is disabled, and the det model input resizing strategy is set to min+1280</td>
+</tr>
+<tr>
+<td>OCR-nopp-min1280-mobile</td>
+<td>Based on the default configuration, document image preprocessing is disabled, det model input resizing strategy is set to min+1280, and mobile det and rec models are used</td>
+</tr>
+</table>
+</details>
+
+
+* Test environment:
+    * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
+    * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
+    * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
+* Test data:
+    * Test data containing 200 images from document and general scenarios.
+* Test strategy:
+    * Warm up with 20 samples, then repeat the full dataset once for performance testing.
+* Note:
+    * Since we did not collect device memory data for NPU and XPU, the corresponding entries in the table are marked as N/A.
+
+**CPU**
+
+| Configuration     | Avg. Time per Image (s) | Avg. Characters Predicted per Second | Avg. CPU Utilization (%) | Peak RAM Usage (MB) | Avg. RAM Usage (MB) |
+| ----------------- | ----------------------- | ------------------------------------ | ------------------------ | ------------------- | ------------------- |
+| mobile_min_1280 | 2.00 | 326.44 | 976.83 | 2233.16 | 1867.94 |
+| mobile_min_736 | 1.75 | 371.82 | 965.89 | 2219.98 | 1830.97 |
+| mobile_max_960 | 1.49 | 422.62 | 969.11 | 2048.67 | 1677.82 |
+| mobile_max_640 | 1.31 | 459.11 | 978.41 | 2023.25 | 1616.42 |
+| server_min_1280 | 5.57 | 117.08 | 991.34 | 4452.39 | 3286.19 |
+| server_min_736 | 4.34 | 149.98 | 990.24 | 4020.85 | 3137.20 |
+| server_max_960 | 3.39 | 186.59 | 984.67 | 3492.62 | 2977.13 |
+| server_max_640 | 2.95 | 201.00 | 980.59 | 3342.38 | 2935.24 |
+
+# Deployment and Secondary Development
+* **Multiple System Support**: Compatible with mainstream operating systems including Windows, Linux, and Mac.
+* **Multiple Hardware Support**: Besides NVIDIA GPUs, it also supports inference and deployment on Intel CPU, Kunlun chips, Ascend, and other new hardware.
+* **High-Performance Inference Plugin**: Recommended to combine with high-performance inference plugins to further improve inference speed. See [High-Performance Inference Guide](../../inference_deployment/local_inference/high_performance_inference.md) for details.
+* **Service Deployment**: Supports highly stable service deployment solutions. See [Service Deployment Guide](../../inference_deployment/serving/serving.md) for details.
+* **Secondary Development Capability**: Supports custom dataset training, dictionary extension, and model fine-tuning. Example: To add Korean recognition, you can extend the dictionary and fine-tune the model, seamlessly integrating into existing pipelines. See [Text Detection Module Usage Tutorial](../../module_usage/text_detection.en.md) and [Text Recognition Module Usage Tutorial](../../module_usage/text_recognition.en.md) for details.
diff --git a/docs/version3.x/algorithm/PP-OCRv5/PP-OCRv5.md b/docs/version3.x/algorithm/PP-OCRv5/PP-OCRv5.md
new file mode 100644
index 0000000..b9724e4
--- /dev/null
+++ b/docs/version3.x/algorithm/PP-OCRv5/PP-OCRv5.md
@@ -0,0 +1,1495 @@
+# 一、PP-OCRv5简介
+**PP-OCRv5** 是PP-OCR新一代文字识别解决方案，该方案聚焦于多场景、多文字类型的文字识别。在文字类型方面，PP-OCRv5支持简体中文、中文拼音、繁体中文、英文、日文5大主流文字类型，在场景方面，PP-OCRv5升级了中英复杂手写体、竖排文本、生僻字等多种挑战性场景的识别能力。在内部多场景复杂评估集上，PP-OCRv5较PP-OCRv4端到端提升13个百分点。
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv5/algorithm_ppocrv5.png" width="600"/>
+</div>
+
+
+# 二、关键指标
+### 1. 文本检测指标
+<table>
+  <thead>
+    <tr>
+      <th>模型</th>
+      <th>手写中文</th>
+      <th>手写英文</th>
+      <th>印刷中文</th>
+      <th>印刷英文</th>
+      <th>繁体中文</th>
+      <th>古籍文本</th>
+      <th>日文</th>
+      <th>通用场景</th>
+      <th>拼音</th>
+      <th>旋转</th>
+      <th>扭曲</th>
+      <th>艺术字</th>
+      <th>平均</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td><b>PP-OCRv5_server_det</b></td>
+      <td><b>0.803</b></td>
+      <td><b>0.841</b></td>
+      <td><b>0.945</b></td>
+      <td><b>0.917</b></td>
+      <td><b>0.815</b></td>
+      <td><b>0.676</b></td>
+      <td><b>0.772</b></td>
+      <td><b>0.797</b></td>
+      <td><b>0.671</b></td>
+      <td><b>0.8</b></td>
+      <td><b>0.876</b></td>
+      <td><b>0.673</b></td>
+      <td><b>0.827</b></td>
+    </tr>
+    <tr>
+      <td>PP-OCRv4_server_det</td>
+      <td>0.706</td>
+      <td>0.249</td>
+      <td>0.888</td>
+      <td>0.690</td>
+      <td>0.759</td>
+      <td>0.473</td>
+      <td>0.685</td>
+      <td>0.715</td>
+      <td>0.542</td>
+      <td>0.366</td>
+      <td>0.775</td>
+      <td>0.583</td>
+      <td>0.662</td>
+    </tr>
+    <tr>
+      <td><b>PP-OCRv5_mobile_det</b></td>
+      <td><b>0.744</b></td>
+      <td><b>0.777</b></td>
+      <td><b>0.905</b></td>
+      <td><b>0.910</b></td>
+      <td><b>0.823</b></td>
+      <td><b>0.581</b></td>
+      <td><b>0.727</b></td>
+      <td><b>0.721</b></td>
+      <td><b>0.575</b></td>
+      <td><b>0.647</b></td>
+      <td><b>0.827</b></td>
+      <td>0.525</td>
+      <td><b>0.770</b></td>
+    </tr>
+    <tr>
+      <td>PP-OCRv4_mobile_det</td>
+      <td>0.583</td>
+      <td>0.369</td>
+      <td>0.872</td>
+      <td>0.773</td>
+      <td>0.663</td>
+      <td>0.231</td>
+      <td>0.634</td>
+      <td>0.710</td>
+      <td>0.430</td>
+      <td>0.299</td>
+      <td>0.715</td>
+      <td><b>0.549</b></td>
+      <td>0.624</td>
+    </tr>
+  </tbody>
+</table>
+
+对比PP-OCRv4，PP-OCRv5在所有检测场景下均有明显提升，尤其在手写、古籍、日文检测能力上表现更优。
+
+### 2. 文本识别指标
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv5/ocrv5_rec_acc.png" width="600"/>
+</div>
+
+
+<table>
+  <thead>
+    <tr>
+      <th>评估集类别</th>
+      <th>手写中文</th>
+      <th>手写英文</th>
+      <th>印刷中文</th>
+      <th>印刷英文</th>
+      <th>繁体中文</th>
+      <th>古籍文本</th>
+      <th>日文</th>
+      <th>易混淆字符</th>
+      <th>通用场景</th>
+      <th>拼音</th>
+      <th>竖直文本</th>
+      <th>艺术字</th>
+      <th>加权平均</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td>PP-OCRv5_server_rec</td>
+      <td><b>0.5807</b></td>
+      <td><b>0.5806</b></td>
+      <td><b>0.9013</b></td>
+      <td><b>0.8679</b></td>
+      <td><b>0.7472</b></td>
+      <td><b>0.6039</b></td>
+      <td><b>0.7372</b></td>
+      <td><b>0.5946</b></td>
+      <td><b>0.8384</b></td>
+      <td><b>0.7435</b></td>
+      <td><b>0.9314</b></td>
+      <td><b>0.6397</b></td>
+      <td><b>0.8401</b></td>
+    </tr>
+    <tr>
+      <td>PP-OCRv4_server_rec</td>
+      <td>0.3626</td>
+      <td>0.2661</td>
+      <td>0.8486</td>
+      <td>0.6677</td>
+      <td>0.4097</td>
+      <td>0.3080</td>
+      <td>0.4623</td>
+      <td>0.5028</td>
+      <td>0.8362</td>
+      <td>0.2694</td>
+      <td>0.5455</td>
+      <td>0.5892</td>
+      <td>0.5735</td>
+    </tr>
+    <tr>
+      <td>PP-OCRv5_mobile_rec</td>
+      <td><b>0.4166</b></td>
+      <td><b>0.4944</b></td>
+      <td><b>0.8605</b></td>
+      <td><b>0.8753</b></td>
+      <td><b>0.7199</b></td>
+      <td><b>0.5786</b></td>
+      <td><b>0.7577</b></td>
+      <td><b>0.5570</b></td>
+      <td>0.7703</td>
+      <td><b>0.7248</b></td>
+      <td><b>0.8089</b></td>
+      <td>0.5398</td>
+      <td><b>0.8015</b></td>
+    </tr>
+    <tr>
+      <td>PP-OCRv4_mobile_rec</td>
+      <td>0.2980</td>
+      <td>0.2550</td>
+      <td>0.8398</td>
+      <td>0.6598</td>
+      <td>0.3218</td>
+      <td>0.2593</td>
+      <td>0.4724</td>
+      <td>0.4599</td>
+      <td><b>0.8106</b></td>
+      <td>0.2593</td>
+      <td>0.5924</td>
+      <td><b>0.5555</b></td>
+      <td>0.5301</td>
+    </tr>
+  </tbody>
+</table>
+
+单模型即可覆盖多语言和多类型文本，识别精度大幅领先前代产品和主流开源方案。
+
+
+# 三、PP-OCRv5 Demo示例
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv5/algorithm_ppocrv5_demo1.png" width="600"/>
+</div>
+
+<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/PaddleX3.0/doc_images/PP-OCRv5/algorithm_ppocrv5_demo.pdf">更多示例</a>
+
+## 四、推理性能参考数据
+
+测试环境：
+
+- NVIDIA Tesla V100
+- Intel Xeon Gold 6271C
+- PaddlePaddle 3.0.0
+
+在 200 张图像（包括通用图像与文档图像）上测试。测试时从磁盘读取图像，因此读图时间及其他额外开销也被包含在总耗时内。如果将图像提前载入到内存，可进一步减少平均每图约 25 ms 的时间开销。
+
+如果不特别说明，则：
+
+- 使用 PP-OCRv4_mobile_det 和 PP-OCRv4_mobile_rec 模型。
+- 不使用文档图像方向分类、文本图像矫正、文本行方向分类。
+- 将 `text_det_limit_type` 设置为 `"min"`、`text_det_limit_side_len` 设置为 `736`。
+
+### 1. PP-OCRv5 与 PP-OCRv4 推理性能对比
+
+| 配置 | 说明 |
+| --- | --- |
+| v5_mobile | 使用 PP-OCRv5_mobile_det 和 PP-OCRv5_mobile_rec 模型。 |
+| v4_mobile | 使用 PP-OCRv4_mobile_det 和 PP-OCRv4_mobile_rec 模型。 |
+| v5_server | 使用 PP-OCRv5_server_det 和 PP-OCRv5_server_rec 模型。 |
+| v4_server | 使用 PP-OCRv4_server_det 和 PP-OCRv4_server_rec 模型。 |
+
+**GPU：**
+
+| 配置      | 平均每图耗时（s） | 平均每秒预测字符数量 | 平均 CPU 利用率（%） | 峰值 RAM 用量（MB） | 平均 RAM 用量（MB） | 平均 GPU 利用率（%） | 峰值 VRAM 用量（MB） | 平均 VRAM 用量（MB） |
+|-----------|------------------|--------------------|--------------------|-------------------|-------------------|-------------------|-------------------|-------------------|
+| v5_mobile | 0.62 | 1054.23 | 106.35 | 1829.36 | 1521.92 | 17.42 | 4190.00  | 3114.02 |
+| v4_mobile | 0.29 | 2062.53 | 112.21 | 1713.10 | 1456.14 | 26.53  | 1304.00 | 1166.68 |
+| v5_server | 0.74 | 878.84 | 105.68 | 1899.80 | 1569.46 | 34.39 | 5402.00 | 4683.93 |
+| v4_server | 0.47 | 1322.06 | 108.06 | 1773.10 | 1518.94 | 55.25 | 6760.67 | 5788.02 |
+
+**CPU：**
+
+| 配置 | 平均每图耗时（s） | 平均每秒预测字符数量 | 平均 CPU 利用率（%） | 峰值 RAM 用量（MB） | 平均 RAM 用量（MB） |
+| --------- | ---- | ------- | ------ | -------- | -------- |
+| v5_mobile | 1.75 | 371.82 | 965.89 | 2219.98 | 1830.97 |
+| v4_mobile | 1.37 | 444.27 | 1007.33 | 2090.53 | 1797.76 |
+| v5_server | 4.34 | 149.98 | 990.24 | 4020.85 | 3137.20 |
+| v4_server | 5.42 | 115.20 | 999.03 | 4018.35 | 3105.29 |
+
+> 说明：PP-OCRv5 的识别模型使用了更大的字典，需要更长的推理时间，导致 PP-OCRv5 的推理速度慢于 PP-OCRv4。
+
+### 2. 使用辅助功能对 PP-OCRv5 推理性能的影响
+
+| 配置 | 说明 |
+| --- | --- |
+| base | 不使用文档图像方向分类、文本图像矫正、文本行方向分类。 |
+| with_textline | 使用文本行方向分类，不使用文档图像方向分类、文本图像矫正。 |
+| with_all | 使用文档图像方向分类、文本图像矫正、文本行方向分类。 |
+
+**GPU：**
+
+| 配置 | 平均每图耗时（s） | 平均每秒预测字符数量 | 平均 CPU 利用率（%） | 峰值 RAM 用量（MB） | 平均 RAM 用量（MB） | 平均GPU利用率（%） | 峰值 VRAM 用量（MB） | 平均 VRAM 用量（MB） |
+| --- | --- | --- | --- | --- | --- | --- | --- | --- |
+| base | 0.62 | 1054.23 | 106.35 | 1829.36 | 1521.92 | 17.42 | 4190.00 | 3114.02 |
+| with_textline | 0.64 | 1012.32 | 106.37 | 1867.69 | 1527.42 | 19.16 | 4198.00 | 3115.05 |
+| with_all | 1.09 | 562.99 | 105.67 | 2381.53 | 1792.48 | 10.77 | 2480.00 | 2065.54 |
+
+**CPU：**
+
+| 配置 | 平均每图耗时（s） | 平均每秒预测字符数量 | 平均 CPU 利用率（%） | 峰值 RAM 用量（MB） | 平均 RAM 用量（MB） |
+| --- | --- | --- | --- | --- | --- |
+| base | 1.75 | 371.82 | 965.89 | 2219.98 | 1830.97 |
+| with_textline | 1.87 | 347.61 | 972.08 | 2232.38 | 1822.13 |
+| with_all | 3.13 | 195.25 | 828.37 | 2751.47 | 2179.70 |
+
+> 说明：文本图像矫正等辅助功能会对端到端推理精度造成影响，因此并不一定使用的辅助功能越多、资源用量越大。
+
+### 3. 文本检测模块输入缩放尺寸策略对 PP-OCRv5 推理性能的影响
+
+| 配置 | 说明 |
+| --- | --- |
+| mobile_min_1280 | 使用 PP-OCRv5_mobile_det 和 PP-OCRv5_mobile_rec 模型，将 `text_det_limit_type` 设置为 `"min"`、`text_det_limit_side_len` 设置为 `1280`。 |
+| mobile_min_736 | 使用 PP-OCRv5_mobile_det 和 PP-OCRv5_mobile_rec 模型，将 `text_det_limit_type` 设置为 `"min"`、`text_det_limit_side_len` 设置为 `736`。 |
+| mobile_max_960 | 使用 PP-OCRv5_mobile_det 和 PP-OCRv5_mobile_rec 模型，将 `text_det_limit_type` 设置为 `"max"`、`text_det_limit_side_len` 设置为 `960`。 |
+| mobile_max_640 | 使用 PP-OCRv5_mobile_det 和 PP-OCRv5_mobile_rec 模型，将 `text_det_limit_type` 设置为 `"max"`、`text_det_limit_side_len` 设置为 `640`。 |
+| server_min_1280 | 使用 PP-OCRv5_server_det 和 PP-OCRv5_server_rec 模型，将 `text_det_limit_type` 设置为 `"min"`、`text_det_limit_side_len` 设置为 `1280`。 |
+| server_min_736 | 使用 PP-OCRv5_server_det 和 PP-OCRv5_server_rec 模型，将 `text_det_limit_type` 设置为 `"min"`、`text_det_limit_side_len` 设置为 `736`。 |
+| server_max_960 | 使用 PP-OCRv5_server_det 和 PP-OCRv5_server_rec 模型，将 `text_det_limit_type` 设置为 `"max"`、`text_det_limit_side_len` 设置为 `960`。 |
+| server_max_640 | 使用 PP-OCRv5_server_det 和 PP-OCRv5_server_rec 模型，将 `text_det_limit_type` 设置为 `"max"`、`text_det_limit_side_len` 设置为 `640`。 |
+
+**GPU：**
+
+| 配置 | 平均每图耗时（s） | 平均每秒预测字符数量 | 平均 CPU 利用率（%） | 峰值 RAM 用量（MB） | 平均 RAM 用量（MB） | 平均GPU利用率（%） | 峰值 VRAM 用量（MB） | 平均 VRAM 用量（MB） |
+| --- | --- | --- | --- | --- | --- | --- | --- | --- |
+| mobile_min_1280 | 0.66 | 985.77 | 109.52 | 1878.74 | 1536.43 | 18.01 | 4050.00 | 3407.33 |
+| mobile_min_736 | 0.62 | 1054.23 | 106.35 | 1829.36 | 1521.92 | 17.42 | 4190.00 | 3114.02 |
+| mobile_max_960 | 0.52 | 1206.68 | 104.01 | 1795.27 | 1484.73 | 18.66 | 2490.00 | 2173.91 |
+| mobile_max_640 | 0.45 | 1353.49 | 103.32 | 1728.91 | 1470.64 | 18.55 | 2378.00 | 1998.62 |
+| server_min_1280 | 0.86 | 759.10 | 107.81 | 1876.31 | 1572.20 | 37.33 | 10368.00 | 8287.41 |
+| server_min_736 | 0.74 | 878.84 | 105.68 | 1899.80 | 1569.46 | 34.39 | 5402.00 | 4683.93 |
+| server_max_960 | 0.64 | 988.85 | 103.61 | 1831.31 | 1544.26 | 30.29 | 2929.33 | 2079.90 |
+| server_max_640 | 0.57 | 1036.90 | 102.89 | 1838.36 | 1532.50 | 28.91 | 3153.33 | 2743.40 |
+
+**CPU：**
+
+| 配置 | 平均每图耗时（s） | 平均每秒预测字符数量 | 平均 CPU 利用率（%） | 峰值 RAM 用量（MB） | 平均 RAM 用量（MB） |
+| --- | --- | --- | --- | --- | --- |
+| mobile_min_1280 | 2.00 | 326.44 | 976.83 | 2233.16 | 1867.94 |
+| mobile_min_736 | 1.75 | 371.82 | 965.89 | 2219.98 | 1830.97 |
+| mobile_max_960 | 1.49 | 422.62 | 969.11 | 2048.67 | 1677.82 |
+| mobile_max_640 | 1.31 | 459.11 | 978.41 | 2023.25 | 1616.42 |
+| server_min_1280 | 5.57 | 117.08 | 991.34 | 4452.39 | 3286.19 |
+| server_min_736 | 4.34 | 149.98 | 990.24 | 4020.85 | 3137.20 |
+| server_max_960 | 3.39 | 186.59 | 984.67 | 3492.62 | 2977.13 |
+| server_max_640 | 2.95 | 201.00 | 980.59 | 3342.38 | 2935.24 |
+
+### 4. 产线基准测试数据
+
+<details>
+<summary>点击展开/折叠表格</summary>
+
+<table border="1">
+<tr><th>流水线配置</th><th>硬件</th><th>平均推理时间 (s)</th><th>峰值CPU利用率 (%)</th><th>平均CPU利用率 (%)</th><th>峰值主机内存 (MB)</th><th>平均主机内存 (MB)</th><th>峰值GPU利用率 (%)</th><th>平均GPU利用率 (%)</th><th>峰值设备内存 (MB)</th><th>平均设备内存 (MB)</th></tr>
+<tr>
+<td rowspan="7">OCR-default</td>
+<td>Intel 6271C</td>
+<td>3.97</td>
+<td>1015.40</td>
+<td>917.61</td>
+<td>4381.22</td>
+<td>3457.78</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>3.79</td>
+<td>1022.50</td>
+<td>921.68</td>
+<td>4675.46</td>
+<td>3585.96</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.65</td>
+<td>113.50</td>
+<td>102.48</td>
+<td>2240.15</td>
+<td>1868.44</td>
+<td>47</td>
+<td>19.60</td>
+<td>7612.00</td>
+<td>6634.15</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>1.06</td>
+<td>114.90</td>
+<td>103.05</td>
+<td>2142.66</td>
+<td>1791.43</td>
+<td>72</td>
+<td>20.01</td>
+<td>5516.00</td>
+<td>4812.81</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.65</td>
+<td>108.90</td>
+<td>101.95</td>
+<td>2456.05</td>
+<td>2080.26</td>
+<td>100</td>
+<td>36.52</td>
+<td>6736.00</td>
+<td>6017.05</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.74</td>
+<td>115.90</td>
+<td>102.22</td>
+<td>2352.88</td>
+<td>1993.39</td>
+<td>100</td>
+<td>25.56</td>
+<td>6762.00</td>
+<td>6039.93</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>1.17</td>
+<td>107.10</td>
+<td>101.78</td>
+<td>2361.88</td>
+<td>1986.61</td>
+<td>100</td>
+<td>51.11</td>
+<td>5282.00</td>
+<td>4585.10</td>
+</tr>
+<tr>
+<td rowspan="9">OCR-nopp-mobile</td>
+<td>Intel 6271C</td>
+<td>1.39</td>
+<td>1019.60</td>
+<td>1007.69</td>
+<td>2178.12</td>
+<td>1873.73</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>1.15</td>
+<td>1015.70</td>
+<td>1006.87</td>
+<td>2184.91</td>
+<td>1916.85</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Hygon 7490 + P800</td>
+<td>0.35</td>
+<td>110.80</td>
+<td>103.77</td>
+<td>2022.49</td>
+<td>1808.11</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.27</td>
+<td>110.90</td>
+<td>103.80</td>
+<td>1762.36</td>
+<td>1525.04</td>
+<td>31</td>
+<td>19.30</td>
+<td>4328.00</td>
+<td>3356.30</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.55</td>
+<td>113.80</td>
+<td>103.68</td>
+<td>1728.02</td>
+<td>1470.52</td>
+<td>38</td>
+<td>18.59</td>
+<td>4198.00</td>
+<td>3199.12</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.22</td>
+<td>111.90</td>
+<td>103.99</td>
+<td>2073.88</td>
+<td>1876.14</td>
+<td>32</td>
+<td>20.25</td>
+<td>4386.00</td>
+<td>3435.86</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.31</td>
+<td>119.90</td>
+<td>104.24</td>
+<td>2037.38</td>
+<td>1771.06</td>
+<td>52</td>
+<td>32.74</td>
+<td>3446.00</td>
+<td>2733.21</td>
+</tr>
+<tr>
+<td>M4</td>
+<td>6.51</td>
+<td>147.30</td>
+<td>106.24</td>
+<td>3550.58</td>
+<td>3236.75</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.46</td>
+<td>111.90</td>
+<td>103.11</td>
+<td>2035.38</td>
+<td>1742.39</td>
+<td>65</td>
+<td>46.77</td>
+<td>3968.00</td>
+<td>2991.91</td>
+</tr>
+<tr>
+<td rowspan="7">OCR-nopp-server</td>
+<td>Intel 6271C</td>
+<td>3.00</td>
+<td>1016.00</td>
+<td>1004.87</td>
+<td>4445.46</td>
+<td>3179.86</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>3.23</td>
+<td>1010.70</td>
+<td>1002.63</td>
+<td>4175.39</td>
+<td>3137.58</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.34</td>
+<td>110.90</td>
+<td>103.30</td>
+<td>1904.99</td>
+<td>1591.10</td>
+<td>57</td>
+<td>32.29</td>
+<td>7494.00</td>
+<td>6551.47</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.69</td>
+<td>108.90</td>
+<td>102.95</td>
+<td>1808.30</td>
+<td>1568.64</td>
+<td>72</td>
+<td>35.30</td>
+<td>5410.00</td>
+<td>4741.18</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.38</td>
+<td>109.40</td>
+<td>102.34</td>
+<td>2100.00</td>
+<td>1863.73</td>
+<td>100</td>
+<td>50.18</td>
+<td>6614.00</td>
+<td>5926.51</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.41</td>
+<td>109.00</td>
+<td>103.18</td>
+<td>2055.21</td>
+<td>1845.14</td>
+<td>100</td>
+<td>47.15</td>
+<td>6654.00</td>
+<td>5951.22</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.82</td>
+<td>104.40</td>
+<td>101.73</td>
+<td>1906.88</td>
+<td>1689.69</td>
+<td>100</td>
+<td>76.41</td>
+<td>5178.00</td>
+<td>4502.64</td>
+</tr>
+<tr>
+<td rowspan="9">OCR-nopp-min736-mobile</td>
+<td>Intel 6271C</td>
+<td>1.41</td>
+<td>1020.10</td>
+<td>1008.14</td>
+<td>2184.16</td>
+<td>1911.86</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>1.20</td>
+<td>1015.70</td>
+<td>1007.08</td>
+<td>2254.04</td>
+<td>1935.18</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Hygon 7490 + P800</td>
+<td>0.36</td>
+<td>112.90</td>
+<td>104.29</td>
+<td>2174.58</td>
+<td>1827.67</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.27</td>
+<td>113.90</td>
+<td>104.48</td>
+<td>1717.55</td>
+<td>1529.77</td>
+<td>30</td>
+<td>19.54</td>
+<td>4328.00</td>
+<td>3388.44</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.57</td>
+<td>118.80</td>
+<td>104.45</td>
+<td>1693.10</td>
+<td>1470.74</td>
+<td>40</td>
+<td>19.83</td>
+<td>4198.00</td>
+<td>3206.91</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.22</td>
+<td>113.40</td>
+<td>104.66</td>
+<td>2037.13</td>
+<td>1797.10</td>
+<td>31</td>
+<td>20.64</td>
+<td>4384.00</td>
+<td>3427.91</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.31</td>
+<td>119.30</td>
+<td>106.05</td>
+<td>1879.15</td>
+<td>1732.39</td>
+<td>49</td>
+<td>30.40</td>
+<td>3446.00</td>
+<td>2751.08</td>
+</tr>
+<tr>
+<td>M4</td>
+<td>6.39</td>
+<td>124.90</td>
+<td>107.16</td>
+<td>3578.98</td>
+<td>3209.90</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.47</td>
+<td>109.60</td>
+<td>103.26</td>
+<td>1961.40</td>
+<td>1742.95</td>
+<td>60</td>
+<td>44.26</td>
+<td>3968.00</td>
+<td>3002.81</td>
+</tr>
+<tr>
+<td rowspan="7">OCR-nopp-min736-server</td>
+<td>Intel 6271C</td>
+<td>3.26</td>
+<td>1068.50</td>
+<td>1004.96</td>
+<td>4582.52</td>
+<td>3135.68</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>3.52</td>
+<td>1010.70</td>
+<td>1002.33</td>
+<td>4723.23</td>
+<td>3209.27</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.35</td>
+<td>108.90</td>
+<td>103.94</td>
+<td>1703.65</td>
+<td>1485.50</td>
+<td>60</td>
+<td>35.54</td>
+<td>7492.00</td>
+<td>6576.97</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.71</td>
+<td>110.80</td>
+<td>103.54</td>
+<td>1800.06</td>
+<td>1559.28</td>
+<td>78</td>
+<td>36.65</td>
+<td>5410.00</td>
+<td>4741.55</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.40</td>
+<td>110.20</td>
+<td>102.75</td>
+<td>2012.64</td>
+<td>1843.45</td>
+<td>100</td>
+<td>55.74</td>
+<td>6614.00</td>
+<td>5940.44</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.44</td>
+<td>114.90</td>
+<td>103.87</td>
+<td>2002.72</td>
+<td>1773.17</td>
+<td>100</td>
+<td>49.28</td>
+<td>6654.00</td>
+<td>5980.68</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.89</td>
+<td>105.00</td>
+<td>101.91</td>
+<td>2149.31</td>
+<td>1795.35</td>
+<td>100</td>
+<td>76.39</td>
+<td>5176.00</td>
+<td>4528.77</td>
+</tr>
+<tr>
+<td rowspan="9">OCR-nopp-max640-mobile</td>
+<td>Intel 6271C</td>
+<td>1.00</td>
+<td>1033.70</td>
+<td>1005.95</td>
+<td>2021.88</td>
+<td>1743.27</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>0.88</td>
+<td>1043.60</td>
+<td>1006.77</td>
+<td>1980.82</td>
+<td>1724.51</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Hygon 7490 + P800</td>
+<td>0.28</td>
+<td>125.70</td>
+<td>101.56</td>
+<td>1962.27</td>
+<td>1782.68</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.21</td>
+<td>122.50</td>
+<td>101.87</td>
+<td>1772.39</td>
+<td>1569.55</td>
+<td>29</td>
+<td>18.74</td>
+<td>2360.00</td>
+<td>2039.07</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.43</td>
+<td>133.80</td>
+<td>101.82</td>
+<td>1636.93</td>
+<td>1464.10</td>
+<td>37</td>
+<td>20.94</td>
+<td>2386.00</td>
+<td>2055.30</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.18</td>
+<td>119.90</td>
+<td>102.12</td>
+<td>2119.93</td>
+<td>1889.49</td>
+<td>29</td>
+<td>20.92</td>
+<td>2636.00</td>
+<td>2321.11</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.24</td>
+<td>126.80</td>
+<td>101.78</td>
+<td>1905.14</td>
+<td>1739.93</td>
+<td>48</td>
+<td>30.71</td>
+<td>2232.00</td>
+<td>1911.18</td>
+</tr>
+<tr>
+<td>M4</td>
+<td>7.08</td>
+<td>137.80</td>
+<td>104.83</td>
+<td>2931.08</td>
+<td>2658.25</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.36</td>
+<td>124.80</td>
+<td>101.70</td>
+<td>1983.21</td>
+<td>1729.43</td>
+<td>61</td>
+<td>46.10</td>
+<td>2162.00</td>
+<td>1836.63</td>
+</tr>
+<tr>
+<td rowspan="9">OCR-nopp-max960-mobile</td>
+<td>Intel 6271C</td>
+<td>1.21</td>
+<td>1020.00</td>
+<td>1008.49</td>
+<td>2200.30</td>
+<td>1800.74</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>1.01</td>
+<td>1024.10</td>
+<td>1007.32</td>
+<td>2038.80</td>
+<td>1800.05</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Hygon 7490 + P800</td>
+<td>0.32</td>
+<td>107.50</td>
+<td>102.00</td>
+<td>2001.21</td>
+<td>1799.01</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.23</td>
+<td>107.70</td>
+<td>102.33</td>
+<td>1727.89</td>
+<td>1490.18</td>
+<td>30</td>
+<td>20.19</td>
+<td>2646.00</td>
+<td>2385.40</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.49</td>
+<td>109.90</td>
+<td>102.26</td>
+<td>1726.01</td>
+<td>1504.90</td>
+<td>38</td>
+<td>20.11</td>
+<td>2498.00</td>
+<td>2227.73</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.20</td>
+<td>109.90</td>
+<td>102.52</td>
+<td>1959.46</td>
+<td>1798.35</td>
+<td>28</td>
+<td>19.38</td>
+<td>2712.00</td>
+<td>2450.10</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.27</td>
+<td>102.90</td>
+<td>101.19</td>
+<td>1938.48</td>
+<td>1741.19</td>
+<td>47</td>
+<td>29.27</td>
+<td>3344.00</td>
+<td>2585.02</td>
+</tr>
+<tr>
+<td>M4</td>
+<td>5.44</td>
+<td>122.10</td>
+<td>105.91</td>
+<td>3094.72</td>
+<td>2686.52</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.41</td>
+<td>106.00</td>
+<td>101.81</td>
+<td>1859.88</td>
+<td>1722.62</td>
+<td>68</td>
+<td>47.05</td>
+<td>2264.00</td>
+<td>2001.07</td>
+</tr>
+<tr>
+<td rowspan="9">OCR-nopp-max640-server</td>
+<td>Intel 6271C</td>
+<td>2.16</td>
+<td>1026.30</td>
+<td>1005.10</td>
+<td>3467.93</td>
+<td>3074.06</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>2.30</td>
+<td>1008.70</td>
+<td>1003.32</td>
+<td>3435.54</td>
+<td>3042.62</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Hygon 7490 + P800</td>
+<td>0.35</td>
+<td>104.70</td>
+<td>101.27</td>
+<td>1948.85</td>
+<td>1779.77</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.25</td>
+<td>104.90</td>
+<td>101.42</td>
+<td>1833.93</td>
+<td>1560.71</td>
+<td>41</td>
+<td>27.61</td>
+<td>4480.00</td>
+<td>3955.14</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.56</td>
+<td>106.20</td>
+<td>101.47</td>
+<td>1669.73</td>
+<td>1500.87</td>
+<td>58</td>
+<td>31.78</td>
+<td>3160.00</td>
+<td>2838.78</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.23</td>
+<td>109.40</td>
+<td>101.45</td>
+<td>1968.77</td>
+<td>1800.81</td>
+<td>58</td>
+<td>30.81</td>
+<td>2602.00</td>
+<td>2588.77</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.30</td>
+<td>106.10</td>
+<td>101.55</td>
+<td>2027.13</td>
+<td>1749.07</td>
+<td>69</td>
+<td>39.10</td>
+<td>3318.00</td>
+<td>2795.54</td>
+</tr>
+<tr>
+<td>M4</td>
+<td>7.26</td>
+<td>133.90</td>
+<td>104.48</td>
+<td>5473.38</td>
+<td>3472.28</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.58</td>
+<td>103.90</td>
+<td>100.86</td>
+<td>1884.23</td>
+<td>1714.48</td>
+<td>84</td>
+<td>63.50</td>
+<td>2852.00</td>
+<td>2540.37</td>
+</tr>
+<tr>
+<td rowspan="9">OCR-nopp-max960-server</td>
+<td>Intel 6271C</td>
+<td>2.53</td>
+<td>1014.50</td>
+<td>1005.22</td>
+<td>3625.57</td>
+<td>3151.73</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>2.66</td>
+<td>1010.60</td>
+<td>1003.39</td>
+<td>3580.64</td>
+<td>3197.09</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Hygon 7490 + P800</td>
+<td>0.40</td>
+<td>105.90</td>
+<td>101.76</td>
+<td>2040.65</td>
+<td>1810.97</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.29</td>
+<td>108.90</td>
+<td>102.12</td>
+<td>1821.03</td>
+<td>1620.02</td>
+<td>44</td>
+<td>30.38</td>
+<td>4290.00</td>
+<td>2928.79</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.60</td>
+<td>109.90</td>
+<td>101.98</td>
+<td>1797.75</td>
+<td>1544.96</td>
+<td>61</td>
+<td>32.48</td>
+<td>2936.00</td>
+<td>2117.71</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.28</td>
+<td>108.80</td>
+<td>101.92</td>
+<td>2016.22</td>
+<td>1811.74</td>
+<td>73</td>
+<td>41.82</td>
+<td>2636.00</td>
+<td>2241.23</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.34</td>
+<td>111.00</td>
+<td>102.75</td>
+<td>1964.21</td>
+<td>1750.21</td>
+<td>68</td>
+<td>41.25</td>
+<td>2722.00</td>
+<td>2293.74</td>
+</tr>
+<tr>
+<td>M4</td>
+<td>6.28</td>
+<td>129.10</td>
+<td>103.74</td>
+<td>7780.70</td>
+<td>3571.92</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.67</td>
+<td>116.90</td>
+<td>101.33</td>
+<td>1941.09</td>
+<td>1693.39</td>
+<td>88</td>
+<td>65.48</td>
+<td>2714.00</td>
+<td>1923.06</td>
+</tr>
+<tr>
+<td rowspan="7">OCR-nopp-min1280-server</td>
+<td>Intel 6271C</td>
+<td>4.13</td>
+<td>1043.40</td>
+<td>1005.45</td>
+<td>5993.70</td>
+<td>3454.00</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>4.46</td>
+<td>1011.70</td>
+<td>996.72</td>
+<td>5633.51</td>
+<td>3489.79</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.42</td>
+<td>113.90</td>
+<td>106.08</td>
+<td>1747.88</td>
+<td>1546.18</td>
+<td>85</td>
+<td>43.73</td>
+<td>13558.00</td>
+<td>11297.98</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.82</td>
+<td>116.80</td>
+<td>105.18</td>
+<td>1873.38</td>
+<td>1609.55</td>
+<td>100</td>
+<td>39.57</td>
+<td>10376.00</td>
+<td>8427.30</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.55</td>
+<td>114.80</td>
+<td>103.14</td>
+<td>2036.36</td>
+<td>1864.45</td>
+<td>100</td>
+<td>69.67</td>
+<td>13224.00</td>
+<td>11411.31</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.55</td>
+<td>105.90</td>
+<td>101.86</td>
+<td>1931.35</td>
+<td>1764.44</td>
+<td>100</td>
+<td>56.16</td>
+<td>12418.00</td>
+<td>10510.77</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>1.13</td>
+<td>105.90</td>
+<td>102.35</td>
+<td>2066.73</td>
+<td>1787.78</td>
+<td>100</td>
+<td>83.50</td>
+<td>10142.00</td>
+<td>8338.80</td>
+</tr>
+<tr>
+<td rowspan="9">OCR-nopp-min1280-mobile</td>
+<td>Intel 6271C</td>
+<td>1.59</td>
+<td>1019.90</td>
+<td>1008.39</td>
+<td>2366.86</td>
+<td>1992.03</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>1.29</td>
+<td>1017.70</td>
+<td>1007.28</td>
+<td>2501.24</td>
+<td>2059.99</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Hygon 7490 + P800</td>
+<td>0.43</td>
+<td>120.90</td>
+<td>107.02</td>
+<td>2108.87</td>
+<td>1821.91</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.29</td>
+<td>117.90</td>
+<td>107.19</td>
+<td>1847.97</td>
+<td>1570.89</td>
+<td>31</td>
+<td>18.98</td>
+<td>3746.00</td>
+<td>3321.86</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>0.61</td>
+<td>122.80</td>
+<td>107.07</td>
+<td>1789.25</td>
+<td>1542.56</td>
+<td>39</td>
+<td>20.52</td>
+<td>4058.00</td>
+<td>3487.46</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.24</td>
+<td>116.80</td>
+<td>106.80</td>
+<td>2092.63</td>
+<td>1882.77</td>
+<td>28</td>
+<td>18.67</td>
+<td>3902.00</td>
+<td>3444.00</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.34</td>
+<td>125.80</td>
+<td>106.79</td>
+<td>1959.45</td>
+<td>1783.97</td>
+<td>49</td>
+<td>32.66</td>
+<td>3532.00</td>
+<td>3094.29</td>
+</tr>
+<tr>
+<td>M4</td>
+<td>6.64</td>
+<td>139.40</td>
+<td>107.63</td>
+<td>4283.97</td>
+<td>3112.59</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>0.51</td>
+<td>116.90</td>
+<td>105.06</td>
+<td>1927.22</td>
+<td>1675.34</td>
+<td>68</td>
+<td>45.78</td>
+<td>3828.00</td>
+<td>3283.78</td>
+</tr>
+</table>
+
+
+<table border="1">
+<tr><th>Pipeline configuration</th><th>description</th></tr>
+<tr>
+<td>OCR-default</td>
+<td>默认配置</td>
+</tr>
+<tr>
+<td>OCR-nopp-mobile</td>
+<td>默认配置基础上，关闭文档图像预处理，使用mobile的det和rec模型</td>
+</tr>
+<tr>
+<td>OCR-nopp-server</td>
+<td>默认配置基础上，关闭文档图像预处理</td>
+</tr>
+<tr>
+<td>OCR-nopp-min736-mobile</td>
+<td>默认配置基础上，关闭文档图像预处理，det模型输入缩放策略为min+736，使用mobile的det和rec模型</td>
+</tr>
+<tr>
+<td>OCR-nopp-min736-server</td>
+<td>默认配置基础上，关闭文档图像预处理，det模型输入缩放策略为min+736</td>
+</tr>
+<tr>
+<td>OCR-nopp-max640-mobile</td>
+<td>默认配置基础上，关闭文档图像预处理，det模型输入缩放策略为max+640，使用mobile的det和rec模型</td>
+</tr>
+<tr>
+<td>OCR-nopp-max960-mobile</td>
+<td>默认配置基础上，关闭文档图像预处理，det模型输入缩放策略为max+960，使用mobile的det和rec模型</td>
+</tr>
+<tr>
+<td>OCR-nopp-max640-server</td>
+<td>默认配置基础上，关闭文档图像预处理，det模型输入缩放策略为max+640</td>
+</tr>
+<tr>
+<td>OCR-nopp-max960-server</td>
+<td>默认配置基础上，关闭文档图像预处理，det模型输入缩放策略为max+960</td>
+</tr>
+<tr>
+<td>OCR-nopp-min1280-server</td>
+<td>默认配置基础上，关闭文档图像预处理，det模型输入缩放策略为min+1280</td>
+</tr>
+<tr>
+<td>OCR-nopp-min1280-mobile</td>
+<td>默认配置基础上，关闭文档图像预处理，det模型输入缩放策略为min+1280，使用mobile的det和rec模型</td>
+</tr>
+</table>
+</details>
+
+
+* 测试环境：
+    * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
+    * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
+    * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
+* 测试数据：
+    * 测试数据包含文档场景和通用场景的200张图像。
+* 测试策略：
+    * 使用 20 个样本进行预热，然后对整个数据集重复 1 次以进行速度性能测试。
+* 备注：
+    * 由于我们没有收集NPU和XPU的设备内存数据，因此表中相应位置的数据标记为N/A。
+
+# 五、部署与二次开发
+* **多系统支持**：兼容Windows、Linux、Mac等主流操作系统。
+* **多硬件支持**：除了英伟达GPU外，还支持Intel CPU、昆仑芯、昇腾等新硬件推理和部署。
+* **高性能推理插件**：推荐结合高性能推理插件进一步提升推理速度，详见[高性能推理指南](../../inference_deployment/local_inference/high_performance_inference.md)。
+* **服务化部署**：支持高稳定性服务化部署方案，详见[服务化部署指南](../../inference_deployment/serving/serving.md)。
+* **二次开发能力**：支持自定义数据集训练、字典扩展、模型微调。举例：如需增加韩文识别，可扩展字典并微调模型，无缝集成到现有产线，详见[文本检测模块使用教程](../../module_usage/text_detection.md)及[文本识别模块使用教程](../../module_usage/text_recognition.md)
diff --git a/docs/version3.x/algorithm/PP-OCRv5/PP-OCRv5_multi_languages.en.md b/docs/version3.x/algorithm/PP-OCRv5/PP-OCRv5_multi_languages.en.md
new file mode 100644
index 0000000..a143a77
--- /dev/null
+++ b/docs/version3.x/algorithm/PP-OCRv5/PP-OCRv5_multi_languages.en.md
@@ -0,0 +1,258 @@
+---
+comments: true
+---
+
+# 1. Introduction to PP-OCRv5 Multilingual Text Recognition
+
+[PP-OCRv5](./PP-OCRv5.md) is the latest generation text recognition solution in the PP-OCR series, focusing on multi-scenario and multilingual text recognition tasks. In terms of supported text types, the default configuration of the recognition model can accurately identify five major types: Simplified Chinese, Pinyin, Traditional Chinese, English, and Japanese. Additionally, PP-OCRv5 offers multilingual text recognition capabilities covering 106 languages, including Korean, Spanish, French, Portuguese, German, Italian, Russian, Thai, Greek and more (for a full list of supported languages and abbreviations, see [Section 4](#4-supported-languages-and-abbreviations)). Compared to the previous PP-OCRv3 version, PP-OCRv5 achieves over a 30% improvement in accuracy for multilingual text recognition.
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/french_0_res.jpg" alt="French recognition result" width="500"/>
+  <br>
+  <b>French Recognition Result</b>
+</div>
+
+<br>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/german_0_res.png" alt="German recognition result" width="500"/>
+  <br>
+  <b>German Recognition Result</b>
+</div>
+
+<br>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/korean_1_res.jpg" alt="Korean recognition result" width="500"/>
+  <br>
+  <b>Korean Recognition Result</b>
+</div>
+
+<br>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/ru_0.jpeg" alt="Russian recognition result" width="500"/>
+  <br>
+  <b>Russian Recognition Result</b>
+</div>
+
+
+<br>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/th_0_res.jpg" alt="Thai recognition result" width="500"/>
+  <br>
+  <b>Thai recognition result</b>
+</div>
+
+<br>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/el_0_res.jpg" alt="Greek recognition result" width="500"/>
+  <br>
+  <b>Greek recognition result</b>
+</div>
+
+<br>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/ar_0_res.jpg" alt="Arabic OCR Result" width="500"/>
+  <br>
+  <b>Arabic recognition Result</b>
+</div>
+
+<br>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/hi_0_res.jpg" alt="Hindi OCR Result" width="500"/>
+  <br>
+  <b>Hindi recognition Result</b>
+</div>
+
+<br>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/ta_0_rec.jpg" alt="Tamil OCR Result" width="500"/>
+  <br>
+  <b>Tamil recognition Result</b>
+</div>
+
+<br>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/te_0_res.png" alt="Telugu OCR Result" width="500"/>
+  <br>
+  <b>Telugu recognition Result</b>
+</div>
+
+
+## 2. Quick Start
+
+You can specify the language for text recognition by using the `--lang` parameter when running the general OCR pipeline in the command line:
+
+```bash
+# Use the `--lang` parameter to specify the French recognition model
+paddleocr ocr -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_french01.png \
+    --lang fr \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False \
+    --use_textline_orientation False \
+    --save_path ./output \
+    --device gpu:0 
+```
+For explanations of the other command-line parameters, please refer to the [Command Line Usage](../../pipeline_usage/OCR.en.md#21-command-line) section of the general OCR pipeline documentation. After running, the results will be displayed in the terminal:
+
+```bash
+{'res': {'input_path': '/root/.paddlex/predict_input/general_ocr_french01.png', 'page_index': None, 'model_settings': {'use_doc_preprocessor': True, 'use_textline_orientation': False}, 'doc_preprocessor_res': {'input_path': None, 'page_index': None, 'model_settings': {'use_doc_orientation_classify': False, 'use_doc_unwarping': False}, 'angle': -1}, 'dt_polys': array([[[119,  23],
+        ...,
+        [118,  75]],
+
+       ...,
+
+       [[109, 506],
+        ...,
+        [108, 556]]], dtype=int16), 'text_det_params': {'limit_side_len': 64, 'limit_type': 'min', 'thresh': 0.3, 'max_side_limit': 4000, 'box_thresh': 0.6, 'unclip_ratio': 1.5}, 'text_type': 'general', 'textline_orientation_angles': array([-1, ..., -1]), 'text_rec_score_thresh': 0.0, 'rec_texts': ['mifere; la profpérité & les fuccès ac-', 'compagnent l’homme induftrieux.', 'Quel eft celui qui a acquis des ri-', 'cheffes, qui eft devenu puiffant, qui', 's’eft couvert de gloire, dont l’éloge', 'retentit par-tout, qui fiege au confeil', "du Roi? C'eft celui qui bannit la pa-", "reffe de fa maifon, & qui a dit à l'oifi-", 'veté : tu es mon ennemie.'], 'rec_scores': array([0.98409832, ..., 0.98091048]), 'rec_polys': array([[[119,  23],
+        ...,
+        [118,  75]],
+
+       ...,
+
+       [[109, 506],
+        ...,
+        [108, 556]]], dtype=int16), 'rec_boxes': array([[118, ...,  81],
+       ...,
+       [108, ..., 562]], dtype=int16)}}
+```
+
+If you specify `save_path`, the visualization results will be saved to the specified path. An example of the visualized result is shown below:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/general_ocr_french01_res.png"/>
+
+
+You can also use Python code to specify the recognition model for a particular language when initializing the general OCR pipeline via the `lang` parameter:
+
+```python
+from paddleocr import PaddleOCR
+
+ocr = PaddleOCR(
+    lang="fr", # Specify French recognition model with the lang parameter
+    use_doc_orientation_classify=False, # Disable document orientation classification model
+    use_doc_unwarping=False, # Disable text image unwarping model
+    use_textline_orientation=False, # Disable text line orientation classification model
+)
+result = ocr.predict("https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_french01.png")
+for res in result:
+    res.print()
+    res.save_to_img("output")
+    res.save_to_json("output")
+```
+For more details on the `PaddleOCR` class parameters, please refer to the [Python Scripting Integration](../../pipeline_usage/OCR.en.md#22-python-script-integration) section of the general OCR pipeline documentation.
+
+
+## 3. Performance Comparison
+
+| Model | Download Link | Accuracy on the corresponding dataset (%) | Improvement over the previous generation model (%) ｜
+|-|-|-|-|
+| korean_PP-OCRv5_mobile_rec |<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/korean_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a> | 88.0|  65.0  | 
+| latin_PP-OCRv5_mobile_rec | <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/latin_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a> | 84.7 | 46.8 |
+| eslav_PP-OCRv5_mobile_rec |<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/eslav_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/eslav_PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a> | 81.6 | 31.4 | 
+| th_PP-OCRv5_mobile_rec |<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/th_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/th_PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a> | 82.68 | - |
+| el_PP-OCRv5_mobile_rec |<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/el_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/el_PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a> | 89.28 | - |
+| en_PP-OCRv5_mobile_rec |<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a> | 85.25 | 11.0 | 
+| cyrillic_PP-OCRv5_mobile_rec | <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/cyrillic_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv5_mobile_rec_pretrained.pdparams">Trained Model</a> | 80.27 | 21.2 |
+| arabic_PP-OCRv5_mobile_rec | <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/arabic_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv5_mobile_rec_pretrained.pdparams">Trained Model</a> | 81.27 | 22.83 |
+| devanagari_PP-OCRv5_mobile_rec | <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/devanagari_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv5_mobile_rec_pretrained.pdparams">Trained Model</a> | 84.96 | 68.26 |
+| te_PP-OCRv5_mobile_rec | <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/te_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv5_mobile_rec_pretrained.pdparams">Trained Model</a> | 87.65 | 43.47 |
+| ta_PP-OCRv5_mobile_rec | <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ta_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv5_mobile_rec_pretrained.pdparams">Trained Model</a> | 94.2 | 39.23 |
+
+ **Notes:**
+ - Korean Dataset: The latest PP-OCRv5 dataset containing 5,007 Korean text images.
+ - Latin Script Language Dataset: The latest PP-OCRv5 dataset containing 3,111 images of Latin script languages.
+ - East Slavic Language Dataset: The latest PP-OCRv5 dataset containing a total of 7,031 text images in Russian, Belarusian, and Ukrainian.
+ - Thai dataset: The latest PP-OCRv5 constructed Thai dataset contains a total of 4,261 text images for recognition.
+ - Greek dataset: The latest PP-OCRv5 constructed Greek dataset contains a total of 2,799 text images for recognition.
+ - English dataset: The latest PP-OCRv5 constructed English dataset contains a total of 6,530 text images for recognition.
+ - Cyrillic Dataset: The latest PP-OCRv5 Cyrillic recognition dataset contains a total of 7,600 text images.
+ - Tamil Dataset: The latest PP-OCRv5 Tamil recognition dataset contains a total of 2,121 text images.
+ - Telugu Dataset: The latest PP-OCRv5 Telugu recognition dataset contains a total of 2,478 text images.
+ - Arabic Dataset: The latest PP-OCRv5 Arabic, Sanskrit, etc. recognition dataset contains a total of 2,676 text images.
+ - Devanagari Dataset: The latest PP-OCRv5 Devanagari recognition dataset contains a total of 3,611 text images.
+
+## 4. Supported Languages and Abbreviations
+
+| Language | Description | Abbreviation | | Language | Description | Abbreviation |
+| --- | --- | --- | ---|--- | --- | --- |
+| Chinese | Chinese & English | ch | | Hungarian | Hungarian | hu |
+| English | English | en | | Serbian (latin) | Serbian (latin) | rs_latin |
+| French | French | fr | | Indonesian | Indonesian | id |
+| German | German | de | | Occitan | Occitan | oc |
+| Japanese | Japanese | japan | | Icelandic | Icelandic | is |
+| Korean | Korean | korean | | Lithuanian | Lithuanian | lt |
+| Traditional Chinese | Chinese Traditional | chinese_cht | | Maori | Maori | mi |
+| Afrikaans | Afrikaans | af | | Malay | Malay | ms |
+| Italian | Italian | it | | Dutch | Dutch | nl |
+| Spanish | Spanish | es | | Norwegian | Norwegian | no |
+| Bosnian | Bosnian | bs | | Polish | Polish | pl |
+| Portuguese | Portuguese | pt | | Slovak | Slovak | sk |
+| Czech | Czech | cs | | Slovenian | Slovenian | sl |
+| Welsh | Welsh | cy | | Albanian | Albanian | sq |
+| Danish | Danish | da | | Swedish | Swedish | sv |
+| Estonian | Estonian | et | | Swahili | Swahili | sw |
+| Irish | Irish | ga | | Tagalog | Tagalog | tl |
+| Croatian | Croatian | hr | | Turkish | Turkish | tr |
+| Uzbek | Uzbek | uz | | Latin | Latin | la |
+| Russian | Russian | ru | | Belarusian | Belarusian | be |
+| Ukrainian | Ukrainian | uk | | Thai | Thai | th | 
+| Greek | Greek | el | | Azerbaijani | Azerbaijani | az |
+| Kurdish | Kurdish | ku | | Latvian | Latvian | lv |
+| Maltese | Maltese | mt | | Pali | Pali | pi |
+| Romanian | Romanian | ro | | Vietnamese | Vietnamese | vi |
+| Finnish | Finnish | fi | | Basque | Basque | eu | 
+| Galician | Galician | gl | | Luxembourgish | Luxembourgish | lb |
+| Romansh | Romansh | rm | | Catalan | Catalan | ca |
+| Quechua | Quechua | qu | | Telugu | Telugu | te |
+| Serbian (Cyrillic) | Serbian (Cyrillic) | rs_cyrillic | | Bulgarian | Bulgarian | bg |
+| Mongolian | Mongolian | mn | | Abkhaz | Abkhaz | ab |
+| Adyghe | Adyghe | ady | | Kabardian | Kabardian | kbd |
+| Avar | Avar | av | | Dargwa | Dargwa | dar |
+| Ingush | Ingush | inh | | Chechen | Chechen | ce |
+| Lak | Lak | lki | | Lezgian | Lezgian | lez |
+| Tabasaran | Tabasaran | tab | | Kazakh | Kazakh | kk |
+| Kyrgyz | Kyrgyz | ky | | Tajik | Tajik | tg |
+| Macedonian | Macedonian | mk | | Tatar | Tatar | tt |
+| Chuvash | Chuvash | cv | | Bashkir | Bashkir | ba |
+| Mari | Mari | mhr | | Moldovan | Moldovan | mo |
+| Udmurt | Udmurt | udm | | Komi | Komi | kv |
+| Ossetian | Ossetian | os | | Buriat | Buriat | bua |
+| Kalmyk | Kalmyk | xal | | Tuvinian | Tuvinian | tyv |
+| Sakha | Sakha | sah | | Karakalpak | Karakalpak | kaa |
+| Arabic | Arabic | ar | | Persian | Persian | fa |
+| Uyghur | Uyghur | ug | | Urdu | Urdu | ur |
+| Pashto | Pashto | ps | | Kurdish | Kurdish | ku |
+| Sindhi | Sindhi | sd | | Balochi | Balochi | bal |
+| Hindi | Hindi | hi | | Marathi | Marathi | mr |
+| Nepali | Nepali | ne | | Bihari | Bihari | bh |
+| Maithili | Maithili | mai | | Old English | Old English | ang |
+| Bhojpuri | Bhojpuri | bho | | Magahi | Magahi | mah |
+| Sadri | Sadri | sck | | Newar | Newar | new |
+| Konkani | Konkani | gom | | Sanskrit | Sanskrit | sa |
+| Haryanvi | Haryanvi | bgc | | Tamil | Tamil | ta |
+
+
+## 5. Models and Their Supported Languages
+
+| Model | Supported Languages |
+|-|-|
+| korean_PP-OCRv5_mobile_rec | Korean, English |
+| latin_PP-OCRv5_mobile_rec | French, German, Afrikaans, Italian, Spanish, Bosnian, Portuguese, Czech, Welsh, Danish, Estonian, Irish, Croatian, Uzbek, Hungarian, Serbian (Latin), Indonesian, Occitan, Icelandic, Lithuanian, Maori, Malay, Dutch, Norwegian, Polish, Slovak, Slovenian, Albanian, Swedish, Swahili, Tagalog, Turkish, Latin, Azerbaijani, Kurdish, Latvian, Maltese, Pali, Romanian, Vietnamese, Finnish, Basque, Galician, Luxembourgish, Romansh, Catalan, Quechua |
+| eslav_PP-OCRv5_mobile_rec | Russian, Belarusian, Ukrainian, English |
+| th_PP-OCRv5_mobile_rec | Thai, English |
+| el_PP-OCRv5_mobile_rec | Greek, English |
+| en_PP-OCRv5_mobile_rec | English |
+| cyrillic_PP-OCRv5_mobile_rec | Russian, Belarusian, Ukrainian, Serbian (Cyrillic), Bulgarian, Mongolian, Abkhazian, Adyghe, Kabardian, Avar, Dargin, Ingush, Chechen, Lak, Lezgin, Tabasaran, Kazakh, Kyrgyz, Tajik, Macedonian, Tatar, Chuvash, Bashkir, Malian, Moldovan, Udmurt, Komi, Ossetian, Buryat, Kalmyk, Tuvan, Sakha, Karakalpak, English |
+| arabic_PP-OCRv5_mobile_rec | Arabic, Persian, Uyghur, Urdu, Pashto, Kurdish, Sindhi, Balochi, English |
+| devanagari_PP-OCRv5_mobile_rec | Hindi, Marathi, Nepali, Bihari, Maithili, Angika, Bhojpuri, Magahi, Santali, Newari, Konkani, Sanskrit, Haryanvi, English |
+| ta_PP-OCRv5_mobile_rec | Tamil, English |
+| te_PP-OCRv5_mobile_rec | Telugu, English |
+
+**note:** `en_PP-OCRv5_mobile_rec` is an optimized version based on the `PP-OCRv5` model, specifically fine-tuned for English scenarios. It demonstrates higher recognition accuracy and better adaptability when processing English text.
diff --git a/docs/version3.x/algorithm/PP-OCRv5/PP-OCRv5_multi_languages.md b/docs/version3.x/algorithm/PP-OCRv5/PP-OCRv5_multi_languages.md
new file mode 100644
index 0000000..4c52108
--- /dev/null
+++ b/docs/version3.x/algorithm/PP-OCRv5/PP-OCRv5_multi_languages.md
@@ -0,0 +1,247 @@
+---
+comments: true
+---
+
+# 一、PP-OCRv5多语种文字识别介绍
+
+
+[PP-OCRv5](./PP-OCRv5.md) 是 PP-OCR 系列的最新一代文字识别解决方案，专注于多场景、多语种的文字识别任务。在文字类型支持方面，默认配置的识别模型可准确识别简体中文、中文拼音、繁体中文、英文和日文这五大主流文字类型。同时，PP-OCRv5还提供了覆盖106种语言的多语种文字识别能力，包括韩文、西班牙文、法文、葡萄牙文、德文、意大利文、俄罗斯文、泰文、希腊文等（具体支持语种及缩写详见[第四节](#_3)）。相较于前代 PP-OCRv3 版本，PP-OCRv5 在多语言文字识别准确率上实现了超过30%的提升。
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/french_0_res.jpg" alt="法文识别结" width="500"/>
+  <br>
+  <b>法文识别结果</b>
+</div>
+
+<br>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/german_0_res.png" alt="德文识别结果" width="500"/>
+  <br>
+  <b>德文识别结果</b>
+</div>
+
+<br>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/korean_1_res.jpg" alt="韩文识别结果" width="500"/>
+  <br>
+  <b>韩文识别结果</b>
+</div>
+
+<br>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/ru_0.jpeg" alt="俄文识别结果" width="500"/>
+  <br>
+  <b>俄文识别结果</b>
+</div>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/th_0_res.jpg" alt="泰文识别结果" width="500"/>
+  <br>
+  <b>泰文识别结果</b>
+</div>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/el_0_res.jpg" alt="希腊文识别结果" width="500"/>
+  <br>
+  <b>希腊文识别结果</b>
+</div>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/ar_0_res.jpg" alt="阿拉伯文文识别结果" width="500"/>
+  <br>
+  <b>阿拉伯文识别结果</b>
+</div>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/hi_0_res.jpg" alt="印地文识别结果" width="500"/>
+  <br>
+  <b>印地文识别结果</b>
+</div>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/ta_0_rec.jpg" alt="泰米尔文识别结果" width="500"/>
+  <br>
+  <b>泰米尔文识别结果</b>
+</div>
+
+<div align="center">
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/te_0_res.png" alt="泰卢固文识别结果" width="500"/>
+  <br>
+  <b>泰卢固文识别结果</b>
+</div>
+
+## 二、快速使用
+
+您可以通过在命令行中使用 `--lang` 参数，来使用指定语种的文本识别模型进行通用 OCR 产线的推理：
+
+```bash
+# 通过 `--lang` 参数指定使用法语的识别模型
+paddleocr ocr -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_french01.png \
+    --lang fr \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False \
+    --use_textline_orientation False \
+    --save_path ./output \
+    --device gpu:0 
+```
+上述命令行的其他参数说明请参考通用 OCR 产线的[命令行使用方式](../../pipeline_usage/OCR.md#21), 运行后结果会被打印到终端上：
+
+```bash
+{'res': {'input_path': '/root/.paddlex/predict_input/general_ocr_french01.png', 'page_index': None, 'model_settings': {'use_doc_preprocessor': True, 'use_textline_orientation': False}, 'doc_preprocessor_res': {'input_path': None, 'page_index': None, 'model_settings': {'use_doc_orientation_classify': False, 'use_doc_unwarping': False}, 'angle': -1}, 'dt_polys': array([[[119,  23],
+        ...,
+        [118,  75]],
+
+       ...,
+
+       [[109, 506],
+        ...,
+        [108, 556]]], dtype=int16), 'text_det_params': {'limit_side_len': 64, 'limit_type': 'min', 'thresh': 0.3, 'max_side_limit': 4000, 'box_thresh': 0.6, 'unclip_ratio': 1.5}, 'text_type': 'general', 'textline_orientation_angles': array([-1, ..., -1]), 'text_rec_score_thresh': 0.0, 'rec_texts': ['mifere; la profpérité & les fuccès ac-', 'compagnent l’homme induftrieux.', 'Quel eft celui qui a acquis des ri-', 'cheffes, qui eft devenu puiffant, qui', 's’eft couvert de gloire, dont l’éloge', 'retentit par-tout, qui fiege au confeil', "du Roi? C'eft celui qui bannit la pa-", "reffe de fa maifon, & qui a dit à l'oifi-", 'veté : tu es mon ennemie.'], 'rec_scores': array([0.98409832, ..., 0.98091048]), 'rec_polys': array([[[119,  23],
+        ...,
+        [118,  75]],
+
+       ...,
+
+       [[109, 506],
+        ...,
+        [108, 556]]], dtype=int16), 'rec_boxes': array([[118, ...,  81],
+       ...,
+       [108, ..., 562]], dtype=int16)}}
+```
+
+若指定了`save_path`，则会保存可视化结果在`save_path`下。可视化结果如下：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/ocr/general_ocr_french01_res.png"/>
+
+
+您也可以使用 Python 代码，在通用 OCR 产线初始化时，通过 `lang` 参数来使用指定语种的识别模型：
+
+```python
+from paddleocr import PaddleOCR
+
+ocr = PaddleOCR(
+    lang="fr" # 通过 lang 参数指定使用法语的识别模型
+    use_doc_orientation_classify=False, # 通过 use_doc_orientation_classify 参数指定不使用文档方向分类模型
+    use_doc_unwarping=False, # 通过 use_doc_unwarping 参数指定不使用文本图像矫正模型
+    use_textline_orientation=False, # 通过 use_textline_orientation 参数指定不使用文本行方向分类模型
+)
+result = ocr.predict("https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_french01.png")
+for res in result:
+    res.print()
+    res.save_to_img("output")
+    res.save_to_json("output")
+```
+更过关于 `PaddleOCR` 类参数的说明参考通用 OCR 产线的[脚本方式集成](../../pipeline_usage/OCR.md#22-python)。
+
+
+## 三、指标对比
+
+| 模型 | 模型下载链接 | 对应数据集精度（%） | 相比前代模型提升幅度 (%) |
+|-|-|-|-|
+| korean_PP-OCRv5_mobile_rec |<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/korean_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a> | 88.0| 65.0 |
+| latin_PP-OCRv5_mobile_rec | <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/latin_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a> | 84.7 | 46.8 |
+| eslav_PP-OCRv5_mobile_rec |<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/eslav_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/eslav_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a> | 81.6 | 31.4 | 
+| th_PP-OCRv5_mobile_rec |<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/th_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/th_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a> | 82.68 | - |
+| el_PP-OCRv5_mobile_rec |<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/el_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/el_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a> | 89.28 | - | 
+| en_PP-OCRv5_mobile_rec |<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a> | 85.25 | 11.0 | 
+| cyrillic_PP-OCRv5_mobile_rec |<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/cyrillic_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a> | 80.27 | 21.2 | 
+| arabic_PP-OCRv5_mobile_rec |<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/arabic_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a> | 81.27 | 22.83 |
+| devanagari_PP-OCRv5_mobile_rec |<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/devanagari_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a> | 84.96 | 68.26 |
+| te_PP-OCRv5_mobile_rec |<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/te_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a> | 87.65 | 43.47 |
+| ta_PP-OCRv5_mobile_rec |<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ta_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a> | 94.2 | 39.23 |
+
+ **注：**
+ - 韩语数据集：PP-OCRv5 最新构建的包含了 5007 张韩语文本图片的识别数据集。
+ - 拉丁字母语言数据集：PP-OCRv5 最新构建的包含了 3111 张拉丁字母语言的文本图片识别数据集。
+ - 东斯拉夫语言数据集：PP-OCRv5 最新构建的包含了俄语、 白俄罗斯语和乌克兰语共计 7031 张文本图片的识别数据集。
+ - 泰文数据集：PP-OCRv5 最新构建的泰文共计 4261 张文本图片的识别数据集。
+ - 希腊文数据集：PP-OCRv5 最新构建的希腊文共计 2799 张文本图片的识别数据集。
+ - 英文数据集：PP-OCRv5 最新构建的英文共计 6530 张文本图片的识别数据集。
+ - 西里尔语数据集：PP-OCRv5 最新构建的西里尔文共计 7600 张文本图片的识别数据集。
+ - 泰米尔语数据集：PP-OCRv5 最新构建的泰米尔语共计 2121 张文本图片的识别数据集。
+ - 泰卢固语数据集：PP-OCRv5 最新构建的泰卢固语共计 2478 张文本图片的识别数据集。
+ - 阿拉伯语数据集：PP-OCRv5 最新构建的阿拉伯、梵语等共计 2676 张文本图片的识别数据集。
+ - 天城文数据集：PP-OCRv5 最新构建的天城文共计 3611 张文本图片的识别数据集。
+
+## 四、 支持语种及缩写
+
+| 语种 | 描述 | 缩写 | | 语种 | 描述 | 缩写 |
+| --- | --- | --- | ---|--- | --- | --- |
+| 中文 | Chinese & English | ch | | 匈牙利文 | Hungarian | hu |
+| 英文 | English | en | | 塞尔维亚文（latin） | Serbian(latin) | rs_latin |
+| 法文 | French | fr | | 印度尼西亚文 | Indonesian | id |
+| 德文 | German | de | | 欧西坦文 | Occitan | oc |
+| 日文 | Japanese | japan | | 冰岛文 | Icelandic | is |
+| 韩文 | Korean | korean | | 立陶宛文 | Lithuanian | lt |
+| 中文繁体 | Chinese Traditional | chinese_cht | | 毛利文 | Maori | mi |
+| 南非荷兰文 | Afrikaans | af | | 马来文 | Malay | ms |
+| 意大利文 | Italian | it | | 荷兰文 | Dutch | nl |
+| 西班牙文 | Spanish | es | | 挪威文 | Norwegian | no |
+| 波斯尼亚文 | Bosnian | bs | | 波兰文 | Polish | pl |
+| 葡萄牙文 | Portuguese | pt | | 斯洛伐克文 | Slovak | sk |
+| 捷克文 | Czech | cs | | 斯洛文尼亚文 | Slovenian | sl |
+| 威尔士文 | Welsh | cy | | 阿尔巴尼亚文 | Albanian | sq |
+| 丹麦文 | Danish | da | | 瑞典文 | Swedish | sv |
+| 爱沙尼亚文 | Estonian | et | | 西瓦希里文 | Swahili | sw |
+| 爱尔兰文 | Irish | ga | | 塔加洛文 | Tagalog | tl |
+| 克罗地亚文 | Croatian | hr | | 土耳其文 | Turkish | tr |
+| 乌兹别克文 | Uzbek | uz | | 拉丁文 | Latin | la |
+| 俄罗斯文 | Russian | ru | | 白俄罗斯文 | Belarusian | be |
+| 乌克兰文 | Ukranian | uk |  | 泰文 | Thai | th | 
+| 希腊文 | Greek | el | | 阿塞拜疆文 | Azerbaijani | az |
+| 库尔德文 | Kurdish | ku | |拉脱维亚文 | Latvian|lv |
+| 马耳他文 | Maltese | mt | |巴利文 | Pali| pi |
+|罗马尼亚文 | Romanian | ro | |越南文 | Vietnamese| vi |
+| 芬兰文 | Finnish | fi | | 巴斯克文 | Basque | eu | 
+| 加利西亚文| Galician | gl | | 卢森堡文 | Luxembourgish | lb |
+| 罗曼什文 | Romansh | rm | | 加泰罗尼亚文 | Catalan | ca |
+| 克丘亚文 | Quechua |qu | | 泰卢固文 |Telugu |te |
+| 塞尔维亚语（西里尔字母） | Serbian (Cyrillic) | rs_cyrillic | | 保加利亚文 | Bulgarian | bg |
+| 蒙古文 | Mongolian | mn | | 阿布哈兹文 | Abkhaz | ab |
+| 阿迪赫文 | Adyghe | ady | | 卡巴尔达文 | Kabardian | kbd |
+| 阿瓦尔文 | Avar | av | | 达尔格瓦文 | Dargwa | dar |
+| 印古什文 | Ingush | inh | | 车臣文 | Chechen | ce |
+| 拉克文 | Lak | lki | | 列兹金文 | Lezgian | lez |
+| 塔巴萨兰文 | Tabasaran | tab | | 哈萨克文 | Kazakh | kk |
+| 吉尔吉斯文 | Kyrgyz | ky | | 塔吉克文 | Tajik | tg |
+| 马其顿文 | Macedonian | mk | | 鞑靼文 | Tatar | tt |
+| 楚瓦什文 | Chuvash | cv | | 巴什基尔文 | Bashkir | ba |
+| 马里文 | Mari | mhr | | 莫尔多瓦文 | Moldovan | mo |
+| 乌德穆尔特文 | Udmurt | udm | | 科米文 | Komi | kv |
+| 奥塞梯文 | Ossetian | os | | 布里亚特文 | Buriat | bua |
+| 卡尔梅克文 | Kalmyk | xal | | 图瓦文 | Tuvinian | tyv |
+| 萨哈文 | Sakha | sah | | 卡拉卡尔帕克语 | Karakalpak | kaa |
+| 阿拉伯文 | Arabic | ar | | 波斯文 | Persian | fa |
+| 维吾尔文 | Uyghur | ug | | 乌尔都文 | Urdu | ur |
+| 普什图文 | Pashto | ps | | 库尔德文 | Kurdish | ku |
+| 信德文 | Sindhi | sd | | 俾路支文 | Balochi | bal |
+| 印地文 | Hindi | hi | | 马拉地文 | Marathi | mr |
+| 尼泊尔文 | Nepali | ne | | 比哈尔文 | Bihari | bh |
+| 迈蒂利文 | Maithili | mai | | 古英文 | Old English | ang |
+| 博杰普尔文 | Bhojpuri | bho | | 马加希文 | Magahi | mah |
+| 萨达里文 | Sadri | sck | | 尼瓦尔文 | Newar | new |
+| 孔卡尼文 | Konkani | gom | | 梵文 | Sanskrit | sa |
+| 哈里亚纳文 | Haryanvi | bgc | | 泰米尔语 | Tamil | ta |
+
+## 五、模型及其支持的语种
+
+| 模型 | 支持语种 |
+|-|-|
+| PP-OCRv5_server_rec | 简体中文、繁体中文、英文、日文  |
+| PP-OCRv5_mobile_rec | 简体中文、繁体中文、英文、日文 |
+| korean_PP-OCRv5_mobile_rec | 韩文、英文 |
+| latin_PP-OCRv5_mobile_rec |法文、德文、南非荷兰文、意大利文、西班牙文、波斯尼亚文、葡萄牙文、捷克文、威尔士文、丹麦文、爱沙尼亚文、爱尔兰文、克罗地亚文、乌兹别克文、匈牙利文、塞尔维亚文（latin）、印度尼西亚文、欧西坦文、冰岛文、立陶宛文、毛利文、马来文、荷兰文、挪威文、波兰文、斯洛伐克文、斯洛文尼亚文、阿尔巴尼亚文、瑞典文、西瓦希里文、塔加洛文、土耳其文、拉丁文、阿塞拜疆文、库尔德文、拉脱维亚文、马耳他文、巴利文、罗马尼亚文、越南文、芬兰文、巴斯克文、加利西亚文、卢森堡文、罗曼什文、加泰罗尼亚文、克丘亚文|
+| eslav_PP-OCRv5_mobile_rec | 俄罗斯文、白俄罗斯文、乌克兰文、英文 |
+| th_PP-OCRv5_mobile_rec | 泰文、英文 |
+| el_PP-OCRv5_mobile_rec | 希腊文、英文 |
+| en_PP-OCRv5_mobile_rec | 英文 |
+| cyrillic_PP-OCRv5_mobile_rec | 俄罗斯文、白俄罗斯文、乌克兰文、塞尔维亚文（cyrillic）、保加利亚文、蒙古文、阿布哈兹文、阿迪赫文、卡巴尔达文、阿瓦尔文、达尔格瓦文、印古什文、车臣文、拉克文、列兹金文、塔巴萨兰文、哈萨克文、吉尔吉斯文、塔吉克文、马其顿文、鞑靼文、楚瓦什文、巴什基尔文、马里文、莫尔多瓦文、乌德穆尔特文、科米文、奥塞梯文、布里亚特文、卡尔梅克文、图瓦文、萨哈文、卡拉卡尔帕克文、英文 | 
+| arabic_PP-OCRv5_mobile_rec | 阿拉伯文、波斯文、维吾尔文、乌尔都文、普什图文、库尔德文、信德文、俾路支文、英文|
+| devanagari_PP-OCRv5_mobile_rec | 印地文，马拉地文，尼泊尔文，比哈尔文，迈蒂利文，古英文，博杰普尔文，马加希文，萨达里文，尼瓦尔文，孔卡尼文，梵文，哈里亚纳文、英文 |
+| ta_PP-OCRv5_mobile_rec | 泰米尔文、英文 |
+| te_PP-OCRv5_mobile_rec| 泰卢固文、英文 |
+
+
+**注：** `en_PP-OCRv5_mobile_rec` 是在 `PP-OCRv5` 模型基础上，针对英文场景进行了定向优化，在处理英文文本时表现出更高的识别精度和更强的场景适应能力。
diff --git a/docs/version3.x/algorithm/PP-OCRv6/PP-OCRv6.en.md b/docs/version3.x/algorithm/PP-OCRv6/PP-OCRv6.en.md
new file mode 100644
index 0000000..7745e0a
--- /dev/null
+++ b/docs/version3.x/algorithm/PP-OCRv6/PP-OCRv6.en.md
@@ -0,0 +1,226 @@
+# 1. PP-OCRv6 Introduction
+
+**PP-OCRv6** is the latest generation of the PP-OCR universal text recognition solution. Built on the newly designed PPLCNetV4 unified backbone, it offers tiny, small, and medium tiers targeting edge/IoT, mobile/desktop, and server scenarios respectively. PP-OCRv6 achieves a major breakthrough in language coverage—the medium/small tiers support 50 languages with a single unified model, including Simplified Chinese, Traditional Chinese, English, Japanese, and 46 Latin-script languages (tiny supports 49, excluding Japanese). On our in-house multi-scenario benchmark, PP-OCRv6_medium achieves +5.1% recognition accuracy and +4.6% detection Hmean over PP-OCRv5_server, with 2.37× GPU inference speedup; with only 34.5M parameters, it surpasses VLMs such as Qwen3-VL-235B and GPT-5.5 in accuracy.
+
+Main contributions:
+
+1. **Unified and Scalable Model Family**: A three-tier OCR model family spanning 1.5M to 34.5M parameters. The medium tier achieves 86.2% detection Hmean and 83.2% recognition accuracy, serving as production-ready infrastructure for industrial deployment and large-scale data pipelines.
+2. **Tailored Lightweight Architectural Innovations**: (i) LCNetV4: a MetaFormer-style lightweight backbone with structural reparameterization; (ii) RepLKFPN: a detection neck with dilated reparameterizable depthwise convolutions for large receptive fields; (iii) EncoderWithLightSVTR: a recognition neck with local-global attention and additive skip connections.
+3. **Extensive Multi-Language and Scenario Generalization**: A single model scaled to support 50 languages and diverse challenging industrial scenes (e.g., digital displays, dot-matrix characters, tire prints), significantly improving OCR performance in scenarios traditionally underserved by general-purpose VLMs.
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/v6acc_opt.png" width="800"/>
+</div>
+
+<p align="center">Performance comparison between PP-OCRv6, PP-OCRv5, and Vision-Language Models. Left: text detection average Hmean (%); Right: text recognition weighted average accuracy (%).</p>
+
+
+# 2. Key Technical Improvements
+
+## 2.1 Unified Backbone: PPLCNetV4
+
+**LCNetV4Block**: Following the MetaFormer paradigm, each layer is decomposed into a Token Mixer and a Channel Mixer. Given input feature $\mathbf{x} \in \mathbb{R}^{C \times H \times W}$:
+
+$$\hat{\mathbf{x}} = \text{SE}(\text{DW}(\mathbf{x})) + \mathbf{x}$$
+
+$$\mathbf{y} = W_2\,\sigma(W_1\,\hat{\mathbf{x}}) + \hat{\mathbf{x}}$$
+
+where $\text{DW}(\cdot)$ is a 3×3 depthwise convolution (Token Mixer), SE is an optional channel attention module, $W_1 \in \mathbb{R}^{2C \times C}$ and $W_2 \in \mathbb{R}^{C \times 2C}$ form the Channel Mixer with expansion ratio 2, and $\sigma$ is GELU activation.
+
+**Task-Adaptive Downsampling**: The same backbone serves both tasks via different stride strategies—detection mode uses standard stride-2 spatial downsampling producing multi-scale feature maps (stride 4/8/16/32); recognition mode uses asymmetric stride $(2,1)$ at Stage 3/4, reducing height only while preserving width, followed by height-axis average pooling to produce 1-D sequential features for CTC/NRTR decoding.
+
+**Comparison with LCNetV3**:
+
+| Design Aspect | LCNetV3 | LCNetV4 |
+|--------------|---------|---------|
+| Architecture | MobileNet-style (DW→SE→PW) | MetaFormer (TokenMixer + ChannelMixer) |
+| Channel Interaction | Single 1×1 PW Conv | Expand(2×)→Act→Compress + residual |
+| Spatial Mixing | Plain DW Conv | RepDWConv (3×3 + 1×1 + identity) |
+| BN Initialization | Standard | Zero-init on compress BN |
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/backbone.png" width="600"/>
+</div>
+
+<p align="center">PPLCNetV4 backbone architecture.</p>
+
+## 2.2 Detection Module
+
+- **RepLKFPN**: Lightweight large-kernel FPN using DilatedReparamBlock (7×7 depthwise conv + dilated branches), 31% fewer parameters than PP-OCRv5's RSEFPN (118K vs 172K) with receptive field expanded from 3×3 to 7×7.
+- **Auxiliary Deep Supervision**: Prediction heads at P2, P3, P4 levels for stronger gradient signals during training.
+- **DiceBCE Loss**: Combined DiceLoss + Focal Loss for better per-pixel supervision on small and dense text.
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/ppocrv6_det_pip_ori.png" width="600"/>
+</div>
+
+<p align="center">PP-OCRv6 detection module architecture.</p>
+
+## 2.3 Recognition Module
+
+- **EncoderWithLightSVTR Neck**: Local context modeling (1×7 depthwise conv) + global self-attention (1-2 Transformer layers), with additive skip connections (instead of concatenation in PP-OCRv5) to reduce parameters.
+- **Multi-Head Decoder**: CTCHead for efficient parallel inference; NRTRHead for auxiliary training supervision (removed at inference).
+- **Tiny Model Design**: No neck (direct reshape + FC), trained with knowledge distillation from the medium model.
+- **Multilingual Unification**: Dictionary extended with ~200 diacritical characters, enabling single-model 50-language coverage.
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/rec.png" width="600"/>
+</div>
+
+<p align="center">PP-OCRv6 recognition module architecture.</p>
+
+# 3. Key Metrics
+
+## 3.1 Text Detection
+
+Text detection Hmean (%) on our in-house multi-scenario benchmark (16 categories):
+
+| Model | AVG | HW-CN | HW-EN | Print-CN | Print-EN | TC | Anc. | JP | Blur | Emo. | Warp | Pin. | Art. | Tab. | Rot. | Indus. | Gen. |
+|-------|-----|-------|-------|----------|----------|-----|------|-----|------|------|------|------|------|------|------|--------|------|
+| **PP-OCRv6_medium** | **86.2** | **83.7** | 84.0 | **95.1** | **93.7** | **86.3** | **80.2** | **84.3** | **94.1** | 99.6 | **88.6** | **74.0** | **69.0** | 96.8 | **93.8** | **73.3** | **82.8** |
+| **PP-OCRv6_small** | **84.1** | 80.5 | **87.1** | 94.2 | 93.6 | 85.7 | 72.6 | 82.3 | 92.6 | 99.7 | 87.6 | 69.6 | 65.3 | 95.6 | 93.7 | 67.6 | 78.2 |
+| **PP-OCRv6_tiny** | **80.6** | 79.4 | 85.9 | 93.1 | 92.3 | 83.7 | 63.0 | 76.6 | 89.3 | **99.8** | 86.1 | 59.0 | 60.1 | 94.7 | 91.0 | 62.0 | 73.8 |
+| PP-OCRv5_server | 81.6 | 80.3 | 84.1 | 94.5 | 91.7 | 81.5 | 67.6 | 77.2 | 90.1 | 96.2 | 87.6 | 67.1 | 67.3 | **97.1** | 80.0 | 64.3 | 79.7 |
+| PP-OCRv5_mobile | 75.2 | 74.4 | 77.7 | 90.5 | 91.0 | 82.3 | 58.1 | 72.7 | 87.4 | 93.6 | 82.7 | 57.5 | 52.5 | 92.8 | 64.7 | 52.8 | 72.1 |
+| Gemini-3.1-Pro | 46.8 | 53.4 | 56.5 | 47.3 | 47.6 | 39.0 | 45.8 | 38.2 | 50.0 | 68.1 | 44.6 | 40.6 | 65.2 | 26.9 | 22.1 | 52.5 | 50.2 |
+| GPT-5.5 | 45.6 | 42.4 | 58.5 | 50.2 | 51.9 | 35.0 | 26.7 | 42.0 | 49.1 | 97.5 | 37.7 | 36.3 | 52.0 | 71.0 | 10.0 | 36.2 | 32.6 |
+| Qwen3-VL-235B | 38.3 | 56.5 | 66.0 | 41.7 | 37.0 | 19.3 | 13.1 | 27.0 | 38.5 | 81.2 | 28.5 | 33.0 | 68.3 | 19.6 | 2.1 | 48.4 | 32.3 |
+
+## 3.2 Text Recognition
+
+Text recognition accuracy (%) on our in-house multi-scenario benchmark (15 categories):
+
+| Model | W-Avg | HW-CN | HW-EN | Print-CN | Print-EN | TC | Anc. | JP | Conf. | Spec. | Gen. | Pin. | Art. | Indus. | Screen | Card |
+|-------|-------|-------|-------|----------|----------|-----|------|-----|-------|-------|------|------|------|--------|--------|------|
+| **PP-OCRv6_medium** | **83.2** | **62.1** | 67.8 | **91.5** | **94.1** | **78.6** | **72.4** | **90.5** | **64.9** | **61.7** | **87.5** | **78.1** | **71.2** | **77.4** | **82.5** | **88.1** |
+| **PP-OCRv6_small** | **81.3** | 57.6 | 61.1 | 90.5 | 93.3 | 77.0 | 71.1 | 88.2 | 64.1 | 60.2 | 85.7 | 75.9 | 68.4 | 76.4 | 79.7 | 86.9 |
+| **PP-OCRv6_tiny** | 73.5 | 40.1 | 39.3 | 86.7 | 88.4 | 65.0 | 68.4 | 89.8 | 52.3 | 57.1 | 78.0 | 65.4 | 54.7 | 62.1 | 71.2 | 80.5 |
+| PP-OCRv5_server | 78.1 | 58.0 | 59.6 | 90.1 | 85.1 | 74.7 | 60.4 | 73.7 | 59.4 | 56.8 | 86.5 | 74.4 | 64.0 | 70.2 | 68.1 | 87.6 |
+| PP-OCRv5_mobile | 73.7 | 41.7 | 50.9 | 86.0 | 86.0 | 72.0 | 57.8 | 75.8 | 55.7 | 54.8 | 80.7 | 72.5 | 54.0 | 59.3 | 57.6 | 81.7 |
+| Qwen3-VL-235B | 74.9 | 49.7 | **73.2** | 82.3 | 86.2 | 76.4 | 33.6 | 66.2 | 56.1 | 49.0 | 82.5 | 76.5 | 69.6 | 74.7 | 73.8 | 78.7 |
+| Gemini-3.1-Pro | 71.4 | 46.4 | 73.0 | 80.0 | 90.5 | 69.5 | 18.0 | 67.2 | 54.4 | 50.3 | 74.6 | 75.9 | 63.1 | 69.1 | 73.2 | 75.9 |
+| GPT-5.5 | 64.2 | 19.2 | 56.9 | 75.7 | 82.2 | 57.5 | 63.7 | 58.6 | 49.1 | 48.3 | 67.7 | 50.4 | 53.0 | 62.4 | 67.7 | 71.1 |
+
+## 3.3 End-to-End Inference Speed (s/image)
+
+Tested on 200 images (general + document scenes), including image I/O, pre/post-processing, and model inference.
+
+| Hardware | Backend | PP-OCRv6_medium | PP-OCRv6_small | PP-OCRv6_tiny | PP-OCRv5_server | PP-OCRv5_mobile | PP-OCRv4_mobile |
+|----------|---------|-----------------|----------------|---------------|-----------------|-----------------|-----------------|
+| NVIDIA A100 | PaddlePaddle | 0.29 | 0.25 | 0.13 | 0.32 | 0.25 | 0.14 |
+| NVIDIA A100 | TensorRT | -- | 0.32 | 0.16 | -- | 0.33 | 0.16 |
+| NVIDIA V100 | PaddlePaddle | 0.72 | 0.49 | 0.21 | 0.66 | 0.50 | 0.25 |
+| NVIDIA V100 | ONNX Runtime | 0.67 | 0.53 | 0.29 | 0.77 | 0.46 | 0.27 |
+| NVIDIA V100 | TensorRT | 0.77 | 0.60 | 0.23 | 0.73 | 0.59 | 0.27 |
+| Intel Xeon 8350C | PaddlePaddle | 2.05 | 0.79 | 0.32 | 2.04 | 0.80 | 0.62 |
+| Intel Xeon 8350C | OpenVINO | 1.40 | 0.59 | 0.20 | 7.30 | 0.78 | 0.60 |
+| Intel Xeon 8350C | ONNX Runtime | 3.31 | 0.61 | 0.22 | 6.36 | 0.61 | 0.49 |
+| Apple M4 | PaddlePaddle | 8.82 | 3.07 | 0.96 | >10 | 5.82 | 5.65 |
+| Apple M4 | ONNX Runtime | 5.55 | 1.29 | 0.35 | 7.20 | 1.10 | 1.02 |
+
+- PP-OCRv6_medium matches or outperforms PP-OCRv5_server on all platforms: 1.1× faster on A100 (0.29s vs 0.32s), 1.15× on V100 ONNX Runtime (0.67s vs 0.77s), 5.2× on Intel Xeon OpenVINO (1.40s vs 7.30s).
+- PP-OCRv6_small matches PP-OCRv5_mobile in latency on most platforms with higher accuracy; 1.9× faster on Apple M4 PaddlePaddle (3.07s vs 5.82s).
+- PP-OCRv6_tiny is the fastest model across all platforms: 6.1× over PP-OCRv5_mobile on Apple M4 PaddlePaddle (0.96s vs 5.82s), 3.9× on Intel Xeon OpenVINO (0.20s vs 0.78s), reaching 0.13s on A100.
+
+# 4. Visualization
+
+## 4.1 Detection Comparison
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/ppocrv6_det_vis.jpg" width="800"/>
+</div>
+
+<p align="center">Text detection comparison. Left to right: PP-OCRv6_medium, PP-OCRv5_server, Gemini-3.1-Pro, GPT-5.5.</p>
+
+## 4.2 Hallucination Comparison
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/huanjue.jpg" width="800"/>
+</div>
+
+<p align="center">PP-OCRv6_medium vs VLMs hallucination comparison. PP-OCRv6 faithfully reproduces visual text content, while VLMs introduce hallucinated corrections based on linguistic priors.</p>
+
+## 4.3 End-to-End OCR Comparison
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/case1.jpg" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/case2.jpg" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/case3.jpg" width="800"/>
+</div>
+
+<p align="center">End-to-end OCR comparison between PP-OCRv6_medium and PP-OCRv5_server across Chinese, English, Japanese, artistic fonts, industrial characters, rotated text, pinyin, and dot-matrix characters.</p>
+
+# 5. Quick Start
+
+```python
+from paddleocr import PaddleOCR
+
+# Default: PP-OCRv6_medium
+ocr = PaddleOCR(
+    use_doc_orientation_classify=False,
+    use_doc_unwarping=False,
+    use_textline_orientation=False,
+)
+result = ocr.predict("https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png")
+
+for res in result:
+    res.print()
+    res.save_to_img("output")
+    res.save_to_json("output")
+```
+
+```bash
+# CLI usage
+paddleocr ocr -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False \
+    --use_textline_orientation False
+```
+
+**Using Transformers Engine:**
+
+PP-OCRv6 supports inference via Hugging Face Transformers (requires `transformers>=5.8.0`):
+
+```python
+from paddleocr import TextRecognition
+
+model = TextRecognition(
+    model_name="PP-OCRv6_medium_rec",
+    engine="transformers",
+)
+output = model.predict(input="general_ocr_rec_001.png", batch_size=1)
+for res in output:
+    res.print()
+```
+
+**Using High-Performance Inference (ONNX Runtime backend):**
+
+Enable the high-performance inference plugin with `enable_hpi=True`:
+
+```python
+from paddleocr import PaddleOCR
+
+ocr = PaddleOCR(
+    use_doc_orientation_classify=False,
+    use_doc_unwarping=False,
+    use_textline_orientation=False,
+    enable_hpi=True,
+)
+result = ocr.predict("general_ocr_002.png")
+```
+
+> The HPI plugin requires additional installation. See [High-Performance Inference Guide](../../inference_deployment/local_inference/high_performance_inference.md).
+
+# 6. Deployment and Custom Development
+
+* **Multi-OS Support**: Compatible with Windows, Linux, and Mac.
+* **Multi-Hardware Support**: Supports NVIDIA GPU, Intel CPU, Kunlun, Ascend, and more.
+* **High-Performance Inference Plugin**: See [High-Performance Inference Guide](../../inference_deployment/local_inference/high_performance_inference.md).
+* **Serving Deployment**: See [Serving Deployment Guide](../../inference_deployment/serving/serving.md).
+* **Custom Development**: Supports custom dataset training, dictionary extension, and model fine-tuning. See [Text Detection Tutorial](../../module_usage/text_detection.md) and [Text Recognition Tutorial](../../module_usage/text_recognition.md).
diff --git a/docs/version3.x/algorithm/PP-OCRv6/PP-OCRv6.md b/docs/version3.x/algorithm/PP-OCRv6/PP-OCRv6.md
new file mode 100644
index 0000000..742f3e9
--- /dev/null
+++ b/docs/version3.x/algorithm/PP-OCRv6/PP-OCRv6.md
@@ -0,0 +1,258 @@
+# 一、PP-OCRv6简介
+
+**PP-OCRv6** 是 PP-OCR 最新一代通用文字识别解决方案。PP-OCRv6 基于全新设计的 PPLCNetV4 统一骨干网络，提供 tiny、small、medium 三档模型，分别面向端侧/IoT、移动端/桌面端、服务端场景。PP-OCRv6 在语言覆盖方面实现重大突破，medium/small 档单一模型统一支持简体中文、繁体中文、英文、日文及 46 种拉丁语系语言共 50 种语言（tiny 档支持 49 种，不含日文）。在内部多场景综合评估集上，PP-OCRv6_medium 相比 PP-OCRv5_server 识别精度提升 5.1%、检测精度提升 4.6%，同时 GPU 推理速度提升 2.37×；以仅 34.5M 参数的规模，精度超越 Qwen3-VL-235B、GPT-5.5 等大型视觉语言模型。
+
+PP-OCRv6 的主要贡献如下：
+
+1. **统一可扩展的模型族**：提供覆盖 1.5M 至 34.5M 参数的三档完整 OCR 模型族。medium 档达到 86.2% 检测 Hmean 和 83.2% 识别准确率，可作为工业部署和大规模数据管线的高效生产级基础设施。
+2. **面向 OCR 的轻量级架构创新**：提出一系列专为 OCR 任务定制的轻量级架构组件——(i) LCNetV4：集成结构重参数化的 MetaFormer 风格轻量骨干；(ii) RepLKFPN：利用膨胀可重参数化深度卷积实现大感受野的检测颈部；(iii) EncoderWithLightSVTR：基于局部-全局注意力和加性跳跃连接的识别颈部。
+3. **广泛的多语言与多场景泛化**：单一模型扩展至支持 50 种语言和多种挑战性工业场景（如数码显示屏、点阵字符、轮胎印字等），显著提升了传统通用视觉语言模型难以覆盖的专业场景 OCR 性能。
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/v6acc_opt.png" width="800"/>
+</div>
+
+<p align="center">图：PP-OCRv6 与 PP-OCRv5 及视觉语言模型的性能对比。左：文本检测平均 Hmean（%）；右：文本识别加权平均准确率（%）。</p>
+
+# 二、核心技术升级
+
+## 1. 统一骨干网络 PPLCNetV4
+
+PP-OCRv6 采用全新设计的 PPLCNetV4 作为检测和识别的统一骨干网络，核心创新包括：
+
+**LCNetV4Block**：遵循 MetaFormer 范式，将每层解耦为 Token Mixer 和 Channel Mixer。设输入特征 $\mathbf{x} \in \mathbb{R}^{C \times H \times W}$，Block 计算如下：
+
+$$\hat{\mathbf{x}} = \text{SE}(\text{DW}(\mathbf{x})) + \mathbf{x}$$
+
+$$\mathbf{y} = W_2\,\sigma(W_1\,\hat{\mathbf{x}}) + \hat{\mathbf{x}}$$
+
+其中 $\text{DW}(\cdot)$ 是 3×3 深度卷积（Token Mixer），SE 是可选的通道注意力模块，$W_1 \in \mathbb{R}^{2C \times C}$、$W_2 \in \mathbb{R}^{C \times 2C}$ 构成扩展比为 2 的 Channel Mixer，$\sigma$ 为 GELU 激活。
+
+**Task-Adaptive Downsampling**：同一骨干通过不同下采样策略服务两个任务——检测模式使用标准 stride-2 空间下采样产出多尺度特征图（stride 4/8/16/32）；识别模式在 Stage 3/4 使用非对称 stride $(2,1)$，仅缩减高度保留宽度，经 height-axis 平均池化后产出 1-D 序列特征用于 CTC/NRTR 解码。
+
+**与 LCNetV3 对比**：
+
+| 设计维度 | LCNetV3 | LCNetV4 |
+|---------|---------|---------|
+| 架构范式 | MobileNet-style (DW→SE→PW) | MetaFormer (TokenMixer + ChannelMixer) |
+| 通道交互 | 单个 1×1 PW Conv | Expand(2×)→Act→Compress + 残差 |
+| 空间混合 | 普通 DW Conv | RepDWConv（3×3 + 1×1 + identity 三分支） |
+| BN 初始化 | 标准 | Compress 层 BN 零初始化 |
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/backbone.png" width="600"/>
+</div>
+
+<p align="center">图：PPLCNetV4 骨干网络结构</p>
+
+## 2. 检测模块升级
+
+- **RepLKFPN**：轻量级大核特征金字塔，使用 DilatedReparamBlock（7×7 深度卷积 + 膨胀分支），相比 PP-OCRv5 的 RSEFPN 参数减少 31%（118K vs 172K），同时感受野从 3×3 扩大到 7×7。
+- **辅助深度监督**：在 P2、P3、P4 层级添加预测头，训练时提供更强梯度信号。
+- **DiceBCE Loss**：组合 DiceLoss + Focal Loss，对小目标和密集文本提供更好的逐像素监督。
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/ppocrv6_det_pip_ori.png" width="600"/>
+</div>
+
+<p align="center">图：PP-OCRv6 检测模块结构</p>
+
+## 3. 识别模块升级
+
+- **EncoderWithLightSVTR 颈部**：局部上下文建模（1×7 深度卷积）+ 全局自注意力（1-2 层 Transformer），通过加性跳跃连接（而非 PP-OCRv5 的拼接）减少参数。
+- **多头解码器**：CTCHead 用于高效并行推理，NRTRHead 用于训练时辅助监督（推理时移除）。
+- **Tiny 模型特殊设计**：无颈部（直接 reshape + FC），使用 medium 模型蒸馏训练。
+- **多语言统一**：字典扩展约 200 个带变音符号字符，实现单模型 48 语言覆盖。
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/rec.png" width="600"/>
+</div>
+
+<p align="center">图：PP-OCRv6 识别模块结构</p>
+
+# 三、关键指标
+
+## 1. 文本检测指标
+
+在内部多场景基准上对 16 类场景进行文本检测 Hmean(%) 评测：
+
+| 模型 | AVG | 手写CN | 手写EN | 印刷CN | 印刷EN | 繁体 | 古籍 | 日文 | 模糊 | 表情 | 扭曲 | 拼音 | 艺术字 | 表格 | 旋转 | 工业 | 通用 |
+|------|-----|--------|--------|--------|--------|------|------|------|------|------|------|------|--------|------|------|------|------|
+| **PP-OCRv6_medium** | **86.2** | **83.7** | 84.0 | **95.1** | **93.7** | **86.3** | **80.2** | **84.3** | **94.1** | 99.6 | **88.6** | **74.0** | **69.0** | 96.8 | **93.8** | **73.3** | **82.8** |
+| **PP-OCRv6_small** | **84.1** | 80.5 | **87.1** | 94.2 | 93.6 | 85.7 | 72.6 | 82.3 | 92.6 | 99.7 | 87.6 | 69.6 | 65.3 | 95.6 | 93.7 | 67.6 | 78.2 |
+| **PP-OCRv6_tiny** | **80.6** | 79.4 | 85.9 | 93.1 | 92.3 | 83.7 | 63.0 | 76.6 | 89.3 | **99.8** | 86.1 | 59.0 | 60.1 | 94.7 | 91.0 | 62.0 | 73.8 |
+| PP-OCRv5_server | 81.6 | 80.3 | 84.1 | 94.5 | 91.7 | 81.5 | 67.6 | 77.2 | 90.1 | 96.2 | 87.6 | 67.1 | 67.3 | **97.1** | 80.0 | 64.3 | 79.7 |
+| PP-OCRv5_mobile | 75.2 | 74.4 | 77.7 | 90.5 | 91.0 | 82.3 | 58.1 | 72.7 | 87.4 | 93.6 | 82.7 | 57.5 | 52.5 | 92.8 | 64.7 | 52.8 | 72.1 |
+| Gemini-3.1-Pro | 46.8 | 53.4 | 56.5 | 47.3 | 47.6 | 39.0 | 45.8 | 38.2 | 50.0 | 68.1 | 44.6 | 40.6 | 65.2 | 26.9 | 22.1 | 52.5 | 50.2 |
+| GPT-5.5 | 45.6 | 42.4 | 58.5 | 50.2 | 51.9 | 35.0 | 26.7 | 42.0 | 49.1 | 97.5 | 37.7 | 36.3 | 52.0 | 71.0 | 10.0 | 36.2 | 32.6 |
+| Qwen3-VL-235B | 38.3 | 56.5 | 66.0 | 41.7 | 37.0 | 19.3 | 13.1 | 27.0 | 38.5 | 81.2 | 28.5 | 33.0 | 68.3 | 19.6 | 2.1 | 48.4 | 32.3 |
+
+PP-OCRv6_medium 平均 Hmean 达 86.2%，相比 PP-OCRv5_server 提升 4.6 个百分点，在日文、古籍、旋转文本、工业字符等场景提升尤为显著。
+
+## 2. 文本识别指标
+
+在内部多场景基准上对 15 类场景进行文本识别准确率(%) 评测：
+
+| 模型 | W-Avg | 手写CN | 手写EN | 印刷CN | 印刷EN | 繁体 | 古籍 | 日文 | 易混淆 | 特殊字符 | 通用 | 拼音 | 艺术字 | 工业 | 屏幕 | 卡片 |
+|------|-------|--------|--------|--------|--------|------|------|------|--------|----------|------|------|--------|------|------|------|
+| **PP-OCRv6_medium** | **83.2** | **62.1** | 67.8 | **91.5** | **94.1** | **78.6** | **72.4** | **90.5** | **64.9** | **61.7** | **87.5** | **78.1** | **71.2** | **77.4** | **82.5** | **88.1** |
+| **PP-OCRv6_small** | **81.3** | 57.6 | 61.1 | 90.5 | 93.3 | 77.0 | 71.1 | 88.2 | 64.1 | 60.2 | 85.7 | 75.9 | 68.4 | 76.4 | 79.7 | 86.9 |
+| **PP-OCRv6_tiny** | 73.5 | 40.1 | 39.3 | 86.7 | 88.4 | 65.0 | 68.4 | 89.8 | 52.3 | 57.1 | 78.0 | 65.4 | 54.7 | 62.1 | 71.2 | 80.5 |
+| PP-OCRv5_server | 78.1 | 58.0 | 59.6 | 90.1 | 85.1 | 74.7 | 60.4 | 73.7 | 59.4 | 56.8 | 86.5 | 74.4 | 64.0 | 70.2 | 68.1 | 87.6 |
+| PP-OCRv5_mobile | 73.7 | 41.7 | 50.9 | 86.0 | 86.0 | 72.0 | 57.8 | 75.8 | 55.7 | 54.8 | 80.7 | 72.5 | 54.0 | 59.3 | 57.6 | 81.7 |
+| Qwen3-VL-235B | 74.9 | 49.7 | **73.2** | 82.3 | 86.2 | 76.4 | 33.6 | 66.2 | 56.1 | 49.0 | 82.5 | 76.5 | 69.6 | 74.7 | 73.8 | 78.7 |
+| Gemini-3.1-Pro | 71.4 | 46.4 | 73.0 | 80.0 | 90.5 | 69.5 | 18.0 | 67.2 | 54.4 | 50.3 | 74.6 | 75.9 | 63.1 | 69.1 | 73.2 | 75.9 |
+| GPT-5.5 | 64.2 | 19.2 | 56.9 | 75.7 | 82.2 | 57.5 | 63.7 | 58.6 | 49.1 | 48.3 | 67.7 | 50.4 | 53.0 | 62.4 | 67.7 | 71.1 |
+
+PP-OCRv6_medium 加权平均准确率 83.2%，相比 PP-OCRv5_server 提升 5.1%，在日文(+16.8%)、古籍(+12.0%)、屏幕显示(+14.4%) 等类别提升显著。即使是仅 1.1M 参数的 PP-OCRv6_tiny，也超越了 4/5 的 VLM 模型。
+
+## 4. 端到端推理速度（s/image）
+
+在 200 张图像（通用场景 + 文档场景）上测试端到端 OCR 产线速度，包含读图、前后处理、模型推理全流程。
+
+| 硬件 | 推理后端 | PP-OCRv6_medium | PP-OCRv6_small | PP-OCRv6_tiny | PP-OCRv5_server | PP-OCRv5_mobile | PP-OCRv4_mobile |
+|------|---------|-----------------|----------------|---------------|-----------------|-----------------|-----------------|
+| NVIDIA A100 | PaddlePaddle | 0.29 | 0.25 | 0.13 | 0.32 | 0.25 | 0.14 |
+| NVIDIA A100 | TensorRT | -- | 0.32 | 0.16 | -- | 0.33 | 0.16 |
+| NVIDIA V100 | PaddlePaddle | 0.72 | 0.49 | 0.21 | 0.66 | 0.50 | 0.25 |
+| NVIDIA V100 | ONNX Runtime | 0.67 | 0.53 | 0.29 | 0.77 | 0.46 | 0.27 |
+| NVIDIA V100 | TensorRT | 0.77 | 0.60 | 0.23 | 0.73 | 0.59 | 0.27 |
+| Intel Xeon 8350C | PaddlePaddle | 2.05 | 0.79 | 0.32 | 2.04 | 0.80 | 0.62 |
+| Intel Xeon 8350C | OpenVINO | 1.40 | 0.59 | 0.20 | 7.30 | 0.78 | 0.60 |
+| Intel Xeon 8350C | ONNX Runtime | 3.31 | 0.61 | 0.22 | 6.36 | 0.61 | 0.49 |
+| Apple M4 | PaddlePaddle | 8.82 | 3.07 | 0.96 | >10 | 5.82 | 5.65 |
+| Apple M4 | ONNX Runtime | 5.55 | 1.29 | 0.35 | 7.20 | 1.10 | 1.02 |
+
+- PP-OCRv6_medium 在所有平台上均匹配或优于 PP-OCRv5_server：A100 上快 1.1×（0.29s vs 0.32s），V100 ONNX Runtime 快 1.15×（0.67s vs 0.77s），Intel Xeon OpenVINO 快 5.2×（1.40s vs 7.30s）。
+- PP-OCRv6_small 在大多数平台上与 PP-OCRv5_mobile 速度持平但精度更高；Apple M4 PaddlePaddle 快 1.9×（3.07s vs 5.82s）。
+- PP-OCRv6_tiny 是所有平台上最快的模型，Apple M4 PaddlePaddle 快 6.1×（0.96s vs 5.82s），Intel Xeon OpenVINO 快 3.9×（0.20s vs 0.78s），A100 上仅需 0.13s。
+
+# 四、语言支持
+
+PP-OCRv6 medium/small 档支持以下 50 种语言：
+
+**核心语言**：简体中文、繁体中文、英文、日文
+
+**拉丁语系（46种）**：法文、德文、意大利文、西班牙文、葡萄牙文、荷兰文、波兰文、罗马尼亚文、捷克文、瑞典文、挪威文、丹麦文、芬兰文、匈牙利文、土耳其文、越南文、印尼文、马来文、阿塞拜疆文、南非荷兰文、波斯尼亚文、克罗地亚文、威尔士文、爱沙尼亚文、爱尔兰文、冰岛文、库尔德文、立陶宛文、拉脱维亚文、马耳他文、毛利文、奥克文、斯洛伐克文、斯洛文尼亚文、阿尔巴尼亚文、斯瓦希里文、他加禄文、乌兹别克文、拉丁文、塞尔维亚文(拉丁)、加泰罗尼亚文、巴斯克文、加利西亚文、卢森堡文、罗曼什文、克丘亚文
+
+> PP-OCRv6_tiny 档支持 49 种语言（不含日文，以避免约 4000 个汉字/假名字符对 1.1M 参数输出层的影响）。
+
+# 五、效果可视化
+
+## 1. 检测效果对比
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/ppocrv6_det_vis.jpg" width="800"/>
+</div>
+
+<p align="center">图：文本检测效果对比。从左到右：PP-OCRv6_medium、PP-OCRv5_server、Gemini-3.1-Pro、GPT-5.5。</p>
+
+## 2. 幻觉对比
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/huanjue.jpg" width="800"/>
+</div>
+
+<p align="center">图：PP-OCRv6_medium 与 VLM 的幻觉对比。PP-OCRv6 忠实还原图像中的文字内容，而 VLM 基于语言先验进行"纠正"，引入了图像中不存在的幻觉。</p>
+
+## 3. 端到端 OCR 效果对比
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/case1.jpg" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/case2.jpg" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-OCRv6/case3.jpg" width="800"/>
+</div>
+
+<p align="center">图：PP-OCRv6_medium 与 PP-OCRv5_server 端到端 OCR 效果对比，涵盖中文、英文、日文、艺术字、工业字符、旋转文本、拼音、点阵字符等场景。</p>
+
+# 六、快速使用
+
+```python
+from paddleocr import PaddleOCR
+
+# 默认使用 PP-OCRv6_medium
+ocr = PaddleOCR(
+    use_doc_orientation_classify=False,
+    use_doc_unwarping=False,
+    use_textline_orientation=False,
+)
+result = ocr.predict("https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png")
+
+for res in result:
+    res.print()
+    res.save_to_img("output")
+    res.save_to_json("output")
+
+```
+
+```bash
+# 命令行使用
+paddleocr ocr -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False \
+    --use_textline_orientation False
+```
+
+**使用 Transformers 引擎推理：**
+
+PP-OCRv6 支持通过 Hugging Face Transformers 引擎进行推理（需安装 `transformers>=5.8.0`）：
+
+```python
+from paddleocr import TextRecognition
+
+model = TextRecognition(
+    model_name="PP-OCRv6_medium_rec",
+    engine="transformers",
+)
+output = model.predict(input="general_ocr_rec_001.png", batch_size=1)
+for res in output:
+    res.print()
+```
+
+```bash
+# 命令行方式
+paddleocr text_recognition -i general_ocr_rec_001.png \
+    --text_recognition_model_name PP-OCRv6_medium_rec \
+    --engine transformers
+```
+
+**使用高性能推理（ONNX Runtime 后端）：**
+
+通过 `enable_hpi=True` 启用高性能推理插件，底层会自动使用 ONNX Runtime 加速：
+
+```python
+from paddleocr import PaddleOCR
+
+ocr = PaddleOCR(
+    use_doc_orientation_classify=False,
+    use_doc_unwarping=False,
+    use_textline_orientation=False,
+    enable_hpi=True,
+)
+result = ocr.predict("general_ocr_002.png")
+```
+
+```bash
+# 命令行方式
+paddleocr ocr -i general_ocr_002.png \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False \
+    --use_textline_orientation False \
+    --enable_hpi True
+```
+
+> 高性能推理插件需额外安装，详见[高性能推理指南](../../inference_deployment/local_inference/high_performance_inference.md)。
+
+# 七、部署与二次开发
+
+- **多系统支持**：兼容 Windows、Linux、Mac 等主流操作系统。
+- **多硬件支持**：支持英伟达 GPU、Intel CPU、昆仑芯、昇腾等硬件推理和部署。
+- **高性能推理插件**：推荐结合高性能推理插件进一步提升推理速度，详见[高性能推理指南](../../inference_deployment/local_inference/high_performance_inference.md)。
+- **服务化部署**：支持高稳定性服务化部署方案，详见[服务化部署指南](../../inference_deployment/serving/serving.md)。
+- **二次开发能力**：支持自定义数据集训练、字典扩展、模型微调，详见[文本检测模块使用教程](../../module_usage/text_detection.md)及[文本识别模块使用教程](../../module_usage/text_recognition.md)。
diff --git a/docs/version3.x/algorithm/PP-StructureV3/PP-StructureV3.en.md b/docs/version3.x/algorithm/PP-StructureV3/PP-StructureV3.en.md
new file mode 100644
index 0000000..e61c9ed
--- /dev/null
+++ b/docs/version3.x/algorithm/PP-StructureV3/PP-StructureV3.en.md
@@ -0,0 +1,1539 @@
+# Introduction to PP-StructureV3
+
+**PP-StructureV3** pipeline, based on the Layout Parsing v1 pipeline, has strengthened the ability of layout detection, table recognition, and formula recognition. It has also added the ability to understand charts and restore reading order, as well as the ability to convert results into Markdown files. In various document data, it performs excellently and can handle more complex document data. This pipeline also provides flexible service-oriented deployment methods, supporting the use of multiple programming languages on various hardware. Moreover, it also provides the ability for secondary development. You can train and optimize on your own dataset based on this pipeline, and the trained model can be seamlessly integrated.
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-StructureV3/algorithm_ppstructurev3.png" width="600"/>
+</div>
+
+# Key Metrics
+<table>
+  <thead>
+    <tr>
+      <th rowspan="2">Method Type</th>
+      <th rowspan="2">Methods</th>
+      <th colspan="2">Overall<sup>Edit</sup>↓</th>
+      <th colspan="2">Text<sup>Edit</sup>↓</th>
+      <th colspan="2">Formula<sup>Edit</sup>↓</th>
+      <th colspan="2">Table<sup>Edit</sup>↓</th>
+      <th colspan="2">Read Order<sup>Edit</sup>↓</th>
+    </tr>
+    <tr>
+      <th>EN</th>
+      <th>ZH</th>
+      <th>EN</th>
+      <th>ZH</th>
+      <th>EN</th>
+      <th>ZH</th>
+      <th>EN</th>
+      <th>ZH</th>
+      <th>EN</th>
+      <th>ZH</th>
+    </tr>
+  </thead>
+ <tbody>
+  <tr> 
+   <td rowspan="9">Pipeline Tools</td> 
+   <td><b>PP-structureV3</b></td> 
+   <td><b>0.145</b></td> 
+   <td><b>0.206</b></td> 
+   <td>0.058</td> 
+   <td><b>0.088</b></td> 
+   <td>0.295</td> 
+   <td>0.535</td> 
+   <td>0.159</td> 
+   <td><b>0.109</b></td> 
+   <td>0.069</td> 
+   <td><b>0.091</b></td> 
+  </tr> 
+  <tr> 
+   <td>MinerU-0.9.3</td> 
+   <td>0.15</td> 
+   <td>0.357</td> 
+   <td>0.061</td> 
+   <td>0.215</td> 
+   <td>0.278</td> 
+   <td>0.577</td> 
+   <td>0.18</td> 
+   <td>0.344</td> 
+   <td>0.079</td> 
+   <td>0.292</td> 
+  </tr> 
+  <tr> 
+   <td>MinerU-1.3.11</td> 
+   <td>0.166</td> 
+   <td>0.310</td> 
+   <td>0.0826</td> 
+   <td>0.2000</td> 
+   <td>0.3368</td> 
+   <td>0.6236</td> 
+   <td>0.1613</td> 
+   <td>0.1833</td> 
+   <td>0.0834</td> 
+   <td>0.2316</td> 
+  </tr> 
+  <tr> 
+   <td>Marker-1.2.3</td> 
+   <td>0.336</td> 
+   <td>0.556</td> 
+   <td>0.08</td> 
+   <td>0.315</td> 
+   <td>0.53</td> 
+   <td>0.883</td> 
+   <td>0.619</td> 
+   <td>0.685</td> 
+   <td>0.114</td> 
+   <td>0.34</td> 
+  </tr> 
+  <tr> 
+   <td>Mathpix</td> 
+   <td>0.191</td> 
+   <td>0.365</td> 
+   <td>0.105</td> 
+   <td>0.384</td> 
+   <td>0.306</td> 
+   <td>0.454</td> 
+   <td>0.243</td> 
+   <td>0.32</td> 
+   <td>0.108</td> 
+   <td>0.304</td> 
+  </tr> 
+  <tr> 
+   <td>Docling-2.14.0</td> 
+   <td>0.589</td> 
+   <td>0.909</td> 
+   <td>0.416</td> 
+   <td>0.987</td> 
+   <td>0.999</td> 
+   <td>1</td> 
+   <td>0.627</td> 
+   <td>0.81</td> 
+   <td>0.313</td> 
+   <td>0.837</td> 
+  </tr> 
+  <tr> 
+   <td>Pix2Text-1.1.2.3</td> 
+   <td>0.32</td> 
+   <td>0.528</td> 
+   <td>0.138</td> 
+   <td>0.356</td> 
+   <td><b>0.276</b></td> 
+   <td>0.611</td> 
+   <td>0.584</td> 
+   <td>0.645</td> 
+   <td>0.281</td> 
+   <td>0.499</td> 
+  </tr> 
+  <tr> 
+   <td>Unstructured-0.17.2</td> 
+   <td>0.586</td> 
+   <td>0.716</td> 
+   <td>0.198</td> 
+   <td>0.481</td> 
+   <td>0.999</td> 
+   <td>1</td> 
+   <td>1</td> 
+   <td>0.998</td> 
+   <td>0.145</td> 
+   <td>0.387</td> 
+  </tr> 
+  <tr> 
+   <td>OpenParse-0.7.0</td> 
+   <td>0.646</td> 
+   <td>0.814</td> 
+   <td>0.681</td> 
+   <td>0.974</td> 
+   <td>0.996</td> 
+   <td>1</td> 
+   <td>0.284</td> 
+   <td>0.639</td> 
+   <td>0.595</td> 
+   <td>0.641</td> 
+  </tr> 
+  <tr> 
+   <td rowspan="5">Expert VLMs</td> 
+   <td>GOT-OCR</td> 
+   <td>0.287</td> 
+   <td>0.411</td> 
+   <td>0.189</td> 
+   <td>0.315</td> 
+   <td>0.36</td> 
+   <td>0.528</td> 
+   <td>0.459</td> 
+   <td>0.52</td> 
+   <td>0.141</td> 
+   <td>0.28</td> 
+  </tr> 
+  <tr> 
+   <td>Nougat</td> 
+   <td>0.452</td> 
+   <td>0.973</td> 
+   <td>0.365</td> 
+   <td>0.998</td> 
+   <td>0.488</td> 
+   <td>0.941</td> 
+   <td>0.572</td> 
+   <td>1</td> 
+   <td>0.382</td> 
+   <td>0.954</td> 
+  </tr> 
+  <tr> 
+   <td>Mistral OCR</td> 
+   <td>0.268</td> 
+   <td>0.439</td> 
+   <td>0.072</td> 
+   <td>0.325</td> 
+   <td>0.318</td> 
+   <td>0.495</td> 
+   <td>0.6</td> 
+   <td>0.65</td> 
+   <td>0.083</td> 
+   <td>0.284</td> 
+  </tr> 
+  <tr> 
+   <td>OLMOCR-sglang</td> 
+   <td>0.326</td> 
+   <td>0.469</td> 
+   <td>0.097</td> 
+   <td>0.293</td> 
+   <td>0.455</td> 
+   <td>0.655</td> 
+   <td>0.608</td> 
+   <td>0.652</td> 
+   <td>0.145</td> 
+   <td>0.277</td> 
+  </tr> 
+  <tr> 
+   <td>SmolDocling-256M_transformer</td> 
+   <td>0.493</td> 
+   <td>0.816</td> 
+   <td>0.262</td> 
+   <td>0.838</td> 
+   <td>0.753</td> 
+   <td>0.997</td> 
+   <td>0.729</td> 
+   <td>0.907</td> 
+   <td>0.227</td> 
+   <td>0.522</td> 
+  </tr> 
+  <tr> 
+   <td rowspan="6">General VLMs</td> 
+   <td>Gemini2.0-flash</td> 
+   <td>0.191</td> 
+   <td>0.264</td> 
+   <td>0.091</td> 
+   <td>0.139</td> 
+   <td>0.389</td> 
+   <td>0.584</td> 
+   <td>0.193</td> 
+   <td>0.206</td> 
+   <td>0.092</td> 
+   <td>0.128</td> 
+  </tr> 
+  <tr> 
+   <td>Gemini2.5-Pro</td> 
+   <td>0.148</td> 
+   <td><b>0.212</b></td> 
+   <td><b>0.055</b></td> 
+   <td>0.168</td> 
+   <td>0.356</td> 
+   <td>0.439</td> 
+   <td><b>0.13</b></td> 
+   <td>0.119</td> 
+   <td><b>0.049</b></td> 
+   <td>0.121</td> 
+  </tr> 
+  <tr> 
+   <td>GPT4o</td> 
+   <td>0.233</td> 
+   <td>0.399</td> 
+   <td>0.144</td> 
+   <td>0.409</td> 
+   <td>0.425</td> 
+   <td>0.606</td> 
+   <td>0.234</td> 
+   <td>0.329</td> 
+   <td>0.128</td> 
+   <td>0.251</td> 
+  </tr> 
+  <tr> 
+   <td>Qwen2-VL-72B</td> 
+   <td>0.252</td> 
+   <td>0.327</td> 
+   <td>0.096</td> 
+   <td>0.218</td> 
+   <td>0.404</td> 
+   <td>0.487</td> 
+   <td>0.387</td> 
+   <td>0.408</td> 
+   <td>0.119</td> 
+   <td>0.193</td> 
+  </tr> 
+  <tr> 
+   <td>Qwen2.5-VL-72B</td> 
+   <td>0.214</td> 
+   <td>0.261</td> 
+   <td>0.092</td> 
+   <td>0.18</td> 
+   <td>0.315</td> 
+   <td><b>0.434</b></td> 
+   <td>0.341</td> 
+   <td>0.262</td> 
+   <td>0.106</td> 
+   <td>0.168</td> 
+  </tr> 
+  <tr> 
+   <td>InternVL2-76B</td> 
+   <td>0.44</td> 
+   <td>0.443</td> 
+   <td>0.353</td> 
+   <td>0.29</td> 
+   <td>0.543</td> 
+   <td>0.701</td> 
+   <td>0.547</td> 
+   <td>0.555</td> 
+   <td>0.317</td> 
+   <td>0.228</td> 
+  </tr> 
+ </tbody>
+</table>
+
+The above data is from:
+* <a href="https://github.com/opendatalab/OmniDocBench">OmniDocBench</a>
+* <a href="https://arxiv.org/abs/2412.07626">OmniDocBench: Benchmarking Diverse PDF Document Parsing with Comprehensive Annotations</a>
+
+# End to End Benchmark
+
+The performance of PP-StructureV3 and MinerU with different configurations under different GPU environments are as follows.
+
+Requirements:
+* Paddle 3.0
+* PaddleOCR 3.0.0
+* MinerU 1.3.10
+* CUDA 11.8
+* cuDNN 8.9
+
+## Local inference
+
+Local inference was tested with both V100 and A100 GPU, evaluating the performance of PP-StructureV3 under 6 different configurations. The test data consists of 15 PDF files, totaling 925 pages, including elements such as tables, formulas, seals, and charts.
+
+In the following PP-StructureV3 configuration, please refer to [PP-OCRv5](../PP-OCRv5/PP-OCRv5.en.md) for OCR model details, see [Formula Recognition](../../module_usage/formula_recognition.en.md) for formula recognition model details, and refer to [Text Detection](../../module_usage/text_detection.en.md) for the max_side_limit setting of the text detection module.
+
+### Env: NVIDIA Tesla V100 + Intel Xeon Gold 6271C
+
+<table border="1">
+ <tr>
+  <td>
+   Methods
+  </td>
+  <td colspan="4">
+   Configurations
+  </td>
+  <td rowspan="2">
+   Average time per
+  page
+    (s)
+  </td>
+  <td rowspan="2">
+   Average CPU
+    （%）
+  </td>
+  <td rowspan="2">
+   Peak RAM Usage
+    （GB）
+  </td>
+  <td rowspan="2">
+   Average RAM
+  Usage
+    （GB）
+  </td>
+  <td rowspan="2">
+   Average GPU
+    （%）
+  </td>
+  <td rowspan="2">
+   Peak VRAM Usage
+    （GB）
+  </td>
+  <td rowspan="2">
+   Average VRAM
+  Usage
+    （GB）
+  </td>
+ </tr>
+ <tr>
+  <td rowspan="7">
+   PP-StructureV3
+  </td>
+  <td>
+   OCR Models
+  </td>
+  <td>
+   Formula Recognition Model
+  </td>
+  <td>
+   Chart Recognition Model
+  </td>
+  <td>
+   text detection module max_side_limit
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Server
+  </td>
+  <td>
+   PP-FormulaNet-L
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   1.77
+  </td>
+  <td>
+   111.4
+  </td>
+  <td>
+   6.7
+  </td>
+  <td>
+   5.2
+  </td>
+  <td>
+   38.9
+  </td>
+  <td>
+   17.0
+  </td>
+  <td>
+   16.5
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Server
+  </td>
+  <td>
+   PP-FormulaNet-L
+  </td>
+  <td>
+   ✔
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   4.09
+  </td>
+  <td>
+   105.3
+  </td>
+  <td>
+   5.5
+  </td>
+  <td>
+   4.0
+  </td>
+  <td>
+   24.7
+  </td>
+  <td>
+   17.0
+  </td>
+  <td>
+   16.6
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Mobile
+  </td>
+  <td>
+   PP-FormulaNet-L
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   1.56
+  </td>
+  <td>
+   113.7
+  </td>
+  <td>
+   6.6
+  </td>
+  <td>
+   4.9
+  </td>
+  <td>
+   29.1
+  </td>
+  <td>
+   10.7
+  </td>
+  <td>
+   10.6
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Server
+  </td>
+  <td>
+   PP-FormulaNet-M
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   1.42
+  </td>
+  <td>
+   112.9
+  </td>
+  <td>
+   6.8
+  </td>
+  <td>
+   5.1
+  </td>
+  <td>
+   38
+  </td>
+  <td>
+   16.0
+  </td>
+  <td>
+   15.5
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Mobile
+  </td>
+  <td>
+   PP-FormulaNet-M
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   1.15
+  </td>
+  <td>
+   114.8
+  </td>
+  <td>
+   6.5
+  </td>
+  <td>
+   5.0
+  </td>
+  <td>
+   26.1
+  </td>
+  <td>
+   8.4
+  </td>
+  <td>
+   8.3
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Mobile
+  </td>
+  <td>
+   PP-FormulaNet-M
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   1200
+  </td>
+  <td>
+   0.99
+  </td>
+  <td>
+   113
+  </td>
+  <td>
+   7.0
+  </td>
+  <td>
+   5.6
+  </td>
+  <td>
+   29.2
+  </td>
+  <td>
+   8.6
+  </td>
+  <td>
+   8.5
+  </td>
+ </tr>
+ <tr>
+  <td>
+   MinerU
+  </td>
+  <td colspan="4">
+   -
+  </td>
+  <td>
+   1.57
+  </td>
+  <td>
+   142.9
+  </td>
+  <td>
+   13.3
+  </td>
+  <td>
+   11.8
+  </td>
+  <td>
+   43.3
+  </td>
+  <td>
+   31.6
+  </td>
+  <td>
+   9.7
+  </td>
+ </tr>
+</table>
+
+### NVIDIA A100 + Intel Xeon Platinum 8350C
+
+<table border="1">
+ <tr>
+  <td>
+   Methods
+  </td>
+  <td colspan="4">
+   Configurations
+  </td>
+  <td rowspan="2">
+   Average time per
+  page
+    (s)
+  </td>
+  <td rowspan="2">
+   Average CPU
+    （%）
+  </td>
+  <td rowspan="2">
+   Peak RAM Usage
+    （GB）
+  </td>
+  <td rowspan="2">
+   Average RAM
+  Usage
+    （GB）
+  </td>
+  <td rowspan="2">
+   Average GPU
+    （%）
+  </td>
+  <td rowspan="2">
+   Peak VRAM Usage
+    （GB）
+  </td>
+  <td rowspan="2">
+   Average VRAM
+  Usage
+    （GB）
+  </td>
+ </tr>
+ <tr>
+  <td rowspan="7">
+   PP-StructureV3
+  </td>
+  <td>
+   OCR Models
+  </td>
+  <td>
+   Formula Recognition Model
+  </td>
+  <td>
+   Chart Recognition Model
+  </td>
+  <td>
+   text detection module max_side_limit
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Server
+  </td>
+  <td>
+   PP-FormulaNet-L
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   1.12
+  </td>
+  <td>
+   109.8
+  </td>
+  <td>
+   9.2
+  </td>
+  <td>
+   7.8
+  </td>
+  <td>
+   29.8
+  </td>
+  <td>
+   21.8
+  </td>
+  <td>
+   21.1
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Server
+  </td>
+  <td>
+   PP-FormulaNet-L
+  </td>
+  <td>
+   ✔
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   2.76
+  </td>
+  <td>
+   103.7
+  </td>
+  <td>
+   9.0
+  </td>
+  <td>
+   7.7
+  </td>
+  <td>
+   24
+  </td>
+  <td>
+   21.8
+  </td>
+  <td>
+   21.1
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Mobile
+  </td>
+  <td>
+   PP-FormulaNet-L
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   1.04
+  </td>
+  <td>
+   110.7
+  </td>
+  <td>
+   9.3
+  </td>
+  <td>
+   7.8
+  </td>
+  <td>
+   22
+  </td>
+  <td>
+   12.2
+  </td>
+  <td>
+   12.1
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Server
+  </td>
+  <td>
+   PP-FormulaNet-M
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   0.95
+  </td>
+  <td>
+   111.4
+  </td>
+  <td>
+   9.1
+  </td>
+  <td>
+   7.8
+  </td>
+  <td>
+   28.1
+  </td>
+  <td>
+   21.8
+  </td>
+  <td>
+   21.0
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Mobile
+  </td>
+  <td>
+   PP-FormulaNet-M
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   0.89
+  </td>
+  <td>
+   112.1
+  </td>
+  <td>
+   9.2
+  </td>
+  <td>
+   7.8
+  </td>
+  <td>
+   18.5
+  </td>
+  <td>
+   11.4
+  </td>
+  <td>
+   11.2
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Mobile
+  </td>
+  <td>
+   PP-FormulaNet-M
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   1200
+  </td>
+  <td>
+   0.64
+  </td>
+  <td>
+   113.5
+  </td>
+  <td>
+   10.2
+  </td>
+  <td>
+   8.5
+  </td>
+  <td>
+   23.7
+  </td>
+  <td>
+   11.4
+  </td>
+  <td>
+   11.2
+  </td>
+ </tr>
+ <tr>
+  <td>
+   MinerU
+  </td>
+  <td colspan="4">
+   -
+  </td>
+  <td>
+   1.06
+  </td>
+  <td>
+   168.3
+  </td>
+  <td>
+   18.3
+  </td>
+  <td>
+   16.8
+  </td>
+  <td>
+   27.5
+  </td>
+  <td>
+   76.9
+  </td>
+  <td>
+   14.8
+  </td>
+ </tr>
+</table>
+
+## Serving Inference
+
+The serving inference test is based on the NVIDIA A100 + Intel Xeon Platinum 8350C environment, with test data consisting of 1500 images, including tables, formulas, seals, charts, and other elements.
+
+<table>
+ <tbody>
+  <tr> 
+   <td>Instances Number</td> 
+   <td>Concurrent Requests Number</td> 
+   <td>Throughput</td> 
+   <td>Average Latency (s)</td> 
+   <td>Success Number/Total Number</td> 
+  </tr> 
+  <tr"> 
+   <td>4 GPUs ✖️ 1 instance/gpu</td> 
+   <td>4</td> 
+   <td>1.69</td> 
+   <td>2.36</td> 
+   <td>100%</td> 
+  </tr> 
+  <tr"> 
+   <td>4 GPUs ✖️ 4 instances/gpu</td> 
+   <td>16</td> 
+   <td>4.05</td> 
+   <td>3.87</td> 
+   <td>100%</td> 
+  </tr> 
+ </tbody>
+</table>
+
+## Pipeline benchmark data
+
+<details>
+<summary>Click to expand/collapse the table</summary>
+
+<table border="1">
+<tr><th>Pipeline configuration</th><th>Hardware</th><th>Avg. inference time (s)</th><th>Peak CPU utilization (%)</th><th>Avg. CPU utilization (%)</th><th>Peak host memory (MB)</th><th>Avg. host memory (MB)</th><th>Peak GPU utilization (%)</th><th>Avg. GPU utilization (%)</th><th>Peak device memory (MB)</th><th>Avg. device memory (MB)</th></tr>
+<tr>
+<td rowspan="5">PP_StructureV3-default</td>
+<td>Intel 8350C + A100</td>
+<td>1.38</td>
+<td>1384.60</td>
+<td>113.26</td>
+<td>5781.59</td>
+<td>3431.21</td>
+<td>100</td>
+<td>32.79</td>
+<td>37370.00</td>
+<td>34165.68</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>2.38</td>
+<td>608.70</td>
+<td>109.96</td>
+<td>6388.91</td>
+<td>3737.19</td>
+<td>100</td>
+<td>39.08</td>
+<td>26824.00</td>
+<td>24581.61</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>1.36</td>
+<td>744.30</td>
+<td>112.82</td>
+<td>6199.01</td>
+<td>3865.78</td>
+<td>100</td>
+<td>43.81</td>
+<td>35132.00</td>
+<td>32077.12</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>1.74</td>
+<td>418.50</td>
+<td>105.96</td>
+<td>6138.25</td>
+<td>3503.41</td>
+<td>100</td>
+<td>48.54</td>
+<td>18536.00</td>
+<td>18353.93</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>3.70</td>
+<td>434.40</td>
+<td>105.45</td>
+<td>6865.87</td>
+<td>3595.68</td>
+<td>100</td>
+<td>71.92</td>
+<td>13970.00</td>
+<td>12668.58</td>
+</tr>
+<tr>
+<td rowspan="3">PP_StructureV3-pp</td>
+<td>Intel 8350C + A100</td>
+<td>3.50</td>
+<td>679.30</td>
+<td>105.96</td>
+<td>13850.20</td>
+<td>5146.50</td>
+<td>100</td>
+<td>14.01</td>
+<td>37656.00</td>
+<td>34716.95</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>5.03</td>
+<td>494.20</td>
+<td>105.63</td>
+<td>13542.94</td>
+<td>4833.55</td>
+<td>100</td>
+<td>20.36</td>
+<td>29402.00</td>
+<td>26607.92</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>3.17</td>
+<td>481.50</td>
+<td>105.13</td>
+<td>14179.97</td>
+<td>5608.80</td>
+<td>100</td>
+<td>19.35</td>
+<td>35454.00</td>
+<td>32512.19</td>
+</tr>
+<tr>
+<td rowspan="2">PP_StructureV3-full</td>
+<td>Intel 8350C + A100</td>
+<td>8.92</td>
+<td>697.30</td>
+<td>102.88</td>
+<td>13777.07</td>
+<td>4573.65</td>
+<td>100</td>
+<td>18.39</td>
+<td>38776.00</td>
+<td>37554.09</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>13.12</td>
+<td>437.40</td>
+<td>102.36</td>
+<td>13974.00</td>
+<td>4484.00</td>
+<td>100</td>
+<td>17.50</td>
+<td>29878.00</td>
+<td>28733.59</td>
+</tr>
+<tr>
+<td rowspan="5">PP_StructureV3-seal</td>
+<td>Intel 8350C + A100</td>
+<td>1.39</td>
+<td>747.50</td>
+<td>112.55</td>
+<td>5788.79</td>
+<td>3742.03</td>
+<td>100</td>
+<td>33.81</td>
+<td>38966.00</td>
+<td>35832.44</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>2.44</td>
+<td>630.10</td>
+<td>110.18</td>
+<td>6343.39</td>
+<td>3725.98</td>
+<td>100</td>
+<td>42.23</td>
+<td>28078.00</td>
+<td>25834.70</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>1.40</td>
+<td>792.20</td>
+<td>113.63</td>
+<td>6673.60</td>
+<td>4417.34</td>
+<td>100</td>
+<td>46.33</td>
+<td>35530.00</td>
+<td>32516.87</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>1.75</td>
+<td>422.40</td>
+<td>106.08</td>
+<td>6068.87</td>
+<td>3973.49</td>
+<td>100</td>
+<td>50.12</td>
+<td>19630.00</td>
+<td>18374.37</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>3.76</td>
+<td>400.30</td>
+<td>105.10</td>
+<td>6296.28</td>
+<td>3651.42</td>
+<td>100</td>
+<td>72.57</td>
+<td>14304.00</td>
+<td>13268.36</td>
+</tr>
+<tr>
+<td rowspan="4">PP_StructureV3-chart</td>
+<td>Intel 8350C + A100</td>
+<td>7.70</td>
+<td>746.80</td>
+<td>102.69</td>
+<td>6355.58</td>
+<td>4006.48</td>
+<td>100</td>
+<td>22.38</td>
+<td>37380.00</td>
+<td>36730.73</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>10.58</td>
+<td>599.20</td>
+<td>102.51</td>
+<td>5754.14</td>
+<td>3333.78</td>
+<td>100</td>
+<td>21.99</td>
+<td>26820.00</td>
+<td>26253.70</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>8.03</td>
+<td>413.30</td>
+<td>101.31</td>
+<td>6473.29</td>
+<td>3689.84</td>
+<td>100</td>
+<td>26.19</td>
+<td>18540.00</td>
+<td>18494.69</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>11.69</td>
+<td>460.90</td>
+<td>101.85</td>
+<td>6503.12</td>
+<td>3524.06</td>
+<td>100</td>
+<td>46.81</td>
+<td>13966.00</td>
+<td>12481.94</td>
+</tr>
+<tr>
+<td rowspan="5">PP_StructureV3-notable</td>
+<td>Intel 8350C + A100</td>
+<td>1.24</td>
+<td>738.30</td>
+<td>110.45</td>
+<td>5638.16</td>
+<td>3278.30</td>
+<td>100</td>
+<td>35.32</td>
+<td>30320.00</td>
+<td>27026.17</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>2.24</td>
+<td>452.40</td>
+<td>107.79</td>
+<td>5579.15</td>
+<td>3635.95</td>
+<td>100</td>
+<td>43.00</td>
+<td>23098.00</td>
+<td>20684.43</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>1.18</td>
+<td>989.00</td>
+<td>107.71</td>
+<td>6041.76</td>
+<td>4024.76</td>
+<td>100</td>
+<td>50.67</td>
+<td>33780.00</td>
+<td>29733.15</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>1.58</td>
+<td>225.00</td>
+<td>102.56</td>
+<td>5518.10</td>
+<td>3333.08</td>
+<td>100</td>
+<td>49.90</td>
+<td>21532.00</td>
+<td>18567.99</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>3.40</td>
+<td>413.30</td>
+<td>103.58</td>
+<td>5874.88</td>
+<td>3662.49</td>
+<td>100</td>
+<td>76.82</td>
+<td>13764.00</td>
+<td>11890.62</td>
+</tr>
+<tr>
+<td rowspan="7">PP_StructureV3-noformula</td>
+<td>Intel 6271C</td>
+<td>7.85</td>
+<td>1172.50</td>
+<td>964.70</td>
+<td>17739.00</td>
+<td>11101.02</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>8.83</td>
+<td>1053.50</td>
+<td>970.64</td>
+<td>15463.48</td>
+<td>9408.19</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.84</td>
+<td>788.60</td>
+<td>124.25</td>
+<td>6246.39</td>
+<td>3674.32</td>
+<td>100</td>
+<td>30.57</td>
+<td>40084.00</td>
+<td>37358.45</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>1.42</td>
+<td>606.20</td>
+<td>115.53</td>
+<td>7015.57</td>
+<td>3707.03</td>
+<td>100</td>
+<td>35.63</td>
+<td>29540.00</td>
+<td>27620.28</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.87</td>
+<td>644.10</td>
+<td>119.23</td>
+<td>6895.76</td>
+<td>4222.85</td>
+<td>100</td>
+<td>50.00</td>
+<td>36878.00</td>
+<td>34104.59</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>1.03</td>
+<td>377.50</td>
+<td>106.87</td>
+<td>5819.88</td>
+<td>3830.19</td>
+<td>100</td>
+<td>42.87</td>
+<td>19340.00</td>
+<td>17550.94</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>2.02</td>
+<td>430.20</td>
+<td>109.21</td>
+<td>6600.62</td>
+<td>3824.18</td>
+<td>100</td>
+<td>65.75</td>
+<td>14332.00</td>
+<td>12712.18</td>
+</tr>
+<tr>
+<td rowspan="9">PP_StructureV3-lightweight</td>
+<td>Intel 6271C</td>
+<td>4.36</td>
+<td>1189.70</td>
+<td>995.78</td>
+<td>14000.50</td>
+<td>9374.97</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>3.74</td>
+<td>1049.60</td>
+<td>967.77</td>
+<td>12960.96</td>
+<td>7644.25</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Hygon 7490 + P800</td>
+<td>0.86</td>
+<td>572.20</td>
+<td>120.84</td>
+<td>8290.49</td>
+<td>3569.44</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.61</td>
+<td>823.40</td>
+<td>126.25</td>
+<td>9258.22</td>
+<td>3776.63</td>
+<td>52</td>
+<td>18.95</td>
+<td>7456.00</td>
+<td>7131.95</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>1.07</td>
+<td>686.80</td>
+<td>116.70</td>
+<td>9381.75</td>
+<td>4126.28</td>
+<td>58</td>
+<td>22.92</td>
+<td>8450.00</td>
+<td>8083.30</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.46</td>
+<td>999.00</td>
+<td>122.21</td>
+<td>9734.78</td>
+<td>4516.40</td>
+<td>61</td>
+<td>24.41</td>
+<td>7524.00</td>
+<td>7167.52</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.70</td>
+<td>355.40</td>
+<td>111.51</td>
+<td>9415.45</td>
+<td>4094.06</td>
+<td>89</td>
+<td>30.85</td>
+<td>7248.00</td>
+<td>6927.58</td>
+</tr>
+<tr>
+<td>M4</td>
+<td>12.22</td>
+<td>223.60</td>
+<td>107.35</td>
+<td>9531.22</td>
+<td>7884.61</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>1.13</td>
+<td>461.40</td>
+<td>112.16</td>
+<td>7923.09</td>
+<td>3837.31</td>
+<td>85</td>
+<td>41.67</td>
+<td>8218.00</td>
+<td>7902.04</td>
+</tr>
+</table>
+
+
+<table border="1">
+<tr><th>Pipeline configuration</th><th>description</th></tr>
+<tr>
+<td>PP_StructureV3-default</td>
+<td>Default configuration</td>
+</tr>
+<tr>
+<td>PP_StructureV3-pp</td>
+<td>Based on the default configuration, document image preprocessing is enabled</td>
+</tr>
+<tr>
+<td>PP_StructureV3-full</td>
+<td>Based on the default configuration, document image preprocessing and chart parsing are enabled</td>
+</tr>
+<tr>
+<td>PP_StructureV3-seal</td>
+<td>Based on the default configuration, seal text recognition is enabled</td>
+</tr>
+<tr>
+<td>PP_StructureV3-chart</td>
+<td>Based on the default configuration, chart parsing is enabled</td>
+</tr>
+<tr>
+<td>PP_StructureV3-notable</td>
+<td>Based on the default configuration, table recognition is disabled</td>
+</tr>
+<tr>
+<td>PP_StructureV3-noformula</td>
+<td>Based on the default configuration, formula recognition is disabled</td>
+</tr>
+<tr>
+<td>PP_StructureV3-lightweight</td>
+<td>Based on the default configuration, all task models are replaced with lightweight versions</td>
+</tr>
+</table>
+</details>
+
+
+* Test environment:
+    * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
+    * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
+    * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
+* Test data:
+    * Test data containing 280 images including tables, seals, formulas, and charts.
+* Test strategy:
+    * Warm up with 20 samples, then repeat the full dataset once for performance testing.
+* Note:
+    * Since we did not collect device memory data for NPU and XPU, the corresponding entries in the table are marked as N/A.
+
+# PP-StructureV3 Demo
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-StructureV3/algorithm_ppstructurev3_demo.png" width="600"/>
+</div>
+
+<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex%2FPaddleX3.0%2Fdoc_images%2FPP-StructureV3%2Falgorithm_ppstructurev3_demo.pdf">More Demos</a>
+
+
+# FAQ
+
+**Q: What is the default model configuration? If I need higher accuracy, faster speed, or lower GPU memory usage, which parameters should I adjust or which models should I switch to? How significant is the impact on results?**
+
+**A:** By default, the largest models for each module are used. Section 3.3 demonstrates how different model selections affect GPU memory consumption and inference speed. You can choose an appropriate model based on your device capabilities and the complexity of your samples. Additionally, in the Python API or CLI, you can set the device parameter as `<device_type>:<device_id1>,<device_id2>...` (e.g., `gpu:0,1,2,3`) to enable multi-GPU parallel inference. If the built-in multi-GPU parallel inference does not meet your speed requirements, you may refer to the example code for multi-process parallel inference and further optimize it for your specific scenario: [Multi-process Parallel Inference](https://www.paddleocr.ai/latest/en/version3.x/inference_deployment/local_inference/parallel_inference.html).
+
+---
+
+**Q: Can PP-StructureV3 run on CPU?**
+
+**A:** While PP-StructureV3 is recommended to run on GPU for optimal performance, it also supports CPU inference. Thanks to a variety of configuration options and sufficient optimization for lightweight models, users can refer to section 3.3 to select lightweight configurations for CPU-only environments. For example, on an Intel 8350C CPU, the inference time per image is about 3.74 seconds.
+
+---
+
+**Q: How can I integrate PP-StructureV3 into my own project?**
+
+**A:**  
+- For Python projects, you can directly integrate using the PaddleOCR Python API.  
+- For projects in other programming languages, it is recommended to use service-based deployment. PaddleOCR supports client-side integration in multiple languages, including C++, C#, Java, Go, and PHP. Please refer to the [official documentation](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PP-StructureV3.html#3-development-integration-deployment) for details.  
+- If you need to interact with large language models, PaddleOCR also provides the MCP service. For more information, please refer to the [MCP Server documentation](https://www.paddleocr.ai/latest/en/version3.x/integrations/mcp_server.html).
+
+---
+
+**Q: Can the serving handle concurrent requests?**
+
+**A:** In the basic service deployment solution, the service processes only one request at a time, which is mainly intended for quick validation, development pipeline integration, or scenarios that do not require concurrent processing. In the high-stability serving solution, the service also processes one request at a time by default, but users can achieve horizontal scaling and concurrent processing by adjusting the configuration as outlined in the service deployment guide.
+
+---
+
+**Q: How can I reduce latency and increase throughput in serving?**
+
+**A:** PaddleOCR offers two types of service deployment solutions. Regardless of the solution used, enabling high-performance inference plugins can accelerate model inference and thus reduce latency. For the high-stability deployment solution, throughput can be further increased by adjusting the service configuration to run multiple instances, making full use of your hardware resources. Please refer to the [documentation](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_deploy/serving.html#22-adjust-configurations) for more details on configuring high-stability serving.
diff --git a/docs/version3.x/algorithm/PP-StructureV3/PP-StructureV3.md b/docs/version3.x/algorithm/PP-StructureV3/PP-StructureV3.md
new file mode 100644
index 0000000..aa9b2ad
--- /dev/null
+++ b/docs/version3.x/algorithm/PP-StructureV3/PP-StructureV3.md
@@ -0,0 +1,1533 @@
+# 一、PP-StructureV3 简介
+PP-StructureV3 能够将文档图像和 PDF 文件高效转换为结构化内容（如 Markdown 格式），并具备版面区域检测、表格识别、公式识别、图表理解以及多栏阅读顺序恢复等强大功能。该工具在多种文档类型下均表现优异，能够处理复杂的文档数据。PP-StructureV3 支持灵活的服务化部署，兼容多种硬件环境，并可通过多种编程语言进行调用。同时，支持二次开发，用户可以基于自有数据集进行模型训练和优化，训练后的模型可实现无缝集成。
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-StructureV3/algorithm_ppstructurev3.png" width="800"/>
+</div>
+
+# 二、关键指标
+
+<div align="center">
+<table>
+  <thead>
+    <tr>
+      <th rowspan="2">Method Type</th>
+      <th rowspan="2">Methods</th>
+      <th colspan="2">Overall<sup>Edit</sup>↓</th>
+      <th colspan="2">Text<sup>Edit</sup>↓</th>
+      <th colspan="2">Formula<sup>Edit</sup>↓</th>
+      <th colspan="2">Table<sup>Edit</sup>↓</th>
+      <th colspan="2">Read Order<sup>Edit</sup>↓</th>
+    </tr>
+    <tr>
+      <th>EN</th>
+      <th>ZH</th>
+      <th>EN</th>
+      <th>ZH</th>
+      <th>EN</th>
+      <th>ZH</th>
+      <th>EN</th>
+      <th>ZH</th>
+      <th>EN</th>
+      <th>ZH</th>
+    </tr>
+  </thead>
+ <tbody>
+  <tr> 
+   <td rowspan="9">Pipeline Tools</td> 
+   <td><b>PP-structureV3</b></td> 
+   <td><b>0.145</b></td> 
+   <td><b>0.206</b></td> 
+   <td>0.058</td> 
+   <td><b>0.088</b></td> 
+   <td>0.295</td> 
+   <td>0.535</td> 
+   <td>0.159</td> 
+   <td><b>0.109</b></td> 
+   <td>0.069</td> 
+   <td><b>0.091</b></td> 
+  </tr> 
+  <tr> 
+   <td>MinerU-0.9.3</td> 
+   <td>0.15</td> 
+   <td>0.357</td> 
+   <td>0.061</td> 
+   <td>0.215</td> 
+   <td>0.278</td> 
+   <td>0.577</td> 
+   <td>0.18</td> 
+   <td>0.344</td> 
+   <td>0.079</td> 
+   <td>0.292</td> 
+  </tr> 
+  <tr> 
+   <td>MinerU-1.3.11</td> 
+   <td>0.166</td> 
+   <td>0.310</td> 
+   <td>0.0826</td> 
+   <td>0.2000</td> 
+   <td>0.3368</td> 
+   <td>0.6236</td> 
+   <td>0.1613</td> 
+   <td>0.1833</td> 
+   <td>0.0834</td> 
+   <td>0.2316</td> 
+  </tr> 
+  <tr> 
+   <td>Marker-1.2.3</td> 
+   <td>0.336</td> 
+   <td>0.556</td> 
+   <td>0.08</td> 
+   <td>0.315</td> 
+   <td>0.53</td> 
+   <td>0.883</td> 
+   <td>0.619</td> 
+   <td>0.685</td> 
+   <td>0.114</td> 
+   <td>0.34</td> 
+  </tr> 
+  <tr> 
+   <td>Mathpix</td> 
+   <td>0.191</td> 
+   <td>0.365</td> 
+   <td>0.105</td> 
+   <td>0.384</td> 
+   <td>0.306</td> 
+   <td>0.454</td> 
+   <td>0.243</td> 
+   <td>0.32</td> 
+   <td>0.108</td> 
+   <td>0.304</td> 
+  </tr> 
+  <tr> 
+   <td>Docling-2.14.0</td> 
+   <td>0.589</td> 
+   <td>0.909</td> 
+   <td>0.416</td> 
+   <td>0.987</td> 
+   <td>0.999</td> 
+   <td>1</td> 
+   <td>0.627</td> 
+   <td>0.81</td> 
+   <td>0.313</td> 
+   <td>0.837</td> 
+  </tr> 
+  <tr> 
+   <td>Pix2Text-1.1.2.3</td> 
+   <td>0.32</td> 
+   <td>0.528</td> 
+   <td>0.138</td> 
+   <td>0.356</td> 
+   <td><b>0.276</b></td> 
+   <td>0.611</td> 
+   <td>0.584</td> 
+   <td>0.645</td> 
+   <td>0.281</td> 
+   <td>0.499</td> 
+  </tr> 
+  <tr> 
+   <td>Unstructured-0.17.2</td> 
+   <td>0.586</td> 
+   <td>0.716</td> 
+   <td>0.198</td> 
+   <td>0.481</td> 
+   <td>0.999</td> 
+   <td>1</td> 
+   <td>1</td> 
+   <td>0.998</td> 
+   <td>0.145</td> 
+   <td>0.387</td> 
+  </tr> 
+  <tr> 
+   <td>OpenParse-0.7.0</td> 
+   <td>0.646</td> 
+   <td>0.814</td> 
+   <td>0.681</td> 
+   <td>0.974</td> 
+   <td>0.996</td> 
+   <td>1</td> 
+   <td>0.284</td> 
+   <td>0.639</td> 
+   <td>0.595</td> 
+   <td>0.641</td> 
+  </tr> 
+  <tr> 
+   <td rowspan="5">Expert VLMs</td> 
+   <td>GOT-OCR</td> 
+   <td>0.287</td> 
+   <td>0.411</td> 
+   <td>0.189</td> 
+   <td>0.315</td> 
+   <td>0.36</td> 
+   <td>0.528</td> 
+   <td>0.459</td> 
+   <td>0.52</td> 
+   <td>0.141</td> 
+   <td>0.28</td> 
+  </tr> 
+  <tr> 
+   <td>Nougat</td> 
+   <td>0.452</td> 
+   <td>0.973</td> 
+   <td>0.365</td> 
+   <td>0.998</td> 
+   <td>0.488</td> 
+   <td>0.941</td> 
+   <td>0.572</td> 
+   <td>1</td> 
+   <td>0.382</td> 
+   <td>0.954</td> 
+  </tr> 
+  <tr> 
+   <td>Mistral OCR</td> 
+   <td>0.268</td> 
+   <td>0.439</td> 
+   <td>0.072</td> 
+   <td>0.325</td> 
+   <td>0.318</td> 
+   <td>0.495</td> 
+   <td>0.6</td> 
+   <td>0.65</td> 
+   <td>0.083</td> 
+   <td>0.284</td> 
+  </tr> 
+  <tr> 
+   <td>OLMOCR-sglang</td> 
+   <td>0.326</td> 
+   <td>0.469</td> 
+   <td>0.097</td> 
+   <td>0.293</td> 
+   <td>0.455</td> 
+   <td>0.655</td> 
+   <td>0.608</td> 
+   <td>0.652</td> 
+   <td>0.145</td> 
+   <td>0.277</td> 
+  </tr> 
+  <tr> 
+   <td>SmolDocling-256M_transformer</td> 
+   <td>0.493</td> 
+   <td>0.816</td> 
+   <td>0.262</td> 
+   <td>0.838</td> 
+   <td>0.753</td> 
+   <td>0.997</td> 
+   <td>0.729</td> 
+   <td>0.907</td> 
+   <td>0.227</td> 
+   <td>0.522</td> 
+  </tr> 
+  <tr> 
+   <td rowspan="6">General VLMs</td> 
+   <td>Gemini2.0-flash</td> 
+   <td>0.191</td> 
+   <td>0.264</td> 
+   <td>0.091</td> 
+   <td>0.139</td> 
+   <td>0.389</td> 
+   <td>0.584</td> 
+   <td>0.193</td> 
+   <td>0.206</td> 
+   <td>0.092</td> 
+   <td>0.128</td> 
+  </tr> 
+  <tr> 
+   <td>Gemini2.5-Pro</td> 
+   <td>0.148</td> 
+   <td><b>0.212</b></td> 
+   <td><b>0.055</b></td> 
+   <td>0.168</td> 
+   <td>0.356</td> 
+   <td>0.439</td> 
+   <td><b>0.13</b></td> 
+   <td>0.119</td> 
+   <td><b>0.049</b></td> 
+   <td>0.121</td> 
+  </tr> 
+  <tr> 
+   <td>GPT4o</td> 
+   <td>0.233</td> 
+   <td>0.399</td> 
+   <td>0.144</td> 
+   <td>0.409</td> 
+   <td>0.425</td> 
+   <td>0.606</td> 
+   <td>0.234</td> 
+   <td>0.329</td> 
+   <td>0.128</td> 
+   <td>0.251</td> 
+  </tr> 
+  <tr> 
+   <td>Qwen2-VL-72B</td> 
+   <td>0.252</td> 
+   <td>0.327</td> 
+   <td>0.096</td> 
+   <td>0.218</td> 
+   <td>0.404</td> 
+   <td>0.487</td> 
+   <td>0.387</td> 
+   <td>0.408</td> 
+   <td>0.119</td> 
+   <td>0.193</td> 
+  </tr> 
+  <tr> 
+   <td>Qwen2.5-VL-72B</td> 
+   <td>0.214</td> 
+   <td>0.261</td> 
+   <td>0.092</td> 
+   <td>0.18</td> 
+   <td>0.315</td> 
+   <td><b>0.434</b></td> 
+   <td>0.341</td> 
+   <td>0.262</td> 
+   <td>0.106</td> 
+   <td>0.168</td> 
+  </tr> 
+  <tr> 
+   <td>InternVL2-76B</td> 
+   <td>0.44</td> 
+   <td>0.443</td> 
+   <td>0.353</td> 
+   <td>0.29</td> 
+   <td>0.543</td> 
+   <td>0.701</td> 
+   <td>0.547</td> 
+   <td>0.555</td> 
+   <td>0.317</td> 
+   <td>0.228</td> 
+  </tr> 
+ </tbody>
+</table>
+</div>
+
+以上部分数据出自：
+* <a href="https://github.com/opendatalab/OmniDocBench">OmniDocBench</a>
+* <a href="https://arxiv.org/abs/2412.07626">OmniDocBench: Benchmarking Diverse PDF Document Parsing with Comprehensive Annotations</a>
+
+
+# 三、推理 Benchmark
+
+在不同GPU环境下，不同配置的 PP-StructureV3 和 MinerU 对比的性能指标如下。
+
+基本测试环境：
+* Paddle 3.0正式版
+* PaddleOCR 3.0.0正式版
+* MinerU 1.3.10
+* CUDA 11.8
+* cuDNN 8.9
+
+## 3.1 本地推理
+
+本地推理分别在 V100 和 A100 两种 GPU机器上，测试了 6 种不同配置下 PP-StructureV3 的性能，测试数据为15个PDF文件，共925页，包含表格、公式、印章、图表等元素。
+
+下述 PP-StructureV3 配置中，OCR 模型详情请见[PP-OCRv5](../PP-OCRv5/PP-OCRv5.md)，公式识别模型详情请见[公式识别](../../module_usage/formula_recognition.md)，文本检测模块 max_side_limit 设置请见[文本检测](../../module_usage/text_detection.md)。
+
+### NVIDIA Tesla V100 + Intel Xeon Gold 6271C
+<table border="1">
+ <tr>
+  <td>
+   方案
+  </td>
+  <td colspan="4">
+   配置
+  </td>
+  <td rowspan="2">
+   平均每页耗时
+    （s）
+  </td>
+  <td rowspan="2">
+   平均CPU利用率
+    （%）
+  </td>
+  <td rowspan="2">
+   峰值RAM用量
+    （GB）
+  </td>
+  <td rowspan="2">
+   平均RAM用量
+    （GB）
+  </td>
+  <td rowspan="2">
+   平均GPU利用率
+    （%）
+  </td>
+  <td rowspan="2">
+   峰值VRAM用量
+    （GB）
+  </td>
+  <td rowspan="2">
+   平均VRAM用量
+    （GB）
+  </td>
+ </tr>
+ <tr>
+  <td rowspan="7">
+   PP-StructureV3
+  </td>
+  <td>
+   OCR模型
+  </td>
+  <td>
+   公式识别模型
+  </td>
+  <td>
+   是否启用图表识别模块
+  </td>
+  <td>
+   文本检测max_side_limit
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Server系列
+  </td>
+  <td>
+   PP-FormulaNet-L
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   1.77
+  </td>
+  <td>
+   111.4
+  </td>
+  <td>
+   6.7
+  </td>
+  <td>
+   5.2
+  </td>
+  <td>
+   38.9
+  </td>
+  <td>
+   17.0
+  </td>
+  <td>
+   16.5
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Server系列
+  </td>
+  <td>
+   PP-FormulaNet-L
+  </td>
+  <td>
+   ✔
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   4.09
+  </td>
+  <td>
+   105.3
+  </td>
+  <td>
+   5.5
+  </td>
+  <td>
+   4.0
+  </td>
+  <td>
+   24.7
+  </td>
+  <td>
+   17.0
+  </td>
+  <td>
+   16.6
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Mobile系列
+  </td>
+  <td>
+   PP-FormulaNet-L
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   1.56
+  </td>
+  <td>
+   113.7
+  </td>
+  <td>
+   6.6
+  </td>
+  <td>
+   4.9
+  </td>
+  <td>
+   29.1
+  </td>
+  <td>
+   10.7
+  </td>
+  <td>
+   10.6
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Server系列
+  </td>
+  <td>
+   PP-FormulaNet-M
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   1.42
+  </td>
+  <td>
+   112.9
+  </td>
+  <td>
+   6.8
+  </td>
+  <td>
+   5.1
+  </td>
+  <td>
+   38
+  </td>
+  <td>
+   16.0
+  </td>
+  <td>
+   15.5
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Mobile系列
+  </td>
+  <td>
+   PP-FormulaNet-M
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   1.15
+  </td>
+  <td>
+   114.8
+  </td>
+  <td>
+   6.5
+  </td>
+  <td>
+   5.0
+  </td>
+  <td>
+   26.1
+  </td>
+  <td>
+   8.4
+  </td>
+  <td>
+   8.3
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Mobile系列
+  </td>
+  <td>
+   PP-FormulaNet-M
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   1200
+  </td>
+  <td>
+   0.99
+  </td>
+  <td>
+   113
+  </td>
+  <td>
+   7.0
+  </td>
+  <td>
+   5.6
+  </td>
+  <td>
+   29.2
+  </td>
+  <td>
+   8.6
+  </td>
+  <td>
+   8.5
+  </td>
+ </tr>
+ <tr>
+  <td>
+   MinerU
+  </td>
+  <td colspan="4">
+   -
+  </td>
+  <td>
+   1.57
+  </td>
+  <td>
+   142.9
+  </td>
+  <td>
+   13.3
+  </td>
+  <td>
+   11.8
+  </td>
+  <td>
+   43.3
+  </td>
+  <td>
+   31.6
+  </td>
+  <td>
+   9.7
+  </td>
+ </tr>
+</table>
+
+### NVIDIA A100 + Intel Xeon Platinum 8350C
+<table border="1">
+ <tr>
+  <td>
+   方案
+  </td>
+  <td colspan="4">
+   配置
+  </td>
+  <td rowspan="2">
+   平均每页耗时
+    （s）
+  </td>
+  <td rowspan="2">
+   平均CPU利用率
+    （%）
+  </td>
+  <td rowspan="2">
+   峰值RAM用量
+    （GB）
+  </td>
+  <td rowspan="2">
+   平均RAM用量
+    （GB）
+  </td>
+  <td rowspan="2">
+   平均GPU利用率
+    （%）
+  </td>
+  <td rowspan="2">
+   峰值VRAM用量
+    （GB）
+  </td>
+  <td rowspan="2">
+   平均VRAM用量
+    （GB）
+  </td>
+ </tr>
+ <tr>
+  <td rowspan="7">
+   PP-StructureV3
+  </td>
+  <td>
+   OCR模型
+  </td>
+  <td>
+   公式识别模型
+  </td>
+  <td>
+   是否启用图表识别模块
+  </td>
+  <td>
+   文本检测max_side_limit
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Server系列
+  </td>
+  <td>
+   PP-FormulaNet-L
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   1.12
+  </td>
+  <td>
+   109.8
+  </td>
+  <td>
+   9.2
+  </td>
+  <td>
+   7.8
+  </td>
+  <td>
+   29.8
+  </td>
+  <td>
+   21.8
+  </td>
+  <td>
+   21.1
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Server系列
+  </td>
+  <td>
+   PP-FormulaNet-L
+  </td>
+  <td>
+   ✔
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   2.76
+  </td>
+  <td>
+   103.7
+  </td>
+  <td>
+   9.0
+  </td>
+  <td>
+   7.7
+  </td>
+  <td>
+   24
+  </td>
+  <td>
+   21.8
+  </td>
+  <td>
+   21.1
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Mobile系列
+  </td>
+  <td>
+   PP-FormulaNet-L
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   1.04
+  </td>
+  <td>
+   110.7
+  </td>
+  <td>
+   9.3
+  </td>
+  <td>
+   7.8
+  </td>
+  <td>
+   22
+  </td>
+  <td>
+   12.2
+  </td>
+  <td>
+   12.1
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Server系列
+  </td>
+  <td>
+   PP-FormulaNet-M
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   0.95
+  </td>
+  <td>
+   111.4
+  </td>
+  <td>
+   9.1
+  </td>
+  <td>
+   7.8
+  </td>
+  <td>
+   28.1
+  </td>
+  <td>
+   21.8
+  </td>
+  <td>
+   21.0
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Mobile系列
+  </td>
+  <td>
+   PP-FormulaNet-M
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   4096
+  </td>
+  <td>
+   0.89
+  </td>
+  <td>
+   112.1
+  </td>
+  <td>
+   9.2
+  </td>
+  <td>
+   7.8
+  </td>
+  <td>
+   18.5
+  </td>
+  <td>
+   11.4
+  </td>
+  <td>
+   11.2
+  </td>
+ </tr>
+ <tr>
+  <td>
+   Mobile系列
+  </td>
+  <td>
+   PP-FormulaNet-M
+  </td>
+  <td>
+   ✗
+  </td>
+  <td>
+   1200
+  </td>
+  <td>
+   0.64
+  </td>
+  <td>
+   113.5
+  </td>
+  <td>
+   10.2
+  </td>
+  <td>
+   8.5
+  </td>
+  <td>
+   23.7
+  </td>
+  <td>
+   11.4
+  </td>
+  <td>
+   11.2
+  </td>
+ </tr>
+ <tr>
+  <td>
+   MinerU
+  </td>
+  <td colspan="4">
+   -
+  </td>
+  <td>
+   1.06
+  </td>
+  <td>
+   168.3
+  </td>
+  <td>
+   18.3
+  </td>
+  <td>
+   16.8
+  </td>
+  <td>
+   27.5
+  </td>
+  <td>
+   76.9
+  </td>
+  <td>
+   14.8
+  </td>
+ </tr>
+</table>
+
+## 3.2 服务化部署
+
+服务化部署测试基于 NVIDIA A100 + Intel Xeon Platinum 8350C 环境，测试数据为 1500 张图像，包含表格、公式、印章、图表等元素。
+
+<table>
+ <tbody>
+  <tr> 
+   <td>实例数</td> 
+   <td>并发请求数</td> 
+   <td>吞吐</td> 
+   <td>平均时延（s）</td> 
+   <td>成功请求数/总请求数</td> 
+  </tr> 
+  <tr"> 
+   <td>4卡 ✖️ 1实例/卡</td>
+   <td>4</td> 
+   <td>1.69</td> 
+   <td>2.36</td> 
+   <td>100%</td> 
+  </tr> 
+  <tr"> 
+   <td>4卡 ✖️ 4实例/卡</td> 
+   <td>16</td> 
+   <td>4.05</td> 
+   <td>3.87</td> 
+   <td>100%</td> 
+  </tr> 
+ </tbody>
+</table>
+
+## 3.3 产线基准测试数据
+
+<details>
+<summary>点击展开/折叠表格</summary>
+
+<table border="1">
+<tr><th>流水线配置</th><th>硬件</th><th>平均推理时间 (s)</th><th>峰值CPU利用率 (%)</th><th>平均CPU利用率 (%)</th><th>峰值主机内存 (MB)</th><th>平均主机内存 (MB)</th><th>峰值GPU利用率 (%)</th><th>平均GPU利用率 (%)</th><th>峰值设备内存 (MB)</th><th>平均设备内存 (MB)</th></tr>
+<tr>
+<td rowspan="5">PP_StructureV3-default</td>
+<td>Intel 8350C + A100</td>
+<td>1.38</td>
+<td>1384.60</td>
+<td>113.26</td>
+<td>5781.59</td>
+<td>3431.21</td>
+<td>100</td>
+<td>32.79</td>
+<td>37370.00</td>
+<td>34165.68</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>2.38</td>
+<td>608.70</td>
+<td>109.96</td>
+<td>6388.91</td>
+<td>3737.19</td>
+<td>100</td>
+<td>39.08</td>
+<td>26824.00</td>
+<td>24581.61</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>1.36</td>
+<td>744.30</td>
+<td>112.82</td>
+<td>6199.01</td>
+<td>3865.78</td>
+<td>100</td>
+<td>43.81</td>
+<td>35132.00</td>
+<td>32077.12</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>1.74</td>
+<td>418.50</td>
+<td>105.96</td>
+<td>6138.25</td>
+<td>3503.41</td>
+<td>100</td>
+<td>48.54</td>
+<td>18536.00</td>
+<td>18353.93</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>3.70</td>
+<td>434.40</td>
+<td>105.45</td>
+<td>6865.87</td>
+<td>3595.68</td>
+<td>100</td>
+<td>71.92</td>
+<td>13970.00</td>
+<td>12668.58</td>
+</tr>
+<tr>
+<td rowspan="3">PP_StructureV3-pp</td>
+<td>Intel 8350C + A100</td>
+<td>3.50</td>
+<td>679.30</td>
+<td>105.96</td>
+<td>13850.20</td>
+<td>5146.50</td>
+<td>100</td>
+<td>14.01</td>
+<td>37656.00</td>
+<td>34716.95</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>5.03</td>
+<td>494.20</td>
+<td>105.63</td>
+<td>13542.94</td>
+<td>4833.55</td>
+<td>100</td>
+<td>20.36</td>
+<td>29402.00</td>
+<td>26607.92</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>3.17</td>
+<td>481.50</td>
+<td>105.13</td>
+<td>14179.97</td>
+<td>5608.80</td>
+<td>100</td>
+<td>19.35</td>
+<td>35454.00</td>
+<td>32512.19</td>
+</tr>
+<tr>
+<td rowspan="2">PP_StructureV3-full</td>
+<td>Intel 8350C + A100</td>
+<td>8.92</td>
+<td>697.30</td>
+<td>102.88</td>
+<td>13777.07</td>
+<td>4573.65</td>
+<td>100</td>
+<td>18.39</td>
+<td>38776.00</td>
+<td>37554.09</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>13.12</td>
+<td>437.40</td>
+<td>102.36</td>
+<td>13974.00</td>
+<td>4484.00</td>
+<td>100</td>
+<td>17.50</td>
+<td>29878.00</td>
+<td>28733.59</td>
+</tr>
+<tr>
+<td rowspan="5">PP_StructureV3-seal</td>
+<td>Intel 8350C + A100</td>
+<td>1.39</td>
+<td>747.50</td>
+<td>112.55</td>
+<td>5788.79</td>
+<td>3742.03</td>
+<td>100</td>
+<td>33.81</td>
+<td>38966.00</td>
+<td>35832.44</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>2.44</td>
+<td>630.10</td>
+<td>110.18</td>
+<td>6343.39</td>
+<td>3725.98</td>
+<td>100</td>
+<td>42.23</td>
+<td>28078.00</td>
+<td>25834.70</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>1.40</td>
+<td>792.20</td>
+<td>113.63</td>
+<td>6673.60</td>
+<td>4417.34</td>
+<td>100</td>
+<td>46.33</td>
+<td>35530.00</td>
+<td>32516.87</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>1.75</td>
+<td>422.40</td>
+<td>106.08</td>
+<td>6068.87</td>
+<td>3973.49</td>
+<td>100</td>
+<td>50.12</td>
+<td>19630.00</td>
+<td>18374.37</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>3.76</td>
+<td>400.30</td>
+<td>105.10</td>
+<td>6296.28</td>
+<td>3651.42</td>
+<td>100</td>
+<td>72.57</td>
+<td>14304.00</td>
+<td>13268.36</td>
+</tr>
+<tr>
+<td rowspan="4">PP_StructureV3-chart</td>
+<td>Intel 8350C + A100</td>
+<td>7.70</td>
+<td>746.80</td>
+<td>102.69</td>
+<td>6355.58</td>
+<td>4006.48</td>
+<td>100</td>
+<td>22.38</td>
+<td>37380.00</td>
+<td>36730.73</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>10.58</td>
+<td>599.20</td>
+<td>102.51</td>
+<td>5754.14</td>
+<td>3333.78</td>
+<td>100</td>
+<td>21.99</td>
+<td>26820.00</td>
+<td>26253.70</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>8.03</td>
+<td>413.30</td>
+<td>101.31</td>
+<td>6473.29</td>
+<td>3689.84</td>
+<td>100</td>
+<td>26.19</td>
+<td>18540.00</td>
+<td>18494.69</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>11.69</td>
+<td>460.90</td>
+<td>101.85</td>
+<td>6503.12</td>
+<td>3524.06</td>
+<td>100</td>
+<td>46.81</td>
+<td>13966.00</td>
+<td>12481.94</td>
+</tr>
+<tr>
+<td rowspan="5">PP_StructureV3-notable</td>
+<td>Intel 8350C + A100</td>
+<td>1.24</td>
+<td>738.30</td>
+<td>110.45</td>
+<td>5638.16</td>
+<td>3278.30</td>
+<td>100</td>
+<td>35.32</td>
+<td>30320.00</td>
+<td>27026.17</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>2.24</td>
+<td>452.40</td>
+<td>107.79</td>
+<td>5579.15</td>
+<td>3635.95</td>
+<td>100</td>
+<td>43.00</td>
+<td>23098.00</td>
+<td>20684.43</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>1.18</td>
+<td>989.00</td>
+<td>107.71</td>
+<td>6041.76</td>
+<td>4024.76</td>
+<td>100</td>
+<td>50.67</td>
+<td>33780.00</td>
+<td>29733.15</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>1.58</td>
+<td>225.00</td>
+<td>102.56</td>
+<td>5518.10</td>
+<td>3333.08</td>
+<td>100</td>
+<td>49.90</td>
+<td>21532.00</td>
+<td>18567.99</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>3.40</td>
+<td>413.30</td>
+<td>103.58</td>
+<td>5874.88</td>
+<td>3662.49</td>
+<td>100</td>
+<td>76.82</td>
+<td>13764.00</td>
+<td>11890.62</td>
+</tr>
+<tr>
+<td rowspan="7">PP_StructureV3-noformula</td>
+<td>Intel 6271C</td>
+<td>7.85</td>
+<td>1172.50</td>
+<td>964.70</td>
+<td>17739.00</td>
+<td>11101.02</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>8.83</td>
+<td>1053.50</td>
+<td>970.64</td>
+<td>15463.48</td>
+<td>9408.19</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.84</td>
+<td>788.60</td>
+<td>124.25</td>
+<td>6246.39</td>
+<td>3674.32</td>
+<td>100</td>
+<td>30.57</td>
+<td>40084.00</td>
+<td>37358.45</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>1.42</td>
+<td>606.20</td>
+<td>115.53</td>
+<td>7015.57</td>
+<td>3707.03</td>
+<td>100</td>
+<td>35.63</td>
+<td>29540.00</td>
+<td>27620.28</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.87</td>
+<td>644.10</td>
+<td>119.23</td>
+<td>6895.76</td>
+<td>4222.85</td>
+<td>100</td>
+<td>50.00</td>
+<td>36878.00</td>
+<td>34104.59</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>1.03</td>
+<td>377.50</td>
+<td>106.87</td>
+<td>5819.88</td>
+<td>3830.19</td>
+<td>100</td>
+<td>42.87</td>
+<td>19340.00</td>
+<td>17550.94</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>2.02</td>
+<td>430.20</td>
+<td>109.21</td>
+<td>6600.62</td>
+<td>3824.18</td>
+<td>100</td>
+<td>65.75</td>
+<td>14332.00</td>
+<td>12712.18</td>
+</tr>
+<tr>
+<td rowspan="9">PP_StructureV3-lightweight</td>
+<td>Intel 6271C</td>
+<td>4.36</td>
+<td>1189.70</td>
+<td>995.78</td>
+<td>14000.50</td>
+<td>9374.97</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C</td>
+<td>3.74</td>
+<td>1049.60</td>
+<td>967.77</td>
+<td>12960.96</td>
+<td>7644.25</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Hygon 7490 + P800</td>
+<td>0.86</td>
+<td>572.20</td>
+<td>120.84</td>
+<td>8290.49</td>
+<td>3569.44</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 8350C + A100</td>
+<td>0.61</td>
+<td>823.40</td>
+<td>126.25</td>
+<td>9258.22</td>
+<td>3776.63</td>
+<td>52</td>
+<td>18.95</td>
+<td>7456.00</td>
+<td>7131.95</td>
+</tr>
+<tr>
+<td>Intel 6271C + V100</td>
+<td>1.07</td>
+<td>686.80</td>
+<td>116.70</td>
+<td>9381.75</td>
+<td>4126.28</td>
+<td>58</td>
+<td>22.92</td>
+<td>8450.00</td>
+<td>8083.30</td>
+</tr>
+<tr>
+<td>Intel 8563C + H20</td>
+<td>0.46</td>
+<td>999.00</td>
+<td>122.21</td>
+<td>9734.78</td>
+<td>4516.40</td>
+<td>61</td>
+<td>24.41</td>
+<td>7524.00</td>
+<td>7167.52</td>
+</tr>
+<tr>
+<td>Intel 8350C + A10</td>
+<td>0.70</td>
+<td>355.40</td>
+<td>111.51</td>
+<td>9415.45</td>
+<td>4094.06</td>
+<td>89</td>
+<td>30.85</td>
+<td>7248.00</td>
+<td>6927.58</td>
+</tr>
+<tr>
+<td>M4</td>
+<td>12.22</td>
+<td>223.60</td>
+<td>107.35</td>
+<td>9531.22</td>
+<td>7884.61</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td>Intel 6271C + T4</td>
+<td>1.13</td>
+<td>461.40</td>
+<td>112.16</td>
+<td>7923.09</td>
+<td>3837.31</td>
+<td>85</td>
+<td>41.67</td>
+<td>8218.00</td>
+<td>7902.04</td>
+</tr>
+</table>
+
+
+<table border="1">
+<tr><th>Pipeline configuration</th><th>description</th></tr>
+<tr>
+<td>PP_StructureV3-default</td>
+<td>默认配置</td>
+</tr>
+<tr>
+<td>PP_StructureV3-pp</td>
+<td>默认配置基础上，开启文档图像预处理</td>
+</tr>
+<tr>
+<td>PP_StructureV3-full</td>
+<td>默认配置基础上，开启文档图像预处理和图表解析</td>
+</tr>
+<tr>
+<td>PP_StructureV3-seal</td>
+<td>默认配置基础上，开启印章文本识别</td>
+</tr>
+<tr>
+<td>PP_StructureV3-chart</td>
+<td>默认配置基础上，开启文档图表解析</td>
+</tr>
+<tr>
+<td>PP_StructureV3-notable</td>
+<td>默认配置基础上，关闭表格识别</td>
+</tr>
+<tr>
+<td>PP_StructureV3-noformula</td>
+<td>默认配置基础上，关闭公式识别</td>
+</tr>
+<tr>
+<td>PP_StructureV3-lightweight</td>
+<td>默认配置基础上，将所有任务模型都换成最轻量版本</td>
+</tr>
+</table>
+</details>
+
+
+* 测试环境：
+    * PaddlePaddle 3.1.0、CUDA 11.8、cuDNN 8.9
+    * PaddleX @ develop (f1eb28e23cfa54ce3e9234d2e61fcb87c93cf407)
+    * Docker image: ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.1.0-gpu-cuda11.8-cudnn8.9
+* 测试数据：
+    * 测试数据包含表格、印章、公式、图表的280张图像。
+* 测试策略：
+    * 使用 20 个样本进行预热，然后对整个数据集重复 1 次以进行速度性能测试。
+* 备注：
+    * 由于我们没有收集NPU和XPU的设备内存数据，因此表中相应位置的数据标记为N/A。
+
+# 四、PP-StructureV3 Demo示例
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/PP-StructureV3/algorithm_ppstructurev3_demo.png" width="600"/>
+</div> 
+
+<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex%2FPaddleX3.0%2Fdoc_images%2FPP-StructureV3%2Falgorithm_ppstructurev3_demo.pdf">更多示例</a>
+
+# 五、使用方法和常见问题
+
+**Q:默认模型是什么配置，如果需要更高精度、更快速度、或者更小显存，应该调哪些参数或者更换哪些模型，对结果影响大概有多大？**
+
+**A:** 默认模型均采用了了各个模块参数量最大的模型，3.3 章节中展示了不同的模型选择对于显存和推理速度的影响。可以根据设备情况和样本难易程度选择合适的模型。另外，在 Python API 或 CLI 设置 device 为<设备类型>:<设备编号1>,<设备编号2>...（例如gpu:0,1,2,3）可实现多卡并行推理。如果内置的多卡并行推理功能提速效果仍不满足预期，可参考多进程并行推理示例代码，结合具体场景进行进一步优化：[多进程并行推理](https://www.paddleocr.ai/latest/version3.x/inference_deployment/local_inference/parallel_inference.html)。
+
+---
+
+**Q: PP-StructureV3 是否可以在 CPU 上运行？**
+
+**A:** PP-StructureV3 虽然更推荐在 GPU 环境下进行推理，但也支持在 CPU 上运行。得益于多种配置选项及对轻量级模型的充分优化，在仅有 CPU 环境时，用户可以参考 3.3 节选择轻量化配置进行推理。例如，在 Intel 8350C CPU 上，每张图片的推理时间约为 3.74 秒。
+
+---
+
+**Q: 如何将 PP-StructureV3 集成到自己的项目中？**
+
+**A:**  
+- 对于 Python 项目，可以直接使用 PaddleOCR 的 Python API 完成集成。  
+- 对于其他编程语言，建议通过服务化部署方式集成。PaddleOCR 支持包括 C++、C#、Java、Go、PHP 等多种语言的客户端调用方式，具体集成方法可参考 [官方文档](https://www.paddleocr.ai/latest/version3.x/pipeline_usage/PP-StructureV3.html#3)。  
+- 如果需要与大模型进行交互，PaddleOCR 还提供了 MCP 服务，详细说明可参考 [MCP 服务器](https://www.paddleocr.ai/latest/version3.x/integrations/mcp_server.html)。
+
+---
+
+**Q:服务化部署可以并发处理请求吗？**
+
+**A:** 对于基础服务化部署方案，服务同一时间只处理一个请求，该方案主要用于快速验证、打通开发链路，或者用在不需要并发请求的场景；对于高稳定性服务化部署方案，服务默认在同一时间只处理一个请求，但用户可以参考服务化部署指南，通过调整配置实现水平扩展，以使服务同时处理多个请求。
+
+---
+
+**Q: 服务化部署如何降低时延、提升吞吐？**
+
+**A:** PaddleOCR 提供的2种服务化部署方案，无论使用哪一种方案，都可以通过启用高性能推理插件提升模型推理速度，从而降低处理时延。此外，对于高稳定性服务化部署方案，通过调整服务配置，设置多个实例，也可以充分利用部署机器的资源，有效提升吞吐。高稳定性服务化部署方案调整配置可以参考[文档](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/serving.html#22)。
diff --git a/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL-1.5.en.md b/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL-1.5.en.md
new file mode 100644
index 0000000..01b0ca5
--- /dev/null
+++ b/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL-1.5.en.md
@@ -0,0 +1,124 @@
+## 1. PaddleOCR-VL-1.5 Introduction
+
+**PaddleOCR-VL-1.5** is an advanced next-generation model of PaddleOCR-VL, achieving a new state-of-the-art accuracy of 94.5% on OmniDocBench v1.5. To rigorously evaluate robustness against real-world physical distortions—including scanning artifacts, skew, warping, screen photography, and illumination—we propose the Real5-OmniDocBench benchmark. Experimental results demonstrate that this enhanced model attains SOTA performance on the newly curated benchmark. Furthermore, we extend the model’s capabilities by incorporating seal recognition and text spotting tasks, while remaining a 0.9B ultra-compact VLM with high efficiency.
+
+### Key Metrics:
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/paddleocr-vl-1.5_metrics.png" width="800"/>
+</div>
+
+
+### **Core Features**
+
+1. With a **parameter size of 0.9B**, PaddleOCR-VL-1.5 **achieves xxx% accuracy on OmniDocBench v1.5**, surpassing the previous SOTA model PaddleOCR-VL. Significant improvements are observed in **table, formula, and text understanding.**
+
+2. **It introduces an innovative approach to document parsing by supporting irregular-shaped localization**, enabling accurate polygonal detection under skewed and curved document conditions. Evaluations across five real-world scenarios—scanning, curving, skewing, screen-photo capture, and light variation—demonstrate superior performance over mainstream open-source and proprietary models.
+
+3. The model introduces **text spotting (text-line localization and recognition)**, along with **seal recognition**, with all corresponding metrics **setting new SOTA results** in their respective tasks.
+
+4. PaddleOCR-VL-1.5 further strengthens its capability in **specialized scenarios and multilingual recognition.** Recognition performance is improved for **rare characters, ancient texts, multilingual tables, underlines, and checkboxes,** and language coverage is extended to include **China's Tibetan script and Bengali.**
+
+5. The model supports **automatic cross-page table merging** and **cross-page paragraph heading recognition**, effectively mitigating content fragmentation issues in **long-document parsing.**
+
+
+## 2. Model Architecture
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/PaddleOCR-VL-1.5.png" width="800"/>
+</div>
+
+
+## 3. Model Performance
+
+### 1. OmniDocBench v1.5
+
+#### PaddleOCR-VL achieves SOTA performance for overall, text, formula, tables and reading order on OmniDocBench v1.5.
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/omnidocbenchv1.5_metrics.png" width="800"/>
+</div>
+
+> **Notes:** 
+> - Performance metrics are cited from the [OmniDocBench official leaderboard](https://opendatalab.com/omnidocbench), except for Gemini-3 Pro, Qwen3-VL-235B-A22B-Instruct and our model, which were evaluated independently.
+
+
+###  2. Real5-OmniDocBench
+
+#### Across all five diverse and challenging scenarios—scanning, warping, screen-photography, illumination, and skew—PaddleOCR-VL-1.5 consistently sets new SOTA records
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/real5-omnidocbench_metrics.png" width="800"/>
+</div>
+
+> **Notes:** 
+> - Real5-OmniDocBench is a brand-new benchmark oriented toward real-world scenarios, which we constructed based on the OmniDocBench v1.5 dataset. The dataset comprises five distinct scenarios: Scanning, Warping, Screen-photography, Illumination, and Skew. For further details, please refer to [Real5-OmniDocBench](https://huggingface.co/datasets/PaddlePaddle/Real5-OmniDocBench).
+
+
+
+## 4、Inference and deployment Performance {#4-inference-and-deployment-performance}
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/inference_performance.png" width="600"/>
+</div>
+
+> **Notes:** 
+> - End-to-End Inference Performance Comparison on OmniDocBench v1.5. PDF documents were processed in batches of 512 on a single NVIDIA A100 GPU. The reported end-to-end runtime includes both PDF rendering and Markdown generation. All methods rely on their built-in PDF parsing modules and default DPI settings to reflect out-of-the-box performance. 
+
+
+## 5. Visualization
+
+### Real-word Document Parsing
+
+
+#### Illumination
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/light.jpg" width="800"/>
+</div>
+
+
+#### Skew
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/skew.jpg" width="800"/>
+</div>
+
+
+#### Screen Photography
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/screen.jpg" width="800"/>
+</div>
+
+
+#### Scanning
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/scaning.jpg" width="800"/>
+</div>
+
+#### Warping
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/curving.jpg" width="800"/>
+</div>
+
+
+### Text Spotting
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/spotting.jpg" width="800"/>
+</div>
+
+
+### Seal Recognition
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/seal.jpg" width="800"/>
+</div>
diff --git a/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL-1.5.md b/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL-1.5.md
new file mode 100644
index 0000000..1f3a0f6
--- /dev/null
+++ b/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL-1.5.md
@@ -0,0 +1,124 @@
+## 1. PaddleOCR-VL-1.5 简介
+
+**PaddleOCR-VL-1.5** 在1.0版本上进行了进一步能力的扩展和升级优化，在文档解析 OmniDocBench v1.5 上取得了 94.5% 的更高的新 SOTA（最佳）结果。为了严格评估其对现实世界物理畸变的鲁棒性——包括扫描伪影、倾斜、弯曲、屏摄和光照变化——我们提出了 Real5-OmniDocBench 基准测试。实验结果表明，该增强模型在这一新构建的基准测试中各个场景都达到了 SOTA 性能。此外，我们通过加入印章识别和文字检测识别任务扩展了模型能力，同时保持了 0.9B 的超紧凑 VLM 规模和高效率。
+
+### **关键指标:**
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/paddleocr-vl-1.5_metrics.png" width="800"/>
+</div>
+
+
+### **核心特性:**
+
+1. **文档解析的SOTA性能：** 凭借 0.9B 的参数量，PaddleOCR-VL-1.5 在 OmniDocBench v1.5 上达到了 94.5% 的准确率，超越了之前的 SOTA 模型 PaddleOCR-VL。在表格、公式和文本识别方面观察到了显著提升。
+
+2. **现实5大场景文档解析的SOTA性能：** 引入了一种创新的文档解析方法，支持不规则形状定位，能够在文档倾斜和弯曲条件下实现精确的多边形检测。在扫描、弯曲、倾斜、屏摄和光照变化这五个现实场景的评估中，表现优于主流的开源和闭源模型。
+
+3. **0.9B紧凑架构扩充能力：** 模型引入了文本行定位与识别 以及 印章识别，所有相关指标均在各自任务中创下了新的 SOTA 结果。
+
+4. **强化多元素识别能力：** PaddleOCR-VL-1.5 进一步增强了在特定场景和多语言识别方面的能力。针对特殊符号、古籍、多语言表格、下划线和复选框的识别性能得到提升，语言覆盖范围扩展至包括中国藏文和孟加拉语。
+
+5. **长文档跨页解析：** 模型支持跨页表格自动合并和跨页段落标题识别，有效缓解了长文档解析中的内容碎片化问题。
+
+
+## 二、技术架构
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/PaddleOCR-VL-1.5.png" width="800"/>
+</div>
+
+
+## 三、 模型性能
+
+### 1. OmniDocBench v1.5
+
+#### PaddleOCR-VL 在 OmniDocBench v1.5 上的整体、文本、公式、表格和阅读顺序中均达到最先进的性能。
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/omnidocbenchv1.5_metrics.png" width="800"/>
+</div>
+
+> **注:** 
+> - 性能指标引自 [OmniDocBench 官方排行榜](https://opendatalab.com/omnidocbench), Gemini-3 Pro、Qwen3-VL-235B-A22B-Instruct 和我们的模型除外。
+
+
+###  2. Real5-OmniDocBench
+
+#### 在扫描、扭曲、屏摄、光照和倾斜这五个多样化且具挑战性的场景中，PaddleOCR-VL-1.5 均创下了新的 SOTA 记录。
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/real5-omnidocbench_metrics.png" width="800"/>
+</div>
+
+> **注:** 
+> - Real5-OmniDocBench 是我们基于 OmniDocBench v1.5 数据集构建的、面向真实场景的全新基准测试。该数据集包含五个不同场景：扫描 (Scanning)、扭曲 (Warping)、屏摄 (Screen-photography)、光照 (Illumination) 和倾斜 (Skew)。更多详情请参阅 [Real5-OmniDocBench](https://huggingface.co/datasets/PaddlePaddle/Real5-OmniDocBench).
+
+
+
+## 4、推理部署性能 {#4推理部署性能}
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/inference_performance.png" width="600"/>
+</div>
+
+> **注:** 
+> - OmniDocBench v1.5 上的端到端推理性能对比。PDF 文档在单张 NVIDIA A100 GPU 上以 512 的 batch size 进行处理。报告的端到端运行时间包含 PDF 渲染和 Markdown 生成。所有方法均依赖其内置的 PDF 解析模块和默认 DPI 设置，以反映开箱即用的性能。 
+
+
+## 5. 可视化
+
+### 现实场景文档
+
+
+#### 光照
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/light.jpg" width="800"/>
+</div>
+
+
+#### 倾斜
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/skew.jpg" width="800"/>
+</div>
+
+
+#### 屏摄
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/screen.jpg" width="800"/>
+</div>
+
+
+#### 扫描
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/scaning.jpg" width="800"/>
+</div>
+
+#### 弯曲/扭曲
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/curving.jpg" width="800"/>
+</div>
+
+
+### 文本定位与识别
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/spotting.jpg" width="800"/>
+</div>
+
+
+### 印章识别
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_5/seal.jpg" width="800"/>
+</div>
diff --git a/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL-1.6.en.md b/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL-1.6.en.md
new file mode 100644
index 0000000..9f39125
--- /dev/null
+++ b/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL-1.6.en.md
@@ -0,0 +1,146 @@
+## 1. Introduction to PaddleOCR-VL-1.6
+
+**PaddleOCR-VL-1.6** further optimizes PaddleOCR-VL-1.5 by systematically analyzing under-optimized areas in the current model, applying targeted data optimization, and adopting refined post-training strategies. It achieves a new state-of-the-art (SOTA) result of 96.33% on the OmniDocBench v1.6 document parsing benchmark. PaddleOCR-VL-1.6 also reaches SOTA performance across all scenarios on Real5-OmniDocBench, a benchmark designed to evaluate robustness against real-world physical distortions. In addition, PaddleOCR-VL-1.6 outperforms PaddleOCR-VL-1.5 on three subtasks: seal recognition, text detection and recognition, and chart recognition, while still maintaining an ultra-compact 0.9B-parameter VLM and high efficiency.
+
+### **Key Metrics:**
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/paddleocr-vl-1.6_metrics.png" width="800"/>
+</div>
+
+
+### **Core Features:**
+
+1. **SOTA performance in document parsing:** With only 0.9B parameters, PaddleOCR-VL-1.6 achieves 96.33% accuracy on OmniDocBench v1.6, surpassing the previous SOTA model, PaddleOCR-VL-1.5. Significant improvements are observed in table, formula, and text recognition.
+
+2. **SOTA performance for document parsing across five real-world scenarios:** PaddleOCR-VL-1.6 offers stronger robustness and practicality in real-world use cases. In evaluations across five real-world distortion scenarios—scanning, warping, skew, screen photography, and illumination variation—it outperforms mainstream open-source and closed-source models.
+
+3. **Enhanced multi-element recognition capabilities:** Beyond improved layout parsing, PaddleOCR-VL-1.6 substantially strengthens recognition of complex tables, ancient books, and rare Chinese characters, while further improving three existing capabilities: chart parsing, seal recognition, and text detection and recognition.
+
+4. **Compact 0.9B architecture:** PaddleOCR-VL-1.6 follows the compact 0.9B architecture of the PaddleOCR-VL series, enabling zero-cost adaptation and drop-in replacement.
+
+
+## 2. Technical Architecture
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/overall.png" width="800"/>
+</div> 
+
+1. **Data engine:** Starting from PaddleOCR-VL-1.5, the data engine systematically identifies under-optimized areas in PaddleOCR-VL-1.5, designs strategies for obtaining high-quality labels, and performs targeted data optimization.
+2. **Progressive post-training strategy:** Data is carefully categorized from three perspectives: quality, difficulty, and improvement value. The training weights of PaddleOCR-VL-1.5 are loaded, and a three-stage post-training strategy—continued pre-training, supervised fine-tuning, and reinforcement learning—is applied according to different data quality levels to steadily improve model performance.
+
+
+## 3. Model Performance
+
+### 1. OmniDocBench v1.6
+
+#### PaddleOCR-VL-1.6 achieves state-of-the-art performance on OmniDocBench v1.6 in overall metrics, text, formulas, and tables. It also delivers leading results in reading order.
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/table2.png" width="800"/>
+</div>
+
+> **Note:** 
+> - Performance metrics are cited from the [official OmniDocBench leaderboard](https://opendatalab.com/omnidocbench).
+
+
+###  2. Real5-OmniDocBench
+
+#### PaddleOCR-VL-1.6 sets new SOTA records across five diverse and challenging scenarios: scanning, warping, screen photography, illumination, and skew.
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/table3.png" width="800"/>
+</div>
+
+> **Note:** 
+> - Real5-OmniDocBench is a new real-world benchmark built by the PaddleOCR team based on the OmniDocBench v1.5 dataset. It contains five scenarios: Scanning, Warping, Screen-photography, Illumination, and Skew. For more details, see [Real5-OmniDocBench](https://huggingface.co/datasets/PaddlePaddle/Real5-OmniDocBench).
+
+
+
+## 4. Inference and Deployment Performance
+
+PaddleOCR-VL-1.6 and PaddleOCR-VL-1.5 use exactly the same model architecture design, so they have identical inference speeds. For details about the inference speed of PaddleOCR-VL-1.5, refer to [PaddleOCR-VL-1.5 inference speed](./PaddleOCR-VL-1.5.md#4-inference-and-deployment-performance).
+
+
+## 5. Visualization
+
+### Comparison with PaddleOCR-VL-1.5
+
+
+#### Ancient Book Recognition
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/ancient1.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/ancient2.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/ancient3.png" width="800"/>
+</div>
+
+
+#### Chart Parsing
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/chart1.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/chart2.png" width="800"/>
+</div>
+
+
+#### Formula Recognition
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/fomula1.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/fomula2.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/fomula3.png" width="800"/>
+</div>
+
+#### Rare Chinese Character Recognition
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/rare-cha1.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/rare-cha2.png" width="800"/>
+</div>
+
+#### Seal Recognition
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/seal1.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/seal2.png" width="800"/>
+</div>
+
+
+### Table Recognition
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/table1.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/table2.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/table3.png" width="800"/>
+</div>
diff --git a/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL-1.6.md b/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL-1.6.md
new file mode 100644
index 0000000..aaf9247
--- /dev/null
+++ b/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL-1.6.md
@@ -0,0 +1,146 @@
+## 1. PaddleOCR-VL-1.6 简介
+
+**PaddleOCR-VL-1.6** 在 PaddleOCR-VL-1.5 的基础上进一步优化，通过系统分析当前模型中的欠优化区域，进行针对性数据优化，并采用精细化的后训练策略，在文档解析基准 OmniDocBench v1.6 上取得 96.33% 的最新 SOTA（最佳）结果。同时，PaddleOCR-VL-1.6 在面向真实世界物理畸变鲁棒性的 Real5-OmniDocBench 基准测试中，也在各个场景下均达到 SOTA 性能。此外，在印章识别、文字检测识别和图表识别三个子任务上，PaddleOCR-VL-1.6 均领先 PaddleOCR-VL-1.5，同时依然保持 0.9B 超紧凑 VLM 参数量和高效率。
+
+### **关键指标:**
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/paddleocr-vl-1.6_metrics.png" width="800"/>
+</div>
+
+
+### **核心特性:**
+
+1. **文档解析的SOTA性能：** 凭借 0.9B 的参数量，PaddleOCR-VL-1.6 在 OmniDocBench v1.6 上达到了 96.33% 的准确率，超越了之前的 SOTA 模型 PaddleOCR-VL-1.5。在表格、公式和文本识别方面观察到了显著提升。
+
+2. **现实5大场景文档解析的SOTA性能：** 具备更强的鲁棒性和真实场景的实用性。在扫描、弯曲、倾斜、屏摄和光照变化这五个真实扰动场景的评估中，表现优于主流的开源和闭源模型。
+
+3. **强化多元素识别能力：** 除了版面解析能力的提升外，PaddleOCR-VL-1.6 进一步大幅度增强了对复杂表格，古籍 和 生僻字的识别能力，同时在 图表解析，印章识别，文字检测识别这三个原有能力上进一步提升。
+
+4. **0.9B紧凑架构：** 沿用 PaddleOCR-VL系列的 0.9B 紧凑构架，零成本适配，即换即用。
+
+
+## 二、技术架构
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/overall.png" width="800"/>
+</div> 
+
+1. **数据引擎:** 以 PaddleOCR-VL-1.5 为出发点，系统性定位 PaddleOCR-VL-1.5 的欠优化区域，并设计高质量标签获取策略，针对性的进行数据优化。
+2. **渐进式后训练策略** 从质量，难度，提升价值三个角度精细化划分数据，加载 PaddleOCR-VL-1.5 训练权重，结合不同的数据质量进行 继续预训练，监督微调，强化学习 三阶段的后训练策略，稳步提升模型性能。
+
+
+## 三、 模型性能
+
+### 1. OmniDocBench v1.6
+
+#### PaddleOCR-VL-1.6 在 OmniDocBench v1.6 上的整体指标、文本、公式、表格均达到最先进的性能，其中在在阅读顺序方面，PaddleOCR-VL-1.6也取得了较为领先的指标。
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/table2.png" width="800"/>
+</div>
+
+> **注:** 
+> - 性能指标引自 [OmniDocBench 官方排行榜](https://opendatalab.com/omnidocbench)。
+
+
+###  2. Real5-OmniDocBench
+
+#### 在扫描、扭曲、屏摄、光照和倾斜这五个多样化且具挑战性的场景中，PaddleOCR-VL-1.6 均创下了新的 SOTA 记录。
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/table3.png" width="800"/>
+</div>
+
+> **注:** 
+> - Real5-OmniDocBench 是 PaddleOCR团队 基于 OmniDocBench v1.5 数据集构建的、面向真实场景的全新基准测试。该数据集包含五个不同场景：扫描 (Scanning)、扭曲 (Warping)、屏摄 (Screen-photography)、光照 (Illumination) 和倾斜 (Skew)。更多详情请参阅 [Real5-OmniDocBench](https://huggingface.co/datasets/PaddlePaddle/Real5-OmniDocBench).
+
+
+
+## 四、推理部署性能
+
+PaddleOCR-VL-1.6 和 PaddleOCR-VL-1.5 采用完全相同的模型架构设计，因此有完全相同的推理速度。关于PaddleOCR-VL-1.5推理速度的说明可以参考 [PaddleOCR-VL-1.5推理速度](./PaddleOCR-VL-1.5.md#4推理部署性能) 。
+
+
+## 5. 可视化
+
+### 和 PaddleOCR-VL-1.5 的对比
+
+
+#### 古籍识别
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/ancient1.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/ancient2.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/ancient3.png" width="800"/>
+</div>
+
+
+#### 图表解析
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/chart1.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/chart2.png" width="800"/>
+</div>
+
+
+#### 公式识别
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/fomula1.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/fomula2.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/fomula3.png" width="800"/>
+</div>
+
+#### 生僻字识别
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/rare-cha1.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/rare-cha2.png" width="800"/>
+</div>
+
+#### 印章识别
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/seal1.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/seal2.png" width="800"/>
+</div>
+
+
+### 表格识别
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/table1.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/table2.png" width="800"/>
+</div>
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl_1_6/comparison/table3.png" width="800"/>
+</div>
diff --git a/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL.en.md b/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL.en.md
new file mode 100644
index 0000000..6151613
--- /dev/null
+++ b/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL.en.md
@@ -0,0 +1,196 @@
+## 1. PaddleOCR-VL Introduction
+
+**PaddleOCR-VL** is a SOTA and resource-efficient model tailored for document parsing. Its core component is PaddleOCR-VL-0.9B, a compact yet powerful vision-language model (VLM) that integrates a NaViT-style dynamic resolution visual encoder with the ERNIE-4.5-0.3B language model to enable accurate element recognition. This innovative model efficiently supports 109 languages and excels in recognizing complex elements (e.g., text, tables, formulas, and charts), while maintaining minimal resource consumption. Through comprehensive evaluations on widely used public benchmarks and in-house benchmarks, PaddleOCR-VL achieves SOTA performance in both page-level document parsing and element-level recognition. It significantly outperforms existing solutions, exhibits strong competitiveness against top-tier VLMs, and delivers fast inference speeds. These strengths make it highly suitable for practical deployment in real-world scenarios.
+
+### Key Metrics:
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/allmetric.png" width="800"/>
+</div>
+
+
+### **Core Features**
+
+1. **Compact yet Powerful VLM Architecture:** We present a novel vision-language model that is specifically designed for resource-efficient inference, achieving outstanding performance in element recognition. By integrating a NaViT-style dynamic high-resolution visual encoder with the lightweight ERNIE-4.5-0.3B language model, we significantly enhance the model’s recognition capabilities and decoding efficiency. This integration maintains high accuracy while reducing computational demands, making it well-suited for efficient and practical document processing applications.
+
+
+2. **SOTA Performance on Document Parsing:** PaddleOCR-VL achieves state-of-the-art performance in both page-level document parsing and element-level recognition. It significantly outperforms existing pipeline-based solutions and exhibiting strong competitiveness against leading vision-language models (VLMs) in document parsing. Moreover, it excels in recognizing complex document elements, such as text, tables, formulas, and charts, making it suitable for a wide range of challenging content types, including handwritten text and historical documents. This makes it highly versatile and suitable for a wide range of document types and scenarios.
+
+
+3. **Multilingual Support:** PaddleOCR-VL Supports 109 languages, covering major global languages, including but not limited to Chinese, English, Japanese, Latin, and Korean, as well as languages with different scripts and structures, such as Russian (Cyrillic script), Arabic, Hindi (Devanagari script), and Thai. This broad language coverage substantially enhances the applicability of our system to multilingual and globalized document processing scenarios.
+
+
+## 2. Model Architecture
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/methods/paddleocrvl.png" width="800"/>
+</div>
+
+
+## 3. Model Performance
+
+### Page-Level Document Parsing 
+
+
+#### OmniDocBench v1.5
+
+##### PaddleOCR-VL achieves SOTA performance for overall, text, formula, tables and reading order on OmniDocBench v1.5.
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/omni15.png" width="800"/>
+</div>
+
+
+
+####  OmniDocBench v1.0
+
+##### PaddleOCR-VL achieves SOTA performance for almost all metrics of overall, text, formula, tables and reading order on OmniDocBench v1.0.
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/omni10.png" width="800"/>
+</div>
+
+
+### Element-level Recognition  
+
+#### Text
+
+**Comparison of OmniDocBench-OCR-block Performance**
+
+PaddleOCR-VL’s robust and versatile capability in handling diverse document types, establishing it as the leading method in the OmniDocBench-OCR-block performance evaluation. 
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/omnibenchocr.png" width="800"/>
+</div>
+
+**Comparison of In-house-OCR-block Performance**
+
+In-house-OCR provides a evaluation of performance across multiple languages and text types. Our model demonstrates outstanding accuracy with the lowest edit distances in all evaluated scripts.
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/inhouseocr.png" width="800"/>
+</div>
+
+
+#### Table
+
+
+**Comparison of In-house-Table Performance**
+
+Our self-built evaluation set contains diverse types of table images, such as Chinese, English, mixed Chinese-English, and tables with various characteristics like full, partial, or no borders, book/manual formats, lists, academic papers, merged cells, as well as low-quality, watermarked, etc. PaddleOCR-VL achieves remarkable performance across all categories.
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/inhousetable.png" width="600"/>
+</div>
+
+
+#### Formula
+
+**Comparison of In-house-Formula Performance**
+
+
+In-house-Formula evaluation set contains simple prints, complex prints, camera scans, and handwritten formulas. PaddleOCR-VL demonstrates the best performance in every category.
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/inhouse-formula.png" width="500"/>
+</div>
+
+#### Chart
+
+**Comparison of In-house-Chart Performance**
+
+The evaluation set is broadly categorized into 11 chart categories, including bar-line hybrid, pie, 100% stacked bar, area, bar, bubble, histogram, line, scatterplot, stacked area, and stacked bar. PaddleOCR-VL not only outperforms expert OCR VLMs but also surpasses some 72B-level multimodal language models.
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/inhousechart.png" width="400"/>
+</div>
+
+
+
+## 4、Inference and deployment Performance
+
+To improve the inference performance of PaddleOCR-VL, we introduce multi-threading asynchronous execution into the inference workflow. The process is divided into three main stages—data loading (e.g., rendering PDF pages as images), layout model processing, and VLM inference—each running in a separate thread. Data is transferred between adjacent stages via queues, enabling concurrent execution for higher efficiency. We measured the end-to-end inference speed and GPU usage on the OmniDocBench v1.0 dataset, processing PDF files in batches of 512 on a single NVIDIA A100 GPU. 
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/inference.png" width="600"/>
+</div>
+
+> **Notes:**  <sup>&#8224;</sup> means vLLM backend，<sup>&#8225;</sup> means sglang backend.
+
+
+## 5. Visualization
+
+### Comprehensive Document Parsing
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/overview1.jpg" width="600"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/overview2.jpg" width="600"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/overview3.jpg" width="600"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/overview4.jpg" width="600"/>
+</div>
+
+
+
+### Text Recognition
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/text_english_arabic.jpg" width="300"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/text_handwriting_02.jpg" width="300"/>
+</div>
+
+
+### Table Recognition
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/table_01.jpg" width="300"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/table_02.jpg" width="300"/>
+</div>
+
+### Formula Recognition
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/formula_EN.jpg" width="300"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/formula_EN.jpg" width="300"/>
+</div>
+
+
+### Chart Recognition
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/chart_01.jpg" width="300"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/chart_02.jpg" width="300"/>
+</div>
+
+
+## 6. FAQ
+
+**1.How to use PaddleOCR-VL for document parsing?**
+
+Please refer to our usage documentation [PaddleOCR-VL Usage](../../pipeline_usage/PaddleOCR-VL.en.md).
+
+**2. How to fine-tune the PaddleOCR-VL model?**
+
+Currently, we do not support fine-tuning of the model, but it is a high-priority feature and will be released soon. Please stay tuned.
+
+**3. Why was my chart not recognized and how can I use chart recognition?**
+
+Because our default chart recognition function is turned off, it needs to be manually turned on. Please refer to [PaddleOCR-VL Usage](../../pipeline_usage/PaddleOCR-VL.en.md) and set the use_chart_recognition为True parameters to True turn it on.
+
+**4. What are the 109 supported languages?**
+
+Chinese, English, Korea, Japanese, Thai, Greek, Tamil, Telugu
+
+Arabic: Arabic, Persian, Uyghur, Urdu, Pashto, Kurdish, Sindhi, Balochi 
+
+Latin: French, German, Afrikaans, Italian, Spanish, Bosnian, Portuguese, Czech, Welsh, Danish, Estonian, Irish, Croatian, Uzbek, Hungarian, Serbian (Latin), Indonesian, Occitan, Icelandic, Lithuanian, Maori, Malay, Dutch, Norwegian, Polish, Slovak, Slovenian, Albanian, Swedish, Swahili, Tagalog, Turkish, Latin, Azerbaijani, Kurdish, Latvian, Maltese, Pali, Romanian, Vietnamese, Finnish, Basque, Galician, Luxembourgish, Romansh, Catalan, Quechua 
+
+Cyrillic: Russian, Belarusian, Ukrainian, Serbian (Cyrillic), Bulgarian, Mongolian, Abkhazian, Adyghe, Kabardian, Avar, Dargin, Ingush, Chechen, Lak, Lezgin, Tabasaran, Kazakh, Kyrgyz, Tajik, Macedonian, Tatar, Chuvash, Bashkir, Malian, Moldovan, Udmurt, Komi, Ossetian, Buryat, Kalmyk, Tuvan, Sakha, Karakalpak 
+
+Devanagari: Hindi, Marathi, Nepali, Bihari, Maithili, Angika, Bhojpuri, Magahi, Santali, Newari, Konkani, Sanskrit, Haryanvi 
+
+**5. If the results of the layout check are not satisfactory, what solutions can be optimized?**
+
+Since layout detection is mainly trained for various document scenarios, if your test data is non-standard documents such as license plates, tickets images or ID Cards and you want to do OCR recognition, you can directly use the PaddleOCR-VL-0.9B model and turn off the layout detection model by setting use_layout_detection to False. If you find any layout detection errors, you can directly try the effect of using PaddleOCR-VL-0.9B alone. 
+We recommend using the [ERNIEKit toolkit](https://github.com/PaddlePaddle/ERNIE/tree/release/v1.4) to perform Supervised Fine-Tuning (SFT) on the PaddleOCR-VL-0.9B model. For detailed steps, please refer to the [ERNIEKit documentation](https://github.com/PaddlePaddle/ERNIE/blob/release/v1.4/docs/paddleocr_vl_sft.md).
diff --git a/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL.md b/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL.md
new file mode 100644
index 0000000..18dd5c7
--- /dev/null
+++ b/docs/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL.md
@@ -0,0 +1,200 @@
+## 一、PaddleOCR-VL简介
+
+**PaddleOCR-VL** 是一款先进、高效的文档解析模型，专为文档中的元素识别设计。其核心组件为 PaddleOCR-VL-0.9B，这是一种紧凑而强大的视觉语言模型（VLM），它由 NaViT 风格的动态分辨率视觉编码器与 ERNIE-4.5-0.3B 语言模型组成，能够实现精准的元素识别。该模型支持 109 种语言，并在识别复杂元素（如文本、表格、公式和图表）方面表现出色，同时保持极低的资源消耗。通过在广泛使用的公开基准与内部基准上的全面评测，PaddleOCR-VL 在页级级文档解析与元素级识别均达到 SOTA 表现。它显著优于现有的基于Pipeline方案和文档解析多模态方案以及先进的通用多模态大模型，并具备更快的推理速度。这些优势使其非常适合在真实场景中落地部署。
+
+### **关键指标:**
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/allmetric.png" width="800"/>
+</div>
+
+### **核心特性:**
+
+1. **紧凑而强大的视觉语言模型架构：** 我们提出了一种新的视觉语言模型，专为资源高效的推理而设计，在元素识别方面表现出色。通过将NaViT风格的动态高分辨率视觉编码器与轻量级的ERNIE-4.5-0.3B语言模型结合，我们显著增强了模型的识别能力和解码效率。这种集成在保持高准确率的同时降低了计算需求，使其非常适合高效且实用的文档处理应用。
+
+2. **文档解析的SOTA性能：** PaddleOCR-VL在页面级文档解析和元素级识别中达到了最先进的性能。它显著优于现有的基于流水线的解决方案，并在文档解析中展现出与领先的视觉语言模型（VLMs）竞争的强劲实力。此外，它在识别复杂的文档元素（如文本、表格、公式和图表）方面表现出色，使其适用于包括手写文本和历史文献在内的各种具有挑战性的内容类型。这使得它具有高度的多功能性，适用于广泛的文档类型和场景。
+
+3. **多语言支持：** PaddleOCR-VL支持109种语言，覆盖了主要的全球语言，包括但不限于中文、英文、日文、拉丁文和韩文，以及使用不同文字和结构的语言，如俄语（西里尔字母）、阿拉伯语、印地语（天城文）和泰语。这种广泛的语言覆盖大大增强了我们系统在多语言和全球化文档处理场景中的适用性。
+
+
+
+
+## 二、技术架构
+
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/methods/paddleocrvl.png" width="800"/>
+</div>
+
+
+
+
+
+## 三、 模型性能
+
+### 页面级文档解析
+
+#### 1. OmniDocBench v1.5
+
+##### PaddleOCR-VL 在 OmniDocBench v1.5 上的整体、文本、公式、表格和阅读顺序中均达到最先进的性能。
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/omni15.png" width="800"/>
+</div>
+
+####  2. OmniDocBench v1.0
+
+##### PaddleOCR-VL 在 OmniDocBench v1.0 的整体、文本、公式、表格以及阅读顺序等几乎所有评估指标上均达到了 SOTA 性能。
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/omni10.png" width="800"/>
+</div>
+
+### 元素级识别
+
+#### 文本
+
+**OmniDocBench-OCR-block**
+
+PaddleOCR-VL 在处理多样化文档类型方面展现出强大而灵活的能力，使其在 OmniDocBench-OCR-block 的性能评估中成为领先方法。
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/omnibenchocr.png" width="800"/>
+</div>
+
+
+**In-house-OCR-block**
+
+我们自建的评测集评估了模型在多语言和多文本类型下的性能。我们的模型在所有评测文字体系中均表现出卓越的准确性，并取得了最低的编辑距离。
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/inhouseocr.png" width="800"/>
+</div>
+
+
+
+#### 表格
+
+**In-house-Table**
+
+我们自建的评测集包含多种类型的表格图像，例如中文、英文、中英混合表格，以及具有不同特征的表格类型，如完整边框、部分边框、无边框、书籍/手册格式、列表、学术论文表格、合并单元格等，还包括低质量和带水印的样本。PaddleOCR-VL 在所有类别中均展现出卓越的性能。
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/inhousetable.png" width="600"/>
+</div>
+
+
+#### 公式
+
+**In-house-Formula**
+
+我们自建的评测集包含简单印刷、复杂印刷、摄像扫描以及手写公式等多种类型。PaddleOCR-VL 在所有类别中均取得了最佳性能。
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/inhouse-formula.png" width="500"/>
+</div>
+
+#### 图表
+
+**In-house-Chart 结果比较**
+
+我们自建的评测集涵盖 11 种主要图表类型，包括柱线混合图、饼图、100% 堆叠柱状图、面积图、柱状图、气泡图、直方图、折线图、散点图、堆叠面积图和堆叠柱状图。PaddleOCR-VL 不仅优于专业 OCR VLM 模型，还超越了一些 72B 级别的多模态语言模型。
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/inhousechart.png" width="400"/>
+</div>
+
+
+## 四、推理部署性能
+
+为了提升PaddleOCR-VL的推理性能，我们在推理工作流程中引入了多线程异步执行。该过程分为三个主要阶段：数据加载（例如，将PDF页面渲染为图像）、布局模型处理和VLM推理——每个阶段都在一个单独的线程中运行。数据通过队列在相邻阶段之间传输，从而实现并发执行以提高效率。在OmniDocBench v1.0数据集上测量了端到端推理速度和GPU使用情况，以512个PDF文件的批次在单个NVIDIA A100 GPU上进行处理。
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/metrics/inference.png" width="600"/>
+</div>
+
+> **Notes:**  <sup>&#8224;</sup> 表示vLLM后端，<sup>&#8225;</sup> 表示SGLang后端
+
+
+
+## 五、可视化
+
+PaddleOCR-VL能够支持多种类型的文档解析，以下是一些预测案例的展示：
+
+### 端到端文档解析
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/overview1.jpg" width="600"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/overview2.jpg" width="600"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/overview3.jpg" width="600"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/overview4.jpg" width="600"/>
+</div>
+
+
+
+### 文本识别
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/text_english_arabic.jpg" width="300"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/text_handwriting_02.jpg" width="300"/>
+</div>
+
+
+### 表格识别
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/table_01.jpg" width="300"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/table_02.jpg" width="300"/>
+</div>
+
+### 公式识别
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/formula_EN.jpg" width="300"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/formula_EN.jpg" width="300"/>
+</div>
+
+### 图表识别
+
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/chart_01.jpg" width="300"/>
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/chart_02.jpg" width="300"/>
+</div>
+
+
+## 六、FAQ
+
+**1.如何使用 PaddleOCR-VL 做文档解析 ？**
+
+参考我们的使用文档 [PaddleOCR-VL使用](../../pipeline_usage/PaddleOCR-VL.md)
+
+**2.如何对 PaddleOCR-VL 模型进行微调 ？**
+
+目前我们暂不支持模型的微调，但已经在高优的支持中，即将发布，请保持关注。
+
+**3.为什么我的图表没有识别出来，如何使用图表识别 ？**
+
+因为我们默认图表识别的功能是关闭的，需要手动开启，请参考 [PaddleOCR-VL使用](../../pipeline_usage/PaddleOCR-VL.md), 设置 use_chart_recognition为True 参数来开启。
+
+**4.支持的109种语言有哪些？**
+
+中文、英语、韩语、日语、泰语、希腊语、泰米尔语、泰卢固语
+
+阿拉伯语：阿拉伯语、波斯语、维吾尔语、乌尔都语、普什图语、库尔德语、信德语、俾路支语
+
+拉丁语：法语、德语、南非荷兰语、意大利语、西班牙语、波斯尼亚语、葡萄牙语、捷克语、威尔士语、丹麦语、爱沙尼亚语、爱尔兰语、克罗地亚语、乌兹别克语、匈牙利语、塞尔维亚语（拉丁语）、印度尼西亚语、奥克语、冰岛语、立陶宛语、毛利语、马来语、荷兰语、挪威语、波兰语、斯洛伐克语、斯洛文尼亚语、阿尔巴尼亚语、瑞典语、斯瓦希里语、他加禄语、土耳其语、拉丁语、阿塞拜疆语、库尔德语、拉脱维亚语、马耳他语、巴利语、罗马尼亚语、越南语、芬兰语、巴斯克语、加利西亚语、卢森堡语、罗曼什语、加泰罗尼亚语、盖丘亚语
+
+西里尔文：俄语、白俄罗斯语、乌克兰语、塞尔维亚语（西里尔文）、保加利亚语、蒙古语、阿布哈兹语、阿迪杰语、卡巴尔达语、阿瓦尔语、达尔金语、印古什语、车臣语、拉克语、列兹金语、塔巴萨兰语、哈萨克语、吉尔吉斯语、塔吉克语、马其顿语、鞑靼语、楚瓦什语、巴什基尔语、马里语、摩尔多瓦语、乌德穆尔特语、科米语、奥塞梯语、布里亚特语、卡尔梅克语、图瓦语、萨哈语、卡拉卡尔帕克语
+
+天城语：印地语、马拉地语、尼泊尔语、比哈里语、迈蒂利语、安吉卡语、博杰普里语、马基语、桑塔利语、纽瓦里语、康卡尼语、梵语、哈里亚维语
+
+**5.如果版面检测的结果不理想，有什么方案可以优化？**
+
+由于版面检测主要针对各种文档场景训练，所以您的测试数据如果是非标准文档，如车牌，火车票或身份证图像想做OCR识别，那可以直接使用PaddleOCR-VL-0.9B的模型，通过设置 use_layout_detection 为 False 关闭版面检测模型。如果您发现有任何版面检测的错误，都可以直接尝试一下单独使用PaddleOCR-VL-0.9B的效果。
+我们推荐使用 [ERNIEKit 套件](https://github.com/PaddlePaddle/ERNIE/tree/release/v1.4) 对 PaddleOCR-VL-0.9B 模型进行有监督微调（SFT）。具体操作步骤可参考 [ERNIEKit 官方文档](https://github.com/PaddlePaddle/ERNIE/blob/release/v1.4/docs/paddleocr_vl_sft_zh.md)。
diff --git a/docs/version3.x/doc2md.en.md b/docs/version3.x/doc2md.en.md
new file mode 100644
index 0000000..d3fe4e9
--- /dev/null
+++ b/docs/version3.x/doc2md.en.md
@@ -0,0 +1,481 @@
+---
+comments: true
+---
+
+# Document to Markdown Tutorial
+
+## 1. Introduction
+
+Document to Markdown (doc2md) is a lightweight Office document conversion feature built into PaddleOCR. It **requires no OCR inference** — it directly parses document structure and outputs well-formed Markdown text. It is well-suited for scenarios where Word, Excel, or PowerPoint files need to be quickly converted to readable text, such as knowledge base construction, document retrieval, and content extraction.
+
+**Supported formats**: `.docx` (Word) / `.xlsx` (Excel) / `.pptx` (PowerPoint)
+
+**Feature overview:**
+
+<table>
+<thead>
+<tr>
+<th>Feature</th>
+<th>Word (.docx)</th>
+<th>Excel (.xlsx)</th>
+<th>PowerPoint (.pptx)</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>Heading levels</td>
+<td>✅ Built-in styles + font-size heuristic + Chinese numbering</td>
+<td>— </td>
+<td>— </td>
+</tr>
+<tr>
+<td>Text formatting (bold / italic / underline / strikethrough)</td>
+<td>✅</td>
+<td>✅</td>
+<td>✅</td>
+</tr>
+<tr>
+<td>Superscript / subscript</td>
+<td>✅</td>
+<td>✅</td>
+<td>✅</td>
+</tr>
+<tr>
+<td>Hyperlinks</td>
+<td>✅</td>
+<td>✅</td>
+<td>✅</td>
+</tr>
+<tr>
+<td>Lists (ordered / unordered / nested)</td>
+<td>✅</td>
+<td>— </td>
+<td>— </td>
+</tr>
+<tr>
+<td>Tables (with merged cells)</td>
+<td>✅ HTML table</td>
+<td>✅ HTML table</td>
+<td>✅ HTML table</td>
+</tr>
+<tr>
+<td>Images</td>
+<td>✅ Proportional width</td>
+<td>✅ Floating images</td>
+<td>✅ Proportional width</td>
+</tr>
+<tr>
+<td>Math formulas (OMML → LaTeX)</td>
+<td>✅ Inline / display formulas</td>
+<td>✅ Drawing-layer formulas</td>
+<td>✅</td>
+</tr>
+<tr>
+<td>Code blocks</td>
+<td>✅ Monospace font auto-detection</td>
+<td>— </td>
+<td>— </td>
+</tr>
+<tr>
+<td>Text boxes</td>
+<td>✅</td>
+<td>— </td>
+<td>— </td>
+</tr>
+<tr>
+<td>Charts</td>
+<td>✅ → HTML table</td>
+<td>— </td>
+<td>✅ 14 chart types</td>
+</tr>
+<tr>
+<td>Headers / footers</td>
+<td>✅ Multi-section + odd/even pages</td>
+<td>— </td>
+<td>— </td>
+</tr>
+<tr>
+<td>Multiple sheets / slides</td>
+<td>— </td>
+<td>✅</td>
+<td>✅ Separated by <code>---</code></td>
+</tr>
+<tr>
+<td>Speaker notes</td>
+<td>— </td>
+<td>— </td>
+<td>✅</td>
+</tr>
+</tbody>
+</table>
+
+## 2. Quick Start
+
+Before using doc2md, make sure you have completed the PaddleOCR base installation as described in the [installation guide](./installation.en.md), then install the doc2md optional dependencies:
+
+```bash
+pip install "paddleocr[doc2md]"
+```
+
+**Dependency details:**
+
+<table>
+<thead>
+<tr>
+<th>Package</th>
+<th>Version constraint</th>
+<th>Purpose</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>python-docx</code></td>
+<td><code>&gt;=0.8.11</code></td>
+<td>Word (.docx) document parsing</td>
+</tr>
+<tr>
+<td><code>python-pptx</code></td>
+<td><code>&gt;=0.6.21</code></td>
+<td>PowerPoint (.pptx) document parsing</td>
+</tr>
+<tr>
+<td><code>openpyxl</code></td>
+<td><code>&gt;=3.0.0</code></td>
+<td>Excel (.xlsx) document parsing</td>
+</tr>
+<tr>
+<td><code>pylatexenc</code></td>
+<td><code>&gt;=2.10,&lt;3</code></td>
+<td>Math formula Unicode → LaTeX symbol mapping</td>
+</tr>
+</tbody>
+</table>
+
+### 2.1 Command Line
+
+```bash
+# Convert a Word document to a file
+paddleocr doc2md -i report.docx -o output.md
+
+# Convert an Excel spreadsheet to a file
+paddleocr doc2md -i data.xlsx -o output.md
+
+# Convert a PowerPoint presentation to a file
+paddleocr doc2md -i slides.pptx -o output.md
+
+# Omit output path to print to stdout
+paddleocr doc2md -i report.docx
+
+# List supported formats and exit
+paddleocr doc2md --formats
+```
+
+<details><summary><b>Click to expand the full list of command line parameters</b></summary>
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>-i</code>, <code>--input</code></td>
+<td><b>Meaning:</b> Input file path. Required (may be omitted when <code>--formats</code> is used).<br/>
+<b>Note:</b> Supports <code>.docx</code>, <code>.xlsx</code>, and <code>.pptx</code> formats.</td>
+<td><code>str</code></td>
+<td>required</td>
+</tr>
+<tr>
+<td><code>-o</code>, <code>--output</code></td>
+<td><b>Meaning:</b> Output Markdown file path.<br/>
+<b>Note:</b> If not set, the result is printed to stdout. When set, Markdown is written to the specified file and images are saved to an <code>images/</code> subdirectory alongside it.</td>
+<td><code>str</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>-q</code>, <code>--quiet</code></td>
+<td><b>Meaning:</b> Quiet mode.<br/>
+<b>Note:</b> When enabled, timing and save-path messages are suppressed.</td>
+<td>flag</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>--formats</code></td>
+<td><b>Meaning:</b> Print the list of supported file formats and exit.<br/>
+<b>Note:</b> <code>--input</code> is not required when this flag is used.</td>
+<td>flag</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>--no-drawings</code></td>
+<td><b>Meaning:</b> Skip text box and drawing-layer content extraction.<br/>
+<b>Note:</b> Applies to <code>.docx</code> and <code>.xlsx</code>. When enabled, text boxes (docx) and drawing-layer math formulas (xlsx) are not extracted.</td>
+<td>flag</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>--no-headers-footers</code></td>
+<td><b>Meaning:</b> Skip header and footer extraction.<br/>
+<b>Note:</b> Applies to <code>.docx</code> only.</td>
+<td>flag</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>--sheet-name</code></td>
+<td><b>Meaning:</b> Convert only the sheet with the given name.<br/>
+<b>Note:</b> Applies to <code>.xlsx</code> only. All sheets are converted when not set.</td>
+<td><code>str</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>--max-rows</code></td>
+<td><b>Meaning:</b> Maximum number of rows to convert per sheet.<br/>
+<b>Note:</b> Applies to <code>.xlsx</code> only. Use this to limit output for large spreadsheets.</td>
+<td><code>int</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+</details>
+<br />
+
+### 2.2 Python API
+
+**Basic usage:**
+
+```python
+from paddleocr._doc2md import convert
+
+# Convert a document and get the result object
+result = convert("report.docx")
+
+# Access the Markdown text
+print(result.markdown)
+
+# List extracted images (dict mapping relative path → bytes)
+print(list(result.images.keys()))
+
+# Get the document title
+print(result.title)
+
+# Get metadata (format, sheet count, etc.)
+print(result.metadata)
+```
+
+**`ConvertResult` fields:**
+
+<table>
+<thead>
+<tr>
+<th>Field</th>
+<th>Type</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>markdown</code></td>
+<td><code>str</code></td>
+<td>The converted Markdown text</td>
+</tr>
+<tr>
+<td><code>images</code></td>
+<td><code>dict[str, bytes]</code></td>
+<td>Extracted images dict; key is a relative path (e.g. <code>images/image1.png</code>), value is raw bytes</td>
+</tr>
+<tr>
+<td><code>title</code></td>
+<td><code>Optional[str]</code></td>
+<td>Document title; may be <code>None</code></td>
+</tr>
+<tr>
+<td><code>metadata</code></td>
+<td><code>dict</code></td>
+<td>Document metadata such as format type and sheet count</td>
+</tr>
+</tbody>
+</table>
+
+**Writing output to a file (images saved automatically):**
+
+```python
+from paddleocr._doc2md import convert
+
+# When output is specified, Markdown is written to the file
+# and images are saved to images/ in the same directory
+result = convert("report.docx", output="output/report.md")
+```
+
+**Available kwargs by format:**
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Type</th>
+<th>Default</th>
+<th>Formats</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>extract_drawings</code></td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+<td>docx, xlsx</td>
+<td>Whether to extract text boxes (docx) / drawing-layer math formulas (xlsx)</td>
+</tr>
+<tr>
+<td><code>extract_headers_footers</code></td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+<td>docx</td>
+<td>Whether to extract headers and footers</td>
+</tr>
+<tr>
+<td><code>sheet_name</code></td>
+<td><code>Optional[str]</code></td>
+<td><code>None</code></td>
+<td>xlsx</td>
+<td>Convert only the named sheet; <code>None</code> converts all sheets</td>
+</tr>
+<tr>
+<td><code>max_rows</code></td>
+<td><code>Optional[int]</code></td>
+<td><code>None</code></td>
+<td>xlsx</td>
+<td>Maximum number of rows to convert per sheet</td>
+</tr>
+</tbody>
+</table>
+
+**Passing kwargs by format:**
+
+```python
+from paddleocr._doc2md import convert
+
+# Word: skip text boxes and headers/footers
+result = convert("report.docx", extract_drawings=False, extract_headers_footers=False)
+
+# Excel: convert only "Sheet1", up to 100 rows
+result = convert("data.xlsx", sheet_name="Sheet1", max_rows=100)
+```
+
+## 3. Supported Features by Format
+
+### 3.1 Word (.docx)
+
+**Heading detection** uses three strategies:
+- **Built-in Heading styles**: Word's built-in Heading 1–6 styles map directly to `#`–`######`
+- **Font-size heuristic**: paragraphs with a font size more than 1.5× the body text and short length are promoted to headings
+- **Chinese numbering**: "一、" patterns are treated as H2; "（一）" patterns as H3
+
+**Text formatting**: bold (`**`), italic (`*`), underline (`<u>`), strikethrough (`~~`), superscript (`<sup>`), subscript (`<sub>`)
+
+**Lists**: ordered, unordered, and nested lists; indentation level is detected automatically
+
+**Tables**: output as HTML `<table>` with `rowspan`/`colspan` for merged cells
+
+**Images**: width is calculated as a percentage of the content area; output as `<img width="75%">`
+
+**Math formulas**: OMML formulas are converted to LaTeX; inline formulas use `$...$` and display formulas use `$$...$$`
+
+**Code blocks**: monospace fonts (Courier New, Consolas, and 7 others) are detected automatically and output as fenced code blocks (` ``` `)
+
+**Other**: text boxes (`wps:txbx`), charts (Chart → HTML table), hyperlinks (two formats supported), headers and footers (multi-section + odd/even pages)
+
+### 3.2 Excel (.xlsx)
+
+**Multiple sheets**: each sheet is output as a section beginning with `## sheet_name`
+
+**Data boundary trimming**: trailing empty rows and columns are removed; only the effective data range is output
+
+**Merged cells**: cell merges are reproduced using `rowspan`/`colspan`
+
+**Text formatting**: bold, italic, underline, strikethrough, superscript, subscript
+
+**Hyperlinks**: cell-level hyperlinks are supported
+
+**Floating images**: both `OneCellAnchor` and `TwoCellAnchor` anchor types are supported
+
+**Math formulas**: OMML formulas are extracted from the drawing-layer XML and converted to LaTeX
+
+### 3.3 PowerPoint (.pptx)
+
+**Multiple slides**: slide content is separated by `---`
+
+**Text formatting**: bold, italic, underline, strikethrough, superscript, subscript
+
+**Images**: width is calculated as a percentage of the slide width; output as `<img>` tags with a width attribute
+
+**Tables**: HTML `<table>` format with merged cell support; tables with background images are handled
+
+**Charts**: 14 chart types are supported; all are converted to HTML table output
+
+**Group shapes**: nested shape groups are processed recursively
+
+**Math formulas**: OMML formulas are extracted from `mc:AlternateContent` and converted to LaTeX
+
+**Speaker notes**: appended at the end of each slide's content
+
+## 4. FAQ
+
+**Q: I get `RuntimeError: python-docx is required` during conversion.**
+
+doc2md uses lazy imports — if a required parsing library is missing, this error is raised. Install the dependency for the format you need:
+```bash
+pip install python-docx     # Word (.docx)
+pip install python-pptx     # PowerPoint (.pptx)
+pip install openpyxl        # Excel (.xlsx)
+pip install pylatexenc      # Math formula support
+```
+Or install all dependencies at once: `pip install "paddleocr[doc2md]"`
+
+---
+
+**Q: I get a `ValueError` saying the format is not supported.**
+
+Run `paddleocr doc2md --formats` to see the currently supported extensions. doc2md only supports `.docx`, `.xlsx`, and `.pptx`. Formats such as `.doc` (legacy Word), `.csv`, and `.pdf` are not supported.
+
+---
+
+**Q: The Excel output is very long because the table has too many rows.**
+
+Use `--max-rows` to limit the number of rows per sheet:
+```bash
+paddleocr doc2md -i data.xlsx -o output.md --max-rows 100
+```
+
+---
+
+**Q: I only want to convert one specific sheet from an Excel file.**
+
+Use `--sheet-name` to specify the sheet by name:
+```bash
+paddleocr doc2md -i data.xlsx -o output.md --sheet-name "Sheet1"
+```
+
+---
+
+**Q: How do I skip headers and footers in a Word document?**
+
+Use the `--no-headers-footers` flag:
+```bash
+paddleocr doc2md -i report.docx -o output.md --no-headers-footers
+```
+
+---
+
+**Q: Where are the extracted images saved?**
+
+When `-o` is specified, images are automatically saved to an `images/` subdirectory in the same directory as the output file. Image references in the Markdown file use relative paths accordingly.
+
+---
+
+**Q: What is the difference between doc2md and PaddleOCR's OCR feature?**
+
+doc2md parses the XML structure of Office documents directly — **no OCR models are used**, making it fast with zero GPU requirement. It is suitable when the original Office file is available. PaddleOCR's OCR feature recognizes text from images or scanned documents, and is suitable when no original document exists.
diff --git a/docs/version3.x/doc2md.md b/docs/version3.x/doc2md.md
new file mode 100644
index 0000000..433b508
--- /dev/null
+++ b/docs/version3.x/doc2md.md
@@ -0,0 +1,480 @@
+---
+comments: true
+---
+
+# 文档转 Markdown 使用教程
+
+## 1. 功能介绍
+
+文档转 Markdown（doc2md）是 PaddleOCR 内置的轻量级 Office 文档结构化转换功能，**无需 OCR 推理**，直接解析文档结构并输出规范的 Markdown 文本。适用于需要将 Word、Excel、PowerPoint 文件快速转为可读文本的场景，如知识库构建、文档检索、内容提取等。
+
+**支持格式**：`.docx`（Word）/ `.xlsx`（Excel）/ `.pptx`（PowerPoint）
+
+**核心能力一览：**
+
+<table>
+<thead>
+<tr>
+<th>功能</th>
+<th>Word (.docx)</th>
+<th>Excel (.xlsx)</th>
+<th>PowerPoint (.pptx)</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>标题层级</td>
+<td>✅ 内置样式 + 字号启发式 + 中文编号</td>
+<td>— </td>
+<td>— </td>
+</tr>
+<tr>
+<td>文本格式化（粗体/斜体/下划线/删除线）</td>
+<td>✅</td>
+<td>✅</td>
+<td>✅</td>
+</tr>
+<tr>
+<td>上标 / 下标</td>
+<td>✅</td>
+<td>✅</td>
+<td>✅</td>
+</tr>
+<tr>
+<td>超链接</td>
+<td>✅</td>
+<td>✅</td>
+<td>✅</td>
+</tr>
+<tr>
+<td>列表（有序 / 无序 / 嵌套）</td>
+<td>✅</td>
+<td>— </td>
+<td>— </td>
+</tr>
+<tr>
+<td>表格（含合并单元格）</td>
+<td>✅ HTML table</td>
+<td>✅ HTML table</td>
+<td>✅ HTML table</td>
+</tr>
+<tr>
+<td>图片</td>
+<td>✅ 按比例宽度</td>
+<td>✅ 浮动图片</td>
+<td>✅ 按比例宽度</td>
+</tr>
+<tr>
+<td>数学公式（OMML → LaTeX）</td>
+<td>✅ 行内 / 显示公式</td>
+<td>✅ drawing 层公式</td>
+<td>✅</td>
+</tr>
+<tr>
+<td>代码块</td>
+<td>✅ 等宽字体自动识别</td>
+<td>— </td>
+<td>— </td>
+</tr>
+<tr>
+<td>文本框</td>
+<td>✅</td>
+<td>— </td>
+<td>— </td>
+</tr>
+<tr>
+<td>图表（Chart）</td>
+<td>✅ → HTML table</td>
+<td>— </td>
+<td>✅ 14 种图表类型</td>
+</tr>
+<tr>
+<td>页眉 / 页脚</td>
+<td>✅ 多节 + 奇偶页</td>
+<td>— </td>
+<td>— </td>
+</tr>
+<tr>
+<td>多 sheet / 多幻灯片</td>
+<td>— </td>
+<td>✅</td>
+<td>✅ <code>---</code> 分隔</td>
+</tr>
+<tr>
+<td>演讲者备注</td>
+<td>— </td>
+<td>— </td>
+<td>✅</td>
+</tr>
+</tbody>
+</table>
+
+## 2. 快速开始
+
+在使用 doc2md 功能前，请确保已按照[安装教程](./installation.md)完成 PaddleOCR 基础安装，然后安装 doc2md 可选依赖：
+
+```bash
+pip install "paddleocr[doc2md]"
+```
+
+**依赖说明：**
+
+<table>
+<thead>
+<tr>
+<th>包名</th>
+<th>版本约束</th>
+<th>用途</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>python-docx</code></td>
+<td><code>&gt;=0.8.11</code></td>
+<td>Word (.docx) 文档解析</td>
+</tr>
+<tr>
+<td><code>python-pptx</code></td>
+<td><code>&gt;=0.6.21</code></td>
+<td>PowerPoint (.pptx) 文档解析</td>
+</tr>
+<tr>
+<td><code>openpyxl</code></td>
+<td><code>&gt;=3.0.0</code></td>
+<td>Excel (.xlsx) 文档解析</td>
+</tr>
+<tr>
+<td><code>pylatexenc</code></td>
+<td><code>&gt;=2.10,&lt;3</code></td>
+<td>数学公式 Unicode → LaTeX 符号映射</td>
+</tr>
+</tbody>
+</table>
+
+### 2.1 命令行方式
+
+```bash
+# 转换 Word 文档，输出到文件
+paddleocr doc2md -i report.docx -o output.md
+
+# 转换 Excel 表格，输出到文件
+paddleocr doc2md -i data.xlsx -o output.md
+
+# 转换 PowerPoint 演示文稿，输出到文件
+paddleocr doc2md -i slides.pptx -o output.md
+
+# 不指定输出路径，结果打印到终端
+paddleocr doc2md -i report.docx
+
+# 查看支持的格式列表
+paddleocr doc2md --formats
+```
+
+<details><summary><b>命令行支持更多参数设置，点击展开以查看命令行参数的详细说明</b></summary>
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>-i</code>, <code>--input</code></td>
+<td><b>含义：</b>输入文件路径，必填（使用 <code>--formats</code> 时可省略）。<br/>
+<b>说明：</b>支持 <code>.docx</code>、<code>.xlsx</code>、<code>.pptx</code> 格式。</td>
+<td><code>str</code></td>
+<td>必填</td>
+</tr>
+<tr>
+<td><code>-o</code>, <code>--output</code></td>
+<td><b>含义：</b>输出 Markdown 文件路径。<br/>
+<b>说明：</b>如果不设置，转换结果将打印到标准输出（stdout）。设置后，Markdown 写入指定文件，图片保存到同目录下的 <code>images/</code> 文件夹。</td>
+<td><code>str</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>-q</code>, <code>--quiet</code></td>
+<td><b>含义：</b>静默模式。<br/>
+<b>说明：</b>启用后不打印耗时、保存路径等提示信息。</td>
+<td>flag</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>--formats</code></td>
+<td><b>含义：</b>列出当前支持的文件格式后退出。<br/>
+<b>说明：</b>启用后不需要 <code>--input</code>，仅输出支持的扩展名列表。</td>
+<td>flag</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>--no-drawings</code></td>
+<td><b>含义：</b>跳过文本框和 drawing 层内容提取。<br/>
+<b>说明：</b>适用于 <code>.docx</code> 和 <code>.xlsx</code>，启用后将跳过文本框（docx）和 drawing 层数学公式（xlsx）的提取。</td>
+<td>flag</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>--no-headers-footers</code></td>
+<td><b>含义：</b>跳过页眉页脚内容提取。<br/>
+<b>说明：</b>仅适用于 <code>.docx</code>，启用后不提取文档的页眉和页脚内容。</td>
+<td>flag</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>--sheet-name</code></td>
+<td><b>含义：</b>仅转换指定名称的 sheet。<br/>
+<b>说明：</b>仅适用于 <code>.xlsx</code>，不设置则转换所有 sheet。</td>
+<td><code>str</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>--max-rows</code></td>
+<td><b>含义：</b>每个 sheet 的最大转换行数。<br/>
+<b>说明：</b>仅适用于 <code>.xlsx</code>，用于限制大表格的输出行数。</td>
+<td><code>int</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+</details>
+<br />
+
+### 2.2 Python API
+
+**基础用法：**
+
+```python
+from paddleocr._doc2md import convert
+
+# 转换文档，返回结果对象
+result = convert("report.docx")
+
+# 访问 Markdown 文本
+print(result.markdown)
+
+# 查看提取的图片（字典，key 为相对路径，value 为图片字节）
+print(list(result.images.keys()))
+
+# 查看文档标题
+print(result.title)
+
+# 查看元信息（格式、sheet 数等）
+print(result.metadata)
+```
+
+**`ConvertResult` 字段说明：**
+
+<table>
+<thead>
+<tr>
+<th>字段</th>
+<th>类型</th>
+<th>说明</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>markdown</code></td>
+<td><code>str</code></td>
+<td>转换后的 Markdown 文本</td>
+</tr>
+<tr>
+<td><code>images</code></td>
+<td><code>dict[str, bytes]</code></td>
+<td>提取的图片字典，key 为相对路径（如 <code>images/image1.png</code>），value 为图片字节</td>
+</tr>
+<tr>
+<td><code>title</code></td>
+<td><code>Optional[str]</code></td>
+<td>文档标题，可能为 <code>None</code></td>
+</tr>
+<tr>
+<td><code>metadata</code></td>
+<td><code>dict</code></td>
+<td>文档元信息，如格式类型、sheet 数量等</td>
+</tr>
+</tbody>
+</table>
+
+**指定输出路径（自动保存文件和图片）：**
+
+```python
+from paddleocr._doc2md import convert
+
+# 指定 output 后，Markdown 写入文件，图片保存到同目录 images/ 下
+result = convert("report.docx", output="output/report.md")
+```
+
+**各格式可用的 kwargs 参数：**
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>类型</th>
+<th>默认值</th>
+<th>适用格式</th>
+<th>说明</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>extract_drawings</code></td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+<td>docx, xlsx</td>
+<td>是否提取文本框（docx）/ drawing 层数学公式（xlsx）</td>
+</tr>
+<tr>
+<td><code>extract_headers_footers</code></td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+<td>docx</td>
+<td>是否提取页眉页脚</td>
+</tr>
+<tr>
+<td><code>sheet_name</code></td>
+<td><code>Optional[str]</code></td>
+<td><code>None</code></td>
+<td>xlsx</td>
+<td>仅转换指定名称的 sheet，<code>None</code> 表示转换全部</td>
+</tr>
+<tr>
+<td><code>max_rows</code></td>
+<td><code>Optional[int]</code></td>
+<td><code>None</code></td>
+<td>xlsx</td>
+<td>每个 sheet 的最大转换行数</td>
+</tr>
+</tbody>
+</table>
+
+**按格式传入 kwargs 示例：**
+
+```python
+from paddleocr._doc2md import convert
+
+# Word：不提取文本框和页眉页脚
+result = convert("report.docx", extract_drawings=False, extract_headers_footers=False)
+
+# Excel：仅转换名为 "Sheet1" 的 sheet，最多 100 行
+result = convert("data.xlsx", sheet_name="Sheet1", max_rows=100)
+```
+
+## 3. 各格式支持特性
+
+### 3.1 Word (.docx)
+
+**标题识别**支持三种方式：
+- **内置 Heading 样式**：Word 内置的 Heading 1–6 样式直接映射为 `#`–`######`
+- **字号启发式**：字号大于正文 1.5 倍且段落较短时，自动识别为标题
+- **中文编号**："一、"格式识别为 H2，"（一）"格式识别为 H3
+
+**文本格式化**：粗体（`**`）、斜体（`*`）、下划线（`<u>`）、删除线（`~~`）、上标（`<sup>`）、下标（`<sub>`）
+
+**列表**：有序列表、无序列表、嵌套列表，自动识别缩进层级
+
+**表格**：输出为 HTML `<table>` 格式，支持 `rowspan`/`colspan` 合并单元格
+
+**图片**：按文档内容区宽度百分比计算，输出 `<img width="75%">` 形式
+
+**数学公式**：OMML 格式公式转为 LaTeX，行内公式为 `$...$`，显示公式为 `$$...$$`
+
+**代码块**：自动检测等宽字体（Courier New、Consolas 等 9 种字体），输出为 fenced code block（` ``` `）
+
+**其他**：文本框内容（`wps:txbx`）、图表（Chart → HTML table）、超链接（两种格式）、页眉页脚（多节 + 奇偶页）
+
+### 3.2 Excel (.xlsx)
+
+**多 sheet**：每个 sheet 输出一个以 `## sheet名称` 开头的章节
+
+**数据边界裁剪**：自动去除尾部空行和空列，只输出有效数据区域
+
+**合并单元格**：使用 `rowspan`/`colspan` 还原单元格合并结构
+
+**字体格式化**：粗体、斜体、下划线、删除线、上标、下标
+
+**超链接**：支持单元格级超链接
+
+**浮动图片**：支持 `OneCellAnchor` 和 `TwoCellAnchor` 两种锚定方式的浮动图片
+
+**数学公式**：通过解析 drawing 层 XML 提取 OMML 格式公式并转为 LaTeX
+
+### 3.3 PowerPoint (.pptx)
+
+**多幻灯片**：每张幻灯片内容以 `---` 分隔
+
+**文本格式化**：粗体、斜体、下划线、删除线、上标、下标
+
+**图片**：按幻灯片宽度百分比计算，输出带宽度的 `<img>` 标签
+
+**表格**：HTML `<table>` 格式，支持合并单元格，支持带背景图片的表格
+
+**图表**：支持 14 种图表类型，转换为 HTML table 输出
+
+**分组形状（GroupShape）**：递归处理嵌套的形状组合
+
+**数学公式**：从 `mc:AlternateContent` 中提取 OMML 格式公式并转为 LaTeX
+
+**演讲者备注**：附加在每张幻灯片内容末尾
+
+## 4. FAQ
+
+**Q：转换时提示 `RuntimeError: python-docx is required`？**
+
+doc2md 采用延迟导入，缺少对应格式的解析库时会抛出此错误。请根据提示安装对应依赖：
+```bash
+pip install python-docx     # Word (.docx)
+pip install python-pptx     # PowerPoint (.pptx)
+pip install openpyxl        # Excel (.xlsx)
+pip install pylatexenc      # 数学公式支持
+```
+或直接安装全部依赖：`pip install "paddleocr[doc2md]"`
+
+---
+
+**Q：格式不支持，提示 `ValueError`？**
+
+运行 `paddleocr doc2md --formats` 查看当前支持的扩展名。doc2md 仅支持 `.docx`、`.xlsx`、`.pptx`，不支持 `.doc`（旧版 Word）、`.csv`、`.pdf` 等格式。
+
+---
+
+**Q：Excel 转换后表格行数很多，输出太长？**
+
+使用 `--max-rows` 限制每个 sheet 的行数：
+```bash
+paddleocr doc2md -i data.xlsx -o output.md --max-rows 100
+```
+
+---
+
+**Q：只想转换 Excel 中的某一个 sheet？**
+
+使用 `--sheet-name` 指定 sheet 名称：
+```bash
+paddleocr doc2md -i data.xlsx -o output.md --sheet-name "Sheet1"
+```
+
+---
+
+**Q：Word 文档中的页眉页脚内容不需要，如何跳过？**
+
+使用 `--no-headers-footers` 参数：
+```bash
+paddleocr doc2md -i report.docx -o output.md --no-headers-footers
+```
+
+---
+
+**Q：图片输出到哪里？**
+
+使用 `-o` 指定输出文件时，图片自动保存在输出文件同目录的 `images/` 文件夹下。Markdown 文件中的图片引用路径也会相应更新为相对路径。
+
+---
+
+**Q：doc2md 与 PaddleOCR 的 OCR 功能有什么区别？**
+
+doc2md 直接解析 Office 文档的 XML 结构，**不使用任何 OCR 模型**，速度快、零 GPU 依赖。适用于有原始 Office 文件的场景。PaddleOCR 的 OCR 功能则针对图片或扫描件进行文字识别，适用于没有原始文档的场景。
diff --git a/docs/version3.x/inference_deployment/cross_platform/android_deployment.en.md b/docs/version3.x/inference_deployment/cross_platform/android_deployment.en.md
new file mode 100644
index 0000000..2618d05
--- /dev/null
+++ b/docs/version3.x/inference_deployment/cross_platform/android_deployment.en.md
@@ -0,0 +1,251 @@
+# PP-OCRv6 Android Demo
+
+## Introduction
+
+This project is an Android deployment example for PaddleOCR v6, implementing mobile OCR inference using ONNX Runtime. The project adopts a **SDK and Demo separation** architecture, where the SDK module can be independently integrated into third-party applications.
+
+## Features
+
+- End-to-end text detection and recognition pipeline
+- Supports PP-OCRv6 series ONNX models
+- Detailed performance timing (detection/recognition stage breakdown)
+- MVVM + Jetpack Compose Demo application
+- AAR integration support
+
+## Project Structure
+
+```
+ppocr-android/
+├── ppocr-sdk/                    # OCR SDK (Android Library)
+│   ├── src/main/
+│   │   ├── assets/models/        # Model files directory
+│   │   │   ├── det/              # Detection model: inference.onnx
+│   │   │   └── rec/              # Recognition model: inference.onnx, inference.yml
+│   │   └── java/com/paddle/ocr/
+│   │       ├── PaddleOCR.kt      # [Public API] SDK entry point
+│   │       ├── PaddleOCRConfig.kt # [Public API] Inference configuration
+│   │       └── ...
+│   └── build.gradle.kts
+├── app/                          # Demo App
+│   ├── src/main/java/com/paddle/ocr/demo/
+│   │   ├── OCRApplication.kt     # Initialize SDK
+│   │   └── ui/                   # Compose UI
+│   └── build.gradle.kts
+├── run_benchmark.sh              # Performance test script
+└── README.md
+```
+
+## Requirements
+
+| Dependency | Version |
+|------------|---------|
+| Android Studio | Ladybug (2024.2+) |
+| JDK | 17 |
+| Kotlin | 2.1.0 |
+| minSdk | 26 (Android 8.0) |
+| ONNX Runtime | 1.21.1 |
+| OpenCV | 4.5.3 |
+
+## Quick Start
+
+### 1. Clone the Project
+
+```bash
+git clone https://github.com/PaddlePaddle/PaddleOCR.git
+cd PaddleOCR/deploy/ppocr-android
+```
+
+### 2. Prepare Models
+
+This project supports the following models:
+
+| Model | HuggingFace | BOS |
+|------|-------------|-----|
+| **PP-OCRv6_small** | [Detection model](https://huggingface.co/PaddlePaddle/PP-OCRv6_small_det_onnx) / [Recognition model](https://huggingface.co/PaddlePaddle/PP-OCRv6_small_rec_onnx) | [Detection model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_det_onnx_infer.tar) / [Recognition model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_rec_onnx_infer.tar) |
+| **PP-OCRv6_tiny** | [Detection model](https://huggingface.co/PaddlePaddle/PP-OCRv6_tiny_det_onnx) / [Recognition model](https://huggingface.co/PaddlePaddle/PP-OCRv6_tiny_rec_onnx) | [Detection model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_det_onnx_infer.tar) / [Recognition model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_rec_onnx_infer.tar) |
+| **PP-OCRv5_mobile** | [Detection model](https://huggingface.co/PaddlePaddle/PP-OCRv5_mobile_det_onnx) / [Recognition model](https://huggingface.co/PaddlePaddle/PP-OCRv5_mobile_rec_onnx) | [Detection model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_onnx_infer.tar) / [Recognition model](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_rec_onnx_infer.tar) |
+
+After downloading and extracting, place the files in `ppocr-sdk/src/main/assets/models/`:
+
+- Detection model: place `inference.onnx` in `models/det/`
+- Recognition model: place `inference.onnx` and `inference.yml` in `models/rec/`
+
+### 3. Build and Run
+
+```bash
+# Build Debug APK
+./gradlew :app:assembleDebug
+
+# Install to device
+./gradlew :app:installDebug
+```
+
+Or run directly from Android Studio.
+
+### 4. Try the Demo
+
+1. Open "PP-OCRv6 Demo" application
+2. Wait for model loading to complete
+3. Tap "Select from Gallery" to choose an image
+4. View recognition results and timing statistics
+
+## SDK Integration
+
+### Option 1: Source Code Dependency
+
+1. Copy `ppocr-sdk/` to your project root
+2. Add to `settings.gradle.kts`:
+   ```kotlin
+   include(":ppocr-sdk")
+   ```
+3. Add to your app module's `build.gradle.kts`:
+   ```kotlin
+   implementation(project(":ppocr-sdk"))
+   ```
+
+### Option 2: AAR Dependency
+
+```bash
+# Build AAR
+./gradlew :ppocr-sdk:assembleRelease
+```
+
+AAR output: `ppocr-sdk/build/outputs/aar/ppocr-sdk-release.aar`
+
+Add to your app module's `build.gradle.kts`:
+
+```kotlin
+dependencies {
+    implementation(files("libs/ppocr-sdk-release.aar"))
+    // AAR doesn't transit dependencies, add manually
+    implementation("com.microsoft.onnxruntime:onnxruntime-android:1.21.1")
+    implementation("com.quickbirdstudios:opencv:4.5.3")
+    implementation("org.jetbrains.kotlinx:kotlinx-coroutines-android:1.9.0")
+}
+```
+
+## API Reference
+
+### Create Instance
+
+```kotlin
+// Default configuration
+val ocr = PaddleOCR.create(context)
+
+// Custom configuration
+val ocr = PaddleOCR.create(
+    context = context,
+    config = PaddleOCRConfig(
+        detThresh = 0.3f,
+        detBoxThresh = 0.6f,
+        recScoreThresh = 0.0f,
+        recBatchSize = 1,
+    ),
+    engineConfig = EngineConfig(numThreads = 4),
+    detModelAssetPath = "models/det/inference.onnx",
+    recModelAssetPath = "models/rec/inference.onnx",
+    recConfigAssetPath = "models/rec/inference.yml",
+)
+```
+
+### Perform OCR
+
+```kotlin
+// Pass Bitmap
+val result = ocr.recognize(bitmap)
+
+// Pass image bytes (recommended, consistent with Python pipeline)
+val result = ocr.recognize(imageBytes)
+
+// Read results
+result.results.forEach { item ->
+    println("Text: ${item.text}, Confidence: ${item.confidence}")
+    println("Box: ${item.box.points}")
+}
+println("Detection: ${result.detectionTimeMs}ms, Recognition: ${result.recognitionTimeMs}ms")
+```
+
+### Release Resources
+
+```kotlin
+ocr.release()
+```
+
+### Configuration Parameters
+
+```kotlin
+data class PaddleOCRConfig(
+    val detImgMode: String = "BGR",         // Input color mode
+    val detLimitSideLen: Int = 64,          // Detection side length limit
+    val detLimitType: String = "min",       // Limit strategy
+    val detMaxSideLimit: Int = 4000,        // Maximum side length
+    val detThresh: Float = 0.3f,            // Binarization threshold
+    val detBoxThresh: Float = 0.6f,         // Detection box confidence threshold
+    val detUnclipRatio: Float = 1.5f,       // Detection box expansion ratio
+    val detMaxCandidates: Int = 3000,       // Maximum candidate boxes
+    val detUseDilation: Boolean = false,    // Whether to dilate
+    val detScoreMode: String = "fast",      // Scoring mode
+    val detBoxType: String = "quad",        // Detection box type
+    val recScoreThresh: Float = 0.0f,       // Recognition confidence threshold
+    val recBatchSize: Int = 1,              // Recognition batch size
+)
+```
+
+### Result Models
+
+```kotlin
+data class OCRRunResult(
+    val results: List<OCRResult>,       // Recognition result list
+    val detectionTimeMs: Long,          // Detection time
+    val recognitionTimeMs: Long,        // Recognition time
+    val totalTimeMs: Long,              // Total time
+    val lineCount: Int,                 // Number of lines
+    // Detailed timing...
+)
+
+data class OCRResult(
+    val box: OCRBox,                    // Detection box coordinates
+    val text: String,                   // Recognized text
+    val confidence: Float,              // Confidence score
+)
+```
+
+## Performance Testing
+
+The project provides an automated performance testing script:
+
+```bash
+# Run benchmark (10 tests, 3 warmup)
+./run_benchmark.sh 10 3
+
+# Sample output
+╔═════════════════════════════════════════════════════════════════════════╗
+║  PP-OCRv6 Speed Benchmark Results                                       ║
+╠═════════════════════════════════════════════════════════════════════════╣
+║  Device: GM1900  |  OS: Android 9  |  Lines: 5                          ║
+║  Cold load: 158ms  |  Warmup: 3  |  Measured: 10                        ║
+╠═════════════════════════════════════════════════════════════════════════╣
++-----------------------------+----------+----------+----------+----------+
+| Stage                       |  Mean ms |    Stdev |      P90 |    Min ms|
++-----------------------------+----------+----------+----------+----------+
+| Total pipeline              |   420.40 |     6.37 |      427 |      413 |
++-----------------------------+----------+----------+----------+----------+
+|   Detection (total)         |   348.70 |     4.67 |      356 |      343 |
+|     Preprocess              |    33.30 |     2.90 |       36 |       28 |
+|     Inference               |   311.00 |     2.93 |      315 |      304 |
+|     Postprocess             |     4.40 |     0.49 |        5 |        4 |
+|   Recognition (total)       |    66.20 |     3.16 |       68 |       64 |
+|     Preprocess              |     3.00 |     0.89 |        4 |        2 |
+|     Inference               |    60.60 |     3.14 |       63 |       58 |
+|     Postprocess             |     2.60 |     0.92 |        4 |        1 |
+|   Pipeline overhead         |     5.50 |     0.50 |        6 |        5 |
++-----------------------------+----------+----------+----------+----------+
+╚═════════════════════════════════════════════════════════════════════════╝
+```
+
+## Notes
+
+1. **OpenCV Initialization**: Call `OpenCVUtils.init(context)` before `PaddleOCR.create()`
+2. **Coroutine Usage**: `create()` and `recognize()` are suspend functions, call them in coroutines
+3. **Memory Management**: Call `release()` when no longer needed
+4. **ProGuard Rules**: Refer to `ppocr-sdk/proguard-rules.pro`
diff --git a/docs/version3.x/inference_deployment/cross_platform/android_deployment.md b/docs/version3.x/inference_deployment/cross_platform/android_deployment.md
new file mode 100644
index 0000000..69008d3
--- /dev/null
+++ b/docs/version3.x/inference_deployment/cross_platform/android_deployment.md
@@ -0,0 +1,251 @@
+# PP-OCRv6 Android Demo
+
+## 简介
+
+本项目是 PaddleOCR v6 在 Android 平台的部署示例，基于 ONNX Runtime 实现移动端 OCR 推理。项目采用 **SDK 与 Demo 分离** 架构，SDK 模块可独立集成到第三方应用。
+
+## 功能特性
+
+- 文本检测 + 文本识别端到端流程
+- 支持 PP-OCRv6 系列 ONNX 模型
+- 详细的性能计时（检测/识别各阶段耗时）
+- MVVM + Jetpack Compose Demo 应用
+- 支持 AAR 方式集成
+
+## 项目结构
+
+```
+ppocr-android/
+├── ppocr-sdk/                    # OCR SDK（Android Library）
+│   ├── src/main/
+│   │   ├── assets/models/        # 模型文件目录
+│   │   │   ├── det/              # 检测模型：inference.onnx
+│   │   │   └── rec/              # 识别模型：inference.onnx, inference.yml
+│   │   └── java/com/paddle/ocr/
+│   │       ├── PaddleOCR.kt      # [公开 API] SDK 入口
+│   │       ├── PaddleOCRConfig.kt # [公开 API] 推理参数配置
+│   │       └── ...
+│   └── build.gradle.kts
+├── app/                          # Demo App
+│   ├── src/main/java/com/paddle/ocr/demo/
+│   │   ├── OCRApplication.kt     # 初始化 SDK
+│   │   └── ui/                   # Compose UI
+│   └── build.gradle.kts
+├── run_benchmark.sh              # 性能测试脚本
+└── README.md
+```
+
+## 环境要求
+
+| 依赖 | 版本 |
+|------|------|
+| Android Studio | Ladybug (2024.2+) |
+| JDK | 17 |
+| Kotlin | 2.1.0 |
+| minSdk | 26 (Android 8.0) |
+| ONNX Runtime | 1.21.1 |
+| OpenCV | 4.5.3 |
+
+## 快速开始
+
+### 1. 克隆项目
+
+```bash
+git clone https://github.com/PaddlePaddle/PaddleOCR.git
+cd PaddleOCR/deploy/ppocr-android
+```
+
+### 2. 准备模型
+
+本项目支持以下模型：
+
+| 模型 | HuggingFace | BOS |
+|------|-------------|-----|
+| **PP-OCRv6_small** | [检测模型](https://huggingface.co/PaddlePaddle/PP-OCRv6_small_det_onnx) / [识别模型](https://huggingface.co/PaddlePaddle/PP-OCRv6_small_rec_onnx) | [检测模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_det_onnx_infer.tar) / [识别模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_rec_onnx_infer.tar) |
+| **PP-OCRv6_tiny** | [检测模型](https://huggingface.co/PaddlePaddle/PP-OCRv6_tiny_det_onnx) / [识别模型](https://huggingface.co/PaddlePaddle/PP-OCRv6_tiny_rec_onnx) | [检测模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_det_onnx_infer.tar) / [识别模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_rec_onnx_infer.tar) |
+| **PP-OCRv5_mobile** | [检测模型](https://huggingface.co/PaddlePaddle/PP-OCRv5_mobile_det_onnx) / [识别模型](https://huggingface.co/PaddlePaddle/PP-OCRv5_mobile_rec_onnx) | [检测模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_onnx_infer.tar) / [识别模型](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_rec_onnx_infer.tar) |
+
+下载并解压后，将文件放入 `ppocr-sdk/src/main/assets/models/` 目录：
+
+- 检测模型：将 `inference.onnx` 放入 `models/det/`
+- 识别模型：将 `inference.onnx` 和 `inference.yml` 放入 `models/rec/`
+
+### 3. 编译运行
+
+```bash
+# 编译 Debug APK
+./gradlew :app:assembleDebug
+
+# 安装到设备
+./gradlew :app:installDebug
+```
+
+或使用 Android Studio 直接运行。
+
+### 4. 体验 Demo
+
+1. 打开 "PP-OCRv6 Demo" 应用
+2. 等待模型加载完成
+3. 点击 "Select from Gallery" 选择图片
+4. 查看识别结果和耗时统计
+
+## SDK 集成
+
+### 方式一：源码依赖
+
+1. 将 `ppocr-sdk/` 复制到项目根目录
+2. 在 `settings.gradle.kts` 添加：
+   ```kotlin
+   include(":ppocr-sdk")
+   ```
+3. 在 App 模块 `build.gradle.kts` 添加：
+   ```kotlin
+   implementation(project(":ppocr-sdk"))
+   ```
+
+### 方式二：AAR 依赖
+
+```bash
+# 构建 AAR
+./gradlew :ppocr-sdk:assembleRelease
+```
+
+AAR 输出：`ppocr-sdk/build/outputs/aar/ppocr-sdk-release.aar`
+
+在 App 模块 `build.gradle.kts` 添加：
+
+```kotlin
+dependencies {
+    implementation(files("libs/ppocr-sdk-release.aar"))
+    // AAR 不传递依赖，需手动添加
+    implementation("com.microsoft.onnxruntime:onnxruntime-android:1.21.1")
+    implementation("com.quickbirdstudios:opencv:4.5.3")
+    implementation("org.jetbrains.kotlinx:kotlinx-coroutines-android:1.9.0")
+}
+```
+
+## API 参考
+
+### 创建实例
+
+```kotlin
+// 默认配置
+val ocr = PaddleOCR.create(context)
+
+// 自定义配置
+val ocr = PaddleOCR.create(
+    context = context,
+    config = PaddleOCRConfig(
+        detThresh = 0.3f,
+        detBoxThresh = 0.6f,
+        recScoreThresh = 0.0f,
+        recBatchSize = 1,
+    ),
+    engineConfig = EngineConfig(numThreads = 4),
+    detModelAssetPath = "models/det/inference.onnx",
+    recModelAssetPath = "models/rec/inference.onnx",
+    recConfigAssetPath = "models/rec/inference.yml",
+)
+```
+
+### 执行 OCR
+
+```kotlin
+// 传入 Bitmap
+val result = ocr.recognize(bitmap)
+
+// 传入图片字节数据（推荐，与 Python 流程一致）
+val result = ocr.recognize(imageBytes)
+
+// 读取结果
+result.results.forEach { item ->
+    println("文本: ${item.text}, 置信度: ${item.confidence}")
+    println("坐标: ${item.box.points}")
+}
+println("检测: ${result.detectionTimeMs}ms, 识别: ${result.recognitionTimeMs}ms")
+```
+
+### 释放资源
+
+```kotlin
+ocr.release()
+```
+
+### 配置参数
+
+```kotlin
+data class PaddleOCRConfig(
+    val detImgMode: String = "BGR",         // 输入色彩模式
+    val detLimitSideLen: Int = 64,          // 检测侧边长限制
+    val detLimitType: String = "min",       // 限制策略
+    val detMaxSideLimit: Int = 4000,        // 最长边上限
+    val detThresh: Float = 0.3f,            // 二值化阈值
+    val detBoxThresh: Float = 0.6f,         // 检测框置信度阈值
+    val detUnclipRatio: Float = 1.5f,       // 检测框扩展比例
+    val detMaxCandidates: Int = 3000,       // 最大候选框数
+    val detUseDilation: Boolean = false,    // 是否膨胀
+    val detScoreMode: String = "fast",      // 打分模式
+    val detBoxType: String = "quad",        // 检测框类型
+    val recScoreThresh: Float = 0.0f,       // 识别置信度阈值
+    val recBatchSize: Int = 1,              // 识别批大小
+)
+```
+
+### 结果模型
+
+```kotlin
+data class OCRRunResult(
+    val results: List<OCRResult>,       // 识别结果列表
+    val detectionTimeMs: Long,          // 检测耗时
+    val recognitionTimeMs: Long,        // 识别耗时
+    val totalTimeMs: Long,              // 总耗时
+    val lineCount: Int,                 // 识别行数
+    // 详细计时...
+)
+
+data class OCRResult(
+    val box: OCRBox,                    // 检测框坐标
+    val text: String,                   // 识别文本
+    val confidence: Float,              // 置信度
+)
+```
+
+## 性能测试
+
+项目提供自动化性能测试脚本：
+
+```bash
+# 运行 benchmark（10次测试，3次预热）
+./run_benchmark.sh 10 3
+
+# 输出示例
+╔═════════════════════════════════════════════════════════════════════════╗
+║  PP-OCRv6 Speed Benchmark Results                                       ║
+╠═════════════════════════════════════════════════════════════════════════╣
+║  Device: GM1900  |  OS: Android 9  |  Lines: 5                          ║
+║  Cold load: 158ms  |  Warmup: 3  |  Measured: 10                        ║
+╠═════════════════════════════════════════════════════════════════════════╣
++-----------------------------+----------+----------+----------+----------+
+| Stage                       |  Mean ms |    Stdev |      P90 |    Min ms|
++-----------------------------+----------+----------+----------+----------+
+| Total pipeline              |   420.40 |     6.37 |      427 |      413 |
++-----------------------------+----------+----------+----------+----------+
+|   Detection (total)         |   348.70 |     4.67 |      356 |      343 |
+|     Preprocess              |    33.30 |     2.90 |       36 |       28 |
+|     Inference               |   311.00 |     2.93 |      315 |      304 |
+|     Postprocess             |     4.40 |     0.49 |        5 |        4 |
+|   Recognition (total)       |    66.20 |     3.16 |       68 |       64 |
+|     Preprocess              |     3.00 |     0.89 |        4 |        2 |
+|     Inference               |    60.60 |     3.14 |       63 |       58 |
+|     Postprocess             |     2.60 |     0.92 |        4 |        1 |
+|   Pipeline overhead         |     5.50 |     0.50 |        6 |        5 |
++-----------------------------+----------+----------+----------+----------+
+╚═════════════════════════════════════════════════════════════════════════╝
+```
+
+## 注意事项
+
+1. **OpenCV 初始化**：调用 `PaddleOCR.create()` 前需先调用 `OpenCVUtils.init(context)`
+2. **协程调用**：`create()` 和 `recognize()` 都是 suspend 函数，需在协程中调用
+3. **内存管理**：不再使用时调用 `release()` 释放资源
+4. **混淆规则**：参考 `ppocr-sdk/proguard-rules.pro`
diff --git a/docs/version3.x/inference_deployment/cross_platform/browser.en.md b/docs/version3.x/inference_deployment/cross_platform/browser.en.md
new file mode 100644
index 0000000..9e03966
--- /dev/null
+++ b/docs/version3.x/inference_deployment/cross_platform/browser.en.md
@@ -0,0 +1,225 @@
+---
+comments: true
+---
+
+# PaddleOCR.js (browser deployment)
+
+PaddleOCR provides **PaddleOCR.js**, a browser OCR SDK for running the PP-OCR pipeline in the browser. You can embed text detection and recognition in web apps and run inference on the client.
+
+The npm package is **`@paddleocr/paddleocr-js`**. Source and demo live under [`paddleocr-js`](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/paddleocr-js) on GitHub.
+
+## Install
+
+```bash
+npm install @paddleocr/paddleocr-js
+```
+
+## Quick start
+
+```js
+import { PaddleOCR } from "@paddleocr/paddleocr-js";
+
+const ocr = await PaddleOCR.create({
+  lang: "ch",
+  ocrVersion: "PP-OCRv5",
+  ortOptions: {
+    backend: "auto"
+  }
+});
+
+const [result] = await ocr.predict(fileOrBlob);
+console.log(result.items);
+```
+
+`predict` resolves to an **array** of `OcrResult` (one per input image). A single `Blob` / `File` still produces a one-element array—use destructuring or `results[0]`.
+
+## Construction options
+
+Two styles: **direct parameters** to `PaddleOCR.create({ ... })`, or a **`pipelineConfig`** object.
+
+### 1. Direct parameters
+
+With direct parameters, you can specify models and set batch sizes, ORT options, and other runtime settings.
+
+**Model selection — `lang` + `ocrVersion`:**
+
+```js
+await PaddleOCR.create({
+  lang: "ch",
+  ocrVersion: "PP-OCRv5"
+});
+```
+
+**Model selection — built-in model names:**
+
+```js
+await PaddleOCR.create({
+  textDetectionModelName: "PP-OCRv5_mobile_det",
+  textRecognitionModelName: "PP-OCRv5_mobile_rec"
+});
+```
+
+**Custom models** — provide a name and asset URL for each of detection and recognition:
+
+```js
+await PaddleOCR.create({
+  textDetectionModelName: "my_det_model",
+  textDetectionModelAsset: {
+    url: "https://example.com/models/my_det_model.tar"
+  },
+  textRecognitionModelName: "my_rec_model",
+  textRecognitionModelAsset: {
+    url: "https://example.com/models/my_rec_model.tar"
+  }
+});
+```
+
+**Batch sizes, ORT options, and other runtime settings:**
+
+```js
+await PaddleOCR.create({
+  lang: "ch",
+  ocrVersion: "PP-OCRv5",
+  textDetectionBatchSize: 2,
+  textRecognitionBatchSize: 8,
+  ortOptions: {
+    backend: "wasm",
+    wasmPaths: "/assets/"
+  }
+});
+```
+
+#### Custom model archive format and validation
+
+The SDK downloads `textDetectionModelAsset.url` / `textRecognitionModelAsset.url` over HTTP(S) and parses the body as a **plain ustar tar (uncompressed)** archive. Ensure that:
+
+| Requirement | Details |
+|-------------|---------|
+| Archive format | The response body must be an **uncompressed `.tar`**. The implementation does **not** gunzip **`.tar.gz`**; if you pass a gzip-compressed tarball, parsing will typically fail and an error will be thrown. |
+| Required files | The tar must contain **`inference.onnx`** and **`inference.yml`** (they may live in a subdirectory; entries are matched by basename). |
+| `model_name` | **`inference.yml`** must define a **`model_name`** that matches the `textDetectionModelName` / `textRecognitionModelName` you pass to `create`. This is checked after load during initialization. |
+
+If you need to convert Paddle models into the ONNX model files used here, see [Obtaining ONNX models](../others/obtaining_onnx_models.en.md). The standard model files produced by that workflow can then be packaged as a `.tar` following the rules above for use with PaddleOCR.js.
+
+If the archive or model files do not meet these rules, initialization typically fails with an **`Error`** that describes the problem, for example: non-2xx download, missing `inference.onnx` / `inference.yml` in the tar, empty resources, missing or mismatched `model_name`, incomplete model config, or ONNX load failure. There is no silent fallback.
+
+All selected OCR models must satisfy the `model_name` rules above.
+
+### 2. Pipeline config
+
+```js
+import { PaddleOCR } from "@paddleocr/paddleocr-js";
+
+const pipelineConfig = `
+pipeline_name: OCR
+SubModules:
+  TextDetection:
+    model_name: PP-OCRv5_mobile_det
+    batch_size: 2
+  TextRecognition:
+    model_name: PP-OCRv5_mobile_rec
+    batch_size: 6
+`;
+
+const ocr = await PaddleOCR.create({ pipelineConfig });
+```
+
+`pipelineConfig` can be YAML text or a parsed object. In the browser, submodule `model_dir` must be **`null` or an asset object** (e.g. `{ url: "..." }`), not a local filesystem path string. If you want to start from a pipeline configuration exported by PaddleOCR / PaddleX, see the "Exporting Pipeline Configuration Files" section in [PaddleOCR and PaddleX](../../paddleocr_and_paddlex.en.md); the exported YAML can be used as the basis for `pipelineConfig`, and any `model_dir` entries should then be adapted to browser-side asset objects.
+
+If both direct parameters and `pipelineConfig` are provided, **direct parameters take precedence**.
+
+## Prediction
+
+### Params
+
+`ocr.predict(image | images[], params?)` accepts both camelCase and PaddleOCR-style snake_case:
+
+- `textDetLimitSideLen` or `text_det_limit_side_len`
+- `textDetLimitType` or `text_det_limit_type`
+- `textDetMaxSideLimit` or `text_det_max_side_limit`
+- `textDetThresh` or `text_det_thresh`
+- `textDetBoxThresh` or `text_det_box_thresh`
+- `textDetUnclipRatio` or `text_det_unclip_ratio`
+- `textRecScoreThresh` or `text_rec_score_thresh`
+
+Supported `image` inputs include `Blob`, `ImageBitmap`, `ImageData`, `HTMLCanvasElement`, `HTMLImageElement`, and `cv.Mat`. Pass an array to run on multiple images in one call.
+
+In **worker mode**, `cv.Mat` is not transferable and is not supported as input.
+
+### Return value
+
+Resolves to `Promise<OcrResult[]>`. Each `OcrResult` contains:
+
+- `image`: `{ width, height }` for that source  
+- `items`: recognized lines (`poly`, `text`, `score`)  
+- `metrics`: `detMs`, `recMs`, `totalMs`, `detectedBoxes`, `recognizedCount` — box and line counts are per image; `detMs`, `recMs`, and `totalMs` cover the **entire** `predict()` call (identical on every element when you pass multiple images)  
+- `runtime`: requested backend and provider metadata  
+
+## Worker mode
+
+You can run the pipeline inside a dedicated Worker while keeping the same high-level API:
+
+```js
+import { PaddleOCR } from "@paddleocr/paddleocr-js";
+
+const ocr = await PaddleOCR.create({
+  lang: "ch",
+  ocrVersion: "PP-OCRv5",
+  worker: true,
+  ortOptions: {
+    backend: "wasm",
+    wasmPaths: "https://cdn.jsdelivr.net/npm/onnxruntime-web/dist/",
+    numThreads: 2,
+    simd: true
+  }
+});
+```
+
+Summary:
+
+- Worker mode uses the package worker entry, not ONNX Runtime Web `env.wasm.proxy`  
+- When `worker: true`, the package forces ORT wasm proxy off to avoid nested workers  
+- Browser inputs are normalized on the main thread, then transferred to the worker  
+- `cv.Mat` is only supported on the main-thread pipeline path  
+
+
+## Visualization
+
+The optional **`@paddleocr/paddleocr-js/viz`** subpath renders OCR results to images.
+
+```js
+import { OcrVisualizer } from "@paddleocr/paddleocr-js/viz";
+
+const viz = new OcrVisualizer({
+  font: { family: "Noto Sans SC", source: "/fonts/NotoSansSC-Regular.ttf" }
+});
+
+const blob = await viz.toBlob(imageBitmap, result);
+const url = URL.createObjectURL(blob);
+const a = document.createElement("a");
+a.href = url;
+a.download = "ocr_result.png";
+a.click();
+URL.revokeObjectURL(url);
+
+viz.dispose();
+```
+
+`renderOcrToBlob` and `deterministicColor` are also exported. Visualization takes a **single** `OcrResult` (for one image, use the first element of the `predict` result array).
+
+## API summary
+
+- `PaddleOCR.create(options)`  
+- `ocr.initialize()` / `ocr.getInitializationSummary()`  
+- `ocr.predict(image | images[], params?)` → `Promise<OcrResult[]>`  
+- `ocr.dispose()`  
+- `parseOcrPipelineConfigText(text)` / `normalizeOcrPipelineConfig(config)`  
+- `OcrVisualizer`, `renderOcrToBlob`, `deterministicColor` (from `@paddleocr/paddleocr-js/viz`)  
+
+## Host application responsibilities
+
+The SDK manages OpenCV.js and ONNX Runtime internally. You still handle:
+
+- **COOP/COEP** (and related headers) when enabling threaded WASM or WebGPU  
+- **ORT environment options** (e.g. `wasmPaths`, threads, SIMD)  
+- A bundler/runtime that can emit and load **module workers** when `worker: true`  
diff --git a/docs/version3.x/inference_deployment/cross_platform/browser.md b/docs/version3.x/inference_deployment/cross_platform/browser.md
new file mode 100644
index 0000000..375a8e2
--- /dev/null
+++ b/docs/version3.x/inference_deployment/cross_platform/browser.md
@@ -0,0 +1,225 @@
+---
+comments: true
+---
+
+# PaddleOCR.js（浏览器端部署）
+
+PaddleOCR 提供可在浏览器中运行 PP-OCR 产线的浏览器端 OCR SDK **PaddleOCR.js**。您可以将文字检测与识别能力嵌入 Web 应用，在客户端完成推理。
+
+npm 上的包名为 **`@paddleocr/paddleocr-js`**。源码与演示应用位于 GitHub 仓库的 [`paddleocr-js`](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/paddleocr-js) 目录。
+
+## 安装
+
+```bash
+npm install @paddleocr/paddleocr-js
+```
+
+## 快速开始
+
+```js
+import { PaddleOCR } from "@paddleocr/paddleocr-js";
+
+const ocr = await PaddleOCR.create({
+  lang: "ch",
+  ocrVersion: "PP-OCRv5",
+  ortOptions: {
+    backend: "auto"
+  }
+});
+
+const [result] = await ocr.predict(fileOrBlob);
+console.log(result.items);
+```
+
+`predict` 返回 **`OcrResult` 组成的数组**（每张输入图像对应一项）。传入单个 `Blob` / `File` 时也会得到长度为 1 的数组，请使用解构或 `results[0]` 取值。
+
+## 构造选项
+
+主要有两种构造方式：向 `PaddleOCR.create({ ... })` 传入**直接参数**，或传入 **产线配置** `pipelineConfig`。
+
+### 1. 直接参数
+
+可通过直接参数指定模型，也可设置推理 batch size、ORT 选项等运行参数。
+
+**按语言与版本选择模型** — `lang` + `ocrVersion`：
+
+```js
+await PaddleOCR.create({
+  lang: "ch",
+  ocrVersion: "PP-OCRv5"
+});
+```
+
+**按内置模型名选择**：
+
+```js
+await PaddleOCR.create({
+  textDetectionModelName: "PP-OCRv5_mobile_det",
+  textRecognitionModelName: "PP-OCRv5_mobile_rec"
+});
+```
+
+**自定义模型** — 为检测 / 识别分别指定模型名与资源地址：
+
+```js
+await PaddleOCR.create({
+  textDetectionModelName: "my_det_model",
+  textDetectionModelAsset: {
+    url: "https://example.com/models/my_det_model.tar"
+  },
+  textRecognitionModelName: "my_rec_model",
+  textRecognitionModelAsset: {
+    url: "https://example.com/models/my_rec_model.tar"
+  }
+});
+```
+
+**batch size 与 ORT 选项等运行参数**：
+
+```js
+await PaddleOCR.create({
+  lang: "ch",
+  ocrVersion: "PP-OCRv5",
+  textDetectionBatchSize: 2,
+  textRecognitionBatchSize: 8,
+  ortOptions: {
+    backend: "wasm",
+    wasmPaths: "/assets/"
+  }
+});
+```
+
+#### 自定义模型包格式与校验行为
+
+SDK 通过 HTTP(S) 下载 `textDetectionModelAsset.url` / `textRecognitionModelAsset.url` 指向的资源，并按 **标准 ustar tar（未压缩）** 解析。请保证：
+
+| 要求 | 说明 |
+|------|------|
+| 归档格式 | 响应体需为 **未压缩的 `.tar`**。当前实现不对 **`.tar.gz` / gzip** 做解压；若误传压缩包，解析结果通常不正确，随后会报错。 |
+| 必需文件 | tar 内必须包含 **`inference.onnx`** 与 **`inference.yml`**（可在子目录中；按文件名匹配）。 |
+| `model_name` | `inference.yml` 中必须能解析出 **`model_name`**，且须与您在 `create` 中传入的 `textDetectionModelName` / `textRecognitionModelName` **完全一致**。初始化加载模型后会校验。 |
+
+如需从 Paddle 模型转换得到这里使用的 ONNX 模型文件，可参考 [获取 ONNX 模型](../others/obtaining_onnx_models.md)。按该文档转换得到的标准模型文件，可按上述要求打包为 `.tar` 后提供给 PaddleOCR.js 使用。
+
+若格式不满足要求，通常会在初始化阶段以 **抛出带明确信息的 `Error`** 失败，例如：下载非 2xx、tar 中找不到 `inference.onnx` / `inference.yml`、资源为空、`model_name` 缺失或与所选名称不一致、模型配置不完整，或 ONNX 无法加载。不会在后台静默失败。
+
+所有通过上述方式选用的 OCR 模型，其 `inference.yml` 都应满足上述 `model_name` 约定。
+
+### 2. 产线配置
+
+```js
+import { PaddleOCR } from "@paddleocr/paddleocr-js";
+
+const pipelineConfig = `
+pipeline_name: OCR
+SubModules:
+  TextDetection:
+    model_name: PP-OCRv5_mobile_det
+    batch_size: 2
+  TextRecognition:
+    model_name: PP-OCRv5_mobile_rec
+    batch_size: 6
+`;
+
+const ocr = await PaddleOCR.create({ pipelineConfig });
+```
+
+`pipelineConfig` 可以是 YAML 文本，也可以是已解析的对象。在浏览器中，子模块的 `model_dir` 仅支持 **`null` 或资源描述对象**（形如 `{ url: "..." }`），不支持本地路径字符串。如需基于 PaddleOCR / PaddleX 导出的产线配置作为起点，可参考 [PaddleOCR 与 PaddleX](../../paddleocr_and_paddlex.md) 中“导出产线配置文件”一节；导出的 YAML 可作为 `pipelineConfig` 的基础，再按浏览器端要求将其中的 `model_dir` 调整为资源描述对象。
+
+若同时提供直接参数与 `pipelineConfig`，**以直接参数为准**。
+
+## 预测
+
+### 参数
+
+`ocr.predict(image | images[], params?)` 同时接受 camelCase 与 PaddleOCR 风格的 snake_case：
+
+- `textDetLimitSideLen` / `text_det_limit_side_len`
+- `textDetLimitType` / `text_det_limit_type`
+- `textDetMaxSideLimit` / `text_det_max_side_limit`
+- `textDetThresh` / `text_det_thresh`
+- `textDetBoxThresh` / `text_det_box_thresh`
+- `textDetUnclipRatio` / `text_det_unclip_ratio`
+- `textRecScoreThresh` / `text_rec_score_thresh`
+
+支持的 `image` 类型包括：`Blob`、`ImageBitmap`、`ImageData`、`HTMLCanvasElement`、`HTMLImageElement`、`cv.Mat`。可传入上述类型的数组以在一次调用中处理多图。
+
+在 **Worker 模式**下，`cv.Mat` 无法传输，不能作为 Worker 路径的输入。
+
+### 返回值
+
+返回 `Promise<OcrResult[]>`。每个 `OcrResult` 包含：
+
+- `image`：该图源尺寸 `{ width, height }`
+- `items`：识别行（`poly`、`text`、`score`）
+- `metrics`：`detMs`、`recMs`、`totalMs`、`detectedBoxes`、`recognizedCount` — 框数与行数为**每张图**统计；`detMs`、`recMs`、`totalMs` 为**整次** `predict()` 调用耗时（多图时每一项上这三个值相同）
+- `runtime`：请求的后端与各阶段 Provider 等元数据
+
+## Worker 模式
+
+可在独立 Worker 中运行产线，同时保持相同的高层 API：
+
+```js
+import { PaddleOCR } from "@paddleocr/paddleocr-js";
+
+const ocr = await PaddleOCR.create({
+  lang: "ch",
+  ocrVersion: "PP-OCRv5",
+  worker: true,
+  ortOptions: {
+    backend: "wasm",
+    wasmPaths: "https://cdn.jsdelivr.net/npm/onnxruntime-web/dist/",
+    numThreads: 2,
+    simd: true
+  }
+});
+```
+
+行为概要：
+
+- Worker 模式使用包内 Worker 脚本路径，而非 ONNX Runtime Web 的 `env.wasm.proxy`
+- 启用 `worker: true` 时，包内会关闭 ORT 的 wasm proxy，避免双层 Worker
+- 浏览器输入先在主线程标准化，再传入 Worker 推理
+- `cv.Mat` 仅支持主线程产线路径
+
+
+## 可视化
+
+可选子路径 **`@paddleocr/paddleocr-js/viz`** 提供将 OCR 结果渲染为图像的工具。
+
+```js
+import { OcrVisualizer } from "@paddleocr/paddleocr-js/viz";
+
+const viz = new OcrVisualizer({
+  font: { family: "Noto Sans SC", source: "/fonts/NotoSansSC-Regular.ttf" }
+});
+
+const blob = await viz.toBlob(imageBitmap, result);
+const url = URL.createObjectURL(blob);
+const a = document.createElement("a");
+a.href = url;
+a.download = "ocr_result.png";
+a.click();
+URL.revokeObjectURL(url);
+
+viz.dispose();
+```
+
+也提供一次性函数 `renderOcrToBlob` 与 `deterministicColor`（用于与内置检测框配色一致）。可视化需传入**单个** `OcrResult`（单张图时取 `predict` 返回数组的首项）。
+
+## API 摘要
+
+- `PaddleOCR.create(options)`
+- `ocr.initialize()` / `ocr.getInitializationSummary()`
+- `ocr.predict(image | images[], params?)` → `Promise<OcrResult[]>`
+- `ocr.dispose()`
+- `parseOcrPipelineConfigText(text)` / `normalizeOcrPipelineConfig(config)`
+- `OcrVisualizer`、`renderOcrToBlob`、`deterministicColor`（`@paddleocr/paddleocr-js/viz`）
+
+## 宿主环境职责
+
+SDK 内部管理 OpenCV.js 与 ONNX Runtime。您仍需自行处理：
+
+- 启用多线程 WASM 或 WebGPU 时所需的 **COOP/COEP** 等响应头
+- **ORT 环境选项**（如 `wasmPaths`、线程数、SIMD）
+- 使用 **`worker: true`** 时，构建工具需能产出并加载 **module worker**
diff --git a/docs/version3.x/inference_deployment/cross_platform/ios_deployment.en.md b/docs/version3.x/inference_deployment/cross_platform/ios_deployment.en.md
new file mode 100644
index 0000000..3c371be
--- /dev/null
+++ b/docs/version3.x/inference_deployment/cross_platform/ios_deployment.en.md
@@ -0,0 +1,171 @@
+---
+comments: true
+---
+
+# iOS Deployment
+
+PaddleOCR ships an iOS OCR sample app under [`deploy/ios_demo`](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/deploy/ios_demo). The **SwiftUI** demo runs exported **ONNX** models on device using the [ONNX Runtime Objective-C API](https://onnxruntime.ai/docs/tutorials/mobile/) for detection and recognition.
+
+This guide covers environment requirements, model asset setup, building and running in Xcode, and optional ORT format conversion and benchmarking.
+
+## Project layout
+
+App sources, bundled resources, and third-party **source** vendored for this demo live under **`PaddleOCRDemo/`**. Unit tests are in **`PaddleOCRDemoTests/`** next to the Xcode project. The project root also contains `Podfile`, **`scripts/`**, `README.md`, and `NOTICE`.
+
+## Prerequisites
+
+- **macOS** with **Xcode 16.0+** (deployment target **iOS 16+**)
+- **CocoaPods** (`gem install cocoapods` or Homebrew)
+- `curl`, `tar`
+
+## Supported model presets
+
+`scripts/fetch_ios_demo_models.sh` downloads ONNX bundles into **`PaddleOCRDemo/Models/`**. Supported presets:
+
+- **`PP-OCRv6_small`** (default)
+- **`PP-OCRv6_tiny`**
+- **`PP-OCRv5_mobile`**
+
+## Quick start
+
+From the **project root** (`deploy/ios_demo`):
+
+```bash
+pod install
+./scripts/fetch_ios_demo_models.sh
+```
+
+Optionally pass the **model preset** (bundle name such as `PP-OCRv6_small`) as a positional argument:
+
+```bash
+./scripts/fetch_ios_demo_models.sh PP-OCRv6_small
+```
+
+Intermediate `.tar` caches are stored under **`.fetch_ios_demo_models_work/`** at the project root.
+
+After `pod install`, open the workspace:
+
+```bash
+open PaddleOCRDemo.xcworkspace
+```
+
+Build the **PaddleOCRDemo** scheme. Ensure **`PaddleOCRDemo/Models/`** and **`PaddleOCRDemo/Resources/SampleImages/`** are included in the app target via folder references / **Copy Bundle Resources**, and **`PaddleOCRDemoTests/Fixtures/`** in the test target (as in the checked-in project). The built-in picker sample is **`general_ocr_002.jpg`**.
+
+### ONNX Runtime execution providers (EPs)
+
+The **CPU** / **XNNPACK** / **Core ML** segmented control chooses which ONNX Runtime EP is preferred first when creating the session. ONNX Runtime may still execute some operators on the default **CPU** EP when another EP lacks an implementation. Selecting **CPU** does not register the XNNPACK or Core ML EPs.
+
+## Convert to ORT model format (optional)
+
+To produce [ORT format](https://onnxruntime.ai/docs/performance/model-optimizations/ort-format-models.html) weights:
+
+```bash
+python3 -m pip install -r requirements-onnx-convert-ort.txt
+./scripts/convert_onnx_to_ort.sh
+```
+
+By default, conversion writes `inference*.ort` next to each `inference.onnx` under `PaddleOCRDemo/Models/`. The demo loads `inference*.ort` when present, so you do not need to ship ONNX in the bundle. To keep only one weight file in the app, delete the `inference.onnx` files manually, or use **`--output-dir`** and replace files under `PaddleOCRDemo/Models/`. Use **`--input-dir`** to point at any ONNX tree on the host; it defaults to `PaddleOCRDemo/Models` when omitted.
+
+By default the script uses **`--optimization-style Fixed`** (one `.ort` per ONNX file). The raw Python tool defaults to both Fixed and Runtime (`*.ort` and `*.with_runtime_opt.ort`). For Runtime-style ORT files (e.g. Core ML–oriented exports), pass **`--optimization-style Runtime`** (alias: `--optimization_style`). You can pass both: **`--optimization-style Fixed Runtime`**.
+
+The converter emits **`required_operators*.config`** for minimal ORT builds; **by default the shell script deletes them** after a successful run. Pass **`--keep-operator-config`** to retain them. They are **not** read at inference time. The iOS demo loads only **`.ort`** (or `.onnx`) weights.
+
+Example (Runtime conversion to a separate folder):
+
+```bash
+./scripts/convert_onnx_to_ort.sh --output-dir ./out/ort_bundles --optimization-style Runtime
+```
+
+## Benchmark
+
+This demo provides a benchmark pipeline for measuring on-device OCR latency and memory.
+
+### Prerequisites
+
+Complete [Quick start](#quick-start) first. For benchmark runs on a **physical device** you also need:
+
+- **Code signing (real device only):** `xcodebuild` requires a valid Apple Development certificate and provisioning profile. Open `PaddleOCRDemo.xcworkspace` in Xcode once, select your Development Team under *Signing & Capabilities*, and let Xcode manage provisioning; later `run_benchmark.sh` invocations reuse the cached profile. On first deploy, trust the developer certificate on the device under **Settings → General → VPN & Device Management** (one-time per device). Simulator runs do not require signing.
+- Optional accuracy precheck: PaddleOCR (ONNX Runtime engine) on the host for reference generation, and `python3 -m pip install -r requirements-accuracy.txt`.
+
+### Full pipeline
+
+From the project root, use **`./scripts/run_benchmark.sh`**. The script resolves the input image, optionally runs an accuracy precheck, runs the XCTest benchmark on a simulator or device, extracts artifacts, and writes the Markdown report. It always invokes `xcodebuild test` with `-configuration Release`.
+
+Prefer script flags. `--image` copies an arbitrary image into test fixtures for this run. `--fixture` selects an existing file under `PaddleOCRDemoTests/Fixtures/`. If neither is set, the script requires exactly one non-`local-*` image under `Fixtures/` (e.g. `ios_ocr_benchmark_reference.jpg`).
+
+| Intent | Flags |
+| --- | --- |
+| Destination | `--udid <id>` for a real device, or `--simulator <name>` |
+| Image from an arbitrary path | `--image <path>` |
+| Image already under `Fixtures/` | `--fixture <name>` |
+| Benchmark intensity | `--warmup <n>`, `--measured-iterations <n>` |
+| Preferred ONNX Runtime EP | `--ort-execution-provider CPU`, `XNNPACK`, or `CORE_ML` |
+| ONNX Runtime profiling JSON | `--ort-profiling` |
+| Optional accuracy precheck | `--accuracy-check`, optionally `--accuracy-reference-json <path>` |
+| Gate benchmark on accuracy `FAIL` | `--accuracy-check --stop-on-accuracy-failure` |
+| Output directory | `--output-dir <dir>` (default: `out/`) |
+| Clean previous artifacts | `--clean` |
+
+```bash
+./scripts/run_benchmark.sh --udid <device-udid> --warmup 5 --measured-iterations 30
+./scripts/run_benchmark.sh --udid <udid> --image /path/to/photo.png --ort-execution-provider CPU
+./scripts/run_benchmark.sh --fixture ios_ocr_benchmark_reference --warmup 2 --measured-iterations 20
+./scripts/run_benchmark.sh
+PADDLEOCR_BENCHMARK_MEASURED_ITERATIONS=30 ./scripts/run_benchmark.sh --warmup 0
+PADDLEOCR_BENCHMARK_ORT_EXECUTION_PROVIDER=XNNPACK ./scripts/run_benchmark.sh --fixture ios_ocr_benchmark_reference --warmup 1
+./scripts/run_benchmark.sh --accuracy-check --udid <device-udid> --measured-iterations 30
+./scripts/run_benchmark.sh --accuracy-check --stop-on-accuracy-failure --udid <device-udid>
+./scripts/run_benchmark.sh --output-dir ./benchmark-out --measured-iterations 30
+./scripts/run_benchmark.sh --udid <device-udid> --ort-profiling
+```
+
+`--accuracy-check` runs before the benchmark in a separate XCTest invocation. Results are `PASS`, `FAIL`, or `ERROR`. By default, `FAIL` records a mismatch but continues; `ERROR` stops the pipeline. `--stop-on-accuracy-failure` also skips the benchmark on `FAIL`.
+
+**ORT session profiling** (attachments `ort_profile_detection`, `ort_profile_recognition`): use `--ort-profiling`. Run profiling in a **separate** run from clean latency measurements.
+
+After the benchmark **completes**, the script overwrites `run-status.json`, `on-device-performance.json`, `xctest-memory-metrics.json`, and `benchmark-report.md` under the output directory (`out/` by default). On **ERROR**, those files may be missing or partial.
+
+The report includes model input tensor shape distributions, first measured run line count, inferred model preset, actual model format (`onnx` or `ort`), det/rec/total weight sizes, app executable size when resolvable, cold model load time, measured latency, memory, etc.
+
+| Artifact | Producer | Purpose |
+|---|---|---|
+| `accuracy-reference.json` | `ocr_reference_run.py` | Optional reference JSON |
+| `accuracy-result.xcresult` | `xcodebuild test` | Optional accuracy precheck |
+| `ios-ocr-export.json` | `extract_xcresult_attachments.py` | Optional iOS accuracy payload |
+| `accuracy-summary.json` | `compare_ocr_json.py` | Optional accuracy summary |
+| `latency-result.xcresult` | `xcodebuild test` | Latency benchmark |
+| `memory-result.xcresult` | `xcodebuild test` | Memory benchmark |
+| `on-device-performance.json` | `extract_xcresult_attachments.py` | Latency stats from `latency-result.xcresult` |
+| `ort_profile_detection`, `ort_profile_recognition` | `extract_xcresult_attachments.py` | Optional ORT profiling JSON |
+| `xctest-memory-metrics.json` | `extract_xctest_metrics.py` | XCTest memory metrics |
+| `logs/*.log` | `run_benchmark.sh` | Per-step logs |
+| `run-status.json` | `run_benchmark.sh` | Step outcomes and metadata |
+| `benchmark-report.md` | `generate_benchmark_report.py` | Human-readable report |
+
+Exit **`0`** when the pipeline reaches **`COMPLETED`**; non-zero on **`ERROR`**. Accuracy `FAIL` changes the exit code only with `--stop-on-accuracy-failure`; `ERROR` always stops the pipeline.
+
+### Advanced: XCTest environment
+
+Most users should use `run_benchmark.sh` flags instead of setting XCTest environment variables directly. This section is for debugging the `PaddleOCRDemoTests` target from Xcode or a custom `xcodebuild test` command.
+
+Benchmark tests read **`PADDLEOCR_BENCHMARK_*`** variables on the XCTest runner process.
+
+| How you launch tests | What to configure |
+| --- | --- |
+| **Xcode** | Scheme **PaddleOCRDemo** → **Test** → **Arguments** → **Environment Variables** (`PADDLEOCR_BENCHMARK_…` as-is). |
+| **Custom `xcodebuild test`** | **`TEST_RUNNER_` + variable name** (e.g. `TEST_RUNNER_PADDLEOCR_BENCHMARK_IMAGE_NAME=…`). |
+
+| Variable | If unset | Role |
+| --- | --- | --- |
+| `PADDLEOCR_BENCHMARK_IMAGE_NAME` | **`ios_ocr_benchmark_reference`** | Bundled test image under **`Fixtures/`**. |
+| `PADDLEOCR_BENCHMARK_WARMUP_ITERATIONS` | **3** | Untimed full OCR runs before timing. |
+| `PADDLEOCR_BENCHMARK_MEASURED_ITERATIONS` | **10** | Timed runs. |
+| `PADDLEOCR_BENCHMARK_ORT_EXECUTION_PROVIDER` | **CPU** | **`CPU`**, **`XNNPACK`**, or **`CORE_ML`**. |
+| `PADDLEOCR_BENCHMARK_ORT_PROFILING` | (unset) | Enable ORT session profiling (`1` / `true` / `yes` / `on`). |
+| `PADDLEOCR_BENCHMARK_ONLY_TESTING_SCOPE` | latency + memory benchmark tests | `run_benchmark.sh` only: comma-separated `-only-testing` scope. |
+
+Non-negative integers for the two iteration variables.
+
+## Third-party licenses
+
+Bundled **Clipper** (polyclipping 6.4.2) is under the [Boost Software License 1.0](https://www.boost.org/LICENSE_1_0.txt); see `NOTICE` and `PaddleOCRDemo/ThirdParty/Clipper1/LICENSE`. CocoaPods pods are governed by their respective licenses (see `Podfile.lock` after `pod install`).
diff --git a/docs/version3.x/inference_deployment/cross_platform/ios_deployment.md b/docs/version3.x/inference_deployment/cross_platform/ios_deployment.md
new file mode 100644
index 0000000..5c86ccf
--- /dev/null
+++ b/docs/version3.x/inference_deployment/cross_platform/ios_deployment.md
@@ -0,0 +1,171 @@
+---
+comments: true
+---
+
+# iOS 部署
+
+PaddleOCR 在 [`deploy/ios_demo`](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/deploy/ios_demo) 提供基于 **SwiftUI** 的 iOS OCR 示例应用。应用在设备上使用导出的 **ONNX** 模型，通过 [ONNX Runtime Objective-C API](https://onnxruntime.ai/docs/tutorials/mobile/) 完成检测与识别推理。
+
+本指南介绍该 Demo 的环境要求、模型资源准备、在 Xcode 中构建运行，以及可选的 ORT 格式转换与基准测试流程。
+
+## 工程结构
+
+应用源码、打包资源及为本 Demo 引入的第三方**源码**均位于 **`PaddleOCRDemo/`** 下。单元测试位于与 Xcode 工程同级的 **`PaddleOCRDemoTests/`**。工程根目录还包含 `Podfile`、`scripts/`、`README.md` 与 `NOTICE`。
+
+## 环境要求
+
+- 搭载 **macOS** 的开发机，**Xcode 16.0+**（部署目标 **iOS 16+**）
+- **CocoaPods**（`gem install cocoapods` 或 Homebrew）
+- `curl`、`tar`
+
+## 支持的模型预设
+
+`scripts/fetch_ios_demo_models.sh` 会将 ONNX 模型包下载到 **`PaddleOCRDemo/Models/`**。当前支持的预设为：
+
+- **`PP-OCRv6_small`**（默认）
+- **`PP-OCRv6_tiny`**
+- **`PP-OCRv5_mobile`**
+
+## 快速开始 {#快速开始}
+
+在 **工程根目录**（`deploy/ios_demo`）执行：
+
+```bash
+pod install
+./scripts/fetch_ios_demo_models.sh
+```
+
+可选：通过位置参数指定模型预设（即 bundle 名称，如 `PP-OCRv6_small`）：
+
+```bash
+./scripts/fetch_ios_demo_models.sh PP-OCRv6_small
+```
+
+中间 `.tar` 缓存目录为工程根下的 **`.fetch_ios_demo_models_work/`**。
+
+使用 CocoaPods 时，需先执行 `pod install` 再打开工作区：
+
+```bash
+open PaddleOCRDemo.xcworkspace
+```
+
+构建 **PaddleOCRDemo** scheme。请确认 **`PaddleOCRDemo/Models/`** 与 **`PaddleOCRDemo/Resources/SampleImages/`** 已通过文件夹引用 / **Copy Bundle Resources** 加入 App target；**`PaddleOCRDemoTests/Fixtures/`** 已加入测试 target（与仓库中工程配置一致）。内置样例图为 **`general_ocr_002.jpg`**。
+
+### ONNX Runtime 执行提供程序（EP）
+
+界面上的 **CPU** / **XNNPACK** / **Core ML** 分段控件用于在创建 Session 时指定优先使用的 EP。当某一 EP 无法覆盖部分算子时，ONNX Runtime 仍可能回退到默认 **CPU** EP。选择 **CPU** 时不会注册 XNNPACK 或 Core ML EP。
+
+## 转换为 ORT 模型格式（可选）
+
+生成 [ORT 格式](https://onnxruntime.ai/docs/performance/model-optimizations/ort-format-models.html) 权重：
+
+```bash
+python3 -m pip install -r requirements-onnx-convert-ort.txt
+./scripts/convert_onnx_to_ort.sh
+```
+
+默认在 `PaddleOCRDemo/Models/` 下每个 `inference.onnx` 旁生成 `inference*.ort`；Demo 在存在 `.ort` 时会优先加载，因此可不将 ONNX 打入 App 包。若只希望保留一种权重文件，可手动删除 `inference.onnx`，或用 **`--output-dir`** 将 `.ort` 输出到单独目录后再替换 `PaddleOCRDemo/Models/` 中的文件。省略 **`--input-dir`** 时默认输入目录为 `PaddleOCRDemo/Models`。
+
+默认使用 **`--optimization-style Fixed`**（每个 ONNX 对应一个 `.ort`）。底层 Python 工具若未指定，会同时生成 Fixed 与 Runtime 两种（`*.ort` 与 `*.with_runtime_opt.ort`）。需要 Runtime 风格（如面向 Core ML 的导出）时，传入 **`--optimization-style Runtime`**（别名 `--optimization_style`）；可同时传入 **`--optimization-style Fixed Runtime`**。
+
+转换会生成 **`required_operators*.config`**（用于裁剪最小 ORT 构建）；**默认在脚本成功结束后删除**。使用 **`--keep-operator-config`** 可保留。这些文件**不参与** Demo 推理，仅加载 **`.ort`**（或 `.onnx`）权重。
+
+示例（Runtime，输出到独立目录）：
+
+```bash
+./scripts/convert_onnx_to_ort.sh --output-dir ./out/ort_bundles --optimization-style Runtime
+```
+
+## 基准测试（Benchmark）
+
+Demo 提供在真机或模拟器上测量 OCR 时延与内存的流水线。
+
+### 前置条件
+
+先完成 [快速开始](#快速开始) 中的资源准备。真机基准测试还需要：
+
+- **代码签名（仅真机）**：`xcodebuild` 部署到物理设备需要有效的 Apple Development 证书与描述文件。建议在 Xcode 中打开 `PaddleOCRDemo.xcworkspace`，在 *Signing & Capabilities* 中选择 Development Team 并启用自动管理；之后 `run_benchmark.sh` 等命令可复用已缓存的配置。首次在设备上启动时，须在 **设置 → 通用 → VPN 与设备管理** 中信任开发者证书（每台设备一次）。模拟器无需签名。
+- 可选精度预检：主机安装 PaddleOCR（ONNX Runtime 引擎）用于生成参考结果，并执行 `python3 -m pip install -r requirements-accuracy.txt`。
+
+### 完整流水线
+
+在工程根目录执行 **`./scripts/run_benchmark.sh`**。脚本会解析输入图像、可选执行精度预检、在模拟器或真机上运行 XCTest 基准、提取产物并生成 Markdown 报告。始终以 **Release** 配置调用 `xcodebuild test`。
+
+**配置方式**：优先使用脚本参数。`--image` 将任意路径图像复制到本次运行的 `Fixtures/local-*`；`--fixture` 选择 `PaddleOCRDemoTests/Fixtures/` 下已有文件。若两者均未指定，则要求 `Fixtures/` 下恰好有一张非 `local-*` 的图像（常见为 `ios_ocr_benchmark_reference.jpg`）。
+
+| 用途 | 参数 |
+| --- | --- |
+| 运行目标 | `--udid <id>`（真机）或 `--simulator <name>` |
+| 任意路径图像 | `--image <path>` |
+| 已有 Fixture | `--fixture <name>`（stem 或 `stem.ext`） |
+| 强度 | `--warmup <n>`、`--measured-iterations <n>` |
+| 优先 ORT EP | `--ort-execution-provider CPU`、`XNNPACK` 或 `CORE_ML` |
+| ORT profiling JSON | `--ort-profiling` |
+| 可选精度预检 | `--accuracy-check`，可选 `--accuracy-reference-json <path>` |
+| 精度 FAIL 时中止 | `--accuracy-check --stop-on-accuracy-failure` |
+| 输出目录 | `--output-dir <dir>`（默认 `out/`） |
+| 清理旧产物 | `--clean` |
+
+```bash
+./scripts/run_benchmark.sh --udid <device-udid> --warmup 5 --measured-iterations 30
+./scripts/run_benchmark.sh --udid <udid> --image /path/to/photo.png --ort-execution-provider CPU
+./scripts/run_benchmark.sh --fixture ios_ocr_benchmark_reference --warmup 2 --measured-iterations 20
+./scripts/run_benchmark.sh
+PADDLEOCR_BENCHMARK_MEASURED_ITERATIONS=30 ./scripts/run_benchmark.sh --warmup 0
+PADDLEOCR_BENCHMARK_ORT_EXECUTION_PROVIDER=XNNPACK ./scripts/run_benchmark.sh --fixture ios_ocr_benchmark_reference --warmup 1
+./scripts/run_benchmark.sh --accuracy-check --udid <device-udid> --measured-iterations 30
+./scripts/run_benchmark.sh --accuracy-check --stop-on-accuracy-failure --udid <device-udid>
+./scripts/run_benchmark.sh --output-dir ./benchmark-out --measured-iterations 30
+./scripts/run_benchmark.sh --udid <device-udid> --ort-profiling
+```
+
+`--accuracy-check` 在基准测试前以独立 XCTest 运行。结果为 `PASS`、`FAIL` 或 `ERROR`。默认 `FAIL` 仅记录并继续基准；`ERROR` 会中止流水线。`--stop-on-accuracy-failure` 会在 `FAIL` 时跳过基准并返回非零退出码。
+
+**ORT Session profiling**（附件 `ort_profile_detection`、`ort_profile_recognition`）：使用 `--ort-profiling`。Profiling 会改变运行时行为，应与干净时延测量**分开**运行。
+
+基准**成功完成**后，脚本会在输出目录（默认 `out/`）覆盖写入 `run-status.json`、`on-device-performance.json`、`xctest-memory-metrics.json`、`benchmark-report.md`。若流水线提前 **ERROR**，这些文件可能缺失或不完整。
+
+报告内容包括模型输入张量形状分布、首次测量运行的行数、推断的模型预设、实际模型格式（`onnx` 或 `ort`）、det/rec/总权重大小、在可解析时的 App 可执行文件大小、冷启动模型加载时间、测量时延与内存等。形状统计在测量循环内按模型调用计数：每次完整 OCR 检测贡献 1 次，识别按 batch 各贡献 1 次。
+
+| 产物 | 来源 | 用途 |
+|---|---|---|
+| `accuracy-reference.json` | `ocr_reference_run.py` | 可选参考 JSON（除非提供 `--accuracy-reference-json`） |
+| `accuracy-result.xcresult` | `xcodebuild test` | 可选精度预检 |
+| `ios-ocr-export.json` | `extract_xcresult_attachments.py` | 从精度预检提取的 iOS 精度 payload |
+| `accuracy-summary.json` | `compare_ocr_json.py` | 精度预检摘要 |
+| `latency-result.xcresult` | `xcodebuild test` | 时延基准 |
+| `memory-result.xcresult` | `xcodebuild test` | 内存基准 |
+| `on-device-performance.json` | `extract_xcresult_attachments.py` | 从 `latency-result.xcresult` 提取的时延统计 |
+| `ort_profile_detection`、`ort_profile_recognition` | `extract_xcresult_attachments.py` | 可选 ORT profiling JSON |
+| `xctest-memory-metrics.json` | `extract_xctest_metrics.py` | XCTest 内存指标 |
+| `logs/*.log` | `run_benchmark.sh` | 各步骤日志 |
+| `run-status.json` | `run_benchmark.sh` | 步骤状态与元数据 |
+| `benchmark-report.md` | `generate_benchmark_report.py` | 可读报告 |
+
+流水线状态为 **`COMPLETED`** 时退出码 **`0`**；**`ERROR`** 时非零。精度预检 `FAIL` 仅在使用 `--stop-on-accuracy-failure` 时改变退出码；`ERROR` 始终中止。
+
+### 进阶：XCTest 环境变量
+
+多数场景应使用 `run_benchmark.sh` 参数，而非直接设置环境变量。本节适用于在 Xcode 或自定义 `xcodebuild test` 中调试 **PaddleOCRDemoTests**。
+
+基准测试通过 **`PADDLEOCR_BENCHMARK_*`** 读取配置（作用于 XCTest runner，不会自动传入交互式 shell）。
+
+| 启动方式 | 配置方法 |
+| --- | --- |
+| **Xcode** | Scheme **PaddleOCRDemo** → **Test** → **Arguments** → **Environment Variables**，变量名如下。 |
+| **自定义 `xcodebuild test`** | 使用 **`TEST_RUNNER_` + 变量名**（如 `TEST_RUNNER_PADDLEOCR_BENCHMARK_IMAGE_NAME=…`）。 |
+
+| 变量 | 未设置时 | 作用 |
+| --- | --- | --- |
+| `PADDLEOCR_BENCHMARK_IMAGE_NAME` | **`ios_ocr_benchmark_reference`** | `Fixtures/` 下 bundled 图像（stem 或 `stem.ext`）。 |
+| `PADDLEOCR_BENCHMARK_WARMUP_ITERATIONS` | **3** | 计时时前的预热完整 OCR 次数。 |
+| `PADDLEOCR_BENCHMARK_MEASURED_ITERATIONS` | **10** | 计时迭代次数。 |
+| `PADDLEOCR_BENCHMARK_ORT_EXECUTION_PROVIDER` | **CPU** | **`CPU`**、**`XNNPACK`** 或 **`CORE_ML`**。 |
+| `PADDLEOCR_BENCHMARK_ORT_PROFILING` | （未设置） | 启用 ORT session profiling（`1` / `true` / `yes` / `on`）。 |
+| `PADDLEOCR_BENCHMARK_ONLY_TESTING_SCOPE` | latency + memory 基准测试 | 仅 `run_benchmark.sh`：传给多次 `xcodebuild -only-testing` 的逗号分隔列表。 |
+
+两个迭代相关变量须为非负整数。
+
+## 第三方许可
+
+随 Demo 打包的 **Clipper**（polyclipping 6.4.2）采用 [Boost Software License 1.0](https://www.boost.org/LICENSE_1_0.txt)；详见 `NOTICE` 与 `PaddleOCRDemo/ThirdParty/Clipper1/LICENSE`。CocoaPods 依赖的许可以 `pod install` 后的 `Podfile.lock` 及各自仓库为准。
diff --git a/docs/version3.x/inference_deployment/local_inference/cpp/OCR.en.md b/docs/version3.x/inference_deployment/local_inference/cpp/OCR.en.md
new file mode 100644
index 0000000..670f71a
--- /dev/null
+++ b/docs/version3.x/inference_deployment/local_inference/cpp/OCR.en.md
@@ -0,0 +1,986 @@
+# C++ Local Deployment for General OCR Pipeline - Linux
+
+- [1. Environment Preparation](#1-environment-preparation)
+    - [1.1 Compile OpenCV Library](#11-compile-opencv-library)
+    - [1.2 Compile Paddle Inference](#12-compile-paddle-inference)
+- [2. Getting Started](#2-getting-started)
+    - [2.1 Compile Prediction Demo](#21-compile-prediction-demo)
+    - [2.2 Prepare Models](#22-prepare-models)
+    - [2.3 Run Prediction Demo](#23-run-the-prediction-demo)
+    - [2.4 C++ API Integration](#24-c-api-integration)
+- [3. Extended Features](#3-extended-features)
+    - [3.1 Multilingual Text Recognition](#31-multilingual-text-recognition)
+    - [3.2 Visualize Text Recognition Results](#32-visualize-text-recognition-results)
+- [4. FAQ](#4-faq)
+
+This section introduces the method for deploying a general OCR pipeline in C++. The general OCR pipeline consists of the following five modules:
+
+1. Document Image Orientation Classification Module (Optional)
+2. Text Image Unwarping Module (Optional)
+3. Text Line Orientation Classification Module (Optional)
+4. Text Detection Module
+5. Text Recognition Module
+
+Below, we will explain how to configure the C++ environment and complete the deployment of the general OCR pipeline in a Linux (CPU/GPU) environment.
+
+- Note:
+    - For specific compilation methods in a Windows environment, please refer to the [Windows Compilation Tutorial](./OCR_windows.en.md). After compilation, the subsequent commands for running the demo are the same as those in Linux.
+
+## 1. Environment Preparation
+
+- **The source code used for compilation and execution in this chapter can be found in the [PaddleOCR/deploy/cpp_infer](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/deploy/cpp_infer) directory.**
+
+- Linux environment.
+    - gcc 8.2 (when compiling with the Paddle Inference GPU version, gcc>=11.2)
+    - cmake 3.18
+
+### 1.1 Compile OpenCV Library
+
+Currently, only OpenCV 4.x versions are supported. Below, we use OpenCV 4.7.0 as an example.
+
+1. Execute the following commands to download the OpenCV source code:
+
+```bash
+cd deploy/cpp_infer
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/cpp/libs/opencv-4.7.0.tgz
+tar -xf opencv-4.7.0.tgz
+```
+
+2. Configure and compile the OpenCV library:
+
+- a. In the `tools/build_opencv.sh` script, set `root_path` to the absolute path of the opencv-4.7.0 source code.
+- b. Set `install_path`, such as the default `${root_path}/opencv4`. `install_path` will be used as the path to the OpenCV library when compiling the prediction demo later.
+- c. After configuration, run the following command to compile OpenCV:
+
+    ```bash
+    sh tools/build_opencv.sh
+    ```
+
+### 1.2 Compile Paddle Inference
+
+You can choose to directly download a pre-compiled package or manually compile the source code.
+
+#### 1.2.1 Directly Download Pre-compiled Package (Recommended)
+
+The [Paddle Inference official website](https://www.paddlepaddle.org.cn/inference/v3.0/guides/install/download_lib.html) provides Linux prediction libraries. You can view and select the appropriate pre-compiled package on the website.
+
+After downloading, extract it:
+
+```shell
+tar -xvf paddle_inference.tgz
+```
+
+This will generate a subfolder `paddle_inference/` in the current directory.
+
+#### 1.2.2 Compile Prediction Library from Source Code
+
+You can choose to compile the prediction library from source code. Compiling from source allows flexible configuration of various features and dependencies to adapt to different hardware and software environments. For detailed steps, please refer to [Source Code Compilation under Linux](https://www.paddlepaddle.org.cn/inference/v3.0/guides/install/compile/source_compile_under_Linux.html).
+
+## 2. Getting Started
+
+### 2.1 Compile Prediction Demo
+
+Before compiling the prediction demo, ensure that you have compiled the OpenCV library and the Paddle Inference prediction library according to sections 1.1 and 1.2.
+
+After modifying the configurations in `tools/build.sh`, execute the following command to compile:
+
+```shell
+sh tools/build.sh
+```
+
+Detailed descriptions of the relevant configuration parameters are as follows:
+
+<table>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Default Value</th>
+</tr>
+<tr>
+<td><code>OPENCV_DIR</code></td>
+<td>The path where OpenCV is compiled and installed (such as the <code>install_path</code> mentioned when compiling OpenCV above, required).</td>
+<td></td>
+</tr>
+<tr>
+<td><code>LIB_DIR</code></td>
+<td>The path to the downloaded <code>Paddle Inference</code> pre-compiled package or the manually compiled Paddle Inference library path (such as the <code>build/paddle_inference_install_dir</code> folder), required.</td>
+<td></td>
+</tr>
+<tr>
+<td><code>CUDA_LIB_DIR</code></td>
+<td>The path to the CUDA library files, usually <code>/usr/local/cuda/lib64</code>. This parameter needs to be set when the Paddle Inference library is the GPU version and <code>-DWITH_GPU=ON</code> is set.</td>
+<td></td>
+</tr>
+<tr>
+<td><code>CUDNN_LIB_DIR</code></td>
+<td>The path to the cuDNN library files, usually <code>/usr/lib/x86_64-linux-gnu/</code>. This parameter needs to be set when the Paddle Inference library is the GPU version and <code>-DWITH_GPU=ON</code> is set.</td>
+<td></td>
+</tr>
+<tr>
+<td><code>WITH_GPU</code></td>
+<td>When set to ON, you can compile the GPU version demo, which requires the Paddle Inference library to be the GPU version.</td>
+<td>OFF</td>
+</tr>
+</table>
+
+**Note: The above paths need to be absolute paths.**
+
+### 2.2 Prepare Models
+
+You can directly download the inference models provided by PaddleOCR:
+
+<details>
+<summary><b>Document Image Orientation Classification Module (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Top-1 Acc (%)</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">Inference Model</a></td>
+<td>99.06</td>
+<td>7</td>
+<td>A document image classification model based on PP-LCNet_x1_0, with four categories: 0 degrees, 90 degrees, 180 degrees, and 270 degrees.</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Text Image Unwrapping Module (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>CER</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">Inference Model</a></td>
+<td>0.179</td>
+<td>30.3</td>
+<td>A high-precision text image unwarping model.</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Text Line Orientation Classification Module (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th>
+<th>Model Download Link</th>
+<th>Top-1 Acc (%)</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<td>PP-LCNet_x1_0_textline_ori (Default)</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_textline_ori_infer.tar">Inference Model</a></td>
+<td>99.42</td>
+<td>6.5</td>
+<td>A text line classification model based on PP-LCNet_x1_0, with two categories: 0 degrees and 180 degrees.</td>
+</tr>
+<tr>
+<td>PP-LCNet_x0_25_textline_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x0_25_textline_ori_infer.tar">Inference Model</a></td>
+<td>98.85</td>
+<td>0.96</td>
+<td>A text line classification model based on PP-LCNet_x0_25, with two categories: 0 degrees and 180 degrees.</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Text Detection Module:</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Detection Hmean (%)</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv5_server_det (Default)</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_det_infer.tar">Inference Model</a></td>
+<td>83.8</td>
+<td>84.3</td>
+<td>A server-side text detection model for PP-OCRv5, with higher precision, suitable for deployment on servers with better performance.</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_infer.tar">Inference Model</a></td>
+<td>79.0</td>
+<td>4.7</td>
+<td>A mobile-side text detection model for PP-OCRv5, with higher efficiency, suitable for deployment on edge devices.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_det_infer.tar">Inference Model</a></td>
+<td>69.2</td>
+<td>109</td>
+<td>A server-side text detection model for PP-OCRv4, with higher precision, suitable for deployment on servers with better performance.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar">Inference Model</a></td>
+<td>63.8</td>
+<td>4.7</td>
+<td>A mobile-side text detection model for PP-OCRv4, with higher efficiency, suitable for deployment on edge devices.</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Text Recognition Module:</b></summary>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Recognition Avg Accuracy (%)</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec (Default)</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">Inference Model</a></td>
+<td>86.38</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec is a new-generation text recognition model. It aims to efficiently and accurately support the recognition of four major languages: simplified Chinese, traditional Chinese, English, and Japanese, as well as complex text scenarios such as handwriting, vertical text, pinyin, and rare characters, with a single model. While maintaining recognition effectiveness, it balances inference speed and model robustness, providing efficient and accurate technical support for document understanding in various scenarios.</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">Inference Model</a></td>
+<td>81.29</td>
+<td>16</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">Inference Model</a></td>
+<td>86.58</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc is trained on a mixed dataset of more Chinese document data and PP-OCR training data, based on PP-OCRv4_server_rec. It enhances the recognition ability of some traditional Chinese characters, Japanese characters, and special characters, supporting over 15,000 characters. In addition to improving the text recognition ability related to documents, it also enhances the general text recognition ability.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">Inference Model</a></td>
+<td>78.74</td>
+<td>10.5</td>
+<td>A lightweight recognition model for PP-OCRv4, with high inference efficiency, suitable for deployment on various hardware devices including edge devices.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">Inference Model</a></td>
+<td>85.19</td>
+<td>173</td>
+<td>A server-side model for PP-OCRv4, with high inference precision, suitable for deployment on various servers.</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+You can also refer to the model export sections of each module, such as [Text Detection Module - Model Export](../../../module_usage/text_detection.en.md#44-model-export), to export the trained models as inference models.
+
+The directory structure of the models is generally as follows:
+
+```
+PP-OCRv5_mobile_det
+|–inference.pdiparams (Model weights file)
+|–inference.json (Model structure file, in JSON format)
+|–inference.yml (Model configuration file, in YAML format)
+```
+
+### 2.3 Run the Prediction Demo
+
+Before using the General OCR Pipeline C++ locally, please first successfully compile the prediction demo. After compilation, you can experience it via the command line or call the API for secondary development and then recompile to generate the application.
+
+**Please note that if you encounter issues such as the program becoming unresponsive, abnormal program exits, memory resource exhaustion, or extremely slow inference speeds during execution, try adjusting the configuration by referring to the documentation, for example, by disabling unused features or using a lighter model.**
+
+This demo supports both system pipeline calls and individual module calls. Before running the following code, please download the [sample image](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png) locally:
+
+Running Method:
+
+```shell
+./build/ppocr <pipeline_or_module> [--param1] [--param2] [...]
+```
+
+Common parameters are as follows:
+
+<li>Input and Output Related</li>
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td>The local image to be predicted, required. Only supports images in <code>jpg</code>, <code>png</code>, <code>jpeg</code>, and <code>bmp</code> formats.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td>Specifies the path where the inference result files will be saved. Both the JSON file and the predicted result image will be saved under this path.</td>
+<td><code>str</code></td>
+<td><code>./output</code></td>
+</tr>
+</tbody>
+</table>
+
+<details><summary><b>Click to expand for detailed descriptions of more parameters</b></summary>
+
+<li>General Parameters</li>
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>device</code></td>
+<td>The device used for inference. Supports specifying a specific card number:
+<ul>
+<li><b>CPU</b>: For example, <code>cpu</code> indicates using the CPU for inference;</li>
+<li><b>GPU</b>: For example, <code>gpu:0</code> indicates using the first GPU for inference;</li>
+</ul>If not set, it will use the default value initialized by the pipeline. During initialization, if <code>-DWITH_GPU=ON</code> is added during compilation, it will prioritize using the local GPU device 0; otherwise, it will use the CPU device.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td>The computation precision, such as <code>fp32</code>, <code>fp16</code>.</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>Whether to enable MKL-DNN for accelerated inference. If MKL-DNN is not available or the model does not support acceleration via MKL-DNN, acceleration will not be used even if this flag is set.</td>
+<td><code>bool</code></td>
+<td><code>true</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td>The number of threads for the PaddleInference CPU acceleration library.</td>
+<td><code>int</code></td>
+<td><code>8</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td>The path to the PaddleX pipeline configuration file.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+</tbody>
+</table>
+
+<li>Module Switches</li>
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td>Whether to load and use the document orientation classification module. If not set, it will use the default value initialized by the pipeline, which is <code>true</code> by default.</td>
+<td><code>bool</code></td>
+<td><code>true</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td>Whether to load and use the text image correction module. If not set, it will use the default value initialized by the pipeline, which is <code>true</code> by default.</td>
+<td><code>bool</code></td>
+<td><code>true</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td>Whether to load and use the text line orientation module. If not set, it will use the default value initialized by the pipeline, which is <code>true</code> by default.</td>
+<td><code>bool</code></td>
+<td><code>true</code></td>
+</tr>
+</tbody>
+</table>
+
+<li>Detection Model Related</li>
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td>The name of the text detection model. If not set, it will use the default model of the pipeline. When the model name passed in via the text detection model path is inconsistent with the configured name of the pipeline's default text recognition model, you need to specify the name of the passed-in model.</td>
+<td><code>str</code></td>
+<td><code>PP-OCRv5_server_det</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td>The directory path of the text detection model, required.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td>The image side length limit for text detection.
+Any integer greater than <code>0</code>. If not set, it will use the default value initialized by the pipeline, which is <code>64</code> by default.
+</td>
+<td><code>int</code></td>
+<td><code>64</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td>The side length limit type for text detection. Supports <code>min</code> and <code>max</code>, where <code>min</code> means ensuring the shortest side of the image is not less than <code>det_limit_side_len</code>, and <code>max</code> means ensuring the longest side of the image is not greater than <code>limit_side_len</code>. If not set, it will use the default value initialized by the pipeline, which is <code>min</code> by default.
+</td>
+<td><code>str</code></td>
+<td><code>min</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td>The pixel threshold for text detection. Only pixels with scores greater than this threshold in the output probability map will be considered as text pixels.
+Any floating-point number greater than <code>0</code>. If not set, it will use the default value initialized by the pipeline.
+</td>
+<td><code>float</code></td>
+<td><code>0.3</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td>The bounding box threshold for text detection. When the average score of all pixels within the detected bounding box is greater than this threshold, the result will be considered as a text region.
+Any floating-point number greater than <code>0</code>. If not set, it will use the default value initialized by the pipeline (default is <code>0.6</code>).
+</td>
+<td><code>float</code></td>
+<td><code>0.6</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td>The expansion coefficient for text detection. This method is used to expand the text region. The larger the value, the greater the expanded area. Any floating-point number greater than <code>0</code>. If not set, it will use the default value initialized by the pipeline.
+</td>
+<td><code>float</code></td>
+<td><code>1.5</code></td>
+</tr>
+<tr>
+<td><code>text_det_input_shape</code></td>
+<td>The input shape for text detection. You can set 3 values representing C, H, W.</td>
+<td><code>str</code></td>
+<td>""</td>
+</tr>
+</tbody>
+</table>
+
+<li>Orientation Classifier Related</li>
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td>The name of the document orientation classification model. If not set, it will use the default model of the pipeline. When the name of the passed-in document orientation classification model is inconsistent with the configured name of the pipeline's default model, you need to specify the name of the passed-in model.</td>
+<td><code>str</code></td>
+<td><code>PP-LCNet_x1_0_doc_ori</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td>The directory path of the document orientation classification model. It can be omitted when setting <code>use_doc_orientation_classify = false</code>.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_name</code></td>
+<td>The name of the text line orientation classification model. If not set, it will use the default model of the pipeline. When the name of the passed-in text line orientation classification model is inconsistent with the configured name of the pipeline's default model, you need to specify the name of the passed-in model.</td>
+<td><code>str</code></td>
+<td><code>PP-LCNet_x1_0_textline_ori</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_dir</code></td>
+<td>The directory path of the text line orientation classification model. It can be omitted when setting <code>use_textline_orientation = false</code>.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_batch_size</code></td>
+<td>The batch size for the text line orientation model. If not set, it will use the default model of the pipeline.</td>
+<td><code>int</code></td>
+<td><code>6</code></td>
+</tr>
+</tbody>
+</table>
+
+<li>Text Recognition Model Related</li>
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td>The name of the text recognition model. If not set, it will use the default model of the pipeline. When the name of the passed-in text recognition model path is inconsistent with the configured name of the pipeline's default text recognition model, you need to specify the name of the passed-in model.</td>
+<td><code>str</code></td>
+<td><code>PP-OCRv5_server_rec</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td>The directory path of the text recognition model, required.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td>The batch size for the text recognition model. If not set, it will use the default value of the pipeline.</td>
+<td><code>int</code></td>
+<td><code>6</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td>The text recognition threshold. Text results with scores greater than this threshold will be retained. Any floating-point number greater than <code>0</code>.</td>
+<td><code>float</code></td>
+<td><code>0.0</code></td>
+</tr>
+<tr>
+<td><code>text_rec_input_shape</code></td>
+<td>The input shape for text recognition. You can set 3 values representing C, H, W.</td>
+<td><code>str</code></td>
+<td>""</td>
+</tr>
+</tbody>
+</table>
+
+</details>
+
+#### 2.3.1 Example of System Pipeline Call
+
+This section provides an example of a system pipeline call. Please refer to Section 2.1 to prepare the models. Assume the model directory structure is as follows:
+
+```
+models
+|--PP-LCNet_x1_0_doc_ori_infer
+|--UVDoc_infer
+|--PP-LCNet_x1_0_textline_ori_infer
+|--PP-OCRv5_server_det_infer
+|--PP-OCRv5_server_rec_infer
+```
+
+=== "Full Pipeline Serialization"
+
+    ```bash
+    ./build/ppocr ocr --input ./general_ocr_002.png --save_path ./output/  \
+    --doc_orientation_classify_model_dir models/PP-LCNet_x1_0_doc_ori_infer \
+    --doc_unwarping_model_dir models/UVDoc_infer \
+    --textline_orientation_model_dir models/PP-LCNet_x1_0_textline_ori_infer \
+    --text_detection_model_dir models/PP-OCRv5_server_det_infer \
+    --text_recognition_model_dir models/PP-OCRv5_server_rec_infer \
+    --device cpu
+    ```
+
+    Example Output (If `save_path` is specified, a standard JSON prediction result file and prediction result image will be generated under this path):
+
+    ```bash
+    {
+       "input_path": "./general_ocr_002.png",
+       "doc_preprocessor_res": {
+           "model_settings": {"use_doc_unwarping": true, "use_doc_orientation_classify": true},
+           "angle": 0
+        },
+       ...,
+       "dt_polys": [[[132, 6], [355, 6], [355, 64], [132, 64]],
+        [[424, 9], [689, 9], [689, 59], [424, 59]],
+         ...,
+        [[664, 8], [867, 4], [868, 55], [665, 60]],
+        [[31, 99], [173, 99], [173, 126], [31, 126]]],
+         ...,
+       "rec_texts": ["登机牌", "BOARDING", "GPASS", ..., ],
+       ...,
+    }        
+    ```
+
+=== "Text Detection + Textline Orientation Classification + Text Recognition"
+
+    ```bash
+    ./build/ppocr ocr --input ./general_ocr_002.png --save_path ./output/  \
+    --doc_orientation_classify_model_dir models/PP-LCNet_x1_0_doc_ori_infer \
+    --doc_unwarping_model_dir models/UVDoc_infer \
+    --textline_orientation_model_dir models/PP-LCNet_x1_0_textline_ori_infer \
+    --text_detection_model_dir models/PP-OCRv5_server_det_infer \
+    --text_recognition_model_dir models/PP-OCRv5_server_rec_infer \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False \
+    --device cpu
+    ```
+
+    Example Output (If `save_path` is specified, a standard JSON prediction result file and prediction result image will be generated under this path):
+
+    ```bash
+    {
+       "input_path": "./general_ocr_002.png",
+       ...,
+        "dt_polys": [[[0, 1], [334, 1], [334, 34], [0, 34]],
+        [[151, 21], [357, 16], [358, 72], [152, 76]],
+         ...,
+        [[675, 97], [740, 97], [740, 121], [675, 121]],
+        [[751, 97], [836, 94], [837, 115], [752, 119]],
+         ...,
+       "rec_texts": ["净小8866-", "登机牌", "BOARDING", "GPASS", ..., ],
+       ...,
+    }    
+    ```
+
+=== "Text Detection + Text Recognition"
+
+    ```bash
+    ./build/ppocr ocr --input ./general_ocr_002.png --save_path ./output/  \
+    --doc_orientation_classify_model_dir models/PP-LCNet_x1_0_doc_ori_infer \
+    --doc_unwarping_model_dir models/UVDoc_infer \
+    --textline_orientation_model_dir models/PP-LCNet_x1_0_textline_ori_infer \
+    --text_detection_model_dir models/PP-OCRv5_server_det_infer \
+    --text_recognition_model_dir models/PP-OCRv5_server_rec_infer \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False \
+    --use_textline_orientation False \
+    --device cpu
+    ```
+
+    Example Output (If `save_path` is specified, a standard JSON prediction result file and prediction result image will be generated under this path):
+
+    ```bash
+    {
+       "input_path": "./general_ocr_002.png",
+       ...,
+       "dt_polys": [[[0, 1], [334, 1], [334, 34], [0, 34]],
+        [[151, 21], [357, 16], [358, 72], [152, 76]],
+         ...,
+        [[61, 109], [194, 106], [194, 132], [61, 135]],
+        [[80, 138], [219, 136], [219, 162], [80, 164]],
+         ...,
+       "rec_texts": ["www.997788.com中国收藏热线","登机牌", "BOARDING", "GPASS", ..., ],
+       ...,
+    }    
+    ```
+
+The above sample code will generate the following text detection result image:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/ocr_res.png"/>
+
+If you want to view the text recognition result image, please refer to the **Visualizing Text Recognition Results** section later.
+
+#### 2.3.2 Example of Single Module Call
+
+=== "Document Image Orientation Classification"   
+
+    ```bash
+    ./build/ppocr doc_img_orientation_classification --input ./general_ocr_002.png --save_path ./output/  \
+    --doc_orientation_classify_model_dir models/PP-LCNet_x1_0_doc_ori_infer \
+    --device cpu 
+    ```
+
+    Example Output (if `save_path` is specified, standard JSON prediction result files and prediction result images will be generated at this path):
+
+    ```bash
+    {
+    "res": {
+        "input_path": {./general_ocr_002.png},
+        "class_ids": {0},
+        "scores": {0.926328},
+        "label_names": {0},
+    }
+    ```
+
+=== "Document Image Unwarping"
+
+    ```bash
+    ./build/ppocr text_image_unwarping --input ./general_ocr_002.png --save_path ./output/  \
+    --doc_unwarping_model_dir models/UVDoc_infer \
+    --device cpu 
+    ```
+
+    Example Output (if `save_path` is specified, standard JSON prediction result files and prediction result images will be generated at this path):
+
+    ```bash
+    {
+    "res": {
+        "input_path": {./general_ocr_002.png},
+        "doctr_img": {...}
+    }    
+    ```   
+
+=== "Text Line Orientation Classification"
+
+    ```bash
+    ./build/ppocr textline_orientation_classification --input ./general_ocr_002.png --save_path ./output/  \
+    --textline_orientation_model_dir models/PP-LCNet_x1_0_textline_ori_infer \
+    --device cpu 
+    ```
+
+    Example Output (if `save_path` is specified, standard JSON prediction result files and prediction result images will be generated at this path):
+
+    ```bash
+    {
+    "res": {
+        "input_path": {./general_ocr_002.png},
+        "class_ids": {0},
+        "scores": {0.719926},
+        "label_names": {0_degree},
+    }
+    ```      
+
+=== "Text Detection"
+
+    ```bash
+    ./build/ppocr text_detection --input ./general_ocr_002.png --save_path ./output/  \
+    --text_detection_model_dir models/PP-OCRv5_server_det_infer \
+    --device cpu 
+    ```
+
+    Example Output (if `save_path` is specified, standard JSON prediction result files and prediction result images will be generated at this path):
+
+    ```bash
+    {
+    "res": {
+        "input_path": {./general_ocr_002.png    },
+        "dt_polys": [
+            [[98, 456], [834, 441], [834, 466], [98, 480]],
+            [[344, 347], [662, 343], [662, 366], [344, 371]],
+            [[66, 341], [165, 337], [167, 363], [67, 367]],
+            ...,
+            [[0, 1], [331, 0], [332, 32], [0, 34]],
+        ]},
+        "dt_scores": [
+            0.812284, 0.8082, 0.848293, ..., 
+        ]
+      }
+    }
+    ```
+    
+=== "Text Recognition"
+
+    ```bash
+    ./build/ppocr text_recognition --input ./general_ocr_rec_001.png --save_path ./output/  \
+    --text_recognition_model_dir models/PP-OCRv5_server_rec_infer \ 
+    --device cpu 
+    ```
+
+    Example Output (if `save_path` is specified, standard JSON prediction result files and prediction result images will be generated at this path):
+
+    ```bash
+    {
+    "res": {
+        "input_path": {./general_ocr_rec_001.png },
+        "rec_text": {绿洲仕格维花园公寓 }
+        "rec_score": {0.982409 }
+    }
+    ```
+
+### 2.4 C++ API Integration
+
+The command-line interface is for quickly experiencing and viewing the results. Generally, in projects, integration through code is often required. You can achieve rapid inference in production lines with just a few lines of code. The inference code is as follows:
+
+Since the general OCR production line has many configuration parameters, a struct is used for parameter passing during instantiation. The naming rule for the struct is `pipeline_class_name + Params`. For example, the corresponding class name for the general OCR production line is `PaddleOCR`, and the struct is `PaddleOCRParams`.
+
+```c++
+#include "src/api/pipelines/ocr.h"
+
+
+int main(){
+    PaddleOCRParams params;
+    params.doc_orientation_classify_model_dir = "models/PP-LCNet_x1_0_doc_ori_infer"; // Path to the document orientation classification model.
+    params.doc_unwarping_model_dir = "models/UVDoc_infer"; // Path to the text image unwarping model.
+    params.textline_orientation_model_dir = "models/PP-LCNet_x1_0_textline_ori_infer"; // Path to the text line orientation classification model.
+    params.text_detection_model_dir = "models/PP-OCRv5_server_det_infer"; // Path to the text detection model.
+    params.text_recognition_model_dir = "models/PP-OCRv5_server_rec_infer"; // Path to the text recognition model.
+
+    // params.device = "gpu"; // Use GPU for inference. Ensure that the -DWITH_GPU=ON option is added during compilation; otherwise, CPU will be used.
+    // params.use_doc_orientation_classify = false; // Do not use the document orientation classification model.
+    // params.use_doc_unwarping = false; // Do not use the text image unwarping model.
+    // params.use_textline_orientation = false; // Do not use the text line orientation classification model.
+    // params.text_recognition_model_name = "PP-OCRv5_server_rec" // Use the PP-OCRv5_server_rec model for recognition.
+    // params.vis_font_dir = "your_vis_font_dir"; // When the -DUSE_FREETYPE=ON option is added during compilation, the corresponding ttf font file path must be provided.
+
+    auto infer = PaddleOCR(params);
+    auto outputs = infer.Predict("./general_ocr_002.png");
+    for (auto& output : outputs) {
+      output->Print();
+      output->SaveToImg("./output/");
+      output->SaveToJson("./output/");
+    }
+}
+```
+
+## 3. Extended Features
+
+### 3.1 Multilingual Text Recognition
+
+PP-OCRv5 also provides multilingual text recognition capabilities covering 39 languages, including Korean, Spanish, French, Portuguese, German, Italian, Russian, Thai, Greek, etc. The specific supported languages are as follows:
+
+<table>
+  <thead>
+    <tr>
+      <th>Model</th>
+      <th>Link</th>
+      <th>Supported Languages</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td>PP-OCRv5_server_rec</td>
+      <td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_rec_infer.tar">Inference Model</a></td>
+      <td>Simplified Chinese, Traditional Chinese, English, Japanese</td>
+    </tr>
+    <tr>
+      <td>PP-OCRv5_mobile_rec</td>
+      <td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_rec_infer.tar">Inference Model</a></td>
+      <td>Simplified Chinese, Traditional Chinese, English, Japanese</td>
+    </tr>
+    <tr>
+      <td>korean_PP-OCRv5_mobile_rec</td>
+      <td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/korean_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a></td>
+      <td>Korean, English</td>
+    </tr>
+    <tr>
+      <td>latin_PP-OCRv5_mobile_rec</td>
+      <td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/latin_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a></td>
+      <td>English, French, German, Afrikaans, Italian, Spanish, Bosnian, Portuguese, Czech, Welsh, Danish, Estonian, Irish, Croatian, Uzbek, Hungarian, Serbian (latin), Indonesian, Occitan, Icelandic, Lithuanian, Maori, Malay, Dutch, Norwegian, Polish, Slovak, Slovenian, Albanian, Swedish, Swahili, Tagalog, Turkish, Latin</td>
+    </tr>
+    <tr>
+      <td>eslav_PP-OCRv5_mobile_rec</td>
+      <td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/eslav_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a></td>
+      <td>Russian, Belarusian, Ukrainian, English</td>
+    </tr>
+    <tr>
+      <td>th_PP-OCRv5_mobile_rec</td>
+      <td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/th_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a></td>
+      <td>Thai, English</td>
+    </tr>
+    <tr>
+      <td>el_PP-OCRv5_mobile_rec</td>
+      <td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/el_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a></td>
+      <td>Greek, English</td>
+    </tr>
+    <tr>
+      <td>en_PP-OCRv5_mobile_rec</td>
+      <td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a></td>
+      <td>English</td>
+    </tr>
+  </tbody>
+</table>
+
+Simply pass the corresponding recognition model when using the pipeline or module. For example, to recognize French text using the text recognition module:
+
+```
+./build/ppocr text_recognition \
+--input ./french.png \
+--text_recognition_model_name latin_PP-OCRv5_mobile_rec \
+--text_recognition_model_dir latin_PP-OCRv5_mobile_rec_infer \
+--save_path ./output/
+```
+
+For more detailed information, refer to the [Introduction to PP-OCRv5 Multilingual Text Recognition](../../../algorithm/PP-OCRv5/PP-OCRv5_multi_languages.en.md).
+
+### 3.2 Visualize Text Recognition Results
+
+We use the FreeType module from the opencv_contrib 4.x version for font rendering. If you want to visualize text recognition results, you need to download the source code of both OpenCV and opencv_contrib and compile OpenCV with the FreeType module included. Make sure that the versions of the two are consistent when downloading the source code. The following instructions use opencv-4.7.0 and opencv_contrib-4.7.0 as examples:
+
+```bash
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/cpp/libs/opencv-4.7.0.tgz
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/cpp/libs/opencv_contrib-4.7.0.tgz
+tar -xf opencv-4.7.0.tgz
+tar -xf opencv_contrib-4.7.0.tgz
+```
+
+Install FreeType dependencies:
+
+```bash
+sudo apt-get update
+sudo apt-get install libfreetype6-dev libharfbuzz-dev
+```
+
+Follow these steps to compile OpenCV with FreeType support:
+
+- a. Add the following three options to the `tools/build_opencv.sh` script:
+    - `-DOPENCV_EXTRA_MODULES_PATH=your_opencv_contrib-4.7.0/modules/`
+    - `-DBUILD_opencv_freetype=ON`
+    - `-DWITH_FREETYPE=ON`
+- b. In `tools/build_opencv.sh`, set `root_path` to the absolute path of your opencv-4.7.0 source code.
+- c. In `tools/build_opencv.sh`, set `install_path` (default: `${root_path}/opencv4`). This path will be used as the OpenCV library path when compiling the prediction demo later.
+- d. After configuration, run the following command to compile OpenCV:
+
+    ```bash
+    sh tools/build_opencv.sh
+    ```
+
+- e. In `tools/build.sh`, add `-DUSE_FREETYPE=ON` to enable text rendering and specify the ttf font file path with `--vis_font_dir your_ttf_path`. Then compile the prediction demo with:
+
+    ```bash
+    sh tools/build.sh
+    ```
+
+After compiling and running the prediction demo, you should see visualized text recognition results like this:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/ocr_res_with_freetype.png"/>
+
+## 4. FAQ
+
+1. If you encounter the error `Model name mismatch, please input the correct model dir. model dir is xxx, but model name is xxx`, it means the specified model name doesn't match the provided model. For example, if the text recognition model expects `PP-OCRv5_server_rec` but you provided `PP-OCRv5_mobile_rec`.
+Solution: Adjust either the model name or the provided model. In the example above, you can specify `--text_recognition_model_name PP-OCRv5_mobile_rec` to match the provided model.
+
+2. If you see garbled text in the Windows console, it may be due to the console's default character encoding (GBK). Change it to UTF-8 encoding.
diff --git a/docs/version3.x/inference_deployment/local_inference/cpp/OCR.md b/docs/version3.x/inference_deployment/local_inference/cpp/OCR.md
new file mode 100644
index 0000000..f07af48
--- /dev/null
+++ b/docs/version3.x/inference_deployment/local_inference/cpp/OCR.md
@@ -0,0 +1,981 @@
+# 通用 OCR 产线 C++ 部署 - Linux
+
+- [1. 环境准备](#1)
+    - [1.1 编译 OpenCV 库](#11-opencv)
+    - [1.2 编译 Paddle Inference](#12-paddle-inference)
+- [2. 开始运行](#2)
+    - [2.1 编译预测 demo](#21-demo)
+    - [2.2 准备模型](#22)
+    - [2.3 运行预测 demo](#23-demo)
+    - [2.4 C++ API 集成](#24-c-api)
+- [3. 拓展功能](#3)
+    - [3.1 多语种文字识别](#31)
+    - [3.2 可视化文本识别结果](#32)
+- [4. FAQ](#4-faq)
+
+本章节介绍通用 OCR 产线 C++ 部署方法。通用 OCR 产线由以下5个模块组成：
+
+1. 文档图像方向分类模块（可选）
+2. 文本图像矫正模块 (可选)
+3. 文本行方向分类模块（可选）
+4. 文本检测模块
+5. 文本识别模块
+
+下面将介绍如何在 Linux (CPU/GPU) 环境下配置 C++ 环境并完成通用 OCR 产线部署。
+
+- 备注
+    - Windows 环境具体编译方法请参考 [Windows 编译教程](./OCR_windows.md)，在编译完成后，后续运行 demo 的指令与 Linux 一致。
+
+## 1. 环境准备
+
+- **本章节编译运行时用到的源代码位于 [PaddleOCR/deploy/cpp_infer](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/deploy/cpp_infer) 目录下。**
+
+- Linux 环境。
+    - gcc   8.2（编译使用 Paddle Inference GPU 版本时，gcc>=11.2）
+    - cmake 3.18
+
+### 1.1 编译 OpenCV 库
+
+目前仅支持 OpenCV 4.x 版本。下面以 OpenCV 4.7.0 为例。
+
+1. 执行如下命令下载 OpenCV 源码：
+
+```bash
+cd deploy/cpp_infer
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/cpp/libs/opencv-4.7.0.tgz
+tar -xf opencv-4.7.0.tgz
+```
+
+2. 配置并编译 OpenCV 库：
+
+- a. 在 `tools/build_opencv.sh` 脚本中，将 `root_path` 设置为 opencv-4.7.0 源码的绝对路径。
+- b. 设置 `install_path`，如默认的 `${root_path}/opencv4`。`install_path` 在后续编译预测 demo 时，将作为 OpenCV 库的路径使用。
+- c. 配置完成后，运行以下命令进行 OpenCV 的编译：
+
+    ```bash
+    sh tools/build_opencv.sh
+    ```
+
+### 1.2 编译 Paddle Inference
+
+可以选择直接下载预编译包或者手动编译源码。
+
+#### 1.2.1 直接下载预编译包（推荐）
+
+[Paddle Inference 官网](https://www.paddlepaddle.org.cn/inference/v3.0/guides/install/download_lib.html) 上提供了 Linux 预测库，可以在官网查看并选择合适的预编译包。
+
+下载之后解压:
+
+```shell
+tar -xvf paddle_inference.tgz
+```
+
+最终会在当前的文件夹中生成 `paddle_inference/` 的子文件夹。
+
+#### 1.2.2 源码编译预测库
+
+可以选择通过源码自行编译预测库，源码编译可灵活配置各类功能和依赖，以适应不同的硬件和软件环境。详细步骤请参考 [Linux 下源码编译](https://www.paddlepaddle.org.cn/inference/v3.0/guides/install/compile/source_compile_under_Linux.html)。
+
+## 2. 开始运行
+
+### 2.1 编译预测 demo
+
+在编译预测demo前，请确保您已经按照 1.1 和 1.2 节编译好 OpenCV 库和 Paddle Inference 预测库。
+
+修改 tools/build.sh 中的配置后，执行以下命令进行编译：
+
+```shell
+sh tools/build.sh
+```
+
+相关配置参数详细介绍如下：
+
+<table>
+<tr>
+<th>参数</th>
+<th>说明</th>
+<th>默认值</th>
+</tr>
+<tr>
+<td><code>OPENCV_DIR</code></td>
+<td>OpenCV编译安装的路径（如上述编译 OpenCV 时的 <code>install_path</code> ，必填。</td>
+<td></td>
+</tr>
+<tr>
+<td><code>LIB_DIR</code></td>
+<td>下载的 <code>Paddle Inference</code> 的预编译包或手动编译生成的Paddle Inference库路径（如 <code>build/paddle_inference_install_dir</code> 文件夹），必填。</td>
+<td></td>
+</tr>
+<tr>
+<td><code>CUDA_LIB_DIR</code></td>
+<td>CUDA库文件路径，通常为<code>/usr/local/cuda/lib64</code>。当Paddle Inference库为GPU版本且设置 <code>-DWITH_GPU=ON</code> 时需要设置该参数。</td>
+<td></td>
+</tr>
+<tr>
+<td><code>CUDNN_LIB_DIR</code></td>
+<td>cuDNN 库文件路径，通常为 <code>/usr/lib/x86_64-linux-gnu/</code> 。当 Paddle Inference 库为 GPU 版本且设置 <code>-DWITH_GPU=ON</code> 时需要设置该参数。</td>
+<td></td>
+</tr>
+<tr>
+<td><code>WITH_GPU</code></td>
+<td>当设置为 ON 时可以进行 GPU 版本 demo 的编译，要求 Paddle Inference 库为 GPU 版本。</td>
+<td>OFF</td>
+</tr>
+</table>
+
+**注意：以上路径需要填绝对路径。**
+
+### 2.2 准备模型
+
+可以直接下载 PaddleOCR 提供的推理模型：
+
+<details>
+<summary><b>文档图像方向分类模块（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>Top-1 Acc（%）</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">推理模型</a></td>
+<td>99.06</td>
+<td>7</td>
+<td>基于 PP-LCNet_x1_0 的文档图像分类模型，含有四个类别，即0度，90度，180度，270度</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>文本图像矫正模块（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>CER </th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">推理模型</a></td>
+<td>0.179</td>
+<td>30.3</td>
+<td>高精度文本图像矫正模型</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>文本行方向分类模块（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>模型下载链接</th>
+<th>Top-1 Acc（%）</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<td>PP-LCNet_x1_0_textline_ori (默认)</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_textline_ori_infer.tar">推理模型</a></td>
+<td>99.42</td>
+<td>6.5</td>
+<td>基于 PP-LCNet_x1_0 的文本行分类模型，含有两个类别，即0度，180度</td>
+</tr>
+<tr>
+<td>PP-LCNet_x0_25_textline_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x0_25_textline_ori_infer.tar">推理模型</a></td>
+<td>98.85</td>
+<td>0.96</td>
+<td>基于 PP-LCNet_x0_25 的文本行分类模型，含有两个类别，即0度，180度</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>文本检测模块：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>检测Hmean（%）</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv5_server_det (默认)</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_det_infer.tar">推理模型</a></td>
+<td>83.8</td>
+<td>84.3</td>
+<td>PP-OCRv5 的服务端文本检测模型，精度更高，适合在性能较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_infer.tar">推理模型</a></td>
+<td>79.0</td>
+<td>4.7</td>
+<td>PP-OCRv5 的移动端文本检测模型，效率更高，适合在端侧设备部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_det_infer.tar">推理模型</a></td>
+<td>69.2</td>
+<td>109</td>
+<td>PP-OCRv4 的服务端文本检测模型，精度更高，适合在性能较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar">推理模型</a></td>
+<td>63.8</td>
+<td>4.7</td>
+<td>PP-OCRv4 的移动端文本检测模型，效率更高，适合在端侧设备部署</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>文本识别模块：</b></summary>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec (默认)</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_rec_infer.tar">推理模型</a></td>
+<td>86.38</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec 是新一代文本识别模型。该模型致力于以单一模型高效、精准地支持简体中文、繁体中文、英文、日文四种主要语言，以及手写、竖版、拼音、生僻字等复杂文本场景的识别。在保持识别效果的同时，兼顾推理速度和模型鲁棒性，为各种场景下的文档理解提供高效、精准的技术支撑。</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_rec_infer.tar">推理模型</a></td>
+<td>81.29</td>
+<td>16</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_doc_infer.tar">推理模型</a></td>
+<td>86.58</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc 是在 PP-OCRv4_server_rec 的基础上，在更多中文文档数据和PP-OCR训练数据的混合数据训练而成，增加了部分繁体字、日文、特殊字符的识别能力，可支持识别的字符为1.5万+，除文档相关的文字识别能力提升外，也同时提升了通用文字的识别能力</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">推理模型</a></td>
+<td>78.74</td>
+<td>10.5</td>
+<td>PP-OCRv4 的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">推理模型</a></td>
+<td>85.19</td>
+<td>173</td>
+<td>PP-OCRv4 的服务器端模型，推理精度高，可以部署在多种不同的服务器上</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+也可以参考各模块的模型导出章节，如[文本检测模块-模型导出](../../../module_usage/text_detection.md#44)，将训练好的模型导出为推理模型。
+
+模型的目录结构一般如下所示：
+
+```
+PP-OCRv5_mobile_det
+|--inference.pdiparams
+|--inference.json
+|--inference.yml
+```
+
+### 2.3 运行预测 demo
+
+在本地使用通用 OCR 产线 C++前，请先成功编译预测 demo。编译后，可通过命令行体验或调用 api 进行二次开发并重新编译生成应用程序。
+
+**请注意，如果在执行过程中遇到程序失去响应、程序异常退出、内存资源耗尽、推理速度极慢等问题，请尝试参考文档调整配置，例如关闭不需要使用的功能或使用更轻量的模型。**
+
+本 demo 支持系统串联调用，也支持单个模块的调用，运行以下代码前，请您下载[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png)到本地：
+
+运行方式：
+
+```shell
+./build/ppocr <pipeline_or_module> [--param1] [--param2] [...]
+```
+
+常用参数如下：
+
+<li>输入输出相关</li>
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td>本地待预测图片，必填。仅支持<code>jpg</code>，<code>png</code>, <code>jpeg</code>,<code>bmp</code>格式的图像。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td>指定推理结果文件保存的路径，该路径下会保存推理结果的 json 文件和预测结果图片。</td>
+<td><code>str</code></td>
+<td><code>./output</code></td>
+</tr>
+</tbody>
+</table>
+
+<details><summary><b>点击展开查看更多参数的详细说明</b></summary>
+
+<li>通用参数</li>
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>device</code></td>
+<td>用于推理的设备。支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+</ul>如果不设置，将默认使用产线初始化的该参数值，初始化时，如果编译时添加<code>-DWITH_GPU=ON</code>，则会优先使用本地的 GPU 0号设备，否则，将使用 CPU 设备。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td>计算精度，如 <code>fp32</code>、<code>fp16</code>。</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>是否启用 MKL-DNN 加速推理。如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>true</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td>PaddleInference CPU 加速库线程数量</td>
+<td><code>int</code></td>
+<td><code>8</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td>PaddleX产线配置文件路径。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+</tbody>
+</table>
+
+<li>模块开关</li>
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td>是否加载并使用文档方向分类模块。如果不设置，将使用产线初始化的该参数值，默认初始化为<code>true</code>。</td>
+<td><code>bool</code></td>
+<td><code>true</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td>是否加载并使用文本图像矫正模块。如果不设置，将使用产线初始化的该参数值，默认初始化为<code>true</code>。</td>
+<td><code>bool</code></td>
+<td><code>true</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td>是否加载并使用文本行方向模块。如果不设置，将使用产线初始化的该参数值，默认初始化为<code>true</code>。</td>
+<td><code>bool</code></td>
+<td><code>true</code></td>
+</tr>
+</tbody>
+</table>
+
+<li>检测模型相关</li>
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td>文本检测模型的名称。如果不设置，将会使用产线默认模型。当传入文本检测模型路径的模型名称与产线默认文本识别模型名称配置不一致时，需指定传入模型的名称。</td>
+<td><code>str</code></td>
+<td><code>PP-OCRv5_server_det</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td>文本检测模型的目录路径，必填。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td>文本检测的图像边长限制。
+大于 <code>0</code> 的任意整数。如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>64</code>。
+</td>
+<td><code>int</code></td>
+<td><code>64</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td>文本检测的边长度限制类型。支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>。如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>min</code>。
+</td>
+<td><code>str</code></td>
+<td><code>min</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td>文本检测像素阈值，输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。
+大于<code>0</code>的任意浮点数。如果不设置，将使用产线初始化的该参数值。
+</td>
+<td><code>float</code></td>
+<td><code>0.3</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td>文本检测框阈值，检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。
+大于 <code>0</code> 的任意浮点数。如果不设置，将使用产线初始化的该参数值（默认为 <code>0.6</code>）。
+</td>
+<td><code>float</code></td>
+<td><code>0.6</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td>文本检测扩张系数，使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。大于 <code>0</code> 的任意浮点数。如果不设置，将使用产线初始化的该参数值。
+</td>
+<td><code>float</code></td>
+<td><code>1.5</code></td>
+</tr>
+<tr>
+<td><code>text_det_input_shape</code></td>
+<td>文本检测的输入形状，您可以设置3个值代表C，H，W。</td>
+<td><code>str</code></td>
+<td>""</td>
+</tr>
+</tbody>
+</table>
+
+<li>方向分类器相关</li>
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td>文档方向分类模型的名称。如果不设置，将会使用产线默认模型。当传入文档方向分类模型与产线默认模型不一致时，需指定传入模型的名称。</td>
+<td><code>str</code></td>
+<td><code>PP-LCNet_x1_0_doc_ori</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td>文档方向分类模型的目录路径。当设置<code>use_doc_orientation_classify = false</code>时，可不添加。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_name</code></td>
+<td>文本行方向分类模型的名称。如果不设置，将会使用产线默认模型。当传入文本行方向分类模型与产线默认模型不一致时，需指定传入模型的名称。</td>
+<td><code>str</code></td>
+<td><code>PP-LCNet_x1_0_textline_ori</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_dir</code></td>
+<td>文本行方向分类模型的目录路径。当设置<code>use_textline_orientation = false</code>时，可不添加。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_batch_size</code></td>
+<td>文本行方向模型的batch size。如果不设置，将会使用产线默认模型。</td>
+<td><code>int</code></td>
+<td><code>6</code></td>
+</tr>
+</tbody>
+</table>
+
+<li>文字识别模型相关</li>
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td>文本识别模型的名称。如果不设置，将会使用产线默认模型。当传入文本识别模型路径的模型名称与产线默认文本识别模型名称配置不一致时，需指定传入模型的名称。</td>
+<td><code>str</code></td>
+<td><code>PP-OCRv5_server_rec</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td>文本识别模型的目录路径，必填。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td>文本识别模型的batch size。如果不设置，将会使用产线默认值。</td>
+<td><code>int</code></td>
+<td><code>6</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td>文本识别阈值，得分大于该阈值的文本结果会被保留。大于<code>0</code>的任意浮点数。</td>
+<td><code>float</code></td>
+<td><code>0.0</code></td>
+</tr>
+<tr>
+<td><code>text_rec_input_shape</code></td>
+<td>文本识别的输入形状，您可以设置3个值代表C，H，W。</td>
+<td><code>str</code></td>
+<td>""</td>
+</tr>
+</tbody>
+</table>
+
+</details>
+
+#### 2.3.1 调用示例-系统串联调用
+
+本节提供了系统串联调用的调用示例，请参考 2.1 节准备好模型，假设模型的目录结构如下所示：
+
+```
+models
+|--PP-LCNet_x1_0_doc_ori_infer
+|--UVDoc_infer
+|--PP-LCNet_x1_0_textline_ori_infer
+|--PP-OCRv5_server_det_infer
+|--PP-OCRv5_server_rec_infer
+```
+
+=== "全模块串联"
+
+    ```bash
+    ./build/ppocr ocr --input ./general_ocr_002.png --save_path ./output/  \
+    --doc_orientation_classify_model_dir models/PP-LCNet_x1_0_doc_ori_infer \
+    --doc_unwarping_model_dir models/UVDoc_infer \
+    --textline_orientation_model_dir models/PP-LCNet_x1_0_textline_ori_infer \
+    --text_detection_model_dir models/PP-OCRv5_server_det_infer \
+    --text_recognition_model_dir models/PP-OCRv5_server_rec_infer \
+    --device cpu
+    ```
+
+    输出示例（若指定了 `save_path`，则会在该路径下生成标准的 json 预测结果文件和预测结果图片）：
+
+    ```bash
+    {
+       "input_path": "./general_ocr_002.png",
+       "doc_preprocessor_res": {
+           "model_settings": {"use_doc_unwarping": true, "use_doc_orientation_classify": true},
+           "angle": 0
+        },
+       ...,
+       "dt_polys": [[[132, 6], [355, 6], [355, 64], [132, 64]],
+        [[424, 9], [689, 9], [689, 59], [424, 59]],
+         ...,
+        [[664, 8], [867, 4], [868, 55], [665, 60]],
+        [[31, 99], [173, 99], [173, 126], [31, 126]]],
+         ...,
+       "rec_texts": ["登机牌", "BOARDING", "GPASS", ..., ],
+       ...,
+    }        
+    ```
+
+=== "文本检测+文本行方向分类+文本识别"
+
+    ```bash
+    ./build/ppocr ocr --input ./general_ocr_002.png --save_path ./output/  \
+    --textline_orientation_model_dir models/PP-LCNet_x1_0_textline_ori_infer \
+    --text_detection_model_dir models/PP-OCRv5_server_det_infer \
+    --text_recognition_model_dir models/PP-OCRv5_server_rec_infer \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False \
+    --device cpu
+    ```
+
+    输出示例（若指定了 `save_path`，则会在该路径下生成标准的 json 预测结果文件和预测结果图片）：
+
+    ```bash
+    {
+       "input_path": "./general_ocr_002.png",
+       ...,
+        "dt_polys": [[[0, 1], [334, 1], [334, 34], [0, 34]],
+        [[151, 21], [357, 16], [358, 72], [152, 76]],
+         ...,
+        [[675, 97], [740, 97], [740, 121], [675, 121]],
+        [[751, 97], [836, 94], [837, 115], [752, 119]],
+         ...,
+       "rec_texts": ["净小8866-", "登机牌", "BOARDING", "GPASS", ..., ],
+       ...,
+    }    
+    ```
+
+=== "文本检测+文本识别"
+
+    ```bash
+    ./build/ppocr ocr --input ./general_ocr_002.png --save_path ./output/  \
+    --text_detection_model_dir models/PP-OCRv5_server_det_infer \
+    --text_recognition_model_dir models/PP-OCRv5_server_rec_infer \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False \
+    --use_textline_orientation False \
+    --device cpu
+    ```
+
+    输出示例（若指定了 `save_path`，则会在该路径下生成标准的 json 预测结果文件和预测结果图片）：
+
+    ```bash
+    {
+       "input_path": "./general_ocr_002.png",
+       ...,
+       "dt_polys": [[[0, 1], [334, 1], [334, 34], [0, 34]],
+        [[151, 21], [357, 16], [358, 72], [152, 76]],
+         ...,
+        [[61, 109], [194, 106], [194, 132], [61, 135]],
+        [[80, 138], [219, 136], [219, 162], [80, 164]],
+         ...,
+       "rec_texts": ["www.997788.com中国收藏热线","登机牌", "BOARDING", "GPASS", ..., ],
+       ...,
+    }    
+    ```
+
+以上示例代码会生成如下文本检测结果图：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/ocr_res.png"/>
+
+若需要查看文本识别结果图，请参考后文 **可视化文本识别结果** 小节。
+
+#### 2.3.2 调用示例-单模块调用
+
+=== "文档图像方向分类"   
+
+    ```bash
+    ./build/ppocr doc_img_orientation_classification --input ./general_ocr_002.png --save_path ./output/  \
+    --doc_orientation_classify_model_dir models/PP-LCNet_x1_0_doc_ori_infer \
+    --device cpu 
+    ```
+
+    输出示例（若指定了 `save_path`，则会在该路径下生成标准的 json 预测结果文件和预测结果图片）：
+
+    ```bash
+    {
+    "res": {
+        "input_path": {./general_ocr_002.png},
+        "class_ids": {0},
+        "scores": {0.926328},
+        "label_names": {0},
+    }
+    ```
+
+=== "文档图像矫正"
+
+    ```bash
+    ./build/ppocr text_image_unwarping --input ./general_ocr_002.png --save_path ./output/  \
+    --doc_unwarping_model_dir models/UVDoc_infer \
+    --device cpu 
+    ```
+
+    输出示例（若指定了 `save_path`，则会在该路径下生成标准的 json 预测结果文件和预测结果图片）：
+
+    ```bash
+    {
+    "res": {
+        "input_path": {./general_ocr_002.png},
+        "doctr_img": {...}
+    }    
+    ```   
+
+=== "文本行方向分类"
+
+    ```bash
+    ./build/ppocr textline_orientation_classification --input ./general_ocr_002.png --save_path ./output/  \
+    --textline_orientation_model_dir models/PP-LCNet_x1_0_textline_ori_infer \
+    --device cpu 
+    ```
+
+    输出示例（若指定了 `save_path`，则会在该路径下生成标准的 json 预测结果文件和预测结果图片）：
+
+    ```bash
+    {
+    "res": {
+        "input_path": {./general_ocr_002.png},
+        "class_ids": {0},
+        "scores": {0.719926},
+        "label_names": {0_degree},
+    }
+    ```      
+
+=== "文本检测"
+
+    ```bash
+    ./build/ppocr text_detection --input ./general_ocr_002.png --save_path ./output/  \
+    --text_detection_model_dir models/PP-OCRv5_server_det_infer \
+    --device cpu 
+    ```
+
+    输出示例（若指定了 `save_path`，则会在该路径下生成标准的 json 预测结果文件和预测结果图片）：
+
+    ```bash
+    {
+    "res": {
+        "input_path": {./general_ocr_002.png    },
+        "dt_polys": [
+            [[98, 456], [834, 441], [834, 466], [98, 480]],
+            [[344, 347], [662, 343], [662, 366], [344, 371]],
+            [[66, 341], [165, 337], [167, 363], [67, 367]],
+            ...,
+            [[0, 1], [331, 0], [332, 32], [0, 34]],
+        ]},
+        "dt_scores": [
+            0.812284, 0.8082, 0.848293, ..., 
+        ]
+      }
+    }
+    ```
+    
+=== "文本识别"
+
+    ```bash
+    ./build/ppocr text_recognition --input ./general_ocr_rec_001.png --save_path ./output/  \
+    --text_recognition_model_dir models/PP-OCRv5_server_rec_infer \ 
+    --device cpu 
+    ```
+
+    输出示例（若指定了 `save_path`，则会在该路径下生成标准的 json 预测结果文件和预测结果图片）：
+
+    ```bash
+    {
+    "res": {
+        "input_path": {./general_ocr_rec_001.png },
+        "rec_text": {绿洲仕格维花园公寓 }
+        "rec_score": {0.982409 }
+    }
+    ```
+
+### 2.4 C++ API 集成
+
+命令行方式是为了快速体验查看效果，一般来说，在项目中，往往需要通过代码集成，您可以通过几行代码即可完成产线的快速推理，推理代码如下：
+
+由于通用 OCR 产线配置参数较多，故采用结构体传参进行实例化，结构体命名规则为 `pipeline_class_name  + Params`，如通用 OCR 产线对应的类名为 `PaddleOCR` ，结构体为 `PaddleOCRParams`。
+
+```c++
+#include "src/api/pipelines/ocr.h"
+
+
+int main(){
+    PaddleOCRParams params;
+    params.doc_orientation_classify_model_dir = "models/PP-LCNet_x1_0_doc_ori_infer"; // 文档方向分类模型路径。
+    params.doc_unwarping_model_dir = "models/UVDoc_infer"; // 文本图像矫正模型路径。
+    params.textline_orientation_model_dir = "models/PP-LCNet_x1_0_textline_ori_infer"; // 文本行方向分类模型路径。
+    params.text_detection_model_dir = "models/PP-OCRv5_server_det_infer"; // 文本检测模型路径
+    params.text_recognition_model_dir = "models/PP-OCRv5_server_rec_infer"; // 文本识别模型路径
+
+    // params.device = "gpu"; // 推理时使用GPU。请确保编译时添加 -DWITH_GPU=ON 选项，否则使用CPU。
+    // params.use_doc_orientation_classify = false;  // 不使用文档方向分类模型。
+    // params.use_doc_unwarping = false; // 不使用文本图像矫正模型。
+    // params.use_textline_orientation = false; // 不使用文本行方向分类模型。
+    // params.text_detection_model_name = "PP-OCRv5_server_det"; // 使用 PP-OCRv5_server_det 模型进行检测。
+    // params.text_recognition_model_name = "PP-OCRv5_server_rec"; // 使用 PP-OCRv5_server_rec 模型进行识别。
+    // params.vis_font_dir = "your_vis_font_dir"; // 当编译时添加 -DUSE_FREETYPE=ON 选项，必须提供相应 ttf 字体文件路径。
+
+    auto infer = PaddleOCR(params);
+    auto outputs = infer.Predict("./general_ocr_002.png");
+    for (auto& output : outputs) {
+      output->Print();
+      output->SaveToImg("./output/");
+      output->SaveToJson("./output/");
+    }
+}
+```
+
+## 3. 拓展功能
+
+### 3.1 多语种文字识别
+
+PP-OCRv5 还提供了覆盖 39 种语言的多语种文字识别能力，包括韩文、西班牙文、法文、葡萄牙文、德文、意大利文、俄罗斯文、泰文、希腊文等，具体支持语种如下：
+
+<table>
+  <thead>
+    <tr>
+      <th>模型</th>
+      <th>链接</th>
+      <th>支持语种</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td>PP-OCRv5_server_rec</td>
+      <td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_rec_infer.tar">推理模型</a></td>
+      <td>简体中文、繁体中文、英文、日文</td>
+    </tr>
+    <tr>
+      <td>PP-OCRv5_mobile_rec</td>
+      <td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_rec_infer.tar">推理模型</a></td>
+      <td>简体中文、繁体中文、英文、日文</td>
+    </tr>
+    <tr>
+      <td>korean_PP-OCRv5_mobile_rec</td>
+      <td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/korean_PP-OCRv5_mobile_rec_infer.tar">推理模型</a></td>
+      <td>韩文、英文</td>
+    </tr>
+    <tr>
+      <td>latin_PP-OCRv5_mobile_rec</td>
+      <td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/latin_PP-OCRv5_mobile_rec_infer.tar">推理模型</a></td>
+      <td>英文、法文、德文、南非荷兰文、意大利文、西班牙文、波斯尼亚文、葡萄牙文、捷克文、威尔士文、丹麦文、爱沙尼亚文、爱尔兰文、克罗地亚文、乌兹别克文、匈牙利文、塞尔维亚文（latin）、印度尼西亚文、欧西坦文、冰岛文、立陶宛文、毛利文、马来文、荷兰文、挪威文、波兰文、斯洛伐克文、斯洛文尼亚文、阿尔巴尼亚文、瑞典文、西瓦希里文、塔加洛文、土耳其文、拉丁文</td>
+    </tr>
+    <tr>
+      <td>eslav_PP-OCRv5_mobile_rec</td>
+      <td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/eslav_PP-OCRv5_mobile_rec_infer.tar">推理模型</a></td>
+      <td>俄罗斯文、白俄罗斯文、乌克兰文、英文</td>
+    </tr>
+    <tr>
+      <td>th_PP-OCRv5_mobile_rec</td>
+      <td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/th_PP-OCRv5_mobile_rec_infer.tar">推理模型</a></td>
+      <td>泰文、英文</td>
+    </tr>
+    <tr>
+      <td>el_PP-OCRv5_mobile_rec</td>
+      <td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/el_PP-OCRv5_mobile_rec_infer.tar">推理模型</a></td>
+      <td>希腊文、英文</td>
+    </tr>
+    <tr>
+      <td>en_PP-OCRv5_mobile_rec</td>
+      <td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv5_mobile_rec_infer.tar">推理模型</a></td>
+      <td>英文</td>
+    </tr>
+  </tbody>
+</table>
+
+在使用产线或模块时传入对应的识别模型即可，如使用文本识别模块进行法文识别：
+
+```
+./build/ppocr text_recognition \
+--input ./french.png \
+--text_recognition_model_name latin_PP-OCRv5_mobile_rec \
+--text_recognition_model_dir latin_PP-OCRv5_mobile_rec_infer \
+--save_path ./output/
+```
+
+更多详细说明可参考 [PP-OCRv5多语种文字识别介绍](../../../algorithm/PP-OCRv5/PP-OCRv5_multi_languages.md)。
+
+### 3.2 可视化文本识别结果
+
+我们使用 4.x 版本的 opencv_contrib 模块中的 FreeType 进行字体渲染，如果想要可视化文本识别结果，需要下载 OpenCV 和 opencv_contrib 的源码并编译包含 FreeType 模块的 OpenCV。下载源码时需确保两者的版本一致。以下以 opencv-4.7.0 和 opencv_contrib-4.7.0 为例进行说明：
+
+```bash
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/cpp/libs/opencv-4.7.0.tgz
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/cpp/libs/opencv_contrib-4.7.0.tgz
+tar -xf opencv-4.7.0.tgz
+tar -xf opencv_contrib-4.7.0.tgz
+```
+
+安装FreeType依赖库
+
+```bash
+sudo apt-get update
+sudo apt-get install libfreetype6-dev libharfbuzz-dev
+```
+
+编译包含 FreeType 模块的 OpenCV 的步骤如下：
+
+- a. 在 `tools/build_opencv.sh` 脚本中增加如下三个选项：
+    - -DOPENCV_EXTRA_MODULES_PATH=your_opencv_contrib-4.7.0/modules/
+    - -DBUILD_opencv_freetype=ON
+    - -DWITH_FREETYPE=ON
+- b. 在 `tools/build_opencv.sh` 脚本中，将 `root_path` 设置为 opencv-4.7.0 源码的绝对路径。
+- c. 在 `tools/build_opencv.sh` 脚本中，设置 `install_path`，如默认的 `${root_path}/opencv4`。`install_path` 在后续编译预测 demo 时，将作为 OpenCV 库的路径使用。
+- d. 配置完成后，运行以下命令进行 OpenCV 的编译：
+
+    ```bash
+    sh tools/build_opencv.sh
+    ```
+
+- e. 在 `tools/build.sh` 设置 `-DUSE_FREETYPE=ON` 开启文字渲染功能，设置 `--vis_font_dir your_ttf_path` 提供相应 ttf 字体文件路径。运行以下命令进行预测 demo 的编译：
+
+    ```bash
+    sh tools/build.sh
+    ```
+
+编译并运行预测 demo 可以得到如下可视化文本识别结果：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/ocr_res_with_freetype.png"/>
+
+## 4. FAQ
+
+1. 如果遇到 `Model name mismatch, please input the correct model dir. model dir is xxx, but model name is xxx` 的报错，说明指定的模型名称和传入模型不匹配。比如文本识别模型指定名称是 `PP-OCRv5_server_rec `，但传入模型是 `PP-OCRv5_mobile_rec`。
+解决：需要调整模型名称或传入的模型。例如上述例子，可以使用 `--text_recognition_model_name PP-OCRv5_mobile_rec` 指定和传入模型匹配的模型名称。
+
+2. 在 Windows 的控制台中输出出现乱码，原因可能是 Windows 控制台的字符编码是 GBK，请设置为 UTF-8 编码。
diff --git a/docs/version3.x/inference_deployment/local_inference/cpp/OCR_windows.en.md b/docs/version3.x/inference_deployment/local_inference/cpp/OCR_windows.en.md
new file mode 100644
index 0000000..35c7cfe
--- /dev/null
+++ b/docs/version3.x/inference_deployment/local_inference/cpp/OCR_windows.en.md
@@ -0,0 +1,240 @@
+# C++ Local Deployment for General OCR Pipeline - Windows
+
+- [1. Environment Preparation](#1-environment-preparation)
+    - [1.1 Compile OpenCV Library](#11-compile-opencv-library)
+    - [1.2 Compile Paddle Inference](#12-compile-paddle-inference)
+- [2. Getting Started](#2-getting-started)
+    - [2.1 Compile Prediction Demo](#21-compile-prediction-demo)
+    - [2.2 Prepare the Model](#22-prepare-the-model)
+    - [2.3 Run Prediction Demo](#23-run-the-prediction-demo)
+    - [2.4 C++ API Integration](#24-c-api-integration)
+- [3. Extended Features](#3-extended-features)
+    - [3.1 Multilingual Text Recognition](#31-multilingual-text-recognition)
+    - [3.2 Visualize Text Recognition Results](#32-visualize-text-recognition-results)
+
+## 1. Environment Preparation
+
+- **The source code used in this compilation and runtime section is located in the [PaddleOCR/deploy/cpp_infer](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/deploy/cpp_infer) directory.**
+
+- Windows Environment:
+    - Visual Studio 2022
+    - CMake 3.29
+
+### 1.1 Compile OpenCV Library
+
+You can choose to directly download a pre-compiled package or manually compile the source code.
+
+#### 1.1.1 Directly Download Pre-compiled Package (Recommended)
+
+Download the .exe pre-compiled package for Windows from the [OpenCV Official Website](https://opencv.org/releases/). Running it will automatically extract the pre-compiled OpenCV library and related folders.
+
+Taking OpenCV 4.7.0 as an example, download [opencv-4.7.0-windows.exe](https://github.com/opencv/opencv/releases/download/4.7.0/opencv-4.7.0-windows.exe). After running it, a `opencv/` subfolder will be generated in the current folder, where `opencv/build` contains the pre-compiled library. This will be used as the path for the OpenCV installation library when compiling the universal OCR pipeline prediction demo later.
+
+#### 1.1.2 Compile from Source Code
+
+First, download the OpenCV source code. Taking OpenCV 4.7.0 as an example, download the [opencv 4.7.0](https://paddle-model-ecology.bj.bcebos.com/paddlex/cpp/libs/opencv-4.7.0.tgz) source code. After extracting it, an `opencv-4.7.0/` folder will be generated in the current folder.
+
+- Step 1: Build Visual Studio Project
+
+  Specify the `opencv-4.7.0` source code path in cmake-gui, and set the compilation output directory to `opencv-4.7.0/build`. The default installation path is `opencv-4.7.0/build/install`. This installation path will be used for subsequent demo compilation.
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_step1.png"/>
+
+- Step 2: Select Target Platform
+
+  Choose x64 as the target platform and then click Finish.
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_step2.png"/>
+
+- Step 3: Generate Visual Studio Project
+
+  Search for `BUILD_opencv_world` and check it.
+  Click Configure -> Generate -> Open in Project sequentially to enter the Visual Studio 2022 compilation interface.
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_step3.png"/>
+
+- Step 4: Execute Compilation
+
+  Click Start Solution Generation. After compilation, click INSTALL to complete the installation.
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_step4.png"/>
+
+### 1.2 Compile Paddle Inference
+
+You can choose to directly download a pre-compiled package or manually compile the source code.
+
+#### 1.2.1 Directly Download Pre-compiled Package (Recommended)
+
+The [Paddle Inference Official Website](https://www.paddlepaddle.org.cn/inference/v3.0/guides/install/download_lib.html#windows) provides Windows prediction libraries. You can view and select the appropriate pre-compiled package on the official website.
+
+After downloading and extracting it, a `paddle_inference/` subfolder will be generated in the current folder. The directory structure is as follows:
+
+```
+paddle_inference
+├── paddle # Paddle core library and header files
+├── third_party # Third-party dependency libraries and header files
+└── version.txt # Version and compilation information
+```
+
+#### 1.2.2 Compile Prediction Library from Source Code
+
+You can choose to compile the prediction library from source code. Compiling from source allows flexible configuration of various features and dependencies to adapt to different hardware and software environments. For detailed steps, please refer to [Compiling from Source under Windows](https://www.paddlepaddle.org.cn/inference/v3.0/guides/install/compile/source_compile_under_Windows.html).
+
+## 2. Getting Started
+
+### 2.1 Compile Prediction Demo
+
+Before compiling the prediction demo, please ensure that you have compiled the OpenCV library and Paddle Inference prediction library according to Sections 1.1 and 1.2.
+
+The compilation steps are as follows:
+
+- Step 1: Build Visual Studio Project
+
+  Specify the `deploy\cpp_infer` source code path in cmake-gui, and set the compilation output directory to `deploy\cpp_infer\build`. The following compilation steps will use `D:\PaddleOCR\deploy\cpp_infer` as an example source code path. It is normal to encounter an error during the first Configure click. In the subsequent compilation options that pop up, add the installation path for OpenCV and the Paddle Inference prediction library path.
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/windows_step1.png"/>- Step 2: Select the target platform
+  Select the target platform as x64 and click Finish.  
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/windows_step2.png"/>  
+
+- Step 3：Configure CMake build options  
+
+    - OPENCV_DIR: Specify the installation path of OpenCV.  
+    - OpenCV_DIR: Same as OPENCV_DIR.  
+    - PADDLE_LIB: Path to the Paddle Inference library.  
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/windows_step3.png"/>  
+
+- Step 4：Generate Visual Studio project  
+
+  Click Configure → Generate → Open in Project to open the Visual Studio 2022 compilation interface.  
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/windows_step4.png"/>  
+
+- Step 5：Compile the code  
+
+  Before building the solution:  
+  1. Change the build configuration from `Debug` to `Release`.  
+  2. Download [dirent.h](https://paddleocr.bj.bcebos.com/deploy/cpp_infer/cpp_files/dirent.h) and copy it to the Visual Studio include folder (e.g., `C:\Program Files (x86)\Microsoft Visual Studio\2022\Community\VC\Auxiliary\VS\include`).  
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/windows_step5.png"/>  
+
+  After compilation, the executable file will be located at `deploy/cpp_infer/build/Release/ppocr.exe`.  
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/windows_step6.png"/>  
+
+- Step 6：Run the prediction demo  
+
+  Copy the following files to the `deploy\cpp_infer\build\Release\` directory, then follow Sections 2.2 and 2.3 to run the demo:  
+  1. `paddle_inference\paddle\lib\paddle_inference.dll`  
+  2. `paddle_inference\paddle\lib\common.dll`  
+  3. `deploy\cpp_infer\build\bin\Release\abseil_dll.dll`  
+  4. `deploy\cpp_infer\build\third_party\clipper_ver6.4.2\cpp\Release\polyclipping.dll`  
+  5. `opencv-4.7.0\build\install\x64\vc16\bin\opencv_world470.dll`  
+
+### 2.2 Prepare the Model  
+
+Refer to the [General OCR Pipeline C++ Deployment - Linux → 2.2 Prepare the Model](./OCR.en.md#22-prepare-models) section.  
+
+### 2.3 Run the Prediction Demo  
+
+Refer to the [General OCR Pipeline C++ Deployment - Linux → 2.3 Run the Prediction Demo](./OCR.en.md#23-run-the-prediction-demo) section.  
+
+### 2.4 C++ API Integration  
+
+Refer to the [General OCR Pipeline C++ Deployment - Linux → 2.4 C++ API Integration](./OCR.en.md#24-c-api-integration) section.  
+
+## 3. Extended Features  
+
+### 3.1 Multilingual Text Recognition
+
+Refer to the [General OCR Pipeline C++ Deployment - Linux → 3.1 Multilingual Text Recognition](./OCR.en.md#31-multilingual-text-recognition) section.  
+
+### 3.2 Visualize Text Recognition Results  
+
+To visualize text recognition results, you need to compile OpenCV with the FreeType module from the `opencv_contrib` repository (version 4.x). Ensure the OpenCV and `opencv_contrib` versions match. Below is an example using `opencv-4.7.0` and `opencv_contrib-4.7.0`:  
+
+[Download opencv-4.7.0](https://paddle-model-ecology.bj.bcebos.com/paddlex/cpp/libs/opencv-4.7.0.tgz)  
+[Download opencv_contrib-4.7.0](https://paddle-model-ecology.bj.bcebos.com/paddlex/cpp/libs/opencv_contrib-4.7.0.tgz)  
+
+- Step 1：Compile FreeType and HarfBuzz  
+
+    - [Download pkg-config](https://sourceforge.net/projects/pkgconfiglite/)  
+    - [Download freetype2](https://download.savannah.gnu.org/releases/freetype/)  
+    - [Download harfbuzz](https://github.com/harfbuzz/harfbuzz)  
+
+  Add the `bin` directory of the extracted `pkg-config` to the system PATH.  
+  For FreeType compilation, manually specify the installation path (example below):  
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_freetype_step1.png"/>  
+
+  Click Configure → Generate, then open the project in Visual Studio to compile.  
+  In VS, build `ALL_BUILD` and `INSTALL` to generate the required `include` and `lib` files in the `install` directory.  
+
+  Add the FreeType installation path to the system environment variables.  
+
+  For HarfBuzz compilation, manually specify the installation path (example below):  
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_freetype_step2.png"/>  
+
+  After setting the paths, click Configure again, go to **Advanced Options**, and specify the FreeType installation path.  
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_freetype_step3.png"/>  
+
+  Add the HarfBuzz installation path to the system environment variables.- Step 2: Modify `modules/freetype/CMakeLists.txt` under opencv_contrib-4.7.0
+
+  ```bash
+  set(the_description "FreeType module. It enables to draw strings with outlines and mono-bitmaps/gray-bitmaps.")
+  
+  find_package(Freetype REQUIRED)
+  
+  # find_package(HarfBuzz) is not included in cmake
+  set(HARFBUZZ_DIR "$ENV{HARFBUZZ_DIR}" CACHE PATH "HarfBuzz directory")
+  find_path(HARFBUZZ_INCLUDE_DIRS
+      NAMES hb-ft.h PATH_SUFFIXES harfbuzz
+      HINTS ${HARFBUZZ_DIR}/include)
+  find_library(HARFBUZZ_LIBRARIES
+      NAMES harfbuzz
+      HINTS ${HARFBUZZ_DIR}/lib)
+  find_package_handle_standard_args(HARFBUZZ
+      DEFAULT_MSG HARFBUZZ_LIBRARIES HARFBUZZ_INCLUDE_DIRS)
+  
+  if(NOT FREETYPE_FOUND)
+    message(STATUS "freetype2:   NO")
+  else()
+    message(STATUS "freetype2:   YES")
+  endif()
+  
+  if(NOT HARFBUZZ_FOUND)
+    message(STATUS "harfbuzz:   NO")
+  else()
+    message(STATUS "harfbuzz:   YES")
+  endif()
+  
+  if(FREETYPE_FOUND AND HARFBUZZ_FOUND)
+    ocv_define_module(freetype opencv_core opencv_imgproc PRIVATE_REQUIRED ${FREETYPE_LIBRARIES} ${HARFBUZZ_LIBRARIES} WRAP python)
+    ocv_include_directories(${FREETYPE_INCLUDE_DIRS} ${HARFBUZZ_INCLUDE_DIRS})
+  else()
+    ocv_module_disable(freetype)
+  endif()
+  ```
+
+- Step 3: Compile OpenCV
+
+  1. Set the `OPENCV_EXTRA_MODULES_PATH` item and fill in the path to the `modules` directory under the opencv-contrib-4.7.0 directory.
+  2. Check the `WITH_FREETYPE` item. Note that you must first compile freetype and harfbuzz.
+  3. If you need to support freetype, you need to add the relevant paths of freetype in the CMake configuration of OpenCV.
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_freetype_step4.png"/>
+
+  Search for harfbuzz, add harfbuzz, and add the relevant paths.
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_freetype_step5.png"/>
+
+  After completing the above steps, click Configure again in the CMake interface. After ensuring there are no errors, click Generate, and then click Open Project to open Visual Studio. Switch from Debug to Release, right-click on ALL_BUILD and select Build. After the compilation is completed, right-click on INSTALL and select Build.
+
+  Note: If you have compiled OpenCV with FreeType included, when compiling the demo for the General OCR Pipeline in Section 2.1 Step 3, you need to check the `USE_FREETYPE` option to enable text rendering functionality. Additionally, when running the demo, you need to provide the path to the corresponding TTF font file using the `--vis_font_dir your_ttf_path` parameter.
+
+After compiling and running the prediction demo, you can obtain the following visualized text recognition results:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/ocr_res_with_freetype.png"/>
diff --git a/docs/version3.x/inference_deployment/local_inference/cpp/OCR_windows.md b/docs/version3.x/inference_deployment/local_inference/cpp/OCR_windows.md
new file mode 100644
index 0000000..4b098b4
--- /dev/null
+++ b/docs/version3.x/inference_deployment/local_inference/cpp/OCR_windows.md
@@ -0,0 +1,247 @@
+# 通用 OCR 产线 C++ 部署 - Windows
+
+- [1. 环境准备](#1)
+    - [1.1 编译 OpenCV 库](#11-opencv)
+    - [1.2 编译 Paddle Inference](#12-paddle-inference)
+- [2. 开始运行](#2)
+    - [2.1 编译预测 demo](#21-demo)
+    - [2.2 准备模型](#22)
+    - [2.3 运行预测 demo](#23-demo)
+    - [2.4 C++ API 集成](#24-c-api)
+- [3. 拓展功能](#3)
+    - [3.1 多语种文字识别](#31)  
+    - [3.2 可视化文本识别结果](#32)  
+
+## 1. 环境准备
+
+- **本章节编译运行时用到的源代码位于 [PaddleOCR/deploy/cpp_infer](https://github.com/PaddlePaddle/PaddleOCR/tree/{{PADDLEOCR_GITHUB_REF}}/deploy/cpp_infer) 目录下。**
+
+- Windows 环境：
+    - visual studio 2022
+    - cmake 3.29
+
+### 1.1 编译 OpenCV 库
+
+可以选择直接下载预编译包或者手动编译源码。
+
+#### 1.1.1 直接下载预编译包（推荐）
+
+在 [OpenCV 官网](https://opencv.org/releases/) 下载适用于 Windows 的 .exe 预编译包，运行后自动解压出 OpenCV 的预编译库和相关文件夹。
+
+以 opencv 4.7.0为例，下载 [opencv-4.7.0-windows.exe](https://github.com/opencv/opencv/releases/download/4.7.0/opencv-4.7.0-windows.exe)，运行后会在当前的文件夹中生成 `opencv/` 的子文件夹，其中 `opencv/build` 为预编译库，在后续编译通用 OCR 产线 预测 demo 时，将作为 OpenCV 安装库的路径使用。
+
+#### 1.1.2 源码编译
+
+首先需要下载 OpenCV 源码，以 opencv 4.7.0 为例，下载 [opencv 4.7.0](https://paddle-model-ecology.bj.bcebos.com/paddlex/cpp/libs/opencv-4.7.0.tgz) 源码，解压后会在当前的文件夹中生成 `opencv-4.7.0/` 的文件夹。
+
+- Step 1：构建 Visual Studio 项目
+
+  在 cmake-gui 中指定 `opencv-4.7.0` 源码路径，并指定编译生成目录为 `opencv-4.7.0/build`，默认安装路径为 `opencv-4.7.0/build/install`，此安装路径用于后续编译 demo。
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_step1.png"/>
+
+- Step 2：选择目标平台
+
+  选择目标平台为 x64 ，然后点击 finish。
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_step2.png"/>
+
+- Step 3 ：生成 Visual Studio 项目
+
+  搜索 `BUILD_opencv_world` 并勾选。
+  依次点击 Configure  ->  Generate  ->  Open in Project，将进入 Visual Studio 2022 编译界面。
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_step3.png"/>
+
+- Step 4：执行编译
+
+  点击开始生成解决方案，完成编译后，点击 INSTALL，运行后完成安装。
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_step4.png"/>
+
+### 1.2 编译 Paddle Inference
+
+可以选择直接下载预编译包或者手动编译源码。
+
+#### 1.2.1 直接下载预编译包（推荐）
+
+[Paddle Inference 官网](https://www.paddlepaddle.org.cn/inference/v3.0/guides/install/download_lib.html#windows) 上提供了 Windows 预测库，可以在官网查看并选择合适的预编译包。
+
+下载后解压，会在当前的文件夹中生成 `paddle_inference/` 的子文件夹。目录结构为：
+
+```
+paddle_inference
+├── paddle # paddle核心库和头文件
+├── third_party # 第三方依赖库和头文件
+└── version.txt # 版本和编译信息
+```
+
+#### 1.2.2 源码编译预测库
+
+可以选择通过源码自行编译预测库，源码编译可灵活配置各类功能和依赖，以适应不同的硬件和软件环境。详细步骤请参考 [Windows 下源码编译](https://www.paddlepaddle.org.cn/inference/v3.0/guides/install/compile/source_compile_under_Windows.html)。
+
+## 2. 开始运行
+
+### 2.1 编译预测 demo
+
+在编译预测 demo 前，请确保您已经按照 1.1 和 1.2 节编译好 OpenCV 库和 Paddle Inference 预测库。
+
+编译步骤如下：
+
+- Step 1：构建 Visual Studio 项目
+
+  在 cmake-gui 中指定 `deploy\cpp_infer` 源码路径，并指定编译生成目录为 `deploy\cpp_infer\build`，以下编译步骤说明均以 `D:\PaddleOCR\deploy\cpp_infer` 作为示例源码路径。第一次点击 Configure 报错是正常的，在后续弹出的编译选项中，添加 OpenCV 的安装路径和 Paddle Inference 预测库路径。
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/windows_step1.png"/>
+
+- Step 2：选择目标平台
+
+  选择目标平台为 x64 ，然后点击 Finish。
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/windows_step2.png"/>
+
+- Step 3：配置 cmake 编译选项
+
+    - OPENCV_DIR：填写 OpenCV 安装路径。
+    - OpenCV_DIR：同 OPENCV_DIR。
+    - PADDLE_LIB：Paddle Inference 预测库路径。
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/windows_step3.png"/>
+
+- Step 4：生成 Visual Studio 项目
+
+  依次点击 Configure  ->  Generate  ->  Open in Project，将进入 Visual Studio 2022 编译界面。
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/windows_step4.png"/>
+
+- Step 5：执行编译
+
+  在开始生成解决方案之前，执行下面步骤：
+
+  1. 将 `Debug` 改为 `Release`。
+  2. 下载[dirent.h](https://paddleocr.bj.bcebos.com/deploy/cpp_infer/cpp_files/dirent.h)，并拷贝到  Visual Studio 的 include 文件夹下，如 `C:\Program Files (x86)\Microsoft Visual Studio\2022\Community\VC\Auxiliary\VS\include`。
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/windows_step5.png"/>
+
+  编译完成后，可执行文件位于 `deploy/cpp_infer/build/Release/ppocr.exe` 。
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/windows_step6.png"/>
+
+- Step 6：运行预测 demo
+
+  将下面文件拷贝到 `deploy\cpp_infer\build\Release\` 路径下后，参考后续 2.2 和 2.3 小节即可运行预测 demo。
+
+  1. `paddle_inference\paddle\lib\paddle_inference.dll`
+  2. `paddle_inference\paddle\lib\common.dll`
+  3. `deploy\cpp_infer\build\bin\Release\abseil_dll.dll`
+  4. `deploy\cpp_infer\build\third_party\clipper_ver6.4.2\cpp\Release\polyclipping.dll`
+  5. `opencv-4.7.0\build\install\x64\vc16\bin\opencv_world470.dll`
+
+### 2.2 准备模型
+
+该步骤参考 [通用 OCR 产线 C++ 部署 - Linux —— 2.2 准备模型](./OCR.md#22) 小节。
+
+### 2.3 运行预测 demo
+
+参考 [通用 OCR 产线 C++ 部署 - Linux —— 2.3 运行预测 demo](./OCR.md#23-demo) 小节。
+
+### 2.4 C++ API 集成
+
+参考 [通用 OCR 产线 C++ 部署 - Linux —— 2.4 C++ API 集成](./OCR.md#24-c-api) 小节。
+
+## 3. 拓展功能
+
+### 3.1 多语种文字识别
+
+参考 [通用 OCR 产线 C++ 部署 - Linux —— 3.1 多语种文字识别](./OCR.md#31) 小节。
+
+### 3.2 可视化文本识别结果
+
+我们使用 4.x 版本的 opencv_contrib 模块中的 FreeType 进行字体渲染，如果想要可视化文本识别结果，需要下载 OpenCV 和 opencv_contrib 的源码并编译包含 FreeType 模块的 OpenCV。下载源码时需确保两者的版本一致。以下以 opencv-4.7.0 和 opencv_contrib-4.7.0 为例进行说明：
+
+[下载 opencv-4.7.0](https://paddle-model-ecology.bj.bcebos.com/paddlex/cpp/libs/opencv-4.7.0.tgz)
+[下载 opencv_contrib-4.7.0](https://paddle-model-ecology.bj.bcebos.com/paddlex/cpp/libs/opencv_contrib-4.7.0.tgz)
+
+- Step 1：编译 freetype 和 harfbuzz
+
+    - [下载pkg-config](https://sourceforge.net/projects/pkgconfiglite/)
+    - [下载freetype2](https://download.savannah.gnu.org/releases/freetype/)
+    - [下载harfbuzz](https://github.com/harfbuzz/harfbuzz)
+
+  解压 pkg-config 后添加其 bin 目录到系统 PATH 环境变量。
+  freetype 编译，需手动更改其安装路径，示例如下：
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_freetype_step1.png"/>
+
+  再次 Configure， 然后点击 Generate， 完成后，点击 Open Project 按钮，打开 VS ，编译。
+  VS里ALL_BUILD, INSTALL. 会在构建文件夹的 install 目录下生成所需的 include 和 lib 文件。
+
+  然后将 freetype 安装路径添加至系统环境变量。
+
+  harfbuzz 编译，需手动更改其安装路径，示例如下：
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_freetype_step2.png"/>
+
+  设置好上面两项后，再次点击 Configure 按钮，选择 Advanced Options ，填写 freetype 安装路径。
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_freetype_step3.png"/>
+
+  然后将 harfbuzz 安装路径添加至系统环境变量。
+
+- Step 2：修改 opencv_contrib-4.7.0 下的 `modules/freetype/CMakeLists.txt`
+
+  ```bash
+  set(the_description "FreeType module. It enables to draw strings with outlines and mono-bitmaps/gray-bitmaps.")
+  
+  find_package(Freetype REQUIRED)
+  
+  # find_package(HarfBuzz) is not included in cmake
+  set(HARFBUZZ_DIR "$ENV{HARFBUZZ_DIR}" CACHE PATH "HarfBuzz directory")
+  find_path(HARFBUZZ_INCLUDE_DIRS
+      NAMES hb-ft.h PATH_SUFFIXES harfbuzz
+      HINTS ${HARFBUZZ_DIR}/include)
+  find_library(HARFBUZZ_LIBRARIES
+      NAMES harfbuzz
+      HINTS ${HARFBUZZ_DIR}/lib)
+  find_package_handle_standard_args(HARFBUZZ
+      DEFAULT_MSG HARFBUZZ_LIBRARIES HARFBUZZ_INCLUDE_DIRS)
+  
+  if(NOT FREETYPE_FOUND)
+    message(STATUS "freetype2:   NO")
+  else()
+    message(STATUS "freetype2:   YES")
+  endif()
+  
+  if(NOT HARFBUZZ_FOUND)
+    message(STATUS "harfbuzz:   NO")
+  else()
+    message(STATUS "harfbuzz:   YES")
+  endif()
+  
+  if(FREETYPE_FOUND AND HARFBUZZ_FOUND)
+    ocv_define_module(freetype opencv_core opencv_imgproc PRIVATE_REQUIRED ${FREETYPE_LIBRARIES} ${HARFBUZZ_LIBRARIES} WRAP python)
+    ocv_include_directories(${FREETYPE_INCLUDE_DIRS} ${HARFBUZZ_INCLUDE_DIRS})
+  else()
+    ocv_module_disable(freetype)
+  endif()
+  ```
+
+- Step 3 编译 OpenCV
+
+  1. 设置 `OPENCV_EXTRA_MODULES_PATH` 项，填入 opencv-contrib-4.7.0 的目录下的 modules 目录。
+  2. 勾选 `WITH_FREETYPE` 项，必须先编译 freetype 和 harfbuzz。
+  3. 如果需要支持 freetype，则需要在 Opencv 的 Cmake 配置中加入 freetype 的相关路径。
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_freetype_step4.png"/>
+
+  搜索 harfbuzz，加入 harfbuzz，加入 的相关路径。
+
+  <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/opencv_freetype_step5.png"/>
+
+  完成后，再次在 Cmake 界面，点击 configure， 确定没报错后，点击 Generate，最后点击 Open Project，打开 Visual studio，将 Debug 切换为 Release，找到 ALL_BUILD 右键 Build， 等待编译完成后， 找到 INSTALL 右键 Build。
+
+  注意：如果完成编译包含 FreeType 的 OpenCV，在编译通用 OCR 产线 demo 时，需要在 2.1节 Step 3 配置编译选项时勾选 `USE_FREETYPE` 开启文字渲染功能，并且在运行 demo 时通过 `--vis_font_dir your_ttf_path` 提供相应 ttf 字体文件路径。
+
+编译并运行预测 demo 可以得到如下可视化文本识别结果：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/deployment/cpp/ocr_res_with_freetype.png"/>
diff --git a/docs/version3.x/inference_deployment/local_inference/cpp/cpp_local_deployment.en.md b/docs/version3.x/inference_deployment/local_inference/cpp/cpp_local_deployment.en.md
new file mode 100644
index 0000000..3017f88
--- /dev/null
+++ b/docs/version3.x/inference_deployment/local_inference/cpp/cpp_local_deployment.en.md
@@ -0,0 +1,9 @@
+# C++ Local Deployment
+
+Linux：
+
+[C++ Local Deployment for General OCR Pipeline - Linux](./OCR.en.md)
+
+Windows:
+
+[C++ Local Deployment for General OCR Pipeline - Windows](./OCR_windows.en.md)
diff --git a/docs/version3.x/inference_deployment/local_inference/cpp/cpp_local_deployment.md b/docs/version3.x/inference_deployment/local_inference/cpp/cpp_local_deployment.md
new file mode 100644
index 0000000..bca6a17
--- /dev/null
+++ b/docs/version3.x/inference_deployment/local_inference/cpp/cpp_local_deployment.md
@@ -0,0 +1,9 @@
+# C++ 本地部署
+
+Linux：
+
+[通用 OCR 产线 C++ 本地部署 - Linux](./OCR.md)
+
+Windows:
+
+[通用 OCR 产线 C++ 本地部署 - Windows](./OCR_windows.md)
diff --git a/docs/version3.x/inference_deployment/local_inference/high_performance_inference.en.md b/docs/version3.x/inference_deployment/local_inference/high_performance_inference.en.md
new file mode 100644
index 0000000..3347283
--- /dev/null
+++ b/docs/version3.x/inference_deployment/local_inference/high_performance_inference.en.md
@@ -0,0 +1,109 @@
+---
+comments: true
+---
+
+# High-Performance Inference
+
+In real-world production environments, many applications have stringent performance requirements for deployment strategies, particularly regarding response speed, to ensure efficient system operation and a smooth user experience. PaddleOCR provides high-performance inference capabilities, allowing users to enhance model inference speed with a single click without worrying about complex configurations or underlying details. Specifically, PaddleOCR's high-performance inference functionality can:
+
+- Automatically select an appropriate inference backend (e.g., Paddle Inference, OpenVINO, ONNX Runtime, TensorRT) based on prior knowledge and configure acceleration strategies (e.g., increasing the number of inference threads, setting FP16 precision inference).
+- Automatically convert PaddlePaddle static graph models to ONNX format as needed to leverage better inference backends for acceleration.
+- Perform inference using ONNX models.
+
+This document primarily introduces the installation and usage methods for high-performance inference.
+
+## 1. Prerequisites
+
+### 1.1 Install High-Performance Inference Dependencies
+
+The high-performance inference functionality depends on the PaddlePaddle framework, so before proceeding to the subsequent steps, you need to ensure that PaddlePaddle is installed in your environment.
+
+Install the dependencies required for high-performance inference using the PaddleOCR CLI:
+
+```bash
+paddleocr install_hpi_deps {device_type}
+```
+
+The supported device types are:
+
+- `cpu`: For CPU-only inference. Currently supports Linux systems, x86-64 architecture processors, and Python 3.8-3.12.
+- `gpu`: For inference using either CPU or NVIDIA GPU. Currently supports Linux systems, x86-64 architecture processors, and Python 3.8-3.12. If you want to use the full high-performance inference capabilities, you also need to ensure that a compatible version of TensorRT is installed in your environment. Refer to the next subsection for detailed instructions.
+
+Only one type of device dependency should exist in the same environment. For Windows systems, it is currently recommended to install within a Docker container or [WSL](https://learn.microsoft.com/en-us/windows/wsl/install) environment.
+
+**It is recommended to use the official PaddlePaddle Docker image to install high-performance inference dependencies.** The corresponding images for each device type are as follows:
+
+- `cpu`：`ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlex/paddlex:paddlex3.0.1-paddlepaddle3.0.0-cpu`
+- `gpu`：
+    - CUDA 11.8：`ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlex/paddlex:paddlex3.0.1-paddlepaddle3.0.0-gpu-cuda11.8-cudnn8.9-trt8.6`
+- `gpu`：
+    - CUDA 12.6：`ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlex/paddlex:paddlex3.0.1-paddlepaddle3.0.0-gpu-cuda12.6-cudnn9.5-trt10.5`
+
+**Notice:**
+
+- **Currently, high-performance inference with CUDA 12.6 and cuDNN 9.5 only supports OpenVINO and ONNX Runtime backends, and does not yet support the TensorRT backend.**
+
+### 1.2 Detailed GPU Environment Instructions
+
+First, ensure that the environment has the required CUDA and cuDNN versions installed. Currently, PaddleOCR supports CUDA and cuDNN versions compatible with CUDA 11.8 + cuDNN 8.9 or CUDA 12.6 + cuDNN 9.5. Below are the installation instructions for CUDA and cuDNN:
+
+- [Install CUDA 11.8](https://developer.nvidia.com/cuda-11-8-0-download-archive)
+- [Install cuDNN 8.9](https://docs.nvidia.com/deeplearning/cudnn/archives/cudnn-890/install-guide/index.html)
+- [Install CUDA 12.6](https://developer.nvidia.com/cuda-12-6-0-download-archive)
+- [Install cuDNN 9.5](https://docs.nvidia.com/deeplearning/cudnn/backend/v9.5.0/installation/linux.html)
+
+If using the official PaddlePaddle image, the CUDA and cuDNN versions in the image already meet the requirements, and no additional installation is needed.
+
+If installing PaddlePaddle via pip, the relevant Python packages for CUDA and cuDNN will typically be installed automatically. In this case, **you still need to install the non-Python-specific CUDA and cuDNN versions.** It is also recommended to install CUDA and cuDNN versions that match the Python package versions in your environment to avoid potential issues caused by coexisting library versions. You can check the versions of the CUDA and cuDNN-related Python packages with the following commands:
+
+```bash
+# CUDA-related Python package versions
+pip list | grep nvidia-cuda
+# cuDNN-related Python package versions
+pip list | grep nvidia-cudnn
+```
+
+Secondly, it is recommended to ensure that a compatible version of TensorRT is installed in the environment; otherwise, the Paddle Inference TensorRT subgraph engine will be unavailable, and the program may not achieve optimal inference performance. Currently, PaddleOCR only supports TensorRT 8.6.1.6 in the CUDA 11.8 environment. If using the official PaddlePaddle 3.0 Docker image, you can install the TensorRT wheel package with the following command:
+
+```bash
+python -m pip install /usr/local/TensorRT-*/python/tensorrt-*-cp310-none-linux_x86_64.whl
+```
+
+For other environments, refer to the [TensorRT documentation](https://docs.nvidia.com/deeplearning/tensorrt/archives/index.html) to install TensorRT. Here is an example:
+
+```bash
+# Download the TensorRT tar file
+wget https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/secure/8.6.1/tars/TensorRT-8.6.1.6.Linux.x86_64-gnu.cuda-11.8.tar.gz
+# Extract the TensorRT tar file
+tar xvf TensorRT-8.6.1.6.Linux.x86_64-gnu.cuda-11.8.tar.gz
+# Install the TensorRT wheel package
+python -m pip install TensorRT-8.6.1.6/python/tensorrt-8.6.1-cp310-none-linux_x86_64.whl
+# Add the absolute path of the TensorRT `lib` directory to LD_LIBRARY_PATH
+export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:TensorRT-8.6.1.6/lib"
+```
+
+## 2. Executing High-Performance Inference
+
+For the PaddleOCR CLI, specify `--enable_hpi` as `True` to execute high-performance inference. For example:
+
+```bash
+paddleocr ocr --enable_hpi True ...
+```
+
+For the PaddleOCR Python API, set `enable_hpi` to `True` when initializing the pipeline or module object to enable high-performance inference when calling the inference method. For example:
+
+```python
+from paddleocr import PaddleOCR
+pipeline = PaddleOCR(enable_hpi=True)
+result = pipeline.predict(...)
+```
+
+## 3. Notes
+
+1. For some models, the first execution of high-performance inference may take longer to complete the construction of the inference engine. Relevant information about the inference engine will be cached in the model directory after the first construction, and subsequent initializations can reuse the cached content to improve speed.
+
+2. Currently, due to reasons such as not using static graph format models or the presence of unsupported operators, some models may not achieve inference acceleration.
+
+3. During high-performance inference, PaddleOCR automatically handles the conversion of model formats and selects the optimal inference backend whenever possible. Additionally, PaddleOCR supports users specifying ONNX models. For information on converting PaddlePaddle static graph models to ONNX format, refer to [Obtaining ONNX Models](../others/obtaining_onnx_models.en.md).
+
+4. The high-performance inference capabilities of PaddleOCR rely on PaddleX and its high-performance inference plugins. By passing in a custom PaddleX production line configuration file, you can configure the inference backend and other related settings. Please refer to [Using PaddleX Production Line Configuration Files](../../paddleocr_and_paddlex.en.md#3-using-paddlex-pipeline-configuration-files) and the [PaddleX High-Performance Inference Guide](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_deploy/high_performance_inference.html#22) to learn how to adjust the high-performance inference configurations.
diff --git a/docs/version3.x/inference_deployment/local_inference/high_performance_inference.md b/docs/version3.x/inference_deployment/local_inference/high_performance_inference.md
new file mode 100644
index 0000000..de161eb
--- /dev/null
+++ b/docs/version3.x/inference_deployment/local_inference/high_performance_inference.md
@@ -0,0 +1,106 @@
+---
+comments: true
+---
+
+# 高性能推理
+
+在实际生产环境中，许多应用对部署策略的性能指标（尤其是响应速度）有着较严苛的标准，以确保系统的高效运行与用户体验的流畅性。PaddleOCR 提供高性能推理能力，让用户无需关注复杂的配置和底层细节，一键提升模型的推理速度。具体而言，PaddleOCR 的高性能推理功能能够：
+
+- 结合先验知识自动选择合适的推理后端（Paddle Inference、OpenVINO、ONNX Runtime、TensorRT等），并配置加速策略（如增大推理线程数、设置 FP16 精度推理）；
+- 根据需要自动将飞桨静态图模型转换为 ONNX 格式，以使用更优的推理后端实现加速；
+- 使用 ONNX 模型完成推理。
+
+本文档主要介绍高性能推理功能的安装与使用方法。
+
+## 1. 前置条件
+
+## 1.1 安装高性能推理依赖
+
+高性能推理功能依赖于飞桨框架，因此，在进行后续步骤前，您需要确保环境中安装有飞桨框架。
+
+通过 PaddleOCR CLI 安装高性能推理所需依赖：
+
+```bash
+paddleocr install_hpi_deps {设备类型}
+```
+
+支持的设备类型包括：
+
+- `cpu`：仅使用 CPU 推理。目前支持 Linux 系统、x86-64 架构处理器、Python 3.8-3.12。
+- `gpu`：使用 CPU 或 NVIDIA GPU 推理。目前支持 Linux 系统、x86-64 架构处理器、Python 3.8-3.12。如果希望使用完整的高性能推理功能，还需要确保环境中安装有符合要求的 TensorRT。请查看下一小节的详细说明。
+
+同一环境中只应该存在一种设备类型的依赖。对于 Windows 系统，目前建议在 Docker 容器或者 [WSL](https://learn.microsoft.com/zh-cn/windows/wsl/install) 环境中安装。
+
+**推荐使用 PaddleX 官方 Docker 镜像安装高性能推理依赖。** 各设备类型对应的镜像如下：
+
+- `cpu`：`ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlex/paddlex:paddlex3.0.1-paddlepaddle3.0.0-cpu`
+- `gpu`：
+    - CUDA 11.8：`ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlex/paddlex:paddlex3.0.1-paddlepaddle3.0.0-gpu-cuda11.8-cudnn8.9-trt8.6`
+- `gpu`：
+    - CUDA 12.6：`ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlex/paddlex:paddlex3.0.1-paddlepaddle3.0.0-gpu-cuda12.6-cudnn9.5-trt10.5`
+
+**注意：**
+
+- **目前 CUDA 12.6 + cuDNN 9.5 的高性能推理仅支持 OpenVINO 和 ONNX Runtime 后端，暂不支持 TensorRT 后端。**
+
+## 1.2 GPU 环境详细说明
+
+首先，需要确保环境中安装有符合要求的 CUDA 与 cuDNN。目前 PaddleOCR 支持与 CUDA 11.8 + cuDNN 8.9 或 CUDA 12.6 + cuDNN 9.5 兼容的 CUDA 和 cuDNN 版本。以下分别是 CUDA 和 cuDNN 的安装说明文档：
+
+- [安装 CUDA 11.8](https://developer.nvidia.com/cuda-11-8-0-download-archive)
+- [安装 cuDNN 8.9](https://docs.nvidia.com/deeplearning/cudnn/archives/cudnn-890/install-guide/index.html)
+- [安装 CUDA 12.6](https://developer.nvidia.com/cuda-12-6-0-download-archive)
+- [安装 cuDNN 9.5](https://docs.nvidia.com/deeplearning/cudnn/backend/v9.5.0/installation/linux.html)
+
+如果使用飞桨官方镜像，则镜像中的 CUDA 和 cuDNN 版本已经是满足要求的，无需额外安装。
+
+如果通过 pip 安装飞桨，通常 CUDA、cuDNN 的相关 Python 包将被自动安装。在这种情况下，**仍需要通过安装非 Python 专用的 CUDA 与 cuDNN**。同时，建议安装的 CUDA 和 cuDNN 版本与环境中存在的 Python 包版本保持一致，以避免不同版本的库共存导致的潜在问题。可以通过如下方式可以查看 CUDA 和 cuDNN 相关 Python 包的版本：
+
+```bash
+# CUDA 相关 Python 包版本
+pip list | grep nvidia-cuda
+# cuDNN 相关 Python 包版本
+pip list | grep nvidia-cudnn
+```
+
+其次，建议确保环境中安装有符合要求的 TensorRT，否则 Paddle Inference TensorRT 子图引擎将不可用，程序可能无法取得最佳推理性能。**目前 PaddleOCR 仅支持在 CUDA 11.8 环境使用 TensorRT 8.6.1.6**。如果使用飞桨官方 3.0 镜像，可执行如下命令安装 TensorRT wheel 包：
+
+```bash
+python -m pip install /usr/local/TensorRT-*/python/tensorrt-*-cp310-none-linux_x86_64.whl
+```
+
+对于其他环境，请参考 [TensorRT 文档](https://docs.nvidia.com/deeplearning/tensorrt/archives/index.html) 安装 TensorRT。示例如下：
+
+```bash
+# 下载 TensorRT tar 文件
+wget https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/secure/8.6.1/tars/TensorRT-8.6.1.6.Linux.x86_64-gnu.cuda-11.8.tar.gz
+# 解压 TensorRT tar 文件
+tar xvf TensorRT-8.6.1.6.Linux.x86_64-gnu.cuda-11.8.tar.gz
+# 安装 TensorRT wheel 包
+python -m pip install TensorRT-8.6.1.6/python/tensorrt-8.6.1-cp310-none-linux_x86_64.whl
+# 添加 TensorRT 的 `lib` 目录的绝对路径到 LD_LIBRARY_PATH 中
+export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:TensorRT-8.6.1.6/lib"
+```
+
+## 2. 执行高性能推理
+
+对于 PaddleOCR CLI，指定 `--enable_hpi` 为 `True` 即可执行高性能推理。例如：
+
+```bash
+paddleocr ocr --enable_hpi True ...
+```
+
+对于 PaddleOCR Python API，在初始化产线对象或者模块对象时，设置 `enable_hpi` 为 `True` 即可在调用推理方法时执行高性能推理。例如：
+
+```python
+from paddleocr import PaddleOCR
+pipeline = PaddleOCR(enable_hpi=True)
+result = pipeline.predict(...)
+```
+
+## 3. 说明
+
+1. 对于部分模型，在首次执行高性能推理时，可能需要花费较长时间完成推理引擎的构建。推理引擎相关信息将在第一次构建完成后被缓存在模型目录，后续可复用缓存中的内容以提升初始化速度。
+2. 目前，由于使用的不是静态图格式模型、存在不支持算子等原因，部分模型可能无法获得推理加速。
+3. 在进行高性能推理时，PaddleOCR 会自动处理模型格式的转换，并尽可能选择最优的推理后端。同时，PaddleOCR 也支持用户指定 ONNX 模型。有关如何飞桨静态图模型转换为 ONNX 格式，可参考 [获取 ONNX 模型](../others/obtaining_onnx_models.md)。
+4. PaddleOCR 的高性能推理能力依托于 PaddleX 及其高性能推理插件。通过传入自定义 PaddleX 产线配置文件，可以对推理后端等进行配置。请参考 [使用 PaddleX 产线配置文件](../../paddleocr_and_paddlex.md#3-paddlex) 和 [PaddleX 高性能推理指南](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/high_performance_inference.html#22) 了解如何调整高性能推理配置。
diff --git a/docs/version3.x/inference_deployment/local_inference/inference_engine.en.md b/docs/version3.x/inference_deployment/local_inference/inference_engine.en.md
new file mode 100644
index 0000000..2a04434
--- /dev/null
+++ b/docs/version3.x/inference_deployment/local_inference/inference_engine.en.md
@@ -0,0 +1,246 @@
+---
+comments: true
+---
+
+# Inference Engine and Configuration
+
+PaddleOCR 3.5 introduces a unified inference-engine configuration mechanism: use `engine` to select the underlying inference engine, and use `engine_config` to pass engine-specific settings. This mechanism applies to both individual models and pipelines.
+
+If `engine` is not explicitly specified, the default behavior remains the same as in earlier versions: except for a few scenarios such as high-performance inference and generative AI client request features, PaddleOCR uses the PaddlePaddle framework for inference in most cases. If `engine` is explicitly specified, initialization follows the selected engine first.
+
+## 1. What Is an Inference Engine
+
+In PaddleOCR, an inference engine refers to the underlying runtime used to execute a model. It determines which runtime loads and runs the model. You can think of it as "the engine actually used during model inference." When using an inference engine, users usually only need to care about two things:
+
+- which type of inference engine to use;
+- how to configure the inference engine.
+
+## 2. Inference Engines Currently Supported by PaddleOCR
+
+| Engine category | `engine` values | Description |
+| - | - | - |
+| PaddlePaddle framework | `paddle`, `paddle_static`, `paddle_dynamic` | Runs on the PaddlePaddle framework. |
+| Transformers | `transformers` | Runs on Hugging Face Transformers. |
+| ONNX Runtime | `onnxruntime` | Runs on ONNX Runtime. |
+
+- `paddle`: The unified entry point for the PaddlePaddle framework. It selects `paddle_static` or `paddle_dynamic` according to the model type and files in the model directory. If both are available, `paddle_static` is preferred.
+- `paddle_static`: PaddlePaddle static-graph inference, suitable for scenarios that require better inference performance or more fine-grained performance tuning.
+- `paddle_dynamic`: PaddlePaddle dynamic-graph inference, which is more flexible and easier to debug compared to static graph.
+- `transformers`: Hugging Face Transformers inference, making it convenient to integrate with the Hugging Face ecosystem.
+- `onnxruntime`: ONNX Runtime inference, used to load and execute ONNX-format models.
+
+## 3. Installation by Inference Engine
+
+> Since different inference engines may have conflicting dependencies, it is recommended to install only one inference engine per environment.
+
+### 3.1 PaddlePaddle framework
+
+When using the PaddlePaddle framework for inference, install PaddlePaddle first. For installation instructions, see [PaddlePaddle Framework Installation](../../paddlepaddle_installation.en.md). For requirements on the PaddlePaddle version, see [here](../../paddleocr_and_paddlex.en.md#1-differences-and-connections-between-paddleocr-and-paddlex).
+
+### 3.2 Transformers
+
+When using Transformers as the inference engine, you need to install Hugging Face Transformers (`>=5.10.0`). Example command:
+
+```bash
+python -m pip install "transformers>=5.10.0"
+```
+
+In many cases, you also need to install the underlying inference framework. For details, see the [Transformers official documentation](https://huggingface.co/docs/transformers/installation).
+
+### 3.3 ONNX Runtime
+
+When using ONNX Runtime as the inference engine, install the ONNX Runtime package. Example command:
+
+```bash
+python -m pip install onnxruntime-gpu
+```
+
+The command above applies to NVIDIA GPUs in CUDA 12.x environments. For installation instructions for other devices or environments, please refer to the [ONNX Runtime official documentation](https://onnxruntime.ai/docs/install/).
+
+## 4. Configuration and Supported Values of `engine` and `engine_config`
+
+### 4.1 `engine`
+
+`engine` is used to specify the inference engine. Supported values are:
+
+| Value | Meaning | Description |
+| - | - | - |
+| `None` | No explicit engine specified | Automatically determines the inference engine. Keep the behavior of PaddleOCR 3.4; in most cases, the PaddlePaddle framework will be used for inference. |
+| `paddle` | Unified PaddlePaddle framework entry | Automatically selects `paddle_static` or `paddle_dynamic`. |
+| `paddle_static` | Static-graph inference | Uses Paddle static-graph inference. |
+| `paddle_dynamic` | Dynamic-graph inference | Uses Paddle dynamic-graph inference. |
+| `transformers` | Transformers inference | Uses Hugging Face Transformers inference. |
+| `onnxruntime` | ONNX Runtime inference | Uses ONNX Runtime inference. |
+
+### 4.2 `engine_config`
+
+`engine_config` is used to configure the inference engine and is recommended to be used together with `engine`. Common `engine_config` fields for each engine are listed below:
+
+#### `paddle_static`
+
+Common fields include:
+
+- `run_mode`: execution mode, such as `paddle`, `trt_fp32`, `trt_fp16`, and `mkldnn`;
+- `device_type` / `device_id`: device type and device index;
+- `cpu_threads`: number of CPU inference threads;
+- `delete_pass`: list of graph optimization passes to disable manually;
+- `enable_new_ir`: whether to enable the new IR;
+- `enable_cinn`: whether to enable CINN;
+- `trt_cfg_setting`: low-level TensorRT configuration;
+- `trt_use_dynamic_shapes`: whether to enable TensorRT dynamic shapes;
+- `trt_collect_shape_range_info`: whether to collect shape range information;
+- `trt_discard_cached_shape_range_info`: whether to discard existing shape range information and recollect it;
+- `trt_dynamic_shapes`: dynamic shape configuration;
+- `trt_dynamic_shape_input_data`: data used to fill input tensors when collecting dynamic shapes;
+- `trt_shape_range_info_path`: path to the shape range information file;
+- `trt_allow_rebuild_at_runtime`: whether rebuilding the TensorRT engine is allowed at runtime;
+- `mkldnn_cache_capacity`: oneDNN (MKLDNN) cache capacity.
+
+#### `paddle_dynamic`
+
+Common fields include:
+
+- `device_type` / `device_id`: device type and device index used during dynamic-graph execution.
+
+#### `transformers`
+
+Common fields include:
+
+- `dtype`: data type used for model weights / inference, such as `float16`;
+- `device_type` / `device_id`: inference device type and device index;
+- `trust_remote_code`: whether to trust and execute custom code in model repositories;
+- `attn_implementation`: attention implementation method, such as `flash_attention_2`;
+- `generation_config`: generation parameters, such as `max_new_tokens` and `temperature`;
+- `model_kwargs`: extra arguments passed to the model loading API;
+- `processor_kwargs`: extra arguments passed to the processor / image processor loading API;
+- `tokenizer_kwargs`: a compatibility-preserved field that is merged with `processor_kwargs`.
+
+#### `onnxruntime`
+
+Common fields include:
+
+- `device_type` / `device_id`: inference device type and device index;
+- `providers`: list of execution providers (e.g., `CUDAExecutionProvider`, `CPUExecutionProvider`);
+- `provider_options`: provider-specific configuration;
+- `graph_optimization_level`: graph optimization level;
+- `intra_op_num_threads`: number of intra-op threads;
+- `inter_op_num_threads`: number of inter-op threads;
+- `execution_mode`: execution mode (e.g., `sequential`, `parallel`);
+- `log_severity_level`: log severity level;
+- `enable_mem_pattern`: whether to enable memory pattern;
+- `enable_cpu_mem_arena`: whether to enable CPU memory arena;
+- `session_options`: ONNX Runtime session options.
+
+#### 4.2.1 Flat vs. bucketed `engine_config`
+
+At the same level `engine_config` may be:
+
+- **Flat**: a dict whose keys are only those required by the **resolved** engine (for example, when using static graph only, top-level keys such as `run_mode` and `cpu_threads`).
+- **Bucketed**: top-level keys are **only** registered engine names (e.g. `paddle_static`, `paddle_dynamic`, `transformers`, `onnxruntime`), each mapping to a nested dict. You **must not** mix bucket keys with flat keys at the same level (e.g. `{"paddle_static": {...}, "run_mode": "paddle"}` is invalid).
+
+When an engine is resolved, only the corresponding config is used: flat configs are validated as a whole; bucketed configs take the entry for that engine.
+
+### 4.3 Priority and Override Rules
+
+- For pipelines, `engine` and `engine_config` passed through CLI arguments or Python API initialization arguments take precedence over fields with the same names in the pipeline configuration file.
+- In pipeline configuration files, top-level `engine` and `engine_config` act as global settings; `engine` and `engine_config` in submodules or sub-pipelines can override upper-level settings.
+- For more complete rules about priority, overriding, and pipeline configuration behavior, refer to the PaddleX documentation: [PaddleX Pipeline Python API Usage](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/instructions/pipeline_python_API.html).
+
+### 4.4 Compatibility Rules
+
+- When `engine` is explicitly set, `enable_hpi` no longer takes effect.
+- When `engine_config` is explicitly provided, compatibility arguments for the selected engine are ignored. For example, in `paddle` / `paddle_static` scenarios, compatibility arguments such as `use_tensorrt`, `precision`, `enable_mkldnn`, `mkldnn_cache_capacity`, `cpu_threads`, and `enable_cinn` no longer take effect.
+
+## 5. Usage Examples
+
+### 5.1 Individual model (CLI): select the engine with `--engine`
+
+```bash
+paddleocr text_detection -i general_ocr_001.png --engine transformers
+paddleocr text_detection -i general_ocr_001.png --engine onnxruntime
+```
+
+### 5.2 Individual model (Python): explicitly specify `transformers`
+
+```python
+from paddleocr import TextDetection
+
+model = TextDetection(
+    model_name="PP-OCRv5_server_det",
+    engine="transformers",
+)
+
+result = model.predict("general_ocr_001.png")
+```
+
+### 5.3 Individual model (Python): explicitly specify `onnxruntime`
+
+```python
+from paddleocr import TextDetection
+
+model = TextDetection(
+    model_name="PP-OCRv5_server_det",
+    engine="onnxruntime",
+)
+
+result = model.predict("general_ocr_001.png")
+```
+
+### 5.4 Individual model (Python): specify `paddle_static` and `engine_config`
+
+```python
+from paddleocr import TextDetection
+
+model = TextDetection(
+    model_name="PP-OCRv5_server_det",
+    engine="paddle_static",
+    engine_config={
+        "device_type": "cpu",
+        "cpu_threads": 4,
+        "run_mode": "mkldnn",
+    },
+)
+
+result = model.predict("general_ocr_001.png")
+```
+
+### 5.5 Pipeline (CLI): select the engine with `--engine`
+
+```bash
+paddleocr ocr -i general_ocr_001.png --engine paddle_static
+```
+
+### 5.6 Pipeline (Python API): configure the inference engine for a specific module
+
+If you want to specify `engine` and `engine_config` for a specific module inside a pipeline, you can first export the configuration file, modify the corresponding module configuration, and then load it. For how to export, edit, and load the configuration file, see [Using PaddleX Pipeline Configuration Files](../../paddleocr_and_paddlex.en.md#3-using-paddlex-pipeline-configuration-files). Example:
+
+First, export the pipeline configuration file:
+
+```python
+from paddleocr import PaddleOCR
+
+pipeline = PaddleOCR()
+pipeline.export_paddlex_config_to_yaml("ocr_config.yaml")
+```
+
+Then, set `engine` and `engine_config` specifically for the `TextDetection` module in `ocr_config.yaml`:
+
+```yaml
+pipeline_name: OCR
+SubModules:
+  TextDetection:
+    engine: paddle_static
+    engine_config:
+      device_type: cpu
+      cpu_threads: 4
+      run_mode: mkldnn
+```
+
+Use the updated configuration file for inference:
+
+```python
+from paddleocr import PaddleOCR
+
+pipeline = PaddleOCR(paddlex_config="ocr_config.yaml")
+result = pipeline.predict("general_ocr_001.png")
+```
diff --git a/docs/version3.x/inference_deployment/local_inference/inference_engine.md b/docs/version3.x/inference_deployment/local_inference/inference_engine.md
new file mode 100644
index 0000000..1871197
--- /dev/null
+++ b/docs/version3.x/inference_deployment/local_inference/inference_engine.md
@@ -0,0 +1,246 @@
+---
+comments: true
+---
+
+# 推理引擎与配置说明
+
+PaddleOCR 3.5 引入了统一的推理引擎配置方式：使用 `engine` 选择底层推理引擎，使用 `engine_config` 传递该引擎的专属配置。无论是单模型还是产线，均可按这套方式声明推理行为。
+
+如果不显式指定 `engine`，默认行为与旧版本保持一致：除高性能推理功能和生成式 AI 客户端请求功能等少数场景外，大多数情况下会优先使用飞桨框架推理；如果显式指定 `engine`，则会优先按指定引擎初始化。
+
+## 1. 什么是推理引擎
+
+在 PaddleOCR 中，推理引擎指模型执行时所使用的底层运行时。它决定了模型由哪套运行时加载与执行，可以将它理解为“模型推理时实际使用的引擎”。在使用推理引擎时，用户通常只需要关心两件事：
+
+- 选择哪类推理引擎；
+- 如何配置推理引擎。
+
+## 2. PaddleOCR 当前支持的推理引擎
+
+| 引擎类别 | `engine` 取值 | 说明 |
+| - | - | - |
+| 飞桨框架 | `paddle`、`paddle_static`、`paddle_dynamic` | 基于飞桨框架运行。 |
+| Transformers | `transformers` | 基于 Hugging Face Transformers 运行。 |
+| ONNX Runtime | `onnxruntime` | 基于 ONNX Runtime 运行。 |
+
+- `paddle`：飞桨框架统一入口。根据模型类型和模型目录中的文件选择 `paddle_static` 或 `paddle_dynamic`，在二者都可用的情况下偏好 `paddle_static`。
+- `paddle_static`：飞桨静态图推理，适合对推理性能有一定要求或者需要进行精细化推理性能调优的场景。
+- `paddle_dynamic`：飞桨动态图推理，相比静态图更加灵活、易于调试。
+- `transformers`：Hugging Face Transformers 推理，便于与 Hugging Face 生态集成。
+- `onnxruntime`：ONNX Runtime 推理，用于加载并执行 ONNX 格式模型。
+
+## 3. 各推理引擎安装方式
+
+> 不同推理引擎之间可能存在依赖冲突，建议每个环境仅安装一种推理引擎。
+
+### 3.1 飞桨框架
+
+当您使用飞桨框架进行推理时，需要先安装飞桨框架。安装方法请参考 [飞桨框架安装](../../paddlepaddle_installation.md)。PaddleOCR 对飞桨框架的版本要求可以参考 [此处](../../paddleocr_and_paddlex.md#1-paddleocr-paddlex)。
+
+### 3.2 Transformers
+
+当您使用 Transformers 作为推理引擎时，需要安装 Hugging Face Transformers（`>=5.10.0`）。示例命令如下：
+
+```bash
+python -m pip install "transformers>=5.10.0"
+```
+
+通常，您还需要安装底层推理框架，详情可参考 [Transformers 官方文档](https://huggingface.co/docs/transformers/installation)。
+
+### 3.3 ONNX Runtime
+
+当您使用 ONNX Runtime 作为推理引擎时，需要安装 ONNX Runtime。示例命令如下：
+
+```bash
+python -m pip install onnxruntime-gpu
+```
+
+以上命令适用于 CUDA 12.x 环境下的 NVIDIA GPU，其他设备或环境的安装方式请参考 [ONNX Runtime 官方文档](https://onnxruntime.ai/docs/install/)。
+
+## 4. `engine` 和 `engine_config` 的设置与取值
+
+### 4.1 `engine`
+
+`engine` 用于指定推理引擎，可取值如下：
+
+| 取值 | 含义 | 说明 |
+| - | - | - |
+| `None` | 不显式指定引擎 | 自动确定推理引擎。保持 PaddleOCR 3.4 的行为，大多数情况下会使用飞桨框架推理。 |
+| `paddle` | 飞桨框架统一入口 | 自动选择 `paddle_static` 或 `paddle_dynamic`。 |
+| `paddle_static` | 静态图推理 | 使用飞桨静态图推理。 |
+| `paddle_dynamic` | 飞桨动态图推理 | 使用飞桨动态图推理。 |
+| `transformers` | Transformers 推理 | 使用 Hugging Face Transformers 推理。 |
+| `onnxruntime` | ONNX Runtime 推理 | 使用 ONNX Runtime 推理。 |
+
+### 4.2 `engine_config`
+
+`engine_config` 用于配置推理引擎，建议与 `engine` 搭配使用。各引擎常见的 `engine_config` 字段如下：
+
+#### `paddle_static`
+
+常见字段包括：
+
+- `run_mode`：运行模式，如 `paddle`、`trt_fp32`、`trt_fp16`、`mkldnn`；
+- `device_type` / `device_id`：设备类型和设备编号；
+- `cpu_threads`：CPU 推理线程数；
+- `delete_pass`：手动禁用的图优化 pass 列表；
+- `enable_new_ir`：是否启用新 IR；
+- `enable_cinn`：是否启用 CINN；
+- `trt_cfg_setting`：TensorRT 底层配置；
+- `trt_use_dynamic_shapes`：是否启用 TensorRT 动态形状；
+- `trt_collect_shape_range_info`：是否采集 shape range 信息；
+- `trt_discard_cached_shape_range_info`：是否丢弃已有 shape range 信息并重新采集；
+- `trt_dynamic_shapes`：动态形状配置；
+- `trt_dynamic_shape_input_data`：采集动态形状时用于填充输入张量的数据；
+- `trt_shape_range_info_path`：shape range 信息文件路径；
+- `trt_allow_rebuild_at_runtime`：运行时是否允许重建 TensorRT 引擎；
+- `mkldnn_cache_capacity`：oneDNN（MKLDNN）缓存容量。
+
+#### `paddle_dynamic`
+
+常见字段包括：
+
+- `device_type` / `device_id`：动态图执行时的设备类型和设备编号。
+
+#### `transformers`
+
+常见字段包括：
+
+- `dtype`：模型权重 / 推理使用的数据类型，如 `float16`；
+- `device_type` / `device_id`：推理设备类型和设备编号；
+- `trust_remote_code`：是否信任并执行模型仓库中的自定义代码；
+- `attn_implementation`：注意力实现方式，如 `flash_attention_2`；
+- `generation_config`：生成参数，如 `max_new_tokens`、`temperature`；
+- `model_kwargs`：传给模型加载接口的额外参数；
+- `processor_kwargs`：传给 processor / image processor 加载接口的额外参数；
+- `tokenizer_kwargs`：兼容保留字段，会与 `processor_kwargs` 合并使用。
+
+#### `onnxruntime`
+
+常见字段包括：
+
+- `device_type` / `device_id`：推理设备类型和设备编号；
+- `providers`：执行提供者列表（如 `CUDAExecutionProvider`、`CPUExecutionProvider`）；
+- `provider_options`：执行提供者专属配置；
+- `graph_optimization_level`：图优化级别；
+- `intra_op_num_threads`：节点内线程数；
+- `inter_op_num_threads`：节点间线程数；
+- `execution_mode`：执行模式（如 `sequential`、`parallel`）；
+- `log_severity_level`：日志严重级别；
+- `enable_mem_pattern`：是否启用内存模式；
+- `enable_cpu_mem_arena`：是否启用 CPU 内存池；
+- `session_options`：ONNX Runtime 会话选项。
+
+#### 4.2.1 扁平与分桶 `engine_config`
+
+同一层级的 `engine_config` 可以是：
+
+- **扁平**：只包含**当前解析得到的引擎**所需的字段（例如仅使用静态图时，顶层直接是 `run_mode`、`cpu_threads` 等）。
+- **分桶**：顶层键**仅**为 PaddleX 已注册的引擎名（如 `paddle_static`、`paddle_dynamic`、`transformers`、`onnxruntime` 等），每个键对应一个嵌套字典。**不得**在同一层级混用「分桶键」与扁平字段（例如 `{"paddle_static": {...}, "run_mode": "paddle"}` 会报错）。
+
+解析为某一引擎时，只会使用与该引擎对应的一份配置：扁平形式直接参与校验；分桶形式则取出对应键下的字典。
+
+### 4.3 优先级与覆盖规则
+
+- 对于产线，命令行参数或 Python API 初始化参数中传入的 `engine`、`engine_config`，优先级高于产线配置文件中的同名字段；
+- 在产线配置文件中，顶层 `engine`、`engine_config` 会作为全局配置，子模块或子产线中的 `engine`、`engine_config` 可覆盖上层配置；
+- 关于产线配置文件中更完整的优先级、覆盖和调用规则，建议参考 PaddleX 文档：[PaddleX 产线 Python 脚本使用说明](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/instructions/pipeline_python_API.html)。
+
+### 4.4 兼容性规则
+
+- 显式设置 `engine` 后，`enable_hpi` 不再生效；
+- 显式传入 `engine_config` 后，与该引擎对应的兼容参数会被忽略。例如在 `paddle` / `paddle_static` 场景下，`use_tensorrt`、`precision`、`enable_mkldnn`、`mkldnn_cache_capacity`、`cpu_threads`、`enable_cinn` 等兼容参数不再生效。
+
+## 5. 调用示例
+
+### 5.1 单模型（CLI）：通过 `--engine` 选择引擎
+
+```bash
+paddleocr text_detection -i general_ocr_001.png --engine transformers
+paddleocr text_detection -i general_ocr_001.png --engine onnxruntime
+```
+
+### 5.2 单模型（Python）：显式指定 `transformers`
+
+```python
+from paddleocr import TextDetection
+
+model = TextDetection(
+    model_name="PP-OCRv5_server_det",
+    engine="transformers",
+)
+
+result = model.predict("general_ocr_001.png")
+```
+
+### 5.3 单模型（Python）：显式指定 `onnxruntime`
+
+```python
+from paddleocr import TextDetection
+
+model = TextDetection(
+    model_name="PP-OCRv5_server_det",
+    engine="onnxruntime",
+)
+
+result = model.predict("general_ocr_001.png")
+```
+
+### 5.4 单模型（Python）：指定 `paddle_static` 与 `engine_config`
+
+```python
+from paddleocr import TextDetection
+
+model = TextDetection(
+    model_name="PP-OCRv5_server_det",
+    engine="paddle_static",
+    engine_config={
+        "device_type": "cpu",
+        "cpu_threads": 4,
+        "run_mode": "mkldnn",
+    },
+)
+
+result = model.predict("general_ocr_001.png")
+```
+
+### 5.5 产线（CLI）：通过 `--engine` 选择引擎
+
+```bash
+paddleocr ocr -i general_ocr_001.png --engine paddle_static
+```
+
+### 5.6 产线（Python API）：为某个模块单独配置推理引擎
+
+如需为产线中的某一个模块单独指定 `engine`、`engine_config`，可先导出配置文件，修改对应模块配置后，再通过加载配置文件。配置文件的导出、编辑与加载方式可参见 [使用 PaddleX 产线配置文件](../../paddleocr_and_paddlex.md#3-paddlex)。示例如下：
+
+首先，导出产线配置文件：
+
+```python
+from paddleocr import PaddleOCR
+
+pipeline = PaddleOCR()
+pipeline.export_paddlex_config_to_yaml("ocr_config.yaml")
+```
+
+然后，在 `ocr_config.yaml` 中为 `TextDetection` 模块单独设置 `engine`、`engine_config`：
+
+```yaml
+pipeline_name: OCR
+SubModules:
+  TextDetection:
+    engine: paddle_static
+    engine_config:
+      device_type: cpu
+      cpu_threads: 4
+      run_mode: mkldnn
+```
+
+使用更新后的配置文件完成推理：
+
+```python
+from paddleocr import PaddleOCR
+
+pipeline = PaddleOCR(paddlex_config="ocr_config.yaml")
+result = pipeline.predict("general_ocr_001.png")
+```
diff --git a/docs/version3.x/inference_deployment/local_inference/parallel_inference.en.md b/docs/version3.x/inference_deployment/local_inference/parallel_inference.en.md
new file mode 100644
index 0000000..428a512
--- /dev/null
+++ b/docs/version3.x/inference_deployment/local_inference/parallel_inference.en.md
@@ -0,0 +1,215 @@
+# Parallel Inference for Pipelines
+
+## Specifying Multiple Inference Devices
+
+For some pipelines, both the CLI and Python API of PaddleOCR support specifying multiple inference devices simultaneously. If multiple devices are specified, during pipeline initialization, an instance of the underlying pipeline class will be created on each device, and the received inputs will be processed using parallel inference. For example, for the document image preprocessing pipeline:
+
+```bash
+paddleocr doc_preprocessor \
+  --input input_images/ \
+  --device 'gpu:0,1,2,3' \
+  --use_doc_orientation_classify True \
+  --use_doc_unwarping True
+  --save_path ./output \
+
+```
+
+```python
+from paddleocr import DocPreprocessor
+
+
+pipeline = DocPreprocessor(device="gpu:0,1,2,3") 
+output = pipeline.predict(    
+    input="input_images/",
+    use_doc_orientation_classify=True,
+    use_doc_unwarping=True)
+
+```
+
+Both examples above use 4 GPUs (numbered 0, 1, 2, 3) to perform parallel inference on the `doc_test_rotated.jpg` image.
+
+When specifying multiple devices, the inference interface remains consistent with that of single-device usage. Please refer to the production line usage tutorial to check whether a specific production line supports multiple inference devices.
+
+## Example of Multi-Process Parallel Inference
+
+Beyond PaddleOCR's built-in multi-device parallel inference capability, users can also implement parallelism by wrapping PaddleOCR pipeline API calls themselves according to their specific scenario, with a view to achieving a better speedup. Below is an example of using Python multiprocessing to perform multi-GPU, multi-instance parallel processing on files in an input directory.
+
+
+
+```python
+import argparse
+import sys
+from multiprocessing import Manager, Process
+from pathlib import Path
+from queue import Empty
+
+import paddleocr
+
+
+def load_pipeline(class_name: str, device: str):
+    if not hasattr(paddleocr, class_name):
+        raise ValueError(f"Class {class_name} not found in paddleocr module.")
+    cls = getattr(paddleocr, class_name)
+    return cls(device=device)
+
+
+def worker(pipeline_class_path, device, task_queue, batch_size, output_dir):
+    pipeline = load_pipeline(pipeline_class_path, device)
+
+    should_end = False
+    batch = []
+
+    while not should_end:
+        try:
+            input_path = task_queue.get_nowait()
+        except Empty:
+            should_end = True
+        else:
+            batch.append(input_path)
+
+        if batch and (len(batch) == batch_size or should_end):
+            try:
+                for result in pipeline.predict(batch):
+                    input_path = Path(result["input_path"])
+                    if result.get("page_index") is not None:
+                        output_path = f"{input_path.stem}_{result['page_index']}.json"
+                    else:
+                        output_path = f"{input_path.stem}.json"
+                    output_path = str(Path(output_dir, output_path))
+                    result.save_to_json(output_path)
+                    print(f"Processed {repr(str(input_path))}")
+            except Exception as e:
+                print(
+                    f"Error processing {batch} on {repr(device)}: {e}",
+                    file=sys.stderr
+                )
+            batch.clear()
+
+
+def main():
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--pipeline",
+        type=str,
+        required=True,
+        help="PaddleOCR pipeline, e.g. 'DocPreprocessor'.",
+    )
+    parser.add_argument(
+        "--input_dir", type=str, required=True, help="Input directory."
+    )
+    parser.add_argument(
+        "--device",
+        type=str,
+        required=True,
+        help="Specifies the devices for performing parallel inference.",
+    )
+    parser.add_argument(
+        "--output_dir", type=str, default="output", help="Output directory."
+    )
+    parser.add_argument(
+        "--instances_per_device",
+        type=int,
+        default=1,
+        help="Number of pipeline instances per device.",
+    )
+    parser.add_argument(
+        "--batch_size",
+        type=int,
+        default=1,
+        help="Inference batch size for each pipeline instance.",
+    )
+    parser.add_argument(
+        "--input_glob_pattern",
+        type=str,
+        default="*",
+        help="Pattern to find the input files.",
+    )
+    args = parser.parse_args()
+
+    input_dir = Path(args.input_dir)
+    if not input_dir.exists():
+        print(f"The input directory does not exist: {input_dir}", file=sys.stderr)
+        return 2
+    if not input_dir.is_dir():
+        print(f"{repr(str(input_dir))} is not a directory.", file=sys.stderr)
+        return 2
+
+    output_dir = Path(args.output_dir)
+    if output_dir.exists() and not output_dir.is_dir():
+        print(f"{repr(str(output_dir))} is not a directory.", file=sys.stderr)
+        return 2
+    output_dir.mkdir(parents=True, exist_ok=True)
+
+    from paddlex.utils.device import constr_device, parse_device
+
+    device_type, device_ids = parse_device(args.device)
+    if device_ids is None or len(device_ids) == 1:
+        print(
+            "Please specify at least two devices for performing parallel inference.",
+            file=sys.stderr,
+        )
+        return 2
+
+    if args.batch_size <= 0:
+        print("Batch size must be greater than 0.", file=sys.stderr)
+        return 2
+
+    with Manager() as manager:
+        task_queue = manager.Queue()
+        for img_path in input_dir.glob(args.input_glob_pattern):
+            task_queue.put(str(img_path))
+
+        processes = []
+        for device_id in device_ids:
+            for _ in range(args.instances_per_device):
+                device = constr_device(device_type, [device_id])
+                p = Process(
+                    target=worker,
+                    args=(
+                        args.pipeline,
+                        device,
+                        task_queue,
+                        args.batch_size,
+                        str(output_dir),
+                    ),
+                )
+                p.start()
+                processes.append(p)
+
+        for p in processes:
+            p.join()
+
+    print("All done")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
+
+```
+Assuming the script is saved as `infer_mp.py`, here are some example commands for running it:
+```bash
+# For the exact value of the `--pipeline` parameter, please refer to the **script** import name of the pipeline
+# This is for the general layout analysis V3 pipeline, corresponding to `PPStructureV3`
+# Process all files in the `input_images` directory
+# Use GPUs 0, 1, 2, and 3, with 1 pipeline instance per GPU, and each instance processes 1 input file at a time
+python infer_mp.py \
+    --pipeline PPStructureV3 \
+    --input_dir input_images/ \
+    --device 'gpu:0,1,2,3' \
+    --output_dir output
+
+# General layout analysis V3 pipeline
+# Process all files with the `.jpg` suffix in the `input_images` directory
+# Use GPUs 0 and 2, with 2 pipeline instances per GPU, and each instance processes 4 input files at a time
+python infer_mp.py \
+    --pipeline PPStructureV3 \
+    --input_dir input_images/ \
+    --device 'gpu:0,2' \
+    --output_dir output \
+    --instances_per_device 2 \
+    --batch_size 4 \
+    --input_glob_pattern '*.jpg'
+
+```
diff --git a/docs/version3.x/inference_deployment/local_inference/parallel_inference.md b/docs/version3.x/inference_deployment/local_inference/parallel_inference.md
new file mode 100644
index 0000000..3bd25b8
--- /dev/null
+++ b/docs/version3.x/inference_deployment/local_inference/parallel_inference.md
@@ -0,0 +1,214 @@
+# 产线并行推理
+
+## 指定多个推理设备
+
+对于部分产线的 CLI 和 Python API，PaddleOCR 支持同时指定多个推理设备。如果指定了多个设备，产线初始化时将在每个设备上创建一个底层产线类对象的实例，并对接收到的输入进行并行推理。例如，对于文档图像预处理产线：
+
+```bash
+paddleocr doc_preprocessor \
+  --input input_images/ \
+  --device 'gpu:0,1,2,3' \
+  --use_doc_orientation_classify True \
+  --use_doc_unwarping True
+  --save_path ./output \
+
+```
+
+```python
+from paddleocr import DocPreprocessor
+
+
+pipeline = DocPreprocessor(device="gpu:0,1,2,3") 
+output = pipeline.predict(    
+    input="input_images/",
+    use_doc_orientation_classify=True,
+    use_doc_unwarping=True)
+    
+```
+
+以上两个例子均使用 4 块 GPU（编号为 0、1、2、3）对 `doc_test_rotated.jpg` 图片进行并行推理。
+
+指定多个设备时，推理接口仍然与指定单设备时保持一致。请查看产线使用教程以了解某一产线是否支持指定多个推理设备。
+
+## 多进程并行推理示例
+
+除了使用 PaddleOCR 内置的多设备并行推理功能外，用户也可以结合实际场景，通过封装 PaddleOCR 产线 API 调用来实现并行处理，从而获得更优的加速效果。如下是使用 Python 多进程实现多卡、多实例并行处理输入目录中的文件的示例代码：
+
+```python
+import argparse
+import sys
+from multiprocessing import Manager, Process
+from pathlib import Path
+from queue import Empty
+
+import paddleocr
+
+
+def load_pipeline(class_name: str, device: str):
+    if not hasattr(paddleocr, class_name):
+        raise ValueError(f"Class {class_name} not found in paddleocr module.")
+    cls = getattr(paddleocr, class_name)
+    return cls(device=device)
+
+
+def worker(pipeline_class_path, device, task_queue, batch_size, output_dir):
+    pipeline = load_pipeline(pipeline_class_path, device)
+
+    should_end = False
+    batch = []
+
+    while not should_end:
+        try:
+            input_path = task_queue.get_nowait()
+        except Empty:
+            should_end = True
+        else:
+            batch.append(input_path)
+
+        if batch and (len(batch) == batch_size or should_end):
+            try:
+                for result in pipeline.predict(batch):
+                    input_path = Path(result["input_path"])
+                    if result.get("page_index") is not None:
+                        output_path = f"{input_path.stem}_{result['page_index']}.json"
+                    else:
+                        output_path = f"{input_path.stem}.json"
+                    output_path = str(Path(output_dir, output_path))
+                    result.save_to_json(output_path)
+                    print(f"Processed {repr(str(input_path))}")
+            except Exception as e:
+                print(
+                    f"Error processing {batch} on {repr(device)}: {e}",
+                    file=sys.stderr
+                )
+            batch.clear()
+
+
+def main():
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--pipeline",
+        type=str,
+        required=True,
+        help="PaddleOCR pipeline, e.g. 'DocPreprocessor'.",
+    )
+    parser.add_argument(
+        "--input_dir", type=str, required=True, help="Input directory."
+    )
+    parser.add_argument(
+        "--device",
+        type=str,
+        required=True,
+        help="Specifies the devices for performing parallel inference.",
+    )
+    parser.add_argument(
+        "--output_dir", type=str, default="output", help="Output directory."
+    )
+    parser.add_argument(
+        "--instances_per_device",
+        type=int,
+        default=1,
+        help="Number of pipeline instances per device.",
+    )
+    parser.add_argument(
+        "--batch_size",
+        type=int,
+        default=1,
+        help="Inference batch size for each pipeline instance.",
+    )
+    parser.add_argument(
+        "--input_glob_pattern",
+        type=str,
+        default="*",
+        help="Pattern to find the input files.",
+    )
+    args = parser.parse_args()
+
+    input_dir = Path(args.input_dir)
+    if not input_dir.exists():
+        print(f"The input directory does not exist: {input_dir}", file=sys.stderr)
+        return 2
+    if not input_dir.is_dir():
+        print(f"{repr(str(input_dir))} is not a directory.", file=sys.stderr)
+        return 2
+
+    output_dir = Path(args.output_dir)
+    if output_dir.exists() and not output_dir.is_dir():
+        print(f"{repr(str(output_dir))} is not a directory.", file=sys.stderr)
+        return 2
+    output_dir.mkdir(parents=True, exist_ok=True)
+
+    from paddlex.utils.device import constr_device, parse_device
+
+    device_type, device_ids = parse_device(args.device)
+    if device_ids is None or len(device_ids) == 1:
+        print(
+            "Please specify at least two devices for performing parallel inference.",
+            file=sys.stderr,
+        )
+        return 2
+
+    if args.batch_size <= 0:
+        print("Batch size must be greater than 0.", file=sys.stderr)
+        return 2
+
+    with Manager() as manager:
+        task_queue = manager.Queue()
+        for img_path in input_dir.glob(args.input_glob_pattern):
+            task_queue.put(str(img_path))
+
+        processes = []
+        for device_id in device_ids:
+            for _ in range(args.instances_per_device):
+                device = constr_device(device_type, [device_id])
+                p = Process(
+                    target=worker,
+                    args=(
+                        args.pipeline,
+                        device,
+                        task_queue,
+                        args.batch_size,
+                        str(output_dir),
+                    ),
+                )
+                p.start()
+                processes.append(p)
+
+        for p in processes:
+            p.join()
+
+    print("All done")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
+
+```
+
+假设将上述脚本存储为 `infer_mp.py`，以下是一些调用示例：
+
+```bash
+# 确定 `--pipeline` 参数需查看其产线 **脚本方式** 导入类名称
+# 此处为通用版面解析 v3 产线，对应PPStructureV3
+# 处理 `input_images` 目录中所有文件
+# 使用 GPU 0、1、2、3，每块 GPU 上 1 个产线实例，每个实例一次处理 1 个输入文件
+python infer_mp.py \
+    --pipeline PPStructureV3 \
+    --input_dir input_images/ \
+    --device 'gpu:0,1,2,3' \
+    --output_dir output
+
+# 通用版面解析 v3 产线
+# 处理 `input_images` 目录中所有后缀为 `.jpg` 的文件
+# 使用 GPU 0、2，每块 GPU 上 2 个产线实例，每个实例一次处理 4 个输入文件
+python infer_mp.py \
+    --pipeline PPStructureV3 \
+    --input_dir input_images/ \
+    --device 'gpu:0,2' \
+    --output_dir output \
+    --instances_per_device 2 \
+    --batch_size 4 \
+    --input_glob_pattern '*.jpg'
+```
diff --git a/docs/version3.x/inference_deployment/others/benchmark.en.md b/docs/version3.x/inference_deployment/others/benchmark.en.md
new file mode 100644
index 0000000..6443f99
--- /dev/null
+++ b/docs/version3.x/inference_deployment/others/benchmark.en.md
@@ -0,0 +1,460 @@
+# Pipeline Benchmark
+
+## Table of Contents
+
+- [1. Instructions](#1-instructions)
+- [2. Usage Examples](#2-usage-examples)
+
+## 1. Instructions
+
+The Benchmark feature calculates the average execution time of all operations during end-to-end pipeline inference and provides summary information. The time unit is milliseconds.
+
+The benchmark feature needs to be enabled via environment variables as follows:
+
+* `PADDLE_PDX_PIPELINE_BENCHMARK`: Set to `True` to enable the benchmark feature. The default is `False`.
+
+The following table describes the methods and parameters related to pipeline inference benchmark:
+
+<table border="1" width="100%" cellpadding="5">
+    <tr>
+        <th>Method Name</th>
+        <th>Description</th>
+    </tr>
+    <tr>
+        <td><code>start_warmup()</code></td>
+        <td>Start the benchmark warmup.</td>
+    </tr>
+    <tr>
+        <td><code>stop_warmup()</code></td>
+        <td>Stop the warmup and clear all benchmark data generated during warmup.</td>
+    </tr>
+        <tr>
+        <td><code>print_detail_data()</code></td>
+        <td>Print detailed benchmark data, including the sequence (Step), name (Operation), and average time (Time) of each operation.</td>
+    </tr>
+    <tr>
+        <td><code>print_summary_data()</code></td>
+        <td>Print summary benchmark data, including the level (Level), name (Operation), and total average time (Time) of each operation. The Time for level 1 represents the total average time.</td>
+    </tr>
+    <tr>
+        <td><code>print_operation_info()</code></td>
+        <td>Print the source code location of each operation.</td>
+    </tr>
+    <tr>
+        <td><code>print_pipeline_data()</code></td>
+        <td>Print the detail data, summary data, and operation_info data of the benchmark to the console.</td>
+    </tr>
+    <tr>
+        <td><code>save_pipeline_data(save_path)</code></td>
+        <td><code>save_path</code>: string <br />Save the benchmark data to the specified file path, including detailed benchmark data <code>detail.csv</code> and summary benchmark data <code>summary.csv</code>.</td>
+    </tr>
+    <tr>
+        <td><code>reset()</code></td>
+        <td>Clear existing benchmark data.</td>
+    </tr>
+</table>
+
+
+## 2. Usage Examples
+
+Create a `test_infer.py` script:
+
+```python
+import os
+
+os.environ["PADDLE_PDX_PIPELINE_BENCHMARK"] = "True"
+
+from paddleocr import PaddleOCR, benchmark
+
+pipeline = PaddleOCR()
+image = "general_ocr_002.png"
+
+benchmark.start_warmup() # Start warmup
+for _ in range(50):
+    pipeline.predict(image)
+benchmark.stop_warmup() # End warmup
+
+for _ in range(100): # Start formal speed measurement
+    pipeline.predict(image)
+
+benchmark.print_pipeline_data()  # Print summary benchmark data
+benchmark.save_pipeline_data("./benchmark") # Save benchmark data to the benchmark folder
+```
+
+Execute the script:
+
+```bash
+python test_infer.py
+```
+
+The benchmark results obtained from running the example program are as follows:
+
+```
+                                                             Operation Info
++-----------------------------------------------------+-------------------------------------------------------------------------------+
+|                      Operation                      |                              Source Code Location                             |
++-----------------------------------------------------+-------------------------------------------------------------------------------+
+|                      ReadImage                      |       /PaddleX/paddlex/inference/common/reader/image_reader.py:47       |
+|                   DocTrPostProcess                  |    /PaddleX/paddlex/inference/models/image_unwarping/processors.py:51   |
+|                   DetResizeForTest                  |    /PaddleX/paddlex/inference/models/text_detection/processors.py:58    |
+|     _DocPreprocessorPipeline.get_model_settings     |  /PaddleX/paddlex/inference/pipelines/doc_preprocessor/pipeline.py:110  |
+|                         Crop                        | /PaddleX/paddlex/inference/models/image_classification/processors.py:45 |
+|                PaddleInferChainLegacy               |       /PaddleX/paddlex/inference/models/common/static_infer.py:248      |
+|              _OCRPipeline.rotate_image              |         /PaddleX/paddlex/inference/pipelines/ocr/pipeline.py:140        |
+|           _DocPreprocessorPipeline.predict          |  /PaddleX/paddlex/inference/pipelines/doc_preprocessor/pipeline.py:133  |
+|                 ClasPredictor.apply                 |  /PaddleX/paddlex/inference/models/base/predictor/base_predictor.py:213 |
+|                 WarpPredictor.apply                 |  /PaddleX/paddlex/inference/models/base/predictor/base_predictor.py:213 |
+|                TextDetPredictor.apply               |  /PaddleX/paddlex/inference/models/base/predictor/base_predictor.py:213 |
+|                    ResizeByShort                    |    /PaddleX/paddlex/inference/models/common/vision/processors.py:203    |
+|                      Normalize                      |    /PaddleX/paddlex/inference/models/common/vision/processors.py:268    |
+|                    DBPostProcess                    |    /PaddleX/paddlex/inference/models/text_detection/processors.py:487   |
+|                TextRecPredictor.apply               |  /PaddleX/paddlex/inference/models/base/predictor/base_predictor.py:213 |
+|                    NormalizeImage                   |    /PaddleX/paddlex/inference/models/text_detection/processors.py:252   |
+| _DocPreprocessorPipeline.check_model_settings_valid |   /PaddleX/paddlex/inference/pipelines/doc_preprocessor/pipeline.py:82  |
+|           _OCRPipeline.get_model_settings           |         /PaddleX/paddlex/inference/pipelines/ocr/pipeline.py:204        |
+|           _OCRPipeline.get_text_det_params          |         /PaddleX/paddlex/inference/pipelines/ocr/pipeline.py:236        |
+|                        Resize                       |    /PaddleX/paddlex/inference/models/common/vision/processors.py:117    |
+|                    CTCLabelDecode                   |   /PaddleX/paddlex/inference/models/text_recognition/processors.py:189  |
+|                         Topk                        | /PaddleX/paddlex/inference/models/image_classification/processors.py:83 |
+|                  OCRReisizeNormImg                  |   /PaddleX/paddlex/inference/models/text_recognition/processors.py:65   |
+|                       ToBatch                       |   /PaddleX/paddlex/inference/models/text_recognition/processors.py:235  |
+|                      ToCHWImage                     |    /PaddleX/paddlex/inference/models/common/vision/processors.py:277    |
+|                 _OCRPipeline.predict                |         /PaddleX/paddlex/inference/pipelines/ocr/pipeline.py:282        |
+|                       ToBatch                       |    /PaddleX/paddlex/inference/models/common/vision/processors.py:284    |
+|       _OCRPipeline.check_model_settings_valid       |         /PaddleX/paddlex/inference/pipelines/ocr/pipeline.py:176        |
++-----------------------------------------------------+-------------------------------------------------------------------------------+
+                                           Detail Data
++------+----------------------------------------------------------------+-----------------------+
+| Step | Operation                                                      | Time                  |
++------+----------------------------------------------------------------+-----------------------+
+|  1   | _OCRPipeline.predict                                           | 375.11244628956774    |
+|  2   |     -> _OCRPipeline.get_model_settings                         | 0.00428391998866573   |
+|  3   |     -> _OCRPipeline.check_model_settings_valid                 | 0.0024828016466926783 |
+|  4   |     -> _OCRPipeline.get_text_det_params                        | 0.005152080120751634  |
+|  5   |     -> ReadImage                                               | 3.2029549301660154    |
+|  6   |     -> _DocPreprocessorPipeline.predict                        | 27.310913350374904    |
+|  7   |         -> _DocPreprocessorPipeline.get_model_settings         | 0.004107539862161502  |
+|  8   |         -> _DocPreprocessorPipeline.check_model_settings_valid | 0.0016830896493047476 |
+|  9   |         -> ReadImage                                           | 0.0029576495580840856 |
+|  10  |         -> ClasPredictor.apply                                 | 4.701614730001893     |
+|  11  |             -> ReadImage                                       | 0.13587839042884298   |
+|  12  |             -> ResizeByShort                                   | 0.3281894406245556    |
+|  13  |             -> Crop                                            | 0.01503000981756486   |
+|  14  |             -> Normalize                                       | 0.3884544402535539    |
+|  15  |             -> ToCHWImage                                      | 0.006330519245238975  |
+|  16  |             -> ToBatch                                         | 0.14169737987685949   |
+|  17  |             -> PaddleInferChainLegacy                          | 3.283889550511958     |
+|  18  |             -> Topk                                            | 0.10010718091507442   |
+|  19  |         -> WarpPredictor.apply                                 | 21.893062600429403    |
+|  20  |             -> ReadImage                                       | 0.004573430051095784  |
+|  21  |             -> Normalize                                       | 4.245691860560328     |
+|  22  |             -> ToCHWImage                                      | 0.005895959911867976  |
+|  23  |             -> ToBatch                                         | 1.7250755491841119    |
+|  24  |             -> PaddleInferChainLegacy                          | 10.887994960212382    |
+|  25  |             -> DocTrPostProcess                                | 1.4253830898087472    |
+|  26  |     -> TextDetPredictor.apply                                  | 49.976056129235076    |
+|  27  |         -> ReadImage                                           | 0.004843260976485908  |
+|  28  |         -> DetResizeForTest                                    | 3.3269549095712136    |
+|  29  |         -> NormalizeImage                                      | 2.9576204597833566    |
+|  30  |         -> ToCHWImage                                          | 0.005182310123927891  |
+|  31  |         -> ToBatch                                             | 1.046062790119322     |
+|  32  |         -> PaddleInferChainLegacy                              | 34.70224040953326     |
+|  33  |         -> DBPostProcess                                       | 5.826775671303039     |
+|  34  |     -> ClasPredictor.apply                                     | 23.43678753997665     |
+|  35  |         -> ReadImage                                           | 0.0633359991479665    |
+|  36  |         -> Resize                                              | 0.24419097986537963   |
+|  37  |         -> Normalize                                           | 0.480741420033155     |
+|  38  |         -> ToCHWImage                                          | 0.0066608507768251    |
+|  39  |         -> ToBatch                                             | 0.18536171046434902   |
+|  40  |         -> PaddleInferChainLegacy                              | 3.3766339404974133    |
+|  41  |         -> Topk                                                | 0.15909907990135252   |
+|  42  |         -> ReadImage                                           | 0.0395357194065582    |
+|  43  |         -> Resize                                              | 0.2085290702234488    |
+|  44  |         -> Normalize                                           | 0.4068155895220116    |
+|  45  |         -> ToCHWImage                                          | 0.005677459557773545  |
+|  46  |         -> ToBatch                                             | 0.11155156986205839   |
+|  47  |         -> PaddleInferChainLegacy                              | 2.7268862597702537    |
+|  48  |         -> Topk                                                | 0.13428127014776692   |
+|  49  |         -> ReadImage                                           | 0.032502070971531793  |
+|  50  |         -> Resize                                              | 0.20152631899691187   |
+|  51  |         -> Normalize                                           | 0.347195100330282     |
+|  52  |         -> ToCHWImage                                          | 0.005517759709618986  |
+|  53  |         -> ToBatch                                             | 0.10656953061698005   |
+|  54  |         -> PaddleInferChainLegacy                              | 2.612808299745666     |
+|  55  |         -> Topk                                                | 0.13188434022595175   |
+|  56  |         -> ReadImage                                           | 0.03589507090509869   |
+|  57  |         -> Resize                                              | 0.2076980892161373    |
+|  58  |         -> Normalize                                           | 0.3592138692329172    |
+|  59  |         -> ToCHWImage                                          | 0.005206359783187509  |
+|  60  |         -> ToBatch                                             | 0.1359267797670327    |
+|  61  |         -> PaddleInferChainLegacy                              | 2.619662079960108     |
+|  62  |         -> Topk                                                | 0.130717080028262     |
+|  63  |         -> ReadImage                                           | 0.038393009890569374  |
+|  64  |         -> Resize                                              | 0.19743553988519125   |
+|  65  |         -> Normalize                                           | 0.33197281998582184   |
+|  66  |         -> ToCHWImage                                          | 0.00512515107402578   |
+|  67  |         -> ToBatch                                             | 0.10293568033375777   |
+|  68  |         -> PaddleInferChainLegacy                              | 2.5824282996472903    |
+|  69  |         -> Topk                                                | 0.129485729848966     |
+|  70  |         -> ReadImage                                           | 0.04028105002362281   |
+|  71  |         -> Resize                                              | 0.10972122952807695   |
+|  72  |         -> Normalize                                           | 0.1787920702190604    |
+|  73  |         -> ToCHWImage                                          | 0.00408922991482541   |
+|  74  |         -> ToBatch                                             | 0.05458273953991011   |
+|  75  |         -> PaddleInferChainLegacy                              | 2.262636839877814     |
+|  76  |         -> Topk                                                | 0.1055472502775956    |
+|  77  |     -> _OCRPipeline.rotate_image                               | 0.05102259965497069   |
+|  78  |     -> TextRecPredictor.apply                                  | 169.44437422047486    |
+|  79  |         -> ReadImage                                           | 0.004737989947898313  |
+|  80  |         -> OCRReisizeNormImg                                   | 0.46037410967983305   |
+|  81  |         -> ToBatch                                             | 0.6405122207070235    |
+|  82  |         -> PaddleInferChainLegacy                              | 15.439773340767715    |
+|  83  |         -> CTCLabelDecode                                      | 10.742378439754248    |
+|  84  |         -> ReadImage                                           | 0.006349970353767276  |
+|  85  |         -> OCRReisizeNormImg                                   | 0.6252558408596087    |
+|  86  |         -> ToBatch                                             | 0.7338531101413537    |
+|  87  |         -> PaddleInferChainLegacy                              | 15.204189889482222    |
+|  88  |         -> CTCLabelDecode                                      | 6.7516070799320005    |
+|  89  |         -> ReadImage                                           | 0.006978959863772616  |
+|  90  |         -> OCRReisizeNormImg                                   | 0.7167729703360237    |
+|  91  |         -> ToBatch                                             | 0.6568272292497568    |
+|  92  |         -> PaddleInferChainLegacy                              | 14.973864750063512    |
+|  93  |         -> CTCLabelDecode                                      | 6.695752280211309     |
+|  94  |         -> ReadImage                                           | 0.0070425499870907515 |
+|  95  |         -> OCRReisizeNormImg                                   | 0.7757280093210284    |
+|  96  |         -> ToBatch                                             | 0.6442721793428063    |
+|  97  |         -> PaddleInferChainLegacy                              | 15.027350780292181    |
+|  98  |         -> CTCLabelDecode                                      | 6.661591530573787     |
+|  99  |         -> ReadImage                                           | 0.007066540565574542  |
+| 100  |         -> OCRReisizeNormImg                                   | 0.9195591000025161    |
+| 101  |         -> ToBatch                                             | 0.7951801503077149    |
+| 102  |         -> PaddleInferChainLegacy                              | 15.379044259898365    |
+| 103  |         -> CTCLabelDecode                                      | 9.372330370388227     |
+| 104  |         -> ReadImage                                           | 0.006225309771252796  |
+| 105  |         -> OCRReisizeNormImg                                   | 1.1437026296334807    |
+| 106  |         -> ToBatch                                             | 1.091715270158602     |
+| 107  |         -> PaddleInferChainLegacy                              | 23.505835609685164    |
+| 108  |         -> CTCLabelDecode                                      | 17.118994210031815    |
++------+----------------------------------------------------------------+-----------------------+
+                                      Summary Data
++-------+-----------------------------------------------------+-----------------------+
+| Level | Operation                                           | Time                  |
++-------+-----------------------------------------------------+-----------------------+
+|   1   | _OCRPipeline.predict                                | 375.11244628956774    |
+|       |                                                     |                       |
+|   2   | Layer                                               | 375.11244628956774    |
+|       | Core                                                | 273.4340275716386     |
+|       | Other                                               | 101.67841871792916    |
+|       | _OCRPipeline.get_model_settings                     | 0.00428391998866573   |
+|       | _OCRPipeline.check_model_settings_valid             | 0.0024828016466926783 |
+|       | _OCRPipeline.get_text_det_params                    | 0.005152080120751634  |
+|       | ReadImage                                           | 3.2029549301660154    |
+|       | _DocPreprocessorPipeline.predict                    | 27.310913350374904    |
+|       | TextDetPredictor.apply                              | 49.976056129235076    |
+|       | ClasPredictor.apply                                 | 23.43678753997665     |
+|       | _OCRPipeline.rotate_image                           | 0.05102259965497069   |
+|       | TextRecPredictor.apply                              | 169.44437422047486    |
+|       |                                                     |                       |
+|   3   | Layer                                               | 270.1681312400615     |
+|       | Core                                                | 261.8130224109336     |
+|       | Other                                               | 8.355108829127857     |
+|       | _DocPreprocessorPipeline.get_model_settings         | 0.004107539862161502  |
+|       | _DocPreprocessorPipeline.check_model_settings_valid | 0.0016830896493047476 |
+|       | ReadImage                                           | 0.29614515136927366   |
+|       | ClasPredictor.apply                                 | 4.701614730001893     |
+|       | WarpPredictor.apply                                 | 21.893062600429403    |
+|       | DetResizeForTest                                    | 3.3269549095712136    |
+|       | NormalizeImage                                      | 2.9576204597833566    |
+|       | ToCHWImage                                          | 0.03745912094018422   |
+|       | ToBatch                                             | 6.305350960610667     |
+|       | PaddleInferChainLegacy                              | 150.41335475922097    |
+|       | DBPostProcess                                       | 5.826775671303039     |
+|       | Resize                                              | 1.1691012277151458    |
+|       | Normalize                                           | 2.104730869323248     |
+|       | Topk                                                | 0.7910147504298948    |
+|       | OCRReisizeNormImg                                   | 4.641392659832491     |
+|       | CTCLabelDecode                                      | 57.342653910891386    |
+|       |                                                     |                       |
+|   4   | Layer                                               | 26.594677330431296    |
+|       | Core                                                | 22.69419176140218     |
+|       | Other                                               | 3.900485569029115     |
+|       | ReadImage                                           | 0.14045182047993876   |
+|       | ResizeByShort                                       | 0.3281894406245556    |
+|       | Crop                                                | 0.01503000981756486   |
+|       | Normalize                                           | 4.634146300813882     |
+|       | ToCHWImage                                          | 0.012226479157106951  |
+|       | ToBatch                                             | 1.8667729290609714    |
+|       | PaddleInferChainLegacy                              | 14.17188451072434     |
+|       | Topk                                                | 0.10010718091507442   |
+|       | DocTrPostProcess                                    | 1.4253830898087472    |
++-------+-----------------------------------------------------+-----------------------+
+```
+
+The results will be saved locally to: `./benchmark/detail.csv` and `./benchmark/summary.csv`:
+
+The content of `detail.csv` is as follows:
+
+```
+Step,Operation,Time
+1,_OCRPipeline.predict,375.11244628956774
+2,    -> _OCRPipeline.get_model_settings,0.00428391998866573
+3,    -> _OCRPipeline.check_model_settings_valid,0.0024828016466926783
+4,    -> _OCRPipeline.get_text_det_params,0.005152080120751634
+5,    -> ReadImage,3.2029549301660154
+6,    -> _DocPreprocessorPipeline.predict,27.310913350374904
+7,        -> _DocPreprocessorPipeline.get_model_settings,0.004107539862161502
+8,        -> _DocPreprocessorPipeline.check_model_settings_valid,0.0016830896493047476
+9,        -> ReadImage,0.0029576495580840856
+10,        -> ClasPredictor.apply,4.701614730001893
+11,            -> ReadImage,0.13587839042884298
+12,            -> ResizeByShort,0.3281894406245556
+13,            -> Crop,0.01503000981756486
+14,            -> Normalize,0.3884544402535539
+15,            -> ToCHWImage,0.006330519245238975
+16,            -> ToBatch,0.14169737987685949
+17,            -> PaddleInferChainLegacy,3.283889550511958
+18,            -> Topk,0.10010718091507442
+19,        -> WarpPredictor.apply,21.893062600429403
+20,            -> ReadImage,0.004573430051095784
+21,            -> Normalize,4.245691860560328
+22,            -> ToCHWImage,0.005895959911867976
+23,            -> ToBatch,1.7250755491841119
+24,            -> PaddleInferChainLegacy,10.887994960212382
+25,            -> DocTrPostProcess,1.4253830898087472
+26,    -> TextDetPredictor.apply,49.976056129235076
+27,        -> ReadImage,0.004843260976485908
+28,        -> DetResizeForTest,3.3269549095712136
+29,        -> NormalizeImage,2.9576204597833566
+30,        -> ToCHWImage,0.005182310123927891
+31,        -> ToBatch,1.046062790119322
+32,        -> PaddleInferChainLegacy,34.70224040953326
+33,        -> DBPostProcess,5.826775671303039
+34,    -> ClasPredictor.apply,23.43678753997665
+35,        -> ReadImage,0.0633359991479665
+36,        -> Resize,0.24419097986537963
+37,        -> Normalize,0.480741420033155
+38,        -> ToCHWImage,0.0066608507768251
+39,        -> ToBatch,0.18536171046434902
+40,        -> PaddleInferChainLegacy,3.3766339404974133
+41,        -> Topk,0.15909907990135252
+42,        -> ReadImage,0.0395357194065582
+43,        -> Resize,0.2085290702234488
+44,        -> Normalize,0.4068155895220116
+45,        -> ToCHWImage,0.005677459557773545
+46,        -> ToBatch,0.11155156986205839
+47,        -> PaddleInferChainLegacy,2.7268862597702537
+48,        -> Topk,0.13428127014776692
+49,        -> ReadImage,0.032502070971531793
+50,        -> Resize,0.20152631899691187
+51,        -> Normalize,0.347195100330282
+52,        -> ToCHWImage,0.005517759709618986
+53,        -> ToBatch,0.10656953061698005
+54,        -> PaddleInferChainLegacy,2.612808299745666
+55,        -> Topk,0.13188434022595175
+56,        -> ReadImage,0.03589507090509869
+57,        -> Resize,0.2076980892161373
+58,        -> Normalize,0.3592138692329172
+59,        -> ToCHWImage,0.005206359783187509
+60,        -> ToBatch,0.1359267797670327
+61,        -> PaddleInferChainLegacy,2.619662079960108
+62,        -> Topk,0.130717080028262
+63,        -> ReadImage,0.038393009890569374
+64,        -> Resize,0.19743553988519125
+65,        -> Normalize,0.33197281998582184
+66,        -> ToCHWImage,0.00512515107402578
+67,        -> ToBatch,0.10293568033375777
+68,        -> PaddleInferChainLegacy,2.5824282996472903
+69,        -> Topk,0.129485729848966
+70,        -> ReadImage,0.04028105002362281
+71,        -> Resize,0.10972122952807695
+72,        -> Normalize,0.1787920702190604
+73,        -> ToCHWImage,0.00408922991482541
+74,        -> ToBatch,0.05458273953991011
+75,        -> PaddleInferChainLegacy,2.262636839877814
+76,        -> Topk,0.1055472502775956
+77,    -> _OCRPipeline.rotate_image,0.05102259965497069
+78,    -> TextRecPredictor.apply,169.44437422047486
+79,        -> ReadImage,0.004737989947898313
+80,        -> OCRReisizeNormImg,0.46037410967983305
+81,        -> ToBatch,0.6405122207070235
+82,        -> PaddleInferChainLegacy,15.439773340767715
+83,        -> CTCLabelDecode,10.742378439754248
+84,        -> ReadImage,0.006349970353767276
+85,        -> OCRReisizeNormImg,0.6252558408596087
+86,        -> ToBatch,0.7338531101413537
+87,        -> PaddleInferChainLegacy,15.204189889482222
+88,        -> CTCLabelDecode,6.7516070799320005
+89,        -> ReadImage,0.006978959863772616
+90,        -> OCRReisizeNormImg,0.7167729703360237
+91,        -> ToBatch,0.6568272292497568
+92,        -> PaddleInferChainLegacy,14.973864750063512
+93,        -> CTCLabelDecode,6.695752280211309
+94,        -> ReadImage,0.0070425499870907515
+95,        -> OCRReisizeNormImg,0.7757280093210284
+96,        -> ToBatch,0.6442721793428063
+97,        -> PaddleInferChainLegacy,15.027350780292181
+98,        -> CTCLabelDecode,6.661591530573787
+99,        -> ReadImage,0.007066540565574542
+100,        -> OCRReisizeNormImg,0.9195591000025161
+101,        -> ToBatch,0.7951801503077149
+102,        -> PaddleInferChainLegacy,15.379044259898365
+103,        -> CTCLabelDecode,9.372330370388227
+104,        -> ReadImage,0.006225309771252796
+105,        -> OCRReisizeNormImg,1.1437026296334807
+106,        -> ToBatch,1.091715270158602
+107,        -> PaddleInferChainLegacy,23.505835609685164
+108,        -> CTCLabelDecode,17.118994210031815
+```
+
+The content of `summary.csv` is as follows:
+
+```
+Level, Operation, Time
+1, _OCRPipeline.predict, 375.11244628956774
+,,
+2, Layer, 375.11244628956774
+, Core, 273.4340275716386
+, Other, 101.67841871792916
+, _OCRPipeline.get_model_settings, 0.00428391998866573
+, _OCRPipeline.check_model_settings_valid, 0.0024828016466926783
+, _OCRPipeline.get_text_det_params, 0.005152080120751634
+, ReadImage, 3.2029549301660154
+, _DocPreprocessorPipeline.predict, 27.310913350374904
+, TextDetPredictor.apply, 49.976056129235076
+, ClasPredictor.apply, 23.43678753997665
+, _OCRPipeline.rotate_image, 0.05102259965497069
+, TextRecPredictor.apply, 169.44437422047486
+,,
+3, Layer, 270.1681312400615
+, Core, 261.8130224109336
+, Other, 8.355108829127857
+, _DocPreprocessorPipeline.get_model_settings, 0.004107539862161502
+, _DocPreprocessorPipeline.check_model_settings_valid, 0.0016830896493047476
+, ReadImage, 0.29614515136927366
+, ClasPredictor.apply, 4.701614730001893
+, WarpPredictor.apply, 21.893062600429403
+, DetResizeForTest, 3.3269549095712136
+, NormalizeImage, 2.9576204597833566
+, ToCHWImage, 0.03745912094018422
+, ToBatch, 6.305350960610667
+, PaddleInferChainLegacy, 150.41335475922097
+, DBPostProcess, 5.826775671303039
+, Resize, 1.1691012277151458
+, Normalize, 2.104730869323248
+, Topk, 0.7910147504298948
+, OCRReisizeNormImg, 4.641392659832491
+, CTCLabelDecode, 57.342653910891386
+,,
+4, Layer, 26.594677330431296
+, Core, 22.69419176140218
+, Other, 3.900485569029115
+, ReadImage, 0.14045182047993876
+, ResizeByShort, 0.3281894406245556
+, Crop, 0.01503000981756486
+, Normalize, 4.634146300813882
+, ToCHWImage, 0.012226479157106951
+, ToBatch, 1.8667729290609714
+, PaddleInferChainLegacy, 14.17188451072434
+, Topk, 0.10010718091507442
+, DocTrPostProcess, 1.4253830898087472
+```
diff --git a/docs/version3.x/inference_deployment/others/benchmark.md b/docs/version3.x/inference_deployment/others/benchmark.md
new file mode 100644
index 0000000..e9b4dfd
--- /dev/null
+++ b/docs/version3.x/inference_deployment/others/benchmark.md
@@ -0,0 +1,460 @@
+# 产线推理 Benchmark
+
+## 目录
+
+- [1. 使用说明](#1)
+- [2. 使用示例](#2)
+
+## 1. 使用说明
+
+Benchmark 功能会统计产线在端到端推理过程中，所有操作的平均执行时间，并给出汇总信息。耗时数据单位为毫秒。
+
+需通过环境变量启用 benchmark 功能，具体如下：
+
+* `PADDLE_PDX_PIPELINE_BENCHMARK`：设置为 `True` 时则开启 benchmark 功能，默认为 `False`；
+
+产线推理benchmark相关方法、参数说明如下：
+
+<table border="1" width="100%" cellpadding="5">
+    <tr>
+        <th>方法名称</th>
+        <th>描述</th>
+    </tr>
+    <tr>
+        <td><code>start_warmup()</code></td>
+        <td>开始benchmark的warmup。</td>
+    </tr>
+    <tr>
+        <td><code>stop_warmup()</code></td>
+        <td>结束warmup，会清除warmup时产生的所有benchmark数据。</td>
+    </tr>
+        <tr>
+        <td><code>print_detail_data()</code></td>
+        <td>打印详细的benchmark数据，包括每个操作的顺序（Step）、名称（Operation）、平均耗时（Time）。</td>
+    </tr>
+    <tr>
+        <td><code>print_summary_data()</code></td>
+        <td>打印汇总的benchmark数据，包括每个操作的层级（Level）、名称（Operation）、总平均耗时（Time）。level为1的Time即为总平均耗时。</td>
+    </tr>
+    <tr>
+        <td><code>print_operation_info()</code></td>
+        <td>打印每个操作的源代码位置。</td>
+    </tr>
+    <tr>
+        <td><code>print_pipeline_data()</code></td>
+        <td>打印benchmark的detail数据、summary数据和operation_info数据到控制台。</td>
+    </tr>
+    <tr>
+        <td><code>save_pipeline_data(save_path)</code></td>
+        <td><code>save_path</code>: string <br />保存benchmark数据的文件路径，包含详细的benchmark数据 <code>detail.csv</code> 和汇总的benchmark数据 <code>summary.csv</code>。</td>
+    </tr>
+    <tr>
+        <td><code>reset()</code></td>
+        <td>清除已有的benchmark数据。</td>
+    </tr>
+</table>
+
+
+## 2. 使用示例
+
+创建 `test_infer.py` 脚本：
+
+```python
+import os
+
+os.environ["PADDLE_PDX_PIPELINE_BENCHMARK"] = "True"
+
+from paddleocr import PaddleOCR, benchmark
+
+pipeline = PaddleOCR()
+image = "general_ocr_002.png"
+
+benchmark.start_warmup() # warmup开始
+for _ in range(50):
+    pipeline.predict(image)
+benchmark.stop_warmup() # warmup结束
+
+for _ in range(100): # 开始正式测速
+    pipeline.predict(image)
+
+benchmark.print_pipeline_data()  # 打印汇总的benchmark数据
+benchmark.save_pipeline_data("./benchmark") # 将benchmark数据保存至benchmark文件夹
+```
+
+执行脚本：
+
+```bash
+python test_infer.py
+```
+
+运行示例程序所得到的 benchmark 结果如下：
+
+```
+                                                             Operation Info
++-----------------------------------------------------+-------------------------------------------------------------------------------+
+|                      Operation                      |                              Source Code Location                             |
++-----------------------------------------------------+-------------------------------------------------------------------------------+
+|                      ReadImage                      |       /PaddleX/paddlex/inference/common/reader/image_reader.py:47       |
+|                   DocTrPostProcess                  |    /PaddleX/paddlex/inference/models/image_unwarping/processors.py:51   |
+|                   DetResizeForTest                  |    /PaddleX/paddlex/inference/models/text_detection/processors.py:58    |
+|     _DocPreprocessorPipeline.get_model_settings     |  /PaddleX/paddlex/inference/pipelines/doc_preprocessor/pipeline.py:110  |
+|                         Crop                        | /PaddleX/paddlex/inference/models/image_classification/processors.py:45 |
+|                PaddleInferChainLegacy               |       /PaddleX/paddlex/inference/models/common/static_infer.py:248      |
+|              _OCRPipeline.rotate_image              |         /PaddleX/paddlex/inference/pipelines/ocr/pipeline.py:140        |
+|           _DocPreprocessorPipeline.predict          |  /PaddleX/paddlex/inference/pipelines/doc_preprocessor/pipeline.py:133  |
+|                 ClasPredictor.apply                 |  /PaddleX/paddlex/inference/models/base/predictor/base_predictor.py:213 |
+|                 WarpPredictor.apply                 |  /PaddleX/paddlex/inference/models/base/predictor/base_predictor.py:213 |
+|                TextDetPredictor.apply               |  /PaddleX/paddlex/inference/models/base/predictor/base_predictor.py:213 |
+|                    ResizeByShort                    |    /PaddleX/paddlex/inference/models/common/vision/processors.py:203    |
+|                      Normalize                      |    /PaddleX/paddlex/inference/models/common/vision/processors.py:268    |
+|                    DBPostProcess                    |    /PaddleX/paddlex/inference/models/text_detection/processors.py:487   |
+|                TextRecPredictor.apply               |  /PaddleX/paddlex/inference/models/base/predictor/base_predictor.py:213 |
+|                    NormalizeImage                   |    /PaddleX/paddlex/inference/models/text_detection/processors.py:252   |
+| _DocPreprocessorPipeline.check_model_settings_valid |   /PaddleX/paddlex/inference/pipelines/doc_preprocessor/pipeline.py:82  |
+|           _OCRPipeline.get_model_settings           |         /PaddleX/paddlex/inference/pipelines/ocr/pipeline.py:204        |
+|           _OCRPipeline.get_text_det_params          |         /PaddleX/paddlex/inference/pipelines/ocr/pipeline.py:236        |
+|                        Resize                       |    /PaddleX/paddlex/inference/models/common/vision/processors.py:117    |
+|                    CTCLabelDecode                   |   /PaddleX/paddlex/inference/models/text_recognition/processors.py:189  |
+|                         Topk                        | /PaddleX/paddlex/inference/models/image_classification/processors.py:83 |
+|                  OCRReisizeNormImg                  |   /PaddleX/paddlex/inference/models/text_recognition/processors.py:65   |
+|                       ToBatch                       |   /PaddleX/paddlex/inference/models/text_recognition/processors.py:235  |
+|                      ToCHWImage                     |    /PaddleX/paddlex/inference/models/common/vision/processors.py:277    |
+|                 _OCRPipeline.predict                |         /PaddleX/paddlex/inference/pipelines/ocr/pipeline.py:282        |
+|                       ToBatch                       |    /PaddleX/paddlex/inference/models/common/vision/processors.py:284    |
+|       _OCRPipeline.check_model_settings_valid       |         /PaddleX/paddlex/inference/pipelines/ocr/pipeline.py:176        |
++-----------------------------------------------------+-------------------------------------------------------------------------------+
+                                           Detail Data
++------+----------------------------------------------------------------+-----------------------+
+| Step | Operation                                                      | Time                  |
++------+----------------------------------------------------------------+-----------------------+
+|  1   | _OCRPipeline.predict                                           | 375.11244628956774    |
+|  2   |     -> _OCRPipeline.get_model_settings                         | 0.00428391998866573   |
+|  3   |     -> _OCRPipeline.check_model_settings_valid                 | 0.0024828016466926783 |
+|  4   |     -> _OCRPipeline.get_text_det_params                        | 0.005152080120751634  |
+|  5   |     -> ReadImage                                               | 3.2029549301660154    |
+|  6   |     -> _DocPreprocessorPipeline.predict                        | 27.310913350374904    |
+|  7   |         -> _DocPreprocessorPipeline.get_model_settings         | 0.004107539862161502  |
+|  8   |         -> _DocPreprocessorPipeline.check_model_settings_valid | 0.0016830896493047476 |
+|  9   |         -> ReadImage                                           | 0.0029576495580840856 |
+|  10  |         -> ClasPredictor.apply                                 | 4.701614730001893     |
+|  11  |             -> ReadImage                                       | 0.13587839042884298   |
+|  12  |             -> ResizeByShort                                   | 0.3281894406245556    |
+|  13  |             -> Crop                                            | 0.01503000981756486   |
+|  14  |             -> Normalize                                       | 0.3884544402535539    |
+|  15  |             -> ToCHWImage                                      | 0.006330519245238975  |
+|  16  |             -> ToBatch                                         | 0.14169737987685949   |
+|  17  |             -> PaddleInferChainLegacy                          | 3.283889550511958     |
+|  18  |             -> Topk                                            | 0.10010718091507442   |
+|  19  |         -> WarpPredictor.apply                                 | 21.893062600429403    |
+|  20  |             -> ReadImage                                       | 0.004573430051095784  |
+|  21  |             -> Normalize                                       | 4.245691860560328     |
+|  22  |             -> ToCHWImage                                      | 0.005895959911867976  |
+|  23  |             -> ToBatch                                         | 1.7250755491841119    |
+|  24  |             -> PaddleInferChainLegacy                          | 10.887994960212382    |
+|  25  |             -> DocTrPostProcess                                | 1.4253830898087472    |
+|  26  |     -> TextDetPredictor.apply                                  | 49.976056129235076    |
+|  27  |         -> ReadImage                                           | 0.004843260976485908  |
+|  28  |         -> DetResizeForTest                                    | 3.3269549095712136    |
+|  29  |         -> NormalizeImage                                      | 2.9576204597833566    |
+|  30  |         -> ToCHWImage                                          | 0.005182310123927891  |
+|  31  |         -> ToBatch                                             | 1.046062790119322     |
+|  32  |         -> PaddleInferChainLegacy                              | 34.70224040953326     |
+|  33  |         -> DBPostProcess                                       | 5.826775671303039     |
+|  34  |     -> ClasPredictor.apply                                     | 23.43678753997665     |
+|  35  |         -> ReadImage                                           | 0.0633359991479665    |
+|  36  |         -> Resize                                              | 0.24419097986537963   |
+|  37  |         -> Normalize                                           | 0.480741420033155     |
+|  38  |         -> ToCHWImage                                          | 0.0066608507768251    |
+|  39  |         -> ToBatch                                             | 0.18536171046434902   |
+|  40  |         -> PaddleInferChainLegacy                              | 3.3766339404974133    |
+|  41  |         -> Topk                                                | 0.15909907990135252   |
+|  42  |         -> ReadImage                                           | 0.0395357194065582    |
+|  43  |         -> Resize                                              | 0.2085290702234488    |
+|  44  |         -> Normalize                                           | 0.4068155895220116    |
+|  45  |         -> ToCHWImage                                          | 0.005677459557773545  |
+|  46  |         -> ToBatch                                             | 0.11155156986205839   |
+|  47  |         -> PaddleInferChainLegacy                              | 2.7268862597702537    |
+|  48  |         -> Topk                                                | 0.13428127014776692   |
+|  49  |         -> ReadImage                                           | 0.032502070971531793  |
+|  50  |         -> Resize                                              | 0.20152631899691187   |
+|  51  |         -> Normalize                                           | 0.347195100330282     |
+|  52  |         -> ToCHWImage                                          | 0.005517759709618986  |
+|  53  |         -> ToBatch                                             | 0.10656953061698005   |
+|  54  |         -> PaddleInferChainLegacy                              | 2.612808299745666     |
+|  55  |         -> Topk                                                | 0.13188434022595175   |
+|  56  |         -> ReadImage                                           | 0.03589507090509869   |
+|  57  |         -> Resize                                              | 0.2076980892161373    |
+|  58  |         -> Normalize                                           | 0.3592138692329172    |
+|  59  |         -> ToCHWImage                                          | 0.005206359783187509  |
+|  60  |         -> ToBatch                                             | 0.1359267797670327    |
+|  61  |         -> PaddleInferChainLegacy                              | 2.619662079960108     |
+|  62  |         -> Topk                                                | 0.130717080028262     |
+|  63  |         -> ReadImage                                           | 0.038393009890569374  |
+|  64  |         -> Resize                                              | 0.19743553988519125   |
+|  65  |         -> Normalize                                           | 0.33197281998582184   |
+|  66  |         -> ToCHWImage                                          | 0.00512515107402578   |
+|  67  |         -> ToBatch                                             | 0.10293568033375777   |
+|  68  |         -> PaddleInferChainLegacy                              | 2.5824282996472903    |
+|  69  |         -> Topk                                                | 0.129485729848966     |
+|  70  |         -> ReadImage                                           | 0.04028105002362281   |
+|  71  |         -> Resize                                              | 0.10972122952807695   |
+|  72  |         -> Normalize                                           | 0.1787920702190604    |
+|  73  |         -> ToCHWImage                                          | 0.00408922991482541   |
+|  74  |         -> ToBatch                                             | 0.05458273953991011   |
+|  75  |         -> PaddleInferChainLegacy                              | 2.262636839877814     |
+|  76  |         -> Topk                                                | 0.1055472502775956    |
+|  77  |     -> _OCRPipeline.rotate_image                               | 0.05102259965497069   |
+|  78  |     -> TextRecPredictor.apply                                  | 169.44437422047486    |
+|  79  |         -> ReadImage                                           | 0.004737989947898313  |
+|  80  |         -> OCRReisizeNormImg                                   | 0.46037410967983305   |
+|  81  |         -> ToBatch                                             | 0.6405122207070235    |
+|  82  |         -> PaddleInferChainLegacy                              | 15.439773340767715    |
+|  83  |         -> CTCLabelDecode                                      | 10.742378439754248    |
+|  84  |         -> ReadImage                                           | 0.006349970353767276  |
+|  85  |         -> OCRReisizeNormImg                                   | 0.6252558408596087    |
+|  86  |         -> ToBatch                                             | 0.7338531101413537    |
+|  87  |         -> PaddleInferChainLegacy                              | 15.204189889482222    |
+|  88  |         -> CTCLabelDecode                                      | 6.7516070799320005    |
+|  89  |         -> ReadImage                                           | 0.006978959863772616  |
+|  90  |         -> OCRReisizeNormImg                                   | 0.7167729703360237    |
+|  91  |         -> ToBatch                                             | 0.6568272292497568    |
+|  92  |         -> PaddleInferChainLegacy                              | 14.973864750063512    |
+|  93  |         -> CTCLabelDecode                                      | 6.695752280211309     |
+|  94  |         -> ReadImage                                           | 0.0070425499870907515 |
+|  95  |         -> OCRReisizeNormImg                                   | 0.7757280093210284    |
+|  96  |         -> ToBatch                                             | 0.6442721793428063    |
+|  97  |         -> PaddleInferChainLegacy                              | 15.027350780292181    |
+|  98  |         -> CTCLabelDecode                                      | 6.661591530573787     |
+|  99  |         -> ReadImage                                           | 0.007066540565574542  |
+| 100  |         -> OCRReisizeNormImg                                   | 0.9195591000025161    |
+| 101  |         -> ToBatch                                             | 0.7951801503077149    |
+| 102  |         -> PaddleInferChainLegacy                              | 15.379044259898365    |
+| 103  |         -> CTCLabelDecode                                      | 9.372330370388227     |
+| 104  |         -> ReadImage                                           | 0.006225309771252796  |
+| 105  |         -> OCRReisizeNormImg                                   | 1.1437026296334807    |
+| 106  |         -> ToBatch                                             | 1.091715270158602     |
+| 107  |         -> PaddleInferChainLegacy                              | 23.505835609685164    |
+| 108  |         -> CTCLabelDecode                                      | 17.118994210031815    |
++------+----------------------------------------------------------------+-----------------------+
+                                      Summary Data
++-------+-----------------------------------------------------+-----------------------+
+| Level | Operation                                           | Time                  |
++-------+-----------------------------------------------------+-----------------------+
+|   1   | _OCRPipeline.predict                                | 375.11244628956774    |
+|       |                                                     |                       |
+|   2   | Layer                                               | 375.11244628956774    |
+|       | Core                                                | 273.4340275716386     |
+|       | Other                                               | 101.67841871792916    |
+|       | _OCRPipeline.get_model_settings                     | 0.00428391998866573   |
+|       | _OCRPipeline.check_model_settings_valid             | 0.0024828016466926783 |
+|       | _OCRPipeline.get_text_det_params                    | 0.005152080120751634  |
+|       | ReadImage                                           | 3.2029549301660154    |
+|       | _DocPreprocessorPipeline.predict                    | 27.310913350374904    |
+|       | TextDetPredictor.apply                              | 49.976056129235076    |
+|       | ClasPredictor.apply                                 | 23.43678753997665     |
+|       | _OCRPipeline.rotate_image                           | 0.05102259965497069   |
+|       | TextRecPredictor.apply                              | 169.44437422047486    |
+|       |                                                     |                       |
+|   3   | Layer                                               | 270.1681312400615     |
+|       | Core                                                | 261.8130224109336     |
+|       | Other                                               | 8.355108829127857     |
+|       | _DocPreprocessorPipeline.get_model_settings         | 0.004107539862161502  |
+|       | _DocPreprocessorPipeline.check_model_settings_valid | 0.0016830896493047476 |
+|       | ReadImage                                           | 0.29614515136927366   |
+|       | ClasPredictor.apply                                 | 4.701614730001893     |
+|       | WarpPredictor.apply                                 | 21.893062600429403    |
+|       | DetResizeForTest                                    | 3.3269549095712136    |
+|       | NormalizeImage                                      | 2.9576204597833566    |
+|       | ToCHWImage                                          | 0.03745912094018422   |
+|       | ToBatch                                             | 6.305350960610667     |
+|       | PaddleInferChainLegacy                              | 150.41335475922097    |
+|       | DBPostProcess                                       | 5.826775671303039     |
+|       | Resize                                              | 1.1691012277151458    |
+|       | Normalize                                           | 2.104730869323248     |
+|       | Topk                                                | 0.7910147504298948    |
+|       | OCRReisizeNormImg                                   | 4.641392659832491     |
+|       | CTCLabelDecode                                      | 57.342653910891386    |
+|       |                                                     |                       |
+|   4   | Layer                                               | 26.594677330431296    |
+|       | Core                                                | 22.69419176140218     |
+|       | Other                                               | 3.900485569029115     |
+|       | ReadImage                                           | 0.14045182047993876   |
+|       | ResizeByShort                                       | 0.3281894406245556    |
+|       | Crop                                                | 0.01503000981756486   |
+|       | Normalize                                           | 4.634146300813882     |
+|       | ToCHWImage                                          | 0.012226479157106951  |
+|       | ToBatch                                             | 1.8667729290609714    |
+|       | PaddleInferChainLegacy                              | 14.17188451072434     |
+|       | Topk                                                | 0.10010718091507442   |
+|       | DocTrPostProcess                                    | 1.4253830898087472    |
++-------+-----------------------------------------------------+-----------------------+
+```
+
+上述结果会保存到到本地：`./benchmark/detail.csv` 和 `./benchmark/summary.csv`：
+
+`detail.csv` 内容如下：
+
+```
+Step,Operation,Time
+1,_OCRPipeline.predict,375.11244628956774
+2,    -> _OCRPipeline.get_model_settings,0.00428391998866573
+3,    -> _OCRPipeline.check_model_settings_valid,0.0024828016466926783
+4,    -> _OCRPipeline.get_text_det_params,0.005152080120751634
+5,    -> ReadImage,3.2029549301660154
+6,    -> _DocPreprocessorPipeline.predict,27.310913350374904
+7,        -> _DocPreprocessorPipeline.get_model_settings,0.004107539862161502
+8,        -> _DocPreprocessorPipeline.check_model_settings_valid,0.0016830896493047476
+9,        -> ReadImage,0.0029576495580840856
+10,        -> ClasPredictor.apply,4.701614730001893
+11,            -> ReadImage,0.13587839042884298
+12,            -> ResizeByShort,0.3281894406245556
+13,            -> Crop,0.01503000981756486
+14,            -> Normalize,0.3884544402535539
+15,            -> ToCHWImage,0.006330519245238975
+16,            -> ToBatch,0.14169737987685949
+17,            -> PaddleInferChainLegacy,3.283889550511958
+18,            -> Topk,0.10010718091507442
+19,        -> WarpPredictor.apply,21.893062600429403
+20,            -> ReadImage,0.004573430051095784
+21,            -> Normalize,4.245691860560328
+22,            -> ToCHWImage,0.005895959911867976
+23,            -> ToBatch,1.7250755491841119
+24,            -> PaddleInferChainLegacy,10.887994960212382
+25,            -> DocTrPostProcess,1.4253830898087472
+26,    -> TextDetPredictor.apply,49.976056129235076
+27,        -> ReadImage,0.004843260976485908
+28,        -> DetResizeForTest,3.3269549095712136
+29,        -> NormalizeImage,2.9576204597833566
+30,        -> ToCHWImage,0.005182310123927891
+31,        -> ToBatch,1.046062790119322
+32,        -> PaddleInferChainLegacy,34.70224040953326
+33,        -> DBPostProcess,5.826775671303039
+34,    -> ClasPredictor.apply,23.43678753997665
+35,        -> ReadImage,0.0633359991479665
+36,        -> Resize,0.24419097986537963
+37,        -> Normalize,0.480741420033155
+38,        -> ToCHWImage,0.0066608507768251
+39,        -> ToBatch,0.18536171046434902
+40,        -> PaddleInferChainLegacy,3.3766339404974133
+41,        -> Topk,0.15909907990135252
+42,        -> ReadImage,0.0395357194065582
+43,        -> Resize,0.2085290702234488
+44,        -> Normalize,0.4068155895220116
+45,        -> ToCHWImage,0.005677459557773545
+46,        -> ToBatch,0.11155156986205839
+47,        -> PaddleInferChainLegacy,2.7268862597702537
+48,        -> Topk,0.13428127014776692
+49,        -> ReadImage,0.032502070971531793
+50,        -> Resize,0.20152631899691187
+51,        -> Normalize,0.347195100330282
+52,        -> ToCHWImage,0.005517759709618986
+53,        -> ToBatch,0.10656953061698005
+54,        -> PaddleInferChainLegacy,2.612808299745666
+55,        -> Topk,0.13188434022595175
+56,        -> ReadImage,0.03589507090509869
+57,        -> Resize,0.2076980892161373
+58,        -> Normalize,0.3592138692329172
+59,        -> ToCHWImage,0.005206359783187509
+60,        -> ToBatch,0.1359267797670327
+61,        -> PaddleInferChainLegacy,2.619662079960108
+62,        -> Topk,0.130717080028262
+63,        -> ReadImage,0.038393009890569374
+64,        -> Resize,0.19743553988519125
+65,        -> Normalize,0.33197281998582184
+66,        -> ToCHWImage,0.00512515107402578
+67,        -> ToBatch,0.10293568033375777
+68,        -> PaddleInferChainLegacy,2.5824282996472903
+69,        -> Topk,0.129485729848966
+70,        -> ReadImage,0.04028105002362281
+71,        -> Resize,0.10972122952807695
+72,        -> Normalize,0.1787920702190604
+73,        -> ToCHWImage,0.00408922991482541
+74,        -> ToBatch,0.05458273953991011
+75,        -> PaddleInferChainLegacy,2.262636839877814
+76,        -> Topk,0.1055472502775956
+77,    -> _OCRPipeline.rotate_image,0.05102259965497069
+78,    -> TextRecPredictor.apply,169.44437422047486
+79,        -> ReadImage,0.004737989947898313
+80,        -> OCRReisizeNormImg,0.46037410967983305
+81,        -> ToBatch,0.6405122207070235
+82,        -> PaddleInferChainLegacy,15.439773340767715
+83,        -> CTCLabelDecode,10.742378439754248
+84,        -> ReadImage,0.006349970353767276
+85,        -> OCRReisizeNormImg,0.6252558408596087
+86,        -> ToBatch,0.7338531101413537
+87,        -> PaddleInferChainLegacy,15.204189889482222
+88,        -> CTCLabelDecode,6.7516070799320005
+89,        -> ReadImage,0.006978959863772616
+90,        -> OCRReisizeNormImg,0.7167729703360237
+91,        -> ToBatch,0.6568272292497568
+92,        -> PaddleInferChainLegacy,14.973864750063512
+93,        -> CTCLabelDecode,6.695752280211309
+94,        -> ReadImage,0.0070425499870907515
+95,        -> OCRReisizeNormImg,0.7757280093210284
+96,        -> ToBatch,0.6442721793428063
+97,        -> PaddleInferChainLegacy,15.027350780292181
+98,        -> CTCLabelDecode,6.661591530573787
+99,        -> ReadImage,0.007066540565574542
+100,        -> OCRReisizeNormImg,0.9195591000025161
+101,        -> ToBatch,0.7951801503077149
+102,        -> PaddleInferChainLegacy,15.379044259898365
+103,        -> CTCLabelDecode,9.372330370388227
+104,        -> ReadImage,0.006225309771252796
+105,        -> OCRReisizeNormImg,1.1437026296334807
+106,        -> ToBatch,1.091715270158602
+107,        -> PaddleInferChainLegacy,23.505835609685164
+108,        -> CTCLabelDecode,17.118994210031815
+```
+
+`summary.csv` 内容如下：
+
+```
+Level,Operation,Time
+1,_OCRPipeline.predict,375.11244628956774
+,,
+2,Layer,375.11244628956774
+,Core,273.4340275716386
+,Other,101.67841871792916
+,_OCRPipeline.get_model_settings,0.00428391998866573
+,_OCRPipeline.check_model_settings_valid,0.0024828016466926783
+,_OCRPipeline.get_text_det_params,0.005152080120751634
+,ReadImage,3.2029549301660154
+,_DocPreprocessorPipeline.predict,27.310913350374904
+,TextDetPredictor.apply,49.976056129235076
+,ClasPredictor.apply,23.43678753997665
+,_OCRPipeline.rotate_image,0.05102259965497069
+,TextRecPredictor.apply,169.44437422047486
+,,
+3,Layer,270.1681312400615
+,Core,261.8130224109336
+,Other,8.355108829127857
+,_DocPreprocessorPipeline.get_model_settings,0.004107539862161502
+,_DocPreprocessorPipeline.check_model_settings_valid,0.0016830896493047476
+,ReadImage,0.29614515136927366
+,ClasPredictor.apply,4.701614730001893
+,WarpPredictor.apply,21.893062600429403
+,DetResizeForTest,3.3269549095712136
+,NormalizeImage,2.9576204597833566
+,ToCHWImage,0.03745912094018422
+,ToBatch,6.305350960610667
+,PaddleInferChainLegacy,150.41335475922097
+,DBPostProcess,5.826775671303039
+,Resize,1.1691012277151458
+,Normalize,2.104730869323248
+,Topk,0.7910147504298948
+,OCRReisizeNormImg,4.641392659832491
+,CTCLabelDecode,57.342653910891386
+,,
+4,Layer,26.594677330431296
+,Core,22.69419176140218
+,Other,3.900485569029115
+,ReadImage,0.14045182047993876
+,ResizeByShort,0.3281894406245556
+,Crop,0.01503000981756486
+,Normalize,4.634146300813882
+,ToCHWImage,0.012226479157106951
+,ToBatch,1.8667729290609714
+,PaddleInferChainLegacy,14.17188451072434
+,Topk,0.10010718091507442
+,DocTrPostProcess,1.4253830898087472
+```
diff --git a/docs/version3.x/inference_deployment/others/obtaining_onnx_models.en.md b/docs/version3.x/inference_deployment/others/obtaining_onnx_models.en.md
new file mode 100644
index 0000000..9b12fc2
--- /dev/null
+++ b/docs/version3.x/inference_deployment/others/obtaining_onnx_models.en.md
@@ -0,0 +1,55 @@
+---
+comments: true
+---
+
+# Obtaining ONNX Models
+
+PaddleOCR provides a rich collection of pre-trained models, all stored in PaddlePaddle's static graph format. To use these models in ONNX format during deployment, you can convert them using the Paddle2ONNX plugin provided by PaddleX. For more information about PaddleX and its relationship with PaddleOCR, refer to [Differences and Connections Between PaddleOCR and PaddleX](../../paddleocr_and_paddlex.en.md#1-differences-and-connections-between-paddleocr-and-paddlex).
+
+First, install the Paddle2ONNX plugin for PaddleX using the following command via the PaddleX CLI:
+
+```bash
+# Windows users need to use the following command to install the dev version of paddlepaddle
+# python -m pip install --pre paddlepaddle -i https://www.paddlepaddle.org.cn/packages/nightly/cpu/
+
+paddlex --install paddle2onnx
+```
+
+Then, execute the following command to complete the model conversion:
+
+```bash
+paddlex \
+    --paddle2onnx \  # Use the paddle2onnx feature
+    --paddle_model_dir /your/paddle_model/dir \  # Specify the directory containing the Paddle model
+    --onnx_model_dir /your/onnx_model/output/dir \  # Specify the output directory for the converted ONNX model
+    --opset_version 7  # Specify the ONNX opset version to use
+```
+
+The parameters are described as follows:
+
+<table>
+    <thead>
+        <tr>
+            <th>Parameter</th>
+            <th>Type</th>
+            <th>Description</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>paddle_model_dir</td>
+            <td>str</td>
+            <td>The directory containing the Paddle model.</td>
+        </tr>
+        <tr>
+            <td>onnx_model_dir</td>
+            <td>str</td>
+            <td>The output directory for the ONNX model. It can be the same as the Paddle model directory. Defaults to <code>onnx</code>.</td>
+        </tr>
+        <tr>
+            <td>opset_version</td>
+            <td>int</td>
+            <td>The ONNX opset version to use. If conversion fails with a lower opset version, a higher version will be automatically selected for conversion. Defaults to <code>7</code>.</td>
+        </tr>
+    </tbody>
+</table>
diff --git a/docs/version3.x/inference_deployment/others/obtaining_onnx_models.md b/docs/version3.x/inference_deployment/others/obtaining_onnx_models.md
new file mode 100644
index 0000000..2521f19
--- /dev/null
+++ b/docs/version3.x/inference_deployment/others/obtaining_onnx_models.md
@@ -0,0 +1,55 @@
+---
+comments: true
+---
+
+# 获取 ONNX 模型
+
+PaddleOCR 提供了丰富的预训练模型，这些模型均采用飞桨的静态图格式进行存储。若需在部署阶段使用 ONNX 格式的模型，可借助 PaddleX 提供的 Paddle2ONNX 插件进行转换。关于 PaddleX 及其与 PaddleOCR 之间的关系，请参考 [PaddleOCR 与 PaddleX 的区别与联系](../../paddleocr_and_paddlex.md#1-paddleocr-paddlex)。
+
+首先，执行如下命令，通过 PaddleX CLI 安装 PaddleX 的 Paddle2ONNX 插件：
+
+```bash
+# Windows 用户需使用以下命令安装 paddlepaddle dev版本
+# python -m pip install --pre paddlepaddle -i https://www.paddlepaddle.org.cn/packages/nightly/cpu/
+
+paddlex --install paddle2onnx
+```
+
+然后，执行如下命令完成模型转换：
+
+```bash
+paddlex \
+    --paddle2onnx \  # 使用paddle2onnx功能
+    --paddle_model_dir /your/paddle_model/dir \  # 指定 Paddle 模型所在的目录
+    --onnx_model_dir /your/onnx_model/output/dir \  # 指定转换后 ONNX 模型的输出目录
+    --opset_version 7  # 指定要使用的 ONNX opset 版本
+```
+
+参数说明如下：
+
+<table>
+    <thead>
+        <tr>
+            <th>参数</th>
+            <th>类型</th>
+            <th>描述</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>paddle_model_dir</td>
+            <td>str</td>
+            <td>包含 Paddle 模型的目录。</td>
+        </tr>
+        <tr>
+            <td>onnx_model_dir</td>
+            <td>str</td>
+            <td>ONNX 模型的输出目录，可以与 Paddle 模型目录相同。默认为 <code>onnx</code>。</td>
+        </tr>
+        <tr>
+            <td>opset_version</td>
+            <td>int</td>
+            <td>使用的 ONNX opset 版本。当使用低版本 opset 无法完成转换时，将自动选择更高版本的 opset 进行转换。默认为 <code>7</code>。</td>
+        </tr>
+    </tbody>
+</table>
diff --git a/docs/version3.x/inference_deployment/others/packaging.en.md b/docs/version3.x/inference_deployment/others/packaging.en.md
new file mode 100644
index 0000000..a482988
--- /dev/null
+++ b/docs/version3.x/inference_deployment/others/packaging.en.md
@@ -0,0 +1,106 @@
+---
+comments: true
+---
+
+# Package PaddleOCR Projects
+
+This guide applies to packaging PaddleOCR projects using PyInstaller.
+
+> Since Nuikta's packaging principle is incompatible with PaddleOCR, packaging with Nuikta is currently not supported.
+
+## Preparing the Environment
+
+- **Complete the PaddleOCR installation according to the [PaddleOCR Installation Documentation](../../installation.en.md).**
+- **Install PyInstaller.**
+
+Install PyInstaller:
+
+```bash
+pip install pyinstaller
+```
+
+> Ensure that all dependencies required by the Python script to be packaged are installed in the current environment to prevent anomalies in the packaged executable due to missing dependencies.
+
+## Running the Packaging Script
+
+Copy the Python script below and save it as a `py` file, for example, `package.py`.
+
+```python
+import paddlex
+import importlib.metadata
+import argparse
+import subprocess
+import sys
+
+parser = argparse.ArgumentParser()
+parser.add_argument('--file', required=True, help='Your file name, e.g. main.py.')
+parser.add_argument('--nvidia', action='store_true', help='Include NVIDIA CUDA and cuDNN dependencies.')
+
+args = parser.parse_args()
+
+main_file = args.file
+
+user_deps = [dist.metadata["Name"] for dist in importlib.metadata.distributions()]
+deps_all = list(paddlex.utils.deps.BASE_DEP_SPECS.keys())
+deps_need = [dep for dep in user_deps if dep in deps_all]
+
+cmd = [
+    "pyinstaller", main_file,
+    "--collect-data", "paddlex",
+    "--collect-binaries", "paddle"
+]
+
+if args.nvidia:
+    cmd += ["--collect-binaries", "nvidia"]
+
+for dep in deps_need:
+    cmd += ["--copy-metadata", dep]
+
+print("PyInstaller command:", " ".join(cmd))
+
+try:
+    result = subprocess.run(cmd, check=True)
+except subprocess.CalledProcessError as e:
+    print("Installation failed:", e)
+    sys.exit(1)
+```
+
+
+**Supported Script Parameters:**
+
+| Parameter         | Required | Description                                                                                                               |
+|--------------|------------------------------------------------------------------------------------------------------------------------------|---------|
+| --file   | Yes     | The name of the file to be packaged (e.g., `main.py`).
+| --nvidia     | No     | Packages NVIDIA CUDA and cuDNN related dependencies into the same directory as the executable. If the system environment paths already include CUDA and cuDNN dependencies or if CUDA and cuDNN dependencies are not required, this option can be omitted.
+
+**Example Usage of the Packaging Script:**
+
+```bash
+python package.py --file main.py
+# Packages NVIDIA CUDA and cuDNN related dependencies into the same directory as the executable.
+python package.py --file main.py --nvidia
+```
+
+**Execution Result**
+
+- The packaging script will execute a command similar to:
+
+    `pyinstaller main.py --collect-data paddlex --collect-binaries paddle [--copy-metadata xxx …]`, where `--copy-metadata xxx` dynamically adds package metadata based on the dependencies required by PaddleOCR installed in the current environment.
+
+- The executable file and related dependency libraries will be generated in the `dist` folder.
+
+## Appendix
+
+**The above packaging process was tested in the following environment:**
+
+- Operating System: **Win 11**
+- Python: **3.10.18**
+- PaddlePaddle: **3.0.0**
+- PaddleX: **3.1.3**
+- PaddleOCR：**3.1.0**
+- PyInstaller: **6.14.2**
+
+**Common Issues**
+
+- When running the executable, if you encounter an error message like `RuntimeError: xxx requires additional dependencies`, it indicates that the current packaging environment lacks the necessary dependencies. Please ensure that the environment is set up correctly as described in the Preparations section.
+- When running the executable, if an error message indicates that CUDA or cuDNN related dynamic link libraries cannot be found, please check whether the system environment variables correctly include the paths to NVIDIA CUDA and cuDNN dependencies, or consider adding `--nvidia` when running the packaging script to include CUDA and cuDNN dependencies in the same directory as the executable.
diff --git a/docs/version3.x/inference_deployment/others/packaging.md b/docs/version3.x/inference_deployment/others/packaging.md
new file mode 100644
index 0000000..575233a
--- /dev/null
+++ b/docs/version3.x/inference_deployment/others/packaging.md
@@ -0,0 +1,106 @@
+---
+comments: true
+---
+
+# 打包PaddleOCR项目
+
+本说明适用于通过PyInstaller打包PaddleOCR项目。
+
+> 由于Nuikta的打包原理与PaddleOCR不适配，当前暂不支持通过Nuikta进行打包。
+
+## 准备环境
+
+- **根据[PaddleOCR安装文档](../../installation.md)完成PaddleOCR安装**
+- **安装PyInstaller**
+
+安装PyInstaller：
+
+```bash
+pip install pyinstaller
+```
+
+> 请确认当前准备环境中安装有待打包的Python脚本所需的全部依赖，以避免缺少依赖导致打包后的可执行程序出现异常。
+
+## 执行打包脚本
+
+将下方Python脚本拷贝后存成`py`文件，文件名可以为`package.py`。
+
+```python
+import paddlex
+import importlib.metadata
+import argparse
+import subprocess
+import sys
+
+parser = argparse.ArgumentParser()
+parser.add_argument('--file', required=True, help='Your file name, e.g. main.py.')
+parser.add_argument('--nvidia', action='store_true', help='Include NVIDIA CUDA and cuDNN dependencies.')
+
+args = parser.parse_args()
+
+main_file = args.file
+
+user_deps = [dist.metadata["Name"] for dist in importlib.metadata.distributions()]
+deps_all = list(paddlex.utils.deps.BASE_DEP_SPECS.keys())
+deps_need = [dep for dep in user_deps if dep in deps_all]
+
+cmd = [
+    "pyinstaller", main_file,
+    "--collect-data", "paddlex",
+    "--collect-binaries", "paddle"
+]
+
+if args.nvidia:
+    cmd += ["--collect-binaries", "nvidia"]
+
+for dep in deps_need:
+    cmd += ["--copy-metadata", dep]
+
+print("PyInstaller command:", " ".join(cmd))
+
+try:
+    result = subprocess.run(cmd, check=True)
+except subprocess.CalledProcessError as e:
+    print("Installation failed:", e)
+    sys.exit(1)
+```
+
+
+**打包脚本支持的参数如下：**
+
+| 参数         | 是否必需 | 说明                                                                                                               |
+|--------------|------------------------------------------------------------------------------------------------------------------------------|---------|
+| --file   | 是     | 你的待打包文件名（如`main.py`）。
+| --nvidia     | 否     | 将NVIDIA的CUDA、cuDNN相关依赖库一同打包到可执行文件的同级目录中。如果系统环境变量路径已包含CUDA、cuDNN相关依赖库或者不需要使用CUDA、cuDNN相关依赖库，则无需开启。
+
+**打包脚本调用示例如下：**
+
+```bash
+python package.py --file main.py
+# 将NVIDIA的CUDA、cuDNN相关依赖库打包至可执行文件的同级目录中。
+python package.py --file main.py --nvidia
+```
+
+**运行结果**
+
+- 安转脚本将执行类似如下命令：
+
+    `pyinstaller main.py --collect-data paddlex --collect-binaries paddle [--copy-metadata xxx …]`，其中`--copy-metadata xxx`会根据当前环境已安装的PaddleOCR需要的依赖动态添加包的元信息。
+
+- 可执行文件和相关依赖库将生成到`dist`文件夹中。
+
+## 附录
+
+**以上打包流程在如下环境中测试：**
+
+- 操作系统：**Win 11**
+- Python：**3.10.18**
+- PaddlePaddle：**3.0.0**
+- PaddleX：**3.1.3**
+- PaddleOCR：**3.1.0**
+- PyInstaller：**6.14.2**
+
+**常见问题**
+
+- 在运行可执行文件时，出现报错信息 `RuntimeError: xxx requires additional dependencies`，说明当前打包环境缺少相关依赖，请确认已按照准备环境部分说明正确安装环境。
+- 在运行可执行文件时，出现报错信息提示CUDA、cuDNN相关动态链接库找不到，请检查系统环境变量中是否正确添加NVIDIA的CUDA、cuDNN相关依赖库路径或者考虑在运行打包脚本时添加 `--nvidia`，将CUDA、cuDNN相关依赖库打包进可执行文件的同级目录中。
diff --git a/docs/version3.x/inference_deployment/serving/paddleocr_official_api/cli.en.md b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/cli.en.md
new file mode 100644
index 0000000..84bbb0d
--- /dev/null
+++ b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/cli.en.md
@@ -0,0 +1,105 @@
+---
+comments: true
+---
+
+# PaddleOCR official API CLI
+
+`paddleocr api` is the PaddleOCR CLI subcommand for calling the PaddleOCR official API. It submits a file URL or local file to hosted services, waits for completion, and outputs the parsing result. It does not run local inference.
+
+## Install And Authenticate
+
+First install the Python package by following [Install paddleocr](../../../installation.en.md#11-install-paddleocr). After installing the core `paddleocr` package, this feature is available out of the box — no additional dependency groups are required.
+
+First obtain an access token from the [AI Studio Access Token page](https://aistudio.baidu.com/account/accessToken).
+
+The CLI reads `PADDLEOCR_ACCESS_TOKEN` by default:
+
+```bash
+export PADDLEOCR_ACCESS_TOKEN="your-access-token"
+```
+
+You can also pass a token explicitly with `--token`.
+
+## Basic Usage
+
+```bash
+paddleocr api \
+  --model_type ocr \
+  --file_url https://example.com/invoice.pdf
+```
+
+`--model_type` is required and accepts `ocr` or `doc_parsing`. Pass exactly one of `--file_url` or `--file_path`.
+
+## Common Options
+
+- `--model_type`: task type, either `ocr` or `doc_parsing`.
+- `--model`: model name.
+- `--file_url`: file URL to process.
+- `--file_path`: local file path to upload and process.
+- `--base_url`: base URL of the PaddleOCR API service; defaults to the official service URL (can also be set via `PADDLEOCR_BASE_URL`).
+- `--request_timeout`: timeout in seconds for one HTTP request.
+- `--poll_timeout`: total timeout in seconds while waiting for the remote job to complete.
+- `--output`: JSON output file path; omitted means print to stdout.
+- `--save_resources`: directory for saving resources referenced by the result object.
+- `--overwrite_resources`: overwrite existing files when saving resources.
+- `--page_ranges`: page ranges such as `2,4-6`.
+- `--use_doc_orientation_classify`: document orientation classification, `True` or `False`.
+- `--use_doc_unwarping`: document unwarping, `True` or `False`.
+- `--use_textline_orientation`: textline orientation detection, `True` or `False`.
+- `--text_det_limit_side_len`: image side length limit for text detection.
+- `--text_det_limit_type`: side length limit type, `min` or `max`.
+- `--text_rec_score_thresh`: score threshold for text recognition results.
+- `--use_layout_detection`: layout detection, `True` or `False`.
+- `--use_seal_recognition`: seal recognition, `True` or `False`.
+- `--use_table_recognition`: table recognition, `True` or `False`.
+- `--use_formula_recognition`: formula recognition, `True` or `False`.
+- `--use_chart_recognition`: chart recognition, `True` or `False`.
+- `--visualize`: result visualization images, `True` or `False`.
+- `--prettify_markdown`: markdown prettification, `True` or `False`.
+
+## OCR Example
+
+```bash
+paddleocr api \
+  --model_type ocr \
+  --model PP-OCRv5 \
+  --file_path ./invoice.pdf \
+  --request_timeout 300 \
+  --poll_timeout 600 \
+  --output ocr-result.json
+```
+
+## Document Parsing Example
+
+```bash
+paddleocr api \
+  --model_type doc_parsing \
+  --file_url https://example.com/report.pdf \
+  --use_chart_recognition True \
+  --save_resources ./doc-assets \
+  --output doc-result.json
+```
+
+## Choose Models
+
+| Task | `--model_type` | Default model | Supported models |
+| --- | --- | --- | --- |
+| OCR | `ocr` | `PP-OCRv6` | `PP-OCRv5`, `PP-OCRv5-latin`, `PP-OCRv6` |
+| Document parsing | `doc_parsing` | `PaddleOCR-VL-1.6` | `PP-StructureV3`, `PaddleOCR-VL`, `PaddleOCR-VL-1.5`, `PaddleOCR-VL-1.6` |
+
+## Output Behavior
+
+On success, the command emits formatted JSON. OCR output includes `jobId` plus each page's `prunedResult` and `ocrImageUrl`; document parsing output includes `jobId` plus each page's `markdownText`, `markdownImages`, and `outputImages`. With `--output`, the CLI writes that file and prints the saved path; otherwise it prints JSON to stdout. With `--save_resources`, the CLI saves resources referenced by the result object into the target directory.
+
+Errors are printed to stderr and return a non-zero exit code. Common causes include a missing `PADDLEOCR_ACCESS_TOKEN`, a model that does not match `--model_type`, request timeout, poll timeout, failed remote job, or malformed response.
+
+## Official API Reference
+
+- [PP-OCRv5 API](https://ai.baidu.com/ai-doc/AISTUDIO/Dmh4onssk)
+- [PP-StructureV3 API](https://ai.baidu.com/ai-doc/AISTUDIO/7mfz6dgx9)
+- [PaddleOCR-VL API](https://ai.baidu.com/ai-doc/AISTUDIO/Vmkz2nz1p)
+- [PaddleOCR-VL-1.5 API](https://ai.baidu.com/ai-doc/AISTUDIO/fml7mozw5)
+
+## Quota and Error Codes
+
+- [API Quota Rules and Error Code Description](https://ai.baidu.com/ai-doc/AISTUDIO/pmjcld5qm)
diff --git a/docs/version3.x/inference_deployment/serving/paddleocr_official_api/cli.md b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/cli.md
new file mode 100644
index 0000000..0745e5a
--- /dev/null
+++ b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/cli.md
@@ -0,0 +1,105 @@
+---
+comments: true
+---
+
+# PaddleOCR 官方 API CLI
+
+`paddleocr api` 是 PaddleOCR CLI 中调用 PaddleOCR 官方 API 的子命令。它把文件 URL 或本地文件提交到官方托管服务，等待任务完成并输出解析结果；它不运行本地推理。
+
+## 安装与认证
+
+先按 [安装 `paddleocr`](../../../installation.md#install-paddleocr) 安装 Python 包。安装 `paddleocr` 本体后即可使用此功能，无需安装额外依赖组。
+
+请先在 [AI Studio Access Token 页面](https://aistudio.baidu.com/account/accessToken) 获取访问令牌。
+
+CLI 默认读取 `PADDLEOCR_ACCESS_TOKEN`：
+
+```bash
+export PADDLEOCR_ACCESS_TOKEN="your-access-token"
+```
+
+也可以使用 `--token` 显式传入 token。
+
+## 基本用法
+
+```bash
+paddleocr api \
+  --model_type ocr \
+  --file_url https://example.com/invoice.pdf
+```
+
+`--model_type` 必填，可选值为 `ocr` 或 `doc_parsing`。`--file_url` 与 `--file_path` 必须二选一。
+
+## 常用参数
+
+- `--model_type`：任务类型，`ocr` 或 `doc_parsing`。
+- `--model`：模型名称。
+- `--file_url`：待处理文件 URL。
+- `--file_path`：待上传并处理的本地文件路径。
+- `--base_url`：PaddleOCR API 服务的 base URL；缺省使用官方服务地址（也可通过 `PADDLEOCR_BASE_URL` 环境变量设置）。
+- `--request_timeout`：一次 HTTP 请求的超时时间，单位为秒。
+- `--poll_timeout`：等待远端任务完成的总超时时间，单位为秒。
+- `--output`：输出 JSON 文件路径；省略时打印到标准输出。
+- `--save_resources`：保存结果对象引用资源的目录。
+- `--overwrite_resources`：保存资源时覆盖已有文件。
+- `--page_ranges`：页码范围，例如 `2,4-6`。
+- `--use_doc_orientation_classify`：文档方向分类，`True` 或 `False`。
+- `--use_doc_unwarping`：文档扭曲矫正，`True` 或 `False`。
+- `--use_textline_orientation`：文本行方向检测，`True` 或 `False`。
+- `--text_det_limit_side_len`：文本检测图像边长限制。
+- `--text_det_limit_type`：边长限制类型，`min` 或 `max`。
+- `--text_rec_score_thresh`：文本识别置信度阈值。
+- `--use_layout_detection`：版面检测，`True` 或 `False`。
+- `--use_seal_recognition`：印章识别，`True` 或 `False`。
+- `--use_table_recognition`：表格识别，`True` 或 `False`。
+- `--use_formula_recognition`：公式识别，`True` 或 `False`。
+- `--use_chart_recognition`：图表识别，`True` 或 `False`。
+- `--visualize`：可视化结果图，`True` 或 `False`。
+- `--prettify_markdown`：markdown 美化，`True` 或 `False`。
+
+## OCR 示例
+
+```bash
+paddleocr api \
+  --model_type ocr \
+  --model PP-OCRv5 \
+  --file_path ./invoice.pdf \
+  --request_timeout 300 \
+  --poll_timeout 600 \
+  --output ocr-result.json
+```
+
+## 文档解析示例
+
+```bash
+paddleocr api \
+  --model_type doc_parsing \
+  --file_url https://example.com/report.pdf \
+  --use_chart_recognition True \
+  --save_resources ./doc-assets \
+  --output doc-result.json
+```
+
+## 模型选择
+
+| 任务 | `--model_type` | 默认模型 | 可选模型 |
+| --- | --- | --- | --- |
+| OCR | `ocr` | `PP-OCRv6` | `PP-OCRv5`、`PP-OCRv5-latin`、`PP-OCRv6` |
+| 文档解析 | `doc_parsing` | `PaddleOCR-VL-1.6` | `PP-StructureV3`、`PaddleOCR-VL`、`PaddleOCR-VL-1.5`、`PaddleOCR-VL-1.6` |
+
+## 输出行为
+
+命令成功时输出格式化 JSON。OCR 结果包含 `jobId` 和每页的 `prunedResult`、`ocrImageUrl`；文档解析结果包含 `jobId` 和每页的 `markdownText`、`markdownImages`、`outputImages`。如果指定 `--output`，CLI 写入该文件并打印保存位置；否则直接打印到标准输出。指定 `--save_resources` 时，CLI 会把结果对象引用的资源保存到目标目录。
+
+错误会输出到标准错误并返回非零退出码。常见原因包括缺少 `PADDLEOCR_ACCESS_TOKEN`、模型与 `--model_type` 不匹配、请求超时、轮询超时、远端任务失败或响应格式异常。
+
+## 官方 API 参考
+
+- [PP-OCRv5 API](https://ai.baidu.com/ai-doc/AISTUDIO/Kmfl2ycs0)
+- [PP-StructureV3 API](https://ai.baidu.com/ai-doc/AISTUDIO/Fmfz6oh2e)
+- [PaddleOCR-VL API](https://ai.baidu.com/ai-doc/AISTUDIO/2mh4okm66)
+- [PaddleOCR-VL-1.5 API](https://ai.baidu.com/ai-doc/AISTUDIO/Cmkz2m0ma)
+
+## 配额与错误码
+
+- [API 配额规则和错误码说明](https://ai.baidu.com/ai-doc/AISTUDIO/Xmjclapam)
diff --git a/docs/version3.x/inference_deployment/serving/paddleocr_official_api/go.en.md b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/go.en.md
new file mode 100644
index 0000000..f8d99a1
--- /dev/null
+++ b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/go.en.md
@@ -0,0 +1,136 @@
+---
+comments: true
+---
+
+# PaddleOCR official API Go SDK
+
+The Go SDK calls the PaddleOCR official API and submits OCR or document parsing jobs to hosted services. It does not run local PaddleOCR inference or load local models.
+
+## Install And Authenticate
+
+First obtain an access token from the [AI Studio Access Token page](https://aistudio.baidu.com/account/accessToken).
+
+```bash
+go get github.com/PaddlePaddle/PaddleOCR/api_sdk/go
+export PADDLEOCR_ACCESS_TOKEN="your-access-token"
+```
+
+`NewClient` reads `PADDLEOCR_ACCESS_TOKEN` by default and also accepts an explicit token through `WithToken`.
+
+## Quick Start
+
+```go
+client, err := paddleocr.NewClient()
+if err != nil {
+	return err
+}
+
+result, err := client.OCR(ctx, &paddleocr.OCRRequest{
+	Model:   paddleocr.PPOCRv5,
+	FileURL: "https://example.com/invoice.pdf",
+})
+if err != nil {
+	return err
+}
+fmt.Println(result.JobID, len(result.Pages))
+```
+
+Use `FilePath` for a local file. Pass exactly one of `FileURL` or `FilePath`.
+
+## Public API
+
+Common Go public methods include:
+
+- `OCR(...)`: submit an OCR job, wait for completion, and return an OCR result.
+- `ParseDocument(...)`: submit a document parsing job, wait for completion, and return a document parsing result.
+- `SubmitOCR(...)`: submit only an OCR job and return a job object.
+- `SubmitDocumentParsing(...)`: submit only a document parsing job and return a job object.
+- `GetStatus(ctx, jobID)`: perform one non-blocking status request.
+- `WaitOCRResult(ctx, job)`: wait for an OCR job and parse its result.
+- `WaitDocumentParsingResult(ctx, job)`: wait for a document parsing job and parse its result.
+- `SaveResource(...)`: save one resource URL.
+- `SaveOCRResultResources(...)`: save resources referenced by an OCR result object.
+- `SaveDocumentParsingResultResources(...)`: save resources referenced by a document parsing result object.
+
+## Choose Models
+
+The model constants in the table are type-safe aliases for the official API model-name strings. They are serialized to the corresponding model name when the request is submitted. You can also pass the official API model-name string directly, for example `Model: "PaddleOCR-VL-1.6"`.
+
+| Task | Interfaces | Default model | Supported models | Option type |
+| --- | --- | --- | --- | --- |
+| OCR | `OCR`, `SubmitOCR`, `WaitOCRResult` | `PPOCRv6` | `PPOCRv5`, `PPOCRv6` | `*OCROptions` |
+| Document parsing | `ParseDocument`, `SubmitDocumentParsing`, `WaitDocumentParsingResult` | `PaddleOCRVL16` | `PPStructureV3`, `PaddleOCRVL`, `PaddleOCRVL15`, `PaddleOCRVL16` | Use `*PPStructureV3Options` with `PPStructureV3`, and `*PaddleOCRVLOptions` with PaddleOCR-VL models. |
+
+## Configuration
+
+### Client Configuration
+
+```go
+client, err := paddleocr.NewClient(
+	paddleocr.WithRequestTimeout(30*time.Second),
+	paddleocr.WithPollTimeout(5*time.Minute),
+)
+```
+
+`WithRequestTimeout` limits one HTTP request, including submit, status, and resource downloads. `WithPollTimeout` limits the total wait time for `OCR`, `ParseDocument`, `WaitOCRResult`, and `WaitDocumentParsingResult`. Callers can also cancel requests through `context.Context`.
+
+Override the service base URL via the `PADDLEOCR_BASE_URL` environment variable or `WithBaseURL`:
+
+```go
+client, err := paddleocr.NewClient(
+    paddleocr.WithBaseURL("https://my-proxy.com/paddle"),
+)
+```
+
+Inject a custom `*http.Client` via `WithHTTPClient` for proxies, custom TLS, or retry policies:
+
+```go
+client, err := paddleocr.NewClient(
+    paddleocr.WithHTTPClient(myHTTPClient),
+)
+```
+
+### Request Options
+
+Go SDK Options struct fields use PascalCase, serialized to camelCase automatically. Pointer fields set to `nil` are omitted. See the struct source or Official API Reference for complete field definitions.
+
+#### OCROptions (common fields)
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `UseDocOrientationClassify` | *bool | Document orientation classification |
+| `UseDocUnwarping` | *bool | Document unwarping |
+| `Visualize` | *bool | Return visualization images |
+
+#### PPStructureV3Options (common fields)
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `UseTableRecognition` | *bool | Table recognition |
+| `UseFormulaRecognition` | *bool | Formula recognition |
+| `UseChartRecognition` | *bool | Chart recognition |
+| `PrettifyMarkdown` | *bool | Markdown prettification |
+
+#### PaddleOCRVLOptions (common fields)
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `UseLayoutDetection` | *bool | Layout detection |
+| `UseChartRecognition` | *bool | Chart recognition |
+| `Temperature` | *float64 | Sampling temperature |
+| `PrettifyMarkdown` | *bool | Markdown prettification |
+
+## Error Handling
+
+The Go SDK exposes typed errors compatible with `errors.As`, including `AuthError`, `InvalidRequestError`, `RateLimitError`, `ServiceUnavailableError`, `APIError`, `NetworkError`, `JobFailedError`, `RequestTimeoutError`, `PollTimeoutError`, `ResponseFormatError`, and `ResultParseError`.
+
+## Official API Reference
+
+- [PP-OCRv5 API](https://ai.baidu.com/ai-doc/AISTUDIO/Dmh4onssk)
+- [PP-StructureV3 API](https://ai.baidu.com/ai-doc/AISTUDIO/7mfz6dgx9)
+- [PaddleOCR-VL API](https://ai.baidu.com/ai-doc/AISTUDIO/Vmkz2nz1p)
+- [PaddleOCR-VL-1.5 API](https://ai.baidu.com/ai-doc/AISTUDIO/fml7mozw5)
+
+## Quota and Error Codes
+
+- [API Quota Rules and Error Code Description](https://ai.baidu.com/ai-doc/AISTUDIO/pmjcld5qm)
diff --git a/docs/version3.x/inference_deployment/serving/paddleocr_official_api/go.md b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/go.md
new file mode 100644
index 0000000..fba2a89
--- /dev/null
+++ b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/go.md
@@ -0,0 +1,136 @@
+---
+comments: true
+---
+
+# PaddleOCR 官方 API Go SDK
+
+Go SDK 调用 PaddleOCR 官方 API，把 OCR 或文档解析任务提交到官方托管服务。它不运行本地 PaddleOCR 推理，也不加载本地模型。
+
+## 安装与认证
+
+请先在 [AI Studio Access Token 页面](https://aistudio.baidu.com/account/accessToken) 获取访问令牌。
+
+```bash
+go get github.com/PaddlePaddle/PaddleOCR/api_sdk/go
+export PADDLEOCR_ACCESS_TOKEN="your-access-token"
+```
+
+`NewClient` 默认读取 `PADDLEOCR_ACCESS_TOKEN`，也可以通过 `WithToken` 显式传入 token。
+
+## 快速开始
+
+```go
+client, err := paddleocr.NewClient()
+if err != nil {
+	return err
+}
+
+result, err := client.OCR(ctx, &paddleocr.OCRRequest{
+	Model:   paddleocr.PPOCRv5,
+	FileURL: "https://example.com/invoice.pdf",
+})
+if err != nil {
+	return err
+}
+fmt.Println(result.JobID, len(result.Pages))
+```
+
+本地文件使用 `FilePath`。`FileURL` 与 `FilePath` 必须二选一。
+
+## 公共 API
+
+Go SDK 常用公共方法包括：
+
+- `OCR(...)`：提交 OCR 任务，等待完成并返回 OCR 结果。
+- `ParseDocument(...)`：提交文档解析任务，等待完成并返回文档解析结果。
+- `SubmitOCR(...)`：只提交 OCR 任务，返回任务对象。
+- `SubmitDocumentParsing(...)`：只提交文档解析任务，返回任务对象。
+- `GetStatus(ctx, jobID)`：执行一次非阻塞状态查询。
+- `WaitOCRResult(ctx, job)`：等待 OCR 任务完成并解析结果。
+- `WaitDocumentParsingResult(ctx, job)`：等待文档解析任务完成并解析结果。
+- `SaveResource(...)`：保存单个资源 URL。
+- `SaveOCRResultResources(...)`：保存 OCR 结果对象引用的资源。
+- `SaveDocumentParsingResultResources(...)`：保存文档解析结果对象引用的资源。
+
+## 模型选择
+
+表中的模型常量是官方 API 模型名字符串的类型安全写法，提交请求时会转换为对应的实际模型名。也可以直接传入官方 API 模型名字符串，例如 `Model: "PaddleOCR-VL-1.6"`。
+
+| 任务 | 适用接口 | 默认模型 | 可选模型 | 参数类型 |
+| --- | --- | --- | --- | --- |
+| OCR | `OCR`、`SubmitOCR`、`WaitOCRResult` | `PPOCRv6` | `PPOCRv5`、`PPOCRv6` | `*OCROptions` |
+| 文档解析 | `ParseDocument`、`SubmitDocumentParsing`、`WaitDocumentParsingResult` | `PaddleOCRVL16` | `PPStructureV3`、`PaddleOCRVL`、`PaddleOCRVL15`、`PaddleOCRVL16` | 选择 `PPStructureV3` 时传入 `*PPStructureV3Options`；选择 PaddleOCR-VL 系列模型时传入 `*PaddleOCRVLOptions`。 |
+
+## 配置与参数
+
+### 客户端配置
+
+```go
+client, err := paddleocr.NewClient(
+	paddleocr.WithRequestTimeout(30*time.Second),
+	paddleocr.WithPollTimeout(5*time.Minute),
+)
+```
+
+`WithRequestTimeout` 限制一次 HTTP 请求，包括提交、查询状态和下载资源。`WithPollTimeout` 限制 `OCR`、`ParseDocument`、`WaitOCRResult` 与 `WaitDocumentParsingResult` 的总等待时间。调用方也可以通过 `context.Context` 取消请求。
+
+通过环境变量 `PADDLEOCR_BASE_URL` 或 `WithBaseURL` 指定自定义服务地址：
+
+```go
+client, err := paddleocr.NewClient(
+    paddleocr.WithBaseURL("https://my-proxy.com/paddle"),
+)
+```
+
+通过 `WithHTTPClient` 注入自定义 `*http.Client`，适用于需要代理、自定义 TLS 或重试策略的场景：
+
+```go
+client, err := paddleocr.NewClient(
+    paddleocr.WithHTTPClient(myHTTPClient),
+)
+```
+
+### 请求参数
+
+Go SDK 的 Options 结构体字段使用 PascalCase，序列化时自动转换为 camelCase。指针类型字段传 `nil` 表示不设置。完整字段定义见结构体源码或官方 API 参考。
+
+#### OCROptions（常用字段）
+
+| 字段 | 类型 | 说明 |
+|------|------|------|
+| `UseDocOrientationClassify` | *bool | 文档方向分类 |
+| `UseDocUnwarping` | *bool | 文档扭曲矫正 |
+| `Visualize` | *bool | 是否返回可视化结果图 |
+
+#### PPStructureV3Options（常用字段）
+
+| 字段 | 类型 | 说明 |
+|------|------|------|
+| `UseTableRecognition` | *bool | 表格识别 |
+| `UseFormulaRecognition` | *bool | 公式识别 |
+| `UseChartRecognition` | *bool | 图表识别 |
+| `PrettifyMarkdown` | *bool | Markdown 美化 |
+
+#### PaddleOCRVLOptions（常用字段）
+
+| 字段 | 类型 | 说明 |
+|------|------|------|
+| `UseLayoutDetection` | *bool | 版面检测 |
+| `UseChartRecognition` | *bool | 图表识别 |
+| `Temperature` | *float64 | 采样温度 |
+| `PrettifyMarkdown` | *bool | Markdown 美化 |
+
+## 错误处理
+
+Go SDK 暴露可与 `errors.As` 配合使用的类型化错误，覆盖 `AuthError`、`InvalidRequestError`、`RateLimitError`、`ServiceUnavailableError`、`APIError`、`NetworkError`、`JobFailedError`、`RequestTimeoutError`、`PollTimeoutError`、`ResponseFormatError` 和 `ResultParseError` 等情况。
+
+## 官方 API 参考
+
+- [PP-OCRv5 API](https://ai.baidu.com/ai-doc/AISTUDIO/Kmfl2ycs0)
+- [PP-StructureV3 API](https://ai.baidu.com/ai-doc/AISTUDIO/Fmfz6oh2e)
+- [PaddleOCR-VL API](https://ai.baidu.com/ai-doc/AISTUDIO/2mh4okm66)
+- [PaddleOCR-VL-1.5 API](https://ai.baidu.com/ai-doc/AISTUDIO/Cmkz2m0ma)
+
+## 配额与错误码
+
+- [API 配额规则和错误码说明](https://ai.baidu.com/ai-doc/AISTUDIO/Xmjclapam)
diff --git a/docs/version3.x/inference_deployment/serving/paddleocr_official_api/overview.en.md b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/overview.en.md
new file mode 100644
index 0000000..de18b5b
--- /dev/null
+++ b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/overview.en.md
@@ -0,0 +1,28 @@
+---
+comments: true
+---
+
+# PaddleOCR official API Overview
+
+The PaddleOCR official API SDKs are client libraries for the PaddleOCR official API. They submit local files or file URLs to hosted PaddleOCR services, poll asynchronous jobs, and parse typed results. They do not run local PaddleOCR inference or load local models.
+
+The current clients include Python, TypeScript, Go, and the `paddleocr api` command in the PaddleOCR CLI.
+
+## Authentication
+
+First obtain an access token from the [AI Studio Access Token page](https://aistudio.baidu.com/account/accessToken).
+
+All SDKs and the CLI read `PADDLEOCR_ACCESS_TOKEN` by default:
+
+```bash
+export PADDLEOCR_ACCESS_TOKEN="your-access-token"
+```
+
+You can also pass the token explicitly when constructing a client or invoking the CLI. Missing or invalid credentials are reported through typed authentication errors.
+
+## Documentation Entry Points
+
+- [Python SDK](python.en.md): best for projects already using the `paddleocr` Python package; provides sync `PaddleOCRClient` and async `AsyncPaddleOCRClient`.
+- [TypeScript SDK](typescript.en.md): best for Node.js 18+ server-side projects.
+- [Go SDK](go.en.md): best for statically typed services that need context cancellation and binary deployment.
+- [CLI](cli.en.md): best for scripts, debugging, and quick no-code validation.
diff --git a/docs/version3.x/inference_deployment/serving/paddleocr_official_api/overview.md b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/overview.md
new file mode 100644
index 0000000..ce74324
--- /dev/null
+++ b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/overview.md
@@ -0,0 +1,16 @@
+---
+comments: true
+---
+
+# PaddleOCR 官方 API 总览
+
+PaddleOCR 官方 API SDK 是面向 PaddleOCR 官方 API 的客户端封装。它们会把本地文件或文件 URL 提交到官方托管服务，轮询异步任务并解析结果；它们不在本机运行 PaddleOCR 推理，也不会加载本地模型。
+
+目前提供 Python、TypeScript、Go SDK，以及集成在 PaddleOCR CLI 中的 `paddleocr api` 命令。
+
+## 文档入口
+
+- [Python SDK](python.md)：适合已经使用 `paddleocr` Python 包的项目，提供同步 `PaddleOCRClient` 与异步 `AsyncPaddleOCRClient`。
+- [TypeScript SDK](typescript.md)：适合 Node.js 18 及以上的服务端项目。
+- [Go SDK](go.md)：适合需要静态类型、上下文取消和二进制部署的服务端项目。
+- [CLI](cli.md)：适合脚本、调试和无代码快速验证。
diff --git a/docs/version3.x/inference_deployment/serving/paddleocr_official_api/python.en.md b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/python.en.md
new file mode 100644
index 0000000..4456119
--- /dev/null
+++ b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/python.en.md
@@ -0,0 +1,129 @@
+---
+comments: true
+---
+
+# PaddleOCR official API Python SDK
+
+The Python SDK calls the PaddleOCR official API through `PaddleOCRClient` and `AsyncPaddleOCRClient` in the `paddleocr` package. It submits OCR or document parsing jobs to hosted services. It does not run local inference or load local models.
+
+## Install And Authenticate
+
+First install the Python package by following [Install paddleocr](../../../installation.en.md#11-install-paddleocr). After installing the core `paddleocr` package, this feature is available out of the box — no additional dependency groups are required.
+
+First obtain an access token from the [AI Studio Access Token page](https://aistudio.baidu.com/account/accessToken).
+
+```bash
+export PADDLEOCR_ACCESS_TOKEN="your-access-token"
+```
+
+`PaddleOCRClient()` reads `PADDLEOCR_ACCESS_TOKEN` by default and also accepts `PaddleOCRClient(token="...")`. Missing credentials raise `AuthError`.
+
+## Quick Start
+
+```python
+from paddleocr import PaddleOCRClient, Model
+
+client = PaddleOCRClient()
+result = client.ocr(
+    file_url="https://example.com/invoice.pdf",
+    model=Model.PP_OCRV5,
+)
+print(result.job_id, len(result.pages))
+client.close()
+```
+
+Use `file_path` for a local file. Pass exactly one of `file_url` or `file_path`.
+
+## Public API
+
+Common Python public methods include:
+
+- `ocr(...)`: submit an OCR job, wait for completion, and return an OCR result.
+- `parse_document(...)`: submit a document parsing job, wait for completion, and return a document parsing result.
+- `submit_ocr(...)`: submit only an OCR job and return a job object.
+- `submit_document_parsing(...)`: submit only a document parsing job and return a job object.
+- `get_status(job_id)`: perform one non-blocking status request without waiting for completion.
+- `wait_ocr_result(job)`: wait for an OCR job and parse its result.
+- `wait_document_parsing_result(job)`: wait for a document parsing job and parse its result.
+- `save_resource(resource_url, destination)`: save one resource URL.
+- `save_ocr_result_resources(result, destination)`: save resources referenced by an OCR result object.
+- `save_document_parsing_result_resources(result, destination)`: save resources referenced by a document parsing result object.
+
+`AsyncPaddleOCRClient` exposes async versions of these job operations and resource-saving methods.
+
+## Choose Models
+
+The `Model` enum values in the table are type-safe aliases for the official API model-name strings. They are serialized to the corresponding model name when the request is submitted. You can also pass the official API model-name string directly, for example `model="PaddleOCR-VL-1.6"`.
+
+| Task | Interfaces | Default model | Supported models | Option type |
+| --- | --- | --- | --- | --- |
+| OCR | `ocr`, `submit_ocr`, `wait_ocr_result` | `Model.PP_OCRV6` | `Model.PP_OCRV5`, `Model.PP_OCRV6` | `OCROptions` |
+| Document parsing | `parse_document`, `submit_document_parsing`, `wait_document_parsing_result` | `Model.PADDLE_OCR_VL_16` | `Model.PP_STRUCTURE_V3`, `Model.PADDLE_OCR_VL`, `Model.PADDLE_OCR_VL_15`, `Model.PADDLE_OCR_VL_16` | Use `PPStructureV3Options` with `PP-StructureV3`, and `PaddleOCRVLOptions` with PaddleOCR-VL models. |
+
+## Configuration
+
+### Client Configuration
+
+```python
+client = PaddleOCRClient(
+    request_timeout=300.0,
+    poll_timeout=600.0,
+)
+```
+
+`request_timeout` limits one HTTP request, including submit, status, and result-resource downloads. `poll_timeout` limits the total wait time for `ocr`, `parse_document`, `wait_ocr_result`, and `wait_document_parsing_result`.
+
+Override the service base URL via the `PADDLEOCR_BASE_URL` environment variable or the `base_url` parameter:
+
+```python
+client = PaddleOCRClient(base_url="https://my-proxy.com/paddle")
+```
+
+### Request Options
+
+SDK parameter names use Python-idiomatic snake_case, automatically converted to camelCase on submission. Only non-`None` fields are sent; omitted fields use server-side defaults. See the Options class source or Official API Reference for complete field definitions.
+
+#### OCROptions (common fields)
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `use_doc_orientation_classify` | bool | Document orientation classification |
+| `use_doc_unwarping` | bool | Document unwarping |
+| `visualize` | bool | Return visualization images |
+
+#### PPStructureV3Options (common fields)
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `use_table_recognition` | bool | Table recognition |
+| `use_formula_recognition` | bool | Formula recognition |
+| `use_chart_recognition` | bool | Chart recognition |
+| `prettify_markdown` | bool | Markdown prettification |
+
+#### PaddleOCRVLOptions (common fields)
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `use_layout_detection` | bool | Layout detection |
+| `use_chart_recognition` | bool | Chart recognition |
+| `temperature` | float | Sampling temperature |
+| `prettify_markdown` | bool | Markdown prettification |
+
+## Error Handling
+
+All SDK errors inherit from `PaddleOCRAPIError`. Common typed errors include `AuthError`, `InvalidRequestError`, `RateLimitError`, `ServiceUnavailableError`, `APIError`, `NetworkError`, `JobFailedError`, `RequestTimeoutError`, `PollTimeoutError`, `ResponseFormatError`, and `ResultParseError`.
+
+## Batch Status
+
+When submitting jobs, pass `batch_id`. Later, use `client.get_batch_status("batch-id")` to inspect each job's state, progress, and result URL in that batch.
+
+## Official API Reference
+
+- [PP-OCRv5 API](https://ai.baidu.com/ai-doc/AISTUDIO/Dmh4onssk)
+- [PP-StructureV3 API](https://ai.baidu.com/ai-doc/AISTUDIO/7mfz6dgx9)
+- [PaddleOCR-VL API](https://ai.baidu.com/ai-doc/AISTUDIO/Vmkz2nz1p)
+- [PaddleOCR-VL-1.5 API](https://ai.baidu.com/ai-doc/AISTUDIO/fml7mozw5)
+
+## Quota and Error Codes
+
+- [API Quota Rules and Error Code Description](https://ai.baidu.com/ai-doc/AISTUDIO/pmjcld5qm)
diff --git a/docs/version3.x/inference_deployment/serving/paddleocr_official_api/python.md b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/python.md
new file mode 100644
index 0000000..9c47071
--- /dev/null
+++ b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/python.md
@@ -0,0 +1,129 @@
+---
+comments: true
+---
+
+# PaddleOCR 官方 API Python SDK
+
+Python SDK 通过 `paddleocr` 包中的 `PaddleOCRClient` 和 `AsyncPaddleOCRClient` 调用 PaddleOCR 官方 API。它提交 OCR 或文档解析任务到官方托管服务，不运行本地推理，也不加载本地模型。
+
+## 安装与认证
+
+先按 [安装 `paddleocr`](../../../installation.md#install-paddleocr) 安装 Python 包。安装 `paddleocr` 本体后即可使用此功能，无需安装额外依赖组。
+
+请先在 [AI Studio Access Token 页面](https://aistudio.baidu.com/account/accessToken) 获取访问令牌。
+
+```bash
+export PADDLEOCR_ACCESS_TOKEN="your-access-token"
+```
+
+`PaddleOCRClient()` 默认读取 `PADDLEOCR_ACCESS_TOKEN`，也支持 `PaddleOCRClient(token="...")`。未提供 token 时会抛出 `AuthError`。
+
+## 快速开始
+
+```python
+from paddleocr import PaddleOCRClient, Model
+
+client = PaddleOCRClient()
+result = client.ocr(
+    file_url="https://example.com/invoice.pdf",
+    model=Model.PP_OCRV5,
+)
+print(result.job_id, len(result.pages))
+client.close()
+```
+
+传入本地文件时使用 `file_path`。`file_url` 与 `file_path` 必须二选一。
+
+## 公共 API
+
+Python SDK 常用公共方法包括：
+
+- `ocr(...)`：提交 OCR 任务，等待完成并返回 OCR 结果。
+- `parse_document(...)`：提交文档解析任务，等待完成并返回文档解析结果。
+- `submit_ocr(...)`：只提交 OCR 任务，返回任务对象。
+- `submit_document_parsing(...)`：只提交文档解析任务，返回任务对象。
+- `get_status(job_id)`：执行一次非阻塞状态查询，不等待完成。
+- `wait_ocr_result(job)`：等待 OCR 任务完成并解析结果。
+- `wait_document_parsing_result(job)`：等待文档解析任务完成并解析结果。
+- `save_resource(resource_url, destination)`：保存单个资源 URL。
+- `save_ocr_result_resources(result, destination)`：保存 OCR 结果对象引用的资源。
+- `save_document_parsing_result_resources(result, destination)`：保存文档解析结果对象引用的资源。
+
+异步客户端 `AsyncPaddleOCRClient` 暴露上述任务操作和资源保存方法的异步版本。
+
+## 模型选择
+
+表中的 `Model` 枚举是官方 API 模型名字符串的类型安全写法，提交请求时会转换为对应的实际模型名。也可以直接传入官方 API 模型名字符串，例如 `model="PaddleOCR-VL-1.6"`。
+
+| 任务 | 适用接口 | 默认模型 | 可选模型 | 参数类型 |
+| --- | --- | --- | --- | --- |
+| OCR | `ocr`、`submit_ocr`、`wait_ocr_result` | `Model.PP_OCRV6` | `Model.PP_OCRV5`、`Model.PP_OCRV6` | `OCROptions` |
+| 文档解析 | `parse_document`、`submit_document_parsing`、`wait_document_parsing_result` | `Model.PADDLE_OCR_VL_16` | `Model.PP_STRUCTURE_V3`、`Model.PADDLE_OCR_VL`、`Model.PADDLE_OCR_VL_15`、`Model.PADDLE_OCR_VL_16` | 选择 `PP-StructureV3` 时传入 `PPStructureV3Options`；选择 `PaddleOCR-VL` 系列模型时传入 `PaddleOCRVLOptions`。 |
+
+## 配置与参数
+
+### 客户端配置
+
+```python
+client = PaddleOCRClient(
+    request_timeout=300.0,
+    poll_timeout=600.0,
+)
+```
+
+`request_timeout` 限制一次 HTTP 请求，包括提交、查询状态和下载结果资源。`poll_timeout` 限制 `ocr`、`parse_document`、`wait_ocr_result` 和 `wait_document_parsing_result` 的总等待时间。
+
+通过环境变量 `PADDLEOCR_BASE_URL` 或 `base_url` 参数指定自定义服务地址：
+
+```python
+client = PaddleOCRClient(base_url="https://my-proxy.com/paddle")
+```
+
+### 请求参数
+
+SDK 的参数名使用 Python 惯用的 snake_case，提交请求时自动转换为官方 API 的 camelCase。只需传入非 `None` 的字段，未设置的字段使用服务端默认值。完整字段定义见各 Options 类型源码或官方 API 参考。
+
+#### OCROptions（常用字段）
+
+| 字段 | 类型 | 说明 |
+|------|------|------|
+| `use_doc_orientation_classify` | bool | 文档方向分类 |
+| `use_doc_unwarping` | bool | 文档扭曲矫正 |
+| `visualize` | bool | 是否返回可视化结果图 |
+
+#### PPStructureV3Options（常用字段）
+
+| 字段 | 类型 | 说明 |
+|------|------|------|
+| `use_table_recognition` | bool | 表格识别 |
+| `use_formula_recognition` | bool | 公式识别 |
+| `use_chart_recognition` | bool | 图表识别 |
+| `prettify_markdown` | bool | Markdown 美化 |
+
+#### PaddleOCRVLOptions（常用字段）
+
+| 字段 | 类型 | 说明 |
+|------|------|------|
+| `use_layout_detection` | bool | 版面检测 |
+| `use_chart_recognition` | bool | 图表识别 |
+| `temperature` | float | 采样温度 |
+| `prettify_markdown` | bool | Markdown 美化 |
+
+## 错误处理
+
+所有 SDK 错误都继承自 `PaddleOCRAPIError`，常见类型包括 `AuthError`、`InvalidRequestError`、`RateLimitError`、`ServiceUnavailableError`、`APIError`、`NetworkError`、`JobFailedError`、`RequestTimeoutError`、`PollTimeoutError`、`ResponseFormatError` 和 `ResultParseError`。
+
+## 批量任务查询
+
+提交任务时可传入 `batch_id`。之后可使用 `client.get_batch_status("batch-id")` 查询该批次下各任务的状态、进度和结果 URL。
+
+## 官方 API 参考
+
+- [PP-OCRv5 API](https://ai.baidu.com/ai-doc/AISTUDIO/Kmfl2ycs0)
+- [PP-StructureV3 API](https://ai.baidu.com/ai-doc/AISTUDIO/Fmfz6oh2e)
+- [PaddleOCR-VL API](https://ai.baidu.com/ai-doc/AISTUDIO/2mh4okm66)
+- [PaddleOCR-VL-1.5 API](https://ai.baidu.com/ai-doc/AISTUDIO/Cmkz2m0ma)
+
+## 配额与错误码
+
+- [API 配额规则和错误码说明](https://ai.baidu.com/ai-doc/AISTUDIO/Xmjclapam)
diff --git a/docs/version3.x/inference_deployment/serving/paddleocr_official_api/typescript.en.md b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/typescript.en.md
new file mode 100644
index 0000000..8ed96fe
--- /dev/null
+++ b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/typescript.en.md
@@ -0,0 +1,139 @@
+---
+comments: true
+---
+
+# PaddleOCR official API TypeScript SDK
+
+The TypeScript SDK targets Node.js 18+ and calls the PaddleOCR official API for OCR and document parsing. It uses hosted PaddleOCR services and does not run local PaddleOCR inference.
+
+## Install And Authenticate
+
+First obtain an access token from the [AI Studio Access Token page](https://aistudio.baidu.com/account/accessToken).
+
+```bash
+npm install @paddleocr/api-sdk
+export PADDLEOCR_ACCESS_TOKEN="your-access-token"
+```
+
+The client reads `PADDLEOCR_ACCESS_TOKEN` by default and also accepts `token`:
+
+```ts
+import { PaddleOCRClient } from "@paddleocr/api-sdk";
+
+const client = new PaddleOCRClient({
+  token: process.env.PADDLEOCR_ACCESS_TOKEN,
+});
+```
+
+## Quick Start
+
+```ts
+import { Model, PaddleOCRClient } from "@paddleocr/api-sdk";
+
+const client = new PaddleOCRClient();
+const result = await client.ocr({
+  fileUrl: "https://example.com/invoice.pdf",
+  model: Model.PPOCRv5,
+});
+console.log(result.jobId, result.pages.length);
+```
+
+Use `filePath` for a local file. Pass exactly one of `fileUrl` or `filePath`.
+
+## Public API
+
+Common TypeScript public methods include:
+
+- `ocr(...)`: submit an OCR job, wait for completion, and return an OCR result.
+- `parseDocument(...)`: submit a document parsing job, wait for completion, and return a document parsing result.
+- `submitOcr(...)`: submit only an OCR job and return a job object.
+- `submitDocumentParsing(...)`: submit only a document parsing job and return a job object.
+- `getStatus(jobId)`: perform one non-blocking status request.
+- `waitOcrResult(job)`: wait for an OCR job and parse its result.
+- `waitDocumentParsingResult(job)`: wait for a document parsing job and parse its result.
+- `saveResource(resourceUrl, destination, options)`: save one resource URL.
+- `saveOcrResultResources(result, destination, options)`: save resources referenced by an OCR result object.
+- `saveDocumentParsingResultResources(result, destination, options)`: save resources referenced by a document parsing result object.
+
+## Choose Models
+
+The `Model` enum values in the table are type-safe aliases for the official API model-name strings. They are serialized to the corresponding model name when the request is submitted. You can also pass the official API model-name string directly, for example `model: "PaddleOCR-VL-1.6"`.
+
+| Task | Interfaces | Default model | Supported models | Option type |
+| --- | --- | --- | --- | --- |
+| OCR | `ocr`, `submitOcr`, `waitOcrResult` | `Model.PPOCRv6` | `Model.PPOCRv5`, `Model.PPOCRv6` | `OCROptions` |
+| Document parsing | `parseDocument`, `submitDocumentParsing`, `waitDocumentParsingResult` | `Model.PaddleOCRVL16` | `Model.PPStructureV3`, `Model.PaddleOCRVL`, `Model.PaddleOCRVL15`, `Model.PaddleOCRVL16` | Use `PPStructureV3Options` with `PPStructureV3`, and `PaddleOCRVLOptions` with PaddleOCR-VL models. |
+
+## Configuration
+
+### Client Configuration
+
+```ts
+const client = new PaddleOCRClient({
+  requestTimeout: 300_000,
+  pollTimeout: 600_000,
+});
+```
+
+`requestTimeout` limits one HTTP request, including submit, status, and resource downloads. `pollTimeout` limits the total wait time for `ocr`, `parseDocument`, `waitOcrResult`, and `waitDocumentParsingResult`. Public methods can also accept an `AbortSignal` for caller-driven cancellation.
+
+Override the service base URL via the `PADDLEOCR_BASE_URL` environment variable or the `baseUrl` option:
+
+```ts
+const client = new PaddleOCRClient({
+  baseUrl: "https://my-proxy.com/paddle",
+});
+```
+
+Inject a custom `fetch` implementation for proxies or custom network layers:
+
+```ts
+const client = new PaddleOCRClient({
+  fetch: myCustomFetch,
+});
+```
+
+### Request Options
+
+TypeScript SDK field names use camelCase, matching the official API directly. Omitted fields are not sent. See the interface source or Official API Reference for complete field definitions.
+
+#### OCROptions (common fields)
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `useDocOrientationClassify` | boolean | Document orientation classification |
+| `useDocUnwarping` | boolean | Document unwarping |
+| `visualize` | boolean | Return visualization images |
+
+#### PPStructureV3Options (common fields)
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `useTableRecognition` | boolean | Table recognition |
+| `useFormulaRecognition` | boolean | Formula recognition |
+| `useChartRecognition` | boolean | Chart recognition |
+| `prettifyMarkdown` | boolean | Markdown prettification |
+
+#### PaddleOCRVLOptions (common fields)
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `useLayoutDetection` | boolean | Layout detection |
+| `useChartRecognition` | boolean | Chart recognition |
+| `temperature` | number | Sampling temperature |
+| `prettifyMarkdown` | boolean | Markdown prettification |
+
+## Error Handling
+
+All SDK errors inherit from `PaddleOCRAPIError`. Common typed errors include `AuthError`, `InvalidRequestError`, `RateLimitError`, `ServiceUnavailableError`, `APIError`, `NetworkError`, `JobFailedError`, `RequestTimeoutError`, `PollTimeoutError`, `ResponseFormatError`, and `ResultParseError`.
+
+## Official API Reference
+
+- [PP-OCRv5 API](https://ai.baidu.com/ai-doc/AISTUDIO/Dmh4onssk)
+- [PP-StructureV3 API](https://ai.baidu.com/ai-doc/AISTUDIO/7mfz6dgx9)
+- [PaddleOCR-VL API](https://ai.baidu.com/ai-doc/AISTUDIO/Vmkz2nz1p)
+- [PaddleOCR-VL-1.5 API](https://ai.baidu.com/ai-doc/AISTUDIO/fml7mozw5)
+
+## Quota and Error Codes
+
+- [API Quota Rules and Error Code Description](https://ai.baidu.com/ai-doc/AISTUDIO/pmjcld5qm)
diff --git a/docs/version3.x/inference_deployment/serving/paddleocr_official_api/typescript.md b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/typescript.md
new file mode 100644
index 0000000..4585187
--- /dev/null
+++ b/docs/version3.x/inference_deployment/serving/paddleocr_official_api/typescript.md
@@ -0,0 +1,139 @@
+---
+comments: true
+---
+
+# PaddleOCR 官方 API TypeScript SDK
+
+TypeScript SDK 面向 Node.js 18 及以上环境，调用 PaddleOCR 官方 API 完成 OCR 与文档解析任务。它使用官方托管服务，不运行本地 PaddleOCR 推理。
+
+## 安装与认证
+
+请先在 [AI Studio Access Token 页面](https://aistudio.baidu.com/account/accessToken) 获取访问令牌。
+
+```bash
+npm install @paddleocr/api-sdk
+export PADDLEOCR_ACCESS_TOKEN="your-access-token"
+```
+
+客户端默认读取 `PADDLEOCR_ACCESS_TOKEN`，也可以传入 `token`：
+
+```ts
+import { PaddleOCRClient } from "@paddleocr/api-sdk";
+
+const client = new PaddleOCRClient({
+  token: process.env.PADDLEOCR_ACCESS_TOKEN,
+});
+```
+
+## 快速开始
+
+```ts
+import { Model, PaddleOCRClient } from "@paddleocr/api-sdk";
+
+const client = new PaddleOCRClient();
+const result = await client.ocr({
+  fileUrl: "https://example.com/invoice.pdf",
+  model: Model.PPOCRv5,
+});
+console.log(result.jobId, result.pages.length);
+```
+
+本地文件使用 `filePath`。`fileUrl` 与 `filePath` 必须二选一。
+
+## 公共 API
+
+TypeScript SDK 常用公共方法包括：
+
+- `ocr(...)`：提交 OCR 任务，等待完成并返回 OCR 结果。
+- `parseDocument(...)`：提交文档解析任务，等待完成并返回文档解析结果。
+- `submitOcr(...)`：只提交 OCR 任务，返回任务对象。
+- `submitDocumentParsing(...)`：只提交文档解析任务，返回任务对象。
+- `getStatus(jobId)`：执行一次非阻塞状态查询。
+- `waitOcrResult(job)`：等待 OCR 任务完成并解析结果。
+- `waitDocumentParsingResult(job)`：等待文档解析任务完成并解析结果。
+- `saveResource(resourceUrl, destination, options)`：保存单个资源 URL。
+- `saveOcrResultResources(result, destination, options)`：保存 OCR 结果对象引用的资源。
+- `saveDocumentParsingResultResources(result, destination, options)`：保存文档解析结果对象引用的资源。
+
+## 模型选择
+
+表中的 `Model` 枚举是官方 API 模型名字符串的类型安全写法，提交请求时会转换为对应的实际模型名。也可以直接传入官方 API 模型名字符串，例如 `model: "PaddleOCR-VL-1.6"`。
+
+| 任务 | 适用接口 | 默认模型 | 可选模型 | 参数类型 |
+| --- | --- | --- | --- | --- |
+| OCR | `ocr`、`submitOcr`、`waitOcrResult` | `Model.PPOCRv6` | `Model.PPOCRv5`、`Model.PPOCRv6` | `OCROptions` |
+| 文档解析 | `parseDocument`、`submitDocumentParsing`、`waitDocumentParsingResult` | `Model.PaddleOCRVL16` | `Model.PPStructureV3`、`Model.PaddleOCRVL`、`Model.PaddleOCRVL15`、`Model.PaddleOCRVL16` | 选择 `PPStructureV3` 时传入 `PPStructureV3Options`；选择 PaddleOCR-VL 系列模型时传入 `PaddleOCRVLOptions`。 |
+
+## 配置与参数
+
+### 客户端配置
+
+```ts
+const client = new PaddleOCRClient({
+  requestTimeout: 300_000,
+  pollTimeout: 600_000,
+});
+```
+
+`requestTimeout` 限制一次 HTTP 请求，包括提交、查询状态和下载资源。`pollTimeout` 限制 `ocr`、`parseDocument`、`waitOcrResult` 与 `waitDocumentParsingResult` 的总等待时间。公共方法还可以接收 `AbortSignal` 以便上层主动取消。
+
+通过环境变量 `PADDLEOCR_BASE_URL` 或 `baseUrl` 参数指定自定义服务地址：
+
+```ts
+const client = new PaddleOCRClient({
+  baseUrl: "https://my-proxy.com/paddle",
+});
+```
+
+通过 `fetch` 选项注入自定义 fetch 实现，适用于需要代理或自定义网络层的场景：
+
+```ts
+const client = new PaddleOCRClient({
+  fetch: myCustomFetch,
+});
+```
+
+### 请求参数
+
+TypeScript SDK 的参数名使用 camelCase，与官方 API 字段名一致。未设置的字段不会发送，使用服务端默认值。完整字段定义见接口源码或官方 API 参考。
+
+#### OCROptions（常用字段）
+
+| 字段 | 类型 | 说明 |
+|------|------|------|
+| `useDocOrientationClassify` | boolean | 文档方向分类 |
+| `useDocUnwarping` | boolean | 文档扭曲矫正 |
+| `visualize` | boolean | 是否返回可视化结果图 |
+
+#### PPStructureV3Options（常用字段）
+
+| 字段 | 类型 | 说明 |
+|------|------|------|
+| `useTableRecognition` | boolean | 表格识别 |
+| `useFormulaRecognition` | boolean | 公式识别 |
+| `useChartRecognition` | boolean | 图表识别 |
+| `prettifyMarkdown` | boolean | Markdown 美化 |
+
+#### PaddleOCRVLOptions（常用字段）
+
+| 字段 | 类型 | 说明 |
+|------|------|------|
+| `useLayoutDetection` | boolean | 版面检测 |
+| `useChartRecognition` | boolean | 图表识别 |
+| `temperature` | number | 采样温度 |
+| `prettifyMarkdown` | boolean | Markdown 美化 |
+
+## 错误处理
+
+所有 SDK 错误都继承自 `PaddleOCRAPIError`，常见类型包括 `AuthError`、`InvalidRequestError`、`RateLimitError`、`ServiceUnavailableError`、`APIError`、`NetworkError`、`JobFailedError`、`RequestTimeoutError`、`PollTimeoutError`、`ResponseFormatError` 和 `ResultParseError`。
+
+## 官方 API 参考
+
+- [PP-OCRv5 API](https://ai.baidu.com/ai-doc/AISTUDIO/Kmfl2ycs0)
+- [PP-StructureV3 API](https://ai.baidu.com/ai-doc/AISTUDIO/Fmfz6oh2e)
+- [PaddleOCR-VL API](https://ai.baidu.com/ai-doc/AISTUDIO/2mh4okm66)
+- [PaddleOCR-VL-1.5 API](https://ai.baidu.com/ai-doc/AISTUDIO/Cmkz2m0ma)
+
+## 配额与错误码
+
+- [API 配额规则和错误码说明](https://ai.baidu.com/ai-doc/AISTUDIO/Xmjclapam)
diff --git a/docs/version3.x/inference_deployment/serving/serving.en.md b/docs/version3.x/inference_deployment/serving/serving.en.md
new file mode 100644
index 0000000..4893916
--- /dev/null
+++ b/docs/version3.x/inference_deployment/serving/serving.en.md
@@ -0,0 +1,118 @@
+---
+comments: true
+---
+
+# Serving
+
+Serving is a common deployment method in real-world production environments. By encapsulating inference capabilities as services, clients can access these services via network requests to obtain inference results. **The client-side code can be written in different programming languages and does not need to match the server-side code.** PaddleOCR recommends using [PaddleX](https://github.com/PaddlePaddle/PaddleX) for serving. Please refer to [Differences and Connections between PaddleOCR and PaddleX](../../paddleocr_and_paddlex.en.md#1-differences-and-connections-between-paddleocr-and-paddlex) to understand the relationship between PaddleOCR and PaddleX.
+
+PaddleX provides the following serving solutions:
+
+- **Basic Serving**: An easy-to-use serving solution with low development costs.
+- **High-Stability Serving**: Built based on [NVIDIA Triton Inference Server](https://developer.nvidia.com/triton-inference-server). Compared to the basic serving, this solution offers higher stability and allows users to adjust configurations to optimize performance.
+
+**It is recommended to first use the basic serving solution for quick validation**, and then evaluate whether to try more complex solutions based on actual needs.
+
+## 1. Basic Serving
+
+### 1.1 Install Dependencies
+
+Run the following command to install the PaddleX serving plugin via PaddleX CLI:
+
+```bash
+paddlex --install serving
+```
+
+### 1.2 Run the Server
+
+Run the server via PaddleX CLI:
+
+```bash
+paddlex --serve --pipeline {PaddleX pipeline registration name or pipeline configuration file path} [{other command-line options}]
+```
+
+Take the general OCR pipeline as an example:
+
+```bash
+paddlex --serve --pipeline OCR
+```
+
+You should see information similar to the following:
+
+```text
+INFO:     Started server process [63108]
+INFO:     Waiting for application startup.
+INFO:     Application startup complete.
+INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+```
+
+To adjust configurations (such as model path, batch size, deployment device, etc.), specify `--pipeline` as a custom configuration file. Refer to [PaddleOCR and PaddleX](../../paddleocr_and_paddlex.en.md) for the mapping between PaddleOCR pipelines and PaddleX pipeline registration names, as well as how to obtain and modify PaddleX pipeline configuration files.
+
+The command-line options related to serving are as follows:
+
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>--pipeline</code></td>
+<td>PaddleX pipeline registration name or pipeline configuration file path.</td>
+</tr>
+<tr>
+<td><code>--device</code></td>
+<td>Deployment device for the pipeline. By default, a GPU will be used if available; otherwise, a CPU will be used."</td>
+</tr>
+<tr>
+<td><code>--host</code></td>
+<td>Hostname or IP address to which the server is bound. Defaults to <code>0.0.0.0</code>.</td>
+</tr>
+<tr>
+<td><code>--port</code></td>
+<td>Port number on which the server listens. Defaults to <code>8080</code>.</td>
+</tr>
+<tr>
+<td><code>--use_hpip</code></td>
+<td>If specified, uses high-performance inference. Refer to the High-Performance Inference documentation for more information.</td>
+</tr>
+<tr>
+<td><code>--hpi_config</code></td>
+<td>High-performance inference configuration. Refer to the High-Performance Inference documentation for more information.</td>
+</tr>
+</tbody>
+</table>
+
+### 1.3 Invoke the Service
+
+The <b>"Development Integration/Deployment"</b> section in the PaddleOCR pipeline tutorial provides API references and multi-language invocation examples for the service.
+
+## 2. High-Stability Serving
+
+Please refer to the [PaddleX Serving Guide](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_deploy/serving.html#2). More information about PaddleX pipeline configuration files can be found in [Using PaddleX Pipeline Configuration Files](../../paddleocr_and_paddlex.en.md#3-using-paddlex-pipeline-configuration-files).
+
+It should be noted that, due to the lack of fine-grained optimization and other reasons, the current high-stability serving deployment solution provided by PaddleOCR may not match the performance of the 2.x version based on PaddleServing. However, this new solution fully supports the PaddlePaddle 3.0 framework. We will continue to optimize it and consider introducing more performant deployment solutions in the future.
+
+## 3. Returning Binary Content as URLs
+
+By default, both basic serving and high-stability serving return images and other binary content in the response inline as Base64-encoded strings. When the response contains large images or a multi-page PDF, Base64 encoding can significantly inflate the payload; you can configure the service to return URLs instead. Enable it in the `Serving` section of the pipeline configuration file (`return_urls` is a top-level switch; object-storage settings live under `Serving.extra`) to return those fields as pre-signed URLs:
+
+```yaml
+Serving:
+  return_urls: true
+  extra:
+    file_storage:
+      type: bos
+      endpoint: <BOS endpoint, e.g. https://bj.bcebos.com>
+      ak: xxx
+      sk: xxx
+      bucket_name: <bucket name>
+    url_expires_in: 3600  # Pre-signed URL lifetime in seconds; -1 means no expiry
+```
+
+- Basic serving: write the configuration to the pipeline config file passed to `paddlex --serve --pipeline`.
+- High-stability serving: the same configuration applies — write it to `server/pipeline_config.yaml` inside the SDK and restart the container.
+
+Currently, URL return is only supported by the `bos` (Baidu Intelligent Cloud object storage) backend. URL return is controlled by the top-level `Serving.return_urls` field, which applies to every Base64-inlined file field in the response (not just images). For the full configuration reference, notes, and use cases, see [PaddleX Serving Guide - Returning Binary Content as URLs](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_deploy/serving.html#3). See the [Baidu Intelligent Cloud documentation](https://cloud.baidu.com/doc/BOS/index.html) for AK/SK retrieval.
diff --git a/docs/version3.x/inference_deployment/serving/serving.md b/docs/version3.x/inference_deployment/serving/serving.md
new file mode 100644
index 0000000..1a9032e
--- /dev/null
+++ b/docs/version3.x/inference_deployment/serving/serving.md
@@ -0,0 +1,118 @@
+---
+comments: true
+---
+
+# 服务化部署
+
+服务化部署是实际生产环境中常见的一种部署形式。通过将推理功能封装为服务，客户端可以通过网络请求来访问这些服务，以获取推理结果。**客户端代码可以由不同的编程语言编写，而不必与服务端代码保持一致。** PaddleOCR 推荐用户使用 [PaddleX](https://github.com/PaddlePaddle/PaddleX) 进行服务化部署。请阅读 [PaddleOCR 与 PaddleX 的区别与联系](../../paddleocr_and_paddlex.md#1-paddleocr-paddlex) 了解 PaddleOCR 与 PaddleX 的关系。
+
+PaddleX 提供以下服务化部署方案：
+
+- **基础服务化部署**：简单易用的服务化部署方案，开发成本低。
+- **高稳定性服务化部署**：基于 [NVIDIA Triton Inference Server](https://developer.nvidia.com/triton-inference-server) 打造。与基础服务化部署相比，该方案提供更高的稳定性，并允许用户调整配置以优化性能。
+
+**建议首先使用基础服务化部署方案进行快速验证**，然后根据实际需要，评估是否尝试更复杂的方案。
+
+## 1. 基础服务化部署
+
+### 1.1 安装依赖
+
+执行如下命令，通过 PaddleX CLI 安装 PaddleX 服务化部署插件：
+
+```bash
+paddlex --install serving
+```
+
+### 1.2 运行服务器
+
+通过 PaddleX CLI 运行服务器：
+
+```bash
+paddlex --serve --pipeline {PaddleX 产线注册名或产线配置文件路径} [{其他命令行选项}]
+```
+
+以通用 OCR 产线为例：
+
+```bash
+paddlex --serve --pipeline OCR
+```
+
+可以看到类似以下展示的信息：
+
+```text
+INFO:     Started server process [63108]
+INFO:     Waiting for application startup.
+INFO:     Application startup complete.
+INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+```
+
+如需调整配置（如模型路径、batch size、部署设备等），可指定 `--pipeline` 为自定义配置文件。请参考 [PaddleOCR 与 PaddleX](../../paddleocr_and_paddlex.md) 了解 PaddleOCR 产线与 PaddleX 产线注册名的对应关系，以及 PaddleX 产线配置文件的获取与修改方式。
+
+与服务化部署相关的命令行选项如下：
+
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>说明</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>--pipeline</code></td>
+<td>PaddleX 产线注册名或产线配置文件路径。</td>
+</tr>
+<tr>
+<td><code>--device</code></td>
+<td>产线部署设备。默认情况下，当 GPU 可用时，将使用 GPU；否则使用 CPU。</td>
+</tr>
+<tr>
+<td><code>--host</code></td>
+<td>服务器绑定的主机名或 IP 地址。默认为 <code>0.0.0.0</code>。</td>
+</tr>
+<tr>
+<td><code>--port</code></td>
+<td>服务器监听的端口号。默认为 <code>8080</code>。</td>
+</tr>
+<tr>
+<td><code>--use_hpip</code></td>
+<td>如果指定，则使用高性能推理。请参考高性能推理文档了解更多信息。</td>
+</tr>
+<tr>
+<td><code>--hpi_config</code></td>
+<td>高性能推理配置。请参考高性能推理文档了解更多信息。</td>
+</tr>
+</tbody>
+</table>
+
+### 1.3 调用服务
+
+PaddleOCR 产线使用教程中的 <b>“开发集成/部署”</b> 部分提供了服务的 API 参考与多语言调用示例。
+
+## 2. 高稳定性服务化部署
+
+请参考 [PaddleX 服务化部署指南](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/serving.html#2)。在 [使用 PaddleX 产线配置文件](../../paddleocr_and_paddlex.md#3-paddlex) 中，可以了解关于 PaddleX 产线配置文件的更多信息。
+
+需要说明的是，由于缺乏细粒度优化等原因，当前 PaddleOCR 提供的高稳定性服务化部署方案在性能上可能不及 2.x 版本基于 PaddleServing 的方案；但该新方案已对飞桨 3.0 框架提供了全面支持，我们也将持续优化，后续考虑推出性能更优的部署方案。
+
+## 3. 以 URL 形式返回二进制内容
+
+基础服务化与高稳定性服务化默认以 Base64 编码内联返回响应中的图像等二进制内容。当响应中包含较大图像或多页 PDF 时，Base64 会显著增加响应体积，可配置服务返回 URL。在产线配置文件的 `Serving` 节中开启（`return_urls` 为顶层字段，对象存储相关配置位于 `Serving.extra`），将相应字段改为以预签名 URL 返回：
+
+```yaml
+Serving:
+  return_urls: true
+  extra:
+    file_storage:
+      type: bos
+      endpoint: <BOS 访问域名，例如 https://bj.bcebos.com>
+      ak: xxx
+      sk: xxx
+      bucket_name: <存储空间名称>
+    url_expires_in: 3600  # 预签名 URL 有效期（秒），-1 表示不过期
+```
+
+- 基础服务化：上述配置写入 `paddlex --serve --pipeline` 指定的产线配置文件。
+- 高稳定性服务化：共用同一组配置项，写入 SDK 内的 `server/pipeline_config.yaml` 后重启容器即可。
+
+当前 URL 返回仅支持 `bos`（百度智能云对象存储）后端。URL 返回由顶层字段 `Serving.return_urls` 控制，作用于响应中所有 Base64 内联文件字段（不仅是图像）。完整配置项、注意事项与适用场景参见 [PaddleX 服务化部署指南 - 以 URL 形式返回二进制内容](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/serving.html#3)；AK/SK 获取等更多信息，请参考 [百度智能云官方文档](https://cloud.baidu.com/doc/BOS/index.html)。
diff --git a/docs/version3.x/installation.en.md b/docs/version3.x/installation.en.md
new file mode 100644
index 0000000..ce30445
--- /dev/null
+++ b/docs/version3.x/installation.en.md
@@ -0,0 +1,79 @@
+---
+comments: true
+---
+
+# Installation
+
+## 1. Install the PaddleOCR Python package and optional dependencies
+
+This section explains how to install, as needed, the `paddleocr` distribution package, optional dependency groups by capability domain, and the inference engine. This path covers running pretrained pipelines for inference locally, as well as auxiliary features such as document format conversion. **Model training and model export** are covered in Section 2 and are independent of the installation path above.
+
+**Python version requirement**: `paddleocr` itself and the `doc2md` dependency group support Python 3.8 and later. The other optional dependency groups (`doc-parser`, `ie`, `trans`, `all`, etc.) require Python 3.9 or later due to upstream dependencies.
+
+### 1.1 Install paddleocr
+
+Install the latest `paddleocr` from PyPI:
+
+```bash
+# Default capabilities only: general OCR and document image preprocessing
+python -m pip install paddleocr
+# All optional capabilities: document parsing, document understanding,
+# document translation, key information extraction, etc.
+# python -m pip install "paddleocr[all]"
+```
+
+Or install from source (tracks the repository’s current default branch by default):
+
+```bash
+# Default capabilities only: general OCR and document image preprocessing
+python -m pip install "paddleocr@git+https://github.com/PaddlePaddle/PaddleOCR.git"
+# All optional capabilities: document parsing, document understanding,
+# document translation, key information extraction, etc.
+# python -m pip install "paddleocr[all]@git+https://github.com/PaddlePaddle/PaddleOCR.git"
+```
+
+### 1.2 Choose dependency groups by capability
+
+Besides `all`, you can enable selected optional capabilities by specifying dependency groups. Each group corresponds to a capability domain (document parsing, information extraction, document translation, etc.). The available groups are:
+
+| Dependency Group Name | Corresponding Functionality |
+| - | - |
+| `doc-parser` | Document parsing. Extracts layout elements such as tables, formulas, seals, and images from documents. Includes model solutions such as PP-StructureV3 |
+| `ie` | Information extraction. Extracts key information such as names, dates, addresses, and amounts from documents. Includes model solutions such as PP-ChatOCRv4 |
+| `trans` | Document translation. Translates documents from one language to another. Includes model solutions such as PP-DocTranslation |
+| `doc2md` | Document-to-Markdown conversion. Quickly turns Word, Excel, and PowerPoint files into readable text |
+| `all` | Full functionality |
+
+The general OCR pipeline and the document image preprocessing pipeline require no extra dependency groups; document parsing, information extraction, document translation, and other capabilities follow the table above. See each pipeline’s documentation for its dependency group. For individual modules, install any dependency group that contains the module to use its basic functionality.
+
+### 1.3 Install the inference engine (as needed)
+
+PaddleOCR 3.5 uses a unified inference-engine configuration and can use backends such as PaddlePaddle and Transformers. To actually run model inference, install your chosen inference engine by following [Inference Engine and Configuration](./inference_deployment/local_inference/inference_engine.en.md).
+
+## 2. Install training and export dependencies
+
+To train models or export models, install the training-related dependencies separately. This path is a different installation dimension from the `paddleocr` package and optional groups in Section 1; both can coexist in one environment without mandatory isolation. Training and export depend on the PaddlePaddle framework. First install PaddlePaddle by following [PaddlePaddle Framework Installation](./paddlepaddle_installation.en.md). If another inference engine (such as Transformers) is already installed in the environment, you may encounter dependency conflicts; installing in a clean environment is recommended.
+
+**Python version requirement**: training and model export support Python 3.8 and later.
+
+Clone this repository locally, then install the remaining dependencies:
+
+```bash
+# Recommended method
+git clone https://github.com/PaddlePaddle/PaddleOCR
+
+# (Optional) Switch to a specific branch
+git checkout release/3.5
+
+# If cloning fails because of network issues, you can also use the Gitee repository:
+git clone https://gitee.com/paddlepaddle/PaddleOCR
+
+# Note: The code hosted on Gitee may lag behind the GitHub repository by 3 to 5 days.
+# Please prioritize the recommended method.
+```
+
+Run the following command to install the remaining training dependencies:
+
+```bash
+python -m pip install -r requirements.txt
+```
diff --git a/docs/version3.x/installation.md b/docs/version3.x/installation.md
new file mode 100644
index 0000000..5508431
--- /dev/null
+++ b/docs/version3.x/installation.md
@@ -0,0 +1,76 @@
+---
+comments: true
+---
+
+# 安装
+
+## 1. 安装 PaddleOCR Python 库及可选依赖
+
+本节说明如何按需安装`paddleocr` 分发包，按能力域启用的可选依赖组，以及安装推理引擎。这条路径适用于在本地调用预训练产线完成推理，或是使用文档格式转换等辅助功能。**模型训练与模型导出**见第 2 节，与上述安装路径相互独立。
+
+**Python 版本要求**：`paddleocr` 本体与 `doc2md` 依赖组支持 Python 3.8 及以上；其他可选依赖组（`doc-parser`、`ie`、`trans`、`all` 等）受上游依赖限制，需要 Python 3.9 及以上。
+
+### 1.1 安装 `paddleocr` {#install-paddleocr}
+
+从 PyPI 安装最新版本的 `paddleocr`：
+
+```bash
+# 仅需要通用 OCR 与文档图像预处理等默认能力
+python -m pip install paddleocr
+# 需要文档解析、文档理解、文档翻译、关键信息抽取等全部可选能力
+# python -m pip install "paddleocr[all]"
+```
+
+或从源码安装（默认跟踪仓库当前默认分支）：
+
+```bash
+# 仅需要通用 OCR 与文档图像预处理等默认能力
+python -m pip install "paddleocr@git+https://github.com/PaddlePaddle/PaddleOCR.git"
+# 需要文档解析、文档理解、文档翻译、关键信息抽取等全部可选能力
+# python -m pip install "paddleocr[all]@git+https://github.com/PaddlePaddle/PaddleOCR.git"
+```
+
+### 1.2 按功能选择依赖组
+
+除 `all` 外，可通过指定依赖组启用部分可选能力。依赖组表示能力域（文档解析、信息抽取、文档翻译等）。各依赖组如下：
+
+| 依赖组名称 | 对应的功能 |
+| - | - |
+| `doc-parser` | 文档解析，可用于提取文档中的表格、公式、印章、图片等版面元素，包含 PP-StructureV3 等模型方案 |
+| `ie` | 信息抽取，可用于从文档中提取关键信息，如姓名、日期、地址、金额等，包含 PP-ChatOCRv4 等模型方案 |
+| `trans` | 文档翻译，可用于将文档从一种语言翻译为另一种语言，包含 PP-DocTranslation 等模型方案 |
+| `doc2md` | 文档转 MarkDown，可用于将 Word、Excel、PowerPoint 文件快速转为可读文本 |
+| `all` | 完整功能 |
+
+通用 OCR 产线与文档图像预处理产线无需额外依赖组；文档解析、信息抽取、文档翻译等按上表安装对应组。各产线所属依赖组见对应产线文档；单功能模块在任一包含该模块的依赖组安装后即可调用其基础能力。
+
+### 1.3 安装推理引擎（按需）
+
+PaddleOCR 3.5 采用统一推理引擎配置，底层可对接飞桨、Transformers 等。若要实际执行模型推理，请参考 [推理引擎与配置说明](./inference_deployment/local_inference/inference_engine.md) 安装所选推理引擎。
+
+## 2. 安装训练与导出依赖
+
+若要进行模型训练或模型导出，需另行安装训练相关依赖。该路径与第 1 节的 `paddleocr` 包及可选依赖组属于不同安装维度，同一环境中可同时存在，无需强制隔离。训练与导出依赖飞桨框架，请先参考 [飞桨框架安装](./paddlepaddle_installation.md) 安装飞桨框架。如果当前环境中已安装飞桨之外的推理引擎（如 Transformers），可能会出现依赖冲突，建议在全新的环境中进行安装。
+
+**Python 版本要求**：训练与模型导出支持 Python 3.8 及以上。
+
+将本仓库克隆到本地后安装其余依赖：
+
+```bash
+# 推荐方式
+git clone https://github.com/PaddlePaddle/PaddleOCR
+
+# （可选）切换到指定分支
+git checkout release/3.5
+
+# 如果因为网络问题无法克隆成功，也可选择使用码云上的仓库：
+git clone https://gitee.com/paddlepaddle/PaddleOCR
+
+# 注：码云托管代码可能无法实时同步本 GitHub 项目更新，存在3~5天延时，请优先使用推荐方式。
+```
+
+执行如下命令安装其余训练依赖：
+
+```bash
+python -m pip install -r requirements.txt
+```
diff --git a/docs/version3.x/integrations/mcp_server.en.md b/docs/version3.x/integrations/mcp_server.en.md
new file mode 100644
index 0000000..53dc8c2
--- /dev/null
+++ b/docs/version3.x/integrations/mcp_server.en.md
@@ -0,0 +1,439 @@
+---
+comments: true
+---
+
+# PaddleOCR MCP Server
+
+[![PaddleOCR](https://img.shields.io/badge/OCR-PaddleOCR-orange)](https://github.com/PaddlePaddle/PaddleOCR)
+[![FastMCP](https://img.shields.io/badge/Built%20with-FastMCP%20v2-blue)](https://gofastmcp.com)
+
+PaddleOCR provides a lightweight [Model Context Protocol (MCP)](https://modelcontextprotocol.io/introduction) server designed to integrate PaddleOCR’s text recognition, layout parsing, and other capabilities into various large-model applications.
+
+### Key features include:
+
+- **Currently Supported Models**
+
+    | Model | MCP tool name | Description |
+    | --- | --- | --- |
+    | `PP-OCRv5`, `PP-OCRv5-latin`, `PP-OCRv6` | `ocr` | Performs text detection and recognition on images and PDF files. |
+    | `PP-StructureV3` | `pp_structurev3` | Identifies and extracts text blocks, titles, paragraphs, images, tables, and other layout elements from images or PDF files, converting the input into Markdown documents. |
+    | `PaddleOCR-VL`, `PaddleOCR-VL-1.5`, `PaddleOCR-VL-1.6` | `paddleocr_vl` | Performs layout parsing with a VLM-based approach and converts the input into Markdown documents. |
+
+- **Supported Inference Methods**
+    - **Local Inference**: Runs PaddleOCR pipelines directly on the local machine. This method has certain requirements for the local environment and hardware performance, and is suitable for offline use and scenarios with strict data privacy requirements.
+    - **Official API**: Invokes the PaddleOCR Official API. This method is suitable for quickly trying out features, validating solutions, and other no-code development scenarios.
+    - **Qianfan API**: Calls the API provided by Baidu AI Cloud's Qianfan platform.
+    - **Self-hosted API**: Invokes the user's self-hosted PaddleOCR inference service. This method offers serving advantages and high flexibility, suitable for scenarios requiring customized service configurations, as well as those with strict data privacy requirements. **Currently, only the basic serving solution is supported.**
+
+## Examples:
+
+The following showcases creative use cases built with the PaddleOCR MCP server combined with other tools:
+
+### Demo 1
+In Claude for Desktop, extract handwritten content from images and save to note-taking software Notion. The PaddleOCR MCP server extracts text, formulas and other information from images while preserving document structure.
+<div align="center">
+  <img width="65%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/mcp_demo/note_to_notion.gif" alt="note_to_notion">
+  <img width="30%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/mcp_demo/note.jpg" alt="note">
+</div>
+
+- Note: In addition to the PaddleOCR MCP server, this demo also uses the [Notion MCP server](https://developers.notion.com/docs/mcp).
+
+---
+
+### Demo 2
+In VSCode, convert handwritten ideas or pseudocode into runnable Python scripts that comply with project coding standards with one click, and upload them to GitHub repositories. The PaddleOCR MCP server extracts explicitly handwritten code from images for subsequent processing.
+
+<div align="center">
+  <img width="70%" img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/mcp_demo/code_to_github.gif" alt="code_to_github">
+</div>
+
+- In addition to the PaddleOCR MCP server, this demo also uses the [filesystem MCP server](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem).
+
+---
+
+### Demo 3
+
+In Claude for Desktop, convert PDF documents or images containing complex tables, formulas, handwritten text and other content into locally editable files.
+
+#### Demo 3.1
+
+Convert complex PDF documents with tables and watermarks to editable doc/Word format:
+<div align="center">
+  <img width="70%" img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/mcp_demo/pdf_to_file.gif" alt="pdf_to_file">
+</div>
+
+#### Demo 3.2
+
+Convert images containing formulas and tables to editable csv/Excel format:
+<div align="center">
+  <img width="70%" img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/00136903a4d0b5f11bd978cb0ef5d3c44f3aa5e9/images/paddleocr/mcp_demo/table_to_excel1.png" alt="table_to_excel1">
+  <img width="50%" img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/00136903a4d0b5f11bd978cb0ef5d3c44f3aa5e9/images/paddleocr/mcp_demo/table_to_excel2.png" alt="table_to_excel2">
+  <img width="45%" img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/00136903a4d0b5f11bd978cb0ef5d3c44f3aa5e9/images/paddleocr/mcp_demo/table_to_excel3.png" alt="table_to_excel3">
+</div>
+
+- In addition to the PaddleOCR MCP server, this demo also uses the [filesystem MCP server](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem).
+
+---
+
+### Table of Contents
+
+- [Table of Contents](#table-of-contents)
+- [1. Installation](#1-installation)
+- [2. Using with Claude for Desktop](#2-using-with-claude-for-desktop)
+    - [2.1 Quick Start](#21-quick-start)
+    - [2.2 MCP Host Configuration Details](#22-mcp-host-configuration-details)
+    - [2.3 Inference Methods](#23-inference-methods)
+    - [2.4 Using `uvx`](#24-using-uvx)
+- [3. Running the Server](#3-running-the-server)
+- [4. Parameter Reference](#4-parameter-reference)
+- [5. Known Limitations](#5-known-limitations)
+
+## 1. Installation
+
+This section explains how to install the `paddleocr-mcp` library via pip.
+
+`paddleocr-mcp` requires Python 3.10 or later.
+`paddleocr-mcp` depends on `paddleocr>=3.7.0` by default, so Official API, Qianfan API, and self-hosted API modes do not require installing PaddleOCR separately. Local inference additionally requires the document-parsing dependencies and an inference engine required to run PaddleOCR pipelines locally; see [Method 1: Local Inference](#method-1-local-inference) for details.
+
+Install from PyPI:
+
+```bash
+pip install -U paddleocr-mcp
+```
+
+Install from source:
+
+```bash
+git clone https://github.com/PaddlePaddle/PaddleOCR.git
+pip install -e mcp_server
+```
+
+For local inference, install the optional extras described in [Method 1: Local Inference](#method-1-local-inference).
+
+To verify successful installation:
+
+```bash
+paddleocr_mcp --help
+```
+
+If help information is printed after running the command above, the installation succeeded.
+
+PaddleOCR also supports running the server without installation through methods like `uvx`; for details, see [2. Using with Claude for Desktop](#2-using-with-claude-for-desktop).
+
+## 2. Using with Claude for Desktop
+
+This section explains how to use the PaddleOCR MCP server within Claude for Desktop. The steps are also applicable to other MCP hosts with minor adjustments.
+
+### 2.1 Quick Start
+
+The following quick start uses **Official API** inference as an example to get you started.
+
+1. **Install `paddleocr-mcp`**
+
+    Refer to [1. Installation](#1-installation).
+
+2. **Obtain an Access Token**
+
+    Obtain your access token from the [AI Studio Access Token page](https://aistudio.baidu.com/account/accessToken).
+
+3. **Add MCP Server Configuration**
+
+    Locate the `claude_desktop_config.json` configuration file:
+
+    - **macOS**: `~/Library/Application Support/Claude/claude_desktop_config.json`
+    - **Windows**: `%APPDATA%\Claude\claude_desktop_config.json`
+    - **Linux**: `~/.config/Claude/claude_desktop_config.json`
+
+    Open the `claude_desktop_config.json` file, adjust the configuration according to the example below, and fill it into `claude_desktop_config.json`.
+
+    ```json
+    {
+      "mcpServers": {
+        "paddleocr": {
+          "command": "paddleocr_mcp",
+          "args": [],
+          "env": {
+            "PADDLEOCR_MCP_MODEL": "PP-OCRv5",
+            "PADDLEOCR_MCP_PPOCR_SOURCE": "aistudio",
+            "PADDLEOCR_MCP_AISTUDIO_ACCESS_TOKEN": "<your-access-token>"
+          }
+        }
+      }
+    }
+    ```
+
+    **Notes**:
+
+    - Replace `<your-access-token>` with your access token.
+    - To use a custom service address, set the `PADDLEOCR_MCP_AISTUDIO_BASE_URL` environment variable.
+
+    **Important**:
+
+    - Do not expose your **access token**.
+    - If `paddleocr_mcp` is not in your system's `PATH`, set `command` to the absolute path of the executable.
+
+4. **Restart the MCP Host**
+
+    Restart Claude for Desktop. The `paddleocr` server should now be available in the application.
+
+### 2.2 MCP Host Configuration Details
+
+In the configuration file for Claude for Desktop, you need to define how the MCP server is started. The key fields are as follows:
+
+- `command`: `paddleocr_mcp` (if the executable can be found in the `PATH`) or the absolute path.
+- `args`: Configurable command-line arguments, such as `["--verbose"]`. See [4. Parameter Reference](#4-parameter-reference) for details.
+- `env`: Configurable environment variables. See [4. Parameter Reference](#4-parameter-reference) for details.
+
+### 2.3 Inference Methods
+
+You can configure the MCP server according to your requirements to use different inference methods. The operational procedures vary for different methods, which will be explained in detail below.
+
+#### Method 1: Local Inference {#method-1-local-inference}
+
+1. Install `paddleocr-mcp` and the local inference dependencies. `paddleocr-mcp` already depends on PaddleOCR; local inference additionally requires the document-parsing dependencies and an inference engine. You can install them manually by referring to the [PaddleOCR installation guide](../installation.en.md), or use the corresponding optional dependencies:
+    - `paddleocr-mcp[local]`: includes `paddleocr[doc-parser]>=3.7.0` (without the inference engine).
+    - `paddleocr-mcp[local-cpu]`: based on `local`, additionally includes the CPU PaddlePaddle inference engine (`paddlepaddle>=3.2.1`).
+
+    ```bash
+    # Install document-parsing dependencies for local inference (inference engine not included):
+    pip install "paddleocr-mcp[local]"
+    # Install the CPU PaddlePaddle framework in addition to local:
+    pip install "paddleocr-mcp[local-cpu]"
+    ```
+
+    To avoid dependency conflicts, **it is strongly recommended to install in an isolated virtual environment**.
+2. Refer to the configuration example below to modify the `claude_desktop_config.json` file.
+3. Restart the MCP host.
+
+Configuration example:
+
+```json
+{
+  "mcpServers": {
+    "paddleocr": {
+      "command": "paddleocr_mcp",
+      "args": [],
+      "env": {
+        "PADDLEOCR_MCP_MODEL": "PP-OCRv5",
+        "PADDLEOCR_MCP_PPOCR_SOURCE": "local"
+      }
+    }
+  }
+}
+```
+
+**Notes**:
+
+- `PADDLEOCR_MCP_MODEL` should be set to the model name. See Section 4 for details.
+- `PADDLEOCR_MCP_PIPELINE_CONFIG` is optional. If not set, the default pipeline configuration is used. To adjust the configuration, such as changing models, refer to the [PaddleOCR documentation](../paddleocr_and_paddlex.md) to export the pipeline configuration file, and set `PADDLEOCR_MCP_PIPELINE_CONFIG` to the absolute path of this file.
+- **Inference Performance Tips**:
+
+    If you encounter long inference time or insufficient memory, consider adjusting the pipeline configuration:
+
+    - **PP-StructureV3 Pipeline**:
+
+        - Disable unused features, such as setting `use_formula_recognition` to `False` to disable formula recognition.
+        - Use lightweight models, such as replacing the OCR model with the `mobile` version or switching to a lightweight formula recognition model like PP-FormulaNet-S.
+
+        The following sample code exports a PP-StructureV3 pipeline configuration with most optional features disabled and some key models replaced with lightweight versions.
+
+        ```python
+        from paddleocr import PPStructureV3
+
+        pipeline = PPStructureV3(
+            use_doc_orientation_classify=False, # Disable document image orientation classification
+            use_doc_unwarping=False,            # Disable text image unwarping
+            use_textline_orientation=False,     # Disable text line orientation classification
+            use_formula_recognition=False,      # Disable formula recognition
+            use_seal_recognition=False,         # Disable seal text recognition
+            use_table_recognition=False,        # Disable table recognition
+            use_chart_recognition=False,        # Disable chart parsing
+            # Use lightweight models
+            text_detection_model_name="PP-OCRv5_mobile_det",
+            text_recognition_model_name="PP-OCRv5_mobile_rec",
+            layout_detection_model_name="PP-DocLayout-S",
+        )
+
+        # The configuration file is saved to `PP-StructureV3.yaml`
+        pipeline.export_paddlex_config_to_yaml("PP-StructureV3.yaml")
+        ```
+
+    **For PaddleOCR-VL series, CPU inference is not recommended.**
+
+#### Method 2: Official API
+
+Refer to [2.1 Quick Start](#21-quick-start).
+
+For tasks other than text recognition, set `PADDLEOCR_MCP_MODEL` correctly (see Section 4 for parameter details).
+
+#### Method 3: Qianfan API
+
+1. Install `paddleocr-mcp`.
+2. Obtain an API key by referring to the [Qianfan Platform Official Documentation](https://cloud.baidu.com/doc/qianfan-api/s/ym9chdsy5).
+3. Refer to the configuration example below to modify the `claude_desktop_config.json` file.
+4. Restart the MCP host.
+
+Configuration example:
+
+```json
+{
+  "mcpServers": {
+    "paddleocr": {
+      "command": "paddleocr_mcp",
+      "args": [],
+      "env": {
+        "PADDLEOCR_MCP_MODEL": "PaddleOCR-VL",
+        "PADDLEOCR_MCP_PPOCR_SOURCE": "qianfan",
+        "PADDLEOCR_MCP_QIANFAN_API_KEY": "<your-api-key>"
+      }
+    }
+  }
+}
+```
+
+**Notes**:
+
+- `PADDLEOCR_MCP_MODEL` should be set to the model name. Qianfan supports only `PP-StructureV3` and `PaddleOCR-VL`.
+- `PADDLEOCR_MCP_QIANFAN_BASE_URL` is the Qianfan API base URL (optional).
+- `PADDLEOCR_MCP_QIANFAN_API_KEY` is your Qianfan API key for authentication.
+
+#### Method 4: Self-hosted API
+
+1. In the environment where you need to run the PaddleOCR inference server, refer to the [PaddleOCR serving documentation](../inference_deployment/serving/serving.en.md) to run the inference server.
+2. Install `paddleocr-mcp` in the environment where you need to run the MCP server.
+3. Refer to the configuration example below to modify the `claude_desktop_config.json` file.
+4. Restart the MCP host.
+
+Configuration example:
+
+```json
+{
+  "mcpServers": {
+    "paddleocr": {
+      "command": "paddleocr_mcp",
+      "args": [],
+      "env": {
+        "PADDLEOCR_MCP_MODEL": "PP-OCRv5",
+        "PADDLEOCR_MCP_PPOCR_SOURCE": "self_hosted",
+        "PADDLEOCR_MCP_SELF_HOSTED_BASE_URL": "<your-server-url>"
+      }
+    }
+  }
+}
+```
+
+**Notes**:
+
+- `PADDLEOCR_MCP_MODEL` should be set to the model name. See Section 4 for details.
+- Replace `<your-server-url>` with the underlying service base URL (e.g. `http://127.0.0.1:8080`, **without** path suffixes such as `/ocr` or `/layout-parsing`; MCP appends them by pipeline).
+
+### 2.4 Using `uvx`
+
+PaddleOCR also supports starting the MCP server via `uvx`. With this approach, manual installation of `paddleocr-mcp` is not required. The main steps are as follows:
+
+1. Install [uv](https://docs.astral.sh/uv/#installation).
+2. Modify `claude_desktop_config.json`. Examples:
+
+  Self-hosted API inference example:
+
+  ```json
+  {
+    "mcpServers": {
+     "paddleocr": {
+      "command": "uvx",
+      "args": [
+        "--from",
+        "paddleocr-mcp",
+        "paddleocr_mcp"
+      ],
+      "env": {
+        "PADDLEOCR_MCP_MODEL": "PP-OCRv5",
+        "PADDLEOCR_MCP_PPOCR_SOURCE": "self_hosted",
+        "PADDLEOCR_MCP_SELF_HOSTED_BASE_URL": "<your-server-url>"
+      }
+     }
+    }
+  }
+  ```
+
+  Local inference (CPU inference, using the optional `local-cpu` extra) example:
+
+  ```json
+  {
+    "mcpServers": {
+     "paddleocr": {
+      "command": "uvx",
+      "args": [
+        "--from",
+        "paddleocr-mcp[local-cpu]",
+        "paddleocr_mcp"
+      ],
+      "env": {
+        "PADDLEOCR_MCP_MODEL": "PP-OCRv5",
+        "PADDLEOCR_MCP_PPOCR_SOURCE": "local"
+      }
+     }
+    }
+  }
+  ```
+
+  For local inference dependencies, performance tuning, and pipeline configuration, refer to [Method 1: Local Inference](#method-1-local-inference).
+
+  Due to the use of a different startup method, the `command` and `args` settings in the configuration file differ from the previously described approach. However, the command-line arguments and environment variables supported by the MCP service (such as `PADDLEOCR_MCP_SELF_HOSTED_BASE_URL`) can still be set in the same way.
+
+## 3. Running the Server
+
+In addition to MCP hosts like Claude for Desktop, you can also run the PaddleOCR MCP server via the CLI.
+
+Run the following command to print help information:
+
+```bash
+paddleocr_mcp --help
+```
+
+Example commands:
+
+```bash
+# PP-OCRv5 + Official API + stdio
+PADDLEOCR_MCP_AISTUDIO_ACCESS_TOKEN=xxxxxx paddleocr_mcp --model PP-OCRv5 --ppocr_source aistudio
+
+# PP-OCRv6 + Official API + stdio
+paddleocr_mcp --model PP-OCRv6 --ppocr_source aistudio
+
+# PP-StructureV3 + Local Inference + stdio
+paddleocr_mcp --model PP-StructureV3 --ppocr_source local
+
+# OCR + Self-hosted API + Streamable HTTP
+paddleocr_mcp --model PP-OCRv5 --ppocr_source self_hosted --self-hosted-base-url http://127.0.0.1:8080 --http
+```
+
+See [4. Parameter Reference](#4-parameter-reference) for all parameters supported by the PaddleOCR MCP server.
+
+## 4. Parameter Reference
+
+You can control the MCP server via environment variables or CLI arguments.
+
+| Environment Variable | CLI Argument | Type | Description | Options | Default |
+|:---------|:-----------|:-----|:-----|:-------|:-------|
+| `PADDLEOCR_MCP_MODEL` | `--model` | `str` | Model to run. MCP selects the tool automatically from the model. | `"PP-OCRv5"`, `"PP-OCRv5-latin"`, `"PP-OCRv6"`, `"PP-StructureV3"`, `"PaddleOCR-VL"`, `"PaddleOCR-VL-1.5"`, `"PaddleOCR-VL-1.6"` | `"PP-OCRv6"` |
+| `PADDLEOCR_MCP_PPOCR_SOURCE` | `--ppocr_source` | `str` | Source of PaddleOCR capabilities. | `"local"` (local inference), `"aistudio"` (Official API), `"qianfan"` (Qianfan API), `"self_hosted"` (self-hosted API) | `"local"` |
+| `PADDLEOCR_MCP_AISTUDIO_BASE_URL` | `--aistudio-base-url` | `str` | AI Studio API base URL (optional for `aistudio` source). | - | `None` |
+| `PADDLEOCR_MCP_QIANFAN_BASE_URL` | `--qianfan-base-url` | `str` | Qianfan API base URL (optional for `qianfan` source). | - | `https://qianfan.baidubce.com/v2/ocr` |
+| `PADDLEOCR_MCP_SELF_HOSTED_BASE_URL` | `--self-hosted-base-url` | `str` | Self-hosted PaddleX serve base URL (required for `self_hosted` source). | - | `None` |
+| `PADDLEOCR_MCP_QIANFAN_API_KEY` | `--qianfan_api_key` | `str` | Qianfan API authentication key (required for `qianfan` source). | - | `None` |
+| `PADDLEOCR_MCP_AISTUDIO_ACCESS_TOKEN` | `--aistudio_access_token` | `str` | AI Studio access token (required for `aistudio` source). | - | `None` |
+| `PADDLEOCR_MCP_HTTP_TIMEOUT` | `--http-timeout` | `int` | HTTP read timeout in seconds for synchronous APIs (`qianfan`, `self_hosted`). | - | `600` |
+| `PADDLEOCR_MCP_AISTUDIO_REQUEST_TIMEOUT` | `--aistudio-request-timeout` | `int` | Per-request HTTP timeout in seconds for AI Studio API calls (job submission, status checks, etc.). | - | `120` |
+| `PADDLEOCR_MCP_AISTUDIO_POLL_TIMEOUT` | `--aistudio-poll-timeout` | `int` | Total job polling timeout in seconds for AI Studio. | - | `600` |
+| `PADDLEOCR_MCP_DEVICE` | `--device` | `str` | Device for inference (only effective for `local` source). | - | `None` |
+| `PADDLEOCR_MCP_PIPELINE_CONFIG` | `--pipeline_config` | `str` | PaddleOCR pipeline configuration file path (only effective for `local` source). | - | `None` |
+| - | `--http` | `bool` | Use Streamable HTTP transport instead of stdio (for remote deployment and multiple clients). | - | `False` |
+| - | `--host` | `str` | Host address for Streamable HTTP mode. | - | `"127.0.0.1"` |
+| - | `--port` | `int` | Port for Streamable HTTP mode. | - | `8000` |
+| - | `--verbose` | `bool` | Enable verbose logging for debugging. | - | `False` |
+
+## 5. Known Limitations
+
+- Under local inference, the exposed MCP tool cannot process PDF document inputs that are Base64 encoded.
+- Under local inference, the exposed MCP tool does not infer file type from the model's `file_type` prompt; some complex URLs may fail to process.
+- For the PP-StructureV3 and PaddleOCR-VL series, if the input file contains images, the returned results may significantly increase token usage. If image content is not needed, you can explicitly exclude it through prompts to reduce resource consumption.
diff --git a/docs/version3.x/integrations/mcp_server.md b/docs/version3.x/integrations/mcp_server.md
new file mode 100644
index 0000000..b5d906f
--- /dev/null
+++ b/docs/version3.x/integrations/mcp_server.md
@@ -0,0 +1,440 @@
+---
+comments: true
+---
+
+# PaddleOCR MCP 服务器
+
+[![PaddleOCR](https://img.shields.io/badge/OCR-PaddleOCR-orange)](https://github.com/PaddlePaddle/PaddleOCR)
+[![FastMCP](https://img.shields.io/badge/Built%20with-FastMCP%20v2-blue)](https://gofastmcp.com)
+
+PaddleOCR 提供轻量级的 [Model Context Protocol（MCP）](https://modelcontextprotocol.io/introduction) 服务器，旨在将 PaddleOCR 的文字识别、版面解析等能力集成到各种大模型应用中。
+
+### 主要功能如下：
+
+- **当前支持的模型**
+
+    | 模型 | MCP 工具名称 | 说明 |
+    | --- | --- | --- |
+    | `PP-OCRv5`、`PP-OCRv5-latin`、`PP-OCRv6` | `ocr` | 对图像和 PDF 文件进行文本检测与识别。 |
+    | `PP-StructureV3` | `pp_structurev3` | 从图像或 PDF 文件中识别和提取文本块、标题、段落、图片、表格以及其他版面元素，将输入转换为 Markdown 文档。 |
+    | `PaddleOCR-VL`、`PaddleOCR-VL-1.5`、`PaddleOCR-VL-1.6` | `paddleocr_vl` | 使用基于多模态大模型的方案进行版面解析，将输入转换为 Markdown 文档。 |
+
+- **支持的推理方式**
+    - **本地推理**：在本机直接运行 PaddleOCR 产线。此方式对本地环境与计算机性能有一定要求，适用于需要离线使用、对数据隐私有严格要求的场景。
+    - **官方 API**：调用 PaddleOCR 官方提供的 API。此方式适合快速体验功能、快速验证方案等，也适用于零代码开发场景。
+    - **千帆 API**：调用百度智能云千帆大模型平台提供的 API。
+    - **自建 API**：调用用户自建的 PaddleOCR 推理服务。此方式具备服务化部署优势及高度灵活性，适用于需要自定义服务配置的场景，同时也适用于对数据隐私有严格要求的场景。**目前暂时只支持基础服务化部署方案。**
+
+## 示例：
+
+以下展示了使用 PaddleOCR MCP 服务器结合其他工具搭建的创意案例：
+
+### Demo 1
+在 Claude for Desktop 中，提取图像中的手写内容，并存到笔记软件 Notion。PaddleOCR MCP 服务器从图像中提取了文字、公式等信息，并保留了文档的结构。
+<div align="center">
+  <img width="65%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/mcp_demo/note_to_notion.gif" alt="note_to_notion">
+  <img width="30%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/mcp_demo/note.jpg" alt="note">
+</div>
+
+- 注: 除 PaddleOCR MCP 服务器外，此 demo 还使用 [Notion MCP 服务器](https://developers.notion.com/docs/mcp)。
+
+---
+
+### Demo 2
+在 VSCode 中，根据手写思路或伪代码一键转换为可运行并符合项目代码风格规范的 Python 脚本，并将其上传到 GitHub 仓库中。PaddleOCR MCP 服务器从图像中准确提取手写代码供后续步骤使用。
+
+<div align="center">
+  <img width="70%" img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/mcp_demo/code_to_github.gif" alt="code_to_github">
+</div>
+
+- 除 PaddleOCR MCP 服务器外，此 demo 还使用 [filesystem MCP 服务器](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem)。
+
+---
+
+### Demo 3
+
+在 Claude for Desktop 中，将含有复杂表格、公式、手写文字等内容的 PDF 文档或图片转存为本地可编辑文件。
+
+#### Demo 3.1
+
+含表格水印复杂文档PDF 转为 doc/Word 可编辑格式：
+<div align="center">
+  <img width="70%" img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/mcp_demo/pdf_to_file.gif" alt="pdf_to_file">
+</div>
+
+#### Demo 3.2
+
+含公式表格图片转为 csv/Excel 可编辑格式：
+<div align="center">
+  <img width="70%" img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/00136903a4d0b5f11bd978cb0ef5d3c44f3aa5e9/images/paddleocr/mcp_demo/table_to_excel1.png" alt="table_to_excel1">
+  <img width="50%" img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/00136903a4d0b5f11bd978cb0ef5d3c44f3aa5e9/images/paddleocr/mcp_demo/table_to_excel2.png" alt="table_to_excel2">
+  <img width="45%" img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/00136903a4d0b5f11bd978cb0ef5d3c44f3aa5e9/images/paddleocr/mcp_demo/table_to_excel3.png" alt="table_to_excel3">
+</div>
+
+
+- 除 PaddleOCR MCP 服务器外，此 demo 还使用 [filesystem MCP 服务器](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem)。
+
+---
+
+### 目录
+
+- [目录](#_3)
+- [1. 安装](#1)
+- [2. 在 Claude for Desktop 中使用](#2-claude-for-desktop)
+    - [2.1 快速开始](#21)
+    - [2.2 MCP 主机配置说明](#22-mcp)
+    - [2.3 推理方式说明](#23)
+    - [2.4 使用 `uvx`](#24-uvx)
+- [3. 运行服务器](#3)
+- [4. 参数说明](#4)
+- [5. 已知局限性](#5)
+
+## 1. 安装
+
+本节将介绍如何通过 pip 安装 `paddleocr-mcp` 库。
+
+`paddleocr-mcp` 要求 Python 版本不低于 3.10。
+`paddleocr-mcp` 默认依赖 `paddleocr>=3.7.0`，因此官方 API、千帆 API 和自建 API 推理方式无需额外安装 PaddleOCR。如需使用本地推理，还需要安装本地运行产线所需的文档解析依赖和推理引擎，详情请参考 [方式一：本地推理](#方式一本地推理)。
+
+从 PyPI 安装：
+
+```bash
+pip install -U paddleocr-mcp
+```
+
+从项目源码安装：
+
+```bash
+git clone https://github.com/PaddlePaddle/PaddleOCR.git
+pip install -e mcp_server
+```
+
+如需使用本地推理，请安装 [方式一：本地推理](#方式一本地推理) 中说明的可选依赖。
+
+可通过以下命令检查是否安装成功：
+
+```bash
+paddleocr_mcp --help
+```
+
+如果执行上述命令后打印出了帮助信息，则说明安装成功。
+
+PaddleOCR 也支持通过 `uvx` 等方式免安装运行服务器，详情请参考 [2. 在 Claude for Desktop 中使用](#2-claude-for-desktop) 中的说明。
+
+## 2. 在 Claude for Desktop 中使用
+
+本节将介绍如何在 Claude for Desktop 中使用 PaddleOCR MCP 服务器。对于其他 MCP 主机，也可参照本节的步骤，并根据实际情况进行相应调整。
+
+### 2.1 快速开始
+
+接下来以 **官方 API** 推理方式为例，引导您快速上手。
+
+1. **安装 `paddleocr-mcp`**
+
+    请参考 [1. 安装](#1)。
+
+2. **获取访问令牌**
+
+    请先在 [AI Studio Access Token 页面](https://aistudio.baidu.com/account/accessToken) 获取访问令牌。
+
+3. **添加 MCP 服务器配置**
+
+    在以下位置之一找到 Claude for Desktop 配置文件：
+
+    - **macOS**：`~/Library/Application Support/Claude/claude_desktop_config.json`
+    - **Windows**：`%APPDATA%\Claude\claude_desktop_config.json`
+    - **Linux**：`~/.config/Claude/claude_desktop_config.json`
+
+    打开 `claude_desktop_config.json` 文件，参考如下示例调整配置，填充到 `claude_desktop_config.json` 中。
+
+    ```json
+    {
+      "mcpServers": {
+        "paddleocr": {
+          "command": "paddleocr_mcp",
+          "args": [],
+          "env": {
+            "PADDLEOCR_MCP_MODEL": "PP-OCRv5",
+            "PADDLEOCR_MCP_PPOCR_SOURCE": "aistudio",
+            "PADDLEOCR_MCP_AISTUDIO_ACCESS_TOKEN": "<your-access-token>"
+          }
+        }
+      }
+    }
+    ```
+
+    **说明**：
+
+    - 将 `<your-access-token>` 替换为您的访问令牌。
+    - 如需使用自定义服务地址，可设置 `PADDLEOCR_MCP_AISTUDIO_BASE_URL` 环境变量。
+
+    **注意**：
+
+    - 请勿泄漏您的 **访问令牌**。
+    - 如果 `paddleocr_mcp` 无法在系统 `PATH` 中找到，请将 `command` 设置为可执行文件的绝对路径。
+
+4. **重启 MCP 主机**
+
+    重启 Claude for Desktop。新的 `paddleocr` 服务现在应该可以在应用中使用了。
+
+### 2.2 MCP 主机配置说明
+
+在 Claude for Desktop 的配置文件中，您需要定义 MCP 服务器的启动方式。关键字段如下：
+
+- `command`：`paddleocr_mcp`（如果可执行文件可在 `PATH` 中找到）或绝对路径。
+- `args`：可配置命令行参数，如 `["--verbose"]`。详见 [4. 参数说明](#4)。
+- `env`：可配置环境变量。详见 [4. 参数说明](#4)。
+
+### 2.3 推理方式说明
+
+您可以根据需求配置 MCP 服务器，使其使用不同的推理方式。不同推理方式需要的操作流程有所不同，下面将详细介绍。
+
+#### 方式一：本地推理 {#方式一本地推理}
+
+1. 安装 `paddleocr-mcp` 及本地推理依赖。`paddleocr-mcp` 已默认依赖 PaddleOCR；本地推理还需要文档解析依赖和推理引擎。可以参考 [PaddleOCR 安装文档](../installation.md) 手动安装，也可以使用相应的可选依赖：
+    - `paddleocr-mcp[local]`：包含 `paddleocr[doc-parser]>=3.7.0`（不包含推理引擎）。
+    - `paddleocr-mcp[local-cpu]`：在 `local` 基础上额外包含 CPU 版本的 PaddlePaddle 推理引擎（`paddlepaddle>=3.2.1`）。
+
+    ```bash
+    # 安装本地推理所需的文档解析依赖（不包含推理引擎）
+    pip install "paddleocr-mcp[local]"
+    # 在 local 基础上额外安装 CPU 版本飞桨框架
+    pip install "paddleocr-mcp[local-cpu]"
+    ```
+
+    为避免依赖冲突，**强烈建议在独立的虚拟环境中安装**。
+2. 参考下方的配置示例更改 `claude_desktop_config.json` 文件内容。
+3. 重启 MCP 主机。
+
+配置示例：
+
+```json
+{
+  "mcpServers": {
+    "paddleocr": {
+      "command": "paddleocr_mcp",
+      "args": [],
+      "env": {
+        "PADDLEOCR_MCP_MODEL": "PP-OCRv5",
+        "PADDLEOCR_MCP_PPOCR_SOURCE": "local"
+      }
+    }
+  }
+}
+```
+
+**说明**：
+
+- `PADDLEOCR_MCP_MODEL` 需要被设置为模型名称。详见第 4 节。
+- `PADDLEOCR_MCP_PIPELINE_CONFIG` 为可选项，不设置时使用产线默认配置。如需调整配置，例如更换模型，请参考 [PaddleOCR 文档](../paddleocr_and_paddlex.md) 导出产线配置文件，并将 `PADDLEOCR_MCP_PIPELINE_CONFIG` 设置为配置文件的绝对路径。
+- **推理性能提示**：
+
+    如果使用过程中出现推理耗时过长、内存不足等问题，可考虑参考如下建议调整产线配置：
+
+    - **PP-StructureV3 产线**：
+    
+        - 关闭不需要用到的功能，例如设置 `use_formula_recognition` 为 `False` 以禁用公式识别。
+        - 使用轻量级的模型，例如将 OCR 模型替换为 `mobile` 版本、换用轻量的公式识别模型 PP-FormulaNet-S 等。
+
+        以下示例代码可用于获取产线配置文件，其中关闭了 PP-StructureV3 产线的大部分可选功能，同时将部分关键模型更换为轻量级版本。
+
+        ```python
+        from paddleocr import PPStructureV3
+
+        pipeline = PPStructureV3(
+            use_doc_orientation_classify=False, # 禁用文档图像方向分类
+            use_doc_unwarping=False,            # 禁用文本图像矫正
+            use_textline_orientation=False,     # 禁用文本行方向分类
+            use_formula_recognition=False,      # 禁用公式识别
+            use_seal_recognition=False,         # 禁用印章文本识别
+            use_table_recognition=False,        # 禁用表格识别
+            use_chart_recognition=False,        # 禁用图表解析
+            # 使用轻量级模型
+            text_detection_model_name="PP-OCRv5_mobile_det",
+            text_recognition_model_name="PP-OCRv5_mobile_rec",
+            layout_detection_model_name="PP-DocLayout-S",
+        )
+
+        # 配置文件保存到 `PP-StructureV3.yaml` 中
+        pipeline.export_paddlex_config_to_yaml("PP-StructureV3.yaml")
+        ```
+    
+    **对于 PaddleOCR-VL 系列，不建议使用 CPU 推理。**
+
+#### 方式二：官方 API
+
+请参考 [2.1 快速开始](#21)。
+
+对于文字识别以外的任务，请正确设置 `PADDLEOCR_MCP_MODEL`（参数说明详见第 4 节）。
+
+#### 方式三：千帆 API
+
+1. 安装 `paddleocr-mcp`。
+2. 参考 [千帆平台官方文档](https://cloud.baidu.com/doc/qianfan-api/s/ym9chdsy5) 获取 API key。
+3. 参考下方的配置示例更改 `claude_desktop_config.json` 文件内容。
+4. 重启 MCP 主机。
+
+配置示例：
+
+```json
+{
+  "mcpServers": {
+    "paddleocr": {
+      "command": "paddleocr_mcp",
+      "args": [],
+      "env": {
+        "PADDLEOCR_MCP_MODEL": "PaddleOCR-VL",
+        "PADDLEOCR_MCP_PPOCR_SOURCE": "qianfan",
+        "PADDLEOCR_MCP_QIANFAN_API_KEY": "<your-api-key>"
+      }
+    }
+  }
+}
+```
+
+**说明**：
+
+- `PADDLEOCR_MCP_MODEL` 需设置为模型名称；千帆仅支持 `PP-StructureV3` 与 `PaddleOCR-VL`。
+- `PADDLEOCR_MCP_QIANFAN_BASE_URL` 为千帆 API 基础 URL（可选）。
+- `PADDLEOCR_MCP_QIANFAN_API_KEY` 是您的千帆 API key，用于认证。
+
+#### 方式四：自建 API
+
+1. 在需要运行 PaddleOCR 推理服务器的环境中，参考 [PaddleOCR 服务化部署文档](../inference_deployment/serving/serving.md) 运行推理服务器。
+2. 在需要运行 MCP 服务器的环境中安装 `paddleocr-mcp`。
+3. 参考下方的配置示例更改 `claude_desktop_config.json` 文件内容。
+4. 重启 MCP 主机。
+
+配置示例：
+
+```json
+{
+  "mcpServers": {
+    "paddleocr": {
+      "command": "paddleocr_mcp",
+      "args": [],
+      "env": {
+        "PADDLEOCR_MCP_MODEL": "PP-OCRv5",
+        "PADDLEOCR_MCP_PPOCR_SOURCE": "self_hosted",
+        "PADDLEOCR_MCP_SELF_HOSTED_BASE_URL": "<your-server-url>"
+      }
+    }
+  }
+}
+```
+
+**说明**：
+
+- `PADDLEOCR_MCP_MODEL` 需要被设置为模型名称。详见第 4 节。
+- 将 `<your-server-url>` 替换为底层服务的基础 URL（如：`http://127.0.0.1:8080`，**不要**包含 `/ocr`、`/layout-parsing` 等路径；MCP 会按产线自动拼接）。
+
+### 2.4 使用 `uvx`
+
+PaddleOCR 也支持通过 `uvx` 启动 MCP 服务器。这种方式不需要手动安装 `paddleocr-mcp`。主要步骤如下：
+
+1. 安装 [uv](https://docs.astral.sh/uv/#installation)。
+2. 修改 `claude_desktop_config.json` 文件。示例如下：
+
+  自建 API 推理示例：
+
+  ```json
+  {
+    "mcpServers": {
+     "paddleocr": {
+      "command": "uvx",
+      "args": [
+        "--from",
+        "paddleocr-mcp",
+        "paddleocr_mcp"
+      ],
+      "env": {
+        "PADDLEOCR_MCP_MODEL": "PP-OCRv5",
+        "PADDLEOCR_MCP_PPOCR_SOURCE": "self_hosted",
+        "PADDLEOCR_MCP_SELF_HOSTED_BASE_URL": "<your-server-url>"
+      }
+     }
+    }
+  }
+  ```
+
+  本地推理（CPU 推理，使用可选依赖 `local-cpu`）示例：
+
+  ```json
+  {
+    "mcpServers": {
+     "paddleocr": {
+      "command": "uvx",
+      "args": [
+        "--from",
+        "paddleocr-mcp[local-cpu]",
+        "paddleocr_mcp"
+      ],
+      "env": {
+        "PADDLEOCR_MCP_MODEL": "PP-OCRv5",
+        "PADDLEOCR_MCP_PPOCR_SOURCE": "local"
+      }
+     }
+    }
+  }
+  ```
+
+  如需了解本地推理的依赖、性能调优及产线配置，请参考 [方式一：本地推理](#方式一本地推理) 部分。
+
+  由于使用了不一样的启动方式，配置文件中 `command` 和 `args` 的设置都与前文介绍的方式存在不同，但 MCP 服务本身支持的命令行参数与环境变量（如 `PADDLEOCR_MCP_SELF_HOSTED_BASE_URL`）仍然可以以相同的方式设置。
+
+## 3. 运行服务器
+
+除了在 Claude for Desktop 等 MCP 主机中使用外，您也可以通过 CLI 运行 PaddleOCR MCP 服务器。
+
+执行以下命令可以打印帮助信息：
+
+```bash
+paddleocr_mcp --help
+```
+
+示例命令如下：
+
+```bash
+# PP-OCRv5 + 官方 API + stdio
+PADDLEOCR_MCP_AISTUDIO_ACCESS_TOKEN=xxxxxx paddleocr_mcp --model PP-OCRv5 --ppocr_source aistudio
+
+# PP-OCRv6 + 官方 API + stdio
+paddleocr_mcp --model PP-OCRv6 --ppocr_source aistudio
+
+# PP-StructureV3 + 本地推理 + stdio
+paddleocr_mcp --model PP-StructureV3 --ppocr_source local
+
+# OCR + 自建 API + Streamable HTTP
+paddleocr_mcp --model PP-OCRv5 --ppocr_source self_hosted --self-hosted-base-url http://127.0.0.1:8080 --http
+```
+
+在 [4. 参数说明](#4) 中可以了解 PaddleOCR MCP 服务器支持的全部参数。
+
+## 4. 参数说明
+
+您可以通过环境变量或命令行参数来控制 MCP 服务器的行为。
+
+| 环境变量 | 命令行参数 | 类型 | 描述 | 可选值 | 默认值 |
+|:---------|:-----------|:-----|:-----|:-------|:-------|
+| `PADDLEOCR_MCP_MODEL` | `--model` | `str` | 要运行的模型。MCP 会根据模型自动选择对应的工具。 | `"PP-OCRv5"`，`"PP-OCRv5-latin"`，`"PP-OCRv6"`，`"PP-StructureV3"`，`"PaddleOCR-VL"`，`"PaddleOCR-VL-1.5"`，`"PaddleOCR-VL-1.6"` | `"PP-OCRv6"` |
+| `PADDLEOCR_MCP_PPOCR_SOURCE` | `--ppocr_source` | `str` | PaddleOCR能力来源。 | `"local"`（本地推理），`"aistudio"`（官方 API），`"qianfan"`（千帆 API），`"self_hosted"`（自建 API） | `"local"` |
+| `PADDLEOCR_MCP_AISTUDIO_BASE_URL` | `--aistudio-base-url` | `str` | AI Studio API 基础 URL（`aistudio` 推理方式下可选）。 | - | `None` |
+| `PADDLEOCR_MCP_QIANFAN_BASE_URL` | `--qianfan-base-url` | `str` | 千帆 API 基础 URL（`qianfan` 推理方式下可选）。 | - | `https://qianfan.baidubce.com/v2/ocr` |
+| `PADDLEOCR_MCP_SELF_HOSTED_BASE_URL` | `--self-hosted-base-url` | `str` | 自建 PaddleX 服务基础 URL（`self_hosted` 推理方式下必需）。 | - | `None` |
+| `PADDLEOCR_MCP_QIANFAN_API_KEY` | `--qianfan_api_key` | `str` | 千帆 API 认证密钥（`qianfan` 推理方式下必需）。 | - | `None` |
+| `PADDLEOCR_MCP_AISTUDIO_ACCESS_TOKEN` | `--aistudio_access_token` | `str` | AI Studio 访问令牌（`aistudio` 推理方式下必需）。 | - | `None` |
+| `PADDLEOCR_MCP_HTTP_TIMEOUT` | `--http-timeout` | `int` | 同步 HTTP API（`qianfan`、`self_hosted`）的单次 POST 读取超时（秒）。 | - | `600` |
+| `PADDLEOCR_MCP_AISTUDIO_REQUEST_TIMEOUT` | `--aistudio-request-timeout` | `int` | AI Studio 单次 HTTP 请求超时（秒），用于提交任务、查询状态等。 | - | `120` |
+| `PADDLEOCR_MCP_AISTUDIO_POLL_TIMEOUT` | `--aistudio-poll-timeout` | `int` | AI Studio 任务轮询总等待上限（秒）。 | - | `600` |
+| `PADDLEOCR_MCP_DEVICE` | `--device` | `str` | 指定运行推理的设备（仅在 `local` 推理方式下生效）。 | - | `None` |
+| `PADDLEOCR_MCP_PIPELINE_CONFIG` | `--pipeline_config` | `str` | PaddleOCR 产线配置文件路径（仅在 `local` 推理方式下生效）。 | - | `None` |
+| - | `--http` | `bool` | 使用 Streamable HTTP 传输而非 stdio（适用于远程部署和多客户端）。 | - | `False` |
+| - | `--host` | `str` | Streamable HTTP 模式的主机地址。 | - | `"127.0.0.1"` |
+| - | `--port` | `int` | Streamable HTTP 模式的端口。 | - | `8000` |
+| - | `--verbose` | `bool` | 启用详细日志记录，便于调试。 | - | `False` |
+
+## 5. 已知局限性
+
+- 在本地推理方式下，对外暴露的 MCP 工具无法处理 Base64 编码的 PDF 文档输入。
+- 在本地推理方式下，对外暴露的 MCP 工具不会根据模型提示的 `file_type` 推断文件类型，对于一些复杂 URL 可能处理失败。
+- 对于 PP-StructureV3 和 PaddleOCR-VL 系列，若输入文件中包含图像，返回结果可能会显著增加 token 使用量。若无需图像内容，可通过提示词明确排除，以降低资源消耗。
diff --git a/docs/version3.x/integrations/skills.en.md b/docs/version3.x/integrations/skills.en.md
new file mode 100644
index 0000000..795483a
--- /dev/null
+++ b/docs/version3.x/integrations/skills.en.md
@@ -0,0 +1,145 @@
+---
+comments: true
+---
+
+# PaddleOCR Agent Skills
+
+PaddleOCR provides official Agent Skills that package the routing rules, calling steps, configuration requirements, and best practices for handling results in OCR and document parsing tasks into on-demand modular capabilities, helping Skills-enabled AI apps complete text recognition and layout parsing more reliably.
+
+## Choose the Right Skill First
+
+| Need | Recommended Skill | Output |
+| --- | --- | --- |
+| Extract plain text from images or PDFs | `paddleocr-text-recognition` | line-level text with bounding boxes and confidence scores |
+| Preserve headings, paragraphs, tables, formulas, and layout structure | `paddleocr-doc-parsing` | Markdown / structured output |
+
+## Included Skills
+
+- `paddleocr-text-recognition`: extracts text from images, scans, and PDF files.
+- `paddleocr-doc-parsing`: parses complex document layouts and converts them to Markdown or structured output.
+
+## Prerequisites
+
+1. Install Python 3.9 or later on the machine that runs the skill.
+2. Install PaddleOCR 3.7.0+: `pip install "paddleocr>=3.7.0"`
+3. Get access token from [AI Studio](https://aistudio.baidu.com/account/accessToken)
+
+## Install into AI Apps
+
+> The instructions below cover both skills. Install and configure only the skill or skills you need.
+
+### Option 1: Install via `skills` CLI
+
+The `skills` CLI installs skills globally on the device so they can be used by supported AI apps. [Node.js](https://nodejs.org/en/download) is required.
+
+```shell
+npx skills add PaddlePaddle/PaddleOCR -g --skill paddleocr-text-recognition -y
+npx skills add PaddlePaddle/PaddleOCR -g --skill paddleocr-doc-parsing -y
+```
+
+> This repository is relatively large. On slower networks, `npx skills add` may time out. If that happens, clone the repository locally first and then install from the local path:
+>
+> ```shell
+> git clone https://github.com/PaddlePaddle/PaddleOCR.git
+> npx skills add ./PaddleOCR/skills/paddleocr-text-recognition
+> npx skills add ./PaddleOCR/skills/paddleocr-doc-parsing
+> ```
+
+### Option 2: Install via `clawhub` (OpenClaw)
+
+```shell
+clawhub install paddleocr-text-recognition
+clawhub install paddleocr-doc-parsing
+```
+
+See the [OpenClaw Skills documentation](https://docs.openclaw.ai/tools/skills) for details.
+
+### Option 3: Manual Installation
+
+If the options above do not fit your environment, clone the repository and copy the skill directories to the location required by your AI app:
+
+```shell
+git clone https://github.com/PaddlePaddle/PaddleOCR.git
+```
+
+The skill source code is located under `PaddleOCR/skills`. Refer to your AI app documentation to complete the installation:
+
+- Claude Code: <https://code.claude.com/docs/en/skills>
+- claude.ai: <https://support.claude.com/en/articles/12512180-use-skills-in-claude>
+- OpenClaw: <https://docs.openclaw.ai/tools/skills>
+
+## Configure Environment Variables
+
+After installation, configure the following environment variables:
+
+- Required: `PADDLEOCR_ACCESS_TOKEN` (access token)
+- Optional: `PADDLEOCR_BASE_URL` (API base URL, defaults to official service)
+
+Get access token: visit [AI Studio Access Token](https://aistudio.baidu.com/account/accessToken)
+
+Examples for some AI apps:
+
+- **Claude Code**: add an `env` field to `.claude/settings.local.json` in your project:
+
+  ```json
+  {
+    "env": {
+      "PADDLEOCR_ACCESS_TOKEN": "<ACCESS_TOKEN>"
+    }
+  }
+  ```
+
+- **OpenClaw**: add skill configuration to `~/.openclaw/openclaw.json`:
+
+  ```json
+  {
+    "skills": {
+      "entries": {
+        "paddleocr-text-recognition": {
+          "enabled": true,
+          "env": {
+            "PADDLEOCR_ACCESS_TOKEN": "<ACCESS_TOKEN>"
+          }
+        },
+        "paddleocr-doc-parsing": {
+          "enabled": true,
+          "env": {
+            "PADDLEOCR_ACCESS_TOKEN": "<ACCESS_TOKEN>"
+          }
+        }
+      }
+    }
+  }
+  ```
+
+## Usage Examples
+
+After configuration, describe the task in natural language and provide a file URL or local path so the AI app can invoke the corresponding skill.
+
+### `paddleocr-text-recognition`
+
+URL example:
+
+```text
+Extract all text from this file: https://example.com/invoice.jpg
+```
+
+Local file example:
+
+```text
+Extract all text from local file C:\docs\invoice.pdf
+```
+
+### `paddleocr-doc-parsing`
+
+URL example:
+
+```text
+Parse this PDF and return the main body plus all tables: https://example.com/report.pdf
+```
+
+Local file example:
+
+```text
+Parse local file C:\docs\report.pdf and return complete structured output.
+```
diff --git a/docs/version3.x/integrations/skills.md b/docs/version3.x/integrations/skills.md
new file mode 100644
index 0000000..97142ed
--- /dev/null
+++ b/docs/version3.x/integrations/skills.md
@@ -0,0 +1,145 @@
+---
+comments: true
+---
+
+# PaddleOCR Agent Skills
+
+PaddleOCR 提供官方 Agent Skills，将 OCR 与文档解析任务的触发规则、调用步骤、配置要求和结果处理最佳实践打包为可按需加载的模块化能力，帮助支持 Skills 的 AI 应用更稳定地完成文字识别与版面解析任务。
+
+## 先选合适的 Skill
+
+| 需求 | 推荐 Skill | 输出 |
+| --- | --- | --- |
+| 只想提取图片或 PDF 中的纯文本 | `paddleocr-text-recognition` | 行级文本及对应边界框和置信度分数 |
+| 需要保留标题、段落、表格、公式等文档结构 | `paddleocr-doc-parsing` | Markdown / 结构化结果 |
+
+## 包含的 Skills
+
+- `paddleocr-text-recognition`：用于识别图片、扫描件与 PDF 中的文字。
+- `paddleocr-doc-parsing`：用于解析复杂文档版面，并转换为 Markdown 或结构化结果。
+
+## 安装前准备
+
+1. 请确保执行 Skill 的设备已安装 Python 3.9 或以上版本。
+2. Skills 依赖 PaddleOCR 3.7.0+。请安装：`pip install "paddleocr>=3.7.0"`
+3. 获取 access token：访问 [AI Studio Access Token 页面](https://aistudio.baidu.com/account/accessToken)
+
+## 安装到 AI 应用
+
+> 以下说明涵盖两个 Skill。只需安装并配置您需要的 Skill 即可。
+
+### 方式一：通过 `skills` CLI 安装
+
+`skills` CLI 可将 Skill 全局安装到设备上，安装后各 AI 应用均可使用。需要先安装 [Node.js](https://nodejs.org/en/download)。
+
+```shell
+npx skills add PaddlePaddle/PaddleOCR -g --skill paddleocr-text-recognition -y
+npx skills add PaddlePaddle/PaddleOCR -g --skill paddleocr-doc-parsing -y
+```
+
+> 由于 PaddleOCR 仓库较大，在网络较慢的环境下 `npx skills add` 可能因超时而失败。如遇此情况，可先将仓库克隆到本地，再从本地路径安装：
+>
+> ```shell
+> git clone https://github.com/PaddlePaddle/PaddleOCR.git
+> npx skills add ./PaddleOCR/skills/paddleocr-text-recognition
+> npx skills add ./PaddleOCR/skills/paddleocr-doc-parsing
+> ```
+
+### 方式二：通过 `clawhub` 安装（OpenClaw）
+
+```shell
+clawhub install paddleocr-text-recognition
+clawhub install paddleocr-doc-parsing
+```
+
+详见 [OpenClaw Skills 文档](https://docs.openclaw.ai/tools/skills)。
+
+### 方式三：手动安装
+
+如果上述方式不适用，也可以先克隆仓库，再手动将 Skill 目录拷贝到 AI 应用指定的位置：
+
+```shell
+git clone https://github.com/PaddlePaddle/PaddleOCR.git
+```
+
+Skill 源码位于 `PaddleOCR/skills` 目录。请参考对应 AI 应用的安装文档：
+
+- Claude Code：<https://code.claude.com/docs/en/skills>
+- claude.ai：<https://support.claude.com/en/articles/12512180-use-skills-in-claude>
+- OpenClaw：<https://docs.openclaw.ai/tools/skills>
+
+## 配置环境变量 {#配置环境变量}
+
+安装完成后，配置如下环境变量：
+
+- 必填：`PADDLEOCR_ACCESS_TOKEN`（access token）
+- 可选：`PADDLEOCR_BASE_URL`（API base URL，默认官方服务）
+
+获取 access token：访问 [AI Studio Access Token 页面](https://aistudio.baidu.com/account/accessToken)
+
+部分 AI 应用的配置方式如下：
+
+- **Claude Code**：在项目的 `.claude/settings.local.json` 中添加 `env` 字段：
+
+  ```json
+  {
+    "env": {
+      "PADDLEOCR_ACCESS_TOKEN": "<ACCESS_TOKEN>"
+    }
+  }
+  ```
+
+- **OpenClaw**：在 `~/.openclaw/openclaw.json` 中添加 Skill 配置：
+
+  ```json
+  {
+    "skills": {
+      "entries": {
+        "paddleocr-text-recognition": {
+          "enabled": true,
+          "env": {
+            "PADDLEOCR_ACCESS_TOKEN": "<ACCESS_TOKEN>"
+          }
+        },
+        "paddleocr-doc-parsing": {
+          "enabled": true,
+          "env": {
+            "PADDLEOCR_ACCESS_TOKEN": "<ACCESS_TOKEN>"
+          }
+        }
+      }
+    }
+  }
+  ```
+
+## 使用示例
+
+配置完成后，可以直接用自然语言描述任务，并附上文件 URL 或本地路径，让 AI 应用调用对应 Skill。
+
+### `paddleocr-text-recognition`
+
+解析 URL 示例：
+
+```text
+提取这个文件中的全部文本：https://example.com/invoice.jpg
+```
+
+解析本地文件示例：
+
+```text
+提取本地文件 C:\docs\invoice.pdf 中的全部文本。
+```
+
+### `paddleocr-doc-parsing`
+
+解析 URL 示例：
+
+```text
+解析这个 PDF，并返回主体内容和全部表格：https://example.com/report.pdf
+```
+
+解析本地文件示例：
+
+```text
+解析本地文件 C:\docs\report.pdf，并返回完整结构化结果。
+```
diff --git a/docs/version3.x/logging.en.md b/docs/version3.x/logging.en.md
new file mode 100644
index 0000000..5cdd9f9
--- /dev/null
+++ b/docs/version3.x/logging.en.md
@@ -0,0 +1,26 @@
+---
+comments: true
+---
+
+# Logging
+
+This document explains how to configure logging for the `paddleocr` Python package. The `paddleocr` package uses a different logging system than training scripts, and this document does not cover training-script logging configuration.
+
+PaddleOCR has built a centralized logging system based on Python's [`logging` standard library](https://docs.python.org/3/library/logging.html#module-logging). In other words, PaddleOCR uses a single logger, which can be accessed and configured via `paddleocr.logger`.
+
+By default, the logging level in PaddleOCR is set to `ERROR`, meaning that log messages will only be output if their level is `ERROR` or higher (e.g., `CRITICAL`). PaddleOCR also configures a `StreamHandler` for this logger, which outputs logs to the standard error stream, and sets the logger's `propagate` attribute to `False` to prevent log messages from being passed to its parent logger.
+
+If you wish to disable PaddleOCR's automatic logging configuration behavior, you can set the environment variable `DISABLE_AUTO_LOGGING_CONFIG` to `1`. In this case, PaddleOCR will not perform any additional configuration of the logger.
+
+For more flexible customization of logging behavior, refer to the relevant documentation of the `logging` standard library. Below is an example of writing logs to a file:
+
+```python
+import logging
+from paddleocr import logger
+
+# Write logs to the file `paddleocr.log`
+fh = logging.FileHandler("paddleocr.log")
+logger.addHandler(fh)
+```
+
+Please note that other libraries that PaddleOCR depends on (such as [PaddleX](./paddleocr_and_paddlex.en.md)) have their own independent logging systems, and the above configuration will not affect the log output of these libraries.
diff --git a/docs/version3.x/logging.md b/docs/version3.x/logging.md
new file mode 100644
index 0000000..1dd5a29
--- /dev/null
+++ b/docs/version3.x/logging.md
@@ -0,0 +1,26 @@
+---
+comments: true
+---
+
+# 日志
+
+本文档主要介绍如何配置通过 PyPI 安装的 `paddleocr` 包的日志系统。需要注意的是，`paddleocr` 与训练脚本使用的是不同的日志系统，本文档不涉及训练脚本所使用的日志系统的配置方法。
+
+PaddleOCR 构建了一个基于 Python [`logging` 标准库](https://docs.python.org/zh-cn/3/library/logging.html#module-logging) 的集中式日志系统。换言之，PaddleOCR 使用唯一的日志记录器（logger），可通过 `paddleocr.logger` 访问和配置。
+
+默认情况下，PaddleOCR 的日志级别设为 `ERROR`，这意味着仅当日志级别为 `ERROR` 或更高（如 `CRITICAL`）时，日志信息才会输出。PaddleOCR 同时为该日志记录器配置了一个 `StreamHandler`，将日志输出到标准错误流，并将记录器的 `propagate` 属性设为 `False`，以避免日志信息传递到其父记录器。
+
+若希望禁止 PaddleOCR 对日志系统的自动配置行为，可将环境变量 `DISABLE_AUTO_LOGGING_CONFIG` 设为 `1`。此时，PaddleOCR 将不会对日志记录器进行任何额外配置。
+
+如需更灵活地定制日志行为，可参考 `logging` 标准库的相关文档。以下是一个将日志写入文件的示例：
+
+```python
+import logging
+from paddleocr import logger
+
+# 将日志写入文件 `paddleocr.log`
+fh = logging.FileHandler("paddleocr.log")
+logger.addHandler(fh)
+```
+
+请注意，PaddleOCR 依赖的其他库（如 [PaddleX](./paddleocr_and_paddlex.md)）拥有各自独立的日志系统，以上配置不会影响这些库的日志输出。
diff --git a/docs/version3.x/logo.jpg b/docs/version3.x/logo.jpg
new file mode 100644
index 0000000..0a05d68
Binary files /dev/null and b/docs/version3.x/logo.jpg differ
diff --git a/docs/version3.x/model_list.md b/docs/version3.x/model_list.md
new file mode 100644
index 0000000..a85910f
--- /dev/null
+++ b/docs/version3.x/model_list.md
@@ -0,0 +1,976 @@
+---
+comments: true
+---
+
+# PaddleOCR模型列表（CPU/GPU）
+
+PaddleOCR 内置了多条产线，每条产线都包含了若干模块，每个模块包含若干模型，具体使用哪些模型，您可以根据下边的 benchmark 数据来选择。如您更考虑模型精度，请选择精度较高的模型，如您更考虑模型推理速度，请选择推理速度较快的模型，如您更考虑模型存储大小，请选择存储大小较小的模型。
+
+## [文本检测模块](./module_usage/text_detection.md)
+
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>检测Hmean（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml 文件</th>
+<th>模型下载链接</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv6_medium_det</td>
+<td>-</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>-</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_detection/PP-OCRv6_medium_det.yaml">PP-OCRv6_medium_det.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_medium_det_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-OCRv6_small_det</td>
+<td>-</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>-</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_detection/PP-OCRv6_small_det.yaml">PP-OCRv6_small_det.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_det_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-OCRv6_tiny_det</td>
+<td>-</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>-</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_detection/PP-OCRv6_tiny_det.yaml">PP-OCRv6_tiny_det.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_det_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-OCRv5_server_det</td>
+<td>83.8</td>
+<td>89.55 / 70.19</td>
+<td>383.15 / 383.15</td>
+<td>101</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_detection/PP-OCRv5_server_det.yaml">PP-OCRv5_server_det.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_det_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_det</td>
+<td>79.0</td>
+<td>10.67 / 6.36</td>
+<td>57.77 / 28.15</td>
+<td>4.7</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_detection/PP-OCRv5_mobile_det.yaml">PP-OCRv5_mobile_det.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_det_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_det</td>
+<td>82.56</td>
+<td>127.82 / 98.87</td>
+<td>585.95 / 489.77</td>
+<td>109</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_detection/PP-OCRv4_server_det.yaml">PP-OCRv4_server_det.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_det_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_det</td>
+<td>63.8</td>
+<td>9.87 / 4.17</td>
+<td>56.60 / 20.79</td>
+<td>4.7</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_detection/PP-OCRv4_mobile_det.yaml">PP-OCRv4_mobile_det.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_det</td>
+<td>78.68</td>
+<td>9.90 / 3.60</td>
+<td>41.93 / 20.76</td>
+<td>2.1</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_detection/PP-OCRv3_mobile_det.yaml">PP-OCRv3_mobile_det.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-OCRv3_server_det</td>
+<td>80.11</td>
+<td>119.50 / 75.00</td>
+<td>379.35 / 318.35</td>
+<td>102.1</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_detection/PP-OCRv3_server_det.yaml">PP-OCRv3_server_det.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_server_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_server_det_pretrained.pdparams">训练模型</a></td>
+</tr>
+</tbody>
+</table>
+<b>注：以上精度指标的评估集是 PaddleOCR 自建的中英文数据集，覆盖街景、网图、文档、手写多个场景，其中文本识别包含 593 张图片。</b>
+
+## [印章文本检测模块](./module_usage/seal_text_detection.md)
+<table>
+<thead>
+<tr>
+<th>模型名称</th>
+<th>检测Hmean（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml 文件</th>
+<th>模型下载链接</th></tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv4_mobile_seal_det</td>
+<td>96.36</td>
+<td>9.70 / 3.56</td>
+<td>50.38 / 19.64</td>
+<td>4.7</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/seal_text_detection/PP-OCRv4_mobile_seal_det.yaml">PP-OCRv4_mobile_seal_det.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_seal_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_seal_det_pretrained.pdparams">训练模型</a></td></tr>
+<tr>
+<td>PP-OCRv4_server_seal_det</td>
+<td>98.40</td>
+<td>124.64 / 91.57</td>
+<td>545.68 / 439.86</td>
+<td>109</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/seal_text_detection/PP-OCRv4_server_seal_det.yaml">PP-OCRv4_server_seal_det.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_seal_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_seal_det_pretrained.pdparams">训练模型</a></td></tr>
+</tbody>
+</table>
+<b>注：以上精度指标的评估集是 PaddleOCR 自建的印章数据集，包含500印章图像。</b>
+
+## [文本识别模块](./module_usage/text_recognition.md)
+
+* <b>中文识别模型</b>
+<table>
+<tr>
+<th>模型</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml 文件</th>
+<th>模型下载链接</th>
+</tr>
+<tr>
+<td>PP-OCRv6_medium_rec</td>
+<td>-</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>-</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/PP-OCRv6_medium_rec.yaml">PP-OCRv6_medium_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_medium_rec_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-OCRv6_small_rec</td>
+<td>-</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>-</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/PP-OCRv6_small_rec.yaml">PP-OCRv6_small_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_rec_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-OCRv6_tiny_rec</td>
+<td>-</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>-</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/PP-OCRv6_tiny_rec.yaml">PP-OCRv6_tiny_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_rec_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td>86.38</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/PP-OCRv5_server_rec.yaml">PP-OCRv5_server_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td>81.29</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/PP-OCRv5_mobile_rec.yaml">PP-OCRv5_mobile_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/PP-OCRv4_server_rec_doc.yaml">PP-OCRv4_server_rec_doc.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/PP-OCRv4_mobile_rec.yaml">PP-OCRv4_mobile_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/PP-OCRv4_server_rec.yaml">PP-OCRv4_server_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_rec</td>
+<td>72.96</td>
+<td>3.89 / 1.16</td>
+<td>8.72 / 3.56</td>
+<td>10.3</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/PP-OCRv3_mobile_rec.yaml">PP-OCRv3_mobile_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+</table>
+<p><b>注：以上精度指标的评估集是 PaddleOCR 自建的中文数据集，覆盖街景、网图、文档、手写多个场景，其中文本识别包含 8367 张图片。</b></p>
+<table>
+<tr>
+<th>模型</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml 文件</th>
+<th>模型下载链接</th>
+</tr>
+<tr>
+<td>ch_SVTRv2_rec</td>
+<td>68.81</td>
+<td>10.38 / 8.31</td>
+<td>66.52 / 30.83</td>
+<td>80.5</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/ch_SVTRv2_rec.yaml">ch_SVTRv2_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_SVTRv2_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_SVTRv2_rec_pretrained.pdparams">训练模型</a></td>
+</tr>
+</table>
+<p><b>注：以上精度指标的评估集是 <a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务</a>A榜。 </b></p>
+<table>
+<tr>
+<th>模型</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml 文件</th>
+<th>模型下载链接</th>
+</tr>
+<tr>
+<td>ch_RepSVTR_rec</td>
+<td>65.07</td>
+<td>6.29 / 1.57</td>
+<td>20.64 / 5.40</td>
+<td>48.8</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/ch_RepSVTR_rec.yaml">ch_RepSVTR_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_RepSVTR_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_RepSVTR_rec_pretrained.pdparams">训练模型</a></td>
+</tr>
+</table>
+<p><b>注：以上精度指标的评估集是 <a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务</a>B榜。 </b></p>
+
+* <b>英文识别模型</b>
+<table>
+<tr>
+<th>模型</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml 文件</th>
+<th>模型下载链接</th>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td> 70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/en_PP-OCRv4_mobile_rec.yaml">en_PP-OCRv4_mobile_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+<tr>
+<td>en_PP-OCRv3_mobile_rec</td>
+<td>70.69</td>
+<td>3.56 / 0.78</td>
+<td>8.44 / 5.78</td>
+<td>17.3</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/en_PP-OCRv3_mobile_rec.yaml">en_PP-OCRv3_mobile_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+</table>
+
+<p><b>注：以上精度指标的评估集是 PaddleOCR 自建的英文数据集。 </b></p>
+
+* <b>多语言识别模型</b>
+<table>
+<tr>
+<th>模型</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml 文件</th>
+<th>模型下载链接</th>
+</tr>
+<tr>
+<td>korean_PP-OCRv3_mobile_rec</td>
+<td>60.21</td>
+<td>3.73 / 0.98</td>
+<td>8.76 / 2.91</td>
+<td>9.6</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/korean_PP-OCRv3_mobile_rec.yaml">korean_PP-OCRv3_mobile_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+<tr>
+<td>japan_PP-OCRv3_mobile_rec</td>
+<td>45.69</td>
+<td>3.86 / 1.01</td>
+<td>8.62 / 2.92</td>
+<td>9.8</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/japan_PP-OCRv3_mobile_rec.yaml">japan_PP-OCRv3_mobile_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+japan_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+<tr>
+<td>chinese_cht_PP-OCRv3_mobile_rec</td>
+<td>82.06</td>
+<td>3.90 / 1.16</td>
+<td>9.24 / 3.18</td>
+<td>10.8</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/chinese_cht_PP-OCRv3_mobile_rec.yaml">chinese_cht_PP-OCRv3_mobile_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+chinese_cht_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+<tr>
+<td>te_PP-OCRv3_mobile_rec</td>
+<td>95.88</td>
+<td>3.59 / 0.81</td>
+<td>8.28 / 6.21</td>
+<td>8.7</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/te_PP-OCRv3_mobile_rec.yaml">te_PP-OCRv3_mobile_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+te_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+<tr>
+<td>ka_PP-OCRv3_mobile_rec</td>
+<td>96.96</td>
+<td>3.49 / 0.89</td>
+<td>8.63 / 2.77</td>
+<td>17.4</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/ka_PP-OCRv3_mobile_rec.yaml">ka_PP-OCRv3_mobile_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ka_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+<tr>
+<td>ta_PP-OCRv3_mobile_rec</td>
+<td>76.83</td>
+<td>3.49 / 0.86</td>
+<td>8.35 / 3.41</td>
+<td>8.7</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/ta_PP-OCRv3_mobile_rec.yaml">ta_PP-OCRv3_mobile_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ta_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+<tr>
+<td>latin_PP-OCRv3_mobile_rec</td>
+<td>76.93</td>
+<td>3.53 / 0.78</td>
+<td>8.50 / 6.83</td>
+<td>8.7</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/latin_PP-OCRv3_mobile_rec.yaml">latin_PP-OCRv3_mobile_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv3_mobile_rec</td>
+<td>73.55</td>
+<td>3.60 / 0.83</td>
+<td>8.44 / 4.69</td>
+<td>17.3</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/arabic_PP-OCRv3_mobile_rec.yaml">arabic_PP-OCRv3_mobile_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+arabic_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv3_mobile_rec</td>
+<td>94.28</td>
+<td>3.56 / 0.79</td>
+<td>8.22 / 2.76</td>
+<td>8.7</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/cyrillic_PP-OCRv3_mobile_rec.yaml">cyrillic_PP-OCRv3_mobile_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+cyrillic_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv3_mobile_rec</td>
+<td>96.44</td>
+<td>3.60 / 0.78</td>
+<td>6.95 / 2.87</td>
+<td>8.7</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/text_recognition/devanagari_PP-OCRv3_mobile_rec.yaml">devanagari_PP-OCRv3_mobile_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="">训练模型</a></td>
+</tr>
+</table>
+<p><b>注：以上精度指标的评估集是 PaddleOCR 自建的多语种数据集。</b></p>
+
+## [公式识别模块](./module_usage/formula_recognition.md)
+
+<table>
+<tr>
+<th>模型</th>
+<th>En-BLEU(%)</th>
+<th>Zh-BLEU(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml 文件</th>
+<th>模型下载链接</th>
+</tr>
+<tr>
+<td>UniMERNet</td>
+<td>85.91</td>
+<td>43.50</td>
+<td>1311.84 / 1311.84</td>
+<td>- / 8288.07</td>
+<td>1530</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/formula_recognition/UniMERNet.yaml">UniMERNet.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UniMERNet_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UniMERNet_pretrained.pdparams">训练模型</a></td>
+</tr>
+<td>PP-FormulaNet-S</td>
+<td>87.00</td>
+<td>45.71</td>
+<td>182.25 / 182.25</td>
+<td>- / 254.39</td>
+<td>224</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/formula_recognition/PP-FormulaNet-S.yaml">PP-FormulaNet-S.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-S_pretrained.pdparams">训练模型</a></td>
+</tr>
+<td>PP-FormulaNet-L</td>
+<td>90.36</td>
+<td>45.78</td>
+<td>1482.03 / 1482.03</td>
+<td>- / 3131.54</td>
+<td>695</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/formula_recognition/PP-FormulaNet-L.yaml">PP-FormulaNet-L.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-L_pretrained.pdparams">训练模型</a></td>
+</tr>
+<td>PP-FormulaNet_plus-S</td>
+<td>88.71</td>
+<td>53.32</td>
+<td>179.20 / 179.20</td>
+<td>- / 260.99</td>
+<td>248</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/formula_recognition/PP-FormulaNet_plus-S.yaml">PP-FormulaNet_plus-S.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-S_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-M</td>
+<td>91.45</td>
+<td>89.76</td>
+<td>1040.27 / 1040.27</td>
+<td>- / 1615.80</td>
+<td>592</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/formula_recognition/PP-FormulaNet_plus-M.yaml">PP-FormulaNet_plus-M.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-M_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-M_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-L</td>
+<td>92.22</td>
+<td>90.64</td>
+<td>1476.07 / 1476.07</td>
+<td>- / 3125.58</td>
+<td>698</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/formula_recognition/PP-FormulaNet_plus-L.yaml">PP-FormulaNet_plus-L.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-L_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>LaTeX_OCR_rec</td>
+<td>74.55</td>
+<td>39.96</td>
+<td>1088.89 / 1088.89</td>
+<td>- / -</td>
+<td>99</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/formula_recognition/LaTeX_OCR_rec.yaml">LaTeX_OCR_rec.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/LaTeX_OCR_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/LaTeX_OCR_rec_pretrained.pdparams">训练模型</a></td>
+</tr>
+</table>
+<b>注：以上精度指标测量自 PaddleX 内部自建公式识别测试集。LaTeX_OCR_rec在LaTeX-OCR公式识别测试集的BLEU score为 0.8821。</b>
+
+## [表格结构识别模块](./module_usage/table_structure_recognition.md)
+
+<table>
+<tr>
+<th>模型</th>
+<th>精度（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml 文件</th>
+<th>模型下载链接</th>
+</tr>
+<tr>
+<td>SLANet</td>
+<td>59.52</td>
+<td>23.96 / 21.75</td>
+<td>- / 43.12</td>
+<td>6.9</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/table_structure_recognition/SLANet.yaml">SLANet.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANet_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>SLANet_plus</td>
+<td>63.69</td>
+<td>23.43 / 22.16</td>
+<td>- / 41.80</td>
+<td>6.9</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/table_structure_recognition/SLANet_plus.yaml">SLANet_plus.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANet_plus_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_plus_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>SLANeXt_wired</td>
+<td rowspan="2">69.65</td>
+<td rowspan="2">85.92 / 85.92</td>
+<td rowspan="2">- / 501.66</td>
+<td rowspan="2">351</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/table_structure_recognition/SLANeXt_wired.yaml">SLANeXt_wired.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wired_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wired_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>SLANeXt_wireless</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/table_structure_recognition/SLANeXt_wireless.yaml">SLANeXt_wireless.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wireless_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wireless_pretrained.pdparams">训练模型</a></td>
+</tr>
+</table>
+<b>注：以上精度指标测量自 PaddleX 内部自建高难度中文表格识别数据集。</b>
+
+
+## [表格单元格检测模块](./module_usage/table_cells_detection.md)
+
+<table>
+<tr>
+<th>模型</th>
+<th>mAP(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml 文件</th>
+<th>模型下载链接</th>
+</tr>
+<tr>
+<td>RT-DETR-L_wired_table_cell_det</td>
+<td rowspan="2">82.7</td>
+<td rowspan="2">33.47 / 27.02</td>
+<td rowspan="2">402.55 / 256.56</td>
+<td rowspan="2">124</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/table_cells_detection/RT-DETR-L_wired_table_cell_det.yaml">RT-DETR-L_wired_table_cell_det.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wired_table_cell_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wired_table_cell_det_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>RT-DETR-L_wireless_table_cell_det</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/table_cells_detection/RT-DETR-L_wireless_table_cell_det.yaml">RT-DETR-L_wireless_table_cell_det.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wireless_table_cell_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wireless_table_cell_det_pretrained.pdparams">训练模型</a></td>
+</tr>
+</table>
+<p><b>注：以上精度指标测量自 PaddleX 内部自建表格单元格检测数据集。</b></p>
+
+## [表格分类模块](./module_usage/table_classification.md)
+
+<table>
+<tr>
+<th>模型</th>
+<th>Top1 Acc(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml文件</th>
+<th>模型下载链接</th>
+</tr>
+<tr>
+<td>PP-LCNet_x1_0_table_cls</td>
+<td>94.2</td>
+<td>2.62 / 0.60</td>
+<td>3.17 / 1.14</td>
+<td>6.6</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/table_classification/PP-LCNet_x1_0_table_cls.yaml">PP-LCNet_x1_0_table_cls.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_table_cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_table_cls_pretrained.pdparams">训练模型</a></td>
+</tr>
+</table>
+<p><b>注：以上精度指标测量自 PaddleX 内部自建表格分类数据集。</b></p>
+
+## [文本图像矫正模块](./module_usage/text_image_unwarping.md)
+<table>
+<thead>
+<tr>
+<th>模型名称</th>
+<th>CER</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml 文件</th>
+<th>模型下载链接</th></tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td>0.179</td>
+<td>19.05 / 19.05</td>
+<td>- / 869.82</td>
+<td>30.3</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/image_unwarping/UVDoc.yaml">UVDoc.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">训练模型</a></td></tr>
+</tbody>
+</table>
+<b>注：以上精度指标测量自 </b><b>PaddleX自建的图像矫正数据集</b><b>。</b>
+
+## [版面区域检测模块](./module_usage/layout_detection.md)
+
+* <b>版面检测模型，包含20个常见的类别：文档标题、段落标题、文本、页码、摘要、目录、参考文献、脚注、页眉、页脚、算法、公式、公式编号、图像、表格、图和表标题（图标题、表格标题和图表标题）、印章、图表、侧栏文本和参考文献内容</b>
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml文件</th>
+<th>模型下载链接</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout_plus-L</td>
+<td>83.2</td>
+<td>53.03 / 17.23</td>
+<td>634.62 / 378.32</td>
+<td>126.01 </td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/layout_detection/PP-DocLayout_plus-L.yaml">PP-DocLayout_plus-L.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout_plus-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout_plus-L_pretrained.pdparams">训练模型</a></td>
+</tr>
+</tbody>
+</table>
+
+<b>注：以上精度指标的评估集是自建的版面区域检测数据集，包含中英文论文、杂志、报纸、研报、PPT、试卷、课本等 1300 张文档类型图片。</b>
+
+* <b>文档图像版面子模块检测，包含1个 版面区域 类别，能检测多栏的报纸、杂志的每个子文章的文本区域:</b>
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml文件</th>
+<th>模型下载链接</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocBlockLayout</td>
+<td>95.9</td>
+<td>34.60 / 28.54</td>
+<td>506.43 / 256.83</td>
+<td>123.92</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/layout_detection/PP-DocBlockLayout.yaml">PP-DocBlockLayout.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBlockLayout_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocBlockLayout_pretrained.pdparams">训练模型</a></td>
+</tr>
+</tbody>
+</table>
+
+<b>注：以上精度指标的评估集是自建的版面子区域检测数据集，包含中英文论文、杂志、报纸、研报、PPT、试卷、课本等 1000 张文档类型图片。</b>
+
+
+* <b>版面检测模型，包含23个常见的类别：文档标题、段落标题、文本、页码、摘要、目录、参考文献、脚注、页眉、页脚、算法、公式、公式编号、图像、图表标题、表格、表格标题、印章、图表标题、图表、页眉图像、页脚图像、侧栏文本</b>
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml文件</th>
+<th>模型下载链接</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout-L</td>
+<td>90.4</td>
+<td>33.59 / 33.59</td>
+<td>503.01 / 251.08</td>
+<td>123.76 </td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/layout_detection/PP-DocLayout-L.yaml">PP-DocLayout-L.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-L_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-DocLayout-M</td>
+<td>75.2</td>
+<td>13.03 / 4.72</td>
+<td>43.39 / 24.44</td>
+<td>22.578</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/layout_detection/PP-DocLayout-M.yaml">PP-DocLayout-M.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-M_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-M_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>PP-DocLayout-S</td>
+<td>70.9</td>
+<td>11.54 / 3.86</td>
+<td>18.53 / 6.29</td>
+<td>4.834</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/layout_detection/PP-DocLayout-S.yaml">PP-DocLayout-S.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-S_pretrained.pdparams">训练模型</a></td>
+</tr>
+</tbody>
+</table>
+
+<b>注：以上精度指标的评估集是自建的版面区域检测数据集，包含中英文论文、杂志和研报等常见的 500 张文档类型图片。</b>
+
+* <b>表格版面检测模型</b>
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml文件</th>
+<th>模型下载链接</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet_layout_1x_table</td>
+<td>97.5</td>
+<td>9.57 / 6.63</td>
+<td>27.66 / 16.75</td>
+<td>7.4</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/layout_detection/PicoDet_layout_1x_table.yaml">PicoDet_layout_1x_table.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_table_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_table_pretrained.pdparams">训练模型</a></td>
+</tr>
+</tbody></table>
+<b>注：以上精度指标的评估集是 PaddleOCR 自建的版面表格区域检测数据集，包含中英文 7835 张带有表格的论文文档类型图片。</b>
+
+* <b>3类版面检测模型，包含表格、图像、印章</b>
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml文件</th>
+<th>模型下载链接</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_3cls</td>
+<td>88.2</td>
+<td>8.43 / 3.44</td>
+<td>17.60 / 6.51</td>
+<td>4.8</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/layout_detection/PicoDet-S_layout_3cls.yaml">PicoDet-S_layout_3cls.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_3cls_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_3cls</td>
+<td>89.0</td>
+<td>12.80 / 9.57</td>
+<td>45.04 / 23.86</td>
+<td>22.6</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/layout_detection/PicoDet-L_layout_3cls.yaml">PicoDet-L_layout_3cls.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_3cls_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_3cls</td>
+<td>95.8</td>
+<td>114.80 / 25.65</td>
+<td>924.38 / 924.38</td>
+<td>470.1</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/layout_detection/RT-DETR-H_layout_3cls.yaml">RT-DETR-H_layout_3cls.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_3cls_pretrained.pdparams">训练模型</a></td>
+</tr>
+</tbody></table>
+<b>注：以上精度指标的评估集是 PaddleOCR 自建的版面区域检测数据集，包含中英文论文、杂志和研报等常见的 1154 张文档类型图片。</b>
+
+* <b>5类英文文档区域检测模型，包含文字、标题、表格、图片以及列表</b>
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml文件</th>
+<th>模型下载链接</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet_layout_1x</td>
+<td>97.8</td>
+<td>9.62 / 6.75</td>
+<td>26.96 / 12.77</td>
+<td>7.4</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/layout_detection/PicoDet_layout_1x.yaml">PicoDet_layout_1x.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_pretrained.pdparams">训练模型</a></td>
+</tr>
+</tbody></table>
+<b>注：以上精度指标的评估集是 [PubLayNet](https://developer.ibm.com/exchanges/data/all/publaynet/) 的评估数据集，包含英文文档的 11245 张图片。</b>
+
+* <b>17类区域检测模型，包含17个版面常见类别，分别是：段落标题、图片、文本、数字、摘要、内容、图表标题、公式、表格、表格标题、参考文献、文档标题、脚注、页眉、算法、页脚、印章</b>
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml文件</th>
+<th>模型下载链接</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_17cls</td>
+<td>87.4</td>
+<td>8.80 / 3.62</td>
+<td>17.51 / 6.35</td>
+<td>4.8</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/layout_detection/PicoDet-S_layout_17cls.yaml">PicoDet-S_layout_17cls.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_17cls_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_17cls</td>
+<td>89.0</td>
+<td>12.60 / 10.27</td>
+<td>43.70 / 24.42</td>
+<td>22.6</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/layout_detection/PicoDet-L_layout_17cls.yaml">PicoDet-L_layout_17cls.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_17cls_pretrained.pdparams">训练模型</a></td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_17cls</td>
+<td>98.3</td>
+<td>115.29 / 101.18</td>
+<td>964.75 / 964.75</td>
+<td>470.2</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/layout_detection/RT-DETR-H_layout_17cls.yaml">RT-DETR-H_layout_17cls.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_17cls_pretrained.pdparams">训练模型</a></td>
+</tr>
+</tbody>
+</table>
+<b>注：以上精度指标的评估集是 PaddleOCR 自建的版面区域检测数据集，包含中英文论文、杂志和研报等常见的 892 张文档类型图片。</b>
+
+## [文档图像方向分类模块](./module_usage/doc_img_orientation_classification.md)
+
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>Top-1 Acc（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml文件</th>
+<th>模型下载链接</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/doc_text_orientation/PP-LCNet_x1_0_doc_ori.yaml">PP-LCNet_x1_0_doc_ori.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">训练模型</a></td>
+</tr>
+</tbody>
+</table>
+<b>注：以上精度指标的评估集是自建的数据集，覆盖证件和文档等多个场景，包含 1000 张图片。</b>
+
+
+## [文本行方向分类模块](./module_usage/doc_img_orientation_classification.md)
+
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>Top-1 Acc（%）</th>
+<th>GPU推理耗时（ms）<br>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>yaml文件</th>
+<th>模型下载链接</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/textline_orientation/PP-LCNet_x0_25_textline_ori.yaml">PP-LCNet_x0_25_textline_ori.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">训练模型</a></td>
+</tr>
+</tbody>
+</table>
+
+<b>注：以上精度指标的评估集是自建的数据集，覆盖证件和文档等多个场景，包含 1000 张图片。</b>
+
+## [文档类视觉语言模型模块](./module_usage/doc_vlm.md)
+
+<table>
+<tr>
+<th>模型</th>
+<th>模型参数尺寸（B）</th>
+<th>模型存储大小（GB）</th>
+<th>yaml文件</th>
+<th>模型下载链接</th>
+</tr>
+<tr>
+<td>PP-DocBee-2B</td>
+<td>2</td>
+<td>4.2</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/doc_vlm/PP-DocBee-2B.yaml">PP-DocBee-2B.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBee-2B_infer.tar">推理模型</a></td>
+</tr>
+<tr>
+<td>PP-DocBee-7B</td>
+<td>7</td>
+<td>15.8</td>
+<td><a href="https://github.com/PaddlePaddle/PaddleX/blob/develop/paddlex/configs/modules/doc_vlm/PP-DocBee-7B.yaml">PP-DocBee-7B.yaml</a></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBee-7B_infer.tar">推理模型</a></td>
+</tr>
+<tr>
+<td>PP-DocBee2-3B</td>
+<td>3</td>
+<td>7.6</td>
+<td></td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBee2-3B_infer.tar">推理模型</a></td>
+</tr>
+</table>
diff --git a/docs/version3.x/module_usage/chart_parsing.en.md b/docs/version3.x/module_usage/chart_parsing.en.md
new file mode 100644
index 0000000..76412e8
--- /dev/null
+++ b/docs/version3.x/module_usage/chart_parsing.en.md
@@ -0,0 +1,358 @@
+---
+comments: true
+---
+
+# Chart Parsing Module Tutorial
+
+## 1. Overview
+
+Multimodal chart parsing is a cutting-edge OCR technology that focuses on automatically converting various types of visual charts (such as bar charts, line charts, pie charts, etc.) into structured data tables with formatted output. Traditional methods rely on complex pipeline designs with chart keypoint detection models, which involve many prior assumptions and tend to lack robustness. The models in this module leverage the latest VLM (Vision-Language Model) techniques and are data-driven, learning robust features from vast real-world datasets. Application scenarios include financial analysis, academic research, business reporting, and more—for instance, quickly extracting growth trend data from financial reports, experimental comparison figures from research papers, or user distribution statistics from market surveys—empowering users to transition from “viewing charts” to “using data”.
+
+## 2. Supported Model List
+
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Model Size (B)</th>
+<th>Storage Size (GB)</th>
+<th>Score</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>PP-Chart2Table</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-Chart2Table_infer.tar">Inference Model</a></td>
+<td>0.58</td>
+<td>1.4</td>
+<th>80.60</th>
+<td>PP-Chart2Table is a multimodal chart parsing model developed by the PaddlePaddle team. It demonstrates exceptional performance on both Chinese and English chart parsing tasks. The team designed a specialized “Shuffled Chart Data Retrieval” training task and adopted a carefully designed token masking strategy, significantly improving performance on chart-to-table conversion. Additionally, the team enhanced the model with a high-quality data synthesis process using seed data, RAG, and LLM persona-driven generation to diversify training data. To handle large amounts of out-of-distribution (OOD) unlabeled data, a two-stage large model distillation process was used to ensure excellent adaptability and generalization to diverse real-world data. In internal Chinese-English use case evaluations, PP-Chart2Table achieved state-of-the-art performance among models of similar size and reached accuracy comparable to 7B-parameter VLMs in key scenarios.</td>
+</tr>
+</table>
+
+**Note:** The scores above are based on internal evaluation on a test set of 1801 samples, covering various chart types (bar, line, pie, etc.) across scenarios such as financial reports, regulations, and contracts. There is currently no plan for public release.
+
+> ❗ **Note:** The PP-Chart2Table model was upgraded on June 27, 2025. To use the previous version, please download it [here](https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-Chart2Table_infer.bak.tar)
+
+## 3. Quick Start
+
+> ❗ Before getting started, please install the PaddleOCR wheel package. Refer to the [Installation Guide](../installation.md) for details.
+
+Run the following command to get started instantly:
+
+```bash
+paddleocr chart_parsing -i "{'image': 'https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/chart_parsing_02.png'}"
+```
+
+The example above uses the <code>paddle_dynamic</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr chart_parsing -i "{'image': 'https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/chart_parsing_02.png'}" \
+    --engine transformers
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following command:
+
+```bash
+# Use the onnxruntime engine for inference
+paddleocr chart_parsing -i "{'image': 'https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/chart_parsing_02.png'}" \
+    --engine onnxruntime
+```
+
+In most scenarios, the default `paddle_dynamic` inference engine delivers better inference performance and is the recommended first choice.
+
+**Note:** By default, PaddleOCR retrieves models from HuggingFace. If HuggingFace access is restricted in your environment, you can switch the model source to BOS by setting the environment variable: `PADDLE_PDX_MODEL_SOURCE="BOS"`. Support for more mainstream sources is planned.
+
+You can also integrate the inference of the vision-language model into your own project. Please download the [example image](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/chart_parsing_02.png) locally before running the following code:
+
+```python
+from paddleocr import ChartParsing
+model = ChartParsing(model_name="PP-Chart2Table")
+results = model.predict(
+    input={"image": "chart_parsing_02.png"},
+    batch_size=1
+)
+for res in results:
+    res.print()
+    res.save_to_json(f"./output/res.json")
+```
+
+The example above uses the <code>paddle_dynamic</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following code:
+
+```python
+from paddleocr import ChartParsing
+model = ChartParsing(
+    model_name="PP-Chart2Table",
+    engine="transformers",
+)
+results = model.predict(
+    input={"image": "chart_parsing_02.png"},
+    batch_size=1
+)
+for res in results:
+    res.print()
+    res.save_to_json(f"./output/res.json")
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following code:
+
+```python
+from paddleocr import ChartParsing
+model = ChartParsing(
+    model_name="PP-Chart2Table",
+    engine="onnxruntime",
+)
+results = model.predict(
+    input={"image": "chart_parsing_02.png"},
+    batch_size=1
+)
+for res in results:
+    res.print()
+    res.save_to_json(f"./output/res.json")
+```
+
+In most scenarios, the default `paddle_dynamic` inference engine delivers better inference performance and is the recommended first choice.
+
+The output result will be:
+
+```bash
+{'res': {'image': 'chart_parsing_02.png', 'result': 'Year | Avg Revenue per 5-star Hotel (Million CNY) | Avg Profit per 5-star Hotel (Million CNY)\n2018 | 104.22 | 9.87\n2019 | 99.11 | 7.47\n2020 | 57.87 | -3.87\n2021 | 68.99 | -2.9\n2022 | 56.29 | -9.48\n2023 | 87.99 | 5.96'}}
+```
+
+Explanation of output parameters:
+<ul>
+<li><code>image</code>: The path to the input image</li>
+<li><code>result</code>:  The model's prediction output</li>
+</ul>
+
+The visualized result is:
+
+```bash
+Year | Avg Revenue per 5-star Hotel (Million CNY) | Avg Profit per 5-star Hotel (Million CNY)
+2018 | 104.22 | 9.87
+2019 | 99.11 | 7.47
+2020 | 57.87 | -3.87
+2021 | 68.99 | -2.9
+2022 | 56.29 | -9.48
+2023 | 87.99 | 5.96
+```
+
+Detailed explanation of related methods and parameters:
+
+* Instantiate a vision-language model with <code>ChartParsing</code>. Parameters:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>Meaning:</b> Model name.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, defaults to <code>PP-Chart2Table</code>.</td>
+<td><code>str | None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>Meaning</b>Model storage path.</td>
+<td><code>str | None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b> Inference device.<br/>
+<b>Description:</b>
+<b>Examples:</b> <code>"cpu"</code>, <code>"gpu"</code>, <code>"npu"</code>, <code>"gpu:0"</code><br/>
+Defaults to GPU 0 if available; otherwise falls back to CPU.
+</td>
+<td><code>str | None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_dynamic</code>, <code>transformers</code>, and <code>onnxruntime</code>. When left as <code>None</code>, local inference uses the <code>paddle_dynamic</code> engine by default. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+* Use the model's <code>predict()</code>  method for inference. This returns a list of results. The module also offers a <code>predict_iter()</code> method, which behaves identically in terms of inputs and outputs but returns a generator—ideal for large datasets or memory-sensitive scenarios. Choose based on your needs.
+
+<code>predict()</code> method parameters:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b> Input data (required). <br/>
+<b>Description:</b>
+Input formats vary by model.<br/>
+<ul>
+<li>For PP-Chart2Table: <code>{'image': image_path}</code></li>
+</ul>
+</td>
+<td><code>dict</code></td>
+<td>N/A</td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>Meaning:</b> Batch size. <br/>
+<b>Description:</b>
+Any positive integer.</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+
+* Prediction results are returned as <code>Result</code> objects for each sample, with support for printing and saving to JSON:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Description</th>
+<th>Parameter</th>
+<th>Type</th>
+<th>Explanation</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">Print results to terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Format output using JSON indentation</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Indentation level for pretty-printed JSON. Only works when <code>format_json=True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Whether to escape non-ASCII characters to Unicode. If <code>False</code>, keeps characters as-is.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">Save results to JSON file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>File path to save. If a directory, file will use input name as filename.</td>
+<td>N/A</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Same as in <code>print()</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Same as in <code>print()</code></td>
+<td><code>False</code></td>
+</tr>
+</table>
+
+* You can also access the result via properties:
+
+<table>
+<thead>
+<tr>
+<th>Property</th>
+<th>Description</th>
+</tr>
+</thead>
+<tr>
+<td><code>json</code></td>
+<td>Returns the result in JSON format</td>
+</tr>
+</table>
+
+## 4. Custom Development
+
+Currently, this module supports inference only and does not yet support fine-tuning. Fine-tuning capabilities are planned for future releases.
+
+## 5. Inference Engine
+
+For detailed descriptions, values, compatibility rules, and examples of the inference engine, please refer to <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration Description</a>.
+
+### 5.1 Speed Data
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="2">PP-Chart2Table</td>
+            <td>paddle_dynamic</td>
+            <td>53.00</td>
+            <td>17863.95</td>
+            <td>0.30</td>
+            <td>17917.78</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>23.95</td>
+            <td>12217.37</td>
+            <td>0.47</td>
+            <td>12269.98</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>Test Environment Description:</strong>
+<ul>
+    <li><strong>Test Data:</strong> <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/medal_table.jpg">Sample Image</a></li>
+    <li><strong>Hardware Configuration:</strong>
+        <ul>
+            <li>GPU: NVIDIA A100 40G</li>
+            <li>CPU: Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>Software Environment:</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+## 6. FAQ
diff --git a/docs/version3.x/module_usage/chart_parsing.md b/docs/version3.x/module_usage/chart_parsing.md
new file mode 100644
index 0000000..689ebe8
--- /dev/null
+++ b/docs/version3.x/module_usage/chart_parsing.md
@@ -0,0 +1,354 @@
+---
+comments: true
+---
+
+# 图表解析模块使用教程
+
+## 一、概述
+
+多模态图表解析是一项OCR领域的前沿技术，专注于将各类可视化图表（如柱状图、折线图、饼图等）自动转化为底层数据表，并进行格式化输出。传统方法依赖于图表关键点检测等模型进行复杂串联编排，先验假设较多，鲁棒性较差，该模块中的模型使用最新的VLM技术，数据驱动，从海量的现实数据中学习鲁棒的特征。其应用场景覆盖金融分析、学术研究、商业报告等场景——例如快速提取财报中的增长趋势数据、科研论文中的实验对比数值，或市场调研中的用户分布统计，助力用户从“看图”转向“用数”。
+
+## 二、支持模型列表
+
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>模型参数规模（B）</th>
+<th>模型存储大小（GB）</th>
+<th>模型分数 </th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-Chart2Table</td><td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-Chart2Table_infer.tar">推理模型</a></td>
+<td>0.58</td>
+<td>1.4</td>
+<th>80.60</th>
+<td>PP-Chart2Table是飞桨团队自研的一款专注于图表解析的多模态模型，在中英文图表解析任务中展现出卓越性能。团队专为图表解析设计了Shuffled Chart Data Retrieval训练任务，并结合精心设计的令牌掩码策略，显著提升其在图表转数据表任务上的性能。此外，团队通过精心设计的数据合成流程增强了PP-Chart2Table的能力，该流程利用高质量的种子数据，并结合RAG和大语言模型人格设计，以生成更丰富多样化的数据。为了处理大量未标记的分布外 (OOD) 数据，团队采用了两阶段大模型蒸馏训练过程，确保模型在广泛的真实世界数据集中具有出色的适应性和泛化能力。在内部业务的中英文场景测试中，PP-Chart2Table不仅达到同参数量级模型中的SOTA水平，更在关键场景中实现了与7B参数量级VLM模型相当的精度。</td>
+</tr>
+</table>
+
+<b>注：以上模型分数为内部评估集模型测试结果，共1801条数据，包括了各个场景（财报、法律法规、合同等）下的各种图表类型（柱状图、折线图、饼图等）的测试样本，暂时未有计划公开。</b>
+
+> ❗ <b>注</b>：PP-Chart2Table模型于 2025.6.27 升级，如需使用升级前的模型权重，请点击<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-Chart2Table_infer.bak.tar">下载链接</a>
+
+## 三、快速开始
+
+> ❗ 在快速开始前，请先安装 PaddleOCR 的 wheel 包，详细请参考 [安装教程](../installation.md)。
+
+使用一行命令即可快速体验：
+
+```bash
+paddleocr chart_parsing -i "{'image': 'https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/chart_parsing_02.png'}"
+```
+
+上述示例默认使用 <code>paddle_dynamic</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下命令：
+
+```bash
+# 使用 transformers 引擎进行推理
+paddleocr chart_parsing -i "{'image': 'https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/chart_parsing_02.png'}" \
+    --engine transformers
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下命令：
+
+```bash
+# 使用 onnxruntime 引擎进行推理
+paddleocr chart_parsing -i "{'image': 'https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/chart_parsing_02.png'}" \
+    --engine onnxruntime
+```
+
+在大多数场景下，默认的 `paddle_dynamic` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+<b>注：</b>PaddleOCR 官方模型默认从 HuggingFace 获取，如运行环境访问 HuggingFace 不便，可通过环境变量修改模型源为 BOS：`PADDLE_PDX_MODEL_SOURCE="BOS"`，未来将支持更多主流模型源；
+
+您也可以将开放文档类视觉语言模型模块中的模型推理集成到您的项目中。运行以下代码前，请您下载[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/chart_parsing_02.png)到本地。
+
+```python
+from paddleocr import ChartParsing
+model = ChartParsing(model_name="PP-Chart2Table")
+results = model.predict(
+    input={"image": "chart_parsing_02.png"},
+    batch_size=1
+)
+for res in results:
+    res.print()
+    res.save_to_json(f"./output/res.json")
+```
+
+上述示例默认使用 <code>paddle_dynamic</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下代码：
+
+```python
+from paddleocr import ChartParsing
+model = ChartParsing(
+    model_name="PP-Chart2Table",
+    engine="transformers",
+)
+results = model.predict(
+    input={"image": "chart_parsing_02.png"},
+    batch_size=1
+)
+for res in results:
+    res.print()
+    res.save_to_json(f"./output/res.json")
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下代码：
+
+```python
+from paddleocr import ChartParsing
+model = ChartParsing(
+    model_name="PP-Chart2Table",
+    engine="onnxruntime",
+)
+results = model.predict(
+    input={"image": "chart_parsing_02.png"},
+    batch_size=1
+)
+for res in results:
+    res.print()
+    res.save_to_json(f"./output/res.json")
+```
+
+在大多数场景下，默认的 `paddle_dynamic` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+运行后，得到的结果为：
+
+```bash
+{'res': {'image': 'chart_parsing_02.png', 'result': '年份 | 单家五星级旅游饭店年平均营收 (百万元) | 单家五星级旅游饭店年平均利润 (百万元)\n2018 | 104.22 | 9.87\n2019 | 99.11 | 7.47\n2020 | 57.87 | -3.87\n2021 | 68.99 | -2.9\n2022 | 56.29 | -9.48\n2023 | 87.99 | 5.96'}}
+```
+
+运行结果参数含义如下：
+<ul>
+<li><code>image</code>: 表示输入待预测图像的路径</li>
+<li><code>result</code>: 模型预测的结果信息</li>
+</ul>
+预测结果打印可视化如下：
+
+```bash
+年份 | 单家五星级旅游饭店年平均营收 (百万元) | 单家五星级旅游饭店年平均利润 (百万元)
+2018 | 104.22 | 9.87
+2019 | 99.11 | 7.47
+2020 | 57.87 | -3.87
+2021 | 68.99 | -2.9
+2022 | 56.29 | -9.48
+2023 | 87.99 | 5.96
+```
+
+相关方法、参数等说明如下：
+
+* <code>ChartParsing</code>实例化文档类视觉语言模型，具体说明如下：
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>含义：</b>模型名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，则使用<code>PP-Chart2Table</code>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>含义：</b>模型存储路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+<b>例如：</b><code>"cpu"</code>、<code>"gpu"</code>、<code>"npu"</code>、<code>"gpu:0"</code></code>。
+默认情况下，优先使用 GPU 0；若不可用则使用 CPU。
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_dynamic</code>、<code>transformers</code>、<code>onnxruntime</code>。保持为默认值 <code>None</code> 时，本地推理默认使用 <code>paddle_dynamic</code> 引擎。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+* 调用图表解析模型的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。另外，本模块还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。<code>predict()</code> 方法参数有 <code>input</code> 、 <code>batch_size</code>，具体说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，必填。<br/>
+<b>说明：</b>
+由于多模态模型对输入要求不同，请根据具体模型设定输入格式。<br/>
+<ul>
+<li>PP-Chart2Table的输入形式为<code>{'image': image_path}</code></li>
+</ul>
+</td>
+<td><code>dict</code></td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>含义：</b>批大小<br/>
+<b>说明：</b>
+可设置为任意正整数。</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+
+* 对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为<code>json</code>文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+</table>
+
+* 此外，也支持通过属性获取预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">获取预测的<code>json</code>格式的结果</td>
+</tr>
+</table>
+
+## 四、二次开发
+
+当前模块暂时不支持微调训练，仅支持推理集成。关于该模块的微调训练，计划在未来支持。
+
+## 五、推理引擎
+
+关于推理引擎的详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。
+
+### 5.1 速度数据
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="2">PP-Chart2Table</td>
+            <td>paddle_dynamic</td>
+            <td>53.00</td>
+            <td>17863.95</td>
+            <td>0.30</td>
+            <td>17917.78</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>23.95</td>
+            <td>12217.37</td>
+            <td>0.47</td>
+            <td>12269.98</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>测试环境说明:</strong>
+<ul>
+    <li><strong>测试数据：</strong><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/medal_table.jpg">示例图片</a></li>
+    <li><strong>硬件配置：</strong>
+        <ul>
+            <li>GPU：NVIDIA A100 40G</li>
+            <li>CPU：Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>软件环境：</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+## 六、FAQ
diff --git a/docs/version3.x/module_usage/doc_img_orientation_classification.en.md b/docs/version3.x/module_usage/doc_img_orientation_classification.en.md
new file mode 100644
index 0000000..b45b48f
--- /dev/null
+++ b/docs/version3.x/module_usage/doc_img_orientation_classification.en.md
@@ -0,0 +1,486 @@
+---
+comments: true
+---
+
+# Document Image Orientation Classification Module Tutorial
+
+## 1. Overview
+
+The Document Image Orientation Classification Module is primarily designed to distinguish the orientation of document images and correct them through post-processing. During processes such as document scanning or ID photo capturing, the device might be rotated to achieve clearer images, resulting in images with various orientations. Standard OCR pipelines may not handle these images effectively. By leveraging image classification techniques, the orientation of documents or IDs containing text regions can be pre-determined and adjusted, thereby improving the accuracy of OCR processing.
+
+## 2. Supported Models List
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing. The "Normal Mode" values correspond to the local <code>paddle_static</code> inference engine.
+
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Links</th>
+<th>Top-1 Acc (%)</th>
+<th>GPU Inference Time (ms)<br>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br>[Normal Mode / High-Performance Mode]</th>
+<th>Model Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">Pretrained Model</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>A document image classification model based on PP-LCNet_x1_0, with four categories: 0°, 90°, 180°, and 270°.</td>
+</tr>
+</tbody>
+</table>
+
+<strong>Test Environment Description:</strong>
+
+<ul>
+    <li><b>Performance Test Environment</b>
+        <ul>
+            <li><strong>Test Dataset:</strong> Self-built multi-scenario dataset (1000 images, including ID/document scenarios)</li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+        </ul>
+    </li>
+    <li><b>Inference Mode Description</b></li>
+</ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration</th>
+            <th>CPU Configuration</th>
+            <th>Acceleration Technology Combination</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Normal Mode</td>
+            <td>FP32 Precision / No TRT Acceleration</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>High-Performance Mode</td>
+            <td>Optimal combination of precision type and acceleration strategy</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Optimal backend selected (Paddle/OpenVINO/TRT, etc.)</td>
+        </tr>
+    </tbody>
+</table>
+
+## 3. Quick Start
+
+> ❗ Before starting, please install the PaddleOCR wheel package. For details, refer to the [Installation Guide](../installation.en.md).
+
+You can quickly experience it with one command:
+
+```bash
+paddleocr doc_img_orientation_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr doc_img_orientation_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg \
+    --engine transformers
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following command:
+
+```bash
+# Use the onnxruntime engine for inference
+paddleocr doc_img_orientation_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg \
+    --engine onnxruntime
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+<b>Note: </b>The official models would be download from HuggingFace by default. If can't access to HuggingFace, please set the environment variable `PADDLE_PDX_MODEL_SOURCE="BOS"` to change the model source to BOS. In the future, more model sources will be supported.
+
+You can also integrate the model inference of the Document Image Orientation Classification Module into your project. Before running the following code, please download the [sample image](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg) to your local machine.
+
+```python
+from paddleocr import DocImgOrientationClassification
+
+model = DocImgOrientationClassification(model_name="PP-LCNet_x1_0_doc_ori")
+output = model.predict("img_rot180_demo.jpg", batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_img("./output/demo.png")
+    res.save_to_json("./output/res.json")
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following code:
+
+```python
+from paddleocr import DocImgOrientationClassification
+
+model = DocImgOrientationClassification(
+    model_name="PP-LCNet_x1_0_doc_ori",
+    engine="transformers",
+)
+output = model.predict("img_rot180_demo.jpg", batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_img("./output/demo.png")
+    res.save_to_json("./output/res.json")
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following code:
+
+```python
+from paddleocr import DocImgOrientationClassification
+
+model = DocImgOrientationClassification(
+    model_name="PP-LCNet_x1_0_doc_ori",
+    engine="onnxruntime",
+)
+output = model.predict("img_rot180_demo.jpg", batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_img("./output/demo.png")
+    res.save_to_json("./output/res.json")
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+If you want to use the trained model with the `paddle_dynamic` or `transformers` engine, refer to the [Weight Conversion](#52-weight-conversion) section in the [Inference Engine](#5-inference-engine) section below to convert the model from the `pdparams` format to the `safetensors` format using PaddleX.
+
+After running, the result will be:
+
+```bash
+{'res': {'input_path': 'img_rot180_demo.jpg', 'page_index': None, 'class_ids': array([2], dtype=int32), 'scores': array([0.88164], dtype=float32), 'label_names': ['180']}}
+```
+
+The meaning of the output parameters is as follows:
+<ul>
+<li><code>input_path</code>：Represents the path of the input image.</li>
+<li><code>class_ids</code>：Represents the predicted class ID, with four categories: 0°, 90°, 180°, and 270°.</li>
+<li><code>scores</code>：Represents the confidence level of the prediction result.</li>
+<li><code>label_names</code>：Represents the category names of the prediction results.</li>
+</ul>
+
+Here is the visualization of the image:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/doc_img_ori_classification/img_rot180_demo_res.jpg">
+
+The explanations of relevant methods and parameters are as follows:
+
+* Instantiate the document image orientation classification model with <code>DocImgOrientationClassification</code> (taking <code>PP-LCNet_x1_0_doc_ori</code> as an example here). The specific explanations are as follows:
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>Meaning：</b>Model name. <br/>
+<b>Description：</b>
+If set to <code>None</code>, <code>PP-LCNet_x1_0_doc_ori</code> will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>Meaning：</b>Model storage path.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning：</b>Device for inference.<br/>
+<b>Description：</b>
+<b>For example:</b><code>"cpu"</code>, <code>"gpu"</code>, <code>"npu"</code>, <code>"gpu:0"</code>, <code>"gpu:0,1"</code>.<br/>
+If multiple devices are specified, parallel inference will be performed.<br/>
+By default, GPU 0 is used if available; otherwise, CPU is used.
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, <code>transformers</code>, and <code>onnxruntime</code>. When left as <code>None</code>, local inference uses the <code>paddle_static</code> engine by default. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning：</b>Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning：</b>Whether to use the Paddle Inference TensorRT subgraph engine.<br/>
+ <b>Description：</b>
+ If the model does not support acceleration through TensorRT, setting this flag will not enable acceleration.<br/>
+For Paddle with CUDA version 11.8, the compatible TensorRT version is 8.x (x>=6), and it is recommended to install TensorRT 8.6.1.6.<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning：</b>Computation precision when using the TensorRT subgraph engine in Paddle Inference.<br/>
+<b>Description：</b>
+<b>Options:</b><code>"fp32"</code>, <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+<b>Meaning：</b>Whether to enable MKL-DNN acceleration for inference. <br/>
+<b>Description：</b>
+If MKL-DNN is unavailable or the model does not support it, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning：</b>MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning：</b>Number of threads to use for inference on CPUs.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+</tbody>
+</table>
+
+* Call the  <code>predict()</code>  method of the document image orientation classification model for inference prediction. This method will return a list of results. In addition, this module also provides the <code>predict_iter()</code> method. The two methods are completely consistent in terms of parameter acceptance and result return. The difference is that  <code>predict_iter()</code>  returns a  <code>generator</code>, which can process and obtain prediction results step by step, suitable for scenarios where large datasets need to be processed or memory needs to be saved. You can choose either of these two methods according to your actual needs. The parameters of the <code>predict()</code>  method are <code>input</code> and <code>batch_size</code>, and the specific explanations are as follows:
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning：</b>Input data to be predicted. Required. <br/>
+<b>Description：</b>
+Supports multiple input types:
+<ul>
+<li><b>Python Var</b>: e.g., <code>numpy.ndarray</code> representing image data</li>
+<li><b>str</b>：Local image or PDF file path: <code>/root/data/img.jpg</code>；
+<b>URL</b> of image or PDF file: e.g., <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">example</a>;
+<b>Local directory</b>: directory containing images for prediction, e.g., <code>/root/data/</code> (Note: directories containing PDF files are not supported; PDFs must be specified by exact file path)</li>
+<li><b>list</b>: Elements must be of the above types, e.g., <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>Meaning：</b>Batch size. <br/>
+<b>Description：</b>
+Positive integer.</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+
+* Process the prediction results. The prediction result for each sample is the corresponding Result object, and it supports operations such as printing, saving as an image, and saving as a <code>json</code> file:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Description</th>
+<th>Parameter</th>
+<th>Parameter Type</th>
+<th>Description</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">Print the result to the terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output content using <code>JSON</code> indentation</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data and make it more readable. It is only valid when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non-<code>ASCII</code> characters as <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; when set to <code>False</code>, the original characters will be retained. It is only valid when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">Save the result as a file in <code>json</code> format</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The file path to save. When it is a directory, the saved file name is consistent with the naming of the input file type.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data and make it more readable. It is only valid when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non-<code>ASCII</code> characters as <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; when set to <code>False</code>, the original characters will be retained. It is only valid when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save the result as a file in image format</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The file path to save. When it is a directory, the saved file name is consistent with the naming of the input file type.</td>
+<td>None</td>
+</tr>
+</table>
+
+* In addition, it also supports obtaining the visualization image with results and the prediction results through attributes. The specifics are as follows:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">Get the prediction result in <code>json</code> format</td>
+</tr>
+<tr>
+<td rowspan = "1"><code>img</code></td>
+<td rowspan = "1">Get the visualization image in <code>dict</code> format</td>
+</tr>
+
+</table>
+
+## 4. Secondary Development
+
+Since PaddleOCR does not directly provide training functionality for document image orientation classification, if you need to train a document image orientation classification model, you can refer to the [PaddleX Secondary Development for Document Image Orientation Classification](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#iv-custom-development) section for training guidance. The trained model can be seamlessly integrated into PaddleOCR's API for inference purposes.
+
+If you want to use the `paddle_dynamic` or `transformers` engine with the trained model, please refer to the [Weight Conversion](#52-weight-conversion) section in [Inference Engine](#5-inference-engine) later in this document to convert the model from the `pdparams` format to the `safetensors` format using PaddleX.
+
+## 5. Inference Engine
+
+For detailed descriptions, values, compatibility rules, and examples of the inference engine, please refer to <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration Description</a>.
+
+### 5.1 Speed Data
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">PP-LCNet_x1_0_doc_ori</td>
+            <td>paddle_static</td>
+            <td>2.21</td>
+            <td>3.36</td>
+            <td>0.06</td>
+            <td>5.74</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>2.15</td>
+            <td>7.54</td>
+            <td>0.07</td>
+            <td>9.87</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>4.46</td>
+            <td>3.44</td>
+            <td>0.14</td>
+            <td>8.36</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>2.02</td>
+            <td>0.87</td>
+            <td>0.05</td>
+            <td>3.03</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>Test Environment Description:</strong>
+<ul>
+    <li><strong>Test Data:</strong> <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">Sample Image</a></li>
+    <li><strong>Hardware Configuration:</strong>
+        <ul>
+            <li>GPU: NVIDIA A100 40G</li>
+            <li>CPU: Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>Software Environment:</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 5.2 Weight Conversion
+
+When using the inference engine, the system will automatically download the official pre-trained model. If you need to use a self-trained model with the `paddle_dynamic` or `transformers` engine, please refer to the [PaddleX Text Image Orientation Classification Module Weight Conversion](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#442) section to convert the model from the `pdparams` format to the `safetensors` format using PaddleX. This allows seamless integration into the PaddleOCR API for inference. If you need to use a self-trained model with the `onnxruntime` engine, refer to [PaddleX Obtain ONNX Models](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html) to obtain the ONNX model, so it can be seamlessly integrated into the PaddleOCR API for inference.
+
+## 6. FAQ
diff --git a/docs/version3.x/module_usage/doc_img_orientation_classification.md b/docs/version3.x/module_usage/doc_img_orientation_classification.md
new file mode 100644
index 0000000..fe4b5ed
--- /dev/null
+++ b/docs/version3.x/module_usage/doc_img_orientation_classification.md
@@ -0,0 +1,477 @@
+---
+comments: true
+---
+
+# 文档图像方向分类模块使用教程
+
+## 一、概述
+
+文档图像方向分类模块主要是将文档图像的方向区分出来，并使用后处理将其矫正。在诸如文档扫描、证照拍摄等过程中，有时为了拍摄更清晰，会将拍摄设备进行旋转，导致得到的图片也是不同方向的。此时，标准的OCR流程无法很好地应对这些数据。利用图像分类技术，可以预先判断含文字区域的文档或证件的方向，并将其进行方向调整，从而提高OCR处理的准确性。
+
+## 二、支持模型列表
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。表格中的“常规模式”耗时对应本地 <code>paddle_static</code> 推理引擎。
+
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>Top-1 Acc（%）</th>
+<th>GPU推理耗时（ms）<br>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">训练模型</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>基于PP-LCNet_x1_0的文档图像分类模型，含有四个类别，即0度，90度，180度，270度</td>
+</tr>
+</tbody>
+</table>
+
+ <strong>测试环境说明:</strong>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+              <li><strong>测试数据集：</strong>自建多场景数据集（1000张图片，含证件/文档等场景）</li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+
+## 三、快速开始
+
+
+> ❗ 在快速开始前，请先安装 PaddleOCR 的 wheel 包，详细请参考 [安装教程](../installation.md)。
+
+
+使用一行命令即可快速体验：
+
+```bash
+paddleocr doc_img_orientation_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下命令：
+
+```bash
+# 使用 transformers 引擎进行推理
+paddleocr doc_img_orientation_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg \
+    --engine transformers
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下命令：
+
+```bash
+# 使用 onnxruntime 引擎进行推理
+paddleocr doc_img_orientation_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg \
+    --engine onnxruntime
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+如需了解推理引擎的介绍、速度数据及权重转换等详细信息，请参考后文 [推理引擎](#五推理引擎) 章节。
+
+您也可以将文档图像方向分类模块中的模型推理集成到您的项目中。运行以下代码前，请您下载[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg)到本地。
+
+```python
+from paddleocr import DocImgOrientationClassification
+
+model = DocImgOrientationClassification(model_name="PP-LCNet_x1_0_doc_ori")
+output = model.predict("img_rot180_demo.jpg",  batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_img("./output/demo.png")
+    res.save_to_json("./output/res.json")
+```
+
+如果希望使用 `transformers` 引擎进行推理，可以参考如下示例：
+
+```python
+from paddleocr import DocImgOrientationClassification
+
+model = DocImgOrientationClassification(
+    model_name="PP-LCNet_x1_0_doc_ori",
+    engine="transformers",
+)
+```
+
+如果希望使用 `onnxruntime` 引擎进行推理，可以参考如下示例：
+
+```python
+from paddleocr import DocImgOrientationClassification
+
+model = DocImgOrientationClassification(
+    model_name="PP-LCNet_x1_0_doc_ori",
+    engine="onnxruntime",
+)
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+随后继续按上文方式调用 `predict()` 即可。
+
+
+运行后，得到的结果为：
+
+```bash
+{'res': {'input_path': 'img_rot180_demo.jpg', 'page_index': None, 'class_ids': array([2], dtype=int32), 'scores': array([0.88164], dtype=float32), 'label_names': ['180']}}
+```
+
+运行结果参数含义如下：
+<ul>
+<li><code>input_path</code>：表示输入图片的路径。</li>
+<li><code>class_ids</code>：表示预测结果的类别id，含有四个类别，即0度，90度，180度和270度。</li>
+<li><code>scores</code>：表示预测结果的置信度。</li>
+<li><code>label_names</code>：表示预测结果的类别名。</li>
+</ul>
+
+可视化图片如下：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/doc_img_ori_classification/img_rot180_demo_res.jpg">
+
+相关方法、参数等说明如下：
+
+* <code>DocImgOrientationClassification</code>实例化文档图像方向分类模型（此处以<code>PP-LCNet_x1_0_doc_ori</code>为例），具体说明如下：
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>说明</th>
+<th>类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>含义：</b>模型名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，则使用<code>PP-LCNet_x1_0_doc_ori</code>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>含义：</b>模型存储路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+<b>例如：</b><code>"cpu"</code>、<code>"gpu"</code>、<code>"npu"</code>、<code>"gpu:0"</code>、<code>"gpu:0,1"</code>。<br/>
+如指定多个设备，将进行并行推理。<br/>
+默认情况下，优先使用 GPU 0；若不可用则使用 CPU。
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>、<code>onnxruntime</code>。保持为默认值 <code>None</code> 时，本地推理默认使用 <code>paddle_static</code> 引擎。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>当使用 Paddle Inference 的 TensorRT 子图引擎时设置的计算精度。<br/>
+<b>说明：</b>
+<b>例如：</b><code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+<b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上推理时使用的线程数量。</td>
+<td><code>int|None</code></td>
+<td><code>10</code></td>
+</tr>
+</tbody>
+</table>
+
+* 调用文档图像方向分类模型的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。另外，本模块还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。<code>predict()</code> 方法参数有 <code>input</code> 和 <code>batch_size</code>，具体说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；
+<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">示例</a>；
+<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>含义：</b>批大小。<br/>
+<b>说明：</b>
+可设置为任意正整数。</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+
+* 对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为<code>json</code>文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将结果保存为图像格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+</table>
+
+* 此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">获取预测的<code>json</code>格式的结果</td>
+</tr>
+<tr>
+<td rowspan = "1"><code>img</code></td>
+<td rowspan = "1">获取格式为<code>dict</code>的可视化图像</td>
+</tr>
+
+</table>
+
+## 四、二次开发
+
+由于 PaddleOCR 并不直接提供文档图像方向分类的训练，因此，如果需要训练文档图像方向分类模型，可以参考 [PaddleX 文档图像方向分类二次开发](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#_5)部分进行训练。训练后的模型可以无缝集成到 PaddleOCR 的 API 中进行推理。
+
+训练后的模型如果想使用 `paddle_dynamic` 或 `transformers` 引擎，请参考后文 [推理引擎](#五推理引擎) 中的 [权重转换](#52-权重转换) 部分将模型由 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式。
+
+## 五、推理引擎 {#五推理引擎}
+
+关于推理引擎的详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。
+
+### 5.1 速度数据
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">PP-LCNet_x1_0_doc_ori</td>
+            <td>paddle_static</td>
+            <td>2.21</td>
+            <td>3.36</td>
+            <td>0.06</td>
+            <td>5.74</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>2.15</td>
+            <td>7.54</td>
+            <td>0.07</td>
+            <td>9.87</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>4.46</td>
+            <td>3.44</td>
+            <td>0.14</td>
+            <td>8.36</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>2.02</td>
+            <td>0.87</td>
+            <td>0.05</td>
+            <td>3.03</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>测试环境说明:</strong>
+<ul>
+    <li><strong>测试数据：</strong><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">示例图片</a></li>
+    <li><strong>硬件配置：</strong>
+        <ul>
+            <li>GPU：NVIDIA A100 40G</li>
+            <li>CPU：Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>软件环境：</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 5.2 权重转换 {#52-权重转换}
+
+使用推理引擎时，系统会自动下载官方预训练模型。若需使用自训练模型配合 `paddle_dynamic` 或 `transformers` 引擎，请参考 [PaddleX 文本图像方向分类模块权重转换](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#442) 部分，将 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式，即可无缝集成到 PaddleOCR 的 API 中进行推理。若需使用自训练模型配合`onnxruntime`引擎，请参考[PaddleX 获取 ONNX 模型](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html)获取onnx模型，即可无缝集成到 PaddleOCR 的 API 中进行推理。
+
+## 六、FAQ
diff --git a/docs/version3.x/module_usage/doc_vlm.en.md b/docs/version3.x/module_usage/doc_vlm.en.md
new file mode 100644
index 0000000..b79b9ab
--- /dev/null
+++ b/docs/version3.x/module_usage/doc_vlm.en.md
@@ -0,0 +1,265 @@
+---
+comments: true
+---
+
+# Document Visual Language Model Module Tutorial
+
+## I. Overview
+
+Document visual language models are a cutting-edge multimodal processing technology aimed at addressing the limitations of traditional document processing methods. Traditional methods are often limited to processing document information in specific formats or predefined categories, whereas document visual language models can integrate visual and linguistic information to understand and handle diverse document content. By combining computer vision and natural language processing technologies, these models can recognize images, text, and their relationships within documents, and even understand semantic information within complex layout structures. This makes document processing more intelligent and flexible, with stronger generalization capabilities, showing broad application prospects in automated office work, information extraction, and other fields.
+
+## II. Supported Model List
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing.
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Model Storage Size (GB)</th>
+<th>Total Score</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>PP-DocBee-2B</td><td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBee-2B_infer.tar">Inference Model</a></td>
+<td>4.2</td>
+<td>765</td>
+<td rowspan="2">PP-DocBee is a self-developed multimodal large model by the PaddlePaddle team, focusing on document understanding, and it performs excellently in Chinese document understanding tasks. The model is fine-tuned and optimized using nearly 5 million multimodal datasets for document understanding, including general VQA, OCR, charts, text-rich documents, mathematics and complex reasoning, synthetic data, and pure text data, with different training data ratios set. On several authoritative English document understanding evaluation lists in academia, PP-DocBee has basically achieved SOTA for models of the same parameter scale. In terms of internal business Chinese scenario indicators, PP-DocBee also outperforms the current popular open-source and closed-source models.</td>
+</tr>
+<tr>
+<td>PP-DocBee-7B</td><td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBee-7B_infer.tar">Inference Model</a></td>
+<td>15.8</td>
+<td>-</td>
+</tr>
+<tr>
+<td>PP-DocBee2-3B</td><td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBee2-3B_infer.tar">Inference Model</a></td>
+<td>7.6</td>
+<td>852</td>
+<td>PP-DocBee2 is a self-developed multimodal large model by the PaddlePaddle team, further optimizing the base model on the foundation of PP-DocBee and introducing a new data optimization scheme to improve data quality. Using a small amount of 470,000 data generated by a self-developed data synthesis strategy, PP-DocBee2 performs better in Chinese document understanding tasks. In terms of internal business Chinese scenario indicators, PP-DocBee2 improves by about 11.4% compared to PP-DocBee, and also outperforms the current popular open-source and closed-source models of the same scale.</td>
+</tr>
+</table>
+
+<b>Note: The total scores of the above models are test results from an internal evaluation set, where all images have a resolution (height, width) of (1680, 1204), with a total of 1196 data entries, covering scenarios such as financial reports, laws and regulations, scientific and technical papers, manuals, humanities papers, contracts, research reports, etc. There are no plans for public release at the moment.</b>
+
+## III. Quick Start
+
+> ❗ Before starting quickly, please install the PaddleOCR wheel package. For details, please refer to the [Installation Guide](../installation.en.md).
+
+You can quickly experience it with one line of command:
+
+```bash
+paddleocr doc_vlm -i "{'image': 'https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/medal_table.png', 'query': '识别这份表格的内容, 以markdown格式输出'}"
+```
+
+The example above uses the <code>paddle_dynamic</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+<b>Note: </b>The official models would be download from HuggingFace by default. If can't access to HuggingFace, please set the environment variable `PADDLE_PDX_MODEL_SOURCE="BOS"` to change the model source to BOS. In the future, more model sources will be supported.
+
+You can also integrate the model inference from the open document visual language model module into your project. Before running the following code, please download the [sample image](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/medal_table.png) locally.
+
+```python
+from paddleocr import DocVLM
+model = DocVLM(model_name="PP-DocBee2-3B")
+results = model.predict(
+    input={"image": "medal_table.png", "query": "识别这份表格的内容, 以markdown格式输出"},
+    batch_size=1
+)
+for res in results:
+    res.print()
+    res.save_to_json(f"./output/res.json")
+```
+
+The example above uses the <code>paddle_dynamic</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+After running, the result is:
+
+```bash
+{'res': {'image': 'medal_table.png', 'query': '识别这份表格的内容, 以markdown格式输出', 'result': '| 名次 | 国家/地区 | 金牌 | 银牌 | 铜牌 | 奖牌总数 |\n| --- | --- | --- | --- | --- | --- |\n| 1 | 中国（CHN） | 48 | 22 | 30 | 100 |\n| 2 | 美国（USA） | 36 | 39 | 37 | 112 |\n| 3 | 俄罗斯（RUS） | 24 | 13 | 23 | 60 |\n| 4 | 英国（GBR） | 19 | 13 | 19 | 51 |\n| 5 | 德国（GER） | 16 | 11 | 14 | 41 |\n| 6 | 澳大利亚（AUS） | 14 | 15 | 17 | 46 |\n| 7 | 韩国（KOR） | 13 | 11 | 8 | 32 |\n| 8 | 日本（JPN） | 9 | 8 | 8 | 25 |\n| 9 | 意大利（ITA） | 8 | 9 | 10 | 27 |\n| 10 | 法国（FRA） | 7 | 16 | 20 | 43 |\n| 11 | 荷兰（NED） | 7 | 5 | 4 | 16 |\n| 12 | 乌克兰（UKR） | 7 | 4 | 11 | 22 |\n| 13 | 肯尼亚（KEN） | 6 | 4 | 6 | 16 |\n| 14 | 西班牙（ESP） | 5 | 11 | 3 | 19 |\n| 15 | 牙买加（JAM） | 5 | 4 | 2 | 11 |\n'}}
+```
+
+The meaning of the result parameters is as follows:
+<ul>
+<li><b>image</b>: Indicates the path of the input image to be predicted</li>
+<li><b>query</b>: Represents the input text information to be predicted</li>
+<li><b>result</b>: Information of the model's prediction result</li>
+</ul>
+
+The visualization of the prediction result is as follows:
+
+```bash
+| 名次 | 国家/地区 | 金牌 | 银牌 | 铜牌 | 奖牌总数 |
+| --- | --- | --- | --- | --- | --- |
+| 1 | 中国（CHN） | 48 | 22 | 30 | 100 |
+| 2 | 美国（USA） | 36 | 39 | 37 | 112 |
+| 3 | 俄罗斯（RUS） | 24 | 13 | 23 | 60 |
+| 4 | 英国（GBR） | 19 | 13 | 19 | 51 |
+| 5 | 德国（GER） | 16 | 11 | 14 | 41 |
+| 6 | 澳大利亚（AUS） | 14 | 15 | 17 | 46 |
+| 7 | 韩国（KOR） | 13 | 11 | 8 | 32 |
+| 8 | 日本（JPN） | 9 | 8 | 8 | 25 |
+| 9 | 意大利（ITA） | 8 | 9 | 10 | 27 |
+| 10 | 法国（FRA） | 7 | 16 | 20 | 43 |
+| 11 | 荷兰（NED） | 7 | 5 | 4 | 16 |
+| 12 | 乌克兰（UKR） | 7 | 4 | 11 | 22 |
+| 13 | 肯尼亚（KEN） | 6 | 4 | 6 | 16 |
+| 14 | 西班牙（ESP） | 5 | 11 | 3 | 19 |
+| 15 | 牙买加（JAM） | 5 | 4 | 2 | 11 |
+```
+
+Explanations of related methods, parameters, etc., are as follows:
+
+* <code>DocVLM</code> instantiates the document visual language model (taking <code>PP-DocBee-2B</code> as an example), with specific explanations as follows:
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>Meaning:</b> Model name.<br/>
+<b>Description:</b>
+ If set to <code>None</code>, <code>PP-DocBee-2B</code> will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>Meaning:</b>Model storage path.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b> Device for inference.<br/>
+<b>Description:</b>
+<b>For example:</b><code>"cpu"</code>, <code>"gpu"</code>, <code>"npu"</code>, <code>"gpu:0"</code>, <code>"gpu:0,1"</code>.<br/>
+By default, GPU 0 is used if available; otherwise, CPU is used.
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, and <code>paddle_dynamic</code>. When left as <code>None</code>, local inference uses the <code>paddle_dynamic</code> engine by default. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+* Call the  <code>predict()</code> method of the document visual language model for inference prediction. This method will return a result list. Additionally, this module also provides the <code>predict_iter()</code> method. Both are completely consistent in terms of parameter acceptance and result return, the difference being that <code>predict_iter()</code> returns a <code>generator</code>, capable of gradually processing and obtaining prediction results, suitable for handling large datasets or scenarios where memory saving is desired. You can choose to use either of these methods based on actual needs. The <code>predict()</code> method parameters include  <code>input</code> , <code>batch_size</code>, with specific explanations as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b> Input data. Required. <br/>
+<b>Description:</b>
+Since multimodal models have different input requirements, please refer to the specific model for the correct format.<br/>
+For example, for the PP-DocBee series models, the input format should be: <code>{'image': image_path, 'query': query_text}</code>
+</td>
+<td><code>dict</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>Meaning:</b>Batch size.<br/>
+<b>Description:</b>
+Positive integer.</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+
+* Process the prediction results. The prediction result for each sample is the corresponding Result object, and it supports operations such as printing and saving as <code>json</code> file:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Description</th>
+<th>Parameter</th>
+<th>Type</th>
+<th>Description</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">Print results to terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output content using <code>JSON</code> indentation</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data, making it more readable, effective only when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether non-<code>ASCII</code> characters are escaped to <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> retains the original characters, effective only when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">Save the result as a json format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Path of the file to be saved. When it is a directory, the naming of the saved file is consistent with the input file type.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data, making it more readable, effective only when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether non-<code>ASCII</code> characters are escaped to <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> retains the original characters, effective only when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+</table>
+
+* Additionally, it also supports obtaining prediction results through attributes, as follows:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">Get the prediction result in <code>json</code> format</td>
+</tr>
+</table>
+
+## IV. Secondary Development
+
+The current module does not support fine-tuning training temporarily, only inference integration is supported. The fine-tuning training of this module is planned to be supported in the future.
+
+## V. FAQ
diff --git a/docs/version3.x/module_usage/doc_vlm.md b/docs/version3.x/module_usage/doc_vlm.md
new file mode 100644
index 0000000..7853e53
--- /dev/null
+++ b/docs/version3.x/module_usage/doc_vlm.md
@@ -0,0 +1,268 @@
+---
+comments: true
+---
+
+# 文档类视觉语言模型模块使用教程
+
+## 一、概述
+
+文档类视觉语言模型是当前一种前沿的多模态处理技术，旨在解决传统文档处理方法的局限性。传统方法往往局限于处理特定格式或预定义类别的文档信息，而文档类视觉语言模型能够融合视觉与语言信息，理解并处理多样化的文档内容。通过结合计算机视觉与自然语言处理技术，模型可以识别文档中的图像、文本及其相互关系，甚至能理解复杂版面结构中的语义信息。这使得文档处理更加智能化、灵活化，具备更强的泛化能力，在自动化办公、信息提取等领域展现出广阔的应用前景。
+
+## 二、支持模型列表
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>模型存储大小（GB）</th>
+<th>模型总分</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-DocBee-2B</td><td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBee-2B_infer.tar">推理模型</a></td>
+<td>4.2</td>
+<td>765</td>
+<td rowspan="2">PP-DocBee 是飞桨团队自研的一款专注于文档理解的多模态大模型，在中文文档理解任务上具有卓越表现。该模型通过近 500 万条文档理解类多模态数据集进行微调优化，各种数据集包括了通用VQA类、OCR类、图表类、text-rich文档类、数学和复杂推理类、合成数据类、纯文本数据等，并设置了不同训练数据配比。在学术界权威的几个英文文档理解评测榜单上，PP-DocBee基本都达到了同参数量级别模型的SOTA。在内部业务中文场景类的指标上，PP-DocBee也高于目前的热门开源和闭源模型。</td>
+</tr>
+<tr>
+<td>PP-DocBee-7B</td><td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBee-7B_infer.tar">推理模型</a></td>
+<td>15.8</td>
+<td>-</td>
+</tr>
+<tr>
+<td>PP-DocBee2-3B</td><td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBee2-3B_infer.tar">推理模型</a></td>
+<td>7.6</td>
+<td>852</td>
+<td>PP-DocBee2 是飞桨团队自研的一款专注于文档理解的多模态大模型，在PP-DocBee的基础上进一步优化了基础模型，并引入了新的数据优化方案，提高了数据质量，使用自研数据合成策略生成的少量的47万数据便使得PP-DocBee2在中文文档理解任务上表现更佳。在内部业务中文场景类的指标上，PP-DocBee2相较于PP-DocBee提升了约11.4%，同时也高于目前的同规模热门开源和闭源模型。</td>
+</tr>
+</table>
+
+<b>注：以上模型总分为内部评估集模型测试结果，内部评估集所有图像分辨率 (height, width) 为 (1680,1204)，共1196条数据，包括了财报、法律法规、理工科论文、说明书、文科论文、合同、研报等场景，暂时未有计划公开。</b>
+
+
+
+## 三、快速开始
+
+> ❗ 在快速开始前，请先安装 PaddleOCR 的 wheel 包，详细请参考 [安装教程](../installation.md)。
+
+使用一行命令即可快速体验：
+
+```bash
+paddleocr doc_vlm -i "{'image': 'https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/medal_table.png', 'query': '识别这份表格的内容, 以markdown格式输出'}"
+```
+
+上述示例默认使用 <code>paddle_dynamic</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+<b>注：</b>PaddleOCR 官方模型默认从 HuggingFace 获取，如运行环境访问 HuggingFace 不便，可通过环境变量修改模型源为 BOS：`PADDLE_PDX_MODEL_SOURCE="BOS"`，未来将支持更多主流模型源；
+
+您也可以将开放文档类视觉语言模型模块中的模型推理集成到您的项目中。运行以下代码前，请您下载[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/medal_table.png)到本地。
+
+```python
+from paddleocr import DocVLM
+model = DocVLM(model_name="PP-DocBee2-3B")
+results = model.predict(
+    input={"image": "medal_table.png", "query": "识别这份表格的内容, 以markdown格式输出"},
+    batch_size=1
+)
+for res in results:
+    res.print()
+    res.save_to_json(f"./output/res.json")
+```
+
+上述示例默认使用 <code>paddle_dynamic</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+运行后，得到的结果为：
+
+```bash
+{'res': {'image': 'medal_table.png', 'query': '识别这份表格的内容, 以markdown格式输出', 'result': '| 名次 | 国家/地区 | 金牌 | 银牌 | 铜牌 | 奖牌总数 |\n| --- | --- | --- | --- | --- | --- |\n| 1 | 中国（CHN） | 48 | 22 | 30 | 100 |\n| 2 | 美国（USA） | 36 | 39 | 37 | 112 |\n| 3 | 俄罗斯（RUS） | 24 | 13 | 23 | 60 |\n| 4 | 英国（GBR） | 19 | 13 | 19 | 51 |\n| 5 | 德国（GER） | 16 | 11 | 14 | 41 |\n| 6 | 澳大利亚（AUS） | 14 | 15 | 17 | 46 |\n| 7 | 韩国（KOR） | 13 | 11 | 8 | 32 |\n| 8 | 日本（JPN） | 9 | 8 | 8 | 25 |\n| 9 | 意大利（ITA） | 8 | 9 | 10 | 27 |\n| 10 | 法国（FRA） | 7 | 16 | 20 | 43 |\n| 11 | 荷兰（NED） | 7 | 5 | 4 | 16 |\n| 12 | 乌克兰（UKR） | 7 | 4 | 11 | 22 |\n| 13 | 肯尼亚（KEN） | 6 | 4 | 6 | 16 |\n| 14 | 西班牙（ESP） | 5 | 11 | 3 | 19 |\n| 15 | 牙买加（JAM） | 5 | 4 | 2 | 11 |\n'}}
+```
+运行结果参数含义如下：
+<ul>
+<li><b>image</b>: 表示输入待预测图像的路径</li>
+<li><b>query</b>: 表述输入待预测的文本信息</li>
+<li><b>result</b>: 模型预测的结果信息</li>
+</ul>
+
+
+预测结果打印可视化如下：
+
+```bash
+| 名次 | 国家/地区 | 金牌 | 银牌 | 铜牌 | 奖牌总数 |
+| --- | --- | --- | --- | --- | --- |
+| 1 | 中国（CHN） | 48 | 22 | 30 | 100 |
+| 2 | 美国（USA） | 36 | 39 | 37 | 112 |
+| 3 | 俄罗斯（RUS） | 24 | 13 | 23 | 60 |
+| 4 | 英国（GBR） | 19 | 13 | 19 | 51 |
+| 5 | 德国（GER） | 16 | 11 | 14 | 41 |
+| 6 | 澳大利亚（AUS） | 14 | 15 | 17 | 46 |
+| 7 | 韩国（KOR） | 13 | 11 | 8 | 32 |
+| 8 | 日本（JPN） | 9 | 8 | 8 | 25 |
+| 9 | 意大利（ITA） | 8 | 9 | 10 | 27 |
+| 10 | 法国（FRA） | 7 | 16 | 20 | 43 |
+| 11 | 荷兰（NED） | 7 | 5 | 4 | 16 |
+| 12 | 乌克兰（UKR） | 7 | 4 | 11 | 22 |
+| 13 | 肯尼亚（KEN） | 6 | 4 | 6 | 16 |
+| 14 | 西班牙（ESP） | 5 | 11 | 3 | 19 |
+| 15 | 牙买加（JAM） | 5 | 4 | 2 | 11 |
+```
+
+
+相关方法、参数等说明如下：
+
+* <code>DocVLM</code>实例化文档类视觉语言模型（此处以<code>PP-DocBee-2B</code>为例），具体说明如下：
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>含义：</b>模型名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，则使用<code>PP-DocBee-2B</code>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>含义：</b>模型存储路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+<b>例如：</b><code>"cpu"</code>、<code>"gpu"</code>、<code>"npu"</code>、<code>"gpu:0"</code></code>。
+默认情况下，优先使用 GPU 0；若不可用则使用 CPU。
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_dynamic</code>。保持为默认值 <code>None</code> 时，本地推理默认使用 <code>paddle_dynamic</code> 引擎。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+* 调用文档类视觉语言模型的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。另外，本模块还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。<code>predict()</code> 方法参数有 <code>input</code> 和 <code>batch_size</code>，具体说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，必填。<br/>
+<b>说明：</b>
+由于多模态模型对输入要求不同，请根据具体模型设定输入格式。<br/>
+例如：对于 PP-DocBee 系列模型，输入形式应为：<code>{'image': image_path, 'query': query_text}</code>
+</td>
+<td><code>dict</code></td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>含义：</b>批大小。<br/>
+<b>说明：</b>
+可设置为任意正整数。</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+
+* 对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为<code>json</code>文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+</table>
+
+* 此外，也支持通过属性获取预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">获取预测的<code>json</code>格式的结果</td>
+</tr>
+</table>
+
+## 四、二次开发
+
+当前模块暂时不支持微调训练，仅支持推理集成。关于该模块的微调训练，计划在未来支持。
+
+## 五、FAQ
diff --git a/docs/version3.x/module_usage/formula_recognition.en.md b/docs/version3.x/module_usage/formula_recognition.en.md
new file mode 100644
index 0000000..d7dfcbb
--- /dev/null
+++ b/docs/version3.x/module_usage/formula_recognition.en.md
@@ -0,0 +1,493 @@
+---
+comments: true
+---
+
+# Formula Recognition Module Tutorial
+
+## I. Overview
+
+The formula recognition module is a key component of an OCR (Optical Character Recognition) system, responsible for converting mathematical formulas in images into editable text or computer-readable formats. The performance of this module directly affects the accuracy and efficiency of the entire OCR system. The formula recognition module typically outputs LaTeX or MathML code of the mathematical formulas, which will be passed as input to the text understanding module for further processing.
+
+## II. Supported Model List
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing. The "Normal Mode" values correspond to the local <code>paddle_static</code> inference engine.
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>En-BLEU(%)</th>
+<th>Zh-BLEU(%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+<tr>
+<td>UniMERNet</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UniMERNet_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UniMERNet_pretrained.pdparams">Training Model</a></td>
+<td>85.91</td>
+<td>43.50</td>
+<td>1311.84 / 1311.84</td>
+<td>- / 8288.07</td>
+<td>1530</td>
+<td>UniMERNet is a formula recognition model developed by Shanghai AI Lab. It uses Donut Swin as the encoder and MBartDecoder as the decoder. The model is trained on a dataset of one million samples, including simple formulas, complex formulas, scanned formulas, and handwritten formulas, significantly improving the recognition accuracy of real-world formulas.</td>
+</tr>
+<td>PP-FormulaNet-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-S_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-S_pretrained.pdparams">Training Model</a></td>
+<td>87.00</td>
+<td>45.71</td>
+<td>182.25 / 182.25</td>
+<td>- / 254.39</td>
+<td>224</td>
+<td rowspan="2">PP-FormulaNet is an advanced formula recognition model developed by the Baidu PaddlePaddle Vision Team. The PP-FormulaNet-S version uses PP-HGNetV2-B4 as its backbone network. Through parallel masking and model distillation techniques, it significantly improves inference speed while maintaining high recognition accuracy, making it suitable for applications requiring fast inference. The PP-FormulaNet-L version, on the other hand, uses Vary_VIT_B as its backbone network and is trained on a large-scale formula dataset, showing significant improvements in recognizing complex formulas compared to PP-FormulaNet-S.</td>
+</tr>
+<td>PP-FormulaNet-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-L_pretrained.pdparams">Training Model</a></td>
+<td>90.36</td>
+<td>45.78</td>
+<td>1482.03 / 1482.03</td>
+<td>- / 3131.54</td>
+<td>695</td>
+</tr>
+<td>PP-FormulaNet_plus-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-S_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-S_pretrained.pdparams">Training Model</a></td>
+<td>88.71</td>
+<td>53.32</td>
+<td>179.20 / 179.20</td>
+<td>- / 260.99</td>
+<td>248</td>
+<td rowspan="3">PP-FormulaNet_plus is an enhanced version of the formula recognition model developed by the Baidu PaddlePaddle Vision Team, building upon the original PP-FormulaNet. Compared to the original version, PP-FormulaNet_plus utilizes a more diverse formula dataset during training, including sources such as Chinese dissertations, professional books, textbooks, exam papers, and mathematics journals. This expansion significantly improves the model’s recognition capabilities. Among the models, PP-FormulaNet_plus-M and PP-FormulaNet_plus-L have added support for Chinese formulas and increased the maximum number of predicted tokens for formulas from 1,024 to 2,560, greatly enhancing the recognition performance for complex formulas. Meanwhile, the PP-FormulaNet_plus-S model focuses on improving the recognition of English formulas. With these improvements, the PP-FormulaNet_plus series models perform exceptionally well in handling complex and diverse formula recognition tasks. </td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-M_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-M_pretrained.pdparams">Training Model</a></td>
+<td>91.45</td>
+<td>89.76</td>
+<td>1040.27 / 1040.27</td>
+<td>- / 1615.80</td>
+<td>592</td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-L_pretrained.pdparams">Training Model</a></td>
+<td>92.22</td>
+<td>90.64</td>
+<td>1476.07 / 1476.07</td>
+<td>- / 3125.58</td>
+<td>698</td>
+</tr>
+<tr>
+<td>LaTeX_OCR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/LaTeX_OCR_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/LaTeX_OCR_rec_pretrained.pdparams">Training Model</a></td>
+<td>74.55</td>
+<td>39.96</td>
+<td>1088.89 / 1088.89</td>
+<td>- / -</td>
+<td>99</td>
+<td>LaTeX-OCR is a formula recognition algorithm based on an autoregressive large model. It uses Hybrid ViT as the backbone network and a transformer as the decoder, significantly improving the accuracy of formula recognition.</td>
+</tr>
+</table>
+<strong>Test Environment Description:</strong>
+    <ul>
+        <li><b>Performance Test Environment</b>
+            <ul>
+                <li><strong>Test Dataset:</strong> PaddleOCR internal custom formula recognition test set</li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+        </li>
+        <li><b>Inference Mode Description</b></li>
+    </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration</th>
+            <th>CPU Configuration</th>
+            <th>Acceleration Technique Combination</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Standard Mode</td>
+            <td>FP32 precision / No TRT acceleration</td>
+            <td>FP32 precision / 8 threads</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>High-Performance Mode</td>
+            <td>Optimal combination of predefined precision type and acceleration strategy</td>
+            <td>FP32 precision / 8 threads</td>
+            <td>Optimal predefined backend (Paddle/OpenVINO/TRT, etc.)</td>
+        </tr>
+    </tbody>
+</table>
+
+## III. Quick Start
+
+> ❗ Before getting started, please install the PaddleOCR wheel package. For details, refer to the [Installation Guide](../installation.en.md).
+
+You can quickly try it out with a single command:
+
+```bash
+paddleocr formula_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_formula_rec_001.png
+
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+<b>Note: </b>The official models would be download from HuggingFace by default. If can't access to HuggingFace, please set the environment variable `PADDLE_PDX_MODEL_SOURCE="BOS"` to change the model source to BOS. In the future, more model sources will be supported.
+
+You can also integrate the model inference from the formula recognition module into your own project.Before running the code below, please download the [example image](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_formula_rec_001.png) locally.
+
+```python
+from paddleocr import FormulaRecognition
+model = FormulaRecognition(model_name="PP-FormulaNet_plus-M")
+output = model.predict(input="general_formula_rec_001.png", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+After running, the output is:
+```bash
+{'res': {'input_path': '/root/.paddlex/predict_input/general_formula_rec_001.png', 'page_index': None, 'rec_formula': '\\zeta_{0}(\\nu)=-\\frac{\\nu\\varrho^{-2\\nu}}{\\pi}\\int_{\\mu}^{\\infty}d\\omega\\int_{C_{+}}d z\\frac{2z^{2}}{(z^{2}+\\omega^{2})^{\\nu+1}}\\breve{\\Psi}(\\omega;z)e^{i\\epsilon z}\\quad,'}}
+```
+Explanation of the result parameters:
+
+- `input_path`： Indicates the path to the input formula image to be predicted
+- `page_index`： If the input is a PDF file, this represents the page number; otherwise, it is None
+- `rec_formula`：Indicates the predicted LaTeX source code of the formula image
+The visualization image is as follows. The left side is the input formula image, and the right side is the rendered formula from the prediction:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/formula_recog/general_formula_rec_001_res_paddleocr3.png">
+
+<b>Note: If you need to visualize the formula recognition module, you must install the LaTeX rendering environment by running the following command. Currently, visualization is only supported on Ubuntu. Other environments are not supported for now. For complex formulas, the LaTeX result may contain advanced representations that may not render successfully in Markdown or similar environments:</b>
+```bash
+sudo apt-get update
+sudo apt-get install texlive texlive-latex-base texlive-xetex latex-cjk-all texlive-latex-extra -y
+```
+
+Related methods and parameter descriptions are as follows:
+
+* `FormulaRecognition` instantiates the formula recognition model (here using `PP-FormulaNet_plus-M` as an example), with detailed description as follows:
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td>Model name. If set to <code>None</code>, <code>PP-FormulaNet_plus-M</code> will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td>Model storage path.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td>Device for inference.<br/>
+<b>For example:</b><code>"cpu"</code>, <code>"gpu"</code>, <code>"npu"</code>, <code>"gpu:0"</code>, <code>"gpu:0,1"</code>.<br/>
+If multiple devices are specified, parallel inference will be performed.<br/>
+By default, GPU 0 is used if available; otherwise, CPU is used.
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, local inference uses the <code>paddle_static</code> engine by default. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td>Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td>Whether to use the Paddle Inference TensorRT subgraph engine. If the model does not support acceleration through TensorRT, setting this flag will not enable acceleration.<br/>
+For Paddle with CUDA version 11.8, the compatible TensorRT version is 8.x (x>=6), and it is recommended to install TensorRT 8.6.1.6.<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td>Computation precision when using the TensorRT subgraph engine in Paddle Inference.<br/><b>Options:</b><code>"fp32"</code>, <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+Whether to enable MKL-DNN acceleration for inference. If MKL-DNN is unavailable or the model does not support it, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td>Number of threads to use for inference on CPUs.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+</tbody>
+</table>
+
+* Call the `predict()` method of the formula recognition model to perform inference, which returns a result list.  
+Additionally, this module provides the `predict_iter()` method. Both accept the same parameters and return the same result format.  
+The difference is that `predict_iter()` returns a `generator`, which can process and retrieve results step-by-step, suitable for large datasets or memory-efficient scenarios.  
+
+You can choose either method based on your actual needs. The `predict()` method takes parameters `input` and `batch_size`, described as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td>Input data to be predicted. Required. Supports multiple input types:
+<ul>
+<li><b>Python Var</b>: e.g., <code>numpy.ndarray</code> representing image data</li>
+<li><b>str</b>: 
+  - Local image or PDF file path: <code>/root/data/img.jpg</code>;
+  - <b>URL</b> of image or PDF file: e.g., <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">example</a>;
+  - <b>Local directory</b>: directory containing images for prediction, e.g., <code>/root/data/</code> (Note: directories containing PDF files are not supported; PDFs must be specified by exact file path)</li>
+<li><b>list</b>: Elements must be of the above types, e.g., <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td>Batch size, positive integer.</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+
+* The prediction results can be processed. Each result corresponds to a `Result` object, which supports printing, saving as an image, and saving as a `json` file:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Description</th>
+<th>Parameter</th>
+<th>Type</th>
+<th>Details</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">Print the result to the terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output using <code>JSON</code> indentation</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the <code>JSON</code> output; only effective when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Controls whether non-<code>ASCII</code> characters are escaped to <code>Unicode</code>. If set to <code>True</code>, all non-ASCII characters are escaped; if <code>False</code>, original characters are kept. Only effective when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">Save the result as a json-formatted file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Path to save the file. If it is a directory, the saved file name will match the input file type</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the <code>JSON</code> output; only effective when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Controls whether non-<code>ASCII</code> characters are escaped to <code>Unicode</code>. If set to <code>True</code>, all non-ASCII characters are escaped; if <code>False</code>, original characters are kept. Only effective when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save the result as an image file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Path to save the file. If it is a directory, the saved file name will match the input file type</td>
+<td>None</td>
+</tr>
+</table>
+
+* In addition, you can also access the visualized image and prediction result via attributes, as follows:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">Get the prediction result in <code>json</code> format</td>
+</tr>
+<tr>
+<td rowspan = "1"><code>img</code></td>
+<td rowspan = "1">Get the visualized image in <code>dict</code> format</td>
+</tr>
+</table>
+
+
+## IV. Custom Development
+
+If the models above do not perform well in your scenario, you can try the following steps for custom development.  
+Here we take training `PP-FormulaNet_plus-M` as an example. For other models, just replace the corresponding config file.  First, you need to prepare a formula recognition dataset. You can follow the format of the [formula recognition demo data](https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_rec_latexocr_dataset_example.tar).  Once the data is ready, follow the steps below to train and export the model. After export, the model can be quickly integrated into the API described above.  This example uses the demo dataset. Before training the model, please ensure you have installed all PaddleOCR dependencies as described in the [installation documentation](../installation.en.md).
+
+### 4.1 Environment Setup
+
+To train the formula recognition model, you need to install additional Python and Linux dependencies. Run the following commands:
+
+```shell
+sudo apt-get update
+sudo apt-get install libmagickwand-dev
+pip install tokenizers==0.19.1 imagesize ftfy Wand
+```
+
+### 4.2 Dataset and Pretrained Model Preparation
+
+#### 4.2.1 Prepare the Dataset
+```shell
+# Download the demo dataset
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_rec_latexocr_dataset_example.tar
+tar -xf ocr_rec_latexocr_dataset_example.tar
+```
+
+#### 4.2.2 Download the Pretrained Model
+```shell
+# Download the PP-FormulaNet_plus-M pre-trained model
+wget https://paddleocr.bj.bcebos.com/contribution/rec_ppformulanet_plus_m_train.tar 
+tar -xf rec_ppformulanet_plus_m_train.tar
+```
+
+### 4.3 Model Training
+PaddleOCR is modularized. To train the `PP-FormulaNet_plus-M`  model, you need to use its [config file](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/configs/rec/PP-FormuaNet/PP-FormulaNet_plus-M.yaml).
+
+Training commands are as follows:
+```bash
+# Single GPU training (default)
+python3 tools/train.py -c configs/rec/PP-FormuaNet/PP-FormulaNet_plus-M.yaml \
+   -o Global.pretrained_model=./rec_ppformulanet_plus_m_train/best_accuracy.pdparams
+
+# Multi-GPU training, specify GPU IDs with --gpus
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/rec/PP-FormuaNet/PP-FormulaNet_plus-M.yaml \
+   -o Global.pretrained_model=./rec_ppformulanet_plus_m_train/best_accuracy.pdparams
+
+```
+**Note:**
+
+- By default, evaluation is performed every 1 epoch.If you change the batch size or dataset, modify the following accordingly:
+```bash
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py -c configs/rec/PP-FormuaNet/PP-FormulaNet_plus-M.yaml \
+  -o Global.eval_batch_step=[0,{length_of_dataset//batch_size//4}] \
+   Global.pretrained_model=./rec_ppformulanet_plus_m_train/best_accuracy.pdparams
+```
+
+### 4.4 Model Evaluation
+You can evaluate trained weights, e.g., output/xxx/xxx.pdparams, or use the downloaded [model](https://paddleocr.bj.bcebos.com/contribution/rec_ppformulanet_plus_m_train.tar ) with the following command:
+
+```bash
+# Make sure pretrained_model is set to the local path.
+# For custom-trained models, modify the path and file name as {path/to/weights}/{model_name}
+# Demo test set evaluation
+python3 tools/eval.py -c configs/rec/PP-FormuaNet/PP-FormulaNet_plus-M.yaml -o \
+Global.pretrained_model=./rec_ppformulanet_plus_m_train/best_accuracy.pdparams
+```
+
+### 4.5 Model Export
+```bash
+ python3 tools/export_model.py -c configs/rec/PP-FormuaNet/PP-FormulaNet_plus-M.yaml -o \
+ Global.pretrained_model=./rec_ppformulanet_plus_m_train/best_accuracy.pdparams \
+ Global.save_inference_dir="./PP-FormulaNet_plus-M_infer/"
+```
+
+After exporting, the static graph model will be saved in `./PP-FormulaNet_plus-M_infer/`, and you will see the following files:
+ ```
+ ./PP-FormulaNet_plus-M_infer/
+ ├── inference.json
+ ├── inference.pdiparams
+ ├── inference.yml
+ ```
+At this point, the secondary development is complete. This static graph model can be directly integrated into the PaddleOCR API.
+
+## V. FAQ
+
+**Q1: Which formula recognition model does PaddleOCR recommend?**
+
+A1: It is recommended to use the PP-FormulaNet series.
+If your scenario is mainly in English and inference speed is not a concern, use PP-FormulaNet-L or PP-FormulaNet_plus-L.
+For mainly Chinese use cases, use PP-FormulaNet_plus-L or PP-FormulaNet_plus-M.
+If your device has limited computing power and you are working with English formulas, use PP-FormulaNet-S.
+
+**Q2: Why does the inference report an error?**
+A2: The formula recognition model depends heavily on Paddle 3.0 official release.
+Please ensure the correct version is installed.
+
+**Q3: Why is there no visualization image after prediction?**
+A3: This may be due to LaTeX not being installed.You need to refer to Section III and install the LaTeX rendering tools.
diff --git a/docs/version3.x/module_usage/formula_recognition.md b/docs/version3.x/module_usage/formula_recognition.md
new file mode 100644
index 0000000..11ec179
--- /dev/null
+++ b/docs/version3.x/module_usage/formula_recognition.md
@@ -0,0 +1,494 @@
+---
+comments: true
+---
+
+# 公式识别模块使用教程
+
+## 一、概述
+
+公式识别模块是OCR（光学字符识别）系统中的关键组成部分，负责将图像中的数学公式转换为可编辑的文本或计算机可识别的格式。该模块的性能直接影响到整个OCR系统的准确性和效率。公式识别模块通常会输出数学公式的 LaTeX 或 MathML 代码，这些代码将作为输入传递给文本理解模块进行后续处理。
+
+## 二、支持模型列表
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。表格中的“常规模式”耗时对应本地 <code>paddle_static</code> 推理引擎。
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>En-BLEU(%)</th>
+<th>Zh-BLEU(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>UniMERNet</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UniMERNet_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UniMERNet_pretrained.pdparams">训练模型</a></td>
+<td>85.91</td>
+<td>43.50</td>
+<td>1311.84 / 1311.84</td>
+<td>- / 8288.07</td>
+<td>1530</td>
+<td>UniMERNet是由上海AI Lab研发的一款公式识别模型。该模型采用Donut Swin作为编码器，MBartDecoder作为解码器，并通过在包含简单公式、复杂公式、扫描捕捉公式和手写公式在内的一百万数据集上进行训练，大幅提升了模型对真实场景公式的识别准确率</td>
+</tr>
+<td>PP-FormulaNet-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-S_pretrained.pdparams">训练模型</a></td>
+<td>87.00</td>
+<td>45.71</td>
+<td>182.25 / 182.25</td>
+<td>- / 254.39</td>
+<td>224</td>
+<td rowspan="2">PP-FormulaNet 是由百度飞桨视觉团队开发的一款先进的公式识别模型，支持5万个常见LateX源码词汇的识别。PP-FormulaNet-S 版本采用了 PP-HGNetV2-B4 作为其骨干网络，通过并行掩码和模型蒸馏等技术，大幅提升了模型的推理速度，同时保持了较高的识别精度，适用于简单印刷公式、跨行简单印刷公式等场景。而 PP-FormulaNet-L 版本则基于 Vary_VIT_B 作为骨干网络，并在大规模公式数据集上进行了深入训练，在复杂公式的识别方面，相较于PP-FormulaNet-S表现出显著的提升，适用于简单印刷公式、复杂印刷公式、手写公式等场景。 </td>
+
+</tr>
+<td>PP-FormulaNet-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-L_pretrained.pdparams">训练模型</a></td>
+<td>90.36</td>
+<td>45.78</td>
+<td>1482.03 / 1482.03</td>
+<td>- / 3131.54</td>
+<td>695</td>
+</tr>
+<td>PP-FormulaNet_plus-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-S_pretrained.pdparams">训练模型</a></td>
+<td>88.71</td>
+<td>53.32</td>
+<td>179.20 / 179.20</td>
+<td>- / 260.99</td>
+<td>248</td>
+<td rowspan="3">PP-FormulaNet_plus 是百度飞桨视觉团队在 PP-FormulaNet 的基础上开发的增强版公式识别模型。与原版相比，PP-FormulaNet_plus 在训练中使用了更为丰富的公式数据集，包括中文学位论文、专业书籍、教材试卷以及数学期刊等多种来源。这一扩展显著提升了模型的识别能力。
+
+其中，PP-FormulaNet_plus-M 和 PP-FormulaNet_plus-L 模型新增了对中文公式的支持，并将公式的最大预测 token 数从 1024 扩大至 2560，大幅提升了对复杂公式的识别性能。同时，PP-FormulaNet_plus-S 模型则专注于增强英文公式的识别能力。通过这些改进，PP-FormulaNet_plus 系列模型在处理复杂多样的公式识别任务时表现更加出色。 </td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-M_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-M_pretrained.pdparams">训练模型</a></td>
+<td>91.45</td>
+<td>89.76</td>
+<td>1040.27 / 1040.27</td>
+<td>- / 1615.80</td>
+<td>592</td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-L_pretrained.pdparams">训练模型</a></td>
+<td>92.22</td>
+<td>90.64</td>
+<td>1476.07 / 1476.07</td>
+<td>- / 3125.58</td>
+<td>698</td>
+</tr>
+<tr>
+<td>LaTeX_OCR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/LaTeX_OCR_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/LaTeX_OCR_rec_pretrained.pdparams">训练模型</a></td>
+<td>74.55</td>
+<td>39.96</td>
+<td>1088.89 / 1088.89</td>
+<td>- / -</td>
+<td>99</td>
+<td>LaTeX-OCR是一种基于自回归大模型的公式识别算法，通过采用 Hybrid ViT 作为骨干网络，transformer作为解码器，显著提升了公式识别的准确性。</td>
+</tr>
+</table>
+ <strong>测试环境说明:</strong>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+              <li><strong>测试数据集：</strong>PaddleOCR 内部自建公式识别测试集</li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+
+
+## 三、快速开始
+
+> ❗ 在快速开始前，请先安装 PaddleOCR 的 wheel 包，详细请参考 [安装教程](../installation.md)。
+
+使用一行命令即可快速体验：
+
+```bash
+paddleocr formula_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_formula_rec_001.png
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+<b>注：</b>PaddleOCR 官方模型默认从 HuggingFace 获取，如运行环境访问 HuggingFace 不便，可通过环境变量修改模型源为 BOS：`PADDLE_PDX_MODEL_SOURCE="BOS"`，未来将支持更多主流模型源；
+
+您也可以将公式识别的模块中的模型推理集成到您的项目中。运行以下代码前，请您下载[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_formula_rec_001.png)到本地。
+
+```python
+from paddleocr import FormulaRecognition
+model = FormulaRecognition(model_name="PP-FormulaNet_plus-M")
+output = model.predict(input="general_formula_rec_001.png", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+运行后，得到的结果为：
+
+```bash
+{'res': {'input_path': '/root/.paddlex/predict_input/general_formula_rec_001.png', 'page_index': None, 'rec_formula': '\\zeta_{0}(\\nu)=-\\frac{\\nu\\varrho^{-2\\nu}}{\\pi}\\int_{\\mu}^{\\infty}d\\omega\\int_{C_{+}}d z\\frac{2z^{2}}{(z^{2}+\\omega^{2})^{\\nu+1}}\\breve{\\Psi}(\\omega;z)e^{i\\epsilon z}\\quad,'}}
+```
+
+运行结果参数含义如下：
+- `input_path`：表示输入待预测公式图像的路径
+- `page_index`：如果输入是PDF文件，则表示当前是PDF的第几页，否则为 `None`
+- `rec_formula`：表示公式图像的预测LaTeX源码
+
+
+可视化图片如下，左侧是待预测的公式图像，右边是预测的结果渲染后的公式图像：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/formula_recog/general_formula_rec_001_res_paddleocr3.png">
+
+<b> 注：如果您需要对公式识别模块进行可视化，需要运行如下命令来对LaTeX渲染环境进行安装。目前公式识别模块可视化只支持Ubuntu环境，其他环境暂不支持。对于复杂公式，LaTeX 结果可能包含部分高级的表示，Markdown等环境中未必可以成功显示：</b>
+```bash
+sudo apt-get update
+sudo apt-get install texlive texlive-latex-base texlive-xetex latex-cjk-all texlive-latex-extra -y
+```
+
+相关方法、参数等说明如下：
+
+* `FormulaRecognition`实例化公式识别模型（此处以`PP-FormulaNet_plus-M`为例），具体说明如下：
+<table><thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr></thead><tbody>
+<tr>
+<td><code>model_name</code></td>
+<td>模型名称。如果设置为<code>None</code>，则使用<code>PP-FormulaNet_plus-M</code>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td>模型存储路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td>用于推理的设备。<br/>
+<b>例如：</b><code>"cpu"</code>、<code>"gpu"</code>、<code>"npu"</code>、<code>"gpu:0"</code>、<code>"gpu:0,1"</code>。<br/>
+如指定多个设备，将进行并行推理。<br/>
+默认情况下，优先使用 GPU 0；若不可用则使用 CPU。
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，本地推理默认使用 <code>paddle_static</code> 引擎。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td>是否启用 Paddle Inference 的 TensorRT 子图引擎。如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td>当使用 Paddle Inference 的 TensorRT 子图引擎时设置的计算精度。<br/><b>可选项：</b><code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+是否启用 MKL-DNN 加速推理。如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td>在 CPU 上推理时使用的线程数量。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+</tbody>
+</table>
+
+* 调用公式识别模型的 `predict()` 方法进行推理预测，该方法会返回一个结果列表。另外，本模块还提供了 `predict_iter()` 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 `predict_iter()` 返回的是一个 `generator`，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。`predict()` 方法参数有 `input` 和 `batch_size`，具体说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td>待预测数据，支持多种输入类型，必填。
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td>批大小，可设置为任意正整数。</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+
+* 对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为`json`文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将结果保存为图像格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+</table>
+
+* 此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">获取预测的<code>json</code>格式的结果</td>
+</tr>
+<tr>
+<td rowspan = "1"><code>img</code></td>
+<td rowspan = "1">获取格式为<code>dict</code>的可视化图像</td>
+</tr>
+
+</table>
+
+
+## 四、二次开发
+如果以上模型在您的场景下效果仍然不理想，您可以尝试以下步骤进行二次开发，此处以训练 `PP-FormulaNet-S` 举例，其他模型替换对应配置文件即可。首先，您需要准备公式识别的数据集，可以参考[公式识别 Demo 数据](https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_rec_latexocr_dataset_example.tar)的格式准备，准备好后，即可按照以下步骤进行模型训练和导出，导出后，可以将模型快速集成到上述API中。此处以公式识别 Demo 数据示例。在训练模型之前，请确保已经按照[安装文档](../installation.md)安装了 PaddleOCR 所需要的依赖。
+
+### 4.1 环境配置
+
+训练公式识别模型需要安装额外的Python依赖和linux依赖，执行如下命令安装：
+```shell
+sudo apt-get update
+sudo apt-get install libmagickwand-dev
+pip install tokenizers==0.19.1 imagesize ftfy Wand
+```
+
+### 4.2 数据集、预训练模型准备
+
+#### 4.2.1 准备数据集
+
+```shell
+# 下载示例数据集
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_rec_latexocr_dataset_example.tar
+tar -xf ocr_rec_latexocr_dataset_example.tar
+```
+
+#### 4.2.2 下载预训练模型
+
+```shell
+# 下载 PP-FormulaNet_plus-M 预训练模型
+wget https://paddleocr.bj.bcebos.com/contribution/rec_ppformulanet_plus_m_train.tar 
+tar -xf rec_ppformulanet_plus_m_train.tar
+```
+
+### 4.3 模型训练
+
+PaddleOCR对代码进行了模块化，训练 `PP-FormulaNet_plus-M` 识别模型时需要使用 `PP-FormulaNet_plus-M` 的[配置文件](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/configs/rec/PP-FormuaNet/PP-FormulaNet_plus-M.yaml)。
+
+
+训练命令如下：
+
+```bash
+#单卡训练 (默认训练方式)
+python3 tools/train.py -c configs/rec/PP-FormuaNet/PP-FormulaNet_plus-M.yaml \
+   -o Global.pretrained_model=./rec_ppformulanet_plus_m_train/best_accuracy.pdparams
+#多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/PP-FormuaNet/PP-FormulaNet_plus-M.yaml \
+        -o Global.pretrained_model=./rec_ppformulanet_plus_m_train/best_accuracy.pdparams
+```
+
+**注意：**
+
+- 默认每训练 1个 epoch 进行 1 次评估，若您更改训练的 batch_size，或更换数据集，请在训练时作出如下修改
+```bash
+python3  -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/PP-FormuaNet/PP-FormulaNet_plus-M.yaml \
+  -o Global.eval_batch_step=[0,{length_of_dataset//batch_size//4}] \
+   Global.pretrained_model=./rec_ppformulanet_plus_m_train/best_accuracy.pdparams
+```
+
+### 4.4 模型评估
+
+您可以评估已经训练好的权重，如，`output/xxx/xxx.pdparams`，也可以使用已经下载的[模型文件](https://paddleocr.bj.bcebos.com/contribution/rec_ppformulanet_s_train.tar)，使用如下命令进行评估：
+
+```bash
+
+#注意将pretrained_model的路径设置为本地路径。若使用自行训练保存的模型，请注意修改路径和文件名为{path/to/weights}/{model_name}。
+#demo 测试集评估
+ python3 tools/eval.py -c configs/rec/PP-FormuaNet/PP-FormulaNet_plus-M.yaml -o \
+ Global.pretrained_model=./rec_ppformulanet_plus_m_train/best_accuracy.pdparams
+```
+
+### 4.5 模型导出
+
+```bash
+python3 tools/export_model.py -c configs/rec/PP-FormuaNet/PP-FormulaNet_plus-M.yaml -o \
+Global.pretrained_model=./rec_ppformulanet_plus_m_train/best_accuracy.pdparams \
+Global.save_inference_dir="./PP-FormulaNet_plus-M_infer/"
+```
+
+ 导出模型后，静态图模型会存放于当前目录的`./PP-FormulaNet_plus-M_infer/`中，在该目录下，您将看到如下文件：
+ ```
+ ./PP-FormulaNet_plus-M_infer/
+ ├── inference.json
+ ├── inference.pdiparams
+ ├── inference.yml
+ ```
+至此，二次开发完成，该静态图模型可以直接集成到 PaddleOCR 的 API 中。
+
+## 五、FAQ
+
+**Q1: PaddleOCR 更推荐哪个公式识别模型？**
+
+A1: 更推荐使用 PP-FormulaNet 系列模型，如果是英文场景居多且不考虑推理耗时，则可以使用 PP-FormulaNet-L 或者 PP-FormulaNet_plus-L 模型，如果中文场景居多，则可以使用 PP-FormulaNet_plus-L 或者 PP-FormulaNet_plus-M，如果推理设备算力有限且是英文场景，则可以使用 PP-FormulaNet-S。
+
+**Q2: 为什么推理报错？**
+
+A2: 公式识别模型的推理强依赖于 Paddle 框架 3.0 正式版，请确保版本一致。
+
+
+**Q3: 为什么预测后没有可视化图像？**
+
+A3: 可能是因为没有安装LaTeX导致，您需要参考第三节安装LaTeX渲染工具。
diff --git a/docs/version3.x/module_usage/layout_analysis.en.md b/docs/version3.x/module_usage/layout_analysis.en.md
new file mode 100644
index 0000000..dc33223
--- /dev/null
+++ b/docs/version3.x/module_usage/layout_analysis.en.md
@@ -0,0 +1,618 @@
+---
+
+comments: true
+
+---
+
+
+
+# Layout Analysis Module User Guide
+
+
+
+## 1. Overview
+
+
+
+Layout analysis is a crucial component in document parsing systems. Its goal is to parse the overall layout of a document, accurately detecting various elements it contains (such as text paragraphs, headings, images, tables, formulas, etc.), and restoring the correct reading order of these elements. The performance of this module directly impacts the overall accuracy and usability of the document parsing system.
+
+
+
+## 2. Supported Model List
+
+
+
+Currently, this module only supports the PP-DocLayoutV2 model. Structurally, PP-DocLayoutV2 is based on the layout detection model [PP-DocLayout_plus-L](./layout_detection.en.md) (built upon the RT-DETR-L model) and cascades a lightweight pointer network with 6 Transformer layers. The PP-DocLayout_plus-L component continues to perform layout detection, identifying different elements in document images (such as text, charts, images, formulas, paragraphs, abstracts, references, etc.), classifying them into predefined categories, and determining their positions within the document. The detected bounding boxes and class labels are then fed into the subsequent pointer network to sort the layout elements and obtain the correct reading order.
+
+
+
+<div align="center">
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/methods/PP-DocLayoutV2.png" width="600"/>
+
+</div>
+
+
+
+As shown in the figure above, PP-DocLayoutV2 embeds the targets detected by RT-DETR using absolute 2D positional encoding and class labels. Additionally, the pointer network’s attention mechanism incorporates the geometric bias mechanism from Relation-DETR to explicitly model pairwise geometric relationships between elements. The pairwise relation head linearly projects element representations into query and key vectors, then computes bilinear similarities to generate pairwise logits, ultimately producing an N×N matrix representing the relative order between every pair of elements. Finally, a deterministic "win-accumulation" decoding algorithm restores a topologically consistent reading order for the detected layout elements.
+
+
+
+The following table only presents the layout detection accuracy of PP-DocLayoutV2. The evaluation dataset is a self-built layout region detection dataset, containing 1,000 images of various document types such as Chinese and English papers, magazines, newspapers, research reports, PPTs, exam papers, textbooks, etc., and covering 25 common layout element categories: document title, section header, text, vertical text, page number, abstract, table of contents, references, footnote, image caption, header, footer, header image, footer image, algorithm, inline formula, display formula, formula number, image, table, figure title (figure title, table title, chart title), seal, chart, aside text, and reference content.
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing. The "Standard" values correspond to the local <code>paddle_static</code> inference engine.
+
+<table>
+
+<thead>
+
+<tr>
+
+<th>Model</th><th>Model Download Link</th>
+
+<th>mAP(0.5) (%)</th>
+
+<th>GPU Inference Time (ms)<br/>[Standard / High-Performance Mode]</th>
+
+<th>CPU Inference Time (ms)<br/>[Standard / High-Performance Mode]</th>
+
+<th>Model Size (MB)</th>
+
+<th>Description</th>
+
+</tr>
+
+</thead>
+
+<tbody>
+
+<tr>
+
+<td>PP-DocLayoutV2</td>
+
+<td><a href="https://huggingface.co/PaddlePaddle/PP-DocLayoutV2">Inference Model</a></td>
+
+<td>81.4</td>
+
+<td> - </td>
+
+<td> - </td>
+
+<td>203.8</td>
+
+<td>In-house layout analysis model trained on a diverse self-built dataset including Chinese and English academic papers, multi-column magazines, newspapers, PPTs, contracts, books, exam papers, research reports, ancient books, Japanese documents, and vertical text documents. Provides high-precision layout region localization and reading order recovery.</td>
+
+</tr>
+
+<tr>
+
+</tbody>
+
+</table>
+
+## 3. Quick Integration  <a id="quick"> </a>
+
+> ❗ Before quick integration, please install the PaddleOCR wheel package. For detailed instructions, refer to [PaddleOCR Local Installation Tutorial](../installation.en.md)。
+
+You can quickly try it out with a single command:
+
+```bash
+paddleocr layout_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg --model_name PP-DocLayoutV3
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following command:
+
+```bash
+paddleocr layout_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg --model_name PP-DocLayoutV3 \
+    --engine transformers
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following command:
+
+```bash
+paddleocr layout_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg --model_name PP-DocLayoutV3 \
+    --engine onnxruntime
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+<b>Note: </b>The official models would be download from HuggingFace by default. If can't access to HuggingFace, please set the environment variable <code>PADDLE_PDX_MODEL_SOURCE="BOS"</code> to change the model source to BOS. In the future, more model sources will be supported.
+
+You can also integrate the model inference from the layout area detection module into your project. Before running the following code, please download [Example Image](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg) Go to the local area.
+
+```python
+from paddleocr import LayoutDetection
+
+model = LayoutDetection(model_name="PP-DocLayoutV2")
+output = model.predict("layout.jpg", batch_size=1, layout_nms=True)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following code:
+
+```python
+from paddleocr import LayoutDetection
+
+model = LayoutDetection(
+    model_name="PP-DocLayoutV2",
+    engine="transformers",
+)
+output = model.predict("layout.jpg", batch_size=1, layout_nms=True)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following code:
+
+```python
+from paddleocr import LayoutDetection
+
+model = LayoutDetection(
+    model_name="PP-DocLayoutV2",
+    engine="onnxruntime",
+)
+output = model.predict("layout.jpg", batch_size=1, layout_nms=True)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+If you want to use the trained model with the `paddle_dynamic` or `transformers` engine, refer to the [Weight Conversion](#42-weight-conversion) section in the [Inference Engine](#4-inference-engine) section below to convert the model from the `pdparams` format to the `safetensors` format using PaddleX.
+
+After running, the result obtained is:
+
+```bash
+{'res': {'input_path': 'layout.jpg', 'page_index': None, 'boxes': [{'cls_id': 7, 'label': 'figure_title', 'score': 0.9764086604118347, 'coordinate': [34.227077, 19.217348, 362.48834, 80.243195]}, {'cls_id': 21, 'label': 'table', 'score': 0.9881672263145447, 'coordinate': [73.759026, 105.72034, 322.67398, 298.82642]}, {'cls_id': 17, 'label': 'paragraph_title', 'score': 0.9405199885368347, 'coordinate': [35.089825, 330.67575, 144.06467, 346.7406]}, {'cls_id': 22, 'label': 'text', 'score': 0.9889925122261047, 'coordinate': [33.540825, 349.37204, 363.5848, 615.04504]}, {'cls_id': 17, 'label': 'paragraph_title', 'score': 0.9430180788040161, 'coordinate': [35.032074, 627.2185, 188.24695, 643.66693]}, {'cls_id': 22, 'label': 'text', 'score': 0.988863468170166, 'coordinate': [33.72388, 646.5885, 363.05005, 852.60236]}, {'cls_id': 7, 'label': 'figure_title', 'score': 0.9749509692192078, 'coordinate': [385.14465, 19.341825, 715.4594, 78.739395]}, {'cls_id': 21, 'label': 'table', 'score': 0.9876241683959961, 'coordinate': [436.68512, 105.67539, 664.1041, 314.0018]}, {'cls_id': 22, 'label': 'text', 'score': 0.9878363013267517, 'coordinate': [385.06494, 345.74847, 715.173, 463.18677]}, {'cls_id': 17, 'label': 'paragraph_title', 'score': 0.951842725276947, 'coordinate': [386.3095, 476.34277, 703.0732, 493.00302]}, {'cls_id': 22, 'label': 'text', 'score': 0.9884033799171448, 'coordinate': [385.1237, 496.70123, 716.09717, 702.33386]}, {'cls_id': 17, 'label': 'paragraph_title', 'score': 0.940105676651001, 'coordinate': [386.66754, 715.5732, 527.5737, 731.77826]}, {'cls_id': 22, 'label': 'text', 'score': 0.9876392483711243, 'coordinate': [384.9688, 734.457, 715.63086, 853.71]}]}}
+```
+
+The meanings of the parameters are as follows:
+<ul>
+<li><code>input_path</code>：The path to the input image for prediction.</li>
+<li><code>page_index</code>：If the input is a PDF file, it indicates which page of the PDF it is; otherwise, it is <code>None</code>.</li>
+<li><code>boxes</code>：Predicted layout element information sorted in reading order, represented as a list of dictionaries. According to the reading order, each dictionary represents a detected layout element and contains the following information:
+    <ol start="1" type="1">
+        <li><code>cls_id</code>：Class ID, an integer.</li>
+        <li><code>label</code>：Class label, a string.</li>
+        <li><code>score</code>：Confidence score of the bounding box, a float.</li>
+        <li><code>coordinate</code>：Coordinates of the bounding box, a list of floats in the format <code>[xmin, ymin, xmax, ymax]</code></li>
+    </ol>
+</li>
+</ul>
+
+
+Relevant methods, parameters, and explanations are as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>Meaning:</b> Model name. <br/>
+<b>Description:</b>
+If set to <code>None</code>, <code>PP-DocLayout-L</code> will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>Meaning:</b>Model storage path.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b> Device for inference. <br/>
+<b>Description:</b>
+If set to <code>None</code>, <code>"cpu"</code> will be used. <br/>
+<b>For example:</b> <code>"cpu"</code>, <code>"gpu"</code>, <code>"npu"</code>, <code>"gpu:0"</code>, <code>"gpu:0,1"</code>.<br/>
+If multiple devices are specified, parallel inference will be performed.<br/>
+By default, GPU 0 is used if available; otherwise, CPU is used.
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, <code>transformers</code>, and <code>onnxruntime</code>. When left as <code>None</code>, local inference uses the <code>paddle_static</code> engine by default. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b>Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b>Whether to use the Paddle Inference TensorRT subgraph engine. <br/>
+<b>Description:</b>
+If the model does not support acceleration through TensorRT, setting this flag will not enable acceleration.<br/>
+For Paddle with CUDA version 11.8, the compatible TensorRT version is 8.x (x>=6), and it is recommended to install TensorRT 8.6.1.6.<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b>Computation precision when using the TensorRT subgraph engine in Paddle Inference. <br/>
+<b>Description:</b>
+<b>Options:</b> <code>"fp32"</code>, <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+<b>Meaning:</b>Whether to enable MKL-DNN acceleration for inference. <br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support it, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b>MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b>Number of threads to use for inference on CPUs.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>img_size</code></td>
+<td><b>Meaning:</b>Input image size.<br/>
+<b>Description:</b>
+<ul>
+<li><b>int</b>: e.g. <code>640</code>, resizes input image to 640x640.</li>
+<li><b>list</b>: e.g. <code>[640, 512]</code>, resizes input image to width 640 and height 512.</li>
+</ul>
+</td>
+<td><code>int|list|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>threshold</code></td>
+<td><b>Meaning:</b>Threshold for filtering low-confidence predictions.<br/>
+<b>Description:</b>
+<ul>
+<li><b>float</b>: e.g. <code>0.2</code>, filters out all boxes with confidence below 0.2.</li>
+<li><b>dict</b>: The key is <code>int</code> (class id), the value is <code>float</code> (threshold). For example, <code>{0: 0.45, 2: 0.48, 7: 0.4}</code> means class 0 uses threshold 0.45, class 2 uses 0.48, class 7 uses 0.4.</li>
+<li><b>None</b>: uses the model's default configuration.</li>
+</ul>
+</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Whether to use NMS post-processing to filter overlapping boxes.<br/>
+<b>Description:</b>
+<ul>
+<li><b>bool</b>: whether to use NMS for post-processing to filter overlapping boxes.</li>
+<li><b>None</b>: uses the model's default configuration.</li>
+</ul>
+</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Scaling factor for the side length of the detection box.<br/>
+<b>Description:</b>
+<ul>
+<li><b>float</b>: A float greater than 0, e.g. <code>1.1</code>, expands width and height by 1.1 times.</li>
+<li><b>list</b>: e.g. <code>[1.2, 1.5]</code>, expands width by 1.2x and height by 1.5x.</li>
+<li><b>dict</b>: The key is <code>int</code> (class id), the value is <code>tuple</code> of two floats (width ratio, height ratio). For example, <code>{0: (1.1, 2.0)}</code> means for class 0, width is expanded by 1.1x and height by 2.0x.</li>
+<li><b>None</b>: uses the model's default configuration.</li>
+</ul>
+</td>
+<td><code>float|list|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>Merge mode for model output bounding boxes.<br/>
+<b>Description:</b>
+<ul>
+<li><b>"large"</b>: Only keep the largest outer box among overlapping boxes, remove inner boxes.</li>
+<li><b>"small"</b>: Only keep the smallest inner box among overlapping boxes, remove outer boxes.</li>
+<li><b>"union"</b>: Keep all boxes, no filtering.</li>
+<li><b>dict</b>: The key is <code>int</code> (class id), the value is <code>str</code> (mode). For example, <code>{0: "large", 2: "small"}</code> means class 0 uses "large" mode, class 2 uses "small" mode.</li>
+<li><b>None</b>: Use the model's default configuration.</li>
+</ul>
+</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+* The <code>predict()</code> method of the target detection model is called for inference prediction. The parameters of the <code>predict()</code> method are <code>input</code>, <code>batch_size</code>, and <code>threshold</code>, which are explained as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Input data to be predicted. Required.<br/>
+ <b>Description:</b>
+ Supports multiple input types:<ul>
+<li><b>Python Var</b>: e.g., <code>numpy.ndarray</code> representing image data</li>
+<li><b>str</b>: 
+  <ul>
+  <li>Local image or PDF file path: <code>/root/data/img.jpg</code>;</li>
+  <li><b>URL</b> of image or PDF file: e.g., <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">example</a>;</li>
+  <li><b>Local directory</b>: directory containing images for prediction, e.g., <code>/root/data/</code> (Note: directories containing PDF files are not supported; PDFs must be specified by exact file path)</li>
+  </ul>
+<li><b>list</b>: Elements must be of the above types, e.g., <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>Meaning:</b>Batch size<br/>
+<b>Description:</b>
+Positive integer.</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+<tr>
+<td><code>threshold</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|dict|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+ If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|list|dict|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>str|dict|None</code></td>
+<td>None</td>
+</tr>
+</tbody>
+</table>
+
+
+* Process the prediction results, with each sample's prediction result being the corresponding Result object, and supporting operations such as printing, saving as an image, and saving as a <code>json</code> file:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Method Description</th>
+<th>Parameters</th>
+<th>Parameter type</th>
+<th>Parameter Description</th>
+<th>Default value</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">Print the result to the terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Do you want to use <code>JSON</code> indentation formatting for the output content</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to enhance the readability of the <code>JSON</code> data output, only valid when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non ASCII characters to Unicode characters. When set to <code>True</code>, all non ASCII </code>characters will be escaped; <code>False</code> preserves the original characters and is only valid when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">Save the result as a JSON format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The saved file path, when it is a directory, the name of the saved file is consistent with the name of the input file type</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to enhance the readability of the <code>JSON</code> data output, only valid when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non ASCII characters to Unicode characters. When set to <code>True</code>, all non <code>ASCII</code> characters will be escaped; <code>False</code> preserves the original characters and is only valid when<code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save the results as an image format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The saved file path, when it is a directory, the name of the saved file is consistent with the name of the input file type</td>
+<td>None</td>
+</tr>
+</table>
+
+
+* Additionally, it also supports obtaining the visualized image with results and the prediction results via attributes, as follows:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">Get the prediction result in <code>json</code> format</td>
+</tr>
+<tr>
+<td rowspan="1"><code>img</code></td>
+<td rowspan="1">Get the visualized image in <code>dict</code> format</td>
+</tr>
+</table>
+
+## 4. Inference Engine
+
+For detailed descriptions, values, compatibility rules, and examples of the inference engine, please refer to <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration Description</a>.
+
+### 4.1 Speed Data
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">PP-DocLayoutV3</td>
+            <td>paddle_static</td>
+            <td>10.95</td>
+            <td>47.99</td>
+            <td>12.97</td>
+            <td>72.33</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>11.33</td>
+            <td>84.48</td>
+            <td>1.31</td>
+            <td>98.01</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>16.94</td>
+            <td>47.11</td>
+            <td>13.83</td>
+            <td>78.97</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>9.26</td>
+            <td>39.11</td>
+            <td>10.23</td>
+            <td>58.80</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-DocLayoutV2</td>
+            <td>paddle_static</td>
+            <td>10.48</td>
+            <td>30.94</td>
+            <td>1.33</td>
+            <td>42.93</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>11.07</td>
+            <td>86.38</td>
+            <td>1.33</td>
+            <td>99.80</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>16.76</td>
+            <td>49.08</td>
+            <td>2.43</td>
+            <td>69.30</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>9.25</td>
+            <td>15.13</td>
+            <td>1.25</td>
+            <td>25.82</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>Test Environment Description:</strong>
+<ul>
+    <li><strong>Test Data:</strong> <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg">Sample Image</a></li>
+    <li><strong>Hardware Configuration:</strong>
+        <ul>
+            <li>GPU: NVIDIA A100 40G</li>
+            <li>CPU: Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>Software Environment:</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 4.2 Weight Conversion
+
+When using the inference engine, the system will automatically download the official pre-trained model. If you need to use a self-trained model with the `paddle_dynamic` or `transformers` engine, please refer to the [PaddleX Layout Analysis Module Weight Conversion](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/layout_analysis.html#442) section to convert the model from the `pdparams` format to the `safetensors` format using PaddleX. This allows seamless integration into the PaddleOCR API for inference. If you need to use a self-trained model with the `onnxruntime` engine, refer to [PaddleX Obtain ONNX Models](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html) to obtain the ONNX model, so it can be seamlessly integrated into the PaddleOCR API for inference.
diff --git a/docs/version3.x/module_usage/layout_analysis.md b/docs/version3.x/module_usage/layout_analysis.md
new file mode 100644
index 0000000..75b37a9
--- /dev/null
+++ b/docs/version3.x/module_usage/layout_analysis.md
@@ -0,0 +1,567 @@
+---
+comments: true
+---
+
+# 版面分析模块使用教程
+
+## 一、概述
+
+版面分析是文档解析系统中的重要组成环节，目标是对文档的整体布局进行解析，准确检测出其中所包含的各种元素（例如，文本段落、标题、图片、表格、公式等），并恢复这些元素正确的阅读顺序。该模块的性能直接影响到整个文档解析系统的准确性和可用性。
+
+
+## 二、支持模型列表
+
+该模块目前仅支持 PP-DocLayoutV2 一个模型。模型结构上，PP-DocLayoutV2 是在版面检测模型 [PP-DocLayout_plus-L](./layout_detection.md)（基于 RT-DETR-L 模型） 的基础上级联一个含 6 层 Transformer 层的轻量级指针网络（Pointer network）组成，原先 PP-DocLayout_plus-L 部分继续用于版面检测，识别文档图像中的不同元素（如文字、图表、图像、公式、段落、摘要、参考文献等），将其归类为预定义的类别并确定这些区域在文档中的位置。检测到的边界框和类别标签作为后续的指针网络的输入来对版面元素进行排序从而得到正确的阅读顺序。
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr_vl/methods/PP-DocLayoutV2.png" width="600"/>
+</div>
+
+具体如上图所示，PP-DocLayoutV2 对来自 RT-DETR 检出的目标利用绝对二维位置编码和类别标签进行嵌入表示。此外，指针网络的注意力机制融合了 Relation-DETR中的几何偏置机制，以显式地建模元素之间的成对几何关系。成对关系头（pairwise relation head）将元素表示线性投影为查询（query）向量和键（key）向量，然后计算双线性相似度以生成成对的 logits，最终得到一个表示每对元素之间相对顺序的 N×N 矩阵。最后，一种确定性的“胜者累积”（win-accumulation）解码算法会为检测到的版面元素恢复出一个拓扑一致的阅读顺序。
+
+下表仅给出 PP-DocLayoutV2 的版面检测精度。该精度指标的评估数据集是自建的版面区域检测数据集，包含了中英文论文、杂志、报纸、研报、PPT、试卷、课本等 1000 张文档类型图片，包含 25 类常见的版面元素：文档标题、段落标题、文本、竖排文本、页码、摘要、目录、参考文献、脚注、图像脚注、页眉、页脚、页眉图像、页脚图像、算法、行内公式、行间公式、公式编号、图像、表格、图和表标题（图标题、表格标题和图表标题）、印章、图表、侧栏文本和参考文献内容。
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。表格中的“常规模式”耗时对应本地 <code>paddle_static</code> 推理引擎。
+
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayoutV2</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayoutV2_infer.tar">推理模型</a></td>
+<td>81.4</td>
+<td> - </td>
+<td> - </td>
+<td>203.8</td>
+<td>自研的版面分析模型在包含中英文论文、多栏杂志、报纸、PPT、合同、书本、试卷、研报、古籍、日文文档、竖版文字文档等场景的自建数据集训练的更高精度版面区域定位和版面阅读顺序恢复模型</td>
+</tr>
+<tr>
+</tbody>
+</table>
+
+
+## 三、快速开始
+
+> ❗ 在快速开始前，请先安装 PaddleOCR 的 wheel 包，详细请参考 [安装教程](../installation.md)。
+
+使用一行命令即可快速体验：
+
+```bash
+paddleocr layout_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg --model_name PP-DocLayoutV3
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下命令：
+
+```bash
+paddleocr layout_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg --model_name PP-DocLayoutV3 \
+    --engine transformers
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下命令：
+
+```bash
+paddleocr layout_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg --model_name PP-DocLayoutV3 \
+    --engine onnxruntime
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+您可以将版面区域检测模块中的模型推理集成到您的项目中。运行以下代码前，请您下载[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg)到本地。
+
+```python
+from paddleocr import LayoutDetection
+
+model = LayoutDetection(model_name="PP-DocLayoutV3")
+output = model.predict("layout.jpg", batch_size=1, layout_nms=True)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下代码：
+
+```python
+from paddleocr import LayoutDetection
+
+model = LayoutDetection(
+    model_name="PP-DocLayoutV3",
+    engine="transformers",
+)
+output = model.predict("layout.jpg", batch_size=1, layout_nms=True)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下代码：
+
+```python
+from paddleocr import LayoutDetection
+
+model = LayoutDetection(
+    model_name="PP-DocLayoutV3",
+    engine="onnxruntime",
+)
+output = model.predict("layout.jpg", batch_size=1, layout_nms=True)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+训练后的模型如果想使用 `paddle_dynamic` 或 `transformers` 引擎，请参考后文 [推理引擎](#四推理引擎) 中的 [权重转换](#42-权重转换) 部分将模型由 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式。
+
+运行后，得到的结果为：
+
+```bash
+{'res': {'input_path': 'layout.jpg', 'page_index': None, 'boxes': [{'cls_id': 7, 'label': 'figure_title', 'score': 0.9764086604118347, 'coordinate': [34.227077, 19.217348, 362.48834, 80.243195]}, {'cls_id': 21, 'label': 'table', 'score': 0.9881672263145447, 'coordinate': [73.759026, 105.72034, 322.67398, 298.82642]}, {'cls_id': 17, 'label': 'paragraph_title', 'score': 0.9405199885368347, 'coordinate': [35.089825, 330.67575, 144.06467, 346.7406]}, {'cls_id': 22, 'label': 'text', 'score': 0.9889925122261047, 'coordinate': [33.540825, 349.37204, 363.5848, 615.04504]}, {'cls_id': 17, 'label': 'paragraph_title', 'score': 0.9430180788040161, 'coordinate': [35.032074, 627.2185, 188.24695, 643.66693]}, {'cls_id': 22, 'label': 'text', 'score': 0.988863468170166, 'coordinate': [33.72388, 646.5885, 363.05005, 852.60236]}, {'cls_id': 7, 'label': 'figure_title', 'score': 0.9749509692192078, 'coordinate': [385.14465, 19.341825, 715.4594, 78.739395]}, {'cls_id': 21, 'label': 'table', 'score': 0.9876241683959961, 'coordinate': [436.68512, 105.67539, 664.1041, 314.0018]}, {'cls_id': 22, 'label': 'text', 'score': 0.9878363013267517, 'coordinate': [385.06494, 345.74847, 715.173, 463.18677]}, {'cls_id': 17, 'label': 'paragraph_title', 'score': 0.951842725276947, 'coordinate': [386.3095, 476.34277, 703.0732, 493.00302]}, {'cls_id': 22, 'label': 'text', 'score': 0.9884033799171448, 'coordinate': [385.1237, 496.70123, 716.09717, 702.33386]}, {'cls_id': 17, 'label': 'paragraph_title', 'score': 0.940105676651001, 'coordinate': [386.66754, 715.5732, 527.5737, 731.77826]}, {'cls_id': 22, 'label': 'text', 'score': 0.9876392483711243, 'coordinate': [384.9688, 734.457, 715.63086, 853.71]}]}}
+```
+
+参数含义如下：
+<ul>
+<li><code>input_path</code>：输入的待预测图像的路径</li>
+<li><code>page_index</code>：如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+<li><code>boxes</code>：预测的目标框信息，一个字典列表。每个字典代表一个检出的目标，包含以下信息：
+    <ol start="1" type="1">
+        <li><code>cls_id</code>：类别ID，一个整数</li>
+        <li><code>label</code>：类别标签，一个字符串</li>
+        <li><code>score</code>：目标框置信度，一个浮点数</li>
+        <li><code>coordinate</code>：目标框坐标，一个浮点数列表，格式为<code>[xmin, ymin, xmax, ymax]</code></li>
+    </ol>
+</li>
+</ul>
+
+
+相关方法、参数等说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>含义：</b>模型名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，则使用<code>PP-DocLayout-L</code></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td>模型存储路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+<b>例如：</b><code>"cpu"</code>、<code>"gpu"</code>、<code>"npu"</code>、<code>"gpu:0"</code>、<code>"gpu:0,1"</code>。<br/>
+如指定多个设备，将进行并行推理。<br/>
+默认情况下，优先使用 GPU 0；若不可用则使用 CPU。
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>、<code>onnxruntime</code>。保持为默认值 <code>None</code> 时，本地推理默认使用 <code>paddle_static</code> 引擎。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>当使用 Paddle Inference 的 TensorRT 子图引擎时设置的计算精度。<br/>
+<b>说明：</b>
+<b>可选项：</b><code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+<b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上推理时使用的线程数量。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>img_size</code></td>
+<td><b>含义：</b>输入图像大小。<br/>
+<b>说明：</b>
+<ul>
+<li><b>int</b>：如<code>640</code>，表示将输入图像resize到640x640大小。</li>
+<li><b>list</b>：如<code>[640, 512]</code>，表示将输入图像resize到宽为640、高为512。</li>
+</ul>
+</td>
+<td><code>int|list|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>threshold</code></td>
+<td><b>含义：</b>用于过滤掉低置信度预测结果的阈值。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：如<code>0.2</code>，表示过滤掉所有阈值小于0.2的目标框。</li>
+<li><b>dict</b>：字典的键为<code>int</code>类型，代表类别ID；值为<code>float</code>类型阈值。如<code>{0: 0.45, 2: 0.48, 7: 0.4}</code>，表示对ID为0的类别应用阈值0.45、ID为1的类别应用阈值0.48、ID为7的类别应用阈值0.4。</li>
+<li><b>None</b>：使用模型默认的配置。</li>
+</ul>
+</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>是否使用NMS后处理，过滤重叠框。<br/>
+<b>说明：</b>
+<ul>
+<li><b>bool</b>表示使用/不使用NMS进行检测框的后处理过滤重叠框。</li>
+<li><b>None</b>使用模型默认的配置。</li>
+</ul>
+</td>
+<td><code>bool|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>检测框的边长缩放倍数。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于0的浮点数，如<code>1.1</code>，表示将模型输出的检测框中心不变，宽和高都扩张1.1倍。</li>
+<li><b>list</b>：如<code>[1.2, 1.5]</code>，表示将模型输出的检测框中心不变，宽度扩张1.2倍，高度扩张1.5倍。</li>
+<li><b>dict</b>：字典的键为<code>int</code>类型，代表类别ID；值为<code>tuple</code>类型，如<code>{0: (1.1, 2.0)}</code>，表示将模型输出的第0类别检测框中心不变，宽度扩张1.1倍，高度扩张2.0倍。</li>
+<li><b>None</b>：使用模型默认的配置。</li>
+</ul>
+</td>
+<td><code>float|list|dict|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>模型输出的检测框的合并处理模式。<br/>
+<b>说明：</b>
+<ul>
+<li><b>"large"</b>：设置为<code>"large"</code>，表示在模型输出的检测框中，对于互相重叠包含的检测框，只保留外部最大的框，删除重叠的内部框。</li>
+<li><b>"small"</b>：设置为<code>"small"</code>，表示在模型输出的检测框中，对于互相重叠包含的检测框，只保留内部被包含的小框，删除重叠的外部框。</li>
+<li><b>"union"</b>：不进行框的过滤处理，内外框都保留。</li>
+<li><b>dict</b>：字典的键为<code>int</code>类型，代表类别ID；值为<code>str</code>类型, 如<code>{0: "large", 2: "small"}</code>, 表示对第0类别检测框使用<code>large</code>模式，对第2类别检测框使用<code>small</code>。</li>
+<li><b>None</b>：使用模型默认的配置。</li>
+</ul>
+</td>
+<td><code>str|dict|None</code></td>
+<td>None</td>
+</tr>
+</tbody>
+</table>
+
+
+* 调用目标检测模型的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。另外，本模块还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。<code>predict()</code> 方法参数有 <code>input</code> 、<code>batch_size</code>和<code>threshold</code>，具体说明如下：
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>含义：</b>批大小。<br/>
+<b>说明：</b>
+可设置为任意正整数。</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+<tr>
+<td><code>threshold</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|dict|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|list|dict|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。
+</td>
+<td><code>str|dict|None</code></td>
+<td>None</td>
+</tr>
+</tbody>
+</table>
+
+* 对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为<code>json</code>文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将结果保存为图像格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+</table>
+
+* 此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">获取预测的<code>json</code>格式的结果</td>
+</tr>
+<tr>
+<td rowspan="1"><code>img</code></td>
+<td rowspan="1">获取格式为<code>dict</code>的可视化图像</td>
+</tr>
+</table>
+
+## 四、推理引擎 {#四推理引擎}
+
+关于推理引擎的详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。
+
+### 4.1 速度数据
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">PP-DocLayoutV3</td>
+            <td>paddle_static</td>
+            <td>10.95</td>
+            <td>47.99</td>
+            <td>12.97</td>
+            <td>72.33</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>11.33</td>
+            <td>84.48</td>
+            <td>1.31</td>
+            <td>98.01</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>16.94</td>
+            <td>47.11</td>
+            <td>13.83</td>
+            <td>78.97</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>9.26</td>
+            <td>39.11</td>
+            <td>10.23</td>
+            <td>58.80</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-DocLayoutV2</td>
+            <td>paddle_static</td>
+            <td>10.48</td>
+            <td>30.94</td>
+            <td>1.33</td>
+            <td>42.93</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>11.07</td>
+            <td>86.38</td>
+            <td>1.33</td>
+            <td>99.80</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>16.76</td>
+            <td>49.08</td>
+            <td>2.43</td>
+            <td>69.30</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>9.25</td>
+            <td>15.13</td>
+            <td>1.25</td>
+            <td>25.82</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>测试环境说明:</strong>
+<ul>
+    <li><strong>测试数据：</strong><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg">示例图片</a></li>
+    <li><strong>硬件配置：</strong>
+        <ul>
+            <li>GPU：NVIDIA A100 40G</li>
+            <li>CPU：Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>软件环境：</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 4.2 权重转换 {#42-权重转换}
+
+使用推理引擎时，系统会自动下载官方预训练模型。若需使用自训练模型配合 `paddle_dynamic` 或 `transformers` 引擎，请参考 [PaddleX 版面分析模块权重转换](https://paddlepaddle.github.io/PaddleX/main/module_usage/tutorials/ocr_modules/layout_analysis.html#442) 部分，将 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式，即可无缝集成到 PaddleOCR 的 API 中进行推理。若需使用自训练模型配合`onnxruntime`引擎，请参考[PaddleX 获取 ONNX 模型](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html)获取onnx模型，即可无缝集成到 PaddleOCR 的 API 中进行推理。
diff --git a/docs/version3.x/module_usage/layout_detection.en.md b/docs/version3.x/module_usage/layout_detection.en.md
new file mode 100644
index 0000000..d895c44
--- /dev/null
+++ b/docs/version3.x/module_usage/layout_detection.en.md
@@ -0,0 +1,852 @@
+---
+comments: true
+---
+
+# Layout Detection Module Tutorial
+
+## 1. Overview
+
+The core task of structure analysis is to parse and segment the content of input document images. By identifying different elements in the image (such as text, charts, images, etc.), they are classified into predefined categories (e.g., pure text area, title area, table area, image area, list area, etc.), and the position and size of these regions in the document are determined.
+
+## 2. Supported Model List
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing. The "Normal Mode" values correspond to the local <code>paddle_static</code> inference engine.
+
+* <b>The layout detection model includes 20 common categories: document title, paragraph title, text, page number, abstract, table, references, footnotes, header, footer, algorithm, formula, formula number, image, table, seal, figure_table title, chart, and sidebar text and lists of references</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout_plus-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout_plus-L_pretrained.pdparams">Training Model</a></td>
+<td>83.2</td>
+<td>53.03 / 17.23</td>
+<td>634.62 / 378.32</td>
+<td>126.01</td>
+<td>A higher-precision layout area localization model trained on a self-built dataset containing Chinese and English papers, PPT, multi-layout magazines, contracts, books, exams, ancient books and research reports using RT-DETR-L</td>
+</tr>
+<tr>
+</tbody>
+</table>
+
+
+* <b>The layout detection model includes 1 category: Block:</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocBlockLayout</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBlockLayout_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocBlockLayout_pretrained.pdparams">Training Model</a></td>
+<td>95.9</td>
+<td>34.60 / 28.54</td>
+<td>506.43 / 256.83</td>
+<td>123.92</td>
+<td>A layout block localization model trained on a self-built dataset containing Chinese and English papers, PPT, multi-layout magazines, contracts, books, exams, ancient books and research reports using RT-DETR-L</td>
+</tr>
+<tr>
+</tbody>
+</table>
+
+
+* <b>The layout detection model includes 23 common categories: document title, paragraph title, text, page number, abstract, table of contents, references, footnotes, header, footer, algorithm, formula, formula number, image, figure caption, table, table caption, seal, figure title, figure, header image, footer image, and sidebar text</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-L_pretrained.pdparams">Training Model</a></td>
+<td>90.4</td>
+<td>33.59 / 33.59</td>
+<td>503.01 / 251.08</td>
+<td>123.76</td>
+<td>A high-precision layout area localization model trained on a self-built dataset containing Chinese and English papers, magazines, contracts, books, exams, and research reports using RT-DETR-L.</td>
+</tr>
+<tr>
+<td>PP-DocLayout-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-M_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-M_pretrained.pdparams">Training Model</a></td>
+<td>75.2</td>
+<td>13.03 / 4.72</td>
+<td>43.39 / 24.44</td>
+<td>22.578</td>
+<td>A layout area localization model with balanced precision and efficiency, trained on a self-built dataset containing Chinese and English papers, magazines, contracts, books, exams, and research reports using PicoDet-L.</td>
+</tr>
+<tr>
+<td>PP-DocLayout-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-S_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-S_pretrained.pdparams">Training Model</a></td>
+<td>70.9</td>
+<td>11.54 / 3.86</td>
+<td>18.53 / 6.29</td>
+<td>4.834</td>
+<td>A high-efficiency layout area localization model trained on a self-built dataset containing Chinese and English papers, magazines, contracts, books, exams, and research reports using PicoDet-S.</td>
+</tr>
+</tbody>
+</table>
+
+> ❗ The above list includes the <b>4 core models</b> that are key supported by the text recognition module. The module actually supports a total of <b>12 full models</b>, including several predefined models with different categories. The complete model list is as follows:
+
+<details><summary> 👉 Details of Model List</summary>
+
+* <b>Table Layout Detection Model</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet_layout_1x_table</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_table_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_table_pretrained.pdparams">Training Model</a></td>
+<td>97.5</td>
+<td>9.57 / 6.63</td>
+<td>27.66 / 16.75</td>
+<td>7.4</td>
+<td>A high-efficiency layout area localization model trained on a self-built dataset using PicoDet-1x, capable of detecting table regions.</td>
+</tr>
+</tbody></table>
+
+* <b>3-Class Layout Detection Model, including Table, Image, and Stamp</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_3cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_3cls_pretrained.pdparams">Training Model</a></td>
+<td>88.2</td>
+<td>8.43 / 3.44</td>
+<td>17.60 / 6.51</td>
+<td>4.8</td>
+<td>A high-efficiency layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using PicoDet-S.</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_3cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_3cls_pretrained.pdparams">Training Model</a></td>
+<td>89.0</td>
+<td>12.80 / 9.57</td>
+<td>45.04 / 23.86</td>
+<td>22.6</td>
+<td>A balanced efficiency and precision layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using PicoDet-L.</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_3cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_3cls_pretrained.pdparams">Training Model</a></td>
+<td>95.8</td>
+<td>114.80 / 25.65</td>
+<td>924.38 / 924.38</td>
+<td>470.1</td>
+<td>A high-precision layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using RT-DETR-H.</td>
+</tr>
+</tbody></table>
+
+* <b>5-Class English Document Area Detection Model, including Text, Title, Table, Image, and List</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet_layout_1x</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_pretrained.pdparams">Training Model</a></td>
+<td>97.8</td>
+<td>9.62 / 6.75</td>
+<td>26.96 / 12.77</td>
+<td>7.4</td>
+<td>A high-efficiency English document layout area localization model trained on the PubLayNet dataset using PicoDet-1x.</td>
+</tr>
+</tbody></table>
+
+* <b>17-Class Area Detection Model, including 17 common layout categories: Paragraph Title, Image, Text, Number, Abstract, Content, Figure Caption, Formula, Table, Table Caption, References, Document Title, Footnote, Header, Algorithm, Footer, and Stamp</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>87.4</td>
+<td>8.80 / 3.62</td>
+<td>17.51 / 6.35</td>
+<td>4.8</td>
+<td>A high-efficiency layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using PicoDet-S.</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>89.0</td>
+<td>12.60 / 10.27</td>
+<td>43.70 / 24.42</td>
+<td>22.6</td>
+<td>A balanced efficiency and precision layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using PicoDet-L.</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>98.3</td>
+<td>115.29 / 101.18</td>
+<td>964.75 / 964.75</td>
+<td>470.2</td>
+<td>A high-precision layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using RT-DETR-H.</td>
+</tr>
+</tbody>
+</table>
+
+<strong>Test Environment Description:</strong>
+
+  <ul>
+      <li><b>Performance Test Environment</b>
+          <ul>
+             <li><strong>Test Dataset：</strong>
+                 <ul>
+                   <li>20 types of layout detection models: PaddleOCR's self built layout area detection dataset, including Chinese and English papers, magazines, newspapers, research papers PPT、 1300 images of document types such as test papers and textbooks. </li>
+                   <li>Type 1 version face region detection model: PaddleOCR's self built version face region detection dataset, including Chinese and English papers, magazines, newspapers, research reports PPT、 1000 document type images such as test papers and textbooks. </li>      
+                   <li>23 categories Layout Detection Model: A self-built layout area detection dataset by PaddleOCR, containing 500 common document type images such as Chinese and English papers, magazines, contracts, books, exam papers, and research reports.</li>
+                   <li>Table Layout Detection Model: A self-built table area detection dataset by PaddleOCR, including 7,835 Chinese and English paper document type images with tables.</li>
+                   <li> 3-Class Layout Detection Model: A self-built layout area detection dataset by PaddleOCR, comprising 1,154 common document type images such as Chinese and English papers, magazines, and research reports.</li>
+                   <li>5-Class English Document Area Detection Model: The evaluation dataset of <a href="https://developer.ibm.com/exchanges/data/all/publaynet">PubLayNet</a>, containing 11,245 images of English documents.</li>
+                   <li>17-Class Area Detection Model: A self-built layout area detection dataset by PaddleOCR, including 892 common document type images such as Chinese and English papers, magazines, and research reports.</li>
+                 </ul>
+             </li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>Inference Mode Description</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration </th>
+            <th>CPU Configuration </th>
+            <th>Acceleration Technology Combination</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Normal Mode</td>
+            <td>FP32 Precision / No TRT Acceleration</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>High-Performance Mode</td>
+            <td>Optimal combination of pre-selected precision types and acceleration strategies</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Pre-selected optimal backend (Paddle/OpenVINO/TRT, etc.)</td>
+        </tr>
+    </tbody>
+</table>
+
+</details>
+
+
+## 3. Quick Integration  <a id="quick"> </a>
+
+> ❗ Before quick integration, please install the PaddleOCR wheel package. For detailed instructions, refer to [PaddleOCR Local Installation Tutorial](../installation.en.md)。
+
+Quickly experience with just one command:
+
+```bash
+paddleocr layout_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg
+
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr layout_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg \
+    --engine transformers
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following command:
+
+```bash
+# Use the onnxruntime engine for inference
+paddleocr layout_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg \
+    --engine onnxruntime
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+<b>Note: </b>The official models would be download from HuggingFace by default. If can't access to HuggingFace, please set the environment variable <code>PADDLE_PDX_MODEL_SOURCE="BOS"</code> to change the model source to BOS. In the future, more model sources will be supported.
+
+You can also integrate the model inference from the layout area detection module into your project. Before running the following code, please download [Example Image](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg) Go to the local area.
+
+```python
+from paddleocr import LayoutDetection
+
+model = LayoutDetection(model_name="PP-DocLayout_plus-L")
+output = model.predict("layout.jpg", batch_size=1, layout_nms=True)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following code:
+
+```python
+from paddleocr import LayoutDetection
+
+model = LayoutDetection(
+    model_name="PP-DocLayout_plus-L",
+    engine="transformers",
+)
+output = model.predict("layout.jpg", batch_size=1, layout_nms=True)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following code:
+
+```python
+from paddleocr import LayoutDetection
+
+model = LayoutDetection(
+    model_name="PP-DocLayout_plus-L",
+    engine="onnxruntime",
+)
+output = model.predict("layout.jpg", batch_size=1, layout_nms=True)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+If you want to use the trained model with the `paddle_dynamic` or `transformers` engine, refer to the [Weight Conversion](#52-weight-conversion) section in the [Inference Engine](#5-inference-engine) section below to convert the model from the `pdparams` format to the `safetensors` format using PaddleX.
+
+After running, the result obtained is:
+
+```bash
+{'res': {'input_path': 'layout.jpg', 'page_index': None, 'boxes': [{'cls_id': 2, 'label': 'text', 'score': 0.9870226979255676, 'coordinate': [34.101906, 349.85275, 358.59213, 611.0772]}, {'cls_id': 2, 'label': 'text', 'score': 0.9866003394126892, 'coordinate': [34.500324, 647.1585, 358.29367, 848.66797]}, {'cls_id': 2, 'label': 'text', 'score': 0.9846674203872681, 'coordinate': [385.71445, 497.40973, 711.2261, 697.84265]}, {'cls_id': 8, 'label': 'table', 'score': 0.984126091003418, 'coordinate': [73.76879, 105.94899, 321.95303, 298.84888]}, {'cls_id': 8, 'label': 'table', 'score': 0.9834211468696594, 'coordinate': [436.95642, 105.81531, 662.7168, 313.48462]}, {'cls_id': 2, 'label': 'text', 'score': 0.9832247495651245, 'coordinate': [385.62787, 346.2288, 710.10095, 458.77127]}, {'cls_id': 2, 'label': 'text', 'score': 0.9816061854362488, 'coordinate': [385.7802, 735.1931, 710.56134, 849.9764]}, {'cls_id': 6, 'label': 'figure_title', 'score': 0.9577341079711914, 'coordinate': [34.421448, 20.055151, 358.71283, 76.53663]}, {'cls_id': 6, 'label': 'figure_title', 'score': 0.9505634307861328, 'coordinate': [385.72278, 20.053688, 711.29333, 74.92744]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.9001723527908325, 'coordinate': [386.46344, 477.03488, 699.4023, 490.07474]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.8845751285552979, 'coordinate': [35.413048, 627.73596, 185.58383, 640.52264]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.8837394118309021, 'coordinate': [387.17603, 716.3423, 524.7841, 729.258]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.8508939743041992, 'coordinate': [35.50064, 331.18445, 141.6444, 344.81097]}]}}
+```
+
+The meanings of the parameters are as follows:
+<ul>
+<li><code>input_path</code>：The path to the input image for prediction.</li>
+<li><code>page_index</code>：If the input is a PDF file, it indicates which page of the PDF it is; otherwise, it is <code>None</code>.</li>
+<li><code>boxes</code>：Information about the predicted bounding boxes, a list of dictionaries. Each dictionary represents a detected object and contains the following information:
+    <ol start="1" type="1">
+        <li><code>cls_id</code>：Class ID, an integer.</li>
+        <li><code>label</code>：Class label, a string.</li>
+        <li><code>score</code>：Confidence score of the bounding box, a float.</li>
+        <li><code>coordinate</code>：Coordinates of the bounding box, a list of floats in the format <code>[xmin, ymin, xmax, ymax]</code>.</li>
+    </ol>
+</li>
+</ul>
+
+
+The visualized image is as follows:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/layout_det/layout_res_plus.jpg"/>
+
+Relevant methods, parameters, and explanations are as follows:
+
+* <code>LayoutDetection</code> instantiates a target detection model (here, <code>PP-DocLayout_plus-L</code> is used as an example). The detailed explanation is as follows:
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>Meaning:</b> Model name.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, <code>PP-DocLayout-L</code> will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>Meaning:</b>Model storage path.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>Device for inference.<br/>
+<b>Description:</b> 
+<b>For example:</b> <code>"cpu"</code>, <code>"gpu"</code>, <code>"npu"</code>, <code>"gpu:0"</code>, <code>"gpu:0,1"</code>.<br/>
+If multiple devices are specified, parallel inference will be performed.<br/>
+By default, GPU 0 is used if available; otherwise, CPU is used.
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, <code>transformers</code>, and <code>onnxruntime</code>. When left as <code>None</code>, local inference uses the <code>paddle_static</code> engine by default. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b>Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b>Whether to use the Paddle Inference TensorRT subgraph engine.<br/>
+<b>Description:</b> 
+If the model does not support acceleration through TensorRT, setting this flag will not enable acceleration.<br/>
+For Paddle with CUDA version 11.8, the compatible TensorRT version is 8.x (x>=6), and it is recommended to install TensorRT 8.6.1.6.<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b>Computation precision when using the TensorRT subgraph engine in Paddle Inference.<br/>
+<b>Description:</b> 
+<b>Options:</b> <code>"fp32"</code>, <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+<b>Meaning:</b>Whether to enable MKL-DNN acceleration for inference. <br/>
+<b>Description:</b> 
+If MKL-DNN is unavailable or the model does not support it, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b>MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b>Number of threads to use for inference on CPUs.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>img_size</code></td>
+<td><b>Meaning:</b>Input image size.<br/>
+<b>Description:</b> 
+<ul>
+<li><b>int</b>: e.g. <code>640</code>, resizes input image to 640x640.</li>
+<li><b>list</b>: e.g. <code>[640, 512]</code>, resizes input image to width 640 and height 512.</li>
+</ul>
+</td>
+<td><code>int|list|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>threshold</code></td>
+<td><b>Meaning:</b>Threshold for filtering low-confidence predictions.<br/>
+<b>Description:</b>
+<ul>
+<li><b>float</b>: e.g. <code>0.2</code>, filters out all boxes with confidence below 0.2.</li>
+<li><b>dict</b>: The key is <code>int</code> (class id), the value is <code>float</code> (threshold). For example, <code>{0: 0.45, 2: 0.48, 7: 0.4}</code> means class 0 uses threshold 0.45, class 2 uses 0.48, class 7 uses 0.4.</li>
+<li><b>None</b>: uses the model's default configuration.</li>
+</ul>
+</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Whether to use NMS post-processing to filter overlapping boxes.<br/>
+<b>Description:</b>
+<ul>
+<li><b>bool</b>: whether to use NMS for post-processing to filter overlapping boxes.</li>
+<li><b>None</b>: uses the model's default configuration.</li>
+</ul>
+</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Scaling factor for the side length of the detection box.<br/>
+<b>Description:</b>
+<ul>
+<li><b>float</b>: A float greater than 0, e.g. <code>1.1</code>, expands width and height by 1.1 times.</li>
+<li><b>list</b>: e.g. <code>[1.2, 1.5]</code>, expands width by 1.2x and height by 1.5x.</li>
+<li><b>dict</b>: The key is <code>int</code> (class id), the value is <code>tuple</code> of two floats (width ratio, height ratio). For example, <code>{0: (1.1, 2.0)}</code> means for class 0, width is expanded by 1.1x and height by 2.0x.</li>
+<li><b>None</b>: uses the model's default configuration.</li>
+</ul>
+</td>
+<td><code>float|list|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>Merge mode for model output bounding boxes.<br/>
+<b>Description:</b>
+<ul>
+<li><b>"large"</b>: Only keep the largest outer box among overlapping boxes, remove inner boxes.</li>
+<li><b>"small"</b>: Only keep the smallest inner box among overlapping boxes, remove outer boxes.</li>
+<li><b>"union"</b>: Keep all boxes, no filtering.</li>
+<li><b>dict</b>: The key is <code>int</code> (class id), the value is <code>str</code> (mode). For example, <code>{0: "large", 2: "small"}</code> means class 0 uses "large" mode, class 2 uses "small" mode.</li>
+<li><b>None</b>: Use the model's default configuration.</li>
+</ul>
+</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+* The <code>predict()</code> method of the target detection model is called for inference prediction. The parameters of the <code>predict()</code> method are <code>input</code> , <code>batch_size</code>, and <code>threshold</code>, which are explained as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Input data to be predicted. Required. <br/>
+<b>Description:</b> 
+Supports multiple input types:<ul>
+<li><b>Python Var</b>: e.g., <code>numpy.ndarray</code> representing image data</li>
+<li><b>str</b>: 
+    <ul>
+  <li>Local image or PDF file path: <code>/root/data/img.jpg</code>;</li>
+  <li><b>URL</b> of image or PDF file: e.g., <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">example</a>;</li>
+  <li><b>Local directory</b>: directory containing images for prediction, e.g., <code>/root/data/</code> (Note: directories containing PDF files are not supported; PDFs must be specified by exact file path)</li>
+  </li>
+    </ul>
+<li><b>list</b>: Elements must be of the above types, e.g., <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>Meaning:</b>Batch size.<br/>
+ <b>Description:</b> 
+ positive integer.</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+<tr>
+<td><code>threshold</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|dict|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|list|dict|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>str|dict|None</code></td>
+<td>None</td>
+</tr>
+</tbody>
+</table>
+
+
+* Process the prediction results, with each sample's prediction result being the corresponding Result object, and supporting operations such as printing, saving as an image, and saving as a <code>json</code> file:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Method Description</th>
+<th>Parameters</th>
+<th>Parameter type</th>
+<th>Parameter Description</th>
+<th>Default value</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">Print the result to the terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Do you want to use <code>JSON</code> indentation formatting for the output content</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to enhance the readability of the <code>JSON</code> data output, only valid when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non ASCII characters to Unicode characters. When set to <code>True</code>, all non ASCII </code>characters will be escaped; <code>False</code> preserves the original characters and is only valid when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">Save the result as a JSON format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The saved file path, when it is a directory, the name of the saved file is consistent with the name of the input file type</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to enhance the readability of the <code>JSON</code> data output, only valid when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non ASCII characters to Unicode characters. When set to <code>True</code>, all non <code>ASCII</code> characters will be escaped; <code>False</code> preserves the original characters and is only valid when<code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save the results as an image format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The saved file path, when it is a directory, the name of the saved file is consistent with the name of the input file type</td>
+<td>None</td>
+</tr>
+</table>
+
+
+* Additionally, it also supports obtaining the visualized image with results and the prediction results via attributes, as follows:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">Get the prediction result in <code>json</code> format</td>
+</tr>
+<tr>
+<td rowspan="1"><code>img</code></td>
+<td rowspan="1">Get the visualized image in <code>dict</code> format</td>
+</tr>
+</table>
+
+
+## 4. Custom Development
+
+Since PaddleOCR does not directly provide training for the layout detection module, if you need to train the layout area detection model, you can refer to [PaddleX Layout Detection Module Secondary Development](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/layout_detection.html#iv-custom-development)Partially conduct training. The trained model can be seamlessly integrated into PaddleOCR's API for inference.
+
+If you want to use the `paddle_dynamic` or `transformers` engine with the trained model, please refer to the [Weight Conversion](#52-weight-conversion) section in [Inference Engine](#5-inference-engine) later in this document to convert the model from the `pdparams` format to the `safetensors` format using PaddleX.
+
+## 5. Inference Engine
+
+For detailed descriptions, values, compatibility rules, and examples of the inference engine, please refer to <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration Description</a>.
+
+### 5.1 Speed Data
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">PP-DocLayout_plus-L</td>
+            <td>paddle_static</td>
+            <td>10.92</td>
+            <td>26.11</td>
+            <td>0.16</td>
+            <td>37.38</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>11.09</td>
+            <td>72.91</td>
+            <td>0.16</td>
+            <td>85.10</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>12.65</td>
+            <td>37.91</td>
+            <td>0.75</td>
+            <td>52.24</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>9.14</td>
+            <td>12.35</td>
+            <td>0.14</td>
+            <td>21.81</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-DocBlockLayout</td>
+            <td>paddle_static</td>
+            <td>9.51</td>
+            <td>27.59</td>
+            <td>0.08</td>
+            <td>37.41</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>8.94</td>
+            <td>70.77</td>
+            <td>0.07</td>
+            <td>80.73</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>11.37</td>
+            <td>37.95</td>
+            <td>0.75</td>
+            <td>50.96</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>7.53</td>
+            <td>9.22</td>
+            <td>0.06</td>
+            <td>16.97</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>Test Environment Description:</strong>
+<ul>
+    <li><strong>Test Data:</strong> <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg">Sample Image</a></li>
+    <li><strong>Hardware Configuration:</strong>
+        <ul>
+            <li>GPU: NVIDIA A100 40G</li>
+            <li>CPU: Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>Software Environment:</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 5.2 Weight Conversion
+
+When using the inference engine, the system will automatically download the official pre-trained model. If you need to use a self-trained model with the `paddle_dynamic` or `transformers` engine, please refer to the [PaddleX Layout Detection Module Weight Conversion](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/layout_detection.html#442) section to convert the model from the `pdparams` format to the `safetensors` format using PaddleX. This allows seamless integration into the PaddleOCR API for inference. If you need to use a self-trained model with the `onnxruntime` engine, refer to [PaddleX Obtain ONNX Models](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html) to obtain the ONNX model, so it can be seamlessly integrated into the PaddleOCR API for inference.
+
+## 6. FAQ
diff --git a/docs/version3.x/module_usage/layout_detection.md b/docs/version3.x/module_usage/layout_detection.md
new file mode 100644
index 0000000..f0766ed
--- /dev/null
+++ b/docs/version3.x/module_usage/layout_detection.md
@@ -0,0 +1,845 @@
+---
+comments: true
+---
+
+# 版面区域检测模块使用教程
+
+## 一、概述
+
+版面区域检测任务的核心是对输入的文档图像进行内容解析和区域划分。通过识别图像中的不同元素（如文字、图表、图像、公式、段落、摘要、参考文献等），将其归类为预定义的类别，并确定这些区域在文档中的位置。
+
+## 二、支持模型列表
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。表格中的“常规模式”耗时对应本地 <code>paddle_static</code> 推理引擎。
+
+* <b>版面检测模型，包含20个常见的类别：文档标题、段落标题、文本、页码、摘要、目录、参考文献、脚注、页眉、页脚、算法、公式、公式编号、图像、表格、图和表标题（图标题、表格标题和图表标题）、印章、图表、侧栏文本和参考文献内容</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout_plus-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout_plus-L_pretrained.pdparams">训练模型</a></td>
+<td>83.2</td>
+<td>53.03 / 17.23</td>
+<td>634.62 / 378.32</td>
+<td>126.01</td>
+<td>基于RT-DETR-L在包含中英文论文、多栏杂志、报纸、PPT、合同、书本、试卷、研报、古籍、日文文档、竖版文字文档等场景的自建数据集训练的更高精度版面区域定位模型</td>
+</tr>
+<tr>
+</tbody>
+</table>
+
+<b>注：以上精度指标的评估集是自建的版面区域检测数据集，包含中英文论文、杂志、报纸、研报、PPT、试卷、课本等 1300 张文档类型图片。</b>
+
+* <b>文档图像版面子模块检测，包含1个 版面区域 类别，能检测多栏的报纸、杂志的每个子文章的文本区域：</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocBlockLayout</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBlockLayout_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocBlockLayout_pretrained.pdparams">训练模型</a></td>
+<td>95.9</td>
+<td>34.60 / 28.54</td>
+<td>506.43 / 256.83</td>
+<td>123.92</td>
+<td>基于RT-DETR-L在包含中英文论文、多栏杂志、报纸、PPT、合同、书本、试卷、研报、古籍、日文文档、竖版文字文档等场景的自建数据集训练的文档图像版面子模块检测模型</td>
+</tr>
+<tr>
+</tbody>
+</table>
+
+<b>注：以上精度指标的评估集是自建的版面子区域检测数据集，包含中英文论文、杂志、报纸、研报、PPT、试卷、课本等 1000 张文档类型图片。</b>
+
+* <b>版面检测模型，包含23个常见的类别：文档标题、段落标题、文本、页码、摘要、目录、参考文献、脚注、页眉、页脚、算法、公式、公式编号、图像、图表标题、表格、表格标题、印章、图表标题、图表、页眉图像、页脚图像、侧栏文本</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-L_pretrained.pdparams">训练模型</a></td>
+<td>90.4</td>
+<td>33.59 / 33.59</td>
+<td>503.01 / 251.08</td>
+<td>123.76</td>
+<td>基于RT-DETR-L在包含中英文论文、杂志、合同、书本、试卷和研报等场景的自建数据集训练的高精度版面区域定位模型</td>
+</tr>
+<tr>
+<td>PP-DocLayout-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-M_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-M_pretrained.pdparams">训练模型</a></td>
+<td>75.2</td>
+<td>13.03 / 4.72</td>
+<td>43.39 / 24.44</td>
+<td>22.578</td>
+<td>基于PicoDet-L在包含中英文论文、杂志、合同、书本、试卷和研报等场景的自建数据集训练的精度效率平衡的版面区域定位模型</td>
+</tr>
+<tr>
+<td>PP-DocLayout-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-S_pretrained.pdparams">训练模型</a></td>
+<td>70.9</td>
+<td>11.54 / 3.86</td>
+<td>18.53 / 6.29</td>
+<td>4.834</td>
+<td>基于PicoDet-S在中英文论文、杂志、合同、书本、试卷和研报等场景上自建数据集训练的高效率版面区域定位模型</td>
+</tr>
+</tbody>
+</table>
+
+<b>注：以上精度指标的评估集是自建的版面区域检测数据集，包含中英文论文、报纸、研报和试卷等 500 张文档类型图片。</b>
+
+> ❗ 以上列出的是版面检测模块重点支持的<b>5个核心模型</b>，该模块总共支持<b>13个全量模型</b>，包含多个预定义了不同类别的模型，完整的模型列表如下：
+
+<details><summary> 👉模型列表详情</summary>
+
+* <b>表格版面检测模型</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet_layout_1x_table</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_table_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_table_pretrained.pdparams">训练模型</a></td>
+<td>97.5</td>
+<td>9.57 / 6.63</td>
+<td>27.66 / 16.75</td>
+<td>7.4</td>
+<td>基于PicoDet-1x在自建数据集训练的高效率版面区域定位模型，可定位表格这1类区域</td>
+</tr>
+</tbody></table>
+
+* <b>3类版面检测模型，包含表格、图像、印章</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_3cls_pretrained.pdparams">训练模型</a></td>
+<td>88.2</td>
+<td>8.43 / 3.44</td>
+<td>17.60 / 6.51</td>
+<td>4.8</td>
+<td>基于PicoDet-S轻量模型在中英文论文、杂志和研报等场景上自建数据集训练的高效率版面区域定位模型</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_3cls_pretrained.pdparams">训练模型</a></td>
+<td>89.0</td>
+<td>12.80 / 9.57</td>
+<td>45.04 / 23.86</td>
+<td>22.6</td>
+<td>基于PicoDet-L在中英文论文、杂志和研报等场景上自建数据集训练的效率精度均衡版面区域定位模型</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_3cls_pretrained.pdparams">训练模型</a></td>
+<td>95.8</td>
+<td>114.80 / 25.65</td>
+<td>924.38 / 924.38</td>
+<td>470.1</td>
+<td>基于RT-DETR-H在中英文论文、杂志和研报等场景上自建数据集训练的高精度版面区域定位模型</td>
+</tr>
+</tbody></table>
+
+* <b>5类英文文档区域检测模型，包含文字、标题、表格、图片以及列表</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet_layout_1x</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_pretrained.pdparams">训练模型</a></td>
+<td>97.8</td>
+<td>9.62 / 6.75</td>
+<td>26.96 / 12.77</td>
+<td>7.4</td>
+<td>基于PicoDet-1x在PubLayNet数据集训练的高效率英文文档版面区域定位模型</td>
+</tr>
+</tbody></table>
+
+* <b>17类区域检测模型，包含17个版面常见类别，分别是：段落标题、图片、文本、数字、摘要、内容、图表标题、公式、表格、表格标题、参考文献、文档标题、脚注、页眉、算法、页脚、印章</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>87.4</td>
+<td>8.80 / 3.62</td>
+<td>17.51 / 6.35</td>
+<td>4.8</td>
+<td>基于PicoDet-S轻量模型在中英文论文、杂志和研报等场景上自建数据集训练的高效率版面区域定位模型</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>89.0</td>
+<td>12.60 / 10.27</td>
+<td>43.70 / 24.42</td>
+<td>22.6</td>
+<td>基于PicoDet-L在中英文论文、杂志和研报等场景上自建数据集训练的效率精度均衡版面区域定位模型</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>98.3</td>
+<td>115.29 / 101.18</td>
+<td>964.75 / 964.75</td>
+<td>470.2</td>
+<td>基于RT-DETR-H在中英文论文、杂志和研报等场景上自建数据集训练的高精度版面区域定位模型</td>
+</tr>
+</tbody>
+</table>
+
+ <strong>测试环境说明:</strong>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+              <li><strong>测试数据集：</strong>
+                 <ul>
+                    <li>20类版面检测模型： PaddleOCR 自建的版面区域检测数据集，包含中英文论文、杂志、报纸、研报、PPT、试卷、课本等 1300 张文档类型图片。</li>
+                    <li>1类版面子区域检测模型： PaddleOCR 自建的版面子区域检测数据集，包含中英文论文、杂志、报纸、研报、PPT、试卷、课本等 1000 张文档类型图片。</li>                 
+                    <li>23类版面检测模型： PaddleOCR 自建的版面区域检测数据集，包含中英文论文、杂志、合同、书本、试卷和研报等常见的 500 张文档类型图片。</li>
+                    <li>3类版面检测模型：PaddleOCR 自建的版面区域检测数据集，包含中英文论文、杂志和研报等常见的 1154 张文档类型图片。</li>
+                    <li>5类英文文档区域检测模型： <a href="https://developer.ibm.com/exchanges/data/all/publaynet" target="_blank">PubLayNet</a> 的评估数据集，包含英文文档的 11245 张图片。</li>
+                    <li>17类区域检测模型：PaddleOCR 自建的版面区域检测数据集，包含中英文论文、杂志和研报等常见的 892 张文档类型图片。</li>
+                 </ul>
+              </li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+</details>
+
+## 三、快速开始
+
+> ❗ 在快速开始前，请先安装 PaddleOCR 的 wheel 包，详细请参考 [安装教程](../installation.md)。
+
+使用一行命令即可快速体验：
+
+```bash
+paddleocr layout_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下命令：
+
+```bash
+# 使用 transformers 引擎进行推理
+paddleocr layout_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg \
+    --engine transformers
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下命令：
+
+```bash
+# 使用 onnxruntime 引擎进行推理
+paddleocr layout_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg \
+    --engine onnxruntime
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+<b>注：</b>PaddleOCR 官方模型默认从 HuggingFace 获取，如运行环境访问 HuggingFace 不便，可通过环境变量修改模型源为 BOS：`PADDLE_PDX_MODEL_SOURCE="BOS"`，未来将支持更多主流模型源；
+
+您也可以将版面区域检测模块中的模型推理集成到您的项目中。运行以下代码前，请您下载[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg)到本地。
+
+```python
+from paddleocr import LayoutDetection
+
+model = LayoutDetection(model_name="PP-DocLayout_plus-L")
+output = model.predict("layout.jpg", batch_size=1, layout_nms=True)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下代码：
+
+```python
+from paddleocr import LayoutDetection
+
+model = LayoutDetection(
+    model_name="PP-DocLayout_plus-L",
+    engine="transformers",
+)
+output = model.predict("layout.jpg", batch_size=1, layout_nms=True)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下代码：
+
+```python
+from paddleocr import LayoutDetection
+
+model = LayoutDetection(
+    model_name="PP-DocLayout_plus-L",
+    engine="onnxruntime",
+)
+output = model.predict("layout.jpg", batch_size=1, layout_nms=True)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+训练后的模型如果想使用 `paddle_dynamic` 或 `transformers` 引擎，请参考后文 [推理引擎](#五推理引擎) 中的 [权重转换](#52-权重转换) 部分将模型由 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式。
+
+运行后，得到的结果为：
+
+```bash
+{'res': {'input_path': 'layout.jpg', 'page_index': None, 'boxes': [{'cls_id': 2, 'label': 'text', 'score': 0.9870226979255676, 'coordinate': [34.101906, 349.85275, 358.59213, 611.0772]}, {'cls_id': 2, 'label': 'text', 'score': 0.9866003394126892, 'coordinate': [34.500324, 647.1585, 358.29367, 848.66797]}, {'cls_id': 2, 'label': 'text', 'score': 0.9846674203872681, 'coordinate': [385.71445, 497.40973, 711.2261, 697.84265]}, {'cls_id': 8, 'label': 'table', 'score': 0.984126091003418, 'coordinate': [73.76879, 105.94899, 321.95303, 298.84888]}, {'cls_id': 8, 'label': 'table', 'score': 0.9834211468696594, 'coordinate': [436.95642, 105.81531, 662.7168, 313.48462]}, {'cls_id': 2, 'label': 'text', 'score': 0.9832247495651245, 'coordinate': [385.62787, 346.2288, 710.10095, 458.77127]}, {'cls_id': 2, 'label': 'text', 'score': 0.9816061854362488, 'coordinate': [385.7802, 735.1931, 710.56134, 849.9764]}, {'cls_id': 6, 'label': 'figure_title', 'score': 0.9577341079711914, 'coordinate': [34.421448, 20.055151, 358.71283, 76.53663]}, {'cls_id': 6, 'label': 'figure_title', 'score': 0.9505634307861328, 'coordinate': [385.72278, 20.053688, 711.29333, 74.92744]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.9001723527908325, 'coordinate': [386.46344, 477.03488, 699.4023, 490.07474]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.8845751285552979, 'coordinate': [35.413048, 627.73596, 185.58383, 640.52264]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.8837394118309021, 'coordinate': [387.17603, 716.3423, 524.7841, 729.258]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.8508939743041992, 'coordinate': [35.50064, 331.18445, 141.6444, 344.81097]}]}}
+```
+
+参数含义如下：
+<ul>
+<li><code>input_path</code>：输入的待预测图像的路径</li>
+<li><code>page_index</code>：如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+<li><code>boxes</code>：预测的目标框信息，一个字典列表。每个字典代表一个检出的目标，包含以下信息：
+    <ol start="1" type="1">
+        <li><code>cls_id</code>：类别ID，一个整数</li>
+        <li><code>label</code>：类别标签，一个字符串</li>
+        <li><code>score</code>：目标框置信度，一个浮点数</li>
+        <li><code>coordinate</code>：目标框坐标，一个浮点数列表，格式为<code>[xmin, ymin, xmax, ymax]</code></li>
+    </ol>
+</li>
+</ul>
+
+可视化图片如下：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/layout_det/layout_res_plus.jpg"/>
+
+
+相关方法、参数等说明如下：
+
+* <code>LayoutDetection</code>实例化目标检测模型（此处以<code>PP-DocLayout_plus-L</code>为例），具体说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>含义：</b>模型名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，则使用<code>PP-DocLayout-L</code></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>含义：</b>模型存储路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+<b>例如：</b><code>"cpu"</code>、<code>"gpu"</code>、<code>"npu"</code>、<code>"gpu:0"</code>、<code>"gpu:0,1"</code>。<br/>
+如指定多个设备，将进行并行推理。<br/>
+默认情况下，优先使用 GPU 0；若不可用则使用 CPU。
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>、<code>onnxruntime</code>。保持为默认值 <code>None</code> 时，本地推理默认使用 <code>paddle_static</code> 引擎。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>当使用 Paddle Inference 的 TensorRT 子图引擎时设置的计算精度。<br/>
+<b>说明：</b>
+<b>可选项：</b><code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+<b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上推理时使用的线程数量。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>img_size</code></td>
+<td><b>含义：</b>输入图像大小。<br/>
+<b>说明：</b>
+<ul>
+<li><b>int</b>：如<code>640</code>，表示将输入图像resize到640x640大小。</li>
+<li><b>list</b>：如<code>[640, 512]</code>，表示将输入图像resize到宽为640、高为512。</li>
+</ul>
+</td>
+<td><code>int|list|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>threshold</code></td>
+<td><b>含义：</b>用于过滤掉低置信度预测结果的阈值。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：如<code>0.2</code>，表示过滤掉所有阈值小于0.2的目标框。</li>
+<li><b>dict</b>：字典的键为<code>int</code>类型，代表类别ID；值为<code>float</code>类型阈值。如<code>{0: 0.45, 2: 0.48, 7: 0.4}</code>，表示对ID为0的类别应用阈值0.45、ID为1的类别应用阈值0.48、ID为7的类别应用阈值0.4。</li>
+<li><b>None</b>：使用模型默认的配置。</li>
+</ul>
+</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>是否使用NMS后处理，过滤重叠框。<br/>
+<b>说明：</b>
+<ul>
+<li><b>bool</b>表示使用/不使用NMS进行检测框的后处理过滤重叠框。</li>
+<li><b>None</b>使用模型默认的配置。</li>
+</ul>
+</td>
+<td><code>bool|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>检测框的边长缩放倍数。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于0的浮点数，如<code>1.1</code>，表示将模型输出的检测框中心不变，宽和高都扩张1.1倍。</li>
+<li><b>list</b>：如<code>[1.2, 1.5]</code>，表示将模型输出的检测框中心不变，宽度扩张1.2倍，高度扩张1.5倍。</li>
+<li><b>dict</b>：字典的键为<code>int</code>类型，代表类别ID；值为<code>tuple</code>类型，如<code>{0: (1.1, 2.0)}</code>，表示将模型输出的第0类别检测框中心不变，宽度扩张1.1倍，高度扩张2.0倍。</li>
+<li><b>None</b>：使用模型默认的配置。</li>
+</ul>
+</td>
+<td><code>float|list|dict|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>模型输出的检测框的合并处理模式。<br/>
+<b>说明：</b>
+<ul>
+<li><b>"large"</b>：设置为<code>"large"</code>，表示在模型输出的检测框中，对于互相重叠包含的检测框，只保留外部最大的框，删除重叠的内部框。</li>
+<li><b>"small"</b>：设置为<code>"small"</code>，表示在模型输出的检测框中，对于互相重叠包含的检测框，只保留内部被包含的小框，删除重叠的外部框。</li>
+<li><b>"union"</b>：不进行框的过滤处理，内外框都保留。</li>
+<li><b>dict</b>：字典的键为<code>int</code>类型，代表类别ID；值为<code>str</code>类型, 如<code>{0: "large", 2: "small"}</code>, 表示对第0类别检测框使用<code>large</code>模式，对第2类别检测框使用<code>small</code>。</li>
+<li><b>None</b>：使用模型默认的配置。</li>
+</ul>
+</td>
+<td><code>str|dict|None</code></td>
+<td>None</td>
+</tr>
+</tbody>
+</table>
+
+
+* 调用目标检测模型的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。另外，本模块还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。<code>predict()</code> 方法参数有 <code>input</code> 、<code>batch_size</code>和<code>threshold</code>，具体说明如下：
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>含义：</b>批大小。<br/>
+<b>说明：</b>
+可设置为任意正整数。</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+<tr>
+<td><code>threshold</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|dict|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|list|dict|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。
+</td>
+<td><code>str|dict|None</code></td>
+<td>None</td>
+</tr>
+</tbody>
+</table>
+
+* 对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为<code>json</code>文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将结果保存为图像格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+</table>
+
+* 此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">获取预测的<code>json</code>格式的结果</td>
+</tr>
+<tr>
+<td rowspan="1"><code>img</code></td>
+<td rowspan="1">获取格式为<code>dict</code>的可视化图像</td>
+</tr>
+</table>
+
+
+## 四、二次开发
+
+由于 PaddleOCR 并不直接提供版面区域检测模块的训练，因此，如果需要训练版面区域测模型，可以参考 [PaddleX 版面区域检测模块二次开发](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/layout_detection.html#_5)部分进行训练。训练后的模型可以无缝集成到 PaddleOCR 的 API 中进行推理。
+
+训练后的模型如果想使用 `paddle_dynamic` 或 `transformers` 引擎，请参考后文 [推理引擎](#五推理引擎) 中的 [权重转换](#52-权重转换) 部分将模型由 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式。
+
+## 五、推理引擎 {#五推理引擎}
+
+关于推理引擎的详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。
+
+### 5.1 速度数据
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">PP-DocLayout_plus-L</td>
+            <td>paddle_static</td>
+            <td>10.92</td>
+            <td>26.11</td>
+            <td>0.16</td>
+            <td>37.38</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>11.09</td>
+            <td>72.91</td>
+            <td>0.16</td>
+            <td>85.10</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>12.65</td>
+            <td>37.91</td>
+            <td>0.75</td>
+            <td>52.24</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>9.14</td>
+            <td>12.35</td>
+            <td>0.14</td>
+            <td>21.81</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-DocBlockLayout</td>
+            <td>paddle_static</td>
+            <td>9.51</td>
+            <td>27.59</td>
+            <td>0.08</td>
+            <td>37.41</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>8.94</td>
+            <td>70.77</td>
+            <td>0.07</td>
+            <td>80.73</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>11.37</td>
+            <td>37.95</td>
+            <td>0.75</td>
+            <td>50.96</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>7.53</td>
+            <td>9.22</td>
+            <td>0.06</td>
+            <td>16.97</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>测试环境说明:</strong>
+<ul>
+    <li><strong>测试数据：</strong><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/layout.jpg">示例图片</a></li>
+    <li><strong>硬件配置：</strong>
+        <ul>
+            <li>GPU：NVIDIA A100 40G</li>
+            <li>CPU：Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>软件环境：</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 5.2 权重转换 {#52-权重转换}
+
+使用推理引擎时，系统会自动下载官方预训练模型。若需使用自训练模型配合 `paddle_dynamic` 或 `transformers` 引擎，请参考 [PaddleX 版面区域检测模块权重转换](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/layout_detection.html#442) 部分，将 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式，即可无缝集成到 PaddleOCR 的 API 中进行推理。若需使用自训练模型配合`onnxruntime`引擎，请参考[PaddleX 获取 ONNX 模型](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html)获取onnx模型，即可无缝集成到 PaddleOCR 的 API 中进行推理。
+
+## 六、FAQ
diff --git a/docs/version3.x/module_usage/module_overview.en.md b/docs/version3.x/module_usage/module_overview.en.md
new file mode 100644
index 0000000..2485629
--- /dev/null
+++ b/docs/version3.x/module_usage/module_overview.en.md
@@ -0,0 +1,3 @@
+# Module Overview
+
+A module is the smallest unit that implements basic functionality. Modules typically use a single model to accomplish specific tasks, such as text detection, image classification, and other basic functions. As fundamental building blocks, modules provide the necessary functional support for more complex application scenarios. This design approach allows users to flexibly select and combine different modules according to their needs, thereby simplifying the development process and enhancing development flexibility and efficiency.
diff --git a/docs/version3.x/module_usage/module_overview.md b/docs/version3.x/module_usage/module_overview.md
new file mode 100644
index 0000000..c60cb76
--- /dev/null
+++ b/docs/version3.x/module_usage/module_overview.md
@@ -0,0 +1,3 @@
+# 模块概述
+
+模块是实现基本功能的最小单位。模块通常使用单个模型去完成特定的任务，比如文本检测、图像分类等基本功能。模块作为基础构建单元，为更复杂的应用场景提供了必要的功能支持。这种设计方式使得用户可以根据需要灵活选择和组合不同的模块，从而简化了开发流程，并提高了开发的灵活性和效率。
diff --git a/docs/version3.x/module_usage/seal_text_detection.en.md b/docs/version3.x/module_usage/seal_text_detection.en.md
new file mode 100644
index 0000000..618e65c
--- /dev/null
+++ b/docs/version3.x/module_usage/seal_text_detection.en.md
@@ -0,0 +1,530 @@
+---
+comments: true
+---
+
+# Seal Text Detection Module Tutorial
+
+## I. Overview
+The seal text detection module typically outputs multi-point bounding boxes around text regions, which are then passed as inputs to the distortion correction and text recognition modules for subsequent processing to identify the textual content of the seal. Recognizing seal text is an integral part of document processing and finds applications in various scenarios such as contract comparison, inventory access auditing, and invoice reimbursement verification. The seal text detection module serves as a subtask within OCR (Optical Character Recognition), responsible for locating and marking the regions containing seal text within an image. The performance of this module directly impacts the accuracy and efficiency of the entire seal text OCR system.
+
+## II. Supported Model List
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing. The "Normal Mode" values correspond to the local <code>paddle_static</code> inference engine.
+
+<table>
+<thead>
+<tr>
+<th>Model Name</th><th>Model Download Link</th>
+<th>Hmean（%）</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv4_server_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_seal_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_seal_det_pretrained.pdparams">Training Model</a></td>
+<td>98.40</td>
+<td>124.64 / 91.57</td>
+<td>545.68 / 439.86</td>
+<td>109</td>
+<td>The server-side seal text detection model of PP-OCRv4 boasts higher accuracy and is suitable for deployment on better-equipped servers.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_seal_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_seal_det_pretrained.pdparams">Training Model</a></td>
+<td>96.36</td>
+<td>9.70 / 3.56</td>
+<td>50.38 / 19.64</td>
+<td>4.6</td>
+<td>The mobile-side seal text detection model of PP-OCRv4, on the other hand, offers greater efficiency and is suitable for deployment on end devices.</td>
+</tr>
+</tbody>
+</table>
+
+<strong>Test Environment Description:</strong>
+
+  <ul>
+      <li><b>Performance Test Environment</b>
+          <ul>
+               <li><strong>Test Dataset：</strong> A Self-built Internal Dataset, Containing 500 Images of Circular Stamps.</li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>Inference Mode Description</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration </th>
+            <th>CPU Configuration </th>
+            <th>Acceleration Technology Combination</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Normal Mode</td>
+            <td>FP32 Precision / No TRT Acceleration</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>High-Performance Mode</td>
+            <td>Optimal combination of pre-selected precision types and acceleration strategies</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Pre-selected optimal backend (Paddle/OpenVINO/TRT, etc.)</td>
+        </tr>
+    </tbody>
+</table>
+
+
+## III. Quick Integration  <a id="quick"> </a>
+
+> ❗ Before quick integration, please install the PaddleOCR wheel package. For detailed instructions, refer to [PaddleOCR Local Installation Tutorial](../installation.en.md)。
+
+Quickly experience with just one command:
+
+```bash
+paddleocr seal_text_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/seal_text_det.png
+
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+<b>Note: </b>The official models would be download from HuggingFace by default. If can't access to HuggingFace, please set the environment variable `PADDLE_PDX_MODEL_SOURCE="BOS"` to change the model source to BOS. In the future, more model sources will be supported.
+
+You can also integrate the model inference from the layout area detection module into your project. Before running the following code, please download [Example Image](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/seal_text_det.png) Go to the local area.
+
+```python
+from paddleocr import SealTextDetection
+model = SealTextDetection(model_name="PP-OCRv4_server_seal_det")
+output = model.predict("seal_text_det.png", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+After running, the result is:
+
+```bash
+{'res': {'input_path': 'seal_text_det.png', 'page_index': None, 'dt_polys': [array([[463, 477],
+       ...,
+       [428, 505]]), array([[297, 444],
+       ...,
+       [230, 443]]), array([[457, 346],
+       ...,
+       [267, 345]]), array([[325,  38],
+       ...,
+       [322,  37]])], 'dt_scores': [0.9912680344777314, 0.9906849624837963, 0.9847219455533163, 0.9914791724153904]}}
+```
+
+The meanings of the parameters are as follows:
+<ul>
+<li> <code>input_path</code>：represents the path of the input image to be predicted</li>
+<li> <code>dt_polys</code>：represents the predicted text detection boxes, where each text detection box contains multiple vertices of a polygon. Each vertex is a list of two elements, representing the x and y coordinates of the vertex respectively</li>
+<li> <code>dt_scores</code>：represents the confidence scores of the predicted text detection boxes</li>
+</ul>
+
+The visualization image is as follows:
+
+<img alt="Visualization Image" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/seal_text_det/seal_text_det_res.png"/>
+
+The explanations of related methods and parameters are as follows:
+
+* <code>SealTextDetection</code> instantiates a text detection model (here we take <code>PP-OCRv4_server_seal_det</code> as an example), and the specific explanations are as follows:
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>Meaning:</b>Model name. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, <code>PP-OCRv4_mobile_seal_det</code> will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>Meaning:</b>Model storage path.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>Device for inference.<br/>
+<b>Description:</b>
+<b>For example:</b> <code>"cpu"</code>, <code>"gpu"</code>, <code>"npu"</code>, <code>"gpu:0"</code>, <code>"gpu:0,1"</code>.<br/>
+If multiple devices are specified, parallel inference will be performed.<br/>
+By default, GPU 0 is used if available; otherwise, CPU is used.
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, local inference uses the <code>paddle_static</code> engine by default. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b>Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b>Whether to use the Paddle Inference TensorRT subgraph engine. <br/>
+<b>Description:</b>
+If the model does not support acceleration through TensorRT, setting this flag will not enable acceleration.<br/>
+For Paddle with CUDA version 11.8, the compatible TensorRT version is 8.x (x>=6), and it is recommended to install TensorRT 8.6.1.6.<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b>Computation precision when using the TensorRT subgraph engine in Paddle Inference.<br/>
+<b>Description:</b>
+<b>Options:</b> <code>"fp32"</code>, <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+<b>Meaning:</b>Whether to enable MKL-DNN acceleration for inference. <br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support it, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b>MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b>Number of threads to use for inference on CPUs.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>limit_side_len</code></td>
+<td><b>Meaning:</b>Limit on the side length of the input image for detection. <br/>
+<b>Description:</b>
+<code>int</code> specifies the value. If set to <code>None</code>, the model's default configuration will be used.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>limit_type</code></td>
+<td><b>Meaning:</b>Type of image side length limitation. <br/>
+<b>Description:</b>
+<code>"min"</code> ensures the shortest side of the image is no less than <code>det_limit_side_len</code>; <code>"max"</code> ensures the longest side is no greater than <code>limit_side_len</code>. If set to <code>None</code>, the model's default configuration will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>thresh</code></td>
+<td><b>Meaning:</b>Pixel score threshold. <br/>
+<b>Description:</b>
+Pixels in the output probability map with scores greater than this threshold are considered text pixels. Accepts any float value greater than 0. If set to <code>None</code>, the model's default configuration will be used.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>box_thresh</code></td>
+<td><b>Meaning:</b>If the average score of all pixels inside the bounding box is greater than this threshold, the result is considered a text region. <br/>
+<b>Description:</b>
+Accepts any float value greater than 0. If set to <code>None</code>, the model's default configuration will be used.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion ratio for the Vatti clipping algorithm, used to expand the text region. <br/>
+<b>Description:</b>Accepts any float value greater than 0. If set to <code>None</code>, the model's default configuration will be used.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>input_shape</code></td>
+<td><b>Meaning:</b>Input image size for the model in the format <code>(C, H, W)</code>. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the model's default size will be used.</td>
+<td><code>tuple|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+* The  <code>predict()</code>  method of the seal text detection model is called for inference prediction. The parameters of the  <code>predict()</code>  method include <code>input</code> 、<code>batch_size</code>、 <code>limit_side_len</code>、 <code>limit_type</code>、 <code>thresh</code>、 <code>box_thresh</code>、 <code>max_candidates</code>、<code>unclip_ratio</code>. The specific descriptions are as follows:
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Input data to be predicted. Required. <br/>
+<b>Description:</b>
+Supports multiple input types:<ul>
+<li><b>Python Var</b>: e.g., <code>numpy.ndarray</code> representing image data</li>
+<li><b>str</b>: 
+    <ul>
+  <li>Local image or PDF file path: <code>/root/data/img.jpg</code>;</li>
+  <li><b>URL</b> of image or PDF file: e.g., <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">example</a>;</li>
+  <li><b>Local directory</b>: directory containing images for prediction, e.g., <code>/root/data/</code> (Note: directories containing PDF files are not supported; PDFs must be specified by exact file path)</li>
+  </li>
+    </ul>
+<li><b>list</b>: Elements must be of the above types, e.g., <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>Meaning:</b>Batch size. <br/>
+<b>Description:</b>
+Can be set to any positive integer.</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+<tr>
+<td><code>limit_side_len</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>limit_type</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+ If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>box_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>unclip_ratio</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+
+* Process the prediction results. Each sample's prediction result is a corresponding Result object, and it supports operations such as printing, saving as an image, and saving as a <code>json</code> file:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Method Description</th>
+<th>Parameter</th>
+<th>Parameter Type</th>
+<th>Parameter Description</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">Print the result to the terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output content using <code>JSON</code> indentation</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data, making it more readable. This is only effective when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non-<code>ASCII</code> characters to <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> retains the original characters. This is only effective when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">Save the result as a file in JSON format</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The file path for saving. When it is a directory, the saved file name will be consistent with the input file name</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data, making it more readable. This is only effective when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non-<code>ASCII</code> characters to <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> retains the original characters. This is only effective when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save the result as a file in image format</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The file path for saving. When it is a directory, the saved file name will be consistent with the input file name</td>
+<td>None</td>
+</tr>
+</table>
+
+* In addition, it also supports obtaining visual images with results and prediction results through attributes, as follows:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Attribute Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">Get the prediction result in <code>json</code> format</td>
+</tr>
+<tr>
+<td rowspan="1"><code>img</code></td>
+<td rowspan="1">Get the visual image in <code>dict</code> format</td>
+</tr>
+</table>
+
+## IV. Custom Development
+
+If the above model is still not performing well in your scenario, you can try the following steps for secondary development. Here, we'll use training `PP-OCRv4_server_seal_det` as an example; you can replace it with the corresponding configuration files for other models. First, you need to prepare a text detection dataset. You can refer to the format of the [seal text detection demo data](https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_curve_det_dataset_examples.tar) for preparation. Once prepared, you can follow the steps below for model training and export. After export, you can quickly integrate the model into the above API. This example uses a seal text detection demo dataset. Before training the model, please ensure that you have installed the dependencies required by PaddleOCR as per the [installation documentation](../installation.en.md).
+
+### 4.1 Dataset and Pre-trained Model Preparation
+
+#### 4.1.1 Preparing the Dataset
+
+```shell
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_curve_det_dataset_examples.tar -P ./dataset
+tar -xf ./dataset/ocr_curve_det_dataset_examples.tar -C ./dataset/
+```
+
+#### 4.1.1 Preparing the pre-trained model
+
+
+```shell
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_seal_det_pretrained.pdparams
+```
+
+### 4.2 Model Training
+
+PaddleOCR has modularized the code, and when training the `PP-OCRv4_server_seal_det` model, you need to use the [configuration file](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/configs/det/PP-OCRv4/PP-OCRv4_server_seal_det.yml) for `PP-OCRv4_server_seal_det`.
+
+The training commands are as follows:
+
+```bash
+# Single GPU training (default training method)
+python3 tools/train.py -c configs/det/PP-OCRv4/PP-OCRv4_server_seal_det.yml \
+   -o Global.pretrained_model=./PP-OCRv4_server_seal_det_pretrained.pdparams \
+   Train.dataset.data_dir=./dataset/ocr_curve_det_dataset_examples Train.dataset.label_file_list=./dataset/ocr_curve_det_dataset_examples/train.txt \
+   Eval.dataset.data_dir=./dataset/ocr_curve_det_dataset_examples Eval.dataset.label_file_list=./dataset/ocr_curve_det_dataset_examples/val.txt
+   
+# Multi-GPU training, specify GPU ids using the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/det/PP-OCRv4/PP-OCRv4_server_seal_det.yml \
+   -o Global.pretrained_model=./PP-OCRv4_server_seal_det_pretrained.pdparams \
+   Train.dataset.data_dir=./dataset/ocr_curve_det_dataset_examples Train.dataset.label_file_list=./dataset/ocr_curve_det_dataset_examples/train.txt \
+   Eval.dataset.data_dir=./dataset/ocr_curve_det_dataset_examples Eval.dataset.label_file_list=./dataset/ocr_curve_det_dataset_examples/val.txt
+```
+
+### 4.3 Model Evaluation
+
+You can evaluate the trained weights, such as `output/xxx/xxx.pdparams`, using the following command:
+
+```bash
+# Make sure to set the pretrained_model path to the local path. If using a model that was trained and saved by yourself, be sure to modify the path and filename to {path/to/weights}/{model_name}.
+# Demo test set evaluation
+python3 tools/eval.py -c configs/det/PP-OCRv4/PP-OCRv4_server_seal_det.yml -o \
+    Global.pretrained_model=output/xxx/xxx.pdparams
+```
+
+### 4.4 Model Export
+
+```bash
+python3 tools/export_model.py -c configs/det/PP-OCRv4/PP-OCRv4_server_seal_det.yml -o \
+    Global.pretrained_model=output/xxx/xxx.pdparams \
+    Global.save_inference_dir="./PP-OCRv4_server_seal_det_infer/"
+```
+
+After exporting the model, the static graph model will be stored in the `./PP-OCRv4_server_seal_det_infer/` directory. In this directory, you will see the following files:
+```
+./PP-OCRv4_server_seal_det_infer/
+├── inference.json
+├── inference.pdiparams
+├── inference.yml
+```
+With this, the secondary development is complete, and the static graph model can be directly integrated into PaddleOCR's API.
+
+## 5. FAQ
diff --git a/docs/version3.x/module_usage/seal_text_detection.md b/docs/version3.x/module_usage/seal_text_detection.md
new file mode 100644
index 0000000..8d5fee1
--- /dev/null
+++ b/docs/version3.x/module_usage/seal_text_detection.md
@@ -0,0 +1,526 @@
+---
+comments: true
+---
+
+# 印章文本检测模块使用教程
+
+## 一、概述
+
+印章文本检测模块通常会输出文本区域的多点边界框（Bounding Boxes），这些边界框将作为输入传递给弯曲矫正和文本检测模块进行后续处理，识别出印章的文字内容。印章文本的识别是文档处理的一部分，在很多场景都有用途，例如合同比对，出入库审核以及发票报销审核等场景。印章文本检测模块是OCR（光学字符识别）中的子任务，负责在图像中定位和标记出包含印章文本的区域。该模块的性能直接影响到整个印章文本OCR系统的准确性和效率。
+
+## 二、支持模型列表
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。表格中的“常规模式”耗时对应本地 <code>paddle_static</code> 推理引擎。
+
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>检测Hmean（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv4_server_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_seal_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_seal_det_pretrained.pdparams">训练模型</a></td>
+<td>98.40</td>
+<td>124.64 / 91.57</td>
+<td>545.68 / 439.86</td>
+<td>109</td>
+<td>PP-OCRv4的服务端印章文本检测模型，精度更高，适合在较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_seal_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_seal_det_pretrained.pdparams">训练模型</a></td>
+<td>96.36</td>
+<td>9.70 / 3.56</td>
+<td>50.38 / 19.64</td>
+<td>4.7</td>
+<td>PP-OCRv4的移动端印章文本检测模型，效率更高，适合在端侧部署</td>
+</tr>
+</tbody>
+</table>
+<strong>测试环境说明:</strong>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+              <li><strong>测试数据集：</strong>自建的内部数据集，包含500张圆形印章图像。</li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+
+## 三、快速开始
+
+> ❗ 在快速开始前，请先安装 PaddleOCR 的 wheel 包，详细请参考 [安装教程](../installation.md)。
+
+使用一行命令即可快速体验：
+
+```bash
+paddleocr seal_text_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/seal_text_det.png
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+<b>注：</b>PaddleOCR 官方模型默认从 HuggingFace 获取，如运行环境访问 HuggingFace 不便，可通过环境变量修改模型源为 BOS：`PADDLE_PDX_MODEL_SOURCE="BOS"`，未来将支持更多主流模型源；
+
+您也可以将印章文本检测的模块中的模型推理集成到您的项目中。运行以下代码前，请您下载[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/seal_text_det.png)到本地。
+
+```python
+from paddleocr import SealTextDetection
+model = SealTextDetection(model_name="PP-OCRv4_server_seal_det")
+output = model.predict("seal_text_det.png", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+运行后，得到的结果为：
+
+```bash
+{'res': {'input_path': 'seal_text_det.png', 'page_index': None, 'dt_polys': [array([[463, 477],
+       ...,
+       [428, 505]]), array([[297, 444],
+       ...,
+       [230, 443]]), array([[457, 346],
+       ...,
+       [267, 345]]), array([[325,  38],
+       ...,
+       [322,  37]])], 'dt_scores': [0.9912680344777314, 0.9906849624837963, 0.9847219455533163, 0.9914791724153904]}}
+```
+
+运行结果参数含义如下：
+<ul>
+<li> <code>input_path</code>：表示输入待预测图像的路径</li>
+<li> <code>dt_polys</code>：表示预测的文本检测框，其中每个文本检测框包含一个多边形的多个顶点。其中每个顶点都是一个列表，分别表示该顶点的x坐标和y坐标</li>
+<li> <code>dt_scores</code>：表示预测的文本检测框的置信度</li>
+</ul>
+
+可视化图片如下：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/seal_text_det/seal_text_det_res.png"/>
+
+相关方法、参数等说明如下：
+
+* <code>SealTextDetection</code>实例化文本检测模型（此处以<code>PP-OCRv4_server_seal_det</code>为例），具体说明如下：
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>含义：</b>模型名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，则使用<code>PP-OCRv4_mobile_seal_det</code>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>含义：</b>模型存储路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+<b>例如：</b><code>"cpu"</code>、<code>"gpu"</code>、<code>"npu"</code>、<code>"gpu:0"</code>、<code>"gpu:0,1"</code>。<br/>
+如指定多个设备，将进行并行推理。<br/>
+默认情况下，优先使用 GPU 0；若不可用则使用 CPU。
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，本地推理默认使用 <code>paddle_static</code> 引擎。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>当使用 Paddle Inference 的 TensorRT 子图引擎时设置的计算精度。<br/>
+<b>说明：</b>
+<b>可选项：</b><code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+<b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上推理时使用的线程数量。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>limit_side_len</code></td>
+<td><b>含义：</b>检测的图像边长限制：<code>int</code> 表示边长限制数值<br/>
+<b>说明：</b>
+如果设置为<code>None</code>, 将使用模型默认配置。</td>
+<td><code>int|None</td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>limit_type</code></td>
+<td><b>含义：</b>检测的图像边长限制，检测的边长限制类型<br/>
+<b>说明：</b>
+<code>"min"</code> 表示保证图像最短边不小于 det_limit_side_len，<code>"max"</code> 表示保证图像最长边不大于 <code>limit_side_len</code>。如果设置为 <code>None</code>，将使用模型默认配置。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>thresh</code></td>
+<td><b>含义：</b>像素得分阈值。<br/>
+<b>说明：</b>
+输出概率图中得分大于该阈值的像素点被认为是文本像素。可选大于0的float任意浮点数。如果设置为<code>None</code>。将使用模型默认配置。</td>
+<td><code>float|None</td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>box_thresh</code></td>
+<td><b>含义：</b>检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。<br/>
+<b>说明：</b>
+可选大于0的float任意浮点数。如果设置为<code>None</code>, 将使用模型默认配置。</td>
+<td><code>float|None</td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>unclip_ratio</code></td>
+<td><b>含义：</b>Vatti clipping算法的扩张系数，使用该方法对文字区域进行扩张。<br/>
+<b>说明：</b>
+可选大于0的任意浮点数。如果设置为<code>None</code>, 将使用模型默认配置。</td>
+<td><code>float|None</td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>input_shape</code></td>
+<td><b>含义：</b>模型输入图像尺寸，格式为 <code>(C, H, W)</code>。<br/>
+<b>说明：</b>
+若为 <code>None</code>，将使用模型默认配置。</td>
+<td><code>tuple|None</td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+* 调用印章文本检测模型的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。另外，本模块还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。<code>predict()</code>  方法参数有 <code>input</code> 、<code>batch_size</code>、 <code>limit_side_len</code>、 <code>limit_type</code>、 <code>thresh</code>、 <code>box_thresh</code>、 <code>max_candidates</code>、<code>unclip_ratio</code>，具体说明如下：
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>含义：</b>批大小。<br/>
+<b>说明：</b>
+可设置为任意正整数。</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+<tr>
+<td><code>limit_side_len</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>limit_type</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>box_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+* 对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为<code>json</code>文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将结果保存为图像格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+</table>
+
+* 此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">获取预测的<code>json</code>格式的结果</td>
+</tr>
+<tr>
+<td rowspan="1"><code>img</code></td>
+<td rowspan="1">获取格式为<code>dict</code>的可视化图像</td>
+</tr>
+</table>
+
+## 四、二次开发
+
+如果以上模型在您的场景上效果仍然不理想，您可以尝试以下步骤进行二次开发，此处以训练 <code>PP-OCRv4_server_seal_det</code> 举例，其他模型替换对应配置文件即可。首先，您需要准备文本检测的数据集，可以参考[印章文本检测 Demo 数据](https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_curve_det_dataset_examples.tar)的格式准备，准备好后，即可按照以下步骤进行模型训练和导出，导出后，可以将模型快速集成到上述 API 中。此处以印章文本检测 Demo 数据示例。在训练模型之前，请确保已经按照[安装文档](../installation.md)安装了 PaddleOCR 所需要的依赖。
+
+
+### 4.1 数据集、预训练模型准备
+
+#### 4.1.1 准备数据集
+
+```shell
+# 下载示例数据集
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_curve_det_dataset_examples.tar -P ./dataset
+tar -xf ./dataset/ocr_curve_det_dataset_examples.tar -C ./dataset/
+```
+
+#### 4.1.2 下载预训练模型
+
+```shell
+# 下载 PP-OCRv4_server_seal_det 预训练模型
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_seal_det_pretrained.pdparams
+```
+
+### 4.2 模型训练
+
+PaddleOCR 对代码进行了模块化，训练 `PP-OCRv4_server_seal_det` 模型时需要使用 `PP-OCRv4_server_seal_det` 的[配置文件](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/configs/det/PP-OCRv4/PP-OCRv4_server_seal_det.yml)。
+
+
+训练命令如下：
+
+```bash
+#单卡训练 (默认训练方式)
+python3 tools/train.py -c configs/det/PP-OCRv4/PP-OCRv4_server_seal_det.yml \
+   -o Global.pretrained_model=./PP-OCRv4_server_seal_det_pretrained.pdparams \
+   Train.dataset.data_dir=./dataset/ocr_curve_det_dataset_examples Train.dataset.label_file_list=./dataset/ocr_curve_det_dataset_examples/train.txt \
+   Eval.dataset.data_dir=./dataset/ocr_curve_det_dataset_examples Eval.dataset.label_file_list=./dataset/ocr_curve_det_dataset_examples/val.txt
+
+#多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/det/PP-OCRv4/PP-OCRv4_server_seal_det.yml \
+   -o Global.pretrained_model=./PP-OCRv4_server_seal_det_pretrained.pdparams \
+   Train.dataset.data_dir=./dataset/ocr_curve_det_dataset_examples Train.dataset.label_file_list=./dataset/ocr_curve_det_dataset_examples/train.txt \
+   Eval.dataset.data_dir=./dataset/ocr_curve_det_dataset_examples Eval.dataset.label_file_list=./dataset/ocr_curve_det_dataset_examples/val.txt
+```
+
+
+### 4.3 模型评估
+
+您可以评估已经训练好的权重，如，`output/xxx/xxx.pdparams`，使用如下命令进行评估：
+
+```bash
+# 注意将pretrained_model的路径设置为本地路径。若使用自行训练保存的模型，请注意修改路径和文件名为{path/to/weights}/{model_name}。
+# demo 测试集评估
+python3 tools/eval.py -c configs/det/PP-OCRv4/PP-OCRv4_server_seal_det.yml -o \
+    Global.pretrained_model=output/xxx/xxx.pdparams
+```
+
+### 4.4 模型导出
+
+```bash
+python3 tools/export_model.py -c configs/det/PP-OCRv4/PP-OCRv4_server_seal_det.yml -o \
+    Global.pretrained_model=output/xxx/xxx.pdparams \
+    Global.save_inference_dir="./PP-OCRv4_server_seal_det_infer/"
+```
+
+ 导出模型后，静态图模型会存放于当前目录的`./PP-OCRv4_server_seal_det_infer/`中，在该目录下，您将看到如下文件：
+ ```
+ ./PP-OCRv4_server_seal_det_infer/
+ ├── inference.json
+ ├── inference.pdiparams
+ ├── inference.yml
+ ```
+至此，二次开发完成，该静态图模型可以直接集成到 PaddleOCR 的 API 中。
+
+## 五、FAQ
diff --git a/docs/version3.x/module_usage/table_cells_detection.en.md b/docs/version3.x/module_usage/table_cells_detection.en.md
new file mode 100644
index 0000000..f01f5fe
--- /dev/null
+++ b/docs/version3.x/module_usage/table_cells_detection.en.md
@@ -0,0 +1,537 @@
+---
+
+comments: true
+
+---
+
+# Table Cell Detection Module Usage Tutorial
+
+## 1. Overview
+
+The Table Cell Detection Module is a key component of the table recognition task, responsible for locating and marking each cell region in table images. The performance of this module directly affects the accuracy and efficiency of the entire table recognition process. The Table Cell Detection Module typically outputs bounding boxes for each cell region, which are then passed as input to the table recognition pipeline for further processing.
+
+## 2. Supported Model List
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing. The "Regular Mode" values correspond to the local <code>paddle_static</code> inference engine.
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>RT-DETR-L_wired_table_cell_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wired_table_cell_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wired_table_cell_det_pretrained.pdparams">Training Model</a></td>
+<td rowspan="2">82.7</td>
+<td rowspan="2">33.47 / 27.02</td>
+<td rowspan="2">402.55 / 256.56</td>
+<td rowspan="2">124</td>
+<td rowspan="2">RT-DETR is a real-time end-to-end object detection model. The Baidu PaddlePaddle Vision team pre-trained on a self-built table cell detection dataset based on the RT-DETR-L as the base model, achieving good performance in detecting both wired and wireless table cells.</td>
+</tr>
+<tr>
+<td>RT-DETR-L_wireless_table_cell_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wireless_table_cell_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wireless_table_cell_det_pretrained.pdparams">Training Model</a></td>
+</tr>
+</table>
+
+<strong>Test Environment Description:</strong>
+
+  <ul>
+      <li><b>Performance Test Environment</b>
+          <ul>
+              <li><strong>Test Dataset:</strong> Internal evaluation set built by PaddleX.</li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>Inference Mode Explanation</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration</th>
+            <th>CPU Configuration</th>
+            <th>Acceleration Technology Combination</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Regular Mode</td>
+            <td>FP32 Precision / No TRT Acceleration</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>High-Performance Mode</td>
+            <td>Optimal combination of prior precision type and acceleration strategy</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Choose the optimal prior backend (Paddle/OpenVINO/TRT, etc.)</td>
+        </tr>
+    </tbody>
+</table>
+
+## 3. Quick Start
+
+> ❗ Before starting quickly, please first install the PaddleOCR wheel package. For details, please refer to the [installation tutorial](../installation.en.md).
+
+You can quickly experience it with one command:
+
+```bash
+paddleocr table_cells_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg
+
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr table_cells_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg \
+    --engine transformers
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following command:
+
+```bash
+# Use the onnxruntime engine for inference
+paddleocr table_cells_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg \
+    --engine onnxruntime
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+<b>Note: </b>The official models would be download from HuggingFace by default. If can't access to HuggingFace, please set the environment variable `PADDLE_PDX_MODEL_SOURCE="BOS"` to change the model source to BOS. In the future, more model sources will be supported.
+
+You can also integrate model inference from the table cell detection module into your project. Before running the following code, please download the [sample image](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg) locally.
+
+```python
+from paddleocr import TableCellsDetection
+model = TableCellsDetection(model_name="RT-DETR-L_wired_table_cell_det")
+output = model.predict("table_recognition.jpg", threshold=0.3, batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_img("./output/")
+    res.save_to_json("./output/res.json")
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following code:
+
+```python
+from paddleocr import TableCellsDetection
+model = TableCellsDetection(
+    model_name="RT-DETR-L_wired_table_cell_det",
+    engine="transformers",
+)
+output = model.predict("table_recognition.jpg", threshold=0.3, batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_img("./output/")
+    res.save_to_json("./output/res.json")
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following code:
+
+```python
+from paddleocr import TableCellsDetection
+model = TableCellsDetection(
+    model_name="RT-DETR-L_wired_table_cell_det",
+    engine="onnxruntime",
+)
+output = model.predict("table_recognition.jpg", threshold=0.3, batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_img("./output/")
+    res.save_to_json("./output/res.json")
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+If you want to use the trained model with the `paddle_dynamic` or `transformers` engine, refer to the [Weight Conversion](#52-weight-conversion) section in the [Inference Engine](#5-inference-engine) section below to convert the model from the `pdparams` format to the `safetensors` format using PaddleX.
+
+After running, the result obtained is:
+
+```
+{'res': {'input_path': 'table_recognition.jpg', 'page_index': None, 'boxes': [{'cls_id': 0, 'label': 'cell', 'score': 0.9698355197906494, 'coordinate': [2.3011515, 0, 546.29926, 30.530712]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9690820574760437, 'coordinate': [212.37508, 64.62493, 403.58868, 95.61413]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9668057560920715, 'coordinate': [212.46791, 30.311079, 403.7182, 64.62613]}, {'cls_id': 0, 'label': 'cell', 'score': 0.966505229473114, 'coordinate': [403.56082, 64.62544, 546.83215, 95.66117]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9662341475486755, 'coordinate': [109.48873, 64.66485, 212.5177, 95.631294]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9654079079627991, 'coordinate': [212.39197, 95.63037, 403.60852, 126.78792]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9653300642967224, 'coordinate': [2.2320926, 64.62229, 109.600494, 95.59732]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9639787673950195, 'coordinate': [403.5752, 30.562355, 546.98975, 64.61531]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9636150002479553, 'coordinate': [2.1537683, 30.410172, 109.568306, 64.62762]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9631900191307068, 'coordinate': [2.0534437, 95.57448, 109.57601, 126.71458]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9631181359291077, 'coordinate': [403.65976, 95.68139, 546.84766, 126.713394]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9614537358283997, 'coordinate': [109.56504, 30.391184, 212.65425, 64.6444]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9607433080673218, 'coordinate': [109.525795, 95.62622, 212.44917, 126.8258]}]}}
+```
+
+The parameter meanings are as follows:
+<ul>
+<li><code>input_path</code>: Path of the input image to be predicted</li>
+<li><code>page_index</code>: If the input is a PDF file, it indicates which page of the PDF it is; otherwise, it is `None`</li>
+<li><code>boxes</code>: Predicted bounding box information, a list of dictionaries. Each dictionary represents a detected object and contains the following information:
+    <ol start = "1" type ="1">
+  <li><code>cls_id</code>: Class ID, an integer</li>
+  <li><code>label</code>: Class label, a string</li>
+  <li><code>score</code>: Confidence of the bounding box, a float</li>
+  <li><code>coordinate</code>: Coordinates of the bounding box, a list of floats in the format <code>[xmin, ymin, xmax, ymax]</code></li>
+    </ol>
+</li>
+</ul>
+The visualized image is as follows:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/table_cells_detection/01.jpg">
+
+The relevant methods, parameters, etc., are described as follows:
+
+* <code>TableCellsDetection</code> instantiates the table cell detection model (taking <code>RT-DETR-L_wired_table_cell_det</code> as an example here), with specific explanations as follows:
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>Meaning:</b> Model name.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, <code>RT-DETR-L_wired_table_cell_det</code> will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>Meaning:</b>Model storage path.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>Device for inference.<br/>
+<b>Description:</b> 
+<b>For example:</b> <code>"cpu"</code>, <code>"gpu"</code>, <code>"npu"</code>, <code>"gpu:0"</code>, <code>"gpu:0,1"</code>.<br/>
+If multiple devices are specified, parallel inference will be performed.<br/>
+By default, GPU 0 is used if available; otherwise, CPU is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, <code>transformers</code>, and <code>onnxruntime</code>. When left as <code>None</code>, local inference uses the <code>paddle_static</code> engine by default. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b>Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b>Whether to use the Paddle Inference TensorRT subgraph engine.<br/> 
+<b>Description:</b> 
+If the model does not support acceleration through TensorRT, setting this flag will not enable acceleration.<br/>
+For Paddle with CUDA version 11.8, the compatible TensorRT version is 8.x (x>=6), and it is recommended to install TensorRT 8.6.1.6.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b>Computation precision when using the Paddle Inference TensorRT subgraph engine.<br/>
+<b>Description:</b> 
+<b>Options:</b> <code>"fp32"</code>, <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b>Whether to enable MKL-DNN acceleration for inference. <br/>
+<b>Description:</b> 
+If MKL-DNN is unavailable or the model does not support it, acceleration will not be used even if this flag is set.</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td><b>Meaning:</b>MKL-DNN cache capacity.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b>Number of threads to use for inference on CPUs.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>img_size</code></td>
+<td><b>Meaning:</b>Input image size.<br/>
+<b>Description:</b> 
+<ul><li><b>int</b>: e.g. <code>640</code>, resizes input image to 640x640</li><li><b>list</b>: e.g. <code>[640, 512]</code>, resizes input image to 640 width and 512 height</li></ul></td>
+<td><code>int|list|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>threshold</code></td>
+<td><b>Meaning:</b>Threshold for filtering out low-confidence prediction results.<br/>
+<b>Description:</b>
+<ul><li><b>float</b>: e.g. <code>0.2</code>, filters out all boxes with confidence below 0.2.</li><li><b>dict</b>: keys are <code>int</code> (class id), values are <code>float</code> thresholds, e.g. <code>{0: 0.45, 2: 0.48, 7: 0.4}</code>, applies thresholds to specific classes.</li><li><b>None</b>: uses the model's default configuration.</li></ul></td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+* Call the <code>predict()</code> method of the table cell detection model for inference prediction. This method will return a result list. Additionally, this module also provides a <code>predict_iter()</code> method. Both methods are consistent in terms of parameter acceptance and result return. The difference is that <code>predict_iter()</code> returns a <code>generator</code>, which can process and obtain prediction results step by step, suitable for handling large datasets or scenarios where memory saving is desired. You can choose to use either of these methods according to your actual needs. The <code>predict()</code> method has parameters <code>input</code>, <code>batch_size</code>, and <code>threshold</code>, with specific explanations as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td>
+<b>Meaning:</b>Input data to be predicted. Required.<br/>
+<b>Description:</b>
+Supports multiple input types:
+<ul>
+  <li><b>Python Var</b>: e.g., <code>numpy.ndarray</code> representing image data</li>
+  <li><b>str</b>: Local image file or PDF file path: <code>/root/data/img.jpg</code>; <b>URL</b>: Image or PDF file network URL: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">Example</a>; <b>Directory</b>: Should contain images for prediction, e.g., <code>/root/data/</code> (currently, PDF files in directories are not supported, PDF files need to be specified by file path)</li>
+  <li><b>list</b>: List elements should be of the above types, e.g., <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>Meaning:</b>Batch size.<br/>
+<b>Description:</b>
+Positive integer.</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+<tr>
+<td><code>threshold</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters.<br/>
+<b>Description:</b>If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|dict|None</code></td>
+<td>None</td>
+</tr>
+</table>
+
+* Process the prediction results. The prediction result for each sample is a corresponding Result object, which supports printing, saving as an image, and saving as a <code>json</code> file:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Description</th>
+<th>Parameter</th>
+<th>Type</th>
+<th>Parameter Description</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">Print result to terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output content using <code>JSON</code> indentation</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specifies the indentation level to beautify the output <code>JSON</code> data, making it more readable, effective only when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Controls whether to escape non-<code>ASCII</code> characters into <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> will retain the original characters, effective only when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">Save the result as a json format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The path to save the file. When specified as a directory, the saved file is named consistent with the input file type.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specifies the indentation level to beautify the output <code>JSON</code> data, making it more readable, effective only when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Controls whether to escape non-<code>ASCII</code> characters into <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> will retain the original characters, effective only when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save the result as an image format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The path to save the file. When specified as a directory, the saved file is named consistent with the input file type.</td>
+<td>None</td>
+</tr>
+</table>
+
+* Additionally, the result can be obtained through attributes that provide the visualized images with results and the prediction results, as follows:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">Get the prediction result in <code>json</code> format</td>
+</tr>
+<tr>
+<td rowspan = "1"><code>img</code></td>
+<td rowspan = "1">Get the visualized image</td>
+</tr>
+
+</table>
+
+## 4. Secondary Development
+
+Since PaddleOCR does not directly provide training for the table cell detection module, if you need to train a table cell detection model, you can refer to the [PaddleX Table Cell Detection Module Secondary Development](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/table_cells_detection.html#iv-secondary-development) section for training. The trained model can be seamlessly integrated into the PaddleOCR API for inference.
+
+If you want to use the `paddle_dynamic` or `transformers` engine with the trained model, please refer to the [Weight Conversion](#52-weight-conversion) section in [Inference Engine](#5-inference-engine) later in this document to convert the model from the `pdparams` format to the `safetensors` format using PaddleX.
+
+## 5. Inference Engine
+
+For detailed descriptions, values, compatibility rules, and examples of the inference engine, please refer to <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration Description</a>.
+
+### 5.1 Speed Data
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">RT-DETR-L_wired_table_cell_det</td>
+            <td>paddle_static</td>
+            <td>3.59</td>
+            <td>23.11</td>
+            <td>0.14</td>
+            <td>27.02</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>4.04</td>
+            <td>70.38</td>
+            <td>0.15</td>
+            <td>75.49</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>3.69</td>
+            <td>37.30</td>
+            <td>0.71</td>
+            <td>42.10</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>2.70</td>
+            <td>9.10</td>
+            <td>0.12</td>
+            <td>12.07</td>
+        </tr>
+        <tr>
+            <td rowspan="4">RT-DETR-L_wireless_table_cell_det</td>
+            <td>paddle_static</td>
+            <td>3.77</td>
+            <td>23.44</td>
+            <td>0.14</td>
+            <td>27.52</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>4.01</td>
+            <td>69.97</td>
+            <td>0.15</td>
+            <td>75.10</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>3.69</td>
+            <td>37.11</td>
+            <td>0.71</td>
+            <td>41.91</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>2.77</td>
+            <td>9.11</td>
+            <td>0.12</td>
+            <td>12.14</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>Test Environment Description:</strong>
+<ul>
+    <li><strong>Test Data:</strong> <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg">Sample Image</a></li>
+    <li><strong>Hardware Configuration:</strong>
+        <ul>
+            <li>GPU: NVIDIA A100 40G</li>
+            <li>CPU: Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>Software Environment:</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 5.2 Weight Conversion
+
+When using the inference engine, the system will automatically download the official pre-trained model. If you need to use a self-trained model with the `paddle_dynamic` or `transformers` engine, please refer to the [PaddleX Table Cell Detection Module Weight Conversion](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/table_cells_detection.html#442) section to convert the model from the `pdparams` format to the `safetensors` format using PaddleX. This allows seamless integration into the PaddleOCR API for inference. If you need to use a self-trained model with the `onnxruntime` engine, refer to [PaddleX Obtain ONNX Models](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html) to obtain the ONNX model, so it can be seamlessly integrated into the PaddleOCR API for inference.
+
+## 6. FAQ
diff --git a/docs/version3.x/module_usage/table_cells_detection.md b/docs/version3.x/module_usage/table_cells_detection.md
new file mode 100644
index 0000000..743515c
--- /dev/null
+++ b/docs/version3.x/module_usage/table_cells_detection.md
@@ -0,0 +1,548 @@
+---
+comments: true
+---
+
+# 表格单元格检测模块使用教程
+
+## 一、概述
+
+表格单元格检测模块是表格识别任务的关键组成部分，负责在表格图像中定位和标记每个单元格区域，该模块的性能直接影响到整个表格识别过程的准确性和效率。表格单元格检测模块通常会输出各个单元格区域的边界框（Bounding Boxes），这些边界框将作为输入传递给表格识别相关产线进行后续处理。
+
+## 二、支持模型列表
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。表格中的“常规模式”耗时对应本地 <code>paddle_static</code> 推理引擎。
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>RT-DETR-L_wired_table_cell_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wired_table_cell_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wired_table_cell_det_pretrained.pdparams">训练模型</a></td>
+<td rowspan="2">82.7</td>
+<td rowspan="2">33.47 / 27.02</td>
+<td rowspan="2">402.55 / 256.56</td>
+<td rowspan="2">124</td>
+<td rowspan="2">RT-DETR 是一个实时的端到端目标检测模型。百度飞桨视觉团队基于 RT-DETR-L 作为基础模型，在自建表格单元格检测数据集上完成预训练，实现了对有线表格、无线表格均有较好性能的表格单元格检测。
+</td>
+</tr>
+<tr>
+<td>RT-DETR-L_wireless_table_cell_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wireless_table_cell_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wireless_table_cell_det_pretrained.pdparams">训练模型</a></td>
+</tr>
+</table>
+
+<strong>测试环境说明:</strong>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+              <li><strong>测试数据集：</strong>自建的内部评测集。</li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+
+## 三、快速开始
+
+> ❗ 在快速开始前，请先安装 PaddleOCR 的 wheel 包，详细请参考 [安装教程](../installation.md)。
+
+使用一行命令即可快速体验：
+
+```bash
+paddleocr table_cells_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下命令：
+
+```bash
+# 使用 transformers 引擎进行推理
+paddleocr table_cells_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg \
+    --engine transformers
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下命令：
+
+```bash
+# 使用 onnxruntime 引擎进行推理
+paddleocr table_cells_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg \
+    --engine onnxruntime
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+<b>注：</b>PaddleOCR 官方模型默认从 HuggingFace 获取，如运行环境访问 HuggingFace 不便，可通过环境变量修改模型源为 BOS：`PADDLE_PDX_MODEL_SOURCE="BOS"`，未来将支持更多主流模型源；
+
+您也可以将表格单元格检测的模块中的模型推理集成到您的项目中。运行以下代码前，请您下载[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg)到本地。
+
+```python
+from paddleocr import TableCellsDetection
+model = TableCellsDetection(model_name="RT-DETR-L_wired_table_cell_det")
+output = model.predict("table_recognition.jpg", threshold=0.3, batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_img("./output/")
+    res.save_to_json("./output/res.json")
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下代码：
+
+```python
+from paddleocr import TableCellsDetection
+model = TableCellsDetection(
+    model_name="RT-DETR-L_wired_table_cell_det",
+    engine="transformers",
+)
+output = model.predict("table_recognition.jpg", threshold=0.3, batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_img("./output/")
+    res.save_to_json("./output/res.json")
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下代码：
+
+```python
+from paddleocr import TableCellsDetection
+model = TableCellsDetection(
+    model_name="RT-DETR-L_wired_table_cell_det",
+    engine="onnxruntime",
+)
+output = model.predict("table_recognition.jpg", threshold=0.3, batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_img("./output/")
+    res.save_to_json("./output/res.json")
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+训练后的模型如果想使用 `paddle_dynamic` 或 `transformers` 引擎，请参考后文 [推理引擎](#五推理引擎) 中的 [权重转换](#52-权重转换) 部分将模型由 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式。
+
+运行后，得到的结果为：
+
+```
+{'res': {'input_path': 'table_recognition.jpg', 'page_index': None, 'boxes': [{'cls_id': 0, 'label': 'cell', 'score': 0.9698355197906494, 'coordinate': [2.3011515, 0, 546.29926, 30.530712]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9690820574760437, 'coordinate': [212.37508, 64.62493, 403.58868, 95.61413]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9668057560920715, 'coordinate': [212.46791, 30.311079, 403.7182, 64.62613]}, {'cls_id': 0, 'label': 'cell', 'score': 0.966505229473114, 'coordinate': [403.56082, 64.62544, 546.83215, 95.66117]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9662341475486755, 'coordinate': [109.48873, 64.66485, 212.5177, 95.631294]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9654079079627991, 'coordinate': [212.39197, 95.63037, 403.60852, 126.78792]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9653300642967224, 'coordinate': [2.2320926, 64.62229, 109.600494, 95.59732]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9639787673950195, 'coordinate': [403.5752, 30.562355, 546.98975, 64.61531]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9636150002479553, 'coordinate': [2.1537683, 30.410172, 109.568306, 64.62762]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9631900191307068, 'coordinate': [2.0534437, 95.57448, 109.57601, 126.71458]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9631181359291077, 'coordinate': [403.65976, 95.68139, 546.84766, 126.713394]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9614537358283997, 'coordinate': [109.56504, 30.391184, 212.65425, 64.6444]}, {'cls_id': 0, 'label': 'cell', 'score': 0.9607433080673218, 'coordinate': [109.525795, 95.62622, 212.44917, 126.8258]}]}}
+```
+
+参数含义如下：
+<ul>
+<li><code>input_path</code>：输入的待预测图像的路径</li>
+<li><code>page_index</code>：如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+<li><code>boxes</code>：预测的目标框信息，一个字典列表。每个字典代表一个检出的目标，包含以下信息：
+    <ol start="1" type="1">
+    <li><code>cls_id</code>：类别ID，一个整数</li>
+    <li><code>label</code>：类别标签，一个字符串</li>
+    <li><code>score</code>：目标框置信度，一个浮点数</li>
+    <li><code>coordinate</code>：目标框坐标，一个浮点数列表，格式为<code>[xmin, ymin, xmax, ymax]</code></li>
+    </ol>
+</li>
+</ul>
+可视化图像如下：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/table_cells_detection/01.jpg">
+
+相关方法、参数等说明如下：
+
+* <code>TableCellsDetection</code>实例化表格单元格检测模型（此处以<code>RT-DETR-L_wired_table_cell_det</code>为例），具体说明如下：
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>含义：</b>模型名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，则使用<code>RT-DETR-L_wired_table_cell_det</code>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>含义：</b>模型存储路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+<b>例如：</b><code>"cpu"</code>、<code>"gpu"</code>、<code>"npu"</code>、<code>"gpu:0"</code>、<code>"gpu:0,1"</code>。<br/>
+如指定多个设备，将进行并行推理。<br/>
+默认情况下，优先使用 GPU 0；若不可用则使用 CPU。
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>、<code>onnxruntime</code>。保持为默认值 <code>None</code> 时，本地推理默认使用 <code>paddle_static</code> 引擎。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>当使用 Paddle Inference 的 TensorRT 子图引擎时设置的计算精度。<br/>
+<b>说明：</b>
+<b>可选项：</b><code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+<b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上推理时使用的线程数量。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>img_size</code></td>
+<td><b>含义：</b>输入图像大小。<br/>
+<b>说明：</b>
+<ul>
+<li><b>int</b>：如<code>640</code>，表示将输入图像resize到640x640大小。</li>
+<li><b>list</b>：如<code>[640, 512]</code>，表示将输入图像resize到宽为640，高为512大小。</li>
+</ul>
+</td>
+<td><code>int|list|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>threshold</code></td>
+<td><b>含义：</b>用于过滤掉低置信度预测结果的阈值。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：如 <code>0.2</code>，表示过滤掉所有阈值小于0.2的目标框。/li>
+<li><b>dict</b>：字典的键为 <code>int</code> 类型，代表类别ID；值为 <code>float</code> 类型阈值。如 <code>{0: 0.45, 2: 0.48, 7: 0.4}</code>，表示对类别ID为0的类别应用阈值0.45、类别ID为1的类别应用阈值0.48、类别ID为7的类别应用阈值0.4。</li>
+<li><b>None</b>：使用模型默认配置。
+</ul>
+</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</table>
+
+* 调用目标检测模型的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。另外，本模块还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。<code>predict()</code> 方法参数有 <code>input</code>、<code>batch_size</code>和<code>threshold</code>，具体说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>含义：</b>批大小。<br/>
+<b>说明：</b>
+可设置为任意正整数。</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+<tr>
+<td><code>threshold</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|dict|None</code></td>
+<td>None</td>
+</tr>
+</table>
+
+* 对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为<code>json</code>文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将结果保存为图像格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+</table>
+
+* 此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">获取预测的<code>json</code>格式的结果</td>
+</tr>
+<tr>
+<td rowspan = "1"><code>img</code></td>
+<td rowspan = "1">获取可视化图像</td>
+</tr>
+
+</table>
+
+## 四、二次开发
+
+由于 PaddleOCR 并不直接提供表格单元格检测模块的训练，因此，如果需要训练表格单元格检测模型，可以参考 [PaddleX 表格单元格检测模块二次开发](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/table_cells_detection.html#_4)部分进行训练。训练后的模型可以无缝集成到 PaddleOCR 的 API 中进行推理。
+
+训练后的模型如果想使用 `paddle_dynamic` 或 `transformers` 引擎，请参考后文 [推理引擎](#五推理引擎) 中的 [权重转换](#52-权重转换) 部分将模型由 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式。
+
+## 五、推理引擎 {#五推理引擎}
+
+关于推理引擎的详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。
+
+### 5.1 速度数据
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">RT-DETR-L_wired_table_cell_det</td>
+            <td>paddle_static</td>
+            <td>3.59</td>
+            <td>23.11</td>
+            <td>0.14</td>
+            <td>27.02</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>4.04</td>
+            <td>70.38</td>
+            <td>0.15</td>
+            <td>75.49</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>3.69</td>
+            <td>37.30</td>
+            <td>0.71</td>
+            <td>42.10</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>2.70</td>
+            <td>9.10</td>
+            <td>0.12</td>
+            <td>12.07</td>
+        </tr>
+        <tr>
+            <td rowspan="4">RT-DETR-L_wireless_table_cell_det</td>
+            <td>paddle_static</td>
+            <td>3.77</td>
+            <td>23.44</td>
+            <td>0.14</td>
+            <td>27.52</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>4.01</td>
+            <td>69.97</td>
+            <td>0.15</td>
+            <td>75.10</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>3.69</td>
+            <td>37.11</td>
+            <td>0.71</td>
+            <td>41.91</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>2.77</td>
+            <td>9.11</td>
+            <td>0.12</td>
+            <td>12.14</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>测试环境说明:</strong>
+<ul>
+    <li><strong>测试数据：</strong><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg">示例图片</a></li>
+    <li><strong>硬件配置：</strong>
+        <ul>
+            <li>GPU：NVIDIA A100 40G</li>
+            <li>CPU：Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>软件环境：</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 5.2 权重转换 {#52-权重转换}
+
+使用推理引擎时，系统会自动下载官方预训练模型。若需使用自训练模型配合 `paddle_dynamic` 或 `transformers` 引擎，请参考 [PaddleX 表格单元格检测模块权重转换](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/table_cells_detection.html#442) 部分，将 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式，即可无缝集成到 PaddleOCR 的 API 中进行推理。若需使用自训练模型配合`onnxruntime`引擎，请参考[PaddleX 获取 ONNX 模型](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html)获取onnx模型，即可无缝集成到 PaddleOCR 的 API 中进行推理。
+
+## 六、FAQ
diff --git a/docs/version3.x/module_usage/table_classification.en.md b/docs/version3.x/module_usage/table_classification.en.md
new file mode 100644
index 0000000..2feb28f
--- /dev/null
+++ b/docs/version3.x/module_usage/table_classification.en.md
@@ -0,0 +1,463 @@
+---
+
+comments: true
+
+---
+
+# Table Classification Module Usage Tutorial
+
+## 1. Overview
+
+The Table Classification Module is a key component in computer vision systems, responsible for classifying input table images. The performance of this module directly affects the accuracy and efficiency of the entire table recognition process. The Table Classification Module typically receives table images as input and, using deep learning algorithms, classifies them into predefined categories based on the characteristics and content of the images, such as wired and wireless tables. The classification results from the Table Classification Module serve as output for use in table recognition pipelines.
+
+## 2. Supported Model List
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing. The "Regular Mode" values correspond to the local <code>paddle_static</code> inference engine.
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Top1 Acc(%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+</tr>
+<tr>
+<td>PP-LCNet_x1_0_table_cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_table_cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_table_cls_pretrained.pdparams">Training Model</a></td>
+<td>94.2</td>
+<td>2.62 / 0.60</td>
+<td>3.17 / 1.14</td>
+<td>6.6</td>
+</tr>
+</table>
+
+<strong>Test Environment Description:</strong>
+
+  <ul>
+      <li><b>Performance Test Environment</b>
+          <ul>
+              <li><strong>Test Dataset:</strong> Internal evaluation dataset built by PaddleX.</li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>Inference Mode Explanation</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration</th>
+            <th>CPU Configuration</th>
+            <th>Acceleration Technology Combination</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Regular Mode</td>
+            <td>FP32 Precision / No TRT Acceleration</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>High-Performance Mode</td>
+            <td>Optimal combination of prior precision type and acceleration strategy</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Choose the optimal prior backend (Paddle/OpenVINO/TRT, etc.)</td>
+        </tr>
+    </tbody>
+</table>
+
+## 3. Quick Start
+
+> ❗ Before starting quickly, please first install the PaddleOCR wheel package. For details, please refer to the [installation tutorial](../installation.en.md).
+
+You can quickly experience it with one command:
+
+```bash
+paddleocr table_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr table_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg \
+    --engine transformers
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following command:
+
+```bash
+# Use the onnxruntime engine for inference
+paddleocr table_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg \
+    --engine onnxruntime
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+<b>Note: </b>The official models would be download from HuggingFace by default. If can't access to HuggingFace, please set the environment variable `PADDLE_PDX_MODEL_SOURCE="BOS"` to change the model source to BOS. In the future, more model sources will be supported.
+
+You can also integrate model inference from the table classification module into your project. Before running the following code, please download the [sample image](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg) locally.
+
+```python
+from paddleocr import TableClassification
+model = TableClassification(model_name="PP-LCNet_x1_0_table_cls")
+output = model.predict("table_recognition.jpg", batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_json("./output/res.json")
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following code:
+
+```python
+from paddleocr import TableClassification
+model = TableClassification(
+    model_name="PP-LCNet_x1_0_table_cls",
+    engine="transformers",
+)
+output = model.predict("table_recognition.jpg", batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_json("./output/res.json")
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following code:
+
+```python
+from paddleocr import TableClassification
+model = TableClassification(
+    model_name="PP-LCNet_x1_0_table_cls",
+    engine="onnxruntime",
+)
+output = model.predict("table_recognition.jpg", batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_json("./output/res.json")
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+If you want to use the trained model with the `paddle_dynamic` or `transformers` engine, refer to the [Weight Conversion](#52-weight-conversion) section in the [Inference Engine](#5-inference-engine) section below to convert the model from the `pdparams` format to the `safetensors` format using PaddleX.
+
+After running, the result obtained is:
+
+```
+{'res': {'input_path': 'table_recognition.jpg', 'page_index': None, 'class_ids': array([0, 1], dtype=int32), 'scores': array([0.84421, 0.15579], dtype=float32), 'label_names': ['wired_table', 'wireless_table']}}
+```
+
+The parameter meanings are as follows:
+<ul>
+<li><code>input_path</code>: Path of the input image</li>
+<li><code>page_index</code>: If the input is a PDF file, it indicates which page of the PDF it is; otherwise, it is <code>None</code></li>
+<li><code>class_ids</code>: Class IDs of the prediction results</li>
+<li><code>scores</code>: Confidence scores of the prediction results</li>
+<li><code>label_names</code>: Class names of the prediction results</li>
+</ul>
+
+  
+The visualized image is as follows:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/table_classification/01.jpg">
+
+The relevant methods, parameters, etc., are described as follows:
+
+* <code>TableClassification</code> instantiates the table classification model (taking <code>PP-LCNet_x1_0_table_cls</code> as an example here), with specific explanations as follows:
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>Meaning:</b>Model name. <br/>
+<b>Description:</b>
+If set to <code>None</code>, <code>PP-LCNet_x1_0_table_cls</code> will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>Meaning:</b>Model storage path.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>Device for inference.<br/>
+<b>Description:</b>
+<b>For example:</b> <code>"cpu"</code>, <code>"gpu"</code>, <code>"npu"</code>, <code>"gpu:0"</code>, <code>"gpu:0,1"</code>.<br/>
+If multiple devices are specified, parallel inference will be performed.<br/>
+By default, GPU 0 is used if available; otherwise, CPU is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, <code>transformers</code>, and <code>onnxruntime</code>. When left as <code>None</code>, local inference uses the <code>paddle_static</code> engine by default. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b>Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b>Whether to use the Paddle Inference TensorRT subgraph engine. <br/>
+<b>Description:</b>
+If the model does not support acceleration through TensorRT, setting this flag will not enable acceleration.<br/>
+For Paddle with CUDA version 11.8, the compatible TensorRT version is 8.x (x>=6), and it is recommended to install TensorRT 8.6.1.6.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b>Computation precision when using the Paddle Inference TensorRT subgraph engine.<br/>
+<b>Description:</b>
+<b>Options:</b> <code>"fp32"</code>, <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b>Whether to enable MKL-DNN acceleration for inference. <br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support it, acceleration will not be used even if this flag is set.</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td><b>Meaning:</b>MKL-DNN cache capacity.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b>Number of threads to use for inference on CPUs.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+</tbody>
+</table>
+
+* Call the <code>predict()</code> method of the table classification model for inference prediction. This method will return a result list. Additionally, this module also provides a  <code>predict_iter()</code> method. Both methods are consistent in terms of parameter acceptance and result return. The difference is that <code>predict_iter()</code>  returns a <code>generator</code>, which can process and obtain prediction results step by step, suitable for handling large datasets or scenarios where memory saving is desired. You can choose to use either of these methods according to your actual needs. The <code>predict()</code> method has parameters <code>input</code> and <code>batch_size</code>, with specific explanations as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td>
+<b>Meaning:</b>Input data to be predicted. Required.<br/>
+<b>Description:</b>
+ Supports multiple input types:
+<ul>
+  <li><b>Python Var</b>: e.g., <code>numpy.ndarray</code> representing image data</li>
+  <li><b>str</b>: Local image file or PDF file path: <code>/root/data/img.jpg</code>; <b>URL</b>: Image or PDF file network URL: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">Example</a>; <b>Directory</b>: Should contain images for prediction, e.g., <code>/root/data/</code> (currently, PDF files in directories are not supported, PDF files need to be specified by file path)</li>
+  <li><b>list</b>: List elements should be of the above types, e.g., <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>Meaning:</b>Batch size.<br/>
+<b>Description:</b>
+ Positive integer.</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+
+* Process the prediction results. The prediction result for each sample is a corresponding Result object, which supports printing, saving as an image, and saving as a <code>json</code> file:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Description</th>
+<th>Parameter</th>
+<th>Type</th>
+<th>Parameter Description</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">Print result to terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output content using <code>JSON</code> indentation</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specifies the indentation level to beautify the output <code>JSON</code> data, making it more readable, effective only when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Controls whether to escape non-<code>ASCII</code> characters into <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> will retain the original characters, effective only when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">Save the result as a json format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The path to save the file. When specified as a directory, the saved file is named consistent with the input file type.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specifies the indentation level to beautify the output <code>JSON</code> data, making it more readable, effective only when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Controls whether to escape non-<code>ASCII</code> characters into <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> will retain the original characters, effective only when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+</table>
+
+* Additionally, the result can be obtained through attributes that provide the visualized images with results and the prediction results, as follows:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">Get the prediction result in <code>json</code> format</td>
+</tr>
+<tr>
+<td rowspan = "1"><code>img</code></td>
+<td rowspan = "1">Get the visualized image</td>
+</tr>
+</table>
+
+## 4. Secondary Development
+
+Since PaddleOCR does not directly provide training for the table classification module, if you need to train a table classification model, you can refer to the [PaddleX Table Classification Module Secondary Development](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/table_classification.html#iv-secondary-development) section for training. The trained model can be seamlessly integrated into the PaddleOCR API for inference.
+
+If you want to use the `paddle_dynamic` or `transformers` engine with the trained model, please refer to the [Weight Conversion](#52-weight-conversion) section in [Inference Engine](#5-inference-engine) later in this document to convert the model from the `pdparams` format to the `safetensors` format using PaddleX.
+
+## 5. Inference Engine
+
+For detailed descriptions, values, compatibility rules, and examples of the inference engine, please refer to <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration Description</a>.
+
+### 5.1 Speed Data
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">PP-LCNet_x1_0_table_cls</td>
+            <td>paddle_static</td>
+            <td>3.56</td>
+            <td>3.38</td>
+            <td>0.06</td>
+            <td>7.11</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>3.57</td>
+            <td>7.77</td>
+            <td>0.07</td>
+            <td>11.52</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>9.30</td>
+            <td>3.72</td>
+            <td>0.15</td>
+            <td>14.05</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>3.23</td>
+            <td>0.88</td>
+            <td>0.05</td>
+            <td>4.25</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>Test Environment Description:</strong>
+<ul>
+    <li><strong>Test Data:</strong> <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_image_classification_001.jpg">Sample Image</a></li>
+    <li><strong>Hardware Configuration:</strong>
+        <ul>
+            <li>GPU: NVIDIA A100 40G</li>
+            <li>CPU: Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>Software Environment:</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 5.2 Weight Conversion
+
+When using the inference engine, the system will automatically download the official pre-trained model. If you need to use a self-trained model with the `paddle_dynamic` or `transformers` engine, please refer to the [PaddleX Table Classification Module Weight Conversion](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/table_classification.html#442) section to convert the model from the `pdparams` format to the `safetensors` format using PaddleX. This allows seamless integration into the PaddleOCR API for inference. If you need to use a self-trained model with the `onnxruntime` engine, refer to [PaddleX Obtain ONNX Models](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html) to obtain the ONNX model, so it can be seamlessly integrated into the PaddleOCR API for inference.
+
+## 6. FAQ
diff --git a/docs/version3.x/module_usage/table_classification.md b/docs/version3.x/module_usage/table_classification.md
new file mode 100644
index 0000000..9c9ba9c
--- /dev/null
+++ b/docs/version3.x/module_usage/table_classification.md
@@ -0,0 +1,464 @@
+---
+comments: true
+---
+
+# 表格分类模块使用教程
+
+## 一、概述
+
+表格分类模块是计算机视觉系统中的关键组成部分，负责对输入的表格图像进行分类，该模块的性能直接影响到整个表格识别过程的准确性和效率。表格分类模块通常会接收表格图像作为输入，然后通过深度学习算法，根据图像的特性和内容，将其分类到预定义的类别中，例如有线表和无线表。表格分类模块的分类结果将作为输出，供表格识别相关产线使用。
+
+## 二、支持模型列表
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。表格中的“常规模式”耗时对应本地 <code>paddle_static</code> 推理引擎。
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>Top1 Acc(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+</tr>
+<tr>
+<td>PP-LCNet_x1_0_table_cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_table_cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_table_cls_pretrained.pdparams">训练模型</a></td>
+<td>94.2</td>
+<td>2.62 / 0.60</td>
+<td>3.17 / 1.14</td>
+<td>6.6</td>
+</tr>
+</table>
+
+<strong>测试环境说明:</strong>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+              <li><strong>测试数据集：</strong>自建的内部评测数据集。</li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+
+## 三、快速开始
+
+> ❗ 在快速开始前，请先安装 PaddleOCR 的 wheel 包，详细请参考 [安装教程](../installation.md)。
+
+使用一行命令即可快速体验：
+
+```bash
+paddleocr table_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下命令：
+
+```bash
+# 使用 transformers 引擎进行推理
+paddleocr table_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg \
+    --engine transformers
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下命令：
+
+```bash
+# 使用 onnxruntime 引擎进行推理
+paddleocr table_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg \
+    --engine onnxruntime
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+<b>注：</b>PaddleOCR 官方模型默认从 HuggingFace 获取，如运行环境访问 HuggingFace 不便，可通过环境变量修改模型源为 BOS：`PADDLE_PDX_MODEL_SOURCE="BOS"`，未来将支持更多主流模型源；
+
+您也可以将表格分类的模块中的模型推理集成到您的项目中。运行以下代码前，请您下载[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg)到本地。
+
+```python
+from paddleocr import TableClassification
+model = TableClassification(model_name="PP-LCNet_x1_0_table_cls")
+output = model.predict("table_recognition.jpg", batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_json("./output/res.json")
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下代码：
+
+```python
+from paddleocr import TableClassification
+model = TableClassification(
+    model_name="PP-LCNet_x1_0_table_cls",
+    engine="transformers",
+)
+output = model.predict("table_recognition.jpg", batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_json("./output/res.json")
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下代码：
+
+```python
+from paddleocr import TableClassification
+model = TableClassification(
+    model_name="PP-LCNet_x1_0_table_cls",
+    engine="onnxruntime",
+)
+output = model.predict("table_recognition.jpg", batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_json("./output/res.json")
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+训练后的模型如果想使用 `paddle_dynamic` 或 `transformers` 引擎，请参考后文 [推理引擎](#五推理引擎) 中的 [权重转换](#52-权重转换) 部分将模型由 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式。
+
+运行后，得到的结果为：
+
+```
+{'res': {'input_path': 'table_recognition.jpg', 'page_index': None, 'class_ids': array([0, 1], dtype=int32), 'scores': array([0.84421, 0.15579], dtype=float32), 'label_names': ['wired_table', 'wireless_table']}}
+```
+
+运行结果参数含义如下：
+<ul>
+    <li><code>input_path</code>：表示输入图片的路径</li>
+    <li><code>page_index</code>：如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+    <li><code>class_ids</code>：表示预测结果的类别id</li>
+    <li><code>scores</code>：表示预测结果的置信度</li>
+    <li><code>label_names</code>：表示预测结果的类别名</li>
+</ul>
+
+可视化图像如下：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/table_classification/01.jpg">
+
+相关方法、参数等说明如下：
+
+* <code>TableClassification</code>实例化表格分类模型（此处以<code>PP-LCNet_x1_0_table_cls</code>为例），具体说明如下：
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>含义：</b>模型名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，则使用<code>PP-LCNet_x1_0_table_cls</code>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>含义：</b>模型存储路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+<b>例如：</b><code>"cpu"</code>、<code>"gpu"</code>、<code>"npu"</code>、<code>"gpu:0"</code>、<code>"gpu:0,1"</code>。<br/>
+如指定多个设备，将进行并行推理。<br/>
+默认情况下，优先使用 GPU 0；若不可用则使用 CPU。
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>、<code>onnxruntime</code>。保持为默认值 <code>None</code> 时，本地推理默认使用 <code>paddle_static</code> 引擎。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>当使用 Paddle Inference 的 TensorRT 子图引擎时设置的计算精度。<br/>
+<b>说明：</b>
+<b>可选项：</b><code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+<b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上推理时使用的线程数量。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+</tbody>
+</table>
+
+* 调用表格分类模型的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。另外，本模块还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。<code>predict()</code> 方法参数有 <code>input</code> 和 <code>batch_size</code>，具体说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_doc_preprocessor_002.png">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>含义：</b>批大小。<br/>
+<b>说明：</b>
+可设置为任意正整数。</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+
+* 对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为<code>json</code>文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+</table>
+
+* 此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">获取预测的<code>json</code>格式的结果</td>
+</tr>
+<tr>
+<td rowspan = "1"><code>img</code></td>
+<td rowspan = "1">获取可视化图像</td>
+</tr>
+</table>
+
+## 四、二次开发
+
+由于 PaddleOCR 并不直接提供表格分类模块的训练，因此，如果需要训练表格分类模型，可以参考 [PaddleX 表格分类模块二次开发](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/table_classification.html#_5)部分进行训练。训练后的模型可以无缝集成到 PaddleOCR 的 API 中进行推理。
+
+训练后的模型如果想使用 `paddle_dynamic` 或 `transformers` 引擎，请参考后文 [推理引擎](#五推理引擎) 中的 [权重转换](#52-权重转换) 部分将模型由 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式。
+
+## 五、推理引擎 {#五推理引擎}
+
+关于推理引擎的详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。
+
+### 5.1 速度数据
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">PP-LCNet_x1_0_table_cls</td>
+            <td>paddle_static</td>
+            <td>3.56</td>
+            <td>3.38</td>
+            <td>0.06</td>
+            <td>7.11</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>3.57</td>
+            <td>7.77</td>
+            <td>0.07</td>
+            <td>11.52</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>9.30</td>
+            <td>3.72</td>
+            <td>0.15</td>
+            <td>14.05</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>3.23</td>
+            <td>0.88</td>
+            <td>0.05</td>
+            <td>4.25</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>测试环境说明:</strong>
+<ul>
+    <li><strong>测试数据：</strong><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_image_classification_001.jpg">示例图片</a></li>
+    <li><strong>硬件配置：</strong>
+        <ul>
+            <li>GPU：NVIDIA A100 40G</li>
+            <li>CPU：Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>软件环境：</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 5.2 权重转换 {#52-权重转换}
+
+使用推理引擎时，系统会自动下载官方预训练模型。若需使用自训练模型配合 `paddle_dynamic` 或 `transformers` 引擎，请参考 [PaddleX 表格分类模块权重转换](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/table_classification.html#442) 部分，将 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式，即可无缝集成到 PaddleOCR 的 API 中进行推理。若需使用自训练模型配合`onnxruntime`引擎，请参考[PaddleX 获取 ONNX 模型](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html)获取onnx模型，即可无缝集成到 PaddleOCR 的 API 中进行推理。
+
+## 六、FAQ
diff --git a/docs/version3.x/module_usage/table_structure_recognition.en.md b/docs/version3.x/module_usage/table_structure_recognition.en.md
new file mode 100644
index 0000000..7e8da27
--- /dev/null
+++ b/docs/version3.x/module_usage/table_structure_recognition.en.md
@@ -0,0 +1,598 @@
+---
+comments: true
+---
+
+# Table Structure Recognition Module Tutorial
+
+## 1. Overview
+
+Table structure recognition is an important component of table recognition systems, capable of converting non-editable table images into editable table formats (such as HTML). The goal of table structure recognition is to identify the positions of rows, columns, and cells in tables. The performance of this module directly affects the accuracy and efficiency of the entire table recognition system. The table structure recognition module usually outputs HTML code for the table area, which is then passed as input to the table recognition pipeline for further processing.
+
+## 2. Supported Model List
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing. The "Normal Mode" values correspond to the local <code>paddle_static</code> inference engine.
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>SLANet</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANet_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_pretrained.pdparams">Training Model</a></td>
+<td>59.52</td>
+<td>23.96 / 21.75</td>
+<td>- / 43.12</td>
+<td>6.9</td>
+<td rowspan="1">SLANet is a table structure recognition model independently developed by Baidu PaddlePaddle Vision Team. By adopting a CPU-friendly lightweight backbone network PP-LCNet, high-low level feature fusion module CSP-PAN, and SLA Head, a feature decoding module aligning structure and position information, this model greatly improves the accuracy and inference speed of table structure recognition.</td>
+</tr>
+<tr>
+<td>SLANet_plus</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANet_plus_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_plus_pretrained.pdparams">Training Model</a></td>
+<td>63.69</td>
+<td>23.43 / 22.16</td>
+<td>- / 41.80</td>
+<td>6.9</td>
+<td rowspan="1">SLANet_plus is an enhanced version of the table structure recognition model SLANet independently developed by the Baidu PaddlePaddle Vision Team. Compared to SLANet, SLANet_plus has greatly improved the recognition ability for wireless and complex tables, and reduced the model's sensitivity to table positioning accuracy. Even if the table positioning is offset, it can still be accurately recognized.
+</td>
+</tr>
+<tr>
+<td>SLANeXt_wired</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wired_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wired_pretrained.pdparams">Training Model</a></td>
+<td rowspan="2">69.65</td>
+<td rowspan="2">85.92 / 85.92</td>
+<td rowspan="2">- / 501.66</td>
+<td rowspan="2">351</td>
+<td rowspan="2">The SLANeXt series is a new generation of table structure recognition models independently developed by the Baidu PaddlePaddle Vision Team. Compared to SLANet and SLANet_plus, SLANeXt focuses on table structure recognition, and trains dedicated weights for wired and wireless tables separately. The recognition ability for all types of tables has been significantly improved, especially for wired tables.</td>
+</tr>
+<tr>
+<td>SLANeXt_wireless</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wireless_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wireless_pretrained.pdparams">Training Model</a></td>
+</tr>
+</table>
+
+<strong>Test Environment Description:</strong>
+
+  <ul>
+      <li><b>Performance Test Environment</b>
+          <ul>
+              <li><strong>Test Dataset:</strong> High-difficulty Chinese table recognition dataset.</li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>Inference Mode Description</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration</th>
+            <th>CPU Configuration</th>
+            <th>Acceleration Technology Combination</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Normal Mode</td>
+            <td>FP32 precision / No TRT acceleration</td>
+            <td>FP32 precision / 8 threads</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>High Performance Mode</td>
+            <td>Optimal combination of prior precision type and acceleration strategy</td>
+            <td>FP32 precision / 8 threads</td>
+            <td>Selects the prior optimal backend (Paddle/OpenVINO/TRT, etc.)</td>
+        </tr>
+    </tbody>
+</table>
+
+
+## 3. Quick Start
+
+> ❗ Before getting started, please install the PaddleOCR wheel package. For details, please refer to the [Installation Tutorial](../installation.en.md).
+
+Quickly experience with a single command:
+
+```bash
+paddleocr table_structure_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr table_structure_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg \
+    --engine transformers
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following command:
+
+```bash
+# Use the onnxruntime engine for inference
+paddleocr table_structure_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg \
+    --engine onnxruntime
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+<b>Note: </b>The official models would be download from HuggingFace by default. If can't access to HuggingFace, please set the environment variable `PADDLE_PDX_MODEL_SOURCE="BOS"` to change the model source to BOS. In the future, more model sources will be supported.
+
+You can also integrate the model inference of the table structure recognition module into your own project. Before running the code below, please download the [sample image](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg) to your local machine.
+
+```python
+from paddleocr import TableStructureRecognition
+model = TableStructureRecognition(model_name="SLANet")
+output = model.predict(input="table_recognition.jpg", batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_json("./output/res.json")
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following code:
+
+```python
+from paddleocr import TableStructureRecognition
+model = TableStructureRecognition(
+    model_name="SLANet",
+    engine="transformers",
+)
+output = model.predict(input="table_recognition.jpg", batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_json("./output/res.json")
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following code:
+
+```python
+from paddleocr import TableStructureRecognition
+model = TableStructureRecognition(
+    model_name="SLANet",
+    engine="onnxruntime",
+)
+output = model.predict(input="table_recognition.jpg", batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_json("./output/res.json")
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+If you want to use the trained model with the `paddle_dynamic` or `transformers` engine, refer to the [Weight Conversion](#52-weight-conversion) section in the [Inference Engine](#5-inference-engine) section below to convert the model from the `pdparams` format to the `safetensors` format using PaddleX.
+
+After running, the result is:
+
+```
+{'res': {'input_path': 'table_recognition.jpg', 'page_index': None, 'bbox': [[42, 2, 390, 2, 388, 27, 40, 26], [11, 35, 89, 35, 87, 63, 11, 63], [113, 34, 192, 34, 186, 64, 109, 64], [219, 33, 399, 33, 393, 62, 212, 62], [413, 33, 544, 33, 544, 64, 407, 64], [12, 67, 98, 68, 96, 93, 12, 93], [115, 66, 205, 66, 200, 91, 111, 91], [234, 65, 390, 65, 385, 92, 227, 92], [414, 66, 537, 67, 537, 95, 409, 95], [7, 97, 106, 97, 104, 128, 7, 128], [113, 96, 206, 95, 201, 127, 109, 127], [236, 96, 386, 96, 381, 128, 230, 128], [413, 96, 534, 95, 533, 127, 408, 127]], 'structure': ['<html>', '<body>', '<table>', '<tr>', '<td', ' colspan="4"', '>', '</td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '</table>', '</body>', '</html>'], 'structure_score': 0.99948007}}
+```
+
+Parameter meanings are as follows:
+<ul>
+<li><code>input_path</code>：The path of the input table image to be predicted</li>
+<li><code>page_index</code>：If the input is a PDF file, indicates the page number of the PDF; otherwise, it is  <code>None</code></li>
+<li><code>boxes</code>： Predicted table cell information, a list consisting of the coordinates of predicted table cells. Notably, table cell predictions for the SLANeXt series models are invalid</li>
+<li><code>structure</code>：Predicted table structure HTML expressions, a list consisting of predicted HTML keywords in order</li>
+<li><code>structure_score</code>：Confidence of the predicted table structure</li>
+</ul>
+
+Descriptions of related methods and parameters are as follows:
+
+* <code>TableStructureRecognition</code> instantiates a table structure recognition model (using <code>SLANet</code> as an example). Details are as follows:
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>Meaning:</b> Model name.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, <code>PP-LCNet_x1_0_table_cls</code> will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>Meaning:</b>Model storage path.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>Device for inference.<br/>
+<b>Description:</b>
+<b>For example:</b> <code>"cpu"</code>, <code>"gpu"</code>, <code>"npu"</code>, <code>"gpu:0"</code>, <code>"gpu:0,1"</code>.<br/>
+If multiple devices are specified, parallel inference will be performed.<br/>
+By default, GPU 0 is used if available; otherwise, CPU is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, <code>transformers</code>, and <code>onnxruntime</code>. When left as <code>None</code>, local inference uses the <code>paddle_static</code> engine by default. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b>Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b>Whether to use the Paddle Inference TensorRT subgraph engine.<br/> 
+<b>Description:</b>
+If the model does not support acceleration through TensorRT, setting this flag will not enable acceleration.<br/>
+For Paddle with CUDA version 11.8, the compatible TensorRT version is 8.x (x>=6), and it is recommended to install TensorRT 8.6.1.6.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b>Computation precision when using the Paddle Inference TensorRT subgraph engine.<br/>
+<b>Description:</b>
+<b>Options:</b> <code>"fp32"</code>, <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b>Whether to enable MKL-DNN acceleration for inference. <br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support it, acceleration will not be used even if this flag is set.</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td><b>Meaning:</b>MKL-DNN cache capacity.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b>Number of threads to use for inference on CPUs.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+</tbody>
+</table>
+
+* Call the <code>predict()</code> method of the table structure recognition model for inference prediction, which returns a result list. In addition, this module also provides the <code>predict_iter()</code> method. The two are completely consistent in parameter acceptance and result return. The difference is that <code>predict_iter()</code> returns a <code>generator</code>, which can process and obtain prediction results step by step, suitable for handling large datasets or scenarios where you want to save memory. You can choose to use either method according to your actual needs. The <code>predict()</code> method has parameters <code>input</code> and <code>batch_size</code>, described as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td>
+<b>Meaning:</b>Data to be predicted. Required. <br/>
+<b>Description:</b>
+Supports multiple input types:
+<ul>
+  <li><b>Python Var</b>: e.g., <code>numpy.ndarray</code> representing image data</li>
+  <li><b>str</b>: Local path to an image or PDF file, e.g., <code>/root/data/img.jpg</code>; <b>URL</b>: Network URL to an image or PDF file, e.g., <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">Example</a>; <b>Directory</b>: A local directory containing images for prediction, e.g., <code>/root/data/</code> (Note: PDF files in directories are not supported; to predict a PDF, specify its file path directly)</li>
+  <li><b>list</b>: A list of the above types, e.g., <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>Meaning:</b>Batch size. <br/>
+<b>Description:</b>
+Can be set to any positive integer.</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+* For processing prediction results, the prediction result of each sample is the corresponding Result object, and supports printing and saving as a <code>json</code> file:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Description</th>
+<th>Parameter</th>
+<th>Type</th>
+<th>Parameter Description</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">Print result to terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to use <code>JSON</code> indentation formatting for the output</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify indentation level to beautify the output <code>JSON</code> data, making it more readable, effective only when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Controls whether to escape non-<code>ASCII</code> characters as <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> keeps the original characters. Effective only when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">Save result as json format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Path to save the file. If it's a directory, the saved file will be named the same as the input file type</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify indentation level to beautify the output <code>JSON</code> data, making it more readable, effective only when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Controls whether to escape non-<code>ASCII</code> characters as <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> keeps the original characters. Effective only when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+</table>
+
+* In addition, it also supports obtaining results through attributes, as follows:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">Get the prediction result in <code>json</code> format</td>
+</tr>
+</table>
+
+## 4. Secondary Development
+
+If the above models are still not ideal for your scenario, you can try the following steps for secondary development. Here, training `SLANet_plus` is used as an example, and for other models, just replace the corresponding configuration file. First, you need to prepare a dataset for table structure recognition, which can be prepared with reference to the format of the [table structure recognition demo data](https://paddle-model-ecology.bj.bcebos.com/paddlex/data/table_rec_dataset_examples.tar). Once ready, you can train and export the model as follows. After exporting, you can quickly integrate the model into the above API. Here, the table structure recognition demo data is used as an example. Before training the model, please make sure you have installed the dependencies required by PaddleOCR according to the [installation documentation](../installation.en.md).
+
+
+## 4.1 Dataset and Pretrained Model Preparation
+
+### 4.1.1 Prepare Dataset
+
+```shell
+# Download sample dataset
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/data/table_rec_dataset_examples.tar
+tar -xf table_rec_dataset_examples.tar
+```
+
+### 4.1.2 Download Pretrained Model
+
+```shell
+# Download SLANet_plus pretrained model
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_plus_pretrained.pdparams
+```
+
+### 4.2 Model Training
+
+PaddleOCR is modularized. When training the `SLANet_plus` recognition model, you need to use the [configuration file](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/configs/table/SLANet_plus.yml) of `SLANet_plus`.
+
+
+The training commands are as follows:
+
+```bash
+# Single card training (default training method)
+python3 tools/train.py -c configs/table/SLANet_plus.yml \
+    -o Global.pretrained_model=./SLANet_plus_pretrained.pdparams
+    Train.dataset.data_dir=./table_rec_dataset_examples \
+    Train.dataset.label_file_list='[./table_rec_dataset_examples/train.txt]' \
+    Eval.dataset.data_dir=./table_rec_dataset_examples \
+    Eval.dataset.label_file_list='[./table_rec_dataset_examples/val.txt]'
+
+# Multi-card training, specify card numbers via --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py \
+    -c configs/table/SLANet_plus.yml \
+    -o Global.pretrained_model=./SLANet_plus_pretrained.pdparams
+    -o Global.pretrained_model=./PP-OCRv5_server_det_pretrained.pdparams \
+    Train.dataset.data_dir=./table_rec_dataset_examples \
+    Train.dataset.label_file_list='[./table_rec_dataset_examples/train.txt]' \
+    Eval.dataset.data_dir=./table_rec_dataset_examples \
+    Eval.dataset.label_file_list='[./table_rec_dataset_examples/val.txt]'
+```
+
+
+### 4.3 Model Evaluation
+
+You can evaluate the trained weights, such as `output/xxx/xxx.pdparams`, using the following command:
+
+```bash
+# Note to set the path of pretrained_model to the local path. If you use the model saved by your own training, please modify the path and file name to {path/to/weights}/{model_name}.
+ # Demo test set evaluation
+ python3 tools/eval.py -c configs/table/SLANet_plus.yml -o \
+    Global.pretrained_model=output/xxx/xxx.pdparams
+    Eval.dataset.data_dir=./table_rec_dataset_examples \
+    Eval.dataset.label_file_list='[./table_rec_dataset_examples/val.txt]'
+```
+
+### 4.4 Model Export
+
+```bash
+ python3 tools/export_model.py -c configs/table/SLANet_plus.yml -o \
+    Global.pretrained_model=output/xxx/xxx.pdparams \
+    Global.save_inference_dir="./SLANet_plus_infer/"
+```
+
+ After exporting the model, the static graph model will be stored in `./SLANet_plus_infer/` in the current directory. In this directory, you will see the following files:
+ ```
+ ./SLANet_plus_infer/
+ ├── inference.json
+ ├── inference.pdiparams
+ ├── inference.yml
+ ```
+At this point, secondary development is complete, and this static graph model can be directly integrated into the PaddleOCR API.
+
+If you want to use the `paddle_dynamic` or `transformers` engine with the trained model, please refer to the [Weight Conversion](#52-weight-conversion) section in [Inference Engine](#5-inference-engine) later in this document to convert the model from the `pdparams` format to the `safetensors` format using PaddleX.
+
+## 5. Inference Engine
+
+For detailed descriptions, values, compatibility rules, and examples of the inference engine, please refer to <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration Description</a>.
+
+### 5.1 Speed Data
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="1">SLANet</td>
+            <td>onnxruntime</td>
+            <td>1.42</td>
+            <td>22.02</td>
+            <td>0.22</td>
+            <td>23.81</td>
+        </tr>
+        <tr>
+            <td rowspan="1">SLANet_plus</td>
+            <td>onnxruntime</td>
+            <td>1.40</td>
+            <td>21.64</td>
+            <td>0.22</td>
+            <td>23.42</td>
+        </tr>
+        <tr>
+            <td rowspan="4">SLANeXt_wired</td>
+            <td>paddle_static</td>
+            <td>1.50</td>
+            <td>30.91</td>
+            <td>0.23</td>
+            <td>32.77</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>1.71</td>
+            <td>57.44</td>
+            <td>0.91</td>
+            <td>60.23</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>4.03</td>
+            <td>45.14</td>
+            <td>0.74</td>
+            <td>51.12</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>1.60</td>
+            <td>30.96</td>
+            <td>0.25</td>
+            <td>32.99</td>
+        </tr>
+        <tr>
+            <td rowspan="4">SLANeXt_wireless</td>
+            <td>paddle_static</td>
+            <td>1.67</td>
+            <td>30.49</td>
+            <td>0.22</td>
+            <td>32.51</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>1.68</td>
+            <td>57.24</td>
+            <td>0.96</td>
+            <td>60.05</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>4.30</td>
+            <td>45.51</td>
+            <td>0.75</td>
+            <td>51.76</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>1.62</td>
+            <td>31.01</td>
+            <td>0.26</td>
+            <td>33.06</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>Test Environment Description:</strong>
+<ul>
+    <li><strong>Test Data:</strong> <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg">Sample Image</a></li>
+    <li><strong>Hardware Configuration:</strong>
+        <ul>
+            <li>GPU: NVIDIA A100 40G</li>
+            <li>CPU: Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>Software Environment:</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 5.2 Weight Conversion
+
+When using the inference engine, the system will automatically download the official pre-trained model. If you need to use a self-trained model with the `paddle_dynamic` or `transformers` engine, please refer to the [PaddleX Table Structure Recognition Module Weight Conversion](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/table_structure_recognition.html#442) section to convert the model from the `pdparams` format to the `safetensors` format using PaddleX. This allows seamless integration into the PaddleOCR API for inference. If you need to use a self-trained model with the `onnxruntime` engine, refer to [PaddleX Obtain ONNX Models](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html) to obtain the ONNX model, so it can be seamlessly integrated into the PaddleOCR API for inference.
+
+## 6. FAQ
diff --git a/docs/version3.x/module_usage/table_structure_recognition.md b/docs/version3.x/module_usage/table_structure_recognition.md
new file mode 100644
index 0000000..12c6845
--- /dev/null
+++ b/docs/version3.x/module_usage/table_structure_recognition.md
@@ -0,0 +1,602 @@
+---
+comments: true
+---
+
+# 表格结构识别模块使用教程
+
+## 一、概述
+
+表格结构识别是表格识别系统中的重要组成部分，能够将不可编辑表格图片转换为可编辑的表格形式（例如html）。表格结构识别的目标是对表格的行、列和单元格位置进行识别，该模块的性能直接影响到整个表格识别系统的准确性和效率。表格结构识别模块会输出表格区域的html代码，这些代码将作为输入传递给表格识别产线进行后续处理。
+
+## 二、支持模型列表
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。表格中的“常规模式”耗时对应本地 <code>paddle_static</code> 推理引擎。
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>精度（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>SLANet</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANet_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_pretrained.pdparams">训练模型</a></td>
+<td>59.52</td>
+<td>23.96 / 21.75</td>
+<td>- / 43.12</td>
+<td>6.9</td>
+<td rowspan="1">SLANet 是百度飞桨视觉团队自研的表格结构识别模型。该模型通过采用 CPU 友好型轻量级骨干网络 PP-LCNet、高低层特征融合模块 CSP-PAN、结构与位置信息对齐的特征解码模块 SLA Head，大幅提升了表格结构识别的精度和推理速度。</td>
+</tr>
+<tr>
+<td>SLANet_plus</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANet_plus_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_plus_pretrained.pdparams">训练模型</a></td>
+<td>63.69</td>
+<td>23.43 / 22.16</td>
+<td>- / 41.80</td>
+<td>6.9</td>
+<td rowspan="1">SLANet_plus 是百度飞桨视觉团队自研的表格结构识别模型 SLANet 的增强版。相较于 SLANet，SLANet_plus 对无线表、复杂表格的识别能力得到了大幅提升，并降低了模型对表格定位准确性的敏感度，即使表格定位出现偏移，也能够较准确地进行识别。
+</td>
+</tr>
+<tr>
+<td>SLANeXt_wired</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wired_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wired_pretrained.pdparams">训练模型</a></td>
+<td rowspan="2">69.65</td>
+<td rowspan="2">85.92 / 85.92</td>
+<td rowspan="2">- / 501.66</td>
+<td rowspan="2">351</td>
+<td rowspan="2">SLANeXt 系列是百度飞桨视觉团队自研的新一代表格结构识别模型。相较于 SLANet 和 SLANet_plus，SLANeXt 专注于对表格结构进行识别，并且对有线表格(wired)和无线表格(wireless)的识别分别训练了专用的权重，对各类型表格的识别能力都得到了明显提高，特别是对有线表格的识别能力得到了大幅提升。</td>
+</tr>
+<tr>
+<td>SLANeXt_wireless</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wireless_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wireless_pretrained.pdparams">训练模型</a></td>
+</tr>
+</table>
+
+<strong>测试环境说明:</strong>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+              <li><strong>测试数据集：</strong>内部自建的高难度中文表格识别数据集。</li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+
+
+## 三、快速开始
+
+> ❗ 在快速开始前，请先安装 PaddleOCR 的 wheel 包，详细请参考 [安装教程](../installation.md)。
+
+使用一行命令即可快速体验：
+
+```bash
+paddleocr table_structure_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下命令：
+
+```bash
+# 使用 transformers 引擎进行推理
+paddleocr table_structure_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg \
+    --engine transformers
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下命令：
+
+```bash
+# 使用 onnxruntime 引擎进行推理
+paddleocr table_structure_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg \
+    --engine onnxruntime
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+<b>注：</b>PaddleOCR 官方模型默认从 HuggingFace 获取，如运行环境访问 HuggingFace 不便，可通过环境变量修改模型源为 BOS：`PADDLE_PDX_MODEL_SOURCE="BOS"`，未来将支持更多主流模型源；
+
+您也可以将表格结构识别的模块中的模型推理集成到您的项目中。运行以下代码前，请您下载[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg)到本地。
+
+```python
+from paddleocr import TableStructureRecognition
+model = TableStructureRecognition(model_name="SLANet")
+output = model.predict(input="table_recognition.jpg", batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_json("./output/res.json")
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下代码：
+
+```python
+from paddleocr import TableStructureRecognition
+model = TableStructureRecognition(
+    model_name="SLANet",
+    engine="transformers",
+)
+output = model.predict(input="table_recognition.jpg", batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_json("./output/res.json")
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下代码：
+
+```python
+from paddleocr import TableStructureRecognition
+model = TableStructureRecognition(
+    model_name="SLANet",
+    engine="onnxruntime",
+)
+output = model.predict(input="table_recognition.jpg", batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_json("./output/res.json")
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+训练后的模型如果想使用 `paddle_dynamic` 或 `transformers` 引擎，请参考后文 [推理引擎](#五推理引擎) 中的 [权重转换](#52-权重转换) 部分将模型由 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式。
+
+运行后，得到的结果为：
+
+```
+{'res': {'input_path': 'table_recognition.jpg', 'page_index': None, 'bbox': [[42, 2, 390, 2, 388, 27, 40, 26], [11, 35, 89, 35, 87, 63, 11, 63], [113, 34, 192, 34, 186, 64, 109, 64], [219, 33, 399, 33, 393, 62, 212, 62], [413, 33, 544, 33, 544, 64, 407, 64], [12, 67, 98, 68, 96, 93, 12, 93], [115, 66, 205, 66, 200, 91, 111, 91], [234, 65, 390, 65, 385, 92, 227, 92], [414, 66, 537, 67, 537, 95, 409, 95], [7, 97, 106, 97, 104, 128, 7, 128], [113, 96, 206, 95, 201, 127, 109, 127], [236, 96, 386, 96, 381, 128, 230, 128], [413, 96, 534, 95, 533, 127, 408, 127]], 'structure': ['<html>', '<body>', '<table>', '<tr>', '<td', ' colspan="4"', '>', '</td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '<tr>', '<td></td>', '<td></td>', '<td></td>', '<td></td>', '</tr>', '</table>', '</body>', '</html>'], 'structure_score': 0.99948007}}
+```
+
+参数含义如下：
+<ul>
+<li><code>input_path</code>：输入的待预测表格图像的路径</li>
+<li><code>page_index</code>：如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+<li><code>boxes</code>：预测的表格单元格信息，一个列表，由预测的若干表格单元格坐标组成。特别地， SLANeXt 系列模型预测的表格单元格无效</li>
+<li><code>structure</code>：预测的表格结构Html表达式，一个列表，由预测的若干Html关键字按顺序组成</li>
+<li><code>structure_score</code>：预测表格结构的置信度</li>
+</ul>
+相关方法、参数等说明如下：
+
+* <code>TableStructureRecognition</code>实例化表格结构识别模型（此处以<code>SLANet</code>为例），具体说明如下：
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>含义：</b>模型名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，则使用<code>PP-LCNet_x1_0_table_cls</code>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>含义：</b>模型存储路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+<b>例如：</b><code>"cpu"</code>、<code>"gpu"</code>、<code>"npu"</code>、<code>"gpu:0"</code>、<code>"gpu:0,1"</code>。<br/>
+如指定多个设备，将进行并行推理。<br/>
+默认情况下，优先使用 GPU 0；若不可用则使用 CPU。
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>、<code>onnxruntime</code>。保持为默认值 <code>None</code> 时，本地推理默认使用 <code>paddle_static</code> 引擎。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>当使用 Paddle Inference 的 TensorRT 子图引擎时设置的计算精度。<br/>
+<b>说明：</b>
+<b>可选项：</b><code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+<b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上推理时使用的线程数量。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+</tbody>
+</table>
+
+* 调用表格结构识别模型的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。另外，本模块还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。<code>predict()</code> 方法参数有 <code>input</code> 和 <code>batch_size</code>，具体说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>含义：</b>批大小。<br/>
+<b>说明：</b>
+可设置为任意正整数。</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+
+* 对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为<code>json</code>文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+</table>
+
+* 此外，也支持通过属性获取结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">获取预测的<code>json</code>格式的结果</td>
+</tr>
+</table>
+
+## 四、二次开发
+
+如果以上模型在您的场景上效果仍然不理想，您可以尝试以下步骤进行二次开发，此处以训练 `SLANet_plus` 举例，其他模型替换对应配置文件即可。首先，您需要准备表格结构识别的数据集，可以参考[表格结构识别 Demo 数据](https://paddle-model-ecology.bj.bcebos.com/paddlex/data/table_rec_dataset_examples.tar)的格式准备，准备好后，即可按照以下步骤进行模型训练和导出，导出后，可以将模型快速集成到上述 API 中。此处以表格结构识别 Demo 数据示例。在训练模型之前，请确保已经按照[安装文档](../installation.md)安装了 PaddleOCR 所需要的依赖。
+
+
+### 4.1 数据集、预训练模型准备
+
+#### 4.1.1 准备数据集
+
+```shell
+# 下载示例数据集
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/data/table_rec_dataset_examples.tar
+tar -xf table_rec_dataset_examples.tar
+```
+
+#### 4.1.2 下载预训练模型
+
+```shell
+# 下载 SLANet_plus 预训练模型
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_plus_pretrained.pdparams
+```
+
+### 4.2 模型训练
+
+PaddleOCR 对代码进行了模块化，训练 `SLANet_plus` 识别模型时需要使用 `SLANet_plus` 的[配置文件](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/configs/table/SLANet_plus.yml)。
+
+
+训练命令如下：
+
+```bash
+#单卡训练 (默认训练方式)
+python3 tools/train.py -c configs/table/SLANet_plus.yml \
+    -o Global.pretrained_model=./SLANet_plus_pretrained.pdparams
+    Train.dataset.data_dir=./table_rec_dataset_examples \
+    Train.dataset.label_file_list='[./table_rec_dataset_examples/train.txt]' \
+    Eval.dataset.data_dir=./table_rec_dataset_examples \
+    Eval.dataset.label_file_list='[./table_rec_dataset_examples/val.txt]'
+
+#多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py \
+    -c configs/table/SLANet_plus.yml \
+    -o Global.pretrained_model=./SLANet_plus_pretrained.pdparams
+    -o Global.pretrained_model=./PP-OCRv5_server_det_pretrained.pdparams \
+    Train.dataset.data_dir=./table_rec_dataset_examples \
+    Train.dataset.label_file_list='[./table_rec_dataset_examples/train.txt]' \
+    Eval.dataset.data_dir=./table_rec_dataset_examples \
+    Eval.dataset.label_file_list='[./table_rec_dataset_examples/val.txt]'
+```
+
+
+### 4.3 模型评估
+
+您可以评估已经训练好的权重，如，`output/xxx/xxx.pdparams`，使用如下命令进行评估：
+
+```bash
+# 注意将pretrained_model的路径设置为本地路径。若使用自行训练保存的模型，请注意修改路径和文件名为{path/to/weights}/{model_name}。
+ # demo 测试集评估
+ python3 tools/eval.py -c configs/table/SLANet_plus.yml -o \
+    Global.pretrained_model=output/xxx/xxx.pdparams
+    Eval.dataset.data_dir=./table_rec_dataset_examples \
+    Eval.dataset.label_file_list='[./table_rec_dataset_examples/val.txt]'
+```
+
+### 4.4 模型导出
+
+```bash
+ python3 tools/export_model.py -c configs/table/SLANet_plus.yml -o \
+    Global.pretrained_model=output/xxx/xxx.pdparams \
+    Global.save_inference_dir="./SLANet_plus_infer/"
+```
+
+ 导出模型后，静态图模型会存放于当前目录的`./SLANet_plus_infer/`中，在该目录下，您将看到如下文件：
+ ```
+ ./SLANet_plus_infer/
+ ├── inference.json
+ ├── inference.pdiparams
+ ├── inference.yml
+ ```
+至此，二次开发完成，该静态图模型可以直接集成到 PaddleOCR 的 API 中。
+
+训练后的模型如果想使用 `paddle_dynamic` 或 `transformers` 引擎，请参考后文 [推理引擎](#五推理引擎) 中的 [权重转换](#52-权重转换) 部分将模型由 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式。
+
+## 五、推理引擎 {#五推理引擎}
+
+关于推理引擎的详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。
+
+### 5.1 速度数据
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="1">SLANet</td>
+            <td>onnxruntime</td>
+            <td>1.42</td>
+            <td>22.02</td>
+            <td>0.22</td>
+            <td>23.81</td>
+        </tr>
+        <tr>
+            <td rowspan="1">SLANet_plus</td>
+            <td>onnxruntime</td>
+            <td>1.40</td>
+            <td>21.64</td>
+            <td>0.22</td>
+            <td>23.42</td>
+        </tr>
+        <tr>
+            <td rowspan="4">SLANeXt_wired</td>
+            <td>paddle_static</td>
+            <td>1.50</td>
+            <td>30.91</td>
+            <td>0.23</td>
+            <td>32.77</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>1.71</td>
+            <td>57.44</td>
+            <td>0.91</td>
+            <td>60.23</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>4.03</td>
+            <td>45.14</td>
+            <td>0.74</td>
+            <td>51.12</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>1.60</td>
+            <td>30.96</td>
+            <td>0.25</td>
+            <td>32.99</td>
+        </tr>
+        <tr>
+            <td rowspan="4">SLANeXt_wireless</td>
+            <td>paddle_static</td>
+            <td>1.67</td>
+            <td>30.49</td>
+            <td>0.22</td>
+            <td>32.51</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>1.68</td>
+            <td>57.24</td>
+            <td>0.96</td>
+            <td>60.05</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>4.30</td>
+            <td>45.51</td>
+            <td>0.75</td>
+            <td>51.76</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>1.62</td>
+            <td>31.01</td>
+            <td>0.26</td>
+            <td>33.06</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>测试环境说明:</strong>
+<ul>
+    <li><strong>测试数据：</strong><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg">示例图片</a></li>
+    <li><strong>硬件配置：</strong>
+        <ul>
+            <li>GPU：NVIDIA A100 40G</li>
+            <li>CPU：Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>软件环境：</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 5.2 权重转换 {#52-权重转换}
+
+使用推理引擎时，系统会自动下载官方预训练模型。若需使用自训练模型配合 `paddle_dynamic` 或 `transformers` 引擎，请参考 [PaddleX 表格结构识别模块权重转换](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/table_structure_recognition.html#442) 部分，将 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式，即可无缝集成到 PaddleOCR 的 API 中进行推理。若需使用自训练模型配合`onnxruntime`引擎，请参考[PaddleX 获取 ONNX 模型](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html)获取onnx模型，即可无缝集成到 PaddleOCR 的 API 中进行推理。
+
+## 六、FAQ
diff --git a/docs/version3.x/module_usage/text_detection.en.md b/docs/version3.x/module_usage/text_detection.en.md
new file mode 100644
index 0000000..3cd8ac9
--- /dev/null
+++ b/docs/version3.x/module_usage/text_detection.en.md
@@ -0,0 +1,822 @@
+---
+comments: true
+---
+
+# Text Detection Module Usage Guide
+
+## 1. Overview
+The text detection module is a critical component of OCR (Optical Character Recognition) systems, responsible for locating and marking text-containing regions in images. The performance of this module directly impacts the accuracy and efficiency of the entire OCR system. The text detection module typically outputs bounding boxes for text regions, which are then passed to the text recognition module for further processing.
+
+## 2. Supported Models List
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing. The "Standard Mode" values correspond to the local <code>paddle_static</code> inference engine.
+
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Detection Hmean (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High-Performance Mode]</th>
+<th>Model Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv6_medium_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_medium_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_medium_det_pretrained.pdparams">Training Model</a></td>
+<td>86.2*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>59.4</td>
+<td>PP-OCRv6 medium-scale text detection model based on PPLCNetV4 + RepLKFPN, highest accuracy, suitable for server deployment</td>
+</tr>
+<tr>
+<td>PP-OCRv6_small_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_small_det_pretrained.pdparams">Training Model</a></td>
+<td>84.1*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>9.6</td>
+<td>PP-OCRv6 small text detection model, balancing accuracy and efficiency, suitable for mobile deployment</td>
+</tr>
+<tr>
+<td>PP-OCRv6_tiny_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_tiny_det_pretrained.pdparams">Training Model</a></td>
+<td>80.6*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>1.9</td>
+<td>PP-OCRv6 ultra-lightweight text detection model (0.43M params), suitable for edge/IoT scenarios</td>
+</tr>
+<tr>
+<td>PP-OCRv5_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_det_pretrained.pdparams">Training Model</a></td>
+<td>83.8</td>
+<td>89.55 / 70.19</td>
+<td>383.15 / 383.15</td>
+<td>84.3</td>
+<td>PP-OCRv5 server-side text detection model with higher accuracy, suitable for deployment on high-performance servers</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_det_pretrained.pdparams">Training Model</a></td>
+<td>79.0</td>
+<td>10.67 / 6.36</td>
+<td>57.77 / 28.15</td>
+<td>4.7</td>
+<td>PP-OCRv5 mobile-side text detection model with higher efficiency, suitable for deployment on edge devices</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_det_pretrained.pdparams">Training Model</a></td>
+<td>69.2</td>
+<td>127.82 / 98.87</td>
+<td>585.95 / 489.77</td>
+<td>109</td>
+<td>PP-OCRv4 server-side text detection model with higher accuracy, suitable for deployment on high-performance servers</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams">Training Model</a></td>
+<td>63.8</td>
+<td>9.87 / 4.17</td>
+<td>56.60 / 20.79</td>
+<td>4.7</td>
+<td>PP-OCRv4 mobile-side text detection model with higher efficiency, suitable for deployment on edge devices</td>
+</tr>
+</tbody>
+</table>
+
+> *Note: PP-OCRv6 metrics are evaluated on an internal multi-scenario evaluation set, while PP-OCRv5/v4 metrics are based on a general evaluation set. As the evaluation sets differ, the metrics are not directly comparable.
+
+<strong>Testing Environment:</strong>
+
+  <ul>
+      <li><b>Performance Testing Environment</b>
+          <ul>
+              <li><strong>Test Dataset:</strong> PaddleOCR3.0 newly constructed multilingual dataset (including Chinese, Traditional Chinese, English, Japanese), covering street scenes, web images, documents, handwriting, blur, rotation, distortion, etc., totaling 2677 images.</li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>Inference Mode Description</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration</th>
+            <th>CPU Configuration</th>
+            <th>Acceleration Techniques</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Standard Mode</td>
+            <td>FP32 precision / No TRT acceleration</td>
+            <td>FP32 precision / 8 threads</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>High-Performance Mode</td>
+            <td>Optimal combination of precision types and acceleration strategies</td>
+            <td>FP32 precision / 8 threads</td>
+            <td>Optimal backend selection (Paddle/OpenVINO/TRT, etc.)</td>
+        </tr>
+    </tbody>
+</table>
+
+## 3. Quick Start
+
+> ❗ Before starting, please install the PaddleOCR wheel package. Refer to the [Installation Guide](../installation.en.md) for details.
+
+Use the following command for a quick experience:
+
+```bash
+paddleocr text_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_001.png
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr text_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_001.png \
+    --engine transformers
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following command:
+
+```bash
+# Use the onnxruntime engine for inference
+paddleocr text_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_001.png \
+    --engine onnxruntime
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+<b>Note: </b>The official models would be download from HuggingFace by default. If can't access to HuggingFace, please set the environment variable `PADDLE_PDX_MODEL_SOURCE="BOS"` to change the model source to BOS. In the future, more model sources will be supported.
+
+You can also integrate the model inference into your project. Before running the following code, download the [example image](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_001.png) locally.
+
+```python
+from paddleocr import TextDetection
+model = TextDetection()
+output = model.predict("general_ocr_001.png", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following code:
+
+```python
+from paddleocr import TextDetection
+model = TextDetection(engine="transformers")
+output = model.predict("general_ocr_001.png", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following code:
+
+```python
+from paddleocr import TextDetection
+model = TextDetection(engine="onnxruntime")
+output = model.predict("general_ocr_001.png", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+If you want to use the trained model with the `paddle_dynamic` or `transformers` engine, refer to the [Weight Conversion](#52-weight-conversion) section in the [Inference Engine](#5-inference-engine) section below to convert the model from the `pdparams` format to the `safetensors` format using PaddleX.
+
+The output will be:
+
+```bash
+{'res': {'input_path': 'general_ocr_001.png', 'page_index': None, 'dt_polys': array([[[ 75, 549],
+        ...,
+        [ 77, 586]],
+
+       ...,
+
+       [[ 31, 406],
+        ...,
+        [ 34, 455]]], dtype=int16), 'dt_scores': [0.873949039891189, 0.8948166013613552, 0.8842595305917041, 0.876953790920377]}}
+```
+
+Output parameter meanings:
+<ul>
+<li><code>input_path</code>：Path of the input image.</li>
+<li><code>page_index</code>：If the input is a PDF, this indicates the current page number; otherwise, it is  <code>None</code></li>
+<li><code>dt_polys</code>：Predicted text detection boxes, where each box contains four vertices (x, y coordinates).</li>
+<li><code>dt_scores</code>：Confidence scores of the predicted text detection boxes.</li>
+</ul>
+
+Visualization example:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/text_det/general_ocr_001_res.png"/>
+
+Method and parameter descriptions:
+
+* Instantiate the text detection model with <code>TextDetection</code>:
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>Meaning:</b>Model name.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, <code>PP-OCRv6_medium_det</code> will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>Meaning:</b>Model storage path.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>Device for inference.<br/>
+<b>Description:</b>
+<b>For example:</b><code>"cpu"</code>, <code>"gpu"</code>, <code>"npu"</code>, <code>"gpu:0"</code>, <code>"gpu:0,1"</code>.<br/>
+If multiple devices are specified, parallel inference will be performed.<br/>
+By default, GPU 0 is used if available; otherwise, CPU is used.
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, <code>transformers</code>, and <code>onnxruntime</code>. When left as <code>None</code>, local inference uses the <code>paddle_static</code> engine by default. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b>Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b>Whether to use the Paddle Inference TensorRT subgraph engine.<br/> 
+<b>Description:</b>
+If the model does not support acceleration through TensorRT, setting this flag will not enable acceleration.<br/>
+For Paddle with CUDA version 11.8, the compatible TensorRT version is 8.x (x>=6), and it is recommended to install TensorRT 8.6.1.6.<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b>Computation precision when using the Paddle Inference TensorRT subgraph engine.<br/>
+<b>Description:</b>
+<b>Options:</b> <code>"fp32"</code>, <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b>Whether to enable MKL-DNN acceleration for inference. <br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support it, acceleration will not be used even if this flag is set.</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td><b>Meaning:</b>MKL-DNN cache capacity.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b>Number of threads to use for inference on CPUs.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>limit_side_len</code></td>
+<td><b>Meaning:</b>Limit on the side length of the input image for detection. 
+<b>Description:</b>
+<code>int</code> specifies the value. If set to <code>None</code>, the model's default configuration will be used.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>limit_type</code></td>
+<td><b>Meaning:</b>Type of image side length limitation. <br/>
+<b>Description:</b>
+<code>"min"</code> ensures the shortest side of the image is no less than <code>det_limit_side_len</code>; <code>"max"</code> ensures the longest side is no greater than <code>limit_side_len</code>. If set to <code>None</code>, the model's default configuration will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>max_side_limit</code></td>
+<td><b>Meaning:</b>Limit on the max length of the input image for detection. <br/>
+<b>Description:</b>
+<code>int</code> limits the longest side of the image for input detection model. If set to <code>None</code>, the model's default configuration will be used.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>thresh</code></td>
+<td><b>Meaning:</b>Pixel score threshold. Pixels in the output probability map with scores greater than this threshold are considered text pixels. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the model's default configuration will be used.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>box_thresh</code></td>
+<td><b>Meaning:</b>If the average score of all pixels inside the bounding box is greater than this threshold, the result is considered a text region. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the model's default configuration will be used.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion ratio for the Vatti clipping algorithm, used to expand the text region. 
+<b>Description:</b>
+If set to <code>None</code>, the model's default configuration will be used.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>input_shape</code></td>
+<td><b>Meaning:</b>Input image size for the model in the format <code>(C, H, W)</code>.</td>
+<td><code>tuple|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+* The <code>predict()</code> method parameters:
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td>
+<b>Meaning:</b>Input data to be predicted. Required.<br/>
+<b>Description:</b> 
+Supports multiple input types:
+<ul>
+  <li><b>Python variable</b>: e.g., <code>numpy.ndarray</code> representing image data</li>
+  <li><b>str</b>: Local image file or PDF file path: <code>/root/data/img.jpg</code>; <b>URL</b>: Image or PDF file network URL: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_rec_001.png">Example</a>; <b>Directory</b>: Should contain images for prediction, e.g., <code>/root/data/</code> (currently, PDF files in directories are not supported, PDF files need to be specified by file path)</li>
+  <li><b>list</b>: List elements should be of the above types, e.g., <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>Meaning:</b>Batch size<br/>
+<b>Description:</b>Positive integer.</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+<tr>
+<td><code>limit_side_len</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>limit_type</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>box_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>unclip_ratio</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+
+* Result processing methods:
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Description</th>
+<th>Parameters</th>
+<th>Type</th>
+<th>Description</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">Print results to terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Format output as JSON</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>JSON indentation level</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Escape non-ASCII characters</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">Save results as JSON file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Output file path</td>
+<td>Required</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>JSON indentation level</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Escape non-ASCII characters</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save results as image</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Output file path</td>
+<td>Required</td>
+</tr>
+</table>
+
+* Additional attributes:
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Description</th>
+</tr>
+</thead>
+<tr>
+<td><code>json</code></td>
+<td>Get prediction results in JSON format</td>
+</tr>
+<tr>
+<td><code>img</code></td>
+<td>Get visualization image as a dictionary</td>
+</tr>
+</table>
+
+## 4. Custom Development
+
+If the above models do not meet your requirements, follow these steps for custom development (using `PP-OCRv5_server_det` as an example). First, prepare a text detection dataset (refer to the [Demo Dataset](https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_det_dataset_examples.tar) format). After preparation, proceed with model training and export. The exported model can be integrated into the API. Ensure PaddleOCR dependencies are installed as per the [Installation Guide](../installation.en.md).
+
+### 4.1 Dataset and Pretrained Model Preparation
+
+#### 4.1.1 Prepare Dataset
+
+```shell
+# Download example dataset
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_det_dataset_examples.tar
+tar -xf ocr_det_dataset_examples.tar
+```
+
+#### 4.1.2 Download Pretrained Model
+
+```shell
+# Download PP-OCRv5_server_det pretrained model
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_det_pretrained.pdparams 
+```
+
+### 4.2 Model Training
+
+PaddleOCR modularizes the code. To train the `PP-OCRv5_server_det` model, use its [configuration file](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/configs/det/PP-OCRv5/PP-OCRv5_server_det.yml).
+
+Training command:
+
+```bash
+# Single-GPU training (default)
+python3 tools/train.py -c configs/det/PP-OCRv5/PP-OCRv5_server_det.yml \
+    -o Global.pretrained_model=./PP-OCRv5_server_det_pretrained.pdparams \
+    Train.dataset.data_dir=./ocr_det_dataset_examples \
+    Train.dataset.label_file_list='[./ocr_det_dataset_examples/train.txt]' \
+    Eval.dataset.data_dir=./ocr_det_dataset_examples \
+    Eval.dataset.label_file_list='[./ocr_det_dataset_examples/val.txt]'
+
+# Multi-GPU training (specify GPUs with --gpus)
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py \
+    -c configs/det/PP-OCRv5/PP-OCRv5_server_det.yml \
+    -o Global.pretrained_model=./PP-OCRv5_server_det_pretrained.pdparams \
+    Train.dataset.data_dir=./ocr_det_dataset_examples \
+    Train.dataset.label_file_list='[./ocr_det_dataset_examples/train.txt]' \
+    Eval.dataset.data_dir=./ocr_det_dataset_examples \
+    Eval.dataset.label_file_list='[./ocr_det_dataset_examples/val.txt]'
+```
+
+### 4.3 Model Evaluation
+
+You can evaluate trained weights (e.g., `output/PP-OCRv5_server_det/best_accuracy.pdparams`) using the following command:
+
+```bash
+# Note: Set pretrained_model to local path. For custom-trained models, modify the path and filename as {path/to/weights}/{model_name}.
+# Demo dataset evaluation
+python3 tools/eval.py -c configs/det/PP-OCRv5/PP-OCRv5_server_det.yml \
+    -o Global.pretrained_model=output/PP-OCRv5_server_det/best_accuracy.pdparams \
+    Eval.dataset.data_dir=./ocr_det_dataset_examples \
+    Eval.dataset.label_file_list='[./ocr_det_dataset_examples/val.txt]' 
+```
+
+### 4.4 Model Export
+
+```bash
+python3 tools/export_model.py -c configs/det/PP-OCRv5/PP-OCRv5_server_det.yml -o \
+    Global.pretrained_model=output/PP-OCRv5_server_det/best_accuracy.pdparams \
+    Global.save_inference_dir="./PP-OCRv5_server_det_infer/"
+```
+
+After export, the static graph model will be saved in `./PP-OCRv5_server_det_infer/` with the following files:
+```
+./PP-OCRv5_server_det_infer/
+├── inference.json
+├── inference.pdiparams
+├── inference.yml
+```
+The custom development is now complete. This static graph model can be directly integrated into PaddleOCR's API.
+
+If you want to use the `paddle_dynamic` or `transformers` engine with the trained model, please refer to the [Weight Conversion](#52-weight-conversion) section in [Inference Engine](#5-inference-engine) later in this document to convert the model from the `pdparams` format to the `safetensors` format using PaddleX.
+
+## 5. Inference Engine
+
+For detailed descriptions, values, compatibility rules, and examples of the inference engine, please refer to <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration Description</a>.
+
+### 5.1 Speed Data
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">PP-OCRv5_mobile_det</td>
+            <td>paddle_static</td>
+            <td>11.43</td>
+            <td>13.80</td>
+            <td>2.15</td>
+            <td>27.58</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>11.70</td>
+            <td>48.36</td>
+            <td>2.47</td>
+            <td>62.71</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>14.05</td>
+            <td>18.45</td>
+            <td>3.98</td>
+            <td>37.54</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>9.98</td>
+            <td>5.70</td>
+            <td>2.04</td>
+            <td>17.90</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-OCRv5_server_det</td>
+            <td>paddle_static</td>
+            <td>13.24</td>
+            <td>26.91</td>
+            <td>2.63</td>
+            <td>43.05</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>11.82</td>
+            <td>45.56</td>
+            <td>2.52</td>
+            <td>60.10</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>14.56</td>
+            <td>13.76</td>
+            <td>7.44</td>
+            <td>36.76</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>10.01</td>
+            <td>13.76</td>
+            <td>1.92</td>
+            <td>25.86</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-OCRv6_medium_det</td>
+            <td>paddle_static</td>
+            <td>13.89</td>
+            <td>16.02</td>
+            <td>2.49</td>
+            <td>33.14</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>11.42</td>
+            <td>26.23</td>
+            <td>2.30</td>
+            <td>40.10</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>11.40</td>
+            <td>8.57</td>
+            <td>8.35</td>
+            <td>29.57</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>10.80</td>
+            <td>13.06</td>
+            <td>2.19</td>
+            <td>26.18</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-OCRv6_small_det</td>
+            <td>paddle_static</td>
+            <td>10.91</td>
+            <td>10.97</td>
+            <td>2.41</td>
+            <td>24.45</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>11.56</td>
+            <td>22.17</td>
+            <td>2.66</td>
+            <td>36.55</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>11.70</td>
+            <td>7.34</td>
+            <td>3.87</td>
+            <td>23.89</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>11.32</td>
+            <td>7.46</td>
+            <td>2.54</td>
+            <td>21.49</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-OCRv6_tiny_det</td>
+            <td>paddle_static</td>
+            <td>11.14</td>
+            <td>10.71</td>
+            <td>2.84</td>
+            <td>24.85</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>11.52</td>
+            <td>21.70</td>
+            <td>2.94</td>
+            <td>36.31</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>10.90</td>
+            <td>6.99</td>
+            <td>4.13</td>
+            <td>23.00</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>11.19</td>
+            <td>6.35</td>
+            <td>2.79</td>
+            <td>20.49</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>Test Environment Description:</strong>
+<ul>
+    <li><strong>Test Data:</strong> <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_001.jpg">Sample Image</a></li>
+    <li><strong>Hardware Configuration:</strong>
+        <ul>
+            <li>GPU: NVIDIA A100 40G</li>
+            <li>CPU: Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>Software Environment:</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 5.2 Weight Conversion
+
+When using the inference engine, the system will automatically download the official pre-trained model. If you need to use a self-trained model with the `paddle_dynamic` or `transformers` engine, please refer to the [PaddleX Text Detection Module Weight Conversion](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/text_detection.html#442) section to convert the model from the `pdparams` format to the `safetensors` format using PaddleX. This allows seamless integration into the PaddleOCR API for inference. If you need to use a self-trained model with the `onnxruntime` engine, refer to [PaddleX Obtain ONNX Models](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html) to obtain the ONNX model, so it can be seamlessly integrated into the PaddleOCR API for inference.
+
+## 6. FAQ
+
+- Use parameters `limit_type` and `limit_side_len` to constrain image dimensions.  
+  - `limit_type` options: [`max`, `min`]  
+  - `limit_side_len`: Positive integer (typically multiples of 32, e.g., 960).  
+  - For lower-resolution images, use `limit_type=min` and `limit_side_len=960` to balance computational efficiency and detection quality.  
+  - For higher-resolution images requiring larger detection scales, set `limit_side_len` to desired values (e.g., 1216).
diff --git a/docs/version3.x/module_usage/text_detection.md b/docs/version3.x/module_usage/text_detection.md
new file mode 100644
index 0000000..981686c
--- /dev/null
+++ b/docs/version3.x/module_usage/text_detection.md
@@ -0,0 +1,936 @@
+---
+comments: true
+---
+
+# 文本检测模块使用教程
+
+## 一、概述
+文本检测模块是OCR（光学字符识别）系统中的关键组成部分，负责在图像中定位和标记出包含文本的区域。该模块的性能直接影响到整个OCR系统的准确性和效率。文本检测模块通常会输出文本区域的边界框（Bounding Boxes），这些边界框将作为输入传递给文本识别模块进行后续处理。
+
+## 二、支持模型列表
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。表格中的“常规模式”耗时对应本地 <code>paddle_static</code> 推理引擎。
+
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>检测Hmean（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv6_medium_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_medium_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_medium_det_pretrained.pdparams">训练模型</a></td>
+<td>86.2*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>59.4</td>
+<td>PP-OCRv6 的中等规模文本检测模型，基于 PPLCNetV4 + RepLKFPN，精度最高，适合服务端部署</td>
+</tr>
+<tr>
+<td>PP-OCRv6_small_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_small_det_pretrained.pdparams">训练模型</a></td>
+<td>84.1*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>9.6</td>
+<td>PP-OCRv6 的小型文本检测模型，兼顾精度与效率，适合移动端部署</td>
+</tr>
+<tr>
+<td>PP-OCRv6_tiny_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_tiny_det_pretrained.pdparams">训练模型</a></td>
+<td>80.6*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>1.9</td>
+<td>PP-OCRv6 的超轻量文本检测模型（0.43M 参数），适合端侧/IoT 场景</td>
+</tr>
+<tr>
+<td>PP-OCRv5_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_det_pretrained.pdparams">训练模型</a></td>
+<td>83.8</td>
+<td>89.55 / 70.19</td>
+<td>383.15 / 383.15</td>
+<td>84.3</td>
+<td>PP-OCRv5 的服务端文本检测模型，精度更高，适合在性能较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_det_pretrained.pdparams">训练模型</a></td>
+<td>79.0</td>
+<td>10.67 / 6.36</td>
+<td>57.77 / 28.15</td>
+<td>4.7</td>
+<td>PP-OCRv5 的移动端文本检测模型，效率更高，适合在端侧设备部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_det_pretrained.pdparams">训练模型</a></td>
+<td>69.2</td>
+<td>127.82 / 98.87</td>
+<td>585.95 / 489.77</td>
+<td>109</td>
+<td>PP-OCRv4 的服务端文本检测模型，精度更高，适合在性能较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams">训练模型</a></td>
+<td>63.8</td>
+<td>9.87 / 4.17</td>
+<td>56.60 / 20.79</td>
+<td>4.7</td>
+<td>PP-OCRv4 的移动端文本检测模型，效率更高，适合在端侧设备部署</td>
+</tr>
+</tbody>
+</table>
+
+> *注：PP-OCRv6 指标基于内部多场景评估集测得，PP-OCRv5/v4 指标基于通用评估集测得，两者评估集不同，指标不可直接对比。
+
+<strong>测试环境说明:</strong>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+              <li><strong>测试数据集：</strong>PaddleOCR3.0 全新构建多语种（包含中、繁、英、日），覆盖街景、网图、文档、手写、模糊、旋转、扭曲等多个场景的文本检测数据集，包含2677 张图片。</li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+
+## 三、快速开始
+
+> ❗ 在快速开始前，请先安装 PaddleOCR 的 wheel 包，详细请参考 [安装教程](../installation.md)。
+
+### 3.1 环境准备
+
+#### 3.1.1 基础安装
+
+```bash
+# 安装基础版本（仅包含文本检测功能）
+pip install paddleocr
+
+# 安装完整版本（包含所有功能）
+pip install "paddleocr[all]"
+```
+
+#### 3.1.2 环境验证
+
+```python
+# 验证安装是否成功
+import paddleocr
+print(f"PaddleOCR版本: {paddleocr.__version__}")
+
+# 验证GPU是否可用
+import paddle
+print(f"Paddle版本: {paddle.__version__}")
+print(f"GPU可用: {paddle.is_compiled_with_cuda()}")
+print(f"GPU数量: {paddle.device.cuda.device_count()}")
+```
+
+### 3.2 命令行快速体验
+
+使用一行命令即可快速体验：
+
+```bash
+# 使用默认模型进行文本检测
+paddleocr text_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_001.png
+
+# 指定模型
+paddleocr text_detection -i general_ocr_001.png --model_name PP-OCRv6_small_det
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下命令：
+
+```bash
+# 使用 transformers 引擎进行推理
+paddleocr text_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_001.png \
+    --engine transformers
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下命令：
+
+```bash
+# 使用 onnxruntime 引擎进行推理
+paddleocr text_detection -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_001.png \
+    --engine onnxruntime
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+<b>注：</b>PaddleOCR 官方模型默认从 HuggingFace 获取，如运行环境访问 HuggingFace 不便，可通过环境变量修改模型源为 BOS：`PADDLE_PDX_MODEL_SOURCE="BOS"`，未来将支持更多主流模型源；
+
+### 3.3 Python API 使用
+
+您也可以将文本检测的模块中的模型推理集成到您的项目中。运行以下代码前，请您下载[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_001.png)到本地。
+
+```python
+from paddleocr import TextDetection
+model = TextDetection()
+output = model.predict("general_ocr_001.png", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下代码：
+
+```python
+from paddleocr import TextDetection
+model = TextDetection(engine="transformers")
+output = model.predict("general_ocr_001.png", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下代码：
+
+```python
+from paddleocr import TextDetection
+model = TextDetection(engine="onnxruntime")
+output = model.predict("general_ocr_001.png", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+训练后的模型如果想使用 `paddle_dynamic` 或 `transformers` 引擎，请参考后文 [推理引擎](#五推理引擎) 中的 [权重转换](#52-权重转换) 部分将模型由 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式。
+
+运行后，得到的结果为：
+
+```bash
+{'res': {'input_path': 'general_ocr_001.png', 'page_index': None, 'dt_polys': array([[[ 75, 549],
+        ...,
+        [ 77, 586]],
+
+       ...,
+
+       [[ 31, 406],
+        ...,
+        [ 34, 455]]], dtype=int16), 'dt_scores': [0.873949039891189, 0.8948166013613552, 0.8842595305917041, 0.876953790920377]}}
+```
+
+运行结果参数含义如下：
+<ul>
+<li><code>input_path</code>：表示输入待预测图像的路径</li>
+<li><code>page_index</code>：如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+<li><code>dt_polys</code>：表示预测的文本检测框，其中每个文本检测框包含一个四边形的四个顶点。其中每个顶点都是一个列表，分别表示该顶点的x坐标和y坐标</li>
+<li><code>dt_scores</code>：表示预测的文本检测框的置信度</li>
+</ul>
+可视化图片如下：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/text_det/general_ocr_001_res.png"/>
+
+相关方法、参数等说明如下：
+
+* <code>TextDetection</code>类实例化文本检测模型，具体说明如下：
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>含义：</b>模型名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，则使用<code>PP-OCRv6_medium_det</code>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>含义：</b>模型存储路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+<b>例如：</b><code>"cpu"</code>、<code>"gpu"</code>、<code>"npu"</code>、<code>"gpu:0"</code>、<code>"gpu:0,1"</code>。<br/>
+如指定多个设备，将进行并行推理。<br/>
+默认情况下，优先使用 GPU 0；若不可用则使用 CPU。
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>、<code>onnxruntime</code>。保持为默认值 <code>None</code> 时，本地推理默认使用 <code>paddle_static</code> 引擎。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>当使用 Paddle Inference 的 TensorRT 子图引擎时设置的计算精度。<br/>
+<b>说明：</b>
+<b>可选项：</b><code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+<b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上推理时使用的线程数量。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>limit_side_len</code></td>
+<td><b>含义：</b>检测的图像边长限制：<code>int</code> 表示边长限制数值。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用模型默认配置。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>limit_type</code></td>
+<td><b>含义：</b>检测的图像边长限制，检测的边长限制类型。<br/>
+<b>说明：</b>
+<code>"min"</code> 表示保证图像最短边不小于det_limit_side_len，<code>"max"</code>表示保证图像最长边不大于limit_side_len。如果设置为<code>None</code>，将使用模型默认配置。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>max_side_limit</code></td>
+<td><b>含义：</b>检测的图像边长最大值限制：<code>int</code> 限制输入检测模型的图片最长边。<br/>
+<b>说明：</b>
+如果设置为 <code>None</code>，将使用模型默认配置。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>thresh</code></td>
+<td><b>含义：</b>像素得分阈值。输出概率图中得分大于该阈值的像素点被认为是文本像素。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用模型默认配置。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>box_thresh</code></td>
+<td><b>含义：</b>检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用模型默认配置。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>unclip_ratio</code></td>
+<td><b>含义：</b>Vatti clipping算法的扩张系数，使用该方法对文字区域进行扩张。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用模型默认配置。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>input_shape</code></td>
+<td><b>含义：</b>模型输入图像尺寸，格式为 <code>(C, H, W)</code>。</td>
+<td><code>tuple|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+
+* 调用文本检测模型的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。另外，本模块还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。<code>predict()</code>  方法参数有 <code>input</code> 、<code>batch_size</code>、 <code>limit_side_len</code>、 <code>limit_type</code>、 <code>thresh</code>、 <code>box_thresh</code>、 <code>max_candidates</code>、<code>unclip_ratio</code>，具体说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_rec_001.png">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>含义：</b>批大小<br/>
+<b>说明：</b>
+可设置为任意正整数。</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+<tr>
+<td><code>limit_side_len</code></td>
+<td><b>含义:</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>limit_type</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>box_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+* 对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为<code>json</code>文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将结果保存为图像格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+</table>
+
+* 此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">获取预测的<code>json</code>格式的结果</td>
+</tr>
+<tr>
+<td rowspan="1"><code>img</code></td>
+<td rowspan="1">获取格式为<code>dict</code>的可视化图像</td>
+</tr>
+</table>
+
+## 四、二次开发
+
+如果以上模型在您的场景上效果仍然不理想，您可以尝试以下步骤进行二次开发，此处以训练 `PP-OCRv5_server_det` 举例，其他模型替换对应配置文件即可。首先，您需要准备文本检测的数据集，可以参考[文本检测 Demo 数据](https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_det_dataset_examples.tar)的格式准备，准备好后，即可按照以下步骤进行模型训练和导出，导出后，可以将模型快速集成到上述 API 中。此处以文本检测 Demo 数据示例。在训练模型之前，请确保已经按照[安装文档](../installation.md)安装了 PaddleOCR 所需要的依赖。
+
+
+### 4.1 数据集、预训练模型准备
+
+#### 4.1.1 准备数据集
+
+```shell
+# 下载示例数据集
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_det_dataset_examples.tar
+tar -xf ocr_det_dataset_examples.tar
+```
+
+#### 4.1.2 下载预训练模型
+
+```shell
+# 下载 PP-OCRv5_server_det 预训练模型
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_det_pretrained.pdparams 
+```
+
+### 4.2 模型训练
+
+PaddleOCR 对代码进行了模块化，训练 `PP-OCRv5_server_det` 识别模型时需要使用 `PP-OCRv5_server_det` 的[配置文件](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/configs/det/PP-OCRv5/PP-OCRv5_server_det.yml)。
+
+
+训练命令如下：
+
+```bash
+#单卡训练 (默认训练方式)
+python3 tools/train.py -c configs/det/PP-OCRv5/PP-OCRv5_server_det.yml \
+    -o Global.pretrained_model=./PP-OCRv5_server_det_pretrained.pdparams \
+    Train.dataset.data_dir=./ocr_det_dataset_examples \
+    Train.dataset.label_file_list='[./ocr_det_dataset_examples/train.txt]' \
+    Eval.dataset.data_dir=./ocr_det_dataset_examples \
+    Eval.dataset.label_file_list='[./ocr_det_dataset_examples/val.txt]'
+
+#多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3' tools/train.py \
+    -c configs/det/PP-OCRv5/PP-OCRv5_server_det.yml \
+    -o Global.pretrained_model=./PP-OCRv5_server_det_pretrained.pdparams \
+    Train.dataset.data_dir=./ocr_det_dataset_examples \
+    Train.dataset.label_file_list='[./ocr_det_dataset_examples/train.txt]' \
+    Eval.dataset.data_dir=./ocr_det_dataset_examples \
+    Eval.dataset.label_file_list='[./ocr_det_dataset_examples/val.txt]'
+```
+
+### 4.3 模型评估
+
+您可以评估已经训练好的权重，如，`output/PP-OCRv5_server_det/best_accuracy.pdprams`，使用如下命令进行评估：
+
+```bash
+# 注意将pretrained_model的路径设置为本地路径。若使用自行训练保存的模型，请注意修改路径和文件名为{path/to/weights}/{model_name}。
+ # demo 测试集评估
+python3 tools/eval.py -c configs/det/PP-OCRv5/PP-OCRv5_server_det.yml \
+    -o Global.pretrained_model=output/PP-OCRv5_server_det/best_accuracy.pdparams \
+    Eval.dataset.data_dir=./ocr_det_dataset_examples \
+    Eval.dataset.label_file_list='[./ocr_det_dataset_examples/val.txt]'
+```
+
+### 4.4 模型导出
+
+```bash
+python3 tools/export_model.py -c configs/det/PP-OCRv5/PP-OCRv5_server_det.yml -o \
+    Global.pretrained_model=output/PP-OCRv5_server_det/best_accuracy.pdparams \
+    Global.save_inference_dir="./PP-OCRv5_server_det_infer/"
+```
+
+ 导出模型后，静态图模型会存放于当前目录的`./PP-OCRv5_server_det_infer/`中，在该目录下，您将看到如下文件：
+ ```
+ ./PP-OCRv5_server_det_infer/
+ ├── inference.json
+ ├── inference.pdiparams
+ ├── inference.yml
+ ```
+至此，二次开发完成，该静态图模型可以直接集成到 PaddleOCR 的 API 中。
+
+训练后的模型如果想使用 `paddle_dynamic` 或 `transformers` 引擎，请参考后文 [推理引擎](#五推理引擎) 中的 [权重转换](#52-权重转换) 部分将模型由 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式。
+
+## 五、推理引擎 {#五推理引擎}
+
+关于推理引擎的详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。
+
+### 5.1 速度数据
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">PP-OCRv5_mobile_det</td>
+            <td>paddle_static</td>
+            <td>11.43</td>
+            <td>13.80</td>
+            <td>2.15</td>
+            <td>27.58</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>11.70</td>
+            <td>48.36</td>
+            <td>2.47</td>
+            <td>62.71</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>14.05</td>
+            <td>18.45</td>
+            <td>3.98</td>
+            <td>37.54</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>9.98</td>
+            <td>5.70</td>
+            <td>2.04</td>
+            <td>17.90</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-OCRv5_server_det</td>
+            <td>paddle_static</td>
+            <td>13.24</td>
+            <td>26.91</td>
+            <td>2.63</td>
+            <td>43.05</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>11.82</td>
+            <td>45.56</td>
+            <td>2.52</td>
+            <td>60.10</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>14.56</td>
+            <td>13.76</td>
+            <td>7.44</td>
+            <td>36.76</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>10.01</td>
+            <td>13.76</td>
+            <td>1.92</td>
+            <td>25.86</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-OCRv6_medium_det</td>
+            <td>paddle_static</td>
+            <td>13.89</td>
+            <td>16.02</td>
+            <td>2.49</td>
+            <td>33.14</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>11.42</td>
+            <td>26.23</td>
+            <td>2.30</td>
+            <td>40.10</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>11.40</td>
+            <td>8.57</td>
+            <td>8.35</td>
+            <td>29.57</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>10.80</td>
+            <td>13.06</td>
+            <td>2.19</td>
+            <td>26.18</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-OCRv6_small_det</td>
+            <td>paddle_static</td>
+            <td>10.91</td>
+            <td>10.97</td>
+            <td>2.41</td>
+            <td>24.45</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>11.56</td>
+            <td>22.17</td>
+            <td>2.66</td>
+            <td>36.55</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>11.70</td>
+            <td>7.34</td>
+            <td>3.87</td>
+            <td>23.89</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>11.32</td>
+            <td>7.46</td>
+            <td>2.54</td>
+            <td>21.49</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-OCRv6_tiny_det</td>
+            <td>paddle_static</td>
+            <td>11.14</td>
+            <td>10.71</td>
+            <td>2.84</td>
+            <td>24.85</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>11.52</td>
+            <td>21.70</td>
+            <td>2.94</td>
+            <td>36.31</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>10.90</td>
+            <td>6.99</td>
+            <td>4.13</td>
+            <td>23.00</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>11.19</td>
+            <td>6.35</td>
+            <td>2.79</td>
+            <td>20.49</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>测试环境说明:</strong>
+<ul>
+    <li><strong>测试数据：</strong><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_001.jpg">示例图片</a></li>
+    <li><strong>硬件配置：</strong>
+        <ul>
+            <li>GPU：NVIDIA A100 40G</li>
+            <li>CPU：Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>软件环境：</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 5.2 权重转换 {#52-权重转换}
+
+使用推理引擎时，系统会自动下载官方预训练模型。若需使用自训练模型配合 `paddle_dynamic` 或 `transformers` 引擎，请参考 [PaddleX 文本检测模块权重转换](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/text_detection.html#442) 部分，将 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式，即可无缝集成到 PaddleOCR 的 API 中进行推理。若需使用自训练模型配合`onnxruntime`引擎，请参考[PaddleX 获取 ONNX 模型](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html)获取onnx模型，即可无缝集成到 PaddleOCR 的 API 中进行推理。
+
+## 六、常见问题与解决方案
+
+### 6.1 性能优化问题
+
+#### Q: GPU推理速度慢怎么办？
+
+**A**: 可以通过以下方式优化：
+（1）启用高性能推理：设置`enable_hpi=True`，自动选择最优加速策略
+（2）启用TensorRT加速：设置`use_tensorrt=True`，需要CUDA 11.8+和TensorRT 8.6+
+（3）使用半精度：设置`precision="fp16"`，可以显著提升速度
+（4）调整批处理大小：根据显存大小设置合适的`batch_size`
+（5）使用轻量模型：在精度要求不高时使用 `PP-OCRv6_small`/`PP-OCRv6_tiny` 等轻量级模型
+
+#### Q: GPU内存不足（CUDA out of memory）怎么办？
+
+**A**: 可以通过以下方式解决：
+（1）减小批处理大小：将`batch_size`设置为1
+（2）减小图像尺寸：设置`det_limit_side_len=640`
+（3）启用内存优化：设置`enable_memory_optim=True`
+（4）限制GPU内存使用：设置`gpu_mem=200`
+（5）使用轻量模型：切换到 `PP-OCRv6_small`/`PP-OCRv6_tiny` 等轻量级模型
+
+### 6.2 检测精度问题
+
+#### Q: 检测框不准确或漏检怎么办？
+
+**A**: 可以通过以下方式优化：
+（1）调整检测参数：
+```python
+model = TextDetection(
+    thresh=0.3,  # 降低像素阈值
+    box_thresh=0.5,  # 降低检测框阈值
+    unclip_ratio=2.0,  # 增大扩张系数
+    limit_side_len=1216  # 增大图像尺寸
+)
+```
+（2）使用更精确的后处理模式：设置`det_db_score_mode="slow"`
+（3）启用膨胀处理：设置`use_dilation=True`
+
+### 6.3 模型选择建议
+
+#### Q: 如何选择合适的模型？
+
+**A**: 根据应用场景选择：
+- 服务器高精度场景：使用 `PP-OCRv6_medium_det`，精度最高
+- 移动端部署：使用 `PP-OCRv6_small_det`，兼顾精度与效率
+- 端侧/IoT：使用 `PP-OCRv6_tiny_det`，模型最小
+- 实时处理：使用 `PP-OCRv6_small_det` 或 `PP-OCRv6_tiny_det`，推理更快
+
+### 6.4 参数调优建议
+
+#### Q: 如何调优检测参数？
+
+**A**: 通过参数`limit_type`和`limit_side_len`来对图片的尺寸进行限制，`limit_type`可选参数为[`max`, `min`]，`limit_side_len` 为正整数，一般设置为 32 的倍数，比如 960。
+
+如果输入图形分辨率不大，建议使用`limit_type=min` 和 `limit_side_len=960` 节省计算资源的同时能获得最佳检测效果。如果输入图片的分辨率比较大，而且想使用更大的分辨率预测，可以设置 `limit_side_len` 为想要的值，比如 1216。
+
+#### Q: 不同场景的参数配置建议？
+
+**A**: 
+- **高精度配置**：`limit_side_len=1216`, `thresh=0.3`, `box_thresh=0.5`, `unclip_ratio=1.5`
+- **高速度配置**：`limit_side_len=640`, `thresh=0.5`, `box_thresh=0.7`, `unclip_ratio=1.2`
+- **平衡配置**：`limit_side_len=960`, `thresh=0.4`, `box_thresh=0.6`, `unclip_ratio=1.5`
+
+### 6.5 错误处理
+
+#### Q: 模型加载失败怎么办？
+
+**A**: 
+（1）检查模型路径是否正确
+（2）确保模型文件完整（inference.pdmodel, inference.pdiparams, inference.json）
+（3）设置模型下载源：`os.environ['PADDLE_PDX_MODEL_SOURCE'] = 'BOS'`
+（4）使用本地模型：指定`model_dir`参数
+
+#### Q: 输入数据格式错误怎么办？
+
+**A**: 
+（1）检查图像文件是否存在
+（2）确保图像格式正确（支持PNG、JPG、JPEG）
+（3）验证图像尺寸（最小10x10像素）
+（4）检查图像是否为3通道格式
diff --git a/docs/version3.x/module_usage/text_image_unwarping.en.md b/docs/version3.x/module_usage/text_image_unwarping.en.md
new file mode 100644
index 0000000..ed826c0
--- /dev/null
+++ b/docs/version3.x/module_usage/text_image_unwarping.en.md
@@ -0,0 +1,478 @@
+---
+
+comments: true
+
+---
+
+# Text Image Rectification Module Usage Tutorial
+
+## 1. Overview
+
+The primary purpose of text image rectification is to perform geometric transformations on images to correct distortions, inclinations, perspective deformations, etc., in the document images for more accurate subsequent text recognition.
+
+## 2. Supported Model List
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing. The "Normal Mode" values correspond to the local <code>paddle_static</code> inference engine.
+
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>CER</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">Training Model</a></td>
+<td>0.179</td>
+<td>19.05 / 19.05</td>
+<td>- / 869.82</td>
+<td>30.3</td>
+<td>High-accuracy text image rectification model</td>
+</tr>
+</tbody>
+</table>
+
+<strong>Test Environment Description:</strong>
+
+  <ul>
+      <li><b>Performance Test Environment</b>
+          <ul>
+              <li><strong>Test Dataset:</strong> <a href="https://www3.cs.stonybrook.edu/~cvl/docunet.html">DocUNet benchmark</a> dataset.</li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>Inference Mode Explanation</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration</th>
+            <th>CPU Configuration</th>
+            <th>Acceleration Technology Combination</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Regular Mode</td>
+            <td>FP32 Precision / No TRT Acceleration</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>High-Performance Mode</td>
+            <td>Choose the optimal combination of prior precision type and acceleration strategy</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Choose the optimal prior backend (Paddle/OpenVINO/TRT, etc.)</td>
+        </tr>
+    </tbody>
+</table>
+
+## 3. Quick Start
+
+> ❗ Before starting quickly, please first install the PaddleOCR wheel package. For details, please refer to the [installation tutorial](../installation.md).
+
+You can quickly experience it with one command:
+
+```bash
+paddleocr text_image_unwarping -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/doc_test.jpg
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr text_image_unwarping -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/doc_test.jpg \
+    --engine transformers
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following command:
+
+```bash
+# Use the onnxruntime engine for inference
+paddleocr text_image_unwarping -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/doc_test.jpg \
+    --engine onnxruntime
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+<b>Note: </b>The official models would be download from HuggingFace by default. If can't access to HuggingFace, please set the environment variable `PADDLE_PDX_MODEL_SOURCE="BOS"` to change the model source to BOS. In the future, more model sources will be supported.
+
+You can also integrate the model inference from the image rectification module into your project. Before running the following code, please download the [sample image](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/doc_test.jpg) locally.
+
+```python
+from paddleocr import TextImageUnwarping
+model = TextImageUnwarping(model_name="UVDoc")
+output = model.predict("doc_test.jpg", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following code:
+
+```python
+from paddleocr import TextImageUnwarping
+model = TextImageUnwarping(
+    model_name="UVDoc",
+    engine="transformers",
+)
+output = model.predict("doc_test.jpg", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following code:
+
+```python
+from paddleocr import TextImageUnwarping
+model = TextImageUnwarping(
+    model_name="UVDoc",
+    engine="onnxruntime",
+)
+output = model.predict("doc_test.jpg", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+After running, the result obtained is:
+
+```bash
+{'res': {'input_path': 'doc_test.jpg', 'page_index': None, 'doctr_img': '...'}}
+```
+
+The meanings of the parameters in the result are as follows:
+<ul>
+<li><code>input_path</code>：Indicates the path of the image to be rectified</li>
+<li><code>doctr_img</code>：Indicates the rectified image result. Due to the large amount of data, it is not convenient to print directly, so it is replaced here with<code>...</code>.You can use<code>res.save_to_img()</code>to save the prediction result as an image, and <code>res.save_to_json()</code> to save the prediction result as a json file.</li>
+</ul>
+
+The visualized image is as follows:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/image_unwarp/doc_test_res.jpg">
+
+The relevant methods, parameters, etc., are described as follows:
+
+* <code>TextImageUnwarping</code> instantiates the image rectification model (taking <code>UVDoc</code> as an example here), with specific explanations as follows:
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>Meaning:</b> Name of the model</td>
+<td><code>str</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>Meaning:</b> Model storage path</td>
+<td><code>str</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b> Device(s) to use for inference.<br/>
+<b>Description:</b>
+<b>Examples:</b> <code>cpu</code>, <code>gpu</code>, <code>npu</code>, <code>gpu:0</code>, <code>gpu:0,1</code>.<br/>
+If multiple devices are specified, inference will be performed in parallel. Note that parallel inference is not always supported.<br/>
+By default, GPU 0 will be used if available; otherwise, the CPU will be used.
+</td>
+<td><code>str</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, <code>transformers</code>, and <code>onnxruntime</code>. When left as <code>None</code>, local inference uses the <code>paddle_static</code> engine by default. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to use the high performance inference.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to use the Paddle Inference TensorRT subgraph engine.<br/> 
+<b>Description:</b>
+If the model does not support acceleration through TensorRT, setting this flag will not enable acceleration.<br/>
+For Paddle with CUDA version 11.8, the compatible TensorRT version is 8.x (x>=6), and it is recommended to install TensorRT 8.6.1.6.<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b>Precision for TensorRT when using the Paddle Inference TensorRT subgraph engine.<br/>
+<b>Description:</b>
+<b>Options:</b> <code>"fp32"</code>, <code>"fp16"</code>, etc.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+<b>Meaning:</b> Whether to enable MKL-DNN acceleration for inference. <br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support it, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b>MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads to use for inference on CPUs.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+</tbody>
+</table>
+
+* Call the  <code>predict()</code> method of the image rectification model for inference prediction. This method will return a result list. Additionally, this module also provides a <code>predict_iter()</code> method. Both methods are consistent in terms of parameter acceptance and result return. The difference is that <code>predict_iter()</code> returns a <code>generator</code>, which can process and obtain prediction results step by step, suitable for handling large datasets or scenarios where memory saving is desired. You can choose to use either of these methods according to your actual needs. The  <code>predict()</code> method has parameters <code>input</code> and <code>batch_size</code>, with specific explanations as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b> Input data to be predicted. Required.<br/> 
+<b>Description:</b>
+Supports multiple input types:
+<ul>
+<li><b>Python Var</b>: e.g., <code>numpy.ndarray</code> representing image data</li>
+<li><b>str</b>: 
+    <ul>
+    <li>Local image or PDF file path: <code>/root/data/img.jpg</code>;</li>
+    <li><b>URL</b> of image or PDF file: e.g., <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_doc_preprocessor_002.png">example</a>;</li>
+    <li><b>Local directory</b>: directory containing images for prediction, e.g., <code>/root/data/</code> (Note: directories containing PDF files are not supported; PDFs must be specified by exact file path)</li>
+    </ul>
+    </li>
+<li><b>list</b>: Elements must be of the above types, e.g., <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>Meaning:</b> Batch size<br/>
+<b>Description:</b>
+Positive integer.</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+
+* Process the prediction results. The prediction result for each sample is a corresponding Result object, which supports printing, saving as an image, and saving as a <code>json</code> file:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Description</th>
+<th>Parameter</th>
+<th>Type</th>
+<th>Parameter Description</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">Print result to terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output content using <code>JSON</code> indentation</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specifies the indentation level to beautify the output <code>JSON</code> data, making it more readable, effective only when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Controls whether to escape non-<code>ASCII</code> characters into <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> will retain the original characters, effective only when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">Save the result as a json format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The path to save the file. When specified as a directory, the saved file is named consistent with the input file type.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specifies the indentation level to beautify the output <code>JSON</code> data, making it more readable, effective only when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Controls whether to escape non-<code>ASCII</code> characters into <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> will retain the original characters, effective only when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save the result as an image format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The path to save the file. When specified as a directory, the saved file is named consistent with the input file type.</td>
+<td>None</td>
+</tr>
+</table>
+
+* Additionally, the result can be obtained through attributes that provide the visualized images with results and the prediction results, as follows:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">Get the prediction result in <code>json</code> format</td>
+</tr>
+<tr>
+<td rowspan = "1"><code>img</code></td>
+<td rowspan = "1">Get the visualized image in <code>dict</code> format</td>
+</tr>
+
+</table>
+
+## 4. Secondary Development
+
+The current module does not support fine-tuning training and only supports inference integration. Concerning fine-tuning training for this module, there are plans to support it in the future.
+
+## 5. Inference Engine
+
+For detailed descriptions, values, compatibility rules, and examples of the inference engine, please refer to <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration Description</a>.
+
+### 5.1 Speed Data
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">UVDoc</td>
+            <td>paddle_static</td>
+            <td>14.96</td>
+            <td>18.60</td>
+            <td>1.93</td>
+            <td>36.66</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>10.90</td>
+            <td>27.59</td>
+            <td>1.96</td>
+            <td>40.94</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>13.54</td>
+            <td>6.74</td>
+            <td>0.91</td>
+            <td>33.07</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>10.60</td>
+            <td>8.44</td>
+            <td>1.75</td>
+            <td>21.30</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>Test Environment Description:</strong>
+<ul>
+    <li><strong>Test Data:</strong> <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/doc_test.jpg">Sample Image</a></li>
+    <li><strong>Hardware Configuration:</strong>
+        <ul>
+            <li>GPU: NVIDIA A100 40G</li>
+            <li>CPU: Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>Software Environment:</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+## 6. FAQ
diff --git a/docs/version3.x/module_usage/text_image_unwarping.md b/docs/version3.x/module_usage/text_image_unwarping.md
new file mode 100644
index 0000000..d3e8028
--- /dev/null
+++ b/docs/version3.x/module_usage/text_image_unwarping.md
@@ -0,0 +1,473 @@
+---
+comments: true
+---
+
+# 文本图像矫正模块使用教程
+
+## 一、概述
+
+文本图像矫正的主要目的是针对图像进行几何变换，以纠正图像中的文档扭曲、倾斜、透视变形等问题，以供后续的文本识别进行更加准确。
+
+## 二、支持模型列表
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。表格中的“常规模式”耗时对应本地 <code>paddle_static</code> 推理引擎。
+
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>CER </th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">训练模型</a></td>
+<td>0.179</td>
+<td>19.05 / 19.05</td>
+<td>- / 869.82</td>
+<td>30.3</td>
+<td>高精度文本图像矫正模型</td>
+</tr>
+</tbody>
+</table>
+
+<strong>测试环境说明:</strong>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+              <li><strong>测试数据集：</strong><a href="https://www3.cs.stonybrook.edu/~cvl/docunet.html">DocUNet benchmark</a>数据集。</li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+
+
+## 三、快速开始
+
+> ❗ 在快速开始前，请先安装 PaddleOCR 的 wheel 包，详细请参考 [安装教程](../installation.md)。
+
+使用一行命令即可快速体验：
+
+```bash
+paddleocr text_image_unwarping -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/doc_test.jpg
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下命令：
+
+```bash
+# 使用 transformers 引擎进行推理
+paddleocr text_image_unwarping -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/doc_test.jpg \
+    --engine transformers
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下命令：
+
+```bash
+# 使用 onnxruntime 引擎进行推理
+paddleocr text_image_unwarping -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/doc_test.jpg \
+    --engine onnxruntime
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+<b>注：</b>PaddleOCR 官方模型默认从 HuggingFace 获取，如运行环境访问 HuggingFace 不便，可通过环境变量修改模型源为 BOS：`PADDLE_PDX_MODEL_SOURCE="BOS"`，未来将支持更多主流模型源；
+
+您也可以将图像矫正的模块中的模型推理集成到您的项目中。运行以下代码前，请您下载[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/doc_test.jpg)到本地。
+
+```python
+from paddleocr import TextImageUnwarping
+model = TextImageUnwarping(model_name="UVDoc")
+output = model.predict("doc_test.jpg", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下代码：
+
+```python
+from paddleocr import TextImageUnwarping
+model = TextImageUnwarping(
+    model_name="UVDoc",
+    engine="transformers",
+)
+output = model.predict("doc_test.jpg", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下代码：
+
+```python
+from paddleocr import TextImageUnwarping
+model = TextImageUnwarping(
+    model_name="UVDoc",
+    engine="onnxruntime",
+)
+output = model.predict("doc_test.jpg", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+运行后，得到的结果为：
+
+```bash
+{'res': {'input_path': 'doc_test.jpg', 'page_index': None, 'doctr_img': '...'}}
+```
+
+运行结果参数含义如下：
+<ul>
+<li><code>input_path</code>：表示输入待矫正图像的路径</li>
+<li><code>doctr_img</code>：表示矫正后的图像结果，由于数据过多不便于直接print，所以此处用<code>...</code>替换，可以通过<code>res.save_to_img()</code>将预测结果保存为图片，通过<code>res.save_to_json()</code>将预测结果保存为json文件。</li>
+</ul>
+
+可视化图片如下：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/image_unwarp/doc_test_res.jpg">
+
+相关方法、参数等说明如下：
+
+* <code>TextImageUnwarping</code>实例化图像矫正模型（此处以<code>UVDoc</code>为例），具体说明如下：
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>含义：</b>模型名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，则使用<code>UVDoc</code>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>含义：</b>模型存储路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+<b>例如：</b><code>"cpu"</code>、<code>"gpu"</code>、<code>"npu"</code>、<code>"gpu:0"</code>、<code>"gpu:0,1"</code>。<br/>
+如指定多个设备，将进行并行推理。<br/>
+默认情况下，优先使用 GPU 0；若不可用则使用 CPU。
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>、<code>onnxruntime</code>。保持为默认值 <code>None</code> 时，本地推理默认使用 <code>paddle_static</code> 引擎。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。</br>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>当使用 Paddle Inference 的 TensorRT 子图引擎时设置的计算精度。<br/>
+<b>说明：</b>
+<b>可选项：</b><code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+<b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上推理时使用的线程数量。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+</tbody>
+</table>
+
+* 调用图像矫正模型的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。另外，本模块还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。<code>predict()</code> 方法参数有 <code>input</code> 和 <code>batch_size</code>，具体说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/img_rot180_demo.jpg">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>含义：</b>批大小<br/>
+<b>说明：</b>
+可设置为任意正整数。</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+
+* 对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为<code>json</code>文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将结果保存为图像格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+</table>
+
+* 此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">获取预测的<code>json</code>格式的结果</td>
+</tr>
+<tr>
+<td rowspan = "1"><code>img</code></td>
+<td rowspan = "1">获取格式为<code>dict</code>的可视化图像</td>
+</tr>
+
+</table>
+
+
+## 四、二次开发
+
+当前模块暂时不支持微调训练，仅支持推理集成。关于该模块的微调训练，计划在未来支持。
+
+## 五、推理引擎
+
+关于推理引擎的详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。
+
+### 5.1 速度数据
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">UVDoc</td>
+            <td>paddle_static</td>
+            <td>14.96</td>
+            <td>18.60</td>
+            <td>1.93</td>
+            <td>36.66</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>10.90</td>
+            <td>27.59</td>
+            <td>1.96</td>
+            <td>40.94</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>13.54</td>
+            <td>6.74</td>
+            <td>0.91</td>
+            <td>33.07</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>10.60</td>
+            <td>8.44</td>
+            <td>1.75</td>
+            <td>21.30</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>测试环境说明:</strong>
+<ul>
+    <li><strong>测试数据：</strong><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/doc_test.jpg">示例图片</a></li>
+    <li><strong>硬件配置：</strong>
+        <ul>
+            <li>GPU：NVIDIA A100 40G</li>
+            <li>CPU：Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>软件环境：</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+## 六、FAQ
diff --git a/docs/version3.x/module_usage/text_recognition.en.md b/docs/version3.x/module_usage/text_recognition.en.md
new file mode 100644
index 0000000..3f0f4bd
--- /dev/null
+++ b/docs/version3.x/module_usage/text_recognition.en.md
@@ -0,0 +1,1159 @@
+---
+comments: true
+---
+
+# Text Recognition Module Tutorial
+
+## 1. Overview
+
+The text recognition module is the core part of the OCR (Optical Character Recognition) system, responsible for extracting text information from text regions in images. The performance of this module directly affects the accuracy and efficiency of the entire OCR system. The text recognition module usually receives the bounding boxes of text regions output by the text detection module as input, and then converts the text in the images into editable and searchable electronic text through complex image processing and deep learning algorithms. The accuracy of text recognition results is crucial for subsequent applications such as information extraction and data mining.
+
+## 2. List of Supported Models
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing. The "Normal Mode" values correspond to the local <code>paddle_static</code> inference engine.
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Links</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+<tr>
+<td>PP-OCRv6_medium_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_medium_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_medium_rec_pretrained.pdparams">Training Model</a></td>
+<td>83.2*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>73.3</td>
+<td rowspan="3">PP-OCRv6 text recognition models based on PPLCNetV4 + LightSVTR + CTC/NRTR multi-head decoder, single model supports 50 languages (tiny: 49). Medium achieves +5.1% over PP-OCRv5_server.</td>
+</tr>
+<tr>
+<td>PP-OCRv6_small_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_small_rec_pretrained.pdparams">Training Model</a></td>
+<td>81.3*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>20.4</td>
+</tr>
+<tr>
+<td>PP-OCRv6_tiny_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_tiny_rec_pretrained.pdparams">Training Model</a></td>
+<td>73.5*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>4.4</td>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>86.38</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec is a new generation text recognition model. It is designed to efficiently and accurately support the recognition of Simplified Chinese, Traditional Chinese, English, Japanese, as well as complex text scenarios such as handwriting, vertical text, pinyin, and rare characters with a single model. While maintaining recognition performance, it also balances inference speed and model robustness, providing efficient and accurate technical support for document understanding in various scenarios.</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>81.29</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">Pretrained Model</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc is trained on a mixed dataset of more Chinese document data and PP-OCR training data, building upon PP-OCRv4_server_rec. It enhances the recognition capabilities for some Traditional Chinese characters, Japanese characters, and special symbols, supporting over 15,000 characters. In addition to improving document-related text recognition, it also enhances general text recognition capabilities.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>A lightweight recognition model of PP-OCRv4 with high inference efficiency, suitable for deployment on various hardware devices, including edge devices.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>The server-side model of PP-OCRv4, offering high inference accuracy and deployable on various servers.</td>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>An ultra-lightweight English recognition model trained based on the PP-OCRv4 recognition model, supporting English and numeric character recognition.</td>
+</tr>
+</table>
+
+> *Note: PP-OCRv6 metrics are evaluated on an internal multi-scenario evaluation set, while PP-OCRv5/v4 metrics are based on a general evaluation set. As the evaluation sets differ, the metrics are not directly comparable.
+
+> ❗ The above lists the <b>4 core models</b> mainly supported by the text recognition module. The module supports a total of <b>20 full models</b>, including multiple multilingual text recognition models. The complete model list is as follows:
+
+<details><summary> 👉Model List Details</summary>
+
+* <b>PP-OCRv6 Multi-Scenario Models</b>
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Links</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</table>
+
+* <b>PP-OCRv5 Multi-Scenario Models</b>
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Links</th>
+<th>Chinese Recognition Avg Accuracy(%)</th>
+<th>English Recognition Avg Accuracy(%)</th>
+<th>Traditional Chinese Recognition Avg Accuracy(%)</th>
+<th>Japanese Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+<tr>
+<td>PP-OCRv6_medium_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_medium_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_medium_rec_pretrained.pdparams">Training Model</a></td>
+<td>83.2*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>73.3</td>
+<td rowspan="3">PP-OCRv6 text recognition models based on PPLCNetV4 + LightSVTR + CTC/NRTR multi-head decoder, single model supports 50 languages (tiny: 49). Medium achieves +5.1% over PP-OCRv5_server.</td>
+</tr>
+<tr>
+<td>PP-OCRv6_small_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_small_rec_pretrained.pdparams">Training Model</a></td>
+<td>81.3*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>20.4</td>
+</tr>
+<tr>
+<td>PP-OCRv6_tiny_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_tiny_rec_pretrained.pdparams">Training Model</a></td>
+<td>73.5*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>4.4</td>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>86.38</td>
+<td>64.70</td>
+<td>93.29</td>
+<td>60.35</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec is a new generation text recognition model. It is designed to efficiently and accurately support the recognition of Simplified Chinese, Traditional Chinese, English, Japanese, as well as complex text scenarios such as handwriting, vertical text, pinyin, and rare characters with a single model. While maintaining recognition performance, it also balances inference speed and model robustness, providing efficient and accurate technical support for document understanding in various scenarios.</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>81.29</td>
+<td>66.00</td>
+<td>83.55</td>
+<td>54.65</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+</table>
+
+* <b>Chinese Recognition Models</b>
+<table>
+<tr>
+<th>Model</th><th>Model Download Links</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">Pretrained Model</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc is trained on a mixed dataset of more Chinese document data and PP-OCR training data, building upon PP-OCRv4_server_rec. It enhances the recognition capabilities for some Traditional Chinese characters, Japanese characters, and special symbols, supporting over 15,000 characters. In addition to improving document-related text recognition, it also enhances general text recognition capabilities.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>A lightweight recognition model of PP-OCRv4 with high inference efficiency, suitable for deployment on various hardware devices, including edge devices.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>The server-side model of PP-OCRv4, offering high inference accuracy and deployable on various servers.</td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>72.96</td>
+<td>3.89 / 1.16</td>
+<td>8.72 / 3.56</td>
+<td>10.3</td>
+<td>A lightweight recognition model of PP-OCRv3 with high inference efficiency, suitable for deployment on various hardware devices, including edge devices.</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Links</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+<tr>
+<td>ch_SVTRv2_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_SVTRv2_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_SVTRv2_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>68.81</td>
+<td>10.38 / 8.31</td>
+<td>66.52 / 30.83</td>
+<td>80.5</td>
+<td rowspan="1">SVTRv2 is a server-side text recognition model developed by the OpenOCR team of the Vision and Learning Lab (FVL) at Fudan University. It won the first prize in the PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition Task, with a 6% improvement in end-to-end recognition accuracy on Leaderboard A compared to PP-OCRv4.</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Links</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+<tr>
+<td>ch_RepSVTR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_RepSVTR_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_RepSVTR_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>65.07</td>
+<td>6.29 / 1.57</td>
+<td>20.64 / 5.40</td>
+<td>22.1</td>
+<td rowspan="1">RepSVTR is a mobile-side text recognition model based on SVTRv2. It won the first prize in the PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition Task, with a 2.5% improvement in end-to-end recognition accuracy on Leaderboard B compared to PP-OCRv4, while maintaining similar inference speed.</td>
+</tr>
+</table>
+
+* <b>English Recognition Models</b>
+<table>
+<tr>
+<th>Model</th><th>Model Download Links</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+<tr>
+<td>en_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv5_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td> 85.25</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>The ultra-lightweight English text recognition model, trained based on the PP-OCRv5 framework, further improves the accuracy of English text recognition, optimizes the omission of spaces, and enhances the recognition performance on handwritten English text.</td>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td> 70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>An ultra-lightweight English recognition model trained based on the PP-OCRv4 recognition model, supporting English and numeric character recognition.</td>
+</tr>
+<tr>
+<td>en_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>70.69</td>
+<td>3.56 / 0.78</td>
+<td>8.44 / 5.78</td>
+<td>17.3</td>
+<td>An ultra-lightweight English recognition model trained based on the PP-OCRv3 recognition model, supporting English and numeric character recognition.</td>
+</tr>
+</table>
+
+* <b>Multilingual Recognition Models</b>
+<table>
+<tr>
+<th>Model</th><th>Model Download Links</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+<tr>
+<td>korean_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv5_mobile_rec_pretrained.pdparams">Pre-trained Model</a></td>
+<td>88.0</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>14</td>
+<td>An ultra-lightweight Korean text recognition model trained based on the PP-OCRv5 recognition framework. Supports Korean, English and numeric text recognition.</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv5_mobile_rec_pretrained.pdparams">Pre-trained Model</a></td>
+<td>84.7</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>14</td>
+<td>A Latin-script text recognition model trained based on the PP-OCRv5 recognition framework. Supports most Latin alphabet languages and numeric text recognition.</td>
+</tr>
+<tr>
+<td>eslav_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+eslav_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/eslav_PP-OCRv5_mobile_rec_pretrained.pdparams">Pre-trained Model</a></td>
+<td>81.6</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>14</td>
+<td>An East Slavic language recognition model trained based on the PP-OCRv5 recognition framework. Supports East Slavic languages, English and numeric text recognition.</td>
+</tr>
+<tr>
+<td>th_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+th_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/th_PP-OCRv5_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>82.68</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>The Thai recognition model trained based on the PP-OCRv5 recognition model supports recognition of Thai, English, and numbers.</td>
+</tr>
+<tr>
+<td>el_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+el_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/el_PP-OCRv5_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>89.28</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>The Greek recognition model trained based on the PP-OCRv5 recognition model supports recognition of Greek, English, and numbers.</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/arabic_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>81.27</td>
+<td>-</td>
+<td>-</td>
+<td>7.6</td>
+<td>Ultra-lightweight Arabic character recognition model trained based on the PP-OCRv5 recognition model, supports Arabic letters and number recognition</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/cyrillic_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>80.27</td>
+<td>-</td>
+<td>-</td>
+<td>7.7</td>
+<td>Ultra-lightweight Cyrillic character recognition model trained based on the PP-OCRv5 recognition model, supports Cyrillic letters and number recognition</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/devanagari_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>84.96</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>Ultra-lightweight Devanagari script recognition model trained based on the PP-OCRv5 recognition model, supports Hindi, Sanskrit and other Devanagari letters, as well as number recognition</td>
+</tr>
+<tr>
+<td>te_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/te_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>87.65</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>Ultra-lightweight Telugu script recognition model trained based on the PP-OCRv5 recognition model, supports Telugu script and number recognition</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ta_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>94.2</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>Ultra-lightweight Tamil script recognition model trained based on the PP-OCRv5 recognition model, supports Tamil script and number recognition</td>
+</tr>
+<tr>
+<td>korean_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>60.21</td>
+<td>3.73 / 0.98</td>
+<td>8.76 / 2.91</td>
+<td>9.6</td>
+<td>An ultra-lightweight Korean recognition model trained based on the PP-OCRv3 recognition model, supporting Korean and numeric character recognition.</td>
+</tr>
+<tr>
+<td>japan_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+japan_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/japan_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>45.69</td>
+<td>3.86 / 1.01</td>
+<td>8.62 / 2.92</td>
+<td>9.8</td>
+<td>An ultra-lightweight Japanese recognition model trained based on the PP-OCRv3 recognition model, supporting Japanese and numeric character recognition.</td>
+</tr>
+<tr>
+<td>chinese_cht_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+chinese_cht_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/chinese_cht_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>82.06</td>
+<td>3.90 / 1.16</td>
+<td>9.24 / 3.18</td>
+<td>10.8</td>
+<td>An ultra-lightweight Traditional Chinese recognition model trained based on the PP-OCRv3 recognition model, supporting Traditional Chinese and numeric character recognition.</td>
+</tr>
+<tr>
+<td>te_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+te_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>95.88</td>
+<td>3.59 / 0.81</td>
+<td>8.28 / 6.21</td>
+<td>8.7</td>
+<td>An ultra-lightweight Telugu recognition model trained based on the PP-OCRv3 recognition model, supporting Telugu and numeric character recognition.</td>
+</tr>
+<tr>
+<td>ka_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ka_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ka_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>96.96</td>
+<td>3.49 / 0.89</td>
+<td>8.63 / 2.77</td>
+<td>17.4</td>
+<td>An ultra-lightweight Kannada recognition model trained based on the PP-OCRv3 recognition model, supporting Kannada and numeric character recognition.</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ta_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>76.83</td>
+<td>3.49 / 0.86</td>
+<td>8.35 / 3.41</td>
+<td>8.7</td>
+<td>An ultra-lightweight Tamil recognition model trained based on the PP-OCRv3 recognition model, supporting Tamil and numeric character recognition.</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>76.93</td>
+<td>3.53 / 0.78</td>
+<td>8.50 / 6.83</td>
+<td>8.7</td>
+<td>An ultra-lightweight Latin recognition model trained based on the PP-OCRv3 recognition model, supporting Latin and numeric character recognition.</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+arabic_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>73.55</td>
+<td>3.60 / 0.83</td>
+<td>8.44 / 4.69</td>
+<td>17.3</td>
+<td>An ultra-lightweight Arabic alphabet recognition model trained based on the PP-OCRv3 recognition model, supporting Arabic alphabet and numeric character recognition.</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+cyrillic_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>94.28</td>
+<td>3.56 / 0.79</td>
+<td>8.22 / 2.76</td>
+<td>8.7</td>
+<td>An ultra-lightweight Cyrillic alphabet recognition model trained based on the PP-OCRv3 recognition model, supporting Cyrillic alphabet and numeric character recognition.</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+devanagari_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>96.44</td>
+<td>3.60 / 0.78</td>
+<td>6.95 / 2.87</td>
+<td>8.7</td>
+<td>An ultra-lightweight Devanagari alphabet recognition model trained based on the PP-OCRv3 recognition model, supporting Devanagari alphabet and numeric character recognition.</td>
+</tr>
+</table>
+
+<strong>Test Environment Description:</strong>
+
+  <ul>
+      <li><b>Performance Test Environment</b>
+          <ul>
+              <li><strong>Test Dataset:</strong>
+                 <ul>
+                    <li>
+                    Chinese Recognition Models: A self-built Chinese dataset by PaddleOCR, covering street views, online images, documents, handwriting, with 11,000 images for text recognition.
+                    </li>
+                    <li>
+                      ch_SVTRv2_rec: <a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition Task</a> Leaderboard A evaluation set.
+                    </li>
+                    <li>
+                      ch_RepSVTR_rec: <a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition Task</a> Leaderboard B evaluation set.
+                    </li>
+                    <li>
+                      English Recognition Models: A self-built English dataset by PaddleOCR.
+                    </li>
+                    <li>
+                      Multilingual Recognition Models: A self-built multilingual dataset by PaddleOCR.
+                    </li>
+                 </ul>
+              </li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>Explanation of Inference Modes</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration</th>
+            <th>CPU Configuration</th>
+            <th>Acceleration Technology Combination</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Normal Mode</td>
+            <td>FP32 Precision / No TRT Acceleration</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>High-Performance Mode</td>
+            <td>Optimal combination of precision type and acceleration strategy</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Selection of the optimal backend (Paddle/OpenVINO/TRT, etc.)</td>
+        </tr>
+    </tbody>
+</table>
+
+</details>
+
+## 3. Quick Start
+
+> ❗ Before starting, please install the PaddleOCR wheel package. For details, please refer to the [Installation Guide](../installation.en.md).
+
+You can quickly experience it with one command:
+
+```bash
+paddleocr text_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_rec_001.png
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr text_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_rec_001.png \
+    --engine transformers
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following command:
+
+```bash
+# Use the onnxruntime engine for inference
+paddleocr text_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_rec_001.png \
+    --engine onnxruntime
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+<b>Note:</b> The official PaddleOCR models are downloaded from HuggingFace by default. If you cannot access HuggingFace, you can change the model source to BOS by setting the environment variable `PADDLE_PDX_MODEL_SOURCE="BOS"`. More mainstream model sources will be supported in the future.
+
+You can also integrate the model inference of the text recognition module into your project. Before running the following code, please download the [sample image](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_rec_001.png) to your local machine.
+
+```python
+from paddleocr import TextRecognition
+model = TextRecognition()
+output = model.predict(input="general_ocr_rec_001.png", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+The example above uses the <code>paddle_static</code> inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following code:
+
+```python
+from paddleocr import TextRecognition
+model = TextRecognition(engine="transformers")
+output = model.predict(input="general_ocr_rec_001.png", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following code:
+
+```python
+from paddleocr import TextRecognition
+model = TextRecognition(engine="onnxruntime")
+output = model.predict(input="general_ocr_rec_001.png", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+If you want to use the trained model with the `paddle_dynamic` or `transformers` engine, refer to the [Weight Conversion](#52-weight-conversion) section in the [Inference Engine](#5-inference-engine) section below to convert the model from the `pdparams` format to the `safetensors` format using PaddleX.
+
+After running, the result is as follows:
+
+```bash
+{'res': {'input_path': 'general_ocr_rec_001.png', 'page_index': None, 'rec_text': '绿洲仕格维花园公寓', 'rec_score': 0.9823867082595825}}
+```
+
+The meanings of the parameters in the result are as follows:
+- `input_path`: The path of the input text line image to be predicted
+- `page_index`: If the input is a PDF file, it indicates which page of the PDF the current text line is from; otherwise, it is `None`
+- `rec_text`: The predicted text of the text line image
+- `rec_score`: The confidence score of the predicted text for the text line image
+
+The visualized image is as follows:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/text_recog/general_ocr_rec_001.png"/>
+
+Descriptions of related methods and parameters are as follows:
+
+* Instantiate the text recognition model using <code>TextRecognition</code>, as follows:
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>Description:</b> If set to <code>None</code>, <code>PP-OCRv6_medium_rec</code> is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>Meaning:</b>Model storage path.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b> Device for inference.<br/>
+<b>Description:</b>
+<b>Examples:</b> <code>"cpu"</code>, <code>"gpu"</code>, <code>"npu"</code>, <code>"gpu:0"</code>, <code>"gpu:0,1"</code>.<br/>
+If multiple devices are specified, inference will be performed in parallel.<br/>
+By default, GPU 0 is used; if unavailable, CPU is used.
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, <code>transformers</code>, and <code>onnxruntime</code>. When left as <code>None</code>, local inference uses the <code>paddle_static</code> engine by default. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high performance inference.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to enable the TensorRT subgraph engine of Paddle Inference.<br/>
+<b>Description:</b>
+For Paddle with CUDA 11.8, the compatible TensorRT version is 8.x (x>=6), recommended 8.6.1.6.<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b>Precision for TensorRT when using the Paddle Inference TensorRT subgraph engine.<br/>
+<b>Description:</b>
+<b>Options:</b> <code>"fp32"</code>, <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b> Whether to enable MKL-DNN acceleration for inference. <br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support it, acceleration will not be used even if this flag is set.</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td><b>Meaning:</b> MKL-DNN cache capacity.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b>Number of threads to use for inference on CPUs.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>input_shape</code></td>
+<td><b>Meaning:</b>Input image size for the model in the format <code>(C, H, W)</code>.</td>
+<td><code>tuple|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+* Call the <code>predict()</code> method of the text recognition model for inference. This method returns a list of results. In addition, this module also provides the <code>predict_iter()</code>  method. The two methods are completely consistent in terms of parameter acceptance and result return. The difference is that <code>predict_iter()</code>  returns a <code>generator</code>, which can process and obtain prediction results step by step. It is suitable for scenarios where large datasets need to be processed or memory savings are desired. You can choose either of these two methods according to your actual needs. The parameters of the <code>predict()</code> method include <code>input</code> and <code>batch_size</code>, with specific descriptions as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted, supporting multiple input types, required.<br/>
+<b>Description:</b>
+<ul>
+<li><b>Python Var</b>: Image data represented by <code>numpy.ndarray</code></li>
+<li><b>str</b>: Local path of image file or PDF file: <code>/root/data/img.jpg</code>; <b>URL link</b>: Network URL of image file or PDF file: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_rec_001.png">Example</a>; <b>Local directory</b>: The directory should contain the images to be predicted, such as <code>/root/data/</code> (currently, prediction of PDF files in the directory is not supported, PDF files need to be specified to a specific file path)</li>
+<li><b>list</b>: The elements of the list should be data of the above types, such as <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td>Batch size, can be set to any positive integer.</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+
+* Process the prediction results. The prediction result for each sample is a corresponding Result object, which supports operations such as printing, saving as an image, and saving as a <code>json</code> file:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Description</th>
+<th>Parameter</th>
+<th>Type</th>
+<th>Description</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">Print the result to the terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output content using <code>JSON</code> indentation</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specifies the indentation level to beautify the output <code>JSON</code> data, making it more readable. Only effective when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Controls whether to escape non-<code>ASCII</code> characters as <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> retains the original characters. Only effective when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">Save the result as a file in <code>json</code> format</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The file path to save the result. When it is a directory, the saved file name is consistent with the naming of the input file type.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specifies the indentation level to beautify the output <code>JSON</code> data, making it more readable. Only effective when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Controls whether to escape non-<code>ASCII</code> characters as <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> retains the original characters. Only effective when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save the result as a file in image format</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The file path to save the result. When it is a directory, the saved file name is consistent with the naming of the input file type.</td>
+<td>None</td>
+</tr>
+</table>
+
+* In addition, it also supports obtaining the visualized image with results and the prediction results through attributes, as follows:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">Obtain the prediction result in <code>json</code> format</td>
+</tr>
+<tr>
+<td rowspan="1"><code>img</code></td>
+<td rowspan="1">Obtain the visualized image in <code>dict</code> format</td>
+</tr>
+</table>
+
+## 4. Secondary Development
+
+If the above models do not perform well in your scenario, you can try the following steps for secondary development. Here, we take training `PP-OCRv5_server_rec` as an example. For other models, just replace the corresponding configuration file. First, you need to prepare a dataset for text recognition. You can refer to the format of the [Text Recognition Demo Data](https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_rec_dataset_examples.tar) for preparation. After preparation, you can train and export the model as follows. After export, the model can be quickly integrated into the above API. This example uses the Text Recognition Demo Data. Before training the model, please make sure you have installed the dependencies required by PaddleOCR as described in the [Installation Guide](../installation.md).
+
+### 4.1 Dataset and Pre-trained Model Preparation
+
+#### 4.1.1 Prepare the Dataset
+
+```shell
+# Download the example dataset
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_rec_dataset_examples.tar
+tar -xf ocr_rec_dataset_examples.tar
+```
+
+#### 4.1.2 Download the Pre-trained Model
+
+```shell
+# Download the PP-OCRv5_server_rec pre-trained model
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams
+```
+
+### 4.2 Model Training
+
+PaddleOCR modularizes its code. To train the `PP-OCRv5_server_rec` recognition model, you need to use its [configuration file](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/configs/rec/PP-OCRv5/PP-OCRv5_server_rec.yml).
+
+The training commands are as follows:
+
+```bash
+# Single-GPU training (default training method)
+python3 tools/train.py -c configs/rec/PP-OCRv5/PP-OCRv5_server_rec.yml \
+   -o Global.pretrained_model=./PP-OCRv5_server_rec_pretrained.pdparams
+
+# Multi-GPU training, specify GPU IDs via the --gpus parameter
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/PP-OCRv5/PP-OCRv5_server_rec.yml \
+        -o Global.pretrained_model=./PP-OCRv5_server_rec_pretrained.pdparams
+```
+
+### 4.3 Model Evaluation
+
+You can evaluate the trained weights, such as `output/xxx/xxx.pdparams`, using the following command:
+
+```bash
+# Note: Set the path of pretrained_model to a local path. If you use a model you trained and saved yourself, please modify the path and file name to {path/to/weights}/{model_name}.
+# Demo test set evaluation
+python3 tools/eval.py -c configs/rec/PP-OCRv5/PP-OCRv5_server_rec.yml -o \
+Global.pretrained_model=output/xxx/xxx.pdparams
+```
+
+### 4.4 Model Export
+
+```bash
+python3 tools/export_model.py -c configs/rec/PP-OCRv5/PP-OCRv5_server_rec.yml -o \
+Global.pretrained_model=output/xxx/xxx.pdparams \
+Global.save_inference_dir="./PP-OCRv5_server_rec_infer/"
+```
+
+After exporting the model, the static graph model will be stored in `./PP-OCRv5_server_rec_infer/` in the current directory. In this directory, you will see the following files:
+```
+./PP-OCRv5_server_rec_infer/
+├── inference.json
+├── inference.pdiparams
+├── inference.yml
+```
+At this point, the secondary development is complete. This static graph model can be directly integrated into the PaddleOCR API.
+
+If you want to use the `paddle_dynamic` or `transformers` engine with the trained model, please refer to the [Weight Conversion](#52-weight-conversion) section in [Inference Engine](#5-inference-engine) later in this document to convert the model from the `pdparams` format to the `safetensors` format using PaddleX.
+
+## 5. Inference Engine
+
+For detailed descriptions, values, compatibility rules, and examples of the inference engine, please refer to <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration Description</a>.
+
+### 5.1 Speed Data
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">PP-OCRv5_mobile_rec</td>
+            <td>paddle_static</td>
+            <td>1.94</td>
+            <td>6.69</td>
+            <td>1.00</td>
+            <td>9.76</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>1.97</td>
+            <td>35.38</td>
+            <td>1.11</td>
+            <td>38.60</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>3.31</td>
+            <td>17.70</td>
+            <td>0.50</td>
+            <td>21.68</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>1.82</td>
+            <td>2.05</td>
+            <td>0.91</td>
+            <td>4.91</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-OCRv5_server_rec</td>
+            <td>paddle_static</td>
+            <td>1.98</td>
+            <td>11.37</td>
+            <td>1.21</td>
+            <td>14.69</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>1.98</td>
+            <td>23.89</td>
+            <td>1.32</td>
+            <td>27.34</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>3.99</td>
+            <td>11.69</td>
+            <td>0.51</td>
+            <td>16.36</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>1.80</td>
+            <td>3.15</td>
+            <td>0.90</td>
+            <td>5.98</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-OCRv6_medium_rec</td>
+            <td>paddle_static</td>
+            <td>1.74</td>
+            <td>5.38</td>
+            <td>0.84</td>
+            <td>8.08</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>1.76</td>
+            <td>13.38</td>
+            <td>0.85</td>
+            <td>16.10</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>2.58</td>
+            <td>7.04</td>
+            <td>0.41</td>
+            <td>10.19</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>1.74</td>
+            <td>2.28</td>
+            <td>0.84</td>
+            <td>4.97</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-OCRv6_small_rec</td>
+            <td>paddle_static</td>
+            <td>1.74</td>
+            <td>4.73</td>
+            <td>0.82</td>
+            <td>7.41</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>1.76</td>
+            <td>12.43</td>
+            <td>0.87</td>
+            <td>15.18</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>2.55</td>
+            <td>6.70</td>
+            <td>0.41</td>
+            <td>9.82</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>1.73</td>
+            <td>1.79</td>
+            <td>0.83</td>
+            <td>4.46</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-OCRv6_tiny_rec</td>
+            <td>paddle_static</td>
+            <td>1.76</td>
+            <td>2.77</td>
+            <td>0.40</td>
+            <td>5.04</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>1.75</td>
+            <td>6.96</td>
+            <td>0.36</td>
+            <td>9.19</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>2.45</td>
+            <td>3.12</td>
+            <td>0.40</td>
+            <td>6.12</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>1.73</td>
+            <td>0.92</td>
+            <td>0.36</td>
+            <td>3.12</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>Test Environment Description:</strong>
+<ul>
+    <li><strong>Test Data:</strong> <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_rec_001.jpg">Sample Image</a></li>
+    <li><strong>Hardware Configuration:</strong>
+        <ul>
+            <li>GPU: NVIDIA A100 40G</li>
+            <li>CPU: Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>Software Environment:</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 5.2 Weight Conversion
+
+When using the inference engine, the system will automatically download the official pre-trained model. If you need to use a self-trained model with the `paddle_dynamic` or `transformers` engine, please refer to the [PaddleX Text Image Orientation Classification Module Weight Conversion](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/text_recognition.html#442) section to convert the model from the `pdparams` format to the `safetensors` format using PaddleX. This allows seamless integration into the PaddleOCR API for inference. If you need to use a self-trained model with the `onnxruntime` engine, refer to [PaddleX Obtain ONNX Models](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html) to obtain the ONNX model, so it can be seamlessly integrated into the PaddleOCR API for inference.
+
+## 6. FAQ
diff --git a/docs/version3.x/module_usage/text_recognition.md b/docs/version3.x/module_usage/text_recognition.md
new file mode 100644
index 0000000..0a03e74
--- /dev/null
+++ b/docs/version3.x/module_usage/text_recognition.md
@@ -0,0 +1,1182 @@
+---
+comments: true
+---
+
+# 文本识别模块使用教程
+
+## 一、概述
+
+文本识别模块是OCR（光学字符识别）系统中的核心部分，负责从图像中的文本区域提取出文本信息。该模块的性能直接影响到整个OCR系统的准确性和效率。文本识别模块通常接收文本检测模块输出的文本区域的边界框（Bounding Boxes）作为输入，然后通过复杂的图像处理和深度学习算法，将图像中的文本转化为可编辑和可搜索的电子文本。文本识别结果的准确性，对于后续的信息提取和数据挖掘等应用至关重要。
+
+## 二、支持模型列表
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。表格中的“常规模式”耗时对应本地 <code>paddle_static</code> 推理引擎。
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv6_medium_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_medium_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_medium_rec_pretrained.pdparams">训练模型</a></td>
+<td>83.2*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>73.3</td>
+<td rowspan="3">PP-OCRv6 文本识别模型，基于 PPLCNetV4 + LightSVTR + CTC/NRTR 多头解码器，单模型支持 50 种语言（tiny 档 49 种）。medium 档相比 PP-OCRv5_server 综合识别精度提升 5.1%。</td>
+</tr>
+<tr>
+<td>PP-OCRv6_small_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_small_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.3*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>20.4</td>
+</tr>
+<tr>
+<td>PP-OCRv6_tiny_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_tiny_rec_pretrained.pdparams">训练模型</a></td>
+<td>73.5*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>4.4</td>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>86.38</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec 是新一代文本识别模型。该模型致力于以单一模型高效、精准地支持简体中文、繁体中文、英文、日文四种主要语言，以及手写、竖版、拼音、生僻字等复杂文本场景的识别。在保持识别效果的同时，兼顾推理速度和模型鲁棒性，为各种场景下的文档理解提供高效、精准的技术支撑。</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.29</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">训练模型</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc是在PP-OCRv4_server_rec的基础上，在更多中文文档数据和PP-OCR训练数据的混合数据训练而成，增加了部分繁体字、日文、特殊字符的识别能力，可支持识别的字符为1.5万+，除文档相关的文字识别能力提升外，也同时提升了通用文字的识别能力</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>PP-OCRv4的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>PP-OCRv4的服务器端模型，推理精度高，可以部署在多种不同的服务器上</td>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>基于PP-OCRv4识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+</table>
+
+> *注：PP-OCRv6 指标基于内部多场景评估集测得，PP-OCRv5/v4 指标基于通用评估集测得，两者评估集不同，指标不可直接对比。
+
+> ❗ 以上列出的是文本识别模块重点支持的<b>4个核心模型</b>，该模块总共支持<b>20个全量模型</b>，包含多个多语言文本识别模型，完整的模型列表如下：
+
+<details><summary> 👉模型列表详情</summary>
+
+* <b>PP-OCRv6 多场景模型</b>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</table>
+
+* <b>PP-OCRv5 多场景模型</b>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>中文识别 Avg Accuracy(%)</th>
+<th>英文识别 Avg Accuracy(%)</th>
+<th>繁体中文识别 Avg Accuracy(%)</th>
+<th>日文识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv6_medium_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_medium_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_medium_rec_pretrained.pdparams">训练模型</a></td>
+<td>83.2*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>73.3</td>
+<td rowspan="3">PP-OCRv6 文本识别模型，基于 PPLCNetV4 + LightSVTR + CTC/NRTR 多头解码器，单模型支持 50 种语言（tiny 档 49 种）。medium 档相比 PP-OCRv5_server 综合识别精度提升 5.1%。</td>
+</tr>
+<tr>
+<td>PP-OCRv6_small_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_small_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.3*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>20.4</td>
+</tr>
+<tr>
+<td>PP-OCRv6_tiny_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_tiny_rec_pretrained.pdparams">训练模型</a></td>
+<td>73.5*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>4.4</td>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>86.38</td>
+<td>64.70</td>
+<td>93.29</td>
+<td>60.35</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec 是新一代文本识别模型。该模型致力于以单一模型高效、精准地支持简体中文、繁体中文、英文、日文四种主要语言，以及手写、竖版、拼音、生僻字等复杂文本场景的识别。在保持识别效果的同时，兼顾推理速度和模型鲁棒性，为各种场景下的文档理解提供高效、精准的技术支撑。</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.29</td>
+<td>66.00</td>
+<td>83.55</td>
+<td>54.65</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+</table>
+
+* <b>中文识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">训练模型</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc是在PP-OCRv4_server_rec的基础上，在更多中文文档数据和PP-OCR训练数据的混合数据训练而成，增加了部分繁体字、日文、特殊字符的识别能力，可支持识别的字符为1.5万+，除文档相关的文字识别能力提升外，也同时提升了通用文字的识别能力</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>PP-OCRv4的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>PP-OCRv4的服务器端模型，推理精度高，可以部署在多种不同的服务器上</td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>72.96</td>
+<td>3.89 / 1.16</td>
+<td>8.72 / 3.56</td>
+<td>10.3</td>
+<td>PP-OCRv3的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>ch_SVTRv2_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_SVTRv2_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_SVTRv2_rec_pretrained.pdparams">训练模型</a></td>
+<td>68.81</td>
+<td>10.38 / 8.31</td>
+<td>66.52 / 30.83</td>
+<td>80.5</td>
+<td rowspan="1">
+SVTRv2 是一种由复旦大学视觉与学习实验室（FVL）的OpenOCR团队研发的服务端文本识别模型，其在PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务中荣获一等奖，A榜端到端识别精度相比PP-OCRv4提升6%。
+</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>ch_RepSVTR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_RepSVTR_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_RepSVTR_rec_pretrained.pdparams">训练模型</a></td>
+<td>65.07</td>
+<td>6.29 / 1.57</td>
+<td>20.64 / 5.40</td>
+<td>48.8</td>
+<td rowspan="1">    RepSVTR 文本识别模型是一种基于SVTRv2 的移动端文本识别模型，其在PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务中荣获一等奖，B榜端到端识别精度相比PP-OCRv4提升2.5%，推理速度持平。</td>
+</tr>
+</table>
+
+* <b>英文识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>en_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td> 85.25</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>基于PP-OCRv5识别模型训练得到的超轻量级英文识别模型，进一步提升英文文本的识别准确率，优化空格漏识别的问题，并提高对手写英文文本的识别效果。</td>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td> 70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>基于PP-OCRv4识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+<tr>
+<td>en_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>70.69</td>
+<td>3.56 / 0.78</td>
+<td>8.44 / 5.78</td>
+<td>17.3</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+</table>
+
+
+* <b>多语言识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>korean_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>88.0</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>14</td>
+<td>基于PP-OCRv5识别模型训练得到的超轻量韩文识别模型，支持韩文、英文和数字识别</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>84.7</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>14</td>
+<td>基于PP-OCRv5识别模型训练得到的拉丁文识别模型，支持大部分拉丁字母语言、数字识别</td>
+</tr>
+<tr>
+<td>eslav_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+eslav_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/eslav_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.6</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>14</td>
+<td>基于PP-OCRv5识别模型训练得到的东斯拉夫语言识别模型， 支持东斯拉夫语言、英文和数字识别</td>
+</tr>
+</tr>
+<tr>
+<td>th_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+th_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/th_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>82.68</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>基于PP-OCRv5识别模型训练得到的泰语识别模型， 支持泰语、英文和数字识别</td>
+</tr>
+<tr>
+<td>el_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+el_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/el_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>89.28</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>基于PP-OCRv5识别模型训练得到的希腊语识别模型， 支持希腊语、英文和数字识别</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+arabic_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.27</td>
+<td>-</td>
+<td>-</td>
+<td>7.6</td>
+<td>基于PP-OCRv5识别模型训练得到的超轻量阿拉伯字母识别模型，支持阿拉伯字母、数字识别</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+cyrillic_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>80.27</td>
+<td>-</td>
+<td>-</td>
+<td>7.7</td>
+<td>基于PP-OCRv5识别模型训练得到的超轻量斯拉夫字母识别模型，支持斯拉夫字母、数字识别</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+devanagari_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>84.96</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>基于PP-OCRv5识别模型训练得到的超轻量天城文识别模型，支持印地文、梵文等字母以及数字识别</td>
+</tr>
+<tr>
+<td>te_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+te_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>87.65</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>基于PP-OCRv5识别模型训练得到的超轻量泰卢固文识别模型，支持泰卢固文、数字识别</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ta_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>94.2</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>基于PP-OCRv5识别模型训练得到的超轻量泰米尔文识别模型，支持泰米尔文、数字识别</td>
+</tr>
+<tr>
+<td>korean_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>60.21</td>
+<td>3.73 / 0.98</td>
+<td>8.76 / 2.91</td>
+<td>9.6</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量韩文识别模型，支持韩文、数字识别</td>
+</tr>
+<tr>
+<td>japan_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+japan_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/japan_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>45.69</td>
+<td>3.86 / 1.01</td>
+<td>8.62 / 2.92</td>
+<td>9.8</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量日文识别模型，支持日文、数字识别</td>
+</tr>
+<tr>
+<td>chinese_cht_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+chinese_cht_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/chinese_cht_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>82.06</td>
+<td>3.90 / 1.16</td>
+<td>9.24 / 3.18</td>
+<td>10.8</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量繁体中文识别模型，支持繁体中文、数字识别</td>
+</tr>
+<tr>
+<td>te_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+te_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>95.88</td>
+<td>3.59 / 0.81</td>
+<td>8.28 / 6.21</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量泰卢固文识别模型，支持泰卢固文、数字识别</td>
+</tr>
+<tr>
+<td>ka_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ka_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ka_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>96.96</td>
+<td>3.49 / 0.89</td>
+<td>8.63 / 2.77</td>
+<td>17.4</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量卡纳达文识别模型，支持卡纳达文、数字识别</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ta_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>76.83</td>
+<td>3.49 / 0.86</td>
+<td>8.35 / 3.41</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量泰米尔文识别模型，支持泰米尔文、数字识别</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>76.93</td>
+<td>3.53 / 0.78</td>
+<td>8.50 / 6.83</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量拉丁文识别模型，支持拉丁文、数字识别</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+arabic_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>73.55</td>
+<td>3.60 / 0.83</td>
+<td>8.44 / 4.69</td>
+<td>17.3</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量阿拉伯字母识别模型，支持阿拉伯字母、数字识别</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+cyrillic_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>94.28</td>
+<td>3.56 / 0.79</td>
+<td>8.22 / 2.76</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量斯拉夫字母识别模型，支持斯拉夫字母、数字识别</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>96.44</td>
+<td>3.60 / 0.78</td>
+<td>6.95 / 2.87</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量梵文字母识别模型，支持梵文字母、数字识别</td>
+</tr>
+</table>
+
+<strong>测试环境说明:</strong>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+              <li><strong>测试数据集：</strong>
+                 <ul>
+                    <li>
+                    中文识别模型： PaddleOCR 自建的中文数据集，覆盖街景、网图、文档、手写多个场景，其中文本识别包含 1.1w 张图片。
+                    </li>
+                    <li>
+                      ch_SVTRv2_rec：<a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务</a>A榜评估集。
+                    </li>
+                    <li>
+                      ch_RepSVTR_rec：<a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务</a>B榜评估集。
+                    </li>
+                    <li>
+                      英文识别模型：PaddleOCR 自建的英文数据集。
+                    </li>
+                    <li>
+                      多语言识别模型：PaddleOCR 自建的多语种数据集。
+                    </li>
+                 </ul>
+              </li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+
+</details>
+
+## 三、快速开始
+
+> ❗ 在快速开始前，请先安装 PaddleOCR 的 wheel 包，详细请参考 [安装教程](../installation.md)。
+
+使用一行命令即可快速体验：
+
+```bash
+paddleocr text_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_rec_001.png
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下命令：
+
+```bash
+# 使用 transformers 引擎进行推理
+paddleocr text_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_rec_001.png \
+    --engine transformers
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下命令：
+
+```bash
+# 使用 onnxruntime 引擎进行推理
+paddleocr text_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_rec_001.png \
+    --engine onnxruntime
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+<b>注：</b>PaddleOCR 官方模型默认从 HuggingFace 获取，如运行环境访问 HuggingFace 不便，可通过环境变量修改模型源为 BOS：`PADDLE_PDX_MODEL_SOURCE="BOS"`，未来将支持更多主流模型源；
+
+您也可以将文本识别的模块中的模型推理集成到您的项目中。运行以下代码前，请您下载[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_rec_001.png)到本地。
+
+```python
+from paddleocr import TextRecognition
+model = TextRecognition()
+output = model.predict(input="general_ocr_rec_001.png", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下代码：
+
+```python
+from paddleocr import TextRecognition
+model = TextRecognition(engine="transformers")
+output = model.predict(input="general_ocr_rec_001.png", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下代码：
+
+```python
+from paddleocr import TextRecognition
+model = TextRecognition(engine="onnxruntime")
+output = model.predict(input="general_ocr_rec_001.png", batch_size=1)
+for res in output:
+    res.print()
+    res.save_to_img(save_path="./output/")
+    res.save_to_json(save_path="./output/res.json")
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+训练后的模型如果想使用 `paddle_dynamic` 或 `transformers` 引擎，请参考后文 [推理引擎](#五推理引擎) 中的 [权重转换](#52-权重转换) 部分将模型由 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式。
+
+运行后，得到的结果为：
+```bash
+{'res': {'input_path': 'general_ocr_rec_001.png', 'page_index': None, 'rec_text': '绿洲仕格维花园公寓', 'rec_score': 0.9823867082595825}}
+```
+
+运行结果参数含义如下：
+<ul>
+<li><code>input_path</code>：表示输入待预测文本行图像的路径</li>
+<li><code>page_index</code>：如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+<li><code>rec_text</code>：表示文本行图像的预测文本</li>
+<li><code>rec_score</code>：表示文本行图像的预测置信度</li>
+</ul>
+
+可视化图片如下：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/text_recog/general_ocr_rec_001.png"/>
+
+相关方法、参数等说明如下：
+
+* <code>TextRecognition</code>实例化文本识别模型，具体说明如下：
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>含义：</b>模型名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，则使用<code>PP-OCRv6_medium_rec</code>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>含义：</b>模型存储路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+<b>例如：</b><code>"cpu"</code>、<code>"gpu"</code>、<code>"npu"</code>、<code>"gpu:0"</code>、<code>"gpu:0,1"</code>。<br/>
+如指定多个设备，将进行并行推理。<br/>
+默认情况下，优先使用 GPU 0；若不可用则使用 CPU。
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>、<code>onnxruntime</code>。保持为默认值 <code>None</code> 时，本地推理默认使用 <code>paddle_static</code> 引擎。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>当使用 Paddle Inference 的 TensorRT 子图引擎时设置的计算精度。<br/>
+<b>说明：</b>
+<b>可选项：</b><code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+<b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上推理时使用的线程数量。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>input_shape</code></td>
+<td><b>含义：</b>模型输入图像尺寸，格式为 <code>(C, H, W)</code>。</td>
+<td><code>tuple|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+* 调用文本识别模型的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。另外，本模块还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。<code>predict()</code> 方法参数有 <code>input</code> 和 <code>batch_size</code>，具体说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_rec_001.png">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>含义：</b>批大小<br/>
+<b>说明：</b>
+可设置为任意正整数。</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+
+* 对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为<code>json</code>文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将结果保存为图像格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+</table>
+
+* 此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">获取预测的<code>json</code>格式的结果</td>
+</tr>
+<tr>
+<td rowspan="1"><code>img</code></td>
+<td rowspan="1">获取格式为<code>dict</code>的可视化图像</td>
+</tr>
+</table>
+
+
+## 四、二次开发
+
+如果以上模型在您的场景上效果仍然不理想，您可以尝试以下步骤进行二次开发，此处以训练 `PP-OCRv5_server_rec` 举例，其他模型替换对应配置文件即可。首先，您需要准备文本识别的数据集，可以参考[文本识别 Demo 数据](https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_rec_dataset_examples.tar)的格式准备，准备好后，即可按照以下步骤进行模型训练和导出，导出后，可以将模型快速集成到上述 API 中。此处以文本识别 Demo 数据示例。在训练模型之前，请确保已经按照[安装文档](../installation.md)安装了 PaddleOCR 所需要的依赖。
+
+
+## 4.1 数据集、预训练模型准备
+
+### 4.1.1 准备数据集
+
+```shell
+# 下载示例数据集
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/data/ocr_rec_dataset_examples.tar
+tar -xf ocr_rec_dataset_examples.tar
+```
+
+### 4.1.2 下载预训练模型
+
+```shell
+# 下载 PP-OCRv5_server_rec 预训练模型
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams 
+```
+
+### 4.2 模型训练
+
+PaddleOCR 对代码进行了模块化，训练 `PP-OCRv5_server_rec` 识别模型时需要使用 `PP-OCRv5_server_rec` 的[配置文件](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/configs/rec/PP-OCRv5/PP-OCRv5_server_rec.yml)。
+
+
+训练命令如下：
+
+```bash
+#单卡训练 (默认训练方式)
+python3 tools/train.py -c configs/rec/PP-OCRv5/PP-OCRv5_server_rec.yml \
+   -o Global.pretrained_model=./PP-OCRv5_server_rec_pretrained.pdparams
+
+#多卡训练，通过--gpus参数指定卡号
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py -c configs/rec/PP-OCRv5/PP-OCRv5_server_rec.yml \
+        -o Global.pretrained_model=./PP-OCRv5_server_rec_pretrained.pdparams
+```
+
+
+### 4.3 模型评估
+
+您可以评估已经训练好的权重，如，`output/xxx/xxx.pdparams`，使用如下命令进行评估：
+
+```bash
+#注意将pretrained_model的路径设置为本地路径。若使用自行训练保存的模型，请注意修改路径和文件名为{path/to/weights}/{model_name}。
+#demo 测试集评估
+python3 tools/eval.py -c configs/rec/PP-OCRv5/PP-OCRv5_server_rec.yml -o \
+Global.pretrained_model=output/xxx/xxx.pdparams
+```
+
+### 4.4 模型导出
+
+```bash
+python3 tools/export_model.py -c configs/rec/PP-OCRv5/PP-OCRv5_server_rec.yml -o \
+Global.pretrained_model=output/xxx/xxx.pdparams \
+Global.save_inference_dir="./PP-OCRv5_server_rec_infer/"
+```
+
+ 导出模型后，静态图模型会存放于当前目录的`./PP-OCRv5_server_rec_infer/`中，在该目录下，您将看到如下文件：
+ ```
+ ./PP-OCRv5_server_rec_infer/
+ ├── inference.json
+ ├── inference.pdiparams
+ ├── inference.yml
+ ```
+至此，二次开发完成，该静态图模型可以直接集成到 PaddleOCR 的 API 中。
+
+训练后的模型如果想使用 `paddle_dynamic` 或 `transformers` 引擎，请参考后文 [推理引擎](#五推理引擎) 中的 [权重转换](#52-权重转换) 部分将模型由 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式。
+
+## 五、推理引擎 {#五推理引擎}
+
+关于推理引擎的详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。
+
+### 5.1 速度数据
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">PP-OCRv5_mobile_rec</td>
+            <td>paddle_static</td>
+            <td>1.94</td>
+            <td>6.69</td>
+            <td>1.00</td>
+            <td>9.76</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>1.97</td>
+            <td>35.38</td>
+            <td>1.11</td>
+            <td>38.60</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>3.31</td>
+            <td>17.70</td>
+            <td>0.50</td>
+            <td>21.68</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>1.82</td>
+            <td>2.05</td>
+            <td>0.91</td>
+            <td>4.91</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-OCRv5_server_rec</td>
+            <td>paddle_static</td>
+            <td>1.98</td>
+            <td>11.37</td>
+            <td>1.21</td>
+            <td>14.69</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>1.98</td>
+            <td>23.89</td>
+            <td>1.32</td>
+            <td>27.34</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>3.99</td>
+            <td>11.69</td>
+            <td>0.51</td>
+            <td>16.36</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>1.80</td>
+            <td>3.15</td>
+            <td>0.90</td>
+            <td>5.98</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-OCRv6_medium_rec</td>
+            <td>paddle_static</td>
+            <td>1.74</td>
+            <td>5.38</td>
+            <td>0.84</td>
+            <td>8.08</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>1.76</td>
+            <td>13.38</td>
+            <td>0.85</td>
+            <td>16.10</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>2.58</td>
+            <td>7.04</td>
+            <td>0.41</td>
+            <td>10.19</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>1.74</td>
+            <td>2.28</td>
+            <td>0.84</td>
+            <td>4.97</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-OCRv6_small_rec</td>
+            <td>paddle_static</td>
+            <td>1.74</td>
+            <td>4.73</td>
+            <td>0.82</td>
+            <td>7.41</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>1.76</td>
+            <td>12.43</td>
+            <td>0.87</td>
+            <td>15.18</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>2.55</td>
+            <td>6.70</td>
+            <td>0.41</td>
+            <td>9.82</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>1.73</td>
+            <td>1.79</td>
+            <td>0.83</td>
+            <td>4.46</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-OCRv6_tiny_rec</td>
+            <td>paddle_static</td>
+            <td>1.76</td>
+            <td>2.77</td>
+            <td>0.40</td>
+            <td>5.04</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>1.75</td>
+            <td>6.96</td>
+            <td>0.36</td>
+            <td>9.19</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>2.45</td>
+            <td>3.12</td>
+            <td>0.40</td>
+            <td>6.12</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>1.73</td>
+            <td>0.92</td>
+            <td>0.36</td>
+            <td>3.12</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>测试环境说明:</strong>
+<ul>
+    <li><strong>测试数据：</strong><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_rec_001.jpg">示例图片</a></li>
+    <li><strong>硬件配置：</strong>
+        <ul>
+            <li>GPU：NVIDIA A100 40G</li>
+            <li>CPU：Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>软件环境：</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 5.2 权重转换 {#52-权重转换}
+
+使用推理引擎时，系统会自动下载官方预训练模型。若需使用自训练模型配合 `paddle_dynamic` 或 `transformers` 引擎，请参考 [PaddleX 文字识别模块权重转换](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/text_recognition.html#442) 部分，将 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式，即可无缝集成到 PaddleOCR 的 API 中进行推理。若需使用自训练模型配合`onnxruntime`引擎，请参考[PaddleX 获取 ONNX 模型](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html)获取onnx模型，即可无缝集成到 PaddleOCR 的 API 中进行推理。
+
+## 六、FAQ
diff --git a/docs/version3.x/module_usage/textline_orientation_classification.en.md b/docs/version3.x/module_usage/textline_orientation_classification.en.md
new file mode 100644
index 0000000..b5ace71
--- /dev/null
+++ b/docs/version3.x/module_usage/textline_orientation_classification.en.md
@@ -0,0 +1,559 @@
+---
+comments: true
+---
+
+# Text Line Orientation Classification Module Tutorial
+
+## 1. Overview
+The text line orientation classification module primarily distinguishes the orientation of text lines and corrects them using post-processing. In processes such as document scanning and license/certificate photography, to capture clearer images, the capture device may be rotated, resulting in text lines in various orientations. Standard OCR pipelines cannot handle such data well. By utilizing image classification technology, the orientation of text lines can be predetermined and adjusted, thereby enhancing the accuracy of OCR processing.
+
+## 2. Supported Model List
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing. The "Normal Mode" values correspond to the local <code>paddle_static</code> inference engine.
+
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Top-1 Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x0_25_textline_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x0_25_textline_ori_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x0_25_textline_ori_pretrained.pdparams">Training Model</a></td>
+<td>98.85</td>
+<td>2.16 / 0.41</td>
+<td>2.37 / 0.73</td>
+<td>0.96</td>
+<td>Text line classification model based on PP-LCNet_x0_25, with two classes: 0 degrees and 180 degrees</td>
+</tr>
+<tr>
+<td>PP-LCNet_x1_0_textline_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_textline_ori_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_textline_ori_pretrained.pdparams">Training Model</a></td>
+<td>99.42</td>
+<td>- / -</td>
+<td>2.98 / 2.98</td>
+<td>6.5</td>
+<td>Text line classification model based on PP-LCNet_x1_0, with two classes: 0 degrees and 180 degrees</td>
+</tr>
+</tbody>
+</table>
+
+> ❗ **Note**: The text line orientation classification model was upgraded on May 26, 2025, and `PP-LCNet_x1_0_textline_ori` has been added. If you need to use the pre-upgrade model weights, please click the <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x0_25_textline_ori_infer.bak.tar">download link</a>.
+
+<strong>Test Environment Description:</strong>
+
+  <ul>
+      <li><b>Performance Test Environment</b>
+          <ul>
+             <li><strong>Test Dataset：</strong> PaddleX Self-built Dataset, Covering Multiple Scenarios Such as Documents and Certificates, Containing 1000 Images.</li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>Inference Mode Description</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration </th>
+            <th>CPU Configuration </th>
+            <th>Acceleration Technology Combination</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Normal Mode</td>
+            <td>FP32 Precision / No TRT Acceleration</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>High-Performance Mode</td>
+            <td>Optimal combination of pre-selected precision types and acceleration strategies</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Pre-selected optimal backend (Paddle/OpenVINO/TRT, etc.)</td>
+        </tr>
+    </tbody>
+</table>
+
+## 3. Quick Integration
+
+> ❗ Before starting, please install the wheel package of PaddleOCR. For detailed instructions, refer to the [Installation Guide](../installation.en.md).
+
+You can quickly experience the functionality with a single command:
+
+```bash
+paddleocr textline_orientation_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/textline_rot180_demo.jpg
+```  
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr textline_orientation_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/textline_rot180_demo.jpg \
+    --engine transformers
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following command:
+
+```bash
+# Use the onnxruntime engine for inference
+paddleocr textline_orientation_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/textline_rot180_demo.jpg \
+    --engine onnxruntime
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+<b>Note: </b>The official models would be download from HuggingFace by default. If can't access to HuggingFace, please set the environment variable <code>PADDLE_PDX_MODEL_SOURCE="BOS"</code> to change the model source to BOS. In the future, more model sources will be supported.
+
+You can also integrate the text line orientation classification model into your project. Run the following code after downloading the [example image](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/textline_rot180_demo.jpg) to your local machine. 
+
+```python
+from paddleocr import TextLineOrientationClassification
+model = TextLineOrientationClassification(model_name="PP-LCNet_x0_25_textline_ori")
+output = model.predict("textline_rot180_demo.jpg",  batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_img("./output/demo.png")
+    res.save_to_json("./output/res.json")
+```
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured, and then run the following code:
+
+```python
+from paddleocr import TextLineOrientationClassification
+model = TextLineOrientationClassification(
+    model_name="PP-LCNet_x0_25_textline_ori",
+    engine="transformers",
+)
+output = model.predict("textline_rot180_demo.jpg",  batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_img("./output/demo.png")
+    res.save_to_json("./output/res.json")
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured, and then run the following code:
+
+```python
+from paddleocr import TextLineOrientationClassification
+model = TextLineOrientationClassification(
+    model_name="PP-LCNet_x0_25_textline_ori",
+    engine="onnxruntime",
+)
+output = model.predict("textline_rot180_demo.jpg",  batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_img("./output/demo.png")
+    res.save_to_json("./output/res.json")
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+If you want to use the trained model with the `paddle_dynamic` or `transformers` engine, refer to the [Weight Conversion](#52-weight-conversion) section in the [Inference Engine](#5-inference-engine) section below to convert the model from the `pdparams` format to the `safetensors` format using PaddleX.
+
+After running, the result obtained is:
+
+```bash
+{'res': {'input_path': 'textline_rot180_demo.jpg', 'page_index': None, 'class_ids': array([1], dtype=int32), 'scores': array([0.99864], dtype=float32), 'label_names': ['180_degree']}}
+```
+
+The meanings of the running results parameters are as follows:
+<ul>
+<li><code>input_path</code>：Indicates the path of the input image.</li>
+<li><code>page_index</code>：If the input is a PDF file, it indicates the current page number of the PDF; otherwise, it is <code>None</code></li>
+<li><code>class_ids</code>：Indicates the class ID of the prediction result.</li>
+<li><code>scores</code>：Indicates the confidence score of the prediction result.</li>
+<li><code>label_names</code>：Indicates the class name of the prediction result.
+The visualization image is as follows:</li>
+</ul>
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/textline_ori_classification/textline_rot180_demo_res.jpg">
+
+The explanations for the methods, parameters, etc., are as follows:
+
+* <code>TextLineOrientationClassification</code> instantiates a textline classification model (here, <code>PP-LCNet_x0_25_textline_ori</code> is used as an example), and the specific explanations are as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>Meaning:</b> Model name.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, <code>PP-LCNet_x0_25_textline_ori</code> will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>Meaning:</b>Model storage path.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b> Device for inference.<br/>
+<b>Description:</b> 
+<b>For example:</b> <code>"cpu"</code>, <code>"gpu"</code>, <code>"npu"</code>, <code>"gpu:0"</code>, <code>"gpu:0,1"</code>.<br/>
+If multiple devices are specified, parallel inference will be performed.<br/>
+By default, GPU 0 is used if available; otherwise, CPU is used.
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, <code>transformers</code>, and <code>onnxruntime</code>. When left as <code>None</code>, local inference uses the <code>paddle_static</code> engine by default. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to use the Paddle Inference TensorRT subgraph engine. <br/>
+<b>Description:</b> 
+If the model does not support acceleration through TensorRT, setting this flag will not enable acceleration.<br/>
+For Paddle with CUDA version 11.8, the compatible TensorRT version is 8.x (x>=6), and it is recommended to install TensorRT 8.6.1.6.<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b> Computation precision when using the TensorRT subgraph engine in Paddle Inference.<br/>
+<b>Description:</b> 
+<b>Options:</b> <code>"fp32"</code>, <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+<b>Meaning:</b> Whether to enable MKL-DNN acceleration for inference. <br/>
+<b>Description:</b> 
+If MKL-DNN is unavailable or the model does not support it, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b>MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads to use for inference on CPUs.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+</tbody>
+</table>
+
+* Use the <code>predict()</code>  method of the text line direction classification model to perform inference. This method returns a list of results. In addition, this module also provides the <code>predict_iter()</code> method. Both methods accept the same parameters and return the same result format. The difference is that <code>predict_iter()</code> returns a <code>generator</code>, which processes and retrieves prediction results step by step. It is suitable for handling large datasets or memory-efficient scenarios. You can choose either method based on your actual needs. The <code>predict()</code>  method accepts the parameters <code>input</code> and <code>batch_size</code>, which are described in detail below:
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Input data to be predicted. Required. <br/>
+<b>Description:</b> 
+Supports multiple input types:<ul>
+<li><b>Python Var</b>: e.g., <code>numpy.ndarray</code> representing image data</li>
+<li><b>str</b>: 
+    <ul>
+  <li>Local image or PDF file path: <code>/root/data/img.jpg</code>;</li>
+  <li><b>URL</b> of image or PDF file: e.g., <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_doc_preprocessor_002.png">example</a>;</li>
+  <li><b>Local directory</b>: directory containing images for prediction, e.g., <code>/root/data/</code> (Note: directories containing PDF files are not supported; PDFs must be specified by exact file path)</li>
+  </ul>
+  </li>
+<li><b>list</b>: Elements must be of the above types, e.g., <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>Meaning:</b>Batch size. <br/>
+<b>Description:</b> 
+Positive integer.</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+
+* Call the <code>predict()</code> method of the text line orientation classification model for inference. This method will return a list of results. In addition, this module also provides a <code>predict_iter()</code> method. Both methods accept the same parameters and return the same results, but <code>predict_iter()</code> returns a <code>generator</code>, which is more suitable for processing large datasets or when you want to save memory. You can choose either method according to your needs. The parameters of the <code>predict()</code> method are <code>input</code> and <code>batch_size</code>, as described below:
+
+
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Parameter Description</th>
+<th>Parameter Type</th>
+<th>Options</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted,<br/>
+<b>Description:</b>
+Supporting multiple input types</td>
+<td><code>Python Var|str|list</code></td>
+<td>
+<ul>
+  <li><b>Python variable</b>, such as image data represented by <code>numpy.ndarray</code></li>
+  <li><b>File path</b>, such as the local path of an image file: <code>/root/data/img.jpg</code></li>
+  <li><b>URL link</b>, such as the network URL of an image file: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/textline_rot180_demo.jpg">Example</a></li>
+  <li><b>Local directory</b>, the directory should contain data files to be predicted, such as the local path: <code>/root/data/</code></li>
+  <li><b>list</b>, the elements of the list should be of the above-mentioned data types, such as <code>[numpy.ndarray, numpy.ndarray]</code>, <code>[\"/root/data/img1.jpg\", \"/root/data/img2.jpg\"]</code>, <code>[\"/root/data1\", \"/root/data2\"]</code></li>
+</ul>
+</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>Meaning:</b>Batch size</td>
+<td><code>int</code></td>
+<td>Any integer</td>
+<td>1</td>
+</tr>
+</table>
+
+* The prediction results are processed, and the prediction result for each sample is of type <code>dict</code>. It supports operations such as printing, saving as an image, and saving as a <code>json</code> file:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Method Description</th>
+<th>Parameter</th>
+<th>Parameter Type</th>
+<th>Parameter Description</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">Print the results to the terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output content using <code>JSON</code> indentation</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data, making it more readable, only effective when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non-<code>ASCII</code> characters to <code>Unicode</code>. If set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> retains the original characters, only effective when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">Save the results as a JSON file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The path to save the file. If it is a directory, the saved file name will be consistent with the input file name</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data, making it more readable, only effective when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non-<code>ASCII</code> characters to <code>Unicode</code>. If set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> retains the original characters, only effective when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save the results as an image file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The path to save the file. If it is a directory, the saved file name will be consistent with the input file name</td>
+<td>None</td>
+</tr>
+</table>
+
+* Additionally, it supports obtaining the visualization image with results and the prediction results through attributes, as follows:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Attribute Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">Get the prediction result in <code>json</code> format</td>
+</tr>
+<tr>
+<td rowspan="1"><code>img</code></td>
+<td rowspan="1">Get the visualization image in <code>dict</code> format</td>
+</tr>
+</table>
+
+## 4. Custom Development  
+
+Since PaddleOCR does not natively support training for text line orientation classification, refer to [PaddleX's Custom Development Guide](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/textline_orientation_classification.html#iv-custom-development) for training. Trained models can seamlessly integrate into PaddleOCR's API for inference.
+
+If you want to use the `paddle_dynamic` or `transformers` engine with the trained model, please refer to the [Weight Conversion](#52-weight-conversion) section in [Inference Engine](#5-inference-engine) later in this document to convert the model from the `pdparams` format to the `safetensors` format using PaddleX.
+
+## 5. Inference Engine
+
+For detailed descriptions, values, compatibility rules, and examples of the inference engine, please refer to <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration Description</a>.
+
+### 5.1 Speed Data
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">PP-LCNet_x0_25_textline_ori</td>
+            <td>paddle_static</td>
+            <td>0.30</td>
+            <td>2.89</td>
+            <td>0.06</td>
+            <td>3.34</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>0.28</td>
+            <td>6.52</td>
+            <td>0.08</td>
+            <td>6.98</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>1.30</td>
+            <td>3.76</td>
+            <td>0.15</td>
+            <td>5.36</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>0.27</td>
+            <td>0.76</td>
+            <td>0.05</td>
+            <td>1.16</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-LCNet_x1_0_textline_ori</td>
+            <td>paddle_static</td>
+            <td>0.33</td>
+            <td>3.20</td>
+            <td>0.06</td>
+            <td>3.69</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>0.29</td>
+            <td>7.60</td>
+            <td>0.07</td>
+            <td>8.06</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>1.28</td>
+            <td>3.47</td>
+            <td>0.14</td>
+            <td>5.04</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>0.27</td>
+            <td>0.77</td>
+            <td>0.05</td>
+            <td>1.16</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>Test Environment Description:</strong>
+<ul>
+    <li><strong>Test Data:</strong> <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/textline_rot180_demo.jpg">Sample Image</a></li>
+    <li><strong>Hardware Configuration:</strong>
+        <ul>
+            <li>GPU: NVIDIA A100 40G</li>
+            <li>CPU: Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>Software Environment:</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 5.2 Weight Conversion
+
+When using the inference engine, the system will automatically download the official pre-trained model. If you need to use a self-trained model with the `paddle_dynamic` or `transformers` engine, please refer to the [PaddleX Text Line Orientation Classification Module Weight Conversion](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/textline_orientation_classification.html#442) section to convert the model from the `pdparams` format to the `safetensors` format using PaddleX. This allows seamless integration into the PaddleOCR API for inference. If you need to use a self-trained model with the `onnxruntime` engine, refer to [PaddleX Obtain ONNX Models](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html) to obtain the ONNX model, so it can be seamlessly integrated into the PaddleOCR API for inference.
diff --git a/docs/version3.x/module_usage/textline_orientation_classification.md b/docs/version3.x/module_usage/textline_orientation_classification.md
new file mode 100644
index 0000000..bc86483
--- /dev/null
+++ b/docs/version3.x/module_usage/textline_orientation_classification.md
@@ -0,0 +1,523 @@
+---
+comments: true
+---
+
+# 文本行方向分类模块使用教程
+
+## 一、概述
+
+文本行方向分类模块主要是将文本行的方向区分出来，并使用后处理将其矫正。在诸如文档扫描、证照拍摄等过程中，有时为了拍摄更清晰，会将拍摄设备进行旋转，导致得到的文本行也是不同方向的。此时，标准的OCR流程无法很好地应对这些数据。利用图像分类技术，可以预先判断文本行方向，并将其进行方向调整，从而提高OCR处理的准确性。
+
+## 二、支持模型列表
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。表格中的“常规模式”耗时对应本地 <code>paddle_static</code> 推理引擎。
+
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>模型下载链接</th>
+<th>Top-1 Acc（%）</th>
+<th>GPU推理耗时（ms）</th>
+<th>CPU推理耗时 (ms)</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x0_25_textline_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x0_25_textline_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x0_25_textline_ori_pretrained.pdparams">训练模型</a></td>
+<td>98.85</td>
+<td>2.16 / 0.41</td>
+<td>2.37 / 0.73</td>
+<td>0.96</td>
+<td>基于PP-LCNet_x0_25的文本行分类模型，含有两个类别，即0度，180度</td>
+</tr>
+<tr>
+<td>PP-LCNet_x1_0_textline_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_textline_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_textline_ori_pretrained.pdparams">训练模型</a></td>
+<td>99.42</td>
+<td>- / -</td>
+<td>2.98 / 2.98</td>
+<td>6.5</td>
+<td>基于PP-LCNet_x1_0的文本行分类模型，含有两个类别，即0度，180度</td>
+</tr>
+</tbody>
+</table>
+
+> ❗ <b>注</b>：文本行方向分类模型于 2025.5.26 升级，并增加 `PP-LCNet_x1_0_textline_ori`，如需使用升级前的模型权重，请点击<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x0_25_textline_ori_infer.bak.tar">下载链接</a>。
+
+<strong>测试环境说明:</strong>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+              <li><strong>测试数据集：</strong>PaddleOCR 自建的数据集，覆盖证件和文档等多个场景，包含 1000 张图片。</li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td>PaddleInference</td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+
+## 三、快速开始
+
+> ❗ 在快速开始前，请先安装 PaddleOCR 的 wheel 包，详细请参考 [安装教程](../installation.md)。
+
+使用一行命令即可快速体验：
+
+```bash
+paddleocr textline_orientation_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/textline_rot180_demo.jpg
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下命令：
+
+```bash
+# 使用 transformers 引擎进行推理
+paddleocr textline_orientation_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/textline_rot180_demo.jpg \
+    --engine transformers
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下命令：
+
+```bash
+# 使用 onnxruntime 引擎进行推理
+paddleocr textline_orientation_classification -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/textline_rot180_demo.jpg \
+    --engine onnxruntime
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+<b>注：</b>PaddleOCR 官方模型默认从 HuggingFace 获取，如运行环境访问 HuggingFace 不便，可通过环境变量修改模型源为 BOS：`PADDLE_PDX_MODEL_SOURCE="BOS"`，未来将支持更多主流模型源；
+
+您也可以将文本行方向分类模块中的模型推理集成到您的项目中。运行以下代码前，请您下载[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/textline_rot180_demo.jpg)到本地。
+
+```python
+from paddleocr import TextLineOrientationClassification
+model = TextLineOrientationClassification(model_name="PP-LCNet_x0_25_textline_ori")
+output = model.predict("textline_rot180_demo.jpg",  batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_img("./output/demo.png")
+    res.save_to_json("./output/res.json")
+```
+
+上述示例默认使用 <code>paddle_static</code> 推理引擎，请先按照[飞桨框架安装](../paddlepaddle_installation.md)完成 PaddlePaddle 安装。
+
+如果选择 `transformers` 作为推理引擎，请确保已配置 Transformers 环境，然后执行如下代码：
+
+```python
+from paddleocr import TextLineOrientationClassification
+model = TextLineOrientationClassification(
+    model_name="PP-LCNet_x0_25_textline_ori",
+    engine="transformers",
+)
+output = model.predict("textline_rot180_demo.jpg",  batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_img("./output/demo.png")
+    res.save_to_json("./output/res.json")
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请确保已配置 ONNX Runtime 环境，然后执行如下代码：
+
+```python
+from paddleocr import TextLineOrientationClassification
+model = TextLineOrientationClassification(
+    model_name="PP-LCNet_x0_25_textline_ori",
+    engine="onnxruntime",
+)
+output = model.predict("textline_rot180_demo.jpg",  batch_size=1)
+for res in output:
+    res.print(json_format=False)
+    res.save_to_img("./output/demo.png")
+    res.save_to_json("./output/res.json")
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+训练后的模型如果想使用 `paddle_dynamic` 或 `transformers` 引擎，请参考后文 [推理引擎](#五推理引擎) 中的 [权重转换](#52-权重转换) 部分将模型由 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式。
+
+运行后，得到的结果为：
+
+```bash
+{'res': {'input_path': 'textline_rot180_demo.jpg', 'page_index': None, 'class_ids': array([1], dtype=int32), 'scores': array([0.99864], dtype=float32), 'label_names': ['180_degree']}}
+```
+
+运行结果参数含义如下：
+<ul>
+<li><code>input_path</code>：表示输入图片的路径。</li>
+<li><code>page_index</code>：如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+<li><code>class_ids</code>：表示预测结果的类别 id，含有两个类别，即0度和180度。</li>
+<li><code>scores</code>：表示预测结果的置信度。</li>
+<li><code>label_names</code>：表示预测结果的类别名。</li>
+</ul>
+可视化图片如下：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/modules/textline_ori_classification/textline_rot180_demo_res.jpg">
+
+相关方法、参数等说明如下：
+
+* <code>TextLineOrientationClassification</code>实例化文本行方向分类模型（此处以<code>PP-LCNet_x0_25_textline_ori</code>为例），具体说明如下：
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model_name</code></td>
+<td><b>含义：</b>模型名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，则使用<code>PP-LCNet_x0_25_textline_ori</code>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>model_dir</code></td>
+<td><b>含义：</b>模型存储路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+<b>例如：</b><code>"cpu"</code>、<code>"gpu"</code>、<code>"npu"</code>、<code>"gpu:0"</code>、<code>"gpu:0,1"</code>。<br/>
+如指定多个设备，将进行并行推理。<br/>
+默认情况下，优先使用 GPU 0；若不可用则使用 CPU。
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>、<code>onnxruntime</code>。保持为默认值 <code>None</code> 时，本地推理默认使用 <code>paddle_static</code> 引擎。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>当使用 Paddle Inference 的 TensorRT 子图引擎时设置的计算精度。<br/>
+<b>说明：</b>
+<b>可选项：</b><code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td>
+<b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上推理时使用的线程数量。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+</tbody>
+</table>
+
+* 调用文本行方向分类模型的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。另外，本模块还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于  <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。<code>predict()</code> 方法参数有 <code>input</code> 和 <code>batch_size</code>，具体说明如下：
+
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_doc_preprocessor_002.png">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)</li>
+<li><b>List</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>batch_size</code></td>
+<td><b>含义：</b>批大小<br/>
+<b>说明：</b>
+可设置为任意正整数。</td>
+<td><code>int</code></td>
+<td>1</td>
+</tr>
+</table>
+
+* 对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为<code>json</code>文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将结果保存为图像格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+</table>
+
+* 此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">获取预测的<code>json</code>格式的结果</td>
+</tr>
+<tr>
+<td rowspan = "1"><code>img</code></td>
+<td rowspan = "1">获取格式为<code>dict</code>的可视化图像</td>
+</tr>
+
+</table>
+
+
+## 四、二次开发
+
+由于 PaddleOCR 并不直接提供文本行方向分类的训练，因此，如果需要训练文档图像方向分类模型，可以参考 [PaddleX 文本行方向分类二次开发](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/textline_orientation_classification.html#_5)部分进行训练。训练后的模型可以无缝集成到 PaddleOCR 的 API 中进行推理。
+
+训练后的模型如果想使用 `paddle_dynamic` 或 `transformers` 引擎，请参考后文 [推理引擎](#五推理引擎) 中的 [权重转换](#52-权重转换) 部分将模型由 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式。
+
+## 五、推理引擎 {#五推理引擎}
+
+关于推理引擎的详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。
+
+### 5.1 速度数据
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>model</th>
+            <th>engine</th>
+            <th>Preprocessing (ms)</th>
+            <th>Inference (ms)</th>
+            <th>PostProcessing (ms)</th>
+            <th>End-to-End (ms)</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td rowspan="4">PP-LCNet_x0_25_textline_ori</td>
+            <td>paddle_static</td>
+            <td>0.30</td>
+            <td>2.89</td>
+            <td>0.06</td>
+            <td>3.34</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>0.28</td>
+            <td>6.52</td>
+            <td>0.08</td>
+            <td>6.98</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>1.30</td>
+            <td>3.76</td>
+            <td>0.15</td>
+            <td>5.36</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>0.27</td>
+            <td>0.76</td>
+            <td>0.05</td>
+            <td>1.16</td>
+        </tr>
+        <tr>
+            <td rowspan="4">PP-LCNet_x1_0_textline_ori</td>
+            <td>paddle_static</td>
+            <td>0.33</td>
+            <td>3.20</td>
+            <td>0.06</td>
+            <td>3.69</td>
+        </tr>
+        <tr>
+            <td>paddle_dynamic</td>
+            <td>0.29</td>
+            <td>7.60</td>
+            <td>0.07</td>
+            <td>8.06</td>
+        </tr>
+        <tr>
+            <td>transformers</td>
+            <td>1.28</td>
+            <td>3.47</td>
+            <td>0.14</td>
+            <td>5.04</td>
+        </tr>
+        <tr>
+            <td>onnxruntime</td>
+            <td>0.27</td>
+            <td>0.77</td>
+            <td>0.05</td>
+            <td>1.16</td>
+        </tr>
+    </tbody>
+</table>
+
+<strong>测试环境说明:</strong>
+<ul>
+    <li><strong>测试数据：</strong><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/textline_rot180_demo.jpg">示例图片</a></li>
+    <li><strong>硬件配置：</strong>
+        <ul>
+            <li>GPU：NVIDIA A100 40G</li>
+            <li>CPU：Intel(R) Xeon(R) Gold 6248 CPU @ 2.50GHz</li>
+        </ul>
+    </li>
+    <li><strong>软件环境：</strong>
+        <ul>
+            <li>Ubuntu 22.04 / CUDA 12.6 / cuDNN 9.5</li>
+            <li>paddlepaddle-gpu 3.2.1 / paddleocr 3.5 / transformers 5.4.0 / torch 2.10 / onnxruntime-gpu 1.23.2</li>
+        </ul>
+    </li>
+</ul>
+
+### 5.2 权重转换 {#52-权重转换}
+
+使用推理引擎时，系统会自动下载官方预训练模型。若需使用自训练模型配合 `paddle_dynamic` 或 `transformers` 引擎，请参考 [PaddleX 文本行方向分类模块权重转换](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/textline_orientation_classification.html#442) 部分，将 `pdparams` 格式通过 PaddleX 转换为 `safetensors` 格式，即可无缝集成到 PaddleOCR 的 API 中进行推理。若需使用自训练模型配合`onnxruntime`引擎，请参考[PaddleX 获取 ONNX 模型](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/paddle2onnx.html)获取onnx模型，即可无缝集成到 PaddleOCR 的 API 中进行推理。
diff --git a/docs/version3.x/other_devices_support/multi_devices_use_guide.en.md b/docs/version3.x/other_devices_support/multi_devices_use_guide.en.md
new file mode 100644
index 0000000..5c229c6
--- /dev/null
+++ b/docs/version3.x/other_devices_support/multi_devices_use_guide.en.md
@@ -0,0 +1,132 @@
+---
+comments: true
+---
+
+# PaddleOCR Multi-Devices Usage Guide
+
+This document focuses on the usage guide of PaddleX for Huawei Ascend NPU and Kunlun XPU hardware platforms.
+This document mainly introduces installation and usage with PaddlePaddle. If you plan to use another inference engine, please follow that engine's official documentation for environment setup and installation.
+
+## 1、Installation
+
+### 1.1 PaddlePaddle Installation
+
+First, please complete the installation of PaddlePaddle according to your hardware platform. The installation tutorials for each hardware are as follows:
+
+Ascend NPU: [Ascend NPU PaddlePaddle Installation Guide](./paddlepaddle_install_NPU.en.md)
+
+Kunlun XPU: [Kunlun XPU PaddlePaddle Installation Guide](./paddlepaddle_install_XPU.en.md)
+
+### 1.2 PaddleOCR Installation
+
+Please refer to [PaddleOCR Installation Guide](../installation.en.md) to install PaddleOCR。
+
+## 2、Usage
+
+The methods for training and inference of PaddleOCR on hardware platforms such as Ascend NPU and Kunlun XPU are the same as those on GPU. You only need to modify the configuration parameters according to the specific hardware platform.
+On these two hardware platforms, quick inference and model fine-tuning are supported for the three major features of PaddleOCR, including the text recognition model PP-OCRv5, the document parsing solution PP-StructureV3, and PP-ChatOCRv4.
+
+### 2.1 Quick Inference
+
+You can quickly experience the OCR pipeline inference with a single command:
+
+* OCR pipeline inference
+
+```bash
+# The default model used is PP-OCRv5
+paddleocr ocr -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png --device npu:0
+```
+
+* PP-StructureV3 pipeline inference
+
+```bash
+paddleocr pp_structurev3 -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/pp_structure_v3_demo.png --device npu:0
+```
+
+To perform quick inference using the production line in your project, you can achieve this with just a few lines of code. Here’s an example of how you might set it up:
+
+* OCR pipeline inference
+
+```python
+from paddleocr import PaddleOCR
+
+ocr = PaddleOCR(device="npu:0")
+
+result = ocr.predict("./general_ocr_002.png")
+for res in result:
+    res.print()
+    res.save_to_img("output")
+    res.save_to_json("output")
+```
+
+* PP-StructureV3 pipeline inference
+
+```python
+from paddleocr import PPStructureV3
+
+pipeline = PPStructureV3(device="npu:0")
+output = pipeline.predict("./pp_structure_v3_demo.png")
+for res in output:
+    res.print() 
+    res.save_to_json(save_path="output") 
+    res.save_to_markdown(save_path="output")
+```
+
+If you want to know more about OCR pipeline inference,please refer to [general OCR pipeline use guide](../pipeline_usage/OCR.en.md).
+If you want to know more about PPStructureV3 pipeline inference,please refer to [PP-StructureV3 pipeline use guide](../pipeline_usage/PP-StructureV3.en.md).
+
+### 2.2 Model Fine-tuning
+
+If you are not satisfied with the performance of the pre-trained model, you can fine-tune it.
+
+* train on Ascend NPU
+```bash
+export FLAGS_npu_storage_format=0
+export FLAGS_npu_jit_compile=0
+export FLAGS_use_stride_kernel=0
+export FLAGS_allocator_strategy=auto_growth
+export FLAGS_npu_split_aclnn=True
+export FLAGS_npu_scale_aclnn=True
+export CUSTOM_DEVICE_BLACK_LIST=pad3d,pad3d_grad
+python3 -m paddle.distributed.launch --devices '0,1,2,3' \
+        tools/train.py -c configs/rec/PP-OCRv5/PP-OCRv5_mobile_rec.yml \
+        -o Global.use_gpu=False Global.use_npu=True
+```
+
+* train on Kunlun XPU
+```bash
+export FLAGS_use_stride_kernel=0
+export BKCL_FORCE_SYNC=1
+export BKCL_TIMEOUT=1800
+export XPU_BLACK_LIST=pad3d,pad3d_grad
+python3 -m paddle.distributed.launch --devices '0,1,2,3' \
+        tools/train.py -c configs/rec/PP-OCRv5/PP-OCRv5_mobile_rec.yml \
+        -o Global.use_gpu=False Global.use_xpu=True
+```
+
+### 2.3 Other Inference methods
+
+On the Ascend NPU, for a small number of inference samples, using the aforementioned pipeline inference method may result in abnormal outcomes (primarily with the PP-StructureV3 pipeline). To address this issue, we support using ONNX models for inference to ensure correct results.
+
+You can use the following command to convert a Paddle model to an ONNX model:
+
+```bash
+paddlex --install paddle2onnx
+paddlex --paddle2onnx --paddle_model_dir /paddle_model_dir --onnx_model_dir /onnx_model_dir --opset_version 7
+```
+
+Meanwhile, some models support Ascend offline OM inference, effectively optimizing inference performance and memory usage. Using models in OM + ONNX format for pipeline inference can ensure both accuracy and speed.
+
+Use the ATC conversion tool to convert an ONNX model to an OM model:
+
+```bash
+atc --model=inference.onnx --framework=5 --output=inference --soc_version="your_device_type" --input_shape "your_input_shape"
+```
+
+We have deeply integrated ONNX and OM models into PaddleX for high-performance inference. By modifying the pipeline configuration file to set the model inference backend to ONNX or OM, you can use the PaddleX high-performance inference API for inference.
+
+For specific modification methods, inference code, and more usage instructions, please refer to [Ascend NPU High-Performance Inference Tutorial](https://github.com/PaddlePaddle/PaddleX/blob/develop/docs/practical_tutorials/high_performance_npu_tutorial.en.md)。
+
+## 3、 FAQ
+### 1.The inference results using PP-StructureV3 on the production line are incorrect. 
+Some models on this line have precision errors in a small number of cases. You can try adjusting the model in the configuration file or use ONNX+OM models for inference.
diff --git a/docs/version3.x/other_devices_support/multi_devices_use_guide.md b/docs/version3.x/other_devices_support/multi_devices_use_guide.md
new file mode 100644
index 0000000..115f94c
--- /dev/null
+++ b/docs/version3.x/other_devices_support/multi_devices_use_guide.md
@@ -0,0 +1,129 @@
+---
+comments: true
+---
+
+# PaddleOCR多硬件使用指南
+
+本文档主要针对昇腾 NPU、昆仑 XPU 等硬件平台，介绍 PaddleOCR 使用指南。
+本文档主要介绍基于飞桨框架的安装与使用方式。若需使用其他推理引擎，请按照对应引擎的官方文档完成环境安装与配置。
+
+## 1、安装
+
+### 1.1 PaddlePaddle安装
+
+首先请您根据所属硬件平台，完成飞桨 PaddlePaddle 的安装，各硬件的飞桨安装教程如下：
+
+昇腾 NPU：[昇腾 NPU 飞桨安装教程](./paddlepaddle_install_NPU.md)
+
+昆仑 XPU：[昆仑 XPU 飞桨安装教程](./paddlepaddle_install_XPU.md)
+
+### 1.2 PaddleOCR安装
+
+请参考[PaddleOCR安装教程](../installation.md)安装 PaddleOCR。
+
+## 2、使用
+
+基于昇腾 NPU、昆仑 XPU 等硬件平台的 PaddleOCR 训练、推理使用方法与 GPU 相同，只需根据所属硬件平台，修改配置设备的参数。在这两款硬件上，支持 PaddleOCR 三大特色能力的快速推理和模型微调，包括 文字识别模型 PP-OCRv5、文档解析方案 PP-StructureV3 和 PP-ChatOCRv4。
+
+### 2.1 快速推理
+
+使用一行命令即可快速体验：
+
+* OCR 产线推理
+
+```bash
+# 默认使用 PP-OCRv5 模型
+paddleocr ocr -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png --device npu:0 # 将设备名修改为 npu或xpu
+```
+
+* PP-StructureV3 产线推理
+
+```bash
+paddleocr pp_structurev3 -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/pp_structure_v3_demo.png --device npu:0
+```
+
+在项目中，您可以通过几行代码即可完成产线的快速推理，推理代码如下：
+
+* OCR 产线推理
+```python
+from paddleocr import PaddleOCR
+
+ocr = PaddleOCR(device="npu:0")
+
+result = ocr.predict("./general_ocr_002.png")
+for res in result:
+    res.print()
+    res.save_to_img("output")
+    res.save_to_json("output")
+```
+
+* PP-StructureV3 产线推理
+```python
+from paddleocr import PPStructureV3
+
+pipeline = PPStructureV3(device="npu:0")
+output = pipeline.predict("./pp_structure_v3_demo.png")
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_json(save_path="output") ## 保存当前图像的结构化json结果
+    res.save_to_markdown(save_path="output") ## 保存当前图像的markdown格式的结果
+```
+
+更多关于 OCR 产线推理的使用说明，请参考[通用OCR产线使用教程](../pipeline_usage/OCR.md)。
+更多关于 PP-StructureV3 产线推理的使用说明，请参考[PP-StructureV3产线使用教程](../pipeline_usage/PP-StructureV3.md)。
+
+### 2.2 模型微调
+
+如果对预训练模型的效果不满意，可以对产线上的模型进行模型微调，示例如下：
+
+* 昇腾 NPU
+```bash
+export FLAGS_npu_storage_format=0
+export FLAGS_npu_jit_compile=0
+export FLAGS_use_stride_kernel=0
+export FLAGS_allocator_strategy=auto_growth
+export FLAGS_npu_split_aclnn=True
+export FLAGS_npu_scale_aclnn=True
+export CUSTOM_DEVICE_BLACK_LIST=pad3d,pad3d_grad
+python3 -m paddle.distributed.launch --devices '0,1,2,3' \
+        tools/train.py -c configs/rec/PP-OCRv5/PP-OCRv5_mobile_rec.yml \
+        -o Global.use_gpu=False Global.use_npu=True
+```
+
+* 昆仑 XPU
+```bash
+export FLAGS_use_stride_kernel=0
+export BKCL_FORCE_SYNC=1
+export BKCL_TIMEOUT=1800
+export XPU_BLACK_LIST=pad3d,pad3d_grad
+python3 -m paddle.distributed.launch --devices '0,1,2,3' \
+        tools/train.py -c configs/rec/PP-OCRv5/PP-OCRv5_mobile_rec.yml \
+        -o Global.use_gpu=False Global.use_xpu=True
+```
+
+### 2.3 其它方式推理
+
+在昇腾 NPU 上，对于少量推理样本，使用上述产线推理方式存在结果异常的可能（主要是 PP-StructureV3 产线），针对这种情况，我们支持使用 ONNX 模型进行推理，保证推理结果正确。
+
+使用如下命令可以将 Paddle 模型转换为 ONNX 模型：
+
+```bash
+paddlex --install paddle2onnx
+paddlex --paddle2onnx --paddle_model_dir /paddle_model_dir --onnx_model_dir /onnx_model_dir --opset_version 7
+```
+
+同时，部分模型支持昇腾离线 OM 推理，有效优化推理性能和内存占用。使用 OM + ONNX 格式的模型进行产线推理，能够同时保障精度和速度。
+
+使用 atc 转换工具将 ONNX 模型转换为 OM 模型：
+
+```bash
+atc --model=inference.onnx --framework=5 --output=inference --soc_version="your_device_type" --input_shape "your_input_shape"
+```
+
+我们已将 ONNX 和 OM 模型深度集成进了 PaddleX 高性能推理，修改产线配置文件，配置模型推理后端为 ONNX 或 OM 即可使用 PaddleX 高性能推理 API 进行推理。
+
+具体修改方法、推理代码以及更多的使用方法请参考[昇腾 NPU 高性能推理教程](https://github.com/PaddlePaddle/PaddleX/blob/develop/docs/practical_tutorials/high_performance_npu_tutorial.md)。
+
+## 3、常见问题
+### 1. 使用 PP-StructureV3 产线推理结果不正确
+该产线上的部分模型在少量 case 上存在精度误差，可以尝试调整配置文件中的模型，或者使用 ONNX+OM 模型进行推理。
diff --git a/docs/version3.x/other_devices_support/paddlepaddle_install_NPU.en.md b/docs/version3.x/other_devices_support/paddlepaddle_install_NPU.en.md
new file mode 100644
index 0000000..065a465
--- /dev/null
+++ b/docs/version3.x/other_devices_support/paddlepaddle_install_NPU.en.md
@@ -0,0 +1,57 @@
+---
+comments: true
+---
+
+# Ascend NPU PaddlePaddle Installation Tutorial
+
+Currently, PaddleOCR supports the Ascend 910B chip (more models are under support. If you have a related need for other models, please submit an issue to inform us). The Ascend driver version is 23.0.3. Considering the differences in environments, we recommend using the <b>Ascend development image provided by PaddlePaddle</b> to complete the environment preparation.
+This guide mainly introduces installation and usage with PaddlePaddle. If you want to use a different inference engine, please follow that engine's official installation and configuration documentation.
+
+## 1. Docker Environment Preparation
+* Pull the image. This image is only for the development environment and does not contain a pre-compiled PaddlePaddle installation package. The image has CANN-8.0.0, the Ascend operator library, installed by default.
+```bash
+# For X86 architecture
+docker pull ccr-2vdh3abv-pub.cnc.bj.baidubce.com/device/paddle-npu:cann800-ubuntu20-npu-910b-base-x86_64-gcc84
+# For Aarch64 architecture
+docker pull ccr-2vdh3abv-pub.cnc.bj.baidubce.com/device/paddle-npu:cann800-ubuntu20-npu-910b-base-aarch64-gcc84
+```
+* Start the container with the following command. ASCEND_RT_VISIBLE_DEVICES specifies the visible NPU card numbers.
+```bash
+docker run -it --name paddle-npu-dev -v $(pwd):/work \
+    --privileged --network=host --shm-size=128G -w=/work \
+    -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
+    -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
+    -v /usr/local/dcmi:/usr/local/dcmi \
+    -e ASCEND_RT_VISIBLE_DEVICES="0,1,2,3,4,5,6,7" \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/device/paddle-npu:cann800-ubuntu20-npu-910b-base-$(uname -m)-gcc84 /bin/bash
+```
+## 2. Install Paddle Package
+* Download and install the Python wheel installation package
+```bash
+# Note: You need to install the CPU version of PaddlePaddle first
+python -m pip install paddlepaddle==3.0.0.dev20250527 -i https://www.paddlepaddle.org.cn/packages/nightly/cpu
+python -m pip install paddle-custom-npu==3.0.0.dev20250527 -i https://www.paddlepaddle.org.cn/packages/nightly/npu
+```
+* CANN-8.0.0 does not support some versions of numpy and opencv, it is recommended to install the specified versions.
+```bash
+python -m pip install numpy==1.26.4
+python -m pip install opencv-python==3.4.18.65
+```
+* Set environment variables on the arm machine (not required for x86 environment)
+```bash
+# Solve the error reported by libgomp on the arm machine
+# "libgomp cannot allocate memory in static TLS block"
+export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libgomp.so.1:$LD_PRELOAD
+```
+* After verifying that the installation package is installed, run the following command
+```bash
+python -c "import paddle; paddle.utils.run_check()"
+```
+The expected output is as follows
+
+```
+Running verify PaddlePaddle program ...
+PaddlePaddle works well on 1 npu.
+PaddlePaddle works well on 8 npus.
+PaddlePaddle is installed successfully! Let's start deep learning with PaddlePaddle now.
+```
diff --git a/docs/version3.x/other_devices_support/paddlepaddle_install_NPU.md b/docs/version3.x/other_devices_support/paddlepaddle_install_NPU.md
new file mode 100644
index 0000000..339905e
--- /dev/null
+++ b/docs/version3.x/other_devices_support/paddlepaddle_install_NPU.md
@@ -0,0 +1,57 @@
+---
+comments: true
+---
+
+# 昇腾 NPU 飞桨安装教程
+
+当前 PaddleOCR 支持昇腾 910B 芯片（更多型号还在支持中，如果您有其他型号的相关需求，请提交issue告知我们），昇腾驱动版本为 23.0.3。考虑到环境差异性，我们推荐使用<b>飞桨官方提供的昇腾开发镜像</b>完成环境准备。
+本文档主要介绍基于飞桨框架的环境安装与使用方式。若您计划使用其他推理引擎，请参考对应引擎的官方文档完成安装与配置。
+
+## 1、docker环境准备
+* 拉取镜像，此镜像仅为开发环境，镜像中不包含预编译的飞桨安装包，镜像中已经默认安装了昇腾算子库 CANN-8.0.0。
+```bash
+# 适用于 X86 架构
+docker pull ccr-2vdh3abv-pub.cnc.bj.baidubce.com/device/paddle-npu:cann800-ubuntu20-npu-910b-base-x86_64-gcc84
+# 适用于 Aarch64 架构
+docker pull ccr-2vdh3abv-pub.cnc.bj.baidubce.com/device/paddle-npu:cann800-ubuntu20-npu-910b-base-aarch64-gcc84
+```
+* 参考如下命令启动容器，ASCEND_RT_VISIBLE_DEVICES 指定可见的 NPU 卡号
+```bash
+docker run -it --name paddle-npu-dev -v $(pwd):/work \
+    --privileged --network=host --shm-size=128G -w=/work \
+    -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
+    -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
+    -v /usr/local/dcmi:/usr/local/dcmi \
+    -e ASCEND_RT_VISIBLE_DEVICES="0,1,2,3,4,5,6,7" \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/device/paddle-npu:cann800-ubuntu20-npu-910b-base-$(uname -m)-gcc84 /bin/bash
+```
+## 2、安装paddle包
+* 下载安装 wheel 安装包
+```bash
+# 注意需要先安装飞桨 cpu 版本
+python -m pip install paddlepaddle==3.0.0.dev20250527 -i https://www.paddlepaddle.org.cn/packages/nightly/cpu
+python -m pip install paddle-custom-npu==3.0.0.dev20250527 -i https://www.paddlepaddle.org.cn/packages/nightly/npu
+```
+* CANN-8.0.0 对 numpy 和 opencv 部分版本不支持，建议安装指定版本
+```bash
+python -m pip install numpy==1.26.4
+python -m pip install opencv-python==3.4.18.65
+```
+* arm机器上需要设置环境变量（x86环境无需设置）
+```bash
+# 解决libgomp在arm机器上报错
+# "libgomp cannot allocate memory in static TLS block"
+export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libgomp.so.1:$LD_PRELOAD
+```
+* 验证安装包安装完成之后，运行如下命令
+```bash
+python -c "import paddle; paddle.utils.run_check()"
+```
+预期得到如下输出结果
+
+```
+Running verify PaddlePaddle program ...
+PaddlePaddle works well on 1 npu.
+PaddlePaddle works well on 8 npus.
+PaddlePaddle is installed successfully! Let's start deep learning with PaddlePaddle now.
+```
diff --git a/docs/version3.x/other_devices_support/paddlepaddle_install_XPU.en.md b/docs/version3.x/other_devices_support/paddlepaddle_install_XPU.en.md
new file mode 100644
index 0000000..5d7dd14
--- /dev/null
+++ b/docs/version3.x/other_devices_support/paddlepaddle_install_XPU.en.md
@@ -0,0 +1,46 @@
+---
+comments: true
+---
+
+# Kunlun XPU PaddlePaddle Installation Tutorial
+
+Currently, PaddleOCR supports Kunlun R200/R300 and other chips. Considering environmental differences, we recommend using the <b>Kunlun XPU development image officially released by PaddlePaddle</b>, which is pre-installed with the Kunlun basic runtime environment library (XRE).
+This guide mainly introduces installation and usage with PaddlePaddle. If you want to use a different inference engine, please follow that engine's official installation and configuration documentation.
+
+## 1. Docker Environment Preparation
+Pull the image. This image is only for the development environment and does not include a pre-compiled PaddlePaddle installation package.
+
+```bash
+docker pull registry.baidubce.com/device/paddle-xpu:ubuntu20-x86_64-gcc84-py310 # For X86 architecture
+docker pull registry.baidubce.com/device/paddle-xpu:kylinv10-aarch64-gcc82-py310 # For ARM architecture
+```
+Refer to the following command to start the container:
+
+```bash
+docker run -it --name=xxx -m 81920M --memory-swap=81920M \
+    --shm-size=128G --privileged --net=host \
+    -v $(pwd):/workspace -w /workspace \
+    registry.baidubce.com/device/paddle-xpu:$(uname -m)-py310 bash
+```
+
+## 2. Install Paddle Package
+Currently, Python3.10 wheel installation packages are provided. If you have a need for other Python versions, you can refer to the [PaddlePaddle official documentation](https://www.paddlepaddle.org.cn/en/install/quick) to compile and install them yourself.
+
+Install the Python3.10 wheel installation package:
+
+```bash
+pip install https://paddle-whl.bj.bcebos.com/paddlex/xpu/paddlepaddle_xpu-2.6.1-cp310-cp310-linux_x86_64.whl # For X86 architecture
+pip install https://paddle-whl.bj.bcebos.com/paddlex/xpu/paddlepaddle_xpu-2.6.1-cp310-cp310-linux_aarch64.whl # For ARM architecture
+```
+
+Verify the installation package. After installation, run the following command:
+
+```bash
+python -c "import paddle; paddle.utils.run_check()"
+```
+
+The expected output is:
+
+```
+PaddlePaddle is installed successfully! Let's start deep learning with PaddlePaddle now.
+```
diff --git a/docs/version3.x/other_devices_support/paddlepaddle_install_XPU.md b/docs/version3.x/other_devices_support/paddlepaddle_install_XPU.md
new file mode 100644
index 0000000..0c0ce60
--- /dev/null
+++ b/docs/version3.x/other_devices_support/paddlepaddle_install_XPU.md
@@ -0,0 +1,43 @@
+---
+comments: true
+---
+
+# 昆仑 XPU 飞桨安装教程
+
+当前 PaddleOCR 支持昆仑 R200/R300 等芯片。考虑到环境差异性，我们推荐使用<b>飞桨官方发布的昆仑 XPU 开发镜像</b>，该镜像预装有昆仑基础运行环境库（XRE）。
+本文档主要介绍基于飞桨框架的环境安装与使用方式。若您计划使用其他推理引擎，请参考对应引擎的官方文档完成安装与配置。
+
+## 1、docker环境准备
+拉取镜像，此镜像仅为开发环境，镜像中不包含预编译的飞桨安装包
+
+```
+docker pull registry.baidubce.com/device/paddle-xpu:ubuntu20-x86_64-gcc84-py310 # X86 架构
+docker pull registry.baidubce.com/device/paddle-xpu:kylinv10-aarch64-gcc82-py310 # ARM 架构
+```
+参考如下命令启动容器
+
+```
+docker run -it --name=xxx -m 81920M --memory-swap=81920M \
+    --shm-size=128G --privileged --net=host \
+    -v $(pwd):/workspace -w /workspace \
+    registry.baidubce.com/device/paddle-xpu:$(uname -m)-py310 bash
+```
+## 2、安装paddle包
+当前提供 Python3.10 的 wheel 安装包。如有其他 Python 版本需求，可以参考[飞桨官方文档](https://www.paddlepaddle.org.cn/install/quick)自行编译安装。
+
+安装 Python3.10 的 wheel 安装包
+
+```
+pip install https://paddle-whl.bj.bcebos.com/paddlex/xpu/paddlepaddle_xpu-2.6.1-cp310-cp310-linux_x86_64.whl # X86 架构
+pip install https://paddle-whl.bj.bcebos.com/paddlex/xpu/paddlepaddle_xpu-2.6.1-cp310-cp310-linux_aarch64.whl # ARM 架构
+```
+验证安装包 安装完成之后，运行如下命令
+
+```
+python -c "import paddle; paddle.utils.run_check()"
+```
+预期得到如下输出结果
+
+```
+PaddlePaddle is installed successfully! Let's start deep learning with PaddlePaddle now.
+```
diff --git a/docs/version3.x/paddleocr_and_paddlex.en.md b/docs/version3.x/paddleocr_and_paddlex.en.md
new file mode 100644
index 0000000..6e10260
--- /dev/null
+++ b/docs/version3.x/paddleocr_and_paddlex.en.md
@@ -0,0 +1,102 @@
+---
+comments: true
+---
+
+# PaddleOCR and PaddleX
+
+[PaddleX](https://github.com/PaddlePaddle/PaddleX) is a low-code development tool built on the PaddlePaddle framework. It integrates numerous out-of-the-box pre-trained models, supports the full-pipeline development from model training to inference, and is compatible with various mainstream hardware both domestically and internationally, empowering AI developers to efficiently deploy solutions in industrial practices.
+
+PaddleOCR leverages PaddleX for inference deployment, enabling seamless collaboration between the two in this regard. When installing PaddleOCR, PaddleX is also installed as a dependency. Additionally, PaddleOCR and PaddleX maintain consistency in pipeline naming conventions. For quick experience, users typically do not need to understand the specific concepts of PaddleX when using basic configurations. However, knowledge of PaddleX can be beneficial in advanced configuration scenarios, service deployment, and other use cases.
+
+This document introduces the relationship between PaddleOCR and PaddleX and explains how to use these two tools collaboratively.
+
+## 1. Differences and Connections Between PaddleOCR and PaddleX
+
+PaddleOCR and PaddleX have distinct focuses and functionalities: PaddleOCR specializes in OCR-related tasks, while PaddleX covers a wide range of task types, including time-series forecasting, face recognition, and more. Furthermore, PaddleX provides rich infrastructure with underlying capabilities for multi-model combined inference, enabling the integration of different models in a unified and flexible manner and supporting the construction of complex model pipelines.
+
+PaddleOCR fully reuses the capabilities of PaddleX in the inference deployment phase, including:
+
+- PaddleOCR primarily relies on PaddleX for underlying capabilities such as model inference, pre- and post-processing, and multi-model combination.
+- The high-performance inference capabilities of PaddleOCR are achieved through PaddleX's Paddle2ONNX plugin and high-performance inference plugins.
+- The service deployment solutions of PaddleOCR are based on PaddleX's implementations.
+
+It is important to note that although PaddleOCR uses PaddleX at the underlying level, thanks to PaddleX’s optional dependency installation feature, **installing the `paddleocr` Python distribution package does not include all of PaddleX’s dependencies—only those required for OCR-related tasks are installed**. Therefore, users generally do not need to worry about excessive expansion of dependency size. Tested in May 2025, in an x86-64 + Linux + Python 3.10 environment, the total size of required dependencies increased only from 717 MB to 738 MB.
+
+The version correspondence between PaddleOCR, PaddleX, and the PaddlePaddle framework is as follows:
+
+| PaddleOCR Version | PaddleX Version | PaddlePaddle Version |
+| --- | --- | --- |
+| `3.0.0` | `3.0.0` | `>= 3.0.0` |
+| `3.0.1` | `3.0.1` | `>= 3.0.0` |
+| `3.0.2` | `3.0.2` | `>= 3.0.0` |
+| `3.0.3` | `>= 3.0.3` | `>= 3.0.0` |
+| `3.1.x` | `>= 3.1.0, < 3.2.0` | `>= 3.0.0` |
+| `3.2.x` | `>= 3.2.0, < 3.3.0` | `>= 3.0.0` |
+| `3.3.x` | `>= 3.3.0, < 3.4.0` | `>= 3.0.0` |
+| `3.4.x` | `>= 3.4.0, < 3.5.0` | `>= 3.0.0` |
+| `3.5.x` | `>= 3.5.0, < 3.6.0` | `>= 3.0.0` |
+| `3.6.x` | `>= 3.6.0, < 3.7.0` | `>= 3.0.0` |
+| `3.7.x` | `>= 3.7.0, < 3.8.0` | `>= 3.0.0` |
+
+## 2. Correspondence Between PaddleOCR Pipelines and PaddleX Pipeline Registration Names
+
+| PaddleOCR Pipeline | PaddleX Pipeline Registration Name |
+| --- | --- |
+| General OCR | `OCR` |
+| PP-StructureV3 | `PP-StructureV3` |
+| PP-ChatOCRv4 | `PP-ChatOCRv4-doc` |
+| General Table Recognition V2 | `table_recognition_v2` |
+| Formula Recognition | `formula_recognition` |
+| Seal Text Recognition | `seal_recognition` |
+| Document Image Preprocessing | `doc_preprocessor` |
+| Document Understanding | `doc_understanding` |
+| PP-DocTranslation | `PP-DocTranslation` |
+| PaddleOCR-VL | `PaddleOCR-VL` |
+| PaddleOCR-VL-1.5 | `PaddleOCR-VL-1.5` |
+| PaddleOCR-VL-1.6 | `PaddleOCR-VL-1.6` |
+
+## 3. Using PaddleX Pipeline Configuration Files
+
+During the inference deployment phase, PaddleOCR supports exporting and loading PaddleX pipeline configuration files. Users can deeply configure inference deployment-related parameters by editing these configuration files.
+
+### 3.1 Exporting Pipeline Configuration Files
+
+You can call the `export_paddlex_config_to_yaml` method of the PaddleOCR pipeline object to export the current pipeline configuration to a YAML file. Here is an example:
+
+```python
+from paddleocr import PaddleOCR
+
+pipeline = PaddleOCR()
+pipeline.export_paddlex_config_to_yaml("ocr_config.yaml")
+```
+
+The above code will generate a pipeline configuration file named `ocr_config.yaml` in the working directory.
+
+You can also obtain the configuration file through the PaddleX CLI. Example:
+
+```bash
+# Specify the pipeline registration name
+paddlex --get_pipeline_config OCR
+```
+
+### 3.2 Editing Pipeline Configuration Files
+
+The exported PaddleX pipeline configuration file not only includes parameters supported by PaddleOCR's CLI and Python API but also allows for more advanced configurations. Please refer to the corresponding pipeline usage tutorials in [PaddleX Pipeline Usage Overview](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/pipeline_develop_guide.html) for detailed instructions on adjusting various configurations according to your needs.
+
+### 3.3 Loading Pipeline Configuration Files in CLI
+
+By specifying the path to the PaddleX pipeline configuration file using the `--paddlex_config` parameter, PaddleOCR will read its contents as the default configuration for the pipeline (this takes precedence over the default values of individual initialization parameters). Here is an example:
+
+```bash
+paddleocr ocr --paddlex_config ocr_config.yaml ...
+```
+
+### 3.4 Loading Pipeline Configuration Files in Python API
+
+When initializing the pipeline object, you can pass the path to the PaddleX pipeline configuration file or a configuration dictionary through the `paddlex_config` parameter, and PaddleOCR will use it as the default configuration (this takes precedence over the default values of individual initialization parameters). Here is an example:
+
+```python
+from paddleocr import PaddleOCR
+
+pipeline = PaddleOCR(paddlex_config="ocr_config.yaml")
+```
diff --git a/docs/version3.x/paddleocr_and_paddlex.md b/docs/version3.x/paddleocr_and_paddlex.md
new file mode 100644
index 0000000..b4dcd37
--- /dev/null
+++ b/docs/version3.x/paddleocr_and_paddlex.md
@@ -0,0 +1,102 @@
+---
+comments: true
+---
+
+# PaddleOCR 与 PaddleX
+
+[PaddleX](https://github.com/PaddlePaddle/PaddleX) 是一款基于飞桨框架构建的低代码开发工具，集成了众多开箱即用的预训练模型，支持模型从训练到推理的全流程开发，兼容多款国内外主流硬件，助力 AI 开发者在产业实践中高效落地。
+
+PaddleOCR 在推理部署方面基于 PaddleX 构建，二者在该环节可实现无缝协同。在安装 PaddleOCR 时，PaddleX 也将作为其依赖一并安装。此外，PaddleOCR 与 PaddleX 在产线名称等方面也保持一致。对于快速体验，如果只使用基础配置，用户通常无需了解 PaddleX 的具体概念；但在涉及高级配置、服务化部署等使用场景时，了解 PaddleX 的相关知识将有所帮助。
+
+本文档将介绍 PaddleOCR 与 PaddleX 之间的关系，并说明如何协同使用这两个工具。
+
+## 1. PaddleOCR 与 PaddleX 的区别与联系
+
+PaddleOCR 与 PaddleX 在定位和功能上各有侧重：PaddleOCR 专注于 OCR 相关任务，而 PaddleX 则覆盖了包括时序预测、人脸识别等在内的多种任务类型。此外，PaddleX 提供了丰富的基础设施，具备多模型组合推理的底层能力，能够以统一且灵活的方式接入不同模型，支持构建复杂的模型产线。
+
+PaddleOCR 在推理部署环节充分复用了 PaddleX 的能力，具体包括：
+
+- PaddleOCR 在模型推理、前后处理及多模型组合等底层能力上，主要依赖于 PaddleX。
+- PaddleOCR 的高性能推理能力通过 PaddleX 的 Paddle2ONNX 插件及高性能推理插件实现。
+- PaddleOCR 的服务化部署方案基于 PaddleX 的实现。
+
+需要特别说明的是，尽管 PaddleOCR 在底层使用了 PaddleX，但得益于 PaddleX 的可选依赖安装功能，**安装 `paddleocr` Python 分发包时并不会安装 PaddleX 的全部依赖，而只会安装 OCR 类任务需要使用到的依赖**，用户通常无需关心依赖体积的过度膨胀问题。2025 年 5 月测试，在 x86-64 + Linux + Python 3.10 环境中，需要安装的依赖总体积仅仅从 717 MB 增加到 738 MB。
+
+PaddleOCR、PaddleX 和飞桨框架的版本存在如下对应关系：
+
+| PaddleOCR 版本 | PaddleX 版本 | 飞桨版本 |
+| --- | --- | --- |
+| `3.0.0` | `3.0.0` | `>= 3.0.0` |
+| `3.0.1` | `3.0.1` | `>= 3.0.0` |
+| `3.0.2` | `3.0.2` | `>= 3.0.0` |
+| `3.0.3` | `>= 3.0.3` | `>= 3.0.0` |
+| `3.1.x` | `>= 3.1.0, < 3.2.0` | `>= 3.0.0` |
+| `3.2.x` | `>= 3.2.0, < 3.3.0` | `>= 3.0.0` |
+| `3.3.x` | `>= 3.3.0, < 3.4.0` | `>= 3.0.0` |
+| `3.4.x` | `>= 3.4.0, < 3.5.0` | `>= 3.0.0` |
+| `3.5.x` | `>= 3.5.0, < 3.6.0` | `>= 3.0.0` |
+| `3.6.x` | `>= 3.6.0, < 3.7.0` | `>= 3.0.0` |
+| `3.7.x` | `>= 3.7.0, < 3.8.0` | `>= 3.0.0` |
+
+## 2. PaddleOCR 产线与 PaddleX 产线注册名的对应关系
+
+| PaddleOCR 产线 | PaddleX 产线注册名 |
+| --- | --- |
+| 通用 OCR | `OCR` |
+| PP-StructureV3 | `PP-StructureV3` |
+| PP-ChatOCRv4 | `PP-ChatOCRv4-doc` |
+| 通用表格识别 v2 | `table_recognition_v2` |
+| 公式识别 | `formula_recognition` |
+| 印章文本识别 | `seal_recognition` |
+| 文档图像预处理 | `doc_preprocessor` |
+| 文档理解 | `doc_understanding` |
+| PP-DocTranslation | `PP-DocTranslation` |
+| PaddleOCR-VL | `PaddleOCR-VL` |
+| PaddleOCR-VL-1.5 | `PaddleOCR-VL-1.5` |
+| PaddleOCR-VL-1.6 | `PaddleOCR-VL-1.6` |
+
+## 3. 使用 PaddleX 产线配置文件
+
+在推理部署阶段，PaddleOCR 支持导出和加载 PaddleX 的产线配置文件。用户可通过编辑配置文件，对推理部署相关参数进行深度配置。
+
+### 3.1 导出产线配置文件
+
+可调用 PaddleOCR 产线对象的 `export_paddlex_config_to_yaml` 方法，将当前产线配置导出为 YAML 文件。示例如下：
+
+```python
+from paddleocr import PaddleOCR
+
+pipeline = PaddleOCR()
+pipeline.export_paddlex_config_to_yaml("ocr_config.yaml")
+```
+
+上述代码会在工作目录下生成名为 `ocr_config.yaml` 的产线配置文件。
+
+也可以通过 PaddleX CLI 获取配置文件。示例如下：
+
+```bash
+# 指定产线注册名
+paddlex --get_pipeline_config OCR
+```
+
+### 3.2 编辑产线配置文件
+
+导出的 PaddleX 产线配置文件不仅包含 PaddleOCR CLI 和 Python API 支持的参数，还可进行更多高级配置。请在 [PaddleX模型产线使用概览](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/pipeline_develop_guide.html) 中找到对应的产线使用教程，参考其中的详细说明，根据需求调整各项配置。
+
+### 3.3 在 CLI 中加载产线配置文件
+
+通过 `--paddlex_config` 参数指定 PaddleX 产线配置文件的路径，PaddleOCR 会读取其中的内容作为产线的默认配置（优先级高于各参数默认初始化的值）。示例如下：
+
+```bash
+paddleocr ocr --paddlex_config ocr_config.yaml ...
+```
+
+### 3.4 在 Python API 中加载产线配置文件
+
+初始化产线对象时，可通过 `paddlex_config` 参数传入 PaddleX 产线配置文件路径或配置字典，PaddleOCR 会将其作为默认配置（优先级高于各参数默认初始化的值）。示例如下：
+
+```python
+from paddleocr import PaddleOCR
+
+pipeline = PaddleOCR(paddlex_config="ocr_config.yaml")
+```
diff --git a/docs/version3.x/paddlepaddle_installation.en.md b/docs/version3.x/paddlepaddle_installation.en.md
new file mode 100644
index 0000000..fa9936d
--- /dev/null
+++ b/docs/version3.x/paddlepaddle_installation.en.md
@@ -0,0 +1,97 @@
+---
+comments: true
+---
+
+# PaddlePaddle Framework Installation
+
+This document explains how to install PaddlePaddle. The following scenarios usually require the PaddlePaddle framework to be installed first:
+
+- using the PaddlePaddle framework as the inference engine for pipeline/model inference;
+- performing development tasks such as model training and export.
+
+## 1. Install PaddlePaddle with Docker
+
+<b>If you install via Docker</b>, use the following commands with the official PaddlePaddle Docker images to create a container named `paddleocr` and mount the current working directory to `/paddle` inside the container.
+
+If your Docker version is >= 19.03, run:
+
+```bash
+# For CPU users:
+docker run --name paddleocr -v $PWD:/paddle --shm-size=8G --network=host -it ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.0.0 /bin/bash
+
+# For GPU users:
+# GPU version, requires driver version >= 450.80.02 (Linux) or >= 452.39 (Windows)
+docker run --gpus all --name paddleocr -v $PWD:/paddle --shm-size=8G --network=host -it ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.0.0-gpu-cuda11.8-cudnn8.9-trt8.6 /bin/bash
+
+# GPU version, requires driver version >= 550.54.14 (Linux) or >= 550.54.14 (Windows)
+docker run --gpus all --name paddleocr -v $PWD:/paddle --shm-size=8G --network=host -it ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.0.0-gpu-cuda12.6-cudnn9.5-trt10.5 /bin/bash
+```
+
+If your Docker version is <= 19.03 but >= 17.06, run:
+
+<details><summary>Click to expand</summary>
+
+<pre><code class="language-bash"># For CPU users:
+docker run --name paddleocr -v $PWD:/paddle --shm-size=8G --network=host -it ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.0.0 /bin/bash
+
+# For GPU users:
+# CUDA 11.8 users
+nvidia-docker run --name paddleocr -v $PWD:/paddle --shm-size=8G --network=host -it ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.0.0-gpu-cuda11.8-cudnn8.9-trt8.6 /bin/bash
+
+# CUDA 12.6 users
+nvidia-docker run --name paddleocr -v $PWD:/paddle --shm-size=8G --network=host -it ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.0.0-gpu-cuda12.6-cudnn9.5-trt10.5 /bin/bash
+</code></pre>
+</details>
+
+If your Docker version is <= 17.06, please upgrade Docker first.
+
+For more official PaddlePaddle Docker images, see the [PaddlePaddle website](https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/en/install/docker/linux-docker.html).
+
+## 2. Install PaddlePaddle with pip
+
+<b>If you install via pip</b>, use the following commands to install PaddlePaddle in the current environment:
+
+```bash
+# CPU version
+python -m pip install paddlepaddle==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+
+# GPU version, requires driver version >= 450.80.02 (Linux) or >= 452.39 (Windows)
+python -m pip install paddlepaddle-gpu==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cu118/
+
+# GPU version, requires driver version >= 550.54.14 (Linux) or >= 550.54.14 (Windows)
+python -m pip install paddlepaddle-gpu==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cu126/
+```
+
+> ❗ <b>Note</b>: You do not need to pay attention to the physical machine's CUDA version. You only need to care about the GPU driver version. For more PaddlePaddle wheel versions, see the [PaddlePaddle website](https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/en/install/pip/linux-pip.html).
+
+After installation, use the following command to verify whether PaddlePaddle is installed successfully:
+
+```bash
+python -c "import paddle; print(paddle.__version__)"
+```
+
+If the installation succeeds, it will output a version number like:
+
+```bash
+3.2.0
+```
+
+## 3. Install PaddlePaddle wheel packages for NVIDIA 50-series GPUs on Windows
+
+PaddlePaddle installed using the methods above does not properly support NVIDIA 50-series GPUs on Windows. Therefore, we provide specially adapted PaddlePaddle packages for this hardware environment. Please choose the corresponding wheel file according to your Python version.
+
+```bash
+# python 3.9
+python -m pip install https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-TagBuild-Training-Windows-Gpu-Cuda12.9-Cudnn9.9-Trt10.5-Mkl-Avx-VS2019-SelfBuiltPypiUse/86d658f56ebf3a5a7b2b33ace48f22d10680d311/paddlepaddle_gpu-3.0.0.dev20250717-cp39-cp39-win_amd64.whl
+
+# python 3.10
+python -m pip install https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-TagBuild-Training-Windows-Gpu-Cuda12.9-Cudnn9.9-Trt10.5-Mkl-Avx-VS2019-SelfBuiltPypiUse/86d658f56ebf3a5a7b2b33ace48f22d10680d311/paddlepaddle_gpu-3.0.0.dev20250717-cp310-cp310-win_amd64.whl
+
+# python 3.11
+python -m pip install https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-TagBuild-Training-Windows-Gpu-Cuda12.9-Cudnn9.9-Trt10.5-Mkl-Avx-VS2019-SelfBuiltPypiUse/86d658f56ebf3a5a7b2b33ace48f22d10680d311/paddlepaddle_gpu-3.0.0.dev20250717-cp311-cp311-win_amd64.whl
+
+# python 3.12
+python -m pip install https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-TagBuild-Training-Windows-Gpu-Cuda12.9-Cudnn9.9-Trt10.5-Mkl-Avx-VS2019-SelfBuiltPypiUse/86d658f56ebf3a5a7b2b33ace48f22d10680d311/paddlepaddle_gpu-3.0.0.dev20250717-cp312-cp312-win_amd64.whl
+```
+
+The currently released PaddlePaddle wheel packages for Windows 50-series GPUs still have known issues in text-recognition model training, and related support is still being adapted and improved.
diff --git a/docs/version3.x/paddlepaddle_installation.md b/docs/version3.x/paddlepaddle_installation.md
new file mode 100644
index 0000000..31e3fb5
--- /dev/null
+++ b/docs/version3.x/paddlepaddle_installation.md
@@ -0,0 +1,97 @@
+---
+comments: true
+---
+
+# 飞桨框架安装
+
+本文档说明如何安装 PaddlePaddle。以下场景通常需要先安装飞桨框架：
+
+- 在进行产线/模型推理时，使用飞桨框架作为推理引擎；
+- 进行模型训练、导出等开发工作。
+
+## 1. 基于 Docker 安装飞桨
+
+<b>若您通过 Docker 安装</b>，请参考下述命令，使用飞桨框架官方 Docker 镜像，创建一个名为 `paddleocr` 的容器，并将当前工作目录映射到容器内的 `/paddle` 目录。
+
+若您使用的 Docker 版本 >= 19.03，请执行：
+
+```bash
+# 对于 CPU 用户:
+docker run --name paddleocr -v $PWD:/paddle --shm-size=8G --network=host -it ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.0.0 /bin/bash
+
+# 对于 GPU 用户:
+# GPU 版本，需显卡驱动程序版本 ≥450.80.02（Linux）或 ≥452.39（Windows）
+docker run --gpus all --name paddleocr -v $PWD:/paddle --shm-size=8G --network=host -it ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.0.0-gpu-cuda11.8-cudnn8.9-trt8.6 /bin/bash
+
+# GPU 版本，需显卡驱动程序版本 ≥550.54.14（Linux）或 ≥550.54.14（Windows）
+docker run --gpus all --name paddleocr -v $PWD:/paddle --shm-size=8G --network=host -it ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.0.0-gpu-cuda12.6-cudnn9.5-trt10.5 /bin/bash
+```
+
+若您使用的 Docker 版本 <= 19.03 但 >= 17.06，请执行：
+
+<details><summary>点击展开</summary>
+
+<pre><code class="language-bash"># 对于 CPU 用户:
+docker run --name paddleocr -v $PWD:/paddle --shm-size=8G --network=host -it ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.0.0 /bin/bash
+
+# 对于 GPU 用户:
+# CUDA11.8 用户
+nvidia-docker run --name paddleocr -v $PWD:/paddle --shm-size=8G --network=host -it ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.0.0-gpu-cuda11.8-cudnn8.9-trt8.6 /bin/bash
+
+# CUDA12.6 用户
+nvidia-docker run --name paddleocr -v $PWD:/paddle --shm-size=8G --network=host -it ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.0.0-gpu-cuda12.6-cudnn9.5-trt10.5 /bin/bash
+</code></pre>
+</details>
+
+若您使用的 Docker 版本 <= 17.06，请升级 Docker 版本。
+
+更多飞桨官方 Docker 镜像请参考[飞桨官网](https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/install/docker/linux-docker.html)。
+
+## 2. 基于 pip 安装飞桨
+
+<b>若您通过 pip 安装</b>，请参考下述命令，用 pip 在当前环境中安装 PaddlePaddle：
+
+```bash
+# CPU 版本
+python -m pip install paddlepaddle==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+
+# GPU 版本，需显卡驱动程序版本 ≥450.80.02（Linux）或 ≥452.39（Windows）
+python -m pip install paddlepaddle-gpu==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cu118/
+
+# GPU 版本，需显卡驱动程序版本 ≥550.54.14（Linux）或 ≥550.54.14（Windows）
+python -m pip install paddlepaddle-gpu==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cu126/
+```
+
+> ❗ <b>注</b>：无需关注物理机上的 CUDA 版本，只需关注显卡驱动程序版本。更多飞桨 Wheel 版本请参考[飞桨官网](https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/install/pip/linux-pip.html)。
+
+安装完成后，使用以下命令验证 PaddlePaddle 是否安装成功：
+
+```bash
+python -c "import paddle; print(paddle.__version__)"
+```
+
+如果已安装成功，将输出如下版本号：
+
+```bash
+3.2.0
+```
+
+## 3. Windows 系统适配 NVIDIA 50 系显卡的 PaddlePaddle wheel 包安装
+
+通过以上方式安装的 PaddlePaddle 在 Windows 操作系统下无法正常支持 NVIDIA 50 系显卡。因此，我们提供了专门适配该硬件环境的 PaddlePaddle 安装包。请根据您的 Python 版本选择对应的 wheel 文件进行安装。
+
+```bash
+# python 3.9
+python -m pip install https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-TagBuild-Training-Windows-Gpu-Cuda12.9-Cudnn9.9-Trt10.5-Mkl-Avx-VS2019-SelfBuiltPypiUse/86d658f56ebf3a5a7b2b33ace48f22d10680d311/paddlepaddle_gpu-3.0.0.dev20250717-cp39-cp39-win_amd64.whl
+
+# python 3.10
+python -m pip install https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-TagBuild-Training-Windows-Gpu-Cuda12.9-Cudnn9.9-Trt10.5-Mkl-Avx-VS2019-SelfBuiltPypiUse/86d658f56ebf3a5a7b2b33ace48f22d10680d311/paddlepaddle_gpu-3.0.0.dev20250717-cp310-cp310-win_amd64.whl
+
+# python 3.11
+python -m pip install https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-TagBuild-Training-Windows-Gpu-Cuda12.9-Cudnn9.9-Trt10.5-Mkl-Avx-VS2019-SelfBuiltPypiUse/86d658f56ebf3a5a7b2b33ace48f22d10680d311/paddlepaddle_gpu-3.0.0.dev20250717-cp311-cp311-win_amd64.whl
+
+# python 3.12
+python -m pip install https://paddle-qa.bj.bcebos.com/paddle-pipeline/Develop-TagBuild-Training-Windows-Gpu-Cuda12.9-Cudnn9.9-Trt10.5-Mkl-Avx-VS2019-SelfBuiltPypiUse/86d658f56ebf3a5a7b2b33ace48f22d10680d311/paddlepaddle_gpu-3.0.0.dev20250717-cp312-cp312-win_amd64.whl
+```
+
+当前发布的适用于 Windows 系统 50 系显卡的 PaddlePaddle wheel 包，其文本识别模型的训练存在已知问题，相关功能仍在持续适配和完善中。
diff --git a/docs/version3.x/paddlex/overview.en.md b/docs/version3.x/paddlex/overview.en.md
new file mode 100644
index 0000000..d2eea0c
--- /dev/null
+++ b/docs/version3.x/paddlex/overview.en.md
@@ -0,0 +1,215 @@
+---
+comments: true
+hide:
+  - toc
+---
+
+## 1. Introduction to All-in-One Development
+
+The All-in-One development tool [PaddleX](https://github.com/PaddlePaddle/PaddleX/tree/release/3.0-beta1), based on the advanced technology of PaddleOCR, supports **low-code full-process** development capabilities in the OCR field. Through low-code development, simple and efficient model use, combination, and customization can be achieved. This will significantly **reduce the time consumption** of model development, **lower its development difficulty**, and greatly accelerate the application and promotion speed of models in the industry. Features include:
+
+* 🎨 [**Rich Model One-Click Call**](https://paddlepaddle.github.io/PaddleOCR/latest/en/paddlex/quick_start.html): Integrates **48 models** related to text image intelligent analysis, general OCR, general layout parsing, table recognition, formula recognition, and seal recognition into 10 pipelines, which can be quickly experienced through a simple **Python API one-click call**. In addition, the same set of APIs also supports a total of **200+ models** in image classification, object detection, image segmentation, and time series forecasting, forming 30+ single-function modules, making it convenient for developers to use **model combinations**.
+
+* 🚀 [**High Efficiency and Low barrier of entry**](https://paddlepaddle.github.io/PaddleOCR/latest/en/paddlex/overview.html): Provides two methods based on **unified commands** and **GUI** to achieve simple and efficient use, combination, and customization of models. Supports multiple deployment methods such as **high-performance inference, service-oriented deployment, and on-device deployment**. Additionally, for various mainstream hardware such as **NVIDIA GPU, Kunlunxin XPU, Ascend NPU, Cambricon MLU, and Haiguang DCU**, models can be developed with **seamless switching**.
+
+> **Note**: PaddleX is committed to achieving pipeline-level model training, inference, and deployment. A model pipeline refers to a series of predefined development processes for specific AI tasks, including combinations of single models (single-function modules) that can independently complete a type of task.
+## 2. OCR-Related Capability Support
+
+In PaddleX, all 6 OCR-related pipelines support **local inference**, and some pipelines support **online experience**. You can quickly experience the pre-trained model effects of each pipeline. If you are satisfied with the pre-trained model effects of a pipeline, you can directly proceed with [high-performance inference](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/pipeline_deploy/high_performance_inference_en.md)/[service-oriented deployment](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/pipeline_deploy/service_deploy_en.md)/[on-device deployment](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_deploy/on_device_deployment.html). If not satisfied, you can also use the **custom development** capabilities of the pipeline to improve the effects. For the complete pipeline development process, please refer to [PaddleX Pipeline Usage Overview](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/pipeline_usage/pipeline_develop_guide_en.md) or the tutorials for each pipeline.
+
+In addition, PaddleX provides developers with a full-process efficient model training and deployment tool based on a [cloud-based GUI](https://aistudio.baidu.com/pipeline/mine). Developers **do not need code development**, just need to prepare a dataset that meets the pipeline requirements to **quickly start model training**. For details, please refer to the tutorial ["Developing Industrial-level AI Models with Zero Barrier"](https://aistudio.baidu.com/practical/introduce/546656605663301).
+
+<table>
+    <tr>
+        <th>Pipeline</th>
+        <th>Online Experience</th>
+        <th>Local Inference</th>
+        <th>High-Performance Inference</th>
+        <th>Service-Oriented Deployment</th>
+        <th>On-Device Deployment</th>
+        <th>Custom Development</th>
+        <th><a href="https://aistudio.baidu.com/pipeline/mine">No-Code Development On AI Studio</a></td> 
+    </tr>
+   <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/doc_preprocessor.html">Document Image Preprocessing</a></td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+    </tr>
+    <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/OCR.html">OCR</a></td>
+        <td><a href = "https://aistudio.baidu.com/community/app/91660/webUI?source=appMineRecent">Link</a></td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+    </tr>
+    <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/table_recognition.html">Table Recognition</a></td>
+        <td><a href = "https://aistudio.baidu.com/community/app/91661?source=appMineRecent">Link</a></td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>✅</td>
+    </tr>
+    <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/table_recognition_v2.html">Table Recognition V2</a></td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+    </tr>
+   </tr>
+        <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/formula_recognition.html">Formula Recognition</a></td>
+        <td><a href = "https://aistudio.baidu.com/community/app/387976/webUI?source=appCenter">Link</a></td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>✅</td>
+    </tr>
+    <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/seal_recognition.html">Seal Recognition</a></td>
+        <td><a href = "https://aistudio.baidu.com/community/app/387977/webUI?source=appCenter">Link</a></td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>✅</td>
+    </tr>
+    <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/layout_parsing.html">Layout Parsing</a></td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+    </tr>
+    <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/layout_parsing_v2.html">Layout Parsing v2</a></td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>🚧</td>
+        <td>🚧</td>
+    </tr>
+    <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction.html">PP-ChatOCRv3-doc</a></td>
+        <td><a href = "https://aistudio.baidu.com/community/app/182491/webUI?source=appCenter">Link</a></td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>✅</td>
+    </tr>
+    <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction.html">PP-ChatOCRv4-doc</a></td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+    </tr>
+</table>
+
+
+</table>
+
+
+> ❗Note: The above capabilities are implemented based on GPU/CPU. PaddleX can also perform local inference and custom development on mainstream hardware such as Kunlunxin, Ascend, Cambricon, and Haiguang. The table below details the support status of the pipelines. For specific supported model lists, please refer to the [Model List (Kunlunxin XPU)](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/support_list/model_list_xpu_en.md)/[Model List (Ascend NPU)](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/support_list/model_list_npu_en.md)/[Model List (Cambricon MLU)](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/support_list/model_list_mlu_en.md)/[Model List (Haiguang DCU)](https://github.com/PaddlePaddle/PaddleX/blob/release/3.0-beta1/docs/support_list/model_list_dcu_en.md). We are continuously adapting more models and promoting the implementation of high-performance and service-oriented deployment on mainstream hardware.
+**🚀 Support for Domestic Hardware Capabilities**
+
+<table>
+  <tr>
+    <th>Pipeline Name</th>
+    <th>Ascend 910B</th>
+    <th>Kunlunxin XPU</th>
+    <th>Cambricon MLU</th>
+    <th>Haiguang DCU</th>
+  </tr>
+  <tr>
+    <td>General OCR</td>
+    <td>✅</td>
+    <td>✅</td>
+    <td>✅</td>
+    <td>🚧</td>
+  </tr>
+  <tr>
+    <td>Table Recognition</td>
+    <td>✅</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+  </tr>
+</table>
+
+## 3. List and Tutorials of OCR-Related Model Pipelines
+
+- **OCR Pipeline**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/OCR.html)
+- **Table Recognition Pipeline**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/table_recognition.html)
+- **PP-ChatOCRv3-doc Pipeline**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction.html)
+- **Layout Parsing Pipeline**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/layout_parsing.html)
+- **Formula Recognition Pipeline**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/formula_recognition.html)
+- **Seal Recognition Pipeline**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/seal_recognition.html)
+
+## 4. List and Tutorials of OCR-Related Modules
+
+- **Text Detection Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/text_detection.html)
+- **Seal Detection Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/seal_text_detection.html)
+- **Text Recognition Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/text_recognition.html)
+- **Formula Recognition Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/formula_recognition.html)
+- **Table Structure Recognition Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/table_structure_recognition.html)
+- **Text Image Unwarping Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/text_image_unwarping.html)
+- **Layout Detection Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/layout_detection.html)
+- **Document Image Orientation Classification Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html)
+
+
+
+
+## 3. List of OCR-related Pipeline Models and Tutorials
+
+- **Document Image Preprocessing Pipeline**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/doc_preprocessor.html)
+- **OCR Pipeline**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/OCR.html)
+- **Table Recognition Pipeline**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/table_recognition.html)
+- **Table Recognition v2 Pipeline**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/table_recognition_v2.html)
+- **Layout Parsing Pipeline**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/layout_parsing.html)
+- **Layout Parsing v2 Pipeline**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/layout_parsing_v2.html)
+- **Formula Recognition**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/formula_recognition.html)
+- **Seal Recognition**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/seal_recognition.html)
+- **PP-ChatOCRv3-doc Pipeline**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction_v3.html)
+- **PP-ChatOCRv4-doc Pipeline**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction_v4.html)
+
+## 4. List of OCR-related Single Function Modules and Tutorials
+
+- **Text Detection Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/text_detection.html)
+- **Seal Detection Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/seal_text_detection.html)
+- **Textline Orientation Classification Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/textline_orientation_classification.html)
+- **Text Recognition Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/text_recognition.html)
+- **Formula Recognition Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/formula_recognition.html)
+- **Table Structure Recognition Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/table_structure_recognition.html)
+- **Text Image Unwarping Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/text_image_unwarping.html)
+- **Layout Detection Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/layout_detection.html)
+- **Document Image Orientation Classification Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html)
+- **Table Cells Detection Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/table_cells_detection.html)
+- **Table Classification Module**: [Tutorial](https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/table_classification.html)
diff --git a/docs/version3.x/paddlex/overview.md b/docs/version3.x/paddlex/overview.md
new file mode 100644
index 0000000..3078842
--- /dev/null
+++ b/docs/version3.x/paddlex/overview.md
@@ -0,0 +1,190 @@
+---
+comments: true
+hide:
+  - toc
+---
+
+## 1. 低代码全流程开发简介
+
+飞桨低代码开发工具 [PaddleX](https://github.com/PaddlePaddle/PaddleX)，依托于 PaddleOCR 的先进技术，支持了 OCR 领域的**低代码全流程**开发能力。通过低代码开发，可实现简单且高效的模型使用、组合与定制。这将显著**减少模型开发的时间消耗**，**降低其开发难度**，大大加快模型在行业中的应用和推广速度。特色如下：
+
+* 🎨 **模型丰富一键调用**：将文本图像智能分析、通用 OCR、通用版面解析、通用表格识别、公式识别、印章文本识别涉及的**48个模型**整合为 10 条模型产线，通过极简的**Python API一键调用**，快速体验模型效果。此外，同一套API，也支持图像分类、目标检测、图像分割、时序预测等共计**200+模型**，形成30+单功能模块，方便开发者进行**模型组合**使用。
+
+* 🚀 **提高效率降低门槛**：提供基于**统一命令**和**图形界面**两种方式，实现模型简洁高效的使用、组合与定制。支持**高性能推理、服务化部署和端侧部署**等多种部署方式。此外，对于各种主流硬件如**英伟达GPU、昆仑芯、昇腾、寒武纪和海光**等，进行模型开发时，都可以**无缝切换**。
+
+>**说明**：PaddleX 致力于实现产线级别的模型训练、推理与部署。模型产线是指一系列预定义好的、针对特定AI任务的开发流程，其中包含能够独立完成某类任务的单模型（单功能模块）组合。
+
+## 2. OCR相关能力支持
+
+PaddleX 中 OCR 相关的 10 条产线均支持本地**快速推理**，部分产线支持**在线体验**，您可以快速体验各个产线的预训练模型效果，如果您对产线的预训练模型效果满意，可以直接对产线进行[高性能推理](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/high_performance_inference.html)/[服务化部署](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/service_deploy.html)/[端侧部署](https://paddlepaddle.github.io/PaddleX/latest/pipeline_deploy/on_device_deployment.html)，如果不满意，您也可以使用产线的**二次开发**能力，提升效果。完整的产线开发流程请参考[PaddleX产线使用概览](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/pipeline_develop_guide.html)或各产线使用教程。
+
+此外，PaddleX为开发者提供了基于[云端图形化开发界面](https://aistudio.baidu.com/pipeline/mine)的全流程开发工具, 详细请参考[教程《零门槛开发产业级AI模型》](https://aistudio.baidu.com/practical/introduce/546656605663301)
+
+<table >
+    <tr>
+        <th>模型产线</th>
+        <th>在线体验</th>
+        <th>快速推理</th>
+        <th>高性能推理</th>
+        <th>服务化部署</th>
+        <th>端侧部署</th>
+        <th>二次开发</th>
+        <th><a href = "https://aistudio.baidu.com/pipeline/mine">星河零代码产线</a></td>
+    </tr>
+    <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/doc_preprocessor.html">文档图像预处理</a></td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+    </tr>
+    <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/OCR.html">通用OCR</a></td>
+        <td><a href = "https://aistudio.baidu.com/community/app/91660/webUI?source=appMineRecent">链接</a></td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+    </tr>
+    <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/table_recognition.html">通用表格识别</a></td>
+        <td><a href = "https://aistudio.baidu.com/community/app/91661?source=appMineRecent">链接</a></td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>✅</td>
+    </tr>
+    <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/table_recognition_v2.html">通用表格识别v2</a></td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+    </tr>
+   </tr>
+        <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/formula_recognition.html">公式识别</a></td>
+        <td><a href = "https://aistudio.baidu.com/community/app/387976/webUI?source=appCenter">链接</a></td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>✅</td>
+    </tr>
+    <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/seal_recognition.html">印章文本识别</a></td>
+        <td><a href = "https://aistudio.baidu.com/community/app/387977/webUI?source=appCenter">链接</a></td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>✅</td>
+    </tr>
+    <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/layout_parsing.html">通用版面解析</a></td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+    </tr>
+    <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/layout_parsing_v2.html">通用版面解析v2</a></td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>🚧</td>
+        <td>🚧</td>
+    </tr>
+    <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction.html">文档场景信息抽取v3</a></td>
+        <td><a href = "https://aistudio.baidu.com/community/app/182491/webUI?source=appCenter">链接</a></td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>✅</td>
+    </tr>
+    <tr>
+        <td><a href="https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction.html">文档场景信息抽取v4</a></td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>✅</td>
+        <td>🚧</td>
+        <td>✅</td>
+        <td>🚧</td>
+    </tr>
+</table>
+
+> ❗注：以上功能均基于GPU/CPU实现。PaddleX还可在昆仑芯、昇腾、寒武纪和海光等主流硬件上进行快速推理和二次开发。下表详细列出了模型产线的支持情况，具体支持的模型列表请参阅 [模型列表(NPU)](https://paddlepaddle.github.io/PaddleX/latest/support_list/model_list_npu.html) // [模型列表(XPU)](https://paddlepaddle.github.io/PaddleX/latest/support_list/model_list_xpu.html) // [模型列表(MLU)](https://paddlepaddle.github.io/PaddleX/latest/support_list/model_list_mlu.html) // [模型列表DCU](https://paddlepaddle.github.io/PaddleX/latest/support_list/model_list_dcu.html)。同时我们也在适配更多的模型，并在主流硬件上推动高性能和服务化部署的实施。
+
+**🚀 国产化硬件能力支持**
+
+<table>
+  <tr>
+    <th>模型产线</th>
+    <th>昇腾 910B</th>
+    <th>昆仑芯 R200/R300</th>
+    <th>寒武纪 MLU370X8</th>
+    <th>海光 Z100/K100AI</th>
+  </tr>
+  <tr>
+    <td>通用OCR</td>
+    <td>✅</td>
+    <td>✅</td>
+    <td>✅</td>
+    <td>✅</td>
+  </tr>
+  <tr>
+    <td>通用表格识别</td>
+    <td>✅</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+  </tr>
+</table>
+
+## 3. OCR相关模型产线列表和教程
+
+* **文档图像预处理产线**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/doc_preprocessor.html)
+* **通用OCR产线**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/OCR.html)
+* **通用表格识别产线**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/table_recognition.html)
+* **通用表格识别v2产线**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/table_recognition_v2.html)
+* **通用版面解析产线**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/layout_parsing.html)
+* **通用版面解析v2产线**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/layout_parsing_v2.html)
+* **公式识别**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/formula_recognition.html)
+* **印章文本识别**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/seal_recognition.html)
+* **文档场景信息抽取v3产线**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction_v3.html)
+* **文档场景信息抽取v4产线**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction_v4.html)
+
+## 4. OCR相关单功能模块列表和教程
+
+* **文本检测模块**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/text_detection.html)
+* **印章文本检测模块**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/seal_text_detection.html)
+* **文本行方向分类模块**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/textline_orientation_classification.html)
+* **文本识别模块**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/text_recognition.html)
+* **公式识别模块**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/formula_recognition.html)
+* **表格结构识别模块**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/table_structure_recognition.html)
+* **文本图像矫正模块**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/text_image_unwarping.html)
+* **版面区域检测模块**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/layout_detection.html)
+* **文档图像方向分类模块**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html)
+* **表格单元格检测模块**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/table_cells_detection.html)
+* **表格分类模块**: [使用教程](https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/table_classification.html)
diff --git a/docs/version3.x/paddlex/quick_start.en.md b/docs/version3.x/paddlex/quick_start.en.md
new file mode 100644
index 0000000..17d4939
--- /dev/null
+++ b/docs/version3.x/paddlex/quick_start.en.md
@@ -0,0 +1,136 @@
+---
+comments: true
+hide:
+  - toc
+---
+
+# ⏭️ Quick Start
+
+> **Note:**
+
+>* The low-code development tool [PaddleX](https://github.com/PaddlePaddle/PaddleX) by PaddlePaddle leverages the advanced technology of PaddleOCR to support **end-to-end low-code development** capabilities in the OCR field. With low-code development, users can achieve simple and efficient model usage, combination, and customization.
+
+>* PaddleX is dedicated to enabling production-level model training, inference, and deployment. A model pipeline refers to a series of predefined development processes for specific AI tasks, including combinations of single models (single-function modules) that can independently accomplish certain tasks. This document provides quick inference usage for **OCR-related pipelines**. For quick usage of single-function modules and more features, please refer to the relevant sections in [Low-Code End-to-End Development with PaddleOCR](https://www.paddleocr.ai/latest/version3.x/paddlex/overview.html).
+
+>* For information about the relationship between PaddleOCR and PaddleX, please refer to the [Introduction to the Relationship between PaddleOCR and PaddleX](https://www.paddleocr.ai/latest/version3.x/paddleocr_and_paddlex.html).
+
+## 🛠️ Installation
+
+> ❗Before installing PaddleX, please make sure you have a basic **Python runtime environment** (Note: Currently, Python 3.8 to Python 3.13 are supported). PaddleX version 3.2 requires PaddlePaddle version 3.0.0 or higher as a dependency.
+
+* **Installing PaddlePaddle**
+
+```bash
+# CPU
+python -m pip install paddlepaddle==3.0.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+
+# gpu，requires GPU driver version ≥450.80.02 (Linux) or ≥452.39 (Windows)
+python -m pip install paddlepaddle-gpu==3.0.0 -i https://www.paddlepaddle.org.cn/packages/stable/cu118/
+
+# gpu，requires GPU driver version ≥550.54.14 (Linux) or ≥550.54.14 (Windows)
+python -m pip install paddlepaddle-gpu==3.0.0 -i https://www.paddlepaddle.org.cn/packages/stable/cu126/
+```
+> ❗No need to focus on the CUDA version on the physical machine, only the GPU driver version needs attention. For more information on PaddlePaddle Wheel versions, please refer to the [PaddlePaddle Official Website](https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation./docs/en/install/pip/linux-pip.html).
+
+* **Installing PaddleX**
+
+```bash
+pip install "paddlex[ocr]"
+```
+
+> ❗For more installation methods, refer to the [PaddleX Installation Guide](https://paddlepaddle.github.io/PaddleX/latest/en/installation/installation.html).
+
+
+## 💻 CLI Usage
+
+One command can quickly experience the pipeline effect, the unified CLI format is:
+
+```bash
+paddlex --pipeline [Pipeline Name] --input [Input Image] --device [Running Device]
+```
+
+Each Pipeline in PaddleX corresponds to specific parameters, which you can view in the respective Pipeline documentation for detailed explanations. Each Pipeline requires specifying three necessary parameters:
+
+* `pipeline`: The name of the Pipeline or the configuration file of the Pipeline
+* `input`: The local path, directory, or URL of the input file (e.g., an image) to be processed
+* `device`: The hardware device and its index to use (e.g., `gpu:0` indicates using the 0th GPU), or you can choose to use NPU (`npu:0`), XPU (`xpu:0`), CPU (`cpu`), etc.
+
+For example, using the  OCR pipeline:
+```bash
+paddlex --pipeline OCR \
+        --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png \
+        --use_doc_orientation_classify False \
+        --use_doc_unwarping False \
+        --use_textline_orientation False \
+        --save_path ./output \
+        --device gpu:0
+```
+  <summary><b>👉 Click to view the running result</b></summary>
+
+```bash
+{'res': {'input_path': 'general_ocr_002.png', 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_textline_orientation': False}, 'doc_preprocessor_res': {'input_path': None, 'model_settings': {'use_doc_orientation_classify': True, 'use_doc_unwarping': False}, 'angle': 0},'dt_polys': [array([[ 3, 10],
+       [82, 10],
+       [82, 33],
+       [ 3, 33]], dtype=int16), ...], 'text_det_params': {'limit_side_len': 960, 'limit_type': 'max', 'thresh': 0.3, 'box_thresh': 0.6, 'unclip_ratio': 2.0}, 'text_type': 'general', 'textline_orientation_angles': [-1, ...], 'text_rec_score_thresh': 0.0, 'rec_texts': ['www.99*', ...], 'rec_scores': [0.8980069160461426,  ...], 'rec_polys': [array([[ 3, 10],
+       [82, 10],
+       [82, 33],
+       [ 3, 33]], dtype=int16), ...], 'rec_boxes': array([[  3,  10,  82,  33], ...], dtype=int16)}}
+```
+
+The visualization result is as follows:
+
+![alt text](https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/boardingpass.png)
+
+
+To use the command line for other pipelines, simply adjust the `pipeline` parameter to the name of the corresponding pipeline and modify the parameters accordingly. Below are the commands for each pipeline:
+
+  <summary><b>👉 More CLI usage for pipelines</b></summary>
+
+| Pipeline Name                | Command                                                                                                                                                                                    |
+|------------------------------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| OCR                      | `paddlex --pipeline OCR --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png --use_doc_orientation_classify False --use_doc_unwarping False --use_textline_orientation False --save_path ./output --device gpu:0`                                      |
+| Document Image Preprocessor      | `paddlex --pipeline doc_preprocessor --input https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/doc_test_rotated.jpg --use_doc_orientation_classify True --use_doc_unwarping True --save_path ./output --device gpu:0`                                                      |
+| Table Recognition        | `paddlex --pipeline table_recognition --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg --save_path ./output --device gpu:0`                                      |
+| Table Recognition v2     | `paddlex --pipeline table_recognition_v2 --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg --save_path ./output --device gpu:0`                                      |
+| Formula Recognition              | `paddlex --pipeline formula_recognition --input https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/general_formula_recognition.png --use_layout_detection True --use_doc_orientation_classify False --use_doc_unwarping False --layout_threshold 0.5 --layout_nms True --layout_unclip_ratio  1.0 --layout_merge_bboxes_mode large --save_path ./output --device gpu:0`                                      |
+| Seal Recognition            | `paddlex --pipeline seal_recognition --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/seal_text_det.png --use_doc_orientation_classify False --use_doc_unwarping False --device gpu:0 --save_path ./output`                                      |
+| Layout Parsing           | `paddlex --pipeline layout_parsing --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/demo_paper.png --use_doc_orientation_classify False --use_doc_unwarping False --use_textline_orientation False --save_path ./output --device gpu:0`                      |
+| PP-StructureV3        | `paddlex --pipeline PP-StructureV3 --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/pp_structure_v3_demo.png --use_doc_orientation_classify False --use_doc_unwarping False --use_textline_orientation False --save_path ./output --device gpu:0`                      |
+
+
+
+## 📝 Python Script Usage
+
+A few lines of code can complete the quick inference of the pipeline, the unified Python script format is as follows:
+```python
+from paddlex import create_pipeline
+
+pipeline = create_pipeline(pipeline=[Pipeline Name])
+output = pipeline.predict([Input Image Name])
+for res in output:
+    res.print()
+    res.save_to_img("./output/")
+    res.save_to_json("./output/")
+```
+The following steps are executed:
+
+* `create_pipeline()` instantiates the pipeline object
+* Passes the image and calls the `predict()` method of the pipeline object for inference prediction
+* Processes the prediction results
+
+To use the Python script for other pipelines, simply adjust the `pipeline` parameter in the `create_pipeline()` method to the name of the corresponding pipeline and modify the parameters accordingly. Below are the parameter names and detailed usage explanations for each pipeline:
+
+👉 More Python script usage for pipelines
+
+| pipeline Name           | Corresponding Parameter               | Detailed Explanation                                                                                                      |
+|-------------------------------|-------------------------------------|---------------------------------------------------------------------------------------------------------------|
+| OCR | `OCR` | [Instructions for Using the General OCR Pipeline Python Script](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/OCR.html#222-python-script-integration) |
+| Document Image Preprocessing | `doc_preprocessor` | [Instructions for Using the Document Image Preprocessing Pipeline Python Script](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/doc_preprocessor.html#212-python-script-integration) |
+| Table Recognition | `table_recognition` | [Instructions for Using the General Table Recognition Pipeline Python Script](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/table_recognition.html#22-python-script-integration) |
+| Table Recognition v2 | `table_recognition_v2` | [Instructions for Using the General Table Recognition v2 Pipeline Python Script](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/table_recognition_v2.html#22-integration-via-python-script) |
+| Formula Recognition | `formula_recognition` | [Instructions for Using the Formula Recognition Pipeline Python Script](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/formula_recognition.html#222-python-script-integration) |
+| Seal Recognition | `seal_recognition` | [Instructions for Using the Seal Text Recognition Pipeline Python Script](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/seal_recognition.html#222-python-script-integration) |
+| Layout Parsing | `layout_parsing` | [Instructions for Using the General Layout Parsing Pipeline Python Script](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/layout_parsing.html#22-integrating-via-python-script) |
+| PP-StructureV3 | `P-StructureV3` | [Instructions for Using the P-StructureV3 Pipeline Python Script](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/ocr_pipelines/PP-StructureV3.html#22-python-script-integration) |
+| PP-ChatOCRv3-doc   | `PP-ChatOCRv3-doc` | [PP-ChatOCRv3-doc Pipeline Python Script Usage Instructions](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction_v3.html#22-local-experience) |
+| PP-ChatOCRv4-doc | `PP-ChatOCRv4-doc` | [PP-ChatOCRv4-doc Pipeline Python Script Usage Instructions](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction_v4.html#21-local-experience) |
diff --git a/docs/version3.x/paddlex/quick_start.md b/docs/version3.x/paddlex/quick_start.md
new file mode 100644
index 0000000..4688334
--- /dev/null
+++ b/docs/version3.x/paddlex/quick_start.md
@@ -0,0 +1,135 @@
+---
+comments: true
+hide:
+  - toc
+---
+
+# 快速开始
+
+>**说明：**
+
+>* 飞桨低代码开发工具[PaddleX](https://github.com/PaddlePaddle/PaddleX)，依托于PaddleOCR的先进技术，支持了OCR领域的**低代码全流程**开发能力。通过低代码开发，可实现简单且高效的模型使用、组合与定制。
+
+>* PaddleX 致力于实现产线级别的模型训练、推理与部署。模型产线是指一系列预定义好的、针对特定AI任务的开发流程，其中包含能够独立完成某类任务的单模型（单功能模块）组合。本文档提供**OCR相关产线**的快速推理使用，单功能模块的快速使用以及更多功能请参考[PaddleOCR低代码全流程开发](https://www.paddleocr.ai/latest/version3.x/paddlex/overview.html)中相关章节。
+
+>* 关于 PaddleOCR 和 PaddleX 的关系，可以查看[ PaddleOCR 与 PaddleX 关系介绍](https://www.paddleocr.ai/latest/version3.x/paddleocr_and_paddlex.html)。
+
+
+### 🛠️ 安装
+
+> ❗在安装 PaddleX 之前，请确保您已具备基本的 **Python 运行环境**（注：目前支持 Python 3.8 至 Python 3.13）。PaddleX 3.2 版本依赖的 PaddlePaddle 版本为 3.0.0 以上版本。
+
+* **安装 PaddlePaddle**
+```bash
+# CPU 版本
+python -m pip install paddlepaddle==3.0.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+
+# GPU 版本，需显卡驱动程序版本 ≥450.80.02（Linux）或 ≥452.39（Windows）
+python -m pip install paddlepaddle-gpu==3.0.0 -i https://www.paddlepaddle.org.cn/packages/stable/cu118/
+
+# GPU 版本，需显卡驱动程序版本 ≥550.54.14（Linux）或 ≥550.54.14（Windows）
+python -m pip install paddlepaddle-gpu==3.0.0 -i https://www.paddlepaddle.org.cn/packages/stable/cu126/
+```
+> ❗无需关注物理机上的 CUDA 版本，只需关注显卡驱动程序版本。更多飞桨 Wheel 版本信息，请参考[飞桨官网](https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation./docs/zh/install/pip/linux-pip.html)。
+
+* **安装PaddleX**
+
+```bash
+pip install "paddlex[ocr]"
+```
+
+> ❗ 更多安装方式参考 [PaddleX 安装教程](https://paddlepaddle.github.io/PaddleX/latest/installation/installation.html)
+
+### 💻 命令行使用
+
+一行命令即可快速体验产线效果，统一的命令行格式为：
+
+```bash
+paddlex --pipeline [产线名称] --input [输入图片] --device [运行设备]
+```
+
+PaddleX的每一条产线对应特定的参数，您可以在各自的产线文档中查看具体的参数说明。每条产线需指定必要的三个参数：
+* `pipeline`：产线名称或产线配置文件
+* `input`：待处理的输入文件（如图片）的本地路径、目录或 URL
+* `device`：使用的硬件设备及序号（例如`gpu:0`表示使用第 0 块 GPU），也可选择使用 NPU(`npu:0`)、 XPU(`xpu:0`)、CPU(`cpu`)等。
+
+
+以通用 OCR 产线为例：
+```bash
+paddlex --pipeline OCR \
+        --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png \
+        --use_doc_orientation_classify False \
+        --use_doc_unwarping False \
+        --use_textline_orientation False \
+        --save_path ./output \
+        --device gpu:0
+```
+
+<b>👉 点击查看运行结果 </b><
+
+```bash
+{'res': {'input_path': 'general_ocr_002.png', 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_textline_orientation': False}, 'doc_preprocessor_res': {'input_path': None, 'model_settings': {'use_doc_orientation_classify': True, 'use_doc_unwarping': False}, 'angle': 0},'dt_polys': [array([[ 3, 10],
+       [82, 10],
+       [82, 33],
+       [ 3, 33]], dtype=int16), ...], 'text_det_params': {'limit_side_len': 960, 'limit_type': 'max', 'thresh': 0.3, 'box_thresh': 0.6, 'unclip_ratio': 2.0}, 'text_type': 'general', 'textline_orientation_angles': [-1, ...], 'text_rec_score_thresh': 0.0, 'rec_texts': ['www.99*', ...], 'rec_scores': [0.8980069160461426,  ...], 'rec_polys': [array([[ 3, 10],
+       [82, 10],
+       [82, 33],
+       [ 3, 33]], dtype=int16), ...], 'rec_boxes': array([[  3,  10,  82,  33], ...], dtype=int16)}}
+```
+
+可视化结果如下：
+
+![alt text](https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/boardingpass.png)
+
+
+其他产线的命令行使用，只需将 `pipeline` 参数调整为相应产线的名称，参数调整为对应的产线的参数即可。下面列出了每个产线对应的命令：
+
+<b>👉 更多产线的命令行使用</b>
+
+| 产线名称           | 使用命令                                                                                                                                                                                    |
+|--------------------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| 文档图像预处理            | `paddlex --pipeline doc_preprocessor --input https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/doc_test_rotated.jpg --use_doc_orientation_classify True --use_doc_unwarping True --save_path ./output --device gpu:0`                                                      |
+| 通用OCR            | `paddlex --pipeline OCR --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png --use_doc_orientation_classify False --use_doc_unwarping False --use_textline_orientation False --save_path ./output --device gpu:0`                                                      |
+| 通用表格识别       | `paddlex --pipeline table_recognition --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg --save_path ./output --device gpu:0`                                      |
+| 通用表格识别v2       | `paddlex --pipeline table_recognition_v2 --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition.jpg --save_path ./output --device gpu:0`                                      |
+| 公式识别       | `paddlex --pipeline formula_recognition --input https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/general_formula_recognition.png --use_layout_detection True --use_doc_orientation_classify False --use_doc_unwarping False --layout_threshold 0.5 --layout_nms True --layout_unclip_ratio  1.0 --layout_merge_bboxes_mode large --save_path ./output --device gpu:0`                                      |
+| 印章文本识别       | `paddlex --pipeline seal_recognition --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/seal_text_det.png --use_doc_orientation_classify False --use_doc_unwarping False --device gpu:0 --save_path ./output`                                      |
+| 通用版面解析       | `paddlex --pipeline layout_parsing --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/demo_paper.png --use_doc_orientation_classify False --use_doc_unwarping False --use_textline_orientation False --save_path ./output --device gpu:0`                                      |
+| 通用版面解析v3       | `paddlex --pipeline PP-StructureV3 --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/pp_structure_v3_demo.png --use_doc_orientation_classify False --use_doc_unwarping False --use_textline_orientation False --save_path ./output --device gpu:0`                                      |
+
+
+### 📝 Python 脚本使用
+
+几行代码即可完成产线的快速推理，统一的 Python 脚本格式如下：
+
+```python
+from paddlex import create_pipeline
+
+pipeline = create_pipeline(pipeline=[产线名称])
+output = pipeline.predict([输入图片名称])
+for res in output:
+    res.print()
+    res.save_to_img("./output/")
+    res.save_to_json("./output/")
+```
+执行了如下几个步骤：
+
+* `create_pipeline()` 实例化产线对象
+* 传入图片并调用产线对象的 `predict()` 方法进行推理预测
+* 对预测结果进行处理
+
+其他产线的 Python 脚本使用，只需将 `create_pipeline()` 方法的 `pipeline` 参数调整为相应产线的名称，参数调整为对应的产线的参数即可。下面列出了每个产线对应的参数名称及详细的使用解释：
+<b>👉 更多产线的Python脚本使用</b>
+
+| 产线名称           | 对应参数                           | 详细说明                                                                                                                                                         |
+|--------------------|------------------------------------|------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| 通用OCR            | `OCR`                              | [通用OCR产线Python脚本使用说明](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/OCR.html#222-python)                                                     |
+| 文档图像预处理            | `doc_preprocessor`                              | [文档图像预处理产线Python脚本使用说明](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/doc_preprocessor.html#212-python脚)                       |
+| 通用表格识别       | `table_recognition`                | [通用表格识别产线Python脚本使用说明](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/table_recognition.html#22-python脚)                                   |
+| 通用表格识别v2      | `table_recognition_v2`                | [通用表格识别v2产线Python脚本使用说明](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/table_recognition_v2.html#22-python脚)                                   |
+| 公式识别       | `formula_recognition`                | [公式识别产线Python脚本使用说明](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/formula_recognition.html#222-python)                                   |
+| 印章文本识别       | `seal_recognition`                | [印章文本识别产线Python脚本使用说明](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/seal_recognition.html#222-python)                                   |
+| 通用版面解析       | `layout_parsing`                | [通用版面解析产线Python脚本使用说明](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/layout_parsing.html#22-python)                                   |
+| 通用版面解析v3      | `PP-StructureV3`                | [通用版面解析v3产线Python脚本使用说明](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/ocr_pipelines/PP-StructureV3.html#22-python)                                   |
+| 文档场景信息抽取v3   | `PP-ChatOCRv3-doc`                 | [文档场景信息抽取v3产线Python脚本使用说明](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction_v3.html#22) |
+| 文档场景信息抽取v4   | `PP-ChatOCRv4-doc`                 | [文档场景信息抽取v4产线Python脚本使用说明](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/tutorials/information_extraction_pipelines/document_scene_information_extraction_v4.html#21) |
diff --git a/docs/version3.x/pipeline_usage/OCR.en.md b/docs/version3.x/pipeline_usage/OCR.en.md
new file mode 100644
index 0000000..d21d58f
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/OCR.en.md
@@ -0,0 +1,2891 @@
+---
+comments: true
+---
+
+# General OCR Pipeline Usage Tutorial
+
+## 1. OCR Pipeline Introduction
+
+OCR is a technology that converts text from images into editable text. It is widely used in fields such as document digitization, information extraction, and data processing. OCR can recognize printed text, handwritten text, and even certain types of fonts and symbols.
+
+The general OCR pipeline is used to solve text recognition tasks by extracting text information from images and outputting it in text form. This pipeline supports PP-OCRv3, PP-OCRv4, PP-OCRv5, and PP-OCRv6 models, with the default model being **PP-OCRv6_medium** released with PaddleOCR 3.7. PP-OCRv6, built on the newly designed PPLCNetV4 unified backbone, offers tiny/small/medium tiers. The medium tier achieves +5.1% (recognition) and +4.6% (detection) over PP-OCRv5_server with faster inference; a single model supports 50 languages including Chinese, English, Japanese, and 46 Latin-script languages.
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/ocr/01.png"/>
+
+<b>The General OCR Pipeline consists of the following 5 modules. Each module can be independently trained and inferred, and includes multiple models. For detailed information, click the corresponding module to view its documentation.</b>
+
+- [Document Image Orientation Classification Module](../module_usage/doc_img_orientation_classification.md) (Optional)
+- [Text Image Unwarping Module](../module_usage/text_image_unwarping.md) (Optional)
+- [Text Line Orientation Classification Module](../module_usage/textline_orientation_classification.md) (Optional)
+- [Text Detection Module](../module_usage/text_detection.md)
+- [Text Recognition Module](../module_usage/text_recognition.md)
+
+In this pipeline, you can select models based on the benchmark test data provided below.
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing.
+> In the inference time columns labeled [Standard Mode / High-Performance Mode], the Standard Mode values correspond to the local `paddle_static` inference engine.
+
+<details>
+<summary><b>Document Image Orientation Classification Module (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Top-1 Acc (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High-Performance Mode]</th>
+<th>Model Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">Training Model</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>Document image classification model based on PP-LCNet_x1_0, with four categories: 0°, 90°, 180°, and 270°.</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Text Image Unwarp Module (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>CER</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">Training Model</a></td>
+<td>0.179</td>
+<td>19.05 / 19.05</td>
+<td>- / 869.82</td>
+<td>30.3</td>
+<td>High-precision Text Image Unwarping model.</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Text Line Orientation Classification Module (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Top-1 Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x0_25_textline_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x0_25_textline_ori_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x0_25_textline_ori_pretrained.pdparams">Training Model</a></td>
+<td>98.85</td>
+<td>2.16 / 0.41</td>
+<td>2.37 / 0.73</td>
+<td>0.96</td>
+<td>Text line classification model based on PP-LCNet_x0_25, with two classes: 0 degrees and 180 degrees</td>
+</tr>
+<tr>
+<td>PP-LCNet_x1_0_textline_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_textline_ori_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_textline_ori_pretrained.pdparams">Training Model</a></td>
+<td>99.42</td>
+<td>- / -</td>
+<td>2.98 / 2.98</td>
+<td>6.5</td>
+<td>Text line classification model based on PP-LCNet_x1_0, with two classes: 0 degrees and 180 degrees</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Text Detection Module:</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Detection Hmean (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High-Performance Mode]</th>
+<th>Model Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv6_medium_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_medium_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_medium_det_pretrained.pdparams">Training Model</a></td>
+<td>86.2*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>59.4</td>
+<td>PP-OCRv6 medium-scale text detection model based on PPLCNetV4 + RepLKFPN, highest accuracy, suitable for server deployment</td>
+</tr>
+<tr>
+<td>PP-OCRv6_small_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_small_det_pretrained.pdparams">Training Model</a></td>
+<td>84.1*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>9.6</td>
+<td>PP-OCRv6 small text detection model, balancing accuracy and efficiency, suitable for mobile deployment</td>
+</tr>
+<tr>
+<td>PP-OCRv6_tiny_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_tiny_det_pretrained.pdparams">Training Model</a></td>
+<td>80.6*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>1.9</td>
+<td>PP-OCRv6 ultra-lightweight text detection model (0.43M params), suitable for edge/IoT scenarios</td>
+</tr>
+<tr>
+<td>PP-OCRv5_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_det_pretrained.pdparams">Training Model</a></td>
+<td>83.8</td>
+<td>89.55 / 70.19</td>
+<td>383.15 / 383.15</td>
+<td>84.3</td>
+<td>PP-OCRv5 server-side text detection model with higher accuracy, suitable for deployment on high-performance servers</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_det_pretrained.pdparams">Training Model</a></td>
+<td>79.0</td>
+<td>10.67 / 6.36</td>
+<td>57.77 / 28.15</td>
+<td>4.7</td>
+<td>PP-OCRv5 mobile-side text detection model with higher efficiency, suitable for deployment on edge devices</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_det_pretrained.pdparams">Training Model</a></td>
+<td>69.2</td>
+<td>127.82 / 98.87</td>
+<td>585.95 / 489.77</td>
+<td>109</td>
+<td>PP-OCRv4 server-side text detection model with higher accuracy, suitable for deployment on high-performance servers</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams">Training Model</a></td>
+<td>63.8</td>
+<td>9.87 / 4.17</td>
+<td>56.60 / 20.79</td>
+<td>4.7</td>
+<td>PP-OCRv4 mobile-side text detection model with higher efficiency, suitable for deployment on edge devices</td>
+</tr>
+</tbody>
+</table>
+
+> *Note: PP-OCRv6 metrics are evaluated on an internal multi-scenario evaluation set, while PP-OCRv5/v4 metrics are based on a general evaluation set. As the evaluation sets differ, the metrics are not directly comparable.
+
+</details>
+
+<details>
+<summary><b>Text Recognition Module:</b></summary>
+<table>
+<tr>
+<th>Model</th><th>Model Download Links</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+<tr>
+<td>PP-OCRv6_medium_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_medium_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_medium_rec_pretrained.pdparams">Training Model</a></td>
+<td>83.2*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>73.3</td>
+<td rowspan="3">PP-OCRv6 text recognition models based on PPLCNetV4 + LightSVTR + CTC/NRTR multi-head decoder, single model supports 50 languages (tiny: 49)</td>
+</tr>
+<tr>
+<td>PP-OCRv6_small_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_small_rec_pretrained.pdparams">Training Model</a></td>
+<td>81.3*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>20.4</td>
+</tr>
+<tr>
+<td>PP-OCRv6_tiny_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_tiny_rec_pretrained.pdparams">Training Model</a></td>
+<td>73.5*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>4.4</td>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>86.38</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec is a next-generation text recognition model. It aims to efficiently and accurately support the recognition of four major languages—Simplified Chinese, Traditional Chinese, English, and Japanese—as well as complex text scenarios such as handwriting, vertical text, pinyin, and rare characters using a single model. While maintaining recognition performance, it balances inference speed and model robustness, providing efficient and accurate technical support for document understanding in various scenarios.</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>81.29</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">Pretrained Model</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc is trained on a mixed dataset of more Chinese document data and PP-OCR training data, building upon PP-OCRv4_server_rec. It enhances the recognition capabilities for some Traditional Chinese characters, Japanese characters, and special symbols, supporting over 15,000 characters. In addition to improving document-related text recognition, it also enhances general text recognition capabilities.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>A lightweight recognition model of PP-OCRv4 with high inference efficiency, suitable for deployment on various hardware devices, including edge devices.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>The server-side model of PP-OCRv4, offering high inference accuracy and deployable on various servers.</td>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>An ultra-lightweight English recognition model trained based on the PP-OCRv4 recognition model, supporting English and numeric character recognition.</td>
+</tr>
+</table>
+
+> *Note: PP-OCRv6 metrics are evaluated on an internal multi-scenario evaluation set, while PP-OCRv5/v4 metrics are based on a general evaluation set. As the evaluation sets differ, the metrics are not directly comparable.
+
+> ❗ The above section lists the <b>6 core models</b> that are primarily supported by the text recognition module. In total, the module supports <b>20 comprehensive models</b>, including multiple multilingual text recognition models. Below is the complete list of models:
+
+<details><summary> 👉Details of the Model List</summary>
+
+* <b>PP-OCRv5 Multi-Scenario Models</b>
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Links</th>
+<th>Avg Accuracy for Chinese Recognition (%)</th>
+<th>Avg Accuracy for English Recognition (%)</th>
+<th>Avg Accuracy for Traditional Chinese Recognition (%)</th>
+<th>Avg Accuracy for Japanese Recognition (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>86.38</td>
+<td>64.70</td>
+<td>93.29</td>
+<td>60.35</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec is a next-generation text recognition model. It aims to efficiently and accurately support the recognition of four major languages—Simplified Chinese, Traditional Chinese, English, and Japanese—as well as complex text scenarios such as handwriting, vertical text, pinyin, and rare characters using a single model. While maintaining recognition performance, it balances inference speed and model robustness, providing efficient and accurate technical support for document understanding in various scenarios.</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>81.29</td>
+<td>66.00</td>
+<td>83.55</td>
+<td>54.65</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+</table>
+
+* <b> Chinese Recognition Models</b>
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">Training Model</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc is built upon PP-OCRv4_server_rec and trained on mixed data including more Chinese document data and PP-OCR training data. It enhances recognition of traditional Chinese characters, Japanese, and special symbols, supporting 15,000+ characters. It improves both document-specific and general text recognition capabilities.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>Lightweight recognition model of PP-OCRv4 with high inference efficiency, deployable on various hardware devices including edge devices</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">Training Model</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>Server-side model of PP-OCRv4 with high inference accuracy, deployable on various server platforms</td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>72.96</td>
+<td>3.89 / 1.16</td>
+<td>8.72 / 3.56</td>
+<td>10.3</td>
+<td>Lightweight recognition model of PP-OCRv3 with high inference efficiency, deployable on various hardware devices including edge devices</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>ch_SVTRv2_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_SVTRv2_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_SVTRv2_rec_pretrained.pdparams">Training Model</a></td>
+<td>68.81</td>
+<td>10.38 / 8.31</td>
+<td>66.52 / 30.83</td>
+<td>80.5</td>
+<td rowspan="1">
+SVTRv2 is a server-side text recognition model developed by the OpenOCR team from Fudan University Vision and Learning Lab (FVL). It won first prize in the PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition, improving end-to-end recognition accuracy by 6% compared to PP-OCRv4 on List A.
+</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>ch_RepSVTR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_RepSVTR_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_RepSVTR_rec_pretrained.pdparams">Training Model</a></td>
+<td>65.07</td>
+<td>6.29 / 1.57</td>
+<td>20.64 / 5.40</td>
+<td>48.8</td>
+<td rowspan="1">RepSVTR is a mobile text recognition model based on SVTRv2. It won first prize in the PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition, improving end-to-end recognition accuracy by 2.5% compared to PP-OCRv4 on List B while maintaining comparable inference speed.</td>
+</tr>
+</table>
+
+* <b> English Recognition Models</b>
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>en_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv5_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td> 85.25</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>The ultra-lightweight English text recognition model, trained based on the PP-OCRv5 framework, further improves the accuracy of English text recognition, optimizes the omission of spaces, and enhances the recognition performance on handwritten English text.</td>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td> 70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>Ultra-lightweight English recognition model based on PP-OCRv4, supporting English and digit recognition</td>
+</tr>
+<tr>
+<td>en_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>70.69</td>
+<td>3.56 / 0.78</td>
+<td>8.44 / 5.78</td>
+<td>17.3</td>
+<td>Ultra-lightweight English recognition model based on PP-OCRv3, supporting English and digit recognition</td>
+</tr>
+</table>
+
+
+* <b>Multilingual Recognition Models</b>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>korean_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv5_mobile_rec_pretrained.pdparams">Pre-trained Model</a></td>
+<td>88.0</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>14</td>
+<td>An ultra-lightweight Korean text recognition model trained based on the PP-OCRv5 recognition framework. Supports Korean, English and numeric text recognition.</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv5_mobile_rec_pretrained.pdparams">Pre-trained Model</a></td>
+<td>84.7</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>14</td>
+<td>A Latin-script text recognition model trained based on the PP-OCRv5 recognition framework. Supports most Latin alphabet languages and numeric text recognition.</td>
+</tr>
+<tr>
+<td>eslav_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+eslav_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/eslav_PP-OCRv5_mobile_rec_pretrained.pdparams">Pre-trained Model</a></td>
+<td>81.6</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>14</td>
+<td>An East Slavic language recognition model trained based on the PP-OCRv5 recognition framework. Supports East Slavic languages, English and numeric text recognition.</td>
+</tr>
+<tr>
+<td>th_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+th_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/th_PP-OCRv5_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>82.68</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>The Thai recognition model trained based on the PP-OCRv5 recognition model supports recognition of Thai, English, and numbers.</td>
+</tr>
+<tr>
+<td>el_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+el_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/el_PP-OCRv5_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>89.28</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>The Greek recognition model trained based on the PP-OCRv5 recognition model supports recognition of Greek, English, and numbers.</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/arabic_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>81.27</td>
+<td>-</td>
+<td>-</td>
+<td>7.6</td>
+<td>Ultra-lightweight Arabic character recognition model trained based on the PP-OCRv5 recognition model, supports Arabic letters and number recognition</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/cyrillic_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>80.27</td>
+<td>-</td>
+<td>-</td>
+<td>7.7</td>
+<td>Ultra-lightweight Cyrillic character recognition model trained based on the PP-OCRv5 recognition model, supports Cyrillic letters and number recognition</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/devanagari_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>84.96</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>Ultra-lightweight Devanagari script recognition model trained based on the PP-OCRv5 recognition model, supports Hindi, Sanskrit and other Devanagari letters, as well as number recognition</td>
+</tr>
+<tr>
+<td>te_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/te_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>87.65</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>Ultra-lightweight Telugu script recognition model trained based on the PP-OCRv5 recognition model, supports Telugu script and number recognition</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ta_PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>94.2</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>Ultra-lightweight Tamil script recognition model trained based on the PP-OCRv5 recognition model, supports Tamil script and number recognition</td>
+</tr>
+<tr>
+<td>korean_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>60.21</td>
+<td>3.73 / 0.98</td>
+<td>8.76 / 2.91</td>
+<td>9.6</td>
+<td>Ultra-lightweight Korean recognition model based on PP-OCRv3, supporting Korean and numeric recognition</td>
+</tr>
+<tr>
+<td>japan_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+japan_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/japan_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>45.69</td>
+<td>3.86 / 1.01</td>
+<td>8.62 / 2.92</td>
+<td>9.8</td>
+<td>Ultra-lightweight Japanese recognition model based on PP-OCRv3, supporting Japanese and numeric recognition</td>
+</tr>
+<tr>
+<td>chinese_cht_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+chinese_cht_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/chinese_cht_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>82.06</td>
+<td>3.90 / 1.16</td>
+<td>9.24 / 3.18</td>
+<td>10.8</td>
+<td>Ultra-lightweight Traditional Chinese recognition model based on PP-OCRv3, supporting Traditional Chinese and numeric recognition</td>
+</tr>
+<tr>
+<td>te_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+te_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>95.88</td>
+<td>3.59 / 0.81</td>
+<td>8.28 / 6.21</td>
+<td>8.7</td>
+<td>Ultra-lightweight Telugu recognition model based on PP-OCRv3, supporting Telugu and numeric recognition</td>
+</tr>
+<tr>
+<td>ka_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ka_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ka_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>96.96</td>
+<td>3.49 / 0.89</td>
+<td>8.63 / 2.77</td>
+<td>17.4</td>
+<td>Ultra-lightweight Kannada recognition model based on PP-OCRv3, supporting Kannada and numeric recognition</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ta_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>76.83</td>
+<td>3.49 / 0.86</td>
+<td>8.35 / 3.41</td>
+<td>8.7</td>
+<td>Ultra-lightweight Tamil recognition model based on PP-OCRv3, supporting Tamil and numeric recognition</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>76.93</td>
+<td>3.53 / 0.78</td>
+<td>8.50 / 6.83</td>
+<td>8.7</td>
+<td>Ultra-lightweight Latin recognition model based on PP-OCRv3, supporting Latin and numeric recognition</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+arabic_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>73.55</td>
+<td>3.60 / 0.83</td>
+<td>8.44 / 4.69</td>
+<td>17.3</td>
+<td>Ultra-lightweight Arabic recognition model based on PP-OCRv3, supporting Arabic and numeric recognition</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+cyrillic_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>94.28</td>
+<td>3.56 / 0.79</td>
+<td>8.22 / 2.76</td>
+<td>8.7</td>
+<td>Ultra-lightweight Cyrillic recognition model based on PP-OCRv3, supporting Cyrillic and numeric recognition</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+devanagari_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>96.44</td>
+<td>3.60 / 0.78</td>
+<td>6.95 / 2.87</td>
+<td>8.7</td>
+<td>Ultra-lightweight Devanagari recognition model based on PP-OCRv3, supporting Devanagari and numeric recognition</td>
+</tr>
+</table>
+</details>
+</details>
+
+<details>
+<summary><strong>Test Environment Details:</strong></summary>
+
+  <ul>
+      <li><b>Performance Test Environment</b>
+          <ul>
+            <li><strong>Test Datasets:
+             </strong>
+                <ul>
+                  <li>Document Image Orientation Classification Model: PaddleX in-house dataset covering ID cards and documents, with 1,000 images.</li>
+                  <li>Text Image Correction Model: <a href="https://www3.cs.stonybrook.edu/~cvl/docunet.html">DocUNet</a>.</li>
+                  <li>Text Detection Model: PaddleOCR in-house Chinese dataset covering street views, web images, documents, and handwriting, with 500 images for detection.</li>
+                  <li>Chinese Recognition Model: PaddleOCR in-house Chinese dataset covering street views, web images, documents, and handwriting, with 11,000 images for recognition.</li>
+                  <li>ch_SVTRv2_rec: <a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR Algorithm Challenge - Task 1: OCR End-to-End Recognition</a> A-set evaluation data.</li>
+                  <li>ch_RepSVTR_rec: <a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR Algorithm Challenge - Task 1: OCR End-to-End Recognition</a> B-set evaluation data.</li>
+                  <li>English Recognition Model: PaddleX in-house English dataset.</li>
+                  <li>Multilingual Recognition Model: PaddleX in-house multilingual dataset.</li>
+                  <li>Text Line Orientation Classification Model: PaddleX in-house dataset covering ID cards and documents, with 1,000 images.</li>
+                </ul>
+             </li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>Inference Mode Description</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+       <tr>
+            <th>Mode</th>
+            <th>GPU Configuration</th>
+            <th>CPU Configuration</th>
+            <th>Acceleration Techniques</th>
+       </tr>
+    </thead>
+    <tbody>
+       <tr>
+           <td>Standard Mode</td>
+           <td>FP32 Precision / No TRT Acceleration</td>
+           <td>FP32 Precision / 8 Threads</td>
+           <td><code>paddle_static</code></td>
+       </tr>
+       <tr>
+           <td>High-Performance Mode</td>
+           <td>Optimal combination of precision types and acceleration strategies</td>
+           <td>FP32 Precision / 8 Threads</td>
+           <td>Optimal backend selection (Paddle/OpenVINO/TRT, etc.)</td>
+       </tr>
+    </tbody>
+</table>
+
+</details>
+
+<br />
+<b>If you prioritize model accuracy, choose models with higher accuracy; if inference speed is critical, select faster models; if model size matters, opt for smaller models.</b>
+
+## 2. Quick Start  
+
+Before using the general OCR pipeline locally, install the dependencies for your chosen inference engine first, then install the `paddleocr` package. See the [Installation Guide](../installation.en.md) for the full procedure. After installation, you can use the command line or Python integration.
+
+### 2.0 Environment Preparation
+
+#### 2.0.1 Install the inference engine
+
+- If you use the local `paddle_static` inference engine, install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+- If you use the `transformers` inference engine, install the required dependencies by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md).
+- If you use the `onnxruntime` inference engine, install the required dependencies by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md).
+
+#### 2.0.2 Install paddleocr
+
+```bash
+# Install the basic package (OCR only)
+pip install paddleocr
+
+# Install the full package (all features)
+pip install "paddleocr[all]"
+```
+
+#### 2.0.3 Environment Verification
+
+```python
+# Verify that PaddleOCR is installed successfully
+import paddleocr
+print(f"PaddleOCR version: {paddleocr.__version__}")
+
+# If you use the local paddle_static inference engine, you can further verify PaddlePaddle and GPU availability
+import paddle
+print(f"Paddle version: {paddle.__version__}")
+print(f"GPU available: {paddle.is_compiled_with_cuda()}")
+print(f"GPU count: {paddle.device.cuda.device_count()}")
+
+# If you use the transformers inference engine, you can further verify the transformers dependency
+import transformers
+print(f"Transformers version: {transformers.__version__}")
+```
+
+#### 2.0.4 Common Installation Issues
+Please note: If you encounter issues such as the program becoming unresponsive, unexpected program termination, running out of memory resources, or extremely slow inference during execution, please try adjusting the configuration according to the documentation, such as disabling unnecessary features or using lighter-weight models.
+
+### 2.1 Command Line  
+
+Run a single command to quickly test the OCR pipeline.  Before running the code below, please download the [example image](https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/pipelines/general_formula_recognition_001.png) locally:  
+
+```bash
+paddleocr ocr -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False \
+    --use_textline_orientation False \
+    --save_path ./output \
+    --device gpu:0
+
+# Use PP-OCRv4 model by --ocr_version PP-OCRv4
+paddleocr ocr -i ./general_ocr_002.png --ocr_version PP-OCRv4
+
+```
+
+The examples above use the local `paddle_static` inference engine by default. To run them, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr ocr -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False \
+    --use_textline_orientation False \
+    --save_path ./output \
+    --device gpu:0 \
+    --engine transformers
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following command:
+
+```bash
+# Use the onnxruntime engine for inference
+paddleocr ocr -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False \
+    --use_textline_orientation False \
+    --save_path ./output \
+    --device gpu:0 \
+    --engine onnxruntime
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+<details><summary><b>Command line supports more parameter settings. Click to expand for detailed instructions on command line parameters.</b></summary>
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Parameter Description</th>
+<th>Parameter Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted, required.<br/>
+<b>Description:</b> 
+Local path of an image file or PDF file: <code>/root/data/img.jpg</code>; <b>URL link</b>, such as the network URL of an image file or PDF file: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png">Example</a>; <b>Local directory</b>, which must contain images to be predicted, such as the local path: <code>/root/data/</code> (currently, predicting PDFs in a directory is not supported; PDFs need to specify the exact file path).
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td><b>Meaning:</b>Path to save inference result files. <br/>
+<b>Description:</b>
+If not set, inference results will not be saved locally.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>Name of the document orientation classification model.<br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the document orientation classification model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>Meaning:</b>Name of the text image unwarping model. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text image unwarping model.<br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the text detection model.<br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text detection model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_name</code></td>
+<td><b>Meaning:</b>Name of the text line orientation model. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text line orientation model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the text line orientation model. <br/>
+<b>Description:</b> 
+If not set, the default batch size will be <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the text recognition model. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text recognition model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the text recognition model. <br/>
+<b>Description:</b> 
+If not set, the default batch size will be <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to load and use the document orientation classification module. <br/>
+<b>Description:</b> 
+If not set, the pipeline's initialized value for this parameter (defaults to <code>True</code>) will be used.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to load and use the text image unwarping module. <br/>
+<b>Description:</b> 
+If not set, the pipeline's initialized value for this parameter (defaults to <code>True</code>) will be used.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>Meaning:</b>Whether to load and use the text line orientation module. <br/>
+<b>Description:</b> 
+If not set, the pipeline's initialized value for this parameter (defaults to <code>True</code>) will be used.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limitation for text detection. <br/>
+<b>Description:</b> 
+Any integer greater than <code>0</code>. If not set, the pipeline's initialized value for this parameter (defaults to <code>64</code>) will be used.
+</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>Meaning:</b>Type of side length limit for text detection. <br/>
+<b>Description:</b> 
+Supports <code>min</code> and <code>max</code>. <code>min</code> means ensuring the shortest side of the image is not smaller than <code>det_limit_side_len</code>, and <code>max</code> means ensuring the longest side of the image is not larger than <code>limit_side_len</code>. If not set, the pipeline's initialized value for this parameter (defaults to <code>min</code>) will be used.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>Meaning:</b>Pixel threshold for text detection. In the output probability map, pixels with scores higher than this threshold will be considered text pixels.<br/>
+<b>Description:</b>
+Any floating-point number greater than <code>0</code>. If not set, the pipeline's initialized value for this parameter (defaults to <code>0.3</code>) will be used.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>Meaning:</b>Text detection box threshold.<br/>
+<b>Description:</b> 
+If the average score of all pixels within the detected result boundary is higher than this threshold, the result will be considered a text region.
+Any floating-point number greater than <code>0</code>. If not set, the pipeline's initialized value for this parameter (defaults to <code>0.6</code>) will be used.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Text detection expansion coefficient. This method is used to expand the text region—the larger the value, the larger the expanded area. <br/>
+<b>Description:</b>
+Any floating-point number greater than <code>0</code>. If not set, the pipeline's initialized value for this parameter (defaults to <code>2.0</code>) will be used.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_input_shape</code></td>
+<td><b>Meaning:</b>Input shape for text detection, you can set three values to represent C, H, and W.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Text recognition threshold. Text results with scores higher than this threshold will be retained.<br/>
+<b>Description:</b>
+Any floating-point number greater than <code>0</code>
+. If not set, the pipeline's initialized value for this parameter (defaults to <code>0.0</code>, i.e., no threshold) will be used.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>return_word_box</code></td>
+<td><b>Meaning:</b>Whether to return word-level boxes.<br/>
+<b>Description:</b> 
+If not set, the pipeline's initialized value for this parameter (defaults to <code>False</code>) will be used.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_rec_input_shape</code></td>
+<td><b>Meaning:</b>Input shape for text recognition.</td>
+<td><code>tuple</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>lang</code></td>
+<td><b>Meaning:</b>OCR model language to use.<br/>
+<b>Description:</b>
+The table in the appendix lists all the supported languages.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>ocr_version</code></td>
+<td><b>Meaning:</b>Version of OCR models.<br/>
+<b>Description:</b>
+<ul>
+<li><b>PP-OCRv6</b>: Use PP-OCRv6 series models;
+<li><b>PP-OCRv5</b>: Use PP-OCRv5 series models;
+<li><b>PP-OCRv4</b>: Use PP-OCRv4 series models;
+<li><b>PP-OCRv3</b>: Use PP-OCRv3 series models.</li>
+</ul>
+Please note that not every <code>ocr_version</code> supports all <code>lang</code> options. Please refer to the correspondence table in the appendix for details.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+
+<td><code>device</code></td>
+<td><b>Meaning:</b>Device for inference. <br/>
+<b>Description:</b>
+Supports specifying a specific card number:
+<ul>
+<li><b>CPU</b>: <code>cpu</code> indicates using CPU for inference;</li>
+<li><b>GPU</b>: <code>gpu:0</code> indicates using the 1st GPU for inference;</li>
+<li><b>NPU</b>: <code>npu:0</code> indicates using the 1st NPU for inference;</li>
+<li><b>XPU</b>: <code>xpu:0</code> indicates using the 1st XPU for inference;</li>
+<li><b>MLU</b>: <code>mlu:0</code> indicates using the 1st MLU for inference;</li>
+<li><b>DCU</b>: <code>dcu:0</code> indicates using the 1st DCU for inference;</li>
+<li><b>MetaX GPU</b>: <code>metax_gpu:0</code> indicates using the 1st MetaX GPU for inference;</li>
+<li><b>Iluvatar GPU</b>: <code>iluvatar_gpu:0</code> indicates using the 1st Iluvatar GPU for inference;</li>
+</ul>If not set, the pipeline initialized value for this parameter will be used. During initialization, the local GPU device 0 will be preferred; if unavailable, the CPU device will be used.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to enable the TensorRT subgraph engine of Paddle Inference.<br/>
+<b>Description:</b>
+If the model does not support TensorRT acceleration, acceleration will not be used even if this flag is set.<br/>
+For CUDA 11.8 versions of PaddlePaddle, the compatible TensorRT version is 8.x (x>=6). TensorRT 8.6.1.6 is recommended.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b> Computation precision, such as <code>fp32</code> or <code>fp16</code>.</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b> Whether to enable MKL-DNN accelerated inference.<br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support MKL-DNN acceleration, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b> MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads used for inference on CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b> Path to the PaddleX pipeline configuration file.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+</tbody>
+</table>
+
+<b>The following parameters have been deprecated in version 2.x and are listed here for the convenience of users of previous versions.</b>
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Parameter Description</th>
+<th>Parameter Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>det_model_dir</code></td>
+<td><b>Description:</b>Deprecated. Please refer <code>text_detection_model_dir</code> , they cannot be specified simultaneously with the new parameters.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>det_limit_side_len</code></td>
+<td><b>Description:</b>Deprecated. Please refer <code>text_det_limit_side_len</code> , they cannot be specified simultaneously with the new parameters.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>det_limit_type</code></td>
+<td><b>Description:</b>Deprecated. Please refer <code>text_det_limit_type</code> , they cannot be specified simultaneously with the new parameters.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>det_db_thresh</code></td>
+<td><b>Description:</b>Deprecated. Please refer <code>text_det_thresh</code> , they cannot be specified simultaneously with the new parameters.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>det_db_box_thresh</code></td>
+<td><b>Description:</b>Deprecated. Please refer <code>text_det_box_thresh</code> , they cannot be specified simultaneously with the new parameters.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>det_db_unclip_ratio</code></td>
+<td>Deprecated. Please refer <code>text_det_unclip_ratio</code> , they cannot be specified simultaneously with the new parameters.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>rec_model_dir</code></td>
+<td><b>Description:</b>Deprecated. Please refer <code>text_recognition_model_dir</code> , they cannot be specified simultaneously with the new parameters.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>rec_batch_num</code></td>
+<td><b>Description:</b>Deprecated. Please refer <code>text_recognition_batch_size</code> , they cannot be specified simultaneously with the new parameters.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_angle_cls</code></td>
+<td><b>Description:</b>Deprecated. Please refer <code>use_textline_orientation</code> , they cannot be specified simultaneously with the new parameters.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>cls_model_dir</code></td>
+<td><b>Description:</b>Deprecated. Please refer <code>textline_orientation_model_dir</code> , they cannot be specified simultaneously with the new parameters.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>cls_batch_num</code></td>
+<td><b>Description:</b>Deprecated. Please refer <code>textline_orientation_batch_size</code> , they cannot be specified simultaneously with the new parameters.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+</tbody>
+</table>
+</details>
+<br/>
+
+Results are printed to the terminal:  
+
+```bash
+{'res': {'input_path': './general_ocr_002.png', 'page_index': None, 'model_settings': {'use_doc_preprocessor': True, 'use_textline_orientation': False}, 'doc_preprocessor_res': {'input_path': None, 'page_index': None, 'model_settings': {'use_doc_orientation_classify': False, 'use_doc_unwarping': False}, 'angle': -1}, 'dt_polys': array([[[  3,  10],
+        ...,
+        [  4,  30]],
+
+       ...,
+
+       [[ 99, 456],
+        ...,
+        [ 99, 479]]], dtype=int16), 'text_det_params': {'limit_side_len': 736, 'limit_type': 'min', 'thresh': 0.3, 'max_side_limit': 4000, 'box_thresh': 0.6, 'unclip_ratio': 1.5}, 'text_type': 'general', 'textline_orientation_angles': array([-1, ..., -1]), 'text_rec_score_thresh': 0.0, 'rec_texts': ['www.997700', '', 'Cm', '登机牌', 'BOARDING', 'PASS', 'CLASS', '序号SERIAL NO.', '座位号', 'SEAT NO.', '航班FLIGHT', '日期DATE', '舱位', '', 'W', '035', '12F', 'MU2379', '03DEc', '始发地', 'FROM', '登机口', 'GATE', '登机时间BDT', '目的地TO', '福州', 'TAIYUAN', 'G11', 'FUZHOU', '身份识别IDNO.', '姓名NAME', 'ZHANGQIWEI', '票号TKT NO.', '张祺伟', '票价FARE', 'ETKT7813699238489/1', '登机口于起飞前10分钟关闭 GATESCL0SE10MINUTESBEFOREDEPARTURETIME'], 'rec_scores': array([0.67634439, ..., 0.97416091]), 'rec_polys': array([[[  3,  10],
+        ...,
+        [  4,  30]],
+
+       ...,
+
+       [[ 99, 456],
+        ...,
+        [ 99, 479]]], dtype=int16), 'rec_boxes': array([[  3, ...,  30],
+       ...,
+       [ 99, ..., 479]], dtype=int16)}}
+```
+
+If `save_path` is specified, the visualization results will be saved under `save_path`. The visualization output is shown below:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/ocr/03.png"/>
+
+### 2.2 Python Script Integration  
+
+The command-line method is for quick testing. For project integration, you can achieve OCR inference with just a few lines of code:  
+
+```python
+from paddleocr import PaddleOCR
+
+# Uses PP-OCRv6 models by default
+ocr = PaddleOCR(
+    use_doc_orientation_classify=False, # Disables document orientation classification model via this parameter
+    use_doc_unwarping=False, # Disables text image rectification model via this parameter
+    use_textline_orientation=False, # Disables text line orientation classification model via this parameter
+)
+# ocr = PaddleOCR(lang="en") # Uses English model by specifying language parameter
+# ocr = PaddleOCR(ocr_version="PP-OCRv5") # Switches to PP-OCRv5 version via ocr_version parameter
+# ocr = PaddleOCR(ocr_version="PP-OCRv4") # Switches to PP-OCRv4 version via ocr_version parameter
+# ocr = PaddleOCR(device="gpu") # Enables GPU acceleration for model inference via device parameter
+# ocr = PaddleOCR(
+#     text_detection_model_name="PP-OCRv5_mobile_det",
+#     text_recognition_model_name="PP-OCRv5_mobile_rec",
+#     use_doc_orientation_classify=False,
+#     use_doc_unwarping=False,
+#     use_textline_orientation=False,
+# ) # Switch to PP-OCRv5 mobile models
+result = ocr.predict("./general_ocr_002.png")  
+for res in result:  
+    res.print()  
+    res.save_to_img("output")  
+    res.save_to_json("output")  
+```  
+
+In the above Python script, the following steps are performed:
+
+<details><summary>(1) Instantiate the OCR pipeline object via <code>PaddleOCR()</code>, with specific parameter descriptions as follows:</summary>
+
+<table>
+  <thead>
+   <tr>
+      <th>Parameter</th>
+      <th>Parameter Description</th>
+      <th>Parameter Type</th>
+      <th>Default Value</th>
+   </tr>
+  </thead>
+  <tbody>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>Name of the document orientation classification model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the document orientation classification model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>Meaning:</b>Name of the text image unwarping model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text image unwarping model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the text detection model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text detection model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_name</code></td>
+<td><b>Meaning:</b>Name of the text line orientation model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text line orientation model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the text line orientation model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the default batch size will be <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the text recognition model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text recognition model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the text recognition model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the default batch size will be <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the text recognition model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text recognition model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the text recognition model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the default batch size will be <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to load and use the document orientation classification module. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the pipeline's initialized value for this parameter (defaults to <code>True</code>) will be used.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to load and use the text image unwarping module. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the pipeline's initialized value for this parameter (defaults to <code>True</code>) will be used.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>Meaning:</b>Whether to load and use the text line orientation module. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the pipeline's initialized value for this parameter (defaults to <code>True</code>) will be used.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limitation for text detection. <br/>
+<b>Description:</b>
+ <ul>
+ <li><b>int</b>: Any integer greater than <code>0</code>;</li>
+ <li><b>None</b>: If set to <code>None</code>, the pipeline's initialized value for this parameter (defaults to <code>64</code>) will be used.</li>
+ </ul>
+ </td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>Meaning:</b>Type of side length limit for text detection. <br/>
+<b>Description:</b>
+ <ul>
+ <li><b>str</b>: Supports <code>min</code> and <code>max</code>, where <code>min</code> means ensuring the shortest side of the image is not smaller than <code>det_limit_side_len</code>, and <code>max</code> means ensuring the longest side of the image is not larger than <code>limit_side_len</code>;</li>
+ <li><b>None</b>: If set to <code>None</code>, the pipeline's initialized value for this parameter (defaults to <code>min</code>) will be used.</li>
+ </ul>
+ </td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>Meaning:</b>Pixel threshold for text detection. Pixels with scores higher than this threshold in the output probability map will be considered text pixels. <br/>
+<b>Description:</b>
+ <ul>
+ <li><b>float</b>: Any floating-point number greater than <code>0</code>;
+<li><b>None</b>: If set to <code>None</code>, the pipeline's initialized value for this parameter (defaults to <code>0.3</code>) will be used.</li></li></ul>
+ </td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>Meaning:</b>Box threshold for text detection. A detection result will be considered a text region if the average score of all pixels within the bounding box is higher than this threshold. <br/>
+<b>Description:</b>
+ <ul>
+ <li><b>float</b>: Any floating-point number greater than <code>0</code>;
+ <li><b>None</b>: If set to <code>None</code>, the pipeline's initialized value for this parameter (defaults to <code>0.6</code>) will be used.</li></li></ul>
+ </td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Dilation coefficient for text detection. This method is used to dilate the text region, and the larger this value, the larger the dilated area. <br/>
+<b>Description:</b>
+ <ul>
+ <li><b>float</b>: Any floating-point number greater than <code>0</code>;
+<li><b>None</b>: If set to <code>None</code>, the pipeline's initialized value for this parameter (defaults to <code>2.0</code>) will be used.</li></li></ul>
+ </td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_input_shape</code></td>
+<td><b>Meaning:</b>Input shape for text detection.</td>
+<td><code>tuple</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Recognition score threshold for text. Text results with scores higher than this threshold will be retained. <br/>
+<b>Description:</b>
+ <ul>
+<li><b>float</b>: Any floating-point number greater than <code>0</code>;
+    <li><b>None</b>: If set to <code>None</code>, the pipeline's initialized value for this parameter (defaults to <code>0.0</code>, i.e., no threshold) will be used.</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_input_shape</code></td>
+<td><b>Meaning:</b>Input shape for text recognition.</td>
+<td><code>tuple</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>lang</code></td>
+<td><b>Meaning:</b>OCR model language to use.<br/>
+<b>Description:</b>
+The table in the appendix lists all the supported languages.
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>ocr_version</code></td>
+<td><b>Meaning:</b>Version of OCR models. <br/>
+<b>Description:</b>
+<ul>
+<li><b>PP-OCRv6</b>: Use PP-OCRv6 series models;</li>
+<li><b>PP-OCRv5</b>: Use PP-OCRv5 series models;</li>
+<li><b>PP-OCRv4</b>: Use PP-OCRv4 series models;</li>
+<li><b>PP-OCRv3</b>: Use PP-OCRv3 series models.</li>
+</ul>
+Please note that not every <code>ocr_version</code> supports all <code>lang</code> options. Please refer to the correspondence table in the appendix for details.
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>Device for inference. <br/>
+<b>Description:</b>
+Supports specifying a specific card number:
+<ul>
+<li><b>CPU</b>: e.g., <code>cpu</code> for CPU inference;</li>
+<li><b>GPU</b>: e.g., <code>gpu:0</code> for inference on the 1st GPU;</li>
+<li><b>NPU</b>: e.g., <code>npu:0</code> for inference on the 1st NPU;</li>
+<li><b>XPU</b>: e.g., <code>xpu:0</code> for inference on the 1st XPU;</li>
+<li><b>MLU</b>: e.g., <code>mlu:0</code> for inference on the 1st MLU;</li>
+<li><b>DCU</b>: e.g., <code>dcu:0</code> for inference on the 1st DCU;</li>
+<li><b>MetaX GPU</b>: e.g., <code>metax_gpu:0</code> for inference on the 1st MetaX GPU;</li>
+<li><b>Iluvatar GPU</b>: e.g., <code>iluvatar_gpu:0</code> for inference on the 1st Iluvatar GPU;</li>
+<li><b>None</b>: If set to <code>None</code>, the pipeline initialized value for this parameter will be used. During initialization, the local GPU device 0 will be preferred; if unavailable, the CPU device will be used.</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b>Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b>Whether to use the Paddle Inference TensorRT subgraph engine. If the model does not support acceleration through TensorRT, setting this flag will not enable acceleration.<br/>
+<b>Description:</b>
+For Paddle with CUDA version 11.8, the compatible TensorRT version is 8.x (x>=6), and it is recommended to install TensorRT 8.6.1.6.<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b>Computational precision, such as <code>"fp32"</code>, <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b>Whether to enable MKL-DNN acceleration for inference.<br/>
+<b>Description:</b> 
+If MKL-DNN is unavailable or the model does not support it, acceleration will not be used even if this flag is set.</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b>MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b>Number of threads used for CPU inference.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b>Path to the PaddleX pipeline configuration file.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details><summary>(2) Invoke the <code>predict()</code> method of the OCR pipeline object for inference prediction, which returns a results list. Additionally, the pipeline provides the <code>predict_iter()</code> method. Both methods are completely consistent in parameter acceptance and result return, except that <code>predict_iter()</code> returns a <code>generator</code>, which can process and obtain prediction results incrementally, suitable for handling large datasets or scenarios where memory saving is desired. You can choose to use either of these two methods according to actual needs. The following are the parameters and descriptions of the <code>predict()</code> method:</summary>
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Parameter Description</th>
+<th>Parameter Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted, supporting multiple input types, required.<br/>
+<b>Description:</b>
+<ul>
+<li><b>Python Var</b>: Image data represented by <code>numpy.ndarray</code>;</li>
+<li><b>str</b>: Local path of an image file or PDF file: <code>/root/data/img.jpg</code>; <b>URL link</b>, such as the network URL of an image file or PDF file: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png">example</a>; <b>local directory</b>, which needs to contain images to be predicted, such as the local path: <code>/root/data/</code> (currently, predicting PDF files in the directory is not supported; PDF files need to specify the specific file path);</li>
+<li><b>list</b>: List elements must be of the above types, such as <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"]</code>.</li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to use the document orientation classification module during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to use the text image unwarping module during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>Meaning:</b>Whether to use the text line orientation classification module during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters.<br/> 
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<td><code>text_det_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</table>
+</details>
+
+<details><summary>(3) Process the prediction results. The prediction result of each sample is a corresponding Result object, which supports operations of printing, saving as an image, and saving as a <code>json</code> file:</summary>
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Method Description</th>
+<th>Parameter</th>
+<th>Parameter Type</th>
+<th>Parameter Description</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">Print the results to the terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output content with <code>JSON</code> indentation.</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data and make it more readable, only valid when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non-<code>ASCII</code> characters as <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> retains the original characters, only valid when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">Save the results as a json-formatted file.</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>File path to save. When it is a directory, the saved file name will be consistent with the input file type name.</td>
+<td>No default</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data and make it more readable, only valid when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non-<code>ASCII</code> characters as <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> retains the original characters, only valid when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save the results as an image-formatted file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>File path to save, supporting directory or file path.</td>
+<td>No default</td>
+</tr>
+</table>
+
+<ul>
+    <li>Calling the <code>print()</code> method will print the results to the terminal. The content printed to the terminal is explained as follows:
+        <ul>
+            <li><code>input_path</code>: <code>(str)</code> Input path of the image to be predicted</li>
+            <li><code>page_index</code>: <code>(Union[int, None])</code> If the input is a PDF file, it indicates which page of the PDF it is; otherwise, it is <code>None</code></li>
+            <li><code>model_settings</code>: <code>(Dict[str, bool])</code> Model parameters configured for the pipeline
+                <ul>
+                    <li><code>use_doc_preprocessor</code>: <code>(bool)</code> Control whether to enable the document preprocessing sub-pipeline</li>
+                    <li><code>use_textline_orientation</code>: <code>(bool)</code> Control whether to enable the text line orientation classification function</li>
+                </ul>
+            </li>
+            <li><code>doc_preprocessor_res</code>: <code>(Dict[str, Union[str, Dict[str, bool], int]])</code> Output results of the document preprocessing sub-pipeline. Only exists when <code>use_doc_preprocessor=True</code>
+                <ul>
+                    <li><code>input_path</code>: <code>(Union[str, None])</code> Image path accepted by the image preprocessing sub-pipeline. When the input is <code>numpy.ndarray</code>, it is saved as <code>None</code></li>
+                    <li><code>model_settings</code>: <code>(Dict)</code> Model configuration parameters of the preprocessing sub-pipeline
+                        <ul>
+                            <li><code>use_doc_orientation_classify</code>: <code>(bool)</code> Control whether to enable document orientation classification</li>
+                            <li><code>use_doc_unwarping</code>: <code>(bool)</code> Control whether to enable text image unwarping</li>
+                        </ul>
+                    </li>
+                    <li><code>angle</code>: <code>(int)</code> Prediction result of document orientation classification. When enabled, the values are [0,1,2,3], corresponding to [0°,90°,180°,270°]; when disabled, it is -1</li>
+                </ul>
+            </li>
+            <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> List of text detection polygon boxes. Each detection box is represented by a numpy array of 4 vertex coordinates, with the array shape being (4, 2) and the data type being int16</li>
+            <li><code>dt_scores</code>: <code>(List[float])</code> List of confidence scores for text detection boxes</li>
+            <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> Configuration parameters for the text detection module
+                <ul>
+                    <li><code>limit_side_len</code>: <code>(int)</code> Side length limit value during image preprocessing</li>
+                    <li><code>limit_type</code>: <code>(str)</code> Processing method for side length limits</li>
+                    <li><code>thresh</code>: <code>(float)</code> Confidence threshold for text pixel classification</li>
+                    <li><code>box_thresh</code>: <code>(float)</code> Confidence threshold for text detection boxes</li>
+                    <li><code>unclip_ratio</code>: <code>(float)</code> Dilation coefficient for text detection boxes</li>
+                    <li><code>text_type</code>: <code>(str)</code> Type of text detection, currently fixed as "general"</li>
+                </ul>
+            </li>
+            <li><code>textline_orientation_angles</code>: <code>(List[int])</code> Prediction results of text line orientation classification. When enabled, actual angle values are returned (e.g., [0,0,1]); when disabled, [-1,-1,-1] is returned</li>
+            <li><code>text_rec_score_thresh</code>: <code>(float)</code> Filtering threshold for text recognition results</li>
+            <li><code>rec_texts</code>: <code>(List[str])</code> List of text recognition results, containing only texts with confidence scores exceeding <code>text_rec_score_thresh</code></li>
+            <li><code>rec_scores</code>: <code>(List[float])</code> List of text recognition confidence scores, filtered by <code>text_rec_score_thresh</code></li>
+            <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> List of text detection boxes filtered by confidence, in the same format as <code>dt_polys</code></li>
+            <li><code>rec_boxes</code>: <code>(numpy.ndarray)</code> Array of rectangular bounding boxes for detection boxes, with shape (n, 4) and dtype int16. Each row represents the [x_min, y_min, x_max, y_max] coordinates of a rectangular box, where (x_min, y_min) is the top-left coordinate and (x_max, y_max) is the bottom-right coordinate</li>
+        </ul>
+    </li>
+    <li>Calling the <code>save_to_json()</code> method will save the above content to the specified <code>save_path</code>. If a directory is specified, the save path will be <code>save_path/{your_img_basename}_res.json</code>. If a file is specified, it will be saved directly to that file. Since json files do not support saving numpy arrays, <code>numpy.array</code> types will be converted to list form.</li>
+    <li>Calling the <code>save_to_img()</code> method will save the visualization results to the specified <code>save_path</code>. If a directory is specified, the save path will be <code>save_path/{your_img_basename}_ocr_res_img.{your_img_extension}</code>. If a file is specified, it will be saved directly to that file. (The pipeline usually generates many result images, so it is not recommended to directly specify a specific file path, as multiple images will be overwritten, leaving only the last one.)</li>
+</ul>
+
+<p>Additionally, you can also obtain the visualized image with results and prediction results through attributes, as follows:</p>
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Attribute Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">Get the prediction results in <code>json</code> format</td>
+</tr>
+<tr>
+<td rowspan="2"><code>img</code></td>
+<td rowspan="2">Get the visualized image in <code>dict</code> format</td>
+</tr>
+</table>
+
+<ul>
+    <li>The prediction results obtained by the <code>json</code> attribute are in dict format, and the content is consistent with that saved by calling the <code>save_to_json()</code> method.</li>
+    <li>The <code>img</code> attribute returns a dictionary-type result. The keys are <code>ocr_res_img</code> and <code>preprocessed_img</code>, with corresponding values being two <code>Image.Image</code> objects: one for displaying the visualized image of OCR results and the other for displaying the visualized image of image preprocessing. If the image preprocessing submodule is not used, only <code>ocr_res_img</code> will be included in the dictionary.</li>
+</ul>
+
+</details>
+
+## 3. Development Integration/Deployment
+
+If the general OCR pipeline meets your requirements for inference speed and accuracy, you can proceed directly with development integration/deployment.
+
+If you need to apply the general OCR pipeline directly in your Python project, you can refer to the sample code in [2.2 Python Script Integration](#22-python-script-integration).
+
+Additionally, PaddleOCR provides two other deployment methods, detailed as follows:
+
+🚀 **High-Performance Inference**: In real-world production environments, many applications have stringent performance requirements (especially for response speed) to ensure system efficiency and smooth user experience. To address this, PaddleOCR offers high-performance inference capabilities, which deeply optimize model inference and pre/post-processing to achieve significant end-to-end speed improvements. For detailed high-performance inference workflows, refer to the [High-Performance Inference Guide](../inference_deployment/local_inference/high_performance_inference.en.md).
+
+☁️ **Service Deployment**: Service deployment is a common form of deployment in production environments. By encapsulating inference functionality as a service, clients can access these services via network requests to obtain inference results. For detailed pipeline service deployment workflows, refer to the [Service Deployment Guide](../inference_deployment/serving/serving.en.md).
+
+Below are the API reference for basic service deployment and examples of multi-language service calls:
+
+<details><summary>API Reference</summary>
+<p>For the main operations provided by the service:</p>
+<ul>
+<li>The HTTP request method is POST.</li>
+<li>Both the request body and response body are JSON data (JSON objects).</li>
+<li>When the request is processed successfully, the response status code is <code>200</code>, and the response body has the following attributes:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>UUID of the request.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Fixed as <code>0</code>.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error message. Fixed as <code>"Success"</code>.</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>Operation result.</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request fails, the response body has the following attributes:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>UUID of the request.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Same as the response status code.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error message.</td>
+</tr>
+</tbody>
+</table>
+<p>The main operations provided by the service are as follows:</p>
+<ul>
+<li><b><code>infer</code></b></li>
+</ul>
+<p>Obtain OCR results for an image.</p>
+<p><code>POST /ocr</code></p>
+<ul>
+<li>The request body has the following attributes:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Description</th>
+<th>Required</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>A server-accessible URL to image files (including TIFF; multi-page TIFF is processed page by page) or PDF file, or the Base64-encoded content of such a file. By default, for PDF or multi-page TIFF files with more than 10 pages, only the first 10 pages are processed.<br /> To remove the page limit, add the following configuration to the pipeline config file:
+<pre><code>Serving:
+  extra:
+    max_num_input_imgs: null
+</code></pre>
+</td>
+<td>Yes</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>File type. <code>0</code> for PDF, <code>1</code> for image (including TIFF). If omitted, the type is inferred from the URL.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Refer to the <code>use_doc_orientation_classify</code> parameter in the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Refer to the <code>use_doc_unwarping</code> parameter in the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<tr>
+<td><code>useTextlineOrientation</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Refer to the <code>use_textline_orientation</code> parameter in the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>Refer to the <code>text_det_limit_side_len</code> parameter in the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Refer to the <code>text_det_limit_type</code> parameter in the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Refer to the <code>text_det_thresh</code> parameter in the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Refer to the <code>text_det_box_thresh</code> parameter in the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Refer to the <code>text_det_unclip_ratio</code> parameter in the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Refer to the <code>text_rec_score_thresh</code> parameter in the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>
+Whether to return the final visualization image and intermediate images during the processing.<br/>
+<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>If <code>true</code> is provided: return images.</li>
+<li>If <code>false</code> is provided: do not return any images.</li>
+<li>If this parameter is omitted from the request body, or if <code>null</code> is explicitly passed, the behavior will follow the value of <code>Serving.visualize</code> in the pipeline configuration.</li>
+</ul>
+<br/>
+For example, adding the following setting to the pipeline config file:<br/>
+<pre><code>Serving:
+  visualize: False
+</code></pre>
+will disable image return by default. This behavior can be overridden by explicitly setting the <code>visualize</code> parameter in the request.<br/>
+If neither the request body nor the configuration file is set (If <code>visualize</code> is set to <code>null</code> in the request and  not defined in the configuration file), the image is returned by default.
+</td>
+<td>No</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is successful, the <code>result</code> in the response body has the following attributes:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>ocrResults</code></td>
+<td><code>object</code></td>
+<td>OCR results. The array length is 1 (for image input) or the number of processed document pages (for PDF input). For PDF input, each element represents the result for a corresponding page.</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>Input data information.</td>
+</tr>
+</tbody>
+</table>
+<p>Image fields in the element schema below (e.g. <code>ocrImage</code>, <code>docPreprocessingImage</code>, <code>inputImage</code>) are returned inline as Base64 strings by default; when the server is configured to return URLs, those values become pre-signed URLs while the field types remain unchanged. See the "Returning Binary Content as URLs" section of the <a href="../inference_deployment/serving/serving.en.md">Serving Deployment Guide</a> for configuration.</p>
+<p>Each element in <code>ocrResults</code> is an <code>object</code> with the following attributes:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>A simplified version of the <code>res</code> field in the JSON output of the pipeline object's <code>predict</code> method, excluding <code>input_path</code> and <code>page_index</code>.</td>
+</tr>
+<tr>
+<td><code>ocrImage</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>OCR result image with detected text regions highlighted. JPEG format, Base64-encoded by default; returned as a pre-signed URL when URL-return mode is enabled.</td>
+</tr>
+<tr>
+<td><code>docPreprocessingImage</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Visualization of preprocessing results. JPEG format, Base64-encoded by default; returned as a pre-signed URL when URL-return mode is enabled.</td>
+</tr>
+<tr>
+<td><code>inputImage</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Input image. JPEG format, Base64-encoded by default; returned as a pre-signed URL when URL-return mode is enabled.</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details><summary>Multi-Language Service Call Examples</summary>
+
+<details>
+<summary>Python</summary>
+
+<pre><code class="language-python">import base64
+import requests
+
+API_URL = "http://localhost:8080/ocr"
+file_path = "./demo.jpg"
+
+with open(file_path, "rb") as file:
+    file_bytes = file.read()
+    file_data = base64.b64encode(file_bytes).decode("ascii")
+
+payload = {"file": file_data, "fileType": 1}
+
+response = requests.post(API_URL, json=payload)
+
+assert response.status_code == 200
+result = response.json()["result"]
+for i, res in enumerate(result["ocrResults"]):
+    print(res["prunedResult"])
+    ocr_img_path = f"ocr_{i}.jpg"
+    with open(ocr_img_path, "wb") as f:
+        f.write(base64.b64decode(res["ocrImage"]))
+    print(f"Output image saved at {ocr_img_path}")
+</code></pre></details>
+
+<details><summary>C++</summary>
+
+<pre><code class="language-cpp">#include &lt;iostream&gt;
+#include &lt;fstream&gt;
+#include &lt;vector&gt;
+#include &lt;string&gt;
+#include "cpp-httplib/httplib.h" // https://github.com/Huiyicc/cpp-httplib
+#include "nlohmann/json.hpp" // https://github.com/nlohmann/json
+#include "base64.hpp" // https://github.com/tobiaslocker/base64
+
+int main() {
+    httplib::Client client("localhost", 8080);  
+    const std::string filePath = "./demo.jpg"; 
+
+    std::ifstream file(filePath, std::ios::binary | std::ios::ate);
+    if (!file) {
+        std::cerr << "Error opening file." << std::endl;
+        return 1;
+    }
+
+    std::streamsize size = file.tellg();
+    file.seekg(0, std::ios::beg);
+    std::vector<char> buffer(size);
+
+    if (!file.read(buffer.data(), size)) {
+        std::cerr << "Error reading file." << std::endl;
+        return 1;
+    }
+
+    std::string bufferStr(buffer.data(), static_cast<size_t>(size));
+    std::string encodedFile = base64::to_base64(bufferStr);
+
+
+    nlohmann::json jsonObj;
+    jsonObj["file"] = encodedFile;
+    jsonObj["fileType"] = 1;  
+
+    auto response = client.Post("/ocr", jsonObj.dump(), "application/json");
+
+    if (response && response->status == 200) {
+        nlohmann::json jsonResponse = nlohmann::json::parse(response->body);
+        auto result = jsonResponse["result"];
+
+        if (!result.is_object() || !result["ocrResults"].is_array()) {
+            std::cerr << "Unexpected response structure." << std::endl;
+            return 1;
+        }
+
+        for (size_t i = 0; i < result["ocrResults"].size(); ++i) {
+            auto ocrResult = result["ocrResults"][i];
+            std::cout << ocrResult["prunedResult"] << std::endl;
+
+            std::string ocrImgPath = "ocr_" + std::to_string(i) + ".jpg";
+            std::string encodedImage = ocrResult["ocrImage"];
+            std::string decodedImage = base64::from_base64(encodedImage);
+
+            std::ofstream outputImage(ocrImgPath, std::ios::binary);
+            if (outputImage.is_open()) {
+                outputImage.write(decodedImage.c_str(), static_cast<std::streamsize>(decodedImage.size()));
+                outputImage.close();
+                std::cout << "Output image saved at " << ocrImgPath << std::endl;
+            } else {
+                std::cerr << "Unable to open file for writing: " << ocrImgPath << std::endl;
+            }
+        }
+    } else {
+        std::cerr << "Failed to send HTTP request." << std::endl;
+        if (response) {
+            std::cerr << "HTTP status code: " << response->status << std::endl;
+            std::cerr << "Response body: " << response->body << std::endl;
+        }
+        return 1;
+    }
+
+    return 0;
+}
+</code></pre></details>
+
+<details><summary>Java</summary>
+
+<pre><code class="language-java">import okhttp3.*;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.util.Base64;
+
+public class Main {
+    public static void main(String[] args) throws IOException {
+        String API_URL = "http://localhost:8080/ocr"; 
+        String imagePath = "./demo.jpg"; 
+
+        File file = new File(imagePath);
+        byte[] fileContent = java.nio.file.Files.readAllBytes(file.toPath());
+        String base64Image = Base64.getEncoder().encodeToString(fileContent);
+
+        ObjectMapper objectMapper = new ObjectMapper();
+        ObjectNode payload = objectMapper.createObjectNode();
+        payload.put("file", base64Image); 
+        payload.put("fileType", 1); 
+
+        OkHttpClient client = new OkHttpClient();
+        MediaType JSON = MediaType.get("application/json; charset=utf-8");
+	RequestBody body = RequestBody.create(JSON, payload.toString());
+
+        Request request = new Request.Builder()
+                .url(API_URL)
+                .post(body)
+                .build();
+
+        try (Response response = client.newCall(request).execute()) {
+            if (response.isSuccessful()) {
+                String responseBody = response.body().string();
+                JsonNode root = objectMapper.readTree(responseBody);
+                JsonNode result = root.get("result");
+
+                JsonNode ocrResults = result.get("ocrResults");
+                for (int i = 0; i < ocrResults.size(); i++) {
+                    JsonNode item = ocrResults.get(i);
+
+                    JsonNode prunedResult = item.get("prunedResult");
+                    System.out.println("Pruned Result [" + i + "]: " + prunedResult.toString());
+
+                    String ocrImageBase64 = item.get("ocrImage").asText();
+                    byte[] ocrImageBytes = Base64.getDecoder().decode(ocrImageBase64);
+                    String ocrImgPath = "ocr_result_" + i + ".jpg";
+                    try (FileOutputStream fos = new FileOutputStream(ocrImgPath)) {
+                        fos.write(ocrImageBytes);
+                        System.out.println("Saved OCR image to: " + ocrImgPath);
+                    }
+                }
+            } else {
+                System.err.println("Request failed with HTTP code: " + response.code());
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Go</summary>
+
+<pre><code class="language-go">package main
+
+import (
+    "bytes"
+    "encoding/base64"
+    "encoding/json"
+    "fmt"
+    "io/ioutil"
+    "net/http"
+)
+
+func main() {
+    API_URL := "http://localhost:8080/ocr"
+    filePath := "./demo.jpg"
+
+    fileBytes, err := ioutil.ReadFile(filePath)
+    if err != nil {
+        fmt.Printf("Error reading file: %v\n", err)
+        return
+    }
+    fileData := base64.StdEncoding.EncodeToString(fileBytes)
+
+    payload := map[string]interface{}{
+        "file":     fileData,
+        "fileType": 1,
+    }
+    payloadBytes, err := json.Marshal(payload)
+    if err != nil {
+        fmt.Printf("Error marshaling payload: %v\n", err)
+        return
+    }
+
+    client := &http.Client{}
+    req, err := http.NewRequest("POST", API_URL, bytes.NewBuffer(payloadBytes))
+    if err != nil {
+        fmt.Printf("Error creating request: %v\n", err)
+        return
+    }
+    req.Header.Set("Content-Type", "application/json")
+
+    res, err := client.Do(req)
+    if err != nil {
+        fmt.Printf("Error sending request: %v\n", err)
+        return
+    }
+    defer res.Body.Close()
+
+    if res.StatusCode != http.StatusOK {
+        fmt.Printf("Unexpected status code: %d\n", res.StatusCode)
+        return
+    }
+
+    body, err := ioutil.ReadAll(res.Body)
+    if err != nil {
+        fmt.Printf("Error reading response body: %v\n", err)
+        return
+    }
+
+    type OcrResult struct {
+        PrunedResult map[string]interface{} `json:"prunedResult"` 
+        OcrImage     *string                `json:"ocrImage"`     
+    }
+
+    type Response struct {
+        Result struct {
+            OcrResults []OcrResult `json:"ocrResults"`
+            DataInfo   interface{} `json:"dataInfo"` 
+        } `json:"result"`
+    }
+
+    var respData Response
+    if err := json.Unmarshal(body, &respData); err != nil {
+        fmt.Printf("Error unmarshaling response: %v\n", err)
+        return
+    }
+
+    for i, res := range respData.Result.OcrResults {
+        
+        if res.OcrImage != nil {
+            imgBytes, err := base64.StdEncoding.DecodeString(*res.OcrImage)
+            if err != nil {
+                fmt.Printf("Error decoding image %d: %v\n", i, err)
+                continue
+            }
+            
+            filename := fmt.Sprintf("ocr_%d.jpg", i)
+            if err := ioutil.WriteFile(filename, imgBytes, 0644); err != nil {
+                fmt.Printf("Error saving image %s: %v\n", filename, err)
+                continue
+            }
+            fmt.Printf("Output image saved at %s\n", filename)
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>C#</summary>
+
+<pre><code class="language-csharp">using System;
+using System.IO;
+using System.Net.Http;
+using System.Text;
+using System.Threading.Tasks;
+using Newtonsoft.Json.Linq;
+
+class Program
+{
+    static readonly string API_URL = "http://localhost:8080/ocr";
+    static readonly string inputFilePath = "./demo.jpg";
+
+    static async Task Main(string[] args)
+    {
+        var httpClient = new HttpClient();
+
+        byte[] fileBytes = File.ReadAllBytes(inputFilePath);
+        string fileData = Convert.ToBase64String(fileBytes);
+
+        var payload = new JObject
+        {
+            { "file", fileData },
+            { "fileType", 1 }
+        };
+        var content = new StringContent(payload.ToString(), Encoding.UTF8, "application/json");
+
+        HttpResponseMessage response = await httpClient.PostAsync(API_URL, content);
+        response.EnsureSuccessStatusCode();
+
+        string responseBody = await response.Content.ReadAsStringAsync();
+        JObject jsonResponse = JObject.Parse(responseBody);
+
+        JArray ocrResults = (JArray)jsonResponse["result"]["ocrResults"];
+        for (int i = 0; i < ocrResults.Count; i++)
+        {
+            var res = ocrResults[i];
+            Console.WriteLine($"[{i}] prunedResult:\n{res["prunedResult"]}");
+
+            string base64Image = res["ocrImage"]?.ToString();
+            if (!string.IsNullOrEmpty(base64Image))
+            {
+                string outputPath = $"ocr_{i}.jpg";
+                byte[] imageBytes = Convert.FromBase64String(base64Image);
+                File.WriteAllBytes(outputPath, imageBytes);
+                Console.WriteLine($"OCR image saved to {outputPath}");
+            }
+            else
+            {
+                Console.WriteLine($"OCR image at index {i} is null.");
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Node.js</summary>
+
+<pre><code class="language-js">const axios = require('axios');
+const fs = require('fs');
+const path = require('path');
+
+const API_URL = 'http://localhost:8080/layout-parsing';
+const imagePath = './demo.jpg';  
+const fileType = 1;             
+
+function encodeImageToBase64(filePath) {
+  const bitmap = fs.readFileSync(filePath);
+  return Buffer.from(bitmap).toString('base64');
+}
+
+const payload = {
+  file: encodeImageToBase64(imagePath),
+  fileType: fileType
+};
+
+axios.post(API_URL, payload)
+  .then(response => {
+    const results = response.data.result.layoutParsingResults;
+    results.forEach((res, index) => {
+      console.log(`\n[${index}] prunedResult:`);
+      console.log(res.prunedResult);
+
+      const outputImages = res.outputImages;
+      if (outputImages) {
+        Object.entries(outputImages).forEach(([imgName, base64Img]) => {
+          const imgPath = `${imgName}_${index}.jpg`;
+          fs.writeFileSync(imgPath, Buffer.from(base64Img, 'base64'));
+          console.log(`Output image saved at ${imgPath}`);
+        });
+      } else {
+        console.log(`[${index}] No outputImages.`);
+      }
+    });
+  })
+  .catch(error => {
+    console.error('Error during API request:', error.message || error);
+  });
+</code></pre></details>
+
+<details><summary>PHP</summary>
+
+<pre><code class="language-php">&lt;?php
+
+$API_URL = "http://localhost:8080/ocr"; 
+$image_path = "./demo.jpg"; 
+
+$image_data = base64_encode(file_get_contents($image_path));
+$payload = array(
+    "file" => $image_data,
+    "fileType" => 1 
+);
+
+$ch = curl_init($API_URL);
+curl_setopt($ch, CURLOPT_POST, true);
+curl_setopt($ch, CURLOPT_POSTFIELDS, json_encode($payload));
+curl_setopt($ch, CURLOPT_HTTPHEADER, array('Content-Type: application/json'));
+curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+$response = curl_exec($ch);
+curl_close($ch);
+
+$result = json_decode($response, true)["result"]["ocrResults"];
+
+foreach ($result as $i => $item) {
+    echo "[$i] prunedResult:\n";
+    print_r($item["prunedResult"]);
+
+    if (!empty($item["ocrImage"])) {
+        $output_img_path = "ocr_{$i}.jpg";
+        file_put_contents($output_img_path, base64_decode($item["ocrImage"]));
+        echo "OCR image saved at $output_img_path\n";
+    } else {
+        echo "No ocrImage found for item $i\n";
+    }
+}
+?&gt;
+</code></pre></details>
+</details>
+
+## 4. Custom Development
+If the default model weights provided by the General OCR Pipeline do not meet your expectations in terms of accuracy or speed for your specific scenario, you can leverage your own domain-specific or application-specific data to further fine-tune the existing models, thereby improving the recognition performance of the General OCR Pipeline in your use case.
+
+### 4.1 Model Fine-Tuning
+The general OCR pipeline consists of multiple modules. If the pipeline's performance does not meet expectations, the issue may stem from any of these modules. You can analyze poorly recognized images to identify the problematic module and refer to the corresponding fine-tuning tutorials in the table below for adjustments.
+
+<table>
+<thead>
+<tr>
+<th>Scenario</th>
+<th>Module to Fine-Tune</th>
+<th>Fine-Tuning Reference</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>Inaccurate whole-image rotation correction</td>
+<td>Document orientation classification module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate image distortion correction</td>
+<td>Text image unwarping module</td>
+<td>Fine-tuning not supported</td>
+</tr>
+<tr>
+<td>Inaccurate textline rotation correction</td>
+<td>Textline orientation classification module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/textline_orientation_classification.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Text detection misses</td>
+<td>Text detection module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/module_usage/text_detection.html#4-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Incorrect text recognition</td>
+<td>Text recognition module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/module_usage/text_recognition.html#v-secondary-development">Link</a></td>
+</tr>
+</tbody>
+</table>
+
+### 4.2 Model Deployment  
+
+After you complete fine-tuning training using a private dataset, you can obtain a local model weight file. You can then use the fine-tuned model weights by specifying the local model save path through parameters or by customizing the pipeline configuration file.
+
+#### 4.2.1 Specify the local model path through parameters
+
+When initializing the pipeline object, specify the local model path through parameters. Take the usage of the weights after fine-tuning the text detection model as an example, as follows:
+
+Command line mode:
+
+```bash
+# Specify the local model path via --text_detection_model_dir
+paddleocr ocr -i ./general_ocr_002.png --text_detection_model_dir your_det_model_path
+
+# Modify the model name via --text_detection_model_name
+paddleocr ocr -i ./general_ocr_002.png --text_detection_model_name PP-OCRv5_mobile_det --text_detection_model_dir your_v5_mobile_det_model_path
+```
+
+Script mode: 
+
+```python
+
+from paddleocr import PaddleOCR
+
+#  Specify the local model path via text_detection_model_dir
+pipeline = PaddleOCR(text_detection_model_dir="./your_det_model_path")
+
+# Modify the model name via text_detection_model_name
+# pipeline = PaddleOCR(text_detection_model_name="PP-OCRv5_mobile_det", text_detection_model_dir="./your_v5_mobile_det_model_path")
+
+```
+
+The example above uses the local `paddle_static` inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following code:
+
+```python
+from paddleocr import PaddleOCR
+
+# Uses PP-OCRv6 models by default
+ocr = PaddleOCR(
+    use_doc_orientation_classify=False, # Disable document orientation classification
+    use_doc_unwarping=False, # Disable document unwarping
+    use_textline_orientation=False, # Disable textline orientation classification
+    engine="transformers",
+)
+# ocr = PaddleOCR(lang="en", engine="transformers") # Use the English model
+# ocr = PaddleOCR(ocr_version="PP-OCRv5", engine="transformers") # Switch to PP-OCRv5 version
+# ocr = PaddleOCR(ocr_version="PP-OCRv4", engine="transformers") # Switch to PP-OCRv4 version
+# ocr = PaddleOCR(device="gpu", engine="transformers") # Use GPU for inference
+# ocr = PaddleOCR(
+#     text_detection_model_name="PP-OCRv5_server_det",
+#     text_recognition_model_name="PP-OCRv5_server_rec",
+#     use_doc_orientation_classify=False,
+#     use_doc_unwarping=False,
+#     use_textline_orientation=False,
+#     engine="transformers",
+# ) # Switch to PP-OCRv5 server models
+result = ocr.predict("./general_ocr_002.png")
+for res in result:
+    res.print()
+    res.save_to_img("output")
+    res.save_to_json("output")
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following code:
+
+```python
+from paddleocr import PaddleOCR
+
+# Uses PP-OCRv6 models by default
+ocr = PaddleOCR(
+    use_doc_orientation_classify=False, # Disable document orientation classification
+    use_doc_unwarping=False, # Disable document unwarping
+    use_textline_orientation=False, # Disable textline orientation classification
+    engine="onnxruntime",
+)
+# ocr = PaddleOCR(lang="en", engine="onnxruntime") # Use the English model
+# ocr = PaddleOCR(ocr_version="PP-OCRv5", engine="onnxruntime") # Switch to PP-OCRv5 version
+# ocr = PaddleOCR(ocr_version="PP-OCRv4", engine="onnxruntime") # Switch to PP-OCRv4 version
+# ocr = PaddleOCR(device="gpu", engine="onnxruntime") # Use GPU for inference
+# ocr = PaddleOCR(
+#     text_detection_model_name="PP-OCRv5_server_det",
+#     text_recognition_model_name="PP-OCRv5_server_rec",
+#     use_doc_orientation_classify=False,
+#     use_doc_unwarping=False,
+#     use_textline_orientation=False,
+#     engine="onnxruntime",
+# ) # Switch to PP-OCRv5 server models
+result = ocr.predict("./general_ocr_002.png")
+for res in result:
+    res.print()
+    res.save_to_img("output")
+    res.save_to_json("output")
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+#### 4.2.2 Specify the local model path through the configuration file
+
+1.Obtain the pipeline configuration file
+
+Call the `export_paddlex_config_to_yaml` method of the **General OCR Pipeline** object in PaddleOCR to export the current pipeline configuration as a YAML file:  
+
+```Python  
+from paddleocr import PaddleOCR  
+
+pipeline = PaddleOCR()  
+pipeline.export_paddlex_config_to_yaml("PaddleOCR.yaml")  
+```  
+
+2.Modify the Configuration File  
+
+After obtaining the default pipeline configuration file, replace the paths of the default model weights with the local paths of your fine-tuned model weights. For example:  
+
+```yaml  
+......  
+SubModules:  
+  TextDetection:  
+    box_thresh: 0.6  
+    limit_side_len: 64  
+    limit_type: min
+    max_side_limit: 4000  
+    model_dir: null # Replace with the path to your fine-tuned text detection model weights  
+    model_name: PP-OCRv6_medium_det  # If the name of the fine-tuned model is different from the default model name, please modify it here as well
+    module_name: text_detection  
+    thresh: 0.3  
+    unclip_ratio: 1.5  
+  TextLineOrientation:  
+    batch_size: 6  
+    model_dir: null  # Replace with the path to your fine-tuned text LineOrientation model weights  
+    model_name: PP-LCNet_x1_0_textline_ori  # If the name of the fine-tuned model is different from the default model name, please modify it here as well
+    module_name: textline_orientation  
+  TextRecognition:  
+    batch_size: 6  
+    model_dir: null # Replace with the path to your fine-tuned text recognition model weights  
+    model_name: PP-OCRv6_medium_rec  # If the name of the fine-tuned model is different from the default model name, please modify it here as well
+    module_name: text_recognition  
+    score_thresh: 0.0  
+......  
+```  
+
+The pipeline configuration file includes not only the parameters supported by the PaddleOCR CLI and Python API but also advanced configurations. For detailed instructions, refer to the [PaddleX Pipeline Usage Overview](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/pipeline_develop_guide.html) and adjust the configurations as needed.  
+
+3.Load the Configuration File in CLI  
+
+After modifying the configuration file, specify its path using the `--paddlex_config` parameter in the command line. PaddleOCR will read the file and apply the configurations. Example:  
+
+```bash  
+paddleocr ocr --paddlex_config PaddleOCR.yaml ...  
+```  
+
+4.Load the Configuration File in Python API  
+
+When initializing the pipeline object, pass the path of the PaddleX pipeline configuration file or a configuration dictionary via the `paddlex_config` parameter. PaddleOCR will read and apply the configurations. Example:  
+
+```python  
+from paddleocr import PaddleOCR  
+
+pipeline = PaddleOCR(paddlex_config="PaddleOCR.yaml")  
+```
+
+## 5. Appendix
+
+<details><summary><b>Supported Languages</b></summary>
+
+<table border="1" cellspacing="0" cellpadding="4">
+  <thead>
+   <tr>
+      <th><code>lang</code></th>
+      <th>Language Name</th>
+   </tr>
+  </thead>
+  <tbody>
+   <tr><td><code>abq</code></td><td>Abaza</td></tr>
+   <tr><td><code>af</code></td><td>Afrikaans</td></tr>
+   <tr><td><code>ang</code></td><td>Old English</td></tr>
+   <tr><td><code>ar</code></td><td>Arabic</td></tr>
+   <tr><td><code>ava</code></td><td>Avaric</td></tr>
+   <tr><td><code>az</code></td><td>Azerbaijani</td></tr>
+   <tr><td><code>be</code></td><td>Belarusian</td></tr>
+   <tr><td><code>bg</code></td><td>Bulgarian</td></tr>
+   <tr><td><code>bgc</code></td><td>Haryanvi</td></tr>
+   <tr><td><code>bh</code></td><td>Bihari</td></tr>
+   <tr><td><code>bho</code></td><td>Bhojpuri</td></tr>
+   <tr><td><code>bs</code></td><td>Bosnian</td></tr>
+   <tr><td><code>ch</code></td><td>Chinese (Simplified)</td></tr>
+   <tr><td><code>che</code></td><td>Chechen</td></tr>
+   <tr><td><code>chinese_cht</code></td><td>Chinese (Traditional)</td></tr>
+   <tr><td><code>cs</code></td><td>Czech</td></tr>
+   <tr><td><code>cy</code></td><td>Welsh</td></tr>
+   <tr><td><code>da</code></td><td>Danish</td></tr>
+   <tr><td><code>dar</code></td><td>Dargwa</td></tr>
+   <tr><td><code>de</code> or <code>german</code></td><td>German</td></tr>
+   <tr><td><code>en</code></td><td>English</td></tr>
+   <tr><td><code>es</code></td><td>Spanish</td></tr>
+   <tr><td><code>et</code></td><td>Estonian</td></tr>
+   <tr><td><code>fa</code></td><td>Persian</td></tr>
+   <tr><td><code>fr</code> or <code>french</code></td><td>French</td></tr>
+   <tr><td><code>ga</code></td><td>Irish</td></tr>
+   <tr><td><code>gom</code></td><td>Konkani</td></tr>
+   <tr><td><code>hi</code></td><td>Hindi</td></tr>
+   <tr><td><code>hr</code></td><td>Croatian</td></tr>
+   <tr><td><code>hu</code></td><td>Hungarian</td></tr>
+   <tr><td><code>id</code></td><td>Indonesian</td></tr>
+   <tr><td><code>inh</code></td><td>Ingush</td></tr>
+   <tr><td><code>is</code></td><td>Icelandic</td></tr>
+   <tr><td><code>it</code></td><td>Italian</td></tr>
+   <tr><td><code>japan</code></td><td>Japanese</td></tr>
+   <tr><td><code>ka</code></td><td>Georgian</td></tr>
+   <tr><td><code>kbd</code></td><td>Kabardian</td></tr>
+   <tr><td><code>korean</code></td><td>Korean</td></tr>
+   <tr><td><code>ku</code></td><td>Kurdish</td></tr>
+   <tr><td><code>la</code></td><td>Latin</td></tr>
+   <tr><td><code>lbe</code></td><td>Lak</td></tr>
+   <tr><td><code>lez</code></td><td>Lezghian</td></tr>
+   <tr><td><code>lt</code></td><td>Lithuanian</td></tr>
+   <tr><td><code>lv</code></td><td>Latvian</td></tr>
+   <tr><td><code>mah</code></td><td>Magahi</td></tr>
+   <tr><td><code>mai</code></td><td>Maithili</td></tr>
+   <tr><td><code>mi</code></td><td>Maori</td></tr>
+   <tr><td><code>mn</code></td><td>Mongolian</td></tr>
+   <tr><td><code>mr</code></td><td>Marathi</td></tr>
+   <tr><td><code>ms</code></td><td>Malay</td></tr>
+   <tr><td><code>mt</code></td><td>Maltese</td></tr>
+   <tr><td><code>ne</code></td><td>Nepali</td></tr>
+   <tr><td><code>new</code></td><td>Newari</td></tr>
+   <tr><td><code>nl</code></td><td>Dutch</td></tr>
+   <tr><td><code>no</code></td><td>Norwegian</td></tr>
+   <tr><td><code>oc</code></td><td>Occitan</td></tr>
+   <tr><td><code>pi</code></td><td>Pali</td></tr>
+   <tr><td><code>pl</code></td><td>Polish</td></tr>
+   <tr><td><code>pt</code></td><td>Portuguese</td></tr>
+   <tr><td><code>ro</code></td><td>Romanian</td></tr>
+   <tr><td><code>rs_cyrillic</code></td><td>Serbian (Cyrillic)</td></tr>
+   <tr><td><code>rs_latin</code></td><td>Serbian (Latin)</td></tr>
+   <tr><td><code>ru</code></td><td>Russian</td></tr>
+   <tr><td><code>sa</code></td><td>Sanskrit</td></tr>
+   <tr><td><code>sck</code></td><td>Sadri</td></tr>
+   <tr><td><code>sk</code></td><td>Slovak</td></tr>
+   <tr><td><code>sl</code></td><td>Slovenian</td></tr>
+   <tr><td><code>sq</code></td><td>Albanian</td></tr>
+   <tr><td><code>sv</code></td><td>Swedish</td></tr>
+   <tr><td><code>sw</code></td><td>Swahili</td></tr>
+   <tr><td><code>tab</code></td><td>Tabassaran</td></tr>
+   <tr><td><code>ta</code></td><td>Tamil</td></tr>
+   <tr><td><code>te</code></td><td>Telugu</td></tr>
+   <tr><td><code>tl</code></td><td>Tagalog</td></tr>
+   <tr><td><code>tr</code></td><td>Turkish</td></tr>
+   <tr><td><code>ug</code></td><td>Uyghur</td></tr>
+   <tr><td><code>uk</code></td><td>Ukrainian</td></tr>
+   <tr><td><code>ur</code></td><td>Urdu</td></tr>
+   <tr><td><code>uz</code></td><td>Uzbek</td></tr>
+   <tr><td><code>vi</code></td><td>Vietnamese</td></tr>
+  </tbody>
+</table>
+
+</details>
+
+<details><summary><b>Correspondence Between OCR Model Versions and Supported Languages</b></summary>
+
+<table border="1" cellspacing="0" cellpadding="4">
+  <thead>
+   <tr>
+      <th><code>ocr_version</code></th>
+      <th>Supported <code>lang</code></th>
+   </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td><code>PP-OCRv6</code></td>
+      <td><code>ch</code>, <code>chinese_cht</code>, <code>en</code>, <code>japan</code>, <code>af</code>, <code>az</code>, <code>bs</code>, <code>ca</code>, <code>cs</code>, <code>cy</code>, <code>da</code>, <code>de</code>, <code>es</code>, <code>et</code>, <code>eu</code>, <code>fi</code>, <code>fr</code>, <code>ga</code>, <code>gl</code>, <code>hr</code>, <code>hu</code>, <code>id</code>, <code>is</code>, <code>it</code>, <code>ku</code>, <code>la</code>, <code>lb</code>, <code>lt</code>, <code>lv</code>, <code>mi</code>, <code>ms</code>, <code>mt</code>, <code>nl</code>, <code>no</code>, <code>oc</code>, <code>pl</code>, <code>pt</code>, <code>qu</code>, <code>rm</code>, <code>ro</code>, <code>rs_latin</code>, <code>sk</code>, <code>sl</code>, <code>sq</code>, <code>sv</code>, <code>sw</code>, <code>tl</code>, <code>tr</code>, <code>uz</code>, <code>vi</code>, <code>french</code>, <code>german</code>. <code>PP-OCRv6_tiny</code> does not support <code>japan</code>; set <code>PP-OCRv6_tiny_det</code>/<code>PP-OCRv6_tiny_rec</code> explicitly.</td>
+    </tr>
+    <tr>
+      <td><code>PP-OCRv5</code></td>
+      <td><code>ch</code>, <code>en</code>, <code>fr</code>, <code>de</code>, <code>japan</code>, <code>korean</code>, <code>chinese_cht</code>, <code>af</code>, <code>it</code>, <code>es</code>, <code>bs</code>, <code>pt</code>, <code>cs</code>, <code>cy</code>, <code>da</code>, <code>et</code>, <code>ga</code>, <code>hr</code>, <code>hu</code>, <code>rslatin</code>, <code>id</code>, <code>oc</code>, <code>is</code>, <code>lt</code>, <code>mi</code>, <code>ms</code>, <code>nl</code>, <code>no</code>, <code>pl</code>, <code>sk</code>, <code>sl</code>, <code>sq</code>, <code>sv</code>, <code>sw</code>, <code>tl</code>, <code>tr</code>, <code>uz</code>, <code>la</code>, <code>ru</code>, <code>be</code>, <code>uk</code>, <code>th</code>, <code>el</code>, <code>az</code>, <code>ku</code>, <code>lv</code>, <code>mt</code>, <code>pi</code>, <code>ro</code>, <code>vi</code>, <code>fi</code>, <code>eu</code>, <code>gl</code>, <code>lb</code>, <code>rm</code>, <code>ca</code>, <code>qu</code>, <code>te</code>, <code>sr</code>, <code>bg</code>, <code>mn</code>, <code>ab</code>, <code>ady</code>, <code>kbd</code>, <code>av</code>, <code>dar</code>, <code>inh</code>, <code>ce</code>, <code>lki</code>, <code>lez</code>, <code>tab</code>, <code>kk</code>, <code>ky</code>, <code>tg</code>, <code>mk</code>, <code>tt</code>, <code>cv</code>, <code>ba</code>, <code>mhr</code>, <code>mo</code>, <code>udm</code>, <code>kv</code>, <code>os</code>, <code>bua</code>, <code>xal</code>, <code>tyv</code>, <code>sah</code>, <code>kaa</code>, <code>ar</code>, <code>fa</code>, <code>ug</code>, <code>ur</code>, <code>ps</code>, <code>ku</code>, <code>sd</code>, <code>bal</code>, <code>hi</code>, <code>mr</code>, <code>ne</code>, <code>bh</code>, <code>mai</code>, <code>ang</code>, <code>bho</code>, <code>mah</code>, <code>sck</code>, <code>new</code>, <code>gom</code>, <code>sa</code>, <code>bgc</code>, <code>ta</code></td>
+    </tr>
+   <tr>
+     <td><code>PP-OCRv4</code></td>
+     <td><code>ch</code>, <code>en</code></td>
+   </tr>
+   <tr>
+     <td><code>PP-OCRv3</code></td>
+     <td>
+        <code>abq</code>, <code>af</code>, <code>ady</code>, <code>ang</code>, <code>ar</code>, <code>ava</code>, <code>az</code>, <code>be</code>,
+        <code>bg</code>, <code>bgc</code>, <code>bh</code>, <code>bho</code>, <code>bs</code>, <code>ch</code>, <code>che</code>,
+        <code>chinese_cht</code>, <code>cs</code>, <code>cy</code>, <code>da</code>, <code>dar</code>, <code>de</code>, <code>german</code>,
+        <code>en</code>, <code>es</code>, <code>et</code>, <code>fa</code>, <code>fr</code>, <code>french</code>, <code>ga</code>, <code>gom</code>,
+        <code>hi</code>, <code>hr</code>, <code>hu</code>, <code>id</code>, <code>inh</code>, <code>is</code>, <code>it</code>, <code>japan</code>,
+        <code>ka</code>, <code>kbd</code>, <code>korean</code>, <code>ku</code>, <code>la</code>, <code>lbe</code>, <code>lez</code>, <code>lt</code>,
+        <code>lv</code>, <code>mah</code>, <code>mai</code>, <code>mi</code>, <code>mn</code>, <code>mr</code>, <code>ms</code>, <code>mt</code>,
+        <code>ne</code>, <code>new</code>, <code>nl</code>, <code>no</code>, <code>oc</code>, <code>pi</code>, <code>pl</code>, <code>pt</code>,
+        <code>ro</code>, <code>rs_cyrillic</code>, <code>rs_latin</code>, <code>ru</code>, <code>sa</code>, <code>sck</code>, <code>sk</code>,
+        <code>sl</code>, <code>sq</code>, <code>sv</code>, <code>sw</code>, <code>ta</code>, <code>tab</code>, <code>te</code>, <code>tl</code>,
+        <code>tr</code>, <code>ug</code>, <code>uk</code>, <code>ur</code>, <code>uz</code>, <code>vi</code>
+      </td>
+   </tr>
+  </tbody>
+</table>
+
+</details>
diff --git a/docs/version3.x/pipeline_usage/OCR.md b/docs/version3.x/pipeline_usage/OCR.md
new file mode 100644
index 0000000..c473efe
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/OCR.md
@@ -0,0 +1,2879 @@
+---
+comments: true
+---
+
+# 通用OCR产线使用教程
+
+## 1. OCR产线介绍
+
+OCR（光学字符识别，Optical Character Recognition）是一种将图像中的文字转换为可编辑文本的技术。它广泛应用于文档数字化、信息提取和数据处理等领域。OCR 可以识别印刷文本、手写文本，甚至某些类型的字体和符号。
+
+通用 OCR 产线用于解决文字识别任务，提取图片中的文字信息以文本形式输出，本产线支持PP-OCRv3、PP-OCRv4、PP-OCRv5、PP-OCRv6模型的使用，其中默认模型为 PaddleOCR3.7 发布的 **PP-OCRv6_medium** 模型。PP-OCRv6 基于全新设计的 PPLCNetV4 统一骨干网络，提供 tiny/small/medium 三档模型，medium 档在综合精度上相比 PP-OCRv5_server 提升 5.1%（识别）和 4.6%（检测），同时推理速度更快；单一模型统一支持中、英、日及 46 种拉丁语系共 50 种语言。
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/ocr/01.png"/>
+
+<b>通用OCR产线中包含以下5个模块。每个模块均可独立进行训练和推理，并包含多个模型。有关详细信息，请点击相应模块以查看文档。</b>
+
+- [文档图像方向分类模块](../module_usage/doc_img_orientation_classification.md) （可选）
+- [文本图像矫正模块](../module_usage/text_image_unwarping.md) （可选）
+- [文本行方向分类模块](../module_usage/textline_orientation_classification.md) （可选）
+- [文本检测模块](../module_usage/text_detection.md)
+- [文本识别模块](../module_usage/text_recognition.md)
+
+在本产线中，您可以根据下方的基准测试数据选择使用的模型。
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。
+> 在带有 [常规模式 / 高性能模式] 标记的推理耗时列中，`常规模式` 对应本地推理引擎 `paddle_static`。
+
+<details>
+<summary><b>文档图像方向分类模块（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>Top-1 Acc（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">训练模型</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>基于PP-LCNet_x1_0的文档图像分类模型，含有四个类别，即0度，90度，180度，270度</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>文本图像矫正模块（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>CER </th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">训练模型</a></td>
+<td>0.179</td>
+<td>19.05 / 19.05</td>
+<td>- / 869.82</td>
+<td>30.3</td>
+<td>高精度文本图像矫正模型</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>文本行方向分类模块（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>模型下载链接</th>
+<th>Top-1 Acc（%）</th>
+<th>GPU推理耗时（ms）</th>
+<th>CPU推理耗时 (ms)</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x0_25_textline_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x0_25_textline_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x0_25_textline_ori_pretrained.pdparams">训练模型</a></td>
+<td>98.85</td>
+<td>2.16 / 0.41</td>
+<td>2.37 / 0.73</td>
+<td>0.96</td>
+<td>基于PP-LCNet_x0_25的文本行分类模型，含有两个类别，即0度，180度</td>
+</tr>
+<tr>
+<td>PP-LCNet_x1_0_textline_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_textline_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_textline_ori_pretrained.pdparams">训练模型</a></td>
+<td>99.42</td>
+<td>- / -</td>
+<td>2.98 / 2.98</td>
+<td>6.5</td>
+<td>基于PP-LCNet_x1_0的文本行分类模型，含有两个类别，即0度，180度</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>文本检测模块：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>检测Hmean（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv6_medium_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_medium_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_medium_det_pretrained.pdparams">训练模型</a></td>
+<td>86.2*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>59.4</td>
+<td>PP-OCRv6 的中等规模文本检测模型，基于 PPLCNetV4 + RepLKFPN，精度最高，适合服务端部署</td>
+</tr>
+<tr>
+<td>PP-OCRv6_small_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_small_det_pretrained.pdparams">训练模型</a></td>
+<td>84.1*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>9.6</td>
+<td>PP-OCRv6 的小型文本检测模型，兼顾精度与效率，适合移动端部署</td>
+</tr>
+<tr>
+<td>PP-OCRv6_tiny_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_tiny_det_pretrained.pdparams">训练模型</a></td>
+<td>80.6*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>1.9</td>
+<td>PP-OCRv6 的超轻量文本检测模型（0.43M 参数），适合端侧/IoT 场景</td>
+</tr>
+<tr>
+<td>PP-OCRv5_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_det_pretrained.pdparams">训练模型</a></td>
+<td>83.8</td>
+<td>89.55 / 70.19</td>
+<td>383.15 / 383.15</td>
+<td>84.3</td>
+<td>PP-OCRv5 的服务端文本检测模型，精度更高，适合在性能较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_det_pretrained.pdparams">训练模型</a></td>
+<td>79.0</td>
+<td>10.67 / 6.36</td>
+<td>57.77 / 28.15</td>
+<td>4.7</td>
+<td>PP-OCRv5 的移动端文本检测模型，效率更高，适合在端侧设备部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_det_pretrained.pdparams">训练模型</a></td>
+<td>69.2</td>
+<td>127.82 / 98.87</td>
+<td>585.95 / 489.77</td>
+<td>109</td>
+<td>PP-OCRv4 的服务端文本检测模型，精度更高，适合在性能较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams">训练模型</a></td>
+<td>63.8</td>
+<td>9.87 / 4.17</td>
+<td>56.60 / 20.79</td>
+<td>4.7</td>
+<td>PP-OCRv4 的移动端文本检测模型，效率更高，适合在端侧设备部署</td>
+</tr>
+</tbody>
+</table>
+
+> *注：PP-OCRv6 指标基于内部多场景评估集测得，PP-OCRv5/v4 指标基于通用评估集测得，两者评估集不同，指标不可直接对比。
+
+</details>
+
+<details>
+<summary><b>文本识别模块：</b></summary>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv6_medium_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_medium_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_medium_rec_pretrained.pdparams">训练模型</a></td>
+<td>83.2*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>73.3</td>
+<td rowspan="3">PP-OCRv6 文本识别模型，基于 PPLCNetV4 + LightSVTR + CTC/NRTR 多头解码器，单模型支持 50 种语言（tiny 档 49 种）</td>
+</tr>
+<tr>
+<td>PP-OCRv6_small_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_small_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.3*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>20.4</td>
+</tr>
+<tr>
+<td>PP-OCRv6_tiny_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv6_tiny_rec_pretrained.pdparams">训练模型</a></td>
+<td>73.5*</td>
+<td>- / -</td>
+<td>- / -</td>
+<td>4.4</td>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>86.38</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec 是新一代文本识别模型。该模型致力于以单一模型高效、精准地支持简体中文、繁体中文、英文、日文四种主要语言，以及手写、竖版、拼音、生僻字等复杂文本场景的识别。在保持识别效果的同时，兼顾推理速度和模型鲁棒性，为各种场景下的文档理解提供高效、精准的技术支撑。</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.29</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">训练模型</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc是在PP-OCRv4_server_rec的基础上，在更多中文文档数据和PP-OCR训练数据的混合数据训练而成，增加了部分繁体字、日文、特殊字符的识别能力，可支持识别的字符为1.5万+，除文档相关的文字识别能力提升外，也同时提升了通用文字的识别能力</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>PP-OCRv4的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>PP-OCRv4的服务器端模型，推理精度高，可以部署在多种不同的服务器上</td>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>基于PP-OCRv4识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+</table>
+
+> *注：PP-OCRv6 指标基于内部多场景评估集测得，PP-OCRv5/v4 指标基于通用评估集测得，两者评估集不同，指标不可直接对比。
+> ❗ 以上列出的是文本识别模块重点支持的<b>6个核心模型</b>，该模块总共支持<b>20个全量模型</b>，包含多个多语言文本识别模型，完整的模型列表如下：
+
+<details><summary> 👉模型列表详情</summary>
+
+* <b>PP-OCRv5 多场景模型</b>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>中文识别 Avg Accuracy(%)</th>
+<th>英文识别 Avg Accuracy(%)</th>
+<th>繁体中文识别 Avg Accuracy(%)</th>
+<th>日文识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>86.38</td>
+<td>64.70</td>
+<td>93.29</td>
+<td>60.35</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec 是新一代文本识别模型。该模型致力于以单一模型高效、精准地支持简体中文、繁体中文、英文、日文四种主要语言，以及手写、竖版、拼音、生僻字等复杂文本场景的识别。在保持识别效果的同时，兼顾推理速度和模型鲁棒性，为各种场景下的文档理解提供高效、精准的技术支撑。</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.29</td>
+<td>66.00</td>
+<td>83.55</td>
+<td>54.65</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+</table>
+
+* <b>中文识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">训练模型</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc是在PP-OCRv4_server_rec的基础上，在更多中文文档数据和PP-OCR训练数据的混合数据训练而成，增加了部分繁体字、日文、特殊字符的识别能力，可支持识别的字符为1.5万+，除文档相关的文字识别能力提升外，也同时提升了通用文字的识别能力</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>PP-OCRv4的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>PP-OCRv4的服务器端模型，推理精度高，可以部署在多种不同的服务器上</td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>72.96</td>
+<td>3.89 / 1.16</td>
+<td>8.72 / 3.56</td>
+<td>10.3</td>
+<td>PP-OCRv3的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>ch_SVTRv2_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_SVTRv2_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_SVTRv2_rec_pretrained.pdparams">训练模型</a></td>
+<td>68.81</td>
+<td>10.38 / 8.31</td>
+<td>66.52 / 30.83</td>
+<td>80.5</td>
+<td rowspan="1">
+SVTRv2 是一种由复旦大学视觉与学习实验室（FVL）的OpenOCR团队研发的服务端文本识别模型，其在PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务中荣获一等奖，A榜端到端识别精度相比PP-OCRv4提升6%。
+</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>ch_RepSVTR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_RepSVTR_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_RepSVTR_rec_pretrained.pdparams">训练模型</a></td>
+<td>65.07</td>
+<td>6.29 / 1.57</td>
+<td>20.64 / 5.40</td>
+<td>48.8</td>
+<td rowspan="1">    RepSVTR 文本识别模型是一种基于SVTRv2 的移动端文本识别模型，其在PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务中荣获一等奖，B榜端到端识别精度相比PP-OCRv4提升2.5%，推理速度持平。</td>
+</tr>
+</table>
+
+* <b>英文识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>en_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td> 85.25</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>基于PP-OCRv5识别模型训练得到的超轻量级英文识别模型，进一步提升英文文本的识别准确率，优化空格漏识别的问题，并提高对手写英文文本的识别效果。</td>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td> 70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>基于PP-OCRv4识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+<tr>
+<td>en_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>70.69</td>
+<td>3.56 / 0.78</td>
+<td>8.44 / 5.78</td>
+<td>17.3</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+</table>
+
+
+* <b>多语言识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>korean_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>88.0</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>14</td>
+<td>基于PP-OCRv5识别模型训练得到的超轻量韩文识别模型，支持韩文、英文和数字识别</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>84.7</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>14</td>
+<td>基于PP-OCRv5识别模型训练得到的拉丁文识别模型，支持大部分拉丁字母语言、数字识别</td>
+</tr>
+<tr>
+<td>eslav_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+eslav_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/eslav_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.6</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>14</td>
+<td>基于PP-OCRv5识别模型训练得到的东斯拉夫语言识别模型， 支持东斯拉夫语言、英文和数字识别</td>
+</tr>
+<tr>
+<td>th_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+th_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/th_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>82.68</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>基于PP-OCRv5识别模型训练得到的泰语识别模型， 支持泰语、英文和数字识别</td>
+</tr>
+<tr>
+<td>el_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+el_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/el_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>89.28</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>基于PP-OCRv5识别模型训练得到的希腊语识别模型， 支持希腊语、英文和数字识别</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+arabic_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.27</td>
+<td>-</td>
+<td>-</td>
+<td>7.6</td>
+<td>基于PP-OCRv5识别模型训练得到的超轻量阿拉伯字母识别模型，支持阿拉伯字母、数字识别</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+cyrillic_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>80.27</td>
+<td>-</td>
+<td>-</td>
+<td>7.7</td>
+<td>基于PP-OCRv5识别模型训练得到的超轻量斯拉夫字母识别模型，支持斯拉夫字母、数字识别</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+devanagari_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>84.96</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>基于PP-OCRv5识别模型训练得到的超轻量天城文识别模型，支持印地文、梵文等字母以及数字识别</td>
+</tr>
+<tr>
+<td>te_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+te_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>87.65</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>基于PP-OCRv5识别模型训练得到的超轻量泰卢固文识别模型，支持泰卢固文、数字识别</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ta_PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>94.2</td>
+<td>-</td>
+<td>-</td>
+<td>7.5</td>
+<td>基于PP-OCRv5识别模型训练得到的超轻量泰米尔文识别模型，支持泰米尔文、数字识别</td>
+</tr>
+<tr>
+<td>korean_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>60.21</td>
+<td>3.73 / 0.98</td>
+<td>8.76 / 2.91</td>
+<td>9.6</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量韩文识别模型，支持韩文、数字识别</td>
+</tr>
+<tr>
+<td>japan_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+japan_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/japan_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>45.69</td>
+<td>3.86 / 1.01</td>
+<td>8.62 / 2.92</td>
+<td>9.8</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量日文识别模型，支持日文、数字识别</td>
+</tr>
+<tr>
+<td>chinese_cht_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+chinese_cht_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/chinese_cht_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>82.06</td>
+<td>3.90 / 1.16</td>
+<td>9.24 / 3.18</td>
+<td>10.8</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量繁体中文识别模型，支持繁体中文、数字识别</td>
+</tr>
+<tr>
+<td>te_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+te_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>95.88</td>
+<td>3.59 / 0.81</td>
+<td>8.28 / 6.21</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量泰卢固文识别模型，支持泰卢固文、数字识别</td>
+</tr>
+<tr>
+<td>ka_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ka_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ka_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>96.96</td>
+<td>3.49 / 0.89</td>
+<td>8.63 / 2.77</td>
+<td>17.4</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量卡纳达文识别模型，支持卡纳达文、数字识别</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ta_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>76.83</td>
+<td>3.49 / 0.86</td>
+<td>8.35 / 3.41</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量泰米尔文识别模型，支持泰米尔文、数字识别</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>76.93</td>
+<td>3.53 / 0.78</td>
+<td>8.50 / 6.83</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量拉丁文识别模型，支持拉丁文、数字识别</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+arabic_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>73.55</td>
+<td>3.60 / 0.83</td>
+<td>8.44 / 4.69</td>
+<td>17.3</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量阿拉伯字母识别模型，支持阿拉伯字母、数字识别</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+cyrillic_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>94.28</td>
+<td>3.56 / 0.79</td>
+<td>8.22 / 2.76</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量斯拉夫字母识别模型，支持斯拉夫字母、数字识别</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>96.44</td>
+<td>3.60 / 0.78</td>
+<td>6.95 / 2.87</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量梵文字母识别模型，支持梵文字母、数字识别</td>
+</tr>
+</table>
+</details>
+</details>
+
+<details>
+<summary><strong>测试环境说明:</strong></summary>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+            <li><strong>测试数据集：
+             </strong>
+                <ul>
+                  <li>文档图像方向分类模型：PaddleOCR 自建的数据集，覆盖证件和文档等多个场景，包含 1000 张图片。</li>
+                  <li>文本图像矫正模型：<a href="https://www3.cs.stonybrook.edu/~cvl/docunet.html">DocUNet</a>。</li>
+                  <li>文本检测模型：PaddleOCR 自建的中文数据集，覆盖街景、网图、文档、手写多个场景，其中检测包含 500 张图片。</li>
+                  <li>中文识别模型： PaddleOCR 自建的中文数据集，覆盖街景、网图、文档、手写多个场景，其中文本识别包含 1.1w 张图片。</li>
+                  <li>ch_SVTRv2_rec：<a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务</a>A榜评估集。</li>
+                  <li>ch_RepSVTR_rec：<a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务</a>B榜评估集。</li>
+                  <li>英文识别模型：PaddleOCR 自建的英文数据集。</li>
+                  <li>多语言识别模型：PaddleOCR 自建的多语种数据集。</li>
+                  <li>文本行方向分类模型：PaddleOCR 自建的数据集，覆盖证件和文档等多个场景，包含 1000 张图片。</li>
+                </ul>
+             </li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td><code>paddle_static</code></td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+
+</details>
+
+<br />
+<b>如果您更注重模型的精度，请选择精度较高的模型；如果您更在意模型的推理速度，请选择推理速度较快的模型；如果您关注模型的存储大小，请选择存储体积较小的模型。</b>
+
+## 2. 快速开始
+
+在本地使用通用 OCR 产线前，请先根据所选推理引擎安装对应依赖，再安装 `paddleocr` Python 包。完整安装说明请参考[安装教程](../installation.md)。安装完成后，即可在本地使用命令行体验或 Python 集成。
+
+**请注意，如果在执行过程中遇到程序失去响应、程序异常退出、内存资源耗尽、推理速度极慢等问题，请尝试参考文档调整配置，例如关闭不需要使用的功能或使用更轻量的模型。**
+
+### 2.0 环境准备
+
+#### 2.0.1 安装推理引擎
+
+- 如需使用本地推理引擎 `paddle_static`，请参考[飞桨框架安装说明](../paddlepaddle_installation.md)安装 PaddlePaddle。
+- 如需使用 `transformers` 推理引擎，请参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)安装相关依赖。
+- 如需使用 `onnxruntime` 推理引擎，请参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)安装相关依赖。
+
+#### 2.0.2 安装 paddleocr
+
+```bash
+# 安装基础版本（仅包含OCR功能）
+pip install paddleocr
+
+# 安装完整版本（包含所有功能）
+pip install "paddleocr[all]"
+```
+
+#### 2.0.3 环境验证
+
+```python
+# 验证安装是否成功
+import paddleocr
+print(f"PaddleOCR版本: {paddleocr.__version__}")
+
+# 若使用本地推理引擎 paddle_static，可继续验证 PaddlePaddle 与 GPU 是否可用
+import paddle
+print(f"Paddle版本: {paddle.__version__}")
+print(f"GPU可用: {paddle.is_compiled_with_cuda()}")
+print(f"GPU数量: {paddle.device.cuda.device_count()}")
+
+# 若使用 transformers 推理引擎，可继续验证 transformers 依赖是否可用
+import transformers
+print(f"Transformers版本: {transformers.__version__}")
+```
+
+#### 2.0.4 常见安装问题解决
+
+**问题1：依赖冲突**
+```bash
+# 解决方案：创建新的虚拟环境
+conda create -n paddleocr python=3.8
+conda activate paddleocr
+pip install paddleocr
+```
+
+**问题2：GPU环境配置**
+```bash
+# 检查CUDA版本
+nvidia-smi
+
+# 使用本地推理引擎 paddle_static 时，安装对应版本的 PaddlePaddle
+pip install paddlepaddle-gpu==3.0.0 -f https://www.paddlepaddle.org.cn/whl/linux/mkl/avx/stable.html
+```
+
+**问题3：模型下载失败**
+```python
+# 设置模型下载源
+import os
+os.environ['PADDLE_PDX_MODEL_SOURCE'] = 'BOS'  # 使用百度云存储
+```
+
+### 2.1 命令行方式
+
+一行命令即可快速体验OCR产线效果。运行以下代码前，请您下载[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png)到本地：
+
+```bash
+paddleocr ocr -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False \
+    --use_textline_orientation False \
+    --save_path ./output \
+    --device gpu:0 
+
+# 通过 --ocr_version 指定 PP-OCR 其他版本
+paddleocr ocr -i ./general_ocr_002.png --ocr_version PP-OCRv4
+```
+
+上述命令默认使用本地推理引擎 `paddle_static`。如需运行，请先参考[飞桨框架安装说明](../paddlepaddle_installation.md)安装 PaddlePaddle。
+
+如果选择 `transformers` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 Transformers 环境配置，然后执行如下命令：
+
+```bash
+# 使用 transformers 引擎进行推理
+paddleocr ocr -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False \
+    --use_textline_orientation False \
+    --save_path ./output \
+    --device gpu:0 \
+    --engine transformers
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 ONNX Runtime 环境配置，然后执行如下命令：
+
+```bash
+# 使用 onnxruntime 引擎进行推理
+paddleocr ocr -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False \
+    --use_textline_orientation False \
+    --save_path ./output \
+    --device gpu:0 \
+    --engine onnxruntime
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+<!-- Luxorion-12修改：表格参数描述 -->
+
+<details><summary><b>命令行支持更多参数设置，点击展开以查看命令行参数的详细说明</b></summary>
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，必填。<br><b>说明：</b>如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<br><b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png">示例</a>；</br><b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)。</br>
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td><b>含义：</b>指定推理结果文件保存的路径。<br><b>说明：</b>如果不设置，推理结果将不会保存到本地。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>含义：</b>文档方向分类模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>文档方向分类模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>含义：</b>文本图像矫正模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>含义：</b>文本检测模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>含义：</b>文本检测模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_name</code></td>
+<td><b>含义：</b>文本行方向模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_dir</code></td>
+<td><b>含义：</b>文本行方向模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_batch_size</code></td>
+<td><b>含义：</b>文本行方向模型的batch size。<br><b>说明：</b>如果不设置，将默认设置batch size为<code>1</code>。</br></td>
+<td><code>int</code></td>
+<td><code>1</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>含义：</b>文本识别模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>含义：</b>文本识别模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>含义：</b>文本识别模型的batch size。<br><b>说明：</b>如果不设置，将默认设置batch size为<code>1</code>。</br></td>
+<td><code>int</code></td>
+<td><code>1</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否加载并使用文档方向分类模块。<br><b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</br></td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。<br><b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</br></td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>含义：</b>是否加载并使用文本行方向模块。<br><b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</br></td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>含义：</b>文本检测的图像边长限制。
+<br><b>说明：</b>大于 <code>0</code> 的任意整数。</br>如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>64</code>。
+</td>
+<td><code>int</code></td>
+<td><code>64</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>含义：</b>文本检测的边长度限制类型。
+<br><b>说明：</b>支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于<code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>。</br>如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>min</code>。
+</td>
+<td><code>str</code></td>
+<td><code>min</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>含义：</b>文本检测像素阈值，输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。
+<br><b>说明：</b>大于<code>0</code>的任意浮点数。</br>如果不设置，将使用产线初始化的该参数值（默认为 <code>0.3</code>）。
+</td>
+<td><code>float</code></td>
+<td><code>0.3</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>含义：</b>文本检测框阈值，检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。
+<br><b>说明：</b>大于 <code>0</code> 的任意浮点数。</br>如果不设置，将使用产线初始化的该参数值（默认为 <code>0.6</code>）。
+</td>
+<td><code>float</code></td>
+<td><code>0.6</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>含义：</b>文本检测扩张系数，使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。
+<br><b>说明：</b>大于 <code>0</code> 的任意浮点数。</br>如果不设置，将使用产线初始化的该参数值（默认为 <code>2.0</code>）。
+</td>
+<td><code>float</code></td>
+<td><code>2.0</code></td>
+</tr>
+<tr>
+<td><code>text_det_input_shape</code></td>
+<td><b>含义：</b>文本检测的输入形状。<br><b>说明：</b>可以设置3个值代表C，H，W。</br></td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>含义：</b>文本识别阈值，得分大于该阈值的文本结果会被保留。
+<br><b>说明：</b>大于<code>0</code>的任意浮点数。</br>如果不设置，将使用产线初始化的该参数值（默认为 <code>0.0</code>，即不设阈值）。
+<td><code>float</code></td>
+<td><code>0.0</code></td>
+</tr>
+<tr>
+<td><code>return_word_box</code></td>
+<td><b>含义：</b>是否返回识别结果的文字框坐标。
+<b>说明：</b><br>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</br></td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>text_rec_input_shape</code></td>
+<td><b>含义：</b>文本识别的输入形状。</td>
+<td><code>tuple</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>lang</code></td>
+<td><b>含义：</b>使用指定语言的 OCR 模型。
+<br><b>说明：</b>附录中的表格中列举了全部支持的语言。</br>
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>ocr_version</code></td>
+<td><b>含义：</b>OCR 模型版本。
+<br><b>说明：</b>
+<ul>
+<li><b>PP-OCRv6</b>：使用PP-OCRv6系列模型；
+<li><b>PP-OCRv5</b>：使用PP-OCRv5系列模型；
+<li><b>PP-OCRv4</b>：使用PP-OCRv4系列模型；
+<li><b>PP-OCRv3</b>：使用PP-OCRv3系列模型。
+</ul>
+注意不是每个<code>ocr_version</code>都支持所有的<code>lang</code>，请查看附录中的对应关系表。
+</td></br>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br><b>说明：</b>支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+</ul>如果不设置，将默认使用产线初始化的该参数值，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。</br>
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>fp32</code>、<code>fp16</code>。</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+
+</tbody>
+</table>
+<b>以下参数在2.x版本已经废弃，为方便之前版本使用者故列出</b>
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<td><code>det_model_dir</code></td>
+<td><b>说明：</b>已废弃，请参考<code>text_detection_model_dir</code>，且与新的参数不能同时指定。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>det_limit_side_len</code></td>
+<td><b>说明：</b>已废弃，请参考<code>text_det_limit_side_len</code>，且与新的参数不能同时指定。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>det_limit_type</code></td>
+<td><b>说明：</b>已废弃，请参考<code>text_det_limit_type</code>，且与新的参数不能同时指定。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>det_db_thresh</code></td>
+<td><b>说明：</b>已废弃，请参考<code>text_det_thresh</code>，且与新的参数不能同时指定。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>det_db_box_thresh</code></td>
+<td><b>说明：</b>已废弃，请参考<code>text_det_box_thresh</code>，且与新的参数不能同时指定。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>det_db_unclip_ratio</code></td>
+<td><b>说明：</b>已废弃，请参考<code>text_det_unclip_ratio</code>，且与新的参数不能同时指定。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>rec_model_dir</code></td>
+<td><b>说明：</b>已废弃，请参考<code>text_recognition_model_dir</code>，且与新的参数不能同时指定。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>rec_batch_num</code></td>
+<td><b>说明：</b>已废弃，请参考<code>text_recognition_batch_size</code>，且与新的参数不能同时指定。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_angle_cls</code></td>
+<td><b>说明：</b>已废弃，请参考<code>use_textline_orientation</code>，且与新的参数不能同时指定。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>cls_model_dir</code></td>
+<td><b>说明：</b>已废弃，请参考<code>textline_orientation_model_dir</code>，且与新的参数不能同时指定。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>cls_batch_num</code></td>
+<td><b>说明：</b>已废弃，请参考<code>textline_orientation_batch_size</code>，且与新的参数不能同时指定。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+</tbody>
+</table>
+</details>
+<br />
+
+运行结果会被打印到终端上：
+
+```bash
+{'res': {'input_path': './general_ocr_002.png', 'page_index': None, 'model_settings': {'use_doc_preprocessor': True, 'use_textline_orientation': False}, 'doc_preprocessor_res': {'input_path': None, 'page_index': None, 'model_settings': {'use_doc_orientation_classify': False, 'use_doc_unwarping': False}, 'angle': -1}, 'dt_polys': array([[[  3,  10],
+        ...,
+        [  4,  30]],
+
+       ...,
+
+       [[ 99, 456],
+        ...,
+        [ 99, 479]]], dtype=int16), 'text_det_params': {'limit_side_len': 736, 'limit_type': 'min', 'thresh': 0.3, 'max_side_limit': 4000, 'box_thresh': 0.6, 'unclip_ratio': 1.5}, 'text_type': 'general', 'textline_orientation_angles': array([-1, ..., -1]), 'text_rec_score_thresh': 0.0, 'rec_texts': ['www.997700', '', 'Cm', '登机牌', 'BOARDING', 'PASS', 'CLASS', '序号SERIAL NO.', '座位号', 'SEAT NO.', '航班FLIGHT', '日期DATE', '舱位', '', 'W', '035', '12F', 'MU2379', '03DEc', '始发地', 'FROM', '登机口', 'GATE', '登机时间BDT', '目的地TO', '福州', 'TAIYUAN', 'G11', 'FUZHOU', '身份识别IDNO.', '姓名NAME', 'ZHANGQIWEI', '票号TKT NO.', '张祺伟', '票价FARE', 'ETKT7813699238489/1', '登机口于起飞前10分钟关闭 GATESCL0SE10MINUTESBEFOREDEPARTURETIME'], 'rec_scores': array([0.67634439, ..., 0.97416091]), 'rec_polys': array([[[  3,  10],
+        ...,
+        [  4,  30]],
+
+       ...,
+
+       [[ 99, 456],
+        ...,
+        [ 99, 479]]], dtype=int16), 'rec_boxes': array([[  3, ...,  30],
+       ...,
+       [ 99, ..., 479]], dtype=int16)}}
+```
+
+若指定了`save_path`，则会保存可视化结果在`save_path`下。可视化结果如下：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/ocr/03.png"/>
+
+
+### 2.2 Python脚本方式集成
+
+命令行方式是为了快速体验查看效果，一般来说，在项目中，往往需要通过代码集成，您可以通过几行代码即可完成产线的快速推理，推理代码如下：
+
+```python
+from paddleocr import PaddleOCR
+
+# 默认使用 PP-OCRv6 模型
+ocr = PaddleOCR(
+    use_doc_orientation_classify=False, # 通过 use_doc_orientation_classify 参数指定不使用文档方向分类模型
+    use_doc_unwarping=False, # 通过 use_doc_unwarping 参数指定不使用文本图像矫正模型
+    use_textline_orientation=False, # 通过 use_textline_orientation 参数指定不使用文本行方向分类模型
+)
+# ocr = PaddleOCR(lang="en") # 通过 lang 参数来使用英文模型
+# ocr = PaddleOCR(ocr_version="PP-OCRv5") # 通过 ocr_version 参数切换为 PP-OCRv5 版本
+# ocr = PaddleOCR(ocr_version="PP-OCRv4") # 通过 ocr_version 参数切换为 PP-OCRv4 版本
+# ocr = PaddleOCR(device="gpu") # 通过 device 参数使得在模型推理时使用 GPU
+# ocr = PaddleOCR(
+#     text_detection_model_name="PP-OCRv5_server_det",
+#     text_recognition_model_name="PP-OCRv5_server_rec",
+#     use_doc_orientation_classify=False,
+#     use_doc_unwarping=False,
+#     use_textline_orientation=False,
+# ) # 使用 PP-OCRv5 的 server 模型
+result = ocr.predict("./general_ocr_002.png")
+for res in result:
+    res.print()
+    res.save_to_img("output")
+    res.save_to_json("output")
+```
+
+上述代码默认使用本地推理引擎 `paddle_static`。如需运行，请先参考[飞桨框架安装说明](../paddlepaddle_installation.md)安装 PaddlePaddle。
+
+如果选择 `transformers` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 Transformers 环境配置，然后执行如下代码：
+
+```python
+from paddleocr import PaddleOCR
+
+# 默认使用 PP-OCRv6 模型
+ocr = PaddleOCR(
+    use_doc_orientation_classify=False, # 通过 use_doc_orientation_classify 参数指定不使用文档方向分类模型
+    use_doc_unwarping=False, # 通过 use_doc_unwarping 参数指定不使用文本图像矫正模型
+    use_textline_orientation=False, # 通过 use_textline_orientation 参数指定不使用文本行方向分类模型
+    engine="transformers",
+)
+# ocr = PaddleOCR(lang="en", engine="transformers") # 通过 lang 参数来使用英文模型
+# ocr = PaddleOCR(ocr_version="PP-OCRv5", engine="transformers") # 通过 ocr_version 参数切换为 PP-OCRv5 版本
+# ocr = PaddleOCR(ocr_version="PP-OCRv4", engine="transformers") # 通过 ocr_version 参数切换为 PP-OCRv4 版本
+# ocr = PaddleOCR(device="gpu", engine="transformers") # 通过 device 参数使得在模型推理时使用 GPU
+# ocr = PaddleOCR(
+#     text_detection_model_name="PP-OCRv5_server_det",
+#     text_recognition_model_name="PP-OCRv5_server_rec",
+#     use_doc_orientation_classify=False,
+#     use_doc_unwarping=False,
+#     use_textline_orientation=False,
+#     engine="transformers",
+# ) # 使用 PP-OCRv5 的 server 模型
+result = ocr.predict("./general_ocr_002.png")
+for res in result:
+    res.print()
+    res.save_to_img("output")
+    res.save_to_json("output")
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 ONNX Runtime 环境配置，然后执行如下代码：
+
+```python
+from paddleocr import PaddleOCR
+
+# 默认使用 PP-OCRv6 模型
+ocr = PaddleOCR(
+    use_doc_orientation_classify=False, # 通过 use_doc_orientation_classify 参数指定不使用文档方向分类模型
+    use_doc_unwarping=False, # 通过 use_doc_unwarping 参数指定不使用文本图像矫正模型
+    use_textline_orientation=False, # 通过 use_textline_orientation 参数指定不使用文本行方向分类模型
+    engine="onnxruntime",
+)
+# ocr = PaddleOCR(lang="en", engine="onnxruntime") # 通过 lang 参数来使用英文模型
+# ocr = PaddleOCR(ocr_version="PP-OCRv5", engine="onnxruntime") # 通过 ocr_version 参数切换为 PP-OCRv5 版本
+# ocr = PaddleOCR(ocr_version="PP-OCRv4", engine="onnxruntime") # 通过 ocr_version 参数切换为 PP-OCRv4 版本
+# ocr = PaddleOCR(device="gpu", engine="onnxruntime") # 通过 device 参数使得在模型推理时使用 GPU
+# ocr = PaddleOCR(
+#     text_detection_model_name="PP-OCRv5_server_det",
+#     text_recognition_model_name="PP-OCRv5_server_rec",
+#     use_doc_orientation_classify=False,
+#     use_doc_unwarping=False,
+#     use_textline_orientation=False,
+#     engine="onnxruntime",
+# ) # 使用 PP-OCRv5 的 server 模型
+result = ocr.predict("./general_ocr_002.png")
+for res in result:
+    res.print()
+    res.save_to_img("output")
+    res.save_to_json("output")
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+<!-- Luxorion-12 修改：表格参数说明-->
+在上述 Python 脚本中，执行了如下几个步骤：
+
+<details><summary>（1）通过 <code>PaddleOCR()</code> 实例化 OCR 产线对象，具体参数说明如下：</summary>
+
+<table>
+  <thead>
+    <tr>
+      <th>参数</th>
+      <th>参数说明</th>
+      <th>参数类型</th>
+      <th>默认值</th>
+    </tr>
+  </thead>
+  <tbody>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>含义：</b>文档方向分类模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>文档方向分类模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>含义：</b>文本图像矫正模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>含义：</b>文本检测模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>含义：</b>文本检测模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_name</code></td>
+<td><b>含义：</b>文本行方向模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_dir</code></td>
+<td><b>含义：</b>文本行方向模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_batch_size</code></td>
+<td><b>含义：</b>文本行方向模型的batch size。
+<br><b>说明：</b>如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</br></td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>含义：</b>文本识别模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>含义：</b>文本识别模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>含义：</b>文本识别模型的batch size。
+<br><b>说明：</b>如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</br></td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否加载并使用文档方向分类模块。
+<br><b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。
+<br><b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>含义：</b>是否加载并使用文本行方向模块。
+<br><b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>含义：</b>文本检测的图像边长限制。
+<br><b>说明：</b>
+<ul>
+<li><b>int</b>：大于 <code>0</code> 的任意整数；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>64</code>。</li>
+</ur>
+</br></td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>含义：</b>文本检测的边长度限制类型。
+<br><b>说明：</b>
+<ul>
+<li><b>str</b>：支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>min</code>。</li>
+</ur>
+</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>含义：</b>文本检测像素阈值，输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。
+<br><b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值（默认为<code>0.3</code>）。</li>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</br></tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>含义：</b>文本检测框阈值，检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。
+<br><b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值（默认为<code>0.6</code>）。
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</br></tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>含义：</b>文本检测扩张系数，使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。
+<br><b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值（默认为<code>2.0</code>）。
+</ur>
+</br></td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_input_shape</code></td>
+<td><b>含义：</b>文本检测的输入形状。</td>
+<td><code>tuple|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>含义：</b>文本识别阈值，得分大于该阈值的文本结果会被保留。
+<br><b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值（默认为<code>0.0</code>，即不设阈值）。
+</ur>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</br></tr>
+<tr>
+<td><code>text_rec_input_shape</code></td>
+<td><b>含义：</b>文本识别的输入形状。</td>
+<td><code>tuple|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>lang</code></td>
+<td><b>含义：</b>使用指定语言的 OCR 模型。
+<br><b>说明：</b>附录中的表格中列举了全部支持的语言。
+</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>ocr_version</code></td>
+<td><b>含义：</b>OCR 模型版本。
+<br><b>说明：</b>
+<ul>
+<li><b>PP-OCRv6</b>：使用PP-OCRv6系列模型；
+<li><b>PP-OCRv5</b>：使用PP-OCRv5系列模型；
+<li><b>PP-OCRv4</b>：使用PP-OCRv4系列模型；
+<li><b>PP-OCRv3</b>：使用PP-OCRv3系列模型。
+</ul>
+注意不是每个<code>ocr_version</code>都支持所有的<code>lang</code>，请查看附录中的对应关系表。
+</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。
+<br><b>说明：</b>
+支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。
+</ur>
+</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。
+<b>说明：</b>如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。
+<br><b>说明：</b>如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。</br></td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details><summary>（2）调用 OCR 产线对象的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。另外，产线还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。以下是 <code>predict()</code> 方法的参数及其说明：</summary>
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据
+<br><b>说明：</b>支持多种输入类型，必填。
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据；</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；
+<br><b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png">示例</a>；</br>
+<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)；</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]。</code></li>
+</ul>
+</br></td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否在推理时使用文档方向分类模块。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否在推理时使用文本图像矫正模块。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>含义：</b>是否在推理时使用文本行方向分类模块。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<td><code>text_det_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</table>
+</details>
+
+<details><summary>（3）对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为<code>json</code>文件的操作:</summary>
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化。</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效。</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效。</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致。</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效。</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效。</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将结果保存为图像格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径。</td>
+<td>无</td>
+</tr>
+</table>
+
+<ul>
+  <li>调用<code>print()</code> 方法会将结果打印到终端，打印到终端的内容解释如下：
+    <ol start="1" type="1">
+      <li><code>input_path</code>: <code>(str)</code> 待预测图像的输入路径</li>
+      <li><code>page_index</code>: <code>(Union[int, None])</code> 如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+      <li><code>model_settings</code>: <code>(Dict[str, bool])</code> 配置产线所需的模型参数
+        <ol >
+          <li><code>use_doc_preprocessor</code>: <code>(bool)</code> 控制是否启用文档预处理子产线</li>
+          <li><code>use_textline_orientation</code>: <code>(bool)</code> 控制是否启用文本行方向分类模块</li>
+        </ol>
+      </li>
+      <li><code>doc_preprocessor_res</code>: <code>(Dict[str, Union[str, Dict[str, bool], int]])</code> 文档预处理子产线的输出结果。仅当<code>use_doc_preprocessor=True</code>时存在
+        <ol >
+          <li><code>input_path</code>: <code>(Union[str, None])</code> 图像预处理子产线接受的图像路径，当输入为<code>numpy.ndarray</code>时，保存为<code>None</code></li>
+          <li><code>model_settings</code>: <code>(Dict)</code> 预处理子产线的模型配置参数
+            <ul>
+              <li><code>use_doc_orientation_classify</code>: <code>(bool)</code> 控制是否启用文档方向分类</li>
+              <li><code>use_doc_unwarping</code>: <code>(bool)</code> 控制是否启用文本图像矫正</li>
+            </ul>
+          </li>
+          <li><code>angle</code>: <code>(int)</code> 文档方向分类的预测结果。启用时取值为[0,1,2,3]，分别对应[0°,90°,180°,270°]；未启用时为-1</li>
+        </ol>
+      </li>
+      <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> 文本检测的多边形框列表。每个检测框由4个顶点坐标构成的numpy数组表示，数组shape为(4, 2)，数据类型为int16</li>
+      <li><code>dt_scores</code>: <code>(List[float])</code> 文本检测框的置信度列表</li>
+      <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> 文本检测模块的配置参数
+        <ul>
+          <li><code>limit_side_len</code>: <code>(int)</code> 图像预处理时的边长限制值</li>
+          <li><code>limit_type</code>: <code>(str)</code> 边长限制的处理方式</li>
+          <li><code>thresh</code>: <code>(float)</code> 文本像素分类的置信度阈值</li>
+          <li><code>box_thresh</code>: <code>(float)</code> 文本检测框的置信度阈值</li>
+          <li><code>unclip_ratio</code>: <code>(float)</code> 文本检测框的膨胀系数</li>
+          <li><code>text_type</code>: <code>(str)</code> 文本检测的类型，当前固定为"general"</li>
+        </ul>
+      </li>
+      <li><code>textline_orientation_angles</code>: <code>(List[int])</code> 文本行方向分类的预测结果。启用时返回实际角度值（如[0,0,1]），未启用时返回[-1,-1,-1]</li>
+      <li><code>text_rec_score_thresh</code>: <code>(float)</code> 文本识别结果的过滤阈值</li>
+      <li><code>rec_texts</code>: <code>(List[str])</code> 文本识别结果列表，仅包含置信度超过<code>text_rec_score_thresh</code>的文本</li>
+      <li><code>rec_scores</code>: <code>(List[float])</code> 文本识别的置信度列表，已按<code>text_rec_score_thresh</code>过滤</li>
+      <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> 经过置信度过滤的文本检测框列表，格式同<code>dt_polys</code></li>
+      <li><code>rec_boxes</code>: <code>(numpy.ndarray)</code> 检测框的矩形边界框数组，shape为(n, 4)，dtype为int16。每一行表示一个矩形框的[x_min, y_min, x_max, y_max]坐标，其中(x_min, y_min)为左上角坐标，(x_max, y_max)为右下角坐标</li>
+    </ol>
+  </li>
+  <li>调用<code>save_to_json()</code> 方法会将上述内容保存到指定的<code>save_path</code>中，如果指定为目录，则保存的路径为<code>save_path/{your_img_basename}_res.json</code>，如果指定为文件，则直接保存到该文件中。由于json文件不支持保存numpy数组，因此会将其中的<code>numpy.array</code>类型转换为列表形式。</li>
+  <li>调用<code>save_to_img()</code> 方法会将可视化结果保存到指定的<code>save_path</code>中，如果指定为目录，则保存的路径为<code>save_path/{your_img_basename}_ocr_res_img.{your_img_extension}</code>，如果指定为文件，则直接保存到该文件中。(产线通常包含较多结果图片，不建议直接指定为具体的文件路径，否则多张图会被覆盖，仅保留最后一张图)</li>
+</ul>
+
+<p>此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：</p>
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">获取预测的 <code>json</code> 格式的结果</td>
+</tr>
+<tr>
+<td rowspan="2"><code>img</code></td>
+<td rowspan="2">获取格式为 <code>dict</code> 的可视化图像</td>
+</tr>
+</table>
+
+<ul>
+  <li><code>json</code> 属性获取的预测结果为dict类型的数据，相关内容与调用 <code>save_to_json()</code> 方法保存的内容一致。</li>
+  <li><code>img</code> 属性返回的预测结果是一个dict类型的数据。其中，键分别为 <code>ocr_res_img</code> 和 <code>preprocessed_img</code>，对应的值是两个 <code>Image.Image</code> 对象：一个用于显示 OCR 结果的可视化图像，另一个用于展示图像预处理的可视化图像。如果没有使用图像预处理子模块，则dict中只包含 <code>ocr_res_img</code>。</li>
+</ul>
+
+</details>
+
+## 3. 开发集成/部署
+
+如果通用 OCR 产线可以达到您对产线推理速度和精度的要求，您可以直接进行开发集成/部署。
+
+若您需要将通用 OCR 产线直接应用在您的Python项目中，可以参考[2.2 Python脚本方式集成](#22-python)中的示例代码。
+
+此外，PaddleOCR 也提供了其他两种部署方式，详细说明如下：
+
+🚀 高性能推理：在实际生产环境中，许多应用对部署策略的性能指标（尤其是响应速度）有着较严苛的标准，以确保系统的高效运行与用户体验的流畅性。为此，PaddleOCR 提供高性能推理功能，旨在对模型推理及前后处理进行深度性能优化，实现端到端流程的显著提速，详细的高性能推理流程请参考[高性能推理](../inference_deployment/local_inference/high_performance_inference.md)。
+
+☁️ 服务化部署：服务化部署是实际生产环境中常见的一种部署形式。通过将推理功能封装为服务，客户端可以通过网络请求来访问这些服务，以获取推理结果。详细的产线服务化部署流程请参考[服务化部署](../inference_deployment/serving/serving.md)。
+
+以下是基础服务化部署的API参考与多语言服务调用示例：
+
+<details><summary>API参考</summary>
+<p>对于服务提供的主要操作：</p>
+<ul>
+<li>HTTP请求方法为POST。</li>
+<li>请求体和响应体均为JSON数据（JSON对象）。</li>
+<li>当请求处理成功时，响应状态码为<code>200</code>，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。固定为<code>0</code>。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。固定为<code>"Success"</code>。</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>操作结果。</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>当请求处理未成功时，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。与响应状态码相同。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。</td>
+</tr>
+</tbody>
+</table>
+<p>服务提供的主要操作如下：</p>
+<ul>
+<li><b><code>infer</code></b></li>
+</ul>
+<p>获取图像OCR结果。</p>
+<p><code>POST /ocr</code></p>
+<ul>
+<li>请求体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>服务器可访问的图像文件（含 TIFF，多页时按页处理）或 PDF 文件的 URL，或上述类型文件内容的 Base64 编码结果。默认对于超过 10 页的 PDF 或多页 TIFF，仅处理前 10 页。<br /> 要解除页数限制，请在产线配置文件中添加以下配置：
+<pre><code>Serving:
+  extra:
+    max_num_input_imgs: null
+</code></pre>
+</td>
+<td>是</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>文件类型。<code>0</code> 表示 PDF 文件，<code>1</code> 表示图像文件（含 TIFF）。若请求体无此属性，则将根据URL推断文件类型。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_doc_orientation_classify</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_doc_unwarping</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<tr>
+<td><code>useTextlineOrientation</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_textline_orientation</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_det_limit_side_len</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_det_limit_type</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_det_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_det_box_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_det_unclip_ratio</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_rec_score_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>是否返回可视化结果图以及处理过程中的中间图像等。
+<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>传入 <code>true</code>：返回图像。</li>
+<li>传入 <code>false</code>：不返回图像。</li>
+<li>若请求体中未提供该参数或传入 <code>null</code>：遵循产线配置文件<code>Serving.visualize</code> 的设置。</li>
+</ul>
+<br/>例如，在产线配置文件中添加如下字段：<br/>
+<pre><code>Serving:
+  visualize: False
+</code></pre>
+将默认不返回图像，通过请求体中的<code>visualize</code>参数可以覆盖默认行为。如果请求体和配置文件中均未设置（或请求体传入<code>null</code>、配置文件中未设置），则默认返回图像。
+</td>
+<td>否</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>请求处理成功时，响应体的<code>result</code>具有如下属性：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>ocrResults</code></td>
+<td><code>object</code></td>
+<td>OCR结果。数组长度为1（对于图像输入）或实际处理的文档页数（对于PDF输入）。对于PDF输入，数组中的每个元素依次表示PDF文件中实际处理的每一页的结果。</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>输入数据信息。</td>
+</tr>
+</tbody>
+</table>
+<p>下表中涉及图像的字段（如 <code>ocrImage</code>、<code>docPreprocessingImage</code>、<code>inputImage</code>）默认以 Base64 字符串内联返回；当服务端开启 URL 返回模式时，相应字段的值变为预签名 URL，字段类型保持不变。配置方式参见 <a href="../inference_deployment/serving/serving.md">服务化部署</a>「以 URL 形式返回二进制内容」一节。</p>
+<p><code>ocrResults</code>中的每个元素为一个<code>object</code>，具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>产线对象的 <code>predict</code> 方法生成结果的 JSON 表示中 <code>res</code> 字段的简化版本，其中去除了 <code>input_path</code> 和 <code>page_index</code> 字段。</td>
+</tr>
+<tr>
+<td><code>ocrImage</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>OCR结果图，其中标注检测到的文本位置。图像为JPEG格式，使用Base64编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+<tr>
+<td><code>docPreprocessingImage</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>可视化结果图像。图像为JPEG格式，使用Base64编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+<tr>
+<td><code>inputImage</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>输入图像。图像为JPEG格式，使用Base64编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details><summary>多语言调用服务示例</summary>
+
+<details>
+<summary>Python</summary>
+
+<pre><code class="language-python">import base64
+import requests
+
+API_URL = "http://localhost:8080/ocr"
+file_path = "./demo.jpg"
+
+with open(file_path, "rb") as file:
+    file_bytes = file.read()
+    file_data = base64.b64encode(file_bytes).decode("ascii")
+
+payload = {"file": file_data, "fileType": 1}
+
+response = requests.post(API_URL, json=payload)
+
+assert response.status_code == 200
+result = response.json()["result"]
+for i, res in enumerate(result["ocrResults"]):
+    print(res["prunedResult"])
+    ocr_img_path = f"ocr_{i}.jpg"
+    with open(ocr_img_path, "wb") as f:
+        f.write(base64.b64decode(res["ocrImage"]))
+    print(f"Output image saved at {ocr_img_path}")
+</code></pre></details>
+
+<details><summary>C++</summary>
+
+<pre><code class="language-cpp">#include &lt;iostream&gt;
+#include &lt;fstream&gt;
+#include &lt;vector&gt;
+#include &lt;string&gt;
+#include "cpp-httplib/httplib.h" // https://github.com/Huiyicc/cpp-httplib
+#include "nlohmann/json.hpp" // https://github.com/nlohmann/json
+#include "base64.hpp" // https://github.com/tobiaslocker/base64
+
+int main() {
+    httplib::Client client("localhost", 8080);  
+    const std::string filePath = "./demo.jpg"; 
+
+    std::ifstream file(filePath, std::ios::binary | std::ios::ate);
+    if (!file) {
+        std::cerr << "Error opening file." << std::endl;
+        return 1;
+    }
+
+    std::streamsize size = file.tellg();
+    file.seekg(0, std::ios::beg);
+    std::vector<char> buffer(size);
+
+    if (!file.read(buffer.data(), size)) {
+        std::cerr << "Error reading file." << std::endl;
+        return 1;
+    }
+
+    std::string bufferStr(buffer.data(), static_cast<size_t>(size));
+    std::string encodedFile = base64::to_base64(bufferStr);
+
+
+    nlohmann::json jsonObj;
+    jsonObj["file"] = encodedFile;
+    jsonObj["fileType"] = 1;  
+
+    auto response = client.Post("/ocr", jsonObj.dump(), "application/json");
+
+    if (response && response->status == 200) {
+        nlohmann::json jsonResponse = nlohmann::json::parse(response->body);
+        auto result = jsonResponse["result"];
+
+        if (!result.is_object() || !result["ocrResults"].is_array()) {
+            std::cerr << "Unexpected response structure." << std::endl;
+            return 1;
+        }
+
+        for (size_t i = 0; i < result["ocrResults"].size(); ++i) {
+            auto ocrResult = result["ocrResults"][i];
+            std::cout << ocrResult["prunedResult"] << std::endl;
+
+            std::string ocrImgPath = "ocr_" + std::to_string(i) + ".jpg";
+            std::string encodedImage = ocrResult["ocrImage"];
+            std::string decodedImage = base64::from_base64(encodedImage);
+
+            std::ofstream outputImage(ocrImgPath, std::ios::binary);
+            if (outputImage.is_open()) {
+                outputImage.write(decodedImage.c_str(), static_cast<std::streamsize>(decodedImage.size()));
+                outputImage.close();
+                std::cout << "Output image saved at " << ocrImgPath << std::endl;
+            } else {
+                std::cerr << "Unable to open file for writing: " << ocrImgPath << std::endl;
+            }
+        }
+    } else {
+        std::cerr << "Failed to send HTTP request." << std::endl;
+        if (response) {
+            std::cerr << "HTTP status code: " << response->status << std::endl;
+            std::cerr << "Response body: " << response->body << std::endl;
+        }
+        return 1;
+    }
+
+    return 0;
+}
+</code></pre></details>
+
+<details><summary>Java</summary>
+
+<pre><code class="language-java">import okhttp3.*;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.util.Base64;
+
+public class Main {
+    public static void main(String[] args) throws IOException {
+        String API_URL = "http://localhost:8080/ocr"; 
+        String imagePath = "./demo.jpg"; 
+
+        File file = new File(imagePath);
+        byte[] fileContent = java.nio.file.Files.readAllBytes(file.toPath());
+        String base64Image = Base64.getEncoder().encodeToString(fileContent);
+
+        ObjectMapper objectMapper = new ObjectMapper();
+        ObjectNode payload = objectMapper.createObjectNode();
+        payload.put("file", base64Image); 
+        payload.put("fileType", 1); 
+
+        OkHttpClient client = new OkHttpClient();
+        MediaType JSON = MediaType.get("application/json; charset=utf-8");
+	RequestBody body = RequestBody.create(JSON, payload.toString());
+
+        Request request = new Request.Builder()
+                .url(API_URL)
+                .post(body)
+                .build();
+
+        try (Response response = client.newCall(request).execute()) {
+            if (response.isSuccessful()) {
+                String responseBody = response.body().string();
+                JsonNode root = objectMapper.readTree(responseBody);
+                JsonNode result = root.get("result");
+
+                JsonNode ocrResults = result.get("ocrResults");
+                for (int i = 0; i < ocrResults.size(); i++) {
+                    JsonNode item = ocrResults.get(i);
+
+                    JsonNode prunedResult = item.get("prunedResult");
+                    System.out.println("Pruned Result [" + i + "]: " + prunedResult.toString());
+
+                    String ocrImageBase64 = item.get("ocrImage").asText();
+                    byte[] ocrImageBytes = Base64.getDecoder().decode(ocrImageBase64);
+                    String ocrImgPath = "ocr_result_" + i + ".jpg";
+                    try (FileOutputStream fos = new FileOutputStream(ocrImgPath)) {
+                        fos.write(ocrImageBytes);
+                        System.out.println("Saved OCR image to: " + ocrImgPath);
+                    }
+                }
+            } else {
+                System.err.println("Request failed with HTTP code: " + response.code());
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Go</summary>
+
+<pre><code class="language-go">package main
+
+import (
+    "bytes"
+    "encoding/base64"
+    "encoding/json"
+    "fmt"
+    "io/ioutil"
+    "net/http"
+)
+
+func main() {
+    API_URL := "http://localhost:8080/ocr"
+    filePath := "./demo.jpg"
+
+    fileBytes, err := ioutil.ReadFile(filePath)
+    if err != nil {
+        fmt.Printf("Error reading file: %v\n", err)
+        return
+    }
+    fileData := base64.StdEncoding.EncodeToString(fileBytes)
+
+    payload := map[string]interface{}{
+        "file":     fileData,
+        "fileType": 1,
+    }
+    payloadBytes, err := json.Marshal(payload)
+    if err != nil {
+        fmt.Printf("Error marshaling payload: %v\n", err)
+        return
+    }
+
+    client := &http.Client{}
+    req, err := http.NewRequest("POST", API_URL, bytes.NewBuffer(payloadBytes))
+    if err != nil {
+        fmt.Printf("Error creating request: %v\n", err)
+        return
+    }
+    req.Header.Set("Content-Type", "application/json")
+
+    res, err := client.Do(req)
+    if err != nil {
+        fmt.Printf("Error sending request: %v\n", err)
+        return
+    }
+    defer res.Body.Close()
+
+    if res.StatusCode != http.StatusOK {
+        fmt.Printf("Unexpected status code: %d\n", res.StatusCode)
+        return
+    }
+
+    body, err := ioutil.ReadAll(res.Body)
+    if err != nil {
+        fmt.Printf("Error reading response body: %v\n", err)
+        return
+    }
+
+    type OcrResult struct {
+        PrunedResult map[string]interface{} `json:"prunedResult"` 
+        OcrImage     *string                `json:"ocrImage"`     
+    }
+
+    type Response struct {
+        Result struct {
+            OcrResults []OcrResult `json:"ocrResults"`
+            DataInfo   interface{} `json:"dataInfo"` 
+        } `json:"result"`
+    }
+
+    var respData Response
+    if err := json.Unmarshal(body, &respData); err != nil {
+        fmt.Printf("Error unmarshaling response: %v\n", err)
+        return
+    }
+
+    for i, res := range respData.Result.OcrResults {
+        
+        if res.OcrImage != nil {
+            imgBytes, err := base64.StdEncoding.DecodeString(*res.OcrImage)
+            if err != nil {
+                fmt.Printf("Error decoding image %d: %v\n", i, err)
+                continue
+            }
+            
+            filename := fmt.Sprintf("ocr_%d.jpg", i)
+            if err := ioutil.WriteFile(filename, imgBytes, 0644); err != nil {
+                fmt.Printf("Error saving image %s: %v\n", filename, err)
+                continue
+            }
+            fmt.Printf("Output image saved at %s\n", filename)
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>C#</summary>
+
+<pre><code class="language-csharp">using System;
+using System.IO;
+using System.Net.Http;
+using System.Text;
+using System.Threading.Tasks;
+using Newtonsoft.Json.Linq;
+
+class Program
+{
+    static readonly string API_URL = "http://localhost:8080/ocr";
+    static readonly string inputFilePath = "./demo.jpg";
+
+    static async Task Main(string[] args)
+    {
+        var httpClient = new HttpClient();
+
+        byte[] fileBytes = File.ReadAllBytes(inputFilePath);
+        string fileData = Convert.ToBase64String(fileBytes);
+
+        var payload = new JObject
+        {
+            { "file", fileData },
+            { "fileType", 1 }
+        };
+        var content = new StringContent(payload.ToString(), Encoding.UTF8, "application/json");
+
+        HttpResponseMessage response = await httpClient.PostAsync(API_URL, content);
+        response.EnsureSuccessStatusCode();
+
+        string responseBody = await response.Content.ReadAsStringAsync();
+        JObject jsonResponse = JObject.Parse(responseBody);
+
+        JArray ocrResults = (JArray)jsonResponse["result"]["ocrResults"];
+        for (int i = 0; i < ocrResults.Count; i++)
+        {
+            var res = ocrResults[i];
+            Console.WriteLine($"[{i}] prunedResult:\n{res["prunedResult"]}");
+
+            string base64Image = res["ocrImage"]?.ToString();
+            if (!string.IsNullOrEmpty(base64Image))
+            {
+                string outputPath = $"ocr_{i}.jpg";
+                byte[] imageBytes = Convert.FromBase64String(base64Image);
+                File.WriteAllBytes(outputPath, imageBytes);
+                Console.WriteLine($"OCR image saved to {outputPath}");
+            }
+            else
+            {
+                Console.WriteLine($"OCR image at index {i} is null.");
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Node.js</summary>
+
+<pre><code class="language-js">const axios = require('axios');
+const fs = require('fs');
+const path = require('path');
+
+const API_URL = 'http://localhost:8080/layout-parsing';
+const imagePath = './demo.jpg';  
+const fileType = 1;             
+
+function encodeImageToBase64(filePath) {
+  const bitmap = fs.readFileSync(filePath);
+  return Buffer.from(bitmap).toString('base64');
+}
+
+const payload = {
+  file: encodeImageToBase64(imagePath),
+  fileType: fileType
+};
+
+axios.post(API_URL, payload)
+  .then(response => {
+    const results = response.data.result.layoutParsingResults;
+    results.forEach((res, index) => {
+      console.log(`\n[${index}] prunedResult:`);
+      console.log(res.prunedResult);
+
+      const outputImages = res.outputImages;
+      if (outputImages) {
+        Object.entries(outputImages).forEach(([imgName, base64Img]) => {
+          const imgPath = `${imgName}_${index}.jpg`;
+          fs.writeFileSync(imgPath, Buffer.from(base64Img, 'base64'));
+          console.log(`Output image saved at ${imgPath}`);
+        });
+      } else {
+        console.log(`[${index}] No outputImages.`);
+      }
+    });
+  })
+  .catch(error => {
+    console.error('Error during API request:', error.message || error);
+  });
+</code></pre></details>
+
+<details><summary>PHP</summary>
+
+<pre><code class="language-php">&lt;?php
+
+$API_URL = "http://localhost:8080/ocr"; 
+$image_path = "./demo.jpg"; 
+
+$image_data = base64_encode(file_get_contents($image_path));
+$payload = array(
+    "file" => $image_data,
+    "fileType" => 1 
+);
+
+$ch = curl_init($API_URL);
+curl_setopt($ch, CURLOPT_POST, true);
+curl_setopt($ch, CURLOPT_POSTFIELDS, json_encode($payload));
+curl_setopt($ch, CURLOPT_HTTPHEADER, array('Content-Type: application/json'));
+curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+$response = curl_exec($ch);
+curl_close($ch);
+
+$result = json_decode($response, true)["result"]["ocrResults"];
+
+foreach ($result as $i => $item) {
+    echo "[$i] prunedResult:\n";
+    print_r($item["prunedResult"]);
+
+    if (!empty($item["ocrImage"])) {
+        $output_img_path = "ocr_{$i}.jpg";
+        file_put_contents($output_img_path, base64_decode($item["ocrImage"]));
+        echo "OCR image saved at $output_img_path\n";
+    } else {
+        echo "No ocrImage found for item $i\n";
+    }
+}
+?&gt;
+</code></pre></details>
+</details>
+
+## 4. 二次开发
+如果 通用OCR 产线提供的默认模型权重在您的场景中，精度或速度不满意，您可以尝试利用<b>您自己拥有的特定领域或应用场景的数据</b>对现有模型进行进一步的<b>微调</b>，以提升 通用OCR 产线的在您的场景中的识别效果。
+
+### 4.1 模型微调
+通用 OCR 产线包含若干模块，模型产线的效果如果不及预期，可能来自于其中任何一个模块。您可以对识别效果差的图片进行分析，进而确定是哪个模块存在问题，并参考以下表格中对应的微调教程链接进行模型微调。
+
+<table>
+<thead>
+<tr>
+<th>情形</th>
+<th>微调模块</th>
+<th>微调参考链接</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>整图旋转矫正不准</td>
+<td>文档图像方向分类模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>图像扭曲矫正不准</td>
+<td>文本图像矫正模块</td>
+<td>暂不支持微调</td>
+</tr>
+<tr>
+<td>文本行旋转矫正不准</td>
+<td>文本行方向分类模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/textline_orientation_classification.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>文本漏检</td>
+<td>文本检测模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/text_detection.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>文本内容不准</td>
+<td>文本识别模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/text_recognition.html#_5">链接</a></td>
+</tr>
+</tbody>
+</table>
+
+### 4.2 模型应用
+
+当您使用私有数据集完成微调训练后，可获得本地模型权重文件，然后可以通过参数指定本地模型保存路径的方式，或者通过自定义产线配置文件的方式，使用微调后的模型权重。
+
+#### 4.2.1 通过参数指定本地模型路径
+
+在初始化产线对象时，通过参数指定本地模型路径。以文本检测模型微调后的权重的使用方法为例，示例如下：
+
+命令行方式:
+
+```bash
+# 通过 --text_detection_model_dir 指定本地模型路径
+paddleocr ocr -i ./general_ocr_002.png --text_detection_model_dir your_det_model_path
+
+# 通过 --text_detection_model_name 修改模型名称
+paddleocr ocr -i ./general_ocr_002.png --text_detection_model_name PP-OCRv5_mobile_det --text_detection_model_dir your_v5_mobile_det_model_path
+```
+
+脚本方式：
+
+```python
+
+from paddleocr import PaddleOCR
+
+# 通过 text_detection_model_dir 指定本地模型路径
+pipeline = PaddleOCR(text_detection_model_dir="./your_det_model_path")
+
+# 通过 text_detection_model_name 修改模型名称
+# pipeline = PaddleOCR(text_detection_model_name="PP-OCRv5_mobile_det", text_detection_model_dir="./your_v5_mobile_det_model_path")
+
+```
+
+
+#### 4.2.2 通过配置文件指定本地模型路径
+
+
+1.获取产线配置文件
+
+可调用 PaddleOCR 中 通用 OCR 产线对象的 `export_paddlex_config_to_yaml` 方法，将当前产线配置导出为 YAML 文件：
+
+```Python
+from paddleocr import PaddleOCR
+
+pipeline = PaddleOCR()
+pipeline.export_paddlex_config_to_yaml("PaddleOCR.yaml")
+```
+
+2.修改配置文件
+
+在得到默认的产线配置文件后，将微调后模型权重的本地路径替换至产线配置文件中的对应位置即可。例如
+
+```yaml
+......
+SubModules:
+  TextDetection:
+    box_thresh: 0.6
+    limit_side_len: 64
+    limit_type: min
+    max_side_limit: 4000
+    model_dir: null # 替换为微调后的文本测模型权重路径
+    model_name: PP-OCRv6_medium_det # 如果微调的模型名称与默认模型名称不同，请一并修改此处
+    module_name: text_detection
+    thresh: 0.3
+    unclip_ratio: 1.5
+  TextLineOrientation:
+    batch_size: 6
+    model_dir: null  # 替换为微调后的文本行方向分类模型权重路径
+    model_name: PP-LCNet_x1_0_textline_ori # 如果微调的模型名称与默认模型名称不同，请一并修改此处
+    module_name: textline_orientation
+  TextRecognition:
+    batch_size: 6
+    model_dir: null # 替换为微调后的文本识模型权重路径
+    model_name: PP-OCRv6_medium_rec # 如果微调的模型名称与默认模型名称不同，请一并修改此处
+    module_name: text_recognition
+    score_thresh: 0.0
+......
+```
+
+在产线配置文件中，不仅包含 PaddleOCR CLI 和 Python API 支持的参数，还可进行更多高级配置，具体信息可在 [PaddleX模型产线使用概览](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/pipeline_develop_guide.html) 中找到对应的产线使用教程，参考其中的详细说明，根据需求调整各项配置。
+
+3.在 CLI 中加载产线配置文件
+
+在修改完成配置文件后，通过命令行的 `--paddlex_config` 参数指定修改后的产线配置文件的路径，PaddleOCR 会读取其中的内容作为产线配置。示例如下：
+
+```bash
+paddleocr ocr --paddlex_config PaddleOCR.yaml ...
+```
+
+4.在 Python API 中加载产线配置文件
+
+初始化产线对象时，可通过 `paddlex_config` 参数传入 PaddleX 产线配置文件路径或配置dict，PaddleOCR 会读取其中的内容作为产线配置。示例如下：
+
+```python
+from paddleocr import PaddleOCR
+
+pipeline = PaddleOCR(paddlex_config="PaddleOCR.yaml")
+```
+
+## 5. 附录
+
+<details><summary><b>支持语言</b></summary>
+
+<table border="1" cellspacing="0" cellpadding="4">
+  <thead>
+    <tr>
+      <th><code>lang</code></th>
+      <th>语言名称</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr><td><code>abq</code></td><td>阿布哈兹文</td></tr>
+    <tr><td><code>af</code></td><td>南非荷兰文</td></tr>
+    <tr><td><code>ang</code></td><td>古英文</td></tr>
+    <tr><td><code>ar</code></td><td>阿拉伯文</td></tr>
+    <tr><td><code>ava</code></td><td>阿瓦尔文</td></tr>
+    <tr><td><code>az</code></td><td>阿塞拜疆文</td></tr>
+    <tr><td><code>be</code></td><td>白俄罗斯文</td></tr>
+    <tr><td><code>bg</code></td><td>保加利亚文</td></tr>
+    <tr><td><code>bgc</code></td><td>哈里亚纳文</td></tr>
+    <tr><td><code>bh</code></td><td>比哈尔文</td></tr>
+    <tr><td><code>bho</code></td><td>博杰普尔文</td></tr>
+    <tr><td><code>bs</code></td><td>波斯尼亚文</td></tr>
+    <tr><td><code>ch</code></td><td>简体中文</td></tr>
+    <tr><td><code>che</code></td><td>车臣文</td></tr>
+    <tr><td><code>chinese_cht</code></td><td>繁体中文</td></tr>
+    <tr><td><code>cs</code></td><td>捷克文</td></tr>
+    <tr><td><code>cy</code></td><td>威尔士文</td></tr>
+    <tr><td><code>da</code></td><td>丹麦文</td></tr>
+    <tr><td><code>dar</code></td><td>达尔格瓦文</td></tr>
+    <tr><td><code>de</code> or <code>german</code></td><td>德文</td></tr>
+    <tr><td><code>en</code></td><td>英文</td></tr>
+    <tr><td><code>es</code></td><td>西班牙文</td></tr>
+    <tr><td><code>et</code></td><td>爱沙尼亚文</td></tr>
+    <tr><td><code>fa</code></td><td>波斯文</td></tr>
+    <tr><td><code>fr</code> or <code>french</code></td><td>法文</td></tr>
+    <tr><td><code>ga</code></td><td>爱尔兰文</td></tr>
+    <tr><td><code>gom</code></td><td>孔卡尼文</td></tr>
+    <tr><td><code>hi</code></td><td>印地文</td></tr>
+    <tr><td><code>hr</code></td><td>克罗地亚文</td></tr>
+    <tr><td><code>hu</code></td><td>匈牙利文</td></tr>
+    <tr><td><code>id</code></td><td>印尼文</td></tr>
+    <tr><td><code>inh</code></td><td>印古什文</td></tr>
+    <tr><td><code>is</code></td><td>冰岛文</td></tr>
+    <tr><td><code>it</code></td><td>意大利文</td></tr>
+    <tr><td><code>japan</code></td><td>日文</td></tr>
+    <tr><td><code>ka</code></td><td>格鲁吉亚文</td></tr>
+    <tr><td><code>kbd</code></td><td>卡巴尔达文</td></tr>
+    <tr><td><code>korean</code></td><td>韩文</td></tr>
+    <tr><td><code>ku</code></td><td>库尔德文</td></tr>
+    <tr><td><code>la</code></td><td>拉丁文</td></tr>
+    <tr><td><code>lbe</code></td><td>拉克文</td></tr>
+    <tr><td><code>lez</code></td><td>列兹金文</td></tr>
+    <tr><td><code>lt</code></td><td>立陶宛文</td></tr>
+    <tr><td><code>lv</code></td><td>拉脱维亚文</td></tr>
+    <tr><td><code>mah</code></td><td>马加希文</td></tr>
+    <tr><td><code>mai</code></td><td>迈蒂利文</td></tr>
+    <tr><td><code>mi</code></td><td>毛利文</td></tr>
+    <tr><td><code>mn</code></td><td>蒙古文</td></tr>
+    <tr><td><code>mr</code></td><td>马拉地文</td></tr>
+    <tr><td><code>ms</code></td><td>马来文</td></tr>
+    <tr><td><code>mt</code></td><td>马耳他文</td></tr>
+    <tr><td><code>ne</code></td><td>尼泊尔文</td></tr>
+    <tr><td><code>new</code></td><td>尼瓦尔文</td></tr>
+    <tr><td><code>nl</code></td><td>荷兰文</td></tr>
+    <tr><td><code>no</code></td><td>挪威文</td></tr>
+    <tr><td><code>oc</code></td><td>奥克文</td></tr>
+    <tr><td><code>pi</code></td><td>巴利文</td></tr>
+    <tr><td><code>pl</code></td><td>波兰文</td></tr>
+    <tr><td><code>pt</code></td><td>葡萄牙文</td></tr>
+    <tr><td><code>ro</code></td><td>罗马尼亚文</td></tr>
+    <tr><td><code>rs_cyrillic</code></td><td>塞尔维亚语西里尔字母</td></tr>
+    <tr><td><code>rs_latin</code></td><td>塞尔维亚语拉丁字母</td></tr>
+    <tr><td><code>ru</code></td><td>俄文</td></tr>
+    <tr><td><code>sa</code></td><td>梵文</td></tr>
+    <tr><td><code>sck</code></td><td>萨达里文</td></tr>
+    <tr><td><code>sk</code></td><td>斯洛伐克文</td></tr>
+    <tr><td><code>sl</code></td><td>斯洛文尼亚文</td></tr>
+    <tr><td><code>sq</code></td><td>阿尔巴尼亚文</td></tr>
+    <tr><td><code>sv</code></td><td>瑞典文</td></tr>
+    <tr><td><code>sw</code></td><td>斯瓦希里文</td></tr>
+    <tr><td><code>tab</code></td><td>塔巴萨兰文</td></tr>
+    <tr><td><code>ta</code></td><td>泰米尔文</td></tr>
+    <tr><td><code>te</code></td><td>泰卢固文</td></tr>
+    <tr><td><code>tl</code></td><td>塔加洛文</td></tr>
+    <tr><td><code>tr</code></td><td>土耳其文</td></tr>
+    <tr><td><code>ug</code></td><td>维吾尔文</td></tr>
+    <tr><td><code>uk</code></td><td>乌克兰文</td></tr>
+    <tr><td><code>ur</code></td><td>乌尔都文</td></tr>
+    <tr><td><code>uz</code></td><td>乌兹别克文</td></tr>
+    <tr><td><code>vi</code></td><td>越南文</td></tr>
+  </tbody>
+</table>
+
+</details>
+
+<details><summary><b>OCR模型版本与支持语言的对应关系</b></summary>
+
+<table border="1" cellspacing="0" cellpadding="4">
+  <thead>
+    <tr>
+      <th><code>ocr_version</code></th>
+      <th>Supported <code>lang</code></th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td><code>PP-OCRv6</code></td>
+      <td><code>ch</code>, <code>chinese_cht</code>, <code>en</code>, <code>japan</code>, <code>af</code>, <code>az</code>, <code>bs</code>, <code>ca</code>, <code>cs</code>, <code>cy</code>, <code>da</code>, <code>de</code>, <code>es</code>, <code>et</code>, <code>eu</code>, <code>fi</code>, <code>fr</code>, <code>ga</code>, <code>gl</code>, <code>hr</code>, <code>hu</code>, <code>id</code>, <code>is</code>, <code>it</code>, <code>ku</code>, <code>la</code>, <code>lb</code>, <code>lt</code>, <code>lv</code>, <code>mi</code>, <code>ms</code>, <code>mt</code>, <code>nl</code>, <code>no</code>, <code>oc</code>, <code>pl</code>, <code>pt</code>, <code>qu</code>, <code>rm</code>, <code>ro</code>, <code>rs_latin</code>, <code>sk</code>, <code>sl</code>, <code>sq</code>, <code>sv</code>, <code>sw</code>, <code>tl</code>, <code>tr</code>, <code>uz</code>, <code>vi</code>, <code>french</code>, <code>german</code>。<code>PP-OCRv6_tiny</code> 不支持 <code>japan</code>，需显式指定 <code>PP-OCRv6_tiny_det</code>/<code>PP-OCRv6_tiny_rec</code>。</td>
+    </tr>
+    <tr>
+      <td><code>PP-OCRv5</code></td>
+      <td><code>ch</code>, <code>en</code>, <code>fr</code>, <code>de</code>, <code>japan</code>, <code>korean</code>, <code>chinese_cht</code>, <code>af</code>, <code>it</code>, <code>es</code>, <code>bs</code>, <code>pt</code>, <code>cs</code>, <code>cy</code>, <code>da</code>, <code>et</code>, <code>ga</code>, <code>hr</code>, <code>hu</code>, <code>rslatin</code>, <code>id</code>, <code>oc</code>, <code>is</code>, <code>lt</code>, <code>mi</code>, <code>ms</code>, <code>nl</code>, <code>no</code>, <code>pl</code>, <code>sk</code>, <code>sl</code>, <code>sq</code>, <code>sv</code>, <code>sw</code>, <code>tl</code>, <code>tr</code>, <code>uz</code>, <code>la</code>, <code>ru</code>, <code>be</code>, <code>uk</code>, <code>th</code>, <code>el</code>, <code>az</code>, <code>ku</code>, <code>lv</code>, <code>mt</code>, <code>pi</code>, <code>ro</code>, <code>vi</code>, <code>fi</code>, <code>eu</code>, <code>gl</code>, <code>lb</code>, <code>rm</code>, <code>ca</code>, <code>qu</code>, <code>te</code>, <code>sr</code>, <code>bg</code>, <code>mn</code>, <code>ab</code>, <code>ady</code>, <code>kbd</code>, <code>av</code>, <code>dar</code>, <code>inh</code>, <code>ce</code>, <code>lki</code>, <code>lez</code>, <code>tab</code>, <code>kk</code>, <code>ky</code>, <code>tg</code>, <code>mk</code>, <code>tt</code>, <code>cv</code>, <code>ba</code>, <code>mhr</code>, <code>mo</code>, <code>udm</code>, <code>kv</code>, <code>os</code>, <code>bua</code>, <code>xal</code>, <code>tyv</code>, <code>sah</code>, <code>kaa</code>, <code>ar</code>, <code>fa</code>, <code>ug</code>, <code>ur</code>, <code>ps</code>, <code>ku</code>, <code>sd</code>, <code>bal</code>, <code>hi</code>, <code>mr</code>, <code>ne</code>, <code>bh</code>, <code>mai</code>, <code>ang</code>, <code>bho</code>, <code>mah</code>, <code>sck</code>, <code>new</code>, <code>gom</code>, <code>sa</code>, <code>bgc</code>, <code>ta</code></td>
+    </tr>
+    <tr>
+      <td><code>PP-OCRv4</code></td>
+      <td><code>ch</code>, <code>en</code></td>
+    </tr>
+    <tr>
+      <td><code>PP-OCRv3</code></td>
+      <td>
+        <code>abq</code>, <code>af</code>, <code>ady</code>, <code>ang</code>, <code>ar</code>, <code>ava</code>, <code>az</code>, <code>be</code>,
+        <code>bg</code>, <code>bgc</code>, <code>bh</code>, <code>bho</code>, <code>bs</code>, <code>ch</code>, <code>che</code>,
+        <code>chinese_cht</code>, <code>cs</code>, <code>cy</code>, <code>da</code>, <code>dar</code>, <code>de</code>, <code>german</code>,
+        <code>en</code>, <code>es</code>, <code>et</code>, <code>fa</code>, <code>fr</code>, <code>french</code>, <code>ga</code>, <code>gom</code>,
+        <code>hi</code>, <code>hr</code>, <code>hu</code>, <code>id</code>, <code>inh</code>, <code>is</code>, <code>it</code>, <code>japan</code>,
+        <code>ka</code>, <code>kbd</code>, <code>korean</code>, <code>ku</code>, <code>la</code>, <code>lbe</code>, <code>lez</code>, <code>lt</code>,
+        <code>lv</code>, <code>mah</code>, <code>mai</code>, <code>mi</code>, <code>mn</code>, <code>mr</code>, <code>ms</code>, <code>mt</code>,
+        <code>ne</code>, <code>new</code>, <code>nl</code>, <code>no</code>, <code>oc</code>, <code>pi</code>, <code>pl</code>, <code>pt</code>,
+        <code>ro</code>, <code>rs_cyrillic</code>, <code>rs_latin</code>, <code>ru</code>, <code>sa</code>, <code>sck</code>, <code>sk</code>,
+        <code>sl</code>, <code>sq</code>, <code>sv</code>, <code>sw</code>, <code>ta</code>, <code>tab</code>, <code>te</code>, <code>tl</code>,
+        <code>tr</code>, <code>ug</code>, <code>uk</code>, <code>ur</code>, <code>uz</code>, <code>vi</code>
+      </td>
+    </tr>
+  </tbody>
+</table>
+
+</details>
+
+## 5. 常见问题与解决方案
+
+### 5.1 性能优化问题
+
+#### Q: GPU推理速度慢怎么办？
+
+**A**: 可以通过以下方式优化：
+（1）启用高性能推理：设置`enable_hpi=True`，自动选择最优加速策略
+（2）启用TensorRT加速：设置`use_tensorrt=True`，需要CUDA 11.8+和TensorRT 8.6+
+（3）使用半精度：设置`precision="fp16"`，可以显著提升速度
+（4）调整批处理大小：根据显存大小设置合适的`batch_size`
+（5）使用移动端模型：在精度要求不高时使用`PP-OCRv5_mobile`系列模型
+
+#### Q: GPU内存不足（CUDA out of memory）怎么办？
+
+**A**: 可以通过以下方式解决：
+（1）减小批处理大小：将`batch_size`设置为1
+（2）减小图像尺寸：设置`det_limit_side_len=640`
+（3）启用内存优化：设置`enable_memory_optim=True`
+（4）限制GPU内存使用：设置`gpu_mem=200`
+（5）使用移动端模型：切换到`PP-OCRv5_mobile`系列模型
+
+### 5.2 模型选择问题
+
+#### Q: 如何选择合适的模型？
+
+**A**: 根据应用场景选择：
+- 服务器高精度场景：使用`PP-OCRv5_server`系列，精度最高
+- 移动端部署：使用`PP-OCRv5_mobile`系列，模型小速度快
+- 实时处理：使用`PP-OCRv5_mobile`系列，推理速度快
+- 批量处理：使用`PP-OCRv5_server`系列，精度高
+- 多语言识别：拉丁语系等可使用 `PP-OCRv6`；阿拉伯文、西里尔文、天城文等语系请使用 `PP-OCRv5`
+
+### 5.3 最佳实践
+
+#### Q: 生产环境部署建议？
+
+**A**: 
+（1）使用高性能推理配置
+（2）启用TensorRT加速
+（3）合理设置批处理大小
+（4）监控GPU内存使用情况
+（5）做好错误处理和日志记录
+
+## 6. 总结
+
+通用OCR产线是PaddleOCR的核心功能，通过合理选择模型和配置参数，可以在不同场景下获得最佳的识别效果。建议根据部署环境选择合适的模型，根据应用需求调整参数，充分利用GPU加速和批处理优化。
diff --git a/docs/version3.x/pipeline_usage/PP-ChatOCRv4.en.md b/docs/version3.x/pipeline_usage/PP-ChatOCRv4.en.md
new file mode 100644
index 0000000..489d30e
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PP-ChatOCRv4.en.md
@@ -0,0 +1,4199 @@
+---
+comments: true
+---
+# PP-ChatOCRv4-doc Pipeline Usage Tutorial
+
+## 1. Introduction to PP-ChatOCRv4-doc Pipeline
+PP-ChatOCRv4-doc is a unique document and image intelligent analysis solution from PaddlePaddle, combining LLM, MLLM, and OCR technologies to address complex document information extraction challenges such as layout analysis, rare characters, multi-page PDFs, tables, and seal text recognition. Integrated with ERNIE Bot, it fuses massive data and knowledge, achieving high accuracy and wide applicability. This pipeline also provides flexible service deployment options, supporting deployment on various hardware. Furthermore, it offers custom development capabilities, allowing you to train and fine-tune models on your own datasets, with seamless integration of trained models.
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/PP-ChatOCRv4/algorithm_ppchatocrv4.png" width="80%"/>
+</div>
+
+The PP-ChatOCRv4 pipeline includes the following 9 modules. Each module can be trained and inferred independently and includes multiple models. For more details, please click on the respective module to view the documentation.
+
+- [Document Image Orientation Classification Module](../module_usage/doc_img_orientation_classification.en.md) (Optional)
+- [Text Image Unwarping Module](../module_usage/text_image_unwarping.en.md) (Optional)
+- [Layout Detection Module](../module_usage/layout_detection.en.md)
+- [Table Structure Recognition Module](../module_usage/table_structure_recognition.en.md) (Optional)
+- [Text Detection Module](../module_usage/text_detection.en.md)
+- [Text Recognition Module](../module_usage/text_recognition.en.md)
+- [Text Line Orientation Classification Module](../module_usage/textline_orientation_classification.en.md)(Optional)
+- [Formula Recognition Module](../module_usage/formula_recognition.en.md) (Optional)
+- [Seal Text Detection Module](../module_usage/seal_text_detection.en.md) (Optional)
+
+
+In this pipeline, you can choose the model to use based on the benchmark data below.
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing.
+> In the inference time columns labeled [Normal Mode / High-Performance Mode], the `Normal Mode` values correspond to local Paddle inference engines. Each module selects the appropriate local Paddle inference engine according to the default model name: models that support only dynamic graph use `paddle_dynamic`, while models that support both static and dynamic graph prefer `paddle_static`.
+
+<details>
+<summary><b>Document Image Orientation Classification Module (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Top-1 Acc (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High-Performance Mode]</th>
+<th>Model Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">Training Model</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>Document image classification model based on PP-LCNet_x1_0, with four categories: 0°, 90°, 180°, and 270°.</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Text Image Unwarp Module (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>CER</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">Training Model</a></td>
+<td>0.179</td>
+<td>19.05 / 19.05</td>
+<td>- / 869.82</td>
+<td>30.3</td>
+<td>High-precision Text Image Unwarping model.</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+
+<details>
+<summary><b>Layout Detection Module Model:</b></summary>
+* <b>The layout detection model includes 20 common categories: document title, paragraph title, text, page number, abstract, table, references, footnotes, header, footer, algorithm, formula, formula number, image, table, seal, figure_table title, chart, and sidebar text and lists of references</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout_plus-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout_plus-L_pretrained.pdparams">Training Model</a></td>
+<td>83.2</td>
+<td>53.03 / 17.23</td>
+<td>634.62 / 378.32</td>
+<td>126.01</td>
+<td>A higher-precision layout area localization model trained on a self-built dataset containing Chinese and English papers, PPT, multi-layout magazines, contracts, books, exams, ancient books and research reports using RT-DETR-L</td>
+</tr>
+<tr>
+</tbody>
+</table>
+
+
+* <b>The layout detection model includes 1 category: Block:</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocBlockLayout</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBlockLayout_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocBlockLayout_pretrained.pdparams">Training Model</a></td>
+<td>95.9</td>
+<td>34.60 / 28.54</td>
+<td>506.43 / 256.83</td>
+<td>123.92</td>
+<td>A layout block localization model trained on a self-built dataset containing Chinese and English papers, PPT, multi-layout magazines, contracts, books, exams, ancient books and research reports using RT-DETR-L</td>
+</tr>
+<tr>
+</tbody>
+</table>
+
+
+* <b>The layout detection model includes 23 common categories: document title, paragraph title, text, page number, abstract, table of contents, references, footnotes, header, footer, algorithm, formula, formula number, image, figure caption, table, table caption, seal, figure title, figure, header image, footer image, and sidebar text</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-L_pretrained.pdparams">Pretrained Model</a></td>
+<td>90.4</td>
+<td>33.59 / 33.59</td>
+<td>503.01 / 251.08</td>
+<td>123.76</td>
+<td>A high-precision layout area localization model trained on a self-built dataset containing Chinese and English papers, magazines, contracts, books, exams, and research reports using RT-DETR-L.</td>
+</tr>
+<tr>
+<td>PP-DocLayout-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-M_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-M_pretrained.pdparams">Pretrained Model</a></td>
+<td>75.2</td>
+<td>13.03 / 4.72</td>
+<td>43.39 / 24.44</td>
+<td>22.578</td>
+<td>A layout area localization model with balanced precision and efficiency, trained on a self-built dataset containing Chinese and English papers, magazines, contracts, books, exams, and research reports using PicoDet-L.</td>
+</tr>
+<tr>
+<td>PP-DocLayout-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-S_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-S_pretrained.pdparams">Pretrained Model</a></td>
+<td>70.9</td>
+<td>11.54 / 3.86</td>
+<td>18.53 / 6.29</td>
+<td>4.834</td>
+<td>A high-efficiency layout area localization model trained on a self-built dataset containing Chinese and English papers, magazines, contracts, books, exams, and research reports using PicoDet-S.</td>
+</tr>
+</tbody>
+</table>
+
+> ❗ The above list includes the <b>4 core models</b> that are key supported by the text recognition module. The module actually supports a total of <b>12 full models</b>, including several predefined models with different categories. The complete model list is as follows:
+
+<details><summary> 👉 Details of Model List</summary>
+
+* <b>Table Layout Detection Model</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet_layout_1x_table</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_table_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_table_pretrained.pdparams">Training Model</a></td>
+<td>97.5</td>
+<td>9.57 / 6.63</td>
+<td>27.66 / 16.75</td>
+<td>7.4</td>
+<td>A high-efficiency layout area localization model trained on a self-built dataset using PicoDet-1x, capable of detecting table regions.</td>
+</tr>
+</tbody></table>
+
+* <b>3-Class Layout Detection Model, including Table, Image, and Stamp</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_3cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_3cls_pretrained.pdparams">Training Model</a></td>
+<td>88.2</td>
+<td>8.43 / 3.44</td>
+<td>17.60 / 6.51</td>
+<td>4.8</td>
+<td>A high-efficiency layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using PicoDet-S.</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_3cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_3cls_pretrained.pdparams">Training Model</a></td>
+<td>89.0</td>
+<td>12.80 / 9.57</td>
+<td>45.04 / 23.86</td>
+<td>22.6</td>
+<td>A balanced efficiency and precision layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using PicoDet-L.</td>
+</tr>
+</table>
+
+<p><b>Table Classification Module Models:</b></p>
+<table>
+<tr>
+<td>RT-DETR-H_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_3cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_3cls_pretrained.pdparams">Training Model</a></td>
+<td>95.8</td>
+<td>114.80 / 25.65</td>
+<td>924.38 / 924.38</td>
+<td>470.1</td>
+<td>A high-precision layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using RT-DETR-H.</td>
+</tr>
+</tbody></table>
+
+* <b>5-Class English Document Area Detection Model, including Text, Title, Table, Image, and List</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet_layout_1x</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_pretrained.pdparams">Training Model</a></td>
+<td>97.8</td>
+<td>9.62 / 6.75</td>
+<td>26.96 / 12.77</td>
+<td>7.4</td>
+<td>A high-efficiency English document layout area localization model trained on the PubLayNet dataset using PicoDet-1x.</td>
+</tr>
+</tbody></table>
+
+* <b>17-Class Area Detection Model, including 17 common layout categories: Paragraph Title, Image, Text, Number, Abstract, Content, Figure Caption, Formula, Table, Table Caption, References, Document Title, Footnote, Header, Algorithm, Footer, and Stamp</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>87.4</td>
+<td>8.80 / 3.62</td>
+<td>17.51 / 6.35</td>
+<td>4.8</td>
+<td>A high-efficiency layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using PicoDet-S.</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>89.0</td>
+<td>12.60 / 10.27</td>
+<td>43.70 / 24.42</td>
+<td>22.6</td>
+<td>A balanced efficiency and precision layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using PicoDet-L.</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>98.3</td>
+<td>115.29 / 101.18</td>
+<td>964.75 / 964.75</td>
+<td>470.2</td>
+<td>A high-precision layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using RT-DETR-H.</td>
+</tr>
+</table>
+</details>
+</details>
+
+<details>
+<summary><b>Table Structure Recognition Module Models (Optional):</b></summary>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>SLANet</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANet_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_pretrained.pdparams">Training Model</a></td>
+<td>59.52</td>
+<td>23.96 / 21.75</td>
+<td>- / 43.12</td>
+<td>6.9</td>
+<td>SLANet is a table structure recognition model developed by Baidu PaddleX Team. The model significantly improves the accuracy and inference speed of table structure recognition by adopting a CPU-friendly lightweight backbone network PP-LCNet, a high-low-level feature fusion module CSP-PAN, and a feature decoding module SLA Head that aligns structural and positional information.</td>
+</tr>
+<tr>
+<td>SLANet_plus</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANet_plus_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_plus_pretrained.pdparams">Training Model</a></td>
+<td>63.69</td>
+<td>23.43 / 22.16</td>
+<td>- / 41.80</td>
+<td>6.9</td>
+<td>SLANet_plus is an enhanced version of SLANet, the table structure recognition model developed by Baidu PaddleX Team. Compared to SLANet, SLANet_plus significantly improves the recognition ability for wireless and complex tables and reduces the model's sensitivity to the accuracy of table positioning, enabling more accurate recognition even with offset table positioning.</td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary><b>Text Detection Module Models</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Detection Hmean (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv5_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_det_pretrained.pdparams">Training Model</a></td>
+<td>83.8</td>
+<td>89.55 / 70.19</td>
+<td>383.15 / 383.15</td>
+<td>84.3</td>
+<td>PP-OCRv5 server-side text detection model with higher accuracy, suitable for deployment on high-performance servers</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_det_pretrained.pdparams">Training Model</a></td>
+<td>79.0</td>
+<td>10.67 / 6.36</td>
+<td>57.77 / 28.15</td>
+<td>4.7</td>
+<td>PP-OCRv5 mobile-side text detection model with higher efficiency, suitable for deployment on edge devices</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_det_pretrained.pdparams">Training Model</a></td>
+<td>69.2</td>
+<td>127.82 / 98.87</td>
+<td>585.95 / 489.77</td>
+<td>109</td>
+<td>PP-OCRv4 server-side text detection model with higher accuracy, suitable for deployment on high-performance servers</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams">Training Model</a></td>
+<td>63.8</td>
+<td>9.87 / 4.17</td>
+<td>56.60 / 20.79</td>
+<td>4.7</td>
+<td>PP-OCRv4 mobile-side text detection model with higher efficiency, suitable for deployment on edge devices</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Text Recognition Module Models</b></summary>
+<table>
+<tr>
+<th>Model</th><th>Model Download Links</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>86.38</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec is a next-generation text recognition model. It aims to efficiently and accurately support the recognition of four major languages—Simplified Chinese, Traditional Chinese, English, and Japanese—as well as complex text scenarios such as handwriting, vertical text, pinyin, and rare characters using a single model. While maintaining recognition performance, it balances inference speed and model robustness, providing efficient and accurate technical support for document understanding in various scenarios.</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>81.29</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">Pretrained Model</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc is trained on a mixed dataset of more Chinese document data and PP-OCR training data, building upon PP-OCRv4_server_rec. It enhances the recognition capabilities for some Traditional Chinese characters, Japanese characters, and special symbols, supporting over 15,000 characters. In addition to improving document-related text recognition, it also enhances general text recognition capabilities.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>A lightweight recognition model of PP-OCRv4 with high inference efficiency, suitable for deployment on various hardware devices, including edge devices.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>The server-side model of PP-OCRv4, offering high inference accuracy and deployable on various servers.</td>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>An ultra-lightweight English recognition model trained based on the PP-OCRv4 recognition model, supporting English and numeric character recognition.</td>
+</tr>
+</table>
+* <b>PP-OCRv5 Multi-Scenario Models</b>
+
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Chinese Avg Accuracy (%)</th>
+<th>English Avg Accuracy (%)</th>
+<th>Traditional Chinese Avg Accuracy (%)</th>
+<th>Japanese Avg Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>86.38</td>
+<td>64.70</td>
+<td>93.29</td>
+<td>60.35</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td>PP-OCRv5_server_rec is a new-generation text recognition model. It efficiently and accurately supports four major languages: Simplified Chinese, Traditional Chinese, English, and Japanese, as well as handwriting, vertical text, pinyin, and rare characters, offering robust and efficient support for document understanding.</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>81.29</td>
+<td>66.00</td>
+<td>83.55</td>
+<td>54.65</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>136</td>
+<td>PP-OCRv5_mobile_rec is a new-generation text recognition model. It efficiently and accurately supports four major languages: Simplified Chinese, Traditional Chinese, English, and Japanese, as well as handwriting, vertical text, pinyin, and rare characters, offering robust and efficient support for document understanding.</td>
+</tr>
+</table>
+
+* <b>Chinese Recognition Models</b>
+
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Avg Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_doc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">Pretrained Model</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>Based on PP-OCRv4_server_rec, trained on additional Chinese documents and PP-OCR mixed data. It supports over 15,000 characters including Traditional Chinese, Japanese, and special symbols, enhancing both document-specific and general text recognition accuracy.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>Lightweight model of PP-OCRv4 with high inference efficiency, suitable for deployment on various edge devices.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>Server-side model of PP-OCRv4 with high recognition accuracy, suitable for deployment on various servers.</td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>72.96</td>
+<td>3.89 / 1.16</td>
+<td>8.72 / 3.56</td>
+<td>10.3</td>
+<td>Lightweight model of PP-OCRv3 with high inference efficiency, suitable for deployment on various edge devices.</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Recognition Avg Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>ch_SVTRv2_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_SVTRv2_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_SVTRv2_rec_pretrained.pdparams">Training Model</a></td>
+<td>68.81</td>
+<td>10.38 / 8.31</td>
+<td>66.52 / 30.83</td>
+<td>80.5</td>
+<td rowspan="1">
+SVTRv2 is a server-side text recognition model developed by the OpenOCR team at the Vision and Learning Lab (FVL) of Fudan University. It won the first prize in the OCR End-to-End Recognition Task of the PaddleOCR Algorithm Model Challenge, with a 6% improvement in end-to-end recognition accuracy compared to PP-OCRv4 on the A-list.
+</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Recognition Avg Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>ch_RepSVTR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_RepSVTR_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_RepSVTR_rec_pretrained.pdparams">Training Model</a></td>
+<td>65.07</td>
+<td>6.29 / 1.57</td>
+<td>20.64 / 5.40</td>
+<td>48.8</td>
+<td rowspan="1">
+The RepSVTR text recognition model is a mobile-oriented text recognition model based on SVTRv2. It won the first prize in the OCR End-to-End Recognition Task of the PaddleOCR Algorithm Model Challenge, with a 2.5% improvement in end-to-end recognition accuracy compared to PP-OCRv4 on the B-list, while maintaining similar inference speed.
+</td>
+</tr>
+</table>
+* <b>English Recognition Models</b>
+
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Avg Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>Ultra-lightweight English recognition model trained on PP-OCRv4, supporting English and number recognition.</td>
+</tr>
+<tr>
+<td>en_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>70.69</td>
+<td>3.56 / 0.78</td>
+<td>8.44 / 5.78</td>
+<td>17.3</td>
+<td>Ultra-lightweight English recognition model trained on PP-OCRv3, supporting English and number recognition.</td>
+</tr>
+</table>
+
+* <b>Multilingual Recognition Models</b>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal / High Performance]</th>
+<th>CPU Inference Time (ms)<br/>[Normal / High Performance]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>korean_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>60.21</td>
+<td>3.73 / 0.98</td>
+<td>8.76 / 2.91</td>
+<td>9.6</td>
+<td>An ultra-lightweight Korean text recognition model trained based on PP-OCRv3, supporting Korean and digits recognition</td>
+</tr>
+<tr>
+<td>japan_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+japan_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/japan_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>45.69</td>
+<td>3.86 / 1.01</td>
+<td>8.62 / 2.92</td>
+<td>9.8</td>
+<td>An ultra-lightweight Japanese text recognition model trained based on PP-OCRv3, supporting Japanese and digits recognition</td>
+</tr>
+<tr>
+<td>chinese_cht_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+chinese_cht_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/chinese_cht_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>82.06</td>
+<td>3.90 / 1.16</td>
+<td>9.24 / 3.18</td>
+<td>10.8</td>
+<td>An ultra-lightweight Traditional Chinese text recognition model trained based on PP-OCRv3, supporting Traditional Chinese and digits recognition</td>
+</tr>
+<tr>
+<td>te_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+te_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>95.88</td>
+<td>3.59 / 0.81</td>
+<td>8.28 / 6.21</td>
+<td>8.7</td>
+<td>An ultra-lightweight Telugu text recognition model trained based on PP-OCRv3, supporting Telugu and digits recognition</td>
+</tr>
+<tr>
+<td>ka_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ka_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ka_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>96.96</td>
+<td>3.49 / 0.89</td>
+<td>8.63 / 2.77</td>
+<td>17.4</td>
+<td>An ultra-lightweight Kannada text recognition model trained based on PP-OCRv3, supporting Kannada and digits recognition</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ta_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>76.83</td>
+<td>3.49 / 0.86</td>
+<td>8.35 / 3.41</td>
+<td>8.7</td>
+<td>An ultra-lightweight Tamil text recognition model trained based on PP-OCRv3, supporting Tamil and digits recognition</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>76.93</td>
+<td>3.53 / 0.78</td>
+<td>8.50 / 6.83</td>
+<td>8.7</td>
+<td>An ultra-lightweight Latin text recognition model trained based on PP-OCRv3, supporting Latin and digits recognition</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+arabic_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>73.55</td>
+<td>3.60 / 0.83</td>
+<td>8.44 / 4.69</td>
+<td>17.3</td>
+<td>An ultra-lightweight Arabic script recognition model trained based on PP-OCRv3, supporting Arabic script and digits recognition</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+cyrillic_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>94.28</td>
+<td>3.56 / 0.79</td>
+<td>8.22 / 2.76</td>
+<td>8.7</td>
+<td>An ultra-lightweight Cyrillic script recognition model trained based on PP-OCRv3, supporting Cyrillic script and digits recognition</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+devanagari_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>96.44</td>
+<td>3.60 / 0.78</td>
+<td>6.95 / 2.87</td>
+<td>8.7</td>
+<td>An ultra-lightweight Devanagari script recognition model trained based on PP-OCRv3, supporting Devanagari script and digits recognition</td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary><b>Text Line Orientation Classification Module (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Top-1 Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x0_25_textline_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x0_25_textline_ori_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x0_25_textline_ori_pretrained.pdparams">Training Model</a></td>
+<td>98.85</td>
+<td>2.16 / 0.41</td>
+<td>2.37 / 0.73</td>
+<td>0.96</td>
+<td>Text line classification model based on PP-LCNet_x0_25, with two classes: 0 degrees and 180 degrees</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Formula Recognition Module Models  (Optional):</b></summary>
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>En-BLEU(%)</th>
+<th>Zh-BLEU(%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+<tr>
+<td>UniMERNet</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UniMERNet_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UniMERNet_pretrained.pdparams">Training Model</a></td>
+<td>85.91</td>
+<td>43.50</td>
+<td>1311.84 / 1311.84</td>
+<td>- / 8288.07</td>
+<td>1530</td>
+<td>UniMERNet is a formula recognition model developed by Shanghai AI Lab. It uses Donut Swin as the encoder and MBartDecoder as the decoder. The model is trained on a dataset of one million samples, including simple formulas, complex formulas, scanned formulas, and handwritten formulas, significantly improving the recognition accuracy of real-world formulas.</td>
+</tr>
+<td>PP-FormulaNet-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-S_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-S_pretrained.pdparams">Training Model</a></td>
+<td>87.00</td>
+<td>45.71</td>
+<td>182.25 / 182.25</td>
+<td>- / 254.39</td>
+<td>224</td>
+<td rowspan="2">PP-FormulaNet is an advanced formula recognition model developed by the Baidu PaddlePaddle Vision Team. The PP-FormulaNet-S version uses PP-HGNetV2-B4 as its backbone network. Through parallel masking and model distillation techniques, it significantly improves inference speed while maintaining high recognition accuracy, making it suitable for applications requiring fast inference. The PP-FormulaNet-L version, on the other hand, uses Vary_VIT_B as its backbone network and is trained on a large-scale formula dataset, showing significant improvements in recognizing complex formulas compared to PP-FormulaNet-S.</td>
+</tr>
+<td>PP-FormulaNet-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-L_pretrained.pdparams">Training Model</a></td>
+<td>90.36</td>
+<td>45.78</td>
+<td>1482.03 / 1482.03</td>
+<td>- / 3131.54</td>
+<td>695</td>
+</tr>
+<td>PP-FormulaNet_plus-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-S_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-S_pretrained.pdparams">Training Model</a></td>
+<td>88.71</td>
+<td>53.32</td>
+<td>179.20 / 179.20</td>
+<td>- / 260.99</td>
+<td>248</td>
+<td rowspan="3">PP-FormulaNet_plus is an enhanced version of the formula recognition model developed by the Baidu PaddlePaddle Vision Team, building upon the original PP-FormulaNet. Compared to the original version, PP-FormulaNet_plus utilizes a more diverse formula dataset during training, including sources such as Chinese dissertations, professional books, textbooks, exam papers, and mathematics journals. This expansion significantly improves the model’s recognition capabilities. Among the models, PP-FormulaNet_plus-M and PP-FormulaNet_plus-L have added support for Chinese formulas and increased the maximum number of predicted tokens for formulas from 1,024 to 2,560, greatly enhancing the recognition performance for complex formulas. Meanwhile, the PP-FormulaNet_plus-S model focuses on improving the recognition of English formulas. With these improvements, the PP-FormulaNet_plus series models perform exceptionally well in handling complex and diverse formula recognition tasks. </td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-M_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-M_pretrained.pdparams">Training Model</a></td>
+<td>91.45</td>
+<td>89.76</td>
+<td>1040.27 / 1040.27</td>
+<td>- / 1615.80</td>
+<td>592</td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-L_pretrained.pdparams">Training Model</a></td>
+<td>92.22</td>
+<td>90.64</td>
+<td>1476.07 / 1476.07</td>
+<td>- / 3125.58</td>
+<td>698</td>
+</tr>
+<tr>
+<td>LaTeX_OCR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/LaTeX_OCR_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/LaTeX_OCR_rec_pretrained.pdparams">Training Model</a></td>
+<td>74.55</td>
+<td>39.96</td>
+<td>1088.89 / 1088.89</td>
+<td>- / -</td>
+<td>99</td>
+<td>LaTeX-OCR is a formula recognition algorithm based on an autoregressive large model. It uses Hybrid ViT as the backbone network and a transformer as the decoder, significantly improving the accuracy of formula recognition.</td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary><b>Seal Text Detection Module Models (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Detection Hmean (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv4_server_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_seal_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_seal_det_pretrained.pdparams">Training Model</a></td>
+<td>98.40</td>
+<td>124.64 / 91.57</td>
+<td>545.68 / 439.86</td>
+<td>109</td>
+<td>PP-OCRv4's server-side seal text detection model, featuring higher accuracy, suitable for deployment on better-equipped servers</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_seal_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_seal_det_pretrained.pdparams">Training Model</a></td>
+<td>96.36</td>
+<td>9.70 / 3.56</td>
+<td>50.38 / 19.64</td>
+<td>4.7</td>
+<td>PP-OCRv4's mobile seal text detection model, offering higher efficiency, suitable for deployment on edge devices</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary> <b>Test Environment Description:</b></summary>
+
+  <ul>
+      <li><b>Performance Test Environment</b>
+          <ul>
+                    <li><strong>Test Dataset: </strong>
+                        <ul>
+                          <li>Text Image Rectification Model: <a href="https://www3.cs.stonybrook.edu/~cvl/docunet.html">DocUNet</a></li>
+                          <li>Layout Region Detection Model: A self-built layout analysis dataset using PaddleOCR, containing 10,000 images of common document types such as Chinese and English papers, magazines, and research reports.</li>
+                          <li>Table Structure Recognition Model: A self-built English table recognition dataset using PaddleX.</li>
+                          <li>Text Detection Model: A self-built Chinese dataset using PaddleOCR, covering multiple scenarios such as street scenes, web images, documents, and handwriting, with 500 images for detection.</li>
+                          <li>Chinese Recognition Model: A self-built Chinese dataset using PaddleOCR, covering multiple scenarios such as street scenes, web images, documents, and handwriting, with 11,000 images for text recognition.</li>
+                          <li>ch_SVTRv2_rec: Evaluation set A for "OCR End-to-End Recognition Task" in the <a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR Algorithm Model Challenge</a></li>
+                          <li>ch_RepSVTR_rec: Evaluation set B for "OCR End-to-End Recognition Task" in the <a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR Algorithm Model Challenge</a></li>
+                          <li>English Recognition Model: A self-built English dataset using PaddleX.</li>
+                          <li>Multilingual Recognition Model: A self-built multilingual dataset using PaddleX.</li>
+                          <li>Text Line Orientation Classification Model: A self-built dataset using PaddleOCR, covering various scenarios such as ID cards and documents, containing 1000 images.</li>
+                          <li>Seal Text Detection Model: A self-built dataset using PaddleOCR, containing 500 images of circular seal textures.</li>
+                        </ul>
+                    </li>
+               <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>Inference Mode Description</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration </th>
+            <th>CPU Configuration </th>
+            <th>Acceleration Technology Combination</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Normal Mode</td>
+            <td>FP32 Precision / No TRT Acceleration</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Local Paddle inference engines (by default, the engine is selected according to the default model name; if both static and dynamic graph are available, <code>paddle_static</code> is preferred)</td>
+        </tr>
+        <tr>
+            <td>High-Performance Mode</td>
+            <td>Optimal combination of pre-selected precision types and acceleration strategies</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Pre-selected optimal backend (Paddle/OpenVINO/TRT, etc.)</td>
+        </tr>
+    </tbody>
+</table>
+
+</details>
+
+<br />
+<b>If you prioritize model accuracy, choose a model with higher accuracy. If you prioritize inference speed, select a model with faster inference. If you prioritize model storage size, choose a model with a smaller storage size.</b>
+
+## 2. Quick Start
+
+Before using the PP-ChatOCRv4-doc pipeline locally, ensure you have completed the installation of the PaddleOCR wheel package according to the [PaddleOCR Local Installation Tutorial](../installation.en.md). If you prefer to install dependencies selectively, please refer to the relevant instructions in the installation documentation. The corresponding dependency group for this pipeline is `ie`.
+
+Please note: If you encounter issues such as the program becoming unresponsive, unexpected program termination, running out of memory resources, or extremely slow inference during execution, please try adjusting the configuration according to the documentation, such as disabling unnecessary features or using lighter-weight models.
+
+Before performing model inference, you first need to prepare the API key for the large language model. PP-ChatOCRv4 supports large model services on the [Baidu Cloud Qianfan Platform](https://console.bce.baidu.com/qianfan/ais/console/onlineService) or the locally deployed standard OpenAI interface. If using the Baidu Cloud Qianfan Platform, refer to [Authentication and Authorization](https://cloud.baidu.com/doc/qianfan-api/s/ym9chdsy5) to obtain the API key. If using a locally deployed large model service, refer to the [PaddleNLP Large Model Deployment Documentation](https://github.com/PaddlePaddle/PaddleNLP/tree/develop/llm) for deployment of the dialogue interface and vectorization interface for large models, and fill in the corresponding `base_url` and `api_key`. If you need to use a multimodal large model for data fusion, refer to the OpenAI service deployment in the [PaddleMIX Model Documentation](https://github.com/PaddlePaddle/PaddleMIX/tree/develop/paddlemix/examples/ppdocbee2) for multimodal large model deployment, and fill in the corresponding `base_url` and `api_key`.
+
+**Note**: If local deployment of a multimodal large model is restricted due to the local environment, you can comment out the lines containing the `mllm` variable in the code and only use the large language model for information extraction.
+
+### 2.1 Command Line Experience
+
+After updating the configuration file, you can complete quick inference using just a few lines of Python code. You can use the [test file](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png) for testing:
+
+
+```bash
+paddleocr pp_chatocrv4_doc -i vehicle_certificate-1.png -k 驾驶室准乘人数 --qianfan_api_key your_api_key
+
+# 通过 --invoke_mllm 和 --pp_docbee_base_url 使用多模态大模型
+paddleocr pp_chatocrv4_doc -i vehicle_certificate-1.png -k 驾驶室准乘人数 --qianfan_api_key your_api_key --invoke_mllm True --pp_docbee_base_url http://127.0.0.1:8080/
+```
+
+The examples above use local Paddle inference engines by default. By default, each module selects the appropriate local Paddle inference engine according to the default model name: models that support only dynamic graph use `paddle_dynamic`, while models that support both static and dynamic graph prefer `paddle_static`. To run them, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr pp_chatocrv4_doc -i vehicle_certificate-1.png -k 驾驶室准乘人数 --qianfan_api_key your_api_key \
+    --engine transformers
+```
+
+<details><summary><b>The command line supports more parameter configurations. Click to expand for a detailed explanation of the command line parameters.</b></summary>
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted, required.<br/> 
+<b>Description:</b>
+Such as the local path of an image file or PDF file: <code>/root/data/img.jpg</code>; <b>URL link</b>, such as the network URL of an image file or PDF file: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png">Example</a>; <b>Local directory</b>, which should contain images to be predicted, such as the local path: <code>/root/data/</code> (currently does not support prediction of PDF files in directories, PDF files need to be specified to the specific file path).
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>keys</code></td>
+<td><b>Meaning:</b>Keys for information extraction.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td>
+<b>Meaning:</b>Specify the path to save the inference results file.<br/>
+<b>Description:</b>
+If not set, the inference results will not be saved locally.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>invoke_mllm</code></td>
+<td><b>Meaning:</b>Whether to load and use a multimodal large model. <br/>
+<b>Description:</b>
+If not set, the default is <code>False</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td>
+<b>Meaning:</b>The name of the layout detection model.<br/> 
+<b>Description:</b>
+If not set, the default model in pipeline will be used. </td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the layout detection model. <br/>
+<b>Description:</b>
+If not set, the official model will be downloaded.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td> 
+<b>Meaning:</b>The name of the document orientation classification model. <br/>
+<b>Description:</b>
+If not set, the default model in pipeline will be used. </td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the document orientation classification model. <br/>
+<b>Description:</b>
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>Meaning:</b>The name of the text image unwarping model. <br/>
+<b>Description:</b>
+If not set, the default model in pipeline will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td> The directory path of the  text image unwarping model. If not set, the official model will be downloaded.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the text detection model. <br/>
+<b>Description:</b>
+If not set, the pipeline's default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text detection model. <br/>
+<b>Description:</b>
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the text recognition model. <br/>
+<b>Description:</b>
+If not set, the pipeline's default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text recognition model. <br/>
+<b>Description:</b>
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the text recognition model. <br/>
+<b>Description:</b>
+If not set, the default batch size will be <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_structure_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the table structure recognition model. <br/>
+<b>Description:</b>
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_structure_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the table structure recognition model. <br/>
+<b>Description:</b>
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_name</code></td>
+<td><b>Meaning:</b>The name of the seal text detection model. <br/>
+<b>Description:</b>
+If not set, the pipeline's default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the seal text detection model. <br/>
+<b>Description:</b>
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_name</code></td>
+<td><b>Meaning:</b>The name of the seal text recognition model. <br/>
+<b>Description:</b>
+If not set, the default model of the pipeline will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the seal text recognition model. <br/>
+<b>Description:</b>
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_batch_size</code></td>
+<td><b>Meaning:</b>The batch size for the seal text recognition model. <br/>
+<b>Description:</b>
+If not set, the batch size will default to <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to load and use the document orientation classification module. <br/>
+<b>Description:</b>
+If not set, the parameter value initialized by the pipeline will be used, which defaults to <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to load and use the text image unwarping module. <br/>
+<b>Description:</b>
+If not set, the parameter value initialized by the pipeline will be used, which defaults to <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>Meaning:</b>Whether to load and use the text line orientation classification module. <br/>
+<b>Description:</b>
+If not set, the parameter value initialized by the pipeline will be used, which defaults to <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>Meaning:</b>Whether to load and use the seal text recognition sub-pipeline. <br/>
+<b>Description:</b>
+If not set, the parameter value initialized by the pipeline will be used, defaulting to <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>Meaning:</b>Whether to load and use the table recognition sub-pipeline.<br/>
+<b>Description:</b> 
+If not set, the parameter's value initialized during pipeline setup will be used, defaulting to <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>Score threshold for the layout model. <br/>
+<b>Description:</b>
+Any value between <code>0-1</code>. If not set, the default value is used, which is <code>0.5</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td>
+<b>Meaning:</b>Whether to use Non-Maximum Suppression (NMS) as post-processing for layout detection.<br/>
+<b>Description:</b> 
+If not set, the parameter will be set to the value initialized in the pipeline, which defaults to <code>True</code> by default.
+</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Unclip ratio for detected boxes in layout detection model. <br/>
+<b>Description:</b>
+Any float > <code>0</code>. If not set, the default is <code>1.0</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>The merging mode for the detection boxes output by the model in layout region detection. <br/>
+<b>Description:</b>
+<ul>
+<li><b>large</b>: When set to "large", only the largest outer bounding box will be retained for overlapping bounding boxes, and the inner overlapping boxes will be removed;</li>
+<li><b>small</b>: When set to "small", only the smallest inner bounding boxes will be retained for overlapping bounding boxes, and the outer overlapping boxes will be removed;</li>
+<li><b>union</b>: No filtering of bounding boxes will be performed, and both inner and outer boxes will be retained;</li>
+</ul>If not set, the default is <code>large</code>.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limitation for text detection. <br/>
+<b>Description:</b>
+Any integer greater than <code>0</code>. If not set, the pipeline's initialized value for this parameter (initialized to <code>960</code>) will be used.
+</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>Meaning:</b>Type of side length limit for text detection. <br/>
+<b>Description:</b>
+Supports <code>min</code> and <code>max</code>. <code>min</code> means ensuring the shortest side of the image is not smaller than <code>det_limit_side_len</code>, and <code>max</code> means ensuring the longest side of the image is not larger than <code>limit_side_len</code>. If not set, the pipeline's initialized value for this parameter (initialized to <code>max</code>) will be used.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>Meaning:</b>Pixel threshold for text detection. In the output probability map, pixels with scores higher than this threshold will be considered text pixels. <br/>
+<b>Description:</b>
+Any floating-point number greater than <code>0</code>
+. If not set, the pipeline's initialized value for this parameter (<code>0.3</code>) will be used.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>Meaning:</b>Text detection box threshold. If the average score of all pixels within the detected result boundary is higher than this threshold, the result will be considered a text region. <br/>
+<b>Description:</b>
+ Any floating-point number greater than <code>0</code>. If not set, the pipeline's initialized value for this parameter (<code>0.6</code>) will be used.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Text detection expansion coefficient. This method is used to expand the text region—the larger the value, the larger the expanded area. <br/>
+<b>Description:</b>
+Any floating-point number greater than <code>0</code>
+. If not set, the pipeline's initialized value for this parameter (<code>2.0</code>) will be used.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Text recognition threshold. Text results with scores higher than this threshold will be retained. <br/>
+<b>Description:</b>
+ Any floating-point number greater than <code>0</code>
+. If not set, the pipeline's initialized value for this parameter (<code>0.0</code>, i.e., no threshold) will be used.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limit for seal text detection.<br/>
+<b>Description:</b>
+Any integer > <code>0</code>. If not set, the default is <code>736</code>.
+</td>
+<td><code>int</code></td>don’t 
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>Meaning:</b>Limit type for image side in seal text detection.<br/>
+<b>Description:</b>
+Supports <code>min</code> and <code>max</code>; <code>min</code> ensures shortest side ≥ <code>det_limit_side_len</code>, <code>max</code> ensures longest side ≤ <code>limit_side_len</code>. If not set, the default is <code>min</code>.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>Meaning:</b>Pixel threshold. Pixels with scores above this value in the probability map are considered text. <br/>
+<b>Description:</b>
+Any float > <code>0</code></li>
+</ul>If not set, the default is <code>0.2</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>Meaning:</b>Box threshold. Boxes with average pixel scores above this value are considered text regions. <br/>
+<b>Description:</b>
+Any float > <code>0</code>. If not set, the default is <code>0.6</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion ratio for seal text detection. Higher value means larger expansion area. <br/>
+<b>Description:</b>
+Any float > <code>0</code>. If not set, the default is <code>0.5</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Recognition score threshold. Text results above this value will be kept. <br/>
+<b>Description:</b>
+Any float > <code>0</code></li>
+</ul>If not set, the default is <code>0.0</code> (no threshold).
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<td><code>qianfan_api_key</code></td>
+<td><b>Meaning:</b>API key for the Qianfan Platform.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<td><code>pp_docbee_base_url</code></td>
+<td><b>Meaning:</b>URL for the multimodal large language model service.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>The device used for inference.<br/>
+<b>Description:</b> 
+You can specify a particular card number:
+<ul>
+<li><b>CPU</b>: e.g., <code>cpu</code> indicates using CPU for inference;</li>
+<li><b>GPU</b>: e.g., <code>gpu:0</code> indicates using the 1st GPU for inference;</li>
+<li><b>NPU</b>: e.g., <code>npu:0</code> indicates using the 1st NPU for inference;</li>
+<li><b>XPU</b>: e.g., <code>xpu:0</code> indicates using the 1st XPU for inference;</li>
+<li><b>MLU</b>: e.g., <code>mlu:0</code> indicates using the 1st MLU for inference;</li>
+<li><b>DCU</b>: e.g., <code>dcu:0</code> indicates using the 1st DCU for inference;</li>
+<li><b>MetaX GPU</b>: e.g., <code>metax_gpu:0</code> indicates using the 1st MetaX GPU for inference;</li>
+<li><b>Iluvatar GPU</b>: e.g., <code>iluvatar_gpu:0</code> indicates using the 1st Iluvatar GPU for inference;</li>
+</ul>If not set, the pipeline initialized value for this parameter will be used. During initialization, the local GPU device 0 will be preferred; if unavailable, the CPU device will be used.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to enable the TensorRT subgraph engine of Paddle Inference.<br/>
+<b>Description:</b>
+If the model does not support TensorRT acceleration, acceleration will not be used even if this flag is set.<br/>
+For CUDA 11.8 versions of PaddlePaddle, the compatible TensorRT version is 8.x (x>=6). TensorRT 8.6.1.6 is recommended.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b> Computation precision, such as <code>fp32</code> or <code>fp16</code>.</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b> Whether to enable MKL-DNN accelerated inference.<br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support MKL-DNN acceleration, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b> MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads used for inference on CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b> Path to the PaddleX pipeline configuration file.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+
+</tbody>
+</table>
+
+</details>
+
+This method will print the results to the terminal. The content printed to the terminal is explained as follows:
+
+
+```
+驾驶室准乘人数 2
+```
+
+
+### 2.2 Python Script Experience
+
+The command-line method is for a quick experience and to view results. Generally, in projects, integration via code is often required. You can download the [Test File](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png) and use the following example code for inference:
+
+```python
+from paddleocr import PPChatOCRv4Doc
+
+chat_bot_config = {
+    "module_name": "chat_bot",
+    "model_name": "ernie-3.5-8k",
+    "base_url": "https://qianfan.baidubce.com/v2",
+    "api_type": "openai",
+    "api_key": "api_key",  # your api_key
+}
+
+retriever_config = {
+    "module_name": "retriever",
+    "model_name": "embedding-v1",
+    "base_url": "https://qianfan.baidubce.com/v2",
+    "api_type": "qianfan",
+    "api_key": "api_key",  # your api_key
+}
+
+mllm_chat_bot_config = {
+    "module_name": "chat_bot",
+    "model_name": "PP-DocBee2",
+    "base_url": "http://127.0.0.1:8080/",  # your local mllm service url
+    "api_type": "openai",
+    "api_key": "api_key",  # your api_key
+}
+
+pipeline = PPChatOCRv4Doc()
+
+visual_predict_res = pipeline.visual_predict(
+    input="vehicle_certificate-1.png",
+    use_doc_orientation_classify=False,
+    use_doc_unwarping=False,
+    use_common_ocr=True,
+    use_seal_recognition=True,
+    use_table_recognition=True,
+)
+
+visual_info_list = []
+for res in visual_predict_res:
+    visual_info_list.append(res["visual_info"])
+    layout_parsing_result = res["layout_parsing_result"]
+
+vector_info = pipeline.build_vector(
+    visual_info_list, flag_save_bytes_vector=True, retriever_config=retriever_config
+)
+mllm_predict_res = pipeline.mllm_pred(
+    input="vehicle_certificate-1.png",
+    key_list=["驾驶室准乘人数"],
+    mllm_chat_bot_config=mllm_chat_bot_config,
+)
+mllm_predict_info = mllm_predict_res["mllm_res"]
+chat_result = pipeline.chat(
+    key_list=["驾驶室准乘人数"],
+    visual_info=visual_info_list,
+    vector_info=vector_info,
+    mllm_predict_info=mllm_predict_info,
+    chat_bot_config=chat_bot_config,
+    retriever_config=retriever_config,
+)
+print(chat_result)
+
+```
+
+The example above uses local Paddle inference engines by default. By default, each module selects the appropriate local Paddle inference engine according to the default model name: models that support only dynamic graph use `paddle_dynamic`, while models that support both static and dynamic graph prefer `paddle_static`. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following code:
+
+```python
+from paddleocr import PPChatOCRv4Doc
+
+chat_bot_config = {
+    "module_name": "chat_bot",
+    "model_name": "ernie-3.5-8k",
+    "base_url": "https://qianfan.baidubce.com/v2",
+    "api_type": "openai",
+    "api_key": "api_key",  # your api_key
+}
+
+retriever_config = {
+    "module_name": "retriever",
+    "model_name": "embedding-v1",
+    "base_url": "https://qianfan.baidubce.com/v2",
+    "api_type": "qianfan",
+    "api_key": "api_key",  # your api_key
+}
+
+mllm_chat_bot_config = {
+    "module_name": "chat_bot",
+    "model_name": "PP-DocBee2",
+    "base_url": "http://127.0.0.1:8080/",  # your local mllm service url
+    "api_type": "openai",
+    "api_key": "api_key",  # your api_key
+}
+
+pipeline = PPChatOCRv4Doc(
+    engine="transformers",
+)
+
+visual_predict_res = pipeline.visual_predict(
+    input="vehicle_certificate-1.png",
+    use_doc_orientation_classify=False,
+    use_doc_unwarping=False,
+    use_common_ocr=True,
+    use_seal_recognition=True,
+    use_table_recognition=True,
+)
+
+visual_info_list = []
+for res in visual_predict_res:
+    visual_info_list.append(res["visual_info"])
+    layout_parsing_result = res["layout_parsing_result"]
+
+vector_info = pipeline.build_vector(
+    visual_info_list, flag_save_bytes_vector=True, retriever_config=retriever_config
+)
+mllm_predict_res = pipeline.mllm_pred(
+    input="vehicle_certificate-1.png",
+    key_list=["驾驶室准乘人数"],
+    mllm_chat_bot_config=mllm_chat_bot_config,
+)
+mllm_predict_info = mllm_predict_res["mllm_res"]
+chat_result = pipeline.chat(
+    key_list=["驾驶室准乘人数"],
+    visual_info=visual_info_list,
+    vector_info=vector_info,
+    mllm_predict_info=mllm_predict_info,
+    chat_bot_config=chat_bot_config,
+    retriever_config=retriever_config,
+)
+print(chat_result)
+
+```
+
+After running, the output is as follows:
+
+```
+{'chat_res': {'驾驶室准乘人数': '2'}}
+```
+
+The prediction process, API description, and output description for PP-ChatOCRv4 are as follows:
+
+<details><summary>(1) Call <code>PPChatOCRv4Doc</code> to instantiate the PP-ChatOCRv4 pipeline object.</summary>
+
+The relevant parameter descriptions are as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Parameter Description</th>
+<th>Parameter Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>Meaning:</b>The name of the model used for layout region detection.<br/>
+<b>Description:</b> 
+If set to<code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the layout region detection model.<br/> 
+<b>Description:</b> 
+If set to<code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>The name of the document orientation classification model.<br/> 
+<b>Description:</b> 
+If set to<code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the document orientation classification model.<br/> 
+<b>Description:</b> 
+If set to<code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>Meaning:</b>The name of the document unwarping model.<br/> 
+<b>Description:</b> 
+If set to<code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the document unwarping model.<br/> 
+<b>Description:</b> 
+If set to<code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>Meaning:</b>The name of the text detection model.<br/> 
+<b>Description:</b> 
+If set to<code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the text detection model.<br/> 
+<b>Description:</b> 
+If set to<code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>Meaning:</b>The name of the text recognition model.<br/> 
+<b>Description:</b> 
+If set to<code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the text recognition model.<br/> 
+<b>Description:</b> 
+If set to<code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>Meaning:</b>The batch size for the text recognition model.<br/> 
+<b>Description:</b> 
+If set to<code>None</code>, the batch size will default to <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_structure_recognition_model_name</code></td>
+<td><b>Meaning:</b>The name of the table structure recognition model.<br/> 
+<b>Description:</b> 
+If set to<code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_structure_recognition_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the table structure recognition model.<br/>
+<b>Description:</b> 
+If set to<code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_name</code></td>
+<td><b>Meaning:</b>The name of the seal text detection model. <br/>
+<b>Description:</b> 
+If set to<code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the seal text detection model. <br/>
+<b>Description:</b> 
+If set to<code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_name</code></td>
+<td><b>Meaning:</b>The name of the seal text recognition model. <br/>
+<b>Description:</b> 
+If set to<code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the seal text recognition model. <br/>
+<b>Description:</b> 
+If set to<code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_batch_size</code></td>
+<td><b>Meaning:</b>The batch size for the seal text recognition model. <br/>
+<b>Description:</b> 
+If set to<code>None</code>, the batch size will default to <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td>Whether to load and use the document orientation classification module. If set to<code>None</code>, the value initialized by the pipeline for this parameter will be used (defaults to <code>True</code>).</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to load and use the document unwarping module. <br/>
+<b>Description:</b> 
+If set to<code>None</code>, the value initialized by the pipeline for this parameter will be used (defaults to <code>True</code>).</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>Meaning:</b>Whether to load and use the text line orientation classification function. <br/>
+<b>Description:</b> 
+If set to<code>None</code>, the value initialized by the pipeline for this parameter will be used (defaults to <code>True</code>).</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>Meaning:</b>Whether to load and use the seal text recognition sub-pipeline. <br/>
+<b>Description:</b> 
+If set to<code>None</code>, the value initialized by the pipeline for this parameter will be used (defaults to <code>True</code>).</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>Meaning:</b>Whether to load and use the table recognition sub-pipeline. <br/>
+<b>Description:</b> 
+If set to<code>None</code>, the value initialized by the pipeline for this parameter will be used (defaults to <code>True</code>).</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>Layout model score threshold. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float between <code>0-1</code>;</li>
+<li><b>dict</b>: <code>{0:0.1}</code> where the key is the class ID and the value is the threshold for that class;</li>
+<li><b>None</b>: If set to <code>None</code>, uses the pipeline default of <code>0.5</code>.</li>
+</ul>
+</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Whether to use Non-Maximum Suppression (NMS) as post-processing for layout detection. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the parameter will be set to the value initialized in the pipeline, which is set to <code>True</code> by default.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion factor for the detection boxes of the layout region detection model. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>Tuple[float,float]</b>: Expansion ratios in horizontal and vertical directions;</li>
+<li><b>dict</b>: A dictionary with <b>int</b> keys representing <code>cls_id</code>, and <b>tuple</b> values, e.g., <code>{0: (1.1, 2.0)}</code> means width is expanded 1.1× and height 2.0× for class 0 boxes;</li>
+<li><b>None</b>: If set to <code>None</code>, uses the pipeline default of <code>1.0</code>.</li>
+</ul>
+</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>Method for filtering overlapping boxes in layout region detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>str</b>: <code>large</code>,<code>small</code>, <code>union</code>, representing whether to keep the large box, small box, or both when filtering overlapping boxes;</li>
+<li><b>dict</b>, where the key is of <b>int</b> type, representing <code>cls_id</code>, and the value is of <b>str</b> type, e.g.,<code>{0: "large", 2: "small"}</code>, meaning use "large" mode for class 0 detection boxes and "small" mode for class 2 detection boxes;</li>
+<li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline for this parameter will be used (defaults to <code>large</code>).</li>
+</ul>
+</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limitation for text detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>int</b>: Any integer greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline for this parameter will be used (defaults to <code>960</code>).</li>
+</ul>
+</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>Meaning:</b>Type of side length limit for text detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>str</b>: Supports <code>min</code> and <code>max</code>. <code>min</code> ensures the shortest side of the image is not less than <code>det_limit_side_len</code>. <code>max</code> ensures the longest side of the image is not greater than <code>limit_side_len</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline for this parameter will be used (defaults to <code>max</code>).</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>Meaning:</b>Detection pixel threshold. In the output probability map, pixels with scores greater than this threshold are considered text pixels. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline for this parameter (defaults to <code>0.3</code>) will be used.</li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>Meaning:</b>Detection box threshold. If the average score of all pixels within a detection result's bounding box is greater than this threshold, the result is considered a text region. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline for this parameter (defaults to <code>0.6</code>) will be used.</li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Text detection expansion factor. This method is used to expand text regions; the larger the value, the larger the expanded area.<br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline for this parameter (defaults to <code>2.0</code>) will be used.</li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Text recognition threshold. Text results with scores greater than this threshold will be kept. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline for this parameter (defaults to <code>0.0</code>, i.e., no threshold) will be used.</li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limit for seal text detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>int</b>: Any integer greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline for this parameter will be used (defaults to <code>736</code>).</li>
+</ul>
+</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>Meaning:</b>Type of image side length limit for seal text detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>str</b>: Supports <code>min</code> and <code>max</code>. <code>min</code> ensures the shortest side of the image is not less than <code>det_limit_side_len</code>. <code>max</code> ensures the longest side of the image is not greater than <code>limit_side_len</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline for this parameter will be used (defaults to <code>min</code>).</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>Meaning:</b>Detection pixel threshold. In the output probability map, pixels with scores greater than this threshold are considered text pixels. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;
+    <li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline for this parameter (defaults to <code>0.2</code>) will be used.</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>Meaning:</b>Detection box threshold. If the average score of all pixels within a detection result's bounding box is greater than this threshold, the result is considered a text region. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;
+    <li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline for this parameter (defaults to <code>0.6</code>) will be used.</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Seal text detection expansion factor. This method is used to expand text regions; the larger the value, the larger the expanded area. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;
+    <li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline for this parameter (defaults to <code>0.5</code>) will be used.</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Seal text recognition threshold. Text results with scores greater than this threshold will be kept. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;
+    <li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline for this parameter (defaults to <code>0.0</code>, i.e., no threshold) will be used.</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>retriever_config</code></td>
+<td><b>Meaning:</b>Configuration parameters for the vector retrieval large model. <br/>
+<b>Description:</b> 
+The configuration content is the following dictionary:
+<pre><code>{
+"module_name": "retriever",
+"model_name": "embedding-v1",
+"base_url": "https://qianfan.baidubce.com/v2",
+"api_type": "qianfan",
+"api_key": "api_key"  # Please set this to your actual API key
+}</code></pre>
+</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>mllm_chat_bot_config</code></td>
+<td><b>Meaning:</b>Configuration parameters for the multimodal large model. <br/>
+<b>Description:</b> 
+The configuration content is the following dictionary:
+<pre><code>{
+"module_name": "chat_bot",
+"model_name": "PP-DocBee",
+"base_url": "http://127.0.0.1:8080/", # Please set this to the actual URL of your multimodal large model service
+"api_type": "openai",
+"api_key": "api_key"  # Please set this to your actual API key
+}</code></pre>
+</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>chat_bot_config</code></td>
+<td><b>Meaning:</b>Configuration information for the large language model. <br/>
+<b>Description:</b> 
+The configuration content is the following dictionary:
+<pre><code>{
+"module_name": "chat_bot",
+"model_name": "ernie-3.5-8k",
+"base_url": "https://qianfan.baidubce.com/v2",
+"api_type": "openai",
+"api_key": "api_key"  # Please set this to your actual API key
+}</code></pre>
+</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>Device used for inference. <br/>
+<b>Description:</b> 
+Supports specifying a specific card number:
+<ul>
+<li><b>CPU</b>: e.g., <code>cpu</code> indicates using CPU for inference;</li>
+<li><b>GPU</b>: e.g., <code>gpu:0</code> indicates using the 1st GPU for inference;</li>
+<li><b>NPU</b>: e.g., <code>npu:0</code> indicates using the 1st NPU for inference;</li>
+<li><b>XPU</b>: e.g., <code>xpu:0</code> indicates using the 1st XPU for inference;</li>
+<li><b>MLU</b>: e.g., <code>mlu:0</code> indicates using the 1st MLU for inference;</li>
+<li><b>DCU</b>: e.g., <code>dcu:0</code> indicates using the 1st DCU for inference;</li>
+<li><b>MetaX GPU</b>: e.g., <code>metax_gpu:0</code> indicates using the 1st MetaX GPU for inference;</li>
+<li><b>Iluvatar GPU</b>: e.g., <code>iluvatar_gpu:0</code> indicates using the 1st Iluvatar GPU for inference;</li>
+<li><b>None</b>: If set to <code>None</code>, the pipeline initialized value for this parameter will be used. During initialization, the local GPU device 0 will be preferred; if unavailable, the CPU device will be used.</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to enable the TensorRT subgraph engine of Paddle Inference.<br/>
+<b>Description:</b>
+If the model does not support TensorRT acceleration, acceleration will not be used even if this flag is set.<br/>
+For CUDA 11.8 versions of PaddlePaddle, the compatible TensorRT version is 8.x (x>=6). TensorRT 8.6.1.6 is recommended.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b> Computation precision, such as <code>"fp32"</code> or <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b> Whether to enable MKL-DNN accelerated inference.<br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support MKL-DNN acceleration, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b> MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads used for inference on CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b> Path to the PaddleX pipeline configuration file.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+</tbody>
+</table>
+</details>
+
+<details><summary>(2) Call the <code>visual_predict()</code> method of the PP-ChatOCRv4 pipeline object to obtain visual prediction results. This method returns a list of results. Additionally, the pipeline also provides the <code>visual_predict_iter()</code> method. Both are identical in terms of parameter acceptance and result return, with the difference being that <code>visual_predict_iter()</code> returns a <code>generator</code>, allowing for step-by-step processing and retrieval of prediction results, suitable for handling large datasets or scenarios where memory saving is desired. You can choose either of these two methods based on your actual needs. The following are the parameters and their descriptions for the <code>visual_predict()</code> method:</summary>
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Parameter Description</th>
+<th>Parameter Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted, supports multiple input types, required. <br/>
+<b>Description:</b> 
+<ul>
+  <li><b>Python Var</b>: e.g., image data represented by <code>numpy.ndarray</code>;</li>
+  <li><b>str</b>: e.g., local path of an image file or PDF file: <code>/root/data/img.jpg</code>; <b>URL link</b>, e.g., network URL of an image file or PDF file: <a href = "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png">Example</a>; <b>Local directory</b>, which must contain images to be predicted, e.g., local path: <code>/root/data/</code> (Currently, prediction from directories containing PDF files is not supported; PDF files need to be specified by their full path);</li>
+  <li><b>list</b>: List elements must be of the above types, e.g.,<code>[numpy.ndarray, numpy.ndarray]</code>,<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>,<code>["/root/data1", "/root/data2"]</code>.</li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to use the document orientation classification module during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to use the document image unwarping module during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>Meaning:</b>Whether to use the text line orientation classification module during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>Meaning:</b>Whether to use the seal text recognition sub-pipeline during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>Meaning:</b>Whether to use the table recognition sub-pipeline during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+</table>
+</details>
+
+<details><summary>(3) Process the visual prediction results.</summary>
+
+The prediction result for each sample is of <code>dict</code> type, containing two fields: <code>visual_info</code> and <code>layout_parsing_result</code>. Visual information (including <code>normal_text_dict</code>, <code>table_text_list</code>, <code>table_html_list</code>, etc.) is obtained through <code>visual_info</code>, and the information for each sample is placed in the <code>visual_info_list</code> list. The content of this list will later be fed into the large language model.
+
+Of course, you can also obtain the layout parsing results through <code>layout_parsing_result</code>. This result contains content such as tables, text, and images found in the file or image, and supports operations like printing, saving as an image, and saving as a <code>json</code> file:
+
+```python
+......
+for res in visual_predict_res:
+    visual_info_list.append(res["visual_info"])
+    layout_parsing_result = res["layout_parsing_result"]
+    layout_parsing_result.print()
+    layout_parsing_result.save_to_img("./output")
+    layout_parsing_result.save_to_json("./output")
+    layout_parsing_result.save_to_xlsx("./output")
+    layout_parsing_result.save_to_html("./output")
+......
+```
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Method Description</th>
+<th>Parameter</th>
+<th>Parameter Type</th>
+<th>Parameter Description</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">Prints the result to the terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output content using <code>JSON</code> indentation.</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specifies the indentation level to beautify the output <code>JSON</code> data for better readability, effective only when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Controls whether to escape non-<code>ASCII</code> characters to <code>Unicode</code>. Set to <code>True</code> to escape all non-<code>ASCII</code> characters; <code>False</code> to preserve original characters, effective only when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">Saves the result as a JSON format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Save file path. When it's a directory, the saved file name will be consistent with the input file name.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specifies the indentation level to beautify the output <code>JSON</code> data for better readability, effective only when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Controls whether to escape non-<code>ASCII</code> characters to <code>Unicode</code>. Set to <code>True</code> to escape all non-<code>ASCII</code> characters; <code>False</code> to preserve original characters, effective only when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Saves the visualization images of various intermediate modules as PNG format images.</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Save file path, supports directory or file path.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>save_to_html()</code></td>
+<td>Saves the tables in the file as HTML format files.</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Save file path, supports directory or file path.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>save_to_xlsx()</code></td>
+<td>Saves the tables in the file as XLSX format files.</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Save file path, supports directory or file path.</td>
+<td>None</td>
+</tr>
+</table>
+
+<ul>
+   <li>Calling the <code>print()</code> method will print the results to the terminal. The content printed to the terminal is explained as follows:</li>
+      <ol start="1" type = "1"> 
+        <li><code>input_path</code>: <code>(str)</code>  Input path of the image to be predicted.</li>
+        <li><code>page_index</code>: <code>(Union[int, None])</code> If the input is a PDF file, it indicates the current page number of the PDF; otherwise, it is <code>None</code>.</li>
+        <li><code>model_settings</code>: <code>(Dict[str, bool])</code>  Model settings required for the pipeline.</li>
+            <ol start = "1" type = "a" > 
+                <li><code>use_doc_preprocessor</code>: <code>(bool)</code>  Whether to enable the document preprocessing sub-pipeline.</li>
+                <li><code>use_seal_recognition</code>: <code>(bool)</code> Whether to enable the seal text recognition sub-pipeline.</li>
+                <li><code>use_table_recognition</code>: <code>(bool)</code> Whether to enable the table recognition sub-pipeline.</li>
+                <li><code>use_formula_recognition</code>: <code>(bool)</code> Whether to enable the formula recognition sub-pipeline.</li>
+            </ol>
+        </li>
+        <li><code>parsing_res_list</code>: <code>(List[Dict])</code> List of parsing results, where each element is a dictionary. The list order is the reading order after parsing.</li>
+            <ol start = "1" type = "a" > 
+                <li><code>block_bbox</code>: <code>(np.ndarray)</code> Bounding box of the layout region.</li>
+                <li><code>block_label</code>: <code>(str)</code> Label of the layout region, e.g., <code>text</code>, <code>table</code> etc.</li>
+                <li><code>block_content</code>: <code>(str)</code> Content within the layout region.</li>
+            </ol>
+            <li><code>overall_ocr_res</code>: <code>(Dict[str, Union[List[str], List[float], numpy.ndarray]])</code> Global OCR results dict.</li>
+                <ol start = "1" type = "a" > 
+                    <li><code>input_path</code>: <code>(Union[str, None])</code> Image path accepted by the image OCR sub-pipeline. When the input is <code>numpy.ndarray</code>, it is saved as <code> None</code> </li>
+                    <li><code>model_settings</code>: <code>(Dict)</code> OCR sub-pipeline model configuration parameters.</li>
+                    <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> List of text detection polygons. Each detection box is represented by a numpy array of shape (4, 2), with data type int16.</li>
+                    <li><code>dt_scores</code>: <code>(List[float])</code> List of text detection box confidence scores.</li>
+                    <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> Text detection module configuration parameters.</li>
+                        <ul>
+                            <li><code>limit_side_len</code>: <code>(int)</code> Side length limit value for image preprocessing.</li>
+                            <li><code>limit_type</code>: <code>(str)</code> Side length limit handling method, e.g., "min", "max", "mean".</li>
+                            <li><code>thresh</code>: <code>(float)</code> Text pixel classification confidence threshold.</li>
+                            <li><code>box_thresh</code>: <code>(float)</code> Text detection box confidence threshold.</li>
+                            <li><code>unclip_ratio</code>: <code>(float)</code> Text detection box expansion coefficient.</li>
+                            <li><code>text_type</code>: <code>(str)</code> Text detection type, currently fixed as "general".</li>
+                        </ul>
+                    <li><code>text_type</code>: <code>(str)</code> Text detection type, currently fixed as "general".</li>
+                    <li><code>textline_orientation_angles</code>: <code>(List[int])</code> Text line orientation classification results. Returns actual angle values (e.g., [0,0,1]) when enabled.</li>
+                    <li><code>text_rec_score_thresh</code>: <code>(float)</code> Text recognition result filtering threshold.</li>
+                    <li><code>rec_texts</code>: <code>(List[str])</code> List of text recognition results, only containing texts with confidence scores higher than <code>text_rec_score_thresh</code>.</li>
+                    <li><code>rec_scores</code>: <code>(List[float])</code> List of text recognition confidence scores, filtered by <code>text_rec_score_thresh</code>.</li>
+                    <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> List of filtered text detection boxes, format same as <code>dt_polys</code>.</li>
+                </ol>
+            <li><code>formula_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> List of formula recognition results, where each element is a dictionary. The list order is the reading order after parsing.</li>
+                <ol start = "1" type = "a" > 
+                    <li><code>rec_formula</code>: <code>(str)</code> Formula recognition result.</li>
+                    <li><code>rec_polys</code>: <code>(numpy.ndarray)</code> Bounding box of the formula region, shape is (4, 2), dtype is int16.</li>
+                    <li><code>formula_region_id</code>: <code>(int)</code> ID of the layout region where the formula belongs.</li>
+                </ol>
+            <li><code>seal_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> List of seal text recognition results, where each element is a dictionary. The list order is the reading order after parsing.</li>
+                <ol start = "1" type = "a" > 
+                    <li><code>input_path</code>: <code>(str)</code> Input path of the seal image.</li>
+                    <li><code>model_settings</code>: <code>(Dict)</code> Seal text recognition sub-pipeline model configuration parameters.</li>
+                    <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> List of seal detection boxes, format same as <code>dt_polys</code></li>
+                    <li><code>dt_scores</code>: <code>(List[float])</code> List of seal detection box confidence scores.</li>
+                    <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> Seal text detection module configuration parameters, same as above.</li>
+                    <li><code>text_type</code>: <code>(str)</code> Seal text detection type, currently fixed as "seal".</li>
+                    <li><code>text_rec_score_thresh</code>: <code>(float)</code> Seal text recognition result filtering threshold.</li>
+                    <li><code>rec_texts</code>: <code>(List[str])</code> List of seal text recognition results, only containing texts with confidence scores higher than <code>text_rec_score_thresh</code>.</li>
+                    <li><code>rec_scores</code>: <code>(List[float])</code> List of seal text recognition confidence scores, filtered by <code>text_rec_score_thresh</code>.</li>
+                    <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> List of filtered seal detection boxes, format same as <code>dt_polys</code></li>
+                    <li><code>rec_boxes</code>: <code>(numpy.ndarray)</code> Array of rectangular bounding boxes for seal detection boxes, shape is (n, 4), dtype is int16. Each row represents a rectangle.</li>
+                </ol>
+            <li><code>table_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> List of table recognition results, where each element is a dictionary. The list order is the reading order after parsing.</li>
+                <ol start = "1" type = "a" > 
+                    <li><code>cell_bbox_list</code>: <code>(List[numpy.ndarray])</code> List of bounding boxes for table cells, shape is (4, 2), dtype is int16.</li>
+                    <li><code>pred_html</code>: <code>(str)</code> HTML format string for the table.</li>
+                    <li><code>table_ocr_pred</code>: <code>(dict)</code> Table OCR recognition results.</li>
+                    <ul>
+                        <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> List of bounding boxes for table cells, shape is (4, 2), dtype is int16.</li>
+                        <li><code>rec_texts</code>: <code>(List[str])</code> List of table cell recognition results.</li>
+                        <li><code>rec_scores</code>: <code>(List[float])</code> List of table cell recognition confidence scores.</li>
+                        <li><code>rec_boxes</code>: <code>(numpy.ndarray)</code> Array of rectangular bounding boxes for table cells, shape is (n, 4), dtype is int16. Each row represents a rectangle.</li>
+                    </ul>
+                </ol>
+            </ol>
+    </li>
+    <li>Calling the <code>save_to_json()</code> method will save the above content to the specified <code>save_path</code>. If a directory is specified, the save path will be <code>save_path/{your_img_basename}.json</code>. If a file is specified, it will be saved directly to that file. Since JSON files do not support saving numpy arrays, <code>numpy.array</code> types will be converted to list form.</li>
+    <li>Calling the <code>save_to_img()</code> method will save the visualization results to the specified <code></code>. If a directory is specified, the save path will be <code>save_path/{your_img_basename}_ocr_res_img.{your_img_extension}</code>. If a file is specified, it will be saved directly to that file. (The pipeline usually contains many result images, so it is not recommended to specify a specific file path directly, otherwise multiple images will be overwritten, and only the last image will be retained).</li>
+
+</ul>
+
+Additionally, it supports obtaining visualization images with results and prediction results through properties, as follows:
+<table>
+<thead>
+<tr>
+<th>Property</th>
+<th>Property Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">Gets the prediction results in <code>json</code> format.</td>
+</tr>
+<tr>
+<td rowspan = "2"><code>img</code></td>
+<td rowspan = "2">Gets the visualization images in <code>dict</code> format.</td>
+</tr>
+</table>
+
+<ul>
+    <li>The prediction result obtained by the <code>json</code> property is dict-type data, and its content is consistent with the content saved by calling the <code>save_to_json()</code> method.</li>
+    <li>The prediction result returned by the <code>img</code> property is a dictionary-type data. The keys are <code>layout_det_res</code>, <code>overall_ocr_res</code>, <code>text_paragraphs_ocr_res</code>, <code>formula_res_region1</code>, <code>table_cell_img</code>, and <code>seal_res_region1</code>, and the corresponding values are <code>Image.Image</code> objects: used to display visualization images of layout region detection, OCR, OCR text paragraphs, formulas, tables, and seal results, respectively. If optional modules are not used, the dictionary will only contain <code>layout_det_res</code>.</li>
+</ul>
+</details>
+
+<details><summary>(4) Call the <code>build_vector()</code> method of the PP-ChatOCRv4 pipeline object to build vectors for the text content.</summary>
+
+The following are the parameters and their descriptions for the <code>build_vector()</code> method:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Parameter Description</th>
+<th>Parameter Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td><code>visual_info</code></td>
+<td><b>Meaning:</b>Visual information, can be a dictionary containing visual information, or a list of such dictionaries.</td>
+<td><code>list|dict</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>min_characters</code></td>
+<td><b>Meaning:</b>Minimum number of characters. <br/>
+<b>Description:</b>
+A positive integer greater than 0, can be determined based on the token length supported by the large language model.</td>
+<td><code>int</code></td>
+<td><code>3500</code></td>
+</tr>
+<tr>
+<td><code>block_size</code></td>
+<td><b>Meaning:</b>Block size when building a vector library for long text. <br/>
+<b>Description:</b>
+A positive integer greater than 0, can be determined based on the token length supported by the large language model.</td>
+<td><code>int</code></td>
+<td><code>300</code></td>
+</tr>
+<tr>
+<td><code>flag_save_bytes_vector</code></td>
+<td><b>Meaning:</b>Whether to save text as a binary file.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>retriever_config</code></td>
+<td><b>Meaning:</b>Configuration parameters for the vector retrieval large model, same as the parameter during instantiation. <br/>
+<b>Description:</b>
+If set to <code>None</code>, uses instantiation parameters; otherwise, this parameter takes precedence.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</table>
+This method returns a dictionary containing visual text information. The content of the dictionary is as follows:
+
+<ul>
+    <li><code>flag_save_bytes_vector</code>：<code>(bool)</code>Whether to save the result as a binary file.</li>
+    <li><code>flag_too_short_text</code>：<code>(bool)</code>Whether the text length is less than the minimum number of characters.</li>
+    <li><code>vector</code>: <code>(str|list)</code> Binary content of the text or the text content itself, depending on the values of <code>flag_save_bytes_vector</code>and<code>min_characters</code>. If<code>flag_save_bytes_vector=True</code>and the text length is greater than or equal to the minimum number of characters, it returns binary content; otherwise, it returns the original text.</li>
+</ul>
+</details>
+
+<details><summary>(5) Call the <code>mllm_pred()</code> method of the PP-ChatOCRv4 pipeline object to get the extraction results from the multimodal large model.</summary>
+
+The following are the parameters and their descriptions for the <code>mllm_pred()</code> method:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Parameter Description</th>
+<th>Parameter Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted, supports multiple input types, required.<br/>
+<b>Description:</b>
+<ul>
+  <li><b>Python Var</b>: e.g., image data represented by <code>numpy.ndarray</code>; </li>
+  <li><b>str</b>: e.g., local path of an image file or single-page PDF file: <code>/root/data/img.jpg</code>;<b>URL link</b>, e.g., network URL of an image file or single-page PDF file: <a href = "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png">Example</a>.</li>
+</ul>
+</td>
+<td><code>Python Var|str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>key_list</code></td>
+<td><b>Meaning:</b>A single key or a list of keys used for extracting information.</td>
+<td><code>Union[str, List[str]]</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>mllm_chat_bot_config</code></td>
+<td><b>Meaning:</b>Configuration parameters for the multimodal large model, same as the parameter during instantiation. <br/>
+<b>Description:</b>
+If set to <code>None</code>, uses instantiation parameters; otherwise, this parameter takes precedence.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+</details>
+
+<details><summary>(6) Call the <code>chat()</code> method of the PP-ChatOCRv4 pipeline object to extract key information.</summary>
+
+The following are the parameters and their descriptions for the <code>chat()</code> method:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Parameter Description</th>
+<th>Parameter Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>key_list</code></td>
+<td><b>Meaning:</b>A single key or a list of keys used for extracting information.</td>
+<td><code>Union[str, List[str]]</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>visual_info</code></td>
+<td><b>Meaning:</b>Visual information result.</td>
+<td><code>List[dict]</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_vector_retrieval</code></td>
+<td><b>Meaning:</b>Whether to use vector retrieval.</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>vector_info</code></td>
+<td><b>Meaning:</b>Vector information used for retrieval.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>min_characters</code></td>
+<td><b>Meaning:</b>Required minimum number of characters. <br/>
+<b>Description:</b>
+A positive integer greater than 0.</td>
+<td><code>int</code></td>
+<td><code>3500</code></td>
+</tr>
+<tr>
+<td><code>text_task_description</code></td>
+<td><b>Meaning:</b>Description of the text task.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_output_format</code></td>
+<td><b>Meaning:</b>Output format for text results.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rules_str</code></td>
+<td><b>Meaning:</b>Rules for generating text results.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_few_shot_demo_text_content</code></td>
+<td><b>Meaning:</b>Text content for few-shot demonstration.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_few_shot_demo_key_value_list</code></td>
+<td><b>Meaning:</b>Key-value list for few-shot demonstration.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_task_description</code></td>
+<td><b>Meaning:</b>Description of the table task.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_output_format</code></td>
+<td><b>Meaning:</b>Output format for table results.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_rules_str</code></td>
+<td><b>Meaning:</b>Rules for generating table results.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_few_shot_demo_text_content</code></td>
+<td><b>Meaning:</b>Text content for table few-shot demonstration.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_few_shot_demo_key_value_list</code></td>
+<td><b>Meaning:</b>Key-value list for table few-shot demonstration.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>mllm_predict_info</code></td>
+<td><b>Meaning:</b>Multimodal large model result.</td>
+<td><code>dict|None</code></td>
+<td>
+<code>None</code>
+</td>
+</tr>
+<td><code>mllm_integration_strategy</code></td>
+<td><b>Meaning:</b>Data fusion strategy for multimodal large model and large language model, supports using one of them separately or fusing the results of both. Options: "integration", "llm_only", and "mllm_only".</td>
+<td><code>str</code></td>
+<td><code>"integration"</code></td>
+</tr>
+<tr>
+<td><code>chat_bot_config</code></td>
+<td><b>Meaning:</b>Configuration information for the large language model, same as the parameter during instantiation.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>retriever_config</code></td>
+<td><b>Meaning:</b>Configuration parameters for the vector retrieval large model, same as the parameter during instantiation. 
+<b>Description:</b>If set to <code>None</code>, uses instantiation parameters; otherwise, this parameter takes precedence.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+This method will print the result to the terminal. The content printed to the terminal is explained as follows:
+<ul>
+    <li><code>chat_res</code>: <code>(dict)</code> The result of information extraction, which is a dictionary containing the keys to be extracted and their corresponding values.</li>
+</ul>
+</details>
+
+
+## 3. Development Integration/Deployment
+If the pipeline meets your requirements for inference speed and accuracy in production, you can proceed directly with development integration/deployment.
+
+If you need to apply the pipeline directly in your Python project, you can refer to the sample code in [2.2  Python Script Experience](#22-python-script-experience).
+
+Additionally, PaddleX provides two other deployment methods, detailed as follows:
+
+🚀 **High-Performance Inference**: In actual production environments, many applications have stringent standards for the performance metrics of deployment strategies (especially response speed) to ensure efficient system operation and smooth user experience. To this end, PaddleX provides a high-performance inference plugin aimed at deeply optimizing model inference and pre/post-processing to significantly speed up the end-to-end process. For detailed instructions on high-performance inference, please refer to the [High-Performance Inference Guide](../inference_deployment/local_inference/high_performance_inference.en.md).
+
+☁️ **Serving**: Serving is a common deployment form in actual production environments. By encapsulating the inference functionality as a service, clients can access these services through network requests to obtain inference results. PaddleX supports multiple serving solutions for pipelines. For detailed instructions on serving, please refer to the [Service Deployment Guide](../inference_deployment/serving/serving.en.md).
+
+Below are the API references for basic serving and multi-language service invocation examples:
+
+<details><summary>API Reference</summary>
+
+<p>For the main operations provided by the service:</p>
+<ul>
+<li>The HTTP request method is POST.</li>
+<li>Both the request body and response body are JSON data (JSON objects).</li>
+<li>When the request is successfully processed, the response status code is <code>200</code>, and the response body has the following properties:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>UUID of the request.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Fixed at <code>0</code>.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error description. Fixed at <code>"Success"</code>.</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>Operation result.</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is not successfully processed, the response body has the following properties:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>UUID of the request.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Same as the response status code.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error description.</td>
+</tr>
+</tbody>
+</table>
+<p>The main operations provided by the service are as follows:</p>
+<ul>
+<li><b><code>analyzeImages</code></b></li>
+</ul>
+<p>Uses computer vision models to analyze images, obtain OCR, table recognition results, etc., and extract key information from the images.</p>
+<p><code>POST /chatocr-visual</code></p>
+<ul>
+<li>Properties of the request body:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+<th>Required</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>URL of image files (including TIFF; multi-page TIFF is processed page by page) or PDF file accessible to the server, or Base64 encoded result of the content of the above file types. By default, for PDF or multi-page TIFF files exceeding 10 pages, only the first 10 pages will be processed.<br />
+To remove the page limit, please add the following configuration to the pipeline configuration file:
+<pre><code>Serving:
+  extra:
+    max_num_input_imgs: null
+</code></pre></td>
+<td>Yes</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>File type. <code>0</code> represents a PDF file, <code>1</code> represents an image file. If this property is not present in the request body, the file type will be inferred based on the URL.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_doc_orientation_classify</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_doc_unwarping</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useSealRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_seal_recognition</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useTableRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_table_recognition</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutThreshold</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>layout_threshold</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutNms</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>layout_nms</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutUnclipRatio</code></td>
+<td><code>number</code> | <code>array</code> | <code>object</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>layout_unclip_ratio</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutMergeBboxesMode</code></td>
+<td><code>string</code> | <code>object</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>layout_merge_bboxes_mode</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>text_det_limit_side_len</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>text_det_limit_type</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>text_det_thresh</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>text_det_box_thresh</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>text_det_unclip_ratio</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>text_rec_score_thresh</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_det_limit_side_len</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_det_limit_type</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_det_thresh</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_det_box_thresh</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_det_unclip_ratio</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_rec_score_thresh</code> parameter of the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>
+Whether to return the final visualization image and intermediate images during the processing.<br/>
+<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>If <code>true</code> is provided: return images.</li>
+<li>If <code>false</code> is provided: do not return any images.</li>
+<li>If this parameter is omitted from the request body, or if <code>null</code> is explicitly passed, the behavior will follow the value of <code>Serving.visualize</code> in the pipeline configuration.</li>
+</ul>
+<br/>
+For example, adding the following setting to the pipeline config file:<br/>
+<pre><code>Serving:
+  visualize: False
+</code></pre>
+will disable image return by default. This behavior can be overridden by explicitly setting the <code>visualize</code> parameter in the request.<br/>
+If neither the request body nor the configuration file is set (If <code>visualize</code> is set to <code>null</code> in the request and  not defined in the configuration file), the image is returned by default.
+</td>
+<td>No</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is successfully processed, the <code>result</code> of the response body has the following properties:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>layoutParsingResults</code></td>
+<td><code>array</code></td>
+<td>Analysis results obtained using computer vision models. The array length is 1 (for image input) or the actual number of document pages processed (for PDF input). For PDF input, each element in the array represents the result of each page actually processed in the PDF file.</td>
+</tr>
+<tr>
+<td><code>visualInfo</code></td>
+<td><code>array</code></td>
+<td>Key information in the image, which can be used as input for other operations.</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>Input data information.</td>
+</tr>
+</tbody>
+</table>
+<p>Image fields in the element schema below (e.g. <code>outputImages</code>, <code>inputImage</code>) are returned inline as Base64 strings by default; when the server is configured to return URLs, those values become pre-signed URLs while the field types remain unchanged. See the "Returning Binary Content as URLs" section of the <a href="../inference_deployment/serving/serving.en.md">Serving Deployment Guide</a> for configuration.</p>
+<p>Each element in <code>layoutParsingResults</code> is an <code>object</code> with the following properties:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>A simplified version of the <code>res</code> field in the JSON representation of the <code>layout_parsing_result</code> generated by the pipeline object's <code>visual_predict</code> method, with the <code>input_path</code> and <code>page_index</code> fields removed.</td>
+</tr>
+<tr>
+<td><code>outputImages</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Refer to the description of <code>img</code> attribute of the pipeline's visual prediction result.</td>
+</tr>
+<tr>
+<td><code>inputImage</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Input image. The image is in JPEG format and encoded using Base64 by default; returned as a pre-signed URL when URL-return mode is enabled.</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li><b><code>buildVectorStore</code></b></li>
+</ul>
+<p>Builds a vector database.</p>
+<p><code>POST /chatocr-vector</code></p>
+<ul>
+<li>Properties of the request body:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+<th>Required</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>visualInfo</code></td>
+<td><code>array</code></td>
+<td>Key information in the image. Provided by the <code>analyzeImages</code> operation.</td>
+<td>Yes</td>
+</tr>
+<tr>
+<td><code>minCharacters</code></td>
+<td><code>integer</code></td>
+<td>Please refer to the description of the <code>min_characters</code> parameter of the pipeline object's <code>build_vector</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>blockSize</code></td>
+<td><code>integer</code></td>
+<td>Please refer to the description of the <code>block_size</code> parameter of the pipeline object's <code>build_vector</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>retrieverConfig</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>retriever_config</code> parameter of the pipeline object's <code>build_vector</code> method.</td>
+<td>No</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is successfully processed, the <code>result</code> of the response body has the following properties:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>vectorInfo</code></td>
+<td><code>object</code></td>
+<td>Serialized result of the vector database, which can be used as input for other operations.</td>
+</tr>
+</tbody>
+</table>
+<li><b><code>invokeMLLM</code></b></li>
+</ul>
+<p>Invoke the MLLM.</p>
+<p><code>POST /chatocr-mllm</code></p>
+<ul>
+<li>Properties of the request body:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+<th>Required</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>image</code></td>
+<td><code>string</code></td>
+<td>URL of an image file accessible by the server or the Base64-encoded content of the image file.</td>
+<td>Yes</td>
+</tr>
+<tr>
+<td><code>keyList</code></td>
+<td><code>array</code></td>
+<td>List of keys.</td>
+<td>Yes</td>
+</tr>
+<tr>
+<td><code>mllmChatBotConfig</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>mllm_chat_bot_config</code> parameter of the pipeline object's <code>mllm_pred</code> method.</td>
+<td>No</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is successfully processed, the <code>result</code> of the response body has the following property:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>mllmPredictInfo</code></td>
+<td><code>object</code></td>
+<td>MLLM invocation result.</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li><b><code>chat</code></b></li>
+</ul>
+<p>Interacts with large language models to extract key information using them.</p>
+<p><code>POST /chatocr-chat</code></p>
+<ul>
+<li>Properties of the request body:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+<th>Required</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>keyList</code></td>
+<td><code>array</code></td>
+<td>List of keys.</td>
+<td>Yes</td>
+</tr>
+<tr>
+<td><code>visualInfo</code></td>
+<td><code>object</code></td>
+<td>Key information in the image. Provided by the <code>analyzeImages</code> operation.</td>
+<td>Yes</td>
+</tr>
+<tr>
+<td><code>useVectorRetrieval</code></td>
+<td><code>boolean</code></td>
+<td>Please refer to the description of the <code>use_vector_retrieval</code> parameter of the pipeline object's <code>chat</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>vectorInfo</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Serialized result of the vector database. Provided by the <code>buildVectorStore</code> operation. Please note that the deserialization process involves performing an unpickle operation. To prevent malicious attacks, be sure to use data from trusted sources.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>minCharacters</code></td>
+<td><code>integer</code></td>
+<td>Please refer to the description of the <code>min_characters</code> parameter of the pipeline object's <code>chat</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textTaskDescription</code></td>
+<td><code>string</code></td>
+<td>Please refer to the description of the <code>text_task_description</code> parameter of the pipeline object's <code>chat</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textOutputFormat</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>text_output_format</code> parameter of the pipeline object's <code>chat</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textRulesStr</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>text_rules_str</code> parameter of the pipeline object's <code>chat</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textFewShotDemoTextContent</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>text_few_shot_demo_text_content</code> parameter of the pipeline object's <code>chat</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textFewShotDemoKeyValueList</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>text_few_shot_demo_key_value_list</code> parameter of the pipeline object's <code>chat</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>tableTaskDescription</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>table_task_description</code> parameter of the pipeline object's <code>chat</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>tableOutputFormat</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>table_output_format</code> parameter of the pipeline object's <code>chat</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>tableRulesStr</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>table_rules_str</code> parameter of the pipeline object's <code>chat</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>tableFewShotDemoTextContent</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>table_few_shot_demo_text_content</code> parameter of the pipeline object's <code>chat</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>tableFewShotDemoKeyValueList</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>table_few_shot_demo_key_value_list</code> parameter of the pipeline object's <code>chat</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>mllmPredictInfo</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>MLLM invocation result. Provided by the <code>invokeMllm</code> operation.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>mllmIntegrationStrategy</code></td>
+<td><code>string</code></td>
+<td>Please refer to the description of the <code>mllm_integration_strategy</code> parameter of the pipeline object's <code>chat</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>chatBotConfig</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>chat_bot_config</code> parameter of the pipeline object's <code>chat</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>retrieverConfig</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>retriever_config</code> parameter of the pipeline object's <code>chat</code> method.</td>
+<td>No</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is successfully processed, the <code>result</code> of the response body has the following properties:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>chatResult</code></td>
+<td><code>object</code></td>
+<td>Key information extraction result.</td>
+</tr>
+</tbody>
+</table>
+<li><b>Note:</b></li>
+Including sensitive parameters such as API key for large model calls in the request body can be a security risk. If not necessary, set these parameters in the configuration file and do not pass them on request.
+<br/><br/>
+</details>
+
+<details><summary>Multi-language Service Invocation Examples</summary>
+
+<details>
+<summary>Python</summary>
+
+<pre><code class="language-python">
+# This script only shows the use case for images. For calling with other file types, please read the API reference and make adjustments.
+
+import base64
+import pprint
+import sys
+import requests
+
+
+API_BASE_URL = "http://127.0.0.1:8080"
+
+image_path = "./demo.jpg"
+keys = ["name"]
+
+with open(image_path, "rb") as file:
+    image_bytes = file.read()
+    image_data = base64.b64encode(image_bytes).decode("ascii")
+
+payload = {
+    "file": image_data,
+    "fileType": 1,
+}
+
+resp_visual = requests.post(url=f"{API_BASE_URL}/chatocr-visual", json=payload)
+if resp_visual.status_code != 200:
+    print(
+        f"Request to chatocr-visual failed with status code {resp_visual.status_code}."
+    )
+    pprint.pp(resp_visual.json())
+    sys.exit(1)
+result_visual = resp_visual.json()["result"]
+
+for i, res in enumerate(result_visual["layoutParsingResults"]):
+    print(res["prunedResult"])
+    for img_name, img in res["outputImages"].items():
+        img_path = f"{img_name}_{i}.jpg"
+        with open(img_path, "wb") as f:
+            f.write(base64.b64decode(img))
+        print(f"Output image saved at {img_path}")
+
+payload = {
+    "visualInfo": result_visual["visualInfo"],
+}
+resp_vector = requests.post(url=f"{API_BASE_URL}/chatocr-vector", json=payload)
+if resp_vector.status_code != 200:
+    print(
+        f"Request to chatocr-vector failed with status code {resp_vector.status_code}."
+    )
+    pprint.pp(resp_vector.json())
+    sys.exit(1)
+result_vector = resp_vector.json()["result"]
+
+payload = {
+    "image": image_data,
+    "keyList": keys,
+}
+resp_mllm = requests.post(url=f"{API_BASE_URL}/chatocr-mllm", json=payload)
+if resp_mllm.status_code != 200:
+    print(
+        f"Request to chatocr-mllm failed with status code {resp_mllm.status_code}."
+    )
+    pprint.pp(resp_mllm.json())
+    sys.exit(1)
+result_mllm = resp_mllm.json()["result"]
+
+payload = {
+    "keyList": keys,
+    "visualInfo": result_visual["visualInfo"],
+    "useVectorRetrieval": True,
+    "vectorInfo": result_vector["vectorInfo"],
+    "mllmPredictInfo": result_mllm["mllmPredictInfo"],
+}
+resp_chat = requests.post(url=f"{API_BASE_URL}/chatocr-chat", json=payload)
+if resp_chat.status_code != 200:
+    print(
+        f"Request to chatocr-chat failed with status code {resp_chat.status_code}."
+    )
+    pprint.pp(resp_chat.json())
+    sys.exit(1)
+result_chat = resp_chat.json()["result"]
+print("Final result:")
+print(result_chat["chatResult"])
+</code></pre>
+</details>
+
+<details><summary>C++</summary>
+
+<pre><code class="language-cpp">#include &lt;iostream&gt;
+#include &lt;fstream&gt;
+#include &lt;vector&gt;
+#include &lt;string&gt;
+#include "cpp-httplib/httplib.h" // https://github.com/Huiyicc/cpp-httplib
+#include "nlohmann/json.hpp" // https://github.com/nlohmann/json
+#include "base64.hpp" // https://github.com/tobiaslocker/base64
+
+using json = nlohmann::json;
+
+std::string encode_image(const std::string& path) {
+    std::ifstream file(path, std::ios::binary | std::ios::ate);
+    if (!file) throw std::runtime_error("File open error.");
+    std::streamsize size = file.tellg();
+    file.seekg(0, std::ios::beg);
+    std::vector<char> buf(size);
+    file.read(buf.data(), size);
+    return base64::to_base64(std::string(buf.data(), buf.size()));
+}
+
+int main() {
+    httplib::Client client("localhost", 8080);
+    std::string imagePath = "./demo.jpg";
+    std::string imageData = encode_image(imagePath);
+    json keys = { "Name" };
+
+    json payload_visual = { {"file", imageData}, {"fileType", 1} };
+    auto resp1 = client.Post("/chatocr-visual", payload_visual.dump(), "application/json");
+    if (!resp1 || resp1->status != 200) {
+        std::cerr << "chatocr-visual failed.\n"; return 1;
+    }
+    json result_visual = json::parse(resp1->body)["result"];
+
+    for (size_t i = 0; i < result_visual["layoutParsingResults"].size(); ++i) {
+        auto& res = result_visual["layoutParsingResults"][i];
+        std::cout << "prunedResult: " << res["prunedResult"].dump() << "\n";
+        if (res.contains("outputImages")) {
+            for (auto& [name, b64] : res["outputImages"].items()) {
+                std::string outPath = name + "_" + std::to_string(i) + ".jpg";
+                std::string decoded = base64::from_base64(b64.get<std::string>());
+                std::ofstream out(outPath, std::ios::binary);
+                out.write(decoded.data(), decoded.size());
+                out.close();
+                std::cout << "Saved: " << outPath << "\n";
+            }
+        }
+    }
+
+    json payload_vector = { {"visualInfo", result_visual["visualInfo"]} };
+    auto resp2 = client.Post("/chatocr-vector", payload_vector.dump(), "application/json");
+    if (!resp2 || resp2->status != 200) {
+        std::cerr << "chatocr-vector failed.\n"; return 1;
+    }
+    json result_vector = json::parse(resp2->body)["result"];
+
+    json payload_mllm = { {"image", imageData}, {"keyList", keys} };
+    auto resp3 = client.Post("/chatocr-mllm", payload_mllm.dump(), "application/json");
+    if (!resp3 || resp3->status != 200) {
+        std::cerr << "chatocr-mllm failed.\n"; return 1;
+    }
+    json result_mllm = json::parse(resp3->body)["result"];
+
+    json payload_chat = {
+        {"keyList", keys},
+        {"visualInfo", result_visual["visualInfo"]},
+        {"useVectorRetrieval", true},
+        {"vectorInfo", result_vector["vectorInfo"]},
+        {"mllmPredictInfo", result_mllm["mllmPredictInfo"]}
+    };
+    auto resp4 = client.Post("/chatocr-chat", payload_chat.dump(), "application/json");
+    if (!resp4 || resp4->status != 200) {
+        std::cerr << "chatocr-chat failed.\n"; return 1;
+    }
+
+    json result_chat = json::parse(resp4->body)["result"];
+    std::cout << "Final chat result: " << result_chat["chatResult"] << std::endl;
+
+    return 0;
+}
+</code></pre></details>
+
+<details><summary>Java</summary>
+
+<pre><code class="language-java">import okhttp3.*;
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.util.Base64;
+import java.util.Iterator;
+
+public class Main {
+    public static void main(String[] args) throws IOException {
+        String API_BASE_URL = "http://localhost:8080";
+        String imagePath = "./demo.jpg";
+        String[] keys = {"Name"};
+
+        OkHttpClient client = new OkHttpClient();
+        ObjectMapper objectMapper = new ObjectMapper();
+        MediaType JSON = MediaType.parse("application/json; charset=utf-8");
+
+        byte[] imageBytes = java.nio.file.Files.readAllBytes(new File(imagePath).toPath());
+        String base64Image = Base64.getEncoder().encodeToString(imageBytes);
+
+        ObjectNode visualPayload = objectMapper.createObjectNode();
+        visualPayload.put("file", base64Image);
+        visualPayload.put("fileType", 1);
+
+        Request requestVisual = new Request.Builder()
+                .url(API_BASE_URL + "/chatocr-visual")
+                .post(RequestBody.create(JSON, visualPayload.toString()))
+                .build();
+
+        Response responseVisual = client.newCall(requestVisual).execute();
+        if (!responseVisual.isSuccessful()) {
+            System.err.println("chatocr-visual failed: " + responseVisual.code());
+            return;
+        }
+
+        JsonNode resultVisual = objectMapper.readTree(responseVisual.body().string()).get("result");
+
+        JsonNode layoutResults = resultVisual.get("layoutParsingResults");
+        for (int i = 0; i < layoutResults.size(); i++) {
+            JsonNode res = layoutResults.get(i);
+            System.out.println("prunedResult [" + i + "]: " + res.get("prunedResult").toString());
+
+            JsonNode outputImages = res.get("outputImages");
+            if (outputImages != null && outputImages.isObject()) {
+                Iterator<String> names = outputImages.fieldNames();
+                while (names.hasNext()) {
+                    String imgName = names.next();
+                    String imgBase64 = outputImages.get(imgName).asText();
+                    byte[] imgBytes = Base64.getDecoder().decode(imgBase64);
+                    String imgPath = imgName + "_" + i + ".jpg";
+                    try (FileOutputStream fos = new FileOutputStream(imgPath)) {
+                        fos.write(imgBytes);
+                        System.out.println("Saved image: " + imgPath);
+                    }
+                }
+            }
+        }
+
+        ObjectNode vectorPayload = objectMapper.createObjectNode();
+        vectorPayload.set("visualInfo", resultVisual.get("visualInfo"));
+
+        Request requestVector = new Request.Builder()
+                .url(API_BASE_URL + "/chatocr-vector")
+                .post(RequestBody.create(JSON, vectorPayload.toString()))
+                .build();
+
+        Response responseVector = client.newCall(requestVector).execute();
+        if (!responseVector.isSuccessful()) {
+            System.err.println("chatocr-vector failed: " + responseVector.code());
+            return;
+        }
+
+        JsonNode resultVector = objectMapper.readTree(responseVector.body().string()).get("result");
+
+        ObjectNode mllmPayload = objectMapper.createObjectNode();
+        mllmPayload.put("image", base64Image);
+        mllmPayload.putArray("keyList").add(keys[0]);
+
+        Request requestMllm = new Request.Builder()
+                .url(API_BASE_URL + "/chatocr-mllm")
+                .post(RequestBody.create(JSON, mllmPayload.toString()))
+                .build();
+
+        Response responseMllm = client.newCall(requestMllm).execute();
+        if (!responseMllm.isSuccessful()) {
+            System.err.println("chatocr-mllm failed: " + responseMllm.code());
+            return;
+        }
+
+        JsonNode resultMllm = objectMapper.readTree(responseMllm.body().string()).get("result");
+
+        ObjectNode chatPayload = objectMapper.createObjectNode();
+        chatPayload.putArray("keyList").add(keys[0]);
+        chatPayload.set("visualInfo", resultVisual.get("visualInfo"));
+        chatPayload.put("useVectorRetrieval", true);
+        chatPayload.set("vectorInfo", resultVector.get("vectorInfo"));
+        chatPayload.set("mllmPredictInfo", resultMllm.get("mllmPredictInfo"));
+
+        Request requestChat = new Request.Builder()
+                .url(API_BASE_URL + "/chatocr-chat")
+                .post(RequestBody.create(JSON, chatPayload.toString()))
+                .build();
+
+        Response responseChat = client.newCall(requestChat).execute();
+        if (!responseChat.isSuccessful()) {
+            System.err.println("chatocr-chat failed: " + responseChat.code());
+            return;
+        }
+
+        JsonNode resultChat = objectMapper.readTree(responseChat.body().string()).get("result");
+        System.out.println("Final result:");
+        System.out.println(resultChat.get("chatResult").toString());
+    }
+}
+</code></pre></details>
+
+<details><summary>Go</summary>
+
+<pre><code class="language-go">package main
+
+import (
+    "bytes"
+    "encoding/base64"
+    "encoding/json"
+    "fmt"
+    "io/ioutil"
+    "net/http"
+    "os"
+)
+
+func sendPostRequest(url string, payload map[string]interface{}) (map[string]interface{}, error) {
+    bodyBytes, err := json.Marshal(payload)
+    if err != nil {
+        return nil, fmt.Errorf("error marshaling payload: %v", err)
+    }
+
+    req, err := http.NewRequest("POST", url, bytes.NewBuffer(bodyBytes))
+    if err != nil {
+        return nil, fmt.Errorf("error creating request: %v", err)
+    }
+    req.Header.Set("Content-Type", "application/json")
+
+    client := &http.Client{}
+    resp, err := client.Do(req)
+    if err != nil {
+        return nil, fmt.Errorf("error sending request: %v", err)
+    }
+    defer resp.Body.Close()
+
+    if resp.StatusCode != http.StatusOK {
+        return nil, fmt.Errorf("status code error: %d", resp.StatusCode)
+    }
+
+    respBytes, err := ioutil.ReadAll(resp.Body)
+    if err != nil {
+        return nil, fmt.Errorf("error reading response: %v", err)
+    }
+
+    var result map[string]interface{}
+    if err := json.Unmarshal(respBytes, &result); err != nil {
+        return nil, fmt.Errorf("error unmarshaling response: %v", err)
+    }
+    return result["result"].(map[string]interface{}), nil
+}
+
+func main() {
+    apiBase := "http://localhost:8080"
+    imagePath := "./demo.jpg"
+    keys := []string{"Name"}
+
+    imageBytes, err := ioutil.ReadFile(imagePath)
+    if err != nil {
+        fmt.Printf("read image failed : %v\n", err)
+        return
+    }
+    imageData := base64.StdEncoding.EncodeToString(imageBytes)
+
+    visualPayload := map[string]interface{}{
+        "file":     imageData,
+        "fileType": 1,
+    }
+    visualResult, err := sendPostRequest(apiBase+"/chatocr-visual", visualPayload)
+    if err != nil {
+        fmt.Printf("chatocr-visual request error: %v\n", err)
+        return
+    }
+
+    layoutResults := visualResult["layoutParsingResults"].([]interface{})
+    for i, res := range layoutResults {
+        layout := res.(map[string]interface{})
+        fmt.Println("PrunedResult:", layout["prunedResult"])
+        outputImages := layout["outputImages"].(map[string]interface{})
+        for name, img := range outputImages {
+            imgBytes, _ := base64.StdEncoding.DecodeString(img.(string))
+            filename := fmt.Sprintf("%s_%d.jpg", name, i)
+            if err := os.WriteFile(filename, imgBytes, 0644); err == nil {
+                fmt.Printf("save image：%s\n", filename)
+            }
+        }
+    }
+
+    vectorPayload := map[string]interface{}{
+        "visualInfo": visualResult["visualInfo"],
+    }
+    vectorResult, err := sendPostRequest(apiBase+"/chatocr-vector", vectorPayload)
+    if err != nil {
+        fmt.Printf("chatocr-vector request error: %v\n", err)
+        return
+    }
+
+    mllmPayload := map[string]interface{}{
+        "image":   imageData,
+        "keyList": keys,
+    }
+    mllmResult, err := sendPostRequest(apiBase+"/chatocr-mllm", mllmPayload)
+    if err != nil {
+        fmt.Printf("chatocr-mllm request error: %v\n", err)
+        return
+    }
+
+    chatPayload := map[string]interface{}{
+        "keyList":           keys,
+        "visualInfo":        visualResult["visualInfo"],
+        "useVectorRetrieval": true,
+        "vectorInfo":        vectorResult["vectorInfo"],
+        "mllmPredictInfo":   mllmResult["mllmPredictInfo"],
+    }
+    chatResult, err := sendPostRequest(apiBase+"/chatocr-chat", chatPayload)
+    if err != nil {
+        fmt.Printf("chatocr-chat request error: %v\n", err)
+        return
+    }
+
+    fmt.Println("final result：", chatResult["chatResult"])
+}
+</code></pre></details>
+
+<details><summary>C#</summary>
+
+<pre><code class="language-csharp">using System;
+using System.IO;
+using System.Net.Http;
+using System.Text;
+using System.Threading.Tasks;
+using Newtonsoft.Json.Linq;
+
+class Program
+{
+    static readonly string API_BASE_URL = "http://localhost:8080";
+    static readonly string inputFilePath = "./demo.jpg";
+    static readonly string[] keys = { "Name" };
+
+    static async Task Main(string[] args)
+    {
+        var httpClient = new HttpClient();
+
+        byte[] imageBytes = File.ReadAllBytes(inputFilePath);
+        string imageData = Convert.ToBase64String(imageBytes);
+
+        var payloadVisual = new JObject
+        {
+            { "file", imageData },
+            { "fileType", 1 }
+        };
+
+        var respVisual = await httpClient.PostAsync($"{API_BASE_URL}/chatocr-visual",
+            new StringContent(payloadVisual.ToString(), Encoding.UTF8, "application/json"));
+
+        if (!respVisual.IsSuccessStatusCode)
+        {
+            Console.Error.WriteLine($"Request to chatocr-visual failed: {respVisual.StatusCode}");
+            Console.Error.WriteLine(await respVisual.Content.ReadAsStringAsync());
+            return;
+        }
+
+        JObject resultVisual = JObject.Parse(await respVisual.Content.ReadAsStringAsync())["result"] as JObject;
+
+        var layoutParsingResults = (JArray)resultVisual["layoutParsingResults"];
+        for (int i = 0; i < layoutParsingResults.Count; i++)
+        {
+            var res = layoutParsingResults[i];
+            Console.WriteLine($"[{i}] prunedResult:\n{res["prunedResult"]}");
+
+            JObject outputImages = res["outputImages"] as JObject;
+            if (outputImages != null)
+            {
+                foreach (var img in outputImages)
+                {
+                    string imgName = img.Key;
+                    string base64Img = img.Value?.ToString();
+                    if (!string.IsNullOrEmpty(base64Img))
+                    {
+                        string imgPath = $"{imgName}_{i}.jpg";
+                        File.WriteAllBytes(imgPath, Convert.FromBase64String(base64Img));
+                        Console.WriteLine($"Output image saved at {imgPath}");
+                    }
+                }
+            }
+        }
+
+        var payloadVector = new JObject
+        {
+            { "visualInfo", resultVisual["visualInfo"] }
+        };
+
+        var respVector = await httpClient.PostAsync($"{API_BASE_URL}/chatocr-vector",
+            new StringContent(payloadVector.ToString(), Encoding.UTF8, "application/json"));
+
+        if (!respVector.IsSuccessStatusCode)
+        {
+            Console.Error.WriteLine($"Request to chatocr-vector failed: {respVector.StatusCode}");
+            Console.Error.WriteLine(await respVector.Content.ReadAsStringAsync());
+            return;
+        }
+
+        JObject resultVector = JObject.Parse(await respVector.Content.ReadAsStringAsync())["result"] as JObject;
+
+        var payloadMllm = new JObject
+        {
+            { "image", imageData },
+            { "keyList", new JArray(keys) }
+        };
+
+        var respMllm = await httpClient.PostAsync($"{API_BASE_URL}/chatocr-mllm",
+            new StringContent(payloadMllm.ToString(), Encoding.UTF8, "application/json"));
+
+        if (!respMllm.IsSuccessStatusCode)
+        {
+            Console.Error.WriteLine($"Request to chatocr-mllm failed: {respMllm.StatusCode}");
+            Console.Error.WriteLine(await respMllm.Content.ReadAsStringAsync());
+            return;
+        }
+
+        JObject resultMllm = JObject.Parse(await respMllm.Content.ReadAsStringAsync())["result"] as JObject;
+
+        var payloadChat = new JObject
+        {
+            { "keyList", new JArray(keys) },
+            { "visualInfo", resultVisual["visualInfo"] },
+            { "useVectorRetrieval", true },
+            { "vectorInfo", resultVector["vectorInfo"] },
+            { "mllmPredictInfo", resultMllm["mllmPredictInfo"] }
+        };
+
+        var respChat = await httpClient.PostAsync($"{API_BASE_URL}/chatocr-chat",
+            new StringContent(payloadChat.ToString(), Encoding.UTF8, "application/json"));
+
+        if (!respChat.IsSuccessStatusCode)
+        {
+            Console.Error.WriteLine($"Request to chatocr-chat failed: {respChat.StatusCode}");
+            Console.Error.WriteLine(await respChat.Content.ReadAsStringAsync());
+            return;
+        }
+
+        JObject resultChat = JObject.Parse(await respChat.Content.ReadAsStringAsync())["result"] as JObject;
+        Console.WriteLine("Final result:");
+        Console.WriteLine(resultChat["chatResult"]);
+    }
+}
+</code></pre></details>
+
+<details><summary>Node.js</summary>
+
+<pre><code class="language-js">const axios = require('axios');
+const fs = require('fs');
+const path = require('path');
+
+const API_BASE_URL = 'http://localhost:8080';
+const imagePath = './demo.jpg';
+const keys = ['Name'];
+
+function encodeImageToBase64(filePath) {
+  const bitmap = fs.readFileSync(filePath);
+  return Buffer.from(bitmap).toString('base64');
+}
+
+(async () => {
+  try {
+    const imageData = encodeImageToBase64(imagePath);
+
+    const respVisual = await axios.post(`${API_BASE_URL}/chatocr-visual`, {
+      file: imageData,
+      fileType: 1
+    });
+
+    const resultVisual = respVisual.data.result;
+    resultVisual.layoutParsingResults.forEach((res, i) => {
+      console.log(`\n[${i}] prunedResult:\n`, res.prunedResult);
+      const outputImages = res.outputImages || {};
+      for (const [imgName, base64Img] of Object.entries(outputImages)) {
+        const fileName = `${imgName}_${i}.jpg`;
+        fs.writeFileSync(fileName, Buffer.from(base64Img, 'base64'));
+        console.log(`Output image saved at ${fileName}`);
+      }
+    });
+
+    const respVector = await axios.post(`${API_BASE_URL}/chatocr-vector`, {
+      visualInfo: resultVisual.visualInfo
+    });
+    const resultVector = respVector.data.result;
+
+    const respMllm = await axios.post(`${API_BASE_URL}/chatocr-mllm`, {
+      image: imageData,
+      keyList: keys
+    });
+    const resultMllm = respMllm.data.result;
+
+    const respChat = await axios.post(`${API_BASE_URL}/chatocr-chat`, {
+      keyList: keys,
+      visualInfo: resultVisual.visualInfo,
+      useVectorRetrieval: true,
+      vectorInfo: resultVector.vectorInfo,
+      mllmPredictInfo: resultMllm.mllmPredictInfo
+    });
+
+    const resultChat = respChat.data.result;
+    console.log('\nFinal result:\n', resultChat.chatResult);
+
+  } catch (error) {
+    if (error.response) {
+      console.error(`❌ Request failed: ${error.response.status}`);
+      console.error(error.response.data);
+    } else {
+      console.error('❌ Error occurred:', error.message);
+    }
+  }
+})();
+</code></pre></details>
+
+<details><summary>PHP</summary>
+
+<pre><code class="language-php">&lt;?php
+
+$API_BASE_URL = "http://localhost:8080";
+$image_path = "./demo.jpg";
+$keys = ["Name"];
+
+$image_data = base64_encode(file_get_contents($image_path));
+
+$payload_visual = [
+    "file" => $image_data,
+    "fileType" => 1
+];
+$response_visual_raw = send_post_raw("$API_BASE_URL/chatocr-visual", $payload_visual);
+$response_visual = json_decode($response_visual_raw, true);
+if (!isset($response_visual["result"])) {
+    echo "chatocr-visual request error\n";
+    print_r($response_visual);
+    exit(1);
+}
+$result_visual_raw = json_decode($response_visual_raw, false)->result;
+$result_visual_arr = $response_visual["result"];
+
+foreach ($result_visual_arr["layoutParsingResults"] as $i => $res) {
+    echo "[$i] prunedResult:\n";
+    print_r($res["prunedResult"]);
+    if (!empty($res["outputImages"])) {
+        foreach ($res["outputImages"] as $img_name => $base64_img) {
+            $img_path = "{$img_name}_{$i}.jpg";
+            file_put_contents($img_path, base64_decode($base64_img));
+            echo "Output image saved at $img_path\n";
+        }
+    }
+}
+
+$payload_vector = [
+    "visualInfo" => $result_visual_raw->visualInfo
+];
+$response_vector_raw = send_post_raw("$API_BASE_URL/chatocr-vector", $payload_vector);
+$response_vector = json_decode($response_vector_raw, true);
+if (!isset($response_vector["result"])) {
+    echo "chatocr-vector request error\n";
+    print_r($response_vector);
+    exit(1);
+}
+$result_vector_raw = json_decode($response_vector_raw, false)->result;
+
+$payload_mllm = [
+    "image" => $image_data,
+    "keyList" => $keys
+];
+$response_mllm_raw = send_post_raw("$API_BASE_URL/chatocr-mllm", $payload_mllm);
+$response_mllm = json_decode($response_mllm_raw, true);
+if (!isset($response_mllm["result"])) {
+    echo "chatocr-mllm request error\n";
+    print_r($response_mllm);
+    exit(1);
+}
+$result_mllm_raw = json_decode($response_mllm_raw, false)->result;
+
+$payload_chat = [
+    "keyList" => $keys,
+    "visualInfo" => $result_visual_raw->visualInfo,
+    "useVectorRetrieval" => true,
+    "vectorInfo" => $result_vector_raw->vectorInfo,
+    "mllmPredictInfo" => $result_mllm_raw->mllmPredictInfo
+];
+$response_chat_raw = send_post_raw("$API_BASE_URL/chatocr-chat", $payload_chat);
+$response_chat = json_decode($response_chat_raw, true);
+if (!isset($response_chat["result"])) {
+    echo "chatocr-chat request error\n";
+    print_r($response_chat);
+    exit(1);
+}
+
+echo "Final result:\n";
+echo json_encode($response_chat["result"]["chatResult"], JSON_UNESCAPED_UNICODE | JSON_PRETTY_PRINT) . "\n";
+
+
+function send_post_raw($url, $data) {
+    $json_str = json_encode($data, JSON_UNESCAPED_UNICODE);
+    $ch = curl_init($url);
+    curl_setopt($ch, CURLOPT_POST, true);
+    curl_setopt($ch, CURLOPT_POSTFIELDS, $json_str);
+    curl_setopt($ch, CURLOPT_HTTPHEADER, ['Content-Type: application/json']);
+    curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+    $response = curl_exec($ch);
+    if ($response === false) {
+        echo "cURL error: " . curl_error($ch) . "\n";
+    }
+    curl_close($ch);
+    return $response;
+}
+?&gt;
+</code></pre></details>
+</details>
+<br/>
+
+## 4. Custom Development
+If the default model weights provided by the PP-ChatOCRv4 pipeline do not meet your requirements in terms of accuracy or speed, you can try to fine-tune the existing model using your own domain-specific or application-specific data to improve the recognition performance of the PP-ChatOCRv4 pipeline in your scenario.
+
+
+### 4.1 Model Fine-Tuning
+Since the PP-ChatOCRv4 pipeline includes several modules, the unsatisfactory performance of the pipeline may originate from any one of these modules. You can analyze the cases with poor extraction results, identify which module is problematic through visualizing the images, and refer to the corresponding fine-tuning tutorial links in the table below to fine-tune the model.
+
+<table>
+<thead>
+<tr>
+<th>Scenario</th>
+<th>Fine-tuning Module</th>
+<th>Fine-tuning Reference Link</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>Inaccurate layout region detection, such as missed detection of seals, tables, etc.</td>
+<td>Layout Region Detection Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/layout_detection.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate table structure recognition</td>
+<td>Table Structure Recognition Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/module_usage/table_structure_recognition.html#4-secondary-development">Link</a></td>
+</tr>
+<tr>
+<td>Missed detection of seal text</td>
+<td>Seal Text Detection Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/module_usage/seal_text_detection.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Missed detection of text</td>
+<td>Text Detection Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/module_usage/text_detection.html#4-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate text content</td>
+<td>Text Recognition Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/module_usage/text_recognition.html#41-dataset-and-pre-trained-model-preparation">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate correction of vertical or rotated text lines</td>
+<td>Text Line Orientation Classification Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/textline_orientation_classification.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate correction of whole-image rotation</td>
+<td>Document Image Orientation Classification Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate correction of image distortion</td>
+<td>Text Image Correction Module</td>
+<td>Fine-tuning not supported</td>
+</tr>
+</tbody>
+</table>
+
+### 4.2 Model Application
+After you complete fine-tuning with your private dataset, you will obtain a local model weight file.
+
+If you need to use the fine-tuned model weights, simply modify the production configuration file by replacing the local directory of the fine-tuned model weights to the corresponding position in the production configuration file:
+
+1. Exporting Pipeline Configuration Files
+
+You can call the `export_paddlex_config_to_yaml` method of the pipeline object to export the current pipeline configuration to a YAML file. Here is an example:
+
+```Python
+from paddleocr import PPChatOCRv4Doc
+
+pipeline = PPChatOCRv4Doc()
+pipeline.export_paddlex_config_to_yaml("PP-ChatOCRv4-doc.yaml")
+```
+
+2. Editing Pipeline Configuration Files
+
+Replace the local directory of the fine-tuned model weights to the corresponding position in the pipeline configuration file. For example:
+
+```yaml
+......
+SubModules:
+    TextDetection:
+    module_name: text_detection
+    model_name: PP-OCRv5_server_det
+    model_dir: null # Replace with the fine-tuned text detection model weights directory
+    limit_side_len: 960
+    limit_type: max
+    thresh: 0.3
+    box_thresh: 0.6
+    unclip_ratio: 1.5
+
+    TextRecognition:
+    module_name: text_recognition
+    model_name: PP-OCRv5_server_rec
+    model_dir: null # Replace with the fine-tuned text recognition model weights directory
+        batch_size: 1
+    batch_size: 1
+            score_thresh: 0
+......
+```
+
+The exported PaddleX pipeline configuration file not only includes parameters supported by PaddleOCR's CLI and Python API but also allows for more advanced settings. Please refer to the corresponding pipeline usage tutorials in [PaddleX Pipeline Usage Overview](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/pipeline_develop_guide.html) for detailed instructions on adjusting various configurations according to your needs.
+
+
+3. Loading Pipeline Configuration Files in CLI
+
+By specifying the path to the PaddleX pipeline configuration file using the `--paddlex_config` parameter, PaddleOCR will read its contents as the configuration for inference. Here is an example:
+
+```bash
+paddleocr pp_chatocrv4_doc --paddlex_config PP-ChatOCRv4-doc.yaml ...
+```
+
+4. Loading Pipeline Configuration Files in Python API
+
+When initializing the pipeline object, you can pass the path to the PaddleX pipeline configuration file or a configuration dictionary through the `paddlex_config` parameter, and PaddleOCR will use it as the configuration for inference. Here is an example:
+
+```python
+from paddleocr import PPChatOCRv4Doc
+
+pipeline = PPChatOCRv4Doc(paddlex_config="PP-ChatOCRv4-doc.yaml")
+```
diff --git a/docs/version3.x/pipeline_usage/PP-ChatOCRv4.md b/docs/version3.x/pipeline_usage/PP-ChatOCRv4.md
new file mode 100644
index 0000000..81a706b
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PP-ChatOCRv4.md
@@ -0,0 +1,4029 @@
+---
+comments: true
+---
+
+# PP-ChatOCRv4 产线使用教程
+
+## 1. PP-ChatOCRv4 产线介绍
+
+PP-ChatOCRv4 是飞桨特色的文档和图像智能分析解决方案，结合了 LLM、MLLM 和 OCR 技术，一站式解决版面分析、生僻字、多页 pdf、表格、印章文本识别等常见的复杂文档信息抽取难点问题，结合文心大模型将海量数据和知识相融合，准确率高且应用广泛。本产线同时提供了灵活的服务化部署方式，支持在多种硬件上部署。不仅如此，本产线也提供了二次开发的能力，您可以基于本产线在您自己的数据集上训练调优，训练后的模型也可以无缝集成。
+
+<div align="center">
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/PP-ChatOCRv4/algorithm_ppchatocrv4.png" width="80%"/>
+</div>
+
+<b>PP-ChatOCRv4 产线中包含以下9个模块。每个模块均可独立进行训练和推理，并包含多个模型。有关详细信息，请点击相应模块以查看文档。</b>
+
+- [文档图像方向分类模块](../module_usage/doc_img_orientation_classification.md)（可选）
+- [文本图像矫正模块](../module_usage/text_image_unwarping.md)（可选）
+- [版面区域检测模块](../module_usage/layout_detection.md)
+- [表格结构识别模块](../module_usage/table_structure_recognition.md)（可选）
+- [文本检测模块](../module_usage/text_detection.md)
+- [文本识别模块](../module_usage/text_recognition.md)
+- [文本行方向分类模块](../module_usage/textline_orientation_classification.md)（可选）
+- [公式识别模块](../module_usage/formula_recognition.md)（可选）
+- [印章文本检测模块](../module_usage/seal_text_detection.md)（可选）
+
+在本产线中，您可以根据下方的基准测试数据选择使用的模型。
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。
+> 在带有 [常规模式 / 高性能模式] 标记的推理耗时列中，`常规模式` 对应本地飞桨推理引擎。各模块会根据默认模型名称选择合适的本地飞桨推理引擎：仅支持动态图的模型使用 `paddle_dynamic`；同时支持静态图和动态图的模型优先使用 `paddle_static`。
+
+<details>
+<summary> <b>文档图像方向分类模块（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>Top-1 Acc（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">训练模型</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>基于PP-LCNet_x1_0的文档图像分类模型，含有四个类别，即0度，90度，180度，270度</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary> <b>文本图像矫正模块（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>CER </th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">训练模型</a></td>
+<td>0.179</td>
+<td>19.05 / 19.05</td>
+<td>- / 869.82</td>
+<td>30.3</td>
+<td>高精度文本图像矫正模型</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary> <b>版面区域检测模块模型：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-L_pretrained.pdparams">训练模型</a></td>
+<td>90.4</td>
+<td>33.59 / 33.59</td>
+<td>503.01 / 251.08</td>
+<td>123.76</td>
+<td>基于RT-DETR-L在包含中英文论文、杂志、合同、书本、试卷和研报等场景的自建数据集训练的高精度版面区域定位模型</td>
+</tr>
+<tr>
+<td>PP-DocLayout-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-M_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-M_pretrained.pdparams">训练模型</a></td>
+<td>75.2</td>
+<td>13.03 / 4.72</td>
+<td>43.39 / 24.44</td>
+<td>22.578</td>
+<td>基于PicoDet-L在包含中英文论文、杂志、合同、书本、试卷和研报等场景的自建数据集训练的精度效率平衡的版面区域定位模型</td>
+</tr>
+<tr>
+<td>PP-DocLayout-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-S_pretrained.pdparams">训练模型</a></td>
+<td>70.9</td>
+<td>11.54 / 3.86</td>
+<td>18.53 / 6.29</td>
+<td>4.834</td>
+<td>基于PicoDet-S在中英文论文、杂志、合同、书本、试卷和研报等场景上自建数据集训练的高效率版面区域定位模型</td>
+</tr>
+<tr>
+<td>PicoDet_layout_1x</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_pretrained.pdparams">训练模型</a></td>
+<td>97.8</td>
+<td>9.62 / 6.75</td>
+<td>26.96 / 12.77</td>
+<td>7.4</td>
+<td>基于PicoDet-1x在PubLayNet数据集训练的高效率版面区域定位模型，可定位包含文字、标题、表格、图片以及列表这5类区域</td>
+</tr>
+<tr>
+<td>PicoDet_layout_1x_table</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_table_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_table_pretrained.pdparams">训练模型</a></td>
+<td>97.5</td>
+<td>9.57 / 6.63</td>
+<td>27.66 / 16.75</td>
+<td>7.4</td>
+<td>基于PicoDet-1x在自建数据集训练的高效率版面区域定位模型，可定位包含表格这1类区域</td>
+</tr>
+<tr>
+<td>PicoDet-S_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_3cls_pretrained.pdparams">训练模型</a></td>
+<td>88.2</td>
+<td>8.43 / 3.44</td>
+<td>17.60 / 6.51</td>
+<td>4.8</td>
+<td>基于PicoDet-S轻量模型在中英文论文、杂志和研报等场景上自建数据集训练的高效率版面区域定位模型，包含3个类别：表格，图像和印章</td>
+</tr>
+<tr>
+<td>PicoDet-S_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>87.4</td>
+<td>8.80 / 3.62</td>
+<td>17.51 / 6.35</td>
+<td>4.8</td>
+<td>基于PicoDet-S轻量模型在中英文论文、杂志和研报等场景上自建数据集训练的高效率版面区域定位模型，包含17个版面常见类别，分别是：段落标题、图片、文本、数字、摘要、内容、图表标题、公式、表格、表格标题、参考文献、文档标题、脚注、页眉、算法、页脚、印章</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_3cls_pretrained.pdparams">训练模型</a></td>
+<td>89.0</td>
+<td>12.80 / 9.57</td>
+<td>45.04 / 23.86</td>
+<td>22.6</td>
+<td>基于PicoDet-L在中英文论文、杂志和研报等场景上自建数据集训练的高效率版面区域定位模型，包含3个类别：表格，图像和印章</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>89.0</td>
+<td>12.60 / 10.27</td>
+<td>43.70 / 24.42</td>
+<td>22.6</td>
+<td>基于PicoDet-L在中英文论文、杂志和研报等场景上自建数据集训练的高效率版面区域定位模型，包含17个版面常见类别，分别是：段落标题、图片、文本、数字、摘要、内容、图表标题、公式、表格、表格标题、参考文献、文档标题、脚注、页眉、算法、页脚、印章</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_3cls_pretrained.pdparams">训练模型</a></td>
+<td>95.8</td>
+<td>114.80 / 25.65</td>
+<td>924.38 / 924.38</td>
+<td>470.1</td>
+<td>基于RT-DETR-H在中英文论文、杂志和研报等场景上自建数据集训练的高精度版面区域定位模型，包含3个类别：表格，图像和印章</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>98.3</td>
+<td>115.29 / 101.18</td>
+<td>964.75 / 964.75</td>
+<td>470.2</td>
+<td>基于RT-DETR-H在中英文论文、杂志和研报等场景上自建数据集训练的高精度版面区域定位模型，包含17个版面常见类别，分别是：段落标题、图片、文本、数字、摘要、内容、图表标题、公式、表格、表格标题、参考文献、文档标题、脚注、页眉、算法、页脚、印章</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary> <b>表格结构识别模块（可选）：</b></summary>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>精度（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>SLANet</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANet_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_pretrained.pdparams">训练模型</a></td>
+<td>59.52</td>
+<td>23.96 / 21.75</td>
+<td>- / 43.12</td>
+<td>6.9</td>
+<td>SLANet 是百度飞桨视觉团队自研的表格结构识别模型。该模型通过采用CPU 友好型轻量级骨干网络 PP-LCNet、高低层特征融合模块CSP-PAN、结构与位置信息对齐的特征解码模块 SLA Head，大幅提升了表格结构识别的精度和推理速度。</td>
+</tr>
+<tr>
+<td>SLANet_plus</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANet_plus_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_plus_pretrained.pdparams">训练模型</a></td>
+<td>63.69</td>
+<td>23.43 / 22.16</td>
+<td>- / 41.80</td>
+<td>6.9</td>
+<td>SLANet_plus 是百度飞桨视觉团队自研的表格结构识别模型SLANet的增强版。相较于SLANet，SLANet_plus 对无线表、复杂表格的识别能力得到了大幅提升，并降低了模型对表格定位准确性的敏感度，即使表格定位出现偏移，也能够较准确地进行识别。</td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary> <b>文本检测模块：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>检测Hmean（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv5_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_det_pretrained.pdparams">训练模型</a></td>
+<td>83.8</td>
+<td>89.55 / 70.19</td>
+<td>383.15 / 383.15</td>
+<td>84.3</td>
+<td>PP-OCRv5 的服务端文本检测模型，精度更高，适合在性能较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_det_pretrained.pdparams">训练模型</a></td>
+<td>79.0</td>
+<td>10.67 / 6.36</td>
+<td>57.77 / 28.15</td>
+<td>4.7</td>
+<td>PP-OCRv5 的移动端文本检测模型，效率更高，适合在端侧设备部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_det_pretrained.pdparams">训练模型</a></td>
+<td>69.2</td>
+<td>127.82 / 98.87</td>
+<td>585.95 / 489.77</td>
+<td>109</td>
+<td>PP-OCRv4 的服务端文本检测模型，精度更高，适合在性能较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams">训练模型</a></td>
+<td>63.8</td>
+<td>9.87 / 4.17</td>
+<td>56.60 / 20.79</td>
+<td>4.7</td>
+<td>PP-OCRv4 的移动端文本检测模型，效率更高，适合在端侧设备部署</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>文本识别模块：</b></summary>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>86.38</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec 是新一代文本识别模型。该模型致力于以单一模型高效、精准地支持简体中文、繁体中文、英文、日文四种主要语言，以及手写、竖版、拼音、生僻字等复杂文本场景的识别。在保持识别效果的同时，兼顾推理速度和模型鲁棒性，为各种场景下的文档理解提供高效、精准的技术支撑。</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.29</td>
+<td> 1.46/5.43 </td>
+<td> 5.32/91.79 </td>
+<td>16</td>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">训练模型</a></td>
+<td>86.58</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>181</td>
+<td>PP-OCRv4_server_rec_doc是在PP-OCRv4_server_rec的基础上，在更多中文文档数据和PP-OCR训练数据的混合数据训练而成，增加了部分繁体字、日文、特殊字符的识别能力，可支持识别的字符为1.5万+，除文档相关的文字识别能力提升外，也同时提升了通用文字的识别能力</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>PP-OCRv4的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>PP-OCRv4的服务器端模型，推理精度高，可以部署在多种不同的服务器上</td>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>基于PP-OCRv4识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+</table>
+
+> ❗ 以上列出的是文本识别模块重点支持的<b>6个核心模型</b>，该模块总共支持<b>20个全量模型</b>，包含多个多语言文本识别模型，完整的模型列表如下：
+
+<details><summary> 👉模型列表详情</summary>
+
+* <b>PP-OCRv5 多场景模型</b>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>中文识别 Avg Accuracy(%)</th>
+<th>英文识别 Avg Accuracy(%)</th>
+<th>繁体中文识别 Avg Accuracy(%)</th>
+<th>日文识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>86.38</td>
+<td>64.70</td>
+<td>93.29</td>
+<td>60.35</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_server_rec 是新一代文本识别模型。该模型致力于以单一模型高效、精准地支持简体中文、繁体中文、英文、日文四种主要语言，以及手写、竖版、拼音、生僻字等复杂文本场景的识别。在保持识别效果的同时，兼顾推理速度和模型鲁棒性，为各种场景下的文档理解提供高效、精准的技术支撑。</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.29</td>
+<td>66.00</td>
+<td>83.55</td>
+<td>54.65</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+</table>
+
+* <b>中文识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">训练模型</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc是在PP-OCRv4_server_rec的基础上，在更多中文文档数据和PP-OCR训练数据的混合数据训练而成，增加了部分繁体字、日文、特殊字符的识别能力，可支持识别的字符为1.5万+，除文档相关的文字识别能力提升外，也同时提升了通用文字的识别能力</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>PP-OCRv4的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>PP-OCRv4的服务器端模型，推理精度高，可以部署在多种不同的服务器上</td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>72.96</td>
+<td>3.89 / 1.16</td>
+<td>8.72 / 3.56</td>
+<td>10.3</td>
+<td>PP-OCRv3的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>ch_SVTRv2_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_SVTRv2_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_SVTRv2_rec_pretrained.pdparams">训练模型</a></td>
+<td>68.81</td>
+<td>10.38 / 8.31</td>
+<td>66.52 / 30.83</td>
+<td>80.5</td>
+<td rowspan="1">
+SVTRv2 是一种由复旦大学视觉与学习实验室（FVL）的OpenOCR团队研发的服务端文本识别模型，其在PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务中荣获一等奖，A榜端到端识别精度相比PP-OCRv4提升6%。
+</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>ch_RepSVTR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_RepSVTR_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_RepSVTR_rec_pretrained.pdparams">训练模型</a></td>
+<td>65.07</td>
+<td>6.29 / 1.57</td>
+<td>20.64 / 5.40</td>
+<td>48.8</td>
+<td rowspan="1">    RepSVTR 文本识别模型是一种基于SVTRv2 的移动端文本识别模型，其在PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务中荣获一等奖，B榜端到端识别精度相比PP-OCRv4提升2.5%，推理速度持平。</td>
+</tr>
+</table>
+
+* <b>英文识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td> 70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>基于PP-OCRv4识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+<tr>
+<td>en_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>70.69</td>
+<td>3.56 / 0.78</td>
+<td>8.44 / 5.78</td>
+<td>17.3</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+</table>
+
+
+* <b>多语言识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>korean_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>60.21</td>
+<td>3.73 / 0.98</td>
+<td>8.76 / 2.91</td>
+<td>9.6</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量韩文识别模型，支持韩文、数字识别</td>
+</tr>
+<tr>
+<td>japan_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+japan_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/japan_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>45.69</td>
+<td>3.86 / 1.01</td>
+<td>8.62 / 2.92</td>
+<td>9.8</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量日文识别模型，支持日文、数字识别</td>
+</tr>
+<tr>
+<td>chinese_cht_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+chinese_cht_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/chinese_cht_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>82.06</td>
+<td>3.90 / 1.16</td>
+<td>9.24 / 3.18</td>
+<td>10.8</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量繁体中文识别模型，支持繁体中文、数字识别</td>
+</tr>
+<tr>
+<td>te_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+te_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>95.88</td>
+<td>3.59 / 0.81</td>
+<td>8.28 / 6.21</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量泰卢固文识别模型，支持泰卢固文、数字识别</td>
+</tr>
+<tr>
+<td>ka_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ka_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ka_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>96.96</td>
+<td>3.49 / 0.89</td>
+<td>8.63 / 2.77</td>
+<td>17.4</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量卡纳达文识别模型，支持卡纳达文、数字识别</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ta_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>76.83</td>
+<td>3.49 / 0.86</td>
+<td>8.35 / 3.41</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量泰米尔文识别模型，支持泰米尔文、数字识别</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>76.93</td>
+<td>3.53 / 0.78</td>
+<td>8.50 / 6.83</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量拉丁文识别模型，支持拉丁文、数字识别</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+arabic_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>73.55</td>
+<td>3.60 / 0.83</td>
+<td>8.44 / 4.69</td>
+<td>17.3</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量阿拉伯字母识别模型，支持阿拉伯字母、数字识别</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+cyrillic_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>94.28</td>
+<td>3.56 / 0.79</td>
+<td>8.22 / 2.76</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量斯拉夫字母识别模型，支持斯拉夫字母、数字识别</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>96.44</td>
+<td>3.60 / 0.78</td>
+<td>6.95 / 2.87</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量梵文字母识别模型，支持梵文字母、数字识别</td>
+</tr>
+</table>
+</details>
+</details>
+
+<details>
+<summary> <b>文本行方向分类模块（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>模型下载链接</th>
+<th>Top-1 Acc（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x0_25_textline_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x0_25_textline_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x0_25_textline_ori_pretrained.pdparams">训练模型</a></td>
+<td>98.85</td>
+<td>2.16 / 0.41</td>
+<td>2.37 / 0.73</td>
+<td>0.96</td>
+<td>基于PP-LCNet_x0_25的文本行分类模型，含有两个类别，即0度，180度</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary> <b>公式识别模块（可选）：</b></summary>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>En-BLEU(%)</th>
+<th>Zh-BLEU(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>UniMERNet</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UniMERNet_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UniMERNet_pretrained.pdparams">训练模型</a></td>
+<td>85.91</td>
+<td>43.50</td>
+<td>1311.84 / 1311.84</td>
+<td>- / 8288.07</td>
+<td>1530</td>
+<td>UniMERNet是由上海AI Lab研发的一款公式识别模型。该模型采用Donut Swin作为编码器，MBartDecoder作为解码器，并通过在包含简单公式、复杂公式、扫描捕捉公式和手写公式在内的一百万数据集上进行训练，大幅提升了模型对真实场景公式的识别准确率</td>
+</tr>
+<tr>
+<td>PP-FormulaNet-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-S_pretrained.pdparams">训练模型</a></td>
+<td>87.00</td>
+<td>45.71</td>
+<td>182.25 / 182.25</td>
+<td>- / 254.39</td>
+<td>224</td>
+<td rowspan="2">PP-FormulaNet 是由百度飞桨视觉团队开发的一款先进的公式识别模型，支持5万个常见LateX源码词汇的识别。PP-FormulaNet-S 版本采用了 PP-HGNetV2-B4 作为其骨干网络，通过并行掩码和模型蒸馏等技术，大幅提升了模型的推理速度，同时保持了较高的识别精度，适用于简单印刷公式、跨行简单印刷公式等场景。而 PP-FormulaNet-L 版本则基于 Vary_VIT_B 作为骨干网络，并在大规模公式数据集上进行了深入训练，在复杂公式的识别方面，相较于PP-FormulaNet-S表现出显著的提升，适用于简单印刷公式、复杂印刷公式、手写公式等场景。 </td>
+
+</tr>
+<td>PP-FormulaNet-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-L_pretrained.pdparams">训练模型</a></td>
+<td>90.36</td>
+<td>45.78</td>
+<td>1482.03 / 1482.03</td>
+<td>- / 3131.54</td>
+<td>695</td>
+</tr>
+<td>PP-FormulaNet_plus-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-S_pretrained.pdparams">训练模型</a></td>
+<td>88.71</td>
+<td>53.32</td>
+<td>179.20 / 179.20</td>
+<td>- / 260.99</td>
+<td>248</td>
+<td rowspan="3">PP-FormulaNet_plus 是百度飞桨视觉团队在 PP-FormulaNet 的基础上开发的增强版公式识别模型。与原版相比，PP-FormulaNet_plus 在训练中使用了更为丰富的公式数据集，包括中文学位论文、专业书籍、教材试卷以及数学期刊等多种来源。这一扩展显著提升了模型的识别能力。
+
+其中，PP-FormulaNet_plus-M 和 PP-FormulaNet_plus-L 模型新增了对中文公式的支持，并将公式的最大预测 token 数从 1024 扩大至 2560，大幅提升了对复杂公式的识别性能。同时，PP-FormulaNet_plus-S 模型则专注于增强英文公式的识别能力。通过这些改进，PP-FormulaNet_plus 系列模型在处理复杂多样的公式识别任务时表现更加出色。 </td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-M_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-M_pretrained.pdparams">训练模型</a></td>
+<td>91.45</td>
+<td>89.76</td>
+<td>1040.27 / 1040.27</td>
+<td>- / 1615.80</td>
+<td>592</td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-L_pretrained.pdparams">训练模型</a></td>
+<td>92.22</td>
+<td>90.64</td>
+<td>1476.07 / 1476.07</td>
+<td>- / 3125.58</td>
+<td>698</td>
+</tr>
+<tr>
+<td>LaTeX_OCR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/LaTeX_OCR_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/LaTeX_OCR_rec_pretrained.pdparams">训练模型</a></td>
+<td>74.55</td>
+<td>39.96</td>
+<td>1088.89 / 1088.89</td>
+<td>- / -</td>
+<td>99</td>
+<td>LaTeX-OCR是一种基于自回归大模型的公式识别算法，通过采用 Hybrid ViT 作为骨干网络，transformer作为解码器，显著提升了公式识别的准确性。</td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary> <b>印章文本检测模块（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>检测Hmean（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv4_server_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_seal_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_seal_det_pretrained.pdparams">训练模型</a></td>
+<td>98.40</td>
+<td>124.64 / 91.57</td>
+<td>545.68 / 439.86</td>
+<td>109</td>
+<td>PP-OCRv4的服务端印章文本检测模型，精度更高，适合在较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_seal_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_seal_det_pretrained.pdparams">训练模型</a></td>
+<td>96.36</td>
+<td>9.70 / 3.56</td>
+<td>50.38 / 19.64</td>
+<td>4.7</td>
+<td>PP-OCRv4的移动端印章文本检测模型，效率更高，适合在端侧部署</td>
+</tr>
+</tbody>
+</table>
+</details>
+</details>
+
+<details>
+<summary> <b>测试环境说明：</b></summary>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+                       <li><strong>测试数据集：
+             </strong>
+                <ul>
+                  <li>文档图像方向分类模型：PaddleOCR 自建的数据集，覆盖证件和文档等多个场景，包含 1000 张图片。</li>
+                  <li>文本图像矫正模型：<a href="https://www3.cs.stonybrook.edu/~cvl/docunet.html">DocUNet</a>。 </li>
+                  <li>版面区域检测模型：PaddleOCR 自建的版面区域分析数据集，包含中英文论文、杂志和研报等常见的 1w 张文档类型图片。 </li>
+                  <li> 表格结构识别模型：内部自建的英文表格识别数据集。</li>
+                  <li> 文本检测模型：PaddleOCR 自建的中文数据集，覆盖街景、网图、文档、手写多个场景，其中检测包含 500 张图片。</li>
+                  <li>中文识别模型： PaddleOCR 自建的中文数据集，覆盖街景、网图、文档、手写多个场景，其中文本识别包含 1.1w 张图片。 </li>
+                  <li>ch_SVTRv2_rec：<a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务</a>A榜评估集。</li>
+                  <li>ch_RepSVTR_rec：<a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务</a>B榜评估集。</li>
+                  <li>英文识别模型：PaddleOCR 自建的英文数据集。</li>
+                  <li>多语言识别模型：PaddleOCR 自建的多语种数据集。</li>
+                  <li>文本行方向分类模型：PaddleOCR 自建的数据集，覆盖证件和文档等多个场景，包含 1000 张图片。</li>
+                  <li>印章文本检测模型：PaddleOCR 自建的数据集，包含500张圆形印章图像。</li>
+                </ul>
+             </li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td>本地飞桨推理引擎（默认情况下根据默认模型名称选择合适引擎；若静态图和动态图均可用则优先 <code>paddle_static</code>）</td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+</details>
+
+<br />
+<b>如您更考虑模型精度，请选择精度较高的模型，如您更考虑模型推理速度，请选择推理速度较快的模型，如您更考虑模型存储大小，请选择存储大小较小的模型</b>
+
+## 2. 快速开始
+
+在本地使用 PP-ChatOCRv4 产线前，请确保您已经按照[安装教程](../installation.md)完成了wheel包安装。如果您希望选择性安装依赖，请参考安装教程中的相关说明。该产线对应的依赖分组为 `ie`。
+
+**请注意，如果在执行过程中遇到程序失去响应、程序异常退出、内存资源耗尽、推理速度极慢等问题，请尝试参考文档调整配置，例如关闭不需要使用的功能或使用更轻量的模型。**
+
+在进行模型推理之前，首先需要准备大语言模型的 api_key，PP-ChatOCRv4 支持在[百度云千帆平台](https://console.bce.baidu.com/qianfan/ais/console/onlineService)或者本地部署的标准 OpenAI 接口大模型服务。如果使用百度云千帆平台，可以参考[认证鉴权](https://cloud.baidu.com/doc/qianfan-api/s/ym9chdsy5) 获取 api_key。如果使用本地部署的大模型服务，可以参考[PaddleNLP大模型部署文档](https://github.com/PaddlePaddle/PaddleNLP/tree/develop/llm)进行大模型部署对话接口部署和向量化接口部署，并填写对应的 base_url 和 api_key 即可。如果需要使用多模态大模型进行数据融合，可以参考[PaddleMIX模型文档](https://github.com/PaddlePaddle/PaddleMIX/tree/develop/paddlemix/examples/ppdocbee2)中的OpenAI服务部署进行多模态大模型部署，并填写对应的 base_url 和 api_key 即可。
+
+**注:** 如果因本地环境限制无法在本地部署多模态大模型，可以将代码中的含有“mllm”变量的行注释掉，仅使用大语言模型完成信息抽取。
+
+### 2.1 命令行方式体验
+
+可以下载 [测试文件](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png)，使用一行命令即可快速体验产线效果：
+
+```bash
+paddleocr pp_chatocrv4_doc -i vehicle_certificate-1.png -k 驾驶室准乘人数 --qianfan_api_key your_api_key
+
+# 通过 --invoke_mllm 和 --pp_docbee_base_url 使用多模态大模型
+paddleocr pp_chatocrv4_doc -i vehicle_certificate-1.png -k 驾驶室准乘人数 --qianfan_api_key your_api_key --invoke_mllm True --pp_docbee_base_url http://127.0.0.1:8080/
+```
+
+上述命令默认使用本地飞桨推理引擎。默认情况下，各模块会根据默认模型名称选择合适的本地飞桨推理引擎：仅支持动态图的模型使用 `paddle_dynamic`；同时支持静态图和动态图的模型优先使用 `paddle_static`。如需运行，请先参考[飞桨框架安装说明](../paddlepaddle_installation.md)安装 PaddlePaddle。
+
+如果选择 `transformers` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 Transformers 环境配置，然后执行如下命令：
+
+```bash
+# 使用 transformers 引擎进行推理
+paddleocr pp_chatocrv4_doc -i vehicle_certificate-1.png -k 驾驶室准乘人数 --qianfan_api_key your_api_key \
+    --engine transformers
+```
+
+<!-- Luxorion-12 修改：表格参数说明 -->
+<details><summary><b>命令行支持更多参数设置，点击展开以查看命令行参数的详细说明</b></summary>
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，必填。<br/>
+<b>说明：</b>如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<br/>
+<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png">示例</a>；<br/>
+<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>keys</code></td>
+<td><b>含义：</b>用于信息提取的键。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td><b>含义：</b>指定推理结果文件保存的路径。<br/>
+<b>说明：</b>如果不设置，推理结果将不会保存到本地。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>invoke_mllm</code></td>
+<td><b>含义：</b>是否加载并使用多模态大模型。<br/>
+<b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>含义：</b>用于版面区域检测的模型名称。<br/>
+<b>说明：</b>如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>含义：</b>版面区域检测模型的目录路径。<br/>
+<b>说明：</b>如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>含义：</b>文档方向分类模型的名称。<br/>
+<b>说明：</b>如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>文档方向分类模型的目录路径。<br/>
+<b>说明：</b>如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>含义：</b>文本图像矫正模型的名称。<br/>
+<b>说明：</b>如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。<br/>
+<b>说明：</b>如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>含义：</b>文本检测模型的名称。<br/>
+<b>说明：</b>如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>含义：</b>文本检测模型的目录路径。<br/>
+<b>说明：</b>如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>含义：</b>文本识别模型的名称。<br/>
+<b>说明：</b>如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>含义：</b>文本识别模型的目录路径。<br/>
+<b>说明：</b>如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>含义：</b>文本识别模型的batch size。<br/>
+<b>说明：</b>如果不设置，将默认设置batch size为<code>1</code>。</td>
+<td><code>int</code></td>
+<td><code>1</code></td>
+</tr>
+<tr>
+<td><code>table_structure_recognition_model_name</code></td>
+<td><b>含义：</b>表格结构识别模型的名称。<br/>
+<b>说明：</b>如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_structure_recognition_model_dir</code></td>
+<td><b>含义：</b>表格结构识别模型的目录路径。<br/>
+<b>说明：</b>如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_name</code></td>
+<td><b>含义：</b>印章文本检测模型的名称。<br/>
+<b>说明：</b>如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_dir</code></td>
+<td><b>含义：</b>印章文本检测模型的目录路径。<br/>
+<b>说明：</b>如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_name</code></td>
+<td><b>含义：</b>印章文本识别模型的名称。<br/>
+<b>说明：</b>如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_dir</code></td>
+<td><b>含义：</b>印章文本识别模型的目录路径。<br/>
+<b>说明：</b>如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_batch_size</code></td>
+<td><b>含义：</b>印章文本识别模型的batch size。<br/>
+<b>说明：</b>如果不设置，将默认设置batch size为<code>1</code>。</td>
+<td><code>int</code></td>
+<td><code>1</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否加载并使用文档方向分类模块。<br/>
+<b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。<br/>
+<b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>含义：</b>是否加载并使用文本行方向分类模块。<br/>
+<b>说明：</b>如果不设置，初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>含义：</b>是否加载并使用印章文本识别子产线。<br/>
+<b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>含义：</b>是否加载并使用表格识别子产线。<br/>
+<b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>版面模型得分阈值。<br/>
+<b>说明：</b><code>0-1</code> 之间的任意浮点数。<br/>
+如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>0.5</code>。
+</td>
+<td><code>float</code></td>
+<td><code>0.5</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>版面检测是否使用后处理NMS。<br/>
+<b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>版面区域检测模型检测框的扩张系数。<br/>
+<b>说明：</b>任意大于 <code>0</code>  浮点数。<br/>
+如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>1.0</code>。
+</td>
+<td><code>float</code></td>
+<td><code>1.0</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>版面检测中模型输出的检测框的合并处理模式。<br/>
+<b>说明：</b>
+<ul>
+<li><b>large</b>，设置为large时，表示在模型输出的检测框中，对于互相重叠包含的检测框，只保留外部最大的框，删除重叠的内部框；</li>
+<li><b>small</b>，设置为small，表示在模型输出的检测框中，对于互相重叠包含的检测框，只保留内部被包含的小框，删除重叠的外部框；</li>
+<li><b>union</b>，不进行框的过滤处理，内外框都保留；</li>
+</ul>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>large</code>。
+</td>
+<td><code>str</code></td>
+<td><code>large</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>含义：</b>文本检测的图像边长限制。大于 <code>0</code> 的任意整数。<br/>
+<b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>960</code>。
+</td>
+<td><code>int</code></td>
+<td><code>960</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>含义：</b>文本检测的边长度限制类型。
+支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code></li>
+</ul>如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>max</code>。
+</td>
+<td><code>str</code></td>
+<td><code>max</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>含义：</b>检测像素阈值。输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。<br/>
+<b>说明：</b>大于 <code>0</code> 的任意浮点数。<br/>
+如果不设置，将默认使用产线初始化的该参数值 <code>0.3</code>。
+</td>
+<td><code>float</code></td>
+<td><code>0.3</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>含义：</b>检测框阈值，检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。<br/>
+<b>说明：</b>大于 <code>0</code> 的任意浮点数。<br/>
+如果不设置，将默认使用产线初始化的该参数值 <code>0.6</code>。
+</td>
+<td><code>float</code></td>
+<td><code>0.6</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>含义：</b>文本检测扩张系数，使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。<br/>
+<b>说明：</b>大于 <code>0</code> 的任意浮点数。<br/>
+如果不设置，将默认使用产线初始化的该参数值 <code>2.0</code>。
+</td>
+<td><code>float</code></td>
+<td><code>2.0</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>含义：</b>文本识别阈值，得分大于该阈值的文本结果会被保留。<br/>
+<b>说明：</b>大于 <code>0</code> 的任意浮点数。<br/>
+如果不设置，将默认使用产线初始化的该参数值 <code>0.0</code>。即不设阈值。
+</td>
+<td><code>float</code></td>
+<td><code>0.0</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>含义：</b>印章文本检测的图像边长限制。<br/>
+<b>说明：</b>大于 <code>0</code> 的任意整数。<br/>
+如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>736</code>。
+</td>
+<td><code>int</code></td>
+<td><code>736</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>含义：</b>印章文本检测的图像边长限制类型。<br/>
+<b>说明：</b>
+支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>。<br/>
+如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>min</code>。
+</td>
+<td><code>str</code></td>
+<td><code>min</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>含义：</b>印章文本检测的检测像素阈值。<br/>
+<b>说明：</b>输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。<br/>
+大于 <code>0</code> 的任意浮点数。<br/>
+如果不设置，将默认使用产线初始化的该参数值 <code>0.2</code>。
+</td>
+<td><code>float</code></td>
+<td><code>0.2</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>含义：</b>印章文本检测的检测框阈值。<br/>
+<b>说明：</b>检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。<br/>
+大于 <code>0</code> 的任意浮点数。<br/>
+如果不设置，将默认使用产线初始化的该参数值 <code>0.6</code>。
+</td>
+<td><code>float</code></td>
+<td><code>0.6</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>含义：</b>印章文本检测扩张系数，使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。<br/>
+<b>说明：</b>大于 <code>0</code> 的任意浮点数。<br/>
+如果不设置，将使用产线初始化的该参数值，默认为 <code>0.5</code>。
+</td>
+<td><code>float</code></td>
+<td><code>0.5</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>含义：</b>印章文本识别阈值，得分大于该阈值的文本结果会被保留。<br/>
+<b>说明：</b>大于 <code>0</code> 的任意浮点数。<br/>
+如果不设置，将使用产线初始化的该参数值，默认为<code>0.0</code>，即不设阈值。
+</td>
+<td><code>float</code></td>
+<td><code>0.0</code></td>
+</tr>
+<tr>
+<td><code>qianfan_api_key</code></td>
+<td><b>含义：</b>千帆平台的API key。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>pp_docbee_base_url</code></td>
+<td><b>含义：</b>多模态大模型服务的URL。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+</ul>如果不设置，将默认使用产线初始化的该参数值，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>fp32</code>、<code>fp16</code>。</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+
+</tbody>
+</table>
+</details>
+<br />
+
+运行结果会被打印到终端上，运行结果如下：
+
+```bash
+驾驶室准乘人数 2
+```
+
+### 2.2 Python脚本方式集成
+
+命令行方式是为了快速体验查看效果，一般来说，在项目中，往往需要通过代码集成，可以下载 [测试文件](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png)，使用如下示例代码进行推理：
+
+```python
+from paddleocr import PPChatOCRv4Doc
+
+chat_bot_config = {
+    "module_name": "chat_bot",
+    "model_name": "ernie-3.5-8k",
+    "base_url": "https://qianfan.baidubce.com/v2",
+    "api_type": "openai",
+    "api_key": "api_key",  # your api_key
+}
+
+retriever_config = {
+    "module_name": "retriever",
+    "model_name": "embedding-v1",
+    "base_url": "https://qianfan.baidubce.com/v2",
+    "api_type": "qianfan",
+    "api_key": "api_key",  # your api_key
+}
+
+mllm_chat_bot_config = {
+    "module_name": "chat_bot",
+    "model_name": "PP-DocBee2",
+    "base_url": "http://127.0.0.1:8080/",  # your local mllm service url
+    "api_type": "openai",
+    "api_key": "api_key",  # your api_key
+}
+
+pipeline = PPChatOCRv4Doc()
+
+visual_predict_res = pipeline.visual_predict(
+    input="vehicle_certificate-1.png",
+    use_doc_orientation_classify=False,
+    use_doc_unwarping=False,
+    use_common_ocr=True,
+    use_seal_recognition=True,
+    use_table_recognition=True,
+)
+
+visual_info_list = []
+for res in visual_predict_res:
+    visual_info_list.append(res["visual_info"])
+    layout_parsing_result = res["layout_parsing_result"]
+
+vector_info = pipeline.build_vector(
+    visual_info_list, flag_save_bytes_vector=True, retriever_config=retriever_config
+)
+mllm_predict_res = pipeline.mllm_pred(
+    input="vehicle_certificate-1.png",
+    key_list=["驾驶室准乘人数"],
+    mllm_chat_bot_config=mllm_chat_bot_config,
+)
+mllm_predict_info = mllm_predict_res["mllm_res"]
+chat_result = pipeline.chat(
+    key_list=["驾驶室准乘人数"],
+    visual_info=visual_info_list,
+    vector_info=vector_info,
+    mllm_predict_info=mllm_predict_info,
+    chat_bot_config=chat_bot_config,
+    retriever_config=retriever_config,
+)
+print(chat_result)
+
+```
+
+上述代码默认使用本地飞桨推理引擎。默认情况下，各模块会根据默认模型名称选择合适的本地飞桨推理引擎：仅支持动态图的模型使用 `paddle_dynamic`；同时支持静态图和动态图的模型优先使用 `paddle_static`。如需运行，请先参考[飞桨框架安装说明](../paddlepaddle_installation.md)安装 PaddlePaddle。
+
+如果选择 `transformers` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 Transformers 环境配置，然后执行如下代码：
+
+```python
+from paddleocr import PPChatOCRv4Doc
+
+chat_bot_config = {
+    "module_name": "chat_bot",
+    "model_name": "ernie-3.5-8k",
+    "base_url": "https://qianfan.baidubce.com/v2",
+    "api_type": "openai",
+    "api_key": "api_key",  # your api_key
+}
+
+retriever_config = {
+    "module_name": "retriever",
+    "model_name": "embedding-v1",
+    "base_url": "https://qianfan.baidubce.com/v2",
+    "api_type": "qianfan",
+    "api_key": "api_key",  # your api_key
+}
+
+mllm_chat_bot_config = {
+    "module_name": "chat_bot",
+    "model_name": "PP-DocBee2",
+    "base_url": "http://127.0.0.1:8080/",  # your local mllm service url
+    "api_type": "openai",
+    "api_key": "api_key",  # your api_key
+}
+
+pipeline = PPChatOCRv4Doc(
+    engine="transformers",
+)
+
+visual_predict_res = pipeline.visual_predict(
+    input="vehicle_certificate-1.png",
+    use_doc_orientation_classify=False,
+    use_doc_unwarping=False,
+    use_common_ocr=True,
+    use_seal_recognition=True,
+    use_table_recognition=True,
+)
+
+visual_info_list = []
+for res in visual_predict_res:
+    visual_info_list.append(res["visual_info"])
+    layout_parsing_result = res["layout_parsing_result"]
+
+vector_info = pipeline.build_vector(
+    visual_info_list, flag_save_bytes_vector=True, retriever_config=retriever_config
+)
+mllm_predict_res = pipeline.mllm_pred(
+    input="vehicle_certificate-1.png",
+    key_list=["驾驶室准乘人数"],
+    mllm_chat_bot_config=mllm_chat_bot_config,
+)
+mllm_predict_info = mllm_predict_res["mllm_res"]
+chat_result = pipeline.chat(
+    key_list=["驾驶室准乘人数"],
+    visual_info=visual_info_list,
+    vector_info=vector_info,
+    mllm_predict_info=mllm_predict_info,
+    chat_bot_config=chat_bot_config,
+    retriever_config=retriever_config,
+)
+print(chat_result)
+
+```
+
+运行后，输出结果如下：
+
+```
+{'chat_res': {'驾驶室准乘人数': '2'}}
+```
+<!--Luxorion-12 修改：表格参数说明-->
+PP-ChatOCRv4 预测的流程、API说明、产出说明如下：
+
+<details><summary>（1）调用 <code>PPChatOCRv4Doc</code> 实例化PP-ChatOCRv4产线对象。</summary>
+
+相关参数说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>含义：</b>用于版面区域检测的模型名称。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>含义：</b>版面区域检测模型的目录路径。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>含义：</b>文档方向分类模型的名称。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>文档方向分类模型的目录路径。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>含义：</b>文本图像矫正模型的名称。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>含义：</b>文本检测模型的名称。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>含义：</b>文本检测模型的目录路径。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>含义：</b>文本识别模型的名称。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>含义：</b>文本识别模型的目录路径。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>含义：</b>文本识别模型的batch size。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_structure_recognition_model_name</code></td>
+<td><b>含义：</b>表格结构识别模型的名称。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_structure_recognition_model_dir</code></td>
+<td><b>含义：</b>表格结构识别模型的目录路径。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_name</code></td>
+<td><b>含义：</b>印章文本检测模型的名称。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_dir</code></td>
+<td><b>含义：</b>印章文本检测模型的目录路径。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_name</code></td>
+<td><b>含义：</b>印章文本识别模型的名称。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_dir</code></td>
+<td><b>含义：</b>印章文本识别模型的目录路径。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_batch_size</code></td>
+<td><b>含义：</b>印章文本识别模型的batch size。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否加载并使用文档方向分类模块。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>含义：</b>是否加载并使用文本行方向分类模块。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>含义：</b>是否加载并使用印章文本识别子产线。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>含义：</b>是否加载并使用表格识别子产线。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>版面模型得分阈值。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：<code>0-1</code> 之间的任意浮点数；</li>
+<li><b>dict</b>： <code>{0:0.1}</code> key为类别ID，value为该类别的阈值；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>0.5</code>。</li>
+</ul>
+</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>版面检测是否使用后处理NMS。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>版面区域检测模型检测框的扩张系数。
+<b>说明：</b>
+<ul>
+<li><b>float</b>：任意大于 <code>0</code>  浮点数；</li>
+<li><b>Tuple[float,float]</b>：在横纵两个方向各自的扩张系数；</li>
+<li><b>dict</b>，dict的key为<b>int</b>类型，代表<code>cls_id</code>，value为<b>tuple</b>类型，如<code>{0: (1.1，2.0)}</code>，表示将模型输出的第0类别检测框中心不变，宽度扩张1.1倍，高度扩张2.0倍</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>1.0</code>。</li>
+</ul>
+</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>版面区域检测的重叠框过滤方式。
+<b>说明：</b>
+<ul>
+<li><b>str</b>：<code>large</code>，<code>small</code>，<code>union</code>，分别表示重叠框过滤时选择保留大框，小框还是同时保留；</li>
+<li><b>dict</b>，dict的key为<b>int</b>类型，代表<code>cls_id</code>，value为<b>str</b>类型，如<code>{0: "large"，2: "small"}</code>，表示对第0类别检测框使用large模式，对第2类别检测框使用small模式；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>large</code>。</li>
+</ul>
+</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>含义：</b>文本检测的图像边长限制。<br/>
+<b>说明：</b>
+<ul>
+<li><b>int</b>：大于 <code>0</code> 的任意整数；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>960</code>。</li>
+</ul>
+</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>含义：</b>文本检测的边长度限制类型。
+<b>说明：</b>
+<ul>
+<li><b>str</b>：支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>max</code>。</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>含义：</b>检测像素阈值，输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认为<code>0.3</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>含义：</b>检测框阈值，检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认为<code>0.6</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>含义：</b>文本检测扩张系数，使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认为<code>2.0</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>含义：</b>文本识别阈值，得分大于该阈值的文本结果会被保留。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认为<code>0.0</code>，即不设阈值。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>含义：</b>印章文本检测的图像边长限制。<br/>
+<b>说明：</b>
+<ul>
+<li><b>int</b>：大于<code>0</code>的任意整数；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>736</code>。</li>
+</ul>
+</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>含义：</b>印章文本检测的图像边长限制类型。<br/>
+<b>说明：</b>
+<ul>
+<li><b>str</b>：支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>min</code>。</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>含义：</b>检测像素阈值，输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认为<code>0.2</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>含义：</b>检测框阈值，检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认为<code>0.6</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>含义：</b>印章文本检测扩张系数，使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认为<code>0.5</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>含义：</b>印章文本识别阈值，得分大于该阈值的文本结果会被保留。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认为<code>0.0</code>，即不设阈值。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>retriever_config</code></td>
+<td><b>含义：</b>向量检索大模型配置参数。<br/>
+<b>说明:</b>
+配置内容为如下dict：
+<pre><code>{
+"module_name": "retriever",
+"model_name": "embedding-v1",
+"base_url": "https://qianfan.baidubce.com/v2",
+"api_type": "qianfan",
+"api_key": "api_key"  # 请将此设置为实际的API密钥
+}</code></pre>
+</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>mllm_chat_bot_config</code></td>
+<td><b>含义：</b>多模态大模型配置参数。<br/>
+<b>说明：</b>
+配置内容为如下dict：
+<pre><code>{
+"module_name": "chat_bot",
+"model_name": "PP-DocBee",
+"base_url": "http://127.0.0.1:8080/", # 请将此设置为实际的多模态大模型服务的url
+"api_type": "openai",
+"api_key": "api_key"  # 请将此设置为实际的API密钥
+}</code></pre>
+</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>chat_bot_config</code></td>
+<td><b>含义：</b>大语言模型配置信息。<br/>
+<b>说明：</b>
+配置内容为如下dict：
+<pre><code>{
+"module_name": "chat_bot",
+"model_name": "ernie-3.5-8k",
+"base_url": "https://qianfan.baidubce.com/v2",
+"api_type": "openai",
+"api_key": "api_key"  # 请将此设置为实际的API密钥
+}</code></pre>
+</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+<li><b>None</b>：如果设置为<code>None</code>，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details><summary>（2）调用 PP-ChatOCRv4 产线对象的 <code>visual_predict()</code> 方法获取视觉预测结果，该方法会返回一个结果列表。另外，产线还提供了 <code>visual_predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>visual_predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。以下是 <code>visual_predict()</code> 方法的参数及其说明：</summary>
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+  <li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据；</li>
+  <li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href = "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)；</li>
+  <li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]</code>。</li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否在推理时使用文档方向分类模块。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否在推理时使用文本图像矫正模块。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>含义：</b>是否加载并使用文本行方向分类模块。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>含义：</b>是否在推理时使用印章文本识别子产线。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>含义：</b>是否在推理时使用表格识别子产线。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明:</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+</table>
+</details>
+
+<details><summary>（3）对视觉预测结果进行处理。</summary>
+
+每个样本的预测结果均为 `dict` 类型，包含 `visual_info` 和 `layout_parsing_result` 两个字段。通过 `visual_info` 得到视觉信息（包含 `normal_text_dict`、`table_text_list`、`table_html_list` 等信息），并将每个样本的信息放到 `visual_info_list` 列表中，该列表的内容会在之后送入大语言模型中。
+
+当然，您也可以通过 `layout_parsing_result` 获取版面解析的结果，该结果包含文件或图片中包含的表格、文字、图片等内容，且支持打印、保存为图片、保存为`json`文件的操作:
+
+```python
+......
+for res in visual_predict_res:
+    visual_info_list.append(res["visual_info"])
+    layout_parsing_result = res["layout_parsing_result"]
+    layout_parsing_result.print()
+    layout_parsing_result.save_to_img("./output")
+    layout_parsing_result.save_to_json("./output")
+    layout_parsing_result.save_to_xlsx("./output")
+    layout_parsing_result.save_to_html("./output")
+......
+```
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化。</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效。</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效。</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致。</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效。</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效。</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将中间各个模块的可视化图像保存在png格式的图像。</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径。</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>save_to_html()</code></td>
+<td>将文件中的表格保存为html格式的文件。</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径。</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>save_to_xlsx()</code></td>
+<td>将文件中的表格保存为xlsx格式的文件。</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径。</td>
+<td>无</td>
+</tr>
+</table>
+
+<ul>
+   <li>调用<code>print()</code> 方法会将结果打印到终端，打印到终端的内容解释如下：</li>
+      <ol start="1" type = "1"> 
+        <li><code>input_path</code>: <code>(str)</code>  待预测图像或者PDF的输入路径</li>
+        <li><code>page_index</code>: <code>(Union[int, None])</code> 如果输入是PDF文件，则表示当前是PDF的第几页，否则为  <code>None</code></li>
+        <li><code>model_settings</code>: <code>(Dict[str, bool])</code>  配置产线所需的模型参数</li>
+            <ol start = "1" type = "a" > 
+                <li><code>use_doc_preprocessor</code>: <code>(bool)</code>  控制是否启用文档预处理子产线</li>
+                <li><code>use_seal_recognition</code>: <code>(bool)</code> 控制是否启用印章文本识别子产线</li>
+                <li><code>use_table_recognition</code>: <code>(bool)</code> 控制是否启用表格识别子产线</li>
+                <li><code>use_formula_recognition</code>: <code>(bool)</code> 控制是否启用公式识别子产线</li>
+            </ol>
+        </li>
+        <li><code>parsing_res_list</code>: <code>(List[Dict])</code> 解析结果的列表，每个元素为一个字典，列表顺序为解析后的阅读顺序。</li>
+            <ol start = "1" type = "a" > 
+                <li><code>block_bbox</code>: <code>(np.ndarray)</code> 版面区域的边界框。</li>
+                <li><code>block_label</code>: <code>(str)</code> 版面区域的标签，例如<code>text</code>, <code>table</code>等。</li>
+                <li><code>block_content</code>: <code>(str)</code> 内容为版面区域内的内容。</li>
+            </ol>
+            <li><code>overall_ocr_res</code>: <code>(Dict[str, Union[List[str], List[float], numpy.ndarray]])</code> 全局 OCR 结果的dict</li>
+                <ol start = "1" type = "a" > 
+                    <li><code>input_path</code>: <code>(Union[str, None])</code> 图像OCR子产线接受的图像路径，当输入为<code>numpy.ndarray</code>时，保存为<code> None</code> </li>
+                    <li><code>model_settings</code>: <code>(Dict)</code> OCR子产线的模型配置参数</li>
+                    <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> 文本检测的多边形框列表。每个检测框由4个顶点坐标构成的numpy数组表示，数组shape为(4, 2)，数据类型为int16</li>
+                    <li><code>dt_scores</code>: <code>(List[float])</code> 文本检测框的置信度列表</li>
+                    <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> 文本检测模块的配置参数</li>
+                        <ul>
+                            <li><code>limit_side_len</code>: <code>(int)</code> 图像预处理时的边长限制值</li>
+                            <li><code>limit_type</code>: <code>(str)</code> 边长限制的处理方式</li>
+                            <li><code>thresh</code>: <code>(float)</code> 文本像素分类的置信度阈值</li>
+                            <li><code>box_thresh</code>: <code>(float)</code> 文本检测框的置信度阈值</li>
+                            <li><code>unclip_ratio</code>: <code>(float)</code> 文本检测框的膨胀系数</li>
+                            <li><code>text_type</code>: <code>(str)</code> 文本检测的类型，当前固定为"general"</li>
+                        </ul>
+                    <li><code>text_type</code>: <code>(str)</code> 文本检测的类型，当前固定为"general"</li>
+                    <li><code>textline_orientation_angles</code>: <code>(List[int])</code> 文本行方向分类的预测结果。启用时返回实际角度值（如[0,0,1]</li>
+                    <li><code>text_rec_score_thresh</code>: <code>(float)</code> 文本识别结果的过滤阈值</li>
+                    <li><code>rec_texts</code>: <code>(List[str])</code> 文本识别结果列表，仅包含置信度超过<code>text_rec_score_thresh</code>的文本</li>
+                    <li><code>rec_scores</code>: <code>(List[float])</code> 文本识别的置信度列表，已按<code>text_rec_score_thresh</code>过滤</li>
+                    <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> 经过置信度过滤的文本检测框列表，格式同<code>dt_polys</code></li>
+                </ol>
+            <li><code>formula_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> 公式识别结果列表，每个元素为一个dict</li>
+                <ol start = "1" type = "a" > 
+                    <li><code>rec_formula</code>: <code>(str)</code> 公式识别结果</li>
+                    <li><code>rec_polys</code>: <code>(numpy.ndarray)</code> 公式检测框，shape为(4, 2)，dtype为int16</li>
+                    <li><code>formula_region_id</code>: <code>(int)</code> 公式所在的区域编号</li>
+                </ol>
+            <li><code>seal_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> 印章文本识别结果列表，每个元素为一个dict</li>
+                <ol start = "1" type = "a" > 
+                    <li><code>input_path</code>: <code>(str)</code> 印章图像的输入路径</li>
+                    <li><code>model_settings</code>: <code>(Dict)</code> 印章文本识别子产线的模型配置参数</li>
+                    <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> 印章检测框列表，格式同<code>dt_polys</code></li>
+                    <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> 印章检测模块的配置参数, 具体参数含义同上</li>
+                    <li><code>text_type</code>: <code>(str)</code> 印章检测的类型，当前固定为"seal"</li>
+                    <li><code>text_rec_score_thresh</code>: <code>(float)</code> 印章文本识别结果的过滤阈值</li>
+                    <li><code>rec_texts</code>: <code>(List[str])</code> 印章文本识别结果列表，仅包含置信度超过<code>text_rec_score_thresh</code>的文本</li>
+                    <li><code>rec_scores</code>: <code>(List[float])</code> 印章文本识别的置信度列表，已按<code>dt_polys</code>过滤</li>
+                    <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> 经过置信度过滤的印章检测框列表，格式同<code>dt_polys</code></li>
+                    <li><code>rec_boxes</code>: <code>(numpy.ndarray)</code> 检测框的矩形边界框数组，shape为(n, 4)，dtype为int16。每一行表示一个矩形</li>
+                </ol>
+            <li><code>table_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> 表格识别结果列表，每个元素为一个dict</li>
+                <ol start = "1" type = "a" > 
+                    <li><code>cell_box_list</code>: <code>(List[numpy.ndarray])</code> 表格单元格的边界框列表</li>
+                    <li><code>pred_html</code>: <code>(str)</code> 表格的HTML格式字符串</li>
+                    <li><code>table_ocr_pred</code>: <code>(dict)</code> 表格的OCR识别结果</li>
+                    <ul>
+                        <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> 单元格的检测框列表</li>
+                        <li><code>rec_texts</code>: <code>(List[str])</code> 单元格的识别结果</li>
+                        <li><code>rec_scores</code>: <code>(List[float])</code>单元格的识别置信度</li>
+                        <li><code>rec_boxes</code>: <code>(numpy.ndarray)</code> 检测框的矩形边界框数组，shape为(n, 4)，dtype为int16。每一行表示一个矩形</li>
+                    </ul>
+                </ol>
+            </ol>
+    </li>
+    <li>调用<code>save_to_json()</code> 方法会将上述内容保存到指定的 <code>save_path</code> 中，如果指定为目录，则保存的路径为<code>save_path/{your_img_basename}_res.json</code>，如果指定为文件，则直接保存到该文件中。由于 json 文件不支持保存numpy数组，因此会将其中的 <code>numpy.array</code> 类型转换为列表形式。</li>
+    <li>调用<code>save_to_img()</code> 方法会将可视化结果保存到指定的 <code>save_path</code> 中，如果指定为目录，则会将版面区域检测可视化图像、全局OCR可视化图像、版面阅读顺序可视化图像等内容保存，如果指定为文件，则直接保存到该文件中。(产线通常包含较多结果图片，不建议直接指定为具体的文件路径，否则多张图会被覆盖，仅保留最后一张图)</li>
+    <li>调用<code>save_to_markdown()</code> 方法会将转化后的 Markdown 文件保存到指定的 <code>save_path</code> 中，保存的文件路径为<code>save_path/{your_img_basename}.md</code>，如果输入是 PDF 文件，建议直接指定目录，否责多个 markdown 文件会被覆盖。
+    调用 <code>concatenate_markdown_pages()</code> 方法将 <code>PP-StructureV3 pipeline</code> 输出的多页Markdown内容<code>markdown_list</code>合并为单个完整文档，并返回合并后的Markdown内容。</li>
+</ul>
+此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">获取预测的 <code>json</code> 格式的结果。</td>
+</tr>
+<tr>
+<td rowspan = "2"><code>img</code></td>
+<td rowspan = "2">获取格式为 <code>dict</code> 的可视化图像。</td>
+</tr>
+</table>
+<ul>
+    <li><code>json</code> 属性获取的预测结果为dict类型的数据，相关内容与调用 <code>save_to_json()</code> 方法保存的内容一致。</li>
+    <li><code>img</code> 属性返回的预测结果是一个dict类型的数据。其中，键分别为 <code>layout_det_res</code>、<code>overall_ocr_res</code>、<code>text_paragraphs_ocr_res</code>、<code>formula_res_region1</code>、<code>table_cell_img</code> 和 <code>seal_res_region1</code>，对应的值是 <code>Image.Image</code> 对象：分别用于显示版面区域检测、OCR、OCR文本段落、公式、表格和印章结果的可视化图像。如果没有使用可选模块，则dict中只包含 <code>layout_det_res</code>。</li>
+</ul>
+</details>
+
+<details><summary>（4）调用PP-ChatOCRv4的产线对象的 <code>build_vector()</code> 方法，对文本内容进行向量构建。</summary>
+
+以下是 <code>build_vector()</code> 方法的参数及其说明：
+
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>visual_info</code></td>
+<td><b>含义：</b>视觉信息。<br/>
+<b>说明:</b>
+可以是包含视觉信息的dict，或者由这些dict组成的列表。</td>
+<td><code>list|dict</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>min_characters</code></td>
+<td><b>含义：</b>最小字符数量。<br/>
+<b>说明：</b>
+为大于0的正整数，可以根据大语言模型支持的token长度来决定。</td>
+<td><code>int</code></td>
+<td><code>3500</code></td>
+</tr>
+<tr>
+<td><code>block_size</code></td>
+<td><b>含义：</b>长文本建立向量库时分块大小。<br/>
+<b>说明：</b>为大于0的正整数，可以根据大语言模型支持的token长度来决定。</td>
+<td><code>int</code></td>
+<td><code>300</code></td>
+</tr>
+<tr>
+<td><code>flag_save_bytes_vector</code></td>
+<td><b>含义：</b>文字是否保存为二进制文件。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>retriever_config</code></td>
+<td><b>含义：</b>向量检索大模型配置参数，参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</table>
+该方法会返回一个包含视觉文本信息的dict，dict的内容如下：
+<ul>
+    <li><code>flag_save_bytes_vector</code>：<code>(bool)</code>是否将结果保存为二进制文件</li>
+    <li><code>flag_too_short_text</code>：<code>(bool)</code>是否文本长度小于最小字符数量</li>
+    <li><code>vector</code>: <code>(str|list)</code> 文本的二进制内容或者文本内容，取决于<code>flag_save_bytes_vector</code>和<code>min_characters</code>的值，如果<code>flag_save_bytes_vector=True</code>且文本长度大于等于最小字符数量，则返回二进制内容；否则返回原始的文本。</li>
+</ul>
+</details>
+
+<details><summary>（5）调用PP-ChatOCRv4的产线对象的 <code>mllm_pred()</code> 方法，获取多模态大模型抽取结果。</summary>
+
+以下是 <code>mllm_pred()</code> 方法的参数及其说明：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。
+<b>说明：</b>
+<ul>
+  <li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据；</li>
+  <li><b>str</b>：如图像文件或者单页PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或单页PDF文件的网络URL：<a href = "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png">示例</a>。</li>
+</ul>
+</td>
+<td><code>Python Var|str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>key_list</code></td>
+<td><b>含义：</b>用于提取信息的单个键或键列表。</td>
+<td><code>Union[str, List[str]]</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>mllm_chat_bot_config</code></td>
+<td><b>含义：</b>多模态大模型配置参数，参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+</details>
+
+<details><summary>（6）调用PP-ChatOCRv4的产线对象的 <code>chat()</code> 方法，对关键信息进行抽取。</summary>
+
+以下是 <code>chat()</code>方法的参数及其说明：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>key_list</code></td>
+<td><b>含义：</b>用于提取信息的单个键或键列表。</td>
+<td><code>Union[str, List[str]]</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>visual_info</code></td>
+<td><b>含义：</b>视觉信息结果。</td>
+<td><code>List[dict]</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_vector_retrieval</code></td>
+<td><b>含义：</b>是否使用向量检索。</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>vector_info</code></td>
+<td><b>含义：</b>用于检索的向量信息。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>min_characters</code></td>
+<td><b>含义：</b>所需的最小字符数。<br/>
+<b>说明：</b>为大于0的正整数。</td>
+<td><code>int</code></td>
+<td><code>3500</code></td>
+</tr>
+<tr>
+<td><code>text_task_description</code></td>
+<td><b>含义：</b>文本任务的描述。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_output_format</code></td>
+<td><b>含义：</b>文本结果的输出格式。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rules_str</code></td>
+<td><b>含义：</b>生成文本结果的规则。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_few_shot_demo_text_content</code></td>
+<td><b>含义：</b>用于少样本演示的文本内容。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_few_shot_demo_key_value_list</code></td>
+<td><b>含义：</b>用于少样本演示的键值列表。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_task_description</code></td>
+<td><b>含义：</b>表任务的描述。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_output_format</code></td>
+<td><b>含义：</b>表结果的输出格式。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_rules_str</code></td>
+<td><b>含义：</b>生成表结果的规则。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_few_shot_demo_text_content</code></td>
+<td><b>含义：</b>表少样本演示的文本内容。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_few_shot_demo_key_value_list</code></td>
+<td><b>含义：</b>表少样本演示的键值列表。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>mllm_predict_info</code></td>
+<td><b>含义：</b>多模态大模型结果。</td>
+<td><code>dict|None</code></td>
+<td>
+<code>None</code>
+</td>
+</tr>
+<td><code>mllm_integration_strategy</code></td>
+<td><b>含义：</b>多模态大模型和大语言模型数据融合策略，支持单独使用其中一个或者融合两者结果。<br/>
+<b>说明：</b>可选：<code>integration</code>, <code>llm_only</code> and <code>mllm_only</code>。</td>
+<td><code>str</code></td>
+<td><code>integration</code></td>
+</tr>
+<tr>
+<td><code>chat_bot_config</code></td>
+<td><b>含义：</b>大语言模型配置信息，参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>retriever_config</code></td>
+<td><b>含义：</b>向量检索大模型配置参数，参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+该方法会将结果打印到终端，打印到终端的内容解释如下：
+<ul><li><code>chat_res</code>: <code>(dict)</code> 提取信息的结果，是一个dict，包含了待抽取的键和对应的值。</li></ul>
+
+</details>
+
+## 3. 开发集成/部署
+
+如果产线可以达到您对产线推理速度和精度的要求，您可以直接进行开发集成/部署。
+
+若您需要将产线直接应用在您的Python项目中，可以参考 [2.2 Python脚本方式](#22-python)中的示例代码。
+
+此外，PaddleOCR 也提供了其他两种部署方式，详细说明如下：
+
+🚀 高性能推理：在实际生产环境中，许多应用对部署策略的性能指标（尤其是响应速度）有着较严苛的标准，以确保系统的高效运行与用户体验的流畅性。为此，PaddleOCR 提供高性能推理功能，旨在对模型推理及前后处理进行深度性能优化，实现端到端流程的显著提速，详细的高性能推理流程请参考[高性能推理](../inference_deployment/local_inference/high_performance_inference.md)。
+
+☁️ 服务化部署：服务化部署是实际生产环境中常见的一种部署形式。通过将推理功能封装为服务，客户端可以通过网络请求来访问这些服务，以获取推理结果。详细的产线服务化部署流程请参考[服务化部署](../inference_deployment/serving/serving.md)。
+
+以下是基础服务化部署的API参考与多语言服务调用示例：
+
+<details><summary>API参考</summary>
+<p>对于服务提供的主要操作：</p>
+<ul>
+<li>HTTP请求方法为POST。</li>
+<li>请求体和响应体均为JSON数据（JSON对象）。</li>
+<li>当请求处理成功时，响应状态码为<code>200</code>，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。固定为<code>0</code>。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。固定为<code>"Success"</code>。</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>操作结果。</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>当请求处理未成功时，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。与响应状态码相同。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。</td>
+</tr>
+</tbody>
+</table>
+<p>服务提供的主要操作如下：</p>
+<ul>
+<li><b><code>analyzeImages</code></b></li>
+</ul>
+<p>使用计算机视觉模型对图像进行分析，获得OCR、表格识别结果等，并提取图像中的关键信息。</p>
+<p><code>POST /chatocr-visual</code></p>
+<ul>
+<li>请求体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>服务器可访问的图像文件（含 TIFF，多页时按页处理）或 PDF 文件的 URL，或上述类型文件内容的 Base64 编码结果。默认对于超过 10 页的 PDF 或多页 TIFF，仅处理前 10 页。<br /> 要解除页数限制，请在产线配置文件中添加以下配置：
+<pre><code>Serving:
+  extra:
+    max_num_input_imgs: null
+</code></pre>
+</td>
+<td>是</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>文件类型。<code>0</code> 表示 PDF 文件，<code>1</code> 表示图像文件（含 TIFF）。若请求体无此属性，则将根据URL推断文件类型。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_doc_orientation_classify</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_doc_unwarping</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useSealRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_seal_recognition</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useTableRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_table_recognition</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutThreshold</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>layout_threshold</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutNms</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>layout_nms</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutUnclipRatio</code></td>
+<td><code>number</code> | <code>array</code> | <code>object</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>layout_unclip_ratio</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutMergeBboxesMode</code></td>
+<td><code>string</code> | <code>object</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>layout_merge_bboxes_mode</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>text_det_limit_side_len</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>text_det_limit_type</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>text_det_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>text_det_box_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>text_det_unclip_ratio</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>text_rec_score_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>seal_det_limit_side_len</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>seal_det_limit_type</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>seal_det_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>seal_det_box_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>seal_det_unclip_ratio</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>seal_rec_score_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>是否返回可视化结果图以及处理过程中的中间图像等。
+<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>传入 <code>true</code>：返回图像。</li>
+<li>传入 <code>false</code>：不返回图像。</li>
+<li>若请求体中未提供该参数或传入 <code>null</code>：遵循产线配置文件<code>Serving.visualize</code> 的设置。</li>
+</ul>
+<br/>例如，在产线配置文件中添加如下字段：<br/>
+<pre><code>Serving:
+  visualize: False
+</code></pre>
+将默认不返回图像，通过请求体中的<code>visualize</code>参数可以覆盖默认行为。如果请求体和配置文件中均未设置（或请求体传入<code>null</code>、配置文件中未设置），则默认返回图像。
+</td>
+<td>否</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>请求处理成功时，响应体的<code>result</code>具有如下属性：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>layoutParsingResults</code></td>
+<td><code>array</code></td>
+<td>使用计算机视觉模型得到的分析结果。数组长度为1（对于图像输入）或实际处理的文档页数（对于PDF输入）。对于PDF输入，数组中的每个元素依次表示PDF文件中实际处理的每一页的结果。</td>
+</tr>
+<tr>
+<td><code>visualInfo</code></td>
+<td><code>array</code></td>
+<td>图像中的关键信息，可用作其他操作的输入。</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>输入数据信息。</td>
+</tr>
+</tbody>
+</table>
+<p>下表中涉及图像的字段（如 <code>outputImages</code>、<code>inputImage</code>）默认以 Base64 字符串内联返回；当服务端开启 URL 返回模式时，相应字段的值变为预签名 URL，字段类型保持不变。配置方式参见 <a href="../inference_deployment/serving/serving.md">服务化部署</a>「以 URL 形式返回二进制内容」一节。</p>
+<p><code>layoutParsingResults</code>中的每个元素为一个<code>object</code>，具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>产线对象的 <code>visual_predict</code> 方法生成的 <code>layout_parsing_result</code> 的 JSON 表示中 <code>res</code> 字段的简化版本，其中去除了 <code>input_path</code> 和 <code>page_index</code> 字段。</td>
+</tr>
+<tr>
+<td><code>outputImages</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>参见产线视觉预测结果的 <code>img</code> 属性说明。</td>
+</tr>
+<tr>
+<td><code>inputImage</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>输入图像。图像为JPEG格式，使用Base64编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li><b><code>buildVectorStore</code></b></li>
+</ul>
+<p>构建向量数据库。</p>
+<p><code>POST /chatocr-vector</code></p>
+<ul>
+<li>请求体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>visualInfo</code></td>
+<td><code>array</code></td>
+<td>图像中的关键信息。由<code>analyzeImages</code>操作提供。</td>
+<td>是</td>
+</tr>
+<tr>
+<td><code>minCharacters</code></td>
+<td><code>integer</code></td>
+<td>请参阅产线对象中 <code>build_vector</code> 方法的 <code>min_characters</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>blockSize</code></td>
+<td><code>integer</code></td>
+<td>请参阅产线对象中 <code>build_vector</code> 方法的 <code>block_size</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>retrieverConfig</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>build_vector</code> 方法的 <code>retriever_config</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>请求处理成功时，响应体的<code>result</code>具有如下属性：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>vectorInfo</code></td>
+<td><code>object</code></td>
+<td>向量数据库序列化结果，可用作其他操作的输入。</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li><b><code>invokeMLLM</code></b></li>
+</ul>
+<p>调用多模态大模型。</p>
+<p><code>POST /chatocr-mllm</code></p>
+<ul>
+<li>请求体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>image</code></td>
+<td><code>string</code></td>
+<td>服务器可访问的图像文件的URL或图像文件内容的Base64编码结果。</td>
+<td>是</td>
+</tr>
+<tr>
+<td><code>keyList</code></td>
+<td><code>array</code></td>
+<td>键列表。</td>
+<td>是</td>
+</tr>
+<tr>
+<td><code>mllmChatBotConfig</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>mllm_pred</code> 方法的 <code>mllm_chat_bot_config</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+</tbody>
+<table>
+<ul>
+<li>请求处理成功时，响应体的<code>result</code>具有如下属性：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>mllmPredictInfo</code></td>
+<td><code>object</code></td>
+<td>多模态大模型调用结果。</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li><b><code>chat</code></b></li>
+</ul>
+<p>与大语言模型交互，利用大语言模型提炼关键信息。</p>
+<p><code>POST /chatocr-chat</code></p>
+<ul>
+<li>请求体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>keyList</code></td>
+<td><code>array</code></td>
+<td>键列表。</td>
+<td>是</td>
+</tr>
+<tr>
+<td><code>visualInfo</code></td>
+<td><code>object</code></td>
+<td>图像中的关键信息。由<code>analyzeImages</code>操作提供。</td>
+<td>是</td>
+</tr>
+<tr>
+<td><code>useVectorRetrieval</code></td>
+<td><code>boolean</code></td>
+<td>请参阅产线对象中 <code>chat</code> 方法的 <code>use_vector_retrieval</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>vectorInfo</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>向量数据库序列化结果。由<code>buildVectorStore</code>操作提供。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>minCharacters</code></td>
+<td><code>integer</code></td>
+<td>请参阅产线对象中 <code>chat</code> 方法的 <code>min_characters</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textTaskDescription</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>chat</code> 方法的 <code>text_task_description</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textOutputFormat</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>chat</code> 方法的 <code>text_output_format</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textRulesStr</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>chat</code> 方法的 <code>text_rules_str</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textFewShotDemoTextContent</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>chat</code> 方法的 <code>text_few_shot_demo_text_content</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textFewShotDemoKeyValueList</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>chat</code> 方法的 <code>text_few_shot_demo_key_value_list</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>tableTaskDescription</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>chat</code> 方法的 <code>table_task_description</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>tableOutputFormat</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>chat</code> 方法的 <code>table_output_format</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>tableRulesStr</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>chat</code> 方法的 <code>table_rules_str</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>tableFewShotDemoTextContent</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td></td>
+<td>请参阅产线对象中 <code>chat</code> 方法的 <code>table_few_shot_demo_text_content</code> 参数相关说明。</td>
+</tr>
+<tr>
+<td><code>tableFewShotDemoKeyValueList</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>chat</code> 方法的 <code>table_few_shot_demo_key_value_list</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>mllmPredictInfo</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>多模态大模型调用结果。由<code>invokeMllm</code>操作提供。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>mllmIntegrationStrategy</code></td>
+<td><code>string</code></td>
+<td>请参阅产线对象中 <code>chat</code> 方法的 <code>mllm_integration_strategy</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>chatBotConfig</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>chat</code> 方法的 <code>chat_bot_config</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>retrieverConfig</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>chat</code> 方法的 <code>retriever_config</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>请求处理成功时，响应体的<code>result</code>具有如下属性：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>chatResult</code></td>
+<td><code>object</code></td>
+<td>关键信息抽取结果。</td>
+</tr>
+</tbody>
+</table>
+<li><b>注意：</b></li>
+在请求体中包含大模型调用的API key等敏感参数可能存在安全风险。如无必要，请在配置文件中设置这些参数，在请求时不传递。
+<br/><br/>
+</details>
+
+<details><summary>多语言调用服务示例</summary>
+
+<details>
+<summary>Python</summary>
+
+
+<pre><code class="language-python">
+# 此脚本只展示了图片的用例，其他文件类型的调用请查看API参考来调整
+
+import base64
+import pprint
+import sys
+import requests
+
+
+API_BASE_URL = "http://127.0.0.1:8080"
+
+image_path = "./demo.jpg"
+keys = ["姓名"]
+
+with open(image_path, "rb") as file:
+    image_bytes = file.read()
+    image_data = base64.b64encode(image_bytes).decode("ascii")
+
+payload = {
+    "file": image_data,
+    "fileType": 1,
+}
+
+resp_visual = requests.post(url=f"{API_BASE_URL}/chatocr-visual", json=payload)
+if resp_visual.status_code != 200:
+    print(
+        f"Request to chatocr-visual failed with status code {resp_visual.status_code}."
+    )
+    pprint.pp(resp_visual.json())
+    sys.exit(1)
+result_visual = resp_visual.json()["result"]
+
+for i, res in enumerate(result_visual["layoutParsingResults"]):
+    print(res["prunedResult"])
+    for img_name, img in res["outputImages"].items():
+        img_path = f"{img_name}_{i}.jpg"
+        with open(img_path, "wb") as f:
+            f.write(base64.b64decode(img))
+        print(f"Output image saved at {img_path}")
+
+payload = {
+    "visualInfo": result_visual["visualInfo"],
+}
+resp_vector = requests.post(url=f"{API_BASE_URL}/chatocr-vector", json=payload)
+if resp_vector.status_code != 200:
+    print(
+        f"Request to chatocr-vector failed with status code {resp_vector.status_code}."
+    )
+    pprint.pp(resp_vector.json())
+    sys.exit(1)
+result_vector = resp_vector.json()["result"]
+
+payload = {
+    "image": image_data,
+    "keyList": keys,
+}
+resp_mllm = requests.post(url=f"{API_BASE_URL}/chatocr-mllm", json=payload)
+if resp_mllm.status_code != 200:
+    print(
+        f"Request to chatocr-mllm failed with status code {resp_mllm.status_code}."
+    )
+    pprint.pp(resp_mllm.json())
+    sys.exit(1)
+result_mllm = resp_mllm.json()["result"]
+
+payload = {
+    "keyList": keys,
+    "visualInfo": result_visual["visualInfo"],
+    "useVectorRetrieval": True,
+    "vectorInfo": result_vector["vectorInfo"],
+    "mllmPredictInfo": result_mllm["mllmPredictInfo"],
+}
+resp_chat = requests.post(url=f"{API_BASE_URL}/chatocr-chat", json=payload)
+if resp_chat.status_code != 200:
+    print(
+        f"Request to chatocr-chat failed with status code {resp_chat.status_code}."
+    )
+    pprint.pp(resp_chat.json())
+    sys.exit(1)
+result_chat = resp_chat.json()["result"]
+print("Final result:")
+print(result_chat["chatResult"])
+</code></pre>
+</details>
+
+<details><summary>C++</summary>
+
+<pre><code class="language-cpp">#include &lt;iostream&gt;
+#include &lt;fstream&gt;
+#include &lt;vector&gt;
+#include &lt;string&gt;
+#include "cpp-httplib/httplib.h" // https://github.com/Huiyicc/cpp-httplib
+#include "nlohmann/json.hpp" // https://github.com/nlohmann/json
+#include "base64.hpp" // https://github.com/tobiaslocker/base64
+
+using json = nlohmann::json;
+
+std::string encode_image(const std::string& path) {
+    std::ifstream file(path, std::ios::binary | std::ios::ate);
+    if (!file) throw std::runtime_error("File open error.");
+    std::streamsize size = file.tellg();
+    file.seekg(0, std::ios::beg);
+    std::vector<char> buf(size);
+    file.read(buf.data(), size);
+    return base64::to_base64(std::string(buf.data(), buf.size()));
+}
+
+int main() {
+    httplib::Client client("localhost", 8080);
+    std::string imagePath = "./demo.jpg";
+    std::string imageData = encode_image(imagePath);
+    json keys = { "合格证编号" };
+
+    json payload_visual = { {"file", imageData}, {"fileType", 1} };
+    auto resp1 = client.Post("/chatocr-visual", payload_visual.dump(), "application/json");
+    if (!resp1 || resp1->status != 200) {
+        std::cerr << "chatocr-visual failed.\n"; return 1;
+    }
+    json result_visual = json::parse(resp1->body)["result"];
+
+    for (size_t i = 0; i < result_visual["layoutParsingResults"].size(); ++i) {
+        auto& res = result_visual["layoutParsingResults"][i];
+        std::cout << "prunedResult: " << res["prunedResult"].dump() << "\n";
+        if (res.contains("outputImages")) {
+            for (auto& [name, b64] : res["outputImages"].items()) {
+                std::string outPath = name + "_" + std::to_string(i) + ".jpg";
+                std::string decoded = base64::from_base64(b64.get<std::string>());
+                std::ofstream out(outPath, std::ios::binary);
+                out.write(decoded.data(), decoded.size());
+                out.close();
+                std::cout << "Saved: " << outPath << "\n";
+            }
+        }
+    }
+
+    json payload_vector = { {"visualInfo", result_visual["visualInfo"]} };
+    auto resp2 = client.Post("/chatocr-vector", payload_vector.dump(), "application/json");
+    if (!resp2 || resp2->status != 200) {
+        std::cerr << "chatocr-vector failed.\n"; return 1;
+    }
+    json result_vector = json::parse(resp2->body)["result"];
+
+    json payload_mllm = { {"image", imageData}, {"keyList", keys} };
+    auto resp3 = client.Post("/chatocr-mllm", payload_mllm.dump(), "application/json");
+    if (!resp3 || resp3->status != 200) {
+        std::cerr << "chatocr-mllm failed.\n"; return 1;
+    }
+    json result_mllm = json::parse(resp3->body)["result"];
+
+    json payload_chat = {
+        {"keyList", keys},
+        {"visualInfo", result_visual["visualInfo"]},
+        {"useVectorRetrieval", true},
+        {"vectorInfo", result_vector["vectorInfo"]},
+        {"mllmPredictInfo", result_mllm["mllmPredictInfo"]}
+    };
+    auto resp4 = client.Post("/chatocr-chat", payload_chat.dump(), "application/json");
+    if (!resp4 || resp4->status != 200) {
+        std::cerr << "chatocr-chat failed.\n"; return 1;
+    }
+
+    json result_chat = json::parse(resp4->body)["result"];
+    std::cout << "Final chat result: " << result_chat["chatResult"] << std::endl;
+
+    return 0;
+}
+</code></pre></details>
+
+<details><summary>Java</summary>
+
+<pre><code class="language-java">import okhttp3.*;
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.util.Base64;
+import java.util.Iterator;
+
+public class Main {
+    public static void main(String[] args) throws IOException {
+        String API_BASE_URL = "http://localhost:8080";
+        String imagePath = "./demo.jpg";
+        String[] keys = {"合格证编号"};
+
+        OkHttpClient client = new OkHttpClient();
+        ObjectMapper objectMapper = new ObjectMapper();
+        MediaType JSON = MediaType.parse("application/json; charset=utf-8");
+
+        byte[] imageBytes = java.nio.file.Files.readAllBytes(new File(imagePath).toPath());
+        String base64Image = Base64.getEncoder().encodeToString(imageBytes);
+
+        ObjectNode visualPayload = objectMapper.createObjectNode();
+        visualPayload.put("file", base64Image);
+        visualPayload.put("fileType", 1);
+
+        Request requestVisual = new Request.Builder()
+                .url(API_BASE_URL + "/chatocr-visual")
+                .post(RequestBody.create(JSON, visualPayload.toString()))
+                .build();
+
+        Response responseVisual = client.newCall(requestVisual).execute();
+        if (!responseVisual.isSuccessful()) {
+            System.err.println("chatocr-visual failed: " + responseVisual.code());
+            return;
+        }
+
+        JsonNode resultVisual = objectMapper.readTree(responseVisual.body().string()).get("result");
+
+        JsonNode layoutResults = resultVisual.get("layoutParsingResults");
+        for (int i = 0; i < layoutResults.size(); i++) {
+            JsonNode res = layoutResults.get(i);
+            System.out.println("prunedResult [" + i + "]: " + res.get("prunedResult").toString());
+
+            JsonNode outputImages = res.get("outputImages");
+            if (outputImages != null && outputImages.isObject()) {
+                Iterator<String> names = outputImages.fieldNames();
+                while (names.hasNext()) {
+                    String imgName = names.next();
+                    String imgBase64 = outputImages.get(imgName).asText();
+                    byte[] imgBytes = Base64.getDecoder().decode(imgBase64);
+                    String imgPath = imgName + "_" + i + ".jpg";
+                    try (FileOutputStream fos = new FileOutputStream(imgPath)) {
+                        fos.write(imgBytes);
+                        System.out.println("Saved image: " + imgPath);
+                    }
+                }
+            }
+        }
+
+        ObjectNode vectorPayload = objectMapper.createObjectNode();
+        vectorPayload.set("visualInfo", resultVisual.get("visualInfo"));
+
+        Request requestVector = new Request.Builder()
+                .url(API_BASE_URL + "/chatocr-vector")
+                .post(RequestBody.create(JSON, vectorPayload.toString()))
+                .build();
+
+        Response responseVector = client.newCall(requestVector).execute();
+        if (!responseVector.isSuccessful()) {
+            System.err.println("chatocr-vector failed: " + responseVector.code());
+            return;
+        }
+
+        JsonNode resultVector = objectMapper.readTree(responseVector.body().string()).get("result");
+
+        ObjectNode mllmPayload = objectMapper.createObjectNode();
+        mllmPayload.put("image", base64Image);
+        mllmPayload.putArray("keyList").add(keys[0]);
+
+        Request requestMllm = new Request.Builder()
+                .url(API_BASE_URL + "/chatocr-mllm")
+                .post(RequestBody.create(JSON, mllmPayload.toString()))
+                .build();
+
+        Response responseMllm = client.newCall(requestMllm).execute();
+        if (!responseMllm.isSuccessful()) {
+            System.err.println("chatocr-mllm failed: " + responseMllm.code());
+            return;
+        }
+
+        JsonNode resultMllm = objectMapper.readTree(responseMllm.body().string()).get("result");
+
+        ObjectNode chatPayload = objectMapper.createObjectNode();
+        chatPayload.putArray("keyList").add(keys[0]);
+        chatPayload.set("visualInfo", resultVisual.get("visualInfo"));
+        chatPayload.put("useVectorRetrieval", true);
+        chatPayload.set("vectorInfo", resultVector.get("vectorInfo"));
+        chatPayload.set("mllmPredictInfo", resultMllm.get("mllmPredictInfo"));
+
+        Request requestChat = new Request.Builder()
+                .url(API_BASE_URL + "/chatocr-chat")
+                .post(RequestBody.create(JSON, chatPayload.toString()))
+                .build();
+
+        Response responseChat = client.newCall(requestChat).execute();
+        if (!responseChat.isSuccessful()) {
+            System.err.println("chatocr-chat failed: " + responseChat.code());
+            return;
+        }
+
+        JsonNode resultChat = objectMapper.readTree(responseChat.body().string()).get("result");
+        System.out.println("Final result:");
+        System.out.println(resultChat.get("chatResult").toString());
+    }
+}
+</code></pre></details>
+
+<details><summary>Go</summary>
+
+<pre><code class="language-go">package main
+
+import (
+    "bytes"
+    "encoding/base64"
+    "encoding/json"
+    "fmt"
+    "io/ioutil"
+    "net/http"
+    "os"
+)
+
+func sendPostRequest(url string, payload map[string]interface{}) (map[string]interface{}, error) {
+    bodyBytes, err := json.Marshal(payload)
+    if err != nil {
+        return nil, fmt.Errorf("error marshaling payload: %v", err)
+    }
+
+    req, err := http.NewRequest("POST", url, bytes.NewBuffer(bodyBytes))
+    if err != nil {
+        return nil, fmt.Errorf("error creating request: %v", err)
+    }
+    req.Header.Set("Content-Type", "application/json")
+
+    client := &http.Client{}
+    resp, err := client.Do(req)
+    if err != nil {
+        return nil, fmt.Errorf("error sending request: %v", err)
+    }
+    defer resp.Body.Close()
+
+    if resp.StatusCode != http.StatusOK {
+        return nil, fmt.Errorf("status code error: %d", resp.StatusCode)
+    }
+
+    respBytes, err := ioutil.ReadAll(resp.Body)
+    if err != nil {
+        return nil, fmt.Errorf("error reading response: %v", err)
+    }
+
+    var result map[string]interface{}
+    if err := json.Unmarshal(respBytes, &result); err != nil {
+        return nil, fmt.Errorf("error unmarshaling response: %v", err)
+    }
+    return result["result"].(map[string]interface{}), nil
+}
+
+func main() {
+    apiBase := "http://localhost:8080"
+    imagePath := "./demo.jpg"
+    keys := []string{"合格证编号"}
+
+    imageBytes, err := ioutil.ReadFile(imagePath)
+    if err != nil {
+        fmt.Printf("read image failed : %v\n", err)
+        return
+    }
+    imageData := base64.StdEncoding.EncodeToString(imageBytes)
+
+    visualPayload := map[string]interface{}{
+        "file":     imageData,
+        "fileType": 1,
+    }
+    visualResult, err := sendPostRequest(apiBase+"/chatocr-visual", visualPayload)
+    if err != nil {
+        fmt.Printf("chatocr-visual request error: %v\n", err)
+        return
+    }
+
+    layoutResults := visualResult["layoutParsingResults"].([]interface{})
+    for i, res := range layoutResults {
+        layout := res.(map[string]interface{})
+        fmt.Println("PrunedResult:", layout["prunedResult"])
+        outputImages := layout["outputImages"].(map[string]interface{})
+        for name, img := range outputImages {
+            imgBytes, _ := base64.StdEncoding.DecodeString(img.(string))
+            filename := fmt.Sprintf("%s_%d.jpg", name, i)
+            if err := os.WriteFile(filename, imgBytes, 0644); err == nil {
+                fmt.Printf("save image：%s\n", filename)
+            }
+        }
+    }
+
+    vectorPayload := map[string]interface{}{
+        "visualInfo": visualResult["visualInfo"],
+    }
+    vectorResult, err := sendPostRequest(apiBase+"/chatocr-vector", vectorPayload)
+    if err != nil {
+        fmt.Printf("chatocr-vector request error: %v\n", err)
+        return
+    }
+
+    mllmPayload := map[string]interface{}{
+        "image":   imageData,
+        "keyList": keys,
+    }
+    mllmResult, err := sendPostRequest(apiBase+"/chatocr-mllm", mllmPayload)
+    if err != nil {
+        fmt.Printf("chatocr-mllm request error: %v\n", err)
+        return
+    }
+
+    chatPayload := map[string]interface{}{
+        "keyList":           keys,
+        "visualInfo":        visualResult["visualInfo"],
+        "useVectorRetrieval": true,
+        "vectorInfo":        vectorResult["vectorInfo"],
+        "mllmPredictInfo":   mllmResult["mllmPredictInfo"],
+    }
+    chatResult, err := sendPostRequest(apiBase+"/chatocr-chat", chatPayload)
+    if err != nil {
+        fmt.Printf("chatocr-chat request error: %v\n", err)
+        return
+    }
+
+    fmt.Println("final result：", chatResult["chatResult"])
+}
+</code></pre></details>
+
+<details><summary>C#</summary>
+
+<pre><code class="language-csharp">using System;
+using System.IO;
+using System.Net.Http;
+using System.Text;
+using System.Threading.Tasks;
+using Newtonsoft.Json.Linq;
+
+class Program
+{
+    static readonly string API_BASE_URL = "http://localhost:8080";
+    static readonly string inputFilePath = "./demo.jpg";
+    static readonly string[] keys = { "合格证编号" };
+
+    static async Task Main(string[] args)
+    {
+        var httpClient = new HttpClient();
+
+        byte[] imageBytes = File.ReadAllBytes(inputFilePath);
+        string imageData = Convert.ToBase64String(imageBytes);
+
+        var payloadVisual = new JObject
+        {
+            { "file", imageData },
+            { "fileType", 1 }
+        };
+
+        var respVisual = await httpClient.PostAsync($"{API_BASE_URL}/chatocr-visual",
+            new StringContent(payloadVisual.ToString(), Encoding.UTF8, "application/json"));
+
+        if (!respVisual.IsSuccessStatusCode)
+        {
+            Console.Error.WriteLine($"Request to chatocr-visual failed: {respVisual.StatusCode}");
+            Console.Error.WriteLine(await respVisual.Content.ReadAsStringAsync());
+            return;
+        }
+
+        JObject resultVisual = JObject.Parse(await respVisual.Content.ReadAsStringAsync())["result"] as JObject;
+
+        var layoutParsingResults = (JArray)resultVisual["layoutParsingResults"];
+        for (int i = 0; i < layoutParsingResults.Count; i++)
+        {
+            var res = layoutParsingResults[i];
+            Console.WriteLine($"[{i}] prunedResult:\n{res["prunedResult"]}");
+
+            JObject outputImages = res["outputImages"] as JObject;
+            if (outputImages != null)
+            {
+                foreach (var img in outputImages)
+                {
+                    string imgName = img.Key;
+                    string base64Img = img.Value?.ToString();
+                    if (!string.IsNullOrEmpty(base64Img))
+                    {
+                        string imgPath = $"{imgName}_{i}.jpg";
+                        File.WriteAllBytes(imgPath, Convert.FromBase64String(base64Img));
+                        Console.WriteLine($"Output image saved at {imgPath}");
+                    }
+                }
+            }
+        }
+
+        var payloadVector = new JObject
+        {
+            { "visualInfo", resultVisual["visualInfo"] }
+        };
+
+        var respVector = await httpClient.PostAsync($"{API_BASE_URL}/chatocr-vector",
+            new StringContent(payloadVector.ToString(), Encoding.UTF8, "application/json"));
+
+        if (!respVector.IsSuccessStatusCode)
+        {
+            Console.Error.WriteLine($"Request to chatocr-vector failed: {respVector.StatusCode}");
+            Console.Error.WriteLine(await respVector.Content.ReadAsStringAsync());
+            return;
+        }
+
+        JObject resultVector = JObject.Parse(await respVector.Content.ReadAsStringAsync())["result"] as JObject;
+
+        var payloadMllm = new JObject
+        {
+            { "image", imageData },
+            { "keyList", new JArray(keys) }
+        };
+
+        var respMllm = await httpClient.PostAsync($"{API_BASE_URL}/chatocr-mllm",
+            new StringContent(payloadMllm.ToString(), Encoding.UTF8, "application/json"));
+
+        if (!respMllm.IsSuccessStatusCode)
+        {
+            Console.Error.WriteLine($"Request to chatocr-mllm failed: {respMllm.StatusCode}");
+            Console.Error.WriteLine(await respMllm.Content.ReadAsStringAsync());
+            return;
+        }
+
+        JObject resultMllm = JObject.Parse(await respMllm.Content.ReadAsStringAsync())["result"] as JObject;
+
+        var payloadChat = new JObject
+        {
+            { "keyList", new JArray(keys) },
+            { "visualInfo", resultVisual["visualInfo"] },
+            { "useVectorRetrieval", true },
+            { "vectorInfo", resultVector["vectorInfo"] },
+            { "mllmPredictInfo", resultMllm["mllmPredictInfo"] }
+        };
+
+        var respChat = await httpClient.PostAsync($"{API_BASE_URL}/chatocr-chat",
+            new StringContent(payloadChat.ToString(), Encoding.UTF8, "application/json"));
+
+        if (!respChat.IsSuccessStatusCode)
+        {
+            Console.Error.WriteLine($"Request to chatocr-chat failed: {respChat.StatusCode}");
+            Console.Error.WriteLine(await respChat.Content.ReadAsStringAsync());
+            return;
+        }
+
+        JObject resultChat = JObject.Parse(await respChat.Content.ReadAsStringAsync())["result"] as JObject;
+        Console.WriteLine("Final result:");
+        Console.WriteLine(resultChat["chatResult"]);
+    }
+}
+</code></pre></details>
+
+<details><summary>Node.js</summary>
+
+<pre><code class="language-js">const axios = require('axios');
+const fs = require('fs');
+const path = require('path');
+
+const API_BASE_URL = 'http://localhost:8080';
+const imagePath = './demo.jpg';
+const keys = ['合格证编号'];
+
+function encodeImageToBase64(filePath) {
+  const bitmap = fs.readFileSync(filePath);
+  return Buffer.from(bitmap).toString('base64');
+}
+
+(async () => {
+  try {
+    const imageData = encodeImageToBase64(imagePath);
+
+    const respVisual = await axios.post(`${API_BASE_URL}/chatocr-visual`, {
+      file: imageData,
+      fileType: 1
+    });
+
+    const resultVisual = respVisual.data.result;
+    resultVisual.layoutParsingResults.forEach((res, i) => {
+      console.log(`\n[${i}] prunedResult:\n`, res.prunedResult);
+      const outputImages = res.outputImages || {};
+      for (const [imgName, base64Img] of Object.entries(outputImages)) {
+        const fileName = `${imgName}_${i}.jpg`;
+        fs.writeFileSync(fileName, Buffer.from(base64Img, 'base64'));
+        console.log(`Output image saved at ${fileName}`);
+      }
+    });
+
+    const respVector = await axios.post(`${API_BASE_URL}/chatocr-vector`, {
+      visualInfo: resultVisual.visualInfo
+    });
+    const resultVector = respVector.data.result;
+
+    const respMllm = await axios.post(`${API_BASE_URL}/chatocr-mllm`, {
+      image: imageData,
+      keyList: keys
+    });
+    const resultMllm = respMllm.data.result;
+
+    const respChat = await axios.post(`${API_BASE_URL}/chatocr-chat`, {
+      keyList: keys,
+      visualInfo: resultVisual.visualInfo,
+      useVectorRetrieval: true,
+      vectorInfo: resultVector.vectorInfo,
+      mllmPredictInfo: resultMllm.mllmPredictInfo
+    });
+
+    const resultChat = respChat.data.result;
+    console.log('\nFinal result:\n', resultChat.chatResult);
+
+  } catch (error) {
+    if (error.response) {
+      console.error(`❌ Request failed: ${error.response.status}`);
+      console.error(error.response.data);
+    } else {
+      console.error('❌ Error occurred:', error.message);
+    }
+  }
+})();
+</code></pre></details>
+
+<details><summary>PHP</summary>
+
+<pre><code class="language-php">&lt;?php
+
+$API_BASE_URL = "http://localhost:8080";
+$image_path = "./demo.jpg";
+$keys = ["合格证编号"];
+
+$image_data = base64_encode(file_get_contents($image_path));
+
+$payload_visual = [
+    "file" => $image_data,
+    "fileType" => 1
+];
+$response_visual_raw = send_post_raw("$API_BASE_URL/chatocr-visual", $payload_visual);
+$response_visual = json_decode($response_visual_raw, true);
+if (!isset($response_visual["result"])) {
+    echo "chatocr-visual request error\n";
+    print_r($response_visual);
+    exit(1);
+}
+$result_visual_raw = json_decode($response_visual_raw, false)->result;
+$result_visual_arr = $response_visual["result"];
+
+foreach ($result_visual_arr["layoutParsingResults"] as $i => $res) {
+    echo "[$i] prunedResult:\n";
+    print_r($res["prunedResult"]);
+    if (!empty($res["outputImages"])) {
+        foreach ($res["outputImages"] as $img_name => $base64_img) {
+            $img_path = "{$img_name}_{$i}.jpg";
+            file_put_contents($img_path, base64_decode($base64_img));
+            echo "Output image saved at $img_path\n";
+        }
+    }
+}
+
+$payload_vector = [
+    "visualInfo" => $result_visual_raw->visualInfo
+];
+$response_vector_raw = send_post_raw("$API_BASE_URL/chatocr-vector", $payload_vector);
+$response_vector = json_decode($response_vector_raw, true);
+if (!isset($response_vector["result"])) {
+    echo "chatocr-vector request error\n";
+    print_r($response_vector);
+    exit(1);
+}
+$result_vector_raw = json_decode($response_vector_raw, false)->result;
+
+$payload_mllm = [
+    "image" => $image_data,
+    "keyList" => $keys
+];
+$response_mllm_raw = send_post_raw("$API_BASE_URL/chatocr-mllm", $payload_mllm);
+$response_mllm = json_decode($response_mllm_raw, true);
+if (!isset($response_mllm["result"])) {
+    echo "chatocr-mllm request error\n";
+    print_r($response_mllm);
+    exit(1);
+}
+$result_mllm_raw = json_decode($response_mllm_raw, false)->result;
+
+$payload_chat = [
+    "keyList" => $keys,
+    "visualInfo" => $result_visual_raw->visualInfo,
+    "useVectorRetrieval" => true,
+    "vectorInfo" => $result_vector_raw->vectorInfo,
+    "mllmPredictInfo" => $result_mllm_raw->mllmPredictInfo
+];
+$response_chat_raw = send_post_raw("$API_BASE_URL/chatocr-chat", $payload_chat);
+$response_chat = json_decode($response_chat_raw, true);
+if (!isset($response_chat["result"])) {
+    echo "chatocr-chat request error\n";
+    print_r($response_chat);
+    exit(1);
+}
+
+echo "Final result:\n";
+echo json_encode($response_chat["result"]["chatResult"], JSON_UNESCAPED_UNICODE | JSON_PRETTY_PRINT) . "\n";
+
+
+function send_post_raw($url, $data) {
+    $json_str = json_encode($data, JSON_UNESCAPED_UNICODE);
+    $ch = curl_init($url);
+    curl_setopt($ch, CURLOPT_POST, true);
+    curl_setopt($ch, CURLOPT_POSTFIELDS, $json_str);
+    curl_setopt($ch, CURLOPT_HTTPHEADER, ['Content-Type: application/json']);
+    curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+    $response = curl_exec($ch);
+    if ($response === false) {
+        echo "cURL error: " . curl_error($ch) . "\n";
+    }
+    curl_close($ch);
+    return $response;
+}
+?&gt;
+</code></pre></details>
+</details>
+<br/>
+
+## 4. 二次开发
+如果 PP-ChatOCRv4 产线提供的默认模型权重在您的场景中，精度或速度不满意，您可以尝试利用<b>您自己拥有的特定领域或应用场景的数据</b>对现有模型进行进一步的<b>微调</b>，以提升在您的场景中的识别效果。
+
+### 4.1 模型微调
+由于 PP-ChatOCRv4 产线包含若干模块，模型产线的效果如果不及预期，可能来自于其中任何一个模块。您可以对提取效果差的 case 进行分析，通过可视化图像，确定是哪个模块存在问题，并参考以下表格中对应的微调教程链接进行模型微调。
+
+
+<table>
+  <thead>
+    <tr>
+      <th>情形</th>
+      <th>微调模块</th>
+      <th>微调参考链接</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td>版面区域检测不准，如印章、表格未检出等</td>
+      <td>版面区域检测模块</td>
+      <td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/layout_detection.html#_5">链接</a></td>
+    </tr>
+    <tr>
+      <td>表格结构识别不准</td>
+      <td>表格结构识别</td>
+      <td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/table_structure_recognition.html#_5">链接</a></td>
+    </tr>
+    <tr>
+      <td>印章文本存在漏检</td>
+      <td>印章文本检测模块</td>
+      <td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/seal_text_detection.html#_5">链接</a></td>
+    </tr>
+    <tr>
+      <td>文本存在漏检</td>
+      <td>文本检测模块</td>
+      <td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/text_detection.html#_5">链接</a></td>
+    </tr>
+    <tr>
+      <td>文本内容都不准</td>
+      <td>文本识别模块</td>
+      <td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/text_recognition.html#_5">链接</a></td>
+    </tr>
+    <tr>
+      <td>垂直或者旋转文本行矫正不准</td>
+      <td>文本行方向分类模块</td>
+      <td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/textline_orientation_classification.html#_5">链接</a></td>
+    </tr>
+    <tr>
+      <td>整图旋转矫正不准</td>
+      <td>文档图像方向分类模块</td>
+      <td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#_5">链接</a></td>
+    </tr>
+    <tr>
+      <td>图像扭曲矫正不准</td>
+      <td>文本图像矫正模块</td>
+      <td>暂不支持微调</td>
+    </tr>
+  </tbody>
+</table>
+
+
+### 4.2 模型应用
+当您使用私有数据集完成微调训练后，可获得本地模型权重文件，然后可以通过自定义产线配置文件的方式，使用微调后的模型权重。
+
+1. 获取产线配置文件
+
+可调用 PaddleOCR 中 PPChatOCRv4 产线对象的 `export_paddlex_config_to_yaml` 方法，将当前产线配置导出为 YAML 文件：
+
+```Python
+from paddleocr import PPChatOCRv4Doc
+
+pipeline = PPChatOCRv4Doc()
+pipeline.export_paddlex_config_to_yaml("PP-ChatOCRv4-doc.yaml")
+```
+
+2. 修改配置文件
+
+在得到默认的产线配置文件后，将微调后模型权重的本地路径替换至产线配置文件中的对应位置即可。例如
+
+```yaml
+......
+SubModules:
+    TextDetection:
+    module_name: text_detection
+    model_name: PP-OCRv5_server_det
+    model_dir: null # 替换为微调后的文本检测模型权重路径
+    limit_side_len: 960
+    limit_type: max
+    thresh: 0.3
+    box_thresh: 0.6
+    unclip_ratio: 1.5
+
+    TextRecognition:
+    module_name: text_recognition
+    model_name: PP-OCRv5_server_rec
+    model_dir: null # 替换为微调后的文本检测模型权重路径
+    batch_size: 1
+            score_thresh: 0
+......
+```
+
+在产线配置文件中，不仅包含 PaddleOCR CLI 和 Python API 支持的参数，还可进行更多高级配置，具体信息可在 [PaddleX模型产线使用概览](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/pipeline_develop_guide.html) 中找到对应的产线使用教程，参考其中的详细说明，根据需求调整各项配置。
+
+3. 在 CLI 中加载产线配置文件
+
+在修改完成配置文件后，通过命令行的 `--paddlex_config` 参数指定修改后的产线配置文件的路径，PaddleOCR 会读取其中的内容作为产线配置。示例如下：
+
+```bash
+paddleocr pp_chatocrv4_doc --paddlex_config PP-ChatOCRv4-doc.yaml ...
+```
+
+4. 在 Python API 中加载产线配置文件
+
+初始化产线对象时，可通过 `paddlex_config` 参数传入 PaddleX 产线配置文件路径或配置dict，PaddleOCR 会读取其中的内容作为产线配置。示例如下：
+
+```python
+from paddleocr import PPChatOCRv4Doc
+
+pipeline = PPChatOCRv4Doc(paddlex_config="PP-ChatOCRv4-doc.yaml")
+```
diff --git a/docs/version3.x/pipeline_usage/PP-DocTranslation.en.md b/docs/version3.x/pipeline_usage/PP-DocTranslation.en.md
new file mode 100644
index 0000000..c17bf04
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PP-DocTranslation.en.md
@@ -0,0 +1,3438 @@
+---
+comments: true
+---
+
+# PP-DocTranslation Pipeline Usage Tutorial
+
+## 1. Introduction to PP-DocTranslation Pipeline
+
+PP-DocTranslation is a document intelligent translation solution provided by PaddlePaddle. It integrates advanced general layout analysis technology and large language model (LLM) capabilities to offer you efficient document intelligent translation services. This solution can accurately identify and extract various elements within documents, including text blocks, headings, paragraphs, images, tables, and other complex layout structures, and on this basis, achieve high-quality multilingual translation. PP-DocTranslation supports mutual translation among multiple mainstream languages, particularly excelling in handling documents with complex layouts and strong contextual dependencies, striving to deliver precise, natural, fluent, and professional translation results. This pipeline also provides flexible serving options, supporting the use of multiple programming languages on various hardware. Moreover, it offers the capability for secondary development, allowing you to train and fine-tune models on your own datasets based on this pipeline, and the trained models can also be seamlessly integrated.
+
+<b>The PP-DocTranslation pipeline uses the PP-StructureV3 sub-pipeline, and thus has all the functions of the PP-StructureV3 pipeline. For more information on the functions and usage details of the PP-StructureV3 pipeline, you can click on the [PP-StructureV3 Pipeline Documentation](./PP-StructureV3.en.md) page.</b>
+
+In this pipeline, you can select the model to use based on the benchmark data below.
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing.
+> In the inference time columns labeled [Regular Mode / High-Performance Mode], the `Regular Mode` values correspond to local Paddle inference engines. Each module selects the appropriate local Paddle inference engine according to the default model name: models that support only dynamic graph use `paddle_dynamic`, while models that support both static and dynamic graph prefer `paddle_static`.
+
+<details><summary> 👉Model List Details</summary>
+<p><b>Document Image Orientation Classification Module:</b></p>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Top-1 Acc (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard / High Performance]</th>
+<th>CPU Inference Time (ms)<br/>[Standard / High Performance]</th>
+<th>Model Size (M)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">Pretrained Model</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>A document image classification model based on PP-LCNet_x1_0 with four classes: 0°, 90°, 180°, and 270°</td>
+</tr>
+</tbody>
+</table>
+
+<p><b>Text Image Unwarping Module:</b></p>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>CER</th>
+<th>Model Size (M)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">Pretrained Model</a></td>
+<td>0.179</td>
+<td>30.3</td>
+<td>High-accuracy text image unwarping model</td>
+</tr>
+</tbody>
+</table>
+
+<p><b>Layout Detection Module Models:</b></p>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard / High Performance]</th>
+<th>CPU Inference Time (ms)<br/>[Standard / High Performance]</th>
+<th>Model Size (M)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout_plus-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout_plus-L_pretrained.pdparams">Pretrained Model</a></td>
+<td>83.2</td>
+<td>53.03 / 17.23</td>
+<td>634.62 / 378.32</td>
+<td>126.01</td>
+<td>High-accuracy layout detection model based on RT-DETR-L, trained on a custom dataset covering scenarios like Chinese/English papers, multi-column magazines, newspapers, PPTs, contracts, books, exams, research reports, ancient books, Japanese documents, and vertical text documents</td>
+</tr>
+<tr>
+<td>PP-DocLayout-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-L_pretrained.pdparams">Pretrained Model</a></td>
+<td>90.4</td>
+<td>33.59 / 33.59</td>
+<td>503.01 / 251.08</td>
+<td>123.76</td>
+<td>High-accuracy layout detection model based on RT-DETR-L, trained on a custom dataset covering papers, magazines, contracts, books, exams, and research reports</td>
+</tr>
+<tr>
+<td>PP-DocLayout-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-M_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-M_pretrained.pdparams">Pretrained Model</a></td>
+<td>75.2</td>
+<td>13.03 / 4.72</td>
+<td>43.39 / 24.44</td>
+<td>22.578</td>
+<td>Balanced accuracy-efficiency layout detection model based on PicoDet-L, trained on a custom dataset covering papers, magazines, contracts, books, exams, and research reports</td>
+</tr>
+<tr>
+<td>PP-DocLayout-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-S_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-S_pretrained.pdparams">Pretrained Model</a></td>
+<td>70.9</td>
+<td>11.54 / 3.86</td>
+<td>18.53 / 6.29</td>
+<td>4.834</td>
+<td>High-efficiency layout detection model based on PicoDet-S, trained on a custom dataset for papers, magazines, contracts, books, exams, and research reports</td>
+</tr>
+</tbody>
+</table>
+
+<p><b>Table Structure Recognition Module:</b></p>
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard / High Performance]</th>
+<th>CPU Inference Time (ms)<br/>[Standard / High Performance]</th>
+<th>Model Size (M)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>SLANeXt_wired</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wired_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wired_pretrained.pdparams">Pretrained Model</a></td>
+<td rowspan="2">69.65</td>
+<td rowspan="2">85.92 / 85.92</td>
+<td rowspan="2">- / 501.66</td>
+<td rowspan="2">351M</td>
+<td rowspan="2">SLANeXt series is a next-generation table structure recognition model developed by Baidu PaddlePaddle Vision Team. Compared with SLANet and SLANet_plus, SLANeXt focuses on recognizing table structures, with dedicated weights for wired and wireless tables, significantly improving performance especially for wired tables.</td>
+</tr>
+<tr>
+<td>SLANeXt_wireless</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wireless_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wireless_pretrained.pdparams">Pretrained Model</a></td>
+</tr>
+</table>
+
+<p><b>Table Classification Module Models:</b></p>
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Top-1 Acc (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard / High Performance]</th>
+<th>CPU Inference Time (ms)<br/>[Standard / High Performance]</th>
+<th>Model Size (M)</th>
+</tr>
+<tr>
+<td>PP-LCNet_x1_0_table_cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/CLIP_vit_base_patch16_224_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_table_cls_pretrained.pdparams">Pretrained Model</a></td>
+<td>94.2</td>
+<td>2.62 / 0.60</td>
+<td>3.17 / 1.14</td>
+<td>6.6M</td>
+</tr>
+</table>
+
+<p><b>Table Cell Detection Module Models:</b></p>
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>mAP (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard / High Performance]</th>
+<th>CPU Inference Time (ms)<br/>[Standard / High Performance]</th>
+<th>Model Size (M)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>RT-DETR-L_wired_table_cell_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wired_table_cell_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wired_table_cell_det_pretrained.pdparams">Pretrained Model</a></td>
+<td rowspan="2">82.7</td>
+<td rowspan="2">33.47 / 27.02</td>
+<td rowspan="2">402.55 / 256.56</td>
+<td rowspan="2">124M</td>
+<td rowspan="2">RT-DETR is the first real-time end-to-end object detection model. Baidu PaddlePaddle Vision Team used RT-DETR-L as the base and pre-trained on a custom table cell detection dataset, achieving strong performance on both wired and wireless tables.</td>
+</tr>
+<tr>
+<td>RT-DETR-L_wireless_table_cell_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wireless_table_cell_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wireless_table_cell_det_pretrained.pdparams">Pretrained Model</a></td>
+</tr>
+</table>
+
+<p><b>Text Detection Module:</b></p>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Detection Hmean (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard / High Performance]</th>
+<th>CPU Inference Time (ms)<br/>[Standard / High Performance]</th>
+<th>Model Size (M)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv5_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_det_pretrained.pdparams">Pretrained Model</a></td>
+<td>83.8</td>
+<td>89.55 / 70.19</td>
+<td>383.15 / 383.15</td>
+<td>84.3</td>
+<td>PP-OCRv5 server-side text detection model, higher accuracy, suitable for deployment on high-performance servers</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_det_pretrained.pdparams">Pretrained Model</a></td>
+<td>79.0</td>
+<td>10.67 / 6.36</td>
+<td>57.77 / 28.15</td>
+<td>4.7</td>
+<td>PP-OCRv5 mobile-side text detection model, more efficient, suitable for edge device deployment</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_det_pretrained.pdparams">Pretrained Model</a></td>
+<td>69.2</td>
+<td>127.82 / 98.87</td>
+<td>585.95 / 489.77</td>
+<td>109</td>
+<td>PP-OCRv4 server-side text detection model, higher accuracy, suitable for deployment on high-performance servers</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams">Pretrained Model</a></td>
+<td>63.8</td>
+<td>9.87 / 4.17</td>
+<td>56.60 / 20.79</td>
+<td>4.7</td>
+<td>PP-OCRv4 mobile-side text detection model, more efficient, suitable for edge device deployment</td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams">Pretrained Model</a></td>
+<td>Accuracy similar to PP-OCRv4_mobile_det</td>
+<td>9.90 / 3.60</td>
+<td>41.93 / 20.76</td>
+<td>2.1</td>
+<td>PP-OCRv3 mobile-side text detection model, more efficient, suitable for edge device deployment</td>
+</tr>
+<tr>
+<td>PP-OCRv3_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_server_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_server_det_pretrained.pdparams">Pretrained Model</a></td>
+<td>Accuracy similar to PP-OCRv4_server_det</td>
+<td>119.50 / 75.00</td>
+<td>379.35 / 318.35</td>
+<td>102.1</td>
+<td>PP-OCRv3 server-side text detection model, higher accuracy, suitable for deployment on high-performance servers</td>
+</tr>
+</tbody>
+</table>
+
+<p><b>Text Recognition Module Models:</b></p>
+
+* <b>Chinese Recognition Models</b>
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Size (M)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">Training Model</a></td>
+<td>86.38</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec is a next-generation text recognition model. It aims to efficiently and accurately support four major languages—Simplified Chinese, Traditional Chinese, English, and Japanese—as well as complex text scenarios such as handwriting, vertical text, pinyin, and rare characters. While maintaining recognition performance, it balances inference speed and model robustness, providing efficient and precise technical support for document understanding in various scenarios.</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>81.29</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">Training Model</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>74.7</td>
+<td>PP-OCRv4_server_rec_doc is trained on a mix of more Chinese document data and PP-OCR training data, based on PP-OCRv4_server_rec. It enhances recognition capabilities for Traditional Chinese, Japanese, and special characters, supporting 15,000+ characters. In addition to improving document-related text recognition, it also enhances general text recognition.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.6</td>
+<td>The lightweight recognition model of PP-OCRv4, with high inference efficiency, deployable on various hardware devices including edge devices.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec </td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">Training Model</a></td>
+<td>80.61 </td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>71.2</td>
+<td>The server-side model of PP-OCRv4, with high inference accuracy, deployable on various servers.</td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>72.96</td>
+<td>3.89 / 1.16</td>
+<td>8.72 / 3.56</td>
+<td>9.2</td>
+<td>The lightweight recognition model of PP-OCRv3, with high inference efficiency, deployable on various hardware devices including edge devices.</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Size (M)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>ch_SVTRv2_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_SVTRv2_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_SVTRv2_rec_pretrained.pdparams">Training Model</a></td>
+<td>68.81</td>
+<td>10.38 / 8.31</td>
+<td>66.52 / 30.83</td>
+<td>73.9</td>
+<td rowspan="1">
+SVTRv2 is a server-side text recognition model developed by the OpenOCR team from Fudan University's Vision and Learning Lab (FVL). It won first prize in the PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition, achieving a 6% improvement in end-to-end recognition accuracy over PP-OCRv4 on the A榜 leaderboard.
+</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Size (M)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>ch_RepSVTR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_RepSVTR_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_RepSVTR_rec_pretrained.pdparams">Training Model</a></td>
+<td>65.07</td>
+<td>6.29 / 1.57</td>
+<td>20.64 / 5.40</td>
+<td>22.1</td>
+<td rowspan="1">RepSVTR is a mobile text recognition model based on SVTRv2. It won first prize in the PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition, achieving a 2.5% improvement in end-to-end recognition accuracy over PP-OCRv4 on the B榜 leaderboard, with comparable inference speed.</td>
+</tr>
+</table>
+
+* <b>English Recognition Models</b>
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Size (M)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td> 70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>6.8</td>
+<td>An ultra-lightweight English recognition model trained based on the PP-OCRv4 recognition model, supporting English and numeric recognition.</td>
+</tr>
+<tr>
+<td>en_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>70.69</td>
+<td>3.56 / 0.78</td>
+<td>8.44 / 5.78</td>
+<td>7.8 M </td>
+<td>An ultra-lightweight English recognition model trained based on the PP-OCRv3 recognition model, supporting English and numeric recognition.</td>
+</tr>
+</table>
+
+* <b>Multilingual Recognition Models</b>
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Size (M)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>korean_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>60.21</td>
+<td>3.73 / 0.98</td>
+<td>8.76 / 2.91</td>
+<td>8.6</td>
+<td>An ultra-lightweight Korean recognition model trained based on the PP-OCRv3 recognition model, supporting Korean and numeric recognition.</td>
+</tr>
+<tr>
+<td>japan_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+japan_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/japan_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>45.69</td>
+<td>3.86 / 1.01</td>
+<td>8.62 / 2.92</td>
+<td>8.8 M </td>
+<td>An ultra-lightweight Japanese recognition model trained based on the PP-OCRv3 recognition model, supporting Japanese and numeric recognition.</td>
+</tr>
+<tr>
+<td>chinese_cht_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+chinese_cht_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/chinese_cht_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>82.06</td>
+<td>3.90 / 1.16</td>
+<td>9.24 / 3.18</td>
+<td>9.7 M </td>
+<td>An ultra-lightweight Traditional Chinese recognition model trained based on the PP-OCRv3 recognition model, supporting Traditional Chinese and numeric recognition.</td>
+</tr>
+<tr>
+<td>te_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+te_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>95.88</td>
+<td>3.59 / 0.81</td>
+<td>8.28 / 6.21</td>
+<td>7.8 M </td>
+<td>An ultra-lightweight Telugu recognition model trained based on the PP-OCRv3 recognition model, supporting Telugu and numeric recognition.</td>
+</tr>
+<tr>
+<td>ka_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ka_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ka_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>96.96</td>
+<td>3.49 / 0.89</td>
+<td>8.63 / 2.77</td>
+<td>8.0 M </td>
+<td>An ultra-lightweight Kannada recognition model trained based on the PP-OCRv3 recognition model, supporting Kannada and numeric recognition.</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ta_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>76.83</td>
+<td>3.49 / 0.86</td>
+<td>8.35 / 3.41</td>
+<td>8.0 M </td>
+<td>An ultra-lightweight Tamil recognition model trained based on the PP-OCRv3 recognition model, supporting Tamil and numeric recognition.</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>76.93</td>
+<td>3.53 / 0.78</td>
+<td>8.50 / 6.83</td>
+<td>7.8</td>
+<td>An ultra-lightweight Latin recognition model trained based on the PP-OCRv3 recognition model, supporting Latin and numeric recognition.</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+arabic_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>73.55</td>
+<td>3.60 / 0.83</td>
+<td>8.44 / 4.69</td>
+<td>7.8</td>
+<td>An ultra-lightweight Arabic script recognition model trained based on the PP-OCRv3 recognition model, supporting Arabic script and numeric recognition.</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+cyrillic_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>94.28</td>
+<td>3.56 / 0.79</td>
+<td>8.22 / 2.76</td>
+<td>7.9 M  </td>
+<td>An ultra-lightweight Cyrillic script recognition model trained based on the PP-OCRv3 recognition model, supporting Cyrillic script and numeric recognition.</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+devanagari_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>96.44</td>
+<td>3.60 / 0.78</td>
+<td>6.95 / 2.87</td>
+<td>7.9</td>
+<td>An ultra-lightweight Devanagari script recognition model trained based on the PP-OCRv3 recognition model, supporting Devanagari script and numeric recognition.</td>
+</tr>
+</table>
+
+<p><b>Text Line Orientation Classification Module (Optional):</b></p>
+<table>
+<thead>
+<tr>
+<th>Model</th>
+<th>Download Link</th>
+<th>Top-1 Acc(%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Size (M)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x0_25_textline_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x0_25_textline_ori_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x0_25_textline_ori_pretrained.pdparams">Training Model</a></td>
+<td>95.54</td>
+<td>2.16 / 0.41</td>
+<td>2.37 / 0.73</td>
+<td>0.32</td>
+<td>A text line classification model based on PP-LCNet_x0_25, with two classes: 0 degrees and 180 degrees.</td>
+</tr>
+</tbody>
+</table>
+
+<p><b>Formula Recognition Module:</b></p>
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Avg-BLEU(%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Size (M)</th>
+<th>Description</th>
+</tr>
+<td>UniMERNet</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UniMERNet_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UniMERNet_pretrained.pdparams">Training Model</a></td>
+<td>86.13</td>
+<td>2266.96/-</td>
+<td>-/-</td>
+<td>1.4 G</td>
+<td>UniMERNet is a formula recognition model developed by Shanghai AI Lab. It uses Donut Swin as the encoder and MBartDecoder as the decoder. Trained on a dataset of one million samples, including simple formulas, complex formulas, scanned formulas, and handwritten formulas, it significantly improves recognition accuracy for real-world scenarios.</td>
+<td>PP-FormulaNet-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-S_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-S_pretrained.pdparams">Training Model</a></td>
+<td>87.12</td>
+<td>1311.84 / 1311.84</td>
+<td>- / 8288.07</td>
+<td>167.9</td>
+<td rowspan="2">PP-FormulaNet is an advanced formula recognition model developed by Baidu's PaddlePaddle Vision team, supporting 50,000 common LaTeX vocabulary items. The PP-FormulaNet-S version uses PP-HGNetV2-B4 as its backbone and employs techniques like parallel masking and model distillation to significantly improve inference speed while maintaining high recognition accuracy, suitable for simple printed formulas, cross-line simple printed formulas, etc. The PP-FormulaNet-L version is based on Vary_VIT_B as its backbone and is trained on a large-scale formula dataset, showing significant improvement in complex formula recognition compared to PP-FormulaNet-S, suitable for simple printed formulas, complex printed formulas, handwritten formulas, etc.</td>
+
+</tr>
+<td>PP-FormulaNet-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-L_pretrained.pdparams">Training Model</a></td>
+<td>92.13</td>
+<td>1976.52/-</td>
+<td>-/-</td>
+<td>535.2</td>
+<td>LaTeX_OCR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/LaTeX_OCR_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/LaTeX_OCR_rec_pretrained.pdparams">Training Model</a></td>
+<td>71.63</td>
+<td>1088.89 / 1088.89</td>
+<td>- / -</td>
+<td>89.7</td>
+<td>LaTeX-OCR is a formula recognition algorithm based on an autoregressive large model. By using Hybrid ViT as the backbone and transformer as the decoder, it significantly improves the accuracy of formula recognition.</td>
+</tr>
+</tbody>
+</table>
+
+<p><b>Seal Text Recognition Module:</b></p>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Detection Hmean(%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Size (M)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv4_server_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_seal_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_seal_det_pretrained.pdparams">Training Model</a></td>
+<td>98.21</td>
+<td>124.64 / 91.57</td>
+<td>545.68 / 439.86</td>
+<td>109</td>
+<td>The server-side seal text detection model of PP-OCRv4, with higher accuracy, suitable for deployment on high-performance servers.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_seal_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_seal_det_pretrained.pdparams">Training Model</a></td>
+<td>96.47</td>
+<td>9.70 / 3.56</td>
+<td>50.38 / 19.64</td>
+<td>4.6</td>
+<td>The mobile-side seal text detection model of PP-OCRv4, with higher efficiency, suitable for deployment on edge devices.</td>
+</tr>
+</tbody>
+</table>
+
+<strong>Testing Environment Description:</strong>
+
+  <ul>
+      <li><b>Performance Testing Environment</b>
+          <ul>
+            <li><strong>Test Datasets:
+             </strong>
+                <ul>
+                  <li>Document Image Orientation Classification Model: A dataset built by PaddleX, covering multiple scenarios such as IDs and documents, containing 1,000 images.</li>
+                  <li>Text Image Unwarping Model: <a href="https://www3.cs.stonybrook.edu/~cvl/docunet.html">DocUNet</a>.</li>
+                  <li>Layout Detection Model: A layout analysis dataset built by PaddleOCR, containing 10,000 common document-type images such as Chinese and English papers, magazines, and reports.</li>
+                  <li>PP-DocLayout_plus-L: A layout detection dataset built by PaddleOCR, containing 1,300 document-type images such as Chinese and English papers, magazines, newspapers, reports, PPTs, exams, and textbooks.</li>
+                  <li>Table Structure Recognition Model: An internal English table recognition dataset built by PaddleX.</li>
+                  <li>Text Detection Model: A Chinese dataset built by PaddleOCR, covering street views, web images, documents, and handwriting, with 500 images for detection.</li>
+                  <li>Chinese Recognition Model: A Chinese dataset built by PaddleOCR, covering street views, web images, documents, and handwriting, with 11,000 images for text recognition.</li>
+                  <li>ch_SVTRv2_rec: <a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition</a> A榜 evaluation set.</li>
+                  <li>ch_RepSVTR_rec: <a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition</a> B榜 evaluation set.</li>
+                  <li>English Recognition Model: An English dataset built by PaddleX.</li>
+                  <li>Multilingual Recognition Model: A multilingual dataset built by PaddleX.</li>
+                  <li>Text Line Orientation Classification Model: A dataset built by PaddleX, covering multiple scenarios such as IDs and documents, containing 1,000 images.</li>
+                  <li>Seal Text Recognition Model: A dataset built by PaddleX, containing 500 circular seal images.</li>
+                </ul>
+             </li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>Inference Mode Description</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration</th>
+            <th>CPU Configuration</th>
+            <th>Acceleration Technology Combination</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Regular Mode</td>
+            <td>FP32 Precision / No TRT Acceleration</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Local Paddle inference engines (by default, the engine is selected according to the default model name; if both static and dynamic graph are available, <code>paddle_static</code> is preferred)</td>
+        </tr>
+        <tr>
+            <td>High-Performance Mode</td>
+            <td>Optimal combination of precision types and acceleration strategies</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Optimal backend selection (Paddle/OpenVINO/TRT, etc.)</td>
+        </tr>
+    </tbody>
+</table>
+
+</details>
+
+## 2. Quick Start
+
+Before using the PP-DocTranslation pipeline locally, please ensure that you have completed the installation of the wheel package according to the [Installation Tutorial](../installation.en.md). If you prefer to install dependencies selectively, please refer to the relevant instructions in the installation documentation. The corresponding dependency group for this pipeline is `trans`.
+
+Please note: If you encounter issues such as the program becoming unresponsive, unexpected program termination, running out of memory resources, or extremely slow inference during execution, please try adjusting the configuration according to the documentation, such as disabling unnecessary features or using lighter-weight models.
+
+Before use, you need to prepare the API key for a large language model, which supports the [Baidu Cloud Qianfan Platform](https://console.bce.baidu.com/qianfan/ais/console/onlineService) or local large model services that comply with the OpenAI interface standards.
+
+### 2.1 Experience via Command Line
+
+You can download the [test file](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png) and quickly experience the pipeline effect with a single command:
+
+```bash
+paddleocr pp_doctranslation -i vehicle_certificate-1.png --target_language en --qianfan_api_key your_api_key
+```
+
+The examples above use local Paddle inference engines by default. By default, each module selects the appropriate local Paddle inference engine according to the default model name: models that support only dynamic graph use `paddle_dynamic`, while models that support both static and dynamic graph prefer `paddle_static`. To run them, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr pp_doctranslation -i vehicle_certificate-1.png --target_language en --qianfan_api_key your_api_key \
+    --engine transformers
+```
+
+<details><summary><b>Command line supports more parameter settings. Click to expand for detailed description of command line parameters</b></summary>
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted, required. <br/>
+<b>Description:</b>
+For example, local path of image file or PDF file: <code>/root/data/img.jpg</code>; <b>URL link</b>, such as network URL of image or PDF file: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png">example</a>; <b>local directory</b>, the directory must contain images to be predicted, such as local path: <code>/root/data/</code> (currently does not support PDF files in the directory, PDF files need to specify the exact file path).
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td><b>Meaning:</b>Specifies the path to save the inference result files. <br/>
+<b>Description:</b>
+If not set, inference results will not be saved locally.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>target_language</code></td>
+<td><b>Meaning:</b>Target language (ISO 639-1 language code).</td>
+<td><code>str</code></td>
+<td><code>zh</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>Meaning:</b>Model name for layout detection.<br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the layout detection model.<br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>Score threshold for layout model. <br/>
+<b>Description:</b>Any float between <code>0-1</code>. If not set, the pipeline initialized value will be used, default initialized as <code>0.5</code>.</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Whether to use post-processing NMS in layout detection. <br/>
+<b>Description:</b> 
+If not set, the pipeline initialized value will be used, default initialized as <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion coefficient for detection boxes in layout detection model. <br/>
+<b>Description:</b> 
+Any float greater than <code>0</code>. If not set, the pipeline initialized value will be used, default initialized as <code>1.0</code>.</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>Mode for merging detection boxes output by the layout detection model. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>large</b>: when set to large, among overlapping boxes, only the largest outer box is kept and the overlapping inner boxes are deleted;</li>
+<li><b>small</b>: when set to small, among overlapping boxes, only the smaller inner boxes are kept and the overlapping outer boxes are deleted;</li>
+<li><b>union</b>: no box filtering, both inner and outer boxes are kept;</li>
+</ul>If not set, the pipeline initialized value will be used, default initialized as <code>large</code>.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>chart_recognition_model_name</code></td>
+<td><b>Meaning:</b>Model name for chart parsing. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>chart_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for chart parsing model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>chart_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for chart parsing model. <br/>
+<b>Description:</b> 
+If not set, batch size defaults to <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>region_detection_model_name</code></td>
+<td><b>Meaning:</b>Model name for region detection. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>region_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for region detection model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>Model name for document orientation classification. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for document orientation classification model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>Meaning:</b>Model name for text image unwarping.<br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for text image unwarping model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>Meaning:</b>Model name for text detection. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for text detection model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limit for text detection. <br/>
+<b>Description:</b> 
+Any integer greater than <code>0</code>. If not set, the pipeline initialized value will be used, default initialized as <code>960</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>Meaning:</b>Type of image side length limit for text detection. <br/>
+<b>Description:</b> 
+Supports <code>min</code> and <code>max</code>. <code>min</code> means ensuring the shortest side of the image is not less than <code>det_limit_side_len</code>, <code>max</code> means ensuring the longest side of the image is not greater than <code>limit_side_len</code>. If not set, the pipeline initialized value will be used, default initialized as <code>max</code>.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>Meaning:</b>Detection pixel threshold. In the output probability map, pixels with score greater than this threshold are considered text pixels. <br/>
+<b>Description:</b> 
+Any float greater than <code>0</code>. If not set, the pipeline initialized value <code>0.3</code> will be used by default.</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>Meaning:</b>Detection box threshold. If the average score of all pixels within the detected bounding box is greater than this threshold, the result is considered a text region. <br/>
+<b>Description:</b> 
+Any float greater than <code>0</code>. If not set, the pipeline initialized value <code>0.6</code> will be used by default.</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Text detection expansion coefficient, used to expand text regions. The larger the value, the larger the expansion area. <br/>
+<b>Description:</b> 
+Any float greater than <code>0</code>. If not set, the pipeline initialized value <code>2.0</code> will be used by default.</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_name</code></td>
+<td><b>Meaning:</b>Model name for textline orientation. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for textline orientation model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for textline orientation model. <br/>
+<b>Description:</b> 
+If not set, batch size defaults to <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>Meaning:</b>Model name for text recognition. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for text recognition model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for text recognition model. <br/>
+<b>Description:</b> 
+If not set, batch size defaults to <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Text recognition threshold. Text results with scores greater than this threshold will be kept. <br/>
+<b>Description:</b> 
+Any float greater than <code>0</code>. If not set, the pipeline initialized value <code>0.0</code> will be used, meaning no threshold.</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_classification_model_name</code></td>
+<td><b>Meaning:</b>Model name for table classification. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_classification_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for table classification model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_name</code></td>
+<td><b>Meaning:</b>Model name for wired table structure recognition. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for wired table structure recognition model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_name</code></td>
+<td><b>Meaning:</b>Model name for wireless table structure recognition. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for wireless table structure recognition model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_name</code></td>
+<td><b>Meaning:</b>Model name for wired table cells detection. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for wired table cells detection model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_name</code></td>
+<td><b>Meaning:</b>Model name for wireless table cells detection. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for wireless table cells detection model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>Model name for table orientation classification. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for table orientation classification model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_name</code></td>
+<td><b>Meaning:</b>Model name for seal text detection. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for seal text detection model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limit for seal text detection. <br/>
+<b>Description:</b> 
+Any integer greater than <code>0</code>. If not set, the pipeline initialized value will be used, default initialized as <code>736</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>Meaning:</b>Type of image side length limit for seal text detection. <br/>
+<b>Description:</b> 
+Supports <code>min</code> and <code>max</code>. <code>min</code> means ensuring the shortest side of the image is not less than <code>det_limit_side_len</code>, <code>max</code> means ensuring the longest side is not greater than <code>limit_side_len</code>. If not set, the pipeline initialized value will be used, default initialized as <code>min</code>.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>Meaning:</b>Detection pixel threshold. In the output probability map, pixels with score greater than this threshold are considered text pixels. <br/>
+<b>Description:</b> 
+Any float greater than <code>0</code>. If not set, the pipeline initialized value <code>0.2</code> will be used by default.</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>Meaning:</b>Detection box threshold. If the average score of all pixels within the detected bounding box is greater than this threshold, the result is considered a text region. <br/>
+<b>Description:</b> 
+Any float greater than <code>0</code>. If not set, the pipeline initialized value <code>0.6</code> will be used by default.</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion coefficient for seal text detection. This method is used to expand the text region; the larger the value, the larger the expansion area. <br/>
+<b>Description:</b> 
+Any float greater than <code>0</code>. If not set, the pipeline initialized value <code>0.5</code> will be used by default.</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_name</code></td>
+<td><b>Meaning:</b>Model name for seal text recognition. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for seal text recognition model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for seal text recognition model. <br/>
+<b>Description:</b> 
+If not set, batch size defaults to <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Text recognition threshold. Text results with scores greater than this threshold will be kept. <br/>
+<b>Description:</b> 
+</b>
+Any float greater than <code>0</code>. If not set, the pipeline initialized value <code>0.0</code> will be used, meaning no threshold.</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_name</code></td>
+<td><b>Meaning:</b>Model name for formula recognition. <br/>
+<b>Description:</b> 
+If not set, the pipeline default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for formula recognition model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>formula_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size of the formula recognition model. <br/>
+<b>Description:</b> 
+If not set, the batch size defaults to <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to load and use the document orientation classification module. <br/>
+<b>Description:</b> 
+If not set, the pipeline initialized value will be used, default is <code>False</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to load and use the text image unwarping module. <br/>
+<b>Description:</b> 
+If not set, the pipeline initialized value will be used, default is <code>False</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>Meaning:</b>Whether to load and use the text line orientation classification module. <br/>
+<b>Description:</b> 
+If not set, the pipeline initialized value will be used, default is <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>Meaning:</b>Whether to load and use the seal text recognition sub-pipeline. <br/>
+<b>Description:</b> 
+If not set, the pipeline initialized value will be used, default is <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>Meaning:</b>Whether to load and use the table recognition sub-pipeline. <br/>
+<b>Description:</b> 
+If not set, the pipeline initialized value will be used, default is <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_formula_recognition</code></td>
+<td><b>Meaning:</b>Whether to load and use the formula recognition sub-pipeline. <br/>
+<b>Description:</b> 
+If not set, the pipeline initialized value will be used, default is <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>Meaning:</b>Whether to load and use the chart parsing module. <br/>
+<b>Description:</b> 
+If not set, the pipeline initialized value will be used, default is <code>False</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_region_detection</code></td>
+<td><b>Meaning:</b>Whether to load and use the region detection module. <br/>
+<b>Description:</b> 
+If not set, the pipeline initialized value will be used, default is <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>qianfan_api_key</code></td>
+<td><b>Meaning:</b>API key for the Qianfan platform.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>Device used for inference.<br/>
+<b>Description:</b> 
+Supports specifying exact card number:
+<ul>
+<li><b>CPU</b>: e.g. <code>cpu</code> means using CPU for inference;</li>
+<li><b>GPU</b>: e.g. <code>gpu:0</code> means using GPU #1 for inference;</li>
+<li><b>NPU</b>: e.g. <code>npu:0</code> means using NPU #1 for inference;</li>
+<li><b>XPU</b>: e.g. <code>xpu:0</code> means using XPU #1 for inference;</li>
+<li><b>MLU</b>: e.g. <code>mlu:0</code> means using MLU #1 for inference;</li>
+<li><b>DCU</b>: e.g. <code>dcu:0</code> means using DCU #1 for inference;</li>
+<li><b>MetaX GPU</b>: e.g. <code>metax_gpu:0</code> means using MetaX GPU #1 for inference;</li>
+<li><b>Iluvatar GPU</b>: e.g. <code>iluvatar_gpu:0</code> means using Iluvatar GPU #1 for inference;</li>
+</ul>If not set, the pipeline initialized value will be used. At initialization, the local GPU device #0 will be preferred, if none, CPU device will be used.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to enable the TensorRT subgraph engine of Paddle Inference.<br/>
+<b>Description:</b>
+If the model does not support TensorRT acceleration, acceleration will not be used even if this flag is set.<br/>
+For CUDA 11.8 versions of PaddlePaddle, the compatible TensorRT version is 8.x (x>=6). TensorRT 8.6.1.6 is recommended.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b> Computation precision, such as <code>fp32</code> or <code>fp16</code>.</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b> Whether to enable MKL-DNN accelerated inference.<br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support MKL-DNN acceleration, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b> MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads used for inference on CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b> Path to the PaddleX pipeline configuration file.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+</tbody>
+</table>
+</details>
+<br />
+
+The execution results will be printed to the terminal.
+
+### 2.2 Integration via Python Script
+
+The command-line method is for quickly experiencing and viewing the results. Generally, in projects, integration via code is often required. You can download the [test file](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png) and use the following sample code for inference:
+
+```python
+from paddleocr import PPDocTranslation
+
+# Create a translation pipeline
+pipeline = PPDocTranslation()
+
+# Document path
+input_path = "document_sample.pdf"
+
+# Output directory
+output_path = "./output"
+
+# Large model configuration
+chat_bot_config = {
+    "module_name": "chat_bot",
+    "model_name": "ernie-3.5-8k",
+    "base_url": "https://qianfan.baidubce.com/v2",
+    "api_type": "openai",
+    "api_key": "api_key",  # your api_key
+}
+
+if input_path.lower().endswith(".md"):
+    # Read markdown documents, supporting passing in directories and url links with the .md suffix
+    ori_md_info_list = pipeline.load_from_markdown(input_path)
+else:
+    # Use PP-StructureV3 to perform layout parsing on PDF/image documents to obtain markdown information
+    visual_predict_res = pipeline.visual_predict(
+        input_path,
+        use_doc_orientation_classify=False,
+        use_doc_unwarping=False,
+        use_common_ocr=True,
+        use_seal_recognition=True,
+        use_table_recognition=True,
+    )
+
+    ori_md_info_list = []
+    for res in visual_predict_res:
+        layout_parsing_result = res["layout_parsing_result"]
+        ori_md_info_list.append(layout_parsing_result.markdown)
+        layout_parsing_result.save_to_img(output_path)
+        layout_parsing_result.save_to_markdown(output_path)
+
+    # Concatenate the markdown information of multi-page documents into a single markdown file, and save the merged original markdown text
+    if input_path.lower().endswith(".pdf"):
+        ori_md_info = pipeline.concatenate_markdown_pages(ori_md_info_list)
+        ori_md_info.save_to_markdown(output_path)
+
+# Perform document translation (target language: English)
+tgt_md_info_list = pipeline.translate(
+    ori_md_info_list=ori_md_info_list,
+    target_language="en",
+    chunk_size=5000,
+    chat_bot_config=chat_bot_config,
+)
+# Save the translation results
+for tgt_md_info in tgt_md_info_list:
+    tgt_md_info.save_to_markdown(output_path)
+```
+
+The example above uses local Paddle inference engines by default. By default, each module selects the appropriate local Paddle inference engine according to the default model name: models that support only dynamic graph use `paddle_dynamic`, while models that support both static and dynamic graph prefer `paddle_static`. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following code:
+
+```python
+from paddleocr import PPDocTranslation
+
+# Create a translation pipeline
+pipeline = PPDocTranslation(
+    engine="transformers",
+)
+
+# Document path
+input_path = "document_sample.pdf"
+
+# Output directory
+output_path = "./output"
+
+# Large model configuration
+chat_bot_config = {
+    "module_name": "chat_bot",
+    "model_name": "ernie-3.5-8k",
+    "base_url": "https://qianfan.baidubce.com/v2",
+    "api_type": "openai",
+    "api_key": "api_key",  # your api_key
+}
+
+if input_path.lower().endswith(".md"):
+    # Read markdown documents, supporting passing in directories and url links with the .md suffix
+    ori_md_info_list = pipeline.load_from_markdown(input_path)
+else:
+    # Use PP-StructureV3 to perform layout parsing on PDF/image documents to obtain markdown information
+    visual_predict_res = pipeline.visual_predict(
+        input_path,
+        use_doc_orientation_classify=False,
+        use_doc_unwarping=False,
+        use_common_ocr=True,
+        use_seal_recognition=True,
+        use_table_recognition=True,
+    )
+
+    ori_md_info_list = []
+    for res in visual_predict_res:
+        layout_parsing_result = res["layout_parsing_result"]
+        ori_md_info_list.append(layout_parsing_result.markdown)
+        layout_parsing_result.save_to_img(output_path)
+        layout_parsing_result.save_to_markdown(output_path)
+
+    # Concatenate the markdown information of multi-page documents into a single markdown file, and save the merged original markdown text
+    if input_path.lower().endswith(".pdf"):
+        ori_md_info = pipeline.concatenate_markdown_pages(ori_md_info_list)
+        ori_md_info.save_to_markdown(output_path)
+
+# Perform document translation (target language: English)
+tgt_md_info_list = pipeline.translate(
+    ori_md_info_list=ori_md_info_list,
+    target_language="en",
+    chunk_size=5000,
+    chat_bot_config=chat_bot_config,
+)
+# Save the translation results
+for tgt_md_info in tgt_md_info_list:
+    tgt_md_info.save_to_markdown(output_path)
+```
+
+After executing the above code, you will obtain the parsed results of the original document to be translated, the Markdown file of the original text to be translated, and the Markdown file of the translated document, all saved in the <code>output</code> directory.
+
+The process, API description, and output description of PP-DocTranslation prediction are as follows:
+
+<details><summary>(1) Instantiate the PP-DocTranslation pipeline object by calling <code>PPDocTranslation</code>.</summary>
+
+Relevant parameter descriptions are as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>Meaning:</b>The model name for layout detection. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the layout detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>Score threshold for the layout model.<br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float between <code>0-1</code>;</li>
+<li><b>dict</b>: <code>{0:0.1}</code>, where the key is the class ID and the value is the threshold for that class;</li>
+<li><b>None</b>: If set to <code>None</code>, the pipeline's initialized value will be used, defaulting to <code>0.5</code>.</li>
+</ul>
+</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Whether to use post-processing NMS for layout detection. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline's initialized value will be used, defaulting to <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion coefficient for detection boxes in the layout detection model. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>Tuple[float,float]</b>: Expansion coefficients in horizontal and vertical directions respectively;</li>
+<li><b>dict</b>: Keys are <b>int</b> representing <code>cls_id</code>, values are <b>tuple</b>, e.g. <code>{0: (1.1, 2.0)}</code>, meaning for class 0 detection boxes, center remains unchanged, width expanded by 1.1 times, height expanded by 2.0 times;</li>
+<li><b>None</b>: If set to <code>None</code>, the pipeline's initialized value will be used, defaulting to <code>1.0</code>.</li>
+</ul>
+</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>Overlap box filtering method for layout detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>str</b>: <code>large</code>, <code>small</code>, <code>union</code>, indicating whether to keep the larger box, smaller box, or both during overlap filtering;</li>
+<li><b>dict</b>: Keys are <b>int</b> <code>cls_id</code>, values are <b>str</b>, e.g. <code>{0: "large", 2: "small"}</code>, meaning use "large" mode for class 0 boxes and "small" mode for class 2 boxes;</li>
+<li><b>None</b>: If set to <code>None</code>, the pipeline's initialized value will be used, defaulting to <code>large</code>.</li>
+</ul>
+</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>chart_recognition_model_name</code></td>
+<td><b>Meaning:</b>The model name for chart parsing. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>chart_recognition_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the chart parsing model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>chart_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the chart parsing model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, batch size defaults to <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>region_detection_model_name</code></td>
+<td><b>Meaning:</b>The model name for region detection. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>region_detection_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the region detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>The model name for document orientation classification. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the document orientation classification model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>Meaning:</b>The model name for text image unwarping. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the text image unwarping model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>Meaning:</b>The model name for text detection. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the text detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limit for text detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>int</b>: Any integer greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the pipeline's initialized value will be used, defaulting to <code>960</code>.</li>
+</ul>
+</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>Meaning:</b>Type of image side length limit for text detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>str</b>: Supports <code>min</code> and <code>max</code>, where <code>min</code> means ensuring the shortest side of the image is not less than <code>det_limit_side_len</code>, and <code>max</code> means ensuring the longest side is not greater than <code>limit_side_len</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the pipeline's initialized value will be used, defaulting to <code>max</code>.</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>Meaning:</b>Pixel threshold for detection; pixels in the output probability map with scores above this threshold are considered text pixels. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the pipeline's initialized value of <code>0.3</code> will be used.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>Meaning:</b>Detection box threshold; when the average score of all pixels inside a detected box exceeds this threshold, it is considered a text region. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the pipeline's initialized value of <code>0.6</code> will be used.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion coefficient for text detection; this method expands the text region, and the larger the value, the larger the expansion area. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the pipeline's initialized value of <code>2.0</code> will be used.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_name</code></td>
+<td><b>Meaning:</b>The model name for text line orientation classification. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the text line orientation model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the text line orientation model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, batch size defaults to <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>Meaning:</b>The model name for text recognition. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the text recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the text recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, batch size defaults to <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Text recognition threshold; text results with scores greater than this threshold will be retained. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the pipeline's initialized value of <code>0.0</code> (no threshold) will be used.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_classification_model_name</code></td>
+<td><b>Meaning:</b>The model name for table classification. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_classification_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the table classification model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_name</code></td>
+<td><b>Meaning:</b>The model name for wired table structure recognition. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the wired table structure recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_name</code></td>
+<td><b>Meaning:</b>The model name for wireless table structure recognition. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the wireless table structure recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_name</code></td>
+<td><b>Meaning:</b>The model name for wired table cell detection. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the wired table cell detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_name</code></td>
+<td><b>Meaning:</b>The model name for wireless table cell detection. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the wireless table cell detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>The model name for table orientation classification. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the table orientation classification model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_name</code></td>
+<td><b>Meaning:</b>The model name for seal text detection. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the seal text detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limit for seal text detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>int</b>: any integer greater than <code>0</code>;</li>
+<li><b>None</b>: if set to <code>None</code>, the parameter value initialized by the pipeline will be used, with a default initialization of <code>736</code>.</li>
+</ul>
+</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>Meaning:</b>Type of image side length limit for seal text detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>str</b>: supports <code>min</code> and <code>max</code>, where <code>min</code> ensures the shortest image side is not less than <code>det_limit_side_len</code>, and <code>max</code> ensures the longest image side is not greater than <code>limit_side_len</code>;</li>
+<li><b>None</b>: if set to <code>None</code>, the parameter value initialized by the pipeline will be used, with a default initialization of <code>min</code>.</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>Meaning:</b>Detection pixel threshold. In the output probability map, pixels with scores above this threshold are considered text pixels. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: any floating number greater than <code>0</code>;</li>
+<li><b>None</b>: if set to <code>None</code>, the pipeline default parameter value <code>0.2</code> will be used.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>Meaning:</b>Detection box threshold. When the average score of all pixels within the detected bounding box is greater than this threshold, the result is considered a text region. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: any floating number greater than <code>0</code>;</li>
+<li><b>None</b>: if set to <code>None</code>, the pipeline default parameter value <code>0.6</code> will be used.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion coefficient for seal text detection. This method expands the text region; the larger the value, the larger the expansion area. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: any floating number greater than <code>0</code>;</li>
+<li><b>None</b>: if set to <code>None</code>, the pipeline default parameter value <code>0.5</code> will be used.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the seal text recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for the seal text recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the seal text recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the batch size defaults to <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Seal text recognition threshold. Text results with scores above this threshold will be retained. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: any floating number greater than <code>0</code>;</li>
+<li><b>None</b>: if set to <code>None</code>, the pipeline default parameter value <code>0.0</code> will be used, meaning no threshold is set.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the formula recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path for the formula recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the formula recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the batch size defaults to <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to load and use the document orientation classification module. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline initialized parameter value will be used, defaulting to <code>False</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td>Whether to load and use the text image unwarping module. 
+If set to <code>None</code>, the pipeline initialized parameter value will be used, defaulting to <code>False</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td>Whether to load and use the text line orientation classification module. If set to <code>None</code>, the pipeline initialized parameter value will be used, defaulting to <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>Meaning:</b>Whether to load and use the seal text recognition sub-pipeline. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline initialized parameter value will be used, defaulting to <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>Meaning:</b>Whether to load and use the table recognition sub-pipeline. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline initialized parameter value will be used, defaulting to <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_formula_recognition</code></td>
+<td><b>Meaning:</b>Whether to load and use the formula recognition sub-pipeline. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline initialized parameter value will be used, defaulting to <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>Meaning:</b>Whether to load and use the chart parsing module. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline initialized parameter value will be used, defaulting to <code>False</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_region_detection</code></td>
+<td><b>Meaning:</b>Whether to load and use the document region detection module. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline initialized parameter value will be used, defaulting to <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>chat_bot_config</code></td>
+<td><b>Meaning:</b>Large language model configuration information. <br/>
+<b>Description:</b>
+The configuration content is the following dict:
+<pre><code>{
+"module_name": "chat_bot",
+"model_name": "ernie-3.5-8k",
+"base_url": "https://qianfan.baidubce.com/v2",
+"api_type": "openai",
+"api_key": "api_key"  # Please set this to the actual API key
+}</code></pre>
+</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>Device used for inference. <br/>
+<b>Description:</b> 
+Supports specifying a specific card number:
+<ul>
+<li><b>CPU</b>: e.g. <code>cpu</code> means using CPU for inference;</li>
+<li><b>GPU</b>: e.g. <code>gpu:0</code> means using the first GPU for inference;</li>
+<li><b>NPU</b>: e.g. <code>npu:0</code> means using the first NPU for inference;</li>
+<li><b>XPU</b>: e.g. <code>xpu:0</code> means using the first XPU for inference;</li>
+<li><b>MLU</b>: e.g. <code>mlu:0</code> means using the first MLU for inference;</li>
+<li><b>DCU</b>: e.g. <code>dcu:0</code> means using the first DCU for inference;</li>
+<li><b>MetaX GPU</b>: e.g. <code>metax_gpu:0</code> means using the first MetaX GPU for inference;</li>
+<li><b>Iluvatar GPU</b>: e.g. <code>iluvatar_gpu:0</code> means using the first Iluvatar GPU for inference;</li>
+<li><b>None</b>: if set to <code>None</code>, initialization will prioritize using the local GPU device 0; if unavailable, CPU will be used.</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to enable the TensorRT subgraph engine of Paddle Inference.<br/>
+<b>Description:</b>
+If the model does not support TensorRT acceleration, acceleration will not be used even if this flag is set.<br/>
+For CUDA 11.8 versions of PaddlePaddle, the compatible TensorRT version is 8.x (x>=6). TensorRT 8.6.1.6 is recommended.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b> Computation precision, such as <code>"fp32"</code> or <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b> Whether to enable MKL-DNN accelerated inference.<br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support MKL-DNN acceleration, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b> MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads used for inference on CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b> Path to the PaddleX pipeline configuration file.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+</tbody>
+</table>
+
+</details>
+
+<details><summary>(2) Call the <code>visual_predict()</code> method of the PP-DocTranslation pipeline object to obtain visual prediction results. This method returns a list of results. Additionally, the pipeline provides a <code>visual_predict_iter()</code> method. Both methods accept the same parameters and return the same results, but <code>visual_predict_iter()</code> returns a <code>generator</code>, which can process and retrieve prediction results step-by-step, suitable for large datasets or memory-saving scenarios. You can choose either method according to your actual needs. Below are the parameters of the <code>visual_predict()</code> method and their descriptions:</summary>
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted, supports multiple input types, required.<br/>
+<b>Description:</b> 
+<ul>
+  <li><b>Python Var</b>: image data such as <code>numpy.ndarray</code>;</li>
+  <li><b>str</b>: local path of image or PDF files, e.g. <code>/root/data/img.jpg</code>; <b>URL link</b>: network URL of image or PDF files, e.g. <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png">example</a>; <b>local directory</b>: directory containing images to be predicted, e.g. <code>/root/data/</code> (currently does not support PDFs in directories, PDF files need to specify exact file path);</li>
+  <li><b>list</b>: list elements must be one of the above types, e.g. <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"]</code>.</li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to use the document orientation classification module during inference. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to use the text image unwarping module during inference. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>Meaning:</b>Whether to use the text line orientation classification module during inference. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>Meaning:</b>Whether to use the seal text recognition sub-pipeline during inference. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>Meaning:</b>Whether to use the table recognition sub-pipeline during inference. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_formula_recognition</code></td>
+<td><b>Meaning:</b>Whether to use the formula recognition sub-pipeline during inference.<br/>
+<b>Description:</b> 
+ Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>Meaning:</b>Whether to use the chart parsing module during inference. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_region_detection</code></td>
+<td><b>Meaning:</b>Whether to use the document layout detection module during inference. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>Parameter meaning is basically the same as the instantiated parameter. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Parameter meaning is basically the same as the instantiated parameter. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Parameter meaning is basically the same as the instantiated parameter. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>Parameter meaning is basically the same as the instantiated parameter. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Parameter meaning is basically the same as the instantiated parameter. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>Meaning:</b>Parameter meaning is basically the same as the instantiated parameter. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>Meaning:</b>Parameter meaning is basically the same as the instantiated parameter. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>Meaning:</b>Parameter meaning is basically the same as the instantiated parameter. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Parameter meaning is basically the same as the instantiated parameter. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Parameter meaning is basically the same as the instantiated parameter. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Parameter meaning is basically the same as the instantiated parameter. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>Meaning:</b>Parameter meaning is basically the same as the instantiated parameter. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>Meaning:</b>Parameter meaning is basically the same as the instantiated parameter. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>Meaning:</b>Parameter meaning is basically the same as the instantiated parameter. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Parameter meaning is basically the same as the instantiated parameter. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Parameter meaning is basically the same as the instantiated parameter. <br/>
+<b>Description:</b> 
+Setting to <code>None</code> means using the instantiated parameter, otherwise this parameter has higher priority.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_wired_table_cells_trans_to_html</code></td>
+<td><b>Meaning:</b>Whether to enable direct conversion of wired table cell detection results to HTML.<br/> 
+<b>Description:</b> 
+When enabled, HTML is constructed directly based on the geometric relations of wired table cell detection results.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_wireless_table_cells_trans_to_html</code></td>
+<td><b>Meaning:</b>Whether to enable direct conversion of wireless table cell detection results to HTML. <br/>
+<b>Description:</b> 
+When enabled, HTML is constructed directly based on the geometric relations of wireless table cell detection results.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_table_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to enable table orientation classification. <br/>
+<b>Description:</b> 
+When enabled, tables with 90/180/270 degree rotations in images can be corrected in orientation and correctly recognized.</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_ocr_results_with_table_cells</code></td>
+<td><b>Meaning:</b>Whether to enable OCR segmentation by table cells. <br/>
+<b>Description:</b> 
+When enabled, OCR detection results are segmented and re-recognized based on cell prediction results to avoid missing text.</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_e2e_wired_table_rec_model</code></td>
+<td><b>Meaning:</b>Whether to enable end-to-end wired table recognition mode. <br/>
+<b>Description:</b> 
+When enabled, the cell detection model is not used, only the table structure recognition model is used.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_e2e_wireless_table_rec_model</code></td>
+<td><b>Meaning:</b>Whether to enable end-to-end wireless table recognition mode. <br/>
+<b>Description:</b> 
+When enabled, the cell detection model is not used, only the table structure recognition model is used.</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+</table>
+</details>
+
+<details><summary>(3) Processing visual prediction results: Each sample's prediction result is a corresponding Result object, supporting operations such as printing, saving as images, and saving as <code>json</code> files:</summary>
+
+<table>
+
+<thead>
+<tr>
+<th>Method</th>
+<th>Description</th>
+<th>Parameter</th>
+<th>Parameter Type</th>
+<th>Parameter Description</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">Print results to terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output content using <code>JSON</code> indentation</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify indentation level to beautify output <code>JSON</code> data for better readability, effective only when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether non-<code>ASCII</code> characters are escaped as <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; if <code>False</code>, original characters are preserved. Effective only when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">Save results as a JSON file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>File path for saving. If a directory is specified, the saved file name matches the input file type name</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify indentation level to beautify output <code>JSON</code> data for better readability, effective only when <code>format_json</code> is <code>True</code></td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether non-<code>ASCII</code> characters are escaped as <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; if <code>False</code>, original characters are preserved. Effective only when <code>format_json</code> is <code>True</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save visualized images from intermediate modules as PNG format images</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>File path for saving, supports directory or file path</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>save_to_markdown()</code></td>
+<td>Save each page of image or PDF files as separate markdown files</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>File path for saving, supports directory or file path</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>save_to_html()</code></td>
+<td>Save tables in the file as HTML format files</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>File path for saving, supports directory or file path</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>save_to_xlsx()</code></td>
+<td>Save tables in the file as XLSX format files</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>File path for saving, supports directory or file path</td>
+<td>None</td>
+</tr>
+</table>
+
+<ul>
+<li>Calling the<code>print()</code> method will print the results to the terminal, with the following explanation of printed content:
+    <ol start="1" type="1">
+    <li><code>input_path</code>: <code>(str)</code> Input path of the image or PDF to be predicted</li>
+    <li><code>page_index</code>: <code>(Union[int, None])</code> If the input is a PDF, this indicates the current page number; otherwise <code>None</code></li>
+    <li><code>model_settings</code>: <code>(Dict[str, bool])</code> Model parameters configured for the pipeline
+        <ol>
+        <li><code>use_doc_preprocessor</code>: <code>(bool)</code> Controls whether to enable the document preprocessing sub-pipeline</li>
+        <li><code>use_general_ocr</code>: <code>(bool)</code> Controls whether to enable the OCR sub-pipeline</li>
+        <li><code>use_seal_recognition</code>: <code>(bool)</code> Controls whether to enable the seal recognition sub-pipeline</li>
+        <li><code>use_table_recognition</code>: <code>(bool)</code> Controls whether to enable the table recognition sub-pipeline</li>
+        <li><code>use_formula_recognition</code>: <code>(bool)</code> Controls whether to enable the formula recognition sub-pipeline</li>
+        </ol>
+    </li>
+    <li><code>doc_preprocessor_res</code>: <code>(Dict[str, Union[List[float], str]])</code> Document preprocessing result dictionary, present only when <code>use_doc_preprocessor=True</code>
+        <ol>
+        <li><code>input_path</code>: <code>(str)</code> Input path of document preprocessing sub-pipeline, when input is <code>numpy.ndarray</code>, saved as <code>None</code>, here it is <code>None</code></li>
+        <li><code>page_index</code>: <code>(None)</code> Page index of document preprocessing sub-pipeline, when input is <code>numpy.ndarray</code>, saved as <code>None</code>, here it is <code>None</code></li>
+        <li><code>model_settings</code>: <code>(Dict[str, bool])</code> Model configuration parameters of document preprocessing sub-pipeline
+          <ul>
+          <li><code>use_doc_orientation_classify</code>: <code>(bool)</code> Controls whether to enable the document image orientation classification sub-pipeline</li>
+          <li><code>use_doc_unwarping</code>: <code>(bool)</code> Controls whether to enable the document image unwarping sub-pipeline</li>
+          </ul>
+        <li><code>angle</code>: <code>(int)</code> Document image orientation classification sub-pipeline prediction result, returns actual angle value when enabled</li>
+        </ul>
+    </li>
+    <li><code>parsing_res_list</code>: <code>(List[Dict])</code> List of parsing results, each element is a dictionary, the order of the list is the reading order after parsing.
+        <ol>
+        <li><code>block_bbox</code>: <code>(np.ndarray)</code> Bounding box of layout area, shape (4, 2), dtype int16; each row represents one rectangle</li>
+        <li><code>block_label</code>: <code>(str)</code> Label of layout area, such as <code>text</code>, <code>table</code> etc.</li>
+        <li><code>block_content</code>: <code>(str)</code> Content of layout area, such as text, table, formula etc.</li>
+        <li><code>block_id</code>: <code>(int)</code> Index of layout area, used to display layout sorting results.</li>
+        <li><code>block_order</code> <code>(int)</code> Order of layout area, used to display layout reading order, default value is <code>None</code> for non-ordered parts.</li>
+        </ol>
+    </li>
+    <li><code>overall_ocr_res</code>: <code>(Dict[str, Union[List[str], List[float], numpy.ndarray]])</code> Global OCR result dictionary, present only when <code>use_general_ocr=True</code>
+        <ol>
+        <li><code>input_path</code>: <code>(Union[str, None])</code> Input path of OCR sub-pipeline, when input is <code>numpy.ndarray</code>, saved as <code>None</code></li>
+        <li><code>page_index</code>: <code>(None)</code> Page index of OCR sub-pipeline, when input is <code>numpy.ndarray</code>, saved as <code>None</code></li>
+        <li><code>model_settings</code>: <code>(Dict)</code> Model configuration parameters of OCR sub-pipeline</li>
+        <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> List of text detection polygons. Each detection box is represented by a numpy array of shape (4, 2), dtype int16, which contains the coordinates of the 4 vertices of the rectangle</li> 
+        <li><code>dt_scores</code>: <code>(List[float])</code> List of confidence scores for text detection boxes</li>    
+        <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> Configuration parameters of the text detection module
+            <ol>
+            <li><code>limit_side_len</code>: <code>(int)</code> Side length limit value for image preprocessing</li>
+            <li><code>limit_type</code>: <code>(str)</code> Side length limit handling method</li>
+            </ol>
+        <li><code>thresh</code>: <code>(float)</code> Text pixel classification confidence threshold</li>
+        <li><code>box_thresh</code>: <code>(float)</code> Confidence threshold for text detection boxes</li>
+        <li><code>unclip_ratio</code>: <code>(float)</code> Expansion factor for text detection boxes</li>
+        <li><code>text_type</code>: <code>(str)</code> Type of text detection, currently fixed as "general"</li>
+        <li><code>textline_orientation_angles</code>: <code>(List[int])</code> Text line direction classification prediction results. Returns actual angle value when enabled, such as [0,0,1]</li>
+        <li><code>text_rec_score_thresh</code>: <code>(float)</code> Text recognition result filtering threshold</li>
+        <li><code>rec_texts</code>: <code>(List[str])</code> Text recognition result list, only contains texts with confidence higher than <code>text_rec_score_thresh</code></li>
+        <li><code>rec_scores</code>: <code>(List[float])</code> Text recognition confidence list, filtered by <code>text_rec_score_thresh</code></li>
+        <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> List of text detection boxes filtered by confidence, format same as <code>dt_polys</code></li>
+         </ol>  
+    </li>
+    <li><code>formula_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> List of formula recognition results, each element is a dictionary
+        <ol>
+        <li><code>rec_formula</code>: <code>(str)</code> Formula recognition result</li>
+        <li><code>rec_polys</code>: <code>(numpy.ndarray)</code> Formula detection boxes, shape (4, 2), dtype int16</li>
+        <li><code>formula_region_id</code>: <code>(int)</code> Region id of formula, used to display formula sorting results.</li>
+        </ol>
+    </li>
+    <li><code>seal_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> List of seal recognition results, each element is a dictionary
+        <ol>
+        <li><code>input_path</code>: <code>(str)</code> Input path of seal image</li>
+        <li><code>page_index</code>: <code>(None)</code> Page index of seal image, when input is <code>numpy.ndarray</code>, saved as <code>None</code></li>        
+        <li><code>model_settings</code>: <code>(Dict)</code> Model configuration parameters of seal recognition sub-pipeline</li>
+        <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> Seal detection boxes list, format same as <code>dt_polys</code></li>
+        <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> Configuration parameters of seal detection module, same as <code>text_det_params</code></li>
+        <li><code>text_type</code>: <code>(str)</code> Type of seal detection, currently fixed as "seal"</li>
+        <li><code>text_rec_score_thresh</code>: <code>(float)</code> Seal recognition result filtering threshold</li>
+        <li><code>rec_texts</code>: <code>(List[str])</code> List of seal recognition results, only contains texts with confidence higher than <code>text_rec_score_thresh</code></li>
+        <li><code>rec_scores</code>: <code>(List[float])</code> List of confidence scores for seal recognition results, filtered by <code>text_rec_score_thresh</code></li>
+        <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> List of seal detection boxes filtered by confidence, format same as <code>dt_polys</code></li>
+        <li> <code>rec_boxes</code>: <code>(numpy.ndarray)</code> Rectangle bounding boxes of seal detection boxes, shape (n, 4), dtype int16. Each row represents a rectangle</li>
+        </ol>
+    </li>
+    <li><code>table_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> List of table recognition results, each element is a dictionary
+        <ol>
+        <li><code>table_id</code>: <code>(int)</code> Index of table, used to display table sorting results.</li>
+        <li><code>cell_box_list</code>: <code>(List[numpy.ndarray])</code> List of table cell detection boxes, format same as <code>dt_polys</code></li>
+        <li><code>pred_html</code>: <code>(str)</code> HTML format string of table</li>
+        <li><code>table_ocr_pred</code>: <code>(dict)</code> OCR recognition results of table
+            <ul>
+            <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> List of table cell detection boxes, format same as <code>dt_polys</code></li>
+            <li><code>rec_texts</code>: <code>(List[str])</code> List of table cell recognition results</li>
+            <li><code>rec_scores</code>: <code>(List[float])</code> List of confidence scores for table cell recognition results</li>
+            <li><code>rec_boxes</code>: <code>(numpy.ndarray)</code> Rectangle bounding boxes of table cell detection boxes, shape (n, 4), dtype int16. Each row represents a rectangle
+            </ul>
+            </li>
+        </li>
+    </li>
+    </ol>
+</li>
+<li>Calling the<code>save_to_json()</code> method will save the above content to the specified <code>save_path</code> .If a directory is specified, the saved path will be <code>save_path/{your_img_basename}_res.json</code>，If a file is specified, it will be saved directly to that file. Since JSON files do not support saving numpy arrays, all <code>numpy.array</code> types will be converted to list format.</li>
+<li>Calling the <code>save_to_img()</code> method will save visualization results to the specified <code>save_path</code>. If a directory is specified, it will save layout detection visual images, global OCR visual images, layout reading order visual images, etc. If a file is specified, it will be saved directly to that file. (The pipeline usually contains many result images, so it is not recommended to specify a specific file path directly, or multiple images will be overwritten, leaving only the last image.)</li>
+<li>Calling the <code>save_to_markdown()</code> method will save the converted Markdown files to the specified <code>save_path</code>. The saved file path will be <code>save_path/{your_img_basename}.md</code>. If the input is a PDF file, it is recommended to specify a directory directly, otherwise multiple markdown files will be overwritten.</li>
+<li>Calling the <code>concatenate_markdown_pages()</code> method merges the multi-page Markdown contents output by the PP-DocTranslation pipeline <code>markdown_list</code> into a single complete document and returns the merged Markdown content.</li>
+</ul>
+</details>
+
+<details><summary>(4) Call the <code>translate()</code> method to perform document translation. This method returns the original and translated markdown content as a markdown object, which can be saved locally by executing the <code>save_to_markdown()</code> method for the desired parts. Below are the relevant parameters of the <code>translate()</code> method:</summary>
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>ori_md_info_list</code></td>
+<td>List of original Markdown data containing content to be translated. Must be a list of dictionaries, each representing a document block</td>
+<td><code>List[Dict]</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>target_language</code></td>
+<td>Target language (ISO 639-1 language code, e.g. <code>"en"</code>/<code>"ja"</code>/<code>"fr"</code>)</td>
+<td><code>str</code></td>
+<td><code>"zh"</code></td>
+</tr>
+<tr>
+<td><code>chunk_size</code></td>
+<td>Character count threshold for chunked translation processing</td>
+<td><code>int</code></td>
+<td><code>5000</code></td>
+</tr>
+<tr>
+<td><code>task_description</code></td>
+<td>Custom task description prompt</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>output_format</code></td>
+<td>Specified output format requirements, e.g. "preserve original Markdown structure"</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>rules_str</code></td>
+<td>Custom translation rule description</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>few_shot_demo_text_content</code></td>
+<td>Few-shot learning example text content</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>few_shot_demo_key_value_list</code></td>
+<td>Structured few-shot example data in key-value pairs, can include professional terminology glossary</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>glossary</code></td>
+<td>Professional terminology glossary for translation</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>llm_request_interval</code></td>
+<td>Interval in seconds between requests to the large language model. This parameter helps prevent too frequent calls to the LLM.</td>
+<td><code>float</code></td>
+<td><code>0.0</code></td>
+</tr>
+<tr>
+<td><code>chat_bot_config</code></td>
+<td>Large language model configuration. Setting to <code>None</code> uses instantiation parameters; otherwise, this parameter takes priority.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+</details>
+
+## 3. Development Integration/Deployment
+
+If the pipeline can meet your requirements for inference speed and accuracy, you can proceed directly with development integration/deployment.
+
+If you need to directly apply the pipeline in your Python project, you can refer to the sample code in [2.2 Python Script Approach](#22-integration-via-python-script).
+
+In addition, PaddleOCR also offers two other deployment methods, detailed as follows:
+
+🚀 High-Performance Inference: In real-world production environments, many applications have stringent performance criteria (especially response speed) for deployment strategies to ensure efficient system operation and a smooth user experience. To this end, PaddleOCR provides high-performance inference capabilities, aiming to deeply optimize model inference and pre/post-processing, achieving significant acceleration in the end-to-end process. For detailed information on the high-performance inference process, please refer to [High-Performance Inference](../inference_deployment/local_inference/high_performance_inference.en.md).
+
+☁️ Serving: Serving is a common deployment form in real-world production environments. By encapsulating inference functions as services, clients can access these services through network requests to obtain inference results. For detailed information on the pipeline serving process, please refer to [Serving](../inference_deployment/serving/serving.en.md).
+
+Below are the API references for basic serving and examples of multi-language service invocation:
+
+<details><summary>API Reference</summary>
+<p>Main operations provided by the serving:</p>
+<ul>
+<li>HTTP request method is POST.</li>
+<li>Both request body and response body are JSON data (JSON objects).</li>
+<li>When the request is processed successfully, the response status code is <code>200</code>, and the response body has the following properties:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>Request UUID.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Fixed as <code>0</code>.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error message. Fixed as <code>"Success"</code>.</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>Operation result.</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is not successful, the response body has the following properties:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>Request UUID.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Same as response status code.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error message.</td>
+</tr>
+</tbody>
+</table>
+<p>Main operations provided by the serving are as follows:</p>
+<ul>
+<li><b><code>analyzeImages</code></b></li>
+</ul>
+<p>Use computer vision models to analyze images, obtaining OCR, table recognition results, etc.</p>
+<p><code>POST /doctrans-visual</code></p>
+<ul>
+<li>Request body properties are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+<th>Required</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>URL of image or PDF file accessible by the server, or Base64 encoding of such file contents. By default, for PDF files over 10 pages, only the first 10 pages are processed.<br /> To remove the page limit, add the following configuration in the pipeline config file:
+<pre><code>Serving:
+  extra:
+    max_num_input_imgs: null
+</code></pre>
+</td>
+<td>Yes</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code>｜<code>null</code></td>
+<td>File type. <code>0</code> means PDF, <code>1</code> means image file. If not present in the request, the file type will be inferred from the URL.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>See the <code>use_doc_orientation_classify</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>See the <code>use_doc_unwarping</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useTextlineOrientation</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>See the <code>use_textline_orientation</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useSealRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>See the <code>use_seal_recognition</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useTableRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>See the <code>use_table_recognition</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useFormulaRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>See the <code>use_formula_recognition</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useChartRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>See the <code>use_chart_recognition</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useRegionDetection</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>See the <code>use_region_detection</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutThreshold</code></td>
+<td><code>number</code> | <code>object</code> | <code>null</code></td>
+<td>See the <code>layout_threshold</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutNms</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>See the <code>layout_nms</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutUnclipRatio</code></td>
+<td><code>number</code> | <code>array</code> | <code>object</code> | <code>null</code></td>
+<td>See the <code>layout_unclip_ratio</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutMergeBboxesMode</code></td>
+<td><code>string</code> | <code>object</code> | <code>null</code></td>
+<td>See the <code>layout_merge_bboxes_mode</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>See the <code>text_det_limit_side_len</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>See the <code>text_det_limit_type</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>See the <code>text_det_thresh</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>See the <code>text_det_box_thresh</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>See the <code>text_det_unclip_ratio</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>See the <code>text_rec_score_thresh</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>See the <code>seal_det_limit_side_len</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>See the <code>seal_det_limit_type</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>See the <code>seal_det_thresh</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>See the <code>seal_det_box_thresh</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>See the <code>seal_det_unclip_ratio</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>See the <code>seal_rec_score_thresh</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useWiredTableCellsTransToHtml</code></td>
+<td><code>boolean</code></td>
+<td>See the <code>use_wired_table_cells_trans_to_html</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useWirelessTableCellsTransToHtml</code></td>
+<td><code>boolean</code></td>
+<td>See the <code>use_wireless_table_cells_trans_to_html</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useTableOrientationClassify</code></td>
+<td><code>boolean</code></td>
+<td>See the <code>use_table_orientation_classify</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useOcrResultsWithTableCells</code></td>
+<td><code>boolean</code></td>
+<td>See the <code>use_ocr_results_with_table_cells</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useE2eWiredTableRecModel</code></td>
+<td><code>boolean</code></td>
+<td>See the <code>use_e2e_wired_table_rec_model</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useE2eWirelessTableRecModel</code></td>
+<td><code>boolean</code></td>
+<td>See the <code>use_e2e_wireless_table_rec_model</code> parameter description in the pipeline object's <code>visual_predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>outputFormats</code></td>
+<td><code>array</code> | <code>null</code></td>
+<td>Optional. Additional export formats; currently only <code>"docx"</code> is supported.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Whether to return visualization result images and intermediate images during processing.
+<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>If <code>true</code> is passed: return images.</li>
+<li>If <code>false</code> is passed: do not return images.</li>
+<li>If this parameter is not provided in the request body or <code>null</code> is passed: follow the pipeline config file setting <code>Serving.visualize</code>.</li>
+</ul>
+<br/>For example, add the following field in the pipeline config file:<br/>
+<pre><code>Serving:
+  visualize: False
+</code></pre>
+By default, images will not be returned; the <code>visualize</code> parameter in the request body can override this default behavior. If neither the request body nor the config file sets it (or the request body passes <code>null</code> and the config file does not set it), images will be returned by default.
+</td>
+<td>No</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is processed successfully, the response body's <code>result</code> has the following properties:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>layoutParsingResults</code></td>
+<td><code>array</code></td>
+<td>Layout parsing results. The array length is 1 (for image input) or the number of document pages actually processed (for PDF input). For PDF input, each element in the array is the result of each processed page, in order.</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>Metadata about the input data.</td>
+</tr>
+</tbody>
+</table>
+<p>Image and other binary file fields in the element schema below (e.g. <code>outputImages</code>, <code>inputImage</code>, <code>markdown.images</code>, <code>exports</code>) are returned inline as Base64 strings by default; when the server is configured to return URLs, those values become pre-signed URLs while the field types remain unchanged. See the "Returning Binary Content as URLs" section of the <a href="../inference_deployment/serving/serving.en.md">Serving Deployment Guide</a> for configuration.</p>
+<p>Each element in <code>layoutParsingResults</code> is an <code>object</code> with the following properties:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>Simplified version of the <code>res</code> field in the JSON representation of the <code>layout_parsing_result</code> generated by the pipeline object's <code>visual_predict</code> method, with <code>input_path</code> and <code>page_index</code> fields removed.</td>
+</tr>
+<tr>
+<td><code>markdown</code></td>
+<td><code>object</code></td>
+<td>Markdown result.</td>
+</tr>
+<tr>
+<td><code>outputImages</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>See the <code>img</code> property description in the pipeline prediction results. Images are in JPEG format and Base64 encoded by default; returned as pre-signed URLs when URL-return mode is enabled.</td>
+</tr>
+<tr>
+<td><code>inputImage</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Input image. JPEG format, Base64 encoded by default; returned as a pre-signed URL when URL-return mode is enabled.</td>
+</tr>
+<tr>
+<td><code>exports</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Optional additional exports such as <code>docx</code>, present only when <code>outputFormats</code> is requested. The <code>content</code> field is Base64-encoded file bytes by default; returned as a pre-signed URL when URL-return mode is enabled.</td>
+</tr>
+</tbody>
+</table>
+<p><code>markdown</code> is an <code>object</code> with the following properties:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>text</code></td>
+<td><code>string</code></td>
+<td>Markdown text.</td>
+</tr>
+<tr>
+<td><code>images</code></td>
+<td><code>object</code></td>
+<td>Mapping from relative Markdown image paths to Base64-encoded image data by default; returned as pre-signed URLs when URL-return mode is enabled.</td>
+</tr>
+<tr>
+<td><code>isStart</code></td>
+<td><code>boolean</code></td>
+<td>Whether the first element on the current page starts a paragraph.</td>
+</tr>
+<tr>
+<td><code>isEnd</code></td>
+<td><code>boolean</code></td>
+<td>Whether the last element on the current page ends a paragraph.</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li><b><code>translate</code></b></li>
+</ul>
+<p>Use a large model to translate documents.</p>
+<p><code>POST /doctrans-translate</code></p>
+<ul>
+<li>Request body properties are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+<th>Required</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>markdownList</code></td>
+<td><code>array</code></td>
+<td>List of Markdown to be translated. Can be obtained from the results of the <code>analyzeImages</code> operation. The <code>images</code> attribute will not be used.</td>
+<td>Yes</td>
+</tr>
+<tr>
+<td><code>targetLanguage</code></td>
+<td><code>string</code></td>
+<td>Please refer to the <code>target_language</code> parameter description in the <code>translate</code> method of the pipeline object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>chunkSize</code></td>
+<td><code>integer</code></td>
+<td>Please refer to the <code>chunk_size</code> parameter description in the <code>translate</code> method of the pipeline object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>taskDescription</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the <code>task_description</code> parameter description in the <code>translate</code> method of the pipeline object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>outputFormat</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the <code>output_format</code> parameter description in the <code>translate</code> method of the pipeline object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>rulesStr</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the <code>rules_str</code> parameter description in the <code>translate</code> method of the pipeline object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>fewShotDemoTextContent</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the <code>few_shot_demo_text_content</code> parameter description in the <code>translate</code> method of the pipeline object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>fewShotDemoKeyValueList</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the <code>few_shot_demo_key_value_list</code> parameter description in the <code>translate</code> method of the pipeline object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>glossary</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Please refer to the <code>glossary</code> parameter description in the <code>translate</code> method of the pipeline object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>llmRequestInterval</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the <code>llm_request_interval</code> parameter description in the <code>translate</code> method of the pipeline object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>chatBotConfig</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Please refer to the <code>chat_bot_config</code> parameter description in the <code>translate</code> method of the pipeline object.</td>
+<td>No</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is successfully processed, the <code>result</code> in the response body has the following attributes:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>translationResults</code></td>
+<td><code>array</code></td>
+<td>Translation results.</td>
+</tr>
+</tbody>
+</table>
+<p>Each element in <code>translationResults</code> is an <code>object</code> with the following attributes:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>language</code></td>
+<td><code>string</code></td>
+<td>Target language.</td>
+</tr>
+<tr>
+<td><code>markdown</code></td>
+<td><code>object</code></td>
+<td>Markdown result; same shape as <code>markdown</code> returned by <code>analyzeImages</code>.</td>
+</tr>
+</tbody>
+</table>
+
+<li><b>Note: </b></li>Including sensitive parameters such as the API key for large model calls in the request body may pose security risks. If not necessary, set these parameters in the configuration file and do not pass them during the request.<br/><br/>
+</details>
+<details><summary>Examples of multi-language service invocation</summary>
+<details>
+<summary>Python</summary>
+<pre><code class="language-python">import base64
+import pathlib
+import pprint
+import sys
+
+import requests
+
+
+API_BASE_URL = "http://127.0.0.1:8080"
+
+file_path = "./demo.jpg"
+target_language = "en"
+
+with open(file_path, "rb") as file:
+    file_bytes = file.read()
+    file_data = base64.b64encode(file_bytes).decode("ascii")
+
+payload = {
+    "file": file_data,
+    "fileType": 1,
+}
+resp_visual = requests.post(url=f"{API_BASE_URL}/doctrans-visual", json=payload)
+if resp_visual.status_code != 200:
+    print(
+        f"Request to doctrans-visual failed with status code {resp_visual.status_code}."
+    )
+    pprint.pp(resp_visual.json())
+    sys.exit(1)
+result_visual = resp_visual.json()["result"]
+
+markdown_list = []
+for i, res in enumerate(result_visual["layoutParsingResults"]):
+    md_dir = pathlib.Path(f"markdown_{i}")
+    md_dir.mkdir(exist_ok=True)
+    (md_dir / "doc.md")
+write_text(res["markdown"]["text"])
+    for img_path, img in res["markdown"]["images"].items():
+        img_path = md_dir / img_path
+        img_path.parent.mkdir(parents=True, exist_ok=True)
+        img_path.write_bytes(base64.b64decode(img))
+    print(f"The Markdown document to be translated is saved at {md_dir / 'doc.md'}")
+    del res["markdown"]["images"]
+    markdown_list.append(res["markdown"])
+    for img_name, img in res["outputImages"].items():
+        img_path = f"{img_name}_{i}.jpg"
+        with open(img_path, "wb") as f:
+            f.write(base64.b64decode(img))
+        print(f"Output image saved at {img_path}")
+
+payload = {
+    "markdownList": markdown_list,
+"targetLanguage": target_language,
+}
+resp_translate = requests.post(url=f"{API_BASE_URL}/doctrans-translate", json=payload)
+if resp_translate.status_code != 200:
+    print(
+        f"Request to doctrans-translate failed with status code {resp_translate.status_code}."
+    )
+    pprint.pprint(resp_translate.json())  # Corrected 'pp' to 'pprint' for proper function call
+    sys.exit(1)
+result_translate = resp_translate.json()["result"]
+
+for i, res in enumerate(result_translate["translationResults"]):
+    md_dir = pathlib.Path(f"markdown_{i}")
+    (md_dir / "doc_translated.md").write_text(res["markdown"]["text"])
+    print(f"Translated markdown document saved at {md_dir / 'doc_translated.md'}")</code></pre></details>
+</details>
+<br/>
+
+## 4. Secondary Development
+If the default model weights provided by the PP-DocTranslation pipeline do not meet your accuracy or speed requirements in your scenario, you can try to use<b>your own data from specific domains or application scenarios</b>to further<b>fine-tune</b>the existing model to improve the recognition effect in your scenario.
+
+### 4.1 Model Fine-tuning
+Since the PP-DocTranslation pipeline contains several modules, if the performance of the model pipeline does not meet expectations, the issue may originate from any one of these modules. You can analyze cases with poor extraction results, use visualized images to determine which module has the problem, and refer to the corresponding fine-tuning tutorial links in the following table to fine-tune the model.
+
+<table>
+<thead>
+<tr>
+<th>Scenario</th>
+<th>Fine-tuning module</th>
+<th>Fine-tuning reference link</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>Inaccurate detection of layout areas, such as failure to detect seals and tables</td>
+<td>Layout detection module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/layout_detection.html#_5">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate recognition of table structures</td>
+<td>Table structure recognition module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/table_structure_recognition.html#_5">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate recognition of formulas</td>
+<td>Formula recognition module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/formula_recognition.html#_5">Link</a></td>
+</tr>
+<tr>
+<td>Omission in detecting seal texts</td>
+<td>Seal text detection module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/seal_text_detection.html#_5">Link</a></td>
+</tr>
+<tr>
+<td>Omission in detecting texts</td>
+<td>Text detection module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/text_detection.html#_5">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate text content</td>
+<td>Text recognition module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/text_recognition.html#_5">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate correction of vertical or rotated text lines</td>
+<td>Text line orientation classification module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/textline_orientation_classification.html#_5">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate correction of whole image rotation</td>
+<td>Document image orientation classification module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#_5">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate correction of image distortion</td>
+<td>Text image unwarping module</td>
+<td>Fine-tuning is temporarily not supported</td>
+</tr>
+</tbody>
+</table>
+
+### 4.2 Model Application
+After completing fine-tuning training with your private dataset, you can obtain a local model weight file. Then, you can use the fine-tuned model weights by customizing the pipeline configuration file.
+
+1. Obtain the pipeline configuration file
+
+You can call the `export_paddlex_config_to_yaml` method of the PP-DocTranslation pipeline object in PaddleOCR to export the current pipeline configuration to a YAML file:
+
+```Python
+from paddleocr import PPDocTranslation
+
+pipeline = PPDocTranslation()
+pipeline.export_paddlex_config_to_yaml("PP-DocTranslation.yaml")
+```
+
+2. Modify the configuration file
+
+After obtaining the default pipeline configuration file, replace the local path of the fine-tuned model weights with the corresponding location in the pipeline configuration file. For example,
+
+```yaml
+......
+SubModules:
+    TextDetection:
+    module_name: text_detection
+    model_name: PP-OCRv5_server_det
+    model_dir: null # Replace with the path to the weights of the fine-tuned text detection model
+    limit_side_len: 960
+    limit_type: max
+    thresh: 0.3
+    box_thresh: 0.6
+    unclip_ratio: 1.5
+
+    TextRecognition:
+    module_name: text_recognition
+    model_name: PP-OCRv5_server_rec
+    model_dir: null # Replace with the path to the weights of the fine-tuned text recognition model
+    batch_size: 1
+            score_thresh: 0
+......
+```
+
+The pipeline configuration file not only includes parameters supported by PaddleOCR CLI and Python API but also allows for more advanced configurations. Detailed information can be found in the corresponding pipeline usage tutorial in the [Overview of PaddleX Model Pipeline Usage](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/pipeline_develop_guide.html). Refer to the detailed instructions therein and adjust the configurations according to your needs.
+
+3. Load the pipeline configuration file in CLI
+
+After modifying the configuration file, specify the path to the modified pipeline configuration file using the `--paddlex_config` parameter in the command line. PaddleOCR will then read its contents as the pipeline configuration. Here is an example:
+
+```bash
+paddleocr pp_doctranslation --paddlex_config PP-DocTranslation.yaml ...
+```
+
+4. Load the pipeline configuration file in the Python API
+
+When initializing the pipeline object, you can pass the path of the PaddleX pipeline configuration file or a configuration dict through the `paddlex_config` parameter, and PaddleOCR will read its content as the pipeline configuration. The example is as follows:
+
+```python
+from paddleocr import PPDocTranslation
+
+pipeline = PPDocTranslation(paddlex_config="PP-DocTranslation.yaml")
+```
diff --git a/docs/version3.x/pipeline_usage/PP-DocTranslation.md b/docs/version3.x/pipeline_usage/PP-DocTranslation.md
new file mode 100644
index 0000000..0f7c2b7
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PP-DocTranslation.md
@@ -0,0 +1,3458 @@
+---
+comments: true
+---
+
+# PP-DocTranslation 产线使用教程
+
+## 1. PP-DocTranslation 产线介绍
+
+PP-DocTranslation 是飞桨提供的文档智能翻译解决方案，融合了先进的通用版面解析技术与大语言模型（LLM）能力，为您提供高效的文档智能翻译服务。该解决方案能够精准识别并提取文档中的各类元素，包括文本块、标题、段落、图片、表格及其他复杂版面结构，并在此基础之上实现高质量的多语种互译。PP-DocTranslation 支持多种主流语言间的相互翻译，尤其擅长处理排版复杂、上下文依赖性强的文档场景，力求输出精准自然、流畅专业的翻译结果。本产线同时提供了灵活的服务化部署方式，支持在多种硬件上使用多种编程语言调用。不仅如此，本产线也提供了二次开发的能力，您可以基于本产线在您自己的数据集上训练调优，训练后的模型也可以无缝集成。
+
+<b>PP-DocTranslation 产线中使用了 PP-StructureV3 子产线，因此具有 PP-StructureV3 产线的所有功能，更多关于 PP-StructureV3 产线的功能介绍和使用细节，可以点击 [PP-StructureV3 产线文档](./PP-StructureV3.md) 页面查看</b>。
+
+在本产线中，您可以根据下方的基准测试数据选择使用的模型。
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。
+> 在带有 [常规模式 / 高性能模式] 标记的推理耗时列中，`常规模式` 对应本地飞桨推理引擎。各模块会根据默认模型名称选择合适的本地飞桨推理引擎：仅支持动态图的模型使用 `paddle_dynamic`；同时支持静态图和动态图的模型优先使用 `paddle_static`。
+
+<details><summary> 👉模型列表详情</summary>
+<p><b>文档图像方向分类模块：</b></p>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>Top-1 Acc（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（M）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">训练模型</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>基于PP-LCNet_x1_0的文档图像分类模型，含有四个类别，即0度，90度，180度，270度</td>
+</tr>
+</tbody>
+</table>
+<p><b>文本图像矫正模块：</b></p>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>CER </th>
+<th>模型存储大小（M）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">训练模型</a></td>
+<td>0.179</td>
+<td>30.3</td>
+<td>高精度文本图像矫正模型</td>
+</tr>
+</tbody>
+</table>
+<p><b>版面区域检测模块模型：</b></p>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（M）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout_plus-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout_plus-L_pretrained.pdparams">训练模型</a></td>
+<td>83.2</td>
+<td>53.03 / 17.23</td>
+<td>634.62 / 378.32</td>
+<td>126.01</td>
+<td>基于RT-DETR-L在包含中英文论文、多栏杂志、报纸、PPT、合同、书本、试卷、研报、古籍、日文文档、竖版文字文档等场景的自建数据集训练的更高精度版面区域定位模型</td>
+</tr>
+<td>PP-DocLayout-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-L_pretrained.pdparams">训练模型</a></td>
+<td>90.4</td>
+<td>33.59 / 33.59</td>
+<td>503.01 / 251.08</td>
+<td>123.76</td>
+<td>基于RT-DETR-L在包含中英文论文、杂志、合同、书本、试卷和研报等场景的自建数据集训练的高精度版面区域定位模型</td>
+</tr>
+<tr>
+<td>PP-DocLayout-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-M_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-M_pretrained.pdparams">训练模型</a></td>
+<td>75.2</td>
+<td>13.03 / 4.72</td>
+<td>43.39 / 24.44</td>
+<td>22.578</td>
+<td>基于PicoDet-L在包含中英文论文、杂志、合同、书本、试卷和研报等场景的自建数据集训练的精度效率平衡的版面区域定位模型</td>
+</tr>
+<tr>
+<td>PP-DocLayout-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-S_pretrained.pdparams">训练模型</a></td>
+<td>70.9</td>
+<td>11.54 / 3.86</td>
+<td>18.53 / 6.29</td>
+<td>4.834</td>
+<td>基于PicoDet-S在中英文论文、杂志、合同、书本、试卷和研报等场景上自建数据集训练的高效率版面区域定位模型</td>
+</tr>
+</tbody>
+</table>
+<p><b>表格结构识别模块：</b></p>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>精度（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小 (M)</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>SLANeXt_wired</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wired_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wired_pretrained.pdparams">训练模型</a></td>
+<td rowspan="2">69.65</td>
+<td rowspan="2">85.92 / 85.92</td>
+<td rowspan="2">- / 501.66</td>
+<td rowspan="2">351M</td>
+<td rowspan="2">SLANeXt 系列是百度飞桨视觉团队自研的新一代表格结构识别模型。相较于 SLANet 和 SLANet_plus，SLANeXt 专注于对表格结构进行识别，并且对有线表格(wired)和无线表格(wireless)的识别分别训练了专用的权重，对各类型表格的识别能力都得到了明显提高，特别是对有线表格的识别能力得到了大幅提升。</td>
+</tr>
+<tr>
+<td>SLANeXt_wireless</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wireless_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wireless_pretrained.pdparams">训练模型</a></td>
+</tr>
+</table>
+
+<p><b>表格分类模块模型：</b></p>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>Top1 Acc(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小 (M)</th>
+</tr>
+<tr>
+<td>PP-LCNet_x1_0_table_cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/CLIP_vit_base_patch16_224_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_table_cls_pretrained.pdparams">训练模型</a></td>
+<td>94.2</td>
+<td>2.62 / 0.60</td>
+<td>3.17 / 1.14</td>
+<td>6.6M</td>
+</tr>
+</table>
+
+<p><b>表格单元格检测模块模型：</b></p>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小 (M)</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>RT-DETR-L_wired_table_cell_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wired_table_cell_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wired_table_cell_det_pretrained.pdparams">训练模型</a></td>
+<td rowspan="2">82.7</td>
+<td rowspan="2">33.47 / 27.02</td>
+<td rowspan="2">402.55 / 256.56</td>
+<td rowspan="2">124M</td>
+<td rowspan="2">RT-DETR 是第一个实时的端到端目标检测模型。百度飞桨视觉团队基于 RT-DETR-L 作为基础模型，在自建表格单元格检测数据集上完成预训练，实现了对有线表格、无线表格均有较好性能的表格单元格检测。
+</td>
+</tr>
+<tr>
+<td>RT-DETR-L_wireless_table_cell_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wireless_table_cell_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wireless_table_cell_det_pretrained.pdparams">训练模型</a></td>
+</tr>
+</table>
+
+<p><b>文本检测模块：</b></p>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>检测Hmean（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（M)</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv5_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_det_pretrained.pdparams">训练模型</a></td>
+<td>83.8</td>
+<td>89.55 / 70.19</td>
+<td>383.15 / 383.15</td>
+<td>84.3</td>
+<td>PP-OCRv5 的服务端文本检测模型，精度更高，适合在性能较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_det_pretrained.pdparams">训练模型</a></td>
+<td>79.0</td>
+<td>10.67 / 6.36</td>
+<td>57.77 / 28.15</td>
+<td>4.7</td>
+<td>PP-OCRv5 的移动端文本检测模型，效率更高，适合在端侧设备部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_det_pretrained.pdparams">训练模型</a></td>
+<td>69.2</td>
+<td>127.82 / 98.87</td>
+<td>585.95 / 489.77</td>
+<td>109</td>
+<td>PP-OCRv4 的服务端文本检测模型，精度更高，适合在性能较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams">训练模型</a></td>
+<td>63.8</td>
+<td>9.87 / 4.17</td>
+<td>56.60 / 20.79</td>
+<td>4.7</td>
+<td>PP-OCRv4 的移动端文本检测模型，效率更高，适合在端侧设备部署</td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams">训练模型</a></td>
+<td>精度接近 PP-OCRv4_mobile_det</td>
+<td>9.90 / 3.60</td>
+<td>41.93 / 20.76</td>
+<td>2.1</td>
+<td>PP-OCRv3 的移动端文本检测模型，效率更高，适合在端侧设备部署</td>
+</tr>
+<tr>
+<td>PP-OCRv3_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_server_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_server_det_pretrained.pdparams">训练模型</a></td>
+<td>精度接近 PP-OCRv4_server_det</td>
+<td>119.50 / 75.00</td>
+<td>379.35 / 318.35</td>
+<td>102.1</td>
+<td>PP-OCRv3 的服务端文本检测模型，精度更高，适合在性能较好的服务器上部署</td>
+</tr>
+</tbody>
+</table>
+
+<p><b>文本识别模块模型：</b></p>
+
+* <b>中文识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（M）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>86.38</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec 是新一代文本识别模型。该模型致力于以单一模型高效、精准地支持简体中文、繁体中文、英文、日文四种主要语言，以及手写、竖版、拼音、生僻字等复杂文本场景的识别。在保持识别效果的同时，兼顾推理速度和模型鲁棒性，为各种场景下的文档理解提供高效、精准的技术支撑。</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.29</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">训练模型</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>74.7</td>
+<td>PP-OCRv4_server_rec_doc是在PP-OCRv4_server_rec的基础上，在更多中文文档数据和PP-OCR训练数据的混合数据训练而成，增加了部分繁体字、日文、特殊字符的识别能力，可支持识别的字符为1.5万+，除文档相关的文字识别能力提升外，也同时提升了通用文字的识别能力</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.6</td>
+<td>PP-OCRv4的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec </td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>80.61 </td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>71.2</td>
+<td>PP-OCRv4的服务器端模型，推理精度高，可以部署在多种不同的服务器上</td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>72.96</td>
+<td>3.89 / 1.16</td>
+<td>8.72 / 3.56</td>
+<td>9.2</td>
+<td>PP-OCRv3的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（M）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>ch_SVTRv2_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_SVTRv2_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_SVTRv2_rec_pretrained.pdparams">训练模型</a></td>
+<td>68.81</td>
+<td>10.38 / 8.31</td>
+<td>66.52 / 30.83</td>
+<td>73.9</td>
+<td rowspan="1">
+SVTRv2 是一种由复旦大学视觉与学习实验室（FVL）的OpenOCR团队研发的服务端文本识别模型，其在PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务中荣获一等奖，A榜端到端识别精度相比PP-OCRv4提升6%。
+</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（M）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>ch_RepSVTR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_RepSVTR_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_RepSVTR_rec_pretrained.pdparams">训练模型</a></td>
+<td>65.07</td>
+<td>6.29 / 1.57</td>
+<td>20.64 / 5.40</td>
+<td>22.1</td>
+<td rowspan="1">    RepSVTR 文本识别模型是一种基于SVTRv2 的移动端文本识别模型，其在PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务中荣获一等奖，B榜端到端识别精度相比PP-OCRv4提升2.5%，推理速度持平。</td>
+</tr>
+</table>
+
+* <b>英文识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（M）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td> 70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>6.8</td>
+<td>基于PP-OCRv4识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+<tr>
+<td>en_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>70.69</td>
+<td>3.56 / 0.78</td>
+<td>8.44 / 5.78</td>
+<td>7.8 M </td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+</table>
+
+* <b>多语言识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（M）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>korean_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>60.21</td>
+<td>3.73 / 0.98</td>
+<td>8.76 / 2.91</td>
+<td>8.6</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量韩文识别模型，支持韩文、数字识别</td>
+</tr>
+<tr>
+<td>japan_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+japan_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/japan_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>45.69</td>
+<td>3.86 / 1.01</td>
+<td>8.62 / 2.92</td>
+<td>8.8 M </td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量日文识别模型，支持日文、数字识别</td>
+</tr>
+<tr>
+<td>chinese_cht_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+chinese_cht_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/chinese_cht_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>82.06</td>
+<td>3.90 / 1.16</td>
+<td>9.24 / 3.18</td>
+<td>9.7 M </td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量繁体中文识别模型，支持繁体中文、数字识别</td>
+</tr>
+<tr>
+<td>te_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+te_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>95.88</td>
+<td>3.59 / 0.81</td>
+<td>8.28 / 6.21</td>
+<td>7.8 M </td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量泰卢固文识别模型，支持泰卢固文、数字识别</td>
+</tr>
+<tr>
+<td>ka_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ka_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ka_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>96.96</td>
+<td>3.49 / 0.89</td>
+<td>8.63 / 2.77</td>
+<td>8.0 M </td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量卡纳达文识别模型，支持卡纳达文、数字识别</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ta_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>76.83</td>
+<td>3.49 / 0.86</td>
+<td>8.35 / 3.41</td>
+<td>8.0 M </td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量泰米尔文识别模型，支持泰米尔文、数字识别</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>76.93</td>
+<td>3.53 / 0.78</td>
+<td>8.50 / 6.83</td>
+<td>7.8</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量拉丁文识别模型，支持拉丁文、数字识别</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+arabic_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>73.55</td>
+<td>3.60 / 0.83</td>
+<td>8.44 / 4.69</td>
+<td>7.8</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量阿拉伯字母识别模型，支持阿拉伯字母、数字识别</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+cyrillic_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>94.28</td>
+<td>3.56 / 0.79</td>
+<td>8.22 / 2.76</td>
+<td>7.9 M  </td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量斯拉夫字母识别模型，支持斯拉夫字母、数字识别</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>96.44</td>
+<td>3.60 / 0.78</td>
+<td>6.95 / 2.87</td>
+<td>7.9</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量梵文字母识别模型，支持梵文字母、数字识别</td>
+</tr>
+</table>
+
+<p><b>文本行方向分类模块（可选）：</b></p>
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>模型下载链接</th>
+<th>Top-1 Acc（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（M)</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x0_25_textline_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x0_25_textline_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x0_25_textline_ori_pretrained.pdparams">训练模型</a></td>
+<td>95.54</td>
+<td>2.16 / 0.41</td>
+<td>2.37 / 0.73</td>
+<td>0.32</td>
+<td>基于PP-LCNet_x0_25的文本行分类模型，含有两个类别，即0度，180度</td>
+</tr>
+</tbody>
+</table>
+
+<p><b>公式识别模块：</b></p>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>Avg-BLEU(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小 (M)</th>
+<th>介绍</th>
+</tr>
+<td>UniMERNet</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UniMERNet_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UniMERNet_pretrained.pdparams">训练模型</a></td>
+<td>86.13</td>
+<td>2266.96/-</td>
+<td>-/-</td>
+<td>1.4 G</td>
+<td>UniMERNet是由上海AI Lab研发的一款公式识别模型。该模型采用Donut Swin作为编码器，MBartDecoder作为解码器，并通过在包含简单公式、复杂公式、扫描捕捉公式和手写公式在内的一百万数据集上进行训练，大幅提升了模型对真实场景公式的识别准确率</td>
+<td>PP-FormulaNet-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-S_pretrained.pdparams">训练模型</a></td>
+<td>87.12</td>
+<td>1311.84 / 1311.84</td>
+<td>- / 8288.07</td>
+<td>167.9</td>
+<td rowspan="2">PP-FormulaNet 是由百度飞桨视觉团队开发的一款先进的公式识别模型，支持5万个常见LateX源码词汇的识别。PP-FormulaNet-S 版本采用了 PP-HGNetV2-B4 作为其骨干网络，通过并行掩码和模型蒸馏等技术，大幅提升了模型的推理速度，同时保持了较高的识别精度，适用于简单印刷公式、跨行简单印刷公式等场景。而 PP-FormulaNet-L 版本则基于 Vary_VIT_B 作为骨干网络，并在大规模公式数据集上进行了深入训练，在复杂公式的识别方面，相较于PP-FormulaNet-S表现出显著的提升，适用于简单印刷公式、复杂印刷公式、手写公式等场景。 </td>
+
+</tr>
+<td>PP-FormulaNet-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-L_pretrained.pdparams">训练模型</a></td>
+<td>92.13</td>
+<td>1976.52/-</td>
+<td>-/-</td>
+<td>535.2</td>
+<td>LaTeX_OCR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/LaTeX_OCR_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/LaTeX_OCR_rec_pretrained.pdparams">训练模型</a></td>
+<td>71.63</td>
+<td>1088.89 / 1088.89</td>
+<td>- / -</td>
+<td>89.7</td>
+<td>LaTeX-OCR是一种基于自回归大模型的公式识别算法，通过采用 Hybrid ViT 作为骨干网络，transformer作为解码器，显著提升了公式识别的准确性。</td>
+</tr>
+</tbody>
+</table>
+
+<p><b>印章文本检测模块：</b></p>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>检测Hmean（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（M)</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv4_server_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_seal_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_seal_det_pretrained.pdparams">训练模型</a></td>
+<td>98.21</td>
+<td>124.64 / 91.57</td>
+<td>545.68 / 439.86</td>
+<td>109</td>
+<td>PP-OCRv4的服务端印章文本检测模型，精度更高，适合在较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_seal_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_seal_det_pretrained.pdparams">训练模型</a></td>
+<td>96.47</td>
+<td>9.70 / 3.56</td>
+<td>50.38 / 19.64</td>
+<td>4.6</td>
+<td>PP-OCRv4的移动端印章文本检测模型，效率更高，适合在端侧部署</td>
+</tr>
+</tbody>
+</table>
+
+<strong>测试环境说明:</strong>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+            <li><strong>测试数据集：
+             </strong>
+                <ul>
+                  <li>文档图像方向分类模型：PaddleX 自建的数据集，覆盖证件和文档等多个场景，包含 1000 张图片。</li>
+                  <li> 文本图像矫正模型：<a href="https://www3.cs.stonybrook.edu/~cvl/docunet.html">DocUNet</a>。</li>
+                  <li>版面区域检测模型：PaddleOCR 自建的版面区域分析数据集，包含中英文论文、杂志和研报等常见的 1w 张文档类型图片。</li>
+                  <li>PP-DocLayout_plus-L：PaddleOCR 自建的版面区域检测数据集，包含中英文论文、杂志、报纸、研报、PPT、试卷、课本等 1300 张文档类型图片。</li>
+                  <li>表格结构识别模型：PaddleX 内部自建英文表格识别数据集。 </li>
+                  <li>文本检测模型：PaddleOCR 自建的中文数据集，覆盖街景、网图、文档、手写多个场景，其中检测包含 500 张图片。</li>
+                  <li> 中文识别模型： PaddleOCR 自建的中文数据集，覆盖街景、网图、文档、手写多个场景，其中文本识别包含 1.1w 张图片。</li>
+                  <li>ch_SVTRv2_rec：<a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务</a>A榜评估集。</li>
+                  <li> ch_RepSVTR_rec：<a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务</a>B榜评估集。</li>
+                  <li>英文识别模型：PaddleX 自建的英文数据集。</li>
+                  <li> 多语言识别模型：PaddleX 自建的多语种数据集。</li>
+                  <li>文本行方向分类模型：PaddleX 自建的数据集，覆盖证件和文档等多个场景，包含 1000 张图片。</li>
+                  <li> 印章文本检测模型：PaddleX 自建的数据集，包含500张圆形印章图像。</li>
+                </ul>
+             </li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td>本地飞桨推理引擎（默认情况下根据默认模型名称选择合适引擎；若静态图和动态图均可用则优先 <code>paddle_static</code>）</td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+
+</details>
+
+## 2. 快速开始
+
+在本地使用 PP-DocTranslation 产线前，请确保您已经按照[安装教程](../installation.md)完成了wheel包安装。如果您希望选择性安装依赖，请参考安装教程中的相关说明。该产线对应的依赖分组为 `trans`。
+
+**请注意，如果在执行过程中遇到程序失去响应、程序异常退出、内存资源耗尽、推理速度极慢等问题，请尝试参考文档调整配置，例如关闭不需要使用的功能或使用更轻量的模型。**
+
+使用前需准备大语言模型的API密钥，支持[百度云千帆平台](https://console.bce.baidu.com/qianfan/ais/console/onlineService)或符合OpenAI接口标准的本地大模型服务。
+
+### 2.1 命令行方式体验
+
+可以下载 [测试文件](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png)，使用一行命令即可快速体验产线效果：
+
+```bash
+paddleocr pp_doctranslation -i vehicle_certificate-1.png --target_language en --qianfan_api_key your_api_key
+```
+
+上述命令默认使用本地飞桨推理引擎。默认情况下，各模块会根据默认模型名称选择合适的本地飞桨推理引擎：仅支持动态图的模型使用 `paddle_dynamic`；同时支持静态图和动态图的模型优先使用 `paddle_static`。如需运行，请先参考[飞桨框架安装说明](../paddlepaddle_installation.md)安装 PaddlePaddle。
+
+如果选择 `transformers` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 Transformers 环境配置，然后执行如下命令：
+
+```bash
+# 使用 transformers 引擎进行推理
+paddleocr pp_doctranslation -i vehicle_certificate-1.png --target_language en --qianfan_api_key your_api_key \
+    --engine transformers
+```
+
+<details><summary><b>命令行支持更多参数设置，点击展开以查看命令行参数的详细说明</b></summary>
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，必填。<br/>
+<b>说明：</b>
+如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td><b>含义：</b>指定推理结果文件保存的路径。<br/>
+<b>说明：</b>
+如果不设置，推理结果将不会保存到本地。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>target_language</code></td>
+<td><b>含义：</b>目标语言（ISO 639-1 语言代码）。</td>
+<td><code>str</code></td>
+<td><code>zh</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>含义：</b>版面区域检测的模型名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>含义：</b>版面区域检测模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>版面模型得分阈值。<br/>
+<b>说明：</b>
+<code>0-1</code> 之间的任意浮点数。如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>0.5</code>。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>版面检测是否使用后处理NMS。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>版面区域检测模型检测框的扩张系数。<br/>
+<b>说明：</b>
+任意大于 <code>0</code>  浮点数。如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>1.0</code>。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>版面检测中模型输出的检测框的合并处理模式。<br/>
+<b>说明：</b>
+<ul>
+<li><b>large</b>，设置为large时，表示在模型输出的检测框中，对于互相重叠包含的检测框，只保留外部最大的框，删除重叠的内部框；</li>
+<li><b>small</b>，设置为small，表示在模型输出的检测框中，对于互相重叠包含的检测框，只保留内部被包含的小框，删除重叠的外部框；</li>
+<li><b>union</b>，不进行框的过滤处理，内外框都保留；</li>
+</ul>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>large</code>。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>chart_recognition_model_name</code></td>
+<td><b>含义：</b>图表解析的模型名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>chart_recognition_model_dir</code></td>
+<td><b>含义：</b>图表解析模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>chart_recognition_batch_size</code></td>
+<td><b>含义：</b>图表解析模型的batch size。<br/>
+<b>说明：</b>
+如果不设置，将默认设置batch size为<code>1</code>。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>region_detection_model_name</code></td>
+<td><b>含义：</b>文档图像版面子模块检测的模型名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>region_detection_model_dir</code></td>
+<td><b>含义：</b>文档图像版面子模块检测模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>含义：</b>文档方向分类模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>文档方向分类模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>含义：</b>文本图像矫正模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>含义：</b>文本检测模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>含义：</b>文本检测模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>含义：</b>文本检测的图像边长限制。<br/>
+<b>说明：</b>
+大于 <code>0</code> 的任意整数。如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>960</code>。
+</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>含义：</b>文本检测的图像边长限制类型。<br/>
+<b>说明：</b>
+支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>。如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>max</code>。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>含义：</b>文本检测像素阈值，输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。<br/>
+<b>说明：</b>
+大于 <code>0</code> 的任意浮点数
+。如果不设置，将默认使用产线初始化的该参数值 <code>0.3</code>。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>含义：</b>文本检测框阈值，检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。<br/>
+<b>说明：</b>
+大于 <code>0</code> 的任意浮点数
+。如果不设置，将默认使用产线初始化的该参数值 <code>0.6</code>。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>含义：</b>文本检测扩张系数，使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。<br/>
+<b>说明：</b>
+大于 <code>0</code> 的任意浮点数
+。如果不设置，将默认使用产线初始化的该参数值 <code>2.0</code>。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_name</code></td>
+<td><b>含义：</b>文本行方向模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_dir</code></td>
+<td><b>含义：</b>文本行方向模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_batch_size</code></td>
+<td><b>含义：</b>文本行方向模型的batch size。<br/>
+<b>说明：</b>
+如果不设置，将默认设置batch size为<code>1</code>。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>含义：</b>文本识别模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>含义：</b>文本识别模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>含义：</b>文本识别模型的batch size。<br/>
+<b>说明：</b>
+如果不设置，将默认设置batch size为<code>1</code>。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>含义：</b>文本识别阈值，得分大于该阈值的文本结果会被保留。<br/>
+<b>说明：</b>
+大于 <code>0</code> 的任意浮点数。如果不设置，将默认使用产线初始化的该参数值 <code>0.0</code>。即不设阈值。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_classification_model_name</code></td>
+<td><b>含义：</b>表格分类模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_classification_model_dir</code></td>
+<td><b>含义：</b>表格分类模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_name</code></td>
+<td><b>含义：</b>有线表格结构识别模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_dir</code></td>
+<td><b>含义：</b>有线表格结构识别模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_name</code></td>
+<td><b>含义：</b>无线表格结构识别模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_dir</code></td>
+<td><b>含义：</b>无线表格结构识别模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_name</code></td>
+<td><b>含义：</b>有线表格单元格检测模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_dir</code></td>
+<td><b>含义：</b>有线表格单元格检测模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_name</code></td>
+<td><b>含义：</b>无线表格单元格检测模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_dir</code></td>
+<td><b>含义：</b>无线表格单元格检测模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_orientation_classify_model_name</code></td>
+<td><b>含义：</b>表格方向分类模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>表格方向分类模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_name</code></td>
+<td>印章文本检测模型的名称。
+<td><b>含义：</b>印章文本检测模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_dir</code></td>
+<td>印章文本检测模型的目录路径。
+<td><b>含义：</b>印章文本检测模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>含义：</b>印章文本检测的图像边长限制。<br/>
+<b>说明：</b>
+大于 <code>0</code> 的任意整数。如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>736</code>。
+</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>含义：</b>印章文本检测的图像边长限制类型。<br/>
+<b>说明：</b>
+支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>。如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>min</code>。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>含义：</b>印章文本检测的检测像素阈值。输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。<br/>
+<b>说明：</b>
+大于 <code>0</code> 的任意浮点数
+。如果不设置，将默认使用产线初始化的该参数值 <code>0.2</code>。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>含义：</b>印章文本检测的检测框阈值。检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。<br/>
+<b>说明：</b>
+大于 <code>0</code> 的任意浮点数
+。如果不设置，将默认使用产线初始化的该参数值 <code>0.6</code>。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>含义：</b>印章文本检测扩张系数。使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。<br/>
+<b>说明：</b>
+大于 <code>0</code> 的任意浮点数
+。如果不设置，将默认使用产线初始化的该参数值 <code>0.5</code>。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_name</code></td>
+<td><b>含义：</b>印章文本识别模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_dir</code></td>
+<td><b>含义：</b>印章文本识别模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_batch_size</code></td>
+<td><b>含义：</b>印章文本识别模型的batch size。<br/>
+<b>说明：</b>
+如果不设置，将默认设置batch size为<code>1</code>。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>含义：</b>印章文本识别阈值，得分大于该阈值的文本结果会被保留。<br/>
+<b>说明：</b>
+大于 <code>0</code> 的任意浮点数
+。如果不设置，将默认使用产线初始化的该参数值 <code>0.0</code>。即不设阈值。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_name</code></td>
+<td><b>含义：</b>公式识别模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_dir</code></td>
+<td><b>含义：</b>公式识别模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>formula_recognition_batch_size</code></td>
+<td><b>含义：</b>公式识别模型的batch size。<br/>
+<b>说明：</b>
+如果不设置，将默认设置batch size为<code>1</code>。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否加载并使用文档方向分类模块。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>含义：</b>是否加载并使用文本行方向分类模块。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>含义：</b>是否加载并使用印章文本识别子产线。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>含义：</b>是否加载并使用表格识别子产线。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_formula_recognition</code></td>
+<td><b>含义：</b>是否加载并使用公式识别子产线。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>含义：</b>是否加载并使用文档区域检测模块。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_region_detection</code></td>
+<td><b>含义：</b>是否加载并使用文档区域检测模块。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>qianfan_api_key</code></td>
+<td><b>含义：</b>千帆平台的API key。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>含义：</b>
+支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+</ul>如果不设置，将默认使用产线初始化的该参数值，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>fp32</code>、<code>fp16</code>。</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+
+</tbody>
+</table>
+</details>
+<br />
+
+运行结果会被打印到终端上。
+
+### 2.2 Python脚本方式集成
+
+命令行方式是为了快速体验查看效果，一般来说，在项目中，往往需要通过代码集成，可以下载 [测试文件](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png)，使用如下示例代码进行推理：
+
+```python
+from paddleocr import PPDocTranslation
+
+# 创建翻译产线
+pipeline = PPDocTranslation()
+
+# 文档路径
+input_path = "document_sample.pdf"
+
+# 输出目录
+output_path = "./output"
+
+# 大模型配置
+chat_bot_config = {
+    "module_name": "chat_bot",
+    "model_name": "ernie-3.5-8k",
+    "base_url": "https://qianfan.baidubce.com/v2",
+    "api_type": "openai",
+    "api_key": "api_key",  # your api_key
+}
+
+if input_path.lower().endswith(".md"):
+    # 读取markdown文档，支持传入目录和以 .md 为后缀的 url 链接
+    ori_md_info_list = pipeline.load_from_markdown(input_path)
+else:
+    # 使用 PP-StructureV3 对 PDF/图片 文档进行版面解析，获取markdown信息
+    visual_predict_res = pipeline.visual_predict(
+        input_path,
+        use_doc_orientation_classify=False,
+        use_doc_unwarping=False,
+        use_common_ocr=True,
+        use_seal_recognition=True,
+        use_table_recognition=True,
+    )
+
+    ori_md_info_list = []
+    for res in visual_predict_res:
+        layout_parsing_result = res["layout_parsing_result"]
+        ori_md_info_list.append(layout_parsing_result.markdown)
+        layout_parsing_result.save_to_img(output_path)
+        layout_parsing_result.save_to_markdown(output_path)
+
+    # 将多页文档的 markdown 信息拼接成一个 markdown 文件，可将合并后的 markdown 原文保存
+    if input_path.lower().endswith(".pdf"):
+        ori_md_info = pipeline.concatenate_markdown_pages(ori_md_info_list)
+        ori_md_info.save_to_markdown(output_path)
+
+# 执行文档翻译（目标语言：英文）
+tgt_md_info_list = pipeline.translate(
+    ori_md_info_list=ori_md_info_list,
+    target_language="en",
+    chunk_size=5000,
+    chat_bot_config=chat_bot_config,
+)
+# 保存翻译结果
+for tgt_md_info in tgt_md_info_list:
+    tgt_md_info.save_to_markdown(output_path)
+```
+
+上述代码默认使用本地飞桨推理引擎。默认情况下，各模块会根据默认模型名称选择合适的本地飞桨推理引擎：仅支持动态图的模型使用 `paddle_dynamic`；同时支持静态图和动态图的模型优先使用 `paddle_static`。如需运行，请先参考[飞桨框架安装说明](../paddlepaddle_installation.md)安装 PaddlePaddle。
+
+如果选择 `transformers` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 Transformers 环境配置，然后执行如下代码：
+
+```python
+from paddleocr import PPDocTranslation
+
+# 创建翻译产线
+pipeline = PPDocTranslation(
+    engine="transformers",
+)
+
+# 文档路径
+input_path = "document_sample.pdf"
+
+# 输出目录
+output_path = "./output"
+
+# 大模型配置
+chat_bot_config = {
+    "module_name": "chat_bot",
+    "model_name": "ernie-3.5-8k",
+    "base_url": "https://qianfan.baidubce.com/v2",
+    "api_type": "openai",
+    "api_key": "api_key",  # your api_key
+}
+
+if input_path.lower().endswith(".md"):
+    # 读取markdown文档，支持传入目录和以 .md 为后缀的 url 链接
+    ori_md_info_list = pipeline.load_from_markdown(input_path)
+else:
+    # 使用 PP-StructureV3 对 PDF/图片 文档进行版面解析，获取markdown信息
+    visual_predict_res = pipeline.visual_predict(
+        input_path,
+        use_doc_orientation_classify=False,
+        use_doc_unwarping=False,
+        use_common_ocr=True,
+        use_seal_recognition=True,
+        use_table_recognition=True,
+    )
+
+    ori_md_info_list = []
+    for res in visual_predict_res:
+        layout_parsing_result = res["layout_parsing_result"]
+        ori_md_info_list.append(layout_parsing_result.markdown)
+        layout_parsing_result.save_to_img(output_path)
+        layout_parsing_result.save_to_markdown(output_path)
+
+    # 将多页文档的 markdown 信息拼接成一个 markdown 文件，可将合并后的 markdown 原文保存
+    if input_path.lower().endswith(".pdf"):
+        ori_md_info = pipeline.concatenate_markdown_pages(ori_md_info_list)
+        ori_md_info.save_to_markdown(output_path)
+
+# 执行文档翻译（目标语言：英文）
+tgt_md_info_list = pipeline.translate(
+    ori_md_info_list=ori_md_info_list,
+    target_language="en",
+    chunk_size=5000,
+    chat_bot_config=chat_bot_config,
+)
+# 保存翻译结果
+for tgt_md_info in tgt_md_info_list:
+    tgt_md_info.save_to_markdown(output_path)
+```
+
+执行上述代码后，您将得到翻译原文的文档解析结果、翻译原文的 Markdown 文件和翻译后文档的 Markdown 文件，保存在 <code>output</code> 目录中。
+
+PP-DocTranslation 预测的流程、API 说明、产出说明如下：
+
+<details><summary>（1）调用 <code>PPDocTranslation</code> 实例化 PP-DocTranslation 产线对象。</summary>
+
+相关参数说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>含义：</b>版面区域检测的模型名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>含义：</b>版面区域检测模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>版面模型得分阈值。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：<code>0-1</code> 之间的任意浮点数；</li>
+<li><b>dict</b>： <code>{0:0.1}</code> key为类别ID，value为该类别的阈值；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>0.5</code>。</li>
+</ul>
+</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>版面检测是否使用后处理NMS。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>版面区域检测模型检测框的扩张系数。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：任意大于 <code>0</code>  浮点数；</li>
+<li><b>Tuple[float,float]</b>：在横纵两个方向各自的扩张系数；</li>
+<li><b>dict</b>，dict的key为<b>int</b>类型，代表<code>cls_id</code>, value为<b>tuple</b>类型，如<code>{0: (1.1, 2.0)}</code>，表示将模型输出的第0类别检测框中心不变，宽度扩张1.1倍，高度扩张2.0倍；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>1.0</code>。</li>
+</ul>
+</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>版面区域检测的重叠框过滤方式。<br/>
+<b>说明：</b>
+<ul>
+<li><b>str</b>：<code>large</code>，<code>small</code>，<code>union</code>，分别表示重叠框过滤时选择保留大框，小框还是同时保留；</li>
+<li><b>dict</b>： dict的key为<b>int</b>类型，代表<code>cls_id</code>，value为<b>str</b>类型，如<code>{0: "large", 2: "small"}</code>，表示对第0类别检测框使用large模式，对第2类别检测框使用small模式；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>large</code>。</li>
+</ul>
+</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>chart_recognition_model_name</code></td>
+<td><b>含义：</b>图表解析的模型名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>chart_recognition_model_dir</code></td>
+<td><b>含义：</b>图表解析模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>chart_recognition_batch_size</code></td>
+<td><b>含义：</b>图表解析模型的batch size。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>region_detection_model_name</code></td>
+<td><b>含义：</b>文档图像版面子模块检测的模型名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>region_detection_model_dir</code></td>
+<td><b>含义：</b>文档图像版面子模块检测模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>含义：</b>文档方向分类模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>文档方向分类模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>含义：</b>文本图像矫正模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>含义：</b>文本检测模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>含义：</b>文本检测模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>含义：</b>文本检测的图像边长限制。<br/>
+<b>说明：</b>
+<ul>
+<li><b>int</b>：大于 <code>0</code> 的任意整数；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>960</code>。</li>
+</ul>
+</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>含义：</b>文本检测的图像边长限制类型。<br/>
+<b>说明：</b>
+<ul>
+<li><b>str</b>：支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>max</code>。</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>含义：</b>检测像素阈值，输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.3</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>含义：</b>检测框阈值，检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.6</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>含义：</b>文本检测扩张系数，使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>2.0</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_name</code></td>
+<td><b>含义：</b>文本行方向模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_dir</code></td>
+<td><b>含义：</b>文本行方向模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_batch_size</code></td>
+<td><b>含义：</b>文本行方向模型的batch size。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>含义：</b>文本识别模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>含义：</b>文本识别模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>含义：</b>文本识别模型的batch size。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>含义：</b>文本识别阈值，得分大于该阈值的文本结果会被保留。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.0</code>，即不设阈值。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_classification_model_name</code></td>
+<td><b>含义：</b>表格分类模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_classification_model_dir</code></td>
+<td><b>含义：</b>表格分类模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_name</code></td>
+<td><b>含义：</b>有线表格结构识别模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_dir</code></td>
+<td><b>含义：</b>有线表格结构识别模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_name</code></td>
+<td><b>含义：</b>无线表格结构识别模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_dir</code></td>
+<td><b>含义：</b>无线表格结构识别模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_name</code></td>
+<td><b>含义：</b>有线表格单元格检测模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_dir</code></td>
+<td><b>含义：</b>有线表格单元格检测模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_name</code></td>
+<td><b>含义：</b>无线表格单元格检测模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_dir</code></td>
+<td><b>含义：</b>无线表格单元格检测模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_orientation_classify_model_name</code></td>
+<td><b>含义：</b>表格方向分类模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>表格方向分类模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_name</code></td>
+<td><b>含义：</b>印章文本检测模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_dir</code></td>
+<td><b>含义：</b>印章文本检测模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>含义：</b>印章文本检测的图像边长限制。<br/>
+<b>说明：</b>
+<ul>
+<li><b>int</b>：大于 <code>0</code> 的任意整数；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>736</code>。</li>
+</ul>
+</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>含义：</b>印章文本检测的图像边长限制类型。<br/>
+<b>说明：</b>
+<ul>
+<li><b>str</b>：支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>min</code>。</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>含义：</b>检测像素阈值，输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于 <code>0</code> 的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.2</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>含义：</b>检测框阈值，检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.6</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>含义：</b>印章文本检测扩张系数，使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.5</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_name</code></td>
+<td><b>含义：</b>印章文本识别模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_dir</code></td>
+<td><b>含义：</b>印章文本识别模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_batch_size</code></td>
+<td><b>含义：</b>印章文本识别模型的batch size。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>含义：</b>印章文本识别阈值，得分大于该阈值的文本结果会被保留。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.0</code>，即不设阈值。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_name</code></td>
+<td><b>含义：</b>公式识别模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_dir</code></td>
+<td><b>含义：</b>公式识别模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_batch_size</code></td>
+<td><b>含义：</b>公式识别模型的batch size。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否加载并使用文档方向分类模块。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>含义：</b>是否加载并使用文本行方向分类模块。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>含义：</b>是否加载并使用印章文本识别子产线。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>含义：</b>是否加载并使用表格识别子产线。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_formula_recognition</code></td>
+<td><b>含义：</b>是否加载并使用公式识别子产线。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>含义：</b>是否加载并使用图表解析模块。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_region_detection</code></td>
+<td><b>含义：</b>是否加载并使用文档区域检测模块。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>chat_bot_config</code></td>
+<td><b>含义：</b>大语言模型配置信息。<br/>
+<b>说明：</b>
+配置内容为如下dict：
+<pre><code>{
+"module_name": "chat_bot",
+"model_name": "ernie-3.5-8k",
+"base_url": "https://qianfan.baidubce.com/v2",
+"api_type": "openai",
+"api_key": "api_key"  # 请将此设置为实际的API密钥
+}</code></pre>
+</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+<li><b>None</b>：如果设置为<code>None</code>，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+</tbody>
+</table>
+
+</details>
+
+<details><summary>（2）调用 PP-DocTranslation 产线对象的 <code>visual_predict()</code> 方法获取视觉预测结果，该方法会返回一个结果列表。另外，产线还提供了 <code>visual_predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>visual_predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。以下是 <code>visual_predict()</code> 方法的参数及其说明：</summary>
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+  <li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据；</li>
+  <li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href = "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/vehicle_certificate-1.png">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)；</li>
+  <li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]</code>。</li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否在推理时使用文档方向分类模块。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否在推理时使用文本图像矫正模块。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>含义：</b>是否在推理时使用文本行方向分类模块。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>含义：</b>是否在推理时使用印章文本识别子产线。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>含义：</b>是否在推理时使用表格识别子产线。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_formula_recognition</code></td>
+<td><b>含义：</b>是否在推理时使用公式识别子产线。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>含义：</b>是否在推理时使用图表解析模块。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_region_detection</code></td>
+<td><b>含义：</b>是否在推理时使用文档区域检测模块。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td></td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_wired_table_cells_trans_to_html</code></td>
+<td><b>含义：</b>是否启用有线表单元格检测结果直转HTML。<br/>
+<b>说明：</b>
+启用则直接基于有线表单元格检测结果的几何关系构建HTML。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_wireless_table_cells_trans_to_html</code></td>
+<td><b>含义：</b>是否启用无线表单元格检测结果直转HTML。<br/>
+<b>说明：</b>
+启用则直接基于无线表单元格检测结果的几何关系构建HTML。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_table_orientation_classify</code></td>
+<td><b>含义：</b>是否启用表格使用表格方向分类。<br/>
+<b>说明：</b>
+启用时当图像中的表格存在90/180/270度旋转时，能够将方向校正并正确完成表格识别。</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_ocr_results_with_table_cells</code></td>
+<td><b>含义：</b>是否启用单元格切分OCR。<br/>
+<b>说明：</b>
+启用时会基于单元格预测结果对OCR检测结果进行切分和重识别，避免出现文字缺失情况。</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_e2e_wired_table_rec_model</code></td>
+<td><b>含义：</b>是否启用有线表端到端表格识别模式。<br/>
+<b>说明：</b>
+启用则不使用单元格检测模型，只使用表格结构识别模型。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_e2e_wireless_table_rec_model</code></td>
+<td><b>含义：</b>是否启用无线表端到端表格识别模式。<br/>
+<b>说明：</b>
+启用则不使用单元格检测模型，只使用表格结构识别模型。</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+</table>
+</details>
+
+<details><summary>（3）对视觉预测结果进行处理：每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为<code>json</code>文件的操作:</summary>
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将中间各个模块的可视化图像保存在png格式的图像</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>save_to_markdown()</code></td>
+<td>将图像或者PDF文件中的每一页分别保存为markdown格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>save_to_html()</code></td>
+<td>将文件中的表格保存为html格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>save_to_xlsx()</code></td>
+<td>将文件中的表格保存为xlsx格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径</td>
+<td>无</td>
+</tr>
+</table>
+<ul>
+<li>调用<code>print()</code> 方法会将结果打印到终端，打印到终端的内容解释如下：
+    <ol start="1" type="1">
+    <li><code>input_path</code>: <code>(str)</code> 待预测图像或者PDF的输入路径</li>
+    <li><code>page_index</code>: <code>(Union[int, None])</code> 如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+    <li><code>model_settings</code>: <code>(Dict[str, bool])</code> 配置产线所需的模型参数
+        <ol>
+        <li><code>use_doc_preprocessor</code>: <code>(bool)</code> 控制是否启用文档预处理子产线</li>
+        <li><code>use_general_ocr</code>: <code>(bool)</code> 控制是否启用 OCR 子产线</li>
+        <li><code>use_seal_recognition</code>: <code>(bool)</code> 控制是否启用印章识别子产线</li>
+        <li><code>use_table_recognition</code>: <code>(bool)</code> 控制是否启用表格识别子产线</li>
+        <li><code>use_formula_recognition</code>: <code>(bool)</code> 控制是否启用公式识别子产线</li>
+        </ol>
+    </li>
+    <li><code>doc_preprocessor_res</code>: <code>(Dict[str, Union[List[float], str]])</code> 文档预处理结果字典，仅当<code>use_doc_preprocessor=True</code>时存在
+        <ol>
+        <li><code>input_path</code>: <code>(str)</code> 文档预处理子产线接受的图像路径，当输入为<code>numpy.ndarray</code>时，保存为<code>None</code>，此处为<code>None</code></li>
+        <li><code>page_index</code>: <code>None</code>，此处的输入为<code>numpy.ndarray</code>，所以值为<code>None</code></li>
+        <li><code>model_settings</code>: <code>(Dict[str, bool])</code> 文档预处理子产线的模型配置参数
+          <ul>
+          <li><code>use_doc_orientation_classify</code>: <code>(bool)</code> 控制是否启用文档图像方向分类子模块</li>
+          <li><code>use_doc_unwarping</code>: <code>(bool)</code> 控制是否启用文本图像扭曲矫正子模块</li>
+          </ul>
+        <li><code>angle</code>: <code>(int)</code> 文档图像方向分类子模块的预测结果，启用时返回实际角度值</li>
+        </ul>
+    </li>
+    <li><code>parsing_res_list</code>: <code>(List[Dict])</code> 解析结果的列表，每个元素为一个字典，列表顺序为解析后的阅读顺序。
+        <ol>
+        <li><code>block_bbox</code>: <code>(np.ndarray)</code> 版面区域的边界框。</li>
+        <li><code>block_label</code>: <code>(str)</code> 版面区域的标签，例如<code>text</code>, <code>table</code>等。</li>
+        <li><code>block_content</code>: <code>(str)</code> 内容为版面区域内的内容。</li>
+        <li><code>block_id</code>: <code>(int)</code> 版面区域的索引，用于显示版面排序结果。</li>
+        <li><code>block_order</code> <code>(int)</code> 版面区域的顺序，用于显示版面阅读顺序,对于非排序部分，默认值为 <code>None</code>。</li>
+        </ol>
+    </li>
+    <li><code>overall_ocr_res</code>: <code>(Dict[str, Union[List[str], List[float], numpy.ndarray]])</code> 全局 OCR 结果的字典
+        <ol>
+        <li><code>input_path</code>: <code>(Union[str, None])</code> 图像OCR子产线接受的图像路径，当输入为<code>numpy.ndarray</code>时，保存为<code>None</code></li>
+        <li><code>page_index</code>: <code>None</code>，此处的输入为<code>numpy.ndarray</code>，所以值为<code>None</code></li>
+        <li><code>model_settings</code>: <code>(Dict)</code> OCR子产线的模型配置参数</li>
+        <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> 文本检测的多边形框列表。每个检测框由4个顶点坐标构成的numpy数组表示，数组shape为(4, 2)，数据类型为int16</li> 
+        <li><code>dt_scores</code>: <code>(List[float])</code> 文本检测框的置信度列表</li>    
+        <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> 文本检测模块的配置参数
+            <ol>
+            <li><code>limit_side_len</code>: <code>(int)</code> 图像预处理时的边长限制值</li>
+            <li><code>limit_type</code>: <code>(str)</code> 边长限制的处理方式</li>
+            </ol>
+        <li><code>thresh</code>: <code>(float)</code> 文本像素分类的置信度阈值</li>
+        <li><code>box_thresh</code>: <code>(float)</code> 文本检测框的置信度阈值</li>
+        <li><code>unclip_ratio</code>: <code>(float)</code> 文本检测框的膨胀系数</li>
+        <li><code>text_type</code>: <code>(str)</code> 文本检测的类型，当前固定为"general"</li> 
+        <li><code>text_type</code>: <code>(str)</code> 文本检测的类型，当前固定为"general"</li>
+        <li><code>textline_orientation_angles</code>: <code>(List[int])</code> 文本行方向分类的预测结果。启用时返回实际角度值（如[0,0,1]</li>
+        <li><code>text_rec_score_thresh</code>: <code>(float)</code> 文本识别结果的过滤阈值</li>
+        <li><code>rec_texts</code>: <code>(List[str])</code> 文本识别结果列表，仅包含置信度超过<code>text_rec_score_thresh</code>的文本</li>
+        <li><code>rec_scores</code>: <code>(List[float])</code> 文本识别的置信度列表，已按<code>text_rec_score_thresh</code>过滤</li>
+        <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> 经过置信度过滤的文本检测框列表，格式同<code>dt_polys</code></li>
+         </ol>  
+    </li>
+    <li><code>formula_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> 公式识别结果列表，每个元素为一个字典
+        <ol>
+        <li><code>rec_formula</code>: <code>(str)</code> 公式识别结果</li>
+        <li><code>rec_polys</code>: <code>(numpy.ndarray)</code> 公式检测框，shape为(4, 2)，dtype为int16</li>
+        <li><code>formula_region_id</code>: <code>(int)</code> 公式所在的区域编号</li>
+        </ol>
+    </li>
+    <li><code>seal_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> 印章识别结果列表，每个元素为一个字典
+        <ol>
+        <li><code>input_path</code>: <code>(str)</code> 印章图像的输入路径</li>
+        <li><code>page_index</code>: <code>None</code>，此处的输入为<code>numpy.ndarray</code>，所以值为<code>None</code></li>
+        <li><code>model_settings</code>: <code>(Dict)</code> 印章识别子产线的模型配置参数</li>
+        <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> 印章检测框列表，格式同<code>dt_polys</code></li>
+        <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> 印章检测模块的配置参数, 具体参数含义同上</li>
+        <li><code>text_type</code>: <code>(str)</code> 印章检测的类型，当前固定为"seal"</li>
+        <li><code>text_rec_score_thresh</code>: <code>(float)</code> 印章识别结果的过滤阈值</li>
+        <li><code>rec_texts</code>: <code>(List[str])</code> 印章识别结果列表，仅包含置信度超过<code>text_rec_score_thresh</code>的文本</li>
+        <li><code>rec_scores</code>: <code>(List[float])</code> 印章识别的置信度列表，已按<code>text_rec_score_thresh</code>过滤</li>
+        <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> 经过置信度过滤的印章检测框列表，格式同<code>dt_polys</code></li>
+        <li> <code>rec_boxes</code>: <code>(numpy.ndarray)</code> 检测框的矩形边界框数组，shape为(n, 4)，dtype为int16。每一行表示一个矩形</li>
+        </ol>
+    </li>
+    <li><code>table_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> 表格识别结果列表，每个元素为一个字典
+        <ol>
+        <li><code>table_id</code>: <code>(int)</code> 表格的索引，用于显示表格排序结果。</li>
+        <li><code>cell_box_list</code>: <code>(List[numpy.ndarray])</code> 表格单元格的边界框列表</li>
+        <li><code>pred_html</code>: <code>(str)</code> 表格的HTML格式字符串</li>
+        <li><code>table_ocr_pred</code>: <code>(dict)</code> 表格的OCR识别结果
+            <ul>
+            <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> 单元格的检测框列表</li>
+            <li><code>rec_texts</code>: <code>(List[str])</code> 单元格的识别结果</li>
+            <li><code>rec_scores</code>: <code>(List[float])</code> 单元格的识别置信度</li>
+            <li><code>rec_boxes</code>: <code>(numpy.ndarray)</code> 检测框的矩形边界框数组，shape为(n, 4)，dtype为int16。每一行表示一个矩形
+            </ul>
+            </li>
+        </li>
+    </li>
+    </ol>
+</li>
+<li>调用<code>save_to_json()</code> 方法会将上述内容保存到指定的 <code>save_path</code> 中，如果指定为目录，则保存的路径为<code>save_path/{your_img_basename}_res.json</code>，如果指定为文件，则直接保存到该文件中。由于 json 文件不支持保存numpy数组，因此会将其中的 <code>numpy.array</code> 类型转换为列表形式。</li>
+<li>调用<code>save_to_img()</code> 方法会将可视化结果保存到指定的 <code>save_path</code> 中，如果指定为目录，则会将版面区域检测可视化图像、全局OCR可视化图像、版面阅读顺序可视化图像等内容保存，如果指定为文件，则直接保存到该文件中。(产线通常包含较多结果图片，不建议直接指定为具体的文件路径，否则多张图会被覆盖，仅保留最后一张图)</li>
+<li>调用<code>save_to_markdown()</code> 方法会将转化后的 Markdown 文件保存到指定的 <code>save_path</code> 中，保存的文件路径为<code>save_path/{your_img_basename}.md</code>，如果输入是 PDF 文件，建议直接指定目录，否责多个 markdown 文件会被覆盖。</li>
+<li>调用 <code>concatenate_markdown_pages()</code> 方法将 PP-DocTranslation 产线输出的多页Markdown内容<code>markdown_list</code>合并为单个完整文档，并返回合并后的Markdown内容。</li>
+</ul>
+</details>
+
+<details><summary>（4）调用 <code>translate()</code> 方法执行文档翻译，该方法会返回翻译的 markdown 原文和译文，是一个markdown对象，可以通过执行 <code>save_to_markdown()</code> 方法将需要的部分保存到本地。以下是 <code>translate()</code> 方法的相关参数说明：</summary>
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>ori_md_info_list</code></td>
+<td>原始Markdown格式的数据列表，包含需要翻译的内容。必须为字典组成的列表，每个字典表示一个文档块</td>
+<td><code>List[Dict]</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>target_language</code></td>
+<td>目标语言（ISO 639-1 语言代码，如 <code>"en"</code>/<code>"ja"</code>/<code>"fr"</code>）。</td>
+<td><code>str</code></td>
+<td><code>"zh"</code></td>
+</tr>
+<tr>
+<td><code>chunk_size</code></td>
+<td>翻译文本分块处理的字符数阈值。</td>
+<td><code>int</code></td>
+<td><code>5000</code></td>
+</tr>
+<tr>
+<td><code>task_description</code></td>
+<td>自定义任务描述提示词。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>output_format</code></td>
+<td>指定输出格式要求，如“保持原Markdown结构”。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>rules_str</code></td>
+<td>自定义翻译规则说明。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>few_shot_demo_text_content</code></td>
+<td>少样本学习的示例文本内容。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>few_shot_demo_key_value_list</code></td>
+<td>结构化少样本示例数据。键值对格式的示例数据，可传入专业术语对照表。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>glossary</code></td>
+<td>翻译专业术语对照表。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>llm_request_interval</code></td>
+<td>向大语言模型发送请求的时间间隔，单位为秒。该参数可用于防止过于频繁地调用大语言模型。</td>
+<td><code>float</code></td>
+<td><code>0.0</code></td>
+</tr>
+<tr>
+<td><code>chat_bot_config</code></td>
+<td>大语言模型配置。设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+</details>
+
+## 3. 开发集成/部署
+
+如果产线可以达到您对产线推理速度和精度的要求，您可以直接进行开发集成/部署。
+
+若您需要将产线直接应用在您的Python项目中，可以参考 [2.2 Python脚本方式](#22-python)中的示例代码。
+
+此外，PaddleOCR 也提供了其他两种部署方式，详细说明如下：
+
+🚀 高性能推理：在实际生产环境中，许多应用对部署策略的性能指标（尤其是响应速度）有着较严苛的标准，以确保系统的高效运行与用户体验的流畅性。为此，PaddleOCR 提供高性能推理功能，旨在对模型推理及前后处理进行深度性能优化，实现端到端流程的显著提速，详细的高性能推理流程请参考[高性能推理](../inference_deployment/local_inference/high_performance_inference.md)。
+
+☁️ 服务化部署：服务化部署是实际生产环境中常见的一种部署形式。通过将推理功能封装为服务，客户端可以通过网络请求来访问这些服务，以获取推理结果。详细的产线服务化部署流程请参考[服务化部署](../inference_deployment/serving/serving.md)。
+
+以下是基础服务化部署的API参考与多语言服务调用示例：
+
+<details><summary>API参考</summary>
+<p>对于服务提供的主要操作：</p>
+<ul>
+<li>HTTP请求方法为POST。</li>
+<li>请求体和响应体均为JSON数据（JSON对象）。</li>
+<li>当请求处理成功时，响应状态码为<code>200</code>，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。固定为<code>0</code>。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。固定为<code>"Success"</code>。</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>操作结果。</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>当请求处理未成功时，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。与响应状态码相同。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。</td>
+</tr>
+</tbody>
+</table>
+<p>服务提供的主要操作如下：</p>
+<ul>
+<li><b><code>analyzeImages</code></b></li>
+</ul>
+<p>使用计算机视觉模型对图像进行分析，获得OCR、表格识别结果等。</p>
+<p><code>POST /doctrans-visual</code></p>
+<ul>
+<li>请求体的属性如下：</li>
+</ul>
+
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>服务器可访问的图像文件（含 TIFF，多页时按页处理）或 PDF 文件的 URL，或上述类型文件内容的 Base64 编码结果。默认对于超过 10 页的 PDF 或多页 TIFF，仅处理前 10 页。<br /> 要解除页数限制，请在产线配置文件中添加以下配置：
+<pre><code>Serving:
+  extra:
+    max_num_input_imgs: null
+</code></pre>
+</td>
+<td>是</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code>｜<code>null</code></td>
+<td>文件类型。<code>0</code> 表示 PDF 文件，<code>1</code> 表示图像文件（含 TIFF）。若请求体无此属性，则将根据URL推断文件类型。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_doc_orientation_classify</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_doc_unwarping</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useTextlineOrientation</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_textline_orientation</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useSealRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_seal_recognition</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useTableRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_table_recognition</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useFormulaRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_formula_recognition</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useChartRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_chart_recognition</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useRegionDetection</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_region_detection</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutThreshold</code></td>
+<td><code>number</code> | <code>object</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>layout_threshold</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutNms</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>layout_nms</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutUnclipRatio</code></td>
+<td><code>number</code> | <code>array</code> | <code>object</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>layout_unclip_ratio</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutMergeBboxesMode</code></td>
+<td><code>string</code> | <code>object</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>layout_merge_bboxes_mode</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>text_det_limit_side_len</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>text_det_limit_type</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>text_det_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>text_det_box_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>text_det_unclip_ratio</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>text_rec_score_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>seal_det_limit_side_len</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>seal_det_limit_type</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>seal_det_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>seal_det_box_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>seal_det_unclip_ratio</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>seal_rec_score_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useWiredTableCellsTransToHtml</code></td>
+<td><code>boolean</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_wired_table_cells_trans_to_html</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useWirelessTableCellsTransToHtml</code></td>
+<td><code>boolean</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_wireless_table_cells_trans_to_html</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useTableOrientationClassify</code></td>
+<td><code>boolean</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_table_orientation_classify</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useOcrResultsWithTableCells</code></td>
+<td><code>boolean</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_ocr_results_with_table_cells</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useE2eWiredTableRecModel</code></td>
+<td><code>boolean</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_e2e_wired_table_rec_model</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useE2eWirelessTableRecModel</code></td>
+<td><code>boolean</code></td>
+<td>请参阅产线对象中 <code>visual_predict</code> 方法的 <code>use_e2e_wireless_table_rec_model</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>outputFormats</code></td>
+<td><code>array</code> | <code>null</code></td>
+<td>可选。附加导出格式列表；当前仅支持 <code>"docx"</code>。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>是否返回可视化结果图以及处理过程中的中间图像等。
+<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>传入 <code>true</code>：返回图像。</li>
+<li>传入 <code>false</code>：不返回图像。</li>
+<li>若请求体中未提供该参数或传入 <code>null</code>：遵循产线配置文件<code>Serving.visualize</code> 的设置。</li>
+</ul>
+<br/>例如，在产线配置文件中添加如下字段：<br/>
+<pre><code>Serving:
+  visualize: False
+</code></pre>
+将默认不返回图像，通过请求体中的<code>visualize</code>参数可以覆盖默认行为。如果请求体和配置文件中均未设置（或请求体传入<code>null</code>、配置文件中未设置），则默认返回图像。
+</td>
+<td>否</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>请求处理成功时，响应体的<code>result</code>具有如下属性：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>layoutParsingResults</code></td>
+<td><code>array</code></td>
+<td>版面解析结果。数组长度为1（对于图像输入）或实际处理的文档页数（对于PDF输入）。对于PDF输入，数组中的每个元素依次表示PDF文件中实际处理的每一页的结果。</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>输入数据信息。</td>
+</tr>
+</tbody>
+</table>
+<p>下表中涉及图像等二进制文件的字段（如 <code>outputImages</code>、<code>inputImage</code>、<code>markdown.images</code>、<code>exports</code>）默认以 Base64 字符串内联返回；当服务端开启 URL 返回模式时，相应字段的值变为预签名 URL，字段类型保持不变。配置方式参见 <a href="../inference_deployment/serving/serving.md">服务化部署</a>「以 URL 形式返回二进制内容」一节。</p>
+<p><code>layoutParsingResults</code>中的每个元素为一个<code>object</code>，具有如下属性：</p>
+<table>
+
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>产线对象的 <code>visual_predict</code> 方法生成的 <code>layout_parsing_result</code> 的 JSON 表示中 <code>res</code> 字段的简化版本，其中去除了 <code>input_path</code> 和 <code>page_index</code> 字段。</td>
+</tr>
+<tr>
+<td><code>markdown</code></td>
+<td><code>object</code></td>
+<td>Markdown结果。</td>
+</tr>
+<tr>
+<td><code>outputImages</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>参见产线预测结果的 <code>img</code> 属性说明。图像为JPEG格式，使用Base64编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+<tr>
+<td><code>inputImage</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>输入图像。图像为JPEG格式，使用Base64编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+<tr>
+<td><code>exports</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>可选的 <code>docx</code> 等附加导出，仅当请求 <code>outputFormats</code> 时出现。其中 <code>content</code> 为文件内容的Base64编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+</tbody>
+</table>
+<p><code>markdown</code>为一个<code>object</code>，具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>text</code></td>
+<td><code>string</code></td>
+<td>Markdown文本。</td>
+</tr>
+<tr>
+<td><code>images</code></td>
+<td><code>object</code></td>
+<td>Markdown图片相对路径和Base64编码图像的键值对；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+<tr>
+<td><code>isStart</code></td>
+<td><code>boolean</code></td>
+<td>当前页面第一个元素是否为段开始。</td>
+</tr>
+<tr>
+<td><code>isEnd</code></td>
+<td><code>boolean</code></td>
+<td>当前页面最后一个元素是否为段结束。</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li><b><code>translate</code></b></li>
+</ul>
+<p>利用大模型翻译文档。</p>
+<p><code>POST /doctrans-translate</code></p>
+<ul>
+<li>请求体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>markdownList</code></td>
+<td><code>array</code></td>
+<td>待翻译的Markdown列表。可从<code>analyzeImages</code>操作的结果中获取。<code>images</code>属性将不会被用到。</td>
+<td>是</td>
+</tr>
+<tr>
+<td><code>targetLanguage</code></td>
+<td><code>string</code></td>
+<td>请参阅产线对象中 <code>translate</code> 方法的 <code>target_language</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>chunkSize</code></td>
+<td><code>integer</code></td>
+<td>请参阅产线对象中 <code>translate</code> 方法的 <code>chunk_size</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>taskDescription</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>translate</code> 方法的 <code>task_description</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>outputFormat</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>translate</code> 方法的 <code>output_format</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>rulesStr</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>translate</code> 方法的 <code>rules_str</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>fewShotDemoTextContent</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>translate</code> 方法的 <code>few_shot_demo_text_content</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>fewShotDemoKeyValueList</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>translate</code> 方法的 <code>few_shot_demo_key_value_list</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>glossary</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>translate</code> 方法的 <code>glossary</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>llmRequestInterval</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>translate</code> 方法的 <code>llm_request_interval</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>chatBotConfig</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>translate</code> 方法的 <code>chat_bot_config</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>请求处理成功时，响应体的<code>result</code>具有如下属性：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>translationResults</code></td>
+<td><code>array</code></td>
+<td>翻译结果。</td>
+</tr>
+</tbody>
+</table>
+<p><code>translationResults</code>中的每个元素为一个<code>object</code>，具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>language</code></td>
+<td><code>string</code></td>
+<td>目标语言。</td>
+</tr>
+<tr>
+<td><code>markdown</code></td>
+<td><code>object</code></td>
+<td>Markdown结果。对象定义与<code>analyzeImages</code>操作返回的<code>markdown</code>一致。</td>
+</tr>
+</tbody>
+</table>
+<li><b>注意：</b></li>
+在请求体中包含大模型调用的API key等敏感参数可能存在安全风险。如无必要，请在配置文件中设置这些参数，在请求时不传递。
+<br/><br/>
+</details>
+<details><summary>多语言调用服务示例</summary>
+<details>
+<summary>Python</summary>
+
+<pre><code class="language-python">import base64
+import pathlib
+import pprint
+import sys
+
+import requests
+
+
+API_BASE_URL = "http://127.0.0.1:8080"
+
+file_path = "./demo.jpg"
+target_language = "en"
+
+with open(file_path, "rb") as file:
+    file_bytes = file.read()
+    file_data = base64.b64encode(file_bytes).decode("ascii")
+
+payload = {
+    "file": file_data,
+    "fileType": 1,
+}
+resp_visual = requests.post(url=f"{API_BASE_URL}/doctrans-visual", json=payload)
+if resp_visual.status_code != 200:
+    print(
+        f"Request to doctrans-visual failed with status code {resp_visual.status_code}."
+    )
+    pprint.pp(resp_visual.json())
+    sys.exit(1)
+result_visual = resp_visual.json()["result"]
+
+markdown_list = []
+for i, res in enumerate(result_visual["layoutParsingResults"]):
+    md_dir = pathlib.Path(f"markdown_{i}")
+    md_dir.mkdir(exist_ok=True)
+    (md_dir / "doc.md").write_text(res["markdown"]["text"])
+    for img_path, img in res["markdown"]["images"].items():
+        img_path = md_dir / img_path
+        img_path.parent.mkdir(parents=True, exist_ok=True)
+        img_path.write_bytes(base64.b64decode(img))
+    print(f"Markdown document to be translated is saved at {md_dir / 'doc.md'}")
+    del res["markdown"]["images"]
+    markdown_list.append(res["markdown"])
+    for img_name, img in res["outputImages"].items():
+        img_path = f"{img_name}_{i}.jpg"
+        with open(img_path, "wb") as f:
+            f.write(base64.b64decode(img))
+        print(f"Output image saved at {img_path}")
+
+payload = {
+    "markdownList": markdown_list,
+    "targetLanguage": target_language,
+}
+resp_translate = requests.post(url=f"{API_BASE_URL}/doctrans-translate", json=payload)
+if resp_translate.status_code != 200:
+    print(
+        f"Request to doctrans-translate failed with status code {resp_translate.status_code}."
+    )
+    pprint.pp(resp_translate.json())
+    sys.exit(1)
+result_translate = resp_translate.json()["result"]
+
+for i, res in enumerate(result_translate["translationResults"]):
+    md_dir = pathlib.Path(f"markdown_{i}")
+    (md_dir / "doc_translated.md").write_text(res["markdown"]["text"])
+    print(f"Translated markdown document saved at {md_dir / 'doc_translated.md'}")
+</code></pre></details>
+</details>
+<br/>
+
+## 4. 二次开发
+如果 PP-DocTranslation 产线提供的默认模型权重在您的场景中，精度或速度不满意，您可以尝试利用<b>您自己拥有的特定领域或应用场景的数据</b>对现有模型进行进一步的<b>微调</b>，以提升在您的场景中的识别效果。
+
+### 4.1 模型微调
+由于 PP-DocTranslation 产线包含若干模块，模型产线的效果如果不及预期，可能来自于其中任何一个模块。您可以对提取效果差的 case 进行分析，通过可视化图像，确定是哪个模块存在问题，并参考以下表格中对应的微调教程链接进行模型微调。
+
+<table>
+<thead>
+<tr>
+<th>情形</th>
+<th>微调模块</th>
+<th>微调参考链接</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>版面区域检测不准，如印章、表格未检出等</td>
+<td>版面区域检测模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/layout_detection.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>表格结构识别不准</td>
+<td>表格结构识别模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/table_structure_recognition.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>公式识别不准</td>
+<td>公式识别模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/formula_recognition.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>印章文本存在漏检</td>
+<td>印章文本检测模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/seal_text_detection.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>文本存在漏检</td>
+<td>文本检测模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/text_detection.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>文本内容都不准</td>
+<td>文本识别模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/text_recognition.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>垂直或者旋转文本行矫正不准</td>
+<td>文本行方向分类模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/textline_orientation_classification.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>整图旋转矫正不准</td>
+<td>文档图像方向分类模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>图像扭曲矫正不准</td>
+<td>文本图像矫正模块</td>
+<td>暂不支持微调</td>
+</tr>
+</tbody>
+</table>
+
+### 4.2 模型应用
+当您使用私有数据集完成微调训练后，可获得本地模型权重文件，然后可以通过自定义产线配置文件的方式，使用微调后的模型权重。
+
+1. 获取产线配置文件
+
+可调用 PaddleOCR 中 PP-DocTranslation 产线对象的 `export_paddlex_config_to_yaml` 方法，将当前产线配置导出为 YAML 文件：
+
+```Python
+from paddleocr import PPDocTranslation
+
+pipeline = PPDocTranslation()
+pipeline.export_paddlex_config_to_yaml("PP-DocTranslation.yaml")
+```
+
+2. 修改配置文件
+
+在得到默认的产线配置文件后，将微调后模型权重的本地路径替换至产线配置文件中的对应位置即可。例如
+
+```yaml
+......
+SubModules:
+    TextDetection:
+    module_name: text_detection
+    model_name: PP-OCRv5_server_det
+    model_dir: null # 替换为微调后的文本检测模型权重路径
+    limit_side_len: 960
+    limit_type: max
+    thresh: 0.3
+    box_thresh: 0.6
+    unclip_ratio: 1.5
+
+    TextRecognition:
+    module_name: text_recognition
+    model_name: PP-OCRv5_server_rec
+    model_dir: null # 替换为微调后的文本检测模型权重路径
+    batch_size: 1
+            score_thresh: 0
+......
+```
+
+在产线配置文件中，不仅包含 PaddleOCR CLI 和 Python API 支持的参数，还可进行更多高级配置，具体信息可在 [PaddleX模型产线使用概览](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/pipeline_develop_guide.html) 中找到对应的产线使用教程，参考其中的详细说明，根据需求调整各项配置。
+
+3. 在 CLI 中加载产线配置文件
+
+在修改完成配置文件后，通过命令行的 `--paddlex_config` 参数指定修改后的产线配置文件的路径，PaddleOCR 会读取其中的内容作为产线配置。示例如下：
+
+```bash
+paddleocr pp_doctranslation --paddlex_config PP-DocTranslation.yaml ...
+```
+
+4. 在 Python API 中加载产线配置文件
+
+初始化产线对象时，可通过 `paddlex_config` 参数传入 PaddleX 产线配置文件路径或配置dict，PaddleOCR 会读取其中的内容作为产线配置。示例如下：
+
+```python
+from paddleocr import PPDocTranslation
+
+pipeline = PPDocTranslation(paddlex_config="PP-DocTranslation.yaml")
+```
diff --git a/docs/version3.x/pipeline_usage/PP-StructureV3.en.md b/docs/version3.x/pipeline_usage/PP-StructureV3.en.md
new file mode 100644
index 0000000..6760451
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PP-StructureV3.en.md
@@ -0,0 +1,4192 @@
+---
+comments: true
+---
+
+# PP-StructureV3 Pipeline Usage Tutorial
+
+## 1. Introduction to PP-StructureV3 Pipeline
+
+Layout analysis is a technique used to extract structured information from document images. It is primarily used to convert complex document layouts into machine-readable data formats. This technology has broad applications in document management, information extraction, and data digitization. Layout analysis combines Optical Character Recognition (OCR), image processing, and machine learning algorithms to identify and extract text blocks, titles, paragraphs, images, tables, and other layout elements from documents. This process generally includes three main steps: layout analysis, element analysis, and data formatting. The final result is structured document data, which enhances the efficiency and accuracy of data processing. <b>PP-StructureV3 improves upon the general layout analysis v1 pipeline by enhancing layout region detection, table recognition, and formula recognition. It also adds capabilities such as multi-column reading order recovery, chart understanding, and result conversion to Markdown files. It performs excellently across various document types and can handle complex document data.</b>  This pipeline also provides flexible service deployment options, supporting invocation using multiple programming languages on various hardware. In addition, it offers secondary development capabilities, allowing you to train and fine-tune models on your own dataset and integrate the trained models seamlessly.
+
+<b>The PP-StructureV3 pipeline consists of the following seven modules or sub-pipelines. Each module or sub-pipeline can be trained and inferred independently and contains multiple models. For more details, please click the corresponding links to view the documentation.</b>
+
+- [Layout Detection Module](../module_usage/layout_detection.en.md)
+- [General OCR Subline](./OCR.en.md)
+- [Document Image Preprocessing Subline](./doc_preprocessor.en.md) （Optional）
+- [Table Recognition Subline ](./table_recognition_v2.en.md) （Optional）
+- [Seal Text Recognition Subline](./seal_recognition.en.md) （Optional）
+- [Formula Recognition Subline](./formula_recognition.en.md) （Optional）
+- [Chart Parsing Module](../module_usage/chart_parsing.en.md) (Optional)
+
+In this pipeline, you can choose the model to use based on the benchmark data below.
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing.
+> In the inference time columns labeled [Standard Mode / High-Performance Mode], [Normal Mode / High-Performance Mode], or [Regular Mode / High-Performance Mode], the Standard Mode, Normal Mode, and Regular Mode values correspond to local Paddle inference engines. Each module selects the appropriate local Paddle inference engine according to the default model name: models that support only dynamic graph use `paddle_dynamic`, while models that support both static and dynamic graph prefer `paddle_static`.
+
+<details>
+<summary><b>Document Image Orientation Classification Module :</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Top-1 Acc (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">Pretrained Model</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>Document image classification model based on PP-LCNet_x1_0, supporting four categories: 0°, 90°, 180°, 270°</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Text Image Rectification Module:</b></summary>
+<p><b>Text Image Rectification Module (Optional):</b></p>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>CER</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">Pretrained Model</a></td>
+<td>0.179</td>
+<td>19.05 / 19.05</td>
+<td>- / 869.82</td>
+<td>30.3</td>
+<td>High-precision text image rectification model</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Layout Detection Module Model:</b></summary>
+* <b>The layout detection model includes 20 common categories: document title, paragraph title, text, page number, abstract, table, references, footnotes, header, footer, algorithm, formula, formula number, image, table, seal, figure_table title, chart, and sidebar text and lists of references</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout_plus-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout_plus-L_pretrained.pdparams">Training Model</a></td>
+<td>83.2</td>
+<td>53.03 / 17.23</td>
+<td>634.62 / 378.32</td>
+<td>126.01</td>
+<td>A higher-precision layout area localization model trained on a self-built dataset containing Chinese and English papers, PPT, multi-layout magazines, contracts, books, exams, ancient books and research reports using RT-DETR-L</td>
+</tr>
+<tr>
+</tbody>
+</table>
+
+
+* <b>The region detection model includes 1 category: Block:</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocBlockLayout</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBlockLayout_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocBlockLayout_pretrained.pdparams">Training Model</a></td>
+<td>95.9</td>
+<td>34.60 / 28.54</td>
+<td>506.43 / 256.83</td>
+<td>123.92</td>
+<td>A layout block localization model trained on a self-built dataset containing Chinese and English papers, PPT, multi-layout magazines, contracts, books, exams, ancient books and research reports using RT-DETR-L</td>
+</tr>
+<tr>
+</tbody>
+</table>
+
+
+* <b>The layout detection model includes 23 common categories: document title, paragraph title, text, page number, abstract, table of contents, references, footnotes, header, footer, algorithm, formula, formula number, image, figure caption, table, table caption, seal, figure title, figure, header image, footer image, and sidebar text</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-L_pretrained.pdparams">Pretrained Model</a></td>
+<td>90.4</td>
+<td>33.59 / 33.59</td>
+<td>503.01 / 251.08</td>
+<td>123.76</td>
+<td>A high-precision layout area localization model trained on a self-built dataset containing Chinese and English papers, magazines, contracts, books, exams, and research reports using RT-DETR-L.</td>
+</tr>
+<tr>
+<td>PP-DocLayout-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-M_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-M_pretrained.pdparams">Pretrained Model</a></td>
+<td>75.2</td>
+<td>13.03 / 4.72</td>
+<td>43.39 / 24.44</td>
+<td>22.578</td>
+<td>A layout area localization model with balanced precision and efficiency, trained on a self-built dataset containing Chinese and English papers, magazines, contracts, books, exams, and research reports using PicoDet-L.</td>
+</tr>
+<tr>
+<td>PP-DocLayout-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-S_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-S_pretrained.pdparams">Pretrained Model</a></td>
+<td>70.9</td>
+<td>11.54 / 3.86</td>
+<td>18.53 / 6.29</td>
+<td>4.834</td>
+<td>A high-efficiency layout area localization model trained on a self-built dataset containing Chinese and English papers, magazines, contracts, books, exams, and research reports using PicoDet-S.</td>
+</tr>
+</tbody>
+</table>
+
+> ❗ The above list includes the <b>4 core models</b> that are key supported by the text recognition module. The module actually supports a total of <b>12 full models</b>, including several predefined models with different categories. The complete model list is as follows:
+
+<details><summary> 👉 Details of Model List</summary>
+
+* <b>Table Layout Detection Model</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet_layout_1x_table</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_table_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_table_pretrained.pdparams">Training Model</a></td>
+<td>97.5</td>
+<td>9.57 / 6.63</td>
+<td>27.66 / 16.75</td>
+<td>7.4</td>
+<td>A high-efficiency layout area localization model trained on a self-built dataset using PicoDet-1x, capable of detecting table regions.</td>
+</tr>
+</tbody></table>
+
+* <b>3-Class Layout Detection Model, including Table, Image, and Stamp</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_3cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_3cls_pretrained.pdparams">Training Model</a></td>
+<td>88.2</td>
+<td>8.43 / 3.44</td>
+<td>17.60 / 6.51</td>
+<td>4.8</td>
+<td>A high-efficiency layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using PicoDet-S.</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_3cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_3cls_pretrained.pdparams">Training Model</a></td>
+<td>89.0</td>
+<td>12.80 / 9.57</td>
+<td>45.04 / 23.86</td>
+<td>22.6</td>
+<td>A balanced efficiency and precision layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using PicoDet-L.</td>
+</tr>
+</table>
+
+<p><b>Table Classification Module Models:</b></p>
+<table>
+<tr>
+<td>RT-DETR-H_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_3cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_3cls_pretrained.pdparams">Training Model</a></td>
+<td>95.8</td>
+<td>114.80 / 25.65</td>
+<td>924.38 / 924.38</td>
+<td>470.1</td>
+<td>A high-precision layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using RT-DETR-H.</td>
+</tr>
+</tbody></table>
+
+* <b>5-Class English Document Area Detection Model, including Text, Title, Table, Image, and List</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet_layout_1x</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_pretrained.pdparams">Training Model</a></td>
+<td>97.8</td>
+<td>9.62 / 6.75</td>
+<td>26.96 / 12.77</td>
+<td>7.4</td>
+<td>A high-efficiency English document layout area localization model trained on the PubLayNet dataset using PicoDet-1x.</td>
+</tr>
+</tbody></table>
+
+* <b>17-Class Area Detection Model, including 17 common layout categories: Paragraph Title, Image, Text, Number, Abstract, Content, Figure Caption, Formula, Table, Table Caption, References, Document Title, Footnote, Header, Algorithm, Footer, and Stamp</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>87.4</td>
+<td>8.80 / 3.62</td>
+<td>17.51 / 6.35</td>
+<td>4.8</td>
+<td>A high-efficiency layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using PicoDet-S.</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>89.0</td>
+<td>12.60 / 10.27</td>
+<td>43.70 / 24.42</td>
+<td>22.6</td>
+<td>A balanced efficiency and precision layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using PicoDet-L.</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>98.3</td>
+<td>115.29 / 101.18</td>
+<td>964.75 / 964.75</td>
+<td>470.2</td>
+<td>A high-precision layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using RT-DETR-H.</td>
+</tr>
+</table>
+</details>
+</details>
+<details>
+<summary><b>Table Structure Recognition Module (Optional):</b></summary>
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>mAP (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>SLANeXt_wired</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wired_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wired_pretrained.pdparams">Training Model</a></td>
+<td rowspan="2">69.65</td>
+<td rowspan="2">85.92 / 85.92</td>
+<td rowspan="2">- / 501.66</td>
+<td rowspan="2">351</td>
+<td rowspan="2">The SLANeXt series is a new generation of table structure recognition models independently developed by the Baidu PaddlePaddle Vision Team. Compared to SLANet and SLANet_plus, SLANeXt focuses on table structure recognition, and trains dedicated weights for wired and wireless tables separately. The recognition ability for all types of tables has been significantly improved, especially for wired tables.</td>
+</tr>
+<tr>
+<td>SLANeXt_wireless</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wireless_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wireless_pretrained.pdparams">Training Model</a></td>
+</tr>
+</table>
+
+<p><b>Table Classification Module Models:</b></p>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Top1 Acc (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+</tr>
+<tr>
+<td>PP-LCNet_x1_0_table_cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/CLIP_vit_base_patch16_224_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_table_cls_pretrained.pdparams">Training Model</a></td>
+<td>94.2</td>
+<td>2.62 / 0.60</td>
+<td>3.17 / 1.14</td>
+<td>6.6</td>
+</tr>
+</table>
+<p><b>Table Cell Detection Module Models:</b></p>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>RT-DETR-L_wired_table_cell_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wired_table_cell_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wired_table_cell_det_pretrained.pdparams">Training Model</a></td>
+<td rowspan="2">82.7</td>
+<td rowspan="2">33.47 / 27.02</td>
+<td rowspan="2">402.55 / 256.56</td>
+<td rowspan="2">124</td>
+<td rowspan="2">RT-DETR is the first real-time end-to-end object detection model. The Baidu PaddlePaddle vision team based RT-DETR-L as the base model, completing pre-training on a self-built table cell detection dataset, achieving good performance in detecting both wired and wireless table cells.</td>
+</tr>
+<tr>
+<td>RT-DETR-L_wireless_table_cell_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wireless_table_cell_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wireless_table_cell_det_pretrained.pdparams">Training Model</a></td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary><b>Text Detection Module (Required):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Detection Hmean (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv5_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_det_pretrained.pdparams">Training Model</a></td>
+<td>83.8</td>
+<td>89.55 / 70.19</td>
+<td>383.15 / 383.15</td>
+<td>84.3</td>
+<td>PP-OCRv5 server-side text detection model with higher accuracy, suitable for deployment on high-performance servers</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_det_pretrained.pdparams">Training Model</a></td>
+<td>79.0</td>
+<td>10.67 / 6.36</td>
+<td>57.77 / 28.15</td>
+<td>4.7</td>
+<td>PP-OCRv5 mobile-side text detection model with higher efficiency, suitable for deployment on edge devices</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_det_pretrained.pdparams">Training Model</a></td>
+<td>82.56</td>
+<td>127.82 / 98.87</td>
+<td>585.95 / 489.77</td>
+<td>109</td>
+<td>The server-side text detection model of PP-OCRv4, with higher accuracy, suitable for deployment on high-performance servers.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams">Training Model</a></td>
+<td>63.8</td>
+<td>9.87 / 4.17</td>
+<td>56.60 / 20.79</td>
+<td>4.7</td>
+<td>The mobile text detection model of PP-OCRv4, with higher efficiency, suitable for deployment on edge devices.</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Text Recognition Module Model (Required):</b></summary>
+
+
+<details><summary> 👉Full Model List</summary>
+
+* <b>PP-OCRv5 Multi-Scenario Models</b>
+
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Chinese Avg Accuracy (%)</th>
+<th>English Avg Accuracy (%)</th>
+<th>Traditional Chinese Avg Accuracy (%)</th>
+<th>Japanese Avg Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>86.38</td>
+<td>64.70</td>
+<td>93.29</td>
+<td>60.35</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td>PP-OCRv5_server_rec is a new-generation text recognition model. It efficiently and accurately supports four major languages: Simplified Chinese, Traditional Chinese, English, and Japanese, as well as handwriting, vertical text, pinyin, and rare characters, offering robust and efficient support for document understanding.</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>81.29</td>
+<td>66.00</td>
+<td>83.55</td>
+<td>54.65</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>136</td>
+<td>PP-OCRv5_mobile_rec is a new-generation text recognition model. It efficiently and accurately supports four major languages: Simplified Chinese, Traditional Chinese, English, and Japanese, as well as handwriting, vertical text, pinyin, and rare characters, offering robust and efficient support for document understanding.</td>
+</tr>
+</table>
+
+* <b>Chinese Recognition Models</b>
+
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Avg Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_doc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">Pretrained Model</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>Based on PP-OCRv4_server_rec, trained on additional Chinese documents and PP-OCR mixed data. It supports over 15,000 characters including Traditional Chinese, Japanese, and special symbols, enhancing both document-specific and general text recognition accuracy.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>Lightweight model of PP-OCRv4 with high inference efficiency, suitable for deployment on various edge devices.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>Server-side model of PP-OCRv4 with high recognition accuracy, suitable for deployment on various servers.</td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>72.96</td>
+<td>3.89 / 1.16</td>
+<td>8.72 / 3.56</td>
+<td>10.3</td>
+<td>Lightweight model of PP-OCRv3 with high inference efficiency, suitable for deployment on various edge devices.</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Avg Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>ch_SVTRv2_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_SVTRv2_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_SVTRv2_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>68.81</td>
+<td>10.38 / 8.31</td>
+<td>66.52 / 30.83</td>
+<td>80.5</td>
+<td>SVTRv2 is a server-side recognition model developed by the OpenOCR team at Fudan University’s FVL Lab. It won first place in the OCR End-to-End Recognition task of the PaddleOCR Model Challenge, improving end-to-end accuracy on Benchmark A by 6% compared to PP-OCRv4.</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Avg Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>ch_RepSVTR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_RepSVTR_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_RepSVTR_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>65.07</td>
+<td>6.29 / 1.57</td>
+<td>20.64 / 5.40</td>
+<td>48.8</td>
+<td>RepSVTR is a mobile text recognition model based on SVTRv2. It won first place in the OCR End-to-End Recognition task of the PaddleOCR Model Challenge, improving accuracy on Benchmark B by 2.5% over PP-OCRv4 with comparable inference speed.</td>
+</tr>
+</table>
+
+* <b>English Recognition Models</b>
+
+<table>
+<tr>
+<th>Model</th><th>Download Link</th>
+<th>Avg Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Standard Mode / High Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>Ultra-lightweight English recognition model trained on PP-OCRv4, supporting English and number recognition.</td>
+</tr>
+<tr>
+<td>en_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/en_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>70.69</td>
+<td>3.56 / 0.78</td>
+<td>8.44 / 5.78</td>
+<td>17.3</td>
+<td>Ultra-lightweight English recognition model trained on PP-OCRv3, supporting English and number recognition.</td>
+</tr>
+</table>
+
+
+
+
+
+
+
+
+* <b>Multilingual Recognition Models</b>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal / High Performance]</th>
+<th>CPU Inference Time (ms)<br/>[Normal / High Performance]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>korean_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>60.21</td>
+<td>3.73 / 0.98</td>
+<td>8.76 / 2.91</td>
+<td>9.6</td>
+<td>An ultra-lightweight Korean text recognition model trained based on PP-OCRv3, supporting Korean and digits recognition</td>
+</tr>
+<tr>
+<td>japan_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+japan_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/japan_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>45.69</td>
+<td>3.86 / 1.01</td>
+<td>8.62 / 2.92</td>
+<td>9.8</td>
+<td>An ultra-lightweight Japanese text recognition model trained based on PP-OCRv3, supporting Japanese and digits recognition</td>
+</tr>
+<tr>
+<td>chinese_cht_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+chinese_cht_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/chinese_cht_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>82.06</td>
+<td>3.90 / 1.16</td>
+<td>9.24 / 3.18</td>
+<td>10.8</td>
+<td>An ultra-lightweight Traditional Chinese text recognition model trained based on PP-OCRv3, supporting Traditional Chinese and digits recognition</td>
+</tr>
+<tr>
+<td>te_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+te_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>95.88</td>
+<td>3.59 / 0.81</td>
+<td>8.28 / 6.21</td>
+<td>8.7</td>
+<td>An ultra-lightweight Telugu text recognition model trained based on PP-OCRv3, supporting Telugu and digits recognition</td>
+</tr>
+<tr>
+<td>ka_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ka_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ka_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>96.96</td>
+<td>3.49 / 0.89</td>
+<td>8.63 / 2.77</td>
+<td>17.4</td>
+<td>An ultra-lightweight Kannada text recognition model trained based on PP-OCRv3, supporting Kannada and digits recognition</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ta_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>76.83</td>
+<td>3.49 / 0.86</td>
+<td>8.35 / 3.41</td>
+<td>8.7</td>
+<td>An ultra-lightweight Tamil text recognition model trained based on PP-OCRv3, supporting Tamil and digits recognition</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>76.93</td>
+<td>3.53 / 0.78</td>
+<td>8.50 / 6.83</td>
+<td>8.7</td>
+<td>An ultra-lightweight Latin text recognition model trained based on PP-OCRv3, supporting Latin and digits recognition</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+arabic_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>73.55</td>
+<td>3.60 / 0.83</td>
+<td>8.44 / 4.69</td>
+<td>17.3</td>
+<td>An ultra-lightweight Arabic script recognition model trained based on PP-OCRv3, supporting Arabic script and digits recognition</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+cyrillic_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>94.28</td>
+<td>3.56 / 0.79</td>
+<td>8.22 / 2.76</td>
+<td>8.7</td>
+<td>An ultra-lightweight Cyrillic script recognition model trained based on PP-OCRv3, supporting Cyrillic script and digits recognition</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+devanagari_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv3_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>96.44</td>
+<td>3.60 / 0.78</td>
+<td>6.95 / 2.87</td>
+<td>8.7</td>
+<td>An ultra-lightweight Devanagari script recognition model trained based on PP-OCRv3, supporting Devanagari script and digits recognition</td>
+</tr>
+</table>
+</details>
+</details>
+
+<details>
+<summary><b>Text Line Orientation Classification Module (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th>
+<th>Model Download Link</th>
+<th>Top-1 Acc (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal / High Performance]</th>
+<th>CPU Inference Time (ms)<br/>[Normal / High Performance]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x0_25_textline_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x0_25_textline_ori_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x0_25_textline_ori_pretrained.pdparams">Pretrained Model</a></td>
+<td>98.85</td>
+<td>2.16 / 0.41</td>
+<td>2.37 / 0.73</td>
+<td>0.96</td>
+<td>A text line classification model based on PP-LCNet_x0_25, containing two categories: 0 degrees and 180 degrees</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Formula Recognition Module (Optional):</b></summary>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>En-BLEU(%)</th>
+<th>Zh-BLEU(%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+<tr>
+<td>UniMERNet</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UniMERNet_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UniMERNet_pretrained.pdparams">Training Model</a></td>
+<td>85.91</td>
+<td>43.50</td>
+<td>1311.84 / 1311.84</td>
+<td>- / 8288.07</td>
+<td>1530</td>
+<td>UniMERNet is a formula recognition model developed by Shanghai AI Lab. It uses Donut Swin as the encoder and MBartDecoder as the decoder. The model is trained on a dataset of one million samples, including simple formulas, complex formulas, scanned formulas, and handwritten formulas, significantly improving the recognition accuracy of real-world formulas.</td>
+</tr>
+<td>PP-FormulaNet-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-S_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-S_pretrained.pdparams">Training Model</a></td>
+<td>87.00</td>
+<td>45.71</td>
+<td>182.25 / 182.25</td>
+<td>- / 254.39</td>
+<td>224</td>
+<td rowspan="2">PP-FormulaNet is an advanced formula recognition model developed by the Baidu PaddlePaddle Vision Team. The PP-FormulaNet-S version uses PP-HGNetV2-B4 as its backbone network. Through parallel masking and model distillation techniques, it significantly improves inference speed while maintaining high recognition accuracy, making it suitable for applications requiring fast inference. The PP-FormulaNet-L version, on the other hand, uses Vary_VIT_B as its backbone network and is trained on a large-scale formula dataset, showing significant improvements in recognizing complex formulas compared to PP-FormulaNet-S.</td>
+</tr>
+<td>PP-FormulaNet-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-L_pretrained.pdparams">Training Model</a></td>
+<td>90.36</td>
+<td>45.78</td>
+<td>1482.03 / 1482.03</td>
+<td>- / 3131.54</td>
+<td>695</td>
+</tr>
+<td>PP-FormulaNet_plus-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-S_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-S_pretrained.pdparams">Training Model</a></td>
+<td>88.71</td>
+<td>53.32</td>
+<td>179.20 / 179.20</td>
+<td>- / 260.99</td>
+<td>248</td>
+<td rowspan="3">PP-FormulaNet_plus is an enhanced version of the formula recognition model developed by the Baidu PaddlePaddle Vision Team, building upon the original PP-FormulaNet. Compared to the original version, PP-FormulaNet_plus utilizes a more diverse formula dataset during training, including sources such as Chinese dissertations, professional books, textbooks, exam papers, and mathematics journals. This expansion significantly improves the model’s recognition capabilities. Among the models, PP-FormulaNet_plus-M and PP-FormulaNet_plus-L have added support for Chinese formulas and increased the maximum number of predicted tokens for formulas from 1,024 to 2,560, greatly enhancing the recognition performance for complex formulas. Meanwhile, the PP-FormulaNet_plus-S model focuses on improving the recognition of English formulas. With these improvements, the PP-FormulaNet_plus series models perform exceptionally well in handling complex and diverse formula recognition tasks. </td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-M_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-M_pretrained.pdparams">Training Model</a></td>
+<td>91.45</td>
+<td>89.76</td>
+<td>1040.27 / 1040.27</td>
+<td>- / 1615.80</td>
+<td>592</td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-L_pretrained.pdparams">Training Model</a></td>
+<td>92.22</td>
+<td>90.64</td>
+<td>1476.07 / 1476.07</td>
+<td>- / 3125.58</td>
+<td>698</td>
+</tr>
+<tr>
+<td>LaTeX_OCR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/LaTeX_OCR_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/LaTeX_OCR_rec_pretrained.pdparams">Training Model</a></td>
+<td>74.55</td>
+<td>39.96</td>
+<td>1088.89 / 1088.89</td>
+<td>- / -</td>
+<td>99</td>
+<td>LaTeX-OCR is a formula recognition algorithm based on an autoregressive large model. It uses Hybrid ViT as the backbone network and a transformer as the decoder, significantly improving the accuracy of formula recognition.</td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary><b>Seal Text Detection Module (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Detection Hmean (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal / High Performance]</th>
+<th>CPU Inference Time (ms)<br/>[Normal / High Performance]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv4_server_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_seal_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_seal_det_pretrained.pdparams">Pretrained Model</a></td>
+<td>98.40</td>
+<td>124.64 / 91.57</td>
+<td>545.68 / 439.86</td>
+<td>109</td>
+<td>Server-side seal text detection model based on PP-OCRv4, offering higher accuracy and suitable for deployment on high-performance servers</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_seal_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_seal_det_pretrained.pdparams">Pretrained Model</a></td>
+<td>96.36</td>
+<td>9.70 / 3.56</td>
+<td>50.38 / 19.64</td>
+<td>4.7</td>
+<td>Mobile-side seal text detection model based on PP-OCRv4, offering higher efficiency and suitable for edge-side deployment</td>
+</tr>
+</tbody>
+</table>
+</details>
+</details>
+
+<details>
+<summary><b>Chart Parsing Module: </b></summary>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Model parameter size（B）</th>
+<th>Model Storage Size (GB)</th>
+<th>Model Score </th>
+<th>Description</th>
+</tr>
+<tr>
+<td>PP-Chart2Table</td><td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-Chart2Table_infer.tar">Inference Model</a></td>
+<td>0.58</td>
+<td>1.4</td>
+<th>75.98</th>
+<td>PP-Chart2Table is a self-developed multimodal model by the PaddlePaddle team, focusing on chart parsing, demonstrating outstanding performance in both Chinese and English chart parsing tasks. The team adopted a carefully designed data generation strategy, constructing a high-quality multimodal dataset of nearly 700,000 entries covering common chart types like pie charts, bar charts, stacked area charts, and various application scenarios. They also designed a two-stage training method, utilizing large model distillation to fully leverage massive unlabeled OOD data. In internal business tests in both Chinese and English scenarios, PP-Chart2Table not only achieved the SOTA level among models of the same parameter scale but also reached accuracy comparable to 7B parameter scale VLM models in critical scenarios.</td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary><b>Test Environment Description:</b></summary>
+  <ul>
+      <li><b>Performance Test Environment</b>
+          <ul>
+          <li><strong>Test Dataset: </strong>
+                        <ul>
+                         <li>Document Image Orientation Classification Module: A self-built dataset using PaddleX, covering multiple scenarios such as ID cards and documents, containing 1000 images.</li>
+                          <li>Text Image Rectification Model: <a href="https://www3.cs.stonybrook.edu/~cvl/docunet.html">DocUNet</a></li>
+                          <li>Layout Region Detection Model: A self-built layout detection dataset using PaddleOCR, containing 10,000 images of common document types such as Chinese and English papers, magazines, and research reports.</li>
+                          <li>Table Structure Recognition Model: A self-built English table recognition dataset using PaddleX.</li>
+                          <li>Text Detection Model: A self-built Chinese dataset using PaddleOCR, covering multiple scenarios such as street scenes, web images, documents, and handwriting, with 500 images for detection.</li>
+                          <li>Chinese Recognition Model: A self-built Chinese dataset using PaddleOCR, covering multiple scenarios such as street scenes, web images, documents, and handwriting, with 11,000 images for text recognition.</li>
+                          <li>ch_SVTRv2_rec: Evaluation set A for "OCR End-to-End Recognition Task" in the <a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR Algorithm Model Challenge</a></li>
+                          <li>ch_RepSVTR_rec: Evaluation set B for "OCR End-to-End Recognition Task" in the <a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR Algorithm Model Challenge</a>.</li>
+                          <li>English Recognition Model: A self-built English dataset using PaddleX.</li>
+                          <li>Multilingual Recognition Model: A self-built multilingual dataset using PaddleX.</li>
+                          <li>Text Line Orientation Classification Model: A self-built dataset using PaddleX, covering various scenarios such as ID cards and documents, containing 1000 images.</li>
+                          <li>Seal Text Detection Model: A self-built dataset using PaddleX, containing 500 images of circular seal textures.</li>
+                        </ul>
+                </li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>Inference Mode Description</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration </th>
+            <th>CPU Configuration </th>
+            <th>Acceleration Technology Combination</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Normal Mode</td>
+            <td>FP32 Precision / No TRT Acceleration</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Local Paddle inference engines (by default, the engine is selected according to the default model name; if both static and dynamic graph are available, <code>paddle_static</code> is preferred)</td>
+        </tr>
+        <tr>
+            <td>High-Performance Mode</td>
+            <td>Optimal combination of pre-selected precision types and acceleration strategies</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Pre-selected optimal backend (Paddle/OpenVINO/TRT, etc.)</td>
+        </tr>
+    </tbody>
+</table>
+
+</details>
+
+## 2. Quick Start
+
+Before using the PP-StructureV3 pipeline locally, please make sure you have completed the installation of the wheel package according to the [installation guide](../installation.en.md). If you prefer to install dependencies selectively, please refer to the relevant instructions in the installation documentation. The corresponding dependency group for this pipeline is <code>doc-parser</code>. After installation, you can use it via command line or Python integration.
+
+Please note: If you encounter issues such as the program becoming unresponsive, unexpected program termination, running out of memory resources, or extremely slow inference during execution, please try adjusting the configuration according to the documentation, such as disabling unnecessary features or using lighter-weight models.
+
+### 2.1 Command Line Usage
+
+Use a single command to quickly experience the PP-StructureV3 pipeline:
+
+```bash
+paddleocr pp_structurev3 -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/pp_structure_v3_demo.png
+
+# Use --use_doc_orientation_classify to enable document orientation classification
+paddleocr pp_structurev3 -i ./pp_structure_v3_demo.png --use_doc_orientation_classify True
+
+# Use --use_doc_unwarping to enable document unwarping module
+paddleocr pp_structurev3 -i ./pp_structure_v3_demo.png --use_doc_unwarping True
+
+# Use --use_textline_orientation to enable text line orientation classification
+paddleocr pp_structurev3 -i ./pp_structure_v3_demo.png --use_textline_orientation False
+
+# Use --device to specify GPU for inference
+paddleocr pp_structurev3 -i ./pp_structure_v3_demo.png --device gpu
+```
+
+The examples above use local Paddle inference engines by default. By default, each module selects the appropriate local Paddle inference engine according to the default model name: models that support only dynamic graph use `paddle_dynamic`, while models that support both static and dynamic graph prefer `paddle_static`. To run them, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr pp_structurev3 -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/pp_structure_v3_demo.png \
+    --engine transformers
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following command:
+
+```bash
+# Use the onnxruntime engine for inference
+# Some models are still being supported. For inference, please disable formula recognition using the following command:
+paddleocr pp_structurev3 -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/pp_structure_v3_demo.png \
+    --engine onnxruntime --use_formula_recognition False
+```
+
+<details><summary><b>Command line supports more parameters. Click to expand for detailed parameter descriptions</b></summary>
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted. Required.<br/>
+<b>Description:</b>
+.e.g., local path to image or PDF file: <code>/root/data/img.jpg</code>; <b>URL</b>, e.g., online image or PDF: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/demo_paper.png">example</a>; <b>local directory</b>: directory containing images to predict, e.g., <code>/root/data/</code> (currently, directories with PDFs are not supported; PDFs must be specified by file path).
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td><b>Meaning:</b>Path to save inference results.<br/>
+<b>Description:</b> 
+If not set, results will not be saved locally.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the layout detection model.<br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the layout detection model.<br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>Score threshold for the layout model.<br/>
+<b>Description:</b> 
+Any value between <code>0-1</code>. If not set, the default value is used, which is <code>0.5</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Whether to use Non-Maximum Suppression (NMS) as post-processing for layout detection. <br/>
+<b>Description:</b> 
+If not set, the parameter will default to the value initialized in the pipeline, which is set to <code>True</code> by default.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Unclip ratio for detected boxes in layout detection model.<br/>
+<b>Description:</b> 
+Any float > <code>0</code>. If not set, the default is <code>1.0</code>.
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>The merging mode for the detection boxes output by the model in layout detection.<br/>
+<b>Description:</b> 
+<ul>
+<li><b>large</b>: When set to "large", only the largest outer bounding box will be retained for overlapping bounding boxes, and the inner overlapping boxes will be removed;</li>
+<li><b>small</b>: When set to "small", only the smallest inner bounding boxes will be retained for overlapping bounding boxes, and the outer overlapping boxes will be removed;</li>
+<li><b>union</b>: No filtering of bounding boxes will be performed, and both inner and outer boxes will be retained;</li>
+</ul>If not set, the default is <code>large</code>.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>chart_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the chart parsing model.<br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>chart_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the chart parsing model.<br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>chart_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the chart parsing model.<br/>
+<b>Description:</b> 
+If not set, the default batch size is <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>region_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the region detection model.<br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>region_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the region detection model.<br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>Name of the document orientation classification model.<br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the document orientation classification model.<br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>Meaning:</b>Name of the document unwarping model.<br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the document unwarping model.<br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the text detection model.<br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text detection model.<br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limitation for text detection.<br/>
+<b>Description:</b> 
+Any integer > <code>0</code>. If not set, the default value will be <code>960</code>.
+</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>Meaning:</b>Type of the image side length limit for text detection.<br/>
+<b>Description:</b> 
+Supports <code>min</code> and <code>max</code>; <code>min</code> means ensuring the shortest side of the image is not less than <code>det_limit_side_len</code>, <code>max</code> means the longest side does not exceed <code>limit_side_len</code>. If not set, the default value will be <code>max</code>.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>Meaning:</b>Pixel threshold for text detection. Pixels with scores above this value in the probability map are considered text.<br/>
+<b>Description:</b> 
+Any float > <code>0</code>. If not set, the default is <code>0.3</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>Meaning:</b>Box threshold for text detection. A bounding box is considered text if the average score of pixels inside is greater than this value.<br/>
+<b>Description:</b> 
+Any float > <code>0</code>. If not set, the default is <code>0.6</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion ratio for text detection. The higher the value, the larger the expansion area.<br/>
+<b>Description:</b> 
+Any float > <code>0</code>. If not set, the default is <code>2.0</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_name</code></td>
+<td><b>Meaning:</b>Name of the text line orientation model.<br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text line orientation model.<br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the text line orientation model.<br/>
+<b>Description:</b> 
+If not set, the default is <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the text recognition model.<br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory of the text recognition model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for text recognition.<br/>
+<b>Description:</b> 
+If not set, the default is <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Score threshold for text recognition. Only results above this value will be kept.<br/>
+<b>Description:</b> 
+Any float > <code>0</code>. If not set, the default is <code>0.0</code> (no threshold).
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_classification_model_name</code></td>
+<td><b>Meaning:</b>Name of the table classification model.<br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_classification_model_dir</code></td>
+<td><b>Meaning:</b>Directory of the table classification model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the wired table structure recognition model.<br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory of the wired table structure recognition model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the wireless table structure recognition model. <br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory of the wireless table structure recognition model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the wired table cell detection model. <br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory of the wired table cell detection model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the wireless table cell detection model. <br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory of the wireless table cell detection model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>Name of the wireless table orientation classification model. <br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>Directory of the table orientation classification model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the seal text detection model. <br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory of the seal text detection model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limit for seal text detection. <br/>
+<b>Description:</b> 
+Any integer > <code>0</code>. If not set, the default is <code>736</code>.
+</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>Meaning:</b>Limit type for image side in seal text detection. <br/>
+<b>Description:</b> 
+Supports <code>min</code> and <code>max</code>; <code>min</code> ensures shortest side ≥ <code>det_limit_side_len</code>, <code>max</code> ensures longest side ≤ <code>limit_side_len</code>. If not set, the default is <code>min</code>.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>Meaning:</b>Pixel threshold. Pixels with scores above this value in the probability map are considered text. <br/>
+<b>Description:</b> 
+Any float > <code>0</code>. If not set, the default is <code>0.2</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>Meaning:</b>Box threshold. Boxes with average pixel scores above this value are considered text regions. <br/>
+<b>Description:</b> 
+Any float > <code>0</code>. If not set, the default is <code>0.6</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion ratio for seal text detection. Higher value means larger expansion area. <br/>
+<b>Description:</b> 
+Any float > <code>0</code>. If not set, the default is <code>0.5</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the seal text recognition model. <br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory of the seal text recognition model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for seal text recognition. <br/>
+<b>Description:</b> 
+If not set, the default is <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Recognition score threshold. Text results above this value will be kept. <br/>
+<b>Description:</b> 
+Any float > <code>0</code>. If not set, the default is <code>0.0</code> (no threshold).
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the formula recognition model. <br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory of the formula recognition model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>formula_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size of the formula recognition model. <br/>
+<b>Description:</b> 
+If not set, the default is <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to load and use the document orientation classification module. <br/>
+<b>Description:</b> 
+If not set, the default is <code>False</code>.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to load and use the document unwarping module. <br/>
+<b>Description:</b> 
+If not set, the default is <code>False</code>.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>Meaning:</b>Whether to load and use the text line orientation classification module. <br/>
+<b>Description:</b> 
+If not set, the default is <code>False</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>Meaning:</b>Whether to load and use seal text recognition subpipeline. <br/>
+<b>Description:</b> 
+If not set, the default is <code>False</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>Meaning:</b>Whether to load and use table recognition subpipeline. <br/>
+<b>Description:</b> 
+If not set, the default is <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_formula_recognition</code></td>
+<td><b>Meaning:</b>Whether to load and use formula recognition subpipeline. <br/>
+<b>Description:</b> 
+If not set, the default is <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>Meaning:</b>Whether to load and use the chart parsing module. <br/>
+<b>Description:</b> 
+If not set, the default is <code>False</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_region_detection</code></td>
+<td><b>Meaning:</b>Whether to load and use the document region detection module. <br/>
+<b>Description:</b> 
+If not set, the default is <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>format_block_content</code></td>
+<td><b>Meaning:</b>Whether to format the content in <code>block_content</code> as Markdown.<br/>
+<b>Description:</b> If not set, the initialized default value will be used, which is <code>False</code> by default.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>markdown_ignore_labels</code></td>
+<td><b>Meaning:</b>Layout tags that need to be ignored in Markdown.<br/>
+<b>Description:</b> If not set, the initialized default value will be used, which is <code>['number','footnote','header','header_image','footer','footer_image','aside_text']</code> by default.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>Device for inference. <br/>
+<b>Description:</b> 
+You can specify a device ID:
+<ul>
+<li><b>CPU</b>: e.g., <code>cpu</code> means using CPU for inference;</li>
+<li><b>GPU</b>: e.g., <code>gpu:0</code> means GPU 0</li>
+<li><b>NPU</b>: e.g., <code>npu:0</code> means NPU 0</li>
+<li><b>XPU</b>: e.g., <code>xpu:0</code> means XPU 0</li>
+<li><b>MLU</b>: e.g., <code>mlu:0</code> means MLU 0</li>
+<li><b>DCU</b>: e.g., <code>dcu:0</code> means DCU 0</li>
+<li><b>MetaX GPU</b>: e.g., <code>metax_gpu:0</code> means MetaX GPU 0</li>
+<li><b>Iluvatar GPU</b>: e.g., <code>iluvatar_gpu:0</code> means Iluvatar GPU 0</li>
+</ul>If not set, the pipeline initialized value for this parameter will be used. During initialization, the local GPU device 0 will be preferred; if unavailable, the CPU device will be used.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to enable the TensorRT subgraph engine of Paddle Inference.<br/>
+<b>Description:</b>
+If the model does not support TensorRT acceleration, acceleration will not be used even if this flag is set.<br/>
+For CUDA 11.8 versions of PaddlePaddle, the compatible TensorRT version is 8.x (x>=6). TensorRT 8.6.1.6 is recommended.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b> Computation precision, such as <code>fp32</code> or <code>fp16</code>.</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b> Whether to enable MKL-DNN accelerated inference.<br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support MKL-DNN acceleration, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b> MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads used for inference on CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b> Path to the PaddleX pipeline configuration file.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+</tbody>
+</table>
+</details>
+<br />
+
+The inference result will be printed in the terminal. The default output of the PP-StructureV3 pipeline is as follows:
+
+<details><summary> 👉Click to expand</summary>
+<pre>
+<code>
+{'res': {'input_path': 'pp_structure_v3_demo.png', 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_general_ocr': True, 'use_seal_recognition': True, 'use_table_recognition': True, 'use_formula_recognition': True}, 'layout_det_res': {'input_path': None, 'page_index': None, 'boxes': [{'cls_id': 2, 'label': 'text', 'score': 0.9853514432907104, 'coordinate': [770.9531, 776.6814, 1122.6057, 1058.7322]}, {'cls_id': 1, 'label': 'image', 'score': 0.9848673939704895, 'coordinate': [775.7434, 202.27979, 1502.8113, 686.02136]}, {'cls_id': 2, 'label': 'text', 'score': 0.983731746673584, 'coordinate': [1152.3197, 1113.3275, 1503.3029, 1346.586]}, {'cls_id': 2, 'label': 'text', 'score': 0.9832221865653992, 'coordinate': [1152.5602, 801.431, 1503.8436, 986.3563]}, {'cls_id': 2, 'label': 'text', 'score': 0.9829439520835876, 'coordinate': [9.549545, 849.5713, 359.1173, 1058.7488]}, {'cls_id': 2, 'label': 'text', 'score': 0.9811657667160034, 'coordinate': [389.58298, 1137.2659, 740.66235, 1346.7488]}, {'cls_id': 2, 'label': 'text', 'score': 0.9775941371917725, 'coordinate': [9.1302185, 201.85, 359.0409, 339.05692]}, {'cls_id': 2, 'label': 'text', 'score': 0.9750366806983948, 'coordinate': [389.71454, 752.96924, 740.544, 889.92456]}, {'cls_id': 2, 'label': 'text', 'score': 0.9738152027130127, 'coordinate': [389.94565, 298.55988, 740.5585, 435.5124]}, {'cls_id': 2, 'label': 'text', 'score': 0.9737328290939331, 'coordinate': [771.50256, 1065.4697, 1122.2582, 1178.7324]}, {'cls_id': 2, 'label': 'text', 'score': 0.9728517532348633, 'coordinate': [1152.5154, 993.3312, 1503.2349, 1106.327]}, {'cls_id': 2, 'label': 'text', 'score': 0.9725610017776489, 'coordinate': [9.372787, 1185.823, 359.31738, 1298.7227]}, {'cls_id': 2, 'label': 'text', 'score': 0.9724331498146057, 'coordinate': [389.62848, 610.7389, 740.83234, 746.2377]}, {'cls_id': 2, 'label': 'text', 'score': 0.9720287322998047, 'coordinate': [389.29898, 897.0936, 741.41516, 1034.6616]}, {'cls_id': 2, 'label': 'text', 'score': 0.9713053703308105, 'coordinate': [10.323685, 1065.4663, 359.6786, 1178.8872]}, {'cls_id': 2, 'label': 'text', 'score': 0.9689728021621704, 'coordinate': [9.336395, 537.6609, 359.2901, 652.1881]}, {'cls_id': 2, 'label': 'text', 'score': 0.9684857130050659, 'coordinate': [10.7608185, 345.95068, 358.93616, 434.64087]}, {'cls_id': 2, 'label': 'text', 'score': 0.9681928753852844, 'coordinate': [9.674866, 658.89075, 359.56528, 770.4319]}, {'cls_id': 2, 'label': 'text', 'score': 0.9634978175163269, 'coordinate': [770.9464, 1281.1785, 1122.6522, 1346.7156]}, {'cls_id': 2, 'label': 'text', 'score': 0.96304851770401, 'coordinate': [390.0113, 201.28055, 740.1684, 291.53073]}, {'cls_id': 2, 'label': 'text', 'score': 0.962053120136261, 'coordinate': [391.21393, 1040.952, 740.5046, 1130.32]}, {'cls_id': 2, 'label': 'text', 'score': 0.9565253853797913, 'coordinate': [10.113251, 777.1482, 359.439, 842.437]}, {'cls_id': 2, 'label': 'text', 'score': 0.9497362375259399, 'coordinate': [390.31357, 537.86285, 740.47595, 603.9285]}, {'cls_id': 2, 'label': 'text', 'score': 0.9371236562728882, 'coordinate': [10.2034, 1305.9753, 359.5958, 1346.7295]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.9338151216506958, 'coordinate': [791.6062, 1200.8479, 1103.3257, 1259.9324]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.9326773285865784, 'coordinate': [408.0737, 457.37024, 718.9509, 516.63464]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.9274250864982605, 'coordinate': [29.448685, 456.6762, 340.99194, 515.6999]}, {'cls_id': 2, 'label': 'text', 'score': 0.8742568492889404, 'coordinate': [1154.7095, 777.3624, 1330.3086, 794.5853]}, {'cls_id': 2, 'label': 'text', 'score': 0.8442489504814148, 'coordinate': [586.49316, 160.15454, 927.468, 179.64203]}, {'cls_id': 11, 'label': 'doc_title', 'score': 0.8332607746124268, 'coordinate': [133.80017, 37.41908, 1380.8601, 124.1429]}, {'cls_id': 6, 'label': 'figure_title', 'score': 0.6770150661468506, 'coordinate': [812.1718, 705.1199, 1484.6973, 747.1692]}]}, 'overall_ocr_res': {'input_path': None, 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_textline_orientation': False}, 'dt_polys': array([[[133,  35],
+        ...,
+        [133, 131]],
+
+       ...,
+
+       [[ 13, 754],
+        ...,
+        [ 13, 777]]], dtype=int16), 'text_det_params': {'limit_side_len': 960, 'limit_type': 'max', 'thresh': 0.3, 'box_thresh': 0.6, 'unclip_ratio': 2.0}, 'text_type': 'general', 'textline_orientation_angles': array([-1, ..., -1]), 'text_rec_score_thresh': 0.0, 'rec_texts': ['助力双方交往', '搭建友谊桥梁', '本报记者', '沈小晓', '任', '彦', '黄培昭', '身着中国传统民族服装的厄立特里亚青', '厄立特里亚高等教育与研究院合作建立，开', '年依次登台表演中国民族舞、现代舞、扇子舞', '设了中国语言课程和中国文化课程，注册学', '等,曼妙的舞姿赢得现场观众阵阵掌声。这', '生2万余人次。10余年来，厄特孔院已成为', '是日前厄立特里亚高等教育与研究院孔子学', '当地民众了解中国的一扇窗口。', '院(以下简称"厄特孔院")举办"喜迎新年"中国', '黄鸣飞表示,随着来学习中文的人日益', '歌舞比赛的场景。', '增多，阿斯马拉大学教学点已难以满足教学', '中国和厄立特里亚传统友谊深厚。近年', '需要。2024年4月，由中企蜀道集团所属四', '来,在高质量共建"一带一路"框架下，中厄两', '川路桥承建的孔院教学楼项目在阿斯马拉开', '国人文交流不断深化，互利合作的民意基础', '工建设,预计今年上半年竣工,建成后将为厄', '日益深厚。', '特孔院提供全新的办学场地。', '“学好中文，我们的', '“在中国学习的经历', '未来不是梦”', '让我看到更广阔的世界”', '多年来,厄立特里亚广大赴华留学生和', '培训人员积极投身国家建设,成为助力该国', '发展的人才和厄中友好的见证者和推动者。', '在厄立特里亚全国妇女联盟工作的约翰', '娜·特韦尔德·凯莱塔就是其中一位。她曾在', '中华女子学院攻读硕士学位,研究方向是女', '性领导力与社会发展。其间，她实地走访中国', '多个地区，获得了观察中国社会发展的第一', '在厄立特里亚不久前举办的第六届中国风筝文化节上，当地小学生体验风筝制作。', '手资料。', '中国驻厄立特里亚大使馆供图', '“这是中文歌曲初级班，共有32人。学', '“不管远近都是客人，请不用客气;相约', '瓦的北红海省博物馆。', '生大部分来自首都阿斯马拉的中小学，年龄', '好了在一起,我们欢迎你"在一场中厄青', '博物馆二层陈列着一个发掘自阿杜利', '最小的仅有6岁。"尤斯拉告诉记者。', '年联谊活动上,四川路桥中方员工同当地大', '斯古城的中国古代陶制酒器,罐身上写着', '尤斯拉今年23岁，是厄立特里亚一所公立', '学生合唱《北京欢迎你》。厄立特里亚技术学', '“万”“和""禅”“山"等汉字。“这件文物证', '学校的艺术老师。她12岁开始在厄特孔院学', '院计算机科学与工程专业学生鲁夫塔·谢拉', '明,很早以前我们就通过海上丝绸之路进行', '习中文,在2017年第十届"汉语桥"世界中学生', '是其中一名演唱者,她很早便在孔院学习中', '贸易往来与文化交流。这也是厄立特里亚', '中文比赛中获得厄立特里亚赛区第一名,并和', '文，一直在为去中国留学作准备。“这句歌词', '与中国友好交往历史的有力证明。"北红海', '同伴代表厄立特里亚前往中国参加决赛,获得', '是我们两国人民友谊的生动写照。无论是投', '省博物馆研究与文献部负责人伊萨亚斯·特', '团体优胜奖。2022年起，尤斯拉开始在厄特孔', '身于厄立特里亚基础设施建设的中企员工，', '斯法兹吉说。', '院兼职教授中文歌曲,每周末两个课时。中国', '还是在中国留学的厄立特里亚学子,两国人', '厄立特里亚国家博物馆考古学和人类学', '文化博大精深,我希望我的学生们能够通过中', '民携手努力,必将推动两国关系不断向前发', '研究员菲尔蒙·特韦尔德十分喜爱中国文', '文歌曲更好地理解中国文化。"她说。', '穆卢盖塔密切关注中国在经济、科技、教', '展。"鲁夫塔说。', '化。他表示：“学习彼此的语言和文化，将帮', '“姐姐,你想去中国吗?"“非常想！我想', '育等领域的发展，“中国在科研等方面的实力', '厄立特里亚高等教育委员会主任助理萨', '助厄中两国人民更好地理解彼此，助力双方', '去看故宫、爬长城。"尤斯拉的学生中有一对', '与日俱增。在中国学习的经历让我看到更广', '马瑞表示：“每年我们都会组织学生到中国访', '交往,搭建友谊桥梁。"', '能歌善舞的姐妹,姐姐露娅今年15岁，妹妹', '阔的世界，从中受益匪浅。', '问学习，目前有超过5000名厄立特里亚学生', '厄立特里亚国家博物馆馆长塔吉丁·努', '莉娅14岁，两人都已在厄特孔院学习多年，', '23岁的莉迪亚·埃斯蒂法诺斯已在厄特', '在中国留学。学习中国的教育经验,有助于', '里达姆·优素福曾多次访问中国，对中华文明', '中文说得格外流利。', '孔院学习3年，在中国书法、中国画等方面表', '提升厄立特里亚的教育水平。”', '的传承与创新、现代化博物馆的建设与发展', '露娅对记者说：“这些年来,怀着对中文', '现十分优秀，在2024年厄立特里亚赛区的', '“共同向世界展示非', '印象深刻。“中国博物馆不仅有许多保存完好', '和中国文化的热爱,我们姐妹俩始终相互鼓', '“汉语桥"比赛中获得一等奖。莉迪亚说：“学', '的文物,还充分运用先进科技手段进行展示，', '励,一起学习。我们的中文一天比一天好,还', '习中国书法让我的内心变得安宁和纯粹。我', '洲和亚洲的灿烂文明”', '帮助人们更好理解中华文明。"塔吉丁说，厄', '学会了中文歌和中国舞。我们一定要到中国', '也喜欢中国的服饰,希望未来能去中国学习，', '立特里亚与中国都拥有悠久的文明,始终相', '去。学好中文,我们的未来不是梦!"', '把中国不同民族元素融入服装设计中，创作', '从阿斯马拉出发,沿着蜿蜒曲折的盘山', '互理解、相互尊重。我希望未来与中国同行', '据厄特孔院中方院长黄鸣飞介绍,这所', '出更多精美作品，也把厄特文化分享给更多', '公路一路向东寻找丝路印迹。驱车两个小', '加强合作,共同向世界展示非洲和亚洲的灿', '孔院成立于2013年3月，由贵州财经大学和', '的中国朋友。”', '时,记者来到位于厄立特里亚港口城市马萨', '烂文明。”', '谈起在中国求学的经历,约翰娜记忆犹', '新：“中国的发展在当今世界是独一无二的。', '沿着中国特色社会主义道路坚定前行，中国', '创造了发展奇迹,这一切都离不开中国共产党', '的领导。中国的发展经验值得许多国家学习', '借鉴，”', '正在西南大学学习的厄立特里亚博士生', '穆卢盖塔·泽穆伊对中国怀有深厚感情。8', '年前，在北京师范大学获得硕士学位后，穆卢', '盖塔在社交媒体上写下这样一段话：“这是我', '人生的重要一步，自此我拥有了一双坚固的', '鞋子.赋予我穿越荆棘的力量。”', '“鲜花曾告诉我你怎样走过，大地知道你', '心中的每一个角落"厄立特里亚阿斯马拉', '大学综合楼二层，一阵优美的歌声在走廊里回', '响。循着熟悉的旋律轻轻推开一间教室的门，', '学生们正跟着老师学唱中文歌曲《同一首歌》。', '这是厄特孔院阿斯马拉大学教学点的一', '节中文歌曲课。为了让学生们更好地理解歌', '词大意，老师尤斯拉·穆罕默德萨尔·侯赛因逐', '字翻译和解释歌词。随着伴奏声响起，学生们', '边唱边随着节拍摇动身体，现场气氛热烈。'], 'rec_scores': array([0.99972075, ..., 0.96241361]), 'rec_polys': array([[[133,  35],
+        ...,
+        [133, 131]],
+
+       ...,
+
+       [[ 13, 754],
+        ...,
+        [ 13, 777]]], dtype=int16), 'rec_boxes': array([[133, ..., 131],
+       ...,
+       [ 13, ..., 777]], dtype=int16)}}}
+</code></pre></details>
+
+For explanation of the result parameters, refer to [2.2 Python Script Integration](#22-python-script-integration).
+
+<b>Note:</b> Due to the large size of the default model in the pipeline, the inference speed may be slow. You can refer to the model list in Section 1 to replace it with a faster model.
+
+### 2.2 Python Script Integration
+
+The command line method is for quick testing and visualization. In actual projects, you usually need to integrate the model via code. You can perform pipeline inference with just a few lines of code as shown below:
+
+```python
+from paddleocr import PPStructureV3
+
+pipeline = PPStructureV3()
+# pipeline = PPStructureV3(lang="en") # Set the lang parameter to use the English text recognition model. For other supported languages, see Section 5: Appendix. By default, both Chinese and English text recognition models are enabled.
+# pipeline = PPStructureV3(use_doc_orientation_classify=True) # Use use_doc_orientation_classify to enable/disable document orientation classification model
+# pipeline = PPStructureV3(use_doc_unwarping=True) # Use use_doc_unwarping to enable/disable document unwarping module
+# pipeline = PPStructureV3(use_textline_orientation=True) # Use use_textline_orientation to enable/disable textline orientation classification model
+# pipeline = PPStructureV3(device="gpu") # Use device to specify GPU for model inference
+output = pipeline.predict("./pp_structure_v3_demo.png")
+for res in output:
+    res.print() ## Print the structured prediction output
+    res.save_to_json(save_path="output") ## Save the current image's structured result in JSON format
+    res.save_to_markdown(save_path="output") ## Save the current image's result in Markdown format
+    res.save_to_word(save_path="output") ## Save the current image's result in Word format
+```
+
+The example above uses local Paddle inference engines by default. By default, each module selects the appropriate local Paddle inference engine according to the default model name: models that support only dynamic graph use `paddle_dynamic`, while models that support both static and dynamic graph prefer `paddle_static`. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following code:
+
+```python
+from paddleocr import PPStructureV3
+
+pipeline = PPStructureV3(
+    engine="transformers",
+)
+# pipeline = PPStructureV3(lang="en") # Set the lang parameter to use the English text recognition model. For other supported languages, see Section 5: Appendix. By default, both Chinese and English text recognition models are enabled.
+# pipeline = PPStructureV3(use_doc_orientation_classify=True) # Use use_doc_orientation_classify to enable/disable document orientation classification model
+# pipeline = PPStructureV3(use_doc_unwarping=True) # Use use_doc_unwarping to enable/disable document unwarping module
+# pipeline = PPStructureV3(use_textline_orientation=True) # Use use_textline_orientation to enable/disable textline orientation classification model
+# pipeline = PPStructureV3(device="gpu") # Use device to specify GPU for model inference
+output = pipeline.predict("./pp_structure_v3_demo.png")
+for res in output:
+    res.print() ## Print the structured prediction output
+    res.save_to_json(save_path="output") ## Save the current image's structured result in JSON format
+    res.save_to_markdown(save_path="output") ## Save the current image's result in Markdown format
+```
+
+If you choose `onnxruntime` as the inference engine, make sure the ONNX Runtime environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following code:
+
+```python
+from paddleocr import PPStructureV3
+
+# Some models are still being supported. For inference, please disable formula recognition using the following code:
+pipeline = PPStructureV3(
+    engine="onnxruntime",
+    use_formula_recognition=False,
+)
+# pipeline = PPStructureV3(lang="en") # Set the lang parameter to use the English text recognition model. For other supported languages, see Section 5: Appendix. By default, both Chinese and English text recognition models are enabled.
+# pipeline = PPStructureV3(use_doc_orientation_classify=True) # Use use_doc_orientation_classify to enable/disable document orientation classification model
+# pipeline = PPStructureV3(use_doc_unwarping=True) # Use use_doc_unwarping to enable/disable document unwarping module
+# pipeline = PPStructureV3(use_textline_orientation=True) # Use use_textline_orientation to enable/disable textline orientation classification model
+# pipeline = PPStructureV3(device="gpu") # Use device to specify GPU for model inference
+output = pipeline.predict("./pp_structure_v3_demo.png")
+for res in output:
+    res.print() ## Print the structured prediction output
+    res.save_to_json(save_path="output") ## Save the current image's structured result in JSON format
+    res.save_to_markdown(save_path="output") ## Save the current image's result in Markdown format
+```
+
+For PDF files, each page will be processed individually and generate a separate Markdown file. If you want to convert the entire PDF to a single Markdown file, use the following method:
+
+```python
+from pathlib import Path
+from paddleocr import PPStructureV3
+
+input_file = "./your_pdf_file.pdf"
+output_path = Path("./output")
+
+pipeline = PPStructureV3()
+output = pipeline.predict(input=input_file)
+
+markdown_list = []
+markdown_images = []
+
+for res in output:
+    md_info = res.markdown
+    markdown_list.append(md_info)
+    markdown_images.append(md_info.get("markdown_images", {}))
+
+markdown_texts = pipeline.concatenate_markdown_pages(markdown_list)
+
+mkd_file_path = output_path / f"{Path(input_file).stem}.md"
+mkd_file_path.parent.mkdir(parents=True, exist_ok=True)
+
+with open(mkd_file_path, "w", encoding="utf-8") as f:
+    f.write(markdown_texts)
+
+for item in markdown_images:
+    if item:
+        for path, image in item.items():
+            file_path = output_path / path
+            file_path.parent.mkdir(parents=True, exist_ok=True)
+            image.save(file_path)
+
+```
+
+**Note:**
+
+- The default text recognition model used by PP-StructureV3 is a **Chinese-English recognition model**, which has limited accuracy for purely English texts. For English-only scenarios, you can set the `text_recognition_model_name` parameter to an English model such as `en_PP-OCRv4_mobile_rec` to achieve better recognition performance. For other languages, refer to the model list above and select the appropriate language recognition model for replacement.
+
+- In the example code, the parameters `use_doc_orientation_classify`, `use_doc_unwarping`, and `use_textline_orientation` are all set to `False` by default. These indicate that document orientation classification, document image unwarping, and textline orientation classification are disabled. You can manually set them to `True` if needed.
+
+The above Python script performs the following steps:
+
+<details><summary>(1) Instantiate <code>PPStructureV3</code> to create the pipeline object. The parameter descriptions are as follows:</summary>
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the layout detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the layout detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>Score threshold for the layout model. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float between <code>0-1</code>;</li>
+<li><b>dict</b>: <code>{0:0.1}</code> where the key is the class ID and the value is the threshold for that class;</li>
+<li><b>None</b>: If set to <code>None</code>, uses the pipeline default of <code>0.5</code>.</li>
+</ul>
+</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Whether to use Non-Maximum Suppression (NMS) as post-processing for layout detection. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the parameter will default to the value initialized in the pipeline, which is set to <code>True</code> by default.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion ratio for the bounding boxes from the layout detection model. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>Tuple[float,float]</b>: Expansion ratios in horizontal and vertical directions;</li>
+<li><b>dict</b>: A dictionary with <b>int</b> keys representing <code>cls_id</code>, and <b>tuple</b> values, e.g., <code>{0: (1.1, 2.0)}</code> means width is expanded 1.1× and height 2.0× for class 0 boxes;</li>
+<li><b>None</b>: If set to <code>None</code>, uses the pipeline default of <code>1.0</code>.</li>
+</ul>
+</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>Filtering method for overlapping boxes in layout detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>str</b>: Options include <code>large</code>, <code>small</code>, and <code>union</code> to retain the larger box, smaller box, or both;</li>
+<li><b>dict</b>: A dictionary with <b>int</b> keys representing <code>cls_id</code>, and <b>str</b> values, e.g., <code>{0: "large", 2: "small"}</code> means using different modes for different classes;</li>
+<li><b>None</b>: If set to <code>None</code>, uses the pipeline default value <code>large</code>.</li>
+</ul>
+</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>chart_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the chart parsing model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>chart_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the chart parsing model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>chart_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the chart parsing model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default is <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>region_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the region detection model for sub-modules in document layout. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>region_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the region detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>Name of the document orientation classification model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the document orientation classification model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>Meaning:</b>Name of the document unwarping model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the document unwarping model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the text detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limitation for text detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>int</b>: Any integer greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, uses the pipeline default of <code>960</code>.</li>
+</ul>
+</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td>
+<b>Meaning:</b>Limit type for text detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>str</b>: Supports <code>min</code> and <code>max</code>. <code>min</code> ensures the shortest side is no less than <code>det_limit_side_len</code>, while <code>max</code> ensures the longest side is no greater than <code>limit_side_len</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, uses the pipeline default of <code>max</code>.</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>Meaning:</b>Pixel threshold for detection. Pixels in the output probability map with scores above this value are considered as text pixels. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, uses the pipeline default value of <code>0.3</code>.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>Meaning:</b>Bounding box threshold. If the average score of all pixels inside the box exceeds this threshold, it is considered a text region. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, uses the pipeline default value of <code>0.6</code>.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion ratio for text detection. The larger the value, the more the text region is expanded. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, uses the pipeline default value of <code>2.0</code>.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_name</code></td>
+<td><b>Meaning:</b>Name of the textline orientation model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the textline orientation model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the textline orientation model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default batch size is <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the text recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the text recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default batch size is <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Score threshold for text recognition. Only results with scores above this threshold will be retained. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, uses the pipeline default of <code>0.0</code> (no threshold).</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_classification_model_name</code></td>
+<td><b>Meaning:</b>Name of the table classification model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_classification_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the table classification model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the wired table structure recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the wired table structure recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the wireless table structure recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the wireless table structure recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the wired table cell detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the wired table cell detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the wireless table cell detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the wireless table cell detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>Name of the wireless table orientation classification model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>Directory of the table orientation classification model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the seal text detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the seal text detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limit for seal text detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>int</b>: Any integer greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the default value is <code>736</code>.</li>
+</ul>
+</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>Meaning:</b>Limit type for seal text detection image side length. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>str</b>: Supports <code>min</code> and <code>max</code>. <code>min</code> ensures the shortest side is no less than <code>det_limit_side_len</code>, while <code>max</code> ensures the longest side is no greater than <code>limit_side_len</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the default value is <code>min</code>.</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>Meaning:</b>Pixel threshold for detection. Pixels with scores greater than this value in the probability map are considered text pixels. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the default value is <code>0.2</code>.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>Meaning:</b>Bounding box threshold. If the average score of all pixels inside a detection box exceeds this threshold, it is considered a text region. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the default value is <code>0.6</code>.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion ratio for seal text detection. The larger the value, the larger the expanded area. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the default value is <code>0.5</code>.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the seal text recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the seal text recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the seal text recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default value is <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Score threshold for seal text recognition. Text results with scores above this threshold will be retained. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the default value is <code>0.0</code> (no threshold).</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the formula recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the formula recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the formula recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default value is <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to enable the document orientation classification module. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default value is <code>False</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to enable the document image unwarping module. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default value is <code>False</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>Meaning:</b>Whether to use the text line orientation classification. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default value is <code>False</code>.</td>
+<td><code>bool|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>Meaning:</b>Whether to enable seal text recognition subpipeline. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default value is <code>False</code>.</td>
+<td><code>bool|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>Meaning:</b>Whether to enable table recognition subpipeline. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default value is <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>use_formula_recognition</code></td>
+<td><b>Meaning:</b>Whether to enable formula recognition subpipeline. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default value is <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>Meaning:</b>Whether to load and use the chart parsing module. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default value is <code>False</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_region_detection</code></td>
+<td><b>Meaning:</b>Whether to load and use the document region detection module. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default value is <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>format_block_content</code></td>
+<td><b>Meaning:</b>Whether to format the content in <code>block_content</code> as Markdown.<br/>
+<b>Description:</b> If set to <code>None</code>, the default value is <code>False</code>.</td>
+<td><code>bool|None</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>markdown_ignore_labels</code></td>
+<td><b>Meaning:</b>Layout tags that need to be ignored in Markdown.<br/>
+<b>Description:</b> If set to <code>None</code>, the default value is <code>['number','footnote','header','header_image','footer','footer_image','aside_text']</code>.</td>
+<td><code>list|None</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>Device used for inference. <br/>
+<b>Description:</b> 
+Supports specifying device ID:
+<ul>
+<li><b>CPU</b>: e.g., <code>cpu</code> means using CPU for inference;</li>
+<li><b>GPU</b>: e.g., <code>gpu:0</code> means using GPU 0;</li>
+<li><b>NPU</b>: e.g., <code>npu:0</code> means using NPU 0;</li>
+<li><b>XPU</b>: e.g., <code>xpu:0</code> means using XPU 0;</li>
+<li><b>MLU</b>: e.g., <code>mlu:0</code> means using MLU 0;</li>
+<li><b>DCU</b>: e.g., <code>dcu:0</code> means using DCU 0;</li>
+<li><b>MetaX GPU</b>: e.g., <code>metax_gpu:0</code> means using MetaX GPU 0;</li>
+<li><b>Iluvatar GPU</b>: e.g., <code>iluvatar_gpu:0</code> means using Iluvatar GPU 0;</li>
+<li><b>None</b>: If set to <code>None</code>, the pipeline initialized value for this parameter will be used. During initialization, the local GPU device 0 will be preferred; if unavailable, the CPU device will be used.</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to enable the TensorRT subgraph engine of Paddle Inference.<br/>
+<b>Description:</b>
+If the model does not support TensorRT acceleration, acceleration will not be used even if this flag is set.<br/>
+For CUDA 11.8 versions of PaddlePaddle, the compatible TensorRT version is 8.x (x>=6). TensorRT 8.6.1.6 is recommended.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b> Computation precision, such as <code>"fp32"</code> or <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b> Whether to enable MKL-DNN accelerated inference.<br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support MKL-DNN acceleration, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b> MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads used for inference on CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b> Path to the PaddleX pipeline configuration file.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+</tbody>
+</table>
+
+</details>
+
+<details><summary>(2) Call the <code>predict()</code> method of the PP-StructureV3 pipeline object for inference. This method returns a result list. The pipeline also provides a <code>predict_iter()</code> method. Both methods accept the same parameters and return the same type of results. The only difference is that <code>predict_iter()</code> returns a <code>generator</code> that allows incremental processing and retrieval of prediction results, which is useful for handling large datasets or saving memory. Choose the method that fits your needs. Below are the parameters of the <code>predict()</code> method:</summary>
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Input data to be predicted. Required. <br/>
+<b>Description:</b> 
+Supports multiple types:
+<ul>
+<li><b>Python Var</b>: Image data represented as <code>numpy.ndarray</code>;</li>
+<li><b>str</b>: Local path to image or PDF file, e.g., <code>/root/data/img.jpg</code>; <b>URL</b> to image or PDF, e.g., <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/demo_paper.png">example</a>; <b>directory</b> containing image files, e.g., <code>/root/data/</code> (directories with PDFs are not supported, use full file path for PDFs);</li>
+<li><b>list</b>: Elements can be any of the above types, e.g., <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"].</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to use document orientation classification during inference. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to use document image unwarping during inference. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>Meaning:</b>Whether to use textline orientation classification during inference. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>Meaning:</b>Whether to use the seal text recognition sub-pipeline during inference. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>Meaning:</b>Whether to use the table recognition sub-pipeline during inference. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_formula_recognition</code></td>
+<td><b>Meaning:</b>Whether to use the formula recognition sub-pipeline during inference. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>Meaning:</b>Whether to use the chart parsing module during inference. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_region_detection</code></td>
+<td><b>Meaning:</b>Whether to use the document region detection module during inference. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>format_block_content</code></td>
+<td>Whether to format the content in <code>block_content</code> as Markdown. If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_wired_table_cells_trans_to_html</code></td>
+<td><b>Meaning:</b>Whether to enable direct conversion of wired table cell detection results to HTML. <br/>
+<b>Description:</b> 
+If enabled, HTML will be constructed directly based on the geometric relationship of wired table cell detection results.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_wireless_table_cells_trans_to_html</code></td>
+<td><b>Meaning:</b>Whether to enable direct conversion of wireless table cell detection results to HTML. <br/>
+<b>Description:</b> 
+If enabled, HTML will be constructed directly based on the geometric relationship of wireless table cell detection results.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_table_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to enable table orientation classification. <br/>
+<b>Description:</b> 
+When enabled, it can correct the orientation and correctly complete table recognition if the table in the image is rotated by 90/180/270 degrees.</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_ocr_results_with_table_cells</code></td>
+<td><b>Meaning:</b>Whether to enable OCR within cell segmentation. <br/>
+<b>Description:</b> 
+When enabled, OCR detection results will be segmented and re-recognized based on cell prediction results to avoid text loss.</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_e2e_wired_table_rec_model</code></td>
+<td><b>Meaning:</b>Whether to enable end-to-end wired table recognition mode. <br/>
+<b>Description:</b> 
+If enabled, the cell detection model will not be used, and only the table structure recognition model will be used.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_e2e_wireless_table_rec_model</code></td>
+<td><b>Meaning:</b>Whether to enable end-to-end wireless table recognition mode. <br/>
+<b>Description:</b> 
+If enabled, the cell detection model will not be used, and only the table structure recognition model will be used.</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>markdown_ignore_labels</code></td>
+<td>Layout tags that need to be ignored in Markdown. If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>list|None</code></td>
+<td></td>
+</tr>
+</table>
+</details>
+
+<details><summary>(3) Process the prediction results: each prediction result corresponds to a Result object, which supports printing, saving as image, or saving as a <code>json</code> file:</summary>
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Description</th>
+<th>Parameter</th>
+<th>Type</th>
+<th>Parameter Description</th>
+<th>Default</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">Print result to terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format output as indented <code>JSON</code>.</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Indentation level to beautify the <code>JSON</code> output. Only effective when <code>format_json=True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Whether to escape non-<code>ASCII</code> characters to <code>Unicode</code>. When <code>True</code>, all non-ASCII characters are escaped. When <code>False</code>, original characters are retained. Only effective when <code>format_json=True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">Save result as a JSON file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Path to save the file. If a directory, the filename will be based on the input type.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Indentation level for beautified <code>JSON</code> output. Only effective when <code>format_json=True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Whether to escape non-<code>ASCII</code> characters to <code>Unicode</code>. Only effective when <code>format_json=True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save intermediate visualization results as PNG image files</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Path to save the file, supports directory or file path.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>save_to_word()</code></td>
+<td>Save the layout parsing results as a Word (.docx) format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Path to save the file, supports directory or file path.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>save_to_markdown()</code></td>
+<td>Save each page of an image or PDF file as a markdown file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Path to save the file, supports directory or file path.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>save_to_html()</code></td>
+<td>Save tables in the file as HTML format</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Path to save the file, supports directory or file path.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>save_to_xlsx()</code></td>
+<td>Save tables in the file as XLSX format</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Path to save the file, supports directory or file path.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>concatenate_markdown_pages()</code></td>
+<td>Concatenate multiple markdown pages into a single document</td>
+<td><code>markdown_list</code></td>
+<td><code>list</code></td>
+<td>List of markdown data for each page.</td>
+<td>Returns the merged markdown text and image list.</td>
+</tr>
+</table>
+
+<ul>
+ <li> Calling <code>print()</code> will print the result to the terminal. Explanation of the printed content:</li>
+     <ul>
+    <li><code>input_path</code>: <code>(str)</code> Input path of the image or PDF to be predicted</li>
+    <li><code>page_index</code>: <code>(Union[int, None])</code> If input is a PDF, indicates the page number; otherwise <code>None</code></li>
+    <li><code>page_count</code>: <code>(Union[int, None])</code> If the input is a PDF file, it indicates the total number of pages in the PDF; otherwise, it is <code>None</code>.</li>
+    <li><code>width</code>: <code>(int)</code> The width of the original input image.</li>
+    <li><code>height</code>: <code>(int)</code> The height of the original input image.</li>
+    <li><code>model_settings</code>: <code>(Dict[str, bool])</code> Model parameters configured for the pipeline</li>
+        <ul>
+            <li><code>use_doc_preprocessor</code>: <code>(bool)</code> Whether to enable document preprocessor sub-pipeline</li>
+            <li><code>use_seal_recognition</code>: <code>(bool)</code> Whether to enable seal text recognition sub-pipeline</li>
+            <li><code>use_table_recognition</code>: <code>(bool)</code> Whether to enable table recognition sub-pipeline</li>
+            <li><code>use_formula_recognition</code>: <code>(bool)</code> Whether to enable formula recognition sub-pipeline</li>
+            <li><code>format_block_content</code>: <code>(bool)</code> Controls whether to format the <code>block_content</code> into Markdown format</li>
+            <li><code>markdown_ignore_labels</code>: <code>(List[str])</code> Labels of layout regions that need to be ignored in Markdown</li>
+        </ul>
+    </li>
+    <li><code>doc_preprocessor_res</code>: <code>(Dict[str, Union[List[float], str]])</code> Document preprocessing result dictionary, only exists if <code>use_doc_preprocessor=True</code></li>
+        <ul>
+        <li><code>input_path</code>: <code>(str)</code> Image path accepted by document preprocessor, <code>None</code> if input is <code>numpy.ndarray</code></li>
+        <li><code>page_index</code>: <code>None</code> since input is <code>numpy.ndarray</code></li>
+        <li><code>model_settings</code>: <code>(Dict[str, bool])</code> Model configuration for the document preprocessor</li>
+            <ul>
+                <li><code>use_doc_orientation_classify</code>: <code>(bool)</code> Whether to enable document orientation classification</li>
+                <li><code>use_doc_unwarping</code>: <code>(bool)</code> Whether to enable image unwarping</li>
+            </ul>
+        <li><code>angle</code>: <code>(int)</code> Predicted angle result if orientation classification is enabled</li>
+        </ul>
+    <li><code>parsing_res_list</code>: <code>(List[Dict])</code> A list of parsing results, where each element is a dictionary. The order of the list is the reading order after parsing.</li>
+        <ul>
+            <li><code>block_bbox</code>: <code>(np.ndarray)</code> The bounding box of the layout area.</li>
+            <li><code>block_label</code>: <code>(str)</code> The label of the layout area, such as <code>text</code>, <code>table</code>, etc.</li>
+            <li><code>block_content</code>: <code>(str)</code> The content within the layout area.</li>
+            <li><code>block_id</code>: <code>(int)</code> The index of the layout area, used to display the layout sorting result.</li>
+            <li><code>block_order</code>: <code>(int)</code> The order of the layout area, used to display the reading order of the layout. For non-ordered parts, the default value is <code>None</code>.</li>
+        </ul>
+    <li><code>overall_ocr_res</code>: <code>(Dict[str, Union[List[str], List[float], numpy.ndarray]])</code>  Dictionary of global OCR results</li>
+        <ul>
+            <li><code>input_path</code>: <code>(Union[str, None])</code> OCR sub-pipeline input path; <code>None</code> if input is <code>numpy.ndarray</code></li>
+            <li><code>page_index</code>: <code>None</code> since input is <code>numpy.ndarray</code></li>
+            <li><code>model_settings</code>: <code>(Dict)</code> OCR model configuration</li>
+            <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> List of polygons for text detection. Each box is a numpy array with shape (4, 2), dtype int16</li>
+            <li><code>dt_scores</code>: <code>(List[float])</code> Confidence scores for detection boxes</li>
+            <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> Text detection module parameters</li>
+                <ul>
+                    <li><code>limit_side_len</code>: <code>(int)</code> Side length limit for image preprocessing</li>
+                    <li><code>limit_type</code>: <code>(str)</code> Limit processing method</li>
+                    <li><code>thresh</code>: <code>(float)</code> Threshold for text pixel classification</li>
+                    <li><code>box_thresh</code>: <code>(float)</code> Threshold for text detection boxes</li>
+                    <li><code>unclip_ratio</code>: <code>(float)</code> Unclip ratio for expanding boxes</li>
+                    <li><code>text_type</code>: <code>(str)</code> Text detection type, currently fixed as "general"</li>
+                </ul>
+            <li><code>text_type</code>: <code>(str)</code> Text detection type, currently fixed as "general"</li>
+            <li><code>textline_orientation_angles</code>: <code>(List[int])</code> Orientation classification results for text lines</li>
+            <li><code>text_rec_score_thresh</code>: <code>(float)</code> Threshold for text recognition filtering</li>
+            <li><code>rec_texts</code>: <code>(List[str])</code> Recognized texts filtered by score threshold</li>
+            <li><code>rec_scores</code>: <code>(List[float])</code> Recognition scores filtered by threshold</li>
+            <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> Filtered detection boxes, same format as <code>dt_polys</code></li>
+         </ul>
+    <li><code>formula_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> List of formula recognition results</li>
+        <ul>
+            <li><code>rec_formula</code>: <code>(str)</code> Recognized formula string</li>
+            <li><code>rec_polys</code>: <code>(numpy.ndarray)</code> Bounding box for the formula, shape (4, 2), dtype int16</li>
+            <li><code>formula_region_id</code>: <code>(int)</code> Region ID of the formula</li>
+        </ul>
+    <li><code>seal_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> List of seal text recognition results</li>
+        <ul>
+            <li><code>input_path</code>: <code>(str)</code> Input path for the seal image</li>
+            <li><code>page_index</code>: <code>None</code> since input is <code>numpy.ndarray</code></li>
+            <li><code>model_settings</code>: <code>(Dict)</code> Model configuration for seal text recognition</li>
+            <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> Seal detection boxes, same format as <code>dt_polys</code></li>
+            <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> Detection parameters, same as above</li>
+            <li><code>text_type</code>: <code>(str)</code> Detection type, currently fixed as "seal"</li>
+            <li><code>text_rec_score_thresh</code>: <code>(float)</code> Score threshold for recognition</li>
+            <li><code>rec_texts</code>: <code>(List[str])</code> Recognized texts filtered by score</li>
+            <li><code>rec_scores</code>: <code>(List[float])</code> Recognition scores filtered by threshold</li>
+            <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> Filtered seal boxes, same format as <code>dt_polys</code></li>
+            <li><code>rec_boxes</code>: <code>(numpy.ndarray)</code> Rectangle boxes, shape (n, 4), dtype int16</li>
+        </ul>
+    <li><code>table_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> List of table recognition results</li>
+        <ul>
+            <li><code>cell_box_list</code>: <code>(List[numpy.ndarray])</code> Bounding boxes of table cells</li>
+            <li><code>pred_html</code>: <code>(str)</code> Table as an HTML string</li>
+            <li><code>table_ocr_pred</code>: <code>(Dict)</code> OCR results for the table</li>
+            <ul>
+                <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> Detected cell boxes</li>
+                <li><code>rec_texts</code>: <code>(List[str])</code> Recognized texts for cells</li>
+                <li><code>rec_scores</code>: <code>(List[float])</code> Confidence scores for cell recognition</li>
+                <li><code>rec_boxes</code>: <code>(numpy.ndarray)</code> Rectangle boxes for detection, shape (n, 4), dtype int16</li>
+            </ul>
+        </ul>
+    </ul>
+</li>
+<li>Calling <code>save_to_json()</code> saves the above content to the specified <code>save_path</code>. If it’s a directory, the saved path will be <code>save_path/{your_img_basename}_res.json</code>. If it’s a file, it saves directly. Numpy arrays are converted to lists since JSON doesn't support them.</li>
+<li>Calling <code>save_to_img()</code> saves visual results to the specified <code>save_path</code>. If a directory, various visualizations such as layout detection, OCR, and reading order are saved. If a file, only the last image is saved and others are overwritten.</li>
+<li>Calling <code>save_to_markdown()</code> saves converted markdown files to <code>save_path/{your_img_basename}.md</code>. For PDF input, it's recommended to specify a directory to avoid file overwriting.</li>
+<li>Calling <code>concatenate_markdown_pages()</code> merges multi-page markdown results from the <code>PP-StructureV3 pipeline</code>  into a single document and returns the merged content.</li>
+
+Additionally, you can access the prediction results and visual images through the following attributes:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>json</code></td>
+<td>Get the prediction result in <code>json</code> format</td>
+</tr>
+<tr>
+<td rowspan="2"><code>img</code></td>
+<td rowspan="2">Get visualized image results as a <code>dict</code></td>
+</tr>
+<tr>
+</tr>
+<tr>
+<td rowspan="3"><code>markdown</code></td>
+<td rowspan="3">Get markdown results as a <code>dict</code></td>
+</tr>
+<tr>
+</tr>
+<tr>
+</tr>
+</tbody>
+</table>
+<ul>
+    <li>The <code>json</code> attribute returns the prediction result as a dictionary, which is consistent with the content saved using the <code>save_to_json()</code> method.</li>
+    <li>The <code>img</code> attribute returns the prediction result as a dictionary. The keys include <code>layout_det_res</code>, <code>overall_ocr_res</code>, <code>text_paragraphs_ocr_res</code>, <code>formula_res_region1</code>, <code>table_cell_img</code>, and <code>seal_res_region1</code>, each corresponding to a visualized <code>Image.Image</code>, object for layout detection, OCR, text paragraph, formula, table, and seal results. If optional modules are not used, the dictionary only contains <code>layout_det_res</code>.</li>
+    <li>The <code>markdown</code> attribute returns the prediction result as a dictionary. The keys include <code>markdown_texts</code>, <code>markdown_images</code>, and <code>page_continuation_flags</code>, where the values represent the markdown text, displayed images (<code>Image.Image</code> objects), and a boolean tuple indicating whether the first and last elements of the current page are paragraph boundaries.</li>
+</ul>
+</details>
+
+## 3. Development Integration / Deployment
+
+If the pipeline meets your requirements for inference speed and accuracy, you can proceed with development integration or deployment.
+
+If you want to directly use the pipeline in your Python project, refer to the example code in [2.2 Python script mode](#22-python-script-integration).
+
+In addition, PaddleOCR provides two other deployment options described in detail below:
+
+🚀 High-Performance Inference: In production environments, many applications have strict performance requirements (especially response speed) to ensure system efficiency and smooth user experience. PaddleOCR offers a high-performance inference option that deeply optimizes model inference and pre/post-processing for significant end-to-end acceleration. For detailed high-performance inference workflow, refer to [High Performance Inference](../inference_deployment/local_inference/high_performance_inference.en.md).
+
+☁️ Service Deployment: Service-based deployment is common in production. It encapsulates the inference logic as a service, allowing clients to access it via network requests to obtain results. For detailed instructions on service deployment, refer to [Service Deployment](../inference_deployment/serving/serving.en.md).
+
+Below is the API reference and multi-language service invocation examples for basic service deployment:
+
+<details><summary>API Reference</summary>
+<p>For the main operations provided by the service:</p>
+<ul>
+<li>The HTTP request method is POST.</li>
+<li>Both the request body and response body are JSON data (JSON objects).</li>
+<li>When the request is processed successfully, the response status code is <code>200</code>, and the attributes of the response body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>The UUID of the request.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Fixed as <code>0</code>.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error message. Fixed as <code>"Success"</code>.</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>The result of the operation.</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is not processed successfully, the attributes of the response body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>The UUID of the request.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Same as the response status code.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error message.</td>
+</tr>
+</tbody>
+</table>
+<p>The main operations provided by the service are as follows:</p>
+<ul>
+<li><b><code>infer</code></b></li>
+</ul>
+<p>Perform layout parsing.</p>
+<p><code>POST /layout-parsing</code></p>
+<ul>
+<li>The attributes of the request body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+<th>Required</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>The URL of image files (including TIFF; multi-page TIFF is processed page by page) or PDF file accessible by the server, or the Base64-encoded content of the above file types. By default, for PDF or multi-page TIFF files exceeding 10 pages, only the first 10 pages will be processed.<br />
+To remove the page limit, please add the following configuration to the pipeline configuration file:
+<pre><code>Serving:
+  extra:
+    max_num_input_imgs: null
+</code></pre></td>
+<td>Yes</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code>｜<code>null</code></td>
+<td>File type. <code>0</code> represents a PDF file, and <code>1</code> represents an image file. If this attribute is missing from the request body, the file type will be inferred based on the URL.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_doc_orientation_classify</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_doc_unwarping</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useTextlineOrientation</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_textline_orientation</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useSealRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_seal_recognition</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useTableRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_table_recognition</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useFormulaRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_formula_recognition</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useChartRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_chart_recognition</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useRegionDetection</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_region_detection</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>formatBlockContent</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>format_block_content</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutThreshold</code></td>
+<td><code>number</code> | <code>object</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>layout_threshold</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutNms</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>layout_nms</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutUnclipRatio</code></td>
+<td><code>number</code> | <code>array</code> | <code>object</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>layout_unclip_ratio</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutMergeBboxesMode</code></td>
+<td><code>string</code> | <code>object</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>layout_merge_bboxes_mode</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>text_det_limit_side_len</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>text_det_limit_type</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>text_det_thresh</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>text_det_box_thresh</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>text_det_unclip_ratio</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>text_rec_score_thresh</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_det_limit_side_len</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_det_limit_type</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_det_thresh</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_det_box_thresh</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_det_unclip_ratio</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_rec_score_thresh</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useWiredTableCellsTransToHtml</code></td>
+<td><code>boolean</code></td>
+<td>Please refer to the description of the <code>use_wired_table_cells_trans_to_html</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useWirelessTableCellsTransToHtml</code></td>
+<td><code>boolean</code></td>
+<td>Please refer to the description of the <code>use_wireless_table_cells_trans_to_html</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useTableOrientationClassify</code></td>
+<td><code>boolean</code></td>
+<td>Please refer to the description of the <code>use_table_orientation_classify</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useOcrResultsWithTableCells</code></td>
+<td><code>boolean</code></td>
+<td>Please refer to the description of the <code>use_ocr_results_with_table_cells</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useE2eWiredTableRecModel</code></td>
+<td><code>boolean</code></td>
+<td>Please refer to the description of the <code>use_e2e_wired_table_rec_model</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useE2eWirelessTableRecModel</code></td>
+<td><code>boolean</code></td>
+<td>Please refer to the description of the <code>use_e2e_wireless_table_rec_model</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>markdownIgnoreLabels</code></td>
+<td><code>array</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>markdown_ignore_labels</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>prettifyMarkdown</code></td>
+<td><code>boolean</code></td>
+<td>Whether to output beautified Markdown text. The default is <code>true</code>.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>showFormulaNumber</code></td>
+<td><code>boolean</code></td>
+<td>Whether to include formula numbers in the output Markdown text. The default is <code>false</code>.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>returnMarkdownImages</code></td>
+<td><code>boolean</code></td>
+<td>Whether to return the images referenced in the Markdown. Default <code>true</code>; when set to <code>false</code>, <code>markdown.images</code> is <code>null</code> or omitted and the server skips image encoding / URL upload.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>outputFormats</code></td>
+<td><code>array</code> | <code>null</code></td>
+<td>Optional list of extra formats to return. Currently only <code>"docx"</code> is supported.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>
+Whether to return the final visualization image and intermediate images during the processing.<br/>
+<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>If <code>true</code> is provided: return images.</li>
+<li>If <code>false</code> is provided: do not return any images.</li>
+<li>If this parameter is omitted from the request body, or if <code>null</code> is explicitly passed, the behavior will follow the value of <code>Serving.visualize</code> in the pipeline configuration.</li>
+</ul>
+<br/>
+For example, adding the following setting to the pipeline config file:<br/>
+<pre><code>Serving:
+  visualize: False
+</code></pre>
+will disable image return by default. This behavior can be overridden by explicitly setting the <code>visualize</code> parameter in the request.<br/>
+If neither the request body nor the configuration file is set (If <code>visualize</code> is set to <code>null</code> in the request and  not defined in the configuration file), the image is returned by default.
+</td>
+<td>No</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is processed successfully, the <code>result</code> in the response body has the following attributes:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>layoutParsingResults</code></td>
+<td><code>array</code></td>
+<td>The layout parsing results. The array length is 1 (for image input) or the actual number of document pages processed (for PDF input). For PDF input, each element in the array represents the result of each page actually processed in the PDF file.</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>Information about the input data.</td>
+</tr>
+</tbody>
+</table>
+<p>Image and other binary file fields in the element schema below (e.g. <code>outputImages</code>, <code>inputImage</code>, <code>markdown.images</code>, <code>exports</code>) are returned inline as Base64 strings by default; when the server is configured to return URLs, those values become pre-signed URLs while the field types remain unchanged. See the "Returning Binary Content as URLs" section of the <a href="../inference_deployment/serving/serving.en.md">Serving Deployment Guide</a> for configuration.</p>
+<p>Each element in <code>layoutParsingResults</code> is an <code>object</code> with the following attributes:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>A simplified version of the <code>res</code> field in the JSON representation of the result generated by the <code>predict</code> method of the pipeline object, with the <code>input_path</code> and the <code>page_index</code> fields removed.</td>
+</tr>
+<tr>
+<td><code>markdown</code></td>
+<td><code>object</code></td>
+<td>The Markdown result.</td>
+</tr>
+<tr>
+<td><code>outputImages</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>See the description of the <code>img</code> attribute of the result of the pipeline prediction. The images are in JPEG format, Base64-encoded by default; returned as pre-signed URLs when URL-return mode is enabled.</td>
+</tr>
+<tr>
+<td><code>inputImage</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>The input image. The image is in JPEG format, Base64-encoded by default; returned as a pre-signed URL when URL-return mode is enabled.</td>
+</tr>
+<tr>
+<td><code>exports</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Optional additional exports when <code>outputFormats</code> is present—for example, <code>{"docx": {"content": "..."}}</code>, where <code>content</code> is the Base64-encoded file content by default, or a pre-signed URL when URL-return mode is enabled.</td>
+</tr>
+</tbody>
+</table>
+<p><code>markdown</code> is an <code>object</code> with the following attributes:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>text</code></td>
+<td><code>string</code></td>
+<td>The Markdown text.</td>
+</tr>
+<tr>
+<td><code>images</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>A key-value pair of relative Markdown image paths and their image data. Values are Base64-encoded by default; returned as pre-signed URLs when URL-return mode is enabled. The field is <code>null</code> or omitted when <code>returnMarkdownImages</code> is <code>false</code> in the request.</td>
+</tr>
+<tr>
+<td><code>isStart</code></td>
+<td><code>boolean</code></td>
+<td>Whether the first element on the current page is the start of a segment.</td>
+</tr>
+<tr>
+<td><code>isEnd</code></td>
+<td><code>boolean</code></td>
+<td>Whether the last element on the current page is the end of a segment.</td>
+</tr>
+</tbody>
+</table>
+</details>
+<details><summary>Multi-language Service Call Examples</summary>
+<details>
+<summary>Python</summary>
+
+<pre><code class="language-python">
+import base64
+import requests
+import pathlib
+
+API_URL = "http://localhost:8080/layout-parsing" # Service URL
+
+image_path = "./demo.jpg"
+
+# Encode the local image with Base64
+with open(image_path, "rb") as file:
+    image_bytes = file.read()
+    image_data = base64.b64encode(image_bytes).decode("ascii")
+
+payload = {
+    "file": image_data, # Base64-encoded file content or file URL
+    "fileType": 1, # file type, 1 represents image file
+}
+
+# Call the API
+response = requests.post(API_URL, json=payload)
+
+# Process the response data
+assert response.status_code == 200
+result = response.json()["result"]
+print("\nDetected layout elements:")
+for i, res in enumerate(result["layoutParsingResults"]):
+    print(res["prunedResult"])
+    md_dir = pathlib.Path(f"markdown_{i}")
+    md_dir.mkdir(exist_ok=True)
+    (md_dir / "doc.md").write_text(res["markdown"]["text"])
+    for img_path, img in res["markdown"]["images"].items():
+        img_path = md_dir / img_path
+        img_path.parent.mkdir(parents=True, exist_ok=True)
+        img_path.write_bytes(base64.b64decode(img))
+    print(f"Markdown document saved at {md_dir / 'doc.md'}")
+    for img_name, img in res["outputImages"].items():
+        img_path = f"{img_name}_{i}.jpg"
+        with open(img_path, "wb") as f:
+            f.write(base64.b64decode(img))
+        print(f"Output image saved at {img_path}")
+</code></pre></details>
+
+<details><summary>C++</summary>
+
+<pre><code class="language-cpp">#include &lt;iostream&gt;
+#include &lt;fstream&gt;
+#include &lt;vector&gt;
+#include &lt;string&gt;
+#include "cpp-httplib/httplib.h" // https://github.com/Huiyicc/cpp-httplib
+#include "nlohmann/json.hpp" // https://github.com/nlohmann/json
+#include "base64.hpp" // https://github.com/tobiaslocker/base64
+
+int main() {
+    httplib::Client client("localhost", 8080);
+
+    const std::string filePath = "./demo.jpg";
+
+    std::ifstream file(filePath, std::ios::binary | std::ios::ate);
+    if (!file) {
+        std::cerr << "Error opening file: " << filePath << std::endl;
+        return 1;
+    }
+
+    std::streamsize size = file.tellg();
+    file.seekg(0, std::ios::beg);
+    std::vector<char> buffer(size);
+    if (!file.read(buffer.data(), size)) {
+        std::cerr << "Error reading file." << std::endl;
+        return 1;
+    }
+
+    std::string bufferStr(buffer.data(), static_cast<size_t>(size));
+    std::string encodedFile = base64::to_base64(bufferStr);
+
+    nlohmann::json jsonObj;
+    jsonObj["file"] = encodedFile;
+    jsonObj["fileType"] = 1;
+
+    auto response = client.Post("/layout-parsing", jsonObj.dump(), "application/json");
+
+    if (response && response->status == 200) {
+        nlohmann::json jsonResponse = nlohmann::json::parse(response->body);
+        auto result = jsonResponse["result"];
+
+        if (!result.is_object() || !result.contains("layoutParsingResults")) {
+            std::cerr << "Unexpected response format." << std::endl;
+            return 1;
+        }
+
+        const auto& results = result["layoutParsingResults"];
+        for (size_t i = 0; i < results.size(); ++i) {
+            const auto& res = results[i];
+
+            if (res.contains("prunedResult")) {
+                std::cout << "Layout result [" << i << "]: " << res["prunedResult"].dump() << std::endl;
+            }
+
+            if (res.contains("outputImages") && res["outputImages"].is_object()) {
+                for (auto& [imgName, imgBase64] : res["outputImages"].items()) {
+                    std::string outputPath = imgName + "_" + std::to_string(i) + ".jpg";
+                    std::string decodedImage = base64::from_base64(imgBase64.get<std::string>());
+
+                    std::ofstream outFile(outputPath, std::ios::binary);
+                    if (outFile.is_open()) {
+                        outFile.write(decodedImage.c_str(), decodedImage.size());
+                        outFile.close();
+                        std::cout << "Saved image: " << outputPath << std::endl;
+                    } else {
+                        std::cerr << "Failed to save image: " << outputPath << std::endl;
+                    }
+                }
+            }
+        }
+    } else {
+        std::cerr << "Request failed." << std::endl;
+        if (response) {
+            std::cerr << "HTTP status: " << response->status << std::endl;
+            std::cerr << "Response body: " << response->body << std::endl;
+        }
+        return 1;
+    }
+
+    return 0;
+}
+</code></pre></details>
+
+<details><summary>Java</summary>
+
+<pre><code class="language-java">import okhttp3.*;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.util.Base64;
+
+public class Main {
+    public static void main(String[] args) throws IOException {
+        String API_URL = "http://localhost:8080/layout-parsing";
+        String imagePath = "./demo.jpg";
+
+        File file = new File(imagePath);
+        byte[] fileContent = java.nio.file.Files.readAllBytes(file.toPath());
+        String base64Image = Base64.getEncoder().encodeToString(fileContent);
+
+        ObjectMapper objectMapper = new ObjectMapper();
+        ObjectNode payload = objectMapper.createObjectNode();
+        payload.put("file", base64Image);
+        payload.put("fileType", 1);
+
+        OkHttpClient client = new OkHttpClient();
+        MediaType JSON = MediaType.get("application/json; charset=utf-8");
+
+        RequestBody body = RequestBody.create(JSON, payload.toString());
+
+        Request request = new Request.Builder()
+                .url(API_URL)
+                .post(body)
+                .build();
+
+        try (Response response = client.newCall(request).execute()) {
+            if (response.isSuccessful()) {
+                String responseBody = response.body().string();
+                JsonNode root = objectMapper.readTree(responseBody);
+                JsonNode result = root.get("result");
+
+                JsonNode layoutParsingResults = result.get("layoutParsingResults");
+                for (int i = 0; i < layoutParsingResults.size(); i++) {
+                    JsonNode item = layoutParsingResults.get(i);
+                    int finalI = i;
+                    JsonNode prunedResult = item.get("prunedResult");
+                    System.out.println("Pruned Result [" + i + "]: " + prunedResult.toString());
+
+                    JsonNode outputImages = item.get("outputImages");
+                    outputImages.fieldNames().forEachRemaining(imgName -> {
+                        try {
+                            String imgBase64 = outputImages.get(imgName).asText();
+                            byte[] imgBytes = Base64.getDecoder().decode(imgBase64);
+                            String imgPath = imgName + "_" + finalI + ".jpg";
+                            try (FileOutputStream fos = new FileOutputStream(imgPath)) {
+                                fos.write(imgBytes);
+                                System.out.println("Saved image: " + imgPath);
+                            }
+                        } catch (IOException e) {
+                            System.err.println("Failed to save image: " + e.getMessage());
+                        }
+                    });
+                }
+            } else {
+                System.err.println("Request failed with HTTP code: " + response.code());
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Go</summary>
+
+<pre><code class="language-go">package main
+
+import (
+    "bytes"
+    "encoding/base64"
+    "encoding/json"
+    "fmt"
+    "io/ioutil"
+    "net/http"
+    "os"
+    "path/filepath"
+)
+
+func main() {
+    API_URL := "http://localhost:8080/layout-parsing"
+    filePath := "./demo.jpg"
+
+    fileBytes, err := ioutil.ReadFile(filePath)
+    if err != nil {
+        fmt.Printf("Error reading file: %v\n", err)
+        return
+    }
+    fileData := base64.StdEncoding.EncodeToString(fileBytes)
+
+    payload := map[string]interface{}{
+        "file":     fileData,
+        "fileType": 1,
+    }
+    payloadBytes, err := json.Marshal(payload)
+    if err != nil {
+        fmt.Printf("Error marshaling payload: %v\n", err)
+        return
+    }
+
+    client := &http.Client{}
+    req, err := http.NewRequest("POST", API_URL, bytes.NewBuffer(payloadBytes))
+    if err != nil {
+        fmt.Printf("Error creating request: %v\n", err)
+        return
+    }
+    req.Header.Set("Content-Type", "application/json")
+
+    res, err := client.Do(req)
+    if err != nil {
+        fmt.Printf("Error sending request: %v\n", err)
+        return
+    }
+    defer res.Body.Close()
+
+    if res.StatusCode != http.StatusOK {
+        fmt.Printf("Unexpected status code: %d\n", res.StatusCode)
+        return
+    }
+
+    body, err := ioutil.ReadAll(res.Body)
+    if err != nil {
+        fmt.Printf("Error reading response: %v\n", err)
+        return
+    }
+
+    type Markdown struct {
+        Text   string            `json:"text"`
+        Images map[string]string `json:"images"`
+    }
+
+    type LayoutResult struct {
+        PrunedResult map[string]interface{} `json:"prunedResult"`
+        Markdown     Markdown               `json:"markdown"`
+        OutputImages map[string]string      `json:"outputImages"`
+        InputImage   *string                `json:"inputImage"`
+    }
+
+    type Response struct {
+        Result struct {
+            LayoutParsingResults []LayoutResult `json:"layoutParsingResults"`
+            DataInfo             interface{}    `json:"dataInfo"`
+        } `json:"result"`
+    }
+
+    var respData Response
+    if err := json.Unmarshal(body, &respData); err != nil {
+        fmt.Printf("Error parsing response: %v\n", err)
+        return
+    }
+
+    for i, res := range respData.Result.LayoutParsingResults {
+        fmt.Printf("Result %d - prunedResult: %+v\n", i, res.PrunedResult)
+
+        mdDir := fmt.Sprintf("markdown_%d", i)
+        os.MkdirAll(mdDir, 0755)
+        mdFile := filepath.Join(mdDir, "doc.md")
+        if err := os.WriteFile(mdFile, []byte(res.Markdown.Text), 0644); err != nil {
+            fmt.Printf("Error writing markdown file: %v\n", err)
+        } else {
+            fmt.Printf("Markdown document saved at %s\n", mdFile)
+        }
+
+        for path, imgBase64 := range res.Markdown.Images {
+            fullPath := filepath.Join(mdDir, path)
+            os.MkdirAll(filepath.Dir(fullPath), 0755)
+            imgBytes, err := base64.StdEncoding.DecodeString(imgBase64)
+            if err != nil {
+                fmt.Printf("Error decoding markdown image: %v\n", err)
+                continue
+            }
+            if err := os.WriteFile(fullPath, imgBytes, 0644); err != nil {
+                fmt.Printf("Error saving markdown image: %v\n", err)
+            }
+        }
+
+        for name, imgBase64 := range res.OutputImages {
+            imgBytes, err := base64.StdEncoding.DecodeString(imgBase64)
+            if err != nil {
+                fmt.Printf("Error decoding output image %s: %v\n", name, err)
+                continue
+            }
+            filename := fmt.Sprintf("%s_%d.jpg", name, i)
+            if err := os.WriteFile(filename, imgBytes, 0644); err != nil {
+                fmt.Printf("Error saving output image %s: %v\n", filename, err)
+            } else {
+                fmt.Printf("Output image saved at %s\n", filename)
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>C#</summary>
+
+<pre><code class="language-csharp">using System;
+using System.IO;
+using System.Net.Http;
+using System.Text;
+using System.Threading.Tasks;
+using Newtonsoft.Json.Linq;
+
+class Program
+{
+    static readonly string API_URL = "http://localhost:8080/layout-parsing";
+    static readonly string inputFilePath = "./demo.jpg";
+
+    static async Task Main(string[] args)
+    {
+        var httpClient = new HttpClient();
+
+        byte[] fileBytes = File.ReadAllBytes(inputFilePath);
+        string fileData = Convert.ToBase64String(fileBytes);
+
+        var payload = new JObject
+        {
+            { "file", fileData },
+            { "fileType", 1 }
+        };
+        var content = new StringContent(payload.ToString(), Encoding.UTF8, "application/json");
+
+        HttpResponseMessage response = await httpClient.PostAsync(API_URL, content);
+        response.EnsureSuccessStatusCode();
+
+        string responseBody = await response.Content.ReadAsStringAsync();
+        JObject jsonResponse = JObject.Parse(responseBody);
+
+        JArray layoutParsingResults = (JArray)jsonResponse["result"]["layoutParsingResults"];
+        for (int i = 0; i < layoutParsingResults.Count; i++)
+        {
+            var res = layoutParsingResults[i];
+            Console.WriteLine($"[{i}] prunedResult:\n{res["prunedResult"]}");
+
+            JObject outputImages = res["outputImages"] as JObject;
+            if (outputImages != null)
+            {
+                foreach (var img in outputImages)
+                {
+                    string imgName = img.Key;
+                    string base64Img = img.Value?.ToString();
+                    if (!string.IsNullOrEmpty(base64Img))
+                    {
+                        string imgPath = $"{imgName}_{i}.jpg";
+                        byte[] imageBytes = Convert.FromBase64String(base64Img);
+                        File.WriteAllBytes(imgPath, imageBytes);
+                        Console.WriteLine($"Output image saved at {imgPath}");
+                    }
+                }
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Node.js</summary>
+
+<pre><code class="language-js">const axios = require('axios');
+const fs = require('fs');
+const path = require('path');
+
+const API_URL = 'http://localhost:8080/layout-parsing';
+const imagePath = './demo.jpg';
+const fileType = 1;
+
+function encodeImageToBase64(filePath) {
+  const bitmap = fs.readFileSync(filePath);
+  return Buffer.from(bitmap).toString('base64');
+}
+
+const payload = {
+  file: encodeImageToBase64(imagePath),
+  fileType: fileType
+};
+
+axios.post(API_URL, payload)
+  .then(response => {
+    const results = response.data.result.layoutParsingResults;
+    results.forEach((res, index) => {
+      console.log(`\n[${index}] prunedResult:`);
+      console.log(res.prunedResult);
+
+      const outputImages = res.outputImages;
+      if (outputImages) {
+        Object.entries(outputImages).forEach(([imgName, base64Img]) => {
+          const imgPath = `${imgName}_${index}.jpg`;
+          fs.writeFileSync(imgPath, Buffer.from(base64Img, 'base64'));
+          console.log(`Output image saved at ${imgPath}`);
+        });
+      } else {
+        console.log(`[${index}] No outputImages.`);
+      }
+    });
+  })
+  .catch(error => {
+    console.error('Error during API request:', error.message || error);
+  });
+</code></pre></details>
+
+<details><summary>PHP</summary>
+
+<pre><code class="language-php">&lt;?php
+
+$API_URL = "http://localhost:8080/layout-parsing";
+$image_path = "./demo.jpg";
+
+$image_data = base64_encode(file_get_contents($image_path));
+$payload = array("file" => $image_data, "fileType" => 1);
+
+$ch = curl_init($API_URL);
+curl_setopt($ch, CURLOPT_POST, true);
+curl_setopt($ch, CURLOPT_POSTFIELDS, json_encode($payload));
+curl_setopt($ch, CURLOPT_HTTPHEADER, array('Content-Type: application/json'));
+curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+$response = curl_exec($ch);
+curl_close($ch);
+
+$result = json_decode($response, true)["result"]["layoutParsingResults"];
+
+foreach ($result as $i => $item) {
+    echo "[$i] prunedResult:\n";
+    print_r($item["prunedResult"]);
+
+    if (!empty($item["outputImages"])) {
+        foreach ($item["outputImages"] as $img_name => $img_base64) {
+            $output_image_path = "{$img_name}_{$i}.jpg";
+            file_put_contents($output_image_path, base64_decode($img_base64));
+            echo "Output image saved at $output_image_path\n";
+        }
+    } else {
+        echo "No outputImages found for item $i\n";
+    }
+}
+?&gt;
+</code></pre></details>
+</details>
+<br/>
+
+## 4. Secondary Development
+
+If the default model weights provided by the PP-StructureV3 pipeline do not meet your accuracy or speed requirements in your scenario, you can try **fine-tuning the existing model** using **your own domain-specific or application-specific data** to improve the performance of the PP-StructureV3 pipeline for your use case.
+
+### 4.1 Model Fine-tuning
+
+Since the PP-StructureV3 pipeline contains multiple modules, unsatisfactory results may originate from any individual module. You can analyze the problematic cases with poor extraction performance, visualize the images, identify the specific module causing the issue, and then refer to the fine-tuning tutorials linked in the table below to perform model fine-tuning.
+
+<table>
+<thead>
+<tr>
+<th>Scenario</th>
+<th>Fine-tuning Module</th>
+<th>Fine-tuning Reference Link</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>Inaccurate layout detection, such as missing seals or tables</td>
+<td>Layout Detection Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/layout_detection.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate table structure recognition</td>
+<td>Table Structure Recognition Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/module_usage/table_structure_recognition.html#4-secondary-development">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate formula recognition</td>
+<td>Formula Recognition Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/module_usage/formula_recognition.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Missing seal text detection</td>
+<td>Seal Text Detection Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/module_usage/seal_text_detection.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Missing text detection</td>
+<td>Text Detection Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/module_usage/text_detection.html#4-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Incorrect text recognition results</td>
+<td>Text Recognition Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/module_usage/text_recognition.html#v-secondary-development">Link</a></td>
+</tr>
+<tr>
+<td>Incorrect correction of vertical or rotated text lines</td>
+<td>Text Line Orientation Classification Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/textline_orientation_classification.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Incorrect correction of full image orientation</td>
+<td>Document Image Orientation Classification Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate image distortion correction</td>
+<td>Text Image Correction Module</td>
+<td>Fine-tuning not supported yet</td>
+</tr>
+</tbody>
+</table>
+
+### 4.2 Model Deployment
+
+Once you have completed fine-tuning with your private dataset, you will obtain the local model weights. You can then use these fine-tuned weights by customizing the pipeline configuration file.
+
+1. Export the pipeline configuration file
+
+You can call the `export_paddlex_config_to_yaml` method of the PPStructureV3 object in PaddleOCR to export the current pipeline configuration as a YAML file:
+
+```Python
+from paddleocr import PPStructureV3
+
+pipeline = PPStructureV3()
+pipeline.export_paddlex_config_to_yaml("PP-StructureV3.yaml")
+```
+
+2. Modify the configuration file
+After obtaining the default pipeline configuration file, replace the corresponding path in the configuration with the local path of your fine-tuned model weights. For example:
+```yaml
+......
+SubModules:
+  LayoutDetection:
+    module_name: layout_detection
+    model_name: PP-DocLayout_plus-L
+    model_dir: null # Replace with the path to the fine-tuned layout detection model weights
+......
+SubPipelines:
+  GeneralOCR:
+    pipeline_name: OCR
+    text_type: general
+    use_doc_preprocessor: False
+    use_textline_orientation: False
+    SubModules:
+      TextDetection:
+        module_name: text_detection
+        model_name: PP-OCRv5_server_det
+        model_dir: null # Replace with the path to the fine-tuned text detection model weights
+        limit_side_len: 960
+        limit_type: max
+        max_side_limit: 4000
+        thresh: 0.3
+        box_thresh: 0.6
+        unclip_ratio: 1.5
+
+      TextRecognition:
+        module_name: text_recognition
+        model_name: PP-OCRv5_server_rec
+        model_dir: null # Replace with the path to the fine-tuned text recognition model weights
+        batch_size: 1
+        score_thresh: 0
+......
+```
+
+The pipeline configuration file not only includes parameters supported by the PaddleOCR CLI and Python API but also allows for more advanced configurations. For more details, refer to the corresponding pipeline usage tutorial in the [PaddleX Pipeline Usage Overview](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/pipeline_develop_guide.html), and adjust the configurations as needed based on your requirements.
+
+3. Load the pipeline configuration file via CLI
+
+After modifying the configuration file, specify the updated pipeline configuration path using the `--paddlex_config` parameter in the command line. PaddleOCR will load its content as the pipeline configuration. Example:
+
+```bash
+paddleocr pp_structurev3 --paddlex_config PP-StructureV3.yaml ...
+```
+
+4. Load the pipeline configuration file via Python API
+When initializing the pipeline object, you can pass the PaddleX pipeline configuration file path or a configuration dictionary using the `paddlex_config` parameter. PaddleOCR will load its content as the pipeline configuration. Example:
+
+```python
+from paddleocr import PPStructureV3
+
+pipeline = PPStructureV3(paddlex_config="PP-StructureV3.yaml")
+```
+
+## 5. Appendix
+
+<details><summary><b>Supported Languages</b></summary>
+
+<table border="1" cellspacing="0" cellpadding="4">
+  <thead>
+    <tr>
+      <th><code>lang</code></th>
+      <th>Language Name</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr><td><code>abq</code></td><td>Abaza</td></tr>
+    <tr><td><code>af</code></td><td>Afrikaans</td></tr>
+    <tr><td><code>ang</code></td><td>Old English</td></tr>
+    <tr><td><code>ar</code></td><td>Arabic</td></tr>
+    <tr><td><code>ava</code></td><td>Avaric</td></tr>
+    <tr><td><code>az</code></td><td>Azerbaijani</td></tr>
+    <tr><td><code>be</code></td><td>Belarusian</td></tr>
+    <tr><td><code>bg</code></td><td>Bulgarian</td></tr>
+    <tr><td><code>bgc</code></td><td>Haryanvi</td></tr>
+    <tr><td><code>bh</code></td><td>Bihari</td></tr>
+    <tr><td><code>bho</code></td><td>Bhojpuri</td></tr>
+    <tr><td><code>bs</code></td><td>Bosnian</td></tr>
+    <tr><td><code>ch</code></td><td>Chinese (Simplified)</td></tr>
+    <tr><td><code>che</code></td><td>Chechen</td></tr>
+    <tr><td><code>chinese_cht</code></td><td>Chinese (Traditional)</td></tr>
+    <tr><td><code>cs</code></td><td>Czech</td></tr>
+    <tr><td><code>cy</code></td><td>Welsh</td></tr>
+    <tr><td><code>da</code></td><td>Danish</td></tr>
+    <tr><td><code>dar</code></td><td>Dargwa</td></tr>
+    <tr><td><code>de</code> or <code>german</code></td><td>German</td></tr>
+    <tr><td><code>en</code></td><td>English</td></tr>
+    <tr><td><code>es</code></td><td>Spanish</td></tr>
+    <tr><td><code>et</code></td><td>Estonian</td></tr>
+    <tr><td><code>fa</code></td><td>Persian</td></tr>
+    <tr><td><code>fr</code> or <code>french</code></td><td>French</td></tr>
+    <tr><td><code>ga</code></td><td>Irish</td></tr>
+    <tr><td><code>gom</code></td><td>Konkani</td></tr>
+    <tr><td><code>hi</code></td><td>Hindi</td></tr>
+    <tr><td><code>hr</code></td><td>Croatian</td></tr>
+    <tr><td><code>hu</code></td><td>Hungarian</td></tr>
+    <tr><td><code>id</code></td><td>Indonesian</td></tr>
+    <tr><td><code>inh</code></td><td>Ingush</td></tr>
+    <tr><td><code>is</code></td><td>Icelandic</td></tr>
+    <tr><td><code>it</code></td><td>Italian</td></tr>
+    <tr><td><code>japan</code></td><td>Japanese</td></tr>
+    <tr><td><code>ka</code></td><td>Georgian</td></tr>
+    <tr><td><code>kbd</code></td><td>Kabardian</td></tr>
+    <tr><td><code>korean</code></td><td>Korean</td></tr>
+    <tr><td><code>ku</code></td><td>Kurdish</td></tr>
+    <tr><td><code>la</code></td><td>Latin</td></tr>
+    <tr><td><code>lbe</code></td><td>Lak</td></tr>
+    <tr><td><code>lez</code></td><td>Lezghian</td></tr>
+    <tr><td><code>lt</code></td><td>Lithuanian</td></tr>
+    <tr><td><code>lv</code></td><td>Latvian</td></tr>
+    <tr><td><code>mah</code></td><td>Magahi</td></tr>
+    <tr><td><code>mai</code></td><td>Maithili</td></tr>
+    <tr><td><code>mi</code></td><td>Maori</td></tr>
+    <tr><td><code>mn</code></td><td>Mongolian</td></tr>
+    <tr><td><code>mr</code></td><td>Marathi</td></tr>
+    <tr><td><code>ms</code></td><td>Malay</td></tr>
+    <tr><td><code>mt</code></td><td>Maltese</td></tr>
+    <tr><td><code>ne</code></td><td>Nepali</td></tr>
+    <tr><td><code>new</code></td><td>Newari</td></tr>
+    <tr><td><code>nl</code></td><td>Dutch</td></tr>
+    <tr><td><code>no</code></td><td>Norwegian</td></tr>
+    <tr><td><code>oc</code></td><td>Occitan</td></tr>
+    <tr><td><code>pi</code></td><td>Pali</td></tr>
+    <tr><td><code>pl</code></td><td>Polish</td></tr>
+    <tr><td><code>pt</code></td><td>Portuguese</td></tr>
+    <tr><td><code>ro</code></td><td>Romanian</td></tr>
+    <tr><td><code>rs_cyrillic</code></td><td>Serbian (Cyrillic)</td></tr>
+    <tr><td><code>rs_latin</code></td><td>Serbian (Latin)</td></tr>
+    <tr><td><code>ru</code></td><td>Russian</td></tr>
+    <tr><td><code>sa</code></td><td>Sanskrit</td></tr>
+    <tr><td><code>sck</code></td><td>Sadri</td></tr>
+    <tr><td><code>sk</code></td><td>Slovak</td></tr>
+    <tr><td><code>sl</code></td><td>Slovenian</td></tr>
+    <tr><td><code>sq</code></td><td>Albanian</td></tr>
+    <tr><td><code>sv</code></td><td>Swedish</td></tr>
+    <tr><td><code>sw</code></td><td>Swahili</td></tr>
+    <tr><td><code>tab</code></td><td>Tabassaran</td></tr>
+    <tr><td><code>ta</code></td><td>Tamil</td></tr>
+    <tr><td><code>te</code></td><td>Telugu</td></tr>
+    <tr><td><code>tl</code></td><td>Tagalog</td></tr>
+    <tr><td><code>tr</code></td><td>Turkish</td></tr>
+    <tr><td><code>ug</code></td><td>Uyghur</td></tr>
+    <tr><td><code>uk</code></td><td>Ukrainian</td></tr>
+    <tr><td><code>ur</code></td><td>Urdu</td></tr>
+    <tr><td><code>uz</code></td><td>Uzbek</td></tr>
+    <tr><td><code>vi</code></td><td>Vietnamese</td></tr>
+  </tbody>
+</table>
+
+</details>
+
+<details><summary><b>Correspondence Between OCR Model Versions and Supported Languages</b></summary>
+
+<table border="1" cellspacing="0" cellpadding="4">
+  <thead>
+    <tr>
+      <th><code>ocr_version</code></th>
+      <th>Supported <code>lang</code></th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td><code>PP-OCRv5</code></td>
+      <td><code>ch</code>, <code>en</code>, <code>fr</code>, <code>de</code>, <code>japan</code>, <code>korean</code>, <code>chinese_cht</code>, <code>af</code>, <code>it</code>, <code>es</code>, <code>bs</code>, <code>pt</code>, <code>cs</code>, <code>cy</code>, <code>da</code>, <code>et</code>, <code>ga</code>, <code>hr</code>, <code>hu</code>, <code>rslatin</code>, <code>id</code>, <code>oc</code>, <code>is</code>, <code>lt</code>, <code>mi</code>, <code>ms</code>, <code>nl</code>, <code>no</code>, <code>pl</code>, <code>sk</code>, <code>sl</code>, <code>sq</code>, <code>sv</code>, <code>sw</code>, <code>tl</code>, <code>tr</code>, <code>uz</code>, <code>la</code>, <code>ru</code>, <code>be</code>, <code>uk</code></td>
+    </tr>
+    <tr>
+      <td><code>PP-OCRv4</code></td>
+      <td><code>ch</code>, <code>en</code></td>
+    </tr>
+    <tr>
+      <td><code>PP-OCRv3</code></td>
+      <td>
+        <code>abq</code>, <code>af</code>, <code>ady</code>, <code>ang</code>, <code>ar</code>, <code>ava</code>, <code>az</code>, <code>be</code>,
+        <code>bg</code>, <code>bgc</code>, <code>bh</code>, <code>bho</code>, <code>bs</code>, <code>ch</code>, <code>che</code>,
+        <code>chinese_cht</code>, <code>cs</code>, <code>cy</code>, <code>da</code>, <code>dar</code>, <code>de</code>, <code>german</code>,
+        <code>en</code>, <code>es</code>, <code>et</code>, <code>fa</code>, <code>fr</code>, <code>french</code>, <code>ga</code>, <code>gom</code>,
+        <code>hi</code>, <code>hr</code>, <code>hu</code>, <code>id</code>, <code>inh</code>, <code>is</code>, <code>it</code>, <code>japan</code>,
+        <code>ka</code>, <code>kbd</code>, <code>korean</code>, <code>ku</code>, <code>la</code>, <code>lbe</code>, <code>lez</code>, <code>lt</code>,
+        <code>lv</code>, <code>mah</code>, <code>mai</code>, <code>mi</code>, <code>mn</code>, <code>mr</code>, <code>ms</code>, <code>mt</code>,
+        <code>ne</code>, <code>new</code>, <code>nl</code>, <code>no</code>, <code>oc</code>, <code>pi</code>, <code>pl</code>, <code>pt</code>,
+        <code>ro</code>, <code>rs_cyrillic</code>, <code>rs_latin</code>, <code>ru</code>, <code>sa</code>, <code>sck</code>, <code>sk</code>,
+        <code>sl</code>, <code>sq</code>, <code>sv</code>, <code>sw</code>, <code>ta</code>, <code>tab</code>, <code>te</code>, <code>tl</code>,
+        <code>tr</code>, <code>ug</code>, <code>uk</code>, <code>ur</code>, <code>uz</code>, <code>vi</code>
+      </td>
+    </tr>
+  </tbody>
+</table>
+
+</details>
diff --git a/docs/version3.x/pipeline_usage/PP-StructureV3.md b/docs/version3.x/pipeline_usage/PP-StructureV3.md
new file mode 100644
index 0000000..93725f9
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PP-StructureV3.md
@@ -0,0 +1,4080 @@
+---
+comments: true
+---
+
+# PP-StructureV3 产线使用教程
+
+## 1. PP-StructureV3 产线介绍
+
+版面解析是一种从文档图像中提取结构化信息的技术，主要用于将复杂的文档版面转换为机器可读的数据格式。这项技术在文档管理、信息提取和数据数字化等领域具有广泛的应用。版面解析通过结合光学字符识别（OCR）、图像处理和机器学习算法，能够识别和提取文档中的文本块、标题、段落、图片、表格以及其他版面元素。此过程通常包括版面分析、元素分析和数据格式化三个主要步骤，最终生成结构化的文档数据，提升数据处理的效率和准确性。<b>PP-StructureV3 产线在通用版面解析v1产线的基础上，强化了版面区域检测、表格识别、公式识别的能力，增加了图表理解能力和多栏阅读顺序的恢复能力、结果转换 Markdown 文件的能力，在多种文档数据中，表现优异，可以处理较复杂的文档数据。</b>本产线同时提供了灵活的服务化部署方式，支持在多种硬件上使用多种编程语言调用。不仅如此，本产线也提供了二次开发的能力，您可以基于本产线在您自己的数据集上训练调优，训练后的模型也可以无缝集成。
+
+<b>PP-StructureV3 产线中包含以下7个模块或子产线。每个模块或子产线均可独立进行训练和推理，并包含多个模型。有关详细信息，请点击相应链接以查看文档。</b>
+
+- [版面区域检测模块](../module_usage/layout_detection.md)
+- [通用OCR子产线](./OCR.md)
+- [文档图像预处理子产线](./doc_preprocessor.md) （可选）
+- [表格识别子产线](./table_recognition_v2.md) （可选）
+- [印章文本识别子产线](./seal_recognition.md) （可选）
+- [公式识别子产线](./formula_recognition.md) （可选）
+- [图表解析模块](../module_usage/chart_parsing.md) （可选）
+
+在本产线中，您可以根据下方的基准测试数据选择使用的模型。
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。
+> 在带有 [常规模式 / 高性能模式] 标记的推理耗时列中，`常规模式` 对应本地飞桨推理引擎。各模块会根据默认模型名称选择合适的本地飞桨推理引擎：仅支持动态图的模型使用 `paddle_dynamic`；同时支持静态图和动态图的模型优先使用 `paddle_static`。
+
+<details>
+<summary><b>文档图像方向分类模块：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>Top-1 Acc（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">训练模型</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>基于PP-LCNet_x1_0的文档图像分类模型，含有四个类别，即0度，90度，180度，270度</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>文本图像矫正模块：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>CER </th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">训练模型</a></td>
+<td>0.179</td>
+<td>19.05 / 19.05</td>
+<td>- / 869.82</td>
+<td>30.3</td>
+<td>高精度文本图像矫正模型</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>版面区域检测模块：</b></summary>
+
+* <b>版面检测模型，包含20个常见的类别：文档标题、段落标题、文本、页码、摘要、目录、参考文献、脚注、页眉、页脚、算法、公式、公式编号、图像、表格、图和表标题（图标题、表格标题和图表标题）、印章、图表、侧栏文本和参考文献内容</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout_plus-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout_plus-L_pretrained.pdparams">训练模型</a></td>
+<td>83.2</td>
+<td>53.03 / 17.23</td>
+<td>634.62 / 378.32</td>
+<td>126.01</td>
+<td>基于RT-DETR-L在包含中英文论文、多栏杂志、报纸、PPT、合同、书本、试卷、研报、古籍、日文文档、竖版文字文档等场景的自建数据集训练的更高精度版面区域定位模型</td>
+</tr>
+<tr>
+</tbody>
+</table>
+
+<b>注：以上精度指标的评估集是自建的版面区域检测数据集，包含中英文论文、杂志、报纸、研报、PPT、试卷、课本等 1300 张文档类型图片。</b>
+
+* <b>文档图像版面子模块检测，包含1个 版面区域 类别，能检测多栏的报纸、杂志的每个子文章的文本区域：</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocBlockLayout</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBlockLayout_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocBlockLayout_pretrained.pdparams">训练模型</a></td>
+<td>95.9</td>
+<td>34.60 / 28.54</td>
+<td>506.43 / 256.83</td>
+<td>123.92</td>
+<td>基于RT-DETR-L在包含中英文论文、多栏杂志、报纸、PPT、合同、书本、试卷、研报、古籍、日文文档、竖版文字文档等场景的自建数据集训练的文档图像版面子模块检测模型</td>
+</tr>
+<tr>
+</tbody>
+</table>
+
+<b>注：以上精度指标的评估集是自建的版面子区域检测数据集，包含中英文论文、杂志、报纸、研报、PPT、试卷、课本等 1000 张文档类型图片。</b>
+
+* <b>版面检测模型，包含23个常见的类别：文档标题、段落标题、文本、页码、摘要、目录、参考文献、脚注、页眉、页脚、算法、公式、公式编号、图像、图表标题、表格、表格标题、印章、图表标题、图表、页眉图像、页脚图像、侧栏文本</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-L_pretrained.pdparams">训练模型</a></td>
+<td>90.4</td>
+<td>33.59 / 33.59</td>
+<td>503.01 / 251.08</td>
+<td>123.76</td>
+<td>基于RT-DETR-L在包含中英文论文、杂志、合同、书本、试卷和研报等场景的自建数据集训练的高精度版面区域定位模型</td>
+</tr>
+<tr>
+<td>PP-DocLayout-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-M_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-M_pretrained.pdparams">训练模型</a></td>
+<td>75.2</td>
+<td>13.03 / 4.72</td>
+<td>43.39 / 24.44</td>
+<td>22.578</td>
+<td>基于PicoDet-L在包含中英文论文、杂志、合同、书本、试卷和研报等场景的自建数据集训练的精度效率平衡的版面区域定位模型</td>
+</tr>
+<tr>
+<td>PP-DocLayout-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-S_pretrained.pdparams">训练模型</a></td>
+<td>70.9</td>
+<td>11.54 / 3.86</td>
+<td>18.53 / 6.29</td>
+<td>4.834</td>
+<td>基于PicoDet-S在中英文论文、杂志、合同、书本、试卷和研报等场景上自建数据集训练的高效率版面区域定位模型</td>
+</tr>
+</tbody>
+</table>
+
+<b>注：以上精度指标的评估集是自建的版面区域检测数据集，包含中英文论文、报纸、研报和试卷等 500 张文档类型图片。</b>
+
+> ❗ 以上列出的是版面检测模块重点支持的<b>5个核心模型</b>，该模块总共支持<b>13个全量模型</b>，包含多个预定义了不同类别的模型，完整的模型列表如下：
+
+<details><summary> 👉模型列表详情</summary>
+
+* <b>表格版面检测模型</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet_layout_1x_table</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_table_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_table_pretrained.pdparams">训练模型</a></td>
+<td>97.5</td>
+<td>9.57 / 6.63</td>
+<td>27.66 / 16.75</td>
+<td>7.4</td>
+<td>基于PicoDet-1x在自建数据集训练的高效率版面区域定位模型，可定位表格这1类区域</td>
+</tr>
+</tbody></table>
+
+* <b>3类版面检测模型，包含表格、图像、印章</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_3cls_pretrained.pdparams">训练模型</a></td>
+<td>88.2</td>
+<td>8.43 / 3.44</td>
+<td>17.60 / 6.51</td>
+<td>4.8</td>
+<td>基于PicoDet-S轻量模型在中英文论文、杂志和研报等场景上自建数据集训练的高效率版面区域定位模型</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_3cls_pretrained.pdparams">训练模型</a></td>
+<td>89.0</td>
+<td>12.80 / 9.57</td>
+<td>45.04 / 23.86</td>
+<td>22.6</td>
+<td>基于PicoDet-L在中英文论文、杂志和研报等场景上自建数据集训练的效率精度均衡版面区域定位模型</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_3cls_pretrained.pdparams">训练模型</a></td>
+<td>95.8</td>
+<td>114.80 / 25.65</td>
+<td>924.38 / 924.38</td>
+<td>470.1</td>
+<td>基于RT-DETR-H在中英文论文、杂志和研报等场景上自建数据集训练的高精度版面区域定位模型</td>
+</tr>
+</tbody></table>
+
+* <b>5类英文文档区域检测模型，包含文字、标题、表格、图片以及列表</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet_layout_1x</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_pretrained.pdparams">训练模型</a></td>
+<td>97.8</td>
+<td>9.62 / 6.75</td>
+<td>26.96 / 12.77</td>
+<td>7.4</td>
+<td>基于PicoDet-1x在PubLayNet数据集训练的高效率英文文档版面区域定位模型</td>
+</tr>
+</tbody></table>
+
+* <b>17类区域检测模型，包含17个版面常见类别，分别是：段落标题、图片、文本、数字、摘要、内容、图表标题、公式、表格、表格标题、参考文献、文档标题、脚注、页眉、算法、页脚、印章</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>87.4</td>
+<td>8.80 / 3.62</td>
+<td>17.51 / 6.35</td>
+<td>4.8</td>
+<td>基于PicoDet-S轻量模型在中英文论文、杂志和研报等场景上自建数据集训练的高效率版面区域定位模型</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>89.0</td>
+<td>12.60 / 10.27</td>
+<td>43.70 / 24.42</td>
+<td>22.6</td>
+<td>基于PicoDet-L在中英文论文、杂志和研报等场景上自建数据集训练的效率精度均衡版面区域定位模型</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>98.3</td>
+<td>115.29 / 101.18</td>
+<td>964.75 / 964.75</td>
+<td>470.2</td>
+<td>基于RT-DETR-H在中英文论文、杂志和研报等场景上自建数据集训练的高精度版面区域定位模型</td>
+</tr>
+</tbody>
+</table>
+</details>
+</details>
+
+<details>
+<summary><b>表格结构识别模块：</b></summary>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>精度（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>SLANeXt_wired</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wired_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wired_pretrained.pdparams">训练模型</a></td>
+<td rowspan="2">69.65</td>
+<td rowspan="2">85.92 / 85.92</td>
+<td rowspan="2">- / 501.66</td>
+<td rowspan="2">351</td>
+<td rowspan="2">SLANeXt 系列是百度飞桨视觉团队自研的新一代表格结构识别模型。相较于 SLANet 和 SLANet_plus，SLANeXt 专注于对表格结构进行识别，并且对有线表格(wired)和无线表格(wireless)的识别分别训练了专用的权重，对各类型表格的识别能力都得到了明显提高，特别是对有线表格的识别能力得到了大幅提升。</td>
+</tr>
+<tr>
+<td>SLANeXt_wireless</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wireless_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wireless_pretrained.pdparams">训练模型</a></td>
+</tr>
+</table>
+
+<p><b>表格分类模块模型：</b></p>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>Top1 Acc(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+</tr>
+<tr>
+<td>PP-LCNet_x1_0_table_cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/CLIP_vit_base_patch16_224_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_table_cls_pretrained.pdparams">训练模型</a></td>
+<td>94.2</td>
+<td>2.62 / 0.60</td>
+<td>3.17 / 1.14</td>
+<td>6.6</td>
+</tr>
+</table>
+
+<p><b>表格单元格检测模块模型：</b></p>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>RT-DETR-L_wired_table_cell_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wired_table_cell_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wired_table_cell_det_pretrained.pdparams">训练模型</a></td>
+<td rowspan="2">82.7</td>
+<td rowspan="2">33.47 / 27.02</td>
+<td rowspan="2">402.55 / 256.56</td>
+<td rowspan="2">124</td>
+<td rowspan="2">RT-DETR 是第一个实时的端到端目标检测模型。百度飞桨视觉团队基于 RT-DETR-L 作为基础模型，在自建表格单元格检测数据集上完成预训练，实现了对有线表格、无线表格均有较好性能的表格单元格检测。
+</td>
+</tr>
+<tr>
+<td>RT-DETR-L_wireless_table_cell_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wireless_table_cell_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wireless_table_cell_det_pretrained.pdparams">训练模型</a></td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary><b>文本检测模块：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>检测Hmean（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv5_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_det_pretrained.pdparams">训练模型</a></td>
+<td>83.8</td>
+<td>89.55 / 70.19</td>
+<td>383.15 / 383.15</td>
+<td>84.3</td>
+<td>PP-OCRv5 的服务端文本检测模型，精度更高，适合在性能较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_det_pretrained.pdparams">训练模型</a></td>
+<td>79.0</td>
+<td>10.67 / 6.36</td>
+<td>57.77 / 28.15</td>
+<td>4.7</td>
+<td>PP-OCRv5 的移动端文本检测模型，效率更高，适合在端侧设备部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_det_pretrained.pdparams">训练模型</a></td>
+<td>69.2</td>
+<td>127.82 / 98.87</td>
+<td>585.95 / 489.77</td>
+<td>109</td>
+<td>PP-OCRv4 的服务端文本检测模型，精度更高，适合在性能较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams">训练模型</a></td>
+<td>63.8</td>
+<td>9.87 / 4.17</td>
+<td>56.60 / 20.79</td>
+<td>4.7</td>
+<td>PP-OCRv4 的移动端文本检测模型，效率更高，适合在端侧设备部署</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>文本识别模块：</b></summary>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>86.38</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec 是新一代文本识别模型。该模型致力于以单一模型高效、精准地支持简体中文、繁体中文、英文、日文四种主要语言，以及手写、竖版、拼音、生僻字等复杂文本场景的识别。在保持识别效果的同时，兼顾推理速度和模型鲁棒性，为各种场景下的文档理解提供高效、精准的技术支撑。</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.29</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">训练模型</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc是在PP-OCRv4_server_rec的基础上，在更多中文文档数据和PP-OCR训练数据的混合数据训练而成，增加了部分繁体字、日文、特殊字符的识别能力，可支持识别的字符为1.5万+，除文档相关的文字识别能力提升外，也同时提升了通用文字的识别能力</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>PP-OCRv4的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>PP-OCRv4的服务器端模型，推理精度高，可以部署在多种不同的服务器上</td>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>基于PP-OCRv4识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+</table>
+
+> ❗ 以上列出的是文本识别模块重点支持的<b>6个核心模型</b>，该模块总共支持<b>20个全量模型</b>，包含多个多语言文本识别模型，完整的模型列表如下：
+
+<details><summary> 👉模型列表详情</summary>
+
+* <b>PP-OCRv5 多场景模型</b>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>中文识别 Avg Accuracy(%)</th>
+<th>英文识别 Avg Accuracy(%)</th>
+<th>繁体中文识别 Avg Accuracy(%)</th>
+<th>日文识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>86.38</td>
+<td>64.70</td>
+<td>93.29</td>
+<td>60.35</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec 是新一代文本识别模型。该模型致力于以单一模型高效、精准地支持简体中文、繁体中文、英文、日文四种主要语言，以及手写、竖版、拼音、生僻字等复杂文本场景的识别。在保持识别效果的同时，兼顾推理速度和模型鲁棒性，为各种场景下的文档理解提供高效、精准的技术支撑。</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.29</td>
+<td>66.00</td>
+<td>83.55</td>
+<td>54.65</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+</table>
+
+* <b>中文识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">训练模型</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc是在PP-OCRv4_server_rec的基础上，在更多中文文档数据和PP-OCR训练数据的混合数据训练而成，增加了部分繁体字、日文、特殊字符的识别能力，可支持识别的字符为1.5万+，除文档相关的文字识别能力提升外，也同时提升了通用文字的识别能力</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>PP-OCRv4的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>PP-OCRv4的服务器端模型，推理精度高，可以部署在多种不同的服务器上</td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>72.96</td>
+<td>3.89 / 1.16</td>
+<td>8.72 / 3.56</td>
+<td>10.3</td>
+<td>PP-OCRv3的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>ch_SVTRv2_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_SVTRv2_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_SVTRv2_rec_pretrained.pdparams">训练模型</a></td>
+<td>68.81</td>
+<td>10.38 / 8.31</td>
+<td>66.52 / 30.83</td>
+<td>80.5</td>
+<td rowspan="1">
+SVTRv2 是一种由复旦大学视觉与学习实验室（FVL）的OpenOCR团队研发的服务端文本识别模型，其在PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务中荣获一等奖，A榜端到端识别精度相比PP-OCRv4提升6%。
+</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>ch_RepSVTR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_RepSVTR_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_RepSVTR_rec_pretrained.pdparams">训练模型</a></td>
+<td>65.07</td>
+<td>6.29 / 1.57</td>
+<td>20.64 / 5.40</td>
+<td>48.8</td>
+<td rowspan="1">    RepSVTR 文本识别模型是一种基于SVTRv2 的移动端文本识别模型，其在PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务中荣获一等奖，B榜端到端识别精度相比PP-OCRv4提升2.5%，推理速度持平。</td>
+</tr>
+</table>
+
+* <b>英文识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td> 70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>基于PP-OCRv4识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+<tr>
+<td>en_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>70.69</td>
+<td>3.56 / 0.78</td>
+<td>8.44 / 5.78</td>
+<td>17.3</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+</table>
+
+
+* <b>多语言识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>korean_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>60.21</td>
+<td>3.73 / 0.98</td>
+<td>8.76 / 2.91</td>
+<td>9.6</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量韩文识别模型，支持韩文、数字识别</td>
+</tr>
+<tr>
+<td>japan_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+japan_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/japan_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>45.69</td>
+<td>3.86 / 1.01</td>
+<td>8.62 / 2.92</td>
+<td>9.8</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量日文识别模型，支持日文、数字识别</td>
+</tr>
+<tr>
+<td>chinese_cht_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+chinese_cht_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/chinese_cht_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>82.06</td>
+<td>3.90 / 1.16</td>
+<td>9.24 / 3.18</td>
+<td>10.8</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量繁体中文识别模型，支持繁体中文、数字识别</td>
+</tr>
+<tr>
+<td>te_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+te_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>95.88</td>
+<td>3.59 / 0.81</td>
+<td>8.28 / 6.21</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量泰卢固文识别模型，支持泰卢固文、数字识别</td>
+</tr>
+<tr>
+<td>ka_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ka_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ka_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>96.96</td>
+<td>3.49 / 0.89</td>
+<td>8.63 / 2.77</td>
+<td>17.4</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量卡纳达文识别模型，支持卡纳达文、数字识别</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ta_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>76.83</td>
+<td>3.49 / 0.86</td>
+<td>8.35 / 3.41</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量泰米尔文识别模型，支持泰米尔文、数字识别</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>76.93</td>
+<td>3.53 / 0.78</td>
+<td>8.50 / 6.83</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量拉丁文识别模型，支持拉丁文、数字识别</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+arabic_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>73.55</td>
+<td>3.60 / 0.83</td>
+<td>8.44 / 4.69</td>
+<td>17.3</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量阿拉伯字母识别模型，支持阿拉伯字母、数字识别</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+cyrillic_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>94.28</td>
+<td>3.56 / 0.79</td>
+<td>8.22 / 2.76</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量斯拉夫字母识别模型，支持斯拉夫字母、数字识别</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>96.44</td>
+<td>3.60 / 0.78</td>
+<td>6.95 / 2.87</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量梵文字母识别模型，支持梵文字母、数字识别</td>
+</tr>
+</table>
+</details>
+</details>
+
+<details>
+<summary><b>文本行方向分类模块（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th>
+<th>模型下载链接</th>
+<th>Top-1 Acc（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x0_25_textline_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x0_25_textline_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x0_25_textline_ori_pretrained.pdparams">训练模型</a></td>
+<td>98.85</td>
+<td>2.16 / 0.41</td>
+<td>2.37 / 0.73</td>
+<td>0.96</td>
+<td>基于PP-LCNet_x0_25的文本行分类模型，含有两个类别，即0度，180度</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>公式识别模块：</b></summary>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>En-BLEU(%)</th>
+<th>Zh-BLEU(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>UniMERNet</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UniMERNet_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UniMERNet_pretrained.pdparams">训练模型</a></td>
+<td>85.91</td>
+<td>43.50</td>
+<td>1311.84 / 1311.84</td>
+<td>- / 8288.07</td>
+<td>1530</td>
+<td>UniMERNet是由上海AI Lab研发的一款公式识别模型。该模型采用Donut Swin作为编码器，MBartDecoder作为解码器，并通过在包含简单公式、复杂公式、扫描捕捉公式和手写公式在内的一百万数据集上进行训练，大幅提升了模型对真实场景公式的识别准确率</td>
+</tr>
+<tr>
+<td>PP-FormulaNet-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-S_pretrained.pdparams">训练模型</a></td>
+<td>87.00</td>
+<td>45.71</td>
+<td>182.25 / 182.25</td>
+<td>- / 254.39</td>
+<td>224</td>
+<td rowspan="2">PP-FormulaNet 是由百度飞桨视觉团队开发的一款先进的公式识别模型，支持5万个常见LateX源码词汇的识别。PP-FormulaNet-S 版本采用了 PP-HGNetV2-B4 作为其骨干网络，通过并行掩码和模型蒸馏等技术，大幅提升了模型的推理速度，同时保持了较高的识别精度，适用于简单印刷公式、跨行简单印刷公式等场景。而 PP-FormulaNet-L 版本则基于 Vary_VIT_B 作为骨干网络，并在大规模公式数据集上进行了深入训练，在复杂公式的识别方面，相较于PP-FormulaNet-S表现出显著的提升，适用于简单印刷公式、复杂印刷公式、手写公式等场景。 </td>
+
+</tr>
+<td>PP-FormulaNet-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-L_pretrained.pdparams">训练模型</a></td>
+<td>90.36</td>
+<td>45.78</td>
+<td>1482.03 / 1482.03</td>
+<td>- / 3131.54</td>
+<td>695</td>
+</tr>
+<td>PP-FormulaNet_plus-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-S_pretrained.pdparams">训练模型</a></td>
+<td>88.71</td>
+<td>53.32</td>
+<td>179.20 / 179.20</td>
+<td>- / 260.99</td>
+<td>248</td>
+<td rowspan="3">PP-FormulaNet_plus 是百度飞桨视觉团队在 PP-FormulaNet 的基础上开发的增强版公式识别模型。与原版相比，PP-FormulaNet_plus 在训练中使用了更为丰富的公式数据集，包括中文学位论文、专业书籍、教材试卷以及数学期刊等多种来源。这一扩展显著提升了模型的识别能力。
+
+其中，PP-FormulaNet_plus-M 和 PP-FormulaNet_plus-L 模型新增了对中文公式的支持，并将公式的最大预测 token 数从 1024 扩大至 2560，大幅提升了对复杂公式的识别性能。同时，PP-FormulaNet_plus-S 模型则专注于增强英文公式的识别能力。通过这些改进，PP-FormulaNet_plus 系列模型在处理复杂多样的公式识别任务时表现更加出色。 </td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-M_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-M_pretrained.pdparams">训练模型</a></td>
+<td>91.45</td>
+<td>89.76</td>
+<td>1040.27 / 1040.27</td>
+<td>- / 1615.80</td>
+<td>592</td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-L_pretrained.pdparams">训练模型</a></td>
+<td>92.22</td>
+<td>90.64</td>
+<td>1476.07 / 1476.07</td>
+<td>- / 3125.58</td>
+<td>698</td>
+</tr>
+<tr>
+<td>LaTeX_OCR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/LaTeX_OCR_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/LaTeX_OCR_rec_pretrained.pdparams">训练模型</a></td>
+<td>74.55</td>
+<td>39.96</td>
+<td>1088.89 / 1088.89</td>
+<td>- / -</td>
+<td>99</td>
+<td>LaTeX-OCR是一种基于自回归大模型的公式识别算法，通过采用 Hybrid ViT 作为骨干网络，transformer作为解码器，显著提升了公式识别的准确性。</td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary><b>印章文本检测模块：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>检测Hmean（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv4_server_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_seal_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_seal_det_pretrained.pdparams">训练模型</a></td>
+<td>98.40</td>
+<td>124.64 / 91.57</td>
+<td>545.68 / 439.86</td>
+<td>109</td>
+<td>PP-OCRv4的服务端印章文本检测模型，精度更高，适合在较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_seal_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_seal_det_pretrained.pdparams">训练模型</a></td>
+<td>96.36</td>
+<td>9.70 / 3.56</td>
+<td>50.38 / 19.64</td>
+<td>4.7</td>
+<td>PP-OCRv4的移动端印章文本检测模型，效率更高，适合在端侧部署</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>图表解析模块：</b></summary>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>模型参数规模（B）</th>
+<th>模型存储大小（GB）</th>
+<th>模型分数 </th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-Chart2Table</td><td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-Chart2Table_infer.tar">推理模型</a></td>
+<td>0.58</td>
+<td>1.4</td>
+<th>75.98</th>
+<td>PP-Chart2Table是飞桨团队自研的一款专注于图表解析的多模态模型，在中英文图表解析任务中展现出卓越性能。团队采用精心设计的数据生成策略，构建了近70万条高质量的图表解析多模态数据集，全面覆盖饼图、柱状图、堆叠面积图等常见图表类型及各类应用场景。同时设计了二阶段训练方法，结合大模型蒸馏实现对海量无标注OOD数据的充分利用。在内部业务的中英文场景测试中，PP-Chart2Table不仅达到同参数量级模型中的SOTA水平，更在关键场景中实现了与7B参数量级VLM模型相当的精度。</td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary><b>测试环境说明：</b></summary>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+            <li><strong>测试数据集：
+             </strong>
+                <ul>
+                  <li>文档图像方向分类模型：自建的内部数据集，覆盖证件和文档等多个场景，包含 1000 张图片。</li>
+                  <li> 文本图像矫正模型：<a href="https://www3.cs.stonybrook.edu/~cvl/docunet.html">DocUNet</a>。</li>
+                  <li>版面区域检测模型：PaddleOCR 自建的版面区域分析数据集，包含中英文论文、杂志和研报等常见的 1w 张文档类型图片。</li>
+                  <li>表格结构识别模型：PaddleX 内部自建英文表格识别数据集。 </li>
+                  <li>文本检测模型：PaddleOCR 自建的中文数据集，覆盖街景、网图、文档、手写多个场景，其中检测包含 500 张图片。</li>
+                  <li> 中文识别模型： PaddleOCR 自建的中文数据集，覆盖街景、网图、文档、手写多个场景，其中文本识别包含 1.1w 张图片。</li>
+                  <li>ch_SVTRv2_rec：<a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务</a>A榜评估集。</li>
+                  <li> ch_RepSVTR_rec：<a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务</a>B榜评估集。</li>
+                  <li>英文识别模型：自建的内部英文数据集。</li>
+                  <li> 多语言识别模型：自建的内部多语种数据集。</li>
+                  <li>文本行方向分类模型：自建的内部数据集，覆盖证件和文档等多个场景，包含 1000 张图片。</li>
+                  <li> 印章文本检测模型：自建的内部数据集，包含500张圆形印章图像。</li>
+                </ul>
+             </li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td>本地飞桨推理引擎（默认情况下根据默认模型名称选择合适引擎；若静态图和动态图均可用则优先 <code>paddle_static</code>）</td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+
+</details>
+
+<br />
+<b>如果您更注重模型的精度，请选择精度较高的模型；如果您更在意模型的推理速度，请选择推理速度较快的模型；如果您关注模型的存储大小，请选择存储体积较小的模型。</b>
+
+## 2. 快速开始
+
+在本地使用 PP-StructureV3 产线前，请确保您已经按照[安装教程](../installation.md)完成了wheel包安装。安装完成后，可以在本地使用命令行体验或 Python 集成。如果您希望选择性安装依赖，请参考安装教程中的相关说明。该产线对应的依赖分组为 `doc-parser`。
+
+**请注意，如果在执行过程中遇到程序失去响应、程序异常退出、内存资源耗尽、推理速度极慢等问题，请尝试参考文档调整配置，例如关闭不需要使用的功能或使用更轻量的模型。**
+<!--Luxorion-12 修改：修改表格参数描述-->
+### 2.1 命令行方式体验
+
+一行命令即可快速体验 PP-StructureV3 产线效果：
+
+```bash
+paddleocr pp_structurev3 -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/pp_structure_v3_demo.png
+
+# 通过 --use_doc_orientation_classify 指定是否使用文档方向分类模型
+paddleocr pp_structurev3 -i ./pp_structure_v3_demo.png --use_doc_orientation_classify True
+
+# 通过 --use_doc_unwarping 指定是否使用文本图像矫正模块
+paddleocr pp_structurev3 -i ./pp_structure_v3_demo.png --use_doc_unwarping True
+
+# 通过 --use_textline_orientation 指定是否使用文本行方向分类模型
+paddleocr pp_structurev3 -i ./pp_structure_v3_demo.png --use_textline_orientation False
+
+# 通过 --device 指定模型推理时使用 GPU
+paddleocr pp_structurev3 -i ./pp_structure_v3_demo.png --device gpu
+```
+
+上述命令默认使用本地飞桨推理引擎。默认情况下，各模块会根据默认模型名称选择合适的本地飞桨推理引擎：仅支持动态图的模型使用 `paddle_dynamic`；同时支持静态图和动态图的模型优先使用 `paddle_static`。如需运行，请先参考[飞桨框架安装说明](../paddlepaddle_installation.md)安装 PaddlePaddle。
+
+如果选择 `transformers` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 Transformers 环境配置，然后执行如下命令：
+
+```bash
+# 部分模型尚在支持中，推理时需关闭公式识别功能并更换无线表格结构识别模型，请使用以下命令：
+paddleocr pp_structurev3 -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/pp_structure_v3_demo.png \
+    --engine transformers
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 ONNX Runtime 环境配置，然后执行如下命令：
+
+```bash
+# 使用 onnxruntime 引擎进行推理
+# 部分模型尚在支持中，推理时需关闭公式识别功能，请使用以下命令：
+paddleocr pp_structurev3 -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/pp_structure_v3_demo.png \
+    --engine onnxruntime --use_formula_recognition False
+```
+
+<details><summary><b>命令行支持更多参数设置，点击展开以查看命令行参数的详细说明</b></summary>
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，必填。<br><b>说明：</b>如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<br><b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/general_ocr_002.png">示例</a>；</br><b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)。</br>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td><b>含义：</b>指定推理结果文件保存的路径。<br><b>说明：</b>如果不设置，推理结果将不会保存到本地。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>含义：</b>版面区域检测的模型名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>含义：</b>版面区域检测模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>版面模型得分阈值。
+<br><b>说明：</b><code>0-1</code> 之间的任意浮点数。</br>如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>0.5</code>。
+</td>
+<td><code>float</code></td>
+<td><code>0.5</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>版面检测是否使用后处理NMS。<br><b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</br></td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>版面区域检测模型检测框的扩张系数。
+<br><b>说明：</b>任意大于 <code>0</code>  浮点数。</br>如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>1.0</code>。
+</td>
+<td><code>float</code></td>
+<td><code>1.0</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>版面检测中模型输出的检测框的合并处理模式。
+<br><b>说明：</b>
+<ul>
+<li><b>large</b>，设置为large时，表示在模型输出的检测框中，对于互相重叠包含的检测框，只保留外部最大的框，删除重叠的内部框；</li>
+<li><b>small</b>，设置为small，表示在模型输出的检测框中，对于互相重叠包含的检测框，只保留内部被包含的小框，删除重叠的外部框；</li>
+<li><b>union</b>，不进行框的过滤处理，内外框都保留；</li>
+</ul>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>large</code>。
+</br></td>
+<td><code>str</code></td>
+<td><code>large</code></td>
+</tr>
+<tr>
+<td><code>chart_recognition_model_name</code></td>
+<td><b>含义：</b>图表解析的模型名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>chart_recognition_model_dir</code></td>
+<td><b>含义：</b>图表解析模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>chart_recognition_batch_size</code></td>
+<td><b>含义：</b>图表解析模型的batch size。<br><b>说明：</b>如果不设置，将默认设置batch size为<code>1</code>。</br></td>
+<td><code>int</code></td>
+<td><code>1</code></td>
+</tr>
+<tr>
+<td><code>region_detection_model_name</code></td>
+<td><b>含义：</b>文档图像版面子模块检测的模型名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>region_detection_model_dir</code></td>
+<td><b>含义：</b>文档图像版面子模块检测模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>含义：</b>文档方向分类模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>文档方向分类模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>含义：</b>文本图像矫正模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>含义：</b>文本检测模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>含义：</b>文本检测模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>含义：</b>文本检测的图像边长限制。
+<br><b>说明：</b>大于 <code>0</code> 的任意整数。</br>如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>960</code>。
+</td>
+<td><code>int</code></td>
+<td><code>960</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>含义：</b>文本检测的图像边长限制类型。
+<br><b>说明：</b>支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>。</br>如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>max</code>。
+</td>
+<td><code>str</code></td>
+<td><code>max</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>含义：</b>检测像素阈值，输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。
+<br><b>说明：</b>大于 <code>0</code> 的任意浮点数。</br>如果不设置，将默认使用产线初始化的该参数值 <code>0.3</code>。
+</td>
+<td><code>float</code></td>
+<td><code>0.3</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>含义：</b>检测框阈值，检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。
+<br><b>说明：</b>大于 <code>0</code> 的任意浮点数。</br>如果不设置，将默认使用产线初始化的该参数值 <code>0.6</code>。
+</td>
+<td><code>float</code></td>
+<td><code>0.6</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>含义：</b>文本检测扩张系数，使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。
+<br><b>说明：</b>大于 <code>0</code> 的任意浮点数。</br>如果不设置，将默认使用产线初始化的该参数值 <code>2.0</code>。
+</td>
+<td><code>float</code></td>
+<td><code>2.0</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_name</code></td>
+<td><b>含义：</b>文本行方向模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_dir</code></td>
+<td><b>含义：</b>文本行方向模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>textline_orientation_batch_size</code></td>
+<td><b>含义：</b>文本行方向模型的batch size。<br><b>说明：</b>如果不设置，将默认设置batch size为<code>1</code>。</br></td>
+<td><code>int</code></td>
+<td><code>1</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>含义：</b>文本识别模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>含义：</b>文本识别模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>含义：</b>文本识别模型的batch size。<br><b>说明：</b>如果不设置，将默认设置batch size为<code>1</code>。</br></td>
+<td><code>int</code></td>
+<td><code>1</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>含义：</b>文本识别阈值，得分大于该阈值的文本结果会被保留。
+<br><b>说明：</b>大于 <code>0</code> 的任意浮点数。如果不设置，将默认使用产线初始化的该参数值 <code>0.0</code>。即不设阈值。</br>
+</td>
+<td><code>float</code></td>
+<td><code>0.0</code></td>
+</tr>
+<tr>
+<td><code>table_classification_model_name</code></td>
+<td><b>含义：</b>表格分类模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_classification_model_dir</code></td>
+<td><b>含义：</b>表格分类模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_name</code></td>
+<td><b>含义：</b>有线表格结构识别模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_dir</code></td>
+<td><b>含义：</b>有线表格结构识别模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_name</code></td>
+<td><b>含义：</b>无线表格结构识别模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_dir</code></td>
+<td><b>含义：</b>无线表格结构识别模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_name</code></td>
+<td><b>含义：</b>有线表格单元格检测模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_dir</code></td>
+<td><b>含义：</b>有线表格单元格检测模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_name</code></td>
+<td><b>含义：</b>无线表格单元格检测模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_dir</code></td>
+<td><b>含义：</b>无线表格单元格检测模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_orientation_classify_model_name</code></td>
+<td><b>含义：</b>表格方向分类模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>表格方向分类模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_name</code></td>
+<td><b>含义：</b>印章文本检测模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_dir</code></td>
+<td><b>含义：</b>印章文本检测模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>含义：</b>印章文本检测的图像边长限制。
+<br><b>说明：</b>大于 <code>0</code> 的任意整数。如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>736</code>。</br>
+</td>
+<td><code>int</code></td>
+<td><code>736</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>含义：</b>印章文本检测的图像边长限制类型。
+<br><b>说明：</b>支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>。</br>
+如果不设置，将使用产线初始化的该参数值，默认初始化为 c。
+</td>
+<td><code>str</code></td>
+<td><code>736</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>含义：</b>印章文本检测的检测像素阈值。
+<br><b>说明：</b>输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。
+大于 <code>0</code> 的任意浮点数。
+</br>如果不设置，将默认使用产线初始化的该参数值 <code>0.2</code>。
+</td>
+<td><code>float</code></td>
+<td><code>0.2</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>含义：</b>印章文本检测的检测框阈值。
+<br><b>说明：</b>检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。</br>
+大于 <code>0</code> 的任意浮点数。<br/>
+如果不设置，将默认使用产线初始化的该参数值 <code>0.6</code>。
+
+</td>
+<td><code>float</code></td>
+<td><code>0.6</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>含义：</b>印章文本检测扩张系数。
+<br><b>说明：</b>使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。</br>
+大于 <code>0</code> 的任意浮点数。如果不设置，将默认使用产线初始化的该参数值 <code>0.5</code>。
+</td>
+<td><code>float</code></td>
+<td><code>0.5</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_name</code></td>
+<td><b>含义：</b>印章文本识别模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_dir</code></td>
+<td><b>含义：</b>印章文本识别模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_batch_size</code></td>
+<td><b>含义：</b>印章文本识别模型的batch size。<br><b>说明：</b>如果不设置，将默认设置batch size为<code>1</code>。</br></td>
+<td><code>int</code></td>
+<td><code>1</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>含义：</b>文本识别阈值，得分大于该阈值的文本结果会被保留。
+<br><b>说明：</b>大于 <code>0</code> 的任意浮点数。</br>如果不设置，将默认使用产线初始化的该参数值 <code>0.0</code>。即不设阈值。
+</td>
+<td><code>float</code></td>
+<td><code>0.0</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_name</code></td>
+<td><b>含义：</b>公式识别模型的名称。<br><b>说明：</b>如果不设置，将会使用产线默认模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_dir</code></td>
+<td><b>含义：</b>公式识别模型的目录路径。<br><b>说明：</b>如果不设置，将会下载官方模型。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>formula_recognition_batch_size</code></td>
+<td><b>含义：</b>公式识别模型的batch size。<br><b>说明：</b>如果不设置，将默认设置batch size为<code>1</code>。</br></td>
+<td><code>int</code></td>
+<td><code>1</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否加载并使用文档方向分类模块。<br><b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</br></td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。<br><b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</br></td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>含义：</b>是否加载并使用文本行方向分类模块。<br><b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</br></td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>含义：</b>是否加载并使用印章文本识别子产线。<br><b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</br></td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>含义：</b>是否加载并使用表格识别子产线。<br><b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</br></td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_formula_recognition</code></td>
+<td><b>含义：</b>是否加载并使用公式识别子产线。<br><b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</br></td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>含义：</b>是否加载并使用图表解析模块。<br><b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</br></td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_region_detection</code></td>
+<td><b>含义：</b>是否加载并使用文档区域检测模块。<br><b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</br></td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>format_block_content</code></td>
+<td><b>含义：</b>是否将<code>block_content</code>中的内容格式化为Markdown格式。<br><b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</br></td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>markdown_ignore_labels</code></td>
+<td><b>含义：</b>需要在Markdown中忽略的版面标签。<br><b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>['number','footnote','header','header_image','footer','footer_image','aside_text']</code>。</br></td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。
+<br><b>说明：</b>支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+</ul>如果不设置，将默认使用产线初始化的该参数值，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。
+</br>
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>fp32</code>、<code>fp16</code>。</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+
+</tbody>
+</table>
+</details>
+<br />
+
+运行结果会被打印到终端上，默认配置的 PP-StructureV3 产线的运行结果如下：
+
+<details><summary> 👉点击展开</summary>
+<pre>
+<code>
+{'res': {'input_path': 'pp_structure_v3_demo.png', 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_general_ocr': True, 'use_seal_recognition': True, 'use_table_recognition': True, 'use_formula_recognition': True}, 'layout_det_res': {'input_path': None, 'page_index': None, 'boxes': [{'cls_id': 2, 'label': 'text', 'score': 0.9853514432907104, 'coordinate': [770.9531, 776.6814, 1122.6057, 1058.7322]}, {'cls_id': 1, 'label': 'image', 'score': 0.9848673939704895, 'coordinate': [775.7434, 202.27979, 1502.8113, 686.02136]}, {'cls_id': 2, 'label': 'text', 'score': 0.983731746673584, 'coordinate': [1152.3197, 1113.3275, 1503.3029, 1346.586]}, {'cls_id': 2, 'label': 'text', 'score': 0.9832221865653992, 'coordinate': [1152.5602, 801.431, 1503.8436, 986.3563]}, {'cls_id': 2, 'label': 'text', 'score': 0.9829439520835876, 'coordinate': [9.549545, 849.5713, 359.1173, 1058.7488]}, {'cls_id': 2, 'label': 'text', 'score': 0.9811657667160034, 'coordinate': [389.58298, 1137.2659, 740.66235, 1346.7488]}, {'cls_id': 2, 'label': 'text', 'score': 0.9775941371917725, 'coordinate': [9.1302185, 201.85, 359.0409, 339.05692]}, {'cls_id': 2, 'label': 'text', 'score': 0.9750366806983948, 'coordinate': [389.71454, 752.96924, 740.544, 889.92456]}, {'cls_id': 2, 'label': 'text', 'score': 0.9738152027130127, 'coordinate': [389.94565, 298.55988, 740.5585, 435.5124]}, {'cls_id': 2, 'label': 'text', 'score': 0.9737328290939331, 'coordinate': [771.50256, 1065.4697, 1122.2582, 1178.7324]}, {'cls_id': 2, 'label': 'text', 'score': 0.9728517532348633, 'coordinate': [1152.5154, 993.3312, 1503.2349, 1106.327]}, {'cls_id': 2, 'label': 'text', 'score': 0.9725610017776489, 'coordinate': [9.372787, 1185.823, 359.31738, 1298.7227]}, {'cls_id': 2, 'label': 'text', 'score': 0.9724331498146057, 'coordinate': [389.62848, 610.7389, 740.83234, 746.2377]}, {'cls_id': 2, 'label': 'text', 'score': 0.9720287322998047, 'coordinate': [389.29898, 897.0936, 741.41516, 1034.6616]}, {'cls_id': 2, 'label': 'text', 'score': 0.9713053703308105, 'coordinate': [10.323685, 1065.4663, 359.6786, 1178.8872]}, {'cls_id': 2, 'label': 'text', 'score': 0.9689728021621704, 'coordinate': [9.336395, 537.6609, 359.2901, 652.1881]}, {'cls_id': 2, 'label': 'text', 'score': 0.9684857130050659, 'coordinate': [10.7608185, 345.95068, 358.93616, 434.64087]}, {'cls_id': 2, 'label': 'text', 'score': 0.9681928753852844, 'coordinate': [9.674866, 658.89075, 359.56528, 770.4319]}, {'cls_id': 2, 'label': 'text', 'score': 0.9634978175163269, 'coordinate': [770.9464, 1281.1785, 1122.6522, 1346.7156]}, {'cls_id': 2, 'label': 'text', 'score': 0.96304851770401, 'coordinate': [390.0113, 201.28055, 740.1684, 291.53073]}, {'cls_id': 2, 'label': 'text', 'score': 0.962053120136261, 'coordinate': [391.21393, 1040.952, 740.5046, 1130.32]}, {'cls_id': 2, 'label': 'text', 'score': 0.9565253853797913, 'coordinate': [10.113251, 777.1482, 359.439, 842.437]}, {'cls_id': 2, 'label': 'text', 'score': 0.9497362375259399, 'coordinate': [390.31357, 537.86285, 740.47595, 603.9285]}, {'cls_id': 2, 'label': 'text', 'score': 0.9371236562728882, 'coordinate': [10.2034, 1305.9753, 359.5958, 1346.7295]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.9338151216506958, 'coordinate': [791.6062, 1200.8479, 1103.3257, 1259.9324]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.9326773285865784, 'coordinate': [408.0737, 457.37024, 718.9509, 516.63464]}, {'cls_id': 0, 'label': 'paragraph_title', 'score': 0.9274250864982605, 'coordinate': [29.448685, 456.6762, 340.99194, 515.6999]}, {'cls_id': 2, 'label': 'text', 'score': 0.8742568492889404, 'coordinate': [1154.7095, 777.3624, 1330.3086, 794.5853]}, {'cls_id': 2, 'label': 'text', 'score': 0.8442489504814148, 'coordinate': [586.49316, 160.15454, 927.468, 179.64203]}, {'cls_id': 11, 'label': 'doc_title', 'score': 0.8332607746124268, 'coordinate': [133.80017, 37.41908, 1380.8601, 124.1429]}, {'cls_id': 6, 'label': 'figure_title', 'score': 0.6770150661468506, 'coordinate': [812.1718, 705.1199, 1484.6973, 747.1692]}]}, 'overall_ocr_res': {'input_path': None, 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_textline_orientation': False}, 'dt_polys': array([[[133,  35],
+        ...,
+        [133, 131]],
+
+       ...,
+
+       [[ 13, 754],
+        ...,
+        [ 13, 777]]], dtype=int16), 'text_det_params': {'limit_side_len': 960, 'limit_type': 'max', 'thresh': 0.3, 'box_thresh': 0.6, 'unclip_ratio': 2.0}, 'text_type': 'general', 'textline_orientation_angles': array([-1, ..., -1]), 'text_rec_score_thresh': 0.0, 'rec_texts': ['助力双方交往', '搭建友谊桥梁', '本报记者', '沈小晓', '任', '彦', '黄培昭', '身着中国传统民族服装的厄立特里亚青', '厄立特里亚高等教育与研究院合作建立，开', '年依次登台表演中国民族舞、现代舞、扇子舞', '设了中国语言课程和中国文化课程，注册学', '等,曼妙的舞姿赢得现场观众阵阵掌声。这', '生2万余人次。10余年来，厄特孔院已成为', '是日前厄立特里亚高等教育与研究院孔子学', '当地民众了解中国的一扇窗口。', '院(以下简称"厄特孔院")举办"喜迎新年"中国', '黄鸣飞表示,随着来学习中文的人日益', '歌舞比赛的场景。', '增多，阿斯马拉大学教学点已难以满足教学', '中国和厄立特里亚传统友谊深厚。近年', '需要。2024年4月，由中企蜀道集团所属四', '来,在高质量共建"一带一路"框架下，中厄两', '川路桥承建的孔院教学楼项目在阿斯马拉开', '国人文交流不断深化，互利合作的民意基础', '工建设,预计今年上半年竣工,建成后将为厄', '日益深厚。', '特孔院提供全新的办学场地。', '“学好中文，我们的', '“在中国学习的经历', '未来不是梦”', '让我看到更广阔的世界”', '多年来,厄立特里亚广大赴华留学生和', '培训人员积极投身国家建设,成为助力该国', '发展的人才和厄中友好的见证者和推动者。', '在厄立特里亚全国妇女联盟工作的约翰', '娜·特韦尔德·凯莱塔就是其中一位。她曾在', '中华女子学院攻读硕士学位,研究方向是女', '性领导力与社会发展。其间，她实地走访中国', '多个地区，获得了观察中国社会发展的第一', '在厄立特里亚不久前举办的第六届中国风筝文化节上，当地小学生体验风筝制作。', '手资料。', '中国驻厄立特里亚大使馆供图', '“这是中文歌曲初级班，共有32人。学', '“不管远近都是客人，请不用客气;相约', '瓦的北红海省博物馆。', '生大部分来自首都阿斯马拉的中小学，年龄', '好了在一起,我们欢迎你"在一场中厄青', '博物馆二层陈列着一个发掘自阿杜利', '最小的仅有6岁。"尤斯拉告诉记者。', '年联谊活动上,四川路桥中方员工同当地大', '斯古城的中国古代陶制酒器,罐身上写着', '尤斯拉今年23岁，是厄立特里亚一所公立', '学生合唱《北京欢迎你》。厄立特里亚技术学', '“万”“和""禅”“山"等汉字。“这件文物证', '学校的艺术老师。她12岁开始在厄特孔院学', '院计算机科学与工程专业学生鲁夫塔·谢拉', '明,很早以前我们就通过海上丝绸之路进行', '习中文,在2017年第十届"汉语桥"世界中学生', '是其中一名演唱者,她很早便在孔院学习中', '贸易往来与文化交流。这也是厄立特里亚', '中文比赛中获得厄立特里亚赛区第一名,并和', '文，一直在为去中国留学作准备。“这句歌词', '与中国友好交往历史的有力证明。"北红海', '同伴代表厄立特里亚前往中国参加决赛,获得', '是我们两国人民友谊的生动写照。无论是投', '省博物馆研究与文献部负责人伊萨亚斯·特', '团体优胜奖。2022年起，尤斯拉开始在厄特孔', '身于厄立特里亚基础设施建设的中企员工，', '斯法兹吉说。', '院兼职教授中文歌曲,每周末两个课时。中国', '还是在中国留学的厄立特里亚学子,两国人', '厄立特里亚国家博物馆考古学和人类学', '文化博大精深,我希望我的学生们能够通过中', '民携手努力,必将推动两国关系不断向前发', '研究员菲尔蒙·特韦尔德十分喜爱中国文', '文歌曲更好地理解中国文化。"她说。', '穆卢盖塔密切关注中国在经济、科技、教', '展。"鲁夫塔说。', '化。他表示：“学习彼此的语言和文化，将帮', '“姐姐,你想去中国吗?"“非常想！我想', '育等领域的发展，“中国在科研等方面的实力', '厄立特里亚高等教育委员会主任助理萨', '助厄中两国人民更好地理解彼此，助力双方', '去看故宫、爬长城。"尤斯拉的学生中有一对', '与日俱增。在中国学习的经历让我看到更广', '马瑞表示：“每年我们都会组织学生到中国访', '交往,搭建友谊桥梁。"', '能歌善舞的姐妹,姐姐露娅今年15岁，妹妹', '阔的世界，从中受益匪浅。', '问学习，目前有超过5000名厄立特里亚学生', '厄立特里亚国家博物馆馆长塔吉丁·努', '莉娅14岁，两人都已在厄特孔院学习多年，', '23岁的莉迪亚·埃斯蒂法诺斯已在厄特', '在中国留学。学习中国的教育经验,有助于', '里达姆·优素福曾多次访问中国，对中华文明', '中文说得格外流利。', '孔院学习3年，在中国书法、中国画等方面表', '提升厄立特里亚的教育水平。”', '的传承与创新、现代化博物馆的建设与发展', '露娅对记者说：“这些年来,怀着对中文', '现十分优秀，在2024年厄立特里亚赛区的', '“共同向世界展示非', '印象深刻。“中国博物馆不仅有许多保存完好', '和中国文化的热爱,我们姐妹俩始终相互鼓', '“汉语桥"比赛中获得一等奖。莉迪亚说：“学', '的文物,还充分运用先进科技手段进行展示，', '励,一起学习。我们的中文一天比一天好,还', '习中国书法让我的内心变得安宁和纯粹。我', '洲和亚洲的灿烂文明”', '帮助人们更好理解中华文明。"塔吉丁说，厄', '学会了中文歌和中国舞。我们一定要到中国', '也喜欢中国的服饰,希望未来能去中国学习，', '立特里亚与中国都拥有悠久的文明,始终相', '去。学好中文,我们的未来不是梦!"', '把中国不同民族元素融入服装设计中，创作', '从阿斯马拉出发,沿着蜿蜒曲折的盘山', '互理解、相互尊重。我希望未来与中国同行', '据厄特孔院中方院长黄鸣飞介绍,这所', '出更多精美作品，也把厄特文化分享给更多', '公路一路向东寻找丝路印迹。驱车两个小', '加强合作,共同向世界展示非洲和亚洲的灿', '孔院成立于2013年3月，由贵州财经大学和', '的中国朋友。”', '时,记者来到位于厄立特里亚港口城市马萨', '烂文明。”', '谈起在中国求学的经历,约翰娜记忆犹', '新：“中国的发展在当今世界是独一无二的。', '沿着中国特色社会主义道路坚定前行，中国', '创造了发展奇迹,这一切都离不开中国共产党', '的领导。中国的发展经验值得许多国家学习', '借鉴，”', '正在西南大学学习的厄立特里亚博士生', '穆卢盖塔·泽穆伊对中国怀有深厚感情。8', '年前，在北京师范大学获得硕士学位后，穆卢', '盖塔在社交媒体上写下这样一段话：“这是我', '人生的重要一步，自此我拥有了一双坚固的', '鞋子.赋予我穿越荆棘的力量。”', '“鲜花曾告诉我你怎样走过，大地知道你', '心中的每一个角落"厄立特里亚阿斯马拉', '大学综合楼二层，一阵优美的歌声在走廊里回', '响。循着熟悉的旋律轻轻推开一间教室的门，', '学生们正跟着老师学唱中文歌曲《同一首歌》。', '这是厄特孔院阿斯马拉大学教学点的一', '节中文歌曲课。为了让学生们更好地理解歌', '词大意，老师尤斯拉·穆罕默德萨尔·侯赛因逐', '字翻译和解释歌词。随着伴奏声响起，学生们', '边唱边随着节拍摇动身体，现场气氛热烈。'], 'rec_scores': array([0.99972075, ..., 0.96241361]), 'rec_polys': array([[[133,  35],
+        ...,
+        [133, 131]],
+
+       ...,
+
+       [[ 13, 754],
+        ...,
+        [ 13, 777]]], dtype=int16), 'rec_boxes': array([[133, ..., 131],
+       ...,
+       [ 13, ..., 777]], dtype=int16)}}}
+</code></pre></details>
+
+运行结果参数说明可以参考[2.2 Python脚本方式集成](#22-python)中的结果解释。
+
+<b>注：</b>由于产线的默认模型较大，推理速度可能较慢，您可以参考第一节的模型列表，替换推理速度更快的模型。
+
+### 2.2 Python脚本方式集成
+
+命令行方式是为了快速体验查看效果，一般来说，在项目中，往往需要通过代码集成，您可以通过几行代码即可完成产线的快速推理，推理代码如下：
+
+```python
+from paddleocr import PPStructureV3
+
+pipeline = PPStructureV3()
+# pipeline = PPStructureV3(lang="en") # 将 lang 参数设置为使用英文文本识别模型。对于其他支持的语言，请参阅第5节：附录部分。默认配置为中英文模型。
+# pipeline = PPStructureV3(use_doc_orientation_classify=True) # 通过 use_doc_orientation_classify 指定是否使用文档方向分类模型
+# pipeline = PPStructureV3(use_doc_unwarping=True) # 通过 use_doc_unwarping 指定是否使用文本图像矫正模块
+# pipeline = PPStructureV3(use_textline_orientation=True) # 通过 use_textline_orientation 指定是否使用文本行方向分类模型
+# pipeline = PPStructureV3(device="gpu") # 通过 device 指定模型推理时使用 GPU
+output = pipeline.predict("./pp_structure_v3_demo.png")
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_json(save_path="output") ## 保存当前图像的结构化json结果
+    res.save_to_markdown(save_path="output") ## 保存当前图像的markdown格式的结果
+    res.save_to_word(save_path="output") ## 保存当前图像的Word格式的结果
+```
+
+上述代码默认使用本地飞桨推理引擎。默认情况下，各模块会根据默认模型名称选择合适的本地飞桨推理引擎：仅支持动态图的模型使用 `paddle_dynamic`；同时支持静态图和动态图的模型优先使用 `paddle_static`。如需运行，请先参考[飞桨框架安装说明](../paddlepaddle_installation.md)安装 PaddlePaddle。
+
+如果选择 `transformers` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 Transformers 环境配置，然后执行如下代码：
+
+```python
+from paddleocr import PPStructureV3
+
+pipeline = PPStructureV3(
+    engine="transformers",
+)
+# pipeline = PPStructureV3(lang="en") # 将 lang 参数设置为使用英文文本识别模型。对于其他支持的语言，请参阅第5节：附录部分。默认配置为中英文模型。
+# pipeline = PPStructureV3(use_doc_orientation_classify=True) # 通过 use_doc_orientation_classify 指定是否使用文档方向分类模型
+# pipeline = PPStructureV3(use_doc_unwarping=True) # 通过 use_doc_unwarping 指定是否使用文本图像矫正模块
+# pipeline = PPStructureV3(use_textline_orientation=True) # 通过 use_textline_orientation 指定是否使用文本行方向分类模型
+# pipeline = PPStructureV3(device="gpu") # 通过 device 指定模型推理时使用 GPU
+output = pipeline.predict("./pp_structure_v3_demo.png")
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_json(save_path="output") ## 保存当前图像的结构化json结果
+    res.save_to_markdown(save_path="output") ## 保存当前图像的markdown格式的结果
+```
+
+如果选择 `onnxruntime` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 ONNX Runtime 环境配置，然后执行如下代码：
+
+```python
+from paddleocr import PPStructureV3
+
+# 部分模型尚在支持中，推理时需关闭公式识别功能，请使用以下代码：
+pipeline = PPStructureV3(
+    engine="onnxruntime",
+    use_formula_recognition=False,
+)
+# pipeline = PPStructureV3(lang="en") # 将 lang 参数设置为使用英文文本识别模型。对于其他支持的语言，请参阅第5节：附录部分。默认配置为中英文模型。
+# pipeline = PPStructureV3(use_doc_orientation_classify=True) # 通过 use_doc_orientation_classify 指定是否使用文档方向分类模型
+# pipeline = PPStructureV3(use_doc_unwarping=True) # 通过 use_doc_unwarping 指定是否使用文本图像矫正模块
+# pipeline = PPStructureV3(use_textline_orientation=True) # 通过 use_textline_orientation 指定是否使用文本行方向分类模型
+# pipeline = PPStructureV3(device="gpu") # 通过 device 指定模型推理时使用 GPU
+output = pipeline.predict("./pp_structure_v3_demo.png")
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_json(save_path="output") ## 保存当前图像的结构化json结果
+    res.save_to_markdown(save_path="output") ## 保存当前图像的markdown格式的结果
+```
+
+如果是 PDF 文件，会将 PDF 的每一页单独处理，每一页的 Markdown 文件也会对应单独的结果。如果希望整个 PDF 文件转换为 Markdown 文件，建议使用以下的方式运行：
+
+```python
+from pathlib import Path
+from paddleocr import PPStructureV3
+
+input_file = "./your_pdf_file.pdf"
+output_path = Path("./output")
+
+pipeline = PPStructureV3()
+output = pipeline.predict(input=input_file)
+
+markdown_list = []
+markdown_images = []
+
+for res in output:
+    md_info = res.markdown
+    markdown_list.append(md_info)
+    markdown_images.append(md_info.get("markdown_images", {}))
+
+markdown_texts = pipeline.concatenate_markdown_pages(markdown_list)
+
+mkd_file_path = output_path / f"{Path(input_file).stem}.md"
+mkd_file_path.parent.mkdir(parents=True, exist_ok=True)
+
+with open(mkd_file_path, "w", encoding="utf-8") as f:
+    f.write(markdown_texts)
+
+for item in markdown_images:
+    if item:
+        for path, image in item.items():
+            file_path = output_path / path
+            file_path.parent.mkdir(parents=True, exist_ok=True)
+            image.save(file_path)
+```
+
+**注：**
+
+- PP-StructureV3 产线使用的默认文本识别模型为 **中英文识别模型**，对于纯英文的识别能力有限，对于全英文场景，您可以设置`text_recognition_model_name`参数为 `en_PP-OCRv4_mobile_rec` 等英文识别模型以取得更好的识别效果。对应其他语言场景，也可以参考前文的模型列表，选择对应的语言识别模型进行替换。
+
+- 在示例代码中，`use_doc_orientation_classify`、`use_doc_unwarping`、`use_textline_orientation` 参数默认均设置为 `False`，分别表示关闭文档方向分类、文本图像矫正、文本行方向分类功能，如果需要使用这些功能，可以手动设置为 `True`。
+<!-- Luxorion-12 修改：表格参数描述-->
+在上述 Python 脚本中，执行了如下几个步骤：
+
+<details><summary>（1）实例化产线对象，具体参数说明如下：</summary>
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>含义：</b>版面区域检测的模型名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>含义：</b>版面区域检测模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>版面模型得分阈值。
+<br><b>说明：</b>
+<ul>
+<li><b>float</b>：<code>0-1</code> 之间的任意浮点数；</li>
+<li><b>dict</b>： <code>{0:0.1}</code> key为类别ID，value为该类别的阈值；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>0.5</code>。</li>
+</ul>
+</br>
+</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>版面检测是否使用后处理NMS。
+<br><b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>版面区域检测模型检测框的扩张系数。
+<br><b>说明：</b>
+<ul>
+<li><b>float</b>：任意大于 <code>0</code>  浮点数；</li>
+<li><b>Tuple[float,float]</b>：在横纵两个方向各自的扩张系数；</li>
+<li><b>dict</b>，dict的key为<b>int</b>类型，代表<code>cls_id</code>, value为<b>tuple</b>类型，如<code>{0: (1.1, 2.0)}</code>，表示将模型输出的第0类别检测框中心不变，宽度扩张1.1倍，高度扩张2.0倍；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>1.0</code>。</li>
+</ul>
+</br>
+</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>版面区域检测的重叠框过滤方式。
+<br><b>说明：</b>
+<ul>
+<li><b>str</b>：<code>large</code>，<code>small</code>，<code>union</code>，分别表示重叠框过滤时选择保留大框，小框还是同时保留；</li>
+<li><b>dict</b>： dict的key为<b>int</b>类型，代表<code>cls_id</code>，value为<b>str</b>类型，如<code>{0: "large", 2: "small"}</code>，表示对第0类别检测框使用large模式，对第2类别检测框使用small模式；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>large</code>。</li>
+</ul>
+</br>
+</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>chart_recognition_model_name</code></td>
+<td><b>含义：</b>图表解析的模型名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>chart_recognition_model_dir</code></td>
+<td><b>含义：</b>图表解析模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>chart_recognition_batch_size</code></td>
+<td><b>含义：</b>图表解析模型的batch size。
+<br><b>说明：</b>如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</br></td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>region_detection_model_name</code></td>
+<td><b>含义：</b>文档图像版面子模块检测的模型名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>region_detection_model_dir</code></td>
+<td><b>含义：</b>文档图像版面子模块检测模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>含义：</b>文档方向分类模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>文档方向分类模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>含义：</b>文本图像矫正模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>含义：</b>文本检测模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>含义：</b>文本检测模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>含义：</b>文本检测的图像边长限制。
+<br><b>说明：</b>
+<ul>
+<li><b>int</b>：大于 <code>0</code> 的任意整数；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>960</code>。</li>
+</ul>
+</br>
+</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>含义：</b>文本检测的图像边长限制类型。
+<br><b>说明：</b>
+<ul>
+<li><b>str</b>：支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>max</code>。</li>
+</ul>
+</br>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>含义：</b>检测像素阈值，输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。
+<br><b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；<br/>
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.3</code>。</li></li></ul>
+</br>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>含义：</b>检测框阈值，检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。
+<br><b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；<br/>
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.6</code>。</li></li></ul>
+</br>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>含义：</b>文本检测扩张系数，使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。
+<br><b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；<br/>
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>2.0</code>。</li></li></ul>
+</br>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_name</code></td>
+<td><b>含义：</b>文本行方向模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_model_dir</code></td>
+<td><b>含义：</b>文本行方向模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>textline_orientation_batch_size</code></td>
+<td><b>含义：</b>文本行方向模型的batch size。
+<br><b>说明：</b>如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</br></td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>含义：</b>文本识别模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>含义：</b>文本识别模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>含义：</b>文本识别模型的batch size。
+<br><b>说明：</b>如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</br></td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>含义：</b>文本识别阈值，得分大于该阈值的文本结果会被保留。
+<br><b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；<br/>
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.0</code>，即不设阈值。</li></li></ul>
+</br>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_classification_model_name</code></td>
+<td><b>含义：</b>表格分类模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_classification_model_dir</code></td>
+<td><b>含义：</b>表格分类模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_name</code></td>
+<td><b>含义：</b>有线表格结构识别模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_dir</code></td>
+<td><b>含义：</b>有线表格结构识别模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_name</code></td>
+<td><b>含义：</b>无线表格结构识别模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_dir</code></td>
+<td><b>含义：</b>无线表格结构识别模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_name</code></td>
+<td><b>含义：</b>有线表格单元格检测模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_dir</code></td>
+<td><b>含义：</b>有线表格单元格检测模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_name</code></td>
+<td><b>含义：</b>无线表格单元格检测模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_dir</code></td>
+<td><b>含义：</b>无线表格单元格检测模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_orientation_classify_model_name</code></td>
+<td><b>含义：</b>表格方向分类模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>表格方向分类模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_name</code></td>
+<td><b>含义：</b>印章文本检测模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_dir</code></td>
+<td><b>含义：</b>印章文本检测模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>含义：</b>印章文本检测的图像边长限制。
+<br><b>说明：</b>
+<ul>
+<li><b>int</b>：大于 <code>0</code> 的任意整数；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>736</code>。</li>
+</ul>
+</br>
+</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>含义：</b>印章文本检测的图像边长限制类型。
+<br><b>说明：</b>
+<ul>
+<li><b>str</b>：支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>min</code>。</li>
+</ul>
+</br>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>含义：</b>印章文本检测的像素阈值。
+<br><b>说明：</b>
+<ul>
+<li><b>float</b>：大于 <code>0</code> 的任意浮点数；<br/>
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.2</code>。</li></li></ul>
+</br>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>含义：</b>印章文本检测的检测框阈值。
+<br><b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；<br/>
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.6</code>。</li></li></ul>
+</br>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>含义：</b>印章文本检测扩张系数，使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。
+<br><b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；<br/>
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.5</code>。</li></li></ul>
+</br>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_name</code></td>
+<td><b>含义：</b>印章文本识别模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_model_dir</code></td>
+<td><b>含义：</b>印章文本识别模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_recognition_batch_size</code></td>
+<td><b>含义：</b>印章文本识别模型的batch size。
+<br><b>说明：</b>如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</br></td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>含义：</b>印章文本识别阈值，得分大于该阈值的文本结果会被保留。
+<br><b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；<br/>
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.0</code>，即不设阈值。</li></li></ul></br>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_name</code></td>
+<td><b>含义：</b>公式识别模型的名称。
+<br><b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_dir</code></td>
+<td><b>含义：</b>公式识别模型的目录路径。
+<br><b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_batch_size</code></td>
+<td><b>含义：</b>公式识别模型的batch size。
+<br><b>说明：</b>如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</br></td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否加载并使用文档方向分类模块。
+<br><b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。
+<br><b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>含义：</b>是否加载并使用文本行方向分类模块。
+<br><b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>含义：</b>是否加载并使用印章文本识别子产线。
+<br><b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>含义：</b>是否加载并使用表格识别子产线。
+<br><b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_formula_recognition</code></td>
+<td><b>含义：</b>是否加载并使用公式识别子产线。
+<br><b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>含义：</b>是否加载并使用图表解析模块。
+<br><b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_region_detection</code></td>
+<td><b>含义：</b>是否加载并使用文档区域检测模块。
+<br><b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>format_block_content</code></td>
+<td>是否将<code>block_content</code>中的内容格式化为Markdown格式。如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>False</code>。</td>
+<td><code>bool|None</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>markdown_ignore_labels</code></td>
+<td>需要在Markdown中忽略的版面标签。如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>['number','footnote','header','header_image','footer','footer_image','aside_text']</code>。</td>
+<td><code>list|None</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。
+<br><b>说明：</b>支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+<li><b>None</b>：如<code>None</code>表示使用 CPU 设备进行推理。</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+</tbody>
+</table>
+
+</details>
+
+<details><summary>（2）调用 PP-StructureV3 产线对象的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。另外，产线还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。以下是 <code>predict()</code> 方法的参数及其说明：</summary>
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。
+<br><b>说明：</b>
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/demo_paper.png">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]。</code></li>
+</ul>
+</br>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否在推理时使用文档方向分类模块。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否在推理时使用文本图像矫正模块。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_textline_orientation</code></td>
+<td><b>含义：</b>是否在推理时使用文本行方向分类模块。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>含义：</b>是否在推理时使用印章文本识别子产线。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_table_recognition</code></td>
+<td><b>含义：</b>是否在推理时使用表格识别子产线。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_formula_recognition</code></td>
+<td><b>含义：</b>是否在推理时使用公式识别子产线。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>含义：</b>是否在推理时使用图表解析模块。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_region_detection</code></td>
+<td><b>含义：</b>是否在推理时使用文档区域检测模块。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>format_block_content</code></td>
+<td><b>含义：</b>是否将<code>block_content</code>中的内容格式化为Markdown格式。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>bool|None</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_wired_table_cells_trans_to_html</code></td>
+<td><b>含义：</b>是否启用有线表单元格检测结果直转HTML，启用则直接基于有线表单元格检测结果的几何关系构建HTML。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_wireless_table_cells_trans_to_html</code></td>
+<td><b>含义：</b>是否启用无线表单元格检测结果直转HTML，启用则直接基于无线表单元格检测结果的几何关系构建HTML。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_table_orientation_classify</code></td>
+<td><b>含义：</b>是否启用表格使用表格方向分类，启用时当图像中的表格存在90/180/270度旋转时，能够将方向校正并正确完成表格识别。</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_ocr_results_with_table_cells</code></td>
+<td><b>含义：</b>是否启用单元格切分OCR，启用时会基于单元格预测结果对OCR检测结果进行切分和重识别，避免出现文字缺失情况。</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_e2e_wired_table_rec_model</code></td>
+<td><b>含义：</b>是否启用有线表端到端表格识别模式，启用则不使用单元格检测模型，只使用表格结构识别模型。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_e2e_wireless_table_rec_model</code></td>
+<td><b>含义：</b>是否启用无线表端到端表格识别模式，启用则不使用单元格检测模型，只使用表格结构识别模型。</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>markdown_ignore_labels</code></td>
+<td><b>含义：</b>需要在Markdown中忽略的版面标签。
+<br><b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</br></td>
+<td><code>list|None</code></td>
+<td></td>
+</tr>
+</table>
+</details>
+
+<details><summary>（3）对预测结果进行处理：每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为<code>json</code>文件的操作:</summary>
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化。</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效。</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效。</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致。</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效。</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效。</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将中间各个模块的可视化图像保存在png格式的图像</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径。</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>save_to_word()</code></td>
+<td>将版面解析结果保存为Word (.docx) 格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>save_to_markdown()</code></td>
+<td>将图像或者PDF文件中的每一页分别保存为markdown格式的文件。</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径。</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>save_to_html()</code></td>
+<td>将文件中的表格保存为html格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径。</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>save_to_xlsx()</code></td>
+<td>将文件中的表格保存为xlsx格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径。</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>concatenate_markdown_pages()</code></td>
+<td>将多页Markdown内容拼接为单一文档</td>
+<td><code>markdown_list</code></td>
+<td><code>list</code></td>
+<td>包含每一页Markdown数据的列表。</td>
+<td>返回处理后的Markdown文本和图像列表。</td>
+</tr>
+</table>
+
+<ul>
+   <li>调用<code>print()</code> 方法会将结果打印到终端，打印到终端的内容解释如下：</li>
+      <ol start="1" type = "1"> 
+        <li><code>input_path</code>: <code>(str)</code>  待预测图像或者PDF的输入路径</li>
+        <li><code>page_index</code>: <code>(Union[int, None])</code> 如果输入是PDF文件，则表示当前是PDF的第几页，否则为  <code>None</code></li>
+        <li><code>page_count</code>: <code>(Union[int, None])</code> 如果输入是PDF文件，表示当前是PDF的总页数，否则为 <code>None</code></li>
+        <li><code>width</code>: <code>(int)</code> 原始输入图像的宽度</li>
+        <li><code>height</code>: <code>(int)</code> 原始输入图像的高度</li>
+        <li><code>model_settings</code>: <code>(Dict[str, bool])</code>  配置产线所需的模型参数</li>
+            <ol start = "1" type = "a" > 
+                <li><code>use_doc_preprocessor</code>: <code>(bool)</code>  控制是否启用文档预处理子产线</li>
+                <li><code>use_seal_recognition</code>: <code>(bool)</code> 控制是否启用印章文本识别子产线</li>
+                <li><code>use_table_recognition</code>: <code>(bool)</code> 控制是否启用表格识别子产线</li>
+                <li><code>use_formula_recognition</code>: <code>(bool)</code> 控制是否启用公式识别子产线</li>
+                <li><code>format_block_content</code>: <code>(bool)</code> 控制是否将 <code>block_content</code> 中的内容格式化为Markdown格式</li>
+                <li><code>markdown_ignore_labels</code>: <code>(List[str])</code> 需要在Markdown中忽略的版面标签</li>
+            </ul>
+        </li>
+        <li><code>doc_preprocessor_res</code>: <code>(Dict[str, Union[List[float], str]])</code> 文档预处理结果dict，仅当<code>use_doc_preprocessor=True</code>时存在</li>
+            <ol start = "1" type = "a" > 
+                <li><code>input_path</code>: <code>(str)</code>  文档预处理子产线接受的图像路径，当输入为 <code>numpy.ndarray</code> 时，保存为<code> None</code> ，此处为<code> None</code> </li>
+                <li><code>page_index</code>: <code>None</code> ，此处的输入为<code>numpy.ndarray</code> ，所以值为<code> None</code> </li>
+                <li><code>model_settings</code>: <code>(Dict[str, bool])</code>  文档预处理子产线的模型配置参数</li>
+                    <ul>
+                        <li><code>use_doc_orientation_classify</code>: <code>(bool)</code>  控制是否启用文档图像方向分类子模块</li>
+                        <li><code>use_doc_unwarping</code>: <code>(bool)</code>  控制是否启用文本图像扭曲矫正子模块</li>
+                    </ul>
+                <li><code>angle</code>: <code>(int)</code>  文档图像方向分类子模块的预测结果，启用时返回实际角度值</li>
+            </ol>
+        <li><code>parsing_res_list</code>: <code>(List[Dict])</code> 解析结果的列表，每个元素为一个字典，列表顺序为解析后的阅读顺序。</li>
+            <ol start = "1" type = "a" > 
+                <li><code>block_bbox</code>: <code>(np.ndarray)</code> 版面区域的边界框。</li>
+                <li><code>block_label</code>: <code>(str)</code> 版面区域的标签，例如<code>text</code>, <code>table</code>等。</li>
+                <li><code>block_content</code>: <code>(str)</code> 内容为版面区域内的内容。</li>
+                <li><code>block_id</code>: <code>(int)</code> 版面区域的索引，用于显示版面排序结果。</li>
+                <li><code>block_order</code> <code>(int)</code> 版面区域的顺序，用于显示版面阅读顺序,对于非排序部分，默认值为 <code> None</code> 。</li>
+            </ol>
+            <li><code>overall_ocr_res</code>: <code>(Dict[str, Union[List[str], List[float], numpy.ndarray]])</code> 全局 OCR 结果的dict</li>
+                <ol start = "1" type = "a" > 
+                    <li><code>input_path</code>: <code>(Union[str, None])</code> 图像OCR子产线接受的图像路径，当输入为<code>numpy.ndarray</code>时，保存为<code> None</code> </li>
+                    <li><code>page_index</code>: <code> None</code> ，此处的输入为<code>numpy.ndarray</code>，所以值为<code> None</code> </li>
+                    <li><code>model_settings</code>: <code>(Dict)</code> OCR子产线的模型配置参数</li>
+                    <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> 文本检测的多边形框列表。每个检测框由4个顶点坐标构成的numpy数组表示，数组shape为(4, 2)，数据类型为int16</li>
+                    <li><code>dt_scores</code>: <code>(List[float])</code> 文本检测框的置信度列表</li>
+                    <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> 文本检测模块的配置参数</li>
+                        <ul>
+                            <li><code>limit_side_len</code>: <code>(int)</code> 图像预处理时的边长限制值</li>
+                            <li><code>limit_type</code>: <code>(str)</code> 边长限制的处理方式</li>
+                            <li><code>thresh</code>: <code>(float)</code> 文本像素分类的置信度阈值</li>
+                            <li><code>box_thresh</code>: <code>(float)</code> 文本检测框的置信度阈值</li>
+                            <li><code>unclip_ratio</code>: <code>(float)</code> 文本检测框的膨胀系数</li>
+                            <li><code>text_type</code>: <code>(str)</code> 文本检测的类型，当前固定为"general"</li>
+                        </ul>
+                    <li><code>text_type</code>: <code>(str)</code> 文本检测的类型，当前固定为"general"</li>
+                    <li><code>textline_orientation_angles</code>: <code>(List[int])</code> 文本行方向分类的预测结果。启用时返回实际角度值（如[0,0,1]</li>
+                    <li><code>text_rec_score_thresh</code>: <code>(float)</code> 文本识别结果的过滤阈值</li>
+                    <li><code>rec_texts</code>: <code>(List[str])</code> 文本识别结果列表，仅包含置信度超过<code>text_rec_score_thresh</code>的文本</li>
+                    <li><code>rec_scores</code>: <code>(List[float])</code> 文本识别的置信度列表，已按<code>text_rec_score_thresh</code>过滤</li>
+                    <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> 经过置信度过滤的文本检测框列表，格式同<code>dt_polys</code></li>
+                </ol>
+            <li><code>formula_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> 公式识别结果列表，每个元素为一个dict</li>
+                <ol start = "1" type = "a" > 
+                    <li><code>rec_formula</code>: <code>(str)</code> 公式识别结果</li>
+                    <li><code>rec_polys</code>: <code>(numpy.ndarray)</code> 公式检测框，shape为(4, 2)，dtype为int16</li>
+                    <li><code>formula_region_id</code>: <code>(int)</code> 公式所在的区域编号</li>
+                </ol>
+            <li><code>seal_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> 印章文本识别结果列表，每个元素为一个dict</li>
+                <ol start = "1" type = "a" > 
+                    <li><code>input_path</code>: <code>(str)</code> 印章图像的输入路径</li>
+                    <li><code>page_index</code>: <code> None</code> ，此处的输入为<code>numpy.ndarray</code>，所以值为<code> None</code> </li>
+                    <li><code>model_settings</code>: <code>(Dict)</code> 印章文本识别子产线的模型配置参数</li>
+                    <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> 印章检测框列表，格式同<code>dt_polys</code></li>
+                    <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> 印章检测模块的配置参数, 具体参数含义同上</li>
+                    <li><code>text_type</code>: <code>(str)</code> 印章检测的类型，当前固定为"seal"</li>
+                    <li><code>text_rec_score_thresh</code>: <code>(float)</code> 印章文本识别结果的过滤阈值</li>
+                    <li><code>rec_texts</code>: <code>(List[str])</code> 印章文本识别结果列表，仅包含置信度超过<code>text_rec_score_thresh</code>的文本</li>
+                    <li><code>rec_scores</code>: <code>(List[float])</code> 印章文本识别的置信度列表，已按<code>dt_polys</code>过滤</li>
+                    <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> 经过置信度过滤的印章检测框列表，格式同<code>dt_polys</code></li>
+                    <li><code>rec_boxes</code>: <code>(numpy.ndarray)</code> 检测框的矩形边界框数组，shape为(n, 4)，dtype为int16。每一行表示一个矩形</li>
+                </ol>
+            <li><code>table_res_list</code>: <code>(List[Dict[str, Union[numpy.ndarray, List[float], str]]])</code> 表格识别结果列表，每个元素为一个dict</li>
+                <ol start = "1" type = "a" > 
+                    <li><code>cell_box_list</code>: <code>(List[numpy.ndarray])</code> 表格单元格的边界框列表</li>
+                    <li><code>pred_html</code>: <code>(str)</code> 表格的HTML格式字符串</li>
+                    <li><code>table_ocr_pred</code>: <code>(dict)</code> 表格的OCR识别结果</li>
+                    <ul>
+                        <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> 单元格的检测框列表</li>
+                        <li><code>rec_texts</code>: <code>(List[str])</code> 单元格的识别结果</li>
+                        <li><code>rec_scores</code>: <code>(List[float])</code>单元格的识别置信度</li>
+                        <li><code>rec_boxes</code>: <code>(numpy.ndarray)</code> 检测框的矩形边界框数组，shape为(n, 4)，dtype为int16。每一行表示一个矩形</li>
+                    </ul>
+                </ol>
+            </ol>
+    </li>
+    <li>调用<code>save_to_json()</code> 方法会将上述内容保存到指定的 <code>save_path</code> 中，如果指定为目录，则保存的路径为<code>save_path/{your_img_basename}_res.json</code>，如果指定为文件，则直接保存到该文件中。由于 json 文件不支持保存numpy数组，因此会将其中的 <code>numpy.array</code> 类型转换为列表形式。</li>
+    <li>调用<code>save_to_img()</code> 方法会将可视化结果保存到指定的 <code>save_path</code> 中，如果指定为目录，则会将版面区域检测可视化图像、全局OCR可视化图像、版面阅读顺序可视化图像等内容保存，如果指定为文件，则直接保存到该文件中。(产线通常包含较多结果图片，不建议直接指定为具体的文件路径，否则多张图会被覆盖，仅保留最后一张图)</li>
+    <li>调用<code>save_to_markdown()</code> 方法会将转化后的 Markdown 文件保存到指定的 <code>save_path</code> 中，保存的文件路径为<code>save_path/{your_img_basename}.md</code>，如果输入是 PDF 文件，建议直接指定目录，否责多个 markdown 文件会被覆盖。
+    调用 <code>concatenate_markdown_pages()</code> 方法将 <code>PP-StructureV3 pipeline</code> 输出的多页Markdown内容<code>markdown_list</code>合并为单个完整文档，并返回合并后的Markdown内容。</li>
+</ul>
+此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>json</code></td>
+<td>获取预测的 <code>json</code> 格式的结果</td>
+</tr>
+<tr>
+<td rowspan="2"><code>img</code></td>
+<td rowspan="2">获取格式为 <code>dict</code> 的可视化图像</td>
+</tr>
+<tr>
+</tr>
+<tr>
+<td rowspan="3"><code>markdown</code></td>
+<td rowspan="3">获取格式为 <code>dict</code> 的 markdown 结果</td>
+</tr>
+<tr>
+</tr>
+<tr>
+</tr>
+</tbody>
+</table>
+<ul>
+    <li><code>json</code> 属性获取的预测结果为dict类型的数据，相关内容与调用 <code>save_to_json()</code> 方法保存的内容一致。</li>
+    <li><code>img</code> 属性返回的预测结果是一个dict类型的数据。其中，键分别为 <code>layout_det_res</code>、<code>overall_ocr_res</code>、<code>text_paragraphs_ocr_res</code>、<code>formula_res_region1</code>、<code>table_cell_img</code> 和 <code>seal_res_region1</code>，对应的值是 <code>Image.Image</code> 对象：分别用于显示版面区域检测、OCR、OCR文本段落、公式、表格和印章结果的可视化图像。如果没有使用可选模块，则dict中只包含 <code>layout_det_res</code>。</li>
+    <li><code>markdown</code> 属性返回的预测结果是一个dict类型的数据。其中，键分别为 <code>markdown_texts</code> 、 <code>markdown_images</code>和<code>page_continuation_flags</code>，对应的值分别是 markdown 文本，在 Markdown 中显示的图像（<code>Image.Image</code> 对象）和用于标识当前页面第一个元素是否为段开始以及最后一个元素是否为段结束的bool元组。</li>
+</ul>
+</details>
+
+
+## 3. 开发集成/部署
+
+如果产线可以达到您对产线推理速度和精度的要求，您可以直接进行开发集成/部署。
+
+若您需要将产线直接应用在您的Python项目中，可以参考 [2.2 Python脚本方式](#22-python)中的示例代码。
+
+此外，PaddleOCR 也提供了其他两种部署方式，详细说明如下：
+
+🚀 高性能推理：在实际生产环境中，许多应用对部署策略的性能指标（尤其是响应速度）有着较严苛的标准，以确保系统的高效运行与用户体验的流畅性。为此，PaddleOCR 提供高性能推理功能，旨在对模型推理及前后处理进行深度性能优化，实现端到端流程的显著提速，详细的高性能推理流程请参考[高性能推理](../inference_deployment/local_inference/high_performance_inference.md)。
+
+☁️ 服务化部署：服务化部署是实际生产环境中常见的一种部署形式。通过将推理功能封装为服务，客户端可以通过网络请求来访问这些服务，以获取推理结果。详细的产线服务化部署流程请参考[服务化部署](../inference_deployment/serving/serving.md)。
+
+以下是基础服务化部署的API参考与多语言服务调用示例：
+
+<details><summary>API参考</summary>
+<p>对于服务提供的主要操作：</p>
+<ul>
+<li>HTTP请求方法为POST。</li>
+<li>请求体和响应体均为JSON数据（JSON对象）。</li>
+<li>当请求处理成功时，响应状态码为<code>200</code>，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。固定为<code>0</code>。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。固定为<code>"Success"</code>。</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>操作结果。</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>当请求处理未成功时，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。与响应状态码相同。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。</td>
+</tr>
+</tbody>
+</table>
+<p>服务提供的主要操作如下：</p>
+<ul>
+<li><b><code>infer</code></b></li>
+</ul>
+<p>进行版面解析。</p>
+<p><code>POST /layout-parsing</code></p>
+<ul>
+<li>请求体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>服务器可访问的图像文件（含 TIFF，多页时按页处理）或 PDF 文件的 URL，或上述类型文件内容的 Base64 编码结果。默认对于超过 10 页的 PDF 或多页 TIFF，仅处理前 10 页。<br /> 要解除页数限制，请在产线配置文件中添加以下配置：
+<pre><code>Serving:
+  extra:
+    max_num_input_imgs: null
+</code></pre>
+</td>
+<td>是</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code>｜<code>null</code></td>
+<td>文件类型。<code>0</code> 表示 PDF 文件，<code>1</code> 表示图像文件（含 TIFF）。若请求体无此属性，则将根据URL推断文件类型。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_doc_orientation_classify</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_doc_unwarping</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useTextlineOrientation</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_textline_orientation</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useSealRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_seal_recognition</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useTableRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_table_recognition</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useFormulaRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_formula_recognition</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useChartRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_chart_recognition</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useRegionDetection</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_region_detection</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>formatBlockContent</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>format_block_content</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutThreshold</code></td>
+<td><code>number</code> | <code>object</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>layout_threshold</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutNms</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>layout_nms</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutUnclipRatio</code></td>
+<td><code>number</code> | <code>array</code> | <code>object</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>layout_unclip_ratio</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutMergeBboxesMode</code></td>
+<td><code>string</code> | <code>object</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>layout_merge_bboxes_mode</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_det_limit_side_len</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_det_limit_type</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_det_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_det_box_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_det_unclip_ratio</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_rec_score_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>seal_det_limit_side_len</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>seal_det_limit_type</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>seal_det_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>seal_det_box_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>seal_det_unclip_ratio</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>seal_rec_score_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useWiredTableCellsTransToHtml</code></td>
+<td><code>boolean</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_wired_table_cells_trans_to_html</code> 参数相关说明。</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useWirelessTableCellsTransToHtml</code></td>
+<td><code>boolean</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_wireless_table_cells_trans_to_html</code> 参数相关说明。</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useTableOrientationClassify</code></td>
+<td><code>boolean</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_table_orientation_classify</code> 参数相关说明。</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useOcrResultsWithTableCells</code></td>
+<td><code>boolean</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_ocr_results_with_table_cells</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useE2eWiredTableRecModel</code></td>
+<td><code>boolean</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_e2e_wired_table_rec_model</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useE2eWirelessTableRecModel</code></td>
+<td><code>boolean</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_e2e_wireless_table_rec_model</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>markdownIgnoreLabels</code></td>
+<td><code>array</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>markdown_ignore_labels</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>prettifyMarkdown</code></td>
+<td><code>boolean</code></td>
+<td>是否输出美化后的 Markdown 文本。默认为 <code>true</code>。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>showFormulaNumber</code></td>
+<td><code>boolean</code></td>
+<td>输出的 Markdown 文本中是否包含公式编号。默认为 <code>false</code>。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>returnMarkdownImages</code></td>
+<td><code>boolean</code></td>
+<td>是否在响应中返回 Markdown 中引用的图片。默认为 <code>true</code>；设为 <code>false</code> 时 <code>markdown.images</code> 为 <code>null</code> 或不出现，且服务端跳过图片编码 / URL 上传。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>outputFormats</code></td>
+<td><code>array</code> | <code>null</code></td>
+<td>可选。附加导出格式列表，默认不返回。当前仅支持 <code>"docx"</code>。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>是否返回可视化结果图以及处理过程中的中间图像等。
+<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>传入 <code>true</code>：返回图像。</li>
+<li>传入 <code>false</code>：不返回图像。</li>
+<li>若请求体中未提供该参数或传入 <code>null</code>：遵循产线配置文件<code>Serving.visualize</code> 的设置。</li>
+</ul>
+<br/>例如，在产线配置文件中添加如下字段：<br/>
+<pre><code>Serving:
+  visualize: False
+</code></pre>
+将默认不返回图像，通过请求体中的<code>visualize</code>参数可以覆盖默认行为。如果请求体和配置文件中均未设置（或请求体传入<code>null</code>、配置文件中未设置），则默认返回图像。
+</td>
+<td>否</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>请求处理成功时，响应体的<code>result</code>具有如下属性：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>layoutParsingResults</code></td>
+<td><code>array</code></td>
+<td>版面解析结果。数组长度为1（对于图像输入）或实际处理的文档页数（对于PDF输入）。对于PDF输入，数组中的每个元素依次表示PDF文件中实际处理的每一页的结果。</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>输入数据信息。</td>
+</tr>
+</tbody>
+</table>
+<p>下表中涉及图像等二进制文件的字段（如 <code>outputImages</code>、<code>inputImage</code>、<code>markdown.images</code>、<code>exports</code>）默认以 Base64 字符串内联返回；当服务端开启 URL 返回模式时，相应字段的值变为预签名 URL，字段类型保持不变。配置方式参见 <a href="../inference_deployment/serving/serving.md">服务化部署</a>「以 URL 形式返回二进制内容」一节。</p>
+<p><code>layoutParsingResults</code>中的每个元素为一个<code>object</code>，具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>产线对象的 <code>predict</code> 方法生成结果的 JSON 表示中 <code>res</code> 字段的简化版本，其中去除了 <code>input_path</code> 和 <code>page_index</code> 字段。</td>
+</tr>
+<tr>
+<td><code>markdown</code></td>
+<td><code>object</code></td>
+<td>Markdown结果。</td>
+</tr>
+<tr>
+<td><code>outputImages</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>参见产线预测结果的 <code>img</code> 属性说明。图像为 JPEG 格式，默认使用 Base64 编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+<tr>
+<td><code>inputImage</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>输入图像。图像为 JPEG 格式，默认使用 Base64 编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+<tr>
+<td><code>exports</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>可选的附加导出结果。仅当请求中包含 <code>outputFormats</code> 时出现，例如 <code>{"docx": {"content": "..."}}</code>，其中 <code>content</code> 默认为文件内容的 Base64 编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+</tbody>
+</table>
+<p><code>markdown</code>为一个<code>object</code>，具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>text</code></td>
+<td><code>string</code></td>
+<td>Markdown文本。</td>
+</tr>
+<tr>
+<td><code>images</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Markdown 图片相对路径与对应图像的键值对。图像默认为 Base64 编码字符串；启用 URL 返回模式时为预签名 URL。当请求中 <code>returnMarkdownImages</code> 为 <code>false</code> 时该字段为 <code>null</code> 或不出现。</td>
+</tr>
+<tr>
+<td><code>isStart</code></td>
+<td><code>boolean</code></td>
+<td>当前页面第一个元素是否为段开始。</td>
+</tr>
+<tr>
+<td><code>isEnd</code></td>
+<td><code>boolean</code></td>
+<td>当前页面最后一个元素是否为段结束。</td>
+</tr>
+</tbody>
+</table>
+</details>
+<details><summary>多语言调用服务示例</summary>
+<details>
+<summary>Python</summary>
+
+<pre><code class="language-python">
+import base64
+import requests
+import pathlib
+
+API_URL = "http://localhost:8080/layout-parsing" # 服务URL
+
+image_path = "./demo.jpg"
+
+# 对本地图像进行Base64编码
+with open(image_path, "rb") as file:
+    image_bytes = file.read()
+    image_data = base64.b64encode(image_bytes).decode("ascii")
+
+payload = {
+    "file": image_data, # Base64编码的文件内容或者文件URL
+    "fileType": 1, # 文件类型，1表示图像文件
+}
+
+# 调用API
+response = requests.post(API_URL, json=payload)
+
+# 处理接口返回数据
+assert response.status_code == 200
+result = response.json()["result"]
+for i, res in enumerate(result["layoutParsingResults"]):
+    print(res["prunedResult"])
+    md_dir = pathlib.Path(f"markdown_{i}")
+    md_dir.mkdir(exist_ok=True)
+    (md_dir / "doc.md").write_text(res["markdown"]["text"])
+    for img_path, img in res["markdown"]["images"].items():
+        img_path = md_dir / img_path
+        img_path.parent.mkdir(parents=True, exist_ok=True)
+        img_path.write_bytes(base64.b64decode(img))
+    print(f"Markdown document saved at {md_dir / 'doc.md'}")
+    for img_name, img in res["outputImages"].items():
+        img_path = f"{img_name}_{i}.jpg"
+        with open(img_path, "wb") as f:
+            f.write(base64.b64decode(img))
+        print(f"Output image saved at {img_path}")
+</code></pre></details>
+
+<details><summary>C++</summary>
+
+<pre><code class="language-cpp">#include &lt;iostream&gt;
+#include &lt;fstream&gt;
+#include &lt;vector&gt;
+#include &lt;string&gt;
+#include "cpp-httplib/httplib.h" // https://github.com/Huiyicc/cpp-httplib
+#include "nlohmann/json.hpp" // https://github.com/nlohmann/json
+#include "base64.hpp" // https://github.com/tobiaslocker/base64
+
+int main() {
+    httplib::Client client("localhost", 8080);
+
+    const std::string filePath = "./demo.jpg";
+
+    std::ifstream file(filePath, std::ios::binary | std::ios::ate);
+    if (!file) {
+        std::cerr << "Error opening file: " << filePath << std::endl;
+        return 1;
+    }
+
+    std::streamsize size = file.tellg();
+    file.seekg(0, std::ios::beg);
+    std::vector<char> buffer(size);
+    if (!file.read(buffer.data(), size)) {
+        std::cerr << "Error reading file." << std::endl;
+        return 1;
+    }
+
+    std::string bufferStr(buffer.data(), static_cast<size_t>(size));
+    std::string encodedFile = base64::to_base64(bufferStr);
+
+    nlohmann::json jsonObj;
+    jsonObj["file"] = encodedFile;
+    jsonObj["fileType"] = 1;
+
+    auto response = client.Post("/layout-parsing", jsonObj.dump(), "application/json");
+
+    if (response && response->status == 200) {
+        nlohmann::json jsonResponse = nlohmann::json::parse(response->body);
+        auto result = jsonResponse["result"];
+
+        if (!result.is_object() || !result.contains("layoutParsingResults")) {
+            std::cerr << "Unexpected response format." << std::endl;
+            return 1;
+        }
+
+        const auto& results = result["layoutParsingResults"];
+        for (size_t i = 0; i < results.size(); ++i) {
+            const auto& res = results[i];
+
+            if (res.contains("prunedResult")) {
+                std::cout << "Layout result [" << i << "]: " << res["prunedResult"].dump() << std::endl;
+            }
+
+            if (res.contains("outputImages") && res["outputImages"].is_object()) {
+                for (auto& [imgName, imgBase64] : res["outputImages"].items()) {
+                    std::string outputPath = imgName + "_" + std::to_string(i) + ".jpg";
+                    std::string decodedImage = base64::from_base64(imgBase64.get<std::string>());
+
+                    std::ofstream outFile(outputPath, std::ios::binary);
+                    if (outFile.is_open()) {
+                        outFile.write(decodedImage.c_str(), decodedImage.size());
+                        outFile.close();
+                        std::cout << "Saved image: " << outputPath << std::endl;
+                    } else {
+                        std::cerr << "Failed to save image: " << outputPath << std::endl;
+                    }
+                }
+            }
+        }
+    } else {
+        std::cerr << "Request failed." << std::endl;
+        if (response) {
+            std::cerr << "HTTP status: " << response->status << std::endl;
+            std::cerr << "Response body: " << response->body << std::endl;
+        }
+        return 1;
+    }
+
+    return 0;
+}
+</code></pre></details>
+
+<details><summary>Java</summary>
+
+<pre><code class="language-java">import okhttp3.*;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.util.Base64;
+
+public class Main {
+    public static void main(String[] args) throws IOException {
+        String API_URL = "http://localhost:8080/layout-parsing";
+        String imagePath = "./demo.jpg";
+
+        File file = new File(imagePath);
+        byte[] fileContent = java.nio.file.Files.readAllBytes(file.toPath());
+        String base64Image = Base64.getEncoder().encodeToString(fileContent);
+
+        ObjectMapper objectMapper = new ObjectMapper();
+        ObjectNode payload = objectMapper.createObjectNode();
+        payload.put("file", base64Image);
+        payload.put("fileType", 1);
+
+        OkHttpClient client = new OkHttpClient();
+        MediaType JSON = MediaType.get("application/json; charset=utf-8");
+
+        RequestBody body = RequestBody.create(JSON, payload.toString());
+
+        Request request = new Request.Builder()
+                .url(API_URL)
+                .post(body)
+                .build();
+
+        try (Response response = client.newCall(request).execute()) {
+            if (response.isSuccessful()) {
+                String responseBody = response.body().string();
+                JsonNode root = objectMapper.readTree(responseBody);
+                JsonNode result = root.get("result");
+
+                JsonNode layoutParsingResults = result.get("layoutParsingResults");
+                for (int i = 0; i < layoutParsingResults.size(); i++) {
+                    JsonNode item = layoutParsingResults.get(i);
+                    int finalI = i;
+                    JsonNode prunedResult = item.get("prunedResult");
+                    System.out.println("Pruned Result [" + i + "]: " + prunedResult.toString());
+
+                    JsonNode outputImages = item.get("outputImages");
+                    outputImages.fieldNames().forEachRemaining(imgName -> {
+                        try {
+                            String imgBase64 = outputImages.get(imgName).asText();
+                            byte[] imgBytes = Base64.getDecoder().decode(imgBase64);
+                            String imgPath = imgName + "_" + finalI + ".jpg";
+                            try (FileOutputStream fos = new FileOutputStream(imgPath)) {
+                                fos.write(imgBytes);
+                                System.out.println("Saved image: " + imgPath);
+                            }
+                        } catch (IOException e) {
+                            System.err.println("Failed to save image: " + e.getMessage());
+                        }
+                    });
+                }
+            } else {
+                System.err.println("Request failed with HTTP code: " + response.code());
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Go</summary>
+
+<pre><code class="language-go">package main
+
+import (
+    "bytes"
+    "encoding/base64"
+    "encoding/json"
+    "fmt"
+    "io/ioutil"
+    "net/http"
+    "os"
+    "path/filepath"
+)
+
+func main() {
+    API_URL := "http://localhost:8080/layout-parsing"
+    filePath := "./demo.jpg"
+
+    fileBytes, err := ioutil.ReadFile(filePath)
+    if err != nil {
+        fmt.Printf("Error reading file: %v\n", err)
+        return
+    }
+    fileData := base64.StdEncoding.EncodeToString(fileBytes)
+
+    payload := map[string]interface{}{
+        "file":     fileData,
+        "fileType": 1,
+    }
+    payloadBytes, err := json.Marshal(payload)
+    if err != nil {
+        fmt.Printf("Error marshaling payload: %v\n", err)
+        return
+    }
+
+    client := &http.Client{}
+    req, err := http.NewRequest("POST", API_URL, bytes.NewBuffer(payloadBytes))
+    if err != nil {
+        fmt.Printf("Error creating request: %v\n", err)
+        return
+    }
+    req.Header.Set("Content-Type", "application/json")
+
+    res, err := client.Do(req)
+    if err != nil {
+        fmt.Printf("Error sending request: %v\n", err)
+        return
+    }
+    defer res.Body.Close()
+
+    if res.StatusCode != http.StatusOK {
+        fmt.Printf("Unexpected status code: %d\n", res.StatusCode)
+        return
+    }
+
+    body, err := ioutil.ReadAll(res.Body)
+    if err != nil {
+        fmt.Printf("Error reading response: %v\n", err)
+        return
+    }
+
+    type Markdown struct {
+        Text   string            `json:"text"`
+        Images map[string]string `json:"images"`
+    }
+
+    type LayoutResult struct {
+        PrunedResult map[string]interface{} `json:"prunedResult"`
+        Markdown     Markdown               `json:"markdown"`
+        OutputImages map[string]string      `json:"outputImages"`
+        InputImage   *string                `json:"inputImage"`
+    }
+
+    type Response struct {
+        Result struct {
+            LayoutParsingResults []LayoutResult `json:"layoutParsingResults"`
+            DataInfo             interface{}    `json:"dataInfo"`
+        } `json:"result"`
+    }
+
+    var respData Response
+    if err := json.Unmarshal(body, &respData); err != nil {
+        fmt.Printf("Error parsing response: %v\n", err)
+        return
+    }
+
+    for i, res := range respData.Result.LayoutParsingResults {
+        fmt.Printf("Result %d - prunedResult: %+v\n", i, res.PrunedResult)
+
+        mdDir := fmt.Sprintf("markdown_%d", i)
+        os.MkdirAll(mdDir, 0755)
+        mdFile := filepath.Join(mdDir, "doc.md")
+        if err := os.WriteFile(mdFile, []byte(res.Markdown.Text), 0644); err != nil {
+            fmt.Printf("Error writing markdown file: %v\n", err)
+        } else {
+            fmt.Printf("Markdown document saved at %s\n", mdFile)
+        }
+
+        for path, imgBase64 := range res.Markdown.Images {
+            fullPath := filepath.Join(mdDir, path)
+            os.MkdirAll(filepath.Dir(fullPath), 0755)
+            imgBytes, err := base64.StdEncoding.DecodeString(imgBase64)
+            if err != nil {
+                fmt.Printf("Error decoding markdown image: %v\n", err)
+                continue
+            }
+            if err := os.WriteFile(fullPath, imgBytes, 0644); err != nil {
+                fmt.Printf("Error saving markdown image: %v\n", err)
+            }
+        }
+
+        for name, imgBase64 := range res.OutputImages {
+            imgBytes, err := base64.StdEncoding.DecodeString(imgBase64)
+            if err != nil {
+                fmt.Printf("Error decoding output image %s: %v\n", name, err)
+                continue
+            }
+            filename := fmt.Sprintf("%s_%d.jpg", name, i)
+            if err := os.WriteFile(filename, imgBytes, 0644); err != nil {
+                fmt.Printf("Error saving output image %s: %v\n", filename, err)
+            } else {
+                fmt.Printf("Output image saved at %s\n", filename)
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>C#</summary>
+
+<pre><code class="language-csharp">using System;
+using System.IO;
+using System.Net.Http;
+using System.Text;
+using System.Threading.Tasks;
+using Newtonsoft.Json.Linq;
+
+class Program
+{
+    static readonly string API_URL = "http://localhost:8080/layout-parsing";
+    static readonly string inputFilePath = "./demo.jpg";
+
+    static async Task Main(string[] args)
+    {
+        var httpClient = new HttpClient();
+
+        byte[] fileBytes = File.ReadAllBytes(inputFilePath);
+        string fileData = Convert.ToBase64String(fileBytes);
+
+        var payload = new JObject
+        {
+            { "file", fileData },
+            { "fileType", 1 }
+        };
+        var content = new StringContent(payload.ToString(), Encoding.UTF8, "application/json");
+
+        HttpResponseMessage response = await httpClient.PostAsync(API_URL, content);
+        response.EnsureSuccessStatusCode();
+
+        string responseBody = await response.Content.ReadAsStringAsync();
+        JObject jsonResponse = JObject.Parse(responseBody);
+
+        JArray layoutParsingResults = (JArray)jsonResponse["result"]["layoutParsingResults"];
+        for (int i = 0; i < layoutParsingResults.Count; i++)
+        {
+            var res = layoutParsingResults[i];
+            Console.WriteLine($"[{i}] prunedResult:\n{res["prunedResult"]}");
+
+            JObject outputImages = res["outputImages"] as JObject;
+            if (outputImages != null)
+            {
+                foreach (var img in outputImages)
+                {
+                    string imgName = img.Key;
+                    string base64Img = img.Value?.ToString();
+                    if (!string.IsNullOrEmpty(base64Img))
+                    {
+                        string imgPath = $"{imgName}_{i}.jpg";
+                        byte[] imageBytes = Convert.FromBase64String(base64Img);
+                        File.WriteAllBytes(imgPath, imageBytes);
+                        Console.WriteLine($"Output image saved at {imgPath}");
+                    }
+                }
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Node.js</summary>
+
+<pre><code class="language-js">const axios = require('axios');
+const fs = require('fs');
+const path = require('path');
+
+const API_URL = 'http://localhost:8080/layout-parsing';
+const imagePath = './demo.jpg';
+const fileType = 1;
+
+function encodeImageToBase64(filePath) {
+  const bitmap = fs.readFileSync(filePath);
+  return Buffer.from(bitmap).toString('base64');
+}
+
+const payload = {
+  file: encodeImageToBase64(imagePath),
+  fileType: fileType
+};
+
+axios.post(API_URL, payload)
+  .then(response => {
+    const results = response.data.result.layoutParsingResults;
+    results.forEach((res, index) => {
+      console.log(`\n[${index}] prunedResult:`);
+      console.log(res.prunedResult);
+
+      const outputImages = res.outputImages;
+      if (outputImages) {
+        Object.entries(outputImages).forEach(([imgName, base64Img]) => {
+          const imgPath = `${imgName}_${index}.jpg`;
+          fs.writeFileSync(imgPath, Buffer.from(base64Img, 'base64'));
+          console.log(`Output image saved at ${imgPath}`);
+        });
+      } else {
+        console.log(`[${index}] No outputImages.`);
+      }
+    });
+  })
+  .catch(error => {
+    console.error('Error during API request:', error.message || error);
+  });
+</code></pre></details>
+
+<details><summary>PHP</summary>
+
+<pre><code class="language-php">&lt;?php
+
+$API_URL = "http://localhost:8080/layout-parsing";
+$image_path = "./demo.jpg";
+
+$image_data = base64_encode(file_get_contents($image_path));
+$payload = array("file" => $image_data, "fileType" => 1);
+
+$ch = curl_init($API_URL);
+curl_setopt($ch, CURLOPT_POST, true);
+curl_setopt($ch, CURLOPT_POSTFIELDS, json_encode($payload));
+curl_setopt($ch, CURLOPT_HTTPHEADER, array('Content-Type: application/json'));
+curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+$response = curl_exec($ch);
+curl_close($ch);
+
+$result = json_decode($response, true)["result"]["layoutParsingResults"];
+
+foreach ($result as $i => $item) {
+    echo "[$i] prunedResult:\n";
+    print_r($item["prunedResult"]);
+
+    if (!empty($item["outputImages"])) {
+        foreach ($item["outputImages"] as $img_name => $img_base64) {
+            $output_image_path = "{$img_name}_{$i}.jpg";
+            file_put_contents($output_image_path, base64_decode($img_base64));
+            echo "Output image saved at $output_image_path\n";
+        }
+    } else {
+        echo "No outputImages found for item $i\n";
+    }
+}
+?&gt;
+</code></pre></details>
+</details>
+<br/>
+
+
+## 4. 二次开发
+如果 PP-StructureV3 产线提供的默认模型权重在您的场景中，精度或速度不满意，您可以尝试利用<b>您自己拥有的特定领域或应用场景的数据</b>对现有模型进行进一步的<b>微调</b>，以提升 PP-StructureV3 产线的在您的场景中的识别效果。
+
+### 4.1 模型微调
+由于 PP-StructureV3 产线包含若干模块，模型产线的效果不及预期可能来自于其中任何一个模块。您可以对提取效果差的 case 进行分析，通过可视化图像，确定是哪个模块存在问题，并参考以下表格中对应的微调教程链接进行模型微调。
+
+
+<table>
+<thead>
+<tr>
+<th>情形</th>
+<th>微调模块</th>
+<th>微调参考链接</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>版面区域检测不准，如印章、表格未检出等</td>
+<td>版面区域检测模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/layout_detection.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>表格结构识别不准</td>
+<td>表格结构识别模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/table_structure_recognition.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>公式识别不准</td>
+<td>公式识别模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/formula_recognition.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>印章文本存在漏检</td>
+<td>印章文本检测模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/seal_text_detection.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>文本存在漏检</td>
+<td>文本检测模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/text_detection.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>文本内容都不准</td>
+<td>文本识别模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/text_recognition.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>垂直或者旋转文本行矫正不准</td>
+<td>文本行方向分类模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/textline_orientation_classification.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>整图旋转矫正不准</td>
+<td>文档图像方向分类模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>图像扭曲矫正不准</td>
+<td>文本图像矫正模块</td>
+<td>暂不支持微调</td>
+</tr>
+</tbody>
+</table>
+
+### 4.2 模型应用
+当您使用私有数据集完成微调训练后，可获得本地模型权重文件，然后可以通过自定义产线配置文件的方式，使用微调后的模型权重。
+
+1. 获取产线配置文件
+
+可调用 PaddleOCR 中 PPStructureV3 产线对象的 `export_paddlex_config_to_yaml` 方法，将当前产线配置导出为 YAML 文件：
+
+```Python
+from paddleocr import PPStructureV3
+
+pipeline = PPStructureV3()
+pipeline.export_paddlex_config_to_yaml("PP-StructureV3.yaml")
+```
+
+2. 修改配置文件
+
+在得到默认的产线配置文件后，将微调后模型权重的本地路径替换至产线配置文件中的对应位置即可。例如
+
+```yaml
+......
+SubModules:
+  LayoutDetection:
+    module_name: layout_detection
+    model_name: PP-DocLayout_plus-L
+    model_dir: null # 替换为微调后的版面区域检测模型权重路径
+......
+SubPipelines:
+  GeneralOCR:
+    pipeline_name: OCR
+    text_type: general
+    use_doc_preprocessor: False
+    use_textline_orientation: False
+    SubModules:
+      TextDetection:
+        module_name: text_detection
+        model_name: PP-OCRv5_server_det
+        model_dir: null # 替换为微调后的文本测模型权重路径
+        limit_side_len: 960
+        limit_type: max
+        max_side_limit: 4000
+        thresh: 0.3
+        box_thresh: 0.6
+        unclip_ratio: 1.5
+
+      TextRecognition:
+        module_name: text_recognition
+        model_name: PP-OCRv5_server_rec
+        model_dir: null # 替换为微调后的文本识别模型权重路径
+        batch_size: 1
+        score_thresh: 0
+......
+```
+
+在产线配置文件中，不仅包含 PaddleOCR CLI 和 Python API 支持的参数，还可进行更多高级配置，具体信息可在 [PaddleX模型产线使用概览](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/pipeline_develop_guide.html) 中找到对应的产线使用教程，参考其中的详细说明，根据需求调整各项配置。
+
+3. 在 CLI 中加载产线配置文件
+
+在修改完成配置文件后，通过命令行的 `--paddlex_config` 参数指定修改后的产线配置文件的路径，PaddleOCR 会读取其中的内容作为产线配置。示例如下：
+
+```bash
+paddleocr pp_structurev3 --paddlex_config PP-StructureV3.yaml ...
+```
+
+4. 在 Python API 中加载产线配置文件
+
+初始化产线对象时，可通过 `paddlex_config` 参数传入 PaddleX 产线配置文件路径或配置dict，PaddleOCR 会读取其中的内容作为产线配置。示例如下：
+
+```python
+from paddleocr import PPStructureV3
+
+pipeline = PPStructureV3(paddlex_config="PP-StructureV3.yaml")
+```
+
+## 5. 附录
+
+<details><summary><b>支持语言</b></summary>
+
+<table border="1" cellspacing="0" cellpadding="4">
+  <thead>
+    <tr>
+      <th><code>lang</code></th>
+      <th>语言名称</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr><td><code>abq</code></td><td>阿布哈兹文</td></tr>
+    <tr><td><code>af</code></td><td>南非荷兰文</td></tr>
+    <tr><td><code>ang</code></td><td>古英文</td></tr>
+    <tr><td><code>ar</code></td><td>阿拉伯文</td></tr>
+    <tr><td><code>ava</code></td><td>阿瓦尔文</td></tr>
+    <tr><td><code>az</code></td><td>阿塞拜疆文</td></tr>
+    <tr><td><code>be</code></td><td>白俄罗斯文</td></tr>
+    <tr><td><code>bg</code></td><td>保加利亚文</td></tr>
+    <tr><td><code>bgc</code></td><td>哈里亚纳文</td></tr>
+    <tr><td><code>bh</code></td><td>比哈尔文</td></tr>
+    <tr><td><code>bho</code></td><td>博杰普尔文</td></tr>
+    <tr><td><code>bs</code></td><td>波斯尼亚文</td></tr>
+    <tr><td><code>ch</code></td><td>简体中文</td></tr>
+    <tr><td><code>che</code></td><td>车臣文</td></tr>
+    <tr><td><code>chinese_cht</code></td><td>繁体中文</td></tr>
+    <tr><td><code>cs</code></td><td>捷克文</td></tr>
+    <tr><td><code>cy</code></td><td>威尔士文</td></tr>
+    <tr><td><code>da</code></td><td>丹麦文</td></tr>
+    <tr><td><code>dar</code></td><td>达尔格瓦文</td></tr>
+    <tr><td><code>de</code> or <code>german</code></td><td>德文</td></tr>
+    <tr><td><code>en</code></td><td>英文</td></tr>
+    <tr><td><code>es</code></td><td>西班牙文</td></tr>
+    <tr><td><code>et</code></td><td>爱沙尼亚文</td></tr>
+    <tr><td><code>fa</code></td><td>波斯文</td></tr>
+    <tr><td><code>fr</code> or <code>french</code></td><td>法文</td></tr>
+    <tr><td><code>ga</code></td><td>爱尔兰文</td></tr>
+    <tr><td><code>gom</code></td><td>孔卡尼文</td></tr>
+    <tr><td><code>hi</code></td><td>印地文</td></tr>
+    <tr><td><code>hr</code></td><td>克罗地亚文</td></tr>
+    <tr><td><code>hu</code></td><td>匈牙利文</td></tr>
+    <tr><td><code>id</code></td><td>印尼文</td></tr>
+    <tr><td><code>inh</code></td><td>印古什文</td></tr>
+    <tr><td><code>is</code></td><td>冰岛文</td></tr>
+    <tr><td><code>it</code></td><td>意大利文</td></tr>
+    <tr><td><code>japan</code></td><td>日文</td></tr>
+    <tr><td><code>ka</code></td><td>格鲁吉亚文</td></tr>
+    <tr><td><code>kbd</code></td><td>卡巴尔达文</td></tr>
+    <tr><td><code>korean</code></td><td>韩文</td></tr>
+    <tr><td><code>ku</code></td><td>库尔德文</td></tr>
+    <tr><td><code>la</code></td><td>拉丁文</td></tr>
+    <tr><td><code>lbe</code></td><td>拉克文</td></tr>
+    <tr><td><code>lez</code></td><td>列兹金文</td></tr>
+    <tr><td><code>lt</code></td><td>立陶宛文</td></tr>
+    <tr><td><code>lv</code></td><td>拉脱维亚文</td></tr>
+    <tr><td><code>mah</code></td><td>马加希文</td></tr>
+    <tr><td><code>mai</code></td><td>迈蒂利文</td></tr>
+    <tr><td><code>mi</code></td><td>毛利文</td></tr>
+    <tr><td><code>mn</code></td><td>蒙古文</td></tr>
+    <tr><td><code>mr</code></td><td>马拉地文</td></tr>
+    <tr><td><code>ms</code></td><td>马来文</td></tr>
+    <tr><td><code>mt</code></td><td>马耳他文</td></tr>
+    <tr><td><code>ne</code></td><td>尼泊尔文</td></tr>
+    <tr><td><code>new</code></td><td>尼瓦尔文</td></tr>
+    <tr><td><code>nl</code></td><td>荷兰文</td></tr>
+    <tr><td><code>no</code></td><td>挪威文</td></tr>
+    <tr><td><code>oc</code></td><td>奥克文</td></tr>
+    <tr><td><code>pi</code></td><td>巴利文</td></tr>
+    <tr><td><code>pl</code></td><td>波兰文</td></tr>
+    <tr><td><code>pt</code></td><td>葡萄牙文</td></tr>
+    <tr><td><code>ro</code></td><td>罗马尼亚文</td></tr>
+    <tr><td><code>rs_cyrillic</code></td><td>塞尔维亚语西里尔字母</td></tr>
+    <tr><td><code>rs_latin</code></td><td>塞尔维亚语拉丁字母</td></tr>
+    <tr><td><code>ru</code></td><td>俄文</td></tr>
+    <tr><td><code>sa</code></td><td>梵文</td></tr>
+    <tr><td><code>sck</code></td><td>萨达里文</td></tr>
+    <tr><td><code>sk</code></td><td>斯洛伐克文</td></tr>
+    <tr><td><code>sl</code></td><td>斯洛文尼亚文</td></tr>
+    <tr><td><code>sq</code></td><td>阿尔巴尼亚文</td></tr>
+    <tr><td><code>sv</code></td><td>瑞典文</td></tr>
+    <tr><td><code>sw</code></td><td>斯瓦希里文</td></tr>
+    <tr><td><code>tab</code></td><td>塔巴萨兰文</td></tr>
+    <tr><td><code>ta</code></td><td>泰米尔文</td></tr>
+    <tr><td><code>te</code></td><td>泰卢固文</td></tr>
+    <tr><td><code>tl</code></td><td>塔加洛文</td></tr>
+    <tr><td><code>tr</code></td><td>土耳其文</td></tr>
+    <tr><td><code>ug</code></td><td>维吾尔文</td></tr>
+    <tr><td><code>uk</code></td><td>乌克兰文</td></tr>
+    <tr><td><code>ur</code></td><td>乌尔都文</td></tr>
+    <tr><td><code>uz</code></td><td>乌兹别克文</td></tr>
+    <tr><td><code>vi</code></td><td>越南文</td></tr>
+  </tbody>
+</table>
+
+</details>
+
+<details><summary><b>OCR模型版本与支持语言的对应关系</b></summary>
+
+<table border="1" cellspacing="0" cellpadding="4">
+  <thead>
+    <tr>
+      <th><code>ocr_version</code></th>
+      <th>Supported <code>lang</code></th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td><code>PP-OCRv5</code></td>
+      <td><code>ch</code>, <code>en</code>, <code>fr</code>, <code>de</code>, <code>japan</code>, <code>korean</code>, <code>chinese_cht</code>, <code>af</code>, <code>it</code>, <code>es</code>, <code>bs</code>, <code>pt</code>, <code>cs</code>, <code>cy</code>, <code>da</code>, <code>et</code>, <code>ga</code>, <code>hr</code>, <code>hu</code>, <code>rslatin</code>, <code>id</code>, <code>oc</code>, <code>is</code>, <code>lt</code>, <code>mi</code>, <code>ms</code>, <code>nl</code>, <code>no</code>, <code>pl</code>, <code>sk</code>, <code>sl</code>, <code>sq</code>, <code>sv</code>, <code>sw</code>, <code>tl</code>, <code>tr</code>, <code>uz</code>, <code>la</code>, <code>ru</code>, <code>be</code>, <code>uk</code></td>
+    </tr>
+    <tr>
+      <td><code>PP-OCRv4</code></td>
+      <td><code>ch</code>, <code>en</code></td>
+    </tr>
+    <tr>
+      <td><code>PP-OCRv3</code></td>
+      <td>
+        <code>abq</code>, <code>af</code>, <code>ady</code>, <code>ang</code>, <code>ar</code>, <code>ava</code>, <code>az</code>, <code>be</code>,
+        <code>bg</code>, <code>bgc</code>, <code>bh</code>, <code>bho</code>, <code>bs</code>, <code>ch</code>, <code>che</code>,
+        <code>chinese_cht</code>, <code>cs</code>, <code>cy</code>, <code>da</code>, <code>dar</code>, <code>de</code>, <code>german</code>,
+        <code>en</code>, <code>es</code>, <code>et</code>, <code>fa</code>, <code>fr</code>, <code>french</code>, <code>ga</code>, <code>gom</code>,
+        <code>hi</code>, <code>hr</code>, <code>hu</code>, <code>id</code>, <code>inh</code>, <code>is</code>, <code>it</code>, <code>japan</code>,
+        <code>ka</code>, <code>kbd</code>, <code>korean</code>, <code>ku</code>, <code>la</code>, <code>lbe</code>, <code>lez</code>, <code>lt</code>,
+        <code>lv</code>, <code>mah</code>, <code>mai</code>, <code>mi</code>, <code>mn</code>, <code>mr</code>, <code>ms</code>, <code>mt</code>,
+        <code>ne</code>, <code>new</code>, <code>nl</code>, <code>no</code>, <code>oc</code>, <code>pi</code>, <code>pl</code>, <code>pt</code>,
+        <code>ro</code>, <code>rs_cyrillic</code>, <code>rs_latin</code>, <code>ru</code>, <code>sa</code>, <code>sck</code>, <code>sk</code>,
+        <code>sl</code>, <code>sq</code>, <code>sv</code>, <code>sw</code>, <code>ta</code>, <code>tab</code>, <code>te</code>, <code>tl</code>,
+        <code>tr</code>, <code>ug</code>, <code>uk</code>, <code>ur</code>, <code>uz</code>, <code>vi</code>
+      </td>
+    </tr>
+  </tbody>
+</table>
+
+</details>
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-AMD-GPU.en.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-AMD-GPU.en.md
new file mode 100644
index 0000000..95dbdb6
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-AMD-GPU.en.md
@@ -0,0 +1,284 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL AMD GPU Usage Tutorial
+
+> INFO:
+> Unless otherwise specified, the term "PaddleOCR-VL" in this tutorial refers to the PaddleOCR-VL model series (e.g., PaddleOCR-VL-1.6). References specific to the PaddleOCR-VL v1 version will be explicitly noted.
+
+This tutorial is a guide for using PaddleOCR-VL on AMD GPU, covering the complete workflow from environment preparation to service deployment.
+
+PaddleOCR-VL has been verified for accuracy and speed on the AMD MI300X. However, due to hardware diversity, compatibility with other AMD GPUs has not yet been confirmed. We welcome the community to test on different hardware setups and share your results.
+
+## Workflow Guide for This Hardware
+
+Use this guide for the workflows below.
+
+| Goal | Support on this hardware | Read this section |
+| --- | --- | --- |
+| Local direct inference | Supported | Read Section 1. Local Runtime Environment Preparation and Section 2. Quick Start. |
+| Client + VLM inference service | Supported | Complete local direct inference first, then read Section 3. Using VLM Inference Services. |
+| Full API service | Supported with Docker Compose deployment | Read Section 4.1 first, then continue with the Section 4.2 client invocation section and the Section 4.3 pipeline configuration section. |
+| Model fine-tuning | Supported | Read Section 5. Model Fine-Tuning. |
+
+If you only need to confirm which inference methods are available on this hardware, refer to the [PaddleOCR-VL Inference Method and Hardware Support Matrix](./PaddleOCR-VL.en.md#inference-device-support-for-paddleocr-vl) in the main guide.
+
+## 1. Local Runtime Environment Preparation
+
+**Local Runtime Environment Setup Methods Supported on This Hardware**
+
+| Local runtime environment setup method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Supported with steps in this guide | Continue with Section 1.1. |
+| Manually install the inference engine and PaddleOCR | Supported with steps in this guide | Continue with Section 1.2. |
+
+This step mainly introduces how to set up the local runtime environment for PaddleOCR-VL. There are two methods available; choose either one:
+
+- Method 1: Use the official Docker image.
+
+- Method 2: Manually install the inference engine and PaddleOCR.
+
+**We strongly recommend using the Docker image to minimize potential environment-related issues.**
+
+### 1.1 Method 1: Using Docker Image
+
+We recommend using the official Docker image (requires Docker version >= 19.03):
+
+```shell
+docker run -it \
+  --user root \
+  --device /dev:/dev \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-amd-gpu \
+  /bin/bash
+# Call PaddleOCR CLI or Python API in the container
+```
+
+If you wish to start the service in an environment without internet access, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-amd-gpu` (image size approximately 15 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-amd-gpu-offline` (image size approximately 17 GB).
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use an image corresponding to a specific PaddleOCR version, you can replace `latest` in the tag with the desired version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.4-amd-gpu-offline`
+
+
+### 1.2 Method 2: Manually Install the Inference Engine and PaddleOCR
+
+If you cannot use Docker, you can also manually install the inference engine and PaddleOCR. This guide documents Python 3.9–3.13 as the verified range.
+
+Local inference on this hardware currently supports only the PaddlePaddle inference engine.
+
+**We strongly recommend installing PaddleOCR-VL in a virtual environment to avoid dependency conflicts.** For example, use the Python venv standard library to create a virtual environment:
+
+```shell
+# Create a virtual environment
+python -m venv .venv_paddleocr
+# Activate the environment
+source .venv_paddleocr/bin/activate
+```
+
+Execute the following commands to complete the installation:
+
+```shell
+python -m pip install paddlepaddle==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **Please note to install PaddlePaddle version 3.2.1 or above.**
+
+## 2. Quick Start
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 2. Quick Start](./PaddleOCR-VL.en.md#2-quick-start).
+
+## 3. Using VLM Inference Services
+
+This section explains how to connect PaddleOCR-VL to a dedicated VLM inference service backend. On this hardware, this is usually used to improve inference performance beyond the default configuration for production use. In this hardware-specific guide, the examples use vLLM as the backend for the VLM inference service.
+
+### 3.1 Starting the VLM Inference Service
+
+> IMPORTANT:
+> The service started according to this section is responsible only for the VLM inference stage in the PaddleOCR-VL workflow. It does not provide a complete end-to-end document parsing API. We strongly recommend that you do not call this service directly via HTTP requests or OpenAI clients to process document images. If you need to deploy a service with the full PaddleOCR-VL capabilities, refer to the service deployment section later in this document.
+
+**Launch Methods Supported on This Hardware**
+
+| Launch method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Supported with steps in this guide | This section provides the vLLM service launch steps. |
+| Install dependencies with the PaddleOCR CLI and launch the service | Not currently supported | This hardware does not currently support this path. |
+| Launch the service directly with the acceleration framework | Not verified | This hardware can start the VLM inference service through the vLLM backend, but launching directly with native vLLM has not been verified. |
+
+PaddleOCR provides a Docker image for quickly starting the vLLM inference service. Use the following command to start the service (requires Docker version >= 19.03):
+
+```shell
+docker run -it \
+  --name paddleocr_vllm \
+  --user root \
+  --device /dev:/dev \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-amd-gpu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm
+```
+
+If you wish to start the service in an environment without internet access, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-amd-gpu` (image size approximately 31 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-amd-gpu-offline` (image size approximately 33 GB).
+
+When launching the vLLM inference service, we provide a set of default parameter settings. If you need to adjust parameters such as GPU memory usage, you can configure additional parameters yourself. Please refer to [3.3.1 Server-side Parameter Adjustment](./PaddleOCR-VL.en.md#331-server-side-parameter-adjustment) to create a configuration file, then mount the file into the container and specify the configuration file using `backend_config` in the command to start the service, for example:
+
+```shell
+docker run -it \
+  --name paddleocr_vllm \
+  --user root \
+  --device /dev:/dev \
+  --shm-size 64g \
+  --network host \
+  -v ./vllm_config.yml:/tmp/vllm_config.yml \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-amd-gpu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /tmp/vllm_config.yml
+```
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use an image corresponding to a specific PaddleOCR version, you can replace `latest` in the tag with the desired version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:paddleocr3.4-amd-gpu-offline`
+
+
+### 3.2 Client Usage Method
+
+For client-side invocation methods, please refer to [PaddleOCR-VL Usage Tutorial - 3.2 Client Usage Methods](./PaddleOCR-VL.en.md#32-client-usage-methods).
+
+### 3.3 Performance Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 3.3 Performance Tuning](./PaddleOCR-VL.en.md#33-performance-tuning).
+
+## 4. Service Deployment
+
+**Deployment Methods Supported on This Hardware**
+
+| Deployment method | Status | Notes |
+| --- | --- | --- |
+| Docker Compose deployment | Supported with steps in this guide | Continue with Section 4.1. |
+| Manual deployment | Not currently supported | This hardware does not currently support this path. |
+
+> IMPORTANT:
+> The PaddleOCR-VL service introduced in this section differs from the VLM inference service in the previous section: the latter is responsible for only one part of the complete process (i.e., VLM inference) and is called as an underlying service by the former.
+
+### 4.1 Deploy Using Docker Compose
+
+This step mainly introduces how to use Docker Compose to deploy PaddleOCR-VL as a service and call it. The specific process is as follows:
+
+1. Download the Compose file and the environment variable configuration file separately from [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/amd-gpu/compose.yaml) and [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/amd-gpu/.env) to your local machine.
+    
+2. Execute the following command in the directory where the `compose.yaml` and `.env` files are located to start the server, which listens on port **8080** by default:
+
+    ```shell
+    # Must be executed in the directory where compose.yaml and .env files are located
+    docker compose up
+    ```
+
+    > TIP:
+    > The image tags used by `compose.yaml` are usually controlled by `API_IMAGE_TAG_SUFFIX` and `VLM_IMAGE_TAG_SUFFIX` in `.env`, and default to tags such as `latest-amd-gpu-offline`. To make sure you pull the newest `latest` images, run `docker compose pull` in the current directory before `docker compose up`.
+    > To use an image corresponding to a specific PaddleOCR version, replace `latest` in these variables with `paddleocr<major>.<minor>`, for example `paddleocr3.3-amd-gpu-offline`.
+
+    After startup, you will see output similar to the following:
+
+    ```text
+    paddleocr-vl-api             | INFO:     Started server process [1]
+    paddleocr-vl-api             | INFO:     Waiting for application startup.
+    paddleocr-vl-api             | INFO:     Application startup complete.
+    paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+    ```
+
+This method accelerates VLM inference using the vLLM framework and is more suitable for production environment deployment.
+
+Additionally, after starting the server in this manner, no internet connection is required except for image pulling. For deployment in an offline environment, you can first pull the images involved in the Compose file on a connected machine, export them, and transfer them to the offline machine for import to start the service in an offline environment.
+
+Docker Compose starts two containers sequentially by reading configurations from the `.env` and `compose.yaml` files, running the underlying VLM inference service and the PaddleOCR-VL service (pipeline service) respectively.
+
+The meanings of each environment variable contained in the `.env` file are as follows:
+
+```
+- `API_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to launch the pipeline service.
+- `VLM_BACKEND`: The VLM inference backend.
+- `VLM_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to launch the VLM inference service.
+```
+
+You can modify `compose.yaml` to meet custom requirements, for example:
+
+<details>
+<summary>1. Change the port of the PaddleOCR-VL service</summary>
+
+Edit <code>paddleocr-vl-api.ports</code> in the <code>compose.yaml</code> file to change the port. For example, if you need to change the service port to 8111, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. Specify the GPU used by the PaddleOCR-VL service</summary>
+
+Edit <code>environment</code> in the <code>compose.yaml</code> file to change the GPU used. For example, if you need to use card 1 for deployment, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    environment:
++     - HIP_VISIBLE_DEVICES: 1
+    ...
+  paddleocr-vlm-server:
+    ...
+    environment:
++     - HIP_VISIBLE_DEVICES: 1
+    ...
+```
+
+
+</details>
+
+<details>
+<summary>3. Adjust VLM server-side configuration</summary>
+
+If you want to adjust the VLM server configuration, refer to <a href="./PaddleOCR-VL.en.md#331-server-side-parameter-adjustment">3.3.1 Server-side Parameter Adjustment</a> to generate a configuration file.
+
+After generating the configuration file, add the following <code>paddleocr-vlm-server.volumes</code> and <code>paddleocr-vlm-server.command</code> fields to your <code>compose.yaml</code>. Replace <code>/path/to/your_config.yaml</code> with your actual configuration file path.
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes: /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. Adjust pipeline-related configurations (such as model path, batch size, deployment device, etc.)</summary>
+
+Refer to the <a href="./PaddleOCR-VL.en.md#44-pipeline-configuration-adjustment-instructions">4.4 Pipeline Configuration Adjustment Instructions</a> section.
+
+</details>
+
+### 4.2 Client Invocation Method
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.3 Client-Side Invocation](./PaddleOCR-VL.en.md#43-client-side-invocation).
+
+### 4.3 Pipeline Configuration Adjustment Instructions
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.4 Pipeline Configuration Adjustment Instructions](./PaddleOCR-VL.en.md#44-pipeline-configuration-adjustment-instructions).
+
+## 5. Model Fine-Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 5. Model Fine-Tuning](./PaddleOCR-VL.en.md#5-model-fine-tuning).
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-AMD-GPU.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-AMD-GPU.md
new file mode 100644
index 0000000..6d049b1
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-AMD-GPU.md
@@ -0,0 +1,280 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL AMD GPU 使用教程
+
+> INFO:
+> 除非另有说明，本教程中提到的 “PaddleOCR-VL” 均指 PaddleOCR-VL 系列模型（如 PaddleOCR-VL-1.6 等）；若特指 PaddleOCR-VL v1 版本，将另行明确标注。
+
+本教程是 PaddleOCR-VL 在 AMD GPU 上的使用指南，涵盖了从本地运行环境准备到服务化部署的完整流程。
+
+目前 PaddleOCR-VL 已在 AMD MI300X 上完成精度、速度验证；鉴于硬件环境的多样性，其他 AMD GPU 的兼容性尚未验证。我们诚挚欢迎社区用户在不同硬件上进行测试并反馈您的运行结果。
+
+## 本硬件支持的使用目标
+
+请在本硬件教程中按下表继续阅读。
+
+| 目标 | 本硬件上的支持情况 | 从哪里开始阅读 |
+| --- | --- | --- |
+| 本地直接推理 | 支持 | 阅读第 1 节“本地运行环境准备”和第 2 节“快速开始”。 |
+| 客户端 + VLM 推理服务 | 支持 | 先完成本地直接推理，再阅读第 3 节“使用 VLM 推理服务”。 |
+| 完整 API 服务 | 支持 Docker Compose 部署 | 先阅读第 4.1 节，再继续阅读第 4.2 节客户端调用部分和第 4.3 节产线配置调整部分。 |
+| 模型微调 | 支持 | 阅读第 5 节“模型微调”。 |
+
+如果你只是想先确认本硬件支持哪些推理方式，请参考主教程中的 [PaddleOCR-VL 推理方式与硬件支持矩阵](./PaddleOCR-VL.md#paddleocr-vl-对推理设备的支持情况)。
+
+## 1. 本地运行环境准备
+
+**当前硬件支持的本地运行环境准备方式**
+
+| 本地运行环境准备方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 支持并提供步骤 | 请继续阅读本节的 1.1。 |
+| 手动安装推理引擎和 PaddleOCR | 支持并提供步骤 | 请继续阅读本节的 1.2。 |
+
+此步骤主要介绍如何搭建 PaddleOCR-VL 的本地运行环境，有以下两种方式，任选一种即可：
+
+- 方法一：使用官方 Docker 镜像。
+
+- 方法二：手动安装推理引擎和 PaddleOCR。
+
+**我们强烈推荐采用 Docker 镜像的方式，以最大程度减少可能出现的环境问题。**
+
+### 1.1 方法一：使用 Docker 镜像
+
+我们推荐使用官方 Docker 镜像（要求 Docker 版本 >= 19.03）：
+
+```shell
+docker run -it \
+  --user root \
+  --device /dev:/dev \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-amd-gpu \
+  /bin/bash
+# 在容器中调用 PaddleOCR CLI 或 Python API
+```
+
+如果您希望在无法连接互联网的环境中启动服务，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-amd-gpu`（镜像的大小约为 15 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-amd-gpu-offline`（镜像的大小约为 17 GB）。
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本的 PaddleOCR 镜像，可以将标签中的 `latest` 替换为对应版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.4-amd-gpu-offline`
+
+### 1.2 方法二：手动安装推理引擎和 PaddleOCR
+
+如果您无法使用 Docker，也可以手动安装推理引擎和 PaddleOCR。本文档验证过的 Python 版本范围为 3.9–3.13。
+
+当前硬件本地推理仅支持 PaddlePaddle 推理引擎。
+
+**我们强烈推荐您在虚拟环境中安装 PaddleOCR-VL，以避免发生依赖冲突。** 例如，使用 Python venv 标准库创建虚拟环境：
+
+```shell
+# 创建虚拟环境
+python -m venv .venv_paddleocr
+# 激活环境
+source .venv_paddleocr/bin/activate
+```
+
+执行如下命令完成安装：
+
+```shell
+python -m pip install paddlepaddle==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **请注意安装 3.2.0 及以上版本的飞桨框架。**
+
+## 2. 快速开始
+
+请参考 [PaddleOCR-VL 使用教程 - 2. 快速开始](./PaddleOCR-VL.md#2)。
+
+## 3. 使用 VLM 推理服务
+
+本节介绍如何通过 VLM 推理服务接入专用后端。对于当前硬件，这通常用于提升默认配置下的推理性能，以更好满足生产需求。在当前硬件文档中，示例使用 vLLM 作为 VLM 推理服务后端。
+
+### 3.1 启动 VLM 推理服务
+
+> IMPORTANT:
+> 按照本节说明启动的服务仅负责 PaddleOCR-VL 流程中的 VLM 推理环节，不提供完整的端到端文档解析 API。强烈不建议直接通过 HTTP 请求或使用 OpenAI 客户端调用该服务处理文档图像。若您需要部署具备 PaddleOCR-VL 完整能力的服务，请参考后文的服务化部署部分。
+
+**当前硬件支持的启动方式**
+
+| 启动方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 支持并提供步骤 | 本节提供 vLLM 推理服务的启动步骤。 |
+| 通过 PaddleOCR CLI 安装依赖后启动 | 当前不支持 | 当前硬件不支持该路径。 |
+| 直接使用推理加速框架启动 | 未验证 | 当前硬件可通过 vLLM 后端启动 VLM 推理服务，但尚未验证直接使用 vLLM 原生方式启动的路径。 |
+
+PaddleOCR 提供了 Docker 镜像，用于快速启动 vLLM 推理服务。可使用以下命令启动服务（要求 Docker 版本 >= 19.03）：
+
+```shell
+docker run -it \
+  --name paddleocr_vllm \
+  --user root \
+  --device /dev:/dev \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-amd-gpu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm
+```
+
+如果您希望在无法连接互联网的环境中启动服务，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-amd-gpu`（镜像的大小约为 31 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-amd-gpu-offline`（镜像的大小约为 33 GB）。
+
+启动 vLLM 推理服务时，我们提供了一套默认参数设置。如果您有调整显存占用等更多参数的需求，可以自行配置更多参数。请参考 [3.3.1 服务端参数调整](./PaddleOCR-VL.md#331) 创建配置文件，然后将该文件挂载到容器中，并在启动服务的命令中使用 `backend_config` 指定配置文件，例如：
+
+```shell
+docker run -it \
+  --name paddleocr_vllm \
+  --user root \
+  --device /dev:/dev \
+  --shm-size 64g \
+  --network host \
+  -v ./vllm_config.yml:/tmp/vllm_config.yml \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-amd-gpu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /tmp/vllm_config.yml
+```
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本的 PaddleOCR 镜像，可以将标签中的 `latest` 替换为对应版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:paddleocr3.4-amd-gpu-offline`
+
+### 3.2 客户端使用方法
+
+客户端调用方式请参考 [PaddleOCR-VL 使用教程 - 3.2 客户端使用方法](./PaddleOCR-VL.md#32)。
+
+### 3.3 性能调优
+
+请参考[PaddleOCR-VL 使用教程 - 3.3 性能调优](./PaddleOCR-VL.md#33)。
+
+## 4. 服务化部署
+
+**当前硬件支持的部署方式**
+
+| 部署方式 | 状态 | 说明 |
+| --- | --- | --- |
+| Docker Compose 部署 | 支持并提供步骤 | 请继续阅读本节的 4.1。 |
+| 手动部署 | 当前不支持 | 当前硬件不支持该路径。 |
+
+> IMPORTANT:
+> 本节所介绍的 PaddleOCR-VL 服务与上一节中的 VLM 推理服务有所区别：后者仅负责完整流程中的一个环节（即 VLM 推理），并作为前者的底层服务被调用。
+
+### 4.1 使用 Docker Compose 部署
+
+此步骤主要介绍如何使用 Docker Compose 将 PaddleOCR-VL 部署为服务并调用，具体流程如下：
+
+1. 分别从 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/amd-gpu/compose.yaml) 和 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/amd-gpu/.env) 获取 Compose 文件与环境变量配置文件并下载到本地。
+
+2. 在 `compose.yaml` 和 `.env` 文件所在目录下执行以下命令启动服务器，默认监听 **8080** 端口：
+
+    ```shell
+    # 必须在 compose.yaml 和 .env 文件所在的目录中执行
+    docker compose up
+    ```
+
+    > 提示：
+    > `compose.yaml` 中使用的镜像标签通常由 `.env` 中的 `API_IMAGE_TAG_SUFFIX` 和 `VLM_IMAGE_TAG_SUFFIX` 控制，默认使用 `latest-amd-gpu-offline` 等标签。如需确保拉取到最新的 `latest` 镜像，可先在当前目录执行 `docker compose pull`，再执行 `docker compose up`。
+    > 如果希望使用特定版本的 PaddleOCR 镜像，可将这两个环境变量中的 `latest` 替换为对应版本号 `paddleocr<major>.<minor>`，例如 `paddleocr3.3-amd-gpu-offline`。
+
+    启动后将看到类似如下输出：
+
+    ```text
+    paddleocr-vl-api             | INFO:     Started server process [1]
+    paddleocr-vl-api             | INFO:     Waiting for application startup.
+    paddleocr-vl-api             | INFO:     Application startup complete.
+    paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+    ```
+
+此方式基于 vLLM 框架对 VLM 推理进行加速，更适合生产环境部署。
+
+此外，使用此方式启动服务器后，除拉取镜像外，无需连接互联网。如需在离线环境中部署，可先在联网机器上拉取 Compose 文件中涉及的镜像，导出并传输至离线机器中导入，即可在离线环境下启动服务。
+
+Docker Compose 通过读取 `.env` 和 `compose.yaml` 文件中配置，先后启动 2 个容器，分别运行底层 VLM 推理服务，以及 PaddleOCR-VL 服务（产线服务）。
+
+`.env` 文件中包含的各环境变量含义如下：
+
+- `API_IMAGE_TAG_SUFFIX`：启动产线服务使用的镜像的标签后缀。
+- `VLM_BACKEND`：VLM 推理后端。
+- `VLM_IMAGE_TAG_SUFFIX`：启动 VLM 推理服务使用的镜像的标签后缀。
+
+您可以通过修改 `compose.yaml` 来满足自定义需求，例如：
+
+<details>
+<summary>1. 更改 PaddleOCR-VL 服务的端口</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>paddleocr-vl-api.ports</code> 来更改端口。例如，如果您需要将服务端口更换为 8111，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. 指定 PaddleOCR-VL 服务所使用的 GPU</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>environment</code> 来更改所使用的 GPU。例如，如果您需要使用卡 1 进行部署，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    environment:
++     - HIP_VISIBLE_DEVICES: 1
+    ...
+  paddleocr-vlm-server:
+    ...
+    environment:
++     - HIP_VISIBLE_DEVICES: 1
+    ...
+```
+
+
+</details>
+
+<details>
+<summary>3. 调整 VLM 服务端配置</summary>
+
+若您想调整 VLM 服务端的配置，可以参考 <a href="./PaddleOCR-VL.md#331">3.3.1 服务端参数调整</a> 生成配置文件。
+
+生成配置文件后，将以下的 <code>paddleocr-vlm-server.volumes</code> 和 <code>paddleocr-vlm-server.command</code> 字段增加到您的 <code>compose.yaml</code> 中。请将 <code>/path/to/your_config.yaml</code> 替换为您的实际配置文件路径。
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes: /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. 调整产线相关配置（如模型路径、批处理大小、部署设备等）</summary>
+
+参考 <a href="./PaddleOCR-VL.md#44">4.4 产线配置调整说明</a> 小节。
+
+</details>
+
+### 4.2 客户端调用方式
+
+请参考[PaddleOCR-VL 使用教程 - 4.3 客户端调用方式](./PaddleOCR-VL.md#43)。
+
+### 4.3 产线配置调整说明
+
+请参考[PaddleOCR-VL 使用教程 - 4.4 产线配置调整说明](./PaddleOCR-VL.md#44)。
+
+## 5. 模型微调
+
+请参考[PaddleOCR-VL 使用教程 - 5. 模型微调](./PaddleOCR-VL.md#5)。
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-Apple-Silicon.en.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Apple-Silicon.en.md
new file mode 100644
index 0000000..5ae8500
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Apple-Silicon.en.md
@@ -0,0 +1,151 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL Apple Silicon Usage Tutorial
+
+> INFO:
+> Unless otherwise specified, the term "PaddleOCR-VL" in this tutorial refers to the PaddleOCR-VL model series (e.g., PaddleOCR-VL-1.6). References specific to the PaddleOCR-VL v1 version will be explicitly noted.
+
+This tutorial is a guide for using PaddleOCR-VL on Apple Silicon, covering the complete workflow from environment preparation to service deployment.
+
+Apple Silicon include, but are not limited to:
+
+- Apple M1
+- Apple M2
+- Apple M3
+- Apple M4
+
+PaddleOCR-VL has been verified for accuracy and speed on the Apple M4. However, due to hardware diversity, compatibility with other Apple Silicon has not yet been confirmed. We welcome the community to test on different hardware setups and share your results.
+
+## Workflow Guide for This Hardware
+
+Use this guide for the workflows below.
+
+| Goal | Support on this hardware | Read this section |
+| --- | --- | --- |
+| Local direct inference | Supported | Read Section 1. Local Runtime Environment Preparation and Section 2. Quick Start. |
+| Client + VLM inference service | Supported | Complete local direct inference first, then read Section 3. Using VLM Inference Services. |
+| Full API service | Supported with manual deployment only | Complete Section 1. Local Runtime Environment Preparation first, then read Section 4.1 Manual Deployment; after that, continue with Section 4.2 Client Invocation Methods and Section 4.3 Pipeline Configuration Adjustment Instructions. |
+| Model fine-tuning | Supported | Read Section 5. Model Fine-Tuning. |
+
+If you only need to confirm which inference methods are available on this hardware, refer to the [PaddleOCR-VL Inference Method and Hardware Support Matrix](./PaddleOCR-VL.en.md#inference-device-support-for-paddleocr-vl) in the main guide.
+
+## 1. Local Runtime Environment Preparation
+
+**Local Runtime Environment Setup Methods Supported on This Hardware**
+
+| Local runtime environment setup method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Not currently supported | This hardware does not currently support this path. |
+| Manually install the inference engine and PaddleOCR | Supported with steps in this guide | Continue reading this section. |
+
+Local inference on this hardware currently supports only the PaddlePaddle inference engine.
+
+**We strongly recommend installing PaddleOCR-VL in a virtual environment to avoid dependency conflicts.** For example, use the Python venv standard library to create a virtual environment:
+
+```shell
+# Create a virtual environment
+python -m venv .venv_paddleocr
+# Activate the environment
+source .venv_paddleocr/bin/activate
+```
+
+Execute the following commands to complete the installation:
+
+```shell
+python -m pip install paddlepaddle==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **Please install PaddlePaddle framework version 3.2.1 or above.**
+
+## 2. Quick Start
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 2. Quick Start](./PaddleOCR-VL.en.md#2-quick-start).
+
+## 3. Using VLM Inference Services
+
+This section explains how to connect PaddleOCR-VL to a dedicated VLM inference service backend. On this hardware, this is usually used to improve inference performance beyond the default configuration for production use. In this hardware-specific guide, the examples use MLX-VLM as the backend for the VLM inference service.
+
+### 3.1 Starting the VLM Inference Service
+
+> IMPORTANT:
+> The service started according to this section is responsible only for the VLM inference stage in the PaddleOCR-VL workflow. It does not provide a complete end-to-end document parsing API. We strongly recommend that you do not call this service directly via HTTP requests or OpenAI clients to process document images. If you need to deploy a service with the full PaddleOCR-VL capabilities, refer to the service deployment section later in this document.
+
+**Launch Methods Supported on This Hardware**
+
+| Launch method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Not currently supported | This hardware does not currently support this path. |
+| Install dependencies with the PaddleOCR CLI and launch the service | Not currently supported | This hardware does not currently support this path. |
+| Launch the service directly with the acceleration framework | Supported with steps in this guide | This section provides the MLX-VLM launch steps. |
+
+Install the MLX-VLM inference framework (v0.3.11 or later):
+
+```shell
+python -m pip install "mlx-vlm>=0.3.11"
+```
+
+Start the MLX-VLM inference service:
+
+```shell
+mlx_vlm.server --port 8111
+```
+
+### 3.2 Client Usage Method
+
+The following invocation methods apply to an already launched MLX-VLM inference service.
+
+#### 3.2.1 Command Line Usage
+
+You can specify the backend type (`mlx-vlm-server`) via `--vl_rec_backend`, the service address via `--vl_rec_server_url`, and the huggingface repo id or server-side model weights path via `--vl_rec_api_model_name`. For example:
+
+```shell
+paddleocr doc_parser \
+  --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png \
+  --vl_rec_backend mlx-vlm-server \
+  --vl_rec_server_url http://localhost:8111/ \
+  --vl_rec_api_model_name PaddlePaddle/PaddleOCR-VL-1.6
+```
+
+#### 3.2.2 Python Script Integration
+
+When creating a `PaddleOCRVL` object, specify the backend type via `vl_rec_backend`, the service address via the `vl_rec_server_url` parameter, and the huggingface repo id or server-side model weights path via `vl_rec_api_model_name`. For example:
+
+```python
+pipeline = PaddleOCRVL(
+    vl_rec_backend="mlx-vlm-server", 
+    vl_rec_server_url="http://localhost:8111/",
+    vl_rec_api_model_name="PaddlePaddle/PaddleOCR-VL-1.6",
+)
+```
+
+### 3.3 Performance Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 3.3 Performance Tuning](./PaddleOCR-VL.en.md#33-performance-tuning).
+
+## 4. Service Deployment
+
+**Deployment Methods Supported on This Hardware**
+
+| Deployment method | Status | Notes |
+| --- | --- | --- |
+| Docker Compose deployment | Not currently supported | This hardware currently supports only the manual deployment path. |
+| Manual deployment | Supported | Complete Section 1. Local Runtime Environment Preparation first, then continue with Section 4.1. |
+
+### 4.1 Manual Deployment
+
+Please complete Section 1. Local Runtime Environment Preparation first, then refer to [PaddleOCR-VL Usage Tutorial - 4.2 Method 2: Manual Deployment](./PaddleOCR-VL.en.md#42-method-2-manual-deployment).
+
+### 4.2 Client Invocation Methods
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.3 Client Invocation Methods](./PaddleOCR-VL.en.md#43-client-side-invocation).
+
+### 4.3 Pipeline Configuration Adjustment Instructions
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.4 Pipeline Configuration Adjustment Instructions](./PaddleOCR-VL.en.md#44-pipeline-configuration-adjustment-instructions).
+
+## 5. Model Fine-Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 5. Model Fine-Tuning](./PaddleOCR-VL.en.md#5-model-fine-tuning).
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-Apple-Silicon.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Apple-Silicon.md
new file mode 100644
index 0000000..acd2104
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Apple-Silicon.md
@@ -0,0 +1,151 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL Apple Silicon 使用教程
+
+> INFO:
+> 除非另有说明，本教程中提到的 “PaddleOCR-VL” 均指 PaddleOCR-VL 系列模型（如 PaddleOCR-VL-1.6 等）；若特指 PaddleOCR-VL v1 版本，将另行明确标注。
+
+本教程是 PaddleOCR-VL 在 Apple Silicon 上的使用指南，涵盖了从本地运行环境准备到服务化部署的完整流程。
+
+Apple Silicon 包括但不限于以下几种：
+
+- Apple M1
+- Apple M2
+- Apple M3
+- Apple M4
+
+目前 PaddleOCR-VL 已在 Apple M4 上完成精度验证；鉴于硬件环境的多样性，其他 Apple Silicon 的兼容性尚未验证。我们诚挚欢迎社区用户在不同硬件上进行测试并反馈您的运行结果。
+
+## 本硬件支持的使用目标
+
+请在本硬件教程中按下表继续阅读。
+
+| 目标 | 本硬件上的支持情况 | 从哪里开始阅读 |
+| --- | --- | --- |
+| 本地直接推理 | 支持 | 阅读第 1 节“本地运行环境准备”和第 2 节“快速开始”。 |
+| 客户端 + VLM 推理服务 | 支持 | 先完成本地直接推理，再阅读第 3 节“使用 VLM 推理服务”。 |
+| 完整 API 服务 | 仅支持手动部署 | 先完成第 1 节“本地运行环境准备”，再阅读第 4.1 节“手动部署”；然后继续阅读第 4.2 节客户端调用部分和第 4.3 节产线配置调整部分。 |
+| 模型微调 | 支持 | 阅读第 5 节“模型微调”。 |
+
+如果你只是想先确认本硬件支持哪些推理方式，请参考主教程中的 [PaddleOCR-VL 推理方式与硬件支持矩阵](./PaddleOCR-VL.md#paddleocr-vl-对推理设备的支持情况)。
+
+## 1. 本地运行环境准备
+
+**当前硬件支持的本地运行环境准备方式**
+
+| 本地运行环境准备方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 当前不支持 | 当前硬件不支持该路径。 |
+| 手动安装推理引擎和 PaddleOCR | 支持并提供步骤 | 请继续阅读本节。 |
+
+当前硬件本地推理仅支持 PaddlePaddle 推理引擎。
+
+**我们强烈推荐您在虚拟环境中安装 PaddleOCR-VL，以避免发生依赖冲突。** 例如，使用 Python venv 标准库创建虚拟环境：
+
+```shell
+# 创建虚拟环境
+python -m venv .venv_paddleocr
+# 激活环境
+source .venv_paddleocr/bin/activate
+```
+
+执行如下命令完成安装：
+
+```shell
+python -m pip install paddlepaddle==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **请注意安装 3.2.1 及以上版本的飞桨框架。**
+
+## 2. 快速开始
+
+请参考[PaddleOCR-VL 使用教程 - 2. 快速开始](./PaddleOCR-VL.md#2)。
+
+## 3. 使用 VLM 推理服务
+
+本节介绍如何通过 VLM 推理服务接入专用后端。对于当前硬件，这通常用于提升默认配置下的推理性能，以更好满足生产需求。在当前硬件文档中，示例使用 MLX-VLM 作为 VLM 推理服务后端。
+
+### 3.1 启动 VLM 推理服务
+
+> IMPORTANT:
+> 按照本节说明启动的服务仅负责 PaddleOCR-VL 流程中的 VLM 推理环节，不提供完整的端到端文档解析 API。强烈不建议直接通过 HTTP 请求或使用 OpenAI 客户端调用该服务处理文档图像。若您需要部署具备 PaddleOCR-VL 完整能力的服务，请参考后文的服务化部署部分。
+
+**当前硬件支持的启动方式**
+
+| 启动方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 当前不支持 | 当前硬件不支持该路径。 |
+| 通过 PaddleOCR CLI 安装依赖后启动 | 当前不支持 | 当前硬件不支持该路径。 |
+| 直接使用推理加速框架启动 | 支持并提供步骤 | 本节提供 MLX-VLM 的启动步骤。 |
+
+安装 MLX-VLM 推理框架（v0.3.11以上版本）：
+
+```shell
+python -m pip install "mlx-vlm>=0.3.11"
+```
+
+启动 MLX-VLM 推理服务：
+
+```shell
+mlx_vlm.server --port 8111
+```
+
+### 3.2 客户端使用方法
+
+以下调用方式适用于已启动的 MLX-VLM 推理服务。
+
+#### 3.2.1 CLI 调用
+
+可通过 `--vl_rec_backend` 指定后端类型（`mlx-vlm-server`），通过 `--vl_rec_server_url` 指定服务地址，通过 `--vl_rec_api_model_name` 指定 huggingface repo id 或服务端模型权重路径，例如：
+
+```shell
+paddleocr doc_parser \
+  --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png \
+  --vl_rec_backend mlx-vlm-server \
+  --vl_rec_server_url http://localhost:8111/ \
+  --vl_rec_api_model_name PaddlePaddle/PaddleOCR-VL-1.6
+```
+
+#### 3.2.2 Python API 调用
+
+创建 `PaddleOCRVL` 对象时传入 `vl_rec_backend` 指定后端类型， `vl_rec_server_url` 参数指定服务地址，`vl_rec_api_model_name` 指定 huggingface repo id 或服务端模型权重路径，例如：
+
+```python
+pipeline = PaddleOCRVL(
+    vl_rec_backend="mlx-vlm-server", 
+    vl_rec_server_url="http://localhost:8111/",
+    vl_rec_api_model_name="PaddlePaddle/PaddleOCR-VL-1.6",
+)
+```
+
+### 3.3 性能调优
+
+请参考[PaddleOCR-VL 使用教程 - 3.3 性能调优](./PaddleOCR-VL.md#33)。
+
+## 4. 服务化部署
+
+**当前硬件支持的部署方式**
+
+| 部署方式 | 状态 | 说明 |
+| --- | --- | --- |
+| Docker Compose 部署 | 当前不支持 | 当前硬件仅支持手动部署路径。 |
+| 手动部署 | 支持 | 请先完成第 1 节“本地运行环境准备”，再继续阅读本节的 4.1。 |
+
+### 4.1 手动部署
+
+请先完成第 1 节“本地运行环境准备”，再参考 [PaddleOCR-VL 使用教程 - 4.2 方法二：手动部署](./PaddleOCR-VL.md#42)。
+
+### 4.2 客户端调用方式
+
+请参考[PaddleOCR-VL 使用教程 - 4.3 客户端调用方式](./PaddleOCR-VL.md#43)。
+
+### 4.3 产线配置调整说明
+
+请参考[PaddleOCR-VL 使用教程 - 4.4 产线配置调整说明](./PaddleOCR-VL.md#44)。
+
+## 5. 模型微调
+
+请参考[PaddleOCR-VL 使用教程 - 5. 模型微调](./PaddleOCR-VL.md#5)。
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-Huawei-Ascend-NPU.en.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Huawei-Ascend-NPU.en.md
new file mode 100644
index 0000000..80215ae
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Huawei-Ascend-NPU.en.md
@@ -0,0 +1,292 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL Huawei Ascend NPU Usage Tutorial
+
+> INFO:
+> Unless otherwise specified, the term "PaddleOCR-VL" in this tutorial refers to the PaddleOCR-VL model series (e.g., PaddleOCR-VL-1.6). References specific to the PaddleOCR-VL v1 version will be explicitly noted.
+
+This tutorial is a guide for using PaddleOCR-VL on Huawei Ascend NPU, covering the complete workflow from environment preparation to service deployment.
+
+PaddleOCR-VL has been verified for accuracy and speed on the Huawei Ascend 910B. However, due to hardware diversity, compatibility with other Huawei Ascend NPUs has not yet been confirmed. We welcome the community to test on different hardware setups and share your results.
+
+## Workflow Guide for This Hardware
+
+Use this guide for the workflows below.
+
+| Goal | Support on this hardware | Read this section |
+| --- | --- | --- |
+| Local direct inference | This hardware does not currently support the local direct inference path | Use the “Client + VLM inference service” path instead: start with Section 1. Local Runtime Environment Preparation, then read Section 3. |
+| Client + VLM inference service | Supported | Start with Section 1. Local Runtime Environment Preparation, then read Section 3. Using VLM Inference Services. |
+| Full API service | Supported with Docker Compose deployment | Read Section 4.1 first, then continue with the Section 4.2 client invocation section and the Section 4.3 pipeline configuration section. |
+| Model fine-tuning | Supported | Read Section 5. Model Fine-Tuning. |
+
+If you only need to confirm which inference methods are available on this hardware, refer to the [PaddleOCR-VL Inference Method and Hardware Support Matrix](./PaddleOCR-VL.en.md#inference-device-support-for-paddleocr-vl) in the main guide.
+
+## 1. Local Runtime Environment Preparation
+
+**Local Runtime Environment Setup Methods Supported on This Hardware**
+
+| Local runtime environment setup method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Supported with steps in this guide | Continue with Section 1.1. |
+| Manually install the inference engine and PaddleOCR | Supported with steps in this guide | Continue with Section 1.2. |
+
+This step mainly introduces how to set up the local runtime environment for PaddleOCR-VL. There are two methods available; choose either one:
+
+- Method 1: Use the official Docker image.
+
+- Method 2: Manually install the inference engine and PaddleOCR.
+
+**We strongly recommend using the Docker image to minimize potential environment-related issues.**
+
+### 1.1 Method 1: Using Docker Image
+
+We recommend using the official Docker image (requires Docker version >= 19.03):
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
+  -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
+  -v /usr/local/dcmi:/usr/local/dcmi \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-huawei-npu \
+  /bin/bash
+# Call PaddleOCR CLI or Python API in the container
+```
+
+If you wish to start the service in an environment without internet access, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-huawei-npu` (image size approximately 28 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-huawei-npu-offline` (image size approximately 30 GB).
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use an image corresponding to a specific PaddleOCR version, you can replace `latest` in the tag with the desired version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.4-huawei-npu-offline`
+
+
+### 1.2 Method 2: Manually Install the Inference Engine and PaddleOCR
+
+If you cannot use Docker, you can also manually install the inference engine and PaddleOCR. This guide documents Python 3.9–3.13 as the verified range.
+
+Local inference on this hardware currently provides PaddlePaddle installation steps only; support for other inference engines is still being validated.
+
+**We strongly recommend installing PaddleOCR-VL in a virtual environment to avoid dependency conflicts.** For example, use the Python venv standard library to create a virtual environment:
+
+```shell
+# Create a virtual environment
+python -m venv .venv_paddleocr
+# Activate the environment
+source .venv_paddleocr/bin/activate
+```
+
+Execute the following commands to complete the installation:
+
+```shell
+python -m pip install paddlepaddle==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+python -m pip install paddle-custom-npu==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/npu/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **Please note to install PaddlePaddle version 3.2.1 or above.**
+
+## 2. Quick Start
+
+This hardware does not currently support the local direct inference path. To use the supported accelerated path on this hardware, continue to the next section and use the vLLM inference service path.
+
+## 3. Using VLM Inference Services
+
+This section explains how to complete the client + VLM inference service path through a VLM inference service. In this hardware-specific guide, the examples use vLLM as the backend for the VLM inference service.
+
+### 3.1 Starting the VLM Inference Service
+
+> IMPORTANT:
+> The service started according to this section is responsible only for the VLM inference stage in the PaddleOCR-VL workflow. It does not provide a complete end-to-end document parsing API. We strongly recommend that you do not call this service directly via HTTP requests or OpenAI clients to process document images. If you need to deploy a service with the full PaddleOCR-VL capabilities, refer to the service deployment section later in this document.
+
+**Launch Methods Supported on This Hardware**
+
+| Launch method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Supported with steps in this guide | This section provides the vLLM service launch steps. |
+| Install dependencies with the PaddleOCR CLI and launch the service | Not currently supported | This hardware does not currently support this path. |
+| Launch the service directly with the acceleration framework | Not verified | This hardware can start the VLM inference service through the vLLM backend, but launching directly with native vLLM has not been verified. |
+
+PaddleOCR provides a Docker image for quickly starting the vLLM inference service. Use the following command to start the service (requires Docker version >= 19.03):
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
+  -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
+  -v /usr/local/dcmi:/usr/local/dcmi \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-huawei-npu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm
+```
+
+If you wish to start the service in an environment without internet access, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-huawei-npu` (image size approximately 18 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-huawei-npu-offline` (image size approximately 20 GB).
+
+When launching the vLLM inference service, we provide a set of default parameter settings. If you need to adjust parameters such as GPU memory usage, you can configure additional parameters yourself. Please refer to [3.3.1 Server-side Parameter Adjustment](./PaddleOCR-VL.en.md#331-server-side-parameter-adjustment) to create a configuration file, then mount the file into the container and specify the configuration file using `backend_config` in the command to start the service, for example:
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
+  -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
+  -v /usr/local/dcmi:/usr/local/dcmi \
+  -v ./vllm_config.yml:/tmp/vllm_config.yml \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-huawei-npu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /tmp/vllm_config.yml
+```
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use an image corresponding to a specific PaddleOCR version, you can replace `latest` in the tag with the desired version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:paddleocr3.4-huawei-npu-offline`
+
+
+### 3.2 Client Usage Method
+
+For client-side invocation methods, please refer to [PaddleOCR-VL Usage Tutorial - 3.2 Client Usage Methods](./PaddleOCR-VL.en.md#32-client-usage-methods). If you run the client on this hardware, make sure to specify `device="npu"`.
+
+### 3.3 Performance Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 3.3 Performance Tuning](./PaddleOCR-VL.en.md#33-performance-tuning).
+
+## 4. Service Deployment
+
+**Deployment Methods Supported on This Hardware**
+
+| Deployment method | Status | Notes |
+| --- | --- | --- |
+| Docker Compose deployment | Supported with steps in this guide | Continue with Section 4.1. |
+| Manual deployment | Not currently supported | This hardware does not currently support this path. |
+
+> IMPORTANT:
+> The PaddleOCR-VL service introduced in this section differs from the VLM inference service in the previous section: the latter is responsible for only one part of the complete process (i.e., VLM inference) and is called as an underlying service by the former.
+
+### 4.1 Deploy Using Docker Compose
+
+This step mainly introduces how to use Docker Compose to deploy PaddleOCR-VL as a service and call it. The specific process is as follows:
+
+1. Download the Compose file and the environment variable configuration file separately from [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/huawei-npu/compose.yaml) and [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/huawei-npu/.env) to your local machine.
+    
+2. Execute the following command in the directory where the `compose.yaml` and `.env` files are located to start the server, which listens on port **8080** by default:
+
+    ```shell
+    # Must be executed in the directory where compose.yaml and .env files are located
+    docker compose up
+    ```
+
+    > TIP:
+    > The image tags used by `compose.yaml` are usually controlled by `API_IMAGE_TAG_SUFFIX` and `VLM_IMAGE_TAG_SUFFIX` in `.env`, and default to tags such as `latest-huawei-npu-offline`. To make sure you pull the newest `latest` images, run `docker compose pull` in the current directory before `docker compose up`.
+    > To use an image corresponding to a specific PaddleOCR version, replace `latest` in these variables with `paddleocr<major>.<minor>`, for example `paddleocr3.3-huawei-npu-offline`.
+
+    After startup, you will see output similar to the following:
+
+    ```text
+    paddleocr-vl-api             | INFO:     Started server process [1]
+    paddleocr-vl-api             | INFO:     Waiting for application startup.
+    paddleocr-vl-api             | INFO:     Application startup complete.
+    paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+    ```
+
+This method accelerates VLM inference using the vLLM framework and is more suitable for production environment deployment.
+
+Additionally, after starting the server in this manner, no internet connection is required except for image pulling. For deployment in an offline environment, you can first pull the images involved in the Compose file on a connected machine, export them, and transfer them to the offline machine for import to start the service in an offline environment.
+
+Docker Compose starts two containers sequentially by reading configurations from the `.env` and `compose.yaml` files, running the underlying VLM inference service and the PaddleOCR-VL service (pipeline service) respectively.
+
+The meanings of each environment variable contained in the `.env` file are as follows:
+
+```
+- `API_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to launch the pipeline service.
+- `VLM_BACKEND`: The VLM inference backend.
+- `VLM_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to launch the VLM inference service.
+```
+
+You can modify `compose.yaml` to meet custom requirements, for example:
+
+<details>
+<summary>1. Change the port of the PaddleOCR-VL service</summary>
+
+Edit <code>paddleocr-vl-api.ports</code> in the <code>compose.yaml</code> file to change the port. For example, if you need to change the service port to 8111, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. Specify the NPU used by the PaddleOCR-VL service</summary>
+
+Edit <code>environment</code> in the <code>compose.yaml</code> file to change the NPU used. For example, if you need to use card 1 for deployment, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    environment:
++     - ASCEND_RT_VISIBLE_DEVICES: 1
+    ...
+  paddleocr-vlm-server:
+    ...
+    environment:
++     - ASCEND_RT_VISIBLE_DEVICES: 1
+    ...
+```
+
+
+</details>
+
+<details>
+<summary>3. Adjust VLM server-side configuration</summary>
+
+If you want to adjust the VLM server configuration, refer to <a href="./PaddleOCR-VL.en.md#331-server-side-parameter-adjustment">3.3.1 Server-side Parameter Adjustment</a> to generate a configuration file.
+
+After generating the configuration file, add the following <code>paddleocr-vlm-server.volumes</code> and <code>paddleocr-vlm-server.command</code> fields to your <code>compose.yaml</code>. Replace <code>/path/to/your_config.yaml</code> with your actual configuration file path.
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes: /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. Adjust pipeline-related configurations (such as model path, batch size, deployment device, etc.)</summary>
+
+Refer to the <a href="./PaddleOCR-VL.en.md#44-pipeline-configuration-adjustment-instructions">4.4 Pipeline Configuration Adjustment Instructions</a> section.
+
+</details>
+
+### 4.2 Client Invocation Method
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.3 Client-Side Invocation](./PaddleOCR-VL.en.md#43-client-side-invocation).
+
+### 4.3 Pipeline Configuration Adjustment Instructions
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.4 Pipeline Configuration Adjustment Instructions](./PaddleOCR-VL.en.md#44-pipeline-configuration-adjustment-instructions).
+
+## 5. Model Fine-Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 5. Model Fine-Tuning](./PaddleOCR-VL.en.md#5-model-fine-tuning).
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-Huawei-Ascend-NPU.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Huawei-Ascend-NPU.md
new file mode 100644
index 0000000..21c1e3c
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Huawei-Ascend-NPU.md
@@ -0,0 +1,288 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL 华为昇腾 NPU 使用教程
+
+> INFO:
+> 除非另有说明，本教程中提到的 “PaddleOCR-VL” 均指 PaddleOCR-VL 系列模型（如 PaddleOCR-VL-1.6 等）；若特指 PaddleOCR-VL v1 版本，将另行明确标注。
+
+本教程是 PaddleOCR-VL 在华为昇腾 NPU 上的使用指南，涵盖了从本地运行环境准备到服务化部署的完整流程。
+
+目前 PaddleOCR-VL 已在华为昇腾 910B 上完成精度、速度验证；鉴于硬件环境的多样性，其他华为昇腾 NPU 的兼容性尚未验证。我们诚挚欢迎社区用户在不同硬件上进行测试并反馈您的运行结果。
+
+## 本硬件支持的使用目标
+
+请在本硬件教程中按下表继续阅读。
+
+| 目标 | 本硬件上的支持情况 | 从哪里开始阅读 |
+| --- | --- | --- |
+| 本地直接推理 | 当前不支持“本地直接推理”路径 | 请改走“客户端 + VLM 推理服务”路径：从第 1 节“本地运行环境准备”开始，然后阅读第 3 节。 |
+| 客户端 + VLM 推理服务 | 支持 | 从第 1 节“本地运行环境准备”开始，然后阅读第 3 节“使用 VLM 推理服务”。 |
+| 完整 API 服务 | 支持 Docker Compose 部署 | 先阅读第 4.1 节，再继续阅读第 4.2 节客户端调用部分和第 4.3 节产线配置调整部分。 |
+| 模型微调 | 支持 | 阅读第 5 节“模型微调”。 |
+
+如果你只是想先确认本硬件支持哪些推理方式，请参考主教程中的 [PaddleOCR-VL 推理方式与硬件支持矩阵](./PaddleOCR-VL.md#paddleocr-vl-对推理设备的支持情况)。
+
+## 1. 本地运行环境准备
+
+**当前硬件支持的本地运行环境准备方式**
+
+| 本地运行环境准备方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 支持并提供步骤 | 请继续阅读本节的 1.1。 |
+| 手动安装推理引擎和 PaddleOCR | 支持并提供步骤 | 请继续阅读本节的 1.2。 |
+
+此步骤主要介绍如何搭建 PaddleOCR-VL 的本地运行环境，有以下两种方式，任选一种即可：
+
+- 方法一：使用官方 Docker 镜像。
+
+- 方法二：手动安装推理引擎和 PaddleOCR。
+
+**我们强烈推荐采用 Docker 镜像的方式，以最大程度减少可能出现的环境问题。**
+
+### 1.1 方法一：使用 Docker 镜像
+
+我们推荐使用官方 Docker 镜像（要求 Docker 版本 >= 19.03）：
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
+  -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
+  -v /usr/local/dcmi:/usr/local/dcmi \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-huawei-npu \
+  /bin/bash
+# 在容器中调用 PaddleOCR CLI 或 Python API
+```
+
+如果您希望在无法连接互联网的环境中启动服务，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-huawei-npu`（镜像的大小约为 28 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-huawei-npu-offline`（镜像的大小约为 30 GB）。
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本的 PaddleOCR 镜像，可以将标签中的 `latest` 替换为对应版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.4-huawei-npu-offline`
+
+### 1.2 方法二：手动安装推理引擎和 PaddleOCR
+
+如果您无法使用 Docker，也可以手动安装推理引擎和 PaddleOCR。本文档验证过的 Python 版本范围为 3.9–3.13。
+
+当前硬件本地推理仅提供 PaddlePaddle 安装步骤，其他推理引擎尚在适配验证中。
+
+**我们强烈推荐您在虚拟环境中安装 PaddleOCR-VL，以避免发生依赖冲突。** 例如，使用 Python venv 标准库创建虚拟环境：
+
+```shell
+# 创建虚拟环境
+python -m venv .venv_paddleocr
+# 激活环境
+source .venv_paddleocr/bin/activate
+```
+
+执行如下命令完成安装：
+
+```shell
+python -m pip install paddlepaddle==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+python -m pip install paddle-custom-npu==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/npu/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **请注意安装 3.2.0 及以上版本的飞桨框架。**
+
+## 2. 快速开始
+
+当前硬件暂不支持“本地直接推理”路径。如需使用当前硬件加速方案，请继续阅读下一节，采用 vLLM 推理服务路径。
+
+## 3. 使用 VLM 推理服务
+
+本节介绍如何通过 VLM 推理服务完成“客户端 + VLM 推理服务”的组合路径。在当前硬件文档中，示例使用 vLLM 作为 VLM 推理服务后端。
+
+### 3.1 启动 VLM 推理服务
+
+> IMPORTANT:
+> 按照本节说明启动的服务仅负责 PaddleOCR-VL 流程中的 VLM 推理环节，不提供完整的端到端文档解析 API。强烈不建议直接通过 HTTP 请求或使用 OpenAI 客户端调用该服务处理文档图像。若您需要部署具备 PaddleOCR-VL 完整能力的服务，请参考后文的服务化部署部分。
+
+**当前硬件支持的启动方式**
+
+| 启动方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 支持并提供步骤 | 本节提供 vLLM 推理服务的启动步骤。 |
+| 通过 PaddleOCR CLI 安装依赖后启动 | 当前不支持 | 当前硬件不支持该路径。 |
+| 直接使用推理加速框架启动 | 未验证 | 当前硬件可通过 vLLM 后端启动 VLM 推理服务，但尚未验证直接使用 vLLM 原生方式启动的路径。 |
+
+PaddleOCR 提供了 Docker 镜像，用于快速启动 vLLM 推理服务。可使用以下命令启动服务（要求 Docker 版本 >= 19.03）：
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
+  -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
+  -v /usr/local/dcmi:/usr/local/dcmi \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-huawei-npu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm
+```
+
+如果您希望在无法连接互联网的环境中启动服务，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-huawei-npu`（镜像的大小约为 18 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-huawei-npu-offline`（镜像的大小约为 20 GB）。
+
+启动 vLLM 推理服务时，我们提供了一套默认参数设置。如果您有调整显存占用等更多参数的需求，可以自行配置更多参数。请参考 [3.3.1 服务端参数调整](./PaddleOCR-VL.md#331) 创建配置文件，然后将该文件挂载到容器中，并在启动服务的命令中使用 `backend_config` 指定配置文件，例如：
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
+  -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
+  -v /usr/local/dcmi:/usr/local/dcmi \
+  -v ./vllm_config.yml:/tmp/vllm_config.yml \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-huawei-npu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /tmp/vllm_config.yml
+```
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本的 PaddleOCR 镜像，可以将标签中的 `latest` 替换为对应版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:paddleocr3.4-huawei-npu-offline`
+
+### 3.2 客户端使用方法
+
+客户端调用方式请参考 [PaddleOCR-VL 使用教程 - 3.2 客户端使用方法](./PaddleOCR-VL.md#32)。如需在当前硬件上运行客户端，请注意指定 `device="npu"`。
+
+### 3.3 性能调优
+
+请参考[PaddleOCR-VL 使用教程 - 3.3 性能调优](./PaddleOCR-VL.md#33)。
+
+## 4. 服务化部署
+
+**当前硬件支持的部署方式**
+
+| 部署方式 | 状态 | 说明 |
+| --- | --- | --- |
+| Docker Compose 部署 | 支持并提供步骤 | 请继续阅读本节的 4.1。 |
+| 手动部署 | 当前不支持 | 当前硬件不支持该路径。 |
+
+> IMPORTANT:
+> 本节所介绍的 PaddleOCR-VL 服务与上一节中的 VLM 推理服务有所区别：后者仅负责完整流程中的一个环节（即 VLM 推理），并作为前者的底层服务被调用。
+
+### 4.1 使用 Docker Compose 部署
+
+此步骤主要介绍如何使用 Docker Compose 将 PaddleOCR-VL 部署为服务并调用，具体流程如下：
+
+1. 分别从 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/huawei-npu/compose.yaml) 和 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/huawei-npu/.env) 获取 Compose 文件与环境变量配置文件并下载到本地。
+
+2. 在 `compose.yaml` 和 `.env` 文件所在目录下执行以下命令启动服务器，默认监听 **8080** 端口：
+
+    ```shell
+    # 必须在 compose.yaml 和 .env 文件所在的目录中执行
+    docker compose up
+    ```
+
+    > 提示：
+    > `compose.yaml` 中使用的镜像标签通常由 `.env` 中的 `API_IMAGE_TAG_SUFFIX` 和 `VLM_IMAGE_TAG_SUFFIX` 控制，默认使用 `latest-huawei-npu-offline` 等标签。如需确保拉取到最新的 `latest` 镜像，可先在当前目录执行 `docker compose pull`，再执行 `docker compose up`。
+    > 如果希望使用特定版本的 PaddleOCR 镜像，可将这两个环境变量中的 `latest` 替换为对应版本号 `paddleocr<major>.<minor>`，例如 `paddleocr3.3-huawei-npu-offline`。
+
+    启动后将看到类似如下输出：
+
+    ```text
+    paddleocr-vl-api             | INFO:     Started server process [1]
+    paddleocr-vl-api             | INFO:     Waiting for application startup.
+    paddleocr-vl-api             | INFO:     Application startup complete.
+    paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+    ```
+
+此方式基于 vLLM 框架对 VLM 推理进行加速，更适合生产环境部署。
+
+此外，使用此方式启动服务器后，除拉取镜像外，无需连接互联网。如需在离线环境中部署，可先在联网机器上拉取 Compose 文件中涉及的镜像，导出并传输至离线机器中导入，即可在离线环境下启动服务。
+
+Docker Compose 通过读取 `.env` 和 `compose.yaml` 文件中配置，先后启动 2 个容器，分别运行底层 VLM 推理服务，以及 PaddleOCR-VL 服务（产线服务）。
+
+`.env` 文件中包含的各环境变量含义如下：
+
+- `API_IMAGE_TAG_SUFFIX`：启动产线服务使用的镜像的标签后缀。
+- `VLM_BACKEND`：VLM 推理后端。
+- `VLM_IMAGE_TAG_SUFFIX`：启动 VLM 推理服务使用的镜像的标签后缀。
+
+您可以通过修改 `compose.yaml` 来满足自定义需求，例如：
+
+<details>
+<summary>1. 更改 PaddleOCR-VL 服务的端口</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>paddleocr-vl-api.ports</code> 来更改端口。例如，如果您需要将服务端口更换为 8111，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. 指定 PaddleOCR-VL 服务所使用的 NPU</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>environment</code> 来更改所使用的 NPU。例如，如果您需要使用卡 1 进行部署，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    environment:
++     - ASCEND_RT_VISIBLE_DEVICES: 1
+    ...
+  paddleocr-vlm-server:
+    ...
+    environment:
++     - ASCEND_RT_VISIBLE_DEVICES: 1
+    ...
+```
+
+
+</details>
+
+<details>
+<summary>3. 调整 VLM 服务端配置</summary>
+
+若您想调整 VLM 服务端的配置，可以参考 <a href="./PaddleOCR-VL.md#331">3.3.1 服务端参数调整</a> 生成配置文件。
+
+生成配置文件后，将以下的 <code>paddleocr-vlm-server.volumes</code> 和 <code>paddleocr-vlm-server.command</code> 字段增加到您的 <code>compose.yaml</code> 中。请将 <code>/path/to/your_config.yaml</code> 替换为您的实际配置文件路径。
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes: /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. 调整产线相关配置（如模型路径、批处理大小、部署设备等）</summary>
+
+参考 <a href="./PaddleOCR-VL.md#44">4.4 产线配置调整说明</a> 小节。
+
+</details>
+
+### 4.2 客户端调用方式
+
+请参考[PaddleOCR-VL 使用教程 - 4.3 客户端调用方式](./PaddleOCR-VL.md#43)。
+
+### 4.3 产线配置调整说明
+
+请参考[PaddleOCR-VL 使用教程 - 4.4 产线配置调整说明](./PaddleOCR-VL.md#44)。
+
+## 5. 模型微调
+
+请参考[PaddleOCR-VL 使用教程 - 5. 模型微调](./PaddleOCR-VL.md#5)。
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-Hygon-DCU.en.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Hygon-DCU.en.md
new file mode 100644
index 0000000..0112f5f
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Hygon-DCU.en.md
@@ -0,0 +1,305 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL Hygon DCU Usage Tutorial
+
+> INFO:
+> Unless otherwise specified, the term "PaddleOCR-VL" in this tutorial refers to the PaddleOCR-VL model series (e.g., PaddleOCR-VL-1.6). References specific to the PaddleOCR-VL v1 version will be explicitly noted.
+
+This tutorial is a guide for using PaddleOCR-VL on Hygon DCU, covering the complete workflow from environment preparation to service deployment.
+
+PaddleOCR-VL has been verified for accuracy and speed on the Hygon K100AI. However, due to hardware diversity, compatibility with other Hygon DCUs has not yet been confirmed. We welcome the community to test on different hardware setups and share your results.
+
+## Workflow Guide for This Hardware
+
+Use this guide for the workflows below.
+
+| Goal | Support on this hardware | Read this section |
+| --- | --- | --- |
+| Local direct inference | Supported | Read Section 1. Local Runtime Environment Preparation and Section 2. Quick Start. |
+| Client + VLM inference service | Supported | Complete local direct inference first, then read Section 3. Using VLM Inference Services. |
+| Full API service | Supported with Docker Compose deployment | Read Section 4.1 first, then continue with the Section 4.2 client invocation section and the Section 4.3 pipeline configuration section. |
+| Model fine-tuning | Supported | Read Section 5. Model Fine-Tuning. |
+
+If you only need to confirm which inference methods are available on this hardware, refer to the [PaddleOCR-VL Inference Method and Hardware Support Matrix](./PaddleOCR-VL.en.md#inference-device-support-for-paddleocr-vl) in the main guide.
+
+## 1. Local Runtime Environment Preparation
+
+**Local Runtime Environment Setup Methods Supported on This Hardware**
+
+| Local runtime environment setup method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Supported with steps in this guide | Continue with Section 1.1. |
+| Manually install the inference engine and PaddleOCR | Supported with steps in this guide | Continue with Section 1.2. |
+
+This step mainly introduces how to set up the local runtime environment for PaddleOCR-VL. There are two methods available; choose either one:
+
+- Method 1: Use the official Docker image.
+
+- Method 2: Manually install the inference engine and PaddleOCR.
+
+**We strongly recommend using the Docker image to minimize potential environment-related issues.**
+
+### 1.1 Method 1: Using Docker Image
+
+We recommend using the official Docker image (requires Docker version >= 19.03):
+
+```shell
+docker run -it \
+  --rm \
+  --user root \
+  --privileged \
+  --device /dev/kfd \
+  --device /dev/dri \
+  --device /dev/mkfd \
+  --group-add video \
+  --cap-add SYS_PTRACE \
+  --security-opt seccomp=unconfined \
+  -v /opt/hyhal/:/opt/hyhal/:ro \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-hygon-dcu \
+  /bin/bash
+# Call PaddleOCR CLI or Python API in the container
+```
+
+If you wish to start the service in an environment without internet access, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-hygon-dcu` (image size approximately 22 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-hygon-dcu-offline` (image size approximately 24 GB).
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use an image corresponding to a specific PaddleOCR version, you can replace `latest` in the tag with the desired version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.3-hygon-dcu-offline`
+
+
+### 1.2 Method 2: Manually Install the Inference Engine and PaddleOCR
+
+If you cannot use Docker, you can also manually install the inference engine and PaddleOCR. This guide documents Python 3.9–3.13 as the verified range.
+
+Local inference on this hardware currently supports only the PaddlePaddle inference engine.
+
+**We strongly recommend installing PaddleOCR-VL in a virtual environment to avoid dependency conflicts.** For example, use the Python venv standard library to create a virtual environment:
+
+```shell
+# Create a virtual environment
+python -m venv .venv_paddleocr
+# Activate the environment
+source .venv_paddleocr/bin/activate
+```
+
+Execute the following commands to complete the installation:
+
+```shell
+python -m pip install paddlepaddle-dcu==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/dcu/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **Please note to install PaddlePaddle version 3.2.1 or above.**
+
+## 2. Quick Start
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 2. Quick Start](./PaddleOCR-VL.en.md#2-quick-start), making sure to specify `device="dcu"`.
+
+## 3. Using VLM Inference Services
+
+This section explains how to connect PaddleOCR-VL to a dedicated VLM inference service backend. On this hardware, this is usually used to improve inference performance beyond the default configuration for production use. In this hardware-specific guide, the examples use vLLM as the backend for the VLM inference service.
+
+### 3.1 Starting the VLM Inference Service
+
+> IMPORTANT:
+> The service started according to this section is responsible only for the VLM inference stage in the PaddleOCR-VL workflow. It does not provide a complete end-to-end document parsing API. We strongly recommend that you do not call this service directly via HTTP requests or OpenAI clients to process document images. If you need to deploy a service with the full PaddleOCR-VL capabilities, refer to the service deployment section later in this document.
+
+**Launch Methods Supported on This Hardware**
+
+| Launch method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Supported with steps in this guide | This section provides the vLLM service launch steps. |
+| Install dependencies with the PaddleOCR CLI and launch the service | Not currently supported | This hardware does not currently support this path. |
+| Launch the service directly with the acceleration framework | Not verified | This hardware can start the VLM inference service through the vLLM backend, but launching directly with native vLLM has not been verified. |
+
+PaddleOCR provides a Docker image for quickly starting the vLLM inference service. Use the following command to start the service (requires Docker version >= 19.03):
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  --device /dev/kfd \
+  --device /dev/dri \
+  --device /dev/mkfd \
+  --group-add video \
+  --cap-add SYS_PTRACE \
+  --security-opt seccomp=unconfined \
+  -v /opt/hyhal/:/opt/hyhal/:ro \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-hygon-dcu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm
+```
+
+If you wish to start the service in an environment without internet access, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-hygon-dcu` (image size approximately 25 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-hygon-dcu-offline` (image size approximately 27 GB).
+
+When launching the vLLM inference service, we provide a set of default parameter settings. If you need to adjust parameters such as GPU memory usage, you can configure additional parameters yourself. Please refer to [3.3.1 Server-side Parameter Adjustment](./PaddleOCR-VL.en.md#331-server-side-parameter-adjustment) to create a configuration file, then mount the file into the container and specify the configuration file using `backend_config` in the command to start the service, for example:
+
+```shell
+docker run -it \
+  --rm \
+  --user root \
+  --privileged \
+  --device /dev/kfd \
+  --device /dev/dri \
+  --device /dev/mkfd \
+  --group-add video \
+  --cap-add SYS_PTRACE \
+  --security-opt seccomp=unconfined \
+  -v /opt/hyhal/:/opt/hyhal/:ro \
+  -v ./vllm_config.yml:/tmp/vllm_config.yml \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-dcu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /tmp/vllm_config.yml
+```
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use an image corresponding to a specific PaddleOCR version, you can replace `latest` in the tag with the desired version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:paddleocr3.3-hygon-dcu-offline`
+
+
+### 3.2 Client Usage Method
+
+For client-side invocation methods, please refer to [PaddleOCR-VL Usage Tutorial - 3.2 Client Usage Methods](./PaddleOCR-VL.en.md#32-client-usage-methods). If you run the client on this hardware, make sure to specify `device="dcu"`.
+
+### 3.3 Performance Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 3.3 Performance Tuning](./PaddleOCR-VL.en.md#33-performance-tuning).
+
+## 4. Service Deployment
+
+**Deployment Methods Supported on This Hardware**
+
+| Deployment method | Status | Notes |
+| --- | --- | --- |
+| Docker Compose deployment | Supported with steps in this guide | Continue with Section 4.1. |
+| Manual deployment | Not currently supported | This hardware does not currently support this path. |
+
+> IMPORTANT:
+> The PaddleOCR-VL service introduced in this section differs from the VLM inference service in the previous section: the latter is responsible for only one part of the complete process (i.e., VLM inference) and is called as an underlying service by the former.
+
+### 4.1 Deploy Using Docker Compose
+
+This step mainly introduces how to use Docker Compose to deploy PaddleOCR-VL as a service and call it. The specific process is as follows:
+
+1. Download the Compose file and the environment variable configuration file separately from [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/hygon-dcu/compose.yaml) and [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/hygon-dcu/.env) to your local machine.
+    
+2. Execute the following command in the directory where the `compose.yaml` and `.env` files are located to start the server, which listens on port **8080** by default:
+
+    ```shell
+    # Must be executed in the directory where compose.yaml and .env files are located
+    docker compose up
+    ```
+
+    > TIP:
+    > The image tags used by `compose.yaml` are usually controlled by `API_IMAGE_TAG_SUFFIX` and `VLM_IMAGE_TAG_SUFFIX` in `.env`, and default to tags such as `latest-hygon-dcu-offline`. To make sure you pull the newest `latest` images, run `docker compose pull` in the current directory before `docker compose up`.
+    > To use an image corresponding to a specific PaddleOCR version, replace `latest` in these variables with `paddleocr<major>.<minor>`, for example `paddleocr3.3-hygon-dcu-offline`.
+
+    After startup, you will see output similar to the following:
+
+    ```text
+    paddleocr-vl-api             | INFO:     Started server process [1]
+    paddleocr-vl-api             | INFO:     Waiting for application startup.
+    paddleocr-vl-api             | INFO:     Application startup complete.
+    paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+    ```
+
+This method accelerates VLM inference using the vLLM framework and is more suitable for production environment deployment.
+
+Additionally, after starting the server in this manner, no internet connection is required except for image pulling. For deployment in an offline environment, you can first pull the images involved in the Compose file on a connected machine, export them, and transfer them to the offline machine for import to start the service in an offline environment.
+
+Docker Compose starts two containers sequentially by reading configurations from the `.env` and `compose.yaml` files, running the underlying VLM inference service and the PaddleOCR-VL service (pipeline service) respectively.
+
+The meanings of each environment variable contained in the `.env` file are as follows:
+
+```
+- `API_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to launch the pipeline service.
+- `VLM_BACKEND`: The VLM inference backend.
+- `VLM_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to launch the VLM inference service.
+```
+
+You can modify `compose.yaml` to meet custom requirements, for example:
+
+<details>
+<summary>1. Change the port of the PaddleOCR-VL service</summary>
+
+Edit <code>paddleocr-vl-api.ports</code> in the <code>compose.yaml</code> file to change the port. For example, if you need to change the service port to 8111, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. Specify the DCU used by the PaddleOCR-VL service</summary>
+
+Edit <code>environment</code> in the <code>compose.yaml</code> file to change the DCU used. For example, if you need to use card 1 for deployment, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    environment:
++     - HIP_VISIBLE_DEVICES: 1
+    ...
+  paddleocr-vlm-server:
+    ...
+    environment:
++     - HIP_VISIBLE_DEVICES: 1
+    ...
+```
+
+
+</details>
+
+<details>
+<summary>3. Adjust VLM server-side configuration</summary>
+
+If you want to adjust the VLM server configuration, refer to <a href="./PaddleOCR-VL.en.md#331-server-side-parameter-adjustment">3.3.1 Server-side Parameter Adjustment</a> to generate a configuration file.
+
+After generating the configuration file, add the following <code>paddleocr-vlm-server.volumes</code> and <code>paddleocr-vlm-server.command</code> fields to your <code>compose.yaml</code>. Replace <code>/path/to/your_config.yaml</code> with your actual configuration file path.
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes: /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. Adjust pipeline-related configurations (such as model path, batch size, deployment device, etc.)</summary>
+
+Refer to the <a href="./PaddleOCR-VL.en.md#44-pipeline-configuration-adjustment-instructions">4.4 Pipeline Configuration Adjustment Instructions</a> section.
+
+</details>
+
+### 4.2 Client Invocation Method
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.3 Client-Side Invocation](./PaddleOCR-VL.en.md#43-client-side-invocation).
+
+### 4.3 Pipeline Configuration Adjustment Instructions
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.4 Pipeline Configuration Adjustment Instructions](./PaddleOCR-VL.en.md#44-pipeline-configuration-adjustment-instructions).
+
+## 5. Model Fine-Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 5. Model Fine-Tuning](./PaddleOCR-VL.en.md#5-model-fine-tuning).
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-Hygon-DCU.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Hygon-DCU.md
new file mode 100644
index 0000000..d61bff1
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Hygon-DCU.md
@@ -0,0 +1,299 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL 海光 DCU 使用教程
+
+> INFO:
+> 除非另有说明，本教程中提到的 “PaddleOCR-VL” 均指 PaddleOCR-VL 系列模型（如 PaddleOCR-VL-1.6 等）；若特指 PaddleOCR-VL v1 版本，将另行明确标注。
+
+本教程是 PaddleOCR-VL 在海光 DCU 上的使用指南，涵盖了从本地运行环境准备到服务化部署的完整流程。
+
+目前 PaddleOCR-VL 已在海光 K100AI 上完成精度、速度验证；鉴于硬件环境的多样性，其他海光 DCU 的兼容性尚未验证。我们诚挚欢迎社区用户在不同硬件上进行测试并反馈您的运行结果。
+
+## 本硬件支持的使用目标
+
+请在本硬件教程中按下表继续阅读。
+
+| 目标 | 本硬件上的支持情况 | 从哪里开始阅读 |
+| --- | --- | --- |
+| 本地直接推理 | 支持 | 阅读第 1 节“本地运行环境准备”和第 2 节“快速开始”。 |
+| 客户端 + VLM 推理服务 | 支持 | 先完成本地直接推理，再阅读第 3 节“使用 VLM 推理服务”。 |
+| 完整 API 服务 | 支持 Docker Compose 部署 | 先阅读第 4.1 节，再继续阅读第 4.2 节客户端调用部分和第 4.3 节产线配置调整部分。 |
+| 模型微调 | 支持 | 阅读第 5 节“模型微调”。 |
+
+如果你只是想先确认本硬件支持哪些推理方式，请参考主教程中的 [PaddleOCR-VL 推理方式与硬件支持矩阵](./PaddleOCR-VL.md#paddleocr-vl-对推理设备的支持情况)。
+
+## 1. 本地运行环境准备
+
+**当前硬件支持的本地运行环境准备方式**
+
+| 本地运行环境准备方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 支持并提供步骤 | 请继续阅读本节的 1.1。 |
+| 手动安装推理引擎和 PaddleOCR | 支持并提供步骤 | 请继续阅读本节的 1.2。 |
+
+此步骤主要介绍如何搭建 PaddleOCR-VL 的本地运行环境，有以下两种方式，任选一种即可：
+
+- 方法一：使用官方 Docker 镜像。
+
+- 方法二：手动安装推理引擎和 PaddleOCR。
+
+**我们强烈推荐采用 Docker 镜像的方式，以最大程度减少可能出现的环境问题。**
+
+### 1.1 方法一：使用 Docker 镜像
+
+我们推荐使用官方 Docker 镜像（要求 Docker 版本 >= 19.03）：
+
+```shell
+docker run -it \
+  --rm \
+  --user root \
+  --privileged \
+  --device /dev/kfd \
+  --device /dev/dri \
+  --device /dev/mkfd \
+  --group-add video \
+  --cap-add SYS_PTRACE \
+  --security-opt seccomp=unconfined \
+  -v /opt/hyhal/:/opt/hyhal/:ro \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-hygon-dcu \
+  /bin/bash
+# 在容器中调用 PaddleOCR CLI 或 Python API
+```
+
+如果您希望在无法连接互联网的环境中启动服务，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-hygon-dcu`（镜像的大小约为 22 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-hygon-dcu-offline`（镜像的大小约为 24 GB）。
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本的 PaddleOCR 镜像，可以将标签中的 `latest` 替换为对应版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.3-hygon-dcu-offline`
+
+### 1.2 方法二：手动安装推理引擎和 PaddleOCR
+
+如果您无法使用 Docker，也可以手动安装推理引擎和 PaddleOCR。本文档验证过的 Python 版本范围为 3.9–3.13。
+
+当前硬件本地推理仅支持 PaddlePaddle 推理引擎。
+
+**我们强烈推荐您在虚拟环境中安装 PaddleOCR-VL，以避免发生依赖冲突。** 例如，使用 Python venv 标准库创建虚拟环境：
+
+```shell
+# 创建虚拟环境
+python -m venv .venv_paddleocr
+# 激活环境
+source .venv_paddleocr/bin/activate
+```
+
+执行如下命令完成安装：
+
+```shell
+python -m pip install paddlepaddle-dcu==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/dcu/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **请注意安装 3.2.1 及以上版本的飞桨框架。**
+
+## 2. 快速开始
+
+请参考[PaddleOCR-VL 使用教程 - 2. 快速开始](./PaddleOCR-VL.md#2)，注意需要指定 `device="dcu"`。
+
+## 3. 使用 VLM 推理服务
+
+本节介绍如何通过 VLM 推理服务接入专用后端。对于当前硬件，这通常用于提升默认配置下的推理性能，以更好满足生产需求。在当前硬件文档中，示例使用 vLLM 作为 VLM 推理服务后端。
+
+### 3.1 启动 VLM 推理服务
+
+> IMPORTANT:
+> 按照本节说明启动的服务仅负责 PaddleOCR-VL 流程中的 VLM 推理环节，不提供完整的端到端文档解析 API。强烈不建议直接通过 HTTP 请求或使用 OpenAI 客户端调用该服务处理文档图像。若您需要部署具备 PaddleOCR-VL 完整能力的服务，请参考后文的服务化部署部分。
+
+**当前硬件支持的启动方式**
+
+| 启动方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 支持并提供步骤 | 本节提供 vLLM 推理服务的启动步骤。 |
+| 通过 PaddleOCR CLI 安装依赖后启动 | 当前不支持 | 当前硬件不支持该路径。 |
+| 直接使用推理加速框架启动 | 未验证 | 当前硬件可通过 vLLM 后端启动 VLM 推理服务，但尚未验证直接使用 vLLM 原生方式启动的路径。 |
+
+PaddleOCR 提供了 Docker 镜像，用于快速启动 vLLM 推理服务。可使用以下命令启动服务（要求 Docker 版本 >= 19.03）：
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  --device /dev/kfd \
+  --device /dev/dri \
+  --device /dev/mkfd \
+  --group-add video \
+  --cap-add SYS_PTRACE \
+  --security-opt seccomp=unconfined \
+  -v /opt/hyhal/:/opt/hyhal/:ro \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-hygon-dcu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm
+```
+
+如果您希望在无法连接互联网的环境中启动服务，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-hygon-dcu`（镜像的大小约为 25 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-hygon-dcu-offline`（镜像的大小约为 27 GB）。
+
+启动 vLLM 推理服务时，我们提供了一套默认参数设置。如果您有调整显存占用等更多参数的需求，可以自行配置更多参数。请参考 [3.3.1 服务端参数调整](./PaddleOCR-VL.md#331) 创建配置文件，然后将该文件挂载到容器中，并在启动服务的命令中使用 `backend_config` 指定配置文件，例如：
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  --device /dev/kfd \
+  --device /dev/dri \
+  --device /dev/mkfd \
+  --group-add video \
+  --cap-add SYS_PTRACE \
+  --security-opt seccomp=unconfined \
+  -v /opt/hyhal/:/opt/hyhal/:ro \
+  -v ./vllm_config.yml:/tmp/vllm_config.yml \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-hygon-dcu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /tmp/vllm_config.yml
+```
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本的 PaddleOCR 镜像，可以将标签中的 `latest` 替换为对应版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:paddleocr3.3-hygon-dcu-offline`
+
+### 3.2 客户端使用方法
+
+客户端调用方式请参考 [PaddleOCR-VL 使用教程 - 3.2 客户端使用方法](./PaddleOCR-VL.md#32)。如需在当前硬件上运行客户端，请注意指定 `device="dcu"`。
+
+### 3.3 性能调优
+
+请参考[PaddleOCR-VL 使用教程 - 3.3 性能调优](./PaddleOCR-VL.md#33)。
+
+## 4. 服务化部署
+
+**当前硬件支持的部署方式**
+
+| 部署方式 | 状态 | 说明 |
+| --- | --- | --- |
+| Docker Compose 部署 | 支持并提供步骤 | 请继续阅读本节的 4.1。 |
+| 手动部署 | 当前不支持 | 当前硬件不支持该路径。 |
+
+> IMPORTANT:
+> 本节所介绍的 PaddleOCR-VL 服务与上一节中的 VLM 推理服务有所区别：后者仅负责完整流程中的一个环节（即 VLM 推理），并作为前者的底层服务被调用。
+
+### 4.1 使用 Docker Compose 部署
+
+此步骤主要介绍如何使用 Docker Compose 将 PaddleOCR-VL 部署为服务并调用，具体流程如下：
+
+1. 分别从 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/hygon-dcu/compose.yaml) 和 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/hygon-dcu/.env) 获取 Compose 文件与环境变量配置文件并下载到本地。
+
+2. 在 `compose.yaml` 和 `.env` 文件所在目录下执行以下命令启动服务器，默认监听 **8080** 端口：
+
+    ```shell
+    # 必须在 compose.yaml 和 .env 文件所在的目录中执行
+    docker compose up
+    ```
+
+    > 提示：
+    > `compose.yaml` 中使用的镜像标签由 `.env` 中的 `API_IMAGE_TAG_SUFFIX` 和 `VLM_IMAGE_TAG_SUFFIX` 控制，默认使用 `latest-hygon-dcu-offline` 等标签，其中 `latest` 表示对应服务在当前硬件上的最新镜像版本。如需确保拉取到该 `latest` 标签对应的最新镜像，可先在当前目录执行 `docker compose pull`，再执行 `docker compose up`。
+    > 如果希望使用特定版本的 PaddleOCR 镜像，可将这两个环境变量中的 `latest` 替换为对应版本号 `paddleocr<major>.<minor>`，例如 `paddleocr3.3-hygon-dcu-offline`。
+
+    启动后将看到类似如下输出：
+
+    ```text
+    paddleocr-vl-api             | INFO:     Started server process [1]
+    paddleocr-vl-api             | INFO:     Waiting for application startup.
+    paddleocr-vl-api             | INFO:     Application startup complete.
+    paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+    ```
+
+此方式基于 vLLM 框架对 VLM 推理进行加速，更适合生产环境部署。
+
+此外，使用此方式启动服务器后，除拉取镜像外，无需连接互联网。如需在离线环境中部署，可先在联网机器上拉取 Compose 文件中涉及的镜像，导出并传输至离线机器中导入，即可在离线环境下启动服务。
+
+Docker Compose 通过读取 `.env` 和 `compose.yaml` 文件中配置，先后启动 2 个容器，分别运行底层 VLM 推理服务，以及 PaddleOCR-VL 服务（产线服务）。
+
+`.env` 文件中包含的各环境变量含义如下：
+
+- `API_IMAGE_TAG_SUFFIX`：启动产线服务使用的镜像的标签后缀。
+- `VLM_BACKEND`：VLM 推理后端。
+- `VLM_IMAGE_TAG_SUFFIX`：启动 VLM 推理服务使用的镜像的标签后缀。
+
+您可以通过修改 `compose.yaml` 来满足自定义需求，例如：
+
+<details>
+<summary>1. 更改 PaddleOCR-VL 服务的端口</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>paddleocr-vl-api.ports</code> 来更改端口。例如，如果您需要将服务端口更换为 8111，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. 指定 PaddleOCR-VL 服务所使用的 DCU</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>environment</code> 来更改所使用的 DCU。例如，如果您需要使用卡 1 进行部署，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    environment:
++     - HIP_VISIBLE_DEVICES: 1
+    ...
+  paddleocr-vlm-server:
+    ...
+    environment:
++     - HIP_VISIBLE_DEVICES: 1
+    ...
+```
+
+</details>
+
+<details>
+<summary>3. 调整 VLM 服务端配置</summary>
+
+若您想调整 VLM 服务端的配置，可以参考 <a href="./PaddleOCR-VL.md#331">3.3.1 服务端参数调整</a> 生成配置文件。
+
+生成配置文件后，将以下的 <code>paddleocr-vlm-server.volumes</code> 和 <code>paddleocr-vlm-server.command</code> 字段增加到您的 <code>compose.yaml</code> 中。请将 <code>/path/to/your_config.yaml</code> 替换为您的实际配置文件路径。
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes: /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. 调整产线相关配置（如模型路径、批处理大小、部署设备等）</summary>
+
+参考 <a href="./PaddleOCR-VL.md#44">4.4 产线配置调整说明</a> 小节。
+
+</details>
+
+### 4.2 客户端调用方式
+
+请参考[PaddleOCR-VL 使用教程 - 4.3 客户端调用方式](./PaddleOCR-VL.md#43)。
+
+### 4.3 产线配置调整说明
+
+请参考[PaddleOCR-VL 使用教程 - 4.4 产线配置调整说明](./PaddleOCR-VL.md#44)。
+
+## 5. 模型微调
+
+请参考[PaddleOCR-VL 使用教程 - 5. 模型微调](./PaddleOCR-VL.md#5)。
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-Iluvatar-GPU.en.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Iluvatar-GPU.en.md
new file mode 100644
index 0000000..49d5716
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Iluvatar-GPU.en.md
@@ -0,0 +1,298 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL Iluvatar GPU Usage Tutorial
+
+> INFO:
+> Unless otherwise specified, the term "PaddleOCR-VL" in this tutorial refers to the PaddleOCR-VL model series (e.g., PaddleOCR-VL-1.6). References specific to the PaddleOCR-VL v1 version will be explicitly noted.
+
+This tutorial is a guide for using PaddleOCR-VL on Iluvatar GPU, covering the complete workflow from environment preparation to service deployment.
+
+PaddleOCR-VL has been verified for accuracy and speed on the Iluvatar BI-V150. However, due to hardware diversity, compatibility with other Iluvatar GPUs has not yet been confirmed. We welcome the community to test on different hardware setups and share your results.
+
+## Workflow Guide for This Hardware
+
+Use this guide for the workflows below.
+
+| Goal | Support on this hardware | Read this section |
+| --- | --- | --- |
+| Local direct inference | Supported | Read Section 1. Local Runtime Environment Preparation and Section 2. Quick Start. |
+| Client + VLM inference service | Supported | Complete local direct inference first, then read Section 3. Using VLM Inference Services. |
+| Full API service | Supported with Docker Compose deployment | Read Section 4.1 first, then continue with the Section 4.2 client invocation section and the Section 4.3 pipeline configuration section. |
+| Model fine-tuning | Supported | Read Section 5. Model Fine-Tuning. |
+
+If you only need to confirm which inference methods are available on this hardware, refer to the [PaddleOCR-VL Inference Method and Hardware Support Matrix](./PaddleOCR-VL.en.md#inference-device-support-for-paddleocr-vl) in the main guide.
+
+## 1. Local Runtime Environment Preparation
+
+**Local Runtime Environment Setup Methods Supported on This Hardware**
+
+| Local runtime environment setup method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Supported with steps in this guide | Continue with Section 1.1. |
+| Manually install the inference engine and PaddleOCR | Supported with steps in this guide | Continue with Section 1.2. |
+
+This step mainly introduces how to set up the local runtime environment for PaddleOCR-VL. There are two methods available; choose either one:
+
+- Method 1: Use the official Docker image.
+
+- Method 2: Manually install the inference engine and PaddleOCR.
+
+**We strongly recommend using the Docker image to minimize potential environment-related issues.**
+
+### 1.1 Method 1: Using Docker Image
+
+We recommend using the official Docker image (requires Docker version >= 19.03):
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  -v /usr/src:/usr/src \
+  -v /lib/modules:/lib/modules \
+  -v /dev:/dev \
+  --cap-add SYS_PTRACE \
+  --pid host \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-iluvatar-gpu \
+  /bin/bash
+# Call PaddleOCR CLI or Python API in the container
+```
+
+If you wish to start the service in an environment without internet access, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-iluvatar-gpu` (image size approximately 37 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-iluvatar-gpu-offline` (image size approximately 39 GB).
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use an image corresponding to a specific PaddleOCR version, you can replace `latest` in the tag with the desired version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.4-iluvatar-gpu-offline`
+
+
+### 1.2 Method 2: Manually Install the Inference Engine and PaddleOCR
+
+If you cannot use Docker, you can also manually install the inference engine and PaddleOCR. This guide documents Python 3.9–3.13 as the verified range.
+
+Local inference on this hardware currently supports only the PaddlePaddle inference engine.
+
+**We strongly recommend installing PaddleOCR-VL in a virtual environment to avoid dependency conflicts.** For example, use the Python venv standard library to create a virtual environment:
+
+```shell
+# Create a virtual environment
+python -m venv .venv_paddleocr
+# Activate the environment
+source .venv_paddleocr/bin/activate
+```
+
+Execute the following commands to complete the installation:
+
+```shell
+python -m pip install paddlepaddle==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+python -m pip install paddle-iluvatar-gpu==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/maca/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **Please note to install PaddlePaddle version 3.2.0 or above.**
+
+## 2. Quick Start
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 2. Quick Start](./PaddleOCR-VL.en.md#2-quick-start), making sure to specify `device="iluvatar_gpu"`.
+
+## 3. Using VLM Inference Services
+
+This section explains how to connect PaddleOCR-VL to a dedicated VLM inference service backend. On this hardware, this is usually used to improve inference performance beyond the default configuration for production use. In this hardware-specific guide, the examples use FastDeploy as the backend for the VLM inference service.
+
+### 3.1 Starting the VLM Inference Service
+
+> IMPORTANT:
+> The service started according to this section is responsible only for the VLM inference stage in the PaddleOCR-VL workflow. It does not provide a complete end-to-end document parsing API. We strongly recommend that you do not call this service directly via HTTP requests or OpenAI clients to process document images. If you need to deploy a service with the full PaddleOCR-VL capabilities, refer to the service deployment section later in this document.
+
+**Launch Methods Supported on This Hardware**
+
+| Launch method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Supported with steps in this guide | This section provides the FastDeploy service launch steps. |
+| Install dependencies with the PaddleOCR CLI and launch the service | Not currently supported | This hardware does not currently support this path. |
+| Launch the service directly with the acceleration framework | Not verified | This hardware can start the VLM inference service through the FastDeploy backend, but launching directly with native FastDeploy has not been verified. |
+
+PaddleOCR provides a Docker image for quickly starting the FastDeploy inference service. Use the following command to start the service (requires Docker version >= 19.03):
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  -v /usr/src:/usr/src \
+  -v /lib/modules:/lib/modules \
+  -v /dev:/dev \
+  --cap-add SYS_PTRACE \
+  --pid host \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-iluvatar-gpu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy
+```
+
+If you wish to start the service in an environment without internet access, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-iluvatar-gpu` (image size approximately 38 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-iluvatar-gpu-offline` (image size approximately 40 GB).
+
+When launching the FastDeploy inference service, we provide a set of default parameter settings. If you need to adjust parameters such as GPU memory usage, you can configure additional parameters yourself. Please refer to [3.3.1 Server-side Parameter Adjustment](./PaddleOCR-VL.en.md#331-server-side-parameter-adjustment) to create a configuration file, then mount the file into the container and specify the configuration file using `backend_config` in the command to start the service, for example:
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  -v /usr/src:/usr/src \
+  -v /lib/modules:/lib/modules \
+  -v /dev:/dev \
+  --cap-add SYS_PTRACE \
+  --pid host \
+  --shm-size 64g \
+  --network host \
+  -v ./fastdeploy_config.yml:/tmp/fastdeploy_config.yml \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-iluvatar-gpu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy --backend_config /tmp/fastdeploy_config.yml
+```
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use an image corresponding to a specific PaddleOCR version, you can replace `latest` in the tag with the desired version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:paddleocr3.4-iluvatar-gpu-offline`
+
+
+### 3.2 Client Usage Method
+
+For client-side invocation methods, please refer to [PaddleOCR-VL Usage Tutorial - 3.2 Client Usage Methods](./PaddleOCR-VL.en.md#32-client-usage-methods). If you run the client on this hardware, make sure to specify `device="iluvatar_gpu"`.
+
+### 3.3 Performance Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 3.3 Performance Tuning](./PaddleOCR-VL.en.md#33-performance-tuning).
+
+## 4. Service Deployment
+
+**Deployment Methods Supported on This Hardware**
+
+| Deployment method | Status | Notes |
+| --- | --- | --- |
+| Docker Compose deployment | Supported with steps in this guide | Continue with Section 4.1. |
+| Manual deployment | Not currently supported | This hardware does not currently support this path. |
+
+> IMPORTANT:
+> The PaddleOCR-VL service introduced in this section differs from the VLM inference service in the previous section: the latter is responsible for only one part of the complete process (i.e., VLM inference) and is called as an underlying service by the former.
+
+### 4.1 Deploy Using Docker Compose
+
+This step mainly introduces how to use Docker Compose to deploy PaddleOCR-VL as a service and call it. The specific process is as follows:
+
+1. Download the Compose file and the environment variable configuration file separately from [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/iluvatar-gpu/compose.yaml) and [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/iluvatar-gpu/.env) to your local machine.
+    
+2. Execute the following command in the directory where the `compose.yaml` and `.env` files are located to start the server, which listens on port **8080** by default:
+
+    ```shell
+    # Must be executed in the directory where compose.yaml and .env files are located
+    docker compose up
+    ```
+
+    > TIP:
+    > The image tags used by `compose.yaml` are usually controlled by `API_IMAGE_TAG_SUFFIX` and `VLM_IMAGE_TAG_SUFFIX` in `.env`, and default to tags such as `latest-iluvatar-gpu-offline`. To make sure you pull the newest `latest` images, run `docker compose pull` in the current directory before `docker compose up`.
+    > To use an image corresponding to a specific PaddleOCR version, replace `latest` in these variables with `paddleocr<major>.<minor>`, for example `paddleocr3.3-iluvatar-gpu-offline`.
+
+    After startup, you will see output similar to the following:
+
+    ```text
+    paddleocr-vl-api             | INFO:     Started server process [1]
+    paddleocr-vl-api             | INFO:     Waiting for application startup.
+    paddleocr-vl-api             | INFO:     Application startup complete.
+    paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+    ```
+
+This method accelerates VLM inference using the FastDeploy framework and is more suitable for production environment deployment.
+
+Additionally, after starting the server in this manner, no internet connection is required except for image pulling. For deployment in an offline environment, you can first pull the images involved in the Compose file on a connected machine, export them, and transfer them to the offline machine for import to start the service in an offline environment.
+
+Docker Compose starts two containers sequentially by reading configurations from the `.env` and `compose.yaml` files, running the underlying VLM inference service and the PaddleOCR-VL service (pipeline service) respectively.
+
+The meanings of each environment variable contained in the `.env` file are as follows:
+
+```
+- `API_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to launch the pipeline service.
+- `VLM_BACKEND`: The VLM inference backend.
+- `VLM_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to launch the VLM inference service.
+```
+
+You can modify `compose.yaml` to meet custom requirements, for example:
+
+<details>
+<summary>1. Change the port of the PaddleOCR-VL service</summary>
+
+Edit <code>paddleocr-vl-api.ports</code> in the <code>compose.yaml</code> file to change the port. For example, if you need to change the service port to 8111, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. Specify the GPU used by the PaddleOCR-VL service</summary>
+
+Edit <code>environment</code> in the <code>compose.yaml</code> file to change the GPU used. For example, if you need to use card 1 for deployment, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    environment:
++     - CUDA_VISIBLE_DEVICES: 1
+    ...
+  paddleocr-vlm-server:
+    ...
+    environment:
++     - CUDA_VISIBLE_DEVICES: 1
+    ...
+```
+
+
+</details>
+
+<details>
+<summary>3. Adjust VLM server-side configuration</summary>
+
+If you want to adjust the VLM server configuration, refer to <a href="./PaddleOCR-VL.en.md#331-server-side-parameter-adjustment">3.3.1 Server-side Parameter Adjustment</a> to generate a configuration file.
+
+After generating the configuration file, add the following <code>paddleocr-vlm-server.volumes</code> and <code>paddleocr-vlm-server.command</code> fields to your <code>compose.yaml</code>. Replace <code>/path/to/your_config.yaml</code> with your actual configuration file path.
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes: /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. Adjust pipeline-related configurations (such as model path, batch size, deployment device, etc.)</summary>
+
+Refer to the <a href="./PaddleOCR-VL.en.md#44-pipeline-configuration-adjustment-instructions">4.4 Pipeline Configuration Adjustment Instructions</a> section.
+
+</details>
+
+### 4.2 Client Invocation Method
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.3 Client-Side Invocation](./PaddleOCR-VL.en.md#43-client-side-invocation).
+
+### 4.3 Pipeline Configuration Adjustment Instructions
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.4 Pipeline Configuration Adjustment Instructions](./PaddleOCR-VL.en.md#44-pipeline-configuration-adjustment-instructions).
+
+## 5. Model Fine-Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 5. Model Fine-Tuning](./PaddleOCR-VL.en.md#5-model-fine-tuning).
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-Iluvatar-GPU.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Iluvatar-GPU.md
new file mode 100644
index 0000000..a6b8950
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Iluvatar-GPU.md
@@ -0,0 +1,294 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL 天数 GPU 使用教程
+
+> INFO:
+> 除非另有说明，本教程中提到的 “PaddleOCR-VL” 均指 PaddleOCR-VL 系列模型（如 PaddleOCR-VL-1.6 等）；若特指 PaddleOCR-VL v1 版本，将另行明确标注。
+
+本教程是 PaddleOCR-VL 在天数 GPU 上的使用指南，涵盖了从本地运行环境准备到服务化部署的完整流程。
+
+目前 PaddleOCR-VL 已在天数天垓 150 上完成精度、速度验证；鉴于硬件环境的多样性，其他天数 GPU 的兼容性尚未验证。我们诚挚欢迎社区用户在不同硬件上进行测试并反馈您的运行结果。
+
+## 本硬件支持的使用目标
+
+请在本硬件教程中按下表继续阅读。
+
+| 目标 | 本硬件上的支持情况 | 从哪里开始阅读 |
+| --- | --- | --- |
+| 本地直接推理 | 支持 | 阅读第 1 节“本地运行环境准备”和第 2 节“快速开始”。 |
+| 客户端 + VLM 推理服务 | 支持 | 先完成本地直接推理，再阅读第 3 节“使用 VLM 推理服务”。 |
+| 完整 API 服务 | 支持 Docker Compose 部署 | 先阅读第 4.1 节，再继续阅读第 4.2 节客户端调用部分和第 4.3 节产线配置调整部分。 |
+| 模型微调 | 支持 | 阅读第 5 节“模型微调”。 |
+
+如果你只是想先确认本硬件支持哪些推理方式，请参考主教程中的 [PaddleOCR-VL 推理方式与硬件支持矩阵](./PaddleOCR-VL.md#paddleocr-vl-对推理设备的支持情况)。
+
+## 1. 本地运行环境准备
+
+**当前硬件支持的本地运行环境准备方式**
+
+| 本地运行环境准备方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 支持并提供步骤 | 请继续阅读本节的 1.1。 |
+| 手动安装推理引擎和 PaddleOCR | 支持并提供步骤 | 请继续阅读本节的 1.2。 |
+
+此步骤主要介绍如何搭建 PaddleOCR-VL 的本地运行环境，有以下两种方式，任选一种即可：
+
+- 方法一：使用官方 Docker 镜像。
+
+- 方法二：手动安装推理引擎和 PaddleOCR。
+
+**我们强烈推荐采用 Docker 镜像的方式，以最大程度减少可能出现的环境问题。**
+
+### 1.1 方法一：使用 Docker 镜像
+
+我们推荐使用官方 Docker 镜像（要求 Docker 版本 >= 19.03）：
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  -v /usr/src:/usr/src \
+  -v /lib/modules:/lib/modules \
+  -v /dev:/dev \
+  --cap-add SYS_PTRACE \
+  --pid host \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-iluvatar-gpu \
+  /bin/bash
+# 在容器中调用 PaddleOCR CLI 或 Python API
+```
+
+如果您希望在无法连接互联网的环境中启动服务，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-iluvatar-gpu`（镜像的大小约为 37 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-iluvatar-gpu-offline`（镜像的大小约为 39 GB）。
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本的 PaddleOCR 镜像，可以将标签中的 `latest` 替换为对应版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.4-iluvatar-gpu-offline`
+
+### 1.2 方法二：手动安装推理引擎和 PaddleOCR
+
+如果您无法使用 Docker，也可以手动安装推理引擎和 PaddleOCR。本文档验证过的 Python 版本范围为 3.9–3.13。
+
+当前硬件本地推理仅支持 PaddlePaddle 推理引擎。
+
+**我们强烈推荐您在虚拟环境中安装 PaddleOCR-VL，以避免发生依赖冲突。** 例如，使用 Python venv 标准库创建虚拟环境：
+
+```shell
+# 创建虚拟环境
+python -m venv .venv_paddleocr
+# 激活环境
+source .venv_paddleocr/bin/activate
+```
+
+执行如下命令完成安装：
+
+```shell
+python -m pip install paddlepaddle==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+python -m pip install paddle-iluvatar-gpu==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/ixuca/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **请注意安装 3.2.0 及以上版本的飞桨框架。**
+
+## 2. 快速开始
+
+请参考 [PaddleOCR-VL 使用教程 - 2. 快速开始](./PaddleOCR-VL.md#2)，注意需要指定 `device="iluvatar_gpu"`。
+
+## 3. 使用 VLM 推理服务
+
+本节介绍如何通过 VLM 推理服务接入专用后端。对于当前硬件，这通常用于提升默认配置下的推理性能，以更好满足生产需求。在当前硬件文档中，示例使用 FastDeploy 作为 VLM 推理服务后端。
+
+### 3.1 启动 VLM 推理服务
+
+> IMPORTANT:
+> 按照本节说明启动的服务仅负责 PaddleOCR-VL 流程中的 VLM 推理环节，不提供完整的端到端文档解析 API。强烈不建议直接通过 HTTP 请求或使用 OpenAI 客户端调用该服务处理文档图像。若您需要部署具备 PaddleOCR-VL 完整能力的服务，请参考后文的服务化部署部分。
+
+**当前硬件支持的启动方式**
+
+| 启动方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 支持并提供步骤 | 本节提供 FastDeploy 推理服务的启动步骤。 |
+| 通过 PaddleOCR CLI 安装依赖后启动 | 当前不支持 | 当前硬件不支持该路径。 |
+| 直接使用推理加速框架启动 | 未验证 | 当前硬件可通过 FastDeploy 后端启动 VLM 推理服务，但尚未验证直接使用 FastDeploy 原生方式启动的路径。 |
+
+PaddleOCR 提供了 Docker 镜像，用于快速启动 FastDeploy 推理服务。可使用以下命令启动服务（要求 Docker 版本 >= 19.03）：
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  -v /usr/src:/usr/src \
+  -v /lib/modules:/lib/modules \
+  -v /dev:/dev \
+  --cap-add SYS_PTRACE \
+  --pid host \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-iluvatar-gpu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy
+```
+
+如果您希望在无法连接互联网的环境中启动服务，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-iluvatar-gpu`（镜像的大小约为 38 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-iluvatar-gpu-offline`（镜像的大小约为 40 GB）。
+
+启动 FastDeploy 推理服务时，我们提供了一套默认参数设置。如果您有调整显存占用等更多参数的需求，可以自行配置更多参数。请参考 [3.3.1 服务端参数调整](./PaddleOCR-VL.md#331) 创建配置文件，然后将该文件挂载到容器中，并在启动服务的命令中使用 `backend_config` 指定配置文件，例如：
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  -v /usr/src:/usr/src \
+  -v /lib/modules:/lib/modules \
+  -v /dev:/dev \
+  --cap-add SYS_PTRACE \
+  --pid host \
+  --shm-size 64g \
+  --network host \
+  -v ./fastdeploy_config.yml:/tmp/fastdeploy_config.yml \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-iluvatar-gpu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy --backend_config /tmp/fastdeploy_config.yml
+```
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本的 PaddleOCR 镜像，可以将标签中的 `latest` 替换为对应版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:paddleocr3.4-iluvatar-gpu-offline`
+
+### 3.2 客户端使用方法
+
+客户端调用方式请参考 [PaddleOCR-VL 使用教程 - 3.2 客户端使用方法](./PaddleOCR-VL.md#32)。如需在当前硬件上运行客户端，请注意指定 `device="iluvatar_gpu"`。
+
+### 3.3 性能调优
+
+请参考 [PaddleOCR-VL 使用教程 - 3.3 性能调优](./PaddleOCR-VL.md#33)。
+
+## 4. 服务化部署
+
+**当前硬件支持的部署方式**
+
+| 部署方式 | 状态 | 说明 |
+| --- | --- | --- |
+| Docker Compose 部署 | 支持并提供步骤 | 请继续阅读本节的 4.1。 |
+| 手动部署 | 当前不支持 | 当前硬件不支持该路径。 |
+
+> IMPORTANT:
+> 本节所介绍的 PaddleOCR-VL 服务与上一节中的 VLM 推理服务有所区别：后者仅负责完整流程中的一个环节（即 VLM 推理），并作为前者的底层服务被调用。
+
+### 4.1 使用 Docker Compose 部署
+
+此步骤主要介绍如何使用 Docker Compose 将 PaddleOCR-VL 部署为服务并调用，具体流程如下：
+
+1. 分别从 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/iluvatar-gpu/compose.yaml) 和 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/iluvatar-gpu/.env) 获取 Compose 文件与环境变量配置文件并下载到本地。
+
+2. 在 `compose.yaml` 和 `.env` 文件所在目录下执行以下命令启动服务器，默认监听 **8080** 端口：
+
+    ```shell
+    # 必须在 compose.yaml 和 .env 文件所在的目录中执行
+    docker compose up
+    ```
+
+    > 提示：
+    > `compose.yaml` 中使用的镜像标签通常由 `.env` 中的 `API_IMAGE_TAG_SUFFIX` 和 `VLM_IMAGE_TAG_SUFFIX` 控制，默认使用 `latest-iluvatar-gpu-offline` 等标签。如需确保拉取到最新的 `latest` 镜像，可先在当前目录执行 `docker compose pull`，再执行 `docker compose up`。
+    > 如果希望使用特定版本的 PaddleOCR 镜像，可将这两个环境变量中的 `latest` 替换为对应版本号 `paddleocr<major>.<minor>`，例如 `paddleocr3.3-iluvatar-gpu-offline`。
+
+    启动后将看到类似如下输出：
+
+    ```text
+    paddleocr-vl-api             | INFO:     Started server process [1]
+    paddleocr-vl-api             | INFO:     Waiting for application startup.
+    paddleocr-vl-api             | INFO:     Application startup complete.
+    paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+    ```
+
+此方式基于 FastDeploy 框架对 VLM 推理进行加速，更适合生产环境部署。
+
+此外，使用此方式启动服务器后，除拉取镜像外，无需连接互联网。如需在离线环境中部署，可先在联网机器上拉取 Compose 文件中涉及的镜像，导出并传输至离线机器中导入，即可在离线环境下启动服务。
+
+Docker Compose 通过读取 `.env` 和 `compose.yaml` 文件中配置，先后启动 2 个容器，分别运行底层 VLM 推理服务，以及 PaddleOCR-VL 服务（产线服务）。
+
+`.env` 文件中包含的各环境变量含义如下：
+
+- `API_IMAGE_TAG_SUFFIX`：启动产线服务使用的镜像的标签后缀。
+- `VLM_BACKEND`：VLM 推理后端。
+- `VLM_IMAGE_TAG_SUFFIX`：启动 VLM 推理服务使用的镜像的标签后缀。
+
+您可以通过修改 `compose.yaml` 来满足自定义需求，例如：
+
+<details>
+<summary>1. 更改 PaddleOCR-VL 服务的端口</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>paddleocr-vl-api.ports</code> 来更改端口。例如，如果您需要将服务端口更换为 8111，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. 指定 PaddleOCR-VL 服务所使用的 GPU</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>environment</code> 来更改所使用的 GPU。例如，如果您需要使用卡 1 进行部署，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    environment:
++     - CUDA_VISIBLE_DEVICES: 1
+    ...
+  paddleocr-vlm-server:
+    ...
+    environment:
++     - CUDA_VISIBLE_DEVICES: 1
+    ...
+```
+
+
+</details>
+
+<details>
+<summary>3. 调整 VLM 服务端配置</summary>
+
+若您想调整 VLM 服务端的配置，可以参考 <a href="./PaddleOCR-VL.md#331">3.3.1 服务端参数调整</a> 生成配置文件。
+
+生成配置文件后，将以下的 <code>paddleocr-vlm-server.volumes</code> 和 <code>paddleocr-vlm-server.command</code> 字段增加到您的 <code>compose.yaml</code> 中。请将 <code>/path/to/your_config.yaml</code> 替换为您的实际配置文件路径。
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes: /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. 调整产线相关配置（如模型路径、批处理大小、部署设备等）</summary>
+
+参考 <a href="./PaddleOCR-VL.md#44">4.4 产线配置调整说明</a> 小节。
+
+</details>
+
+### 4.2 客户端调用方式
+
+请参考 [PaddleOCR-VL 使用教程 - 4.3 客户端调用方式](./PaddleOCR-VL.md#43)。
+
+### 4.3 产线配置调整说明
+
+请参考 [PaddleOCR-VL 使用教程 - 4.4 产线配置调整说明](./PaddleOCR-VL.md#44)。
+
+## 5. 模型微调
+
+请参考 [PaddleOCR-VL 使用教程 - 5. 模型微调](./PaddleOCR-VL.md#5)。
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-Intel-Arc-GPU.en.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Intel-Arc-GPU.en.md
new file mode 100644
index 0000000..ca2cccd
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Intel-Arc-GPU.en.md
@@ -0,0 +1,284 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL Intel Arc GPU Usage Tutorial
+
+> INFO:
+> Unless otherwise specified, the term "PaddleOCR-VL" in this tutorial refers to the PaddleOCR-VL model series (e.g., PaddleOCR-VL-1.6). References specific to the PaddleOCR-VL v1 version will be explicitly noted.
+
+This tutorial is a guide for using PaddleOCR-VL on Intel Arc GPU, covering the complete workflow from environment preparation to service deployment.
+
+PaddleOCR-VL has been verified for accuracy and speed on the Intel Arc B60 Pro. However, due to hardware diversity, compatibility with other Intel Arc GPUs has not yet been confirmed. We welcome the community to test on different hardware setups and share your results.
+
+## Workflow Guide for This Hardware
+
+Use this guide for the workflows below.
+
+| Goal | Support on this hardware | Read this section |
+| --- | --- | --- |
+| Local direct inference | This hardware does not currently support the local direct inference path | Use the “Client + VLM inference service” path instead: start with Section 1. Local Runtime Environment Preparation, then read Section 3. |
+| Client + VLM inference service | Supported | Start with Section 1. Local Runtime Environment Preparation, then read Section 3. Using VLM Inference Services. |
+| Full API service | Supported with Docker Compose deployment | Read Section 4.1 first, then continue with the Section 4.2 client invocation section and the Section 4.3 pipeline configuration section. |
+| Model fine-tuning | Supported | Read Section 5. Model Fine-Tuning. |
+
+If you only need to confirm which inference methods are available on this hardware, refer to the [PaddleOCR-VL Inference Method and Hardware Support Matrix](./PaddleOCR-VL.en.md#inference-device-support-for-paddleocr-vl) in the main guide.
+
+## 1. Local Runtime Environment Preparation
+
+**Local Runtime Environment Setup Methods Supported on This Hardware**
+
+| Local runtime environment setup method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Supported with steps in this guide | Continue with Section 1.1. |
+| Manually install the inference engine and PaddleOCR | Supported with steps in this guide | Continue with Section 1.2. |
+
+This step mainly introduces how to set up the local runtime environment for PaddleOCR-VL. There are two methods available; choose either one:
+
+- Method 1: Use the official Docker image.
+
+- Method 2: Manually install the inference engine and PaddleOCR.
+
+**We strongly recommend using the Docker image to minimize potential environment-related issues.**
+
+### 1.1 Method 1: Using Docker Image
+
+We recommend using the official Docker image (requires Docker version >= 19.03):
+
+```shell
+docker run -it \
+  --user root \
+  --device /dev:/dev \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-intel-gpu \
+  /bin/bash
+# Call PaddleOCR CLI or Python API in the container
+```
+
+If you wish to start the service in an environment without internet access, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-intel-gpu` (image size approximately 31 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-intel-gpu-offline` (image size approximately 33 GB).
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use an image corresponding to a specific PaddleOCR version, you can replace `latest` in the tag with the desired version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.4-intel-gpu-offline`
+
+
+### 1.2 Method 2: Manually Install the Inference Engine and PaddleOCR
+
+If you cannot use Docker, you can also manually install the inference engine and PaddleOCR. This guide documents Python 3.9–3.13 as the verified range.
+
+Local inference on this hardware currently provides PaddlePaddle installation steps only; support for other inference engines is still being validated.
+
+**We strongly recommend installing PaddleOCR-VL in a virtual environment to avoid dependency conflicts.** For example, use the Python venv standard library to create a virtual environment:
+
+```shell
+# Create a virtual environment
+python -m venv .venv_paddleocr
+# Activate the environment
+source .venv_paddleocr/bin/activate
+```
+
+Execute the following commands to complete the installation:
+
+```shell
+python -m pip install paddlepaddle==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **Please note to install PaddlePaddle version 3.2.1 or above.**
+
+## 2. Quick Start
+
+This hardware does not currently support the local direct inference path. To use the supported accelerated path on this hardware, continue to the next section and use the vLLM inference service path.
+
+## 3. Using VLM Inference Services
+
+This section explains how to complete the client + VLM inference service path through a VLM inference service. In this hardware-specific guide, the examples use vLLM as the backend for the VLM inference service.
+
+### 3.1 Starting the VLM Inference Service
+
+> IMPORTANT:
+> The service started according to this section is responsible only for the VLM inference stage in the PaddleOCR-VL workflow. It does not provide a complete end-to-end document parsing API. We strongly recommend that you do not call this service directly via HTTP requests or OpenAI clients to process document images. If you need to deploy a service with the full PaddleOCR-VL capabilities, refer to the service deployment section later in this document.
+
+**Launch Methods Supported on This Hardware**
+
+| Launch method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Supported with steps in this guide | This section provides the vLLM service launch steps. |
+| Install dependencies with the PaddleOCR CLI and launch the service | Not currently supported | This hardware does not currently support this path. |
+| Launch the service directly with the acceleration framework | Not verified | This hardware can start the VLM inference service through the vLLM backend, but launching directly with native vLLM has not been verified. |
+
+PaddleOCR provides a Docker image for quickly starting the vLLM inference service. Use the following command to start the service (requires Docker version >= 19.03):
+
+```shell
+docker run -it \
+  --name paddleocr_vllm \
+  --user root \
+  --device /dev:/dev \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-intel-gpu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm
+```
+
+If you wish to start the service in an environment without internet access, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-intel-gpu` (image size approximately 30 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-intel-gpu-offline` (image size approximately 32 GB).
+
+When launching the vLLM inference service, we provide a set of default parameter settings. If you need to adjust parameters such as GPU memory usage, you can configure additional parameters yourself. Please refer to [3.3.1 Server-side Parameter Adjustment](./PaddleOCR-VL.en.md#331-server-side-parameter-adjustment) to create a configuration file, then mount the file into the container and specify the configuration file using `backend_config` in the command to start the service, for example:
+
+```shell
+docker run -it \
+  --name paddleocr_vllm \
+  --user root \
+  --device /dev:/dev \
+  --shm-size 64g \
+  --network host \
+  -v ./vllm_config.yml:/tmp/vllm_config.yml \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-intel-gpu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /tmp/vllm_config.yml
+```
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use an image corresponding to a specific PaddleOCR version, you can replace `latest` in the tag with the desired version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:paddleocr3.4-intel-gpu-offline`
+
+
+### 3.2 Client Usage Method
+
+For client-side invocation methods, please refer to [PaddleOCR-VL Usage Tutorial - 3.2 Client Usage Methods](./PaddleOCR-VL.en.md#32-client-usage-methods).
+
+### 3.3 Performance Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 3.3 Performance Tuning](./PaddleOCR-VL.en.md#33-performance-tuning).
+
+## 4. Service Deployment
+
+**Deployment Methods Supported on This Hardware**
+
+| Deployment method | Status | Notes |
+| --- | --- | --- |
+| Docker Compose deployment | Supported with steps in this guide | Continue with Section 4.1. |
+| Manual deployment | Not currently supported | This hardware does not currently support this path. |
+
+> IMPORTANT:
+> The PaddleOCR-VL service introduced in this section differs from the VLM inference service in the previous section: the latter is responsible for only one part of the complete process (i.e., VLM inference) and is called as an underlying service by the former.
+
+### 4.1 Deploy Using Docker Compose
+
+This step mainly introduces how to use Docker Compose to deploy PaddleOCR-VL as a service and call it. The specific process is as follows:
+
+1. Download the Compose file and the environment variable configuration file separately from [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/intel-gpu/compose.yaml) and [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/intel-gpu/.env) to your local machine.
+    
+2. Execute the following command in the directory where the `compose.yaml` and `.env` files are located to start the server, which listens on port **8080** by default:
+
+    ```shell
+    # Must be executed in the directory where compose.yaml and .env files are located
+    docker compose up
+    ```
+
+    > TIP:
+    > The image tags used by `compose.yaml` are usually controlled by `API_IMAGE_TAG_SUFFIX` and `VLM_IMAGE_TAG_SUFFIX` in `.env`, and default to tags such as `latest-intel-gpu-offline`. To make sure you pull the newest `latest` images, run `docker compose pull` in the current directory before `docker compose up`.
+    > To use an image corresponding to a specific PaddleOCR version, replace `latest` in these variables with `paddleocr<major>.<minor>`, for example `paddleocr3.3-intel-gpu-offline`.
+
+    After startup, you will see output similar to the following:
+
+    ```text
+    paddleocr-vl-api             | INFO:     Started server process [1]
+    paddleocr-vl-api             | INFO:     Waiting for application startup.
+    paddleocr-vl-api             | INFO:     Application startup complete.
+    paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+    ```
+
+This method accelerates VLM inference using the vLLM framework and is more suitable for production environment deployment.
+
+Additionally, after starting the server in this manner, no internet connection is required except for image pulling. For deployment in an offline environment, you can first pull the images involved in the Compose file on a connected machine, export them, and transfer them to the offline machine for import to start the service in an offline environment.
+
+Docker Compose starts two containers sequentially by reading configurations from the `.env` and `compose.yaml` files, running the underlying VLM inference service and the PaddleOCR-VL service (pipeline service) respectively.
+
+The meanings of each environment variable contained in the `.env` file are as follows:
+
+```
+- `API_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to launch the pipeline service.
+- `VLM_BACKEND`: The VLM inference backend.
+- `VLM_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to launch the VLM inference service.
+```
+
+You can modify `compose.yaml` to meet custom requirements, for example:
+
+<details>
+<summary>1. Change the port of the PaddleOCR-VL service</summary>
+
+Edit <code>paddleocr-vl-api.ports</code> in the <code>compose.yaml</code> file to change the port. For example, if you need to change the service port to 8111, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. Specify the GPU used by the PaddleOCR-VL service</summary>
+
+Edit <code>environment</code> in the <code>compose.yaml</code> file to change the GPU used. For example, if you need to use card 1 for deployment, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    environment:
++     - XPU_AFFINITY_MASK: 1
+    ...
+  paddleocr-vlm-server:
+    ...
+    environment:
++     - XPU_AFFINITY_MASK: 1
+    ...
+```
+
+
+</details>
+
+<details>
+<summary>3. Adjust VLM server-side configuration</summary>
+
+If you want to adjust the VLM server configuration, refer to <a href="./PaddleOCR-VL.en.md#331-server-side-parameter-adjustment">3.3.1 Server-side Parameter Adjustment</a> to generate a configuration file.
+
+After generating the configuration file, add the following <code>paddleocr-vlm-server.volumes</code> and <code>paddleocr-vlm-server.command</code> fields to your <code>compose.yaml</code>. Replace <code>/path/to/your_config.yaml</code> with your actual configuration file path.
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes: /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. Adjust pipeline-related configurations (such as model path, batch size, deployment device, etc.)</summary>
+
+Refer to the <a href="./PaddleOCR-VL.en.md#44-pipeline-configuration-adjustment-instructions">4.4 Pipeline Configuration Adjustment Instructions</a> section.
+
+</details>
+
+### 4.2 Client Invocation Method
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.3 Client-Side Invocation](./PaddleOCR-VL.en.md#43-client-side-invocation).
+
+### 4.3 Pipeline Configuration Adjustment Instructions
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.4 Pipeline Configuration Adjustment Instructions](./PaddleOCR-VL.en.md#44-pipeline-configuration-adjustment-instructions).
+
+## 5. Model Fine-Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 5. Model Fine-Tuning](./PaddleOCR-VL.en.md#5-model-fine-tuning).
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-Intel-Arc-GPU.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Intel-Arc-GPU.md
new file mode 100644
index 0000000..038bf1a
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Intel-Arc-GPU.md
@@ -0,0 +1,280 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL Intel Arc GPU 使用教程
+
+> INFO:
+> 除非另有说明，本教程中提到的 “PaddleOCR-VL” 均指 PaddleOCR-VL 系列模型（如 PaddleOCR-VL-1.6 等）；若特指 PaddleOCR-VL v1 版本，将另行明确标注。
+
+本教程是 PaddleOCR-VL 在 Intel Arc GPU 上的使用指南，涵盖了从本地运行环境准备到服务化部署的完整流程。
+
+目前 PaddleOCR-VL 已在 Intel Arc B60 Pro 上完成精度、速度验证；鉴于硬件环境的多样性，其他 Intel Arc GPU 的兼容性尚未验证。我们诚挚欢迎社区用户在不同硬件上进行测试并反馈您的运行结果。
+
+## 本硬件支持的使用目标
+
+请在本硬件教程中按下表继续阅读。
+
+| 目标 | 本硬件上的支持情况 | 从哪里开始阅读 |
+| --- | --- | --- |
+| 本地直接推理 | 当前不支持“本地直接推理”路径 | 请改走“客户端 + VLM 推理服务”路径：从第 1 节“本地运行环境准备”开始，然后阅读第 3 节。 |
+| 客户端 + VLM 推理服务 | 支持 | 从第 1 节“本地运行环境准备”开始，然后阅读第 3 节“使用 VLM 推理服务”。 |
+| 完整 API 服务 | 支持 Docker Compose 部署 | 先阅读第 4.1 节，再继续阅读第 4.2 节客户端调用部分和第 4.3 节产线配置调整部分。 |
+| 模型微调 | 支持 | 阅读第 5 节“模型微调”。 |
+
+如果你只是想先确认本硬件支持哪些推理方式，请参考主教程中的 [PaddleOCR-VL 推理方式与硬件支持矩阵](./PaddleOCR-VL.md#paddleocr-vl-对推理设备的支持情况)。
+
+## 1. 本地运行环境准备
+
+**当前硬件支持的本地运行环境准备方式**
+
+| 本地运行环境准备方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 支持并提供步骤 | 请继续阅读本节的 1.1。 |
+| 手动安装推理引擎和 PaddleOCR | 支持并提供步骤 | 请继续阅读本节的 1.2。 |
+
+此步骤主要介绍如何搭建 PaddleOCR-VL 的本地运行环境，有以下两种方式，任选一种即可：
+
+- 方法一：使用官方 Docker 镜像。
+
+- 方法二：手动安装推理引擎和 PaddleOCR。
+
+**我们强烈推荐采用 Docker 镜像的方式，以最大程度减少可能出现的环境问题。**
+
+### 1.1 方法一：使用 Docker 镜像
+
+我们推荐使用官方 Docker 镜像（要求 Docker 版本 >= 19.03）：
+
+```shell
+docker run -it \
+  --user root \
+  --device /dev:/dev \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-intel-gpu \
+  /bin/bash
+# 在容器中调用 PaddleOCR CLI 或 Python API
+```
+
+如果您希望在无法连接互联网的环境中启动服务，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-intel-gpu`（镜像的大小约为 31 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-intel-gpu-offline`（镜像的大小约为 33 GB）。
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本 PaddleOCR 对应的镜像，可以将标签中的 `latest` 替换为对应版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.4-intel-gpu-offline`
+
+### 1.2 方法二：手动安装推理引擎和 PaddleOCR
+
+如果您无法使用 Docker，也可以手动安装推理引擎和 PaddleOCR。本文档验证过的 Python 版本范围为 3.9–3.13。
+
+当前硬件本地推理仅提供 PaddlePaddle 安装步骤，其他推理引擎尚在适配验证中。
+
+**我们强烈推荐您在虚拟环境中安装 PaddleOCR-VL，以避免发生依赖冲突。** 例如，使用 Python venv 标准库创建虚拟环境：
+
+```shell
+# 创建虚拟环境
+python -m venv .venv_paddleocr
+# 激活环境
+source .venv_paddleocr/bin/activate
+```
+
+执行如下命令完成安装：
+
+```shell
+python -m pip install paddlepaddle==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **请注意安装 3.2.0 及以上版本的飞桨框架。**
+
+## 2. 快速开始
+
+当前硬件暂不支持“本地直接推理”路径。如需使用当前硬件加速方案，请继续阅读下一节，采用 vLLM 推理服务路径。
+
+## 3. 使用 VLM 推理服务
+
+本节介绍如何通过 VLM 推理服务完成“客户端 + VLM 推理服务”的组合路径。在当前硬件文档中，示例使用 vLLM 作为 VLM 推理服务后端。
+
+### 3.1 启动 VLM 推理服务
+
+> IMPORTANT:
+> 按照本节说明启动的服务仅负责 PaddleOCR-VL 流程中的 VLM 推理环节，不提供完整的端到端文档解析 API。强烈不建议直接通过 HTTP 请求或使用 OpenAI 客户端调用该服务处理文档图像。若您需要部署具备 PaddleOCR-VL 完整能力的服务，请参考后文的服务化部署部分。
+
+**当前硬件支持的启动方式**
+
+| 启动方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 支持并提供步骤 | 本节提供 vLLM 推理服务的启动步骤。 |
+| 通过 PaddleOCR CLI 安装依赖后启动 | 当前不支持 | 当前硬件不支持该路径。 |
+| 直接使用推理加速框架启动 | 未验证 | 当前硬件可通过 vLLM 后端启动 VLM 推理服务，但尚未验证直接使用 vLLM 原生方式启动的路径。 |
+
+PaddleOCR 提供了 Docker 镜像，用于快速启动 vLLM 推理服务。可使用以下命令启动服务（要求 Docker 版本 >= 19.03）：
+
+```shell
+docker run -it \
+  --name paddleocr_vllm \
+  --user root \
+  --device /dev:/dev \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-intel-gpu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm
+```
+
+如果您希望在无法连接互联网的环境中启动服务，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-intel-gpu`（镜像的大小约为 30 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-intel-gpu-offline`（镜像的大小约为 32 GB）。
+
+启动 vLLM 推理服务时，我们提供了一套默认参数设置。如果您有调整显存占用等更多参数的需求，可以自行配置更多参数。请参考 [3.3.1 服务端参数调整](./PaddleOCR-VL.md#331) 创建配置文件，然后将该文件挂载到容器中，并在启动服务的命令中使用 `backend_config` 指定配置文件，例如：
+
+```shell
+docker run -it \
+  --name paddleocr_vllm \
+  --user root \
+  --device /dev:/dev \
+  --shm-size 64g \
+  --network host \
+  -v ./vllm_config.yml:/tmp/vllm_config.yml \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-intel-gpu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /tmp/vllm_config.yml
+```
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本 PaddleOCR 对应的镜像，可以将标签中的 `latest` 替换为对应版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:paddleocr3.4-intel-gpu-offline`
+
+### 3.2 客户端使用方法
+
+客户端调用方式请参考 [PaddleOCR-VL 使用教程 - 3.2 客户端使用方法](./PaddleOCR-VL.md#32)。
+
+### 3.3 性能调优
+
+请参考[PaddleOCR-VL 使用教程 - 3.3 性能调优](./PaddleOCR-VL.md#33)。
+
+## 4. 服务化部署
+
+**当前硬件支持的部署方式**
+
+| 部署方式 | 状态 | 说明 |
+| --- | --- | --- |
+| Docker Compose 部署 | 支持并提供步骤 | 请继续阅读本节的 4.1。 |
+| 手动部署 | 当前不支持 | 当前硬件不支持该路径。 |
+
+> IMPORTANT:
+> 本节所介绍的 PaddleOCR-VL 服务与上一节中的 VLM 推理服务有所区别：后者仅负责完整流程中的一个环节（即 VLM 推理），并作为前者的底层服务被调用。
+
+### 4.1 使用 Docker Compose 部署
+
+此步骤主要介绍如何使用 Docker Compose 将 PaddleOCR-VL 部署为服务并调用，具体流程如下：
+
+1. 分别从 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/intel-gpu/compose.yaml) 和 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/intel-gpu/.env) 获取 Compose 文件与环境变量配置文件并下载到本地。
+
+2. 在 `compose.yaml` 和 `.env` 文件所在目录下执行以下命令启动服务器，默认监听 **8080** 端口：
+
+    ```shell
+    # 必须在 compose.yaml 和 .env 文件所在的目录中执行
+    docker compose up
+    ```
+
+    > 提示：
+    > `compose.yaml` 中使用的镜像标签通常由 `.env` 中的 `API_IMAGE_TAG_SUFFIX` 和 `VLM_IMAGE_TAG_SUFFIX` 控制，默认使用 `latest-intel-gpu-offline` 等标签。如需确保拉取到最新的 `latest` 镜像，可先在当前目录执行 `docker compose pull`，再执行 `docker compose up`。
+    > 如果希望使用特定版本 PaddleOCR 对应的镜像，可将这两个环境变量中的 `latest` 替换为对应版本号 `paddleocr<major>.<minor>`，例如 `paddleocr3.3-intel-gpu-offline`。
+
+    启动后将看到类似如下输出：
+
+    ```text
+    paddleocr-vl-api             | INFO:     Started server process [1]
+    paddleocr-vl-api             | INFO:     Waiting for application startup.
+    paddleocr-vl-api             | INFO:     Application startup complete.
+    paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+    ```
+
+此方式基于 vLLM 框架对 VLM 推理进行加速，更适合生产环境部署。
+
+此外，使用此方式启动服务器后，除拉取镜像外，无需连接互联网。如需在离线环境中部署，可先在联网机器上拉取 Compose 文件中涉及的镜像，导出并传输至离线机器中导入，即可在离线环境下启动服务。
+
+Docker Compose 通过读取 `.env` 和 `compose.yaml` 文件中配置，先后启动 2 个容器，分别运行底层 VLM 推理服务，以及 PaddleOCR-VL 服务（产线服务）。
+
+`.env` 文件中包含的各环境变量含义如下：
+
+- `API_IMAGE_TAG_SUFFIX`：启动产线服务使用的镜像的标签后缀。
+- `VLM_BACKEND`：VLM 推理后端。
+- `VLM_IMAGE_TAG_SUFFIX`：启动 VLM 推理服务使用的镜像的标签后缀。
+
+您可以通过修改 `compose.yaml` 来满足自定义需求，例如：
+
+<details>
+<summary>1. 更改 PaddleOCR-VL 服务的端口</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>paddleocr-vl-api.ports</code> 来更改端口。例如，如果您需要将服务端口更换为 8111，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. 指定 PaddleOCR-VL 服务所使用的 GPU</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>environment</code> 来更改所使用的 GPU。例如，如果您需要使用卡 1 进行部署，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    environment:
++     - XPU_AFFINITY_MASK: 1
+    ...
+  paddleocr-vlm-server:
+    ...
+    environment:
++     - XPU_AFFINITY_MASK: 1
+    ...
+```
+
+
+</details>
+
+<details>
+<summary>3. 调整 VLM 服务端配置</summary>
+
+若您想调整 VLM 服务端的配置，可以参考 <a href="./PaddleOCR-VL.md#331">3.3.1 服务端参数调整</a> 生成配置文件。
+
+生成配置文件后，将以下的 <code>paddleocr-vlm-server.volumes</code> 和 <code>paddleocr-vlm-server.command</code> 字段增加到您的 <code>compose.yaml</code> 中。请将 <code>/path/to/your_config.yaml</code> 替换为您的实际配置文件路径。
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes: /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. 调整产线相关配置（如模型路径、批处理大小、部署设备等）</summary>
+
+参考 <a href="./PaddleOCR-VL.md#44">4.4 产线配置调整说明</a> 小节。
+
+</details>
+
+### 4.2 客户端调用方式
+
+请参考[PaddleOCR-VL 使用教程 - 4.3 客户端调用方式](./PaddleOCR-VL.md#43)。
+
+### 4.3 产线配置调整说明
+
+请参考[PaddleOCR-VL 使用教程 - 4.4 产线配置调整说明](./PaddleOCR-VL.md#44)。
+
+## 5. 模型微调
+
+请参考[PaddleOCR-VL 使用教程 - 5. 模型微调](./PaddleOCR-VL.md#5)。
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-Kunlunxin-XPU.en.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Kunlunxin-XPU.en.md
new file mode 100644
index 0000000..7bfc12f
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Kunlunxin-XPU.en.md
@@ -0,0 +1,285 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL Kunlunxin XPU Usage Tutorial
+
+> INFO:
+> Unless otherwise specified, the term "PaddleOCR-VL" in this tutorial refers to the PaddleOCR-VL model series (e.g., PaddleOCR-VL-1.6). References specific to the PaddleOCR-VL v1 version will be explicitly noted.
+
+This tutorial is a guide for using PaddleOCR-VL on Kunlunxin XPU, covering the complete workflow from environment preparation to service deployment.
+
+PaddleOCR-VL has been verified for accuracy and speed on the Kunlunxin P800. However, due to hardware diversity, compatibility with other Kunlunxin XPUs has not yet been confirmed. We welcome the community to test on different hardware setups and share your results.
+
+## Workflow Guide for This Hardware
+
+Use this guide for the workflows below.
+
+| Goal | Support on this hardware | Read this section |
+| --- | --- | --- |
+| Local direct inference | Supported | Read Section 1. Local Runtime Environment Preparation and Section 2. Quick Start. |
+| Client + VLM inference service | Supported | Complete local direct inference first, then read Section 3. Using VLM Inference Services. |
+| Full API service | Supported with Docker Compose deployment | Read Section 4.1 first, then continue with the Section 4.2 client invocation section and the Section 4.3 pipeline configuration section. |
+| Model fine-tuning | Supported | Read Section 5. Model Fine-Tuning. |
+
+If you only need to confirm which inference methods are available on this hardware, refer to the [PaddleOCR-VL Inference Method and Hardware Support Matrix](./PaddleOCR-VL.en.md#inference-device-support-for-paddleocr-vl) in the main guide.
+
+## 1. Local Runtime Environment Preparation
+
+**Local Runtime Environment Setup Methods Supported on This Hardware**
+
+| Local runtime environment setup method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Supported with steps in this guide | Continue with Section 1.1. |
+| Manually install the inference engine and PaddleOCR | Supported with steps in this guide | Continue with Section 1.2. |
+
+This step mainly introduces how to set up the local runtime environment for PaddleOCR-VL. There are two methods available; choose one as needed:
+
+- Method 1: Use the official Docker image.
+
+- Method 2: Manually install the inference engine and PaddleOCR.
+
+**We strongly recommend using the Docker image to minimize potential environment-related issues.**
+
+### 1.1 Method 1: Using Docker Image
+
+We recommend using the official Docker image (requires Docker version >= 19.03):
+
+```shell
+docker run \
+    -it \
+    --network host \
+    --user root \
+    --privileged \
+    --shm-size 64g \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-kunlunxin-xpu \
+    /bin/bash
+# Call PaddleOCR CLI or Python API in the container
+```
+
+If you wish to start the service in an environment without internet access, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-kunlunxin-xpu` (image size approximately 13 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-kunlunxin-xpu-offline` (image size approximately 15 GB).
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use an image corresponding to a specific PaddleOCR version, you can replace `latest` in the tag with the desired version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.3-kunlunxin-xpu-offline`
+
+
+### 1.2 Method 2: Manually Install the Inference Engine and PaddleOCR
+
+If you cannot use Docker, you can also manually install the inference engine and PaddleOCR. This guide documents Python 3.9–3.13 as the verified range.
+
+Local inference on this hardware currently supports only the PaddlePaddle inference engine.
+
+**We strongly recommend installing PaddleOCR-VL in a virtual environment to avoid dependency conflicts.** For example, use the Python venv standard library to create a virtual environment:
+
+```shell
+# Create a virtual environment
+python -m venv .venv_paddleocr
+# Activate the environment
+source .venv_paddleocr/bin/activate
+```
+
+Execute the following commands to complete the installation:
+
+```shell
+python -m pip install paddlepaddle-xpu==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/xpu-p800/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **Please ensure to install PaddlePaddle version 3.2.1 or above.**
+
+## 2. Quick Start
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 2. Quick Start](./PaddleOCR-VL.en.md#2-quick-start), making sure to specify `device='xpu'`.
+
+## 3. Using VLM Inference Services
+
+This section explains how to connect PaddleOCR-VL to a dedicated VLM inference service backend. On this hardware, this is usually used to improve inference performance beyond the default configuration for production use. In this hardware-specific guide, the examples use FastDeploy as the backend for the VLM inference service.
+
+### 3.1 Starting the VLM Inference Service
+
+> IMPORTANT:
+> The service started according to this section is responsible only for the VLM inference stage in the PaddleOCR-VL workflow. It does not provide a complete end-to-end document parsing API. We strongly recommend that you do not call this service directly via HTTP requests or OpenAI clients to process document images. If you need to deploy a service with the full PaddleOCR-VL capabilities, refer to the service deployment section later in this document.
+
+**Launch Methods Supported on This Hardware**
+
+| Launch method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Supported with steps in this guide | This section provides the FastDeploy service launch steps. |
+| Install dependencies with the PaddleOCR CLI and launch the service | Not currently supported | This hardware does not currently support this path. |
+| Launch the service directly with the acceleration framework | Not verified | This hardware can start the VLM inference service through the FastDeploy backend, but launching directly with native FastDeploy has not been verified. |
+
+PaddleOCR provides a Docker image for quickly starting the FastDeploy inference service. Use the following command to start the service (requires Docker version >= 19.03):
+
+```shell
+docker run \
+    -it \
+    --network host \
+    --user root \
+    --privileged \
+    --shm-size 64g \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-kunlunxin-xpu \
+    paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy
+```
+
+If you wish to start the service in an environment without internet access, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-kunlunxin-xpu` (image size approximately 53 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-kunlunxin-xpu-offline` (image size approximately 55 GB).
+
+When launching the FastDeploy inference service, we provide a set of default parameter settings. If you need to adjust parameters such as GPU memory usage, you can configure additional parameters yourself. Please refer to [3.3.1 Server Parameter Adjustment](./PaddleOCR-VL.en.md#331-server-side-parameter-adjustment) to create a configuration file, then mount this file into the container and specify the configuration file using `backend_config` in the command to start the service, for example:
+
+```shell
+docker run \
+    -it \
+    --rm \
+    --network host \
+    --user root \
+    --privileged \
+    --shm-size 64G \
+    -v ./fastdeploy_config.yml:/tmp/fastdeploy_config.yml \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-kunlunxin-xpu \
+    paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy --backend_config /tmp/fastdeploy_config.yml
+```
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use an image corresponding to a specific PaddleOCR version, you can replace `latest` in the tag with the desired version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:paddleocr3.3-kunlunxin-xpu-offline`
+
+
+### 3.2 Client Usage Method
+
+For client-side invocation methods, please refer to [PaddleOCR-VL Usage Tutorial - 3.2 Client Usage Methods](./PaddleOCR-VL.en.md#32-client-usage-methods). If you run the client on this hardware, make sure to specify `device="xpu"`.
+
+### 3.3 Performance Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 3.3 Performance Tuning](./PaddleOCR-VL.en.md#33-performance-tuning).
+
+## 4. Service Deployment
+
+**Deployment Methods Supported on This Hardware**
+
+| Deployment method | Status | Notes |
+| --- | --- | --- |
+| Docker Compose deployment | Supported with steps in this guide | Continue with Section 4.1. |
+| Manual deployment | Not currently supported | This hardware does not currently support this path. |
+
+> IMPORTANT:
+> The PaddleOCR-VL service introduced in this section differs from the VLM inference service in the previous section: the latter is responsible for only one part of the complete process (i.e., VLM inference) and is called as an underlying service by the former.
+
+### 4.1 Deploy Using Docker Compose
+
+This step mainly introduces how to deploy PaddleOCR-VL as a service and call it using Docker Compose. The specific process is as follows:
+
+1. Download the Compose file and the environment variable configuration file separately from [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/kunlunxin-xpu/compose.yaml) and [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/kunlunxin-xpu/.env) to your local machine.
+
+2. Execute the following command in the directory where `compose.yaml` and `.env` files are located to start the server, which listens on port **8080** by default:
+
+    ```shell
+    # Must be executed in the directory where compose.yaml and .env files are located
+    docker compose up
+    ```
+
+    > TIP:
+    > The image tags used by `compose.yaml` are usually controlled by `API_IMAGE_TAG_SUFFIX` and `VLM_IMAGE_TAG_SUFFIX` in `.env`, and default to tags such as `latest-kunlunxin-xpu-offline`. To make sure you pull the newest `latest` images, run `docker compose pull` in the current directory before `docker compose up`.
+    > To use an image corresponding to a specific PaddleOCR version, replace `latest` in these variables with `paddleocr<major>.<minor>`, for example `paddleocr3.3-kunlunxin-xpu-offline`.
+
+    After starting, you will see output similar to the following:
+
+    ```text
+    paddleocr-vl-api             | INFO:     Started server process [1]
+    paddleocr-vl-api             | INFO:     Waiting for application startup.
+    paddleocr-vl-api             | INFO:     Application startup complete.
+    paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+    ```
+
+This method accelerates VLM inference based on the FastDeploy framework and is more suitable for production environment deployment.
+
+Additionally, after starting the server using this method, no internet connection is required except for pulling the image. If you need to deploy in an offline environment, you can first pull the images involved in the Compose file on a connected machine, export them, and transfer them to the offline machine for import to start the service in an offline environment.
+
+Docker Compose starts two containers sequentially by reading configurations from the `.env` and `compose.yaml` files, running the underlying VLM inference service and the PaddleOCR-VL service (pipeline service) respectively.
+
+The meanings of each environment variable contained in the `.env` file are as follows:
+
+```
+- `API_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to launch the pipeline service.
+- `VLM_BACKEND`: The VLM inference backend.
+- `VLM_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to launch the VLM inference service.
+```
+
+You can modify `compose.yaml` to meet custom requirements, for example:
+
+<details>
+<summary>1. Change the port of the PaddleOCR-VL service</summary>
+
+Edit <code>paddleocr-vl-api.ports</code> in the <code>compose.yaml</code> file to change the port. For example, if you need to change the service port to 8111, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. Specify the XPU used by the PaddleOCR-VL service</summary>
+
+Edit <code>environment</code> in the <code>compose.yaml</code> file to change the XPU used. For example, if you need to use card 1 for deployment, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    environment:
++     - XPU_VISIBLE_DEVICES: 1
+    ...
+  paddleocr-vlm-server:
+    ...
+    environment:
++     - XPU_VISIBLE_DEVICES: 1
+    ...
+```
+
+</details>
+
+<details>
+<summary>3. Adjust VLM server configuration</summary>
+
+If you want to adjust the VLM server configuration, refer to <a href="./PaddleOCR-VL.en.md#331-server-side-parameter-adjustment">3.3.1 Server Parameter Adjustment</a> to generate a configuration file.
+
+After generating the configuration file, add the following <code>paddleocr-vlm-server.volumes</code> and <code>paddleocr-vlm-server.command</code> fields to your <code>compose.yaml</code>. Replace <code>/path/to/your_config.yaml</code> with your actual configuration file path.
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes: /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. Adjust pipeline-related configurations (such as model path, batch size, deployment device, etc.)</summary>
+
+Refer to the <a href="./PaddleOCR-VL.en.md#44-pipeline-configuration-adjustment-instructions">4.4 Pipeline Configuration Adjustment Instructions</a> section.
+
+</details>
+
+### 4.2 Client Invocation Methods
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.3 Client Invocation Methods](./PaddleOCR-VL.en.md#43-client-side-invocation).
+
+### 4.3 Pipeline Configuration Adjustment Instructions
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.4 Pipeline Configuration Adjustment Instructions](./PaddleOCR-VL.en.md#44-pipeline-configuration-adjustment-instructions).
+
+## 5. Model Fine-Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 5. Model Fine-Tuning](./PaddleOCR-VL.en.md#5-model-fine-tuning).
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-Kunlunxin-XPU.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Kunlunxin-XPU.md
new file mode 100644
index 0000000..33b7640
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-Kunlunxin-XPU.md
@@ -0,0 +1,281 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL 昆仑芯 XPU 使用教程
+
+> INFO:
+> 除非另有说明，本教程中提到的 “PaddleOCR-VL” 均指 PaddleOCR-VL 系列模型（如 PaddleOCR-VL-1.6 等）；若特指 PaddleOCR-VL v1 版本，将另行明确标注。
+
+本教程是 PaddleOCR-VL 在昆仑芯 XPU 上的使用指南，涵盖了从本地运行环境准备到服务化部署的完整流程。
+
+目前 PaddleOCR-VL 已在昆仑芯 P800 上完成精度、速度验证；鉴于硬件环境的多样性，其他昆仑芯 XPU 的兼容性尚未验证。我们诚挚欢迎社区用户在不同硬件上进行测试并反馈您的运行结果。
+
+## 本硬件支持的使用目标
+
+请在本硬件教程中按下表继续阅读。
+
+| 目标 | 本硬件上的支持情况 | 从哪里开始阅读 |
+| --- | --- | --- |
+| 本地直接推理 | 支持 | 阅读第 1 节“本地运行环境准备”和第 2 节“快速开始”。 |
+| 客户端 + VLM 推理服务 | 支持 | 先完成本地直接推理，再阅读第 3 节“使用 VLM 推理服务”。 |
+| 完整 API 服务 | 支持 Docker Compose 部署 | 先阅读第 4.1 节，再继续阅读第 4.2 节客户端调用部分和第 4.3 节产线配置调整部分。 |
+| 模型微调 | 支持 | 阅读第 5 节“模型微调”。 |
+
+如果你只是想先确认本硬件支持哪些推理方式，请参考主教程中的 [PaddleOCR-VL 推理方式与硬件支持矩阵](./PaddleOCR-VL.md#paddleocr-vl-对推理设备的支持情况)。
+
+## 1. 本地运行环境准备
+
+**当前硬件支持的本地运行环境准备方式**
+
+| 本地运行环境准备方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 支持并提供步骤 | 请继续阅读本节的 1.1。 |
+| 手动安装推理引擎和 PaddleOCR | 支持并提供步骤 | 请继续阅读本节的 1.2。 |
+
+此步骤主要介绍如何搭建 PaddleOCR-VL 的本地运行环境，有以下两种方式，任选一种即可：
+
+- 方法一：使用官方 Docker 镜像。
+
+- 方法二：手动安装推理引擎和 PaddleOCR。
+
+**我们强烈推荐采用 Docker 镜像的方式，以最大程度减少可能出现的环境问题。**
+
+### 1.1 方法一：使用 Docker 镜像
+
+我们推荐使用官方 Docker 镜像（要求 Docker 版本 >= 19.03）：
+
+```shell
+docker run \
+    -it \
+    --network host \
+    --user root \
+    --privileged \
+    --shm-size 64g \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-kunlunxin-xpu \
+    /bin/bash
+# 在容器中调用 PaddleOCR CLI 或 Python API
+```
+
+如果您希望在无法连接互联网的环境中启动服务，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-kunlunxin-xpu`（镜像的大小约为 13 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-kunlunxin-xpu-offline`（镜像的大小约为 15 GB）。
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本的 PaddleOCR 镜像，可以将标签中的 `latest` 替换为对应版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.3-kunlunxin-xpu-offline`
+
+### 1.2 方法二：手动安装推理引擎和 PaddleOCR
+
+如果您无法使用 Docker，也可以手动安装推理引擎和 PaddleOCR。本文档验证过的 Python 版本范围为 3.9–3.13。
+
+当前硬件本地推理仅支持 PaddlePaddle 推理引擎。
+
+**我们强烈推荐您在虚拟环境中安装 PaddleOCR-VL，以避免发生依赖冲突。** 例如，使用 Python venv 标准库创建虚拟环境：
+
+```shell
+# 创建虚拟环境
+python -m venv .venv_paddleocr
+# 激活环境
+source .venv_paddleocr/bin/activate
+```
+
+执行如下命令完成安装：
+
+```shell
+python -m pip install paddlepaddle-xpu==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/xpu-p800/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **请注意安装 3.2.1 及以上版本的飞桨框架。**
+
+## 2. 快速开始
+
+请参考 [PaddleOCR-VL 使用教程 - 2. 快速开始](./PaddleOCR-VL.md#2)，注意需要指定 `device="xpu"`。
+
+## 3. 使用 VLM 推理服务
+
+本节介绍如何通过 VLM 推理服务接入专用后端。对于当前硬件，这通常用于提升默认配置下的推理性能，以更好满足生产需求。在当前硬件文档中，示例使用 FastDeploy 作为 VLM 推理服务后端。
+
+### 3.1 启动 VLM 推理服务
+
+> IMPORTANT:
+> 按照本节说明启动的服务仅负责 PaddleOCR-VL 流程中的 VLM 推理环节，不提供完整的端到端文档解析 API。强烈不建议直接通过 HTTP 请求或使用 OpenAI 客户端调用该服务处理文档图像。若您需要部署具备 PaddleOCR-VL 完整能力的服务，请参考后文的服务化部署部分。
+
+**当前硬件支持的启动方式**
+
+| 启动方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 支持并提供步骤 | 本节提供 FastDeploy 推理服务的启动步骤。 |
+| 通过 PaddleOCR CLI 安装依赖后启动 | 当前不支持 | 当前硬件不支持该路径。 |
+| 直接使用推理加速框架启动 | 未验证 | 当前硬件可通过 FastDeploy 后端启动 VLM 推理服务，但尚未验证直接使用 FastDeploy 原生方式启动的路径。 |
+
+PaddleOCR 提供了 Docker 镜像，用于快速启动 FastDeploy 推理服务。可使用以下命令启动服务（要求 Docker 版本 >= 19.03）：
+
+```shell
+docker run \
+    -it \
+    --network host \
+    --user root \
+    --privileged \
+    --shm-size 64g \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-kunlunxin-xpu \
+    paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy
+```
+
+如果您希望在无法连接互联网的环境中启动服务，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-kunlunxin-xpu`（镜像的大小约为 53 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-kunlunxin-xpu-offline`（镜像的大小约为 55 GB）。
+
+启动 FastDeploy 推理服务时，我们提供了一套默认参数设置。如果您有调整显存占用等更多参数的需求，可以自行配置更多参数。请参考 [3.3.1 服务端参数调整](./PaddleOCR-VL.md#331) 创建配置文件，然后将该文件挂载到容器中，并在启动服务的命令中使用 `backend_config` 指定配置文件，例如：
+
+```shell
+docker run \
+    -it \
+    --rm \
+    --network host \
+    --user root \
+    --privileged \
+    --shm-size 64G \
+    -v ./fastdeploy_config.yml:/tmp/fastdeploy_config.yml \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-kunlunxin-xpu \
+    paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy --backend_config /tmp/fastdeploy_config.yml
+```
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本的 PaddleOCR 镜像，可以将标签中的 `latest` 替换为对应版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:paddleocr3.3-kunlunxin-xpu-offline`
+
+### 3.2 客户端使用方法
+
+客户端调用方式请参考 [PaddleOCR-VL 使用教程 - 3.2 客户端使用方法](./PaddleOCR-VL.md#32)。如需在当前硬件上运行客户端，请注意指定 `device="xpu"`。
+
+### 3.3 性能调优
+
+请参考 [PaddleOCR-VL 使用教程 - 3.3 性能调优](./PaddleOCR-VL.md#33)。
+
+## 4. 服务化部署
+
+**当前硬件支持的部署方式**
+
+| 部署方式 | 状态 | 说明 |
+| --- | --- | --- |
+| Docker Compose 部署 | 支持并提供步骤 | 请继续阅读本节的 4.1。 |
+| 手动部署 | 当前不支持 | 当前硬件不支持该路径。 |
+
+> IMPORTANT:
+> 本节所介绍的 PaddleOCR-VL 服务与上一节中的 VLM 推理服务有所区别：后者仅负责完整流程中的一个环节（即 VLM 推理），并作为前者的底层服务被调用。
+
+### 4.1 使用 Docker Compose 部署
+
+此步骤主要介绍如何使用 Docker Compose 将 PaddleOCR-VL 部署为服务并调用，具体流程如下：
+
+1. 分别从 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/kunlunxin-xpu/compose.yaml) 和 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/kunlunxin-xpu/.env) 获取 Compose 文件与环境变量配置文件并下载到本地。
+
+2. 在 `compose.yaml` 和 `.env` 文件所在目录下执行以下命令启动服务器，默认监听 **8080** 端口：
+
+    ```shell
+    # 必须在 compose.yaml 和 .env 文件所在的目录中执行
+    docker compose up
+    ```
+
+    > 提示：
+    > `compose.yaml` 中使用的镜像标签通常由 `.env` 中的 `API_IMAGE_TAG_SUFFIX` 和 `VLM_IMAGE_TAG_SUFFIX` 控制，默认使用 `latest-kunlunxin-xpu-offline` 等标签。如需确保拉取到最新的 `latest` 镜像，可先在当前目录执行 `docker compose pull`，再执行 `docker compose up`。
+    > 如果希望使用特定版本的 PaddleOCR 镜像，可将这两个环境变量中的 `latest` 替换为对应版本号 `paddleocr<major>.<minor>`，例如 `paddleocr3.3-kunlunxin-xpu-offline`。
+
+    启动后将看到类似如下输出：
+
+    ```text
+    paddleocr-vl-api             | INFO:     Started server process [1]
+    paddleocr-vl-api             | INFO:     Waiting for application startup.
+    paddleocr-vl-api             | INFO:     Application startup complete.
+    paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+    ```
+
+此方式基于 FastDeploy 框架对 VLM 推理进行加速，更适合生产环境部署。
+
+此外，使用此方式启动服务器后，除拉取镜像外，无需连接互联网。如需在离线环境中部署，可先在联网机器上拉取 Compose 文件中涉及的镜像，导出并传输至离线机器中导入，即可在离线环境下启动服务。
+
+Docker Compose 通过读取 `.env` 和 `compose.yaml` 文件中配置，先后启动 2 个容器，分别运行底层 VLM 推理服务，以及 PaddleOCR-VL 服务（产线服务）。
+
+`.env` 文件中包含的各环境变量含义如下：
+
+- `API_IMAGE_TAG_SUFFIX`：启动产线服务使用的镜像的标签后缀。
+- `VLM_BACKEND`：VLM 推理后端。
+- `VLM_IMAGE_TAG_SUFFIX`：启动 VLM 推理服务使用的镜像的标签后缀。
+
+您可以通过修改 `compose.yaml` 来满足自定义需求，例如：
+
+<details>
+<summary>1. 更改 PaddleOCR-VL 服务的端口</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>paddleocr-vl-api.ports</code> 来更改端口。例如，如果您需要将服务端口更换为 8111，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. 指定 PaddleOCR-VL 服务所使用的 XPU</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>environment</code> 来更改所使用的 XPU。例如，如果您需要使用卡 1 进行部署，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    environment:
++     - XPU_VISIBLE_DEVICES: 1
+    ...
+  paddleocr-vlm-server:
+    ...
+    environment:
++     - XPU_VISIBLE_DEVICES: 1
+    ...
+```
+
+</details>
+
+<details>
+<summary>3. 调整 VLM 服务端配置</summary>
+
+若您想调整 VLM 服务端的配置，可以参考 <a href="./PaddleOCR-VL.md#331">3.3.1 服务端参数调整</a> 生成配置文件。
+
+生成配置文件后，将以下的 <code>paddleocr-vlm-server.volumes</code> 和 <code>paddleocr-vlm-server.command</code> 字段增加到您的 <code>compose.yaml</code> 中。请将 <code>/path/to/your_config.yaml</code> 替换为您的实际配置文件路径。
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes: /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. 调整产线相关配置（如模型路径、批处理大小、部署设备等）</summary>
+
+参考 <a href="./PaddleOCR-VL.md#44">4.4 产线配置调整说明</a> 小节。
+
+</details>
+
+### 4.2 客户端调用方式
+
+请参考 [PaddleOCR-VL 使用教程 - 4.3 客户端调用方式](./PaddleOCR-VL.md#43)。
+
+### 4.3 产线配置调整说明
+
+请参考 [PaddleOCR-VL 使用教程 - 4.4 产线配置调整说明](./PaddleOCR-VL.md#44)。
+
+## 5. 模型微调
+
+请参考 [PaddleOCR-VL 使用教程 - 5. 模型微调](./PaddleOCR-VL.md#5)。
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-MetaX-GPU.en.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-MetaX-GPU.en.md
new file mode 100644
index 0000000..7c71d55
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-MetaX-GPU.en.md
@@ -0,0 +1,293 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL MetaX GPU Usage Tutorial
+
+> INFO:
+> Unless otherwise specified, the term "PaddleOCR-VL" in this tutorial refers to the PaddleOCR-VL model series (e.g., PaddleOCR-VL-1.6). References specific to the PaddleOCR-VL v1 version will be explicitly noted.
+
+This tutorial is a guide for using PaddleOCR-VL on MetaX GPU, covering the complete workflow from environment preparation to service deployment.
+
+PaddleOCR-VL has been verified for accuracy and speed on the MetaX C550. However, due to hardware diversity, compatibility with other MetaX GPUs has not yet been confirmed. We welcome the community to test on different hardware setups and share your results.
+
+## Workflow Guide for This Hardware
+
+Use this guide for the workflows below.
+
+| Goal | Support on this hardware | Read this section |
+| --- | --- | --- |
+| Local direct inference | Supported | Read Section 1. Local Runtime Environment Preparation and Section 2. Quick Start. |
+| Client + VLM inference service | Supported | Complete local direct inference first, then read Section 3. Using VLM Inference Services. |
+| Full API service | Supported with Docker Compose deployment | Read Section 4.1 first, then continue with the Section 4.2 client invocation section and the Section 4.3 pipeline configuration section. |
+| Model fine-tuning | Supported | Read Section 5. Model Fine-Tuning. |
+
+If you only need to confirm which inference methods are available on this hardware, refer to the [PaddleOCR-VL Inference Method and Hardware Support Matrix](./PaddleOCR-VL.en.md#inference-device-support-for-paddleocr-vl) in the main guide.
+
+## 1. Local Runtime Environment Preparation
+
+**Local Runtime Environment Setup Methods Supported on This Hardware**
+
+| Local runtime environment setup method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Supported with steps in this guide | Continue with Section 1.1. |
+| Manually install the inference engine and PaddleOCR | Supported with steps in this guide | Continue with Section 1.2. |
+
+This step mainly introduces how to set up the local runtime environment for PaddleOCR-VL. There are two methods available; choose either one:
+
+- Method 1: Use the official Docker image.
+
+- Method 2: Manually install the inference engine and PaddleOCR.
+
+### 1.1 Method 1: Using Docker Image
+
+We recommend using the official Docker image (requires Docker version >= 19.03):
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  --device /dev/dri:/dev/dri \
+  --device /dev/mxcd:/dev/mxcd \
+  --security-opt seccomp=unconfined \
+  --security-opt apparmor=unconfined \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-metax-gpu \
+  /bin/bash
+# Call PaddleOCR CLI or Python API in the container
+```
+
+If you wish to start the service in an environment without internet access, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-metax-gpu` (image size approximately 32 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-metax-gpu-offline` (image size approximately 34 GB).
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use an image corresponding to a specific PaddleOCR version, you can replace `latest` in the tag with the desired version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.3-metax-gpu-offline`
+
+
+### 1.2 Method 2: Manually Install the Inference Engine and PaddleOCR
+
+If you cannot use Docker, you can also manually install the inference engine and PaddleOCR. This guide documents Python 3.9–3.13 as the verified range.
+
+Local inference on this hardware currently supports only the PaddlePaddle inference engine.
+
+**We strongly recommend installing PaddleOCR-VL in a virtual environment to avoid dependency conflicts.** For example, use the Python venv standard library to create a virtual environment:
+
+```shell
+# Create a virtual environment
+python -m venv .venv_paddleocr
+# Activate the environment
+source .venv_paddleocr/bin/activate
+```
+
+Execute the following commands to complete the installation:
+
+```shell
+python -m pip install paddlepaddle==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+python -m pip install paddle-metax-gpu==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/maca/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **Please note to install PaddlePaddle version 3.2.0 or above.**
+
+## 2. Quick Start
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 2. Quick Start](./PaddleOCR-VL.en.md#2-quick-start), making sure to specify `device='metax_gpu'`.
+
+## 3. Using VLM Inference Services
+
+This section explains how to connect PaddleOCR-VL to a dedicated VLM inference service backend. On this hardware, this is usually used to improve inference performance beyond the default configuration for production use. In this hardware-specific guide, the examples use FastDeploy as the backend for the VLM inference service.
+
+### 3.1 Starting the VLM Inference Service
+
+> IMPORTANT:
+> The service started according to this section is responsible only for the VLM inference stage in the PaddleOCR-VL workflow. It does not provide a complete end-to-end document parsing API. We strongly recommend that you do not call this service directly via HTTP requests or OpenAI clients to process document images. If you need to deploy a service with the full PaddleOCR-VL capabilities, refer to the service deployment section later in this document.
+
+**Launch Methods Supported on This Hardware**
+
+| Launch method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Supported with steps in this guide | This section provides the FastDeploy service launch steps. |
+| Install dependencies with the PaddleOCR CLI and launch the service | Not currently supported | This hardware does not currently support this path. |
+| Launch the service directly with the acceleration framework | Not verified | This hardware can start the VLM inference service through the FastDeploy backend, but launching directly with native FastDeploy has not been verified. |
+
+PaddleOCR provides a Docker image for quickly starting the FastDeploy inference service. Use the following command to start the service (requires Docker version >= 19.03):
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  --device /dev/dri:/dev/dri \
+  --device /dev/mxcd:/dev/mxcd \
+  --security-opt seccomp=unconfined \
+  --security-opt apparmor=unconfined \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-metax-gpu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy
+```
+
+If you wish to start the service in an environment without internet access, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-metax-gpu` (image size approximately 37 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-metax-gpu-offline` (image size approximately 39 GB).
+
+When launching the FastDeploy inference service, we provide a set of default parameter settings. If you need to adjust parameters such as GPU memory usage, you can configure additional parameters yourself. Please refer to [3.3.1 Server-side Parameter Adjustment](./PaddleOCR-VL.en.md#331-server-side-parameter-adjustment) to create a configuration file, then mount the file into the container and specify the configuration file using `backend_config` in the command to start the service, for example:
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  --device /dev/dri:/dev/dri \
+  --device /dev/mxcd:/dev/mxcd \
+  --security-opt seccomp=unconfined \
+  --security-opt apparmor=unconfined \
+  --shm-size 64g \
+  --network host \
+  -v ./fastdeploy_config.yml:/tmp/fastdeploy_config.yml \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-metax-gpu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy --backend_config /tmp/fastdeploy_config.yml
+```
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use an image corresponding to a specific PaddleOCR version, you can replace `latest` in the tag with the desired version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:paddleocr3.3-metax-gpu-offline`
+
+
+### 3.2 Client Usage Method
+
+For client-side invocation methods, please refer to [PaddleOCR-VL Usage Tutorial - 3.2 Client Usage Methods](./PaddleOCR-VL.en.md#32-client-usage-methods). If you run the client on this hardware, make sure to specify `device="metax_gpu"`.
+
+### 3.3 Performance Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 3.3 Performance Tuning](./PaddleOCR-VL.en.md#33-performance-tuning).
+
+## 4. Service Deployment
+
+**Deployment Methods Supported on This Hardware**
+
+| Deployment method | Status | Notes |
+| --- | --- | --- |
+| Docker Compose deployment | Supported with steps in this guide | Continue with Section 4.1. |
+| Manual deployment | Not currently supported | This hardware does not currently support this path. |
+
+> IMPORTANT:
+> The PaddleOCR-VL service introduced in this section differs from the VLM inference service in the previous section: the latter is responsible for only one part of the complete process (i.e., VLM inference) and is called as an underlying service by the former.
+
+### 4.1 Deploy Using Docker Compose
+
+This step mainly introduces how to use Docker Compose to deploy PaddleOCR-VL as a service and call it. The specific process is as follows:
+
+1. Download the Compose file and the environment variable configuration file separately from [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/metax-gpu/compose.yaml) and [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/metax-gpu/.env) to your local machine.
+    
+2. Execute the following command in the directory where the `compose.yaml` and `.env` files are located to start the server, which listens on port **8080** by default:
+
+    ```shell
+    # Must be executed in the directory where compose.yaml and .env files are located
+    docker compose up
+    ```
+
+    > TIP:
+    > The image tags used by `compose.yaml` are usually controlled by `API_IMAGE_TAG_SUFFIX` and `VLM_IMAGE_TAG_SUFFIX` in `.env`, and default to tags such as `latest-metax-gpu-offline`. To make sure you pull the newest `latest` images, run `docker compose pull` in the current directory before `docker compose up`.
+    > To use an image corresponding to a specific PaddleOCR version, replace `latest` in these variables with `paddleocr<major>.<minor>`, for example `paddleocr3.3-metax-gpu-offline`.
+
+    After startup, you will see output similar to the following:
+
+    ```text
+    paddleocr-vl-api             | INFO:     Started server process [1]
+    paddleocr-vl-api             | INFO:     Waiting for application startup.
+    paddleocr-vl-api             | INFO:     Application startup complete.
+    paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+    ```
+
+This method accelerates VLM inference using the FastDeploy framework and is more suitable for production environment deployment.
+
+Additionally, after starting the server in this manner, no internet connection is required except for image pulling. For deployment in an offline environment, you can first pull the images involved in the Compose file on a connected machine, export them, and transfer them to the offline machine for import to start the service in an offline environment.
+
+Docker Compose starts two containers sequentially by reading configurations from the `.env` and `compose.yaml` files, running the underlying VLM inference service and the PaddleOCR-VL service (pipeline service) respectively.
+
+The meanings of each environment variable contained in the `.env` file are as follows:
+
+```
+- `API_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to launch the pipeline service.
+- `VLM_BACKEND`: The VLM inference backend.
+- `VLM_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to launch the VLM inference service.
+```
+
+You can modify `compose.yaml` to meet custom requirements, for example:
+
+<details>
+<summary>1. Change the port of the PaddleOCR-VL service</summary>
+
+Edit <code>paddleocr-vl-api.ports</code> in the <code>compose.yaml</code> file to change the port. For example, if you need to change the service port to 8111, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. Specify the GPU used by the PaddleOCR-VL service</summary>
+
+Edit <code>environment</code> in the <code>compose.yaml</code> file to change the GPU used. For example, if you need to use card 1 for deployment, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    environment:
++     - MACA_VISIBLE_DEVICES: 1
+    ...
+  paddleocr-vlm-server:
+    ...
+    environment:
++     - MACA_VISIBLE_DEVICES: 1
+    ...
+```
+
+
+</details>
+
+<details>
+<summary>3. Adjust VLM server-side configuration</summary>
+
+If you want to adjust the VLM server configuration, refer to <a href="./PaddleOCR-VL.en.md#331-server-side-parameter-adjustment">3.3.1 Server Parameter Adjustment</a> to generate a configuration file.
+
+After generating the configuration file, add the following <code>paddleocr-vlm-server.volumes</code> and <code>paddleocr-vlm-server.command</code> fields to your <code>compose.yaml</code>. Replace <code>/path/to/your_config.yaml</code> with your actual configuration file path.
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes: /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. Adjust pipeline-related configurations (such as model path, batch size, deployment device, etc.)</summary>
+
+Refer to the <a href="./PaddleOCR-VL.en.md#44-pipeline-configuration-adjustment-instructions">4.4 Pipeline Configuration Adjustment Instructions</a> section.
+
+</details>
+
+### 4.2 Client Invocation Method
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.3 Client Invocation Method](./PaddleOCR-VL.en.md#43-client-side-invocation).
+
+### 4.3 Pipeline Configuration Adjustment Instructions
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.4 Pipeline Configuration Adjustment Instructions](./PaddleOCR-VL.en.md#44-pipeline-configuration-adjustment-instructions).
+
+## 5. Model Fine-Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 5. Model Fine-Tuning](./PaddleOCR-VL.en.md#5-model-fine-tuning).
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-MetaX-GPU.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-MetaX-GPU.md
new file mode 100644
index 0000000..11625cb
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-MetaX-GPU.md
@@ -0,0 +1,289 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL 沐曦 GPU 使用教程
+
+> INFO:
+> 除非另有说明，本教程中提到的 “PaddleOCR-VL” 均指 PaddleOCR-VL 系列模型（如 PaddleOCR-VL-1.6 等）；若特指 PaddleOCR-VL v1 版本，将另行明确标注。
+
+本教程是 PaddleOCR-VL 在沐曦 GPU 上的使用指南，涵盖了从本地运行环境准备到服务化部署的完整流程。
+
+目前 PaddleOCR-VL 已在沐曦 C550 上完成精度、速度验证；鉴于硬件环境的多样性，其他沐曦 GPU 的兼容性尚未验证。我们诚挚欢迎社区用户在不同硬件上进行测试并反馈您的运行结果。
+
+## 本硬件支持的使用目标
+
+请在本硬件教程中按下表继续阅读。
+
+| 目标 | 本硬件上的支持情况 | 从哪里开始阅读 |
+| --- | --- | --- |
+| 本地直接推理 | 支持 | 阅读第 1 节“本地运行环境准备”和第 2 节“快速开始”。 |
+| 客户端 + VLM 推理服务 | 支持 | 先完成本地直接推理，再阅读第 3 节“使用 VLM 推理服务”。 |
+| 完整 API 服务 | 支持 Docker Compose 部署 | 先阅读第 4.1 节，再继续阅读第 4.2 节客户端调用部分和第 4.3 节产线配置调整部分。 |
+| 模型微调 | 支持 | 阅读第 5 节“模型微调”。 |
+
+如果你只是想先确认本硬件支持哪些推理方式，请参考主教程中的 [PaddleOCR-VL 推理方式与硬件支持矩阵](./PaddleOCR-VL.md#paddleocr-vl-对推理设备的支持情况)。
+
+## 1. 本地运行环境准备
+
+**当前硬件支持的本地运行环境准备方式**
+
+| 本地运行环境准备方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 支持并提供步骤 | 请继续阅读本节的 1.1。 |
+| 手动安装推理引擎和 PaddleOCR | 支持并提供步骤 | 请继续阅读本节的 1.2。 |
+
+此步骤主要介绍如何搭建 PaddleOCR-VL 的本地运行环境，有以下两种方式，任选一种即可：
+
+- 方法一：使用官方 Docker 镜像。
+
+- 方法二：手动安装推理引擎和 PaddleOCR。
+
+### 1.1 方法一：使用 Docker 镜像
+
+我们推荐使用官方 Docker 镜像（要求 Docker 版本 >= 19.03）：
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  --device /dev/dri:/dev/dri \
+  --device /dev/mxcd:/dev/mxcd \
+  --security-opt seccomp=unconfined \
+  --security-opt apparmor=unconfined \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-metax-gpu \
+  /bin/bash
+# 在容器中调用 PaddleOCR CLI 或 Python API
+```
+
+如果您希望在无法连接互联网的环境中启动服务，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-metax-gpu`（镜像的大小约为 32 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-metax-gpu-offline`（镜像的大小约为 34 GB）。
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本的 PaddleOCR 镜像，可以将标签中的 `latest` 替换为对应版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.3-metax-gpu-offline`
+
+### 1.2 方法二：手动安装推理引擎和 PaddleOCR
+
+如果您无法使用 Docker，也可以手动安装推理引擎和 PaddleOCR。本文档验证过的 Python 版本范围为 3.9–3.13。
+
+当前硬件本地推理仅支持 PaddlePaddle 推理引擎。
+
+**我们强烈推荐您在虚拟环境中安装 PaddleOCR-VL，以避免发生依赖冲突。** 例如，使用 Python venv 标准库创建虚拟环境：
+
+```shell
+# 创建虚拟环境
+python -m venv .venv_paddleocr
+# 激活环境
+source .venv_paddleocr/bin/activate
+```
+
+执行如下命令完成安装：
+
+```shell
+python -m pip install paddlepaddle==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+python -m pip install paddle-metax-gpu==3.2.0 -i https://www.paddlepaddle.org.cn/packages/stable/maca/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **请注意安装 3.2.0 及以上版本的飞桨框架。**
+
+## 2. 快速开始
+
+请参考 [PaddleOCR-VL 使用教程 - 2. 快速开始](./PaddleOCR-VL.md#2)，注意需要指定 `device="metax_gpu"`。
+
+## 3. 使用 VLM 推理服务
+
+本节介绍如何通过 VLM 推理服务接入专用后端。对于当前硬件，这通常用于提升默认配置下的推理性能，以更好满足生产需求。在当前硬件文档中，示例使用 FastDeploy 作为 VLM 推理服务后端。
+
+### 3.1 启动 VLM 推理服务
+
+> IMPORTANT:
+> 按照本节说明启动的服务仅负责 PaddleOCR-VL 流程中的 VLM 推理环节，不提供完整的端到端文档解析 API。强烈不建议直接通过 HTTP 请求或使用 OpenAI 客户端调用该服务处理文档图像。若您需要部署具备 PaddleOCR-VL 完整能力的服务，请参考后文的服务化部署部分。
+
+**当前硬件支持的启动方式**
+
+| 启动方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 支持并提供步骤 | 本节提供 FastDeploy 推理服务的启动步骤。 |
+| 通过 PaddleOCR CLI 安装依赖后启动 | 当前不支持 | 当前硬件不支持该路径。 |
+| 直接使用推理加速框架启动 | 未验证 | 当前硬件可通过 FastDeploy 后端启动 VLM 推理服务，但尚未验证直接使用 FastDeploy 原生方式启动的路径。 |
+
+PaddleOCR 提供了 Docker 镜像，用于快速启动 FastDeploy 推理服务。可使用以下命令启动服务（要求 Docker 版本 >= 19.03）：
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  --device /dev/dri:/dev/dri \
+  --device /dev/mxcd:/dev/mxcd \
+  --security-opt seccomp=unconfined \
+  --security-opt apparmor=unconfined \
+  --shm-size 64g \
+  --network host \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-metax-gpu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy
+```
+
+如果您希望在无法连接互联网的环境中启动服务，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-metax-gpu`（镜像的大小约为 37 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-metax-gpu-offline`（镜像的大小约为 39 GB）。
+
+启动 FastDeploy 推理服务时，我们提供了一套默认参数设置。如果您有调整显存占用等更多参数的需求，可以自行配置更多参数。请参考 [3.3.1 服务端参数调整](./PaddleOCR-VL.md#331) 创建配置文件，然后将该文件挂载到容器中，并在启动服务的命令中使用 `backend_config` 指定配置文件，例如：
+
+```shell
+docker run -it \
+  --user root \
+  --privileged \
+  --device /dev/dri:/dev/dri \
+  --device /dev/mxcd:/dev/mxcd \
+  --security-opt seccomp=unconfined \
+  --security-opt apparmor=unconfined \
+  --shm-size 64g \
+  --network host \
+  -v ./fastdeploy_config.yml:/tmp/fastdeploy_config.yml \
+  ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-metax-gpu \
+  paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy --backend_config /tmp/fastdeploy_config.yml
+```
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本的 PaddleOCR 镜像，可以将标签中的 `latest` 替换为对应版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:paddleocr3.3-metax-gpu-offline`
+
+### 3.2 客户端使用方法
+
+客户端调用方式请参考 [PaddleOCR-VL 使用教程 - 3.2 客户端使用方法](./PaddleOCR-VL.md#32)。如需在当前硬件上运行客户端，请注意指定 `device="metax_gpu"`。
+
+### 3.3 性能调优
+
+请参考 [PaddleOCR-VL 使用教程 - 3.3 性能调优](./PaddleOCR-VL.md#33)。
+
+## 4. 服务化部署
+
+**当前硬件支持的部署方式**
+
+| 部署方式 | 状态 | 说明 |
+| --- | --- | --- |
+| Docker Compose 部署 | 支持并提供步骤 | 请继续阅读本节的 4.1。 |
+| 手动部署 | 当前不支持 | 当前硬件不支持该路径。 |
+
+> IMPORTANT:
+> 本节所介绍的 PaddleOCR-VL 服务与上一节中的 VLM 推理服务有所区别：后者仅负责完整流程中的一个环节（即 VLM 推理），并作为前者的底层服务被调用。
+
+### 4.1 使用 Docker Compose 部署
+
+此步骤主要介绍如何使用 Docker Compose 将 PaddleOCR-VL 部署为服务并调用，具体流程如下：
+
+1. 分别从 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/metax-gpu/compose.yaml) 和 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/metax-gpu/.env) 获取 Compose 文件与环境变量配置文件并下载到本地。
+
+2. 在 `compose.yaml` 和 `.env` 文件所在目录下执行以下命令启动服务器，默认监听 **8080** 端口：
+
+    ```shell
+    # 必须在 compose.yaml 和 .env 文件所在的目录中执行
+    docker compose up
+    ```
+
+    > 提示：
+    > `compose.yaml` 中使用的镜像标签通常由 `.env` 中的 `API_IMAGE_TAG_SUFFIX` 和 `VLM_IMAGE_TAG_SUFFIX` 控制，默认使用 `latest-metax-gpu-offline` 等标签。如需确保拉取到最新的 `latest` 镜像，可先在当前目录执行 `docker compose pull`，再执行 `docker compose up`。
+    > 如果希望使用特定版本的 PaddleOCR 镜像，可将这两个环境变量中的 `latest` 替换为对应版本号 `paddleocr<major>.<minor>`，例如 `paddleocr3.3-metax-gpu-offline`。
+
+    启动后将看到类似如下输出：
+
+    ```text
+    paddleocr-vl-api             | INFO:     Started server process [1]
+    paddleocr-vl-api             | INFO:     Waiting for application startup.
+    paddleocr-vl-api             | INFO:     Application startup complete.
+    paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+    ```
+
+此方式基于 FastDeploy 框架对 VLM 推理进行加速，更适合生产环境部署。
+
+此外，使用此方式启动服务器后，除拉取镜像外，无需连接互联网。如需在离线环境中部署，可先在联网机器上拉取 Compose 文件中涉及的镜像，导出并传输至离线机器中导入，即可在离线环境下启动服务。
+
+Docker Compose 通过读取 `.env` 和 `compose.yaml` 文件中配置，先后启动 2 个容器，分别运行底层 VLM 推理服务，以及 PaddleOCR-VL 服务（产线服务）。
+
+`.env` 文件中包含的各环境变量含义如下：
+
+- `API_IMAGE_TAG_SUFFIX`：启动产线服务使用的镜像的标签后缀。
+- `VLM_BACKEND`：VLM 推理后端。
+- `VLM_IMAGE_TAG_SUFFIX`：启动 VLM 推理服务使用的镜像的标签后缀。
+
+您可以通过修改 `compose.yaml` 来满足自定义需求，例如：
+
+<details>
+<summary>1. 更改 PaddleOCR-VL 服务的端口</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>paddleocr-vl-api.ports</code> 来更改端口。例如，如果您需要将服务端口更换为 8111，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. 指定 PaddleOCR-VL 服务所使用的 GPU</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>environment</code> 来更改所使用的 GPU。例如，如果您需要使用卡 1 进行部署，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    environment:
++     - MACA_VISIBLE_DEVICES: 1
+    ...
+  paddleocr-vlm-server:
+    ...
+    environment:
++     - MACA_VISIBLE_DEVICES: 1
+    ...
+```
+
+
+</details>
+
+<details>
+<summary>3. 调整 VLM 服务端配置</summary>
+
+若您想调整 VLM 服务端的配置，可以参考 <a href="./PaddleOCR-VL.md#331">3.3.1 服务端参数调整</a> 生成配置文件。
+
+生成配置文件后，将以下的 <code>paddleocr-vlm-server.volumes</code> 和 <code>paddleocr-vlm-server.command</code> 字段增加到您的 <code>compose.yaml</code> 中。请将 <code>/path/to/your_config.yaml</code> 替换为您的实际配置文件路径。
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes: /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. 调整产线相关配置（如模型路径、批处理大小、部署设备等）</summary>
+
+参考 <a href="./PaddleOCR-VL.md#44">4.4 产线配置调整说明</a> 小节。
+
+</details>
+
+### 4.2 客户端调用方式
+
+请参考 [PaddleOCR-VL 使用教程 - 4.3 客户端调用方式](./PaddleOCR-VL.md#43)。
+
+### 4.3 产线配置调整说明
+
+请参考 [PaddleOCR-VL 使用教程 - 4.4 产线配置调整说明](./PaddleOCR-VL.md#44)。
+
+## 5. 模型微调
+
+请参考 [PaddleOCR-VL 使用教程 - 5. 模型微调](./PaddleOCR-VL.md#5)。
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-NVIDIA-Blackwell.en.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-NVIDIA-Blackwell.en.md
new file mode 100644
index 0000000..68da339
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-NVIDIA-Blackwell.en.md
@@ -0,0 +1,372 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL NVIDIA Blackwell-Architecture GPUs Usage Tutorial
+
+> INFO:
+> Unless otherwise specified, the term "PaddleOCR-VL" in this tutorial refers to the PaddleOCR-VL model series (e.g., PaddleOCR-VL-1.6). References specific to the PaddleOCR-VL v1 version will be explicitly noted.
+
+This tutorial provides guidance on using PaddleOCR-VL on NVIDIA Blackwell-architecture GPUs, covering the complete workflow from environment preparation to service deployment.
+
+NVIDIA Blackwell-architecture GPUs include, but are not limited to:
+
+- RTX 5090
+- RTX 5080
+- RTX 5070、RTX 5070 Ti
+- RTX 5060、RTX 5060 Ti
+- RTX 5050
+
+PaddleOCR-VL has been verified for accuracy and speed on the RTX 5070. However, due to hardware diversity, compatibility with other NVIDIA Blackwell-architecture GPUs has not yet been confirmed. We welcome the community to test on different hardware setups and share your results.
+
+Before starting the tutorial, **please ensure that your NVIDIA driver supports CUDA 12.9 or higher**.
+
+## Workflow Guide for This Hardware
+
+Use this guide for the workflows below.
+
+| Goal | Support on this hardware | Read this section |
+| --- | --- | --- |
+| Local direct inference | Supported | Read Section 1. Local Runtime Environment Preparation and Section 2. Quick Start. |
+| Client + VLM inference service | Supported | Complete local direct inference first, then read Section 3. Using VLM Inference Services. |
+| Full API service | Supported with Docker Compose or manual deployment | Use Section 4.1 for Docker Compose, or Section 4.2 for manual deployment (complete Section 1. Local Runtime Environment Preparation first), then continue with the Section 4.3 client invocation section and the Section 4.4 pipeline configuration section. |
+| Model fine-tuning | Supported | Read Section 5. Model Fine-Tuning. |
+
+If you only need to confirm which inference methods are available on this hardware, refer to the [PaddleOCR-VL Inference Method and Hardware Support Matrix](./PaddleOCR-VL.en.md#inference-device-support-for-paddleocr-vl) in the main guide.
+
+## 1. Local Runtime Environment Preparation
+
+**Local Runtime Environment Setup Methods Supported on This Hardware**
+
+| Local runtime environment setup method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Supported with steps in this guide | Continue with Section 1.1. |
+| Manually install the inference engine and PaddleOCR | Supported with steps in this guide | Continue with Section 1.2. |
+
+This section introduces how to set up the PaddleOCR-VL local runtime environment using one of the following two methods:
+
+- Method 1: Use the official Docker image.
+
+- Method 2: Manually install the inference engine and PaddleOCR.
+
+**We strongly recommend using the Docker image to minimize potential environment-related issues.**
+
+### 1.1 Method 1: Using Docker Image
+
+We recommend using the official Docker image (requires Docker version >= 19.03, GPU-equipped machine with NVIDIA driver supporting CUDA 12.9 or higher):
+
+```shell
+docker run \
+    -it \
+    --gpus all \
+    --network host \
+    --user root \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu-sm120 \
+    /bin/bash
+# Call PaddleOCR CLI or Python API in the container
+```
+
+If you wish to use PaddleOCR-VL in an offline environment, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu-sm120` (image size approximately 10 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu-sm120-offline` (image size approximately 12 GB).
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use an image corresponding to a specific PaddleOCR version, you can replace `latest` in the tag with the desired version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.3-nvidia-gpu-sm120-offline`
+
+
+### 1.2 Method 2: Manually Install the Inference Engine and PaddleOCR
+
+If Docker is not an option, you can manually install the inference engine and PaddleOCR. This guide documents Python 3.9–3.13 as the verified range.
+
+This guide provides PaddlePaddle installation steps. To use Transformers or other inference engines, see [Section 1.2 of the main tutorial](./PaddleOCR-VL.en.md#manual-install-inference-engine-and-paddleocr).
+
+**We strongly recommend installing PaddleOCR-VL in a virtual environment to avoid dependency conflicts.** For example, create a virtual environment using Python's standard venv library:
+
+```shell
+# Create a virtual environment
+python -m venv .venv_paddleocr
+# Activate the environment
+source .venv_paddleocr/bin/activate
+```
+
+Run the following commands to complete the installation:
+
+```shell
+# Note that PaddlePaddle for cu129 is being installed here
+python -m pip install paddlepaddle-gpu==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/cu129/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **Please ensure that PaddlePaddle framework version 3.2.1 or higher is installed.**
+
+## 2. Quick Start
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 2. Quick Start](./PaddleOCR-VL.en.md#2-quick-start).
+
+## 3. Using VLM Inference Services
+
+This section explains how to connect PaddleOCR-VL to a dedicated VLM inference service backend. On this hardware, this is usually used to improve inference performance beyond the default configuration for production use. In this hardware-specific guide, the examples use vLLM and SGLang as the backends for the VLM inference service.
+
+### 3.1 Starting the VLM Inference Service
+
+> IMPORTANT:
+> The service started according to this section is responsible only for the VLM inference stage in the PaddleOCR-VL workflow. It does not provide a complete end-to-end document parsing API. We strongly recommend that you do not call this service directly via HTTP requests or OpenAI clients to process document images. If you need to deploy a service with the full PaddleOCR-VL capabilities, refer to the service deployment section later in this document.
+
+**Launch Methods Supported on This Hardware**
+
+| Launch method | Status | Notes |
+| --- | --- | --- |
+| Official Docker image | Supported with steps in this guide | Continue with Section 3.1.1. |
+| Install dependencies with the PaddleOCR CLI and launch the service | Supported with steps in this guide | Continue with Section 3.1.2. |
+| Launch the service directly with the acceleration framework | Not verified | This hardware can start the VLM inference service through the vLLM or SGLang backend, but launching directly with the native framework has not been verified. |
+
+There are two methods to start the VLM inference service; choose one:
+
+- Method 1: Start the service using the official Docker image.
+
+- Method 2: Manually install dependencies and start the service via PaddleOCR CLI.
+
+**We strongly recommend using the Docker image to minimize potential environment-related issues.**
+
+#### 3.1.1 Method 1: Using Docker Image
+
+PaddleOCR provides a Docker image for quickly starting the vLLM inference service. Use the following command to start the service (requires Docker version >= 19.03, GPU-equipped machine with NVIDIA driver supporting CUDA 12.9 or higher):
+
+```shell
+docker run \
+    -it \
+    --gpus all \
+    --network host \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu-sm120 \
+    paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm
+```
+
+If you wish to start the service in an offline environment, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu-sm120` (image size approximately 13 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu-sm120-offline` (image size approximately 15 GB).
+
+When launching the vLLM inference service, we provide a set of default parameter settings. If you need to adjust parameters such as GPU memory usage, you can configure additional parameters yourself. Please refer to [3.3.1 Server-side Parameter Adjustment](./PaddleOCR-VL.en.md#331-server-side-parameter-adjustment) to create a configuration file, then mount the file into the container and specify the configuration file using `backend_config` in the command to start the service, for example:
+
+```shell
+docker run \
+    -it \
+    --rm \
+    --gpus all \
+    --network host \
+    -v ./vllm_config.yml:/tmp/vllm_config.yml \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu-sm120 \
+    paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /tmp/vllm_config.yml
+```
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use an image corresponding to a specific PaddleOCR version, you can replace `latest` in the tag with the desired version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:paddleocr3.3-nvidia-gpu-sm120-offline`
+
+
+#### 3.1.2 Method 2: Installation and Usage via PaddleOCR CLI
+
+Since inference acceleration frameworks may conflict with packages already installed in the current environment, it is recommended to install them in a virtual environment:
+
+```shell
+# If a virtual environment is currently activated, deactivate it first using `deactivate`
+# Create a virtual environment
+python -m venv .venv_vlm
+# Activate the environment
+source .venv_vlm/bin/activate
+```
+
+vLLM and SGLang depend on FlashAttention, and installing FlashAttention may require CUDA compilation tools such as `nvcc`. If these tools are not available in your environment (for example, when using the `paddleocr-vl` image), you can obtain a prebuilt FlashAttention package (version 2.8.3 required) from [this repository](https://github.com/mjun0812/flash-attention-prebuild-wheels), install it first, and then proceed with subsequent commands. For example, in the `paddleocr-vl` image, run `python -m pip install https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.3.14/flash_attn-2.8.3+cu128torch2.8-cp310-cp310-linux_x86_64.whl`. This step is not required for FastDeploy.
+
+Install PaddleOCR and the dependencies of inference acceleration services, using vLLM as an example:
+
+```shell
+# Install PaddleOCR
+python -m pip install "paddleocr[doc-parser]"
+# Install inference acceleration service dependencies
+paddleocr install_genai_server_deps vllm
+```
+
+Usage of the `paddleocr install_genai_server_deps` command:
+
+```shell
+paddleocr install_genai_server_deps <inference acceleration framework name>
+```
+
+Currently supported framework names are `vllm` and `sglang`, corresponding to vLLM and SGLang, respectively.
+
+> WARNING:
+> The transformers library versions required by vLLM, SGLang and Transformers engine are currently incompatible, so Transformers engine cannot be installed together with vLLM or SGLang in the same environment. If using Transformers + vLLM or Transformers + SGLang inference, please deploy the layout analysis model and VLM service in different environments.
+
+After installation, you can start the service using the `paddleocr genai_server` command:
+
+```shell
+paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --backend vllm --port 8118
+```
+
+The parameters supported by this command are as follows:
+
+| Parameter         | Description                                                                 |
+|------------------|-----------------------------------------------------------------------------|
+| `--model_name`   | Name of the model                                                            |
+| `--model_dir`    | Directory containing the model                                               |
+| `--host`         | Server hostname                                                             |
+| `--port`         | Server port number                                                           |
+| `--backend`      | Backend name, i.e., the name of the inference acceleration framework being used; options are `vllm` or `sglang` |
+| `--backend_config`| YAML file specifying backend configuration                                   |
+
+### 3.2 Client Usage
+
+For client-side invocation methods, please refer to [PaddleOCR-VL Usage Tutorial - 3.2 Client Usage Methods](./PaddleOCR-VL.en.md#32-client-usage-methods). If you run the client on this hardware, make sure to specify `device="gpu"`.
+
+### 3.3 Performance Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 3.3 Performance Tuning](./PaddleOCR-VL.en.md#33-performance-tuning).
+
+## 4. Service Deployment
+
+**Deployment Methods Supported on This Hardware**
+
+| Deployment method | Status | Notes |
+| --- | --- | --- |
+| Docker Compose deployment | Supported with steps in this guide | Continue with Section 4.1. |
+| Manual deployment | Supported | Complete Section 1. Local Runtime Environment Preparation first, then continue with Section 4.2. |
+
+This section mainly introduces how to deploy PaddleOCR-VL as a service and invoke it. There are two methods available; choose one:
+
+- Method 1: Deploy using Docker Compose.
+
+- Method 2: Manually install dependencies for deployment.
+
+> IMPORTANT:
+> The PaddleOCR-VL service introduced in this section differs from the VLM inference service in the previous section: the latter is responsible for only one part of the complete process (i.e., VLM inference) and is called as an underlying service by the former.
+
+### 4.1 Method 1: Deploy Using Docker Compose
+
+1. Download the Compose file and the environment variable configuration file separately from [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu-sm120/compose.yaml) and [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu-sm120/.env) to your local machine.
+
+2. Execute the following command in the directory containing the `compose.yaml` and `.env` files to start the server, which will listen on port **8080** by default:
+
+    ```shell
+    # Must be executed in the directory containing compose.yaml and .env files
+    docker compose up
+    ```
+
+    > TIP:
+    > The image tags used by `compose.yaml` are usually controlled by `API_IMAGE_TAG_SUFFIX` and `VLM_IMAGE_TAG_SUFFIX` in `.env`, and default to tags such as `latest-nvidia-gpu-offline`. To make sure you pull the newest `latest` images, run `docker compose pull` in the current directory before `docker compose up`.
+    > To use an image corresponding to a specific PaddleOCR version, replace `latest` in these variables with `paddleocr<major>.<minor>`, for example `paddleocr3.3-nvidia-gpu-offline`.
+
+    After startup, you will see output similar to the following:
+
+    ```text
+    paddleocr-vl-api             | INFO:     Started server process [1]
+    paddleocr-vl-api             | INFO:     Waiting for application startup.
+    paddleocr-vl-api             | INFO:     Application startup complete.
+    paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+    ```
+
+This method accelerates VLM inference using the vLLM framework and is more suitable for production environment deployment.
+
+Additionally, after starting the server in this manner, no internet connection is required except for image pulling. For deployment in an offline environment, you can first pull the images involved in the Compose file on a connected machine, export them, and transfer them to the offline machine for import to start the service in an offline environment.
+
+Docker Compose starts two containers sequentially by reading configurations from the `.env` and `compose.yaml` files, running the underlying VLM inference service and the PaddleOCR-VL service (pipeline service) respectively.
+
+The meanings of each environment variable contained in the `.env` file are as follows:
+
+```
+- `API_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to launch the pipeline service.
+- `VLM_BACKEND`: The VLM inference backend.
+- `VLM_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to launch the VLM inference service.
+```
+
+You can modify `compose.yaml` to meet custom requirements, for example:
+
+<details>
+<summary>1. Change the port of the PaddleOCR-VL service</summary>
+
+Edit <code>paddleocr-vl-api.ports</code> in the <code>compose.yaml</code> file to change the port. For example, if you need to change the service port to 8111, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. Specify the GPU used by the PaddleOCR-VL service</summary>
+
+Edit <code>environment</code> in the <code>compose.yaml</code> file to change the GPU used. For example, if you need to use card 1 for deployment, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+-             device_ids: ["0"]
++             device_ids: ["1"]
+              capabilities: [gpu]
+    ...
+  paddleocr-vlm-server:
+    ...
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+-             device_ids: ["0"]
++             device_ids: ["1"]
+              capabilities: [gpu]
+    ...
+```
+
+</details>
+
+<details>
+<summary>3. Adjust VLM server-side configuration</summary>
+
+If you want to adjust the VLM server configuration, refer to <a href="./PaddleOCR-VL.en.md#331-server-side-parameter-adjustment">3.3.1 Server Parameter Adjustment</a> to generate a configuration file.
+
+After generating the configuration file, add the following <code>paddleocr-vlm-server.volumes</code> and <code>paddleocr-vlm-server.command</code> fields to your <code>compose.yaml</code>. Replace <code>/path/to/your_config.yaml</code> with your actual configuration file path.
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes: /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. Adjust pipeline-related configurations (such as model path, batch size, deployment device, etc.)</summary>
+
+Refer to the <a href="./PaddleOCR-VL.en.md#44-pipeline-configuration-adjustment-instructions">4.4 Pipeline Configuration Adjustment Instructions</a> section.
+
+</details>
+
+### 4.2 Method 2: Manually Deployment
+
+Please complete Section 1. Local Runtime Environment Preparation first, then refer to [PaddleOCR-VL Usage Tutorial - 4.2 Method 2: Manual Deployment](./PaddleOCR-VL.en.md#42-method-2-manual-deployment).
+
+### 4.3 Client Invocation Methods
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.3 Client Invocation Methods](./PaddleOCR-VL.en.md#43-client-side-invocation).
+
+### 4.4 Pipeline Configuration Adjustment Instructions
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 4.4 Pipeline Configuration Adjustment Instructions](./PaddleOCR-VL.en.md#44-pipeline-configuration-adjustment-instructions).
+
+## 5. Model Fine-Tuning
+
+Please refer to [PaddleOCR-VL Usage Tutorial - 5. Model Fine-Tuning](./PaddleOCR-VL.en.md#5-model-fine-tuning).
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL-NVIDIA-Blackwell.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL-NVIDIA-Blackwell.md
new file mode 100644
index 0000000..6b577ff
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL-NVIDIA-Blackwell.md
@@ -0,0 +1,368 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL NVIDIA Blackwell 架构 GPU 使用教程
+
+> INFO:
+> 除非另有说明，本教程中提到的 “PaddleOCR-VL” 均指 PaddleOCR-VL 系列模型（如 PaddleOCR-VL-1.6 等）；若特指 PaddleOCR-VL v1 版本，将另行明确标注。
+
+本教程是 PaddleOCR-VL 在 NVIDIA Blackwell 架构 GPU 上的使用指南，涵盖了从本地运行环境准备到服务化部署的完整流程。
+
+NVIDIA Blackwell 架构 GPU 包括但不限于以下几种：
+
+- RTX 5090
+- RTX 5080
+- RTX 5070、RTX 5070 Ti
+- RTX 5060、RTX 5060 Ti
+- RTX 5050
+
+目前 PaddleOCR-VL 已在 RTX 5070 上完成精度、速度验证；鉴于硬件环境的多样性，其他 NVIDIA Blackwell 架构 GPU 的兼容性尚未验证。我们诚挚欢迎社区用户在不同硬件上进行测试并反馈您的运行结果。
+
+教程开始前，**请确认您的 NVIDIA 驱动支持 CUDA 12.9 或以上版本**。
+
+## 本硬件支持的使用目标
+
+请在本硬件教程中按下表继续阅读。
+
+| 目标 | 本硬件上的支持情况 | 从哪里开始阅读 |
+| --- | --- | --- |
+| 本地直接推理 | 支持 | 阅读第 1 节“本地运行环境准备”和第 2 节“快速开始”。 |
+| 客户端 + VLM 推理服务 | 支持 | 先完成本地直接推理，再阅读第 3 节“使用 VLM 推理服务”。 |
+| 完整 API 服务 | 支持 Docker Compose 和手动部署 | 如需 Docker Compose，阅读第 4.1 节；如需手动部署，先完成第 1 节“本地运行环境准备”，然后阅读第 4.2 节；随后继续阅读第 4.3 节客户端调用部分和第 4.4 节产线配置调整部分。 |
+| 模型微调 | 支持 | 阅读第 5 节“模型微调”。 |
+
+如果你只是想先确认本硬件支持哪些推理方式，请参考主教程中的 [PaddleOCR-VL 推理方式与硬件支持矩阵](./PaddleOCR-VL.md#paddleocr-vl-对推理设备的支持情况)。
+
+## 1. 本地运行环境准备
+
+**当前硬件支持的本地运行环境准备方式**
+
+| 本地运行环境准备方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 支持并提供步骤 | 请继续阅读本节的 1.1。 |
+| 手动安装推理引擎和 PaddleOCR | 支持并提供步骤 | 请继续阅读本节的 1.2。 |
+
+此步骤主要介绍如何搭建 PaddleOCR-VL 的本地运行环境，有以下两种方式，任选一种即可：
+
+- 方法一：使用官方 Docker 镜像。
+
+- 方法二：手动安装推理引擎和 PaddleOCR。
+
+**我们强烈推荐采用 Docker 镜像的方式，以最大程度减少可能出现的环境问题。**
+
+### 1.1 方法一：使用 Docker 镜像
+
+我们推荐使用官方 Docker 镜像（要求 Docker 版本 >= 19.03，机器装配有 GPU 且 NVIDIA 驱动支持 CUDA 12.9 或以上版本）：
+
+```shell
+docker run \
+    -it \
+    --gpus all \
+    --network host \
+    --user root \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu-sm120 \
+    /bin/bash
+# 在容器中调用 PaddleOCR CLI 或 Python API
+```
+
+如果您希望在无法连接互联网的环境中使用 PaddleOCR-VL，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu-sm120`（镜像的大小约为 10 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu-sm120-offline`（镜像的大小约为 12 GB）。
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本的 PaddleOCR 镜像，可以将标签中的 `latest` 替换为对应版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.3-nvidia-gpu-sm120-offline`
+
+### 1.2 方法二：手动安装推理引擎和 PaddleOCR
+
+如果您无法使用 Docker，也可以手动安装推理引擎和 PaddleOCR。本文档验证过的 Python 版本范围为 3.9–3.13。
+
+本教程提供 PaddlePaddle 安装步骤；若需使用 Transformers 等其他推理引擎，请参考 [主教程第 1.2 节](./PaddleOCR-VL.md#manual-install-inference-engine-and-paddleocr)。
+
+**我们强烈推荐您在虚拟环境中安装 PaddleOCR-VL，以避免发生依赖冲突。** 例如，使用 Python venv 标准库创建虚拟环境：
+
+```shell
+# 创建虚拟环境
+python -m venv .venv_paddleocr
+# 激活环境
+source .venv_paddleocr/bin/activate
+```
+
+执行如下命令完成安装：
+
+```shell
+# 注意这里安装的是 cu129 的 PaddlePaddle
+python -m pip install paddlepaddle-gpu==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/cu129/
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+> **请注意安装 3.2.1 及以上版本的飞桨框架。**
+
+## 2. 快速开始
+
+请参考 [PaddleOCR-VL 使用教程 - 2. 快速开始](./PaddleOCR-VL.md#2)。
+
+## 3. 使用 VLM 推理服务
+
+本节介绍如何通过 VLM 推理服务接入专用后端。对于当前硬件，这通常用于提升默认配置下的推理性能，以更好满足生产需求。在当前硬件文档中，示例使用 vLLM 和 SGLang 作为 VLM 推理服务后端。
+
+### 3.1 启动 VLM 推理服务
+
+> IMPORTANT:
+> 按照本节说明启动的服务仅负责 PaddleOCR-VL 流程中的 VLM 推理环节，不提供完整的端到端文档解析 API。强烈不建议直接通过 HTTP 请求或使用 OpenAI 客户端调用该服务处理文档图像。若您需要部署具备 PaddleOCR-VL 完整能力的服务，请参考后文的服务化部署部分。
+
+**当前硬件支持的启动方式**
+
+| 启动方式 | 状态 | 说明 |
+| --- | --- | --- |
+| 官方 Docker 镜像 | 支持并提供步骤 | 请继续阅读本节的 3.1.1。 |
+| 通过 PaddleOCR CLI 安装依赖后启动 | 支持并提供步骤 | 请继续阅读本节的 3.1.2。 |
+| 直接使用推理加速框架启动 | 未验证 | 当前硬件可通过 vLLM 或 SGLang 后端启动 VLM 推理服务，但尚未验证直接使用原生框架启动的路径。 |
+
+启动 VLM 推理服务有以下两种方式，任选一种即可：
+
+- 方法一：使用官方 Docker 镜像启动服务。
+
+- 方法二：通过 PaddleOCR CLI 手动安装依赖后启动服务。
+
+**我们强烈推荐采用 Docker 镜像的方式，以最大程度减少可能出现的环境问题。**
+
+#### 3.1.1 方法一：使用 Docker 镜像
+
+PaddleOCR 提供了 Docker 镜像，用于快速启动 vLLM 推理服务。可使用以下命令启动服务（要求 Docker 版本 >= 19.03，机器装配有 GPU 且 NVIDIA 驱动支持 CUDA 12.9 或以上版本）：
+
+```shell
+docker run \
+    -it \
+    --gpus all \
+    --network host \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu-sm120 \
+    paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm
+```
+
+如果您希望在无法连接互联网的环境中启动服务，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu-sm120`（镜像的大小约为 13 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu-sm120-offline`（镜像的大小约为 15 GB）。
+
+启动 vLLM 推理服务时，我们提供了一套默认参数设置。如果您有调整显存占用等更多参数的需求，可以自行配置更多参数。请参考 [3.3.1 服务端参数调整](./PaddleOCR-VL.md#331) 创建配置文件，然后将该文件挂载到容器中，并在启动服务的命令中使用 `backend_config` 指定配置文件，例如：
+
+```shell
+docker run \
+    -it \
+    --rm \
+    --gpus all \
+    --network host \
+    -v ./vllm_config.yml:/tmp/vllm_config.yml \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu-sm120 \
+    paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /tmp/vllm_config.yml
+```
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本的 PaddleOCR 镜像，可以将标签中的 `latest` 替换为对应版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:paddleocr3.3-nvidia-gpu-sm120-offline`
+
+#### 3.1.2 方法二：通过 PaddleOCR CLI 安装和使用
+
+由于推理加速框架可能与当前环境中的包存在依赖冲突，建议在虚拟环境中安装：
+
+```shell
+# 如果当前存在已激活的虚拟环境，先通过 `deactivate` 取消激活
+# 创建虚拟环境
+python -m venv .venv_vlm
+# 激活环境
+source .venv_vlm/bin/activate
+```
+
+vLLM 和 SGLang 依赖 FlashAttention，而安装 FlashAttention 时可能需要使用 nvcc 等 CUDA 编译工具。如果您的环境中没有这些工具（例如在使用 `paddleocr-vl` 镜像），可以从 [此仓库](https://github.com/mjun0812/flash-attention-prebuild-wheels) 获取 FlashAttention 的预编译版本（要求 2.8.3 版本），先安装预编译包，再执行后续命令。例如，在 `paddleocr-vl` 镜像中，执行 `python -m pip install https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.3.14/flash_attn-2.8.3+cu128torch2.8-cp310-cp310-linux_x86_64.whl`。对于 FastDeploy，无需执行此步骤。
+
+安装 PaddleOCR 及推理加速服务依赖，以 vLLM 为例：
+
+```shell
+# 安装 PaddleOCR
+python -m pip install "paddleocr[doc-parser]"
+# 安装推理加速服务依赖
+paddleocr install_genai_server_deps vllm
+```
+
+`paddleocr install_genai_server_deps` 命令用法：
+
+```shell
+paddleocr install_genai_server_deps <推理加速框架名称>
+```
+
+当前支持的框架名称为 `vllm` 和 `sglang`，分别对应 vLLM 和 SGLang。
+
+> WARNING:
+> 目前 vLLM 和 SGLang 与 Transformers 引擎所需的 transformers 库版本存在冲突，因此同一环境中无法同时安装 Transformers 引擎与 vLLM 或 SGLang。如果使用 Transformers + vLLM 或 Transformers + SGLang 的推理方式，请将版面分析模型和 VLM 服务部署在不同环境中。
+
+安装完成后，可通过 `paddleocr genai_server` 命令启动服务：
+
+```shell
+paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --backend vllm --port 8118
+```
+
+该命令支持的参数如下：
+
+| 参数                 | 说明                        |
+| ------------------ | ------------------------- |
+| `--model_name`     | 模型名称                      |
+| `--model_dir`      | 模型目录                      |
+| `--host`           | 服务器主机名                    |
+| `--port`           | 服务器端口号                    |
+| `--backend`        | 后端名称，即使用的推理加速框架名称，可选 `vllm` 或 `sglang` |
+| `--backend_config` | 可指定 YAML 文件，包含后端配置        |
+
+### 3.2 客户端使用方法
+
+客户端调用方式请参考 [PaddleOCR-VL 使用教程 - 3.2 客户端使用方法](./PaddleOCR-VL.md#32)。如需在当前硬件上运行客户端，请注意指定 `device="gpu"`。
+
+### 3.3 性能调优
+
+请参考[PaddleOCR-VL 使用教程 - 3.3 性能调优](./PaddleOCR-VL.md#33)。
+
+## 4. 服务化部署
+
+**当前硬件支持的部署方式**
+
+| 部署方式 | 状态 | 说明 |
+| --- | --- | --- |
+| Docker Compose 部署 | 支持并提供步骤 | 请继续阅读本节的 4.1。 |
+| 手动部署 | 支持 | 请先完成第 1 节“本地运行环境准备”，再继续阅读本节的 4.2。 |
+
+此步骤主要介绍如何将 PaddleOCR-VL 部署为服务并调用，有以下两种方式，任选一种即可：
+
+- 方法一：使用 Docker Compose 部署。
+
+- 方法二：手动安装依赖部署。
+
+> IMPORTANT:
+> 本节所介绍的 PaddleOCR-VL 服务与上一节中的 VLM 推理服务有所区别：后者仅负责完整流程中的一个环节（即 VLM 推理），并作为前者的底层服务被调用。
+
+### 4.1 方法一：使用 Docker Compose 部署
+
+1. 分别从 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu-sm120/compose.yaml) 和 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu-sm120/.env) 获取 Compose 文件与环境变量配置文件并下载到本地。
+
+2. 在 `compose.yaml` 和 `.env` 文件所在目录下执行以下命令启动服务器，默认监听 **8080** 端口：
+
+    ```shell
+    # 必须在 compose.yaml 和 .env 文件所在的目录中执行
+    docker compose up
+    ```
+
+    > 提示：
+    > `compose.yaml` 中使用的镜像标签通常由 `.env` 中的 `API_IMAGE_TAG_SUFFIX` 和 `VLM_IMAGE_TAG_SUFFIX` 控制，默认使用 `latest-nvidia-gpu-offline` 等标签。如需确保拉取到最新的 `latest` 镜像，可先在当前目录执行 `docker compose pull`，再执行 `docker compose up`。
+    > 如果希望使用特定版本的 PaddleOCR 镜像，可将这两个环境变量中的 `latest` 替换为对应版本号 `paddleocr<major>.<minor>`，例如 `paddleocr3.3-nvidia-gpu-offline`。
+
+    启动后将看到类似如下输出：
+
+    ```text
+    paddleocr-vl-api             | INFO:     Started server process [1]
+    paddleocr-vl-api             | INFO:     Waiting for application startup.
+    paddleocr-vl-api             | INFO:     Application startup complete.
+    paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+    ```
+
+此方式基于 vLLM 框架对 VLM 推理进行加速，更适合生产环境部署。
+
+此外，使用此方式启动服务器后，除拉取镜像外，无需连接互联网。如需在离线环境中部署，可先在联网机器上拉取 Compose 文件中涉及的镜像，导出并传输至离线机器中导入，即可在离线环境下启动服务。
+
+Docker Compose 通过读取 `.env` 和 `compose.yaml` 文件中配置，先后启动 2 个容器，分别运行底层 VLM 推理服务，以及 PaddleOCR-VL 服务（产线服务）。
+
+`.env` 文件中包含的各环境变量含义如下：
+
+- `API_IMAGE_TAG_SUFFIX`：启动产线服务使用的镜像的标签后缀。
+- `VLM_BACKEND`：VLM 推理后端。
+- `VLM_IMAGE_TAG_SUFFIX`：启动 VLM 推理服务使用的镜像的标签后缀。
+
+您可以通过修改 `compose.yaml` 来满足自定义需求，例如：
+
+<details>
+<summary>1. 更改 PaddleOCR-VL 服务的端口</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>paddleocr-vl-api.ports</code> 来更改端口。例如，如果您需要将服务端口更换为 8111，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. 指定 PaddleOCR-VL 服务所使用的 GPU</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>environment</code> 来更改所使用的 GPU。例如，如果您需要使用卡 1 进行部署，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+-             device_ids: ["0"]
++             device_ids: ["1"]
+              capabilities: [gpu]
+    ...
+  paddleocr-vlm-server:
+    ...
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+-             device_ids: ["0"]
++             device_ids: ["1"]
+              capabilities: [gpu]
+    ...
+```
+
+</details>
+
+<details>
+<summary>3. 调整 VLM 服务端配置</summary>
+
+若您想调整 VLM 服务端的配置，可以参考 <a href="./PaddleOCR-VL.md#331">3.3.1 服务端参数调整</a> 生成配置文件。
+
+生成配置文件后，将以下的 <code>paddleocr-vlm-server.volumes</code> 和 <code>paddleocr-vlm-server.command</code> 字段增加到您的 <code>compose.yaml</code> 中。请将 <code>/path/to/your_config.yaml</code> 替换为您的实际配置文件路径。
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes: /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. 调整产线相关配置（如模型路径、批处理大小、部署设备等）</summary>
+
+参考 <a href="./PaddleOCR-VL.md#44">4.4 产线配置调整说明</a> 小节。
+
+</details>
+
+### 4.2 方法二：手动部署
+
+请先完成第 1 节“本地运行环境准备”，再参考[PaddleOCR-VL 使用教程 - 4.2 方法二：手动部署](./PaddleOCR-VL.md#42)。
+
+### 4.3 客户端调用方式
+
+请参考[PaddleOCR-VL 使用教程 - 4.3 客户端调用方式](./PaddleOCR-VL.md#43)。
+
+### 4.4 产线配置调整说明
+
+请参考[PaddleOCR-VL 使用教程 - 4.4 产线配置调整说明](./PaddleOCR-VL.md#44)。
+
+## 5. 模型微调
+
+请参考[PaddleOCR-VL 使用教程 - 5. 模型微调](./PaddleOCR-VL.md#5)。
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL.en.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL.en.md
new file mode 100644
index 0000000..719c3a4
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL.en.md
@@ -0,0 +1,3369 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL Usage Tutorial
+
+> INFO:
+> PaddleOCR provides a unified interface for the PaddleOCR-VL model series to facilitate quick setup and usage. Unless otherwise specified, the term "PaddleOCR-VL" in this tutorial and related hardware usage tutorials refers to the PaddleOCR-VL model series (e.g., PaddleOCR-VL-1.6). References specific to the PaddleOCR-VL v1 version will be explicitly noted.
+
+PaddleOCR-VL is an advanced and efficient document parsing model designed specifically for element recognition in documents. Taking its initial version (PaddleOCR-VL v1) as an example, its core component is PaddleOCR-VL-0.9B, a compact yet powerful Vision-Language Model (VLM) composed of a NaViT-style dynamic resolution visual encoder and the ERNIE-4.5-0.3B language model, enabling precise element recognition. The model series supports 109 languages and excels in recognizing complex elements (such as text, tables, formulas, and charts) while maintaining extremely low resource consumption. Comprehensive evaluations on widely used public benchmarks and internal benchmarks demonstrate that PaddleOCR-VL achieves SOTA performance in both page-level document parsing and element-level recognition. It significantly outperforms existing Pipeline-based solutions, document parsing multimodal schemes, and advanced general-purpose multimodal large models, while offering faster inference speeds. These advantages make it highly suitable for deployment in real-world scenarios.
+
+On January 29, 2026, we released PaddleOCR-VL-1.5. PaddleOCR-VL-1.5 not only significantly improved the accuracy on the OmniDocBench v1.5 evaluation set to 94.5%, but also innovatively supports irregular-shaped bounding box localization. As a result, PaddleOCR-VL-1.5 demonstrates outstanding performance in real-world scenarios such as Skew, Warping, Screen Photography, Illumination, and Scanning. In addition, the model has added new capabilities for seal (stamp) recognition and text detection and recognition, with key metrics continuing to lead the industry.
+
+**On May 28, 2026, we released PaddleOCR-VL-1.6. With an accuracy of 96.3%, PaddleOCR-VL-1.6 once again set a new benchmark on OmniDocBench v1.6, while also achieving new state-of-the-art (SOTA) results on OmniDocBench v1.5 and Real5-OmniDocBench. It delivers industry-leading performance in text, formula, and table recognition across both open-source and proprietary solutions. In addition, the model shows substantial improvements in ancient document and rare character recognition, as well as significantly enhanced capabilities in multiple scenarios such as seal recognition, spotting, and chart understanding. The model architecture remains fully consistent with PaddleOCR-VL-1.5, enabling seamless migration at zero cost.**
+
+PaddleOCR-VL consists of two core stages: layout analysis and VLM-based recognition. The simplified workflow is illustrated as follows:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/5f2c42665e6a97d5726fe553241acd79361159a0/images/paddleocr_vl_1_5/process_step_EN.png"/>
+
+In this pipeline, the first stage is layout analysis: the model takes the entire image as input, detects and localizes various layout elements (e.g., tables and formulas), determines their reading order, and crops the corresponding element-level sub-images based on the detection results. The second stage is VLM-based recognition: each sub-image is independently fed into the VLM to produce its recognition result (e.g., Markdown text), after which all element-level outputs are merged according to the reading order determined in the layout analysis stage to form the complete parsing result of the entire image. Therefore, **to fully leverage the capabilities of PaddleOCR-VL, it is necessary to adopt the complete pipeline that integrates layout analysis and VLM-based recognition, rather than using the VLM component alone.** This distinction will be referenced multiple times in the following sections, so it is important to clearly differentiate between the full PaddleOCR-VL pipeline and its VLM component. Taking PaddleOCR-VL v1 as an example, the layout analysis model is PP-DocLayoutV2, and the VLM component is PaddleOCR-VL-0.9B. It is important to note that “PaddleOCR-VL-0.9B” does not represent a standalone variant of PaddleOCR-VL, but rather refers specifically to the VLM component within the complete PaddleOCR-VL v1 pipeline. This differs from common naming conventions of LLMs/VLMs—for instance, Qwen2-72B typically denotes a specific model variant within the Qwen2 series.
+
+**If issues arise during usage—such as failure to reproduce the performance reported in the paper or on the PaddleOCR official website, or the generation of excessive hallucinated text—the first step is to verify whether the complete PaddleOCR-VL pipeline is being used, rather than only the VLM component.** For example, directly running the PaddleOCR-VL-0.9B model locally via Transformers, or sending requests to services such as vLLM, SGLang, or FastDeploy, is not equivalent to executing the full PaddleOCR-VL pipeline.
+
+## Start Here
+
+Choose the guide that matches your hardware first.
+
+| Hardware                      | Read this guide                                                                                                          |
+| -------------------------------| --------------------------------------------------------------------------------------------------------------------------|
+| x64 CPU                       | Continue with this tutorial. Use the manual installation path in Section 1.2; the NVIDIA-only Docker steps do not apply. |
+| NVIDIA GPU (except Blackwell) | Continue with this tutorial.                                                                                             |
+| NVIDIA Blackwell GPU          | Read [PaddleOCR-VL NVIDIA Blackwell-Architecture GPUs Usage Tutorial](./PaddleOCR-VL-NVIDIA-Blackwell.en.md).            |
+| Apple Silicon                 | Read [PaddleOCR-VL Apple Silicon Usage Tutorial](./PaddleOCR-VL-Apple-Silicon.en.md).                                    |
+| Kunlunxin XPU                 | Read [PaddleOCR-VL Kunlunxin XPU Usage Tutorial](./PaddleOCR-VL-Kunlunxin-XPU.en.md).                                    |
+| Hygon DCU                     | Read [PaddleOCR-VL Hygon DCU Usage Tutorial](./PaddleOCR-VL-Hygon-DCU.en.md).                                            |
+| MetaX GPU                     | Read [PaddleOCR-VL MetaX GPU Usage Tutorial](./PaddleOCR-VL-MetaX-GPU.en.md).                                            |
+| Iluvatar GPU                  | Read [PaddleOCR-VL Iluvatar GPU Usage Tutorial](./PaddleOCR-VL-Iluvatar-GPU.en.md).                                      |
+| Huawei Ascend NPU             | Read [PaddleOCR-VL Huawei Ascend NPU Usage Tutorial](./PaddleOCR-VL-Huawei-Ascend-NPU.en.md).                            |
+| AMD GPU                       | Read [PaddleOCR-VL AMD GPU Usage Tutorial](./PaddleOCR-VL-AMD-GPU.en.md).                                                |
+| Intel Arc GPU                 | Read [PaddleOCR-VL Intel Arc GPU Usage Tutorial](./PaddleOCR-VL-Intel-Arc-GPU.en.md).                                    |
+
+If you just want to first confirm which hardware PaddleOCR-VL can be deployed on, or which inference methods are supported by specific hardware, read the [PaddleOCR-VL Inference Method and Hardware Support Matrix](#inference-device-support-for-paddleocr-vl) before continuing.
+
+## PaddleOCR-VL Inference Method and Hardware Support Matrix {#inference-device-support-for-paddleocr-vl}
+
+
+<table border="1">
+<thead>
+  <tr>
+    <th>Inference Method</th>
+    <th>NVIDIA GPU</th>
+    <th>Kunlunxin XPU</th>
+    <th>Hygon DCU</th>
+    <th>MetaX GPU</th>
+    <th>Iluvatar GPU</th>
+    <th>Huawei Ascend NPU</th>
+    <th>x64 CPU</th>
+    <th>Apple Silicon</th>
+    <th>AMD GPU</th>
+    <th>Intel Arc GPU</th>
+  </tr>
+</thead>
+<tbody>
+  <tr style="text-align: center;">
+    <td>PaddlePaddle</td>
+    <td>✅</td>
+    <td>✅</td>
+    <td>✅</td>
+    <td>✅</td>
+    <td>✅</td>
+    <td>🚧</td>
+    <td>✅</td>
+    <td>✅</td>
+    <td>✅</td>
+    <td>🚧</td>
+  </tr>
+  <tr style="text-align: center;">
+    <td>Transformers</td>
+    <td>✅</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>✅</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+  </tr>
+  <tr style="text-align: center;">
+    <td>PaddlePaddle + vLLM</td>
+    <td>✅</td>
+    <td>🚧</td>
+    <td>✅</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>✅</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>✅</td>
+    <td>✅</td>
+  </tr>
+  <tr style="text-align: center;">
+    <td>PaddlePaddle + SGLang</td>
+    <td>✅</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>🚧</td>
+    <td>🚧</td>
+  </tr>
+  <tr style="text-align: center;">
+    <td>PaddlePaddle + FastDeploy</td>
+    <td>✅</td>
+    <td>✅</td>
+    <td>🚧</td>
+    <td>✅</td>
+    <td>✅</td>
+    <td>🚧</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>🚧</td>
+    <td>🚧</td>
+  </tr>
+  <tr style="text-align: center;">
+    <td>PaddlePaddle + MLX-VLM</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>✅</td>
+    <td>❌</td>
+    <td>❌</td>
+  </tr>
+  <tr style="text-align: center;">
+    <td>PaddlePaddle + llama.cpp</td>
+    <td>✅</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>✅</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+  </tr>
+  <tr style="text-align: center;">
+    <td>Transformers + vLLM</td>
+    <td>✅</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>🚧</td>
+    <td>🚧</td>
+  </tr>
+  <tr style="text-align: center;">
+    <td>Transformers + SGLang</td>
+    <td>✅</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>🚧</td>
+    <td>🚧</td>
+  </tr>
+  <tr style="text-align: center;">
+    <td>Transformers + FastDeploy</td>
+    <td>✅</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>🚧</td>
+    <td>🚧</td>
+  </tr>
+  <tr style="text-align: center;">
+    <td>Transformers + MLX-VLM</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>❌</td>
+    <td>✅</td>
+    <td>❌</td>
+    <td>❌</td>
+  </tr>
+  <tr style="text-align: center;">
+    <td>Transformers + llama.cpp</td>
+    <td>✅</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>✅</td>
+    <td>🚧</td>
+    <td>🚧</td>
+    <td>🚧</td>
+  </tr>
+</tbody>
+</table>
+
+**Status Legend**
+
+- `✅`: Supported.
+- `🚧`: In progress or pending further verification.
+- `❌`: Not currently supported or not applicable.
+
+**Inference Method Notes**
+
+`PaddlePaddle` means that both the layout analysis model and the VLM run locally using the PaddlePaddle framework. In practice, each module is resolved to `paddle_static` or `paddle_dynamic` depending on the model form. `Transformers` means that both the layout analysis model and the VLM run locally via the `transformers` engine. The remaining methods follow the format `layout analysis model inference method + VLM inference method`, where the VLM part is handled by a separate inference service; for example, `PaddlePaddle + vLLM` means the layout analysis model runs locally on the client using PaddlePaddle, while the VLM is served by vLLM.
+
+**Additional Notes for Interpreting the Matrix**
+
+- vLLM, SGLang, and FastDeploy cannot run natively on Windows. Please use our Docker images.
+- Due to dependency conflicts between different libraries, mixed inference methods such as Transformers + vLLM usually require deploying the layout analysis model and the VLM service in separate environments.
+
+## Environment Requirements
+
+- When using NVIDIA GPUs for inference, ensure that the Compute Capability (CC) and CUDA version meet the requirements:
+  - PaddlePaddle: CC ≥ 7.0, CUDA ≥ 11.8
+  - Transformers: CC ≥ 7.0, CUDA ≥ 11.8
+  - vLLM: CC ≥ 8.0, CUDA ≥ 12.6
+  - SGLang: 8.0 ≤ CC < 12.0, CUDA ≥ 12.6
+  - FastDeploy: 8.0 ≤ CC < 12.0, CUDA ≥ 12.6
+  - Common GPUs with CC ≥ 8 include the RTX 30/40/50 series and A10/A100. See [CUDA GPU Compute Capability](https://developer.nvidia.com/cuda-gpus) for more models.
+- Although vLLM can start on NVIDIA GPUs with CC 7.x, such as T4/V100, it is prone to timeouts or OOM and is therefore not recommended.
+
+## Workflow Guide for This Tutorial
+
+This tutorial is the default path for x64 CPU users and NVIDIA GPU users other than Blackwell.
+
+| Goal | Availability in this tutorial | Read this section |
+| --- | --- | --- |
+| Local direct inference | Supported. x64 CPU users should use the manual installation path in Section 1.2. | Read Section 1. Local Runtime Environment Preparation and Section 2. Quick Start. |
+| Client + VLM inference service | Supported. | Complete local direct inference first, then read Section 3. Using VLM Inference Services. |
+| Full API service | Supported. x64 CPU users should use the manual deployment path; non-Blackwell NVIDIA GPU users can use Docker Compose or manual deployment. | Read Section 4. Service Deployment: use Section 4.1 for Docker Compose, or Section 4.2 for manual deployment (complete Section 1. Local Runtime Environment Preparation first). For higher concurrency, refer to the [High-Performance Service Deployment solution](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/hps/README_en.md). |
+| Model fine-tuning | Supported. | Read Section 5. Model Fine-Tuning. |
+
+## 1. Local Runtime Environment Preparation
+
+This section explains how to set up the local runtime environment for PaddleOCR-VL. This tutorial mainly applies to **x64 CPU** users and **NVIDIA GPU** users other than Blackwell. For other hardware, please refer first to the dedicated tutorials listed above.
+
+This tutorial provides the following two methods for local runtime environment preparation:
+
+- Method 1: Use the official Docker image (NVIDIA GPU only).
+
+- Method 2: Manually install the inference engine and PaddleOCR (available for both x64 CPU and NVIDIA GPU).
+
+**We strongly recommend using the Docker image to minimize potential environment-related issues.**
+
+### 1.1 Method 1: Using Docker Image
+
+We recommend using the official Docker image (requires Docker version >= 19.03, GPU-equipped machine with NVIDIA drivers supporting CUDA 12.6 or later):
+
+```shell
+docker run \
+    -it \
+    --gpus all \
+    --network host \
+    --user root \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu \
+    /bin/bash
+# Invoke PaddleOCR CLI or Python API within the container
+```
+
+If you need to use PaddleOCR-VL in an offline environment, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu` (image size approximately 8 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu-offline` (image size is approximately 10 GB). You will need to pull the image on an internet-connected machine, import it into the offline machine, and then start the container using this image on the offline machine. For example:
+
+```shell
+# Execute on an internet-connected machine
+docker pull ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu-offline
+# Save the image to a file
+docker save ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu-offline -o paddleocr-vl-latest-nvidia-gpu-offline.tar
+
+# Transfer the image file to the offline machine
+
+# Execute on the offline machine
+docker load -i paddleocr-vl-latest-nvidia-gpu-offline.tar
+# After that, you can use `docker run` to start the container on the offline machine
+```
+
+The image comes preinstalled with the PaddlePaddle framework and does not include any other inference engines. If you want to use other inference engines, it is recommended to install them manually using Method 2 (it is not recommended to install them in an environment where the PaddlePaddle framework is preinstalled).
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use a specific version of the image, you can replace `latest` in the tag with the desired PaddleOCR version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.3-nvidia-gpu-offline`
+
+
+### 1.2 Method 2: Manually Install the Inference Engine and PaddleOCR {#manual-install-inference-engine-and-paddleocr}
+
+If you cannot use Docker, you can manually install the inference engine and PaddleOCR. This guide documents Python 3.9–3.13 as the verified range.
+
+**We strongly recommend installing PaddleOCR-VL in a virtual environment to avoid dependency conflicts.** For example, use the Python venv standard library to create a virtual environment:
+
+```shell
+# Create a virtual environment
+python -m venv .venv_paddleocr
+# Activate the environment
+source .venv_paddleocr/bin/activate
+```
+
+Please first install the dependencies corresponding to your chosen inference engine:
+
+- If you use PaddlePaddle for inference, install PaddlePaddle 3.2.1 or later (**do not install both the CPU and GPU versions of PaddlePaddle at the same time**). Common installation commands are as follows:
+
+```shell
+# NVIDIA GPU (CUDA 12.6 as an example)
+python -m pip install paddlepaddle-gpu==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/cu126/
+
+# x64 CPU
+python -m pip install paddlepaddle==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+```
+
+  For other CUDA versions, please refer to the PaddlePaddle installation guide: <https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/en/develop/install/pip/linux-pip_en.html>
+
+- If you use `transformers` for inference, refer to the [official Transformers documentation](https://huggingface.co/docs/transformers/installation) to install `transformers` and the required low-level inference framework dependencies.
+
+After installing the inference engine, run the following command to install the base package required by PaddleOCR-VL:
+
+```shell
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+## 2. Quick Start
+
+This section introduces how to use PaddleOCR-VL through the CLI and Python API.
+
+PaddleOCR-VL supports both CLI and Python API usage. The CLI method is simpler and suitable for quick verification, while the Python API is more flexible and suitable for integration into existing projects. The examples below use local PaddlePaddle by default, with each module resolving to `paddle_static` or `paddle_dynamic` as appropriate. To switch to the `transformers` engine, append `--engine transformers` in the CLI, or pass `engine="transformers"` when initializing the Python API.
+
+> IMPORTANT:
+> The methods introduced in this section are primarily for rapid validation. Their inference speed, memory usage, and stability may not meet the requirements of a production environment. **If deployment to a production environment is needed, we strongly recommend using a dedicated VLM inference service**. For specific methods, please refer to the next section.
+
+### 2.1 Command Line Usage
+
+When you run PaddleOCR-VL for the first time, it will automatically download the official model files. Please make sure the current environment has internet access and allow some extra time for downloading and initialization.
+
+The following are ready-to-copy example commands. For the first try, we recommend adding `--save_path ./output` so that you can inspect the saved results in the current directory:
+
+```shell
+# NVIDIA GPU
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --save_path ./output
+
+# Kunlunxin XPU
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --device xpu --save_path ./output
+
+# Hygon DCU
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --device dcu --save_path ./output
+
+# MetaX GPU
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --device metax_gpu --save_path ./output
+
+# Apple Silicon
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --device cpu --save_path ./output
+
+# Huawei Ascend NPU
+# For Huawei Ascend NPU, please refer to Chapter 3 and use PaddlePaddle + vLLM for inference
+
+# Use --use_doc_orientation_classify to enable document orientation classification
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --use_doc_orientation_classify True --save_path ./output
+
+# Use --use_doc_unwarping to enable the document unwarping module
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --use_doc_unwarping True --save_path ./output
+
+# Use --use_layout_detection to disable the layout analysis and ordering module
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --use_layout_detection False --save_path ./output
+```
+
+After successful execution, the terminal will print the structured result. If you set `--save_path ./output`, the result files will also be saved under the `output` directory in the current working directory for further inspection and debugging.
+
+To switch to the `transformers` engine, use:
+
+```bash
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --engine transformers --save_path ./output
+```
+
+<details><summary><b>Command line supports more parameters. Click to expand for detailed parameter descriptions</b></summary>
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted, required. <br/>
+<b>Description:</b> 
+For example, the local path of an image file or PDF file: <code>/root/data/img.jpg</code>;<b>Such as a URL link</b>, for example, the network URL of an image file or PDF file:<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/demo_paper.png">Example</a>;<b>Such as a local directory</b>, which should contain the images to be predicted, for example, the local path: <code>/root/data/</code>(Currently, prediction for directories containing PDF files is not supported. PDF files need to be specified with a specific file path).</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td><b>Meaning:</b>Specify the path where the inference result file will be saved. <br/>
+<b>Description:</b> 
+If not set, the inference results will not be saved locally.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+  <td><code>pipeline_version</code></td>
+  <td>
+    <b>Meaning:</b> Specifies the pipeline version.<br/>
+    <b>Description:</b> The currently available values are <code>"v1"</code>, <code>"v1.5"</code>, and <code>"v1.6"</code>.
+  </td>
+  <td><code>str</code></td>
+  <td>"v1.6"</td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the layout analysis model.<br/>
+<b>Description:</b>
+ If not set, the default model of the production line will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the layout analysis model.<br/> 
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>Score threshold for the layout model. <br/>
+<b>Description:</b> 
+Any value between  <code>0-1</code>. If not set, the default value is used, which is  <code>0.5</code>.
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Whether to use post-processing NMS for layout analysis. <br/>
+<b>Description:</b> 
+If not set, the initialized default value will be used.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion coefficient for the detection boxes of the layout area detection model. <br/>
+<b>Description:</b> 
+Any floating-point number greater than <code>0</code>. If not set, the initialized default value will be used.</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>Merging mode for the detection boxes output by the model in layout analysis. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>large</b> when set to large, it means that among the detection boxes output by the model, for overlapping and contained boxes, only the outermost largest box is retained, and the overlapping inner boxes are deleted;</li>
+<li><b>small</b>, when set to small, it means that among the detection boxes output by the model, for overlapping and contained boxes, only the innermost contained small box is retained, and the overlapping outer boxes are deleted;</li>
+<li><b>union</b>,no filtering is performed on the boxes, and both inner and outer boxes are retained;</li></ul>
+If not set, the initialized parameter value will be used.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>vl_rec_model_name</code></td>
+<td><b>Meaning:</b>Name of the multimodal recognition model. <br/>
+<b>Description:</b> 
+If not set, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>vl_rec_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the multimodal recognition model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>vl_rec_backend</code></td>
+<td><b>Meaning:</b>Inference backend used by the multimodal recognition model.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>vl_rec_server_url</code></td>
+<td><b>Description:</b>If the multimodal recognition model uses an inference service, this parameter is used to specify the server URL.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>vl_rec_max_concurrency</code></td>
+<td><b>Meaning:</b>If the multimodal recognition model uses an inference service, this parameter is used to specify the maximum number of concurrent requests.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>vl_rec_api_model_name</code></td>
+<td><b>Meaning:</b>If the multimodal recognition model uses an inference service, this parameter is used to specify the model name of the service.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>vl_rec_api_key</code></td>
+<td><b>Meaning:</b>If the multimodal recognition model uses an inference service, this parameter is used to specify the API key of the service.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>Name of the document orientation classification model.<br/>
+<b>Description:</b>
+If not set, the initialized default value will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the document orientation classification model. <br/>
+<b>Description:</b>
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>Meaning:</b>Name of the text image rectification model. <br/>
+<b>Description:</b>
+If not set, the initialized default value will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text image rectification model. <br/>
+<b>Description:</b>
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to load and use the document orientation classification module. <br/>
+<b>Description:</b>
+If not set, the initialized default value will be used, which is initialized to<code>False</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to load and use the text image rectification module. <br/>
+<b>Description:</b>
+If not set, the initialized default value will be used, which is initialized to <code>False</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>Meaning:</b>Whether to load and use the layout analysis module. <br/>
+<b>Description:</b>
+If not set, the initialized default value will be used, which is initialized to <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>Meaning:</b>Whether to use the chart parsing function. <br/>
+<b>Description:</b>
+If not set, the initialized default value will be used, which is initialized to <code>False</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>Meaning:</b>Whether to use the seal recognition function.<br/>
+<b>Description:</b> If not set, the initialized default value will be used, which defaults to initialization as <code>False</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_ocr_for_image_block</code></td>
+<td><b>Meaning:</b>Whether to perform OCR on text within image blocks.<br/>
+<b>Description:</b> If not set, the initialized default value will be used, which defaults to initialization as <code>False</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>format_block_content</code></td>
+<td><b>Meaning:</b>Controls whether to format the <code>block_content</code> content within as Markdown. <br/>
+<b>Description:</b>
+If not set, the initialized default value will be used, which defaults to initialization as<code>False</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>merge_layout_blocks</code></td>
+<td><b>Meaning:</b>Control whether to merge the layout detection boxes for cross-column or staggered top and bottom columns.<br/>
+<b>Description:</b> If not set, the initialized default value will be used, which defaults to initialization as<code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>markdown_ignore_labels</code></td>
+<td><b>Meaning:</b>Layout labels that need to be ignored in Markdown.<br/>
+<b>Description:</b> If not set, the initialized default value will be used, which defaults to initialization as<code>['number','footnote','header','header_image','footer','footer_image','aside_text']</code>.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+  <td><code>layout_shape_mode</code></td>
+  <td>
+    <b>Meaning:</b>Specifies the geometric representation mode for layout analysis results. It defines how the boundaries of detected regions (e.g., text blocks, images, tables) are calculated and displayed.<br/>
+    <b>Description:</b> Value descriptions:
+    <ul>
+      <li>
+        <b>rect (rectangle)</b>:
+        Outputs an axis-aligned bounding box (including x1, y1, x2, y2).
+        Suitable for standard horizontally aligned layouts.
+      </li>
+      <li>
+        <b>quad (quadrilateral)</b>:
+        Outputs an arbitrary quadrilateral composed of four vertices.
+        Suitable for regions with skew or perspective distortion.
+      </li>
+      <li>
+        <b>poly (polygon)</b>:
+        Outputs a closed contour composed of multiple coordinate points.
+        Suitable for irregularly shaped or curved layout elements,
+        offering the highest precision.
+      </li>
+      <li>
+        <b>auto (automatic)</b>:
+        The system automatically selects the most appropriate shape
+        representation based on the complexity and confidence of the
+        detected targets.
+      </li>
+    </ul>
+  </td>
+  <td><code>str</code></td>
+  <td>"auto"</td>
+</tr>
+<tr>
+<td><code>use_queues</code></td>
+<td><b>Meaning:</b>Used to control whether to enable internal queues.<br/>
+<b>Description:</b> When set to <code>True</code>, data loading (such as rendering PDF pages as images), layout analysis model processing, and VLM inference will be executed asynchronously in separate threads, with data passed through queues, thereby improving efficiency. This approach is particularly efficient for PDF documents with a large number of pages or directories containing a large number of images or PDF files. If not set, the initialized default value will be used, which defaults to initialization as <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>prompt_label</code></td>
+<td><b>Meaning:</b>The prompt type setting for the VL model, which takes effect if and only if <code>use_layout_detection=False</code>.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>repetition_penalty</code></td>
+<td><b>Meaning:</b>The repetition penalty parameter used in VL model sampling.</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>temperature</code></td>
+<td><b>Meaning:</b>The temperature parameter used in VL model sampling.</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>top_p</code></td>
+<td><b>Meaning:</b>The top-p parameter used in VL model sampling.</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>min_pixels</code></td>
+<td><b>Meaning:</b>The minimum number of pixels allowed when the VL model preprocesses images.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>max_pixels</code></td>
+<td><b>Meaning:</b>The maximum number of pixels allowed when the VL model preprocesses images.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>The device used for inference. <br/>
+<b>Description:</b>
+Supports specifying specific card numbers:<ul>
+<li><b>CPU</b>: For example,<code>cpu</code> indicates using the CPU for inference;</li>
+<li><b>GPU</b>: For example,<code>gpu:0</code> indicates using the first GPU for inference;</li>
+<li><b>NPU</b>: For example,<code>npu:0</code> indicates using the first NPU for inference;</li>
+<li><b>XPU</b>: For example,<code>xpu:0</code> indicates using the first XPU for inference;</li>
+<li><b>MLU</b>: For example,<code>mlu:0</code> indicates using the first MLU for inference;</li>
+<li><b>DCU</b>: For example,<code>dcu:0</code> indicates using the first DCU for inference;</li>
+<li><b>MetaX GPU</b>: For example,<code>metax_gpu:0</code> indicates using the first MetaX GPU for inference;</li>
+<li><b>Iluvatar GPU</b>: For example,<code>iluvatar_gpu:0</code> indicates using the first Iluvatar GPU for inference;</li>
+</ul>If not set, the initialized default value will be used. During initialization, the local GPU device 0 will be used preferentially. If it is not available, the CPU device will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to enable the TensorRT subgraph engine of Paddle Inference.<br/>
+<b>Description:</b>
+If the model does not support TensorRT acceleration, acceleration will not be used even if this flag is set.<br/>
+For CUDA 11.8 versions of PaddlePaddle, the compatible TensorRT version is 8.x (x>=6). TensorRT 8.6.1.6 is recommended.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b> Computation precision, such as <code>fp32</code> or <code>fp16</code>.</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b> Whether to enable MKL-DNN accelerated inference.<br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support MKL-DNN acceleration, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b> MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads used for inference on CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b> Path to the PaddleX pipeline configuration file.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+
+</tbody>
+</table>
+</details>
+<br/>
+
+The inference result will be printed in the terminal. The default output of PaddleOCR-VL is as follows:
+
+<details><summary> 👉Click to expand</summary>
+<pre>
+ <code>
+{'res': {'input_path': 'paddleocr_vl_demo.png', 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_layout_detection': True, 'use_chart_recognition': False, 'format_block_content': False}, 'layout_det_res': {'input_path': None, 'page_index': None, 'boxes': [{'cls_id': 6, 'label': 'doc_title', 'score': 0.9636914134025574, 'coordinate': [np.float32(131.31366), np.float32(36.450516), np.float32(1384.522), np.float32(127.984665)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9281806349754333, 'coordinate': [np.float32(585.39465), np.float32(158.438), np.float32(930.2184), np.float32(182.57469)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9840355515480042, 'coordinate': [np.float32(9.023666), np.float32(200.86115), np.float32(361.41583), np.float32(343.8828)]}, {'cls_id': 14, 'label': 'image', 'score': 0.9871416091918945, 'coordinate': [np.float32(775.50574), np.float32(200.66502), np.float32(1503.3807), np.float32(684.9304)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9801855087280273, 'coordinate': [np.float32(9.532196), np.float32(344.90594), np.float32(361.4413), np.float32(440.8244)]}, {'cls_id': 17, 'label': 'paragraph_title', 'score': 0.9708921313285828, 'coordinate': [np.float32(28.040405), np.float32(455.87976), np.float32(341.7215), np.float32(520.7117)]}, {'cls_id': 24, 'label': 'vision_footnote', 'score': 0.9002962708473206, 'coordinate': [np.float32(809.0692), np.float32(703.70044), np.float32(1488.3016), np.float32(750.5238)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9825374484062195, 'coordinate': [np.float32(8.896561), np.float32(536.54895), np.float32(361.05237), np.float32(655.8058)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9822263717651367, 'coordinate': [np.float32(8.971573), np.float32(657.4949), np.float32(362.01715), np.float32(774.625)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9767460823059082, 'coordinate': [np.float32(9.407074), np.float32(776.5216), np.float32(361.31067), np.float32(846.82874)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9868153929710388, 'coordinate': [np.float32(8.669495), np.float32(848.2543), np.float32(361.64703), np.float32(1062.8568)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9826608300209045, 'coordinate': [np.float32(8.8025055), np.float32(1063.8615), np.float32(361.46588), np.float32(1182.8524)]}, {'cls_id': 22, 'label': 'text', 'score': 0.982555627822876, 'coordinate': [np.float32(8.820602), np.float32(1184.4663), np.float32(361.66394), np.float32(1302.4507)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9584776759147644, 'coordinate': [np.float32(9.170288), np.float32(1304.2161), np.float32(361.48898), np.float32(1351.7483)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9782056212425232, 'coordinate': [np.float32(389.1618), np.float32(200.38202), np.float32(742.7591), np.float32(295.65146)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9844875931739807, 'coordinate': [np.float32(388.73303), np.float32(297.18463), np.float32(744.00024), np.float32(441.3034)]}, {'cls_id': 17, 'label': 'paragraph_title', 'score': 0.9680547714233398, 'coordinate': [np.float32(409.39468), np.float32(455.89386), np.float32(721.7174), np.float32(520.9387)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9741666913032532, 'coordinate': [np.float32(389.71606), np.float32(536.8138), np.float32(742.7112), np.float32(608.00165)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9840384721755981, 'coordinate': [np.float32(389.30988), np.float32(609.39636), np.float32(743.09247), np.float32(750.3231)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9845995306968689, 'coordinate': [np.float32(389.13272), np.float32(751.7772), np.float32(743.058), np.float32(894.8815)]}, {'cls_id': 22, 'label': 'text', 'score': 0.984852135181427, 'coordinate': [np.float32(388.83267), np.float32(896.0371), np.float32(743.58215), np.float32(1038.7345)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9804865717887878, 'coordinate': [np.float32(389.08478), np.float32(1039.9119), np.float32(742.7585), np.float32(1134.4897)]}, {'cls_id': 22, 'label': 'text', 'score': 0.986461341381073, 'coordinate': [np.float32(388.52643), np.float32(1135.8137), np.float32(743.451), np.float32(1352.0085)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9869391918182373, 'coordinate': [np.float32(769.8341), np.float32(775.66235), np.float32(1124.9813), np.float32(1063.207)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9822869896888733, 'coordinate': [np.float32(770.30383), np.float32(1063.938), np.float32(1124.8295), np.float32(1184.2192)]}, {'cls_id': 17, 'label': 'paragraph_title', 'score': 0.9689218997955322, 'coordinate': [np.float32(791.3042), np.float32(1199.3169), np.float32(1104.4521), np.float32(1264.6985)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9713128209114075, 'coordinate': [np.float32(770.4253), np.float32(1279.6072), np.float32(1124.6917), np.float32(1351.8672)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9236552119255066, 'coordinate': [np.float32(1153.9058), np.float32(775.5814), np.float32(1334.0654), np.float32(798.1581)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9857938885688782, 'coordinate': [np.float32(1151.5197), np.float32(799.28015), np.float32(1506.3619), np.float32(991.1156)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9820687174797058, 'coordinate': [np.float32(1151.5686), np.float32(991.91095), np.float32(1506.6023), np.float32(1110.8875)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9866049885749817, 'coordinate': [np.float32(1151.6919), np.float32(1112.1301), np.float32(1507.1611), np.float32(1351.9504)]}]}}}
+</code></pre></details>
+
+For detailed descriptions of the running results and saving interfaces, refer to the result explanation in [2.2 Python Script Integration](#22-python-script-integration).
+
+<b>Note: </b> Since the default model of PaddleOCR-VL is relatively large, inference may be slow. For actual use, it is recommended to use [3. Using VLM Inference Services](#3-vlm) for faster inference.
+
+### 2.2 Python Script Integration
+
+The command line method is intended for quick testing and visualization. In real projects, you usually integrate the model through code. You can quickly run PaddleOCR-VL inference with just a few lines of code:
+
+```python
+from pathlib import Path
+
+from paddleocr import PaddleOCRVL
+
+output_dir = Path("./output")
+output_dir.mkdir(parents=True, exist_ok=True)
+
+# NVIDIA GPU
+pipeline = PaddleOCRVL()
+# Kunlunxin XPU
+# pipeline = PaddleOCRVL(device="xpu")
+# Hygon DCU
+# pipeline = PaddleOCRVL(device="dcu")
+# MetaX GPU
+# pipeline = PaddleOCRVL(device="metax_gpu")
+# Apple Silicon
+# pipeline = PaddleOCRVL(device="cpu")
+# Huawei Ascend NPU 
+# Huawei Ascend NPU please refer to Chapter 3 for inference using PaddlePaddle + vLLM
+
+# pipeline = PaddleOCRVL(use_doc_orientation_classify=True) # Use use_doc_orientation_classify to enable/disable document orientation classification model
+# pipeline = PaddleOCRVL(use_doc_unwarping=True) # Use use_doc_unwarping to enable/disable document unwarping module
+# pipeline = PaddleOCRVL(use_layout_detection=False) # Use use_layout_detection to enable/disable layout analysis module
+
+output = pipeline.predict("https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png")
+for res in output:
+    res.print() ## Print the structured prediction output
+    res.save_to_json(save_path=output_dir) ## Save the current image's structured result in JSON format
+    res.save_to_markdown(save_path=output_dir) ## Save the current image's result in Markdown format
+    res.save_to_word(save_path="output") ## Save the current image's result in Word format
+```
+
+To switch to the `transformers` engine, use:
+
+```python
+from pathlib import Path
+
+from paddleocr import PaddleOCRVL
+
+output_dir = Path("./output")
+output_dir.mkdir(parents=True, exist_ok=True)
+
+pipeline = PaddleOCRVL(engine="transformers")
+output = pipeline.predict("https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png")
+for res in output:
+    res.print() ## Print the structured prediction output
+    res.save_to_json(save_path=output_dir) ## Save the current image's structured result in JSON format
+    res.save_to_markdown(save_path=output_dir) ## Save the current image's result in Markdown format
+    res.save_to_word(save_path="output") ## Save the current image's result in Word format
+```
+
+For PDF files, each page will be processed individually, and a separate Markdown file will be generated for each page. If you wish to perform cross-page table merging, reconstruct multi-level headings, or merge multi-page results, you can achieve this using the following method:
+
+```python
+from pathlib import Path
+
+from paddleocr import PaddleOCRVL
+
+input_file = "./your_pdf_file.pdf"
+output_dir = Path("./output")
+output_dir.mkdir(parents=True, exist_ok=True)
+
+pipeline = PaddleOCRVL()
+
+output = pipeline.predict(input=input_file)
+
+pages_res = list(output)
+
+output = pipeline.restructure_pages(pages_res)
+# output = pipeline.restructure_pages(pages_res, merge_tables=True) # Merge tables across pages
+# output = pipeline.restructure_pages(pages_res, merge_tables=True, relevel_titles=True) # Merge tables across pages and reconstruct multi-level titles
+# output = pipeline.restructure_pages(pages_res, merge_tables=True, relevel_titles=True, concatenate_pages=True) # Merge tables across pages, reconstruct multi-level titles, and merge multiple pages
+
+for res in output:
+    res.print() ## Print the structured prediction output
+    res.save_to_json(save_path=output_dir) ## Save the current image's structured result in JSON format
+    res.save_to_markdown(save_path=output_dir) ## Save the current image's result in Markdown format
+```
+
+If you need to process multiple files, **it is recommended to pass the directory path containing the files or a list of file paths to the `predict` method** to maximize processing efficiency. For example:
+
+```python
+# The `imgs` directory contains multiple images to be processed: file1.png, file2.png, file3.png
+# Pass the directory path
+output = pipeline.predict("imgs")
+# Or pass a list of file paths
+output = pipeline.predict(["imgs/file1.png", "imgs/file2.png", "imgs/file3.png"])
+# Both of the above methods are more efficient than the following approach:
+# for file in ["imgs/file1.png", "imgs/file2.png", "imgs/file3.png"]:
+#     output = pipeline.predict(file)
+```
+
+**Note:**
+
+- In the example code above, the `use_doc_orientation_classify` and `use_doc_unwarping` parameters are both set to `False` by default, meaning document orientation classification and document unwarping are disabled. If you need these features, set them to `True` manually.
+
+
+The above Python script performs the following steps:
+
+<details><summary>(1) Instantiate the pipeline object. Specific parameter descriptions are as follows:</summary>
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Parameter Description</th>
+<th>Parameter Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+  <td><code>pipeline_version</code></td>
+  <td>
+    <b>Meaning:</b> Specifies the pipeline version.<br/>
+    <b>Description:</b> The currently available values are <code>"v1"</code>, <code>"v1.5"</code>, and <code>"v1.6"</code>.
+  </td>
+  <td><code>str</code></td>
+  <td>"v1.6"</td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the layout analysis model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the default model of the production line will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the layout analysis model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>Score threshold for the layout model. <br/>
+<b>Description:</b>
+<ul>
+<li><b>float</b>: Any floating-point number between <code>0-1</code>;</li>
+<li><b>dict</b>: <code>{0:0.1}</code> The key is the class ID, and the value is the threshold for that class;</li>
+<li><b>None</b>: If set to <code>None</code>, the parameter value initialized by the pipeline will be used.</li>
+</ul>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Whether to use post-processing NMS for layout analysis. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the parameter value initialized by the production line will be used.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td>
+<b>Meaning:</b>Expansion coefficient for the detection box of the layout area detection model. <br/>
+<b>Description:</b>
+<ul>
+<li><b>float</b>: Any floating-point number greater than <code>0</code></li>
+<li><b>Tuple[float,float]</b>: The respective expansion coefficients in the horizontal and vertical directions;</li>
+<li><b>dict</b>: where the key of the dict is of <b>int</b> type, representing <code>cls_id</code>, and the value is of</code>tuple <code>type, such as</code>{0: (1.1, 2.0)}, indicating that the center of the detection box for class 0 output by the model remains unchanged, with the width expanded by 1.1 times and the height expanded by 2.0 times;</li>
+<li><b>None</b>: If set to <code>None</code>, the parameter value initialized by the pipeline will be used.</li>
+</ul>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code><ul>
+<td><b>Meaning:</b>Merging mode for the detection boxes output by the model in layout analysis. <br/>
+<b>Description:</b>
+<ul>
+<li><b>large</b> when set to large, it means that among the detection boxes output by the model, for overlapping and contained boxes, only the outermost largest box is retained, and the overlapping inner boxes are deleted;</li>
+<li><b>small</b>, when set to small, it means that among the detection boxes output by the model, for overlapping and contained boxes, only the innermost contained small box is retained, and the overlapping outer boxes are deleted;</li>
+<li><b>union</b>,no filtering is performed on the boxes, and both inner and outer boxes are retained;</li></ul>
+If set to <code>None</code>, the initialized parameter value will be used.
+</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>vl_rec_model_name</code></td>
+<td><b>Meaning:</b>Name of the multimodal recognition model.<br/>
+<b>Description:</b> If set to <code>None</code>, the default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>vl_rec_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the multimodal recognition model.<br/>
+<b>Description:</b> If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>vl_rec_backend</code></td>
+<td><b>Meaning:</b>Inference backend used by the multimodal recognition model.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>vl_rec_server_url</code></td>
+<td><b>Meaning:</b>If the multimodal recognition model uses an inference service, this parameter is used to specify the server URL.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>vl_rec_max_concurrency</code></td>
+<td><b>Meaning:</b>If the multimodal recognition model uses an inference service, this parameter is used to specify the maximum number of concurrent requests.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>vl_rec_api_model_name</code></td>
+<td><b>Meaning:</b>If the multimodal recognition model uses an inference service, this parameter is used to specify the model name of the service.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>vl_rec_api_key</code></td>
+<td><b>Meaning:</b>If the multimodal recognition model uses an inference service, this parameter is used to specify the API key of the service.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>Name of the document orientation classification model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the initialized default value will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the document orientation classification model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>Meaning:</b>Name of the text image rectification model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the initialized default value will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text image rectification model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to load and use the document orientation classification module. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the initialized default value will be used, which is initialized to<code>False</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to load and use the text image rectification module. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the initialized default value will be used, which is initialized to <code>False</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>Meaning:</b>Whether to load and use the layout analysis module. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the initialized default value will be used, which is initialized to <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>Meaning:</b>Whether to use the chart parsing function. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the initialized default value will be used, which is initialized to <code>False</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>Meaning:</b>Whether to use the seal recognition function.<br/>
+<b>Description:</b> If set to <code>None</code>, the initialized default value will be used, which is initialized to <code>False</code>.</td>
+<td><code>bool|None</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_ocr_for_image_block</code></td>
+<td><b>Meaning:</b>Whether to perform OCR on text within image blocks.<br/>
+<b>Description:</b> If set to <code>None</code>, the initialized default value will be used, which is initialized to <code>False</code>.</td>
+<td><code>bool|None</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>format_block_content</code></td>
+<td><b>Meaning:</b>Controls whether to format the <code>block_content</code> content within as Markdown. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the initialized default value will be used, which defaults to initialization as<code>False</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>merge_layout_blocks</code></td>
+<td><b>Meaning:</b>Control whether to merge the layout detection boxes for cross-column or staggered top and bottom columns.<br/>
+<b>Description:</b> If set to <code>None</code>, the initialized default value will be used, which defaults to initialization as<code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>markdown_ignore_labels</code></td>
+<td><b>Meaning:</b>Layout labels that need to be ignored in Markdown.<br/>
+<b>Description:</b> If set to <code>None</code>, the initialized default value will be used, which defaults to initialization as <code>['number','footnote','header','header_image','footer','footer_image','aside_text']</code>.</td>
+<td><code>list|None</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_queues</code></td>
+<td><b>Meaning:</b>Used to control whether to enable internal queues.<br/>
+<b>Description:</b> When set to <code>True</code>, data loading (such as rendering PDF pages as images), layout analysis model processing, and VLM inference will be executed asynchronously in separate threads, with data passed through queues, thereby improving efficiency. This approach is particularly efficient for PDF documents with many pages or directories containing a large number of images or PDF files. If set to <code>None</code>, the initialized default value will be used, which defaults to initialization as <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>The device used for inference.<br/>
+<b>Description:</b> 
+Supports specifying specific card numbers:<ul>
+<li><b>CPU</b>: For example,<code>cpu</code> indicates using the CPU for inference;</li>
+<li><b>GPU</b>: For example,<code>gpu:0</code> indicates using the first GPU for inference;</li>
+<li><b>NPU</b>: For example,<code>npu:0</code> indicates using the first NPU for inference;</li>
+<li><b>XPU</b>: For example,<code>xpu:0</code> indicates using the first XPU for inference;</li>
+<li><b>MLU</b>: For example,<code>mlu:0</code> indicates using the first MLU for inference;</li>
+<li><b>DCU</b>: For example,<code>dcu:0</code> indicates using the first DCU for inference;</li>
+<li><b>MetaX GPU</b>: For example,<code>metax_gpu:0</code> indicates using the first MetaX GPU for inference;</li>
+<li><b>Iluvatar GPU</b>: For example,<code>iluvatar_gpu:0</code> indicates using the first Iluvatar GPU for inference;</li>
+</ul>If not set, the initialized default value will be used. During initialization, the local GPU device 0 will be used preferentially. If it is not available, the CPU device will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to enable the TensorRT subgraph engine of Paddle Inference.<br/>
+<b>Description:</b>
+If the model does not support TensorRT acceleration, acceleration will not be used even if this flag is set.<br/>
+For CUDA 11.8 versions of PaddlePaddle, the compatible TensorRT version is 8.x (x>=6). TensorRT 8.6.1.6 is recommended.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b> Computation precision, such as <code>"fp32"</code> or <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b> Whether to enable MKL-DNN accelerated inference.<br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support MKL-DNN acceleration, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b> MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads used for inference on CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b> Path to the PaddleX pipeline configuration file.</td>
+<td><code>str</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details><summary>(2) Call the <code>predict()</code>method of the PaddleOCR-VL pipeline object for inference prediction. This method will return a list of results. Additionally, the pipeline also provides the <code>predict_iter()</code>Method. The two are completely consistent in terms of parameter acceptance and result return. The difference lies in that <code>predict_iter()</code>returns a <code>generator</code>, which can process and obtain prediction results step by step. It is suitable for scenarios involving large datasets or where memory conservation is desired. You can choose either of these two methods based on actual needs. Below are the parameters of the <code>predict()</code>method and their descriptions:</summary>
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Parameter Description</th>
+<th>Parameter Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted, supporting multiple input types. Required.<br/>
+<b>Description:</b> 
+<ul>
+<li><b>Python Var</b>: such as <code>numpy.ndarray</code> representing image data</li>
+<li><b>str</b>: such as the local path of an image file or PDF file: <code>/root/data/img.jpg</code>;<b>such as a URL link</b>, such as the network URL of an image file or PDF file:<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/demo_paper.png">Example</a>;<b>such as a local directory</b>, which should contain the images to be predicted, such as the local path: <code>/root/data/</code>(Currently, prediction for directories containing PDF files is not supported. PDF files need to be specified with a specific file path)</li>
+<li><b>list</b>: List elements should be of the aforementioned data types, such as <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"].</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to use the document orientation classification module during inference. <br/>
+<b>Description:</b> 
+Setting it to <code>None</code> means using the instantiation parameter; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to use the text image rectification module during inference. <br/>
+<b>Description:</b> 
+Setting it to <code>None</code> means using the instantiation parameter; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>Meaning:</b>Whether to use the layout region detection and sorting module during inference.<br/>
+<b>Description:</b> 
+Setting it to <code>None</code> means using the instantiation parameter; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>Meaning:</b>Whether to use the chart parsing function. Setting it to <code>None</code> means using the instantiation parameter; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>Meaning:</b>Whether to use the seal recognition function. Setting it to <code>None</code> means using the instantiation parameter; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_ocr_for_image_block</code></td>
+<td><b>Meaning:</b>Whether to perform OCR on text within image blocks. Setting it to <code>None</code> means using the instantiation parameter; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>The parameter meaning is basically the same as the instantiation parameter. <br/>
+<b>Description:</b> 
+Setting it to <code>None</code> means using the instantiation parameter; otherwise, this parameter takes precedence.</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>The parameter meaning is basically the same as the instantiation parameter. <br/>
+<b>Description:</b> 
+Setting it to <code>None</code> means using the instantiation parameter; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>The parameter meaning is basically the same as the instantiation parameter. <br/>
+<b>Description:</b> 
+Setting it to <code>None</code> means using the instantiation parameter; otherwise, this parameter takes precedence.</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>The parameter meaning is basically the same as the instantiation parameter. <br/>
+<b>Description:</b> 
+Setting it to <code>None</code> means using the instantiation parameter; otherwise, this parameter takes precedence.</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+  <td><code>layout_shape_mode</code></td>
+  <td>
+    <b>Meaning:</b>Specifies the geometric representation mode for layout analysis results. It defines how the boundaries of detected regions (e.g., text blocks, images, tables) are calculated and displayed.<br/>
+    <b>Description:</b> Value descriptions:
+    <ul>
+      <li>
+        <b>rect (rectangle)</b>:
+        Outputs an axis-aligned bounding box (including x1, y1, x2, y2).
+        Suitable for standard horizontally aligned layouts.
+      </li>
+      <li>
+        <b>quad (quadrilateral)</b>:
+        Outputs an arbitrary quadrilateral composed of four vertices.
+        Suitable for regions with skew or perspective distortion.
+      </li>
+      <li>
+        <b>poly (polygon)</b>:
+        Outputs a closed contour composed of multiple coordinate points.
+        Suitable for irregularly shaped or curved layout elements,
+        offering the highest precision.
+      </li>
+      <li>
+        <b>auto (automatic)</b>:
+        The system automatically selects the most appropriate shape
+        representation based on the complexity and confidence of the
+        detected targets.
+      </li>
+    </ul>
+  </td>
+  <td><code>str</code></td>
+  <td>"auto"</td>
+</tr>
+<tr>
+<td><code>use_queues</code></td>
+<td><b>Meaning:</b>The parameter meaning is basically the same as the instantiation parameter.<br/>
+<b>Description:</b> Setting it to <code>None</code> means using the instantiation parameter; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>prompt_label</code></td>
+<td><b>Meaning:</b>The prompt type setting for the VL model, which takes effect only when <code>use_layout_detection=False</code>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+<tr>
+<td><code>format_block_content</code></td>
+<td><b>Meaning:</b>The parameter meaning is basically the same as the instantiation parameter. <br/>
+<b>Description:</b> 
+Setting it to <code>None</code> means using the instantiation parameter; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>repetition_penalty</code></td>
+<td><b>Meaning:</b>The repetition penalty parameter used for VL model sampling.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>temperature</code></td>
+<td><b>Meaning:</b>Temperature parameter used for VL model sampling.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>top_p</code></td>
+<td><b>Meaning:</b>Top-p parameter used for VL model sampling.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>min_pixels</code></td>
+<td><b>Meaning:</b>The minimum number of pixels allowed when the VL model preprocesses images.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>max_pixels</code></td>
+<td><b>Meaning:</b>The maximum number of pixels allowed when the VL model preprocesses images.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>max_new_tokens</code></td>
+<td><b>Meaning:</b>The maximum number of tokens generated by the VL model.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>merge_layout_blocks</code></td>
+<td><b>Meaning:</b>Control whether to merge the layout detection boxes for cross-column or staggered top and bottom columns.</td>
+<td><code>bool|None</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>markdown_ignore_labels</code></td>
+<td><b>Meaning:</b>Layout labels that need to be ignored in Markdown.</td>
+<td><code>list|None</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>vlm_extra_args</code></td>
+<td><b>Meaning:</b>Additional configuration parameters for the VLM. The currently supported custom parameters are as follows:
+<ul>
+  <li><code>ocr_min_pixels</code>: Minimum resolution for OCR</li>
+  <li><code>ocr_max_pixels</code>: Maximum resolution for OCR</li>
+  <li><code>table_min_pixels</code>: Minimum resolution for tables</li>
+  <li><code>table_max_pixels</code>: Maximum resolution for tables</li>
+  <li><code>chart_min_pixels</code>: Minimum resolution for charts</li>
+  <li><code>chart_max_pixels</code>: Maximum resolution for charts</li>
+  <li><code>formula_min_pixels</code>: Minimum resolution for formulas</li>
+  <li><code>formula_max_pixels</code>: Maximum resolution for formulas</li>
+  <li><code>seal_min_pixels</code>: Minimum resolution for seals</li>
+  <li><code>seal_max_pixels</code>: Maximum resolution for seals</li>
+</ul></td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</table>
+</details>
+
+<details><summary>(3) Invoke the <code>restructure_pages()</code> method of the PaddleOCR-VL object to reconstruct pages from the multi-page results list of inference predictions. This method will return a reconstructed multi-page result or a merged single-page result. Below are the parameters of the <code>restructure_pages()</code> method and their descriptions:</summary>
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>res_list</code></td>
+<td><b>Meaning:</b> The list of results predicted from a multi-page PDF inference.</td>
+<td><code>list|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>merge_tables</code></td>
+<td><b>Meaning:</b> Controls whether to merge tables across pages.</td>
+<td><code>Bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>relevel_titles</code></td>
+<td><b>Meaning:</b> Controls whether to parse multi-level headings.</td>
+<td><code>Bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>concatenate_pages</code></td>
+<td><b>Meaning:</b> Controls whether to concatenate multi-page results into one page.</td>
+<td><code>Bool</code></td>
+<td><code>False</code></td>
+</tr>
+</tbody>
+</table>
+</details>
+
+
+<details><summary>(4) Process the prediction results: The prediction result for each sample is a corresponding Result object, supporting operations such as printing, saving as an image, and saving as a <code>json</code> file:</summary>
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Method Description</th>
+<th>Parameter</th>
+<th>Parameter Type</th>
+<th>Parameter Description</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"> <code>print()</code></td>
+<td rowspan="3">Print results to the terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output content using <code>JSON</code> indentation.</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code>  data, making it more readable. Only valid when <code>format_json</code> is <code>True</code>.</td>
+<td><code>4</code></td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether non- <code>ASCII</code> characters are escaped as <code>Unicode</code>. When set to <code>True</code>, all non- <code>ASCII</code> characters will be escaped; <code>False</code> retains the original characters. Only valid when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"> <code>save_to_json()</code></td>
+<td rowspan="3">Save the results as a json format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The file path for saving. When it is a directory, the saved file name will be consistent with the input file type naming.</td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code>data, making it more readable. Only valid when <code>format_json</code>is <code>True</code>.</td>
+<td><code>4</code></td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether non- <code>ASCII</code> characters are escaped as <code>Unicode</code>. When set to <code>True</code>, all non- <code>ASCII</code> characters will be escaped; <code>False</code> retains the original characters. Only valid when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save the visualized images of each intermediate module in png format</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The file path for saving, supporting directory or file paths.</td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td rowspan="3"> <code>save_to_markdown()</code></td>
+<td rowspan="3">Save each page in an image or PDF file as a markdown format file separately</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The file path for saving. When it is a directory, the saved file name will be consistent with the input file type naming</td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>pretty</code></td>
+<td><code>bool</code></td>
+<td>Whether to beautify the <code>markdown</code> output results, centering charts, etc., to make the <code>markdown</code> rendering more aesthetically pleasing.</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>show_formula_number</code></td>
+<td><code>bool</code></td>
+<td>Control whether to retain formula numbers in <code>markdown</code>. When set to <code>True</code>, all formula numbers are retained; <code>False</code> retains only the formulas</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_html()</code></td>
+<td>Save the tables in the file as html format files</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The file path for saving, supporting directory or file paths.</td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>save_to_xlsx()</code></td>
+<td>Save the tables in the file as xlsx format files</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The file path for saving, supporting directory or file paths.</td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>save_to_word()</code></td>
+<td>Save the layout parsing results as a Word (.docx) format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The file path for saving, supporting directory or file paths.</td>
+<td><code>None</code></td>
+</tr>
+</tr>
+</table>
+
+
+- Calling the `print()` method will print the results to the terminal. The content printed to the terminal is explained as follows:
+    - `input_path`: `(str)` The input path of the image or PDF to be predicted.
+
+    - `page_index`: `(Union[int, None])` If the input is a PDF file, it indicates the current page number of the PDF; otherwise, it is `None`.
+
+    - `page_count`: `(Union[int, None])` If the input is a PDF file, it indicates the total number of pages in the PDF; otherwise, it is `None`.
+
+    - `width`: `(int)` The width of the original input image.
+
+    - `height`: `(int)` The height of the original input image.
+
+    - `model_settings`: `(Dict[str, bool])` Model parameters required for configuring PaddleOCR-VL.
+        - `use_doc_preprocessor`: `(bool)` Controls whether to enable the document preprocessing sub-pipeline.
+        - `use_layout_detection`: `(bool)` Controls whether to enable the layout analysis module.
+        - `use_chart_recognition`: `(bool)` Controls whether to enable the chart recognition function.
+        - `format_block_content`: `(bool)` Controls whether to save the formatted markdown content in `JSON`.
+        - `markdown_ignore_labels`: `(List[str])` Labels of layout regions that need to be ignored in Markdown
+
+    - `doc_preprocessor_res`: `(Dict[str, Union[List[float], str]])` A dictionary of document preprocessing results, which exists only when `use_doc_preprocessor=True`.
+        - `input_path`: `(str)` The image path accepted by the document preprocessing sub-pipeline. When the input is a `numpy.ndarray`, it is saved as `None`; here, it is `None`.
+        - `page_index`: `None`. Since the input here is a `numpy.ndarray`, the value is `None`.
+        - `model_settings`: `(Dict[str, bool])` Model configuration parameters for the document preprocessing sub-pipeline.
+          - `use_doc_orientation_classify`: `(bool)` Controls whether to enable the document image orientation classification sub-module.
+          - `use_doc_unwarping`: `(bool)` Controls whether to enable the text image distortion correction sub-module.
+        - `angle`: `(int)` The prediction result of the document image orientation classification sub-module. When enabled, it returns the actual angle value.
+
+    - `parsing_res_list`: `(List[Dict])` A list of parsing results, where each element is a dictionary. The list order is the reading order after parsing.
+        - `block_bbox`: `(np.ndarray)` The bounding box of the layout area.
+        - `block_label`: `(str)` The label of the layout area, such as `text`, `table`, etc.
+        - `block_content`: `(str)` The content within the layout area.
+        - `block_id`: `(int)` The index of the layout area, used to display the layout sorting results.
+        - `block_order` `(int)` The order of the layout area, used to display the layout reading order. For non-sorted parts, the default value is `None`.
+- Calling the `save_to_json()` method will save the above content to the specified `save_path`. If a directory is specified, the saved path will be `save_path/{your_img_basename}_res.json`. If a file is specified, it will be saved directly to that file. Since json files do not support saving numpy arrays, the `numpy.array` types within will be converted to list form.
+    - `input_path`: `(str)` The input path of the image or PDF to be predicted.
+
+    - `page_index`: `(Union[int, None])` If the input is a PDF file, it indicates the current page number of the PDF; otherwise, it is `None`.
+
+    - `model_settings`: `(Dict[str, bool])` Model parameters required for configuring PaddleOCR-VL.
+
+        - `use_doc_preprocessor`: `(bool)` Controls whether to enable the document preprocessing sub-pipeline.
+        - `use_layout_detection`: `(bool)` Controls whether to enable the layout analysis module.
+        - `use_chart_recognition`: `(bool)` Controls whether to enable the chart recognition function.
+        - `format_block_content`: `(bool)` Controls whether to save the formatted markdown content in `JSON`.
+
+    - `doc_preprocessor_res`: `(Dict[str, Union[List[float], str]])` A dictionary of document preprocessing results, which exists only when `use_doc_preprocessor=True`.
+        - `input_path`: `(str)` The image path accepted by the document preprocessing sub-pipeline. When the input is a `numpy.ndarray`, it is saved as `None`; here, it is `None`.
+        - `page_index`: `None`. Since the input here is a `numpy.ndarray`, the value is `None`.
+        - `model_settings`: `(Dict[str, bool])` Model configuration parameters for the document preprocessing sub-pipeline.
+          - `use_doc_orientation_classify`: `(bool)` Controls whether to enable the document image orientation classification sub-module.
+          - `use_doc_unwarping`: `(bool)` Controls whether to enable the text image distortion correction sub-module.
+        - `angle`: `(int)` The prediction result of the document image orientation classification sub-module. When enabled, it returns the actual angle value.
+
+    - `parsing_res_list`: `(List[Dict])` A list of parsing results, where each element is a dictionary. The list order represents the reading order after parsing.
+        - `block_bbox`: `(np.ndarray)` The bounding box of the layout region.
+        - `block_label`: `(str)` The label of the layout region, such as `text`, `table`, etc.
+        - `block_content`: `(str)` The content within the layout region.
+        - `block_id`: `(int)` The index of the layout region, used to display the layout sorting results.
+        - `block_order` `(int)` The order of the layout region, used to display the layout reading order. For non-sorted parts, the default value is `None`.
+
+
+- Calling the `save_to_img()` method will save the visualization results to the specified `save_path`. If a directory is specified, visualized images for layout region detection, global OCR, layout reading order, etc., will be saved. If a file is specified, it will be saved directly to that file. (Pipelines typically contain many result images, so it is not recommended to directly specify a specific file path, as multiple images will be overwritten, retaining only the last one.)
+- Calling the `save_to_markdown()` method will save the converted Markdown file to the specified `save_path`. The saved file path will be `save_path/{your_img_basename}.md`. If the input is a PDF file, it is recommended to directly specify a directory; otherwise, multiple markdown files will be overwritten.
+
+<li>Additionally, it also supports obtaining visualized images and prediction results with results through attributes, as follows:
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Attribute Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>json</code></td>
+<td>Obtain the prediction <code>json</code>result in the format</td>
+</tr>
+<tr>
+<td><code>img</code></td>
+<td>Obtain visualized images in <code>dict</code> format</td>
+</tr>
+<tr>
+<td><code>markdown</code></td>
+<td>Obtain markdown results in <code>dict</code> format</td>
+</tr>
+</tbody>
+</table>
+
+<ul>
+  <li>The prediction result obtained through the <code>json</code> attribute is data of dict type, with relevant content consistent with that saved by calling the <code>save_to_json()</code> method.</li>
+  <li>The prediction result returned by the <code>img</code> attribute is data of dict type. The keys are <code>layout_det_res</code>, <code>overall_ocr_res</code>, <code>text_paragraphs_ocr_res</code>, <code>formula_res_region1</code>, <code>table_cell_img</code>, and <code>seal_res_region1</code>, with corresponding values being <code>Image.Image</code> objects: used to display visualized images of layout region detection, OCR, OCR text paragraphs, formulas, tables, and seal results, respectively. If optional modules are not used, the dict only contains <code>layout_det_res</code>.</li>
+  <li>The prediction result returned by the <code>markdown</code> attribute is data of dict type. The keys are <code>markdown_texts</code>, <code>markdown_images</code>, and <code>page_continuation_flags</code>, with corresponding values being markdown text, images displayed in Markdown (<code>Image.Image</code> objects), and a bool tuple used to identify whether the first element on the current page is the start of a paragraph and whether the last element is the end of a paragraph, respectively.</li>
+</ul>
+</li>
+</details>
+
+<a id="3-vlm"></a>
+## 3. Using VLM Inference Services
+
+This section explains how to connect a VLM inference service into the PaddleOCR-VL workflow. This usually improves inference performance and helps achieve better resource utilization and service stability in production environments. You can either deploy your own VLM inference service based on backends such as vLLM, SGLang, FastDeploy, MLX-VLM, and llama.cpp, or directly use compatible managed services. For the applicable path on other hardware, refer to the corresponding hardware-specific tutorial. This section corresponds to combinations of "layout analysis inference method + VLM inference service". Its core idea is that **the client continues to handle the other stages in the full workflow, such as layout analysis, while only the VLM stage is delegated to a dedicated service**.
+
+### 3.1 Launching the VLM Inference Service
+
+> IMPORTANT:
+> The services launched according to this section are responsible only for the VLM inference stage in the PaddleOCR-VL workflow and do not provide a complete end-to-end document parsing API. It is strongly discouraged to directly call such services through plain HTTP requests or OpenAI clients to process document images. If you need to deploy a service with the full PaddleOCR-VL capability, please refer to the service deployment section later in this document.
+
+> TIP:
+> The default recommended vLLM / SGLang / FastDeploy routes in this section are only suitable for hardware that meets the corresponding CC/CUDA requirements. For NVIDIA GPUs with CC 7.x, such as T4/V100, do not follow the default recommended route directly. Review the matrix above and choose an available backend based on the support status.
+
+There are three methods to launch the VLM inference service; choose either one:
+
+- Method 1: Launch the service using the official Docker image. Currently supported:
+    - vLLM
+    - FastDeploy
+
+- Method 2: Launch the service by manually installing dependencies via the PaddleOCR CLI. Currently supported:
+    - vLLM
+    - SGLang
+    - FastDeploy
+
+- Method 3: Launch service directly using inference acceleration frameworks (the pre-configured performance tuning parameters provided by PaddleOCR will not be applied). Currently supported:
+    - vLLM
+    - FastDeploy
+    - MLX-VLM
+    - llama.cpp
+
+**We strongly recommend using the Docker image to minimize potential environment-related issues.**
+
+In addition, cloud platforms such as [SiliconFlow](https://siliconflow.cn/) and [Novita AI](https://novita.ai/models-console/model-detail/paddlepaddle-paddleocr-vl) also provide managed services. If you choose to use such services, you can skip this subsection and directly read [3.2 Client Usage Methods](#32-client-usage-methods).
+
+#### 3.1.1 Method 1: Using Docker Image
+
+PaddleOCR provides Docker images for quickly launching vLLM or FastDeploy inference services. You can use the following commands to start the services (requires Docker version >= 19.03, a machine equipped with a GPU, and NVIDIA drivers supporting CUDA 12.6 or later):
+
+=== "Launch vLLM Service"
+
+    ```shell
+    docker run \
+        -it \
+        --rm \
+        --gpus all \
+        --network host \
+        ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu \
+        paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm
+    ```
+
+    If you wish to start the service in an environment without internet access, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu` (image size approximately 13 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu-offline` (image size approximately 15 GB).
+
+=== "Launch FastDeploy Service"
+
+    ```shell
+    docker run \
+        -it \
+        --rm \
+        --gpus all \
+        --network host \
+        ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-nvidia-gpu \
+        paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy
+    ```
+
+    If you wish to start the service in an environment without internet access, replace `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-nvidia-gpu` (image size approximately 43 GB) in the above command with the offline version image `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-nvidia-gpu-offline` (image size approximately 45 GB).
+
+When starting the vLLM or FastDeploy inference service, we provide a set of default parameter settings. If you have additional requirements for adjusting parameters such as GPU memory usage, you can configure more parameters yourself. Please refer to [3.3.1 Server-side Parameter Adjustment](#331-server-side-parameter-adjustment) to create a configuration file, then mount this file into the container, and specify the configuration file using `backend_config` in the command to start the service. Taking vLLM as an example:
+
+```shell
+docker run \
+    -it \
+    --rm \
+    --gpus all \
+    --network host \
+    -v ./vllm_config.yml:/tmp/vllm_config.yml \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu \
+    paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /tmp/vllm_config.yml
+```
+
+Here, `./vllm_config.yml` refers to a local configuration file in the current working directory on the host machine. If the file is located elsewhere, replace it with an actual absolute path or a relative path with an explicit directory prefix.
+
+> TIP:
+> Images with the `latest-xxx` tag correspond to the latest version.
+> If the corresponding `latest` image already exists locally and you want the newest features or fixes, we recommend running `docker pull` again before using it.
+> If you want to use an image corresponding to a specific PaddleOCR version, you can replace `latest` in the tag with the desired version number: `paddleocr<major>.<minor>`.
+> For example:
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:paddleocr3.3-nvidia-gpu-offline`
+
+
+#### 3.1.2 Method 2: Installation and Usage via PaddleOCR CLI
+
+**The PaddleOCR CLI has already resolved complex version compatibility issues. Instead of spending time studying framework documentation, you can install the necessary environment with a single command.**
+
+Since inference acceleration frameworks may conflict with packages already installed in the current environment, it is recommended to install them in a virtual environment:
+
+```shell
+# If a virtual environment is currently activated, deactivate it first using `deactivate`
+# Create a virtual environment
+python -m venv .venv_vlm
+# Activate the environment
+source .venv_vlm/bin/activate
+```
+
+vLLM and SGLang depend on FlashAttention, and installing FlashAttention may require CUDA compilation tools such as `nvcc`. If these tools are not available in your environment (for example, when using the `paddleocr-vl` image), you can obtain a prebuilt FlashAttention package (version 2.8.2 required) from [this repository](https://github.com/mjun0812/flash-attention-prebuild-wheels), install it first, and then proceed with subsequent commands. For example, in the `paddleocr-vl` image, run `python -m pip install https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.3.14/flash_attn-2.8.2+cu128torch2.8-cp310-cp310-linux_x86_64.whl`. This step is not required for FastDeploy.
+
+Install PaddleOCR and the dependencies of inference acceleration services, using vLLM as an example:
+
+```shell
+# Install PaddleOCR
+python -m pip install "paddleocr[doc-parser]"
+# Install inference acceleration service dependencies
+paddleocr install_genai_server_deps vllm
+```
+
+The usage of `paddleocr install_genai_server_deps` is:
+
+```shell
+paddleocr install_genai_server_deps <inference-acceleration-framework-name>
+```
+
+Currently supported framework names are `vllm`, `sglang`, and `fastdeploy`, corresponding to vLLM, SGLang, and FastDeploy, respectively.
+
+Both vLLM and SGLang installed through `paddleocr install_genai_server_deps` are **CUDA 12.6** versions. Please ensure that your local NVIDIA driver supports this version or a later one.
+
+> WARNING:
+> The transformers library versions required by vLLM, SGLang and Transformers engine are currently incompatible, so Transformers engine and vLLM cannot be installed together with vLLM or SGLang in the same environment. If using Transformers + vLLM or Transformers + SGLang inference, please deploy the layout analysis model and VLM service in different environments.
+
+After installation, you can launch the service using the `paddleocr genai_server` command:
+
+```shell
+paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --backend vllm --port 8118
+```
+
+The parameters supported by this command are as follows:
+
+| Parameter          | Description                                                  |
+|--------------------|--------------------------------------------------------------|
+| `--model_name`     | Model name                                                    |
+| `--model_dir`      | Model directory                                               |
+| `--host`           | Server hostname                                               |
+| `--port`           | Server port number                                             |
+| `--backend`        | Backend name, i.e., the name of the inference acceleration framework used; options are `vllm`, `sglang`, or `fastdeploy` |
+| `--backend_config` | Can specify a YAML file containing backend configurations      |
+
+#### 3.1.3 Launch Service Directly Using Inference Acceleration Frameworks
+
+**If you need to install a custom version of an inference framework and launch the service natively, please refer to the following guidelines. Please note that when launching natively, the pre-configured performance tuning parameters provided by PaddleOCR will not be applied.**
+
+- vLLM: [Refer to this document](https://docs.vllm.ai/projects/recipes/en/latest/PaddlePaddle/PaddleOCR-VL.html)
+- SGLang: [Refer to the SGLang official documentation](https://docs.sglang.ai/)
+- FastDeploy: [Refer to this document](https://paddlepaddle.github.io/FastDeploy/best_practices/PaddleOCR-VL-0.9B/)
+- MLX-VLM: [Refer to this document](./PaddleOCR-VL-Apple-Silicon.en.md)
+- llama.cpp:
+    1. Install llama.cpp by referring to the `Quick start` section in the [llama.cpp github](https://github.com/ggml-org/llama.cpp).
+    2. Download the model files in gguf format: [PaddlePaddle/PaddleOCR-VL-1.6-GGUF](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.6-GGUF).
+    3. Execute the following command to start the inference service. For an introduction to the parameters, please refer to [LLaMA.cpp HTTP Server](https://github.com/ggml-org/llama.cpp/blob/master/tools/server/README.md):
+
+        ```shell
+        ./build/bin/llama-server \
+          -m /path/to/PaddleOCR-VL-1.6-GGUF.gguf \
+          --mmproj /path/to/PaddleOCR-VL-1.6-GGUF-mmproj.gguf  \
+          --port 8111  \
+          --host 0.0.0.0 \
+          --temp 0
+        ```
+
+### 3.2 Client Usage Methods
+
+After launching the VLM inference service, the client can call the service through PaddleOCR. This section applies both to self-hosted VLM inference services launched in 3.1 and to compatible managed services provided by third parties. **Please note that because the client still needs to call the layout analysis model and complete the other stages in the workflow, it is still recommended to run the client on GPU or other acceleration devices to achieve more stable and efficient performance. Please refer to Section 1 for the client-side environment configuration. The configuration described in Section 3.1 applies only to starting the service and is not applicable to the client. If you want the client to invoke the full PaddleOCR-VL capability only through an HTTP interface, please directly refer to Section 4, "Service Deployment".**
+
+#### 3.2.1 CLI Invocation
+
+Specify the backend type (`vllm-server`, `sglang-server`, `fastdeploy-server`, `mlx-vlm-server` or `llama-cpp-server`) using `--vl_rec_backend` and the service address using `--vl_rec_server_url`, for example:
+
+```shell
+paddleocr doc_parser --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --vl_rec_backend vllm-server --vl_rec_server_url http://localhost:8118/v1
+```
+
+In addition, you can specify the model name used by the service via `--vl_rec_api_model_name`, and specify the API key used for authentication via `--vl_rec_api_key`. Examples are as follows:
+
+Using a service started with the default parameters of `vllm serve`:
+
+```shell
+paddleocr doc_parser \
+    --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png \
+    --vl_rec_backend vllm-server \
+    --vl_rec_server_url http://localhost:8000/v1 \
+    --vl_rec_api_model_name 'PaddlePaddle/PaddleOCR-VL-1.6'
+```
+
+SiliconFlow platform (currently only PaddleOCR-VL-0.9B and PaddleOCR-VL-1.5-0.9B are supported):
+
+```shell
+paddleocr doc_parser \
+    --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png \
+    --pipeline_version v1.5 \
+    --vl_rec_backend vllm-server \
+    --vl_rec_server_url https://api.siliconflow.cn/v1 \
+    --vl_rec_api_model_name 'PaddlePaddle/PaddleOCR-VL-1.5' \
+    --vl_rec_api_key xxxxxx
+```
+
+Novita AI platform (currently only PaddleOCR-VL-0.9B is supported, i.e., the v1 model):
+
+```shell
+paddleocr doc_parser \
+    --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png \
+    --pipeline_version v1 \
+    --vl_rec_backend vllm-server \
+    --vl_rec_server_url https://api.novita.ai/openai \
+    --vl_rec_api_model_name 'paddlepaddle/paddleocr-vl' \
+    --vl_rec_api_key xxxxxx
+```
+
+#### 3.2.2 Python API Invocation
+
+When creating a `PaddleOCRVL` object, specify the backend type (`vllm-server`, `sglang-server`, `fastdeploy-server`, `mlx-vlm-server` or `llama-cpp-server`) using `vl_rec_backend` and the service address using `vl_rec_server_url`, for example:
+
+```python
+pipeline = PaddleOCRVL(vl_rec_backend="vllm-server", vl_rec_server_url="http://localhost:8118/v1")
+```
+
+In addition, you can specify the model name used by the service via `vl_rec_api_model_name`, and specify the API key used for authentication via `vl_rec_api_key`.
+
+Using a service started with the default parameters of `vllm serve`:
+
+```python
+pipeline = PaddleOCRVL(
+    vl_rec_backend="vllm-server", 
+    vl_rec_server_url="http://localhost:8000/v1",
+    vl_rec_api_model_name="PaddlePaddle/PaddleOCR-VL-1.6",
+)
+```
+
+SiliconFlow platform (currently only PaddleOCR-VL-0.9B and PaddleOCR-VL-1.5-0.9B are supported):
+
+```python
+pipeline = PaddleOCRVL(
+    pipeline_version="v1.5",
+    vl_rec_backend="vllm-server", 
+    vl_rec_server_url="https://api.siliconflow.cn/v1",
+    vl_rec_api_model_name="PaddlePaddle/PaddleOCR-VL-1.5",
+    vl_rec_api_key="xxxxxx",
+)
+```
+
+Novita AI platform (currently only PaddleOCR-VL-0.9B is supported, i.e., the v1 model):
+
+```python
+pipeline = PaddleOCRVL(
+    pipeline_version="v1",
+    vl_rec_backend="vllm-server", 
+    vl_rec_server_url="https://api.novita.ai/openai",
+    vl_rec_api_model_name="paddlepaddle/paddleocr-vl",
+    vl_rec_api_key="xxxxxx",
+)
+```
+
+### 3.3 Performance Tuning
+
+The default configurations cannot guarantee optimal performance in all environments. If you encounter performance issues in actual use, you can try the following optimization methods.
+
+#### 3.3.1 Server-Side Parameter Adjustment
+
+Different inference acceleration frameworks support different parameters. Refer to their official documentation for available parameters and adjustment timing:
+
+- [vLLM Official Parameter Tuning Guide](https://docs.vllm.ai/en/latest/configuration/optimization.html)
+- [SGLang Hyperparameter Tuning Documentation](https://docs.sglang.ai/advanced_features/hyperparameter_tuning.html)
+- [FastDeploy Best Practices](https://paddlepaddle.github.io/FastDeploy/best_practices/PaddleOCR-VL-0.9B/)
+
+The PaddleOCR VLM inference service supports parameter tuning through configuration files. The following example shows how to adjust the `gpu-memory-utilization` and `max-num-seqs` parameters for the vLLM server:
+
+1. Create a YAML file `vllm_config.yaml` with the following content:
+
+   ```yaml
+   gpu-memory-utilization: 0.3
+   max-num-seqs: 128
+   ```
+
+2. Specify the configuration file path when starting the service, for example, using the `paddleocr genai_server` command:
+
+   ```shell
+   paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --backend vllm --backend_config vllm_config.yaml
+   ```
+
+If using a shell that supports process substitution (like Bash), you can also pass configuration items directly without creating a configuration file:
+
+```bash
+paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --backend vllm --backend_config <(echo -e 'gpu-memory-utilization: 0.3\nmax-num-seqs: 128')
+```
+
+#### 3.3.2 Client-Side Parameter Adjustment
+
+PaddleOCR groups sub-images from single or multiple input images and sends concurrent requests to the server, so the number of concurrent requests significantly impacts performance.
+
+- For CLI and Python API, adjust the maximum number of concurrent requests using the `vl_rec_max_concurrency` parameter;
+- For service deployment, modify the `VLRecognition.genai_config.max_concurrency` field in the configuration file.
+
+When there is a 1:1 client-to-VLM inference service ratio and sufficient server resources, increasing concurrency can improve performance. If the server needs to support multiple clients or has limited computing resources, reduce concurrency to avoid resource overload and service abnormalities.
+
+#### 3.3.3 Common Hardware Performance Tuning Recommendations
+
+The following configurations are for scenarios with a 1:1 client-to-VLM inference service ratio.
+
+**NVIDIA RTX 3060**
+
+- **Server-Side**
+    - vLLM: `gpu-memory-utilization: 0.7`
+    - FastDeploy: 
+        - `gpu-memory-utilization: 0.7`
+        - `max-concurrency: 2048`
+
+## 4. Service Deployment
+
+This step mainly introduces how to deploy PaddleOCR-VL as a service and invoke it. If concurrent request processing is not required, choose either of the following two methods:
+
+- Method 1: Deploy using Docker Compose (recommended).
+
+- Method 2: Manual Deployment.
+
+Both methods can handle only one request at a time. If you need concurrent request processing, please refer to the [High-Performance Service Deployment solution](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/hps/README_en.md).
+
+> IMPORTANT:
+> The PaddleOCR-VL service introduced in this section differs from the VLM inference service in the previous section: the latter is responsible for only one part of the complete process (i.e., VLM inference) and is called as an underlying service by the former.
+
+### 4.1 Method 1: Deploy Using Docker Compose (Recommended)
+
+> TIP:
+> The default Docker Compose solution in this section is only suitable for NVIDIA GPUs that meet the corresponding CC/CUDA requirements, and uses vLLM or FastDeploy as the underlying VLM backend. For CC 7.x devices such as T4/V100, do not follow this default route directly. Instead, read Section 4.2 and combine it with the matrix above to choose an available inference method.
+
+You can obtain the Compose file and the environment variables configuration file from [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu/compose.yaml) and [here](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu/.env), respectively, and download them to your local machine. Then, in the directory where the files were just downloaded, execute the following command to start the server, which will listen on port **8080** by default:
+
+```shell
+# Must be executed in the directory containing the compose.yaml and .env files
+docker compose up
+```
+
+> TIP:
+> The image tags used by `compose.yaml` are usually controlled by `API_IMAGE_TAG_SUFFIX` and `VLM_IMAGE_TAG_SUFFIX` in `.env`, and default to tags such as `latest-nvidia-gpu-offline`. To make sure you pull the newest `latest` images, run `docker compose pull` in the current directory before `docker compose up`.
+> To use an image corresponding to a specific PaddleOCR version, replace `latest` in these variables with `paddleocr<major>.<minor>`, for example `paddleocr3.3-nvidia-gpu-offline`.
+
+After startup, you will see output similar to the following:
+
+```text
+paddleocr-vl-api             | INFO:     Started server process [1]
+paddleocr-vl-api             | INFO:     Waiting for application startup.
+paddleocr-vl-api             | INFO:     Application startup complete.
+paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+```
+
+This solution accelerates VLM inference based on frameworks like vLLM, making it more suitable for production environment deployment. However, it requires the machine to be equipped with a GPU and the NVIDIA driver to support CUDA 12.6 or higher.
+
+Additionally, after starting the server using this method, no internet connection is required except for pulling the image. For offline environment deployment, you can first pull the images involved in the Compose file on an online machine, export and transfer them to the offline machine for import, and then start the service in the offline environment.
+
+Docker Compose starts two containers in sequence by reading the configurations in the `.env` and `compose.yaml` files, running the underlying VLM inference service and the PaddleOCR-VL service (Pipeline) respectively.
+
+The meanings of each environment variable contained in the `.env` file are as follows:
+
+- `API_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to start the pipeline service. The default is `latest-nvidia-gpu-offline`, indicating the use of the latest offline GPU image. To use an image corresponding to a specific version of PaddleOCR, replace `latest` with the desired version `paddleocr<major>.<minor>`, for example `paddleocr3.3-nvidia-gpu-offline`.
+- `VLM_BACKEND`: The VLM inference backend, currently supporting `vllm` and `fastdeploy`. The default is `vllm`.
+- `VLM_IMAGE_TAG_SUFFIX`: The tag suffix of the image used to start the VLM inference service. The default is `latest-nvidia-gpu-offline`, indicating the use of the latest offline GPU image. If you want to use a non-offline version of the image, you can remove the `-offline` suffix. To use an image corresponding to a specific version of PaddleOCR, replace `latest` with the desired version `paddleocr<major>.<minor>`, for example `paddleocr3.3-nvidia-gpu-offline`.
+
+You can meet custom requirements by modifying `.env` and `compose.yaml`, for example:
+
+<details>
+<summary>1. Change the port of the PaddleOCR-VL service</summary>
+
+Edit <code>paddleocr-vl-api.ports</code> in the <code>compose.yaml</code> file to change the port. For example, if you need to change the service port to 8111, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. Specify the GPU used by the PaddleOCR-VL service</summary>
+
+Edit <code>device_ids</code> in the <code>compose.yaml</code> file to change the GPU used. For example, if you need to use GPU card 1 for deployment, make the following modifications:
+
+```diff
+  paddleocr-vl-api:
+    ...
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+-             device_ids: ["0"]
++             device_ids: ["1"]
+              capabilities: [gpu]
+    ...
+  paddleocr-vlm-server:
+    ...
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+-             device_ids: ["0"]
++             device_ids: ["1"]
+              capabilities: [gpu]
+    ...
+```
+
+</details>
+
+<details>
+<summary>3. Adjust VLM server-side configuration</summary>
+
+If you want to adjust the VLM server-side configuration, please refer to <a href="#331-server-side-parameter-adjustment">3.3.1 Server-side Parameter Adjustment</a> to generate a configuration file.
+
+After generating the configuration file, add the following <code>paddleocr-vlm-server.volumes</code> and <code>paddleocr-vlm-server.command</code> fields to your <code>compose.yaml</code>. Please replace <code>/path/to/your_config.yaml</code> with your actual configuration file path.
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes:
+      - /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. Change the VLM inference backend</summary>
+
+Modify <code>VLM_BACKEND</code> in the <code>.env</code> file, for example, to change the VLM inference backend to <code>fastdeploy</code>:
+
+```diff
+  API_IMAGE_TAG_SUFFIX=latest-nvidia-gpu-offline
+- VLM_BACKEND=vllm
++ VLM_BACKEND=fastdeploy
+  VLM_IMAGE_TAG_SUFFIX=latest-nvidia-gpu-offline
+```
+
+</details>
+
+<details>
+<summary>5. Adjust pipeline configurations (such as model path, batch size, deployment device, etc.)</summary>
+
+Refer to section <a href="#44-pipeline-configuration-adjustment-instructions">4.4 Pipeline Configuration Adjustment Instructions</a> in this document.
+
+</details>
+
+### 4.2 Method 2: Manual Deployment
+
+> IMPORTANT:
+> Before following this section, complete the PaddleOCR local runtime environment setup in Section 1.
+
+Execute the following command to install the service deployment plugin via the PaddleX CLI:
+
+> The `paddlex` command is installed together with `paddleocr`. Therefore, if you have already installed PaddleOCR in the previous steps, you usually do not need to install PaddleX separately.
+
+```shell
+paddlex --install serving
+```
+
+Then, start the server using the PaddleX CLI:
+
+```shell
+paddlex --serve --pipeline PaddleOCR-VL
+```
+
+To switch to the `transformers` engine for service deployment, use:
+
+```shell
+paddlex --serve --pipeline PaddleOCR-VL --engine transformers
+```
+
+After startup, you will see output similar to the following, with the server listening on port **8080** by default:
+
+```text
+INFO:     Started server process [63108]
+INFO:     Waiting for application startup.
+INFO:     Application startup complete.
+INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+```
+
+The command-line options related to serving are as follows:
+
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>--pipeline</code></td>
+<td>PaddleX pipeline registration name or pipeline configuration file path.</td>
+</tr>
+<tr>
+<td><code>--device</code></td>
+<td>Deployment device for the pipeline. By default, a GPU will be used if available; otherwise, a CPU will be used."</td>
+</tr>
+<tr>
+<td><code>--host</code></td>
+<td>Hostname or IP address to which the server is bound. Defaults to <code>0.0.0.0</code>.</td>
+</tr>
+<tr>
+<td><code>--port</code></td>
+<td>Port number on which the server listens. Defaults to <code>8080</code>.</td>
+</tr>
+<tr>
+<td><code>--use_hpip</code></td>
+<td>If specified, uses high-performance inference. Refer to the High-Performance Inference documentation for more information.</td>
+</tr>
+<tr>
+<td><code>--hpi_config</code></td>
+<td>High-performance inference configuration. Refer to the High-Performance Inference documentation for more information.</td>
+</tr>
+</tbody>
+</table>
+
+If you need to adjust pipeline configurations (such as model path, batch size, deployment device, etc.), you can specify the `--pipeline` parameter as a custom configuration file path. For the correspondence between PaddleOCR pipelines and PaddleX pipeline registration names, as well as how to obtain and modify PaddleX pipeline configuration files, please refer to [PaddleOCR and PaddleX](../paddleocr_and_paddlex.en.md). Furthermore, section 4.1.3 will introduce how to adjust the pipeline configuration based on common requirements.
+
+### 4.3 Client-Side Invocation
+
+Below are the API reference and examples of multi-language service invocation:
+
+<details><summary>API Reference</summary>
+<p>Main operations provided by the service:</p>
+<ul>
+<li>The HTTP request method is POST.</li>
+<li>Both the request body and response body are JSON data (JSON objects).</li>
+<li>When the request is processed successfully, the response status code is<code>200</code>, and the properties of the response body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>The UUID of the request.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Fixed as <code>0</code>.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error description. Fixed as <code>"Success"</code>.</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>Operation result.</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is not processed successfully, the properties of the response body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>The UUID of the request.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Same as the response status code.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error description.</td>
+</tr>
+</tbody>
+</table>
+<p>The main operations provided by the service are as follows:</p>
+<ul>
+<li><b><code>infer</code></b></li>
+</ul>
+<p>Perform layout parsing.</p>
+<p><code>POST /layout-parsing</code></p>
+<ul>
+<li>The properties of the request body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+<th>Required</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>The URL of image files (including TIFF; multi-page TIFF is processed page by page) or PDF file accessible to the server, or the Base64-encoded result of the content of the aforementioned file types.
+</td>
+<td>Yes</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code>|<code>null</code></td>
+<td>File type. <code>0</code> represents a PDF file, <code>1</code> represents an image file (including TIFF). If this property is not present in the request body, the file type will be inferred from the URL.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_doc_orientation_classify</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>use_doc_unwarping</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useLayoutDetection</code></td>
+<td><code>boolean</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>use_layout_detection</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useChartRecognition</code></td>
+<td><code>boolean</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>use_chart_recognition</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useSealRecognition</code></td>
+<td><code>boolean</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>use_seal_recognition</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useOcrForImageBlock</code></td>
+<td><code>boolean</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>use_ocr_for_image_block</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutThreshold</code></td>
+<td><code>number</code>|<code>object</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>layout_threshold</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutNms</code></td>
+<td><code>boolean</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>layout_nms</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutUnclipRatio</code></td>
+<td><code>number</code>|<code>array</code>|<code>object</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>layout_unclip_ratio</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutMergeBboxesMode</code></td>
+<td><code>string</code>|<code>object</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>layout_merge_bboxes_mode</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutShapeMode</code></td>
+<td><code>string</code></td>
+<td>Please refer to the description of the <code>layout_shape_mode</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>promptLabel</code></td>
+<td><code>string</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>prompt_label</code> parameter in the  <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>formatBlockContent</code></td>
+<td><code>boolean</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>format_block_content</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>repetitionPenalty</code></td>
+<td><code>number</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>repetition_penalty</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>temperature</code></td>
+<td><code>number</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>temperature</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>topP</code></td>
+<td><code>number</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>top_p</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>minPixels</code></td>
+<td><code>number</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>min_pixels</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>maxPixels</code></td>
+<td><code>number</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>max_pixels</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>maxNewTokens</code></td>
+<td><code>number</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>max_new_tokens</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>mergeLayoutBlocks</code></td>
+<td><code>boolean</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>merge_layout_blocks</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>markdownIgnoreLabels</code></td>
+<td><code>array</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>markdown_ignore_labels</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>vlmExtraArgs</code></td>
+<td><code>object</code>|<code>null</code></td>
+<td>Please refer to the description of the <code>vlm_extra_args</code> parameter in the <code>predict</code> method of the PaddleOCR-VL object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>prettifyMarkdown</code></td>
+<td><code>boolean</code></td>
+<td>Whether to output beautified Markdown text. The default is <code>true</code>.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>showFormulaNumber</code></td>
+<td><code>boolean</code></td>
+<td>Whether to include formula numbers in the output Markdown text. The default is <code>false</code>.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>restructurePages</code></td>
+<td><code>boolean</code></td>
+<td>Whether to restructure results across multiple pages. The default is <code>false</code>.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>mergeTables</code></td>
+<td><code>boolean</code></td>
+<td>Please refer to the description of the <code>merge_tables</code> parameter in the <code>restructure_pages</code> method of the PaddleOCR-VL object. Valid only when <code>restructurePages</code> is <code>true</code>.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>relevelTitles</code></td>
+<td><code>boolean</code></td>
+<td>Please refer to the description of the <code>relevel_titles</code> parameter in the <code>restructure_pages</code> method of the PaddleOCR-VL object. Valid only when <code>restructurePages</code> is <code>true</code>.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>returnMarkdownImages</code></td>
+<td><code>boolean</code></td>
+<td>Whether to return the images referenced in the Markdown. Default <code>true</code>; when set to <code>false</code>, <code>markdown.images</code> is <code>null</code> or omitted and the server skips image encoding / URL upload.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>outputFormats</code></td>
+<td><code>array</code> | <code>null</code></td>
+<td>Optional. List of extra document formats to return. By default, no extra formats are returned. Currently only <code>"docx"</code> is supported.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code>|<code>null</code></td>
+<td>Whether to return visualization result images and intermediate images during the processing.<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>Pass <code>true</code>: Return images.</li>
+<li>Pass <code>false</code>: Do not return images.</li>
+<li>If this parameter is not provided in the request body or <code>null</code> is passed: Follow the setting in the configuration file <code>Serving.visualize</code>.</li>
+</ul>
+<br/>For example, add the following field in the configuration file:<br/>
+<pre><code>Serving:
+  visualize: False</code></pre>Images will not be returned by default, and the default behavior can be overridden by the <code>visualize</code> parameter in the request body. If this parameter is not set in either the request body or the configuration file (or <code>null</code> is passed in the request body and the configuration file is not set), images will be returned by default.</td>
+<td>No</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is processed successfully, the <code>result</code> in the response body has the following attributes:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>layoutParsingResults</code></td>
+<td><code>array</code></td>
+<td>Layout parsing results. The array length is 1 (for image input) or the actual number of document pages processed (for PDF input). For PDF input, each element in the array represents the result of each actual page processed in the PDF file.</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>Input data information.</td>
+</tr>
+</tbody>
+</table>
+<p>Image and other binary file fields in the element schema below (e.g. <code>outputImages</code>, <code>inputImage</code>, <code>markdown.images</code>, <code>exports</code>) are returned inline as Base64 strings by default; when the server is configured to return URLs, those values become pre-signed URLs while the field types remain unchanged. See the "Returning Binary Content as URLs" section of the <a href="../inference_deployment/serving/serving.en.md">Serving Deployment Guide</a> for configuration.</p>
+<p>Each element in <code>layoutParsingResults</code> is an <code>object</code> with the following attributes:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>A simplified version of the <code>res</code> field in the JSON representation of the results generated by the <code>predict</code> method of the object, with the <code>input_path</code> and <code>page_index</code> fields removed.</td>
+</tr>
+<tr>
+<td><code>markdown</code></td>
+<td><code>object</code></td>
+<td>Markdown results.</td>
+</tr>
+<tr>
+<td><code>outputImages</code></td>
+<td><code>object</code>|<code>null</code></td>
+<td>Refer to the <code>img</code> property description of the prediction results. The image is in JPEG format, Base64-encoded by default; returned as a pre-signed URL when URL-return mode is enabled.</td>
+</tr>
+<tr>
+<td><code>inputImage</code></td>
+<td><code>string</code>|<code>null</code></td>
+<td>Input image. The image is in JPEG format, Base64-encoded by default; returned as a pre-signed URL when URL-return mode is enabled.</td>
+</tr>
+<tr>
+<td><code>exports</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Optional additional exports. Present only when <code>outputFormats</code> is set. Example: <code>{"docx": {"content": "..."}}</code>, where <code>content</code> is the Base64-encoded file content by default, or a pre-signed URL when URL-return mode is enabled.</td>
+</tr>
+</tbody>
+</table>
+<p><code>markdown</code> is an <code>object</code> with the following properties:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>text</code></td>
+<td><code>string</code></td>
+<td>Markdown text.</td>
+</tr>
+<tr>
+<td><code>images</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Key-value pairs of relative paths to Markdown images and their image data. Values are Base64-encoded by default; returned as pre-signed URLs when URL-return mode is enabled. The field is <code>null</code> or omitted when <code>returnMarkdownImages</code> is <code>false</code> in the request.</td>
+</tr>
+</tbody>
+</table>
+<ul>
+  <li><b><code>restructurePages</code></b></li>
+</ul>
+<p>Restructure results across multiple pages.</p>
+<p><code>POST /restructure-pages</code></p>
+
+<ul>
+  <li>The request body has the following properties:</li>
+</ul>
+
+<table>
+  <thead>
+    <tr>
+      <th>Name</th>
+      <th>Type</th>
+      <th>Description</th>
+      <th>Required</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td><code>pages</code></td>
+      <td><code>array</code></td>
+      <td>An array of pages.</td>
+      <td>Yes</td>
+    </tr>
+    <tr>
+    <td><code>mergeTables</code></td>
+    <td><code>boolean</code></td>
+    <td>Please refer to the description of the <code>merge_tables</code> parameter in the <code>restructure_pages</code> method of the PaddleOCR-VL object.</td>
+    <td>No</td>
+    </tr>
+    <tr>
+    <td><code>relevelTitles</code></td>
+    <td><code>boolean</code></td>
+    <td>Please refer to the description of the <code>relevel_titles</code> parameter in the <code>restructure_pages</code> method of the PaddleOCR-VL object.</td>
+    <td>No</td>
+    </tr>
+    <tr>
+    <td><code>concatenatePages</code></td>
+    <td><code>boolean</code></td>
+    <td>Please refer to the description of the <code>concatenate_pages</code> parameter in the <code>restructure_pages</code> method of the PaddleOCR-VL object.</td>
+    <td>No</td>
+    </tr>
+    <tr>
+    <td><code>prettifyMarkdown</code></td>
+    <td><code>boolean</code></td>
+    <td>Whether to output beautified Markdown text. The default is <code>true</code>.</td>
+    <td>No</td>
+    </tr>
+    <tr>
+    <td><code>showFormulaNumber</code></td>
+    <td><code>boolean</code></td>
+    <td>Whether to include formula numbers in the output Markdown text. The default is <code>false</code>.</td>
+    <td>No</td>
+    </tr>
+    <tr>
+    <td><code>returnMarkdownImages</code></td>
+    <td><code>boolean</code></td>
+    <td>Whether to return the images referenced in the Markdown. Default <code>true</code>; when set to <code>false</code>, <code>markdown.images</code> is <code>null</code> or omitted and the server skips image encoding / URL upload.</td>
+    <td>No</td>
+    </tr>
+    <tr>
+    <td><code>outputFormats</code></td>
+    <td><code>array</code> | <code>null</code></td>
+    <td>Optional extra export formats; same meaning as <code>outputFormats</code> on <code>infer</code>. Only <code>"docx"</code> is supported.</td>
+    <td>No</td>
+    </tr>
+  </tbody>
+</table>
+
+<p>Each element in <code>pages</code> is an <code>object</code> with the following properties:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>The <code>prunedResult</code> object returned by the <code>infer</code> operation.</td>
+</tr>
+<tr>
+<td><code>markdownImages</code></td>
+<td><code>object</code>|<code>null</code></td>
+<td>The <code>images</code> property of the <code>markdown</code> object returned by the <code>infer</code> operation.</td>
+</tr>
+</tbody>
+</table>
+
+<ul>
+  <li>When the request is processed successfully, the <code>result</code> field in the response body has the following properties:</li>
+</ul>
+
+<table>
+  <thead>
+    <tr>
+      <th>Name</th>
+      <th>Type</th>
+      <th>Description</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td><code>layoutParsingResults</code></td>
+      <td><code>array</code></td>
+      <td>The restructured layout parsing results. For the fields that every element contains, please refer to the description of the result returned by the <code>infer</code> operation (excluding visualization result images and intermediate images).</td>
+    </tr>
+  </tbody>
+</table>
+</details>
+<details><summary>Multi-Language Service Invocation Examples</summary>
+<details>
+<summary>Python</summary>
+
+<pre><code class="language-python">
+import base64
+import requests
+import pathlib
+
+BASE_URL = "http://localhost:8080"
+
+image_path = "./demo.jpg"
+
+# Encode the local image in Base64
+with open(image_path, "rb") as file:
+    image_bytes = file.read()
+    image_data = base64.b64encode(image_bytes).decode("ascii")
+
+payload = {
+    "file": image_data, # Base64-encoded file content or file URL
+    "fileType": 1, # File type, 1 indicates an image file
+}
+
+response = requests.post(BASE_URL + "/layout-parsing", json=payload)
+assert response.status_code == 200, (response.status_code, response.text)
+
+result = response.json()["result"]
+pages = []
+for i, res in enumerate(result["layoutParsingResults"]):
+    pages.append({"prunedResult": res["prunedResult"], "markdownImages": res["markdown"].get("images")})
+    for img_name, img in res["outputImages"].items():
+        img_path = f"{img_name}_{i}.jpg"
+        pathlib.Path(img_path).parent.mkdir(exist_ok=True)
+        with open(img_path, "wb") as f:
+            f.write(base64.b64decode(img))
+        print(f"Output image saved at {img_path}")
+
+payload = {
+    "pages": pages,
+    "concatenatePages": True,
+}
+
+response = requests.post(BASE_URL + "/restructure-pages", json=payload)
+assert response.status_code == 200, (response.status_code, response.text)
+
+result = response.json()["result"]
+res = result["layoutParsingResults"][0]
+print(res["prunedResult"])
+md_dir = pathlib.Path("markdown")
+md_dir.mkdir(exist_ok=True)
+(md_dir / "doc.md").write_text(res["markdown"]["text"])
+for img_path, img in res["markdown"]["images"].items():
+    img_path = md_dir / img_path
+    img_path.parent.mkdir(parents=True, exist_ok=True)
+    img_path.write_bytes(base64.b64decode(img))
+print(f"Markdown document saved at {md_dir / 'doc.md'}")
+</code></pre></details>
+
+<details><summary>C++</summary>
+
+<pre><code class="language-cpp">#include &lt;iostream&gt;
+#include &lt;filesystem&gt;
+#include &lt;fstream&gt;
+#include &lt;vector&gt;
+#include &lt;string&gt;
+#include "cpp-httplib/httplib.h" // https://github.com/Huiyicc/cpp-httplib
+#include "nlohmann/json.hpp" // https://github.com/nlohmann/json
+#include "base64.hpp" // https://github.com/tobiaslocker/base64
+
+namespace fs = std::filesystem;
+
+int main() {
+    httplib::Client client("localhost", 8080);
+
+    const std::string filePath = "./demo.jpg";
+
+    std::ifstream file(filePath, std::ios::binary | std::ios::ate);
+    if (!file) {
+        std::cerr << "Error opening file: " << filePath << std::endl;
+        return 1;
+    }
+
+    std::streamsize size = file.tellg();
+    file.seekg(0, std::ios::beg);
+    std::vector<char> buffer(size);
+    if (!file.read(buffer.data(), size)) {
+        std::cerr << "Error reading file." << std::endl;
+        return 1;
+    }
+
+    std::string bufferStr(buffer.data(), static_cast<size_t>(size));
+    std::string encodedFile = base64::to_base64(bufferStr);
+
+    nlohmann::json jsonObj;
+    jsonObj["file"] = encodedFile;
+    jsonObj["fileType"] = 1;
+
+    auto response = client.Post("/layout-parsing", jsonObj.dump(), "application/json");
+
+    if (response && response->status == 200) {
+        nlohmann::json jsonResponse = nlohmann::json::parse(response->body);
+        auto result = jsonResponse["result"];
+
+        if (!result.is_object() || !result.contains("layoutParsingResults")) {
+            std::cerr << "Unexpected response format." << std::endl;
+            return 1;
+        }
+
+        const auto& results = result["layoutParsingResults"];
+        for (size_t i = 0; i < results.size(); ++i) {
+            const auto& res = results[i];
+
+            if (res.contains("prunedResult")) {
+                std::cout << "Layout result [" << i << "]: " << res["prunedResult"].dump() << std::endl;
+            }
+
+            if (res.contains("outputImages") && res["outputImages"].is_object()) {
+                for (auto& [imgName, imgBase64] : res["outputImages"].items()) {
+                    std::string outputPath = imgName + "_" + std::to_string(i) + ".jpg";
+                    fs::path pathObj(outputPath);
+                    fs::path parentDir = pathObj.parent_path();
+                    if (!parentDir.empty() && !fs::exists(parentDir)) {
+                        fs::create_directories(parentDir);
+                    }
+
+                    std::string decodedImage = base64::from_base64(imgBase64.get<std::string>());
+
+                    std::ofstream outFile(outputPath, std::ios::binary);
+                    if (outFile.is_open()) {
+                        outFile.write(decodedImage.c_str(), decodedImage.size());
+                        outFile.close();
+                        std::cout << "Saved image: " << outputPath << std::endl;
+                    } else {
+                        std::cerr << "Failed to save image: " << outputPath << std::endl;
+                    }
+                }
+            }
+        }
+    } else {
+        std::cerr << "Request failed." << std::endl;
+        if (response) {
+            std::cerr << "HTTP status: " << response->status << std::endl;
+            std::cerr << "Response body: " << response->body << std::endl;
+        }
+        return 1;
+    }
+
+    return 0;
+}
+</code></pre></details>
+
+<details><summary>Java</summary>
+
+<pre><code class="language-java">import okhttp3.*;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.util.Base64;
+import java.nio.file.Paths;
+import java.nio.file.Files;
+
+public class Main {
+    public static void main(String[] args) throws IOException {
+        String API_URL = "http://localhost:8080/layout-parsing";
+        String imagePath = "./demo.jpg";
+
+        File file = new File(imagePath);
+        byte[] fileContent = java.nio.file.Files.readAllBytes(file.toPath());
+        String base64Image = Base64.getEncoder().encodeToString(fileContent);
+
+        ObjectMapper objectMapper = new ObjectMapper();
+        ObjectNode payload = objectMapper.createObjectNode();
+        payload.put("file", base64Image);
+        payload.put("fileType", 1);
+
+        OkHttpClient client = new OkHttpClient();
+        MediaType JSON = MediaType.get("application/json; charset=utf-8");
+
+        RequestBody body = RequestBody.create(JSON, payload.toString());
+
+        Request request = new Request.Builder()
+                .url(API_URL)
+                .post(body)
+                .build();
+
+        try (Response response = client.newCall(request).execute()) {
+            if (response.isSuccessful()) {
+                String responseBody = response.body().string();
+                JsonNode root = objectMapper.readTree(responseBody);
+                JsonNode result = root.get("result");
+
+                JsonNode layoutParsingResults = result.get("layoutParsingResults");
+                for (int i = 0; i < layoutParsingResults.size(); i++) {
+                    JsonNode item = layoutParsingResults.get(i);
+                    int finalI = i;
+                    JsonNode prunedResult = item.get("prunedResult");
+                    System.out.println("Pruned Result [" + i + "]: " + prunedResult.toString());
+
+                    JsonNode outputImages = item.get("outputImages");
+                    outputImages.fieldNames().forEachRemaining(imgName -> {
+                        try {
+                            String imgBase64 = outputImages.get(imgName).asText();
+                            byte[] imgBytes = Base64.getDecoder().decode(imgBase64);
+                            String imgPath = imgName + "_" + finalI + ".jpg";
+                            
+                            File outputFile = new File(imgPath);
+                            File parentDir = outputFile.getParentFile();
+                            if (parentDir != null && !parentDir.exists()) {
+                                parentDir.mkdirs();
+                                System.out.println("Created directory: " + parentDir.getAbsolutePath());
+                            }
+                            
+                            try (FileOutputStream fos = new FileOutputStream(outputFile)) {
+                                fos.write(imgBytes);
+                                System.out.println("Saved image: " + imgPath);
+                            }
+                        } catch (IOException e) {
+                            System.err.println("Failed to save image: " + e.getMessage());
+                        }
+                    });
+                }
+            } else {
+                System.err.println("Request failed with HTTP code: " + response.code());
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Go</summary>
+
+<pre><code class="language-go">package main
+
+import (
+    "bytes"
+    "encoding/base64"
+    "encoding/json"
+    "fmt"
+    "io/ioutil"
+    "net/http"
+    "os"
+    "path/filepath"
+)
+
+func main() {
+    API_URL := "http://localhost:8080/layout-parsing"
+    filePath := "./demo.jpg"
+
+    fileBytes, err := ioutil.ReadFile(filePath)
+    if err != nil {
+        fmt.Printf("Error reading file: %v\n", err)
+        return
+    }
+    fileData := base64.StdEncoding.EncodeToString(fileBytes)
+
+    payload := map[string]interface{}{
+        "file":     fileData,
+        "fileType": 1,
+    }
+    payloadBytes, err := json.Marshal(payload)
+    if err != nil {
+        fmt.Printf("Error marshaling payload: %v\n", err)
+        return
+    }
+
+    client := &http.Client{}
+    req, err := http.NewRequest("POST", API_URL, bytes.NewBuffer(payloadBytes))
+    if err != nil {
+        fmt.Printf("Error creating request: %v\n", err)
+        return
+    }
+    req.Header.Set("Content-Type", "application/json")
+
+    res, err := client.Do(req)
+    if err != nil {
+        fmt.Printf("Error sending request: %v\n", err)
+        return
+    }
+    defer res.Body.Close()
+
+    if res.StatusCode != http.StatusOK {
+        fmt.Printf("Unexpected status code: %d\n", res.StatusCode)
+        return
+    }
+
+    body, err := ioutil.ReadAll(res.Body)
+    if err != nil {
+        fmt.Printf("Error reading response: %v\n", err)
+        return
+    }
+
+    type Markdown struct {
+        Text   string            `json:"text"`
+        Images map[string]string `json:"images"`
+    }
+
+    type LayoutResult struct {
+        PrunedResult map[string]interface{} `json:"prunedResult"`
+        Markdown     Markdown               `json:"markdown"`
+        OutputImages map[string]string      `json:"outputImages"`
+        InputImage   *string                `json:"inputImage"`
+    }
+
+    type Response struct {
+        Result struct {
+            LayoutParsingResults []LayoutResult `json:"layoutParsingResults"`
+            DataInfo             interface{}    `json:"dataInfo"`
+        } `json:"result"`
+    }
+
+    var respData Response
+    if err := json.Unmarshal(body, &respData); err != nil {
+        fmt.Printf("Error parsing response: %v\n", err)
+        return
+    }
+
+    for i, res := range respData.Result.LayoutParsingResults {
+        fmt.Printf("Result %d - prunedResult: %+v\n", i, res.PrunedResult)
+
+        mdDir := fmt.Sprintf("markdown_%d", i)
+        os.MkdirAll(mdDir, 0755)
+        mdFile := filepath.Join(mdDir, "doc.md")
+        if err := os.WriteFile(mdFile, []byte(res.Markdown.Text), 0644); err != nil {
+            fmt.Printf("Error writing markdown file: %v\n", err)
+        } else {
+            fmt.Printf("Markdown document saved at %s\n", mdFile)
+        }
+
+        for path, imgBase64 := range res.Markdown.Images {
+            fullPath := filepath.Join(mdDir, path)
+            if err := os.MkdirAll(filepath.Dir(fullPath), 0755); err != nil {
+                fmt.Printf("Error creating directory for markdown image: %v\n", err)
+                continue
+            }
+            imgBytes, err := base64.StdEncoding.DecodeString(imgBase64)
+            if err != nil {
+                fmt.Printf("Error decoding markdown image: %v\n", err)
+                continue
+            }
+            if err := os.WriteFile(fullPath, imgBytes, 0644); err != nil {
+                fmt.Printf("Error saving markdown image: %v\n", err)
+            }
+        }
+
+        for name, imgBase64 := range res.OutputImages {
+            imgBytes, err := base64.StdEncoding.DecodeString(imgBase64)
+            if err != nil {
+                fmt.Printf("Error decoding output image %s: %v\n", name, err)
+                continue
+            }
+            filename := fmt.Sprintf("%s_%d.jpg", name, i)
+            
+            if err := os.MkdirAll(filepath.Dir(filename), 0755); err != nil {
+                fmt.Printf("Error creating directory for output image: %v\n", err)
+                continue
+            }
+            
+            if err := os.WriteFile(filename, imgBytes, 0644); err != nil {
+                fmt.Printf("Error saving output image %s: %v\n", filename, err)
+            } else {
+                fmt.Printf("Output image saved at %s\n", filename)
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>C#</summary>
+
+<pre><code class="language-csharp">using System;
+using System.IO;
+using System.Net.Http;
+using System.Text;
+using System.Threading.Tasks;
+using Newtonsoft.Json.Linq;
+
+class Program
+{
+    static readonly string API_URL = "http://localhost:8080/layout-parsing";
+    static readonly string inputFilePath = "./demo.jpg";
+
+    static async Task Main(string[] args)
+    {
+        var httpClient = new HttpClient();
+
+        byte[] fileBytes = File.ReadAllBytes(inputFilePath);
+        string fileData = Convert.ToBase64String(fileBytes);
+
+        var payload = new JObject
+        {
+            { "file", fileData },
+            { "fileType", 1 }
+        };
+        var content = new StringContent(payload.ToString(), Encoding.UTF8, "application/json");
+
+        HttpResponseMessage response = await httpClient.PostAsync(API_URL, content);
+        response.EnsureSuccessStatusCode();
+
+        string responseBody = await response.Content.ReadAsStringAsync();
+        JObject jsonResponse = JObject.Parse(responseBody);
+
+        JArray layoutParsingResults = (JArray)jsonResponse["result"]["layoutParsingResults"];
+        for (int i = 0; i < layoutParsingResults.Count; i++)
+        {
+            var res = layoutParsingResults[i];
+            Console.WriteLine($"[{i}] prunedResult:\n{res["prunedResult"]}");
+
+            JObject outputImages = res["outputImages"] as JObject;
+            if (outputImages != null)
+            {
+                foreach (var img in outputImages)
+                {
+                    string imgName = img.Key;
+                    string base64Img = img.Value?.ToString();
+                    if (!string.IsNullOrEmpty(base64Img))
+                    {
+                        string imgPath = $"{imgName}_{i}.jpg";
+                        byte[] imageBytes = Convert.FromBase64String(base64Img);
+                        
+                        string directory = Path.GetDirectoryName(imgPath);
+                        if (!string.IsNullOrEmpty(directory) && !Directory.Exists(directory))
+                        {
+                            Directory.CreateDirectory(directory);
+                            Console.WriteLine($"Created directory: {directory}");
+                        }
+                        
+                        File.WriteAllBytes(imgPath, imageBytes);
+                        Console.WriteLine($"Output image saved at {imgPath}");
+                    }
+                }
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Node.js</summary>
+
+<pre><code class="language-js">const axios = require('axios');
+const fs = require('fs');
+const path = require('path');
+
+const API_URL = 'http://localhost:8080/layout-parsing';
+const imagePath = './demo.jpg';
+const fileType = 1;
+
+function encodeImageToBase64(filePath) {
+  const bitmap = fs.readFileSync(filePath);
+  return Buffer.from(bitmap).toString('base64');
+}
+
+const payload = {
+  file: encodeImageToBase64(imagePath),
+  fileType: fileType
+};
+
+axios.post(API_URL, payload)
+  .then(response => {
+    const results = response.data.result.layoutParsingResults;
+    results.forEach((res, index) => {
+      console.log(`\n[${index}] prunedResult:`);
+      console.log(res.prunedResult);
+
+      const outputImages = res.outputImages;
+      if (outputImages) {
+        Object.entries(outputImages).forEach(([imgName, base64Img]) => {
+          const imgPath = `${imgName}_${index}.jpg`;
+          
+          const directory = path.dirname(imgPath);
+          if (!fs.existsSync(directory)) {
+            fs.mkdirSync(directory, { recursive: true });
+            console.log(`Created directory: ${directory}`);
+          }
+          
+          fs.writeFileSync(imgPath, Buffer.from(base64Img, 'base64'));
+          console.log(`Output image saved at ${imgPath}`);
+        });
+      } else {
+        console.log(`[${index}] No outputImages.`);
+      }
+    });
+  })
+  .catch(error => {
+    console.error('Error during API request:', error.message || error);
+  });
+</code></pre></details>
+
+<details><summary>PHP</summary>
+
+<pre><code class="language-php">&lt;?php
+
+$API_URL = "http://localhost:8080/layout-parsing";
+$image_path = "./demo.jpg";
+
+$image_data = base64_encode(file_get_contents($image_path));
+$payload = array("file" => $image_data, "fileType" => 1);
+
+$ch = curl_init($API_URL);
+curl_setopt($ch, CURLOPT_POST, true);
+curl_setopt($ch, CURLOPT_POSTFIELDS, json_encode($payload));
+curl_setopt($ch, CURLOPT_HTTPHEADER, array('Content-Type: application/json'));
+curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+$response = curl_exec($ch);
+curl_close($ch);
+
+$result = json_decode($response, true)["result"]["layoutParsingResults"];
+
+foreach ($result as $i => $item) {
+    echo "[$i] prunedResult:\n";
+    print_r($item["prunedResult"]);
+
+    if (!empty($item["outputImages"])) {
+        foreach ($item["outputImages"] as $img_name => $img_base64) {
+            $output_image_path = "{$img_name}_{$i}.jpg";
+            
+            $directory = dirname($output_image_path);
+            if (!is_dir($directory)) {
+                mkdir($directory, 0777, true);
+                echo "Created directory: $directory\n";
+            }
+            
+            file_put_contents($output_image_path, base64_decode($img_base64));
+            echo "Output image saved at $output_image_path\n";
+        }
+    } else {
+        echo "No outputImages found for item $i\n";
+    }
+}
+?&gt;
+</code></pre></details>
+</details>
+
+### 4.4 Pipeline Configuration Adjustment Instructions
+
+> NOTE:
+> If you do not need to adjust pipeline configurations, you can ignore this section.
+
+Adjusting the PaddleOCR-VL configuration for service deployment involves only three steps:
+
+1. Obtain the configuration file
+2. Modify the configuration file
+3. Apply the configuration file
+
+#### 4.4.1 Obtain the Configuration File
+
+**If you are deploying using Docker Compose:**
+
+Download the corresponding pipeline configuration file based on the backend you are using:
+
+- vLLM: [pipeline_config_vllm.yaml](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/pipeline_config_vllm.yaml)
+- FastDeploy: [pipeline_config_fastdeploy.yaml](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/pipeline_config_fastdeploy.yaml)
+
+**If you are deploying by manually installing dependencies:**
+
+Execute the following command to generate the pipeline configuration file:
+
+```shell
+paddlex --get_pipeline_config PaddleOCR-VL
+```
+
+#### 4.4.2 Modify the Configuration File
+
+**Enhance VLM Inference Performance Using Acceleration Frameworks**
+
+To improve VLM inference performance using acceleration frameworks such as vLLM (refer to Section 3 for detailed instructions on starting the VLM inference service), modify the `VLRecognition.genai_config.backend` and `VLRecognition.genai_config.server_url` fields in the pipeline configuration file, as shown below:
+
+```yaml
+VLRecognition:
+  ...
+  genai_config:
+    backend: vllm-server
+    server_url: http://localhost:8118/v1
+```
+
+The Docker Compose solution already uses an acceleration framework by default.
+
+**Enable Document Image Preprocessing Functionality**
+
+The service started with default configurations does not support document preprocessing. If a client attempts to invoke this functionality, an error message will be returned. To enable document preprocessing, set `use_doc_preprocessor` to `True` in the pipeline configuration file and start the service using the modified configuration file.
+
+**Disable Result Visualization Functionality**
+
+The service returns visualized results by default, which introduces additional overhead. To disable this functionality, add the following configuration to the pipeline configuration file (`Serving` is a top-level field):
+
+```yaml
+Serving:
+  visualize: False
+```
+
+Additionally, you can set the `visualize` field to `false` in the request body to disable visualization for a single request.
+
+**Configure Returning Binary Content as URLs**
+
+By default, the service returns images and other binary content in the response inline as Base64. To return them as URLs instead, add the following configuration to the pipeline configuration file (`Serving` is a top-level field):
+
+```yaml
+Serving:
+  return_urls: True
+  extra:
+    file_storage:
+      type: bos
+      endpoint: https://bj.bcebos.com
+      bucket_name: some-bucket
+      ak: xxx
+      sk: xxx
+      key_prefix: deploy
+    url_expires_in: 3600
+```
+
+Currently, storing generated files in Baidu Intelligent Cloud Object Storage (BOS) and returning URLs is supported. The parameters are described as follows:
+
+- `endpoint`: Access domain name (required).
+- `ak`: Baidu Intelligent Cloud AK (required).
+- `sk`: Baidu Intelligent Cloud SK (required).
+- `bucket_name`: Storage bucket name (required).
+- `key_prefix`: Unified prefix for object keys.
+- `connection_timeout_in_mills`: Request timeout in milliseconds.
+- `url_expires_in`: URL validity period (in seconds). `-1` indicates it never expires.
+
+For more information on obtaining AK/SK and other details, refer to the [Baidu Intelligent Cloud Official Documentation](https://cloud.baidu.com/doc/BOS/index.html).
+
+**Limit the Number of PDF and Multi-Page TIFF Pages Parsed**
+
+By default, the service processes the entire PDF file. Multi-page TIFF files (`fileType=1`) are expanded page by page before processing. In production environments, if a PDF or multi-page TIFF contains too many pages, it may affect system stability, leading to processing timeouts or excessive resource usage. To ensure stable service operation, it is recommended to set a reasonable page limit based on actual needs. You can add the following configuration to the production configuration file (`Serving` is the top-level field):
+
+```yaml
+Serving:
+  extra:
+    max_num_input_imgs: <page limit, e.g., 100>
+```
+
+`max_num_input_imgs` caps the maximum number of pages processed for both PDF and multi-page TIFF. When it is set to `null`, there is no page limit for either.
+
+#### 4.4.3 Apply the Configuration File
+
+**If you deployed using Docker Compose**:
+
+**If you are deploying using Docker Compose:**
+
+Set the `services.paddleocr-vl-api.volumes` field in the Compose file to mount the pipeline configuration file to the `/home/paddleocr` directory. For example:
+
+```yaml
+services:
+  paddleocr-vl-api:
+    ...
+    volumes:
+      - ./pipeline_config_vllm.yaml:/home/paddleocr/pipeline_config_vllm.yaml
+...
+```
+
+> In a production environment, you can also build the image yourself and package the configuration file into the image.
+
+**If you are deploying by manually installing dependencies:**
+
+When starting the service, specify the `--pipeline` parameter as the path to your custom configuration file.
+
+## 5. Model Fine-Tuning
+
+If you find that PaddleOCR-VL does not meet accuracy expectations in specific business scenarios, we recommend using the [ERNIEKit suite](https://github.com/PaddlePaddle/ERNIE/tree/release/v1.4) to perform supervised fine-tuning (SFT) on the VLM (e.g. PaddleOCR-VL-0.9B). For detailed instructions, refer to the [ERNIEKit Official Documentation](https://github.com/PaddlePaddle/ERNIE/blob/release/v1.4/docs/paddleocr_vl_sft.md).
+
+> Currently, fine-tuning of layout analysis and ranking models is not supported.
diff --git a/docs/version3.x/pipeline_usage/PaddleOCR-VL.md b/docs/version3.x/pipeline_usage/PaddleOCR-VL.md
new file mode 100644
index 0000000..2949817
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/PaddleOCR-VL.md
@@ -0,0 +1,3129 @@
+---
+comments: true
+---
+
+# PaddleOCR-VL 使用教程
+
+> INFO:
+> PaddleOCR 为 PaddleOCR-VL 系列模型提供了统一的接口，方便用户快速上手和使用。除非另有说明，本教程及相关硬件使用教程中提到的 “PaddleOCR-VL” 均指 PaddleOCR-VL 系列模型（如 PaddleOCR-VL-1.6 等）；若特指 PaddleOCR-VL v1 版本，将另行明确标注。
+
+PaddleOCR-VL 是一款先进、高效的文档解析模型，专为文档中的元素识别设计。以其初代版本（PaddleOCR-VL v1）为例，其核心组件为 PaddleOCR-VL-0.9B，这是一种紧凑而强大的视觉语言模型（VLM），它由 NaViT 风格的动态分辨率视觉编码器与 ERNIE-4.5-0.3B 语言模型组成，能够实现精准的元素识别。该系列模型支持 109 种语言，并在识别复杂元素（如文本、表格、公式和图表）方面表现出色，同时保持极低的资源消耗。通过在广泛使用的公开基准与内部基准上的全面评测，PaddleOCR-VL 在页级文档解析与元素级识别均达到 SOTA 表现。它显著优于现有的基于Pipeline方案和文档解析多模态方案以及先进的通用多模态大模型，并具备更快的推理速度。这些优势使其非常适合在真实场景中落地部署。
+
+2026年1月29日，我们发布了 PaddleOCR-VL-1.5。PaddleOCR-VL-1.5 不仅以 94.5% 精度大幅刷新了评测集 OmniDocBench v1.5，更创新性地支持了异形框定位，使得 PaddleOCR-VL-1.5 在扫描、倾斜、弯折、屏幕拍摄及复杂光照等真实场景中均表现优异。此外，模型还新增了印章识别与文本检测识别能力，关键指标持续领跑。
+
+**2026年5月28日，我们发布了 PaddleOCR-VL-1.6。PaddleOCR-VL-1.6 以 96.3% 精度再次刷新评测集 OmniDocBench v1.6，并在 OmniDocBench v1.5、Real5-OmniDocBench 上同步达到全新 SOTA，文本、公式、表格识别全面领先开源与闭源方案。此外，模型在古籍、生僻字识别上大幅提升，印章、spotting、图表识别等多场景能力也显著增强，且模型结构与 PaddleOCR-VL-1.5 完全一致，支持零成本无缝迁移。**
+
+PaddleOCR-VL 整体由版面分析与 VLM 识别两个核心阶段组成。下图展示了一个简化的流程：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/5f2c42665e6a97d5726fe553241acd79361159a0/images/paddleocr_vl_1_5/process_step_ZH.png"/>
+
+在该流程中，第一阶段为版面分析：模型以整图作为输入，检测并定位图像中的各类版面元素（例如表格、公式等），同时确定其阅读顺序，并根据检测结果裁剪出对应的元素子图；第二阶段为 VLM 识别：将每个子图独立输入 VLM，生成对应的识别结果（例如 Markdown 文本），随后再按照版面分析阶段给出的顺序对各元素结果进行合并，得到整幅图像的完整解析结果。因此，**若需使用 PaddleOCR-VL 的完整能力，必须采用版面分析与 VLM 识别协同的完整流程，而不能仅单独使用 VLM。** 后文会多次涉及相关概念，请注意区分完整的 PaddleOCR-VL 流程与其中的 VLM 组件。以 PaddleOCR-VL v1 为例，版面分析模型为 PP-DocLayoutV2，VLM 为 PaddleOCR-VL-0.9B。需要特别说明的是，“PaddleOCR-VL-0.9B” 并不是 PaddleOCR-VL 的一个模型变种，而是 PaddleOCR-VL v1 完整流程中的 VLM 组件；这与常见 LLM / VLM 的命名习惯不同，例如 Qwen2-72B 通常表示 Qwen2 系列下的一个具体模型变体。
+
+**如果在使用过程中出现无法复现论文或 PaddleOCR 官网精度、模型输出大量幻觉文本等问题，首先应确认当前使用的是完整的 PaddleOCR-VL 流程，而不是仅使用其中的 VLM 组件。** 例如，直接通过 Transformers 本地执行 PaddleOCR-VL-0.9B 模型，或直接请求 vLLM / SGLang / FastDeploy 服务，都不等同于运行完整的 PaddleOCR-VL 流程。
+
+## 从这里开始
+
+请先根据硬件选择对应教程。
+
+| 硬件　　　　　　　　　　　　　 | 阅读哪篇教程　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　 |
+| --------------------------------| ----------------------------------------------------------------------------------------------|
+| x64 CPU　　　　　　　　　　　　| 继续阅读本教程。请使用第 1.2 节的手动安装路径；仅适用于 NVIDIA GPU 的 Docker 步骤不适用。　　|
+| 除 Blackwell 之外的 NVIDIA GPU | 继续阅读本教程。　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　 |
+| NVIDIA Blackwell GPU　　　　　 | 阅读 [PaddleOCR-VL NVIDIA Blackwell 架构 GPU 使用教程](./PaddleOCR-VL-NVIDIA-Blackwell.md)。 |
+| Apple Silicon　　　　　　　　　| 阅读 [PaddleOCR-VL Apple Silicon 使用教程](./PaddleOCR-VL-Apple-Silicon.md)。　　　　　　　　|
+| 昆仑芯 XPU　　　　　　　　　　 | 阅读 [PaddleOCR-VL 昆仑芯 XPU 使用教程](./PaddleOCR-VL-Kunlunxin-XPU.md)。　　　　　　　　　 |
+| 海光 DCU　　　　　　　　　　　 | 阅读 [PaddleOCR-VL 海光 DCU 使用教程](./PaddleOCR-VL-Hygon-DCU.md)。　　　　　　　　　　　　 |
+| 沐曦 GPU　　　　　　　　　　　 | 阅读 [PaddleOCR-VL 沐曦 GPU 使用教程](./PaddleOCR-VL-MetaX-GPU.md)。　　　　　　　　　　　　 |
+| 天数 GPU　　　　　　　　　　　 | 阅读 [PaddleOCR-VL 天数 GPU 使用教程](./PaddleOCR-VL-Iluvatar-GPU.md)。　　　　　　　　　　　|
+| 华为昇腾 NPU　　　　　　　　　 | 阅读 [PaddleOCR-VL 华为昇腾 NPU 使用教程](./PaddleOCR-VL-Huawei-Ascend-NPU.md)。　　　　　　 |
+| AMD GPU　　　　　　　　　　　　| 阅读 [PaddleOCR-VL AMD GPU 使用教程](./PaddleOCR-VL-AMD-GPU.md)。　　　　　　　　　　　　　　|
+| Intel Arc GPU　　　　　　　　　| 阅读 [PaddleOCR-VL Intel Arc GPU 使用教程](./PaddleOCR-VL-Intel-Arc-GPU.md)。　　　　　　　　|
+
+如果你只是想先确认 PaddleOCR-VL 支持在哪些硬件上部署，或是特定硬件支持哪些推理方式，请先阅读下方的 [PaddleOCR-VL 推理方式与硬件支持矩阵](#paddleocr-vl-对推理设备的支持情况)。
+
+## PaddleOCR-VL 推理方式与硬件支持矩阵 {#paddleocr-vl-对推理设备的支持情况}
+
+| 推理方式                      | NVIDIA GPU | 昆仑芯 XPU | 海光 DCU | 沐曦 GPU | 天数 GPU | 华为昇腾 NPU | x64 CPU | Apple Silicon | AMD GPU | Intel Arc GPU |
+| ------------------------- | ------- | ------- | ------ | ------ | ------ | -------- | ------- | ------------- | ------- | ------------- |
+| PaddlePaddle              | ✅       | ✅       | ✅      | ✅      | ✅      | 🚧       | ✅       | ✅             | ✅       | 🚧            |
+| Transformers              | ✅       | 🚧      | 🚧     | 🚧     | 🚧     | 🚧       | ✅       | 🚧            | 🚧      | 🚧            |
+| PaddlePaddle + vLLM       | ✅       | 🚧      | ✅      | 🚧     | 🚧     | ✅        | ❌       | ❌             | ✅       | ✅             |
+| PaddlePaddle + SGLang     | ✅       | 🚧      | 🚧     | 🚧     | 🚧     | 🚧       | ❌       | ❌             | 🚧      | 🚧            |
+| PaddlePaddle + FastDeploy | ✅       | ✅       | 🚧     | ✅      | ✅      | 🚧       | ❌       | ❌             | 🚧      | 🚧            |
+| PaddlePaddle + MLX-VLM    | ❌       | ❌       | ❌      | ❌      | ❌      | ❌        | ❌       | ✅             | ❌       | ❌             |
+| PaddlePaddle + llama.cpp  | ✅       | 🚧      | 🚧     | 🚧     | 🚧     | 🚧       | ✅       | 🚧            | 🚧      | 🚧            |
+| Transformers + vLLM       | ✅       | 🚧      | 🚧     | 🚧     | 🚧     | 🚧       | ❌       | ❌             | 🚧      | 🚧            |
+| Transformers + SGLang     | ✅       | 🚧      | 🚧     | 🚧     | 🚧     | 🚧       | ❌       | ❌             | 🚧      | 🚧            |
+| Transformers + FastDeploy | ✅       | 🚧      | 🚧     | 🚧     | 🚧     | 🚧       | ❌       | ❌             | 🚧      | 🚧            |
+| Transformers + MLX-VLM    | ❌       | ❌       | ❌      | ❌      | ❌      | ❌        | ❌       | ✅             | ❌       | ❌             |
+| Transformers + llama.cpp  | ✅       | 🚧      | 🚧     | 🚧     | 🚧     | 🚧       | ✅       | 🚧            | 🚧      | 🚧            |
+
+**状态说明**
+
+- `✅`：支持。
+- `🚧`：适配中或待进一步验证。
+- `❌`：当前不支持或不适用。
+
+**推理方式说明**
+
+`PaddlePaddle` 表示版面分析模型与 VLM 均在本地使用飞桨框架推理；在实际执行时，各模块会根据模型形态解析为 `paddle_static` 或 `paddle_dynamic`。`Transformers` 表示版面分析模型与 VLM 均在本地通过 `transformers` 引擎推理；其余推理方式遵循 `版面分析模型推理方式 + VLM 推理方式` 的格式，后者中的 VLM 部分由独立推理服务承担，例如 `PaddlePaddle + vLLM` 表示版面分析模型在客户端本地使用飞桨框架推理，VLM 由 vLLM 服务推理。
+
+**判定补充说明**
+
+- vLLM、SGLang 和 FastDeploy 无法在 Windows 上原生运行，请使用我们提供的 Docker 镜像。
+- 由于不同库之间存在依赖冲突，使用 Transformers + vLLM 等混合推理方式时，通常需要将版面分析模型和 VLM 服务部署在不同环境中。
+
+## 环境要求
+
+使用 NVIDIA GPU 推理时，需要注意 Compute Capability（简称 CC）和 CUDA 版本（简称 CUDA）是否满足要求：
+
+- PaddlePaddle：CC ≥ 7.0，CUDA ≥ 11.8
+- Transformers：CC ≥ 7.0，CUDA ≥ 11.8
+- vLLM：CC ≥ 8.0，CUDA ≥ 12.6
+- SGLang：8.0 ≤ CC < 12.0，CUDA ≥ 12.6
+- FastDeploy：8.0 ≤ CC < 12.0，CUDA ≥ 12.6
+- CC ≥ 8 的常见显卡包括 RTX 30/40/50 系列及 A10/A100 等，更多型号可查看 [CUDA GPU 计算能力](https://developer.nvidia.cn/cuda-gpus)。
+
+此外，虽然 vLLM 可在 T4/V100 等 CC 7.x 的 NVIDIA GPU 上启动，但容易出现超时或 OOM，不推荐使用。
+
+## 本教程支持的使用目标
+
+本教程是 x64 CPU 用户和除 Blackwell 之外的 NVIDIA GPU 用户的默认路径。
+
+
+| 目标             | 在本教程中的支持情况                                                                    | 从哪里开始阅读                                                      |
+| -------------- | ----------------------------------------------------------------------------- | ------------------------------------------------------------ |
+| 本地直接推理         | 支持。x64 CPU 用户应使用第 1.2 节的手动安装路径。                                               | 阅读第 1 节“本地运行环境准备”和第 2 节“快速开始”。                                   |
+| 客户端 + VLM 推理服务 | 支持。                                | 先完成本地直接推理，再阅读第 3 节“使用 VLM 推理服务”。                       |
+| 完整 API 服务      | 支持。x64 CPU 用户应使用手动部署路径；除 Blackwell 之外的 NVIDIA GPU 用户可使用 Docker Compose 或手动部署。 | 阅读第 4 节“服务化部署”：如需 Docker Compose，阅读第 4.1 节；如需手动部署，先完成第 1 节“本地运行环境准备”，然后阅读第 4.2 节。如需支持更高并发，请参考[高性能服务化部署方案](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/hps/README.md)。 |
+| 模型微调           | 支持。                                                                           | 阅读第 5 节“模型微调”。                                               |
+
+## 1. 本地运行环境准备
+
+此步骤主要介绍如何搭建 PaddleOCR-VL 的本地运行环境。本教程主要适用于 **x64 CPU** 和 **除 Blackwell 之外的 NVIDIA GPU** 用户，其他硬件请优先参考上文列出的专用教程。
+
+在本教程中，有以下两种本地运行环境准备方式：
+
+- 方法一：使用官方 Docker 镜像（仅适用于 NVIDIA GPU）。
+
+- 方法二：手动安装推理引擎和 PaddleOCR（x64 CPU 和 NVIDIA GPU 均可使用）。
+
+**我们强烈推荐采用 Docker 镜像的方式，以最大程度减少可能出现的环境问题。**
+
+### 1.1 方法一：使用 Docker 镜像
+
+我们推荐使用官方 Docker 镜像（要求 Docker 版本 >= 19.03，机器装配有 GPU 且 NVIDIA 驱动支持 CUDA 12.6 或以上版本）：
+
+```shell
+docker run \
+    -it \
+    --gpus all \
+    --network host \
+    --user root \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu \
+    /bin/bash
+# 在容器中调用 PaddleOCR CLI 或 Python API
+```
+
+如果您希望在无法连接互联网的环境中使用 PaddleOCR-VL，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu` （镜像的大小约为 8 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu-offline`（镜像大小约为 10 GB）。您需要在可以联网的机器上拉取镜像，将镜像导入到离线机器，然后在离线机器使用该镜像启动容器。例如：
+
+```shell
+# 在能够联网的机器上执行
+docker pull ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu-offline
+# 将镜像保存到文件中
+docker save ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:latest-nvidia-gpu-offline -o paddleocr-vl-latest-nvidia-gpu-offline.tar
+
+# 将镜像文件传输到离线机器
+
+# 在离线机器上执行
+docker load -i paddleocr-vl-latest-nvidia-gpu-offline.tar
+# 之后可以在离线机器上使用 `docker run` 启动容器
+```
+
+镜像中仅预装飞桨框架，未安装其他推理引擎（如 Transformers）。如果希望使用其他推理引擎，建议采用方法二手动安装（不建议在预装飞桨框架的环境中安装）。
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本的镜像，可以将标签中的 `latest` 替换为 PaddleOCR 版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-vl:paddleocr3.3-nvidia-gpu-offline`
+
+### 1.2 方法二：手动安装推理引擎和 PaddleOCR {#manual-install-inference-engine-and-paddleocr}
+
+如果您无法使用 Docker，也可以手动安装推理引擎和 PaddleOCR。本文档验证过的 Python 版本范围为 3.9–3.13。
+
+**我们强烈推荐您在虚拟环境中安装 PaddleOCR-VL，以避免发生依赖冲突。** 例如，使用 Python venv 标准库创建虚拟环境：
+
+```shell
+# 创建虚拟环境
+python -m venv .venv_paddleocr
+# 激活环境
+source .venv_paddleocr/bin/activate
+```
+
+请先根据所选推理引擎安装对应依赖：
+
+- 使用 PaddlePaddle 推理时：请安装 3.2.1 及以上版本的 PaddlePaddle。常见安装方式如下（**注意不允许同时安装 CPU 和 GPU 版本的 PaddlePaddle**）：
+
+```shell
+# NVIDIA GPU（以 CUDA 12.6 为例）
+python -m pip install paddlepaddle-gpu==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/cu126/
+
+# x64 CPU
+python -m pip install paddlepaddle==3.2.1 -i https://www.paddlepaddle.org.cn/packages/stable/cpu/
+```
+
+  对于其他 CUDA 版本，请参考飞桨安装文档：[https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/develop/install/pip/linux-pip.html](https://www.paddlepaddle.org.cn/install/quick?docurl=/documentation/docs/zh/develop/install/pip/linux-pip.html)
+
+- 使用 `transformers` 推理时：请参考 [Transformers 官方文档](https://huggingface.co/docs/transformers/installation) 安装 `transformers` 及其依赖的底层推理框架。
+
+完成推理引擎安装后，再执行如下命令安装 PaddleOCR-VL 所需的基础包：
+
+```shell
+python -m pip install -U "paddleocr[doc-parser]"
+```
+
+## 2. 快速开始
+
+此步骤主要介绍如何使用 PaddleOCR-VL，包括如何通过 CLI 命令行方式和 Python API 方式进行使用。
+
+PaddleOCR-VL 支持 CLI 命令行方式和 Python API 两种使用方式，其中 CLI 命令行方式更简单，适合快速验证功能，而 Python API 方式更灵活，适合集成到现有项目中。下文示例默认使用本地飞桨框架；在实际执行时，各模块会根据模型形态解析为 `paddle_static` 或 `paddle_dynamic`。如需切换到 `transformers` 引擎，可在 CLI 中追加 `--engine transformers`，或在 Python API 初始化时传入 `engine="transformers"`。
+
+> IMPORTANT:
+> 本节所介绍的方法主要用于快速验证，其推理速度、显存占用及稳定性表现未必能满足生产环境的要求。**若需部署至生产环境，我们强烈建议使用专门的 VLM 推理服务**，具体方法请参考下一节。
+
+### 2.1 命令行方式体验
+
+首次运行时，PaddleOCR-VL 会自动下载官方模型，请确保当前环境可以联网，并预留一定的下载和初始化时间。
+
+下面给出一组可直接复制的示例命令。建议首次体验时附加 `--save_path ./output`，便于在当前目录下查看保存结果：
+
+```shell
+# NVIDIA GPU
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --save_path ./output
+
+# 昆仑芯 XPU
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --device xpu --save_path ./output
+
+# 海光 DCU
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --device dcu --save_path ./output
+
+# 沐曦 GPU
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --device metax_gpu --save_path ./output
+
+# Apple Silicon
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --device cpu --save_path ./output
+
+# 华为昇腾 NPU 
+# 华为昇腾 NPU 请参考第 3 章节使用 PaddlePaddle + vLLM 的方式进行推理
+
+# 通过 --use_doc_orientation_classify 指定是否使用文档方向分类模型
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --use_doc_orientation_classify True --save_path ./output
+
+# 通过 --use_doc_unwarping 指定是否使用文本图像矫正模块
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --use_doc_unwarping True --save_path ./output
+
+# 通过 --use_layout_detection 指定是否使用版面分析模块
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --use_layout_detection False --save_path ./output
+```
+
+执行成功后，终端会打印结构化结果；如果设置了 `--save_path ./output`，结果文件也会保存到当前目录下的 `output` 中，便于继续查看和调试。
+
+若需切换到 `transformers` 引擎，可参考以下示例：
+
+```bash
+paddleocr doc_parser -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --engine transformers --save_path ./output
+```
+
+<details><summary><b>命令行支持更多参数设置，点击展开以查看命令行参数的详细说明</b></summary>
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，必填。<br/>
+<b>说明：</b>如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<br/>
+<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/demo_paper.png">示例</a>；<br/>
+<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td><b>含义：</b>指定推理结果文件保存的路径。<br/>
+<b>说明：</b>如果不设置，推理结果将不会保存到本地。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>pipeline_version</code></td>
+<td><b>含义：</b>指定产线版本。<br/>
+<b>说明：</b>当前可选值为<code>"v1"</code>、<code>"v1.5"</code>和<code>"v1.6"</code>。</td>
+<td><code>str</code></td>
+<td>"v1.6"</td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>含义：</b>版面分析模型名称。<br/>
+<b>说明：</b>如果不设置，将会使用默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>含义：</b>版面分析模型的目录路径。<br/>
+<b>说明：</b>如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>版面模型得分阈值。<br/>
+<b>说明：</b><code>0-1</code> 之间的任意浮点数。<br/>
+如果不设置，将使用初始化的默认值。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>版面分析是否使用后处理NMS。<br/>
+<b>说明：</b>如果不设置，将使用初始化的默认值。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>版面区域检测模型检测框的扩张系数。<br/>
+<b>说明：</b>任意大于 <code>0</code>  浮点数。<br/>
+如果不设置，将使用初始化的默认值
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>版面分析中模型输出的检测框的合并处理模式。<br/>
+<b>说明：</b>
+<ul>
+<li><b>large</b>，设置为large时，表示在模型输出的检测框中，对于互相重叠包含的检测框，只保留外部最大的框，删除重叠的内部框；</li>
+<li><b>small</b>，设置为small，表示在模型输出的检测框中，对于互相重叠包含的检测框，只保留内部被包含的小框，删除重叠的外部框；</li>
+<li><b>union</b>，不进行框的过滤处理，内外框都保留；</li>
+</ul>
+如果不设置，将使用初始化的参数值。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>vl_rec_model_name</code></td>
+<td><b>含义：</b>多模态识别模型名称。<br/>
+<b>说明：</b>如果不设置，将会使用默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>vl_rec_model_dir</code></td>
+<td><b>含义：</b>多模态识别模型目录路径。<br/>
+<b>说明：</b>如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+<td></td>
+</tr>
+<tr>
+<td><code>vl_rec_backend</code></td>
+<td><b>含义：</b>多模态识别模型使用的推理后端。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>vl_rec_server_url</code></td>
+<td><b>含义：</b>如果多模态识别模型使用推理服务，该参数用于指定服务器URL。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>vl_rec_max_concurrency</code></td>
+<td><b>含义：</b>如果多模态识别模型使用推理服务，该参数用于指定最大并发请求数。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>vl_rec_api_model_name</code></td>
+<td><b>含义：</b>如果多模态识别模型使用推理服务，该参数用于指定服务的模型名称。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>vl_rec_api_key</code></td>
+<td><b>含义：</b>如果多模态识别模型使用推理服务，该参数用于指定服务的 API key。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>含义：</b>文档方向分类模型的名称。<br/>
+<b>说明：</b>如果不设置，将使用初始化的默认值。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>文档方向分类模型的目录路径。<br/>
+<b>说明：</b>如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>含义：</b>文本图像矫正模型的名称。<br/>
+<b>说明：</b>如果不设置，将使用初始化的默认值。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。<br/>
+<b>说明：</b>如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否加载并使用文档方向分类模块。<br/>
+<b>说明：</b>如果不设置，将使用初始化的默认值，默认初始化为<code>False</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。<br/>
+<b>说明：</b>如果不设置，将使用初始化的默认值，默认初始化为<code>False</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>含义：</b>是否加载并使用版面分析模块。<br/>
+<b>说明：</b>如果不设置，将使用初始化的默认值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>含义：</b>是否使用图表解析功能。<br/>
+<b>说明：</b>如果不设置，将使用初始化的默认值，默认初始化为<code>False</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>含义：</b>是否使用印章识别功能。<br/>
+<b>说明：</b>如果不设置，将使用初始化的默认值，默认初始化为<code>False</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_ocr_for_image_block</code></td>
+<td><b>含义：</b>是否对图片中的文字进行识别。<br/>
+<b>说明：</b>如果不设置，将使用初始化的默认值，默认初始化为<code>False</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>format_block_content</code></td>
+<td><b>含义：</b>控制是否将 <code>block_content</code> 中的内容格式化为Markdown格式。<br/>
+<b>说明：</b>如果不设置，将使用初始化的默认值，默认初始化为<code>False</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>merge_layout_blocks</code></td>
+<td><b>含义：</b>控制是否对跨栏或上下交错分栏的版面检测框进行合并。<br/>
+<b>说明：</b>如果不设置，将使用初始化的默认值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>markdown_ignore_labels</code></td>
+<td><b>含义：</b>需要在Markdown中忽略的版面标签。<br/>
+<b>说明：</b>如果不设置，将使用初始化的默认值，默认初始化为<code>['number','footnote','header','header_image','footer','footer_image','aside_text']</code>。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_shape_mode</code></td>
+<td><b>含义：</b>用于指定版面分析结果的几何形状表示模式。该参数决定了检测区域（如文本块、图片、表格等）边界的计算方式及展示形态。<br/>
+<b>说明：</b>取值说明：
+    <ul>
+    <li><b>rect (矩形)</b>: 输出水平正向的边界框（包含 x1, y1, x2, y2）。适用于标准的水平排版版面。</li>
+    <li><b>quad (四边形)</b>: 输出由四个顶点组成的任意四边形。适用于存在倾斜、透视变形的区域。</li>
+    <li><b>poly (多边形)</b>: 输出由多个坐标点组成的闭合轮廓。适用于形状不规则或具有弧度的版面元素，精度最高。</li>
+    <li><b>auto (自动)</b>: 系统根据检测目标的复杂程度和置信度，自动选择最合适的形状表达方式。</li>
+    </ul>
+</td>
+<td><code>str</code></td>
+<td>"auto"</td>
+</tr>
+<tr>
+<td><code>use_queues</code></td>
+<td><b>含义：</b>用于控制是否启用内部队列。<br/>
+<b>说明：</b>当设置为 <code>True</code> 时，数据加载（如将 PDF 页面渲染为图像）、版面分析模型处理以及 VLM 推理将分别在独立线程中异步执行，通过队列传递数据，从而提升效率。对于页数较多的 PDF 文档，或是包含大量图像或 PDF 文件的目录，这种方式尤其高效。如果不设置，将使用初始化的默认值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>prompt_label</code></td>
+<td><b>含义：</b>VL模型的 prompt 类型设置。<br/>
+<b>说明：</b>当且仅当 <code>use_layout_detection=False</code> 时生效。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>repetition_penalty</code></td>
+<td><b>含义：</b>VL模型采样使用的重复惩罚参数。</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>temperature</code></td>
+<td><b>含义：</b>VL模型采样使用的温度参数。</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>top_p</code></td>
+<td><b>含义：</b>VL模型采样使用的top-p参数。</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>min_pixels</code></td>
+<td><b>含义：</b>VL模型预处理图像时允许的最小像素数。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>max_pixels</code></td>
+<td><b>含义：</b>VL模型预处理图像时允许的最大像素数。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+</ul>
+如果不设置，将使用初始化的默认值，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>fp32</code>、<code>fp16</code>。</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+
+</tbody>
+</table>
+</details>
+
+运行结果会被打印到终端上，默认配置的 PaddleOCR-VL 的运行结果如下：
+
+<details><summary> 👉点击展开</summary>
+<pre>
+<code>
+{'res': {'input_path': 'paddleocr_vl_demo.png', 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_layout_detection': True, 'use_chart_recognition': False, 'format_block_content': False}, 'layout_det_res': {'input_path': None, 'page_index': None, 'boxes': [{'cls_id': 6, 'label': 'doc_title', 'score': 0.9636914134025574, 'coordinate': [np.float32(131.31366), np.float32(36.450516), np.float32(1384.522), np.float32(127.984665)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9281806349754333, 'coordinate': [np.float32(585.39465), np.float32(158.438), np.float32(930.2184), np.float32(182.57469)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9840355515480042, 'coordinate': [np.float32(9.023666), np.float32(200.86115), np.float32(361.41583), np.float32(343.8828)]}, {'cls_id': 14, 'label': 'image', 'score': 0.9871416091918945, 'coordinate': [np.float32(775.50574), np.float32(200.66502), np.float32(1503.3807), np.float32(684.9304)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9801855087280273, 'coordinate': [np.float32(9.532196), np.float32(344.90594), np.float32(361.4413), np.float32(440.8244)]}, {'cls_id': 17, 'label': 'paragraph_title', 'score': 0.9708921313285828, 'coordinate': [np.float32(28.040405), np.float32(455.87976), np.float32(341.7215), np.float32(520.7117)]}, {'cls_id': 24, 'label': 'vision_footnote', 'score': 0.9002962708473206, 'coordinate': [np.float32(809.0692), np.float32(703.70044), np.float32(1488.3016), np.float32(750.5238)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9825374484062195, 'coordinate': [np.float32(8.896561), np.float32(536.54895), np.float32(361.05237), np.float32(655.8058)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9822263717651367, 'coordinate': [np.float32(8.971573), np.float32(657.4949), np.float32(362.01715), np.float32(774.625)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9767460823059082, 'coordinate': [np.float32(9.407074), np.float32(776.5216), np.float32(361.31067), np.float32(846.82874)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9868153929710388, 'coordinate': [np.float32(8.669495), np.float32(848.2543), np.float32(361.64703), np.float32(1062.8568)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9826608300209045, 'coordinate': [np.float32(8.8025055), np.float32(1063.8615), np.float32(361.46588), np.float32(1182.8524)]}, {'cls_id': 22, 'label': 'text', 'score': 0.982555627822876, 'coordinate': [np.float32(8.820602), np.float32(1184.4663), np.float32(361.66394), np.float32(1302.4507)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9584776759147644, 'coordinate': [np.float32(9.170288), np.float32(1304.2161), np.float32(361.48898), np.float32(1351.7483)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9782056212425232, 'coordinate': [np.float32(389.1618), np.float32(200.38202), np.float32(742.7591), np.float32(295.65146)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9844875931739807, 'coordinate': [np.float32(388.73303), np.float32(297.18463), np.float32(744.00024), np.float32(441.3034)]}, {'cls_id': 17, 'label': 'paragraph_title', 'score': 0.9680547714233398, 'coordinate': [np.float32(409.39468), np.float32(455.89386), np.float32(721.7174), np.float32(520.9387)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9741666913032532, 'coordinate': [np.float32(389.71606), np.float32(536.8138), np.float32(742.7112), np.float32(608.00165)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9840384721755981, 'coordinate': [np.float32(389.30988), np.float32(609.39636), np.float32(743.09247), np.float32(750.3231)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9845995306968689, 'coordinate': [np.float32(389.13272), np.float32(751.7772), np.float32(743.058), np.float32(894.8815)]}, {'cls_id': 22, 'label': 'text', 'score': 0.984852135181427, 'coordinate': [np.float32(388.83267), np.float32(896.0371), np.float32(743.58215), np.float32(1038.7345)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9804865717887878, 'coordinate': [np.float32(389.08478), np.float32(1039.9119), np.float32(742.7585), np.float32(1134.4897)]}, {'cls_id': 22, 'label': 'text', 'score': 0.986461341381073, 'coordinate': [np.float32(388.52643), np.float32(1135.8137), np.float32(743.451), np.float32(1352.0085)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9869391918182373, 'coordinate': [np.float32(769.8341), np.float32(775.66235), np.float32(1124.9813), np.float32(1063.207)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9822869896888733, 'coordinate': [np.float32(770.30383), np.float32(1063.938), np.float32(1124.8295), np.float32(1184.2192)]}, {'cls_id': 17, 'label': 'paragraph_title', 'score': 0.9689218997955322, 'coordinate': [np.float32(791.3042), np.float32(1199.3169), np.float32(1104.4521), np.float32(1264.6985)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9713128209114075, 'coordinate': [np.float32(770.4253), np.float32(1279.6072), np.float32(1124.6917), np.float32(1351.8672)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9236552119255066, 'coordinate': [np.float32(1153.9058), np.float32(775.5814), np.float32(1334.0654), np.float32(798.1581)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9857938885688782, 'coordinate': [np.float32(1151.5197), np.float32(799.28015), np.float32(1506.3619), np.float32(991.1156)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9820687174797058, 'coordinate': [np.float32(1151.5686), np.float32(991.91095), np.float32(1506.6023), np.float32(1110.8875)]}, {'cls_id': 22, 'label': 'text', 'score': 0.9866049885749817, 'coordinate': [np.float32(1151.6919), np.float32(1112.1301), np.float32(1507.1611), np.float32(1351.9504)]}]}}}
+</code></pre></details>
+
+运行结果及保存接口的详细说明可参考 [2.2 Python 脚本方式集成](#22-python) 中的结果解释。
+
+**注：**由于 PaddleOCR-VL 的默认模型较大，推理速度可能较慢，建议实际推理使用 [3. 使用 VLM 推理服务](#3-vlm) 方式进行快速推理。
+
+### 2.2 Python 脚本方式集成
+
+命令行方式是为了快速体验查看效果，一般来说，在项目中，往往需要通过代码集成。您可以通过几行代码即可完成 PaddleOCR-VL 的快速推理：
+
+```python
+from pathlib import Path
+
+from paddleocr import PaddleOCRVL
+
+output_dir = Path("./output")
+output_dir.mkdir(parents=True, exist_ok=True)
+
+# NVIDIA GPU
+pipeline = PaddleOCRVL()
+# 昆仑芯 XPU
+# pipeline = PaddleOCRVL(device="xpu")
+# 海光 DCU
+# pipeline = PaddleOCRVL(device="dcu")
+# 沐曦 GPU
+# pipeline = PaddleOCRVL(device="metax_gpu")
+# Apple Silicon
+# pipeline = PaddleOCRVL(device="cpu")
+# 华为昇腾 NPU 
+# 华为昇腾 NPU 请参考第 3 章节使用 PaddlePaddle + vLLM 的方式进行推理
+
+# pipeline = PaddleOCRVL(use_doc_orientation_classify=True) # 通过 use_doc_orientation_classify 指定是否使用文档方向分类模型
+# pipeline = PaddleOCRVL(use_doc_unwarping=True) # 通过 use_doc_unwarping 指定是否使用文本图像矫正模块
+# pipeline = PaddleOCRVL(use_layout_detection=False) # 通过 use_layout_detection 指定是否使用版面分析模块
+
+output = pipeline.predict("https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png")
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_json(save_path=output_dir) ## 保存当前图像的结构化json结果
+    res.save_to_markdown(save_path=output_dir) ## 保存当前图像的markdown格式的结果
+    res.save_to_word(save_path=output_dir) ## 保存当前图像的Word格式的结果
+```
+
+若需切换到 `transformers` 引擎，可参考以下示例：
+
+```python
+from pathlib import Path
+
+from paddleocr import PaddleOCRVL
+
+output_dir = Path("./output")
+output_dir.mkdir(parents=True, exist_ok=True)
+
+pipeline = PaddleOCRVL(engine="transformers")
+output = pipeline.predict("https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png")
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_json(save_path=output_dir) ## 保存当前图像的结构化json结果
+    res.save_to_markdown(save_path=output_dir) ## 保存当前图像的markdown格式的结果
+    res.save_to_word(save_path=output_dir) ## 保存当前图像的Word格式的结果
+```
+
+如果是 PDF 文件，会将 PDF 的每一页单独处理，每一页的 Markdown 文件也会对应单独的结果。如果您希望对多页的推理结果进行跨页表格合并、重建多级标题和合并多页结果等需求，可以通过如下方式实现：
+
+```python
+from pathlib import Path
+
+from paddleocr import PaddleOCRVL
+
+input_file = "./your_pdf_file.pdf"
+output_dir = Path("./output")
+output_dir.mkdir(parents=True, exist_ok=True)
+
+pipeline = PaddleOCRVL()
+
+output = pipeline.predict(input=input_file)
+
+pages_res = list(output)
+
+output = pipeline.restructure_pages(pages_res)
+
+# output = pipeline.restructure_pages(pages_res, merge_tables=True) # 合并跨页表格
+# output = pipeline.restructure_pages(pages_res, merge_tables=True, relevel_titles=True) # 合并跨页表格，重建多级标题
+# output = pipeline.restructure_pages(pages_res, merge_tables=True, relevel_titles=True, concatenate_pages=True) # 合并跨页表格，重建多级标题，合并多页结果为一页
+
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_json(save_path=output_dir) ## 保存当前图像的结构化json结果
+    res.save_to_markdown(save_path=output_dir) ## 保存当前图像的markdown格式的结果
+```
+
+如果您需要处理多个文件，**建议将包含文件的目录路径，或者文件路径列表传入 `predict` 方法**，以最大化处理效率。例如：
+
+```python
+# `imgs` 目录中包含多张待处理图像：file1.png、file2.png、file3.png
+# 传入目录路径
+output = pipeline.predict("imgs")
+# 或者传入文件路径列表
+output = pipeline.predict(["imgs/file1.png", "imgs/file2.png", "imgs/file3.png"])
+# 以上两种方式的处理效率高于下列方式：
+# for file in ["imgs/file1.png", "imgs/file2.png", "imgs/file3.png"]:
+#     output = pipeline.predict(file)
+```
+
+**注：**
+
+- 在示例代码中，`use_doc_orientation_classify`、`use_doc_unwarping` 参数默认均设置为 `False`，分别表示关闭文档方向分类、文本图像矫正功能，如果需要使用这些功能，可以手动设置为 `True`。
+
+在上述 Python 脚本中，执行了如下几个步骤：
+
+<details><summary>（1）实例化对象，具体参数说明如下：</summary>
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>pipeline_version</code></td>
+<td><b>含义：</b>指定产线版本。<br/>
+<b>说明：</b>当前可选值为<code>"v1"</code>、<code>"v1.5"</code>和<code>"v1.6"</code>。</td>
+<td><code>str</code></td>
+<td>"v1.6"</td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>含义：</b>版面分析模型名称。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会使用默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>含义：</b>版面分析模型的目录路径。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>版面模型得分阈值。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：<code>0-1</code> 之间的任意浮点数；</li>
+<li><b>dict</b>： <code>{0:0.1}</code> key为类别ID，value为该类别的阈值；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用初始化的默认值。</li>
+</ul>
+</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>版面分析是否使用后处理NMS。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用初始化的默认值。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>版面区域检测模型检测框的扩张系数。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：任意大于 <code>0</code>  浮点数；</li>
+<li><b>Tuple[float,float]</b>：在横纵两个方向各自的扩张系数；</li>
+<li><b>dict</b>，dict的key为<b>int</b>类型，代表<code>cls_id</code>, value为<b>tuple</b>类型，如<code>{0: (1.1, 2.0)}</code>，表示将模型输出的第0类别检测框中心不变，宽度扩张1.1倍，高度扩张2.0倍；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用初始化的默认值。</li>
+</ul>
+</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>版面区域检测的重叠框过滤方式。<br/>
+<b>说明：</b>
+<ul>
+<li><b>str</b>：<code>large</code>，<code>small</code>，<code>union</code>，分别表示重叠框过滤时选择保留大框，小框还是同时保留；</li>
+<li><b>dict</b>： dict的key为<b>int</b>类型，代表<code>cls_id</code>，value为<b>str</b>类型，如<code>{0: "large", 2: "small"}</code>，表示对第0类别检测框使用large模式，对第2类别检测框使用small模式；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用初始化的默认值。</li>
+</ul>
+</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>vl_rec_model_name</code></td>
+<td><b>含义：</b>多模态识别模型名称。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会使用默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>vl_rec_model_dir</code></td>
+<td><b>含义：</b>多模态识别模型目录路径。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>vl_rec_backend</code></td>
+<td><b>含义：</b>多模态识别模型使用的推理后端。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>vl_rec_server_url</code></td>
+<td><b>含义：</b>如果多模态识别模型使用推理服务，该参数用于指定服务器URL。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>vl_rec_max_concurrency</code></td>
+<td><b>含义：</b>如果多模态识别模型使用推理服务，该参数用于指定最大并发请求数。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>vl_rec_api_model_name</code></td>
+<td><b>含义：</b>如果多模态识别模型使用推理服务，该参数用于指定服务的模型名称。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>vl_rec_api_key</code></td>
+<td><b>含义：</b>如果多模态识别模型使用推理服务，该参数用于指定服务的 API key。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>含义：</b>文档方向分类模型的名称。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会使用默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>文档方向分类模型的目录路径。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>含义：</b>文本图像矫正模型的名称。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会使用默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否加载并使用文档方向分类模块。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用初始化的默认值，默认初始化为<code>False</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用初始化的默认值，默认初始化为<code>False</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>含义：</b>是否加载并使用版面分析模块。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用初始化的默认值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>含义：</b>是否使用图表解析功能。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用初始化的默认值，默认初始化为<code>False</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>含义：</b>是否使用印章识别功能。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用初始化的默认值，默认初始化为<code>False</code>。</td>
+<td><code>bool|None</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_ocr_for_image_block</code></td>
+<td><b>含义：</b>是否对图片中的文字进行识别。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用初始化的默认值，默认初始化为<code>False</code>。</td>
+<td><code>bool|None</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>format_block_content</code></td>
+<td><b>含义：</b>控制是否将 <code>block_content</code> 中的内容格式化为Markdown格式。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用初始化的默认值，默认初始化为<code>False</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>merge_layout_blocks</code></td>
+<td><b>含义：</b>控制是否对跨栏或上下交错分栏的版面检测框进行合并。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用初始化的默认值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>markdown_ignore_labels</code></td>
+<td><b>含义：</b>需要在Markdown中忽略的版面标签。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用初始化的默认值，默认初始化为<code>['number','footnote','header','header_image','footer','footer_image','aside_text']</code>。</td>
+<td><code>list|None</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_queues</code></td>
+<td><b>含义：</b>用于控制是否启用内部队列。<br/>
+<b>说明：</b>当设置为 <code>True</code> 时，数据加载（如将 PDF 页面渲染为图像）、版面分析模型处理以及 VLM 推理将分别在独立线程中异步执行，通过队列传递数据，从而提升效率。对于页数较多的 PDF 文档，或是包含大量图像或 PDF 文件的目录，这种方式尤其高效。如果设置为<code>None</code>，将使用初始化的默认值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+<li><b>None</b>：如果设置为<code>None</code>，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+</tbody>
+</table>
+
+</details>
+
+<details><summary>（2）调用 PaddleOCR-VL 对象的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。另外，PaddleOCR-VL 还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。以下是 <code>predict()</code> 方法的参数及其说明：</summary>
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/demo_paper.png">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]。</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否在推理时使用文档方向分类模块。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否在推理时使用文本图像矫正模块。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>含义：</b>是否在推理时使用版面区域检测排序模块。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_chart_recognition</code></td>
+<td><b>含义：</b>是否使用图表解析功能。<br/>
+<b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_seal_recognition</code></td>
+<td><b>含义：</b>是否使用印章识别功能。<br/>
+<b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_ocr_for_image_block</code></td>
+<td><b>含义：</b>是否对图片中的文字进行识别。<br/>
+<b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_shape_mode</code></td>
+<td><b>含义：</b>用于指定版面分析结果的几何形状表示模式。该参数决定了检测区域（如文本块、图片、表格等）边界的计算方式及展示形态。<br/>
+<b>说明：</b>取值说明：
+    <ul>
+    <li><b>rect (矩形)</b>: 输出水平正向的边界框（包含 x1, y1, x2, y2）。适用于标准的水平排版版面。</li>
+    <li><b>quad (四边形)</b>: 输出由四个顶点组成的任意四边形。适用于存在倾斜、透视变形的区域。</li>
+    <li><b>poly (多边形)</b>: 输出由多个坐标点组成的闭合轮廓。适用于形状不规则或具有弧度的版面元素，精度最高。</li>
+    <li><b>auto (自动)</b>: 系统根据检测目标的复杂程度和置信度，自动选择最合适的形状表达方式。</li>
+    </ul>
+</td>
+<td><code>str</code></td>
+<td>"auto"</td>
+</tr>
+<tr>
+<td><code>use_queues</code></td>
+<td>参数含义与实例化参数基本相同。设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>prompt_label</code></td>
+<td><b>含义：</b>VL模型的 prompt 类型设置。<br/>
+<b>说明：</b>
+当且仅当 <code>use_layout_detection=False</code> 时生效。可填写参数为 <code>ocr</code>、<code>formula</code>、<code>table</code> 、<code>seal</code>、<code>chart</code>和 <code>spotting</code>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>format_block_content</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>repetition_penalty</code></td>
+<td><b>含义：</b>VL模型采样使用的重复惩罚参数。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>temperature</code></td>
+<td><b>含义：</b>VL模型采样使用的温度参数。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>top_p</code></td>
+<td><b>含义：</b>VL模型采样使用的top-p参数。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>min_pixels</code></td>
+<td><b>含义：</b>VL模型预处理图像时允许的最小像素数。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>max_pixels</code></td>
+<td><b>含义：</b>VL模型预处理图像时允许的最大像素数。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>max_new_tokens</code></td>
+<td><b>含义：</b>VL模型生成的最大token数。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>merge_layout_blocks</code></td>
+<td><b>含义：</b>控制是否对跨栏或上下交错分栏的版面检测框进行合并。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>markdown_ignore_labels</code></td>
+<td><b>含义：</b>需要在Markdown中忽略的版面标签。</td>
+<td><code>list|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>vlm_extra_args</code></td>
+<td><b>含义：</b>VLM额外配置参数。<br/>
+<b>说明：</b>目前支持的自定义参数如下：
+<ul>
+  <li><code>ocr_min_pixels</code>：OCR 最小分辨率</li>
+  <li><code>ocr_max_pixels</code>：OCR 最大分辨率</li>
+  <li><code>table_min_pixels</code>：表格最小分辨率</li>
+  <li><code>table_max_pixels</code>：表格最大分辨率</li>
+  <li><code>chart_min_pixels</code>：图表最小分辨率</li>
+  <li><code>chart_max_pixels</code>：图表最大分辨率</li>
+  <li><code>formula_min_pixels</code>：公式最小分辨率</li>
+  <li><code>formula_max_pixels</code>：公式最大分辨率</li>
+  <li><code>seal_min_pixels</code>：印章最小分辨率</li>
+  <li><code>seal_max_pixels</code>：印章最大分辨率</li>
+</ul>
+</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</table>
+</details>
+
+<details><summary>（3）调用 PaddleOCR-VL 对象的 <code>restructure_pages()</code> 方法对推理预测的多页结果列表进行页面重建，该方法会返回一个重建后的多页结果或合并后的单页结果。以下是 <code>restructure_pages()</code> 方法的参数及其说明：</summary>
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>res_list</code></td>
+<td><b>含义：</b>多页 PDF 推理预测出的结果列表。</td>
+<td><code>list|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>merge_tables</code></td>
+<td><b>含义：</b>控制是否进行跨页表格合并。</td>
+<td><code>Bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>relevel_titles</code></td>
+<td><b>含义：</b>控制是否进行多级标题分级</td>
+<td><code>Bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>concatenate_pages</code></td>
+<td><b>含义：</b>控制是否拼接多页结果为一页</td>
+<td><code>Bool</code></td>
+<td><code>False</code></td>
+</tr>
+</table>
+</details>
+
+<details><summary>（4）对预测结果进行处理：每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为<code>json</code>文件的操作:</summary>
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化。</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效。</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效。</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致。</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效。</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效。</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将中间各个模块的可视化图像保存在png格式的图像</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径。</td>
+<td>无</td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_markdown()</code></td>
+<td rowspan="3">将图像或者PDF文件中的每一页分别保存为markdown格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>pretty</code></td>
+<td><code>bool</code></td>
+<td>是否美化 <code>markdown</code> 输出结果，将图表等进行居中操作，使 <code>markdown</code> 渲染后更美观。</td>
+<td>True</td>
+</tr>
+<tr>
+<td><code>show_formula_number</code></td>
+<td><code>bool</code></td>
+<td>控制是否在 <code>markdown</code> 中将保留公式编号。设置为 <code>True</code> 时，保留全部公式编号；<code>False</code> 则仅保留公式</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_html()</code></td>
+<td>将文件中的表格保存为html格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径。</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>save_to_xlsx()</code></td>
+<td>将文件中的表格保存为xlsx格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径。</td>
+<td>无</td>
+</tr>
+</table>
+
+<ul>
+  <li>调用<code>print()</code> 方法会将结果打印到终端，打印到终端的内容解释如下：
+    <ol start="1" type="1">
+      <li><code>input_path</code>: <code>(str)</code> 待预测图像的输入路径</li>
+      <li><code>page_index</code>: <code>(Union[int, None])</code> 如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+      <li><code>model_settings</code>: <code>(Dict[str, bool])</code> 配置产线所需的模型参数
+        <ol>
+          <li><code>use_doc_preprocessor</code>: <code>(bool)</code> 控制是否启用文档预处理子产线</li>
+          <li><code>use_layout_detection</code>: <code>(bool)</code> 控制是否启用版面分析模块</li>
+          <li><code>use_chart_recognition</code>: <code>(bool)</code> 控制是否开启图表识别功能</li>
+          <li><code>format_block_content</code>: <code>(bool)</code> 控制是否在<code>JSON</code>中保存格式化后的markdown内容</li>
+        </ol>
+      </li>
+      <li><code>doc_preprocessor_res</code>: <code>(Dict[str, Union[str, Dict[str, bool], int]])</code> 文档预处理子产线的输出结果。仅当<code>use_doc_preprocessor=True</code>时存在
+        <ol>
+          <li><code>input_path</code>: <code>(str)</code> 文档预处理子接受的图像路径，当输入为<code>numpy.ndarray</code>时，保存为<code>None</code>，此处为<code>None</code></li>
+          <li><code>page_index</code>: <code>None</code> 此处的输入为<code>numpy.ndarray</code>时，所以值为<code>None</code></li>
+          <li><code>model_settings</code>: <code>(Dict[str, bool])</code> 文档预处理子的模型配置参数
+            <ul>
+              <li><code>use_doc_orientation_classify</code>: <code>(bool)</code> 控制是否启用文档方向分类</li>
+              <li><code>use_doc_unwarping</code>: <code>(bool)</code> 控制是否启用文本图像扭曲矫正子模块</li>
+            </ul>
+          </li>
+          <li><code>angle</code>: <code>(int)</code> 文档图像方向分类子模块的预测结果，启用时返回实际角度值</li>
+        </ol>
+      </li>
+      <li><code>parsing_res_list</code>: <code>(List[Dict])</code> 解析结果的列表，每个元素为一个字典，列表顺序为解析后的阅读顺序。
+        <ol>
+          <li><code>block_bbox</code>: <code>(np.ndarray)</code> 版面区域的边界框。</li>
+          <li><code>block_label</code>: <code>(str)</code> 版面区域的标签，例如<code>text</code>、<code>table</code>等</li>
+          <li><code>block_content</code>: <code>(str)</code> 内容为版面区域内的内容。</li>
+          <li><code>block_id</code>: <code>(int)</code> 版面区域的索引，用于显示版面排序结果。</li>
+          <li><code>block_order</code>: <code>(int)</code> 版面区域的顺序，用于显示版面阅读顺序，对于非排序部分，默认值为 <code>None</code>。</li>
+        </ol>
+      </li>
+    </ol>
+  </li>
+  <li>调用<code>save_to_json()</code> 方法会将上述内容保存到指定的<code>save_path</code>中，如果指定为目录，则保存的路径为<code>save_path/{your_img_basename}_res.json</code>；如果指定为文件，则直接保存到该文件中。由于 JSON 文件不支持保存 numpy 数组，因此会将其中的<code>numpy.array</code>类型转换为列表形式。JSON 中的字段内容如下：
+    <ol start="1" type="1">
+      <li><code>input_path</code>: <code>(str)</code> 待预测图像的输入路径</li>
+      <li><code>page_index</code>: <code>(Union[int, None])</code> 如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+      <li><code>model_settings</code>: <code>(Dict[str, bool])</code> 配置产线所需的模型参数
+        <ol>
+          <li><code>use_doc_preprocessor</code>: <code>(bool)</code> 控制是否启用文档预处理子产线</li>
+          <li><code>use_layout_detection</code>: <code>(bool)</code> 控制是否启用版面分析模块</li>
+          <li><code>use_chart_recognition</code>: <code>(bool)</code> 控制是否开启图表识别功能</li>
+          <li><code>format_block_content</code>: <code>(bool)</code> 控制是否在<code>JSON</code>中保存格式化后的markdown内容</li>
+        </ol>
+      </li>
+      <li><code>doc_preprocessor_res</code>: <code>(Dict[str, Union[str, Dict[str, bool], int]])</code> 文档预处理子产线的输出结果。仅当<code>use_doc_preprocessor=True</code>时存在
+        <ol>
+          <li><code>input_path</code>: <code>(str)</code> 文档预处理子接受的图像路径，当输入为<code>numpy.ndarray</code>时，保存为<code>None</code>，此处为<code>None</code></li>
+          <li><code>page_index</code>: <code>None</code> 此处的输入为<code>numpy.ndarray</code>时，所以值为<code>None</code></li>
+          <li><code>model_settings</code>: <code>(Dict[str, bool])</code> 文档预处理子的模型配置参数
+            <ul>
+              <li><code>use_doc_orientation_classify</code>: <code>(bool)</code> 控制是否启用文档方向分类</li>
+              <li><code>use_doc_unwarping</code>: <code>(bool)</code> 控制是否启用文本图像扭曲矫正子模块</li>
+            </ul>
+          </li>
+          <li><code>angle</code>: <code>(int)</code> 文档图像方向分类子模块的预测结果，启用时返回实际角度值</li>
+        </ol>
+      </li>
+      <li><code>parsing_res_list</code>: <code>(List[Dict])</code> 解析结果的列表，每个元素为一个字典，列表顺序为解析后的阅读顺序。
+        <ol>
+          <li><code>block_bbox</code>: <code>(np.ndarray)</code> 版面区域的边界框。</li>
+          <li><code>block_label</code>: <code>(str)</code> 版面区域的标签，例如<code>text</code>、<code>table</code>等</li>
+          <li><code>block_content</code>: <code>(str)</code> 内容为版面区域内的内容。</li>
+          <li><code>block_id</code>: <code>(int)</code> 版面区域的索引，用于显示版面排序结果。</li>
+          <li><code>block_order</code>: <code>(int)</code> 版面区域的顺序，用于显示版面阅读顺序，对于非排序部分，默认值为 <code>None</code>。</li>
+        </ol>
+      </li>
+    </ol>
+  </li>
+  <li>调用<code>save_to_img()</code> 方法会将可视化结果保存到指定的<code>save_path</code>中，如果指定为目录，则会将版面区域检测可视化图像、全局OCR可视化图像、版面阅读顺序可视化图像等内容保存；如果指定为文件，则直接保存到该文件中。</li>
+  <li>调用<code>save_to_markdown()</code> 方法会将转化后的 Markdown 文件保存到指定的<code>save_path</code>中，保存的文件路径为<code>save_path/{your_img_basename}.md</code>。如果输入是 PDF 文件，建议直接指定目录，否则多个 Markdown 文件会被覆盖。</li>
+  <li>此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+    <table>
+      <thead>
+        <tr>
+          <th>属性</th>
+          <th>属性说明</th>
+        </tr>
+      </thead>
+      <tbody>
+        <tr>
+          <td><code>json</code></td>
+          <td>获取预测的 <code>json</code> 格式的结果</td>
+        </tr>
+        <tr>
+          <td><code>img</code></td>
+          <td>获取格式为 <code>dict</code> 的可视化图像</td>
+        </tr>
+        <tr>
+          <td><code>markdown</code></td>
+          <td>获取格式为 <code>dict</code> 的 markdown 结果</td>
+        </tr>
+      </tbody>
+    </table>
+    <ul>
+      <li><code>json</code> 属性获取的预测结果为 dict 类型的数据，相关内容与调用 <code>save_to_json()</code> 方法保存的内容一致。</li>
+      <li><code>img</code> 属性返回的预测结果是一个 dict 类型的数据。其中，键分别为 <code>ocr_res_img</code> 和 <code>preprocessed_img</code>，对应的值是两个 <code>Image.Image</code> 对象：一个用于显示 OCR 结果的可视化图像，另一个用于展示图像预处理的可视化图像。如果没有使用图像预处理子模块，则 dict 中只包含 <code>ocr_res_img</code>。</li>
+      <li><code>markdown</code> 属性获取的预测结果为 dict 类型的数据，相关内容与调用 <code>save_to_markdown()</code> 方法保存的内容一致。</li>
+    </ul>
+  </li>
+</ul>
+</details>
+
+<a id="3-vlm"></a>
+## 3. 使用 VLM 推理服务
+
+本节主要介绍如何在 PaddleOCR-VL 流程中接入 VLM 推理服务。这通常用于提升推理性能，并改善生产环境下的资源利用和服务稳定性。您既可以自行部署基于 vLLM、SGLang、FastDeploy、MLX-VLM、llama.cpp 等后端的 VLM 推理服务，也可以直接使用兼容的托管服务。不同硬件上的适用路径请以对应硬件教程为准。这一节对应“版面分析推理方式 + VLM 推理服务”类组合，其核心思路是：**客户端继续负责版面分析等完整流程中的其他环节，仅将 VLM 推理交给专用服务处理。**
+
+### 3.1 启动 VLM 推理服务
+
+> IMPORTANT:
+> 按照本节说明启动的服务仅负责 PaddleOCR-VL 流程中的 VLM 推理环节，不提供完整的端到端文档解析 API。强烈不建议直接通过 HTTP 请求或使用 OpenAI 客户端调用该服务处理文档图像。若您需要部署具备 PaddleOCR-VL 完整能力的服务，请参考后文的服务化部署部分。
+
+启动 VLM 推理服务有以下三种方式，任选一种即可：
+
+- 方法一：使用官方 Docker 镜像启动服务，目前支持：
+    - vLLM
+    - FastDeploy
+
+- 方法二：通过 PaddleOCR CLI 手动安装依赖后启动服务，目前支持：
+    - vLLM
+    - SGLang
+    - FastDeploy
+
+- 方法三：直接使用推理加速框架启动服务（此方法无法应用 PaddleOCR 预置的性能调优参数），目前支持：
+    - vLLM
+    - FastDeploy
+    - MLX-VLM
+    - llama.cpp
+
+**我们强烈推荐采用 Docker 镜像的方式，以最大程度减少可能出现的环境问题。**
+
+此外，[硅基流动](https://siliconflow.cn/)、[Novita AI](https://novita.ai/models-console/model-detail/paddlepaddle-paddleocr-vl) 等云平台还提供托管服务。若采用此类服务，可跳过本小节，直接阅读 [3.2 客户端使用方法](#32)。
+
+#### 3.1.1 方法一：使用 Docker 镜像
+
+PaddleOCR 提供了 Docker 镜像，用于快速启动 vLLM 或 FastDeploy 推理服务。可使用以下命令启动服务（要求 Docker 版本 >= 19.03，机器装配有 GPU 且NVIDIA驱动支持 CUDA 12.6 或以上版本）：
+
+=== "启动 vLLM 服务"
+
+    ```shell
+    docker run \
+        -it \
+        --rm \
+        --gpus all \
+        --network host \
+        ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu \
+        paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm
+    ```
+
+    如果您希望在无法连接互联网的环境中启动服务，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu`（镜像大小约为 13 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu-offline`（镜像大小约为 15 GB）。
+
+=== "启动 FastDeploy 服务"
+
+    ```shell
+    docker run \
+        -it \
+        --rm \
+        --gpus all \
+        --network host \
+        ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-nvidia-gpu \
+        paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend fastdeploy
+    ```
+
+    如果您希望在无法连接互联网的环境中启动服务，请将上述命令中的 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-nvidia-gpu`（镜像大小约为 43 GB）更换为离线版本镜像 `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-fastdeploy-server:latest-nvidia-gpu-offline`（镜像大小约为 45 GB）。
+
+启动 vLLM 或 FastDeploy 推理服务时，我们提供了一套默认参数设置。如果您有调整显存占用等更多参数的需求，可以自行配置更多参数。请参考 [3.3.1 服务端参数调整](#331) 创建配置文件，然后将该文件挂载到容器中，并在启动服务的命令中使用 `backend_config` 指定配置文件，以 vLLM 为例：
+
+```shell
+docker run \
+    -it \
+    --rm \
+    --gpus all \
+    --network host \
+    -v ./vllm_config.yml:/tmp/vllm_config.yml \
+    ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:latest-nvidia-gpu \
+    paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /tmp/vllm_config.yml
+```
+
+其中，`./vllm_config.yml` 表示宿主机当前目录下的本地配置文件路径；如果文件位于其他位置，请替换为实际绝对路径或带目录前缀的相对路径。
+
+> TIP:
+> 标签后缀为 `latest-xxx` 的镜像对应最新版本。
+> 如果本地已经存在对应的 `latest` 镜像，但希望使用最新功能或修复，建议在继续使用前重新执行一次 `docker pull` 更新镜像。
+> 如果希望使用特定版本的 PaddleOCR 镜像，可以将标签中的 `latest` 替换为对应版本号：`paddleocr<major>.<minor>`。
+> 例如：
+> `ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddleocr-genai-vllm-server:paddleocr3.3-nvidia-gpu-offline`
+
+#### 3.1.2 方法二：通过 PaddleOCR CLI 安装和使用
+
+**PaddleOCR CLI 已经为您解决了复杂的版本兼容性问题。您无需花费时间研究推理框架的文档，只需一条简单的命令即可安装推理框架所需的依赖环境。**
+
+由于推理加速框架可能与当前环境中的包存在依赖冲突，建议在虚拟环境中安装：
+
+```shell
+# 如果当前存在已激活的虚拟环境，先通过 `deactivate` 取消激活
+# 创建虚拟环境
+python -m venv .venv_vlm
+# 激活环境
+source .venv_vlm/bin/activate
+```
+
+vLLM 和 SGLang 依赖 FlashAttention，而安装 FlashAttention 时可能需要使用 nvcc 等 CUDA 编译工具。如果您的环境中没有这些工具（例如在使用 `paddleocr-vl` 镜像），可以从 [此仓库](https://github.com/mjun0812/flash-attention-prebuild-wheels) 获取 FlashAttention 的预编译版本（要求 2.8.2 版本），先安装预编译包，再执行后续命令。例如，在 `paddleocr-vl` 镜像中，执行 `python -m pip install https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.3.14/flash_attn-2.8.2+cu128torch2.8-cp310-cp310-linux_x86_64.whl`。对于 FastDeploy，无需执行此步骤。
+
+安装 PaddleOCR 及推理加速服务依赖，以 vLLM 为例：
+
+```shell
+# 安装 PaddleOCR
+python -m pip install "paddleocr[doc-parser]"
+# 安装推理加速服务依赖
+paddleocr install_genai_server_deps vllm
+```
+
+`paddleocr install_genai_server_deps` 命令用法：
+
+```shell
+paddleocr install_genai_server_deps <推理加速框架名称>
+```
+
+当前支持的框架名称为 `vllm`、`sglang` 和 `fastdeploy`，分别对应 vLLM、SGLang 和 FastDeploy。
+
+通过 `paddleocr install_genai_server_deps` 安装的 vLLM 与 SGLang 均为 **CUDA 12.6** 版本，请确保本地 NVIDIA 驱动与此版本一致或更高。
+
+> WARNING:
+> 目前 vLLM 和 SGLang 与 Transformers 引擎所需的 transformers 库版本存在冲突，因此同一环境中无法同时安装 Transformers 引擎与 vLLM 或 SGLang。如果使用 Transformers + vLLM 或 Transformers + SGLang 的推理方式，请将版面分析模型和 VLM 服务部署在不同环境中。
+
+安装完成后，可通过 `paddleocr genai_server` 命令启动服务：
+
+```shell
+paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --backend vllm --port 8118
+```
+
+该命令支持的参数如下：
+
+
+| 参数                 | 说明                                                  |
+| ------------------ | --------------------------------------------------- |
+| `--model_name`     | 模型名称                                                |
+| `--model_dir`      | 模型目录                                                |
+| `--host`           | 服务器主机名                                              |
+| `--port`           | 服务器端口号                                              |
+| `--backend`        | 后端名称，即使用的推理加速框架名称，可选 `vllm`、`sglang` 或 `fastdeploy` |
+| `--backend_config` | 可指定 YAML 文件，包含后端配置                                  |
+
+
+#### 3.1.3 方法三：直接使用推理加速框架启动服务
+
+**如果您需要安装自定义版本的推理框架并使用原生方式启动服务，请参考以下指引。请注意，使用原生方式启动时，将无法应用 PaddleOCR 预置的性能调优参数。**
+
+- vLLM：[参考此文档](https://docs.vllm.ai/projects/recipes/en/latest/PaddlePaddle/PaddleOCR-VL.html)
+- SGLang：[参考 SGLang 官方文档](https://docs.sglang.ai/)
+- FastDeploy：[参考此文档](https://paddlepaddle.github.io/FastDeploy/zh/best_practices/PaddleOCR-VL-0.9B/)
+- MLX-VLM：[参考此文档](./PaddleOCR-VL-Apple-Silicon.md)
+- llama.cpp：
+    1. 参考 [llama.cpp github](https://github.com/ggml-org/llama.cpp) 中的 `Quick start` 安装 llama.cpp。
+    2. 下载 gguf 格式的模型文件：[PaddlePaddle/PaddleOCR-VL-1.6-GGUF](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.6-GGUF)。
+    3. 执行以下命令启动推理服务，参数介绍可参考 [LLaMA.cpp HTTP Server](https://github.com/ggml-org/llama.cpp/blob/master/tools/server/README.md)：
+
+        ```shell
+        ./build/bin/llama-server \
+          -m /path/to/PaddleOCR-VL-1.6-GGUF.gguf \
+          --mmproj /path/to/PaddleOCR-VL-1.6-GGUF-mmproj.gguf  \
+          --port 8111  \
+          --host 0.0.0.0 \
+          --temp 0
+        ```
+
+### 3.2 客户端使用方法
+
+启动 VLM 推理服务后，客户端即可通过 PaddleOCR 调用该服务。本节既适用于调用 3.1 中自建的 VLM 推理服务，也适用于调用第三方提供的兼容托管服务。**请注意，由于客户端仍需要调用版面分析模型并完成其他流程环节，仍建议在 GPU 等加速设备上运行客户端，以获得更稳定和高效的性能。客户端环境配置请参考第 1 节，3.1 节介绍的环境配置仅适用于启动服务，不适用于客户端。若您希望客户端只通过 HTTP 接口调用 PaddleOCR-VL 的完整能力，请直接参考第 4 节“服务化部署”。**
+
+#### 3.2.1 CLI 调用
+
+可通过 `--vl_rec_backend` 指定后端类型（`vllm-server`、`sglang-server`、`fastdeploy-server`、`mlx-vlm-server` 或 `llama-cpp-server`），通过 `--vl_rec_server_url` 指定服务地址，例如：
+
+```shell
+paddleocr doc_parser --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png --vl_rec_backend vllm-server --vl_rec_server_url http://localhost:8118/v1
+```
+
+此外，可通过 `--vl_rec_api_model_name` 指定服务使用的模型名称，`--vl_rec_api_key` 指定鉴权使用的 API key。示例如下：
+
+使用通过 `vllm serve` 默认参数启动的服务：
+
+```shell
+paddleocr doc_parser \
+    --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png \
+    --vl_rec_backend vllm-server \
+    --vl_rec_server_url http://localhost:8000/v1 \
+    --vl_rec_api_model_name 'PaddlePaddle/PaddleOCR-VL-1.6'
+```
+
+硅基流动平台（目前只支持 PaddleOCR-VL-0.9B 和 PaddleOCR-VL-1.5-0.9B）：
+
+```shell
+paddleocr doc_parser \
+    --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png \
+    --pipeline_version v1.5 \
+    --vl_rec_backend vllm-server \
+    --vl_rec_server_url https://api.siliconflow.cn/v1 \
+    --vl_rec_api_model_name 'PaddlePaddle/PaddleOCR-VL-1.5' \
+    --vl_rec_api_key xxxxxx
+```
+
+Novita AI 平台（目前只支持 PaddleOCR-VL-0.9B，即 v1 版本模型）：
+
+```shell
+paddleocr doc_parser \
+    --input https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/paddleocr_vl_demo.png \
+    --pipeline_version v1 \
+    --vl_rec_backend vllm-server \
+    --vl_rec_server_url https://api.novita.ai/openai \
+    --vl_rec_api_model_name 'paddlepaddle/paddleocr-vl' \
+    --vl_rec_api_key xxxxxx
+```
+
+#### 3.2.2 Python API 调用
+
+创建 `PaddleOCRVL` 对象时传入 `vl_rec_backend` 指定后端类型（`vllm-server`、`sglang-server`、`fastdeploy-server`、`mlx-vlm-server` 或 `llama-cpp-server`）， `vl_rec_server_url` 指定服务地址，例如：
+
+```python
+pipeline = PaddleOCRVL(vl_rec_backend="vllm-server", vl_rec_server_url="http://localhost:8118/v1")
+```
+
+此外，可通过 `vl_rec_api_model_name` 指定服务使用的模型名称，`vl_rec_api_key` 指定鉴权使用的 API key。
+
+使用通过 `vllm serve` 默认参数启动的服务：
+
+```python
+pipeline = PaddleOCRVL(
+    vl_rec_backend="vllm-server",
+    vl_rec_server_url="http://localhost:8000/v1",
+    vl_rec_api_model_name="PaddlePaddle/PaddleOCR-VL-1.6",
+)
+```
+
+硅基流动平台（目前只支持 PaddleOCR-VL-0.9B 和 PaddleOCR-VL-1.5-0.9B）：
+
+```python
+pipeline = PaddleOCRVL(
+    pipeline_version="v1.5",
+    vl_rec_backend="vllm-server", 
+    vl_rec_server_url="https://api.siliconflow.cn/v1",
+    vl_rec_api_model_name="PaddlePaddle/PaddleOCR-VL-1.5",
+    vl_rec_api_key="xxxxxx",
+)
+```
+
+Novita AI 平台（目前只支持 PaddleOCR-VL-0.9B，即 v1 版本模型）：
+
+```python
+pipeline = PaddleOCRVL(
+    pipeline_version="v1",
+    vl_rec_backend="vllm-server", 
+    vl_rec_server_url="https://api.novita.ai/openai",
+    vl_rec_api_model_name="paddlepaddle/paddleocr-vl",
+    vl_rec_api_key="xxxxxx",
+)
+```
+
+### 3.3 性能调优
+
+默认配置无法保证在所有环境取得最优性能。如果您在实际使用中遇到性能问题，可以尝试以下优化方法。
+
+#### 3.3.1 服务端参数调整
+
+不同推理加速框架支持的参数不同，可参考各自官方文档了解可用参数及其调整时机：
+
+- [vLLM 官方参数调优指南](https://docs.vllm.ai/en/latest/configuration/optimization.html)
+- [SGLang 超参数调整文档](https://docs.sglang.ai/advanced_features/hyperparameter_tuning.html)
+- [FastDeploy 最佳实践文档](https://paddlepaddle.github.io/FastDeploy/zh/best_practices/PaddleOCR-VL-0.9B/)
+
+PaddleOCR VLM 推理服务支持通过配置文件进行调参。以下示例展示如何调整 vLLM 服务器的 `gpu-memory-utilization` 和 `max-num-seqs` 参数：
+
+1. 创建 YAML 文件 `vllm_config.yaml`，内容如下：
+
+   ```yaml
+   gpu-memory-utilization: 0.3
+   max-num-seqs: 128
+   ```
+
+2. 启动服务时指定配置文件路径，例如使用 `paddleocr genai_server` 命令：
+
+   ```shell
+   paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --backend vllm --backend_config vllm_config.yaml
+   ```
+
+如果使用支持进程替换（process substitution）的 shell（如 Bash），也可以无需创建配置文件，直接在启动服务时传入配置项：
+
+```bash
+paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --backend vllm --backend_config <(echo -e 'gpu-memory-utilization: 0.3\nmax-num-seqs: 128')
+```
+
+#### 3.3.2 客户端参数调整
+
+PaddleOCR 会将来自单张或多张输入图像中的子图分组并对服务器发起并发请求，因此并发请求数对性能影响显著。
+
+- 对 CLI 和 Python API，可通过 `vl_rec_max_concurrency` 参数调整最大并发请求数；
+- 对服务化部署，可修改配置文件中 `VLRecognition.genai_config.max_concurrency` 字段。
+
+当客户端与 VLM 推理服务为 1 对 1 且服务端资源充足时，可适当增加并发数以提升性能；若服务端需支持多个客户端或计算资源有限，则应降低并发数，以避免资源过载导致服务异常。
+
+#### 3.3.3 常用硬件性能调优建议
+
+以下配置均针对客户端与 VLM 推理服务为 1 对 1 的场景。
+
+**NVIDIA RTX 3060**
+
+- **服务端**
+    - vLLM：`gpu-memory-utilization: 0.7`
+    - FastDeploy：
+        - `gpu-memory-utilization: 0.7`
+        - `max-concurrency: 2048`
+
+## 4. 服务化部署
+
+此步骤主要介绍如何将 PaddleOCR-VL 部署为服务并调用。如果不要求服务具备并发处理请求的能力，可选择以下两种方式中的任一种：
+
+- 方法一：使用 Docker Compose 部署（推荐使用）。
+
+- 方法二：手动部署。
+
+上述两种方式一次仅能处理一个请求，如需支持并发请求，请参考[高性能服务化部署方案](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/hps/README.md)。
+
+> IMPORTANT:
+> 本节所介绍的 PaddleOCR-VL 服务与上一节中的 VLM 推理服务有所区别：后者仅负责完整流程中的一个环节（即 VLM 推理），并作为前者的底层服务被调用。
+
+### 4.1 方法一：使用 Docker Compose 部署（推荐使用）
+
+您可以分别从 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu/compose.yaml) 和 [此处](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/accelerators/nvidia-gpu/.env) 获取 Compose 文件与环境变量配置文件并下载到本地，然后在刚刚下载的文件所在目录下执行以下命令启动服务器，默认监听 **8080** 端口：
+
+```shell
+# 必须在 compose.yaml 和 .env 文件所在的目录中执行
+docker compose up
+```
+
+> 提示：
+> `compose.yaml` 中使用的镜像标签通常由 `.env` 中的 `API_IMAGE_TAG_SUFFIX` 和 `VLM_IMAGE_TAG_SUFFIX` 控制，默认使用 `latest-nvidia-gpu-offline` 等标签。如需确保拉取到最新的 `latest` 镜像，可先在当前目录执行 `docker compose pull`，再执行 `docker compose up`。
+> 如果希望使用特定版本的 PaddleOCR 镜像，可将这两个环境变量中的 `latest` 替换为对应版本号 `paddleocr<major>.<minor>`，例如 `paddleocr3.3-nvidia-gpu-offline`。
+
+启动后将看到类似如下输出：
+
+```text
+paddleocr-vl-api             | INFO:     Started server process [1]
+paddleocr-vl-api             | INFO:     Waiting for application startup.
+paddleocr-vl-api             | INFO:     Application startup complete.
+paddleocr-vl-api             | INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+```
+
+此方式基于 vLLM 等框架对 VLM 推理进行加速，更适合生产环境部署，但要求机器配备 GPU，并且NVIDIA驱动程序支持 CUDA 12.6 或以上版本。
+
+此外，使用此方式启动服务器后，除拉取镜像外，无需连接互联网。如需在离线环境中部署，可先在联网机器上拉取 Compose 文件中涉及的镜像，导出并传输至离线机器中导入，即可在离线环境下启动服务。
+
+Docker Compose 通过读取 `.env` 和 `compose.yaml` 文件中配置，先后启动 2 个容器，分别运行底层 VLM 推理服务，以及 PaddleOCR-VL 服务（产线服务）。
+
+`.env` 文件中包含的各环境变量含义如下：
+
+- `API_IMAGE_TAG_SUFFIX`：启动产线服务使用的镜像的标签后缀。默认为 `latest-nvidia-gpu-offline`，表示使用最新版本离线 GPU 镜像。如果需要使用非离线版本镜像，可以去除 `-offline` 后缀；如果希望使用特定版本的 PaddleOCR 对应的镜像，可将 `latest` 换成具体版本 `paddleocr<major>.<minor>`，例如 `paddleocr3.3-nvidia-gpu-offline`。
+- `VLM_BACKEND`：VLM 推理后端，目前支持 `vllm` 和 `fastdeploy`。默认为 `vllm`。
+- `VLM_IMAGE_TAG_SUFFIX`：启动 VLM 推理服务使用的镜像的标签后缀。默认为 `latest-nvidia-gpu-offline`，表示使用最新版本离线 GPU 镜像。如果需要使用非离线版本镜像，可以去除 `-offline` 后缀；如果希望使用特定版本的 PaddleOCR 对应的镜像，可将 `latest` 换成具体版本 `paddleocr<major>.<minor>`，例如 `paddleocr3.3-nvidia-gpu-offline`。
+
+您可以通过修改 `.env` 和 `compose.yaml` 来满足自定义需求，例如：
+
+<details>
+<summary>1. 更改 PaddleOCR-VL 服务的端口</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>paddleocr-vl-api.ports</code> 来更改端口。例如，如果您需要将服务端口更换为 8111，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    ports:
+-     - 8080:8080
++     - 8111:8080
+    ...
+```
+
+</details>
+
+<details>
+<summary>2. 指定 PaddleOCR-VL 服务所使用的 GPU</summary>
+
+编辑 <code>compose.yaml</code> 文件中的 <code>device_ids</code> 来更改所使用的 GPU。例如，如果您需要使用卡 1 进行部署，可以进行以下修改：
+
+```diff
+  paddleocr-vl-api:
+    ...
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+-             device_ids: ["0"]
++             device_ids: ["1"]
+              capabilities: [gpu]
+    ...
+  paddleocr-vlm-server:
+    ...
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+-             device_ids: ["0"]
++             device_ids: ["1"]
+              capabilities: [gpu]
+    ...
+```
+
+</details>
+
+<details>
+<summary>3. 调整 VLM 服务端配置</summary>
+
+若您想调整 VLM 服务端的配置，可以参考 <a href="#331">3.3.1 服务端参数调整</a> 生成配置文件。
+
+生成配置文件后，将以下的 <code>paddleocr-vlm-server.volumes</code> 和 <code>paddleocr-vlm-server.command</code> 字段增加到您的 <code>compose.yaml</code> 中。请将 <code>/path/to/your_config.yaml</code> 替换为您的实际配置文件路径。
+
+```yaml
+  paddleocr-vlm-server:
+    ...
+    volumes:
+      - /path/to/your_config.yaml:/home/paddleocr/vlm_server_config.yaml
+    command: paddleocr genai_server --model_name PaddleOCR-VL-1.6-0.9B --host 0.0.0.0 --port 8118 --backend vllm --backend_config /home/paddleocr/vlm_server_config.yaml
+    ...
+```
+
+</details>
+
+<details>
+<summary>4. 更改 VLM 推理后端</summary>
+
+修改 <code>.env</code> 文件中的 <code>VLM_BACKEND</code>，例如将 VLM 推理后端修改为 <code>fastdeploy</code>：
+
+```diff
+  API_IMAGE_TAG_SUFFIX=latest-nvidia-gpu-offline
+- VLM_BACKEND=vllm
++ VLM_BACKEND=fastdeploy
+  VLM_IMAGE_TAG_SUFFIX=latest-nvidia-gpu-offline
+```
+
+</details>
+
+<details>
+<summary>5. 调整产线相关配置（如模型路径、批处理大小、部署设备等）</summary>
+
+参考本文中 <a href="#44">4.4 产线配置调整说明</a> 小节。
+
+</details>
+
+### 4.2 方法二：手动部署
+
+> IMPORTANT:
+> 执行本节前，请先按第 1 节完成 PaddleOCR 本地运行环境准备。
+
+执行以下命令，通过 PaddleX CLI 安装服务化部署插件：
+
+> `paddlex` 命令会在安装 `paddleocr` 时一并安装，因此如果您已按前文完成 PaddleOCR 安装，通常无需额外安装 PaddleX。
+
+```shell
+paddlex --install serving
+```
+
+然后，使用 PaddleX CLI 启动服务器：
+
+```shell
+paddlex --serve --pipeline PaddleOCR-VL
+```
+
+如需在服务化部署中切换到 `transformers` 引擎，可参考如下示例：
+
+```shell
+paddlex --serve --pipeline PaddleOCR-VL --engine transformers
+```
+
+启动后将看到类似如下输出，服务器默认监听 **8080** 端口：
+
+```text
+INFO:     Started server process [63108]
+INFO:     Waiting for application startup.
+INFO:     Application startup complete.
+INFO:     Uvicorn running on http://0.0.0.0:8080 (Press CTRL+C to quit)
+```
+
+与服务化部署相关的命令行参数如下：
+
+
+| 名称             | 说明                                     |
+| -------------- | -------------------------------------- |
+| `--pipeline`   | PaddleX 产线注册名或产线配置文件路径。                |
+| `--device`     | 产线部署设备。默认情况下，若 GPU 可用则使用 GPU，否则使用 CPU。 |
+| `--host`       | 服务器绑定的主机名或 IP 地址，默认为 `0.0.0.0`。        |
+| `--port`       | 服务器监听的端口号，默认为 `8080`。                  |
+| `--use_hpip`   | 启用高性能推理模式。请参考高性能推理文档了解更多信息。            |
+| `--hpi_config` | 高性能推理配置。请参考高性能推理文档了解更多信息。              |
+
+
+如需调整产线相关配置（如模型路径、批处理大小、部署设备等），可参考 4.4 小节。
+
+### 4.3 客户端调用方式
+
+以下是服务化部署的 API 参考与多语言服务调用示例：
+
+<details><summary>API 参考</summary>
+<p>对于服务提供的主要操作：</p>
+<ul>
+<li>HTTP请求方法为POST。</li>
+<li>请求体和响应体均为JSON数据（JSON对象）。</li>
+<li>当请求处理成功时，响应状态码为<code>200</code>，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。固定为<code>0</code>。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。固定为<code>"Success"</code>。</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>操作结果。</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>当请求处理未成功时，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。与响应状态码相同。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。</td>
+</tr>
+</tbody>
+</table>
+<p>服务提供的主要操作如下：</p>
+<ul>
+<li><b><code>infer</code></b></li>
+</ul>
+<p>进行版面解析。</p>
+<p><code>POST /layout-parsing</code></p>
+<ul>
+<li>请求体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>服务器可访问的图像文件（含 TIFF，多页时按页处理）或 PDF 文件的 URL，或上述类型文件内容的 Base64 编码结果。
+</td>
+<td>是</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code>｜<code>null</code></td>
+<td>文件类型。<code>0</code> 表示 PDF 文件，<code>1</code> 表示图像文件（含 TIFF）。若请求体无此属性，则将根据URL推断文件类型。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_doc_orientation_classify</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>use_doc_unwarping</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useLayoutDetection</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>use_layout_detection</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useChartRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>use_chart_recognition</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useSealRecognition</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>use_seal_recognition</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useOcrForImageBlock</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>use_ocr_for_image_block</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutThreshold</code></td>
+<td><code>number</code> | <code>object</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>layout_threshold</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutNms</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>layout_nms</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutUnclipRatio</code></td>
+<td><code>number</code> | <code>array</code> | <code>object</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>layout_unclip_ratio</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutMergeBboxesMode</code></td>
+<td><code>string</code> | <code>object</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>layout_merge_bboxes_mode</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutShapeMode</code></td>
+<td><code>string</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>layout_shape_mode</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>promptLabel</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>prompt_label</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>formatBlockContent</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>format_block_content</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>repetitionPenalty</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>repetition_penalty</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>temperature</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>temperature</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>topP</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>top_p</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>minPixels</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>min_pixels</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>maxPixels</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>max_pixels</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>maxNewTokens</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>max_new_tokens</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>mergeLayoutBlocks</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>merge_layout_blocks</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>markdownIgnoreLabels</code></td>
+<td><code>array</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>markdown_ignore_labels</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>vlmExtraArgs</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>predict</code> 方法的 <code>vlm_extra_args</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>prettifyMarkdown</code></td>
+<td><code>boolean</code></td>
+<td>是否输出美化后的 Markdown 文本。默认为 <code>true</code>。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>showFormulaNumber</code></td>
+<td><code>boolean</code></td>
+<td>输出的 Markdown 文本中是否包含公式编号。默认为 <code>false</code>。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>restructurePages</code></td>
+<td><code>boolean</code></td>
+<td>是否重构多页结果。默认为 <code>false</code>。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>mergeTables</code></td>
+<td><code>boolean</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>restructure_pages</code> 方法的 <code>merge_tables</code> 参数相关说明。仅当<code>restructurePages</code>为<code>true</code>时生效。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>relevelTitles</code></td>
+<td><code>boolean</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>restructure_pages</code> 方法的 <code>relevel_titles</code> 参数相关说明。仅当<code>restructurePages</code>为<code>true</code>时生效。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>returnMarkdownImages</code></td>
+<td><code>boolean</code></td>
+<td>是否在响应中返回 Markdown 中引用的图片。默认为 <code>true</code>；设为 <code>false</code> 时 <code>markdown.images</code> 为 <code>null</code> 或不出现，且服务端跳过图片编码 / URL 上传。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>outputFormats</code></td>
+<td><code>array</code> | <code>null</code></td>
+<td>可选。需要额外返回的文档格式列表。默认不返回任何附加格式。当前仅支持 <code>"docx"</code>。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>是否返回可视化结果图以及处理过程中的中间图像等。
+<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>传入 <code>true</code>：返回图像。</li>
+<li>传入 <code>false</code>：不返回图像。</li>
+<li>若请求体中未提供该参数或传入 <code>null</code>：遵循配置文件<code>Serving.visualize</code> 的设置。</li>
+</ul>
+<br/>例如，在配置文件中添加如下字段：<br/>
+<pre><code>Serving:
+  visualize: False
+</code></pre>
+将默认不返回图像，通过请求体中的<code>visualize</code>参数可以覆盖默认行为。如果请求体和配置文件中均未设置（或请求体传入<code>null</code>、配置文件中未设置），则默认返回图像。
+</td>
+<td>否</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>请求处理成功时，响应体的<code>result</code>具有如下属性：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>layoutParsingResults</code></td>
+<td><code>array</code></td>
+<td>版面解析结果。数组长度为1（对于图像输入）或实际处理的文档页数（对于PDF输入）。对于PDF输入，数组中的每个元素依次表示PDF文件中实际处理的每一页的结果。</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>输入数据信息。</td>
+</tr>
+</tbody>
+</table>
+<p>下表中涉及图像等二进制文件的字段（如 <code>outputImages</code>、<code>inputImage</code>、<code>markdown.images</code>、<code>exports</code>）默认以 Base64 字符串内联返回；当服务端开启 URL 返回模式时，相应字段的值变为预签名 URL，字段类型保持不变。配置方式参见 <a href="../inference_deployment/serving/serving.md">服务化部署</a>「以 URL 形式返回二进制内容」一节。</p>
+<p><code>layoutParsingResults</code>中的每个元素为一个<code>object</code>，具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>对象的 <code>predict</code> 方法生成结果的 JSON 表示中 <code>res</code> 字段的简化版本，其中去除了 <code>input_path</code> 和 <code>page_index</code> 字段。</td>
+</tr>
+<tr>
+<td><code>markdown</code></td>
+<td><code>object</code></td>
+<td>Markdown结果。</td>
+</tr>
+<tr>
+<td><code>outputImages</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>参见预测结果的 <code>img</code> 属性说明。图像为 JPEG 格式，默认使用 Base64 编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+<tr>
+<td><code>inputImage</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>输入图像。图像为 JPEG 格式，默认使用 Base64 编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+<tr>
+<td><code>exports</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>可选的附加导出结果。仅当请求体中包含 <code>outputFormats</code> 且列出相应格式时出现。例如 <code>{"docx": {"content": "..."}}</code>，其中 <code>content</code> 默认为文件内容的 Base64 编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+</tbody>
+</table>
+<p><code>markdown</code>为一个<code>object</code>，具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>text</code></td>
+<td><code>string</code></td>
+<td>Markdown文本。</td>
+</tr>
+<tr>
+<td><code>images</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Markdown 图片相对路径与对应图像的键值对。图像默认为 Base64 编码字符串；启用 URL 返回模式时为预签名 URL。当请求中 <code>returnMarkdownImages</code> 为 <code>false</code> 时该字段为 <code>null</code> 或不出现。</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li><b><code>restructurePages</code></b></li>
+</ul>
+<p>重构多页结果。</p>
+<p><code>POST /restructure-pages</code></p>
+<ul>
+<li>请求体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>pages</code></td>
+<td><code>array</code></td>
+<td>页面数组。
+</td>
+<td>是</td>
+</tr>
+<tr>
+<td><code>mergeTables</code></td>
+<td><code>boolean</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>restructure_pages</code> 方法的 <code>merge_tables</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>relevelTitles</code></td>
+<td><code>boolean</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>restructure_pages</code> 方法的 <code>relevel_titles</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>concatenatePages</code></td>
+<td><code>boolean</code></td>
+<td>请参阅PaddleOCR-VL对象中 <code>restructure_pages</code> 方法的 <code>concatenate_pages</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>prettifyMarkdown</code></td>
+<td><code>boolean</code></td>
+<td>是否输出美化后的 Markdown 文本。默认为 <code>true</code>。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>showFormulaNumber</code></td>
+<td><code>boolean</code></td>
+<td>输出的 Markdown 文本中是否包含公式编号。默认为 <code>false</code>。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>returnMarkdownImages</code></td>
+<td><code>boolean</code></td>
+<td>是否在响应中返回 Markdown 中引用的图片。默认为 <code>true</code>；设为 <code>false</code> 时 <code>markdown.images</code> 为 <code>null</code> 或不出现，且服务端跳过图片编码 / URL 上传。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>outputFormats</code></td>
+<td><code>array</code> | <code>null</code></td>
+<td>可选。附加导出格式，含义与 <code>infer</code> 中的 <code>outputFormats</code> 相同。当前仅支持 <code>"docx"</code>。</td>
+<td>否</td>
+</tr>
+</tbody>
+</table>
+<p><code>pages</code>中的每个元素为一个<code>object</code>，具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>对应<code>infer</code>操作返回的<code>prunedResult</code>对象。</td>
+</tr>
+<tr>
+<td><code>markdownImages</code></td>
+<td><code>object</code>|<code>null</code></td>
+<td>对应<code>infer</code>操作返回的<code>markdown</code>对象的<code>images</code>属性。</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>请求处理成功时，响应体的<code>result</code>具有如下属性：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>layoutParsingResults</code></td>
+<td><code>array</code></td>
+<td>重构后的版面解析结果。其中每个元素包含的字段请参见对 <code>infer</code> 操作返回结果的说明（不含可视化结果图和中间图像）。</td>
+</tr>
+</tbody>
+</table>
+</details>
+<details><summary>多语言调用服务示例</summary>
+<details>
+<summary>Python</summary>
+
+<pre><code class="language-python">
+import base64
+import requests
+import pathlib
+
+BASE_URL = "http://localhost:8080"
+
+image_path = "./demo.jpg"
+
+# 对本地图像进行Base64编码
+with open(image_path, "rb") as file:
+    image_bytes = file.read()
+    image_data = base64.b64encode(image_bytes).decode("ascii")
+
+payload = {
+    "file": image_data, # Base64编码的文件内容或者文件URL
+    "fileType": 1, # 文件类型，1表示图像文件
+}
+
+response = requests.post(BASE_URL + "/layout-parsing", json=payload)
+assert response.status_code == 200, (response.status_code, response.text)
+
+result = response.json()["result"]
+pages = []
+for i, res in enumerate(result["layoutParsingResults"]):
+    pages.append({"prunedResult": res["prunedResult"], "markdownImages": res["markdown"].get("images")})
+    for img_name, img in res["outputImages"].items():
+        img_path = f"{img_name}_{i}.jpg"
+        pathlib.Path(img_path).parent.mkdir(exist_ok=True)
+        with open(img_path, "wb") as f:
+            f.write(base64.b64decode(img))
+        print(f"Output image saved at {img_path}")
+
+payload = {
+    "pages": pages,
+    "concatenatePages": True,
+}
+
+response = requests.post(BASE_URL + "/restructure-pages", json=payload)
+assert response.status_code == 200, (response.status_code, response.text)
+
+result = response.json()["result"]
+res = result["layoutParsingResults"][0]
+print(res["prunedResult"])
+md_dir = pathlib.Path("markdown")
+md_dir.mkdir(exist_ok=True)
+(md_dir / "doc.md").write_text(res["markdown"]["text"])
+for img_path, img in res["markdown"]["images"].items():
+    img_path = md_dir / img_path
+    img_path.parent.mkdir(parents=True, exist_ok=True)
+    img_path.write_bytes(base64.b64decode(img))
+print(f"Markdown document saved at {md_dir / 'doc.md'}")
+</code></pre></details>
+
+<details><summary>C++</summary>
+
+<pre><code class="language-cpp">#include &lt;iostream&gt;
+#include &lt;filesystem&gt;
+#include &lt;fstream&gt;
+#include &lt;vector&gt;
+#include &lt;string&gt;
+#include "cpp-httplib/httplib.h" // https://github.com/Huiyicc/cpp-httplib
+#include "nlohmann/json.hpp" // https://github.com/nlohmann/json
+#include "base64.hpp" // https://github.com/tobiaslocker/base64
+
+namespace fs = std::filesystem;
+
+int main() {
+    httplib::Client client("localhost", 8080);
+
+    const std::string filePath = "./demo.jpg";
+
+    std::ifstream file(filePath, std::ios::binary | std::ios::ate);
+    if (!file) {
+        std::cerr << "Error opening file: " << filePath << std::endl;
+        return 1;
+    }
+
+    std::streamsize size = file.tellg();
+    file.seekg(0, std::ios::beg);
+    std::vector<char> buffer(size);
+    if (!file.read(buffer.data(), size)) {
+        std::cerr << "Error reading file." << std::endl;
+        return 1;
+    }
+
+    std::string bufferStr(buffer.data(), static_cast<size_t>(size));
+    std::string encodedFile = base64::to_base64(bufferStr);
+
+    nlohmann::json jsonObj;
+    jsonObj["file"] = encodedFile;
+    jsonObj["fileType"] = 1;
+
+    auto response = client.Post("/layout-parsing", jsonObj.dump(), "application/json");
+
+    if (response && response->status == 200) {
+        nlohmann::json jsonResponse = nlohmann::json::parse(response->body);
+        auto result = jsonResponse["result"];
+
+        if (!result.is_object() || !result.contains("layoutParsingResults")) {
+            std::cerr << "Unexpected response format." << std::endl;
+            return 1;
+        }
+
+        const auto& results = result["layoutParsingResults"];
+        for (size_t i = 0; i < results.size(); ++i) {
+            const auto& res = results[i];
+
+            if (res.contains("prunedResult")) {
+                std::cout << "Layout result [" << i << "]: " << res["prunedResult"].dump() << std::endl;
+            }
+
+            if (res.contains("outputImages") && res["outputImages"].is_object()) {
+                for (auto& [imgName, imgBase64] : res["outputImages"].items()) {
+                    std::string outputPath = imgName + "_" + std::to_string(i) + ".jpg";
+                    fs::path pathObj(outputPath);
+                    fs::path parentDir = pathObj.parent_path();
+                    if (!parentDir.empty() && !fs::exists(parentDir)) {
+                        fs::create_directories(parentDir);
+                    }
+
+                    std::string decodedImage = base64::from_base64(imgBase64.get<std::string>());
+
+                    std::ofstream outFile(outputPath, std::ios::binary);
+                    if (outFile.is_open()) {
+                        outFile.write(decodedImage.c_str(), decodedImage.size());
+                        outFile.close();
+                        std::cout << "Saved image: " << outputPath << std::endl;
+                    } else {
+                        std::cerr << "Failed to save image: " << outputPath << std::endl;
+                    }
+                }
+            }
+        }
+    } else {
+        std::cerr << "Request failed." << std::endl;
+        if (response) {
+            std::cerr << "HTTP status: " << response->status << std::endl;
+            std::cerr << "Response body: " << response->body << std::endl;
+        }
+        return 1;
+    }
+
+    return 0;
+}
+</code></pre></details>
+
+<details><summary>Java</summary>
+
+<pre><code class="language-java">import okhttp3.*;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.util.Base64;
+import java.nio.file.Paths;
+import java.nio.file.Files;
+
+public class Main {
+    public static void main(String[] args) throws IOException {
+        String API_URL = "http://localhost:8080/layout-parsing";
+        String imagePath = "./demo.jpg";
+
+        File file = new File(imagePath);
+        byte[] fileContent = java.nio.file.Files.readAllBytes(file.toPath());
+        String base64Image = Base64.getEncoder().encodeToString(fileContent);
+
+        ObjectMapper objectMapper = new ObjectMapper();
+        ObjectNode payload = objectMapper.createObjectNode();
+        payload.put("file", base64Image);
+        payload.put("fileType", 1);
+
+        OkHttpClient client = new OkHttpClient();
+        MediaType JSON = MediaType.get("application/json; charset=utf-8");
+
+        RequestBody body = RequestBody.create(JSON, payload.toString());
+
+        Request request = new Request.Builder()
+                .url(API_URL)
+                .post(body)
+                .build();
+
+        try (Response response = client.newCall(request).execute()) {
+            if (response.isSuccessful()) {
+                String responseBody = response.body().string();
+                JsonNode root = objectMapper.readTree(responseBody);
+                JsonNode result = root.get("result");
+
+                JsonNode layoutParsingResults = result.get("layoutParsingResults");
+                for (int i = 0; i < layoutParsingResults.size(); i++) {
+                    JsonNode item = layoutParsingResults.get(i);
+                    int finalI = i;
+                    JsonNode prunedResult = item.get("prunedResult");
+                    System.out.println("Pruned Result [" + i + "]: " + prunedResult.toString());
+
+                    JsonNode outputImages = item.get("outputImages");
+                    outputImages.fieldNames().forEachRemaining(imgName -> {
+                        try {
+                            String imgBase64 = outputImages.get(imgName).asText();
+                            byte[] imgBytes = Base64.getDecoder().decode(imgBase64);
+                            String imgPath = imgName + "_" + finalI + ".jpg";
+
+                            File outputFile = new File(imgPath);
+                            File parentDir = outputFile.getParentFile();
+                            if (parentDir != null && !parentDir.exists()) {
+                                parentDir.mkdirs();
+                                System.out.println("Created directory: " + parentDir.getAbsolutePath());
+                            }
+
+                            try (FileOutputStream fos = new FileOutputStream(outputFile)) {
+                                fos.write(imgBytes);
+                                System.out.println("Saved image: " + imgPath);
+                            }
+                        } catch (IOException e) {
+                            System.err.println("Failed to save image: " + e.getMessage());
+                        }
+                    });
+                }
+            } else {
+                System.err.println("Request failed with HTTP code: " + response.code());
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Go</summary>
+
+<pre><code class="language-go">package main
+
+import (
+    "bytes"
+    "encoding/base64"
+    "encoding/json"
+    "fmt"
+    "io/ioutil"
+    "net/http"
+    "os"
+    "path/filepath"
+)
+
+func main() {
+    API_URL := "http://localhost:8080/layout-parsing"
+    filePath := "./demo.jpg"
+
+    fileBytes, err := ioutil.ReadFile(filePath)
+    if err != nil {
+        fmt.Printf("Error reading file: %v\n", err)
+        return
+    }
+    fileData := base64.StdEncoding.EncodeToString(fileBytes)
+
+    payload := map[string]interface{}{
+        "file":     fileData,
+        "fileType": 1,
+    }
+    payloadBytes, err := json.Marshal(payload)
+    if err != nil {
+        fmt.Printf("Error marshaling payload: %v\n", err)
+        return
+    }
+
+    client := &http.Client{}
+    req, err := http.NewRequest("POST", API_URL, bytes.NewBuffer(payloadBytes))
+    if err != nil {
+        fmt.Printf("Error creating request: %v\n", err)
+        return
+    }
+    req.Header.Set("Content-Type", "application/json")
+
+    res, err := client.Do(req)
+    if err != nil {
+        fmt.Printf("Error sending request: %v\n", err)
+        return
+    }
+    defer res.Body.Close()
+
+    if res.StatusCode != http.StatusOK {
+        fmt.Printf("Unexpected status code: %d\n", res.StatusCode)
+        return
+    }
+
+    body, err := ioutil.ReadAll(res.Body)
+    if err != nil {
+        fmt.Printf("Error reading response: %v\n", err)
+        return
+    }
+
+    type Markdown struct {
+        Text   string            `json:"text"`
+        Images map[string]string `json:"images"`
+    }
+
+    type LayoutResult struct {
+        PrunedResult map[string]interface{} `json:"prunedResult"`
+        Markdown     Markdown               `json:"markdown"`
+        OutputImages map[string]string      `json:"outputImages"`
+        InputImage   *string                `json:"inputImage"`
+    }
+
+    type Response struct {
+        Result struct {
+            LayoutParsingResults []LayoutResult `json:"layoutParsingResults"`
+            DataInfo             interface{}    `json:"dataInfo"`
+        } `json:"result"`
+    }
+
+    var respData Response
+    if err := json.Unmarshal(body, &respData); err != nil {
+        fmt.Printf("Error parsing response: %v\n", err)
+        return
+    }
+
+    for i, res := range respData.Result.LayoutParsingResults {
+        fmt.Printf("Result %d - prunedResult: %+v\n", i, res.PrunedResult)
+
+        mdDir := fmt.Sprintf("markdown_%d", i)
+        os.MkdirAll(mdDir, 0755)
+        mdFile := filepath.Join(mdDir, "doc.md")
+        if err := os.WriteFile(mdFile, []byte(res.Markdown.Text), 0644); err != nil {
+            fmt.Printf("Error writing markdown file: %v\n", err)
+        } else {
+            fmt.Printf("Markdown document saved at %s\n", mdFile)
+        }
+
+        for path, imgBase64 := range res.Markdown.Images {
+            fullPath := filepath.Join(mdDir, path)
+            if err := os.MkdirAll(filepath.Dir(fullPath), 0755); err != nil {
+                fmt.Printf("Error creating directory for markdown image: %v\n", err)
+                continue
+            }
+            imgBytes, err := base64.StdEncoding.DecodeString(imgBase64)
+            if err != nil {
+                fmt.Printf("Error decoding markdown image: %v\n", err)
+                continue
+            }
+            if err := os.WriteFile(fullPath, imgBytes, 0644); err != nil {
+                fmt.Printf("Error saving markdown image: %v\n", err)
+            }
+        }
+
+        for name, imgBase64 := range res.OutputImages {
+            imgBytes, err := base64.StdEncoding.DecodeString(imgBase64)
+            if err != nil {
+                fmt.Printf("Error decoding output image %s: %v\n", name, err)
+                continue
+            }
+            filename := fmt.Sprintf("%s_%d.jpg", name, i)
+
+            if err := os.MkdirAll(filepath.Dir(filename), 0755); err != nil {
+                fmt.Printf("Error creating directory for output image: %v\n", err)
+                continue
+            }
+
+            if err := os.WriteFile(filename, imgBytes, 0644); err != nil {
+                fmt.Printf("Error saving output image %s: %v\n", filename, err)
+            } else {
+                fmt.Printf("Output image saved at %s\n", filename)
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>C#</summary>
+
+<pre><code class="language-csharp">using System;
+using System.IO;
+using System.Net.Http;
+using System.Text;
+using System.Threading.Tasks;
+using Newtonsoft.Json.Linq;
+
+class Program
+{
+    static readonly string API_URL = "http://localhost:8080/layout-parsing";
+    static readonly string inputFilePath = "./demo.jpg";
+
+    static async Task Main(string[] args)
+    {
+        var httpClient = new HttpClient();
+
+        byte[] fileBytes = File.ReadAllBytes(inputFilePath);
+        string fileData = Convert.ToBase64String(fileBytes);
+
+        var payload = new JObject
+        {
+            { "file", fileData },
+            { "fileType", 1 }
+        };
+        var content = new StringContent(payload.ToString(), Encoding.UTF8, "application/json");
+
+        HttpResponseMessage response = await httpClient.PostAsync(API_URL, content);
+        response.EnsureSuccessStatusCode();
+
+        string responseBody = await response.Content.ReadAsStringAsync();
+        JObject jsonResponse = JObject.Parse(responseBody);
+
+        JArray layoutParsingResults = (JArray)jsonResponse["result"]["layoutParsingResults"];
+        for (int i = 0; i < layoutParsingResults.Count; i++)
+        {
+            var res = layoutParsingResults[i];
+            Console.WriteLine($"[{i}] prunedResult:\n{res["prunedResult"]}");
+
+            JObject outputImages = res["outputImages"] as JObject;
+            if (outputImages != null)
+            {
+                foreach (var img in outputImages)
+                {
+                    string imgName = img.Key;
+                    string base64Img = img.Value?.ToString();
+                    if (!string.IsNullOrEmpty(base64Img))
+                    {
+                        string imgPath = $"{imgName}_{i}.jpg";
+                        byte[] imageBytes = Convert.FromBase64String(base64Img);
+
+                        string directory = Path.GetDirectoryName(imgPath);
+                        if (!string.IsNullOrEmpty(directory) && !Directory.Exists(directory))
+                        {
+                            Directory.CreateDirectory(directory);
+                            Console.WriteLine($"Created directory: {directory}");
+                        }
+
+                        File.WriteAllBytes(imgPath, imageBytes);
+                        Console.WriteLine($"Output image saved at {imgPath}");
+                    }
+                }
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Node.js</summary>
+
+<pre><code class="language-js">const axios = require('axios');
+const fs = require('fs');
+const path = require('path');
+
+const API_URL = 'http://localhost:8080/layout-parsing';
+const imagePath = './demo.jpg';
+const fileType = 1;
+
+function encodeImageToBase64(filePath) {
+  const bitmap = fs.readFileSync(filePath);
+  return Buffer.from(bitmap).toString('base64');
+}
+
+const payload = {
+  file: encodeImageToBase64(imagePath),
+  fileType: fileType
+};
+
+axios.post(API_URL, payload)
+  .then(response => {
+    const results = response.data.result.layoutParsingResults;
+    results.forEach((res, index) => {
+      console.log(`\n[${index}] prunedResult:`);
+      console.log(res.prunedResult);
+
+      const outputImages = res.outputImages;
+      if (outputImages) {
+        Object.entries(outputImages).forEach(([imgName, base64Img]) => {
+          const imgPath = `${imgName}_${index}.jpg`;
+
+          const directory = path.dirname(imgPath);
+          if (!fs.existsSync(directory)) {
+            fs.mkdirSync(directory, { recursive: true });
+            console.log(`Created directory: ${directory}`);
+          }
+
+          fs.writeFileSync(imgPath, Buffer.from(base64Img, 'base64'));
+          console.log(`Output image saved at ${imgPath}`);
+        });
+      } else {
+        console.log(`[${index}] No outputImages.`);
+      }
+    });
+  })
+  .catch(error => {
+    console.error('Error during API request:', error.message || error);
+  });
+</code></pre></details>
+
+<details><summary>PHP</summary>
+
+<pre><code class="language-php">&lt;?php
+
+$API_URL = "http://localhost:8080/layout-parsing";
+$image_path = "./demo.jpg";
+
+$image_data = base64_encode(file_get_contents($image_path));
+$payload = array("file" => $image_data, "fileType" => 1);
+
+$ch = curl_init($API_URL);
+curl_setopt($ch, CURLOPT_POST, true);
+curl_setopt($ch, CURLOPT_POSTFIELDS, json_encode($payload));
+curl_setopt($ch, CURLOPT_HTTPHEADER, array('Content-Type: application/json'));
+curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+$response = curl_exec($ch);
+curl_close($ch);
+
+$result = json_decode($response, true)["result"]["layoutParsingResults"];
+
+foreach ($result as $i => $item) {
+    echo "[$i] prunedResult:\n";
+    print_r($item["prunedResult"]);
+
+    if (!empty($item["outputImages"])) {
+        foreach ($item["outputImages"] as $img_name => $img_base64) {
+            $output_image_path = "{$img_name}_{$i}.jpg";
+
+            $directory = dirname($output_image_path);
+            if (!is_dir($directory)) {
+                mkdir($directory, 0777, true);
+                echo "Created directory: $directory\n";
+            }
+
+            file_put_contents($output_image_path, base64_decode($img_base64));
+            echo "Output image saved at $output_image_path\n";
+        }
+    } else {
+        echo "No outputImages found for item $i\n";
+    }
+}
+?&gt;
+</code></pre></details>
+</details>
+
+### 4.4 产线配置调整说明
+
+> NOTE:
+> 若您无需调整产线配置，可忽略此小节。
+
+调整服务化部署的 PaddleOCR-VL 配置只需以下三步：
+
+1. 获取配置文件
+2. 修改配置文件
+3. 应用配置文件
+
+#### 4.4.1 获取配置文件
+
+**若您使用 Docker Compose 部署：**
+
+根据使用的后端，下载对应的产线配置文件：
+
+- vLLM：[pipeline_config_vllm.yaml](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/pipeline_config_vllm.yaml)
+- FastDeploy：[pipeline_config_fastdeploy.yaml](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/paddleocr_vl_docker/pipeline_config_fastdeploy.yaml)
+
+**若您是手动部署：**
+
+执行以下命令生成产线配置文件：
+
+```shell
+paddlex --get_pipeline_config PaddleOCR-VL
+```
+
+#### 4.4.2 修改配置文件
+
+**使用加速框架提升 VLM 推理性能**
+
+如需使用 vLLM 等加速框架提升 VLM 推理性能（第 3 节详细介绍如何启动 VLM 推理服务），可在产线配置文件中修改 `VLRecognition.genai_config.backend` 和 `VLRecognition.genai_config.server_url` 字段，例如：
+
+```yaml
+VLRecognition:
+  ...
+  genai_config:
+    backend: vllm-server
+    server_url: http://localhost:8118/v1
+```
+
+Docker Compose 方案默认已使用加速框架。
+
+**启用文档图像预处理功能**
+
+默认配置启动的服务不支持文档预处理功能。若客户端调用该功能，将返回错误信息。如需启用文档预处理，请在产线配置文件中将 `use_doc_preprocessor` 设置为 `True`，并使用修改后的配置文件启动服务。
+
+**禁用结果可视化功能**
+
+服务默认返回可视化结果，这会引入额外开销。如需禁用该功能，可在产线配置文件中添加如下配置（`Serving` 为顶层字段）：
+
+```yaml
+Serving:
+  visualize: False
+```
+
+此外，也可在请求体中设置 `visualize` 字段为 `false`，以针对单次请求禁用可视化。
+
+**配置以 URL 形式返回二进制内容**
+
+服务默认以 Base64 编码内联返回响应中的图像等二进制内容。如需改为以 URL 形式返回，可在产线配置文件中添加如下配置（`Serving` 为顶层字段）：
+
+```yaml
+Serving:
+  return_urls: True
+  extra:
+    file_storage:
+      type: bos
+      endpoint: https://bj.bcebos.com
+      bucket_name: some-bucket
+      ak: xxx
+      sk: xxx
+      key_prefix: deploy
+    url_expires_in: 3600
+```
+
+目前支持将生成的文件存储至百度智能云对象存储（BOS）并返回 URL。相关参数说明如下：
+
+- `endpoint`：访问域名，必须配置。
+- `ak`：百度智能云 AK，必须配置。
+- `sk`：百度智能云 SK，必须配置。
+- `bucket_name`：存储空间名称，必须配置。
+- `key_prefix`：Object key 的统一前缀。
+- `connection_timeout_in_mills`：请求超时时间（单位：毫秒）。
+- `url_expires_in`：URL 有效期（单位：秒）。`-1` 表示永不过期。
+
+有关 AK/SK 获取等更多信息，请参考 [百度智能云官方文档](https://cloud.baidu.com/doc/BOS/index.html)。
+
+**限制 PDF 与多页 TIFF 解析页数**
+
+服务默认处理完整的 PDF 文件；多页 TIFF（`fileType=1`）会按页展开后逐页处理。在实际生产环境中，若 PDF 或多页 TIFF 页数过多，可能会影响系统稳定性，导致处理超时或资源占用过高。为保障服务的稳定运行，建议根据实际情况合理设置页数上限。可在产线配置文件中添加如下配置（`Serving` 为顶层字段）：
+
+```yaml
+Serving:
+  extra:
+    max_num_input_imgs: <页数限制，例如 100>
+```
+
+`max_num_input_imgs` 同时限制 PDF 与多页 TIFF 的最大处理页数。将其设置为 `null` 时，不对上述页数进行限制。
+
+#### 4.4.3 应用配置文件
+
+**若您使用 Docker Compose 部署：**
+
+设置 Compose 文件中的 `services.paddleocr-vl-api.volumes` 字段，将产线配置文件挂载到 `/home/paddleocr` 目录。例如：
+
+```yaml
+services:
+  paddleocr-vl-api:
+    ...
+    volumes:
+      - ./pipeline_config_vllm.yaml:/home/paddleocr/pipeline_config_vllm.yaml
+...
+```
+
+> 在生产环境中，您也可以自行构建镜像，将配置文件打包到镜像中。
+
+**若您是手动部署：**
+
+在启动服务时，将 `--pipeline` 参数指定为自定义配置文件路径。
+
+## 5. 模型微调
+
+若您发现 PaddleOCR-VL 在特定业务场景中的精度表现未达预期，我们推荐使用 [ERNIEKit 套件](https://github.com/PaddlePaddle/ERNIE/tree/release/v1.4) 对视觉语言模型（例如 PaddleOCR-VL-0.9B）进行有监督微调（SFT）。具体操作步骤可参考 [ERNIEKit 官方文档](https://github.com/PaddlePaddle/ERNIE/blob/release/v1.4/docs/paddleocr_vl_sft_zh.md)。
+
+> 目前暂不支持对版面分析排序模型进行微调。
diff --git a/docs/version3.x/pipeline_usage/doc_preprocessor.en.md b/docs/version3.x/pipeline_usage/doc_preprocessor.en.md
new file mode 100644
index 0000000..79e5ebd
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/doc_preprocessor.en.md
@@ -0,0 +1,1339 @@
+---
+comments: true
+---
+
+# Document Image Preprocessing Pipeline Tutorial
+
+## 1. Introduction to Document Image Preprocessing Pipeline
+
+The Document Image Preprocessing Pipeline integrates two key functions: document orientation classification and geometric distortion correction. The document orientation classification module automatically identifies the four possible orientations of a document (0°, 90°, 180°, 270°), ensuring that the document is processed in the correct direction. The text image unwarping model is designed to correct geometric distortions that occur during document photography or scanning, restoring the document's original shape and proportions. This pipeline is suitable for digital document management, preprocessing tasks for OCR, and any scenario requiring improved document image quality. By automating orientation correction and geometric distortion correction, this module significantly enhances the accuracy and efficiency of document processing, providing a more reliable foundation for image analysis. The pipeline also offers flexible service-oriented deployment options, supporting calls from various programming languages on multiple hardware platforms. Additionally, the pipeline supports secondary development, allowing you to fine-tune the models on your own datasets and seamlessly integrate the trained models.
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/doc_preprocessor/02.jpg">
+
+<b>The General Document Image Preprocessing Pipeline includes the following two modules. Each module can be trained and inferred independently and contains multiple models. For detailed information, please click on the corresponding module to view the documentation.</b>
+
+- [Document Image Orientation Classification Module](../module_usage/doc_img_orientation_classification.md) (Optional)
+- [Text Image Unwarping Module](../module_usage/text_image_unwarping.md) (Optional)
+
+In this pipeline, you can select the models to use based on the benchmark data provided below.
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing.
+> In the inference time columns labeled [Normal Mode / High-Performance Mode], the Normal Mode values correspond to the local `paddle_static` inference engine.
+
+<details>
+<summary> <b>Document Image Orientation Classification Module (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Links</th>
+<th>Top-1 Acc (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">Training Model</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>A document image classification model based on PP-LCNet_x1_0, which includes four categories: 0°, 90°, 180°, and 270°.</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary> <b>Text Image Unwarping Module (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>CER</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">Training Model</a></td>
+<td>0.179</td>
+<td>19.05 / 19.05</td>
+<td>- / 869.82</td>
+<td>30.3</td>
+<td>A high-precision text image unwarping model.</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary> <b>Test Environment Description:</b></summary>
+
+  <ul>
+      <li><b>Performance Test Environment</b>
+          <ul>
+                      <li><strong>Test Datasets:
+             </strong>
+                <ul>
+                  <li>Document Image Orientation Classification Model: A self-built dataset by PaddleX, covering various scenarios including ID cards and documents, containing 1000 images.</li>
+                  <li>Text Image Unwarping Model: <a href="https://www3.cs.stonybrook.edu/~cvl/docunet.html">DocUNet.</a></li>
+                </ul>
+             </li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>Inference Mode Description</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration</th>
+            <th>CPU Configuration</th>
+            <th>Acceleration Technology Combination</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Normal Mode</td>
+            <td>FP32 Precision / No TRT Acceleration</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td><code>paddle_static</code></td>
+        </tr>
+        <tr>
+            <td>High-Performance Mode</td>
+            <td>Optimal combination of precision type and acceleration strategy selected in advance</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Optimal backend (Paddle/OpenVINO/TRT, etc.) selected in advance</td>
+        </tr>
+    </tbody>
+</table>
+</details>
+
+## 2. Quick Start
+
+Before using the General Document Image Preprocessing Pipeline locally, ensure that you have completed the wheel package installation according to the [Installation Guide](../installation.en.md). After installation, you can experience it via the command line or integrate it into Python locally.
+
+Please note: If you encounter issues such as the program becoming unresponsive, unexpected program termination, running out of memory resources, or extremely slow inference during execution, please try adjusting the configuration according to the documentation, such as disabling unnecessary features or using lighter-weight models.
+
+### 2.1 Command Line Experience
+
+You can quickly experience the `doc_preprocessor` pipeline with a single command:
+
+```bash
+paddleocr doc_preprocessor -i https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/doc_test_rotated.jpg
+
+# Specify whether to use the document orientation classification model via --use_doc_orientation_classify
+paddleocr doc_preprocessor -i ./doc_test_rotated.jpg --use_doc_orientation_classify True
+
+# Specify whether to use the text image unwarping module via --use_doc_unwarping
+paddleocr doc_preprocessor -i ./doc_test_rotated.jpg --use_doc_unwarping True
+
+# Specify the use of GPU for model inference via --device
+paddleocr doc_preprocessor -i ./doc_test_rotated.jpg --device gpu
+```
+
+The examples above use the local `paddle_static` inference engine by default. To run them, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr doc_preprocessor -i https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/doc_test_rotated.jpg \
+    --engine transformers
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+<details><summary><b>The command line supports more parameter settings. Click to expand for detailed explanations of command line parameters.</b></summary>
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Parameter Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>The data to be predicted. This parameter is required.<br/>
+<b>Description:</b>
+For example, the local path of an image file or PDF file: <code>/root/data/img.jpg</code>; <b>or a URL link</b>, such as the network URL of an image file or PDF file: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/doc_test_rotated.jpg">example</a>; <b>or a local directory</b>, which should contain the images to be predicted, such as the local path: <code>/root/data/</code> (currently does not support prediction of PDF files in directories; PDF files need to be specified to a specific file path).
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td><b>Meaning:</b>Specify the path to save the inference result file. <br/>
+<b>Description:</b>
+If not set, the inference result will not be saved locally.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>The name of the document orientation classification model.<br/>
+<b>Description:</b>
+If not set, the pipeline's default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the document orientation classification model. <br/>
+<b>Description:</b>
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>Meaning:</b>The name of the text image unwarping model. <br/>
+<b>Description:</b>
+If not set, the pipeline's default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the text image unwarping model. <br/>
+<b>Description:</b>
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to load and use  the document orientation classification module. <br/>
+<b>Description:</b>
+If not set, the parameter value initialized by the pipeline will be used, which defaults to <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to load and use  the text image unwarping module. <br/>
+<b>Description:</b>
+If not set, the parameter value initialized by the pipeline will be used, which defaults to <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>The device used for inference. <br/>
+<b>Description:</b>
+Support for specifying specific card numbers:
+<ul>
+<li><b>CPU</b>: For example, <code>cpu</code> indicates using the CPU for inference;</li>
+<li><b>GPU</b>: For example, <code>gpu:0</code> indicates using the first GPU for inference;</li>
+<li><b>NPU</b>: For example, <code>npu:0</code> indicates using the first NPU for inference;</li>
+<li><b>XPU</b>: For example, <code>xpu:0</code> indicates using the first XPU for inference;</li>
+<li><b>MLU</b>: For example, <code>mlu:0</code> indicates using the first MLU for inference;</li>
+<li><b>DCU</b>: For example, <code>dcu:0</code> indicates using the first DCU for inference;</li>
+<li><b>MetaX GPU</b>: For example, <code>metax_gpu:0</code> indicates using the first MetaX GPU for inference;</li>
+<li><b>Iluvatar GPU</b>: For example, <code>iluvatar_gpu:0</code> indicates using the first Iluvatar GPU for inference;</li>
+</ul>
+If not set, the pipeline initialized value for this parameter will be used. During initialization, the local GPU device 0 will be preferred; if unavailable, the CPU device will be used.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to enable the TensorRT subgraph engine of Paddle Inference.<br/>
+<b>Description:</b>
+If the model does not support TensorRT acceleration, acceleration will not be used even if this flag is set.<br/>
+For CUDA 11.8 versions of PaddlePaddle, the compatible TensorRT version is 8.x (x>=6). TensorRT 8.6.1.6 is recommended.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b> Computation precision, such as <code>fp32</code> or <code>fp16</code>.</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b> Whether to enable MKL-DNN accelerated inference.<br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support MKL-DNN acceleration, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b> MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads used for inference on CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b> Path to the PaddleX pipeline configuration file.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+
+</tbody>
+</table>
+</details>
+<br />
+
+The running results will be printed to the terminal. The running results of the <code>doc_preprocessor</code> pipeline with default configuration are as follows:
+```bash
+{'res': {'input_path': '/root/.paddlex/predict_input/doc_test_rotated.jpg', 'page_index': None, 'model_settings': {'use_doc_orientation_classify': True, 'use_doc_unwarping': True}, 'angle': 180}}
+```
+
+The visualization results are saved under the <code>save_path</code>. The visualization results are as follows:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/doc_preprocessor/02.jpg"/>
+
+
+### 2.2 Integration via Python Script
+
+The command-line approach is for quick experience and viewing results. Generally, in projects, integration through code is often required. You can achieve rapid inference in pipelines with just a few lines of code. The inference code is as follows:
+
+```python
+from paddleocr import DocPreprocessor
+
+pipeline = DocPreprocessor()
+# docpp = DocPreprocessor(use_doc_orientation_classify=True) # Specify whether to use the document orientation classification model via use_doc_orientation_classify
+# docpp = DocPreprocessor(use_doc_unwarping=True) # Specify whether to use the text image unwarping module via use_doc_unwarping
+# docpp = DocPreprocessor(device="gpu") # Specify whether to use GPU for model inference via device
+output = pipeline.predict("./doc_test_rotated.jpg")
+for res in output:
+    res.print()  ## Print the structured output of the prediction
+    res.save_to_img("./output/")
+    res.save_to_json("./output/")
+```
+
+The example above uses the local `paddle_static` inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following code:
+
+```python
+from paddleocr import DocPreprocessor
+
+pipeline = DocPreprocessor(
+    engine="transformers",
+)
+# docpp = DocPreprocessor(use_doc_orientation_classify=True) # Specify whether to use the document orientation classification model via use_doc_orientation_classify
+# docpp = DocPreprocessor(use_doc_unwarping=True) # Specify whether to use the text image unwarping module via use_doc_unwarping
+# docpp = DocPreprocessor(device="gpu") # Specify whether to use GPU for model inference via device
+output = pipeline.predict("./doc_test_rotated.jpg")
+for res in output:
+    res.print()  ## Print the structured output of the prediction
+    res.save_to_img("./output/")
+    res.save_to_json("./output/")
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+In the above Python script, the following steps are executed:
+
+(1) Instantiate the <code>doc_preprocessor</code> pipeline object via <code>DocPreprocessor()</code>. The specific parameter descriptions are as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Parameter Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>The name of the document orientation classification model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the document orientation classification model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>Meaning:</b>The name of the text image unwarping model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the pipeline's default model will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the text image unwarping model. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to load and use the document orientation classification module. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the parameter value initialized by the pipeline will be used, which defaults to <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to load and use the text image unwarping module. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the parameter value initialized by the pipeline will be used, which defaults to <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>The device used for inference. <br/>
+<b>Description:</b>
+Support for specifying specific card numbers:
+<ul>
+<li><b>CPU</b>: For example, <code>cpu</code> indicates using the CPU for inference;</li>
+<li><b>GPU</b>: For example, <code>gpu:0</code> indicates using the first GPU for inference;</li>
+<li><b>NPU</b>: For example, <code>npu:0</code> indicates using the first NPU for inference;</li>
+<li><b>XPU</b>: For example, <code>xpu:0</code> indicates using the first XPU for inference;</li>
+<li><b>MLU</b>: For example, <code>mlu:0</code> indicates using the first MLU for inference;</li>
+<li><b>DCU</b>: For example, <code>dcu:0</code> indicates using the first DCU for inference;</li>
+<li><b>MetaX GPU</b>: For example, <code>metax_gpu:0</code> indicates using the first MetaX GPU for inference;</li>
+<li><b>Iluvatar GPU</b>: For example, <code>iluvatar_gpu:0</code> indicates using the first Iluvatar GPU for inference;</li>
+<li><b>None</b>: If set to <code>None</code>,  the pipeline initialized value for this parameter will be used. During initialization, the local GPU device 0 will be preferred; if unavailable, the CPU device will be used.</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to enable the TensorRT subgraph engine of Paddle Inference.<br/>
+<b>Description:</b>
+If the model does not support TensorRT acceleration, acceleration will not be used even if this flag is set.<br/>
+For CUDA 11.8 versions of PaddlePaddle, the compatible TensorRT version is 8.x (x>=6). TensorRT 8.6.1.6 is recommended.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b> Computation precision, such as <code>"fp32"</code> or <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b> Whether to enable MKL-DNN accelerated inference.<br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support MKL-DNN acceleration, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b> MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads used for inference on CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b> Path to the PaddleX pipeline configuration file.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+</tbody>
+</table>
+
+(2) Call the <code>predict()</code> method of the <code>doc_preprocessor</code> pipeline object for inference prediction. This method will return a list of results.
+
+In addition, the pipeline also provides the <code>predict_iter()</code> method. The two methods are completely consistent in terms of parameter acceptance and result return. The difference is that <code>predict_iter()</code> returns a <code>generator</code>, which can process and obtain prediction results step by step, suitable for scenarios with large datasets or where memory savings are desired. You can choose either of the two methods according to your actual needs.
+
+The following are the parameters and their descriptions of the <code>predict()</code> method:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Parameter Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>The data to be predicted, supporting multiple input types. This parameter is required.<br/>
+<b>Description:</b>
+<ul>
+<li><b>Python Var</b>: For example, image data represented as <code>numpy.ndarray</code>;</li>
+<li><b>str</b>: For example, the local path of an image file or PDF file: <code>/root/data/img.jpg</code>; <b>or a URL link</b>, such as the network URL of an image file or PDF file: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/doc_test_rotated.jpg">example</a>; <b>or a local directory</b>, which should contain the images to be predicted, such as the local path: <code>/root/data/</code> (currently does not support prediction of PDF files in directories; PDF files need to be specified to a specific file path);</li>
+<li><b>list</b>: The list elements should be of the above types, such as <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"]</code>.</li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to use the document orientation classification module during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to use the text image unwarping module during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+</table>
+
+(3) Process the prediction results. The prediction result for each sample is a corresponding Result object, which supports operations such as printing, saving as an image, and saving as a <code>json</code> file:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Description</th>
+<th>Parameter</th>
+<th>Parameter Type</th>
+<th>Description</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">Print the result to the terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output content using <code>JSON</code> indentation</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data for better readability. Only valid when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non-<code>ASCII</code> characters to <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> retains the original characters. Only valid when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">Save the result as a JSON file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The file path for saving. When it is a directory, the saved file name will be consistent with the input file type name.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data for better readability. Only valid when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non-<code>ASCII</code> characters to <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> retains the original characters. Only valid when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save the result as an image file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The file path for saving. Supports directory or file paths.</td>
+<td>None</td>
+</tr>
+</table>Here's the continuation of the translation:
+
+<ul>
+<li>Calling the<code>print()</code> method will output the results to the terminal. The content printed to the terminal is explained as follows:</li>
+    <ol start="1" type="1">
+     <li><code>input_path</code>: <code>(str)</code> The input path of the image to be predicted
+</li>
+     <li><code>page_index</code>: <code>(Union[int, None])</code> If the input is a PDF file, it indicates the current page number of the PDF; otherwise, it is  <code>None</code></li>  
+     <li><code>model_settings</code>: <code>(Dict[str, bool])</code> Model parameters configured for the pipeline
+</li>
+        <ol>
+        <li><code>use_doc_orientation_classify</code>: <code>(bool)</code> Controls whether to enable the document orientation classification module</li>
+        <li><code>use_doc_unwarping</code>: <code>(bool)</code> Controls whether to enable the text image rectification module</li>
+        </ol>
+    <li><code>angle</code>: <code>(int)</code> The prediction result of the document orientation classification. When enabled, the value is one of [0,90,180,270]; when disabled, it is -1</li>
+    </ol>
+<li> Calling the<code>save_to_json()</code> method will save the above content to the specified<code>save_path</code>. If a directory is specified, the saved path will be <code>save_path/{your_img_basename}.json</code>，If a file is specified, it will be saved directly to that file. Since JSON files do not support saving numpy arrays,<code>numpy.array</code>types will be converted to list form.
+</li>
+<li>Calling the<code>save_to_img()</code> method will save the visualization results to the specified<code>save_path</code>. If a directory is specified, the saved path will be <code>save_path/{your_img_basename}_doc_preprocessor_res_img.{your_img_extension}</code>. If a file is specified, it will be saved directly to that file. (Production lines usually contain many result images, so it is not recommended to specify a specific file path directly, as multiple images will be overwritten, and only the last image will be retained)</li>
+
+
+* In addition, it also supports obtaining visualization images and prediction results with results through attributes, as follows:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Attribute Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">Obtain the prediction result in JSON format</td>
+</tr>
+<tr>
+<td rowspan="2"><code>img</code></td>
+<td rowspan="2">Obtain visualization images in dictionary format</td>
+</tr>
+</table>
+<ul>
+<li>The prediction result obtained by the <code>json</code> attribute is data of type dict, and the content is consistent with that saved by calling the <code>save_to_json()</code> method.</li>
+<li>The prediction result returned by the <code>img</code> attribute is a dictionary-type data. The key is <code>preprocessed_img</code>, and the corresponding value is an <code>Image.Image</code> object: a visualization image for displaying the doc_preprocessor result.</li>
+</ul>
+
+## 3. Development Integration/Deployment
+
+If the pipeline meets your requirements for inference speed and accuracy, you can proceed directly to development integration/deployment.
+
+If you need to apply the pipeline directly to your Python project, you can refer to the example code in [2.2 Python Script Integration](#22-integration-via-python-script).
+
+In addition, PaddleOCR also provides two other deployment methods, which are detailed as follows:
+
+🚀 High-performance inference: In actual production environments, many applications have strict performance requirements (especially response speed) to ensure efficient system operation and smooth user experience. To this end, PaddleOCR provides high-performance inference functionality, aiming to deeply optimize model inference and pre/post-processing to achieve significant end-to-end process acceleration. For detailed high-performance inference procedures, please refer to the [High-Performance Inference Guide](../inference_deployment/local_inference/high_performance_inference.md).
+
+☁️ Service-oriented deployment: Service-oriented deployment is a common form of deployment in actual production environments. By encapsulating inference functions as services, clients can access these services through network requests to obtain inference results. For detailed pipeline service-oriented deployment procedures, please refer to the [Service-Oriented Deployment Guide](../inference_deployment/serving/serving.md).
+
+Below are the API references for basic service-oriented deployment and examples of multi-language service calls:
+
+<details><summary>API Reference</summary>
+<p>Main operations provided by the service:</p>
+<ul>
+<li>The HTTP request method is POST.</li>
+<li>The request body and response body are both JSON data (JSON objects).</li>
+<li>When the request is processed successfully, the response status code is <code>200</code>, and the properties of the response body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>The UUID of the request.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Fixed to <code>0</code>.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error description. Fixed to <code>"Success"</code>.</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>Operation result.</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is not processed successfully, the properties of the response body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>The UUID of the request.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Same as the response status code.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error description.</td>
+</tr>
+</tbody>
+</table>
+<p>Main operations provided by the service:</p>
+<ul>
+<li><b><code>infer</code></b></li>
+</ul>
+<p>Obtain the preprocessing result of the image document image.</p>
+<p><code>POST /document-preprocessing</code></p>
+<ul>
+<li>Properties of the request body:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Description</th>
+<th>Required</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>The URL of image files (including TIFF; multi-page TIFF is processed page by page) or PDF file accessible to the server, or the Base64 encoding result of the content of the above types of files. By default, for PDF or multi-page TIFF files with more than 10 pages, only the first 10 pages will be processed.<br /> To remove the page limit, please add the following configuration to the pipeline configuration file:
+<pre><code>Serving:
+  extra:
+    max_num_input_imgs: null
+</code></pre>
+</td>
+<td>Yes</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>File type. <code>0</code> indicates a PDF file, and <code>1</code> indicates an image file. If this property is not present in the request body, the file type will be inferred based on the URL.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_doc_orientation_classify</code> parameter in the <code>predict</code> method of the pipeline object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_doc_unwarping</code> parameter in the <code>predict</code> method of the pipeline object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>
+Whether to return the final visualization image and intermediate images during the processing.<br/>
+<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>If <code>true</code> is provided: return images.</li>
+<li>If <code>false</code> is provided: do not return any images.</li>
+<li>If this parameter is omitted from the request body, or if <code>null</code> is explicitly passed, the behavior will follow the value of <code>Serving.visualize</code> in the pipeline configuration.</li>
+</ul>
+<br/>
+For example, adding the following setting to the pipeline config file:<br/>
+<pre><code>Serving:
+  visualize: False
+</code></pre>
+will disable image return by default. This behavior can be overridden by explicitly setting the <code>visualize</code> parameter in the request.<br/>
+If neither the request body nor the configuration file is set (If <code>visualize</code> is set to <code>null</code> in the request and  not defined in the configuration file), the image is returned by default.
+</td>
+<td>No</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is processed successfully, the <code>result</code> in the response body has the following properties:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>docPreprocessingResults</code></td>
+<td><code>object</code></td>
+<td>Document image preprocessing results. The array length is 1 (for image input) or the actual number of processed document pages (for PDF input). For PDF input, each element in the array represents the result of each page actually processed in the PDF file.</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>Input data information.</td>
+</tr>
+</tbody>
+</table>
+<p>Each element in <code>docPreprocessingResults</code> is an <code>object</code> with the following properties:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>outputImage</code></td>
+<td><code>string</code></td>
+<td>The preprocessed image. The image is in PNG format and uses Base64 encoding.</td>
+</tr>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>A simplified version of the <code>res</code> field in the JSON representation of the result generated by the <code>predict</code> method of the pipeline object, with the <code>input_path</code> and <code>page_index</code> fields removed.</td>
+</tr>
+<tr>
+<td><code>docPreprocessingImage</code></td>
+<td><code>string</code> ｜ <code>null</code></td>
+<td>Visualization result image. The image is in JPEG format and uses Base64 encoding.</td>
+</tr>
+<tr>
+<td><code>inputImage</code></td>
+<td><code>string</code> ｜ <code>null</code></td>
+<td>Input image. The image is in JPEG format and uses Base64 encoding.</td>
+</tr>
+</tbody>
+</table>
+</details>
+<details><summary>Multi-language Service Call Examples</summary>
+<details>
+<summary>Python</summary>
+
+<pre><code class="language-python">import base64
+import requests
+
+API_URL = "http://localhost:8080/document-preprocessing"
+file_path = "./demo.jpg"
+
+with open(file_path, "rb") as file:
+    file_bytes = file.read()
+    file_data = base64.b64encode(file_bytes).decode("ascii")
+
+payload = {"file": file_data, "fileType": 1}
+
+response = requests.post(API_URL, json=payload)
+
+assert response.status_code == 200
+result = response.json()["result"]
+for i, res in enumerate(result["docPreprocessingResults"]):
+    print(res["prunedResult"])
+    output_img_path = f"out_{i}.png"
+    with open(output_img_path, "wb") as f:
+        f.write(base64.b64decode(res["outputImage"]))
+    print(f"Output image saved at {output_img_path}")
+</code></pre></details>
+
+<details><summary>C++</summary>
+
+<pre><code class="language-cpp">#include &lt;iostream&gt;
+#include &lt;fstream&gt;
+#include &lt;vector&gt;
+#include &lt;string&gt;
+#include "cpp-httplib/httplib.h" // https://github.com/Huiyicc/cpp-httplib
+#include "nlohmann/json.hpp" // https://github.com/nlohmann/json
+#include "base64.hpp" // https://github.com/tobiaslocker/base64
+
+int main() {
+
+    httplib::Client client("localhost", 8080);
+    const std::string filePath = "./demo.jpg";
+    std::ifstream file(filePath, std::ios::binary | std::ios::ate);
+    if (!file) {
+        std::cerr << "Error opening file: " << filePath << std::endl;
+        return 1;
+    }
+
+    std::streamsize size = file.tellg();
+    file.seekg(0, std::ios::beg);
+    std::vector<char> buffer(size);
+    if (!file.read(buffer.data(), size)) {
+        std::cerr << "Error reading file." << std::endl;
+        return 1;
+    }
+
+    std::string bufferStr(buffer.data(), static_cast<size_t>(size));
+    std::string encodedFile = base64::to_base64(bufferStr);
+
+    nlohmann::json jsonObj;
+    jsonObj["file"] = encodedFile;
+    jsonObj["fileType"] = 1;
+
+    auto response = client.Post("/document-preprocessing", jsonObj.dump(), "application/json");
+
+    if (response && response->status == 200) {
+        nlohmann::json jsonResponse = nlohmann::json::parse(response->body);
+        auto result = jsonResponse["result"];
+
+        if (!result.is_object() || !result["docPreprocessingResults"].is_array()) {
+            std::cerr << "Unexpected response format." << std::endl;
+            return 1;
+        }
+
+        for (size_t i = 0; i < result["docPreprocessingResults"].size(); ++i) {
+            auto res = result["docPreprocessingResults"][i];
+
+            if (res.contains("prunedResult")) {
+                std::cout << "Preprocessed result: " << res["prunedResult"].dump() << std::endl;
+            }
+
+            if (res.contains("outputImage")) {
+                std::string outputImgPath = "out_" + std::to_string(i) + ".png";
+                std::string decodedImage = base64::from_base64(res["outputImage"].get<std::string>());
+
+                std::ofstream outFile(outputImgPath, std::ios::binary);
+                if (outFile.is_open()) {
+                    outFile.write(decodedImage.c_str(), decodedImage.size());
+                    outFile.close();
+                    std::cout << "Saved image: " << outputImgPath << std::endl;
+                } else {
+                    std::cerr << "Failed to write image: " << outputImgPath << std::endl;
+                }
+            }
+        }
+    } else {
+        std::cerr << "Request failed." << std::endl;
+        if (response) {
+            std::cerr << "HTTP status: " << response->status << std::endl;
+            std::cerr << "Response body: " << response->body << std::endl;
+        }
+        return 1;
+    }
+
+    return 0;
+}
+
+</code></pre></details>
+
+<details><summary>Java</summary>
+
+<pre><code class="language-java">import okhttp3.*;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.util.Base64;
+
+public class Main {
+    public static void main(String[] args) throws IOException {
+        String API_URL = "http://localhost:8080/document-preprocessing";
+        String imagePath = "./demo.jpg";
+
+        File file = new File(imagePath);
+        byte[] fileContent = java.nio.file.Files.readAllBytes(file.toPath());
+        String base64Image = Base64.getEncoder().encodeToString(fileContent);
+
+        ObjectMapper objectMapper = new ObjectMapper();
+        ObjectNode payload = objectMapper.createObjectNode();
+        payload.put("file", base64Image);
+        payload.put("fileType", 1);
+
+        OkHttpClient client = new OkHttpClient();
+        MediaType JSON = MediaType.get("application/json; charset=utf-8");
+        RequestBody body = RequestBody.create(JSON, payload.toString());
+
+        Request request = new Request.Builder()
+                .url(API_URL)
+                .post(body)
+                .build();
+
+        try (Response response = client.newCall(request).execute()) {
+            if (response.isSuccessful()) {
+                String responseBody = response.body().string();
+                JsonNode root = objectMapper.readTree(responseBody);
+                JsonNode result = root.get("result");
+
+                JsonNode docPreprocessingResults = result.get("docPreprocessingResults");
+                for (int i = 0; i < docPreprocessingResults.size(); i++) {
+                    JsonNode item = docPreprocessingResults.get(i);
+                    int finalI = i;
+
+                    JsonNode prunedResult = item.get("prunedResult");
+                    System.out.println("Pruned Result [" + i + "]: " + prunedResult.toString());
+
+                    String outputImgBase64 = item.get("outputImage").asText();
+                    byte[] outputImgBytes = Base64.getDecoder().decode(outputImgBase64);
+                    String outputImgPath = "out_" + finalI + ".png";
+                    try (FileOutputStream fos = new FileOutputStream(outputImgPath)) {
+                        fos.write(outputImgBytes);
+                        System.out.println("Saved output image: " + outputImgPath);
+                    }
+
+                    JsonNode inputImageNode = item.get("inputImage");
+                    if (inputImageNode != null && !inputImageNode.isNull()) {
+                        String inputImageBase64 = inputImageNode.asText();
+                        byte[] inputImageBytes = Base64.getDecoder().decode(inputImageBase64);
+                        String inputImgPath = "inputImage_" + i + ".jpg";
+                        try (FileOutputStream fos = new FileOutputStream(inputImgPath)) {
+                            fos.write(inputImageBytes);
+                            System.out.println("Saved input image to: " + inputImgPath);
+                        }
+                    }
+                }
+            } else {
+                System.err.println("Request failed with HTTP code: " + response.code());
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Go</summary>
+
+<pre><code class="language-go">package main
+
+import (
+    "bytes"
+    "encoding/base64"
+    "encoding/json"
+    "fmt"
+    "io/ioutil"
+    "net/http"
+    "os"
+)
+
+func main() {
+    API_URL := "http://localhost:8080/document-preprocessing"
+    filePath := "./demo.jpg"
+
+    fileBytes, err := ioutil.ReadFile(filePath)
+    if err != nil {
+        fmt.Printf("Error reading file: %v\n", err)
+        return
+    }
+    fileData := base64.StdEncoding.EncodeToString(fileBytes)
+
+    payload := map[string]interface{}{
+        "file":     fileData,
+        "fileType": 1,
+    }
+    payloadBytes, err := json.Marshal(payload)
+    if err != nil {
+        fmt.Printf("Error marshaling payload: %v\n", err)
+        return
+    }
+
+    client := &http.Client{}
+    req, err := http.NewRequest("POST", API_URL, bytes.NewBuffer(payloadBytes))
+    if err != nil {
+        fmt.Printf("Error creating request: %v\n", err)
+        return
+    }
+    req.Header.Set("Content-Type", "application/json")
+
+    res, err := client.Do(req)
+    if err != nil {
+        fmt.Printf("Error sending request: %v\n", err)
+        return
+    }
+    defer res.Body.Close()
+
+    if res.StatusCode != http.StatusOK {
+        fmt.Printf("Unexpected status code: %d\n", res.StatusCode)
+        return
+    }
+
+    body, err := ioutil.ReadAll(res.Body)
+    if err != nil {
+        fmt.Printf("Error reading response body: %v\n", err)
+        return
+    }
+
+    type DocPreprocessingResult struct {
+        PrunedResult         map[string]interface{} `json:"prunedResult"`
+        OutputImage          string                 `json:"outputImage"`
+        DocPreprocessingImage *string               `json:"docPreprocessingImage"`
+        InputImage           *string                `json:"inputImage"`
+    }
+
+    type Response struct {
+        Result struct {
+            DocPreprocessingResults []DocPreprocessingResult `json:"docPreprocessingResults"`
+            DataInfo                interface{}              `json:"dataInfo"`
+        } `json:"result"`
+    }
+
+    var respData Response
+    if err := json.Unmarshal(body, &respData); err != nil {
+        fmt.Printf("Error unmarshaling response: %v\n", err)
+        return
+    }
+
+    for i, res := range respData.Result.DocPreprocessingResults {
+        fmt.Printf("Result %d - prunedResult: %+v\n", i, res.PrunedResult)
+
+        imgBytes, err := base64.StdEncoding.DecodeString(res.OutputImage)
+        if err != nil {
+            fmt.Printf("Error decoding outputImage at index %d: %v\n", i, err)
+            continue
+        }
+
+        filename := fmt.Sprintf("out_%d.png", i)
+        if err := os.WriteFile(filename, imgBytes, 0644); err != nil {
+            fmt.Printf("Error saving image %s: %v\n", filename, err)
+            continue
+        }
+        fmt.Printf("Saved output image to %s\n", filename)
+    }
+}
+</code></pre></details>
+
+<details><summary>C#</summary>
+
+<pre><code class="language-csharp">using System;
+using System.IO;
+using System.Net.Http;
+using System.Text;
+using System.Threading.Tasks;
+using Newtonsoft.Json.Linq;
+
+class Program
+{
+    static readonly string API_URL = "http://localhost:8080/document-preprocessing";
+    static readonly string inputFilePath = "./demo.jpg";
+
+    static async Task Main(string[] args)
+    {
+        var httpClient = new HttpClient();
+
+        byte[] fileBytes = File.ReadAllBytes(inputFilePath);
+        string fileData = Convert.ToBase64String(fileBytes);
+
+        var payload = new JObject
+        {
+            { "file", fileData },
+            { "fileType", 1 }
+        };
+        var content = new StringContent(payload.ToString(), Encoding.UTF8, "application/json");
+
+        HttpResponseMessage response = await httpClient.PostAsync(API_URL, content);
+        response.EnsureSuccessStatusCode();
+
+        string responseBody = await response.Content.ReadAsStringAsync();
+        JObject jsonResponse = JObject.Parse(responseBody);
+
+        JArray docPreResults = (JArray)jsonResponse["result"]["docPreprocessingResults"];
+        for (int i = 0; i < docPreResults.Count; i++)
+        {
+            var res = docPreResults[i];
+            Console.WriteLine($"[{i}] prunedResult:\n{res["prunedResult"]}");
+
+            string base64Image = res["outputImage"]?.ToString();
+            if (!string.IsNullOrEmpty(base64Image))
+            {
+                string outputPath = $"out_{i}.png";
+                byte[] imageBytes = Convert.FromBase64String(base64Image);
+                File.WriteAllBytes(outputPath, imageBytes);
+                Console.WriteLine($"Output image saved at {outputPath}");
+            }
+            else
+            {
+                Console.WriteLine($"outputImage at index {i} is null.");
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Node.js</summary>
+
+<pre><code class="language-js">const axios = require('axios');
+const fs = require('fs');
+const path = require('path');
+
+const API_URL = 'http://localhost:8080/document-preprocessing';
+const imagePath = './demo.jpg';
+
+function encodeImageToBase64(filePath) {
+  const bitmap = fs.readFileSync(filePath);
+  return Buffer.from(bitmap).toString('base64');
+}
+
+const payload = {
+  file: encodeImageToBase64(imagePath),
+  fileType: 1
+};
+
+axios.post(API_URL, payload, {
+  headers: {
+    'Content-Type': 'application/json'
+  },
+  maxBodyLength: Infinity
+})
+.then((response) => {
+  const results = response.data.result.docPreprocessingResults;
+
+  results.forEach((res, index) => {
+    console.log(`\n[${index}] prunedResult:`);
+    console.log(res.prunedResult);
+
+    const base64Image = res.outputImage;
+    if (base64Image) {
+      const outputImagePath = `out_${index}.png`;
+      const imageBuffer = Buffer.from(base64Image, 'base64');
+      fs.writeFileSync(outputImagePath, imageBuffer);
+      console.log(`Output image saved at ${outputImagePath}`);
+    } else {
+      console.log(`outputImage at index ${index} is null.`);
+    }
+  });
+})
+.catch((error) => {
+  console.error('API error:', error.message);
+});
+</code></pre></details>
+
+<details><summary>PHP</summary>
+
+<pre><code class="language-php">&lt;?php
+
+$API_URL = "http://localhost:8080/document-preprocessing";
+$image_path = "./demo.jpg";
+$output_image_path = "./out_0.png";
+
+$image_data = base64_encode(file_get_contents($image_path));
+$payload = array("file" => $image_data, "fileType" => 1);
+
+$ch = curl_init($API_URL);
+curl_setopt($ch, CURLOPT_POST, true);
+curl_setopt($ch, CURLOPT_POSTFIELDS, json_encode($payload));
+curl_setopt($ch, CURLOPT_HTTPHEADER, array('Content-Type: application/json'));
+curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+$response = curl_exec($ch);
+curl_close($ch);
+
+$result = json_decode($response, true)["result"]["docPreprocessingResults"];
+
+foreach ($result as $i => $item) {
+    echo "[$i] prunedResult:\n";
+    print_r($item["prunedResult"]);
+
+    if (!empty($item["outputImage"])) {
+        $output_image_path = "out_" . $i . ".png";
+        file_put_contents($output_image_path, base64_decode($item["outputImage"]));
+        echo "Output image saved at $output_image_path\n";
+    } else {
+        echo "No outputImage found for item $i\n";
+    }
+}
+?&gt;
+</code></pre></details>
+</details>
+<br/>
+
+## 4. Secondary Development
+
+If the default model weights provided by the document image preprocessing pipeline do not meet your accuracy or speed requirements in your specific scenario, you can attempt to further **fine-tune** the existing model using **your own domain-specific or application-specific data** to enhance the recognition performance of the document image preprocessing pipeline in your context.
+
+### 4.1 Model Fine-Tuning
+
+Since the document image preprocessing pipeline comprises multiple modules, any module could potentially contribute to suboptimal performance if the overall pipeline does not meet expectations. You can analyze images with poor recognition results to identify which module is causing the issue and then refer to the corresponding fine-tuning tutorial links in the table below to perform model fine-tuning.
+
+<table>
+<thead>
+<tr>
+<th>Scenario</th>
+<th>Module to Fine-Tune</th>
+<th>Fine-Tuning Reference Link</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>Inaccurate rotation correction of the entire image</td>
+<td>Document Image Orientation Classification Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate distortion correction of the image</td>
+<td>Text Image Rectification Module</td>
+<td>Fine-tuning is currently not supported</td>
+</tr>
+</tbody>
+</table>
diff --git a/docs/version3.x/pipeline_usage/doc_preprocessor.md b/docs/version3.x/pipeline_usage/doc_preprocessor.md
new file mode 100644
index 0000000..b96f967
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/doc_preprocessor.md
@@ -0,0 +1,1339 @@
+---
+comments: true
+---
+
+# 文档图像预处理产线使用教程
+
+## 1. 文档图像预处理产线介绍
+
+文档图像预处理产线集成了文档方向分类和形变矫正两大功能。文档方向分类可自动识别文档的四个方向（0°、90°、180°、270°），确保文档以正确的方向进行后续处理。文本图像矫正模型则用于修正文档拍摄或扫描过程中的几何扭曲，恢复文档的原始形状和比例。适用于数字化文档管理、OCR类任务前处理、以及任何需要提高文档图像质量的场景。通过自动化的方向校正与形变矫正，该模块显著提升了文档处理的准确性和效率，为用户提供更为可靠的图像分析基础。本产线同时提供了灵活的服务化部署方式，支持在多种硬件上使用多种编程语言调用。不仅如此，本产线也提供了二次开发的能力，您可以基于本产线在您自己的数据集上训练调优，训练后的模型也可以无缝集成。
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/doc_preprocessor/02.jpg">
+
+<b>通用文档图像预处理产线中包含以下2个模块。每个模块均可独立进行训练和推理，并包含多个模型。有关详细信息，请点击相应模块以查看文档。</b>
+
+- [文档图像方向分类模块](../module_usage/doc_img_orientation_classification.md)（可选）
+- [文本图像矫正模块](../module_usage/text_image_unwarping.md)（可选）
+
+在本产线中，您可以根据下方的基准测试数据选择使用的模型。
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。
+> 在带有 [常规模式 / 高性能模式] 标记的推理耗时列中，`常规模式` 对应本地推理引擎 `paddle_static`。
+
+<details>
+<summary> <b>文档图像方向分类模块（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>Top-1 Acc（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">训练模型</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>基于PP-LCNet_x1_0的文档图像分类模型，含有四个类别，即0度，90度，180度，270度</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary> <b>文本图像矫正模块（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>CER </th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">训练模型</a></td>
+<td>0.179</td>
+<td>19.05 / 19.05</td>
+<td>- / 869.82</td>
+<td>30.3</td>
+<td>高精度文本图像矫正模型</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary> <b>测试环境说明：</b></summary>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+                      <li><strong>测试数据集：
+             </strong>
+                <ul>
+                  <li>文档图像方向分类模型：自建的内部数据集，覆盖证件和文档等多个场景，包含 1000 张图片。</li>
+                  <li> 文本图像矫正模型：<a href="https://www3.cs.stonybrook.edu/~cvl/docunet.html">DocUNet。</a></li>
+                </ul>
+             </li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td><code>paddle_static</code></td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+</details>
+
+## 2. 快速开始
+
+
+在本地使用通用文档图像预处理产线前，请确保您已经按照[安装教程](../installation.md)完成了wheel包安装。安装完成后，可以在本地使用命令行体验或 Python 集成。
+
+**请注意，如果在执行过程中遇到程序失去响应、程序异常退出、内存资源耗尽、推理速度极慢等问题，请尝试参考文档调整配置，例如关闭不需要使用的功能或使用更轻量的模型。**
+
+
+### 2.1 命令行方式体验
+
+一行命令即可快速体验 doc_preprocessor 产线效果：
+
+```bash
+paddleocr doc_preprocessor -i https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/doc_test_rotated.jpg
+
+# 通过 --use_doc_orientation_classify 指定是否使用文档方向分类模型
+paddleocr doc_preprocessor -i ./doc_test_rotated.jpg --use_doc_orientation_classify True
+
+# 通过 --use_doc_unwarping 指定是否使用文本图像矫正模块
+paddleocr doc_preprocessor -i ./doc_test_rotated.jpg --use_doc_unwarping True
+
+# 通过 --device 指定模型推理时使用 GPU
+paddleocr doc_preprocessor -i ./doc_test_rotated.jpg --device gpu
+```
+
+上述命令默认使用本地推理引擎 `paddle_static`。如需运行，请先参考[飞桨框架安装说明](../paddlepaddle_installation.md)安装 PaddlePaddle。
+
+如果选择 `transformers` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 Transformers 环境配置，然后执行如下命令：
+
+```bash
+# 使用 transformers 引擎进行推理
+paddleocr doc_preprocessor -i https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/doc_test_rotated.jpg \
+    --engine transformers
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+<details><summary><b>命令行支持更多参数设置，点击展开以查看命令行参数的详细说明</b></summary>
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，必填。<br/>
+<b>说明：</b>
+如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；
+<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/doc_test_rotated.jpg">示例</a>；
+<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td><b>含义：</b>指定推理结果文件保存的路径。<br/>
+<b>说明：</b>
+如果不设置，推理结果将不会保存到本地。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>含义：</b>文档方向分类模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>文档方向分类模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>含义：</b>文本图像矫正模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否加载并使用文档方向分类模块。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。
+</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。
+</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+</ul>如果不设置，将默认使用产线初始化的该参数值，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>fp32</code>、<code>fp16</code>。</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+
+</tbody>
+</table>
+</details>
+<br />
+
+运行结果会被打印到终端上，默认配置的 doc_preprocessor 产线的运行结果如下：
+
+```bash
+{'res': {'input_path': '/root/.paddlex/predict_input/doc_test_rotated.jpg', 'page_index': None, 'model_settings': {'use_doc_orientation_classify': True, 'use_doc_unwarping': True}, 'angle': 180}}
+```
+
+可视化结果保存在`save_path`下，可视化结果如下：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/doc_preprocessor/02.jpg"/>
+
+
+### 2.2 Python脚本方式集成
+
+命令行方式是为了快速体验查看效果，一般来说，在项目中，往往需要通过代码集成，您可以通过几行代码即可完成产线的快速推理，推理代码如下：
+
+```python
+from paddleocr import DocPreprocessor
+
+pipeline = DocPreprocessor()
+# docpp = DocPreprocessor(use_doc_orientation_classify=True) # 通过 use_doc_orientation_classify 指定是否使用文档方向分类模型
+# docpp = DocPreprocessor(use_doc_unwarping=True) # 通过 use_doc_unwarping 指定是否使用文本图像矫正模块
+# docpp = DocPreprocessor(device="gpu") # 通过 device 指定模型推理时使用 GPU
+output = pipeline.predict("./doc_test_rotated.jpg")
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_img("./output/")
+    res.save_to_json("./output/")
+```
+
+上述代码默认使用本地推理引擎 `paddle_static`。如需运行，请先参考[飞桨框架安装说明](../paddlepaddle_installation.md)安装 PaddlePaddle。
+
+如果选择 `transformers` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 Transformers 环境配置，然后执行如下代码：
+
+```python
+from paddleocr import DocPreprocessor
+
+pipeline = DocPreprocessor(
+    engine="transformers",
+)
+# docpp = DocPreprocessor(use_doc_orientation_classify=True) # 通过 use_doc_orientation_classify 指定是否使用文档方向分类模型
+# docpp = DocPreprocessor(use_doc_unwarping=True) # 通过 use_doc_unwarping 指定是否使用文本图像矫正模块
+# docpp = DocPreprocessor(device="gpu") # 通过 device 指定模型推理时使用 GPU
+output = pipeline.predict("./doc_test_rotated.jpg")
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_img("./output/")
+    res.save_to_json("./output/")
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+在上述 Python 脚本中，执行了如下几个步骤：
+（1）通过 <code>DocPreprocessor()</code> 实例化 doc_preprocessor 产线对象：具体参数说明如下：
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>含义：</b>文档方向分类模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>文档方向分类模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>含义：</b>文本图像矫正模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否加载并使用文档方向分类模块。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。
+</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。
+</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+</tbody>
+</table>
+
+（2）调用 doc_preprocessor 产线对象的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。以下是 <code>predict()</code> 方法的参数及其说明：
+
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据；</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/doc_test_rotated.jpg">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)；</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]</code>。</li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否在推理时使用文档方向分类模块。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否在推理时使用文本图像矫正模块。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+</table>
+
+
+
+（3）对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为<code>json</code>文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将结果保存为图像格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径</td>
+<td>无</td>
+</tr>
+</table>
+
+<ul>
+<li>调用<code>print()</code> 方法会将结果打印到终端，打印到终端的内容解释如下：</li>
+    <ol start="1" type="1">
+     <li><code>input_path</code>: <code>(str)</code> 待预测图像的输入路径</li>
+     <li><code>page_index</code>: <code>(Union[int, None])</code> 如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>  
+     <li><code>model_settings</code>: <code>(Dict[str, bool])</code> 配置产线所需的模型参数</li>
+        <ol>
+        <li><code>use_doc_orientation_classify</code>: <code>(bool)</code> 控制是否启用文档方向分类模块</li>
+        <li><code>use_doc_unwarping</code>: <code>(bool)</code> 控制是否启用文本图像矫正模块</li>
+        </ol>
+    <li><code>angle</code>: <code>(int)</code> 文档方向分类的预测结果。启用时取值为[0,90,180,270]；未启用时为-1</li>
+    </ol>
+<li>调用<code>save_to_json()</code> 方法会将上述内容保存到指定的<code>save_path</code>中，如果指定为目录，则保存的路径为<code>save_path/{your_img_basename}.json</code>，如果指定为文件，则直接保存到该文件中。由于json文件不支持保存numpy数组，因此会将其中的<code>numpy.array</code>类型转换为列表形式。</li>
+<li>调用<code>save_to_img()</code> 方法会将可视化结果保存到指定的<code>save_path</code>中，如果指定为目录，则保存的路径为<code>save_path/{your_img_basename}_doc_preprocessor_res_img.{your_img_extension}</code>，如果指定为文件，则直接保存到该文件中。(产线通常包含较多结果图片，不建议直接指定为具体的文件路径，否则多张图会被覆盖，仅保留最后一张图)</li>
+
+* 此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">获取预测的 <code>json</code> 格式的结果</td>
+</tr>
+<tr>
+<td rowspan="2"><code>img</code></td>
+<td rowspan="2">获取格式为 <code>dict</code> 的可视化图像</td>
+</tr>
+</table>
+<ul>
+<li><code>json</code> 属性获取的预测结果为dict类型的数据，相关内容与调用 <code>save_to_json()</code> 方法保存的内容一致。</li>
+<li><code>img</code> 属性返回的预测结果是一个dict类型的数据。其中，键为 <code>preprocessed_img</code>，对应的值是 <code>Image.Image</code> 对象：用于显示 doc_preprocessor 结果的可视化图像。</li>
+</ul>
+
+
+## 3. 开发集成/部署
+
+如果产线可以达到您对产线推理速度和精度的要求，您可以直接进行开发集成/部署。
+
+若您需要将产线直接应用在您的Python项目中，可以参考 [2.2 Python脚本方式](#22-python)中的示例代码。
+
+此外，PaddleOCR 也提供了其他两种部署方式，详细说明如下：
+
+🚀 高性能推理：在实际生产环境中，许多应用对部署策略的性能指标（尤其是响应速度）有着较严苛的标准，以确保系统的高效运行与用户体验的流畅性。为此，PaddleOCR 提供高性能推理功能，旨在对模型推理及前后处理进行深度性能优化，实现端到端流程的显著提速，详细的高性能推理流程请参考[高性能推理](../inference_deployment/local_inference/high_performance_inference.md)。
+
+☁️ 服务化部署：服务化部署是实际生产环境中常见的一种部署形式。通过将推理功能封装为服务，客户端可以通过网络请求来访问这些服务，以获取推理结果。详细的产线服务化部署流程请参考[服务化部署](../inference_deployment/serving/serving.md)。
+
+以下是基础服务化部署的API参考与多语言服务调用示例：
+
+<details><summary>API参考</summary>
+<p>对于服务提供的主要操作：</p>
+<ul>
+<li>HTTP请求方法为POST。</li>
+<li>请求体和响应体均为JSON数据（JSON对象）。</li>
+<li>当请求处理成功时，响应状态码为<code>200</code>，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。固定为<code>0</code>。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。固定为<code>"Success"</code>。</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>操作结果。</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>当请求处理未成功时，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。与响应状态码相同。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。</td>
+</tr>
+</tbody>
+</table>
+<p>服务提供的主要操作如下：</p>
+<ul>
+<li><b><code>infer</code></b></li>
+</ul>
+<p>获取图像文档图像预处理结果。</p>
+<p><code>POST /document-preprocessing</code></p>
+<ul>
+<li>请求体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>服务器可访问的图像文件（含 TIFF，多页时按页处理）或 PDF 文件的 URL，或上述类型文件内容的 Base64 编码结果。默认对于超过 10 页的 PDF 或多页 TIFF，仅处理前 10 页。<br /> 要解除页数限制，请在产线配置文件中添加以下配置：
+<pre><code>Serving:
+  extra:
+    max_num_input_imgs: null
+</code></pre>
+</td>
+<td>是</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>文件类型。<code>0</code> 表示 PDF 文件，<code>1</code> 表示图像文件（含 TIFF）。若请求体无此属性，则将根据URL推断文件类型。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_doc_orientation_classify</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_doc_unwarping</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>是否返回可视化结果图以及处理过程中的中间图像等。
+<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>传入 <code>true</code>：返回图像。</li>
+<li>传入 <code>false</code>：不返回图像。</li>
+<li>若请求体中未提供该参数或传入 <code>null</code>：遵循产线配置文件<code>Serving.visualize</code> 的设置。</li>
+</ul>
+<br/>例如，在产线配置文件中添加如下字段：<br/>
+<pre><code>Serving:
+  visualize: False
+</code></pre>
+将默认不返回图像，通过请求体中的<code>visualize</code>参数可以覆盖默认行为。如果请求体和配置文件中均未设置（或请求体传入<code>null</code>、配置文件中未设置），则默认返回图像。
+</td>
+<td>否</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>请求处理成功时，响应体的<code>result</code>具有如下属性：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>docPreprocessingResults</code></td>
+<td><code>object</code></td>
+<td>文档图像预处理结果。数组长度为1（对于图像输入）或实际处理的文档页数（对于PDF输入）。对于PDF输入，数组中的每个元素依次表示PDF文件中实际处理的每一页的结果。</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>输入数据信息。</td>
+</tr>
+</tbody>
+</table>
+<p><code>docPreprocessingResults</code>中的每个元素为一个<code>object</code>，具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>outputImage</code></td>
+<td><code>string</code></td>
+<td>经过预处理的图像。图像为PNG格式，使用Base64编码。</td>
+</tr>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>产线对象的 <code>predict</code> 方法生成结果的 JSON 表示中 <code>res</code> 字段的简化版本，其中去除了 <code>input_path</code> 和 <code>page_index</code> 字段。</td>
+</tr>
+<tr>
+<td><code>docPreprocessingImage</code></td>
+<td><code>string</code> ｜ <code>null</code></td>
+<td>可视化结果图。图像为JPEG格式，使用Base64编码。</td>
+</tr>
+<tr>
+<td><code>inputImage</code></td>
+<td><code>string</code> ｜ <code>null</code></td>
+<td>输入图像。图像为JPEG格式，使用Base64编码。</td>
+</tr>
+</tbody>
+</table>
+</details>
+<details><summary>多语言调用服务示例</summary>
+<details>
+<summary>Python</summary>
+
+<pre><code class="language-python">import base64
+import requests
+
+API_URL = "http://localhost:8080/document-preprocessing"
+file_path = "./demo.jpg"
+
+with open(file_path, "rb") as file:
+    file_bytes = file.read()
+    file_data = base64.b64encode(file_bytes).decode("ascii")
+
+payload = {"file": file_data, "fileType": 1}
+
+response = requests.post(API_URL, json=payload)
+
+assert response.status_code == 200
+result = response.json()["result"]
+for i, res in enumerate(result["docPreprocessingResults"]):
+    print(res["prunedResult"])
+    output_img_path = f"out_{i}.png"
+    with open(output_img_path, "wb") as f:
+        f.write(base64.b64decode(res["outputImage"]))
+    print(f"Output image saved at {output_img_path}")
+</code></pre></details>
+
+<details><summary>C++</summary>
+
+<pre><code class="language-cpp">#include &lt;iostream&gt;
+#include &lt;fstream&gt;
+#include &lt;vector&gt;
+#include &lt;string&gt;
+#include "cpp-httplib/httplib.h" // https://github.com/Huiyicc/cpp-httplib
+#include "nlohmann/json.hpp" // https://github.com/nlohmann/json
+#include "base64.hpp" // https://github.com/tobiaslocker/base64
+
+int main() {
+
+    httplib::Client client("localhost", 8080);
+    const std::string filePath = "./demo.jpg";
+    std::ifstream file(filePath, std::ios::binary | std::ios::ate);
+    if (!file) {
+        std::cerr << "Error opening file: " << filePath << std::endl;
+        return 1;
+    }
+
+    std::streamsize size = file.tellg();
+    file.seekg(0, std::ios::beg);
+    std::vector<char> buffer(size);
+    if (!file.read(buffer.data(), size)) {
+        std::cerr << "Error reading file." << std::endl;
+        return 1;
+    }
+
+    std::string bufferStr(buffer.data(), static_cast<size_t>(size));
+    std::string encodedFile = base64::to_base64(bufferStr);
+
+    nlohmann::json jsonObj;
+    jsonObj["file"] = encodedFile;
+    jsonObj["fileType"] = 1;
+
+    auto response = client.Post("/document-preprocessing", jsonObj.dump(), "application/json");
+
+    if (response && response->status == 200) {
+        nlohmann::json jsonResponse = nlohmann::json::parse(response->body);
+        auto result = jsonResponse["result"];
+
+        if (!result.is_object() || !result["docPreprocessingResults"].is_array()) {
+            std::cerr << "Unexpected response format." << std::endl;
+            return 1;
+        }
+
+        for (size_t i = 0; i < result["docPreprocessingResults"].size(); ++i) {
+            auto res = result["docPreprocessingResults"][i];
+
+            if (res.contains("prunedResult")) {
+                std::cout << "Preprocessed result: " << res["prunedResult"].dump() << std::endl;
+            }
+
+            if (res.contains("outputImage")) {
+                std::string outputImgPath = "out_" + std::to_string(i) + ".png";
+                std::string decodedImage = base64::from_base64(res["outputImage"].get<std::string>());
+
+                std::ofstream outFile(outputImgPath, std::ios::binary);
+                if (outFile.is_open()) {
+                    outFile.write(decodedImage.c_str(), decodedImage.size());
+                    outFile.close();
+                    std::cout << "Saved image: " << outputImgPath << std::endl;
+                } else {
+                    std::cerr << "Failed to write image: " << outputImgPath << std::endl;
+                }
+            }
+        }
+    } else {
+        std::cerr << "Request failed." << std::endl;
+        if (response) {
+            std::cerr << "HTTP status: " << response->status << std::endl;
+            std::cerr << "Response body: " << response->body << std::endl;
+        }
+        return 1;
+    }
+
+    return 0;
+}
+
+</code></pre></details>
+
+<details><summary>Java</summary>
+
+<pre><code class="language-java">import okhttp3.*;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.util.Base64;
+
+public class Main {
+    public static void main(String[] args) throws IOException {
+        String API_URL = "http://localhost:8080/document-preprocessing";
+        String imagePath = "./demo.jpg";
+
+        File file = new File(imagePath);
+        byte[] fileContent = java.nio.file.Files.readAllBytes(file.toPath());
+        String base64Image = Base64.getEncoder().encodeToString(fileContent);
+
+        ObjectMapper objectMapper = new ObjectMapper();
+        ObjectNode payload = objectMapper.createObjectNode();
+        payload.put("file", base64Image);
+        payload.put("fileType", 1);
+
+        OkHttpClient client = new OkHttpClient();
+        MediaType JSON = MediaType.get("application/json; charset=utf-8");
+        RequestBody body = RequestBody.create(JSON, payload.toString());
+
+        Request request = new Request.Builder()
+                .url(API_URL)
+                .post(body)
+                .build();
+
+        try (Response response = client.newCall(request).execute()) {
+            if (response.isSuccessful()) {
+                String responseBody = response.body().string();
+                JsonNode root = objectMapper.readTree(responseBody);
+                JsonNode result = root.get("result");
+
+                JsonNode docPreprocessingResults = result.get("docPreprocessingResults");
+                for (int i = 0; i < docPreprocessingResults.size(); i++) {
+                    JsonNode item = docPreprocessingResults.get(i);
+                    int finalI = i;
+
+                    JsonNode prunedResult = item.get("prunedResult");
+                    System.out.println("Pruned Result [" + i + "]: " + prunedResult.toString());
+
+                    String outputImgBase64 = item.get("outputImage").asText();
+                    byte[] outputImgBytes = Base64.getDecoder().decode(outputImgBase64);
+                    String outputImgPath = "out_" + finalI + ".png";
+                    try (FileOutputStream fos = new FileOutputStream(outputImgPath)) {
+                        fos.write(outputImgBytes);
+                        System.out.println("Saved output image: " + outputImgPath);
+                    }
+
+                    JsonNode inputImageNode = item.get("inputImage");
+                    if (inputImageNode != null && !inputImageNode.isNull()) {
+                        String inputImageBase64 = inputImageNode.asText();
+                        byte[] inputImageBytes = Base64.getDecoder().decode(inputImageBase64);
+                        String inputImgPath = "inputImage_" + i + ".jpg";
+                        try (FileOutputStream fos = new FileOutputStream(inputImgPath)) {
+                            fos.write(inputImageBytes);
+                            System.out.println("Saved input image to: " + inputImgPath);
+                        }
+                    }
+                }
+            } else {
+                System.err.println("Request failed with HTTP code: " + response.code());
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Go</summary>
+
+<pre><code class="language-go">package main
+
+import (
+    "bytes"
+    "encoding/base64"
+    "encoding/json"
+    "fmt"
+    "io/ioutil"
+    "net/http"
+    "os"
+)
+
+func main() {
+    API_URL := "http://localhost:8080/document-preprocessing"
+    filePath := "./demo.jpg"
+
+    fileBytes, err := ioutil.ReadFile(filePath)
+    if err != nil {
+        fmt.Printf("Error reading file: %v\n", err)
+        return
+    }
+    fileData := base64.StdEncoding.EncodeToString(fileBytes)
+
+    payload := map[string]interface{}{
+        "file":     fileData,
+        "fileType": 1,
+    }
+    payloadBytes, err := json.Marshal(payload)
+    if err != nil {
+        fmt.Printf("Error marshaling payload: %v\n", err)
+        return
+    }
+
+    client := &http.Client{}
+    req, err := http.NewRequest("POST", API_URL, bytes.NewBuffer(payloadBytes))
+    if err != nil {
+        fmt.Printf("Error creating request: %v\n", err)
+        return
+    }
+    req.Header.Set("Content-Type", "application/json")
+
+    res, err := client.Do(req)
+    if err != nil {
+        fmt.Printf("Error sending request: %v\n", err)
+        return
+    }
+    defer res.Body.Close()
+
+    if res.StatusCode != http.StatusOK {
+        fmt.Printf("Unexpected status code: %d\n", res.StatusCode)
+        return
+    }
+
+    body, err := ioutil.ReadAll(res.Body)
+    if err != nil {
+        fmt.Printf("Error reading response body: %v\n", err)
+        return
+    }
+
+    type DocPreprocessingResult struct {
+        PrunedResult         map[string]interface{} `json:"prunedResult"`
+        OutputImage          string                 `json:"outputImage"`
+        DocPreprocessingImage *string               `json:"docPreprocessingImage"`
+        InputImage           *string                `json:"inputImage"`
+    }
+
+    type Response struct {
+        Result struct {
+            DocPreprocessingResults []DocPreprocessingResult `json:"docPreprocessingResults"`
+            DataInfo                interface{}              `json:"dataInfo"`
+        } `json:"result"`
+    }
+
+    var respData Response
+    if err := json.Unmarshal(body, &respData); err != nil {
+        fmt.Printf("Error unmarshaling response: %v\n", err)
+        return
+    }
+
+    for i, res := range respData.Result.DocPreprocessingResults {
+        fmt.Printf("Result %d - prunedResult: %+v\n", i, res.PrunedResult)
+
+        imgBytes, err := base64.StdEncoding.DecodeString(res.OutputImage)
+        if err != nil {
+            fmt.Printf("Error decoding outputImage at index %d: %v\n", i, err)
+            continue
+        }
+
+        filename := fmt.Sprintf("out_%d.png", i)
+        if err := os.WriteFile(filename, imgBytes, 0644); err != nil {
+            fmt.Printf("Error saving image %s: %v\n", filename, err)
+            continue
+        }
+        fmt.Printf("Saved output image to %s\n", filename)
+    }
+}
+</code></pre></details>
+
+<details><summary>C#</summary>
+
+<pre><code class="language-csharp">using System;
+using System.IO;
+using System.Net.Http;
+using System.Text;
+using System.Threading.Tasks;
+using Newtonsoft.Json.Linq;
+
+class Program
+{
+    static readonly string API_URL = "http://localhost:8080/document-preprocessing";
+    static readonly string inputFilePath = "./demo.jpg";
+
+    static async Task Main(string[] args)
+    {
+        var httpClient = new HttpClient();
+
+        byte[] fileBytes = File.ReadAllBytes(inputFilePath);
+        string fileData = Convert.ToBase64String(fileBytes);
+
+        var payload = new JObject
+        {
+            { "file", fileData },
+            { "fileType", 1 }
+        };
+        var content = new StringContent(payload.ToString(), Encoding.UTF8, "application/json");
+
+        HttpResponseMessage response = await httpClient.PostAsync(API_URL, content);
+        response.EnsureSuccessStatusCode();
+
+        string responseBody = await response.Content.ReadAsStringAsync();
+        JObject jsonResponse = JObject.Parse(responseBody);
+
+        JArray docPreResults = (JArray)jsonResponse["result"]["docPreprocessingResults"];
+        for (int i = 0; i < docPreResults.Count; i++)
+        {
+            var res = docPreResults[i];
+            Console.WriteLine($"[{i}] prunedResult:\n{res["prunedResult"]}");
+
+            string base64Image = res["outputImage"]?.ToString();
+            if (!string.IsNullOrEmpty(base64Image))
+            {
+                string outputPath = $"out_{i}.png";
+                byte[] imageBytes = Convert.FromBase64String(base64Image);
+                File.WriteAllBytes(outputPath, imageBytes);
+                Console.WriteLine($"Output image saved at {outputPath}");
+            }
+            else
+            {
+                Console.WriteLine($"outputImage at index {i} is null.");
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Node.js</summary>
+
+<pre><code class="language-js">const axios = require('axios');
+const fs = require('fs');
+const path = require('path');
+
+const API_URL = 'http://localhost:8080/document-preprocessing';
+const imagePath = './demo.jpg';
+
+function encodeImageToBase64(filePath) {
+  const bitmap = fs.readFileSync(filePath);
+  return Buffer.from(bitmap).toString('base64');
+}
+
+const payload = {
+  file: encodeImageToBase64(imagePath),
+  fileType: 1
+};
+
+axios.post(API_URL, payload, {
+  headers: {
+    'Content-Type': 'application/json'
+  },
+  maxBodyLength: Infinity
+})
+.then((response) => {
+  const results = response.data.result.docPreprocessingResults;
+
+  results.forEach((res, index) => {
+    console.log(`\n[${index}] prunedResult:`);
+    console.log(res.prunedResult);
+
+    const base64Image = res.outputImage;
+    if (base64Image) {
+      const outputImagePath = `out_${index}.png`;
+      const imageBuffer = Buffer.from(base64Image, 'base64');
+      fs.writeFileSync(outputImagePath, imageBuffer);
+      console.log(`Output image saved at ${outputImagePath}`);
+    } else {
+      console.log(`outputImage at index ${index} is null.`);
+    }
+  });
+})
+.catch((error) => {
+  console.error('API error:', error.message);
+});
+</code></pre></details>
+
+<details><summary>PHP</summary>
+
+<pre><code class="language-php">&lt;?php
+
+$API_URL = "http://localhost:8080/document-preprocessing";
+$image_path = "./demo.jpg";
+$output_image_path = "./out_0.png";
+
+$image_data = base64_encode(file_get_contents($image_path));
+$payload = array("file" => $image_data, "fileType" => 1);
+
+$ch = curl_init($API_URL);
+curl_setopt($ch, CURLOPT_POST, true);
+curl_setopt($ch, CURLOPT_POSTFIELDS, json_encode($payload));
+curl_setopt($ch, CURLOPT_HTTPHEADER, array('Content-Type: application/json'));
+curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+$response = curl_exec($ch);
+curl_close($ch);
+
+$result = json_decode($response, true)["result"]["docPreprocessingResults"];
+
+foreach ($result as $i => $item) {
+    echo "[$i] prunedResult:\n";
+    print_r($item["prunedResult"]);
+
+    if (!empty($item["outputImage"])) {
+        $output_image_path = "out_" . $i . ".png";
+        file_put_contents($output_image_path, base64_decode($item["outputImage"]));
+        echo "Output image saved at $output_image_path\n";
+    } else {
+        echo "No outputImage found for item $i\n";
+    }
+}
+?&gt;
+</code></pre></details>
+</details>
+<br/>
+
+## 4. 二次开发
+
+如果文档图像预处理产线提供的默认模型权重在您的场景中，精度或速度不满意，您可以尝试利用<b>您自己拥有的特定领域或应用场景的数据</b>对现有模型进行进一步的<b>微调</b>，以提升文档图像预处理产线的在您的场景中的识别效果。
+
+### 4.1 模型微调
+
+由于文档图像预处理产线包含若干模块，模型产线的效果如果不及预期，可能来自于其中任何一个模块。您可以对识别效果差的图片进行分析，进而确定是哪个模块存在问题，并参考以下表格中对应的微调教程链接进行模型微调。
+
+
+<table>
+<thead>
+<tr>
+<th>情形</th>
+<th>微调模块</th>
+<th>微调参考链接</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>整图旋转矫正不准</td>
+<td>文档图像方向分类模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>图像扭曲矫正不准</td>
+<td>文本图像矫正模块</td>
+<td>暂不支持微调</td>
+</tr>
+</tbody>
+</table>
diff --git a/docs/version3.x/pipeline_usage/doc_understanding.en.md b/docs/version3.x/pipeline_usage/doc_understanding.en.md
new file mode 100644
index 0000000..81c9b53
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/doc_understanding.en.md
@@ -0,0 +1,900 @@
+---
+comments: true
+---
+
+# Document Understanding Pipeline Usage Tutorial
+
+## 1. Introduction to the Document Understanding Pipeline
+
+The Document Understanding Pipeline is an advanced document processing technology based on Visual-Language Models (VLM), designed to overcome the limitations of traditional document processing. Traditional methods rely on fixed templates or predefined rules to parse documents, whereas this pipeline leverages the multimodal capabilities of VLM to accurately answer user queries by inputting document images and user questions, integrating visual and language information. This technology does not require pre-training for specific document formats, allowing it to flexibly handle diverse document content, significantly enhancing the generalization and practicality of document processing. It has broad application prospects in intelligent Q&A, information extraction, and other scenarios. Currently, the pipeline does not support secondary development of VLM models, but plans to support it in the future.
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/doc_understanding/doc_understanding.png">
+
+<b>The document understanding pipeline includes the following module. Each module can be trained and inferred independently and contains multiple models. For more details, click the corresponding module to view the documentation.</b>
+
+- [Document-like Vision Language Model Module](../module_usage/doc_vlm.md)
+
+In this pipeline, you can choose the model to use based on the benchmark data below.
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing.
+
+<details>
+<summary> <b>Document-like Vision Language Model Module:</b></summary>
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Model Storage Size (GB)</th>
+<th>Total Score</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>PP-DocBee-2B</td><td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBee-2B_infer.tar">Inference Model</a></td>
+<td>4.2</td>
+<td>765</td>
+<td rowspan="2">PP-DocBee is a multimodal large model independently developed by the PaddlePaddle team, focusing on document understanding, with excellent performance in Chinese document understanding tasks. The model is fine-tuned and optimized using nearly 5 million multimodal datasets related to document understanding, including general VQA, OCR, chart, text-rich documents, math and complex reasoning, synthetic data, pure text data, etc., with different training data ratios set. In several authoritative English document understanding evaluation lists in academia, PP-DocBee has generally achieved SOTA for models of the same parameter scale. In internal business Chinese scenarios, PP-DocBee also outperforms current popular open and closed-source models.</td>
+</tr>
+<tr>
+<td>PP-DocBee-7B</td><td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBee-7B_infer.tar">Inference Model</a></td>
+<td>15.8</td>
+<td>-</td>
+</tr>
+<tr>
+<td>PP-DocBee2-3B</td><td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBee2-3B_infer.tar">Inference Model</a></td>
+<td>7.6</td>
+<td>852</td>
+<td>PP-DocBee2 is a multimodal large model independently developed by the PaddlePaddle team, focusing on document understanding. It further optimizes the basic model based on PP-DocBee and introduces new data optimization schemes to improve data quality. With only 470,000 data generated using self-developed data synthesis strategy, PP-DocBee2 performs better in Chinese document understanding tasks. In internal business Chinese scenarios, PP-DocBee2 improves by about 11.4% compared to PP-DocBee and also outperforms current popular open and closed-source models of the same scale.</td>
+</tr>
+</table>
+
+<b>Note: The above total scores are the model test results of the internal evaluation set. All images in the internal evaluation set have a resolution (height, width) of (1680,1204), with a total of 1196 data, including scenarios such as financial reports, laws and regulations, science and engineering papers, manuals, humanities papers, contracts, and research reports, with no plans to make it public for now.</b>
+</details>
+
+<br />
+<b>If you focus more on model accuracy, choose a model with higher accuracy; if you care more about inference speed, choose a model with faster inference speed; if you are concerned about storage size, choose a model with a smaller storage volume.</b>
+
+## 2. Quick Start
+
+Before using the document understanding pipeline locally, ensure that you have completed the installation of the wheel package according to the [installation tutorial](../installation.en.md). If you prefer to install dependencies selectively, please refer to the relevant instructions in the installation documentation. The corresponding dependency group for this pipeline is <code>doc-parser</code>. After installation, you can experience it locally using the command line or Python integration.
+
+Please note: If you encounter issues such as the program becoming unresponsive, unexpected program termination, running out of memory resources, or extremely slow inference during execution, please try adjusting the configuration according to the documentation, such as disabling unnecessary features or using lighter-weight models.
+
+### 2.1 Command Line Experience
+
+Experience the doc_understanding pipeline with just one command line:
+
+```bash
+paddleocr doc_understanding -i "{'image': 'https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/medal_table.png', 'query': '识别这份表格的内容，以markdown格式输出'}"
+```
+
+<details><summary><b>The command line supports more parameter settings, click to expand for a detailed explanation of the command line parameters</b></summary>
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted,  required.<br/>
+<b>Description:</b>
+"{'image': 'https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/medal_table.png', 'query': 'Recognize the content of this table and output it in markdown format'}".
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td><b>Meaning:</b>Specify the path for saving the inference result file.<br/>
+<b>Description:</b> 
+If not set, the inference result will not be saved locally.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_understanding_model_name</code></td>
+<td><b>Meaning:</b>The name of the document understanding model.<br/>
+<b>Description:</b> 
+If not set, the default model of the pipeline will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_understanding_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the document understanding model.<br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_understanding_batch_size</code></td>
+<td><b>Meaning:</b>The batch size of the document understanding model.<br/>
+<b>Description:</b> 
+If not set, the default batch size will be set to <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>The device used for inference.<br/> 
+<b>Description:</b> 
+Supports specifying a specific card number:
+<ul>
+<li><b>CPU</b>: For example, <code>cpu</code> indicates using the CPU for inference;</li>
+<li><b>GPU</b>: For example, <code>gpu:0</code> indicates using the first GPU for inference;</li>
+<li><b>NPU</b>: For example, <code>npu:0</code> indicates using the first NPU for inference;</li>
+<li><b>XPU</b>: For example, <code>xpu:0</code> indicates using the first XPU for inference;</li>
+<li><b>MLU</b>: For example, <code>mlu:0</code> indicates using the first MLU for inference;</li>
+<li><b>DCU</b>: For example, <code>dcu:0</code> indicates using the first DCU for inference;</li>
+<li><b>MetaX GPU</b>: For example, <code>metax_gpu:0</code> indicates using the first MetaX GPU for inference;</li>
+<li><b>Iluvatar GPU</b>: For example, <code>iluvatar_gpu:0</code> indicates using the first Iluvatar GPU for inference;</li>
+</ul>If not set,  the pipeline initialized value for this parameter will be used. During initialization, the local GPU device 0 will be preferred; if unavailable, the CPU device will be used.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, and <code>paddle_dynamic</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to enable the TensorRT subgraph engine of Paddle Inference.<br/>
+<b>Description:</b>
+If the model does not support TensorRT acceleration, acceleration will not be used even if this flag is set.<br/>
+For CUDA 11.8 versions of PaddlePaddle, the compatible TensorRT version is 8.x (x>=6). TensorRT 8.6.1.6 is recommended.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b> Computation precision, such as <code>"fp32"</code> or <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b> Whether to enable MKL-DNN accelerated inference.<br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support MKL-DNN acceleration, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b> MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads used for inference on CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b> Path to the PaddleX pipeline configuration file.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+</tbody>
+</table>
+</details>
+<br />
+
+The results will be printed to the terminal, and the default configuration of the doc_understanding pipeline will produce the following output:
+
+```bash
+{'res': {'image': 'https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/medal_table.png', 'query': '识别这份表格的内容，以markdown格式输出', 'result': '| 名次 | 国家/地区 | 金牌 | 银牌 | 铜牌 | 奖牌总数 |\n| --- | --- | --- | --- | --- | --- |\n| 1 | 中国（CHN） | 48 | 22 | 30 | 100 |\n| 2 | 美国（USA） | 36 | 39 | 37 | 112 |\n| 3 | 俄罗斯（RUS） | 24 | 13 | 23 | 60 |\n| 4 | 英国（GBR） | 19 | 13 | 19 | 51 |\n| 5 | 德国（GER） | 16 | 11 | 14 | 41 |\n| 6 | 澳大利亚（AUS） | 14 | 15 | 17 | 46 |\n| 7 | 韩国（KOR） | 13 | 11 | 8 | 32 |\n| 8 | 日本（JPN） | 9 | 8 | 8 | 25 |\n| 9 | 意大利（ITA） | 8 | 9 | 10 | 27 |\n| 10 | 法国（FRA） | 7 | 16 | 20 | 43 |\n| 11 | 荷兰（NED） | 7 | 5 | 4 | 16 |\n| 12 | 乌克兰（UKR） | 7 | 4 | 11 | 22 |\n| 13 | 肯尼亚（KEN） | 6 | 4 | 6 | 16 |\n| 14 | 西班牙（ESP） | 5 | 11 | 3 | 19 |\n| 15 | 牙买加（JAM） | 5 | 4 | 2 | 11 |\n'}}
+```
+
+### 2.2 Python Script Integration
+
+The command line method is for quickly experiencing the effect. Generally, in projects, code integration is often required. You can complete quick inference of the pipeline with just a few lines of code. The inference code is as follows:
+
+```python
+from paddleocr import DocUnderstanding
+
+pipeline = DocUnderstanding()
+output = pipeline.predict(
+    {
+        "image": "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/medal_table.png",
+        "query": "识别这份表格的内容，以markdown格式输出"
+    }
+)
+for res in output:
+    res.print() ## Print the structured output of the prediction
+    res.save_to_json("./output/")
+```
+
+In the above Python script, the following steps are performed:
+
+(1) Instantiate a Document Understanding Pipeline object through <code>DocUnderstanding()</code>. The specific parameter descriptions are as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>doc_understanding_model_name</code></td>
+<td><b>Meaning:</b>The name of the document understanding model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default model of the pipeline will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_understanding_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the document understanding model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_understanding_batch_size</code></td>
+<td><b>Meaning:</b>The batch size of the document understanding model.<br/> 
+<b>Description:</b> 
+If set to <code>None</code>, the default batch size will be set to <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>The device used for inference. <br/>
+<b>Description:</b> 
+Supports specifying a specific card number:
+<ul>
+<li><b>CPU</b>: For example, <code>cpu</code> indicates using the CPU for inference;</li>
+<li><b>GPU</b>: For example, <code>gpu:0</code> indicates using the first GPU for inference;</li>
+<li><b>NPU</b>: For example, <code>npu:0</code> indicates using the first NPU for inference;</li>
+<li><b>XPU</b>: For example, <code>xpu:0</code> indicates using the first XPU for inference;</li>
+<li><b>MLU</b>: For example, <code>mlu:0</code> indicates using the first MLU for inference;</li>
+<li><b>DCU</b>: For example, <code>dcu:0</code> indicates using the first DCU for inference;</li>
+<li><b>MetaX GPU</b>: For example, <code>metax_gpu:0</code> indicates using the first MetaX GPU for inference;</li>
+<li><b>Iluvatar GPU</b>: For example, <code>iluvatar_gpu:0</code> indicates using the first Iluvatar GPU for inference;</li>
+<li><b>None</b>: If set to <code>None</code>,  the pipeline initialized value for this parameter will be used. During initialization, the local GPU device 0 will be preferred; if unavailable, the CPU device will be used.</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, and <code>paddle_dynamic</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b>Path to PaddleX pipeline configuration file.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+(2) Call the <code>predict()</code> method of the Document Understanding Pipeline object for inference prediction, which will return a result list.
+
+Additionally, the pipeline also provides a <code>predict_iter()</code> method. Both methods are consistent in terms of parameter acceptance and result return. The difference is that <code>predict_iter()</code> returns a <code>generator</code> that can process and obtain prediction results step by step, suitable for handling large datasets or scenarios where memory saving is desired. You can choose to use either method according to your actual needs.
+
+Below are the parameters and their descriptions for the <code>predict()</code> method:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted, currently only supports dictionary type input<br/>
+<b>Description:</b>
+<ul>
+  <li><b>Python Dict</b>: The input format for PP-DocBee is: <code>{"image":/path/to/image, "query": user question}</code>, representing the input image and corresponding user question.</li>
+</ul>
+</td>
+<td><code>Python Dict</code></td>
+<td></td>
+</tr>
+
+</table>
+
+(3) Process the prediction results. The prediction result for each sample is a corresponding Result object, which supports printing and saving as a <code>json</code> file:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Description</th>
+<th>Parameter</th>
+<th>Type</th>
+<th>Parameter Description</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">Print the result to the terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output content using <code>JSON</code> indentation.</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specifies the indentation level to beautify the output <code>JSON</code> data, making it more readable, effective only when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Controls whether to escape non-<code>ASCII</code> characters into <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> will retain the original characters, effective only when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">Save the result as a JSON format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The path to save the file. When specified as a directory, the saved file is named consistent with the input file type.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specifies the indentation level to beautify the output <code>JSON</code> data, making it more readable, effective only when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Controls whether to escape non-<code>ASCII</code> characters into <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> will retain the original characters, effective only when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+</table>
+
+- Calling the `print()` method will print the result to the terminal. The content printed to the terminal is explained as follows:
+
+    - `image`: `(str)` Input path of the image
+
+    - `query`: `(str)` Question regarding the input image
+
+    - `result`: `(str)` Output result of the model
+
+- Calling the `save_to_json()` method will save the above content to the specified `save_path`. If specified as a directory, the path saved will be `save_path/{your_img_basename}_res.json`, and if specified as a file, it will be saved directly to that file.
+
+* Additionally, the result can be obtained through attributes that provide the visualized images with results and the prediction results, as follows:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">Get the prediction result in <code>json</code> format</td>
+</tr>
+<tr>
+<td rowspan = "2"><code>img</code></td>
+<td rowspan = "2">Get the visualized image in <code>dict</code> format</td>
+</tr>
+</table>
+
+- The prediction result obtained through the `json` attribute is data of the dict type, consistent with the content saved by calling the `save_to_json()` method.
+
+## 3. Development Integration/Deployment
+
+If the pipeline meets your requirements for pipeline inference speed and accuracy, you can proceed with development integration/deployment directly.
+
+If you need to apply the pipeline directly to your Python project, you can refer to the example code in [2.2 Python Script Integration](#22-python-script-integration).
+
+In addition, PaddleOCR also provides two other deployment methods, detailed descriptions are as follows:
+
+🚀 High-Performance Inference: In real production environments, many applications have strict standards for the performance indicators of deployment strategies (especially response speed) to ensure efficient system operation and smooth user experience. To this end, PaddleOCR provides high-performance inference capabilities, aiming to deeply optimize the performance of model inference and pre-and post-processing, achieving significant acceleration of the end-to-end process. For detailed high-performance inference processes, refer to [High-Performance Inference](../inference_deployment/local_inference/high_performance_inference.md).
+
+☁️ Service Deployment: Service deployment is a common form of deployment in real production environments. By encapsulating inference functions as services, clients can access these services through network requests to obtain inference results. For detailed pipeline service deployment processes, refer to [Serving](../inference_deployment/serving/serving.md).
+
+Below is the API reference for basic service deployment and examples of service invocation in multiple languages:
+
+<details><summary>API Reference</summary>
+
+<p>For the main operations provided by the service:</p>
+<ul>
+<li>The HTTP request method is POST.</li>
+<li>Both the request body and response body are JSON data (JSON object).</li>
+<li>When the request is processed successfully, the response status code is <code>200</code>, and the response body has the following attributes:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>UUID of the request.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Fixed as <code>0</code>.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error description. Fixed as <code>"Success"</code>.</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>Operation result.</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is not processed successfully, the response body has the following attributes:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>UUID of the request.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Same as the response status code.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error description.</td>
+</tr>
+</tbody>
+</table>
+<p>The main operations provided by the service are as follows:</p>
+<ul>
+<li><b><code>infer</code></b></li>
+</ul>
+<p>Perform inference on the input message to generate a response.</p>
+<p><code>POST /document-understanding</code></p>
+<p>Note: The above interface is also known as /chat/completion, compatible with OpenAI interfaces.</p>
+
+<ul>
+<li>The attributes of the request body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+<th>Required</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model</code></td>
+<td><code>string</code></td>
+<td>The name of the model to use</td>
+<td>Yes</td>
+<td>-</td>
+</tr>
+<tr>
+<td><code>messages</code></td>
+<td><code>array</code></td>
+<td>List of dialogue messages</td>
+<td>Yes</td>
+<td>-</td>
+</tr>
+<tr>
+<td><code>max_tokens</code></td>
+<td><code>integer</code></td>
+<td>Maximum number of tokens to generate</td>
+<td>No</td>
+<td>1024</td>
+</tr>
+<tr>
+<td><code>temperature</code></td>
+<td><code>float</code></td>
+<td>Sampling temperature</td>
+<td>No</td>
+<td>0.1</td>
+</tr>
+<tr>
+<td><code>top_p</code></td>
+<td><code>float</code></td>
+<td>Core sampling probability</td>
+<td>No</td>
+<td>0.95</td>
+</tr>
+<tr>
+<td><code>stream</code></td>
+<td><code>boolean</code></td>
+<td>Whether to output in streaming mode</td>
+<td>No</td>
+<td>false</td>
+</tr>
+<tr>
+<td><code>max_image_tokens</code></td>
+<td><code>int</code></td>
+<td>Maximum number of input tokens for images</td>
+<td>No</td>
+<td>None</td>
+</tr>
+</tbody>
+</table>
+
+<p>Each element in <code>messages</code> is an <code>object</code> with the following attributes:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+<th>Required</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>role</code></td>
+<td><code>string</code></td>
+<td>Message role (user/assistant/system)</td>
+<td>Yes</td>
+</tr>
+<tr>
+<td><code>content</code></td>
+<td><code>string</code> or <code>array</code></td>
+<td>Message content (text or mixed media)</td>
+<td>Yes</td>
+</tr>
+</tbody>
+</table>
+
+<p>When <code>content</code> is an array, each element is an <code>object</code> with the following attributes:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+<th>Required</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>type</code></td>
+<td><code>string</code></td>
+<td>Content type (text/image_url)</td>
+<td>Yes</td>
+<td>-</td>
+</tr>
+<tr>
+<td><code>text</code></td>
+<td><code>string</code></td>
+<td>Text content (when type is text)</td>
+<td>Conditionally required</td>
+<td>-</td>
+</tr>
+<tr>
+<td><code>image_url</code></td>
+<td><code>string</code> or <code>object</code></td>
+<td>Image URL or object (when type is image_url)</td>
+<td>Conditionally required</td>
+<td>-</td>
+</tr>
+</tbody>
+</table>
+
+<p>When <code>image_url</code> is an object, it has the following attributes:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+<th>Required</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>url</code></td>
+<td><code>string</code></td>
+<td>Image URL</td>
+<td>Yes</td>
+<td>-</td>
+</tr>
+<tr>
+<td><code>detail</code></td>
+<td><code>string</code></td>
+<td>Image detail processing method (low/high/auto)</td>
+<td>No</td>
+<td>auto</td>
+</tr>
+</tbody>
+</table>
+
+<p>When the request is processed successfully, the <code>result</code> in the response body has the following attributes:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>id</code></td>
+<td><code>string</code></td>
+<td>Request ID</td>
+</tr>
+<tr>
+<td><code>object</code></td>
+<td><code>string</code></td>
+<td>Object type (chat.completion)</td>
+</tr>
+<tr>
+<td><code>created</code></td>
+<td><code>integer</code></td>
+<td>Creation timestamp</td>
+</tr>
+<tr>
+<td><code>choices</code></td>
+<td><code>array</code></td>
+<td>Generated result options</td>
+</tr>
+<tr>
+<td><code>usage</code></td>
+<td><code>object</code></td>
+<td>Token usage</td>
+</tr>
+</tbody>
+</table>
+
+<p>Each element in <code>choices</code> is a <code>Choice</code> object with the following attributes:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+<th>Optional Values</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>finish_reason</code></td>
+<td><code>string</code></td>
+<td>Reason for the model to stop generating tokens</td>
+<td><code>stop</code> (natural stop)<br><code>length</code> (reached max token count)<br><code>tool_calls</code> (called a tool)<br><code>content_filter</code> (content filtering)<br><code>function_call</code> (called a function, deprecated)</td>
+</tr>
+<tr>
+<td><code>index</code></td>
+<td><code>integer</code></td>
+<td>Index of the option in the list</td>
+<td>-</td>
+</tr>
+<tr>
+<td><code>logprobs</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Log probability information of the option</td>
+<td>-</td>
+</tr>
+<tr>
+<td><code>message</code></td>
+<td><code>ChatCompletionMessage</code></td>
+<td>Chat message generated by the model</td>
+<td>-</td>
+</tr>
+</tbody>
+</table>
+
+<p>The <code>message</code> object has the following attributes:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+<th>Remarks</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>content</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Message content</td>
+<td>May be empty</td>
+</tr>
+<tr>
+<td><code>refusal</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Refusal message generated by the model</td>
+<td>Provided when content is refused</td>
+</tr>
+<tr>
+<td><code>role</code></td>
+<td><code>string</code></td>
+<td>Role of the message author</td>
+<td>Fixed as <code>"assistant"</code></td>
+</tr>
+<tr>
+<td><code>audio</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Audio output data</td>
+<td>Provided when audio output is requested<br><a href="https://platform.openai.com/docs/guides/audio">Learn more</a></td>
+</tr>
+<tr>
+<td><code>function_call</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Name and parameters of the function to be called</td>
+<td>Deprecated, recommended to use <code>tool_calls</code></td>
+</tr>
+<tr>
+<td><code>tool_calls</code></td>
+<td><code>array</code> | <code>null</code></td>
+<td>Tool calls generated by the model</td>
+<td>Such as function calls</td>
+</tr>
+</tbody>
+</table>
+
+<p>The <code>usage</code> object has the following attributes:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prompt_tokens</code></td>
+<td><code>integer</code></td>
+<td>Number of prompt tokens</td>
+</tr>
+<tr>
+<td><code>completion_tokens</code></td>
+<td><code>integer</code></td>
+<td>Number of generated tokens</td>
+</tr>
+<tr>
+<td><code>total_tokens</code></td>
+<td><code>integer</code></td>
+<td>Total number of tokens</td>
+</tr>
+</tbody>
+</table>
+<p>An example of a <code>result</code> is as follows:</p>
+<pre><code class="language-json">{
+    "id": "ed960013-eb19-43fa-b826-3c1b59657e35",
+    "choices": [
+        {
+            "finish_reason": "stop",
+            "index": 0,
+            "message": {
+                "content": "| 名次 | 国家/地区 | 金牌 | 银牌 | 铜牌 | 奖牌总数 |\n| --- | --- | --- | --- | --- | --- |\n| 1 | 中国（CHN） | 48 | 22 | 30 | 100 |\n| 2 | 美国（USA） | 36 | 39 | 37 | 112 |\n| 3 | 俄罗斯（RUS） | 24 | 13 | 23 | 60 |\n| 4 | 英国（GBR） | 19 | 13 | 19 | 51 |\n| 5 | 德国（GER） | 16 | 11 | 14 | 41 |\n| 6 | 澳大利亚（AUS） | 14 | 15 | 17 | 46 |\n| 7 | 韩国（KOR） | 13 | 11 | 8 | 32 |\n| 8 | 日本（JPN） | 9 | 8 | 8 | 25 |\n| 9 | 意大利（ITA） | 8 | 9 | 10 | 27 |\n| 10 | 法国（FRA） | 7 | 16 | 20 | 43 |\n| 11 | 荷兰（NED） | 7 | 5 | 4 | 16 |\n| 12 | 乌克兰（UKR） | 7 | 4 | 11 | 22 |\n| 13 | 肯尼亚（KEN） | 6 | 4 | 6 | 16 |\n| 14 | 西班牙（ESP） | 5 | 11 | 3 | 19 |\n| 15 | 牙买加（JAM） | 5 | 4 | 2 | 11 |\n",
+                "role": "assistant"
+            }
+        }
+    ],
+    "created": 1745218041,
+    "model": "pp-docbee",
+    "object": "chat.completion"
+}
+</code></pre></details>
+
+<details><summary>Multi-language Service Invocation Examples</summary>
+
+<details>
+<summary>Python</summary>
+OpenAI interface invocation example
+
+<pre><code class="language-python">import base64
+from openai import OpenAI
+
+API_BASE_URL = "http://127.0.0.1:8080"
+
+# Initialize OpenAI client
+client = OpenAI(
+    api_key='xxxxxxxxx',
+    base_url=f'{API_BASE_URL}'
+)
+
+# Function to convert image to base64
+def encode_image(image_path):
+  with open(image_path, "rb") as image_file:
+    return base64.b64encode(image_file.read()).decode('utf-8')
+
+# Input image path
+image_path = "medal_table.png"
+
+# Convert original image to base64
+base64_image = encode_image(image_path)
+
+# Submit information to PP-DocBee model
+response = client.chat.completions.create(
+    model="pp-docbee",# Choose Model
+    messages=[
+        {
+            "role": "system",
+            "content": "You are a helpful assistant."
+        },
+        {
+            "role": "user",
+            "content":[
+                {
+                    "type": "text",
+                    "text": "识别这份表格的内容,输出html格式的内容"
+                },
+                {
+                    "type": "image_url",
+                    "image_url": {"url": f"data:image/jpeg;base64,{base64_image}"}
+                },
+            ]
+        },
+    ],
+)
+content = response.choices[0].message.content
+print('Reply:', content)
+</code></pre></details>
+</details>
+<br/>
+
+## 4. Secondary Development
+
+The current pipeline does not support fine-tuning training and only supports inference integration. Concerning fine-tuning training for this pipeline, there are plans to support it in the future.
diff --git a/docs/version3.x/pipeline_usage/doc_understanding.md b/docs/version3.x/pipeline_usage/doc_understanding.md
new file mode 100644
index 0000000..2a461bc
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/doc_understanding.md
@@ -0,0 +1,895 @@
+---
+comments: true
+---
+
+# 文档理解产线使用教程
+
+## 1. 文档理解产线介绍
+
+文档理解产线是基于视觉-语言模型（VLM）打造的先进文档处理技术，旨在突破传统文档处理的局限。传统方法依赖固定模板或预定义规则解析文档，而该产线借助VLM的多模态能力，仅需输入文档图片和用户问题，即可通过融合视觉与语言信息，精准回答用户提问。这种技术无需针对特定文档格式预训练，能够灵活应对多样化文档内容，显著提升文档处理的泛化性与实用性，在智能问答、信息提取等场景中具有广阔应用前景。本产线目前暂不支持对VLM模型的二次开发，后续计划支持。
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/pipelines/doc_understanding/doc_understanding.png">
+
+<b>文档理解产线中包含以下1个模块。每个模块均可独立进行训练和推理，并包含多个模型。有关详细信息，请点击相应模块以查看文档。</b>
+
+- [文档类视觉语言模型模块](../module_usage/doc_vlm.md)
+
+在本产线中，您可以根据下方的基准测试数据选择使用的模型。
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。
+
+<details>
+<summary> <b>文档类视觉语言模型模块：</b></summary>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>模型存储大小（GB）</th>
+<th>模型总分</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-DocBee-2B</td><td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBee-2B_infer.tar">推理模型</a></td>
+<td>4.2</td>
+<td>765</td>
+<td rowspan="2">PP-DocBee 是飞桨团队自研的一款专注于文档理解的多模态大模型，在中文文档理解任务上具有卓越表现。该模型通过近 500 万条文档理解类多模态数据集进行微调优化，各种数据集包括了通用VQA类、OCR类、图表类、text-rich文档类、数学和复杂推理类、合成数据类、纯文本数据等，并设置了不同训练数据配比。在学术界权威的几个英文文档理解评测榜单上，PP-DocBee基本都达到了同参数量级别模型的SOTA。在内部业务中文场景类的指标上，PP-DocBee也高于目前的热门开源和闭源模型。</td>
+</tr>
+<tr>
+<td>PP-DocBee-7B</td><td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBee-7B_infer.tar">推理模型</a></td>
+<td>15.8</td>
+<td>-</td>
+</tr>
+<tr>
+<td>PP-DocBee2-3B</td><td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocBee2-3B_infer.tar">推理模型</a></td>
+<td>7.6</td>
+<td>852</td>
+<td>PP-DocBee2 是飞桨团队自研的一款专注于文档理解的多模态大模型，在PP-DocBee的基础上进一步优化了基础模型，并引入了新的数据优化方案，提高了数据质量，使用自研数据合成策略生成的少量的47万数据便使得PP-DocBee2在中文文档理解任务上表现更佳。在内部业务中文场景类的指标上，PP-DocBee2相较于PP-DocBee提升了约11.4%，同时也高于目前的同规模热门开源和闭源模型。</td>
+</tr>
+</table>
+
+<b>注：以上模型总分为内部评估集模型测试结果，内部评估集所有图像分辨率 (height，width) 为 (1680,1204)，共1196条数据，包括了财报、法律法规、理工科论文、说明书、文科论文、合同、研报等场景，暂时未有计划公开。</b>
+</details>
+
+<br />
+<b>如果您更注重模型的精度，请选择精度较高的模型；如果您更在意模型的推理速度，请选择推理速度较快的模型；如果您关注模型的存储大小，请选择存储体积较小的模型。</b>
+
+## 2. 快速开始
+
+在本地使用文档理解产线前，请确保您已经按照[安装教程](../installation.md)完成了wheel包安装。如果您希望选择性安装依赖，请参考安装教程中的相关说明。该产线对应的依赖分组为 `doc-parser`。安装完成后，可以在本地使用命令行体验或 Python 集成。
+
+**请注意，如果在执行过程中遇到程序失去响应、程序异常退出、内存资源耗尽、推理速度极慢等问题，请尝试参考文档调整配置，例如关闭不需要使用的功能或使用更轻量的模型。**
+
+### 2.1 命令行方式体验
+
+一行命令即可快速体验 doc_understanding 产线效果：
+
+```bash
+paddleocr doc_understanding -i "{'image': 'https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/medal_table.png', 'query': '识别这份表格的内容，以markdown格式输出'}"
+```
+
+<details><summary><b>命令行支持更多参数设置，点击展开以查看命令行参数的详细说明</b></summary>
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，必填。<br/>
+<b>说明：</b>
+如<code>{'image': 'https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/medal_table.png', 'query': '识别这份表格的内容，以markdown格式输出'}</code>。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td><b>含义：</b>指定推理结果文件保存的路径。<br/>
+<b>说明：</b>如果不设置，推理结果将不会保存到本地。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_understanding_model_name</code></td>
+<td><b>含义：</b>文档理解模型的名称。<br/>
+<b>说明：</b>如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_understanding_model_dir</code></td>
+<td><b>含义：</b>文档理解模型的目录路径。<br/>
+<b>说明：</b>如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_understanding_batch_size</code></td>
+<td><b>含义：</b>文档理解模型的batch size。<br/>
+<b>说明：</b>如果不设置，将默认设置batch size为<code>1</code>。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+</ul>如果不设置，将默认使用产线初始化的该参数值，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_dynamic</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>fp32</code>、<code>fp16</code>。</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+
+</tbody>
+</table>
+</details>
+<br />
+
+运行结果会被打印到终端上，默认配置的 doc_understanding 产线的运行结果如下：
+
+```bash
+{'res': {'image': 'https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/medal_table.png', 'query': '识别这份表格的内容，以markdown格式输出', 'result': '| 名次 | 国家/地区 | 金牌 | 银牌 | 铜牌 | 奖牌总数 |\n| --- | --- | --- | --- | --- | --- |\n| 1 | 中国（CHN） | 48 | 22 | 30 | 100 |\n| 2 | 美国（USA） | 36 | 39 | 37 | 112 |\n| 3 | 俄罗斯（RUS） | 24 | 13 | 23 | 60 |\n| 4 | 英国（GBR） | 19 | 13 | 19 | 51 |\n| 5 | 德国（GER） | 16 | 11 | 14 | 41 |\n| 6 | 澳大利亚（AUS） | 14 | 15 | 17 | 46 |\n| 7 | 韩国（KOR） | 13 | 11 | 8 | 32 |\n| 8 | 日本（JPN） | 9 | 8 | 8 | 25 |\n| 9 | 意大利（ITA） | 8 | 9 | 10 | 27 |\n| 10 | 法国（FRA） | 7 | 16 | 20 | 43 |\n| 11 | 荷兰（NED） | 7 | 5 | 4 | 16 |\n| 12 | 乌克兰（UKR） | 7 | 4 | 11 | 22 |\n| 13 | 肯尼亚（KEN） | 6 | 4 | 6 | 16 |\n| 14 | 西班牙（ESP） | 5 | 11 | 3 | 19 |\n| 15 | 牙买加（JAM） | 5 | 4 | 2 | 11 |\n'}}
+```
+
+### 2.2 Python脚本方式集成
+
+命令行方式是为了快速体验查看效果，一般来说，在项目中，往往需要通过代码集成，您可以通过几行代码即可完成产线的快速推理，推理代码如下：
+
+```python
+from paddleocr import DocUnderstanding
+
+pipeline = DocUnderstanding()
+output = pipeline.predict(
+    {
+        "image": "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/medal_table.png",
+        "query": "识别这份表格的内容，以markdown格式输出"
+    }
+)
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_json("./output/")
+```
+
+在上述 Python 脚本中，执行了如下几个步骤：
+
+（1）通过 <code>DocUnderstanding()</code> 实例化文档理解产线产线对象，具体参数说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>doc_understanding_model_name</code></td>
+<td><b>含义：</b>文档理解模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_understanding_model_dir</code></td>
+<td><b>含义：</b>文档理解模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_understanding_batch_size</code></td>
+<td><b>含义：</b>文档理解模型的batch size。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_dynamic</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+（2）调用 文档理解产线 产线对象的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。
+
+另外，产线还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。
+
+以下是 <code>predict()</code> 方法的参数及其说明：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，目前仅支持dict类型的输入。<br/>
+<b>说明：</b>   
+<ul>
+  <li><b>Python Dict</b>：如PP-DocBee的输入形式为: <code>{"image":/path/to/image, "query": user question}</code> ,分别表示输入的图像和对应的用户问题。</li>
+</ul>
+</td>
+<td><code>Python Dict</code></td>
+<td></td>
+</tr>
+</table>
+
+（3）对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为 <code>json</code> 文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "3"><code>print()</code></td>
+<td rowspan = "3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化。</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效。</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效。</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan = "3"><code>save_to_json()</code></td>
+<td rowspan = "3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致。</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效。</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效。</td>
+<td><code>False</code></td>
+</tr>
+</table>
+
+- 调用`print()` 方法会将结果打印到终端，打印到终端的内容解释如下：
+
+    - `image`: `(str)` 图像的输入路径
+
+    - `query`: `(str)` 针对输入图像的问题
+
+    - `result`: `(str)` 模型的输出结果
+
+- 调用`save_to_json()` 方法会将上述内容保存到指定的`save_path`中，如果指定为目录，则保存的路径为`save_path/{your_img_basename}_res.json`，如果指定为文件，则直接保存到该文件中。
+
+* 此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan = "1"><code>json</code></td>
+<td rowspan = "1">获取预测的 <code>json</code> 格式的结果</td>
+</tr>
+<tr>
+<td rowspan = "2"><code>img</code></td>
+<td rowspan = "2">获取格式为 <code>dict</code> 的可视化图像</td>
+</tr>
+</table>
+
+- `json` 属性获取的预测结果为dict类型的数据，相关内容与调用 `save_to_json()` 方法保存的内容一致。
+
+## 3. 开发集成/部署
+
+如果产线可以达到您对产线推理速度和精度的要求，您可以直接进行开发集成/部署。
+
+若您需要将产线直接应用在您的Python项目中，可以参考 [2.2 Python脚本方式](#22-python) 中的示例代码。
+
+此外，PaddleOCR 也提供了其他两种部署方式，详细说明如下：
+
+🚀 高性能推理：在实际生产环境中，许多应用对部署策略的性能指标（尤其是响应速度）有着较严苛的标准，以确保系统的高效运行与用户体验的流畅性。为此，PaddleOCR 提供高性能推理功能，旨在对模型推理及前后处理进行深度性能优化，实现端到端流程的显著提速，详细的高性能推理流程请参考[高性能推理](../inference_deployment/local_inference/high_performance_inference.md)。
+
+☁️ 服务化部署：服务化部署是实际生产环境中常见的一种部署形式。通过将推理功能封装为服务，客户端可以通过网络请求来访问这些服务，以获取推理结果。详细的产线服务化部署流程请参考[服务化部署](../inference_deployment/serving/serving.md)。
+
+以下是基础服务化部署的API参考与多语言服务调用示例：
+
+<details><summary>API参考</summary>
+
+<p>对于服务提供的主要操作：</p>
+<ul>
+<li>HTTP请求方法为POST。</li>
+<li>请求体和响应体均为JSON数据（JSON对象）。</li>
+<li>当请求处理成功时，响应状态码为<code>200</code>，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。固定为<code>0</code>。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。固定为<code>"Success"</code>。</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>操作结果。</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>当请求处理未成功时，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。与响应状态码相同。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。</td>
+</tr>
+</tbody>
+</table>
+<p>服务提供的主要操作如下：</p>
+<ul>
+<li><b><code>infer</code></b></li>
+</ul>
+<p>对输入消息进行推理生成响应。</p>
+<p><code>POST /document-understanding</code></p>
+<p>说明 以上接口别名/chat/completion，openai兼容的接口</p>
+
+<ul>
+<li>请求体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>model</code></td>
+<td><code>string</code></td>
+<td>要使用的模型名称</td>
+<td>是</td>
+<td>-</td>
+</tr>
+<tr>
+<td><code>messages</code></td>
+<td><code>array</code></td>
+<td>对话消息列表</td>
+<td>是</td>
+<td>-</td>
+</tr>
+<tr>
+<td><code>max_tokens</code></td>
+<td><code>integer</code></td>
+<td>生成的最大token数</td>
+<td>否</td>
+<td>1024</td>
+</tr>
+<tr>
+<td><code>temperature</code></td>
+<td><code>float</code></td>
+<td>采样温度</td>
+<td>否</td>
+<td>0.1</td>
+</tr>
+<tr>
+<td><code>top_p</code></td>
+<td><code>float</code></td>
+<td>核心采样概率</td>
+<td>否</td>
+<td>0.95</td>
+</tr>
+<tr>
+<td><code>stream</code></td>
+<td><code>boolean</code></td>
+<td>是否流式输出</td>
+<td>否</td>
+<td>false</td>
+</tr>
+<tr>
+<td><code>max_image_tokens</code></td>
+<td><code>int</code></td>
+<td>图像的最大输入token数</td>
+<td>否</td>
+<td>None</td>
+</tr>
+</tbody>
+</table>
+
+<p><code>messages</code>中的每个元素为一个<code>object</code>，具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>role</code></td>
+<td><code>string</code></td>
+<td>消息角色（user/assistant/system）</td>
+<td>是</td>
+</tr>
+<tr>
+<td><code>content</code></td>
+<td><code>string</code>或<code>array</code></td>
+<td>消息内容（文本或图文混合）</td>
+<td>是</td>
+</tr>
+</tbody>
+</table>
+
+<p>当<code>content</code>为数组时，每个元素为一个<code>object</code>，具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>type</code></td>
+<td><code>string</code></td>
+<td>内容类型（text/image_url）</td>
+<td>是</td>
+<td>-</td>
+</tr>
+<tr>
+<td><code>text</code></td>
+<td><code>string</code></td>
+<td>文本内容（当type为text时）</td>
+<td>条件必填</td>
+<td>-</td>
+</tr>
+<tr>
+<td><code>image_url</code></td>
+<td><code>string</code>或<code>object</code></td>
+<td>图片URL或对象（当type为image_url时）</td>
+<td>条件必填</td>
+<td>-</td>
+</tr>
+</tbody>
+</table>
+
+<p>当<code>image_url</code>为对象时，具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>url</code></td>
+<td><code>string</code></td>
+<td>图片URL</td>
+<td>是</td>
+<td>-</td>
+</tr>
+<tr>
+<td><code>detail</code></td>
+<td><code>string</code></td>
+<td>图片细节处理方式（low/high/auto）</td>
+<td>否</td>
+<td>auto</td>
+</tr>
+</tbody>
+</table>
+
+<p>请求处理成功时，响应体的<code>result</code>具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>id</code></td>
+<td><code>string</code></td>
+<td>请求ID</td>
+</tr>
+<tr>
+<td><code>object</code></td>
+<td><code>string</code></td>
+<td>对象类型（chat.completion）</td>
+</tr>
+<tr>
+<td><code>created</code></td>
+<td><code>integer</code></td>
+<td>创建时间戳</td>
+</tr>
+<tr>
+<td><code>choices</code></td>
+<td><code>array</code></td>
+<td>生成结果选项</td>
+</tr>
+<tr>
+<td><code>usage</code></td>
+<td><code>object</code></td>
+<td>token使用情况</td>
+</tr>
+</tbody>
+</table>
+
+<p><code>choices</code>中的每个元素为一个<code>Choice</code>对象，具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>可选值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>finish_reason</code></td>
+<td><code>string</code></td>
+<td>模型停止生成token的原因</td>
+<td><code>stop</code>（自然停止）<br><code>length</code>（达到最大token数）<br><code>tool_calls</code>（调用了工具）<br><code>content_filter</code>（内容过滤）<br><code>function_call</code>（调用了函数，已弃用）</td>
+</tr>
+<tr>
+<td><code>index</code></td>
+<td><code>integer</code></td>
+<td>选项在列表中的索引</td>
+<td>-</td>
+</tr>
+<tr>
+<td><code>logprobs</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>选项的log概率信息</td>
+<td>-</td>
+</tr>
+<tr>
+<td><code>message</code></td>
+<td><code>ChatCompletionMessage</code></td>
+<td>模型生成的聊天消息</td>
+<td>-</td>
+</tr>
+</tbody>
+</table>
+
+<p><code>message</code>对象具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>备注</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>content</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>消息内容</td>
+<td>可能为空</td>
+</tr>
+<tr>
+<td><code>refusal</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>模型生成的拒绝消息</td>
+<td>当内容被拒绝时提供</td>
+</tr>
+<tr>
+<td><code>role</code></td>
+<td><code>string</code></td>
+<td>消息作者角色</td>
+<td>固定为<code>"assistant"</code></td>
+</tr>
+<tr>
+<td><code>audio</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>音频输出数据</td>
+<td>当请求音频输出时提供<br><a href="https://platform.openai.com/docs/guides/audio">了解更多</a></td>
+</tr>
+<tr>
+<td><code>function_call</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>应调用的函数名称和参数</td>
+<td>已弃用，推荐使用<code>tool_calls</code></td>
+</tr>
+<tr>
+<td><code>tool_calls</code></td>
+<td><code>array</code> | <code>null</code></td>
+<td>模型生成的工具调用</td>
+<td>如函数调用等</td>
+</tr>
+</tbody>
+</table>
+
+<p><code>usage</code>对象具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prompt_tokens</code></td>
+<td><code>integer</code></td>
+<td>提示token数</td>
+</tr>
+<tr>
+<td><code>completion_tokens</code></td>
+<td><code>integer</code></td>
+<td>生成token数</td>
+</tr>
+<tr>
+<td><code>total_tokens</code></td>
+<td><code>integer</code></td>
+<td>总token数</td>
+</tr>
+</tbody>
+</table>
+<p><code>result</code>示例如下：</p>
+<pre><code class="language-json">{
+    "id": "ed960013-eb19-43fa-b826-3c1b59657e35",
+    "choices": [
+        {
+            "finish_reason": "stop",
+            "index": 0,
+            "message": {
+                "content": "| 名次 | 国家/地区 | 金牌 | 银牌 | 铜牌 | 奖牌总数 |\n| --- | --- | --- | --- | --- | --- |\n| 1 | 中国（CHN） | 48 | 22 | 30 | 100 |\n| 2 | 美国（USA） | 36 | 39 | 37 | 112 |\n| 3 | 俄罗斯（RUS） | 24 | 13 | 23 | 60 |\n| 4 | 英国（GBR） | 19 | 13 | 19 | 51 |\n| 5 | 德国（GER） | 16 | 11 | 14 | 41 |\n| 6 | 澳大利亚（AUS） | 14 | 15 | 17 | 46 |\n| 7 | 韩国（KOR） | 13 | 11 | 8 | 32 |\n| 8 | 日本（JPN） | 9 | 8 | 8 | 25 |\n| 9 | 意大利（ITA） | 8 | 9 | 10 | 27 |\n| 10 | 法国（FRA） | 7 | 16 | 20 | 43 |\n| 11 | 荷兰（NED） | 7 | 5 | 4 | 16 |\n| 12 | 乌克兰（UKR） | 7 | 4 | 11 | 22 |\n| 13 | 肯尼亚（KEN） | 6 | 4 | 6 | 16 |\n| 14 | 西班牙（ESP） | 5 | 11 | 3 | 19 |\n| 15 | 牙买加（JAM） | 5 | 4 | 2 | 11 |\n",
+                "role": "assistant"
+            }
+        }
+    ],
+    "created": 1745218041,
+    "model": "pp-docbee",
+    "object": "chat.completion"
+}
+</code></pre></details>
+
+<details><summary>多语言调用服务示例</summary>
+
+<details>
+<summary>Python</summary>
+openai接口调用示例
+
+<pre><code class="language-python">import base64
+from openai import OpenAI
+
+API_BASE_URL = "http://127.0.0.1:8080"
+
+# 初始化OpenAI客户端
+client = OpenAI(
+    api_key='xxxxxxxxx',
+    base_url=f'{API_BASE_URL}'
+)
+
+#图片转base64函数
+def encode_image(image_path):
+  with open(image_path, "rb") as image_file:
+    return base64.b64encode(image_file.read()).decode('utf-8')
+
+#输入图片路径
+image_path = "medal_table.png"
+
+#原图片转base64
+base64_image = encode_image(image_path)
+
+#提交信息至PP-DocBee模型
+response = client.chat.completions.create(
+    model="pp-docbee",#选择模型
+    messages=[
+        {
+            "role": "system",
+            "content": "You are a helpful assistant."
+        },
+        {
+            "role": "user",
+            "content":[
+                {
+                    "type": "text",
+                    "text": "识别这份表格的内容,输出html格式的内容"
+                },
+                {
+                    "type": "image_url",
+                    "image_url": {"url": f"data:image/jpeg;base64,{base64_image}"}
+                },
+            ]
+        },
+    ],
+)
+content = response.choices[0].message.content
+print('Reply:', content)
+</code></pre></details>
+</details>
+<br/>
+
+## 4. 二次开发
+
+当前产线暂时不支持微调训练，仅支持推理集成。关于该产线的微调训练，计划在未来支持。
diff --git a/docs/version3.x/pipeline_usage/formula_recognition.en.md b/docs/version3.x/pipeline_usage/formula_recognition.en.md
new file mode 100644
index 0000000..e67eab1
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/formula_recognition.en.md
@@ -0,0 +1,2110 @@
+---
+comments: true
+---
+
+# Formula Recognition Pipeline Usage Tutorial
+
+## 1. Introduction to Formula Recognition Pipeline
+
+Formula recognition is a technology that automatically identifies and extracts LaTeX formula content and structure from documents or images. It is widely used in fields such as mathematics, physics, and computer science for document editing and data analysis. By using computer vision and machine learning algorithms, formula recognition can convert complex mathematical formula information into editable LaTeX format, facilitating further processing and analysis of data.
+
+The formula recognition pipeline is designed to solve formula recognition tasks by extracting formula information from images and outputting it in LaTeX source code format. This pipeline integrates the advanced formula recognition model PP-FormulaNet developed by the PaddlePaddle Vision Team and the well-known formula recognition model UniMERNet. It is an end-to-end formula recognition system that supports the recognition of simple printed formulas, complex printed formulas, and handwritten formulas. Additionally, it includes functions for image orientation correction and distortion correction. Based on this pipeline, precise formula content prediction can be achieved, covering various application scenarios in education, research, finance, manufacturing, and other fields. The pipeline also provides flexible deployment options, supporting multiple hardware devices and programming languages. Moreover, it offers the capability for custom development. You can train and optimize the pipeline on your own dataset, and the trained model can be seamlessly integrated.
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/formula_recognition/03.png" style="width: 70%"/>
+
+<b>
+The formula recognition pipeline includes the following four modules. Each module can be trained and inferred independently and contains multiple models. For more details, please click on the respective module to view the documentation.</b>
+
+- [Formula Recognition Module](../module_usage/formula_recognition.en.md)
+- [Layout Detection Module](../module_usage/layout_detection.en.md)（Optional）
+- [Document Image Orientation Classification Module](../module_usage/doc_img_orientation_classification.en.md) （Optional）
+- [Text Image Correction Module](../module_usage/text_image_unwarping.en.md) （Optional）
+
+
+In this pipeline, you can choose the model you want to use based on the benchmark data provided below.
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing.
+> In the inference time columns labeled [Normal Mode / High-Performance Mode], the Normal Mode values correspond to the local `paddle_static` inference engine.
+
+<details>
+<summary><b>Document Image Orientation Classification Module (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Top-1 Acc (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">Training Model</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>A document image classification model based on PP-LCNet_x1_0, with four categories: 0 degrees, 90 degrees, 180 degrees, and 270 degrees.</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Text Image Correction Module (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>CER</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">Training Model</a></td>
+<td>0.179</td>
+<td>19.05 / 19.05</td>
+<td>- / 869.82</td>
+<td>30.3</td>
+<td>High-precision text image correction model</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Layout Detection Module (Optional):</b></summary>
+
+* <b>The layout detection model includes 20 common categories: document title, paragraph title, text, page number, abstract, table, references, footnotes, header, footer, algorithm, formula, formula number, image, table, seal, figure_table title, chart, and sidebar text and lists of references</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout_plus-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout_plus-L_pretrained.pdparams">Training Model</a></td>
+<td>83.2</td>
+<td>53.03 / 17.23</td>
+<td>634.62 / 378.32</td>
+<td>126.01</td>
+<td>A higher-precision layout area localization model trained on a self-built dataset containing Chinese and English papers, PPT, multi-layout magazines, contracts, books, exams, ancient books and research reports using RT-DETR-L</td>
+</tr>
+<tr>
+</tbody>
+</table>
+
+* <b>The layout detection model includes 23 common categories: document title, paragraph title, text, page number, abstract, table of contents, references, footnotes, header, footer, algorithm, formula, formula number, image, figure caption, table, table caption, seal, figure title, figure, header image, footer image, and sidebar text</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-L_pretrained.pdparams">Training Model</a></td>
+<td>90.4</td>
+<td>33.59 / 33.59</td>
+<td>503.01 / 251.08</td>
+<td>123.76</td>
+<td>A high-precision layout area localization model trained on a self-built dataset containing Chinese and English papers, magazines, contracts, books, exams, and research reports using RT-DETR-L.</td>
+</tr>
+<tr>
+<td>PP-DocLayout-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-M_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-M_pretrained.pdparams">Training Model</a></td>
+<td>75.2</td>
+<td>13.03 / 4.72</td>
+<td>43.39 / 24.44</td>
+<td>22.578</td>
+<td>A layout area localization model with balanced precision and efficiency, trained on a self-built dataset containing Chinese and English papers, magazines, contracts, books, exams, and research reports using PicoDet-L.</td>
+</tr>
+<tr>
+<td>PP-DocLayout-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-S_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-S_pretrained.pdparams">Training Model</a></td>
+<td>70.9</td>
+<td>11.54 / 3.86</td>
+<td>18.53 / 6.29</td>
+<td>4.834</td>
+<td>A high-efficiency layout area localization model trained on a self-built dataset containing Chinese and English papers, magazines, contracts, books, exams, and research reports using PicoDet-S.</td>
+</tr>
+</tbody>
+</table>
+
+>❗ The above list includes the <b>4 core models</b> that are key supported by the layout detection module. The module actually supports a total of <b>7 full models</b>, including several predefined models with different categories. The complete model list is as follows:
+
+<details><summary> 👉Details of Model List</summary>
+
+* <b>Layout Detection Model, including 17 common layout categories: Paragraph Title, Image, Text, Number, Abstract, Content, Figure Caption, Formula, Table, Table Caption, References, Document Title, Footnote, Header, Algorithm, Footer, and Stamp</b>
+
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>87.4</td>
+<td>8.80 / 3.62</td>
+<td>17.51 / 6.35</td>
+<td>4.8</td>
+<td>A high-efficiency layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using PicoDet-S.</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>89.0</td>
+<td>12.60 / 10.27</td>
+<td>43.70 / 24.42</td>
+<td>22.6</td>
+<td>A balanced efficiency and precision layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using PicoDet-L.</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>98.3</td>
+<td>115.29 / 101.18</td>
+<td>964.75 / 964.75</td>
+<td>470.2</td>
+<td>A high-precision layout area localization model trained on a self-built dataset of Chinese and English papers, magazines, and research reports using RT-DETR-H.</td>
+</tr>
+</tbody></table>
+
+* <b>Layout Detection Model, including 23 common layout categories: Document Title, Section Title, Text, Page Number, Abstract, Table of Contents, References, Footnotes, Header, Footer, Algorithm, Formula, Formula Number, Image, Figure Caption, Table, Table Caption, Seal, Chart Caption, Chart, Header Image, Footer Image, Sidebar Text</b>
+
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-L_pretrained.pdparams">Training Model</a></td>
+<td>90.4</td>
+<td>33.59 / 33.59</td>
+<td>503.01 / 251.08</td>
+<td>123.76</td>
+<td>A high-precision layout area localization model trained on a self-built dataset containing Chinese and English papers, magazines, contracts, books, exams, and research reports using RT-DETR-L.</td>
+</tr>
+<tr>
+<td>PP-DocLayout-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-M_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-M_pretrained.pdparams">Training Model</a></td>
+<td>75.2</td>
+<td>13.03 / 4.72</td>
+<td>43.39 / 24.44</td>
+<td>22.578</td>
+<td>A layout area localization model with balanced precision and efficiency, trained on a self-built dataset containing Chinese and English papers, magazines, contracts, books, exams, and research reports using PicoDet-L.</td>
+</tr>
+<tr>
+<td>PP-DocLayout-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-S_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-S_pretrained.pdparams">Training Model</a></td>
+<td>70.9</td>
+<td>11.54 / 3.86</td>
+<td>18.53 / 6.29</td>
+<td>4.834</td>
+<td>A high-efficiency layout area localization model trained on a self-built dataset containing Chinese and English papers, magazines, contracts, books, exams, and research reports using PicoDet-S.</td>
+</tr>
+</tbody>
+</table>
+
+* <b>The layout detection model includes 20 common categories: document title, paragraph title, text, page number, abstract, table, references, footnotes, header, footer, algorithm, formula, formula number, image, table, seal, figure_table title, chart, and sidebar text and lists of references</b>
+
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout_plus-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout_plus-L_pretrained.pdparams">Training Model</a></td>
+<td>83.2</td>
+<td>53.03 / 17.23</td>
+<td>634.62 / 378.32</td>
+<td>126.01</td>
+<td>A higher-precision layout area localization model trained on a self-built dataset containing Chinese and English papers, PPT, multi-layout magazines, contracts, books, exams, ancient books and research reports using RT-DETR-L</td>
+</tr>
+<tr>
+</tbody>
+</table>
+
+
+</details>
+</details>
+
+<details>
+<summary><b>Formula Recognition Module : </b></summary>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>En-BLEU(%)</th>
+<th>Zh-BLEU(%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+<td>UniMERNet</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UniMERNet_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UniMERNet_pretrained.pdparams">Training Model</a></td>
+<td>85.91</td>
+<td>43.50</td>
+<td>1311.84 / 1311.84</td>
+<td>- / 8288.07</td>
+<td>1530</td>
+<td>UniMERNet is a formula recognition model developed by Shanghai AI Lab. It uses Donut Swin as the encoder and MBartDecoder as the decoder. The model is trained on a dataset of one million samples, including simple formulas, complex formulas, scanned formulas, and handwritten formulas, significantly improving the recognition accuracy of real-world formulas.</td>
+</tr>
+<td>PP-FormulaNet-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-S_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-S_pretrained.pdparams">Training Model</a></td>
+<td>87.00</td>
+<td>45.71</td>
+<td>182.25 / 182.25</td>
+<td>- / 254.39</td>
+<td>224</td>
+<td rowspan="2">PP-FormulaNet is an advanced formula recognition model developed by the Baidu PaddlePaddle Vision Team. The PP-FormulaNet-S version uses PP-HGNetV2-B4 as its backbone network. Through parallel masking and model distillation techniques, it significantly improves inference speed while maintaining high recognition accuracy, making it suitable for applications requiring fast inference. The PP-FormulaNet-L version, on the other hand, uses Vary_VIT_B as its backbone network and is trained on a large-scale formula dataset, showing significant improvements in recognizing complex formulas compared to PP-FormulaNet-S.</td>
+</tr>
+<td>PP-FormulaNet-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-L_pretrained.pdparams">Training Model</a></td>
+<td>90.36</td>
+<td>45.78</td>
+<td>1482.03 / 1482.03</td>
+<td>- / 3131.54</td>
+<td>695</td>
+</tr>
+<td>PP-FormulaNet_plus-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-S_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-S_pretrained.pdparams">Training Model</a></td>
+<td>88.71</td>
+<td>53.32</td>
+<td>179.20 / 179.20</td>
+<td>- / 260.99</td>
+<td>248</td>
+<td rowspan="3">PP-FormulaNet_plus is an enhanced version of the formula recognition model developed by the Baidu PaddlePaddle Vision Team, building upon the original PP-FormulaNet. Compared to the original version, PP-FormulaNet_plus utilizes a more diverse formula dataset during training, including sources such as Chinese dissertations, professional books, textbooks, exam papers, and mathematics journals. This expansion significantly improves the model’s recognition capabilities. Among the models, PP-FormulaNet_plus-M and PP-FormulaNet_plus-L have added support for Chinese formulas and increased the maximum number of predicted tokens for formulas from 1,024 to 2,560, greatly enhancing the recognition performance for complex formulas. Meanwhile, the PP-FormulaNet_plus-S model focuses on improving the recognition of English formulas. With these improvements, the PP-FormulaNet_plus series models perform exceptionally well in handling complex and diverse formula recognition tasks. </td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-M_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-M_pretrained.pdparams">Training Model</a></td>
+<td>91.45</td>
+<td>89.76</td>
+<td>1040.27 / 1040.27</td>
+<td>- / 1615.80</td>
+<td>592</td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-L_pretrained.pdparams">Training Model</a></td>
+<td>92.22</td>
+<td>90.64</td>
+<td>1476.07 / 1476.07</td>
+<td>- / 3125.58</td>
+<td>698</td>
+</tr>
+<tr>
+<td>LaTeX_OCR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/LaTeX_OCR_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/LaTeX_OCR_rec_pretrained.pdparams">Training Model</a></td>
+<td>74.55</td>
+<td>39.96</td>
+<td>1088.89 / 1088.89</td>
+<td>- / -</td>
+<td>99</td>
+<td>LaTeX-OCR is a formula recognition algorithm based on an autoregressive large model. It uses Hybrid ViT as the backbone network and a transformer as the decoder, significantly improving the accuracy of formula recognition.</td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary><b>Test Environment Description: </b></summary>
+
+  <ul>
+      <li><b>Performance Test Environment</b>
+          <ul>
+            <li><strong>Test Dataset: 
+             </strong>
+                <ul>
+                  <li>Document Image Orientation Classification Module: A self-built dataset using PaddleOCR, covering multiple scenarios such as ID cards and documents, containing 1000 images.</li>
+                  <li> Text Image Rectification Module: <a href="https://www3.cs.stonybrook.edu/~cvl/docunet.html">DocUNet</a>。</li>
+                  <li>Layout Region Detection Module: A self-built layout region detection dataset using PaddleOCR, including 500 images of common document types such as Chinese and English papers, magazines, contracts, books, exam papers, and research reports.</li>
+                  <li>17-Class Region Detection Model: A self-built layout region detection dataset using PaddleOCR, including 892 images of common document types such as Chinese and English papers, magazines, and research reports.</li>
+                  <li>Formula Recognition Module: A self-built formula recognition test set using PaddleX.</li>
+                </ul>
+             </li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>             
+          </ul>
+      </li>
+      <li><b>Inference Mode Description</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration </th>
+            <th>CPU Configuration </th>
+            <th>Acceleration Technology Combination</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Normal Mode</td>
+            <td>FP32 Precision / No TRT Acceleration</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td><code>paddle_static</code></td>
+        </tr>
+        <tr>
+            <td>High-Performance Mode</td>
+            <td>Optimal combination of pre-selected precision types and acceleration strategies</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Pre-selected optimal backend (Paddle/OpenVINO/TRT, etc.)</td>
+        </tr>
+    </tbody>
+</table>
+</details>
+<br />
+<b>If you prioritize model accuracy, choose a model with higher precision; if you care more about inference speed, choose a faster model; if you are concerned about model storage size, choose a smaller model.</b>
+
+## 2. Quick Start
+
+Before using the formula recognition pipeline locally, please ensure that you have completed the wheel package installation according to the [installation guide](../installation.en.md). If you prefer to install dependencies selectively, please refer to the relevant instructions in the installation documentation. The corresponding dependency group for this pipeline is <code>doc-parser</code>. Once installed, you can experience it locally via the command line or integrate it with Python.
+
+Please note: If you encounter issues such as the program becoming unresponsive, unexpected program termination, running out of memory resources, or extremely slow inference during execution, please try adjusting the configuration according to the documentation, such as disabling unnecessary features or using lighter-weight models.
+
+### 2.1 Command Line Experience
+
+You can quickly experience the effect of the formula recognition pipeline with one command. Before running the code below, please download the [example image](https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/pipelines/general_formula_recognition_001.png) locally：
+
+```bash
+paddleocr formula_recognition_pipeline -i https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/pipelines/general_formula_recognition_001.png
+
+# Specify whether to use the document orientation classification model with --use_doc_orientation_classify.
+paddleocr formula_recognition_pipeline -i ./general_formula_recognition_001.png --use_doc_orientation_classify True
+
+# Specify whether to use the text image unwarping module with --use_doc_unwarping.
+paddleocr formula_recognition_pipeline -i ./general_formula_recognition_001.png --use_doc_unwarping True
+
+# Specify the use of GPU for model inference with --device.
+paddleocr formula_recognition_pipeline -i ./general_formula_recognition_001.png --device gpu
+```
+
+The examples above use the local `paddle_static` inference engine by default. To run them, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr formula_recognition_pipeline -i https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/pipelines/general_formula_recognition_001.png \
+    --engine transformers
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+<details><summary><b>The command line supports more parameter settings. Click to expand for detailed descriptions of the command line parameters.</b></summary>
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted, required.<br/>
+<b>Description:</b>
+Local path of image or PDF file, e.g., <code>/root/data/img.jpg</code>; <b>URL link</b>, e.g., network URL of image or PDF file: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/pipelines/general_formula_recognition_001.png">Example</a>; <b>Local directory</b>, the directory should contain images to be predicted, e.g., local path: <code>/root/data/</code> (currently does not support prediction of PDF files in directories; PDF files must be specified with a specific file path).
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td>
+<b>Meaning:</b>Specify the path to save the inference results file. <br/>
+<b>Description:</b>
+If not set, the inference results will not be saved locally.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td> 
+<b>Meaning:</b>The name of the document orientation classification model.<br/>
+<b>Description:</b> 
+If not set, the default model in pipeline will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the document orientation classification model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_batch_size</code></td>
+<td><b>Meaning:</b>The batch size of the document orientation classification model. <br/>
+<b>Description:</b> 
+If not set, the default batch size will be set to <code>1</code>.
+</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>Meaning:</b> The name of the text image unwarping model. <br/>
+<b>Description:</b> 
+If not set, the default model in pipeline will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td> <b>Meaning:</b>The directory path of the  text image unwarping model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_batch_size</code></td>
+<td><b>Meaning:</b>The batch size of the text image unwarping model. <br/>
+<b>Description:</b> 
+If not set, the default batch size will be set to <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to load and use the document orientation classification module. <br/>
+<b>Description:</b> 
+If not set, the parameter will be set to the value initialized in the pipeline, which is <code>True</code> by default.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td>
+<b>Meaning:</b>Whether to load and use the text image unwarping module. <br/>
+<b>Description:</b> 
+If not set, the parameter will be set to the value initialized in the pipeline, which is <code>True</code> by default.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td>
+<b>Meaning:</b>The name of the layout detection model. <br/>
+<b>Description:</b> 
+If not set, the default model in pipeline will be used. </td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td> <b>Meaning:</b>The directory path of the  layout detection model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>Score threshold for the layout model. <br/>
+<b>Description:</b> 
+Any value between <code>0-1</code>. If not set, the default value is used, which is <code>0.5</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td>
+<b>Meaning:</b>Whether to use Non-Maximum Suppression (NMS) as post-processing for layout detection. <br/>
+<b>Description:</b> 
+If not set, the parameter will be set to the value initialized in the pipeline, which is <code>True</code> by default.
+</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Unclip ratio for detected boxes in layout detection model. <br/>
+<b>Description:</b> 
+Any float > <code>0</code>. If not set, the default is <code>1.0</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>The merging mode for the detection boxes output by the model in layout region detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>large</b>: When set to "large", only the largest outer bounding box will be retained for overlapping bounding boxes, and the inner overlapping boxes will be removed;</li>
+<li><b>small</b>: When set to "small", only the smallest inner bounding boxes will be retained for overlapping bounding boxes, and the outer overlapping boxes will be removed;</li>
+<li><b>union</b>: No filtering of bounding boxes will be performed, and both inner and outer boxes will be retained;</li>
+</ul>If not set, the default is <code>large</code>.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_batch_size</code></td>
+<td><b>Meaning:</b>The batch size for the layout region detection model. <br/>
+<b>Description:</b> 
+If not set, the default batch size will be set to <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td>
+<b>Meaning:</b>Whether to load and use the layout detection module. <br/>
+<b>Description:</b> 
+If not set, the parameter will be set to the value initialized in the pipeline, which is <code>True</code> by default.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_name</code></td>
+<td>
+<b>Meaning:</b>The name of the formula recognition model. <br/>
+<b>Description:</b> 
+If not set, the default model from the pipeline will be used.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the formula recognition model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>formula_recognition_batch_size</code></td>
+<td><b>Meaning:</b>The batch size for the formula recognition model. <br/>
+<b>Description:</b> 
+If not set, the batch size will default to <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>The device used for inference. <br/>
+<b>Description:</b> 
+You can specify a particular card number:
+<ul>
+<li><b>CPU</b>: e.g., <code>cpu</code> indicates using CPU for inference;</li>
+<li><b>GPU</b>: e.g., <code>gpu:0</code> indicates using the 1st GPU for inference;</li>
+<li><b>NPU</b>: e.g., <code>npu:0</code> indicates using the 1st NPU for inference;</li>
+<li><b>XPU</b>: e.g., <code>xpu:0</code> indicates using the 1st XPU for inference;</li>
+<li><b>MLU</b>: e.g., <code>mlu:0</code> indicates using the 1st MLU for inference;</li>
+<li><b>DCU</b>: e.g., <code>dcu:0</code> indicates using the 1st DCU for inference;</li>
+<li><b>MetaX GPU</b>: e.g., <code>metax_gpu:0</code> indicates using the 1st MetaX GPU for inference;</li>
+<li><b>Iluvatar GPU</b>: e.g., <code>iluvatar_gpu:0</code> indicates using the 1st Iluvatar GPU for inference;</li>
+</ul>If not set, the pipeline initialized value for this parameter will be used. During initialization, the local GPU device 0 will be preferred; if unavailable, the CPU device will be used.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to enable the TensorRT subgraph engine of Paddle Inference.<br/>
+<b>Description:</b>
+If the model does not support TensorRT acceleration, acceleration will not be used even if this flag is set.<br/>
+For CUDA 11.8 versions of PaddlePaddle, the compatible TensorRT version is 8.x (x>=6). TensorRT 8.6.1.6 is recommended.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b> Computation precision, such as <code>fp32</code> or <code>fp16</code>.</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b> Whether to enable MKL-DNN accelerated inference.<br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support MKL-DNN acceleration, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b> MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads used for inference on CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b> Path to the PaddleX pipeline configuration file.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+
+</tbody>
+</table>
+</details>
+<br />
+
+
+The results of running the default configuration of the formula recognition pipeline will be printed to the terminal as follows:
+
+```bash
+{'res': {'input_path': './general_formula_recognition_001.png', 'page_index': None, 'model_settings': {'use_doc_preprocessor': True, 'use_layout_detection': True}, 'doc_preprocessor_res': {'input_path': None, 'page_index': None, 'model_settings': {'use_doc_orientation_classify': True, 'use_doc_unwarping': True}, 'angle': 0}, 'layout_det_res': {'input_path': None, 'page_index': None, 'boxes': [{'cls_id': 2, 'label': 'text', 'score': 0.9855189323425293, 'coordinate': [90.56131, 1086.7773, 658.8992, 1553.2681]}, {'cls_id': 2, 'label': 'text', 'score': 0.9814704060554504, 'coordinate': [93.04651, 127.988556, 664.8587, 396.60892]}, {'cls_id': 2, 'label': 'text', 'score': 0.9767388105392456, 'coordinate': [698.4391, 591.0454, 1293.3676, 748.28345]}, {'cls_id': 2, 'label': 'text', 'score': 0.9712911248207092, 'coordinate': [701.4946, 286.61566, 1299.0099, 391.87457]}, {'cls_id': 2, 'label': 'text', 'score': 0.9709068536758423, 'coordinate': [697.0126, 751.93604, 1290.2236, 883.64453]}, {'cls_id': 2, 'label': 'text', 'score': 0.9689271450042725, 'coordinate': [704.01196, 79.645935, 1304.7493, 187.96674]}, {'cls_id': 2, 'label': 'text', 'score': 0.9683637619018555, 'coordinate': [93.063385, 799.3567, 660.6935, 902.0344]}, {'cls_id': 7, 'label': 'formula', 'score': 0.9660536646842957, 'coordinate': [728.5045, 440.9215, 1224.0634, 570.8518]}, {'cls_id': 7, 'label': 'formula', 'score': 0.9616329669952393, 'coordinate': [722.9789, 1333.5085, 1257.1136, 1468.0432]}, {'cls_id': 7, 'label': 'formula', 'score': 0.9610316753387451, 'coordinate': [756.4525, 1211.323, 1188.0428, 1268.2336]}, {'cls_id': 7, 'label': 'formula', 'score': 0.960993230342865, 'coordinate': [777.51355, 207.87927, 1222.8966, 267.33014]}, {'cls_id': 2, 'label': 'text', 'score': 0.9594196677207947, 'coordinate': [697.5154, 957.6764, 1288.6238, 1033.5211]}, {'cls_id': 2, 'label': 'text', 'score': 0.9593432545661926, 'coordinate': [691.333, 1511.8015, 1282.0968, 1642.5906]}, {'cls_id': 7, 'label': 'formula', 'score': 0.9589930176734924, 'coordinate': [153.89856, 924.2046, 601.0946, 1036.9038]}, {'cls_id': 2, 'label': 'text', 'score': 0.9582098722457886, 'coordinate': [87.02347, 1557.2971, 655.9584, 1632.6912]}, {'cls_id': 7, 'label': 'formula', 'score': 0.9579620957374573, 'coordinate': [810.86975, 1057.0771, 1175.101, 1117.6631]}, {'cls_id': 7, 'label': 'formula', 'score': 0.9557801485061646, 'coordinate': [165.26271, 557.8495, 598.1803, 614.35]}, {'cls_id': 7, 'label': 'formula', 'score': 0.953873872756958, 'coordinate': [116.48187, 713.88416, 614.2181, 774.02576]}, {'cls_id': 2, 'label': 'text', 'score': 0.9521227478981018, 'coordinate': [96.6882, 478.32745, 662.573, 536.5877]}, {'cls_id': 2, 'label': 'text', 'score': 0.944242000579834, 'coordinate': [96.12866, 639.1591, 661.7959, 692.4849]}, {'cls_id': 2, 'label': 'text', 'score': 0.9403323531150818, 'coordinate': [695.9436, 1138.6748, 1286.7242, 1188.0049]}, {'cls_id': 7, 'label': 'formula', 'score': 0.9249663949012756, 'coordinate': [852.90137, 908.64386, 1131.1882, 933.81793]}, {'cls_id': 7, 'label': 'formula', 'score': 0.9249223470687866, 'coordinate': [195.28397, 424.81024, 567.697, 451.1291]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.9173304438591003, 'coordinate': [1246.2393, 1079.0535, 1286.3281, 1104.3323]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.9169507026672363, 'coordinate': [1246.9003, 908.6482, 1288.2013, 934.61426]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.915979266166687, 'coordinate': [1247.0374, 1229.1572, 1287.094, 1254.9805]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.9085646867752075, 'coordinate': [1252.864, 492.1079, 1294.6238, 518.47095]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.9017605781555176, 'coordinate': [1242.1719, 1473.6951, 1283.02, 1498.6316]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.8999755382537842, 'coordinate': [1269.8164, 220.34933, 1299.8589, 247.01102]}, {'cls_id': 7, 'label': 'formula', 'score': 0.8965252041816711, 'coordinate': [96.00711, 235.49493, 295.43823, 265.60016]}, {'cls_id': 2, 'label': 'text', 'score': 0.8954343199729919, 'coordinate': [696.85693, 1286.2236, 1083.3921, 1310.8643]}, {'cls_id': 7, 'label': 'formula', 'score': 0.8952110409736633, 'coordinate': [166.60979, 129.20242, 511.65692, 156.29672]}, {'cls_id': 2, 'label': 'text', 'score': 0.893648624420166, 'coordinate': [725.64575, 396.18964, 1263.0391, 422.76813]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.8922948837280273, 'coordinate': [634.14124, 427.77087, 661.1686, 454.10022]}, {'cls_id': 2, 'label': 'text', 'score': 0.8892256617546082, 'coordinate': [94.483246, 1058.7595, 441.92313, 1082.4875]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.8878197073936462, 'coordinate': [630.4175, 939.3015, 657.7135, 965.36426]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.8831961154937744, 'coordinate': [630.5835, 1000.95715, 657.4309, 1026.2128]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.8767948150634766, 'coordinate': [634.1024, 575.3833, 660.59094, 601.1677]}, {'cls_id': 7, 'label': 'formula', 'score': 0.873543918132782, 'coordinate': [95.29655, 1320.3627, 264.93008, 1345.8473]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.8702306151390076, 'coordinate': [633.82825, 730.31525, 659.83215, 755.5485]}, {'cls_id': 7, 'label': 'formula', 'score': 0.8387619853019714, 'coordinate': [365.19897, 268.29675, 515.7938, 296.07013]}, {'cls_id': 7, 'label': 'formula', 'score': 0.8314349055290222, 'coordinate': [1090.509, 1599.1382, 1276.6736, 1622.156]}, {'cls_id': 7, 'label': 'formula', 'score': 0.817135751247406, 'coordinate': [246.175, 161.22958, 314.3764, 186.40591]}, {'cls_id': 3, 'label': 'number', 'score': 0.8042846322059631, 'coordinate': [1297.4036, 7.1497707, 1310.5969, 27.737753]}, {'cls_id': 7, 'label': 'formula', 'score': 0.7970448136329651, 'coordinate': [538.45593, 478.09354, 661.8812, 508.50778]}, {'cls_id': 7, 'label': 'formula', 'score': 0.7644855976104736, 'coordinate': [916.51746, 1618.5188, 1009.62537, 1640.8206]}, {'cls_id': 7, 'label': 'formula', 'score': 0.7423419952392578, 'coordinate': [694.8439, 1612.2507, 861.05334, 1635.9768]}, {'cls_id': 7, 'label': 'formula', 'score': 0.7072376608848572, 'coordinate': [99.72007, 508.21167, 254.91953, 535.74744]}, {'cls_id': 7, 'label': 'formula', 'score': 0.6976271867752075, 'coordinate': [696.8011, 1561.4375, 899.79584, 1586.7349]}, {'cls_id': 7, 'label': 'formula', 'score': 0.6707713007926941, 'coordinate': [1117.0862, 1571.9763, 1191.502, 1594.742]}, {'cls_id': 7, 'label': 'formula', 'score': 0.6338322162628174, 'coordinate': [577.33484, 1274.4131, 602.5636, 1296.7021]}, {'cls_id': 7, 'label': 'formula', 'score': 0.6199935674667358, 'coordinate': [175.28284, 349.82376, 241.24683, 376.6708]}, {'cls_id': 7, 'label': 'formula', 'score': 0.612853467464447, 'coordinate': [773.06287, 595.202, 800.43884, 617.3812]}, {'cls_id': 7, 'label': 'formula', 'score': 0.6107096672058105, 'coordinate': [706.6776, 316.87082, 736.69714, 339.9352]}, {'cls_id': 7, 'label': 'formula', 'score': 0.5520269870758057, 'coordinate': [1263.9711, 314.65167, 1292.7728, 337.3896]}, {'cls_id': 7, 'label': 'formula', 'score': 0.5346108675003052, 'coordinate': [1219.2955, 316.599, 1243.9181, 339.71802]}, {'cls_id': 7, 'label': 'formula', 'score': 0.5195119380950928, 'coordinate': [254.65729, 323.6553, 326.57758, 349.53494]}, {'cls_id': 7, 'label': 'formula', 'score': 0.501812219619751, 'coordinate': [255.8518, 1350.6472, 301.74304, 1375.5286]}]}, 'formula_res_list': [{'rec_formula': '\\begin{aligned}{\\psi_{0}(M)-\\psi_{}(M,z)=}&{{}\\frac{(1-\\epsilon_{r})}{\\epsilon_{r}}\\frac{\\lambda^{2}c^{2}}{t_{\\operatorname{E}}^{2}\\operatorname{l n}(10)}\\times}\\\\ {}&{{}\\int_{0}^{z}d z^{\\prime}\\frac{d t}{d z^{\\prime}}\\left.\\frac{\\partial\\phi}{\\partial L}\\right|_{L=\\lambda M c^{2}/t_{\\operatorname{E}}},}\\\\ \\end{aligned}', 'formula_region_id': 1, 'dt_polys': ([728.5045, 440.9215, 1224.0634, 570.8518],)}, {'rec_formula': '\\begin{aligned}{p(\\operatorname{l o g}_{10}}&{{}M|\\operatorname{l o g}_{10}\\sigma)=\\frac{1}{\\sqrt{2\\pi}\\epsilon_{0}}}\\\\ {}&{{}\\times\\operatorname{e x p}\\left[-\\frac{1}{2}\\left(\\frac{\\operatorname{l o g}_{10}M-a_{\\bullet}-b_{\\bullet}\\operatorname{l o g}_{10}\\sigma}{\\epsilon_{0}}\\right)^{2}\\right].}\\\\ \\end{aligned}', 'formula_region_id': 2, 'dt_polys': ([722.9789, 1333.5085, 1257.1136, 1468.0432],)}, {'rec_formula': '\\psi_{0}(M)=\\int d\\sigma\\frac{p(\\operatorname{l o g}_{10}M|\\operatorname{l o g}_{10}\\sigma)}{M\\operatorname{l o g}(10)}\\frac{d n}{d\\sigma}(\\sigma),', 'formula_region_id': 3, 'dt_polys': ([756.4525, 1211.323, 1188.0428, 1268.2336],)}, {'rec_formula': '\\phi(L)\\equiv\\frac{d n}{d\\operatorname{l o g}_{10}L}=\\frac{\\phi_{*}}{(L/L_{*})^{\\gamma_{1}}+(L/L_{*})^{\\gamma_{2}}}.', 'formula_region_id': 4, 'dt_polys': ([777.51355, 207.87927, 1222.8966, 267.33014],)}, {'rec_formula': '\\begin{aligned}{\\rho_{\\operatorname{B H}}}&{{}=\\int d M\\psi(M)M}\\\\ {}&{{}=\\frac{1-\\epsilon_{r}}{\\epsilon_{r}c^{2}}\\int_{0}^{\\infty}d z\\frac{d t}{d z}\\int d\\operatorname{l o g}_{10}L\\phi(L,z)L,}\\\\ \\end{aligned}', 'formula_region_id': 5, 'dt_polys': ([153.89856, 924.2046, 601.0946, 1036.9038],)}, {'rec_formula': '\\frac{d n}{d\\sigma}d\\sigma=\\psi_{*}\\left(\\frac{\\sigma}{\\sigma_{*}}\\right)^{\\alpha}\\frac{e^{-(\\sigma/\\sigma_{*})^{\\beta}}}{\\Gamma(\\alpha/\\beta)}\\beta\\frac{d\\sigma}{\\sigma}.', 'formula_region_id': 6, 'dt_polys': ([810.86975, 1057.0771, 1175.101, 1117.6631],)}, {'rec_formula': '\\langle\\dot{M}(M,t)\\rangle\\psi(M,t)=\\frac{(1-\\epsilon_{r})}{\\epsilon_{r}c^{2}\\operatorname{l n}(10)}\\phi(L,t)\\frac{d L}{d M}.', 'formula_region_id': 7, 'dt_polys': ([165.26271, 557.8495, 598.1803, 614.35],)}, {'rec_formula': '\\frac{\\partial\\psi}{\\partial t}(M,t)+\\frac{(1-\\epsilon_{r})}{\\epsilon_{r}}\\frac{\\lambda^{2}c^{2}}{t_{\\operatorname{E}}^{2}\\operatorname{l n}(10)}\\left.\\frac{\\partial\\phi}{\\partial L}\\right|_{L=\\lambda M c^{2}/t_{\\operatorname{E}}}=0,', 'formula_region_id': 8, 'dt_polys': ([116.48187, 713.88416, 614.2181, 774.02576],)}, {'rec_formula': '\\operatorname{l o g}_{10}M=a_{\\bullet}+b_{\\bullet}\\operatorname{l o g}_{10}X.', 'formula_region_id': 9, 'dt_polys': ([852.90137, 908.64386, 1131.1882, 933.81793],)}, {'rec_formula': '\\phi(L,t)d\\operatorname{l o g}_{10}L=\\delta(M,t)\\psi(M,t)d M.', 'formula_region_id': 10, 'dt_polys': ([195.28397, 424.81024, 567.697, 451.1291],)}, {'rec_formula': '\\dot{M}\\:=\\:(1\\:-\\:\\epsilon_{r})\\dot{M}_{\\mathrm{a c c}}^{\\mathrm{~\\tiny~\\cdot~}}', 'formula_region_id': 11, 'dt_polys': ([96.00711, 235.49493, 295.43823, 265.60016],)}, {'rec_formula': 't_{E}=\\sigma_{T}c/4\\pi G m_{p}=4.5\\times10^{8}\\mathrm{y r}', 'formula_region_id': 12, 'dt_polys': ([166.60979, 129.20242, 511.65692, 156.29672],)}, {'rec_formula': 'M_{*}=L_{*}t_{E}/\\tilde{\\lambda}c^{2}', 'formula_region_id': 13, 'dt_polys': ([95.29655, 1320.3627, 264.93008, 1345.8473],)}, {'rec_formula': '\\phi(L,t)d\\operatorname{l o g}_{10}L', 'formula_region_id': 14, 'dt_polys': ([365.19897, 268.29675, 515.7938, 296.07013],)}, {'rec_formula': 'a_{\\bullet}=8.32\\pm0.05', 'formula_region_id': 15, 'dt_polys': ([1090.509, 1599.1382, 1276.6736, 1622.156],)}, {'rec_formula': '\\epsilon_{r}\\dot{M}_{\\mathrm{a c c}}', 'formula_region_id': 16, 'dt_polys': ([246.175, 161.22958, 314.3764, 186.40591],)}, {'rec_formula': '\\langle\\dot{M}(M,t)\\rangle=', 'formula_region_id': 17, 'dt_polys': ([538.45593, 478.09354, 661.8812, 508.50778],)}, {'rec_formula': '\\epsilon_{0}=0.38', 'formula_region_id': 18, 'dt_polys': ([916.51746, 1618.5188, 1009.62537, 1640.8206],)}, {'rec_formula': 'b_{\\bullet}=5.64\\dot{\\pm}\\dot{0.32}', 'formula_region_id': 19, 'dt_polys': ([694.8439, 1612.2507, 861.05334, 1635.9768],)}, {'rec_formula': '\\delta(M,t)\\dot{M}(M,t)', 'formula_region_id': 20, 'dt_polys': ([99.72007, 508.21167, 254.91953, 535.74744],)}, {'rec_formula': 'X=\\sigma/200\\mathrm{k m}\\mathrm{~s^{-1}~}', 'formula_region_id': 21, 'dt_polys': ([696.8011, 1561.4375, 899.79584, 1586.7349],)}, {'rec_formula': 'M-\\sigma', 'formula_region_id': 22, 'dt_polys': ([1117.0862, 1571.9763, 1191.502, 1594.742],)}, {'rec_formula': 'L_{*}', 'formula_region_id': 23, 'dt_polys': ([577.33484, 1274.4131, 602.5636, 1296.7021],)}, {'rec_formula': '\\phi(L,t)', 'formula_region_id': 24, 'dt_polys': ([175.28284, 349.82376, 241.24683, 376.6708],)}, {'rec_formula': '\\psi_{0}', 'formula_region_id': 25, 'dt_polys': ([773.06287, 595.202, 800.43884, 617.3812],)}, {'rec_formula': '\\mathrm{A^{\\prime\\prime}}', 'formula_region_id': 26, 'dt_polys': ([706.6776, 316.87082, 736.69714, 339.9352],)}, {'rec_formula': 'L_{*}', 'formula_region_id': 27, 'dt_polys': ([1263.9711, 314.65167, 1292.7728, 337.3896],)}, {'rec_formula': '\\phi_{*}', 'formula_region_id': 28, 'dt_polys': ([1219.2955, 316.599, 1243.9181, 339.71802],)}, {'rec_formula': '\\delta(M,t)', 'formula_region_id': 29, 'dt_polys': ([254.65729, 323.6553, 326.57758, 349.53494],)}, {'rec_formula': '\\phi(L)', 'formula_region_id': 30, 'dt_polys': ([255.8518, 1350.6472, 301.74304, 1375.5286],)}]}}
+```
+
+The explanation of the running result parameters can refer to the result interpretation in [ 2.2 Python Script Integration](#22-python-script-integration).
+
+The visualization results are saved under <code>save_path</code>, where the visualization result of formula recognition is as follows:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/formula_recognition/04_paddleocr3.png" style="width: 70%"/>
+
+<b> If you need to visualize the formula recognition pipeline, you need to run the following command to install the LaTeX rendering environment. Currently, visualization of the formula recognition pipeline only supports the Ubuntu environment, and other environments are not supported. For complex formulas, the LaTeX result may contain some advanced representations that may not be successfully displayed in environments such as Markdown:</b>
+
+```bash
+sudo apt-get update
+sudo apt-get install texlive texlive-latex-base texlive-xetex latex-cjk-all texlive-latex-extra -y
+```
+
+<b>Note</b>: Due to the need to render each formula image during the formula recognition visualization process, the process takes a long time. Please be patient.
+
+### 2.2 Python Script Integration
+
+Using the command line is a quick way to experience and check the results. Generally, in a project, you often need to integrate it through code. You can perform quick inference with just a few lines of code. The inference code is as follows:
+
+```python
+from paddleocr import FormulaRecognitionPipeline
+
+pipeline = FormulaRecognitionPipeline()
+# ocr = FormulaRecognitionPipeline(use_doc_orientation_classify=True) # Specify whether to use the document orientation classification model with use_doc_orientation_classify.
+# ocr = FormulaRecognitionPipeline(use_doc_unwarping=True) # Specify whether to use the text image unwarping module with use_doc_unwarping.
+# ocr = FormulaRecognitionPipeline(device="gpu") # Specify the use of GPU for model inference with device.
+output = pipeline.predict("./general_formula_recognition_001.png")
+for res in output:
+    res.print() ##  Print the structured output of the prediction
+    res.save_to_img(save_path="output") ## Save the formula visualization result of the current image.
+    res.save_to_json(save_path="output") ## Save the structured JSON result of the current image
+```
+
+The example above uses the local `paddle_static` inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following code:
+
+```python
+from paddleocr import FormulaRecognitionPipeline
+
+pipeline = FormulaRecognitionPipeline(
+    engine="transformers",
+)
+# ocr = FormulaRecognitionPipeline(use_doc_orientation_classify=True) # Specify whether to use the document orientation classification model with use_doc_orientation_classify.
+# ocr = FormulaRecognitionPipeline(use_doc_unwarping=True) # Specify whether to use the text image unwarping module with use_doc_unwarping.
+# ocr = FormulaRecognitionPipeline(device="gpu") # Specify the use of GPU for model inference with device.
+output = pipeline.predict("./general_formula_recognition_001.png")
+for res in output:
+    res.print() ##  Print the structured output of the prediction
+    res.save_to_img(save_path="output") ## Save the formula visualization result of the current image.
+    res.save_to_json(save_path="output") ## Save the structured JSON result of the current image
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+In the above Python script, the following steps are executed:
+
+（1）Instantiate the formula recognition pipeline object through <code>create_pipeline()</code>, with specific parameters as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>The name of the document orientation classification model.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default model in pipeline will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the document orientation classification model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_batch_size</code></td>
+<td><b>Meaning:</b>The batch size of the document orientation classification model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default batch size will be set to <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>Meaning:</b>The name of the text image unwarping model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default model in pipeline will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the  text image unwarping model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_batch_size</code></td>
+<td><b>Meaning:</b>The batch size of the text image unwarping model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default batch size will be set to <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to load and use the document orientation classification module. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the parameter will be set to the value initialized in the pipeline, which is <code>True</code> by default.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to load and use the text image unwarping module.<br/> 
+<b>Description:</b> 
+If set to <code>None</code>, the parameter will be set to the value initialized in the pipeline, which is <code>True</code> by default.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>Meaning:</b>The name of the layout detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default model in pipeline will be used. </td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the  layout detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>Threshold for layout detection, used to filter out predictions with low confidence. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Such as 0.2, indicates filtering out all bounding boxes with a confidence score less than 0.2;</li>
+<li><b>Dictionary</b>: With <b>int</b> keys representing <code>cls_id</code> and <b>float</b> values as thresholds. For example, <code>{0: 0.45, 2: 0.48, 7: 0.4}</code> indicates applying a threshold of 0.45 for class ID 0, 0.48 for class ID 2, and 0.4 for class ID 7;</li>
+<li><b>None</b>: If set to <code>None</code>, the default is <code>0.5</code>.</li>
+</ul>
+</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Whether to use Non-Maximum Suppression (NMS) as post-processing for layout detection. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the parameter will be set to the value initialized in the pipeline, which is set to <code>True</code> by default.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion factor for the detection boxes of the layout region detection model. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>Tuple[float,float]</b>: Expansion ratios in horizontal and vertical directions;</li>
+<li><b>dict</b>: A dictionary with <b>int</b> keys representing <code>cls_id</code>, and <b>tuple</b> values, e.g., <code>{0: (1.1, 2.0)}</code> means width is expanded 1.1× and height 2.0× for class 0 boxes;</li>
+<li><b>None</b>: If set to <code>None</code>, uses the pipeline default of <code>1.0</code>.</li>
+</ul>
+</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>Filtering method for overlapping boxes in layout detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>str</b>: Options include <code>large</code>, <code>small</code>, and <code>union</code> to retain the larger box, smaller box, or both;</li>
+<li><b>dict</b>: A dictionary with <b>int</b> keys representing <code>cls_id</code>, and <b>str</b> values, e.g., <code>{0: "large", 2: "small"}</code> means using different modes for different classes;</li>
+<li><b>None</b>: If set to <code>None</code>, uses the pipeline default value <code>large</code>.</li>
+</ul>
+</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_batch_size</code></td>
+<td><b>Meaning:</b>The batch size for the layout region detection model.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default batch size will be set to <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>Meaning:</b>Whether to load and use the layout detection module. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the parameter will be set to the value initialized in the pipeline, which is <code>True</code> by default.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_name</code></td>
+<td><b>Meaning:</b>The name of the formula recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default model from the pipeline will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the formula recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_batch_size</code></td>
+<td><b>Meaning:</b>The batch size for the formula recognition model. <br/>
+<b>Description:</b> 
+If set to  <code>None</code>, the batch size will default to <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>The device used for inference. <br/>
+<b>Description:</b> 
+You can specify a particular card number:
+<ul>
+<li><b>CPU</b>: e.g., <code>cpu</code> indicates using CPU for inference;</li>
+<li><b>GPU</b>: e.g., <code>gpu:0</code> indicates using the 1st GPU for inference;</li>
+<li><b>NPU</b>: e.g., <code>npu:0</code> indicates using the 1st NPU for inference;</li>
+<li><b>XPU</b>: e.g., <code>xpu:0</code> indicates using the 1st XPU for inference;</li>
+<li><b>MLU</b>: e.g., <code>mlu:0</code> indicates using the 1st MLU for inference;</li>
+<li><b>DCU</b>: e.g., <code>dcu:0</code> indicates using the 1st DCU for inference;</li>
+<li><b>MetaX GPU</b>: e.g., <code>metax_gpu:0</code> indicates using the 1st MetaX GPU for inference;</li>
+<li><b>Iluvatar GPU</b>: e.g., <code>iluvatar_gpu:0</code> indicates using the 1st Iluvatar GPU for inference;</li>
+<li><b>None</b>: If set to <code>None</code>, the pipeline initialized value for this parameter will be used. During initialization, the local GPU device 0 will be preferred; if unavailable, the CPU device will be used.
+</ur>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to enable the TensorRT subgraph engine of Paddle Inference.<br/>
+<b>Description:</b>
+If the model does not support TensorRT acceleration, acceleration will not be used even if this flag is set.<br/>
+For CUDA 11.8 versions of PaddlePaddle, the compatible TensorRT version is 8.x (x>=6). TensorRT 8.6.1.6 is recommended.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b> Computation precision, such as <code>"fp32"</code> or <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b> Whether to enable MKL-DNN accelerated inference.<br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support MKL-DNN acceleration, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b> MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads used for inference on CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b> Path to the PaddleX pipeline configuration file.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+</tbody>
+</table>
+
+（2）Call the <code>predict()</code> method of the formula recognition pipeline object to perform inference prediction. This method will return a list of results.
+
+Additionally, the pipeline also provides the <code>predict_iter()</code> method. Both methods are completely consistent in terms of parameter acceptance and result return. The difference is that <code>predict_iter()</code> returns a <code>generator</code>, which allows for step-by-step processing and retrieval of prediction results. This is suitable for handling large datasets or scenarios where memory saving is desired. You can choose to use either of these methods based on your actual needs.
+
+Here are the parameters of the <code>predict()</code> method and their descriptions:
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted, supporting multiple input types, required.<br/>
+<b>Description:</b> 
+<ul>
+<li><b>Python Var</b>: Image data represented by <code>numpy.ndarray;</code></li>
+<li><b>str</b>: Local path of image or PDF file, e.g., <code>/root/data/img.jpg</code>; <b>URL link</b>, e.g., network URL of image or PDF file: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/pipelines/general_formula_recognition_001.png">Example</a>; <b>Local directory</b>, the directory should contain images to be predicted, e.g., local path: <code>/root/data/</code> (currently does not support prediction of PDF files in directories; PDF files must be specified with a specific file path);</li>
+<li><b>list</b>: Elements of the list must be of the above types, e.g., <code>[numpy.ndarray, numpy.ndarray]</code>, <code>[\"/root/data/img1.jpg\", \"/root/data/img2.jpg\"]</code>, <code>[\"/root/data1\", \"/root/data2\"].</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td>
+<b>Meaning:</b>Whether to use the layout detection module during inference. </td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td>
+<b>Meaning:</b>Whether to use the document orientation classification module during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to use the text image unwarping module during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>string</code></td>
+<td><code>None</code></td>
+</tr>
+</tr></tr></tbody>
+</table>
+
+（3）Process the prediction results, where the prediction result for each sample corresponds to a Result object, and supports operations such as printing, saving as an image, and saving as a <code>json</code> file:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Description</th>
+<th>Parameter</th>
+<th>Parameter Type</th>
+<th>Parameter Description</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">Print results to terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output content using <code>JSON</code> indentation.</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data, making it more readable. Effective only when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non-<code>ASCII</code> characters to <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> retains the original characters. Effective only when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">Save results as a JSON file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Path to save the file. If it is a directory, the saved file will be named the same as the input file type.</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data, making it more readable. Effective only when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non-<code>ASCII</code> characters to <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> retains the original characters. Effective only when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save results as an image file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>Path to save the file, supports directory or file path.</td>
+<td>无</td>
+</tr>
+</table>
+<ul>
+  <li>Calling the <code>print()</code> method will print the results to the terminal. The content printed to the terminal is explained as follows:
+    <ol start="1" type="1">
+      <li><code>input_path</code>: <code>(str)</code> The input path of the image to be predicted.</li>
+      <li><code>page_index</code>: <code>(Union[int, None])</code> If the input is a PDF file, this indicates the current page number of the PDF. Otherwise, it is <code>None</code></li>
+      <li><code>model_settings</code>: <code>(Dict[str, bool])</code> The model parameters required for the pipeline configuration.
+        <ol >
+          <li><code>use_doc_preprocessor</code>: <code>(bool)</code> Controls whether to enable the document preprocessing sub-pipeline.</li>
+          <li><code>use_layout_detection</code>: <code>(bool)</code> Controls whether to enable the layout detection module.</li>
+        </ol>
+      </li>
+      <li><code>doc_preprocessor_res</code>: <code>(Dict[str, Union[str, Dict[str, bool], int]])</code> The output result of the document preprocessing sub-pipeline. It exists only when <code>use_doc_preprocessor=True</code>
+        <ol>
+          <li><code>input_path</code>: <code>(Union[str, None])</code> T <code>numpy.ndarray</code>, it is saved as <code>None</code>.</li>
+          <li><code>model_settings</code>: <code>(Dict)</code> he model configuration parameters of the preprocessing sub-pipeline.
+            <ul>
+              <li><code>use_doc_orientation_classify</code>: <code>(bool)</code> Controls whether to enable document orientation classification.</li>
+              <li><code>use_doc_unwarping</code>: <code>(bool)</code> Controls whether to enable document distortion correction.</li>
+            </ul>
+          </li>
+          <li><code>angle</code>: <code>(int)</code> The prediction result of document orientation classification. When enabled, it takes values from [0,1,2,3], corresponding to [0°,90°,180°,270°]; when disabled, it is -1.</li>
+        </ol>
+      </li>
+      <li><code>layout_det_res</code>: <code>(Dict[str, List[Dict]])</code> The output result of the layout area detection module. It exists only when <code>use_layout_detection=True</code></li>
+        <ol>
+          <li><code>input_path</code>: <code>(Union[str, None])</code> The image path accepted by the layout area detection module. When the input is a <code>numpy.ndarray</code>, it is saved as <code>None</code>.</li>
+            <ol>
+                <li><code>boxes</code>: <code>(List[Dict[int, str, float, List[float]]])</code> A list of layout area detection prediction results.</li>
+                <li><code>cls_id</code>: <code>(int)</code> The class ID predicted by layout area detection.</li>
+                <li><code>label</code>: <code>(str)</code> The class label predicted by layout area detection.</li>
+                <li><code>score</code>: <code>(float)</code> The confidence score of the predicted class.</li>
+                <li><code>coordinate</code>: <code>(List[float])</code> The bounding box coordinates predicted by layout area detection, in the format [x_min, y_min, x_max, y_max], where (x_min, y_min) is the top-left corner and (x_max, y_max) is the bottom-right corner.</li>
+            </ol>
+          <li><code>formula_res_list</code>: <code>(List[Dict[str, int, List[float]]])</code>  A list of formula recognition prediction results.</li>
+            <ol>
+                <li><code>rec_formula</code>: <code>(str)</code> The LaTeX source code predicted by formula recognition.</li>
+                <li><code>formula_region_id</code>: <code>(int)</code>  The ID number predicted by formula recognition.</li>
+                <li><code>dt_polys</code>:  <code>(List[float])</code> The bounding box coordinates predicted by formula recognition, in the format [x_min, y_min, x_max, y_max], where (x_min, y_min) is the top-left corner and (x_max, y_max) is the bottom-right corner.</li>
+            </ol>
+        </li>
+      </li>
+    </ol>
+  </li>
+  <li>Calling the <code>save_to_json()</code> method will save the above content to the specified <code>save_path</code>. If a directory is specified, the saved path will be <code>save_path/{your_img_basename}_res.json</code>. If a file is specified, it will be saved directly to that file. Since JSON files do not support saving numpy arrays, <code>numpy.array</code> types will be converted to list format.</li>
+<li>Calling the <code>save_to_img()</code> method will save the visualization results to the specified <code>save_path</code>. If a directory is specified, the saved path will be <code>save_path/{your_img_basename}_formula_res_img.{your_img_extension}</code>. If a file is specified, it will be saved directly to that file. (The pipeline usually contains many result images, so it is not recommended to specify a specific file path directly, otherwise multiple images will be overwritten and only the last one will be retained.)</li>
+
+
+* In addition, you can also obtain the visualization image with results and the prediction results through attributes, as follows:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Attribute Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">Get the prediction results in <code>json</code> format</td>
+</tr>
+<tr>
+<td rowspan="2"><code>img</code></td>
+<td rowspan="2">Get the visualization image in <code>dict</code> format</td>
+</tr>
+</table>
+
+
+<ul>
+<li>The prediction result obtained from the <code>json</code> attribute is of the dict type, and its content is consistent with what is saved by calling the <code>save_to_json()</code> method.</li>
+<li>The prediction result returned by the  <code>img</code> attribute is a dictionary-type data. The keys are  <code>preprocessed_img</code>、 <code>layout_det_res</code> and <code>formula_res_img</code>, and the corresponding values are three <code>Image.Image</code> objects: the first one is used to display the visualization of image preprocessing, the second one is for displaying the visualization of layout region detection, and the third one is for displaying the visualization of formula recognition. If the image preprocessing submodule is not used, the dictionary will not contain the <code>preprocessed_img</code> key. Similarly, if the layout region detection submodule is not used, the dictionary will not contain the <code>layout_det_res</code> key.</li>
+</ul>
+
+## 3. Development Integration/Deployment
+
+If the formula recognition pipeline meets your requirements for inference speed and accuracy, you can proceed directly with development integration/deployment.
+
+If you need to integrate the formula recognition pipeline into your Python project, you can refer to the example code in [ 2.2 Python Script Integration](#22-python-script-integration).
+
+In addition, PaddleOCR also provides two other deployment methods, which are detailed as follows:
+
+🚀 High-Performance Inference: In real-world production environments, many applications have stringent standards for performance metrics of deployment strategies, particularly regarding response speed, to ensure efficient system operation and a smooth user experience. To address this, PaddleOCR offers high-performance inference capabilities designed to deeply optimize the performance of model inference and pre/post-processing, significantly accelerating the end-to-end process. For detailed information on the high-performance inference process, please refer to the [High-Performance Inference Guide](../inference_deployment/local_inference/high_performance_inference.en.md).
+
+
+☁️ Service-Based Deployment: 
+Service-Based Deployment is a common deployment form in real-world production environments. By encapsulating inference capabilities as a service, clients can access these services via network requests to obtain inference results. For detailed instructions on Service-Based Deployment in pipelines, please refer to the [Service-Based Deployment Guide](../inference_deployment/serving/serving.en.md).
+
+Below are the API references for basic service-based deployment and multi-language service invocation examples:
+
+
+<details><summary>API Reference</summary>
+<p>For the main operations provided by the service:</p>
+<ul>
+<li>The HTTP request method is POST.</li>
+<li>Both the request body and response body are JSON data (JSON objects).</li>
+<li>When the request is processed successfully, the response status code is <code>200</code>, and the attributes of the response body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>The UUID of the request.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Fixed as <code>0</code>.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error message. Fixed as <code>"Success"</code>.</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>The result of the operation.</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is not processed successfully, the attributes of the response body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>The UUID of the request.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Same as the response status code.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error message.</td>
+</tr>
+</tbody>
+</table>
+<p>The main operations provided by the service are as follows:</p>
+<ul>
+<li><b><code>infer</code></b></li>
+</ul>
+<p>Obtain the formula recognition results from images.</p>
+<p><code>POST /formula-recognition</code></p>
+<ul>
+<li>The attributes of the request body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+<th>Required</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>The URL of image files (including TIFF; multi-page TIFF is processed page by page) or PDF file accessible by the server, or the Base64-encoded content of the file. By default, for PDF or multi-page TIFF files exceeding 10 pages, only the first 10 pages will be processed.<br />
+To remove the page limit, please add the following configuration to the pipeline configuration file:
+<pre><code>Serving:
+  extra:
+    max_num_input_imgs: null
+</code></pre>
+</td>
+<td>Yes</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>The type of the file. <code>0</code> for PDF files, <code>1</code> for image files (including TIFF). If this attribute is missing, the file type will be inferred from the URL.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_doc_orientation_classify</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_doc_unwarping</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useLayoutDetection</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_layout_detection</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutThreshold</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>layout_threshold</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutNms</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>layout_nms</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutUnclipRatio</code></td>
+<td><code>number</code> | <code>array</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>layout_unclip_ratio</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutMergeBboxesMode</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>layout_merge_bboxes_mode</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>
+Whether to return the final visualization image and intermediate images during the processing.<br/>
+<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>If <code>true</code> is provided: return images.</li>
+<li>If <code>false</code> is provided: do not return any images.</li>
+<li>If this parameter is omitted from the request body, or if <code>null</code> is explicitly passed, the behavior will follow the value of <code>Serving.visualize</code> in the pipeline configuration.</li>
+</ul>
+<br/>
+For example, adding the following setting to the pipeline config file:<br/>
+<pre><code>Serving:
+  visualize: False
+</code></pre>
+will disable image return by default. This behavior can be overridden by explicitly setting the <code>visualize</code> parameter in the request.<br/>
+If neither the request body nor the configuration file is set (If <code>visualize</code> is set to <code>null</code> in the request and  not defined in the configuration file), the image is returned by default.
+</td>
+<td>No</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is processed successfully, the <code>result</code> in the response body has the following attributes:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>formulaRecResults</code></td>
+<td><code>object</code></td>
+<td>The formula recognition results. The array length is 1 (for image input) or the actual number of document pages processed (for PDF input). For PDF input, each element in the array represents the result of each page actually processed in the PDF file.</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>Information about the input data.</td>
+</tr>
+</tbody>
+</table>
+<p>Image fields in the element schema below (e.g. <code>outputImages</code>, <code>inputImage</code>) are returned inline as Base64 strings by default; when the server is configured to return URLs, those values become pre-signed URLs while the field types remain unchanged. See the "Returning Binary Content as URLs" section of the <a href="../inference_deployment/serving/serving.en.md">Serving Deployment Guide</a> for configuration.</p>
+<p>Each element in <code>formulaRecResults</code> is an <code>object</code> with the following attributes:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>A simplified version of the <code>res</code> field in the JSON representation of the result generated by the pipeline object's <code>predict</code> method, excluding the <code>input_path</code> and the <code>page_index</code> fields.</td>
+</tr>
+</tr>
+<tr>
+<td><code>outputImages</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>See the description of the <code>img</code> attribute of the result of the pipeline prediction. The images are in JPEG format and are Base64-encoded by default; returned as pre-signed URLs when URL-return mode is enabled.</td>
+</tr>
+<tr>
+<td><code>inputImage</code> | <code>null</code></td>
+<td><code>string</code></td>
+<td>The input image. The image is in JPEG format and is Base64-encoded by default; returned as a pre-signed URL when URL-return mode is enabled.</td>
+</tr>
+</tbody>
+</table>
+</details>
+<details><summary>Multi-language Service Invocation Example</summary>
+<details>
+<summary>Python</summary>
+
+<pre><code class="language-python">import base64
+import requests
+
+API_URL = "http://localhost:8080/formula-recognition"
+file_path = "./demo.jpg"
+
+with open(file_path, "rb") as file:
+    file_bytes = file.read()
+    file_data = base64.b64encode(file_bytes).decode("ascii")
+
+payload = {"file": file_data, "fileType": 1}
+
+response = requests.post(API_URL, json=payload)
+
+assert response.status_code == 200
+result = response.json()["result"]
+for i, res in enumerate(result["formulaRecResults"]):
+    print(res["prunedResult"])
+    for img_name, img in res["outputImages"].items():
+        img_path = f"{img_name}_{i}.jpg"
+        with open(img_path, "wb") as f:
+            f.write(base64.b64decode(img))
+        print(f"Output image saved at {img_path}")
+</code></pre></details>
+
+<details><summary>C++</summary>
+
+<pre><code class="language-cpp">#include &lt;iostream&gt;
+#include &lt;fstream&gt;
+#include &lt;vector&gt;
+#include &lt;string&gt;
+#include "cpp-httplib/httplib.h" // https://github.com/Huiyicc/cpp-httplib
+#include "nlohmann/json.hpp" // https://github.com/nlohmann/json
+#include "base64.hpp" // https://github.com/tobiaslocker/base64
+
+int main() {
+    httplib::Client client("localhost", 8080);
+
+    const std::string filePath = "./demo.jpg";
+
+    std::ifstream file(filePath, std::ios::binary | std::ios::ate);
+    if (!file) {
+        std::cerr << "Error opening file: " << filePath << std::endl;
+        return 1;
+    }
+
+    std::streamsize size = file.tellg();
+    file.seekg(0, std::ios::beg);
+    std::vector<char> buffer(size);
+    if (!file.read(buffer.data(), size)) {
+        std::cerr << "Error reading file." << std::endl;
+        return 1;
+    }
+
+    std::string bufferStr(buffer.data(), static_cast<size_t>(size));
+    std::string encodedFile = base64::to_base64(bufferStr);
+
+    nlohmann::json jsonObj;
+    jsonObj["file"] = encodedFile;
+    jsonObj["fileType"] = 1;
+
+    auto response = client.Post("/formula-recognition", jsonObj.dump(), "application/json");
+
+    if (response && response->status == 200) {
+        nlohmann::json jsonResponse = nlohmann::json::parse(response->body);
+        auto result = jsonResponse["result"];
+
+        if (!result.is_object() || !result["formulaRecResults"].is_array()) {
+            std::cerr << "Unexpected response format." << std::endl;
+            return 1;
+        }
+
+        for (size_t i = 0; i < result["formulaRecResults"].size(); ++i) {
+            auto res = result["formulaRecResults"][i];
+
+            if (res.contains("prunedResult")) {
+                std::cout << "Recognized formula: " << res["prunedResult"].dump() << std::endl;
+            }
+
+            if (res.contains("outputImages") && res["outputImages"].is_object()) {
+                for (auto& [imgName, imgData] : res["outputImages"].items()) {
+                    std::string outputPath = imgName + "_" + std::to_string(i) + ".jpg";
+                    std::string decodedImage = base64::from_base64(imgData.get<std::string>());
+
+                    std::ofstream outFile(outputPath, std::ios::binary);
+                    if (outFile.is_open()) {
+                        outFile.write(decodedImage.c_str(), decodedImage.size());
+                        outFile.close();
+                        std::cout << "Saved image: " << outputPath << std::endl;
+                    } else {
+                        std::cerr << "Failed to write image: " << outputPath << std::endl;
+                    }
+                }
+            }
+        }
+    } else {
+        std::cerr << "Request failed." << std::endl;
+        if (response) {
+            std::cerr << "HTTP status: " << response->status << std::endl;
+            std::cerr << "Response body: " << response->body << std::endl;
+        }
+        return 1;
+    }
+
+    return 0;
+}
+</code></pre></details>
+
+<details><summary>Java</summary>
+
+<pre><code class="language-java">import okhttp3.*;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.util.Base64;
+
+public class Main {
+    public static void main(String[] args) throws IOException {
+        String API_URL = "http://localhost:8080/formula-recognition";
+        String imagePath = "./demo.jpg";
+
+        File file = new File(imagePath);
+        byte[] fileContent = java.nio.file.Files.readAllBytes(file.toPath());
+        String base64Image = Base64.getEncoder().encodeToString(fileContent);
+
+        ObjectMapper objectMapper = new ObjectMapper();
+        ObjectNode payload = objectMapper.createObjectNode();
+        payload.put("file", base64Image);
+        payload.put("fileType", 1);
+
+        OkHttpClient client = new OkHttpClient();
+        MediaType JSON = MediaType.get("application/json; charset=utf-8");
+        RequestBody body = RequestBody.create(JSON, payload.toString());
+
+        Request request = new Request.Builder()
+                .url(API_URL)
+                .post(body)
+                .build();
+
+        try (Response response = client.newCall(request).execute()) {
+            if (response.isSuccessful()) {
+                String responseBody = response.body().string();
+                JsonNode root = objectMapper.readTree(responseBody);
+                JsonNode result = root.get("result");
+
+                JsonNode formulaRecResults = result.get("formulaRecResults");
+                for (int i = 0; i < formulaRecResults.size(); i++) {
+                    JsonNode item = formulaRecResults.get(i);
+                    int finalI = i;
+
+                    JsonNode prunedResult = item.get("prunedResult");
+                    System.out.println("Pruned Result [" + i + "]: " + prunedResult.toString());
+
+                    JsonNode outputImages = item.get("outputImages");
+                    if (outputImages != null && outputImages.isObject()) {
+                        outputImages.fieldNames().forEachRemaining(imgName -> {
+                            try {
+                                String imgBase64 = outputImages.get(imgName).asText();
+                                byte[] imgBytes = Base64.getDecoder().decode(imgBase64);
+                                String imgPath = imgName + "_" + finalI + ".jpg";
+                                try (FileOutputStream fos = new FileOutputStream(imgPath)) {
+                                    fos.write(imgBytes);
+                                    System.out.println("Saved image: " + imgPath);
+                                }
+                            } catch (IOException e) {
+                                System.err.println("Failed to save image: " + e.getMessage());
+                            }
+                        });
+                    }
+                }
+            } else {
+                System.err.println("Request failed with HTTP code: " + response.code());
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Go</summary>
+
+<pre><code class="language-go">package main
+
+import (
+    "bytes"
+    "encoding/base64"
+    "encoding/json"
+    "fmt"
+    "io/ioutil"
+    "net/http"
+    "os"
+)
+
+func main() {
+    API_URL := "http://localhost:8080/formula-recognition"
+    filePath := "./demo.jpg"
+
+    fileBytes, err := ioutil.ReadFile(filePath)
+    if err != nil {
+        fmt.Printf("Error reading file: %v\n", err)
+        return
+    }
+    fileData := base64.StdEncoding.EncodeToString(fileBytes)
+
+    payload := map[string]interface{}{
+        "file":     fileData,
+        "fileType": 1,
+    }
+    payloadBytes, err := json.Marshal(payload)
+    if err != nil {
+        fmt.Printf("Error marshaling payload: %v\n", err)
+        return
+    }
+
+    client := &http.Client{}
+    req, err := http.NewRequest("POST", API_URL, bytes.NewBuffer(payloadBytes))
+    if err != nil {
+        fmt.Printf("Error creating request: %v\n", err)
+        return
+    }
+    req.Header.Set("Content-Type", "application/json")
+
+    res, err := client.Do(req)
+    if err != nil {
+        fmt.Printf("Error sending request: %v\n", err)
+        return
+    }
+    defer res.Body.Close()
+
+    if res.StatusCode != http.StatusOK {
+        fmt.Printf("Unexpected status code: %d\n", res.StatusCode)
+        return
+    }
+
+    body, err := ioutil.ReadAll(res.Body)
+    if err != nil {
+        fmt.Printf("Error reading response body: %v\n", err)
+        return
+    }
+
+    type FormulaRecResult struct {
+        PrunedResult map[string]interface{} `json:"prunedResult"`
+        OutputImages map[string]string      `json:"outputImages"`
+        InputImage   *string                `json:"inputImage"`
+    }
+
+    type Response struct {
+        Result struct {
+            FormulaRecResults []FormulaRecResult `json:"formulaRecResults"`
+            DataInfo          interface{}        `json:"dataInfo"`
+        } `json:"result"`
+    }
+
+    var respData Response
+    if err := json.Unmarshal(body, &respData); err != nil {
+        fmt.Printf("Error unmarshaling response: %v\n", err)
+        return
+    }
+
+    for i, res := range respData.Result.FormulaRecResults {
+        fmt.Printf("Result %d - prunedResult: %+v\n", i, res.PrunedResult)
+
+        for imgName, imgData := range res.OutputImages {
+            imgBytes, err := base64.StdEncoding.DecodeString(imgData)
+            if err != nil {
+                fmt.Printf("Error decoding image %s_%d: %v\n", imgName, i, err)
+                continue
+            }
+
+            filename := fmt.Sprintf("%s_%d.jpg", imgName, i)
+            if err := os.WriteFile(filename, imgBytes, 0644); err != nil {
+                fmt.Printf("Error saving image %s: %v\n", filename, err)
+                continue
+            }
+            fmt.Printf("Saved image to %s\n", filename)
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>C#</summary>
+
+<pre><code class="language-csharp">using System;
+using System.IO;
+using System.Net.Http;
+using System.Text;
+using System.Threading.Tasks;
+using Newtonsoft.Json.Linq;
+
+class Program
+{
+    static readonly string API_URL = "http://localhost:8080/formula-recognition";
+    static readonly string inputFilePath = "./demo.jpg";
+
+    static async Task Main(string[] args)
+    {
+        var httpClient = new HttpClient();
+
+        byte[] fileBytes = File.ReadAllBytes(inputFilePath);
+        string fileData = Convert.ToBase64String(fileBytes);
+
+        var payload = new JObject
+        {
+            { "file", fileData },
+            { "fileType", 1 }
+        };
+        var content = new StringContent(payload.ToString(), Encoding.UTF8, "application/json");
+
+        HttpResponseMessage response = await httpClient.PostAsync(API_URL, content);
+        response.EnsureSuccessStatusCode();
+
+        string responseBody = await response.Content.ReadAsStringAsync();
+        JObject jsonResponse = JObject.Parse(responseBody);
+
+        JArray formulaRecResults = (JArray)jsonResponse["result"]["formulaRecResults"];
+        for (int i = 0; i < formulaRecResults.Count; i++)
+        {
+            var res = formulaRecResults[i];
+            Console.WriteLine($"[{i}] prunedResult:\n{res["prunedResult"]}");
+
+            JObject outputImages = res["outputImages"] as JObject;
+            if (outputImages != null)
+            {
+                foreach (var img in outputImages)
+                {
+                    string imgName = img.Key;
+                    string base64Img = img.Value?.ToString();
+                    if (!string.IsNullOrEmpty(base64Img))
+                    {
+                        string imgPath = $"{imgName}_{i}.jpg";
+                        byte[] imageBytes = Convert.FromBase64String(base64Img);
+                        File.WriteAllBytes(imgPath, imageBytes);
+                        Console.WriteLine($"Output image saved at {imgPath}");
+                    }
+                }
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Node.js</summary>
+
+<pre><code class="language-js">const axios = require('axios');
+const fs = require('fs');
+const path = require('path');
+
+const API_URL = 'http://localhost:8080/formula-recognition';
+const inputFilePath = './demo.jpg';
+const fileType = 1;
+
+function encodeImageToBase64(filePath) {
+  const bitmap = fs.readFileSync(filePath);
+  return Buffer.from(bitmap).toString('base64');
+}
+
+const payload = {
+  file: encodeImageToBase64(inputFilePath),
+  fileType: fileType
+};
+
+axios.post(API_URL, payload)
+  .then((response) => {
+    const resultArray = response.data.result.formulaRecResults;
+
+    resultArray.forEach((res, index) => {
+      console.log(`\n[${index}] prunedResult:`);
+      console.log(res.prunedResult);
+
+      const outputImages = res.outputImages;
+      if (outputImages) {
+        Object.entries(outputImages).forEach(([imgName, base64Img]) => {
+          const outputPath = `${imgName}_${index}.jpg`;
+          fs.writeFileSync(outputPath, Buffer.from(base64Img, 'base64'));
+          console.log(`Saved output image: ${outputPath}`);
+        });
+      } else {
+        console.log(`[${index}] outputImages is null`);
+      }
+    });
+  })
+  .catch((error) => {
+    console.error('API error：', error.message);
+  });
+</code></pre></details>
+
+<details><summary>PHP</summary>
+
+<pre><code class="language-php">&lt;?php
+
+$API_URL = "http://localhost:8080/formula-recognition";
+$image_path = "./demo.jpg";
+
+$image_data = base64_encode(file_get_contents($image_path));
+$payload = array("file" => $image_data, "fileType" => 1);
+
+$ch = curl_init($API_URL);
+curl_setopt($ch, CURLOPT_POST, true);
+curl_setopt($ch, CURLOPT_POSTFIELDS, json_encode($payload));
+curl_setopt($ch, CURLOPT_HTTPHEADER, array('Content-Type: application/json'));
+curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+$response = curl_exec($ch);
+curl_close($ch);
+
+$result = json_decode($response, true)["result"]["formulaRecResults"];
+
+foreach ($result as $i => $item) {
+    echo "[$i] prunedResult:\n";
+    print_r($item["prunedResult"]);
+
+    if (!empty($item["outputImages"])) {
+        foreach ($item["outputImages"] as $img_name => $base64_img) {
+            $img_path = "{$img_name}_{$i}.jpg";
+            file_put_contents($img_path, base64_decode($base64_img));
+            echo "Output image saved at $img_path\n";
+        }
+    } else {
+        echo "No outputImages found for item $i\n";
+    }
+}
+?&gt;
+</code></pre></details>
+</details>
+<br/>
+
+
+## 4. Custom Development
+
+If the default model weights provided by the formula recognition pipeline do not meet your requirements in terms of accuracy or speed, you can try to <b>fine-tune</b> the existing models using <b>your own domain-specific or application-specific data</b> to improve the recognition performance of the formula recognition pipeline in your scenario.
+
+### 4.1 Model Fine-Tuning
+
+Since the formula recognition pipeline consists of several modules, if the pipeline's performance is not satisfactory, the issue may arise from any one of these modules. You can analyze the poorly recognized images to determine which module is problematic and refer to the corresponding fine-tuning tutorial links in the table below for model fine-tuning.
+
+<table>
+<thead>
+<tr>
+<th>Scenario</th>
+<th>Fine-Tuning Module</th>
+<th>Reference Link</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>Formulas are missing</td>
+<td>Layout Detection Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/module_usage/layout_detection.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Formula content is inaccurate</td>
+<td>Formula Recognition Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/module_usage/formula_recognition.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Whole-image rotation correction is inaccurate</td>
+<td>Document Image Orientation Classification Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/module_usage/doc_img_orientation_classification.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Image distortion correction is inaccurate</td>
+<td>Text Image Correction Module</td>
+<td>Fine-tuning not supported</td>
+</tr>
+</tbody>
+</table>
+
+### 4.2  Model Deployment
+
+After you complete fine-tuning training using a private dataset, you can obtain a local model weight file. You can then use the fine-tuned model weights by specifying the local model save path through parameters or by customizing the pipeline configuration file.
+
+#### 4.2.1 Specify the local model path through parameters
+
+When initializing the pipeline object, specify the local model path through parameters. Take the usage of the weights after fine-tuning the text detection model as an example, as follows:
+
+Command line mode:
+
+```bash
+# Specify the local model path via --formula_recognition_model_dir
+paddleocr formula_recognition_pipeline -i ./general_formula_recognition_001.png --formula_recognition_model_dir your_formula_recognition_model_path
+
+# PP-FormulaNet_plus-M model is used as the default formula recognition model. If you do not fine-tune this model, modify the model name by using --formula_recognition_model_name
+paddleocr formula_recognition_pipeline -i ./general_formula_recognition_001.png --formula_recognition_model_name PP-FormulaNet_plus-M --formula_recognition_model_dir your_ppformulanet_plus-m_formula_recognition_model_path
+```
+
+Script mode：
+
+```python
+
+from paddleocr import FormulaRecognitionPipeline
+
+#  Specify the local model path via formula_recognition_model_dir
+pipeline = FormulaRecognitionPipeline(formula_recognition_model_dir="./your_formula_recognition_model_path")
+output = pipeline.predict("./general_formula_recognition_001.png")
+for res in output:
+    res.print() ## Print the structured output of the prediction
+    res.save_to_img(save_path="output") ## Save the formula visualization result of the current image.
+    res.save_to_json(save_path="output") ## Save the structured JSON result of the current image
+
+# PP-FormulaNet_plus-M model is used as the default formula recognition model. If you do not fine-tune this model, modify the model name by using  formula_recognition_model_name
+# pipeline = FormulaRecognitionPipeline(formula_recognition_model_name="PP-FormulaNet_plus-M", formula_recognition_model_dir="./your_ppformulanet_plus-m_formula_recognition_model_path")
+
+```
+
+
+#### 4.2.2 Specify the local model path through the configuration file
+
+
+1.Obtain the pipeline configuration file
+
+Call the `export_paddlex_config_to_yaml` method of the **Formula Recognition Pipeline** object in PaddleOCR to export the current pipeline configuration as a YAML file:  
+
+```Python
+from paddleocr import FormulaRecognitionPipeline
+
+pipeline = FormulaRecognitionPipeline()
+pipeline.export_paddlex_config_to_yaml("FormulaRecognitionPipeline.yaml")
+```
+
+2.Modify the Configuration File 
+
+After obtaining the default pipeline configuration file, replace the paths of the default model weights with the local paths of your fine-tuned model weights. For example:  
+
+```yaml
+......
+SubModules:
+  FormulaRecognition:
+    batch_size: 5
+    model_dir: null # Replace with the path to your fine-tuned formula recognition model weights  
+    model_name: PP-FormulaNet_plus-M # If the name of the fine-tuned model is different from the default model name, please modify it here as well
+    module_name: formula_recognition
+  LayoutDetection:
+    batch_size: 1
+    layout_merge_bboxes_mode: large
+    layout_nms: true
+    layout_unclip_ratio: 1.0
+    model_dir: null # Replace with the path to your fine-tuned layout detection model weights 
+    model_name: PP-DocLayout_plus-L # If the name of the fine-tuned model is different from the default model name, please modify it here as well
+    module_name: layout_detection
+    threshold: 0.5
+SubPipelines:
+  DocPreprocessor:
+    SubModules:
+      DocOrientationClassify:
+        batch_size: 1
+        model_dir: null # Replace with the path to your fine-tuned document image orientation classification model weights 
+        model_name: PP-LCNet_x1_0_doc_ori # If the name of the fine-tuned model is different from the default model name, please modify it here as well
+        module_name: doc_text_orientation
+      DocUnwarping:
+        batch_size: 1
+        model_dir: null 
+        model_name: UVDoc 
+        module_name: image_unwarping
+    pipeline_name: doc_preprocessor
+    use_doc_orientation_classify: true
+    use_doc_unwarping: true
+pipeline_name: formula_recognition
+use_doc_preprocessor: true
+use_layout_detection: true
+......
+```
+
+The pipeline configuration file includes not only the parameters supported by the PaddleOCR CLI and Python API but also advanced configurations. For detailed instructions, refer to the [PaddleX Pipeline Usage Overview](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/pipeline_develop_guide.html) and adjust the configurations as needed.  
+
+3.Load the Configuration File in CLI  
+
+After modifying the configuration file, specify its path using the `--paddlex_config` parameter in the command line. PaddleOCR will read the file and apply the configurations. Example:  
+
+```bash
+paddleocr formula_recognition_pipeline -i ./general_formula_recognition_001.png --paddlex_config FormulaRecognitionPipeline.yaml 
+```
+4.Load the Configuration File in Python API  
+
+When initializing the pipeline object, pass the path of the PaddleX pipeline configuration file or a configuration dictionary via the `paddlex_config` parameter. PaddleOCR will read and apply the configurations. Example: 
+
+```python
+from paddleocr import  FormulaRecognitionPipeline
+
+pipeline =  FormulaRecognitionPipeline(paddlex_config="FormulaRecognitionPipeline.yaml")
+output = pipeline.predict("./general_formula_recognition_001.png")
+for res in output:
+    res.print() ## Print the structured output of the prediction
+    res.save_to_img(save_path="output") ## Save the formula visualization result of the current image.
+    res.save_to_json(save_path="output") ## Save the structured JSON result of the current image
+```
diff --git a/docs/version3.x/pipeline_usage/formula_recognition.md b/docs/version3.x/pipeline_usage/formula_recognition.md
new file mode 100644
index 0000000..8e0f712
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/formula_recognition.md
@@ -0,0 +1,2043 @@
+---
+comments: true
+---
+
+# 公式识别产线使用教程
+
+## 1. 公式识别产线介绍
+
+公式识别是一种自动从文档或图像中识别和提取LaTeX公式内容及其结构的技术，广泛应用于数学、物理、计算机科学等领域的文档编辑和数据分析。通过使用计算机视觉和机器学习算法，公式识别能够将复杂的数学公式信息转换为可编辑的LaTeX格式，方便用户进一步处理和分析数据。
+
+公式识别产线用于解决公式识别任务，提取图片中的公式信息以LaTeX源码形式输出，本产线是一个集成了百度飞桨视觉团队自研的先进公式识别模型PP-FormulaNet 和业界知名公式识别模型 UniMERNet的端到端公式识别系统，支持简单印刷公式、复杂印刷公式、手写公式的识别，并在此基础上，增加了对图像的方向矫正和扭曲矫正功能。基于本产线，可实现公式内容精准预测，使用场景覆盖教育、科研、金融、制造等各个领域。本产线同时提供了灵活的服务化部署方式，支持在多种硬件上使用多种编程语言调用。不仅如此，本产线也提供了二次开发的能力，您可以基于本产线在您自己的数据集上训练调优，训练后的模型也可以无缝集成。
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/formula_recognition/03.png" style="width: 70%"/>
+
+<b>公式识别产线中包含以下4个模块。每个模块均可独立进行训练和推理，并包含多个模型。有关详细信息，请点击相应模块以查看文档。</b>
+
+- [公式识别模块](../module_usage/formula_recognition.md)
+- [版面区域检测模块](../module_usage/layout_detection.md)（可选）
+- [文档图像方向分类模块](../module_usage/doc_img_orientation_classification.md) （可选）
+- [文本图像矫正模块](../module_usage/text_image_unwarping.md) （可选）
+
+在本产线中，您可以根据下方的基准测试数据选择使用的模型。
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。
+> 在带有 [常规模式 / 高性能模式] 标记的推理耗时列中，`常规模式` 对应本地推理引擎 `paddle_static`。
+
+<details>
+<summary><b>文档图像方向分类模块（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>Top-1 Acc（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">训练模型</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>基于PP-LCNet_x1_0的文档图像分类模型，含有四个类别，即0度，90度，180度，270度</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>文本图像矫正模块（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>CER </th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">训练模型</a></td>
+<td>0.179</td>
+<td>19.05 / 19.05</td>
+<td>- / 869.82</td>
+<td>30.3</td>
+<td>高精度文本图像矫正模型</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>版面区域检测模块（可选）：</b></summary>
+
+* <b>版面检测模型，包含20个常见的类别：文档标题、段落标题、文本、页码、摘要、目录、参考文献、脚注、页眉、页脚、算法、公式、公式编号、图像、表格、图和表标题（图标题、表格标题和图表标题）、印章、图表、侧栏文本和参考文献内容</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout_plus-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout_plus-L_pretrained.pdparams">训练模型</a></td>
+<td>83.2</td>
+<td>53.03 / 17.23</td>
+<td>634.62 / 378.32</td>
+<td>126.01</td>
+<td>基于RT-DETR-L在包含中英文论文、多栏杂志、报纸、PPT、合同、书本、试卷、研报、古籍、日文文档、竖版文字文档等场景的自建数据集训练的更高精度版面区域定位模型</td>
+</tr>
+<tr>
+</tbody>
+</table>
+
+* <b>版面检测模型，包含23个常见的类别：文档标题、段落标题、文本、页码、摘要、目录、参考文献、脚注、页眉、页脚、算法、公式、公式编号、图像、图表标题、表格、表格标题、印章、图表标题、图表、页眉图像、页脚图像、侧栏文本</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-L_pretrained.pdparams">训练模型</a></td>
+<td>90.4</td>
+<td>33.59 / 33.59</td>
+<td>503.01 / 251.08</td>
+<td>123.76</td>
+<td>基于RT-DETR-L在包含中英文论文、杂志、合同、书本、试卷和研报等场景的自建数据集训练的高精度版面区域定位模型</td>
+</tr>
+<tr>
+<td>PP-DocLayout-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-M_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-M_pretrained.pdparams">训练模型</a></td>
+<td>75.2</td>
+<td>13.03 / 4.72</td>
+<td>43.39 / 24.44</td>
+<td>22.578</td>
+<td>基于PicoDet-L在包含中英文论文、杂志、合同、书本、试卷和研报等场景的自建数据集训练的精度效率平衡的版面区域定位模型</td>
+</tr>
+<tr>
+<td>PP-DocLayout-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-S_pretrained.pdparams">训练模型</a></td>
+<td>70.9</td>
+<td>11.54 / 3.86</td>
+<td>18.53 / 6.29</td>
+<td>4.834</td>
+<td>基于PicoDet-S在中英文论文、杂志、合同、书本、试卷和研报等场景上自建数据集训练的高效率版面区域定位模型</td>
+</tr>
+</tbody>
+</table>
+
+>❗ 以上列出的是版面区域检测模块重点支持的<b>4个核心模型</b>，该模块总共支持<b>7个全量模型</b>，包含多个预定义了不同类别的模型，完整的模型列表如下：
+
+<details><summary> 👉模型列表详情</summary>
+
+* <b>版面区域检测模型，包含17个版面常见类别，分别是：段落标题、图片、文本、数字、摘要、内容、图表标题、公式、表格、表格标题、参考文献、文档标题、脚注、页眉、算法、页脚、印章</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）</th>
+<th>CPU推理耗时 (ms)</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>87.4</td>
+<td>8.80 / 3.62</td>
+<td>17.51 / 6.35</td>
+<td>4.8</td>
+<td>基于PicoDet-S轻量模型在中英文论文、杂志和研报等场景上自建数据集训练的高效率版面区域定位模型</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>89.0</td>
+<td>12.60 / 10.27</td>
+<td>43.70 / 24.42</td>
+<td>22.6</td>
+<td>基于PicoDet-L在中英文论文、杂志和研报等场景上自建数据集训练的效率精度均衡版面区域定位模型</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>98.3</td>
+<td>115.29 / 101.18</td>
+<td>964.75 / 964.75</td>
+<td>470.2</td>
+<td>基于RT-DETR-H在中英文论文、杂志和研报等场景上自建数据集训练的高精度版面区域定位模型</td>
+</tr>
+</tbody>
+</table>
+
+
+* <b>版面区域检测模型，包含23个常见的类别：文档标题、段落标题、文本、页码、摘要、目录、参考文献、脚注、页眉、页脚、算法、公式、公式编号、图像、图表标题、表格、表格标题、印章、图表标题、图表、页眉图像、页脚图像、侧栏文本</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-L_pretrained.pdparams">训练模型</a></td>
+<td>90.4</td>
+<td>33.59 / 33.59</td>
+<td>503.01 / 251.08</td>
+<td>123.76</td>
+<td>基于RT-DETR-L在包含中英文论文、杂志、合同、书本、试卷和研报等场景的自建数据集训练的高精度版面区域定位模型</td>
+</tr>
+<tr>
+<td>PP-DocLayout-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-M_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-M_pretrained.pdparams">训练模型</a></td>
+<td>75.2</td>
+<td>13.03 / 4.72</td>
+<td>43.39 / 24.44</td>
+<td>22.578</td>
+<td>基于PicoDet-L在包含中英文论文、杂志、合同、书本、试卷和研报等场景的自建数据集训练的精度效率平衡的版面区域定位模型</td>
+</tr>
+<tr>
+<td>PP-DocLayout-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-S_pretrained.pdparams">训练模型</a></td>
+<td>70.9</td>
+<td>11.54 / 3.86</td>
+<td>18.53 / 6.29</td>
+<td>4.834</td>
+<td>基于PicoDet-S在中英文论文、杂志、合同、书本、试卷和研报等场景上自建数据集训练的高效率版面区域定位模型</td>
+</tr>
+</tbody>
+</table>
+
+* <b>版面检测模型，包含20个常见的类别：文档标题、段落标题、文本、页码、摘要、目录、参考文献、脚注、页眉、页脚、算法、公式、公式编号、图像、表格、图和表标题（图标题、表格标题和图表标题）、印章、图表、侧栏文本和参考文献内容</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout_plus-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout_plus-L_pretrained.pdparams">训练模型</a></td>
+<td>83.2</td>
+<td>53.03 / 17.23</td>
+<td>634.62 / 378.32</td>
+<td>126.01</td>
+<td>基于RT-DETR-L在包含中英文论文、多栏杂志、报纸、PPT、合同、书本、试卷、研报、古籍、日文文档、竖版文字文档等场景的自建数据集训练的更高精度版面区域定位模型</td>
+</tr>
+<tr>
+</tbody>
+</table>
+
+</details>
+</details>
+
+<details>
+<summary><b>公式识别模块：</b></summary>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>En-BLEU(%)</th>
+<th>Zh-BLEU(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>UniMERNet</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UniMERNet_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UniMERNet_pretrained.pdparams">训练模型</a></td>
+<td>85.91</td>
+<td>43.50</td>
+<td>1311.84 / 1311.84</td>
+<td>- / 8288.07</td>
+<td>1530</td>
+<td>UniMERNet是由上海AI Lab研发的一款公式识别模型。该模型采用Donut Swin作为编码器，MBartDecoder作为解码器，并通过在包含简单公式、复杂公式、扫描捕捉公式和手写公式在内的一百万数据集上进行训练，大幅提升了模型对真实场景公式的识别准确率</td>
+</tr>
+<td>PP-FormulaNet-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-S_pretrained.pdparams">训练模型</a></td>
+<td>87.00</td>
+<td>45.71</td>
+<td>182.25 / 182.25</td>
+<td>- / 254.39</td>
+<td>224</td>
+<td rowspan="2">PP-FormulaNet 是由百度飞桨视觉团队开发的一款先进的公式识别模型，支持5万个常见LateX源码词汇的识别。PP-FormulaNet-S 版本采用了 PP-HGNetV2-B4 作为其骨干网络，通过并行掩码和模型蒸馏等技术，大幅提升了模型的推理速度，同时保持了较高的识别精度，适用于简单印刷公式、跨行简单印刷公式等场景。而 PP-FormulaNet-L 版本则基于 Vary_VIT_B 作为骨干网络，并在大规模公式数据集上进行了深入训练，在复杂公式的识别方面，相较于PP-FormulaNet-S表现出显著的提升，适用于简单印刷公式、复杂印刷公式、手写公式等场景。 </td>
+
+</tr>
+<td>PP-FormulaNet-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet-L_pretrained.pdparams">训练模型</a></td>
+<td>90.36</td>
+<td>45.78</td>
+<td>1482.03 / 1482.03</td>
+<td>- / 3131.54</td>
+<td>695</td>
+</tr>
+<td>PP-FormulaNet_plus-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-S_pretrained.pdparams">训练模型</a></td>
+<td>88.71</td>
+<td>53.32</td>
+<td>179.20 / 179.20</td>
+<td>- / 260.99</td>
+<td>248</td>
+<td rowspan="3">PP-FormulaNet_plus 是百度飞桨视觉团队在 PP-FormulaNet 的基础上开发的增强版公式识别模型。与原版相比，PP-FormulaNet_plus 在训练中使用了更为丰富的公式数据集，包括中文学位论文、专业书籍、教材试卷以及数学期刊等多种来源。这一扩展显著提升了模型的识别能力。
+
+其中，PP-FormulaNet_plus-M 和 PP-FormulaNet_plus-L 模型新增了对中文公式的支持，并将公式的最大预测 token 数从 1024 扩大至 2560，大幅提升了对复杂公式的识别性能。同时，PP-FormulaNet_plus-S 模型则专注于增强英文公式的识别能力。通过这些改进，PP-FormulaNet_plus 系列模型在处理复杂多样的公式识别任务时表现更加出色。 </td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-M_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-M_pretrained.pdparams">训练模型</a></td>
+<td>91.45</td>
+<td>89.76</td>
+<td>1040.27 / 1040.27</td>
+<td>- / 1615.80</td>
+<td>592</td>
+</tr>
+<tr>
+<td>PP-FormulaNet_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-FormulaNet_plus-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-FormulaNet_plus-L_pretrained.pdparams">训练模型</a></td>
+<td>92.22</td>
+<td>90.64</td>
+<td>1476.07 / 1476.07</td>
+<td>- / 3125.58</td>
+<td>698</td>
+</tr>
+<tr>
+<td>LaTeX_OCR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/LaTeX_OCR_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/LaTeX_OCR_rec_pretrained.pdparams">训练模型</a></td>
+<td>74.55</td>
+<td>39.96</td>
+<td>1088.89 / 1088.89</td>
+<td>- / -</td>
+<td>99</td>
+<td>LaTeX-OCR是一种基于自回归大模型的公式识别算法，通过采用 Hybrid ViT 作为骨干网络，transformer作为解码器，显著提升了公式识别的准确性。</td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary><b>测试环境说明：</b></summary>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+            <li><strong>测试数据集：
+             </strong>
+                <ul>
+                  <li>文档图像方向分类模型：自建的内部数据集，覆盖证件和文档等多个场景，包含 1000 张图片。</li>
+                  <li> 文本图像矫正模型：<a href="https://www3.cs.stonybrook.edu/~cvl/docunet.html">DocUNet</a>。</li>
+                  <li>版面区域检测模型：PaddleOCR 自建的版面区域检测数据集，包含中英文论文、杂志、合同、书本、试卷和研报等常见的 500 张文档类型图片。</li>
+                  <li>17类区域检测模型：PaddleOCR 自建的版面区域检测数据集，包含中英文论文、杂志和研报等常见的 892 张文档类型图片。</li>
+                  <li>公式识别模型：自建的内部公式识别测试集。</li>
+                </ul>
+             </li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td><code>paddle_static</code></td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+</details>
+
+<br />
+<b>如果您更注重模型的精度，请选择精度较高的模型；如果您更在意模型的推理速度，请选择推理速度较快的模型；如果您关注模型的存储大小，请选择存储体积较小的模型。</b>
+
+## 2. 快速开始
+
+在本地使用公式识别产线前，请确保您已经按照[安装教程](../installation.md)完成了wheel包安装。如果您希望选择性安装依赖，请参考安装教程中的相关说明。该产线对应的依赖分组为 `doc-parser`。安装完成后，可以在本地使用命令行体验或 Python 集成。
+
+**请注意，如果在执行过程中遇到程序失去响应、程序异常退出、内存资源耗尽、推理速度极慢等问题，请尝试参考文档调整配置，例如关闭不需要使用的功能或使用更轻量的模型。**
+
+### 2.1 命令行方式体验
+
+一行命令即可快速体验 formula_recognition 产线效果。运行以下代码前，请您下载[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/pipelines/general_formula_recognition_001.png)到本地：
+
+```bash
+paddleocr formula_recognition_pipeline -i https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/pipelines/general_formula_recognition_001.png
+
+# 通过 --use_doc_orientation_classify 指定是否使用文档方向分类模型
+paddleocr formula_recognition_pipeline -i ./general_formula_recognition_001.png --use_doc_orientation_classify True
+
+# 通过 --use_doc_unwarping 指定是否使用文本图像矫正模块
+paddleocr formula_recognition_pipeline -i ./general_formula_recognition_001.png --use_doc_unwarping True
+
+# 通过 --device 指定模型推理时使用 GPU
+paddleocr formula_recognition_pipeline -i ./general_formula_recognition_001.png --device gpu
+```
+
+上述命令默认使用本地推理引擎 `paddle_static`。如需运行，请先参考[飞桨框架安装说明](../paddlepaddle_installation.md)安装 PaddlePaddle。
+
+如果选择 `transformers` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 Transformers 环境配置，然后执行如下命令：
+
+```bash
+# 使用 transformers 引擎进行推理
+paddleocr formula_recognition_pipeline -i https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/pipelines/general_formula_recognition_001.png \
+    --engine transformers
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+<details><summary><b>命令行支持更多参数设置，点击展开以查看命令行参数的详细说明</b></summary>
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，必填。<br/>
+<b>说明：</b>如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<br/>
+<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/demo_paper.png">示例</a>；<br/>
+<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td><b>含义：</b>指定推理结果文件保存的路径。<br/>
+<b>说明：</b>如果不设置，推理结果将不会保存到本地。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>含义：</b>文档方向分类模型的名称。<br/>
+<b>说明：</b>如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>文档方向分类模型的目录路径。<br/>
+<b>说明：</b>如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_batch_size</code></td>
+<td><b>含义：</b>文档方向分类模型的batch size。<br/>
+<b>说明：</b>如果不设置，将默认设置batch size为<code>1</code>。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>含义：</b>文本图像矫正模型的名称。<br/>
+<b>说明：</b>如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。<br/>
+<b>说明：</b>如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_batch_size</code></td>
+<td><b>含义：</b>文本图像矫正模型的batch size。<br/>
+<b>说明：</b>如果不设置，将默认设置batch size为<code>1</code>。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否加载并使用文档方向分类模块。<br/>
+<b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。<br/>
+<b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>含义：</b>版面区域检测模型的名称。<br/>
+<b>说明：</b>如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>含义：</b>版面区域检测模型的目录路径。<br/>
+<b>说明：</b>如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>版面区域检测的阈值，用于过滤掉低置信度预测结果的阈值。<br/>
+<b>说明：</b>如 0.2，表示过滤掉所有阈值小于0.2的目标框。如果不设置，将默认使用默认值。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>版面区域检测是否使用后处理NMS。<br/>
+<b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>版面区域检测中检测框的边长缩放倍数。<br/>
+<b>说明：</b>大于0的浮点数，如 1.1 ，表示将模型输出的检测框中心不变，宽和高都扩张1.1倍。如果不设置，将使用默认值：1.0。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>版面区域检测中模型输出的检测框的合并处理模式。<br/>
+<b>说明：</b>
+<ul>
+<li><b>large</b>，设置为large时，表示在模型输出的检测框中，对于互相重叠包含的检测框，只保留外部最大的框，删除重叠的内部框；</li>
+<li><b>small</b>，设置为small，表示在模型输出的检测框中，对于互相重叠包含的检测框，只保留内部被包含的小框，删除重叠的外部框；</li>
+<li><b>union</b>，不进行框的过滤处理，内外框都保留</li>
+</ul>如果不设置，将使用默认值：“large”；
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_batch_size</code></td>
+<td><b>含义：</b>版面区域检测模型的batch size。<br/>
+<b>说明：</b>如果不设置，将默认设置batch size为<code>1</code>。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<td><code>use_layout_detection</code></td>
+<td><b>含义：</b>是否加载并使用版面区域检测模块。<br/>
+<b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_name</code></td>
+<td><b>含义：</b>公式识别模型的名称。<br/>
+<b>说明：</b>如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_dir</code></td>
+<td><b>含义：</b>公式识别模型的目录路径。<br/>
+<b>说明：</b>如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>formula_recognition_batch_size</code></td>
+<td><b>含义：</b>公式识别模型的batch size。<br/>
+<b>说明：</b>如果不设置，将默认设置batch size为<code>1</code>。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+</ul>如果不设置， 将默认使用产线初始化的该参数值，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。
+
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>fp32</code>、<code>fp16</code>。</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+
+</tbody>
+</table>
+</details>
+<br />
+
+运行结果会被打印到终端上，默认配置的 formula_recognition 产线的运行结果如下：
+
+```bash
+{'res': {'input_path': './general_formula_recognition_001.png', 'page_index': None, 'model_settings': {'use_doc_preprocessor': True, 'use_layout_detection': True}, 'doc_preprocessor_res': {'input_path': None, 'page_index': None, 'model_settings': {'use_doc_orientation_classify': True, 'use_doc_unwarping': True}, 'angle': 0}, 'layout_det_res': {'input_path': None, 'page_index': None, 'boxes': [{'cls_id': 2, 'label': 'text', 'score': 0.9855189323425293, 'coordinate': [90.56131, 1086.7773, 658.8992, 1553.2681]}, {'cls_id': 2, 'label': 'text', 'score': 0.9814704060554504, 'coordinate': [93.04651, 127.988556, 664.8587, 396.60892]}, {'cls_id': 2, 'label': 'text', 'score': 0.9767388105392456, 'coordinate': [698.4391, 591.0454, 1293.3676, 748.28345]}, {'cls_id': 2, 'label': 'text', 'score': 0.9712911248207092, 'coordinate': [701.4946, 286.61566, 1299.0099, 391.87457]}, {'cls_id': 2, 'label': 'text', 'score': 0.9709068536758423, 'coordinate': [697.0126, 751.93604, 1290.2236, 883.64453]}, {'cls_id': 2, 'label': 'text', 'score': 0.9689271450042725, 'coordinate': [704.01196, 79.645935, 1304.7493, 187.96674]}, {'cls_id': 2, 'label': 'text', 'score': 0.9683637619018555, 'coordinate': [93.063385, 799.3567, 660.6935, 902.0344]}, {'cls_id': 7, 'label': 'formula', 'score': 0.9660536646842957, 'coordinate': [728.5045, 440.9215, 1224.0634, 570.8518]}, {'cls_id': 7, 'label': 'formula', 'score': 0.9616329669952393, 'coordinate': [722.9789, 1333.5085, 1257.1136, 1468.0432]}, {'cls_id': 7, 'label': 'formula', 'score': 0.9610316753387451, 'coordinate': [756.4525, 1211.323, 1188.0428, 1268.2336]}, {'cls_id': 7, 'label': 'formula', 'score': 0.960993230342865, 'coordinate': [777.51355, 207.87927, 1222.8966, 267.33014]}, {'cls_id': 2, 'label': 'text', 'score': 0.9594196677207947, 'coordinate': [697.5154, 957.6764, 1288.6238, 1033.5211]}, {'cls_id': 2, 'label': 'text', 'score': 0.9593432545661926, 'coordinate': [691.333, 1511.8015, 1282.0968, 1642.5906]}, {'cls_id': 7, 'label': 'formula', 'score': 0.9589930176734924, 'coordinate': [153.89856, 924.2046, 601.0946, 1036.9038]}, {'cls_id': 2, 'label': 'text', 'score': 0.9582098722457886, 'coordinate': [87.02347, 1557.2971, 655.9584, 1632.6912]}, {'cls_id': 7, 'label': 'formula', 'score': 0.9579620957374573, 'coordinate': [810.86975, 1057.0771, 1175.101, 1117.6631]}, {'cls_id': 7, 'label': 'formula', 'score': 0.9557801485061646, 'coordinate': [165.26271, 557.8495, 598.1803, 614.35]}, {'cls_id': 7, 'label': 'formula', 'score': 0.953873872756958, 'coordinate': [116.48187, 713.88416, 614.2181, 774.02576]}, {'cls_id': 2, 'label': 'text', 'score': 0.9521227478981018, 'coordinate': [96.6882, 478.32745, 662.573, 536.5877]}, {'cls_id': 2, 'label': 'text', 'score': 0.944242000579834, 'coordinate': [96.12866, 639.1591, 661.7959, 692.4849]}, {'cls_id': 2, 'label': 'text', 'score': 0.9403323531150818, 'coordinate': [695.9436, 1138.6748, 1286.7242, 1188.0049]}, {'cls_id': 7, 'label': 'formula', 'score': 0.9249663949012756, 'coordinate': [852.90137, 908.64386, 1131.1882, 933.81793]}, {'cls_id': 7, 'label': 'formula', 'score': 0.9249223470687866, 'coordinate': [195.28397, 424.81024, 567.697, 451.1291]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.9173304438591003, 'coordinate': [1246.2393, 1079.0535, 1286.3281, 1104.3323]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.9169507026672363, 'coordinate': [1246.9003, 908.6482, 1288.2013, 934.61426]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.915979266166687, 'coordinate': [1247.0374, 1229.1572, 1287.094, 1254.9805]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.9085646867752075, 'coordinate': [1252.864, 492.1079, 1294.6238, 518.47095]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.9017605781555176, 'coordinate': [1242.1719, 1473.6951, 1283.02, 1498.6316]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.8999755382537842, 'coordinate': [1269.8164, 220.34933, 1299.8589, 247.01102]}, {'cls_id': 7, 'label': 'formula', 'score': 0.8965252041816711, 'coordinate': [96.00711, 235.49493, 295.43823, 265.60016]}, {'cls_id': 2, 'label': 'text', 'score': 0.8954343199729919, 'coordinate': [696.85693, 1286.2236, 1083.3921, 1310.8643]}, {'cls_id': 7, 'label': 'formula', 'score': 0.8952110409736633, 'coordinate': [166.60979, 129.20242, 511.65692, 156.29672]}, {'cls_id': 2, 'label': 'text', 'score': 0.893648624420166, 'coordinate': [725.64575, 396.18964, 1263.0391, 422.76813]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.8922948837280273, 'coordinate': [634.14124, 427.77087, 661.1686, 454.10022]}, {'cls_id': 2, 'label': 'text', 'score': 0.8892256617546082, 'coordinate': [94.483246, 1058.7595, 441.92313, 1082.4875]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.8878197073936462, 'coordinate': [630.4175, 939.3015, 657.7135, 965.36426]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.8831961154937744, 'coordinate': [630.5835, 1000.95715, 657.4309, 1026.2128]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.8767948150634766, 'coordinate': [634.1024, 575.3833, 660.59094, 601.1677]}, {'cls_id': 7, 'label': 'formula', 'score': 0.873543918132782, 'coordinate': [95.29655, 1320.3627, 264.93008, 1345.8473]}, {'cls_id': 17, 'label': 'formula_number', 'score': 0.8702306151390076, 'coordinate': [633.82825, 730.31525, 659.83215, 755.5485]}, {'cls_id': 7, 'label': 'formula', 'score': 0.8387619853019714, 'coordinate': [365.19897, 268.29675, 515.7938, 296.07013]}, {'cls_id': 7, 'label': 'formula', 'score': 0.8314349055290222, 'coordinate': [1090.509, 1599.1382, 1276.6736, 1622.156]}, {'cls_id': 7, 'label': 'formula', 'score': 0.817135751247406, 'coordinate': [246.175, 161.22958, 314.3764, 186.40591]}, {'cls_id': 3, 'label': 'number', 'score': 0.8042846322059631, 'coordinate': [1297.4036, 7.1497707, 1310.5969, 27.737753]}, {'cls_id': 7, 'label': 'formula', 'score': 0.7970448136329651, 'coordinate': [538.45593, 478.09354, 661.8812, 508.50778]}, {'cls_id': 7, 'label': 'formula', 'score': 0.7644855976104736, 'coordinate': [916.51746, 1618.5188, 1009.62537, 1640.8206]}, {'cls_id': 7, 'label': 'formula', 'score': 0.7423419952392578, 'coordinate': [694.8439, 1612.2507, 861.05334, 1635.9768]}, {'cls_id': 7, 'label': 'formula', 'score': 0.7072376608848572, 'coordinate': [99.72007, 508.21167, 254.91953, 535.74744]}, {'cls_id': 7, 'label': 'formula', 'score': 0.6976271867752075, 'coordinate': [696.8011, 1561.4375, 899.79584, 1586.7349]}, {'cls_id': 7, 'label': 'formula', 'score': 0.6707713007926941, 'coordinate': [1117.0862, 1571.9763, 1191.502, 1594.742]}, {'cls_id': 7, 'label': 'formula', 'score': 0.6338322162628174, 'coordinate': [577.33484, 1274.4131, 602.5636, 1296.7021]}, {'cls_id': 7, 'label': 'formula', 'score': 0.6199935674667358, 'coordinate': [175.28284, 349.82376, 241.24683, 376.6708]}, {'cls_id': 7, 'label': 'formula', 'score': 0.612853467464447, 'coordinate': [773.06287, 595.202, 800.43884, 617.3812]}, {'cls_id': 7, 'label': 'formula', 'score': 0.6107096672058105, 'coordinate': [706.6776, 316.87082, 736.69714, 339.9352]}, {'cls_id': 7, 'label': 'formula', 'score': 0.5520269870758057, 'coordinate': [1263.9711, 314.65167, 1292.7728, 337.3896]}, {'cls_id': 7, 'label': 'formula', 'score': 0.5346108675003052, 'coordinate': [1219.2955, 316.599, 1243.9181, 339.71802]}, {'cls_id': 7, 'label': 'formula', 'score': 0.5195119380950928, 'coordinate': [254.65729, 323.6553, 326.57758, 349.53494]}, {'cls_id': 7, 'label': 'formula', 'score': 0.501812219619751, 'coordinate': [255.8518, 1350.6472, 301.74304, 1375.5286]}]}, 'formula_res_list': [{'rec_formula': '\\begin{aligned}{\\psi_{0}(M)-\\psi_{}(M,z)=}&{{}\\frac{(1-\\epsilon_{r})}{\\epsilon_{r}}\\frac{\\lambda^{2}c^{2}}{t_{\\operatorname{E}}^{2}\\operatorname{l n}(10)}\\times}\\\\ {}&{{}\\int_{0}^{z}d z^{\\prime}\\frac{d t}{d z^{\\prime}}\\left.\\frac{\\partial\\phi}{\\partial L}\\right|_{L=\\lambda M c^{2}/t_{\\operatorname{E}}},}\\\\ \\end{aligned}', 'formula_region_id': 1, 'dt_polys': ([728.5045, 440.9215, 1224.0634, 570.8518],)}, {'rec_formula': '\\begin{aligned}{p(\\operatorname{l o g}_{10}}&{{}M|\\operatorname{l o g}_{10}\\sigma)=\\frac{1}{\\sqrt{2\\pi}\\epsilon_{0}}}\\\\ {}&{{}\\times\\operatorname{e x p}\\left[-\\frac{1}{2}\\left(\\frac{\\operatorname{l o g}_{10}M-a_{\\bullet}-b_{\\bullet}\\operatorname{l o g}_{10}\\sigma}{\\epsilon_{0}}\\right)^{2}\\right].}\\\\ \\end{aligned}', 'formula_region_id': 2, 'dt_polys': ([722.9789, 1333.5085, 1257.1136, 1468.0432],)}, {'rec_formula': '\\psi_{0}(M)=\\int d\\sigma\\frac{p(\\operatorname{l o g}_{10}M|\\operatorname{l o g}_{10}\\sigma)}{M\\operatorname{l o g}(10)}\\frac{d n}{d\\sigma}(\\sigma),', 'formula_region_id': 3, 'dt_polys': ([756.4525, 1211.323, 1188.0428, 1268.2336],)}, {'rec_formula': '\\phi(L)\\equiv\\frac{d n}{d\\operatorname{l o g}_{10}L}=\\frac{\\phi_{*}}{(L/L_{*})^{\\gamma_{1}}+(L/L_{*})^{\\gamma_{2}}}.', 'formula_region_id': 4, 'dt_polys': ([777.51355, 207.87927, 1222.8966, 267.33014],)}, {'rec_formula': '\\begin{aligned}{\\rho_{\\operatorname{B H}}}&{{}=\\int d M\\psi(M)M}\\\\ {}&{{}=\\frac{1-\\epsilon_{r}}{\\epsilon_{r}c^{2}}\\int_{0}^{\\infty}d z\\frac{d t}{d z}\\int d\\operatorname{l o g}_{10}L\\phi(L,z)L,}\\\\ \\end{aligned}', 'formula_region_id': 5, 'dt_polys': ([153.89856, 924.2046, 601.0946, 1036.9038],)}, {'rec_formula': '\\frac{d n}{d\\sigma}d\\sigma=\\psi_{*}\\left(\\frac{\\sigma}{\\sigma_{*}}\\right)^{\\alpha}\\frac{e^{-(\\sigma/\\sigma_{*})^{\\beta}}}{\\Gamma(\\alpha/\\beta)}\\beta\\frac{d\\sigma}{\\sigma}.', 'formula_region_id': 6, 'dt_polys': ([810.86975, 1057.0771, 1175.101, 1117.6631],)}, {'rec_formula': '\\langle\\dot{M}(M,t)\\rangle\\psi(M,t)=\\frac{(1-\\epsilon_{r})}{\\epsilon_{r}c^{2}\\operatorname{l n}(10)}\\phi(L,t)\\frac{d L}{d M}.', 'formula_region_id': 7, 'dt_polys': ([165.26271, 557.8495, 598.1803, 614.35],)}, {'rec_formula': '\\frac{\\partial\\psi}{\\partial t}(M,t)+\\frac{(1-\\epsilon_{r})}{\\epsilon_{r}}\\frac{\\lambda^{2}c^{2}}{t_{\\operatorname{E}}^{2}\\operatorname{l n}(10)}\\left.\\frac{\\partial\\phi}{\\partial L}\\right|_{L=\\lambda M c^{2}/t_{\\operatorname{E}}}=0,', 'formula_region_id': 8, 'dt_polys': ([116.48187, 713.88416, 614.2181, 774.02576],)}, {'rec_formula': '\\operatorname{l o g}_{10}M=a_{\\bullet}+b_{\\bullet}\\operatorname{l o g}_{10}X.', 'formula_region_id': 9, 'dt_polys': ([852.90137, 908.64386, 1131.1882, 933.81793],)}, {'rec_formula': '\\phi(L,t)d\\operatorname{l o g}_{10}L=\\delta(M,t)\\psi(M,t)d M.', 'formula_region_id': 10, 'dt_polys': ([195.28397, 424.81024, 567.697, 451.1291],)}, {'rec_formula': '\\dot{M}\\:=\\:(1\\:-\\:\\epsilon_{r})\\dot{M}_{\\mathrm{a c c}}^{\\mathrm{~\\tiny~\\cdot~}}', 'formula_region_id': 11, 'dt_polys': ([96.00711, 235.49493, 295.43823, 265.60016],)}, {'rec_formula': 't_{E}=\\sigma_{T}c/4\\pi G m_{p}=4.5\\times10^{8}\\mathrm{y r}', 'formula_region_id': 12, 'dt_polys': ([166.60979, 129.20242, 511.65692, 156.29672],)}, {'rec_formula': 'M_{*}=L_{*}t_{E}/\\tilde{\\lambda}c^{2}', 'formula_region_id': 13, 'dt_polys': ([95.29655, 1320.3627, 264.93008, 1345.8473],)}, {'rec_formula': '\\phi(L,t)d\\operatorname{l o g}_{10}L', 'formula_region_id': 14, 'dt_polys': ([365.19897, 268.29675, 515.7938, 296.07013],)}, {'rec_formula': 'a_{\\bullet}=8.32\\pm0.05', 'formula_region_id': 15, 'dt_polys': ([1090.509, 1599.1382, 1276.6736, 1622.156],)}, {'rec_formula': '\\epsilon_{r}\\dot{M}_{\\mathrm{a c c}}', 'formula_region_id': 16, 'dt_polys': ([246.175, 161.22958, 314.3764, 186.40591],)}, {'rec_formula': '\\langle\\dot{M}(M,t)\\rangle=', 'formula_region_id': 17, 'dt_polys': ([538.45593, 478.09354, 661.8812, 508.50778],)}, {'rec_formula': '\\epsilon_{0}=0.38', 'formula_region_id': 18, 'dt_polys': ([916.51746, 1618.5188, 1009.62537, 1640.8206],)}, {'rec_formula': 'b_{\\bullet}=5.64\\dot{\\pm}\\dot{0.32}', 'formula_region_id': 19, 'dt_polys': ([694.8439, 1612.2507, 861.05334, 1635.9768],)}, {'rec_formula': '\\delta(M,t)\\dot{M}(M,t)', 'formula_region_id': 20, 'dt_polys': ([99.72007, 508.21167, 254.91953, 535.74744],)}, {'rec_formula': 'X=\\sigma/200\\mathrm{k m}\\mathrm{~s^{-1}~}', 'formula_region_id': 21, 'dt_polys': ([696.8011, 1561.4375, 899.79584, 1586.7349],)}, {'rec_formula': 'M-\\sigma', 'formula_region_id': 22, 'dt_polys': ([1117.0862, 1571.9763, 1191.502, 1594.742],)}, {'rec_formula': 'L_{*}', 'formula_region_id': 23, 'dt_polys': ([577.33484, 1274.4131, 602.5636, 1296.7021],)}, {'rec_formula': '\\phi(L,t)', 'formula_region_id': 24, 'dt_polys': ([175.28284, 349.82376, 241.24683, 376.6708],)}, {'rec_formula': '\\psi_{0}', 'formula_region_id': 25, 'dt_polys': ([773.06287, 595.202, 800.43884, 617.3812],)}, {'rec_formula': '\\mathrm{A^{\\prime\\prime}}', 'formula_region_id': 26, 'dt_polys': ([706.6776, 316.87082, 736.69714, 339.9352],)}, {'rec_formula': 'L_{*}', 'formula_region_id': 27, 'dt_polys': ([1263.9711, 314.65167, 1292.7728, 337.3896],)}, {'rec_formula': '\\phi_{*}', 'formula_region_id': 28, 'dt_polys': ([1219.2955, 316.599, 1243.9181, 339.71802],)}, {'rec_formula': '\\delta(M,t)', 'formula_region_id': 29, 'dt_polys': ([254.65729, 323.6553, 326.57758, 349.53494],)}, {'rec_formula': '\\phi(L)', 'formula_region_id': 30, 'dt_polys': ([255.8518, 1350.6472, 301.74304, 1375.5286],)}]}}
+```
+
+运行结果参数说明可以参考[2.2 Python脚本方式集成](#22-python)中的结果解释。
+
+可视化结果保存在`save_path`下，其中公式识别的可视化结果如下：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/formula_recognition/04_paddleocr3.png" style="width: 70%"/>
+
+<b> 如果您需要对公式识别产线进行可视化，需要运行如下命令来对LaTeX渲染环境进行安装。目前公式识别产线可视化只支持Ubuntu环境，其他环境暂不支持。对于复杂公式，LaTeX 结果可能包含部分高级的表示，Markdown等环境中未必可以成功显示：</b>
+
+```bash
+sudo apt-get update
+sudo apt-get install texlive texlive-latex-base texlive-xetex latex-cjk-all texlive-latex-extra -y
+```
+
+<b>备注</b>： 由于公式识别可视化过程中需要对每张公式图片进行渲染，因此耗时较长，请您耐心等待。
+
+### 2.2 Python脚本方式集成
+
+命令行方式是为了快速体验查看效果，一般来说，在项目中，往往需要通过代码集成，您可以通过几行代码即可完成产线的快速推理，推理代码如下：
+
+```python
+from paddleocr import FormulaRecognitionPipeline
+
+pipeline = FormulaRecognitionPipeline()
+# ocr = FormulaRecognitionPipeline(use_doc_orientation_classify=True) # 通过 use_doc_orientation_classify 指定是否使用文档方向分类模型
+# ocr = FormulaRecognitionPipeline(use_doc_unwarping=True) # 通过 use_doc_unwarping 指定是否使用文本图像矫正模块
+# ocr = FormulaRecognitionPipeline(device="gpu") # 通过 device 指定模型推理时使用 GPU
+output = pipeline.predict("./general_formula_recognition_001.png")
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_img(save_path="output") ## 保存当前图像的公式可视化结果
+    res.save_to_json(save_path="output") ## 保存当前图像的结构化json结果
+```
+
+上述代码默认使用本地推理引擎 `paddle_static`。如需运行，请先参考[飞桨框架安装说明](../paddlepaddle_installation.md)安装 PaddlePaddle。
+
+如果选择 `transformers` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 Transformers 环境配置，然后执行如下代码：
+
+```python
+from paddleocr import FormulaRecognitionPipeline
+
+pipeline = FormulaRecognitionPipeline(
+    engine="transformers",
+)
+# ocr = FormulaRecognitionPipeline(use_doc_orientation_classify=True) # 通过 use_doc_orientation_classify 指定是否使用文档方向分类模型
+# ocr = FormulaRecognitionPipeline(use_doc_unwarping=True) # 通过 use_doc_unwarping 指定是否使用文本图像矫正模块
+# ocr = FormulaRecognitionPipeline(device="gpu") # 通过 device 指定模型推理时使用 GPU
+output = pipeline.predict("./general_formula_recognition_001.png")
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_img(save_path="output") ## 保存当前图像的公式可视化结果
+    res.save_to_json(save_path="output") ## 保存当前图像的结构化json结果
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+在上述 Python 脚本中，执行了如下几个步骤：
+
+（1）通过 `FormulaRecognitionPipeline()` 实例化公式识别产线对象，具体参数说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>含义：</b>文档方向分类模型的名称。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td>文档方向分类模型的目录路径。如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_batch_size</code></td>
+<td><b>含义：</b>文档方向分类模型的batch size。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td>文本图像矫正模型的名称。如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_batch_size</code></td>
+<td><b>含义：</b>文本图像矫正模型的batch size。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否加载并使用文档方向分类模块。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>含义：</b>版面区域检测模型的名称。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>含义：</b>版面区域检测模型的目录路径。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>版面区域检测的阈值，用于过滤掉低置信度预测结果的阈值。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：如 0.2，表示过滤掉所有阈值小于0.2的目标框；</li>
+<li><b>dict</b>：dict的key为<b>int</b>类型，代表<code>cls_id</code>，val为<b>float</b>类型阈值。如 <code>{0: 0.45，2: 0.48，7: 0.4}</code>，表示对cls_id为0的类别应用阈值0.45、cls_id为2的类别应用阈值0.48、cls_id为7的类别应用阈值0.4；</li>
+<li><b>None</b>：不指定，将使用默认值：0.5。</li>
+</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>版面区域检测是否使用后处理NMS。<br/>
+<b>说明：</b>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>版面区域检测模型检测框的扩张系数。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：任意大于 <code>0</code>  浮点数；</li>
+<li><b>Tuple[float,float]</b>：在横纵两个方向各自的扩张系数；</li>
+<li><b>dict</b>，dict的key为<b>int</b>类型，代表<code>cls_id</code>，value为<b>tuple</b>类型，如<code>{0: (1.1，2.0)}</code>，表示将模型输出的第0类别检测框中心不变，宽度扩张1.1倍，高度扩张2.0倍</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>1.0</code>。</li>
+</ul>
+</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>版面区域检测的重叠框过滤方式。<br/>
+<b>说明：</b>
+<ul>
+<li><b>str</b>：<code>large</code>，<code>small</code>，<code>union</code>，分别表示重叠框过滤时选择保留大框，小框还是同时保留；</li>
+<li><b>dict</b>： dict的key为<b>int</b>类型，代表<code>cls_id</code>，value为<b>str</b>类型，如<code>{0: "large", 2: "small"}</code>，表示对第0类别检测框使用large模式，对第2类别检测框使用small模式；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>large</code>。</li>
+</ul>
+</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_batch_size</code></td>
+<td><b>含义：</b>版面区域检测模型的batch size。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>含义：</b>是否加载并使用版面区域检测模块。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_name</code></td>
+<td><b>含义：</b>公式识别模型的名称。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_model_dir</code></td>
+<td><b>含义：</b>公式识别模型的目录路径。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>formula_recognition_batch_size</code></td>
+<td><b>含义：</b>公式识别模型的batch size。<br/>
+<b>说明：</b>如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+</tbody>
+</table>
+
+（2）调用 公式识别产线对象的 `predict()` 方法进行推理预测，该方法会返回一个结果列表。
+
+另外，产线还提供了 `predict_iter()` 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 `predict_iter()` 返回的是一个 `generator`，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。
+
+以下是 `predict()` 方法的参数及其说明：
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据；</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/demo_image/pipelines/general_formula_recognition_001.png">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)；</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]。</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>含义：</b>是否在推理时使用文档区域检测模块。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否在推理时使用文档方向分类模块。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否在推理时使用文本图像矫正模块。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tr></tr></tbody>
+</table>
+
+（3）对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为`json`文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化。</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效。</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效。</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致。</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效。</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效。</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将结果保存为图像格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径。</td>
+<td>无</td>
+</tr>
+</table>
+<!-- Luxorion-12 修改：markdown格式转换为html格式-->
+<ul>
+  <li>调用<code>print()</code> 方法会将结果打印到终端，打印到终端的内容解释如下：
+    <ol start="1" type="1">
+      <li><code>input_path</code>: <code>(str)</code> 待预测图像的输入路径</li>
+      <li><code>page_index</code>: <code>(Union[int, None])</code> 如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+      <li><code>model_settings</code>: <code>(Dict[str, bool])</code> 配置产线所需的模型参数
+        <ol >
+          <li><code>use_doc_preprocessor</code>: <code>(bool)</code> 控制是否启用文档预处理子产线</li>
+          <li><code>use_layout_detection</code>: <code>(bool)</code> 控制是否启用版面检测模块</li>
+        </ol>
+      </li>
+      <li><code>doc_preprocessor_res</code>: <code>(Dict[str, Union[str, Dict[str, bool], int]])</code> 文档预处理子产线的输出结果。仅当<code>use_doc_preprocessor=True</code>时存在
+        <ol>
+          <li><code>input_path</code>: <code>(str)</code> 文档预处理子接受的图像路径，当输入为<code>numpy.ndarray</code>时，保存为<code>None</code>,此处为<code>None</code></li>
+          <li><code>model_settings</code>: <code>(Dict[str, bool])</code> 文档预处理子的模型配置参数
+            <ul>
+              <li><code>use_doc_orientation_classify</code>: <code>(bool)</code> 控制是否启用文档方向分类</li>
+              <li><code>use_doc_unwarping</code>: <code>(bool)</code> 控制是否启用文本图像扭曲矫正子模块</li>
+            </ul>
+          </li>
+          <li><code>angle</code>: <code>(int)</code> 文文档方向分类的预测结果。启用时取值为[0,1,2,3]，分别对应[0°,90°,180°,270°]；未启用时为-1</li>
+        </ol>
+      </li>
+      <li><code>layout_det_res</code>: <code>(Dict[str, List[Dict]])</code> 版面区域检测模块的输出结果。仅当<code>use_layout_detection=True</code>时存在</li>
+        <ol>
+          <li><code>input_path</code>: <code>(Union[str, None])</code> 版面区域检测模块接收的图像路径，当输入为<code>numpy.ndarray</code>时，保存为<code>None</code></li>
+            <ol>
+                <li><code>cls_id</code>: <code>(int)</code> 版面区域检测预测的类别id</li>
+                <li><code>label</code>: <code>(str)</code> 版面区域检测预测的类别</li>
+                <li><code>score</code>: <code>(float)</code> 版面区域检测预测的类别置信度分数</li>
+                <li><code>coordinate</code>: <code>(List[float])</code> 版面区域检测预测的边界框坐标，格式为[x_min, y_min, x_max, y_max]，其中(x_min, y_min)为左上角坐标，(x_max, y_max) 为右上角坐标</li>
+            </ol>
+          <li><code>formula_res_list</code>: <code>(List[Dict[str, int, List[float]]])</code> 公式识别的预测结果列表</li>
+            <ol>
+                <li><code>rec_formula</code>: <code>(str)</code> 公式识别预测的LaTeX源码</li>
+                <li><code>formula_region_id</code>: <code>(int)</code> 公式识别预测的id编号</li>
+                <li><code>dt_polys</code>:  <code>(List[float])</code> 公式识别预测的边界框坐标，格式为[x_min, y_min, x_max, y_max]，其中(x_min, y_min)为左上角坐标，(x_max, y_max) 为右上角坐标</li>
+            </ol>
+        </li>
+      </li>
+    </ol>
+  </li>
+  <li>调用<code>save_to_json()</code> 方法会将上述内容保存到指定的<code>save_path</code>中，如果指定为目录，则保存的路径为<code>save_path/{your_img_basename}_res.json</code>，如果指定为文件，则直接保存到该文件中。由于json文件不支持保存numpy数组，因此会将其中的<code>numpy.array</code>类型转换为列表形式。</li>
+<li>调用<code>save_to_img()</code> 方法会将可视化结果保存到指定的<code>save_path</code>中，如果指定为目录，则会将版面区域检测可视化图像、全局OCR可视化图像、版面阅读顺序可视化图像等内容保存，如果指定为文件，则直接保存到该文件中。</li>
+
+* 此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">获取预测的 <code>json</code> 格式的结果</td>
+</tr>
+<tr>
+<td rowspan="2"><code>img</code></td>
+<td rowspan="2">获取格式为 <code>dict</code> 的可视化图像</td>
+</tr>
+</table>
+<ul>
+  <li><code>json</code> 属性获取的预测结果为dict类型的数据，相关内容与调用 <code>save_to_json()</code> 方法保存的内容一致。</li>
+  <li><code>img</code> 属性返回的预测结果是一个dict类型的数据。其中，键分别为 <code>ocr_res_img</code> 和 <code>preprocessed_img</code>，对应的值是两个 <code>Image.Image</code> 对象：一个用于显示 OCR 结果的可视化图像，另一个用于展示图像预处理的可视化图像。如果没有使用图像预处理子模块，则dict中只包含 <code>ocr_res_img</code>。</li>
+</ul>
+
+## 3. 开发集成/部署
+
+如果产线可以达到您对产线推理速度和精度的要求，您可以直接进行开发集成/部署。
+
+若您需要将产线直接应用在您的Python项目中，可以参考 [2.2 Python脚本方式](#22-python)中的示例代码。
+
+此外，PaddleOCR 也提供了其他两种部署方式，详细说明如下：
+
+🚀 高性能推理：在实际生产环境中，许多应用对部署策略的性能指标（尤其是响应速度）有着较严苛的标准，以确保系统的高效运行与用户体验的流畅性。为此，PaddleOCR 提供高性能推理功能，旨在对模型推理及前后处理进行深度性能优化，实现端到端流程的显著提速，详细的高性能推理流程请参考[高性能推理](../inference_deployment/local_inference/high_performance_inference.md)。
+
+☁️ 服务化部署：服务化部署是实际生产环境中常见的一种部署形式。通过将推理功能封装为服务，客户端可以通过网络请求来访问这些服务，以获取推理结果。详细的产线服务化部署流程请参考[服务化部署](../inference_deployment/serving/serving.md)。
+
+以下是基础服务化部署的API参考与多语言服务调用示例：
+
+<details><summary>API参考</summary>
+<p>对于服务提供的主要操作：</p>
+<ul>
+<li>HTTP请求方法为POST。</li>
+<li>请求体和响应体均为JSON数据（JSON对象）。</li>
+<li>当请求处理成功时，响应状态码为<code>200</code>，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。固定为<code>0</code>。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。固定为<code>"Success"</code>。</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>操作结果。</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>当请求处理未成功时，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。与响应状态码相同。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。</td>
+</tr>
+</tbody>
+</table>
+<p>服务提供的主要操作如下：</p>
+<ul>
+<li><b><code>infer</code></b></li>
+</ul>
+<p>获取图像公式识别结果。</p>
+<p><code>POST /formula-recognition</code></p>
+<ul>
+<li>请求体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>服务器可访问的图像文件（含 TIFF，多页时按页处理）或 PDF 文件的 URL，或上述类型文件内容的 Base64 编码结果。默认对于超过 10 页的 PDF 或多页 TIFF，仅处理前 10 页。<br /> 要解除页数限制，请在产线配置文件中添加以下配置：
+<pre><code>Serving:
+  extra:
+    max_num_input_imgs: null
+</code></pre>
+</td>
+<td>是</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>文件类型。<code>0</code> 表示 PDF 文件，<code>1</code> 表示图像文件（含 TIFF）。若请求体无此属性，则将根据URL推断文件类型。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_doc_orientation_classify</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_doc_unwarping</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useLayoutDetection</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code>方法的 <code>use_layout_detection</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutThreshold</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中<code>predict</code> 法的<code>layout_threshold</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutNms</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中<code>predict</code> 法的<code>layout_nms</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutUnclipRatio</code></td>
+<td><code>number</code> | <code>array</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code>方法的 <code>layout_unclip_ratio</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutMergeBboxesMode</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中<code>predict</code>方法的 <code>layout_merge_bboxes_mode</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>是否返回可视化结果图以及处理过程中的中间图像等。
+<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>传入 <code>true</code>：返回图像。</li>
+<li>传入 <code>false</code>：不返回图像。</li>
+<li>若请求体中未提供该参数或传入 <code>null</code>：遵循产线配置文件<code>Serving.visualize</code> 的设置。</li>
+</ul>
+<br/>例如，在产线配置文件中添加如下字段：<br/>
+<pre><code>Serving:
+  visualize: False
+</code></pre>
+将默认不返回图像，通过请求体中的<code>visualize</code>参数可以覆盖默认行为。如果请求体和配置文件中均未设置（或请求体传入<code>null</code>、配置文件中未设置），则默认返回图像。
+</td>
+<td>否</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>请求处理成功时，响应体的<code>result</code>具有如下属性：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>formulaRecResults</code></td>
+<td><code>object</code></td>
+<td>公式识别结果。数组长度为1（对于图像输入）或实际处理的文档页数（对于PDF输入）。对于PDF输入，数组中的每个元素依次表示PDF文件中实际处理的每一页的结果。</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>输入数据信息。</td>
+</tr>
+</tbody>
+</table>
+<p>下表中涉及图像的字段（如 <code>outputImages</code>、<code>inputImage</code>）默认以 Base64 字符串内联返回；当服务端开启 URL 返回模式时，相应字段的值变为预签名 URL，字段类型保持不变。配置方式参见 <a href="../inference_deployment/serving/serving.md">服务化部署</a>「以 URL 形式返回二进制内容」一节。</p>
+<p><code>formulaRecResults</code>中的每个元素为一个<code>object</code>，具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>产线对象的 <code>predict</code> 方法生成结果的JSON表示中<code>res</code>字段的简化版本，其中去除了<code>input_path</code>和 <code>page_index</code>字段。</td>
+</tr>
+<tr>
+<td><code>outputImages</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>参见产线预测结果的<code>img</code>属性说明。图像为JPEG格式，使用Base64编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+<tr>
+<td><code>inputImage</code> | <code>null</code></td>
+<td><code>string</code></td>
+<td>输入图像。图像为JPEG格式，使用Base64编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+</tbody>
+</table>
+</details>
+<details><summary>多语言调用服务示例</summary>
+<details>
+<summary>Python</summary>
+
+<pre><code class="language-python">import base64
+import requests
+
+API_URL = "http://localhost:8080/formula-recognition"
+file_path = "./demo.jpg"
+
+with open(file_path, "rb") as file:
+    file_bytes = file.read()
+    file_data = base64.b64encode(file_bytes).decode("ascii")
+
+payload = {"file": file_data, "fileType": 1}
+
+response = requests.post(API_URL, json=payload)
+
+assert response.status_code == 200
+result = response.json()["result"]
+for i, res in enumerate(result["formulaRecResults"]):
+    print(res["prunedResult"])
+    for img_name, img in res["outputImages"].items():
+        img_path = f"{img_name}_{i}.jpg"
+        with open(img_path, "wb") as f:
+            f.write(base64.b64decode(img))
+        print(f"Output image saved at {img_path}")
+</code></pre></details>
+
+<details><summary>C++</summary>
+
+<pre><code class="language-cpp">#include &lt;iostream&gt;
+#include &lt;fstream&gt;
+#include &lt;vector&gt;
+#include &lt;string&gt;
+#include "cpp-httplib/httplib.h" // https://github.com/Huiyicc/cpp-httplib
+#include "nlohmann/json.hpp" // https://github.com/nlohmann/json
+#include "base64.hpp" // https://github.com/tobiaslocker/base64
+
+int main() {
+    httplib::Client client("localhost", 8080);
+
+    const std::string filePath = "./demo.jpg";
+
+    std::ifstream file(filePath, std::ios::binary | std::ios::ate);
+    if (!file) {
+        std::cerr << "Error opening file: " << filePath << std::endl;
+        return 1;
+    }
+
+    std::streamsize size = file.tellg();
+    file.seekg(0, std::ios::beg);
+    std::vector<char> buffer(size);
+    if (!file.read(buffer.data(), size)) {
+        std::cerr << "Error reading file." << std::endl;
+        return 1;
+    }
+
+    std::string bufferStr(buffer.data(), static_cast<size_t>(size));
+    std::string encodedFile = base64::to_base64(bufferStr);
+
+    nlohmann::json jsonObj;
+    jsonObj["file"] = encodedFile;
+    jsonObj["fileType"] = 1;
+
+    auto response = client.Post("/formula-recognition", jsonObj.dump(), "application/json");
+
+    if (response && response->status == 200) {
+        nlohmann::json jsonResponse = nlohmann::json::parse(response->body);
+        auto result = jsonResponse["result"];
+
+        if (!result.is_object() || !result["formulaRecResults"].is_array()) {
+            std::cerr << "Unexpected response format." << std::endl;
+            return 1;
+        }
+
+        for (size_t i = 0; i < result["formulaRecResults"].size(); ++i) {
+            auto res = result["formulaRecResults"][i];
+
+            if (res.contains("prunedResult")) {
+                std::cout << "Recognized formula: " << res["prunedResult"].dump() << std::endl;
+            }
+
+            if (res.contains("outputImages") && res["outputImages"].is_object()) {
+                for (auto& [imgName, imgData] : res["outputImages"].items()) {
+                    std::string outputPath = imgName + "_" + std::to_string(i) + ".jpg";
+                    std::string decodedImage = base64::from_base64(imgData.get<std::string>());
+
+                    std::ofstream outFile(outputPath, std::ios::binary);
+                    if (outFile.is_open()) {
+                        outFile.write(decodedImage.c_str(), decodedImage.size());
+                        outFile.close();
+                        std::cout << "Saved image: " << outputPath << std::endl;
+                    } else {
+                        std::cerr << "Failed to write image: " << outputPath << std::endl;
+                    }
+                }
+            }
+        }
+    } else {
+        std::cerr << "Request failed." << std::endl;
+        if (response) {
+            std::cerr << "HTTP status: " << response->status << std::endl;
+            std::cerr << "Response body: " << response->body << std::endl;
+        }
+        return 1;
+    }
+
+    return 0;
+}
+</code></pre></details>
+
+<details><summary>Java</summary>
+
+<pre><code class="language-java">import okhttp3.*;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.util.Base64;
+
+public class Main {
+    public static void main(String[] args) throws IOException {
+        String API_URL = "http://localhost:8080/formula-recognition";
+        String imagePath = "./demo.jpg";
+
+        File file = new File(imagePath);
+        byte[] fileContent = java.nio.file.Files.readAllBytes(file.toPath());
+        String base64Image = Base64.getEncoder().encodeToString(fileContent);
+
+        ObjectMapper objectMapper = new ObjectMapper();
+        ObjectNode payload = objectMapper.createObjectNode();
+        payload.put("file", base64Image);
+        payload.put("fileType", 1);
+
+        OkHttpClient client = new OkHttpClient();
+        MediaType JSON = MediaType.get("application/json; charset=utf-8");
+        RequestBody body = RequestBody.create(JSON, payload.toString());
+
+        Request request = new Request.Builder()
+                .url(API_URL)
+                .post(body)
+                .build();
+
+        try (Response response = client.newCall(request).execute()) {
+            if (response.isSuccessful()) {
+                String responseBody = response.body().string();
+                JsonNode root = objectMapper.readTree(responseBody);
+                JsonNode result = root.get("result");
+
+                JsonNode formulaRecResults = result.get("formulaRecResults");
+                for (int i = 0; i < formulaRecResults.size(); i++) {
+                    JsonNode item = formulaRecResults.get(i);
+                    int finalI = i;
+
+                    JsonNode prunedResult = item.get("prunedResult");
+                    System.out.println("Pruned Result [" + i + "]: " + prunedResult.toString());
+
+                    JsonNode outputImages = item.get("outputImages");
+                    if (outputImages != null && outputImages.isObject()) {
+                        outputImages.fieldNames().forEachRemaining(imgName -> {
+                            try {
+                                String imgBase64 = outputImages.get(imgName).asText();
+                                byte[] imgBytes = Base64.getDecoder().decode(imgBase64);
+                                String imgPath = imgName + "_" + finalI + ".jpg";
+                                try (FileOutputStream fos = new FileOutputStream(imgPath)) {
+                                    fos.write(imgBytes);
+                                    System.out.println("Saved image: " + imgPath);
+                                }
+                            } catch (IOException e) {
+                                System.err.println("Failed to save image: " + e.getMessage());
+                            }
+                        });
+                    }
+                }
+            } else {
+                System.err.println("Request failed with HTTP code: " + response.code());
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Go</summary>
+
+<pre><code class="language-go">package main
+
+import (
+    "bytes"
+    "encoding/base64"
+    "encoding/json"
+    "fmt"
+    "io/ioutil"
+    "net/http"
+    "os"
+)
+
+func main() {
+    API_URL := "http://localhost:8080/formula-recognition"
+    filePath := "./demo.jpg"
+
+    fileBytes, err := ioutil.ReadFile(filePath)
+    if err != nil {
+        fmt.Printf("Error reading file: %v\n", err)
+        return
+    }
+    fileData := base64.StdEncoding.EncodeToString(fileBytes)
+
+    payload := map[string]interface{}{
+        "file":     fileData,
+        "fileType": 1,
+    }
+    payloadBytes, err := json.Marshal(payload)
+    if err != nil {
+        fmt.Printf("Error marshaling payload: %v\n", err)
+        return
+    }
+
+    client := &http.Client{}
+    req, err := http.NewRequest("POST", API_URL, bytes.NewBuffer(payloadBytes))
+    if err != nil {
+        fmt.Printf("Error creating request: %v\n", err)
+        return
+    }
+    req.Header.Set("Content-Type", "application/json")
+
+    res, err := client.Do(req)
+    if err != nil {
+        fmt.Printf("Error sending request: %v\n", err)
+        return
+    }
+    defer res.Body.Close()
+
+    if res.StatusCode != http.StatusOK {
+        fmt.Printf("Unexpected status code: %d\n", res.StatusCode)
+        return
+    }
+
+    body, err := ioutil.ReadAll(res.Body)
+    if err != nil {
+        fmt.Printf("Error reading response body: %v\n", err)
+        return
+    }
+
+    type FormulaRecResult struct {
+        PrunedResult map[string]interface{} `json:"prunedResult"`
+        OutputImages map[string]string      `json:"outputImages"`
+        InputImage   *string                `json:"inputImage"`
+    }
+
+    type Response struct {
+        Result struct {
+            FormulaRecResults []FormulaRecResult `json:"formulaRecResults"`
+            DataInfo          interface{}        `json:"dataInfo"`
+        } `json:"result"`
+    }
+
+    var respData Response
+    if err := json.Unmarshal(body, &respData); err != nil {
+        fmt.Printf("Error unmarshaling response: %v\n", err)
+        return
+    }
+
+    for i, res := range respData.Result.FormulaRecResults {
+        fmt.Printf("Result %d - prunedResult: %+v\n", i, res.PrunedResult)
+
+        for imgName, imgData := range res.OutputImages {
+            imgBytes, err := base64.StdEncoding.DecodeString(imgData)
+            if err != nil {
+                fmt.Printf("Error decoding image %s_%d: %v\n", imgName, i, err)
+                continue
+            }
+
+            filename := fmt.Sprintf("%s_%d.jpg", imgName, i)
+            if err := os.WriteFile(filename, imgBytes, 0644); err != nil {
+                fmt.Printf("Error saving image %s: %v\n", filename, err)
+                continue
+            }
+            fmt.Printf("Saved image to %s\n", filename)
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>C#</summary>
+
+<pre><code class="language-csharp">using System;
+using System.IO;
+using System.Net.Http;
+using System.Text;
+using System.Threading.Tasks;
+using Newtonsoft.Json.Linq;
+
+class Program
+{
+    static readonly string API_URL = "http://localhost:8080/formula-recognition";
+    static readonly string inputFilePath = "./demo.jpg";
+
+    static async Task Main(string[] args)
+    {
+        var httpClient = new HttpClient();
+
+        byte[] fileBytes = File.ReadAllBytes(inputFilePath);
+        string fileData = Convert.ToBase64String(fileBytes);
+
+        var payload = new JObject
+        {
+            { "file", fileData },
+            { "fileType", 1 }
+        };
+        var content = new StringContent(payload.ToString(), Encoding.UTF8, "application/json");
+
+        HttpResponseMessage response = await httpClient.PostAsync(API_URL, content);
+        response.EnsureSuccessStatusCode();
+
+        string responseBody = await response.Content.ReadAsStringAsync();
+        JObject jsonResponse = JObject.Parse(responseBody);
+
+        JArray formulaRecResults = (JArray)jsonResponse["result"]["formulaRecResults"];
+        for (int i = 0; i < formulaRecResults.Count; i++)
+        {
+            var res = formulaRecResults[i];
+            Console.WriteLine($"[{i}] prunedResult:\n{res["prunedResult"]}");
+
+            JObject outputImages = res["outputImages"] as JObject;
+            if (outputImages != null)
+            {
+                foreach (var img in outputImages)
+                {
+                    string imgName = img.Key;
+                    string base64Img = img.Value?.ToString();
+                    if (!string.IsNullOrEmpty(base64Img))
+                    {
+                        string imgPath = $"{imgName}_{i}.jpg";
+                        byte[] imageBytes = Convert.FromBase64String(base64Img);
+                        File.WriteAllBytes(imgPath, imageBytes);
+                        Console.WriteLine($"Output image saved at {imgPath}");
+                    }
+                }
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Node.js</summary>
+
+<pre><code class="language-js">const axios = require('axios');
+const fs = require('fs');
+const path = require('path');
+
+const API_URL = 'http://localhost:8080/formula-recognition';
+const inputFilePath = './demo.jpg';
+const fileType = 1;
+
+function encodeImageToBase64(filePath) {
+  const bitmap = fs.readFileSync(filePath);
+  return Buffer.from(bitmap).toString('base64');
+}
+
+const payload = {
+  file: encodeImageToBase64(inputFilePath),
+  fileType: fileType
+};
+
+axios.post(API_URL, payload)
+  .then((response) => {
+    const resultArray = response.data.result.formulaRecResults;
+
+    resultArray.forEach((res, index) => {
+      console.log(`\n[${index}] prunedResult:`);
+      console.log(res.prunedResult);
+
+      const outputImages = res.outputImages;
+      if (outputImages) {
+        Object.entries(outputImages).forEach(([imgName, base64Img]) => {
+          const outputPath = `${imgName}_${index}.jpg`;
+          fs.writeFileSync(outputPath, Buffer.from(base64Img, 'base64'));
+          console.log(`Saved output image: ${outputPath}`);
+        });
+      } else {
+        console.log(`[${index}] outputImages is null`);
+      }
+    });
+  })
+  .catch((error) => {
+    console.error('API error：', error.message);
+  });
+</code></pre></details>
+
+<details><summary>PHP</summary>
+
+<pre><code class="language-php">&lt;?php
+
+$API_URL = "http://localhost:8080/formula-recognition";
+$image_path = "./demo.jpg";
+
+$image_data = base64_encode(file_get_contents($image_path));
+$payload = array("file" => $image_data, "fileType" => 1);
+
+$ch = curl_init($API_URL);
+curl_setopt($ch, CURLOPT_POST, true);
+curl_setopt($ch, CURLOPT_POSTFIELDS, json_encode($payload));
+curl_setopt($ch, CURLOPT_HTTPHEADER, array('Content-Type: application/json'));
+curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+$response = curl_exec($ch);
+curl_close($ch);
+
+$result = json_decode($response, true)["result"]["formulaRecResults"];
+
+foreach ($result as $i => $item) {
+    echo "[$i] prunedResult:\n";
+    print_r($item["prunedResult"]);
+
+    if (!empty($item["outputImages"])) {
+        foreach ($item["outputImages"] as $img_name => $base64_img) {
+            $img_path = "{$img_name}_{$i}.jpg";
+            file_put_contents($img_path, base64_decode($base64_img));
+            echo "Output image saved at $img_path\n";
+        }
+    } else {
+        echo "No outputImages found for item $i\n";
+    }
+}
+?&gt;
+</code></pre></details>
+</details>
+<br/>
+
+
+## 4. 二次开发
+
+如果公式识别产线提供的默认模型权重在您的场景中，精度或速度不满意，您可以尝试利用<b>您自己拥有的特定领域或应用场景的数据</b>对现有模型进行进一步的<b>微调</b>，以提升公式识别产线的在您的场景中的识别效果。
+
+### 4.1 模型微调
+由于公式识别产线包含若干模块，模型产线的效果如果不及预期，可能来自于其中任何一个模块。您可以对识别效果差的图片进行分析，进而确定是哪个模块存在问题，并参考以下表格中对应的微调教程链接进行模型微调。
+
+
+<table>
+<thead>
+<tr>
+<th>情形</th>
+<th>微调模块</th>
+<th>微调参考链接</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>公式存在漏检</td>
+<td>版面区域检测模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/module_usage/layout_detection.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>公式内容不准</td>
+<td>公式识别模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/module_usage/formula_recognition.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>整图旋转矫正不准</td>
+<td>文档图像方向分类模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/module_usage/doc_img_orientation_classification.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>图像扭曲矫正不准</td>
+<td>文本图像矫正模块</td>
+<td>暂不支持微调</td>
+</tr>
+</tbody>
+</table>
+
+### 4.2 模型应用
+
+当您使用私有数据集完成微调训练后，可获得本地模型权重文件，然后可以通过参数指定本地模型保存路径的方式，或者通过自定义产线配置文件的方式，使用微调后的模型权重。
+
+#### 4.2.1 通过参数指定本地模型路径
+
+在初始化产线对象时，通过参数指定本地模型路径。以公式识别模型微调后的权重的使用方法为例，示例如下：
+
+命令行方式:
+
+```bash
+# 通过 --formula_recognition_model_dir 指定本地模型路径
+paddleocr formula_recognition_pipeline -i ./general_formula_recognition_001.png --formula_recognition_model_dir your_formula_recognition_model_path
+
+# 默认使用 PP-FormulaNet_plus-M 模型作为默认公式识别模型，如果微调的不是该模型，通过 --formula_recognition_model_name 修改模型名称
+paddleocr formula_recognition_pipeline -i ./general_formula_recognition_001.png --formula_recognition_model_name PP-FormulaNet_plus-M --formula_recognition_model_dir your_ppformulanet_plus-m_formula_recognition_model_path
+```
+
+脚本方式：
+
+```python
+
+from paddleocr import FormulaRecognitionPipeline
+
+# 通过 formula_recognition_model_dir 指定本地模型路径
+pipeline = FormulaRecognitionPipeline(formula_recognition_model_dir="./your_formula_recognition_model_path")
+output = pipeline.predict("./general_formula_recognition_001.png")
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_img(save_path="output") ## 保存当前图像的公式可视化结果
+    res.save_to_json(save_path="output") ## 保存当前图像的结构化json结果
+# 默认使用 PP-FormulaNet_plus-M 模型作为默认公式识别模型，如果微调的不是该模型，通过 formula_recognition_model_name 修改模型名称
+# pipeline = FormulaRecognitionPipeline(formula_recognition_model_name="PP-FormulaNet_plus-M", formula_recognition_model_dir="./your_ppformulanet_plus-m_formula_recognition_model_path")
+
+```
+
+
+#### 4.2.2 通过配置文件指定本地模型路径
+
+
+1.获取产线配置文件
+
+可调用 PaddleOCR 中 公式识别 产线对象的 `export_paddlex_config_to_yaml` 方法，将当前产线配置导出为 YAML 文件：
+
+```Python
+from paddleocr import FormulaRecognitionPipeline
+
+pipeline = FormulaRecognitionPipeline()
+pipeline.export_paddlex_config_to_yaml("FormulaRecognitionPipeline.yaml")
+```
+
+2.修改配置文件
+
+在得到默认的产线配置文件后，将微调后模型权重的本地路径替换至产线配置文件中的对应位置即可。例如
+
+```yaml
+......
+SubModules:
+  FormulaRecognition:
+    batch_size: 5
+    model_dir: null # 替换为微调后的公式模型权重路径
+    model_name: PP-FormulaNet_plus-M # 如果微调的模型名称与默认模型名称不同，请一并修改此处
+    module_name: formula_recognition
+  LayoutDetection:
+    batch_size: 1
+    layout_merge_bboxes_mode: large
+    layout_nms: true
+    layout_unclip_ratio: 1.0
+    model_dir: null # 替换为微调后的版面区域检测模型权重路径
+    model_name: PP-DocLayout_plus-L # 如果微调的模型名称与默认模型名称不同，请一并修改此处
+    module_name: layout_detection
+    threshold: 0.5
+SubPipelines:
+  DocPreprocessor:
+    SubModules:
+      DocOrientationClassify:
+        batch_size: 1
+        model_dir: null # 替换为微调后的文档图像方向分类模型权重路径
+        model_name: PP-LCNet_x1_0_doc_ori # 如果微调的模型名称与默认模型名称不同，请一并修改此处
+        module_name: doc_text_orientation
+      DocUnwarping:
+        batch_size: 1
+        model_dir: null 
+        model_name: UVDoc  
+        module_name: image_unwarping
+    pipeline_name: doc_preprocessor
+    use_doc_orientation_classify: true
+    use_doc_unwarping: true
+pipeline_name: formula_recognition
+use_doc_preprocessor: true
+use_layout_detection: true
+......
+```
+
+在产线配置文件中，不仅包含 PaddleOCR CLI 和 Python API 支持的参数，还可进行更多高级配置，具体信息可在 [PaddleX模型产线使用概览](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/pipeline_develop_guide.html) 中找到对应的产线使用教程，参考其中的详细说明，根据需求调整各项配置。
+
+3.在 CLI 中加载产线配置文件
+
+在修改完成配置文件后，通过命令行的 `--paddlex_config` 参数指定修改后的产线配置文件的路径，PaddleOCR 会读取其中的内容作为产线配置。示例如下：
+
+```bash
+paddleocr formula_recognition_pipeline -i ./general_formula_recognition_001.png --paddlex_config FormulaRecognitionPipeline.yaml 
+```
+
+4.在 Python API 中加载产线配置文件
+
+初始化产线对象时，可通过 `paddlex_config` 参数传入 PaddleX 产线配置文件路径或配置dict，PaddleOCR 会读取其中的内容作为产线配置。示例如下：
+
+```python
+from paddleocr import  FormulaRecognitionPipeline
+
+pipeline =  FormulaRecognitionPipeline(paddlex_config="FormulaRecognitionPipeline.yaml")
+output = pipeline.predict("./general_formula_recognition_001.png")
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_img(save_path="output") ## 保存当前图像的公式可视化结果
+    res.save_to_json(save_path="output") ## 保存当前图像的结构化json结果
+```
diff --git a/docs/version3.x/pipeline_usage/pipeline_overview.en.md b/docs/version3.x/pipeline_usage/pipeline_overview.en.md
new file mode 100644
index 0000000..402b5da
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/pipeline_overview.en.md
@@ -0,0 +1,3 @@
+# Pipeline Overview
+
+A pipeline is a practical functional implementation composed of one or more modules. Through reasonable module combination and configuration, pipelines can meet the needs of complex application scenarios, such as technological applications like Optical Character Recognition (OCR). Pipelines not only demonstrate the integrated application of basic modules but also support capabilities such as high-performance inference and service-oriented deployment, providing users with higher development efficiency and broader application possibilities.
diff --git a/docs/version3.x/pipeline_usage/pipeline_overview.md b/docs/version3.x/pipeline_usage/pipeline_overview.md
new file mode 100644
index 0000000..71d797d
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/pipeline_overview.md
@@ -0,0 +1,3 @@
+# 产线概述
+
+产线是由一个或多个模块组合而成的实际功能实现。通过合理的模块组合与配置，产线能够满足复杂应用场景的需求，比如光学字符识别（OCR）等技术应用。产线不仅体现了基础模块的整合应用，还支持高性能推理和服务化部署等能力，赋予用户更高的开发效率和更广泛的应用可能性。
diff --git a/docs/version3.x/pipeline_usage/seal_recognition.en.md b/docs/version3.x/pipeline_usage/seal_recognition.en.md
new file mode 100644
index 0000000..ed07df4
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/seal_recognition.en.md
@@ -0,0 +1,2565 @@
+---
+comments: true
+---
+
+# Seal Text Recognition Pipeline Usage Tutorial
+
+## 1. Introduction to Seal Text Recognition Pipeline
+Seal text recognition is a technology that automatically extracts and recognizes the content of seals from documents or images. The recognition of seal text is part of document processing and has many applications in various scenarios, such as contract comparison, warehouse entry and exit review, and invoice reimbursement review.
+
+The seal text recognition pipeline is used to recognize the text content of seals, extracting the text information from seal images and outputting it in text form. This pipeline integrates the industry-renowned end-to-end OCR system PP-OCRv4, supporting the detection and recognition of curved seal text. Additionally, this pipeline integrates an optional layout region localization module, which can accurately locate the layout position of the seal within the entire document. It also includes optional document image orientation correction and distortion correction functions. Based on this pipeline, millisecond-level accurate text content prediction can be achieved on a CPU. This pipeline also provides flexible service deployment methods, supporting the use of multiple programming languages on various hardware. Moreover, it offers custom development capabilities, allowing you to train and fine-tune on your own dataset based on this pipeline, and the trained model can be seamlessly integrated.
+
+<img src="https://paddle-model-ecology.bj.bcebos.com/paddlex/PaddleX3.0/doc_images/practical_tutorial/PP-ChatOCRv3_doc_seal/01.png"/>
+
+<b>The seal text recognition</b> pipeline includes a seal text detection module and a text recognition module, as well as optional layout detection module, document image orientation classification module, and text image correction module.
+
+- [Seal Text Detection Module](../module_usage/seal_text_detection.en.md)
+- [Text Recognition Module](../module_usage/text_recognition.en.md)
+- [Layout Detection Module](../module_usage/layout_detection.en.md) (Optional)
+- [Document Image Orientation Classification Module](../module_usage/doc_img_orientation_classification.en.md) (Optional)
+- [Text Image Unwarping Module](../module_usage/text_image_unwarping.en.md) (Optional)
+
+In this pipeline, you can choose the model to use based on the benchmark data below.
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing.
+> In the inference time columns labeled [Regular Mode / High-Performance Mode], the Regular Mode values correspond to the local `paddle_static` inference engine.
+
+<details>
+<summary> <b>Layout Region Detection Module (Optional):</b></summary>
+
+* <b>Layout detection model, including 20 common categories: document title, paragraph title, text, page number, abstract, table of contents, references, footnotes, header, footer, algorithm, formula, formula number, image, table, figure and table title (figure title, table title, and chart title), seal, chart, sidebar text, and reference content</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout_plus-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout_plus-L_pretrained.pdparams">Training Model</a></td>
+<td>83.2</td>
+<td>53.03 / 17.23</td>
+<td>634.62 / 378.32</td>
+<td>126.01</td>
+<td>A higher precision layout region localization model based on RT-DETR-L trained on a self-built dataset including Chinese and English papers, multi-column magazines, newspapers, PPTs, contracts, books, exam papers, research reports, ancient books, Japanese documents, and vertical text documents</td>
+</tr>
+<tr>
+</tbody>
+</table>
+
+* <b>Layout detection model, including 23 common categories: document title, paragraph title, text, page number, abstract, table of contents, references, footnotes, header, footer, algorithm, formula, formula number, image, chart title, table, table title, seal, chart title, chart, header image, footer image, sidebar text</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-L_pretrained.pdparams">Training Model</a></td>
+<td>90.4</td>
+<td>33.59 / 33.59</td>
+<td>503.01 / 251.08</td>
+<td>123.76</td>
+<td>A high precision layout region localization model based on RT-DETR-L trained on a self-built dataset including Chinese and English papers, magazines, contracts, books, exam papers, and research reports</td>
+</tr>
+<tr>
+<td>PP-DocLayout-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-M_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-M_pretrained.pdparams">Training Model</a></td>
+<td>75.2</td>
+<td>13.03 / 4.72</td>
+<td>43.39 / 24.44</td>
+<td>22.578</td>
+<td>A balanced model of accuracy and efficiency based on PicoDet-L trained on a self-built dataset including Chinese and English papers, magazines, contracts, books, exam papers, and research reports</td>
+</tr>
+<tr>
+<td>PP-DocLayout-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-S_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-S_pretrained.pdparams">Training Model</a></td>
+<td>70.9</td>
+<td>11.54 / 3.86</td>
+<td>18.53 / 6.29</td>
+<td>4.834</td>
+<td>A highly efficient layout region localization model based on PicoDet-S trained on a self-built dataset including Chinese and English papers, magazines, contracts, books, exam papers, and research reports</td>
+</tr>
+</tbody>
+</table>
+
+>❗ Listed above are the <b>4 core models</b> that are the focus of the layout detection module, which supports a total of <b>13 full models</b>, including multiple models with pre-defined different categories, among which 9 models include the seal category. Apart from the 3 core models mentioned above, the remaining models are as follows:
+
+<details><summary> 👉Details of the Model List</summary>
+
+* <b>3-class layout detection model, including table, image, seal</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_3cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_3cls_pretrained.pdparams">Training Model</a></td>
+<td>88.2</td>
+<td>8.43 / 3.44</td>
+<td>17.60 / 6.51</td>
+<td>4.8</td>
+<td>A highly efficient layout region localization model based on the lightweight PicoDet-S model trained on a self-built dataset including Chinese and English papers, magazines, and research reports</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_3cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_3cls_pretrained.pdparams">Training Model</a></td>
+<td>89.0</td>
+<td>12.80 / 9.57</td>
+<td>45.04 / 23.86</td>
+<td>22.6</td>
+<td>An efficiency-accuracy balanced layout region localization model based on PicoDet-L trained on a self-built dataset including Chinese and English papers, magazines, and research reports</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_3cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_3cls_pretrained.pdparams">Training Model</a></td>
+<td>95.8</td>
+<td>114.80 / 25.65</td>
+<td>924.38 / 924.38</td>
+<td>470.1</td>
+<td>A high precision layout region localization model based on RT-DETR-H trained on a self-built dataset including Chinese and English papers, magazines, and research reports</td>
+</tr>
+</tbody></table>
+
+* <b>17-class region detection model, including 17 common layout categories: paragraph title, image, text, number, abstract, content, chart title, formula, table, table title, references, document title, footnote, header, algorithm, footer, seal</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>87.4</td>
+<td>8.80 / 3.62</td>
+<td>17.51 / 6.35</td>
+<td>4.8</td>
+<td>A highly efficient layout region localization model based on the lightweight PicoDet-S model trained on a self-built dataset including Chinese and English papers, magazines, and research reports</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>89.0</td>
+<td>12.60 / 10.27</td>
+<td>43.70 / 24.42</td>
+<td>22.6</td>
+<td>An efficiency-accuracy balanced layout region localization model based on PicoDet-L trained on a self-built dataset including Chinese and English papers, magazines, and research reports</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>98.3</td>
+<td>115.29 / 101.18</td>
+<td>964.75 / 964.75</td>
+<td>470.2</td>
+<td>A high precision layout region localization model based on RT-DETR-H trained on a self-built dataset including Chinese and English papers, magazines, and research reports</td>
+</tr>
+</tbody>
+</table>
+</details>
+</details>
+
+<details>
+<summary> <b>Document Image Orientation Classification Module (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Top-1 Acc (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">Training Model</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>A document image classification model based on PP-LCNet_x1_0, containing four categories: 0 degrees, 90 degrees, 180 degrees, and 270 degrees</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary> <b>Text Image Correction Module (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>CER</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">Training Model</a></td>
+<td>0.179</td>
+<td>19.05 / 19.05</td>
+<td>- / 869.82</td>
+<td>30.3</td>
+<td>A high precision text image correction model</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary> <b>Seal Text Detection Module:</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Detection Hmean (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv4_server_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_seal_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_seal_det_pretrained.pdparams">Training Model</a></td>
+<td>98.40</td>
+<td>124.64 / 91.57</td>
+<td>545.68 / 439.86</td>
+<td>109</td>
+<td>PP-OCRv4 server-side seal text detection model, with higher accuracy, suitable for deployment on better servers</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_seal_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_seal_det_pretrained.pdparams">Training Model</a></td>
+<td>96.36</td>
+<td>9.70 / 3.56</td>
+<td>50.38 / 19.64</td>
+<td>4.7</td>
+<td>PP-OCRv4 mobile-side seal text detection model, with higher efficiency, suitable for deployment on the edge</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Text Recognition Module:</b></summary>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">Training Model</a></td>
+<td>86.38</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec is a new generation text recognition model. This model aims to efficiently and accurately support the recognition of four major languages: Simplified Chinese, Traditional Chinese, English, and Japanese, as well as complex text scenes like handwriting, vertical text, pinyin, and rare characters with a single model. It balances recognition effectiveness, inference speed, and model robustness, providing efficient and accurate technical support for document understanding in various scenarios.</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>81.29</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">Training Model</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc is trained on a mix of more Chinese document data and PP-OCR training data based on PP-OCRv4_server_rec, enhancing recognition capabilities for some traditional Chinese characters, Japanese, and special characters, supporting over 15,000+ characters. Besides improving document-related text recognition, it also enhances general text recognition capabilities</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>PP-OCRv4 lightweight recognition model, with high inference efficiency, can be deployed on multiple hardware devices, including edge devices</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">Training Model</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>PP-OCRv4 server-side model, with high inference accuracy, can be deployed on various servers</td>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>An ultra-lightweight English recognition model trained based on the PP-OCRv4 recognition model, supporting English and number recognition</td>
+</tr>
+</table>
+
+> ❗ Listed above are the <b>6 core models</b> that are the focus of the text recognition module, which supports a total of <b>20 full models</b>, including multiple multi-language text recognition models, with the complete model list as follows:
+
+<details><summary> 👉Details of the Model List</summary>
+
+* <b>PP-OCRv5 Multi-Scene Model</b>
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Chinese Recognition Avg Accuracy(%)</th>
+<th>English Recognition Avg Accuracy(%)</th>
+<th>Traditional Chinese Recognition Avg Accuracy(%)</th>
+<th>Japanese Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">Training Model</a></td>
+<td>86.38</td>
+<td>64.70</td>
+<td>93.29</td>
+<td>60.35</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec is a new generation text recognition model. This model aims to efficiently and accurately support the recognition of four major languages: Simplified Chinese, Traditional Chinese, English, and Japanese, as well as complex text scenes like handwriting, vertical text, pinyin, and rare characters with a single model. It balances recognition effectiveness, inference speed, and model robustness, providing efficient and accurate technical support for document understanding in various scenarios.</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>81.29</td>
+<td>66.00</td>
+<td>83.55</td>
+<td>54.65</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+</table>
+
+* <b>Chinese Recognition Model</b>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">Training Model</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc is trained on a mix of more Chinese document data and PP-OCR training data based on PP-OCRv4_server_rec, enhancing recognition capabilities for some traditional Chinese characters, Japanese, and special characters, supporting over 15,000+ characters. Besides improving document-related text recognition, it also enhances general text recognition capabilities</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>PP-OCRv4 lightweight recognition model, with high inference efficiency, can be deployed on multiple hardware devices, including edge devices</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">Training Model</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>PP-OCRv4 server-side model, with high inference accuracy, can be deployed on various servers</td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>72.96</td>
+<td>3.89 / 1.16</td>
+<td>8.72 / 3.56</td>
+<td>10.3</td>
+<td>PP-OCRv3 lightweight recognition model, with high inference efficiency, can be deployed on multiple hardware devices, including edge devices</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>ch_SVTRv2_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_SVTRv2_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_SVTRv2_rec_pretrained.pdparams">Training Model</a></td>
+<td>68.81</td>
+<td>10.38 / 8.31</td>
+<td>66.52 / 30.83</td>
+<td>80.5</td>
+<td rowspan="1">
+SVTRv2 is a server-side text recognition model developed by the OpenOCR team from Fudan University's Visual and Learning Lab (FVL), which won first place in the PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition, improving the end-to-end recognition accuracy on the A leaderboard by 6% compared to PP-OCRv4.
+</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>ch_RepSVTR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_RepSVTR_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_RepSVTR_rec_pretrained.pdparams">Training Model</a></td>
+<td>65.07</td>
+<td>6.29 / 1.57</td>
+<td>20.64 / 5.40</td>
+<td>48.8</td>
+<td rowspan="1">RepSVTR text recognition model is a mobile-side text recognition model based on SVTRv2, which won first place in the PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition, improving the end-to-end recognition accuracy on the B leaderboard by 2.5% compared to PP-OCRv4, with comparable inference speed.</td>
+</tr>
+</table>
+
+* <b>English Recognition Model</b>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td> 70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>An ultra-lightweight English recognition model trained based on the PP-OCRv4 recognition model, supporting English and number recognition</td>
+</tr>
+<tr>
+<td>en_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>70.69</td>
+<td>3.56 / 0.78</td>
+<td>8.44 / 5.78</td>
+<td>17.3</td>
+<td>An ultra-lightweight English recognition model trained based on the PP-OCRv3 recognition model, supporting English and number recognition</td>
+</tr>
+</table>
+
+* <b>Multilingual Recognition Model</b>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>korean_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>60.21</td>
+<td>3.73 / 0.98</td>
+<td>8.76 / 2.91</td>
+<td>9.6</td>
+<td>An ultra-lightweight Korean recognition model trained based on the PP-OCRv3 recognition model, supporting Korean and number recognition</td>
+</tr>
+<tr>
+<td>japan_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+japan_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/japan_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>45.69</td>
+<td>3.86 / 1.01</td>
+<td>8.62 / 2.92</td>
+<td>9.8</td>
+<td>An ultra-lightweight Japanese recognition model trained based on the PP-OCRv3 recognition model, supporting Japanese and number recognition</td>
+</tr>
+<tr>
+<td>chinese_cht_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+chinese_cht_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/chinese_cht_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>82.06</td>
+<td>3.90 / 1.16</td>
+<td>9.24 / 3.18</td>
+<td>10.8</td>
+<td>An ultra-lightweight Traditional Chinese recognition model trained based on the PP-OCRv3 recognition model, supporting Traditional Chinese and number recognition</td>
+</tr>
+<tr>
+<td>te_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+te_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>95.88</td>
+<td>3.59 / 0.81</td>
+<td>8.28 / 6.21</td>
+<td>8.7</td>
+<td>An ultra-lightweight Telugu recognition model trained based on the PP-OCRv3 recognition model, supporting Telugu and number recognition</td>
+</tr>
+<tr>
+<td>ka_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ka_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ka_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>96.96</td>
+<td>3.49 / 0.89</td>
+<td>8.63 / 2.77</td>
+<td>17.4</td>
+<td>An ultra-lightweight Kannada recognition model trained based on the PP-OCRv3 recognition model, supporting Kannada and number recognition</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ta_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>76.83</td>
+<td>3.49 / 0.86</td>
+<td>8.35 / 3.41</td>
+<td>8.7</td>
+<td>An ultra-lightweight Tamil recognition model trained based on the PP-OCRv3 recognition model, supporting Tamil and number recognition</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>76.93</td>
+<td>3.53 / 0.78</td>
+<td>8.50 / 6.83</td>
+<td>8.7</td>
+<td>An ultra-lightweight Latin recognition model trained based on the PP-OCRv3 recognition model, supporting Latin and number recognition</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+arabic_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>73.55</td>
+<td>3.60 / 0.83</td>
+<td>8.44 / 4.69</td>
+<td>17.3</td>
+<td>An ultra-lightweight Arabic letter recognition model trained based on the PP-OCRv3 recognition model, supporting Arabic letters and number recognition</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+cyrillic_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>94.28</td>
+<td>3.56 / 0.79</td>
+<td>8.22 / 2.76</td>
+<td>8.7</td>
+<td>An ultra-lightweight Cyrillic letter recognition model trained based on the PP-OCRv3 recognition model, supporting Cyrillic letters and number recognition</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+devanagari_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>96.44</td>
+<td>3.60 / 0.78</td>
+<td>6.95 / 2.87</td>
+<td>8.7</td>
+<td>An ultra-lightweight Devanagari letter recognition model trained based on the PP-OCRv3 recognition model, supporting Devanagari letters and number recognition</td>
+</tr>
+</table>
+</details>
+</details>
+
+<details>
+<summary> <b>Test Environment Description:</b></summary>
+
+  <ul>
+      <li><b>Performance Test Environment</b>
+          <ul>
+            <li><strong>Test Dataset:
+             </strong>
+                <ul>
+                  <li>Document Image Orientation Classification Model: Self-built internal dataset covering multiple scenarios such as documents and certificates, containing 1000 images.</li>
+                  <li>Text Image Correction Model: <a href="https://www3.cs.stonybrook.edu/~cvl/docunet.html">DocUNet</a>.</li>
+                   <li>Layout Region Detection Model: PaddleOCR self-built layout region detection dataset, containing 500 common document type images such as Chinese and English papers, magazines, contracts, books, exam papers, and research reports.</li>
+                    <li>3-Class Layout Detection Model: PaddleOCR self-built layout region detection dataset, containing 1154 common document type images such as Chinese and English papers, magazines, and research reports.</li>
+                  <li>17-Class Region Detection Model: PaddleOCR self-built layout region detection dataset, containing 892 common document type images such as Chinese and English papers, magazines, and research reports.</li>
+                  <li>Text Detection Model: PaddleOCR self-built Chinese dataset covering multiple scenarios such as street scenes, web images, documents, and handwriting, where detection includes 500 images.</li>
+                   <li>Chinese Recognition Model: PaddleOCR self-built Chinese dataset covering multiple scenarios such as street scenes, web images, documents, and handwriting, where text recognition includes 11,000 images.</li>
+                    <li>ch_SVTRv2_rec: <a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition</a> A leaderboard evaluation set.</li>
+                  <li>ch_RepSVTR_rec: <a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition</a> B leaderboard evaluation set.</li>
+                  <li>English Recognition Model: Self-built internal English dataset.</li>
+                   <li>Multilingual Recognition Model: Self-built internal multilingual dataset.</li>
+                    <li>Text Line Orientation Classification Model: Self-built internal dataset covering multiple scenarios such as documents and certificates, containing 1000 images.</li>
+                   <li>Seal Text Detection Model: Self-built internal dataset containing 500 circular seal images.</li>
+                </ul>
+             </li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>Inference Mode Description</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration</th>
+            <th>CPU Configuration</th>
+            <th>Acceleration Technology Combination</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Regular Mode</td>
+            <td>FP32 Precision / No TRT Acceleration</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td><code>paddle_static</code></td>
+        </tr>
+        <tr>
+            <td>High-Performance Mode</td>
+            <td>Optimal combination of prior precision type and acceleration strategy</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Select optimal prior backend (Paddle/OpenVINO/TRT, etc.)</td>
+        </tr>
+    </tbody>
+</table>
+</details>
+
+<br />
+<b>If you are more concerned with model accuracy, please choose a model with higher accuracy. If you are more concerned with inference speed, please choose a model with faster inference speed. If you are more concerned with model storage size, please choose a model with smaller storage size</b>.
+
+
+## 2. Quick Start
+
+Before using the seal text recognition pipeline locally, please ensure that you have completed the installation of the wheel package according to the [installation tutorial](../installation.md). If you prefer to install dependencies selectively, please refer to the relevant instructions in the installation documentation. The corresponding dependency group for this pipeline is `doc-parser`. Once the installation is complete, you can experience it locally via the command line or integrate it with Python.
+
+Please note: If you encounter issues such as the program becoming unresponsive, unexpected program termination, running out of memory resources, or extremely slow inference during execution, please try adjusting the configuration according to the documentation, such as disabling unnecessary features or using lighter-weight models.
+
+### 2.1 Command Line Experience
+
+You can quickly experience the seal_recognition pipeline effect with a single command:
+
+```bash
+paddleocr seal_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/seal_text_det.png \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False
+
+# Use --device to specify the use of GPU for model inference.
+paddleocr seal_recognition -i ./seal_text_det.png --device gpu
+```
+
+The examples above use the local `paddle_static` inference engine by default. To run them, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr seal_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/seal_text_det.png \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False \
+    --engine transformers
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+<details><summary><b>The command line supports more parameter settings. Click to expand for detailed explanations of command line parameters.</b></summary>
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Parameter Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted, required.<br/>
+<b>Description:</b>
+Local path of image or PDF file, e.g., <code>/root/data/img.jpg</code>; <b>URL link</b>, e.g., network URL of image or PDF file: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/seal_text_det.png">Example</a>; <b>Local directory</b>, the directory should contain images to be predicted, e.g., local path: <code>/root/data/</code> (currently does not support prediction of PDF files in directories; PDF files must be specified with a specific file path).
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td>
+<b>Meaning:</b>Specify the path to save the inference results file.<br/>
+<b>Description:</b> 
+If not set, the inference results will not be saved locally.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td> 
+<b>Meaning:</b>The name of the document orientation classification model.<br/>
+<b>Description:</b> 
+If not set, the default model in pipeline will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the document orientation classification model.<br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td> <b>Meaning:</b>The name of the text image unwarping model. <br/>
+<b>Description:</b> 
+If not set, the default model in pipeline will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td> <b>Meaning:</b>The directory path of the  text image unwarping model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td>
+<b>Meaning:</b>The name of the layout detection model. <br/>
+<b>Description:</b> 
+If not set, the default model in pipeline will be used. </td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td> <b>Meaning:</b>The directory path of the  layout detection model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_name</code></td>
+<td><b>Meaning:</b>The name of the seal text detection model.<br/>
+<b>Description:</b> 
+If not set, the pipeline's default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_dir</code></td>
+<td><b>Meaning:</b>The directory path of the seal text detection model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>  
+<td><code>text_recognition_model_name</code></td>  
+<td><b>Meaning:</b>Name of the text recognition model. <br/>
+<b>Description:</b> 
+If not set, the default pipeline model is used.</td>  
+<td><code>str</code></td>  
+<td></td>  
+</tr>  
+<tr>  
+<td><code>text_recognition_model_dir</code></td>  
+<td><b>Meaning:</b>Directory path of the text recognition model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>  
+<td><code>str</code></td>  
+<td></td>  
+</tr>
+<tr>  
+<td><code>text_recognition_batch_size</code></td>  
+<td><b>Meaning:</b>Batch size for the text recognition model. <br/>
+<b>Description:</b> 
+If not set, defaults to <code>1</code>.</td>  
+<td><code>int</code></td>  
+<td></td>  
+</tr>
+<tr>  
+<td><code>use_doc_orientation_classify</code></td>  
+<td><b>Meaning:</b>Whether to load and use document orientation classification module. <br/>
+<b>Description:</b> 
+If not set, defaults to pipeline initialization value (<code>True</code>).</td>  
+<td><code>bool</code></td>  
+<td></td>  
+</tr>  
+<tr>  
+<td><code>use_doc_unwarping</code></td>  
+<td><b>Meaning:</b>Whether to load and use text image correction module. <br/>
+<b>Description:</b> 
+If not set, defaults to pipeline initialization value (<code>True</code>).</td>  
+<td><code>bool</code></td>  
+<td></td>  
+</tr>   
+<tr>
+<td><code>use_layout_detection</code></td>
+<td>
+<b>Meaning:</b>Whether to load and use the layout detection module. <br/>
+<b>Description:</b> 
+If not set, the parameter will be set to the value initialized in the pipeline, which is <code>True</code> by default.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>Score threshold for the layout model. <br/>
+<b>Description:</b>
+Any value between <code>0-1</code>. If not set, the default value is used, which is <code>0.5</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Whether to use Non-Maximum Suppression (NMS) as post-processing for layout detection. <br/>
+<b>Description:</b> 
+If not set, the parameter will be set to the value initialized in the pipeline, which is set to <code>True</code> by default.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Unclip ratio for detected boxes in layout detection model. <br/>
+<b>Description:</b>
+Any float > <code>0</code>. If not set, the default is <code>1.0</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>The merging mode for the detection boxes output by the model in layout region detection. <br/>
+<b>Description:</b>
+<ul>
+<li><b>large</b>: When set to "large", only the largest outer bounding box will be retained for overlapping bounding boxes, and the inner overlapping boxes will be removed;</li>
+<li><b>small</b>: When set to "small", only the smallest inner bounding boxes will be retained for overlapping bounding boxes, and the outer overlapping boxes will be removed;</li>
+<li><b>union</b>: No filtering of bounding boxes will be performed, and both inner and outer boxes will be retained;</li>
+</ul>If not set, the default is <code>large</code>.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limit for seal text detection. <br/>
+<b>Description:</b>
+Any integer > <code>0</code>. If not set, the default is <code>736</code>.
+</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>Meaning:</b>Limit type for image side in seal text detection. <br/>
+<b>Description:</b>
+Supports <code>min</code> and <code>max</code>; <code>min</code> ensures shortest side ≥ <code>det_limit_side_len</code>, <code>max</code> ensures longest side ≤ <code>limit_side_len</code>. If not set, the default is <code>min</code>.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>Meaning:</b>Pixel threshold. Pixels with scores above this value in the probability map are considered text. <br/>
+<b>Description:</b>
+Any float > <code>0</code>. If not set, the default is <code>0.2</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>Meaning:</b>Box threshold. Boxes with average pixel scores above this value are considered text regions. <br/>
+<b>Description:</b>
+Any float > <code>0</code>. If not set, the default is <code>0.6</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion ratio for seal text detection. Higher value means larger expansion area. <br/>
+<b>Description:</b>
+Any float > <code>0</code>. If not set, the default is <code>0.5</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Recognition score threshold. Text results above this value will be kept. <br/>
+<b>Description:</b>
+Any float > <code>0</code>. If not set, the default is <code>0.0</code> (no threshold).
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>The device used for inference.<br/>
+<b>Description:</b>
+Support for specifying specific card numbers:
+<ul>
+<li><b>CPU</b>: For example, <code>cpu</code> indicates using the CPU for inference.</li>
+<li><b>GPU</b>: For example, <code>gpu:0</code> indicates using the first GPU for inference.</li>
+<li><b>NPU</b>: For example, <code>npu:0</code> indicates using the first NPU for inference.</li>
+<li><b>XPU</b>: For example, <code>xpu:0</code> indicates using the first XPU for inference.</li>
+<li><b>MLU</b>: For example, <code>mlu:0</code> indicates using the first MLU for inference.</li>
+<li><b>DCU</b>: For example, <code>dcu:0</code> indicates using the first DCU for inference.</li>
+<li><b>MetaX GPU</b>: For example, <code>metax_gpu:0</code> indicates using the first MetaX GPU for inference.</li>
+<li><b>Iluvatar GPU</b>: For example, <code>iluvatar_gpu:0</code> indicates using the first Iluvatar GPU for inference.</li>
+</ul>If not set, the pipeline initialized value for this parameter will be used. During initialization, the local GPU device 0 will be preferred; if unavailable, the CPU device will be used.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to enable the TensorRT subgraph engine of Paddle Inference.<br/>
+<b>Description:</b>
+If the model does not support TensorRT acceleration, acceleration will not be used even if this flag is set.<br/>
+For CUDA 11.8 versions of PaddlePaddle, the compatible TensorRT version is 8.x (x>=6). TensorRT 8.6.1.6 is recommended.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b> Computation precision, such as <code>fp32</code> or <code>fp16</code>.</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b> Whether to enable MKL-DNN accelerated inference.<br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support MKL-DNN acceleration, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b> MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads used for inference on CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b> Path to the PaddleX pipeline configuration file.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+
+</tbody>
+</table>
+</details>
+<br/>
+
+
+
+After running, the results will be printed to the terminal, as follows:
+
+```bash
+{'res': {'input_path': './seal_text_det.png', 'model_settings': {'use_doc_preprocessor': True, 'use_layout_detection': True}, 'doc_preprocessor_res': {'input_path': None, 'page_index': None, 'model_settings': {'use_doc_orientation_classify': False, 'use_doc_unwarping': False}, 'angle': -1}, 'layout_det_res': {'input_path': None, 'page_index': None, 'boxes': [{'cls_id': 16, 'label': 'seal', 'score': 0.975529670715332, 'coordinate': [6.191284, 0.16680908, 634.39325, 628.85345]}]}, 'seal_res_list': [{'input_path': None, 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_textline_orientation': False}, 'dt_polys': [array([[320,  38],
+       ...,
+       [315,  38]]), array([[461, 347],
+       ...,
+       [456, 346]]), array([[439, 445],
+       ...,
+       [434, 444]]), array([[158, 468],
+       ...,
+       [154, 466]])], 'text_det_params': {'limit_side_len': 736, 'limit_type': 'min', 'thresh': 0.2, 'max_side_limit': 4000, 'box_thresh': 0.6, 'unclip_ratio': 0.5}, 'text_type': 'seal', 'textline_orientation_angles': array([-1, ..., -1]), 'text_rec_score_thresh': 0, 'rec_texts': ['天津君和缘商贸有限公司', '发票专用章', '吗繁物', '5263647368706'], 'rec_scores': array([0.99340463, ..., 0.9916274 ]), 'rec_polys': [array([[320,  38],
+       ...,
+       [315,  38]]), array([[461, 347],
+       ...,
+       [456, 346]]), array([[439, 445],
+       ...,
+       [434, 444]]), array([[158, 468],
+       ...,
+       [154, 466]])], 'rec_boxes': array([], dtype=float64)}]}}
+```
+The visualized results are saved under `save_path`, and the visualized result of seal OCR is as follows:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/seal_recognition/04.png"/>
+
+### 2.2 Python Script Integration
+
+* The above command line is for quickly experiencing and viewing the effect. Generally, in a project, you often need to integrate through code. You can complete the quick inference of the pipeline with just a few lines of code. The inference code is as follows:
+
+```python
+from paddleocr import SealRecognition
+
+pipeline = SealRecognition(
+    use_doc_orientation_classify=False, # Set whether to use document orientation classification model
+    use_doc_unwarping=False, # Set whether to use document image unwarping module
+)
+# ocr = SealRecognition(device="gpu") # Specify GPU for model inference
+output = pipeline.predict("./seal_text_det.png")
+for res in output:
+    res.print() ## Print structured prediction results
+    res.save_to_img("./output/")
+    res.save_to_json("./output/")
+```
+
+The example above uses the local `paddle_static` inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following code:
+
+```python
+from paddleocr import SealRecognition
+
+pipeline = SealRecognition(
+    engine="transformers",
+)
+# ocr = SealRecognition(device="gpu") # Specify GPU for model inference
+output = pipeline.predict("./seal_text_det.png")
+for res in output:
+    res.print() ## Print structured prediction results
+    res.save_to_img("./output/")
+    res.save_to_json("./output/")
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+In the above Python script, the following steps were executed:
+
+(1) Instantiate a pipeline object for seal text recognition using the SealRecognition() class, with specific parameter descriptions as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>Name of the document orientation classification model.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the document orientation classification model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>Meaning:</b>Name of the document unwarping model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the document unwarping model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the layout detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the layout detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the seal text detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default model will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory of the seal text detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the text recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the pipeline default model is used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the text recognition model.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default batch size is <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to enable the document orientation classification module. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default value is <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to enable the document image unwarping module. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default value is <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>Meaning:</b>Whether to load and use the layout detection module. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the parameter will be set to the value initialized in the pipeline, which is <code>True</code> by default.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>Score threshold for the layout model. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float between <code>0-1</code>;</li>
+<li><b>dict</b>: <code>{0:0.1}</code> where the key is the class ID and the value is the threshold for that class;</li>
+<li><b>None</b>: If set to <code>None</code>, uses the pipeline default of <code>0.5</code>.</li>
+</ul>
+</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Whether to use Non-Maximum Suppression (NMS) as post-processing for layout detection. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the parameter will be set to the value initialized in the pipeline, which is set to <code>True</code> by default.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion ratio for the bounding boxes from the layout detection model. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>Tuple[float,float]</b>: Expansion ratios in horizontal and vertical directions;</li>
+<li><b>dict</b>: A dictionary with <b>int</b> keys representing <code>cls_id</code>, and <b>tuple</b> values, e.g., <code>{0: (1.1, 2.0)}</code> means width is expanded 1.1× and height 2.0× for class 0 boxes;</li>
+<li><b>None</b>: If set to <code>None</code>, uses the pipeline default of <code>1.0</code>.</li>
+</ul>
+</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>Filtering method for overlapping boxes in layout detection.<br/>
+<b>Description:</b> 
+<ul>
+<li><b>str</b>: Options include <code>large</code>, <code>small</code>, and <code>union</code> to retain the larger box, smaller box, or both;</li>
+<li><b>dict</b>: A dictionary with <b>int</b> keys representing <code>cls_id</code>, and <b>str</b> values, e.g., <code>{0: "large", 2: "small"}</code> means using different modes for different classes;</li>
+<li><b>None</b>: If set to <code>None</code>, uses the pipeline default value <code>large</code>.</li>
+</ul>
+</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limit for seal text detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>int</b>: Any integer greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the default value is <code>736</code>.</li>
+</ul>
+</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>Meaning:</b>Limit type for seal text detection image side length. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>str</b>: Supports <code>min</code> and <code>max</code>. <code>min</code> ensures the shortest side is no less than <code>det_limit_side_len</code>, while <code>max</code> ensures the longest side is no greater than <code>limit_side_len</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the default value is <code>min</code>.</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>Meaning:</b>Pixel threshold for detection. Pixels with scores greater than this value in the probability map are considered text pixels. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the default value is <code>0.2</code>.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>Meaning:</b>Bounding box threshold. If the average score of all pixels inside a detection box exceeds this threshold, it is considered a text region. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the default value is <code>0.6</code>.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Expansion ratio for seal text detection. The larger the value, the larger the expanded area. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the default value is <code>0.5</code>.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Score threshold for seal text recognition. Text results with scores above this threshold will be retained. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any float greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the default value is <code>0.0</code> (no threshold).</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>Device used for inference. <br/>
+<b>Description:</b> 
+Supports specifying device ID:
+<ul>
+<li><b>CPU</b>: e.g., <code>cpu</code> means using CPU for inference;</li>
+<li><b>GPU</b>: e.g., <code>gpu:0</code> means using GPU 0;</li>
+<li><b>NPU</b>: e.g., <code>npu:0</code> means using NPU 0;</li>
+<li><b>XPU</b>: e.g., <code>xpu:0</code> means using XPU 0;</li>
+<li><b>MLU</b>: e.g., <code>mlu:0</code> means using MLU 0;</li>
+<li><b>DCU</b>: e.g., <code>dcu:0</code> means using DCU 0;</li>
+<li><b>MetaX GPU</b>: e.g., <code>metax_gpu:0</code> means using MetaX GPU 0;</li>
+<li><b>Iluvatar GPU</b>: e.g., <code>iluvatar_gpu:0</code> means using Iluvatar GPU 0;</li>
+<li><b>None</b>: If set to <code>None</code>, the pipeline initialized value for this parameter will be used. During initialization, the local GPU device 0 will be preferred; if unavailable, the CPU device will be used.</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>Meaning:</b> Inference-engine configuration.<br/><b>Description:</b> Recommended together with <code>engine</code>. For supported fields, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b> Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b> Whether to enable the TensorRT subgraph engine of Paddle Inference.<br/>
+<b>Description:</b>
+If the model does not support TensorRT acceleration, acceleration will not be used even if this flag is set.<br/>
+For CUDA 11.8 versions of PaddlePaddle, the compatible TensorRT version is 8.x (x>=6). TensorRT 8.6.1.6 is recommended.<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b> Computation precision, such as <code>"fp32"</code> or <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b> Whether to enable MKL-DNN accelerated inference.<br/>
+<b>Description:</b>
+If MKL-DNN is unavailable or the model does not support MKL-DNN acceleration, acceleration will not be used even if this flag is set.
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b> MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b> Number of threads used for inference on CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b> Path to the PaddleX pipeline configuration file.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+</tbody>
+</table>
+
+(2) Call the <code>predict()</code> method of the Seal Text Recognition pipeline object for inference prediction. This method will return a <code>generator</code>. Below are the parameters and their descriptions for the <code>predict()</code> method:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Parameter Description</th>
+<th>Parameter Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Input data to be predicted. Required. <br/>
+<b>Description:</b> 
+Supports multiple types:
+<ul>
+<li><b>Python Var</b>: Image data represented by <code>numpy.ndarray</code>;</li>
+<li><b>str</b>: Local path of an image or PDF file, e.g., <code>/root/data/img.jpg</code>; <b>URL link</b>, e.g., the network URL of an image or PDF file: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/seal_text_det.png">Example</a>; <b>Local directory</b>, containing images to be predicted, e.g., <code>/root/data/</code> (currently does not support prediction of PDF files in directories; PDF files must be specified with an exact file path);</li>
+<li><b>list</b>: Elements of the list must be of the above types, e.g., <code>[numpy.ndarray, numpy.ndarray]</code>, <code>[\"/root/data/img1.jpg\", \"/root/data/img2.jpg\"]</code>, <code>[\"/root/data1\", \"/root/data2\"]</code>.</li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to use the document orientation classification module during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to use the text image correction module during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td>
+<b>Meaning:</b>Whether to use the layout detection module during inference. </td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters.<br/> 
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td>Same meaning as the instantiation parameters. 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+</table>
+</details>
+
+(3) Process the prediction results. The prediction result for each sample is of <code>dict</code> type and supports operations such as printing, saving as an image, and saving as a <code>json</code> file:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Description</th>
+<th>Parameter</th>
+<th>Parameter Type</th>
+<th>Parameter Description</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">Print results to the terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output content using <code>JSON</code> indentation.</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data for better readability, effective only when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non-<code>ASCII</code> characters to <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> will retain the original characters, effective only when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">Save results as a json file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The file path to save the results. When it is a directory, the saved file name will be consistent with the input file type.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data for better readability, effective only when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non-<code>ASCII</code> characters to <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> will retain the original characters, effective only when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save results as an image file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The file path to save the results, supports directory or file path.</td>
+<td>None</td>
+</tr>
+</table>
+<ul>
+<li>Calling the<code>print()</code> method will print the results to the terminal, and the explanations of the printed content are as follows:
+    <ol start="1" type="1">
+    <li><code>input_path</code>: <code>(str)</code> The input path of the image to be predicted.</li>
+    <li><code>model_settings</code>: <code>(Dict[str, bool])</code> The model parameters required for pipeline configuration.
+        <ol>
+        <li><code>use_doc_preprocessor</code>: <code>(bool)</code> Controls whether to enable the document preprocessing sub-pipeline.</li>
+        <li><code>use_layout_detection</code>: <code>(bool)</code> Controls whether to enable the layout detection sub-module.</li>
+        </ol>
+    </li>
+    <li><code>layout_det_res</code>: <code>(Dict[str, Union[List[numpy.ndarray], List[float]]])</code> The output result of the layout detection sub-module. Only exists when <code>use_layout_detection=True</code>
+        <ol>
+        <li><code>input_path</code>: <code>(Union[str, None])</code> The image path accepted by the layout detection module. Saved as <code>None</code> when the input is a <code>numpy.ndarray</code>.</li>
+        <li><code>page_index</code>: <code>(Union[int, None])</code> Indicates the current page number of the PDF if the input is a PDF file; otherwise, it is<code>None</code>.</li>
+        <li><code>boxes</code>: <code>(List[Dict])</code> A list of detected layout seal regions, with each element containing the following fields:</li>
+            <ul>
+            <li><code>cls_id</code>: <code>(int)</code> The class ID of the detected seal region.</li>
+            <li><code>score</code>: <code>(float)</code> The confidence score of the detected region.</li>
+            <li><code>coordinate</code>: <code>(List[float])</code> The coordinates of the four corners of the detection box, in the order of x1, y1, x2, y2, representing the x-coordinate of the top-left corner, the y-coordinate of the top-left corner, the x-coordinate of the bottom-right corner, and the y-coordinate of the bottom-right corner.</li>
+            </ul>
+        </ol>
+    </li>
+    <li><code>seal_res_list</code>: <code>List[Dict]</code> A list of seal text recognition results, with each element containing the following fields:
+        <ol>
+        <li><code>input_path</code>: <code>(Union[str, None])</code> The image path accepted by the seal text recognition pipeline. Saved as <code>None</code> when the input is a <code>numpy.ndarray</code>.</li>
+        <li><code>page_index</code>: <code>(Union[int, None])</code> Indicates the current page number of the PDF if the input is a PDF file; otherwise, it is <code>None</code>.</li>
+        <li><code>model_settings</code>: <code>(Dict[str, bool])</code> The model configuration parameters for the seal text recognition pipeline.</li>
+            <ul>
+            <li><code>use_doc_preprocessor</code>: <code>(bool)</code> Controls whether to enable the document preprocessing sub-pipeline.</li>
+            <li><code>use_textline_orientation</code>: <code>(bool)</code> Controls whether to enable the text line orientation classification sub-module.</li>
+            </ul>
+        </ol>
+    </li>
+    <li><code>doc_preprocessor_res</code>: <code>(Dict[str, Union[str, Dict[str, bool], int]])</code> The output result of the document preprocessing sub-pipeline. Only exists when <code>use_doc_preprocessor=True</code>.
+        <ol>
+        <li><code>input_path</code>: <code>(Union[str, None])</code> The image path accepted by the document preprocessing sub-pipeline. Saved as <code>None</code> when the input is a <code>numpy.ndarray</code>.</li>
+        <li><code>model_settings</code>: <code>(Dict[str, bool])</code> The model configuration parameters for the preprocessing sub-pipeline.</li>
+            <ul>
+            <li><code>use_doc_orientation_classify</code>: <code>(bool)</code> Controls whether to enable the document orientation classification sub-module.</li>
+            <li><code>use_doc_unwarping</code>: <code>(bool)</code> Controls whether to enable the document unwarping sub-module.</li>
+            </ul>
+        <li><code>angle</code>: <code>(int)</code> The predicted result of document orientation classification. When enabled, it takes values [0, 1, 2, 3], corresponding to [0°, 90°, 180°, 270°]; when disabled, it is -1.</li>
+        </ol>
+    </li>
+    <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> A list of polygon boxes for text detection. Each box is represented by a numpy array of shape (n, 2), where n is the number of vertices.</li>
+    <li><code>dt_scores</code>: <code>(List[float])</code> A list of confidence scores for text detection boxes.</li>
+    <li><code>dt_scores</code>: <code>(List[float])</code> A list of confidence scores for text detection boxes.</li>
+    <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> Configuration parameters for the text detection module.
+        <ol>
+        <li><code>limit_side_len</code>: <code>(int)</code> The side length limit value during image preprocessing.</li>
+        <li><code>limit_type</code>: <code>(str)</code> The type of side length limit, which can be "min" or "max".</li>
+        <li><code>thresh</code>: <code>(float)</code> The confidence threshold for text pixel classification.</li>
+        <li><code>box_thresh</code>: <code>(float)</code> The confidence threshold for text detection boxes.</li>
+        <li><code>unclip_ratio</code>: <code>(float)</code> The expansion ratio for text detection boxes.</li>
+        <li><code>text_type</code>: <code>(str)</code> The type of text to be detected, which is fixed as "seal" for seal text detection.</li>
+        </ol>
+    </li>
+    <li><code>text_rec_score_thresh</code>: <code>(float)</code> The filtering threshold for text recognition results.</li>
+    <li><code>rec_texts</code>: <code>(List[str])</code> A list of text recognition results, containing only texts with confidence scores above <code>text_rec_score_thresh</code>.</li>
+     <li><code>rec_scores</code>: <code>(List[float])</code> A list of confidence scores for text recognition, filtered by <code>text_rec_score_thresh</code>.</li>
+     <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> A list of text detection boxes filtered by confidence score, in the same format as <code>dt_polys</code>.</li>
+     <li><code>rec_boxes</code>: <code>(numpy.ndarray)</code> 检An array of rectangular bounding boxes for detection boxes; the seal recognition pipeline returns an empty array.</li>
+    </ol>
+</li>
+<li>Calling the <code>save_to_json()</code> method will save the above content to the specified <code>save_path</code>. If a directory is specified, the saved path will be <code>save_path/{your_img_basename}_res.json</code>. If a file is specified, it will be saved directly to that file. Since JSON files do not support saving numpy arrays, <code>numpy.array</code> types will be converted to list format.</li>
+<li>Calling the <code>save_to_img()</code> method will save the visualization results to the specified <code>save_path</code>. If a directory is specified, the saved path will be <code>save_path/{your_img_basename}_seal_res_region1.{your_img_extension}</code>. If a file is specified, it will be saved directly to that file. (The pipeline usually contains multiple result images, so it is not recommended to specify a specific file path directly, as multiple images will be overwritten, and only the last image will be retained.)</li>
+
+* Additionally, you can obtain visualized images with results and prediction results through attributes, as follows:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">Get the prediction results in <code>json</code> format.</td>
+</tr>
+<tr>
+<td rowspan="2"><code>img</code></td>
+<td rowspan="2">Get the visualization results in <code>dict</code> format.</td>
+</tr>
+</table>
+
+<li>The prediction results obtained through the <code>json</code> attribute are of dict type, with content consistent with what is saved by calling the <code>save_to_json()</code> method.</li>
+<li>The prediction results returned by the <code>img</code> attribute are of dict type. The keys are <code>layout_det_res</code>, <code>seal_res_region1</code>, and <code>preprocessed_img</code>, corresponding to three <code>Image.Image</code> objects: one for visualizing layout detection, one for visualizing seal text recognition results, and one for visualizing image preprocessing. If the image preprocessing sub-module is not used, <code>preprocessed_img</code> will not be included in the dictionary. If the layout region detection module is not used, <code>layout_det_res</code> will not be included.</li>
+
+## 3. Development Integration/Deployment
+If the pipeline meets your requirements for inference speed and accuracy, you can proceed directly with development integration/deployment.
+
+If you need to integrate the pipeline into your Python project, you can refer to the example code in [2.2 Python Script Method](#22-python-script-integration).
+
+In addition, PaddleOCR also provides three other deployment methods, which are detailed as follows:
+
+🚀 High-Performance Inference: In real-world production environments, many applications have stringent performance requirements for deployment strategies, especially in terms of response speed, to ensure efficient system operation and a smooth user experience. To address this, PaddleOCR offers high-performance inference capabilities aimed at deeply optimizing the performance of model inference and pre/post-processing, thereby significantly accelerating the end-to-end process. For detailed high-performance inference procedures, please refer to [High-Performance Inference](../inference_deployment/local_inference/high_performance_inference.md).
+
+☁️ Service Deployment: Service deployment is a common form of deployment in real-world production environments. By encapsulating inference functionality into a service, clients can access these services via network requests to obtain inference results. For detailed production service deployment procedures, please refer to [Serving](../inference_deployment/serving/serving.md).
+
+Below are the API references for basic serving deployment and multi-language service invocation examples:
+
+<details><summary>API Reference</summary>
+<p>For the main operations provided by the service:</p>
+<ul>
+<li>The HTTP request method is POST.</li>
+<li>The request body and response body are both JSON data (JSON objects).</li>
+<li>When the request is processed successfully, the response status code is <code>200</code>, and the attributes of the response body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>The UUID of the request.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Fixed as <code>0</code>.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error message. Fixed as <code>"Success"</code>.</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>The result of the operation.</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is not processed successfully, the attributes of the response body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>The UUID of the request.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Same as the response status code.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error message.</td>
+</tr>
+</tbody>
+</table>
+<p>The main operations provided by the service are as follows:</p>
+<ul>
+<li><b><code>infer</code></b></li>
+</ul>
+<p>Obtain the seal text recognition result.</p>
+<p><code>POST /seal-recognition</code></p>
+<ul>
+<li>The attributes of the request body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Description</th>
+<th>Required</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>The URL of image files (including TIFF; multi-page TIFF is processed page by page) or PDF file accessible by the server, or the Base64-encoded content of the file. By default, for PDF or multi-page TIFF files exceeding 10 pages, only the first 10 pages will be processed.<br />
+To remove the page limit, please add the following configuration to the pipeline configuration file:
+<pre><code>Serving:
+  extra:
+    max_num_input_imgs: null
+</code></pre></td>
+<td>Yes</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>The type of file. <code>0</code> indicates a PDF file, <code>1</code> indicates an image file. If this attribute is not present in the request body, the file type will be inferred from the URL.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_doc_orientation_classify</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_doc_unwarping</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useLayoutDetection</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>use_layout_detection</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutThreshold</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>layout_threshold</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutNms</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>layout_nms</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutUnclipRatio</code></td>
+<td><code>number</code> | <code>array</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>layout_unclip_ratio</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>layoutMergeBboxesMode</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>layout_merge_bboxes_mode</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_det_limit_side_len</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_det_limit_type</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_det_thresh</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_det_box_thresh</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_det_unclip_ratio</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>sealRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the description of the <code>seal_rec_score_thresh</code> parameter of the pipeline object's <code>predict</code> method.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>
+Whether to return the final visualization image and intermediate images during the processing.<br/>
+<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>If <code>true</code> is provided: return images.</li>
+<li>If <code>false</code> is provided: do not return any images.</li>
+<li>If this parameter is omitted from the request body, or if <code>null</code> is explicitly passed, the behavior will follow the value of <code>Serving.visualize</code> in the pipeline configuration.</li>
+</ul>
+<br/>
+For example, adding the following setting to the pipeline config file:<br/>
+<pre><code>Serving:
+  visualize: False
+</code></pre>
+will disable image return by default. This behavior can be overridden by explicitly setting the <code>visualize</code> parameter in the request.<br/>
+If neither the request body nor the configuration file is set (If <code>visualize</code> is set to <code>null</code> in the request and  not defined in the configuration file), the image is returned by default.
+</td>
+<td>No</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is processed successfully, the <code>result</code> in the response body has the following properties:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>sealRecResults</code></td>
+<td><code>object</code></td>
+<td>The seal text recognition result. The array length is 1 (for image input) or the actual number of document pages processed (for PDF input). For PDF input, each element in the array represents the result of each page actually processed in the PDF file.</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>Information about the input data.</td>
+</tr>
+</tbody>
+</table>
+<p>Image fields in the element schema below (e.g. <code>outputImages</code>, <code>inputImage</code>) are returned inline as Base64 strings by default; when the server is configured to return URLs, those values become pre-signed URLs while the field types remain unchanged. See the "Returning Binary Content as URLs" section of the <a href="../inference_deployment/serving/serving.en.md">Serving Deployment Guide</a> for configuration.</p>
+<p>Each element in <code>sealRecResults</code> is an <code>object</code> with the following properties:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>A simplified version of the <code>res</code> field in the JSON representation generated by the <code>predict</code> method of the production object, where the <code>input_path</code> and the <code>page_index</code> fields are removed.</td>
+</tr>
+<tr>
+<td><code>outputImages</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>See the description of the <code>img</code> attribute of the result of the pipeline prediction. The images are in JPEG format and encoded in Base64 by default; returned as pre-signed URLs when URL-return mode is enabled.</td>
+</tr>
+<tr>
+<td><code>inputImage</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>The input image. The image is in JPEG format and encoded in Base64 by default; returned as a pre-signed URL when URL-return mode is enabled.</td>
+</tr>
+</tbody>
+</table></details>
+<details><summary>Multi-language Service Invocation Example</summary>
+<details>
+<summary>Python</summary>
+
+<pre><code class="language-python">import base64
+import requests
+
+API_URL = "http://localhost:8080/seal-recognition"
+file_path = "./demo.jpg"
+
+with open(file_path, "rb") as file:
+    file_bytes = file.read()
+    file_data = base64.b64encode(file_bytes).decode("ascii")
+
+payload = {"file": file_data, "fileType": 1}
+
+response = requests.post(API_URL, json=payload)
+
+assert response.status_code == 200
+result = response.json()["result"]
+for i, res in enumerate(result["sealRecResults"]):
+    print(res["prunedResult"])
+    for img_name, img in res["outputImages"].items():
+        img_path = f"{img_name}_{i}.jpg"
+        with open(img_path, "wb") as f:
+            f.write(base64.b64decode(img))
+        print(f"Output image saved at {img_path}")
+</code></pre></details>
+
+<details><summary>C++</summary>
+
+<pre><code class="language-cpp">#include &lt;iostream&gt;
+#include &lt;fstream&gt;
+#include &lt;vector&gt;
+#include &lt;string&gt;
+#include "cpp-httplib/httplib.h" // https://github.com/Huiyicc/cpp-httplib
+#include "nlohmann/json.hpp" // https://github.com/nlohmann/json
+#include "base64.hpp" // https://github.com/tobiaslocker/base64
+
+int main() {
+    httplib::Client client("localhost", 8080);
+
+    const std::string filePath = "./demo.jpg";
+
+    std::ifstream file(filePath, std::ios::binary | std::ios::ate);
+    if (!file) {
+        std::cerr << "Error opening file: " << filePath << std::endl;
+        return 1;
+    }
+
+    std::streamsize size = file.tellg();
+    file.seekg(0, std::ios::beg);
+    std::vector<char> buffer(size);
+    if (!file.read(buffer.data(), size)) {
+        std::cerr << "Error reading file." << std::endl;
+        return 1;
+    }
+
+    std::string bufferStr(buffer.data(), static_cast<size_t>(size));
+    std::string encodedFile = base64::to_base64(bufferStr);
+
+    nlohmann::json jsonObj;
+    jsonObj["file"] = encodedFile;
+    jsonObj["fileType"] = 1;
+
+    auto response = client.Post("/seal-recognition", jsonObj.dump(), "application/json");
+
+    if (response && response->status == 200) {
+        nlohmann::json jsonResponse = nlohmann::json::parse(response->body);
+        auto result = jsonResponse["result"];
+
+        if (!result.is_object() || !result["sealRecResults"].is_array()) {
+            std::cerr << "Unexpected response format." << std::endl;
+            return 1;
+        }
+
+        for (size_t i = 0; i < result["sealRecResults"].size(); ++i) {
+            auto res = result["sealRecResults"][i];
+
+            if (res.contains("prunedResult")) {
+                std::cout << "Recognized seal result: " << res["prunedResult"].dump() << std::endl;
+            }
+
+            if (res.contains("outputImages") && res["outputImages"].is_object()) {
+                for (auto& [imgName, imgData] : res["outputImages"].items()) {
+                    std::string outputPath = imgName + "_" + std::to_string(i) + ".jpg";
+                    std::string decodedImage = base64::from_base64(imgData.get<std::string>());
+
+                    std::ofstream outFile(outputPath, std::ios::binary);
+                    if (outFile.is_open()) {
+                        outFile.write(decodedImage.c_str(), decodedImage.size());
+                        outFile.close();
+                        std::cout << "Saved image: " << outputPath << std::endl;
+                    } else {
+                        std::cerr << "Failed to write image: " << outputPath << std::endl;
+                    }
+                }
+            }
+        }
+    } else {
+        std::cerr << "Request failed." << std::endl;
+        if (response) {
+            std::cerr << "HTTP status: " << response->status << std::endl;
+            std::cerr << "Response body: " << response->body << std::endl;
+        }
+        return 1;
+    }
+
+    return 0;
+}
+</code></pre></details>
+
+<details><summary>Java</summary>
+
+<pre><code class="language-java">import okhttp3.*;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.util.Base64;
+
+public class Main {
+    public static void main(String[] args) throws IOException {
+        String API_URL = "http://localhost:8080/seal-recognition";
+        String imagePath = "./demo.jpg";
+
+        File file = new File(imagePath);
+        byte[] fileContent = java.nio.file.Files.readAllBytes(file.toPath());
+        String base64Image = Base64.getEncoder().encodeToString(fileContent);
+
+        ObjectMapper objectMapper = new ObjectMapper();
+        ObjectNode payload = objectMapper.createObjectNode();
+        payload.put("file", base64Image);
+        payload.put("fileType", 1);
+
+        OkHttpClient client = new OkHttpClient();
+        MediaType JSON = MediaType.get("application/json; charset=utf-8");
+        RequestBody body = RequestBody.create(JSON, payload.toString());
+
+        Request request = new Request.Builder()
+                .url(API_URL)
+                .post(body)
+                .build();
+
+        try (Response response = client.newCall(request).execute()) {
+            if (response.isSuccessful()) {
+                String responseBody = response.body().string();
+                JsonNode root = objectMapper.readTree(responseBody);
+                JsonNode result = root.get("result");
+
+                JsonNode sealRecResults = result.get("sealRecResults");
+                for (int i = 0; i < sealRecResults.size(); i++) {
+                    JsonNode item = sealRecResults.get(i);
+                    int finalI = i;
+
+                    JsonNode prunedResult = item.get("prunedResult");
+                    System.out.println("Pruned Result [" + i + "]: " + prunedResult.toString());
+
+                    JsonNode outputImages = item.get("outputImages");
+                    if (outputImages != null && outputImages.isObject()) {
+                        outputImages.fieldNames().forEachRemaining(imgName -> {
+                            try {
+                                String imgBase64 = outputImages.get(imgName).asText();
+                                byte[] imgBytes = Base64.getDecoder().decode(imgBase64);
+                                String imgPath = imgName + "_" + finalI + ".jpg";
+                                try (FileOutputStream fos = new FileOutputStream(imgPath)) {
+                                    fos.write(imgBytes);
+                                    System.out.println("Saved image: " + imgPath);
+                                }
+                            } catch (IOException e) {
+                                System.err.println("Failed to save image: " + e.getMessage());
+                            }
+                        });
+                    }
+                }
+            } else {
+                System.err.println("Request failed with HTTP code: " + response.code());
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Go</summary>
+
+<pre><code class="language-go">package main
+
+import (
+	"bytes"
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"io/ioutil"
+	"net/http"
+)
+
+func main() {
+	API_URL := "http://localhost:8080/seal-recognition"
+	filePath := "./demo.jpg"
+
+	fileBytes, err := ioutil.ReadFile(filePath)
+	if err != nil {
+		fmt.Printf("Error reading file: %v\n", err)
+		return
+	}
+	fileData := base64.StdEncoding.EncodeToString(fileBytes)
+
+	payload := map[string]interface{}{
+		"file":     fileData,
+		"fileType": 1,
+	}
+	payloadBytes, err := json.Marshal(payload)
+	if err != nil {
+		fmt.Printf("Error marshaling payload: %v\n", err)
+		return
+	}
+
+	client := &http.Client{}
+	req, err := http.NewRequest("POST", API_URL, bytes.NewBuffer(payloadBytes))
+	if err != nil {
+		fmt.Printf("Error creating request: %v\n", err)
+		return
+	}
+	req.Header.Set("Content-Type", "application/json")
+
+	resp, err := client.Do(req)
+	if err != nil {
+		fmt.Printf("Error sending request: %v\n", err)
+		return
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		fmt.Printf("Unexpected status code: %d\n", resp.StatusCode)
+		return
+	}
+
+	body, err := ioutil.ReadAll(resp.Body)
+	if err != nil {
+		fmt.Printf("Error reading response body: %v\n", err)
+		return
+	}
+
+	type SealResult struct {
+		PrunedResult map[string]interface{}   `json:"prunedResult"`
+		OutputImages map[string]string        `json:"outputImages"`
+		InputImage   *string                  `json:"inputImage"`
+	}
+
+	type Response struct {
+		Result struct {
+			SealRecResults []SealResult  `json:"sealRecResults"`
+			DataInfo       interface{}   `json:"dataInfo"`
+		} `json:"result"`
+	}
+
+	var respData Response
+	if err := json.Unmarshal(body, &respData); err != nil {
+		fmt.Printf("Error unmarshaling response: %v\n", err)
+		return
+	}
+
+	for i, res := range respData.Result.SealRecResults {
+		fmt.Printf("Pruned Result %d: %+v\n", i, res.PrunedResult)
+
+		for name, imgBase64 := range res.OutputImages {
+			imgBytes, err := base64.StdEncoding.DecodeString(imgBase64)
+			if err != nil {
+				fmt.Printf("Error decoding image %s: %v\n", name, err)
+				continue
+			}
+
+			filename := fmt.Sprintf("%s_%d.jpg", name, i)
+			if err := ioutil.WriteFile(filename, imgBytes, 0644); err != nil {
+				fmt.Printf("Error saving image %s: %v\n", filename, err)
+				continue
+			}
+			fmt.Printf("Output image saved at %s\n", filename)
+		}
+	}
+}
+</code></pre></details>
+
+<details><summary>C#</summary>
+
+<pre><code class="language-csharp">using System;
+using System.IO;
+using System.Net.Http;
+using System.Text;
+using System.Threading.Tasks;
+using Newtonsoft.Json.Linq;
+
+class Program
+{
+    static readonly string API_URL = "http://localhost:8080/seal-recognition";
+    static readonly string inputFilePath = "./demo.jpg";
+
+    static async Task Main(string[] args)
+    {
+        var httpClient = new HttpClient();
+
+        byte[] fileBytes = File.ReadAllBytes(inputFilePath);
+        string fileData = Convert.ToBase64String(fileBytes);
+
+        var payload = new JObject
+        {
+            { "file", fileData },
+            { "fileType", 1 }
+        };
+        var content = new StringContent(payload.ToString(), Encoding.UTF8, "application/json");
+
+        HttpResponseMessage response = await httpClient.PostAsync(API_URL, content);
+        response.EnsureSuccessStatusCode();
+
+        string responseBody = await response.Content.ReadAsStringAsync();
+        JObject jsonResponse = JObject.Parse(responseBody);
+
+        JArray sealRecResults = (JArray)jsonResponse["result"]["sealRecResults"];
+        for (int i = 0; i < sealRecResults.Count; i++)
+        {
+            var res = sealRecResults[i];
+            Console.WriteLine($"[{i}] prunedResult:\n{res["prunedResult"]}");
+
+            JObject outputImages = res["outputImages"] as JObject;
+            if (outputImages != null)
+            {
+                foreach (var img in outputImages)
+                {
+                    string imgName = img.Key;
+                    string base64Img = img.Value?.ToString();
+                    if (!string.IsNullOrEmpty(base64Img))
+                    {
+                        string imgPath = $"{imgName}_{i}.jpg";
+                        byte[] imageBytes = Convert.FromBase64String(base64Img);
+                        File.WriteAllBytes(imgPath, imageBytes);
+                        Console.WriteLine($"Output image saved at {imgPath}");
+                    }
+                }
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Node.js</summary>
+
+<pre><code class="language-js">const axios = require('axios');
+const fs = require('fs');
+const path = require('path');
+
+const API_URL = 'http://localhost:8080/seal-recognition';
+const imagePath = './demo.jpg';
+
+function encodeImageToBase64(filePath) {
+  const bitmap = fs.readFileSync(filePath);
+  return Buffer.from(bitmap).toString('base64');
+}
+
+const payload = {
+  file: encodeImageToBase64(imagePath),
+  fileType: 1
+};
+
+axios.post(API_URL, payload)
+  .then((response) => {
+    const result = response.data["result"];
+    const sealRecResults = result["sealRecResults"];
+
+    sealRecResults.forEach((res, i) => {
+      console.log(`\n[${i}] prunedResult:\n`, res["prunedResult"]);
+
+      const outputImages = res["outputImages"];
+      if (outputImages) {
+        for (const [imgName, base64Img] of Object.entries(outputImages)) {
+          const imgBuffer = Buffer.from(base64Img, 'base64');
+          const fileName = `${imgName}_${i}.jpg`;
+          fs.writeFileSync(fileName, imgBuffer);
+          console.log(`Output image saved at ${fileName}`);
+        }
+      } else {
+        console.log(`[${i}] No outputImages found.`);
+      }
+    });
+  })
+  .catch((error) => {
+    console.error('Error occurred while calling the API:', error.message);
+  });
+</code></pre></details>
+
+<details><summary>PHP</summary>
+
+<pre><code class="language-php">&lt;?php
+
+$API_URL = "http://localhost:8080/seal-recognition";
+$image_path = "./demo.jpg";
+
+$image_data = base64_encode(file_get_contents($image_path));
+$payload = array("file" => $image_data, "fileType" => 1);
+
+$ch = curl_init($API_URL);
+curl_setopt($ch, CURLOPT_POST, true);
+curl_setopt($ch, CURLOPT_POSTFIELDS, json_encode($payload));
+curl_setopt($ch, CURLOPT_HTTPHEADER, array('Content-Type: application/json'));
+curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+$response = curl_exec($ch);
+curl_close($ch);
+
+$result = json_decode($response, true)["result"]["sealRecResults"];
+
+foreach ($result as $i => $item) {
+    echo "[$i] prunedResult:\n";
+    print_r($item["prunedResult"]);
+
+    if (!empty($item["outputImages"])) {
+        foreach ($item["outputImages"] as $img_name => $base64_img) {
+            if (!empty($base64_img)) {
+                $output_path = "{$img_name}_{$i}.jpg";
+                file_put_contents($output_path, base64_decode($base64_img));
+                echo "Output image saved at $output_path\n";
+            }
+        }
+    } else {
+        echo "No outputImages found for item $i\n";
+    }
+}
+?&gt;
+</code></pre></details>
+</details>
+<br/>
+
+
+## 4. Custom Development
+If the default model weights provided by the seal text recognition pipeline do not meet your requirements in terms of accuracy or speed, you can try to <b>fine-tune</b> the existing models using <b>your own domain-specific or application data</b> to improve the recognition performance of the seal text recognition pipeline in your scenario.
+
+### 4.1 Model Fine-Tuning
+Since the seal text recognition pipeline consists of several modules, if the pipeline's performance does not meet expectations, the issue may arise from any one of these modules. You can analyze images with poor recognition results to identify which module is problematic and refer to the corresponding fine-tuning tutorial links in the table below for model fine-tuning.
+
+<table>
+<thead>
+<tr>
+<th>Scenario</th>
+<th>Fine-Tuning Module</th>
+<th>Fine-Tuning Reference Link</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>Inaccurate or missing seal position detection</td>
+<td>Layout Detection Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/layout_detection.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Missing text detection</td>
+<td>Text Detection Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/module_usage/text_detection.html#4-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate text content</td>
+<td>Text Recognition Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/module_usage/text_recognition.html#41-dataset-and-pre-trained-model-preparation">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate full-image rotation correction</td>
+<td>Document Image Orientation Classification Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Inaccurate image distortion correction</td>
+<td>Text Image Correction Module</td>
+<td>Not supported for fine-tuning</td>
+</tr>
+</tbody>
+</table>
+
+
+### 4.2 Model Application
+
+After you complete the fine-tuning training with a private dataset, you can obtain the local model weight files. You can then use the fine-tuned model weights by specifying the local model save path through parameters or by using a custom pipeline configuration file.
+
+#### 4.2.1 Specify Local Model Path via Parameters
+
+When initializing the pipeline object, specify the local model path through parameters. Taking the usage of fine-tuned weights for a text detection model as an example, the demonstration is as follows:
+
+Command line method:
+
+```bash
+# Specify the local model path through --doc_orientation_classify_model_dir
+paddleocr seal_recognition -i ./seal_text_det.png --doc_orientation_classify_model_dir your_orientation_classify_model_path
+
+# By default, the PP-LCNet_x1_0_doc_ori model is used as the default text detection model. If the fine-tuned model is not this one, modify the model name with --text_detection_model_name
+paddleocr seal_recognition -i ./seal_text_det.png --doc_orientation_classify_model_name PP-LCNet_x1_0_doc_ori --doc_orientation_classify_model_dir your_orientation_classify_model_path
+```
+
+Script method:
+
+```python
+
+from paddleocr import SealRecognition
+
+# Specify the local model path through doc_orientation_classify_model_dir
+pipeline = SealRecognition(doc_orientation_classify_model_dir ="./your_orientation_classify_model_path")
+
+# By default, the PP-LCNet_x1_0_doc_ori model is used as the default text detection model. If the fine-tuned model is not this one, modify the model name with doc_orientation_classify_model_name
+# pipeline = SealRecognition(doc_orientation_classify_model_name="PP-LCNet_x1_0_doc_ori", doc_orientation_classify_model_dir="./your_orientation_classify_model_path")
+
+```
+
+#### 4.2.2 Specify Local Model Path via Configuration File
+
+1. Obtain pipeline Configuration File
+
+You can call the `export_paddlex_config_to_yaml` method of the general OCR pipeline object in PaddleOCR to export the current pipeline configuration to a YAML file:
+
+```Python
+from paddleocr import SealRecognition
+
+pipeline = SealRecognition()
+pipeline.export_paddlex_config_to_yaml("SealRecognition.yaml")
+```
+
+2. Modify Configuration File
+
+After obtaining the default pipeline configuration file, replace the local path of the fine-tuned model weights in the corresponding position of the pipeline configuration file. For example:
+
+```yaml
+......
+SubPipelines:
+  DocPreprocessor:
+    SubModules:
+      DocOrientationClassify:
+        model_dir: null  # Replace with the path of the fine-tuned document orientation classification model weights
+        model_name: PP-LCNet_x1_0_doc_ori # If the name of the fine-tuned model is different from the default model name, please also modify here
+        module_name: doc_text_orientation
+      DocUnwarping:
+        model_dir: null  # Replace with the path of the fine-tuned document unwarping model weights
+        model_name: UVDoc # If the name of the fine-tuned model is different from the default model name, please also modify here
+        module_name: image_unwarping
+    pipeline_name: doc_preprocessor
+    use_doc_orientation_classify: true
+    use_doc_unwarping: true
+......
+```
+
+The pipeline configuration file not only contains the parameters supported by the SealRecognition CLI and Python API but also allows for more advanced configurations. Detailed information can be found in the [PaddleX Model pipeline Usage Overview](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/pipeline_develop_guide.html), where you can find the corresponding pipeline usage tutorial and adjust various configurations as needed.
+
+3. Load pipeline Configuration File in CLI
+
+After modifying the configuration file, specify the path of the modified pipeline configuration file using the `--paddlex_config` parameter in the command line. PaddleOCR will read its contents as the pipeline configuration. Example:
+
+```bash
+paddleocr seal_recognition --paddlex_config SealRecognition.yaml ...
+```
+
+4. Load pipeline Configuration File in Python API
+
+When initializing the pipeline object, you can pass the PaddleX pipeline configuration file path or configuration dictionary through the `paddlex_config` parameter. PaddleOCR will read its contents as the pipeline configuration. Example:
+
+```python
+from paddleocr import SealRecognition
+
+pipeline = SealRecognition(paddlex_config="SealRecognition.yaml")
+```
diff --git a/docs/version3.x/pipeline_usage/seal_recognition.md b/docs/version3.x/pipeline_usage/seal_recognition.md
new file mode 100644
index 0000000..d489577
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/seal_recognition.md
@@ -0,0 +1,2563 @@
+---
+comments: true
+---
+
+# 印章文本识别产线使用教程
+
+## 1. 印章文本识别产线介绍
+印章文本识别是一种自动从文档或图像中提取和识别印章内容的技术，印章文本的识别是文档处理的一部分，在很多场景都有用途，例如合同比对，出入库审核以及发票报销审核等场景。
+
+印章文本识别产线用于识别印章的文本内容，提取印章图像中的文字信息以文本形式输出，本产线集成了业界知名的 PP-OCRv4 的端到端 OCR 串联系统，支持弯曲印章文本的检测和识别。同时，本产线集成了可选的版面区域定位模块，可以在整个文档中准确定位印章所在的版面位置。此外也增加可选的文档图像的方向矫正和扭曲矫正功能。基于本产线，可实现 CPU 上毫秒级的文本内容精准预测。本产线同时提供了灵活的服务化部署方式，支持在多种硬件上使用多种编程语言调用。不仅如此，本产线也提供了二次开发的能力，您可以基于本产线在您自己的数据集上训练调优，训练后的模型也可以无缝集成。
+
+<img src="https://paddle-model-ecology.bj.bcebos.com/paddlex/PaddleX3.0/doc_images/practical_tutorial/PP-ChatOCRv3_doc_seal/01.png"/>
+
+<b>印章文本识别产线中包含以下5个模块。每个模块均可独立进行训练和推理，并包含多个模型。有关详细信息，请点击相应模块以查看文档。</b>
+
+- [印章文本检测模块](../module_usage/seal_text_detection.md)
+- [文本识别模块](../module_usage/text_recognition.md)
+- [版面区域检测模块](../module_usage/layout_detection.md)（可选）
+- [文档图像方向分类模块](../module_usage/doc_img_orientation_classification.md) （可选）
+- [文本图像矫正模块](../module_usage/text_image_unwarping.md) （可选）
+
+在本产线中，您可以根据下方的基准测试数据选择使用的模型。
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。
+> 在带有 [常规模式 / 高性能模式] 标记的推理耗时列中，`常规模式` 对应本地推理引擎 `paddle_static`。
+
+<details>
+<summary> <b>版面区域检测模块（可选）：</b></summary>
+
+* <b>版面检测模型，包含20个常见的类别：文档标题、段落标题、文本、页码、摘要、目录、参考文献、脚注、页眉、页脚、算法、公式、公式编号、图像、表格、图和表标题（图标题、表格标题和图表标题）、印章、图表、侧栏文本和参考文献内容</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout_plus-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout_plus-L_pretrained.pdparams">训练模型</a></td>
+<td>83.2</td>
+<td>53.03 / 17.23</td>
+<td>634.62 / 378.32</td>
+<td>126.01</td>
+<td>基于RT-DETR-L在包含中英文论文、多栏杂志、报纸、PPT、合同、书本、试卷、研报、古籍、日文文档、竖版文字文档等场景的自建数据集训练的更高精度版面区域定位模型</td>
+</tr>
+<tr>
+</tbody>
+</table>
+
+* <b>版面检测模型，包含23个常见的类别：文档标题、段落标题、文本、页码、摘要、目录、参考文献、脚注、页眉、页脚、算法、公式、公式编号、图像、图表标题、表格、表格标题、印章、图表标题、图表、页眉图像、页脚图像、侧栏文本</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-L_pretrained.pdparams">训练模型</a></td>
+<td>90.4</td>
+<td>33.59 / 33.59</td>
+<td>503.01 / 251.08</td>
+<td>123.76</td>
+<td>基于RT-DETR-L在包含中英文论文、杂志、合同、书本、试卷和研报等场景的自建数据集训练的高精度版面区域定位模型</td>
+</tr>
+<tr>
+<td>PP-DocLayout-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-M_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-M_pretrained.pdparams">训练模型</a></td>
+<td>75.2</td>
+<td>13.03 / 4.72</td>
+<td>43.39 / 24.44</td>
+<td>22.578</td>
+<td>基于PicoDet-L在包含中英文论文、杂志、合同、书本、试卷和研报等场景的自建数据集训练的精度效率平衡的版面区域定位模型</td>
+</tr>
+<tr>
+<td>PP-DocLayout-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-S_pretrained.pdparams">训练模型</a></td>
+<td>70.9</td>
+<td>11.54 / 3.86</td>
+<td>18.53 / 6.29</td>
+<td>4.834</td>
+<td>基于PicoDet-S在中英文论文、杂志、合同、书本、试卷和研报等场景上自建数据集训练的高效率版面区域定位模型</td>
+</tr>
+</tbody>
+</table>
+
+
+>❗ 以上列出的是版面检测模块重点支持的<b>4个核心模型</b>，该模块总共支持<b>13个全量模型</b>，包含多个预定义了不同类别的模型，其中包含印章类别的模型有9个，除上述3个核心模型外，其余模型列表如下：
+
+
+<details><summary> 👉模型列表详情</summary>
+
+* <b>3类版面检测模型，包含表格、图像、印章</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_3cls_pretrained.pdparams">训练模型</a></td>
+<td>88.2</td>
+<td>8.43 / 3.44</td>
+<td>17.60 / 6.51</td>
+<td>4.8</td>
+<td>基于PicoDet-S轻量模型在中英文论文、杂志和研报等场景上自建数据集训练的高效率版面区域定位模型</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_3cls_pretrained.pdparams">训练模型</a></td>
+<td>89.0</td>
+<td>12.80 / 9.57</td>
+<td>45.04 / 23.86</td>
+<td>22.6</td>
+<td>基于PicoDet-L在中英文论文、杂志和研报等场景上自建数据集训练的效率精度均衡版面区域定位模型</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_3cls_pretrained.pdparams">训练模型</a></td>
+<td>95.8</td>
+<td>114.80 / 25.65</td>
+<td>924.38 / 924.38</td>
+<td>470.1</td>
+<td>基于RT-DETR-H在中英文论文、杂志和研报等场景上自建数据集训练的高精度版面区域定位模型</td>
+</tr>
+</tbody></table>
+
+* <b>17类区域检测模型，包含17个版面常见类别，分别是：段落标题、图片、文本、数字、摘要、内容、图表标题、公式、表格、表格标题、参考文献、文档标题、脚注、页眉、算法、页脚、印章</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>87.4</td>
+<td>8.80 / 3.62</td>
+<td>17.51 / 6.35</td>
+<td>4.8</td>
+<td>基于PicoDet-S轻量模型在中英文论文、杂志和研报等场景上自建数据集训练的高效率版面区域定位模型</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>89.0</td>
+<td>12.60 / 10.27</td>
+<td>43.70 / 24.42</td>
+<td>22.6</td>
+<td>基于PicoDet-L在中英文论文、杂志和研报等场景上自建数据集训练的效率精度均衡版面区域定位模型</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>98.3</td>
+<td>115.29 / 101.18</td>
+<td>964.75 / 964.75</td>
+<td>470.2</td>
+<td>基于RT-DETR-H在中英文论文、杂志和研报等场景上自建数据集训练的高精度版面区域定位模型</td>
+</tr>
+</tbody>
+</table>
+</details>
+</details>
+
+<details>
+<summary> <b>文档图像方向分类模块（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>Top-1 Acc（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">训练模型</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>基于PP-LCNet_x1_0的文档图像分类模型，含有四个类别，即0度，90度，180度，270度</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary> <b>文本图像矫正模块（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>CER </th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">训练模型</a></td>
+<td>0.179</td>
+<td>19.05 / 19.05</td>
+<td>- / 869.82</td>
+<td>30.3</td>
+<td>高精度文本图像矫正模型</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary> <b>印章文本检测模块：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>检测Hmean（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-OCRv4_server_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_seal_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_seal_det_pretrained.pdparams">训练模型</a></td>
+<td>98.40</td>
+<td>124.64 / 91.57</td>
+<td>545.68 / 439.86</td>
+<td>109</td>
+<td>PP-OCRv4的服务端印章文本检测模型，精度更高，适合在较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_seal_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_seal_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_seal_det_pretrained.pdparams">训练模型</a></td>
+<td>96.36</td>
+<td>9.70 / 3.56</td>
+<td>50.38 / 19.64</td>
+<td>4.7</td>
+<td>PP-OCRv4的移动端印章文本检测模型，效率更高，适合在端侧部署</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>文本识别模块：</b></summary>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>86.38</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec 是新一代文本识别模型。该模型致力于以单一模型高效、精准地支持简体中文、繁体中文、英文、日文四种主要语言，以及手写、竖版、拼音、生僻字等复杂文本场景的识别。在保持识别效果的同时，兼顾推理速度和模型鲁棒性，为各种场景下的文档理解提供高效、精准的技术支撑。</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.29</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">训练模型</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc是在PP-OCRv4_server_rec的基础上，在更多中文文档数据和PP-OCR训练数据的混合数据训练而成，增加了部分繁体字、日文、特殊字符的识别能力，可支持识别的字符为1.5万+，除文档相关的文字识别能力提升外，也同时提升了通用文字的识别能力</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>PP-OCRv4的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>PP-OCRv4的服务器端模型，推理精度高，可以部署在多种不同的服务器上</td>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>基于PP-OCRv4识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+</table>
+
+> ❗ 以上列出的是文本识别模块重点支持的<b>6个核心模型</b>，该模块总共支持<b>20个全量模型</b>，包含多个多语言文本识别模型，完整的模型列表如下：
+
+<details><summary> 👉模型列表详情</summary>
+
+* <b>PP-OCRv5 多场景模型</b>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>中文识别 Avg Accuracy(%)</th>
+<th>英文识别 Avg Accuracy(%)</th>
+<th>繁体中文识别 Avg Accuracy(%)</th>
+<th>日文识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>86.38</td>
+<td>64.70</td>
+<td>93.29</td>
+<td>60.35</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec 是新一代文本识别模型。该模型致力于以单一模型高效、精准地支持简体中文、繁体中文、英文、日文四种主要语言，以及手写、竖版、拼音、生僻字等复杂文本场景的识别。在保持识别效果的同时，兼顾推理速度和模型鲁棒性，为各种场景下的文档理解提供高效、精准的技术支撑。</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.29</td>
+<td>66.00</td>
+<td>83.55</td>
+<td>54.65</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+</table>
+
+* <b>中文识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">训练模型</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc是在PP-OCRv4_server_rec的基础上，在更多中文文档数据和PP-OCR训练数据的混合数据训练而成，增加了部分繁体字、日文、特殊字符的识别能力，可支持识别的字符为1.5万+，除文档相关的文字识别能力提升外，也同时提升了通用文字的识别能力</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>PP-OCRv4的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>PP-OCRv4的服务器端模型，推理精度高，可以部署在多种不同的服务器上</td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>72.96</td>
+<td>3.89 / 1.16</td>
+<td>8.72 / 3.56</td>
+<td>10.3</td>
+<td>PP-OCRv3的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>ch_SVTRv2_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_SVTRv2_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_SVTRv2_rec_pretrained.pdparams">训练模型</a></td>
+<td>68.81</td>
+<td>10.38 / 8.31</td>
+<td>66.52 / 30.83</td>
+<td>80.5</td>
+<td rowspan="1">
+SVTRv2 是一种由复旦大学视觉与学习实验室（FVL）的OpenOCR团队研发的服务端文本识别模型，其在PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务中荣获一等奖，A榜端到端识别精度相比PP-OCRv4提升6%。
+</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>ch_RepSVTR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_RepSVTR_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_RepSVTR_rec_pretrained.pdparams">训练模型</a></td>
+<td>65.07</td>
+<td>6.29 / 1.57</td>
+<td>20.64 / 5.40</td>
+<td>48.8</td>
+<td rowspan="1">    RepSVTR 文本识别模型是一种基于SVTRv2 的移动端文本识别模型，其在PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务中荣获一等奖，B榜端到端识别精度相比PP-OCRv4提升2.5%，推理速度持平。</td>
+</tr>
+</table>
+
+* <b>英文识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td> 70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>基于PP-OCRv4识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+<tr>
+<td>en_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>70.69</td>
+<td>3.56 / 0.78</td>
+<td>8.44 / 5.78</td>
+<td>17.3</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+</table>
+
+
+* <b>多语言识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>korean_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>60.21</td>
+<td>3.73 / 0.98</td>
+<td>8.76 / 2.91</td>
+<td>9.6</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量韩文识别模型，支持韩文、数字识别</td>
+</tr>
+<tr>
+<td>japan_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+japan_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/japan_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>45.69</td>
+<td>3.86 / 1.01</td>
+<td>8.62 / 2.92</td>
+<td>9.8</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量日文识别模型，支持日文、数字识别</td>
+</tr>
+<tr>
+<td>chinese_cht_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+chinese_cht_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/chinese_cht_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>82.06</td>
+<td>3.90 / 1.16</td>
+<td>9.24 / 3.18</td>
+<td>10.8</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量繁体中文识别模型，支持繁体中文、数字识别</td>
+</tr>
+<tr>
+<td>te_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+te_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>95.88</td>
+<td>3.59 / 0.81</td>
+<td>8.28 / 6.21</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量泰卢固文识别模型，支持泰卢固文、数字识别</td>
+</tr>
+<tr>
+<td>ka_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ka_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ka_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>96.96</td>
+<td>3.49 / 0.89</td>
+<td>8.63 / 2.77</td>
+<td>17.4</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量卡纳达文识别模型，支持卡纳达文、数字识别</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ta_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>76.83</td>
+<td>3.49 / 0.86</td>
+<td>8.35 / 3.41</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量泰米尔文识别模型，支持泰米尔文、数字识别</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>76.93</td>
+<td>3.53 / 0.78</td>
+<td>8.50 / 6.83</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量拉丁文识别模型，支持拉丁文、数字识别</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+arabic_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>73.55</td>
+<td>3.60 / 0.83</td>
+<td>8.44 / 4.69</td>
+<td>17.3</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量阿拉伯字母识别模型，支持阿拉伯字母、数字识别</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+cyrillic_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>94.28</td>
+<td>3.56 / 0.79</td>
+<td>8.22 / 2.76</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量斯拉夫字母识别模型，支持斯拉夫字母、数字识别</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>96.44</td>
+<td>3.60 / 0.78</td>
+<td>6.95 / 2.87</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量梵文字母识别模型，支持梵文字母、数字识别</td>
+</tr>
+</table>
+</details>
+</details>
+
+<details>
+<summary> <b>测试环境说明：</b></summary>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+            <li><strong>测试数据集：
+             </strong>
+                <ul>
+                  <li>文档图像方向分类模型：自建的内部数据集，覆盖证件和文档等多个场景，包含 1000 张图片。</li>
+                  <li>文本图像矫正模型：<a href="https://www3.cs.stonybrook.edu/~cvl/docunet.html">DocUNet</a>。</li>
+                   <li>版面区域检测模型：PaddleOCR 自建的版面区域检测数据集，包含中英文论文、杂志、合同、书本、试卷和研报等常见的 500 张文档类型图片。</li>
+                    <li>3类版面检测模型：PaddleOCR 自建的版面区域检测数据集，包含中英文论文、杂志和研报等常见的 1154 张文档类型图片。</li>
+                  <li> 17类区域检测模型：PaddleOCR 自建的版面区域检测数据集，包含中英文论文、杂志和研报等常见的 892 张文档类型图片。</li>
+                  <li>文本检测模型：PaddleOCR 自建的中文数据集，覆盖街景、网图、文档、手写多个场景，其中检测包含 500 张图片。</li>
+                   <li>中文识别模型： PaddleOCR 自建的中文数据集，覆盖街景、网图、文档、手写多个场景，其中文本识别包含 1.1w 张图片。</li>
+                    <li>ch_SVTRv2_rec：<a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务</a>A榜评估集。</li>
+                  <li>ch_RepSVTR_rec：<a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务</a>B榜评估集。</li>
+                  <li>英文识别模型：自建的内部英文数据集。</li>
+                   <li>多语言识别模型：自建的内部多语种数据集。</li>
+                    <li>文本行方向分类模型：自建的内部数据集，覆盖证件和文档等多个场景，包含 1000 张图片。</li>
+                   <li>印章文本检测模型：自建的内部数据集，包含500张圆形印章图像。</li>
+                </ul>
+             </li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td><code>paddle_static</code></td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+</details>
+
+<br />
+<b>如您更考虑模型精度，请选择精度较高的模型，如您更考虑模型推理速度，请选择推理速度较快的模型，如您更考虑模型存储大小，请选择存储大小较小的模型</b>。
+
+## 2. 快速开始
+
+在本地使用印章文本识别产线前，请确保您已经按照[安装教程](../installation.md)完成了wheel包安装。如果您希望选择性安装依赖，请参考安装教程中的相关说明。该产线对应的依赖分组为 `doc-parser`。安装完成后，可以在本地使用命令行体验或 Python 集成。
+
+**请注意，如果在执行过程中遇到程序失去响应、程序异常退出、内存资源耗尽、推理速度极慢等问题，请尝试参考文档调整配置，例如关闭不需要使用的功能或使用更轻量的模型。**
+
+### 2.1 命令行方式体验
+
+一行命令即可快速体验 seal_recognition 产线效果：
+
+```bash
+paddleocr seal_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/seal_text_det.png \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False
+
+# 通过 --device 指定模型推理时使用 GPU
+paddleocr seal_recognition -i ./seal_text_det.png --device gpu
+```
+
+上述命令默认使用本地推理引擎 `paddle_static`。如需运行，请先参考[飞桨框架安装说明](../paddlepaddle_installation.md)安装 PaddlePaddle。
+
+如果选择 `transformers` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 Transformers 环境配置，然后执行如下命令：
+
+```bash
+# 使用 transformers 引擎进行推理
+paddleocr seal_recognition -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/seal_text_det.png \
+    --use_doc_orientation_classify False \
+    --use_doc_unwarping False \
+    --engine transformers
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+<details><summary><b>命令行支持更多参数设置，点击展开以查看命令行参数的详细说明</b></summary>
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，必填。<br/>
+<b>说明：</b>
+如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；
+<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/seal_text_det.png">示例</a>；
+<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td><b>含义：</b>指定推理结果文件保存的路径。<br/>
+<b>说明：</b>
+如果不设置，推理结果将不会保存到本地。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>含义：</b>文档方向分类模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>文档方向分类模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>含义：</b>文本图像矫正模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>含义：</b>版面检测模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>含义：</b>版面检测模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_name</code></td>
+<td><b>含义：</b>印章文本检测模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_dir</code></td>
+<td><b>含义：</b>印章文本检测模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>含义：</b>文本识别模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>含义：</b>文本识别模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>含义：</b>文本识别模型的batch size。<br/>
+<b>说明：</b>
+如果不设置，将默认设置batch size为<code>1</code>。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否加载并使用文档方向分类模块。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>含义：</b>是否加载并使用版面检测模块。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>版面模型得分阈值。<br/>
+<b>说明：</b>
+<code>0-1</code> 之间的任意浮点数。如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>0.5</code>。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>版面检测是否使用后处理NMS。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>检测框的边长缩放倍数。<br/>
+<b>说明：</b>
+大于0的浮点数，如 1.1 ，表示将模型输出的检测框中心不变，宽和高都扩张1.1倍如果不设置，将使用产线初始化的该参数值，默认初始化为1.0。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>版面检测中模型输出的检测框的合并处理模式。<br/>
+<b>说明：</b>
+<ul>
+<li><b>large</b>，设置为large时，表示在模型输出的检测框中，对于互相重叠包含的检测框，只保留外部最大的框，删除重叠的内部框；</li>
+<li><b>small</b>，设置为small，表示在模型输出的检测框中，对于互相重叠包含的检测框，只保留内部被包含的小框，删除重叠的外部框；</li>
+<li><b>union</b>，不进行框的过滤处理，内外框都保留；</li>
+</ul>如果不设置，将使用产线初始化的该参数值，默认初始化为<code>large</code>。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>含义：</b>印章文本检测的图像边长限制。<br/>
+<b>说明：</b>
+大于 <code>0</code> 的任意整数。如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>736</code>。
+</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>含义：</b>印章文本检测的图像边长限制类型。<br/>
+<b>说明：</b>
+支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>。如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>min</code>。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>含义：</b>检测像素阈值，输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。<br/>
+<b>说明：</b>大于<code>0</code>的任意浮点数。
+如果不设置，将默认使用产线初始化的该参数值 <code>0.2</code>。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>含义：</b>检测框阈值，检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。<br/>
+<b>说明：</b>大于<code>0</code>的任意浮点数
+。如果不设置，将默认使用产线初始化的该参数值 <code>0.6</code>。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>含义：</b>印章文本检测扩张系数，使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。<br/>
+<b>说明：</b>
+大于<code>0</code>的任意浮点数
+。如果不设置，将默认使用产线初始化的该参数值 <code>0.5</code>。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>含义：</b>文本识别阈值，得分大于该阈值的文本结果会被保留。<br/>
+<b>说明：</b>
+大于<code>0</code>的任意浮点数
+。如果不设置，将默认使用产线初始化的该参数值 <code>0.0</code>。即不设阈值。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+</ul>如果不设置，将默认使用产线初始化的该参数值，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>fp32</code>、<code>fp16</code>。</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+
+</tbody>
+</table>
+</details>
+<br />
+
+运行结果会被打印到终端上，默认配置的 seal_recognition 产线的运行结果如下：
+
+```bash
+{'res': {'input_path': './seal_text_det.png', 'model_settings': {'use_doc_preprocessor': True, 'use_layout_detection': True}, 'doc_preprocessor_res': {'input_path': None, 'page_index': None, 'model_settings': {'use_doc_orientation_classify': False, 'use_doc_unwarping': False}, 'angle': -1}, 'layout_det_res': {'input_path': None, 'page_index': None, 'boxes': [{'cls_id': 16, 'label': 'seal', 'score': 0.975529670715332, 'coordinate': [6.191284, 0.16680908, 634.39325, 628.85345]}]}, 'seal_res_list': [{'input_path': None, 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_textline_orientation': False}, 'dt_polys': [array([[320,  38],
+       ...,
+       [315,  38]]), array([[461, 347],
+       ...,
+       [456, 346]]), array([[439, 445],
+       ...,
+       [434, 444]]), array([[158, 468],
+       ...,
+       [154, 466]])], 'text_det_params': {'limit_side_len': 736, 'limit_type': 'min', 'thresh': 0.2, 'max_side_limit': 4000, 'box_thresh': 0.6, 'unclip_ratio': 0.5}, 'text_type': 'seal', 'textline_orientation_angles': array([-1, ..., -1]), 'text_rec_score_thresh': 0, 'rec_texts': ['天津君和缘商贸有限公司', '发票专用章', '吗繁物', '5263647368706'], 'rec_scores': array([0.99340463, ..., 0.9916274 ]), 'rec_polys': [array([[320,  38],
+       ...,
+       [315,  38]]), array([[461, 347],
+       ...,
+       [456, 346]]), array([[439, 445],
+       ...,
+       [434, 444]]), array([[158, 468],
+       ...,
+       [154, 466]])], 'rec_boxes': array([], dtype=float64)}]}}
+```
+
+可视化结果保存在`save_path`下，其中印章OCR的可视化结果如下：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/seal_recognition/04.png"/>
+
+### 2.2 Python脚本方式集成
+
+命令行方式是为了快速体验查看效果，一般来说，在项目中，往往需要通过代码集成，您可以通过几行代码即可完成产线的快速推理，推理代码如下：
+
+```python
+from paddleocr import SealRecognition
+
+pipeline = SealRecognition(
+    use_doc_orientation_classify=False, # 通过 use_doc_orientation_classify 指定是否使用文档方向分类模型
+    use_doc_unwarping=False, # 通过 use_doc_unwarping 指定是否使用文本图像矫正模块
+)
+# ocr = SealRecognition(device="gpu") # 通过 device 指定模型推理时使用 GPU
+output = pipeline.predict("./seal_text_det.png")
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_img("./output/")
+    res.save_to_json("./output/")
+```
+
+上述代码默认使用本地推理引擎 `paddle_static`。如需运行，请先参考[飞桨框架安装说明](../paddlepaddle_installation.md)安装 PaddlePaddle。
+
+如果选择 `transformers` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 Transformers 环境配置，然后执行如下代码：
+
+```python
+from paddleocr import SealRecognition
+
+pipeline = SealRecognition(
+    engine="transformers",
+)
+# ocr = SealRecognition(device="gpu") # 通过 device 指定模型推理时使用 GPU
+output = pipeline.predict("./seal_text_det.png")
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_img("./output/")
+    res.save_to_json("./output/")
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+在上述 Python 脚本中，执行了如下几个步骤：
+
+（1）通过 <code>SealRecognition()</code> 实例化 印章文本识别 产线对象，具体参数说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td>v<b>含义：</b>文档方向分类模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>文档方向分类模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td></td><b>含义：</b>文本图像矫正模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>含义：</b>版面检测模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>含义：</b>版面检测模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_name</code></td>
+<td><b>含义：</b>印章文本检测模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_text_detection_model_dir</code></td>
+<td><b>含义：</b>印章文本检测模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>含义：</b>文本识别模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>含义：</b>文本识别模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>含义：</b>文本识别模型的batch size。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否加载并使用文档方向分类模块。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>含义：</b>是否加载并使用版面检测模块。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>版面检测置信度阈值，得分大于该阈值才会被输出。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>dict</b>：key是int类别id，value是大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.5</code>。</li></li></li></ul>
+</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>版面检测是否使用后处理NMS。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>版面区域检测模型检测框的扩张系数。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：任意大于 <code>0</code>  浮点数；</li>
+<li><b>Tuple[float,float]</b>：在横纵两个方向各自的扩张系数；</li>
+<li><b>dict</b>，dict的key为<b>int</b>类型，代表<code>cls_id</code>, value为<b>tuple</b>类型，如<code>{0: (1.1, 2.0)}</code>，表示将模型输出的第0类别检测框中心不变，宽度扩张1.1倍，高度扩张2.0倍；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>1.0</code>。</li>
+</ul>
+</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>版面区域检测的重叠框过滤方式。<br/>
+<b>说明：</b>
+<ul>
+<li><b>str</b>：<code>large</code>，<code>small</code>，<code>union</code>，分别表示重叠框过滤时选择保留大框，小框还是同时保留；</li>
+<li><b>dict</b>： dict的key为<b>int</b>类型，代表<code>cls_id</code>，value为<b>str</b>类型，如<code>{0: "large", 2: "small"}</code>，表示对第0类别检测框使用large模式，对第2类别检测框使用small模式；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>large</code>。</li>
+</ul>
+</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>含义：</b>印章文本检测的图像边长限制。<br/>
+<b>说明：</b>
+<ul>
+<li><b>int</b>：大于 <code>0</code> 的任意整数；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>736</code>。</li>
+</ul>
+</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>含义：</b>印章文本检测的图像边长限制类型。<br/>
+<b>说明：</b>
+<ul>
+<li><b>str</b>：支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>min</code>。</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>含义：</b>检测像素阈值，输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.2</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>含义：</b>检测框阈值，检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+ <li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.6</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>含义：</b>印章文本检测扩张系数，使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.5</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>含义：</b>文本识别阈值，得分大于该阈值的文本结果会被保留。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+    <li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.0</code>，即不设阈值。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+<li><b>None</b>：如果设置为<code>None</code>，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>engine_config</code></td>
+<td><b>含义：</b>推理引擎配置。<br><b>说明：</b>推荐与 <code>engine</code> 搭配使用。详细字段、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+
+</tbody>
+</table>
+
+（2）调用 印章文本识别 产线对象的 <code>predict()</code>方法进行推理预测，该方法会返回一个结果列表。
+
+另外，产线还提供了 <code>predict_iter() </code>方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。
+
+以下是 <code>predict_iter()</code> 方法的参数及其说明：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据；</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/seal_text_det.png">示例</a>；<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)；</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]。</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否在推理时使用文档方向分类模块。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否在推理时使用文本图像矫正模块。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>含义：</b>是否在推理时使用版面检测模块。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_threshold</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_nms</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|Tuple[float,float]|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_merge_bboxes_mode</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>str|dict|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_side_len</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_limit_type</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_box_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_det_unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>seal_rec_score_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tr></table>
+
+（3）对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为<code>json</code>文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化。</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效。</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效。</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致。</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效。</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效。</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将结果保存为图像格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径。</td>
+<td>无</td>
+</tr>
+</table>
+
+<ul>
+<li>调用<code>print()</code> 方法会将结果打印到终端，打印到终端的内容解释如下：
+    <ol start="1" type="1">
+    <li><code>input_path</code>: <code>(str)</code> 待预测图像的输入路径</li>
+    <li><code>page_index</code>: <code>(Union[int, None])</code> 如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+    <li><code>model_settings</code>: <code>(Dict[str, bool])</code> 配置产线所需的模型参数
+        <ol>
+        <li><code>use_doc_preprocessor</code>: <code>(bool)</code> 控制是否启用文档预处理子产线</li>
+        <li><code>use_layout_detection</code>: <code>(bool)</code> 控制是否启用版面检测子模块</li>
+        </ol>
+    </li>
+    <li><code>layout_det_res</code>: <code>(Dict[str, Union[List[numpy.ndarray], List[float]]])</code> 版面检测子模块的输出结果。仅当<code>use_layout_detection=True</code>时存在
+        <ol>
+        <li><code>input_path</code>: <code>(Union[str, None])</code> 版面检测区域模块接受的图像路径，当输入为<code>numpy.ndarray</code>时，保存为<code>None</code></li>
+        <li><code>page_index</code>: <code>(Union[int, None])</code> 如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+        <li><code>boxes</code>: <code>(List[Dict])</code> 版面印章区域的检测框列表，每个列表中的元素，包含以下字段</li>
+            <ul>
+            <li><code>cls_id</code>: <code>(int)</code> 检测框的印章类别id</li>
+            <li><code>score</code>: <code>(float)</code> 检测框的置信度</li>
+            <li><code>coordinate</code>: <code>(List[float])</code> 检测框的四个顶点坐标，顺序为x1，y1，x2，y2表示左上角的x坐标，左上角的y坐标，右下角x坐标，右下角的y坐标</li>
+            </ul>
+        </ol>
+    </li>
+    <li><code>seal_res_list</code>: <code>List[Dict]</code> 印章文本识别的结果列表，每个元素包含以下字段
+        <ol>
+        <li><code>input_path</code>: <code>(Union[str, None])</code> 印章文本识别产线接受的图像路径，当输入为<code>numpy.ndarray</code>时，保存为<code>None</code></li>
+        <li><code>page_index</code>: <code>(Union[int, None])</code> 如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+        <li><code>model_settings</code>: <code>(Dict[str, bool])</code> 印章文本识别产线的模型配置参数</li>
+            <ul>
+            <li><code>use_doc_preprocessor</code>: <code>(bool)</code> 控制是否启用文档预处理子产线</li>
+            <li><code>use_textline_orientation</code>: <code>(bool)</code> 控制是否启用文本行方向分类子模块</li>
+            </ul>
+        </ol>
+    </li>
+    <li><code>doc_preprocessor_res</code>: <code>(Dict[str, Union[str, Dict[str, bool], int]])</code> 文档预处理子产线的输出结果。仅当<code>use_doc_preprocessor=True</code>时存在
+        <ol>
+        <li><code>input_path</code>: <code>(Union[str, None])</code> 图像预处理子产线接受的图像路径，当输入为<code>numpy.ndarray</code>时，保存为<code>None</code></li>
+        <li><code>model_settings</code>: <code>(Dict[str, bool])</code> 预处理子产线的模型配置参数</li>
+            <ul>
+            <li><code>use_doc_orientation_classify</code>: <code>(bool)</code> 控制是否启用文档方向分类</li>
+            <li><code>use_doc_unwarping</code>: <code>(bool)</code> 控制是否启用文本图像矫正</li>
+            </ul>
+        <li><code>angle</code>: <code>(int)</code> 文档方向分类的预测结果。启用时取值为[0,1,2,3]，分别对应[0°,90°,180°,270°]；未启用时为-1</li>
+        </ol>
+    </li>
+    <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> 印章文本检测的多边形框列表。每个检测框由多个顶点坐标构成的numpy数组表示，数组shape为(n, 2)</li>
+    <li><code>dt_scores</code>: <code>(List[float])</code> 文本检测框的置信度列表</li>
+    <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> 文本检测模块的配置参数
+        <ol>
+        <li><code>limit_side_len</code>: <code>(int)</code> 图像预处理时的边长限制值</li>
+        <li><code>limit_type</code>: <code>(str)</code> 边长限制的处理方式</li>
+        <li><code>thresh</code>: <code>(float)</code> 文本像素分类的置信度阈值</li>
+        <li><code>box_thresh</code>: <code>(float)</code> 文本检测框的置信度阈值</li>
+        <li><code>unclip_ratio</code>: <code>(float)</code> 文本检测框的膨胀系数</li>
+        <li><code>text_type</code>: <code>(str)</code> 印章文本检测的类型，当前固定为"seal"</li>
+        </ol>
+    </li>
+    <li><code>text_rec_score_thresh</code>: <code>(float)</code> 文本识别结果的过滤阈值</li>
+    <li><code>rec_texts</code>: <code>(List[str])</code> 文本识别结果列表，仅包含置信度超过<code>text_rec_score_thresh</code>的文本</li>
+     <li><code>rec_scores</code>: <code>(List[float])</code> 文本识别的置信度列表，已按<code>text_rec_score_thresh</code>过滤</li>
+     <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> 经过置信度过滤的文本检测框列表，格式同<code>dt_polys</code></li>
+     <li><code>rec_boxes</code>: <code>(numpy.ndarray)</code> 检测框的矩形边界框数组，印章识别产线为空数组</li>
+    </ol>
+</li>
+<li>调用<code>save_to_json()</code> 方法会将上述内容保存到指定的<code>save_path</code>中，如果指定为目录，则保存的路径为<code>save_path/{your_img_basename}_res.json</code>，如果指定为文件，则直接保存到该文件中。由于json文件不支持保存numpy数组，因此会将其中的<code>numpy.array</code>类型转换为列表形式。</li>
+<li>调用<code>save_to_img()</code> 方法会将可视化结果保存到指定的<code>save_path</code>中，如果指定为目录，则保存的路径为<code>save_path/{your_img_basename}_seal_res_region1.{your_img_extension}</code>，如果指定为文件，则直接保存到该文件中。(产线通常包含较多结果图片，不建议直接指定为具体的文件路径，否则多张图会被覆盖，仅保留最后一张图)</li>
+
+* 此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">获取预测的 <code>json</code> 格式的结果</td>
+</tr>
+<tr>
+<td rowspan="2"><code>img</code></td>
+<td rowspan="2">获取格式为 <code>dict</code> 的可视化图像</td>
+</tr>
+</table>
+
+<li><code>json</code> 属性获取的预测结果为dict类型的数据，相关内容与调用 <code>save_to_json()</code> 方法保存的内容一致。</li>   
+<li><code>img</code> 属性返回的预测结果是一个dict类型的数据。其中，键分别为 <code>layout_det_res</code> 、  <code>seal_res_region1</code>和 <code>preprocessed_img</code>，对应的值是三个 <code>Image.Image</code> 对象：一个用于显示版面检测可视化，一个用于显示印章文本识别结果的可视化图像，另一个用于展示图像预处理的可视化图像。如果没有使用图像预处理子模块，则dict中不包含preprocessed_img，如果没有使用版面区域检测模块，则dict中不包含layout_det_res。</li>
+</ul>
+
+## 3. 开发集成/部署
+
+如果产线可以达到您对产线推理速度和精度的要求，您可以直接进行开发集成/部署。
+
+若您需要将产线直接应用在您的Python项目中，可以参考 [2.2 Python脚本方式](#22-python)中的示例代码。
+
+此外，PaddleOCR 也提供了其他两种部署方式，详细说明如下：
+
+🚀 高性能推理：在实际生产环境中，许多应用对部署策略的性能指标（尤其是响应速度）有着较严苛的标准，以确保系统的高效运行与用户体验的流畅性。为此，PaddleOCR 提供高性能推理功能，旨在对模型推理及前后处理进行深度性能优化，实现端到端流程的显著提速，详细的高性能推理流程请参考[高性能推理](../inference_deployment/local_inference/high_performance_inference.md)。
+
+☁️ 服务化部署：服务化部署是实际生产环境中常见的一种部署形式。通过将推理功能封装为服务，客户端可以通过网络请求来访问这些服务，以获取推理结果。详细的产线服务化部署流程请参考[服务化部署](../inference_deployment/serving/serving.md)。
+
+以下是基础服务化部署的API参考与多语言服务调用示例：
+
+<details><summary>API参考</summary>
+<p>对于服务提供的主要操作：</p>
+<ul>
+<li>HTTP请求方法为POST。</li>
+<li>请求体和响应体均为JSON数据（JSON对象）。</li>
+<li>当请求处理成功时，响应状态码为<code>200</code>，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。固定为<code>0</code>。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。固定为<code>"Success"</code>。</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>操作结果。</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>当请求处理未成功时，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。与响应状态码相同。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。</td>
+</tr>
+</tbody>
+</table>
+<p>服务提供的主要操作如下：</p>
+<ul>
+<li><b><code>infer</code></b></li>
+</ul>
+<p>获取印章文本识别结果。</p>
+<p><code>POST /seal-recognition</code></p>
+<ul>
+<li>请求体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>服务器可访问的图像文件（含 TIFF，多页时按页处理）或 PDF 文件的 URL，或上述类型文件内容的 Base64 编码结果。默认对于超过 10 页的 PDF 或多页 TIFF，仅处理前 10 页。<br /> 要解除页数限制，请在产线配置文件中添加以下配置：
+<pre><code>Serving:
+  extra:
+    max_num_input_imgs: null
+</code></pre>
+</td>
+<td>是</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>文件类型。<code>0</code> 表示 PDF 文件，<code>1</code> 表示图像文件（含 TIFF）。若请求体无此属性，则将根据URL推断文件类型。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_doc_orientation_classify</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_doc_unwarping</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useLayoutDetection</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_layout_detection</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutThreshold</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>layout_threshold</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutNms</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>layout_nms</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutUnclipRatio</code></td>
+<td><code>number</code> | <code>array</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>layout_unclip_ratio</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>layoutMergeBboxesMode</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>layout_merge_bboxes_mode</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>seal_det_limit_side_len</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>seal_det_limit_type</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>seal_det_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>seal_det_box_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>seal_det_unclip_ratio</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>sealRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>seal_rec_score_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>是否返回可视化结果图以及处理过程中的中间图像等。
+<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>传入 <code>true</code>：返回图像。</li>
+<li>传入 <code>false</code>：不返回图像。</li>
+<li>若请求体中未提供该参数或传入 <code>null</code>：遵循产线配置文件<code>Serving.visualize</code> 的设置。</li>
+</ul>
+<br/>例如，在产线配置文件中添加如下字段：<br/>
+<pre><code>Serving:
+  visualize: False
+</code></pre>
+将默认不返回图像，通过请求体中的<code>visualize</code>参数可以覆盖默认行为。如果请求体和配置文件中均未设置（或请求体传入<code>null</code>、配置文件中未设置），则默认返回图像。
+</td>
+<td>否</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>请求处理成功时，响应体的<code>result</code>具有如下属性：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>sealRecResults</code></td>
+<td><code>object</code></td>
+<td>印章文本识别结果。数组长度为1（对于图像输入）或实际处理的文档页数（对于PDF输入）。对于PDF输入，数组中的每个元素依次表示PDF文件中实际处理的每一页的结果。</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>输入数据信息。</td>
+</tr>
+</tbody>
+</table>
+<p>下表中涉及图像的字段（如 <code>outputImages</code>、<code>inputImage</code>）默认以 Base64 字符串内联返回；当服务端开启 URL 返回模式时，相应字段的值变为预签名 URL，字段类型保持不变。配置方式参见 <a href="../inference_deployment/serving/serving.md">服务化部署</a>「以 URL 形式返回二进制内容」一节。</p>
+<p><code>sealRecResults</code>中的每个元素为一个<code>object</code>，具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>产线对象的 <code>predict</code> 方法生成结果的 JSON 表示中 <code>res</code> 字段的简化版本，其中去除了 <code>input_path</code> 和 <code>page_index</code> 字段。</td>
+</tr>
+<tr>
+<td><code>outputImages</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>参见产线预测结果的 <code>img</code> 属性说明。图像为JPEG格式，使用Base64编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+<tr>
+<td><code>inputImage</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>输入图像。图像为JPEG格式，使用Base64编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+</tbody>
+</table></details>
+<details><summary>多语言调用服务示例</summary>
+<details>
+<summary>Python</summary>
+
+<pre><code class="language-python">import base64
+import requests
+
+API_URL = "http://localhost:8080/seal-recognition"
+file_path = "./demo.jpg"
+
+with open(file_path, "rb") as file:
+    file_bytes = file.read()
+    file_data = base64.b64encode(file_bytes).decode("ascii")
+
+payload = {"file": file_data, "fileType": 1}
+
+response = requests.post(API_URL, json=payload)
+
+assert response.status_code == 200
+result = response.json()["result"]
+for i, res in enumerate(result["sealRecResults"]):
+    print(res["prunedResult"])
+    for img_name, img in res["outputImages"].items():
+        img_path = f"{img_name}_{i}.jpg"
+        with open(img_path, "wb") as f:
+            f.write(base64.b64decode(img))
+        print(f"Output image saved at {img_path}")
+</code></pre></details>
+
+<details><summary>C++</summary>
+
+<pre><code class="language-cpp">#include &lt;iostream&gt;
+#include &lt;fstream&gt;
+#include &lt;vector&gt;
+#include &lt;string&gt;
+#include "cpp-httplib/httplib.h" // https://github.com/Huiyicc/cpp-httplib
+#include "nlohmann/json.hpp" // https://github.com/nlohmann/json
+#include "base64.hpp" // https://github.com/tobiaslocker/base64
+
+int main() {
+    httplib::Client client("localhost", 8080);
+
+    const std::string filePath = "./demo.jpg";
+
+    std::ifstream file(filePath, std::ios::binary | std::ios::ate);
+    if (!file) {
+        std::cerr << "Error opening file: " << filePath << std::endl;
+        return 1;
+    }
+
+    std::streamsize size = file.tellg();
+    file.seekg(0, std::ios::beg);
+    std::vector<char> buffer(size);
+    if (!file.read(buffer.data(), size)) {
+        std::cerr << "Error reading file." << std::endl;
+        return 1;
+    }
+
+    std::string bufferStr(buffer.data(), static_cast<size_t>(size));
+    std::string encodedFile = base64::to_base64(bufferStr);
+
+    nlohmann::json jsonObj;
+    jsonObj["file"] = encodedFile;
+    jsonObj["fileType"] = 1;
+
+    auto response = client.Post("/seal-recognition", jsonObj.dump(), "application/json");
+
+    if (response && response->status == 200) {
+        nlohmann::json jsonResponse = nlohmann::json::parse(response->body);
+        auto result = jsonResponse["result"];
+
+        if (!result.is_object() || !result["sealRecResults"].is_array()) {
+            std::cerr << "Unexpected response format." << std::endl;
+            return 1;
+        }
+
+        for (size_t i = 0; i < result["sealRecResults"].size(); ++i) {
+            auto res = result["sealRecResults"][i];
+
+            if (res.contains("prunedResult")) {
+                std::cout << "Recognized seal result: " << res["prunedResult"].dump() << std::endl;
+            }
+
+            if (res.contains("outputImages") && res["outputImages"].is_object()) {
+                for (auto& [imgName, imgData] : res["outputImages"].items()) {
+                    std::string outputPath = imgName + "_" + std::to_string(i) + ".jpg";
+                    std::string decodedImage = base64::from_base64(imgData.get<std::string>());
+
+                    std::ofstream outFile(outputPath, std::ios::binary);
+                    if (outFile.is_open()) {
+                        outFile.write(decodedImage.c_str(), decodedImage.size());
+                        outFile.close();
+                        std::cout << "Saved image: " << outputPath << std::endl;
+                    } else {
+                        std::cerr << "Failed to write image: " << outputPath << std::endl;
+                    }
+                }
+            }
+        }
+    } else {
+        std::cerr << "Request failed." << std::endl;
+        if (response) {
+            std::cerr << "HTTP status: " << response->status << std::endl;
+            std::cerr << "Response body: " << response->body << std::endl;
+        }
+        return 1;
+    }
+
+    return 0;
+}
+</code></pre></details>
+
+<details><summary>Java</summary>
+
+<pre><code class="language-java">import okhttp3.*;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.util.Base64;
+
+public class Main {
+    public static void main(String[] args) throws IOException {
+        String API_URL = "http://localhost:8080/seal-recognition";
+        String imagePath = "./demo.jpg";
+
+        File file = new File(imagePath);
+        byte[] fileContent = java.nio.file.Files.readAllBytes(file.toPath());
+        String base64Image = Base64.getEncoder().encodeToString(fileContent);
+
+        ObjectMapper objectMapper = new ObjectMapper();
+        ObjectNode payload = objectMapper.createObjectNode();
+        payload.put("file", base64Image);
+        payload.put("fileType", 1);
+
+        OkHttpClient client = new OkHttpClient();
+        MediaType JSON = MediaType.get("application/json; charset=utf-8");
+        RequestBody body = RequestBody.create(JSON, payload.toString());
+
+        Request request = new Request.Builder()
+                .url(API_URL)
+                .post(body)
+                .build();
+
+        try (Response response = client.newCall(request).execute()) {
+            if (response.isSuccessful()) {
+                String responseBody = response.body().string();
+                JsonNode root = objectMapper.readTree(responseBody);
+                JsonNode result = root.get("result");
+
+                JsonNode sealRecResults = result.get("sealRecResults");
+                for (int i = 0; i < sealRecResults.size(); i++) {
+                    JsonNode item = sealRecResults.get(i);
+                    int finalI = i;
+
+                    JsonNode prunedResult = item.get("prunedResult");
+                    System.out.println("Pruned Result [" + i + "]: " + prunedResult.toString());
+
+                    JsonNode outputImages = item.get("outputImages");
+                    if (outputImages != null && outputImages.isObject()) {
+                        outputImages.fieldNames().forEachRemaining(imgName -> {
+                            try {
+                                String imgBase64 = outputImages.get(imgName).asText();
+                                byte[] imgBytes = Base64.getDecoder().decode(imgBase64);
+                                String imgPath = imgName + "_" + finalI + ".jpg";
+                                try (FileOutputStream fos = new FileOutputStream(imgPath)) {
+                                    fos.write(imgBytes);
+                                    System.out.println("Saved image: " + imgPath);
+                                }
+                            } catch (IOException e) {
+                                System.err.println("Failed to save image: " + e.getMessage());
+                            }
+                        });
+                    }
+                }
+            } else {
+                System.err.println("Request failed with HTTP code: " + response.code());
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Go</summary>
+
+<pre><code class="language-go">package main
+
+import (
+	"bytes"
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"io/ioutil"
+	"net/http"
+)
+
+func main() {
+	API_URL := "http://localhost:8080/seal-recognition"
+	filePath := "./demo.jpg"
+
+	fileBytes, err := ioutil.ReadFile(filePath)
+	if err != nil {
+		fmt.Printf("Error reading file: %v\n", err)
+		return
+	}
+	fileData := base64.StdEncoding.EncodeToString(fileBytes)
+
+	payload := map[string]interface{}{
+		"file":     fileData,
+		"fileType": 1,
+	}
+	payloadBytes, err := json.Marshal(payload)
+	if err != nil {
+		fmt.Printf("Error marshaling payload: %v\n", err)
+		return
+	}
+
+	client := &http.Client{}
+	req, err := http.NewRequest("POST", API_URL, bytes.NewBuffer(payloadBytes))
+	if err != nil {
+		fmt.Printf("Error creating request: %v\n", err)
+		return
+	}
+	req.Header.Set("Content-Type", "application/json")
+
+	resp, err := client.Do(req)
+	if err != nil {
+		fmt.Printf("Error sending request: %v\n", err)
+		return
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		fmt.Printf("Unexpected status code: %d\n", resp.StatusCode)
+		return
+	}
+
+	body, err := ioutil.ReadAll(resp.Body)
+	if err != nil {
+		fmt.Printf("Error reading response body: %v\n", err)
+		return
+	}
+
+	type SealResult struct {
+		PrunedResult map[string]interface{}   `json:"prunedResult"`
+		OutputImages map[string]string        `json:"outputImages"`
+		InputImage   *string                  `json:"inputImage"`
+	}
+
+	type Response struct {
+		Result struct {
+			SealRecResults []SealResult  `json:"sealRecResults"`
+			DataInfo       interface{}   `json:"dataInfo"`
+		} `json:"result"`
+	}
+
+	var respData Response
+	if err := json.Unmarshal(body, &respData); err != nil {
+		fmt.Printf("Error unmarshaling response: %v\n", err)
+		return
+	}
+
+	for i, res := range respData.Result.SealRecResults {
+		fmt.Printf("Pruned Result %d: %+v\n", i, res.PrunedResult)
+
+		for name, imgBase64 := range res.OutputImages {
+			imgBytes, err := base64.StdEncoding.DecodeString(imgBase64)
+			if err != nil {
+				fmt.Printf("Error decoding image %s: %v\n", name, err)
+				continue
+			}
+
+			filename := fmt.Sprintf("%s_%d.jpg", name, i)
+			if err := ioutil.WriteFile(filename, imgBytes, 0644); err != nil {
+				fmt.Printf("Error saving image %s: %v\n", filename, err)
+				continue
+			}
+			fmt.Printf("Output image saved at %s\n", filename)
+		}
+	}
+}
+</code></pre></details>
+
+<details><summary>C#</summary>
+
+<pre><code class="language-csharp">using System;
+using System.IO;
+using System.Net.Http;
+using System.Text;
+using System.Threading.Tasks;
+using Newtonsoft.Json.Linq;
+
+class Program
+{
+    static readonly string API_URL = "http://localhost:8080/seal-recognition";
+    static readonly string inputFilePath = "./demo.jpg";
+
+    static async Task Main(string[] args)
+    {
+        var httpClient = new HttpClient();
+
+        byte[] fileBytes = File.ReadAllBytes(inputFilePath);
+        string fileData = Convert.ToBase64String(fileBytes);
+
+        var payload = new JObject
+        {
+            { "file", fileData },
+            { "fileType", 1 }
+        };
+        var content = new StringContent(payload.ToString(), Encoding.UTF8, "application/json");
+
+        HttpResponseMessage response = await httpClient.PostAsync(API_URL, content);
+        response.EnsureSuccessStatusCode();
+
+        string responseBody = await response.Content.ReadAsStringAsync();
+        JObject jsonResponse = JObject.Parse(responseBody);
+
+        JArray sealRecResults = (JArray)jsonResponse["result"]["sealRecResults"];
+        for (int i = 0; i < sealRecResults.Count; i++)
+        {
+            var res = sealRecResults[i];
+            Console.WriteLine($"[{i}] prunedResult:\n{res["prunedResult"]}");
+
+            JObject outputImages = res["outputImages"] as JObject;
+            if (outputImages != null)
+            {
+                foreach (var img in outputImages)
+                {
+                    string imgName = img.Key;
+                    string base64Img = img.Value?.ToString();
+                    if (!string.IsNullOrEmpty(base64Img))
+                    {
+                        string imgPath = $"{imgName}_{i}.jpg";
+                        byte[] imageBytes = Convert.FromBase64String(base64Img);
+                        File.WriteAllBytes(imgPath, imageBytes);
+                        Console.WriteLine($"Output image saved at {imgPath}");
+                    }
+                }
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Node.js</summary>
+
+<pre><code class="language-js">const axios = require('axios');
+const fs = require('fs');
+const path = require('path');
+
+const API_URL = 'http://localhost:8080/seal-recognition';
+const imagePath = './demo.jpg';
+
+function encodeImageToBase64(filePath) {
+  const bitmap = fs.readFileSync(filePath);
+  return Buffer.from(bitmap).toString('base64');
+}
+
+const payload = {
+  file: encodeImageToBase64(imagePath),
+  fileType: 1
+};
+
+axios.post(API_URL, payload)
+  .then((response) => {
+    const result = response.data["result"];
+    const sealRecResults = result["sealRecResults"];
+
+    sealRecResults.forEach((res, i) => {
+      console.log(`\n[${i}] prunedResult:\n`, res["prunedResult"]);
+
+      const outputImages = res["outputImages"];
+      if (outputImages) {
+        for (const [imgName, base64Img] of Object.entries(outputImages)) {
+          const imgBuffer = Buffer.from(base64Img, 'base64');
+          const fileName = `${imgName}_${i}.jpg`;
+          fs.writeFileSync(fileName, imgBuffer);
+          console.log(`Output image saved at ${fileName}`);
+        }
+      } else {
+        console.log(`[${i}] No outputImages found.`);
+      }
+    });
+  })
+  .catch((error) => {
+    console.error('Error occurred while calling the API:', error.message);
+  });
+</code></pre></details>
+
+<details><summary>PHP</summary>
+
+<pre><code class="language-php">&lt;?php
+
+$API_URL = "http://localhost:8080/seal-recognition";
+$image_path = "./demo.jpg";
+
+$image_data = base64_encode(file_get_contents($image_path));
+$payload = array("file" => $image_data, "fileType" => 1);
+
+$ch = curl_init($API_URL);
+curl_setopt($ch, CURLOPT_POST, true);
+curl_setopt($ch, CURLOPT_POSTFIELDS, json_encode($payload));
+curl_setopt($ch, CURLOPT_HTTPHEADER, array('Content-Type: application/json'));
+curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+$response = curl_exec($ch);
+curl_close($ch);
+
+$result = json_decode($response, true)["result"]["sealRecResults"];
+
+foreach ($result as $i => $item) {
+    echo "[$i] prunedResult:\n";
+    print_r($item["prunedResult"]);
+
+    if (!empty($item["outputImages"])) {
+        foreach ($item["outputImages"] as $img_name => $base64_img) {
+            if (!empty($base64_img)) {
+                $output_path = "{$img_name}_{$i}.jpg";
+                file_put_contents($output_path, base64_decode($base64_img));
+                echo "Output image saved at $output_path\n";
+            }
+        }
+    } else {
+        echo "No outputImages found for item $i\n";
+    }
+}
+?&gt;
+</code></pre></details>
+</details>
+<br/>
+
+## 4. 二次开发
+如果印章文本识别产线提供的默认模型权重在您的场景中，精度或速度不满意，您可以尝试利用<b>您自己拥有的特定领域或应用场景的数据</b>对现有模型进行进一步的<b>微调</b>，以提升印章文本识别产线的在您的场景中的识别效果。
+
+### 4.1 模型微调
+由于印章文本识别产线包含若干模块，模型产线的效果如果不及预期，可能来自于其中任何一个模块。您可以对识别效果差的图片进行分析，进而确定是哪个模块存在问题，并参考以下表格中对应的微调教程链接进行模型微调。
+
+<table>
+<thead>
+<tr>
+<th>情形</th>
+<th>微调模块</th>
+<th>微调参考链接</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>印章位置检测不准或未检出</td>
+<td>版面检测模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/layout_detection.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>印章文本存在漏检</td>
+<td>印章文本检测模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/seal_text_detection.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>文本内容不准</td>
+<td>文本识别模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/text_recognition.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>整图旋转矫正不准</td>
+<td>文档图像方向分类模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>图像扭曲矫正不准</td>
+<td>文本图像矫正模块</td>
+<td>暂不支持微调</td>
+</tr>
+</tbody>
+</table>
+
+### 4.2 模型应用
+
+当您使用私有数据集完成微调训练后，可获得本地模型权重文件，然后可以通过参数指定本地模型保存路径的方式，或者通过自定义产线配置文件的方式，使用微调后的模型权重。
+
+#### 4.2.1 通过参数指定本地模型路径
+
+在初始化产线对象时，通过参数指定本地模型路径。以文本检测模型微调后的权重的使用方法为例，示例如下：
+
+命令行方式:
+
+```bash
+# 通过 --doc_orientation_classify_model_dir 指定本地模型路径
+paddleocr seal_recognition -i ./seal_text_det.png --doc_orientation_classify_model_dir your_orientation_classify_model_path
+
+# 默认使用 PP-LCNet_x1_0_doc_ori 模型作为默认文本检测模型，如果微调的不是该模型，通过 --text_detection_model_name 修改模型名称
+paddleocr seal_recognition -i ./seal_text_det.png --doc_orientation_classify_model_name PP-LCNet_x1_0_doc_ori --doc_orientation_classify_model_dir your_orientation_classify_model_path
+```
+
+脚本方式：
+
+```python
+
+from paddleocr import SealRecognition
+
+# 通过 doc_orientation_classify_model_dir 指定本地模型路径
+pipeline = SealRecognition(doc_orientation_classify_model_dir ="./your_orientation_classify_model_path")
+
+# 默认使用 PP-LCNet_x1_0_doc_ori 模型作为默认文本检测模型，如果微调的不是该模型，通过 doc_orientation_classify_model_name 修改模型名称
+# pipeline = SealRecognition(doc_orientation_classify_model_name="PP-LCNet_x1_0_doc_ori", doc_orientation_classify_model_dir="./your_orientation_classify_model_path")
+
+```
+
+
+#### 4.2.2 通过配置文件指定本地模型路径
+
+
+1. 获取产线配置文件
+
+可调用 PaddleOCR 中 通用OCR 产线对象的 `export_paddlex_config_to_yaml` 方法，将当前产线配置导出为 YAML 文件：
+
+```Python
+from paddleocr import SealRecognition
+
+pipeline = SealRecognition()
+pipeline.export_paddlex_config_to_yaml("SealRecognition.yaml")
+```
+
+2. 修改配置文件
+
+在得到默认的产线配置文件后，将微调后模型权重的本地路径替换至产线配置文件中的对应位置即可。例如
+
+```yaml
+......
+SubPipelines:
+  DocPreprocessor:
+    SubModules:
+      DocOrientationClassify:
+        model_dir: null  # 替换为微调后的文档方向分类模型权重路径
+        model_name: PP-LCNet_x1_0_doc_ori # 如果微调的模型名称与默认模型名称不同，请一并修改此处
+        module_name: doc_text_orientation
+      DocUnwarping:
+        model_dir: null  # 替换为微调后的文档矫正模型权重路径
+        model_name: UVDoc # 如果微调的模型名称与默认模型名称不同，请一并修改此处
+        module_name: image_unwarping
+    pipeline_name: doc_preprocessor
+    use_doc_orientation_classify: true
+    use_doc_unwarping: true
+......
+```
+
+在产线配置文件中，不仅包含 SealRecognition CLI 和 Python API 支持的参数，还可进行更多高级配置，具体信息可在 [PaddleX模型产线使用概览](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/pipeline_develop_guide.html) 中找到对应的产线使用教程，参考其中的详细说明，根据需求调整各项配置。
+
+3. 在 CLI 中加载产线配置文件
+
+在修改完成配置文件后，通过命令行的 `--paddlex_config` 参数指定修改后的产线配置文件的路径，PaddleOCR 会读取其中的内容作为产线配置。示例如下：
+
+```bash
+paddleocr seal_recognition --paddlex_config SealRecognition.yaml ...
+```
+
+4. 在 Python API 中加载产线配置文件
+
+初始化产线对象时，可通过 `paddlex_config` 参数传入 PaddleX 产线配置文件路径或配置dict，PaddleOCR 会读取其中的内容作为产线配置。示例如下：
+
+```python
+from paddleocr import SealRecognition
+
+pipeline = SealRecognition(paddlex_config="SealRecognition.yaml")
+```
diff --git a/docs/version3.x/pipeline_usage/table_recognition_v2.en.md b/docs/version3.x/pipeline_usage/table_recognition_v2.en.md
new file mode 100644
index 0000000..97f44e4
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/table_recognition_v2.en.md
@@ -0,0 +1,2923 @@
+---
+comments: true
+---
+
+# General Table Recognition V2 Pipeline Usage Tutorial
+
+## 1. Introduction to General Table Recognition v2 pipeline
+
+Table recognition is a technology that automatically identifies and extracts table content and its structure from documents or images. It is widely used in fields such as data entry, information retrieval, and document analysis. By using computer vision and machine learning algorithms, table recognition can convert complex table information into an editable format, making it easier for users to further process and analyze data.
+
+The General Table Recognition v2 pipeline (PP-TableMagic) is designed to tackle table recognition tasks, identifying tables in images and outputting them in HTML format. Unlike the original General Table Recognition pipeline, this version introduces two new modules: table classification and table cell detection. By adopting a <b>multi-model pipeline combining "table classification + table structure recognition + cell detection"</b>, it achieves better end-to-end table recognition performance compared to the previous version. Based on this, the General Table Recognition v2 pipeline <b>natively supports targeted model fine-tuning</b>, allowing developers to customize it to varying degrees for satisfactory performance in different application scenarios. <b>Furthermore, the General Table Recognition v2 pipeline also supports end-to-end table structure recognition models (e.g., SLANet, SLANet_plus, etc.) and allows independent configuration for wired and wireless table recognition methods, enabling developers to freely select and combine the best table recognition solutions.</b>
+
+This pipeline is applicable in a variety of fields, including general, manufacturing, finance, and transportation. It also provides flexible service deployment options, supporting multiple programming languages on various hardware. Additionally, it offers capabilities for secondary development, allowing you to train and fine-tune your own datasets based on this pipeline, with the trained models seamlessly integrated.
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/table_recognition_v2/01.png"/>
+
+<b>The General Table Recognition Pipeline v2 includes the following 8 modules. Each module can be trained and inferred independently and contains multiple models. For detailed information, please click on the corresponding module to view the documentation.</b>
+
+- [Table Structure Recognition Module](../module_usage/table_structure_recognition.md)
+- [Table Classification Module](../module_usage/table_classification.md)
+- [Table Cell Detection Module](../module_usage/table_cells_detection.md)
+- [Text Detection Module](../module_usage/text_detection.md)
+- [Text Recognition Module](../module_usage/text_recognition.md)
+- [Layout Region Detection Module](../module_usage/layout_detection.md) (optional)
+- [Document Image Orientation Classification Module](../module_usage/doc_img_orientation_classification.md) (optional)
+- [Text Image Unwarping Module](../module_usage/text_image_unwarping.md) (optional)
+
+In this pipeline, you can choose the models to use based on the benchmark data below.
+
+> The inference time only includes the model inference time and does not include the time for pre- or post-processing.
+> In the inference time columns labeled [Regular Mode / High-Performance Mode], the Regular Mode values correspond to the local `paddle_static` inference engine.
+
+<details>
+<summary> <b>Table Structure Recognition Module Models:</b></summary>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>SLANet</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANet_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_pretrained.pdparams">Training Model</a></td>
+<td>59.52</td>
+<td>23.96 / 21.75</td>
+<td>- / 43.12</td>
+<td>6.9</td>
+<td rowspan="1">SLANet is a table structure recognition model developed by Baidu PaddlePaddle's vision team. This model significantly improves the accuracy and inference speed of table structure recognition by using a CPU-friendly lightweight backbone network PP-LCNet, a high-low feature fusion module CSP-PAN, and a feature decoding module SLA Head that aligns structure and location information.</td>
+</tr>
+<tr>
+<td>SLANet_plus</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANet_plus_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_plus_pretrained.pdparams">Training Model</a></td>
+<td>63.69</td>
+<td>23.43 / 22.16</td>
+<td>- / 41.80</td>
+<td>6.9</td>
+<td rowspan="1">SLANet_plus is an enhanced version of the SLANet table structure recognition model developed by Baidu PaddlePaddle's vision team. Compared to SLANet, SLANet_plus significantly improves the recognition capabilities for wireless tables and complex tables, while reducing the model's sensitivity to table positioning accuracy, allowing for accurate recognition even if the table is slightly misaligned.</td>
+</tr>
+<tr>
+<td>SLANeXt_wired</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wired_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wired_pretrained.pdparams">Training Model</a></td>
+<td rowspan="2">69.65</td>
+<td rowspan="2">85.92 / 85.92</td>
+<td rowspan="2">- / 501.66</td>
+<td rowspan="2">351</td>
+<td rowspan="2">The SLANeXt series is a new generation of table structure recognition models developed by Baidu PaddlePaddle's vision team. Compared to SLANet and SLANet_plus, SLANeXt focuses on recognizing table structures and has been trained with dedicated weights for recognizing wired and wireless tables, significantly enhancing recognition capabilities across both types, especially for wired tables.</td>
+</tr>
+<tr>
+<td>SLANeXt_wireless</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wireless_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wireless_pretrained.pdparams">Training Model</a></td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary> <b>Table Classification Module Models:</b></summary>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Top1 Acc (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+</tr>
+<tr>
+<td>PP-LCNet_x1_0_table_cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/CLIP_vit_base_patch16_224_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_table_cls_pretrained.pdparams">Training Model</a></td>
+<td>94.2</td>
+<td>2.62 / 0.60</td>
+<td>3.17 / 1.14</td>
+<td>6.6</td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary> <b>Table Cell Detection Module Models:</b></summary>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>RT-DETR-L_wired_table_cell_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wired_table_cell_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wired_table_cell_det_pretrained.pdparams">Training Model</a></td>
+<td rowspan="2">82.7</td>
+<td rowspan="2">33.47 / 27.02</td>
+<td rowspan="2">402.55 / 256.56</td>
+<td rowspan="2">124</td>
+<td rowspan="2">RT-DETR is the first real-time end-to-end object detection model. The Baidu PaddlePaddle vision team based RT-DETR-L as the base model, completing pre-training on a self-built table cell detection dataset, achieving good performance in detecting both wired and wireless table cells.</td>
+</tr>
+<tr>
+<td>RT-DETR-L_wireless_table_cell_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wireless_table_cell_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wireless_table_cell_det_pretrained.pdparams">Training Model</a></td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary> <b>Text Detection Module Models:</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Detection Hmean (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+<tr>
+<td>PP-OCRv5_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_det_pretrained.pdparams">Training Model</a></td>
+<td>83.8</td>
+<td>89.55 / 70.19</td>
+<td>383.15 / 383.15</td>
+<td>84.3</td>
+<td>PP-OCRv5 server-side text detection model with higher accuracy, suitable for deployment on high-performance servers</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_det_pretrained.pdparams">Training Model</a></td>
+<td>79.0</td>
+<td>10.67 / 6.36</td>
+<td>57.77 / 28.15</td>
+<td>4.7</td>
+<td>PP-OCRv5 mobile-side text detection model with higher efficiency, suitable for deployment on edge devices</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_det_pretrained.pdparams">Training Model</a></td>
+<td>69.2</td>
+<td>127.82 / 98.87</td>
+<td>585.95 / 489.77</td>
+<td>109</td>
+<td>PP-OCRv4 server-side text detection model with higher accuracy, suitable for deployment on high-performance servers</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams">Training Model</a></td>
+<td>63.8</td>
+<td>9.87 / 4.17</td>
+<td>56.60 / 20.79</td>
+<td>4.7</td>
+<td>PP-OCRv4 mobile-side text detection model with higher efficiency, suitable for deployment on edge devices</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>Text Recognition Module:</b></summary>
+<table>
+<tr>
+<th>Model</th><th>Model Download Links</th>
+<th>Recognition Avg Accuracy(%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>86.38</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec is a next-generation text recognition model. It aims to efficiently and accurately support the recognition of four major languages—Simplified Chinese, Traditional Chinese, English, and Japanese—as well as complex text scenarios such as handwriting, vertical text, pinyin, and rare characters using a single model. While maintaining recognition performance, it balances inference speed and model robustness, providing efficient and accurate technical support for document understanding in various scenarios.</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>81.29</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">Pretrained Model</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc is trained on a mixed dataset of more Chinese document data and PP-OCR training data, building upon PP-OCRv4_server_rec. It enhances the recognition capabilities for some Traditional Chinese characters, Japanese characters, and special symbols, supporting over 15,000 characters. In addition to improving document-related text recognition, it also enhances general text recognition capabilities.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>A lightweight recognition model of PP-OCRv4 with high inference efficiency, suitable for deployment on various hardware devices, including edge devices.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>The server-side model of PP-OCRv4, offering high inference accuracy and deployable on various servers.</td>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>An ultra-lightweight English recognition model trained based on the PP-OCRv4 recognition model, supporting English and numeric character recognition.</td>
+</tr>
+</table>
+
+> ❗ The above section lists the **6 core models** that are primarily supported by the text recognition module. In total, the module supports **20 comprehensive models**, including multiple multilingual text recognition models. Below is the complete list of models:
+
+<details><summary> 👉Details of the Model List</summary>
+
+* <b>PP-OCRv5 Multi-Scenario Models</b>
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Links</th>
+<th>Avg Accuracy for Chinese Recognition (%)</th>
+<th>Avg Accuracy for English Recognition (%)</th>
+<th>Avg Accuracy for Traditional Chinese Recognition (%)</th>
+<th>Avg Accuracy for Japanese Recognition (%)</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Introduction</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>86.38</td>
+<td>64.70</td>
+<td>93.29</td>
+<td>60.35</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec is a next-generation text recognition model. It aims to efficiently and accurately support the recognition of four major languages—Simplified Chinese, Traditional Chinese, English, and Japanese—as well as complex text scenarios such as handwriting, vertical text, pinyin, and rare characters using a single model. While maintaining recognition performance, it balances inference speed and model robustness, providing efficient and accurate technical support for document understanding in various scenarios.</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">Pretrained Model</a></td>
+<td>81.29</td>
+<td>66.00</td>
+<td>83.55</td>
+<td>54.65</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+</table>
+
+* <b>Chinese Recognition Models</b>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Recognition Avg Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">Training Model</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc is based on PP-OCRv4_server_rec, trained with a mix of more Chinese document data and PP-OCR training data, increasing the recognition capabilities for some Traditional Chinese, Japanese, and special characters, supporting recognition of over 15,000 characters. In addition to improving the document-related text recognition capabilities, it also enhances general text recognition capabilities.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>PP-OCRv4's lightweight recognition model has high inference efficiency and can be deployed on various hardware, including edge devices.</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">Training Model</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>PP-OCRv4's server-side model has high inference accuracy and can be deployed on various servers.</td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>72.96</td>
+<td>3.89 / 1.16</td>
+<td>8.72 / 3.56</td>
+<td>10.3</td>
+<td>PP-OCRv3's lightweight recognition model has high inference efficiency and can be deployed on various hardware, including edge devices.</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Recognition Avg Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>ch_SVTRv2_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_SVTRv2_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_SVTRv2_rec_pretrained.pdparams">Training Model</a></td>
+<td>68.81</td>
+<td>10.38 / 8.31</td>
+<td>66.52 / 30.83</td>
+<td>80.5</td>
+<td rowspan="1">SVTRv2 is a server-side text recognition model developed by the OpenOCR team at Fudan University's Vision and Learning Laboratory (FVL). It won first place in the PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition Task, achieving a 6% improvement in end-to-end recognition accuracy compared to PP-OCRv4.</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Recognition Avg Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>ch_RepSVTR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_RepSVTR_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_RepSVTR_rec_pretrained.pdparams">Training Model</a></td>
+<td>65.07</td>
+<td>6.29 / 1.57</td>
+<td>20.64 / 5.40</td>
+<td>48.8</td>
+<td rowspan="1">RepSVTR is a mobile text recognition model based on SVTRv2, which won first place in the PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition Task, achieving a 2.5% improvement in end-to-end recognition accuracy compared to PP-OCRv4, while maintaining the same inference speed.</td>
+</tr>
+</table>
+
+* <b>English Recognition Models</b>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Recognition Avg Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td> 70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>This ultra-lightweight English recognition model is trained based on the PP-OCRv4 recognition model, supporting English and digit recognition.</td>
+</tr>
+<tr>
+<td>en_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>70.69</td>
+<td>3.56 / 0.78</td>
+<td>8.44 / 5.78</td>
+<td>17.3</td>
+<td>This ultra-lightweight English recognition model is trained based on the PP-OCRv3 recognition model, supporting English and digit recognition.</td>
+</tr>
+</table>
+
+* <b>Multilingual Recognition Models</b>
+<table>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Recognition Avg Accuracy (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+<tr>
+<td>korean_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>60.21</td>
+<td>3.73 / 0.98</td>
+<td>8.76 / 2.91</td>
+<td>9.6</td>
+<td>This ultra-lightweight Korean recognition model is trained based on the PP-OCRv3 recognition model, supporting Korean and digit recognition.</td>
+</tr>
+<tr>
+<td>japan_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+japan_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/japan_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>45.69</td>
+<td>3.86 / 1.01</td>
+<td>8.62 / 2.92</td>
+<td>9.8</td>
+<td>This ultra-lightweight Japanese recognition model is trained based on the PP-OCRv3 recognition model, supporting Japanese and digit recognition.</td>
+</tr>
+<tr>
+<td>chinese_cht_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+chinese_cht_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/chinese_cht_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>82.06</td>
+<td>3.90 / 1.16</td>
+<td>9.24 / 3.18</td>
+<td>10.8</td>
+<td>This ultra-lightweight Traditional Chinese recognition model is trained based on the PP-OCRv3 recognition model, supporting Traditional Chinese and digit recognition.</td>
+</tr>
+<tr>
+<td>te_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+te_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>95.88</td>
+<td>3.59 / 0.81</td>
+<td>8.28 / 6.21</td>
+<td>85 M </td>
+<td>This ultra-lightweight Telugu recognition model is trained based on the PP-OCRv3 recognition model, supporting Telugu and digit recognition.</td>
+</tr>
+<tr>
+<td>ka_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ka_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ka_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>96.96</td>
+<td>3.49 / 0.89</td>
+<td>8.63 / 2.77</td>
+<td>17.4</td>
+<td>This ultra-lightweight Kannada recognition model is trained based on the PP-OCRv3 recognition model, supporting Kannada and digit recognition.</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ta_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>76.83</td>
+<td>3.49 / 0.86</td>
+<td>8.35 / 3.41</td>
+<td>8.7</td>
+<td>This ultra-lightweight Tamil recognition model is trained based on the PP-OCRv3 recognition model, supporting Tamil and digit recognition.</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>76.93</td>
+<td>3.53 / 0.78</td>
+<td>8.50 / 6.83</td>
+<td>8.7</td>
+<td>This ultra-lightweight Latin recognition model is trained based on the PP-OCRv3 recognition model, supporting Latin and digit recognition.</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+arabic_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>73.55</td>
+<td>3.60 / 0.83</td>
+<td>8.44 / 4.69</td>
+<td>17.3</td>
+<td>This ultra-lightweight Arabic alphabet recognition model is trained based on the PP-OCRv3 recognition model, supporting Arabic letters and digit recognition.</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+cyrillic_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>94.28</td>
+<td>3.56 / 0.79</td>
+<td>8.22 / 2.76</td>
+<td>8.7</td>
+<td>This ultra-lightweight Slavic alphabet recognition model is trained based on the PP-OCRv3 recognition model, supporting Slavic letters and digit recognition.</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+devanagari_PP-OCRv3_mobile_rec_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv3_mobile_rec_pretrained.pdparams">Training Model</a></td>
+<td>96.44</td>
+<td>3.60 / 0.78</td>
+<td>6.95 / 2.87</td>
+<td>8.7</td>
+<td>This ultra-lightweight Devanagari alphabet recognition model is trained based on the PP-OCRv3 recognition model, supporting Devanagari letters and digit recognition.</td>
+</tr>
+</table>
+</details>
+</details>
+
+<details>
+<summary> <b>Layout Region Detection Module Models:</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout_plus-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout_plus-L_pretrained.pdparams">Training Model</a></td>
+<td>83.2</td>
+<td>53.03 / 17.23</td>
+<td>634.62 / 378.32</td>
+<td>126.01</td>
+<td>This higher-precision layout region localization model is trained based on RT-DETR-L on a self-built dataset that includes Chinese and English papers, multi-column magazines, newspapers, PPTs, contracts, books, exam papers, research reports, ancient texts, Japanese documents, and vertical text documents.</td>
+</tr>
+<tr>
+<tr>
+<td>PP-DocLayout-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-L_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-L_pretrained.pdparams">Training Model</a></td>
+<td>90.4</td>
+<td>33.59 / 33.59</td>
+<td>503.01 / 251.08</td>
+<td>123.76</td>
+<td>This high-precision layout region localization model is trained based on RT-DETR-L on a self-built dataset that includes Chinese and English papers, magazines, contracts, books, exam papers, and research reports.</td>
+</tr>
+<tr>
+<td>PP-DocLayout-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-M_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-M_pretrained.pdparams">Training Model</a></td>
+<td>75.2</td>
+<td>13.03 / 4.72</td>
+<td>43.39 / 24.44</td>
+<td>22.578</td>
+<td>This layout region localization model balances accuracy and efficiency, trained based on PicoDet-L on a self-built dataset that includes Chinese and English papers, magazines, contracts, books, exam papers, and research reports.</td>
+</tr>
+<tr>
+<td>PP-DocLayout-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-S_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-S_pretrained.pdparams">Training Model</a></td>
+<td>70.9</td>
+<td>11.54 / 3.86</td>
+<td>18.53 / 6.29</td>
+<td>4.834</td>
+<td>This highly efficient layout region localization model is trained based on PicoDet-S on a self-built dataset that includes Chinese and English papers, magazines, contracts, books, exam papers, and research reports.</td>
+</tr>
+</tbody>
+</table>
+
+> ❗ The above lists the <b>4 core models</b> that are key to the layout detection module. The module supports a total of <b>12 complete models</b>, including multiple pre-defined models for different categories. The complete model list is as follows:
+<details><summary> 👉 Model List Details</summary>
+* <b>Table Layout Detection Models</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet_layout_1x_table</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_table_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_table_pretrained.pdparams">Training Model</a></td>
+<td>97.5</td>
+<td>9.57 / 6.63</td>
+<td>27.66 / 16.75</td>
+<td>7.4</td>
+<td>This high-efficiency layout region localization model is trained based on PicoDet-1x on a self-built dataset, capable of locating tables as one type of region.</td>
+</tr>
+</tbody></table>
+
+* <b>3-Class Layout Detection Models, Including Tables, Images, and Stamps</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_3cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_3cls_pretrained.pdparams">Training Model</a></td>
+<td>88.2</td>
+<td>8.43 / 3.44</td>
+<td>17.60 / 6.51</td>
+<td>4.8</td>
+<td>This high-efficiency layout region localization model is trained using a self-built dataset of Chinese and English papers, magazines, and research reports based on the lightweight model PicoDet-S.</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_3cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_3cls_pretrained.pdparams">Training Model</a></td>
+<td>89.0</td>
+<td>12.80 / 9.57</td>
+<td>45.04 / 23.86</td>
+<td>22.6</td>
+<td>This layout region localization model balances efficiency and accuracy, trained using a self-built dataset of Chinese and English papers, magazines, and research reports based on the model PicoDet-L.</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_3cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_3cls_pretrained.pdparams">Training Model</a></td>
+<td>95.8</td>
+<td>114.80 / 25.65</td>
+<td>924.38 / 924.38</td>
+<td>470.1</td>
+<td>This high-precision layout region localization model is trained using a self-built dataset of Chinese and English papers, magazines, and research reports based on the model RT-DETR-H.</td>
+</tr>
+</tbody>
+</table>
+
+* <b>5-Class English Document Region Detection Models, Including Text, Titles, Tables, Images, and Lists</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet_layout_1x</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_pretrained.pdparams">Training Model</a></td>
+<td>97.8</td>
+<td>9.62 / 6.75</td>
+<td>26.96 / 12.77</td>
+<td>7.4</td>
+<td>This high-efficiency English document layout region localization model is trained on the PubLayNet dataset.</td>
+</tr>
+</tbody></table>
+</b>
+
+* <b>17-Class Region Detection Models, Including 17 Common Layout Categories: Title, Image, Text, Number, Abstract, Content, Chart Title, Formula, Table, Table Title, References, Document Title, Footnote, Header, Algorithm, Footer, and Stamp</b>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>mAP(0.5) (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>87.4</td>
+<td>8.80 / 3.62</td>
+<td>17.51 / 6.35</td>
+<td>4.8</td>
+<td>This high-efficiency layout region localization model is trained using a self-built dataset of Chinese and English papers, magazines, and research reports based on the lightweight model PicoDet-S.</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>89.0</td>
+<td>12.60 / 10.27</td>
+<td>43.70 / 24.42</td>
+<td>22.6</td>
+<td>This layout region localization model balances efficiency and accuracy, trained using a self-built dataset of Chinese and English papers, magazines, and research reports based on the model PicoDet-L.</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_17cls_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_17cls_pretrained.pdparams">Training Model</a></td>
+<td>98.3</td>
+<td>115.29 / 101.18</td>
+<td>964.75 / 964.75</td>
+<td>470.2</td>
+<td>This high-precision layout region localization model is trained using a self-built dataset of Chinese and English papers, magazines, and research reports based on the model RT-DETR-H.</td>
+</tr>
+</tbody>
+</table>
+</details>
+</details>
+
+<details>
+<summary> <b>Text Image Unwarping Module Models (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>CER</th>
+<th>GPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Normal Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">Training Model</a></td>
+<td>0.179</td>
+<td>19.05 / 19.05</td>
+<td>- / 869.82</td>
+<td>30.3</td>
+<td>High-precision text image correction model.</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary> <b>Document Image Orientation Classification Module Models (Optional):</b></summary>
+<table>
+<thead>
+<tr>
+<th>Model</th><th>Model Download Link</th>
+<th>Top-1 Acc (%)</th>
+<th>GPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>CPU Inference Time (ms)<br/>[Regular Mode / High-Performance Mode]</th>
+<th>Model Storage Size (MB)</th>
+<th>Description</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">Inference Model</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">Training Model</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>Based on PP-LCNet_x1_0, this document image classification model includes four categories: 0 degrees, 90 degrees, 180 degrees, and 270 degrees.</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary> <b>Testing Environment Information:</b></summary>
+
+  <ul>
+      <li><b>Performance Testing Environment</b>
+          <ul>
+            <li><strong>Test Dataset:
+             </strong>
+                <ul>
+                  <li>Document Image Orientation Classification Model: A self-built dataset by PaddleX, covering multiple scenarios including identification documents and documents, containing 1,000 images.</li>
+                  <li>Layout Region Detection Model: A self-built layout region detection dataset by PaddleOCR, containing 500 common document type images, including Chinese and English papers, magazines, contracts, books, exam papers, and research reports.</li>
+                  <li>Table Layout Detection Model: A self-built layout table region detection dataset by PaddleOCR, containing 7,835 images with tables in Chinese and English papers.</li>
+                  <li>3-Class Layout Detection Model: A self-built layout region detection dataset by PaddleOCR, containing 1,154 common document type images including Chinese and English papers, magazines, and research reports.</li>
+                  <li>5-Class English Document Region Detection Model: The evaluation dataset from <a href="https://developer.ibm.com/exchanges/data/all/publaynet">PubLayNet</a>, containing 11,245 images of English documents.</li>
+                  <li>17-Class Region Detection Model: A self-built layout region detection dataset by PaddleOCR, containing 892 common document type images including Chinese and English papers, magazines, and research reports.</li>
+                  <li>Table Structure Recognition Model: A high-difficulty Chinese table recognition dataset built internally by PaddleX.</li>
+                  <li>Table Cell Detection Model: An evaluation set built internally by PaddleX.</li>
+                  <li>Table Classification Model: An evaluation set built internally by PaddleX.</li>
+                  <li>Text Detection Model: A Chinese dataset built by PaddleOCR, covering multiple scenarios including street scenes, web images, documents, and handwriting, with 500 images for detection.</li>
+                  <li>Chinese Recognition Model: A Chinese dataset built by PaddleOCR, covering multiple scenarios including street scenes, web images, documents, and handwriting, with 11,000 images for text recognition.</li>
+                  <li>ch_SVTRv2_rec: The evaluation set from <a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition Task</a> A-list.</li>
+                  <li>ch_RepSVTR_rec: The evaluation set from <a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR Algorithm Model Challenge - Task 1: OCR End-to-End Recognition Task</a> B-list.</li>
+                  <li>English Recognition Model: An English dataset built internally by PaddleX.</li>
+                  <li>Multilingual Recognition Model: A multilingual dataset built internally by PaddleX.</li>
+                </ul>
+             </li>
+              <li><strong>Hardware Configuration:</strong>
+                  <ul>
+                      <li>GPU: NVIDIA Tesla T4</li>
+                      <li>CPU: Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>Software Environment:</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>Inference Mode Information</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>Mode</th>
+            <th>GPU Configuration</th>
+            <th>CPU Configuration</th>
+            <th>Acceleration Technology Combination</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>Regular Mode</td>
+            <td>FP32 Precision / No TRT Acceleration</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td><code>paddle_static</code></td>
+        </tr>
+        <tr>
+            <td>High-Performance Mode</td>
+            <td>Optimal combination of prior precision type and acceleration strategy</td>
+            <td>FP32 Precision / 8 Threads</td>
+            <td>Optimal backend (Paddle/OpenVINO/TRT, etc.) selected based on prior knowledge</td>
+        </tr>
+    </tbody>
+</table>
+
+</details>
+</details>
+
+<br />
+<b>If you prioritize model accuracy, please choose models with higher accuracy; if you care more about inference speed, please select models with faster inference speeds; if you focus on model storage size, please choose models with smaller storage volumes.</b>
+
+## 2. Quick Start
+
+Before using the table structure recognition V2 pipeline locally, please ensure that you have completed the installation of the wheel package according to the [installation guide](../installation.md). If you prefer to install dependencies selectively, please refer to the relevant instructions in the installation documentation. The corresponding dependency group for this pipeline is <code>doc-parser</code>. After installation, you can experience it locally using the command line or Python integration.
+
+Please note: If you encounter issues such as the program becoming unresponsive, unexpected program termination, running out of memory resources, or extremely slow inference during execution, please try adjusting the configuration according to the documentation, such as disabling unnecessary features or using lighter-weight models.
+
+### 2.1 Command Line Experience
+
+A single command allows you to quickly experience the effects of the table_recognition_v2 pipeline:
+
+```bash
+paddleocr table_recognition_v2 -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition_v2.jpg
+
+# Specify whether to use the document orientation classification model with --use_doc_orientation_classify
+paddleocr table_recognition_v2 -i ./table_recognition_v2.jpg --use_doc_orientation_classify True
+
+# Specify whether to use the text image unwarping module with --use_doc_unwarping
+paddleocr table_recognition_v2 -i ./table_recognition_v2.jpg --use_doc_unwarping True
+
+# Specify the device to use GPU for model inference with --device
+paddleocr table_recognition_v2 -i ./table_recognition_v2.jpg --device gpu
+```
+
+The examples above use the local `paddle_static` inference engine by default. To run them, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following command:
+
+```bash
+# Use the transformers engine for inference
+paddleocr table_recognition_v2 -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition_v2.jpg \
+    --engine transformers
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+<details><summary><b>More command line parameters are supported. Click to expand for detailed descriptions of the command line parameters</b></summary>
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted, required.<br/>
+<b>Description:</b>
+Local path to image files or PDF files: <code>/root/data/img.jpg</code>; <b>as URL links</b>, such as network URLs for image files or PDF files: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition_v2.jpg">example</a>; <b>as local directories</b>, the directory must contain images to be predicted, such as local path: <code>/root/data/</code> (currently, predictions do not support directories that contain PDF files; the PDF file must be specified to the specific file path).
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td><b>Meaning:</b>Specify the path to save the inference result file. <br/>
+<b>Description:</b>
+If not set, the inference result will not be saved locally.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the layout detection model.<br/>
+<b>Description:</b> 
+If not set, the default model of the pipeline will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td>Directory path of the layout detection model. If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_classification_model_name</code></td>
+<td><b>Meaning:</b>Name of the table classification model. <br/>
+<b>Description:</b> 
+If not set, the default model of the pipeline will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_classification_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the table classification model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the wired table structure recognition model.<br/>
+<b>Description:</b> 
+If not set, the default model of the pipeline will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the wired table structure recognition model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the wireless table structure recognition model. <br/>
+<b>Description:</b> 
+If not set, the default model of the pipeline will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the wireless table structure recognition model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the wired table cell detection model. <br/>
+<b>Description:</b> 
+If not set, the default model of the pipeline will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the wired table cell detection model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the wireless table cell detection model. <br/>
+<b>Description:</b> 
+If not set, the default model of the pipeline will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the wireless table cell detection model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>Meaning:</b>Name of the document orientation classification model. <br/>
+<b>Description:</b> 
+If not set, the default model of the pipeline will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the document orientation classification model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>Meaning:</b>Name of the text image unwarping model. <br/>
+<b>Description:</b> 
+If not set, the default model of the pipeline will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text image unwarping model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the text detection model. <br/>
+<b>Description:</b> 
+If not set, the default model of the pipeline will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text detection model. <br/>
+<b>Description:</b> 
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limit for text detection.<br/>
+<b>Description:</b>
+Any integer greater than <code>0</code>. If not set, the value initialized by the pipeline will be used, which defaults to <code>960</code>.
+</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>Meaning:</b>Type of the image side length limit for text detection.<br/>
+<b>Description:</b>
+Supports <code>min</code> and <code>max</code>. <code>min</code> ensures that the shortest side of the image is not less than <code>det_limit_side_len</code>, while <code>max</code> ensures that the longest side of the image is not greater than <code>limit_side_len</code>. If not set, the value initialized by the pipeline will be used, which defaults to <code>max</code>.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>Meaning:</b>Detection pixel threshold. In the output probability map, only pixels with a score greater than this threshold will be considered text pixels.<br/>
+<b>Description:</b>
+Any floating-point number greater than <code>0</code>. If not set, the value initialized by the pipeline will be used, which defaults to <code>0.3</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>Meaning:</b>Detection box threshold. When the average score of all pixels within the detection result box is greater than this threshold, the result is considered a text area.<br/>
+<b>Description:</b>
+Any floating-point number greater than <code>0</code>. If not set, the value initialized by the pipeline will be used, which defaults to <code>0.6</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Text detection expansion coefficient. This method expands the text area; the larger this value, the larger the expanded area.<br/>
+<b>Description:</b>
+Any floating-point number greater than <code>0</code>. If not set, the value initialized by the pipeline will be used, which defaults to <code>2.0</code>.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the text recognition model. <br/>
+<b>Description:</b>
+If not set, the default model of the pipeline will be used.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the text recognition model. <br/>
+<b>Description:</b>
+If not set, the official model will be downloaded.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>Meaning:</b>Batch size for the text recognition model.<br/>
+<b>Description:</b> 
+If not set, the default batch size will be set to <code>1</code>.</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Text recognition threshold. Text results with a score greater than this threshold will be retained.<br/>
+<b>Description:</b>
+Any floating-point number greater than <code>0</code>. If not set, the value initialized by the pipeline will be used, which defaults to <code>0.0</code>. That is, no threshold is set.
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to load and use the document orientation classification module.<br/>
+<b>Description:</b> 
+If not set, the value initialized by the pipeline will be used, which defaults to <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to load and use the text image unwarping module. <br/>
+<b>Description:</b> 
+If not set, the value initialized by the pipeline will be used, which defaults to <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>Meaning:</b>Whether to load and use the layout detection module.<br/>
+<b>Description:</b> 
+If not set, the value initialized by the pipeline will be used, which defaults to <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_ocr_model</code></td>
+<td><b>Meaning:</b>Whether to load and use the OCR module. <br/>
+<b>Description:</b> 
+If not set, the value initialized by the pipeline will be used, which defaults to <code>True</code>.</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>The device used for inference.<br/>
+<b>Description:</b> 
+Supports specifying a specific card number:
+<ul>
+<li><b>CPU</b>: For example, <code>cpu</code> indicates using CPU for inference;</li>
+<li><b>GPU</b>: For example, <code>gpu:0</code> indicates using the first GPU for inference;</li>
+<li><b>NPU</b>: For example, <code>npu:0</code> indicates using the first NPU for inference;</li>
+<li><b>XPU</b>: For example, <code>xpu:0</code> indicates using the first XPU for inference;</li>
+<li><b>MLU</b>: For example, <code>mlu:0</code> indicates using the first MLU for inference;</li>
+<li><b>DCU</b>: For example, <code>dcu:0</code> indicates using the first DCU for inference;</li>
+<li><b>MetaX GPU</b>: For example, <code>metax_gpu:0</code> indicates using the first MetaX GPU for inference;</li>
+<li><b>Iluvatar GPU</b>: For example, <code>iluvatar_gpu:0</code> indicates using the first Iluvatar GPU for inference;</li>
+</ul>If not set, the pipeline initialized value for this parameter will be used. During initialization, the local GPU device 0 will be preferred; if unavailable, the CPU device will be used.
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>Meaning:</b> Inference engine.<br/><b>Description:</b> Supports <code>None</code> (the default), <code>paddle</code>, <code>paddle_static</code>, <code>paddle_dynamic</code>, and <code>transformers</code>. When left as <code>None</code>, PaddleOCR preserves the behavior of earlier versions, which in most configurations is equivalent to <code>paddle</code>. For detailed descriptions, supported values, compatibility rules, and examples, see <a href="../inference_deployment/local_inference/inference_engine.en.md">Inference Engine and Configuration</a>.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b>Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b>Whether to use the Paddle Inference TensorRT subgraph engine. <br/>
+<b>Description:</b>
+If the model does not support acceleration through TensorRT, setting this flag will not enable acceleration.<br/>
+For Paddle with CUDA version 11.8, the compatible TensorRT version is 8.x (x>=6), and it is recommended to install TensorRT 8.6.1.6.<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b>Computation precision, such as <code>fp32</code>, <code>fp16</code>.</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b>Whether to enable MKL-DNN acceleration for inference.<br/>
+<b>Description:</b> 
+If MKL-DNN is unavailable or the model does not support it, acceleration will not be used even if this flag is set.</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b>MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b>Number of threads to use for inference on the CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b> Path to the PaddleX pipeline configuration file.<br/><b>Description:</b> Use this parameter when you need to configure advanced options such as <code>engine_config</code> through a configuration file. See <a href="../paddleocr_and_paddlex.en.md">PaddleOCR and PaddleX</a>.</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+</tbody>
+</table>
+</details>
+<br />
+
+To run inference on the [example image](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition_v2.jpg), you can use the following command:
+
+```bash
+paddleocr table_recognition_v2 -i ./table_recognition_v2.jpg --use_doc_orientation_classify False --use_doc_unwarping False
+```
+
+The running results will be printed to the terminal. The default configuration of the table_recognition_v2 pipeline's running results is as follows:
+
+```
+{'res': {'input_path': 'table_recognition_v2.jpg', 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_layout_detection': True, 'use_ocr_model': True}, 'layout_det_res': {'input_path': None, 'page_index': None, 'boxes': [{'cls_id': 8, 'label': 'table', 'score': 0.86655592918396, 'coordinate': [0.0125130415, 0.41920784, 1281.3737, 585.3884]}]}, 'overall_ocr_res': {'input_path': None, 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_textline_orientation': False}, 'dt_polys': array([[[   9,   21],
+        ...,
+        [   9,   59]],
+
+       ...,
+
+       [[1046,  536],
+        ...,
+        [1046,  573]]], dtype=int16), 'text_det_params': {'limit_side_len': 960, 'limit_type': 'max', 'thresh': 0.3, 'box_thresh': 0.6, 'unclip_ratio': 2.0}, 'text_type': 'general', 'textline_orientation_angles': array([-1, ..., -1]), 'text_rec_score_thresh': 0, 'rec_texts': ['部门', '报销人', '报销事由', '批准人：', '单据', '张', '合计金额', '元', '车费票', '其', '火车费票', '飞机票', '中', '旅住宿费', '其他', '补贴'], 'rec_scores': array([0.99958128, ..., 0.99317062]), 'rec_polys': array([[[   9,   21],
+        ...,
+        [   9,   59]],
+
+       ...,
+
+       [[1046,  536],
+        ...,
+        [1046,  573]]], dtype=int16), 'rec_boxes': array([[   9, ...,   59],
+       ...,
+       [1046, ...,  573]], dtype=int16)}, 'table_res_list': [{'cell_box_list': [array([ 0.13052222, ..., 73.08310249]), array([104.43082511, ...,  73.27777413]), array([319.39041221, ...,  73.30439308]), array([424.2436837 , ...,  73.44736794]), array([580.75836265, ...,  73.24003914]), array([723.04370201, ...,  73.22717598]), array([984.67315757, ...,  73.20420387]), array([1.25130415e-02, ..., 5.85419208e+02]), array([984.37072837, ..., 137.02281502]), array([984.26586998, ..., 201.22290352]), array([984.24017417, ..., 585.30775765]), array([1039.90606773, ...,  265.44664314]), array([1039.69549644, ...,  329.30540779]), array([1039.66546714, ...,  393.57319954]), array([1039.5122689 , ...,  457.74644783]), array([1039.55535972, ...,  521.73030403]), array([1039.58612144, ...,  585.09468392])], 'pred_html': '<html><body><table><tbody><tr><td>部门</td><td></td><td>报销人</td><td></td><td>报销事由</td><td></td><td colspan="2">批准人：</td></tr><tr><td colspan="6" rowspan="8"></td><td colspan="2">单据 张</td></tr><tr><td colspan="2">合计金额 元</td></tr><tr><td rowspan="6">其 中</td><td>车费票</td></tr><tr><td>火车费票</td></tr><tr><td>飞机票</td></tr><tr><td>旅住宿费</td></tr><tr><td>其他</td></tr><tr><td>补贴</td></tr></tbody></table></body></html>', 'table_ocr_pred': {'rec_polys': array([[[   9,   21],
+        ...,
+        [   9,   59]],
+
+       ...,
+
+       [[1046,  536],
+        ...,
+        [1046,  573]]], dtype=int16), 'rec_texts': ['部门', '报销人', '报销事由', '批准人：', '单据', '张', '合计金额', '元', '车费票', '其', '火车费票', '飞机票', '中', '旅住宿费', '其他', '补贴'], 'rec_scores': array([0.99958128, ..., 0.99317062]), 'rec_boxes': array([[   9, ...,   59],
+       ...,
+       [1046, ...,  573]], dtype=int16)}}]}}
+```
+
+The visualization results are saved under <code>save_path</code>, and the visualization results are as follows:
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/table_recognition_v2/02.jpg">
+
+### 2.2 Python Script Integration
+
+The command line method is designed for quick experience and viewing effects. Generally, in a project, it is often necessary to integrate through code. You can complete quick inference of the pipeline with just a few lines of code. The inference code is as follows:
+
+```python
+from paddleocr import TableRecognitionPipelineV2
+
+pipeline = TableRecognitionPipelineV2()
+# ocr = TableRecognitionPipelineV2(use_doc_orientation_classify=True) # Specify whether to use the document orientation classification model with use_doc_orientation_classify
+# ocr = TableRecognitionPipelineV2(use_doc_unwarping=True) # Specify whether to use the text image unwarping module with use_doc_unwarping
+# ocr = TableRecognitionPipelineV2(device="gpu") # Specify the device to use GPU for model inference
+output = pipeline.predict("./table_recognition_v2.jpg")
+for res in output:
+    res.print() ## Print the predicted structured output
+    res.save_to_img("./output/")
+    res.save_to_xlsx("./output/")
+    res.save_to_html("./output/")
+    res.save_to_json("./output/")
+```
+
+The example above uses the local `paddle_static` inference engine by default. To run it, first install PaddlePaddle by following [PaddlePaddle Framework Installation](../paddlepaddle_installation.en.md).
+
+If you choose `transformers` as the inference engine, make sure the Transformers environment is configured by following [Inference Engine and Configuration](../inference_deployment/local_inference/inference_engine.en.md), and then run the following code:
+
+```python
+from paddleocr import TableRecognitionPipelineV2
+
+pipeline = TableRecognitionPipelineV2(
+    engine="transformers",
+)
+# ocr = TableRecognitionPipelineV2(use_doc_orientation_classify=True) # Specify whether to use the document orientation classification model with use_doc_orientation_classify
+# ocr = TableRecognitionPipelineV2(use_doc_unwarping=True) # Specify whether to use the text image unwarping module with use_doc_unwarping
+# ocr = TableRecognitionPipelineV2(device="gpu") # Specify the device to use GPU for model inference
+output = pipeline.predict("./table_recognition_v2.jpg")
+for res in output:
+    res.print() ## Print the predicted structured output
+    res.save_to_img("./output/")
+    res.save_to_xlsx("./output/")
+    res.save_to_html("./output/")
+    res.save_to_json("./output/")
+```
+
+In most scenarios, the default `paddle_static` inference engine delivers better inference performance and is the recommended first choice.
+
+In the above Python script, the following steps are performed:
+
+(1) Instantiate the general table recognition V2 pipeline object using <code>TableRecognitionPipelineV2()</code>. The specific parameter descriptions are as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>Meaning:</b>Name of the layout detection model.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default model of the pipeline will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the layout detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_classification_model_name</code></td>
+<td><b>Meaning:</b>Name of the table classification model.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default model of the pipeline will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_classification_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the table classification model.<br/> 
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the wired table structure recognition model.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default model of the pipeline will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the wired table structure recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the wireless table structure recognition model.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default model of the pipeline will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_dir</code></td>
+<td><b>Meaning:</b>Directory path of the wireless table structure recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_name</code></td>
+<td>Name of the wired table cell detection model. 
+<td><b>Meaning:</b>Name of the wired table cell detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default model of the pipeline will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_dir</code></td>
+<td>Directory path of the wired table cell detection model. 
+<td><b>Meaning:</b>Directory path of the wired table cell detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_name</code></td>
+<td>Name of the wireless table cell detection model. 
+<td><b>Meaning:</b>Name of the wireless table cell detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default model of the pipeline will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_dir</code></td>
+<td>Directory path of the wireless table cell detection model. 
+<td><b>Meaning:</b>Directory path of the wireless table cell detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td>Name of the document orientation classification model. 
+<td><b>Meaning:</b>Name of the document orientation classification model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default model of the pipeline will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td>Directory path of the document orientation classification model. 
+<td><b>Meaning:</b>Directory path of the document orientation classification model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td>Name of the text image unwarping model. 
+<td><b>Meaning:</b>Name of the text image unwarping model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default model of the pipeline will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td>Directory path of the text image unwarping model. 
+<td><b>Meaning:</b>Directory path of the text image unwarping model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td>Name of the text detection model. 
+<td><b>Meaning:</b>Name of the text detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default model of the pipeline will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td>Directory path of the text detection model. 
+<td><b>Meaning:</b>Directory path of the text detection model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Image side length limit for text detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>int</b>: Any integer greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline will be used, which defaults to <code>960</code>.</li>
+</ul>
+</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>Meaning:</b>Type of the image side length limit for text detection. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>str</b>: Supports <code>min</code> and <code>max</code>. <code>min</code> ensures that the shortest side of the image is not less than <code>det_limit_side_len</code>, while <code>max</code> ensures that the longest side of the image is not greater than <code>limit_side_len</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline will be used, which defaults to <code>max</code>.</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>Meaning:</b>Detection pixel threshold. In the output probability map, only pixels with a score greater than this threshold will be considered text pixels. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any floating-point number greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline will be used, which defaults to <code>0.3</code>.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>Meaning:</b>Detection box threshold. When the average score of all pixels within the detection result box is greater than this threshold, the result is considered a text area. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any floating-point number greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline will be used, which defaults to <code>0.6</code>.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Text detection expansion coefficient. This method expands the text area; the larger this value, the larger the expanded area. <br/>
+<b>Description:</b> 
+<ul>
+<li><b>float</b>: Any floating-point number greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline will be used, which defaults to <code>2.0</code>.</li>
+</ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>Meaning:</b>Name of the text recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default model of the pipeline will be used.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td>Directory path of the text recognition model. 
+<td><b>Meaning:</b>Directory path of the text recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the official model will be downloaded.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td>Batch size for the text recognition model. 
+<td><b>Meaning:</b>Batch size for the text recognition model. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the default batch size will be set to <code>1</code>.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Text recognition threshold. Text results with a score greater than this threshold will be retained.<br/>
+<b>Description:</b>
+<ul>
+<li><b>float</b>: Any floating-point number greater than <code>0</code>;</li>
+<li><b>None</b>: If set to <code>None</code>, the value initialized by the pipeline will be used, which defaults to <code>0.0</code>. That is, no threshold is set.
+</ur>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to load and use the document orientation classification module. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the value initialized by the pipeline will be used, which defaults to <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td>Whether to load and use the text image unwarping module. 
+<td><b>Meaning:</b>Whether to load and use the text image unwarping module. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the value initialized by the pipeline will be used, which defaults to <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td>Whether to load and use the layout detection module. 
+<td><b>Meaning:</b>Whether to load and use the layout detection module. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the value initialized by the pipeline will be used, which defaults to <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_ocr_model</code></td>
+<td>Whether to load and use the OCR module. 
+<td><b>Meaning:</b>Whether to load and use the OCR module. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the value initialized by the pipeline will be used, which defaults to <code>True</code>.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>Meaning:</b>The device used for inference. <br/>
+<b>Description:</b> 
+Supports specifying a specific card number:
+<ul>
+<li><b>CPU</b>: For example, <code>cpu</code> indicates using CPU for inference;</li>
+<li><b>GPU</b>: For example, <code>gpu:0</code> indicates using the first GPU for inference;</li>
+<li><b>NPU</b>: For example, <code>npu:0</code> indicates using the first NPU for inference;</li>
+<li><b>XPU</b>: For example, <code>xpu:0</code> indicates using the first XPU for inference;</li>
+<li><b>MLU</b>: For example, <code>mlu:0</code> indicates using the first MLU for inference;</li>
+<li><b>DCU</b>: For example, <code>dcu:0</code> indicates using the first DCU for inference;</li>
+<li><b>MetaX GPU</b>: For example, <code>metax_gpu:0</code> indicates using the first MetaX GPU for inference;</li>
+<li><b>Iluvatar GPU</b>: For example, <code>iluvatar_gpu:0</code> indicates using the first Iluvatar GPU for inference;</li>
+<li><b>None</b>: If set to <code>None</code>, the pipeline initialized value for this parameter will be used. During initialization, the local GPU device 0 will be preferred; if unavailable, the CPU device will be used.</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>Meaning:</b>Whether to enable high-performance inference.</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>Meaning:</b>Whether to use the Paddle Inference TensorRT subgraph engine. If the model does not support acceleration through TensorRT, setting this flag will not enable acceleration.<br/>
+<b>Description:</b>
+For Paddle with CUDA version 11.8, the compatible TensorRT version is 8.x (x>=6), and it is recommended to install TensorRT 8.6.1.6.<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>Meaning:</b>Computation precision, such as <code>"fp32"</code>, <code>"fp16"</code>.</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>Meaning:</b>Whether to enable MKL-DNN acceleration for inference.<br/>
+<b>Description:</b> 
+If MKL-DNN is unavailable or the model does not support it, acceleration will not be used even if this flag is set.</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>Meaning:</b>MKL-DNN cache capacity.
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>Meaning:</b>Number of threads to use for inference on the CPU.</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>Meaning:</b>Path to PaddleX pipeline configuration file.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+(2) Call the <code>predict()</code> method of the general table recognition V2 pipeline object to perform inference prediction, which returns a result list.
+
+Additionally, the pipeline also provides the <code>predict_iter()</code> method. Both methods accept the same parameters and return results in the same way; the difference is that <code>predict_iter()</code> returns a <code>generator</code>, allowing for gradual processing and retrieval of prediction results, suitable for handling large datasets or for scenarios where memory savings are desired. You can choose to use either method based on your actual needs.
+
+The parameters and descriptions of the <code>predict()</code> method are as follows:
+
+<table>
+<thead>
+<tr>
+<th>Parameter</th>
+<th>Description</th>
+<th>Type</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>Meaning:</b>Data to be predicted, supports multiple input types, required.<br/>
+<b>Description:</b>
+<ul>
+<li><b>Python Var</b>: For example, image data represented as <code>numpy.ndarray</code>;</li>
+<li><b>str</b>: Local path to image files or PDF files: <code>/root/data/img.jpg</code>; <b>as URL links</b>, such as network URLs for image files or PDF files: <a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition_v2.jpg">example</a>; <b>as local directories</b>, the directory must contain images to be predicted, such as local path: <code>/root/data/</code> (currently, predictions do not support directories that contain PDF files; the PDF file must be specified to the specific file path);</li>
+<li><b>list</b>: The elements of the list must be of the above types, such as <code>[numpy.ndarray, numpy.ndarray]</code>, <code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>, <code>["/root/data1", "/root/data2"]</code>.</li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to use the document orientation classification module during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>Meaning:</b>Whether to use the text image unwarping module during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>Meaning:</b>Whether to use the layout detection module during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_ocr_model</code></td>
+<td><b>Meaning:</b>Whether to use the <code>ocr</code> model during inference.</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b>
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters.<br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>Meaning:</b>Same meaning as the instantiation parameters. <br/>
+<b>Description:</b> 
+If set to <code>None</code>, the instantiation value is used; otherwise, this parameter takes precedence.</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_e2e_wired_table_rec_model</code></td>
+<td><b>Meaning:</b>Whether to use the wired end-to-end table recognition mode during inference.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_e2e_wireless_table_rec_model</code></td>
+<td><b>Meaning:</b>Whether to use the wireless end-to-end table recognition mode during inference.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_wired_table_cells_trans_to_html</code></td>
+<td><b>Meaning:</b>Whether to use the wired table cell detection result direct-to-HTML mode during inference. <br/>
+<b>Description:</b> 
+If enabled, it directly constructs the HTML based on the geometric relationships of the wired table cell detection results.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_wireless_table_cells_trans_to_html</code></td>
+<td><b>Meaning:</b>Whether to use the wireless table cell detection result direct-to-HTML mode during inference.<br/>
+<b>Description:</b> 
+If enabled, it directly constructs the HTML based on the geometric relationships of the wireless table cell detection results.</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_table_orientation_classify</code></td>
+<td><b>Meaning:</b>Whether to use the table orientation classification mode during inference.<br/> 
+<b>Description:</b> 
+If enabled, it can correct the direction and correctly complete table recognition when the table in the image has 90/180/270-degree rotation.</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_ocr_results_with_table_cells</code></td>
+<td><b>Meaning:</b>Whether to use the cell-split OCR mode during inference.<br/>
+<b>Description:</b> 
+If enabled, it will split and re-recognize OCR detection results based on the cell prediction results to avoid missing text.</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+</tbody>
+</table>
+
+(3) Process the prediction results. The prediction result for each sample is a corresponding Result object, which supports printing, saving as an image, saving as an <code>xlsx</code> file, saving as an <code>HTML</code> file, and saving as a <code>json</code> file:
+
+<table>
+<thead>
+<tr>
+<th>Method</th>
+<th>Description</th>
+<th>Parameter</th>
+<th>Type</th>
+<th>Parameter Description</th>
+<th>Default Value</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">Print results to the terminal</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>Whether to format the output content using <code>JSON</code> indentation.</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data, making it more readable. Effective only when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non-<code>ASCII</code> characters to <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> keeps the original characters. Effective only when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">Save results as a json format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The path to save the file. When it is a directory, the saved file will be named the same as the input file type.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>Specify the indentation level to beautify the output <code>JSON</code> data, making it more readable. Effective only when <code>format_json</code> is <code>True</code>.</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>Control whether to escape non-<code>ASCII</code> characters to <code>Unicode</code>. When set to <code>True</code>, all non-<code>ASCII</code> characters will be escaped; <code>False</code> keeps the original characters. Effective only when <code>format_json</code> is <code>True</code>.</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>Save results as an image format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The path to save the file, supporting directory or file path.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>save_to_xlsx()</code></td>
+<td>Save results as an xlsx format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The path to save the file, supporting directory or file path.</td>
+<td>None</td>
+</tr>
+<tr>
+<td><code>save_to_html()</code></td>
+<td>Save results as an html format file</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>The path to save the file, supporting directory or file path.</td>
+<td>None</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>Calling the <code>print()</code> method will print the results to the terminal. The content printed to the terminal is explained as follows:
+    <ol start="1" type="1">
+    <li><code>input_path</code>: <code>(str)</code> The input path of the image to be predicted.</li>
+    <li><code>page_index</code>: <code>(Union[int, None])</code> If the input is a PDF file, it represents the current page number of the PDF file; otherwise, it is <code>None</code>.</li>
+    <li><code>model_settings</code>: <code>(Dict[str, bool])</code> Configuration parameters required for the production line.
+        <ol>
+        <li><code>use_doc_preprocessor</code>: <code>(bool)</code> Control whether to enable the document preprocessing sub-line.</li>
+        <li><code>use_layout_detection</code>: <code>(bool)</code> Control whether to enable the layout detection sub-line.</li>
+        <li><code>use_ocr_model</code>: <code>(bool)</code> Control whether to enable the OCR sub-line.</li>
+        </ol>
+    </li>
+    <li><code>layout_det_res</code>: <code>(Dict[str, Union[List[numpy.ndarray], List[float]]])</code> Output results of the layout detection sub-module. Only exists when <code>use_layout_detection=True</code>
+        <ol>
+        <li><code>input_path</code>: <code>(Union[str, None])</code> The image path accepted by the layout detection area module, saved as <code>None</code> when the input is <code>numpy.ndarray</code>.</li>
+        <li><code>page_index</code>: <code>(Union[int, None])</code> If the input is a PDF file, it represents the current page number of the PDF file; otherwise, it is <code>None</code>.</li>
+        <li><code>boxes</code>: <code>(List[Dict])</code> A list of layout seal area detection boxes. Each element in the list contains the following fields
+            <ul>
+            <li><code>cls_id</code>: <code>(int)</code> The class ID of the layout seal area detection box.</li>
+            <li><code>score</code>: <code>(float)</code> The confidence score of the layout seal area detection box.</li>
+            <li><code>coordinate</code>: <code>(List[float])</code> The coordinates of the four vertices of the layout seal area detection box, in the order of x1, y1, x2, y2, representing the top-left x coordinate, top-left y coordinate, bottom-right x coordinate, and bottom-right y coordinate.</li>
+            </ul>
+        </li>
+        </ol>
+    </li>
+    <li><code>doc_preprocessor_res</code>: <code>(Dict[str, Union[str, Dict[str, bool], int]])</code> Output results of the document preprocessing sub-module. Only exists when <code>use_doc_preprocessor=True</code>
+        <ol>
+        <li><code>input_path</code>: <code>(Union[str, None])</code> The image path accepted by the document preprocessing area module, saved as <code>None</code> when the input is <code>numpy.ndarray</code>.</li>
+        <li><code>page_index</code>: <code>(Union[int, None])</code> If the input is a PDF file, it represents the current page number of the PDF file; otherwise, it is <code>None</code>.</li>
+        <li><code>model_settings</code>: <code>(Dict)</code> Model configuration parameters of the document preprocessing sub-line.</li>
+        <li><code>model_settings</code>: <code>(Dict)</code> Model configuration parameters of the document preprocessing sub-line.
+            <ul>
+            <li><code>use_doc_orientation_classify</code>: <code>(bool)</code> Control whether to enable document orientation classification.</li>
+            <li><code>use_doc_unwarping</code>: <code>(bool)</code> Control whether to enable text image unwarping.</li>
+            </ul>
+        </li>
+        <li><code>angle</code>: <code>(int)</code> The predicted result of document orientation classification. Enabled values are [0,1,2,3], corresponding to [0°,90°,180°,270°]; disabled value is -1.</li>
+        </ol>
+    </li>
+    <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> List of text detection polygons. Each detection box is represented by a numpy array of shape (4, 2), with data type int16.</li>
+    <li><code>dt_scores</code>: <code>(List[float])</code> List of text detection box confidence scores</li>
+    <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> Configuration parameters for the text detection module.
+        <ol>
+        <li><code>limit_side_len</code>: <code>(int)</code> Side length limit value during image preprocessing.</li>
+        <li><code>limit_type</code>: <code>(str)</code> Type of side length limit processing.</li>
+        <li><code>thresh</code>: <code>(float)</code> Confidence threshold for text pixel classification.</li>
+        <li><code>box_thresh</code>: <code>(float)</code> Confidence threshold for text detection boxes.</li>
+        <li><code>unclip_ratio</code>: <code>(float)</code> Expansion coefficient for text detection boxes.</li>
+        <li><code>text_type</code>: <code>(str)</code> Type of text detection, currently fixed as "general".</li>
+        </ol>
+    </li>
+    <li><code>text_rec_score_thresh</code>: <code>(float)</code> Confidence threshold for text recognition results filtering.</li>
+    <li><code>rec_texts</code>: <code>(List[str])</code> List of text recognition results, only containing texts with confidence scores higher than <code>text_rec_score_thresh</code>.</li>
+    <li><code>rec_scores</code>: <code>(List[float])</code> List of text recognition confidence scores, filtered by <code>text_rec_score_thresh</code>.</li>
+    <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> List of text detection boxes after confidence filtering, in the same format as <code>dt_polys</code>.</li>
+    <li><code>rec_boxes</code>: <code>(numpy.ndarray)</code> Array of rectangular bounding boxes for text detection boxes, with shape (n, 4) and dtype int16. Each row represents the coordinates of a rectangular box in the format [x_min, y_min, x_max, y_max], where (x_min, y_min) is the top-left coordinate and (x_max, y_max) is the bottom-right coordinate.</li>
+    </ol>
+</li>
+<li>Calling the <code>save_to_json()</code> method will save the above content to the specified <code>save_path</code>. If a directory is specified, the saved path will be <code>save_path/{your_img_basename}_res.json</code>; if a file is specified, it will be saved directly to that file. Since json files do not support saving numpy arrays, the <code>numpy.array</code> types will be converted to list format.</li>
+<li>Calling the <code>save_to_img()</code> method will save the visualization results to the specified <code>save_path</code>. If a directory is specified, the saved path will be <code>save_path/{your_img_basename}_ocr_res_img.{your_img_extension}</code>; if a file is specified, it will be saved directly to that file. (The pipeline usually contains many result images, so it is not recommended to specify a specific file path directly, as multiple images will be overwritten and only the last image will be retained.)</li>
+<li>Calling the <code>save_to_html()</code> method will save the above content to the specified <code>save_path</code>. If a directory is specified, the saved path will be <code>save_path/{your_img_basename}_table_1.html</code>; if a file is specified, it will be saved directly to that file. In the general table recognition V2 pipeline, the HTML format of the table in the image will be written to the specified HTML file.</li>
+<li>Calling the <code>save_to_xlsx()</code> method will save the above content to the specified <code>save_path</code>. If a directory is specified, the saved path will be <code>save_path/{your_img_basename}_res.xlsx</code>; if a file is specified, it will be saved directly to that file. In the general table recognition V2 pipeline, the Excel format of the table in the image will be written to the specified xlsx file.</li> 
+
+* Additionally, it also supports obtaining visualization images and prediction results through attributes, as follows:
+
+<table>
+<thead>
+<tr>
+<th>Attribute</th>
+<th>Description</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">Get the prediction results in <code>json</code> format</td>
+</tr>
+<tr>
+<td rowspan="2"><code>img</code></td>
+<td rowspan="2">Get visualization images in <code>dict</code> format</td>
+</tr>
+</table>
+<ul>
+<li>The prediction result obtained by the <code>json</code> attribute is of dict type, and the relevant content is consistent with the content saved by calling the <code>save_to_json()</code> method.</li>
+<li>The prediction result returned by the <code>img</code> attribute is a dictionary type data. The keys are <code>table_res_img</code>、<code>ocr_res_img</code> 、<code>layout_res_img</code>, and <code>preprocessed_img</code>, and the corresponding values are four <code>Image.Image</code> objects, in the order of: visualization image of the table recognition result, visualization image of the OCR result, visualization image of the layout area detection result, and visualization image of the image preprocessing. If a certain sub-module is not used, the corresponding result image will not be included in the dictionary.</li>
+</ul>
+## 3. Development Integration/Deployment
+
+If the model can meet your requirements for inference speed and accuracy, you can proceed directly with development integration/deployment.
+
+If you need to apply the model directly in your Python project, you can refer to the example code in [2.2 Python Script Integration](#22-python-script-integration).
+
+Additionally, PaddleOCR provides two other deployment methods, which are described in detail below:
+
+🚀 High-Performance Inference: In actual production environments, many applications have strict performance criteria (especially response speed) for deployment strategies to ensure efficient system operation and smooth user experience. To address this, PaddleOCR provides high-performance inference capabilities aimed at optimizing model inference and preprocessing to significantly speed up the end-to-end process. For detailed high-performance inference procedures, please refer to [High-Performance Inference](../inference_deployment/local_inference/high_performance_inference.md).
+
+☁️ Service-Oriented Deployment: Service-oriented deployment is a common form of deployment in actual production environments. By encapsulating inference functions as services, clients can access these services via network requests to obtain inference results. For detailed service-oriented deployment procedures, please refer to [Serving](../inference_deployment/serving/serving.md).
+
+Below is the API reference for basic service-oriented deployment and examples of multilingual service calls:
+
+<details><summary>API Reference</summary>
+<p>The main operations provided by the service are as follows:</p>
+<ul>
+<li>HTTP request method is POST.</li>
+<li>Both request and response bodies are JSON data (JSON objects).</li>
+<li>When the request is processed successfully, the response status code is <code>200</code>, and the properties of the response body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>Request UUID.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Fixed to <code>0</code>.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error description. Fixed to <code>"Success"</code>.</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>Operation result.</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request processing is unsuccessful, the properties of the response body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>Request UUID.</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>Error code. Same as the response status code.</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>Error description.</td>
+</tr>
+</tbody>
+</table>
+<p>The main operation provided by the service is as follows:</p>
+<ul>
+<li><b><code>infer</code></b></li>
+</ul>
+<p>Locate and identify tables in the image.</p>
+<p><code>POST /table-recognition</code></p>
+<ul>
+<li>The properties of the request body are as follows:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+<th>Required</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>URL of an accessible image file or PDF file on the server, or the Base64 encoded result of the content of the above types of files. By default, for PDF files with more than 10 pages, only the first 10 pages will be processed.<br /> To lift the page limit, please add the following configuration in the model configuration file:
+<pre><code>Serving:
+  extra:
+    max_num_input_imgs: null
+</code></pre>
+</td>
+<td>Yes</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>File type. <code>0</code> represents a PDF file, <code>1</code> represents an image file. If this property is not present in the request body, the file type will be inferred from the URL.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the <code>use_doc_orientation_classify</code> parameter description in the <code>predict</code> method of the model object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the <code>use_doc_unwarping</code> parameter description in the <code>predict</code> method of the model object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useLayoutDetection</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the <code>use_layout_detection</code> parameter description in the <code>predict</code> method of the model object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useOcrModel</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>Please refer to the <code>use_ocr_model</code> parameter description in the <code>predict</code> method of the model object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>Please refer to the <code>text_det_limit_side_len</code> parameter description in the <code>predict</code> method of the model object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Please refer to the <code>text_det_limit_type</code> parameter description in the <code>predict</code> method of the model object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the <code>text_det_thresh</code> parameter description in the <code>predict</code> method of the model object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the <code>text_det_box_thresh</code> parameter description in the <code>predict</code> method of the model object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the <code>text_det_unclip_ratio</code> parameter description in the <code>predict</code> method of the model object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>textRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>Please refer to the <code>text_rec_score_thresh</code> parameter description in the <code>predict</code> method of the model object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useTableCellsOcrResults</code></td>
+<td><code>boolean</code></td>
+<td>Please refer to the <code>use_table_cells_ocr_results</code> parameter description in the <code>predict</code> method of the model object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useE2eWiredTableRecModel</code></td>
+<td><code>boolean</code></td>
+<td>Please refer to the <code>use_e2e_wired_table_rec_model</code> parameter description in the <code>predict</code> method of the model object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>useE2eWirelessTableRecModel</code></td>
+<td><code>boolean</code></td>
+<td>Please refer to the <code>use_e2e_wireless_table_rec_model</code> parameter description in the <code>predict</code> method of the model object.</td>
+<td>No</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>
+Whether to return the final visualization image and intermediate images during the processing.<br/>
+<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>If <code>true</code> is provided: return images.</li>
+<li>If <code>false</code> is provided: do not return any images.</li>
+<li>If this parameter is omitted from the request body, or if <code>null</code> is explicitly passed, the behavior will follow the value of <code>Serving.visualize</code> in the pipeline configuration.</li>
+</ul>
+<br/>
+For example, adding the following setting to the pipeline config file:<br/>
+<pre><code>Serving:
+  visualize: False
+</code></pre>
+will disable image return by default. This behavior can be overridden by explicitly setting the <code>visualize</code> parameter in the request.<br/>
+If neither the request body nor the configuration file is set (If <code>visualize</code> is set to <code>null</code> in the request and  not defined in the configuration file), the image is returned by default.
+</td>
+<td>No</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>When the request is processed successfully, the <code>result</code> in the response body has the following properties:</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>tableRecResults</code></td>
+<td><code>object</code></td>
+<td>Table recognition results. The length of the array is 1 (for image input) or the actual number of processed document pages (for PDF input). For PDF input, each element in the array represents the result of each processed page in the PDF file.</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>Input data information.</td>
+</tr>
+</tbody>
+</table>
+<p>Image fields in the element schema below (e.g. <code>outputImages</code>, <code>inputImage</code>) are returned inline as Base64 strings by default; when the server is configured to return URLs, those values become pre-signed URLs while the field types remain unchanged. See the "Returning Binary Content as URLs" section of the <a href="../inference_deployment/serving/serving.en.md">Serving Deployment Guide</a> for configuration.</p>
+<p>Each element in <code>tableRecResults</code> is an <code>object</code> with the following properties:</p>
+<table>
+<thead>
+<tr>
+<th>Name</th>
+<th>Type</th>
+<th>Meaning</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>A simplified version of the JSON representation of the result generated by the <code>predict</code> method of the model object, where the <code>input_path</code> and <code>page_index</code> fields are removed.</td>
+</tr>
+<tr>
+<td><code>outputImages</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>Refer to the <code>img</code> property description of the model prediction results. The images are in JPEG format and encoded in Base64 by default; returned as pre-signed URLs when URL-return mode is enabled.</td>
+</tr>
+<tr>
+<td><code>inputImage</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>Input image. The image is in JPEG format and encoded in Base64 by default; returned as a pre-signed URL when URL-return mode is enabled.</td>
+</tr>
+</tbody>
+</table></details>
+<details><summary>Multilingual Service Call Examples</summary>
+<details>
+<summary>Python</summary>
+
+<pre><code class="language-python">import base64
+import requests
+
+API_URL = "http://localhost:8080/table-recognition"
+file_path = "./demo.jpg"
+
+with open(file_path, "rb") as file:
+    file_bytes = file.read()
+    file_data = base64.b64encode(file_bytes).decode("ascii")
+
+payload = {"file": file_data, "fileType": 1}
+
+response = requests.post(API_URL, json=payload)
+
+assert response.status_code == 200
+result = response.json()["result"]
+for i, res in enumerate(result["tableRecResults"]):
+    print(res["prunedResult"])
+    for img_name, img in res["outputImages"].items():
+        img_path = f"{img_name}_{i}.jpg"
+        with open(img_path, "wb") as f:
+            f.write(base64.b64decode(img))
+        print(f"Output image saved at {img_path}")
+</code></pre></details>
+
+<details><summary>C++</summary>
+
+<pre><code class="language-cpp">#include &lt;iostream&gt;
+#include &lt;fstream&gt;
+#include &lt;vector&gt;
+#include &lt;string&gt;
+#include "cpp-httplib/httplib.h" // https://github.com/Huiyicc/cpp-httplib
+#include "nlohmann/json.hpp" // https://github.com/nlohmann/json
+#include "base64.hpp" // https://github.com/tobiaslocker/base64
+
+int main() {
+    httplib::Client client("localhost", 8080);
+    const std::string filePath = "./demo.jpg";
+    std::ifstream file(filePath, std::ios::binary | std::ios::ate);
+    if (!file) {
+        std::cerr << "Error opening file." << std::endl;
+        return 1;
+    }
+
+    std::streamsize size = file.tellg();
+    file.seekg(0, std::ios::beg);
+    std::vector<char> buffer(size);
+
+    if (!file.read(buffer.data(), size)) {
+        std::cerr << "Error reading file." << std::endl;
+        return 1;
+    }
+
+    std::string bufferStr(buffer.data(), static_cast<size_t>(size));
+    std::string encodedFile = base64::to_base64(bufferStr);
+
+    nlohmann::json jsonObj;
+    jsonObj["file"] = encodedFile;
+    jsonObj["fileType"] = 1;
+
+    auto response = client.Post("/table-recognition", jsonObj.dump(), "application/json");
+
+    if (response && response->status == 200) {
+        nlohmann::json jsonResponse = nlohmann::json::parse(response->body);
+        auto result = jsonResponse["result"];
+
+        if (!result.is_object() || !result["tableRecResults"].is_array()) {
+            std::cerr << "Unexpected response structure." << std::endl;
+            return 1;
+        }
+
+        for (size_t i = 0; i < result["tableRecResults"].size(); ++i) {
+            auto tableRecResult = result["tableRecResults"][i];
+            std::cout << tableRecResult["prunedResult"] << std::endl;
+
+            if (tableRecResult["outputImages"].is_object()) {
+                for (auto& img : tableRecResult["outputImages"].items()) {
+                    std::string imgName = img.key();
+                    std::string encodedImage = img.value();
+                    std::string decodedImage = base64::from_base64(encodedImage);
+
+                    std::string imgPath = imgName + "_" + std::to_string(i) + ".jpg";
+                    std::ofstream outputImage(imgPath, std::ios::binary);
+                    if (outputImage.is_open()) {
+                        outputImage.write(decodedImage.c_str(), static_cast<std::streamsize>(decodedImage.size()));
+                        outputImage.close();
+                        std::cout << "Output image saved at " << imgPath << std::endl;
+                    } else {
+                        std::cerr << "Unable to open file for writing: " << imgPath << std::endl;
+                    }
+                }
+            }
+        }
+    } else {
+        std::cerr << "Failed to send HTTP request." << std::endl;
+        if (response) {
+            std::cerr << "HTTP status code: " << response->status << std::endl;
+            std::cerr << "Response body: " << response->body << std::endl;
+        }
+        return 1;
+    }
+
+    return 0;
+}
+</code></pre></details>
+
+<details><summary>Java</summary>
+
+<pre><code class="language-java">import okhttp3.*;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.util.Base64;
+
+public class Main {
+    public static void main(String[] args) throws IOException {
+        String API_URL = "http://localhost:8080/table-recognition";
+        String imagePath = "./demo.jpg";
+
+        File file = new File(imagePath);
+        byte[] fileContent = java.nio.file.Files.readAllBytes(file.toPath());
+        String base64Image = Base64.getEncoder().encodeToString(fileContent);
+
+        ObjectMapper objectMapper = new ObjectMapper();
+        ObjectNode payload = objectMapper.createObjectNode();
+        payload.put("file", base64Image);
+        payload.put("fileType", 1);
+
+        OkHttpClient client = new OkHttpClient();
+        MediaType JSON = MediaType.get("application/json; charset=utf-8");
+        RequestBody body = RequestBody.create(JSON, payload.toString());
+
+        Request request = new Request.Builder()
+                .url(API_URL)
+                .post(body)
+                .build();
+
+        try (Response response = client.newCall(request).execute()) {
+            if (response.isSuccessful()) {
+                String responseBody = response.body().string();
+                JsonNode root = objectMapper.readTree(responseBody);
+                JsonNode result = root.get("result");
+
+                JsonNode tableRecResults = result.get("tableRecResults");
+                for (int i = 0; i < tableRecResults.size(); i++) {
+                    JsonNode item = tableRecResults.get(i);
+
+                    JsonNode prunedResult = item.get("prunedResult");
+                    System.out.println("Pruned Result [" + i + "]: " + prunedResult.toString());
+
+                    JsonNode outputImages = item.get("outputImages");
+
+                    outputImages.fieldNames().forEachRemaining(imgName -> {
+                        String imgBase64 = outputImages.get(imgName).asText();
+                        byte[] imgBytes = Base64.getDecoder().decode(imgBase64);
+                        String imgPath = "output_" + imgName +  ".jpg";
+                        try (FileOutputStream fos = new FileOutputStream(imgPath)) {
+                            fos.write(imgBytes);
+                            System.out.println("Saved image to: " + imgPath);
+                        } catch (IOException e) {
+                            e.printStackTrace();
+                        }
+                    });
+                }
+            } else {
+                System.err.println("Request failed with HTTP code: " + response.code());
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Go</summary>
+
+<pre><code class="language-go">package main
+
+import (
+    "bytes"
+    "encoding/base64"
+    "encoding/json"
+    "fmt"
+    "io/ioutil"
+    "net/http"
+)
+
+func main() {
+    API_URL := "http://localhost:8080/table-recognition"
+    filePath := "./demo.jpg"
+
+    fileBytes, err := ioutil.ReadFile(filePath)
+    if err != nil {
+        fmt.Printf("Error reading file: %v\n", err)
+        return
+    }
+    fileData := base64.StdEncoding.EncodeToString(fileBytes)
+
+    payload := map[string]interface{}{
+        "file":     fileData,
+        "fileType": 1,
+    }
+    payloadBytes, err := json.Marshal(payload)
+    if err != nil {
+        fmt.Printf("Error marshaling payload: %v\n", err)
+        return
+    }
+
+    client := &http.Client{}
+    req, err := http.NewRequest("POST", API_URL, bytes.NewBuffer(payloadBytes))
+    if err != nil {
+        fmt.Printf("Error creating request: %v\n", err)
+        return
+    }
+    req.Header.Set("Content-Type", "application/json")
+
+    res, err := client.Do(req)
+    if err != nil {
+        fmt.Printf("Error sending request: %v\n", err)
+        return
+    }
+    defer res.Body.Close()
+
+    if res.StatusCode != http.StatusOK {
+        fmt.Printf("Unexpected status code: %d\n", res.StatusCode)
+        return
+    }
+
+    body, err := ioutil.ReadAll(res.Body)
+    if err != nil {
+        fmt.Printf("Error reading response body: %v\n", err)
+        return
+    }
+
+    type TableRecResult struct {
+        PrunedResult  map[string]interface{} `json:"prunedResult"`
+        OutputImages  map[string]string      `json:"outputImages"`
+        InputImage    *string                `json:"inputImage"`
+    }
+
+    type Response struct {
+        Result struct {
+            TableRecResults []TableRecResult `json:"tableRecResults"`
+            DataInfo        interface{}      `json:"dataInfo"`
+        } `json:"result"`
+    }
+
+    var respData Response
+    if err := json.Unmarshal(body, &respData); err != nil {
+        fmt.Printf("Error unmarshaling response: %v\n", err)
+        return
+    }
+
+    for i, res := range respData.Result.TableRecResults {
+        fmt.Printf("Result %d - prunedResult: %+v\n", i, res.PrunedResult)
+
+        for imgName, imgData := range res.OutputImages {
+            imgBytes, err := base64.StdEncoding.DecodeString(imgData)
+            if err != nil {
+                fmt.Printf("Error decoding image %s_%d: %v\n", imgName, i, err)
+                continue
+            }
+
+            filename := fmt.Sprintf("%s_%d.jpg", imgName, i)
+            if err := ioutil.WriteFile(filename, imgBytes, 0644); err != nil {
+                fmt.Printf("Error saving image %s: %v\n", filename, err)
+                continue
+            }
+            fmt.Printf("Saved image to %s\n", filename)
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>C#</summary>
+
+<pre><code class="language-csharp">using System;
+using System.IO;
+using System.Net.Http;
+using System.Text;
+using System.Threading.Tasks;
+using Newtonsoft.Json.Linq;
+
+class Program
+{
+    static readonly string API_URL = "http://localhost:8080/table-recognition";
+    static readonly string inputFilePath = "./demo.jpg";
+
+    static async Task Main(string[] args)
+    {
+        var httpClient = new HttpClient();
+
+        byte[] fileBytes = File.ReadAllBytes(inputFilePath);
+        string fileData = Convert.ToBase64String(fileBytes);
+
+        var payload = new JObject
+        {
+            { "file", fileData },
+            { "fileType", 1 }
+        };
+        var content = new StringContent(payload.ToString(), Encoding.UTF8, "application/json");
+
+        HttpResponseMessage response = await httpClient.PostAsync(API_URL, content);
+        response.EnsureSuccessStatusCode();
+
+        string responseBody = await response.Content.ReadAsStringAsync();
+        JObject jsonResponse = JObject.Parse(responseBody);
+
+        JArray tableRecResults = (JArray)jsonResponse["result"]["tableRecResults"];
+        for (int i = 0; i < tableRecResults.Count; i++)
+        {
+            var res = tableRecResults[i];
+            Console.WriteLine($"[{i}] prunedResult:\n{res["prunedResult"]}");
+
+            JObject outputImages = res["outputImages"] as JObject;
+            if (outputImages != null)
+            {
+                foreach (var img in outputImages)
+                {
+                    string imgName = img.Key;
+                    string base64Img = img.Value?.ToString();
+                    if (!string.IsNullOrEmpty(base64Img))
+                    {
+                        string imgPath = $"{imgName}_{i}.jpg";
+                        byte[] imageBytes = Convert.FromBase64String(base64Img);
+                        File.WriteAllBytes(imgPath, imageBytes);
+                        Console.WriteLine($"Output image saved at {imgPath}");
+                    }
+                }
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Node.js</summary>
+
+<pre><code class="language-js">const axios = require('axios');
+const fs = require('fs');
+const path = require('path');
+
+const API_URL = 'http://localhost:8080/table-recognition';
+const inputImagePath = './demo.jpg';
+
+function encodeImageToBase64(filePath) {
+  const bitmap = fs.readFileSync(filePath);
+  return Buffer.from(bitmap).toString('base64');
+}
+
+async function callTableRecognitionAPI() {
+  const payload = {
+    file: encodeImageToBase64(inputImagePath),
+    fileType: 1
+  };
+
+  try {
+    const response = await axios.post(API_URL, payload, {
+      headers: {
+        'Content-Type': 'application/json'
+      },
+      maxBodyLength: Infinity
+    });
+
+    const results = response.data.result.tableRecResults;
+
+    results.forEach((res, index) => {
+      console.log(`Result [${index}] prunedResult:\n`, res.prunedResult);
+
+      const outputImages = res.outputImages || {};
+      Object.entries(outputImages).forEach(([imgName, base64Img]) => {
+        const outputPath = `${imgName}_${index}.jpg`;
+        fs.writeFileSync(outputPath, Buffer.from(base64Img, 'base64'));
+        console.log(`Saved image: ${outputPath}`);
+      });
+    });
+
+  } catch (error) {
+    console.error('API request failed:', error.message);
+  }
+}
+
+callTableRecognitionAPI();
+
+</code></pre></details>
+
+<details><summary>PHP</summary>
+
+<pre><code class="language-php">&lt;?php
+
+$API_URL = "http://localhost:8080/table-recognition";
+$image_path = "./demo.jpg";
+
+$image_data = base64_encode(file_get_contents($image_path));
+$payload = array(
+    "file" => $image_data,
+    "fileType" => 1
+);
+
+$ch = curl_init($API_URL);
+curl_setopt($ch, CURLOPT_POST, true);
+curl_setopt($ch, CURLOPT_POSTFIELDS, json_encode($payload));
+curl_setopt($ch, CURLOPT_HTTPHEADER, array('Content-Type: application/json'));
+curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+$response = curl_exec($ch);
+curl_close($ch);
+
+$result_array = json_decode($response, true);
+$results = $result_array["result"]["tableRecResults"];
+
+foreach ($results as $i => $item) {
+    echo "[$i] prunedResult:\n";
+    print_r($item["prunedResult"]);
+
+    if (!empty($item["outputImages"])) {
+        foreach ($item["outputImages"] as $img_name => $base64_img) {
+            $img_path = $img_name . "_" . $i . ".jpg";
+            file_put_contents($img_path, base64_decode($base64_img));
+            echo "Output image saved at $img_path\n";
+        }
+    } else {
+        echo "No outputImages found for item $i\n";
+    }
+}
+?&gt;
+</code></pre></details>
+</details>
+<br/>
+
+## 4. Secondary Development
+
+If the default model weights provided by the General Table Recognition v2 model are not satisfactory in terms of accuracy or speed for your scenario, you can try further <b>fine-tuning</b> the existing model using <b>your own domain-specific or application scenario data</b> to improve the recognition performance of the General Table Recognition v2 model in your scenario.
+
+Since the General Table Recognition v2 model consists of several modules, if the model's performance is not as expected, it may be due to any of these modules. You can analyze images with poor recognition performance to determine which module is problematic and refer to the corresponding fine-tuning tutorial links in the following table for model fine-tuning.
+
+<table>
+<thead>
+<tr>
+<th>Situation</th>
+<th>Fine-Tuning Module</th>
+<th>Fine-Tuning Reference Link</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>Table classification error</td>
+<td>Table Classification Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/table_classification.html#_5">Link</a></td>
+</tr>
+<tr>
+<td>Table cell location error</td>
+<td>Table Cell Detection Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/table_cells_detection.html#iv-secondary-development">Link</a></td>
+</tr>
+<tr>
+<td>Table structure recognition error</td>
+<td>Table Structure Recognition Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/module_usage/table_structure_recognition.html#4-secondary-development">Link</a></td>
+</tr>
+<tr>
+<td>Failed to detect the area where the table is located</td>
+<td>Layout Area Detection Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/layout_detection.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Text detection missed</td>
+<td>Text Detection Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/module_usage/text_detection.html#4-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Incorrect text content</td>
+<td>Text Recognition Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/en/version3.x/module_usage/text_recognition.html#v-secondary-development">Link</a></td>
+</tr>
+<tr>
+<td>Overall image rotation/table rotation correction is inaccurate</td>
+<td>Document Image Orientation Classification Module</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/en/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#iv-custom-development">Link</a></td>
+</tr>
+<tr>
+<td>Image distortion correction is inaccurate</td>
+<td>Text Image Correction Module</td>
+<td>Fine-tuning not supported</td>
+</tr>
+</tbody>
+</table>
+
+
+### 4.2 Model Application
+
+When you complete fine-tuning with your private dataset, you’ll obtain local model weight files. You can then use these fine-tuned model weights either by specifying the local model save path through parameters or by customizing the pipeline configuration file.
+
+#### 4.2.1 Specifying Local Model Path Through Parameters
+
+When initializing the pipeline object, you can specify the local model path via parameters. Taking the use of fine-tuned weights for the SLANeXt_wired table structure recognition model as an example:
+
+Command line method:
+
+```bash
+# Specify local model path via --wired_table_structure_recognition_model_dir
+paddleocr table_recognition_v2_pipeline -i ./table_recognition_v2.jpg --wired_table_structure_recognition_model_dir your_model_path
+```
+
+```bash
+# If using SLANeXt_wired model as the default wired table structure recognition model, and if you fine-tuned a different model, modify the model name via --wired_table_structure_recognition_model_name
+paddleocr table_recognition_v2_pipeline -i ./table_recognition_v2.jpg --wired_table_structure_recognition_model_name SLANeXt_wired --wired_table_structure_recognition_model_dir your_model_path
+```
+
+Python script method:
+
+```python
+from paddleocr import TableRecognitionPipelineV2
+
+# Specify local model path via wired_table_structure_recognition_model_dir
+pipeline = TableRecognitionPipelineV2(wired_table_structure_recognition_model_dir="./your_model_path")
+
+# By default, SLANeXt_wired is used as the default table recognition model. If you fine-tuned a different model, modify the model name via wired_table_structure_recognition_model_name
+# pipeline = PaddleOCR(wired_table_structure_recognition_model_name="SLANeXt_wired", wired_table_structure_recognition_model_dir="./your_model_path")
+```
+
+#### 4.2.2 Specifying Local Model Path Through Configuration File
+
+1.Obtain the pipeline configuration file
+You can call the `export_paddlex_config_to_yaml method` of the TableRecognitionPipelineV2 object in PaddleOCR to export the current pipeline configuration to a YAML file:
+
+```Python
+from paddleocr import TableRecognitionPipelineV2
+
+pipeline = TableRecognitionPipelineV2()
+pipeline.export_paddlex_config_to_yaml("TableRecognitionPipelineV2.yaml")
+```
+
+2.Modify the configuration file
+
+After obtaining the default pipeline configuration file, simply replace the corresponding paths in the pipeline configuration file with the local paths of your fine-tuned model weights. For example:
+
+```yaml
+......
+SubModules:
+  LayoutDetection:
+    module_name: layout_detection
+    model_name: PicoDet_layout_1x_table
+    model_dir: null # Replace with the path to fine-tuned layout detection model weights
+
+  TableClassification:
+    module_name: table_classification
+    model_name: PP-LCNet_x1_0_table_cls
+    model_dir: null # Replace with the path to fine-tuned table classification model weights
+
+  WiredTableStructureRecognition:
+    module_name: table_structure_recognition
+    model_name: SLANeXt_wired
+    model_dir: null # Replace with the path to fine-tuned wired table structure recognition model weights
+
+  WirelessTableStructureRecognition:
+    module_name: table_structure_recognition
+    model_name: SLANeXt_wireless
+    model_dir: null # Replace with the path to fine-tuned wireless table structure recognition model weights
+
+  WiredTableCellsDetection:
+    module_name: table_cells_detection
+    model_name: RT-DETR-L_wired_table_cell_det
+    model_dir: null # Replace with the path to fine-tuned wired table cell detection model weights
+
+  WirelessTableCellsDetection:
+    module_name: table_cells_detection
+    model_name: RT-DETR-L_wireless_table_cell_det
+    model_dir: null # Replace with the path to fine-tuned wireless table cell detection model weights
+
+SubPipelines:
+  DocPreprocessor:
+    pipeline_name: doc_preprocessor
+    use_doc_orientation_classify: True
+    use_doc_unwarping: True
+    SubModules:
+      DocOrientationClassify:
+        module_name: doc_text_orientation
+        model_name: PP-LCNet_x1_0_doc_ori
+        model_dir: null # Replace with the path to fine-tuned document orientation classification model weights
+
+      DocUnwarping:
+        module_name: image_unwarping
+        model_name: UVDoc
+        model_dir: null
+
+  GeneralOCR:
+    pipeline_name: OCR
+    text_type: general
+    use_doc_preprocessor: False
+    use_textline_orientation: False
+    SubModules:
+      TextDetection:
+        module_name: text_detection
+        model_name: PP-OCRv5_server_det
+        model_dir: null # Replace with the path to fine-tuned text detection model weights
+        limit_side_len: 960
+        limit_type: max
+        max_side_limit: 4000
+        thresh: 0.3
+        box_thresh: 0.4
+        unclip_ratio: 1.5
+
+      TextRecognition:
+        module_name: text_recognition
+        model_name: PP-OCRv5_server_rec
+        model_dir: null # Replace with the path to fine-tuned text recognition model weights
+        batch_size: 1
+        score_thresh: 0
+......
+```
+
+The pipeline configuration file includes not only the parameters supported by PaddleOCR CLI and Python API but also allows for more advanced configurations. For detailed information, you can find the corresponding pipeline usage tutorial in [Overview of PaddleX Model Pipeline Usage](https://paddlepaddle.github.io/PaddleX/latest/en/pipeline_usage/pipeline_develop_guide.html), and refer to the detailed instructions to adjust the configurations according to your needs.
+
+3. Loading the pipeline configuration file in CLI
+
+After completing the configuration file modifications, specify the path to the modified pipeline configuration file using the `--paddlex_config` parameter in the command line. PaddleOCR will read its contents as the pipeline configuration. For example:
+
+```bash
+paddleocr table_recognition_v2_pipeline --paddlex_config PaddleOCR.yaml ...
+```
+
+4. Loading the pipeline configuration file in Python API
+
+When initializing the pipeline object, you can pass the PaddleX pipeline configuration file path or configuration dictionary through the `paddlex_config` parameter. PaddleOCR will read its contents as the pipeline configuration. For example:
+
+```python
+from paddleocr import TableRecognitionPipelineV2
+
+pipeline = TableRecognitionPipelineV2(paddlex_config="TableRecognitionPipelineV2.yaml")
+```
diff --git a/docs/version3.x/pipeline_usage/table_recognition_v2.md b/docs/version3.x/pipeline_usage/table_recognition_v2.md
new file mode 100644
index 0000000..209bd2a
--- /dev/null
+++ b/docs/version3.x/pipeline_usage/table_recognition_v2.md
@@ -0,0 +1,2915 @@
+---
+comments: true
+---
+
+# 通用表格识别v2产线使用教程
+
+## 1. 通用表格识别v2产线介绍
+
+表格识别是一种自动从文档或图像中识别和提取表格内容及其结构的技术，广泛应用于数据录入、信息检索和文档分析等领域。通过使用计算机视觉和机器学习算法，表格识别能够将复杂的表格信息转换为可编辑的格式，方便用户进一步处理和分析数据。
+
+通用表格识别v2产线（PP-TableMagic）用于解决表格识别任务，对图片中的表格进行识别，并以HTML格式输出。与通用表格识别产线不同，本产线新引入了表格分类和表格单元格检测两个模块，通过<b>采用“表格分类+表格结构识别+单元格检测”多模型串联组网方案</b>，实现了相比通用表格识别产线更好的端到端表格识别性能。基于此，通用表格识别v2产线<b>原生支持针对性地模型微调</b>，各类开发者均能对通用表格识别v2产线进行不同程度的自定义微调，使其在不同应用场景下都能得到令人满意的性能。<b>除此之外，通用表格识别v2产线同样支持使用端到端表格结构识别模型（例如 SLANet、SLANet_plus 等），并且支持有线表、无线表独立配置表格识别方式，开发者可以自由选取和组合最佳的表格识别方案。</b>
+
+本产线的使用场景覆盖通用、制造、金融、交通等各个领域。本产线同时提供了灵活的服务化部署方式，支持在多种硬件上使用多种编程语言调用。不仅如此，本产线也提供了二次开发的能力，您可以基于本产线在您自己的数据集上训练调优，训练后的模型也可以无缝集成。
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/table_recognition_v2/01.png"/>
+
+<b>通用表格识别产线v2中包含以下8个模块。每个模块均可独立进行训练和推理，并包含多个模型。有关详细信息，请点击相应模块以查看文档。</b>
+
+- [表格结构识别模块](../module_usage/table_structure_recognition.md)
+- [表格分类模块](../module_usage/table_classification.md)
+- [表格单元格检测模块](../module_usage/table_cells_detection.md)
+- [文本检测模块](../module_usage/text_detection.md)
+- [文本识别模块](../module_usage/text_recognition.md)
+- [版面区域检测模块](../module_usage/layout_detection.md)（可选）
+- [文档图像方向分类模块](../module_usage/doc_img_orientation_classification.md) （可选）
+- [文本图像矫正模块](../module_usage/text_image_unwarping.md) （可选）
+
+在本产线中，您可以根据下方的基准测试数据选择使用的模型。
+
+> 推理耗时仅包含模型推理耗时，不包含前后处理耗时。
+> 在带有 [常规模式 / 高性能模式] 标记的推理耗时列中，`常规模式` 对应本地推理引擎 `paddle_static`。
+
+<details>
+<summary> <b>表格结构识别模块模型：</b></summary>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>精度（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>SLANet</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANet_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_pretrained.pdparams">训练模型</a></td>
+<td>59.52</td>
+<td>23.96 / 21.75</td>
+<td>- / 43.12</td>
+<td>6.9</td>
+<td rowspan="1">SLANet 是百度飞桨视觉团队自研的表格结构识别模型。该模型通过采用 CPU 友好型轻量级骨干网络 PP-LCNet、高低层特征融合模块 CSP-PAN、结构与位置信息对齐的特征解码模块 SLA Head，大幅提升了表格结构识别的精度和推理速度。</td>
+</tr>
+<tr>
+<td>SLANet_plus</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANet_plus_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANet_plus_pretrained.pdparams">训练模型</a></td>
+<td>63.69</td>
+<td>23.43 / 22.16</td>
+<td>- / 41.80</td>
+<td>6.9</td>
+<td rowspan="1">SLANet_plus 是百度飞桨视觉团队自研的表格结构识别模型 SLANet 的增强版。相较于 SLANet，SLANet_plus 对无线表、复杂表格的识别能力得到了大幅提升，并降低了模型对表格定位准确性的敏感度，即使表格定位出现偏移，也能够较准确地进行识别。
+</td>
+</tr>
+<tr>
+<td>SLANeXt_wired</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wired_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wired_pretrained.pdparams">训练模型</a></td>
+<td rowspan="2">69.65</td>
+<td rowspan="2">85.92 / 85.92</td>
+<td rowspan="2">- / 501.66</td>
+<td rowspan="2">351</td>
+<td rowspan="2">SLANeXt 系列是百度飞桨视觉团队自研的新一代表格结构识别模型。相较于 SLANet 和 SLANet_plus，SLANeXt 专注于对表格结构进行识别，并且对有线表格(wired)和无线表格(wireless)的识别分别训练了专用的权重，对各类型表格的识别能力都得到了明显提高，特别是对有线表格的识别能力得到了大幅提升。</td>
+</tr>
+<tr>
+<td>SLANeXt_wireless</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/SLANeXt_wireless_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/SLANeXt_wireless_pretrained.pdparams">训练模型</a></td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary> <b>表格分类模块模型：</b></summary>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>Top1 Acc(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+</tr>
+<tr>
+<td>PP-LCNet_x1_0_table_cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/CLIP_vit_base_patch16_224_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_table_cls_pretrained.pdparams">训练模型</a></td>
+<td>94.2</td>
+<td>2.62 / 0.60</td>
+<td>3.17 / 1.14</td>
+<td>6.6</td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary> <b>表格单元格检测模块模型：</b></summary>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>RT-DETR-L_wired_table_cell_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wired_table_cell_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wired_table_cell_det_pretrained.pdparams">训练模型</a></td>
+<td rowspan="2">82.7</td>
+<td rowspan="2">33.47 / 27.02</td>
+<td rowspan="2">402.55 / 256.56</td>
+<td rowspan="2">124</td>
+<td rowspan="2">RT-DETR 是第一个实时的端到端目标检测模型。百度飞桨视觉团队基于 RT-DETR-L 作为基础模型，在自建表格单元格检测数据集上完成预训练，实现了对有线表格、无线表格均有较好性能的表格单元格检测。
+</td>
+</tr>
+<tr>
+<td>RT-DETR-L_wireless_table_cell_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-L_wireless_table_cell_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-L_wireless_table_cell_det_pretrained.pdparams">训练模型</a></td>
+</tr>
+</table>
+</details>
+
+<details>
+<summary> <b>文本检测模块模型：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>检测Hmean（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+<tr>
+<td>PP-OCRv5_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_server_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_det_pretrained.pdparams">训练模型</a></td>
+<td>83.8</td>
+<td>89.55 / 70.19</td>
+<td>383.15 / 383.15</td>
+<td>84.3</td>
+<td>PP-OCRv5 的服务端文本检测模型，精度更高，适合在性能较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_det_pretrained.pdparams">训练模型</a></td>
+<td>79.0</td>
+<td>10.67 / 6.36</td>
+<td>57.77 / 28.15</td>
+<td>4.7</td>
+<td>PP-OCRv5 的移动端文本检测模型，效率更高，适合在端侧设备部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_det_pretrained.pdparams">训练模型</a></td>
+<td>69.2</td>
+<td>127.82 / 98.87</td>
+<td>585.95 / 489.77</td>
+<td>109</td>
+<td>PP-OCRv4 的服务端文本检测模型，精度更高，适合在性能较好的服务器上部署</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_det</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_det_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams">训练模型</a></td>
+<td>63.8</td>
+<td>9.87 / 4.17</td>
+<td>56.60 / 20.79</td>
+<td>4.7</td>
+<td>PP-OCRv4 的移动端文本检测模型，效率更高，适合在端侧设备部署</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary><b>文本识别模块：</b></summary>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>86.38</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec 是新一代文本识别模型。该模型致力于以单一模型高效、精准地支持简体中文、繁体中文、英文、日文四种主要语言，以及手写、竖版、拼音、生僻字等复杂文本场景的识别。在保持识别效果的同时，兼顾推理速度和模型鲁棒性，为各种场景下的文档理解提供高效、精准的技术支撑。</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.29</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">训练模型</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc是在PP-OCRv4_server_rec的基础上，在更多中文文档数据和PP-OCR训练数据的混合数据训练而成，增加了部分繁体字、日文、特殊字符的识别能力，可支持识别的字符为1.5万+，除文档相关的文字识别能力提升外，也同时提升了通用文字的识别能力</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>PP-OCRv4的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>PP-OCRv4的服务器端模型，推理精度高，可以部署在多种不同的服务器上</td>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>基于PP-OCRv4识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+</table>
+
+> ❗ 以上列出的是文本识别模块重点支持的<b>6个核心模型</b>，该模块总共支持<b>20个全量模型</b>，包含多个多语言文本识别模型，完整的模型列表如下：
+
+<details><summary> 👉模型列表详情</summary>
+
+* <b>PP-OCRv5 多场景模型</b>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>中文识别 Avg Accuracy(%)</th>
+<th>英文识别 Avg Accuracy(%)</th>
+<th>繁体中文识别 Avg Accuracy(%)</th>
+<th>日文识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv5_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>86.38</td>
+<td>64.70</td>
+<td>93.29</td>
+<td>60.35</td>
+<td>8.46 / 2.36</td>
+<td>31.21 / 31.21</td>
+<td>81</td>
+<td rowspan="2">PP-OCRv5_rec 是新一代文本识别模型。该模型致力于以单一模型高效、精准地支持简体中文、繁体中文、英文、日文四种主要语言，以及手写、竖版、拼音、生僻字等复杂文本场景的识别。在保持识别效果的同时，兼顾推理速度和模型鲁棒性，为各种场景下的文档理解提供高效、精准的技术支撑。</td>
+</tr>
+<tr>
+<td>PP-OCRv5_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv5_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv5_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>81.29</td>
+<td>66.00</td>
+<td>83.55</td>
+<td>54.65</td>
+<td>5.43 / 1.46</td>
+<td>21.20 / 5.32</td>
+<td>16</td>
+</tr>
+</table>
+
+* <b>中文识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec_doc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv4_server_rec_doc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_doc_pretrained.pdparams">训练模型</a></td>
+<td>86.58</td>
+<td>8.69 / 2.78</td>
+<td>37.93 / 37.93</td>
+<td>182</td>
+<td>PP-OCRv4_server_rec_doc是在PP-OCRv4_server_rec的基础上，在更多中文文档数据和PP-OCR训练数据的混合数据训练而成，增加了部分繁体字、日文、特殊字符的识别能力，可支持识别的字符为1.5万+，除文档相关的文字识别能力提升外，也同时提升了通用文字的识别能力</td>
+</tr>
+<tr>
+<td>PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>78.74</td>
+<td>5.26 / 1.12</td>
+<td>17.48 / 3.61</td>
+<td>10.5</td>
+<td>PP-OCRv4的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+<tr>
+<td>PP-OCRv4_server_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv4_server_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_rec_pretrained.pdparams">训练模型</a></td>
+<td>85.19</td>
+<td>8.75 / 2.49</td>
+<td>36.93 / 36.93</td>
+<td>173</td>
+<td>PP-OCRv4的服务器端模型，推理精度高，可以部署在多种不同的服务器上</td>
+</tr>
+<tr>
+<td>PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>72.96</td>
+<td>3.89 / 1.16</td>
+<td>8.72 / 3.56</td>
+<td>10.3</td>
+<td>PP-OCRv3的轻量级识别模型，推理效率高，可以部署在包含端侧设备的多种硬件设备中</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>ch_SVTRv2_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_SVTRv2_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_SVTRv2_rec_pretrained.pdparams">训练模型</a></td>
+<td>68.81</td>
+<td>10.38 / 8.31</td>
+<td>66.52 / 30.83</td>
+<td>80.5</td>
+<td rowspan="1">
+SVTRv2 是一种由复旦大学视觉与学习实验室（FVL）的OpenOCR团队研发的服务端文本识别模型，其在PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务中荣获一等奖，A榜端到端识别精度相比PP-OCRv4提升6%。
+</td>
+</tr>
+</table>
+
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>ch_RepSVTR_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/ch_RepSVTR_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ch_RepSVTR_rec_pretrained.pdparams">训练模型</a></td>
+<td>65.07</td>
+<td>6.29 / 1.57</td>
+<td>20.64 / 5.40</td>
+<td>48.8</td>
+<td rowspan="1">    RepSVTR 文本识别模型是一种基于SVTRv2 的移动端文本识别模型，其在PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务中荣获一等奖，B榜端到端识别精度相比PP-OCRv4提升2.5%，推理速度持平。</td>
+</tr>
+</table>
+
+* <b>英文识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>en_PP-OCRv4_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv4_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv4_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td> 70.39</td>
+<td>4.81 / 1.23</td>
+<td>17.20 / 4.18</td>
+<td>7.5</td>
+<td>基于PP-OCRv4识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+<tr>
+<td>en_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+en_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/en_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>70.69</td>
+<td>3.56 / 0.78</td>
+<td>8.44 / 5.78</td>
+<td>17.3</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量英文识别模型，支持英文、数字识别</td>
+</tr>
+</table>
+
+
+* <b>多语言识别模型</b>
+<table>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>识别 Avg Accuracy(%)</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+<tr>
+<td>korean_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+korean_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/korean_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>60.21</td>
+<td>3.73 / 0.98</td>
+<td>8.76 / 2.91</td>
+<td>9.6</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量韩文识别模型，支持韩文、数字识别</td>
+</tr>
+<tr>
+<td>japan_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+japan_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/japan_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>45.69</td>
+<td>3.86 / 1.01</td>
+<td>8.62 / 2.92</td>
+<td>9.8</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量日文识别模型，支持日文、数字识别</td>
+</tr>
+<tr>
+<td>chinese_cht_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+chinese_cht_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/chinese_cht_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>82.06</td>
+<td>3.90 / 1.16</td>
+<td>9.24 / 3.18</td>
+<td>10.8</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量繁体中文识别模型，支持繁体中文、数字识别</td>
+</tr>
+<tr>
+<td>te_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+te_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/te_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>95.88</td>
+<td>3.59 / 0.81</td>
+<td>8.28 / 6.21</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量泰卢固文识别模型，支持泰卢固文、数字识别</td>
+</tr>
+<tr>
+<td>ka_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ka_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ka_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>96.96</td>
+<td>3.49 / 0.89</td>
+<td>8.63 / 2.77</td>
+<td>17.4</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量卡纳达文识别模型，支持卡纳达文、数字识别</td>
+</tr>
+<tr>
+<td>ta_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+ta_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/ta_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>76.83</td>
+<td>3.49 / 0.86</td>
+<td>8.35 / 3.41</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量泰米尔文识别模型，支持泰米尔文、数字识别</td>
+</tr>
+<tr>
+<td>latin_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+latin_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/latin_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>76.93</td>
+<td>3.53 / 0.78</td>
+<td>8.50 / 6.83</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量拉丁文识别模型，支持拉丁文、数字识别</td>
+</tr>
+<tr>
+<td>arabic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+arabic_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/arabic_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>73.55</td>
+<td>3.60 / 0.83</td>
+<td>8.44 / 4.69</td>
+<td>17.3</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量阿拉伯字母识别模型，支持阿拉伯字母、数字识别</td>
+</tr>
+<tr>
+<td>cyrillic_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+cyrillic_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/cyrillic_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>94.28</td>
+<td>3.56 / 0.79</td>
+<td>8.22 / 2.76</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量斯拉夫字母识别模型，支持斯拉夫字母、数字识别</td>
+</tr>
+<tr>
+<td>devanagari_PP-OCRv3_mobile_rec</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/\
+devanagari_PP-OCRv3_mobile_rec_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/devanagari_PP-OCRv3_mobile_rec_pretrained.pdparams">训练模型</a></td>
+<td>96.44</td>
+<td>3.60 / 0.78</td>
+<td>6.95 / 2.87</td>
+<td>8.7</td>
+<td>基于PP-OCRv3识别模型训练得到的超轻量梵文字母识别模型，支持梵文字母、数字识别</td>
+</tr>
+</table>
+</details>
+</details>
+
+<details>
+<summary> <b>版面区域检测模块模型：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-DocLayout_plus-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout_plus-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout_plus-L_pretrained.pdparams">训练模型</a></td>
+<td>83.2</td>
+<td>53.03 / 17.23</td>
+<td>634.62 / 378.32</td>
+<td>126.01</td>
+<td>基于RT-DETR-L在包含中英文论文、多栏杂志、报纸、PPT、合同、书本、试卷、研报、古籍、日文文档、竖版文字文档等场景的自建数据集训练的更高精度版面区域定位模型</td>
+</tr>
+<tr>
+<tr>
+<td>PP-DocLayout-L</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-L_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-L_pretrained.pdparams">训练模型</a></td>
+<td>90.4</td>
+<td>33.59 / 33.59</td>
+<td>503.01 / 251.08</td>
+<td>123.76</td>
+<td>基于RT-DETR-L在包含中英文论文、杂志、合同、书本、试卷和研报等场景的自建数据集训练的高精度版面区域定位模型</td>
+</tr>
+<tr>
+<td>PP-DocLayout-M</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-M_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-M_pretrained.pdparams">训练模型</a></td>
+<td>75.2</td>
+<td>13.03 / 4.72</td>
+<td>43.39 / 24.44</td>
+<td>22.578</td>
+<td>基于PicoDet-L在包含中英文论文、杂志、合同、书本、试卷和研报等场景的自建数据集训练的精度效率平衡的版面区域定位模型</td>
+</tr>
+<tr>
+<td>PP-DocLayout-S</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-DocLayout-S_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-DocLayout-S_pretrained.pdparams">训练模型</a></td>
+<td>70.9</td>
+<td>11.54 / 3.86</td>
+<td>18.53 / 6.29</td>
+<td>4.834</td>
+<td>基于PicoDet-S在中英文论文、杂志、合同、书本、试卷和研报等场景上自建数据集训练的高效率版面区域定位模型</td>
+</tr>
+</tbody>
+</table>
+
+> ❗ 以上列出的是版面检测模块重点支持的<b>4个核心模型</b>，该模块总共支持<b>12个全量模型</b>，包含多个预定义了不同类别的模型，完整的模型列表如下：
+<details><summary> 👉模型列表详情</summary>
+* <b>表格版面检测模型</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet_layout_1x_table</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_table_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_table_pretrained.pdparams">训练模型</a></td>
+<td>97.5</td>
+<td>9.57 / 6.63</td>
+<td>27.66 / 16.75</td>
+<td>7.4</td>
+<td>基于PicoDet-1x在自建数据集训练的高效率版面区域定位模型，可定位表格这1类区域</td>
+</tr>
+</tbody></table>
+
+* <b>3类版面检测模型，包含表格、图像、印章</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_3cls_pretrained.pdparams">训练模型</a></td>
+<td>88.2</td>
+<td>8.43 / 3.44</td>
+<td>17.60 / 6.51</td>
+<td>4.8</td>
+<td>基于PicoDet-S轻量模型在中英文论文、杂志和研报等场景上自建数据集训练的高效率版面区域定位模型</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_3cls_pretrained.pdparams">训练模型</a></td>
+<td>89.0</td>
+<td>12.80 / 9.57</td>
+<td>45.04 / 23.86</td>
+<td>22.6</td>
+<td>基于PicoDet-L在中英文论文、杂志和研报等场景上自建数据集训练的效率精度均衡版面区域定位模型</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_3cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_3cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_3cls_pretrained.pdparams">训练模型</a></td>
+<td>95.8</td>
+<td>114.80 / 25.65</td>
+<td>924.38 / 924.38</td>
+<td>470.1</td>
+<td>基于RT-DETR-H在中英文论文、杂志和研报等场景上自建数据集训练的高精度版面区域定位模型</td>
+</tr>
+</tbody></table>
+
+* <b>5类英文文档区域检测模型，包含文字、标题、表格、图片以及列表</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet_layout_1x</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet_layout_1x_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet_layout_1x_pretrained.pdparams">训练模型</a></td>
+<td>97.8</td>
+<td>9.62 / 6.75</td>
+<td>26.96 / 12.77</td>
+<td>7.4</td>
+<td>基于PicoDet-1x在PubLayNet数据集训练的高效率英文文档版面区域定位模型</td>
+</tr>
+</tbody></table>
+</b>
+
+* <b>17类区域检测模型，包含17个版面常见类别，分别是：段落标题、图片、文本、数字、摘要、内容、图表标题、公式、表格、表格标题、参考文献、文档标题、脚注、页眉、算法、页脚、印章</b>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>mAP(0.5)（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PicoDet-S_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-S_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-S_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>87.4</td>
+<td>8.80 / 3.62</td>
+<td>17.51 / 6.35</td>
+<td>4.8</td>
+<td>基于PicoDet-S轻量模型在中英文论文、杂志和研报等场景上自建数据集训练的高效率版面区域定位模型</td>
+</tr>
+<tr>
+<td>PicoDet-L_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PicoDet-L_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PicoDet-L_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>89.0</td>
+<td>12.60 / 10.27</td>
+<td>43.70 / 24.42</td>
+<td>22.6</td>
+<td>基于PicoDet-L在中英文论文、杂志和研报等场景上自建数据集训练的效率精度均衡版面区域定位模型</td>
+</tr>
+<tr>
+<td>RT-DETR-H_layout_17cls</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/RT-DETR-H_layout_17cls_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/RT-DETR-H_layout_17cls_pretrained.pdparams">训练模型</a></td>
+<td>98.3</td>
+<td>115.29 / 101.18</td>
+<td>964.75 / 964.75</td>
+<td>470.2</td>
+<td>基于RT-DETR-H在中英文论文、杂志和研报等场景上自建数据集训练的高精度版面区域定位模型</td>
+</tr>
+</tbody>
+</table>
+</details>
+</details>
+
+<details>
+<summary> <b>文本图像矫正模块模型（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>CER</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>UVDoc</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/UVDoc_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/UVDoc_pretrained.pdparams">训练模型</a></td>
+<td>0.179</td>
+<td>19.05 / 19.05</td>
+<td>- / 869.82</td>
+<td>30.3</td>
+<td>高精度文本图像矫正模型</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary> <b>文档图像方向分类模块模型（可选）：</b></summary>
+<table>
+<thead>
+<tr>
+<th>模型</th><th>模型下载链接</th>
+<th>Top-1 Acc（%）</th>
+<th>GPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>CPU推理耗时（ms）<br/>[常规模式 / 高性能模式]</th>
+<th>模型存储大小（MB）</th>
+<th>介绍</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>PP-LCNet_x1_0_doc_ori</td>
+<td><a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-LCNet_x1_0_doc_ori_infer.tar">推理模型</a>/<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-LCNet_x1_0_doc_ori_pretrained.pdparams">训练模型</a></td>
+<td>99.06</td>
+<td>2.62 / 0.59</td>
+<td>3.24 / 1.19</td>
+<td>7</td>
+<td>基于PP-LCNet_x1_0的文档图像分类模型，含有四个类别，即0度，90度，180度，270度</td>
+</tr>
+</tbody>
+</table>
+</details>
+
+<details>
+<summary> <b>测试环境说明：</b></summary>
+
+  <ul>
+      <li><b>性能测试环境</b>
+          <ul>
+            <li><strong>测试数据集：
+             </strong>
+                <ul>
+                  <li>文档图像方向分类模型：PaddleOCR 自建的数据集，覆盖证件和文档等多个场景，包含 1000 张图片。</li>
+                  <li>版面区域检测模型：PaddleOCR 自建的版面区域检测数据集，包含中英文论文、杂志、合同、书本、试卷和研报等常见的 500 张文档类型图片。</li>
+                  <li>表格版面检测模型：PaddleOCR 自建的版面表格区域检测数据集，包含中英文 7835 张带有表格的论文文档类型图片。</li>
+                  <li>3类版面检测模型：PaddleOCR 自建的版面区域检测数据集，包含中英文论文、杂志和研报等常见的 1154 张文档类型图片。</li>
+                  <li> 5类英文文档区域检测模型：<a href="https://developer.ibm.com/exchanges/data/all/publaynet">PubLayNet</a> 的评估数据集，包含英文文档的 11245 张图片。</li>
+                  <li>17类区域检测模型：PaddleOCR 自建的版面区域检测数据集，包含中英文论文、杂志和研报等常见的 892 张文档类型图片。</li>
+                  <li>表格结构识别模型：自建的内部高难度中文表格识别数据集。</li>
+                  <li>表格单元格检测模型：自建的内部评测集。</li>
+                  <li>表格分类模型：自建的内部评测集。</li>
+                  <li>文本检测模型：PaddleOCR 自建的中文数据集，覆盖街景、网图、文档、手写多个场景，其中检测包含 500 张图片。</li>
+                  <li>中文识别模型： PaddleOCR 自建的中文数据集，覆盖街景、网图、文档、手写多个场景，其中文本识别包含 1.1w 张图片。</li>
+                  <li>ch_SVTRv2_rec：<a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务</a>A榜评估集。</li>
+                  <li>ch_RepSVTR_rec：<a href="https://aistudio.baidu.com/competition/detail/1131/0/introduction">PaddleOCR算法模型挑战赛 - 赛题一：OCR端到端识别任务</a>B榜评估集。</li>
+                  <li>英文识别模型：PaddleOCR 自建的英文数据集。</li>
+                  <li>多语言识别模型：PaddleOCR 自建的多语种数据集。</li>
+                </ul>
+             </li>
+              <li><strong>硬件配置：</strong>
+                  <ul>
+                      <li>GPU：NVIDIA Tesla T4</li>
+                      <li>CPU：Intel Xeon Gold 6271C @ 2.60GHz</li>
+                  </ul>
+              </li>
+              <li><strong>软件环境：</strong>
+                  <ul>
+                      <li>Ubuntu 20.04 / CUDA 11.8 / cuDNN 8.9 / TensorRT 8.6.1.6</li>
+                      <li>paddlepaddle-gpu 3.0.0 / paddleocr 3.0.3</li>
+                  </ul>
+              </li>
+          </ul>
+      </li>
+      <li><b>推理模式说明</b></li>
+  </ul>
+
+<table border="1">
+    <thead>
+        <tr>
+            <th>模式</th>
+            <th>GPU配置</th>
+            <th>CPU配置</th>
+            <th>加速技术组合</th>
+        </tr>
+    </thead>
+    <tbody>
+        <tr>
+            <td>常规模式</td>
+            <td>FP32精度 / 无TRT加速</td>
+            <td>FP32精度 / 8线程</td>
+            <td><code>paddle_static</code></td>
+        </tr>
+        <tr>
+            <td>高性能模式</td>
+            <td>选择先验精度类型和加速策略的最优组合</td>
+            <td>FP32精度 / 8线程</td>
+            <td>选择先验最优后端（Paddle/OpenVINO/TRT等）</td>
+        </tr>
+    </tbody>
+</table>
+
+</details>
+</details>
+
+<br />
+<b>如果您更注重模型的精度，请选择精度较高的模型；如果您更在意模型的推理速度，请选择推理速度较快的模型；如果您关注模型的存储大小，请选择存储体积较小的模型。</b>
+
+## 2. 快速开始
+
+在本地使用表格结构识别v2产线前，请确保您已经按照[安装教程](../installation.md)完成了wheel包安装。如果您希望选择性安装依赖，请参考安装教程中的相关说明。该产线对应的依赖分组为 `doc-parser`。安装完成后，可以在本地使用命令行体验或 Python 集成。
+
+**请注意，如果在执行过程中遇到程序失去响应、程序异常退出、内存资源耗尽、推理速度极慢等问题，请尝试参考文档调整配置，例如关闭不需要使用的功能或使用更轻量的模型。**
+
+### 2.1 命令行方式体验
+
+一行命令即可快速体验 table_recognition_v2 产线效果：
+
+```bash
+paddleocr table_recognition_v2 -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition_v2.jpg
+
+# 通过 --use_doc_orientation_classify 指定是否使用文档方向分类模型
+paddleocr table_recognition_v2 -i ./table_recognition_v2.jpg --use_doc_orientation_classify True
+
+# 通过 --use_doc_unwarping 指定是否使用文本图像矫正模块
+paddleocr table_recognition_v2 -i ./table_recognition_v2.jpg --use_doc_unwarping True
+
+# 通过 --device 指定模型推理时使用 GPU
+paddleocr table_recognition_v2 -i ./table_recognition_v2.jpg --device gpu
+```
+
+上述命令默认使用本地推理引擎 `paddle_static`。如需运行，请先参考[飞桨框架安装说明](../paddlepaddle_installation.md)安装 PaddlePaddle。
+
+如果选择 `transformers` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 Transformers 环境配置，然后执行如下命令：
+
+```bash
+# 使用 transformers 引擎进行推理
+paddleocr table_recognition_v2 -i https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition_v2.jpg \
+    --engine transformers
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+<details><summary><b>命令行支持更多参数设置，点击展开以查看命令行参数的详细说明</b></summary>
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，必填。<br/>
+<b>说明：</b>
+如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；
+<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition_v2.jpg">示例</a>；
+<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>save_path</code></td>
+<td><b>含义：</b>指定推理结果文件保存的路径。<br/>
+<b>说明：</b>
+如果不设置，推理结果将不会保存到本地。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>含义：</b>版面检测模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>含义：</b>版面检测模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_classification_model_name</code></td>
+<td><b>含义：</b>表格分类模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>table_classification_model_dir</code></td>
+<td><b>含义：</b>表格分类模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_name</code></td>
+<td><b>含义：</b>有线表格结构识别模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_dir</code></td>
+<td><b>含义：</b>有线表格结构识别模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_name</code></td>
+<td><b>含义：</b>无线表格结构识别模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_dir</code></td>
+<td><b>含义：</b>无线表格结构识别模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_name</code></td>
+<td><b>含义：</b>有线表格单元检测模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_dir</code></td>
+<td><b>含义：</b>有线表格单元检测模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_name</code></td>
+<td><b>含义：</b>无线表格单元检测模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_dir</code></td>
+<td><b>含义：</b>无线表格单元检测模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>含义：</b>文档方向分类模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>文档方向分类模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>含义：</b>文本图像矫正模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>含义：</b>文本检测模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>含义：</b>文本检测模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>含义：</b>文本检测的图像边长限制。<br/>
+<b>说明：</b>
+大于 <code>0</code> 的任意整数。如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>960</code>。
+</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>含义：</b>文本检测的图像边长限制类型。<br/>
+<b>说明：</b>
+支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>。如果不设置，将使用产线初始化的该参数值，默认初始化为 <code>max</code>。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>含义：</b>文本检测像素阈值，输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。<br/>
+<b>说明：</b>
+大于<code>0</code>的任意浮点数
+。如果不设置，将默认使用产线初始化的该参数值 <code>0.3</code>。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>含义：</b>文本检测框阈值，检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。<br/>
+<b>说明：</b>
+大于<code>0</code>的任意浮点数
+。如果不设置，将默认使用产线初始化的该参数值 <code>0.6</code>。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>含义：</b>文本检测扩张系数，使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。<br/>
+<b>说明：</b>
+大于<code>0</code>的任意浮点数
+。如果不设置，将默认使用产线初始化的该参数值 <code>2.0</code>。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>含义：</b>文本识别模型的名称。<br/>
+<b>说明：</b>
+如果不设置，将会使用产线默认模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>含义：</b>文本识别模型的目录路径。<br/>
+<b>说明：</b>
+如果不设置，将会下载官方模型。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>含义：</b>文本识别模型的batch size。<br/>
+<b>说明：</b>
+如果不设置，将默认设置batch size为<code>1</code>。</td>
+<td><code>int</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>含义：</b>文本识别阈值，得分大于该阈值的文本结果会被保留。<br/>
+<b>说明：</b>
+大于<code>0</code>的任意浮点数
+。如果不设置，将默认使用产线初始化的该参数值 <code>0.0</code>。即不设阈值。
+</td>
+<td><code>float</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否加载并使用文档方向分类模块。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>含义：</b>是否加载并使用版面检测模块。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_ocr_model</code></td>
+<td><b>含义：</b>是否加载并使用OCR模块。<br/>
+<b>说明：</b>
+如果不设置，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+</ul>如果不设置，将默认使用产线初始化的该参数值，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。
+</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>engine</code></td>
+<td><b>含义：</b>推理引擎。<br><b>说明：</b>支持 <code>None</code>（默认值）、<code>paddle</code>、<code>paddle_static</code>、<code>paddle_dynamic</code>、<code>transformers</code>。保持为默认值 <code>None</code> 时，PaddleOCR 保留旧版本的行为，在大多数配置下等价于 <code>paddle</code>。详细说明、取值、兼容性规则与示例请参见 <a href="../inference_deployment/local_inference/inference_engine.md">推理引擎与配置说明</a>。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>fp32</code>、<code>fp16</code>。</td>
+<td><code>str</code></td>
+<td><code>fp32</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。<br/><b>说明：</b>如需通过配置文件设置 <code>engine_config</code> 等高级参数，可结合该参数使用。详见 <a href="../paddleocr_and_paddlex.md">PaddleOCR 与 PaddleX</a>。</td>
+<td><code>str</code></td>
+<td></td>
+</tr>
+</tbody>
+</table>
+</details>
+<br />
+
+接下来，通过下列一行命令来对[示例图片](https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition_v2.jpg)进行推理：
+```bash
+paddleocr table_recognition_v2 -i ./table_recognition_v2.jpg --use_doc_orientation_classify False --use_doc_unwarping False
+```
+
+运行结果会被打印到终端上，默认配置的 table_recognition_v2 产线的运行结果如下：
+
+```
+{'res': {'input_path': 'table_recognition_v2.jpg', 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_layout_detection': True, 'use_ocr_model': True}, 'layout_det_res': {'input_path': None, 'page_index': None, 'boxes': [{'cls_id': 8, 'label': 'table', 'score': 0.86655592918396, 'coordinate': [0.0125130415, 0.41920784, 1281.3737, 585.3884]}]}, 'overall_ocr_res': {'input_path': None, 'page_index': None, 'model_settings': {'use_doc_preprocessor': False, 'use_textline_orientation': False}, 'dt_polys': array([[[   9,   21],
+        ...,
+        [   9,   59]],
+
+       ...,
+
+       [[1046,  536],
+        ...,
+        [1046,  573]]], dtype=int16), 'text_det_params': {'limit_side_len': 960, 'limit_type': 'max', 'thresh': 0.3, 'box_thresh': 0.6, 'unclip_ratio': 2.0}, 'text_type': 'general', 'textline_orientation_angles': array([-1, ..., -1]), 'text_rec_score_thresh': 0, 'rec_texts': ['部门', '报销人', '报销事由', '批准人：', '单据', '张', '合计金额', '元', '车费票', '其', '火车费票', '飞机票', '中', '旅住宿费', '其他', '补贴'], 'rec_scores': array([0.99958128, ..., 0.99317062]), 'rec_polys': array([[[   9,   21],
+        ...,
+        [   9,   59]],
+
+       ...,
+
+       [[1046,  536],
+        ...,
+        [1046,  573]]], dtype=int16), 'rec_boxes': array([[   9, ...,   59],
+       ...,
+       [1046, ...,  573]], dtype=int16)}, 'table_res_list': [{'cell_box_list': [array([ 0.13052222, ..., 73.08310249]), array([104.43082511, ...,  73.27777413]), array([319.39041221, ...,  73.30439308]), array([424.2436837 , ...,  73.44736794]), array([580.75836265, ...,  73.24003914]), array([723.04370201, ...,  73.22717598]), array([984.67315757, ...,  73.20420387]), array([1.25130415e-02, ..., 5.85419208e+02]), array([984.37072837, ..., 137.02281502]), array([984.26586998, ..., 201.22290352]), array([984.24017417, ..., 585.30775765]), array([1039.90606773, ...,  265.44664314]), array([1039.69549644, ...,  329.30540779]), array([1039.66546714, ...,  393.57319954]), array([1039.5122689 , ...,  457.74644783]), array([1039.55535972, ...,  521.73030403]), array([1039.58612144, ...,  585.09468392])], 'pred_html': '<html><body><table><tbody><tr><td>部门</td><td></td><td>报销人</td><td></td><td>报销事由</td><td></td><td colspan="2">批准人：</td></tr><tr><td colspan="6" rowspan="8"></td><td colspan="2">单据 张</td></tr><tr><td colspan="2">合计金额 元</td></tr><tr><td rowspan="6">其 中</td><td>车费票</td></tr><tr><td>火车费票</td></tr><tr><td>飞机票</td></tr><tr><td>旅住宿费</td></tr><tr><td>其他</td></tr><tr><td>补贴</td></tr></tbody></table></body></html>', 'table_ocr_pred': {'rec_polys': array([[[   9,   21],
+        ...,
+        [   9,   59]],
+
+       ...,
+
+       [[1046,  536],
+        ...,
+        [1046,  573]]], dtype=int16), 'rec_texts': ['部门', '报销人', '报销事由', '批准人：', '单据', '张', '合计金额', '元', '车费票', '其', '火车费票', '飞机票', '中', '旅住宿费', '其他', '补贴'], 'rec_scores': array([0.99958128, ..., 0.99317062]), 'rec_boxes': array([[   9, ...,   59],
+       ...,
+       [1046, ...,  573]], dtype=int16)}}]}}
+```
+
+可视化结果保存在`save_path`下，可视化结果如下：
+
+<img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/pipelines/table_recognition_v2/02.jpg">
+
+### 2.2 Python脚本方式集成
+
+命令行方式是为了快速体验查看效果，一般来说，在项目中，往往需要通过代码集成，您可以通过几行代码即可完成产线的快速推理，推理代码如下：
+
+```python
+from paddleocr import TableRecognitionPipelineV2
+
+pipeline = TableRecognitionPipelineV2()
+# ocr = TableRecognitionPipelineV2(use_doc_orientation_classify=True) # 通过 use_doc_orientation_classify 指定是否使用文档方向分类模型
+# ocr = TableRecognitionPipelineV2(use_doc_unwarping=True) # 通过 use_doc_unwarping 指定是否使用文本图像矫正模块
+# ocr = TableRecognitionPipelineV2(device="gpu") # 通过 device 指定模型推理时使用 GPU
+output = pipeline.predict("./table_recognition_v2.jpg")
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_img("./output/")
+    res.save_to_xlsx("./output/")
+    res.save_to_html("./output/")
+    res.save_to_json("./output/")
+```
+
+上述代码默认使用本地推理引擎 `paddle_static`。如需运行，请先参考[飞桨框架安装说明](../paddlepaddle_installation.md)安装 PaddlePaddle。
+
+如果选择 `transformers` 作为推理引擎，请先参考[推理引擎文档](../inference_deployment/local_inference/inference_engine.md)完成 Transformers 环境配置，然后执行如下代码：
+
+```python
+from paddleocr import TableRecognitionPipelineV2
+
+pipeline = TableRecognitionPipelineV2(
+    engine="transformers",
+)
+# ocr = TableRecognitionPipelineV2(use_doc_orientation_classify=True) # 通过 use_doc_orientation_classify 指定是否使用文档方向分类模型
+# ocr = TableRecognitionPipelineV2(use_doc_unwarping=True) # 通过 use_doc_unwarping 指定是否使用文本图像矫正模块
+# ocr = TableRecognitionPipelineV2(device="gpu") # 通过 device 指定模型推理时使用 GPU
+output = pipeline.predict("./table_recognition_v2.jpg")
+for res in output:
+    res.print() ## 打印预测的结构化输出
+    res.save_to_img("./output/")
+    res.save_to_xlsx("./output/")
+    res.save_to_html("./output/")
+    res.save_to_json("./output/")
+```
+
+在大多数场景下，默认的 `paddle_static` 推理引擎通常具备更好的推理性能，建议优先使用。
+
+在上述 Python 脚本中，执行了如下几个步骤：
+
+（1）通过 <code>TableRecognitionPipelineV2()</code> 实例化通用表格识别v2产线对象，具体参数说明如下：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>layout_detection_model_name</code></td>
+<td><b>含义：</b>版面检测模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>layout_detection_model_dir</code></td>
+<td><b>含义：</b>版面检测模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_classification_model_name</code></td>
+<td><b>含义：</b>表格分类模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>table_classification_model_dir</code></td>
+<td><b>含义：</b>表格分类模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_name</code></td>
+<td><b>含义：</b>有线表格结构识别模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_structure_recognition_model_dir</code></td>
+<td><b>含义：</b>有线表格结构识别模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_name</code></td>
+<td><b>含义：</b>无线表格结构识别模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_structure_recognition_model_dir</code></td>
+<td><b>含义：</b>无线表格结构识别模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_name</code></td>
+<td><b>含义：</b>有线表格单元检测模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wired_table_cells_detection_model_dir</code></td>
+<td><b>含义：</b>有线表格单元检测模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_name</code></td>
+<td><b>含义：</b>无线表格单元检测模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>wireless_table_cells_detection_model_dir</code></td>
+<td><b>含义：</b>无线表格单元检测模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_name</code></td>
+<td><b>含义：</b>文档方向分类模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_orientation_classify_model_dir</code></td>
+<td><b>含义：</b>文档方向分类模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_name</code></td>
+<td><b>含义：</b>文本图像矫正模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>doc_unwarping_model_dir</code></td>
+<td><b>含义：</b>文本图像矫正模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_name</code></td>
+<td><b>含义：</b>文本检测模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_detection_model_dir</code></td>
+<td><b>含义：</b>文本检测模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>含义：</b>文本检测的图像边长限制。<br/>
+<b>说明：</b>
+<ul>
+<li><b>int</b>：大于 <code>0</code> 的任意整数；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>960</code>。</li>
+</ul>
+</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>含义：</b>文本检测的图像边长限制类型。<br/>
+<b>说明：</b>
+<ul>
+<li><b>str</b>：支持 <code>min</code> 和 <code>max</code>，<code>min</code> 表示保证图像最短边不小于 <code>det_limit_side_len</code>，<code>max</code> 表示保证图像最长边不大于 <code>limit_side_len</code>；</li>
+<li><b>None</b>：如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为 <code>max</code>。</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>含义：</b>文本检测像素阈值，输出的概率图中，得分大于该阈值的像素点才会被认为是文字像素点。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于 <code>0</code> 的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.3</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>含义：</b>文本检测框阈值，检测结果边框内，所有像素点的平均得分大于该阈值时，该结果会被认为是文字区域。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.6</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>含义：</b>文本检测扩张系数，使用该方法对文字区域进行扩张，该值越大，扩张的面积越大。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>2.0</code>。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_name</code></td>
+<td><b>含义：</b>文本识别模型的名称。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会使用产线默认模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_model_dir</code></td>
+<td><b>含义：</b>文本识别模型的目录路径。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将会下载官方模型。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_recognition_batch_size</code></td>
+<td><b>含义：</b>文本识别模型的batch size。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将默认设置batch size为<code>1</code>。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>含义：</b>文本识别阈值，得分大于该阈值的文本结果会被保留。<br/>
+<b>说明：</b>
+<ul>
+<li><b>float</b>：大于<code>0</code>的任意浮点数；
+<li><b>None</b>：如果设置为<code>None</code>，将默认使用产线初始化的该参数值 <code>0.0</code>，即不设阈值。</li></li></ul>
+</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b></b>含义：</b>是否加载并使用文档方向分类模块。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否加载并使用文本图像矫正模块。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>含义：</b>是否加载并使用版面检测模块。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_ocr_model</code></td>
+<td><b>含义：</b>是否加载并使用OCR模块。<br/>
+<b>说明：</b>
+如果设置为<code>None</code>，将使用产线初始化的该参数值，默认初始化为<code>True</code>。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>device</code></td>
+<td><b>含义：</b>用于推理的设备。<br/>
+<b>说明：</b>
+支持指定具体卡号：
+<ul>
+<li><b>CPU</b>：如 <code>cpu</code> 表示使用 CPU 进行推理；</li>
+<li><b>GPU</b>：如 <code>gpu:0</code> 表示使用第 1 块 GPU 进行推理；</li>
+<li><b>NPU</b>：如 <code>npu:0</code> 表示使用第 1 块 NPU 进行推理；</li>
+<li><b>XPU</b>：如 <code>xpu:0</code> 表示使用第 1 块 XPU 进行推理；</li>
+<li><b>MLU</b>：如 <code>mlu:0</code> 表示使用第 1 块 MLU 进行推理；</li>
+<li><b>DCU</b>：如 <code>dcu:0</code> 表示使用第 1 块 DCU 进行推理；</li>
+<li><b>沐曦 GPU</b>：如 <code>metax_gpu:0</code> 表示使用第 1 块沐曦 GPU 进行推理；</li>
+<li><b>天数 GPU</b>：如 <code>iluvatar_gpu:0</code> 表示使用第 1 块天数 GPU 进行推理；</li>
+<li><b>None</b>：如果设置为<code>None</code>，初始化时，会优先使用本地的 GPU 0号设备，如果没有，则使用 CPU 设备。</li>
+</ul>
+</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>enable_hpi</code></td>
+<td><b>含义：</b>是否启用高性能推理。</td>
+<td><code>bool</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_tensorrt</code></td>
+<td><b>含义：</b>是否启用 Paddle Inference 的 TensorRT 子图引擎。<br/>
+<b>说明：</b>
+如果模型不支持通过 TensorRT 加速，即使设置了此标志，也不会使用加速。<br/>
+对于 CUDA 11.8 版本的飞桨，兼容的 TensorRT 版本为 8.x（x>=6），建议安装 TensorRT 8.6.1.6。<br/>
+
+</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>precision</code></td>
+<td><b>含义：</b>计算精度，如 <code>"fp32"</code>、<code>"fp16"</code>。</td>
+<td><code>str</code></td>
+<td><code>"fp32"</code></td>
+</tr>
+<tr>
+<td><code>enable_mkldnn</code></td>
+<td><b>含义：</b>是否启用 MKL-DNN 加速推理。<br/>
+<b>说明：</b>
+如果 MKL-DNN 不可用或模型不支持通过 MKL-DNN 加速，即使设置了此标志，也不会使用加速。
+</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>mkldnn_cache_capacity</code></td>
+<td>
+<b>含义：</b>MKL-DNN 缓存容量。
+</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>cpu_threads</code></td>
+<td><b>含义：</b>在 CPU 上进行推理时使用的线程数。</td>
+<td><code>int</code></td>
+<td><code>10</code></td>
+</tr>
+<tr>
+<td><code>paddlex_config</code></td>
+<td><b>含义：</b>PaddleX产线配置文件路径。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+</tbody>
+</table>
+
+（2）调用通用表格识别v2产线对象的 <code>predict()</code> 方法进行推理预测，该方法会返回一个结果列表。
+
+另外，产线还提供了 <code>predict_iter()</code> 方法。两者在参数接受和结果返回方面是完全一致的，区别在于 <code>predict_iter()</code> 返回的是一个 <code>generator</code>，能够逐步处理和获取预测结果，适合处理大型数据集或希望节省内存的场景。可以根据实际需求选择使用这两种方法中的任意一种。
+
+以下是 <code>predict()</code> 方法的参数及其说明：
+
+<table>
+<thead>
+<tr>
+<th>参数</th>
+<th>参数说明</th>
+<th>参数类型</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td><code>input</code></td>
+<td><b>含义：</b>待预测数据，支持多种输入类型，必填。<br/>
+<b>说明：</b>
+<ul>
+<li><b>Python Var</b>：如 <code>numpy.ndarray</code> 表示的图像数据；</li>
+<li><b>str</b>：如图像文件或者PDF文件的本地路径：<code>/root/data/img.jpg</code>；
+<b>如URL链接</b>，如图像文件或PDF文件的网络URL：<a href="https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/table_recognition_v2.jpg">示例</a>；
+<b>如本地目录</b>，该目录下需包含待预测图像，如本地路径：<code>/root/data/</code>(当前不支持目录中包含PDF文件的预测，PDF文件需要指定到具体文件路径)；</li>
+<li><b>list</b>：列表元素需为上述类型数据，如<code>[numpy.ndarray, numpy.ndarray]</code>，<code>["/root/data/img1.jpg", "/root/data/img2.jpg"]</code>，<code>["/root/data1", "/root/data2"]。</code></li>
+</ul>
+</td>
+<td><code>Python Var|str|list</code></td>
+<td></td>
+</tr>
+<tr>
+<td><code>use_doc_orientation_classify</code></td>
+<td><b>含义：</b>是否在推理时使用文档方向分类模块。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_doc_unwarping</code></td>
+<td><b>含义：</b>是否在推理时使用文本图像矫正模块。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_layout_detection</code></td>
+<td><b>含义：</b>是否在推理时使用版面检测模块。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_ocr_model</code></td>
+<td><b>含义：</b>是否在推理时使用OCR模型。</td>
+<td><code>bool|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_side_len</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>int|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_limit_type</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>str|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_box_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_det_unclip_ratio</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>text_rec_score_thresh</code></td>
+<td><b>含义：</b>参数含义与实例化参数基本相同。<br/>
+<b>说明：</b>
+设置为<code>None</code>表示使用实例化参数，否则该参数优先级更高。</td>
+<td><code>float|None</code></td>
+<td><code>None</code></td>
+</tr>
+<tr>
+<td><code>use_e2e_wired_table_rec_model</code></td>
+<td><b>含义：</b>是否在推理时使用有线表端到端表格识别模式。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_e2e_wireless_table_rec_model</code></td>
+<td><b>含义：</b>是否在推理时使用无线表端到端表格识别模式。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_wired_table_cells_trans_to_html</code></td>
+<td><b>含义：</b>是否在推理时使用有线表单元格检测结果直转HTML模式。<br/>
+<b>说明：</b>
+启用则直接基于有线表单元格检测结果的几何关系构建HTML。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_wireless_table_cells_trans_to_html</code></td>
+<td><b>含义：</b>是否在推理时使用无线表单元格检测结果直转HTML模式。<br/>
+<b>说明：</b>
+启用则直接基于无线表单元格检测结果的几何关系构建HTML。</td>
+<td><code>bool</code></td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>use_table_orientation_classify</code></td>
+<td><b>含义：</b>是否在推理时使用表格方向分类模式。<br/>
+<b>说明：</b>
+启用时当图像中的表格存在90/180/270度旋转时，能够将方向校正并正确完成表格识别。</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>use_ocr_results_with_table_cells</code></td>
+<td><b>含义：</b>是否在推理时使用单元格切分OCR模式。<br/>
+<b>说明：</b>
+启用时会基于单元格预测结果对OCR检测结果进行切分和重识别，避免出现文字缺失情况。</td>
+<td><code>bool</code></td>
+<td><code>True</code></td>
+</tr>
+</table>
+
+（3）对预测结果进行处理，每个样本的预测结果均为对应的Result对象，且支持打印、保存为图片、保存为<code>xlsx</code>文件、保存为<code>HTML</code>文件、保存为<code>json</code>文件的操作:
+
+<table>
+<thead>
+<tr>
+<th>方法</th>
+<th>方法说明</th>
+<th>参数</th>
+<th>参数类型</th>
+<th>参数说明</th>
+<th>默认值</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="3"><code>print()</code></td>
+<td rowspan="3">打印结果到终端</td>
+<td><code>format_json</code></td>
+<td><code>bool</code></td>
+<td>是否对输出内容进行使用 <code>JSON</code> 缩进格式化。</td>
+<td><code>True</code></td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效。</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效。</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td rowspan="3"><code>save_to_json()</code></td>
+<td rowspan="3">将结果保存为json格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，当为目录时，保存文件命名与输入文件类型命名一致。</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>indent</code></td>
+<td><code>int</code></td>
+<td>指定缩进级别，以美化输出的 <code>JSON</code> 数据，使其更具可读性，仅当 <code>format_json</code> 为 <code>True</code> 时有效。</td>
+<td>4</td>
+</tr>
+<tr>
+<td><code>ensure_ascii</code></td>
+<td><code>bool</code></td>
+<td>控制是否将非 <code>ASCII</code> 字符转义为 <code>Unicode</code>。设置为 <code>True</code> 时，所有非 <code>ASCII</code> 字符将被转义；<code>False</code> 则保留原始字符，仅当<code>format_json</code>为<code>True</code>时有效。</td>
+<td><code>False</code></td>
+</tr>
+<tr>
+<td><code>save_to_img()</code></td>
+<td>将结果保存为图像格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径。</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>save_to_xlsx()</code></td>
+<td>将结果保存为xlsx格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径。</td>
+<td>无</td>
+</tr>
+<tr>
+<td><code>save_to_html()</code></td>
+<td>将结果保存为html格式的文件</td>
+<td><code>save_path</code></td>
+<td><code>str</code></td>
+<td>保存的文件路径，支持目录或文件路径。</td>
+<td>无</td>
+</tr>
+</table>
+
+<ul>
+<li>调用<code>print()</code> 方法会将结果打印到终端，打印到终端的内容解释如下：
+    <ol start="1" type="1">
+    <li><code>input_path</code>: <code>(str)</code> 待预测图像的输入路径</li>
+    <li><code>page_index</code>: <code>(Union[int, None])</code> 如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+    <li><code>model_settings</code>: <code>(Dict[str, bool])</code> 配置产线所需的模型参数
+        <ol>
+        <li><code>use_doc_preprocessor</code>: <code>(bool)</code> 控制是否启用文档预处理子产线</li>
+        <li><code>use_layout_detection</code>: <code>(bool)</code> 控制是否启用版面区域检测子产线</li>
+        <li><code>use_ocr_model</code>: <code>(bool)</code> 控制是否启用OCR子产线</li>
+        </ol>
+    </li>
+    <li><code>layout_det_res</code>: <code>(Dict[str, Union[List[numpy.ndarray], List[float]]])</code> 版面检测子模块的输出结果。仅当<code>use_layout_detection=True</code>时存在
+        <ol>
+        <li><code>input_path</code>: <code>(Union[str, None])</code> 版面检测区域模块接受的图像路径，当输入为<code>numpy.ndarray</code>时，保存为<code>None</code></li>
+        <li><code>page_index</code>: <code>(Union[int, None])</code> 如果输入是PDF文件，则表示当前是PDF的第几页，否则为 <code>None</code></li>
+        <li><code>boxes</code>: <code>(List[Dict])</code> 版面印章区域的检测框列表，每个列表中的元素，包含以下字段
+            <ul>
+            <li><code>cls_id</code>: <code>(int)</code> 检测框的印章类别id</li>
+            <li><code>score</code>: <code>(float)</code> 检测框的置信度</li>
+            <li><code>coordinate</code>: <code>(List[float])</code> 检测框的四个顶点坐标，顺序为x1，y1，x2，y2表示左上角的x坐标，左上角的y坐标，右下角x坐标，右下角的y坐标</li>
+            </ul>
+        </li>
+        </ol>
+    </li>
+    <li><code>doc_preprocessor_res</code>: <code>(Dict[str, Union[str, Dict[str, bool], int]])</code> 文档预处理子产线的输出结果。仅当<code>use_doc_preprocessor=True</code>时存在
+        <ol>
+        <li><code>input_path</code>: <code>(Union[str, None])</code> 图像预处理子产线接受的图像路径，当输入为<code>numpy.ndarray</code>时，保存为<code>None</code></li>
+        <li><code>model_settings</code>: <code>(Dict)</code> 预处理子产线的模型配置参数</li>
+        <li><code>model_settings</code>: <code>(Dict)</code> 预处理子产线的模型配置参数
+            <ul>
+            <li><code>use_doc_orientation_classify</code>: <code>(bool)</code> 控制是否启用文档方向分类</li>
+            <li><code>use_doc_unwarping</code>: <code>(bool)</code> 控制是否启用文本图像矫正</li>
+            </ul>
+        </li>
+        <li><code>angle</code>: <code>(int)</code> 文档方向分类的预测结果。启用时取值为[0,1,2,3]，分别对应[0°,90°,180°,270°]；未启用时为-1</li>
+        </ol>
+    </li>
+    <li><code>dt_polys</code>: <code>(List[numpy.ndarray])</code> 文本检测的多边形框列表。每个检测框由4个顶点坐标构成的numpy数组表示，数组shape为(4, 2)，数据类型为int16</li>
+    <li><code>dt_scores</code>: <code>(List[float])</code> 文本检测框的置信度列表</li>
+    <li><code>text_det_params</code>: <code>(Dict[str, Dict[str, int, float]])</code> 文本检测模块的配置参数
+        <ol>
+        <li><code>limit_side_len</code>: <code>(int)</code> 图像预处理时的边长限制值</li>
+        <li><code>limit_type</code>: <code>(str)</code> 边长限制的处理方式</li>
+        <li><code>thresh</code>: <code>(float)</code> 文本像素分类的置信度阈值</li>
+        <li><code>box_thresh</code>: <code>(float)</code> 文本检测框的置信度阈值</li>
+        <li><code>unclip_ratio</code>: <code>(float)</code> 文本检测框的膨胀系数</li>
+        <li><code>text_type</code>: <code>(str)</code> 文本检测的类型，当前固定为"general"</li>
+        </ol>
+    </li>
+    <li><code>text_rec_score_thresh</code>: <code>(float)</code> 文本识别结果的过滤阈值</li>
+    <li><code>rec_texts</code>: <code>(List[str])</code> 文本识别结果列表，仅包含置信度超过<code>text_rec_score_thresh</code>的文本</li>
+    <li><code>rec_scores</code>: <code>(List[float])</code> 文本识别的置信度列表，已按<code>text_rec_score_thresh</code>过滤</li>
+    <li><code>rec_polys</code>: <code>(List[numpy.ndarray])</code> 经过置信度过滤的文本检测框列表，格式同<code>dt_polys</code></li>
+    <li><code>rec_boxes</code>: <code>(numpy.ndarray)</code> 检测框的矩形边界框数组，shape为(n, 4)，dtype为int16。每一行表示一个矩形框的[x_min, y_min, x_max, y_max]坐标
+    ，其中(x_min, y_min)为左上角坐标，(x_max, y_max)为右下角坐标</li>
+    </ol>
+</li>
+<li>调用<code>save_to_json()</code> 方法会将上述内容保存到指定的<code>save_path</code>中，如果指定为目录，则保存的路径为<code>save_path/{your_img_basename}_res.json</code>，如果指定为文件，则直接保存到该文件中。由于json文件不支持保存numpy数组，因此会将其中的<code>numpy.array</code>类型转换为列表形式。</li>
+<li>调用<code>save_to_img()</code> 方法会将可视化结果保存到指定的<code>save_path</code>中，如果指定为目录，则保存的路径为<code>save_path/{your_img_basename}_ocr_res_img.{your_img_extension}</code>，如果指定为文件，则直接保存到该文件中。(产线通常包含较多结果图片，不建议直接指定为具体的文件路径，否则多张图会被覆盖，仅保留最后一张图)</li>
+<li>调用<code>save_to_html()</code> 方法会将上述内容保存到指定的<code>save_path</code>中，如果指定为目录，则保存的路径为<code>save_path/{your_img_basename}_table_1.html</code>，如果指定为文件，则直接保存到该文件中。在通用表格识别v2产线中，将会把图像中表格的HTML形式写入到指定的html文件中。</li>
+<li>调用<code>save_to_xlsx()</code> 方法会将上述内容保存到指定的<code>save_path</code>中，如果指定为目录，则保存的路径为<code>save_path/{your_img_basename}_res.xlsx</code>，如果指定为文件，则直接保存到该文件中。在通用表格识别v2产线中，将会把图像中表格的Excel表格形式写入到指定的xlsx文件中。</li>
+
+* 此外，也支持通过属性获取带结果的可视化图像和预测结果，具体如下：
+
+<table>
+<thead>
+<tr>
+<th>属性</th>
+<th>属性说明</th>
+</tr>
+</thead>
+<tr>
+<td rowspan="1"><code>json</code></td>
+<td rowspan="1">获取预测的 <code>json</code> 格式的结果</td>
+</tr>
+<tr>
+<td rowspan="2"><code>img</code></td>
+<td rowspan="2">获取格式为 <code>dict</code> 的可视化图像</td>
+</tr>
+</table>
+
+<li><code>json</code> 属性获取的预测结果为dict类型的数据，相关内容与调用 <code>save_to_json()</code> 方法保存的内容一致。</li>
+<li><code>img</code> 属性返回的预测结果是一个dict类型的数据。其中，键分别为 <code>table_res_img</code>、<code>ocr_res_img</code> 、<code>layout_res_img</code> 和 <code>preprocessed_img</code>，对应的值是四个 <code>Image.Image</code> 对象，按顺序分别为：表格识别结果的可视化图像、OCR 结果的可视化图像、版面区域检测结果的可视化图像、图像预处理的可视化图像。如果没有使用某个子模块，则dict中不包含对应的结果图像。</li>
+</ul>
+
+## 3. 开发集成/部署
+
+如果产线可以达到您对产线推理速度和精度的要求，您可以直接进行开发集成/部署。
+
+若您需要将产线直接应用在您的Python项目中，可以参考 [2.2 Python脚本方式集成](#22-python)中的示例代码。
+
+此外，PaddleOCR 也提供了其他两种部署方式，详细说明如下：
+
+🚀 高性能推理：在实际生产环境中，许多应用对部署策略的性能指标（尤其是响应速度）有着较严苛的标准，以确保系统的高效运行与用户体验的流畅性。为此，PaddleOCR 提供高性能推理功能，旨在对模型推理及前后处理进行深度性能优化，实现端到端流程的显著提速，详细的高性能推理流程请参考[高性能推理](../inference_deployment/local_inference/high_performance_inference.md)。
+
+☁️ 服务化部署：服务化部署是实际生产环境中常见的一种部署形式。通过将推理功能封装为服务，客户端可以通过网络请求来访问这些服务，以获取推理结果。详细的产线服务化部署流程请参考[服务化部署](../inference_deployment/serving/serving.md)。
+
+以下是基础服务化部署的API参考与多语言服务调用示例：
+
+
+<details><summary>API参考</summary>
+<p>对于服务提供的主要操作：</p>
+<ul>
+<li>HTTP请求方法为POST。</li>
+<li>请求体和响应体均为JSON数据（JSON对象）。</li>
+<li>当请求处理成功时，响应状态码为<code>200</code>，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。固定为<code>0</code>。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。固定为<code>"Success"</code>。</td>
+</tr>
+<tr>
+<td><code>result</code></td>
+<td><code>object</code></td>
+<td>操作结果。</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>当请求处理未成功时，响应体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>logId</code></td>
+<td><code>string</code></td>
+<td>请求的UUID。</td>
+</tr>
+<tr>
+<td><code>errorCode</code></td>
+<td><code>integer</code></td>
+<td>错误码。与响应状态码相同。</td>
+</tr>
+<tr>
+<td><code>errorMsg</code></td>
+<td><code>string</code></td>
+<td>错误说明。</td>
+</tr>
+</tbody>
+</table>
+<p>服务提供的主要操作如下：</p>
+<ul>
+<li><b><code>infer</code></b></li>
+</ul>
+<p>定位并识别图中的表格。</p>
+<p><code>POST /table-recognition</code></p>
+<ul>
+<li>请求体的属性如下：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+<th>是否必填</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>file</code></td>
+<td><code>string</code></td>
+<td>服务器可访问的图像文件（含 TIFF，多页时按页处理）或 PDF 文件的 URL，或上述类型文件内容的 Base64 编码结果。默认对于超过 10 页的 PDF 或多页 TIFF，仅处理前 10 页。<br /> 要解除页数限制，请在产线配置文件中添加以下配置：
+<pre><code>Serving:
+  extra:
+    max_num_input_imgs: null
+</code></pre>
+</td>
+<td>是</td>
+</tr>
+<tr>
+<td><code>fileType</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>文件类型。<code>0</code> 表示 PDF 文件，<code>1</code> 表示图像文件（含 TIFF）。若请求体无此属性，则将根据URL推断文件类型。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocOrientationClassify</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_doc_orientation_classify</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useDocUnwarping</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_doc_unwarping</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useLayoutDetection</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_layout_detection</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useOcrModel</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_ocr_model</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetLimitSideLen</code></td>
+<td><code>integer</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_det_limit_side_len</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetLimitType</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_det_limit_type</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_det_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetBoxThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_det_box_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textDetUnclipRatio</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_det_unclip_ratio</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>textRecScoreThresh</code></td>
+<td><code>number</code> | <code>null</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>text_rec_score_thresh</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useTableCellsOcrResults</code></td>
+<td><code>boolean</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_table_cells_ocr_results</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useE2eWiredTableRecModel</code></td>
+<td><code>boolean</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_e2e_wired_table_rec_model</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>useE2eWirelessTableRecModel</code></td>
+<td><code>boolean</code></td>
+<td>请参阅产线对象中 <code>predict</code> 方法的 <code>use_e2e_wireless_table_rec_model</code> 参数相关说明。</td>
+<td>否</td>
+</tr>
+<tr>
+<td><code>visualize</code></td>
+<td><code>boolean</code> | <code>null</code></td>
+<td>是否返回可视化结果图以及处理过程中的中间图像等。
+<ul style="margin: 0 0 0 1em; padding-left: 0em;">
+<li>传入 <code>true</code>：返回图像。</li>
+<li>传入 <code>false</code>：不返回图像。</li>
+<li>若请求体中未提供该参数或传入 <code>null</code>：遵循产线配置文件<code>Serving.visualize</code> 的设置。</li>
+</ul>
+<br/>例如，在产线配置文件中添加如下字段：<br/>
+<pre><code>Serving:
+  visualize: False
+</code></pre>
+将默认不返回图像，通过请求体中的<code>visualize</code>参数可以覆盖默认行为。如果请求体和配置文件中均未设置（或请求体传入<code>null</code>、配置文件中未设置），则默认返回图像。
+</td>
+<td>否</td>
+</tr>
+</tbody>
+</table>
+<ul>
+<li>请求处理成功时，响应体的<code>result</code>具有如下属性：</li>
+</ul>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>tableRecResults</code></td>
+<td><code>object</code></td>
+<td>表格识别结果。数组长度为1（对于图像输入）或实际处理的文档页数（对于PDF输入）。对于PDF输入，数组中的每个元素依次表示PDF文件中实际处理的每一页的结果。</td>
+</tr>
+<tr>
+<td><code>dataInfo</code></td>
+<td><code>object</code></td>
+<td>输入数据信息。</td>
+</tr>
+</tbody>
+</table>
+<p>下表中涉及图像的字段（如 <code>outputImages</code>、<code>inputImage</code>）默认以 Base64 字符串内联返回；当服务端开启 URL 返回模式时，相应字段的值变为预签名 URL，字段类型保持不变。配置方式参见 <a href="../inference_deployment/serving/serving.md">服务化部署</a>「以 URL 形式返回二进制内容」一节。</p>
+<p><code>tableRecResults</code>中的每个元素为一个<code>object</code>，具有如下属性：</p>
+<table>
+<thead>
+<tr>
+<th>名称</th>
+<th>类型</th>
+<th>含义</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td><code>prunedResult</code></td>
+<td><code>object</code></td>
+<td>产线对象的 <code>predict</code> 方法生成结果的 JSON 表示中 <code>res</code> 字段的简化版本，其中去除了 <code>input_path</code> 和 <code>page_index</code> 字段。</td>
+</tr>
+<tr>
+<td><code>outputImages</code></td>
+<td><code>object</code> | <code>null</code></td>
+<td>参见产线预测结果的 <code>img</code> 属性说明。图像为JPEG格式，使用Base64编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+<tr>
+<td><code>inputImage</code></td>
+<td><code>string</code> | <code>null</code></td>
+<td>输入图像。图像为JPEG格式，使用Base64编码；启用 URL 返回模式时为预签名 URL。</td>
+</tr>
+</tbody>
+</table></details>
+<details><summary>多语言调用服务示例</summary>
+<details>
+<summary>Python</summary>
+
+<pre><code class="language-python">import base64
+import requests
+
+API_URL = "http://localhost:8080/table-recognition"
+file_path = "./demo.jpg"
+
+with open(file_path, "rb") as file:
+    file_bytes = file.read()
+    file_data = base64.b64encode(file_bytes).decode("ascii")
+
+payload = {"file": file_data, "fileType": 1}
+
+response = requests.post(API_URL, json=payload)
+
+assert response.status_code == 200
+result = response.json()["result"]
+for i, res in enumerate(result["tableRecResults"]):
+    print(res["prunedResult"])
+    for img_name, img in res["outputImages"].items():
+        img_path = f"{img_name}_{i}.jpg"
+        with open(img_path, "wb") as f:
+            f.write(base64.b64decode(img))
+        print(f"Output image saved at {img_path}")
+</code></pre></details>
+
+<details><summary>C++</summary>
+
+<pre><code class="language-cpp">#include &lt;iostream&gt;
+#include &lt;fstream&gt;
+#include &lt;vector&gt;
+#include &lt;string&gt;
+#include "cpp-httplib/httplib.h" // https://github.com/Huiyicc/cpp-httplib
+#include "nlohmann/json.hpp" // https://github.com/nlohmann/json
+#include "base64.hpp" // https://github.com/tobiaslocker/base64
+
+int main() {
+    httplib::Client client("localhost", 8080);
+    const std::string filePath = "./demo.jpg";
+    std::ifstream file(filePath, std::ios::binary | std::ios::ate);
+    if (!file) {
+        std::cerr << "Error opening file." << std::endl;
+        return 1;
+    }
+
+    std::streamsize size = file.tellg();
+    file.seekg(0, std::ios::beg);
+    std::vector<char> buffer(size);
+
+    if (!file.read(buffer.data(), size)) {
+        std::cerr << "Error reading file." << std::endl;
+        return 1;
+    }
+
+    std::string bufferStr(buffer.data(), static_cast<size_t>(size));
+    std::string encodedFile = base64::to_base64(bufferStr);
+
+    nlohmann::json jsonObj;
+    jsonObj["file"] = encodedFile;
+    jsonObj["fileType"] = 1;
+
+    auto response = client.Post("/table-recognition", jsonObj.dump(), "application/json");
+
+    if (response && response->status == 200) {
+        nlohmann::json jsonResponse = nlohmann::json::parse(response->body);
+        auto result = jsonResponse["result"];
+
+        if (!result.is_object() || !result["tableRecResults"].is_array()) {
+            std::cerr << "Unexpected response structure." << std::endl;
+            return 1;
+        }
+
+        for (size_t i = 0; i < result["tableRecResults"].size(); ++i) {
+            auto tableRecResult = result["tableRecResults"][i];
+            std::cout << tableRecResult["prunedResult"] << std::endl;
+
+            if (tableRecResult["outputImages"].is_object()) {
+                for (auto& img : tableRecResult["outputImages"].items()) {
+                    std::string imgName = img.key();
+                    std::string encodedImage = img.value();
+                    std::string decodedImage = base64::from_base64(encodedImage);
+
+                    std::string imgPath = imgName + "_" + std::to_string(i) + ".jpg";
+                    std::ofstream outputImage(imgPath, std::ios::binary);
+                    if (outputImage.is_open()) {
+                        outputImage.write(decodedImage.c_str(), static_cast<std::streamsize>(decodedImage.size()));
+                        outputImage.close();
+                        std::cout << "Output image saved at " << imgPath << std::endl;
+                    } else {
+                        std::cerr << "Unable to open file for writing: " << imgPath << std::endl;
+                    }
+                }
+            }
+        }
+    } else {
+        std::cerr << "Failed to send HTTP request." << std::endl;
+        if (response) {
+            std::cerr << "HTTP status code: " << response->status << std::endl;
+            std::cerr << "Response body: " << response->body << std::endl;
+        }
+        return 1;
+    }
+
+    return 0;
+}
+</code></pre></details>
+
+<details><summary>Java</summary>
+
+<pre><code class="language-java">import okhttp3.*;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.JsonNode;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+
+import java.io.File;
+import java.io.FileOutputStream;
+import java.io.IOException;
+import java.util.Base64;
+
+public class Main {
+    public static void main(String[] args) throws IOException {
+        String API_URL = "http://localhost:8080/table-recognition";
+        String imagePath = "./demo.jpg";
+
+        File file = new File(imagePath);
+        byte[] fileContent = java.nio.file.Files.readAllBytes(file.toPath());
+        String base64Image = Base64.getEncoder().encodeToString(fileContent);
+
+        ObjectMapper objectMapper = new ObjectMapper();
+        ObjectNode payload = objectMapper.createObjectNode();
+        payload.put("file", base64Image);
+        payload.put("fileType", 1);
+
+        OkHttpClient client = new OkHttpClient();
+        MediaType JSON = MediaType.get("application/json; charset=utf-8");
+        RequestBody body = RequestBody.create(JSON, payload.toString());
+
+        Request request = new Request.Builder()
+                .url(API_URL)
+                .post(body)
+                .build();
+
+        try (Response response = client.newCall(request).execute()) {
+            if (response.isSuccessful()) {
+                String responseBody = response.body().string();
+                JsonNode root = objectMapper.readTree(responseBody);
+                JsonNode result = root.get("result");
+
+                JsonNode tableRecResults = result.get("tableRecResults");
+                for (int i = 0; i < tableRecResults.size(); i++) {
+                    JsonNode item = tableRecResults.get(i);
+
+                    JsonNode prunedResult = item.get("prunedResult");
+                    System.out.println("Pruned Result [" + i + "]: " + prunedResult.toString());
+
+                    JsonNode outputImages = item.get("outputImages");
+
+                    outputImages.fieldNames().forEachRemaining(imgName -> {
+                        String imgBase64 = outputImages.get(imgName).asText();
+                        byte[] imgBytes = Base64.getDecoder().decode(imgBase64);
+                        String imgPath = "output_" + imgName +  ".jpg";
+                        try (FileOutputStream fos = new FileOutputStream(imgPath)) {
+                            fos.write(imgBytes);
+                            System.out.println("Saved image to: " + imgPath);
+                        } catch (IOException e) {
+                            e.printStackTrace();
+                        }
+                    });
+                }
+            } else {
+                System.err.println("Request failed with HTTP code: " + response.code());
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Go</summary>
+
+<pre><code class="language-go">package main
+
+import (
+    "bytes"
+    "encoding/base64"
+    "encoding/json"
+    "fmt"
+    "io/ioutil"
+    "net/http"
+)
+
+func main() {
+    API_URL := "http://localhost:8080/table-recognition"
+    filePath := "./demo.jpg"
+
+    fileBytes, err := ioutil.ReadFile(filePath)
+    if err != nil {
+        fmt.Printf("Error reading file: %v\n", err)
+        return
+    }
+    fileData := base64.StdEncoding.EncodeToString(fileBytes)
+
+    payload := map[string]interface{}{
+        "file":     fileData,
+        "fileType": 1,
+    }
+    payloadBytes, err := json.Marshal(payload)
+    if err != nil {
+        fmt.Printf("Error marshaling payload: %v\n", err)
+        return
+    }
+
+    client := &http.Client{}
+    req, err := http.NewRequest("POST", API_URL, bytes.NewBuffer(payloadBytes))
+    if err != nil {
+        fmt.Printf("Error creating request: %v\n", err)
+        return
+    }
+    req.Header.Set("Content-Type", "application/json")
+
+    res, err := client.Do(req)
+    if err != nil {
+        fmt.Printf("Error sending request: %v\n", err)
+        return
+    }
+    defer res.Body.Close()
+
+    if res.StatusCode != http.StatusOK {
+        fmt.Printf("Unexpected status code: %d\n", res.StatusCode)
+        return
+    }
+
+    body, err := ioutil.ReadAll(res.Body)
+    if err != nil {
+        fmt.Printf("Error reading response body: %v\n", err)
+        return
+    }
+
+    type TableRecResult struct {
+        PrunedResult  map[string]interface{} `json:"prunedResult"`
+        OutputImages  map[string]string      `json:"outputImages"`
+        InputImage    *string                `json:"inputImage"`
+    }
+
+    type Response struct {
+        Result struct {
+            TableRecResults []TableRecResult `json:"tableRecResults"`
+            DataInfo        interface{}      `json:"dataInfo"`
+        } `json:"result"`
+    }
+
+    var respData Response
+    if err := json.Unmarshal(body, &respData); err != nil {
+        fmt.Printf("Error unmarshaling response: %v\n", err)
+        return
+    }
+
+    for i, res := range respData.Result.TableRecResults {
+        fmt.Printf("Result %d - prunedResult: %+v\n", i, res.PrunedResult)
+
+        for imgName, imgData := range res.OutputImages {
+            imgBytes, err := base64.StdEncoding.DecodeString(imgData)
+            if err != nil {
+                fmt.Printf("Error decoding image %s_%d: %v\n", imgName, i, err)
+                continue
+            }
+
+            filename := fmt.Sprintf("%s_%d.jpg", imgName, i)
+            if err := ioutil.WriteFile(filename, imgBytes, 0644); err != nil {
+                fmt.Printf("Error saving image %s: %v\n", filename, err)
+                continue
+            }
+            fmt.Printf("Saved image to %s\n", filename)
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>C#</summary>
+
+<pre><code class="language-csharp">using System;
+using System.IO;
+using System.Net.Http;
+using System.Text;
+using System.Threading.Tasks;
+using Newtonsoft.Json.Linq;
+
+class Program
+{
+    static readonly string API_URL = "http://localhost:8080/table-recognition";
+    static readonly string inputFilePath = "./demo.jpg";
+
+    static async Task Main(string[] args)
+    {
+        var httpClient = new HttpClient();
+
+        byte[] fileBytes = File.ReadAllBytes(inputFilePath);
+        string fileData = Convert.ToBase64String(fileBytes);
+
+        var payload = new JObject
+        {
+            { "file", fileData },
+            { "fileType", 1 }
+        };
+        var content = new StringContent(payload.ToString(), Encoding.UTF8, "application/json");
+
+        HttpResponseMessage response = await httpClient.PostAsync(API_URL, content);
+        response.EnsureSuccessStatusCode();
+
+        string responseBody = await response.Content.ReadAsStringAsync();
+        JObject jsonResponse = JObject.Parse(responseBody);
+
+        JArray tableRecResults = (JArray)jsonResponse["result"]["tableRecResults"];
+        for (int i = 0; i < tableRecResults.Count; i++)
+        {
+            var res = tableRecResults[i];
+            Console.WriteLine($"[{i}] prunedResult:\n{res["prunedResult"]}");
+
+            JObject outputImages = res["outputImages"] as JObject;
+            if (outputImages != null)
+            {
+                foreach (var img in outputImages)
+                {
+                    string imgName = img.Key;
+                    string base64Img = img.Value?.ToString();
+                    if (!string.IsNullOrEmpty(base64Img))
+                    {
+                        string imgPath = $"{imgName}_{i}.jpg";
+                        byte[] imageBytes = Convert.FromBase64String(base64Img);
+                        File.WriteAllBytes(imgPath, imageBytes);
+                        Console.WriteLine($"Output image saved at {imgPath}");
+                    }
+                }
+            }
+        }
+    }
+}
+</code></pre></details>
+
+<details><summary>Node.js</summary>
+
+<pre><code class="language-js">const axios = require('axios');
+const fs = require('fs');
+const path = require('path');
+
+const API_URL = 'http://localhost:8080/table-recognition';
+const inputImagePath = './demo.jpg';
+
+function encodeImageToBase64(filePath) {
+  const bitmap = fs.readFileSync(filePath);
+  return Buffer.from(bitmap).toString('base64');
+}
+
+async function callTableRecognitionAPI() {
+  const payload = {
+    file: encodeImageToBase64(inputImagePath),
+    fileType: 1
+  };
+
+  try {
+    const response = await axios.post(API_URL, payload, {
+      headers: {
+        'Content-Type': 'application/json'
+      },
+      maxBodyLength: Infinity
+    });
+
+    const results = response.data.result.tableRecResults;
+
+    results.forEach((res, index) => {
+      console.log(`Result [${index}] prunedResult:\n`, res.prunedResult);
+
+      const outputImages = res.outputImages || {};
+      Object.entries(outputImages).forEach(([imgName, base64Img]) => {
+        const outputPath = `${imgName}_${index}.jpg`;
+        fs.writeFileSync(outputPath, Buffer.from(base64Img, 'base64'));
+        console.log(`Saved image: ${outputPath}`);
+      });
+    });
+
+  } catch (error) {
+    console.error('API request failed:', error.message);
+  }
+}
+
+callTableRecognitionAPI();
+
+</code></pre></details>
+
+<details><summary>PHP</summary>
+
+<pre><code class="language-php">&lt;?php
+
+$API_URL = "http://localhost:8080/table-recognition";
+$image_path = "./demo.jpg";
+
+$image_data = base64_encode(file_get_contents($image_path));
+$payload = array(
+    "file" => $image_data,
+    "fileType" => 1
+);
+
+$ch = curl_init($API_URL);
+curl_setopt($ch, CURLOPT_POST, true);
+curl_setopt($ch, CURLOPT_POSTFIELDS, json_encode($payload));
+curl_setopt($ch, CURLOPT_HTTPHEADER, array('Content-Type: application/json'));
+curl_setopt($ch, CURLOPT_RETURNTRANSFER, true);
+$response = curl_exec($ch);
+curl_close($ch);
+
+$result_array = json_decode($response, true);
+$results = $result_array["result"]["tableRecResults"];
+
+foreach ($results as $i => $item) {
+    echo "[$i] prunedResult:\n";
+    print_r($item["prunedResult"]);
+
+    if (!empty($item["outputImages"])) {
+        foreach ($item["outputImages"] as $img_name => $base64_img) {
+            $img_path = $img_name . "_" . $i . ".jpg";
+            file_put_contents($img_path, base64_decode($base64_img));
+            echo "Output image saved at $img_path\n";
+        }
+    } else {
+        echo "No outputImages found for item $i\n";
+    }
+}
+?&gt;
+</code></pre></details>
+</details>
+<br/>
+
+## 4. 二次开发
+
+如果通用表格识别v2产线提供的默认模型权重在您的场景中，精度或速度不满意，您可以尝试利用<b>您自己拥有的特定领域或应用场景的数据</b>对现有模型进行进一步的<b>微调</b>，以提升通用表格识别v2产线的在您的场景中的识别效果。
+
+由于通用表格识别v2产线包含若干模块，模型产线的效果如果不及预期，可能来自于其中任何一个模块。您可以对识别效果差的图片进行分析，进而确定是哪个模块存在问题，并参考以下表格中对应的微调教程链接进行模型微调。
+
+<table>
+<thead>
+<tr>
+<th>情形</th>
+<th>微调模块</th>
+<th>微调参考链接</th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td>表格分类错误</td>
+<td>表格分类模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/table_classification.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>表格单元格定位错误</td>
+<td>表格单元格检测模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/table_cells_detection.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>表格结构识别错误</td>
+<td>表格结构识别模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/table_structure_recognition.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>未能成功检测到表格所在区域</td>
+<td>版面区域检测模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/layout_detection.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>文本存在漏检</td>
+<td>文本检测模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/text_detection.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>文本内容都不准</td>
+<td>文本识别模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleOCR/main/version3.x/module_usage/text_recognition.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>整图旋转/表格旋转矫正不准</td>
+<td>文档图像方向分类模块</td>
+<td><a href="https://paddlepaddle.github.io/PaddleX/latest/module_usage/tutorials/ocr_modules/doc_img_orientation_classification.html#_5">链接</a></td>
+</tr>
+<tr>
+<td>图像扭曲矫正不准</td>
+<td>文本图像矫正模块</td>
+<td>暂不支持微调</td>
+</tr>
+</tbody>
+</table>
+
+
+### 4.2 模型应用
+
+当您使用私有数据集完成微调训练后，可获得本地模型权重文件，然后可以通过参数指定本地模型保存路径的方式，或者通过自定义产线配置文件的方式，使用微调后的模型权重。
+
+#### 4.2.1 通过参数指定本地模型路径
+
+在初始化产线对象时，通过参数指定本地模型路径。以有线表结构识别模型 SLANeXt_wired 微调后的权重的使用方法为例，示例如下：
+
+命令行方式:
+
+```bash
+# 通过 --wired_table_structure_recognition_model_dir 指定本地模型路径
+paddleocr table_recognition_v2_pipeline -i ./table_recognition_v2.jpg --wired_table_structure_recognition_model_dir your_model_path
+
+# 假设使用 SLANeXt_wired 模型作为默认有线表结构识别模型，如果微调的不是该模型，通过 --wired_table_structure_recognition_model_name 修改模型名称
+paddleocr table_recognition_v2_pipeline -i ./table_recognition_v2.jpg --wired_table_structure_recognition_model_name SLANeXt_wired --wired_table_structure_recognition_model_dir your_model_path
+```
+
+脚本方式：
+
+```python
+
+from paddleocr import TableRecognitionPipelineV2
+
+# 通过 wired_table_structure_recognition_model_dir 指定本地模型路径
+pipeline = TableRecognitionPipelineV2(wired_table_structure_recognition_model_dir="./your_model_path")
+
+# 默认使用 SLANeXt_wired 模型作为默认表格识别模型，如果微调的不是该模型，通过 wired_table_structure_recognition_model_name 修改模型名称
+# pipeline = PaddleOCR(wired_table_structure_recognition_model_name="SLANeXt_wired", wired_table_structure_recognition_model_dir="./your_model_path")
+
+```
+
+#### 4.2.2 通过配置文件指定本地模型路径
+
+1.获取产线配置文件
+
+可调用 PaddleOCR 中 通用表格识别v2产线对象的 `export_paddlex_config_to_yaml` 方法，将当前产线配置导出为 YAML 文件：
+
+```Python
+from paddleocr import TableRecognitionPipelineV2
+
+pipeline = TableRecognitionPipelineV2()
+pipeline.export_paddlex_config_to_yaml("TableRecognitionPipelineV2.yaml")
+```
+
+2.修改配置文件
+
+在得到默认的产线配置文件后，将微调后模型权重的本地路径替换至产线配置文件中的对应位置即可。例如
+
+```yaml
+......
+SubModules:
+  LayoutDetection:
+    module_name: layout_detection
+    model_name: PicoDet_layout_1x_table
+    model_dir: null # 替换为微调后的版面区域检测模型权重路径
+
+  TableClassification:
+    module_name: table_classification
+    model_name: PP-LCNet_x1_0_table_cls
+    model_dir: null # 替换为微调后的表格分类模型权重路径
+
+  WiredTableStructureRecognition:
+    module_name: table_structure_recognition
+    model_name: SLANeXt_wired
+    model_dir: null # 替换为微调后的有线表格结构识别模型权重路径
+
+  WirelessTableStructureRecognition:
+    module_name: table_structure_recognition
+    model_name: SLANeXt_wireless
+    model_dir: null # 替换为微调后的无线表格结构识别模型权重路径
+
+  WiredTableCellsDetection:
+    module_name: table_cells_detection
+    model_name: RT-DETR-L_wired_table_cell_det
+    model_dir: null # 替换为微调后的有线表格单元格检测模型权重路径
+
+  WirelessTableCellsDetection:
+    module_name: table_cells_detection
+    model_name: RT-DETR-L_wireless_table_cell_det
+    model_dir: null # 替换为微调后的无线表格单元格检测模型权重路径
+
+SubPipelines:
+  DocPreprocessor:
+    pipeline_name: doc_preprocessor
+    use_doc_orientation_classify: True
+    use_doc_unwarping: True
+    SubModules:
+      DocOrientationClassify:
+        module_name: doc_text_orientation
+        model_name: PP-LCNet_x1_0_doc_ori
+        model_dir: null # 替换为微调后的文档图像方向分类模型权重路径
+
+      DocUnwarping:
+        module_name: image_unwarping
+        model_name: UVDoc
+        model_dir: null
+
+  GeneralOCR:
+    pipeline_name: OCR
+    text_type: general
+    use_doc_preprocessor: False
+    use_textline_orientation: False
+    SubModules:
+      TextDetection:
+        module_name: text_detection
+        model_name: PP-OCRv5_server_det
+        model_dir: null # 替换为微调后的文本检测模型权重路径
+        limit_side_len: 960
+        limit_type: max
+        max_side_limit: 4000
+        thresh: 0.3
+        box_thresh: 0.4
+        unclip_ratio: 1.5
+
+      TextRecognition:
+        module_name: text_recognition
+        model_name: PP-OCRv5_server_rec
+        model_dir: null # 替换为微调后文本识别的模型权重路径
+        batch_size: 1
+        score_thresh: 0
+......
+```
+
+在产线配置文件中，不仅包含 PaddleOCR CLI 和 Python API 支持的参数，还可进行更多高级配置，具体信息可在 [PaddleX模型产线使用概览](https://paddlepaddle.github.io/PaddleX/latest/pipeline_usage/pipeline_develop_guide.html) 中找到对应的产线使用教程，参考其中的详细说明，根据需求调整各项配置。
+
+3.在 CLI 中加载产线配置文件
+
+在修改完成配置文件后，通过命令行的 `--paddlex_config` 参数指定修改后的产线配置文件的路径，PaddleOCR 会读取其中的内容作为产线配置。示例如下：
+
+```bash
+paddleocr table_recognition_v2_pipeline --paddlex_config PaddleOCR.yaml ...
+```
+
+4.在 Python API 中加载产线配置文件
+
+初始化产线对象时，可通过 `paddlex_config` 参数传入 PaddleX 产线配置文件路径或配置dict，PaddleOCR 会读取其中的内容作为产线配置。示例如下：
+
+```python
+from paddleocr import TableRecognitionPipelineV2
+
+pipeline = TableRecognitionPipelineV2(paddlex_config="TableRecognitionPipelineV2.yaml")
+```
diff --git a/langchain-paddleocr/Makefile b/langchain-paddleocr/Makefile
new file mode 100644
index 0000000..eca1ba8
--- /dev/null
+++ b/langchain-paddleocr/Makefile
@@ -0,0 +1,71 @@
+.PHONY: all format lint test tests integration_tests help
+
+# Default target executed when no arguments are given to make.
+all: help
+
+.EXPORT_ALL_VARIABLES:
+UV_FROZEN = true
+
+# Define a variable for the test file path.
+TEST_FILE ?= tests/unit_tests/
+
+integration_test integration_tests: TEST_FILE = tests/integration_tests/
+
+test tests:
+	uv run --group test pytest --disable-socket --allow-unix-socket $(TEST_FILE)
+
+integration_test integration_tests:
+	uv run --group test --group test_integration pytest --retries 3 --retry-delay 1 $(TEST_FILE)
+
+check_imports: $(shell find langchain_paddleocr -name '*.py')
+	uv run --all-groups python ./scripts/check_imports.py $^
+
+test_watch:
+	uv run ptw --snapshot-update --now . -- -vv $(TEST_FILE)
+
+# Run unit tests and generate a coverage report.
+coverage:
+	uv run --with pytest-cov pytest --cov \
+		--cov-report xml \
+		--cov-report html \
+		--cov-report term-missing:skip-covered \
+		$(TEST_FILE)
+	open htmlcov/index.html
+
+######################
+# LINTING AND FORMATTING
+######################
+
+# Define a variable for Python and notebook files.
+PYTHON_FILES=.
+MYPY_CACHE=.mypy_cache
+lint format: PYTHON_FILES=.
+lint_diff format_diff: PYTHON_FILES=$(shell git diff --name-only --diff-filter=d main | grep -E '\.py$$|\.ipynb$$')
+lint_package: PYTHON_FILES=langchain_paddleocr
+lint_tests: PYTHON_FILES=tests
+lint_tests: MYPY_CACHE=.mypy_cache_test
+
+lint lint_diff lint_package lint_tests:
+	./scripts/lint_imports.sh
+	[ "$(PYTHON_FILES)" = "" ] || uv run --all-groups ruff check $(PYTHON_FILES)
+	[ "$(PYTHON_FILES)" = "" ] || uv run --all-groups ruff format $(PYTHON_FILES) --diff
+	[ "$(PYTHON_FILES)" = "" ] || uv run --all-groups ruff check --select I $(PYTHON_FILES)
+	[ "$(PYTHON_FILES)" = "" ] || mkdir -p $(MYPY_CACHE) && uv run --all-groups mypy $(PYTHON_FILES) --cache-dir $(MYPY_CACHE)
+
+format format_diff:
+	[ "$(PYTHON_FILES)" = "" ] || uv run --all-groups ruff format $(PYTHON_FILES)
+	[ "$(PYTHON_FILES)" = "" ] || uv run --all-groups ruff check --fix $(PYTHON_FILES)
+
+######################
+# HELP
+######################
+
+help:
+	@echo '----'
+	@echo 'check_imports                - check imports'
+	@echo 'format                       - run code formatters'
+	@echo 'lint                         - run linters'
+	@echo 'test                         - run unit tests'
+	@echo 'tests                        - run unit tests'
+	@echo 'integration_test             - run integration tests (NOTE: "export PADDLEOCR_ACCESS_TOKEN=..." is needed.)'
+	@echo 'test TEST_FILE=<test_file>   - run all tests in file'
diff --git a/langchain-paddleocr/README.md b/langchain-paddleocr/README.md
new file mode 100644
index 0000000..5ebb9d8
--- /dev/null
+++ b/langchain-paddleocr/README.md
@@ -0,0 +1,47 @@
+# langchain-paddleocr
+
+[![PyPI - Version](https://img.shields.io/pypi/v/langchain-paddleocr?label=%20)](https://pypi.org/project/langchain-paddleocr/#history)
+[![PyPI - License](https://img.shields.io/pypi/l/langchain-paddleocr)](https://opensource.org/license/apache-2-0)
+[![PyPI - Downloads](https://img.shields.io/pepy/dt/langchain-paddleocr)](https://pypistats.org/packages/langchain-paddleocr)
+
+This package provides access to PaddleOCR's capabilities within the LangChain ecosystem.
+
+## Quick Install
+
+```bash
+pip install langchain-paddleocr
+```
+
+## Basic Usage
+
+### `PaddleOCRVLLoader`
+
+The `PaddleOCRVLLoader` enables you to:
+
+- Extract text and layout information from PDF and image files using models from Baidu's PaddleOCR-VL series (e.g., PaddleOCR-VL, PaddleOCR-VL-1.5, PaddleOCR-VL-1.6)
+- Process documents from local files or remote URLs
+
+Basic usage of `PaddleOCRVLLoader` looks as follows:
+
+```python
+from langchain_paddleocr import PaddleOCRVLLoader
+from pydantic import SecretStr
+
+loader = PaddleOCRVLLoader(
+    file_path="path/to/document.pdf",
+    base_url="your-api-endpoint",
+    model="PaddleOCR-VL-1.5",
+    access_token=SecretStr("your-access-token")  # Optional if using environment variable `PADDLEOCR_ACCESS_TOKEN`
+)
+
+docs = loader.load()
+
+for doc in docs[:2]:
+    print(f"Content: {doc.page_content[:200]}...")
+    print(f"Source: {doc.metadata['source']}")
+    print("---")
+```
+
+## 📖 Documentation
+
+For full documentation, see the [LangChain Docs](https://docs.langchain.com/oss/python/integrations/providers/baidu).
diff --git a/langchain-paddleocr/README_cn.md b/langchain-paddleocr/README_cn.md
new file mode 100644
index 0000000..ef08372
--- /dev/null
+++ b/langchain-paddleocr/README_cn.md
@@ -0,0 +1,48 @@
+# langchain-paddleocr
+
+[![PyPI - 版本](https://img.shields.io/pypi/v/langchain-paddleocr?label=%20)](https://pypi.org/project/langchain-paddleocr/#history)
+[![PyPI - 许可证](https://img.shields.io/pypi/l/langchain-paddleocr)](https://opensource.org/license/apache-2-0)
+[![PyPI - 下载量](https://img.shields.io/pepy/dt/langchain-paddleocr)](https://pypistats.org/packages/langchain-paddleocr)
+
+本 Python 包在 LangChain 生态系统中提供对 PaddleOCR 功能的访问。
+
+## 快速安装
+
+```bash
+pip install langchain-paddleocr
+```
+
+## 基本用法
+
+### `PaddleOCRVLLoader`
+
+`PaddleOCRVLLoader` 允许你：
+
+- 使用百度 PaddleOCR-VL 系列模型（例如 PaddleOCR-VL、PaddleOCR-VL-1.5、PaddleOCR-VL-1.6）从 PDF 和图像文件中提取文本和版面布局信息
+- 处理来自本地文件或远程 URL 的文档
+
+`PaddleOCRVLLoader` 的基本用法如下：
+
+```python
+from langchain_paddleocr import PaddleOCRVLLoader
+from pydantic import SecretStr
+
+loader = PaddleOCRVLLoader(
+    file_path="path/to/document.pdf",
+    base_url="your-api-endpoint",
+    model="PaddleOCR-VL-1.5",
+    access_token=SecretStr("your-access-token")  # 如果使用环境变量 `PADDLEOCR_ACCESS_TOKEN`，则此项为可选
+)
+
+docs = loader.load()
+
+for doc in docs[:2]:
+    print(f"Content: {doc.page_content[:200]}...")
+    print(f"Source: {doc.metadata['source']}")
+    print("---")
+```
+
+
+## 📖 文档
+
+完整文档请参阅 [LangChain 文档](https://docs.langchain.com/oss/python/integrations/providers/baidu)。
diff --git a/langchain-paddleocr/langchain_paddleocr/__init__.py b/langchain-paddleocr/langchain_paddleocr/__init__.py
new file mode 100644
index 0000000..e761d4c
--- /dev/null
+++ b/langchain-paddleocr/langchain_paddleocr/__init__.py
@@ -0,0 +1,17 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .document_loaders import PaddleOCRVLLoader
+
+__all__ = ["PaddleOCRVLLoader"]
diff --git a/langchain-paddleocr/langchain_paddleocr/document_loaders/__init__.py b/langchain-paddleocr/langchain_paddleocr/document_loaders/__init__.py
new file mode 100644
index 0000000..dc6517f
--- /dev/null
+++ b/langchain-paddleocr/langchain_paddleocr/document_loaders/__init__.py
@@ -0,0 +1,17 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .paddleocr_vl import PaddleOCRVLLoader
+
+__all__ = ["PaddleOCRVLLoader"]
diff --git a/langchain-paddleocr/langchain_paddleocr/document_loaders/paddleocr_vl.py b/langchain-paddleocr/langchain_paddleocr/document_loaders/paddleocr_vl.py
new file mode 100644
index 0000000..e5a0e78
--- /dev/null
+++ b/langchain-paddleocr/langchain_paddleocr/document_loaders/paddleocr_vl.py
@@ -0,0 +1,264 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import annotations
+
+import logging
+import os
+from collections.abc import Iterable, Iterator
+from pathlib import Path
+from typing import Any
+
+from langchain_core.document_loaders import BaseLoader
+from langchain_core.documents import Document
+from paddleocr import Model, PaddleOCRClient, PaddleOCRVLOptions
+from pydantic import SecretStr
+
+logger = logging.getLogger(__name__)
+
+_DEFAULT_BASE_URL = "https://paddleocr.aistudio-app.com"
+
+_PAGES_DELIMITER = "\n\f"
+
+
+def _resolve_vl_model(model: str | Model) -> Model:
+    if isinstance(model, Model):
+        resolved = model
+    else:
+        try:
+            resolved = Model(model)
+        except ValueError as exc:
+            msg = f"Unsupported model: {model!r}"
+            raise ValueError(msg) from exc
+
+    return resolved
+
+
+class PaddleOCRVLLoader(BaseLoader):
+    """Load documents using the PaddleOCR-VL document parsing API via SDK."""
+
+    def __init__(
+        self,
+        file_path: str | Iterable[str],
+        *,
+        access_token: SecretStr | None = None,
+        base_url: str | None = None,
+        model: str | Model | None = None,
+        use_doc_orientation_classify: bool | None = False,
+        use_doc_unwarping: bool | None = False,
+        use_layout_detection: bool | None = None,
+        use_chart_recognition: bool | None = None,
+        use_seal_recognition: bool | None = None,
+        use_ocr_for_image_block: bool | None = None,
+        layout_threshold: float | dict[int, float] | None = None,
+        layout_nms: bool | None = None,
+        layout_unclip_ratio: (
+            tuple[float, float] | dict[int, tuple[float, float]] | float | None
+        ) = None,
+        layout_merge_bboxes_mode: str | dict[str, float] | None = None,
+        layout_shape_mode: str | None = None,
+        prompt_label: str | None = None,
+        format_block_content: bool | None = None,
+        repetition_penalty: float | None = None,
+        temperature: float | None = None,
+        top_p: float | None = None,
+        min_pixels: int | None = None,
+        max_pixels: int | None = None,
+        max_new_tokens: int | None = None,
+        vlm_extra_args: dict[str, Any] | None = None,
+        merge_layout_blocks: bool | None = None,
+        markdown_ignore_labels: list[str] | None = None,
+        prettify_markdown: bool | None = None,
+        show_formula_number: bool | None = None,
+        restructure_pages: bool | None = None,
+        merge_tables: bool | None = None,
+        relevel_titles: bool | None = None,
+        return_markdown_images: bool | None = None,
+        output_formats: list[str] | None = None,
+        visualize: bool | None = None,
+        extra_options: dict[str, Any] | None = None,
+        timeout: int = 300,
+    ) -> None:
+        """Initialize the PaddleOCR-VL loader.
+
+        Args:
+            file_path: Single path/URL or an iterable of paths/URLs to PDF or
+                image files.
+            access_token: Optional access token as ``SecretStr``. If ``None``, the value
+                from the ``PADDLEOCR_ACCESS_TOKEN`` environment variable will be used.
+            base_url: Base URL of the PaddleOCR API service. Defaults to the official
+                service.
+            model: PaddleOCR-VL model name or ``Model`` enum value.
+            use_doc_orientation_classify: Whether to enable document orientation
+                classification.
+            use_doc_unwarping: Whether to enable document unwarping.
+            use_layout_detection: Whether to enable layout detection.
+            use_chart_recognition: Whether to enable chart recognition.
+            use_seal_recognition: Whether to enable seal recognition.
+            use_ocr_for_image_block: Whether to use OCR for image layout blocks.
+            layout_threshold: Layout detection threshold.
+            layout_nms: Whether to apply non-maximum suppression for layout detection.
+            layout_unclip_ratio: Layout unclip ratio.
+            layout_merge_bboxes_mode: Mode for merging layout bounding boxes.
+            layout_shape_mode: Layout shape mode.
+            prompt_label: Prompt label for the VLM.
+            format_block_content: Whether to save formatted Markdown in JSON.
+            repetition_penalty: Repetition penalty for VLM sampling.
+            temperature: Temperature for VLM sampling.
+            top_p: Top-p sampling value for VLM.
+            min_pixels: Minimum number of pixels allowed in preprocessing.
+            max_pixels: Maximum number of pixels allowed in preprocessing.
+            max_new_tokens: Maximum number of tokens generated by the VLM.
+            vlm_extra_args: Additional VLM arguments passed through to the service.
+            merge_layout_blocks: Whether to merge layout blocks across columns.
+            markdown_ignore_labels: Layout labels to ignore in Markdown.
+            prettify_markdown: Whether to prettify the Markdown output.
+            show_formula_number: Whether to include formula numbers in Markdown.
+            restructure_pages: Whether to restructure results across pages.
+            merge_tables: Whether to merge tables across pages.
+            relevel_titles: Whether to relevel titles.
+            return_markdown_images: Whether to return images referenced by Markdown.
+            output_formats: Additional export formats such as ``docx``.
+            visualize: Whether to include visualization results.
+            extra_options: Additional service options passed through as-is.
+            timeout: Poll timeout in seconds.
+        """
+        self._file_paths = (
+            file_path
+            if isinstance(file_path, Iterable) and not isinstance(file_path, str)
+            else [file_path]
+        )
+
+        if access_token is None:
+            env_value = os.getenv("PADDLEOCR_ACCESS_TOKEN")
+            self._token = env_value
+        else:
+            self._token = access_token.get_secret_value()
+
+        self._base_url = base_url or _DEFAULT_BASE_URL
+        self._timeout = timeout
+
+        self._model = _resolve_vl_model(model) if model is not None else None
+
+        options_kwargs: dict[str, Any] = {
+            key: val
+            for key, val in {
+                "use_doc_orientation_classify": use_doc_orientation_classify,
+                "use_doc_unwarping": use_doc_unwarping,
+                "use_layout_detection": use_layout_detection,
+                "use_chart_recognition": use_chart_recognition,
+                "use_seal_recognition": use_seal_recognition,
+                "use_ocr_for_image_block": use_ocr_for_image_block,
+                "layout_threshold": layout_threshold,
+                "layout_nms": layout_nms,
+                "layout_unclip_ratio": layout_unclip_ratio,
+                "layout_merge_bboxes_mode": layout_merge_bboxes_mode,
+                "layout_shape_mode": layout_shape_mode,
+                "prompt_label": prompt_label,
+                "format_block_content": format_block_content,
+                "repetition_penalty": repetition_penalty,
+                "temperature": temperature,
+                "top_p": top_p,
+                "min_pixels": min_pixels,
+                "max_pixels": max_pixels,
+                "max_new_tokens": max_new_tokens,
+                "vlm_extra_args": vlm_extra_args,
+                "merge_layout_blocks": merge_layout_blocks,
+                "markdown_ignore_labels": markdown_ignore_labels,
+                "prettify_markdown": prettify_markdown,
+                "show_formula_number": show_formula_number,
+                "restructure_pages": restructure_pages,
+                "merge_tables": merge_tables,
+                "relevel_titles": relevel_titles,
+                "return_markdown_images": return_markdown_images,
+                "output_formats": output_formats,
+                "visualize": visualize,
+                "extra_options": extra_options,
+            }.items()
+            if val is not None
+        }
+
+        self._options = PaddleOCRVLOptions(**options_kwargs) if options_kwargs else None
+
+    def _is_url(self, path: str) -> bool:
+        """Check if the path is a URL."""
+        return path.startswith(("http://", "https://"))
+
+    def _process_file(self, file_path: str) -> tuple[str, dict[str, Any]]:
+        """Process a single file through the SDK and return text + raw result."""
+        with PaddleOCRClient(
+            token=self._token,
+            base_url=self._base_url,
+            client_platform="langchain",
+            poll_timeout=self._timeout,
+        ) as client:
+            parse_kwargs: dict[str, Any] = {"options": self._options}
+            if self._model is not None:
+                parse_kwargs["model"] = self._model
+            if self._is_url(file_path):
+                result = client.parse_document(file_url=file_path, **parse_kwargs)
+            else:
+                local_path = Path(file_path)
+                if not local_path.exists():
+                    msg = f"File not found: '{file_path}'"
+                    raise ValueError(msg)
+                result = client.parse_document(
+                    file_path=str(local_path),
+                    **parse_kwargs,
+                )
+
+        text_parts = [page.markdown_text for page in result.pages if page.markdown_text]
+        text = _PAGES_DELIMITER.join(text_parts)
+
+        raw_response = {
+            "job_id": result.job_id,
+            "data_info": result.data_info,
+            "pages": [
+                {
+                    "markdown_text": page.markdown_text,
+                    "markdown_images": page.markdown_images,
+                    "output_images": page.output_images,
+                    "pruned_result": page.pruned_result,
+                    "input_image_url": page.input_image_url,
+                    "exports": page.exports,
+                    "markdown": page.markdown,
+                    "raw": page.raw,
+                }
+                for page in result.pages
+            ],
+        }
+
+        return text, raw_response
+
+    def lazy_load(self) -> Iterator[Document]:
+        """Lazily load documents from the configured file paths."""
+        for file_path in self._file_paths:
+            text, raw_response = self._process_file(file_path)
+
+            if not text:
+                logger.warning(
+                    "%s could not extract any text from '%s'."
+                    " Returning an empty document.",
+                    self.__class__.__name__,
+                    file_path,
+                )
+
+            metadata: dict[str, Any] = {
+                "source": str(file_path),
+                "paddleocr_vl_raw_response": raw_response,
+            }
+            if self._model is not None:
+                metadata["model"] = self._model.value
+
+            yield Document(page_content=text, metadata=metadata)
diff --git a/langchain-paddleocr/langchain_paddleocr/py.typed b/langchain-paddleocr/langchain_paddleocr/py.typed
new file mode 100644
index 0000000..e69de29
diff --git a/langchain-paddleocr/pyproject.toml b/langchain-paddleocr/pyproject.toml
new file mode 100644
index 0000000..4e691ac
--- /dev/null
+++ b/langchain-paddleocr/pyproject.toml
@@ -0,0 +1,143 @@
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[project]
+name = "langchain-paddleocr"
+version = "0.2.2"
+description = "An integration package connecting PaddleOCR and LangChain"
+authors = []
+readme = "README.md"
+license = { text = "Apache License 2.0" }
+requires-python = ">=3.10"
+dependencies = [
+    "paddleocr>=3.7.0",
+    "langchain-core>=1.2.5,<2.0.0",
+    "pydantic>=2.0.0,<3.0.0",
+]
+
+[project.urls]
+"Source Code" = "https://github.com/PaddlePaddle/PaddleOCR/tree/main/langchain-paddleocr"
+"Release Notes" = "https://github.com/PaddlePaddle/PaddleOCR/releases"
+Repository = "https://github.com/PaddlePaddle/PaddleOCR"
+
+[dependency-groups]
+lint = ["ruff>=0.13.1,<0.14.0"]
+typing = [
+    "mypy>=1.18.1,<1.19.0",
+    "types-requests>=2.31.0,<3.0.0",
+]
+test = [
+    "pytest>=8.4.0,<9.0.0",
+    "freezegun>=1.5.0,<2.0.0",
+    "pytest-mock>=3.14.0,<4.0.0",
+    "syrupy>=4.9.0,<5.0.0",
+    "pytest-watcher>=0.4.0,<1.0.0",
+    "pytest-asyncio>=0.21.0,<1.0.0",
+    "pytest-retry>=1.7.0,<2.0.0",
+    "pytest-socket>=0.7.0,<1.0.0",
+    "langchain-tests>=1.1.2,<2.0.0",
+]
+test_integration = [
+    "pytest>=8.4.0,<9.0.0",
+]
+
+[tool.ruff]
+fix = true
+
+[tool.ruff.format]
+docstring-code-format = true
+
+[tool.ruff.lint]
+select = [ "ALL",]
+ignore = [
+    "C90",     # McCabe complexity
+    "COM812",  # Messes with the formatter
+    "CPY",     # No copyright
+    "FIX002",  # Line contains TODO
+    "ISC001",  # Messes with the formatter
+    "PERF203", # Rarely useful
+    "PLR09",   # Too many something (arg, statements, etc)
+    "RUF012",  # Doesn't play well with Pydantic
+    "TC001",   # Doesn't play well with Pydantic
+    "TC002",   # Doesn't play well with Pydantic
+    "TC003",   # Doesn't play well with Pydantic
+    "TD002",   # Missing author in TODO
+    "TD003",   # Missing issue link in TODO
+
+    # TODO rules
+    "ANN401",  # No Any types
+    "BLE",     # Blind exceptions
+    "DOC",     # Docstrings (preview)
+    "ERA",     # No commented-out code
+    "PLR2004", # Comparison to magic number
+    "PLC0415", # Import top level
+    "SIM117",  # Single with statement
+    "FBT",     # Bool stuff
+    "PT",
+    "ARG",     # Argument related
+    "SLF",     # Self
+    "D",       # Docstring related
+    "G",
+    "S",
+    "A",
+    "N",
+    "TRY",
+    "B",
+    "PGH",
+    "SIM",
+    "ASYNC",
+    "PTH",
+    "DTZ",
+    "PYI",
+    "PLW",
+]
+unfixable = [
+    "B028",    # People should intentionally tune the stacklevel
+    "PLW1510", # People should intentionally set the check argument
+]
+
+[tool.ruff.lint.per-file-ignores]
+"tests/**" = [ "D1", "S", "SLF", "ARG001", "PGH003", "PT011"]
+"scripts/**" = [ "INP", "S",]
+
+[tool.mypy]
+plugins = ["pydantic.mypy"]
+strict = true
+disallow_untyped_defs = true
+
+[[tool.mypy.overrides]]
+module = ["paddleocr.*"]
+ignore_missing_imports = true
+
+# TODO: activate for 'strict' checking
+disallow_any_generics = false
+warn_return_any = false
+
+[tool.coverage.run]
+omit = [
+    "tests/*",
+    "*/site-packages/*",
+    "*/.venv/*",
+    "*/venv/*",
+]
+
+[tool.pytest.ini_options]
+# --strict-markers will raise errors on unknown marks.
+# https://docs.pytest.org/en/7.1.x/how-to/mark.html#raising-errors-on-unknown-marks
+#
+# https://docs.pytest.org/en/7.1.x/reference/reference.html
+# --strict-config       any warnings encountered while parsing the `pytest`
+#                       section of the configuration file raise errors.
+#
+# https://github.com/tophat/syrupy
+# --snapshot-warn-unused    Prints a warning on unused snapshots rather than fail the test suite.
+#addopts = "--snapshot-warn-unused --strict-markers --strict-config --durations=5"
+# Registering custom markers.
+# https://docs.pytest.org/en/7.1.x/example/markers.html#registering-markers
+markers = [
+    "requires: mark tests as requiring a specific library",
+    "asyncio: mark tests as requiring asyncio",
+    "compile: mark placeholder test used to compile integration tests without running them",
+]
+asyncio_mode = "auto"
diff --git a/langchain-paddleocr/scripts/check_imports.py b/langchain-paddleocr/scripts/check_imports.py
new file mode 100644
index 0000000..36f2ff6
--- /dev/null
+++ b/langchain-paddleocr/scripts/check_imports.py
@@ -0,0 +1,31 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import sys
+import traceback
+from importlib.machinery import SourceFileLoader
+
+if __name__ == "__main__":
+    files = sys.argv[1:]
+    has_failure = False
+    for file in files:
+        try:
+            SourceFileLoader("x", file).load_module()
+        except Exception:
+            has_failure = True
+            print(file)  # noqa: T201
+            traceback.print_exc()
+            print()  # noqa: T201
+
+    sys.exit(1 if has_failure else 0)
diff --git a/langchain-paddleocr/scripts/lint_imports.sh b/langchain-paddleocr/scripts/lint_imports.sh
new file mode 100644
index 0000000..695613c
--- /dev/null
+++ b/langchain-paddleocr/scripts/lint_imports.sh
@@ -0,0 +1,17 @@
+#!/bin/bash
+
+set -eu
+
+# Initialize a variable to keep track of errors
+errors=0
+
+# make sure not importing from langchain or langchain_experimental
+git --no-pager grep '^from langchain\.' . && errors=$((errors+1))
+git --no-pager grep '^from langchain_experimental\.' . && errors=$((errors+1))
+
+# Decide on an exit status based on the errors
+if [ "$errors" -gt 0 ]; then
+    exit 1
+else
+    exit 0
+fi
diff --git a/langchain-paddleocr/tests/__init__.py b/langchain-paddleocr/tests/__init__.py
new file mode 100644
index 0000000..b4abe68
--- /dev/null
+++ b/langchain-paddleocr/tests/__init__.py
@@ -0,0 +1,14 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
diff --git a/langchain-paddleocr/tests/data/sample_img.jpg b/langchain-paddleocr/tests/data/sample_img.jpg
new file mode 100644
index 0000000..b030a75
Binary files /dev/null and b/langchain-paddleocr/tests/data/sample_img.jpg differ
diff --git a/langchain-paddleocr/tests/data/sample_pdf.pdf b/langchain-paddleocr/tests/data/sample_pdf.pdf
new file mode 100644
index 0000000..6384246
Binary files /dev/null and b/langchain-paddleocr/tests/data/sample_pdf.pdf differ
diff --git a/langchain-paddleocr/tests/integration_tests/__init__.py b/langchain-paddleocr/tests/integration_tests/__init__.py
new file mode 100644
index 0000000..b4abe68
--- /dev/null
+++ b/langchain-paddleocr/tests/integration_tests/__init__.py
@@ -0,0 +1,14 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
diff --git a/langchain-paddleocr/tests/integration_tests/document_loaders/__init__.py b/langchain-paddleocr/tests/integration_tests/document_loaders/__init__.py
new file mode 100644
index 0000000..b4abe68
--- /dev/null
+++ b/langchain-paddleocr/tests/integration_tests/document_loaders/__init__.py
@@ -0,0 +1,14 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
diff --git a/langchain-paddleocr/tests/integration_tests/document_loaders/test_paddleocr_vl_loader.py b/langchain-paddleocr/tests/integration_tests/document_loaders/test_paddleocr_vl_loader.py
new file mode 100644
index 0000000..5f94daf
--- /dev/null
+++ b/langchain-paddleocr/tests/integration_tests/document_loaders/test_paddleocr_vl_loader.py
@@ -0,0 +1,59 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import annotations
+
+import os
+from pathlib import Path
+
+import pytest
+from pydantic import SecretStr
+
+from langchain_paddleocr import PaddleOCRVLLoader
+
+
+def test_paddleocr_vl_loader_live_integration() -> None:
+    """Live integration test against a real PaddleOCR-VL endpoint.
+
+    This test requires the following environment variables to be set:
+
+    - ``PADDLEOCR_ACCESS_TOKEN``: Access token for the endpoint.
+    - ``PADDLEOCR_BASE_URL`` (optional): Override the default base URL.
+    """
+    access_token = os.getenv("PADDLEOCR_ACCESS_TOKEN")
+
+    if not access_token:
+        pytest.skip("PADDLEOCR_ACCESS_TOKEN must be set for integration tests.")
+
+    base_url = os.getenv("PADDLEOCR_BASE_URL")
+
+    tests_dir = Path(__file__).resolve().parents[2]
+    sample_paths = [
+        str(tests_dir / "data" / "sample_pdf.pdf"),
+        str(tests_dir / "data" / "sample_img.jpg"),
+    ]
+
+    loader = PaddleOCRVLLoader(
+        file_path=sample_paths,
+        access_token=SecretStr(access_token),
+        base_url=base_url,
+    )
+
+    docs = list(loader.lazy_load())
+    assert len(docs) == 2
+
+    for doc, input_path in zip(docs, sample_paths):
+        assert isinstance(doc.page_content, str)
+        assert doc.metadata.get("source") == input_path
+        assert "paddleocr_vl_raw_response" in doc.metadata
diff --git a/langchain-paddleocr/tests/unit_tests/__init__.py b/langchain-paddleocr/tests/unit_tests/__init__.py
new file mode 100644
index 0000000..b4abe68
--- /dev/null
+++ b/langchain-paddleocr/tests/unit_tests/__init__.py
@@ -0,0 +1,14 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
diff --git a/langchain-paddleocr/tests/unit_tests/document_loaders/__init__.py b/langchain-paddleocr/tests/unit_tests/document_loaders/__init__.py
new file mode 100644
index 0000000..b4abe68
--- /dev/null
+++ b/langchain-paddleocr/tests/unit_tests/document_loaders/__init__.py
@@ -0,0 +1,14 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
diff --git a/langchain-paddleocr/tests/unit_tests/document_loaders/test_paddleocr_vl_loader.py b/langchain-paddleocr/tests/unit_tests/document_loaders/test_paddleocr_vl_loader.py
new file mode 100644
index 0000000..10c676b
--- /dev/null
+++ b/langchain-paddleocr/tests/unit_tests/document_loaders/test_paddleocr_vl_loader.py
@@ -0,0 +1,148 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import annotations
+
+from unittest.mock import MagicMock
+
+import pytest
+from paddleocr import Model
+
+from langchain_paddleocr import PaddleOCRVLLoader
+
+
+def test_default_base_url() -> None:
+    """Ensure default base_url is set when not provided."""
+    loader = PaddleOCRVLLoader(
+        file_path="dummy.pdf",
+        access_token=None,
+    )
+    assert loader._base_url == "https://paddleocr.aistudio-app.com"
+
+
+def test_custom_base_url() -> None:
+    """Ensure custom base_url is respected."""
+    loader = PaddleOCRVLLoader(
+        file_path="dummy.pdf",
+        base_url="https://custom.example.com",
+    )
+    assert loader._base_url == "https://custom.example.com"
+
+
+def test_access_token_from_env(monkeypatch: pytest.MonkeyPatch) -> None:
+    """Ensure token is read from environment variable."""
+    monkeypatch.setenv("PADDLEOCR_ACCESS_TOKEN", "env-token")
+    loader = PaddleOCRVLLoader(file_path="dummy.pdf")
+    assert loader._token == "env-token"
+
+
+def test_options_are_built_correctly() -> None:
+    """Ensure options with non-None values are passed to PaddleOCRVLOptions."""
+    loader = PaddleOCRVLLoader(
+        file_path="dummy.pdf",
+        use_layout_detection=True,
+        temperature=0.7,
+        top_p=0.9,
+    )
+    assert loader._options is not None
+    assert loader._options.use_layout_detection is True
+    assert loader._options.temperature == 0.7
+    assert loader._options.top_p == 0.9
+
+
+def test_options_include_default_flags() -> None:
+    """Ensure default boolean flags are included in options."""
+    loader = PaddleOCRVLLoader(file_path="dummy.pdf")
+    assert loader._options is not None
+    assert loader._options.use_doc_orientation_classify is False
+    assert loader._options.use_doc_unwarping is False
+
+
+def test_lazy_load_raises_for_missing_file(monkeypatch: pytest.MonkeyPatch) -> None:
+    """Ensure lazy_load raises when a local file does not exist."""
+    monkeypatch.setenv("PADDLEOCR_ACCESS_TOKEN", "test-token")
+    loader = PaddleOCRVLLoader(
+        file_path="nonexistent-file.pdf",
+    )
+    with pytest.raises(ValueError, match="File not found"):
+        list(loader.lazy_load())
+
+
+def test_default_model() -> None:
+    loader = PaddleOCRVLLoader(file_path="dummy.pdf")
+    assert loader._model is None
+
+
+def test_parse_document_omits_model_by_default(monkeypatch: pytest.MonkeyPatch) -> None:
+    mock_result = MagicMock()
+    mock_result.job_id = "job-1"
+    mock_result.pages = []
+
+    mock_client = MagicMock()
+    mock_client.parse_document.return_value = mock_result
+    mock_client.__enter__.return_value = mock_client
+    mock_client.__exit__.return_value = None
+
+    monkeypatch.setattr(
+        "langchain_paddleocr.document_loaders.paddleocr_vl.PaddleOCRClient",
+        lambda **kwargs: mock_client,
+    )
+    monkeypatch.setenv("PADDLEOCR_ACCESS_TOKEN", "test-token")
+
+    loader = PaddleOCRVLLoader(file_path="https://example.com/doc.pdf")
+    list(loader.lazy_load())
+
+    mock_client.parse_document.assert_called_once_with(
+        file_url="https://example.com/doc.pdf",
+        options=loader._options,
+    )
+
+
+def test_custom_model_string() -> None:
+    loader = PaddleOCRVLLoader(
+        file_path="dummy.pdf",
+        model="PaddleOCR-VL-1.5",
+    )
+    assert loader._model == Model.PADDLE_OCR_VL_15
+
+
+def test_parse_document_receives_model(monkeypatch: pytest.MonkeyPatch) -> None:
+    mock_result = MagicMock()
+    mock_result.job_id = "job-1"
+    mock_result.pages = []
+
+    mock_client = MagicMock()
+    mock_client.parse_document.return_value = mock_result
+    mock_client.__enter__.return_value = mock_client
+    mock_client.__exit__.return_value = None
+
+    monkeypatch.setattr(
+        "langchain_paddleocr.document_loaders.paddleocr_vl.PaddleOCRClient",
+        lambda **kwargs: mock_client,
+    )
+
+    loader = PaddleOCRVLLoader(
+        file_path="https://example.com/doc.pdf",
+        access_token=None,
+        model=Model.PADDLE_OCR_VL_15,
+    )
+    monkeypatch.setenv("PADDLEOCR_ACCESS_TOKEN", "test-token")
+
+    list(loader.lazy_load())
+
+    mock_client.parse_document.assert_called_once_with(
+        file_url="https://example.com/doc.pdf",
+        model=Model.PADDLE_OCR_VL_15,
+        options=loader._options,
+    )
diff --git a/langchain-paddleocr/tests/unit_tests/document_loaders/test_paddleocr_vl_param_mapping.py b/langchain-paddleocr/tests/unit_tests/document_loaders/test_paddleocr_vl_param_mapping.py
new file mode 100644
index 0000000..c772105
--- /dev/null
+++ b/langchain-paddleocr/tests/unit_tests/document_loaders/test_paddleocr_vl_param_mapping.py
@@ -0,0 +1,48 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import annotations
+
+from paddleocr import PaddleOCRVLOptions
+
+from langchain_paddleocr import PaddleOCRVLLoader
+
+
+def test_vl_options_are_mapped_to_paddleocr_vl_options() -> None:
+    loader = PaddleOCRVLLoader(
+        file_path="dummy.pdf",
+        use_doc_orientation_classify=True,
+        layout_unclip_ratio=(0.1, 0.9),
+        prompt_label="ocr",
+        use_ocr_for_image_block=True,
+        format_block_content=True,
+        markdown_ignore_labels=["image"],
+        vlm_extra_args={"temperature": 0.1},
+        return_markdown_images=False,
+        output_formats=["docx"],
+        extra_options={"futureOption": "enabled"},
+    )
+
+    assert isinstance(loader._options, PaddleOCRVLOptions)
+    assert loader._options.use_doc_orientation_classify is True
+    assert loader._options.layout_unclip_ratio == (0.1, 0.9)
+    assert loader._options.prompt_label == "ocr"
+    assert loader._options.use_ocr_for_image_block is True
+    assert loader._options.format_block_content is True
+    assert loader._options.markdown_ignore_labels == ["image"]
+    assert loader._options.vlm_extra_args == {"temperature": 0.1}
+    assert loader._options.return_markdown_images is False
+    assert loader._options.output_formats == ["docx"]
+    assert loader._options.extra_options == {"futureOption": "enabled"}
+    assert loader._options.to_payload()["futureOption"] == "enabled"
diff --git a/langchain-paddleocr/uv.lock b/langchain-paddleocr/uv.lock
new file mode 100644
index 0000000..12ac51a
--- /dev/null
+++ b/langchain-paddleocr/uv.lock
@@ -0,0 +1,3403 @@
+version = 1
+revision = 3
+requires-python = ">=3.10"
+resolution-markers = [
+    "python_full_version >= '3.14' and sys_platform == 'darwin'",
+    "python_full_version == '3.13.*' and sys_platform == 'darwin'",
+    "python_full_version >= '3.14' and platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "python_full_version == '3.13.*' and platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "python_full_version >= '3.14' and sys_platform == 'win32'",
+    "python_full_version >= '3.14' and sys_platform == 'emscripten'",
+    "(python_full_version >= '3.14' and platform_machine != 'aarch64' and sys_platform == 'linux') or (python_full_version >= '3.14' and sys_platform != 'darwin' and sys_platform != 'emscripten' and sys_platform != 'linux' and sys_platform != 'win32')",
+    "python_full_version == '3.13.*' and sys_platform == 'win32'",
+    "python_full_version == '3.13.*' and sys_platform == 'emscripten'",
+    "(python_full_version == '3.13.*' and platform_machine != 'aarch64' and sys_platform == 'linux') or (python_full_version == '3.13.*' and sys_platform != 'darwin' and sys_platform != 'emscripten' and sys_platform != 'linux' and sys_platform != 'win32')",
+    "python_full_version == '3.12.*' and sys_platform == 'darwin'",
+    "python_full_version == '3.12.*' and platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "python_full_version == '3.12.*' and sys_platform == 'win32'",
+    "python_full_version == '3.12.*' and sys_platform == 'emscripten'",
+    "(python_full_version == '3.12.*' and platform_machine != 'aarch64' and sys_platform == 'linux') or (python_full_version == '3.12.*' and sys_platform != 'darwin' and sys_platform != 'emscripten' and sys_platform != 'linux' and sys_platform != 'win32')",
+    "python_full_version == '3.11.*' and sys_platform == 'darwin'",
+    "python_full_version == '3.11.*' and platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "python_full_version == '3.11.*' and sys_platform == 'win32'",
+    "python_full_version == '3.11.*' and sys_platform == 'emscripten'",
+    "(python_full_version == '3.11.*' and platform_machine != 'aarch64' and sys_platform == 'linux') or (python_full_version == '3.11.*' and sys_platform != 'darwin' and sys_platform != 'emscripten' and sys_platform != 'linux' and sys_platform != 'win32')",
+    "python_full_version < '3.11' and sys_platform == 'darwin'",
+    "python_full_version < '3.11' and platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "(python_full_version < '3.11' and platform_machine != 'aarch64' and sys_platform == 'linux') or (python_full_version < '3.11' and sys_platform != 'darwin' and sys_platform != 'linux')",
+]
+
+[[package]]
+name = "aiohappyeyeballs"
+version = "2.6.1"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/26/30/f84a107a9c4331c14b2b586036f40965c128aa4fee4dda5d3d51cb14ad54/aiohappyeyeballs-2.6.1.tar.gz", hash = "sha256:c3f9d0113123803ccadfdf3f0faa505bc78e6a72d1cc4806cbd719826e943558", size = 22760, upload-time = "2025-03-12T01:42:48.764Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/0f/15/5bf3b99495fb160b63f95972b81750f18f7f4e02ad051373b669d17d44f2/aiohappyeyeballs-2.6.1-py3-none-any.whl", hash = "sha256:f349ba8f4b75cb25c99c5c2d84e997e485204d2902a9597802b0371f09331fb8", size = 15265, upload-time = "2025-03-12T01:42:47.083Z" },
+]
+
+[[package]]
+name = "aiohttp"
+version = "3.13.3"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "aiohappyeyeballs" },
+    { name = "aiosignal" },
+    { name = "async-timeout", marker = "python_full_version < '3.11'" },
+    { name = "attrs" },
+    { name = "frozenlist" },
+    { name = "multidict" },
+    { name = "propcache" },
+    { name = "yarl" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/50/42/32cf8e7704ceb4481406eb87161349abb46a57fee3f008ba9cb610968646/aiohttp-3.13.3.tar.gz", hash = "sha256:a949eee43d3782f2daae4f4a2819b2cb9b0c5d3b7f7a927067cc84dafdbb9f88", size = 7844556, upload-time = "2026-01-03T17:33:05.204Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/36/d6/5aec9313ee6ea9c7cde8b891b69f4ff4001416867104580670a31daeba5b/aiohttp-3.13.3-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:d5a372fd5afd301b3a89582817fdcdb6c34124787c70dbcc616f259013e7eef7", size = 738950, upload-time = "2026-01-03T17:29:13.002Z" },
+    { url = "https://files.pythonhosted.org/packages/68/03/8fa90a7e6d11ff20a18837a8e2b5dd23db01aabc475aa9271c8ad33299f5/aiohttp-3.13.3-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:147e422fd1223005c22b4fe080f5d93ced44460f5f9c105406b753612b587821", size = 496099, upload-time = "2026-01-03T17:29:15.268Z" },
+    { url = "https://files.pythonhosted.org/packages/d2/23/b81f744d402510a8366b74eb420fc0cc1170d0c43daca12d10814df85f10/aiohttp-3.13.3-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:859bd3f2156e81dd01432f5849fc73e2243d4a487c4fd26609b1299534ee1845", size = 491072, upload-time = "2026-01-03T17:29:16.922Z" },
+    { url = "https://files.pythonhosted.org/packages/d5/e1/56d1d1c0dd334cd203dd97706ce004c1aa24b34a813b0b8daf3383039706/aiohttp-3.13.3-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:dca68018bf48c251ba17c72ed479f4dafe9dbd5a73707ad8d28a38d11f3d42af", size = 1671588, upload-time = "2026-01-03T17:29:18.539Z" },
+    { url = "https://files.pythonhosted.org/packages/5f/34/8d7f962604f4bc2b4e39eb1220dac7d4e4cba91fb9ba0474b4ecd67db165/aiohttp-3.13.3-cp310-cp310-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:fee0c6bc7db1de362252affec009707a17478a00ec69f797d23ca256e36d5940", size = 1640334, upload-time = "2026-01-03T17:29:21.028Z" },
+    { url = "https://files.pythonhosted.org/packages/94/1d/fcccf2c668d87337ddeef9881537baee13c58d8f01f12ba8a24215f2b804/aiohttp-3.13.3-cp310-cp310-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:c048058117fd649334d81b4b526e94bde3ccaddb20463a815ced6ecbb7d11160", size = 1722656, upload-time = "2026-01-03T17:29:22.531Z" },
+    { url = "https://files.pythonhosted.org/packages/aa/98/c6f3b081c4c606bc1e5f2ec102e87d6411c73a9ef3616fea6f2d5c98c062/aiohttp-3.13.3-cp310-cp310-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:215a685b6fbbfcf71dfe96e3eba7a6f58f10da1dfdf4889c7dd856abe430dca7", size = 1817625, upload-time = "2026-01-03T17:29:24.276Z" },
+    { url = "https://files.pythonhosted.org/packages/2c/c0/cfcc3d2e11b477f86e1af2863f3858c8850d751ce8dc39c4058a072c9e54/aiohttp-3.13.3-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:de2c184bb1fe2cbd2cefba613e9db29a5ab559323f994b6737e370d3da0ac455", size = 1672604, upload-time = "2026-01-03T17:29:26.099Z" },
+    { url = "https://files.pythonhosted.org/packages/1e/77/6b4ffcbcac4c6a5d041343a756f34a6dd26174ae07f977a64fe028dda5b0/aiohttp-3.13.3-cp310-cp310-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:75ca857eba4e20ce9f546cd59c7007b33906a4cd48f2ff6ccf1ccfc3b646f279", size = 1554370, upload-time = "2026-01-03T17:29:28.121Z" },
+    { url = "https://files.pythonhosted.org/packages/f2/f0/e3ddfa93f17d689dbe014ba048f18e0c9f9b456033b70e94349a2e9048be/aiohttp-3.13.3-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:81e97251d9298386c2b7dbeb490d3d1badbdc69107fb8c9299dd04eb39bddc0e", size = 1642023, upload-time = "2026-01-03T17:29:30.002Z" },
+    { url = "https://files.pythonhosted.org/packages/eb/45/c14019c9ec60a8e243d06d601b33dcc4fd92379424bde3021725859d7f99/aiohttp-3.13.3-cp310-cp310-musllinux_1_2_armv7l.whl", hash = "sha256:c0e2d366af265797506f0283487223146af57815b388623f0357ef7eac9b209d", size = 1649680, upload-time = "2026-01-03T17:29:31.782Z" },
+    { url = "https://files.pythonhosted.org/packages/9c/fd/09c9451dae5aa5c5ed756df95ff9ef549d45d4be663bafd1e4954fd836f0/aiohttp-3.13.3-cp310-cp310-musllinux_1_2_ppc64le.whl", hash = "sha256:4e239d501f73d6db1522599e14b9b321a7e3b1de66ce33d53a765d975e9f4808", size = 1692407, upload-time = "2026-01-03T17:29:33.392Z" },
+    { url = "https://files.pythonhosted.org/packages/a6/81/938bc2ec33c10efd6637ccb3d22f9f3160d08e8f3aa2587a2c2d5ab578eb/aiohttp-3.13.3-cp310-cp310-musllinux_1_2_riscv64.whl", hash = "sha256:0db318f7a6f065d84cb1e02662c526294450b314a02bd9e2a8e67f0d8564ce40", size = 1543047, upload-time = "2026-01-03T17:29:34.855Z" },
+    { url = "https://files.pythonhosted.org/packages/f7/23/80488ee21c8d567c83045e412e1d9b7077d27171591a4eb7822586e8c06a/aiohttp-3.13.3-cp310-cp310-musllinux_1_2_s390x.whl", hash = "sha256:bfc1cc2fe31a6026a8a88e4ecfb98d7f6b1fec150cfd708adbfd1d2f42257c29", size = 1715264, upload-time = "2026-01-03T17:29:36.389Z" },
+    { url = "https://files.pythonhosted.org/packages/e2/83/259a8da6683182768200b368120ab3deff5370bed93880fb9a3a86299f34/aiohttp-3.13.3-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:af71fff7bac6bb7508956696dce8f6eec2bbb045eceb40343944b1ae62b5ef11", size = 1657275, upload-time = "2026-01-03T17:29:38.162Z" },
+    { url = "https://files.pythonhosted.org/packages/3f/4f/2c41f800a0b560785c10fb316216ac058c105f9be50bdc6a285de88db625/aiohttp-3.13.3-cp310-cp310-win32.whl", hash = "sha256:37da61e244d1749798c151421602884db5270faf479cf0ef03af0ff68954c9dd", size = 434053, upload-time = "2026-01-03T17:29:40.074Z" },
+    { url = "https://files.pythonhosted.org/packages/80/df/29cd63c7ecfdb65ccc12f7d808cac4fa2a19544660c06c61a4a48462de0c/aiohttp-3.13.3-cp310-cp310-win_amd64.whl", hash = "sha256:7e63f210bc1b57ef699035f2b4b6d9ce096b5914414a49b0997c839b2bd2223c", size = 456687, upload-time = "2026-01-03T17:29:41.819Z" },
+    { url = "https://files.pythonhosted.org/packages/f1/4c/a164164834f03924d9a29dc3acd9e7ee58f95857e0b467f6d04298594ebb/aiohttp-3.13.3-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:5b6073099fb654e0a068ae678b10feff95c5cae95bbfcbfa7af669d361a8aa6b", size = 746051, upload-time = "2026-01-03T17:29:43.287Z" },
+    { url = "https://files.pythonhosted.org/packages/82/71/d5c31390d18d4f58115037c432b7e0348c60f6f53b727cad33172144a112/aiohttp-3.13.3-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:1cb93e166e6c28716c8c6aeb5f99dfb6d5ccf482d29fe9bf9a794110e6d0ab64", size = 499234, upload-time = "2026-01-03T17:29:44.822Z" },
+    { url = "https://files.pythonhosted.org/packages/0e/c9/741f8ac91e14b1d2e7100690425a5b2b919a87a5075406582991fb7de920/aiohttp-3.13.3-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:28e027cf2f6b641693a09f631759b4d9ce9165099d2b5d92af9bd4e197690eea", size = 494979, upload-time = "2026-01-03T17:29:46.405Z" },
+    { url = "https://files.pythonhosted.org/packages/75/b5/31d4d2e802dfd59f74ed47eba48869c1c21552c586d5e81a9d0d5c2ad640/aiohttp-3.13.3-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:3b61b7169ababd7802f9568ed96142616a9118dd2be0d1866e920e77ec8fa92a", size = 1748297, upload-time = "2026-01-03T17:29:48.083Z" },
+    { url = "https://files.pythonhosted.org/packages/1a/3e/eefad0ad42959f226bb79664826883f2687d602a9ae2941a18e0484a74d3/aiohttp-3.13.3-cp311-cp311-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:80dd4c21b0f6237676449c6baaa1039abae86b91636b6c91a7f8e61c87f89540", size = 1707172, upload-time = "2026-01-03T17:29:49.648Z" },
+    { url = "https://files.pythonhosted.org/packages/c5/3a/54a64299fac2891c346cdcf2aa6803f994a2e4beeaf2e5a09dcc54acc842/aiohttp-3.13.3-cp311-cp311-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:65d2ccb7eabee90ce0503c17716fc77226be026dcc3e65cce859a30db715025b", size = 1805405, upload-time = "2026-01-03T17:29:51.244Z" },
+    { url = "https://files.pythonhosted.org/packages/6c/70/ddc1b7169cf64075e864f64595a14b147a895a868394a48f6a8031979038/aiohttp-3.13.3-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:5b179331a481cb5529fca8b432d8d3c7001cb217513c94cd72d668d1248688a3", size = 1899449, upload-time = "2026-01-03T17:29:53.938Z" },
+    { url = "https://files.pythonhosted.org/packages/a1/7e/6815aab7d3a56610891c76ef79095677b8b5be6646aaf00f69b221765021/aiohttp-3.13.3-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:9d4c940f02f49483b18b079d1c27ab948721852b281f8b015c058100e9421dd1", size = 1748444, upload-time = "2026-01-03T17:29:55.484Z" },
+    { url = "https://files.pythonhosted.org/packages/6b/f2/073b145c4100da5511f457dc0f7558e99b2987cf72600d42b559db856fbc/aiohttp-3.13.3-cp311-cp311-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:f9444f105664c4ce47a2a7171a2418bce5b7bae45fb610f4e2c36045d85911d3", size = 1606038, upload-time = "2026-01-03T17:29:57.179Z" },
+    { url = "https://files.pythonhosted.org/packages/0a/c1/778d011920cae03ae01424ec202c513dc69243cf2db303965615b81deeea/aiohttp-3.13.3-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:694976222c711d1d00ba131904beb60534f93966562f64440d0c9d41b8cdb440", size = 1724156, upload-time = "2026-01-03T17:29:58.914Z" },
+    { url = "https://files.pythonhosted.org/packages/0e/cb/3419eabf4ec1e9ec6f242c32b689248365a1cf621891f6f0386632525494/aiohttp-3.13.3-cp311-cp311-musllinux_1_2_armv7l.whl", hash = "sha256:f33ed1a2bf1997a36661874b017f5c4b760f41266341af36febaf271d179f6d7", size = 1722340, upload-time = "2026-01-03T17:30:01.962Z" },
+    { url = "https://files.pythonhosted.org/packages/7a/e5/76cf77bdbc435bf233c1f114edad39ed4177ccbfab7c329482b179cff4f4/aiohttp-3.13.3-cp311-cp311-musllinux_1_2_ppc64le.whl", hash = "sha256:e636b3c5f61da31a92bf0d91da83e58fdfa96f178ba682f11d24f31944cdd28c", size = 1783041, upload-time = "2026-01-03T17:30:03.609Z" },
+    { url = "https://files.pythonhosted.org/packages/9d/d4/dd1ca234c794fd29c057ce8c0566b8ef7fd6a51069de5f06fa84b9a1971c/aiohttp-3.13.3-cp311-cp311-musllinux_1_2_riscv64.whl", hash = "sha256:5d2d94f1f5fcbe40838ac51a6ab5704a6f9ea42e72ceda48de5e6b898521da51", size = 1596024, upload-time = "2026-01-03T17:30:05.132Z" },
+    { url = "https://files.pythonhosted.org/packages/55/58/4345b5f26661a6180afa686c473620c30a66afdf120ed3dd545bbc809e85/aiohttp-3.13.3-cp311-cp311-musllinux_1_2_s390x.whl", hash = "sha256:2be0e9ccf23e8a94f6f0650ce06042cefc6ac703d0d7ab6c7a917289f2539ad4", size = 1804590, upload-time = "2026-01-03T17:30:07.135Z" },
+    { url = "https://files.pythonhosted.org/packages/7b/06/05950619af6c2df7e0a431d889ba2813c9f0129cec76f663e547a5ad56f2/aiohttp-3.13.3-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:9af5e68ee47d6534d36791bbe9b646d2a7c7deb6fc24d7943628edfbb3581f29", size = 1740355, upload-time = "2026-01-03T17:30:09.083Z" },
+    { url = "https://files.pythonhosted.org/packages/3e/80/958f16de79ba0422d7c1e284b2abd0c84bc03394fbe631d0a39ffa10e1eb/aiohttp-3.13.3-cp311-cp311-win32.whl", hash = "sha256:a2212ad43c0833a873d0fb3c63fa1bacedd4cf6af2fee62bf4b739ceec3ab239", size = 433701, upload-time = "2026-01-03T17:30:10.869Z" },
+    { url = "https://files.pythonhosted.org/packages/dc/f2/27cdf04c9851712d6c1b99df6821a6623c3c9e55956d4b1e318c337b5a48/aiohttp-3.13.3-cp311-cp311-win_amd64.whl", hash = "sha256:642f752c3eb117b105acbd87e2c143de710987e09860d674e068c4c2c441034f", size = 457678, upload-time = "2026-01-03T17:30:12.719Z" },
+    { url = "https://files.pythonhosted.org/packages/a0/be/4fc11f202955a69e0db803a12a062b8379c970c7c84f4882b6da17337cc1/aiohttp-3.13.3-cp312-cp312-macosx_10_13_universal2.whl", hash = "sha256:b903a4dfee7d347e2d87697d0713be59e0b87925be030c9178c5faa58ea58d5c", size = 739732, upload-time = "2026-01-03T17:30:14.23Z" },
+    { url = "https://files.pythonhosted.org/packages/97/2c/621d5b851f94fa0bb7430d6089b3aa970a9d9b75196bc93bb624b0db237a/aiohttp-3.13.3-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:a45530014d7a1e09f4a55f4f43097ba0fd155089372e105e4bff4ca76cb1b168", size = 494293, upload-time = "2026-01-03T17:30:15.96Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/43/4be01406b78e1be8320bb8316dc9c42dbab553d281c40364e0f862d5661c/aiohttp-3.13.3-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:27234ef6d85c914f9efeb77ff616dbf4ad2380be0cda40b4db086ffc7ddd1b7d", size = 493533, upload-time = "2026-01-03T17:30:17.431Z" },
+    { url = "https://files.pythonhosted.org/packages/8d/a8/5a35dc56a06a2c90d4742cbf35294396907027f80eea696637945a106f25/aiohttp-3.13.3-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:d32764c6c9aafb7fb55366a224756387cd50bfa720f32b88e0e6fa45b27dcf29", size = 1737839, upload-time = "2026-01-03T17:30:19.422Z" },
+    { url = "https://files.pythonhosted.org/packages/bf/62/4b9eeb331da56530bf2e198a297e5303e1c1ebdceeb00fe9b568a65c5a0c/aiohttp-3.13.3-cp312-cp312-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:b1a6102b4d3ebc07dad44fbf07b45bb600300f15b552ddf1851b5390202ea2e3", size = 1703932, upload-time = "2026-01-03T17:30:21.756Z" },
+    { url = "https://files.pythonhosted.org/packages/7c/f6/af16887b5d419e6a367095994c0b1332d154f647e7dc2bd50e61876e8e3d/aiohttp-3.13.3-cp312-cp312-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:c014c7ea7fb775dd015b2d3137378b7be0249a448a1612268b5a90c2d81de04d", size = 1771906, upload-time = "2026-01-03T17:30:23.932Z" },
+    { url = "https://files.pythonhosted.org/packages/ce/83/397c634b1bcc24292fa1e0c7822800f9f6569e32934bdeef09dae7992dfb/aiohttp-3.13.3-cp312-cp312-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:2b8d8ddba8f95ba17582226f80e2de99c7a7948e66490ef8d947e272a93e9463", size = 1871020, upload-time = "2026-01-03T17:30:26Z" },
+    { url = "https://files.pythonhosted.org/packages/86/f6/a62cbbf13f0ac80a70f71b1672feba90fdb21fd7abd8dbf25c0105fb6fa3/aiohttp-3.13.3-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:9ae8dd55c8e6c4257eae3a20fd2c8f41edaea5992ed67156642493b8daf3cecc", size = 1755181, upload-time = "2026-01-03T17:30:27.554Z" },
+    { url = "https://files.pythonhosted.org/packages/0a/87/20a35ad487efdd3fba93d5843efdfaa62d2f1479eaafa7453398a44faf13/aiohttp-3.13.3-cp312-cp312-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:01ad2529d4b5035578f5081606a465f3b814c542882804e2e8cda61adf5c71bf", size = 1561794, upload-time = "2026-01-03T17:30:29.254Z" },
+    { url = "https://files.pythonhosted.org/packages/de/95/8fd69a66682012f6716e1bc09ef8a1a2a91922c5725cb904689f112309c4/aiohttp-3.13.3-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:bb4f7475e359992b580559e008c598091c45b5088f28614e855e42d39c2f1033", size = 1697900, upload-time = "2026-01-03T17:30:31.033Z" },
+    { url = "https://files.pythonhosted.org/packages/e5/66/7b94b3b5ba70e955ff597672dad1691333080e37f50280178967aff68657/aiohttp-3.13.3-cp312-cp312-musllinux_1_2_armv7l.whl", hash = "sha256:c19b90316ad3b24c69cd78d5c9b4f3aa4497643685901185b65166293d36a00f", size = 1728239, upload-time = "2026-01-03T17:30:32.703Z" },
+    { url = "https://files.pythonhosted.org/packages/47/71/6f72f77f9f7d74719692ab65a2a0252584bf8d5f301e2ecb4c0da734530a/aiohttp-3.13.3-cp312-cp312-musllinux_1_2_ppc64le.whl", hash = "sha256:96d604498a7c782cb15a51c406acaea70d8c027ee6b90c569baa6e7b93073679", size = 1740527, upload-time = "2026-01-03T17:30:34.695Z" },
+    { url = "https://files.pythonhosted.org/packages/fa/b4/75ec16cbbd5c01bdaf4a05b19e103e78d7ce1ef7c80867eb0ace42ff4488/aiohttp-3.13.3-cp312-cp312-musllinux_1_2_riscv64.whl", hash = "sha256:084911a532763e9d3dd95adf78a78f4096cd5f58cdc18e6fdbc1b58417a45423", size = 1554489, upload-time = "2026-01-03T17:30:36.864Z" },
+    { url = "https://files.pythonhosted.org/packages/52/8f/bc518c0eea29f8406dcf7ed1f96c9b48e3bc3995a96159b3fc11f9e08321/aiohttp-3.13.3-cp312-cp312-musllinux_1_2_s390x.whl", hash = "sha256:7a4a94eb787e606d0a09404b9c38c113d3b099d508021faa615d70a0131907ce", size = 1767852, upload-time = "2026-01-03T17:30:39.433Z" },
+    { url = "https://files.pythonhosted.org/packages/9d/f2/a07a75173124f31f11ea6f863dc44e6f09afe2bca45dd4e64979490deab1/aiohttp-3.13.3-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:87797e645d9d8e222e04160ee32aa06bc5c163e8499f24db719e7852ec23093a", size = 1722379, upload-time = "2026-01-03T17:30:41.081Z" },
+    { url = "https://files.pythonhosted.org/packages/3c/4a/1a3fee7c21350cac78e5c5cef711bac1b94feca07399f3d406972e2d8fcd/aiohttp-3.13.3-cp312-cp312-win32.whl", hash = "sha256:b04be762396457bef43f3597c991e192ee7da460a4953d7e647ee4b1c28e7046", size = 428253, upload-time = "2026-01-03T17:30:42.644Z" },
+    { url = "https://files.pythonhosted.org/packages/d9/b7/76175c7cb4eb73d91ad63c34e29fc4f77c9386bba4a65b53ba8e05ee3c39/aiohttp-3.13.3-cp312-cp312-win_amd64.whl", hash = "sha256:e3531d63d3bdfa7e3ac5e9b27b2dd7ec9df3206a98e0b3445fa906f233264c57", size = 455407, upload-time = "2026-01-03T17:30:44.195Z" },
+    { url = "https://files.pythonhosted.org/packages/97/8a/12ca489246ca1faaf5432844adbfce7ff2cc4997733e0af120869345643a/aiohttp-3.13.3-cp313-cp313-macosx_10_13_universal2.whl", hash = "sha256:5dff64413671b0d3e7d5918ea490bdccb97a4ad29b3f311ed423200b2203e01c", size = 734190, upload-time = "2026-01-03T17:30:45.832Z" },
+    { url = "https://files.pythonhosted.org/packages/32/08/de43984c74ed1fca5c014808963cc83cb00d7bb06af228f132d33862ca76/aiohttp-3.13.3-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:87b9aab6d6ed88235aa2970294f496ff1a1f9adcd724d800e9b952395a80ffd9", size = 491783, upload-time = "2026-01-03T17:30:47.466Z" },
+    { url = "https://files.pythonhosted.org/packages/17/f8/8dd2cf6112a5a76f81f81a5130c57ca829d101ad583ce57f889179accdda/aiohttp-3.13.3-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:425c126c0dc43861e22cb1c14ba4c8e45d09516d0a3ae0a3f7494b79f5f233a3", size = 490704, upload-time = "2026-01-03T17:30:49.373Z" },
+    { url = "https://files.pythonhosted.org/packages/6d/40/a46b03ca03936f832bc7eaa47cfbb1ad012ba1be4790122ee4f4f8cba074/aiohttp-3.13.3-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:7f9120f7093c2a32d9647abcaf21e6ad275b4fbec5b55969f978b1a97c7c86bf", size = 1720652, upload-time = "2026-01-03T17:30:50.974Z" },
+    { url = "https://files.pythonhosted.org/packages/f7/7e/917fe18e3607af92657e4285498f500dca797ff8c918bd7d90b05abf6c2a/aiohttp-3.13.3-cp313-cp313-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:697753042d57f4bf7122cab985bf15d0cef23c770864580f5af4f52023a56bd6", size = 1692014, upload-time = "2026-01-03T17:30:52.729Z" },
+    { url = "https://files.pythonhosted.org/packages/71/b6/cefa4cbc00d315d68973b671cf105b21a609c12b82d52e5d0c9ae61d2a09/aiohttp-3.13.3-cp313-cp313-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:6de499a1a44e7de70735d0b39f67c8f25eb3d91eb3103be99ca0fa882cdd987d", size = 1759777, upload-time = "2026-01-03T17:30:54.537Z" },
+    { url = "https://files.pythonhosted.org/packages/fb/e3/e06ee07b45e59e6d81498b591fc589629be1553abb2a82ce33efe2a7b068/aiohttp-3.13.3-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:37239e9f9a7ea9ac5bf6b92b0260b01f8a22281996da609206a84df860bc1261", size = 1861276, upload-time = "2026-01-03T17:30:56.512Z" },
+    { url = "https://files.pythonhosted.org/packages/7c/24/75d274228acf35ceeb2850b8ce04de9dd7355ff7a0b49d607ee60c29c518/aiohttp-3.13.3-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:f76c1e3fe7d7c8afad7ed193f89a292e1999608170dcc9751a7462a87dfd5bc0", size = 1743131, upload-time = "2026-01-03T17:30:58.256Z" },
+    { url = "https://files.pythonhosted.org/packages/04/98/3d21dde21889b17ca2eea54fdcff21b27b93f45b7bb94ca029c31ab59dc3/aiohttp-3.13.3-cp313-cp313-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:fc290605db2a917f6e81b0e1e0796469871f5af381ce15c604a3c5c7e51cb730", size = 1556863, upload-time = "2026-01-03T17:31:00.445Z" },
+    { url = "https://files.pythonhosted.org/packages/9e/84/da0c3ab1192eaf64782b03971ab4055b475d0db07b17eff925e8c93b3aa5/aiohttp-3.13.3-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:4021b51936308aeea0367b8f006dc999ca02bc118a0cc78c303f50a2ff6afb91", size = 1682793, upload-time = "2026-01-03T17:31:03.024Z" },
+    { url = "https://files.pythonhosted.org/packages/ff/0f/5802ada182f575afa02cbd0ec5180d7e13a402afb7c2c03a9aa5e5d49060/aiohttp-3.13.3-cp313-cp313-musllinux_1_2_armv7l.whl", hash = "sha256:49a03727c1bba9a97d3e93c9f93ca03a57300f484b6e935463099841261195d3", size = 1716676, upload-time = "2026-01-03T17:31:04.842Z" },
+    { url = "https://files.pythonhosted.org/packages/3f/8c/714d53bd8b5a4560667f7bbbb06b20c2382f9c7847d198370ec6526af39c/aiohttp-3.13.3-cp313-cp313-musllinux_1_2_ppc64le.whl", hash = "sha256:3d9908a48eb7416dc1f4524e69f1d32e5d90e3981e4e37eb0aa1cd18f9cfa2a4", size = 1733217, upload-time = "2026-01-03T17:31:06.868Z" },
+    { url = "https://files.pythonhosted.org/packages/7d/79/e2176f46d2e963facea939f5be2d26368ce543622be6f00a12844d3c991f/aiohttp-3.13.3-cp313-cp313-musllinux_1_2_riscv64.whl", hash = "sha256:2712039939ec963c237286113c68dbad80a82a4281543f3abf766d9d73228998", size = 1552303, upload-time = "2026-01-03T17:31:08.958Z" },
+    { url = "https://files.pythonhosted.org/packages/ab/6a/28ed4dea1759916090587d1fe57087b03e6c784a642b85ef48217b0277ae/aiohttp-3.13.3-cp313-cp313-musllinux_1_2_s390x.whl", hash = "sha256:7bfdc049127717581866fa4708791220970ce291c23e28ccf3922c700740fdc0", size = 1763673, upload-time = "2026-01-03T17:31:10.676Z" },
+    { url = "https://files.pythonhosted.org/packages/e8/35/4a3daeb8b9fab49240d21c04d50732313295e4bd813a465d840236dd0ce1/aiohttp-3.13.3-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:8057c98e0c8472d8846b9c79f56766bcc57e3e8ac7bfd510482332366c56c591", size = 1721120, upload-time = "2026-01-03T17:31:12.575Z" },
+    { url = "https://files.pythonhosted.org/packages/bc/9f/d643bb3c5fb99547323e635e251c609fbbc660d983144cfebec529e09264/aiohttp-3.13.3-cp313-cp313-win32.whl", hash = "sha256:1449ceddcdbcf2e0446957863af03ebaaa03f94c090f945411b61269e2cb5daf", size = 427383, upload-time = "2026-01-03T17:31:14.382Z" },
+    { url = "https://files.pythonhosted.org/packages/4e/f1/ab0395f8a79933577cdd996dd2f9aa6014af9535f65dddcf88204682fe62/aiohttp-3.13.3-cp313-cp313-win_amd64.whl", hash = "sha256:693781c45a4033d31d4187d2436f5ac701e7bbfe5df40d917736108c1cc7436e", size = 453899, upload-time = "2026-01-03T17:31:15.958Z" },
+    { url = "https://files.pythonhosted.org/packages/99/36/5b6514a9f5d66f4e2597e40dea2e3db271e023eb7a5d22defe96ba560996/aiohttp-3.13.3-cp314-cp314-macosx_10_13_universal2.whl", hash = "sha256:ea37047c6b367fd4bd632bff8077449b8fa034b69e812a18e0132a00fae6e808", size = 737238, upload-time = "2026-01-03T17:31:17.909Z" },
+    { url = "https://files.pythonhosted.org/packages/f7/49/459327f0d5bcd8c6c9ca69e60fdeebc3622861e696490d8674a6d0cb90a6/aiohttp-3.13.3-cp314-cp314-macosx_10_13_x86_64.whl", hash = "sha256:6fc0e2337d1a4c3e6acafda6a78a39d4c14caea625124817420abceed36e2415", size = 492292, upload-time = "2026-01-03T17:31:19.919Z" },
+    { url = "https://files.pythonhosted.org/packages/e8/0b/b97660c5fd05d3495b4eb27f2d0ef18dc1dc4eff7511a9bf371397ff0264/aiohttp-3.13.3-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:c685f2d80bb67ca8c3837823ad76196b3694b0159d232206d1e461d3d434666f", size = 493021, upload-time = "2026-01-03T17:31:21.636Z" },
+    { url = "https://files.pythonhosted.org/packages/54/d4/438efabdf74e30aeceb890c3290bbaa449780583b1270b00661126b8aae4/aiohttp-3.13.3-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:48e377758516d262bde50c2584fc6c578af272559c409eecbdd2bae1601184d6", size = 1717263, upload-time = "2026-01-03T17:31:23.296Z" },
+    { url = "https://files.pythonhosted.org/packages/71/f2/7bddc7fd612367d1459c5bcf598a9e8f7092d6580d98de0e057eb42697ad/aiohttp-3.13.3-cp314-cp314-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:34749271508078b261c4abb1767d42b8d0c0cc9449c73a4df494777dc55f0687", size = 1669107, upload-time = "2026-01-03T17:31:25.334Z" },
+    { url = "https://files.pythonhosted.org/packages/00/5a/1aeaecca40e22560f97610a329e0e5efef5e0b5afdf9f857f0d93839ab2e/aiohttp-3.13.3-cp314-cp314-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:82611aeec80eb144416956ec85b6ca45a64d76429c1ed46ae1b5f86c6e0c9a26", size = 1760196, upload-time = "2026-01-03T17:31:27.394Z" },
+    { url = "https://files.pythonhosted.org/packages/f8/f8/0ff6992bea7bd560fc510ea1c815f87eedd745fe035589c71ce05612a19a/aiohttp-3.13.3-cp314-cp314-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:2fff83cfc93f18f215896e3a190e8e5cb413ce01553901aca925176e7568963a", size = 1843591, upload-time = "2026-01-03T17:31:29.238Z" },
+    { url = "https://files.pythonhosted.org/packages/e3/d1/e30e537a15f53485b61f5be525f2157da719819e8377298502aebac45536/aiohttp-3.13.3-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:bbe7d4cecacb439e2e2a8a1a7b935c25b812af7a5fd26503a66dadf428e79ec1", size = 1720277, upload-time = "2026-01-03T17:31:31.053Z" },
+    { url = "https://files.pythonhosted.org/packages/84/45/23f4c451d8192f553d38d838831ebbc156907ea6e05557f39563101b7717/aiohttp-3.13.3-cp314-cp314-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:b928f30fe49574253644b1ca44b1b8adbd903aa0da4b9054a6c20fc7f4092a25", size = 1548575, upload-time = "2026-01-03T17:31:32.87Z" },
+    { url = "https://files.pythonhosted.org/packages/6a/ed/0a42b127a43712eda7807e7892c083eadfaf8429ca8fb619662a530a3aab/aiohttp-3.13.3-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:7b5e8fe4de30df199155baaf64f2fcd604f4c678ed20910db8e2c66dc4b11603", size = 1679455, upload-time = "2026-01-03T17:31:34.76Z" },
+    { url = "https://files.pythonhosted.org/packages/2e/b5/c05f0c2b4b4fe2c9d55e73b6d3ed4fd6c9dc2684b1d81cbdf77e7fad9adb/aiohttp-3.13.3-cp314-cp314-musllinux_1_2_armv7l.whl", hash = "sha256:8542f41a62bcc58fc7f11cf7c90e0ec324ce44950003feb70640fc2a9092c32a", size = 1687417, upload-time = "2026-01-03T17:31:36.699Z" },
+    { url = "https://files.pythonhosted.org/packages/c9/6b/915bc5dad66aef602b9e459b5a973529304d4e89ca86999d9d75d80cbd0b/aiohttp-3.13.3-cp314-cp314-musllinux_1_2_ppc64le.whl", hash = "sha256:5e1d8c8b8f1d91cd08d8f4a3c2b067bfca6ec043d3ff36de0f3a715feeedf926", size = 1729968, upload-time = "2026-01-03T17:31:38.622Z" },
+    { url = "https://files.pythonhosted.org/packages/11/3b/e84581290a9520024a08640b63d07673057aec5ca548177a82026187ba73/aiohttp-3.13.3-cp314-cp314-musllinux_1_2_riscv64.whl", hash = "sha256:90455115e5da1c3c51ab619ac57f877da8fd6d73c05aacd125c5ae9819582aba", size = 1545690, upload-time = "2026-01-03T17:31:40.57Z" },
+    { url = "https://files.pythonhosted.org/packages/f5/04/0c3655a566c43fd647c81b895dfe361b9f9ad6d58c19309d45cff52d6c3b/aiohttp-3.13.3-cp314-cp314-musllinux_1_2_s390x.whl", hash = "sha256:042e9e0bcb5fba81886c8b4fbb9a09d6b8a00245fd8d88e4d989c1f96c74164c", size = 1746390, upload-time = "2026-01-03T17:31:42.857Z" },
+    { url = "https://files.pythonhosted.org/packages/1f/53/71165b26978f719c3419381514c9690bd5980e764a09440a10bb816ea4ab/aiohttp-3.13.3-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:2eb752b102b12a76ca02dff751a801f028b4ffbbc478840b473597fc91a9ed43", size = 1702188, upload-time = "2026-01-03T17:31:44.984Z" },
+    { url = "https://files.pythonhosted.org/packages/29/a7/cbe6c9e8e136314fa1980da388a59d2f35f35395948a08b6747baebb6aa6/aiohttp-3.13.3-cp314-cp314-win32.whl", hash = "sha256:b556c85915d8efaed322bf1bdae9486aa0f3f764195a0fb6ee962e5c71ef5ce1", size = 433126, upload-time = "2026-01-03T17:31:47.463Z" },
+    { url = "https://files.pythonhosted.org/packages/de/56/982704adea7d3b16614fc5936014e9af85c0e34b58f9046655817f04306e/aiohttp-3.13.3-cp314-cp314-win_amd64.whl", hash = "sha256:9bf9f7a65e7aa20dd764151fb3d616c81088f91f8df39c3893a536e279b4b984", size = 459128, upload-time = "2026-01-03T17:31:49.2Z" },
+    { url = "https://files.pythonhosted.org/packages/6c/2a/3c79b638a9c3d4658d345339d22070241ea341ed4e07b5ac60fb0f418003/aiohttp-3.13.3-cp314-cp314t-macosx_10_13_universal2.whl", hash = "sha256:05861afbbec40650d8a07ea324367cb93e9e8cc7762e04dd4405df99fa65159c", size = 769512, upload-time = "2026-01-03T17:31:51.134Z" },
+    { url = "https://files.pythonhosted.org/packages/29/b9/3e5014d46c0ab0db8707e0ac2711ed28c4da0218c358a4e7c17bae0d8722/aiohttp-3.13.3-cp314-cp314t-macosx_10_13_x86_64.whl", hash = "sha256:2fc82186fadc4a8316768d61f3722c230e2c1dcab4200d52d2ebdf2482e47592", size = 506444, upload-time = "2026-01-03T17:31:52.85Z" },
+    { url = "https://files.pythonhosted.org/packages/90/03/c1d4ef9a054e151cd7839cdc497f2638f00b93cbe8043983986630d7a80c/aiohttp-3.13.3-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:0add0900ff220d1d5c5ebbf99ed88b0c1bbf87aa7e4262300ed1376a6b13414f", size = 510798, upload-time = "2026-01-03T17:31:54.91Z" },
+    { url = "https://files.pythonhosted.org/packages/ea/76/8c1e5abbfe8e127c893fe7ead569148a4d5a799f7cf958d8c09f3eedf097/aiohttp-3.13.3-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:568f416a4072fbfae453dcf9a99194bbb8bdeab718e08ee13dfa2ba0e4bebf29", size = 1868835, upload-time = "2026-01-03T17:31:56.733Z" },
+    { url = "https://files.pythonhosted.org/packages/8e/ac/984c5a6f74c363b01ff97adc96a3976d9c98940b8969a1881575b279ac5d/aiohttp-3.13.3-cp314-cp314t-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:add1da70de90a2569c5e15249ff76a631ccacfe198375eead4aadf3b8dc849dc", size = 1720486, upload-time = "2026-01-03T17:31:58.65Z" },
+    { url = "https://files.pythonhosted.org/packages/b2/9a/b7039c5f099c4eb632138728828b33428585031a1e658d693d41d07d89d1/aiohttp-3.13.3-cp314-cp314t-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:10b47b7ba335d2e9b1239fa571131a87e2d8ec96b333e68b2a305e7a98b0bae2", size = 1847951, upload-time = "2026-01-03T17:32:00.989Z" },
+    { url = "https://files.pythonhosted.org/packages/3c/02/3bec2b9a1ba3c19ff89a43a19324202b8eb187ca1e928d8bdac9bbdddebd/aiohttp-3.13.3-cp314-cp314t-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:3dd4dce1c718e38081c8f35f323209d4c1df7d4db4bab1b5c88a6b4d12b74587", size = 1941001, upload-time = "2026-01-03T17:32:03.122Z" },
+    { url = "https://files.pythonhosted.org/packages/37/df/d879401cedeef27ac4717f6426c8c36c3091c6e9f08a9178cc87549c537f/aiohttp-3.13.3-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:34bac00a67a812570d4a460447e1e9e06fae622946955f939051e7cc895cfab8", size = 1797246, upload-time = "2026-01-03T17:32:05.255Z" },
+    { url = "https://files.pythonhosted.org/packages/8d/15/be122de1f67e6953add23335c8ece6d314ab67c8bebb3f181063010795a7/aiohttp-3.13.3-cp314-cp314t-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:a19884d2ee70b06d9204b2727a7b9f983d0c684c650254679e716b0b77920632", size = 1627131, upload-time = "2026-01-03T17:32:07.607Z" },
+    { url = "https://files.pythonhosted.org/packages/12/12/70eedcac9134cfa3219ab7af31ea56bc877395b1ac30d65b1bc4b27d0438/aiohttp-3.13.3-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:5f8ca7f2bb6ba8348a3614c7918cc4bb73268c5ac2a207576b7afea19d3d9f64", size = 1795196, upload-time = "2026-01-03T17:32:09.59Z" },
+    { url = "https://files.pythonhosted.org/packages/32/11/b30e1b1cd1f3054af86ebe60df96989c6a414dd87e27ad16950eee420bea/aiohttp-3.13.3-cp314-cp314t-musllinux_1_2_armv7l.whl", hash = "sha256:b0d95340658b9d2f11d9697f59b3814a9d3bb4b7a7c20b131df4bcef464037c0", size = 1782841, upload-time = "2026-01-03T17:32:11.445Z" },
+    { url = "https://files.pythonhosted.org/packages/88/0d/d98a9367b38912384a17e287850f5695c528cff0f14f791ce8ee2e4f7796/aiohttp-3.13.3-cp314-cp314t-musllinux_1_2_ppc64le.whl", hash = "sha256:a1e53262fd202e4b40b70c3aff944a8155059beedc8a89bba9dc1f9ef06a1b56", size = 1795193, upload-time = "2026-01-03T17:32:13.705Z" },
+    { url = "https://files.pythonhosted.org/packages/43/a5/a2dfd1f5ff5581632c7f6a30e1744deda03808974f94f6534241ef60c751/aiohttp-3.13.3-cp314-cp314t-musllinux_1_2_riscv64.whl", hash = "sha256:d60ac9663f44168038586cab2157e122e46bdef09e9368b37f2d82d354c23f72", size = 1621979, upload-time = "2026-01-03T17:32:15.965Z" },
+    { url = "https://files.pythonhosted.org/packages/fa/f0/12973c382ae7c1cccbc4417e129c5bf54c374dfb85af70893646e1f0e749/aiohttp-3.13.3-cp314-cp314t-musllinux_1_2_s390x.whl", hash = "sha256:90751b8eed69435bac9ff4e3d2f6b3af1f57e37ecb0fbeee59c0174c9e2d41df", size = 1822193, upload-time = "2026-01-03T17:32:18.219Z" },
+    { url = "https://files.pythonhosted.org/packages/3c/5f/24155e30ba7f8c96918af1350eb0663e2430aad9e001c0489d89cd708ab1/aiohttp-3.13.3-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:fc353029f176fd2b3ec6cfc71be166aba1936fe5d73dd1992ce289ca6647a9aa", size = 1769801, upload-time = "2026-01-03T17:32:20.25Z" },
+    { url = "https://files.pythonhosted.org/packages/eb/f8/7314031ff5c10e6ece114da79b338ec17eeff3a079e53151f7e9f43c4723/aiohttp-3.13.3-cp314-cp314t-win32.whl", hash = "sha256:2e41b18a58da1e474a057b3d35248d8320029f61d70a37629535b16a0c8f3767", size = 466523, upload-time = "2026-01-03T17:32:22.215Z" },
+    { url = "https://files.pythonhosted.org/packages/b4/63/278a98c715ae467624eafe375542d8ba9b4383a016df8fdefe0ae28382a7/aiohttp-3.13.3-cp314-cp314t-win_amd64.whl", hash = "sha256:44531a36aa2264a1860089ffd4dce7baf875ee5a6079d5fb42e261c704ef7344", size = 499694, upload-time = "2026-01-03T17:32:24.546Z" },
+]
+
+[[package]]
+name = "aiosignal"
+version = "1.4.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "frozenlist" },
+    { name = "typing-extensions", marker = "python_full_version < '3.13'" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/61/62/06741b579156360248d1ec624842ad0edf697050bbaf7c3e46394e106ad1/aiosignal-1.4.0.tar.gz", hash = "sha256:f47eecd9468083c2029cc99945502cb7708b082c232f9aca65da147157b251c7", size = 25007, upload-time = "2025-07-03T22:54:43.528Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/fb/76/641ae371508676492379f16e2fa48f4e2c11741bd63c48be4b12a6b09cba/aiosignal-1.4.0-py3-none-any.whl", hash = "sha256:053243f8b92b990551949e63930a839ff0cf0b0ebbe0597b0f3fb19e1a0fe82e", size = 7490, upload-time = "2025-07-03T22:54:42.156Z" },
+]
+
+[[package]]
+name = "aistudio-sdk"
+version = "0.3.8"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "bce-python-sdk" },
+    { name = "click" },
+    { name = "prettytable" },
+    { name = "psutil" },
+    { name = "requests" },
+    { name = "tqdm" },
+]
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/cb/77/cd71a481bb7a76b0e9d0b6bf47711c627b1dd079001ea246893f19a9d04c/aistudio_sdk-0.3.8-py3-none-any.whl", hash = "sha256:bfc96af7243ac2ee3303014849aa4028717cce5afa622af8cfe3a1d44d1941d6", size = 62972, upload-time = "2025-09-19T11:42:39.384Z" },
+]
+
+[[package]]
+name = "annotated-doc"
+version = "0.0.4"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/57/ba/046ceea27344560984e26a590f90bc7f4a75b06701f653222458922b558c/annotated_doc-0.0.4.tar.gz", hash = "sha256:fbcda96e87e9c92ad167c2e53839e57503ecfda18804ea28102353485033faa4", size = 7288, upload-time = "2025-11-10T22:07:42.062Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/1e/d3/26bf1008eb3d2daa8ef4cacc7f3bfdc11818d111f7e2d0201bc6e3b49d45/annotated_doc-0.0.4-py3-none-any.whl", hash = "sha256:571ac1dc6991c450b25a9c2d84a3705e2ae7a53467b5d111c24fa8baabbed320", size = 5303, upload-time = "2025-11-10T22:07:40.673Z" },
+]
+
+[[package]]
+name = "annotated-types"
+version = "0.7.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/ee/67/531ea369ba64dcff5ec9c3402f9f51bf748cec26dde048a2f973a4eea7f5/annotated_types-0.7.0.tar.gz", hash = "sha256:aff07c09a53a08bc8cfccb9c85b05f1aa9a2a6f23728d790723543408344ce89", size = 16081, upload-time = "2024-05-20T21:33:25.928Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/78/b6/6307fbef88d9b5ee7421e68d78a9f162e0da4900bc5f5793f6d3d0e34fb8/annotated_types-0.7.0-py3-none-any.whl", hash = "sha256:1f02e8b43a8fbbc3f3e0d4f0f4bfc8131bcb4eebe8849b8e5c773f3a1c582a53", size = 13643, upload-time = "2024-05-20T21:33:24.1Z" },
+]
+
+[[package]]
+name = "anyio"
+version = "4.12.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "exceptiongroup", marker = "python_full_version < '3.11'" },
+    { name = "idna" },
+    { name = "typing-extensions", marker = "python_full_version < '3.13'" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/96/f0/5eb65b2bb0d09ac6776f2eb54adee6abe8228ea05b20a5ad0e4945de8aac/anyio-4.12.1.tar.gz", hash = "sha256:41cfcc3a4c85d3f05c932da7c26d0201ac36f72abd4435ba90d0464a3ffed703", size = 228685, upload-time = "2026-01-06T11:45:21.246Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/38/0e/27be9fdef66e72d64c0cdc3cc2823101b80585f8119b5c112c2e8f5f7dab/anyio-4.12.1-py3-none-any.whl", hash = "sha256:d405828884fc140aa80a3c667b8beed277f1dfedec42ba031bd6ac3db606ab6c", size = 113592, upload-time = "2026-01-06T11:45:19.497Z" },
+]
+
+[[package]]
+name = "async-timeout"
+version = "5.0.1"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/a5/ae/136395dfbfe00dfc94da3f3e136d0b13f394cba8f4841120e34226265780/async_timeout-5.0.1.tar.gz", hash = "sha256:d9321a7a3d5a6a5e187e824d2fa0793ce379a202935782d555d6e9d2735677d3", size = 9274, upload-time = "2024-11-06T16:41:39.6Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/fe/ba/e2081de779ca30d473f21f5b30e0e737c438205440784c7dfc81efc2b029/async_timeout-5.0.1-py3-none-any.whl", hash = "sha256:39e3809566ff85354557ec2398b55e096c8364bacac9405a7a1fa429e77fe76c", size = 6233, upload-time = "2024-11-06T16:41:37.9Z" },
+]
+
+[[package]]
+name = "attrs"
+version = "25.4.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/6b/5c/685e6633917e101e5dcb62b9dd76946cbb57c26e133bae9e0cd36033c0a9/attrs-25.4.0.tar.gz", hash = "sha256:16d5969b87f0859ef33a48b35d55ac1be6e42ae49d5e853b597db70c35c57e11", size = 934251, upload-time = "2025-10-06T13:54:44.725Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/3a/2a/7cc015f5b9f5db42b7d48157e23356022889fc354a2813c15934b7cb5c0e/attrs-25.4.0-py3-none-any.whl", hash = "sha256:adcf7e2a1fb3b36ac48d97835bb6d8ade15b8dcce26aba8bf1d14847b57a3373", size = 67615, upload-time = "2025-10-06T13:54:43.17Z" },
+]
+
+[[package]]
+name = "bce-python-sdk"
+version = "0.9.60"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "future" },
+    { name = "pycryptodome" },
+    { name = "six" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/dc/00/7b84673285ede23fd3ca8d33a90a6963cd7f16755f4e8228025710acb078/bce_python_sdk-0.9.60.tar.gz", hash = "sha256:e0d04b8377cdfa264b1c217db3208dcb8ba58d02c9bad052dc3cbecf61c9eb0d", size = 279370, upload-time = "2026-01-27T03:05:29.502Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/39/45/1ef7b8db8716bf072e13e3857c2aa5f62e36b904cf88ceb796adbe7957e7/bce_python_sdk-0.9.60-py3-none-any.whl", hash = "sha256:50f13df97e79ff8e8b5ab22fbf38a78ff711e878b5976b8950e1b318d3d6df61", size = 395377, upload-time = "2026-01-27T03:05:26.404Z" },
+]
+
+[[package]]
+name = "certifi"
+version = "2026.1.4"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/e0/2d/a891ca51311197f6ad14a7ef42e2399f36cf2f9bd44752b3dc4eab60fdc5/certifi-2026.1.4.tar.gz", hash = "sha256:ac726dd470482006e014ad384921ed6438c457018f4b3d204aea4281258b2120", size = 154268, upload-time = "2026-01-04T02:42:41.825Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/e6/ad/3cc14f097111b4de0040c83a525973216457bbeeb63739ef1ed275c1c021/certifi-2026.1.4-py3-none-any.whl", hash = "sha256:9943707519e4add1115f44c2bc244f782c0249876bf51b6599fee1ffbedd685c", size = 152900, upload-time = "2026-01-04T02:42:40.15Z" },
+]
+
+[[package]]
+name = "cffi"
+version = "2.0.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "pycparser", marker = "implementation_name != 'PyPy'" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/eb/56/b1ba7935a17738ae8453301356628e8147c79dbb825bcbc73dc7401f9846/cffi-2.0.0.tar.gz", hash = "sha256:44d1b5909021139fe36001ae048dbdde8214afa20200eda0f64c068cac5d5529", size = 523588, upload-time = "2025-09-08T23:24:04.541Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/93/d7/516d984057745a6cd96575eea814fe1edd6646ee6efd552fb7b0921dec83/cffi-2.0.0-cp310-cp310-macosx_10_13_x86_64.whl", hash = "sha256:0cf2d91ecc3fcc0625c2c530fe004f82c110405f101548512cce44322fa8ac44", size = 184283, upload-time = "2025-09-08T23:22:08.01Z" },
+    { url = "https://files.pythonhosted.org/packages/9e/84/ad6a0b408daa859246f57c03efd28e5dd1b33c21737c2db84cae8c237aa5/cffi-2.0.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:f73b96c41e3b2adedc34a7356e64c8eb96e03a3782b535e043a986276ce12a49", size = 180504, upload-time = "2025-09-08T23:22:10.637Z" },
+    { url = "https://files.pythonhosted.org/packages/50/bd/b1a6362b80628111e6653c961f987faa55262b4002fcec42308cad1db680/cffi-2.0.0-cp310-cp310-manylinux1_i686.manylinux2014_i686.manylinux_2_17_i686.manylinux_2_5_i686.whl", hash = "sha256:53f77cbe57044e88bbd5ed26ac1d0514d2acf0591dd6bb02a3ae37f76811b80c", size = 208811, upload-time = "2025-09-08T23:22:12.267Z" },
+    { url = "https://files.pythonhosted.org/packages/4f/27/6933a8b2562d7bd1fb595074cf99cc81fc3789f6a6c05cdabb46284a3188/cffi-2.0.0-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:3e837e369566884707ddaf85fc1744b47575005c0a229de3327f8f9a20f4efeb", size = 216402, upload-time = "2025-09-08T23:22:13.455Z" },
+    { url = "https://files.pythonhosted.org/packages/05/eb/b86f2a2645b62adcfff53b0dd97e8dfafb5c8aa864bd0d9a2c2049a0d551/cffi-2.0.0-cp310-cp310-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:5eda85d6d1879e692d546a078b44251cdd08dd1cfb98dfb77b670c97cee49ea0", size = 203217, upload-time = "2025-09-08T23:22:14.596Z" },
+    { url = "https://files.pythonhosted.org/packages/9f/e0/6cbe77a53acf5acc7c08cc186c9928864bd7c005f9efd0d126884858a5fe/cffi-2.0.0-cp310-cp310-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:9332088d75dc3241c702d852d4671613136d90fa6881da7d770a483fd05248b4", size = 203079, upload-time = "2025-09-08T23:22:15.769Z" },
+    { url = "https://files.pythonhosted.org/packages/98/29/9b366e70e243eb3d14a5cb488dfd3a0b6b2f1fb001a203f653b93ccfac88/cffi-2.0.0-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:fc7de24befaeae77ba923797c7c87834c73648a05a4bde34b3b7e5588973a453", size = 216475, upload-time = "2025-09-08T23:22:17.427Z" },
+    { url = "https://files.pythonhosted.org/packages/21/7a/13b24e70d2f90a322f2900c5d8e1f14fa7e2a6b3332b7309ba7b2ba51a5a/cffi-2.0.0-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:cf364028c016c03078a23b503f02058f1814320a56ad535686f90565636a9495", size = 218829, upload-time = "2025-09-08T23:22:19.069Z" },
+    { url = "https://files.pythonhosted.org/packages/60/99/c9dc110974c59cc981b1f5b66e1d8af8af764e00f0293266824d9c4254bc/cffi-2.0.0-cp310-cp310-musllinux_1_2_i686.whl", hash = "sha256:e11e82b744887154b182fd3e7e8512418446501191994dbf9c9fc1f32cc8efd5", size = 211211, upload-time = "2025-09-08T23:22:20.588Z" },
+    { url = "https://files.pythonhosted.org/packages/49/72/ff2d12dbf21aca1b32a40ed792ee6b40f6dc3a9cf1644bd7ef6e95e0ac5e/cffi-2.0.0-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:8ea985900c5c95ce9db1745f7933eeef5d314f0565b27625d9a10ec9881e1bfb", size = 218036, upload-time = "2025-09-08T23:22:22.143Z" },
+    { url = "https://files.pythonhosted.org/packages/e2/cc/027d7fb82e58c48ea717149b03bcadcbdc293553edb283af792bd4bcbb3f/cffi-2.0.0-cp310-cp310-win32.whl", hash = "sha256:1f72fb8906754ac8a2cc3f9f5aaa298070652a0ffae577e0ea9bd480dc3c931a", size = 172184, upload-time = "2025-09-08T23:22:23.328Z" },
+    { url = "https://files.pythonhosted.org/packages/33/fa/072dd15ae27fbb4e06b437eb6e944e75b068deb09e2a2826039e49ee2045/cffi-2.0.0-cp310-cp310-win_amd64.whl", hash = "sha256:b18a3ed7d5b3bd8d9ef7a8cb226502c6bf8308df1525e1cc676c3680e7176739", size = 182790, upload-time = "2025-09-08T23:22:24.752Z" },
+    { url = "https://files.pythonhosted.org/packages/12/4a/3dfd5f7850cbf0d06dc84ba9aa00db766b52ca38d8b86e3a38314d52498c/cffi-2.0.0-cp311-cp311-macosx_10_13_x86_64.whl", hash = "sha256:b4c854ef3adc177950a8dfc81a86f5115d2abd545751a304c5bcf2c2c7283cfe", size = 184344, upload-time = "2025-09-08T23:22:26.456Z" },
+    { url = "https://files.pythonhosted.org/packages/4f/8b/f0e4c441227ba756aafbe78f117485b25bb26b1c059d01f137fa6d14896b/cffi-2.0.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:2de9a304e27f7596cd03d16f1b7c72219bd944e99cc52b84d0145aefb07cbd3c", size = 180560, upload-time = "2025-09-08T23:22:28.197Z" },
+    { url = "https://files.pythonhosted.org/packages/b1/b7/1200d354378ef52ec227395d95c2576330fd22a869f7a70e88e1447eb234/cffi-2.0.0-cp311-cp311-manylinux1_i686.manylinux2014_i686.manylinux_2_17_i686.manylinux_2_5_i686.whl", hash = "sha256:baf5215e0ab74c16e2dd324e8ec067ef59e41125d3eade2b863d294fd5035c92", size = 209613, upload-time = "2025-09-08T23:22:29.475Z" },
+    { url = "https://files.pythonhosted.org/packages/b8/56/6033f5e86e8cc9bb629f0077ba71679508bdf54a9a5e112a3c0b91870332/cffi-2.0.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:730cacb21e1bdff3ce90babf007d0a0917cc3e6492f336c2f0134101e0944f93", size = 216476, upload-time = "2025-09-08T23:22:31.063Z" },
+    { url = "https://files.pythonhosted.org/packages/dc/7f/55fecd70f7ece178db2f26128ec41430d8720f2d12ca97bf8f0a628207d5/cffi-2.0.0-cp311-cp311-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:6824f87845e3396029f3820c206e459ccc91760e8fa24422f8b0c3d1731cbec5", size = 203374, upload-time = "2025-09-08T23:22:32.507Z" },
+    { url = "https://files.pythonhosted.org/packages/84/ef/a7b77c8bdc0f77adc3b46888f1ad54be8f3b7821697a7b89126e829e676a/cffi-2.0.0-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:9de40a7b0323d889cf8d23d1ef214f565ab154443c42737dfe52ff82cf857664", size = 202597, upload-time = "2025-09-08T23:22:34.132Z" },
+    { url = "https://files.pythonhosted.org/packages/d7/91/500d892b2bf36529a75b77958edfcd5ad8e2ce4064ce2ecfeab2125d72d1/cffi-2.0.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:8941aaadaf67246224cee8c3803777eed332a19d909b47e29c9842ef1e79ac26", size = 215574, upload-time = "2025-09-08T23:22:35.443Z" },
+    { url = "https://files.pythonhosted.org/packages/44/64/58f6255b62b101093d5df22dcb752596066c7e89dd725e0afaed242a61be/cffi-2.0.0-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:a05d0c237b3349096d3981b727493e22147f934b20f6f125a3eba8f994bec4a9", size = 218971, upload-time = "2025-09-08T23:22:36.805Z" },
+    { url = "https://files.pythonhosted.org/packages/ab/49/fa72cebe2fd8a55fbe14956f9970fe8eb1ac59e5df042f603ef7c8ba0adc/cffi-2.0.0-cp311-cp311-musllinux_1_2_i686.whl", hash = "sha256:94698a9c5f91f9d138526b48fe26a199609544591f859c870d477351dc7b2414", size = 211972, upload-time = "2025-09-08T23:22:38.436Z" },
+    { url = "https://files.pythonhosted.org/packages/0b/28/dd0967a76aab36731b6ebfe64dec4e981aff7e0608f60c2d46b46982607d/cffi-2.0.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:5fed36fccc0612a53f1d4d9a816b50a36702c28a2aa880cb8a122b3466638743", size = 217078, upload-time = "2025-09-08T23:22:39.776Z" },
+    { url = "https://files.pythonhosted.org/packages/2b/c0/015b25184413d7ab0a410775fdb4a50fca20f5589b5dab1dbbfa3baad8ce/cffi-2.0.0-cp311-cp311-win32.whl", hash = "sha256:c649e3a33450ec82378822b3dad03cc228b8f5963c0c12fc3b1e0ab940f768a5", size = 172076, upload-time = "2025-09-08T23:22:40.95Z" },
+    { url = "https://files.pythonhosted.org/packages/ae/8f/dc5531155e7070361eb1b7e4c1a9d896d0cb21c49f807a6c03fd63fc877e/cffi-2.0.0-cp311-cp311-win_amd64.whl", hash = "sha256:66f011380d0e49ed280c789fbd08ff0d40968ee7b665575489afa95c98196ab5", size = 182820, upload-time = "2025-09-08T23:22:42.463Z" },
+    { url = "https://files.pythonhosted.org/packages/95/5c/1b493356429f9aecfd56bc171285a4c4ac8697f76e9bbbbb105e537853a1/cffi-2.0.0-cp311-cp311-win_arm64.whl", hash = "sha256:c6638687455baf640e37344fe26d37c404db8b80d037c3d29f58fe8d1c3b194d", size = 177635, upload-time = "2025-09-08T23:22:43.623Z" },
+    { url = "https://files.pythonhosted.org/packages/ea/47/4f61023ea636104d4f16ab488e268b93008c3d0bb76893b1b31db1f96802/cffi-2.0.0-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:6d02d6655b0e54f54c4ef0b94eb6be0607b70853c45ce98bd278dc7de718be5d", size = 185271, upload-time = "2025-09-08T23:22:44.795Z" },
+    { url = "https://files.pythonhosted.org/packages/df/a2/781b623f57358e360d62cdd7a8c681f074a71d445418a776eef0aadb4ab4/cffi-2.0.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:8eca2a813c1cb7ad4fb74d368c2ffbbb4789d377ee5bb8df98373c2cc0dee76c", size = 181048, upload-time = "2025-09-08T23:22:45.938Z" },
+    { url = "https://files.pythonhosted.org/packages/ff/df/a4f0fbd47331ceeba3d37c2e51e9dfc9722498becbeec2bd8bc856c9538a/cffi-2.0.0-cp312-cp312-manylinux1_i686.manylinux2014_i686.manylinux_2_17_i686.manylinux_2_5_i686.whl", hash = "sha256:21d1152871b019407d8ac3985f6775c079416c282e431a4da6afe7aefd2bccbe", size = 212529, upload-time = "2025-09-08T23:22:47.349Z" },
+    { url = "https://files.pythonhosted.org/packages/d5/72/12b5f8d3865bf0f87cf1404d8c374e7487dcf097a1c91c436e72e6badd83/cffi-2.0.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:b21e08af67b8a103c71a250401c78d5e0893beff75e28c53c98f4de42f774062", size = 220097, upload-time = "2025-09-08T23:22:48.677Z" },
+    { url = "https://files.pythonhosted.org/packages/c2/95/7a135d52a50dfa7c882ab0ac17e8dc11cec9d55d2c18dda414c051c5e69e/cffi-2.0.0-cp312-cp312-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:1e3a615586f05fc4065a8b22b8152f0c1b00cdbc60596d187c2a74f9e3036e4e", size = 207983, upload-time = "2025-09-08T23:22:50.06Z" },
+    { url = "https://files.pythonhosted.org/packages/3a/c8/15cb9ada8895957ea171c62dc78ff3e99159ee7adb13c0123c001a2546c1/cffi-2.0.0-cp312-cp312-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:81afed14892743bbe14dacb9e36d9e0e504cd204e0b165062c488942b9718037", size = 206519, upload-time = "2025-09-08T23:22:51.364Z" },
+    { url = "https://files.pythonhosted.org/packages/78/2d/7fa73dfa841b5ac06c7b8855cfc18622132e365f5b81d02230333ff26e9e/cffi-2.0.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:3e17ed538242334bf70832644a32a7aae3d83b57567f9fd60a26257e992b79ba", size = 219572, upload-time = "2025-09-08T23:22:52.902Z" },
+    { url = "https://files.pythonhosted.org/packages/07/e0/267e57e387b4ca276b90f0434ff88b2c2241ad72b16d31836adddfd6031b/cffi-2.0.0-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:3925dd22fa2b7699ed2617149842d2e6adde22b262fcbfada50e3d195e4b3a94", size = 222963, upload-time = "2025-09-08T23:22:54.518Z" },
+    { url = "https://files.pythonhosted.org/packages/b6/75/1f2747525e06f53efbd878f4d03bac5b859cbc11c633d0fb81432d98a795/cffi-2.0.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:2c8f814d84194c9ea681642fd164267891702542f028a15fc97d4674b6206187", size = 221361, upload-time = "2025-09-08T23:22:55.867Z" },
+    { url = "https://files.pythonhosted.org/packages/7b/2b/2b6435f76bfeb6bbf055596976da087377ede68df465419d192acf00c437/cffi-2.0.0-cp312-cp312-win32.whl", hash = "sha256:da902562c3e9c550df360bfa53c035b2f241fed6d9aef119048073680ace4a18", size = 172932, upload-time = "2025-09-08T23:22:57.188Z" },
+    { url = "https://files.pythonhosted.org/packages/f8/ed/13bd4418627013bec4ed6e54283b1959cf6db888048c7cf4b4c3b5b36002/cffi-2.0.0-cp312-cp312-win_amd64.whl", hash = "sha256:da68248800ad6320861f129cd9c1bf96ca849a2771a59e0344e88681905916f5", size = 183557, upload-time = "2025-09-08T23:22:58.351Z" },
+    { url = "https://files.pythonhosted.org/packages/95/31/9f7f93ad2f8eff1dbc1c3656d7ca5bfd8fb52c9d786b4dcf19b2d02217fa/cffi-2.0.0-cp312-cp312-win_arm64.whl", hash = "sha256:4671d9dd5ec934cb9a73e7ee9676f9362aba54f7f34910956b84d727b0d73fb6", size = 177762, upload-time = "2025-09-08T23:22:59.668Z" },
+    { url = "https://files.pythonhosted.org/packages/4b/8d/a0a47a0c9e413a658623d014e91e74a50cdd2c423f7ccfd44086ef767f90/cffi-2.0.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:00bdf7acc5f795150faa6957054fbbca2439db2f775ce831222b66f192f03beb", size = 185230, upload-time = "2025-09-08T23:23:00.879Z" },
+    { url = "https://files.pythonhosted.org/packages/4a/d2/a6c0296814556c68ee32009d9c2ad4f85f2707cdecfd7727951ec228005d/cffi-2.0.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:45d5e886156860dc35862657e1494b9bae8dfa63bf56796f2fb56e1679fc0bca", size = 181043, upload-time = "2025-09-08T23:23:02.231Z" },
+    { url = "https://files.pythonhosted.org/packages/b0/1e/d22cc63332bd59b06481ceaac49d6c507598642e2230f201649058a7e704/cffi-2.0.0-cp313-cp313-manylinux1_i686.manylinux2014_i686.manylinux_2_17_i686.manylinux_2_5_i686.whl", hash = "sha256:07b271772c100085dd28b74fa0cd81c8fb1a3ba18b21e03d7c27f3436a10606b", size = 212446, upload-time = "2025-09-08T23:23:03.472Z" },
+    { url = "https://files.pythonhosted.org/packages/a9/f5/a2c23eb03b61a0b8747f211eb716446c826ad66818ddc7810cc2cc19b3f2/cffi-2.0.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:d48a880098c96020b02d5a1f7d9251308510ce8858940e6fa99ece33f610838b", size = 220101, upload-time = "2025-09-08T23:23:04.792Z" },
+    { url = "https://files.pythonhosted.org/packages/f2/7f/e6647792fc5850d634695bc0e6ab4111ae88e89981d35ac269956605feba/cffi-2.0.0-cp313-cp313-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:f93fd8e5c8c0a4aa1f424d6173f14a892044054871c771f8566e4008eaa359d2", size = 207948, upload-time = "2025-09-08T23:23:06.127Z" },
+    { url = "https://files.pythonhosted.org/packages/cb/1e/a5a1bd6f1fb30f22573f76533de12a00bf274abcdc55c8edab639078abb6/cffi-2.0.0-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:dd4f05f54a52fb558f1ba9f528228066954fee3ebe629fc1660d874d040ae5a3", size = 206422, upload-time = "2025-09-08T23:23:07.753Z" },
+    { url = "https://files.pythonhosted.org/packages/98/df/0a1755e750013a2081e863e7cd37e0cdd02664372c754e5560099eb7aa44/cffi-2.0.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:c8d3b5532fc71b7a77c09192b4a5a200ea992702734a2e9279a37f2478236f26", size = 219499, upload-time = "2025-09-08T23:23:09.648Z" },
+    { url = "https://files.pythonhosted.org/packages/50/e1/a969e687fcf9ea58e6e2a928ad5e2dd88cc12f6f0ab477e9971f2309b57c/cffi-2.0.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:d9b29c1f0ae438d5ee9acb31cadee00a58c46cc9c0b2f9038c6b0b3470877a8c", size = 222928, upload-time = "2025-09-08T23:23:10.928Z" },
+    { url = "https://files.pythonhosted.org/packages/36/54/0362578dd2c9e557a28ac77698ed67323ed5b9775ca9d3fe73fe191bb5d8/cffi-2.0.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:6d50360be4546678fc1b79ffe7a66265e28667840010348dd69a314145807a1b", size = 221302, upload-time = "2025-09-08T23:23:12.42Z" },
+    { url = "https://files.pythonhosted.org/packages/eb/6d/bf9bda840d5f1dfdbf0feca87fbdb64a918a69bca42cfa0ba7b137c48cb8/cffi-2.0.0-cp313-cp313-win32.whl", hash = "sha256:74a03b9698e198d47562765773b4a8309919089150a0bb17d829ad7b44b60d27", size = 172909, upload-time = "2025-09-08T23:23:14.32Z" },
+    { url = "https://files.pythonhosted.org/packages/37/18/6519e1ee6f5a1e579e04b9ddb6f1676c17368a7aba48299c3759bbc3c8b3/cffi-2.0.0-cp313-cp313-win_amd64.whl", hash = "sha256:19f705ada2530c1167abacb171925dd886168931e0a7b78f5bffcae5c6b5be75", size = 183402, upload-time = "2025-09-08T23:23:15.535Z" },
+    { url = "https://files.pythonhosted.org/packages/cb/0e/02ceeec9a7d6ee63bb596121c2c8e9b3a9e150936f4fbef6ca1943e6137c/cffi-2.0.0-cp313-cp313-win_arm64.whl", hash = "sha256:256f80b80ca3853f90c21b23ee78cd008713787b1b1e93eae9f3d6a7134abd91", size = 177780, upload-time = "2025-09-08T23:23:16.761Z" },
+    { url = "https://files.pythonhosted.org/packages/92/c4/3ce07396253a83250ee98564f8d7e9789fab8e58858f35d07a9a2c78de9f/cffi-2.0.0-cp314-cp314-macosx_10_13_x86_64.whl", hash = "sha256:fc33c5141b55ed366cfaad382df24fe7dcbc686de5be719b207bb248e3053dc5", size = 185320, upload-time = "2025-09-08T23:23:18.087Z" },
+    { url = "https://files.pythonhosted.org/packages/59/dd/27e9fa567a23931c838c6b02d0764611c62290062a6d4e8ff7863daf9730/cffi-2.0.0-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:c654de545946e0db659b3400168c9ad31b5d29593291482c43e3564effbcee13", size = 181487, upload-time = "2025-09-08T23:23:19.622Z" },
+    { url = "https://files.pythonhosted.org/packages/d6/43/0e822876f87ea8a4ef95442c3d766a06a51fc5298823f884ef87aaad168c/cffi-2.0.0-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:24b6f81f1983e6df8db3adc38562c83f7d4a0c36162885ec7f7b77c7dcbec97b", size = 220049, upload-time = "2025-09-08T23:23:20.853Z" },
+    { url = "https://files.pythonhosted.org/packages/b4/89/76799151d9c2d2d1ead63c2429da9ea9d7aac304603de0c6e8764e6e8e70/cffi-2.0.0-cp314-cp314-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:12873ca6cb9b0f0d3a0da705d6086fe911591737a59f28b7936bdfed27c0d47c", size = 207793, upload-time = "2025-09-08T23:23:22.08Z" },
+    { url = "https://files.pythonhosted.org/packages/bb/dd/3465b14bb9e24ee24cb88c9e3730f6de63111fffe513492bf8c808a3547e/cffi-2.0.0-cp314-cp314-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:d9b97165e8aed9272a6bb17c01e3cc5871a594a446ebedc996e2397a1c1ea8ef", size = 206300, upload-time = "2025-09-08T23:23:23.314Z" },
+    { url = "https://files.pythonhosted.org/packages/47/d9/d83e293854571c877a92da46fdec39158f8d7e68da75bf73581225d28e90/cffi-2.0.0-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:afb8db5439b81cf9c9d0c80404b60c3cc9c3add93e114dcae767f1477cb53775", size = 219244, upload-time = "2025-09-08T23:23:24.541Z" },
+    { url = "https://files.pythonhosted.org/packages/2b/0f/1f177e3683aead2bb00f7679a16451d302c436b5cbf2505f0ea8146ef59e/cffi-2.0.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:737fe7d37e1a1bffe70bd5754ea763a62a066dc5913ca57e957824b72a85e205", size = 222828, upload-time = "2025-09-08T23:23:26.143Z" },
+    { url = "https://files.pythonhosted.org/packages/c6/0f/cafacebd4b040e3119dcb32fed8bdef8dfe94da653155f9d0b9dc660166e/cffi-2.0.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:38100abb9d1b1435bc4cc340bb4489635dc2f0da7456590877030c9b3d40b0c1", size = 220926, upload-time = "2025-09-08T23:23:27.873Z" },
+    { url = "https://files.pythonhosted.org/packages/3e/aa/df335faa45b395396fcbc03de2dfcab242cd61a9900e914fe682a59170b1/cffi-2.0.0-cp314-cp314-win32.whl", hash = "sha256:087067fa8953339c723661eda6b54bc98c5625757ea62e95eb4898ad5e776e9f", size = 175328, upload-time = "2025-09-08T23:23:44.61Z" },
+    { url = "https://files.pythonhosted.org/packages/bb/92/882c2d30831744296ce713f0feb4c1cd30f346ef747b530b5318715cc367/cffi-2.0.0-cp314-cp314-win_amd64.whl", hash = "sha256:203a48d1fb583fc7d78a4c6655692963b860a417c0528492a6bc21f1aaefab25", size = 185650, upload-time = "2025-09-08T23:23:45.848Z" },
+    { url = "https://files.pythonhosted.org/packages/9f/2c/98ece204b9d35a7366b5b2c6539c350313ca13932143e79dc133ba757104/cffi-2.0.0-cp314-cp314-win_arm64.whl", hash = "sha256:dbd5c7a25a7cb98f5ca55d258b103a2054f859a46ae11aaf23134f9cc0d356ad", size = 180687, upload-time = "2025-09-08T23:23:47.105Z" },
+    { url = "https://files.pythonhosted.org/packages/3e/61/c768e4d548bfa607abcda77423448df8c471f25dbe64fb2ef6d555eae006/cffi-2.0.0-cp314-cp314t-macosx_10_13_x86_64.whl", hash = "sha256:9a67fc9e8eb39039280526379fb3a70023d77caec1852002b4da7e8b270c4dd9", size = 188773, upload-time = "2025-09-08T23:23:29.347Z" },
+    { url = "https://files.pythonhosted.org/packages/2c/ea/5f76bce7cf6fcd0ab1a1058b5af899bfbef198bea4d5686da88471ea0336/cffi-2.0.0-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:7a66c7204d8869299919db4d5069a82f1561581af12b11b3c9f48c584eb8743d", size = 185013, upload-time = "2025-09-08T23:23:30.63Z" },
+    { url = "https://files.pythonhosted.org/packages/be/b4/c56878d0d1755cf9caa54ba71e5d049479c52f9e4afc230f06822162ab2f/cffi-2.0.0-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:7cc09976e8b56f8cebd752f7113ad07752461f48a58cbba644139015ac24954c", size = 221593, upload-time = "2025-09-08T23:23:31.91Z" },
+    { url = "https://files.pythonhosted.org/packages/e0/0d/eb704606dfe8033e7128df5e90fee946bbcb64a04fcdaa97321309004000/cffi-2.0.0-cp314-cp314t-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:92b68146a71df78564e4ef48af17551a5ddd142e5190cdf2c5624d0c3ff5b2e8", size = 209354, upload-time = "2025-09-08T23:23:33.214Z" },
+    { url = "https://files.pythonhosted.org/packages/d8/19/3c435d727b368ca475fb8742ab97c9cb13a0de600ce86f62eab7fa3eea60/cffi-2.0.0-cp314-cp314t-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:b1e74d11748e7e98e2f426ab176d4ed720a64412b6a15054378afdb71e0f37dc", size = 208480, upload-time = "2025-09-08T23:23:34.495Z" },
+    { url = "https://files.pythonhosted.org/packages/d0/44/681604464ed9541673e486521497406fadcc15b5217c3e326b061696899a/cffi-2.0.0-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:28a3a209b96630bca57cce802da70c266eb08c6e97e5afd61a75611ee6c64592", size = 221584, upload-time = "2025-09-08T23:23:36.096Z" },
+    { url = "https://files.pythonhosted.org/packages/25/8e/342a504ff018a2825d395d44d63a767dd8ebc927ebda557fecdaca3ac33a/cffi-2.0.0-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:7553fb2090d71822f02c629afe6042c299edf91ba1bf94951165613553984512", size = 224443, upload-time = "2025-09-08T23:23:37.328Z" },
+    { url = "https://files.pythonhosted.org/packages/e1/5e/b666bacbbc60fbf415ba9988324a132c9a7a0448a9a8f125074671c0f2c3/cffi-2.0.0-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:6c6c373cfc5c83a975506110d17457138c8c63016b563cc9ed6e056a82f13ce4", size = 223437, upload-time = "2025-09-08T23:23:38.945Z" },
+    { url = "https://files.pythonhosted.org/packages/a0/1d/ec1a60bd1a10daa292d3cd6bb0b359a81607154fb8165f3ec95fe003b85c/cffi-2.0.0-cp314-cp314t-win32.whl", hash = "sha256:1fc9ea04857caf665289b7a75923f2c6ed559b8298a1b8c49e59f7dd95c8481e", size = 180487, upload-time = "2025-09-08T23:23:40.423Z" },
+    { url = "https://files.pythonhosted.org/packages/bf/41/4c1168c74fac325c0c8156f04b6749c8b6a8f405bbf91413ba088359f60d/cffi-2.0.0-cp314-cp314t-win_amd64.whl", hash = "sha256:d68b6cef7827e8641e8ef16f4494edda8b36104d79773a334beaa1e3521430f6", size = 191726, upload-time = "2025-09-08T23:23:41.742Z" },
+    { url = "https://files.pythonhosted.org/packages/ae/3a/dbeec9d1ee0844c679f6bb5d6ad4e9f198b1224f4e7a32825f47f6192b0c/cffi-2.0.0-cp314-cp314t-win_arm64.whl", hash = "sha256:0a1527a803f0a659de1af2e1fd700213caba79377e27e4693648c2923da066f9", size = 184195, upload-time = "2025-09-08T23:23:43.004Z" },
+]
+
+[[package]]
+name = "chardet"
+version = "5.2.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/f3/0d/f7b6ab21ec75897ed80c17d79b15951a719226b9fababf1e40ea74d69079/chardet-5.2.0.tar.gz", hash = "sha256:1b3b6ff479a8c414bc3fa2c0852995695c4a026dcd6d0633b2dd092ca39c1cf7", size = 2069618, upload-time = "2023-08-01T19:23:02.662Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/38/6f/f5fbc992a329ee4e0f288c1fe0e2ad9485ed064cac731ed2fe47dcc38cbf/chardet-5.2.0-py3-none-any.whl", hash = "sha256:e1cf59446890a00105fe7b7912492ea04b6e6f06d4b742b2c788469e34c82970", size = 199385, upload-time = "2023-08-01T19:23:00.661Z" },
+]
+
+[[package]]
+name = "charset-normalizer"
+version = "3.4.4"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/13/69/33ddede1939fdd074bce5434295f38fae7136463422fe4fd3e0e89b98062/charset_normalizer-3.4.4.tar.gz", hash = "sha256:94537985111c35f28720e43603b8e7b43a6ecfb2ce1d3058bbe955b73404e21a", size = 129418, upload-time = "2025-10-14T04:42:32.879Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/1f/b8/6d51fc1d52cbd52cd4ccedd5b5b2f0f6a11bbf6765c782298b0f3e808541/charset_normalizer-3.4.4-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:e824f1492727fa856dd6eda4f7cee25f8518a12f3c4a56a74e8095695089cf6d", size = 209709, upload-time = "2025-10-14T04:40:11.385Z" },
+    { url = "https://files.pythonhosted.org/packages/5c/af/1f9d7f7faafe2ddfb6f72a2e07a548a629c61ad510fe60f9630309908fef/charset_normalizer-3.4.4-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:4bd5d4137d500351a30687c2d3971758aac9a19208fc110ccb9d7188fbe709e8", size = 148814, upload-time = "2025-10-14T04:40:13.135Z" },
+    { url = "https://files.pythonhosted.org/packages/79/3d/f2e3ac2bbc056ca0c204298ea4e3d9db9b4afe437812638759db2c976b5f/charset_normalizer-3.4.4-cp310-cp310-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:027f6de494925c0ab2a55eab46ae5129951638a49a34d87f4c3eda90f696b4ad", size = 144467, upload-time = "2025-10-14T04:40:14.728Z" },
+    { url = "https://files.pythonhosted.org/packages/ec/85/1bf997003815e60d57de7bd972c57dc6950446a3e4ccac43bc3070721856/charset_normalizer-3.4.4-cp310-cp310-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:f820802628d2694cb7e56db99213f930856014862f3fd943d290ea8438d07ca8", size = 162280, upload-time = "2025-10-14T04:40:16.14Z" },
+    { url = "https://files.pythonhosted.org/packages/3e/8e/6aa1952f56b192f54921c436b87f2aaf7c7a7c3d0d1a765547d64fd83c13/charset_normalizer-3.4.4-cp310-cp310-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:798d75d81754988d2565bff1b97ba5a44411867c0cf32b77a7e8f8d84796b10d", size = 159454, upload-time = "2025-10-14T04:40:17.567Z" },
+    { url = "https://files.pythonhosted.org/packages/36/3b/60cbd1f8e93aa25d1c669c649b7a655b0b5fb4c571858910ea9332678558/charset_normalizer-3.4.4-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:9d1bb833febdff5c8927f922386db610b49db6e0d4f4ee29601d71e7c2694313", size = 153609, upload-time = "2025-10-14T04:40:19.08Z" },
+    { url = "https://files.pythonhosted.org/packages/64/91/6a13396948b8fd3c4b4fd5bc74d045f5637d78c9675585e8e9fbe5636554/charset_normalizer-3.4.4-cp310-cp310-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:9cd98cdc06614a2f768d2b7286d66805f94c48cde050acdbbb7db2600ab3197e", size = 151849, upload-time = "2025-10-14T04:40:20.607Z" },
+    { url = "https://files.pythonhosted.org/packages/b7/7a/59482e28b9981d105691e968c544cc0df3b7d6133152fb3dcdc8f135da7a/charset_normalizer-3.4.4-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:077fbb858e903c73f6c9db43374fd213b0b6a778106bc7032446a8e8b5b38b93", size = 151586, upload-time = "2025-10-14T04:40:21.719Z" },
+    { url = "https://files.pythonhosted.org/packages/92/59/f64ef6a1c4bdd2baf892b04cd78792ed8684fbc48d4c2afe467d96b4df57/charset_normalizer-3.4.4-cp310-cp310-musllinux_1_2_armv7l.whl", hash = "sha256:244bfb999c71b35de57821b8ea746b24e863398194a4014e4c76adc2bbdfeff0", size = 145290, upload-time = "2025-10-14T04:40:23.069Z" },
+    { url = "https://files.pythonhosted.org/packages/6b/63/3bf9f279ddfa641ffa1962b0db6a57a9c294361cc2f5fcac997049a00e9c/charset_normalizer-3.4.4-cp310-cp310-musllinux_1_2_ppc64le.whl", hash = "sha256:64b55f9dce520635f018f907ff1b0df1fdc31f2795a922fb49dd14fbcdf48c84", size = 163663, upload-time = "2025-10-14T04:40:24.17Z" },
+    { url = "https://files.pythonhosted.org/packages/ed/09/c9e38fc8fa9e0849b172b581fd9803bdf6e694041127933934184e19f8c3/charset_normalizer-3.4.4-cp310-cp310-musllinux_1_2_riscv64.whl", hash = "sha256:faa3a41b2b66b6e50f84ae4a68c64fcd0c44355741c6374813a800cd6695db9e", size = 151964, upload-time = "2025-10-14T04:40:25.368Z" },
+    { url = "https://files.pythonhosted.org/packages/d2/d1/d28b747e512d0da79d8b6a1ac18b7ab2ecfd81b2944c4c710e166d8dd09c/charset_normalizer-3.4.4-cp310-cp310-musllinux_1_2_s390x.whl", hash = "sha256:6515f3182dbe4ea06ced2d9e8666d97b46ef4c75e326b79bb624110f122551db", size = 161064, upload-time = "2025-10-14T04:40:26.806Z" },
+    { url = "https://files.pythonhosted.org/packages/bb/9a/31d62b611d901c3b9e5500c36aab0ff5eb442043fb3a1c254200d3d397d9/charset_normalizer-3.4.4-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:cc00f04ed596e9dc0da42ed17ac5e596c6ccba999ba6bd92b0e0aef2f170f2d6", size = 155015, upload-time = "2025-10-14T04:40:28.284Z" },
+    { url = "https://files.pythonhosted.org/packages/1f/f3/107e008fa2bff0c8b9319584174418e5e5285fef32f79d8ee6a430d0039c/charset_normalizer-3.4.4-cp310-cp310-win32.whl", hash = "sha256:f34be2938726fc13801220747472850852fe6b1ea75869a048d6f896838c896f", size = 99792, upload-time = "2025-10-14T04:40:29.613Z" },
+    { url = "https://files.pythonhosted.org/packages/eb/66/e396e8a408843337d7315bab30dbf106c38966f1819f123257f5520f8a96/charset_normalizer-3.4.4-cp310-cp310-win_amd64.whl", hash = "sha256:a61900df84c667873b292c3de315a786dd8dac506704dea57bc957bd31e22c7d", size = 107198, upload-time = "2025-10-14T04:40:30.644Z" },
+    { url = "https://files.pythonhosted.org/packages/b5/58/01b4f815bf0312704c267f2ccb6e5d42bcc7752340cd487bc9f8c3710597/charset_normalizer-3.4.4-cp310-cp310-win_arm64.whl", hash = "sha256:cead0978fc57397645f12578bfd2d5ea9138ea0fac82b2f63f7f7c6877986a69", size = 100262, upload-time = "2025-10-14T04:40:32.108Z" },
+    { url = "https://files.pythonhosted.org/packages/ed/27/c6491ff4954e58a10f69ad90aca8a1b6fe9c5d3c6f380907af3c37435b59/charset_normalizer-3.4.4-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:6e1fcf0720908f200cd21aa4e6750a48ff6ce4afe7ff5a79a90d5ed8a08296f8", size = 206988, upload-time = "2025-10-14T04:40:33.79Z" },
+    { url = "https://files.pythonhosted.org/packages/94/59/2e87300fe67ab820b5428580a53cad894272dbb97f38a7a814a2a1ac1011/charset_normalizer-3.4.4-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:5f819d5fe9234f9f82d75bdfa9aef3a3d72c4d24a6e57aeaebba32a704553aa0", size = 147324, upload-time = "2025-10-14T04:40:34.961Z" },
+    { url = "https://files.pythonhosted.org/packages/07/fb/0cf61dc84b2b088391830f6274cb57c82e4da8bbc2efeac8c025edb88772/charset_normalizer-3.4.4-cp311-cp311-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:a59cb51917aa591b1c4e6a43c132f0cdc3c76dbad6155df4e28ee626cc77a0a3", size = 142742, upload-time = "2025-10-14T04:40:36.105Z" },
+    { url = "https://files.pythonhosted.org/packages/62/8b/171935adf2312cd745d290ed93cf16cf0dfe320863ab7cbeeae1dcd6535f/charset_normalizer-3.4.4-cp311-cp311-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:8ef3c867360f88ac904fd3f5e1f902f13307af9052646963ee08ff4f131adafc", size = 160863, upload-time = "2025-10-14T04:40:37.188Z" },
+    { url = "https://files.pythonhosted.org/packages/09/73/ad875b192bda14f2173bfc1bc9a55e009808484a4b256748d931b6948442/charset_normalizer-3.4.4-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:d9e45d7faa48ee908174d8fe84854479ef838fc6a705c9315372eacbc2f02897", size = 157837, upload-time = "2025-10-14T04:40:38.435Z" },
+    { url = "https://files.pythonhosted.org/packages/6d/fc/de9cce525b2c5b94b47c70a4b4fb19f871b24995c728e957ee68ab1671ea/charset_normalizer-3.4.4-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:840c25fb618a231545cbab0564a799f101b63b9901f2569faecd6b222ac72381", size = 151550, upload-time = "2025-10-14T04:40:40.053Z" },
+    { url = "https://files.pythonhosted.org/packages/55/c2/43edd615fdfba8c6f2dfbd459b25a6b3b551f24ea21981e23fb768503ce1/charset_normalizer-3.4.4-cp311-cp311-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:ca5862d5b3928c4940729dacc329aa9102900382fea192fc5e52eb69d6093815", size = 149162, upload-time = "2025-10-14T04:40:41.163Z" },
+    { url = "https://files.pythonhosted.org/packages/03/86/bde4ad8b4d0e9429a4e82c1e8f5c659993a9a863ad62c7df05cf7b678d75/charset_normalizer-3.4.4-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:d9c7f57c3d666a53421049053eaacdd14bbd0a528e2186fcb2e672effd053bb0", size = 150019, upload-time = "2025-10-14T04:40:42.276Z" },
+    { url = "https://files.pythonhosted.org/packages/1f/86/a151eb2af293a7e7bac3a739b81072585ce36ccfb4493039f49f1d3cae8c/charset_normalizer-3.4.4-cp311-cp311-musllinux_1_2_armv7l.whl", hash = "sha256:277e970e750505ed74c832b4bf75dac7476262ee2a013f5574dd49075879e161", size = 143310, upload-time = "2025-10-14T04:40:43.439Z" },
+    { url = "https://files.pythonhosted.org/packages/b5/fe/43dae6144a7e07b87478fdfc4dbe9efd5defb0e7ec29f5f58a55aeef7bf7/charset_normalizer-3.4.4-cp311-cp311-musllinux_1_2_ppc64le.whl", hash = "sha256:31fd66405eaf47bb62e8cd575dc621c56c668f27d46a61d975a249930dd5e2a4", size = 162022, upload-time = "2025-10-14T04:40:44.547Z" },
+    { url = "https://files.pythonhosted.org/packages/80/e6/7aab83774f5d2bca81f42ac58d04caf44f0cc2b65fc6db2b3b2e8a05f3b3/charset_normalizer-3.4.4-cp311-cp311-musllinux_1_2_riscv64.whl", hash = "sha256:0d3d8f15c07f86e9ff82319b3d9ef6f4bf907608f53fe9d92b28ea9ae3d1fd89", size = 149383, upload-time = "2025-10-14T04:40:46.018Z" },
+    { url = "https://files.pythonhosted.org/packages/4f/e8/b289173b4edae05c0dde07f69f8db476a0b511eac556dfe0d6bda3c43384/charset_normalizer-3.4.4-cp311-cp311-musllinux_1_2_s390x.whl", hash = "sha256:9f7fcd74d410a36883701fafa2482a6af2ff5ba96b9a620e9e0721e28ead5569", size = 159098, upload-time = "2025-10-14T04:40:47.081Z" },
+    { url = "https://files.pythonhosted.org/packages/d8/df/fe699727754cae3f8478493c7f45f777b17c3ef0600e28abfec8619eb49c/charset_normalizer-3.4.4-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:ebf3e58c7ec8a8bed6d66a75d7fb37b55e5015b03ceae72a8e7c74495551e224", size = 152991, upload-time = "2025-10-14T04:40:48.246Z" },
+    { url = "https://files.pythonhosted.org/packages/1a/86/584869fe4ddb6ffa3bd9f491b87a01568797fb9bd8933f557dba9771beaf/charset_normalizer-3.4.4-cp311-cp311-win32.whl", hash = "sha256:eecbc200c7fd5ddb9a7f16c7decb07b566c29fa2161a16cf67b8d068bd21690a", size = 99456, upload-time = "2025-10-14T04:40:49.376Z" },
+    { url = "https://files.pythonhosted.org/packages/65/f6/62fdd5feb60530f50f7e38b4f6a1d5203f4d16ff4f9f0952962c044e919a/charset_normalizer-3.4.4-cp311-cp311-win_amd64.whl", hash = "sha256:5ae497466c7901d54b639cf42d5b8c1b6a4fead55215500d2f486d34db48d016", size = 106978, upload-time = "2025-10-14T04:40:50.844Z" },
+    { url = "https://files.pythonhosted.org/packages/7a/9d/0710916e6c82948b3be62d9d398cb4fcf4e97b56d6a6aeccd66c4b2f2bd5/charset_normalizer-3.4.4-cp311-cp311-win_arm64.whl", hash = "sha256:65e2befcd84bc6f37095f5961e68a6f077bf44946771354a28ad434c2cce0ae1", size = 99969, upload-time = "2025-10-14T04:40:52.272Z" },
+    { url = "https://files.pythonhosted.org/packages/f3/85/1637cd4af66fa687396e757dec650f28025f2a2f5a5531a3208dc0ec43f2/charset_normalizer-3.4.4-cp312-cp312-macosx_10_13_universal2.whl", hash = "sha256:0a98e6759f854bd25a58a73fa88833fba3b7c491169f86ce1180c948ab3fd394", size = 208425, upload-time = "2025-10-14T04:40:53.353Z" },
+    { url = "https://files.pythonhosted.org/packages/9d/6a/04130023fef2a0d9c62d0bae2649b69f7b7d8d24ea5536feef50551029df/charset_normalizer-3.4.4-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:b5b290ccc2a263e8d185130284f8501e3e36c5e02750fc6b6bdeb2e9e96f1e25", size = 148162, upload-time = "2025-10-14T04:40:54.558Z" },
+    { url = "https://files.pythonhosted.org/packages/78/29/62328d79aa60da22c9e0b9a66539feae06ca0f5a4171ac4f7dc285b83688/charset_normalizer-3.4.4-cp312-cp312-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:74bb723680f9f7a6234dcf67aea57e708ec1fbdf5699fb91dfd6f511b0a320ef", size = 144558, upload-time = "2025-10-14T04:40:55.677Z" },
+    { url = "https://files.pythonhosted.org/packages/86/bb/b32194a4bf15b88403537c2e120b817c61cd4ecffa9b6876e941c3ee38fe/charset_normalizer-3.4.4-cp312-cp312-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:f1e34719c6ed0b92f418c7c780480b26b5d9c50349e9a9af7d76bf757530350d", size = 161497, upload-time = "2025-10-14T04:40:57.217Z" },
+    { url = "https://files.pythonhosted.org/packages/19/89/a54c82b253d5b9b111dc74aca196ba5ccfcca8242d0fb64146d4d3183ff1/charset_normalizer-3.4.4-cp312-cp312-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:2437418e20515acec67d86e12bf70056a33abdacb5cb1655042f6538d6b085a8", size = 159240, upload-time = "2025-10-14T04:40:58.358Z" },
+    { url = "https://files.pythonhosted.org/packages/c0/10/d20b513afe03acc89ec33948320a5544d31f21b05368436d580dec4e234d/charset_normalizer-3.4.4-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:11d694519d7f29d6cd09f6ac70028dba10f92f6cdd059096db198c283794ac86", size = 153471, upload-time = "2025-10-14T04:40:59.468Z" },
+    { url = "https://files.pythonhosted.org/packages/61/fa/fbf177b55bdd727010f9c0a3c49eefa1d10f960e5f09d1d887bf93c2e698/charset_normalizer-3.4.4-cp312-cp312-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:ac1c4a689edcc530fc9d9aa11f5774b9e2f33f9a0c6a57864e90908f5208d30a", size = 150864, upload-time = "2025-10-14T04:41:00.623Z" },
+    { url = "https://files.pythonhosted.org/packages/05/12/9fbc6a4d39c0198adeebbde20b619790e9236557ca59fc40e0e3cebe6f40/charset_normalizer-3.4.4-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:21d142cc6c0ec30d2efee5068ca36c128a30b0f2c53c1c07bd78cb6bc1d3be5f", size = 150647, upload-time = "2025-10-14T04:41:01.754Z" },
+    { url = "https://files.pythonhosted.org/packages/ad/1f/6a9a593d52e3e8c5d2b167daf8c6b968808efb57ef4c210acb907c365bc4/charset_normalizer-3.4.4-cp312-cp312-musllinux_1_2_armv7l.whl", hash = "sha256:5dbe56a36425d26d6cfb40ce79c314a2e4dd6211d51d6d2191c00bed34f354cc", size = 145110, upload-time = "2025-10-14T04:41:03.231Z" },
+    { url = "https://files.pythonhosted.org/packages/30/42/9a52c609e72471b0fc54386dc63c3781a387bb4fe61c20231a4ebcd58bdd/charset_normalizer-3.4.4-cp312-cp312-musllinux_1_2_ppc64le.whl", hash = "sha256:5bfbb1b9acf3334612667b61bd3002196fe2a1eb4dd74d247e0f2a4d50ec9bbf", size = 162839, upload-time = "2025-10-14T04:41:04.715Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/5b/c0682bbf9f11597073052628ddd38344a3d673fda35a36773f7d19344b23/charset_normalizer-3.4.4-cp312-cp312-musllinux_1_2_riscv64.whl", hash = "sha256:d055ec1e26e441f6187acf818b73564e6e6282709e9bcb5b63f5b23068356a15", size = 150667, upload-time = "2025-10-14T04:41:05.827Z" },
+    { url = "https://files.pythonhosted.org/packages/e4/24/a41afeab6f990cf2daf6cb8c67419b63b48cf518e4f56022230840c9bfb2/charset_normalizer-3.4.4-cp312-cp312-musllinux_1_2_s390x.whl", hash = "sha256:af2d8c67d8e573d6de5bc30cdb27e9b95e49115cd9baad5ddbd1a6207aaa82a9", size = 160535, upload-time = "2025-10-14T04:41:06.938Z" },
+    { url = "https://files.pythonhosted.org/packages/2a/e5/6a4ce77ed243c4a50a1fecca6aaaab419628c818a49434be428fe24c9957/charset_normalizer-3.4.4-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:780236ac706e66881f3b7f2f32dfe90507a09e67d1d454c762cf642e6e1586e0", size = 154816, upload-time = "2025-10-14T04:41:08.101Z" },
+    { url = "https://files.pythonhosted.org/packages/a8/ef/89297262b8092b312d29cdb2517cb1237e51db8ecef2e9af5edbe7b683b1/charset_normalizer-3.4.4-cp312-cp312-win32.whl", hash = "sha256:5833d2c39d8896e4e19b689ffc198f08ea58116bee26dea51e362ecc7cd3ed26", size = 99694, upload-time = "2025-10-14T04:41:09.23Z" },
+    { url = "https://files.pythonhosted.org/packages/3d/2d/1e5ed9dd3b3803994c155cd9aacb60c82c331bad84daf75bcb9c91b3295e/charset_normalizer-3.4.4-cp312-cp312-win_amd64.whl", hash = "sha256:a79cfe37875f822425b89a82333404539ae63dbdddf97f84dcbc3d339aae9525", size = 107131, upload-time = "2025-10-14T04:41:10.467Z" },
+    { url = "https://files.pythonhosted.org/packages/d0/d9/0ed4c7098a861482a7b6a95603edce4c0d9db2311af23da1fb2b75ec26fc/charset_normalizer-3.4.4-cp312-cp312-win_arm64.whl", hash = "sha256:376bec83a63b8021bb5c8ea75e21c4ccb86e7e45ca4eb81146091b56599b80c3", size = 100390, upload-time = "2025-10-14T04:41:11.915Z" },
+    { url = "https://files.pythonhosted.org/packages/97/45/4b3a1239bbacd321068ea6e7ac28875b03ab8bc0aa0966452db17cd36714/charset_normalizer-3.4.4-cp313-cp313-macosx_10_13_universal2.whl", hash = "sha256:e1f185f86a6f3403aa2420e815904c67b2f9ebc443f045edd0de921108345794", size = 208091, upload-time = "2025-10-14T04:41:13.346Z" },
+    { url = "https://files.pythonhosted.org/packages/7d/62/73a6d7450829655a35bb88a88fca7d736f9882a27eacdca2c6d505b57e2e/charset_normalizer-3.4.4-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:6b39f987ae8ccdf0d2642338faf2abb1862340facc796048b604ef14919e55ed", size = 147936, upload-time = "2025-10-14T04:41:14.461Z" },
+    { url = "https://files.pythonhosted.org/packages/89/c5/adb8c8b3d6625bef6d88b251bbb0d95f8205831b987631ab0c8bb5d937c2/charset_normalizer-3.4.4-cp313-cp313-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:3162d5d8ce1bb98dd51af660f2121c55d0fa541b46dff7bb9b9f86ea1d87de72", size = 144180, upload-time = "2025-10-14T04:41:15.588Z" },
+    { url = "https://files.pythonhosted.org/packages/91/ed/9706e4070682d1cc219050b6048bfd293ccf67b3d4f5a4f39207453d4b99/charset_normalizer-3.4.4-cp313-cp313-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:81d5eb2a312700f4ecaa977a8235b634ce853200e828fbadf3a9c50bab278328", size = 161346, upload-time = "2025-10-14T04:41:16.738Z" },
+    { url = "https://files.pythonhosted.org/packages/d5/0d/031f0d95e4972901a2f6f09ef055751805ff541511dc1252ba3ca1f80cf5/charset_normalizer-3.4.4-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:5bd2293095d766545ec1a8f612559f6b40abc0eb18bb2f5d1171872d34036ede", size = 158874, upload-time = "2025-10-14T04:41:17.923Z" },
+    { url = "https://files.pythonhosted.org/packages/f5/83/6ab5883f57c9c801ce5e5677242328aa45592be8a00644310a008d04f922/charset_normalizer-3.4.4-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:a8a8b89589086a25749f471e6a900d3f662d1d3b6e2e59dcecf787b1cc3a1894", size = 153076, upload-time = "2025-10-14T04:41:19.106Z" },
+    { url = "https://files.pythonhosted.org/packages/75/1e/5ff781ddf5260e387d6419959ee89ef13878229732732ee73cdae01800f2/charset_normalizer-3.4.4-cp313-cp313-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:bc7637e2f80d8530ee4a78e878bce464f70087ce73cf7c1caf142416923b98f1", size = 150601, upload-time = "2025-10-14T04:41:20.245Z" },
+    { url = "https://files.pythonhosted.org/packages/d7/57/71be810965493d3510a6ca79b90c19e48696fb1ff964da319334b12677f0/charset_normalizer-3.4.4-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:f8bf04158c6b607d747e93949aa60618b61312fe647a6369f88ce2ff16043490", size = 150376, upload-time = "2025-10-14T04:41:21.398Z" },
+    { url = "https://files.pythonhosted.org/packages/e5/d5/c3d057a78c181d007014feb7e9f2e65905a6c4ef182c0ddf0de2924edd65/charset_normalizer-3.4.4-cp313-cp313-musllinux_1_2_armv7l.whl", hash = "sha256:554af85e960429cf30784dd47447d5125aaa3b99a6f0683589dbd27e2f45da44", size = 144825, upload-time = "2025-10-14T04:41:22.583Z" },
+    { url = "https://files.pythonhosted.org/packages/e6/8c/d0406294828d4976f275ffbe66f00266c4b3136b7506941d87c00cab5272/charset_normalizer-3.4.4-cp313-cp313-musllinux_1_2_ppc64le.whl", hash = "sha256:74018750915ee7ad843a774364e13a3db91682f26142baddf775342c3f5b1133", size = 162583, upload-time = "2025-10-14T04:41:23.754Z" },
+    { url = "https://files.pythonhosted.org/packages/d7/24/e2aa1f18c8f15c4c0e932d9287b8609dd30ad56dbe41d926bd846e22fb8d/charset_normalizer-3.4.4-cp313-cp313-musllinux_1_2_riscv64.whl", hash = "sha256:c0463276121fdee9c49b98908b3a89c39be45d86d1dbaa22957e38f6321d4ce3", size = 150366, upload-time = "2025-10-14T04:41:25.27Z" },
+    { url = "https://files.pythonhosted.org/packages/e4/5b/1e6160c7739aad1e2df054300cc618b06bf784a7a164b0f238360721ab86/charset_normalizer-3.4.4-cp313-cp313-musllinux_1_2_s390x.whl", hash = "sha256:362d61fd13843997c1c446760ef36f240cf81d3ebf74ac62652aebaf7838561e", size = 160300, upload-time = "2025-10-14T04:41:26.725Z" },
+    { url = "https://files.pythonhosted.org/packages/7a/10/f882167cd207fbdd743e55534d5d9620e095089d176d55cb22d5322f2afd/charset_normalizer-3.4.4-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:9a26f18905b8dd5d685d6d07b0cdf98a79f3c7a918906af7cc143ea2e164c8bc", size = 154465, upload-time = "2025-10-14T04:41:28.322Z" },
+    { url = "https://files.pythonhosted.org/packages/89/66/c7a9e1b7429be72123441bfdbaf2bc13faab3f90b933f664db506dea5915/charset_normalizer-3.4.4-cp313-cp313-win32.whl", hash = "sha256:9b35f4c90079ff2e2edc5b26c0c77925e5d2d255c42c74fdb70fb49b172726ac", size = 99404, upload-time = "2025-10-14T04:41:29.95Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/26/b9924fa27db384bdcd97ab83b4f0a8058d96ad9626ead570674d5e737d90/charset_normalizer-3.4.4-cp313-cp313-win_amd64.whl", hash = "sha256:b435cba5f4f750aa6c0a0d92c541fb79f69a387c91e61f1795227e4ed9cece14", size = 107092, upload-time = "2025-10-14T04:41:31.188Z" },
+    { url = "https://files.pythonhosted.org/packages/af/8f/3ed4bfa0c0c72a7ca17f0380cd9e4dd842b09f664e780c13cff1dcf2ef1b/charset_normalizer-3.4.4-cp313-cp313-win_arm64.whl", hash = "sha256:542d2cee80be6f80247095cc36c418f7bddd14f4a6de45af91dfad36d817bba2", size = 100408, upload-time = "2025-10-14T04:41:32.624Z" },
+    { url = "https://files.pythonhosted.org/packages/2a/35/7051599bd493e62411d6ede36fd5af83a38f37c4767b92884df7301db25d/charset_normalizer-3.4.4-cp314-cp314-macosx_10_13_universal2.whl", hash = "sha256:da3326d9e65ef63a817ecbcc0df6e94463713b754fe293eaa03da99befb9a5bd", size = 207746, upload-time = "2025-10-14T04:41:33.773Z" },
+    { url = "https://files.pythonhosted.org/packages/10/9a/97c8d48ef10d6cd4fcead2415523221624bf58bcf68a802721a6bc807c8f/charset_normalizer-3.4.4-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:8af65f14dc14a79b924524b1e7fffe304517b2bff5a58bf64f30b98bbc5079eb", size = 147889, upload-time = "2025-10-14T04:41:34.897Z" },
+    { url = "https://files.pythonhosted.org/packages/10/bf/979224a919a1b606c82bd2c5fa49b5c6d5727aa47b4312bb27b1734f53cd/charset_normalizer-3.4.4-cp314-cp314-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:74664978bb272435107de04e36db5a9735e78232b85b77d45cfb38f758efd33e", size = 143641, upload-time = "2025-10-14T04:41:36.116Z" },
+    { url = "https://files.pythonhosted.org/packages/ba/33/0ad65587441fc730dc7bd90e9716b30b4702dc7b617e6ba4997dc8651495/charset_normalizer-3.4.4-cp314-cp314-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:752944c7ffbfdd10c074dc58ec2d5a8a4cd9493b314d367c14d24c17684ddd14", size = 160779, upload-time = "2025-10-14T04:41:37.229Z" },
+    { url = "https://files.pythonhosted.org/packages/67/ed/331d6b249259ee71ddea93f6f2f0a56cfebd46938bde6fcc6f7b9a3d0e09/charset_normalizer-3.4.4-cp314-cp314-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:d1f13550535ad8cff21b8d757a3257963e951d96e20ec82ab44bc64aeb62a191", size = 159035, upload-time = "2025-10-14T04:41:38.368Z" },
+    { url = "https://files.pythonhosted.org/packages/67/ff/f6b948ca32e4f2a4576aa129d8bed61f2e0543bf9f5f2b7fc3758ed005c9/charset_normalizer-3.4.4-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:ecaae4149d99b1c9e7b88bb03e3221956f68fd6d50be2ef061b2381b61d20838", size = 152542, upload-time = "2025-10-14T04:41:39.862Z" },
+    { url = "https://files.pythonhosted.org/packages/16/85/276033dcbcc369eb176594de22728541a925b2632f9716428c851b149e83/charset_normalizer-3.4.4-cp314-cp314-manylinux_2_31_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:cb6254dc36b47a990e59e1068afacdcd02958bdcce30bb50cc1700a8b9d624a6", size = 149524, upload-time = "2025-10-14T04:41:41.319Z" },
+    { url = "https://files.pythonhosted.org/packages/9e/f2/6a2a1f722b6aba37050e626530a46a68f74e63683947a8acff92569f979a/charset_normalizer-3.4.4-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:c8ae8a0f02f57a6e61203a31428fa1d677cbe50c93622b4149d5c0f319c1d19e", size = 150395, upload-time = "2025-10-14T04:41:42.539Z" },
+    { url = "https://files.pythonhosted.org/packages/60/bb/2186cb2f2bbaea6338cad15ce23a67f9b0672929744381e28b0592676824/charset_normalizer-3.4.4-cp314-cp314-musllinux_1_2_armv7l.whl", hash = "sha256:47cc91b2f4dd2833fddaedd2893006b0106129d4b94fdb6af1f4ce5a9965577c", size = 143680, upload-time = "2025-10-14T04:41:43.661Z" },
+    { url = "https://files.pythonhosted.org/packages/7d/a5/bf6f13b772fbb2a90360eb620d52ed8f796f3c5caee8398c3b2eb7b1c60d/charset_normalizer-3.4.4-cp314-cp314-musllinux_1_2_ppc64le.whl", hash = "sha256:82004af6c302b5d3ab2cfc4cc5f29db16123b1a8417f2e25f9066f91d4411090", size = 162045, upload-time = "2025-10-14T04:41:44.821Z" },
+    { url = "https://files.pythonhosted.org/packages/df/c5/d1be898bf0dc3ef9030c3825e5d3b83f2c528d207d246cbabe245966808d/charset_normalizer-3.4.4-cp314-cp314-musllinux_1_2_riscv64.whl", hash = "sha256:2b7d8f6c26245217bd2ad053761201e9f9680f8ce52f0fcd8d0755aeae5b2152", size = 149687, upload-time = "2025-10-14T04:41:46.442Z" },
+    { url = "https://files.pythonhosted.org/packages/a5/42/90c1f7b9341eef50c8a1cb3f098ac43b0508413f33affd762855f67a410e/charset_normalizer-3.4.4-cp314-cp314-musllinux_1_2_s390x.whl", hash = "sha256:799a7a5e4fb2d5898c60b640fd4981d6a25f1c11790935a44ce38c54e985f828", size = 160014, upload-time = "2025-10-14T04:41:47.631Z" },
+    { url = "https://files.pythonhosted.org/packages/76/be/4d3ee471e8145d12795ab655ece37baed0929462a86e72372fd25859047c/charset_normalizer-3.4.4-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:99ae2cffebb06e6c22bdc25801d7b30f503cc87dbd283479e7b606f70aff57ec", size = 154044, upload-time = "2025-10-14T04:41:48.81Z" },
+    { url = "https://files.pythonhosted.org/packages/b0/6f/8f7af07237c34a1defe7defc565a9bc1807762f672c0fde711a4b22bf9c0/charset_normalizer-3.4.4-cp314-cp314-win32.whl", hash = "sha256:f9d332f8c2a2fcbffe1378594431458ddbef721c1769d78e2cbc06280d8155f9", size = 99940, upload-time = "2025-10-14T04:41:49.946Z" },
+    { url = "https://files.pythonhosted.org/packages/4b/51/8ade005e5ca5b0d80fb4aff72a3775b325bdc3d27408c8113811a7cbe640/charset_normalizer-3.4.4-cp314-cp314-win_amd64.whl", hash = "sha256:8a6562c3700cce886c5be75ade4a5db4214fda19fede41d9792d100288d8f94c", size = 107104, upload-time = "2025-10-14T04:41:51.051Z" },
+    { url = "https://files.pythonhosted.org/packages/da/5f/6b8f83a55bb8278772c5ae54a577f3099025f9ade59d0136ac24a0df4bde/charset_normalizer-3.4.4-cp314-cp314-win_arm64.whl", hash = "sha256:de00632ca48df9daf77a2c65a484531649261ec9f25489917f09e455cb09ddb2", size = 100743, upload-time = "2025-10-14T04:41:52.122Z" },
+    { url = "https://files.pythonhosted.org/packages/0a/4c/925909008ed5a988ccbb72dcc897407e5d6d3bd72410d69e051fc0c14647/charset_normalizer-3.4.4-py3-none-any.whl", hash = "sha256:7a32c560861a02ff789ad905a2fe94e3f840803362c84fecf1851cb4cf3dc37f", size = 53402, upload-time = "2025-10-14T04:42:31.76Z" },
+]
+
+[[package]]
+name = "click"
+version = "8.3.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "colorama", marker = "sys_platform == 'win32'" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/3d/fa/656b739db8587d7b5dfa22e22ed02566950fbfbcdc20311993483657a5c0/click-8.3.1.tar.gz", hash = "sha256:12ff4785d337a1bb490bb7e9c2b1ee5da3112e94a8622f26a6c77f5d2fc6842a", size = 295065, upload-time = "2025-11-15T20:45:42.706Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/98/78/01c019cdb5d6498122777c1a43056ebb3ebfeef2076d9d026bfe15583b2b/click-8.3.1-py3-none-any.whl", hash = "sha256:981153a64e25f12d547d3426c367a4857371575ee7ad18df2a6183ab0545b2a6", size = 108274, upload-time = "2025-11-15T20:45:41.139Z" },
+]
+
+[[package]]
+name = "colorama"
+version = "0.4.6"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/d8/53/6f443c9a4a8358a93a6792e2acffb9d9d5cb0a5cfd8802644b7b1c9a02e4/colorama-0.4.6.tar.gz", hash = "sha256:08695f5cb7ed6e0531a20572697297273c47b8cae5a63ffc6d6ed5c201be6e44", size = 27697, upload-time = "2022-10-25T02:36:22.414Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/d1/d6/3965ed04c63042e047cb6a3e6ed1a63a35087b6a609aa3a15ed8ac56c221/colorama-0.4.6-py2.py3-none-any.whl", hash = "sha256:4f1d9991f5acc0ca119f9d443620b77f9d6b33703e51011c16baf57afb285fc6", size = 25335, upload-time = "2022-10-25T02:36:20.889Z" },
+]
+
+[[package]]
+name = "colorlog"
+version = "6.10.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "colorama", marker = "sys_platform == 'win32'" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/a2/61/f083b5ac52e505dfc1c624eafbf8c7589a0d7f32daa398d2e7590efa5fda/colorlog-6.10.1.tar.gz", hash = "sha256:eb4ae5cb65fe7fec7773c2306061a8e63e02efc2c72eba9d27b0fa23c94f1321", size = 17162, upload-time = "2025-10-16T16:14:11.978Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/6d/c1/e419ef3723a074172b68aaa89c9f3de486ed4c2399e2dbd8113a4fdcaf9e/colorlog-6.10.1-py3-none-any.whl", hash = "sha256:2d7e8348291948af66122cff006c9f8da6255d224e7cf8e37d8de2df3bad8c9c", size = 11743, upload-time = "2025-10-16T16:14:10.512Z" },
+]
+
+[[package]]
+name = "exceptiongroup"
+version = "1.3.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "typing-extensions", marker = "python_full_version < '3.11'" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/50/79/66800aadf48771f6b62f7eb014e352e5d06856655206165d775e675a02c9/exceptiongroup-1.3.1.tar.gz", hash = "sha256:8b412432c6055b0b7d14c310000ae93352ed6754f70fa8f7c34141f91c4e3219", size = 30371, upload-time = "2025-11-21T23:01:54.787Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/8a/0e/97c33bf5009bdbac74fd2beace167cab3f978feb69cc36f1ef79360d6c4e/exceptiongroup-1.3.1-py3-none-any.whl", hash = "sha256:a7a39a3bd276781e98394987d3a5701d0c4edffb633bb7a5144577f82c773598", size = 16740, upload-time = "2025-11-21T23:01:53.443Z" },
+]
+
+[[package]]
+name = "fastapi"
+version = "0.128.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "annotated-doc" },
+    { name = "pydantic" },
+    { name = "starlette" },
+    { name = "typing-extensions" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/52/08/8c8508db6c7b9aae8f7175046af41baad690771c9bcde676419965e338c7/fastapi-0.128.0.tar.gz", hash = "sha256:1cc179e1cef10a6be60ffe429f79b829dce99d8de32d7acb7e6c8dfdf7f2645a", size = 365682, upload-time = "2025-12-27T15:21:13.714Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/5c/05/5cbb59154b093548acd0f4c7c474a118eda06da25aa75c616b72d8fcd92a/fastapi-0.128.0-py3-none-any.whl", hash = "sha256:aebd93f9716ee3b4f4fcfe13ffb7cf308d99c9f3ab5622d8877441072561582d", size = 103094, upload-time = "2025-12-27T15:21:12.154Z" },
+]
+
+[[package]]
+name = "filelock"
+version = "3.20.3"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/1d/65/ce7f1b70157833bf3cb851b556a37d4547ceafc158aa9b34b36782f23696/filelock-3.20.3.tar.gz", hash = "sha256:18c57ee915c7ec61cff0ecf7f0f869936c7c30191bb0cf406f1341778d0834e1", size = 19485, upload-time = "2026-01-09T17:55:05.421Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/b5/36/7fb70f04bf00bc646cd5bb45aa9eddb15e19437a28b8fb2b4a5249fac770/filelock-3.20.3-py3-none-any.whl", hash = "sha256:4b0dda527ee31078689fc205ec4f1c1bf7d56cf88b6dc9426c4f230e46c2dce1", size = 16701, upload-time = "2026-01-09T17:55:04.334Z" },
+]
+
+[[package]]
+name = "filetype"
+version = "1.2.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/bb/29/745f7d30d47fe0f251d3ad3dc2978a23141917661998763bebb6da007eb1/filetype-1.2.0.tar.gz", hash = "sha256:66b56cd6474bf41d8c54660347d37afcc3f7d1970648de365c102ef77548aadb", size = 998020, upload-time = "2022-11-02T17:34:04.141Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/18/79/1b8fa1bb3568781e84c9200f951c735f3f157429f44be0495da55894d620/filetype-1.2.0-py2.py3-none-any.whl", hash = "sha256:7ce71b6880181241cf7ac8697a2f1eb6a8bd9b429f7ad6d27b8db9ba5f1c2d25", size = 19970, upload-time = "2022-11-02T17:34:01.425Z" },
+]
+
+[[package]]
+name = "freezegun"
+version = "1.5.5"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "python-dateutil" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/95/dd/23e2f4e357f8fd3bdff613c1fe4466d21bfb00a6177f238079b17f7b1c84/freezegun-1.5.5.tar.gz", hash = "sha256:ac7742a6cc6c25a2c35e9292dfd554b897b517d2dec26891a2e8debf205cb94a", size = 35914, upload-time = "2025-08-09T10:39:08.338Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/5e/2e/b41d8a1a917d6581fc27a35d05561037b048e47df50f27f8ac9c7e27a710/freezegun-1.5.5-py3-none-any.whl", hash = "sha256:cd557f4a75cf074e84bc374249b9dd491eaeacd61376b9eb3c423282211619d2", size = 19266, upload-time = "2025-08-09T10:39:06.636Z" },
+]
+
+[[package]]
+name = "frozenlist"
+version = "1.8.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/2d/f5/c831fac6cc817d26fd54c7eaccd04ef7e0288806943f7cc5bbf69f3ac1f0/frozenlist-1.8.0.tar.gz", hash = "sha256:3ede829ed8d842f6cd48fc7081d7a41001a56f1f38603f9d49bf3020d59a31ad", size = 45875, upload-time = "2025-10-06T05:38:17.865Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/83/4a/557715d5047da48d54e659203b9335be7bfaafda2c3f627b7c47e0b3aaf3/frozenlist-1.8.0-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:b37f6d31b3dcea7deb5e9696e529a6aa4a898adc33db82da12e4c60a7c4d2011", size = 86230, upload-time = "2025-10-06T05:35:23.699Z" },
+    { url = "https://files.pythonhosted.org/packages/a2/fb/c85f9fed3ea8fe8740e5b46a59cc141c23b842eca617da8876cfce5f760e/frozenlist-1.8.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:ef2b7b394f208233e471abc541cc6991f907ffd47dc72584acee3147899d6565", size = 49621, upload-time = "2025-10-06T05:35:25.341Z" },
+    { url = "https://files.pythonhosted.org/packages/63/70/26ca3f06aace16f2352796b08704338d74b6d1a24ca38f2771afbb7ed915/frozenlist-1.8.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:a88f062f072d1589b7b46e951698950e7da00442fc1cacbe17e19e025dc327ad", size = 49889, upload-time = "2025-10-06T05:35:26.797Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/ed/c7895fd2fde7f3ee70d248175f9b6cdf792fb741ab92dc59cd9ef3bd241b/frozenlist-1.8.0-cp310-cp310-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl", hash = "sha256:f57fb59d9f385710aa7060e89410aeb5058b99e62f4d16b08b91986b9a2140c2", size = 219464, upload-time = "2025-10-06T05:35:28.254Z" },
+    { url = "https://files.pythonhosted.org/packages/6b/83/4d587dccbfca74cb8b810472392ad62bfa100bf8108c7223eb4c4fa2f7b3/frozenlist-1.8.0-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:799345ab092bee59f01a915620b5d014698547afd011e691a208637312db9186", size = 221649, upload-time = "2025-10-06T05:35:29.454Z" },
+    { url = "https://files.pythonhosted.org/packages/6a/c6/fd3b9cd046ec5fff9dab66831083bc2077006a874a2d3d9247dea93ddf7e/frozenlist-1.8.0-cp310-cp310-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:c23c3ff005322a6e16f71bf8692fcf4d5a304aaafe1e262c98c6d4adc7be863e", size = 219188, upload-time = "2025-10-06T05:35:30.951Z" },
+    { url = "https://files.pythonhosted.org/packages/ce/80/6693f55eb2e085fc8afb28cf611448fb5b90e98e068fa1d1b8d8e66e5c7d/frozenlist-1.8.0-cp310-cp310-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:8a76ea0f0b9dfa06f254ee06053d93a600865b3274358ca48a352ce4f0798450", size = 231748, upload-time = "2025-10-06T05:35:32.101Z" },
+    { url = "https://files.pythonhosted.org/packages/97/d6/e9459f7c5183854abd989ba384fe0cc1a0fb795a83c033f0571ec5933ca4/frozenlist-1.8.0-cp310-cp310-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:c7366fe1418a6133d5aa824ee53d406550110984de7637d65a178010f759c6ef", size = 236351, upload-time = "2025-10-06T05:35:33.834Z" },
+    { url = "https://files.pythonhosted.org/packages/97/92/24e97474b65c0262e9ecd076e826bfd1d3074adcc165a256e42e7b8a7249/frozenlist-1.8.0-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:13d23a45c4cebade99340c4165bd90eeb4a56c6d8a9d8aa49568cac19a6d0dc4", size = 218767, upload-time = "2025-10-06T05:35:35.205Z" },
+    { url = "https://files.pythonhosted.org/packages/ee/bf/dc394a097508f15abff383c5108cb8ad880d1f64a725ed3b90d5c2fbf0bb/frozenlist-1.8.0-cp310-cp310-musllinux_1_2_armv7l.whl", hash = "sha256:e4a3408834f65da56c83528fb52ce7911484f0d1eaf7b761fc66001db1646eff", size = 235887, upload-time = "2025-10-06T05:35:36.354Z" },
+    { url = "https://files.pythonhosted.org/packages/40/90/25b201b9c015dbc999a5baf475a257010471a1fa8c200c843fd4abbee725/frozenlist-1.8.0-cp310-cp310-musllinux_1_2_ppc64le.whl", hash = "sha256:42145cd2748ca39f32801dad54aeea10039da6f86e303659db90db1c4b614c8c", size = 228785, upload-time = "2025-10-06T05:35:37.949Z" },
+    { url = "https://files.pythonhosted.org/packages/84/f4/b5bc148df03082f05d2dd30c089e269acdbe251ac9a9cf4e727b2dbb8a3d/frozenlist-1.8.0-cp310-cp310-musllinux_1_2_s390x.whl", hash = "sha256:e2de870d16a7a53901e41b64ffdf26f2fbb8917b3e6ebf398098d72c5b20bd7f", size = 230312, upload-time = "2025-10-06T05:35:39.178Z" },
+    { url = "https://files.pythonhosted.org/packages/db/4b/87e95b5d15097c302430e647136b7d7ab2398a702390cf4c8601975709e7/frozenlist-1.8.0-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:20e63c9493d33ee48536600d1a5c95eefc870cd71e7ab037763d1fbb89cc51e7", size = 217650, upload-time = "2025-10-06T05:35:40.377Z" },
+    { url = "https://files.pythonhosted.org/packages/e5/70/78a0315d1fea97120591a83e0acd644da638c872f142fd72a6cebee825f3/frozenlist-1.8.0-cp310-cp310-win32.whl", hash = "sha256:adbeebaebae3526afc3c96fad434367cafbfd1b25d72369a9e5858453b1bb71a", size = 39659, upload-time = "2025-10-06T05:35:41.863Z" },
+    { url = "https://files.pythonhosted.org/packages/66/aa/3f04523fb189a00e147e60c5b2205126118f216b0aa908035c45336e27e4/frozenlist-1.8.0-cp310-cp310-win_amd64.whl", hash = "sha256:667c3777ca571e5dbeb76f331562ff98b957431df140b54c85fd4d52eea8d8f6", size = 43837, upload-time = "2025-10-06T05:35:43.205Z" },
+    { url = "https://files.pythonhosted.org/packages/39/75/1135feecdd7c336938bd55b4dc3b0dfc46d85b9be12ef2628574b28de776/frozenlist-1.8.0-cp310-cp310-win_arm64.whl", hash = "sha256:80f85f0a7cc86e7a54c46d99c9e1318ff01f4687c172ede30fd52d19d1da1c8e", size = 39989, upload-time = "2025-10-06T05:35:44.596Z" },
+    { url = "https://files.pythonhosted.org/packages/bc/03/077f869d540370db12165c0aa51640a873fb661d8b315d1d4d67b284d7ac/frozenlist-1.8.0-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:09474e9831bc2b2199fad6da3c14c7b0fbdd377cce9d3d77131be28906cb7d84", size = 86912, upload-time = "2025-10-06T05:35:45.98Z" },
+    { url = "https://files.pythonhosted.org/packages/df/b5/7610b6bd13e4ae77b96ba85abea1c8cb249683217ef09ac9e0ae93f25a91/frozenlist-1.8.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:17c883ab0ab67200b5f964d2b9ed6b00971917d5d8a92df149dc2c9779208ee9", size = 50046, upload-time = "2025-10-06T05:35:47.009Z" },
+    { url = "https://files.pythonhosted.org/packages/6e/ef/0e8f1fe32f8a53dd26bdd1f9347efe0778b0fddf62789ea683f4cc7d787d/frozenlist-1.8.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:fa47e444b8ba08fffd1c18e8cdb9a75db1b6a27f17507522834ad13ed5922b93", size = 50119, upload-time = "2025-10-06T05:35:48.38Z" },
+    { url = "https://files.pythonhosted.org/packages/11/b1/71a477adc7c36e5fb628245dfbdea2166feae310757dea848d02bd0689fd/frozenlist-1.8.0-cp311-cp311-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl", hash = "sha256:2552f44204b744fba866e573be4c1f9048d6a324dfe14475103fd51613eb1d1f", size = 231067, upload-time = "2025-10-06T05:35:49.97Z" },
+    { url = "https://files.pythonhosted.org/packages/45/7e/afe40eca3a2dc19b9904c0f5d7edfe82b5304cb831391edec0ac04af94c2/frozenlist-1.8.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:957e7c38f250991e48a9a73e6423db1bb9dd14e722a10f6b8bb8e16a0f55f695", size = 233160, upload-time = "2025-10-06T05:35:51.729Z" },
+    { url = "https://files.pythonhosted.org/packages/a6/aa/7416eac95603ce428679d273255ffc7c998d4132cfae200103f164b108aa/frozenlist-1.8.0-cp311-cp311-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:8585e3bb2cdea02fc88ffa245069c36555557ad3609e83be0ec71f54fd4abb52", size = 228544, upload-time = "2025-10-06T05:35:53.246Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/3d/2a2d1f683d55ac7e3875e4263d28410063e738384d3adc294f5ff3d7105e/frozenlist-1.8.0-cp311-cp311-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:edee74874ce20a373d62dc28b0b18b93f645633c2943fd90ee9d898550770581", size = 243797, upload-time = "2025-10-06T05:35:54.497Z" },
+    { url = "https://files.pythonhosted.org/packages/78/1e/2d5565b589e580c296d3bb54da08d206e797d941a83a6fdea42af23be79c/frozenlist-1.8.0-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:c9a63152fe95756b85f31186bddf42e4c02c6321207fd6601a1c89ebac4fe567", size = 247923, upload-time = "2025-10-06T05:35:55.861Z" },
+    { url = "https://files.pythonhosted.org/packages/aa/c3/65872fcf1d326a7f101ad4d86285c403c87be7d832b7470b77f6d2ed5ddc/frozenlist-1.8.0-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:b6db2185db9be0a04fecf2f241c70b63b1a242e2805be291855078f2b404dd6b", size = 230886, upload-time = "2025-10-06T05:35:57.399Z" },
+    { url = "https://files.pythonhosted.org/packages/a0/76/ac9ced601d62f6956f03cc794f9e04c81719509f85255abf96e2510f4265/frozenlist-1.8.0-cp311-cp311-musllinux_1_2_armv7l.whl", hash = "sha256:f4be2e3d8bc8aabd566f8d5b8ba7ecc09249d74ba3c9ed52e54dc23a293f0b92", size = 245731, upload-time = "2025-10-06T05:35:58.563Z" },
+    { url = "https://files.pythonhosted.org/packages/b9/49/ecccb5f2598daf0b4a1415497eba4c33c1e8ce07495eb07d2860c731b8d5/frozenlist-1.8.0-cp311-cp311-musllinux_1_2_ppc64le.whl", hash = "sha256:c8d1634419f39ea6f5c427ea2f90ca85126b54b50837f31497f3bf38266e853d", size = 241544, upload-time = "2025-10-06T05:35:59.719Z" },
+    { url = "https://files.pythonhosted.org/packages/53/4b/ddf24113323c0bbcc54cb38c8b8916f1da7165e07b8e24a717b4a12cbf10/frozenlist-1.8.0-cp311-cp311-musllinux_1_2_s390x.whl", hash = "sha256:1a7fa382a4a223773ed64242dbe1c9c326ec09457e6b8428efb4118c685c3dfd", size = 241806, upload-time = "2025-10-06T05:36:00.959Z" },
+    { url = "https://files.pythonhosted.org/packages/a7/fb/9b9a084d73c67175484ba2789a59f8eebebd0827d186a8102005ce41e1ba/frozenlist-1.8.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:11847b53d722050808926e785df837353bd4d75f1d494377e59b23594d834967", size = 229382, upload-time = "2025-10-06T05:36:02.22Z" },
+    { url = "https://files.pythonhosted.org/packages/95/a3/c8fb25aac55bf5e12dae5c5aa6a98f85d436c1dc658f21c3ac73f9fa95e5/frozenlist-1.8.0-cp311-cp311-win32.whl", hash = "sha256:27c6e8077956cf73eadd514be8fb04d77fc946a7fe9f7fe167648b0b9085cc25", size = 39647, upload-time = "2025-10-06T05:36:03.409Z" },
+    { url = "https://files.pythonhosted.org/packages/0a/f5/603d0d6a02cfd4c8f2a095a54672b3cf967ad688a60fb9faf04fc4887f65/frozenlist-1.8.0-cp311-cp311-win_amd64.whl", hash = "sha256:ac913f8403b36a2c8610bbfd25b8013488533e71e62b4b4adce9c86c8cea905b", size = 44064, upload-time = "2025-10-06T05:36:04.368Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/16/c2c9ab44e181f043a86f9a8f84d5124b62dbcb3a02c0977ec72b9ac1d3e0/frozenlist-1.8.0-cp311-cp311-win_arm64.whl", hash = "sha256:d4d3214a0f8394edfa3e303136d0575eece0745ff2b47bd2cb2e66dd92d4351a", size = 39937, upload-time = "2025-10-06T05:36:05.669Z" },
+    { url = "https://files.pythonhosted.org/packages/69/29/948b9aa87e75820a38650af445d2ef2b6b8a6fab1a23b6bb9e4ef0be2d59/frozenlist-1.8.0-cp312-cp312-macosx_10_13_universal2.whl", hash = "sha256:78f7b9e5d6f2fdb88cdde9440dc147259b62b9d3b019924def9f6478be254ac1", size = 87782, upload-time = "2025-10-06T05:36:06.649Z" },
+    { url = "https://files.pythonhosted.org/packages/64/80/4f6e318ee2a7c0750ed724fa33a4bdf1eacdc5a39a7a24e818a773cd91af/frozenlist-1.8.0-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:229bf37d2e4acdaf808fd3f06e854a4a7a3661e871b10dc1f8f1896a3b05f18b", size = 50594, upload-time = "2025-10-06T05:36:07.69Z" },
+    { url = "https://files.pythonhosted.org/packages/2b/94/5c8a2b50a496b11dd519f4a24cb5496cf125681dd99e94c604ccdea9419a/frozenlist-1.8.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:f833670942247a14eafbb675458b4e61c82e002a148f49e68257b79296e865c4", size = 50448, upload-time = "2025-10-06T05:36:08.78Z" },
+    { url = "https://files.pythonhosted.org/packages/6a/bd/d91c5e39f490a49df14320f4e8c80161cfcce09f1e2cde1edd16a551abb3/frozenlist-1.8.0-cp312-cp312-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl", hash = "sha256:494a5952b1c597ba44e0e78113a7266e656b9794eec897b19ead706bd7074383", size = 242411, upload-time = "2025-10-06T05:36:09.801Z" },
+    { url = "https://files.pythonhosted.org/packages/8f/83/f61505a05109ef3293dfb1ff594d13d64a2324ac3482be2cedc2be818256/frozenlist-1.8.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:96f423a119f4777a4a056b66ce11527366a8bb92f54e541ade21f2374433f6d4", size = 243014, upload-time = "2025-10-06T05:36:11.394Z" },
+    { url = "https://files.pythonhosted.org/packages/d8/cb/cb6c7b0f7d4023ddda30cf56b8b17494eb3a79e3fda666bf735f63118b35/frozenlist-1.8.0-cp312-cp312-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:3462dd9475af2025c31cc61be6652dfa25cbfb56cbbf52f4ccfe029f38decaf8", size = 234909, upload-time = "2025-10-06T05:36:12.598Z" },
+    { url = "https://files.pythonhosted.org/packages/31/c5/cd7a1f3b8b34af009fb17d4123c5a778b44ae2804e3ad6b86204255f9ec5/frozenlist-1.8.0-cp312-cp312-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:c4c800524c9cd9bac5166cd6f55285957fcfc907db323e193f2afcd4d9abd69b", size = 250049, upload-time = "2025-10-06T05:36:14.065Z" },
+    { url = "https://files.pythonhosted.org/packages/c0/01/2f95d3b416c584a1e7f0e1d6d31998c4a795f7544069ee2e0962a4b60740/frozenlist-1.8.0-cp312-cp312-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:d6a5df73acd3399d893dafc71663ad22534b5aa4f94e8a2fabfe856c3c1b6a52", size = 256485, upload-time = "2025-10-06T05:36:15.39Z" },
+    { url = "https://files.pythonhosted.org/packages/ce/03/024bf7720b3abaebcff6d0793d73c154237b85bdf67b7ed55e5e9596dc9a/frozenlist-1.8.0-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:405e8fe955c2280ce66428b3ca55e12b3c4e9c336fb2103a4937e891c69a4a29", size = 237619, upload-time = "2025-10-06T05:36:16.558Z" },
+    { url = "https://files.pythonhosted.org/packages/69/fa/f8abdfe7d76b731f5d8bd217827cf6764d4f1d9763407e42717b4bed50a0/frozenlist-1.8.0-cp312-cp312-musllinux_1_2_armv7l.whl", hash = "sha256:908bd3f6439f2fef9e85031b59fd4f1297af54415fb60e4254a95f75b3cab3f3", size = 250320, upload-time = "2025-10-06T05:36:17.821Z" },
+    { url = "https://files.pythonhosted.org/packages/f5/3c/b051329f718b463b22613e269ad72138cc256c540f78a6de89452803a47d/frozenlist-1.8.0-cp312-cp312-musllinux_1_2_ppc64le.whl", hash = "sha256:294e487f9ec720bd8ffcebc99d575f7eff3568a08a253d1ee1a0378754b74143", size = 246820, upload-time = "2025-10-06T05:36:19.046Z" },
+    { url = "https://files.pythonhosted.org/packages/0f/ae/58282e8f98e444b3f4dd42448ff36fa38bef29e40d40f330b22e7108f565/frozenlist-1.8.0-cp312-cp312-musllinux_1_2_s390x.whl", hash = "sha256:74c51543498289c0c43656701be6b077f4b265868fa7f8a8859c197006efb608", size = 250518, upload-time = "2025-10-06T05:36:20.763Z" },
+    { url = "https://files.pythonhosted.org/packages/8f/96/007e5944694d66123183845a106547a15944fbbb7154788cbf7272789536/frozenlist-1.8.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:776f352e8329135506a1d6bf16ac3f87bc25b28e765949282dcc627af36123aa", size = 239096, upload-time = "2025-10-06T05:36:22.129Z" },
+    { url = "https://files.pythonhosted.org/packages/66/bb/852b9d6db2fa40be96f29c0d1205c306288f0684df8fd26ca1951d461a56/frozenlist-1.8.0-cp312-cp312-win32.whl", hash = "sha256:433403ae80709741ce34038da08511d4a77062aa924baf411ef73d1146e74faf", size = 39985, upload-time = "2025-10-06T05:36:23.661Z" },
+    { url = "https://files.pythonhosted.org/packages/b8/af/38e51a553dd66eb064cdf193841f16f077585d4d28394c2fa6235cb41765/frozenlist-1.8.0-cp312-cp312-win_amd64.whl", hash = "sha256:34187385b08f866104f0c0617404c8eb08165ab1272e884abc89c112e9c00746", size = 44591, upload-time = "2025-10-06T05:36:24.958Z" },
+    { url = "https://files.pythonhosted.org/packages/a7/06/1dc65480ab147339fecc70797e9c2f69d9cea9cf38934ce08df070fdb9cb/frozenlist-1.8.0-cp312-cp312-win_arm64.whl", hash = "sha256:fe3c58d2f5db5fbd18c2987cba06d51b0529f52bc3a6cdc33d3f4eab725104bd", size = 40102, upload-time = "2025-10-06T05:36:26.333Z" },
+    { url = "https://files.pythonhosted.org/packages/2d/40/0832c31a37d60f60ed79e9dfb5a92e1e2af4f40a16a29abcc7992af9edff/frozenlist-1.8.0-cp313-cp313-macosx_10_13_universal2.whl", hash = "sha256:8d92f1a84bb12d9e56f818b3a746f3efba93c1b63c8387a73dde655e1e42282a", size = 85717, upload-time = "2025-10-06T05:36:27.341Z" },
+    { url = "https://files.pythonhosted.org/packages/30/ba/b0b3de23f40bc55a7057bd38434e25c34fa48e17f20ee273bbde5e0650f3/frozenlist-1.8.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:96153e77a591c8adc2ee805756c61f59fef4cf4073a9275ee86fe8cba41241f7", size = 49651, upload-time = "2025-10-06T05:36:28.855Z" },
+    { url = "https://files.pythonhosted.org/packages/0c/ab/6e5080ee374f875296c4243c381bbdef97a9ac39c6e3ce1d5f7d42cb78d6/frozenlist-1.8.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:f21f00a91358803399890ab167098c131ec2ddd5f8f5fd5fe9c9f2c6fcd91e40", size = 49417, upload-time = "2025-10-06T05:36:29.877Z" },
+    { url = "https://files.pythonhosted.org/packages/d5/4e/e4691508f9477ce67da2015d8c00acd751e6287739123113a9fca6f1604e/frozenlist-1.8.0-cp313-cp313-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl", hash = "sha256:fb30f9626572a76dfe4293c7194a09fb1fe93ba94c7d4f720dfae3b646b45027", size = 234391, upload-time = "2025-10-06T05:36:31.301Z" },
+    { url = "https://files.pythonhosted.org/packages/40/76/c202df58e3acdf12969a7895fd6f3bc016c642e6726aa63bd3025e0fc71c/frozenlist-1.8.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:eaa352d7047a31d87dafcacbabe89df0aa506abb5b1b85a2fb91bc3faa02d822", size = 233048, upload-time = "2025-10-06T05:36:32.531Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/c0/8746afb90f17b73ca5979c7a3958116e105ff796e718575175319b5bb4ce/frozenlist-1.8.0-cp313-cp313-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:03ae967b4e297f58f8c774c7eabcce57fe3c2434817d4385c50661845a058121", size = 226549, upload-time = "2025-10-06T05:36:33.706Z" },
+    { url = "https://files.pythonhosted.org/packages/7e/eb/4c7eefc718ff72f9b6c4893291abaae5fbc0c82226a32dcd8ef4f7a5dbef/frozenlist-1.8.0-cp313-cp313-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:f6292f1de555ffcc675941d65fffffb0a5bcd992905015f85d0592201793e0e5", size = 239833, upload-time = "2025-10-06T05:36:34.947Z" },
+    { url = "https://files.pythonhosted.org/packages/c2/4e/e5c02187cf704224f8b21bee886f3d713ca379535f16893233b9d672ea71/frozenlist-1.8.0-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:29548f9b5b5e3460ce7378144c3010363d8035cea44bc0bf02d57f5a685e084e", size = 245363, upload-time = "2025-10-06T05:36:36.534Z" },
+    { url = "https://files.pythonhosted.org/packages/1f/96/cb85ec608464472e82ad37a17f844889c36100eed57bea094518bf270692/frozenlist-1.8.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:ec3cc8c5d4084591b4237c0a272cc4f50a5b03396a47d9caaf76f5d7b38a4f11", size = 229314, upload-time = "2025-10-06T05:36:38.582Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/6f/4ae69c550e4cee66b57887daeebe006fe985917c01d0fff9caab9883f6d0/frozenlist-1.8.0-cp313-cp313-musllinux_1_2_armv7l.whl", hash = "sha256:517279f58009d0b1f2e7c1b130b377a349405da3f7621ed6bfae50b10adf20c1", size = 243365, upload-time = "2025-10-06T05:36:40.152Z" },
+    { url = "https://files.pythonhosted.org/packages/7a/58/afd56de246cf11780a40a2c28dc7cbabbf06337cc8ddb1c780a2d97e88d8/frozenlist-1.8.0-cp313-cp313-musllinux_1_2_ppc64le.whl", hash = "sha256:db1e72ede2d0d7ccb213f218df6a078a9c09a7de257c2fe8fcef16d5925230b1", size = 237763, upload-time = "2025-10-06T05:36:41.355Z" },
+    { url = "https://files.pythonhosted.org/packages/cb/36/cdfaf6ed42e2644740d4a10452d8e97fa1c062e2a8006e4b09f1b5fd7d63/frozenlist-1.8.0-cp313-cp313-musllinux_1_2_s390x.whl", hash = "sha256:b4dec9482a65c54a5044486847b8a66bf10c9cb4926d42927ec4e8fd5db7fed8", size = 240110, upload-time = "2025-10-06T05:36:42.716Z" },
+    { url = "https://files.pythonhosted.org/packages/03/a8/9ea226fbefad669f11b52e864c55f0bd57d3c8d7eb07e9f2e9a0b39502e1/frozenlist-1.8.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:21900c48ae04d13d416f0e1e0c4d81f7931f73a9dfa0b7a8746fb2fe7dd970ed", size = 233717, upload-time = "2025-10-06T05:36:44.251Z" },
+    { url = "https://files.pythonhosted.org/packages/1e/0b/1b5531611e83ba7d13ccc9988967ea1b51186af64c42b7a7af465dcc9568/frozenlist-1.8.0-cp313-cp313-win32.whl", hash = "sha256:8b7b94a067d1c504ee0b16def57ad5738701e4ba10cec90529f13fa03c833496", size = 39628, upload-time = "2025-10-06T05:36:45.423Z" },
+    { url = "https://files.pythonhosted.org/packages/d8/cf/174c91dbc9cc49bc7b7aab74d8b734e974d1faa8f191c74af9b7e80848e6/frozenlist-1.8.0-cp313-cp313-win_amd64.whl", hash = "sha256:878be833caa6a3821caf85eb39c5ba92d28e85df26d57afb06b35b2efd937231", size = 43882, upload-time = "2025-10-06T05:36:46.796Z" },
+    { url = "https://files.pythonhosted.org/packages/c1/17/502cd212cbfa96eb1388614fe39a3fc9ab87dbbe042b66f97acb57474834/frozenlist-1.8.0-cp313-cp313-win_arm64.whl", hash = "sha256:44389d135b3ff43ba8cc89ff7f51f5a0bb6b63d829c8300f79a2fe4fe61bcc62", size = 39676, upload-time = "2025-10-06T05:36:47.8Z" },
+    { url = "https://files.pythonhosted.org/packages/d2/5c/3bbfaa920dfab09e76946a5d2833a7cbdf7b9b4a91c714666ac4855b88b4/frozenlist-1.8.0-cp313-cp313t-macosx_10_13_universal2.whl", hash = "sha256:e25ac20a2ef37e91c1b39938b591457666a0fa835c7783c3a8f33ea42870db94", size = 89235, upload-time = "2025-10-06T05:36:48.78Z" },
+    { url = "https://files.pythonhosted.org/packages/d2/d6/f03961ef72166cec1687e84e8925838442b615bd0b8854b54923ce5b7b8a/frozenlist-1.8.0-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:07cdca25a91a4386d2e76ad992916a85038a9b97561bf7a3fd12d5d9ce31870c", size = 50742, upload-time = "2025-10-06T05:36:49.837Z" },
+    { url = "https://files.pythonhosted.org/packages/1e/bb/a6d12b7ba4c3337667d0e421f7181c82dda448ce4e7ad7ecd249a16fa806/frozenlist-1.8.0-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:4e0c11f2cc6717e0a741f84a527c52616140741cd812a50422f83dc31749fb52", size = 51725, upload-time = "2025-10-06T05:36:50.851Z" },
+    { url = "https://files.pythonhosted.org/packages/bc/71/d1fed0ffe2c2ccd70b43714c6cab0f4188f09f8a67a7914a6b46ee30f274/frozenlist-1.8.0-cp313-cp313t-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl", hash = "sha256:b3210649ee28062ea6099cfda39e147fa1bc039583c8ee4481cb7811e2448c51", size = 284533, upload-time = "2025-10-06T05:36:51.898Z" },
+    { url = "https://files.pythonhosted.org/packages/c9/1f/fb1685a7b009d89f9bf78a42d94461bc06581f6e718c39344754a5d9bada/frozenlist-1.8.0-cp313-cp313t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:581ef5194c48035a7de2aefc72ac6539823bb71508189e5de01d60c9dcd5fa65", size = 292506, upload-time = "2025-10-06T05:36:53.101Z" },
+    { url = "https://files.pythonhosted.org/packages/e6/3b/b991fe1612703f7e0d05c0cf734c1b77aaf7c7d321df4572e8d36e7048c8/frozenlist-1.8.0-cp313-cp313t-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:3ef2d026f16a2b1866e1d86fc4e1291e1ed8a387b2c333809419a2f8b3a77b82", size = 274161, upload-time = "2025-10-06T05:36:54.309Z" },
+    { url = "https://files.pythonhosted.org/packages/ca/ec/c5c618767bcdf66e88945ec0157d7f6c4a1322f1473392319b7a2501ded7/frozenlist-1.8.0-cp313-cp313t-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:5500ef82073f599ac84d888e3a8c1f77ac831183244bfd7f11eaa0289fb30714", size = 294676, upload-time = "2025-10-06T05:36:55.566Z" },
+    { url = "https://files.pythonhosted.org/packages/7c/ce/3934758637d8f8a88d11f0585d6495ef54b2044ed6ec84492a91fa3b27aa/frozenlist-1.8.0-cp313-cp313t-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:50066c3997d0091c411a66e710f4e11752251e6d2d73d70d8d5d4c76442a199d", size = 300638, upload-time = "2025-10-06T05:36:56.758Z" },
+    { url = "https://files.pythonhosted.org/packages/fc/4f/a7e4d0d467298f42de4b41cbc7ddaf19d3cfeabaf9ff97c20c6c7ee409f9/frozenlist-1.8.0-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:5c1c8e78426e59b3f8005e9b19f6ff46e5845895adbde20ece9218319eca6506", size = 283067, upload-time = "2025-10-06T05:36:57.965Z" },
+    { url = "https://files.pythonhosted.org/packages/dc/48/c7b163063d55a83772b268e6d1affb960771b0e203b632cfe09522d67ea5/frozenlist-1.8.0-cp313-cp313t-musllinux_1_2_armv7l.whl", hash = "sha256:eefdba20de0d938cec6a89bd4d70f346a03108a19b9df4248d3cf0d88f1b0f51", size = 292101, upload-time = "2025-10-06T05:36:59.237Z" },
+    { url = "https://files.pythonhosted.org/packages/9f/d0/2366d3c4ecdc2fd391e0afa6e11500bfba0ea772764d631bbf82f0136c9d/frozenlist-1.8.0-cp313-cp313t-musllinux_1_2_ppc64le.whl", hash = "sha256:cf253e0e1c3ceb4aaff6df637ce033ff6535fb8c70a764a8f46aafd3d6ab798e", size = 289901, upload-time = "2025-10-06T05:37:00.811Z" },
+    { url = "https://files.pythonhosted.org/packages/b8/94/daff920e82c1b70e3618a2ac39fbc01ae3e2ff6124e80739ce5d71c9b920/frozenlist-1.8.0-cp313-cp313t-musllinux_1_2_s390x.whl", hash = "sha256:032efa2674356903cd0261c4317a561a6850f3ac864a63fc1583147fb05a79b0", size = 289395, upload-time = "2025-10-06T05:37:02.115Z" },
+    { url = "https://files.pythonhosted.org/packages/e3/20/bba307ab4235a09fdcd3cc5508dbabd17c4634a1af4b96e0f69bfe551ebd/frozenlist-1.8.0-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:6da155091429aeba16851ecb10a9104a108bcd32f6c1642867eadaee401c1c41", size = 283659, upload-time = "2025-10-06T05:37:03.711Z" },
+    { url = "https://files.pythonhosted.org/packages/fd/00/04ca1c3a7a124b6de4f8a9a17cc2fcad138b4608e7a3fc5877804b8715d7/frozenlist-1.8.0-cp313-cp313t-win32.whl", hash = "sha256:0f96534f8bfebc1a394209427d0f8a63d343c9779cda6fc25e8e121b5fd8555b", size = 43492, upload-time = "2025-10-06T05:37:04.915Z" },
+    { url = "https://files.pythonhosted.org/packages/59/5e/c69f733a86a94ab10f68e496dc6b7e8bc078ebb415281d5698313e3af3a1/frozenlist-1.8.0-cp313-cp313t-win_amd64.whl", hash = "sha256:5d63a068f978fc69421fb0e6eb91a9603187527c86b7cd3f534a5b77a592b888", size = 48034, upload-time = "2025-10-06T05:37:06.343Z" },
+    { url = "https://files.pythonhosted.org/packages/16/6c/be9d79775d8abe79b05fa6d23da99ad6e7763a1d080fbae7290b286093fd/frozenlist-1.8.0-cp313-cp313t-win_arm64.whl", hash = "sha256:bf0a7e10b077bf5fb9380ad3ae8ce20ef919a6ad93b4552896419ac7e1d8e042", size = 41749, upload-time = "2025-10-06T05:37:07.431Z" },
+    { url = "https://files.pythonhosted.org/packages/f1/c8/85da824b7e7b9b6e7f7705b2ecaf9591ba6f79c1177f324c2735e41d36a2/frozenlist-1.8.0-cp314-cp314-macosx_10_13_universal2.whl", hash = "sha256:cee686f1f4cadeb2136007ddedd0aaf928ab95216e7691c63e50a8ec066336d0", size = 86127, upload-time = "2025-10-06T05:37:08.438Z" },
+    { url = "https://files.pythonhosted.org/packages/8e/e8/a1185e236ec66c20afd72399522f142c3724c785789255202d27ae992818/frozenlist-1.8.0-cp314-cp314-macosx_10_13_x86_64.whl", hash = "sha256:119fb2a1bd47307e899c2fac7f28e85b9a543864df47aa7ec9d3c1b4545f096f", size = 49698, upload-time = "2025-10-06T05:37:09.48Z" },
+    { url = "https://files.pythonhosted.org/packages/a1/93/72b1736d68f03fda5fdf0f2180fb6caaae3894f1b854d006ac61ecc727ee/frozenlist-1.8.0-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:4970ece02dbc8c3a92fcc5228e36a3e933a01a999f7094ff7c23fbd2beeaa67c", size = 49749, upload-time = "2025-10-06T05:37:10.569Z" },
+    { url = "https://files.pythonhosted.org/packages/a7/b2/fabede9fafd976b991e9f1b9c8c873ed86f202889b864756f240ce6dd855/frozenlist-1.8.0-cp314-cp314-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl", hash = "sha256:cba69cb73723c3f329622e34bdbf5ce1f80c21c290ff04256cff1cd3c2036ed2", size = 231298, upload-time = "2025-10-06T05:37:11.993Z" },
+    { url = "https://files.pythonhosted.org/packages/3a/3b/d9b1e0b0eed36e70477ffb8360c49c85c8ca8ef9700a4e6711f39a6e8b45/frozenlist-1.8.0-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:778a11b15673f6f1df23d9586f83c4846c471a8af693a22e066508b77d201ec8", size = 232015, upload-time = "2025-10-06T05:37:13.194Z" },
+    { url = "https://files.pythonhosted.org/packages/dc/94/be719d2766c1138148564a3960fc2c06eb688da592bdc25adcf856101be7/frozenlist-1.8.0-cp314-cp314-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:0325024fe97f94c41c08872db482cf8ac4800d80e79222c6b0b7b162d5b13686", size = 225038, upload-time = "2025-10-06T05:37:14.577Z" },
+    { url = "https://files.pythonhosted.org/packages/e4/09/6712b6c5465f083f52f50cf74167b92d4ea2f50e46a9eea0523d658454ae/frozenlist-1.8.0-cp314-cp314-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:97260ff46b207a82a7567b581ab4190bd4dfa09f4db8a8b49d1a958f6aa4940e", size = 240130, upload-time = "2025-10-06T05:37:15.781Z" },
+    { url = "https://files.pythonhosted.org/packages/f8/d4/cd065cdcf21550b54f3ce6a22e143ac9e4836ca42a0de1022da8498eac89/frozenlist-1.8.0-cp314-cp314-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:54b2077180eb7f83dd52c40b2750d0a9f175e06a42e3213ce047219de902717a", size = 242845, upload-time = "2025-10-06T05:37:17.037Z" },
+    { url = "https://files.pythonhosted.org/packages/62/c3/f57a5c8c70cd1ead3d5d5f776f89d33110b1addae0ab010ad774d9a44fb9/frozenlist-1.8.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:2f05983daecab868a31e1da44462873306d3cbfd76d1f0b5b69c473d21dbb128", size = 229131, upload-time = "2025-10-06T05:37:18.221Z" },
+    { url = "https://files.pythonhosted.org/packages/6c/52/232476fe9cb64f0742f3fde2b7d26c1dac18b6d62071c74d4ded55e0ef94/frozenlist-1.8.0-cp314-cp314-musllinux_1_2_armv7l.whl", hash = "sha256:33f48f51a446114bc5d251fb2954ab0164d5be02ad3382abcbfe07e2531d650f", size = 240542, upload-time = "2025-10-06T05:37:19.771Z" },
+    { url = "https://files.pythonhosted.org/packages/5f/85/07bf3f5d0fb5414aee5f47d33c6f5c77bfe49aac680bfece33d4fdf6a246/frozenlist-1.8.0-cp314-cp314-musllinux_1_2_ppc64le.whl", hash = "sha256:154e55ec0655291b5dd1b8731c637ecdb50975a2ae70c606d100750a540082f7", size = 237308, upload-time = "2025-10-06T05:37:20.969Z" },
+    { url = "https://files.pythonhosted.org/packages/11/99/ae3a33d5befd41ac0ca2cc7fd3aa707c9c324de2e89db0e0f45db9a64c26/frozenlist-1.8.0-cp314-cp314-musllinux_1_2_s390x.whl", hash = "sha256:4314debad13beb564b708b4a496020e5306c7333fa9a3ab90374169a20ffab30", size = 238210, upload-time = "2025-10-06T05:37:22.252Z" },
+    { url = "https://files.pythonhosted.org/packages/b2/60/b1d2da22f4970e7a155f0adde9b1435712ece01b3cd45ba63702aea33938/frozenlist-1.8.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:073f8bf8becba60aa931eb3bc420b217bb7d5b8f4750e6f8b3be7f3da85d38b7", size = 231972, upload-time = "2025-10-06T05:37:23.5Z" },
+    { url = "https://files.pythonhosted.org/packages/3f/ab/945b2f32de889993b9c9133216c068b7fcf257d8595a0ac420ac8677cab0/frozenlist-1.8.0-cp314-cp314-win32.whl", hash = "sha256:bac9c42ba2ac65ddc115d930c78d24ab8d4f465fd3fc473cdedfccadb9429806", size = 40536, upload-time = "2025-10-06T05:37:25.581Z" },
+    { url = "https://files.pythonhosted.org/packages/59/ad/9caa9b9c836d9ad6f067157a531ac48b7d36499f5036d4141ce78c230b1b/frozenlist-1.8.0-cp314-cp314-win_amd64.whl", hash = "sha256:3e0761f4d1a44f1d1a47996511752cf3dcec5bbdd9cc2b4fe595caf97754b7a0", size = 44330, upload-time = "2025-10-06T05:37:26.928Z" },
+    { url = "https://files.pythonhosted.org/packages/82/13/e6950121764f2676f43534c555249f57030150260aee9dcf7d64efda11dd/frozenlist-1.8.0-cp314-cp314-win_arm64.whl", hash = "sha256:d1eaff1d00c7751b7c6662e9c5ba6eb2c17a2306ba5e2a37f24ddf3cc953402b", size = 40627, upload-time = "2025-10-06T05:37:28.075Z" },
+    { url = "https://files.pythonhosted.org/packages/c0/c7/43200656ecc4e02d3f8bc248df68256cd9572b3f0017f0a0c4e93440ae23/frozenlist-1.8.0-cp314-cp314t-macosx_10_13_universal2.whl", hash = "sha256:d3bb933317c52d7ea5004a1c442eef86f426886fba134ef8cf4226ea6ee1821d", size = 89238, upload-time = "2025-10-06T05:37:29.373Z" },
+    { url = "https://files.pythonhosted.org/packages/d1/29/55c5f0689b9c0fb765055629f472c0de484dcaf0acee2f7707266ae3583c/frozenlist-1.8.0-cp314-cp314t-macosx_10_13_x86_64.whl", hash = "sha256:8009897cdef112072f93a0efdce29cd819e717fd2f649ee3016efd3cd885a7ed", size = 50738, upload-time = "2025-10-06T05:37:30.792Z" },
+    { url = "https://files.pythonhosted.org/packages/ba/7d/b7282a445956506fa11da8c2db7d276adcbf2b17d8bb8407a47685263f90/frozenlist-1.8.0-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:2c5dcbbc55383e5883246d11fd179782a9d07a986c40f49abe89ddf865913930", size = 51739, upload-time = "2025-10-06T05:37:32.127Z" },
+    { url = "https://files.pythonhosted.org/packages/62/1c/3d8622e60d0b767a5510d1d3cf21065b9db874696a51ea6d7a43180a259c/frozenlist-1.8.0-cp314-cp314t-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl", hash = "sha256:39ecbc32f1390387d2aa4f5a995e465e9e2f79ba3adcac92d68e3e0afae6657c", size = 284186, upload-time = "2025-10-06T05:37:33.21Z" },
+    { url = "https://files.pythonhosted.org/packages/2d/14/aa36d5f85a89679a85a1d44cd7a6657e0b1c75f61e7cad987b203d2daca8/frozenlist-1.8.0-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:92db2bf818d5cc8d9c1f1fc56b897662e24ea5adb36ad1f1d82875bd64e03c24", size = 292196, upload-time = "2025-10-06T05:37:36.107Z" },
+    { url = "https://files.pythonhosted.org/packages/05/23/6bde59eb55abd407d34f77d39a5126fb7b4f109a3f611d3929f14b700c66/frozenlist-1.8.0-cp314-cp314t-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:2dc43a022e555de94c3b68a4ef0b11c4f747d12c024a520c7101709a2144fb37", size = 273830, upload-time = "2025-10-06T05:37:37.663Z" },
+    { url = "https://files.pythonhosted.org/packages/d2/3f/22cff331bfad7a8afa616289000ba793347fcd7bc275f3b28ecea2a27909/frozenlist-1.8.0-cp314-cp314t-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:cb89a7f2de3602cfed448095bab3f178399646ab7c61454315089787df07733a", size = 294289, upload-time = "2025-10-06T05:37:39.261Z" },
+    { url = "https://files.pythonhosted.org/packages/a4/89/5b057c799de4838b6c69aa82b79705f2027615e01be996d2486a69ca99c4/frozenlist-1.8.0-cp314-cp314t-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:33139dc858c580ea50e7e60a1b0ea003efa1fd42e6ec7fdbad78fff65fad2fd2", size = 300318, upload-time = "2025-10-06T05:37:43.213Z" },
+    { url = "https://files.pythonhosted.org/packages/30/de/2c22ab3eb2a8af6d69dc799e48455813bab3690c760de58e1bf43b36da3e/frozenlist-1.8.0-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:168c0969a329b416119507ba30b9ea13688fafffac1b7822802537569a1cb0ef", size = 282814, upload-time = "2025-10-06T05:37:45.337Z" },
+    { url = "https://files.pythonhosted.org/packages/59/f7/970141a6a8dbd7f556d94977858cfb36fa9b66e0892c6dd780d2219d8cd8/frozenlist-1.8.0-cp314-cp314t-musllinux_1_2_armv7l.whl", hash = "sha256:28bd570e8e189d7f7b001966435f9dac6718324b5be2990ac496cf1ea9ddb7fe", size = 291762, upload-time = "2025-10-06T05:37:46.657Z" },
+    { url = "https://files.pythonhosted.org/packages/c1/15/ca1adae83a719f82df9116d66f5bb28bb95557b3951903d39135620ef157/frozenlist-1.8.0-cp314-cp314t-musllinux_1_2_ppc64le.whl", hash = "sha256:b2a095d45c5d46e5e79ba1e5b9cb787f541a8dee0433836cea4b96a2c439dcd8", size = 289470, upload-time = "2025-10-06T05:37:47.946Z" },
+    { url = "https://files.pythonhosted.org/packages/ac/83/dca6dc53bf657d371fbc88ddeb21b79891e747189c5de990b9dfff2ccba1/frozenlist-1.8.0-cp314-cp314t-musllinux_1_2_s390x.whl", hash = "sha256:eab8145831a0d56ec9c4139b6c3e594c7a83c2c8be25d5bcf2d86136a532287a", size = 289042, upload-time = "2025-10-06T05:37:49.499Z" },
+    { url = "https://files.pythonhosted.org/packages/96/52/abddd34ca99be142f354398700536c5bd315880ed0a213812bc491cff5e4/frozenlist-1.8.0-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:974b28cf63cc99dfb2188d8d222bc6843656188164848c4f679e63dae4b0708e", size = 283148, upload-time = "2025-10-06T05:37:50.745Z" },
+    { url = "https://files.pythonhosted.org/packages/af/d3/76bd4ed4317e7119c2b7f57c3f6934aba26d277acc6309f873341640e21f/frozenlist-1.8.0-cp314-cp314t-win32.whl", hash = "sha256:342c97bf697ac5480c0a7ec73cd700ecfa5a8a40ac923bd035484616efecc2df", size = 44676, upload-time = "2025-10-06T05:37:52.222Z" },
+    { url = "https://files.pythonhosted.org/packages/89/76/c615883b7b521ead2944bb3480398cbb07e12b7b4e4d073d3752eb721558/frozenlist-1.8.0-cp314-cp314t-win_amd64.whl", hash = "sha256:06be8f67f39c8b1dc671f5d83aaefd3358ae5cdcf8314552c57e7ed3e6475bdd", size = 49451, upload-time = "2025-10-06T05:37:53.425Z" },
+    { url = "https://files.pythonhosted.org/packages/e0/a3/5982da14e113d07b325230f95060e2169f5311b1017ea8af2a29b374c289/frozenlist-1.8.0-cp314-cp314t-win_arm64.whl", hash = "sha256:102e6314ca4da683dca92e3b1355490fed5f313b768500084fbe6371fddfdb79", size = 42507, upload-time = "2025-10-06T05:37:54.513Z" },
+    { url = "https://files.pythonhosted.org/packages/9a/9a/e35b4a917281c0b8419d4207f4334c8e8c5dbf4f3f5f9ada73958d937dcc/frozenlist-1.8.0-py3-none-any.whl", hash = "sha256:0c18a16eab41e82c295618a77502e17b195883241c563b00f0aa5106fc4eaa0d", size = 13409, upload-time = "2025-10-06T05:38:16.721Z" },
+]
+
+[[package]]
+name = "fsspec"
+version = "2026.1.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/d5/7d/5df2650c57d47c57232af5ef4b4fdbff182070421e405e0d62c6cdbfaa87/fsspec-2026.1.0.tar.gz", hash = "sha256:e987cb0496a0d81bba3a9d1cee62922fb395e7d4c3b575e57f547953334fe07b", size = 310496, upload-time = "2026-01-09T15:21:35.562Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/01/c9/97cc5aae1648dcb851958a3ddf73ccd7dbe5650d95203ecb4d7720b4cdbf/fsspec-2026.1.0-py3-none-any.whl", hash = "sha256:cb76aa913c2285a3b49bdd5fc55b1d7c708d7208126b60f2eb8194fe1b4cbdcc", size = 201838, upload-time = "2026-01-09T15:21:34.041Z" },
+]
+
+[[package]]
+name = "future"
+version = "1.0.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/a7/b2/4140c69c6a66432916b26158687e821ba631a4c9273c474343badf84d3ba/future-1.0.0.tar.gz", hash = "sha256:bd2968309307861edae1458a4f8a4f3598c03be43b97521076aebf5d94c07b05", size = 1228490, upload-time = "2024-02-21T11:52:38.461Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/da/71/ae30dadffc90b9006d77af76b393cb9dfbfc9629f339fc1574a1c52e6806/future-1.0.0-py3-none-any.whl", hash = "sha256:929292d34f5872e70396626ef385ec22355a1fae8ad29e1a734c3e43f9fbc216", size = 491326, upload-time = "2024-02-21T11:52:35.956Z" },
+]
+
+[[package]]
+name = "h11"
+version = "0.16.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/01/ee/02a2c011bdab74c6fb3c75474d40b3052059d95df7e73351460c8588d963/h11-0.16.0.tar.gz", hash = "sha256:4e35b956cf45792e4caa5885e69fba00bdbc6ffafbfa020300e549b208ee5ff1", size = 101250, upload-time = "2025-04-24T03:35:25.427Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/04/4b/29cac41a4d98d144bf5f6d33995617b185d14b22401f75ca86f384e87ff1/h11-0.16.0-py3-none-any.whl", hash = "sha256:63cf8bbe7522de3bf65932fda1d9c2772064ffb3dae62d55932da54b31cb6c86", size = 37515, upload-time = "2025-04-24T03:35:24.344Z" },
+]
+
+[[package]]
+name = "hf-xet"
+version = "1.2.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/5e/6e/0f11bacf08a67f7fb5ee09740f2ca54163863b07b70d579356e9222ce5d8/hf_xet-1.2.0.tar.gz", hash = "sha256:a8c27070ca547293b6890c4bf389f713f80e8c478631432962bb7f4bc0bd7d7f", size = 506020, upload-time = "2025-10-24T19:04:32.129Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/9e/a5/85ef910a0aa034a2abcfadc360ab5ac6f6bc4e9112349bd40ca97551cff0/hf_xet-1.2.0-cp313-cp313t-macosx_10_12_x86_64.whl", hash = "sha256:ceeefcd1b7aed4956ae8499e2199607765fbd1c60510752003b6cc0b8413b649", size = 2861870, upload-time = "2025-10-24T19:04:11.422Z" },
+    { url = "https://files.pythonhosted.org/packages/ea/40/e2e0a7eb9a51fe8828ba2d47fe22a7e74914ea8a0db68a18c3aa7449c767/hf_xet-1.2.0-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:b70218dd548e9840224df5638fdc94bd033552963cfa97f9170829381179c813", size = 2717584, upload-time = "2025-10-24T19:04:09.586Z" },
+    { url = "https://files.pythonhosted.org/packages/a5/7d/daf7f8bc4594fdd59a8a596f9e3886133fdc68e675292218a5e4c1b7e834/hf_xet-1.2.0-cp313-cp313t-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:7d40b18769bb9a8bc82a9ede575ce1a44c75eb80e7375a01d76259089529b5dc", size = 3315004, upload-time = "2025-10-24T19:04:00.314Z" },
+    { url = "https://files.pythonhosted.org/packages/b1/ba/45ea2f605fbf6d81c8b21e4d970b168b18a53515923010c312c06cd83164/hf_xet-1.2.0-cp313-cp313t-manylinux_2_28_aarch64.whl", hash = "sha256:cd3a6027d59cfb60177c12d6424e31f4b5ff13d8e3a1247b3a584bf8977e6df5", size = 3222636, upload-time = "2025-10-24T19:03:58.111Z" },
+    { url = "https://files.pythonhosted.org/packages/4a/1d/04513e3cab8f29ab8c109d309ddd21a2705afab9d52f2ba1151e0c14f086/hf_xet-1.2.0-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:6de1fc44f58f6dd937956c8d304d8c2dea264c80680bcfa61ca4a15e7b76780f", size = 3408448, upload-time = "2025-10-24T19:04:20.951Z" },
+    { url = "https://files.pythonhosted.org/packages/f0/7c/60a2756d7feec7387db3a1176c632357632fbe7849fce576c5559d4520c7/hf_xet-1.2.0-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:f182f264ed2acd566c514e45da9f2119110e48a87a327ca271027904c70c5832", size = 3503401, upload-time = "2025-10-24T19:04:22.549Z" },
+    { url = "https://files.pythonhosted.org/packages/4e/64/48fffbd67fb418ab07451e4ce641a70de1c40c10a13e25325e24858ebe5a/hf_xet-1.2.0-cp313-cp313t-win_amd64.whl", hash = "sha256:293a7a3787e5c95d7be1857358a9130694a9c6021de3f27fa233f37267174382", size = 2900866, upload-time = "2025-10-24T19:04:33.461Z" },
+    { url = "https://files.pythonhosted.org/packages/e2/51/f7e2caae42f80af886db414d4e9885fac959330509089f97cccb339c6b87/hf_xet-1.2.0-cp314-cp314t-macosx_10_12_x86_64.whl", hash = "sha256:10bfab528b968c70e062607f663e21e34e2bba349e8038db546646875495179e", size = 2861861, upload-time = "2025-10-24T19:04:19.01Z" },
+    { url = "https://files.pythonhosted.org/packages/6e/1d/a641a88b69994f9371bd347f1dd35e5d1e2e2460a2e350c8d5165fc62005/hf_xet-1.2.0-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:2a212e842647b02eb6a911187dc878e79c4aa0aa397e88dd3b26761676e8c1f8", size = 2717699, upload-time = "2025-10-24T19:04:17.306Z" },
+    { url = "https://files.pythonhosted.org/packages/df/e0/e5e9bba7d15f0318955f7ec3f4af13f92e773fbb368c0b8008a5acbcb12f/hf_xet-1.2.0-cp314-cp314t-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:30e06daccb3a7d4c065f34fc26c14c74f4653069bb2b194e7f18f17cbe9939c0", size = 3314885, upload-time = "2025-10-24T19:04:07.642Z" },
+    { url = "https://files.pythonhosted.org/packages/21/90/b7fe5ff6f2b7b8cbdf1bd56145f863c90a5807d9758a549bf3d916aa4dec/hf_xet-1.2.0-cp314-cp314t-manylinux_2_28_aarch64.whl", hash = "sha256:29c8fc913a529ec0a91867ce3d119ac1aac966e098cf49501800c870328cc090", size = 3221550, upload-time = "2025-10-24T19:04:05.55Z" },
+    { url = "https://files.pythonhosted.org/packages/6f/cb/73f276f0a7ce46cc6a6ec7d6c7d61cbfe5f2e107123d9bbd0193c355f106/hf_xet-1.2.0-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:66e159cbfcfbb29f920db2c09ed8b660eb894640d284f102ada929b6e3dc410a", size = 3408010, upload-time = "2025-10-24T19:04:28.598Z" },
+    { url = "https://files.pythonhosted.org/packages/b8/1e/d642a12caa78171f4be64f7cd9c40e3ca5279d055d0873188a58c0f5fbb9/hf_xet-1.2.0-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:9c91d5ae931510107f148874e9e2de8a16052b6f1b3ca3c1b12f15ccb491390f", size = 3503264, upload-time = "2025-10-24T19:04:30.397Z" },
+    { url = "https://files.pythonhosted.org/packages/17/b5/33764714923fa1ff922770f7ed18c2daae034d21ae6e10dbf4347c854154/hf_xet-1.2.0-cp314-cp314t-win_amd64.whl", hash = "sha256:210d577732b519ac6ede149d2f2f34049d44e8622bf14eb3d63bbcd2d4b332dc", size = 2901071, upload-time = "2025-10-24T19:04:37.463Z" },
+    { url = "https://files.pythonhosted.org/packages/96/2d/22338486473df5923a9ab7107d375dbef9173c338ebef5098ef593d2b560/hf_xet-1.2.0-cp37-abi3-macosx_10_12_x86_64.whl", hash = "sha256:46740d4ac024a7ca9b22bebf77460ff43332868b661186a8e46c227fdae01848", size = 2866099, upload-time = "2025-10-24T19:04:15.366Z" },
+    { url = "https://files.pythonhosted.org/packages/7f/8c/c5becfa53234299bc2210ba314eaaae36c2875e0045809b82e40a9544f0c/hf_xet-1.2.0-cp37-abi3-macosx_11_0_arm64.whl", hash = "sha256:27df617a076420d8845bea087f59303da8be17ed7ec0cd7ee3b9b9f579dff0e4", size = 2722178, upload-time = "2025-10-24T19:04:13.695Z" },
+    { url = "https://files.pythonhosted.org/packages/9a/92/cf3ab0b652b082e66876d08da57fcc6fa2f0e6c70dfbbafbd470bb73eb47/hf_xet-1.2.0-cp37-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:3651fd5bfe0281951b988c0facbe726aa5e347b103a675f49a3fa8144c7968fd", size = 3320214, upload-time = "2025-10-24T19:04:03.596Z" },
+    { url = "https://files.pythonhosted.org/packages/46/92/3f7ec4a1b6a65bf45b059b6d4a5d38988f63e193056de2f420137e3c3244/hf_xet-1.2.0-cp37-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:d06fa97c8562fb3ee7a378dd9b51e343bc5bc8190254202c9771029152f5e08c", size = 3229054, upload-time = "2025-10-24T19:04:01.949Z" },
+    { url = "https://files.pythonhosted.org/packages/0b/dd/7ac658d54b9fb7999a0ccb07ad863b413cbaf5cf172f48ebcd9497ec7263/hf_xet-1.2.0-cp37-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:4c1428c9ae73ec0939410ec73023c4f842927f39db09b063b9482dac5a3bb737", size = 3413812, upload-time = "2025-10-24T19:04:24.585Z" },
+    { url = "https://files.pythonhosted.org/packages/92/68/89ac4e5b12a9ff6286a12174c8538a5930e2ed662091dd2572bbe0a18c8a/hf_xet-1.2.0-cp37-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:a55558084c16b09b5ed32ab9ed38421e2d87cf3f1f89815764d1177081b99865", size = 3508920, upload-time = "2025-10-24T19:04:26.927Z" },
+    { url = "https://files.pythonhosted.org/packages/cb/44/870d44b30e1dcfb6a65932e3e1506c103a8a5aea9103c337e7a53180322c/hf_xet-1.2.0-cp37-abi3-win_amd64.whl", hash = "sha256:e6584a52253f72c9f52f9e549d5895ca7a471608495c4ecaa6cc73dba2b24d69", size = 2905735, upload-time = "2025-10-24T19:04:35.928Z" },
+]
+
+[[package]]
+name = "httpcore"
+version = "1.0.9"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "certifi" },
+    { name = "h11" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/06/94/82699a10bca87a5556c9c59b5963f2d039dbd239f25bc2a63907a05a14cb/httpcore-1.0.9.tar.gz", hash = "sha256:6e34463af53fd2ab5d807f399a9b45ea31c3dfa2276f15a2c3f00afff6e176e8", size = 85484, upload-time = "2025-04-24T22:06:22.219Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/7e/f5/f66802a942d491edb555dd61e3a9961140fd64c90bce1eafd741609d334d/httpcore-1.0.9-py3-none-any.whl", hash = "sha256:2d400746a40668fc9dec9810239072b40b4484b640a8c38fd654a024c7a1bf55", size = 78784, upload-time = "2025-04-24T22:06:20.566Z" },
+]
+
+[[package]]
+name = "httpx"
+version = "0.28.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "anyio" },
+    { name = "certifi" },
+    { name = "httpcore" },
+    { name = "idna" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/b1/df/48c586a5fe32a0f01324ee087459e112ebb7224f646c0b5023f5e79e9956/httpx-0.28.1.tar.gz", hash = "sha256:75e98c5f16b0f35b567856f597f06ff2270a374470a5c2392242528e3e3e42fc", size = 141406, upload-time = "2024-12-06T15:37:23.222Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/2a/39/e50c7c3a983047577ee07d2a9e53faf5a69493943ec3f6a384bdc792deb2/httpx-0.28.1-py3-none-any.whl", hash = "sha256:d909fcccc110f8c7faf814ca82a9a4d816bc5a6dbfea25d6591d6985b8ba59ad", size = 73517, upload-time = "2024-12-06T15:37:21.509Z" },
+]
+
+[[package]]
+name = "huggingface-hub"
+version = "1.3.7"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "filelock" },
+    { name = "fsspec" },
+    { name = "hf-xet", marker = "platform_machine == 'AMD64' or platform_machine == 'aarch64' or platform_machine == 'amd64' or platform_machine == 'arm64' or platform_machine == 'x86_64'" },
+    { name = "httpx" },
+    { name = "packaging" },
+    { name = "pyyaml" },
+    { name = "shellingham" },
+    { name = "tqdm" },
+    { name = "typer-slim" },
+    { name = "typing-extensions" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/6d/3f/352efd52136bfd8aa9280c6d4a445869226ae2ccd49ddad4f62e90cfd168/huggingface_hub-1.3.7.tar.gz", hash = "sha256:5f86cd48f27131cdbf2882699cbdf7a67dd4cbe89a81edfdc31211f42e4a5fd1", size = 627537, upload-time = "2026-02-02T10:40:10.61Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/54/89/bfbfde252d649fae8d5f09b14a2870e5672ed160c1a6629301b3e5302621/huggingface_hub-1.3.7-py3-none-any.whl", hash = "sha256:8155ce937038fa3d0cb4347d752708079bc85e6d9eb441afb44c84bcf48620d2", size = 536728, upload-time = "2026-02-02T10:40:08.274Z" },
+]
+
+[[package]]
+name = "idna"
+version = "3.11"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/6f/6d/0703ccc57f3a7233505399edb88de3cbd678da106337b9fcde432b65ed60/idna-3.11.tar.gz", hash = "sha256:795dafcc9c04ed0c1fb032c2aa73654d8e8c5023a7df64a53f39190ada629902", size = 194582, upload-time = "2025-10-12T14:55:20.501Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/0e/61/66938bbb5fc52dbdf84594873d5b51fb1f7c7794e9c0f5bd885f30bc507b/idna-3.11-py3-none-any.whl", hash = "sha256:771a87f49d9defaf64091e6e6fe9c18d4833f140bd19464795bc32d966ca37ea", size = 71008, upload-time = "2025-10-12T14:55:18.883Z" },
+]
+
+[[package]]
+name = "imagesize"
+version = "1.4.1"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/a7/84/62473fb57d61e31fef6e36d64a179c8781605429fd927b5dd608c997be31/imagesize-1.4.1.tar.gz", hash = "sha256:69150444affb9cb0d5cc5a92b3676f0b2fb7cd9ae39e947a5e11a36b4497cd4a", size = 1280026, upload-time = "2022-07-01T12:21:05.687Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/ff/62/85c4c919272577931d407be5ba5d71c20f0b616d31a0befe0ae45bb79abd/imagesize-1.4.1-py2.py3-none-any.whl", hash = "sha256:0d8d18d08f840c19d0ee7ca1fd82490fdc3729b7ac93f49870406ddde8ef8d8b", size = 8769, upload-time = "2022-07-01T12:21:02.467Z" },
+]
+
+[[package]]
+name = "iniconfig"
+version = "2.3.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/72/34/14ca021ce8e5dfedc35312d08ba8bf51fdd999c576889fc2c24cb97f4f10/iniconfig-2.3.0.tar.gz", hash = "sha256:c76315c77db068650d49c5b56314774a7804df16fee4402c1f19d6d15d8c4730", size = 20503, upload-time = "2025-10-18T21:55:43.219Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/cb/b1/3846dd7f199d53cb17f49cba7e651e9ce294d8497c8c150530ed11865bb8/iniconfig-2.3.0-py3-none-any.whl", hash = "sha256:f631c04d2c48c52b84d0d0549c99ff3859c98df65b3101406327ecc7d53fbf12", size = 7484, upload-time = "2025-10-18T21:55:41.639Z" },
+]
+
+[[package]]
+name = "jsonpatch"
+version = "1.33"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "jsonpointer" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/42/78/18813351fe5d63acad16aec57f94ec2b70a09e53ca98145589e185423873/jsonpatch-1.33.tar.gz", hash = "sha256:9fcd4009c41e6d12348b4a0ff2563ba56a2923a7dfee731d004e212e1ee5030c", size = 21699, upload-time = "2023-06-26T12:07:29.144Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/73/07/02e16ed01e04a374e644b575638ec7987ae846d25ad97bcc9945a3ee4b0e/jsonpatch-1.33-py2.py3-none-any.whl", hash = "sha256:0ae28c0cd062bbd8b8ecc26d7d164fbbea9652a1a3693f3b956c1eae5145dade", size = 12898, upload-time = "2023-06-16T21:01:28.466Z" },
+]
+
+[[package]]
+name = "jsonpointer"
+version = "3.0.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/6a/0a/eebeb1fa92507ea94016a2a790b93c2ae41a7e18778f85471dc54475ed25/jsonpointer-3.0.0.tar.gz", hash = "sha256:2b2d729f2091522d61c3b31f82e11870f60b68f43fbc705cb76bf4b832af59ef", size = 9114, upload-time = "2024-06-10T19:24:42.462Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/71/92/5e77f98553e9e75130c78900d000368476aed74276eb8ae8796f65f00918/jsonpointer-3.0.0-py2.py3-none-any.whl", hash = "sha256:13e088adc14fca8b6aa8177c044e12701e6ad4b28ff10e65f2267a90109c9942", size = 7595, upload-time = "2024-06-10T19:24:40.698Z" },
+]
+
+[[package]]
+name = "langchain-core"
+version = "1.2.7"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "jsonpatch" },
+    { name = "langsmith" },
+    { name = "packaging" },
+    { name = "pydantic" },
+    { name = "pyyaml" },
+    { name = "tenacity" },
+    { name = "typing-extensions" },
+    { name = "uuid-utils" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/a2/0e/664d8d81b3493e09cbab72448d2f9d693d1fa5aa2bcc488602203a9b6da0/langchain_core-1.2.7.tar.gz", hash = "sha256:e1460639f96c352b4a41c375f25aeb8d16ffc1769499fb1c20503aad59305ced", size = 837039, upload-time = "2026-01-09T17:44:25.505Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/6e/6f/34a9fba14d191a67f7e2ee3dbce3e9b86d2fa7310e2c7f2c713583481bd2/langchain_core-1.2.7-py3-none-any.whl", hash = "sha256:452f4fef7a3d883357b22600788d37e3d8854ef29da345b7ac7099f33c31828b", size = 490232, upload-time = "2026-01-09T17:44:24.236Z" },
+]
+
+[[package]]
+name = "langchain-paddleocr"
+version = "0.2.1"
+source = { editable = "." }
+dependencies = [
+    { name = "langchain-core" },
+    { name = "paddleocr" },
+    { name = "paddlex", extra = ["serving"] },
+    { name = "pydantic" },
+    { name = "requests" },
+]
+
+[package.dev-dependencies]
+lint = [
+    { name = "ruff" },
+]
+test = [
+    { name = "freezegun" },
+    { name = "langchain-tests" },
+    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
+    { name = "numpy", version = "2.4.2", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
+    { name = "protobuf" },
+    { name = "pytest" },
+    { name = "pytest-asyncio" },
+    { name = "pytest-mock" },
+    { name = "pytest-retry" },
+    { name = "pytest-socket" },
+    { name = "pytest-watcher" },
+    { name = "syrupy" },
+]
+test-integration = [
+    { name = "pytest" },
+]
+typing = [
+    { name = "mypy" },
+    { name = "types-requests" },
+]
+
+[package.metadata]
+requires-dist = [
+    { name = "langchain-core", specifier = ">=1.2.5,<2.0.0" },
+    { name = "paddleocr", specifier = ">=3.4.0,<3.5.0" },
+    { name = "paddlex", extras = ["serving"], specifier = ">=3.4.1,<3.5.0" },
+    { name = "pydantic", specifier = ">=2.0.0,<3.0.0" },
+    { name = "requests", specifier = ">=2.32.5" },
+]
+
+[package.metadata.requires-dev]
+lint = [{ name = "ruff", specifier = ">=0.13.1,<0.14.0" }]
+test = [
+    { name = "freezegun", specifier = ">=1.5.0,<2.0.0" },
+    { name = "langchain-tests", specifier = ">=1.1.2,<2.0.0" },
+    { name = "numpy", marker = "python_full_version < '3.13'", specifier = ">=1.26.4" },
+    { name = "numpy", marker = "python_full_version >= '3.13'", specifier = ">=2.1.0" },
+    { name = "protobuf", specifier = ">=4.25.0,<7.0.0" },
+    { name = "pytest", specifier = ">=8.4.0,<9.0.0" },
+    { name = "pytest-asyncio", specifier = ">=0.21.0,<1.0.0" },
+    { name = "pytest-mock", specifier = ">=3.14.0,<4.0.0" },
+    { name = "pytest-retry", specifier = ">=1.7.0,<2.0.0" },
+    { name = "pytest-socket", specifier = ">=0.7.0,<1.0.0" },
+    { name = "pytest-watcher", specifier = ">=0.4.0,<1.0.0" },
+    { name = "syrupy", specifier = ">=4.9.0,<5.0.0" },
+]
+test-integration = [{ name = "pytest", specifier = ">=8.4.0,<9.0.0" }]
+typing = [
+    { name = "mypy", specifier = ">=1.18.1,<1.19.0" },
+    { name = "types-requests", specifier = ">=2.31.0,<3.0.0" },
+]
+
+[[package]]
+name = "langchain-tests"
+version = "1.1.3"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "httpx" },
+    { name = "langchain-core" },
+    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
+    { name = "numpy", version = "2.4.2", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
+    { name = "pytest" },
+    { name = "pytest-asyncio" },
+    { name = "pytest-benchmark" },
+    { name = "pytest-codspeed" },
+    { name = "pytest-recording" },
+    { name = "pytest-socket" },
+    { name = "syrupy" },
+    { name = "vcrpy" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/23/6b/0168122d0fc7e8f62af910a076d87769311fff270b47d0b4a8a27fadfeca/langchain_tests-1.1.3.tar.gz", hash = "sha256:4afd15984783a718e114ac99b66a8ef822628d8ac372a69a83fa3b539c8df8cc", size = 150811, upload-time = "2026-01-31T21:41:40.059Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/86/a6/37c4cbb718f67932f8dca77d86b77d068281756440d9da6ed087a93fcd7b/langchain_tests-1.1.3-py3-none-any.whl", hash = "sha256:155ff912a611c28d313e00f1f0ad0099c0689d036797580bd89a2c6a11815cb5", size = 52511, upload-time = "2026-01-31T21:41:39.109Z" },
+]
+
+[[package]]
+name = "langsmith"
+version = "0.6.7"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "httpx" },
+    { name = "orjson", marker = "platform_python_implementation != 'PyPy'" },
+    { name = "packaging" },
+    { name = "pydantic" },
+    { name = "requests" },
+    { name = "requests-toolbelt" },
+    { name = "uuid-utils" },
+    { name = "xxhash" },
+    { name = "zstandard" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/66/bb/b8a196c9b9a7ca8b8845eaec7dbae35bcfcb0da3068794e76b29211eae2b/langsmith-0.6.7.tar.gz", hash = "sha256:d89c604a18fc606b7835d8e7924f7cdbe130ca2207bdff8f989590e50d65b802", size = 963940, upload-time = "2026-01-31T02:10:34.069Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/ec/9a/5bc17ea3c746363e73c11df5e89068fea1ed175ca9c00fdb6886efc35b21/langsmith-0.6.7-py3-none-any.whl", hash = "sha256:4bd4372b8bf724b86314f64644562b5598407614e04e74b536c09490d153bd61", size = 309369, upload-time = "2026-01-31T02:10:32.6Z" },
+]
+
+[[package]]
+name = "markdown-it-py"
+version = "4.0.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "mdurl" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/5b/f5/4ec618ed16cc4f8fb3b701563655a69816155e79e24a17b651541804721d/markdown_it_py-4.0.0.tar.gz", hash = "sha256:cb0a2b4aa34f932c007117b194e945bd74e0ec24133ceb5bac59009cda1cb9f3", size = 73070, upload-time = "2025-08-11T12:57:52.854Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/94/54/e7d793b573f298e1c9013b8c4dade17d481164aa517d1d7148619c2cedbf/markdown_it_py-4.0.0-py3-none-any.whl", hash = "sha256:87327c59b172c5011896038353a81343b6754500a08cd7a4973bb48c6d578147", size = 87321, upload-time = "2025-08-11T12:57:51.923Z" },
+]
+
+[[package]]
+name = "mdurl"
+version = "0.1.2"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/d6/54/cfe61301667036ec958cb99bd3efefba235e65cdeb9c84d24a8293ba1d90/mdurl-0.1.2.tar.gz", hash = "sha256:bb413d29f5eea38f31dd4754dd7377d4465116fb207585f97bf925588687c1ba", size = 8729, upload-time = "2022-08-14T12:40:10.846Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/b3/38/89ba8ad64ae25be8de66a6d463314cf1eb366222074cfda9ee839c56a4b4/mdurl-0.1.2-py3-none-any.whl", hash = "sha256:84008a41e51615a49fc9966191ff91509e3c40b939176e643fd50a5c2196b8f8", size = 9979, upload-time = "2022-08-14T12:40:09.779Z" },
+]
+
+[[package]]
+name = "modelscope"
+version = "1.34.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "filelock" },
+    { name = "requests" },
+    { name = "setuptools" },
+    { name = "tqdm" },
+    { name = "urllib3" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/f4/89/7a13bf70090e631f81a943797e875272d19d47e0a6984eda0a5a3f04e7e3/modelscope-1.34.0.tar.gz", hash = "sha256:c3041af301334aa9ca3f66f5b23e11ca33a2bdf28cc415dcceb75f68e4732aac", size = 4560273, upload-time = "2026-01-19T02:50:23.274Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/46/df/2c112a7c4160aa5e74dad87060019be5eca197d910af3f5b12e68ec090a9/modelscope-1.34.0-py3-none-any.whl", hash = "sha256:4629ace145972520b71b0ad02e4604282426c0cfae6a4b0922509898f3b269c8", size = 6050825, upload-time = "2026-01-19T02:50:20.018Z" },
+]
+
+[[package]]
+name = "multidict"
+version = "6.7.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "typing-extensions", marker = "python_full_version < '3.11'" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/1a/c2/c2d94cbe6ac1753f3fc980da97b3d930efe1da3af3c9f5125354436c073d/multidict-6.7.1.tar.gz", hash = "sha256:ec6652a1bee61c53a3e5776b6049172c53b6aaba34f18c9ad04f82712bac623d", size = 102010, upload-time = "2026-01-26T02:46:45.979Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/84/0b/19348d4c98980c4851d2f943f8ebafdece2ae7ef737adcfa5994ce8e5f10/multidict-6.7.1-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:c93c3db7ea657dd4637d57e74ab73de31bccefe144d3d4ce370052035bc85fb5", size = 77176, upload-time = "2026-01-26T02:42:59.784Z" },
+    { url = "https://files.pythonhosted.org/packages/ef/04/9de3f8077852e3d438215c81e9b691244532d2e05b4270e89ce67b7d103c/multidict-6.7.1-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:974e72a2474600827abaeda71af0c53d9ebbc3c2eb7da37b37d7829ae31232d8", size = 44996, upload-time = "2026-01-26T02:43:01.674Z" },
+    { url = "https://files.pythonhosted.org/packages/31/5c/08c7f7fe311f32e83f7621cd3f99d805f45519cd06fafb247628b861da7d/multidict-6.7.1-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:cdea2e7b2456cfb6694fb113066fd0ec7ea4d67e3a35e1f4cbeea0b448bf5872", size = 44631, upload-time = "2026-01-26T02:43:03.169Z" },
+    { url = "https://files.pythonhosted.org/packages/b7/7f/0e3b1390ae772f27501199996b94b52ceeb64fe6f9120a32c6c3f6b781be/multidict-6.7.1-cp310-cp310-manylinux1_i686.manylinux_2_28_i686.manylinux_2_5_i686.whl", hash = "sha256:17207077e29342fdc2c9a82e4b306f1127bf1ea91f8b71e02d4798a70bb99991", size = 242561, upload-time = "2026-01-26T02:43:04.733Z" },
+    { url = "https://files.pythonhosted.org/packages/dd/f4/8719f4f167586af317b69dd3e90f913416c91ca610cac79a45c53f590312/multidict-6.7.1-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:d4f49cb5661344764e4c7c7973e92a47a59b8fc19b6523649ec9dc4960e58a03", size = 242223, upload-time = "2026-01-26T02:43:06.695Z" },
+    { url = "https://files.pythonhosted.org/packages/47/ab/7c36164cce64a6ad19c6d9a85377b7178ecf3b89f8fd589c73381a5eedfd/multidict-6.7.1-cp310-cp310-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:a9fc4caa29e2e6ae408d1c450ac8bf19892c5fca83ee634ecd88a53332c59981", size = 222322, upload-time = "2026-01-26T02:43:08.472Z" },
+    { url = "https://files.pythonhosted.org/packages/f5/79/a25add6fb38035b5337bc5734f296d9afc99163403bbcf56d4170f97eb62/multidict-6.7.1-cp310-cp310-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:c5f0c21549ab432b57dcc82130f388d84ad8179824cc3f223d5e7cfbfd4143f6", size = 254005, upload-time = "2026-01-26T02:43:10.127Z" },
+    { url = "https://files.pythonhosted.org/packages/4a/7b/64a87cf98e12f756fc8bd444b001232ffff2be37288f018ad0d3f0aae931/multidict-6.7.1-cp310-cp310-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:7dfb78d966b2c906ae1d28ccf6e6712a3cd04407ee5088cd276fe8cb42186190", size = 251173, upload-time = "2026-01-26T02:43:11.731Z" },
+    { url = "https://files.pythonhosted.org/packages/4b/ac/b605473de2bb404e742f2cc3583d12aedb2352a70e49ae8fce455b50c5aa/multidict-6.7.1-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:9b0d9b91d1aa44db9c1f1ecd0d9d2ae610b2f4f856448664e01a3b35899f3f92", size = 243273, upload-time = "2026-01-26T02:43:13.063Z" },
+    { url = "https://files.pythonhosted.org/packages/03/65/11492d6a0e259783720f3bc1d9ea55579a76f1407e31ed44045c99542004/multidict-6.7.1-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:dd96c01a9dcd4889dcfcf9eb5544ca0c77603f239e3ffab0524ec17aea9a93ee", size = 238956, upload-time = "2026-01-26T02:43:14.843Z" },
+    { url = "https://files.pythonhosted.org/packages/5f/a7/7ee591302af64e7c196fb63fe856c788993c1372df765102bd0448e7e165/multidict-6.7.1-cp310-cp310-musllinux_1_2_armv7l.whl", hash = "sha256:067343c68cd6612d375710f895337b3a98a033c94f14b9a99eff902f205424e2", size = 233477, upload-time = "2026-01-26T02:43:16.025Z" },
+    { url = "https://files.pythonhosted.org/packages/9c/99/c109962d58756c35fd9992fed7f2355303846ea2ff054bb5f5e9d6b888de/multidict-6.7.1-cp310-cp310-musllinux_1_2_i686.whl", hash = "sha256:5884a04f4ff56c6120f6ccf703bdeb8b5079d808ba604d4d53aec0d55dc33568", size = 243615, upload-time = "2026-01-26T02:43:17.84Z" },
+    { url = "https://files.pythonhosted.org/packages/d5/5f/1973e7c771c86e93dcfe1c9cc55a5481b610f6614acfc28c0d326fe6bfad/multidict-6.7.1-cp310-cp310-musllinux_1_2_ppc64le.whl", hash = "sha256:8affcf1c98b82bc901702eb73b6947a1bfa170823c153fe8a47b5f5f02e48e40", size = 249930, upload-time = "2026-01-26T02:43:19.06Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/a5/f170fc2268c3243853580203378cd522446b2df632061e0a5409817854c7/multidict-6.7.1-cp310-cp310-musllinux_1_2_s390x.whl", hash = "sha256:0d17522c37d03e85c8098ec8431636309b2682cf12e58f4dbc76121fb50e4962", size = 243807, upload-time = "2026-01-26T02:43:20.286Z" },
+    { url = "https://files.pythonhosted.org/packages/de/01/73856fab6d125e5bc652c3986b90e8699a95e84b48d72f39ade6c0e74a8c/multidict-6.7.1-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:24c0cf81544ca5e17cfcb6e482e7a82cd475925242b308b890c9452a074d4505", size = 239103, upload-time = "2026-01-26T02:43:21.508Z" },
+    { url = "https://files.pythonhosted.org/packages/e7/46/f1220bd9944d8aa40d8ccff100eeeee19b505b857b6f603d6078cb5315b0/multidict-6.7.1-cp310-cp310-win32.whl", hash = "sha256:d82dd730a95e6643802f4454b8fdecdf08667881a9c5670db85bc5a56693f122", size = 41416, upload-time = "2026-01-26T02:43:22.703Z" },
+    { url = "https://files.pythonhosted.org/packages/68/00/9b38e272a770303692fc406c36e1a4c740f401522d5787691eb38a8925a8/multidict-6.7.1-cp310-cp310-win_amd64.whl", hash = "sha256:cf37cbe5ced48d417ba045aca1b21bafca67489452debcde94778a576666a1df", size = 46022, upload-time = "2026-01-26T02:43:23.77Z" },
+    { url = "https://files.pythonhosted.org/packages/64/65/d8d42490c02ee07b6bbe00f7190d70bb4738b3cce7629aaf9f213ef730dd/multidict-6.7.1-cp310-cp310-win_arm64.whl", hash = "sha256:59bc83d3f66b41dac1e7460aac1d196edc70c9ba3094965c467715a70ecb46db", size = 43238, upload-time = "2026-01-26T02:43:24.882Z" },
+    { url = "https://files.pythonhosted.org/packages/ce/f1/a90635c4f88fb913fbf4ce660b83b7445b7a02615bda034b2f8eb38fd597/multidict-6.7.1-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:7ff981b266af91d7b4b3793ca3382e53229088d193a85dfad6f5f4c27fc73e5d", size = 76626, upload-time = "2026-01-26T02:43:26.485Z" },
+    { url = "https://files.pythonhosted.org/packages/a6/9b/267e64eaf6fc637a15b35f5de31a566634a2740f97d8d094a69d34f524a4/multidict-6.7.1-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:844c5bca0b5444adb44a623fb0a1310c2f4cd41f402126bb269cd44c9b3f3e1e", size = 44706, upload-time = "2026-01-26T02:43:27.607Z" },
+    { url = "https://files.pythonhosted.org/packages/dd/a4/d45caf2b97b035c57267791ecfaafbd59c68212004b3842830954bb4b02e/multidict-6.7.1-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:f2a0a924d4c2e9afcd7ec64f9de35fcd96915149b2216e1cb2c10a56df483855", size = 44356, upload-time = "2026-01-26T02:43:28.661Z" },
+    { url = "https://files.pythonhosted.org/packages/fd/d2/0a36c8473f0cbaeadd5db6c8b72d15bbceeec275807772bfcd059bef487d/multidict-6.7.1-cp311-cp311-manylinux1_i686.manylinux_2_28_i686.manylinux_2_5_i686.whl", hash = "sha256:8be1802715a8e892c784c0197c2ace276ea52702a0ede98b6310c8f255a5afb3", size = 244355, upload-time = "2026-01-26T02:43:31.165Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/16/8c65be997fd7dd311b7d39c7b6e71a0cb449bad093761481eccbbe4b42a2/multidict-6.7.1-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:2e2d2ed645ea29f31c4c7ea1552fcfd7cb7ba656e1eafd4134a6620c9f5fdd9e", size = 246433, upload-time = "2026-01-26T02:43:32.581Z" },
+    { url = "https://files.pythonhosted.org/packages/01/fb/4dbd7e848d2799c6a026ec88ad39cf2b8416aa167fcc903baa55ecaa045c/multidict-6.7.1-cp311-cp311-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:95922cee9a778659e91db6497596435777bd25ed116701a4c034f8e46544955a", size = 225376, upload-time = "2026-01-26T02:43:34.417Z" },
+    { url = "https://files.pythonhosted.org/packages/b6/8a/4a3a6341eac3830f6053062f8fbc9a9e54407c80755b3f05bc427295c2d0/multidict-6.7.1-cp311-cp311-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:6b83cabdc375ffaaa15edd97eb7c0c672ad788e2687004990074d7d6c9b140c8", size = 257365, upload-time = "2026-01-26T02:43:35.741Z" },
+    { url = "https://files.pythonhosted.org/packages/f7/a2/dd575a69c1aa206e12d27d0770cdf9b92434b48a9ef0cd0d1afdecaa93c4/multidict-6.7.1-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:38fb49540705369bab8484db0689d86c0a33a0a9f2c1b197f506b71b4b6c19b0", size = 254747, upload-time = "2026-01-26T02:43:36.976Z" },
+    { url = "https://files.pythonhosted.org/packages/5a/56/21b27c560c13822ed93133f08aa6372c53a8e067f11fbed37b4adcdac922/multidict-6.7.1-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:439cbebd499f92e9aa6793016a8acaa161dfa749ae86d20960189f5398a19144", size = 246293, upload-time = "2026-01-26T02:43:38.258Z" },
+    { url = "https://files.pythonhosted.org/packages/5a/a4/23466059dc3854763423d0ad6c0f3683a379d97673b1b89ec33826e46728/multidict-6.7.1-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:6d3bc717b6fe763b8be3f2bee2701d3c8eb1b2a8ae9f60910f1b2860c82b6c49", size = 242962, upload-time = "2026-01-26T02:43:40.034Z" },
+    { url = "https://files.pythonhosted.org/packages/1f/67/51dd754a3524d685958001e8fa20a0f5f90a6a856e0a9dcabff69be3dbb7/multidict-6.7.1-cp311-cp311-musllinux_1_2_armv7l.whl", hash = "sha256:619e5a1ac57986dbfec9f0b301d865dddf763696435e2962f6d9cf2fdff2bb71", size = 237360, upload-time = "2026-01-26T02:43:41.752Z" },
+    { url = "https://files.pythonhosted.org/packages/64/3f/036dfc8c174934d4b55d86ff4f978e558b0e585cef70cfc1ad01adc6bf18/multidict-6.7.1-cp311-cp311-musllinux_1_2_i686.whl", hash = "sha256:0b38ebffd9be37c1170d33bc0f36f4f262e0a09bc1aac1c34c7aa51a7293f0b3", size = 245940, upload-time = "2026-01-26T02:43:43.042Z" },
+    { url = "https://files.pythonhosted.org/packages/3d/20/6214d3c105928ebc353a1c644a6ef1408bc5794fcb4f170bb524a3c16311/multidict-6.7.1-cp311-cp311-musllinux_1_2_ppc64le.whl", hash = "sha256:10ae39c9cfe6adedcdb764f5e8411d4a92b055e35573a2eaa88d3323289ef93c", size = 253502, upload-time = "2026-01-26T02:43:44.371Z" },
+    { url = "https://files.pythonhosted.org/packages/b1/e2/c653bc4ae1be70a0f836b82172d643fcf1dade042ba2676ab08ec08bff0f/multidict-6.7.1-cp311-cp311-musllinux_1_2_s390x.whl", hash = "sha256:25167cc263257660290fba06b9318d2026e3c910be240a146e1f66dd114af2b0", size = 247065, upload-time = "2026-01-26T02:43:45.745Z" },
+    { url = "https://files.pythonhosted.org/packages/c8/11/a854b4154cd3bd8b1fd375e8a8ca9d73be37610c361543d56f764109509b/multidict-6.7.1-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:128441d052254f42989ef98b7b6a6ecb1e6f708aa962c7984235316db59f50fa", size = 241870, upload-time = "2026-01-26T02:43:47.054Z" },
+    { url = "https://files.pythonhosted.org/packages/13/bf/9676c0392309b5fdae322333d22a829715b570edb9baa8016a517b55b558/multidict-6.7.1-cp311-cp311-win32.whl", hash = "sha256:d62b7f64ffde3b99d06b707a280db04fb3855b55f5a06df387236051d0668f4a", size = 41302, upload-time = "2026-01-26T02:43:48.753Z" },
+    { url = "https://files.pythonhosted.org/packages/c9/68/f16a3a8ba6f7b6dc92a1f19669c0810bd2c43fc5a02da13b1cbf8e253845/multidict-6.7.1-cp311-cp311-win_amd64.whl", hash = "sha256:bdbf9f3b332abd0cdb306e7c2113818ab1e922dc84b8f8fd06ec89ed2a19ab8b", size = 45981, upload-time = "2026-01-26T02:43:49.921Z" },
+    { url = "https://files.pythonhosted.org/packages/ac/ad/9dd5305253fa00cd3c7555dbef69d5bf4133debc53b87ab8d6a44d411665/multidict-6.7.1-cp311-cp311-win_arm64.whl", hash = "sha256:b8c990b037d2fff2f4e33d3f21b9b531c5745b33a49a7d6dbe7a177266af44f6", size = 43159, upload-time = "2026-01-26T02:43:51.635Z" },
+    { url = "https://files.pythonhosted.org/packages/8d/9c/f20e0e2cf80e4b2e4b1c365bf5fe104ee633c751a724246262db8f1a0b13/multidict-6.7.1-cp312-cp312-macosx_10_13_universal2.whl", hash = "sha256:a90f75c956e32891a4eda3639ce6dd86e87105271f43d43442a3aedf3cddf172", size = 76893, upload-time = "2026-01-26T02:43:52.754Z" },
+    { url = "https://files.pythonhosted.org/packages/fe/cf/18ef143a81610136d3da8193da9d80bfe1cb548a1e2d1c775f26b23d024a/multidict-6.7.1-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:3fccb473e87eaa1382689053e4a4618e7ba7b9b9b8d6adf2027ee474597128cd", size = 45456, upload-time = "2026-01-26T02:43:53.893Z" },
+    { url = "https://files.pythonhosted.org/packages/a9/65/1caac9d4cd32e8433908683446eebc953e82d22b03d10d41a5f0fefe991b/multidict-6.7.1-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:b0fa96985700739c4c7853a43c0b3e169360d6855780021bfc6d0f1ce7c123e7", size = 43872, upload-time = "2026-01-26T02:43:55.041Z" },
+    { url = "https://files.pythonhosted.org/packages/cf/3b/d6bd75dc4f3ff7c73766e04e705b00ed6dbbaccf670d9e05a12b006f5a21/multidict-6.7.1-cp312-cp312-manylinux1_i686.manylinux_2_28_i686.manylinux_2_5_i686.whl", hash = "sha256:cb2a55f408c3043e42b40cc8eecd575afa27b7e0b956dfb190de0f8499a57a53", size = 251018, upload-time = "2026-01-26T02:43:56.198Z" },
+    { url = "https://files.pythonhosted.org/packages/fd/80/c959c5933adedb9ac15152e4067c702a808ea183a8b64cf8f31af8ad3155/multidict-6.7.1-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:eb0ce7b2a32d09892b3dd6cc44877a0d02a33241fafca5f25c8b6b62374f8b75", size = 258883, upload-time = "2026-01-26T02:43:57.499Z" },
+    { url = "https://files.pythonhosted.org/packages/86/85/7ed40adafea3d4f1c8b916e3b5cc3a8e07dfcdcb9cd72800f4ed3ca1b387/multidict-6.7.1-cp312-cp312-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:c3a32d23520ee37bf327d1e1a656fec76a2edd5c038bf43eddfa0572ec49c60b", size = 242413, upload-time = "2026-01-26T02:43:58.755Z" },
+    { url = "https://files.pythonhosted.org/packages/d2/57/b8565ff533e48595503c785f8361ff9a4fde4d67de25c207cd0ba3befd03/multidict-6.7.1-cp312-cp312-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:9c90fed18bffc0189ba814749fdcc102b536e83a9f738a9003e569acd540a733", size = 268404, upload-time = "2026-01-26T02:44:00.216Z" },
+    { url = "https://files.pythonhosted.org/packages/e0/50/9810c5c29350f7258180dfdcb2e52783a0632862eb334c4896ac717cebcb/multidict-6.7.1-cp312-cp312-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:da62917e6076f512daccfbbde27f46fed1c98fee202f0559adec8ee0de67f71a", size = 269456, upload-time = "2026-01-26T02:44:02.202Z" },
+    { url = "https://files.pythonhosted.org/packages/f3/8d/5e5be3ced1d12966fefb5c4ea3b2a5b480afcea36406559442c6e31d4a48/multidict-6.7.1-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:bfde23ef6ed9db7eaee6c37dcec08524cb43903c60b285b172b6c094711b3961", size = 256322, upload-time = "2026-01-26T02:44:03.56Z" },
+    { url = "https://files.pythonhosted.org/packages/31/6e/d8a26d81ac166a5592782d208dd90dfdc0a7a218adaa52b45a672b46c122/multidict-6.7.1-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:3758692429e4e32f1ba0df23219cd0b4fc0a52f476726fff9337d1a57676a582", size = 253955, upload-time = "2026-01-26T02:44:04.845Z" },
+    { url = "https://files.pythonhosted.org/packages/59/4c/7c672c8aad41534ba619bcd4ade7a0dc87ed6b8b5c06149b85d3dd03f0cd/multidict-6.7.1-cp312-cp312-musllinux_1_2_armv7l.whl", hash = "sha256:398c1478926eca669f2fd6a5856b6de9c0acf23a2cb59a14c0ba5844fa38077e", size = 251254, upload-time = "2026-01-26T02:44:06.133Z" },
+    { url = "https://files.pythonhosted.org/packages/7b/bd/84c24de512cbafbdbc39439f74e967f19570ce7924e3007174a29c348916/multidict-6.7.1-cp312-cp312-musllinux_1_2_i686.whl", hash = "sha256:c102791b1c4f3ab36ce4101154549105a53dc828f016356b3e3bcae2e3a039d3", size = 252059, upload-time = "2026-01-26T02:44:07.518Z" },
+    { url = "https://files.pythonhosted.org/packages/fa/ba/f5449385510825b73d01c2d4087bf6d2fccc20a2d42ac34df93191d3dd03/multidict-6.7.1-cp312-cp312-musllinux_1_2_ppc64le.whl", hash = "sha256:a088b62bd733e2ad12c50dad01b7d0166c30287c166e137433d3b410add807a6", size = 263588, upload-time = "2026-01-26T02:44:09.382Z" },
+    { url = "https://files.pythonhosted.org/packages/d7/11/afc7c677f68f75c84a69fe37184f0f82fce13ce4b92f49f3db280b7e92b3/multidict-6.7.1-cp312-cp312-musllinux_1_2_s390x.whl", hash = "sha256:3d51ff4785d58d3f6c91bdbffcb5e1f7ddfda557727043aa20d20ec4f65e324a", size = 259642, upload-time = "2026-01-26T02:44:10.73Z" },
+    { url = "https://files.pythonhosted.org/packages/2b/17/ebb9644da78c4ab36403739e0e6e0e30ebb135b9caf3440825001a0bddcb/multidict-6.7.1-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:fc5907494fccf3e7d3f94f95c91d6336b092b5fc83811720fae5e2765890dfba", size = 251377, upload-time = "2026-01-26T02:44:12.042Z" },
+    { url = "https://files.pythonhosted.org/packages/ca/a4/840f5b97339e27846c46307f2530a2805d9d537d8b8bd416af031cad7fa0/multidict-6.7.1-cp312-cp312-win32.whl", hash = "sha256:28ca5ce2fd9716631133d0e9a9b9a745ad7f60bac2bccafb56aa380fc0b6c511", size = 41887, upload-time = "2026-01-26T02:44:14.245Z" },
+    { url = "https://files.pythonhosted.org/packages/80/31/0b2517913687895f5904325c2069d6a3b78f66cc641a86a2baf75a05dcbb/multidict-6.7.1-cp312-cp312-win_amd64.whl", hash = "sha256:fcee94dfbd638784645b066074b338bc9cc155d4b4bffa4adce1615c5a426c19", size = 46053, upload-time = "2026-01-26T02:44:15.371Z" },
+    { url = "https://files.pythonhosted.org/packages/0c/5b/aba28e4ee4006ae4c7df8d327d31025d760ffa992ea23812a601d226e682/multidict-6.7.1-cp312-cp312-win_arm64.whl", hash = "sha256:ba0a9fb644d0c1a2194cf7ffb043bd852cea63a57f66fbd33959f7dae18517bf", size = 43307, upload-time = "2026-01-26T02:44:16.852Z" },
+    { url = "https://files.pythonhosted.org/packages/f2/22/929c141d6c0dba87d3e1d38fbdf1ba8baba86b7776469f2bc2d3227a1e67/multidict-6.7.1-cp313-cp313-macosx_10_13_universal2.whl", hash = "sha256:2b41f5fed0ed563624f1c17630cb9941cf2309d4df00e494b551b5f3e3d67a23", size = 76174, upload-time = "2026-01-26T02:44:18.509Z" },
+    { url = "https://files.pythonhosted.org/packages/c7/75/bc704ae15fee974f8fccd871305e254754167dce5f9e42d88a2def741a1d/multidict-6.7.1-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:84e61e3af5463c19b67ced91f6c634effb89ef8bfc5ca0267f954451ed4bb6a2", size = 45116, upload-time = "2026-01-26T02:44:19.745Z" },
+    { url = "https://files.pythonhosted.org/packages/79/76/55cd7186f498ed080a18440c9013011eb548f77ae1b297206d030eb1180a/multidict-6.7.1-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:935434b9853c7c112eee7ac891bc4cb86455aa631269ae35442cb316790c1445", size = 43524, upload-time = "2026-01-26T02:44:21.571Z" },
+    { url = "https://files.pythonhosted.org/packages/e9/3c/414842ef8d5a1628d68edee29ba0e5bcf235dbfb3ccd3ea303a7fe8c72ff/multidict-6.7.1-cp313-cp313-manylinux1_i686.manylinux_2_28_i686.manylinux_2_5_i686.whl", hash = "sha256:432feb25a1cb67fe82a9680b4d65fb542e4635cb3166cd9c01560651ad60f177", size = 249368, upload-time = "2026-01-26T02:44:22.803Z" },
+    { url = "https://files.pythonhosted.org/packages/f6/32/befed7f74c458b4a525e60519fe8d87eef72bb1e99924fa2b0f9d97a221e/multidict-6.7.1-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:e82d14e3c948952a1a85503817e038cba5905a3352de76b9a465075d072fba23", size = 256952, upload-time = "2026-01-26T02:44:24.306Z" },
+    { url = "https://files.pythonhosted.org/packages/03/d6/c878a44ba877f366630c860fdf74bfb203c33778f12b6ac274936853c451/multidict-6.7.1-cp313-cp313-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:4cfb48c6ea66c83bcaaf7e4dfa7ec1b6bbcf751b7db85a328902796dfde4c060", size = 240317, upload-time = "2026-01-26T02:44:25.772Z" },
+    { url = "https://files.pythonhosted.org/packages/68/49/57421b4d7ad2e9e60e25922b08ceb37e077b90444bde6ead629095327a6f/multidict-6.7.1-cp313-cp313-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:1d540e51b7e8e170174555edecddbd5538105443754539193e3e1061864d444d", size = 267132, upload-time = "2026-01-26T02:44:27.648Z" },
+    { url = "https://files.pythonhosted.org/packages/b7/fe/ec0edd52ddbcea2a2e89e174f0206444a61440b40f39704e64dc807a70bd/multidict-6.7.1-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:273d23f4b40f3dce4d6c8a821c741a86dec62cded82e1175ba3d99be128147ed", size = 268140, upload-time = "2026-01-26T02:44:29.588Z" },
+    { url = "https://files.pythonhosted.org/packages/b0/73/6e1b01cbeb458807aa0831742232dbdd1fa92bfa33f52a3f176b4ff3dc11/multidict-6.7.1-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:9d624335fd4fa1c08a53f8b4be7676ebde19cd092b3895c421045ca87895b429", size = 254277, upload-time = "2026-01-26T02:44:30.902Z" },
+    { url = "https://files.pythonhosted.org/packages/6a/b2/5fb8c124d7561a4974c342bc8c778b471ebbeb3cc17df696f034a7e9afe7/multidict-6.7.1-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:12fad252f8b267cc75b66e8fc51b3079604e8d43a75428ffe193cd9e2195dfd6", size = 252291, upload-time = "2026-01-26T02:44:32.31Z" },
+    { url = "https://files.pythonhosted.org/packages/5a/96/51d4e4e06bcce92577fcd488e22600bd38e4fd59c20cb49434d054903bd2/multidict-6.7.1-cp313-cp313-musllinux_1_2_armv7l.whl", hash = "sha256:03ede2a6ffbe8ef936b92cb4529f27f42be7f56afcdab5ab739cd5f27fb1cbf9", size = 250156, upload-time = "2026-01-26T02:44:33.734Z" },
+    { url = "https://files.pythonhosted.org/packages/db/6b/420e173eec5fba721a50e2a9f89eda89d9c98fded1124f8d5c675f7a0c0f/multidict-6.7.1-cp313-cp313-musllinux_1_2_i686.whl", hash = "sha256:90efbcf47dbe33dcf643a1e400d67d59abeac5db07dc3f27d6bdeae497a2198c", size = 249742, upload-time = "2026-01-26T02:44:35.222Z" },
+    { url = "https://files.pythonhosted.org/packages/44/a3/ec5b5bd98f306bc2aa297b8c6f11a46714a56b1e6ef5ebda50a4f5d7c5fb/multidict-6.7.1-cp313-cp313-musllinux_1_2_ppc64le.whl", hash = "sha256:5c4b9bfc148f5a91be9244d6264c53035c8a0dcd2f51f1c3c6e30e30ebaa1c84", size = 262221, upload-time = "2026-01-26T02:44:36.604Z" },
+    { url = "https://files.pythonhosted.org/packages/cd/f7/e8c0d0da0cd1e28d10e624604e1a36bcc3353aaebdfdc3a43c72bc683a12/multidict-6.7.1-cp313-cp313-musllinux_1_2_s390x.whl", hash = "sha256:401c5a650f3add2472d1d288c26deebc540f99e2fb83e9525007a74cd2116f1d", size = 258664, upload-time = "2026-01-26T02:44:38.008Z" },
+    { url = "https://files.pythonhosted.org/packages/52/da/151a44e8016dd33feed44f730bd856a66257c1ee7aed4f44b649fb7edeb3/multidict-6.7.1-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:97891f3b1b3ffbded884e2916cacf3c6fc87b66bb0dde46f7357404750559f33", size = 249490, upload-time = "2026-01-26T02:44:39.386Z" },
+    { url = "https://files.pythonhosted.org/packages/87/af/a3b86bf9630b732897f6fc3f4c4714b90aa4361983ccbdcd6c0339b21b0c/multidict-6.7.1-cp313-cp313-win32.whl", hash = "sha256:e1c5988359516095535c4301af38d8a8838534158f649c05dd1050222321bcb3", size = 41695, upload-time = "2026-01-26T02:44:41.318Z" },
+    { url = "https://files.pythonhosted.org/packages/b2/35/e994121b0e90e46134673422dd564623f93304614f5d11886b1b3e06f503/multidict-6.7.1-cp313-cp313-win_amd64.whl", hash = "sha256:960c83bf01a95b12b08fd54324a4eb1d5b52c88932b5cba5d6e712bb3ed12eb5", size = 45884, upload-time = "2026-01-26T02:44:42.488Z" },
+    { url = "https://files.pythonhosted.org/packages/ca/61/42d3e5dbf661242a69c97ea363f2d7b46c567da8eadef8890022be6e2ab0/multidict-6.7.1-cp313-cp313-win_arm64.whl", hash = "sha256:563fe25c678aaba333d5399408f5ec3c383ca5b663e7f774dd179a520b8144df", size = 43122, upload-time = "2026-01-26T02:44:43.664Z" },
+    { url = "https://files.pythonhosted.org/packages/6d/b3/e6b21c6c4f314bb956016b0b3ef2162590a529b84cb831c257519e7fde44/multidict-6.7.1-cp313-cp313t-macosx_10_13_universal2.whl", hash = "sha256:c76c4bec1538375dad9d452d246ca5368ad6e1c9039dadcf007ae59c70619ea1", size = 83175, upload-time = "2026-01-26T02:44:44.894Z" },
+    { url = "https://files.pythonhosted.org/packages/fb/76/23ecd2abfe0957b234f6c960f4ade497f55f2c16aeb684d4ecdbf1c95791/multidict-6.7.1-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:57b46b24b5d5ebcc978da4ec23a819a9402b4228b8a90d9c656422b4bdd8a963", size = 48460, upload-time = "2026-01-26T02:44:46.106Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/57/a0ed92b23f3a042c36bc4227b72b97eca803f5f1801c1ab77c8a212d455e/multidict-6.7.1-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:e954b24433c768ce78ab7929e84ccf3422e46deb45a4dc9f93438f8217fa2d34", size = 46930, upload-time = "2026-01-26T02:44:47.278Z" },
+    { url = "https://files.pythonhosted.org/packages/b5/66/02ec7ace29162e447f6382c495dc95826bf931d3818799bbef11e8f7df1a/multidict-6.7.1-cp313-cp313t-manylinux1_i686.manylinux_2_28_i686.manylinux_2_5_i686.whl", hash = "sha256:3bd231490fa7217cc832528e1cd8752a96f0125ddd2b5749390f7c3ec8721b65", size = 242582, upload-time = "2026-01-26T02:44:48.604Z" },
+    { url = "https://files.pythonhosted.org/packages/58/18/64f5a795e7677670e872673aca234162514696274597b3708b2c0d276cce/multidict-6.7.1-cp313-cp313t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:253282d70d67885a15c8a7716f3a73edf2d635793ceda8173b9ecc21f2fb8292", size = 250031, upload-time = "2026-01-26T02:44:50.544Z" },
+    { url = "https://files.pythonhosted.org/packages/c8/ed/e192291dbbe51a8290c5686f482084d31bcd9d09af24f63358c3d42fd284/multidict-6.7.1-cp313-cp313t-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:0b4c48648d7649c9335cf1927a8b87fa692de3dcb15faa676c6a6f1f1aabda43", size = 228596, upload-time = "2026-01-26T02:44:51.951Z" },
+    { url = "https://files.pythonhosted.org/packages/1e/7e/3562a15a60cf747397e7f2180b0a11dc0c38d9175a650e75fa1b4d325e15/multidict-6.7.1-cp313-cp313t-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:98bc624954ec4d2c7cb074b8eefc2b5d0ce7d482e410df446414355d158fe4ca", size = 257492, upload-time = "2026-01-26T02:44:53.902Z" },
+    { url = "https://files.pythonhosted.org/packages/24/02/7d0f9eae92b5249bb50ac1595b295f10e263dd0078ebb55115c31e0eaccd/multidict-6.7.1-cp313-cp313t-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:1b99af4d9eec0b49927b4402bcbb58dea89d3e0db8806a4086117019939ad3dd", size = 255899, upload-time = "2026-01-26T02:44:55.316Z" },
+    { url = "https://files.pythonhosted.org/packages/00/e3/9b60ed9e23e64c73a5cde95269ef1330678e9c6e34dd4eb6b431b85b5a10/multidict-6.7.1-cp313-cp313t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:6aac4f16b472d5b7dc6f66a0d49dd57b0e0902090be16594dc9ebfd3d17c47e7", size = 247970, upload-time = "2026-01-26T02:44:56.783Z" },
+    { url = "https://files.pythonhosted.org/packages/3e/06/538e58a63ed5cfb0bd4517e346b91da32fde409d839720f664e9a4ae4f9d/multidict-6.7.1-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:21f830fe223215dffd51f538e78c172ed7c7f60c9b96a2bf05c4848ad49921c3", size = 245060, upload-time = "2026-01-26T02:44:58.195Z" },
+    { url = "https://files.pythonhosted.org/packages/b2/2f/d743a3045a97c895d401e9bd29aaa09b94f5cbdf1bd561609e5a6c431c70/multidict-6.7.1-cp313-cp313t-musllinux_1_2_armv7l.whl", hash = "sha256:f5dd81c45b05518b9aa4da4aa74e1c93d715efa234fd3e8a179df611cc85e5f4", size = 235888, upload-time = "2026-01-26T02:44:59.57Z" },
+    { url = "https://files.pythonhosted.org/packages/38/83/5a325cac191ab28b63c52f14f1131f3b0a55ba3b9aa65a6d0bf2a9b921a0/multidict-6.7.1-cp313-cp313t-musllinux_1_2_i686.whl", hash = "sha256:eb304767bca2bb92fb9c5bd33cedc95baee5bb5f6c88e63706533a1c06ad08c8", size = 243554, upload-time = "2026-01-26T02:45:01.054Z" },
+    { url = "https://files.pythonhosted.org/packages/20/1f/9d2327086bd15da2725ef6aae624208e2ef828ed99892b17f60c344e57ed/multidict-6.7.1-cp313-cp313t-musllinux_1_2_ppc64le.whl", hash = "sha256:c9035dde0f916702850ef66460bc4239d89d08df4d02023a5926e7446724212c", size = 252341, upload-time = "2026-01-26T02:45:02.484Z" },
+    { url = "https://files.pythonhosted.org/packages/e8/2c/2a1aa0280cf579d0f6eed8ee5211c4f1730bd7e06c636ba2ee6aafda302e/multidict-6.7.1-cp313-cp313t-musllinux_1_2_s390x.whl", hash = "sha256:af959b9beeb66c822380f222f0e0a1889331597e81f1ded7f374f3ecb0fd6c52", size = 246391, upload-time = "2026-01-26T02:45:03.862Z" },
+    { url = "https://files.pythonhosted.org/packages/e5/03/7ca022ffc36c5a3f6e03b179a5ceb829be9da5783e6fe395f347c0794680/multidict-6.7.1-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:41f2952231456154ee479651491e94118229844dd7226541788be783be2b5108", size = 243422, upload-time = "2026-01-26T02:45:05.296Z" },
+    { url = "https://files.pythonhosted.org/packages/dc/1d/b31650eab6c5778aceed46ba735bd97f7c7d2f54b319fa916c0f96e7805b/multidict-6.7.1-cp313-cp313t-win32.whl", hash = "sha256:df9f19c28adcb40b6aae30bbaa1478c389efd50c28d541d76760199fc1037c32", size = 47770, upload-time = "2026-01-26T02:45:06.754Z" },
+    { url = "https://files.pythonhosted.org/packages/ac/5b/2d2d1d522e51285bd61b1e20df8f47ae1a9d80839db0b24ea783b3832832/multidict-6.7.1-cp313-cp313t-win_amd64.whl", hash = "sha256:d54ecf9f301853f2c5e802da559604b3e95bb7a3b01a9c295c6ee591b9882de8", size = 53109, upload-time = "2026-01-26T02:45:08.044Z" },
+    { url = "https://files.pythonhosted.org/packages/3d/a3/cc409ba012c83ca024a308516703cf339bdc4b696195644a7215a5164a24/multidict-6.7.1-cp313-cp313t-win_arm64.whl", hash = "sha256:5a37ca18e360377cfda1d62f5f382ff41f2b8c4ccb329ed974cc2e1643440118", size = 45573, upload-time = "2026-01-26T02:45:09.349Z" },
+    { url = "https://files.pythonhosted.org/packages/91/cc/db74228a8be41884a567e88a62fd589a913708fcf180d029898c17a9a371/multidict-6.7.1-cp314-cp314-macosx_10_15_universal2.whl", hash = "sha256:8f333ec9c5eb1b7105e3b84b53141e66ca05a19a605368c55450b6ba208cb9ee", size = 75190, upload-time = "2026-01-26T02:45:10.651Z" },
+    { url = "https://files.pythonhosted.org/packages/d5/22/492f2246bb5b534abd44804292e81eeaf835388901f0c574bac4eeec73c5/multidict-6.7.1-cp314-cp314-macosx_10_15_x86_64.whl", hash = "sha256:a407f13c188f804c759fc6a9f88286a565c242a76b27626594c133b82883b5c2", size = 44486, upload-time = "2026-01-26T02:45:11.938Z" },
+    { url = "https://files.pythonhosted.org/packages/f1/4f/733c48f270565d78b4544f2baddc2fb2a245e5a8640254b12c36ac7ac68e/multidict-6.7.1-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:0e161ddf326db5577c3a4cc2d8648f81456e8a20d40415541587a71620d7a7d1", size = 43219, upload-time = "2026-01-26T02:45:14.346Z" },
+    { url = "https://files.pythonhosted.org/packages/24/bb/2c0c2287963f4259c85e8bcbba9182ced8d7fca65c780c38e99e61629d11/multidict-6.7.1-cp314-cp314-manylinux1_i686.manylinux_2_28_i686.manylinux_2_5_i686.whl", hash = "sha256:1e3a8bb24342a8201d178c3b4984c26ba81a577c80d4d525727427460a50c22d", size = 245132, upload-time = "2026-01-26T02:45:15.712Z" },
+    { url = "https://files.pythonhosted.org/packages/a7/f9/44d4b3064c65079d2467888794dea218d1601898ac50222ab8a9a8094460/multidict-6.7.1-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:97231140a50f5d447d3164f994b86a0bed7cd016e2682f8650d6a9158e14fd31", size = 252420, upload-time = "2026-01-26T02:45:17.293Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/13/78f7275e73fa17b24c9a51b0bd9d73ba64bb32d0ed51b02a746eb876abe7/multidict-6.7.1-cp314-cp314-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:6b10359683bd8806a200fd2909e7c8ca3a7b24ec1d8132e483d58e791d881048", size = 233510, upload-time = "2026-01-26T02:45:19.356Z" },
+    { url = "https://files.pythonhosted.org/packages/4b/25/8167187f62ae3cbd52da7893f58cb036b47ea3fb67138787c76800158982/multidict-6.7.1-cp314-cp314-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:283ddac99f7ac25a4acadbf004cb5ae34480bbeb063520f70ce397b281859362", size = 264094, upload-time = "2026-01-26T02:45:20.834Z" },
+    { url = "https://files.pythonhosted.org/packages/a1/e7/69a3a83b7b030cf283fb06ce074a05a02322359783424d7edf0f15fe5022/multidict-6.7.1-cp314-cp314-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:538cec1e18c067d0e6103aa9a74f9e832904c957adc260e61cd9d8cf0c3b3d37", size = 260786, upload-time = "2026-01-26T02:45:22.818Z" },
+    { url = "https://files.pythonhosted.org/packages/fe/3b/8ec5074bcfc450fe84273713b4b0a0dd47c0249358f5d82eb8104ffe2520/multidict-6.7.1-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:7eee46ccb30ff48a1e35bb818cc90846c6be2b68240e42a78599166722cea709", size = 248483, upload-time = "2026-01-26T02:45:24.368Z" },
+    { url = "https://files.pythonhosted.org/packages/48/5a/d5a99e3acbca0e29c5d9cba8f92ceb15dce78bab963b308ae692981e3a5d/multidict-6.7.1-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:fa263a02f4f2dd2d11a7b1bb4362aa7cb1049f84a9235d31adf63f30143469a0", size = 248403, upload-time = "2026-01-26T02:45:25.982Z" },
+    { url = "https://files.pythonhosted.org/packages/35/48/e58cd31f6c7d5102f2a4bf89f96b9cf7e00b6c6f3d04ecc44417c00a5a3c/multidict-6.7.1-cp314-cp314-musllinux_1_2_armv7l.whl", hash = "sha256:2e1425e2f99ec5bd36c15a01b690a1a2456209c5deed58f95469ffb46039ccbb", size = 240315, upload-time = "2026-01-26T02:45:27.487Z" },
+    { url = "https://files.pythonhosted.org/packages/94/33/1cd210229559cb90b6786c30676bb0c58249ff42f942765f88793b41fdce/multidict-6.7.1-cp314-cp314-musllinux_1_2_i686.whl", hash = "sha256:497394b3239fc6f0e13a78a3e1b61296e72bf1c5f94b4c4eb80b265c37a131cd", size = 245528, upload-time = "2026-01-26T02:45:28.991Z" },
+    { url = "https://files.pythonhosted.org/packages/64/f2/6e1107d226278c876c783056b7db43d800bb64c6131cec9c8dfb6903698e/multidict-6.7.1-cp314-cp314-musllinux_1_2_ppc64le.whl", hash = "sha256:233b398c29d3f1b9676b4b6f75c518a06fcb2ea0b925119fb2c1bc35c05e1601", size = 258784, upload-time = "2026-01-26T02:45:30.503Z" },
+    { url = "https://files.pythonhosted.org/packages/4d/c1/11f664f14d525e4a1b5327a82d4de61a1db604ab34c6603bb3c2cc63ad34/multidict-6.7.1-cp314-cp314-musllinux_1_2_s390x.whl", hash = "sha256:93b1818e4a6e0930454f0f2af7dfce69307ca03cdcfb3739bf4d91241967b6c1", size = 251980, upload-time = "2026-01-26T02:45:32.603Z" },
+    { url = "https://files.pythonhosted.org/packages/e1/9f/75a9ac888121d0c5bbd4ecf4eead45668b1766f6baabfb3b7f66a410e231/multidict-6.7.1-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:f33dc2a3abe9249ea5d8360f969ec7f4142e7ac45ee7014d8f8d5acddf178b7b", size = 243602, upload-time = "2026-01-26T02:45:34.043Z" },
+    { url = "https://files.pythonhosted.org/packages/9a/e7/50bf7b004cc8525d80dbbbedfdc7aed3e4c323810890be4413e589074032/multidict-6.7.1-cp314-cp314-win32.whl", hash = "sha256:3ab8b9d8b75aef9df299595d5388b14530839f6422333357af1339443cff777d", size = 40930, upload-time = "2026-01-26T02:45:36.278Z" },
+    { url = "https://files.pythonhosted.org/packages/e0/bf/52f25716bbe93745595800f36fb17b73711f14da59ed0bb2eba141bc9f0f/multidict-6.7.1-cp314-cp314-win_amd64.whl", hash = "sha256:5e01429a929600e7dab7b166062d9bb54a5eed752384c7384c968c2afab8f50f", size = 45074, upload-time = "2026-01-26T02:45:37.546Z" },
+    { url = "https://files.pythonhosted.org/packages/97/ab/22803b03285fa3a525f48217963da3a65ae40f6a1b6f6cf2768879e208f9/multidict-6.7.1-cp314-cp314-win_arm64.whl", hash = "sha256:4885cb0e817aef5d00a2e8451d4665c1808378dc27c2705f1bf4ef8505c0d2e5", size = 42471, upload-time = "2026-01-26T02:45:38.889Z" },
+    { url = "https://files.pythonhosted.org/packages/e0/6d/f9293baa6146ba9507e360ea0292b6422b016907c393e2f63fc40ab7b7b5/multidict-6.7.1-cp314-cp314t-macosx_10_15_universal2.whl", hash = "sha256:0458c978acd8e6ea53c81eefaddbbee9c6c5e591f41b3f5e8e194780fe026581", size = 82401, upload-time = "2026-01-26T02:45:40.254Z" },
+    { url = "https://files.pythonhosted.org/packages/7a/68/53b5494738d83558d87c3c71a486504d8373421c3e0dbb6d0db48ad42ee0/multidict-6.7.1-cp314-cp314t-macosx_10_15_x86_64.whl", hash = "sha256:c0abd12629b0af3cf590982c0b413b1e7395cd4ec026f30986818ab95bfaa94a", size = 48143, upload-time = "2026-01-26T02:45:41.635Z" },
+    { url = "https://files.pythonhosted.org/packages/37/e8/5284c53310dcdc99ce5d66563f6e5773531a9b9fe9ec7a615e9bc306b05f/multidict-6.7.1-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:14525a5f61d7d0c94b368a42cff4c9a4e7ba2d52e2672a7b23d84dc86fb02b0c", size = 46507, upload-time = "2026-01-26T02:45:42.99Z" },
+    { url = "https://files.pythonhosted.org/packages/e4/fc/6800d0e5b3875568b4083ecf5f310dcf91d86d52573160834fb4bfcf5e4f/multidict-6.7.1-cp314-cp314t-manylinux1_i686.manylinux_2_28_i686.manylinux_2_5_i686.whl", hash = "sha256:17307b22c217b4cf05033dabefe68255a534d637c6c9b0cc8382718f87be4262", size = 239358, upload-time = "2026-01-26T02:45:44.376Z" },
+    { url = "https://files.pythonhosted.org/packages/41/75/4ad0973179361cdf3a113905e6e088173198349131be2b390f9fa4da5fc6/multidict-6.7.1-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:7a7e590ff876a3eaf1c02a4dfe0724b6e69a9e9de6d8f556816f29c496046e59", size = 246884, upload-time = "2026-01-26T02:45:47.167Z" },
+    { url = "https://files.pythonhosted.org/packages/c3/9c/095bb28b5da139bd41fb9a5d5caff412584f377914bd8787c2aa98717130/multidict-6.7.1-cp314-cp314t-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:5fa6a95dfee63893d80a34758cd0e0c118a30b8dcb46372bf75106c591b77889", size = 225878, upload-time = "2026-01-26T02:45:48.698Z" },
+    { url = "https://files.pythonhosted.org/packages/07/d0/c0a72000243756e8f5a277b6b514fa005f2c73d481b7d9e47cd4568aa2e4/multidict-6.7.1-cp314-cp314t-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:a0543217a6a017692aa6ae5cc39adb75e587af0f3a82288b1492eb73dd6cc2a4", size = 253542, upload-time = "2026-01-26T02:45:50.164Z" },
+    { url = "https://files.pythonhosted.org/packages/c0/6b/f69da15289e384ecf2a68837ec8b5ad8c33e973aa18b266f50fe55f24b8c/multidict-6.7.1-cp314-cp314t-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:f99fe611c312b3c1c0ace793f92464d8cd263cc3b26b5721950d977b006b6c4d", size = 252403, upload-time = "2026-01-26T02:45:51.779Z" },
+    { url = "https://files.pythonhosted.org/packages/a2/76/b9669547afa5a1a25cd93eaca91c0da1c095b06b6d2d8ec25b713588d3a1/multidict-6.7.1-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:9004d8386d133b7e6135679424c91b0b854d2d164af6ea3f289f8f2761064609", size = 244889, upload-time = "2026-01-26T02:45:53.27Z" },
+    { url = "https://files.pythonhosted.org/packages/7e/a9/a50d2669e506dad33cfc45b5d574a205587b7b8a5f426f2fbb2e90882588/multidict-6.7.1-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:e628ef0e6859ffd8273c69412a2465c4be4a9517d07261b33334b5ec6f3c7489", size = 241982, upload-time = "2026-01-26T02:45:54.919Z" },
+    { url = "https://files.pythonhosted.org/packages/c5/bb/1609558ad8b456b4827d3c5a5b775c93b87878fd3117ed3db3423dfbce1b/multidict-6.7.1-cp314-cp314t-musllinux_1_2_armv7l.whl", hash = "sha256:841189848ba629c3552035a6a7f5bf3b02eb304e9fea7492ca220a8eda6b0e5c", size = 232415, upload-time = "2026-01-26T02:45:56.981Z" },
+    { url = "https://files.pythonhosted.org/packages/d8/59/6f61039d2aa9261871e03ab9dc058a550d240f25859b05b67fd70f80d4b3/multidict-6.7.1-cp314-cp314t-musllinux_1_2_i686.whl", hash = "sha256:ce1bbd7d780bb5a0da032e095c951f7014d6b0a205f8318308140f1a6aba159e", size = 240337, upload-time = "2026-01-26T02:45:58.698Z" },
+    { url = "https://files.pythonhosted.org/packages/a1/29/fdc6a43c203890dc2ae9249971ecd0c41deaedfe00d25cb6564b2edd99eb/multidict-6.7.1-cp314-cp314t-musllinux_1_2_ppc64le.whl", hash = "sha256:b26684587228afed0d50cf804cc71062cc9c1cdf55051c4c6345d372947b268c", size = 248788, upload-time = "2026-01-26T02:46:00.862Z" },
+    { url = "https://files.pythonhosted.org/packages/a9/14/a153a06101323e4cf086ecee3faadba52ff71633d471f9685c42e3736163/multidict-6.7.1-cp314-cp314t-musllinux_1_2_s390x.whl", hash = "sha256:9f9af11306994335398293f9958071019e3ab95e9a707dc1383a35613f6abcb9", size = 242842, upload-time = "2026-01-26T02:46:02.824Z" },
+    { url = "https://files.pythonhosted.org/packages/41/5f/604ae839e64a4a6efc80db94465348d3b328ee955e37acb24badbcd24d83/multidict-6.7.1-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:b4938326284c4f1224178a560987b6cf8b4d38458b113d9b8c1db1a836e640a2", size = 240237, upload-time = "2026-01-26T02:46:05.898Z" },
+    { url = "https://files.pythonhosted.org/packages/5f/60/c3a5187bf66f6fb546ff4ab8fb5a077cbdd832d7b1908d4365c7f74a1917/multidict-6.7.1-cp314-cp314t-win32.whl", hash = "sha256:98655c737850c064a65e006a3df7c997cd3b220be4ec8fe26215760b9697d4d7", size = 48008, upload-time = "2026-01-26T02:46:07.468Z" },
+    { url = "https://files.pythonhosted.org/packages/0c/f7/addf1087b860ac60e6f382240f64fb99f8bfb532bb06f7c542b83c29ca61/multidict-6.7.1-cp314-cp314t-win_amd64.whl", hash = "sha256:497bde6223c212ba11d462853cfa4f0ae6ef97465033e7dc9940cdb3ab5b48e5", size = 53542, upload-time = "2026-01-26T02:46:08.809Z" },
+    { url = "https://files.pythonhosted.org/packages/4c/81/4629d0aa32302ef7b2ec65c75a728cc5ff4fa410c50096174c1632e70b3e/multidict-6.7.1-cp314-cp314t-win_arm64.whl", hash = "sha256:2bbd113e0d4af5db41d5ebfe9ccaff89de2120578164f86a5d17d5a576d1e5b2", size = 44719, upload-time = "2026-01-26T02:46:11.146Z" },
+    { url = "https://files.pythonhosted.org/packages/81/08/7036c080d7117f28a4af526d794aab6a84463126db031b007717c1a6676e/multidict-6.7.1-py3-none-any.whl", hash = "sha256:55d97cc6dae627efa6a6e548885712d4864b81110ac76fa4e534c03819fa4a56", size = 12319, upload-time = "2026-01-26T02:46:44.004Z" },
+]
+
+[[package]]
+name = "mypy"
+version = "1.18.2"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "mypy-extensions" },
+    { name = "pathspec" },
+    { name = "tomli", marker = "python_full_version < '3.11'" },
+    { name = "typing-extensions" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/c0/77/8f0d0001ffad290cef2f7f216f96c814866248a0b92a722365ed54648e7e/mypy-1.18.2.tar.gz", hash = "sha256:06a398102a5f203d7477b2923dda3634c36727fa5c237d8f859ef90c42a9924b", size = 3448846, upload-time = "2025-09-19T00:11:10.519Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/03/6f/657961a0743cff32e6c0611b63ff1c1970a0b482ace35b069203bf705187/mypy-1.18.2-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:c1eab0cf6294dafe397c261a75f96dc2c31bffe3b944faa24db5def4e2b0f77c", size = 12807973, upload-time = "2025-09-19T00:10:35.282Z" },
+    { url = "https://files.pythonhosted.org/packages/10/e9/420822d4f661f13ca8900f5fa239b40ee3be8b62b32f3357df9a3045a08b/mypy-1.18.2-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:7a780ca61fc239e4865968ebc5240bb3bf610ef59ac398de9a7421b54e4a207e", size = 11896527, upload-time = "2025-09-19T00:10:55.791Z" },
+    { url = "https://files.pythonhosted.org/packages/aa/73/a05b2bbaa7005f4642fcfe40fb73f2b4fb6bb44229bd585b5878e9a87ef8/mypy-1.18.2-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:448acd386266989ef11662ce3c8011fd2a7b632e0ec7d61a98edd8e27472225b", size = 12507004, upload-time = "2025-09-19T00:11:05.411Z" },
+    { url = "https://files.pythonhosted.org/packages/4f/01/f6e4b9f0d031c11ccbd6f17da26564f3a0f3c4155af344006434b0a05a9d/mypy-1.18.2-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:f9e171c465ad3901dc652643ee4bffa8e9fef4d7d0eece23b428908c77a76a66", size = 13245947, upload-time = "2025-09-19T00:10:46.923Z" },
+    { url = "https://files.pythonhosted.org/packages/d7/97/19727e7499bfa1ae0773d06afd30ac66a58ed7437d940c70548634b24185/mypy-1.18.2-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:592ec214750bc00741af1f80cbf96b5013d81486b7bb24cb052382c19e40b428", size = 13499217, upload-time = "2025-09-19T00:09:39.472Z" },
+    { url = "https://files.pythonhosted.org/packages/9f/4f/90dc8c15c1441bf31cf0f9918bb077e452618708199e530f4cbd5cede6ff/mypy-1.18.2-cp310-cp310-win_amd64.whl", hash = "sha256:7fb95f97199ea11769ebe3638c29b550b5221e997c63b14ef93d2e971606ebed", size = 9766753, upload-time = "2025-09-19T00:10:49.161Z" },
+    { url = "https://files.pythonhosted.org/packages/88/87/cafd3ae563f88f94eec33f35ff722d043e09832ea8530ef149ec1efbaf08/mypy-1.18.2-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:807d9315ab9d464125aa9fcf6d84fde6e1dc67da0b6f80e7405506b8ac72bc7f", size = 12731198, upload-time = "2025-09-19T00:09:44.857Z" },
+    { url = "https://files.pythonhosted.org/packages/0f/e0/1e96c3d4266a06d4b0197ace5356d67d937d8358e2ee3ffac71faa843724/mypy-1.18.2-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:776bb00de1778caf4db739c6e83919c1d85a448f71979b6a0edd774ea8399341", size = 11817879, upload-time = "2025-09-19T00:09:47.131Z" },
+    { url = "https://files.pythonhosted.org/packages/72/ef/0c9ba89eb03453e76bdac5a78b08260a848c7bfc5d6603634774d9cd9525/mypy-1.18.2-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:1379451880512ffce14505493bd9fe469e0697543717298242574882cf8cdb8d", size = 12427292, upload-time = "2025-09-19T00:10:22.472Z" },
+    { url = "https://files.pythonhosted.org/packages/1a/52/ec4a061dd599eb8179d5411d99775bec2a20542505988f40fc2fee781068/mypy-1.18.2-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:1331eb7fd110d60c24999893320967594ff84c38ac6d19e0a76c5fd809a84c86", size = 13163750, upload-time = "2025-09-19T00:09:51.472Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/5f/2cf2ceb3b36372d51568f2208c021870fe7834cf3186b653ac6446511839/mypy-1.18.2-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:3ca30b50a51e7ba93b00422e486cbb124f1c56a535e20eff7b2d6ab72b3b2e37", size = 13351827, upload-time = "2025-09-19T00:09:58.311Z" },
+    { url = "https://files.pythonhosted.org/packages/c8/7d/2697b930179e7277529eaaec1513f8de622818696857f689e4a5432e5e27/mypy-1.18.2-cp311-cp311-win_amd64.whl", hash = "sha256:664dc726e67fa54e14536f6e1224bcfce1d9e5ac02426d2326e2bb4e081d1ce8", size = 9757983, upload-time = "2025-09-19T00:10:09.071Z" },
+    { url = "https://files.pythonhosted.org/packages/07/06/dfdd2bc60c66611dd8335f463818514733bc763e4760dee289dcc33df709/mypy-1.18.2-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:33eca32dd124b29400c31d7cf784e795b050ace0e1f91b8dc035672725617e34", size = 12908273, upload-time = "2025-09-19T00:10:58.321Z" },
+    { url = "https://files.pythonhosted.org/packages/81/14/6a9de6d13a122d5608e1a04130724caf9170333ac5a924e10f670687d3eb/mypy-1.18.2-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:a3c47adf30d65e89b2dcd2fa32f3aeb5e94ca970d2c15fcb25e297871c8e4764", size = 11920910, upload-time = "2025-09-19T00:10:20.043Z" },
+    { url = "https://files.pythonhosted.org/packages/5f/a9/b29de53e42f18e8cc547e38daa9dfa132ffdc64f7250e353f5c8cdd44bee/mypy-1.18.2-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:5d6c838e831a062f5f29d11c9057c6009f60cb294fea33a98422688181fe2893", size = 12465585, upload-time = "2025-09-19T00:10:33.005Z" },
+    { url = "https://files.pythonhosted.org/packages/77/ae/6c3d2c7c61ff21f2bee938c917616c92ebf852f015fb55917fd6e2811db2/mypy-1.18.2-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:01199871b6110a2ce984bde85acd481232d17413868c9807e95c1b0739a58914", size = 13348562, upload-time = "2025-09-19T00:10:11.51Z" },
+    { url = "https://files.pythonhosted.org/packages/4d/31/aec68ab3b4aebdf8f36d191b0685d99faa899ab990753ca0fee60fb99511/mypy-1.18.2-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:a2afc0fa0b0e91b4599ddfe0f91e2c26c2b5a5ab263737e998d6817874c5f7c8", size = 13533296, upload-time = "2025-09-19T00:10:06.568Z" },
+    { url = "https://files.pythonhosted.org/packages/9f/83/abcb3ad9478fca3ebeb6a5358bb0b22c95ea42b43b7789c7fb1297ca44f4/mypy-1.18.2-cp312-cp312-win_amd64.whl", hash = "sha256:d8068d0afe682c7c4897c0f7ce84ea77f6de953262b12d07038f4d296d547074", size = 9828828, upload-time = "2025-09-19T00:10:28.203Z" },
+    { url = "https://files.pythonhosted.org/packages/5f/04/7f462e6fbba87a72bc8097b93f6842499c428a6ff0c81dd46948d175afe8/mypy-1.18.2-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:07b8b0f580ca6d289e69209ec9d3911b4a26e5abfde32228a288eb79df129fcc", size = 12898728, upload-time = "2025-09-19T00:10:01.33Z" },
+    { url = "https://files.pythonhosted.org/packages/99/5b/61ed4efb64f1871b41fd0b82d29a64640f3516078f6c7905b68ab1ad8b13/mypy-1.18.2-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:ed4482847168439651d3feee5833ccedbf6657e964572706a2adb1f7fa4dfe2e", size = 11910758, upload-time = "2025-09-19T00:10:42.607Z" },
+    { url = "https://files.pythonhosted.org/packages/3c/46/d297d4b683cc89a6e4108c4250a6a6b717f5fa96e1a30a7944a6da44da35/mypy-1.18.2-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:c3ad2afadd1e9fea5cf99a45a822346971ede8685cc581ed9cd4d42eaf940986", size = 12475342, upload-time = "2025-09-19T00:11:00.371Z" },
+    { url = "https://files.pythonhosted.org/packages/83/45/4798f4d00df13eae3bfdf726c9244bcb495ab5bd588c0eed93a2f2dd67f3/mypy-1.18.2-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:a431a6f1ef14cf8c144c6b14793a23ec4eae3db28277c358136e79d7d062f62d", size = 13338709, upload-time = "2025-09-19T00:11:03.358Z" },
+    { url = "https://files.pythonhosted.org/packages/d7/09/479f7358d9625172521a87a9271ddd2441e1dab16a09708f056e97007207/mypy-1.18.2-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:7ab28cc197f1dd77a67e1c6f35cd1f8e8b73ed2217e4fc005f9e6a504e46e7ba", size = 13529806, upload-time = "2025-09-19T00:10:26.073Z" },
+    { url = "https://files.pythonhosted.org/packages/71/cf/ac0f2c7e9d0ea3c75cd99dff7aec1c9df4a1376537cb90e4c882267ee7e9/mypy-1.18.2-cp313-cp313-win_amd64.whl", hash = "sha256:0e2785a84b34a72ba55fb5daf079a1003a34c05b22238da94fcae2bbe46f3544", size = 9833262, upload-time = "2025-09-19T00:10:40.035Z" },
+    { url = "https://files.pythonhosted.org/packages/5a/0c/7d5300883da16f0063ae53996358758b2a2df2a09c72a5061fa79a1f5006/mypy-1.18.2-cp314-cp314-macosx_10_13_x86_64.whl", hash = "sha256:62f0e1e988ad41c2a110edde6c398383a889d95b36b3e60bcf155f5164c4fdce", size = 12893775, upload-time = "2025-09-19T00:10:03.814Z" },
+    { url = "https://files.pythonhosted.org/packages/50/df/2cffbf25737bdb236f60c973edf62e3e7b4ee1c25b6878629e88e2cde967/mypy-1.18.2-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:8795a039bab805ff0c1dfdb8cd3344642c2b99b8e439d057aba30850b8d3423d", size = 11936852, upload-time = "2025-09-19T00:10:51.631Z" },
+    { url = "https://files.pythonhosted.org/packages/be/50/34059de13dd269227fb4a03be1faee6e2a4b04a2051c82ac0a0b5a773c9a/mypy-1.18.2-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:6ca1e64b24a700ab5ce10133f7ccd956a04715463d30498e64ea8715236f9c9c", size = 12480242, upload-time = "2025-09-19T00:11:07.955Z" },
+    { url = "https://files.pythonhosted.org/packages/5b/11/040983fad5132d85914c874a2836252bbc57832065548885b5bb5b0d4359/mypy-1.18.2-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:d924eef3795cc89fecf6bedc6ed32b33ac13e8321344f6ddbf8ee89f706c05cb", size = 13326683, upload-time = "2025-09-19T00:09:55.572Z" },
+    { url = "https://files.pythonhosted.org/packages/e9/ba/89b2901dd77414dd7a8c8729985832a5735053be15b744c18e4586e506ef/mypy-1.18.2-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:20c02215a080e3a2be3aa50506c67242df1c151eaba0dcbc1e4e557922a26075", size = 13514749, upload-time = "2025-09-19T00:10:44.827Z" },
+    { url = "https://files.pythonhosted.org/packages/25/bc/cc98767cffd6b2928ba680f3e5bc969c4152bf7c2d83f92f5a504b92b0eb/mypy-1.18.2-cp314-cp314-win_amd64.whl", hash = "sha256:749b5f83198f1ca64345603118a6f01a4e99ad4bf9d103ddc5a3200cc4614adf", size = 9982959, upload-time = "2025-09-19T00:10:37.344Z" },
+    { url = "https://files.pythonhosted.org/packages/87/e3/be76d87158ebafa0309946c4a73831974d4d6ab4f4ef40c3b53a385a66fd/mypy-1.18.2-py3-none-any.whl", hash = "sha256:22a1748707dd62b58d2ae53562ffc4d7f8bcc727e8ac7cbc69c053ddc874d47e", size = 2352367, upload-time = "2025-09-19T00:10:15.489Z" },
+]
+
+[[package]]
+name = "mypy-extensions"
+version = "1.1.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/a2/6e/371856a3fb9d31ca8dac321cda606860fa4548858c0cc45d9d1d4ca2628b/mypy_extensions-1.1.0.tar.gz", hash = "sha256:52e68efc3284861e772bbcd66823fde5ae21fd2fdb51c62a211403730b916558", size = 6343, upload-time = "2025-04-22T14:54:24.164Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/79/7b/2c79738432f5c924bef5071f933bcc9efd0473bac3b4aa584a6f7c1c8df8/mypy_extensions-1.1.0-py3-none-any.whl", hash = "sha256:1be4cccdb0f2482337c4743e60421de3a356cd97508abadd57d47403e94f5505", size = 4963, upload-time = "2025-04-22T14:54:22.983Z" },
+]
+
+[[package]]
+name = "numpy"
+version = "2.2.6"
+source = { registry = "https://pypi.org/simple" }
+resolution-markers = [
+    "python_full_version < '3.11' and sys_platform == 'darwin'",
+    "python_full_version < '3.11' and platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "(python_full_version < '3.11' and platform_machine != 'aarch64' and sys_platform == 'linux') or (python_full_version < '3.11' and sys_platform != 'darwin' and sys_platform != 'linux')",
+]
+sdist = { url = "https://files.pythonhosted.org/packages/76/21/7d2a95e4bba9dc13d043ee156a356c0a8f0c6309dff6b21b4d71a073b8a8/numpy-2.2.6.tar.gz", hash = "sha256:e29554e2bef54a90aa5cc07da6ce955accb83f21ab5de01a62c8478897b264fd", size = 20276440, upload-time = "2025-05-17T22:38:04.611Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/9a/3e/ed6db5be21ce87955c0cbd3009f2803f59fa08df21b5df06862e2d8e2bdd/numpy-2.2.6-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:b412caa66f72040e6d268491a59f2c43bf03eb6c96dd8f0307829feb7fa2b6fb", size = 21165245, upload-time = "2025-05-17T21:27:58.555Z" },
+    { url = "https://files.pythonhosted.org/packages/22/c2/4b9221495b2a132cc9d2eb862e21d42a009f5a60e45fc44b00118c174bff/numpy-2.2.6-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:8e41fd67c52b86603a91c1a505ebaef50b3314de0213461c7a6e99c9a3beff90", size = 14360048, upload-time = "2025-05-17T21:28:21.406Z" },
+    { url = "https://files.pythonhosted.org/packages/fd/77/dc2fcfc66943c6410e2bf598062f5959372735ffda175b39906d54f02349/numpy-2.2.6-cp310-cp310-macosx_14_0_arm64.whl", hash = "sha256:37e990a01ae6ec7fe7fa1c26c55ecb672dd98b19c3d0e1d1f326fa13cb38d163", size = 5340542, upload-time = "2025-05-17T21:28:30.931Z" },
+    { url = "https://files.pythonhosted.org/packages/7a/4f/1cb5fdc353a5f5cc7feb692db9b8ec2c3d6405453f982435efc52561df58/numpy-2.2.6-cp310-cp310-macosx_14_0_x86_64.whl", hash = "sha256:5a6429d4be8ca66d889b7cf70f536a397dc45ba6faeb5f8c5427935d9592e9cf", size = 6878301, upload-time = "2025-05-17T21:28:41.613Z" },
+    { url = "https://files.pythonhosted.org/packages/eb/17/96a3acd228cec142fcb8723bd3cc39c2a474f7dcf0a5d16731980bcafa95/numpy-2.2.6-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:efd28d4e9cd7d7a8d39074a4d44c63eda73401580c5c76acda2ce969e0a38e83", size = 14297320, upload-time = "2025-05-17T21:29:02.78Z" },
+    { url = "https://files.pythonhosted.org/packages/b4/63/3de6a34ad7ad6646ac7d2f55ebc6ad439dbbf9c4370017c50cf403fb19b5/numpy-2.2.6-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:fc7b73d02efb0e18c000e9ad8b83480dfcd5dfd11065997ed4c6747470ae8915", size = 16801050, upload-time = "2025-05-17T21:29:27.675Z" },
+    { url = "https://files.pythonhosted.org/packages/07/b6/89d837eddef52b3d0cec5c6ba0456c1bf1b9ef6a6672fc2b7873c3ec4e2e/numpy-2.2.6-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:74d4531beb257d2c3f4b261bfb0fc09e0f9ebb8842d82a7b4209415896adc680", size = 15807034, upload-time = "2025-05-17T21:29:51.102Z" },
+    { url = "https://files.pythonhosted.org/packages/01/c8/dc6ae86e3c61cfec1f178e5c9f7858584049b6093f843bca541f94120920/numpy-2.2.6-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:8fc377d995680230e83241d8a96def29f204b5782f371c532579b4f20607a289", size = 18614185, upload-time = "2025-05-17T21:30:18.703Z" },
+    { url = "https://files.pythonhosted.org/packages/5b/c5/0064b1b7e7c89137b471ccec1fd2282fceaae0ab3a9550f2568782d80357/numpy-2.2.6-cp310-cp310-win32.whl", hash = "sha256:b093dd74e50a8cba3e873868d9e93a85b78e0daf2e98c6797566ad8044e8363d", size = 6527149, upload-time = "2025-05-17T21:30:29.788Z" },
+    { url = "https://files.pythonhosted.org/packages/a3/dd/4b822569d6b96c39d1215dbae0582fd99954dcbcf0c1a13c61783feaca3f/numpy-2.2.6-cp310-cp310-win_amd64.whl", hash = "sha256:f0fd6321b839904e15c46e0d257fdd101dd7f530fe03fd6359c1ea63738703f3", size = 12904620, upload-time = "2025-05-17T21:30:48.994Z" },
+    { url = "https://files.pythonhosted.org/packages/da/a8/4f83e2aa666a9fbf56d6118faaaf5f1974d456b1823fda0a176eff722839/numpy-2.2.6-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:f9f1adb22318e121c5c69a09142811a201ef17ab257a1e66ca3025065b7f53ae", size = 21176963, upload-time = "2025-05-17T21:31:19.36Z" },
+    { url = "https://files.pythonhosted.org/packages/b3/2b/64e1affc7972decb74c9e29e5649fac940514910960ba25cd9af4488b66c/numpy-2.2.6-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:c820a93b0255bc360f53eca31a0e676fd1101f673dda8da93454a12e23fc5f7a", size = 14406743, upload-time = "2025-05-17T21:31:41.087Z" },
+    { url = "https://files.pythonhosted.org/packages/4a/9f/0121e375000b5e50ffdd8b25bf78d8e1a5aa4cca3f185d41265198c7b834/numpy-2.2.6-cp311-cp311-macosx_14_0_arm64.whl", hash = "sha256:3d70692235e759f260c3d837193090014aebdf026dfd167834bcba43e30c2a42", size = 5352616, upload-time = "2025-05-17T21:31:50.072Z" },
+    { url = "https://files.pythonhosted.org/packages/31/0d/b48c405c91693635fbe2dcd7bc84a33a602add5f63286e024d3b6741411c/numpy-2.2.6-cp311-cp311-macosx_14_0_x86_64.whl", hash = "sha256:481b49095335f8eed42e39e8041327c05b0f6f4780488f61286ed3c01368d491", size = 6889579, upload-time = "2025-05-17T21:32:01.712Z" },
+    { url = "https://files.pythonhosted.org/packages/52/b8/7f0554d49b565d0171eab6e99001846882000883998e7b7d9f0d98b1f934/numpy-2.2.6-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:b64d8d4d17135e00c8e346e0a738deb17e754230d7e0810ac5012750bbd85a5a", size = 14312005, upload-time = "2025-05-17T21:32:23.332Z" },
+    { url = "https://files.pythonhosted.org/packages/b3/dd/2238b898e51bd6d389b7389ffb20d7f4c10066d80351187ec8e303a5a475/numpy-2.2.6-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:ba10f8411898fc418a521833e014a77d3ca01c15b0c6cdcce6a0d2897e6dbbdf", size = 16821570, upload-time = "2025-05-17T21:32:47.991Z" },
+    { url = "https://files.pythonhosted.org/packages/83/6c/44d0325722cf644f191042bf47eedad61c1e6df2432ed65cbe28509d404e/numpy-2.2.6-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:bd48227a919f1bafbdda0583705e547892342c26fb127219d60a5c36882609d1", size = 15818548, upload-time = "2025-05-17T21:33:11.728Z" },
+    { url = "https://files.pythonhosted.org/packages/ae/9d/81e8216030ce66be25279098789b665d49ff19eef08bfa8cb96d4957f422/numpy-2.2.6-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:9551a499bf125c1d4f9e250377c1ee2eddd02e01eac6644c080162c0c51778ab", size = 18620521, upload-time = "2025-05-17T21:33:39.139Z" },
+    { url = "https://files.pythonhosted.org/packages/6a/fd/e19617b9530b031db51b0926eed5345ce8ddc669bb3bc0044b23e275ebe8/numpy-2.2.6-cp311-cp311-win32.whl", hash = "sha256:0678000bb9ac1475cd454c6b8c799206af8107e310843532b04d49649c717a47", size = 6525866, upload-time = "2025-05-17T21:33:50.273Z" },
+    { url = "https://files.pythonhosted.org/packages/31/0a/f354fb7176b81747d870f7991dc763e157a934c717b67b58456bc63da3df/numpy-2.2.6-cp311-cp311-win_amd64.whl", hash = "sha256:e8213002e427c69c45a52bbd94163084025f533a55a59d6f9c5b820774ef3303", size = 12907455, upload-time = "2025-05-17T21:34:09.135Z" },
+    { url = "https://files.pythonhosted.org/packages/82/5d/c00588b6cf18e1da539b45d3598d3557084990dcc4331960c15ee776ee41/numpy-2.2.6-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:41c5a21f4a04fa86436124d388f6ed60a9343a6f767fced1a8a71c3fbca038ff", size = 20875348, upload-time = "2025-05-17T21:34:39.648Z" },
+    { url = "https://files.pythonhosted.org/packages/66/ee/560deadcdde6c2f90200450d5938f63a34b37e27ebff162810f716f6a230/numpy-2.2.6-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:de749064336d37e340f640b05f24e9e3dd678c57318c7289d222a8a2f543e90c", size = 14119362, upload-time = "2025-05-17T21:35:01.241Z" },
+    { url = "https://files.pythonhosted.org/packages/3c/65/4baa99f1c53b30adf0acd9a5519078871ddde8d2339dc5a7fde80d9d87da/numpy-2.2.6-cp312-cp312-macosx_14_0_arm64.whl", hash = "sha256:894b3a42502226a1cac872f840030665f33326fc3dac8e57c607905773cdcde3", size = 5084103, upload-time = "2025-05-17T21:35:10.622Z" },
+    { url = "https://files.pythonhosted.org/packages/cc/89/e5a34c071a0570cc40c9a54eb472d113eea6d002e9ae12bb3a8407fb912e/numpy-2.2.6-cp312-cp312-macosx_14_0_x86_64.whl", hash = "sha256:71594f7c51a18e728451bb50cc60a3ce4e6538822731b2933209a1f3614e9282", size = 6625382, upload-time = "2025-05-17T21:35:21.414Z" },
+    { url = "https://files.pythonhosted.org/packages/f8/35/8c80729f1ff76b3921d5c9487c7ac3de9b2a103b1cd05e905b3090513510/numpy-2.2.6-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:f2618db89be1b4e05f7a1a847a9c1c0abd63e63a1607d892dd54668dd92faf87", size = 14018462, upload-time = "2025-05-17T21:35:42.174Z" },
+    { url = "https://files.pythonhosted.org/packages/8c/3d/1e1db36cfd41f895d266b103df00ca5b3cbe965184df824dec5c08c6b803/numpy-2.2.6-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:fd83c01228a688733f1ded5201c678f0c53ecc1006ffbc404db9f7a899ac6249", size = 16527618, upload-time = "2025-05-17T21:36:06.711Z" },
+    { url = "https://files.pythonhosted.org/packages/61/c6/03ed30992602c85aa3cd95b9070a514f8b3c33e31124694438d88809ae36/numpy-2.2.6-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:37c0ca431f82cd5fa716eca9506aefcabc247fb27ba69c5062a6d3ade8cf8f49", size = 15505511, upload-time = "2025-05-17T21:36:29.965Z" },
+    { url = "https://files.pythonhosted.org/packages/b7/25/5761d832a81df431e260719ec45de696414266613c9ee268394dd5ad8236/numpy-2.2.6-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:fe27749d33bb772c80dcd84ae7e8df2adc920ae8297400dabec45f0dedb3f6de", size = 18313783, upload-time = "2025-05-17T21:36:56.883Z" },
+    { url = "https://files.pythonhosted.org/packages/57/0a/72d5a3527c5ebffcd47bde9162c39fae1f90138c961e5296491ce778e682/numpy-2.2.6-cp312-cp312-win32.whl", hash = "sha256:4eeaae00d789f66c7a25ac5f34b71a7035bb474e679f410e5e1a94deb24cf2d4", size = 6246506, upload-time = "2025-05-17T21:37:07.368Z" },
+    { url = "https://files.pythonhosted.org/packages/36/fa/8c9210162ca1b88529ab76b41ba02d433fd54fecaf6feb70ef9f124683f1/numpy-2.2.6-cp312-cp312-win_amd64.whl", hash = "sha256:c1f9540be57940698ed329904db803cf7a402f3fc200bfe599334c9bd84a40b2", size = 12614190, upload-time = "2025-05-17T21:37:26.213Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/5c/6657823f4f594f72b5471f1db1ab12e26e890bb2e41897522d134d2a3e81/numpy-2.2.6-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:0811bb762109d9708cca4d0b13c4f67146e3c3b7cf8d34018c722adb2d957c84", size = 20867828, upload-time = "2025-05-17T21:37:56.699Z" },
+    { url = "https://files.pythonhosted.org/packages/dc/9e/14520dc3dadf3c803473bd07e9b2bd1b69bc583cb2497b47000fed2fa92f/numpy-2.2.6-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:287cc3162b6f01463ccd86be154f284d0893d2b3ed7292439ea97eafa8170e0b", size = 14143006, upload-time = "2025-05-17T21:38:18.291Z" },
+    { url = "https://files.pythonhosted.org/packages/4f/06/7e96c57d90bebdce9918412087fc22ca9851cceaf5567a45c1f404480e9e/numpy-2.2.6-cp313-cp313-macosx_14_0_arm64.whl", hash = "sha256:f1372f041402e37e5e633e586f62aa53de2eac8d98cbfb822806ce4bbefcb74d", size = 5076765, upload-time = "2025-05-17T21:38:27.319Z" },
+    { url = "https://files.pythonhosted.org/packages/73/ed/63d920c23b4289fdac96ddbdd6132e9427790977d5457cd132f18e76eae0/numpy-2.2.6-cp313-cp313-macosx_14_0_x86_64.whl", hash = "sha256:55a4d33fa519660d69614a9fad433be87e5252f4b03850642f88993f7b2ca566", size = 6617736, upload-time = "2025-05-17T21:38:38.141Z" },
+    { url = "https://files.pythonhosted.org/packages/85/c5/e19c8f99d83fd377ec8c7e0cf627a8049746da54afc24ef0a0cb73d5dfb5/numpy-2.2.6-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:f92729c95468a2f4f15e9bb94c432a9229d0d50de67304399627a943201baa2f", size = 14010719, upload-time = "2025-05-17T21:38:58.433Z" },
+    { url = "https://files.pythonhosted.org/packages/19/49/4df9123aafa7b539317bf6d342cb6d227e49f7a35b99c287a6109b13dd93/numpy-2.2.6-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:1bc23a79bfabc5d056d106f9befb8d50c31ced2fbc70eedb8155aec74a45798f", size = 16526072, upload-time = "2025-05-17T21:39:22.638Z" },
+    { url = "https://files.pythonhosted.org/packages/b2/6c/04b5f47f4f32f7c2b0e7260442a8cbcf8168b0e1a41ff1495da42f42a14f/numpy-2.2.6-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:e3143e4451880bed956e706a3220b4e5cf6172ef05fcc397f6f36a550b1dd868", size = 15503213, upload-time = "2025-05-17T21:39:45.865Z" },
+    { url = "https://files.pythonhosted.org/packages/17/0a/5cd92e352c1307640d5b6fec1b2ffb06cd0dabe7d7b8227f97933d378422/numpy-2.2.6-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:b4f13750ce79751586ae2eb824ba7e1e8dba64784086c98cdbbcc6a42112ce0d", size = 18316632, upload-time = "2025-05-17T21:40:13.331Z" },
+    { url = "https://files.pythonhosted.org/packages/f0/3b/5cba2b1d88760ef86596ad0f3d484b1cbff7c115ae2429678465057c5155/numpy-2.2.6-cp313-cp313-win32.whl", hash = "sha256:5beb72339d9d4fa36522fc63802f469b13cdbe4fdab4a288f0c441b74272ebfd", size = 6244532, upload-time = "2025-05-17T21:43:46.099Z" },
+    { url = "https://files.pythonhosted.org/packages/cb/3b/d58c12eafcb298d4e6d0d40216866ab15f59e55d148a5658bb3132311fcf/numpy-2.2.6-cp313-cp313-win_amd64.whl", hash = "sha256:b0544343a702fa80c95ad5d3d608ea3599dd54d4632df855e4c8d24eb6ecfa1c", size = 12610885, upload-time = "2025-05-17T21:44:05.145Z" },
+    { url = "https://files.pythonhosted.org/packages/6b/9e/4bf918b818e516322db999ac25d00c75788ddfd2d2ade4fa66f1f38097e1/numpy-2.2.6-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:0bca768cd85ae743b2affdc762d617eddf3bcf8724435498a1e80132d04879e6", size = 20963467, upload-time = "2025-05-17T21:40:44Z" },
+    { url = "https://files.pythonhosted.org/packages/61/66/d2de6b291507517ff2e438e13ff7b1e2cdbdb7cb40b3ed475377aece69f9/numpy-2.2.6-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:fc0c5673685c508a142ca65209b4e79ed6740a4ed6b2267dbba90f34b0b3cfda", size = 14225144, upload-time = "2025-05-17T21:41:05.695Z" },
+    { url = "https://files.pythonhosted.org/packages/e4/25/480387655407ead912e28ba3a820bc69af9adf13bcbe40b299d454ec011f/numpy-2.2.6-cp313-cp313t-macosx_14_0_arm64.whl", hash = "sha256:5bd4fc3ac8926b3819797a7c0e2631eb889b4118a9898c84f585a54d475b7e40", size = 5200217, upload-time = "2025-05-17T21:41:15.903Z" },
+    { url = "https://files.pythonhosted.org/packages/aa/4a/6e313b5108f53dcbf3aca0c0f3e9c92f4c10ce57a0a721851f9785872895/numpy-2.2.6-cp313-cp313t-macosx_14_0_x86_64.whl", hash = "sha256:fee4236c876c4e8369388054d02d0e9bb84821feb1a64dd59e137e6511a551f8", size = 6712014, upload-time = "2025-05-17T21:41:27.321Z" },
+    { url = "https://files.pythonhosted.org/packages/b7/30/172c2d5c4be71fdf476e9de553443cf8e25feddbe185e0bd88b096915bcc/numpy-2.2.6-cp313-cp313t-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:e1dda9c7e08dc141e0247a5b8f49cf05984955246a327d4c48bda16821947b2f", size = 14077935, upload-time = "2025-05-17T21:41:49.738Z" },
+    { url = "https://files.pythonhosted.org/packages/12/fb/9e743f8d4e4d3c710902cf87af3512082ae3d43b945d5d16563f26ec251d/numpy-2.2.6-cp313-cp313t-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:f447e6acb680fd307f40d3da4852208af94afdfab89cf850986c3ca00562f4fa", size = 16600122, upload-time = "2025-05-17T21:42:14.046Z" },
+    { url = "https://files.pythonhosted.org/packages/12/75/ee20da0e58d3a66f204f38916757e01e33a9737d0b22373b3eb5a27358f9/numpy-2.2.6-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:389d771b1623ec92636b0786bc4ae56abafad4a4c513d36a55dce14bd9ce8571", size = 15586143, upload-time = "2025-05-17T21:42:37.464Z" },
+    { url = "https://files.pythonhosted.org/packages/76/95/bef5b37f29fc5e739947e9ce5179ad402875633308504a52d188302319c8/numpy-2.2.6-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:8e9ace4a37db23421249ed236fdcdd457d671e25146786dfc96835cd951aa7c1", size = 18385260, upload-time = "2025-05-17T21:43:05.189Z" },
+    { url = "https://files.pythonhosted.org/packages/09/04/f2f83279d287407cf36a7a8053a5abe7be3622a4363337338f2585e4afda/numpy-2.2.6-cp313-cp313t-win32.whl", hash = "sha256:038613e9fb8c72b0a41f025a7e4c3f0b7a1b5d768ece4796b674c8f3fe13efff", size = 6377225, upload-time = "2025-05-17T21:43:16.254Z" },
+    { url = "https://files.pythonhosted.org/packages/67/0e/35082d13c09c02c011cf21570543d202ad929d961c02a147493cb0c2bdf5/numpy-2.2.6-cp313-cp313t-win_amd64.whl", hash = "sha256:6031dd6dfecc0cf9f668681a37648373bddd6421fff6c66ec1624eed0180ee06", size = 12771374, upload-time = "2025-05-17T21:43:35.479Z" },
+    { url = "https://files.pythonhosted.org/packages/9e/3b/d94a75f4dbf1ef5d321523ecac21ef23a3cd2ac8b78ae2aac40873590229/numpy-2.2.6-pp310-pypy310_pp73-macosx_10_15_x86_64.whl", hash = "sha256:0b605b275d7bd0c640cad4e5d30fa701a8d59302e127e5f79138ad62762c3e3d", size = 21040391, upload-time = "2025-05-17T21:44:35.948Z" },
+    { url = "https://files.pythonhosted.org/packages/17/f4/09b2fa1b58f0fb4f7c7963a1649c64c4d315752240377ed74d9cd878f7b5/numpy-2.2.6-pp310-pypy310_pp73-macosx_14_0_x86_64.whl", hash = "sha256:7befc596a7dc9da8a337f79802ee8adb30a552a94f792b9c9d18c840055907db", size = 6786754, upload-time = "2025-05-17T21:44:47.446Z" },
+    { url = "https://files.pythonhosted.org/packages/af/30/feba75f143bdc868a1cc3f44ccfa6c4b9ec522b36458e738cd00f67b573f/numpy-2.2.6-pp310-pypy310_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:ce47521a4754c8f4593837384bd3424880629f718d87c5d44f8ed763edd63543", size = 16643476, upload-time = "2025-05-17T21:45:11.871Z" },
+    { url = "https://files.pythonhosted.org/packages/37/48/ac2a9584402fb6c0cd5b5d1a91dcf176b15760130dd386bbafdbfe3640bf/numpy-2.2.6-pp310-pypy310_pp73-win_amd64.whl", hash = "sha256:d042d24c90c41b54fd506da306759e06e568864df8ec17ccc17e9e884634fd00", size = 12812666, upload-time = "2025-05-17T21:45:31.426Z" },
+]
+
+[[package]]
+name = "numpy"
+version = "2.4.2"
+source = { registry = "https://pypi.org/simple" }
+resolution-markers = [
+    "python_full_version >= '3.14' and sys_platform == 'darwin'",
+    "python_full_version == '3.13.*' and sys_platform == 'darwin'",
+    "python_full_version >= '3.14' and platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "python_full_version == '3.13.*' and platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "python_full_version >= '3.14' and sys_platform == 'win32'",
+    "python_full_version >= '3.14' and sys_platform == 'emscripten'",
+    "(python_full_version >= '3.14' and platform_machine != 'aarch64' and sys_platform == 'linux') or (python_full_version >= '3.14' and sys_platform != 'darwin' and sys_platform != 'emscripten' and sys_platform != 'linux' and sys_platform != 'win32')",
+    "python_full_version == '3.13.*' and sys_platform == 'win32'",
+    "python_full_version == '3.13.*' and sys_platform == 'emscripten'",
+    "(python_full_version == '3.13.*' and platform_machine != 'aarch64' and sys_platform == 'linux') or (python_full_version == '3.13.*' and sys_platform != 'darwin' and sys_platform != 'emscripten' and sys_platform != 'linux' and sys_platform != 'win32')",
+    "python_full_version == '3.12.*' and sys_platform == 'darwin'",
+    "python_full_version == '3.12.*' and platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "python_full_version == '3.12.*' and sys_platform == 'win32'",
+    "python_full_version == '3.12.*' and sys_platform == 'emscripten'",
+    "(python_full_version == '3.12.*' and platform_machine != 'aarch64' and sys_platform == 'linux') or (python_full_version == '3.12.*' and sys_platform != 'darwin' and sys_platform != 'emscripten' and sys_platform != 'linux' and sys_platform != 'win32')",
+    "python_full_version == '3.11.*' and sys_platform == 'darwin'",
+    "python_full_version == '3.11.*' and platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "python_full_version == '3.11.*' and sys_platform == 'win32'",
+    "python_full_version == '3.11.*' and sys_platform == 'emscripten'",
+    "(python_full_version == '3.11.*' and platform_machine != 'aarch64' and sys_platform == 'linux') or (python_full_version == '3.11.*' and sys_platform != 'darwin' and sys_platform != 'emscripten' and sys_platform != 'linux' and sys_platform != 'win32')",
+]
+sdist = { url = "https://files.pythonhosted.org/packages/57/fd/0005efbd0af48e55eb3c7208af93f2862d4b1a56cd78e84309a2d959208d/numpy-2.4.2.tar.gz", hash = "sha256:659a6107e31a83c4e33f763942275fd278b21d095094044eb35569e86a21ddae", size = 20723651, upload-time = "2026-01-31T23:13:10.135Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/d3/44/71852273146957899753e69986246d6a176061ea183407e95418c2aa4d9a/numpy-2.4.2-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:e7e88598032542bd49af7c4747541422884219056c268823ef6e5e89851c8825", size = 16955478, upload-time = "2026-01-31T23:10:25.623Z" },
+    { url = "https://files.pythonhosted.org/packages/74/41/5d17d4058bd0cd96bcbd4d9ff0fb2e21f52702aab9a72e4a594efa18692f/numpy-2.4.2-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:7edc794af8b36ca37ef5fcb5e0d128c7e0595c7b96a2318d1badb6fcd8ee86b1", size = 14965467, upload-time = "2026-01-31T23:10:28.186Z" },
+    { url = "https://files.pythonhosted.org/packages/49/48/fb1ce8136c19452ed15f033f8aee91d5defe515094e330ce368a0647846f/numpy-2.4.2-cp311-cp311-macosx_14_0_arm64.whl", hash = "sha256:6e9f61981ace1360e42737e2bae58b27bf28a1b27e781721047d84bd754d32e7", size = 5475172, upload-time = "2026-01-31T23:10:30.848Z" },
+    { url = "https://files.pythonhosted.org/packages/40/a9/3feb49f17bbd1300dd2570432961f5c8a4ffeff1db6f02c7273bd020a4c9/numpy-2.4.2-cp311-cp311-macosx_14_0_x86_64.whl", hash = "sha256:cb7bbb88aa74908950d979eeaa24dbdf1a865e3c7e45ff0121d8f70387b55f73", size = 6805145, upload-time = "2026-01-31T23:10:32.352Z" },
+    { url = "https://files.pythonhosted.org/packages/3f/39/fdf35cbd6d6e2fcad42fcf85ac04a85a0d0fbfbf34b30721c98d602fd70a/numpy-2.4.2-cp311-cp311-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:4f069069931240b3fc703f1e23df63443dbd6390614c8c44a87d96cd0ec81eb1", size = 15966084, upload-time = "2026-01-31T23:10:34.502Z" },
+    { url = "https://files.pythonhosted.org/packages/1b/46/6fa4ea94f1ddf969b2ee941290cca6f1bfac92b53c76ae5f44afe17ceb69/numpy-2.4.2-cp311-cp311-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:c02ef4401a506fb60b411467ad501e1429a3487abca4664871d9ae0b46c8ba32", size = 16899477, upload-time = "2026-01-31T23:10:37.075Z" },
+    { url = "https://files.pythonhosted.org/packages/09/a1/2a424e162b1a14a5bd860a464ab4e07513916a64ab1683fae262f735ccd2/numpy-2.4.2-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:2653de5c24910e49c2b106499803124dde62a5a1fe0eedeaecf4309a5f639390", size = 17323429, upload-time = "2026-01-31T23:10:39.704Z" },
+    { url = "https://files.pythonhosted.org/packages/ce/a2/73014149ff250628df72c58204822ac01d768697913881aacf839ff78680/numpy-2.4.2-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:1ae241bbfc6ae276f94a170b14785e561cb5e7f626b6688cf076af4110887413", size = 18635109, upload-time = "2026-01-31T23:10:41.924Z" },
+    { url = "https://files.pythonhosted.org/packages/6c/0c/73e8be2f1accd56df74abc1c5e18527822067dced5ec0861b5bb882c2ce0/numpy-2.4.2-cp311-cp311-win32.whl", hash = "sha256:df1b10187212b198dd45fa943d8985a3c8cf854aed4923796e0e019e113a1bda", size = 6237915, upload-time = "2026-01-31T23:10:45.26Z" },
+    { url = "https://files.pythonhosted.org/packages/76/ae/e0265e0163cf127c24c3969d29f1c4c64551a1e375d95a13d32eab25d364/numpy-2.4.2-cp311-cp311-win_amd64.whl", hash = "sha256:b9c618d56a29c9cb1c4da979e9899be7578d2e0b3c24d52079c166324c9e8695", size = 12607972, upload-time = "2026-01-31T23:10:47.021Z" },
+    { url = "https://files.pythonhosted.org/packages/29/a5/c43029af9b8014d6ea157f192652c50042e8911f4300f8f6ed3336bf437f/numpy-2.4.2-cp311-cp311-win_arm64.whl", hash = "sha256:47c5a6ed21d9452b10227e5e8a0e1c22979811cad7dcc19d8e3e2fb8fa03f1a3", size = 10485763, upload-time = "2026-01-31T23:10:50.087Z" },
+    { url = "https://files.pythonhosted.org/packages/51/6e/6f394c9c77668153e14d4da83bcc247beb5952f6ead7699a1a2992613bea/numpy-2.4.2-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:21982668592194c609de53ba4933a7471880ccbaadcc52352694a59ecc860b3a", size = 16667963, upload-time = "2026-01-31T23:10:52.147Z" },
+    { url = "https://files.pythonhosted.org/packages/1f/f8/55483431f2b2fd015ae6ed4fe62288823ce908437ed49db5a03d15151678/numpy-2.4.2-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:40397bda92382fcec844066efb11f13e1c9a3e2a8e8f318fb72ed8b6db9f60f1", size = 14693571, upload-time = "2026-01-31T23:10:54.789Z" },
+    { url = "https://files.pythonhosted.org/packages/2f/20/18026832b1845cdc82248208dd929ca14c9d8f2bac391f67440707fff27c/numpy-2.4.2-cp312-cp312-macosx_14_0_arm64.whl", hash = "sha256:b3a24467af63c67829bfaa61eecf18d5432d4f11992688537be59ecd6ad32f5e", size = 5203469, upload-time = "2026-01-31T23:10:57.343Z" },
+    { url = "https://files.pythonhosted.org/packages/7d/33/2eb97c8a77daaba34eaa3fa7241a14ac5f51c46a6bd5911361b644c4a1e2/numpy-2.4.2-cp312-cp312-macosx_14_0_x86_64.whl", hash = "sha256:805cc8de9fd6e7a22da5aed858e0ab16be5a4db6c873dde1d7451c541553aa27", size = 6550820, upload-time = "2026-01-31T23:10:59.429Z" },
+    { url = "https://files.pythonhosted.org/packages/b1/91/b97fdfd12dc75b02c44e26c6638241cc004d4079a0321a69c62f51470c4c/numpy-2.4.2-cp312-cp312-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:6d82351358ffbcdcd7b686b90742a9b86632d6c1c051016484fa0b326a0a1548", size = 15663067, upload-time = "2026-01-31T23:11:01.291Z" },
+    { url = "https://files.pythonhosted.org/packages/f5/c6/a18e59f3f0b8071cc85cbc8d80cd02d68aa9710170b2553a117203d46936/numpy-2.4.2-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:9e35d3e0144137d9fdae62912e869136164534d64a169f86438bc9561b6ad49f", size = 16619782, upload-time = "2026-01-31T23:11:03.669Z" },
+    { url = "https://files.pythonhosted.org/packages/b7/83/9751502164601a79e18847309f5ceec0b1446d7b6aa12305759b72cf98b2/numpy-2.4.2-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:adb6ed2ad29b9e15321d167d152ee909ec73395901b70936f029c3bc6d7f4460", size = 17013128, upload-time = "2026-01-31T23:11:05.913Z" },
+    { url = "https://files.pythonhosted.org/packages/61/c4/c4066322256ec740acc1c8923a10047818691d2f8aec254798f3dd90f5f2/numpy-2.4.2-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:8906e71fd8afcb76580404e2a950caef2685df3d2a57fe82a86ac8d33cc007ba", size = 18345324, upload-time = "2026-01-31T23:11:08.248Z" },
+    { url = "https://files.pythonhosted.org/packages/ab/af/6157aa6da728fa4525a755bfad486ae7e3f76d4c1864138003eb84328497/numpy-2.4.2-cp312-cp312-win32.whl", hash = "sha256:ec055f6dae239a6299cace477b479cca2fc125c5675482daf1dd886933a1076f", size = 5960282, upload-time = "2026-01-31T23:11:10.497Z" },
+    { url = "https://files.pythonhosted.org/packages/92/0f/7ceaaeaacb40567071e94dbf2c9480c0ae453d5bb4f52bea3892c39dc83c/numpy-2.4.2-cp312-cp312-win_amd64.whl", hash = "sha256:209fae046e62d0ce6435fcfe3b1a10537e858249b3d9b05829e2a05218296a85", size = 12314210, upload-time = "2026-01-31T23:11:12.176Z" },
+    { url = "https://files.pythonhosted.org/packages/2f/a3/56c5c604fae6dd40fa2ed3040d005fca97e91bd320d232ac9931d77ba13c/numpy-2.4.2-cp312-cp312-win_arm64.whl", hash = "sha256:fbde1b0c6e81d56f5dccd95dd4a711d9b95df1ae4009a60887e56b27e8d903fa", size = 10220171, upload-time = "2026-01-31T23:11:14.684Z" },
+    { url = "https://files.pythonhosted.org/packages/a1/22/815b9fe25d1d7ae7d492152adbc7226d3eff731dffc38fe970589fcaaa38/numpy-2.4.2-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:25f2059807faea4b077a2b6837391b5d830864b3543627f381821c646f31a63c", size = 16663696, upload-time = "2026-01-31T23:11:17.516Z" },
+    { url = "https://files.pythonhosted.org/packages/09/f0/817d03a03f93ba9c6c8993de509277d84e69f9453601915e4a69554102a1/numpy-2.4.2-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:bd3a7a9f5847d2fb8c2c6d1c862fa109c31a9abeca1a3c2bd5a64572955b2979", size = 14688322, upload-time = "2026-01-31T23:11:19.883Z" },
+    { url = "https://files.pythonhosted.org/packages/da/b4/f805ab79293c728b9a99438775ce51885fd4f31b76178767cfc718701a39/numpy-2.4.2-cp313-cp313-macosx_14_0_arm64.whl", hash = "sha256:8e4549f8a3c6d13d55041925e912bfd834285ef1dd64d6bc7d542583355e2e98", size = 5198157, upload-time = "2026-01-31T23:11:22.375Z" },
+    { url = "https://files.pythonhosted.org/packages/74/09/826e4289844eccdcd64aac27d13b0fd3f32039915dd5b9ba01baae1f436c/numpy-2.4.2-cp313-cp313-macosx_14_0_x86_64.whl", hash = "sha256:aea4f66ff44dfddf8c2cffd66ba6538c5ec67d389285292fe428cb2c738c8aef", size = 6546330, upload-time = "2026-01-31T23:11:23.958Z" },
+    { url = "https://files.pythonhosted.org/packages/19/fb/cbfdbfa3057a10aea5422c558ac57538e6acc87ec1669e666d32ac198da7/numpy-2.4.2-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:c3cd545784805de05aafe1dde61752ea49a359ccba9760c1e5d1c88a93bbf2b7", size = 15660968, upload-time = "2026-01-31T23:11:25.713Z" },
+    { url = "https://files.pythonhosted.org/packages/04/dc/46066ce18d01645541f0186877377b9371b8fa8017fa8262002b4ef22612/numpy-2.4.2-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:d0d9b7c93578baafcbc5f0b83eaf17b79d345c6f36917ba0c67f45226911d499", size = 16607311, upload-time = "2026-01-31T23:11:28.117Z" },
+    { url = "https://files.pythonhosted.org/packages/14/d9/4b5adfc39a43fa6bf918c6d544bc60c05236cc2f6339847fc5b35e6cb5b0/numpy-2.4.2-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:f74f0f7779cc7ae07d1810aab8ac6b1464c3eafb9e283a40da7309d5e6e48fbb", size = 17012850, upload-time = "2026-01-31T23:11:30.888Z" },
+    { url = "https://files.pythonhosted.org/packages/b7/20/adb6e6adde6d0130046e6fdfb7675cc62bc2f6b7b02239a09eb58435753d/numpy-2.4.2-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:c7ac672d699bf36275c035e16b65539931347d68b70667d28984c9fb34e07fa7", size = 18334210, upload-time = "2026-01-31T23:11:33.214Z" },
+    { url = "https://files.pythonhosted.org/packages/78/0e/0a73b3dff26803a8c02baa76398015ea2a5434d9b8265a7898a6028c1591/numpy-2.4.2-cp313-cp313-win32.whl", hash = "sha256:8e9afaeb0beff068b4d9cd20d322ba0ee1cecfb0b08db145e4ab4dd44a6b5110", size = 5958199, upload-time = "2026-01-31T23:11:35.385Z" },
+    { url = "https://files.pythonhosted.org/packages/43/bc/6352f343522fcb2c04dbaf94cb30cca6fd32c1a750c06ad6231b4293708c/numpy-2.4.2-cp313-cp313-win_amd64.whl", hash = "sha256:7df2de1e4fba69a51c06c28f5a3de36731eb9639feb8e1cf7e4a7b0daf4cf622", size = 12310848, upload-time = "2026-01-31T23:11:38.001Z" },
+    { url = "https://files.pythonhosted.org/packages/6e/8d/6da186483e308da5da1cc6918ce913dcfe14ffde98e710bfeff2a6158d4e/numpy-2.4.2-cp313-cp313-win_arm64.whl", hash = "sha256:0fece1d1f0a89c16b03442eae5c56dc0be0c7883b5d388e0c03f53019a4bfd71", size = 10221082, upload-time = "2026-01-31T23:11:40.392Z" },
+    { url = "https://files.pythonhosted.org/packages/25/a1/9510aa43555b44781968935c7548a8926274f815de42ad3997e9e83680dd/numpy-2.4.2-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:5633c0da313330fd20c484c78cdd3f9b175b55e1a766c4a174230c6b70ad8262", size = 14815866, upload-time = "2026-01-31T23:11:42.495Z" },
+    { url = "https://files.pythonhosted.org/packages/36/30/6bbb5e76631a5ae46e7923dd16ca9d3f1c93cfa8d4ed79a129814a9d8db3/numpy-2.4.2-cp313-cp313t-macosx_14_0_arm64.whl", hash = "sha256:d9f64d786b3b1dd742c946c42d15b07497ed14af1a1f3ce840cce27daa0ce913", size = 5325631, upload-time = "2026-01-31T23:11:44.7Z" },
+    { url = "https://files.pythonhosted.org/packages/46/00/3a490938800c1923b567b3a15cd17896e68052e2145d8662aaf3e1ffc58f/numpy-2.4.2-cp313-cp313t-macosx_14_0_x86_64.whl", hash = "sha256:b21041e8cb6a1eb5312dd1d2f80a94d91efffb7a06b70597d44f1bd2dfc315ab", size = 6646254, upload-time = "2026-01-31T23:11:46.341Z" },
+    { url = "https://files.pythonhosted.org/packages/d3/e9/fac0890149898a9b609caa5af7455a948b544746e4b8fe7c212c8edd71f8/numpy-2.4.2-cp313-cp313t-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:00ab83c56211a1d7c07c25e3217ea6695e50a3e2f255053686b081dc0b091a82", size = 15720138, upload-time = "2026-01-31T23:11:48.082Z" },
+    { url = "https://files.pythonhosted.org/packages/ea/5c/08887c54e68e1e28df53709f1893ce92932cc6f01f7c3d4dc952f61ffd4e/numpy-2.4.2-cp313-cp313t-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:2fb882da679409066b4603579619341c6d6898fc83a8995199d5249f986e8e8f", size = 16655398, upload-time = "2026-01-31T23:11:50.293Z" },
+    { url = "https://files.pythonhosted.org/packages/4d/89/253db0fa0e66e9129c745e4ef25631dc37d5f1314dad2b53e907b8538e6d/numpy-2.4.2-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:66cb9422236317f9d44b67b4d18f44efe6e9c7f8794ac0462978513359461554", size = 17079064, upload-time = "2026-01-31T23:11:52.927Z" },
+    { url = "https://files.pythonhosted.org/packages/2a/d5/cbade46ce97c59c6c3da525e8d95b7abe8a42974a1dc5c1d489c10433e88/numpy-2.4.2-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:0f01dcf33e73d80bd8dc0f20a71303abbafa26a19e23f6b68d1aa9990af90257", size = 18379680, upload-time = "2026-01-31T23:11:55.22Z" },
+    { url = "https://files.pythonhosted.org/packages/40/62/48f99ae172a4b63d981babe683685030e8a3df4f246c893ea5c6ef99f018/numpy-2.4.2-cp313-cp313t-win32.whl", hash = "sha256:52b913ec40ff7ae845687b0b34d8d93b60cb66dcee06996dd5c99f2fc9328657", size = 6082433, upload-time = "2026-01-31T23:11:58.096Z" },
+    { url = "https://files.pythonhosted.org/packages/07/38/e054a61cfe48ad9f1ed0d188e78b7e26859d0b60ef21cd9de4897cdb5326/numpy-2.4.2-cp313-cp313t-win_amd64.whl", hash = "sha256:5eea80d908b2c1f91486eb95b3fb6fab187e569ec9752ab7d9333d2e66bf2d6b", size = 12451181, upload-time = "2026-01-31T23:11:59.782Z" },
+    { url = "https://files.pythonhosted.org/packages/6e/a4/a05c3a6418575e185dd84d0b9680b6bb2e2dc3e4202f036b7b4e22d6e9dc/numpy-2.4.2-cp313-cp313t-win_arm64.whl", hash = "sha256:fd49860271d52127d61197bb50b64f58454e9f578cb4b2c001a6de8b1f50b0b1", size = 10290756, upload-time = "2026-01-31T23:12:02.438Z" },
+    { url = "https://files.pythonhosted.org/packages/18/88/b7df6050bf18fdcfb7046286c6535cabbdd2064a3440fca3f069d319c16e/numpy-2.4.2-cp314-cp314-macosx_10_15_x86_64.whl", hash = "sha256:444be170853f1f9d528428eceb55f12918e4fda5d8805480f36a002f1415e09b", size = 16663092, upload-time = "2026-01-31T23:12:04.521Z" },
+    { url = "https://files.pythonhosted.org/packages/25/7a/1fee4329abc705a469a4afe6e69b1ef7e915117747886327104a8493a955/numpy-2.4.2-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:d1240d50adff70c2a88217698ca844723068533f3f5c5fa6ee2e3220e3bdb000", size = 14698770, upload-time = "2026-01-31T23:12:06.96Z" },
+    { url = "https://files.pythonhosted.org/packages/fb/0b/f9e49ba6c923678ad5bc38181c08ac5e53b7a5754dbca8e581aa1a56b1ff/numpy-2.4.2-cp314-cp314-macosx_14_0_arm64.whl", hash = "sha256:7cdde6de52fb6664b00b056341265441192d1291c130e99183ec0d4b110ff8b1", size = 5208562, upload-time = "2026-01-31T23:12:09.632Z" },
+    { url = "https://files.pythonhosted.org/packages/7d/12/d7de8f6f53f9bb76997e5e4c069eda2051e3fe134e9181671c4391677bb2/numpy-2.4.2-cp314-cp314-macosx_14_0_x86_64.whl", hash = "sha256:cda077c2e5b780200b6b3e09d0b42205a3d1c68f30c6dceb90401c13bff8fe74", size = 6543710, upload-time = "2026-01-31T23:12:11.969Z" },
+    { url = "https://files.pythonhosted.org/packages/09/63/c66418c2e0268a31a4cf8a8b512685748200f8e8e8ec6c507ce14e773529/numpy-2.4.2-cp314-cp314-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:d30291931c915b2ab5717c2974bb95ee891a1cf22ebc16a8006bd59cd210d40a", size = 15677205, upload-time = "2026-01-31T23:12:14.33Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/6c/7f237821c9642fb2a04d2f1e88b4295677144ca93285fd76eff3bcba858d/numpy-2.4.2-cp314-cp314-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:bba37bc29d4d85761deed3954a1bc62be7cf462b9510b51d367b769a8c8df325", size = 16611738, upload-time = "2026-01-31T23:12:16.525Z" },
+    { url = "https://files.pythonhosted.org/packages/c2/a7/39c4cdda9f019b609b5c473899d87abff092fc908cfe4d1ecb2fcff453b0/numpy-2.4.2-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:b2f0073ed0868db1dcd86e052d37279eef185b9c8db5bf61f30f46adac63c909", size = 17028888, upload-time = "2026-01-31T23:12:19.306Z" },
+    { url = "https://files.pythonhosted.org/packages/da/b3/e84bb64bdfea967cc10950d71090ec2d84b49bc691df0025dddb7c26e8e3/numpy-2.4.2-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:7f54844851cdb630ceb623dcec4db3240d1ac13d4990532446761baede94996a", size = 18339556, upload-time = "2026-01-31T23:12:21.816Z" },
+    { url = "https://files.pythonhosted.org/packages/88/f5/954a291bc1192a27081706862ac62bb5920fbecfbaa302f64682aa90beed/numpy-2.4.2-cp314-cp314-win32.whl", hash = "sha256:12e26134a0331d8dbd9351620f037ec470b7c75929cb8a1537f6bfe411152a1a", size = 6006899, upload-time = "2026-01-31T23:12:24.14Z" },
+    { url = "https://files.pythonhosted.org/packages/05/cb/eff72a91b2efdd1bc98b3b8759f6a1654aa87612fc86e3d87d6fe4f948c4/numpy-2.4.2-cp314-cp314-win_amd64.whl", hash = "sha256:068cdb2d0d644cdb45670810894f6a0600797a69c05f1ac478e8d31670b8ee75", size = 12443072, upload-time = "2026-01-31T23:12:26.33Z" },
+    { url = "https://files.pythonhosted.org/packages/37/75/62726948db36a56428fce4ba80a115716dc4fad6a3a4352487f8bb950966/numpy-2.4.2-cp314-cp314-win_arm64.whl", hash = "sha256:6ed0be1ee58eef41231a5c943d7d1375f093142702d5723ca2eb07db9b934b05", size = 10494886, upload-time = "2026-01-31T23:12:28.488Z" },
+    { url = "https://files.pythonhosted.org/packages/36/2f/ee93744f1e0661dc267e4b21940870cabfae187c092e1433b77b09b50ac4/numpy-2.4.2-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:98f16a80e917003a12c0580f97b5f875853ebc33e2eaa4bccfc8201ac6869308", size = 14818567, upload-time = "2026-01-31T23:12:30.709Z" },
+    { url = "https://files.pythonhosted.org/packages/a7/24/6535212add7d76ff938d8bdc654f53f88d35cddedf807a599e180dcb8e66/numpy-2.4.2-cp314-cp314t-macosx_14_0_arm64.whl", hash = "sha256:20abd069b9cda45874498b245c8015b18ace6de8546bf50dfa8cea1696ed06ef", size = 5328372, upload-time = "2026-01-31T23:12:32.962Z" },
+    { url = "https://files.pythonhosted.org/packages/5e/9d/c48f0a035725f925634bf6b8994253b43f2047f6778a54147d7e213bc5a7/numpy-2.4.2-cp314-cp314t-macosx_14_0_x86_64.whl", hash = "sha256:e98c97502435b53741540a5717a6749ac2ada901056c7db951d33e11c885cc7d", size = 6649306, upload-time = "2026-01-31T23:12:34.797Z" },
+    { url = "https://files.pythonhosted.org/packages/81/05/7c73a9574cd4a53a25907bad38b59ac83919c0ddc8234ec157f344d57d9a/numpy-2.4.2-cp314-cp314t-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:da6cad4e82cb893db4b69105c604d805e0c3ce11501a55b5e9f9083b47d2ffe8", size = 15722394, upload-time = "2026-01-31T23:12:36.565Z" },
+    { url = "https://files.pythonhosted.org/packages/35/fa/4de10089f21fc7d18442c4a767ab156b25c2a6eaf187c0db6d9ecdaeb43f/numpy-2.4.2-cp314-cp314t-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:9e4424677ce4b47fe73c8b5556d876571f7c6945d264201180db2dc34f676ab5", size = 16653343, upload-time = "2026-01-31T23:12:39.188Z" },
+    { url = "https://files.pythonhosted.org/packages/b8/f9/d33e4ffc857f3763a57aa85650f2e82486832d7492280ac21ba9efda80da/numpy-2.4.2-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:2b8f157c8a6f20eb657e240f8985cc135598b2b46985c5bccbde7616dc9c6b1e", size = 17078045, upload-time = "2026-01-31T23:12:42.041Z" },
+    { url = "https://files.pythonhosted.org/packages/c8/b8/54bdb43b6225badbea6389fa038c4ef868c44f5890f95dd530a218706da3/numpy-2.4.2-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:5daf6f3914a733336dab21a05cdec343144600e964d2fcdabaac0c0269874b2a", size = 18380024, upload-time = "2026-01-31T23:12:44.331Z" },
+    { url = "https://files.pythonhosted.org/packages/a5/55/6e1a61ded7af8df04016d81b5b02daa59f2ea9252ee0397cb9f631efe9e5/numpy-2.4.2-cp314-cp314t-win32.whl", hash = "sha256:8c50dd1fc8826f5b26a5ee4d77ca55d88a895f4e4819c7ecc2a9f5905047a443", size = 6153937, upload-time = "2026-01-31T23:12:47.229Z" },
+    { url = "https://files.pythonhosted.org/packages/45/aa/fa6118d1ed6d776b0983f3ceac9b1a5558e80df9365b1c3aa6d42bf9eee4/numpy-2.4.2-cp314-cp314t-win_amd64.whl", hash = "sha256:fcf92bee92742edd401ba41135185866f7026c502617f422eb432cfeca4fe236", size = 12631844, upload-time = "2026-01-31T23:12:48.997Z" },
+    { url = "https://files.pythonhosted.org/packages/32/0a/2ec5deea6dcd158f254a7b372fb09cfba5719419c8d66343bab35237b3fb/numpy-2.4.2-cp314-cp314t-win_arm64.whl", hash = "sha256:1f92f53998a17265194018d1cc321b2e96e900ca52d54c7c77837b71b9465181", size = 10565379, upload-time = "2026-01-31T23:12:51.345Z" },
+    { url = "https://files.pythonhosted.org/packages/f4/f8/50e14d36d915ef64d8f8bc4a087fc8264d82c785eda6711f80ab7e620335/numpy-2.4.2-pp311-pypy311_pp73-macosx_10_15_x86_64.whl", hash = "sha256:89f7268c009bc492f506abd6f5265defa7cb3f7487dc21d357c3d290add45082", size = 16833179, upload-time = "2026-01-31T23:12:53.5Z" },
+    { url = "https://files.pythonhosted.org/packages/17/17/809b5cad63812058a8189e91a1e2d55a5a18fd04611dbad244e8aeae465c/numpy-2.4.2-pp311-pypy311_pp73-macosx_11_0_arm64.whl", hash = "sha256:e6dee3bb76aa4009d5a912180bf5b2de012532998d094acee25d9cb8dee3e44a", size = 14889755, upload-time = "2026-01-31T23:12:55.933Z" },
+    { url = "https://files.pythonhosted.org/packages/3e/ea/181b9bcf7627fc8371720316c24db888dcb9829b1c0270abf3d288b2e29b/numpy-2.4.2-pp311-pypy311_pp73-macosx_14_0_arm64.whl", hash = "sha256:cd2bd2bbed13e213d6b55dc1d035a4f91748a7d3edc9480c13898b0353708920", size = 5399500, upload-time = "2026-01-31T23:12:58.671Z" },
+    { url = "https://files.pythonhosted.org/packages/33/9f/413adf3fc955541ff5536b78fcf0754680b3c6d95103230252a2c9408d23/numpy-2.4.2-pp311-pypy311_pp73-macosx_14_0_x86_64.whl", hash = "sha256:cf28c0c1d4c4bf00f509fa7eb02c58d7caf221b50b467bcb0d9bbf1584d5c821", size = 6714252, upload-time = "2026-01-31T23:13:00.518Z" },
+    { url = "https://files.pythonhosted.org/packages/91/da/643aad274e29ccbdf42ecd94dafe524b81c87bcb56b83872d54827f10543/numpy-2.4.2-pp311-pypy311_pp73-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:e04ae107ac591763a47398bb45b568fc38f02dbc4aa44c063f67a131f99346cb", size = 15797142, upload-time = "2026-01-31T23:13:02.219Z" },
+    { url = "https://files.pythonhosted.org/packages/66/27/965b8525e9cb5dc16481b30a1b3c21e50c7ebf6e9dbd48d0c4d0d5089c7e/numpy-2.4.2-pp311-pypy311_pp73-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:602f65afdef699cda27ec0b9224ae5dc43e328f4c24c689deaf77133dbee74d0", size = 16727979, upload-time = "2026-01-31T23:13:04.62Z" },
+    { url = "https://files.pythonhosted.org/packages/de/e5/b7d20451657664b07986c2f6e3be564433f5dcaf3482d68eaecd79afaf03/numpy-2.4.2-pp311-pypy311_pp73-win_amd64.whl", hash = "sha256:be71bf1edb48ebbbf7f6337b5bfd2f895d1902f6335a5830b20141fc126ffba0", size = 12502577, upload-time = "2026-01-31T23:13:07.08Z" },
+]
+
+[[package]]
+name = "opencv-contrib-python"
+version = "4.10.0.84"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
+    { name = "numpy", version = "2.4.2", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/1d/33/7b8ec6c4d45e678b26297e4a5e76464a93033a9adcc8c17eac01097065f6/opencv-contrib-python-4.10.0.84.tar.gz", hash = "sha256:4a3eae0ed9cadf1abe9293a6938a25a540e2fd6d7fc308595caa5896c8b36a0c", size = 150433857, upload-time = "2024-06-17T18:30:50.217Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/92/64/c1194510eaed272d86b53a08c790ca6ed1c450f06d401c49c8145fc46d40/opencv_contrib_python-4.10.0.84-cp37-abi3-macosx_11_0_arm64.whl", hash = "sha256:ee4b0919026d8c533aeb69b16c6ec4a891a2f6844efaa14121bf68838753209c", size = 63667391, upload-time = "2024-06-18T04:57:54.718Z" },
+    { url = "https://files.pythonhosted.org/packages/09/94/d077c4c976c2d7a88812fd55396e92edae0e0c708689dbd8c8f508920e47/opencv_contrib_python-4.10.0.84-cp37-abi3-macosx_12_0_x86_64.whl", hash = "sha256:dea80d4db73b8acccf9e16b5744bf3654f47b22745074263f0a6c10de26c5ef5", size = 66278032, upload-time = "2024-06-17T19:34:23.718Z" },
+    { url = "https://files.pythonhosted.org/packages/f8/76/f76fe74b864f3cfa737173ca12e8890aad8369e980006fb8a0b6cd14c6c7/opencv_contrib_python-4.10.0.84-cp37-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:040575b69e4f3aa761676bace4e3d1b8485fbfaf77ef77b266ab6bda5a3b5e9b", size = 47384495, upload-time = "2024-06-17T20:00:39.027Z" },
+    { url = "https://files.pythonhosted.org/packages/b0/e0/8f5d065ebb2e5941d289c5f653f944318f9e418bc5167bc6a346ab5e0f6a/opencv_contrib_python-4.10.0.84-cp37-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:a261223db41f6e512d76deaf21c8fcfb4fbbcbc2de62ca7f74a05f2c9ee489ef", size = 68681489, upload-time = "2024-06-17T18:30:32.918Z" },
+    { url = "https://files.pythonhosted.org/packages/36/30/7041bd7350cb1a26fa80415a7664b6f04f7ccbf0c12b9318d564cdf35932/opencv_contrib_python-4.10.0.84-cp37-abi3-win32.whl", hash = "sha256:2a36257ec1375d1bec2a62177ea39828ff9804de6831ee39646bdc875c343cec", size = 34506122, upload-time = "2024-06-17T18:28:29.922Z" },
+    { url = "https://files.pythonhosted.org/packages/a7/9e/7110d2c5d543ab03b9581dbb1f8e2429863e44e0c9b4960b766f230c1279/opencv_contrib_python-4.10.0.84-cp37-abi3-win_amd64.whl", hash = "sha256:47ec3160dae75f70e099b286d1a2e086d20dac8b06e759f60eaf867e6bdecba7", size = 45541421, upload-time = "2024-06-17T18:28:46.012Z" },
+]
+
+[[package]]
+name = "orjson"
+version = "3.11.6"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/70/a3/4e09c61a5f0c521cba0bb433639610ae037437669f1a4cbc93799e731d78/orjson-3.11.6.tar.gz", hash = "sha256:0a54c72259f35299fd033042367df781c2f66d10252955ca1efb7db309b954cb", size = 6175856, upload-time = "2026-01-29T15:13:07.942Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/30/3c/098ed0e49c565fdf1ccc6a75b190115d1ca74148bf5b6ab036554a550650/orjson-3.11.6-cp310-cp310-macosx_10_15_x86_64.macosx_11_0_arm64.macosx_10_15_universal2.whl", hash = "sha256:a613fc37e007143d5b6286dccb1394cd114b07832417006a02b620ddd8279e37", size = 250411, upload-time = "2026-01-29T15:11:17.941Z" },
+    { url = "https://files.pythonhosted.org/packages/15/7c/cb11a360fd228ceebade03b1e8e9e138dd4b1b3b11602b72dbdad915aded/orjson-3.11.6-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:46ebee78f709d3ba7a65384cfe285bb0763157c6d2f836e7bde2f12d33a867a2", size = 138147, upload-time = "2026-01-29T15:11:19.659Z" },
+    { url = "https://files.pythonhosted.org/packages/4e/4b/e57b5c45ffe69fbef7cbd56e9f40e2dc0d5de920caafefcc6981d1a7efc5/orjson-3.11.6-cp310-cp310-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:a726fa86d2368cd57990f2bd95ef5495a6e613b08fc9585dfe121ec758fb08d1", size = 135110, upload-time = "2026-01-29T15:11:21.231Z" },
+    { url = "https://files.pythonhosted.org/packages/b0/6e/4f21c6256f8cee3c0c69926cf7ac821cfc36f218512eedea2e2dc4a490c8/orjson-3.11.6-cp310-cp310-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:150f12e59d6864197770c78126e1a6e07a3da73d1728731bf3bc1e8b96ffdbe6", size = 140995, upload-time = "2026-01-29T15:11:22.902Z" },
+    { url = "https://files.pythonhosted.org/packages/d0/78/92c36205ba2f6094ba1eea60c8e646885072abe64f155196833988c14b74/orjson-3.11.6-cp310-cp310-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:9a2d9746a5b5ce20c0908ada451eb56da4ffa01552a50789a0354d8636a02953", size = 144435, upload-time = "2026-01-29T15:11:24.124Z" },
+    { url = "https://files.pythonhosted.org/packages/4d/52/1b518d164005811eb3fea92650e76e7d9deadb0b41e92c483373b1e82863/orjson-3.11.6-cp310-cp310-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:afd177f5dd91666d31e9019f1b06d2fcdf8a409a1637ddcb5915085dede85680", size = 142734, upload-time = "2026-01-29T15:11:25.708Z" },
+    { url = "https://files.pythonhosted.org/packages/4b/11/60ea7885a2b7c1bf60ed8b5982356078a73785bd3bab392041a5bcf8de7c/orjson-3.11.6-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:8d777ec41a327bd3b7de97ba7bce12cc1007815ca398e4e4de9ec56c022c090b", size = 145802, upload-time = "2026-01-29T15:11:26.917Z" },
+    { url = "https://files.pythonhosted.org/packages/41/7f/15a927e7958fd4f7560fb6dbb9346bee44a168e40168093c46020d866098/orjson-3.11.6-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:f3a135f83185c87c13ff231fcb7dbb2fa4332a376444bd65135b50ff4cc5265c", size = 147504, upload-time = "2026-01-29T15:11:28.07Z" },
+    { url = "https://files.pythonhosted.org/packages/66/1f/cabb9132a533f4f913e29294d0a1ca818b1a9a52e990526fe3f7ddd75f1c/orjson-3.11.6-cp310-cp310-musllinux_1_2_armv7l.whl", hash = "sha256:2a8eeed7d4544cf391a142b0dd06029dac588e96cc692d9ab1c3f05b1e57c7f6", size = 421408, upload-time = "2026-01-29T15:11:29.314Z" },
+    { url = "https://files.pythonhosted.org/packages/4c/b9/09bda9257a982e300313e4a9fc9b9c3aaff424d07bcf765bf045e4e3ed03/orjson-3.11.6-cp310-cp310-musllinux_1_2_i686.whl", hash = "sha256:9d576865a21e5cc6695be8fb78afc812079fd361ce6a027a7d41561b61b33a90", size = 155801, upload-time = "2026-01-29T15:11:30.575Z" },
+    { url = "https://files.pythonhosted.org/packages/98/19/4e40ea3e5f4c6a8d51f31fd2382351ee7b396fecca915b17cd1af588175b/orjson-3.11.6-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:925e2df51f60aa50f8797830f2adfc05330425803f4105875bb511ced98b7f89", size = 147647, upload-time = "2026-01-29T15:11:31.856Z" },
+    { url = "https://files.pythonhosted.org/packages/5a/73/ef4bd7dd15042cf33a402d16b87b9e969e71edb452b63b6e2b05025d1f7d/orjson-3.11.6-cp310-cp310-win32.whl", hash = "sha256:09dded2de64e77ac0b312ad59f35023548fb87393a57447e1bb36a26c181a90f", size = 139770, upload-time = "2026-01-29T15:11:33.031Z" },
+    { url = "https://files.pythonhosted.org/packages/b4/ac/daab6e10467f7fffd7081ba587b492505b49313130ff5446a6fe28bf076e/orjson-3.11.6-cp310-cp310-win_amd64.whl", hash = "sha256:3a63b5e7841ca8635214c6be7c0bf0246aa8c5cd4ef0c419b14362d0b2fb13de", size = 136783, upload-time = "2026-01-29T15:11:34.686Z" },
+    { url = "https://files.pythonhosted.org/packages/f3/fd/d6b0a36854179b93ed77839f107c4089d91cccc9f9ba1b752b6e3bac5f34/orjson-3.11.6-cp311-cp311-macosx_10_15_x86_64.macosx_11_0_arm64.macosx_10_15_universal2.whl", hash = "sha256:e259e85a81d76d9665f03d6129e09e4435531870de5961ddcd0bf6e3a7fde7d7", size = 250029, upload-time = "2026-01-29T15:11:35.942Z" },
+    { url = "https://files.pythonhosted.org/packages/a3/bb/22902619826641cf3b627c24aab62e2ad6b571bdd1d34733abb0dd57f67a/orjson-3.11.6-cp311-cp311-macosx_15_0_arm64.whl", hash = "sha256:52263949f41b4a4822c6b1353bcc5ee2f7109d53a3b493501d3369d6d0e7937a", size = 134518, upload-time = "2026-01-29T15:11:37.347Z" },
+    { url = "https://files.pythonhosted.org/packages/72/90/7a818da4bba1de711a9653c420749c0ac95ef8f8651cbc1dca551f462fe0/orjson-3.11.6-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:6439e742fa7834a24698d358a27346bb203bff356ae0402e7f5df8f749c621a8", size = 137917, upload-time = "2026-01-29T15:11:38.511Z" },
+    { url = "https://files.pythonhosted.org/packages/59/0f/02846c1cac8e205cb3822dd8aa8f9114acda216f41fd1999ace6b543418d/orjson-3.11.6-cp311-cp311-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:b81ffd68f084b4e993e3867acb554a049fa7787cc8710bbcc1e26965580d99be", size = 134923, upload-time = "2026-01-29T15:11:39.711Z" },
+    { url = "https://files.pythonhosted.org/packages/94/cf/aeaf683001b474bb3c3c757073a4231dfdfe8467fceaefa5bfd40902c99f/orjson-3.11.6-cp311-cp311-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:a5a5468e5e60f7ef6d7f9044b06c8f94a3c56ba528c6e4f7f06ae95164b595ec", size = 140752, upload-time = "2026-01-29T15:11:41.347Z" },
+    { url = "https://files.pythonhosted.org/packages/fc/fe/dad52d8315a65f084044a0819d74c4c9daf9ebe0681d30f525b0d29a31f0/orjson-3.11.6-cp311-cp311-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:72c5005eb45bd2535632d4f3bec7ad392832cfc46b62a3021da3b48a67734b45", size = 144201, upload-time = "2026-01-29T15:11:42.537Z" },
+    { url = "https://files.pythonhosted.org/packages/36/bc/ab070dd421565b831801077f1e390c4d4af8bfcecafc110336680a33866b/orjson-3.11.6-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:0b14dd49f3462b014455a28a4d810d3549bf990567653eb43765cd847df09145", size = 142380, upload-time = "2026-01-29T15:11:44.309Z" },
+    { url = "https://files.pythonhosted.org/packages/e6/d8/4b581c725c3a308717f28bf45a9fdac210bca08b67e8430143699413ff06/orjson-3.11.6-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:6e0bb2c1ea30ef302f0f89f9bf3e7f9ab5e2af29dc9f80eb87aa99788e4e2d65", size = 145582, upload-time = "2026-01-29T15:11:45.506Z" },
+    { url = "https://files.pythonhosted.org/packages/5b/a2/09aab99b39f9a7f175ea8fa29adb9933a3d01e7d5d603cdee7f1c40c8da2/orjson-3.11.6-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:825e0a85d189533c6bff7e2fc417a28f6fcea53d27125c4551979aecd6c9a197", size = 147270, upload-time = "2026-01-29T15:11:46.782Z" },
+    { url = "https://files.pythonhosted.org/packages/b8/2f/5ef8eaf7829dc50da3bf497c7775b21ee88437bc8c41f959aa3504ca6631/orjson-3.11.6-cp311-cp311-musllinux_1_2_armv7l.whl", hash = "sha256:b04575417a26530637f6ab4b1f7b4f666eb0433491091da4de38611f97f2fcf3", size = 421222, upload-time = "2026-01-29T15:11:48.106Z" },
+    { url = "https://files.pythonhosted.org/packages/3b/b0/dd6b941294c2b5b13da5fdc7e749e58d0c55a5114ab37497155e83050e95/orjson-3.11.6-cp311-cp311-musllinux_1_2_i686.whl", hash = "sha256:b83eb2e40e8c4da6d6b340ee6b1d6125f5195eb1b0ebb7eac23c6d9d4f92d224", size = 155562, upload-time = "2026-01-29T15:11:49.408Z" },
+    { url = "https://files.pythonhosted.org/packages/8e/09/43924331a847476ae2f9a16bd6d3c9dab301265006212ba0d3d7fd58763a/orjson-3.11.6-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:1f42da604ee65a6b87eef858c913ce3e5777872b19321d11e6fc6d21de89b64f", size = 147432, upload-time = "2026-01-29T15:11:50.635Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/e9/d9865961081816909f6b49d880749dbbd88425afd7c5bbce0549e2290d77/orjson-3.11.6-cp311-cp311-win32.whl", hash = "sha256:5ae45df804f2d344cffb36c43fdf03c82fb6cd247f5faa41e21891b40dfbf733", size = 139623, upload-time = "2026-01-29T15:11:51.82Z" },
+    { url = "https://files.pythonhosted.org/packages/b4/f9/6836edb92f76eec1082919101eb1145d2f9c33c8f2c5e6fa399b82a2aaa8/orjson-3.11.6-cp311-cp311-win_amd64.whl", hash = "sha256:f4295948d65ace0a2d8f2c4ccc429668b7eb8af547578ec882e16bf79b0050b2", size = 136647, upload-time = "2026-01-29T15:11:53.454Z" },
+    { url = "https://files.pythonhosted.org/packages/b3/0c/4954082eea948c9ae52ee0bcbaa2f99da3216a71bcc314ab129bde22e565/orjson-3.11.6-cp311-cp311-win_arm64.whl", hash = "sha256:314e9c45e0b81b547e3a1cfa3df3e07a815821b3dac9fe8cb75014071d0c16a4", size = 135327, upload-time = "2026-01-29T15:11:56.616Z" },
+    { url = "https://files.pythonhosted.org/packages/14/ba/759f2879f41910b7e5e0cdbd9cf82a4f017c527fb0e972e9869ca7fe4c8e/orjson-3.11.6-cp312-cp312-macosx_10_15_x86_64.macosx_11_0_arm64.macosx_10_15_universal2.whl", hash = "sha256:6f03f30cd8953f75f2a439070c743c7336d10ee940da918d71c6f3556af3ddcf", size = 249988, upload-time = "2026-01-29T15:11:58.294Z" },
+    { url = "https://files.pythonhosted.org/packages/f0/70/54cecb929e6c8b10104fcf580b0cc7dc551aa193e83787dd6f3daba28bb5/orjson-3.11.6-cp312-cp312-macosx_15_0_arm64.whl", hash = "sha256:af44baae65ef386ad971469a8557a0673bb042b0b9fd4397becd9c2dfaa02588", size = 134445, upload-time = "2026-01-29T15:11:59.819Z" },
+    { url = "https://files.pythonhosted.org/packages/f2/6f/ec0309154457b9ba1ad05f11faa4441f76037152f75e1ac577db3ce7ca96/orjson-3.11.6-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:c310a48542094e4f7dbb6ac076880994986dda8ca9186a58c3cb70a3514d3231", size = 137708, upload-time = "2026-01-29T15:12:01.488Z" },
+    { url = "https://files.pythonhosted.org/packages/20/52/3c71b80840f8bab9cb26417302707b7716b7d25f863f3a541bcfa232fe6e/orjson-3.11.6-cp312-cp312-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:d8dfa7a5d387f15ecad94cb6b2d2d5f4aeea64efd8d526bfc03c9812d01e1cc0", size = 134798, upload-time = "2026-01-29T15:12:02.705Z" },
+    { url = "https://files.pythonhosted.org/packages/30/51/b490a43b22ff736282360bd02e6bded455cf31dfc3224e01cd39f919bbd2/orjson-3.11.6-cp312-cp312-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:ba8daee3e999411b50f8b50dbb0a3071dd1845f3f9a1a0a6fa6de86d1689d84d", size = 140839, upload-time = "2026-01-29T15:12:03.956Z" },
+    { url = "https://files.pythonhosted.org/packages/95/bc/4bcfe4280c1bc63c5291bb96f98298845b6355da2226d3400e17e7b51e53/orjson-3.11.6-cp312-cp312-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:f89d104c974eafd7436d7a5fdbc57f7a1e776789959a2f4f1b2eab5c62a339f4", size = 144080, upload-time = "2026-01-29T15:12:05.151Z" },
+    { url = "https://files.pythonhosted.org/packages/01/74/22970f9ead9ab1f1b5f8c227a6c3aa8d71cd2c5acd005868a1d44f2362fa/orjson-3.11.6-cp312-cp312-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:b2e2e2456788ca5ea75616c40da06fc885a7dc0389780e8a41bf7c5389ba257b", size = 142435, upload-time = "2026-01-29T15:12:06.641Z" },
+    { url = "https://files.pythonhosted.org/packages/29/34/d564aff85847ab92c82ee43a7a203683566c2fca0723a5f50aebbe759603/orjson-3.11.6-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:2a42efebc45afabb1448001e90458c4020d5c64fbac8a8dc4045b777db76cb5a", size = 145631, upload-time = "2026-01-29T15:12:08.351Z" },
+    { url = "https://files.pythonhosted.org/packages/e7/ef/016957a3890752c4aa2368326ea69fa53cdc1fdae0a94a542b6410dbdf52/orjson-3.11.6-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:71b7cbef8471324966c3738c90ba38775563ef01b512feb5ad4805682188d1b9", size = 147058, upload-time = "2026-01-29T15:12:10.023Z" },
+    { url = "https://files.pythonhosted.org/packages/56/cc/9a899c3972085645b3225569f91a30e221f441e5dc8126e6d060b971c252/orjson-3.11.6-cp312-cp312-musllinux_1_2_armv7l.whl", hash = "sha256:f8515e5910f454fe9a8e13c2bb9dc4bae4c1836313e967e72eb8a4ad874f0248", size = 421161, upload-time = "2026-01-29T15:12:11.308Z" },
+    { url = "https://files.pythonhosted.org/packages/21/a8/767d3fbd6d9b8fdee76974db40619399355fd49bf91a6dd2c4b6909ccf05/orjson-3.11.6-cp312-cp312-musllinux_1_2_i686.whl", hash = "sha256:300360edf27c8c9bf7047345a94fddf3a8b8922df0ff69d71d854a170cb375cf", size = 155757, upload-time = "2026-01-29T15:12:12.776Z" },
+    { url = "https://files.pythonhosted.org/packages/ad/0b/205cd69ac87e2272e13ef3f5f03a3d4657e317e38c1b08aaa2ef97060bbc/orjson-3.11.6-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:caaed4dad39e271adfadc106fab634d173b2bb23d9cf7e67bd645f879175ebfc", size = 147446, upload-time = "2026-01-29T15:12:14.166Z" },
+    { url = "https://files.pythonhosted.org/packages/de/c5/dd9f22aa9f27c54c7d05cc32f4580c9ac9b6f13811eeb81d6c4c3f50d6b1/orjson-3.11.6-cp312-cp312-win32.whl", hash = "sha256:955368c11808c89793e847830e1b1007503a5923ddadc108547d3b77df761044", size = 139717, upload-time = "2026-01-29T15:12:15.7Z" },
+    { url = "https://files.pythonhosted.org/packages/23/a1/e62fc50d904486970315a1654b8cfb5832eb46abb18cd5405118e7e1fc79/orjson-3.11.6-cp312-cp312-win_amd64.whl", hash = "sha256:2c68de30131481150073d90a5d227a4a421982f42c025ecdfb66157f9579e06f", size = 136711, upload-time = "2026-01-29T15:12:17.055Z" },
+    { url = "https://files.pythonhosted.org/packages/04/3d/b4fefad8bdf91e0fe212eb04975aeb36ea92997269d68857efcc7eb1dda3/orjson-3.11.6-cp312-cp312-win_arm64.whl", hash = "sha256:65dfa096f4e3a5e02834b681f539a87fbe85adc82001383c0db907557f666bfc", size = 135212, upload-time = "2026-01-29T15:12:18.3Z" },
+    { url = "https://files.pythonhosted.org/packages/ae/45/d9c71c8c321277bc1ceebf599bc55ba826ae538b7c61f287e9a7e71bd589/orjson-3.11.6-cp313-cp313-macosx_10_15_x86_64.macosx_11_0_arm64.macosx_10_15_universal2.whl", hash = "sha256:e4ae1670caabb598a88d385798692ce2a1b2f078971b3329cfb85253c6097f5b", size = 249828, upload-time = "2026-01-29T15:12:20.14Z" },
+    { url = "https://files.pythonhosted.org/packages/ac/7e/4afcf4cfa9c2f93846d70eee9c53c3c0123286edcbeb530b7e9bd2aea1b2/orjson-3.11.6-cp313-cp313-macosx_15_0_arm64.whl", hash = "sha256:2c6b81f47b13dac2caa5d20fbc953c75eb802543abf48403a4703ed3bff225f0", size = 134339, upload-time = "2026-01-29T15:12:22.01Z" },
+    { url = "https://files.pythonhosted.org/packages/40/10/6d2b8a064c8d2411d3d0ea6ab43125fae70152aef6bea77bb50fa54d4097/orjson-3.11.6-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:647d6d034e463764e86670644bdcaf8e68b076e6e74783383b01085ae9ab334f", size = 137662, upload-time = "2026-01-29T15:12:23.307Z" },
+    { url = "https://files.pythonhosted.org/packages/5a/50/5804ea7d586baf83ee88969eefda97a24f9a5bdba0727f73e16305175b26/orjson-3.11.6-cp313-cp313-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:8523b9cc4ef174ae52414f7699e95ee657c16aa18b3c3c285d48d7966cce9081", size = 134626, upload-time = "2026-01-29T15:12:25.099Z" },
+    { url = "https://files.pythonhosted.org/packages/9e/2e/f0492ed43e376722bb4afd648e06cc1e627fc7ec8ff55f6ee739277813ea/orjson-3.11.6-cp313-cp313-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:313dfd7184cde50c733fc0d5c8c0e2f09017b573afd11dc36bd7476b30b4cb17", size = 140873, upload-time = "2026-01-29T15:12:26.369Z" },
+    { url = "https://files.pythonhosted.org/packages/10/15/6f874857463421794a303a39ac5494786ad46a4ab46d92bda6705d78c5aa/orjson-3.11.6-cp313-cp313-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:905ee036064ff1e1fd1fb800055ac477cdcb547a78c22c1bc2bbf8d5d1a6fb42", size = 144044, upload-time = "2026-01-29T15:12:28.082Z" },
+    { url = "https://files.pythonhosted.org/packages/d2/c7/b7223a3a70f1d0cc2d86953825de45f33877ee1b124a91ca1f79aa6e643f/orjson-3.11.6-cp313-cp313-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:ce374cb98411356ba906914441fc993f271a7a666d838d8de0e0900dd4a4bc12", size = 142396, upload-time = "2026-01-29T15:12:30.529Z" },
+    { url = "https://files.pythonhosted.org/packages/87/e3/aa1b6d3ad3cd80f10394134f73ae92a1d11fdbe974c34aa199cc18bb5fcf/orjson-3.11.6-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:cded072b9f65fcfd188aead45efa5bd528ba552add619b3ad2a81f67400ec450", size = 145600, upload-time = "2026-01-29T15:12:31.848Z" },
+    { url = "https://files.pythonhosted.org/packages/f6/cf/e4aac5a46cbd39d7e769ef8650efa851dfce22df1ba97ae2b33efe893b12/orjson-3.11.6-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:7ab85bdbc138e1f73a234db6bb2e4cc1f0fcec8f4bd2bd2430e957a01aadf746", size = 146967, upload-time = "2026-01-29T15:12:33.203Z" },
+    { url = "https://files.pythonhosted.org/packages/0b/04/975b86a4bcf6cfeda47aad15956d52fbeda280811206e9967380fa9355c8/orjson-3.11.6-cp313-cp313-musllinux_1_2_armv7l.whl", hash = "sha256:351b96b614e3c37a27b8ab048239ebc1e0be76cc17481a430d70a77fb95d3844", size = 421003, upload-time = "2026-01-29T15:12:35.097Z" },
+    { url = "https://files.pythonhosted.org/packages/28/d1/0369d0baf40eea5ff2300cebfe209883b2473ab4aa4c4974c8bd5ee42bb2/orjson-3.11.6-cp313-cp313-musllinux_1_2_i686.whl", hash = "sha256:f9959c85576beae5cdcaaf39510b15105f1ee8b70d5dacd90152617f57be8c83", size = 155695, upload-time = "2026-01-29T15:12:36.589Z" },
+    { url = "https://files.pythonhosted.org/packages/ab/1f/d10c6d6ae26ff1d7c3eea6fd048280ef2e796d4fb260c5424fd021f68ecf/orjson-3.11.6-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:75682d62b1b16b61a30716d7a2ec1f4c36195de4a1c61f6665aedd947b93a5d5", size = 147392, upload-time = "2026-01-29T15:12:37.876Z" },
+    { url = "https://files.pythonhosted.org/packages/8d/43/7479921c174441a0aa5277c313732e20713c0969ac303be9f03d88d3db5d/orjson-3.11.6-cp313-cp313-win32.whl", hash = "sha256:40dc277999c2ef227dcc13072be879b4cfd325502daeb5c35ed768f706f2bf30", size = 139718, upload-time = "2026-01-29T15:12:39.274Z" },
+    { url = "https://files.pythonhosted.org/packages/88/bc/9ffe7dfbf8454bc4e75bb8bf3a405ed9e0598df1d3535bb4adcd46be07d0/orjson-3.11.6-cp313-cp313-win_amd64.whl", hash = "sha256:f0f6e9f8ff7905660bc3c8a54cd4a675aa98f7f175cf00a59815e2ff42c0d916", size = 136635, upload-time = "2026-01-29T15:12:40.593Z" },
+    { url = "https://files.pythonhosted.org/packages/6f/7e/51fa90b451470447ea5023b20d83331ec741ae28d1e6d8ed547c24e7de14/orjson-3.11.6-cp313-cp313-win_arm64.whl", hash = "sha256:1608999478664de848e5900ce41f25c4ecdfc4beacbc632b6fd55e1a586e5d38", size = 135175, upload-time = "2026-01-29T15:12:41.997Z" },
+    { url = "https://files.pythonhosted.org/packages/31/9f/46ca908abaeeec7560638ff20276ab327b980d73b3cc2f5b205b4a1c60b3/orjson-3.11.6-cp314-cp314-macosx_10_15_x86_64.macosx_11_0_arm64.macosx_10_15_universal2.whl", hash = "sha256:6026db2692041d2a23fe2545606df591687787825ad5821971ef0974f2c47630", size = 249823, upload-time = "2026-01-29T15:12:43.332Z" },
+    { url = "https://files.pythonhosted.org/packages/ff/78/ca478089818d18c9cd04f79c43f74ddd031b63c70fa2a946eb5e85414623/orjson-3.11.6-cp314-cp314-macosx_15_0_arm64.whl", hash = "sha256:132b0ab2e20c73afa85cf142e547511feb3d2f5b7943468984658f3952b467d4", size = 134328, upload-time = "2026-01-29T15:12:45.171Z" },
+    { url = "https://files.pythonhosted.org/packages/39/5e/cbb9d830ed4e47f4375ad8eef8e4fff1bf1328437732c3809054fc4e80be/orjson-3.11.6-cp314-cp314-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:b376fb05f20a96ec117d47987dd3b39265c635725bda40661b4c5b73b77b5fde", size = 137651, upload-time = "2026-01-29T15:12:46.602Z" },
+    { url = "https://files.pythonhosted.org/packages/7c/3a/35df6558c5bc3a65ce0961aefee7f8364e59af78749fc796ea255bfa0cf5/orjson-3.11.6-cp314-cp314-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:954dae4e080574672a1dfcf2a840eddef0f27bd89b0e94903dd0824e9c1db060", size = 134596, upload-time = "2026-01-29T15:12:47.95Z" },
+    { url = "https://files.pythonhosted.org/packages/cd/8e/3d32dd7b7f26a19cc4512d6ed0ae3429567c71feef720fe699ff43c5bc9e/orjson-3.11.6-cp314-cp314-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:fe515bb89d59e1e4b48637a964f480b35c0a2676de24e65e55310f6016cca7ce", size = 140923, upload-time = "2026-01-29T15:12:49.333Z" },
+    { url = "https://files.pythonhosted.org/packages/6c/9c/1efbf5c99b3304f25d6f0d493a8d1492ee98693637c10ce65d57be839d7b/orjson-3.11.6-cp314-cp314-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:380f9709c275917af28feb086813923251e11ee10687257cd7f1ea188bcd4485", size = 144068, upload-time = "2026-01-29T15:12:50.927Z" },
+    { url = "https://files.pythonhosted.org/packages/82/83/0d19eeb5be797de217303bbb55dde58dba26f996ed905d301d98fd2d4637/orjson-3.11.6-cp314-cp314-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:a8173e0d3f6081e7034c51cf984036d02f6bab2a2126de5a759d79f8e5a140e7", size = 142493, upload-time = "2026-01-29T15:12:52.432Z" },
+    { url = "https://files.pythonhosted.org/packages/32/a7/573fec3df4dc8fc259b7770dc6c0656f91adce6e19330c78d23f87945d1e/orjson-3.11.6-cp314-cp314-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:6dddf9ba706294906c56ef5150a958317b09aa3a8a48df1c52ccf22ec1907eac", size = 145616, upload-time = "2026-01-29T15:12:53.903Z" },
+    { url = "https://files.pythonhosted.org/packages/c2/0e/23551b16f21690f7fd5122e3cf40fdca5d77052a434d0071990f97f5fe2f/orjson-3.11.6-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:cbae5c34588dc79938dffb0b6fbe8c531f4dc8a6ad7f39759a9eb5d2da405ef2", size = 146951, upload-time = "2026-01-29T15:12:55.698Z" },
+    { url = "https://files.pythonhosted.org/packages/b8/63/5e6c8f39805c39123a18e412434ea364349ee0012548d08aa586e2bd6aa9/orjson-3.11.6-cp314-cp314-musllinux_1_2_armv7l.whl", hash = "sha256:f75c318640acbddc419733b57f8a07515e587a939d8f54363654041fd1f4e465", size = 421024, upload-time = "2026-01-29T15:12:57.434Z" },
+    { url = "https://files.pythonhosted.org/packages/1d/4d/724975cf0087f6550bd01fd62203418afc0ea33fd099aed318c5bcc52df8/orjson-3.11.6-cp314-cp314-musllinux_1_2_i686.whl", hash = "sha256:e0ab8d13aa2a3e98b4a43487c9205b2c92c38c054b4237777484d503357c8437", size = 155774, upload-time = "2026-01-29T15:12:59.397Z" },
+    { url = "https://files.pythonhosted.org/packages/a8/a3/f4c4e3f46b55db29e0a5f20493b924fc791092d9a03ff2068c9fe6c1002f/orjson-3.11.6-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:f884c7fb1020d44612bd7ac0db0babba0e2f78b68d9a650c7959bf99c783773f", size = 147393, upload-time = "2026-01-29T15:13:00.769Z" },
+    { url = "https://files.pythonhosted.org/packages/ee/86/6f5529dd27230966171ee126cecb237ed08e9f05f6102bfaf63e5b32277d/orjson-3.11.6-cp314-cp314-win32.whl", hash = "sha256:8d1035d1b25732ec9f971e833a3e299d2b1a330236f75e6fd945ad982c76aaf3", size = 139760, upload-time = "2026-01-29T15:13:02.173Z" },
+    { url = "https://files.pythonhosted.org/packages/d3/b5/91ae7037b2894a6b5002fb33f4fbccec98424a928469835c3837fbb22a9b/orjson-3.11.6-cp314-cp314-win_amd64.whl", hash = "sha256:931607a8865d21682bb72de54231655c86df1870502d2962dbfd12c82890d077", size = 136633, upload-time = "2026-01-29T15:13:04.267Z" },
+    { url = "https://files.pythonhosted.org/packages/55/74/f473a3ec7a0a7ebc825ca8e3c86763f7d039f379860c81ba12dcdd456547/orjson-3.11.6-cp314-cp314-win_arm64.whl", hash = "sha256:fe71f6b283f4f1832204ab8235ce07adad145052614f77c876fcf0dac97bc06f", size = 135168, upload-time = "2026-01-29T15:13:05.932Z" },
+]
+
+[[package]]
+name = "packaging"
+version = "25.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/a1/d4/1fc4078c65507b51b96ca8f8c3ba19e6a61c8253c72794544580a7b6c24d/packaging-25.0.tar.gz", hash = "sha256:d443872c98d677bf60f6a1f2f8c1cb748e8fe762d2bf9d3148b5599295b0fc4f", size = 165727, upload-time = "2025-04-19T11:48:59.673Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/20/12/38679034af332785aac8774540895e234f4d07f7545804097de4b666afd8/packaging-25.0-py3-none-any.whl", hash = "sha256:29572ef2b1f17581046b3a2227d5c611fb25ec70ca1ba8554b24b0e69331a484", size = 66469, upload-time = "2025-04-19T11:48:57.875Z" },
+]
+
+[[package]]
+name = "paddleocr"
+version = "3.4.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "paddlex", extra = ["ocr-core"] },
+    { name = "pyyaml" },
+    { name = "requests" },
+    { name = "typing-extensions" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/b2/78/15f43c2fd98d12320f9d771df27beaf49464fd151bf8dffd58ede0a4f019/paddleocr-3.4.0.tar.gz", hash = "sha256:8002a03ae88b5cd9366b7969897e6f76ae23a45a6314f4f6da35c081a979bb54", size = 2504177, upload-time = "2026-01-29T10:15:25.251Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/32/25/75a7aa8409d9a6ece6bd7ff9a896a25b75c5a78b2948d2119ab7a99db2f2/paddleocr-3.4.0-py3-none-any.whl", hash = "sha256:67b28a98c8ce58668473702fa22a6053d959df1105ec430a578280356b889ba1", size = 87786, upload-time = "2026-01-29T10:15:22.356Z" },
+]
+
+[[package]]
+name = "paddlex"
+version = "3.4.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "aistudio-sdk" },
+    { name = "chardet" },
+    { name = "colorlog" },
+    { name = "filelock" },
+    { name = "huggingface-hub" },
+    { name = "modelscope" },
+    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
+    { name = "numpy", version = "2.4.2", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
+    { name = "packaging" },
+    { name = "pandas", version = "2.3.3", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
+    { name = "pandas", version = "3.0.0", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
+    { name = "pillow" },
+    { name = "prettytable" },
+    { name = "py-cpuinfo" },
+    { name = "pydantic" },
+    { name = "pyyaml" },
+    { name = "requests" },
+    { name = "ruamel-yaml" },
+    { name = "typing-extensions" },
+    { name = "ujson" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/22/bc/1cbf54cb3692311bf243ec24cc07998a574976479f6edbf3a61b33367480/paddlex-3.4.1.tar.gz", hash = "sha256:6746d07e12326e1cbba4fb170856867e25d07f2cfc85d36b7e617f0164eb3cab", size = 3994513, upload-time = "2026-01-30T06:51:40.933Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/26/c4/44b2ebd14b58915ef3ae30409cb550ef16da30a0845a1ecab53c44978089/paddlex-3.4.1-py3-none-any.whl", hash = "sha256:901f64954a2d1d0484d5fb922fa74a3b088eb7611e245a9f1a756ea7ff104806", size = 2005787, upload-time = "2026-01-30T06:51:35.099Z" },
+]
+
+[package.optional-dependencies]
+ocr-core = [
+    { name = "imagesize" },
+    { name = "opencv-contrib-python" },
+    { name = "pyclipper" },
+    { name = "pypdfium2" },
+    { name = "python-bidi" },
+    { name = "shapely" },
+]
+serving = [
+    { name = "aiohttp" },
+    { name = "bce-python-sdk" },
+    { name = "fastapi" },
+    { name = "filetype" },
+    { name = "opencv-contrib-python" },
+    { name = "pypdfium2" },
+    { name = "starlette" },
+    { name = "uvicorn" },
+    { name = "yarl" },
+]
+
+[[package]]
+name = "pandas"
+version = "2.3.3"
+source = { registry = "https://pypi.org/simple" }
+resolution-markers = [
+    "python_full_version < '3.11' and sys_platform == 'darwin'",
+    "python_full_version < '3.11' and platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "(python_full_version < '3.11' and platform_machine != 'aarch64' and sys_platform == 'linux') or (python_full_version < '3.11' and sys_platform != 'darwin' and sys_platform != 'linux')",
+]
+dependencies = [
+    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
+    { name = "python-dateutil", marker = "python_full_version < '3.11'" },
+    { name = "pytz", marker = "python_full_version < '3.11'" },
+    { name = "tzdata", marker = "python_full_version < '3.11'" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/33/01/d40b85317f86cf08d853a4f495195c73815fdf205eef3993821720274518/pandas-2.3.3.tar.gz", hash = "sha256:e05e1af93b977f7eafa636d043f9f94c7ee3ac81af99c13508215942e64c993b", size = 4495223, upload-time = "2025-09-29T23:34:51.853Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/3d/f7/f425a00df4fcc22b292c6895c6831c0c8ae1d9fac1e024d16f98a9ce8749/pandas-2.3.3-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:376c6446ae31770764215a6c937f72d917f214b43560603cd60da6408f183b6c", size = 11555763, upload-time = "2025-09-29T23:16:53.287Z" },
+    { url = "https://files.pythonhosted.org/packages/13/4f/66d99628ff8ce7857aca52fed8f0066ce209f96be2fede6cef9f84e8d04f/pandas-2.3.3-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:e19d192383eab2f4ceb30b412b22ea30690c9e618f78870357ae1d682912015a", size = 10801217, upload-time = "2025-09-29T23:17:04.522Z" },
+    { url = "https://files.pythonhosted.org/packages/1d/03/3fc4a529a7710f890a239cc496fc6d50ad4a0995657dccc1d64695adb9f4/pandas-2.3.3-cp310-cp310-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:5caf26f64126b6c7aec964f74266f435afef1c1b13da3b0636c7518a1fa3e2b1", size = 12148791, upload-time = "2025-09-29T23:17:18.444Z" },
+    { url = "https://files.pythonhosted.org/packages/40/a8/4dac1f8f8235e5d25b9955d02ff6f29396191d4e665d71122c3722ca83c5/pandas-2.3.3-cp310-cp310-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:dd7478f1463441ae4ca7308a70e90b33470fa593429f9d4c578dd00d1fa78838", size = 12769373, upload-time = "2025-09-29T23:17:35.846Z" },
+    { url = "https://files.pythonhosted.org/packages/df/91/82cc5169b6b25440a7fc0ef3a694582418d875c8e3ebf796a6d6470aa578/pandas-2.3.3-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:4793891684806ae50d1288c9bae9330293ab4e083ccd1c5e383c34549c6e4250", size = 13200444, upload-time = "2025-09-29T23:17:49.341Z" },
+    { url = "https://files.pythonhosted.org/packages/10/ae/89b3283800ab58f7af2952704078555fa60c807fff764395bb57ea0b0dbd/pandas-2.3.3-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:28083c648d9a99a5dd035ec125d42439c6c1c525098c58af0fc38dd1a7a1b3d4", size = 13858459, upload-time = "2025-09-29T23:18:03.722Z" },
+    { url = "https://files.pythonhosted.org/packages/85/72/530900610650f54a35a19476eca5104f38555afccda1aa11a92ee14cb21d/pandas-2.3.3-cp310-cp310-win_amd64.whl", hash = "sha256:503cf027cf9940d2ceaa1a93cfb5f8c8c7e6e90720a2850378f0b3f3b1e06826", size = 11346086, upload-time = "2025-09-29T23:18:18.505Z" },
+    { url = "https://files.pythonhosted.org/packages/c1/fa/7ac648108144a095b4fb6aa3de1954689f7af60a14cf25583f4960ecb878/pandas-2.3.3-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:602b8615ebcc4a0c1751e71840428ddebeb142ec02c786e8ad6b1ce3c8dec523", size = 11578790, upload-time = "2025-09-29T23:18:30.065Z" },
+    { url = "https://files.pythonhosted.org/packages/9b/35/74442388c6cf008882d4d4bdfc4109be87e9b8b7ccd097ad1e7f006e2e95/pandas-2.3.3-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:8fe25fc7b623b0ef6b5009149627e34d2a4657e880948ec3c840e9402e5c1b45", size = 10833831, upload-time = "2025-09-29T23:38:56.071Z" },
+    { url = "https://files.pythonhosted.org/packages/fe/e4/de154cbfeee13383ad58d23017da99390b91d73f8c11856f2095e813201b/pandas-2.3.3-cp311-cp311-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:b468d3dad6ff947df92dcb32ede5b7bd41a9b3cceef0a30ed925f6d01fb8fa66", size = 12199267, upload-time = "2025-09-29T23:18:41.627Z" },
+    { url = "https://files.pythonhosted.org/packages/bf/c9/63f8d545568d9ab91476b1818b4741f521646cbdd151c6efebf40d6de6f7/pandas-2.3.3-cp311-cp311-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:b98560e98cb334799c0b07ca7967ac361a47326e9b4e5a7dfb5ab2b1c9d35a1b", size = 12789281, upload-time = "2025-09-29T23:18:56.834Z" },
+    { url = "https://files.pythonhosted.org/packages/f2/00/a5ac8c7a0e67fd1a6059e40aa08fa1c52cc00709077d2300e210c3ce0322/pandas-2.3.3-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:1d37b5848ba49824e5c30bedb9c830ab9b7751fd049bc7914533e01c65f79791", size = 13240453, upload-time = "2025-09-29T23:19:09.247Z" },
+    { url = "https://files.pythonhosted.org/packages/27/4d/5c23a5bc7bd209231618dd9e606ce076272c9bc4f12023a70e03a86b4067/pandas-2.3.3-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:db4301b2d1f926ae677a751eb2bd0e8c5f5319c9cb3f88b0becbbb0b07b34151", size = 13890361, upload-time = "2025-09-29T23:19:25.342Z" },
+    { url = "https://files.pythonhosted.org/packages/8e/59/712db1d7040520de7a4965df15b774348980e6df45c129b8c64d0dbe74ef/pandas-2.3.3-cp311-cp311-win_amd64.whl", hash = "sha256:f086f6fe114e19d92014a1966f43a3e62285109afe874f067f5abbdcbb10e59c", size = 11348702, upload-time = "2025-09-29T23:19:38.296Z" },
+    { url = "https://files.pythonhosted.org/packages/9c/fb/231d89e8637c808b997d172b18e9d4a4bc7bf31296196c260526055d1ea0/pandas-2.3.3-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:6d21f6d74eb1725c2efaa71a2bfc661a0689579b58e9c0ca58a739ff0b002b53", size = 11597846, upload-time = "2025-09-29T23:19:48.856Z" },
+    { url = "https://files.pythonhosted.org/packages/5c/bd/bf8064d9cfa214294356c2d6702b716d3cf3bb24be59287a6a21e24cae6b/pandas-2.3.3-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:3fd2f887589c7aa868e02632612ba39acb0b8948faf5cc58f0850e165bd46f35", size = 10729618, upload-time = "2025-09-29T23:39:08.659Z" },
+    { url = "https://files.pythonhosted.org/packages/57/56/cf2dbe1a3f5271370669475ead12ce77c61726ffd19a35546e31aa8edf4e/pandas-2.3.3-cp312-cp312-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:ecaf1e12bdc03c86ad4a7ea848d66c685cb6851d807a26aa245ca3d2017a1908", size = 11737212, upload-time = "2025-09-29T23:19:59.765Z" },
+    { url = "https://files.pythonhosted.org/packages/e5/63/cd7d615331b328e287d8233ba9fdf191a9c2d11b6af0c7a59cfcec23de68/pandas-2.3.3-cp312-cp312-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:b3d11d2fda7eb164ef27ffc14b4fcab16a80e1ce67e9f57e19ec0afaf715ba89", size = 12362693, upload-time = "2025-09-29T23:20:14.098Z" },
+    { url = "https://files.pythonhosted.org/packages/a6/de/8b1895b107277d52f2b42d3a6806e69cfef0d5cf1d0ba343470b9d8e0a04/pandas-2.3.3-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:a68e15f780eddf2b07d242e17a04aa187a7ee12b40b930bfdd78070556550e98", size = 12771002, upload-time = "2025-09-29T23:20:26.76Z" },
+    { url = "https://files.pythonhosted.org/packages/87/21/84072af3187a677c5893b170ba2c8fbe450a6ff911234916da889b698220/pandas-2.3.3-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:371a4ab48e950033bcf52b6527eccb564f52dc826c02afd9a1bc0ab731bba084", size = 13450971, upload-time = "2025-09-29T23:20:41.344Z" },
+    { url = "https://files.pythonhosted.org/packages/86/41/585a168330ff063014880a80d744219dbf1dd7a1c706e75ab3425a987384/pandas-2.3.3-cp312-cp312-win_amd64.whl", hash = "sha256:a16dcec078a01eeef8ee61bf64074b4e524a2a3f4b3be9326420cabe59c4778b", size = 10992722, upload-time = "2025-09-29T23:20:54.139Z" },
+    { url = "https://files.pythonhosted.org/packages/cd/4b/18b035ee18f97c1040d94debd8f2e737000ad70ccc8f5513f4eefad75f4b/pandas-2.3.3-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:56851a737e3470de7fa88e6131f41281ed440d29a9268dcbf0002da5ac366713", size = 11544671, upload-time = "2025-09-29T23:21:05.024Z" },
+    { url = "https://files.pythonhosted.org/packages/31/94/72fac03573102779920099bcac1c3b05975c2cb5f01eac609faf34bed1ca/pandas-2.3.3-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:bdcd9d1167f4885211e401b3036c0c8d9e274eee67ea8d0758a256d60704cfe8", size = 10680807, upload-time = "2025-09-29T23:21:15.979Z" },
+    { url = "https://files.pythonhosted.org/packages/16/87/9472cf4a487d848476865321de18cc8c920b8cab98453ab79dbbc98db63a/pandas-2.3.3-cp313-cp313-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:e32e7cc9af0f1cc15548288a51a3b681cc2a219faa838e995f7dc53dbab1062d", size = 11709872, upload-time = "2025-09-29T23:21:27.165Z" },
+    { url = "https://files.pythonhosted.org/packages/15/07/284f757f63f8a8d69ed4472bfd85122bd086e637bf4ed09de572d575a693/pandas-2.3.3-cp313-cp313-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:318d77e0e42a628c04dc56bcef4b40de67918f7041c2b061af1da41dcff670ac", size = 12306371, upload-time = "2025-09-29T23:21:40.532Z" },
+    { url = "https://files.pythonhosted.org/packages/33/81/a3afc88fca4aa925804a27d2676d22dcd2031c2ebe08aabd0ae55b9ff282/pandas-2.3.3-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:4e0a175408804d566144e170d0476b15d78458795bb18f1304fb94160cabf40c", size = 12765333, upload-time = "2025-09-29T23:21:55.77Z" },
+    { url = "https://files.pythonhosted.org/packages/8d/0f/b4d4ae743a83742f1153464cf1a8ecfafc3ac59722a0b5c8602310cb7158/pandas-2.3.3-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:93c2d9ab0fc11822b5eece72ec9587e172f63cff87c00b062f6e37448ced4493", size = 13418120, upload-time = "2025-09-29T23:22:10.109Z" },
+    { url = "https://files.pythonhosted.org/packages/4f/c7/e54682c96a895d0c808453269e0b5928a07a127a15704fedb643e9b0a4c8/pandas-2.3.3-cp313-cp313-win_amd64.whl", hash = "sha256:f8bfc0e12dc78f777f323f55c58649591b2cd0c43534e8355c51d3fede5f4dee", size = 10993991, upload-time = "2025-09-29T23:25:04.889Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/ca/3f8d4f49740799189e1395812f3bf23b5e8fc7c190827d55a610da72ce55/pandas-2.3.3-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:75ea25f9529fdec2d2e93a42c523962261e567d250b0013b16210e1d40d7c2e5", size = 12048227, upload-time = "2025-09-29T23:22:24.343Z" },
+    { url = "https://files.pythonhosted.org/packages/0e/5a/f43efec3e8c0cc92c4663ccad372dbdff72b60bdb56b2749f04aa1d07d7e/pandas-2.3.3-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:74ecdf1d301e812db96a465a525952f4dde225fdb6d8e5a521d47e1f42041e21", size = 11411056, upload-time = "2025-09-29T23:22:37.762Z" },
+    { url = "https://files.pythonhosted.org/packages/46/b1/85331edfc591208c9d1a63a06baa67b21d332e63b7a591a5ba42a10bb507/pandas-2.3.3-cp313-cp313t-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:6435cb949cb34ec11cc9860246ccb2fdc9ecd742c12d3304989017d53f039a78", size = 11645189, upload-time = "2025-09-29T23:22:51.688Z" },
+    { url = "https://files.pythonhosted.org/packages/44/23/78d645adc35d94d1ac4f2a3c4112ab6f5b8999f4898b8cdf01252f8df4a9/pandas-2.3.3-cp313-cp313t-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:900f47d8f20860de523a1ac881c4c36d65efcb2eb850e6948140fa781736e110", size = 12121912, upload-time = "2025-09-29T23:23:05.042Z" },
+    { url = "https://files.pythonhosted.org/packages/53/da/d10013df5e6aaef6b425aa0c32e1fc1f3e431e4bcabd420517dceadce354/pandas-2.3.3-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:a45c765238e2ed7d7c608fc5bc4a6f88b642f2f01e70c0c23d2224dd21829d86", size = 12712160, upload-time = "2025-09-29T23:23:28.57Z" },
+    { url = "https://files.pythonhosted.org/packages/bd/17/e756653095a083d8a37cbd816cb87148debcfcd920129b25f99dd8d04271/pandas-2.3.3-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:c4fc4c21971a1a9f4bdb4c73978c7f7256caa3e62b323f70d6cb80db583350bc", size = 13199233, upload-time = "2025-09-29T23:24:24.876Z" },
+    { url = "https://files.pythonhosted.org/packages/04/fd/74903979833db8390b73b3a8a7d30d146d710bd32703724dd9083950386f/pandas-2.3.3-cp314-cp314-macosx_10_13_x86_64.whl", hash = "sha256:ee15f284898e7b246df8087fc82b87b01686f98ee67d85a17b7ab44143a3a9a0", size = 11540635, upload-time = "2025-09-29T23:25:52.486Z" },
+    { url = "https://files.pythonhosted.org/packages/21/00/266d6b357ad5e6d3ad55093a7e8efc7dd245f5a842b584db9f30b0f0a287/pandas-2.3.3-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:1611aedd912e1ff81ff41c745822980c49ce4a7907537be8692c8dbc31924593", size = 10759079, upload-time = "2025-09-29T23:26:33.204Z" },
+    { url = "https://files.pythonhosted.org/packages/ca/05/d01ef80a7a3a12b2f8bbf16daba1e17c98a2f039cbc8e2f77a2c5a63d382/pandas-2.3.3-cp314-cp314-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:6d2cefc361461662ac48810cb14365a365ce864afe85ef1f447ff5a1e99ea81c", size = 11814049, upload-time = "2025-09-29T23:27:15.384Z" },
+    { url = "https://files.pythonhosted.org/packages/15/b2/0e62f78c0c5ba7e3d2c5945a82456f4fac76c480940f805e0b97fcbc2f65/pandas-2.3.3-cp314-cp314-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:ee67acbbf05014ea6c763beb097e03cd629961c8a632075eeb34247120abcb4b", size = 12332638, upload-time = "2025-09-29T23:27:51.625Z" },
+    { url = "https://files.pythonhosted.org/packages/c5/33/dd70400631b62b9b29c3c93d2feee1d0964dc2bae2e5ad7a6c73a7f25325/pandas-2.3.3-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:c46467899aaa4da076d5abc11084634e2d197e9460643dd455ac3db5856b24d6", size = 12886834, upload-time = "2025-09-29T23:28:21.289Z" },
+    { url = "https://files.pythonhosted.org/packages/d3/18/b5d48f55821228d0d2692b34fd5034bb185e854bdb592e9c640f6290e012/pandas-2.3.3-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:6253c72c6a1d990a410bc7de641d34053364ef8bcd3126f7e7450125887dffe3", size = 13409925, upload-time = "2025-09-29T23:28:58.261Z" },
+    { url = "https://files.pythonhosted.org/packages/a6/3d/124ac75fcd0ecc09b8fdccb0246ef65e35b012030defb0e0eba2cbbbe948/pandas-2.3.3-cp314-cp314-win_amd64.whl", hash = "sha256:1b07204a219b3b7350abaae088f451860223a52cfb8a6c53358e7948735158e5", size = 11109071, upload-time = "2025-09-29T23:32:27.484Z" },
+    { url = "https://files.pythonhosted.org/packages/89/9c/0e21c895c38a157e0faa1fb64587a9226d6dd46452cac4532d80c3c4a244/pandas-2.3.3-cp314-cp314t-macosx_10_13_x86_64.whl", hash = "sha256:2462b1a365b6109d275250baaae7b760fd25c726aaca0054649286bcfbb3e8ec", size = 12048504, upload-time = "2025-09-29T23:29:31.47Z" },
+    { url = "https://files.pythonhosted.org/packages/d7/82/b69a1c95df796858777b68fbe6a81d37443a33319761d7c652ce77797475/pandas-2.3.3-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:0242fe9a49aa8b4d78a4fa03acb397a58833ef6199e9aa40a95f027bb3a1b6e7", size = 11410702, upload-time = "2025-09-29T23:29:54.591Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/88/702bde3ba0a94b8c73a0181e05144b10f13f29ebfc2150c3a79062a8195d/pandas-2.3.3-cp314-cp314t-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:a21d830e78df0a515db2b3d2f5570610f5e6bd2e27749770e8bb7b524b89b450", size = 11634535, upload-time = "2025-09-29T23:30:21.003Z" },
+    { url = "https://files.pythonhosted.org/packages/a4/1e/1bac1a839d12e6a82ec6cb40cda2edde64a2013a66963293696bbf31fbbb/pandas-2.3.3-cp314-cp314t-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:2e3ebdb170b5ef78f19bfb71b0dc5dc58775032361fa188e814959b74d726dd5", size = 12121582, upload-time = "2025-09-29T23:30:43.391Z" },
+    { url = "https://files.pythonhosted.org/packages/44/91/483de934193e12a3b1d6ae7c8645d083ff88dec75f46e827562f1e4b4da6/pandas-2.3.3-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:d051c0e065b94b7a3cea50eb1ec32e912cd96dba41647eb24104b6c6c14c5788", size = 12699963, upload-time = "2025-09-29T23:31:10.009Z" },
+    { url = "https://files.pythonhosted.org/packages/70/44/5191d2e4026f86a2a109053e194d3ba7a31a2d10a9c2348368c63ed4e85a/pandas-2.3.3-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:3869faf4bd07b3b66a9f462417d0ca3a9df29a9f6abd5d0d0dbab15dac7abe87", size = 13202175, upload-time = "2025-09-29T23:31:59.173Z" },
+]
+
+[[package]]
+name = "pandas"
+version = "3.0.0"
+source = { registry = "https://pypi.org/simple" }
+resolution-markers = [
+    "python_full_version >= '3.14' and sys_platform == 'darwin'",
+    "python_full_version == '3.13.*' and sys_platform == 'darwin'",
+    "python_full_version >= '3.14' and platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "python_full_version == '3.13.*' and platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "python_full_version >= '3.14' and sys_platform == 'win32'",
+    "python_full_version >= '3.14' and sys_platform == 'emscripten'",
+    "(python_full_version >= '3.14' and platform_machine != 'aarch64' and sys_platform == 'linux') or (python_full_version >= '3.14' and sys_platform != 'darwin' and sys_platform != 'emscripten' and sys_platform != 'linux' and sys_platform != 'win32')",
+    "python_full_version == '3.13.*' and sys_platform == 'win32'",
+    "python_full_version == '3.13.*' and sys_platform == 'emscripten'",
+    "(python_full_version == '3.13.*' and platform_machine != 'aarch64' and sys_platform == 'linux') or (python_full_version == '3.13.*' and sys_platform != 'darwin' and sys_platform != 'emscripten' and sys_platform != 'linux' and sys_platform != 'win32')",
+    "python_full_version == '3.12.*' and sys_platform == 'darwin'",
+    "python_full_version == '3.12.*' and platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "python_full_version == '3.12.*' and sys_platform == 'win32'",
+    "python_full_version == '3.12.*' and sys_platform == 'emscripten'",
+    "(python_full_version == '3.12.*' and platform_machine != 'aarch64' and sys_platform == 'linux') or (python_full_version == '3.12.*' and sys_platform != 'darwin' and sys_platform != 'emscripten' and sys_platform != 'linux' and sys_platform != 'win32')",
+    "python_full_version == '3.11.*' and sys_platform == 'darwin'",
+    "python_full_version == '3.11.*' and platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "python_full_version == '3.11.*' and sys_platform == 'win32'",
+    "python_full_version == '3.11.*' and sys_platform == 'emscripten'",
+    "(python_full_version == '3.11.*' and platform_machine != 'aarch64' and sys_platform == 'linux') or (python_full_version == '3.11.*' and sys_platform != 'darwin' and sys_platform != 'emscripten' and sys_platform != 'linux' and sys_platform != 'win32')",
+]
+dependencies = [
+    { name = "numpy", version = "2.4.2", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
+    { name = "python-dateutil", marker = "python_full_version >= '3.11'" },
+    { name = "tzdata", marker = "(python_full_version >= '3.11' and sys_platform == 'emscripten') or (python_full_version >= '3.11' and sys_platform == 'win32')" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/de/da/b1dc0481ab8d55d0f46e343cfe67d4551a0e14fcee52bd38ca1bd73258d8/pandas-3.0.0.tar.gz", hash = "sha256:0facf7e87d38f721f0af46fe70d97373a37701b1c09f7ed7aeeb292ade5c050f", size = 4633005, upload-time = "2026-01-21T15:52:04.726Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/46/1e/b184654a856e75e975a6ee95d6577b51c271cd92cb2b020c9378f53e0032/pandas-3.0.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:d64ce01eb9cdca96a15266aa679ae50212ec52757c79204dbc7701a222401850", size = 10313247, upload-time = "2026-01-21T15:50:15.775Z" },
+    { url = "https://files.pythonhosted.org/packages/dd/5e/e04a547ad0f0183bf151fd7c7a477468e3b85ff2ad231c566389e6cc9587/pandas-3.0.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:613e13426069793aa1ec53bdcc3b86e8d32071daea138bbcf4fa959c9cdaa2e2", size = 9913131, upload-time = "2026-01-21T15:50:18.611Z" },
+    { url = "https://files.pythonhosted.org/packages/a2/93/bb77bfa9fc2aba9f7204db807d5d3fb69832ed2854c60ba91b4c65ba9219/pandas-3.0.0-cp311-cp311-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:0192fee1f1a8e743b464a6607858ee4b071deb0b118eb143d71c2a1d170996d5", size = 10741925, upload-time = "2026-01-21T15:50:21.058Z" },
+    { url = "https://files.pythonhosted.org/packages/62/fb/89319812eb1d714bfc04b7f177895caeba8ab4a37ef6712db75ed786e2e0/pandas-3.0.0-cp311-cp311-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:f0b853319dec8d5e0c8b875374c078ef17f2269986a78168d9bd57e49bf650ae", size = 11245979, upload-time = "2026-01-21T15:50:23.413Z" },
+    { url = "https://files.pythonhosted.org/packages/a9/63/684120486f541fc88da3862ed31165b3b3e12b6a1c7b93be4597bc84e26c/pandas-3.0.0-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:707a9a877a876c326ae2cb640fbdc4ef63b0a7b9e2ef55c6df9942dcee8e2af9", size = 11756337, upload-time = "2026-01-21T15:50:25.932Z" },
+    { url = "https://files.pythonhosted.org/packages/39/92/7eb0ad232312b59aec61550c3c81ad0743898d10af5df7f80bc5e5065416/pandas-3.0.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:afd0aa3d0b5cda6e0b8ffc10dbcca3b09ef3cbcd3fe2b27364f85fdc04e1989d", size = 12325517, upload-time = "2026-01-21T15:50:27.952Z" },
+    { url = "https://files.pythonhosted.org/packages/51/27/bf9436dd0a4fc3130acec0828951c7ef96a0631969613a9a35744baf27f6/pandas-3.0.0-cp311-cp311-win_amd64.whl", hash = "sha256:113b4cca2614ff7e5b9fee9b6f066618fe73c5a83e99d721ffc41217b2bf57dd", size = 9881576, upload-time = "2026-01-21T15:50:30.149Z" },
+    { url = "https://files.pythonhosted.org/packages/e7/2b/c618b871fce0159fd107516336e82891b404e3f340821853c2fc28c7830f/pandas-3.0.0-cp311-cp311-win_arm64.whl", hash = "sha256:c14837eba8e99a8da1527c0280bba29b0eb842f64aa94982c5e21227966e164b", size = 9140807, upload-time = "2026-01-21T15:50:32.308Z" },
+    { url = "https://files.pythonhosted.org/packages/0b/38/db33686f4b5fa64d7af40d96361f6a4615b8c6c8f1b3d334eee46ae6160e/pandas-3.0.0-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:9803b31f5039b3c3b10cc858c5e40054adb4b29b4d81cb2fd789f4121c8efbcd", size = 10334013, upload-time = "2026-01-21T15:50:34.771Z" },
+    { url = "https://files.pythonhosted.org/packages/a5/7b/9254310594e9774906bacdd4e732415e1f86ab7dbb4b377ef9ede58cd8ec/pandas-3.0.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:14c2a4099cd38a1d18ff108168ea417909b2dea3bd1ebff2ccf28ddb6a74d740", size = 9874154, upload-time = "2026-01-21T15:50:36.67Z" },
+    { url = "https://files.pythonhosted.org/packages/63/d4/726c5a67a13bc66643e66d2e9ff115cead482a44fc56991d0c4014f15aaf/pandas-3.0.0-cp312-cp312-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:d257699b9a9960e6125686098d5714ac59d05222bef7a5e6af7a7fd87c650801", size = 10384433, upload-time = "2026-01-21T15:50:39.132Z" },
+    { url = "https://files.pythonhosted.org/packages/bf/2e/9211f09bedb04f9832122942de8b051804b31a39cfbad199a819bb88d9f3/pandas-3.0.0-cp312-cp312-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:69780c98f286076dcafca38d8b8eee1676adf220199c0a39f0ecbf976b68151a", size = 10864519, upload-time = "2026-01-21T15:50:41.043Z" },
+    { url = "https://files.pythonhosted.org/packages/00/8d/50858522cdc46ac88b9afdc3015e298959a70a08cd21e008a44e9520180c/pandas-3.0.0-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:4a66384f017240f3858a4c8a7cf21b0591c3ac885cddb7758a589f0f71e87ebb", size = 11394124, upload-time = "2026-01-21T15:50:43.377Z" },
+    { url = "https://files.pythonhosted.org/packages/86/3f/83b2577db02503cd93d8e95b0f794ad9d4be0ba7cb6c8bcdcac964a34a42/pandas-3.0.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:be8c515c9bc33989d97b89db66ea0cececb0f6e3c2a87fcc8b69443a6923e95f", size = 11920444, upload-time = "2026-01-21T15:50:45.932Z" },
+    { url = "https://files.pythonhosted.org/packages/64/2d/4f8a2f192ed12c90a0aab47f5557ece0e56b0370c49de9454a09de7381b2/pandas-3.0.0-cp312-cp312-win_amd64.whl", hash = "sha256:a453aad8c4f4e9f166436994a33884442ea62aa8b27d007311e87521b97246e1", size = 9730970, upload-time = "2026-01-21T15:50:47.962Z" },
+    { url = "https://files.pythonhosted.org/packages/d4/64/ff571be435cf1e643ca98d0945d76732c0b4e9c37191a89c8550b105eed1/pandas-3.0.0-cp312-cp312-win_arm64.whl", hash = "sha256:da768007b5a33057f6d9053563d6b74dd6d029c337d93c6d0d22a763a5c2ecc0", size = 9041950, upload-time = "2026-01-21T15:50:50.422Z" },
+    { url = "https://files.pythonhosted.org/packages/6f/fa/7f0ac4ca8877c57537aaff2a842f8760e630d8e824b730eb2e859ffe96ca/pandas-3.0.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:b78d646249b9a2bc191040988c7bb524c92fa8534fb0898a0741d7e6f2ffafa6", size = 10307129, upload-time = "2026-01-21T15:50:52.877Z" },
+    { url = "https://files.pythonhosted.org/packages/6f/11/28a221815dcea4c0c9414dfc845e34a84a6a7dabc6da3194498ed5ba4361/pandas-3.0.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:bc9cba7b355cb4162442a88ce495e01cb605f17ac1e27d6596ac963504e0305f", size = 9850201, upload-time = "2026-01-21T15:50:54.807Z" },
+    { url = "https://files.pythonhosted.org/packages/ba/da/53bbc8c5363b7e5bd10f9ae59ab250fc7a382ea6ba08e4d06d8694370354/pandas-3.0.0-cp313-cp313-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:3c9a1a149aed3b6c9bf246033ff91e1b02d529546c5d6fb6b74a28fea0cf4c70", size = 10354031, upload-time = "2026-01-21T15:50:57.463Z" },
+    { url = "https://files.pythonhosted.org/packages/f7/a3/51e02ebc2a14974170d51e2410dfdab58870ea9bcd37cda15bd553d24dc4/pandas-3.0.0-cp313-cp313-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:95683af6175d884ee89471842acfca29172a85031fccdabc35e50c0984470a0e", size = 10861165, upload-time = "2026-01-21T15:50:59.32Z" },
+    { url = "https://files.pythonhosted.org/packages/a5/fe/05a51e3cac11d161472b8297bd41723ea98013384dd6d76d115ce3482f9b/pandas-3.0.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:1fbbb5a7288719e36b76b4f18d46ede46e7f916b6c8d9915b756b0a6c3f792b3", size = 11359359, upload-time = "2026-01-21T15:51:02.014Z" },
+    { url = "https://files.pythonhosted.org/packages/ee/56/ba620583225f9b85a4d3e69c01df3e3870659cc525f67929b60e9f21dcd1/pandas-3.0.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:8e8b9808590fa364416b49b2a35c1f4cf2785a6c156935879e57f826df22038e", size = 11912907, upload-time = "2026-01-21T15:51:05.175Z" },
+    { url = "https://files.pythonhosted.org/packages/c9/8c/c6638d9f67e45e07656b3826405c5cc5f57f6fd07c8b2572ade328c86e22/pandas-3.0.0-cp313-cp313-win_amd64.whl", hash = "sha256:98212a38a709feb90ae658cb6227ea3657c22ba8157d4b8f913cd4c950de5e7e", size = 9732138, upload-time = "2026-01-21T15:51:07.569Z" },
+    { url = "https://files.pythonhosted.org/packages/7b/bf/bd1335c3bf1770b6d8fed2799993b11c4971af93bb1b729b9ebbc02ca2ec/pandas-3.0.0-cp313-cp313-win_arm64.whl", hash = "sha256:177d9df10b3f43b70307a149d7ec49a1229a653f907aa60a48f1877d0e6be3be", size = 9033568, upload-time = "2026-01-21T15:51:09.484Z" },
+    { url = "https://files.pythonhosted.org/packages/8e/c6/f5e2171914d5e29b9171d495344097d54e3ffe41d2d85d8115baba4dc483/pandas-3.0.0-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:2713810ad3806767b89ad3b7b69ba153e1c6ff6d9c20f9c2140379b2a98b6c98", size = 10741936, upload-time = "2026-01-21T15:51:11.693Z" },
+    { url = "https://files.pythonhosted.org/packages/51/88/9a0164f99510a1acb9f548691f022c756c2314aad0d8330a24616c14c462/pandas-3.0.0-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:15d59f885ee5011daf8335dff47dcb8a912a27b4ad7826dc6cbe809fd145d327", size = 10393884, upload-time = "2026-01-21T15:51:14.197Z" },
+    { url = "https://files.pythonhosted.org/packages/e0/53/b34d78084d88d8ae2b848591229da8826d1e65aacf00b3abe34023467648/pandas-3.0.0-cp313-cp313t-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:24e6547fb64d2c92665dd2adbfa4e85fa4fd70a9c070e7cfb03b629a0bbab5eb", size = 10310740, upload-time = "2026-01-21T15:51:16.093Z" },
+    { url = "https://files.pythonhosted.org/packages/5b/d3/bee792e7c3d6930b74468d990604325701412e55d7aaf47460a22311d1a5/pandas-3.0.0-cp313-cp313t-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:48ee04b90e2505c693d3f8e8f524dab8cb8aaf7ddcab52c92afa535e717c4812", size = 10700014, upload-time = "2026-01-21T15:51:18.818Z" },
+    { url = "https://files.pythonhosted.org/packages/55/db/2570bc40fb13aaed1cbc3fbd725c3a60ee162477982123c3adc8971e7ac1/pandas-3.0.0-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:66f72fb172959af42a459e27a8d8d2c7e311ff4c1f7db6deb3b643dbc382ae08", size = 11323737, upload-time = "2026-01-21T15:51:20.784Z" },
+    { url = "https://files.pythonhosted.org/packages/bc/2e/297ac7f21c8181b62a4cccebad0a70caf679adf3ae5e83cb676194c8acc3/pandas-3.0.0-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:4a4a400ca18230976724a5066f20878af785f36c6756e498e94c2a5e5d57779c", size = 11771558, upload-time = "2026-01-21T15:51:22.977Z" },
+    { url = "https://files.pythonhosted.org/packages/0a/46/e1c6876d71c14332be70239acce9ad435975a80541086e5ffba2f249bcf6/pandas-3.0.0-cp313-cp313t-win_amd64.whl", hash = "sha256:940eebffe55528074341a5a36515f3e4c5e25e958ebbc764c9502cfc35ba3faa", size = 10473771, upload-time = "2026-01-21T15:51:25.285Z" },
+    { url = "https://files.pythonhosted.org/packages/c0/db/0270ad9d13c344b7a36fa77f5f8344a46501abf413803e885d22864d10bf/pandas-3.0.0-cp314-cp314-macosx_10_15_x86_64.whl", hash = "sha256:597c08fb9fef0edf1e4fa2f9828dd27f3d78f9b8c9b4a748d435ffc55732310b", size = 10312075, upload-time = "2026-01-21T15:51:28.5Z" },
+    { url = "https://files.pythonhosted.org/packages/09/9f/c176f5e9717f7c91becfe0f55a52ae445d3f7326b4a2cf355978c51b7913/pandas-3.0.0-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:447b2d68ac5edcbf94655fe909113a6dba6ef09ad7f9f60c80477825b6c489fe", size = 9900213, upload-time = "2026-01-21T15:51:30.955Z" },
+    { url = "https://files.pythonhosted.org/packages/d9/e7/63ad4cc10b257b143e0a5ebb04304ad806b4e1a61c5da25f55896d2ca0f4/pandas-3.0.0-cp314-cp314-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:debb95c77ff3ed3ba0d9aa20c3a2f19165cc7956362f9873fce1ba0a53819d70", size = 10428768, upload-time = "2026-01-21T15:51:33.018Z" },
+    { url = "https://files.pythonhosted.org/packages/9e/0e/4e4c2d8210f20149fd2248ef3fff26623604922bd564d915f935a06dd63d/pandas-3.0.0-cp314-cp314-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:fedabf175e7cd82b69b74c30adbaa616de301291a5231138d7242596fc296a8d", size = 10882954, upload-time = "2026-01-21T15:51:35.287Z" },
+    { url = "https://files.pythonhosted.org/packages/c6/60/c9de8ac906ba1f4d2250f8a951abe5135b404227a55858a75ad26f84db47/pandas-3.0.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:412d1a89aab46889f3033a386912efcdfa0f1131c5705ff5b668dda88305e986", size = 11430293, upload-time = "2026-01-21T15:51:37.57Z" },
+    { url = "https://files.pythonhosted.org/packages/a1/69/806e6637c70920e5787a6d6896fd707f8134c2c55cd761e7249a97b7dc5a/pandas-3.0.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:e979d22316f9350c516479dd3a92252be2937a9531ed3a26ec324198a99cdd49", size = 11952452, upload-time = "2026-01-21T15:51:39.618Z" },
+    { url = "https://files.pythonhosted.org/packages/cb/de/918621e46af55164c400ab0ef389c9d969ab85a43d59ad1207d4ddbe30a5/pandas-3.0.0-cp314-cp314-win_amd64.whl", hash = "sha256:083b11415b9970b6e7888800c43c82e81a06cd6b06755d84804444f0007d6bb7", size = 9851081, upload-time = "2026-01-21T15:51:41.758Z" },
+    { url = "https://files.pythonhosted.org/packages/91/a1/3562a18dd0bd8c73344bfa26ff90c53c72f827df119d6d6b1dacc84d13e3/pandas-3.0.0-cp314-cp314-win_arm64.whl", hash = "sha256:5db1e62cb99e739fa78a28047e861b256d17f88463c76b8dafc7c1338086dca8", size = 9174610, upload-time = "2026-01-21T15:51:44.312Z" },
+    { url = "https://files.pythonhosted.org/packages/ce/26/430d91257eaf366f1737d7a1c158677caaf6267f338ec74e3a1ec444111c/pandas-3.0.0-cp314-cp314t-macosx_10_15_x86_64.whl", hash = "sha256:697b8f7d346c68274b1b93a170a70974cdc7d7354429894d5927c1effdcccd73", size = 10761999, upload-time = "2026-01-21T15:51:46.899Z" },
+    { url = "https://files.pythonhosted.org/packages/ec/1a/954eb47736c2b7f7fe6a9d56b0cb6987773c00faa3c6451a43db4beb3254/pandas-3.0.0-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:8cb3120f0d9467ed95e77f67a75e030b67545bcfa08964e349252d674171def2", size = 10410279, upload-time = "2026-01-21T15:51:48.89Z" },
+    { url = "https://files.pythonhosted.org/packages/20/fc/b96f3a5a28b250cd1b366eb0108df2501c0f38314a00847242abab71bb3a/pandas-3.0.0-cp314-cp314t-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:33fd3e6baa72899746b820c31e4b9688c8e1b7864d7aec2de7ab5035c285277a", size = 10330198, upload-time = "2026-01-21T15:51:51.015Z" },
+    { url = "https://files.pythonhosted.org/packages/90/b3/d0e2952f103b4fbef1ef22d0c2e314e74fc9064b51cee30890b5e3286ee6/pandas-3.0.0-cp314-cp314t-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:a8942e333dc67ceda1095227ad0febb05a3b36535e520154085db632c40ad084", size = 10728513, upload-time = "2026-01-21T15:51:53.387Z" },
+    { url = "https://files.pythonhosted.org/packages/76/81/832894f286df828993dc5fd61c63b231b0fb73377e99f6c6c369174cf97e/pandas-3.0.0-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:783ac35c4d0fe0effdb0d67161859078618b1b6587a1af15928137525217a721", size = 11345550, upload-time = "2026-01-21T15:51:55.329Z" },
+    { url = "https://files.pythonhosted.org/packages/34/a0/ed160a00fb4f37d806406bc0a79a8b62fe67f29d00950f8d16203ff3409b/pandas-3.0.0-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:125eb901e233f155b268bbef9abd9afb5819db74f0e677e89a61b246228c71ac", size = 11799386, upload-time = "2026-01-21T15:51:57.457Z" },
+    { url = "https://files.pythonhosted.org/packages/36/c8/2ac00d7255252c5e3cf61b35ca92ca25704b0188f7454ca4aec08a33cece/pandas-3.0.0-cp314-cp314t-win_amd64.whl", hash = "sha256:b86d113b6c109df3ce0ad5abbc259fe86a1bd4adfd4a31a89da42f84f65509bb", size = 10873041, upload-time = "2026-01-21T15:52:00.034Z" },
+    { url = "https://files.pythonhosted.org/packages/e6/3f/a80ac00acbc6b35166b42850e98a4f466e2c0d9c64054161ba9620f95680/pandas-3.0.0-cp314-cp314t-win_arm64.whl", hash = "sha256:1c39eab3ad38f2d7a249095f0a3d8f8c22cc0f847e98ccf5bbe732b272e2d9fa", size = 9441003, upload-time = "2026-01-21T15:52:02.281Z" },
+]
+
+[[package]]
+name = "pathspec"
+version = "1.0.4"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/fa/36/e27608899f9b8d4dff0617b2d9ab17ca5608956ca44461ac14ac48b44015/pathspec-1.0.4.tar.gz", hash = "sha256:0210e2ae8a21a9137c0d470578cb0e595af87edaa6ebf12ff176f14a02e0e645", size = 131200, upload-time = "2026-01-27T03:59:46.938Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/ef/3c/2c197d226f9ea224a9ab8d197933f9da0ae0aac5b6e0f884e2b8d9c8e9f7/pathspec-1.0.4-py3-none-any.whl", hash = "sha256:fb6ae2fd4e7c921a165808a552060e722767cfa526f99ca5156ed2ce45a5c723", size = 55206, upload-time = "2026-01-27T03:59:45.137Z" },
+]
+
+[[package]]
+name = "pillow"
+version = "12.1.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/d0/02/d52c733a2452ef1ffcc123b68e6606d07276b0e358db70eabad7e40042b7/pillow-12.1.0.tar.gz", hash = "sha256:5c5ae0a06e9ea030ab786b0251b32c7e4ce10e58d983c0d5c56029455180b5b9", size = 46977283, upload-time = "2026-01-02T09:13:29.892Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/fe/41/f73d92b6b883a579e79600d391f2e21cb0df767b2714ecbd2952315dfeef/pillow-12.1.0-cp310-cp310-macosx_10_10_x86_64.whl", hash = "sha256:fb125d860738a09d363a88daa0f59c4533529a90e564785e20fe875b200b6dbd", size = 5304089, upload-time = "2026-01-02T09:10:24.953Z" },
+    { url = "https://files.pythonhosted.org/packages/94/55/7aca2891560188656e4a91ed9adba305e914a4496800da6b5c0a15f09edf/pillow-12.1.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:cad302dc10fac357d3467a74a9561c90609768a6f73a1923b0fd851b6486f8b0", size = 4657815, upload-time = "2026-01-02T09:10:27.063Z" },
+    { url = "https://files.pythonhosted.org/packages/e9/d2/b28221abaa7b4c40b7dba948f0f6a708bd7342c4d47ce342f0ea39643974/pillow-12.1.0-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:a40905599d8079e09f25027423aed94f2823adaf2868940de991e53a449e14a8", size = 6222593, upload-time = "2026-01-02T09:10:29.115Z" },
+    { url = "https://files.pythonhosted.org/packages/71/b8/7a61fb234df6a9b0b479f69e66901209d89ff72a435b49933f9122f94cac/pillow-12.1.0-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:92a7fe4225365c5e3a8e598982269c6d6698d3e783b3b1ae979e7819f9cd55c1", size = 8027579, upload-time = "2026-01-02T09:10:31.182Z" },
+    { url = "https://files.pythonhosted.org/packages/ea/51/55c751a57cc524a15a0e3db20e5cde517582359508d62305a627e77fd295/pillow-12.1.0-cp310-cp310-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:f10c98f49227ed8383d28174ee95155a675c4ed7f85e2e573b04414f7e371bda", size = 6335760, upload-time = "2026-01-02T09:10:33.02Z" },
+    { url = "https://files.pythonhosted.org/packages/dc/7c/60e3e6f5e5891a1a06b4c910f742ac862377a6fe842f7184df4a274ce7bf/pillow-12.1.0-cp310-cp310-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:8637e29d13f478bc4f153d8daa9ffb16455f0a6cb287da1b432fdad2bfbd66c7", size = 7027127, upload-time = "2026-01-02T09:10:35.009Z" },
+    { url = "https://files.pythonhosted.org/packages/06/37/49d47266ba50b00c27ba63a7c898f1bb41a29627ced8c09e25f19ebec0ff/pillow-12.1.0-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:21e686a21078b0f9cb8c8a961d99e6a4ddb88e0fc5ea6e130172ddddc2e5221a", size = 6449896, upload-time = "2026-01-02T09:10:36.793Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/e5/67fd87d2913902462cd9b79c6211c25bfe95fcf5783d06e1367d6d9a741f/pillow-12.1.0-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:2415373395a831f53933c23ce051021e79c8cd7979822d8cc478547a3f4da8ef", size = 7151345, upload-time = "2026-01-02T09:10:39.064Z" },
+    { url = "https://files.pythonhosted.org/packages/bd/15/f8c7abf82af68b29f50d77c227e7a1f87ce02fdc66ded9bf603bc3b41180/pillow-12.1.0-cp310-cp310-win32.whl", hash = "sha256:e75d3dba8fc1ddfec0cd752108f93b83b4f8d6ab40e524a95d35f016b9683b09", size = 6325568, upload-time = "2026-01-02T09:10:41.035Z" },
+    { url = "https://files.pythonhosted.org/packages/d4/24/7d1c0e160b6b5ac2605ef7d8be537e28753c0db5363d035948073f5513d7/pillow-12.1.0-cp310-cp310-win_amd64.whl", hash = "sha256:64efdf00c09e31efd754448a383ea241f55a994fd079866b92d2bbff598aad91", size = 7032367, upload-time = "2026-01-02T09:10:43.09Z" },
+    { url = "https://files.pythonhosted.org/packages/f4/03/41c038f0d7a06099254c60f618d0ec7be11e79620fc23b8e85e5b31d9a44/pillow-12.1.0-cp310-cp310-win_arm64.whl", hash = "sha256:f188028b5af6b8fb2e9a76ac0f841a575bd1bd396e46ef0840d9b88a48fdbcea", size = 2452345, upload-time = "2026-01-02T09:10:44.795Z" },
+    { url = "https://files.pythonhosted.org/packages/43/c4/bf8328039de6cc22182c3ef007a2abfbbdab153661c0a9aa78af8d706391/pillow-12.1.0-cp311-cp311-macosx_10_10_x86_64.whl", hash = "sha256:a83e0850cb8f5ac975291ebfc4170ba481f41a28065277f7f735c202cd8e0af3", size = 5304057, upload-time = "2026-01-02T09:10:46.627Z" },
+    { url = "https://files.pythonhosted.org/packages/43/06/7264c0597e676104cc22ca73ee48f752767cd4b1fe084662620b17e10120/pillow-12.1.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:b6e53e82ec2db0717eabb276aa56cf4e500c9a7cec2c2e189b55c24f65a3e8c0", size = 4657811, upload-time = "2026-01-02T09:10:49.548Z" },
+    { url = "https://files.pythonhosted.org/packages/72/64/f9189e44474610daf83da31145fa56710b627b5c4c0b9c235e34058f6b31/pillow-12.1.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:40a8e3b9e8773876d6e30daed22f016509e3987bab61b3b7fe309d7019a87451", size = 6232243, upload-time = "2026-01-02T09:10:51.62Z" },
+    { url = "https://files.pythonhosted.org/packages/ef/30/0df458009be6a4caca4ca2c52975e6275c387d4e5c95544e34138b41dc86/pillow-12.1.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:800429ac32c9b72909c671aaf17ecd13110f823ddb7db4dfef412a5587c2c24e", size = 8037872, upload-time = "2026-01-02T09:10:53.446Z" },
+    { url = "https://files.pythonhosted.org/packages/e4/86/95845d4eda4f4f9557e25381d70876aa213560243ac1a6d619c46caaedd9/pillow-12.1.0-cp311-cp311-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:0b022eaaf709541b391ee069f0022ee5b36c709df71986e3f7be312e46f42c84", size = 6345398, upload-time = "2026-01-02T09:10:55.426Z" },
+    { url = "https://files.pythonhosted.org/packages/5c/1f/8e66ab9be3aaf1435bc03edd1ebdf58ffcd17f7349c1d970cafe87af27d9/pillow-12.1.0-cp311-cp311-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:1f345e7bc9d7f368887c712aa5054558bad44d2a301ddf9248599f4161abc7c0", size = 7034667, upload-time = "2026-01-02T09:10:57.11Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/f6/683b83cb9b1db1fb52b87951b1c0b99bdcfceaa75febf11406c19f82cb5e/pillow-12.1.0-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:d70347c8a5b7ccd803ec0c85c8709f036e6348f1e6a5bf048ecd9c64d3550b8b", size = 6458743, upload-time = "2026-01-02T09:10:59.331Z" },
+    { url = "https://files.pythonhosted.org/packages/9a/7d/de833d63622538c1d58ce5395e7c6cb7e7dce80decdd8bde4a484e095d9f/pillow-12.1.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:1fcc52d86ce7a34fd17cb04e87cfdb164648a3662a6f20565910a99653d66c18", size = 7159342, upload-time = "2026-01-02T09:11:01.82Z" },
+    { url = "https://files.pythonhosted.org/packages/8c/40/50d86571c9e5868c42b81fe7da0c76ca26373f3b95a8dd675425f4a92ec1/pillow-12.1.0-cp311-cp311-win32.whl", hash = "sha256:3ffaa2f0659e2f740473bcf03c702c39a8d4b2b7ffc629052028764324842c64", size = 6328655, upload-time = "2026-01-02T09:11:04.556Z" },
+    { url = "https://files.pythonhosted.org/packages/6c/af/b1d7e301c4cd26cd45d4af884d9ee9b6fab893b0ad2450d4746d74a6968c/pillow-12.1.0-cp311-cp311-win_amd64.whl", hash = "sha256:806f3987ffe10e867bab0ddad45df1148a2b98221798457fa097ad85d6e8bc75", size = 7031469, upload-time = "2026-01-02T09:11:06.538Z" },
+    { url = "https://files.pythonhosted.org/packages/48/36/d5716586d887fb2a810a4a61518a327a1e21c8b7134c89283af272efe84b/pillow-12.1.0-cp311-cp311-win_arm64.whl", hash = "sha256:9f5fefaca968e700ad1a4a9de98bf0869a94e397fe3524c4c9450c1445252304", size = 2452515, upload-time = "2026-01-02T09:11:08.226Z" },
+    { url = "https://files.pythonhosted.org/packages/20/31/dc53fe21a2f2996e1b7d92bf671cdb157079385183ef7c1ae08b485db510/pillow-12.1.0-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:a332ac4ccb84b6dde65dbace8431f3af08874bf9770719d32a635c4ef411b18b", size = 5262642, upload-time = "2026-01-02T09:11:10.138Z" },
+    { url = "https://files.pythonhosted.org/packages/ab/c1/10e45ac9cc79419cedf5121b42dcca5a50ad2b601fa080f58c22fb27626e/pillow-12.1.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:907bfa8a9cb790748a9aa4513e37c88c59660da3bcfffbd24a7d9e6abf224551", size = 4657464, upload-time = "2026-01-02T09:11:12.319Z" },
+    { url = "https://files.pythonhosted.org/packages/ad/26/7b82c0ab7ef40ebede7a97c72d473bda5950f609f8e0c77b04af574a0ddb/pillow-12.1.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:efdc140e7b63b8f739d09a99033aa430accce485ff78e6d311973a67b6bf3208", size = 6234878, upload-time = "2026-01-02T09:11:14.096Z" },
+    { url = "https://files.pythonhosted.org/packages/76/25/27abc9792615b5e886ca9411ba6637b675f1b77af3104710ac7353fe5605/pillow-12.1.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:bef9768cab184e7ae6e559c032e95ba8d07b3023c289f79a2bd36e8bf85605a5", size = 8044868, upload-time = "2026-01-02T09:11:15.903Z" },
+    { url = "https://files.pythonhosted.org/packages/0a/ea/f200a4c36d836100e7bc738fc48cd963d3ba6372ebc8298a889e0cfc3359/pillow-12.1.0-cp312-cp312-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:742aea052cf5ab5034a53c3846165bc3ce88d7c38e954120db0ab867ca242661", size = 6349468, upload-time = "2026-01-02T09:11:17.631Z" },
+    { url = "https://files.pythonhosted.org/packages/11/8f/48d0b77ab2200374c66d344459b8958c86693be99526450e7aee714e03e4/pillow-12.1.0-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:a6dfc2af5b082b635af6e08e0d1f9f1c4e04d17d4e2ca0ef96131e85eda6eb17", size = 7041518, upload-time = "2026-01-02T09:11:19.389Z" },
+    { url = "https://files.pythonhosted.org/packages/1d/23/c281182eb986b5d31f0a76d2a2c8cd41722d6fb8ed07521e802f9bba52de/pillow-12.1.0-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:609e89d9f90b581c8d16358c9087df76024cf058fa693dd3e1e1620823f39670", size = 6462829, upload-time = "2026-01-02T09:11:21.28Z" },
+    { url = "https://files.pythonhosted.org/packages/25/ef/7018273e0faac099d7b00982abdcc39142ae6f3bd9ceb06de09779c4a9d6/pillow-12.1.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:43b4899cfd091a9693a1278c4982f3e50f7fb7cff5153b05174b4afc9593b616", size = 7166756, upload-time = "2026-01-02T09:11:23.559Z" },
+    { url = "https://files.pythonhosted.org/packages/8f/c8/993d4b7ab2e341fe02ceef9576afcf5830cdec640be2ac5bee1820d693d4/pillow-12.1.0-cp312-cp312-win32.whl", hash = "sha256:aa0c9cc0b82b14766a99fbe6084409972266e82f459821cd26997a488a7261a7", size = 6328770, upload-time = "2026-01-02T09:11:25.661Z" },
+    { url = "https://files.pythonhosted.org/packages/a7/87/90b358775a3f02765d87655237229ba64a997b87efa8ccaca7dd3e36e7a7/pillow-12.1.0-cp312-cp312-win_amd64.whl", hash = "sha256:d70534cea9e7966169ad29a903b99fc507e932069a881d0965a1a84bb57f6c6d", size = 7033406, upload-time = "2026-01-02T09:11:27.474Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/cf/881b457eccacac9e5b2ddd97d5071fb6d668307c57cbf4e3b5278e06e536/pillow-12.1.0-cp312-cp312-win_arm64.whl", hash = "sha256:65b80c1ee7e14a87d6a068dd3b0aea268ffcabfe0498d38661b00c5b4b22e74c", size = 2452612, upload-time = "2026-01-02T09:11:29.309Z" },
+    { url = "https://files.pythonhosted.org/packages/dd/c7/2530a4aa28248623e9d7f27316b42e27c32ec410f695929696f2e0e4a778/pillow-12.1.0-cp313-cp313-ios_13_0_arm64_iphoneos.whl", hash = "sha256:7b5dd7cbae20285cdb597b10eb5a2c13aa9de6cde9bb64a3c1317427b1db1ae1", size = 4062543, upload-time = "2026-01-02T09:11:31.566Z" },
+    { url = "https://files.pythonhosted.org/packages/8f/1f/40b8eae823dc1519b87d53c30ed9ef085506b05281d313031755c1705f73/pillow-12.1.0-cp313-cp313-ios_13_0_arm64_iphonesimulator.whl", hash = "sha256:29a4cef9cb672363926f0470afc516dbf7305a14d8c54f7abbb5c199cd8f8179", size = 4138373, upload-time = "2026-01-02T09:11:33.367Z" },
+    { url = "https://files.pythonhosted.org/packages/d4/77/6fa60634cf06e52139fd0e89e5bbf055e8166c691c42fb162818b7fda31d/pillow-12.1.0-cp313-cp313-ios_13_0_x86_64_iphonesimulator.whl", hash = "sha256:681088909d7e8fa9e31b9799aaa59ba5234c58e5e4f1951b4c4d1082a2e980e0", size = 3601241, upload-time = "2026-01-02T09:11:35.011Z" },
+    { url = "https://files.pythonhosted.org/packages/4f/bf/28ab865de622e14b747f0cd7877510848252d950e43002e224fb1c9ababf/pillow-12.1.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:983976c2ab753166dc66d36af6e8ec15bb511e4a25856e2227e5f7e00a160587", size = 5262410, upload-time = "2026-01-02T09:11:36.682Z" },
+    { url = "https://files.pythonhosted.org/packages/1c/34/583420a1b55e715937a85bd48c5c0991598247a1fd2eb5423188e765ea02/pillow-12.1.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:db44d5c160a90df2d24a24760bbd37607d53da0b34fb546c4c232af7192298ac", size = 4657312, upload-time = "2026-01-02T09:11:38.535Z" },
+    { url = "https://files.pythonhosted.org/packages/1d/fd/f5a0896839762885b3376ff04878f86ab2b097c2f9a9cdccf4eda8ba8dc0/pillow-12.1.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:6b7a9d1db5dad90e2991645874f708e87d9a3c370c243c2d7684d28f7e133e6b", size = 6232605, upload-time = "2026-01-02T09:11:40.602Z" },
+    { url = "https://files.pythonhosted.org/packages/98/aa/938a09d127ac1e70e6ed467bd03834350b33ef646b31edb7452d5de43792/pillow-12.1.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:6258f3260986990ba2fa8a874f8b6e808cf5abb51a94015ca3dc3c68aa4f30ea", size = 8041617, upload-time = "2026-01-02T09:11:42.721Z" },
+    { url = "https://files.pythonhosted.org/packages/17/e8/538b24cb426ac0186e03f80f78bc8dc7246c667f58b540bdd57c71c9f79d/pillow-12.1.0-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:e115c15e3bc727b1ca3e641a909f77f8ca72a64fff150f666fcc85e57701c26c", size = 6346509, upload-time = "2026-01-02T09:11:44.955Z" },
+    { url = "https://files.pythonhosted.org/packages/01/9a/632e58ec89a32738cabfd9ec418f0e9898a2b4719afc581f07c04a05e3c9/pillow-12.1.0-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:6741e6f3074a35e47c77b23a4e4f2d90db3ed905cb1c5e6e0d49bff2045632bc", size = 7038117, upload-time = "2026-01-02T09:11:46.736Z" },
+    { url = "https://files.pythonhosted.org/packages/c7/a2/d40308cf86eada842ca1f3ffa45d0ca0df7e4ab33c83f81e73f5eaed136d/pillow-12.1.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:935b9d1aed48fcfb3f838caac506f38e29621b44ccc4f8a64d575cb1b2a88644", size = 6460151, upload-time = "2026-01-02T09:11:48.625Z" },
+    { url = "https://files.pythonhosted.org/packages/f1/88/f5b058ad6453a085c5266660a1417bdad590199da1b32fb4efcff9d33b05/pillow-12.1.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:5fee4c04aad8932da9f8f710af2c1a15a83582cfb884152a9caa79d4efcdbf9c", size = 7164534, upload-time = "2026-01-02T09:11:50.445Z" },
+    { url = "https://files.pythonhosted.org/packages/19/ce/c17334caea1db789163b5d855a5735e47995b0b5dc8745e9a3605d5f24c0/pillow-12.1.0-cp313-cp313-win32.whl", hash = "sha256:a786bf667724d84aa29b5db1c61b7bfdde380202aaca12c3461afd6b71743171", size = 6332551, upload-time = "2026-01-02T09:11:52.234Z" },
+    { url = "https://files.pythonhosted.org/packages/e5/07/74a9d941fa45c90a0d9465098fe1ec85de3e2afbdc15cc4766622d516056/pillow-12.1.0-cp313-cp313-win_amd64.whl", hash = "sha256:461f9dfdafa394c59cd6d818bdfdbab4028b83b02caadaff0ffd433faf4c9a7a", size = 7040087, upload-time = "2026-01-02T09:11:54.822Z" },
+    { url = "https://files.pythonhosted.org/packages/88/09/c99950c075a0e9053d8e880595926302575bc742b1b47fe1bbcc8d388d50/pillow-12.1.0-cp313-cp313-win_arm64.whl", hash = "sha256:9212d6b86917a2300669511ed094a9406888362e085f2431a7da985a6b124f45", size = 2452470, upload-time = "2026-01-02T09:11:56.522Z" },
+    { url = "https://files.pythonhosted.org/packages/b5/ba/970b7d85ba01f348dee4d65412476321d40ee04dcb51cd3735b9dc94eb58/pillow-12.1.0-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:00162e9ca6d22b7c3ee8e61faa3c3253cd19b6a37f126cad04f2f88b306f557d", size = 5264816, upload-time = "2026-01-02T09:11:58.227Z" },
+    { url = "https://files.pythonhosted.org/packages/10/60/650f2fb55fdba7a510d836202aa52f0baac633e50ab1cf18415d332188fb/pillow-12.1.0-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:7d6daa89a00b58c37cb1747ec9fb7ac3bc5ffd5949f5888657dfddde6d1312e0", size = 4660472, upload-time = "2026-01-02T09:12:00.798Z" },
+    { url = "https://files.pythonhosted.org/packages/2b/c0/5273a99478956a099d533c4f46cbaa19fd69d606624f4334b85e50987a08/pillow-12.1.0-cp313-cp313t-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:e2479c7f02f9d505682dc47df8c0ea1fc5e264c4d1629a5d63fe3e2334b89554", size = 6268974, upload-time = "2026-01-02T09:12:02.572Z" },
+    { url = "https://files.pythonhosted.org/packages/b4/26/0bf714bc2e73d5267887d47931d53c4ceeceea6978148ed2ab2a4e6463c4/pillow-12.1.0-cp313-cp313t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:f188d580bd870cda1e15183790d1cc2fa78f666e76077d103edf048eed9c356e", size = 8073070, upload-time = "2026-01-02T09:12:04.75Z" },
+    { url = "https://files.pythonhosted.org/packages/43/cf/1ea826200de111a9d65724c54f927f3111dc5ae297f294b370a670c17786/pillow-12.1.0-cp313-cp313t-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:0fde7ec5538ab5095cc02df38ee99b0443ff0e1c847a045554cf5f9af1f4aa82", size = 6380176, upload-time = "2026-01-02T09:12:06.626Z" },
+    { url = "https://files.pythonhosted.org/packages/03/e0/7938dd2b2013373fd85d96e0f38d62b7a5a262af21ac274250c7ca7847c9/pillow-12.1.0-cp313-cp313t-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:0ed07dca4a8464bada6139ab38f5382f83e5f111698caf3191cb8dbf27d908b4", size = 7067061, upload-time = "2026-01-02T09:12:08.624Z" },
+    { url = "https://files.pythonhosted.org/packages/86/ad/a2aa97d37272a929a98437a8c0ac37b3cf012f4f8721e1bd5154699b2518/pillow-12.1.0-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:f45bd71d1fa5e5749587613037b172e0b3b23159d1c00ef2fc920da6f470e6f0", size = 6491824, upload-time = "2026-01-02T09:12:10.488Z" },
+    { url = "https://files.pythonhosted.org/packages/a4/44/80e46611b288d51b115826f136fb3465653c28f491068a72d3da49b54cd4/pillow-12.1.0-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:277518bf4fe74aa91489e1b20577473b19ee70fb97c374aa50830b279f25841b", size = 7190911, upload-time = "2026-01-02T09:12:12.772Z" },
+    { url = "https://files.pythonhosted.org/packages/86/77/eacc62356b4cf81abe99ff9dbc7402750044aed02cfd6a503f7c6fc11f3e/pillow-12.1.0-cp313-cp313t-win32.whl", hash = "sha256:7315f9137087c4e0ee73a761b163fc9aa3b19f5f606a7fc08d83fd3e4379af65", size = 6336445, upload-time = "2026-01-02T09:12:14.775Z" },
+    { url = "https://files.pythonhosted.org/packages/e7/3c/57d81d0b74d218706dafccb87a87ea44262c43eef98eb3b164fd000e0491/pillow-12.1.0-cp313-cp313t-win_amd64.whl", hash = "sha256:0ddedfaa8b5f0b4ffbc2fa87b556dc59f6bb4ecb14a53b33f9189713ae8053c0", size = 7045354, upload-time = "2026-01-02T09:12:16.599Z" },
+    { url = "https://files.pythonhosted.org/packages/ac/82/8b9b97bba2e3576a340f93b044a3a3a09841170ab4c1eb0d5c93469fd32f/pillow-12.1.0-cp313-cp313t-win_arm64.whl", hash = "sha256:80941e6d573197a0c28f394753de529bb436b1ca990ed6e765cf42426abc39f8", size = 2454547, upload-time = "2026-01-02T09:12:18.704Z" },
+    { url = "https://files.pythonhosted.org/packages/8c/87/bdf971d8bbcf80a348cc3bacfcb239f5882100fe80534b0ce67a784181d8/pillow-12.1.0-cp314-cp314-ios_13_0_arm64_iphoneos.whl", hash = "sha256:5cb7bc1966d031aec37ddb9dcf15c2da5b2e9f7cc3ca7c54473a20a927e1eb91", size = 4062533, upload-time = "2026-01-02T09:12:20.791Z" },
+    { url = "https://files.pythonhosted.org/packages/ff/4f/5eb37a681c68d605eb7034c004875c81f86ec9ef51f5be4a63eadd58859a/pillow-12.1.0-cp314-cp314-ios_13_0_arm64_iphonesimulator.whl", hash = "sha256:97e9993d5ed946aba26baf9c1e8cf18adbab584b99f452ee72f7ee8acb882796", size = 4138546, upload-time = "2026-01-02T09:12:23.664Z" },
+    { url = "https://files.pythonhosted.org/packages/11/6d/19a95acb2edbace40dcd582d077b991646b7083c41b98da4ed7555b59733/pillow-12.1.0-cp314-cp314-ios_13_0_x86_64_iphonesimulator.whl", hash = "sha256:414b9a78e14ffeb98128863314e62c3f24b8a86081066625700b7985b3f529bd", size = 3601163, upload-time = "2026-01-02T09:12:26.338Z" },
+    { url = "https://files.pythonhosted.org/packages/fc/36/2b8138e51cb42e4cc39c3297713455548be855a50558c3ac2beebdc251dd/pillow-12.1.0-cp314-cp314-macosx_10_15_x86_64.whl", hash = "sha256:e6bdb408f7c9dd2a5ff2b14a3b0bb6d4deb29fb9961e6eb3ae2031ae9a5cec13", size = 5266086, upload-time = "2026-01-02T09:12:28.782Z" },
+    { url = "https://files.pythonhosted.org/packages/53/4b/649056e4d22e1caa90816bf99cef0884aed607ed38075bd75f091a607a38/pillow-12.1.0-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:3413c2ae377550f5487991d444428f1a8ae92784aac79caa8b1e3b89b175f77e", size = 4657344, upload-time = "2026-01-02T09:12:31.117Z" },
+    { url = "https://files.pythonhosted.org/packages/6c/6b/c5742cea0f1ade0cd61485dc3d81f05261fc2276f537fbdc00802de56779/pillow-12.1.0-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:e5dcbe95016e88437ecf33544ba5db21ef1b8dd6e1b434a2cb2a3d605299e643", size = 6232114, upload-time = "2026-01-02T09:12:32.936Z" },
+    { url = "https://files.pythonhosted.org/packages/bf/8f/9f521268ce22d63991601aafd3d48d5ff7280a246a1ef62d626d67b44064/pillow-12.1.0-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:d0a7735df32ccbcc98b98a1ac785cc4b19b580be1bdf0aeb5c03223220ea09d5", size = 8042708, upload-time = "2026-01-02T09:12:34.78Z" },
+    { url = "https://files.pythonhosted.org/packages/1a/eb/257f38542893f021502a1bbe0c2e883c90b5cff26cc33b1584a841a06d30/pillow-12.1.0-cp314-cp314-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:0c27407a2d1b96774cbc4a7594129cc027339fd800cd081e44497722ea1179de", size = 6347762, upload-time = "2026-01-02T09:12:36.748Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/5a/8ba375025701c09b309e8d5163c5a4ce0102fa86bbf8800eb0d7ac87bc51/pillow-12.1.0-cp314-cp314-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:15c794d74303828eaa957ff8070846d0efe8c630901a1c753fdc63850e19ecd9", size = 7039265, upload-time = "2026-01-02T09:12:39.082Z" },
+    { url = "https://files.pythonhosted.org/packages/cf/dc/cf5e4cdb3db533f539e88a7bbf9f190c64ab8a08a9bc7a4ccf55067872e4/pillow-12.1.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:c990547452ee2800d8506c4150280757f88532f3de2a58e3022e9b179107862a", size = 6462341, upload-time = "2026-01-02T09:12:40.946Z" },
+    { url = "https://files.pythonhosted.org/packages/d0/47/0291a25ac9550677e22eda48510cfc4fa4b2ef0396448b7fbdc0a6946309/pillow-12.1.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:b63e13dd27da389ed9475b3d28510f0f954bca0041e8e551b2a4eb1eab56a39a", size = 7165395, upload-time = "2026-01-02T09:12:42.706Z" },
+    { url = "https://files.pythonhosted.org/packages/4f/4c/e005a59393ec4d9416be06e6b45820403bb946a778e39ecec62f5b2b991e/pillow-12.1.0-cp314-cp314-win32.whl", hash = "sha256:1a949604f73eb07a8adab38c4fe50791f9919344398bdc8ac6b307f755fc7030", size = 6431413, upload-time = "2026-01-02T09:12:44.944Z" },
+    { url = "https://files.pythonhosted.org/packages/1c/af/f23697f587ac5f9095d67e31b81c95c0249cd461a9798a061ed6709b09b5/pillow-12.1.0-cp314-cp314-win_amd64.whl", hash = "sha256:4f9f6a650743f0ddee5593ac9e954ba1bdbc5e150bc066586d4f26127853ab94", size = 7176779, upload-time = "2026-01-02T09:12:46.727Z" },
+    { url = "https://files.pythonhosted.org/packages/b3/36/6a51abf8599232f3e9afbd16d52829376a68909fe14efe29084445db4b73/pillow-12.1.0-cp314-cp314-win_arm64.whl", hash = "sha256:808b99604f7873c800c4840f55ff389936ef1948e4e87645eaf3fccbc8477ac4", size = 2543105, upload-time = "2026-01-02T09:12:49.243Z" },
+    { url = "https://files.pythonhosted.org/packages/82/54/2e1dd20c8749ff225080d6ba465a0cab4387f5db0d1c5fb1439e2d99923f/pillow-12.1.0-cp314-cp314t-macosx_10_15_x86_64.whl", hash = "sha256:bc11908616c8a283cf7d664f77411a5ed2a02009b0097ff8abbba5e79128ccf2", size = 5268571, upload-time = "2026-01-02T09:12:51.11Z" },
+    { url = "https://files.pythonhosted.org/packages/57/61/571163a5ef86ec0cf30d265ac2a70ae6fc9e28413d1dc94fa37fae6bda89/pillow-12.1.0-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:896866d2d436563fa2a43a9d72f417874f16b5545955c54a64941e87c1376c61", size = 4660426, upload-time = "2026-01-02T09:12:52.865Z" },
+    { url = "https://files.pythonhosted.org/packages/5e/e1/53ee5163f794aef1bf84243f755ee6897a92c708505350dd1923f4afec48/pillow-12.1.0-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:8e178e3e99d3c0ea8fc64b88447f7cac8ccf058af422a6cedc690d0eadd98c51", size = 6269908, upload-time = "2026-01-02T09:12:54.884Z" },
+    { url = "https://files.pythonhosted.org/packages/bc/0b/b4b4106ff0ee1afa1dc599fde6ab230417f800279745124f6c50bcffed8e/pillow-12.1.0-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:079af2fb0c599c2ec144ba2c02766d1b55498e373b3ac64687e43849fbbef5bc", size = 8074733, upload-time = "2026-01-02T09:12:56.802Z" },
+    { url = "https://files.pythonhosted.org/packages/19/9f/80b411cbac4a732439e629a26ad3ef11907a8c7fc5377b7602f04f6fe4e7/pillow-12.1.0-cp314-cp314t-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:bdec5e43377761c5dbca620efb69a77f6855c5a379e32ac5b158f54c84212b14", size = 6381431, upload-time = "2026-01-02T09:12:58.823Z" },
+    { url = "https://files.pythonhosted.org/packages/8f/b7/d65c45db463b66ecb6abc17c6ba6917a911202a07662247e1355ce1789e7/pillow-12.1.0-cp314-cp314t-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:565c986f4b45c020f5421a4cea13ef294dde9509a8577f29b2fc5edc7587fff8", size = 7068529, upload-time = "2026-01-02T09:13:00.885Z" },
+    { url = "https://files.pythonhosted.org/packages/50/96/dfd4cd726b4a45ae6e3c669fc9e49deb2241312605d33aba50499e9d9bd1/pillow-12.1.0-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:43aca0a55ce1eefc0aefa6253661cb54571857b1a7b2964bd8a1e3ef4b729924", size = 6492981, upload-time = "2026-01-02T09:13:03.314Z" },
+    { url = "https://files.pythonhosted.org/packages/4d/1c/b5dc52cf713ae46033359c5ca920444f18a6359ce1020dd3e9c553ea5bc6/pillow-12.1.0-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:0deedf2ea233722476b3a81e8cdfbad786f7adbed5d848469fa59fe52396e4ef", size = 7191878, upload-time = "2026-01-02T09:13:05.276Z" },
+    { url = "https://files.pythonhosted.org/packages/53/26/c4188248bd5edaf543864fe4834aebe9c9cb4968b6f573ce014cc42d0720/pillow-12.1.0-cp314-cp314t-win32.whl", hash = "sha256:b17fbdbe01c196e7e159aacb889e091f28e61020a8abeac07b68079b6e626988", size = 6438703, upload-time = "2026-01-02T09:13:07.491Z" },
+    { url = "https://files.pythonhosted.org/packages/b8/0e/69ed296de8ea05cb03ee139cee600f424ca166e632567b2d66727f08c7ed/pillow-12.1.0-cp314-cp314t-win_amd64.whl", hash = "sha256:27b9baecb428899db6c0de572d6d305cfaf38ca1596b5c0542a5182e3e74e8c6", size = 7182927, upload-time = "2026-01-02T09:13:09.841Z" },
+    { url = "https://files.pythonhosted.org/packages/fc/f5/68334c015eed9b5cff77814258717dec591ded209ab5b6fb70e2ae873d1d/pillow-12.1.0-cp314-cp314t-win_arm64.whl", hash = "sha256:f61333d817698bdcdd0f9d7793e365ac3d2a21c1f1eb02b32ad6aefb8d8ea831", size = 2545104, upload-time = "2026-01-02T09:13:12.068Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/bc/224b1d98cffd7164b14707c91aac83c07b047fbd8f58eba4066a3e53746a/pillow-12.1.0-pp311-pypy311_pp73-macosx_10_15_x86_64.whl", hash = "sha256:ca94b6aac0d7af2a10ba08c0f888b3d5114439b6b3ef39968378723622fed377", size = 5228605, upload-time = "2026-01-02T09:13:14.084Z" },
+    { url = "https://files.pythonhosted.org/packages/0c/ca/49ca7769c4550107de049ed85208240ba0f330b3f2e316f24534795702ce/pillow-12.1.0-pp311-pypy311_pp73-macosx_11_0_arm64.whl", hash = "sha256:351889afef0f485b84078ea40fe33727a0492b9af3904661b0abbafee0355b72", size = 4622245, upload-time = "2026-01-02T09:13:15.964Z" },
+    { url = "https://files.pythonhosted.org/packages/73/48/fac807ce82e5955bcc2718642b94b1bd22a82a6d452aea31cbb678cddf12/pillow-12.1.0-pp311-pypy311_pp73-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:bb0984b30e973f7e2884362b7d23d0a348c7143ee559f38ef3eaab640144204c", size = 5247593, upload-time = "2026-01-02T09:13:17.913Z" },
+    { url = "https://files.pythonhosted.org/packages/d2/95/3e0742fe358c4664aed4fd05d5f5373dcdad0b27af52aa0972568541e3f4/pillow-12.1.0-pp311-pypy311_pp73-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:84cabc7095dd535ca934d57e9ce2a72ffd216e435a84acb06b2277b1de2689bd", size = 6989008, upload-time = "2026-01-02T09:13:20.083Z" },
+    { url = "https://files.pythonhosted.org/packages/5a/74/fe2ac378e4e202e56d50540d92e1ef4ff34ed687f3c60f6a121bcf99437e/pillow-12.1.0-pp311-pypy311_pp73-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:53d8b764726d3af1a138dd353116f774e3862ec7e3794e0c8781e30db0f35dfc", size = 5313824, upload-time = "2026-01-02T09:13:22.405Z" },
+    { url = "https://files.pythonhosted.org/packages/f3/77/2a60dee1adee4e2655ac328dd05c02a955c1cd683b9f1b82ec3feb44727c/pillow-12.1.0-pp311-pypy311_pp73-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:5da841d81b1a05ef940a8567da92decaa15bc4d7dedb540a8c219ad83d91808a", size = 5963278, upload-time = "2026-01-02T09:13:24.706Z" },
+    { url = "https://files.pythonhosted.org/packages/2d/71/64e9b1c7f04ae0027f788a248e6297d7fcc29571371fe7d45495a78172c0/pillow-12.1.0-pp311-pypy311_pp73-win_amd64.whl", hash = "sha256:75af0b4c229ac519b155028fa1be632d812a519abba9b46b20e50c6caa184f19", size = 7029809, upload-time = "2026-01-02T09:13:26.541Z" },
+]
+
+[[package]]
+name = "pluggy"
+version = "1.6.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/f9/e2/3e91f31a7d2b083fe6ef3fa267035b518369d9511ffab804f839851d2779/pluggy-1.6.0.tar.gz", hash = "sha256:7dcc130b76258d33b90f61b658791dede3486c3e6bfb003ee5c9bfb396dd22f3", size = 69412, upload-time = "2025-05-15T12:30:07.975Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/54/20/4d324d65cc6d9205fabedc306948156824eb9f0ee1633355a8f7ec5c66bf/pluggy-1.6.0-py3-none-any.whl", hash = "sha256:e920276dd6813095e9377c0bc5566d94c932c33b27a3e3945d8389c374dd4746", size = 20538, upload-time = "2025-05-15T12:30:06.134Z" },
+]
+
+[[package]]
+name = "prettytable"
+version = "3.17.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "wcwidth" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/79/45/b0847d88d6cfeb4413566738c8bbf1e1995fad3d42515327ff32cc1eb578/prettytable-3.17.0.tar.gz", hash = "sha256:59f2590776527f3c9e8cf9fe7b66dd215837cca96a9c39567414cbc632e8ddb0", size = 67892, upload-time = "2025-11-14T17:33:20.212Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/ee/8c/83087ebc47ab0396ce092363001fa37c17153119ee282700c0713a195853/prettytable-3.17.0-py3-none-any.whl", hash = "sha256:aad69b294ddbe3e1f95ef8886a060ed1666a0b83018bbf56295f6f226c43d287", size = 34433, upload-time = "2025-11-14T17:33:19.093Z" },
+]
+
+[[package]]
+name = "propcache"
+version = "0.4.1"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/9e/da/e9fc233cf63743258bff22b3dfa7ea5baef7b5bc324af47a0ad89b8ffc6f/propcache-0.4.1.tar.gz", hash = "sha256:f48107a8c637e80362555f37ecf49abe20370e557cc4ab374f04ec4423c97c3d", size = 46442, upload-time = "2025-10-08T19:49:02.291Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/3c/0e/934b541323035566a9af292dba85a195f7b78179114f2c6ebb24551118a9/propcache-0.4.1-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:7c2d1fa3201efaf55d730400d945b5b3ab6e672e100ba0f9a409d950ab25d7db", size = 79534, upload-time = "2025-10-08T19:46:02.083Z" },
+    { url = "https://files.pythonhosted.org/packages/a1/6b/db0d03d96726d995dc7171286c6ba9d8d14251f37433890f88368951a44e/propcache-0.4.1-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:1eb2994229cc8ce7fe9b3db88f5465f5fd8651672840b2e426b88cdb1a30aac8", size = 45526, upload-time = "2025-10-08T19:46:03.884Z" },
+    { url = "https://files.pythonhosted.org/packages/e4/c3/82728404aea669e1600f304f2609cde9e665c18df5a11cdd57ed73c1dceb/propcache-0.4.1-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:66c1f011f45a3b33d7bcb22daed4b29c0c9e2224758b6be00686731e1b46f925", size = 47263, upload-time = "2025-10-08T19:46:05.405Z" },
+    { url = "https://files.pythonhosted.org/packages/df/1b/39313ddad2bf9187a1432654c38249bab4562ef535ef07f5eb6eb04d0b1b/propcache-0.4.1-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:9a52009f2adffe195d0b605c25ec929d26b36ef986ba85244891dee3b294df21", size = 201012, upload-time = "2025-10-08T19:46:07.165Z" },
+    { url = "https://files.pythonhosted.org/packages/5b/01/f1d0b57d136f294a142acf97f4ed58c8e5b974c21e543000968357115011/propcache-0.4.1-cp310-cp310-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:5d4e2366a9c7b837555cf02fb9be2e3167d333aff716332ef1b7c3a142ec40c5", size = 209491, upload-time = "2025-10-08T19:46:08.909Z" },
+    { url = "https://files.pythonhosted.org/packages/a1/c8/038d909c61c5bb039070b3fb02ad5cccdb1dde0d714792e251cdb17c9c05/propcache-0.4.1-cp310-cp310-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:9d2b6caef873b4f09e26ea7e33d65f42b944837563a47a94719cc3544319a0db", size = 215319, upload-time = "2025-10-08T19:46:10.7Z" },
+    { url = "https://files.pythonhosted.org/packages/08/57/8c87e93142b2c1fa2408e45695205a7ba05fb5db458c0bf5c06ba0e09ea6/propcache-0.4.1-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:2b16ec437a8c8a965ecf95739448dd938b5c7f56e67ea009f4300d8df05f32b7", size = 196856, upload-time = "2025-10-08T19:46:12.003Z" },
+    { url = "https://files.pythonhosted.org/packages/42/df/5615fec76aa561987a534759b3686008a288e73107faa49a8ae5795a9f7a/propcache-0.4.1-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:296f4c8ed03ca7476813fe666c9ea97869a8d7aec972618671b33a38a5182ef4", size = 193241, upload-time = "2025-10-08T19:46:13.495Z" },
+    { url = "https://files.pythonhosted.org/packages/d5/21/62949eb3a7a54afe8327011c90aca7e03547787a88fb8bd9726806482fea/propcache-0.4.1-cp310-cp310-musllinux_1_2_armv7l.whl", hash = "sha256:1f0978529a418ebd1f49dad413a2b68af33f85d5c5ca5c6ca2a3bed375a7ac60", size = 190552, upload-time = "2025-10-08T19:46:14.938Z" },
+    { url = "https://files.pythonhosted.org/packages/30/ee/ab4d727dd70806e5b4de96a798ae7ac6e4d42516f030ee60522474b6b332/propcache-0.4.1-cp310-cp310-musllinux_1_2_ppc64le.whl", hash = "sha256:fd138803047fb4c062b1c1dd95462f5209456bfab55c734458f15d11da288f8f", size = 200113, upload-time = "2025-10-08T19:46:16.695Z" },
+    { url = "https://files.pythonhosted.org/packages/8a/0b/38b46208e6711b016aa8966a3ac793eee0d05c7159d8342aa27fc0bc365e/propcache-0.4.1-cp310-cp310-musllinux_1_2_s390x.whl", hash = "sha256:8c9b3cbe4584636d72ff556d9036e0c9317fa27b3ac1f0f558e7e84d1c9c5900", size = 200778, upload-time = "2025-10-08T19:46:18.023Z" },
+    { url = "https://files.pythonhosted.org/packages/cf/81/5abec54355ed344476bee711e9f04815d4b00a311ab0535599204eecc257/propcache-0.4.1-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:f93243fdc5657247533273ac4f86ae106cc6445a0efacb9a1bfe982fcfefd90c", size = 193047, upload-time = "2025-10-08T19:46:19.449Z" },
+    { url = "https://files.pythonhosted.org/packages/ec/b6/1f237c04e32063cb034acd5f6ef34ef3a394f75502e72703545631ab1ef6/propcache-0.4.1-cp310-cp310-win32.whl", hash = "sha256:a0ee98db9c5f80785b266eb805016e36058ac72c51a064040f2bc43b61101cdb", size = 38093, upload-time = "2025-10-08T19:46:20.643Z" },
+    { url = "https://files.pythonhosted.org/packages/a6/67/354aac4e0603a15f76439caf0427781bcd6797f370377f75a642133bc954/propcache-0.4.1-cp310-cp310-win_amd64.whl", hash = "sha256:1cdb7988c4e5ac7f6d175a28a9aa0c94cb6f2ebe52756a3c0cda98d2809a9e37", size = 41638, upload-time = "2025-10-08T19:46:21.935Z" },
+    { url = "https://files.pythonhosted.org/packages/e0/e1/74e55b9fd1a4c209ff1a9a824bf6c8b3d1fc5a1ac3eabe23462637466785/propcache-0.4.1-cp310-cp310-win_arm64.whl", hash = "sha256:d82ad62b19645419fe79dd63b3f9253e15b30e955c0170e5cebc350c1844e581", size = 38229, upload-time = "2025-10-08T19:46:23.368Z" },
+    { url = "https://files.pythonhosted.org/packages/8c/d4/4e2c9aaf7ac2242b9358f98dccd8f90f2605402f5afeff6c578682c2c491/propcache-0.4.1-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:60a8fda9644b7dfd5dece8c61d8a85e271cb958075bfc4e01083c148b61a7caf", size = 80208, upload-time = "2025-10-08T19:46:24.597Z" },
+    { url = "https://files.pythonhosted.org/packages/c2/21/d7b68e911f9c8e18e4ae43bdbc1e1e9bbd971f8866eb81608947b6f585ff/propcache-0.4.1-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:c30b53e7e6bda1d547cabb47c825f3843a0a1a42b0496087bb58d8fedf9f41b5", size = 45777, upload-time = "2025-10-08T19:46:25.733Z" },
+    { url = "https://files.pythonhosted.org/packages/d3/1d/11605e99ac8ea9435651ee71ab4cb4bf03f0949586246476a25aadfec54a/propcache-0.4.1-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:6918ecbd897443087a3b7cd978d56546a812517dcaaca51b49526720571fa93e", size = 47647, upload-time = "2025-10-08T19:46:27.304Z" },
+    { url = "https://files.pythonhosted.org/packages/58/1a/3c62c127a8466c9c843bccb503d40a273e5cc69838805f322e2826509e0d/propcache-0.4.1-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:3d902a36df4e5989763425a8ab9e98cd8ad5c52c823b34ee7ef307fd50582566", size = 214929, upload-time = "2025-10-08T19:46:28.62Z" },
+    { url = "https://files.pythonhosted.org/packages/56/b9/8fa98f850960b367c4b8fe0592e7fc341daa7a9462e925228f10a60cf74f/propcache-0.4.1-cp311-cp311-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:a9695397f85973bb40427dedddf70d8dc4a44b22f1650dd4af9eedf443d45165", size = 221778, upload-time = "2025-10-08T19:46:30.358Z" },
+    { url = "https://files.pythonhosted.org/packages/46/a6/0ab4f660eb59649d14b3d3d65c439421cf2f87fe5dd68591cbe3c1e78a89/propcache-0.4.1-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:2bb07ffd7eaad486576430c89f9b215f9e4be68c4866a96e97db9e97fead85dc", size = 228144, upload-time = "2025-10-08T19:46:32.607Z" },
+    { url = "https://files.pythonhosted.org/packages/52/6a/57f43e054fb3d3a56ac9fc532bc684fc6169a26c75c353e65425b3e56eef/propcache-0.4.1-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:fd6f30fdcf9ae2a70abd34da54f18da086160e4d7d9251f81f3da0ff84fc5a48", size = 210030, upload-time = "2025-10-08T19:46:33.969Z" },
+    { url = "https://files.pythonhosted.org/packages/40/e2/27e6feebb5f6b8408fa29f5efbb765cd54c153ac77314d27e457a3e993b7/propcache-0.4.1-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:fc38cba02d1acba4e2869eef1a57a43dfbd3d49a59bf90dda7444ec2be6a5570", size = 208252, upload-time = "2025-10-08T19:46:35.309Z" },
+    { url = "https://files.pythonhosted.org/packages/9e/f8/91c27b22ccda1dbc7967f921c42825564fa5336a01ecd72eb78a9f4f53c2/propcache-0.4.1-cp311-cp311-musllinux_1_2_armv7l.whl", hash = "sha256:67fad6162281e80e882fb3ec355398cf72864a54069d060321f6cd0ade95fe85", size = 202064, upload-time = "2025-10-08T19:46:36.993Z" },
+    { url = "https://files.pythonhosted.org/packages/f2/26/7f00bd6bd1adba5aafe5f4a66390f243acab58eab24ff1a08bebb2ef9d40/propcache-0.4.1-cp311-cp311-musllinux_1_2_ppc64le.whl", hash = "sha256:f10207adf04d08bec185bae14d9606a1444715bc99180f9331c9c02093e1959e", size = 212429, upload-time = "2025-10-08T19:46:38.398Z" },
+    { url = "https://files.pythonhosted.org/packages/84/89/fd108ba7815c1117ddca79c228f3f8a15fc82a73bca8b142eb5de13b2785/propcache-0.4.1-cp311-cp311-musllinux_1_2_s390x.whl", hash = "sha256:e9b0d8d0845bbc4cfcdcbcdbf5086886bc8157aa963c31c777ceff7846c77757", size = 216727, upload-time = "2025-10-08T19:46:39.732Z" },
+    { url = "https://files.pythonhosted.org/packages/79/37/3ec3f7e3173e73f1d600495d8b545b53802cbf35506e5732dd8578db3724/propcache-0.4.1-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:981333cb2f4c1896a12f4ab92a9cc8f09ea664e9b7dbdc4eff74627af3a11c0f", size = 205097, upload-time = "2025-10-08T19:46:41.025Z" },
+    { url = "https://files.pythonhosted.org/packages/61/b0/b2631c19793f869d35f47d5a3a56fb19e9160d3c119f15ac7344fc3ccae7/propcache-0.4.1-cp311-cp311-win32.whl", hash = "sha256:f1d2f90aeec838a52f1c1a32fe9a619fefd5e411721a9117fbf82aea638fe8a1", size = 38084, upload-time = "2025-10-08T19:46:42.693Z" },
+    { url = "https://files.pythonhosted.org/packages/f4/78/6cce448e2098e9f3bfc91bb877f06aa24b6ccace872e39c53b2f707c4648/propcache-0.4.1-cp311-cp311-win_amd64.whl", hash = "sha256:364426a62660f3f699949ac8c621aad6977be7126c5807ce48c0aeb8e7333ea6", size = 41637, upload-time = "2025-10-08T19:46:43.778Z" },
+    { url = "https://files.pythonhosted.org/packages/9c/e9/754f180cccd7f51a39913782c74717c581b9cc8177ad0e949f4d51812383/propcache-0.4.1-cp311-cp311-win_arm64.whl", hash = "sha256:e53f3a38d3510c11953f3e6a33f205c6d1b001129f972805ca9b42fc308bc239", size = 38064, upload-time = "2025-10-08T19:46:44.872Z" },
+    { url = "https://files.pythonhosted.org/packages/a2/0f/f17b1b2b221d5ca28b4b876e8bb046ac40466513960646bda8e1853cdfa2/propcache-0.4.1-cp312-cp312-macosx_10_13_universal2.whl", hash = "sha256:e153e9cd40cc8945138822807139367f256f89c6810c2634a4f6902b52d3b4e2", size = 80061, upload-time = "2025-10-08T19:46:46.075Z" },
+    { url = "https://files.pythonhosted.org/packages/76/47/8ccf75935f51448ba9a16a71b783eb7ef6b9ee60f5d14c7f8a8a79fbeed7/propcache-0.4.1-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:cd547953428f7abb73c5ad82cbb32109566204260d98e41e5dfdc682eb7f8403", size = 46037, upload-time = "2025-10-08T19:46:47.23Z" },
+    { url = "https://files.pythonhosted.org/packages/0a/b6/5c9a0e42df4d00bfb4a3cbbe5cf9f54260300c88a0e9af1f47ca5ce17ac0/propcache-0.4.1-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:f048da1b4f243fc44f205dfd320933a951b8d89e0afd4c7cacc762a8b9165207", size = 47324, upload-time = "2025-10-08T19:46:48.384Z" },
+    { url = "https://files.pythonhosted.org/packages/9e/d3/6c7ee328b39a81ee877c962469f1e795f9db87f925251efeb0545e0020d0/propcache-0.4.1-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:ec17c65562a827bba85e3872ead335f95405ea1674860d96483a02f5c698fa72", size = 225505, upload-time = "2025-10-08T19:46:50.055Z" },
+    { url = "https://files.pythonhosted.org/packages/01/5d/1c53f4563490b1d06a684742cc6076ef944bc6457df6051b7d1a877c057b/propcache-0.4.1-cp312-cp312-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:405aac25c6394ef275dee4c709be43745d36674b223ba4eb7144bf4d691b7367", size = 230242, upload-time = "2025-10-08T19:46:51.815Z" },
+    { url = "https://files.pythonhosted.org/packages/20/e1/ce4620633b0e2422207c3cb774a0ee61cac13abc6217763a7b9e2e3f4a12/propcache-0.4.1-cp312-cp312-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:0013cb6f8dde4b2a2f66903b8ba740bdfe378c943c4377a200551ceb27f379e4", size = 238474, upload-time = "2025-10-08T19:46:53.208Z" },
+    { url = "https://files.pythonhosted.org/packages/46/4b/3aae6835b8e5f44ea6a68348ad90f78134047b503765087be2f9912140ea/propcache-0.4.1-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:15932ab57837c3368b024473a525e25d316d8353016e7cc0e5ba9eb343fbb1cf", size = 221575, upload-time = "2025-10-08T19:46:54.511Z" },
+    { url = "https://files.pythonhosted.org/packages/6e/a5/8a5e8678bcc9d3a1a15b9a29165640d64762d424a16af543f00629c87338/propcache-0.4.1-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:031dce78b9dc099f4c29785d9cf5577a3faf9ebf74ecbd3c856a7b92768c3df3", size = 216736, upload-time = "2025-10-08T19:46:56.212Z" },
+    { url = "https://files.pythonhosted.org/packages/f1/63/b7b215eddeac83ca1c6b934f89d09a625aa9ee4ba158338854c87210cc36/propcache-0.4.1-cp312-cp312-musllinux_1_2_armv7l.whl", hash = "sha256:ab08df6c9a035bee56e31af99be621526bd237bea9f32def431c656b29e41778", size = 213019, upload-time = "2025-10-08T19:46:57.595Z" },
+    { url = "https://files.pythonhosted.org/packages/57/74/f580099a58c8af587cac7ba19ee7cb418506342fbbe2d4a4401661cca886/propcache-0.4.1-cp312-cp312-musllinux_1_2_ppc64le.whl", hash = "sha256:4d7af63f9f93fe593afbf104c21b3b15868efb2c21d07d8732c0c4287e66b6a6", size = 220376, upload-time = "2025-10-08T19:46:59.067Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/ee/542f1313aff7eaf19c2bb758c5d0560d2683dac001a1c96d0774af799843/propcache-0.4.1-cp312-cp312-musllinux_1_2_s390x.whl", hash = "sha256:cfc27c945f422e8b5071b6e93169679e4eb5bf73bbcbf1ba3ae3a83d2f78ebd9", size = 226988, upload-time = "2025-10-08T19:47:00.544Z" },
+    { url = "https://files.pythonhosted.org/packages/8f/18/9c6b015dd9c6930f6ce2229e1f02fb35298b847f2087ea2b436a5bfa7287/propcache-0.4.1-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:35c3277624a080cc6ec6f847cbbbb5b49affa3598c4535a0a4682a697aaa5c75", size = 215615, upload-time = "2025-10-08T19:47:01.968Z" },
+    { url = "https://files.pythonhosted.org/packages/80/9e/e7b85720b98c45a45e1fca6a177024934dc9bc5f4d5dd04207f216fc33ed/propcache-0.4.1-cp312-cp312-win32.whl", hash = "sha256:671538c2262dadb5ba6395e26c1731e1d52534bfe9ae56d0b5573ce539266aa8", size = 38066, upload-time = "2025-10-08T19:47:03.503Z" },
+    { url = "https://files.pythonhosted.org/packages/54/09/d19cff2a5aaac632ec8fc03737b223597b1e347416934c1b3a7df079784c/propcache-0.4.1-cp312-cp312-win_amd64.whl", hash = "sha256:cb2d222e72399fcf5890d1d5cc1060857b9b236adff2792ff48ca2dfd46c81db", size = 41655, upload-time = "2025-10-08T19:47:04.973Z" },
+    { url = "https://files.pythonhosted.org/packages/68/ab/6b5c191bb5de08036a8c697b265d4ca76148efb10fa162f14af14fb5f076/propcache-0.4.1-cp312-cp312-win_arm64.whl", hash = "sha256:204483131fb222bdaaeeea9f9e6c6ed0cac32731f75dfc1d4a567fc1926477c1", size = 37789, upload-time = "2025-10-08T19:47:06.077Z" },
+    { url = "https://files.pythonhosted.org/packages/bf/df/6d9c1b6ac12b003837dde8a10231a7344512186e87b36e855bef32241942/propcache-0.4.1-cp313-cp313-macosx_10_13_universal2.whl", hash = "sha256:43eedf29202c08550aac1d14e0ee619b0430aaef78f85864c1a892294fbc28cf", size = 77750, upload-time = "2025-10-08T19:47:07.648Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/e8/677a0025e8a2acf07d3418a2e7ba529c9c33caf09d3c1f25513023c1db56/propcache-0.4.1-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:d62cdfcfd89ccb8de04e0eda998535c406bf5e060ffd56be6c586cbcc05b3311", size = 44780, upload-time = "2025-10-08T19:47:08.851Z" },
+    { url = "https://files.pythonhosted.org/packages/89/a4/92380f7ca60f99ebae761936bc48a72a639e8a47b29050615eef757cb2a7/propcache-0.4.1-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:cae65ad55793da34db5f54e4029b89d3b9b9490d8abe1b4c7ab5d4b8ec7ebf74", size = 46308, upload-time = "2025-10-08T19:47:09.982Z" },
+    { url = "https://files.pythonhosted.org/packages/2d/48/c5ac64dee5262044348d1d78a5f85dd1a57464a60d30daee946699963eb3/propcache-0.4.1-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:333ddb9031d2704a301ee3e506dc46b1fe5f294ec198ed6435ad5b6a085facfe", size = 208182, upload-time = "2025-10-08T19:47:11.319Z" },
+    { url = "https://files.pythonhosted.org/packages/c6/0c/cd762dd011a9287389a6a3eb43aa30207bde253610cca06824aeabfe9653/propcache-0.4.1-cp313-cp313-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:fd0858c20f078a32cf55f7e81473d96dcf3b93fd2ccdb3d40fdf54b8573df3af", size = 211215, upload-time = "2025-10-08T19:47:13.146Z" },
+    { url = "https://files.pythonhosted.org/packages/30/3e/49861e90233ba36890ae0ca4c660e95df565b2cd15d4a68556ab5865974e/propcache-0.4.1-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:678ae89ebc632c5c204c794f8dab2837c5f159aeb59e6ed0539500400577298c", size = 218112, upload-time = "2025-10-08T19:47:14.913Z" },
+    { url = "https://files.pythonhosted.org/packages/f1/8b/544bc867e24e1bd48f3118cecd3b05c694e160a168478fa28770f22fd094/propcache-0.4.1-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:d472aeb4fbf9865e0c6d622d7f4d54a4e101a89715d8904282bb5f9a2f476c3f", size = 204442, upload-time = "2025-10-08T19:47:16.277Z" },
+    { url = "https://files.pythonhosted.org/packages/50/a6/4282772fd016a76d3e5c0df58380a5ea64900afd836cec2c2f662d1b9bb3/propcache-0.4.1-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:4d3df5fa7e36b3225954fba85589da77a0fe6a53e3976de39caf04a0db4c36f1", size = 199398, upload-time = "2025-10-08T19:47:17.962Z" },
+    { url = "https://files.pythonhosted.org/packages/3e/ec/d8a7cd406ee1ddb705db2139f8a10a8a427100347bd698e7014351c7af09/propcache-0.4.1-cp313-cp313-musllinux_1_2_armv7l.whl", hash = "sha256:ee17f18d2498f2673e432faaa71698032b0127ebf23ae5974eeaf806c279df24", size = 196920, upload-time = "2025-10-08T19:47:19.355Z" },
+    { url = "https://files.pythonhosted.org/packages/f6/6c/f38ab64af3764f431e359f8baf9e0a21013e24329e8b85d2da32e8ed07ca/propcache-0.4.1-cp313-cp313-musllinux_1_2_ppc64le.whl", hash = "sha256:580e97762b950f993ae618e167e7be9256b8353c2dcd8b99ec100eb50f5286aa", size = 203748, upload-time = "2025-10-08T19:47:21.338Z" },
+    { url = "https://files.pythonhosted.org/packages/d6/e3/fa846bd70f6534d647886621388f0a265254d30e3ce47e5c8e6e27dbf153/propcache-0.4.1-cp313-cp313-musllinux_1_2_s390x.whl", hash = "sha256:501d20b891688eb8e7aa903021f0b72d5a55db40ffaab27edefd1027caaafa61", size = 205877, upload-time = "2025-10-08T19:47:23.059Z" },
+    { url = "https://files.pythonhosted.org/packages/e2/39/8163fc6f3133fea7b5f2827e8eba2029a0277ab2c5beee6c1db7b10fc23d/propcache-0.4.1-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:9a0bd56e5b100aef69bd8562b74b46254e7c8812918d3baa700c8a8009b0af66", size = 199437, upload-time = "2025-10-08T19:47:24.445Z" },
+    { url = "https://files.pythonhosted.org/packages/93/89/caa9089970ca49c7c01662bd0eeedfe85494e863e8043565aeb6472ce8fe/propcache-0.4.1-cp313-cp313-win32.whl", hash = "sha256:bcc9aaa5d80322bc2fb24bb7accb4a30f81e90ab8d6ba187aec0744bc302ad81", size = 37586, upload-time = "2025-10-08T19:47:25.736Z" },
+    { url = "https://files.pythonhosted.org/packages/f5/ab/f76ec3c3627c883215b5c8080debb4394ef5a7a29be811f786415fc1e6fd/propcache-0.4.1-cp313-cp313-win_amd64.whl", hash = "sha256:381914df18634f5494334d201e98245c0596067504b9372d8cf93f4bb23e025e", size = 40790, upload-time = "2025-10-08T19:47:26.847Z" },
+    { url = "https://files.pythonhosted.org/packages/59/1b/e71ae98235f8e2ba5004d8cb19765a74877abf189bc53fc0c80d799e56c3/propcache-0.4.1-cp313-cp313-win_arm64.whl", hash = "sha256:8873eb4460fd55333ea49b7d189749ecf6e55bf85080f11b1c4530ed3034cba1", size = 37158, upload-time = "2025-10-08T19:47:27.961Z" },
+    { url = "https://files.pythonhosted.org/packages/83/ce/a31bbdfc24ee0dcbba458c8175ed26089cf109a55bbe7b7640ed2470cfe9/propcache-0.4.1-cp313-cp313t-macosx_10_13_universal2.whl", hash = "sha256:92d1935ee1f8d7442da9c0c4fa7ac20d07e94064184811b685f5c4fada64553b", size = 81451, upload-time = "2025-10-08T19:47:29.445Z" },
+    { url = "https://files.pythonhosted.org/packages/25/9c/442a45a470a68456e710d96cacd3573ef26a1d0a60067e6a7d5e655621ed/propcache-0.4.1-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:473c61b39e1460d386479b9b2f337da492042447c9b685f28be4f74d3529e566", size = 46374, upload-time = "2025-10-08T19:47:30.579Z" },
+    { url = "https://files.pythonhosted.org/packages/f4/bf/b1d5e21dbc3b2e889ea4327044fb16312a736d97640fb8b6aa3f9c7b3b65/propcache-0.4.1-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:c0ef0aaafc66fbd87842a3fe3902fd889825646bc21149eafe47be6072725835", size = 48396, upload-time = "2025-10-08T19:47:31.79Z" },
+    { url = "https://files.pythonhosted.org/packages/f4/04/5b4c54a103d480e978d3c8a76073502b18db0c4bc17ab91b3cb5092ad949/propcache-0.4.1-cp313-cp313t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:f95393b4d66bfae908c3ca8d169d5f79cd65636ae15b5e7a4f6e67af675adb0e", size = 275950, upload-time = "2025-10-08T19:47:33.481Z" },
+    { url = "https://files.pythonhosted.org/packages/b4/c1/86f846827fb969c4b78b0af79bba1d1ea2156492e1b83dea8b8a6ae27395/propcache-0.4.1-cp313-cp313t-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:c07fda85708bc48578467e85099645167a955ba093be0a2dcba962195676e859", size = 273856, upload-time = "2025-10-08T19:47:34.906Z" },
+    { url = "https://files.pythonhosted.org/packages/36/1d/fc272a63c8d3bbad6878c336c7a7dea15e8f2d23a544bda43205dfa83ada/propcache-0.4.1-cp313-cp313t-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:af223b406d6d000830c6f65f1e6431783fc3f713ba3e6cc8c024d5ee96170a4b", size = 280420, upload-time = "2025-10-08T19:47:36.338Z" },
+    { url = "https://files.pythonhosted.org/packages/07/0c/01f2219d39f7e53d52e5173bcb09c976609ba30209912a0680adfb8c593a/propcache-0.4.1-cp313-cp313t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:a78372c932c90ee474559c5ddfffd718238e8673c340dc21fe45c5b8b54559a0", size = 263254, upload-time = "2025-10-08T19:47:37.692Z" },
+    { url = "https://files.pythonhosted.org/packages/2d/18/cd28081658ce597898f0c4d174d4d0f3c5b6d4dc27ffafeef835c95eb359/propcache-0.4.1-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:564d9f0d4d9509e1a870c920a89b2fec951b44bf5ba7d537a9e7c1ccec2c18af", size = 261205, upload-time = "2025-10-08T19:47:39.659Z" },
+    { url = "https://files.pythonhosted.org/packages/7a/71/1f9e22eb8b8316701c2a19fa1f388c8a3185082607da8e406a803c9b954e/propcache-0.4.1-cp313-cp313t-musllinux_1_2_armv7l.whl", hash = "sha256:17612831fda0138059cc5546f4d12a2aacfb9e47068c06af35c400ba58ba7393", size = 247873, upload-time = "2025-10-08T19:47:41.084Z" },
+    { url = "https://files.pythonhosted.org/packages/4a/65/3d4b61f36af2b4eddba9def857959f1016a51066b4f1ce348e0cf7881f58/propcache-0.4.1-cp313-cp313t-musllinux_1_2_ppc64le.whl", hash = "sha256:41a89040cb10bd345b3c1a873b2bf36413d48da1def52f268a055f7398514874", size = 262739, upload-time = "2025-10-08T19:47:42.51Z" },
+    { url = "https://files.pythonhosted.org/packages/2a/42/26746ab087faa77c1c68079b228810436ccd9a5ce9ac85e2b7307195fd06/propcache-0.4.1-cp313-cp313t-musllinux_1_2_s390x.whl", hash = "sha256:e35b88984e7fa64aacecea39236cee32dd9bd8c55f57ba8a75cf2399553f9bd7", size = 263514, upload-time = "2025-10-08T19:47:43.927Z" },
+    { url = "https://files.pythonhosted.org/packages/94/13/630690fe201f5502d2403dd3cfd451ed8858fe3c738ee88d095ad2ff407b/propcache-0.4.1-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:6f8b465489f927b0df505cbe26ffbeed4d6d8a2bbc61ce90eb074ff129ef0ab1", size = 257781, upload-time = "2025-10-08T19:47:45.448Z" },
+    { url = "https://files.pythonhosted.org/packages/92/f7/1d4ec5841505f423469efbfc381d64b7b467438cd5a4bbcbb063f3b73d27/propcache-0.4.1-cp313-cp313t-win32.whl", hash = "sha256:2ad890caa1d928c7c2965b48f3a3815c853180831d0e5503d35cf00c472f4717", size = 41396, upload-time = "2025-10-08T19:47:47.202Z" },
+    { url = "https://files.pythonhosted.org/packages/48/f0/615c30622316496d2cbbc29f5985f7777d3ada70f23370608c1d3e081c1f/propcache-0.4.1-cp313-cp313t-win_amd64.whl", hash = "sha256:f7ee0e597f495cf415bcbd3da3caa3bd7e816b74d0d52b8145954c5e6fd3ff37", size = 44897, upload-time = "2025-10-08T19:47:48.336Z" },
+    { url = "https://files.pythonhosted.org/packages/fd/ca/6002e46eccbe0e33dcd4069ef32f7f1c9e243736e07adca37ae8c4830ec3/propcache-0.4.1-cp313-cp313t-win_arm64.whl", hash = "sha256:929d7cbe1f01bb7baffb33dc14eb5691c95831450a26354cd210a8155170c93a", size = 39789, upload-time = "2025-10-08T19:47:49.876Z" },
+    { url = "https://files.pythonhosted.org/packages/8e/5c/bca52d654a896f831b8256683457ceddd490ec18d9ec50e97dfd8fc726a8/propcache-0.4.1-cp314-cp314-macosx_10_13_universal2.whl", hash = "sha256:3f7124c9d820ba5548d431afb4632301acf965db49e666aa21c305cbe8c6de12", size = 78152, upload-time = "2025-10-08T19:47:51.051Z" },
+    { url = "https://files.pythonhosted.org/packages/65/9b/03b04e7d82a5f54fb16113d839f5ea1ede58a61e90edf515f6577c66fa8f/propcache-0.4.1-cp314-cp314-macosx_10_13_x86_64.whl", hash = "sha256:c0d4b719b7da33599dfe3b22d3db1ef789210a0597bc650b7cee9c77c2be8c5c", size = 44869, upload-time = "2025-10-08T19:47:52.594Z" },
+    { url = "https://files.pythonhosted.org/packages/b2/fa/89a8ef0468d5833a23fff277b143d0573897cf75bd56670a6d28126c7d68/propcache-0.4.1-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:9f302f4783709a78240ebc311b793f123328716a60911d667e0c036bc5dcbded", size = 46596, upload-time = "2025-10-08T19:47:54.073Z" },
+    { url = "https://files.pythonhosted.org/packages/86/bd/47816020d337f4a746edc42fe8d53669965138f39ee117414c7d7a340cfe/propcache-0.4.1-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:c80ee5802e3fb9ea37938e7eecc307fb984837091d5fd262bb37238b1ae97641", size = 206981, upload-time = "2025-10-08T19:47:55.715Z" },
+    { url = "https://files.pythonhosted.org/packages/df/f6/c5fa1357cc9748510ee55f37173eb31bfde6d94e98ccd9e6f033f2fc06e1/propcache-0.4.1-cp314-cp314-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:ed5a841e8bb29a55fb8159ed526b26adc5bdd7e8bd7bf793ce647cb08656cdf4", size = 211490, upload-time = "2025-10-08T19:47:57.499Z" },
+    { url = "https://files.pythonhosted.org/packages/80/1e/e5889652a7c4a3846683401a48f0f2e5083ce0ec1a8a5221d8058fbd1adf/propcache-0.4.1-cp314-cp314-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:55c72fd6ea2da4c318e74ffdf93c4fe4e926051133657459131a95c846d16d44", size = 215371, upload-time = "2025-10-08T19:47:59.317Z" },
+    { url = "https://files.pythonhosted.org/packages/b2/f2/889ad4b2408f72fe1a4f6a19491177b30ea7bf1a0fd5f17050ca08cfc882/propcache-0.4.1-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:8326e144341460402713f91df60ade3c999d601e7eb5ff8f6f7862d54de0610d", size = 201424, upload-time = "2025-10-08T19:48:00.67Z" },
+    { url = "https://files.pythonhosted.org/packages/27/73/033d63069b57b0812c8bd19f311faebeceb6ba31b8f32b73432d12a0b826/propcache-0.4.1-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:060b16ae65bc098da7f6d25bf359f1f31f688384858204fe5d652979e0015e5b", size = 197566, upload-time = "2025-10-08T19:48:02.604Z" },
+    { url = "https://files.pythonhosted.org/packages/dc/89/ce24f3dc182630b4e07aa6d15f0ff4b14ed4b9955fae95a0b54c58d66c05/propcache-0.4.1-cp314-cp314-musllinux_1_2_armv7l.whl", hash = "sha256:89eb3fa9524f7bec9de6e83cf3faed9d79bffa560672c118a96a171a6f55831e", size = 193130, upload-time = "2025-10-08T19:48:04.499Z" },
+    { url = "https://files.pythonhosted.org/packages/a9/24/ef0d5fd1a811fb5c609278d0209c9f10c35f20581fcc16f818da959fc5b4/propcache-0.4.1-cp314-cp314-musllinux_1_2_ppc64le.whl", hash = "sha256:dee69d7015dc235f526fe80a9c90d65eb0039103fe565776250881731f06349f", size = 202625, upload-time = "2025-10-08T19:48:06.213Z" },
+    { url = "https://files.pythonhosted.org/packages/f5/02/98ec20ff5546f68d673df2f7a69e8c0d076b5abd05ca882dc7ee3a83653d/propcache-0.4.1-cp314-cp314-musllinux_1_2_s390x.whl", hash = "sha256:5558992a00dfd54ccbc64a32726a3357ec93825a418a401f5cc67df0ac5d9e49", size = 204209, upload-time = "2025-10-08T19:48:08.432Z" },
+    { url = "https://files.pythonhosted.org/packages/a0/87/492694f76759b15f0467a2a93ab68d32859672b646aa8a04ce4864e7932d/propcache-0.4.1-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:c9b822a577f560fbd9554812526831712c1436d2c046cedee4c3796d3543b144", size = 197797, upload-time = "2025-10-08T19:48:09.968Z" },
+    { url = "https://files.pythonhosted.org/packages/ee/36/66367de3575db1d2d3f3d177432bd14ee577a39d3f5d1b3d5df8afe3b6e2/propcache-0.4.1-cp314-cp314-win32.whl", hash = "sha256:ab4c29b49d560fe48b696cdcb127dd36e0bc2472548f3bf56cc5cb3da2b2984f", size = 38140, upload-time = "2025-10-08T19:48:11.232Z" },
+    { url = "https://files.pythonhosted.org/packages/0c/2a/a758b47de253636e1b8aef181c0b4f4f204bf0dd964914fb2af90a95b49b/propcache-0.4.1-cp314-cp314-win_amd64.whl", hash = "sha256:5a103c3eb905fcea0ab98be99c3a9a5ab2de60228aa5aceedc614c0281cf6153", size = 41257, upload-time = "2025-10-08T19:48:12.707Z" },
+    { url = "https://files.pythonhosted.org/packages/34/5e/63bd5896c3fec12edcbd6f12508d4890d23c265df28c74b175e1ef9f4f3b/propcache-0.4.1-cp314-cp314-win_arm64.whl", hash = "sha256:74c1fb26515153e482e00177a1ad654721bf9207da8a494a0c05e797ad27b992", size = 38097, upload-time = "2025-10-08T19:48:13.923Z" },
+    { url = "https://files.pythonhosted.org/packages/99/85/9ff785d787ccf9bbb3f3106f79884a130951436f58392000231b4c737c80/propcache-0.4.1-cp314-cp314t-macosx_10_13_universal2.whl", hash = "sha256:824e908bce90fb2743bd6b59db36eb4f45cd350a39637c9f73b1c1ea66f5b75f", size = 81455, upload-time = "2025-10-08T19:48:15.16Z" },
+    { url = "https://files.pythonhosted.org/packages/90/85/2431c10c8e7ddb1445c1f7c4b54d886e8ad20e3c6307e7218f05922cad67/propcache-0.4.1-cp314-cp314t-macosx_10_13_x86_64.whl", hash = "sha256:c2b5e7db5328427c57c8e8831abda175421b709672f6cfc3d630c3b7e2146393", size = 46372, upload-time = "2025-10-08T19:48:16.424Z" },
+    { url = "https://files.pythonhosted.org/packages/01/20/b0972d902472da9bcb683fa595099911f4d2e86e5683bcc45de60dd05dc3/propcache-0.4.1-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:6f6ff873ed40292cd4969ef5310179afd5db59fdf055897e282485043fc80ad0", size = 48411, upload-time = "2025-10-08T19:48:17.577Z" },
+    { url = "https://files.pythonhosted.org/packages/e2/e3/7dc89f4f21e8f99bad3d5ddb3a3389afcf9da4ac69e3deb2dcdc96e74169/propcache-0.4.1-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:49a2dc67c154db2c1463013594c458881a069fcf98940e61a0569016a583020a", size = 275712, upload-time = "2025-10-08T19:48:18.901Z" },
+    { url = "https://files.pythonhosted.org/packages/20/67/89800c8352489b21a8047c773067644e3897f02ecbbd610f4d46b7f08612/propcache-0.4.1-cp314-cp314t-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:005f08e6a0529984491e37d8dbc3dd86f84bd78a8ceb5fa9a021f4c48d4984be", size = 273557, upload-time = "2025-10-08T19:48:20.762Z" },
+    { url = "https://files.pythonhosted.org/packages/e2/a1/b52b055c766a54ce6d9c16d9aca0cad8059acd9637cdf8aa0222f4a026ef/propcache-0.4.1-cp314-cp314t-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:5c3310452e0d31390da9035c348633b43d7e7feb2e37be252be6da45abd1abcc", size = 280015, upload-time = "2025-10-08T19:48:22.592Z" },
+    { url = "https://files.pythonhosted.org/packages/48/c8/33cee30bd890672c63743049f3c9e4be087e6780906bfc3ec58528be59c1/propcache-0.4.1-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:4c3c70630930447f9ef1caac7728c8ad1c56bc5015338b20fed0d08ea2480b3a", size = 262880, upload-time = "2025-10-08T19:48:23.947Z" },
+    { url = "https://files.pythonhosted.org/packages/0c/b1/8f08a143b204b418285c88b83d00edbd61afbc2c6415ffafc8905da7038b/propcache-0.4.1-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:8e57061305815dfc910a3634dcf584f08168a8836e6999983569f51a8544cd89", size = 260938, upload-time = "2025-10-08T19:48:25.656Z" },
+    { url = "https://files.pythonhosted.org/packages/cf/12/96e4664c82ca2f31e1c8dff86afb867348979eb78d3cb8546a680287a1e9/propcache-0.4.1-cp314-cp314t-musllinux_1_2_armv7l.whl", hash = "sha256:521a463429ef54143092c11a77e04056dd00636f72e8c45b70aaa3140d639726", size = 247641, upload-time = "2025-10-08T19:48:27.207Z" },
+    { url = "https://files.pythonhosted.org/packages/18/ed/e7a9cfca28133386ba52278136d42209d3125db08d0a6395f0cba0c0285c/propcache-0.4.1-cp314-cp314t-musllinux_1_2_ppc64le.whl", hash = "sha256:120c964da3fdc75e3731aa392527136d4ad35868cc556fd09bb6d09172d9a367", size = 262510, upload-time = "2025-10-08T19:48:28.65Z" },
+    { url = "https://files.pythonhosted.org/packages/f5/76/16d8bf65e8845dd62b4e2b57444ab81f07f40caa5652b8969b87ddcf2ef6/propcache-0.4.1-cp314-cp314t-musllinux_1_2_s390x.whl", hash = "sha256:d8f353eb14ee3441ee844ade4277d560cdd68288838673273b978e3d6d2c8f36", size = 263161, upload-time = "2025-10-08T19:48:30.133Z" },
+    { url = "https://files.pythonhosted.org/packages/e7/70/c99e9edb5d91d5ad8a49fa3c1e8285ba64f1476782fed10ab251ff413ba1/propcache-0.4.1-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:ab2943be7c652f09638800905ee1bab2c544e537edb57d527997a24c13dc1455", size = 257393, upload-time = "2025-10-08T19:48:31.567Z" },
+    { url = "https://files.pythonhosted.org/packages/08/02/87b25304249a35c0915d236575bc3574a323f60b47939a2262b77632a3ee/propcache-0.4.1-cp314-cp314t-win32.whl", hash = "sha256:05674a162469f31358c30bcaa8883cb7829fa3110bf9c0991fe27d7896c42d85", size = 42546, upload-time = "2025-10-08T19:48:32.872Z" },
+    { url = "https://files.pythonhosted.org/packages/cb/ef/3c6ecf8b317aa982f309835e8f96987466123c6e596646d4e6a1dfcd080f/propcache-0.4.1-cp314-cp314t-win_amd64.whl", hash = "sha256:990f6b3e2a27d683cb7602ed6c86f15ee6b43b1194736f9baaeb93d0016633b1", size = 46259, upload-time = "2025-10-08T19:48:34.226Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/2d/346e946d4951f37eca1e4f55be0f0174c52cd70720f84029b02f296f4a38/propcache-0.4.1-cp314-cp314t-win_arm64.whl", hash = "sha256:ecef2343af4cc68e05131e45024ba34f6095821988a9d0a02aa7c73fcc448aa9", size = 40428, upload-time = "2025-10-08T19:48:35.441Z" },
+    { url = "https://files.pythonhosted.org/packages/5b/5a/bc7b4a4ef808fa59a816c17b20c4bef6884daebbdf627ff2a161da67da19/propcache-0.4.1-py3-none-any.whl", hash = "sha256:af2a6052aeb6cf17d3e46ee169099044fd8224cbaf75c76a2ef596e8163e2237", size = 13305, upload-time = "2025-10-08T19:49:00.792Z" },
+]
+
+[[package]]
+name = "protobuf"
+version = "6.33.5"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/ba/25/7c72c307aafc96fa87062aa6291d9f7c94836e43214d43722e86037aac02/protobuf-6.33.5.tar.gz", hash = "sha256:6ddcac2a081f8b7b9642c09406bc6a4290128fce5f471cddd165960bb9119e5c", size = 444465, upload-time = "2026-01-29T21:51:33.494Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/b1/79/af92d0a8369732b027e6d6084251dd8e782c685c72da161bd4a2e00fbabb/protobuf-6.33.5-cp310-abi3-win32.whl", hash = "sha256:d71b040839446bac0f4d162e758bea99c8251161dae9d0983a3b88dee345153b", size = 425769, upload-time = "2026-01-29T21:51:21.751Z" },
+    { url = "https://files.pythonhosted.org/packages/55/75/bb9bc917d10e9ee13dee8607eb9ab963b7cf8be607c46e7862c748aa2af7/protobuf-6.33.5-cp310-abi3-win_amd64.whl", hash = "sha256:3093804752167bcab3998bec9f1048baae6e29505adaf1afd14a37bddede533c", size = 437118, upload-time = "2026-01-29T21:51:24.022Z" },
+    { url = "https://files.pythonhosted.org/packages/a2/6b/e48dfc1191bc5b52950246275bf4089773e91cb5ba3592621723cdddca62/protobuf-6.33.5-cp39-abi3-macosx_10_9_universal2.whl", hash = "sha256:a5cb85982d95d906df1e2210e58f8e4f1e3cdc088e52c921a041f9c9a0386de5", size = 427766, upload-time = "2026-01-29T21:51:25.413Z" },
+    { url = "https://files.pythonhosted.org/packages/4e/b1/c79468184310de09d75095ed1314b839eb2f72df71097db9d1404a1b2717/protobuf-6.33.5-cp39-abi3-manylinux2014_aarch64.whl", hash = "sha256:9b71e0281f36f179d00cbcb119cb19dec4d14a81393e5ea220f64b286173e190", size = 324638, upload-time = "2026-01-29T21:51:26.423Z" },
+    { url = "https://files.pythonhosted.org/packages/c5/f5/65d838092fd01c44d16037953fd4c2cc851e783de9b8f02b27ec4ffd906f/protobuf-6.33.5-cp39-abi3-manylinux2014_s390x.whl", hash = "sha256:8afa18e1d6d20af15b417e728e9f60f3aa108ee76f23c3b2c07a2c3b546d3afd", size = 339411, upload-time = "2026-01-29T21:51:27.446Z" },
+    { url = "https://files.pythonhosted.org/packages/9b/53/a9443aa3ca9ba8724fdfa02dd1887c1bcd8e89556b715cfbacca6b63dbec/protobuf-6.33.5-cp39-abi3-manylinux2014_x86_64.whl", hash = "sha256:cbf16ba3350fb7b889fca858fb215967792dc125b35c7976ca4818bee3521cf0", size = 323465, upload-time = "2026-01-29T21:51:28.925Z" },
+    { url = "https://files.pythonhosted.org/packages/57/bf/2086963c69bdac3d7cff1cc7ff79b8ce5ea0bec6797a017e1be338a46248/protobuf-6.33.5-py3-none-any.whl", hash = "sha256:69915a973dd0f60f31a08b8318b73eab2bd6a392c79184b3612226b0a3f8ec02", size = 170687, upload-time = "2026-01-29T21:51:32.557Z" },
+]
+
+[[package]]
+name = "psutil"
+version = "7.2.2"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/aa/c6/d1ddf4abb55e93cebc4f2ed8b5d6dbad109ecb8d63748dd2b20ab5e57ebe/psutil-7.2.2.tar.gz", hash = "sha256:0746f5f8d406af344fd547f1c8daa5f5c33dbc293bb8d6a16d80b4bb88f59372", size = 493740, upload-time = "2026-01-28T18:14:54.428Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/51/08/510cbdb69c25a96f4ae523f733cdc963ae654904e8db864c07585ef99875/psutil-7.2.2-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:2edccc433cbfa046b980b0df0171cd25bcaeb3a68fe9022db0979e7aa74a826b", size = 130595, upload-time = "2026-01-28T18:14:57.293Z" },
+    { url = "https://files.pythonhosted.org/packages/d6/f5/97baea3fe7a5a9af7436301f85490905379b1c6f2dd51fe3ecf24b4c5fbf/psutil-7.2.2-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:e78c8603dcd9a04c7364f1a3e670cea95d51ee865e4efb3556a3a63adef958ea", size = 131082, upload-time = "2026-01-28T18:14:59.732Z" },
+    { url = "https://files.pythonhosted.org/packages/37/d6/246513fbf9fa174af531f28412297dd05241d97a75911ac8febefa1a53c6/psutil-7.2.2-cp313-cp313t-manylinux2010_x86_64.manylinux_2_12_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:1a571f2330c966c62aeda00dd24620425d4b0cc86881c89861fbc04549e5dc63", size = 181476, upload-time = "2026-01-28T18:15:01.884Z" },
+    { url = "https://files.pythonhosted.org/packages/b8/b5/9182c9af3836cca61696dabe4fd1304e17bc56cb62f17439e1154f225dd3/psutil-7.2.2-cp313-cp313t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:917e891983ca3c1887b4ef36447b1e0873e70c933afc831c6b6da078ba474312", size = 184062, upload-time = "2026-01-28T18:15:04.436Z" },
+    { url = "https://files.pythonhosted.org/packages/16/ba/0756dca669f5a9300d0cbcbfae9a4c30e446dfc7440ffe43ded5724bfd93/psutil-7.2.2-cp313-cp313t-win_amd64.whl", hash = "sha256:ab486563df44c17f5173621c7b198955bd6b613fb87c71c161f827d3fb149a9b", size = 139893, upload-time = "2026-01-28T18:15:06.378Z" },
+    { url = "https://files.pythonhosted.org/packages/1c/61/8fa0e26f33623b49949346de05ec1ddaad02ed8ba64af45f40a147dbfa97/psutil-7.2.2-cp313-cp313t-win_arm64.whl", hash = "sha256:ae0aefdd8796a7737eccea863f80f81e468a1e4cf14d926bd9b6f5f2d5f90ca9", size = 135589, upload-time = "2026-01-28T18:15:08.03Z" },
+    { url = "https://files.pythonhosted.org/packages/81/69/ef179ab5ca24f32acc1dac0c247fd6a13b501fd5534dbae0e05a1c48b66d/psutil-7.2.2-cp314-cp314t-macosx_10_15_x86_64.whl", hash = "sha256:eed63d3b4d62449571547b60578c5b2c4bcccc5387148db46e0c2313dad0ee00", size = 130664, upload-time = "2026-01-28T18:15:09.469Z" },
+    { url = "https://files.pythonhosted.org/packages/7b/64/665248b557a236d3fa9efc378d60d95ef56dd0a490c2cd37dafc7660d4a9/psutil-7.2.2-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:7b6d09433a10592ce39b13d7be5a54fbac1d1228ed29abc880fb23df7cb694c9", size = 131087, upload-time = "2026-01-28T18:15:11.724Z" },
+    { url = "https://files.pythonhosted.org/packages/d5/2e/e6782744700d6759ebce3043dcfa661fb61e2fb752b91cdeae9af12c2178/psutil-7.2.2-cp314-cp314t-manylinux2010_x86_64.manylinux_2_12_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:1fa4ecf83bcdf6e6c8f4449aff98eefb5d0604bf88cb883d7da3d8d2d909546a", size = 182383, upload-time = "2026-01-28T18:15:13.445Z" },
+    { url = "https://files.pythonhosted.org/packages/57/49/0a41cefd10cb7505cdc04dab3eacf24c0c2cb158a998b8c7b1d27ee2c1f5/psutil-7.2.2-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:e452c464a02e7dc7822a05d25db4cde564444a67e58539a00f929c51eddda0cf", size = 185210, upload-time = "2026-01-28T18:15:16.002Z" },
+    { url = "https://files.pythonhosted.org/packages/dd/2c/ff9bfb544f283ba5f83ba725a3c5fec6d6b10b8f27ac1dc641c473dc390d/psutil-7.2.2-cp314-cp314t-win_amd64.whl", hash = "sha256:c7663d4e37f13e884d13994247449e9f8f574bc4655d509c3b95e9ec9e2b9dc1", size = 141228, upload-time = "2026-01-28T18:15:18.385Z" },
+    { url = "https://files.pythonhosted.org/packages/f2/fc/f8d9c31db14fcec13748d373e668bc3bed94d9077dbc17fb0eebc073233c/psutil-7.2.2-cp314-cp314t-win_arm64.whl", hash = "sha256:11fe5a4f613759764e79c65cf11ebdf26e33d6dd34336f8a337aa2996d71c841", size = 136284, upload-time = "2026-01-28T18:15:19.912Z" },
+    { url = "https://files.pythonhosted.org/packages/e7/36/5ee6e05c9bd427237b11b3937ad82bb8ad2752d72c6969314590dd0c2f6e/psutil-7.2.2-cp36-abi3-macosx_10_9_x86_64.whl", hash = "sha256:ed0cace939114f62738d808fdcecd4c869222507e266e574799e9c0faa17d486", size = 129090, upload-time = "2026-01-28T18:15:22.168Z" },
+    { url = "https://files.pythonhosted.org/packages/80/c4/f5af4c1ca8c1eeb2e92ccca14ce8effdeec651d5ab6053c589b074eda6e1/psutil-7.2.2-cp36-abi3-macosx_11_0_arm64.whl", hash = "sha256:1a7b04c10f32cc88ab39cbf606e117fd74721c831c98a27dc04578deb0c16979", size = 129859, upload-time = "2026-01-28T18:15:23.795Z" },
+    { url = "https://files.pythonhosted.org/packages/b5/70/5d8df3b09e25bce090399cf48e452d25c935ab72dad19406c77f4e828045/psutil-7.2.2-cp36-abi3-manylinux2010_x86_64.manylinux_2_12_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:076a2d2f923fd4821644f5ba89f059523da90dc9014e85f8e45a5774ca5bc6f9", size = 155560, upload-time = "2026-01-28T18:15:25.976Z" },
+    { url = "https://files.pythonhosted.org/packages/63/65/37648c0c158dc222aba51c089eb3bdfa238e621674dc42d48706e639204f/psutil-7.2.2-cp36-abi3-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:b0726cecd84f9474419d67252add4ac0cd9811b04d61123054b9fb6f57df6e9e", size = 156997, upload-time = "2026-01-28T18:15:27.794Z" },
+    { url = "https://files.pythonhosted.org/packages/8e/13/125093eadae863ce03c6ffdbae9929430d116a246ef69866dad94da3bfbc/psutil-7.2.2-cp36-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:fd04ef36b4a6d599bbdb225dd1d3f51e00105f6d48a28f006da7f9822f2606d8", size = 148972, upload-time = "2026-01-28T18:15:29.342Z" },
+    { url = "https://files.pythonhosted.org/packages/04/78/0acd37ca84ce3ddffaa92ef0f571e073faa6d8ff1f0559ab1272188ea2be/psutil-7.2.2-cp36-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:b58fabe35e80b264a4e3bb23e6b96f9e45a3df7fb7eed419ac0e5947c61e47cc", size = 148266, upload-time = "2026-01-28T18:15:31.597Z" },
+    { url = "https://files.pythonhosted.org/packages/b4/90/e2159492b5426be0c1fef7acba807a03511f97c5f86b3caeda6ad92351a7/psutil-7.2.2-cp37-abi3-win_amd64.whl", hash = "sha256:eb7e81434c8d223ec4a219b5fc1c47d0417b12be7ea866e24fb5ad6e84b3d988", size = 137737, upload-time = "2026-01-28T18:15:33.849Z" },
+    { url = "https://files.pythonhosted.org/packages/8c/c7/7bb2e321574b10df20cbde462a94e2b71d05f9bbda251ef27d104668306a/psutil-7.2.2-cp37-abi3-win_arm64.whl", hash = "sha256:8c233660f575a5a89e6d4cb65d9f938126312bca76d8fe087b947b3a1aaac9ee", size = 134617, upload-time = "2026-01-28T18:15:36.514Z" },
+]
+
+[[package]]
+name = "py-cpuinfo"
+version = "9.0.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/37/a8/d832f7293ebb21690860d2e01d8115e5ff6f2ae8bbdc953f0eb0fa4bd2c7/py-cpuinfo-9.0.0.tar.gz", hash = "sha256:3cdbbf3fac90dc6f118bfd64384f309edeadd902d7c8fb17f02ffa1fc3f49690", size = 104716, upload-time = "2022-10-25T20:38:06.303Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/e0/a9/023730ba63db1e494a271cb018dcd361bd2c917ba7004c3e49d5daf795a2/py_cpuinfo-9.0.0-py3-none-any.whl", hash = "sha256:859625bc251f64e21f077d099d4162689c762b5d6a4c3c97553d56241c9674d5", size = 22335, upload-time = "2022-10-25T20:38:27.636Z" },
+]
+
+[[package]]
+name = "pyclipper"
+version = "1.4.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/f6/21/3c06205bb407e1f79b73b7b4dfb3950bd9537c4f625a68ab5cc41177f5bc/pyclipper-1.4.0.tar.gz", hash = "sha256:9882bd889f27da78add4dd6f881d25697efc740bf840274e749988d25496c8e1", size = 54489, upload-time = "2025-12-01T13:15:35.015Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/8a/9f/a10173d32ecc2ce19a04d018163f3ca22a04c0c6ad03b464dcd32f9152a8/pyclipper-1.4.0-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:bafad70d2679c187120e8c44e1f9a8b06150bad8c0aecf612ad7dfbfa9510f73", size = 264510, upload-time = "2025-12-01T13:14:46.551Z" },
+    { url = "https://files.pythonhosted.org/packages/e0/c2/5490ddc4a1f7ceeaa0258f4266397e720c02db515b2ca5bc69b85676f697/pyclipper-1.4.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:0b74a9dd44b22a7fd35d65fb1ceeba57f3817f34a97a28c3255556362e491447", size = 139498, upload-time = "2025-12-01T13:14:48.31Z" },
+    { url = "https://files.pythonhosted.org/packages/3b/0a/bea9102d1d75634b1a5702b0e92982451a1eafca73c4845d3dbe27eba13d/pyclipper-1.4.0-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:0a4d2736fb3c42e8eb1d38bf27a720d1015526c11e476bded55138a977c17d9d", size = 970974, upload-time = "2025-12-01T13:14:49.799Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/1b/097f8776d5b3a10eb7b443b632221f4ed825d892e79e05682f4b10a1a59c/pyclipper-1.4.0-cp310-cp310-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:b3b3630051b53ad2564cb079e088b112dd576e3d91038338ad1cc7915e0f14dc", size = 943315, upload-time = "2025-12-01T13:14:51.266Z" },
+    { url = "https://files.pythonhosted.org/packages/fd/4d/17d6a3f1abf0f368d58f2309e80ee3761afb1fd1342f7780ab32ba4f0b1d/pyclipper-1.4.0-cp310-cp310-win32.whl", hash = "sha256:8d42b07a2f6cfe2d9b87daf345443583f00a14e856927782fde52f3a255e305a", size = 95286, upload-time = "2025-12-01T13:14:52.922Z" },
+    { url = "https://files.pythonhosted.org/packages/53/ca/b30138427ed122ec9b47980b943164974a2ec606fa3f71597033b9a9f9a6/pyclipper-1.4.0-cp310-cp310-win_amd64.whl", hash = "sha256:6a97b961f182b92d899ca88c1bb3632faea2e00ce18d07c5f789666ebb021ca4", size = 104227, upload-time = "2025-12-01T13:14:54.013Z" },
+    { url = "https://files.pythonhosted.org/packages/de/e3/64cf7794319b088c288706087141e53ac259c7959728303276d18adc665d/pyclipper-1.4.0-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:adcb7ca33c5bdc33cd775e8b3eadad54873c802a6d909067a57348bcb96e7a2d", size = 264281, upload-time = "2025-12-01T13:14:55.47Z" },
+    { url = "https://files.pythonhosted.org/packages/34/cd/44ec0da0306fa4231e76f1c2cb1fa394d7bde8db490a2b24d55b39865f69/pyclipper-1.4.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:fd24849d2b94ec749ceac7c34c9f01010d23b6e9d9216cf2238b8481160e703d", size = 139426, upload-time = "2025-12-01T13:14:56.683Z" },
+    { url = "https://files.pythonhosted.org/packages/ad/88/d8f6c6763ea622fe35e19c75d8b39ed6c55191ddc82d65e06bc46b26cb8e/pyclipper-1.4.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:1b6c8d75ba20c6433c9ea8f1a0feb7e4d3ac06a09ad1fd6d571afc1ddf89b869", size = 989649, upload-time = "2025-12-01T13:14:58.28Z" },
+    { url = "https://files.pythonhosted.org/packages/ff/e9/ea7d68c8c4af3842d6515bedcf06418610ad75f111e64c92c1d4785a1513/pyclipper-1.4.0-cp311-cp311-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:58e29d7443d7cc0e83ee9daf43927730386629786d00c63b04fe3b53ac01462c", size = 962842, upload-time = "2025-12-01T13:15:00.044Z" },
+    { url = "https://files.pythonhosted.org/packages/4e/b7/0b4a272d8726e51ab05e2b933d8cc47f29757fb8212e38b619e170e6015c/pyclipper-1.4.0-cp311-cp311-win32.whl", hash = "sha256:a8d2b5fb75ebe57e21ce61e79a9131edec2622ff23cc665e4d1d1f201bc1a801", size = 95098, upload-time = "2025-12-01T13:15:01.359Z" },
+    { url = "https://files.pythonhosted.org/packages/3a/76/4901de2919198bb2bd3d989f86d4a1dff363962425bb2d63e24e6c990042/pyclipper-1.4.0-cp311-cp311-win_amd64.whl", hash = "sha256:e9b973467d9c5fa9bc30bb6ac95f9f4d7c3d9fc25f6cf2d1cc972088e5955c01", size = 104362, upload-time = "2025-12-01T13:15:02.439Z" },
+    { url = "https://files.pythonhosted.org/packages/90/1b/7a07b68e0842324d46c03e512d8eefa9cb92ba2a792b3b4ebf939dafcac3/pyclipper-1.4.0-cp312-cp312-macosx_10_13_universal2.whl", hash = "sha256:222ac96c8b8281b53d695b9c4fedc674f56d6d4320ad23f1bdbd168f4e316140", size = 265676, upload-time = "2025-12-01T13:15:04.15Z" },
+    { url = "https://files.pythonhosted.org/packages/6b/dd/8bd622521c05d04963420ae6664093f154343ed044c53ea260a310c8bb4d/pyclipper-1.4.0-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:f3672dbafbb458f1b96e1ee3e610d174acb5ace5bd2ed5d1252603bb797f2fc6", size = 140458, upload-time = "2025-12-01T13:15:05.76Z" },
+    { url = "https://files.pythonhosted.org/packages/7a/06/6e3e241882bf7d6ab23d9c69ba4e85f1ec47397cbbeee948a16cf75e21ed/pyclipper-1.4.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:d1f807e2b4760a8e5c6d6b4e8c1d71ef52b7fe1946ff088f4fa41e16a881a5ca", size = 978235, upload-time = "2025-12-01T13:15:06.993Z" },
+    { url = "https://files.pythonhosted.org/packages/cf/f4/3418c1cd5eea640a9fa2501d4bc0b3655fa8d40145d1a4f484b987990a75/pyclipper-1.4.0-cp312-cp312-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:ce1f83c9a4e10ea3de1959f0ae79e9a5bd41346dff648fee6228ba9eaf8b3872", size = 961388, upload-time = "2025-12-01T13:15:08.467Z" },
+    { url = "https://files.pythonhosted.org/packages/ac/94/c85401d24be634af529c962dd5d781f3cb62a67cd769534df2cb3feee97a/pyclipper-1.4.0-cp312-cp312-win32.whl", hash = "sha256:3ef44b64666ebf1cb521a08a60c3e639d21b8c50bfbe846ba7c52a0415e936f4", size = 95169, upload-time = "2025-12-01T13:15:10.098Z" },
+    { url = "https://files.pythonhosted.org/packages/97/77/dfea08e3b230b82ee22543c30c35d33d42f846a77f96caf7c504dd54fab1/pyclipper-1.4.0-cp312-cp312-win_amd64.whl", hash = "sha256:d1e5498d883b706a4ce636247f0d830c6eb34a25b843a1b78e2c969754ca9037", size = 104619, upload-time = "2025-12-01T13:15:11.592Z" },
+    { url = "https://files.pythonhosted.org/packages/67/d0/cbce7d47de1e6458f66a4d999b091640134deb8f2c7351eab993b70d2e10/pyclipper-1.4.0-cp313-cp313-macosx_10_13_universal2.whl", hash = "sha256:d49df13cbb2627ccb13a1046f3ea6ebf7177b5504ec61bdef87d6a704046fd6e", size = 264342, upload-time = "2025-12-01T13:15:12.697Z" },
+    { url = "https://files.pythonhosted.org/packages/ce/cc/742b9d69d96c58ac156947e1b56d0f81cbacbccf869e2ac7229f2f86dc4e/pyclipper-1.4.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:37bfec361e174110cdddffd5ecd070a8064015c99383d95eb692c253951eee8a", size = 139839, upload-time = "2025-12-01T13:15:13.911Z" },
+    { url = "https://files.pythonhosted.org/packages/db/48/dd301d62c1529efdd721b47b9e5fb52120fcdac5f4d3405cfc0d2f391414/pyclipper-1.4.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:14c8bdb5a72004b721c4e6f448d2c2262d74a7f0c9e3076aeff41e564a92389f", size = 972142, upload-time = "2025-12-01T13:15:15.477Z" },
+    { url = "https://files.pythonhosted.org/packages/07/bf/d493fd1b33bb090fa64e28c1009374d5d72fa705f9331cd56517c35e381e/pyclipper-1.4.0-cp313-cp313-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:f2a50c22c3a78cb4e48347ecf06930f61ce98cf9252f2e292aa025471e9d75b1", size = 952789, upload-time = "2025-12-01T13:15:17.042Z" },
+    { url = "https://files.pythonhosted.org/packages/cf/88/b95ea8ea21ddca34aa14b123226a81526dd2faaa993f9aabd3ed21231604/pyclipper-1.4.0-cp313-cp313-win32.whl", hash = "sha256:c9a3faa416ff536cee93417a72bfb690d9dea136dc39a39dbbe1e5dadf108c9c", size = 94817, upload-time = "2025-12-01T13:15:18.724Z" },
+    { url = "https://files.pythonhosted.org/packages/ba/42/0a1920d276a0e1ca21dc0d13ee9e3ba10a9a8aa3abac76cd5e5a9f503306/pyclipper-1.4.0-cp313-cp313-win_amd64.whl", hash = "sha256:d4b2d7c41086f1927d14947c563dfc7beed2f6c0d9af13c42fe3dcdc20d35832", size = 104007, upload-time = "2025-12-01T13:15:19.763Z" },
+    { url = "https://files.pythonhosted.org/packages/1a/20/04d58c70f3ccd404f179f8dd81d16722a05a3bf1ab61445ee64e8218c1f8/pyclipper-1.4.0-cp314-cp314-macosx_10_15_universal2.whl", hash = "sha256:7c87480fc91a5af4c1ba310bdb7de2f089a3eeef5fe351a3cedc37da1fcced1c", size = 265167, upload-time = "2025-12-01T13:15:20.844Z" },
+    { url = "https://files.pythonhosted.org/packages/bd/2e/a570c1abe69b7260ca0caab4236ce6ea3661193ebf8d1bd7f78ccce537a5/pyclipper-1.4.0-cp314-cp314-macosx_10_15_x86_64.whl", hash = "sha256:81d8bb2d1fb9d66dc7ea4373b176bb4b02443a7e328b3b603a73faec088b952e", size = 139966, upload-time = "2025-12-01T13:15:22.036Z" },
+    { url = "https://files.pythonhosted.org/packages/e8/3b/e0859e54adabdde8a24a29d3f525ebb31c71ddf2e8d93edce83a3c212ffc/pyclipper-1.4.0-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:773c0e06b683214dcfc6711be230c83b03cddebe8a57eae053d4603dd63582f9", size = 968216, upload-time = "2025-12-01T13:15:23.18Z" },
+    { url = "https://files.pythonhosted.org/packages/f6/6b/e3c4febf0a35ae643ee579b09988dd931602b5bf311020535fd9e5b7e715/pyclipper-1.4.0-cp314-cp314-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:9bc45f2463d997848450dbed91c950ca37c6cf27f84a49a5cad4affc0b469e39", size = 954198, upload-time = "2025-12-01T13:15:24.522Z" },
+    { url = "https://files.pythonhosted.org/packages/fc/74/728efcee02e12acb486ce9d56fa037120c9bf5b77c54bbdbaa441c14a9d9/pyclipper-1.4.0-cp314-cp314-win32.whl", hash = "sha256:0b8c2105b3b3c44dbe1a266f64309407fe30bf372cf39a94dc8aaa97df00da5b", size = 96951, upload-time = "2025-12-01T13:15:25.79Z" },
+    { url = "https://files.pythonhosted.org/packages/e3/d7/7f4354e69f10a917e5c7d5d72a499ef2e10945312f5e72c414a0a08d2ae4/pyclipper-1.4.0-cp314-cp314-win_amd64.whl", hash = "sha256:6c317e182590c88ec0194149995e3d71a979cfef3b246383f4e035f9d4a11826", size = 106782, upload-time = "2025-12-01T13:15:26.945Z" },
+    { url = "https://files.pythonhosted.org/packages/63/60/fc32c7a3d7f61a970511ec2857ecd09693d8ac80d560ee7b8e67a6d268c9/pyclipper-1.4.0-cp314-cp314t-macosx_10_15_universal2.whl", hash = "sha256:f160a2c6ba036f7eaf09f1f10f4fbfa734234af9112fb5187877efed78df9303", size = 269880, upload-time = "2025-12-01T13:15:28.117Z" },
+    { url = "https://files.pythonhosted.org/packages/49/df/c4a72d3f62f0ba03ec440c4fff56cd2d674a4334d23c5064cbf41c9583f6/pyclipper-1.4.0-cp314-cp314t-macosx_10_15_x86_64.whl", hash = "sha256:a9f11ad133257c52c40d50de7a0ca3370a0cdd8e3d11eec0604ad3c34ba549e9", size = 141706, upload-time = "2025-12-01T13:15:30.134Z" },
+    { url = "https://files.pythonhosted.org/packages/c5/0b/cf55df03e2175e1e2da9db585241401e0bc98f76bee3791bed39d0313449/pyclipper-1.4.0-cp314-cp314t-win32.whl", hash = "sha256:bbc827b77442c99deaeee26e0e7f172355ddb097a5e126aea206d447d3b26286", size = 105308, upload-time = "2025-12-01T13:15:31.225Z" },
+    { url = "https://files.pythonhosted.org/packages/8f/dc/53df8b6931d47080b4fe4ee8450d42e660ee1c5c1556c7ab73359182b769/pyclipper-1.4.0-cp314-cp314t-win_amd64.whl", hash = "sha256:29dae3e0296dff8502eeb7639fcfee794b0eec8590ba3563aee28db269da6b04", size = 117608, upload-time = "2025-12-01T13:15:32.69Z" },
+    { url = "https://files.pythonhosted.org/packages/18/59/81050abdc9e5b90ffc2c765738c5e40e9abd8e44864aaa737b600f16c562/pyclipper-1.4.0-pp311-pypy311_pp73-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:98b2a40f98e1fc1b29e8a6094072e7e0c7dfe901e573bf6cfc6eb7ce84a7ae87", size = 126495, upload-time = "2025-12-01T13:15:33.743Z" },
+]
+
+[[package]]
+name = "pycparser"
+version = "3.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/1b/7d/92392ff7815c21062bea51aa7b87d45576f649f16458d78b7cf94b9ab2e6/pycparser-3.0.tar.gz", hash = "sha256:600f49d217304a5902ac3c37e1281c9fe94e4d0489de643a9504c5cdfdfc6b29", size = 103492, upload-time = "2026-01-21T14:26:51.89Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/0c/c3/44f3fbbfa403ea2a7c779186dc20772604442dde72947e7d01069cbe98e3/pycparser-3.0-py3-none-any.whl", hash = "sha256:b727414169a36b7d524c1c3e31839a521725078d7b2ff038656844266160a992", size = 48172, upload-time = "2026-01-21T14:26:50.693Z" },
+]
+
+[[package]]
+name = "pycryptodome"
+version = "3.23.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/8e/a6/8452177684d5e906854776276ddd34eca30d1b1e15aa1ee9cefc289a33f5/pycryptodome-3.23.0.tar.gz", hash = "sha256:447700a657182d60338bab09fdb27518f8856aecd80ae4c6bdddb67ff5da44ef", size = 4921276, upload-time = "2025-05-17T17:21:45.242Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/04/5d/bdb09489b63cd34a976cc9e2a8d938114f7a53a74d3dd4f125ffa49dce82/pycryptodome-3.23.0-cp313-cp313t-macosx_10_13_universal2.whl", hash = "sha256:0011f7f00cdb74879142011f95133274741778abba114ceca229adbf8e62c3e4", size = 2495152, upload-time = "2025-05-17T17:20:20.833Z" },
+    { url = "https://files.pythonhosted.org/packages/a7/ce/7840250ed4cc0039c433cd41715536f926d6e86ce84e904068eb3244b6a6/pycryptodome-3.23.0-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:90460fc9e088ce095f9ee8356722d4f10f86e5be06e2354230a9880b9c549aae", size = 1639348, upload-time = "2025-05-17T17:20:23.171Z" },
+    { url = "https://files.pythonhosted.org/packages/ee/f0/991da24c55c1f688d6a3b5a11940567353f74590734ee4a64294834ae472/pycryptodome-3.23.0-cp313-cp313t-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:4764e64b269fc83b00f682c47443c2e6e85b18273712b98aa43bcb77f8570477", size = 2184033, upload-time = "2025-05-17T17:20:25.424Z" },
+    { url = "https://files.pythonhosted.org/packages/54/16/0e11882deddf00f68b68dd4e8e442ddc30641f31afeb2bc25588124ac8de/pycryptodome-3.23.0-cp313-cp313t-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:eb8f24adb74984aa0e5d07a2368ad95276cf38051fe2dc6605cbcf482e04f2a7", size = 2270142, upload-time = "2025-05-17T17:20:27.808Z" },
+    { url = "https://files.pythonhosted.org/packages/d5/fc/4347fea23a3f95ffb931f383ff28b3f7b1fe868739182cb76718c0da86a1/pycryptodome-3.23.0-cp313-cp313t-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:d97618c9c6684a97ef7637ba43bdf6663a2e2e77efe0f863cce97a76af396446", size = 2309384, upload-time = "2025-05-17T17:20:30.765Z" },
+    { url = "https://files.pythonhosted.org/packages/6e/d9/c5261780b69ce66d8cfab25d2797bd6e82ba0241804694cd48be41add5eb/pycryptodome-3.23.0-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:9a53a4fe5cb075075d515797d6ce2f56772ea7e6a1e5e4b96cf78a14bac3d265", size = 2183237, upload-time = "2025-05-17T17:20:33.736Z" },
+    { url = "https://files.pythonhosted.org/packages/5a/6f/3af2ffedd5cfa08c631f89452c6648c4d779e7772dfc388c77c920ca6bbf/pycryptodome-3.23.0-cp313-cp313t-musllinux_1_2_i686.whl", hash = "sha256:763d1d74f56f031788e5d307029caef067febf890cd1f8bf61183ae142f1a77b", size = 2343898, upload-time = "2025-05-17T17:20:36.086Z" },
+    { url = "https://files.pythonhosted.org/packages/9a/dc/9060d807039ee5de6e2f260f72f3d70ac213993a804f5e67e0a73a56dd2f/pycryptodome-3.23.0-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:954af0e2bd7cea83ce72243b14e4fb518b18f0c1649b576d114973e2073b273d", size = 2269197, upload-time = "2025-05-17T17:20:38.414Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/34/e6c8ca177cb29dcc4967fef73f5de445912f93bd0343c9c33c8e5bf8cde8/pycryptodome-3.23.0-cp313-cp313t-win32.whl", hash = "sha256:257bb3572c63ad8ba40b89f6fc9d63a2a628e9f9708d31ee26560925ebe0210a", size = 1768600, upload-time = "2025-05-17T17:20:40.688Z" },
+    { url = "https://files.pythonhosted.org/packages/e4/1d/89756b8d7ff623ad0160f4539da571d1f594d21ee6d68be130a6eccb39a4/pycryptodome-3.23.0-cp313-cp313t-win_amd64.whl", hash = "sha256:6501790c5b62a29fcb227bd6b62012181d886a767ce9ed03b303d1f22eb5c625", size = 1799740, upload-time = "2025-05-17T17:20:42.413Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/61/35a64f0feaea9fd07f0d91209e7be91726eb48c0f1bfc6720647194071e4/pycryptodome-3.23.0-cp313-cp313t-win_arm64.whl", hash = "sha256:9a77627a330ab23ca43b48b130e202582e91cc69619947840ea4d2d1be21eb39", size = 1703685, upload-time = "2025-05-17T17:20:44.388Z" },
+    { url = "https://files.pythonhosted.org/packages/db/6c/a1f71542c969912bb0e106f64f60a56cc1f0fabecf9396f45accbe63fa68/pycryptodome-3.23.0-cp37-abi3-macosx_10_9_universal2.whl", hash = "sha256:187058ab80b3281b1de11c2e6842a357a1f71b42cb1e15bce373f3d238135c27", size = 2495627, upload-time = "2025-05-17T17:20:47.139Z" },
+    { url = "https://files.pythonhosted.org/packages/6e/4e/a066527e079fc5002390c8acdd3aca431e6ea0a50ffd7201551175b47323/pycryptodome-3.23.0-cp37-abi3-macosx_10_9_x86_64.whl", hash = "sha256:cfb5cd445280c5b0a4e6187a7ce8de5a07b5f3f897f235caa11f1f435f182843", size = 1640362, upload-time = "2025-05-17T17:20:50.392Z" },
+    { url = "https://files.pythonhosted.org/packages/50/52/adaf4c8c100a8c49d2bd058e5b551f73dfd8cb89eb4911e25a0c469b6b4e/pycryptodome-3.23.0-cp37-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:67bd81fcbe34f43ad9422ee8fd4843c8e7198dd88dd3d40e6de42ee65fbe1490", size = 2182625, upload-time = "2025-05-17T17:20:52.866Z" },
+    { url = "https://files.pythonhosted.org/packages/5f/e9/a09476d436d0ff1402ac3867d933c61805ec2326c6ea557aeeac3825604e/pycryptodome-3.23.0-cp37-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:c8987bd3307a39bc03df5c8e0e3d8be0c4c3518b7f044b0f4c15d1aa78f52575", size = 2268954, upload-time = "2025-05-17T17:20:55.027Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/c5/ffe6474e0c551d54cab931918127c46d70cab8f114e0c2b5a3c071c2f484/pycryptodome-3.23.0-cp37-abi3-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:aa0698f65e5b570426fc31b8162ed4603b0c2841cbb9088e2b01641e3065915b", size = 2308534, upload-time = "2025-05-17T17:20:57.279Z" },
+    { url = "https://files.pythonhosted.org/packages/18/28/e199677fc15ecf43010f2463fde4c1a53015d1fe95fb03bca2890836603a/pycryptodome-3.23.0-cp37-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:53ecbafc2b55353edcebd64bf5da94a2a2cdf5090a6915bcca6eca6cc452585a", size = 2181853, upload-time = "2025-05-17T17:20:59.322Z" },
+    { url = "https://files.pythonhosted.org/packages/ce/ea/4fdb09f2165ce1365c9eaefef36625583371ee514db58dc9b65d3a255c4c/pycryptodome-3.23.0-cp37-abi3-musllinux_1_2_i686.whl", hash = "sha256:156df9667ad9f2ad26255926524e1c136d6664b741547deb0a86a9acf5ea631f", size = 2342465, upload-time = "2025-05-17T17:21:03.83Z" },
+    { url = "https://files.pythonhosted.org/packages/22/82/6edc3fc42fe9284aead511394bac167693fb2b0e0395b28b8bedaa07ef04/pycryptodome-3.23.0-cp37-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:dea827b4d55ee390dc89b2afe5927d4308a8b538ae91d9c6f7a5090f397af1aa", size = 2267414, upload-time = "2025-05-17T17:21:06.72Z" },
+    { url = "https://files.pythonhosted.org/packages/59/fe/aae679b64363eb78326c7fdc9d06ec3de18bac68be4b612fc1fe8902693c/pycryptodome-3.23.0-cp37-abi3-win32.whl", hash = "sha256:507dbead45474b62b2bbe318eb1c4c8ee641077532067fec9c1aa82c31f84886", size = 1768484, upload-time = "2025-05-17T17:21:08.535Z" },
+    { url = "https://files.pythonhosted.org/packages/54/2f/e97a1b8294db0daaa87012c24a7bb714147c7ade7656973fd6c736b484ff/pycryptodome-3.23.0-cp37-abi3-win_amd64.whl", hash = "sha256:c75b52aacc6c0c260f204cbdd834f76edc9fb0d8e0da9fbf8352ef58202564e2", size = 1799636, upload-time = "2025-05-17T17:21:10.393Z" },
+    { url = "https://files.pythonhosted.org/packages/18/3d/f9441a0d798bf2b1e645adc3265e55706aead1255ccdad3856dbdcffec14/pycryptodome-3.23.0-cp37-abi3-win_arm64.whl", hash = "sha256:11eeeb6917903876f134b56ba11abe95c0b0fd5e3330def218083c7d98bbcb3c", size = 1703675, upload-time = "2025-05-17T17:21:13.146Z" },
+    { url = "https://files.pythonhosted.org/packages/d9/12/e33935a0709c07de084d7d58d330ec3f4daf7910a18e77937affdb728452/pycryptodome-3.23.0-pp310-pypy310_pp73-macosx_10_15_x86_64.whl", hash = "sha256:ddb95b49df036ddd264a0ad246d1be5b672000f12d6961ea2c267083a5e19379", size = 1623886, upload-time = "2025-05-17T17:21:20.614Z" },
+    { url = "https://files.pythonhosted.org/packages/22/0b/aa8f9419f25870889bebf0b26b223c6986652bdf071f000623df11212c90/pycryptodome-3.23.0-pp310-pypy310_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:d8e95564beb8782abfd9e431c974e14563a794a4944c29d6d3b7b5ea042110b4", size = 1672151, upload-time = "2025-05-17T17:21:22.666Z" },
+    { url = "https://files.pythonhosted.org/packages/d4/5e/63f5cbde2342b7f70a39e591dbe75d9809d6338ce0b07c10406f1a140cdc/pycryptodome-3.23.0-pp310-pypy310_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:14e15c081e912c4b0d75632acd8382dfce45b258667aa3c67caf7a4d4c13f630", size = 1664461, upload-time = "2025-05-17T17:21:25.225Z" },
+    { url = "https://files.pythonhosted.org/packages/d6/92/608fbdad566ebe499297a86aae5f2a5263818ceeecd16733006f1600403c/pycryptodome-3.23.0-pp310-pypy310_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:a7fc76bf273353dc7e5207d172b83f569540fc9a28d63171061c42e361d22353", size = 1702440, upload-time = "2025-05-17T17:21:27.991Z" },
+    { url = "https://files.pythonhosted.org/packages/d1/92/2eadd1341abd2989cce2e2740b4423608ee2014acb8110438244ee97d7ff/pycryptodome-3.23.0-pp310-pypy310_pp73-win_amd64.whl", hash = "sha256:45c69ad715ca1a94f778215a11e66b7ff989d792a4d63b68dc586a1da1392ff5", size = 1803005, upload-time = "2025-05-17T17:21:31.37Z" },
+]
+
+[[package]]
+name = "pydantic"
+version = "2.12.5"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "annotated-types" },
+    { name = "pydantic-core" },
+    { name = "typing-extensions" },
+    { name = "typing-inspection" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/69/44/36f1a6e523abc58ae5f928898e4aca2e0ea509b5aa6f6f392a5d882be928/pydantic-2.12.5.tar.gz", hash = "sha256:4d351024c75c0f085a9febbb665ce8c0c6ec5d30e903bdb6394b7ede26aebb49", size = 821591, upload-time = "2025-11-26T15:11:46.471Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/5a/87/b70ad306ebb6f9b585f114d0ac2137d792b48be34d732d60e597c2f8465a/pydantic-2.12.5-py3-none-any.whl", hash = "sha256:e561593fccf61e8a20fc46dfc2dfe075b8be7d0188df33f221ad1f0139180f9d", size = 463580, upload-time = "2025-11-26T15:11:44.605Z" },
+]
+
+[[package]]
+name = "pydantic-core"
+version = "2.41.5"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "typing-extensions" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/71/70/23b021c950c2addd24ec408e9ab05d59b035b39d97cdc1130e1bce647bb6/pydantic_core-2.41.5.tar.gz", hash = "sha256:08daa51ea16ad373ffd5e7606252cc32f07bc72b28284b6bc9c6df804816476e", size = 460952, upload-time = "2025-11-04T13:43:49.098Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/c6/90/32c9941e728d564b411d574d8ee0cf09b12ec978cb22b294995bae5549a5/pydantic_core-2.41.5-cp310-cp310-macosx_10_12_x86_64.whl", hash = "sha256:77b63866ca88d804225eaa4af3e664c5faf3568cea95360d21f4725ab6e07146", size = 2107298, upload-time = "2025-11-04T13:39:04.116Z" },
+    { url = "https://files.pythonhosted.org/packages/fb/a8/61c96a77fe28993d9a6fb0f4127e05430a267b235a124545d79fea46dd65/pydantic_core-2.41.5-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:dfa8a0c812ac681395907e71e1274819dec685fec28273a28905df579ef137e2", size = 1901475, upload-time = "2025-11-04T13:39:06.055Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/b6/338abf60225acc18cdc08b4faef592d0310923d19a87fba1faf05af5346e/pydantic_core-2.41.5-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:5921a4d3ca3aee735d9fd163808f5e8dd6c6972101e4adbda9a4667908849b97", size = 1918815, upload-time = "2025-11-04T13:39:10.41Z" },
+    { url = "https://files.pythonhosted.org/packages/d1/1c/2ed0433e682983d8e8cba9c8d8ef274d4791ec6a6f24c58935b90e780e0a/pydantic_core-2.41.5-cp310-cp310-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:e25c479382d26a2a41b7ebea1043564a937db462816ea07afa8a44c0866d52f9", size = 2065567, upload-time = "2025-11-04T13:39:12.244Z" },
+    { url = "https://files.pythonhosted.org/packages/b3/24/cf84974ee7d6eae06b9e63289b7b8f6549d416b5c199ca2d7ce13bbcf619/pydantic_core-2.41.5-cp310-cp310-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:f547144f2966e1e16ae626d8ce72b4cfa0caedc7fa28052001c94fb2fcaa1c52", size = 2230442, upload-time = "2025-11-04T13:39:13.962Z" },
+    { url = "https://files.pythonhosted.org/packages/fd/21/4e287865504b3edc0136c89c9c09431be326168b1eb7841911cbc877a995/pydantic_core-2.41.5-cp310-cp310-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:6f52298fbd394f9ed112d56f3d11aabd0d5bd27beb3084cc3d8ad069483b8941", size = 2350956, upload-time = "2025-11-04T13:39:15.889Z" },
+    { url = "https://files.pythonhosted.org/packages/a8/76/7727ef2ffa4b62fcab916686a68a0426b9b790139720e1934e8ba797e238/pydantic_core-2.41.5-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:100baa204bb412b74fe285fb0f3a385256dad1d1879f0a5cb1499ed2e83d132a", size = 2068253, upload-time = "2025-11-04T13:39:17.403Z" },
+    { url = "https://files.pythonhosted.org/packages/d5/8c/a4abfc79604bcb4c748e18975c44f94f756f08fb04218d5cb87eb0d3a63e/pydantic_core-2.41.5-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:05a2c8852530ad2812cb7914dc61a1125dc4e06252ee98e5638a12da6cc6fb6c", size = 2177050, upload-time = "2025-11-04T13:39:19.351Z" },
+    { url = "https://files.pythonhosted.org/packages/67/b1/de2e9a9a79b480f9cb0b6e8b6ba4c50b18d4e89852426364c66aa82bb7b3/pydantic_core-2.41.5-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:29452c56df2ed968d18d7e21f4ab0ac55e71dc59524872f6fc57dcf4a3249ed2", size = 2147178, upload-time = "2025-11-04T13:39:21Z" },
+    { url = "https://files.pythonhosted.org/packages/16/c1/dfb33f837a47b20417500efaa0378adc6635b3c79e8369ff7a03c494b4ac/pydantic_core-2.41.5-cp310-cp310-musllinux_1_1_armv7l.whl", hash = "sha256:d5160812ea7a8a2ffbe233d8da666880cad0cbaf5d4de74ae15c313213d62556", size = 2341833, upload-time = "2025-11-04T13:39:22.606Z" },
+    { url = "https://files.pythonhosted.org/packages/47/36/00f398642a0f4b815a9a558c4f1dca1b4020a7d49562807d7bc9ff279a6c/pydantic_core-2.41.5-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:df3959765b553b9440adfd3c795617c352154e497a4eaf3752555cfb5da8fc49", size = 2321156, upload-time = "2025-11-04T13:39:25.843Z" },
+    { url = "https://files.pythonhosted.org/packages/7e/70/cad3acd89fde2010807354d978725ae111ddf6d0ea46d1ea1775b5c1bd0c/pydantic_core-2.41.5-cp310-cp310-win32.whl", hash = "sha256:1f8d33a7f4d5a7889e60dc39856d76d09333d8a6ed0f5f1190635cbec70ec4ba", size = 1989378, upload-time = "2025-11-04T13:39:27.92Z" },
+    { url = "https://files.pythonhosted.org/packages/76/92/d338652464c6c367e5608e4488201702cd1cbb0f33f7b6a85a60fe5f3720/pydantic_core-2.41.5-cp310-cp310-win_amd64.whl", hash = "sha256:62de39db01b8d593e45871af2af9e497295db8d73b085f6bfd0b18c83c70a8f9", size = 2013622, upload-time = "2025-11-04T13:39:29.848Z" },
+    { url = "https://files.pythonhosted.org/packages/e8/72/74a989dd9f2084b3d9530b0915fdda64ac48831c30dbf7c72a41a5232db8/pydantic_core-2.41.5-cp311-cp311-macosx_10_12_x86_64.whl", hash = "sha256:a3a52f6156e73e7ccb0f8cced536adccb7042be67cb45f9562e12b319c119da6", size = 2105873, upload-time = "2025-11-04T13:39:31.373Z" },
+    { url = "https://files.pythonhosted.org/packages/12/44/37e403fd9455708b3b942949e1d7febc02167662bf1a7da5b78ee1ea2842/pydantic_core-2.41.5-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:7f3bf998340c6d4b0c9a2f02d6a400e51f123b59565d74dc60d252ce888c260b", size = 1899826, upload-time = "2025-11-04T13:39:32.897Z" },
+    { url = "https://files.pythonhosted.org/packages/33/7f/1d5cab3ccf44c1935a359d51a8a2a9e1a654b744b5e7f80d41b88d501eec/pydantic_core-2.41.5-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:378bec5c66998815d224c9ca994f1e14c0c21cb95d2f52b6021cc0b2a58f2a5a", size = 1917869, upload-time = "2025-11-04T13:39:34.469Z" },
+    { url = "https://files.pythonhosted.org/packages/6e/6a/30d94a9674a7fe4f4744052ed6c5e083424510be1e93da5bc47569d11810/pydantic_core-2.41.5-cp311-cp311-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:e7b576130c69225432866fe2f4a469a85a54ade141d96fd396dffcf607b558f8", size = 2063890, upload-time = "2025-11-04T13:39:36.053Z" },
+    { url = "https://files.pythonhosted.org/packages/50/be/76e5d46203fcb2750e542f32e6c371ffa9b8ad17364cf94bb0818dbfb50c/pydantic_core-2.41.5-cp311-cp311-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:6cb58b9c66f7e4179a2d5e0f849c48eff5c1fca560994d6eb6543abf955a149e", size = 2229740, upload-time = "2025-11-04T13:39:37.753Z" },
+    { url = "https://files.pythonhosted.org/packages/d3/ee/fed784df0144793489f87db310a6bbf8118d7b630ed07aa180d6067e653a/pydantic_core-2.41.5-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:88942d3a3dff3afc8288c21e565e476fc278902ae4d6d134f1eeda118cc830b1", size = 2350021, upload-time = "2025-11-04T13:39:40.94Z" },
+    { url = "https://files.pythonhosted.org/packages/c8/be/8fed28dd0a180dca19e72c233cbf58efa36df055e5b9d90d64fd1740b828/pydantic_core-2.41.5-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:f31d95a179f8d64d90f6831d71fa93290893a33148d890ba15de25642c5d075b", size = 2066378, upload-time = "2025-11-04T13:39:42.523Z" },
+    { url = "https://files.pythonhosted.org/packages/b0/3b/698cf8ae1d536a010e05121b4958b1257f0b5522085e335360e53a6b1c8b/pydantic_core-2.41.5-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:c1df3d34aced70add6f867a8cf413e299177e0c22660cc767218373d0779487b", size = 2175761, upload-time = "2025-11-04T13:39:44.553Z" },
+    { url = "https://files.pythonhosted.org/packages/b8/ba/15d537423939553116dea94ce02f9c31be0fa9d0b806d427e0308ec17145/pydantic_core-2.41.5-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:4009935984bd36bd2c774e13f9a09563ce8de4abaa7226f5108262fa3e637284", size = 2146303, upload-time = "2025-11-04T13:39:46.238Z" },
+    { url = "https://files.pythonhosted.org/packages/58/7f/0de669bf37d206723795f9c90c82966726a2ab06c336deba4735b55af431/pydantic_core-2.41.5-cp311-cp311-musllinux_1_1_armv7l.whl", hash = "sha256:34a64bc3441dc1213096a20fe27e8e128bd3ff89921706e83c0b1ac971276594", size = 2340355, upload-time = "2025-11-04T13:39:48.002Z" },
+    { url = "https://files.pythonhosted.org/packages/e5/de/e7482c435b83d7e3c3ee5ee4451f6e8973cff0eb6007d2872ce6383f6398/pydantic_core-2.41.5-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:c9e19dd6e28fdcaa5a1de679aec4141f691023916427ef9bae8584f9c2fb3b0e", size = 2319875, upload-time = "2025-11-04T13:39:49.705Z" },
+    { url = "https://files.pythonhosted.org/packages/fe/e6/8c9e81bb6dd7560e33b9053351c29f30c8194b72f2d6932888581f503482/pydantic_core-2.41.5-cp311-cp311-win32.whl", hash = "sha256:2c010c6ded393148374c0f6f0bf89d206bf3217f201faa0635dcd56bd1520f6b", size = 1987549, upload-time = "2025-11-04T13:39:51.842Z" },
+    { url = "https://files.pythonhosted.org/packages/11/66/f14d1d978ea94d1bc21fc98fcf570f9542fe55bfcc40269d4e1a21c19bf7/pydantic_core-2.41.5-cp311-cp311-win_amd64.whl", hash = "sha256:76ee27c6e9c7f16f47db7a94157112a2f3a00e958bc626e2f4ee8bec5c328fbe", size = 2011305, upload-time = "2025-11-04T13:39:53.485Z" },
+    { url = "https://files.pythonhosted.org/packages/56/d8/0e271434e8efd03186c5386671328154ee349ff0354d83c74f5caaf096ed/pydantic_core-2.41.5-cp311-cp311-win_arm64.whl", hash = "sha256:4bc36bbc0b7584de96561184ad7f012478987882ebf9f9c389b23f432ea3d90f", size = 1972902, upload-time = "2025-11-04T13:39:56.488Z" },
+    { url = "https://files.pythonhosted.org/packages/5f/5d/5f6c63eebb5afee93bcaae4ce9a898f3373ca23df3ccaef086d0233a35a7/pydantic_core-2.41.5-cp312-cp312-macosx_10_12_x86_64.whl", hash = "sha256:f41a7489d32336dbf2199c8c0a215390a751c5b014c2c1c5366e817202e9cdf7", size = 2110990, upload-time = "2025-11-04T13:39:58.079Z" },
+    { url = "https://files.pythonhosted.org/packages/aa/32/9c2e8ccb57c01111e0fd091f236c7b371c1bccea0fa85247ac55b1e2b6b6/pydantic_core-2.41.5-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:070259a8818988b9a84a449a2a7337c7f430a22acc0859c6b110aa7212a6d9c0", size = 1896003, upload-time = "2025-11-04T13:39:59.956Z" },
+    { url = "https://files.pythonhosted.org/packages/68/b8/a01b53cb0e59139fbc9e4fda3e9724ede8de279097179be4ff31f1abb65a/pydantic_core-2.41.5-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:e96cea19e34778f8d59fe40775a7a574d95816eb150850a85a7a4c8f4b94ac69", size = 1919200, upload-time = "2025-11-04T13:40:02.241Z" },
+    { url = "https://files.pythonhosted.org/packages/38/de/8c36b5198a29bdaade07b5985e80a233a5ac27137846f3bc2d3b40a47360/pydantic_core-2.41.5-cp312-cp312-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:ed2e99c456e3fadd05c991f8f437ef902e00eedf34320ba2b0842bd1c3ca3a75", size = 2052578, upload-time = "2025-11-04T13:40:04.401Z" },
+    { url = "https://files.pythonhosted.org/packages/00/b5/0e8e4b5b081eac6cb3dbb7e60a65907549a1ce035a724368c330112adfdd/pydantic_core-2.41.5-cp312-cp312-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:65840751b72fbfd82c3c640cff9284545342a4f1eb1586ad0636955b261b0b05", size = 2208504, upload-time = "2025-11-04T13:40:06.072Z" },
+    { url = "https://files.pythonhosted.org/packages/77/56/87a61aad59c7c5b9dc8caad5a41a5545cba3810c3e828708b3d7404f6cef/pydantic_core-2.41.5-cp312-cp312-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:e536c98a7626a98feb2d3eaf75944ef6f3dbee447e1f841eae16f2f0a72d8ddc", size = 2335816, upload-time = "2025-11-04T13:40:07.835Z" },
+    { url = "https://files.pythonhosted.org/packages/0d/76/941cc9f73529988688a665a5c0ecff1112b3d95ab48f81db5f7606f522d3/pydantic_core-2.41.5-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:eceb81a8d74f9267ef4081e246ffd6d129da5d87e37a77c9bde550cb04870c1c", size = 2075366, upload-time = "2025-11-04T13:40:09.804Z" },
+    { url = "https://files.pythonhosted.org/packages/d3/43/ebef01f69baa07a482844faaa0a591bad1ef129253ffd0cdaa9d8a7f72d3/pydantic_core-2.41.5-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:d38548150c39b74aeeb0ce8ee1d8e82696f4a4e16ddc6de7b1d8823f7de4b9b5", size = 2171698, upload-time = "2025-11-04T13:40:12.004Z" },
+    { url = "https://files.pythonhosted.org/packages/b1/87/41f3202e4193e3bacfc2c065fab7706ebe81af46a83d3e27605029c1f5a6/pydantic_core-2.41.5-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:c23e27686783f60290e36827f9c626e63154b82b116d7fe9adba1fda36da706c", size = 2132603, upload-time = "2025-11-04T13:40:13.868Z" },
+    { url = "https://files.pythonhosted.org/packages/49/7d/4c00df99cb12070b6bccdef4a195255e6020a550d572768d92cc54dba91a/pydantic_core-2.41.5-cp312-cp312-musllinux_1_1_armv7l.whl", hash = "sha256:482c982f814460eabe1d3bb0adfdc583387bd4691ef00b90575ca0d2b6fe2294", size = 2329591, upload-time = "2025-11-04T13:40:15.672Z" },
+    { url = "https://files.pythonhosted.org/packages/cc/6a/ebf4b1d65d458f3cda6a7335d141305dfa19bdc61140a884d165a8a1bbc7/pydantic_core-2.41.5-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:bfea2a5f0b4d8d43adf9d7b8bf019fb46fdd10a2e5cde477fbcb9d1fa08c68e1", size = 2319068, upload-time = "2025-11-04T13:40:17.532Z" },
+    { url = "https://files.pythonhosted.org/packages/49/3b/774f2b5cd4192d5ab75870ce4381fd89cf218af999515baf07e7206753f0/pydantic_core-2.41.5-cp312-cp312-win32.whl", hash = "sha256:b74557b16e390ec12dca509bce9264c3bbd128f8a2c376eaa68003d7f327276d", size = 1985908, upload-time = "2025-11-04T13:40:19.309Z" },
+    { url = "https://files.pythonhosted.org/packages/86/45/00173a033c801cacf67c190fef088789394feaf88a98a7035b0e40d53dc9/pydantic_core-2.41.5-cp312-cp312-win_amd64.whl", hash = "sha256:1962293292865bca8e54702b08a4f26da73adc83dd1fcf26fbc875b35d81c815", size = 2020145, upload-time = "2025-11-04T13:40:21.548Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/22/91fbc821fa6d261b376a3f73809f907cec5ca6025642c463d3488aad22fb/pydantic_core-2.41.5-cp312-cp312-win_arm64.whl", hash = "sha256:1746d4a3d9a794cacae06a5eaaccb4b8643a131d45fbc9af23e353dc0a5ba5c3", size = 1976179, upload-time = "2025-11-04T13:40:23.393Z" },
+    { url = "https://files.pythonhosted.org/packages/87/06/8806241ff1f70d9939f9af039c6c35f2360cf16e93c2ca76f184e76b1564/pydantic_core-2.41.5-cp313-cp313-macosx_10_12_x86_64.whl", hash = "sha256:941103c9be18ac8daf7b7adca8228f8ed6bb7a1849020f643b3a14d15b1924d9", size = 2120403, upload-time = "2025-11-04T13:40:25.248Z" },
+    { url = "https://files.pythonhosted.org/packages/94/02/abfa0e0bda67faa65fef1c84971c7e45928e108fe24333c81f3bfe35d5f5/pydantic_core-2.41.5-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:112e305c3314f40c93998e567879e887a3160bb8689ef3d2c04b6cc62c33ac34", size = 1896206, upload-time = "2025-11-04T13:40:27.099Z" },
+    { url = "https://files.pythonhosted.org/packages/15/df/a4c740c0943e93e6500f9eb23f4ca7ec9bf71b19e608ae5b579678c8d02f/pydantic_core-2.41.5-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:0cbaad15cb0c90aa221d43c00e77bb33c93e8d36e0bf74760cd00e732d10a6a0", size = 1919307, upload-time = "2025-11-04T13:40:29.806Z" },
+    { url = "https://files.pythonhosted.org/packages/9a/e3/6324802931ae1d123528988e0e86587c2072ac2e5394b4bc2bc34b61ff6e/pydantic_core-2.41.5-cp313-cp313-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:03ca43e12fab6023fc79d28ca6b39b05f794ad08ec2feccc59a339b02f2b3d33", size = 2063258, upload-time = "2025-11-04T13:40:33.544Z" },
+    { url = "https://files.pythonhosted.org/packages/c9/d4/2230d7151d4957dd79c3044ea26346c148c98fbf0ee6ebd41056f2d62ab5/pydantic_core-2.41.5-cp313-cp313-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:dc799088c08fa04e43144b164feb0c13f9a0bc40503f8df3e9fde58a3c0c101e", size = 2214917, upload-time = "2025-11-04T13:40:35.479Z" },
+    { url = "https://files.pythonhosted.org/packages/e6/9f/eaac5df17a3672fef0081b6c1bb0b82b33ee89aa5cec0d7b05f52fd4a1fa/pydantic_core-2.41.5-cp313-cp313-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:97aeba56665b4c3235a0e52b2c2f5ae9cd071b8a8310ad27bddb3f7fb30e9aa2", size = 2332186, upload-time = "2025-11-04T13:40:37.436Z" },
+    { url = "https://files.pythonhosted.org/packages/cf/4e/35a80cae583a37cf15604b44240e45c05e04e86f9cfd766623149297e971/pydantic_core-2.41.5-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:406bf18d345822d6c21366031003612b9c77b3e29ffdb0f612367352aab7d586", size = 2073164, upload-time = "2025-11-04T13:40:40.289Z" },
+    { url = "https://files.pythonhosted.org/packages/bf/e3/f6e262673c6140dd3305d144d032f7bd5f7497d3871c1428521f19f9efa2/pydantic_core-2.41.5-cp313-cp313-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:b93590ae81f7010dbe380cdeab6f515902ebcbefe0b9327cc4804d74e93ae69d", size = 2179146, upload-time = "2025-11-04T13:40:42.809Z" },
+    { url = "https://files.pythonhosted.org/packages/75/c7/20bd7fc05f0c6ea2056a4565c6f36f8968c0924f19b7d97bbfea55780e73/pydantic_core-2.41.5-cp313-cp313-musllinux_1_1_aarch64.whl", hash = "sha256:01a3d0ab748ee531f4ea6c3e48ad9dac84ddba4b0d82291f87248f2f9de8d740", size = 2137788, upload-time = "2025-11-04T13:40:44.752Z" },
+    { url = "https://files.pythonhosted.org/packages/3a/8d/34318ef985c45196e004bc46c6eab2eda437e744c124ef0dbe1ff2c9d06b/pydantic_core-2.41.5-cp313-cp313-musllinux_1_1_armv7l.whl", hash = "sha256:6561e94ba9dacc9c61bce40e2d6bdc3bfaa0259d3ff36ace3b1e6901936d2e3e", size = 2340133, upload-time = "2025-11-04T13:40:46.66Z" },
+    { url = "https://files.pythonhosted.org/packages/9c/59/013626bf8c78a5a5d9350d12e7697d3d4de951a75565496abd40ccd46bee/pydantic_core-2.41.5-cp313-cp313-musllinux_1_1_x86_64.whl", hash = "sha256:915c3d10f81bec3a74fbd4faebe8391013ba61e5a1a8d48c4455b923bdda7858", size = 2324852, upload-time = "2025-11-04T13:40:48.575Z" },
+    { url = "https://files.pythonhosted.org/packages/1a/d9/c248c103856f807ef70c18a4f986693a46a8ffe1602e5d361485da502d20/pydantic_core-2.41.5-cp313-cp313-win32.whl", hash = "sha256:650ae77860b45cfa6e2cdafc42618ceafab3a2d9a3811fcfbd3bbf8ac3c40d36", size = 1994679, upload-time = "2025-11-04T13:40:50.619Z" },
+    { url = "https://files.pythonhosted.org/packages/9e/8b/341991b158ddab181cff136acd2552c9f35bd30380422a639c0671e99a91/pydantic_core-2.41.5-cp313-cp313-win_amd64.whl", hash = "sha256:79ec52ec461e99e13791ec6508c722742ad745571f234ea6255bed38c6480f11", size = 2019766, upload-time = "2025-11-04T13:40:52.631Z" },
+    { url = "https://files.pythonhosted.org/packages/73/7d/f2f9db34af103bea3e09735bb40b021788a5e834c81eedb541991badf8f5/pydantic_core-2.41.5-cp313-cp313-win_arm64.whl", hash = "sha256:3f84d5c1b4ab906093bdc1ff10484838aca54ef08de4afa9de0f5f14d69639cd", size = 1981005, upload-time = "2025-11-04T13:40:54.734Z" },
+    { url = "https://files.pythonhosted.org/packages/ea/28/46b7c5c9635ae96ea0fbb779e271a38129df2550f763937659ee6c5dbc65/pydantic_core-2.41.5-cp314-cp314-macosx_10_12_x86_64.whl", hash = "sha256:3f37a19d7ebcdd20b96485056ba9e8b304e27d9904d233d7b1015db320e51f0a", size = 2119622, upload-time = "2025-11-04T13:40:56.68Z" },
+    { url = "https://files.pythonhosted.org/packages/74/1a/145646e5687e8d9a1e8d09acb278c8535ebe9e972e1f162ed338a622f193/pydantic_core-2.41.5-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:1d1d9764366c73f996edd17abb6d9d7649a7eb690006ab6adbda117717099b14", size = 1891725, upload-time = "2025-11-04T13:40:58.807Z" },
+    { url = "https://files.pythonhosted.org/packages/23/04/e89c29e267b8060b40dca97bfc64a19b2a3cf99018167ea1677d96368273/pydantic_core-2.41.5-cp314-cp314-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:25e1c2af0fce638d5f1988b686f3b3ea8cd7de5f244ca147c777769e798a9cd1", size = 1915040, upload-time = "2025-11-04T13:41:00.853Z" },
+    { url = "https://files.pythonhosted.org/packages/84/a3/15a82ac7bd97992a82257f777b3583d3e84bdb06ba6858f745daa2ec8a85/pydantic_core-2.41.5-cp314-cp314-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:506d766a8727beef16b7adaeb8ee6217c64fc813646b424d0804d67c16eddb66", size = 2063691, upload-time = "2025-11-04T13:41:03.504Z" },
+    { url = "https://files.pythonhosted.org/packages/74/9b/0046701313c6ef08c0c1cf0e028c67c770a4e1275ca73131563c5f2a310a/pydantic_core-2.41.5-cp314-cp314-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:4819fa52133c9aa3c387b3328f25c1facc356491e6135b459f1de698ff64d869", size = 2213897, upload-time = "2025-11-04T13:41:05.804Z" },
+    { url = "https://files.pythonhosted.org/packages/8a/cd/6bac76ecd1b27e75a95ca3a9a559c643b3afcd2dd62086d4b7a32a18b169/pydantic_core-2.41.5-cp314-cp314-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:2b761d210c9ea91feda40d25b4efe82a1707da2ef62901466a42492c028553a2", size = 2333302, upload-time = "2025-11-04T13:41:07.809Z" },
+    { url = "https://files.pythonhosted.org/packages/4c/d2/ef2074dc020dd6e109611a8be4449b98cd25e1b9b8a303c2f0fca2f2bcf7/pydantic_core-2.41.5-cp314-cp314-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:22f0fb8c1c583a3b6f24df2470833b40207e907b90c928cc8d3594b76f874375", size = 2064877, upload-time = "2025-11-04T13:41:09.827Z" },
+    { url = "https://files.pythonhosted.org/packages/18/66/e9db17a9a763d72f03de903883c057b2592c09509ccfe468187f2a2eef29/pydantic_core-2.41.5-cp314-cp314-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:2782c870e99878c634505236d81e5443092fba820f0373997ff75f90f68cd553", size = 2180680, upload-time = "2025-11-04T13:41:12.379Z" },
+    { url = "https://files.pythonhosted.org/packages/d3/9e/3ce66cebb929f3ced22be85d4c2399b8e85b622db77dad36b73c5387f8f8/pydantic_core-2.41.5-cp314-cp314-musllinux_1_1_aarch64.whl", hash = "sha256:0177272f88ab8312479336e1d777f6b124537d47f2123f89cb37e0accea97f90", size = 2138960, upload-time = "2025-11-04T13:41:14.627Z" },
+    { url = "https://files.pythonhosted.org/packages/a6/62/205a998f4327d2079326b01abee48e502ea739d174f0a89295c481a2272e/pydantic_core-2.41.5-cp314-cp314-musllinux_1_1_armv7l.whl", hash = "sha256:63510af5e38f8955b8ee5687740d6ebf7c2a0886d15a6d65c32814613681bc07", size = 2339102, upload-time = "2025-11-04T13:41:16.868Z" },
+    { url = "https://files.pythonhosted.org/packages/3c/0d/f05e79471e889d74d3d88f5bd20d0ed189ad94c2423d81ff8d0000aab4ff/pydantic_core-2.41.5-cp314-cp314-musllinux_1_1_x86_64.whl", hash = "sha256:e56ba91f47764cc14f1daacd723e3e82d1a89d783f0f5afe9c364b8bb491ccdb", size = 2326039, upload-time = "2025-11-04T13:41:18.934Z" },
+    { url = "https://files.pythonhosted.org/packages/ec/e1/e08a6208bb100da7e0c4b288eed624a703f4d129bde2da475721a80cab32/pydantic_core-2.41.5-cp314-cp314-win32.whl", hash = "sha256:aec5cf2fd867b4ff45b9959f8b20ea3993fc93e63c7363fe6851424c8a7e7c23", size = 1995126, upload-time = "2025-11-04T13:41:21.418Z" },
+    { url = "https://files.pythonhosted.org/packages/48/5d/56ba7b24e9557f99c9237e29f5c09913c81eeb2f3217e40e922353668092/pydantic_core-2.41.5-cp314-cp314-win_amd64.whl", hash = "sha256:8e7c86f27c585ef37c35e56a96363ab8de4e549a95512445b85c96d3e2f7c1bf", size = 2015489, upload-time = "2025-11-04T13:41:24.076Z" },
+    { url = "https://files.pythonhosted.org/packages/4e/bb/f7a190991ec9e3e0ba22e4993d8755bbc4a32925c0b5b42775c03e8148f9/pydantic_core-2.41.5-cp314-cp314-win_arm64.whl", hash = "sha256:e672ba74fbc2dc8eea59fb6d4aed6845e6905fc2a8afe93175d94a83ba2a01a0", size = 1977288, upload-time = "2025-11-04T13:41:26.33Z" },
+    { url = "https://files.pythonhosted.org/packages/92/ed/77542d0c51538e32e15afe7899d79efce4b81eee631d99850edc2f5e9349/pydantic_core-2.41.5-cp314-cp314t-macosx_10_12_x86_64.whl", hash = "sha256:8566def80554c3faa0e65ac30ab0932b9e3a5cd7f8323764303d468e5c37595a", size = 2120255, upload-time = "2025-11-04T13:41:28.569Z" },
+    { url = "https://files.pythonhosted.org/packages/bb/3d/6913dde84d5be21e284439676168b28d8bbba5600d838b9dca99de0fad71/pydantic_core-2.41.5-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:b80aa5095cd3109962a298ce14110ae16b8c1aece8b72f9dafe81cf597ad80b3", size = 1863760, upload-time = "2025-11-04T13:41:31.055Z" },
+    { url = "https://files.pythonhosted.org/packages/5a/f0/e5e6b99d4191da102f2b0eb9687aaa7f5bea5d9964071a84effc3e40f997/pydantic_core-2.41.5-cp314-cp314t-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:3006c3dd9ba34b0c094c544c6006cc79e87d8612999f1a5d43b769b89181f23c", size = 1878092, upload-time = "2025-11-04T13:41:33.21Z" },
+    { url = "https://files.pythonhosted.org/packages/71/48/36fb760642d568925953bcc8116455513d6e34c4beaa37544118c36aba6d/pydantic_core-2.41.5-cp314-cp314t-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:72f6c8b11857a856bcfa48c86f5368439f74453563f951e473514579d44aa612", size = 2053385, upload-time = "2025-11-04T13:41:35.508Z" },
+    { url = "https://files.pythonhosted.org/packages/20/25/92dc684dd8eb75a234bc1c764b4210cf2646479d54b47bf46061657292a8/pydantic_core-2.41.5-cp314-cp314t-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:5cb1b2f9742240e4bb26b652a5aeb840aa4b417c7748b6f8387927bc6e45e40d", size = 2218832, upload-time = "2025-11-04T13:41:37.732Z" },
+    { url = "https://files.pythonhosted.org/packages/e2/09/f53e0b05023d3e30357d82eb35835d0f6340ca344720a4599cd663dca599/pydantic_core-2.41.5-cp314-cp314t-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:bd3d54f38609ff308209bd43acea66061494157703364ae40c951f83ba99a1a9", size = 2327585, upload-time = "2025-11-04T13:41:40Z" },
+    { url = "https://files.pythonhosted.org/packages/aa/4e/2ae1aa85d6af35a39b236b1b1641de73f5a6ac4d5a7509f77b814885760c/pydantic_core-2.41.5-cp314-cp314t-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:2ff4321e56e879ee8d2a879501c8e469414d948f4aba74a2d4593184eb326660", size = 2041078, upload-time = "2025-11-04T13:41:42.323Z" },
+    { url = "https://files.pythonhosted.org/packages/cd/13/2e215f17f0ef326fc72afe94776edb77525142c693767fc347ed6288728d/pydantic_core-2.41.5-cp314-cp314t-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:d0d2568a8c11bf8225044aa94409e21da0cb09dcdafe9ecd10250b2baad531a9", size = 2173914, upload-time = "2025-11-04T13:41:45.221Z" },
+    { url = "https://files.pythonhosted.org/packages/02/7a/f999a6dcbcd0e5660bc348a3991c8915ce6599f4f2c6ac22f01d7a10816c/pydantic_core-2.41.5-cp314-cp314t-musllinux_1_1_aarch64.whl", hash = "sha256:a39455728aabd58ceabb03c90e12f71fd30fa69615760a075b9fec596456ccc3", size = 2129560, upload-time = "2025-11-04T13:41:47.474Z" },
+    { url = "https://files.pythonhosted.org/packages/3a/b1/6c990ac65e3b4c079a4fb9f5b05f5b013afa0f4ed6780a3dd236d2cbdc64/pydantic_core-2.41.5-cp314-cp314t-musllinux_1_1_armv7l.whl", hash = "sha256:239edca560d05757817c13dc17c50766136d21f7cd0fac50295499ae24f90fdf", size = 2329244, upload-time = "2025-11-04T13:41:49.992Z" },
+    { url = "https://files.pythonhosted.org/packages/d9/02/3c562f3a51afd4d88fff8dffb1771b30cfdfd79befd9883ee094f5b6c0d8/pydantic_core-2.41.5-cp314-cp314t-musllinux_1_1_x86_64.whl", hash = "sha256:2a5e06546e19f24c6a96a129142a75cee553cc018ffee48a460059b1185f4470", size = 2331955, upload-time = "2025-11-04T13:41:54.079Z" },
+    { url = "https://files.pythonhosted.org/packages/5c/96/5fb7d8c3c17bc8c62fdb031c47d77a1af698f1d7a406b0f79aaa1338f9ad/pydantic_core-2.41.5-cp314-cp314t-win32.whl", hash = "sha256:b4ececa40ac28afa90871c2cc2b9ffd2ff0bf749380fbdf57d165fd23da353aa", size = 1988906, upload-time = "2025-11-04T13:41:56.606Z" },
+    { url = "https://files.pythonhosted.org/packages/22/ed/182129d83032702912c2e2d8bbe33c036f342cc735737064668585dac28f/pydantic_core-2.41.5-cp314-cp314t-win_amd64.whl", hash = "sha256:80aa89cad80b32a912a65332f64a4450ed00966111b6615ca6816153d3585a8c", size = 1981607, upload-time = "2025-11-04T13:41:58.889Z" },
+    { url = "https://files.pythonhosted.org/packages/9f/ed/068e41660b832bb0b1aa5b58011dea2a3fe0ba7861ff38c4d4904c1c1a99/pydantic_core-2.41.5-cp314-cp314t-win_arm64.whl", hash = "sha256:35b44f37a3199f771c3eaa53051bc8a70cd7b54f333531c59e29fd4db5d15008", size = 1974769, upload-time = "2025-11-04T13:42:01.186Z" },
+    { url = "https://files.pythonhosted.org/packages/11/72/90fda5ee3b97e51c494938a4a44c3a35a9c96c19bba12372fb9c634d6f57/pydantic_core-2.41.5-graalpy311-graalpy242_311_native-macosx_10_12_x86_64.whl", hash = "sha256:b96d5f26b05d03cc60f11a7761a5ded1741da411e7fe0909e27a5e6a0cb7b034", size = 2115441, upload-time = "2025-11-04T13:42:39.557Z" },
+    { url = "https://files.pythonhosted.org/packages/1f/53/8942f884fa33f50794f119012dc6a1a02ac43a56407adaac20463df8e98f/pydantic_core-2.41.5-graalpy311-graalpy242_311_native-macosx_11_0_arm64.whl", hash = "sha256:634e8609e89ceecea15e2d61bc9ac3718caaaa71963717bf3c8f38bfde64242c", size = 1930291, upload-time = "2025-11-04T13:42:42.169Z" },
+    { url = "https://files.pythonhosted.org/packages/79/c8/ecb9ed9cd942bce09fc888ee960b52654fbdbede4ba6c2d6e0d3b1d8b49c/pydantic_core-2.41.5-graalpy311-graalpy242_311_native-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:93e8740d7503eb008aa2df04d3b9735f845d43ae845e6dcd2be0b55a2da43cd2", size = 1948632, upload-time = "2025-11-04T13:42:44.564Z" },
+    { url = "https://files.pythonhosted.org/packages/2e/1b/687711069de7efa6af934e74f601e2a4307365e8fdc404703afc453eab26/pydantic_core-2.41.5-graalpy311-graalpy242_311_native-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:f15489ba13d61f670dcc96772e733aad1a6f9c429cc27574c6cdaed82d0146ad", size = 2138905, upload-time = "2025-11-04T13:42:47.156Z" },
+    { url = "https://files.pythonhosted.org/packages/09/32/59b0c7e63e277fa7911c2fc70ccfb45ce4b98991e7ef37110663437005af/pydantic_core-2.41.5-graalpy312-graalpy250_312_native-macosx_10_12_x86_64.whl", hash = "sha256:7da7087d756b19037bc2c06edc6c170eeef3c3bafcb8f532ff17d64dc427adfd", size = 2110495, upload-time = "2025-11-04T13:42:49.689Z" },
+    { url = "https://files.pythonhosted.org/packages/aa/81/05e400037eaf55ad400bcd318c05bb345b57e708887f07ddb2d20e3f0e98/pydantic_core-2.41.5-graalpy312-graalpy250_312_native-macosx_11_0_arm64.whl", hash = "sha256:aabf5777b5c8ca26f7824cb4a120a740c9588ed58df9b2d196ce92fba42ff8dc", size = 1915388, upload-time = "2025-11-04T13:42:52.215Z" },
+    { url = "https://files.pythonhosted.org/packages/6e/0d/e3549b2399f71d56476b77dbf3cf8937cec5cd70536bdc0e374a421d0599/pydantic_core-2.41.5-graalpy312-graalpy250_312_native-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:c007fe8a43d43b3969e8469004e9845944f1a80e6acd47c150856bb87f230c56", size = 1942879, upload-time = "2025-11-04T13:42:56.483Z" },
+    { url = "https://files.pythonhosted.org/packages/f7/07/34573da085946b6a313d7c42f82f16e8920bfd730665de2d11c0c37a74b5/pydantic_core-2.41.5-graalpy312-graalpy250_312_native-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:76d0819de158cd855d1cbb8fcafdf6f5cf1eb8e470abe056d5d161106e38062b", size = 2139017, upload-time = "2025-11-04T13:42:59.471Z" },
+    { url = "https://files.pythonhosted.org/packages/e6/b0/1a2aa41e3b5a4ba11420aba2d091b2d17959c8d1519ece3627c371951e73/pydantic_core-2.41.5-pp310-pypy310_pp73-macosx_10_12_x86_64.whl", hash = "sha256:b5819cd790dbf0c5eb9f82c73c16b39a65dd6dd4d1439dcdea7816ec9adddab8", size = 2103351, upload-time = "2025-11-04T13:43:02.058Z" },
+    { url = "https://files.pythonhosted.org/packages/a4/ee/31b1f0020baaf6d091c87900ae05c6aeae101fa4e188e1613c80e4f1ea31/pydantic_core-2.41.5-pp310-pypy310_pp73-macosx_11_0_arm64.whl", hash = "sha256:5a4e67afbc95fa5c34cf27d9089bca7fcab4e51e57278d710320a70b956d1b9a", size = 1925363, upload-time = "2025-11-04T13:43:05.159Z" },
+    { url = "https://files.pythonhosted.org/packages/e1/89/ab8e86208467e467a80deaca4e434adac37b10a9d134cd2f99b28a01e483/pydantic_core-2.41.5-pp310-pypy310_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:ece5c59f0ce7d001e017643d8d24da587ea1f74f6993467d85ae8a5ef9d4f42b", size = 2135615, upload-time = "2025-11-04T13:43:08.116Z" },
+    { url = "https://files.pythonhosted.org/packages/99/0a/99a53d06dd0348b2008f2f30884b34719c323f16c3be4e6cc1203b74a91d/pydantic_core-2.41.5-pp310-pypy310_pp73-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:16f80f7abe3351f8ea6858914ddc8c77e02578544a0ebc15b4c2e1a0e813b0b2", size = 2175369, upload-time = "2025-11-04T13:43:12.49Z" },
+    { url = "https://files.pythonhosted.org/packages/6d/94/30ca3b73c6d485b9bb0bc66e611cff4a7138ff9736b7e66bcf0852151636/pydantic_core-2.41.5-pp310-pypy310_pp73-musllinux_1_1_aarch64.whl", hash = "sha256:33cb885e759a705b426baada1fe68cbb0a2e68e34c5d0d0289a364cf01709093", size = 2144218, upload-time = "2025-11-04T13:43:15.431Z" },
+    { url = "https://files.pythonhosted.org/packages/87/57/31b4f8e12680b739a91f472b5671294236b82586889ef764b5fbc6669238/pydantic_core-2.41.5-pp310-pypy310_pp73-musllinux_1_1_armv7l.whl", hash = "sha256:c8d8b4eb992936023be7dee581270af5c6e0697a8559895f527f5b7105ecd36a", size = 2329951, upload-time = "2025-11-04T13:43:18.062Z" },
+    { url = "https://files.pythonhosted.org/packages/7d/73/3c2c8edef77b8f7310e6fb012dbc4b8551386ed575b9eb6fb2506e28a7eb/pydantic_core-2.41.5-pp310-pypy310_pp73-musllinux_1_1_x86_64.whl", hash = "sha256:242a206cd0318f95cd21bdacff3fcc3aab23e79bba5cac3db5a841c9ef9c6963", size = 2318428, upload-time = "2025-11-04T13:43:20.679Z" },
+    { url = "https://files.pythonhosted.org/packages/2f/02/8559b1f26ee0d502c74f9cca5c0d2fd97e967e083e006bbbb4e97f3a043a/pydantic_core-2.41.5-pp310-pypy310_pp73-win_amd64.whl", hash = "sha256:d3a978c4f57a597908b7e697229d996d77a6d3c94901e9edee593adada95ce1a", size = 2147009, upload-time = "2025-11-04T13:43:23.286Z" },
+    { url = "https://files.pythonhosted.org/packages/5f/9b/1b3f0e9f9305839d7e84912f9e8bfbd191ed1b1ef48083609f0dabde978c/pydantic_core-2.41.5-pp311-pypy311_pp73-macosx_10_12_x86_64.whl", hash = "sha256:b2379fa7ed44ddecb5bfe4e48577d752db9fc10be00a6b7446e9663ba143de26", size = 2101980, upload-time = "2025-11-04T13:43:25.97Z" },
+    { url = "https://files.pythonhosted.org/packages/a4/ed/d71fefcb4263df0da6a85b5d8a7508360f2f2e9b3bf5814be9c8bccdccc1/pydantic_core-2.41.5-pp311-pypy311_pp73-macosx_11_0_arm64.whl", hash = "sha256:266fb4cbf5e3cbd0b53669a6d1b039c45e3ce651fd5442eff4d07c2cc8d66808", size = 1923865, upload-time = "2025-11-04T13:43:28.763Z" },
+    { url = "https://files.pythonhosted.org/packages/ce/3a/626b38db460d675f873e4444b4bb030453bbe7b4ba55df821d026a0493c4/pydantic_core-2.41.5-pp311-pypy311_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:58133647260ea01e4d0500089a8c4f07bd7aa6ce109682b1426394988d8aaacc", size = 2134256, upload-time = "2025-11-04T13:43:31.71Z" },
+    { url = "https://files.pythonhosted.org/packages/83/d9/8412d7f06f616bbc053d30cb4e5f76786af3221462ad5eee1f202021eb4e/pydantic_core-2.41.5-pp311-pypy311_pp73-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:287dad91cfb551c363dc62899a80e9e14da1f0e2b6ebde82c806612ca2a13ef1", size = 2174762, upload-time = "2025-11-04T13:43:34.744Z" },
+    { url = "https://files.pythonhosted.org/packages/55/4c/162d906b8e3ba3a99354e20faa1b49a85206c47de97a639510a0e673f5da/pydantic_core-2.41.5-pp311-pypy311_pp73-musllinux_1_1_aarch64.whl", hash = "sha256:03b77d184b9eb40240ae9fd676ca364ce1085f203e1b1256f8ab9984dca80a84", size = 2143141, upload-time = "2025-11-04T13:43:37.701Z" },
+    { url = "https://files.pythonhosted.org/packages/1f/f2/f11dd73284122713f5f89fc940f370d035fa8e1e078d446b3313955157fe/pydantic_core-2.41.5-pp311-pypy311_pp73-musllinux_1_1_armv7l.whl", hash = "sha256:a668ce24de96165bb239160b3d854943128f4334822900534f2fe947930e5770", size = 2330317, upload-time = "2025-11-04T13:43:40.406Z" },
+    { url = "https://files.pythonhosted.org/packages/88/9d/b06ca6acfe4abb296110fb1273a4d848a0bfb2ff65f3ee92127b3244e16b/pydantic_core-2.41.5-pp311-pypy311_pp73-musllinux_1_1_x86_64.whl", hash = "sha256:f14f8f046c14563f8eb3f45f499cc658ab8d10072961e07225e507adb700e93f", size = 2316992, upload-time = "2025-11-04T13:43:43.602Z" },
+    { url = "https://files.pythonhosted.org/packages/36/c7/cfc8e811f061c841d7990b0201912c3556bfeb99cdcb7ed24adc8d6f8704/pydantic_core-2.41.5-pp311-pypy311_pp73-win_amd64.whl", hash = "sha256:56121965f7a4dc965bff783d70b907ddf3d57f6eba29b6d2e5dabfaf07799c51", size = 2145302, upload-time = "2025-11-04T13:43:46.64Z" },
+]
+
+[[package]]
+name = "pygments"
+version = "2.19.2"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/b0/77/a5b8c569bf593b0140bde72ea885a803b82086995367bf2037de0159d924/pygments-2.19.2.tar.gz", hash = "sha256:636cb2477cec7f8952536970bc533bc43743542f70392ae026374600add5b887", size = 4968631, upload-time = "2025-06-21T13:39:12.283Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/c7/21/705964c7812476f378728bdf590ca4b771ec72385c533964653c68e86bdc/pygments-2.19.2-py3-none-any.whl", hash = "sha256:86540386c03d588bb81d44bc3928634ff26449851e99741617ecb9037ee5ec0b", size = 1225217, upload-time = "2025-06-21T13:39:07.939Z" },
+]
+
+[[package]]
+name = "pypdfium2"
+version = "5.3.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/18/83/173dab58beb6c7e772b838199014c173a2436018dd7cfde9bbf4a3be15da/pypdfium2-5.3.0.tar.gz", hash = "sha256:2873ffc95fcb01f329257ebc64a5fdce44b36447b6b171fe62f7db5dc3269885", size = 268742, upload-time = "2026-01-05T16:29:03.02Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/e3/a4/6bb5b5918c7fc236ec426be8a0205a984fe0a26ae23d5e4dd497398a6571/pypdfium2-5.3.0-py3-none-android_23_arm64_v8a.whl", hash = "sha256:885df6c78d41600cb086dc0c76b912d165b5bd6931ca08138329ea5a991b3540", size = 2763287, upload-time = "2026-01-05T16:28:24.21Z" },
+    { url = "https://files.pythonhosted.org/packages/3e/64/24b41b906006bf07099b095f0420ee1f01a3a83a899f3e3731e4da99c06a/pypdfium2-5.3.0-py3-none-android_23_armeabi_v7a.whl", hash = "sha256:6e53dee6b333ee77582499eff800300fb5aa0c7eb8f52f95ccb5ca35ebc86d48", size = 2303285, upload-time = "2026-01-05T16:28:26.274Z" },
+    { url = "https://files.pythonhosted.org/packages/c2/c0/3ec73f4ded83ba6c02acf6e9d228501759d5d74fe57f1b93849ab92dcc20/pypdfium2-5.3.0-py3-none-macosx_11_0_arm64.whl", hash = "sha256:ce4466bdd62119fe25a5f74d107acc9db8652062bf217057630c6ff0bb419523", size = 2816066, upload-time = "2026-01-05T16:28:28.099Z" },
+    { url = "https://files.pythonhosted.org/packages/62/ca/e553b3b8b5c2cdc3d955cc313493ac27bbe63fc22624769d56ded585dd5e/pypdfium2-5.3.0-py3-none-macosx_11_0_x86_64.whl", hash = "sha256:cc2647fd03db42b8a56a8835e8bc7899e604e2042cd6fedeea53483185612907", size = 2945545, upload-time = "2026-01-05T16:28:29.489Z" },
+    { url = "https://files.pythonhosted.org/packages/a1/56/615b776071e95c8570d579038256d0c77969ff2ff381e427be4ab8967f44/pypdfium2-5.3.0-py3-none-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:35e205f537ddb4069e4b4e22af7ffe84fcf2d686c3fee5e5349f73268a0ef1ca", size = 2979892, upload-time = "2026-01-05T16:28:31.088Z" },
+    { url = "https://files.pythonhosted.org/packages/df/10/27114199b765bdb7d19a9514c07036ad2fc3a579b910e7823ba167ead6de/pypdfium2-5.3.0-py3-none-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:b5795298f44050797ac030994fc2525ea35d2d714efe70058e0ee22e5f613f27", size = 2765738, upload-time = "2026-01-05T16:28:33.18Z" },
+    { url = "https://files.pythonhosted.org/packages/b4/d7/2a3afa35e6c205a4f6264c33b8d2f659707989f93c30b336aa58575f66fa/pypdfium2-5.3.0-py3-none-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:b7cd43dfceb77137e69e74c933d41506da1dddaff70f3a794fb0ad0d73e90d75", size = 3064338, upload-time = "2026-01-05T16:28:34.731Z" },
+    { url = "https://files.pythonhosted.org/packages/a2/f1/6658755cf6e369bb51d0bccb81c51c300404fbe67c2f894c90000b6442dd/pypdfium2-5.3.0-py3-none-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:d5956867558fd3a793e58691cf169718864610becb765bfe74dd83f05cbf1ae3", size = 3415059, upload-time = "2026-01-05T16:28:37.313Z" },
+    { url = "https://files.pythonhosted.org/packages/f5/34/f86482134fa641deb1f524c45ec7ebd6fc8d404df40c5657ddfce528593e/pypdfium2-5.3.0-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:b3ff1071e9a782625822658dfe6e29e3a644a66960f8713bb17819f5a0ac5987", size = 2998517, upload-time = "2026-01-05T16:28:38.873Z" },
+    { url = "https://files.pythonhosted.org/packages/09/34/40ab99425dcf503c172885904c5dc356c052bfdbd085f9f3cc920e0b8b25/pypdfium2-5.3.0-py3-none-manylinux_2_27_s390x.manylinux_2_28_s390x.whl", hash = "sha256:f319c46ead49d289ab8c1ed2ea63c91e684f35bdc4cf4dc52191c441182ac481", size = 3673154, upload-time = "2026-01-05T16:28:40.347Z" },
+    { url = "https://files.pythonhosted.org/packages/a5/67/0f7532f80825a7728a5cbff3f1104857f8f9fe49ebfd6cb25582a89ae8e1/pypdfium2-5.3.0-py3-none-manylinux_2_38_riscv64.manylinux_2_39_riscv64.whl", hash = "sha256:6dc67a186da0962294321cace6ccc0a4d212dbc5e9522c640d35725a812324b8", size = 2965002, upload-time = "2026-01-05T16:28:42.143Z" },
+    { url = "https://files.pythonhosted.org/packages/ce/6c/c03d2a3d6621b77aac9604bce1c060de2af94950448787298501eac6c6a2/pypdfium2-5.3.0-py3-none-musllinux_1_2_aarch64.whl", hash = "sha256:0ad0afd3d2b5b54d86287266fd6ae3fef0e0a1a3df9d2c4984b3e3f8f70e6330", size = 4130530, upload-time = "2026-01-05T16:28:44.264Z" },
+    { url = "https://files.pythonhosted.org/packages/af/39/9ad1f958cbe35d4693ae87c09ebafda4bb3e4709c7ccaec86c1a829163a3/pypdfium2-5.3.0-py3-none-musllinux_1_2_armv7l.whl", hash = "sha256:1afe35230dc3951b3e79b934c0c35a2e79e2372d06503fce6cf1926d2a816f47", size = 3746568, upload-time = "2026-01-05T16:28:45.897Z" },
+    { url = "https://files.pythonhosted.org/packages/2a/e2/4d32310166c2d6955d924737df8b0a3e3efc8d133344a98b10f96320157d/pypdfium2-5.3.0-py3-none-musllinux_1_2_i686.whl", hash = "sha256:00385793030cadce08469085cd21b168fd8ff981b009685fef3103bdc5fc4686", size = 4336683, upload-time = "2026-01-05T16:28:47.584Z" },
+    { url = "https://files.pythonhosted.org/packages/14/ea/38c337ff12a8cec4b00fd4fdb0a63a70597a344581e20b02addbd301ab56/pypdfium2-5.3.0-py3-none-musllinux_1_2_ppc64le.whl", hash = "sha256:d911e82676398949697fef80b7f412078df14d725a91c10e383b727051530285", size = 4375030, upload-time = "2026-01-05T16:28:49.5Z" },
+    { url = "https://files.pythonhosted.org/packages/a1/77/9d8de90c35d2fc383be8819bcde52f5821dacbd7404a0225e4010b99d080/pypdfium2-5.3.0-py3-none-musllinux_1_2_riscv64.whl", hash = "sha256:ca1dc625ed347fac3d9002a3ed33d521d5803409bd572e7b3f823c12ab2ef58f", size = 3928914, upload-time = "2026-01-05T16:28:51.433Z" },
+    { url = "https://files.pythonhosted.org/packages/a5/39/9d4a6fbd78fcb6803b0ea5e4952a31d6182a0aaa2609cfcd0eb88446fdb8/pypdfium2-5.3.0-py3-none-musllinux_1_2_s390x.whl", hash = "sha256:ea4f9db2d3575f22cd41f4c7a855240ded842f135e59a961b5b1351a65ce2b6e", size = 4997777, upload-time = "2026-01-05T16:28:53.589Z" },
+    { url = "https://files.pythonhosted.org/packages/9d/38/cdd4ed085c264234a59ad32df1dfe432c77a7403da2381e0fcc1ba60b74e/pypdfium2-5.3.0-py3-none-musllinux_1_2_x86_64.whl", hash = "sha256:0ea24409613df350223c6afc50911c99dca0d43ddaf2616c5a1ebdffa3e1bcb5", size = 4179895, upload-time = "2026-01-05T16:28:55.322Z" },
+    { url = "https://files.pythonhosted.org/packages/93/4c/d2f40145c9012482699664f615d7ae540a346c84f68a8179449e69dcc4d8/pypdfium2-5.3.0-py3-none-win32.whl", hash = "sha256:5bf695d603f9eb8fdd7c1786add5cf420d57fbc81df142ed63c029ce29614df9", size = 2993570, upload-time = "2026-01-05T16:28:58.37Z" },
+    { url = "https://files.pythonhosted.org/packages/2c/dc/1388ea650020c26ef3f68856b9227e7f153dcaf445e7e4674a0b8f26891e/pypdfium2-5.3.0-py3-none-win_amd64.whl", hash = "sha256:8365af22a39d4373c265f8e90e561cd64d4ddeaf5e6a66546a8caed216ab9574", size = 3102340, upload-time = "2026-01-05T16:28:59.933Z" },
+    { url = "https://files.pythonhosted.org/packages/c8/71/a433668d33999b3aeb2c2dda18aaf24948e862ea2ee148078a35daac6c1c/pypdfium2-5.3.0-py3-none-win_arm64.whl", hash = "sha256:0b2c6bf825e084d91d34456be54921da31e9199d9530b05435d69d1a80501a12", size = 2940987, upload-time = "2026-01-05T16:29:01.511Z" },
+]
+
+[[package]]
+name = "pytest"
+version = "8.4.2"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "colorama", marker = "sys_platform == 'win32'" },
+    { name = "exceptiongroup", marker = "python_full_version < '3.11'" },
+    { name = "iniconfig" },
+    { name = "packaging" },
+    { name = "pluggy" },
+    { name = "pygments" },
+    { name = "tomli", marker = "python_full_version < '3.11'" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/a3/5c/00a0e072241553e1a7496d638deababa67c5058571567b92a7eaa258397c/pytest-8.4.2.tar.gz", hash = "sha256:86c0d0b93306b961d58d62a4db4879f27fe25513d4b969df351abdddb3c30e01", size = 1519618, upload-time = "2025-09-04T14:34:22.711Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/a8/a4/20da314d277121d6534b3a980b29035dcd51e6744bd79075a6ce8fa4eb8d/pytest-8.4.2-py3-none-any.whl", hash = "sha256:872f880de3fc3a5bdc88a11b39c9710c3497a547cfa9320bc3c5e62fbf272e79", size = 365750, upload-time = "2025-09-04T14:34:20.226Z" },
+]
+
+[[package]]
+name = "pytest-asyncio"
+version = "0.23.8"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "pytest" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/de/b4/0b378b7bf26a8ae161c3890c0b48a91a04106c5713ce81b4b080ea2f4f18/pytest_asyncio-0.23.8.tar.gz", hash = "sha256:759b10b33a6dc61cce40a8bd5205e302978bbbcc00e279a8b61d9a6a3c82e4d3", size = 46920, upload-time = "2024-07-17T17:39:34.617Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/ee/82/62e2d63639ecb0fbe8a7ee59ef0bc69a4669ec50f6d3459f74ad4e4189a2/pytest_asyncio-0.23.8-py3-none-any.whl", hash = "sha256:50265d892689a5faefb84df80819d1ecef566eb3549cf915dfb33569359d1ce2", size = 17663, upload-time = "2024-07-17T17:39:32.478Z" },
+]
+
+[[package]]
+name = "pytest-benchmark"
+version = "5.2.3"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "py-cpuinfo" },
+    { name = "pytest" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/24/34/9f732b76456d64faffbef6232f1f9dbec7a7c4999ff46282fa418bd1af66/pytest_benchmark-5.2.3.tar.gz", hash = "sha256:deb7317998a23c650fd4ff76e1230066a76cb45dcece0aca5607143c619e7779", size = 341340, upload-time = "2025-11-09T18:48:43.215Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/33/29/e756e715a48959f1c0045342088d7ca9762a2f509b945f362a316e9412b7/pytest_benchmark-5.2.3-py3-none-any.whl", hash = "sha256:bc839726ad20e99aaa0d11a127445457b4219bdb9e80a1afc4b51da7f96b0803", size = 45255, upload-time = "2025-11-09T18:48:39.765Z" },
+]
+
+[[package]]
+name = "pytest-codspeed"
+version = "4.2.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "cffi" },
+    { name = "pytest" },
+    { name = "rich" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/e2/e8/27fcbe6516a1c956614a4b61a7fccbf3791ea0b992e07416e8948184327d/pytest_codspeed-4.2.0.tar.gz", hash = "sha256:04b5d0bc5a1851ba1504d46bf9d7dbb355222a69f2cd440d54295db721b331f7", size = 113263, upload-time = "2025-10-24T09:02:55.704Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/6c/b8/d599a466c50af3f04001877ae8b17c12b803f3b358235736b91a0769de0d/pytest_codspeed-4.2.0-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:609828b03972966b75b9b7416fa2570c4a0f6124f67e02d35cd3658e64312a7b", size = 261943, upload-time = "2025-10-24T09:02:37.962Z" },
+    { url = "https://files.pythonhosted.org/packages/74/19/ccc1a2fcd28357a8db08ba6b60f381832088a3850abc262c8e0b3406491a/pytest_codspeed-4.2.0-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:23a0c0fbf8bb4de93a3454fd9e5efcdca164c778aaef0a9da4f233d85cb7f5b8", size = 250782, upload-time = "2025-10-24T09:02:39.617Z" },
+    { url = "https://files.pythonhosted.org/packages/b9/2d/f0083a2f14ecf008d961d40439a71da0ae0d568e5f8dc2fccd3e8a2ab3e4/pytest_codspeed-4.2.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:2de87bde9fbc6fd53f0fd21dcf2599c89e0b8948d49f9bad224edce51c47e26b", size = 261960, upload-time = "2025-10-24T09:02:40.665Z" },
+    { url = "https://files.pythonhosted.org/packages/5f/0c/1f514c553db4ea5a69dfbe2706734129acd0eca8d5101ec16f1dd00dbc0f/pytest_codspeed-4.2.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:95aeb2479ca383f6b18e2cc9ebcd3b03ab184980a59a232aea6f370bbf59a1e3", size = 250808, upload-time = "2025-10-24T09:02:42.07Z" },
+    { url = "https://files.pythonhosted.org/packages/81/04/479905bd6653bc981c0554fcce6df52d7ae1594e1eefd53e6cf31810ec7f/pytest_codspeed-4.2.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:7d4fefbd4ae401e2c60f6be920a0be50eef0c3e4a1f0a1c83962efd45be38b39", size = 262084, upload-time = "2025-10-24T09:02:43.155Z" },
+    { url = "https://files.pythonhosted.org/packages/d2/46/d6f345d7907bac6cbb6224bd697ecbc11cf7427acc9e843c3618f19e3476/pytest_codspeed-4.2.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:309b4227f57fcbb9df21e889ea1ae191d0d1cd8b903b698fdb9ea0461dbf1dfe", size = 251100, upload-time = "2025-10-24T09:02:44.168Z" },
+    { url = "https://files.pythonhosted.org/packages/de/dc/e864f45e994a50390ff49792256f1bdcbf42f170e3bc0470ee1a7d2403f3/pytest_codspeed-4.2.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:72aab8278452a6d020798b9e4f82780966adb00f80d27a25d1274272c54630d5", size = 262057, upload-time = "2025-10-24T09:02:45.791Z" },
+    { url = "https://files.pythonhosted.org/packages/1d/1c/f1d2599784486879cf6579d8d94a3e22108f0e1f130033dab8feefd29249/pytest_codspeed-4.2.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:684fcd9491d810ded653a8d38de4835daa2d001645f4a23942862950664273f8", size = 251013, upload-time = "2025-10-24T09:02:46.937Z" },
+    { url = "https://files.pythonhosted.org/packages/0c/fd/eafd24db5652a94b4d00fe9b309b607de81add0f55f073afb68a378a24b6/pytest_codspeed-4.2.0-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:50794dabea6ec90d4288904452051e2febace93e7edf4ca9f2bce8019dd8cd37", size = 262065, upload-time = "2025-10-24T09:02:48.018Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/14/8d9340d7dc0ae647991b28a396e16b3403e10def883cde90d6b663d3f7ec/pytest_codspeed-4.2.0-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:a0ebd87f2a99467a1cfd8e83492c4712976e43d353ee0b5f71cbb057f1393aca", size = 251057, upload-time = "2025-10-24T09:02:49.102Z" },
+    { url = "https://files.pythonhosted.org/packages/4b/39/48cf6afbca55bc7c8c93c3d4ae926a1068bcce3f0241709db19b078d5418/pytest_codspeed-4.2.0-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:dbbb2d61b85bef8fc7e2193f723f9ac2db388a48259d981bbce96319043e9830", size = 267983, upload-time = "2025-10-24T09:02:50.558Z" },
+    { url = "https://files.pythonhosted.org/packages/33/86/4407341efb5dceb3e389635749ce1d670542d6ca148bd34f9d5334295faf/pytest_codspeed-4.2.0-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:748411c832147bfc85f805af78a1ab1684f52d08e14aabe22932bbe46c079a5f", size = 256732, upload-time = "2025-10-24T09:02:51.603Z" },
+    { url = "https://files.pythonhosted.org/packages/25/0e/8cb71fd3ed4ed08c07aec1245aea7bc1b661ba55fd9c392db76f1978d453/pytest_codspeed-4.2.0-py3-none-any.whl", hash = "sha256:e81bbb45c130874ef99aca97929d72682733527a49f84239ba575b5cb843bab0", size = 113726, upload-time = "2025-10-24T09:02:54.785Z" },
+]
+
+[[package]]
+name = "pytest-mock"
+version = "3.15.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "pytest" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/68/14/eb014d26be205d38ad5ad20d9a80f7d201472e08167f0bb4361e251084a9/pytest_mock-3.15.1.tar.gz", hash = "sha256:1849a238f6f396da19762269de72cb1814ab44416fa73a8686deac10b0d87a0f", size = 34036, upload-time = "2025-09-16T16:37:27.081Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/5a/cc/06253936f4a7fa2e0f48dfe6d851d9c56df896a9ab09ac019d70b760619c/pytest_mock-3.15.1-py3-none-any.whl", hash = "sha256:0a25e2eb88fe5168d535041d09a4529a188176ae608a6d249ee65abc0949630d", size = 10095, upload-time = "2025-09-16T16:37:25.734Z" },
+]
+
+[[package]]
+name = "pytest-recording"
+version = "0.13.4"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "pytest" },
+    { name = "vcrpy" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/32/9c/f4027c5f1693847b06d11caf4b4f6bb09f22c1581ada4663877ec166b8c6/pytest_recording-0.13.4.tar.gz", hash = "sha256:568d64b2a85992eec4ae0a419c855d5fd96782c5fb016784d86f18053792768c", size = 26576, upload-time = "2025-05-08T10:41:11.231Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/42/c2/ce34735972cc42d912173e79f200fe66530225190c06655c5632a9d88f1e/pytest_recording-0.13.4-py3-none-any.whl", hash = "sha256:ad49a434b51b1c4f78e85b1e6b74fdcc2a0a581ca16e52c798c6ace971f7f439", size = 13723, upload-time = "2025-05-08T10:41:09.684Z" },
+]
+
+[[package]]
+name = "pytest-retry"
+version = "1.7.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "pytest" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/c5/5b/607b017994cca28de3a1ad22a3eee8418e5d428dcd8ec25b26b18e995a73/pytest_retry-1.7.0.tar.gz", hash = "sha256:f8d52339f01e949df47c11ba9ee8d5b362f5824dff580d3870ec9ae0057df80f", size = 19977, upload-time = "2025-01-19T01:56:13.115Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/7c/ff/3266c8a73b9b93c4b14160a7e2b31d1e1088e28ed29f4c2d93ae34093bfd/pytest_retry-1.7.0-py3-none-any.whl", hash = "sha256:a2dac85b79a4e2375943f1429479c65beb6c69553e7dae6b8332be47a60954f4", size = 13775, upload-time = "2025-01-19T01:56:11.199Z" },
+]
+
+[[package]]
+name = "pytest-socket"
+version = "0.7.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "pytest" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/05/ff/90c7e1e746baf3d62ce864c479fd53410b534818b9437413903596f81580/pytest_socket-0.7.0.tar.gz", hash = "sha256:71ab048cbbcb085c15a4423b73b619a8b35d6a307f46f78ea46be51b1b7e11b3", size = 12389, upload-time = "2024-01-28T20:17:23.177Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/19/58/5d14cb5cb59409e491ebe816c47bf81423cd03098ea92281336320ae5681/pytest_socket-0.7.0-py3-none-any.whl", hash = "sha256:7e0f4642177d55d317bbd58fc68c6bd9048d6eadb2d46a89307fa9221336ce45", size = 6754, upload-time = "2024-01-28T20:17:22.105Z" },
+]
+
+[[package]]
+name = "pytest-watcher"
+version = "0.6.3"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "tomli", marker = "python_full_version < '3.11'" },
+    { name = "watchdog" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/e6/d2/80606077b7fa8784417687f494ff801d7ab817d9a17fc94305811d5919bb/pytest_watcher-0.6.3.tar.gz", hash = "sha256:842dc904264df0ad2d5264153a66bb452fccfa46598cd6e0a5ef1d19afed9b13", size = 601878, upload-time = "2026-01-10T23:28:18.805Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/fc/3f/172d73600ad2771774cda108efb813fc724fc345e5240a81a1085f1ade5d/pytest_watcher-0.6.3-py3-none-any.whl", hash = "sha256:83e7748c933087e8276edb6078663e6afa9926434b4fd8b85cf6b32b1d5bec89", size = 12431, upload-time = "2026-01-10T23:28:17.64Z" },
+]
+
+[[package]]
+name = "python-bidi"
+version = "0.6.7"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/ed/e3/c0c8bf6fca79ac946a28d57f116e3b9e5b10a4469b6f70bf73f3744c49bf/python_bidi-0.6.7.tar.gz", hash = "sha256:c10065081c0e137975de5d9ba2ff2306286dbf5e0c586d4d5aec87c856239b41", size = 45503, upload-time = "2025-10-22T09:52:49.624Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/5d/c3/cdbece686fab47d4d04f2c15d372b3d3f3308da2e535657bf4bbd5afef50/python_bidi-0.6.7-cp310-cp310-macosx_10_12_x86_64.whl", hash = "sha256:94dbfd6a6ec0ae64b5262290bf014d6063f9ac8688bda9ec668dc175378d2c80", size = 274857, upload-time = "2025-10-22T09:51:57.298Z" },
+    { url = "https://files.pythonhosted.org/packages/aa/19/1cd52f04345717613eafe8b23dd1ce8799116f7cc54b23aaefa27db298d6/python_bidi-0.6.7-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:d8274ff02d447cca026ba00f56070ba15f95e184b2d028ee0e4b6c9813d2aaf9", size = 264682, upload-time = "2025-10-22T09:51:48.203Z" },
+    { url = "https://files.pythonhosted.org/packages/c7/39/f46dae8bd298ffecaf169ea8871c1e63c6116e1b0178ca4eab2cb99d1c13/python_bidi-0.6.7-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:24afff65c581a5d6f658a9ec027d6719d19a1d8a4401000fdb22d2eeb677b8e3", size = 293680, upload-time = "2025-10-22T09:50:57.091Z" },
+    { url = "https://files.pythonhosted.org/packages/96/ed/c4e2c684bf8f226de4d0070780073fc7f3f97def3ad06f11b4c021bfa965/python_bidi-0.6.7-cp310-cp310-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:8678c2272e7bd60a75f781409e900c9ddb9f01f55c625d83ae0d49dfc6a2674f", size = 302625, upload-time = "2025-10-22T09:51:05.378Z" },
+    { url = "https://files.pythonhosted.org/packages/83/fa/3b5be9187515a4c28ad358c2f2785f968d4de090389f08a11c826ae1c17f/python_bidi-0.6.7-cp310-cp310-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:d4cd82e65b5aeb31bd73534e61ece1cab625f4bcbdc13bc4ddc5f8cbfb37c24a", size = 441183, upload-time = "2025-10-22T09:51:14.014Z" },
+    { url = "https://files.pythonhosted.org/packages/d7/c7/023028ca45e674b67abee29a049fb3b7aac74873181940a1d34ad27e23cd/python_bidi-0.6.7-cp310-cp310-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:dde1c3f3edb1f0095dcbf79cf8a0bb768f9539e809d0ad010d78200eea97d42a", size = 326788, upload-time = "2025-10-22T09:51:22.58Z" },
+    { url = "https://files.pythonhosted.org/packages/d3/30/0753601fdad405e806c89cfa9603ff75241f8c7196cfe2cb37c43e34cdbd/python_bidi-0.6.7-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:9c463ae15e94b1c6a8a50bd671d6166b0b0d779fd1e56cbf46d8a4a84c9aa2d0", size = 302036, upload-time = "2025-10-22T09:51:40.341Z" },
+    { url = "https://files.pythonhosted.org/packages/c6/38/e83901206c7161e4fa14f52d1244eb54bad2b9a959be62af7b472cded20a/python_bidi-0.6.7-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:6f9fa1257e075eeeed67d21f95e411036b7ca2b5c78f757d4ac66485c191720a", size = 315484, upload-time = "2025-10-22T09:51:32.285Z" },
+    { url = "https://files.pythonhosted.org/packages/98/89/cd73185ad92990261b050a30753a693ad22a72ad5dc61b4e3845c58eff75/python_bidi-0.6.7-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:9adeec7cab0f2c2c291bd7faf9fa3fa233365fd0bf1c1c27a6ddd6cc563d4b32", size = 474003, upload-time = "2025-10-22T09:52:06.535Z" },
+    { url = "https://files.pythonhosted.org/packages/9f/38/03fd74c68cae08d08a32a4bc2031300a882a7ceab39b7e7fc5a5e37f5b7c/python_bidi-0.6.7-cp310-cp310-musllinux_1_2_armv7l.whl", hash = "sha256:3b96744e4709f4445788a3645cea7ef8d7520ccd4fa8bbbfb3b650702e12c1e6", size = 567114, upload-time = "2025-10-22T09:52:17.534Z" },
+    { url = "https://files.pythonhosted.org/packages/98/44/e196002ba8317d48ebab4750092a61287574195a3f685232059aa776edf4/python_bidi-0.6.7-cp310-cp310-musllinux_1_2_i686.whl", hash = "sha256:8860d67dc04dc530b8b4f588f38b7341a76f2ec44a45685a2d54e9dcffa5d15a", size = 493810, upload-time = "2025-10-22T09:52:28.683Z" },
+    { url = "https://files.pythonhosted.org/packages/e8/e2/1d495515d3fea0ecdd8bbb50e573282826ba074bceb2c0430206f94cde68/python_bidi-0.6.7-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:a4319f478ab1b90bbbe9921606ecb7baa0ebf0b332e821d41c3abdf1a30f0c35", size = 465208, upload-time = "2025-10-22T09:52:39.411Z" },
+    { url = "https://files.pythonhosted.org/packages/89/c7/fc5b25d017677793435c415c7884f9c60ce7705bd35565280cca3be69fa9/python_bidi-0.6.7-cp310-cp310-win32.whl", hash = "sha256:8d4e621caadfdbc73d36eabdb2f392da850d28c58b020738411d09dda6208509", size = 157426, upload-time = "2025-10-22T09:52:58.114Z" },
+    { url = "https://files.pythonhosted.org/packages/85/be/bd323950b98d40ab45f97630c3bfb5ed3a7416b2f71c250bcc1ed1267eb0/python_bidi-0.6.7-cp310-cp310-win_amd64.whl", hash = "sha256:fd87d112eda1f0528074e1f7c0312881816cb75854133021124269a27c6c48dc", size = 161038, upload-time = "2025-10-22T09:52:50.44Z" },
+    { url = "https://files.pythonhosted.org/packages/ec/de/c30a13ad95239507af472a5fc2cadd2e5e172055068f12ac39b37922c7f8/python_bidi-0.6.7-cp311-cp311-macosx_10_12_x86_64.whl", hash = "sha256:a8892a7da0f617135fe9c92dc7070d13a0f96ab3081f9db7ff5b172a3905bd78", size = 274420, upload-time = "2025-10-22T09:51:58.262Z" },
+    { url = "https://files.pythonhosted.org/packages/ad/9f/be5efef7eea5f1e2a6415c4052a988f594dcf5a11a15103f2718d324a35b/python_bidi-0.6.7-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:06650a164e63e94dc8a291cc9d415b4027cb1cce125bc9b02dac0f34d535ed47", size = 264586, upload-time = "2025-10-22T09:51:49.255Z" },
+    { url = "https://files.pythonhosted.org/packages/87/ec/2c374b6de35870817ffb3512c0666ea8c3794ef923b5586c69451e0e5395/python_bidi-0.6.7-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:6df7be07af867ec1d121c92ea827efad4d77b25457c06eeab477b601e82b2340", size = 293672, upload-time = "2025-10-22T09:50:58.504Z" },
+    { url = "https://files.pythonhosted.org/packages/29/1a/722d7d7128bdc9a530351a0d2fdf2ff5f4af66a865a6bca925f99832e2cc/python_bidi-0.6.7-cp311-cp311-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:73a88dc333efc42281bd800d5182c8625c6e11d109fc183fe3d7a11d48ab1150", size = 302643, upload-time = "2025-10-22T09:51:06.419Z" },
+    { url = "https://files.pythonhosted.org/packages/24/d7/5b9b593dd58fc745233d8476e9f4e0edd437547c78c58340619868470349/python_bidi-0.6.7-cp311-cp311-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:f24189dc3aea3a0a94391a047076e1014306b39ba17d7a38ebab510553cd1a97", size = 441692, upload-time = "2025-10-22T09:51:15.39Z" },
+    { url = "https://files.pythonhosted.org/packages/08/b9/16e7a1db5f022da6654e89875d231ec2e044d42ef7b635feeff61cee564c/python_bidi-0.6.7-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:a507fe6928a27a308e04ebf2065719b7850d1bf9ff1924f4e601ef77758812bd", size = 326933, upload-time = "2025-10-22T09:51:23.631Z" },
+    { url = "https://files.pythonhosted.org/packages/e0/a6/45aaec301292c6a07a9cc3168f5d1a92c8adc2ef36a3cd1f227b9caa980c/python_bidi-0.6.7-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:fbbffb948a32f9783d1a28bc0c53616f0a76736ed1e7c1d62e3e99a8dfaab869", size = 302034, upload-time = "2025-10-22T09:51:41.347Z" },
+    { url = "https://files.pythonhosted.org/packages/71/a3/7e42cce6e153c21b4e5cc96d429a5910909823f6fedd174b64ff67bc76a7/python_bidi-0.6.7-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:f7e507e1e798ebca77ddc9774fd405107833315ad802cfdaa1ab07b6d9154fc8", size = 315738, upload-time = "2025-10-22T09:51:33.409Z" },
+    { url = "https://files.pythonhosted.org/packages/43/7c/a5e4c0acc8e6ca61953b4add0576f0483f63b809b5389154e5da13927b0b/python_bidi-0.6.7-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:849a57d39feaf897955d0b19bbf4796bea53d1bcdf83b82e0a7b059167eb2049", size = 473968, upload-time = "2025-10-22T09:52:07.624Z" },
+    { url = "https://files.pythonhosted.org/packages/b1/aa/a18bc3cbab7a0e598cbe7b89f2c0913aedcc66dcafce9a4c357465c87859/python_bidi-0.6.7-cp311-cp311-musllinux_1_2_armv7l.whl", hash = "sha256:5ebc19f24e65a1f5c472e26d88e78b9d316e293bc6f205f32de4c4e99276336e", size = 567038, upload-time = "2025-10-22T09:52:18.594Z" },
+    { url = "https://files.pythonhosted.org/packages/92/46/fc6c54a8b5bfbee50e650f885ddef4f8c4f92880467ea0bc2bf133747048/python_bidi-0.6.7-cp311-cp311-musllinux_1_2_i686.whl", hash = "sha256:24388c77cb00b8aa0f9c84beb7e3e523a3dac4f786ece64a1d8175a07b24da72", size = 493970, upload-time = "2025-10-22T09:52:29.815Z" },
+    { url = "https://files.pythonhosted.org/packages/e3/f1/2c15f5b938b2e087e4e950cc14dcead5bedbaabfc6c576dac15739bc0c91/python_bidi-0.6.7-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:19737d217088ef27014f98eac1827c5913e6fb1dea96332ed84ede61791070d9", size = 465161, upload-time = "2025-10-22T09:52:40.517Z" },
+    { url = "https://files.pythonhosted.org/packages/56/d7/73a70a1fb819152485521b8dfe627e14ba9d3d5a65213244ab099adf3600/python_bidi-0.6.7-cp311-cp311-win32.whl", hash = "sha256:95c9de7ebc55ffb777548f2ecaf4b96b0fa0c92f42bf4d897b9f4cd164ec7394", size = 157033, upload-time = "2025-10-22T09:52:59.228Z" },
+    { url = "https://files.pythonhosted.org/packages/68/84/06999dc54ea047fe33209af7150df4202ab7ad52deeb66b2c2040ac07884/python_bidi-0.6.7-cp311-cp311-win_amd64.whl", hash = "sha256:898db0ea3e4aaa95b7fecba02a7560dfbf368f9d85053f2875f6d610c4d4ec2c", size = 161282, upload-time = "2025-10-22T09:52:51.467Z" },
+    { url = "https://files.pythonhosted.org/packages/e5/03/5b2f3e73501d0f41ebc2b075b49473047c6cdfc3465cf890263fc69e3915/python_bidi-0.6.7-cp312-cp312-macosx_10_12_x86_64.whl", hash = "sha256:11c51579e01f768446a7e13a0059fea1530936a707abcbeaad9467a55cb16073", size = 272536, upload-time = "2025-10-22T09:51:59.721Z" },
+    { url = "https://files.pythonhosted.org/packages/31/77/c6048e938a73e5a7c6fa3d5e3627a5961109daa728c2e7d050567cecdc26/python_bidi-0.6.7-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:47deaada8949af3a790f2cd73b613f9bfa153b4c9450f91c44a60c3109a81f73", size = 263258, upload-time = "2025-10-22T09:51:50.328Z" },
+    { url = "https://files.pythonhosted.org/packages/57/56/ed4dc501cab7de70ce35cd435c86278e4eb1caf238c80bc72297767c9219/python_bidi-0.6.7-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:b38ddfab41d10e780edb431edc30aec89bee4ce43d718e3896e99f33dae5c1d3", size = 292700, upload-time = "2025-10-22T09:50:59.628Z" },
+    { url = "https://files.pythonhosted.org/packages/77/6a/1bf06d7544c940ffddd97cd0e02c55348a92163c5495fa18e34217dfbebe/python_bidi-0.6.7-cp312-cp312-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:2a93b0394cc684d64356b0475858c116f1e335ffbaba388db93bf47307deadfa", size = 300881, upload-time = "2025-10-22T09:51:07.507Z" },
+    { url = "https://files.pythonhosted.org/packages/22/1d/ce7577a8f50291c06e94f651ac5de0d1678fc2642af26a5dad9901a0244f/python_bidi-0.6.7-cp312-cp312-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:ec1694134961b71ac05241ac989b49ccf08e232b5834d5fc46f8a7c3bb1c13a9", size = 439125, upload-time = "2025-10-22T09:51:16.559Z" },
+    { url = "https://files.pythonhosted.org/packages/a3/87/4cf6dcd58e22f0fd904e7a161c6b73a5f9d17d4d49073fcb089ba62f1469/python_bidi-0.6.7-cp312-cp312-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:8047c33b85f7790474a1f488bef95689f049976a4e1c6f213a8d075d180a93e4", size = 325816, upload-time = "2025-10-22T09:51:25.12Z" },
+    { url = "https://files.pythonhosted.org/packages/2a/0a/4028a088e29ce8f1673e85ec9f64204fc368355c3207e6a71619c2b4579a/python_bidi-0.6.7-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:9d9de35eb5987da27dd81e371c52142dd8e924bd61c1006003071ea05a735587", size = 300550, upload-time = "2025-10-22T09:51:42.739Z" },
+    { url = "https://files.pythonhosted.org/packages/1f/05/cac15eba462d5a2407ac4ef1c792c45a948652b00c6bd81eaab3834a62d2/python_bidi-0.6.7-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:a99d898ad1a399d9c8cab5561b3667fd24f4385820ac90c3340aa637aa5adfc9", size = 313017, upload-time = "2025-10-22T09:51:34.905Z" },
+    { url = "https://files.pythonhosted.org/packages/4b/b1/3ba91b9ea60fa54a9aa730a5fe432bd73095d55be371244584fc6818eae1/python_bidi-0.6.7-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:5debaab33562fdfc79ffdbd8d9c51cf07b8529de0e889d8cd145d78137aab21e", size = 472798, upload-time = "2025-10-22T09:52:09.079Z" },
+    { url = "https://files.pythonhosted.org/packages/50/40/4bf5fb7255e35c218174f322a4d4c80b63b2604d73adc6e32f843e700824/python_bidi-0.6.7-cp312-cp312-musllinux_1_2_armv7l.whl", hash = "sha256:c11c62a3cdb9d1426b1536de9e3446cb09c7d025bd4df125275cae221f214899", size = 565234, upload-time = "2025-10-22T09:52:19.703Z" },
+    { url = "https://files.pythonhosted.org/packages/bd/81/ad23fb85bff69d0a25729cd3834254b87c3c7caa93d657c8f8edcbed08f6/python_bidi-0.6.7-cp312-cp312-musllinux_1_2_i686.whl", hash = "sha256:6c051f2d28ca542092d01da8b5fe110fb6191ff58d298a54a93dc183bece63bf", size = 491844, upload-time = "2025-10-22T09:52:31.216Z" },
+    { url = "https://files.pythonhosted.org/packages/65/85/103baaf142b2838f583b71904a2454fa31bd2a912ff505c25874f45d6c3e/python_bidi-0.6.7-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:95867a07c5dee0ea2340fe1d0e4f6d9f5c5687d473193b6ee6f86fa44aac45d1", size = 463753, upload-time = "2025-10-22T09:52:41.943Z" },
+    { url = "https://files.pythonhosted.org/packages/54/c3/6a5c3b9f42a6b188430c83a7e70a76bc7c0db3354302fce7c8ed94a0c062/python_bidi-0.6.7-cp312-cp312-win32.whl", hash = "sha256:4c73cd980d45bb967799c7f0fc98ea93ae3d65b21ef2ba6abef6a057720bf483", size = 155820, upload-time = "2025-10-22T09:53:00.254Z" },
+    { url = "https://files.pythonhosted.org/packages/45/c4/683216398ee3abf6b9bb0f26ae15c696fabbe36468ba26d5271f0c11b343/python_bidi-0.6.7-cp312-cp312-win_amd64.whl", hash = "sha256:d524a4ba765bae9b950706472a77a887a525ed21144fe4b41f6190f6e57caa2c", size = 159966, upload-time = "2025-10-22T09:52:52.547Z" },
+    { url = "https://files.pythonhosted.org/packages/25/a5/8ad0a448d42fd5d01dd127c1dc5ab974a8ea6e20305ac89a3356dacd3bdf/python_bidi-0.6.7-cp313-cp313-macosx_10_12_x86_64.whl", hash = "sha256:1c061207212cd1db27bf6140b96dcd0536246f1e13e99bb5d03f4632f8e2ad7f", size = 272129, upload-time = "2025-10-22T09:52:00.761Z" },
+    { url = "https://files.pythonhosted.org/packages/e6/c0/a13981fc0427a0d35e96fc4e31fbb0f981b28d0ce08416f98f42d51ea3bc/python_bidi-0.6.7-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:a2eb8fca918c7381531035c3aae31c29a1c1300ab8a63cad1ec3a71331096c78", size = 263174, upload-time = "2025-10-22T09:51:51.401Z" },
+    { url = "https://files.pythonhosted.org/packages/9c/32/74034239d0bca32c315cac5c3ec07ef8eb44fa0e8cea1585cad85f5b8651/python_bidi-0.6.7-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:414004fe9cba33d288ff4a04e1c9afe6a737f440595d01b5bbed00d750296bbd", size = 292496, upload-time = "2025-10-22T09:51:00.708Z" },
+    { url = "https://files.pythonhosted.org/packages/83/fa/d6c853ed2668b1c12d66e71d4f843d0710d1ccaecc17ce09b35d2b1382a7/python_bidi-0.6.7-cp313-cp313-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:5013ba963e9da606c4c03958cc737ebd5f8b9b8404bd71ab0d580048c746f875", size = 300727, upload-time = "2025-10-22T09:51:09.152Z" },
+    { url = "https://files.pythonhosted.org/packages/9c/8d/55685bddfc1fbfa6e28e1c0be7df4023e504de7d2ac1355a3fa610836bc1/python_bidi-0.6.7-cp313-cp313-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:ad5f0847da00687f52d2b81828e8d887bdea9eb8686a9841024ea7a0e153028e", size = 438823, upload-time = "2025-10-22T09:51:17.844Z" },
+    { url = "https://files.pythonhosted.org/packages/9f/54/db9e70443f89e3ec6fa70dcd16809c3656d1efe7946076dcd59832f722df/python_bidi-0.6.7-cp313-cp313-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:26a8fe0d532b966708fc5f8aea0602107fde4745a8a5ae961edd3cf02e807d07", size = 325721, upload-time = "2025-10-22T09:51:26.132Z" },
+    { url = "https://files.pythonhosted.org/packages/55/c5/98ac9c00f17240f9114c756791f0cd9ba59a5d4b5d84fd1a6d0d50604e82/python_bidi-0.6.7-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:6323e943c7672b271ad9575a2232508f17e87e81a78d7d10d6e93040e210eddf", size = 300493, upload-time = "2025-10-22T09:51:43.783Z" },
+    { url = "https://files.pythonhosted.org/packages/0b/cb/382538dd7c656eb50408802b9a9466dbd3432bea059410e65a6c14bc79f9/python_bidi-0.6.7-cp313-cp313-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:349b89c3110bd25aa56d79418239ca4785d4bcc7a596e63bb996a9696fc6a907", size = 312889, upload-time = "2025-10-22T09:51:36.011Z" },
+    { url = "https://files.pythonhosted.org/packages/50/8d/dbc784cecd9b2950ba99c8fef0387ae588837e4e2bfd543be191d18bf9f6/python_bidi-0.6.7-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:e7cad66317f12f0fd755fe41ee7c6b06531d2189a9048a8f37addb5109f7e3e3", size = 472798, upload-time = "2025-10-22T09:52:10.446Z" },
+    { url = "https://files.pythonhosted.org/packages/83/e6/398d59075265717d2950622ede1d366aff88ffcaa67a30b85709dea72206/python_bidi-0.6.7-cp313-cp313-musllinux_1_2_armv7l.whl", hash = "sha256:49639743f1230648fd4fb47547f8a48ada9c5ca1426b17ac08e3be607c65394c", size = 564974, upload-time = "2025-10-22T09:52:22.416Z" },
+    { url = "https://files.pythonhosted.org/packages/7c/8e/2b939be0651bc2b69c234dc700723a26b93611d5bdd06b253d67d9da3557/python_bidi-0.6.7-cp313-cp313-musllinux_1_2_i686.whl", hash = "sha256:4636d572b357ab9f313c5340915c1cf51e3e54dd069351e02b6b76577fd1a854", size = 491711, upload-time = "2025-10-22T09:52:32.322Z" },
+    { url = "https://files.pythonhosted.org/packages/8f/05/f53739ab2ce2eee0c855479a31b64933f6ff6164f3ddc611d04e4b79d922/python_bidi-0.6.7-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:d7310312a68fdb1a8249cf114acb5435aa6b6a958b15810f053c1df5f98476e4", size = 463536, upload-time = "2025-10-22T09:52:43.142Z" },
+    { url = "https://files.pythonhosted.org/packages/77/c6/800899e2764f723c2ea9172eabcc1a31ffb8b4bb71ea5869158fd83bd437/python_bidi-0.6.7-cp313-cp313-win32.whl", hash = "sha256:ec985386bc3cd54155f2ef0434fccbfd743617ed6fc1a84dae2ab1de6062e0c6", size = 155786, upload-time = "2025-10-22T09:53:01.357Z" },
+    { url = "https://files.pythonhosted.org/packages/30/ba/a811c12c1a4b8fa7c0c0963d92c042284c2049b1586615af6b1774b786d9/python_bidi-0.6.7-cp313-cp313-win_amd64.whl", hash = "sha256:f57726b5a90d818625e6996f5116971b7a4ceb888832337d0e2cf43d1c362a90", size = 159863, upload-time = "2025-10-22T09:52:53.537Z" },
+    { url = "https://files.pythonhosted.org/packages/6f/a5/cda302126e878be162bf183eb0bd6dc47ca3e680fb52111e49c62a8ea1eb/python_bidi-0.6.7-cp314-cp314-macosx_10_12_x86_64.whl", hash = "sha256:b0bee27fb596a0f518369c275a965d0448c39a0730e53a030b311bb10562d4d5", size = 271899, upload-time = "2025-10-22T09:52:01.758Z" },
+    { url = "https://files.pythonhosted.org/packages/4d/4b/9c15ca0fe795a5c55a39daa391524ac74e26d9187493632d455257771023/python_bidi-0.6.7-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:6c19ab378fefb1f09623f583fcfa12ed42369a998ddfbd39c40908397243c56b", size = 262235, upload-time = "2025-10-22T09:51:52.379Z" },
+    { url = "https://files.pythonhosted.org/packages/0f/5e/25b25be64bff05272aa28d8bef2fbbad8415db3159a41703eb2e63dc9824/python_bidi-0.6.7-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:630cee960ba9e3016f95a8e6f725a621ddeff6fd287839f5693ccfab3f3a9b5c", size = 471983, upload-time = "2025-10-22T09:52:12.182Z" },
+    { url = "https://files.pythonhosted.org/packages/4d/78/a9363f5da1b10d9211514b96ea47ecc95c797ed5ac566684bfece0666082/python_bidi-0.6.7-cp314-cp314-musllinux_1_2_armv7l.whl", hash = "sha256:0dbb4bbae212cca5bcf6e522fe8f572aff7d62544557734c2f810ded844d9eea", size = 565016, upload-time = "2025-10-22T09:52:23.515Z" },
+    { url = "https://files.pythonhosted.org/packages/0d/ed/37dcb7d3dc250ecdff8120b026c37fcdbeada4111e4d7148c053180bcf54/python_bidi-0.6.7-cp314-cp314-musllinux_1_2_i686.whl", hash = "sha256:1dd0a5ec0d8710905cebb4c9e5018aa8464395a33cb32a3a6c2a951bf1984fe5", size = 491180, upload-time = "2025-10-22T09:52:33.505Z" },
+    { url = "https://files.pythonhosted.org/packages/40/a3/50d1f6060a7a500768768f5f8735cb68deba36391248dbf13d5d2c9c0885/python_bidi-0.6.7-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:4ea928c31c7364098f853f122868f6f2155d6840661f7ea8b2ccfdf6084eb9f4", size = 463126, upload-time = "2025-10-22T09:52:44.28Z" },
+    { url = "https://files.pythonhosted.org/packages/d2/47/712cd7d1068795c57fdf6c4acca00716688aa8b4e353b30de2ed8f599fd6/python_bidi-0.6.7-cp314-cp314-win32.whl", hash = "sha256:f7c055a50d068b3a924bd33a327646346839f55bcb762a26ec3fde8ea5d40564", size = 155793, upload-time = "2025-10-22T09:53:02.7Z" },
+    { url = "https://files.pythonhosted.org/packages/c3/e8/1f86bf699b20220578351f9b7b635ed8b6e84dd51ad3cca08b89513ae971/python_bidi-0.6.7-cp314-cp314-win_amd64.whl", hash = "sha256:8a17631e3e691eec4ae6a370f7b035cf0a5767f4457bd615d11728c23df72e43", size = 159821, upload-time = "2025-10-22T09:52:54.95Z" },
+    { url = "https://files.pythonhosted.org/packages/b8/4e/6135798d84b62eea70c0f9435301c2a4ba854e87be93a3fcd1d935266d24/python_bidi-0.6.7-pp311-pypy311_pp73-macosx_10_12_x86_64.whl", hash = "sha256:c9a679b24f5c6f366a0dec75745e1abeae2f597f033d0d54c74cbe62e7e6ae28", size = 276275, upload-time = "2025-10-22T09:52:05.078Z" },
+    { url = "https://files.pythonhosted.org/packages/74/83/2123596d43e552af9e2806e361646fa579f34a1d1e9e2c1707a0ab6a02dd/python_bidi-0.6.7-pp311-pypy311_pp73-macosx_11_0_arm64.whl", hash = "sha256:05fe5971110013610f0db40505d0b204edc756e92eafac1372a464f8b9162b11", size = 266951, upload-time = "2025-10-22T09:51:56.216Z" },
+    { url = "https://files.pythonhosted.org/packages/5c/8c/8d1e1501717227a6d52fc7b9c47a3de61486b024fbdd4821bfad724c0699/python_bidi-0.6.7-pp311-pypy311_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:17572944e6d8fb616d111fc702c759da2bf7cedab85a3e4fa2af0c9eb95ed438", size = 295745, upload-time = "2025-10-22T09:51:04.438Z" },
+    { url = "https://files.pythonhosted.org/packages/fd/ff/ef04e7f9067c2c5d862b9f8d9a192486c500c8aa295f0fb756c25ab47fc8/python_bidi-0.6.7-pp311-pypy311_pp73-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:3b63d19f3f56ff7f99bce5ca9ef8c811dbf0f509d8e84c1bc06105ed26a49528", size = 304123, upload-time = "2025-10-22T09:51:12.559Z" },
+    { url = "https://files.pythonhosted.org/packages/be/72/b973895e257a7d4cc8365ab094612f6ee885df863a4964d8865b9f534b67/python_bidi-0.6.7-pp311-pypy311_pp73-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:f1350033431d75be749273236dcfc808e54404cd6ece6204cdb1bc4ccc163455", size = 442484, upload-time = "2025-10-22T09:51:21.575Z" },
+    { url = "https://files.pythonhosted.org/packages/c1/1a/68ca9d10bc309828e8cdb2d57a30dd7e5753ac8520c8d7a0322daeb9eef7/python_bidi-0.6.7-pp311-pypy311_pp73-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:1c5fb99f774748de283fadf915106f130b74be1bade934b7f73a7a8488b95da1", size = 329149, upload-time = "2025-10-22T09:51:31.232Z" },
+    { url = "https://files.pythonhosted.org/packages/03/40/ab450c06167a7de596d99b1ba5cee2c605b3ff184baccf08210ede706b1b/python_bidi-0.6.7-pp311-pypy311_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:2d28e2bdcadf5b6161bb4ee9313ce41eac746ba57e744168bf723a415a11af05", size = 303529, upload-time = "2025-10-22T09:51:46.997Z" },
+    { url = "https://files.pythonhosted.org/packages/ec/c5/585b5c413e3b77a32500fb877ea30aa23c45a6064dbd7fe77d87b72cd90b/python_bidi-0.6.7-pp311-pypy311_pp73-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:c3777ae3e088e94df854fbcbd8d59f9239b74aac036cb6bbd19f8035c8e42478", size = 317753, upload-time = "2025-10-22T09:51:39.272Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/05/b7b4b447890d614ccb40633f4d65f334bcf9fe3ad13be33aaa54dcbc34f3/python_bidi-0.6.7-pp311-pypy311_pp73-musllinux_1_2_aarch64.whl", hash = "sha256:77bb4cbadf4121db395189065c58c9dd5d1950257cc1983004e6df4a3e2f97ad", size = 476054, upload-time = "2025-10-22T09:52:15.856Z" },
+    { url = "https://files.pythonhosted.org/packages/ca/94/64f6d2c09c4426918345b54ca8902f94b663eadd744c9dd89070f546c9bc/python_bidi-0.6.7-pp311-pypy311_pp73-musllinux_1_2_armv7l.whl", hash = "sha256:f1fe71c203f66bc169a393964d5702f9251cfd4d70279cb6453fdd42bd2e675f", size = 568365, upload-time = "2025-10-22T09:52:27.556Z" },
+    { url = "https://files.pythonhosted.org/packages/fc/d2/c39a6b82aa0fcedac7cbe6078b78bb9089b43d903f8e00859e42b504bb8e/python_bidi-0.6.7-pp311-pypy311_pp73-musllinux_1_2_i686.whl", hash = "sha256:d87ed09e5c9b6d2648e8856a4e556147b9d3cd4d63905fa664dd6706bc414256", size = 495292, upload-time = "2025-10-22T09:52:38.306Z" },
+    { url = "https://files.pythonhosted.org/packages/0a/8d/a80f37ab92118e305d7b574306553599f81534c50b4eb23ef34ebe09c09c/python_bidi-0.6.7-pp311-pypy311_pp73-musllinux_1_2_x86_64.whl", hash = "sha256:766d5f5a686eb99b53168a7bdfb338035931a609bdbbcb537cef9e050a86f359", size = 467159, upload-time = "2025-10-22T09:52:48.603Z" },
+]
+
+[[package]]
+name = "python-dateutil"
+version = "2.9.0.post0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "six" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/66/c0/0c8b6ad9f17a802ee498c46e004a0eb49bc148f2fd230864601a86dcf6db/python-dateutil-2.9.0.post0.tar.gz", hash = "sha256:37dd54208da7e1cd875388217d5e00ebd4179249f90fb72437e91a35459a0ad3", size = 342432, upload-time = "2024-03-01T18:36:20.211Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/ec/57/56b9bcc3c9c6a792fcbaf139543cee77261f3651ca9da0c93f5c1221264b/python_dateutil-2.9.0.post0-py2.py3-none-any.whl", hash = "sha256:a8b2bc7bffae282281c8140a97d3aa9c14da0b136dfe83f850eea9a5f7470427", size = 229892, upload-time = "2024-03-01T18:36:18.57Z" },
+]
+
+[[package]]
+name = "pytz"
+version = "2025.2"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/f8/bf/abbd3cdfb8fbc7fb3d4d38d320f2441b1e7cbe29be4f23797b4a2b5d8aac/pytz-2025.2.tar.gz", hash = "sha256:360b9e3dbb49a209c21ad61809c7fb453643e048b38924c765813546746e81c3", size = 320884, upload-time = "2025-03-25T02:25:00.538Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/81/c4/34e93fe5f5429d7570ec1fa436f1986fb1f00c3e0f43a589fe2bbcd22c3f/pytz-2025.2-py2.py3-none-any.whl", hash = "sha256:5ddf76296dd8c44c26eb8f4b6f35488f3ccbf6fbbd7adee0b7262d43f0ec2f00", size = 509225, upload-time = "2025-03-25T02:24:58.468Z" },
+]
+
+[[package]]
+name = "pyyaml"
+version = "6.0.2"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/54/ed/79a089b6be93607fa5cdaedf301d7dfb23af5f25c398d5ead2525b063e17/pyyaml-6.0.2.tar.gz", hash = "sha256:d584d9ec91ad65861cc08d42e834324ef890a082e591037abe114850ff7bbc3e", size = 130631, upload-time = "2024-08-06T20:33:50.674Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/9b/95/a3fac87cb7158e231b5a6012e438c647e1a87f09f8e0d123acec8ab8bf71/PyYAML-6.0.2-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:0a9a2848a5b7feac301353437eb7d5957887edbf81d56e903999a75a3d743086", size = 184199, upload-time = "2024-08-06T20:31:40.178Z" },
+    { url = "https://files.pythonhosted.org/packages/c7/7a/68bd47624dab8fd4afbfd3c48e3b79efe09098ae941de5b58abcbadff5cb/PyYAML-6.0.2-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:29717114e51c84ddfba879543fb232a6ed60086602313ca38cce623c1d62cfbf", size = 171758, upload-time = "2024-08-06T20:31:42.173Z" },
+    { url = "https://files.pythonhosted.org/packages/49/ee/14c54df452143b9ee9f0f29074d7ca5516a36edb0b4cc40c3f280131656f/PyYAML-6.0.2-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:8824b5a04a04a047e72eea5cec3bc266db09e35de6bdfe34c9436ac5ee27d237", size = 718463, upload-time = "2024-08-06T20:31:44.263Z" },
+    { url = "https://files.pythonhosted.org/packages/4d/61/de363a97476e766574650d742205be468921a7b532aa2499fcd886b62530/PyYAML-6.0.2-cp310-cp310-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:7c36280e6fb8385e520936c3cb3b8042851904eba0e58d277dca80a5cfed590b", size = 719280, upload-time = "2024-08-06T20:31:50.199Z" },
+    { url = "https://files.pythonhosted.org/packages/6b/4e/1523cb902fd98355e2e9ea5e5eb237cbc5f3ad5f3075fa65087aa0ecb669/PyYAML-6.0.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:ec031d5d2feb36d1d1a24380e4db6d43695f3748343d99434e6f5f9156aaa2ed", size = 751239, upload-time = "2024-08-06T20:31:52.292Z" },
+    { url = "https://files.pythonhosted.org/packages/b7/33/5504b3a9a4464893c32f118a9cc045190a91637b119a9c881da1cf6b7a72/PyYAML-6.0.2-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:936d68689298c36b53b29f23c6dbb74de12b4ac12ca6cfe0e047bedceea56180", size = 695802, upload-time = "2024-08-06T20:31:53.836Z" },
+    { url = "https://files.pythonhosted.org/packages/5c/20/8347dcabd41ef3a3cdc4f7b7a2aff3d06598c8779faa189cdbf878b626a4/PyYAML-6.0.2-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:23502f431948090f597378482b4812b0caae32c22213aecf3b55325e049a6c68", size = 720527, upload-time = "2024-08-06T20:31:55.565Z" },
+    { url = "https://files.pythonhosted.org/packages/be/aa/5afe99233fb360d0ff37377145a949ae258aaab831bde4792b32650a4378/PyYAML-6.0.2-cp310-cp310-win32.whl", hash = "sha256:2e99c6826ffa974fe6e27cdb5ed0021786b03fc98e5ee3c5bfe1fd5015f42b99", size = 144052, upload-time = "2024-08-06T20:31:56.914Z" },
+    { url = "https://files.pythonhosted.org/packages/b5/84/0fa4b06f6d6c958d207620fc60005e241ecedceee58931bb20138e1e5776/PyYAML-6.0.2-cp310-cp310-win_amd64.whl", hash = "sha256:a4d3091415f010369ae4ed1fc6b79def9416358877534caf6a0fdd2146c87a3e", size = 161774, upload-time = "2024-08-06T20:31:58.304Z" },
+    { url = "https://files.pythonhosted.org/packages/f8/aa/7af4e81f7acba21a4c6be026da38fd2b872ca46226673c89a758ebdc4fd2/PyYAML-6.0.2-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:cc1c1159b3d456576af7a3e4d1ba7e6924cb39de8f67111c735f6fc832082774", size = 184612, upload-time = "2024-08-06T20:32:03.408Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/62/b9faa998fd185f65c1371643678e4d58254add437edb764a08c5a98fb986/PyYAML-6.0.2-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:1e2120ef853f59c7419231f3bf4e7021f1b936f6ebd222406c3b60212205d2ee", size = 172040, upload-time = "2024-08-06T20:32:04.926Z" },
+    { url = "https://files.pythonhosted.org/packages/ad/0c/c804f5f922a9a6563bab712d8dcc70251e8af811fce4524d57c2c0fd49a4/PyYAML-6.0.2-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:5d225db5a45f21e78dd9358e58a98702a0302f2659a3c6cd320564b75b86f47c", size = 736829, upload-time = "2024-08-06T20:32:06.459Z" },
+    { url = "https://files.pythonhosted.org/packages/51/16/6af8d6a6b210c8e54f1406a6b9481febf9c64a3109c541567e35a49aa2e7/PyYAML-6.0.2-cp311-cp311-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:5ac9328ec4831237bec75defaf839f7d4564be1e6b25ac710bd1a96321cc8317", size = 764167, upload-time = "2024-08-06T20:32:08.338Z" },
+    { url = "https://files.pythonhosted.org/packages/75/e4/2c27590dfc9992f73aabbeb9241ae20220bd9452df27483b6e56d3975cc5/PyYAML-6.0.2-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:3ad2a3decf9aaba3d29c8f537ac4b243e36bef957511b4766cb0057d32b0be85", size = 762952, upload-time = "2024-08-06T20:32:14.124Z" },
+    { url = "https://files.pythonhosted.org/packages/9b/97/ecc1abf4a823f5ac61941a9c00fe501b02ac3ab0e373c3857f7d4b83e2b6/PyYAML-6.0.2-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:ff3824dc5261f50c9b0dfb3be22b4567a6f938ccce4587b38952d85fd9e9afe4", size = 735301, upload-time = "2024-08-06T20:32:16.17Z" },
+    { url = "https://files.pythonhosted.org/packages/45/73/0f49dacd6e82c9430e46f4a027baa4ca205e8b0a9dce1397f44edc23559d/PyYAML-6.0.2-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:797b4f722ffa07cc8d62053e4cff1486fa6dc094105d13fea7b1de7d8bf71c9e", size = 756638, upload-time = "2024-08-06T20:32:18.555Z" },
+    { url = "https://files.pythonhosted.org/packages/22/5f/956f0f9fc65223a58fbc14459bf34b4cc48dec52e00535c79b8db361aabd/PyYAML-6.0.2-cp311-cp311-win32.whl", hash = "sha256:11d8f3dd2b9c1207dcaf2ee0bbbfd5991f571186ec9cc78427ba5bd32afae4b5", size = 143850, upload-time = "2024-08-06T20:32:19.889Z" },
+    { url = "https://files.pythonhosted.org/packages/ed/23/8da0bbe2ab9dcdd11f4f4557ccaf95c10b9811b13ecced089d43ce59c3c8/PyYAML-6.0.2-cp311-cp311-win_amd64.whl", hash = "sha256:e10ce637b18caea04431ce14fabcf5c64a1c61ec9c56b071a4b7ca131ca52d44", size = 161980, upload-time = "2024-08-06T20:32:21.273Z" },
+    { url = "https://files.pythonhosted.org/packages/86/0c/c581167fc46d6d6d7ddcfb8c843a4de25bdd27e4466938109ca68492292c/PyYAML-6.0.2-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:c70c95198c015b85feafc136515252a261a84561b7b1d51e3384e0655ddf25ab", size = 183873, upload-time = "2024-08-06T20:32:25.131Z" },
+    { url = "https://files.pythonhosted.org/packages/a8/0c/38374f5bb272c051e2a69281d71cba6fdb983413e6758b84482905e29a5d/PyYAML-6.0.2-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:ce826d6ef20b1bc864f0a68340c8b3287705cae2f8b4b1d932177dcc76721725", size = 173302, upload-time = "2024-08-06T20:32:26.511Z" },
+    { url = "https://files.pythonhosted.org/packages/c3/93/9916574aa8c00aa06bbac729972eb1071d002b8e158bd0e83a3b9a20a1f7/PyYAML-6.0.2-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:1f71ea527786de97d1a0cc0eacd1defc0985dcf6b3f17bb77dcfc8c34bec4dc5", size = 739154, upload-time = "2024-08-06T20:32:28.363Z" },
+    { url = "https://files.pythonhosted.org/packages/95/0f/b8938f1cbd09739c6da569d172531567dbcc9789e0029aa070856f123984/PyYAML-6.0.2-cp312-cp312-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:9b22676e8097e9e22e36d6b7bda33190d0d400f345f23d4065d48f4ca7ae0425", size = 766223, upload-time = "2024-08-06T20:32:30.058Z" },
+    { url = "https://files.pythonhosted.org/packages/b9/2b/614b4752f2e127db5cc206abc23a8c19678e92b23c3db30fc86ab731d3bd/PyYAML-6.0.2-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:80bab7bfc629882493af4aa31a4cfa43a4c57c83813253626916b8c7ada83476", size = 767542, upload-time = "2024-08-06T20:32:31.881Z" },
+    { url = "https://files.pythonhosted.org/packages/d4/00/dd137d5bcc7efea1836d6264f049359861cf548469d18da90cd8216cf05f/PyYAML-6.0.2-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:0833f8694549e586547b576dcfaba4a6b55b9e96098b36cdc7ebefe667dfed48", size = 731164, upload-time = "2024-08-06T20:32:37.083Z" },
+    { url = "https://files.pythonhosted.org/packages/c9/1f/4f998c900485e5c0ef43838363ba4a9723ac0ad73a9dc42068b12aaba4e4/PyYAML-6.0.2-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:8b9c7197f7cb2738065c481a0461e50ad02f18c78cd75775628afb4d7137fb3b", size = 756611, upload-time = "2024-08-06T20:32:38.898Z" },
+    { url = "https://files.pythonhosted.org/packages/df/d1/f5a275fdb252768b7a11ec63585bc38d0e87c9e05668a139fea92b80634c/PyYAML-6.0.2-cp312-cp312-win32.whl", hash = "sha256:ef6107725bd54b262d6dedcc2af448a266975032bc85ef0172c5f059da6325b4", size = 140591, upload-time = "2024-08-06T20:32:40.241Z" },
+    { url = "https://files.pythonhosted.org/packages/0c/e8/4f648c598b17c3d06e8753d7d13d57542b30d56e6c2dedf9c331ae56312e/PyYAML-6.0.2-cp312-cp312-win_amd64.whl", hash = "sha256:7e7401d0de89a9a855c839bc697c079a4af81cf878373abd7dc625847d25cbd8", size = 156338, upload-time = "2024-08-06T20:32:41.93Z" },
+    { url = "https://files.pythonhosted.org/packages/ef/e3/3af305b830494fa85d95f6d95ef7fa73f2ee1cc8ef5b495c7c3269fb835f/PyYAML-6.0.2-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:efdca5630322a10774e8e98e1af481aad470dd62c3170801852d752aa7a783ba", size = 181309, upload-time = "2024-08-06T20:32:43.4Z" },
+    { url = "https://files.pythonhosted.org/packages/45/9f/3b1c20a0b7a3200524eb0076cc027a970d320bd3a6592873c85c92a08731/PyYAML-6.0.2-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:50187695423ffe49e2deacb8cd10510bc361faac997de9efef88badc3bb9e2d1", size = 171679, upload-time = "2024-08-06T20:32:44.801Z" },
+    { url = "https://files.pythonhosted.org/packages/7c/9a/337322f27005c33bcb656c655fa78325b730324c78620e8328ae28b64d0c/PyYAML-6.0.2-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:0ffe8360bab4910ef1b9e87fb812d8bc0a308b0d0eef8c8f44e0254ab3b07133", size = 733428, upload-time = "2024-08-06T20:32:46.432Z" },
+    { url = "https://files.pythonhosted.org/packages/a3/69/864fbe19e6c18ea3cc196cbe5d392175b4cf3d5d0ac1403ec3f2d237ebb5/PyYAML-6.0.2-cp313-cp313-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:17e311b6c678207928d649faa7cb0d7b4c26a0ba73d41e99c4fff6b6c3276484", size = 763361, upload-time = "2024-08-06T20:32:51.188Z" },
+    { url = "https://files.pythonhosted.org/packages/04/24/b7721e4845c2f162d26f50521b825fb061bc0a5afcf9a386840f23ea19fa/PyYAML-6.0.2-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:70b189594dbe54f75ab3a1acec5f1e3faa7e8cf2f1e08d9b561cb41b845f69d5", size = 759523, upload-time = "2024-08-06T20:32:53.019Z" },
+    { url = "https://files.pythonhosted.org/packages/2b/b2/e3234f59ba06559c6ff63c4e10baea10e5e7df868092bf9ab40e5b9c56b6/PyYAML-6.0.2-cp313-cp313-musllinux_1_1_aarch64.whl", hash = "sha256:41e4e3953a79407c794916fa277a82531dd93aad34e29c2a514c2c0c5fe971cc", size = 726660, upload-time = "2024-08-06T20:32:54.708Z" },
+    { url = "https://files.pythonhosted.org/packages/fe/0f/25911a9f080464c59fab9027482f822b86bf0608957a5fcc6eaac85aa515/PyYAML-6.0.2-cp313-cp313-musllinux_1_1_x86_64.whl", hash = "sha256:68ccc6023a3400877818152ad9a1033e3db8625d899c72eacb5a668902e4d652", size = 751597, upload-time = "2024-08-06T20:32:56.985Z" },
+    { url = "https://files.pythonhosted.org/packages/14/0d/e2c3b43bbce3cf6bd97c840b46088a3031085179e596d4929729d8d68270/PyYAML-6.0.2-cp313-cp313-win32.whl", hash = "sha256:bc2fa7c6b47d6bc618dd7fb02ef6fdedb1090ec036abab80d4681424b84c1183", size = 140527, upload-time = "2024-08-06T20:33:03.001Z" },
+    { url = "https://files.pythonhosted.org/packages/fa/de/02b54f42487e3d3c6efb3f89428677074ca7bf43aae402517bc7cca949f3/PyYAML-6.0.2-cp313-cp313-win_amd64.whl", hash = "sha256:8388ee1976c416731879ac16da0aff3f63b286ffdd57cdeb95f3f2e085687563", size = 156446, upload-time = "2024-08-06T20:33:04.33Z" },
+]
+
+[[package]]
+name = "requests"
+version = "2.32.5"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "certifi" },
+    { name = "charset-normalizer" },
+    { name = "idna" },
+    { name = "urllib3" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/c9/74/b3ff8e6c8446842c3f5c837e9c3dfcfe2018ea6ecef224c710c85ef728f4/requests-2.32.5.tar.gz", hash = "sha256:dbba0bac56e100853db0ea71b82b4dfd5fe2bf6d3754a8893c3af500cec7d7cf", size = 134517, upload-time = "2025-08-18T20:46:02.573Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/1e/db/4254e3eabe8020b458f1a747140d32277ec7a271daf1d235b70dc0b4e6e3/requests-2.32.5-py3-none-any.whl", hash = "sha256:2462f94637a34fd532264295e186976db0f5d453d1cdd31473c85a6a161affb6", size = 64738, upload-time = "2025-08-18T20:46:00.542Z" },
+]
+
+[[package]]
+name = "requests-toolbelt"
+version = "1.0.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "requests" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/f3/61/d7545dafb7ac2230c70d38d31cbfe4cc64f7144dc41f6e4e4b78ecd9f5bb/requests-toolbelt-1.0.0.tar.gz", hash = "sha256:7681a0a3d047012b5bdc0ee37d7f8f07ebe76ab08caeccfc3921ce23c88d5bc6", size = 206888, upload-time = "2023-05-01T04:11:33.229Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/3f/51/d4db610ef29373b879047326cbf6fa98b6c1969d6f6dc423279de2b1be2c/requests_toolbelt-1.0.0-py2.py3-none-any.whl", hash = "sha256:cccfdd665f0a24fcf4726e690f65639d272bb0637b9b92dfd91a5568ccf6bd06", size = 54481, upload-time = "2023-05-01T04:11:28.427Z" },
+]
+
+[[package]]
+name = "rich"
+version = "14.3.2"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "markdown-it-py" },
+    { name = "pygments" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/74/99/a4cab2acbb884f80e558b0771e97e21e939c5dfb460f488d19df485e8298/rich-14.3.2.tar.gz", hash = "sha256:e712f11c1a562a11843306f5ed999475f09ac31ffb64281f73ab29ffdda8b3b8", size = 230143, upload-time = "2026-02-01T16:20:47.908Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/ef/45/615f5babd880b4bd7d405cc0dc348234c5ffb6ed1ea33e152ede08b2072d/rich-14.3.2-py3-none-any.whl", hash = "sha256:08e67c3e90884651da3239ea668222d19bea7b589149d8014a21c633420dbb69", size = 309963, upload-time = "2026-02-01T16:20:46.078Z" },
+]
+
+[[package]]
+name = "ruamel-yaml"
+version = "0.19.1"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/c7/3b/ebda527b56beb90cb7652cb1c7e4f91f48649fbcd8d2eb2fb6e77cd3329b/ruamel_yaml-0.19.1.tar.gz", hash = "sha256:53eb66cd27849eff968ebf8f0bf61f46cdac2da1d1f3576dd4ccee9b25c31993", size = 142709, upload-time = "2026-01-02T16:50:31.84Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/b8/0c/51f6841f1d84f404f92463fc2b1ba0da357ca1e3db6b7fbda26956c3b82a/ruamel_yaml-0.19.1-py3-none-any.whl", hash = "sha256:27592957fedf6e0b62f281e96effd28043345e0e66001f97683aa9a40c667c93", size = 118102, upload-time = "2026-01-02T16:50:29.201Z" },
+]
+
+[[package]]
+name = "ruff"
+version = "0.13.3"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/c7/8e/f9f9ca747fea8e3ac954e3690d4698c9737c23b51731d02df999c150b1c9/ruff-0.13.3.tar.gz", hash = "sha256:5b0ba0db740eefdfbcce4299f49e9eaefc643d4d007749d77d047c2bab19908e", size = 5438533, upload-time = "2025-10-02T19:29:31.582Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/d2/33/8f7163553481466a92656d35dea9331095122bb84cf98210bef597dd2ecd/ruff-0.13.3-py3-none-linux_armv6l.whl", hash = "sha256:311860a4c5e19189c89d035638f500c1e191d283d0cc2f1600c8c80d6dcd430c", size = 12484040, upload-time = "2025-10-02T19:28:49.199Z" },
+    { url = "https://files.pythonhosted.org/packages/b0/b5/4a21a4922e5dd6845e91896b0d9ef493574cbe061ef7d00a73c61db531af/ruff-0.13.3-py3-none-macosx_10_12_x86_64.whl", hash = "sha256:2bdad6512fb666b40fcadb65e33add2b040fc18a24997d2e47fee7d66f7fcae2", size = 13122975, upload-time = "2025-10-02T19:28:52.446Z" },
+    { url = "https://files.pythonhosted.org/packages/40/90/15649af836d88c9f154e5be87e64ae7d2b1baa5a3ef317cb0c8fafcd882d/ruff-0.13.3-py3-none-macosx_11_0_arm64.whl", hash = "sha256:fc6fa4637284708d6ed4e5e970d52fc3b76a557d7b4e85a53013d9d201d93286", size = 12346621, upload-time = "2025-10-02T19:28:54.712Z" },
+    { url = "https://files.pythonhosted.org/packages/a5/42/bcbccb8141305f9a6d3f72549dd82d1134299177cc7eaf832599700f95a7/ruff-0.13.3-py3-none-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:1c9e6469864f94a98f412f20ea143d547e4c652f45e44f369d7b74ee78185838", size = 12574408, upload-time = "2025-10-02T19:28:56.679Z" },
+    { url = "https://files.pythonhosted.org/packages/ce/19/0f3681c941cdcfa2d110ce4515624c07a964dc315d3100d889fcad3bfc9e/ruff-0.13.3-py3-none-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:5bf62b705f319476c78891e0e97e965b21db468b3c999086de8ffb0d40fd2822", size = 12285330, upload-time = "2025-10-02T19:28:58.79Z" },
+    { url = "https://files.pythonhosted.org/packages/10/f8/387976bf00d126b907bbd7725219257feea58650e6b055b29b224d8cb731/ruff-0.13.3-py3-none-manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:78cc1abed87ce40cb07ee0667ce99dbc766c9f519eabfd948ed87295d8737c60", size = 13980815, upload-time = "2025-10-02T19:29:01.577Z" },
+    { url = "https://files.pythonhosted.org/packages/0c/a6/7c8ec09d62d5a406e2b17d159e4817b63c945a8b9188a771193b7e1cc0b5/ruff-0.13.3-py3-none-manylinux_2_17_ppc64.manylinux2014_ppc64.whl", hash = "sha256:4fb75e7c402d504f7a9a259e0442b96403fa4a7310ffe3588d11d7e170d2b1e3", size = 14987733, upload-time = "2025-10-02T19:29:04.036Z" },
+    { url = "https://files.pythonhosted.org/packages/97/e5/f403a60a12258e0fd0c2195341cfa170726f254c788673495d86ab5a9a9d/ruff-0.13.3-py3-none-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:17b951f9d9afb39330b2bdd2dd144ce1c1335881c277837ac1b50bfd99985ed3", size = 14439848, upload-time = "2025-10-02T19:29:06.684Z" },
+    { url = "https://files.pythonhosted.org/packages/39/49/3de381343e89364c2334c9f3268b0349dc734fc18b2d99a302d0935c8345/ruff-0.13.3-py3-none-manylinux_2_17_s390x.manylinux2014_s390x.whl", hash = "sha256:6052f8088728898e0a449f0dde8fafc7ed47e4d878168b211977e3e7e854f662", size = 13421890, upload-time = "2025-10-02T19:29:08.767Z" },
+    { url = "https://files.pythonhosted.org/packages/ab/b5/c0feca27d45ae74185a6bacc399f5d8920ab82df2d732a17213fb86a2c4c/ruff-0.13.3-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:dc742c50f4ba72ce2a3be362bd359aef7d0d302bf7637a6f942eaa763bd292af", size = 13444870, upload-time = "2025-10-02T19:29:11.234Z" },
+    { url = "https://files.pythonhosted.org/packages/50/a1/b655298a1f3fda4fdc7340c3f671a4b260b009068fbeb3e4e151e9e3e1bf/ruff-0.13.3-py3-none-manylinux_2_31_riscv64.whl", hash = "sha256:8e5640349493b378431637019366bbd73c927e515c9c1babfea3e932f5e68e1d", size = 13691599, upload-time = "2025-10-02T19:29:13.353Z" },
+    { url = "https://files.pythonhosted.org/packages/32/b0/a8705065b2dafae007bcae21354e6e2e832e03eb077bb6c8e523c2becb92/ruff-0.13.3-py3-none-musllinux_1_2_aarch64.whl", hash = "sha256:6b139f638a80eae7073c691a5dd8d581e0ba319540be97c343d60fb12949c8d0", size = 12421893, upload-time = "2025-10-02T19:29:15.668Z" },
+    { url = "https://files.pythonhosted.org/packages/0d/1e/cbe7082588d025cddbb2f23e6dfef08b1a2ef6d6f8328584ad3015b5cebd/ruff-0.13.3-py3-none-musllinux_1_2_armv7l.whl", hash = "sha256:6b547def0a40054825de7cfa341039ebdfa51f3d4bfa6a0772940ed351d2746c", size = 12267220, upload-time = "2025-10-02T19:29:17.583Z" },
+    { url = "https://files.pythonhosted.org/packages/a5/99/4086f9c43f85e0755996d09bdcb334b6fee9b1eabdf34e7d8b877fadf964/ruff-0.13.3-py3-none-musllinux_1_2_i686.whl", hash = "sha256:9cc48a3564423915c93573f1981d57d101e617839bef38504f85f3677b3a0a3e", size = 13177818, upload-time = "2025-10-02T19:29:19.943Z" },
+    { url = "https://files.pythonhosted.org/packages/9b/de/7b5db7e39947d9dc1c5f9f17b838ad6e680527d45288eeb568e860467010/ruff-0.13.3-py3-none-musllinux_1_2_x86_64.whl", hash = "sha256:1a993b17ec03719c502881cb2d5f91771e8742f2ca6de740034433a97c561989", size = 13618715, upload-time = "2025-10-02T19:29:22.527Z" },
+    { url = "https://files.pythonhosted.org/packages/28/d3/bb25ee567ce2f61ac52430cf99f446b0e6d49bdfa4188699ad005fdd16aa/ruff-0.13.3-py3-none-win32.whl", hash = "sha256:f14e0d1fe6460f07814d03c6e32e815bff411505178a1f539a38f6097d3e8ee3", size = 12334488, upload-time = "2025-10-02T19:29:24.782Z" },
+    { url = "https://files.pythonhosted.org/packages/cf/49/12f5955818a1139eed288753479ba9d996f6ea0b101784bb1fe6977ec128/ruff-0.13.3-py3-none-win_amd64.whl", hash = "sha256:621e2e5812b691d4f244638d693e640f188bacbb9bc793ddd46837cea0503dd2", size = 13455262, upload-time = "2025-10-02T19:29:26.882Z" },
+    { url = "https://files.pythonhosted.org/packages/fe/72/7b83242b26627a00e3af70d0394d68f8f02750d642567af12983031777fc/ruff-0.13.3-py3-none-win_arm64.whl", hash = "sha256:9e9e9d699841eaf4c2c798fa783df2fabc680b72059a02ca0ed81c460bc58330", size = 12538484, upload-time = "2025-10-02T19:29:28.951Z" },
+]
+
+[[package]]
+name = "setuptools"
+version = "80.10.2"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/76/95/faf61eb8363f26aa7e1d762267a8d602a1b26d4f3a1e758e92cb3cb8b054/setuptools-80.10.2.tar.gz", hash = "sha256:8b0e9d10c784bf7d262c4e5ec5d4ec94127ce206e8738f29a437945fbc219b70", size = 1200343, upload-time = "2026-01-25T22:38:17.252Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/94/b8/f1f62a5e3c0ad2ff1d189590bfa4c46b4f3b6e49cef6f26c6ee4e575394d/setuptools-80.10.2-py3-none-any.whl", hash = "sha256:95b30ddfb717250edb492926c92b5221f7ef3fbcc2b07579bcd4a27da21d0173", size = 1064234, upload-time = "2026-01-25T22:38:15.216Z" },
+]
+
+[[package]]
+name = "shapely"
+version = "2.1.2"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
+    { name = "numpy", version = "2.4.2", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/4d/bc/0989043118a27cccb4e906a46b7565ce36ca7b57f5a18b78f4f1b0f72d9d/shapely-2.1.2.tar.gz", hash = "sha256:2ed4ecb28320a433db18a5bf029986aa8afcfd740745e78847e330d5d94922a9", size = 315489, upload-time = "2025-09-24T13:51:41.432Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/05/89/c3548aa9b9812a5d143986764dededfa48d817714e947398bdda87c77a72/shapely-2.1.2-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:7ae48c236c0324b4e139bea88a306a04ca630f49be66741b340729d380d8f52f", size = 1825959, upload-time = "2025-09-24T13:50:00.682Z" },
+    { url = "https://files.pythonhosted.org/packages/ce/8a/7ebc947080442edd614ceebe0ce2cdbd00c25e832c240e1d1de61d0e6b38/shapely-2.1.2-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:eba6710407f1daa8e7602c347dfc94adc02205ec27ed956346190d66579eb9ea", size = 1629196, upload-time = "2025-09-24T13:50:03.447Z" },
+    { url = "https://files.pythonhosted.org/packages/c8/86/c9c27881c20d00fc409e7e059de569d5ed0abfcec9c49548b124ebddea51/shapely-2.1.2-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:ef4a456cc8b7b3d50ccec29642aa4aeda959e9da2fe9540a92754770d5f0cf1f", size = 2951065, upload-time = "2025-09-24T13:50:05.266Z" },
+    { url = "https://files.pythonhosted.org/packages/50/8a/0ab1f7433a2a85d9e9aea5b1fbb333f3b09b309e7817309250b4b7b2cc7a/shapely-2.1.2-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:e38a190442aacc67ff9f75ce60aec04893041f16f97d242209106d502486a142", size = 3058666, upload-time = "2025-09-24T13:50:06.872Z" },
+    { url = "https://files.pythonhosted.org/packages/bb/c6/5a30ffac9c4f3ffd5b7113a7f5299ccec4713acd5ee44039778a7698224e/shapely-2.1.2-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:40d784101f5d06a1fd30b55fc11ea58a61be23f930d934d86f19a180909908a4", size = 3966905, upload-time = "2025-09-24T13:50:09.417Z" },
+    { url = "https://files.pythonhosted.org/packages/9c/72/e92f3035ba43e53959007f928315a68fbcf2eeb4e5ededb6f0dc7ff1ecc3/shapely-2.1.2-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:f6f6cd5819c50d9bcf921882784586aab34a4bd53e7553e175dece6db513a6f0", size = 4129260, upload-time = "2025-09-24T13:50:11.183Z" },
+    { url = "https://files.pythonhosted.org/packages/42/24/605901b73a3d9f65fa958e63c9211f4be23d584da8a1a7487382fac7fdc5/shapely-2.1.2-cp310-cp310-win32.whl", hash = "sha256:fe9627c39c59e553c90f5bc3128252cb85dc3b3be8189710666d2f8bc3a5503e", size = 1544301, upload-time = "2025-09-24T13:50:12.521Z" },
+    { url = "https://files.pythonhosted.org/packages/e1/89/6db795b8dd3919851856bd2ddd13ce434a748072f6fdee42ff30cbd3afa3/shapely-2.1.2-cp310-cp310-win_amd64.whl", hash = "sha256:1d0bfb4b8f661b3b4ec3565fa36c340bfb1cda82087199711f86a88647d26b2f", size = 1722074, upload-time = "2025-09-24T13:50:13.909Z" },
+    { url = "https://files.pythonhosted.org/packages/8f/8d/1ff672dea9ec6a7b5d422eb6d095ed886e2e523733329f75fdcb14ee1149/shapely-2.1.2-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:91121757b0a36c9aac3427a651a7e6567110a4a67c97edf04f8d55d4765f6618", size = 1820038, upload-time = "2025-09-24T13:50:15.628Z" },
+    { url = "https://files.pythonhosted.org/packages/4f/ce/28fab8c772ce5db23a0d86bf0adaee0c4c79d5ad1db766055fa3dab442e2/shapely-2.1.2-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:16a9c722ba774cf50b5d4541242b4cce05aafd44a015290c82ba8a16931ff63d", size = 1626039, upload-time = "2025-09-24T13:50:16.881Z" },
+    { url = "https://files.pythonhosted.org/packages/70/8b/868b7e3f4982f5006e9395c1e12343c66a8155c0374fdc07c0e6a1ab547d/shapely-2.1.2-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:cc4f7397459b12c0b196c9efe1f9d7e92463cbba142632b4cc6d8bbbbd3e2b09", size = 3001519, upload-time = "2025-09-24T13:50:18.606Z" },
+    { url = "https://files.pythonhosted.org/packages/13/02/58b0b8d9c17c93ab6340edd8b7308c0c5a5b81f94ce65705819b7416dba5/shapely-2.1.2-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:136ab87b17e733e22f0961504d05e77e7be8c9b5a8184f685b4a91a84efe3c26", size = 3110842, upload-time = "2025-09-24T13:50:21.77Z" },
+    { url = "https://files.pythonhosted.org/packages/af/61/8e389c97994d5f331dcffb25e2fa761aeedfb52b3ad9bcdd7b8671f4810a/shapely-2.1.2-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:16c5d0fc45d3aa0a69074979f4f1928ca2734fb2e0dde8af9611e134e46774e7", size = 4021316, upload-time = "2025-09-24T13:50:23.626Z" },
+    { url = "https://files.pythonhosted.org/packages/d3/d4/9b2a9fe6039f9e42ccf2cb3e84f219fd8364b0c3b8e7bbc857b5fbe9c14c/shapely-2.1.2-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:6ddc759f72b5b2b0f54a7e7cde44acef680a55019eb52ac63a7af2cf17cb9cd2", size = 4178586, upload-time = "2025-09-24T13:50:25.443Z" },
+    { url = "https://files.pythonhosted.org/packages/16/f6/9840f6963ed4decf76b08fd6d7fed14f8779fb7a62cb45c5617fa8ac6eab/shapely-2.1.2-cp311-cp311-win32.whl", hash = "sha256:2fa78b49485391224755a856ed3b3bd91c8455f6121fee0db0e71cefb07d0ef6", size = 1543961, upload-time = "2025-09-24T13:50:26.968Z" },
+    { url = "https://files.pythonhosted.org/packages/38/1e/3f8ea46353c2a33c1669eb7327f9665103aa3a8dfe7f2e4ef714c210b2c2/shapely-2.1.2-cp311-cp311-win_amd64.whl", hash = "sha256:c64d5c97b2f47e3cd9b712eaced3b061f2b71234b3fc263e0fcf7d889c6559dc", size = 1722856, upload-time = "2025-09-24T13:50:28.497Z" },
+    { url = "https://files.pythonhosted.org/packages/24/c0/f3b6453cf2dfa99adc0ba6675f9aaff9e526d2224cbd7ff9c1a879238693/shapely-2.1.2-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:fe2533caae6a91a543dec62e8360fe86ffcdc42a7c55f9dfd0128a977a896b94", size = 1833550, upload-time = "2025-09-24T13:50:30.019Z" },
+    { url = "https://files.pythonhosted.org/packages/86/07/59dee0bc4b913b7ab59ab1086225baca5b8f19865e6101db9ebb7243e132/shapely-2.1.2-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:ba4d1333cc0bc94381d6d4308d2e4e008e0bd128bdcff5573199742ee3634359", size = 1643556, upload-time = "2025-09-24T13:50:32.291Z" },
+    { url = "https://files.pythonhosted.org/packages/26/29/a5397e75b435b9895cd53e165083faed5d12fd9626eadec15a83a2411f0f/shapely-2.1.2-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:0bd308103340030feef6c111d3eb98d50dc13feea33affc8a6f9fa549e9458a3", size = 2988308, upload-time = "2025-09-24T13:50:33.862Z" },
+    { url = "https://files.pythonhosted.org/packages/b9/37/e781683abac55dde9771e086b790e554811a71ed0b2b8a1e789b7430dd44/shapely-2.1.2-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:1e7d4d7ad262a48bb44277ca12c7c78cb1b0f56b32c10734ec9a1d30c0b0c54b", size = 3099844, upload-time = "2025-09-24T13:50:35.459Z" },
+    { url = "https://files.pythonhosted.org/packages/d8/f3/9876b64d4a5a321b9dc482c92bb6f061f2fa42131cba643c699f39317cb9/shapely-2.1.2-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:e9eddfe513096a71896441a7c37db72da0687b34752c4e193577a145c71736fc", size = 3988842, upload-time = "2025-09-24T13:50:37.478Z" },
+    { url = "https://files.pythonhosted.org/packages/d1/a0/704c7292f7014c7e74ec84eddb7b109e1fbae74a16deae9c1504b1d15565/shapely-2.1.2-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:980c777c612514c0cf99bc8a9de6d286f5e186dcaf9091252fcd444e5638193d", size = 4152714, upload-time = "2025-09-24T13:50:39.9Z" },
+    { url = "https://files.pythonhosted.org/packages/53/46/319c9dc788884ad0785242543cdffac0e6530e4d0deb6c4862bc4143dcf3/shapely-2.1.2-cp312-cp312-win32.whl", hash = "sha256:9111274b88e4d7b54a95218e243282709b330ef52b7b86bc6aaf4f805306f454", size = 1542745, upload-time = "2025-09-24T13:50:41.414Z" },
+    { url = "https://files.pythonhosted.org/packages/ec/bf/cb6c1c505cb31e818e900b9312d514f381fbfa5c4363edfce0fcc4f8c1a4/shapely-2.1.2-cp312-cp312-win_amd64.whl", hash = "sha256:743044b4cfb34f9a67205cee9279feaf60ba7d02e69febc2afc609047cb49179", size = 1722861, upload-time = "2025-09-24T13:50:43.35Z" },
+    { url = "https://files.pythonhosted.org/packages/c3/90/98ef257c23c46425dc4d1d31005ad7c8d649fe423a38b917db02c30f1f5a/shapely-2.1.2-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:b510dda1a3672d6879beb319bc7c5fd302c6c354584690973c838f46ec3e0fa8", size = 1832644, upload-time = "2025-09-24T13:50:44.886Z" },
+    { url = "https://files.pythonhosted.org/packages/6d/ab/0bee5a830d209adcd3a01f2d4b70e587cdd9fd7380d5198c064091005af8/shapely-2.1.2-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:8cff473e81017594d20ec55d86b54bc635544897e13a7cfc12e36909c5309a2a", size = 1642887, upload-time = "2025-09-24T13:50:46.735Z" },
+    { url = "https://files.pythonhosted.org/packages/2d/5e/7d7f54ba960c13302584c73704d8c4d15404a51024631adb60b126a4ae88/shapely-2.1.2-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:fe7b77dc63d707c09726b7908f575fc04ff1d1ad0f3fb92aec212396bc6cfe5e", size = 2970931, upload-time = "2025-09-24T13:50:48.374Z" },
+    { url = "https://files.pythonhosted.org/packages/f2/a2/83fc37e2a58090e3d2ff79175a95493c664bcd0b653dd75cb9134645a4e5/shapely-2.1.2-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:7ed1a5bbfb386ee8332713bf7508bc24e32d24b74fc9a7b9f8529a55db9f4ee6", size = 3082855, upload-time = "2025-09-24T13:50:50.037Z" },
+    { url = "https://files.pythonhosted.org/packages/44/2b/578faf235a5b09f16b5f02833c53822294d7f21b242f8e2d0cf03fb64321/shapely-2.1.2-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:a84e0582858d841d54355246ddfcbd1fce3179f185da7470f41ce39d001ee1af", size = 3979960, upload-time = "2025-09-24T13:50:51.74Z" },
+    { url = "https://files.pythonhosted.org/packages/4d/04/167f096386120f692cc4ca02f75a17b961858997a95e67a3cb6a7bbd6b53/shapely-2.1.2-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:dc3487447a43d42adcdf52d7ac73804f2312cbfa5d433a7d2c506dcab0033dfd", size = 4142851, upload-time = "2025-09-24T13:50:53.49Z" },
+    { url = "https://files.pythonhosted.org/packages/48/74/fb402c5a6235d1c65a97348b48cdedb75fb19eca2b1d66d04969fc1c6091/shapely-2.1.2-cp313-cp313-win32.whl", hash = "sha256:9c3a3c648aedc9f99c09263b39f2d8252f199cb3ac154fadc173283d7d111350", size = 1541890, upload-time = "2025-09-24T13:50:55.337Z" },
+    { url = "https://files.pythonhosted.org/packages/41/47/3647fe7ad990af60ad98b889657a976042c9988c2807cf322a9d6685f462/shapely-2.1.2-cp313-cp313-win_amd64.whl", hash = "sha256:ca2591bff6645c216695bdf1614fca9c82ea1144d4a7591a466fef64f28f0715", size = 1722151, upload-time = "2025-09-24T13:50:57.153Z" },
+    { url = "https://files.pythonhosted.org/packages/3c/49/63953754faa51ffe7d8189bfbe9ca34def29f8c0e34c67cbe2a2795f269d/shapely-2.1.2-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:2d93d23bdd2ed9dc157b46bc2f19b7da143ca8714464249bef6771c679d5ff40", size = 1834130, upload-time = "2025-09-24T13:50:58.49Z" },
+    { url = "https://files.pythonhosted.org/packages/7f/ee/dce001c1984052970ff60eb4727164892fb2d08052c575042a47f5a9e88f/shapely-2.1.2-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:01d0d304b25634d60bd7cf291828119ab55a3bab87dc4af1e44b07fb225f188b", size = 1642802, upload-time = "2025-09-24T13:50:59.871Z" },
+    { url = "https://files.pythonhosted.org/packages/da/e7/fc4e9a19929522877fa602f705706b96e78376afb7fad09cad5b9af1553c/shapely-2.1.2-cp313-cp313t-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:8d8382dd120d64b03698b7298b89611a6ea6f55ada9d39942838b79c9bc89801", size = 3018460, upload-time = "2025-09-24T13:51:02.08Z" },
+    { url = "https://files.pythonhosted.org/packages/a1/18/7519a25db21847b525696883ddc8e6a0ecaa36159ea88e0fef11466384d0/shapely-2.1.2-cp313-cp313t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:19efa3611eef966e776183e338b2d7ea43569ae99ab34f8d17c2c054d3205cc0", size = 3095223, upload-time = "2025-09-24T13:51:04.472Z" },
+    { url = "https://files.pythonhosted.org/packages/48/de/b59a620b1f3a129c3fecc2737104a0a7e04e79335bd3b0a1f1609744cf17/shapely-2.1.2-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:346ec0c1a0fcd32f57f00e4134d1200e14bf3f5ae12af87ba83ca275c502498c", size = 4030760, upload-time = "2025-09-24T13:51:06.455Z" },
+    { url = "https://files.pythonhosted.org/packages/96/b3/c6655ee7232b417562bae192ae0d3ceaadb1cc0ffc2088a2ddf415456cc2/shapely-2.1.2-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:6305993a35989391bd3476ee538a5c9a845861462327efe00dd11a5c8c709a99", size = 4170078, upload-time = "2025-09-24T13:51:08.584Z" },
+    { url = "https://files.pythonhosted.org/packages/a0/8e/605c76808d73503c9333af8f6cbe7e1354d2d238bda5f88eea36bfe0f42a/shapely-2.1.2-cp313-cp313t-win32.whl", hash = "sha256:c8876673449f3401f278c86eb33224c5764582f72b653a415d0e6672fde887bf", size = 1559178, upload-time = "2025-09-24T13:51:10.73Z" },
+    { url = "https://files.pythonhosted.org/packages/36/f7/d317eb232352a1f1444d11002d477e54514a4a6045536d49d0c59783c0da/shapely-2.1.2-cp313-cp313t-win_amd64.whl", hash = "sha256:4a44bc62a10d84c11a7a3d7c1c4fe857f7477c3506e24c9062da0db0ae0c449c", size = 1739756, upload-time = "2025-09-24T13:51:12.105Z" },
+    { url = "https://files.pythonhosted.org/packages/fc/c4/3ce4c2d9b6aabd27d26ec988f08cb877ba9e6e96086eff81bfea93e688c7/shapely-2.1.2-cp314-cp314-macosx_10_13_x86_64.whl", hash = "sha256:9a522f460d28e2bf4e12396240a5fc1518788b2fcd73535166d748399ef0c223", size = 1831290, upload-time = "2025-09-24T13:51:13.56Z" },
+    { url = "https://files.pythonhosted.org/packages/17/b9/f6ab8918fc15429f79cb04afa9f9913546212d7fb5e5196132a2af46676b/shapely-2.1.2-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:1ff629e00818033b8d71139565527ced7d776c269a49bd78c9df84e8f852190c", size = 1641463, upload-time = "2025-09-24T13:51:14.972Z" },
+    { url = "https://files.pythonhosted.org/packages/a5/57/91d59ae525ca641e7ac5551c04c9503aee6f29b92b392f31790fcb1a4358/shapely-2.1.2-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:f67b34271dedc3c653eba4e3d7111aa421d5be9b4c4c7d38d30907f796cb30df", size = 2970145, upload-time = "2025-09-24T13:51:16.961Z" },
+    { url = "https://files.pythonhosted.org/packages/8a/cb/4948be52ee1da6927831ab59e10d4c29baa2a714f599f1f0d1bc747f5777/shapely-2.1.2-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:21952dc00df38a2c28375659b07a3979d22641aeb104751e769c3ee825aadecf", size = 3073806, upload-time = "2025-09-24T13:51:18.712Z" },
+    { url = "https://files.pythonhosted.org/packages/03/83/f768a54af775eb41ef2e7bec8a0a0dbe7d2431c3e78c0a8bdba7ab17e446/shapely-2.1.2-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:1f2f33f486777456586948e333a56ae21f35ae273be99255a191f5c1fa302eb4", size = 3980803, upload-time = "2025-09-24T13:51:20.37Z" },
+    { url = "https://files.pythonhosted.org/packages/9f/cb/559c7c195807c91c79d38a1f6901384a2878a76fbdf3f1048893a9b7534d/shapely-2.1.2-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:cf831a13e0d5a7eb519e96f58ec26e049b1fad411fc6fc23b162a7ce04d9cffc", size = 4133301, upload-time = "2025-09-24T13:51:21.887Z" },
+    { url = "https://files.pythonhosted.org/packages/80/cd/60d5ae203241c53ef3abd2ef27c6800e21afd6c94e39db5315ea0cbafb4a/shapely-2.1.2-cp314-cp314-win32.whl", hash = "sha256:61edcd8d0d17dd99075d320a1dd39c0cb9616f7572f10ef91b4b5b00c4aeb566", size = 1583247, upload-time = "2025-09-24T13:51:23.401Z" },
+    { url = "https://files.pythonhosted.org/packages/74/d4/135684f342e909330e50d31d441ace06bf83c7dc0777e11043f99167b123/shapely-2.1.2-cp314-cp314-win_amd64.whl", hash = "sha256:a444e7afccdb0999e203b976adb37ea633725333e5b119ad40b1ca291ecf311c", size = 1773019, upload-time = "2025-09-24T13:51:24.873Z" },
+    { url = "https://files.pythonhosted.org/packages/a3/05/a44f3f9f695fa3ada22786dc9da33c933da1cbc4bfe876fe3a100bafe263/shapely-2.1.2-cp314-cp314t-macosx_10_13_x86_64.whl", hash = "sha256:5ebe3f84c6112ad3d4632b1fd2290665aa75d4cef5f6c5d77c4c95b324527c6a", size = 1834137, upload-time = "2025-09-24T13:51:26.665Z" },
+    { url = "https://files.pythonhosted.org/packages/52/7e/4d57db45bf314573427b0a70dfca15d912d108e6023f623947fa69f39b72/shapely-2.1.2-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:5860eb9f00a1d49ebb14e881f5caf6c2cf472c7fd38bd7f253bbd34f934eb076", size = 1642884, upload-time = "2025-09-24T13:51:28.029Z" },
+    { url = "https://files.pythonhosted.org/packages/5a/27/4e29c0a55d6d14ad7422bf86995d7ff3f54af0eba59617eb95caf84b9680/shapely-2.1.2-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:b705c99c76695702656327b819c9660768ec33f5ce01fa32b2af62b56ba400a1", size = 3018320, upload-time = "2025-09-24T13:51:29.903Z" },
+    { url = "https://files.pythonhosted.org/packages/9f/bb/992e6a3c463f4d29d4cd6ab8963b75b1b1040199edbd72beada4af46bde5/shapely-2.1.2-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:a1fd0ea855b2cf7c9cddaf25543e914dd75af9de08785f20ca3085f2c9ca60b0", size = 3094931, upload-time = "2025-09-24T13:51:32.699Z" },
+    { url = "https://files.pythonhosted.org/packages/9c/16/82e65e21070e473f0ed6451224ed9fa0be85033d17e0c6e7213a12f59d12/shapely-2.1.2-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:df90e2db118c3671a0754f38e36802db75fe0920d211a27481daf50a711fdf26", size = 4030406, upload-time = "2025-09-24T13:51:34.189Z" },
+    { url = "https://files.pythonhosted.org/packages/7c/75/c24ed871c576d7e2b64b04b1fe3d075157f6eb54e59670d3f5ffb36e25c7/shapely-2.1.2-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:361b6d45030b4ac64ddd0a26046906c8202eb60d0f9f53085f5179f1d23021a0", size = 4169511, upload-time = "2025-09-24T13:51:36.297Z" },
+    { url = "https://files.pythonhosted.org/packages/b1/f7/b3d1d6d18ebf55236eec1c681ce5e665742aab3c0b7b232720a7d43df7b6/shapely-2.1.2-cp314-cp314t-win32.whl", hash = "sha256:b54df60f1fbdecc8ebc2c5b11870461a6417b3d617f555e5033f1505d36e5735", size = 1602607, upload-time = "2025-09-24T13:51:37.757Z" },
+    { url = "https://files.pythonhosted.org/packages/9a/f6/f09272a71976dfc138129b8faf435d064a811ae2f708cb147dccdf7aacdb/shapely-2.1.2-cp314-cp314t-win_amd64.whl", hash = "sha256:0036ac886e0923417932c2e6369b6c52e38e0ff5d9120b90eef5cd9a5fc5cae9", size = 1796682, upload-time = "2025-09-24T13:51:39.233Z" },
+]
+
+[[package]]
+name = "shellingham"
+version = "1.5.4"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/58/15/8b3609fd3830ef7b27b655beb4b4e9c62313a4e8da8c676e142cc210d58e/shellingham-1.5.4.tar.gz", hash = "sha256:8dbca0739d487e5bd35ab3ca4b36e11c4078f3a234bfce294b0a0291363404de", size = 10310, upload-time = "2023-10-24T04:13:40.426Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/e0/f9/0595336914c5619e5f28a1fb793285925a8cd4b432c9da0a987836c7f822/shellingham-1.5.4-py2.py3-none-any.whl", hash = "sha256:7ecfff8f2fd72616f7481040475a65b2bf8af90a56c89140852d1120324e8686", size = 9755, upload-time = "2023-10-24T04:13:38.866Z" },
+]
+
+[[package]]
+name = "six"
+version = "1.17.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/94/e7/b2c673351809dca68a0e064b6af791aa332cf192da575fd474ed7d6f16a2/six-1.17.0.tar.gz", hash = "sha256:ff70335d468e7eb6ec65b95b99d3a2836546063f63acc5171de367e834932a81", size = 34031, upload-time = "2024-12-04T17:35:28.174Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/b7/ce/149a00dd41f10bc29e5921b496af8b574d8413afcd5e30dfa0ed46c2cc5e/six-1.17.0-py2.py3-none-any.whl", hash = "sha256:4721f391ed90541fddacab5acf947aa0d3dc7d27b2e1e8eda2be8970586c3274", size = 11050, upload-time = "2024-12-04T17:35:26.475Z" },
+]
+
+[[package]]
+name = "starlette"
+version = "0.50.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "anyio" },
+    { name = "typing-extensions", marker = "python_full_version < '3.13'" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/ba/b8/73a0e6a6e079a9d9cfa64113d771e421640b6f679a52eeb9b32f72d871a1/starlette-0.50.0.tar.gz", hash = "sha256:a2a17b22203254bcbc2e1f926d2d55f3f9497f769416b3190768befe598fa3ca", size = 2646985, upload-time = "2025-11-01T15:25:27.516Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/d9/52/1064f510b141bd54025f9b55105e26d1fa970b9be67ad766380a3c9b74b0/starlette-0.50.0-py3-none-any.whl", hash = "sha256:9e5391843ec9b6e472eed1365a78c8098cfceb7a74bfd4d6b1c0c0095efb3bca", size = 74033, upload-time = "2025-11-01T15:25:25.461Z" },
+]
+
+[[package]]
+name = "syrupy"
+version = "4.9.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "pytest" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/8c/f8/022d8704a3314f3e96dbd6bbd16ebe119ce30e35f41aabfa92345652fceb/syrupy-4.9.1.tar.gz", hash = "sha256:b7d0fcadad80a7d2f6c4c71917918e8ebe2483e8c703dfc8d49cdbb01081f9a4", size = 52492, upload-time = "2025-03-24T01:36:37.225Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/ec/9d/aef9ec5fd5a4ee2f6a96032c4eda5888c5c7cec65cef6b28c4fc37671d88/syrupy-4.9.1-py3-none-any.whl", hash = "sha256:b94cc12ed0e5e75b448255430af642516842a2374a46936dd2650cfb6dd20eda", size = 52214, upload-time = "2025-03-24T01:36:35.278Z" },
+]
+
+[[package]]
+name = "tenacity"
+version = "9.1.2"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/0a/d4/2b0cd0fe285e14b36db076e78c93766ff1d529d70408bd1d2a5a84f1d929/tenacity-9.1.2.tar.gz", hash = "sha256:1169d376c297e7de388d18b4481760d478b0e99a777cad3a9c86e556f4b697cb", size = 48036, upload-time = "2025-04-02T08:25:09.966Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/e5/30/643397144bfbfec6f6ef821f36f33e57d35946c44a2352d3c9f0ae847619/tenacity-9.1.2-py3-none-any.whl", hash = "sha256:f77bf36710d8b73a50b2dd155c97b870017ad21afe6ab300326b0371b3b05138", size = 28248, upload-time = "2025-04-02T08:25:07.678Z" },
+]
+
+[[package]]
+name = "tomli"
+version = "2.4.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/82/30/31573e9457673ab10aa432461bee537ce6cef177667deca369efb79df071/tomli-2.4.0.tar.gz", hash = "sha256:aa89c3f6c277dd275d8e243ad24f3b5e701491a860d5121f2cdd399fbb31fc9c", size = 17477, upload-time = "2026-01-11T11:22:38.165Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/3c/d9/3dc2289e1f3b32eb19b9785b6a006b28ee99acb37d1d47f78d4c10e28bf8/tomli-2.4.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:b5ef256a3fd497d4973c11bf142e9ed78b150d36f5773f1ca6088c230ffc5867", size = 153663, upload-time = "2026-01-11T11:21:45.27Z" },
+    { url = "https://files.pythonhosted.org/packages/51/32/ef9f6845e6b9ca392cd3f64f9ec185cc6f09f0a2df3db08cbe8809d1d435/tomli-2.4.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:5572e41282d5268eb09a697c89a7bee84fae66511f87533a6f88bd2f7b652da9", size = 148469, upload-time = "2026-01-11T11:21:46.873Z" },
+    { url = "https://files.pythonhosted.org/packages/d6/c2/506e44cce89a8b1b1e047d64bd495c22c9f71f21e05f380f1a950dd9c217/tomli-2.4.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:551e321c6ba03b55676970b47cb1b73f14a0a4dce6a3e1a9458fd6d921d72e95", size = 236039, upload-time = "2026-01-11T11:21:48.503Z" },
+    { url = "https://files.pythonhosted.org/packages/b3/40/e1b65986dbc861b7e986e8ec394598187fa8aee85b1650b01dd925ca0be8/tomli-2.4.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:5e3f639a7a8f10069d0e15408c0b96a2a828cfdec6fca05296ebcdcc28ca7c76", size = 243007, upload-time = "2026-01-11T11:21:49.456Z" },
+    { url = "https://files.pythonhosted.org/packages/9c/6f/6e39ce66b58a5b7ae572a0f4352ff40c71e8573633deda43f6a379d56b3e/tomli-2.4.0-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:1b168f2731796b045128c45982d3a4874057626da0e2ef1fdd722848b741361d", size = 240875, upload-time = "2026-01-11T11:21:50.755Z" },
+    { url = "https://files.pythonhosted.org/packages/aa/ad/cb089cb190487caa80204d503c7fd0f4d443f90b95cf4ef5cf5aa0f439b0/tomli-2.4.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:133e93646ec4300d651839d382d63edff11d8978be23da4cc106f5a18b7d0576", size = 246271, upload-time = "2026-01-11T11:21:51.81Z" },
+    { url = "https://files.pythonhosted.org/packages/0b/63/69125220e47fd7a3a27fd0de0c6398c89432fec41bc739823bcc66506af6/tomli-2.4.0-cp311-cp311-win32.whl", hash = "sha256:b6c78bdf37764092d369722d9946cb65b8767bfa4110f902a1b2542d8d173c8a", size = 96770, upload-time = "2026-01-11T11:21:52.647Z" },
+    { url = "https://files.pythonhosted.org/packages/1e/0d/a22bb6c83f83386b0008425a6cd1fa1c14b5f3dd4bad05e98cf3dbbf4a64/tomli-2.4.0-cp311-cp311-win_amd64.whl", hash = "sha256:d3d1654e11d724760cdb37a3d7691f0be9db5fbdaef59c9f532aabf87006dbaa", size = 107626, upload-time = "2026-01-11T11:21:53.459Z" },
+    { url = "https://files.pythonhosted.org/packages/2f/6d/77be674a3485e75cacbf2ddba2b146911477bd887dda9d8c9dfb2f15e871/tomli-2.4.0-cp311-cp311-win_arm64.whl", hash = "sha256:cae9c19ed12d4e8f3ebf46d1a75090e4c0dc16271c5bce1c833ac168f08fb614", size = 94842, upload-time = "2026-01-11T11:21:54.831Z" },
+    { url = "https://files.pythonhosted.org/packages/3c/43/7389a1869f2f26dba52404e1ef13b4784b6b37dac93bac53457e3ff24ca3/tomli-2.4.0-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:920b1de295e72887bafa3ad9f7a792f811847d57ea6b1215154030cf131f16b1", size = 154894, upload-time = "2026-01-11T11:21:56.07Z" },
+    { url = "https://files.pythonhosted.org/packages/e9/05/2f9bf110b5294132b2edf13fe6ca6ae456204f3d749f623307cbb7a946f2/tomli-2.4.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:7d6d9a4aee98fac3eab4952ad1d73aee87359452d1c086b5ceb43ed02ddb16b8", size = 149053, upload-time = "2026-01-11T11:21:57.467Z" },
+    { url = "https://files.pythonhosted.org/packages/e8/41/1eda3ca1abc6f6154a8db4d714a4d35c4ad90adc0bcf700657291593fbf3/tomli-2.4.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:36b9d05b51e65b254ea6c2585b59d2c4cb91c8a3d91d0ed0f17591a29aaea54a", size = 243481, upload-time = "2026-01-11T11:21:58.661Z" },
+    { url = "https://files.pythonhosted.org/packages/d2/6d/02ff5ab6c8868b41e7d4b987ce2b5f6a51d3335a70aa144edd999e055a01/tomli-2.4.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:1c8a885b370751837c029ef9bc014f27d80840e48bac415f3412e6593bbc18c1", size = 251720, upload-time = "2026-01-11T11:22:00.178Z" },
+    { url = "https://files.pythonhosted.org/packages/7b/57/0405c59a909c45d5b6f146107c6d997825aa87568b042042f7a9c0afed34/tomli-2.4.0-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:8768715ffc41f0008abe25d808c20c3d990f42b6e2e58305d5da280ae7d1fa3b", size = 247014, upload-time = "2026-01-11T11:22:01.238Z" },
+    { url = "https://files.pythonhosted.org/packages/2c/0e/2e37568edd944b4165735687cbaf2fe3648129e440c26d02223672ee0630/tomli-2.4.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:7b438885858efd5be02a9a133caf5812b8776ee0c969fea02c45e8e3f296ba51", size = 251820, upload-time = "2026-01-11T11:22:02.727Z" },
+    { url = "https://files.pythonhosted.org/packages/5a/1c/ee3b707fdac82aeeb92d1a113f803cf6d0f37bdca0849cb489553e1f417a/tomli-2.4.0-cp312-cp312-win32.whl", hash = "sha256:0408e3de5ec77cc7f81960c362543cbbd91ef883e3138e81b729fc3eea5b9729", size = 97712, upload-time = "2026-01-11T11:22:03.777Z" },
+    { url = "https://files.pythonhosted.org/packages/69/13/c07a9177d0b3bab7913299b9278845fc6eaaca14a02667c6be0b0a2270c8/tomli-2.4.0-cp312-cp312-win_amd64.whl", hash = "sha256:685306e2cc7da35be4ee914fd34ab801a6acacb061b6a7abca922aaf9ad368da", size = 108296, upload-time = "2026-01-11T11:22:04.86Z" },
+    { url = "https://files.pythonhosted.org/packages/18/27/e267a60bbeeee343bcc279bb9e8fbed0cbe224bc7b2a3dc2975f22809a09/tomli-2.4.0-cp312-cp312-win_arm64.whl", hash = "sha256:5aa48d7c2356055feef06a43611fc401a07337d5b006be13a30f6c58f869e3c3", size = 94553, upload-time = "2026-01-11T11:22:05.854Z" },
+    { url = "https://files.pythonhosted.org/packages/34/91/7f65f9809f2936e1f4ce6268ae1903074563603b2a2bd969ebbda802744f/tomli-2.4.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:84d081fbc252d1b6a982e1870660e7330fb8f90f676f6e78b052ad4e64714bf0", size = 154915, upload-time = "2026-01-11T11:22:06.703Z" },
+    { url = "https://files.pythonhosted.org/packages/20/aa/64dd73a5a849c2e8f216b755599c511badde80e91e9bc2271baa7b2cdbb1/tomli-2.4.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:9a08144fa4cba33db5255f9b74f0b89888622109bd2776148f2597447f92a94e", size = 149038, upload-time = "2026-01-11T11:22:07.56Z" },
+    { url = "https://files.pythonhosted.org/packages/9e/8a/6d38870bd3d52c8d1505ce054469a73f73a0fe62c0eaf5dddf61447e32fa/tomli-2.4.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:c73add4bb52a206fd0c0723432db123c0c75c280cbd67174dd9d2db228ebb1b4", size = 242245, upload-time = "2026-01-11T11:22:08.344Z" },
+    { url = "https://files.pythonhosted.org/packages/59/bb/8002fadefb64ab2669e5b977df3f5e444febea60e717e755b38bb7c41029/tomli-2.4.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:1fb2945cbe303b1419e2706e711b7113da57b7db31ee378d08712d678a34e51e", size = 250335, upload-time = "2026-01-11T11:22:09.951Z" },
+    { url = "https://files.pythonhosted.org/packages/a5/3d/4cdb6f791682b2ea916af2de96121b3cb1284d7c203d97d92d6003e91c8d/tomli-2.4.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:bbb1b10aa643d973366dc2cb1ad94f99c1726a02343d43cbc011edbfac579e7c", size = 245962, upload-time = "2026-01-11T11:22:11.27Z" },
+    { url = "https://files.pythonhosted.org/packages/f2/4a/5f25789f9a460bd858ba9756ff52d0830d825b458e13f754952dd15fb7bb/tomli-2.4.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:4cbcb367d44a1f0c2be408758b43e1ffb5308abe0ea222897d6bfc8e8281ef2f", size = 250396, upload-time = "2026-01-11T11:22:12.325Z" },
+    { url = "https://files.pythonhosted.org/packages/aa/2f/b73a36fea58dfa08e8b3a268750e6853a6aac2a349241a905ebd86f3047a/tomli-2.4.0-cp313-cp313-win32.whl", hash = "sha256:7d49c66a7d5e56ac959cb6fc583aff0651094ec071ba9ad43df785abc2320d86", size = 97530, upload-time = "2026-01-11T11:22:13.865Z" },
+    { url = "https://files.pythonhosted.org/packages/3b/af/ca18c134b5d75de7e8dc551c5234eaba2e8e951f6b30139599b53de9c187/tomli-2.4.0-cp313-cp313-win_amd64.whl", hash = "sha256:3cf226acb51d8f1c394c1b310e0e0e61fecdd7adcb78d01e294ac297dd2e7f87", size = 108227, upload-time = "2026-01-11T11:22:15.224Z" },
+    { url = "https://files.pythonhosted.org/packages/22/c3/b386b832f209fee8073c8138ec50f27b4460db2fdae9ffe022df89a57f9b/tomli-2.4.0-cp313-cp313-win_arm64.whl", hash = "sha256:d20b797a5c1ad80c516e41bc1fb0443ddb5006e9aaa7bda2d71978346aeb9132", size = 94748, upload-time = "2026-01-11T11:22:16.009Z" },
+    { url = "https://files.pythonhosted.org/packages/f3/c4/84047a97eb1004418bc10bdbcfebda209fca6338002eba2dc27cc6d13563/tomli-2.4.0-cp314-cp314-macosx_10_15_x86_64.whl", hash = "sha256:26ab906a1eb794cd4e103691daa23d95c6919cc2fa9160000ac02370cc9dd3f6", size = 154725, upload-time = "2026-01-11T11:22:17.269Z" },
+    { url = "https://files.pythonhosted.org/packages/a8/5d/d39038e646060b9d76274078cddf146ced86dc2b9e8bbf737ad5983609a0/tomli-2.4.0-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:20cedb4ee43278bc4f2fee6cb50daec836959aadaf948db5172e776dd3d993fc", size = 148901, upload-time = "2026-01-11T11:22:18.287Z" },
+    { url = "https://files.pythonhosted.org/packages/73/e5/383be1724cb30f4ce44983d249645684a48c435e1cd4f8b5cded8a816d3c/tomli-2.4.0-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:39b0b5d1b6dd03684b3fb276407ebed7090bbec989fa55838c98560c01113b66", size = 243375, upload-time = "2026-01-11T11:22:19.154Z" },
+    { url = "https://files.pythonhosted.org/packages/31/f0/bea80c17971c8d16d3cc109dc3585b0f2ce1036b5f4a8a183789023574f2/tomli-2.4.0-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:a26d7ff68dfdb9f87a016ecfd1e1c2bacbe3108f4e0f8bcd2228ef9a766c787d", size = 250639, upload-time = "2026-01-11T11:22:20.168Z" },
+    { url = "https://files.pythonhosted.org/packages/2c/8f/2853c36abbb7608e3f945d8a74e32ed3a74ee3a1f468f1ffc7d1cb3abba6/tomli-2.4.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:20ffd184fb1df76a66e34bd1b36b4a4641bd2b82954befa32fe8163e79f1a702", size = 246897, upload-time = "2026-01-11T11:22:21.544Z" },
+    { url = "https://files.pythonhosted.org/packages/49/f0/6c05e3196ed5337b9fe7ea003e95fd3819a840b7a0f2bf5a408ef1dad8ed/tomli-2.4.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:75c2f8bbddf170e8effc98f5e9084a8751f8174ea6ccf4fca5398436e0320bc8", size = 254697, upload-time = "2026-01-11T11:22:23.058Z" },
+    { url = "https://files.pythonhosted.org/packages/f3/f5/2922ef29c9f2951883525def7429967fc4d8208494e5ab524234f06b688b/tomli-2.4.0-cp314-cp314-win32.whl", hash = "sha256:31d556d079d72db7c584c0627ff3a24c5d3fb4f730221d3444f3efb1b2514776", size = 98567, upload-time = "2026-01-11T11:22:24.033Z" },
+    { url = "https://files.pythonhosted.org/packages/7b/31/22b52e2e06dd2a5fdbc3ee73226d763b184ff21fc24e20316a44ccc4d96b/tomli-2.4.0-cp314-cp314-win_amd64.whl", hash = "sha256:43e685b9b2341681907759cf3a04e14d7104b3580f808cfde1dfdb60ada85475", size = 108556, upload-time = "2026-01-11T11:22:25.378Z" },
+    { url = "https://files.pythonhosted.org/packages/48/3d/5058dff3255a3d01b705413f64f4306a141a8fd7a251e5a495e3f192a998/tomli-2.4.0-cp314-cp314-win_arm64.whl", hash = "sha256:3d895d56bd3f82ddd6faaff993c275efc2ff38e52322ea264122d72729dca2b2", size = 96014, upload-time = "2026-01-11T11:22:26.138Z" },
+    { url = "https://files.pythonhosted.org/packages/b8/4e/75dab8586e268424202d3a1997ef6014919c941b50642a1682df43204c22/tomli-2.4.0-cp314-cp314t-macosx_10_15_x86_64.whl", hash = "sha256:5b5807f3999fb66776dbce568cc9a828544244a8eb84b84b9bafc080c99597b9", size = 163339, upload-time = "2026-01-11T11:22:27.143Z" },
+    { url = "https://files.pythonhosted.org/packages/06/e3/b904d9ab1016829a776d97f163f183a48be6a4deb87304d1e0116a349519/tomli-2.4.0-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:c084ad935abe686bd9c898e62a02a19abfc9760b5a79bc29644463eaf2840cb0", size = 159490, upload-time = "2026-01-11T11:22:28.399Z" },
+    { url = "https://files.pythonhosted.org/packages/e3/5a/fc3622c8b1ad823e8ea98a35e3c632ee316d48f66f80f9708ceb4f2a0322/tomli-2.4.0-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:0f2e3955efea4d1cfbcb87bc321e00dc08d2bcb737fd1d5e398af111d86db5df", size = 269398, upload-time = "2026-01-11T11:22:29.345Z" },
+    { url = "https://files.pythonhosted.org/packages/fd/33/62bd6152c8bdd4c305ad9faca48f51d3acb2df1f8791b1477d46ff86e7f8/tomli-2.4.0-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:0e0fe8a0b8312acf3a88077a0802565cb09ee34107813bba1c7cd591fa6cfc8d", size = 276515, upload-time = "2026-01-11T11:22:30.327Z" },
+    { url = "https://files.pythonhosted.org/packages/4b/ff/ae53619499f5235ee4211e62a8d7982ba9e439a0fb4f2f351a93d67c1dd2/tomli-2.4.0-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:413540dce94673591859c4c6f794dfeaa845e98bf35d72ed59636f869ef9f86f", size = 273806, upload-time = "2026-01-11T11:22:32.56Z" },
+    { url = "https://files.pythonhosted.org/packages/47/71/cbca7787fa68d4d0a9f7072821980b39fbb1b6faeb5f5cf02f4a5559fa28/tomli-2.4.0-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:0dc56fef0e2c1c470aeac5b6ca8cc7b640bb93e92d9803ddaf9ea03e198f5b0b", size = 281340, upload-time = "2026-01-11T11:22:33.505Z" },
+    { url = "https://files.pythonhosted.org/packages/f5/00/d595c120963ad42474cf6ee7771ad0d0e8a49d0f01e29576ee9195d9ecdf/tomli-2.4.0-cp314-cp314t-win32.whl", hash = "sha256:d878f2a6707cc9d53a1be1414bbb419e629c3d6e67f69230217bb663e76b5087", size = 108106, upload-time = "2026-01-11T11:22:34.451Z" },
+    { url = "https://files.pythonhosted.org/packages/de/69/9aa0c6a505c2f80e519b43764f8b4ba93b5a0bbd2d9a9de6e2b24271b9a5/tomli-2.4.0-cp314-cp314t-win_amd64.whl", hash = "sha256:2add28aacc7425117ff6364fe9e06a183bb0251b03f986df0e78e974047571fd", size = 120504, upload-time = "2026-01-11T11:22:35.764Z" },
+    { url = "https://files.pythonhosted.org/packages/b3/9f/f1668c281c58cfae01482f7114a4b88d345e4c140386241a1a24dcc9e7bc/tomli-2.4.0-cp314-cp314t-win_arm64.whl", hash = "sha256:2b1e3b80e1d5e52e40e9b924ec43d81570f0e7d09d11081b797bc4692765a3d4", size = 99561, upload-time = "2026-01-11T11:22:36.624Z" },
+    { url = "https://files.pythonhosted.org/packages/23/d1/136eb2cb77520a31e1f64cbae9d33ec6df0d78bdf4160398e86eec8a8754/tomli-2.4.0-py3-none-any.whl", hash = "sha256:1f776e7d669ebceb01dee46484485f43a4048746235e683bcdffacdf1fb4785a", size = 14477, upload-time = "2026-01-11T11:22:37.446Z" },
+]
+
+[[package]]
+name = "tqdm"
+version = "4.67.2"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "colorama", marker = "sys_platform == 'win32'" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/27/89/4b0001b2dab8df0a5ee2787dcbe771de75ded01f18f1f8d53dedeea2882b/tqdm-4.67.2.tar.gz", hash = "sha256:649aac53964b2cb8dec76a14b405a4c0d13612cb8933aae547dd144eacc99653", size = 169514, upload-time = "2026-01-30T23:12:06.555Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/f5/e2/31eac96de2915cf20ccaed0225035db149dfb9165a9ed28d4b252ef3f7f7/tqdm-4.67.2-py3-none-any.whl", hash = "sha256:9a12abcbbff58b6036b2167d9d3853042b9d436fe7330f06ae047867f2f8e0a7", size = 78354, upload-time = "2026-01-30T23:12:04.368Z" },
+]
+
+[[package]]
+name = "typer-slim"
+version = "0.21.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "click" },
+    { name = "typing-extensions" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/17/d4/064570dec6358aa9049d4708e4a10407d74c99258f8b2136bb8702303f1a/typer_slim-0.21.1.tar.gz", hash = "sha256:73495dd08c2d0940d611c5a8c04e91c2a0a98600cbd4ee19192255a233b6dbfd", size = 110478, upload-time = "2026-01-06T11:21:11.176Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/c8/0a/4aca634faf693e33004796b6cee0ae2e1dba375a800c16ab8d3eff4bb800/typer_slim-0.21.1-py3-none-any.whl", hash = "sha256:6e6c31047f171ac93cc5a973c9e617dbc5ab2bddc4d0a3135dc161b4e2020e0d", size = 47444, upload-time = "2026-01-06T11:21:12.441Z" },
+]
+
+[[package]]
+name = "types-requests"
+version = "2.32.4.20260107"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "urllib3" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/0f/f3/a0663907082280664d745929205a89d41dffb29e89a50f753af7d57d0a96/types_requests-2.32.4.20260107.tar.gz", hash = "sha256:018a11ac158f801bfa84857ddec1650750e393df8a004a8a9ae2a9bec6fcb24f", size = 23165, upload-time = "2026-01-07T03:20:54.091Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/1c/12/709ea261f2bf91ef0a26a9eed20f2623227a8ed85610c1e54c5805692ecb/types_requests-2.32.4.20260107-py3-none-any.whl", hash = "sha256:b703fe72f8ce5b31ef031264fe9395cac8f46a04661a79f7ed31a80fb308730d", size = 20676, upload-time = "2026-01-07T03:20:52.929Z" },
+]
+
+[[package]]
+name = "typing-extensions"
+version = "4.15.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/72/94/1a15dd82efb362ac84269196e94cf00f187f7ed21c242792a923cdb1c61f/typing_extensions-4.15.0.tar.gz", hash = "sha256:0cea48d173cc12fa28ecabc3b837ea3cf6f38c6d1136f85cbaaf598984861466", size = 109391, upload-time = "2025-08-25T13:49:26.313Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/18/67/36e9267722cc04a6b9f15c7f3441c2363321a3ea07da7ae0c0707beb2a9c/typing_extensions-4.15.0-py3-none-any.whl", hash = "sha256:f0fa19c6845758ab08074a0cfa8b7aecb71c999ca73d62883bc25cc018c4e548", size = 44614, upload-time = "2025-08-25T13:49:24.86Z" },
+]
+
+[[package]]
+name = "typing-inspection"
+version = "0.4.2"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "typing-extensions" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/55/e3/70399cb7dd41c10ac53367ae42139cf4b1ca5f36bb3dc6c9d33acdb43655/typing_inspection-0.4.2.tar.gz", hash = "sha256:ba561c48a67c5958007083d386c3295464928b01faa735ab8547c5692e87f464", size = 75949, upload-time = "2025-10-01T02:14:41.687Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/dc/9b/47798a6c91d8bdb567fe2698fe81e0c6b7cb7ef4d13da4114b41d239f65d/typing_inspection-0.4.2-py3-none-any.whl", hash = "sha256:4ed1cacbdc298c220f1bd249ed5287caa16f34d44ef4e9c3d0cbad5b521545e7", size = 14611, upload-time = "2025-10-01T02:14:40.154Z" },
+]
+
+[[package]]
+name = "tzdata"
+version = "2025.3"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/5e/a7/c202b344c5ca7daf398f3b8a477eeb205cf3b6f32e7ec3a6bac0629ca975/tzdata-2025.3.tar.gz", hash = "sha256:de39c2ca5dc7b0344f2eba86f49d614019d29f060fc4ebc8a417896a620b56a7", size = 196772, upload-time = "2025-12-13T17:45:35.667Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/c7/b0/003792df09decd6849a5e39c28b513c06e84436a54440380862b5aeff25d/tzdata-2025.3-py2.py3-none-any.whl", hash = "sha256:06a47e5700f3081aab02b2e513160914ff0694bce9947d6b76ebd6bf57cfc5d1", size = 348521, upload-time = "2025-12-13T17:45:33.889Z" },
+]
+
+[[package]]
+name = "ujson"
+version = "5.11.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/43/d9/3f17e3c5773fb4941c68d9a37a47b1a79c9649d6c56aefbed87cc409d18a/ujson-5.11.0.tar.gz", hash = "sha256:e204ae6f909f099ba6b6b942131cee359ddda2b6e4ea39c12eb8b991fe2010e0", size = 7156583, upload-time = "2025-08-20T11:57:02.452Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/86/0c/8bf7a4fabfd01c7eed92d9b290930ce6d14910dec708e73538baa38885d1/ujson-5.11.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:446e8c11c06048611c9d29ef1237065de0af07cabdd97e6b5b527b957692ec25", size = 55248, upload-time = "2025-08-20T11:55:02.368Z" },
+    { url = "https://files.pythonhosted.org/packages/7b/2e/eeab0b8b641817031ede4f790db4c4942df44a12f44d72b3954f39c6a115/ujson-5.11.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:16ccb973b7ada0455201808ff11d48fe9c3f034a6ab5bd93b944443c88299f89", size = 53157, upload-time = "2025-08-20T11:55:04.012Z" },
+    { url = "https://files.pythonhosted.org/packages/21/1b/a4e7a41870797633423ea79618526747353fd7be9191f3acfbdee0bf264b/ujson-5.11.0-cp310-cp310-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:3134b783ab314d2298d58cda7e47e7a0f7f71fc6ade6ac86d5dbeaf4b9770fa6", size = 57657, upload-time = "2025-08-20T11:55:05.169Z" },
+    { url = "https://files.pythonhosted.org/packages/94/ae/4e0d91b8f6db7c9b76423b3649612189506d5a06ddd3b6334b6d37f77a01/ujson-5.11.0-cp310-cp310-manylinux_2_24_i686.manylinux_2_28_i686.whl", hash = "sha256:185f93ebccffebc8baf8302c869fac70dd5dd78694f3b875d03a31b03b062cdb", size = 59780, upload-time = "2025-08-20T11:55:06.325Z" },
+    { url = "https://files.pythonhosted.org/packages/b3/cc/46b124c2697ca2da7c65c4931ed3cb670646978157aa57a7a60f741c530f/ujson-5.11.0-cp310-cp310-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:d06e87eded62ff0e5f5178c916337d2262fdbc03b31688142a3433eabb6511db", size = 57307, upload-time = "2025-08-20T11:55:07.493Z" },
+    { url = "https://files.pythonhosted.org/packages/39/eb/20dd1282bc85dede2f1c62c45b4040bc4c389c80a05983515ab99771bca7/ujson-5.11.0-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:181fb5b15703a8b9370b25345d2a1fd1359f0f18776b3643d24e13ed9c036d4c", size = 1036369, upload-time = "2025-08-20T11:55:09.192Z" },
+    { url = "https://files.pythonhosted.org/packages/64/a2/80072439065d493e3a4b1fbeec991724419a1b4c232e2d1147d257cac193/ujson-5.11.0-cp310-cp310-musllinux_1_2_i686.whl", hash = "sha256:a4df61a6df0a4a8eb5b9b1ffd673429811f50b235539dac586bb7e9e91994138", size = 1195738, upload-time = "2025-08-20T11:55:11.402Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/7e/d77f9e9c039d58299c350c978e086a804d1fceae4fd4a1cc6e8d0133f838/ujson-5.11.0-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:6eff24e1abd79e0ec6d7eae651dd675ddbc41f9e43e29ef81e16b421da896915", size = 1088718, upload-time = "2025-08-20T11:55:13.297Z" },
+    { url = "https://files.pythonhosted.org/packages/ab/f1/697559d45acc849cada6b3571d53522951b1a64027400507aabc6a710178/ujson-5.11.0-cp310-cp310-win32.whl", hash = "sha256:30f607c70091483550fbd669a0b37471e5165b317d6c16e75dba2aa967608723", size = 39653, upload-time = "2025-08-20T11:55:14.869Z" },
+    { url = "https://files.pythonhosted.org/packages/86/a2/70b73a0f55abe0e6b8046d365d74230c20c5691373e6902a599b2dc79ba1/ujson-5.11.0-cp310-cp310-win_amd64.whl", hash = "sha256:3d2720e9785f84312b8e2cb0c2b87f1a0b1c53aaab3b2af3ab817d54409012e0", size = 43720, upload-time = "2025-08-20T11:55:15.897Z" },
+    { url = "https://files.pythonhosted.org/packages/1c/5f/b19104afa455630b43efcad3a24495b9c635d92aa8f2da4f30e375deb1a2/ujson-5.11.0-cp310-cp310-win_arm64.whl", hash = "sha256:85e6796631165f719084a9af00c79195d3ebf108151452fefdcb1c8bb50f0105", size = 38410, upload-time = "2025-08-20T11:55:17.556Z" },
+    { url = "https://files.pythonhosted.org/packages/da/ea/80346b826349d60ca4d612a47cdf3533694e49b45e9d1c07071bb867a184/ujson-5.11.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:d7c46cb0fe5e7056b9acb748a4c35aa1b428025853032540bb7e41f46767321f", size = 55248, upload-time = "2025-08-20T11:55:19.033Z" },
+    { url = "https://files.pythonhosted.org/packages/57/df/b53e747562c89515e18156513cc7c8ced2e5e3fd6c654acaa8752ffd7cd9/ujson-5.11.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:d8951bb7a505ab2a700e26f691bdfacf395bc7e3111e3416d325b513eea03a58", size = 53156, upload-time = "2025-08-20T11:55:20.174Z" },
+    { url = "https://files.pythonhosted.org/packages/41/b8/ab67ec8c01b8a3721fd13e5cb9d85ab2a6066a3a5e9148d661a6870d6293/ujson-5.11.0-cp311-cp311-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:952c0be400229940248c0f5356514123d428cba1946af6fa2bbd7503395fef26", size = 57657, upload-time = "2025-08-20T11:55:21.296Z" },
+    { url = "https://files.pythonhosted.org/packages/7b/c7/fb84f27cd80a2c7e2d3c6012367aecade0da936790429801803fa8d4bffc/ujson-5.11.0-cp311-cp311-manylinux_2_24_i686.manylinux_2_28_i686.whl", hash = "sha256:94fcae844f1e302f6f8095c5d1c45a2f0bfb928cccf9f1b99e3ace634b980a2a", size = 59779, upload-time = "2025-08-20T11:55:22.772Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/7c/48706f7c1e917ecb97ddcfb7b1d756040b86ed38290e28579d63bd3fcc48/ujson-5.11.0-cp311-cp311-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:7e0ec1646db172beb8d3df4c32a9d78015e671d2000af548252769e33079d9a6", size = 57284, upload-time = "2025-08-20T11:55:24.01Z" },
+    { url = "https://files.pythonhosted.org/packages/ec/ce/48877c6eb4afddfd6bd1db6be34456538c07ca2d6ed233d3f6c6efc2efe8/ujson-5.11.0-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:da473b23e3a54448b008d33f742bcd6d5fb2a897e42d1fc6e7bf306ea5d18b1b", size = 1036395, upload-time = "2025-08-20T11:55:25.725Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/7a/2c20dc97ad70cd7c31ad0596ba8e2cf8794d77191ba4d1e0bded69865477/ujson-5.11.0-cp311-cp311-musllinux_1_2_i686.whl", hash = "sha256:aa6b3d4f1c0d3f82930f4cbd7fe46d905a4a9205a7c13279789c1263faf06dba", size = 1195731, upload-time = "2025-08-20T11:55:27.915Z" },
+    { url = "https://files.pythonhosted.org/packages/15/f5/ca454f2f6a2c840394b6f162fff2801450803f4ff56c7af8ce37640b8a2a/ujson-5.11.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:4843f3ab4fe1cc596bb7e02228ef4c25d35b4bb0809d6a260852a4bfcab37ba3", size = 1088710, upload-time = "2025-08-20T11:55:29.426Z" },
+    { url = "https://files.pythonhosted.org/packages/fe/d3/9ba310e07969bc9906eb7548731e33a0f448b122ad9705fed699c9b29345/ujson-5.11.0-cp311-cp311-win32.whl", hash = "sha256:e979fbc469a7f77f04ec2f4e853ba00c441bf2b06720aa259f0f720561335e34", size = 39648, upload-time = "2025-08-20T11:55:31.194Z" },
+    { url = "https://files.pythonhosted.org/packages/57/f7/da05b4a8819f1360be9e71fb20182f0bb3ec611a36c3f213f4d20709e099/ujson-5.11.0-cp311-cp311-win_amd64.whl", hash = "sha256:683f57f0dd3acdd7d9aff1de0528d603aafcb0e6d126e3dc7ce8b020a28f5d01", size = 43717, upload-time = "2025-08-20T11:55:32.241Z" },
+    { url = "https://files.pythonhosted.org/packages/9a/cc/f3f9ac0f24f00a623a48d97dc3814df5c2dc368cfb00031aa4141527a24b/ujson-5.11.0-cp311-cp311-win_arm64.whl", hash = "sha256:7855ccea3f8dad5e66d8445d754fc1cf80265a4272b5f8059ebc7ec29b8d0835", size = 38402, upload-time = "2025-08-20T11:55:33.641Z" },
+    { url = "https://files.pythonhosted.org/packages/b9/ef/a9cb1fce38f699123ff012161599fb9f2ff3f8d482b4b18c43a2dc35073f/ujson-5.11.0-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:7895f0d2d53bd6aea11743bd56e3cb82d729980636cd0ed9b89418bf66591702", size = 55434, upload-time = "2025-08-20T11:55:34.987Z" },
+    { url = "https://files.pythonhosted.org/packages/b1/05/dba51a00eb30bd947791b173766cbed3492269c150a7771d2750000c965f/ujson-5.11.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:12b5e7e22a1fe01058000d1b317d3b65cc3daf61bd2ea7a2b76721fe160fa74d", size = 53190, upload-time = "2025-08-20T11:55:36.384Z" },
+    { url = "https://files.pythonhosted.org/packages/03/3c/fd11a224f73fbffa299fb9644e425f38b38b30231f7923a088dd513aabb4/ujson-5.11.0-cp312-cp312-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:0180a480a7d099082501cad1fe85252e4d4bf926b40960fb3d9e87a3a6fbbc80", size = 57600, upload-time = "2025-08-20T11:55:37.692Z" },
+    { url = "https://files.pythonhosted.org/packages/55/b9/405103cae24899df688a3431c776e00528bd4799e7d68820e7ebcf824f92/ujson-5.11.0-cp312-cp312-manylinux_2_24_i686.manylinux_2_28_i686.whl", hash = "sha256:fa79fdb47701942c2132a9dd2297a1a85941d966d8c87bfd9e29b0cf423f26cc", size = 59791, upload-time = "2025-08-20T11:55:38.877Z" },
+    { url = "https://files.pythonhosted.org/packages/17/7b/2dcbc2bbfdbf68f2368fb21ab0f6735e872290bb604c75f6e06b81edcb3f/ujson-5.11.0-cp312-cp312-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:8254e858437c00f17cb72e7a644fc42dad0ebb21ea981b71df6e84b1072aaa7c", size = 57356, upload-time = "2025-08-20T11:55:40.036Z" },
+    { url = "https://files.pythonhosted.org/packages/d1/71/fea2ca18986a366c750767b694430d5ded6b20b6985fddca72f74af38a4c/ujson-5.11.0-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:1aa8a2ab482f09f6c10fba37112af5f957689a79ea598399c85009f2f29898b5", size = 1036313, upload-time = "2025-08-20T11:55:41.408Z" },
+    { url = "https://files.pythonhosted.org/packages/a3/bb/d4220bd7532eac6288d8115db51710fa2d7d271250797b0bfba9f1e755af/ujson-5.11.0-cp312-cp312-musllinux_1_2_i686.whl", hash = "sha256:a638425d3c6eed0318df663df44480f4a40dc87cc7c6da44d221418312f6413b", size = 1195782, upload-time = "2025-08-20T11:55:43.357Z" },
+    { url = "https://files.pythonhosted.org/packages/80/47/226e540aa38878ce1194454385701d82df538ccb5ff8db2cf1641dde849a/ujson-5.11.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:7e3cff632c1d78023b15f7e3a81c3745cd3f94c044d1e8fa8efbd6b161997bbc", size = 1088817, upload-time = "2025-08-20T11:55:45.262Z" },
+    { url = "https://files.pythonhosted.org/packages/7e/81/546042f0b23c9040d61d46ea5ca76f0cc5e0d399180ddfb2ae976ebff5b5/ujson-5.11.0-cp312-cp312-win32.whl", hash = "sha256:be6b0eaf92cae8cdee4d4c9e074bde43ef1c590ed5ba037ea26c9632fb479c88", size = 39757, upload-time = "2025-08-20T11:55:46.522Z" },
+    { url = "https://files.pythonhosted.org/packages/44/1b/27c05dc8c9728f44875d74b5bfa948ce91f6c33349232619279f35c6e817/ujson-5.11.0-cp312-cp312-win_amd64.whl", hash = "sha256:b7b136cc6abc7619124fd897ef75f8e63105298b5ca9bdf43ebd0e1fa0ee105f", size = 43859, upload-time = "2025-08-20T11:55:47.987Z" },
+    { url = "https://files.pythonhosted.org/packages/22/2d/37b6557c97c3409c202c838aa9c960ca3896843b4295c4b7bb2bbd260664/ujson-5.11.0-cp312-cp312-win_arm64.whl", hash = "sha256:6cd2df62f24c506a0ba322d5e4fe4466d47a9467b57e881ee15a31f7ecf68ff6", size = 38361, upload-time = "2025-08-20T11:55:49.122Z" },
+    { url = "https://files.pythonhosted.org/packages/1c/ec/2de9dd371d52c377abc05d2b725645326c4562fc87296a8907c7bcdf2db7/ujson-5.11.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:109f59885041b14ee9569bf0bb3f98579c3fa0652317b355669939e5fc5ede53", size = 55435, upload-time = "2025-08-20T11:55:50.243Z" },
+    { url = "https://files.pythonhosted.org/packages/5b/a4/f611f816eac3a581d8a4372f6967c3ed41eddbae4008d1d77f223f1a4e0a/ujson-5.11.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:a31c6b8004438e8c20fc55ac1c0e07dad42941db24176fe9acf2815971f8e752", size = 53193, upload-time = "2025-08-20T11:55:51.373Z" },
+    { url = "https://files.pythonhosted.org/packages/e9/c5/c161940967184de96f5cbbbcce45b562a4bf851d60f4c677704b1770136d/ujson-5.11.0-cp313-cp313-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:78c684fb21255b9b90320ba7e199780f653e03f6c2528663768965f4126a5b50", size = 57603, upload-time = "2025-08-20T11:55:52.583Z" },
+    { url = "https://files.pythonhosted.org/packages/2b/d6/c7b2444238f5b2e2d0e3dab300b9ddc3606e4b1f0e4bed5a48157cebc792/ujson-5.11.0-cp313-cp313-manylinux_2_24_i686.manylinux_2_28_i686.whl", hash = "sha256:4c9f5d6a27d035dd90a146f7761c2272cf7103de5127c9ab9c4cd39ea61e878a", size = 59794, upload-time = "2025-08-20T11:55:53.69Z" },
+    { url = "https://files.pythonhosted.org/packages/fe/a3/292551f936d3d02d9af148f53e1bc04306b00a7cf1fcbb86fa0d1c887242/ujson-5.11.0-cp313-cp313-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:837da4d27fed5fdc1b630bd18f519744b23a0b5ada1bbde1a36ba463f2900c03", size = 57363, upload-time = "2025-08-20T11:55:54.843Z" },
+    { url = "https://files.pythonhosted.org/packages/90/a6/82cfa70448831b1a9e73f882225980b5c689bf539ec6400b31656a60ea46/ujson-5.11.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:787aff4a84da301b7f3bac09bc696e2e5670df829c6f8ecf39916b4e7e24e701", size = 1036311, upload-time = "2025-08-20T11:55:56.197Z" },
+    { url = "https://files.pythonhosted.org/packages/84/5c/96e2266be50f21e9b27acaee8ca8f23ea0b85cb998c33d4f53147687839b/ujson-5.11.0-cp313-cp313-musllinux_1_2_i686.whl", hash = "sha256:6dd703c3e86dc6f7044c5ac0b3ae079ed96bf297974598116aa5fb7f655c3a60", size = 1195783, upload-time = "2025-08-20T11:55:58.081Z" },
+    { url = "https://files.pythonhosted.org/packages/8d/20/78abe3d808cf3bb3e76f71fca46cd208317bf461c905d79f0d26b9df20f1/ujson-5.11.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:3772e4fe6b0c1e025ba3c50841a0ca4786825a4894c8411bf8d3afe3a8061328", size = 1088822, upload-time = "2025-08-20T11:55:59.469Z" },
+    { url = "https://files.pythonhosted.org/packages/d8/50/8856e24bec5e2fc7f775d867aeb7a3f137359356200ac44658f1f2c834b2/ujson-5.11.0-cp313-cp313-win32.whl", hash = "sha256:8fa2af7c1459204b7a42e98263b069bd535ea0cd978b4d6982f35af5a04a4241", size = 39753, upload-time = "2025-08-20T11:56:01.345Z" },
+    { url = "https://files.pythonhosted.org/packages/5b/d8/1baee0f4179a4d0f5ce086832147b6cc9b7731c24ca08e14a3fdb8d39c32/ujson-5.11.0-cp313-cp313-win_amd64.whl", hash = "sha256:34032aeca4510a7c7102bd5933f59a37f63891f30a0706fb46487ab6f0edf8f0", size = 43866, upload-time = "2025-08-20T11:56:02.552Z" },
+    { url = "https://files.pythonhosted.org/packages/a9/8c/6d85ef5be82c6d66adced3ec5ef23353ed710a11f70b0b6a836878396334/ujson-5.11.0-cp313-cp313-win_arm64.whl", hash = "sha256:ce076f2df2e1aa62b685086fbad67f2b1d3048369664b4cdccc50707325401f9", size = 38363, upload-time = "2025-08-20T11:56:03.688Z" },
+    { url = "https://files.pythonhosted.org/packages/28/08/4518146f4984d112764b1dfa6fb7bad691c44a401adadaa5e23ccd930053/ujson-5.11.0-cp314-cp314-macosx_10_13_x86_64.whl", hash = "sha256:65724738c73645db88f70ba1f2e6fb678f913281804d5da2fd02c8c5839af302", size = 55462, upload-time = "2025-08-20T11:56:04.873Z" },
+    { url = "https://files.pythonhosted.org/packages/29/37/2107b9a62168867a692654d8766b81bd2fd1e1ba13e2ec90555861e02b0c/ujson-5.11.0-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:29113c003ca33ab71b1b480bde952fbab2a0b6b03a4ee4c3d71687cdcbd1a29d", size = 53246, upload-time = "2025-08-20T11:56:06.054Z" },
+    { url = "https://files.pythonhosted.org/packages/9b/f8/25583c70f83788edbe3ca62ce6c1b79eff465d78dec5eb2b2b56b3e98b33/ujson-5.11.0-cp314-cp314-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:c44c703842024d796b4c78542a6fcd5c3cb948b9fc2a73ee65b9c86a22ee3638", size = 57631, upload-time = "2025-08-20T11:56:07.374Z" },
+    { url = "https://files.pythonhosted.org/packages/ed/ca/19b3a632933a09d696f10dc1b0dfa1d692e65ad507d12340116ce4f67967/ujson-5.11.0-cp314-cp314-manylinux_2_24_i686.manylinux_2_28_i686.whl", hash = "sha256:e750c436fb90edf85585f5c62a35b35082502383840962c6983403d1bd96a02c", size = 59877, upload-time = "2025-08-20T11:56:08.534Z" },
+    { url = "https://files.pythonhosted.org/packages/55/7a/4572af5324ad4b2bfdd2321e898a527050290147b4ea337a79a0e4e87ec7/ujson-5.11.0-cp314-cp314-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:f278b31a7c52eb0947b2db55a5133fbc46b6f0ef49972cd1a80843b72e135aba", size = 57363, upload-time = "2025-08-20T11:56:09.758Z" },
+    { url = "https://files.pythonhosted.org/packages/7b/71/a2b8c19cf4e1efe53cf439cdf7198ac60ae15471d2f1040b490c1f0f831f/ujson-5.11.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:ab2cb8351d976e788669c8281465d44d4e94413718af497b4e7342d7b2f78018", size = 1036394, upload-time = "2025-08-20T11:56:11.168Z" },
+    { url = "https://files.pythonhosted.org/packages/7a/3e/7b98668cba3bb3735929c31b999b374ebc02c19dfa98dfebaeeb5c8597ca/ujson-5.11.0-cp314-cp314-musllinux_1_2_i686.whl", hash = "sha256:090b4d11b380ae25453100b722d0609d5051ffe98f80ec52853ccf8249dfd840", size = 1195837, upload-time = "2025-08-20T11:56:12.6Z" },
+    { url = "https://files.pythonhosted.org/packages/a1/ea/8870f208c20b43571a5c409ebb2fe9b9dba5f494e9e60f9314ac01ea8f78/ujson-5.11.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:80017e870d882d5517d28995b62e4e518a894f932f1e242cbc802a2fd64d365c", size = 1088837, upload-time = "2025-08-20T11:56:14.15Z" },
+    { url = "https://files.pythonhosted.org/packages/63/b6/c0e6607e37fa47929920a685a968c6b990a802dec65e9c5181e97845985d/ujson-5.11.0-cp314-cp314-win32.whl", hash = "sha256:1d663b96eb34c93392e9caae19c099ec4133ba21654b081956613327f0e973ac", size = 41022, upload-time = "2025-08-20T11:56:15.509Z" },
+    { url = "https://files.pythonhosted.org/packages/4e/56/f4fe86b4c9000affd63e9219e59b222dc48b01c534533093e798bf617a7e/ujson-5.11.0-cp314-cp314-win_amd64.whl", hash = "sha256:849e65b696f0d242833f1df4182096cedc50d414215d1371fca85c541fbff629", size = 45111, upload-time = "2025-08-20T11:56:16.597Z" },
+    { url = "https://files.pythonhosted.org/packages/0a/f3/669437f0280308db4783b12a6d88c00730b394327d8334cc7a32ef218e64/ujson-5.11.0-cp314-cp314-win_arm64.whl", hash = "sha256:e73df8648c9470af2b6a6bf5250d4744ad2cf3d774dcf8c6e31f018bdd04d764", size = 39682, upload-time = "2025-08-20T11:56:17.763Z" },
+    { url = "https://files.pythonhosted.org/packages/6e/cd/e9809b064a89fe5c4184649adeb13c1b98652db3f8518980b04227358574/ujson-5.11.0-cp314-cp314t-macosx_10_13_x86_64.whl", hash = "sha256:de6e88f62796372fba1de973c11138f197d3e0e1d80bcb2b8aae1e826096d433", size = 55759, upload-time = "2025-08-20T11:56:18.882Z" },
+    { url = "https://files.pythonhosted.org/packages/1b/be/ae26a6321179ebbb3a2e2685b9007c71bcda41ad7a77bbbe164005e956fc/ujson-5.11.0-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:49e56ef8066f11b80d620985ae36869a3ff7e4b74c3b6129182ec5d1df0255f3", size = 53634, upload-time = "2025-08-20T11:56:20.012Z" },
+    { url = "https://files.pythonhosted.org/packages/ae/e9/fb4a220ee6939db099f4cfeeae796ecb91e7584ad4d445d4ca7f994a9135/ujson-5.11.0-cp314-cp314t-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:1a325fd2c3a056cf6c8e023f74a0c478dd282a93141356ae7f16d5309f5ff823", size = 58547, upload-time = "2025-08-20T11:56:21.175Z" },
+    { url = "https://files.pythonhosted.org/packages/bd/f8/fc4b952b8f5fea09ea3397a0bd0ad019e474b204cabcb947cead5d4d1ffc/ujson-5.11.0-cp314-cp314t-manylinux_2_24_i686.manylinux_2_28_i686.whl", hash = "sha256:a0af6574fc1d9d53f4ff371f58c96673e6d988ed2b5bf666a6143c782fa007e9", size = 60489, upload-time = "2025-08-20T11:56:22.342Z" },
+    { url = "https://files.pythonhosted.org/packages/2e/e5/af5491dfda4f8b77e24cf3da68ee0d1552f99a13e5c622f4cef1380925c3/ujson-5.11.0-cp314-cp314t-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:10f29e71ecf4ecd93a6610bd8efa8e7b6467454a363c3d6416db65de883eb076", size = 58035, upload-time = "2025-08-20T11:56:23.92Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/09/0945349dd41f25cc8c38d78ace49f14c5052c5bbb7257d2f466fa7bdb533/ujson-5.11.0-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:1a0a9b76a89827a592656fe12e000cf4f12da9692f51a841a4a07aa4c7ecc41c", size = 1037212, upload-time = "2025-08-20T11:56:25.274Z" },
+    { url = "https://files.pythonhosted.org/packages/49/44/8e04496acb3d5a1cbee3a54828d9652f67a37523efa3d3b18a347339680a/ujson-5.11.0-cp314-cp314t-musllinux_1_2_i686.whl", hash = "sha256:b16930f6a0753cdc7d637b33b4e8f10d5e351e1fb83872ba6375f1e87be39746", size = 1196500, upload-time = "2025-08-20T11:56:27.517Z" },
+    { url = "https://files.pythonhosted.org/packages/64/ae/4bc825860d679a0f208a19af2f39206dfd804ace2403330fdc3170334a2f/ujson-5.11.0-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:04c41afc195fd477a59db3a84d5b83a871bd648ef371cf8c6f43072d89144eef", size = 1089487, upload-time = "2025-08-20T11:56:29.07Z" },
+    { url = "https://files.pythonhosted.org/packages/30/ed/5a057199fb0a5deabe0957073a1c1c1c02a3e99476cd03daee98ea21fa57/ujson-5.11.0-cp314-cp314t-win32.whl", hash = "sha256:aa6d7a5e09217ff93234e050e3e380da62b084e26b9f2e277d2606406a2fc2e5", size = 41859, upload-time = "2025-08-20T11:56:30.495Z" },
+    { url = "https://files.pythonhosted.org/packages/aa/03/b19c6176bdf1dc13ed84b886e99677a52764861b6cc023d5e7b6ebda249d/ujson-5.11.0-cp314-cp314t-win_amd64.whl", hash = "sha256:48055e1061c1bb1f79e75b4ac39e821f3f35a9b82de17fce92c3140149009bec", size = 46183, upload-time = "2025-08-20T11:56:31.574Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/ca/a0413a3874b2dc1708b8796ca895bf363292f9c70b2e8ca482b7dbc0259d/ujson-5.11.0-cp314-cp314t-win_arm64.whl", hash = "sha256:1194b943e951092db611011cb8dbdb6cf94a3b816ed07906e14d3bc6ce0e90ab", size = 40264, upload-time = "2025-08-20T11:56:32.773Z" },
+    { url = "https://files.pythonhosted.org/packages/50/17/30275aa2933430d8c0c4ead951cc4fdb922f575a349aa0b48a6f35449e97/ujson-5.11.0-pp311-pypy311_pp73-macosx_10_15_x86_64.whl", hash = "sha256:abae0fb58cc820092a0e9e8ba0051ac4583958495bfa5262a12f628249e3b362", size = 51206, upload-time = "2025-08-20T11:56:48.797Z" },
+    { url = "https://files.pythonhosted.org/packages/c3/15/42b3924258eac2551f8f33fa4e35da20a06a53857ccf3d4deb5e5d7c0b6c/ujson-5.11.0-pp311-pypy311_pp73-macosx_11_0_arm64.whl", hash = "sha256:fac6c0649d6b7c3682a0a6e18d3de6857977378dce8d419f57a0b20e3d775b39", size = 48907, upload-time = "2025-08-20T11:56:50.136Z" },
+    { url = "https://files.pythonhosted.org/packages/94/7e/0519ff7955aba581d1fe1fb1ca0e452471250455d182f686db5ac9e46119/ujson-5.11.0-pp311-pypy311_pp73-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:4b42c115c7c6012506e8168315150d1e3f76e7ba0f4f95616f4ee599a1372bbc", size = 50319, upload-time = "2025-08-20T11:56:51.63Z" },
+    { url = "https://files.pythonhosted.org/packages/74/cf/209d90506b7d6c5873f82c5a226d7aad1a1da153364e9ebf61eff0740c33/ujson-5.11.0-pp311-pypy311_pp73-manylinux_2_24_i686.manylinux_2_28_i686.whl", hash = "sha256:86baf341d90b566d61a394869ce77188cc8668f76d7bb2c311d77a00f4bdf844", size = 56584, upload-time = "2025-08-20T11:56:52.89Z" },
+    { url = "https://files.pythonhosted.org/packages/e9/97/bd939bb76943cb0e1d2b692d7e68629f51c711ef60425fa5bb6968037ecd/ujson-5.11.0-pp311-pypy311_pp73-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:4598bf3965fc1a936bd84034312bcbe00ba87880ef1ee33e33c1e88f2c398b49", size = 51588, upload-time = "2025-08-20T11:56:54.054Z" },
+    { url = "https://files.pythonhosted.org/packages/52/5b/8c5e33228f7f83f05719964db59f3f9f276d272dc43752fa3bbf0df53e7b/ujson-5.11.0-pp311-pypy311_pp73-win_amd64.whl", hash = "sha256:416389ec19ef5f2013592f791486bef712ebce0cd59299bf9df1ba40bb2f6e04", size = 43835, upload-time = "2025-08-20T11:56:55.237Z" },
+]
+
+[[package]]
+name = "urllib3"
+version = "2.6.3"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/c7/24/5f1b3bdffd70275f6661c76461e25f024d5a38a46f04aaca912426a2b1d3/urllib3-2.6.3.tar.gz", hash = "sha256:1b62b6884944a57dbe321509ab94fd4d3b307075e0c2eae991ac71ee15ad38ed", size = 435556, upload-time = "2026-01-07T16:24:43.925Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/39/08/aaaad47bc4e9dc8c725e68f9d04865dbcb2052843ff09c97b08904852d84/urllib3-2.6.3-py3-none-any.whl", hash = "sha256:bf272323e553dfb2e87d9bfd225ca7b0f467b919d7bbd355436d3fd37cb0acd4", size = 131584, upload-time = "2026-01-07T16:24:42.685Z" },
+]
+
+[[package]]
+name = "uuid-utils"
+version = "0.14.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/57/7c/3a926e847516e67bc6838634f2e54e24381105b4e80f9338dc35cca0086b/uuid_utils-0.14.0.tar.gz", hash = "sha256:fc5bac21e9933ea6c590433c11aa54aaca599f690c08069e364eb13a12f670b4", size = 22072, upload-time = "2026-01-20T20:37:15.729Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/a7/42/42d003f4a99ddc901eef2fd41acb3694163835e037fb6dde79ad68a72342/uuid_utils-0.14.0-cp39-abi3-macosx_10_12_x86_64.macosx_11_0_arm64.macosx_10_12_universal2.whl", hash = "sha256:f6695c0bed8b18a904321e115afe73b34444bc8451d0ce3244a1ec3b84deb0e5", size = 601786, upload-time = "2026-01-20T20:37:09.843Z" },
+    { url = "https://files.pythonhosted.org/packages/96/e6/775dfb91f74b18f7207e3201eb31ee666d286579990dc69dd50db2d92813/uuid_utils-0.14.0-cp39-abi3-macosx_10_12_x86_64.whl", hash = "sha256:4f0a730bbf2d8bb2c11b93e1005e91769f2f533fa1125ed1f00fd15b6fcc732b", size = 303943, upload-time = "2026-01-20T20:37:18.767Z" },
+    { url = "https://files.pythonhosted.org/packages/17/82/ea5f5e85560b08a1f30cdc65f75e76494dc7aba9773f679e7eaa27370229/uuid_utils-0.14.0-cp39-abi3-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:40ce3fd1a4fdedae618fc3edc8faf91897012469169d600133470f49fd699ed3", size = 340467, upload-time = "2026-01-20T20:37:11.794Z" },
+    { url = "https://files.pythonhosted.org/packages/ca/33/54b06415767f4569882e99b6470c6c8eeb97422686a6d432464f9967fd91/uuid_utils-0.14.0-cp39-abi3-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:09ae4a98416a440e78f7d9543d11b11cae4bab538b7ed94ec5da5221481748f2", size = 346333, upload-time = "2026-01-20T20:37:12.818Z" },
+    { url = "https://files.pythonhosted.org/packages/cb/10/a6bce636b8f95e65dc84bf4a58ce8205b8e0a2a300a38cdbc83a3f763d27/uuid_utils-0.14.0-cp39-abi3-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:971e8c26b90d8ae727e7f2ac3ee23e265971d448b3672882f2eb44828b2b8c3e", size = 470859, upload-time = "2026-01-20T20:37:01.512Z" },
+    { url = "https://files.pythonhosted.org/packages/8a/27/84121c51ea72f013f0e03d0886bcdfa96b31c9b83c98300a7bd5cc4fa191/uuid_utils-0.14.0-cp39-abi3-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:d5cde1fa82804a8f9d2907b7aec2009d440062c63f04abbdb825fce717a5e860", size = 341988, upload-time = "2026-01-20T20:37:22.881Z" },
+    { url = "https://files.pythonhosted.org/packages/90/a4/01c1c7af5e6a44f20b40183e8dac37d6ed83e7dc9e8df85370a15959b804/uuid_utils-0.14.0-cp39-abi3-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:c7343862a2359e0bd48a7f3dfb5105877a1728677818bb694d9f40703264a2db", size = 365784, upload-time = "2026-01-20T20:37:10.808Z" },
+    { url = "https://files.pythonhosted.org/packages/04/f0/65ee43ec617b8b6b1bf2a5aecd56a069a08cca3d9340c1de86024331bde3/uuid_utils-0.14.0-cp39-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:c51e4818fdb08ccec12dc7083a01f49507b4608770a0ab22368001685d59381b", size = 523750, upload-time = "2026-01-20T20:37:06.152Z" },
+    { url = "https://files.pythonhosted.org/packages/95/d3/6bf503e3f135a5dfe705a65e6f89f19bccd55ac3fb16cb5d3ec5ba5388b8/uuid_utils-0.14.0-cp39-abi3-musllinux_1_2_armv7l.whl", hash = "sha256:181bbcccb6f93d80a8504b5bd47b311a1c31395139596edbc47b154b0685b533", size = 615818, upload-time = "2026-01-20T20:37:21.816Z" },
+    { url = "https://files.pythonhosted.org/packages/df/6c/99937dd78d07f73bba831c8dc9469dfe4696539eba2fc269ae1b92752f9e/uuid_utils-0.14.0-cp39-abi3-musllinux_1_2_i686.whl", hash = "sha256:5c8ae96101c3524ba8dbf762b6f05e9e9d896544786c503a727c5bf5cb9af1a7", size = 580831, upload-time = "2026-01-20T20:37:19.691Z" },
+    { url = "https://files.pythonhosted.org/packages/44/fa/bbc9e2c25abd09a293b9b097a0d8fc16acd6a92854f0ec080f1ea7ad8bb3/uuid_utils-0.14.0-cp39-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:00ac3c6edfdaff7e1eed041f4800ae09a3361287be780d7610a90fdcde9befdc", size = 546333, upload-time = "2026-01-20T20:37:03.117Z" },
+    { url = "https://files.pythonhosted.org/packages/e7/9b/e5e99b324b1b5f0c62882230455786df0bc66f67eff3b452447e703f45d2/uuid_utils-0.14.0-cp39-abi3-win32.whl", hash = "sha256:ec2fd80adf8e0e6589d40699e6f6df94c93edcc16dd999be0438dd007c77b151", size = 177319, upload-time = "2026-01-20T20:37:04.208Z" },
+    { url = "https://files.pythonhosted.org/packages/d3/28/2c7d417ea483b6ff7820c948678fdf2ac98899dc7e43bb15852faa95acaf/uuid_utils-0.14.0-cp39-abi3-win_amd64.whl", hash = "sha256:efe881eb43a5504fad922644cb93d725fd8a6a6d949bd5a4b4b7d1a1587c7fd1", size = 182566, upload-time = "2026-01-20T20:37:16.868Z" },
+    { url = "https://files.pythonhosted.org/packages/b8/86/49e4bdda28e962fbd7266684171ee29b3d92019116971d58783e51770745/uuid_utils-0.14.0-cp39-abi3-win_arm64.whl", hash = "sha256:32b372b8fd4ebd44d3a219e093fe981af4afdeda2994ee7db208ab065cfcd080", size = 182809, upload-time = "2026-01-20T20:37:05.139Z" },
+    { url = "https://files.pythonhosted.org/packages/f1/03/1f1146e32e94d1f260dfabc81e1649102083303fb4ad549775c943425d9a/uuid_utils-0.14.0-pp311-pypy311_pp73-macosx_10_12_x86_64.macosx_11_0_arm64.macosx_10_12_universal2.whl", hash = "sha256:762e8d67992ac4d2454e24a141a1c82142b5bde10409818c62adbe9924ebc86d", size = 587430, upload-time = "2026-01-20T20:37:24.998Z" },
+    { url = "https://files.pythonhosted.org/packages/87/ba/d5a7469362594d885fd9219fe9e851efbe65101d3ef1ef25ea321d7ce841/uuid_utils-0.14.0-pp311-pypy311_pp73-macosx_10_12_x86_64.whl", hash = "sha256:40be5bf0b13aa849d9062abc86c198be6a25ff35316ce0b89fc25f3bac6d525e", size = 298106, upload-time = "2026-01-20T20:37:23.896Z" },
+    { url = "https://files.pythonhosted.org/packages/8a/11/3dafb2a5502586f59fd49e93f5802cd5face82921b3a0f3abb5f357cb879/uuid_utils-0.14.0-pp311-pypy311_pp73-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:191a90a6f3940d1b7322b6e6cceff4dd533c943659e0a15f788674407856a515", size = 333423, upload-time = "2026-01-20T20:37:17.828Z" },
+    { url = "https://files.pythonhosted.org/packages/7c/f2/c8987663f0cdcf4d717a36d85b5db2a5589df0a4e129aa10f16f4380ef48/uuid_utils-0.14.0-pp311-pypy311_pp73-manylinux_2_17_armv7l.manylinux2014_armv7l.whl", hash = "sha256:4aa4525f4ad82f9d9c842f9a3703f1539c1808affbaec07bb1b842f6b8b96aa5", size = 338659, upload-time = "2026-01-20T20:37:14.286Z" },
+    { url = "https://files.pythonhosted.org/packages/d1/c8/929d81665d83f0b2ffaecb8e66c3091a50f62c7cb5b65e678bd75a96684e/uuid_utils-0.14.0-pp311-pypy311_pp73-manylinux_2_17_ppc64le.manylinux2014_ppc64le.whl", hash = "sha256:cdbd82ff20147461caefc375551595ecf77ebb384e46267f128aca45a0f2cdfc", size = 467029, upload-time = "2026-01-20T20:37:08.277Z" },
+    { url = "https://files.pythonhosted.org/packages/8e/a0/27d7daa1bfed7163f4ccaf52d7d2f4ad7bb1002a85b45077938b91ee584f/uuid_utils-0.14.0-pp311-pypy311_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:eff57e8a5d540006ce73cf0841a643d445afe78ba12e75ac53a95ca2924a56be", size = 333298, upload-time = "2026-01-20T20:37:07.271Z" },
+    { url = "https://files.pythonhosted.org/packages/63/d4/acad86ce012b42ce18a12f31ee2aa3cbeeb98664f865f05f68c882945913/uuid_utils-0.14.0-pp311-pypy311_pp73-manylinux_2_5_i686.manylinux1_i686.whl", hash = "sha256:3fd9112ca96978361201e669729784f26c71fecc9c13a7f8a07162c31bd4d1e2", size = 359217, upload-time = "2026-01-20T20:36:59.687Z" },
+]
+
+[[package]]
+name = "uvicorn"
+version = "0.40.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "click" },
+    { name = "h11" },
+    { name = "typing-extensions", marker = "python_full_version < '3.11'" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/c3/d1/8f3c683c9561a4e6689dd3b1d345c815f10f86acd044ee1fb9a4dcd0b8c5/uvicorn-0.40.0.tar.gz", hash = "sha256:839676675e87e73694518b5574fd0f24c9d97b46bea16df7b8c05ea1a51071ea", size = 81761, upload-time = "2025-12-21T14:16:22.45Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/3d/d8/2083a1daa7439a66f3a48589a57d576aa117726762618f6bb09fe3798796/uvicorn-0.40.0-py3-none-any.whl", hash = "sha256:c6c8f55bc8bf13eb6fa9ff87ad62308bbbc33d0b67f84293151efe87e0d5f2ee", size = 68502, upload-time = "2025-12-21T14:16:21.041Z" },
+]
+
+[[package]]
+name = "vcrpy"
+version = "8.1.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "pyyaml" },
+    { name = "wrapt" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/b3/07/bcfd5ebd7cb308026ab78a353e091bd699593358be49197d39d004e5ad83/vcrpy-8.1.1.tar.gz", hash = "sha256:58e3053e33b423f3594031cb758c3f4d1df931307f1e67928e30cf352df7709f", size = 85770, upload-time = "2026-01-04T19:22:03.886Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/3a/d7/f79b05a5d728f8786876a7d75dfb0c5cae27e428081b2d60152fb52f155f/vcrpy-8.1.1-py3-none-any.whl", hash = "sha256:2d16f31ad56493efb6165182dd99767207031b0da3f68b18f975545ede8ac4b9", size = 42445, upload-time = "2026-01-04T19:22:02.532Z" },
+]
+
+[[package]]
+name = "watchdog"
+version = "6.0.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/db/7d/7f3d619e951c88ed75c6037b246ddcf2d322812ee8ea189be89511721d54/watchdog-6.0.0.tar.gz", hash = "sha256:9ddf7c82fda3ae8e24decda1338ede66e1c99883db93711d8fb941eaa2d8c282", size = 131220, upload-time = "2024-11-01T14:07:13.037Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/0c/56/90994d789c61df619bfc5ce2ecdabd5eeff564e1eb47512bd01b5e019569/watchdog-6.0.0-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:d1cdb490583ebd691c012b3d6dae011000fe42edb7a82ece80965b42abd61f26", size = 96390, upload-time = "2024-11-01T14:06:24.793Z" },
+    { url = "https://files.pythonhosted.org/packages/55/46/9a67ee697342ddf3c6daa97e3a587a56d6c4052f881ed926a849fcf7371c/watchdog-6.0.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:bc64ab3bdb6a04d69d4023b29422170b74681784ffb9463ed4870cf2f3e66112", size = 88389, upload-time = "2024-11-01T14:06:27.112Z" },
+    { url = "https://files.pythonhosted.org/packages/44/65/91b0985747c52064d8701e1075eb96f8c40a79df889e59a399453adfb882/watchdog-6.0.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:c897ac1b55c5a1461e16dae288d22bb2e412ba9807df8397a635d88f671d36c3", size = 89020, upload-time = "2024-11-01T14:06:29.876Z" },
+    { url = "https://files.pythonhosted.org/packages/e0/24/d9be5cd6642a6aa68352ded4b4b10fb0d7889cb7f45814fb92cecd35f101/watchdog-6.0.0-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:6eb11feb5a0d452ee41f824e271ca311a09e250441c262ca2fd7ebcf2461a06c", size = 96393, upload-time = "2024-11-01T14:06:31.756Z" },
+    { url = "https://files.pythonhosted.org/packages/63/7a/6013b0d8dbc56adca7fdd4f0beed381c59f6752341b12fa0886fa7afc78b/watchdog-6.0.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:ef810fbf7b781a5a593894e4f439773830bdecb885e6880d957d5b9382a960d2", size = 88392, upload-time = "2024-11-01T14:06:32.99Z" },
+    { url = "https://files.pythonhosted.org/packages/d1/40/b75381494851556de56281e053700e46bff5b37bf4c7267e858640af5a7f/watchdog-6.0.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:afd0fe1b2270917c5e23c2a65ce50c2a4abb63daafb0d419fde368e272a76b7c", size = 89019, upload-time = "2024-11-01T14:06:34.963Z" },
+    { url = "https://files.pythonhosted.org/packages/39/ea/3930d07dafc9e286ed356a679aa02d777c06e9bfd1164fa7c19c288a5483/watchdog-6.0.0-cp312-cp312-macosx_10_13_universal2.whl", hash = "sha256:bdd4e6f14b8b18c334febb9c4425a878a2ac20efd1e0b231978e7b150f92a948", size = 96471, upload-time = "2024-11-01T14:06:37.745Z" },
+    { url = "https://files.pythonhosted.org/packages/12/87/48361531f70b1f87928b045df868a9fd4e253d9ae087fa4cf3f7113be363/watchdog-6.0.0-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:c7c15dda13c4eb00d6fb6fc508b3c0ed88b9d5d374056b239c4ad1611125c860", size = 88449, upload-time = "2024-11-01T14:06:39.748Z" },
+    { url = "https://files.pythonhosted.org/packages/5b/7e/8f322f5e600812e6f9a31b75d242631068ca8f4ef0582dd3ae6e72daecc8/watchdog-6.0.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:6f10cb2d5902447c7d0da897e2c6768bca89174d0c6e1e30abec5421af97a5b0", size = 89054, upload-time = "2024-11-01T14:06:41.009Z" },
+    { url = "https://files.pythonhosted.org/packages/68/98/b0345cabdce2041a01293ba483333582891a3bd5769b08eceb0d406056ef/watchdog-6.0.0-cp313-cp313-macosx_10_13_universal2.whl", hash = "sha256:490ab2ef84f11129844c23fb14ecf30ef3d8a6abafd3754a6f75ca1e6654136c", size = 96480, upload-time = "2024-11-01T14:06:42.952Z" },
+    { url = "https://files.pythonhosted.org/packages/85/83/cdf13902c626b28eedef7ec4f10745c52aad8a8fe7eb04ed7b1f111ca20e/watchdog-6.0.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:76aae96b00ae814b181bb25b1b98076d5fc84e8a53cd8885a318b42b6d3a5134", size = 88451, upload-time = "2024-11-01T14:06:45.084Z" },
+    { url = "https://files.pythonhosted.org/packages/fe/c4/225c87bae08c8b9ec99030cd48ae9c4eca050a59bf5c2255853e18c87b50/watchdog-6.0.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:a175f755fc2279e0b7312c0035d52e27211a5bc39719dd529625b1930917345b", size = 89057, upload-time = "2024-11-01T14:06:47.324Z" },
+    { url = "https://files.pythonhosted.org/packages/30/ad/d17b5d42e28a8b91f8ed01cb949da092827afb9995d4559fd448d0472763/watchdog-6.0.0-pp310-pypy310_pp73-macosx_10_15_x86_64.whl", hash = "sha256:c7ac31a19f4545dd92fc25d200694098f42c9a8e391bc00bdd362c5736dbf881", size = 87902, upload-time = "2024-11-01T14:06:53.119Z" },
+    { url = "https://files.pythonhosted.org/packages/5c/ca/c3649991d140ff6ab67bfc85ab42b165ead119c9e12211e08089d763ece5/watchdog-6.0.0-pp310-pypy310_pp73-macosx_11_0_arm64.whl", hash = "sha256:9513f27a1a582d9808cf21a07dae516f0fab1cf2d7683a742c498b93eedabb11", size = 88380, upload-time = "2024-11-01T14:06:55.19Z" },
+    { url = "https://files.pythonhosted.org/packages/a9/c7/ca4bf3e518cb57a686b2feb4f55a1892fd9a3dd13f470fca14e00f80ea36/watchdog-6.0.0-py3-none-manylinux2014_aarch64.whl", hash = "sha256:7607498efa04a3542ae3e05e64da8202e58159aa1fa4acddf7678d34a35d4f13", size = 79079, upload-time = "2024-11-01T14:06:59.472Z" },
+    { url = "https://files.pythonhosted.org/packages/5c/51/d46dc9332f9a647593c947b4b88e2381c8dfc0942d15b8edc0310fa4abb1/watchdog-6.0.0-py3-none-manylinux2014_armv7l.whl", hash = "sha256:9041567ee8953024c83343288ccc458fd0a2d811d6a0fd68c4c22609e3490379", size = 79078, upload-time = "2024-11-01T14:07:01.431Z" },
+    { url = "https://files.pythonhosted.org/packages/d4/57/04edbf5e169cd318d5f07b4766fee38e825d64b6913ca157ca32d1a42267/watchdog-6.0.0-py3-none-manylinux2014_i686.whl", hash = "sha256:82dc3e3143c7e38ec49d61af98d6558288c415eac98486a5c581726e0737c00e", size = 79076, upload-time = "2024-11-01T14:07:02.568Z" },
+    { url = "https://files.pythonhosted.org/packages/ab/cc/da8422b300e13cb187d2203f20b9253e91058aaf7db65b74142013478e66/watchdog-6.0.0-py3-none-manylinux2014_ppc64.whl", hash = "sha256:212ac9b8bf1161dc91bd09c048048a95ca3a4c4f5e5d4a7d1b1a7d5752a7f96f", size = 79077, upload-time = "2024-11-01T14:07:03.893Z" },
+    { url = "https://files.pythonhosted.org/packages/2c/3b/b8964e04ae1a025c44ba8e4291f86e97fac443bca31de8bd98d3263d2fcf/watchdog-6.0.0-py3-none-manylinux2014_ppc64le.whl", hash = "sha256:e3df4cbb9a450c6d49318f6d14f4bbc80d763fa587ba46ec86f99f9e6876bb26", size = 79078, upload-time = "2024-11-01T14:07:05.189Z" },
+    { url = "https://files.pythonhosted.org/packages/62/ae/a696eb424bedff7407801c257d4b1afda455fe40821a2be430e173660e81/watchdog-6.0.0-py3-none-manylinux2014_s390x.whl", hash = "sha256:2cce7cfc2008eb51feb6aab51251fd79b85d9894e98ba847408f662b3395ca3c", size = 79077, upload-time = "2024-11-01T14:07:06.376Z" },
+    { url = "https://files.pythonhosted.org/packages/b5/e8/dbf020b4d98251a9860752a094d09a65e1b436ad181faf929983f697048f/watchdog-6.0.0-py3-none-manylinux2014_x86_64.whl", hash = "sha256:20ffe5b202af80ab4266dcd3e91aae72bf2da48c0d33bdb15c66658e685e94e2", size = 79078, upload-time = "2024-11-01T14:07:07.547Z" },
+    { url = "https://files.pythonhosted.org/packages/07/f6/d0e5b343768e8bcb4cda79f0f2f55051bf26177ecd5651f84c07567461cf/watchdog-6.0.0-py3-none-win32.whl", hash = "sha256:07df1fdd701c5d4c8e55ef6cf55b8f0120fe1aef7ef39a1c6fc6bc2e606d517a", size = 79065, upload-time = "2024-11-01T14:07:09.525Z" },
+    { url = "https://files.pythonhosted.org/packages/db/d9/c495884c6e548fce18a8f40568ff120bc3a4b7b99813081c8ac0c936fa64/watchdog-6.0.0-py3-none-win_amd64.whl", hash = "sha256:cbafb470cf848d93b5d013e2ecb245d4aa1c8fd0504e863ccefa32445359d680", size = 79070, upload-time = "2024-11-01T14:07:10.686Z" },
+    { url = "https://files.pythonhosted.org/packages/33/e8/e40370e6d74ddba47f002a32919d91310d6074130fe4e17dabcafc15cbf1/watchdog-6.0.0-py3-none-win_ia64.whl", hash = "sha256:a1914259fa9e1454315171103c6a30961236f508b9b623eae470268bbcc6a22f", size = 79067, upload-time = "2024-11-01T14:07:11.845Z" },
+]
+
+[[package]]
+name = "wcwidth"
+version = "0.5.3"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/c2/62/a7c072fbfefb2980a00f99ca994279cb9ecf310cb2e6b2a4d2a28fe192b3/wcwidth-0.5.3.tar.gz", hash = "sha256:53123b7af053c74e9fe2e92ac810301f6139e64379031f7124574212fb3b4091", size = 157587, upload-time = "2026-01-31T03:52:10.92Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/3c/c1/d73f12f8cdb1891334a2ccf7389eed244d3941e74d80dd220badb937f3fb/wcwidth-0.5.3-py3-none-any.whl", hash = "sha256:d584eff31cd4753e1e5ff6c12e1edfdb324c995713f75d26c29807bb84bf649e", size = 92981, upload-time = "2026-01-31T03:52:09.14Z" },
+]
+
+[[package]]
+name = "wrapt"
+version = "2.1.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/86/31/afb4cf08b9892430ec419a3f0f469fb978cb013f4432e0edb9c2cf06f081/wrapt-2.1.0.tar.gz", hash = "sha256:757ff1de7e1d8db1839846672aaecf4978af433cc57e808255b83980e9651914", size = 80924, upload-time = "2026-01-31T23:25:58.917Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/60/6f/9773ddbf70d2f787d049fb5a412c18fd8140b8a33e90e8b911f0d512a7b5/wrapt-2.1.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:ba00229045bc0ec808f12f7d2fd02166631657c56d5b7acbbb8f03ea70fc1cd6", size = 60561, upload-time = "2026-01-31T23:26:51.314Z" },
+    { url = "https://files.pythonhosted.org/packages/a1/91/f6cc8762153ebcdccf7d7aa7ca3a75fe688b7cebc250f1eac72229943d09/wrapt-2.1.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:657c7d0dcca7df8cfdce9d4e9062d51d2a2b2c8f4bdd41dc908a717099cf552a", size = 61501, upload-time = "2026-01-31T23:26:30.857Z" },
+    { url = "https://files.pythonhosted.org/packages/f1/74/ce91f1e9cd77bf6c11700ac0643ccb747d7c4cbd948f63fba90d345aa85a/wrapt-2.1.0-cp310-cp310-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl", hash = "sha256:cb21ff015afe80cc30daca53136427463c364fb7c1ca96e4b7013dc6f56b2829", size = 113531, upload-time = "2026-01-31T23:26:36.105Z" },
+    { url = "https://files.pythonhosted.org/packages/91/be/231563aaf305c930705b4455023155bf485974b431bb4bf9ddc53be5ae9b/wrapt-2.1.0-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:f8d25f31cf032bfa70ec1872cdf0f7e1f1154c5a5bc6c73444bb3375b904f97f", size = 115538, upload-time = "2026-01-31T23:25:31.53Z" },
+    { url = "https://files.pythonhosted.org/packages/9b/9e/576265c0173e85e77eb9713ccecedeab34a1785d493bfa511fd98b7154bc/wrapt-2.1.0-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:099c88ce146134786577b892d2748ac79c16c9f70304367eee17295732907045", size = 114719, upload-time = "2026-01-31T23:25:25.153Z" },
+    { url = "https://files.pythonhosted.org/packages/77/07/b374bd08739bf2f5c1accbb4c77e34bd21c3e9a0c5c49f54269d014c263f/wrapt-2.1.0-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:dfe3f09f5ce33a4e54a3340c3cde774fd19eca0da8a83343889a3673a33ee579", size = 113204, upload-time = "2026-01-31T23:26:02.779Z" },
+    { url = "https://files.pythonhosted.org/packages/8f/e6/09285cbca4467c9701bea4c8bece8bf7cebdf2721516e6b6ccc8737d086f/wrapt-2.1.0-cp310-cp310-win32.whl", hash = "sha256:73d77cc1698bf2f0580616a2eadb94aa15b47ae09ade7d9828a5c413dbbabab8", size = 57878, upload-time = "2026-01-31T23:25:18.276Z" },
+    { url = "https://files.pythonhosted.org/packages/b5/6c/b027af42f6c8aacf1ab83c0a4b278ded3488f452b8b0478c5204637338dc/wrapt-2.1.0-cp310-cp310-win_amd64.whl", hash = "sha256:8012f863320ece76c6b95527b8ee831b818e186cafa356620cba15ba19c904de", size = 60226, upload-time = "2026-01-31T23:26:38.631Z" },
+    { url = "https://files.pythonhosted.org/packages/22/ab/1fc44e40f4a7277f67eac33c645c88d54192fa2a8c6cad2735f8eb86fe3b/wrapt-2.1.0-cp310-cp310-win_arm64.whl", hash = "sha256:ccd99596ae95bc7b844196e6691b4987749ba7832c9ba437fdd99885ee5e7a84", size = 58648, upload-time = "2026-01-31T23:26:20.163Z" },
+    { url = "https://files.pythonhosted.org/packages/97/0a/de541b2543e33144043cd58da09bda8d837ba42e13ae90baca32b0553023/wrapt-2.1.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:d877003dbc601e1365bd03f6a980965a20d585f90c056f33e1fc241b63a6f0e7", size = 60558, upload-time = "2026-01-31T23:25:27.784Z" },
+    { url = "https://files.pythonhosted.org/packages/84/2e/7e48207420e6ca7e7a05c0e4ebe9464ec9965c8face256f3ef8cc2acd862/wrapt-2.1.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:771ec962fe3ccb078177c9b8f3529e204ffcbb11d62d509e0a438e6a83f7ca68", size = 61501, upload-time = "2026-01-31T23:26:46.477Z" },
+    { url = "https://files.pythonhosted.org/packages/67/2b/639a4970ecdc7143acb69a1162c76b0f1620218ad502c33e1a88d28f00b1/wrapt-2.1.0-cp311-cp311-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl", hash = "sha256:73e742368b52f9cf0921e1d2bcb8a6a44ede2e372e33df6e77caa136a942099f", size = 113954, upload-time = "2026-01-31T23:26:01.493Z" },
+    { url = "https://files.pythonhosted.org/packages/81/5d/8d9177c8c0ecaf5313b462be63c5aa9672044b02bfd644dd65c6cb420d2a/wrapt-2.1.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:0e9129d1b582c55ad0dfb9e29e221daa0e02b18c67d8642bc8d08dd7038b3aed", size = 115994, upload-time = "2026-01-31T23:25:57.118Z" },
+    { url = "https://files.pythonhosted.org/packages/e3/e3/c5a514a0ed1dc463f5b6b4e31abbaa3b8df48b9fd391a6e8412608155a29/wrapt-2.1.0-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:cc9e37bfe67f6ea738851dd606640a87692ff81bcc76df313fb75d08e05e855f", size = 115245, upload-time = "2026-01-31T23:26:11.171Z" },
+    { url = "https://files.pythonhosted.org/packages/35/9c/2fc6a31f5758266de2cf9dc6111d3bda7b7dd6cbdcabfd755103bbcda08f/wrapt-2.1.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:46583aae3c807aa76f96355c4943031225785ed160c84052612bba0e9d456639", size = 113679, upload-time = "2026-01-31T23:25:19.475Z" },
+    { url = "https://files.pythonhosted.org/packages/6c/81/ce52694dc8184f4898c01c8af20e145b348fc7a0e4766a7345c45f0e9ce6/wrapt-2.1.0-cp311-cp311-win32.whl", hash = "sha256:e3958ba70aef2895d8c62c2d31f51ced188f60451212294677b92f4b32c12978", size = 57865, upload-time = "2026-01-31T23:25:50.947Z" },
+    { url = "https://files.pythonhosted.org/packages/85/31/0df5d38243c2a538e7bd481e676d286b41f98a729e0d37cfed9f4421ad4d/wrapt-2.1.0-cp311-cp311-win_amd64.whl", hash = "sha256:0ff9797e6e0b82b330ef80b0cdba7fcd0ca056d4c7af2ca44e3d05fd47929ede", size = 60227, upload-time = "2026-01-31T23:25:35.954Z" },
+    { url = "https://files.pythonhosted.org/packages/a3/79/b587edbab21d6b8a7460234440c784e08344bcdf4fdfd9a6e9125ea14923/wrapt-2.1.0-cp311-cp311-win_arm64.whl", hash = "sha256:4b0a29509ef7b501abe47b693a3c91d1f21c9a948711f6ce7afa81eb274c7eae", size = 58648, upload-time = "2026-01-31T23:25:32.887Z" },
+    { url = "https://files.pythonhosted.org/packages/f8/6f/c731b1fbbcdf9bd202809c6fa354c4237b663dd82a95035a7cbe899cfd25/wrapt-2.1.0-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:a64c0fb29c89810973f312a04c067b63523e7303b9a2653820cbf16474c2e5cf", size = 61149, upload-time = "2026-01-31T23:25:29.092Z" },
+    { url = "https://files.pythonhosted.org/packages/b2/da/7022458a1d99f0c59720a0b0fd4b1966f8df6d41e741aadfe43bc5350547/wrapt-2.1.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:5509d9150ed01c4149e40020fa68e917d5c4bb77d311e79535565c2a0418afcb", size = 61743, upload-time = "2026-01-31T23:26:14.338Z" },
+    { url = "https://files.pythonhosted.org/packages/b5/f4/57cc12c3fc6f4fe6ccfc15567cc1ac8aeb53a9946a675adc3df7a1ee4e6a/wrapt-2.1.0-cp312-cp312-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl", hash = "sha256:52bb58b3207ace156b6134235fd43140994597704fd07d148cbcfb474ee084ea", size = 121331, upload-time = "2026-01-31T23:25:37.294Z" },
+    { url = "https://files.pythonhosted.org/packages/5e/a4/a96ea114298f81f02c07313da85fd46a2a57bbe12389d0619ac3371f691c/wrapt-2.1.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:7112cbf72fc4035afe1e3314a311654c41dd92c2932021ef76f5ca87583917b3", size = 122907, upload-time = "2026-01-31T23:26:49.604Z" },
+    { url = "https://files.pythonhosted.org/packages/ac/43/df73362b6e47f92aaff0fc3fc459314025c795f75d61724c83232dee199c/wrapt-2.1.0-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:e90656b433808a0ab68e95aaf9f588aea5c8c7a514e180849dfc638ba00ec449", size = 121337, upload-time = "2026-01-31T23:26:04.072Z" },
+    { url = "https://files.pythonhosted.org/packages/51/4f/8147e3b9a7887cee4eeb3a3414265ad4649a156832a08063f55aa7842af0/wrapt-2.1.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:e45f54903da38fc4f6f66397fd550fc0dac6164b4c5e721c1b4eb05664181821", size = 120461, upload-time = "2026-01-31T23:26:43.055Z" },
+    { url = "https://files.pythonhosted.org/packages/35/b1/eea720fcca8a05dec848a6d11a47c20f59bdabdcc444ba3be0589350eb7a/wrapt-2.1.0-cp312-cp312-win32.whl", hash = "sha256:6653bf30dbbafd55cb4553195cc60b94920b6711a8835866c0e02aa9f22c5598", size = 58089, upload-time = "2026-01-31T23:26:47.773Z" },
+    { url = "https://files.pythonhosted.org/packages/af/79/8a8f3f8c71ee3379191b69e47f32115fa25cdb6d5b581d74c64d5c897fa7/wrapt-2.1.0-cp312-cp312-win_amd64.whl", hash = "sha256:d61238a072501ed071a9f4b9567d10c2eb3d2f1a0258ae79b47160871d8f29c3", size = 60330, upload-time = "2026-01-31T23:26:12.518Z" },
+    { url = "https://files.pythonhosted.org/packages/08/4e/e992d05c3d2f7163883a65ead2620ff5fe7b3d44d7c2136ce981e40e453d/wrapt-2.1.0-cp312-cp312-win_arm64.whl", hash = "sha256:9e971000347f61271725e801ef44fa5d01b52720e59737f0d96280bffb98c5d1", size = 58727, upload-time = "2026-01-31T23:26:53.222Z" },
+    { url = "https://files.pythonhosted.org/packages/30/93/b414826a5aaf2fdcfe73c2e649cbeb2e098fef4820d1217554ee64f45666/wrapt-2.1.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:875a10a6f3b667f90a39010af26acf684ba831d9b18a86b242899d57c74550fa", size = 61155, upload-time = "2026-01-31T23:26:24.462Z" },
+    { url = "https://files.pythonhosted.org/packages/58/9e/8b21ea776bf2a3c858e3377ecde4b348893ec44dc1726baaf583ca22c56e/wrapt-2.1.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:e00f8559ceac0fb45091daad5f15d37f2c22bdc28ed71521d47ff01aad8fff3d", size = 61747, upload-time = "2026-01-31T23:25:53.987Z" },
+    { url = "https://files.pythonhosted.org/packages/da/ec/48cd2470ad09557dfe6fccfe9de98698cc0df3786a6d4d97e8edd574d67a/wrapt-2.1.0-cp313-cp313-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl", hash = "sha256:ce0cf4c79c19904aaf2e822af280d7b3c23ad902f57e31c5a19433bc86e5d36d", size = 121342, upload-time = "2026-01-31T23:26:32.156Z" },
+    { url = "https://files.pythonhosted.org/packages/3b/4e/e8447b31be27b6057cdfc904a38632a765c3407fb4d10d11e5c1d0c203d5/wrapt-2.1.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:d3dd4f8c2256fcde1a85037a1837afc52e8d32d086fd669ae469455fd9a988d6", size = 122951, upload-time = "2026-01-31T23:25:08.936Z" },
+    { url = "https://files.pythonhosted.org/packages/7e/b6/73a6c9277e844ffe11f3002ad27a84ff5418248def33af9435d24dfe6c5b/wrapt-2.1.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:737e1e491473047cb66944b8b8fd23f3f542019afd6cf0569d1356d18a7ea6d5", size = 121373, upload-time = "2026-01-31T23:26:18.322Z" },
+    { url = "https://files.pythonhosted.org/packages/85/04/869384435fecf829dc05621ffa02dab0f2f830be5d42fa8d8ac7b0b4c9fa/wrapt-2.1.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:38de19e30e266c15d542ceb0603e657db4e82c53e7f47fd70674ae5da2b41180", size = 120468, upload-time = "2026-01-31T23:25:13.689Z" },
+    { url = "https://files.pythonhosted.org/packages/80/ac/42a5378d9b5b486122ae0572c46ae8d69ab6486b9f13961e6b9706297ff5/wrapt-2.1.0-cp313-cp313-win32.whl", hash = "sha256:bc7d496b6e16bd2f77e37e8969b21a7b58d6954e46c6689986fb67b9078100e5", size = 58095, upload-time = "2026-01-31T23:26:33.481Z" },
+    { url = "https://files.pythonhosted.org/packages/86/de/538fcef30f70a1aaadab4cab7d0396037518d7ec2b064557171147ce297f/wrapt-2.1.0-cp313-cp313-win_amd64.whl", hash = "sha256:57df799e67b011847ef7ac64b05ed4633e56b64e7e7cab5eb83dc9689dbe0acf", size = 60344, upload-time = "2026-01-31T23:25:10.615Z" },
+    { url = "https://files.pythonhosted.org/packages/08/13/27884668b21e9f0a625c13ebd6a8d70ad8371250ec8519881858404686bf/wrapt-2.1.0-cp313-cp313-win_arm64.whl", hash = "sha256:01559d2961c29edc6263849fd9d32b29a20737da67648c7fd752a67bd96208c7", size = 58734, upload-time = "2026-01-31T23:26:00.099Z" },
+    { url = "https://files.pythonhosted.org/packages/c9/a3/e558c5b8f3a097aa1e942e2d75923adebfdfafb5a51ec425d1d062e49ab0/wrapt-2.1.0-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:66f588c8b3a44863156cfaccb516f946a64b3b03a6880822ab0b878135ca1f5c", size = 62972, upload-time = "2026-01-31T23:26:08.576Z" },
+    { url = "https://files.pythonhosted.org/packages/93/b6/7157e98107099fad846f1e79308cc0954e26b25b01c03f1624ba7f57ec54/wrapt-2.1.0-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:355779ff720c11a2a5cffd03332dbce1005cb4747dca65b0fc8cdd5f8bf1037e", size = 63610, upload-time = "2026-01-31T23:26:39.9Z" },
+    { url = "https://files.pythonhosted.org/packages/e4/8e/b8992671e4b4d3ce2a53af930588c204bf37b66eb212bd1722f2a5a8cf62/wrapt-2.1.0-cp313-cp313t-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl", hash = "sha256:7a0471df3fb4e85a9ff62f7142cdb169e31172467cdb79a713f9b1319c555903", size = 152538, upload-time = "2026-01-31T23:26:27.696Z" },
+    { url = "https://files.pythonhosted.org/packages/8c/f6/79f9fd4b3c0a8715e651fff1cc1182a983fd971376d5688a06fa94e31acd/wrapt-2.1.0-cp313-cp313t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:5bacf063143fa86f15b00a21259a81c95c527a18d504b8c820835366d361c879", size = 158702, upload-time = "2026-01-31T23:25:11.848Z" },
+    { url = "https://files.pythonhosted.org/packages/9e/46/f88b52beb813eeb830d9134bc6eaf3e53cde4e3cfa1804e383754d4104fe/wrapt-2.1.0-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:c87cd4f61a3b7cd65113e74006e1cd6352b74807fcc65d440e8342f001f8de5e", size = 155564, upload-time = "2026-01-31T23:25:15.033Z" },
+    { url = "https://files.pythonhosted.org/packages/93/31/97145ea71e3e5a1b419af5c410b07b258155dc7cc1a6302791a93e991c83/wrapt-2.1.0-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:2893498fe898719ac8fb6b4fe36ca86892bec1e2480d94e3bd1bc592c00527ad", size = 150165, upload-time = "2026-01-31T23:26:09.848Z" },
+    { url = "https://files.pythonhosted.org/packages/10/bd/f33551d5bfbb0ddab81296cffc15570570039a973c0f99bba474be0fadf2/wrapt-2.1.0-cp313-cp313t-win32.whl", hash = "sha256:cbc07f101f5f1e7c23ec06a07e45715f459de992108eeb381b21b76d94dbaf4f", size = 59785, upload-time = "2026-01-31T23:25:52.23Z" },
+    { url = "https://files.pythonhosted.org/packages/5f/3a/9a76be7a36442f43841bb6336e262e09a915b2fb5dfc2822ffce1fb903d2/wrapt-2.1.0-cp313-cp313t-win_amd64.whl", hash = "sha256:2ccc89cd504fc29c32f0b24046e8edf3ef0fcbc5d5efe8c91b303c099863d2c8", size = 63085, upload-time = "2026-01-31T23:26:05.363Z" },
+    { url = "https://files.pythonhosted.org/packages/7a/35/65a13c2df008d189ebca5fec534011c5dd69ab4f47e6923b403321816fbf/wrapt-2.1.0-cp313-cp313t-win_arm64.whl", hash = "sha256:0b660be1c9cdfb4c711baab4ccbd0e9d1b65a0480d38729ec8cdbf3b29cb7f15", size = 60254, upload-time = "2026-01-31T23:25:06.052Z" },
+    { url = "https://files.pythonhosted.org/packages/6f/eb/7c9eb1ea9b10ea98d9983a147c877a2ae927acb4a86e2dc4a0b548f05ad1/wrapt-2.1.0-cp314-cp314-macosx_10_15_x86_64.whl", hash = "sha256:7f7bf95bae7ac5f2bbcb307464b3b0ff70569dd3b036a87b1cf7efb2c76e66e5", size = 61316, upload-time = "2026-01-31T23:25:20.739Z" },
+    { url = "https://files.pythonhosted.org/packages/6d/c2/1c3d16d6b644f688913a00e2dc10f59adca817b5b3ee034ce4e9a692ab63/wrapt-2.1.0-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:be2f541a242818829526e5d08c716b6730970ed0dc1b76ba962a546947d0f005", size = 61813, upload-time = "2026-01-31T23:25:49.714Z" },
+    { url = "https://files.pythonhosted.org/packages/8c/51/b6170084b6b771cc62374d924e328df2e81f687399a835f003497cad1110/wrapt-2.1.0-cp314-cp314-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl", hash = "sha256:ad3aa174d06a14b4758d5a1678b9adde8b8e657c6695de9a3d4c223f4fcbbcce", size = 120309, upload-time = "2026-01-31T23:25:16.866Z" },
+    { url = "https://files.pythonhosted.org/packages/f8/34/467829f0dd79f50878b2e67b67c67c816a6326a27d252d4192ef815b4a09/wrapt-2.1.0-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:bffa584240d41bc3127510e07a752f94223d73bb1283ac2e99ac44235762efd2", size = 122690, upload-time = "2026-01-31T23:26:16.914Z" },
+    { url = "https://files.pythonhosted.org/packages/df/5b/244c61a65e0bc9d4a18cfa2a2b3b05f8065290284fc60436a7ea5047ee10/wrapt-2.1.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:9b2da9c8f1723994b335dbf9f496fbfabc76bcdd001f73772b8eb2118a714cea", size = 121115, upload-time = "2026-01-31T23:26:44.518Z" },
+    { url = "https://files.pythonhosted.org/packages/86/7d/f9b5e103d3caf23a72c04a1baf2b61c4a14d1feb440d3c98c26725b4503a/wrapt-2.1.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:eabe95ea5fbe1524a53c0f3fc535c99f2aa376ec1451b0b79d943d2240d80e36", size = 119487, upload-time = "2026-01-31T23:25:34.186Z" },
+    { url = "https://files.pythonhosted.org/packages/f8/49/b61fdc4680dd5cd6828977341b9fd729e2c623338bfe65647f5c0ff8195e/wrapt-2.1.0-cp314-cp314-win32.whl", hash = "sha256:2cd647097df1df78f027ac7d5d663f05daa1a117b69cf7f476cb299f90557747", size = 58519, upload-time = "2026-01-31T23:25:04.426Z" },
+    { url = "https://files.pythonhosted.org/packages/6a/4f/42ab43e496d0d19caed9f69366d0f28f7f08c139297e78b17dab6ecbb6d5/wrapt-2.1.0-cp314-cp314-win_amd64.whl", hash = "sha256:c0fc3e388a14ef8101c685dc80b4d2932924a639a03e5c44b5ffabbda2f1f2dc", size = 60767, upload-time = "2026-01-31T23:25:21.954Z" },
+    { url = "https://files.pythonhosted.org/packages/ef/15/0337768ac97a8758bc0fc1afdf5f656075a7facf198f62bbe8a22b789277/wrapt-2.1.0-cp314-cp314-win_arm64.whl", hash = "sha256:7c06653908a23a85c4b2455b9d37c085f9756c09058df87b4a2fce2b2f8d58c2", size = 59056, upload-time = "2026-01-31T23:26:25.814Z" },
+    { url = "https://files.pythonhosted.org/packages/d6/f1/58f4674d1db44912003a51b34e8d9823a832fbbb39162e9dbe06e5f6424e/wrapt-2.1.0-cp314-cp314t-macosx_10_15_x86_64.whl", hash = "sha256:c70b4829c6f2f4af4cdaa16442032fcaf882063304160555e4a19b43fd2c6c9d", size = 63061, upload-time = "2026-01-31T23:26:06.601Z" },
+    { url = "https://files.pythonhosted.org/packages/02/c1/07f6bf6619285f39cd616314217170c6160da99a46ad6ae4a60044f6ab5a/wrapt-2.1.0-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:d7fd4c4ee51ebdf245549d54a7c2181a4f39caac97c9dc8a050b5ba814067a29", size = 63620, upload-time = "2026-01-31T23:25:30.326Z" },
+    { url = "https://files.pythonhosted.org/packages/46/82/f7df1648762260f60c4e22c066a17d95f20267c94bfe653fab4f08e2c297/wrapt-2.1.0-cp314-cp314t-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl", hash = "sha256:a7b158558438874e5fd5cb505b5a635bd08c84857bc937973d9e12e1166cdf3b", size = 152546, upload-time = "2026-01-31T23:25:02.102Z" },
+    { url = "https://files.pythonhosted.org/packages/78/b7/d953336e09bac13a9ffa9073e167c5dec8aaa4a717a8551bf64cb4683590/wrapt-2.1.0-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:3e2e156fe2d41700b837be9b1d8d80ebab44e9891589bc7c41578ef110184e29", size = 158704, upload-time = "2026-01-31T23:25:43.269Z" },
+    { url = "https://files.pythonhosted.org/packages/39/a1/2ed57e46b30af2a5a750c85a9dd30d2244ef10e2f8db150560126d8cbd24/wrapt-2.1.0-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:9f1e9bac6a6c1ba65e0ac50e32c575266734a07b6c17e718c4babd91e2faa69b", size = 155563, upload-time = "2026-01-31T23:25:39.17Z" },
+    { url = "https://files.pythonhosted.org/packages/d0/8c/4f54f7ea5addf208be44459393185aaa193bd2d0b8ecf4683b159fcc5238/wrapt-2.1.0-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:12687e6271df7ae5706bee44cc1f77fecb7805976ec9f14f58381b30ae2aceb5", size = 150189, upload-time = "2026-01-31T23:25:44.654Z" },
+    { url = "https://files.pythonhosted.org/packages/b7/cc/e8290a1cd94297fbc1e9fbad06481b5a7c918f2db6645c550f05ee47f359/wrapt-2.1.0-cp314-cp314t-win32.whl", hash = "sha256:38bbe336ee32f67eb99f886bd4f040d91310b7e660061bb03b9083d26e8cf915", size = 60431, upload-time = "2026-01-31T23:25:48.34Z" },
+    { url = "https://files.pythonhosted.org/packages/d0/df/af5d244938853e3adb1251ca1397e9fa78d3e92adc808a0af0a8547585d3/wrapt-2.1.0-cp314-cp314t-win_amd64.whl", hash = "sha256:0fa64a9a07df7f85b352adc42b43e7f44085fb11191b8f5b9b77219f7aaf7e17", size = 63859, upload-time = "2026-01-31T23:26:23.2Z" },
+    { url = "https://files.pythonhosted.org/packages/39/c4/28b6f2804e8bc05d17114dfed03a80bce5b83ca2113fd44eecbef12275d1/wrapt-2.1.0-cp314-cp314t-win_arm64.whl", hash = "sha256:da379cbdf3b7d97ace33a69a391b7a7e2130b1aca94dc447246217994233974c", size = 60446, upload-time = "2026-01-31T23:25:41.001Z" },
+    { url = "https://files.pythonhosted.org/packages/57/e9/70983b75d4abd6f85cffc6df79c623220ec5a579ceaacabac35c904b7b52/wrapt-2.1.0-py3-none-any.whl", hash = "sha256:e035693a0d25ea5bf5826df3e203dff7d091b0d5442aaefec9ca8f2bab38417f", size = 43886, upload-time = "2026-01-31T23:25:07.22Z" },
+]
+
+[[package]]
+name = "xxhash"
+version = "3.6.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/02/84/30869e01909fb37a6cc7e18688ee8bf1e42d57e7e0777636bd47524c43c7/xxhash-3.6.0.tar.gz", hash = "sha256:f0162a78b13a0d7617b2845b90c763339d1f1d82bb04a4b07f4ab535cc5e05d6", size = 85160, upload-time = "2025-10-02T14:37:08.097Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/34/ee/f9f1d656ad168681bb0f6b092372c1e533c4416b8069b1896a175c46e484/xxhash-3.6.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:87ff03d7e35c61435976554477a7f4cd1704c3596a89a8300d5ce7fc83874a71", size = 32845, upload-time = "2025-10-02T14:33:51.573Z" },
+    { url = "https://files.pythonhosted.org/packages/a3/b1/93508d9460b292c74a09b83d16750c52a0ead89c51eea9951cb97a60d959/xxhash-3.6.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:f572dfd3d0e2eb1a57511831cf6341242f5a9f8298a45862d085f5b93394a27d", size = 30807, upload-time = "2025-10-02T14:33:52.964Z" },
+    { url = "https://files.pythonhosted.org/packages/07/55/28c93a3662f2d200c70704efe74aab9640e824f8ce330d8d3943bf7c9b3c/xxhash-3.6.0-cp310-cp310-manylinux1_i686.manylinux_2_28_i686.manylinux_2_5_i686.whl", hash = "sha256:89952ea539566b9fed2bbd94e589672794b4286f342254fad28b149f9615fef8", size = 193786, upload-time = "2025-10-02T14:33:54.272Z" },
+    { url = "https://files.pythonhosted.org/packages/c1/96/fec0be9bb4b8f5d9c57d76380a366f31a1781fb802f76fc7cda6c84893c7/xxhash-3.6.0-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:48e6f2ffb07a50b52465a1032c3cf1f4a5683f944acaca8a134a2f23674c2058", size = 212830, upload-time = "2025-10-02T14:33:55.706Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/a0/c706845ba77b9611f81fd2e93fad9859346b026e8445e76f8c6fd057cc6d/xxhash-3.6.0-cp310-cp310-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:b5b848ad6c16d308c3ac7ad4ba6bede80ed5df2ba8ed382f8932df63158dd4b2", size = 211606, upload-time = "2025-10-02T14:33:57.133Z" },
+    { url = "https://files.pythonhosted.org/packages/67/1e/164126a2999e5045f04a69257eea946c0dc3e86541b400d4385d646b53d7/xxhash-3.6.0-cp310-cp310-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:a034590a727b44dd8ac5914236a7b8504144447a9682586c3327e935f33ec8cc", size = 444872, upload-time = "2025-10-02T14:33:58.446Z" },
+    { url = "https://files.pythonhosted.org/packages/2d/4b/55ab404c56cd70a2cf5ecfe484838865d0fea5627365c6c8ca156bd09c8f/xxhash-3.6.0-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:8a8f1972e75ebdd161d7896743122834fe87378160c20e97f8b09166213bf8cc", size = 193217, upload-time = "2025-10-02T14:33:59.724Z" },
+    { url = "https://files.pythonhosted.org/packages/45/e6/52abf06bac316db33aa269091ae7311bd53cfc6f4b120ae77bac1b348091/xxhash-3.6.0-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:ee34327b187f002a596d7b167ebc59a1b729e963ce645964bbc050d2f1b73d07", size = 210139, upload-time = "2025-10-02T14:34:02.041Z" },
+    { url = "https://files.pythonhosted.org/packages/34/37/db94d490b8691236d356bc249c08819cbcef9273a1a30acf1254ff9ce157/xxhash-3.6.0-cp310-cp310-musllinux_1_2_i686.whl", hash = "sha256:339f518c3c7a850dd033ab416ea25a692759dc7478a71131fe8869010d2b75e4", size = 197669, upload-time = "2025-10-02T14:34:03.664Z" },
+    { url = "https://files.pythonhosted.org/packages/b7/36/c4f219ef4a17a4f7a64ed3569bc2b5a9c8311abdb22249ac96093625b1a4/xxhash-3.6.0-cp310-cp310-musllinux_1_2_ppc64le.whl", hash = "sha256:bf48889c9630542d4709192578aebbd836177c9f7a4a2778a7d6340107c65f06", size = 210018, upload-time = "2025-10-02T14:34:05.325Z" },
+    { url = "https://files.pythonhosted.org/packages/fd/06/bfac889a374fc2fc439a69223d1750eed2e18a7db8514737ab630534fa08/xxhash-3.6.0-cp310-cp310-musllinux_1_2_s390x.whl", hash = "sha256:5576b002a56207f640636056b4160a378fe36a58db73ae5c27a7ec8db35f71d4", size = 413058, upload-time = "2025-10-02T14:34:06.925Z" },
+    { url = "https://files.pythonhosted.org/packages/c9/d1/555d8447e0dd32ad0930a249a522bb2e289f0d08b6b16204cfa42c1f5a0c/xxhash-3.6.0-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:af1f3278bd02814d6dedc5dec397993b549d6f16c19379721e5a1d31e132c49b", size = 190628, upload-time = "2025-10-02T14:34:08.669Z" },
+    { url = "https://files.pythonhosted.org/packages/d1/15/8751330b5186cedc4ed4b597989882ea05e0408b53fa47bcb46a6125bfc6/xxhash-3.6.0-cp310-cp310-win32.whl", hash = "sha256:aed058764db109dc9052720da65fafe84873b05eb8b07e5e653597951af57c3b", size = 30577, upload-time = "2025-10-02T14:34:10.234Z" },
+    { url = "https://files.pythonhosted.org/packages/bb/cc/53f87e8b5871a6eb2ff7e89c48c66093bda2be52315a8161ddc54ea550c4/xxhash-3.6.0-cp310-cp310-win_amd64.whl", hash = "sha256:e82da5670f2d0d98950317f82a0e4a0197150ff19a6df2ba40399c2a3b9ae5fb", size = 31487, upload-time = "2025-10-02T14:34:11.618Z" },
+    { url = "https://files.pythonhosted.org/packages/9f/00/60f9ea3bb697667a14314d7269956f58bf56bb73864f8f8d52a3c2535e9a/xxhash-3.6.0-cp310-cp310-win_arm64.whl", hash = "sha256:4a082ffff8c6ac07707fb6b671caf7c6e020c75226c561830b73d862060f281d", size = 27863, upload-time = "2025-10-02T14:34:12.619Z" },
+    { url = "https://files.pythonhosted.org/packages/17/d4/cc2f0400e9154df4b9964249da78ebd72f318e35ccc425e9f403c392f22a/xxhash-3.6.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:b47bbd8cf2d72797f3c2772eaaac0ded3d3af26481a26d7d7d41dc2d3c46b04a", size = 32844, upload-time = "2025-10-02T14:34:14.037Z" },
+    { url = "https://files.pythonhosted.org/packages/5e/ec/1cc11cd13e26ea8bc3cb4af4eaadd8d46d5014aebb67be3f71fb0b68802a/xxhash-3.6.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:2b6821e94346f96db75abaa6e255706fb06ebd530899ed76d32cd99f20dc52fa", size = 30809, upload-time = "2025-10-02T14:34:15.484Z" },
+    { url = "https://files.pythonhosted.org/packages/04/5f/19fe357ea348d98ca22f456f75a30ac0916b51c753e1f8b2e0e6fb884cce/xxhash-3.6.0-cp311-cp311-manylinux1_i686.manylinux_2_28_i686.manylinux_2_5_i686.whl", hash = "sha256:d0a9751f71a1a65ce3584e9cae4467651c7e70c9d31017fa57574583a4540248", size = 194665, upload-time = "2025-10-02T14:34:16.541Z" },
+    { url = "https://files.pythonhosted.org/packages/90/3b/d1f1a8f5442a5fd8beedae110c5af7604dc37349a8e16519c13c19a9a2de/xxhash-3.6.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:8b29ee68625ab37b04c0b40c3fafdf24d2f75ccd778333cfb698f65f6c463f62", size = 213550, upload-time = "2025-10-02T14:34:17.878Z" },
+    { url = "https://files.pythonhosted.org/packages/c4/ef/3a9b05eb527457d5db13a135a2ae1a26c80fecd624d20f3e8dcc4cb170f3/xxhash-3.6.0-cp311-cp311-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:6812c25fe0d6c36a46ccb002f40f27ac903bf18af9f6dd8f9669cb4d176ab18f", size = 212384, upload-time = "2025-10-02T14:34:19.182Z" },
+    { url = "https://files.pythonhosted.org/packages/0f/18/ccc194ee698c6c623acbf0f8c2969811a8a4b6185af5e824cd27b9e4fd3e/xxhash-3.6.0-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:4ccbff013972390b51a18ef1255ef5ac125c92dc9143b2d1909f59abc765540e", size = 445749, upload-time = "2025-10-02T14:34:20.659Z" },
+    { url = "https://files.pythonhosted.org/packages/a5/86/cf2c0321dc3940a7aa73076f4fd677a0fb3e405cb297ead7d864fd90847e/xxhash-3.6.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:297b7fbf86c82c550e12e8fb71968b3f033d27b874276ba3624ea868c11165a8", size = 193880, upload-time = "2025-10-02T14:34:22.431Z" },
+    { url = "https://files.pythonhosted.org/packages/82/fb/96213c8560e6f948a1ecc9a7613f8032b19ee45f747f4fca4eb31bb6d6ed/xxhash-3.6.0-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:dea26ae1eb293db089798d3973a5fc928a18fdd97cc8801226fae705b02b14b0", size = 210912, upload-time = "2025-10-02T14:34:23.937Z" },
+    { url = "https://files.pythonhosted.org/packages/40/aa/4395e669b0606a096d6788f40dbdf2b819d6773aa290c19e6e83cbfc312f/xxhash-3.6.0-cp311-cp311-musllinux_1_2_i686.whl", hash = "sha256:7a0b169aafb98f4284f73635a8e93f0735f9cbde17bd5ec332480484241aaa77", size = 198654, upload-time = "2025-10-02T14:34:25.644Z" },
+    { url = "https://files.pythonhosted.org/packages/67/74/b044fcd6b3d89e9b1b665924d85d3f400636c23590226feb1eb09e1176ce/xxhash-3.6.0-cp311-cp311-musllinux_1_2_ppc64le.whl", hash = "sha256:08d45aef063a4531b785cd72de4887766d01dc8f362a515693df349fdb825e0c", size = 210867, upload-time = "2025-10-02T14:34:27.203Z" },
+    { url = "https://files.pythonhosted.org/packages/bc/fd/3ce73bf753b08cb19daee1eb14aa0d7fe331f8da9c02dd95316ddfe5275e/xxhash-3.6.0-cp311-cp311-musllinux_1_2_s390x.whl", hash = "sha256:929142361a48ee07f09121fe9e96a84950e8d4df3bb298ca5d88061969f34d7b", size = 414012, upload-time = "2025-10-02T14:34:28.409Z" },
+    { url = "https://files.pythonhosted.org/packages/ba/b3/5a4241309217c5c876f156b10778f3ab3af7ba7e3259e6d5f5c7d0129eb2/xxhash-3.6.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:51312c768403d8540487dbbfb557454cfc55589bbde6424456951f7fcd4facb3", size = 191409, upload-time = "2025-10-02T14:34:29.696Z" },
+    { url = "https://files.pythonhosted.org/packages/c0/01/99bfbc15fb9abb9a72b088c1d95219fc4782b7d01fc835bd5744d66dd0b8/xxhash-3.6.0-cp311-cp311-win32.whl", hash = "sha256:d1927a69feddc24c987b337ce81ac15c4720955b667fe9b588e02254b80446fd", size = 30574, upload-time = "2025-10-02T14:34:31.028Z" },
+    { url = "https://files.pythonhosted.org/packages/65/79/9d24d7f53819fe301b231044ea362ce64e86c74f6e8c8e51320de248b3e5/xxhash-3.6.0-cp311-cp311-win_amd64.whl", hash = "sha256:26734cdc2d4ffe449b41d186bbeac416f704a482ed835d375a5c0cb02bc63fef", size = 31481, upload-time = "2025-10-02T14:34:32.062Z" },
+    { url = "https://files.pythonhosted.org/packages/30/4e/15cd0e3e8772071344eab2961ce83f6e485111fed8beb491a3f1ce100270/xxhash-3.6.0-cp311-cp311-win_arm64.whl", hash = "sha256:d72f67ef8bf36e05f5b6c65e8524f265bd61071471cd4cf1d36743ebeeeb06b7", size = 27861, upload-time = "2025-10-02T14:34:33.555Z" },
+    { url = "https://files.pythonhosted.org/packages/9a/07/d9412f3d7d462347e4511181dea65e47e0d0e16e26fbee2ea86a2aefb657/xxhash-3.6.0-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:01362c4331775398e7bb34e3ab403bc9ee9f7c497bc7dee6272114055277dd3c", size = 32744, upload-time = "2025-10-02T14:34:34.622Z" },
+    { url = "https://files.pythonhosted.org/packages/79/35/0429ee11d035fc33abe32dca1b2b69e8c18d236547b9a9b72c1929189b9a/xxhash-3.6.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:b7b2df81a23f8cb99656378e72501b2cb41b1827c0f5a86f87d6b06b69f9f204", size = 30816, upload-time = "2025-10-02T14:34:36.043Z" },
+    { url = "https://files.pythonhosted.org/packages/b7/f2/57eb99aa0f7d98624c0932c5b9a170e1806406cdbcdb510546634a1359e0/xxhash-3.6.0-cp312-cp312-manylinux1_i686.manylinux_2_28_i686.manylinux_2_5_i686.whl", hash = "sha256:dc94790144e66b14f67b10ac8ed75b39ca47536bf8800eb7c24b50271ea0c490", size = 194035, upload-time = "2025-10-02T14:34:37.354Z" },
+    { url = "https://files.pythonhosted.org/packages/4c/ed/6224ba353690d73af7a3f1c7cdb1fc1b002e38f783cb991ae338e1eb3d79/xxhash-3.6.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:93f107c673bccf0d592cdba077dedaf52fe7f42dcd7676eba1f6d6f0c3efffd2", size = 212914, upload-time = "2025-10-02T14:34:38.6Z" },
+    { url = "https://files.pythonhosted.org/packages/38/86/fb6b6130d8dd6b8942cc17ab4d90e223653a89aa32ad2776f8af7064ed13/xxhash-3.6.0-cp312-cp312-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:2aa5ee3444c25b69813663c9f8067dcfaa2e126dc55e8dddf40f4d1c25d7effa", size = 212163, upload-time = "2025-10-02T14:34:39.872Z" },
+    { url = "https://files.pythonhosted.org/packages/ee/dc/e84875682b0593e884ad73b2d40767b5790d417bde603cceb6878901d647/xxhash-3.6.0-cp312-cp312-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:f7f99123f0e1194fa59cc69ad46dbae2e07becec5df50a0509a808f90a0f03f0", size = 445411, upload-time = "2025-10-02T14:34:41.569Z" },
+    { url = "https://files.pythonhosted.org/packages/11/4f/426f91b96701ec2f37bb2b8cec664eff4f658a11f3fa9d94f0a887ea6d2b/xxhash-3.6.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:49e03e6fe2cac4a1bc64952dd250cf0dbc5ef4ebb7b8d96bce82e2de163c82a2", size = 193883, upload-time = "2025-10-02T14:34:43.249Z" },
+    { url = "https://files.pythonhosted.org/packages/53/5a/ddbb83eee8e28b778eacfc5a85c969673e4023cdeedcfcef61f36731610b/xxhash-3.6.0-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:bd17fede52a17a4f9a7bc4472a5867cb0b160deeb431795c0e4abe158bc784e9", size = 210392, upload-time = "2025-10-02T14:34:45.042Z" },
+    { url = "https://files.pythonhosted.org/packages/1e/c2/ff69efd07c8c074ccdf0a4f36fcdd3d27363665bcdf4ba399abebe643465/xxhash-3.6.0-cp312-cp312-musllinux_1_2_i686.whl", hash = "sha256:6fb5f5476bef678f69db04f2bd1efbed3030d2aba305b0fc1773645f187d6a4e", size = 197898, upload-time = "2025-10-02T14:34:46.302Z" },
+    { url = "https://files.pythonhosted.org/packages/58/ca/faa05ac19b3b622c7c9317ac3e23954187516298a091eb02c976d0d3dd45/xxhash-3.6.0-cp312-cp312-musllinux_1_2_ppc64le.whl", hash = "sha256:843b52f6d88071f87eba1631b684fcb4b2068cd2180a0224122fe4ef011a9374", size = 210655, upload-time = "2025-10-02T14:34:47.571Z" },
+    { url = "https://files.pythonhosted.org/packages/d4/7a/06aa7482345480cc0cb597f5c875b11a82c3953f534394f620b0be2f700c/xxhash-3.6.0-cp312-cp312-musllinux_1_2_s390x.whl", hash = "sha256:7d14a6cfaf03b1b6f5f9790f76880601ccc7896aff7ab9cd8978a939c1eb7e0d", size = 414001, upload-time = "2025-10-02T14:34:49.273Z" },
+    { url = "https://files.pythonhosted.org/packages/23/07/63ffb386cd47029aa2916b3d2f454e6cc5b9f5c5ada3790377d5430084e7/xxhash-3.6.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:418daf3db71e1413cfe211c2f9a528456936645c17f46b5204705581a45390ae", size = 191431, upload-time = "2025-10-02T14:34:50.798Z" },
+    { url = "https://files.pythonhosted.org/packages/0f/93/14fde614cadb4ddf5e7cebf8918b7e8fac5ae7861c1875964f17e678205c/xxhash-3.6.0-cp312-cp312-win32.whl", hash = "sha256:50fc255f39428a27299c20e280d6193d8b63b8ef8028995323bf834a026b4fbb", size = 30617, upload-time = "2025-10-02T14:34:51.954Z" },
+    { url = "https://files.pythonhosted.org/packages/13/5d/0d125536cbe7565a83d06e43783389ecae0c0f2ed037b48ede185de477c0/xxhash-3.6.0-cp312-cp312-win_amd64.whl", hash = "sha256:c0f2ab8c715630565ab8991b536ecded9416d615538be8ecddce43ccf26cbc7c", size = 31534, upload-time = "2025-10-02T14:34:53.276Z" },
+    { url = "https://files.pythonhosted.org/packages/54/85/6ec269b0952ec7e36ba019125982cf11d91256a778c7c3f98a4c5043d283/xxhash-3.6.0-cp312-cp312-win_arm64.whl", hash = "sha256:eae5c13f3bc455a3bbb68bdc513912dc7356de7e2280363ea235f71f54064829", size = 27876, upload-time = "2025-10-02T14:34:54.371Z" },
+    { url = "https://files.pythonhosted.org/packages/33/76/35d05267ac82f53ae9b0e554da7c5e281ee61f3cad44c743f0fcd354f211/xxhash-3.6.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:599e64ba7f67472481ceb6ee80fa3bd828fd61ba59fb11475572cc5ee52b89ec", size = 32738, upload-time = "2025-10-02T14:34:55.839Z" },
+    { url = "https://files.pythonhosted.org/packages/31/a8/3fbce1cd96534a95e35d5120637bf29b0d7f5d8fa2f6374e31b4156dd419/xxhash-3.6.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:7d8b8aaa30fca4f16f0c84a5c8d7ddee0e25250ec2796c973775373257dde8f1", size = 30821, upload-time = "2025-10-02T14:34:57.219Z" },
+    { url = "https://files.pythonhosted.org/packages/0c/ea/d387530ca7ecfa183cb358027f1833297c6ac6098223fd14f9782cd0015c/xxhash-3.6.0-cp313-cp313-manylinux1_i686.manylinux_2_28_i686.manylinux_2_5_i686.whl", hash = "sha256:d597acf8506d6e7101a4a44a5e428977a51c0fadbbfd3c39650cca9253f6e5a6", size = 194127, upload-time = "2025-10-02T14:34:59.21Z" },
+    { url = "https://files.pythonhosted.org/packages/ba/0c/71435dcb99874b09a43b8d7c54071e600a7481e42b3e3ce1eb5226a5711a/xxhash-3.6.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:858dc935963a33bc33490128edc1c12b0c14d9c7ebaa4e387a7869ecc4f3e263", size = 212975, upload-time = "2025-10-02T14:35:00.816Z" },
+    { url = "https://files.pythonhosted.org/packages/84/7a/c2b3d071e4bb4a90b7057228a99b10d51744878f4a8a6dd643c8bd897620/xxhash-3.6.0-cp313-cp313-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:ba284920194615cb8edf73bf52236ce2e1664ccd4a38fdb543506413529cc546", size = 212241, upload-time = "2025-10-02T14:35:02.207Z" },
+    { url = "https://files.pythonhosted.org/packages/81/5f/640b6eac0128e215f177df99eadcd0f1b7c42c274ab6a394a05059694c5a/xxhash-3.6.0-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:4b54219177f6c6674d5378bd862c6aedf64725f70dd29c472eaae154df1a2e89", size = 445471, upload-time = "2025-10-02T14:35:03.61Z" },
+    { url = "https://files.pythonhosted.org/packages/5e/1e/3c3d3ef071b051cc3abbe3721ffb8365033a172613c04af2da89d5548a87/xxhash-3.6.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:42c36dd7dbad2f5238950c377fcbf6811b1cdb1c444fab447960030cea60504d", size = 193936, upload-time = "2025-10-02T14:35:05.013Z" },
+    { url = "https://files.pythonhosted.org/packages/2c/bd/4a5f68381939219abfe1c22a9e3a5854a4f6f6f3c4983a87d255f21f2e5d/xxhash-3.6.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:f22927652cba98c44639ffdc7aaf35828dccf679b10b31c4ad72a5b530a18eb7", size = 210440, upload-time = "2025-10-02T14:35:06.239Z" },
+    { url = "https://files.pythonhosted.org/packages/eb/37/b80fe3d5cfb9faff01a02121a0f4d565eb7237e9e5fc66e73017e74dcd36/xxhash-3.6.0-cp313-cp313-musllinux_1_2_i686.whl", hash = "sha256:b45fad44d9c5c119e9c6fbf2e1c656a46dc68e280275007bbfd3d572b21426db", size = 197990, upload-time = "2025-10-02T14:35:07.735Z" },
+    { url = "https://files.pythonhosted.org/packages/d7/fd/2c0a00c97b9e18f72e1f240ad4e8f8a90fd9d408289ba9c7c495ed7dc05c/xxhash-3.6.0-cp313-cp313-musllinux_1_2_ppc64le.whl", hash = "sha256:6f2580ffab1a8b68ef2b901cde7e55fa8da5e4be0977c68f78fc80f3c143de42", size = 210689, upload-time = "2025-10-02T14:35:09.438Z" },
+    { url = "https://files.pythonhosted.org/packages/93/86/5dd8076a926b9a95db3206aba20d89a7fc14dd5aac16e5c4de4b56033140/xxhash-3.6.0-cp313-cp313-musllinux_1_2_s390x.whl", hash = "sha256:40c391dd3cd041ebc3ffe6f2c862f402e306eb571422e0aa918d8070ba31da11", size = 414068, upload-time = "2025-10-02T14:35:11.162Z" },
+    { url = "https://files.pythonhosted.org/packages/af/3c/0bb129170ee8f3650f08e993baee550a09593462a5cddd8e44d0011102b1/xxhash-3.6.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:f205badabde7aafd1a31e8ca2a3e5a763107a71c397c4481d6a804eb5063d8bd", size = 191495, upload-time = "2025-10-02T14:35:12.971Z" },
+    { url = "https://files.pythonhosted.org/packages/e9/3a/6797e0114c21d1725e2577508e24006fd7ff1d8c0c502d3b52e45c1771d8/xxhash-3.6.0-cp313-cp313-win32.whl", hash = "sha256:2577b276e060b73b73a53042ea5bd5203d3e6347ce0d09f98500f418a9fcf799", size = 30620, upload-time = "2025-10-02T14:35:14.129Z" },
+    { url = "https://files.pythonhosted.org/packages/86/15/9bc32671e9a38b413a76d24722a2bf8784a132c043063a8f5152d390b0f9/xxhash-3.6.0-cp313-cp313-win_amd64.whl", hash = "sha256:757320d45d2fbcce8f30c42a6b2f47862967aea7bf458b9625b4bbe7ee390392", size = 31542, upload-time = "2025-10-02T14:35:15.21Z" },
+    { url = "https://files.pythonhosted.org/packages/39/c5/cc01e4f6188656e56112d6a8e0dfe298a16934b8c47a247236549a3f7695/xxhash-3.6.0-cp313-cp313-win_arm64.whl", hash = "sha256:457b8f85dec5825eed7b69c11ae86834a018b8e3df5e77783c999663da2f96d6", size = 27880, upload-time = "2025-10-02T14:35:16.315Z" },
+    { url = "https://files.pythonhosted.org/packages/f3/30/25e5321c8732759e930c555176d37e24ab84365482d257c3b16362235212/xxhash-3.6.0-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:a42e633d75cdad6d625434e3468126c73f13f7584545a9cf34e883aa1710e702", size = 32956, upload-time = "2025-10-02T14:35:17.413Z" },
+    { url = "https://files.pythonhosted.org/packages/9f/3c/0573299560d7d9f8ab1838f1efc021a280b5ae5ae2e849034ef3dee18810/xxhash-3.6.0-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:568a6d743219e717b07b4e03b0a828ce593833e498c3b64752e0f5df6bfe84db", size = 31072, upload-time = "2025-10-02T14:35:18.844Z" },
+    { url = "https://files.pythonhosted.org/packages/7a/1c/52d83a06e417cd9d4137722693424885cc9878249beb3a7c829e74bf7ce9/xxhash-3.6.0-cp313-cp313t-manylinux1_i686.manylinux_2_28_i686.manylinux_2_5_i686.whl", hash = "sha256:bec91b562d8012dae276af8025a55811b875baace6af510412a5e58e3121bc54", size = 196409, upload-time = "2025-10-02T14:35:20.31Z" },
+    { url = "https://files.pythonhosted.org/packages/e3/8e/c6d158d12a79bbd0b878f8355432075fc82759e356ab5a111463422a239b/xxhash-3.6.0-cp313-cp313t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:78e7f2f4c521c30ad5e786fdd6bae89d47a32672a80195467b5de0480aa97b1f", size = 215736, upload-time = "2025-10-02T14:35:21.616Z" },
+    { url = "https://files.pythonhosted.org/packages/bc/68/c4c80614716345d55071a396cf03d06e34b5f4917a467faf43083c995155/xxhash-3.6.0-cp313-cp313t-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:3ed0df1b11a79856df5ffcab572cbd6b9627034c1c748c5566fa79df9048a7c5", size = 214833, upload-time = "2025-10-02T14:35:23.32Z" },
+    { url = "https://files.pythonhosted.org/packages/7e/e9/ae27c8ffec8b953efa84c7c4a6c6802c263d587b9fc0d6e7cea64e08c3af/xxhash-3.6.0-cp313-cp313t-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:0e4edbfc7d420925b0dd5e792478ed393d6e75ff8fc219a6546fb446b6a417b1", size = 448348, upload-time = "2025-10-02T14:35:25.111Z" },
+    { url = "https://files.pythonhosted.org/packages/d7/6b/33e21afb1b5b3f46b74b6bd1913639066af218d704cc0941404ca717fc57/xxhash-3.6.0-cp313-cp313t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:fba27a198363a7ef87f8c0f6b171ec36b674fe9053742c58dd7e3201c1ab30ee", size = 196070, upload-time = "2025-10-02T14:35:26.586Z" },
+    { url = "https://files.pythonhosted.org/packages/96/b6/fcabd337bc5fa624e7203aa0fa7d0c49eed22f72e93229431752bddc83d9/xxhash-3.6.0-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:794fe9145fe60191c6532fa95063765529770edcdd67b3d537793e8004cabbfd", size = 212907, upload-time = "2025-10-02T14:35:28.087Z" },
+    { url = "https://files.pythonhosted.org/packages/4b/d3/9ee6160e644d660fcf176c5825e61411c7f62648728f69c79ba237250143/xxhash-3.6.0-cp313-cp313t-musllinux_1_2_i686.whl", hash = "sha256:6105ef7e62b5ac73a837778efc331a591d8442f8ef5c7e102376506cb4ae2729", size = 200839, upload-time = "2025-10-02T14:35:29.857Z" },
+    { url = "https://files.pythonhosted.org/packages/0d/98/e8de5baa5109394baf5118f5e72ab21a86387c4f89b0e77ef3e2f6b0327b/xxhash-3.6.0-cp313-cp313t-musllinux_1_2_ppc64le.whl", hash = "sha256:f01375c0e55395b814a679b3eea205db7919ac2af213f4a6682e01220e5fe292", size = 213304, upload-time = "2025-10-02T14:35:31.222Z" },
+    { url = "https://files.pythonhosted.org/packages/7b/1d/71056535dec5c3177eeb53e38e3d367dd1d16e024e63b1cee208d572a033/xxhash-3.6.0-cp313-cp313t-musllinux_1_2_s390x.whl", hash = "sha256:d706dca2d24d834a4661619dcacf51a75c16d65985718d6a7d73c1eeeb903ddf", size = 416930, upload-time = "2025-10-02T14:35:32.517Z" },
+    { url = "https://files.pythonhosted.org/packages/dc/6c/5cbde9de2cd967c322e651c65c543700b19e7ae3e0aae8ece3469bf9683d/xxhash-3.6.0-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:5f059d9faeacd49c0215d66f4056e1326c80503f51a1532ca336a385edadd033", size = 193787, upload-time = "2025-10-02T14:35:33.827Z" },
+    { url = "https://files.pythonhosted.org/packages/19/fa/0172e350361d61febcea941b0cc541d6e6c8d65d153e85f850a7b256ff8a/xxhash-3.6.0-cp313-cp313t-win32.whl", hash = "sha256:1244460adc3a9be84731d72b8e80625788e5815b68da3da8b83f78115a40a7ec", size = 30916, upload-time = "2025-10-02T14:35:35.107Z" },
+    { url = "https://files.pythonhosted.org/packages/ad/e6/e8cf858a2b19d6d45820f072eff1bea413910592ff17157cabc5f1227a16/xxhash-3.6.0-cp313-cp313t-win_amd64.whl", hash = "sha256:b1e420ef35c503869c4064f4a2f2b08ad6431ab7b229a05cce39d74268bca6b8", size = 31799, upload-time = "2025-10-02T14:35:36.165Z" },
+    { url = "https://files.pythonhosted.org/packages/56/15/064b197e855bfb7b343210e82490ae672f8bc7cdf3ddb02e92f64304ee8a/xxhash-3.6.0-cp313-cp313t-win_arm64.whl", hash = "sha256:ec44b73a4220623235f67a996c862049f375df3b1052d9899f40a6382c32d746", size = 28044, upload-time = "2025-10-02T14:35:37.195Z" },
+    { url = "https://files.pythonhosted.org/packages/7e/5e/0138bc4484ea9b897864d59fce9be9086030825bc778b76cb5a33a906d37/xxhash-3.6.0-cp314-cp314-macosx_10_13_x86_64.whl", hash = "sha256:a40a3d35b204b7cc7643cbcf8c9976d818cb47befcfac8bbefec8038ac363f3e", size = 32754, upload-time = "2025-10-02T14:35:38.245Z" },
+    { url = "https://files.pythonhosted.org/packages/18/d7/5dac2eb2ec75fd771957a13e5dda560efb2176d5203f39502a5fc571f899/xxhash-3.6.0-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:a54844be970d3fc22630b32d515e79a90d0a3ddb2644d8d7402e3c4c8da61405", size = 30846, upload-time = "2025-10-02T14:35:39.6Z" },
+    { url = "https://files.pythonhosted.org/packages/fe/71/8bc5be2bb00deb5682e92e8da955ebe5fa982da13a69da5a40a4c8db12fb/xxhash-3.6.0-cp314-cp314-manylinux1_i686.manylinux_2_28_i686.manylinux_2_5_i686.whl", hash = "sha256:016e9190af8f0a4e3741343777710e3d5717427f175adfdc3e72508f59e2a7f3", size = 194343, upload-time = "2025-10-02T14:35:40.69Z" },
+    { url = "https://files.pythonhosted.org/packages/e7/3b/52badfb2aecec2c377ddf1ae75f55db3ba2d321c5e164f14461c90837ef3/xxhash-3.6.0-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:4f6f72232f849eb9d0141e2ebe2677ece15adfd0fa599bc058aad83c714bb2c6", size = 213074, upload-time = "2025-10-02T14:35:42.29Z" },
+    { url = "https://files.pythonhosted.org/packages/a2/2b/ae46b4e9b92e537fa30d03dbc19cdae57ed407e9c26d163895e968e3de85/xxhash-3.6.0-cp314-cp314-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:63275a8aba7865e44b1813d2177e0f5ea7eadad3dd063a21f7cf9afdc7054063", size = 212388, upload-time = "2025-10-02T14:35:43.929Z" },
+    { url = "https://files.pythonhosted.org/packages/f5/80/49f88d3afc724b4ac7fbd664c8452d6db51b49915be48c6982659e0e7942/xxhash-3.6.0-cp314-cp314-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:3cd01fa2aa00d8b017c97eb46b9a794fbdca53fc14f845f5a328c71254b0abb7", size = 445614, upload-time = "2025-10-02T14:35:45.216Z" },
+    { url = "https://files.pythonhosted.org/packages/ed/ba/603ce3961e339413543d8cd44f21f2c80e2a7c5cfe692a7b1f2cccf58f3c/xxhash-3.6.0-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:0226aa89035b62b6a86d3c68df4d7c1f47a342b8683da2b60cedcddb46c4d95b", size = 194024, upload-time = "2025-10-02T14:35:46.959Z" },
+    { url = "https://files.pythonhosted.org/packages/78/d1/8e225ff7113bf81545cfdcd79eef124a7b7064a0bba53605ff39590b95c2/xxhash-3.6.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:c6e193e9f56e4ca4923c61238cdaced324f0feac782544eb4c6d55ad5cc99ddd", size = 210541, upload-time = "2025-10-02T14:35:48.301Z" },
+    { url = "https://files.pythonhosted.org/packages/6f/58/0f89d149f0bad89def1a8dd38feb50ccdeb643d9797ec84707091d4cb494/xxhash-3.6.0-cp314-cp314-musllinux_1_2_i686.whl", hash = "sha256:9176dcaddf4ca963d4deb93866d739a343c01c969231dbe21680e13a5d1a5bf0", size = 198305, upload-time = "2025-10-02T14:35:49.584Z" },
+    { url = "https://files.pythonhosted.org/packages/11/38/5eab81580703c4df93feb5f32ff8fa7fe1e2c51c1f183ee4e48d4bb9d3d7/xxhash-3.6.0-cp314-cp314-musllinux_1_2_ppc64le.whl", hash = "sha256:c1ce4009c97a752e682b897aa99aef84191077a9433eb237774689f14f8ec152", size = 210848, upload-time = "2025-10-02T14:35:50.877Z" },
+    { url = "https://files.pythonhosted.org/packages/5e/6b/953dc4b05c3ce678abca756416e4c130d2382f877a9c30a20d08ee6a77c0/xxhash-3.6.0-cp314-cp314-musllinux_1_2_s390x.whl", hash = "sha256:8cb2f4f679b01513b7adbb9b1b2f0f9cdc31b70007eaf9d59d0878809f385b11", size = 414142, upload-time = "2025-10-02T14:35:52.15Z" },
+    { url = "https://files.pythonhosted.org/packages/08/a9/238ec0d4e81a10eb5026d4a6972677cbc898ba6c8b9dbaec12ae001b1b35/xxhash-3.6.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:653a91d7c2ab54a92c19ccf43508b6a555440b9be1bc8be553376778be7f20b5", size = 191547, upload-time = "2025-10-02T14:35:53.547Z" },
+    { url = "https://files.pythonhosted.org/packages/f1/ee/3cf8589e06c2164ac77c3bf0aa127012801128f1feebf2a079272da5737c/xxhash-3.6.0-cp314-cp314-win32.whl", hash = "sha256:a756fe893389483ee8c394d06b5ab765d96e68fbbfe6fde7aa17e11f5720559f", size = 31214, upload-time = "2025-10-02T14:35:54.746Z" },
+    { url = "https://files.pythonhosted.org/packages/02/5d/a19552fbc6ad4cb54ff953c3908bbc095f4a921bc569433d791f755186f1/xxhash-3.6.0-cp314-cp314-win_amd64.whl", hash = "sha256:39be8e4e142550ef69629c9cd71b88c90e9a5db703fecbcf265546d9536ca4ad", size = 32290, upload-time = "2025-10-02T14:35:55.791Z" },
+    { url = "https://files.pythonhosted.org/packages/b1/11/dafa0643bc30442c887b55baf8e73353a344ee89c1901b5a5c54a6c17d39/xxhash-3.6.0-cp314-cp314-win_arm64.whl", hash = "sha256:25915e6000338999236f1eb68a02a32c3275ac338628a7eaa5a269c401995679", size = 28795, upload-time = "2025-10-02T14:35:57.162Z" },
+    { url = "https://files.pythonhosted.org/packages/2c/db/0e99732ed7f64182aef4a6fb145e1a295558deec2a746265dcdec12d191e/xxhash-3.6.0-cp314-cp314t-macosx_10_13_x86_64.whl", hash = "sha256:c5294f596a9017ca5a3e3f8884c00b91ab2ad2933cf288f4923c3fd4346cf3d4", size = 32955, upload-time = "2025-10-02T14:35:58.267Z" },
+    { url = "https://files.pythonhosted.org/packages/55/f4/2a7c3c68e564a099becfa44bb3d398810cc0ff6749b0d3cb8ccb93f23c14/xxhash-3.6.0-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:1cf9dcc4ab9cff01dfbba78544297a3a01dafd60f3bde4e2bfd016cf7e4ddc67", size = 31072, upload-time = "2025-10-02T14:35:59.382Z" },
+    { url = "https://files.pythonhosted.org/packages/c6/d9/72a29cddc7250e8a5819dad5d466facb5dc4c802ce120645630149127e73/xxhash-3.6.0-cp314-cp314t-manylinux1_i686.manylinux_2_28_i686.manylinux_2_5_i686.whl", hash = "sha256:01262da8798422d0685f7cef03b2bd3f4f46511b02830861df548d7def4402ad", size = 196579, upload-time = "2025-10-02T14:36:00.838Z" },
+    { url = "https://files.pythonhosted.org/packages/63/93/b21590e1e381040e2ca305a884d89e1c345b347404f7780f07f2cdd47ef4/xxhash-3.6.0-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:51a73fb7cb3a3ead9f7a8b583ffd9b8038e277cdb8cb87cf890e88b3456afa0b", size = 215854, upload-time = "2025-10-02T14:36:02.207Z" },
+    { url = "https://files.pythonhosted.org/packages/ce/b8/edab8a7d4fa14e924b29be877d54155dcbd8b80be85ea00d2be3413a9ed4/xxhash-3.6.0-cp314-cp314t-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:b9c6df83594f7df8f7f708ce5ebeacfc69f72c9fbaaababf6cf4758eaada0c9b", size = 214965, upload-time = "2025-10-02T14:36:03.507Z" },
+    { url = "https://files.pythonhosted.org/packages/27/67/dfa980ac7f0d509d54ea0d5a486d2bb4b80c3f1bb22b66e6a05d3efaf6c0/xxhash-3.6.0-cp314-cp314t-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:627f0af069b0ea56f312fd5189001c24578868643203bca1abbc2c52d3a6f3ca", size = 448484, upload-time = "2025-10-02T14:36:04.828Z" },
+    { url = "https://files.pythonhosted.org/packages/8c/63/8ffc2cc97e811c0ca5d00ab36604b3ea6f4254f20b7bc658ca825ce6c954/xxhash-3.6.0-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:aa912c62f842dfd013c5f21a642c9c10cd9f4c4e943e0af83618b4a404d9091a", size = 196162, upload-time = "2025-10-02T14:36:06.182Z" },
+    { url = "https://files.pythonhosted.org/packages/4b/77/07f0e7a3edd11a6097e990f6e5b815b6592459cb16dae990d967693e6ea9/xxhash-3.6.0-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:b465afd7909db30168ab62afe40b2fcf79eedc0b89a6c0ab3123515dc0df8b99", size = 213007, upload-time = "2025-10-02T14:36:07.733Z" },
+    { url = "https://files.pythonhosted.org/packages/ae/d8/bc5fa0d152837117eb0bef6f83f956c509332ce133c91c63ce07ee7c4873/xxhash-3.6.0-cp314-cp314t-musllinux_1_2_i686.whl", hash = "sha256:a881851cf38b0a70e7c4d3ce81fc7afd86fbc2a024f4cfb2a97cf49ce04b75d3", size = 200956, upload-time = "2025-10-02T14:36:09.106Z" },
+    { url = "https://files.pythonhosted.org/packages/26/a5/d749334130de9411783873e9b98ecc46688dad5db64ca6e04b02acc8b473/xxhash-3.6.0-cp314-cp314t-musllinux_1_2_ppc64le.whl", hash = "sha256:9b3222c686a919a0f3253cfc12bb118b8b103506612253b5baeaac10d8027cf6", size = 213401, upload-time = "2025-10-02T14:36:10.585Z" },
+    { url = "https://files.pythonhosted.org/packages/89/72/abed959c956a4bfc72b58c0384bb7940663c678127538634d896b1195c10/xxhash-3.6.0-cp314-cp314t-musllinux_1_2_s390x.whl", hash = "sha256:c5aa639bc113e9286137cec8fadc20e9cd732b2cc385c0b7fa673b84fc1f2a93", size = 417083, upload-time = "2025-10-02T14:36:12.276Z" },
+    { url = "https://files.pythonhosted.org/packages/0c/b3/62fd2b586283b7d7d665fb98e266decadf31f058f1cf6c478741f68af0cb/xxhash-3.6.0-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:5c1343d49ac102799905e115aee590183c3921d475356cb24b4de29a4bc56518", size = 193913, upload-time = "2025-10-02T14:36:14.025Z" },
+    { url = "https://files.pythonhosted.org/packages/9a/9a/c19c42c5b3f5a4aad748a6d5b4f23df3bed7ee5445accc65a0fb3ff03953/xxhash-3.6.0-cp314-cp314t-win32.whl", hash = "sha256:5851f033c3030dd95c086b4a36a2683c2ff4a799b23af60977188b057e467119", size = 31586, upload-time = "2025-10-02T14:36:15.603Z" },
+    { url = "https://files.pythonhosted.org/packages/03/d6/4cc450345be9924fd5dc8c590ceda1db5b43a0a889587b0ae81a95511360/xxhash-3.6.0-cp314-cp314t-win_amd64.whl", hash = "sha256:0444e7967dac37569052d2409b00a8860c2135cff05502df4da80267d384849f", size = 32526, upload-time = "2025-10-02T14:36:16.708Z" },
+    { url = "https://files.pythonhosted.org/packages/0f/c9/7243eb3f9eaabd1a88a5a5acadf06df2d83b100c62684b7425c6a11bcaa8/xxhash-3.6.0-cp314-cp314t-win_arm64.whl", hash = "sha256:bb79b1e63f6fd84ec778a4b1916dfe0a7c3fdb986c06addd5db3a0d413819d95", size = 28898, upload-time = "2025-10-02T14:36:17.843Z" },
+    { url = "https://files.pythonhosted.org/packages/93/1e/8aec23647a34a249f62e2398c42955acd9b4c6ed5cf08cbea94dc46f78d2/xxhash-3.6.0-pp311-pypy311_pp73-macosx_10_15_x86_64.whl", hash = "sha256:0f7b7e2ec26c1666ad5fc9dbfa426a6a3367ceaf79db5dd76264659d509d73b0", size = 30662, upload-time = "2025-10-02T14:37:01.743Z" },
+    { url = "https://files.pythonhosted.org/packages/b8/0b/b14510b38ba91caf43006209db846a696ceea6a847a0c9ba0a5b1adc53d6/xxhash-3.6.0-pp311-pypy311_pp73-manylinux1_i686.manylinux_2_28_i686.manylinux_2_5_i686.whl", hash = "sha256:5dc1e14d14fa0f5789ec29a7062004b5933964bb9b02aae6622b8f530dc40296", size = 41056, upload-time = "2025-10-02T14:37:02.879Z" },
+    { url = "https://files.pythonhosted.org/packages/50/55/15a7b8a56590e66ccd374bbfa3f9ffc45b810886c8c3b614e3f90bd2367c/xxhash-3.6.0-pp311-pypy311_pp73-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:881b47fc47e051b37d94d13e7455131054b56749b91b508b0907eb07900d1c13", size = 36251, upload-time = "2025-10-02T14:37:04.44Z" },
+    { url = "https://files.pythonhosted.org/packages/62/b2/5ac99a041a29e58e95f907876b04f7067a0242cb85b5f39e726153981503/xxhash-3.6.0-pp311-pypy311_pp73-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:c6dc31591899f5e5666f04cc2e529e69b4072827085c1ef15294d91a004bc1bd", size = 32481, upload-time = "2025-10-02T14:37:05.869Z" },
+    { url = "https://files.pythonhosted.org/packages/7b/d9/8d95e906764a386a3d3b596f3c68bb63687dfca806373509f51ce8eea81f/xxhash-3.6.0-pp311-pypy311_pp73-win_amd64.whl", hash = "sha256:15e0dac10eb9309508bfc41f7f9deaa7755c69e35af835db9cb10751adebc35d", size = 31565, upload-time = "2025-10-02T14:37:06.966Z" },
+]
+
+[[package]]
+name = "yarl"
+version = "1.22.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "idna" },
+    { name = "multidict" },
+    { name = "propcache" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/57/63/0c6ebca57330cd313f6102b16dd57ffaf3ec4c83403dcb45dbd15c6f3ea1/yarl-1.22.0.tar.gz", hash = "sha256:bebf8557577d4401ba8bd9ff33906f1376c877aa78d1fe216ad01b4d6745af71", size = 187169, upload-time = "2025-10-06T14:12:55.963Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/d1/43/a2204825342f37c337f5edb6637040fa14e365b2fcc2346960201d457579/yarl-1.22.0-cp310-cp310-macosx_10_9_universal2.whl", hash = "sha256:c7bd6683587567e5a49ee6e336e0612bec8329be1b7d4c8af5687dcdeb67ee1e", size = 140517, upload-time = "2025-10-06T14:08:42.494Z" },
+    { url = "https://files.pythonhosted.org/packages/44/6f/674f3e6f02266428c56f704cd2501c22f78e8b2eeb23f153117cc86fb28a/yarl-1.22.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:5cdac20da754f3a723cceea5b3448e1a2074866406adeb4ef35b469d089adb8f", size = 93495, upload-time = "2025-10-06T14:08:46.2Z" },
+    { url = "https://files.pythonhosted.org/packages/b8/12/5b274d8a0f30c07b91b2f02cba69152600b47830fcfb465c108880fcee9c/yarl-1.22.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:07a524d84df0c10f41e3ee918846e1974aba4ec017f990dc735aad487a0bdfdf", size = 94400, upload-time = "2025-10-06T14:08:47.855Z" },
+    { url = "https://files.pythonhosted.org/packages/e2/7f/df1b6949b1fa1aa9ff6de6e2631876ad4b73c4437822026e85d8acb56bb1/yarl-1.22.0-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:e1b329cb8146d7b736677a2440e422eadd775d1806a81db2d4cded80a48efc1a", size = 347545, upload-time = "2025-10-06T14:08:49.683Z" },
+    { url = "https://files.pythonhosted.org/packages/84/09/f92ed93bd6cd77872ab6c3462df45ca45cd058d8f1d0c9b4f54c1704429f/yarl-1.22.0-cp310-cp310-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:75976c6945d85dbb9ee6308cd7ff7b1fb9409380c82d6119bd778d8fcfe2931c", size = 319598, upload-time = "2025-10-06T14:08:51.215Z" },
+    { url = "https://files.pythonhosted.org/packages/c3/97/ac3f3feae7d522cf7ccec3d340bb0b2b61c56cb9767923df62a135092c6b/yarl-1.22.0-cp310-cp310-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:80ddf7a5f8c86cb3eb4bc9028b07bbbf1f08a96c5c0bc1244be5e8fefcb94147", size = 363893, upload-time = "2025-10-06T14:08:53.144Z" },
+    { url = "https://files.pythonhosted.org/packages/06/49/f3219097403b9c84a4d079b1d7bda62dd9b86d0d6e4428c02d46ab2c77fc/yarl-1.22.0-cp310-cp310-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:d332fc2e3c94dad927f2112395772a4e4fedbcf8f80efc21ed7cdfae4d574fdb", size = 371240, upload-time = "2025-10-06T14:08:55.036Z" },
+    { url = "https://files.pythonhosted.org/packages/35/9f/06b765d45c0e44e8ecf0fe15c9eacbbde342bb5b7561c46944f107bfb6c3/yarl-1.22.0-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:0cf71bf877efeac18b38d3930594c0948c82b64547c1cf420ba48722fe5509f6", size = 346965, upload-time = "2025-10-06T14:08:56.722Z" },
+    { url = "https://files.pythonhosted.org/packages/c5/69/599e7cea8d0fcb1694323b0db0dda317fa3162f7b90166faddecf532166f/yarl-1.22.0-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:663e1cadaddae26be034a6ab6072449a8426ddb03d500f43daf952b74553bba0", size = 342026, upload-time = "2025-10-06T14:08:58.563Z" },
+    { url = "https://files.pythonhosted.org/packages/95/6f/9dfd12c8bc90fea9eab39832ee32ea48f8e53d1256252a77b710c065c89f/yarl-1.22.0-cp310-cp310-musllinux_1_2_armv7l.whl", hash = "sha256:6dcbb0829c671f305be48a7227918cfcd11276c2d637a8033a99a02b67bf9eda", size = 335637, upload-time = "2025-10-06T14:09:00.506Z" },
+    { url = "https://files.pythonhosted.org/packages/57/2e/34c5b4eb9b07e16e873db5b182c71e5f06f9b5af388cdaa97736d79dd9a6/yarl-1.22.0-cp310-cp310-musllinux_1_2_ppc64le.whl", hash = "sha256:f0d97c18dfd9a9af4490631905a3f131a8e4c9e80a39353919e2cfed8f00aedc", size = 359082, upload-time = "2025-10-06T14:09:01.936Z" },
+    { url = "https://files.pythonhosted.org/packages/31/71/fa7e10fb772d273aa1f096ecb8ab8594117822f683bab7d2c5a89914c92a/yarl-1.22.0-cp310-cp310-musllinux_1_2_s390x.whl", hash = "sha256:437840083abe022c978470b942ff832c3940b2ad3734d424b7eaffcd07f76737", size = 357811, upload-time = "2025-10-06T14:09:03.445Z" },
+    { url = "https://files.pythonhosted.org/packages/26/da/11374c04e8e1184a6a03cf9c8f5688d3e5cec83ed6f31ad3481b3207f709/yarl-1.22.0-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:a899cbd98dce6f5d8de1aad31cb712ec0a530abc0a86bd6edaa47c1090138467", size = 351223, upload-time = "2025-10-06T14:09:05.401Z" },
+    { url = "https://files.pythonhosted.org/packages/82/8f/e2d01f161b0c034a30410e375e191a5d27608c1f8693bab1a08b089ca096/yarl-1.22.0-cp310-cp310-win32.whl", hash = "sha256:595697f68bd1f0c1c159fcb97b661fc9c3f5db46498043555d04805430e79bea", size = 82118, upload-time = "2025-10-06T14:09:11.148Z" },
+    { url = "https://files.pythonhosted.org/packages/62/46/94c76196642dbeae634c7a61ba3da88cd77bed875bf6e4a8bed037505aa6/yarl-1.22.0-cp310-cp310-win_amd64.whl", hash = "sha256:cb95a9b1adaa48e41815a55ae740cfda005758104049a640a398120bf02515ca", size = 86852, upload-time = "2025-10-06T14:09:12.958Z" },
+    { url = "https://files.pythonhosted.org/packages/af/af/7df4f179d3b1a6dcb9a4bd2ffbc67642746fcafdb62580e66876ce83fff4/yarl-1.22.0-cp310-cp310-win_arm64.whl", hash = "sha256:b85b982afde6df99ecc996990d4ad7ccbdbb70e2a4ba4de0aecde5922ba98a0b", size = 82012, upload-time = "2025-10-06T14:09:14.664Z" },
+    { url = "https://files.pythonhosted.org/packages/4d/27/5ab13fc84c76a0250afd3d26d5936349a35be56ce5785447d6c423b26d92/yarl-1.22.0-cp311-cp311-macosx_10_9_universal2.whl", hash = "sha256:1ab72135b1f2db3fed3997d7e7dc1b80573c67138023852b6efb336a5eae6511", size = 141607, upload-time = "2025-10-06T14:09:16.298Z" },
+    { url = "https://files.pythonhosted.org/packages/6a/a1/d065d51d02dc02ce81501d476b9ed2229d9a990818332242a882d5d60340/yarl-1.22.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:669930400e375570189492dc8d8341301578e8493aec04aebc20d4717f899dd6", size = 94027, upload-time = "2025-10-06T14:09:17.786Z" },
+    { url = "https://files.pythonhosted.org/packages/c1/da/8da9f6a53f67b5106ffe902c6fa0164e10398d4e150d85838b82f424072a/yarl-1.22.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:792a2af6d58177ef7c19cbf0097aba92ca1b9cb3ffdd9c7470e156c8f9b5e028", size = 94963, upload-time = "2025-10-06T14:09:19.662Z" },
+    { url = "https://files.pythonhosted.org/packages/68/fe/2c1f674960c376e29cb0bec1249b117d11738db92a6ccc4a530b972648db/yarl-1.22.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:3ea66b1c11c9150f1372f69afb6b8116f2dd7286f38e14ea71a44eee9ec51b9d", size = 368406, upload-time = "2025-10-06T14:09:21.402Z" },
+    { url = "https://files.pythonhosted.org/packages/95/26/812a540e1c3c6418fec60e9bbd38e871eaba9545e94fa5eff8f4a8e28e1e/yarl-1.22.0-cp311-cp311-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:3e2daa88dc91870215961e96a039ec73e4937da13cf77ce17f9cad0c18df3503", size = 336581, upload-time = "2025-10-06T14:09:22.98Z" },
+    { url = "https://files.pythonhosted.org/packages/0b/f5/5777b19e26fdf98563985e481f8be3d8a39f8734147a6ebf459d0dab5a6b/yarl-1.22.0-cp311-cp311-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:ba440ae430c00eee41509353628600212112cd5018d5def7e9b05ea7ac34eb65", size = 388924, upload-time = "2025-10-06T14:09:24.655Z" },
+    { url = "https://files.pythonhosted.org/packages/86/08/24bd2477bd59c0bbd994fe1d93b126e0472e4e3df5a96a277b0a55309e89/yarl-1.22.0-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:e6438cc8f23a9c1478633d216b16104a586b9761db62bfacb6425bac0a36679e", size = 392890, upload-time = "2025-10-06T14:09:26.617Z" },
+    { url = "https://files.pythonhosted.org/packages/46/00/71b90ed48e895667ecfb1eaab27c1523ee2fa217433ed77a73b13205ca4b/yarl-1.22.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:4c52a6e78aef5cf47a98ef8e934755abf53953379b7d53e68b15ff4420e6683d", size = 365819, upload-time = "2025-10-06T14:09:28.544Z" },
+    { url = "https://files.pythonhosted.org/packages/30/2d/f715501cae832651d3282387c6a9236cd26bd00d0ff1e404b3dc52447884/yarl-1.22.0-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:3b06bcadaac49c70f4c88af4ffcfbe3dc155aab3163e75777818092478bcbbe7", size = 363601, upload-time = "2025-10-06T14:09:30.568Z" },
+    { url = "https://files.pythonhosted.org/packages/f8/f9/a678c992d78e394e7126ee0b0e4e71bd2775e4334d00a9278c06a6cce96a/yarl-1.22.0-cp311-cp311-musllinux_1_2_armv7l.whl", hash = "sha256:6944b2dc72c4d7f7052683487e3677456050ff77fcf5e6204e98caf785ad1967", size = 358072, upload-time = "2025-10-06T14:09:32.528Z" },
+    { url = "https://files.pythonhosted.org/packages/2c/d1/b49454411a60edb6fefdcad4f8e6dbba7d8019e3a508a1c5836cba6d0781/yarl-1.22.0-cp311-cp311-musllinux_1_2_ppc64le.whl", hash = "sha256:d5372ca1df0f91a86b047d1277c2aaf1edb32d78bbcefffc81b40ffd18f027ed", size = 385311, upload-time = "2025-10-06T14:09:34.634Z" },
+    { url = "https://files.pythonhosted.org/packages/87/e5/40d7a94debb8448c7771a916d1861d6609dddf7958dc381117e7ba36d9e8/yarl-1.22.0-cp311-cp311-musllinux_1_2_s390x.whl", hash = "sha256:51af598701f5299012b8416486b40fceef8c26fc87dc6d7d1f6fc30609ea0aa6", size = 381094, upload-time = "2025-10-06T14:09:36.268Z" },
+    { url = "https://files.pythonhosted.org/packages/35/d8/611cc282502381ad855448643e1ad0538957fc82ae83dfe7762c14069e14/yarl-1.22.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:b266bd01fedeffeeac01a79ae181719ff848a5a13ce10075adbefc8f1daee70e", size = 370944, upload-time = "2025-10-06T14:09:37.872Z" },
+    { url = "https://files.pythonhosted.org/packages/2d/df/fadd00fb1c90e1a5a8bd731fa3d3de2e165e5a3666a095b04e31b04d9cb6/yarl-1.22.0-cp311-cp311-win32.whl", hash = "sha256:a9b1ba5610a4e20f655258d5a1fdc7ebe3d837bb0e45b581398b99eb98b1f5ca", size = 81804, upload-time = "2025-10-06T14:09:39.359Z" },
+    { url = "https://files.pythonhosted.org/packages/b5/f7/149bb6f45f267cb5c074ac40c01c6b3ea6d8a620d34b337f6321928a1b4d/yarl-1.22.0-cp311-cp311-win_amd64.whl", hash = "sha256:078278b9b0b11568937d9509b589ee83ef98ed6d561dfe2020e24a9fd08eaa2b", size = 86858, upload-time = "2025-10-06T14:09:41.068Z" },
+    { url = "https://files.pythonhosted.org/packages/2b/13/88b78b93ad3f2f0b78e13bfaaa24d11cbc746e93fe76d8c06bf139615646/yarl-1.22.0-cp311-cp311-win_arm64.whl", hash = "sha256:b6a6f620cfe13ccec221fa312139135166e47ae169f8253f72a0abc0dae94376", size = 81637, upload-time = "2025-10-06T14:09:42.712Z" },
+    { url = "https://files.pythonhosted.org/packages/75/ff/46736024fee3429b80a165a732e38e5d5a238721e634ab41b040d49f8738/yarl-1.22.0-cp312-cp312-macosx_10_13_universal2.whl", hash = "sha256:e340382d1afa5d32b892b3ff062436d592ec3d692aeea3bef3a5cfe11bbf8c6f", size = 142000, upload-time = "2025-10-06T14:09:44.631Z" },
+    { url = "https://files.pythonhosted.org/packages/5a/9a/b312ed670df903145598914770eb12de1bac44599549b3360acc96878df8/yarl-1.22.0-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:f1e09112a2c31ffe8d80be1b0988fa6a18c5d5cad92a9ffbb1c04c91bfe52ad2", size = 94338, upload-time = "2025-10-06T14:09:46.372Z" },
+    { url = "https://files.pythonhosted.org/packages/ba/f5/0601483296f09c3c65e303d60c070a5c19fcdbc72daa061e96170785bc7d/yarl-1.22.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:939fe60db294c786f6b7c2d2e121576628468f65453d86b0fe36cb52f987bd74", size = 94909, upload-time = "2025-10-06T14:09:48.648Z" },
+    { url = "https://files.pythonhosted.org/packages/60/41/9a1fe0b73dbcefce72e46cf149b0e0a67612d60bfc90fb59c2b2efdfbd86/yarl-1.22.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:e1651bf8e0398574646744c1885a41198eba53dc8a9312b954073f845c90a8df", size = 372940, upload-time = "2025-10-06T14:09:50.089Z" },
+    { url = "https://files.pythonhosted.org/packages/17/7a/795cb6dfee561961c30b800f0ed616b923a2ec6258b5def2a00bf8231334/yarl-1.22.0-cp312-cp312-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:b8a0588521a26bf92a57a1705b77b8b59044cdceccac7151bd8d229e66b8dedb", size = 345825, upload-time = "2025-10-06T14:09:52.142Z" },
+    { url = "https://files.pythonhosted.org/packages/d7/93/a58f4d596d2be2ae7bab1a5846c4d270b894958845753b2c606d666744d3/yarl-1.22.0-cp312-cp312-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:42188e6a615c1a75bcaa6e150c3fe8f3e8680471a6b10150c5f7e83f47cc34d2", size = 386705, upload-time = "2025-10-06T14:09:54.128Z" },
+    { url = "https://files.pythonhosted.org/packages/61/92/682279d0e099d0e14d7fd2e176bd04f48de1484f56546a3e1313cd6c8e7c/yarl-1.22.0-cp312-cp312-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:f6d2cb59377d99718913ad9a151030d6f83ef420a2b8f521d94609ecc106ee82", size = 396518, upload-time = "2025-10-06T14:09:55.762Z" },
+    { url = "https://files.pythonhosted.org/packages/db/0f/0d52c98b8a885aeda831224b78f3be7ec2e1aa4a62091f9f9188c3c65b56/yarl-1.22.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:50678a3b71c751d58d7908edc96d332af328839eea883bb554a43f539101277a", size = 377267, upload-time = "2025-10-06T14:09:57.958Z" },
+    { url = "https://files.pythonhosted.org/packages/22/42/d2685e35908cbeaa6532c1fc73e89e7f2efb5d8a7df3959ea8e37177c5a3/yarl-1.22.0-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:1e8fbaa7cec507aa24ea27a01456e8dd4b6fab829059b69844bd348f2d467124", size = 365797, upload-time = "2025-10-06T14:09:59.527Z" },
+    { url = "https://files.pythonhosted.org/packages/a2/83/cf8c7bcc6355631762f7d8bdab920ad09b82efa6b722999dfb05afa6cfac/yarl-1.22.0-cp312-cp312-musllinux_1_2_armv7l.whl", hash = "sha256:433885ab5431bc3d3d4f2f9bd15bfa1614c522b0f1405d62c4f926ccd69d04fa", size = 365535, upload-time = "2025-10-06T14:10:01.139Z" },
+    { url = "https://files.pythonhosted.org/packages/25/e1/5302ff9b28f0c59cac913b91fe3f16c59a033887e57ce9ca5d41a3a94737/yarl-1.22.0-cp312-cp312-musllinux_1_2_ppc64le.whl", hash = "sha256:b790b39c7e9a4192dc2e201a282109ed2985a1ddbd5ac08dc56d0e121400a8f7", size = 382324, upload-time = "2025-10-06T14:10:02.756Z" },
+    { url = "https://files.pythonhosted.org/packages/bf/cd/4617eb60f032f19ae3a688dc990d8f0d89ee0ea378b61cac81ede3e52fae/yarl-1.22.0-cp312-cp312-musllinux_1_2_s390x.whl", hash = "sha256:31f0b53913220599446872d757257be5898019c85e7971599065bc55065dc99d", size = 383803, upload-time = "2025-10-06T14:10:04.552Z" },
+    { url = "https://files.pythonhosted.org/packages/59/65/afc6e62bb506a319ea67b694551dab4a7e6fb7bf604e9bd9f3e11d575fec/yarl-1.22.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:a49370e8f711daec68d09b821a34e1167792ee2d24d405cbc2387be4f158b520", size = 374220, upload-time = "2025-10-06T14:10:06.489Z" },
+    { url = "https://files.pythonhosted.org/packages/e7/3d/68bf18d50dc674b942daec86a9ba922d3113d8399b0e52b9897530442da2/yarl-1.22.0-cp312-cp312-win32.whl", hash = "sha256:70dfd4f241c04bd9239d53b17f11e6ab672b9f1420364af63e8531198e3f5fe8", size = 81589, upload-time = "2025-10-06T14:10:09.254Z" },
+    { url = "https://files.pythonhosted.org/packages/c8/9a/6ad1a9b37c2f72874f93e691b2e7ecb6137fb2b899983125db4204e47575/yarl-1.22.0-cp312-cp312-win_amd64.whl", hash = "sha256:8884d8b332a5e9b88e23f60bb166890009429391864c685e17bd73a9eda9105c", size = 87213, upload-time = "2025-10-06T14:10:11.369Z" },
+    { url = "https://files.pythonhosted.org/packages/44/c5/c21b562d1680a77634d748e30c653c3ca918beb35555cff24986fff54598/yarl-1.22.0-cp312-cp312-win_arm64.whl", hash = "sha256:ea70f61a47f3cc93bdf8b2f368ed359ef02a01ca6393916bc8ff877427181e74", size = 81330, upload-time = "2025-10-06T14:10:13.112Z" },
+    { url = "https://files.pythonhosted.org/packages/ea/f3/d67de7260456ee105dc1d162d43a019ecad6b91e2f51809d6cddaa56690e/yarl-1.22.0-cp313-cp313-macosx_10_13_universal2.whl", hash = "sha256:8dee9c25c74997f6a750cd317b8ca63545169c098faee42c84aa5e506c819b53", size = 139980, upload-time = "2025-10-06T14:10:14.601Z" },
+    { url = "https://files.pythonhosted.org/packages/01/88/04d98af0b47e0ef42597b9b28863b9060bb515524da0a65d5f4db160b2d5/yarl-1.22.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:01e73b85a5434f89fc4fe27dcda2aff08ddf35e4d47bbbea3bdcd25321af538a", size = 93424, upload-time = "2025-10-06T14:10:16.115Z" },
+    { url = "https://files.pythonhosted.org/packages/18/91/3274b215fd8442a03975ce6bee5fe6aa57a8326b29b9d3d56234a1dca244/yarl-1.22.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:22965c2af250d20c873cdbee8ff958fb809940aeb2e74ba5f20aaf6b7ac8c70c", size = 93821, upload-time = "2025-10-06T14:10:17.993Z" },
+    { url = "https://files.pythonhosted.org/packages/61/3a/caf4e25036db0f2da4ca22a353dfeb3c9d3c95d2761ebe9b14df8fc16eb0/yarl-1.22.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:b4f15793aa49793ec8d1c708ab7f9eded1aa72edc5174cae703651555ed1b601", size = 373243, upload-time = "2025-10-06T14:10:19.44Z" },
+    { url = "https://files.pythonhosted.org/packages/6e/9e/51a77ac7516e8e7803b06e01f74e78649c24ee1021eca3d6a739cb6ea49c/yarl-1.22.0-cp313-cp313-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:e5542339dcf2747135c5c85f68680353d5cb9ffd741c0f2e8d832d054d41f35a", size = 342361, upload-time = "2025-10-06T14:10:21.124Z" },
+    { url = "https://files.pythonhosted.org/packages/d4/f8/33b92454789dde8407f156c00303e9a891f1f51a0330b0fad7c909f87692/yarl-1.22.0-cp313-cp313-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:5c401e05ad47a75869c3ab3e35137f8468b846770587e70d71e11de797d113df", size = 387036, upload-time = "2025-10-06T14:10:22.902Z" },
+    { url = "https://files.pythonhosted.org/packages/d9/9a/c5db84ea024f76838220280f732970aa4ee154015d7f5c1bfb60a267af6f/yarl-1.22.0-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:243dda95d901c733f5b59214d28b0120893d91777cb8aa043e6ef059d3cddfe2", size = 397671, upload-time = "2025-10-06T14:10:24.523Z" },
+    { url = "https://files.pythonhosted.org/packages/11/c9/cd8538dc2e7727095e0c1d867bad1e40c98f37763e6d995c1939f5fdc7b1/yarl-1.22.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:bec03d0d388060058f5d291a813f21c011041938a441c593374da6077fe21b1b", size = 377059, upload-time = "2025-10-06T14:10:26.406Z" },
+    { url = "https://files.pythonhosted.org/packages/a1/b9/ab437b261702ced75122ed78a876a6dec0a1b0f5e17a4ac7a9a2482d8abe/yarl-1.22.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:b0748275abb8c1e1e09301ee3cf90c8a99678a4e92e4373705f2a2570d581273", size = 365356, upload-time = "2025-10-06T14:10:28.461Z" },
+    { url = "https://files.pythonhosted.org/packages/b2/9d/8e1ae6d1d008a9567877b08f0ce4077a29974c04c062dabdb923ed98e6fe/yarl-1.22.0-cp313-cp313-musllinux_1_2_armv7l.whl", hash = "sha256:47fdb18187e2a4e18fda2c25c05d8251a9e4a521edaed757fef033e7d8498d9a", size = 361331, upload-time = "2025-10-06T14:10:30.541Z" },
+    { url = "https://files.pythonhosted.org/packages/ca/5a/09b7be3905962f145b73beb468cdd53db8aa171cf18c80400a54c5b82846/yarl-1.22.0-cp313-cp313-musllinux_1_2_ppc64le.whl", hash = "sha256:c7044802eec4524fde550afc28edda0dd5784c4c45f0be151a2d3ba017daca7d", size = 382590, upload-time = "2025-10-06T14:10:33.352Z" },
+    { url = "https://files.pythonhosted.org/packages/aa/7f/59ec509abf90eda5048b0bc3e2d7b5099dffdb3e6b127019895ab9d5ef44/yarl-1.22.0-cp313-cp313-musllinux_1_2_s390x.whl", hash = "sha256:139718f35149ff544caba20fce6e8a2f71f1e39b92c700d8438a0b1d2a631a02", size = 385316, upload-time = "2025-10-06T14:10:35.034Z" },
+    { url = "https://files.pythonhosted.org/packages/e5/84/891158426bc8036bfdfd862fabd0e0fa25df4176ec793e447f4b85cf1be4/yarl-1.22.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:e1b51bebd221006d3d2f95fbe124b22b247136647ae5dcc8c7acafba66e5ee67", size = 374431, upload-time = "2025-10-06T14:10:37.76Z" },
+    { url = "https://files.pythonhosted.org/packages/bb/49/03da1580665baa8bef5e8ed34c6df2c2aca0a2f28bf397ed238cc1bbc6f2/yarl-1.22.0-cp313-cp313-win32.whl", hash = "sha256:d3e32536234a95f513bd374e93d717cf6b2231a791758de6c509e3653f234c95", size = 81555, upload-time = "2025-10-06T14:10:39.649Z" },
+    { url = "https://files.pythonhosted.org/packages/9a/ee/450914ae11b419eadd067c6183ae08381cfdfcb9798b90b2b713bbebddda/yarl-1.22.0-cp313-cp313-win_amd64.whl", hash = "sha256:47743b82b76d89a1d20b83e60d5c20314cbd5ba2befc9cda8f28300c4a08ed4d", size = 86965, upload-time = "2025-10-06T14:10:41.313Z" },
+    { url = "https://files.pythonhosted.org/packages/98/4d/264a01eae03b6cf629ad69bae94e3b0e5344741e929073678e84bf7a3e3b/yarl-1.22.0-cp313-cp313-win_arm64.whl", hash = "sha256:5d0fcda9608875f7d052eff120c7a5da474a6796fe4d83e152e0e4d42f6d1a9b", size = 81205, upload-time = "2025-10-06T14:10:43.167Z" },
+    { url = "https://files.pythonhosted.org/packages/88/fc/6908f062a2f77b5f9f6d69cecb1747260831ff206adcbc5b510aff88df91/yarl-1.22.0-cp313-cp313t-macosx_10_13_universal2.whl", hash = "sha256:719ae08b6972befcba4310e49edb1161a88cdd331e3a694b84466bd938a6ab10", size = 146209, upload-time = "2025-10-06T14:10:44.643Z" },
+    { url = "https://files.pythonhosted.org/packages/65/47/76594ae8eab26210b4867be6f49129861ad33da1f1ebdf7051e98492bf62/yarl-1.22.0-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:47d8a5c446df1c4db9d21b49619ffdba90e77c89ec6e283f453856c74b50b9e3", size = 95966, upload-time = "2025-10-06T14:10:46.554Z" },
+    { url = "https://files.pythonhosted.org/packages/ab/ce/05e9828a49271ba6b5b038b15b3934e996980dd78abdfeb52a04cfb9467e/yarl-1.22.0-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:cfebc0ac8333520d2d0423cbbe43ae43c8838862ddb898f5ca68565e395516e9", size = 97312, upload-time = "2025-10-06T14:10:48.007Z" },
+    { url = "https://files.pythonhosted.org/packages/d1/c5/7dffad5e4f2265b29c9d7ec869c369e4223166e4f9206fc2243ee9eea727/yarl-1.22.0-cp313-cp313t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:4398557cbf484207df000309235979c79c4356518fd5c99158c7d38203c4da4f", size = 361967, upload-time = "2025-10-06T14:10:49.997Z" },
+    { url = "https://files.pythonhosted.org/packages/50/b2/375b933c93a54bff7fc041e1a6ad2c0f6f733ffb0c6e642ce56ee3b39970/yarl-1.22.0-cp313-cp313t-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:2ca6fd72a8cd803be290d42f2dec5cdcd5299eeb93c2d929bf060ad9efaf5de0", size = 323949, upload-time = "2025-10-06T14:10:52.004Z" },
+    { url = "https://files.pythonhosted.org/packages/66/50/bfc2a29a1d78644c5a7220ce2f304f38248dc94124a326794e677634b6cf/yarl-1.22.0-cp313-cp313t-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:ca1f59c4e1ab6e72f0a23c13fca5430f889634166be85dbf1013683e49e3278e", size = 361818, upload-time = "2025-10-06T14:10:54.078Z" },
+    { url = "https://files.pythonhosted.org/packages/46/96/f3941a46af7d5d0f0498f86d71275696800ddcdd20426298e572b19b91ff/yarl-1.22.0-cp313-cp313t-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:6c5010a52015e7c70f86eb967db0f37f3c8bd503a695a49f8d45700144667708", size = 372626, upload-time = "2025-10-06T14:10:55.767Z" },
+    { url = "https://files.pythonhosted.org/packages/c1/42/8b27c83bb875cd89448e42cd627e0fb971fa1675c9ec546393d18826cb50/yarl-1.22.0-cp313-cp313t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:9d7672ecf7557476642c88497c2f8d8542f8e36596e928e9bcba0e42e1e7d71f", size = 341129, upload-time = "2025-10-06T14:10:57.985Z" },
+    { url = "https://files.pythonhosted.org/packages/49/36/99ca3122201b382a3cf7cc937b95235b0ac944f7e9f2d5331d50821ed352/yarl-1.22.0-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:3b7c88eeef021579d600e50363e0b6ee4f7f6f728cd3486b9d0f3ee7b946398d", size = 346776, upload-time = "2025-10-06T14:10:59.633Z" },
+    { url = "https://files.pythonhosted.org/packages/85/b4/47328bf996acd01a4c16ef9dcd2f59c969f495073616586f78cd5f2efb99/yarl-1.22.0-cp313-cp313t-musllinux_1_2_armv7l.whl", hash = "sha256:f4afb5c34f2c6fecdcc182dfcfc6af6cccf1aa923eed4d6a12e9d96904e1a0d8", size = 334879, upload-time = "2025-10-06T14:11:01.454Z" },
+    { url = "https://files.pythonhosted.org/packages/c2/ad/b77d7b3f14a4283bffb8e92c6026496f6de49751c2f97d4352242bba3990/yarl-1.22.0-cp313-cp313t-musllinux_1_2_ppc64le.whl", hash = "sha256:59c189e3e99a59cf8d83cbb31d4db02d66cda5a1a4374e8a012b51255341abf5", size = 350996, upload-time = "2025-10-06T14:11:03.452Z" },
+    { url = "https://files.pythonhosted.org/packages/81/c8/06e1d69295792ba54d556f06686cbd6a7ce39c22307100e3fb4a2c0b0a1d/yarl-1.22.0-cp313-cp313t-musllinux_1_2_s390x.whl", hash = "sha256:5a3bf7f62a289fa90f1990422dc8dff5a458469ea71d1624585ec3a4c8d6960f", size = 356047, upload-time = "2025-10-06T14:11:05.115Z" },
+    { url = "https://files.pythonhosted.org/packages/4b/b8/4c0e9e9f597074b208d18cef227d83aac36184bfbc6eab204ea55783dbc5/yarl-1.22.0-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:de6b9a04c606978fdfe72666fa216ffcf2d1a9f6a381058d4378f8d7b1e5de62", size = 342947, upload-time = "2025-10-06T14:11:08.137Z" },
+    { url = "https://files.pythonhosted.org/packages/e0/e5/11f140a58bf4c6ad7aca69a892bff0ee638c31bea4206748fc0df4ebcb3a/yarl-1.22.0-cp313-cp313t-win32.whl", hash = "sha256:1834bb90991cc2999f10f97f5f01317f99b143284766d197e43cd5b45eb18d03", size = 86943, upload-time = "2025-10-06T14:11:10.284Z" },
+    { url = "https://files.pythonhosted.org/packages/31/74/8b74bae38ed7fe6793d0c15a0c8207bbb819cf287788459e5ed230996cdd/yarl-1.22.0-cp313-cp313t-win_amd64.whl", hash = "sha256:ff86011bd159a9d2dfc89c34cfd8aff12875980e3bd6a39ff097887520e60249", size = 93715, upload-time = "2025-10-06T14:11:11.739Z" },
+    { url = "https://files.pythonhosted.org/packages/69/66/991858aa4b5892d57aef7ee1ba6b4d01ec3b7eb3060795d34090a3ca3278/yarl-1.22.0-cp313-cp313t-win_arm64.whl", hash = "sha256:7861058d0582b847bc4e3a4a4c46828a410bca738673f35a29ba3ca5db0b473b", size = 83857, upload-time = "2025-10-06T14:11:13.586Z" },
+    { url = "https://files.pythonhosted.org/packages/46/b3/e20ef504049f1a1c54a814b4b9bed96d1ac0e0610c3b4da178f87209db05/yarl-1.22.0-cp314-cp314-macosx_10_13_universal2.whl", hash = "sha256:34b36c2c57124530884d89d50ed2c1478697ad7473efd59cfd479945c95650e4", size = 140520, upload-time = "2025-10-06T14:11:15.465Z" },
+    { url = "https://files.pythonhosted.org/packages/e4/04/3532d990fdbab02e5ede063676b5c4260e7f3abea2151099c2aa745acc4c/yarl-1.22.0-cp314-cp314-macosx_10_13_x86_64.whl", hash = "sha256:0dd9a702591ca2e543631c2a017e4a547e38a5c0f29eece37d9097e04a7ac683", size = 93504, upload-time = "2025-10-06T14:11:17.106Z" },
+    { url = "https://files.pythonhosted.org/packages/11/63/ff458113c5c2dac9a9719ac68ee7c947cb621432bcf28c9972b1c0e83938/yarl-1.22.0-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:594fcab1032e2d2cc3321bb2e51271e7cd2b516c7d9aee780ece81b07ff8244b", size = 94282, upload-time = "2025-10-06T14:11:19.064Z" },
+    { url = "https://files.pythonhosted.org/packages/a7/bc/315a56aca762d44a6aaaf7ad253f04d996cb6b27bad34410f82d76ea8038/yarl-1.22.0-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:f3d7a87a78d46a2e3d5b72587ac14b4c16952dd0887dbb051451eceac774411e", size = 372080, upload-time = "2025-10-06T14:11:20.996Z" },
+    { url = "https://files.pythonhosted.org/packages/3f/3f/08e9b826ec2e099ea6e7c69a61272f4f6da62cb5b1b63590bb80ca2e4a40/yarl-1.22.0-cp314-cp314-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:852863707010316c973162e703bddabec35e8757e67fcb8ad58829de1ebc8590", size = 338696, upload-time = "2025-10-06T14:11:22.847Z" },
+    { url = "https://files.pythonhosted.org/packages/e3/9f/90360108e3b32bd76789088e99538febfea24a102380ae73827f62073543/yarl-1.22.0-cp314-cp314-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:131a085a53bfe839a477c0845acf21efc77457ba2bcf5899618136d64f3303a2", size = 387121, upload-time = "2025-10-06T14:11:24.889Z" },
+    { url = "https://files.pythonhosted.org/packages/98/92/ab8d4657bd5b46a38094cfaea498f18bb70ce6b63508fd7e909bd1f93066/yarl-1.22.0-cp314-cp314-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:078a8aefd263f4d4f923a9677b942b445a2be970ca24548a8102689a3a8ab8da", size = 394080, upload-time = "2025-10-06T14:11:27.307Z" },
+    { url = "https://files.pythonhosted.org/packages/f5/e7/d8c5a7752fef68205296201f8ec2bf718f5c805a7a7e9880576c67600658/yarl-1.22.0-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:bca03b91c323036913993ff5c738d0842fc9c60c4648e5c8d98331526df89784", size = 372661, upload-time = "2025-10-06T14:11:29.387Z" },
+    { url = "https://files.pythonhosted.org/packages/b6/2e/f4d26183c8db0bb82d491b072f3127fb8c381a6206a3a56332714b79b751/yarl-1.22.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:68986a61557d37bb90d3051a45b91fa3d5c516d177dfc6dd6f2f436a07ff2b6b", size = 364645, upload-time = "2025-10-06T14:11:31.423Z" },
+    { url = "https://files.pythonhosted.org/packages/80/7c/428e5812e6b87cd00ee8e898328a62c95825bf37c7fa87f0b6bb2ad31304/yarl-1.22.0-cp314-cp314-musllinux_1_2_armv7l.whl", hash = "sha256:4792b262d585ff0dff6bcb787f8492e40698443ec982a3568c2096433660c694", size = 355361, upload-time = "2025-10-06T14:11:33.055Z" },
+    { url = "https://files.pythonhosted.org/packages/ec/2a/249405fd26776f8b13c067378ef4d7dd49c9098d1b6457cdd152a99e96a9/yarl-1.22.0-cp314-cp314-musllinux_1_2_ppc64le.whl", hash = "sha256:ebd4549b108d732dba1d4ace67614b9545b21ece30937a63a65dd34efa19732d", size = 381451, upload-time = "2025-10-06T14:11:35.136Z" },
+    { url = "https://files.pythonhosted.org/packages/67/a8/fb6b1adbe98cf1e2dd9fad71003d3a63a1bc22459c6e15f5714eb9323b93/yarl-1.22.0-cp314-cp314-musllinux_1_2_s390x.whl", hash = "sha256:f87ac53513d22240c7d59203f25cc3beac1e574c6cd681bbfd321987b69f95fd", size = 383814, upload-time = "2025-10-06T14:11:37.094Z" },
+    { url = "https://files.pythonhosted.org/packages/d9/f9/3aa2c0e480fb73e872ae2814c43bc1e734740bb0d54e8cb2a95925f98131/yarl-1.22.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:22b029f2881599e2f1b06f8f1db2ee63bd309e2293ba2d566e008ba12778b8da", size = 370799, upload-time = "2025-10-06T14:11:38.83Z" },
+    { url = "https://files.pythonhosted.org/packages/50/3c/af9dba3b8b5eeb302f36f16f92791f3ea62e3f47763406abf6d5a4a3333b/yarl-1.22.0-cp314-cp314-win32.whl", hash = "sha256:6a635ea45ba4ea8238463b4f7d0e721bad669f80878b7bfd1f89266e2ae63da2", size = 82990, upload-time = "2025-10-06T14:11:40.624Z" },
+    { url = "https://files.pythonhosted.org/packages/ac/30/ac3a0c5bdc1d6efd1b41fa24d4897a4329b3b1e98de9449679dd327af4f0/yarl-1.22.0-cp314-cp314-win_amd64.whl", hash = "sha256:0d6e6885777af0f110b0e5d7e5dda8b704efed3894da26220b7f3d887b839a79", size = 88292, upload-time = "2025-10-06T14:11:42.578Z" },
+    { url = "https://files.pythonhosted.org/packages/df/0a/227ab4ff5b998a1b7410abc7b46c9b7a26b0ca9e86c34ba4b8d8bc7c63d5/yarl-1.22.0-cp314-cp314-win_arm64.whl", hash = "sha256:8218f4e98d3c10d683584cb40f0424f4b9fd6e95610232dd75e13743b070ee33", size = 82888, upload-time = "2025-10-06T14:11:44.863Z" },
+    { url = "https://files.pythonhosted.org/packages/06/5e/a15eb13db90abd87dfbefb9760c0f3f257ac42a5cac7e75dbc23bed97a9f/yarl-1.22.0-cp314-cp314t-macosx_10_13_universal2.whl", hash = "sha256:45c2842ff0e0d1b35a6bf1cd6c690939dacb617a70827f715232b2e0494d55d1", size = 146223, upload-time = "2025-10-06T14:11:46.796Z" },
+    { url = "https://files.pythonhosted.org/packages/18/82/9665c61910d4d84f41a5bf6837597c89e665fa88aa4941080704645932a9/yarl-1.22.0-cp314-cp314t-macosx_10_13_x86_64.whl", hash = "sha256:d947071e6ebcf2e2bee8fce76e10faca8f7a14808ca36a910263acaacef08eca", size = 95981, upload-time = "2025-10-06T14:11:48.845Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/9a/2f65743589809af4d0a6d3aa749343c4b5f4c380cc24a8e94a3c6625a808/yarl-1.22.0-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:334b8721303e61b00019474cc103bdac3d7b1f65e91f0bfedeec2d56dfe74b53", size = 97303, upload-time = "2025-10-06T14:11:50.897Z" },
+    { url = "https://files.pythonhosted.org/packages/b0/ab/5b13d3e157505c43c3b43b5a776cbf7b24a02bc4cccc40314771197e3508/yarl-1.22.0-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:1e7ce67c34138a058fd092f67d07a72b8e31ff0c9236e751957465a24b28910c", size = 361820, upload-time = "2025-10-06T14:11:52.549Z" },
+    { url = "https://files.pythonhosted.org/packages/fb/76/242a5ef4677615cf95330cfc1b4610e78184400699bdda0acb897ef5e49a/yarl-1.22.0-cp314-cp314t-manylinux2014_armv7l.manylinux_2_17_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:d77e1b2c6d04711478cb1c4ab90db07f1609ccf06a287d5607fcd90dc9863acf", size = 323203, upload-time = "2025-10-06T14:11:54.225Z" },
+    { url = "https://files.pythonhosted.org/packages/8c/96/475509110d3f0153b43d06164cf4195c64d16999e0c7e2d8a099adcd6907/yarl-1.22.0-cp314-cp314t-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:c4647674b6150d2cae088fc07de2738a84b8bcedebef29802cf0b0a82ab6face", size = 363173, upload-time = "2025-10-06T14:11:56.069Z" },
+    { url = "https://files.pythonhosted.org/packages/c9/66/59db471aecfbd559a1fd48aedd954435558cd98c7d0da8b03cc6c140a32c/yarl-1.22.0-cp314-cp314t-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:efb07073be061c8f79d03d04139a80ba33cbd390ca8f0297aae9cce6411e4c6b", size = 373562, upload-time = "2025-10-06T14:11:58.783Z" },
+    { url = "https://files.pythonhosted.org/packages/03/1f/c5d94abc91557384719da10ff166b916107c1b45e4d0423a88457071dd88/yarl-1.22.0-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:e51ac5435758ba97ad69617e13233da53908beccc6cfcd6c34bbed8dcbede486", size = 339828, upload-time = "2025-10-06T14:12:00.686Z" },
+    { url = "https://files.pythonhosted.org/packages/5f/97/aa6a143d3afba17b6465733681c70cf175af89f76ec8d9286e08437a7454/yarl-1.22.0-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:33e32a0dd0c8205efa8e83d04fc9f19313772b78522d1bdc7d9aed706bfd6138", size = 347551, upload-time = "2025-10-06T14:12:02.628Z" },
+    { url = "https://files.pythonhosted.org/packages/43/3c/45a2b6d80195959239a7b2a8810506d4eea5487dce61c2a3393e7fc3c52e/yarl-1.22.0-cp314-cp314t-musllinux_1_2_armv7l.whl", hash = "sha256:bf4a21e58b9cde0e401e683ebd00f6ed30a06d14e93f7c8fd059f8b6e8f87b6a", size = 334512, upload-time = "2025-10-06T14:12:04.871Z" },
+    { url = "https://files.pythonhosted.org/packages/86/a0/c2ab48d74599c7c84cb104ebd799c5813de252bea0f360ffc29d270c2caa/yarl-1.22.0-cp314-cp314t-musllinux_1_2_ppc64le.whl", hash = "sha256:e4b582bab49ac33c8deb97e058cd67c2c50dac0dd134874106d9c774fd272529", size = 352400, upload-time = "2025-10-06T14:12:06.624Z" },
+    { url = "https://files.pythonhosted.org/packages/32/75/f8919b2eafc929567d3d8411f72bdb1a2109c01caaab4ebfa5f8ffadc15b/yarl-1.22.0-cp314-cp314t-musllinux_1_2_s390x.whl", hash = "sha256:0b5bcc1a9c4839e7e30b7b30dd47fe5e7e44fb7054ec29b5bb8d526aa1041093", size = 357140, upload-time = "2025-10-06T14:12:08.362Z" },
+    { url = "https://files.pythonhosted.org/packages/cf/72/6a85bba382f22cf78add705d8c3731748397d986e197e53ecc7835e76de7/yarl-1.22.0-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:c0232bce2170103ec23c454e54a57008a9a72b5d1c3105dc2496750da8cfa47c", size = 341473, upload-time = "2025-10-06T14:12:10.994Z" },
+    { url = "https://files.pythonhosted.org/packages/35/18/55e6011f7c044dc80b98893060773cefcfdbf60dfefb8cb2f58b9bacbd83/yarl-1.22.0-cp314-cp314t-win32.whl", hash = "sha256:8009b3173bcd637be650922ac455946197d858b3630b6d8787aa9e5c4564533e", size = 89056, upload-time = "2025-10-06T14:12:13.317Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/86/0f0dccb6e59a9e7f122c5afd43568b1d31b8ab7dda5f1b01fb5c7025c9a9/yarl-1.22.0-cp314-cp314t-win_amd64.whl", hash = "sha256:9fb17ea16e972c63d25d4a97f016d235c78dd2344820eb35bc034bc32012ee27", size = 96292, upload-time = "2025-10-06T14:12:15.398Z" },
+    { url = "https://files.pythonhosted.org/packages/48/b7/503c98092fb3b344a179579f55814b613c1fbb1c23b3ec14a7b008a66a6e/yarl-1.22.0-cp314-cp314t-win_arm64.whl", hash = "sha256:9f6d73c1436b934e3f01df1e1b21ff765cd1d28c77dfb9ace207f746d4610ee1", size = 85171, upload-time = "2025-10-06T14:12:16.935Z" },
+    { url = "https://files.pythonhosted.org/packages/73/ae/b48f95715333080afb75a4504487cbe142cae1268afc482d06692d605ae6/yarl-1.22.0-py3-none-any.whl", hash = "sha256:1380560bdba02b6b6c90de54133c81c9f2a453dee9912fe58c1dcced1edb7cff", size = 46814, upload-time = "2025-10-06T14:12:53.872Z" },
+]
+
+[[package]]
+name = "zstandard"
+version = "0.25.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/fd/aa/3e0508d5a5dd96529cdc5a97011299056e14c6505b678fd58938792794b1/zstandard-0.25.0.tar.gz", hash = "sha256:7713e1179d162cf5c7906da876ec2ccb9c3a9dcbdffef0cc7f70c3667a205f0b", size = 711513, upload-time = "2025-09-14T22:15:54.002Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/56/7a/28efd1d371f1acd037ac64ed1c5e2b41514a6cc937dd6ab6a13ab9f0702f/zstandard-0.25.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:e59fdc271772f6686e01e1b3b74537259800f57e24280be3f29c8a0deb1904dd", size = 795256, upload-time = "2025-09-14T22:15:56.415Z" },
+    { url = "https://files.pythonhosted.org/packages/96/34/ef34ef77f1ee38fc8e4f9775217a613b452916e633c4f1d98f31db52c4a5/zstandard-0.25.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:4d441506e9b372386a5271c64125f72d5df6d2a8e8a2a45a0ae09b03cb781ef7", size = 640565, upload-time = "2025-09-14T22:15:58.177Z" },
+    { url = "https://files.pythonhosted.org/packages/9d/1b/4fdb2c12eb58f31f28c4d28e8dc36611dd7205df8452e63f52fb6261d13e/zstandard-0.25.0-cp310-cp310-manylinux2010_i686.manylinux2014_i686.manylinux_2_12_i686.manylinux_2_17_i686.whl", hash = "sha256:ab85470ab54c2cb96e176f40342d9ed41e58ca5733be6a893b730e7af9c40550", size = 5345306, upload-time = "2025-09-14T22:16:00.165Z" },
+    { url = "https://files.pythonhosted.org/packages/73/28/a44bdece01bca027b079f0e00be3b6bd89a4df180071da59a3dd7381665b/zstandard-0.25.0-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:e05ab82ea7753354bb054b92e2f288afb750e6b439ff6ca78af52939ebbc476d", size = 5055561, upload-time = "2025-09-14T22:16:02.22Z" },
+    { url = "https://files.pythonhosted.org/packages/e9/74/68341185a4f32b274e0fc3410d5ad0750497e1acc20bd0f5b5f64ce17785/zstandard-0.25.0-cp310-cp310-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:78228d8a6a1c177a96b94f7e2e8d012c55f9c760761980da16ae7546a15a8e9b", size = 5402214, upload-time = "2025-09-14T22:16:04.109Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/67/f92e64e748fd6aaffe01e2b75a083c0c4fd27abe1c8747fee4555fcee7dd/zstandard-0.25.0-cp310-cp310-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:2b6bd67528ee8b5c5f10255735abc21aa106931f0dbaf297c7be0c886353c3d0", size = 5449703, upload-time = "2025-09-14T22:16:06.312Z" },
+    { url = "https://files.pythonhosted.org/packages/fd/e5/6d36f92a197c3c17729a2125e29c169f460538a7d939a27eaaa6dcfcba8e/zstandard-0.25.0-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:4b6d83057e713ff235a12e73916b6d356e3084fd3d14ced499d84240f3eecee0", size = 5556583, upload-time = "2025-09-14T22:16:08.457Z" },
+    { url = "https://files.pythonhosted.org/packages/d7/83/41939e60d8d7ebfe2b747be022d0806953799140a702b90ffe214d557638/zstandard-0.25.0-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:9174f4ed06f790a6869b41cba05b43eeb9a35f8993c4422ab853b705e8112bbd", size = 5045332, upload-time = "2025-09-14T22:16:10.444Z" },
+    { url = "https://files.pythonhosted.org/packages/b3/87/d3ee185e3d1aa0133399893697ae91f221fda79deb61adbe998a7235c43f/zstandard-0.25.0-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:25f8f3cd45087d089aef5ba3848cd9efe3ad41163d3400862fb42f81a3a46701", size = 5572283, upload-time = "2025-09-14T22:16:12.128Z" },
+    { url = "https://files.pythonhosted.org/packages/0a/1d/58635ae6104df96671076ac7d4ae7816838ce7debd94aecf83e30b7121b0/zstandard-0.25.0-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:3756b3e9da9b83da1796f8809dd57cb024f838b9eeafde28f3cb472012797ac1", size = 4959754, upload-time = "2025-09-14T22:16:14.225Z" },
+    { url = "https://files.pythonhosted.org/packages/75/d6/57e9cb0a9983e9a229dd8fd2e6e96593ef2aa82a3907188436f22b111ccd/zstandard-0.25.0-cp310-cp310-musllinux_1_2_i686.whl", hash = "sha256:81dad8d145d8fd981b2962b686b2241d3a1ea07733e76a2f15435dfb7fb60150", size = 5266477, upload-time = "2025-09-14T22:16:16.343Z" },
+    { url = "https://files.pythonhosted.org/packages/d1/a9/ee891e5edf33a6ebce0a028726f0bbd8567effe20fe3d5808c42323e8542/zstandard-0.25.0-cp310-cp310-musllinux_1_2_ppc64le.whl", hash = "sha256:a5a419712cf88862a45a23def0ae063686db3d324cec7edbe40509d1a79a0aab", size = 5440914, upload-time = "2025-09-14T22:16:18.453Z" },
+    { url = "https://files.pythonhosted.org/packages/58/08/a8522c28c08031a9521f27abc6f78dbdee7312a7463dd2cfc658b813323b/zstandard-0.25.0-cp310-cp310-musllinux_1_2_s390x.whl", hash = "sha256:e7360eae90809efd19b886e59a09dad07da4ca9ba096752e61a2e03c8aca188e", size = 5819847, upload-time = "2025-09-14T22:16:20.559Z" },
+    { url = "https://files.pythonhosted.org/packages/6f/11/4c91411805c3f7b6f31c60e78ce347ca48f6f16d552fc659af6ec3b73202/zstandard-0.25.0-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:75ffc32a569fb049499e63ce68c743155477610532da1eb38e7f24bf7cd29e74", size = 5363131, upload-time = "2025-09-14T22:16:22.206Z" },
+    { url = "https://files.pythonhosted.org/packages/ef/d6/8c4bd38a3b24c4c7676a7a3d8de85d6ee7a983602a734b9f9cdefb04a5d6/zstandard-0.25.0-cp310-cp310-win32.whl", hash = "sha256:106281ae350e494f4ac8a80470e66d1fe27e497052c8d9c3b95dc4cf1ade81aa", size = 436469, upload-time = "2025-09-14T22:16:25.002Z" },
+    { url = "https://files.pythonhosted.org/packages/93/90/96d50ad417a8ace5f841b3228e93d1bb13e6ad356737f42e2dde30d8bd68/zstandard-0.25.0-cp310-cp310-win_amd64.whl", hash = "sha256:ea9d54cc3d8064260114a0bbf3479fc4a98b21dffc89b3459edd506b69262f6e", size = 506100, upload-time = "2025-09-14T22:16:23.569Z" },
+    { url = "https://files.pythonhosted.org/packages/2a/83/c3ca27c363d104980f1c9cee1101cc8ba724ac8c28a033ede6aab89585b1/zstandard-0.25.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:933b65d7680ea337180733cf9e87293cc5500cc0eb3fc8769f4d3c88d724ec5c", size = 795254, upload-time = "2025-09-14T22:16:26.137Z" },
+    { url = "https://files.pythonhosted.org/packages/ac/4d/e66465c5411a7cf4866aeadc7d108081d8ceba9bc7abe6b14aa21c671ec3/zstandard-0.25.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:a3f79487c687b1fc69f19e487cd949bf3aae653d181dfb5fde3bf6d18894706f", size = 640559, upload-time = "2025-09-14T22:16:27.973Z" },
+    { url = "https://files.pythonhosted.org/packages/12/56/354fe655905f290d3b147b33fe946b0f27e791e4b50a5f004c802cb3eb7b/zstandard-0.25.0-cp311-cp311-manylinux2010_i686.manylinux2014_i686.manylinux_2_12_i686.manylinux_2_17_i686.whl", hash = "sha256:0bbc9a0c65ce0eea3c34a691e3c4b6889f5f3909ba4822ab385fab9057099431", size = 5348020, upload-time = "2025-09-14T22:16:29.523Z" },
+    { url = "https://files.pythonhosted.org/packages/3b/13/2b7ed68bd85e69a2069bcc72141d378f22cae5a0f3b353a2c8f50ef30c1b/zstandard-0.25.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:01582723b3ccd6939ab7b3a78622c573799d5d8737b534b86d0e06ac18dbde4a", size = 5058126, upload-time = "2025-09-14T22:16:31.811Z" },
+    { url = "https://files.pythonhosted.org/packages/c9/dd/fdaf0674f4b10d92cb120ccff58bbb6626bf8368f00ebfd2a41ba4a0dc99/zstandard-0.25.0-cp311-cp311-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:5f1ad7bf88535edcf30038f6919abe087f606f62c00a87d7e33e7fc57cb69fcc", size = 5405390, upload-time = "2025-09-14T22:16:33.486Z" },
+    { url = "https://files.pythonhosted.org/packages/0f/67/354d1555575bc2490435f90d67ca4dd65238ff2f119f30f72d5cde09c2ad/zstandard-0.25.0-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:06acb75eebeedb77b69048031282737717a63e71e4ae3f77cc0c3b9508320df6", size = 5452914, upload-time = "2025-09-14T22:16:35.277Z" },
+    { url = "https://files.pythonhosted.org/packages/bb/1f/e9cfd801a3f9190bf3e759c422bbfd2247db9d7f3d54a56ecde70137791a/zstandard-0.25.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:9300d02ea7c6506f00e627e287e0492a5eb0371ec1670ae852fefffa6164b072", size = 5559635, upload-time = "2025-09-14T22:16:37.141Z" },
+    { url = "https://files.pythonhosted.org/packages/21/88/5ba550f797ca953a52d708c8e4f380959e7e3280af029e38fbf47b55916e/zstandard-0.25.0-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:bfd06b1c5584b657a2892a6014c2f4c20e0db0208c159148fa78c65f7e0b0277", size = 5048277, upload-time = "2025-09-14T22:16:38.807Z" },
+    { url = "https://files.pythonhosted.org/packages/46/c0/ca3e533b4fa03112facbe7fbe7779cb1ebec215688e5df576fe5429172e0/zstandard-0.25.0-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:f373da2c1757bb7f1acaf09369cdc1d51d84131e50d5fa9863982fd626466313", size = 5574377, upload-time = "2025-09-14T22:16:40.523Z" },
+    { url = "https://files.pythonhosted.org/packages/12/9b/3fb626390113f272abd0799fd677ea33d5fc3ec185e62e6be534493c4b60/zstandard-0.25.0-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:6c0e5a65158a7946e7a7affa6418878ef97ab66636f13353b8502d7ea03c8097", size = 4961493, upload-time = "2025-09-14T22:16:43.3Z" },
+    { url = "https://files.pythonhosted.org/packages/cb/d3/23094a6b6a4b1343b27ae68249daa17ae0651fcfec9ed4de09d14b940285/zstandard-0.25.0-cp311-cp311-musllinux_1_2_i686.whl", hash = "sha256:c8e167d5adf59476fa3e37bee730890e389410c354771a62e3c076c86f9f7778", size = 5269018, upload-time = "2025-09-14T22:16:45.292Z" },
+    { url = "https://files.pythonhosted.org/packages/8c/a7/bb5a0c1c0f3f4b5e9d5b55198e39de91e04ba7c205cc46fcb0f95f0383c1/zstandard-0.25.0-cp311-cp311-musllinux_1_2_ppc64le.whl", hash = "sha256:98750a309eb2f020da61e727de7d7ba3c57c97cf6213f6f6277bb7fb42a8e065", size = 5443672, upload-time = "2025-09-14T22:16:47.076Z" },
+    { url = "https://files.pythonhosted.org/packages/27/22/503347aa08d073993f25109c36c8d9f029c7d5949198050962cb568dfa5e/zstandard-0.25.0-cp311-cp311-musllinux_1_2_s390x.whl", hash = "sha256:22a086cff1b6ceca18a8dd6096ec631e430e93a8e70a9ca5efa7561a00f826fa", size = 5822753, upload-time = "2025-09-14T22:16:49.316Z" },
+    { url = "https://files.pythonhosted.org/packages/e2/be/94267dc6ee64f0f8ba2b2ae7c7a2df934a816baaa7291db9e1aa77394c3c/zstandard-0.25.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:72d35d7aa0bba323965da807a462b0966c91608ef3a48ba761678cb20ce5d8b7", size = 5366047, upload-time = "2025-09-14T22:16:51.328Z" },
+    { url = "https://files.pythonhosted.org/packages/7b/a3/732893eab0a3a7aecff8b99052fecf9f605cf0fb5fb6d0290e36beee47a4/zstandard-0.25.0-cp311-cp311-win32.whl", hash = "sha256:f5aeea11ded7320a84dcdd62a3d95b5186834224a9e55b92ccae35d21a8b63d4", size = 436484, upload-time = "2025-09-14T22:16:55.005Z" },
+    { url = "https://files.pythonhosted.org/packages/43/a3/c6155f5c1cce691cb80dfd38627046e50af3ee9ddc5d0b45b9b063bfb8c9/zstandard-0.25.0-cp311-cp311-win_amd64.whl", hash = "sha256:daab68faadb847063d0c56f361a289c4f268706b598afbf9ad113cbe5c38b6b2", size = 506183, upload-time = "2025-09-14T22:16:52.753Z" },
+    { url = "https://files.pythonhosted.org/packages/8c/3e/8945ab86a0820cc0e0cdbf38086a92868a9172020fdab8a03ac19662b0e5/zstandard-0.25.0-cp311-cp311-win_arm64.whl", hash = "sha256:22a06c5df3751bb7dc67406f5374734ccee8ed37fc5981bf1ad7041831fa1137", size = 462533, upload-time = "2025-09-14T22:16:53.878Z" },
+    { url = "https://files.pythonhosted.org/packages/82/fc/f26eb6ef91ae723a03e16eddb198abcfce2bc5a42e224d44cc8b6765e57e/zstandard-0.25.0-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:7b3c3a3ab9daa3eed242d6ecceead93aebbb8f5f84318d82cee643e019c4b73b", size = 795738, upload-time = "2025-09-14T22:16:56.237Z" },
+    { url = "https://files.pythonhosted.org/packages/aa/1c/d920d64b22f8dd028a8b90e2d756e431a5d86194caa78e3819c7bf53b4b3/zstandard-0.25.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:913cbd31a400febff93b564a23e17c3ed2d56c064006f54efec210d586171c00", size = 640436, upload-time = "2025-09-14T22:16:57.774Z" },
+    { url = "https://files.pythonhosted.org/packages/53/6c/288c3f0bd9fcfe9ca41e2c2fbfd17b2097f6af57b62a81161941f09afa76/zstandard-0.25.0-cp312-cp312-manylinux2010_i686.manylinux2014_i686.manylinux_2_12_i686.manylinux_2_17_i686.whl", hash = "sha256:011d388c76b11a0c165374ce660ce2c8efa8e5d87f34996aa80f9c0816698b64", size = 5343019, upload-time = "2025-09-14T22:16:59.302Z" },
+    { url = "https://files.pythonhosted.org/packages/1e/15/efef5a2f204a64bdb5571e6161d49f7ef0fffdbca953a615efbec045f60f/zstandard-0.25.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:6dffecc361d079bb48d7caef5d673c88c8988d3d33fb74ab95b7ee6da42652ea", size = 5063012, upload-time = "2025-09-14T22:17:01.156Z" },
+    { url = "https://files.pythonhosted.org/packages/b7/37/a6ce629ffdb43959e92e87ebdaeebb5ac81c944b6a75c9c47e300f85abdf/zstandard-0.25.0-cp312-cp312-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:7149623bba7fdf7e7f24312953bcf73cae103db8cae49f8154dd1eadc8a29ecb", size = 5394148, upload-time = "2025-09-14T22:17:03.091Z" },
+    { url = "https://files.pythonhosted.org/packages/e3/79/2bf870b3abeb5c070fe2d670a5a8d1057a8270f125ef7676d29ea900f496/zstandard-0.25.0-cp312-cp312-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:6a573a35693e03cf1d67799fd01b50ff578515a8aeadd4595d2a7fa9f3ec002a", size = 5451652, upload-time = "2025-09-14T22:17:04.979Z" },
+    { url = "https://files.pythonhosted.org/packages/53/60/7be26e610767316c028a2cbedb9a3beabdbe33e2182c373f71a1c0b88f36/zstandard-0.25.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:5a56ba0db2d244117ed744dfa8f6f5b366e14148e00de44723413b2f3938a902", size = 5546993, upload-time = "2025-09-14T22:17:06.781Z" },
+    { url = "https://files.pythonhosted.org/packages/85/c7/3483ad9ff0662623f3648479b0380d2de5510abf00990468c286c6b04017/zstandard-0.25.0-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:10ef2a79ab8e2974e2075fb984e5b9806c64134810fac21576f0668e7ea19f8f", size = 5046806, upload-time = "2025-09-14T22:17:08.415Z" },
+    { url = "https://files.pythonhosted.org/packages/08/b3/206883dd25b8d1591a1caa44b54c2aad84badccf2f1de9e2d60a446f9a25/zstandard-0.25.0-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:aaf21ba8fb76d102b696781bddaa0954b782536446083ae3fdaa6f16b25a1c4b", size = 5576659, upload-time = "2025-09-14T22:17:10.164Z" },
+    { url = "https://files.pythonhosted.org/packages/9d/31/76c0779101453e6c117b0ff22565865c54f48f8bd807df2b00c2c404b8e0/zstandard-0.25.0-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:1869da9571d5e94a85a5e8d57e4e8807b175c9e4a6294e3b66fa4efb074d90f6", size = 4953933, upload-time = "2025-09-14T22:17:11.857Z" },
+    { url = "https://files.pythonhosted.org/packages/18/e1/97680c664a1bf9a247a280a053d98e251424af51f1b196c6d52f117c9720/zstandard-0.25.0-cp312-cp312-musllinux_1_2_i686.whl", hash = "sha256:809c5bcb2c67cd0ed81e9229d227d4ca28f82d0f778fc5fea624a9def3963f91", size = 5268008, upload-time = "2025-09-14T22:17:13.627Z" },
+    { url = "https://files.pythonhosted.org/packages/1e/73/316e4010de585ac798e154e88fd81bb16afc5c5cb1a72eeb16dd37e8024a/zstandard-0.25.0-cp312-cp312-musllinux_1_2_ppc64le.whl", hash = "sha256:f27662e4f7dbf9f9c12391cb37b4c4c3cb90ffbd3b1fb9284dadbbb8935fa708", size = 5433517, upload-time = "2025-09-14T22:17:16.103Z" },
+    { url = "https://files.pythonhosted.org/packages/5b/60/dd0f8cfa8129c5a0ce3ea6b7f70be5b33d2618013a161e1ff26c2b39787c/zstandard-0.25.0-cp312-cp312-musllinux_1_2_s390x.whl", hash = "sha256:99c0c846e6e61718715a3c9437ccc625de26593fea60189567f0118dc9db7512", size = 5814292, upload-time = "2025-09-14T22:17:17.827Z" },
+    { url = "https://files.pythonhosted.org/packages/fc/5f/75aafd4b9d11b5407b641b8e41a57864097663699f23e9ad4dbb91dc6bfe/zstandard-0.25.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:474d2596a2dbc241a556e965fb76002c1ce655445e4e3bf38e5477d413165ffa", size = 5360237, upload-time = "2025-09-14T22:17:19.954Z" },
+    { url = "https://files.pythonhosted.org/packages/ff/8d/0309daffea4fcac7981021dbf21cdb2e3427a9e76bafbcdbdf5392ff99a4/zstandard-0.25.0-cp312-cp312-win32.whl", hash = "sha256:23ebc8f17a03133b4426bcc04aabd68f8236eb78c3760f12783385171b0fd8bd", size = 436922, upload-time = "2025-09-14T22:17:24.398Z" },
+    { url = "https://files.pythonhosted.org/packages/79/3b/fa54d9015f945330510cb5d0b0501e8253c127cca7ebe8ba46a965df18c5/zstandard-0.25.0-cp312-cp312-win_amd64.whl", hash = "sha256:ffef5a74088f1e09947aecf91011136665152e0b4b359c42be3373897fb39b01", size = 506276, upload-time = "2025-09-14T22:17:21.429Z" },
+    { url = "https://files.pythonhosted.org/packages/ea/6b/8b51697e5319b1f9ac71087b0af9a40d8a6288ff8025c36486e0c12abcc4/zstandard-0.25.0-cp312-cp312-win_arm64.whl", hash = "sha256:181eb40e0b6a29b3cd2849f825e0fa34397f649170673d385f3598ae17cca2e9", size = 462679, upload-time = "2025-09-14T22:17:23.147Z" },
+    { url = "https://files.pythonhosted.org/packages/35/0b/8df9c4ad06af91d39e94fa96cc010a24ac4ef1378d3efab9223cc8593d40/zstandard-0.25.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:ec996f12524f88e151c339688c3897194821d7f03081ab35d31d1e12ec975e94", size = 795735, upload-time = "2025-09-14T22:17:26.042Z" },
+    { url = "https://files.pythonhosted.org/packages/3f/06/9ae96a3e5dcfd119377ba33d4c42a7d89da1efabd5cb3e366b156c45ff4d/zstandard-0.25.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:a1a4ae2dec3993a32247995bdfe367fc3266da832d82f8438c8570f989753de1", size = 640440, upload-time = "2025-09-14T22:17:27.366Z" },
+    { url = "https://files.pythonhosted.org/packages/d9/14/933d27204c2bd404229c69f445862454dcc101cd69ef8c6068f15aaec12c/zstandard-0.25.0-cp313-cp313-manylinux2010_i686.manylinux2014_i686.manylinux_2_12_i686.manylinux_2_17_i686.whl", hash = "sha256:e96594a5537722fdfb79951672a2a63aec5ebfb823e7560586f7484819f2a08f", size = 5343070, upload-time = "2025-09-14T22:17:28.896Z" },
+    { url = "https://files.pythonhosted.org/packages/6d/db/ddb11011826ed7db9d0e485d13df79b58586bfdec56e5c84a928a9a78c1c/zstandard-0.25.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:bfc4e20784722098822e3eee42b8e576b379ed72cca4a7cb856ae733e62192ea", size = 5063001, upload-time = "2025-09-14T22:17:31.044Z" },
+    { url = "https://files.pythonhosted.org/packages/db/00/87466ea3f99599d02a5238498b87bf84a6348290c19571051839ca943777/zstandard-0.25.0-cp313-cp313-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:457ed498fc58cdc12fc48f7950e02740d4f7ae9493dd4ab2168a47c93c31298e", size = 5394120, upload-time = "2025-09-14T22:17:32.711Z" },
+    { url = "https://files.pythonhosted.org/packages/2b/95/fc5531d9c618a679a20ff6c29e2b3ef1d1f4ad66c5e161ae6ff847d102a9/zstandard-0.25.0-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:fd7a5004eb1980d3cefe26b2685bcb0b17989901a70a1040d1ac86f1d898c551", size = 5451230, upload-time = "2025-09-14T22:17:34.41Z" },
+    { url = "https://files.pythonhosted.org/packages/63/4b/e3678b4e776db00f9f7b2fe58e547e8928ef32727d7a1ff01dea010f3f13/zstandard-0.25.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:8e735494da3db08694d26480f1493ad2cf86e99bdd53e8e9771b2752a5c0246a", size = 5547173, upload-time = "2025-09-14T22:17:36.084Z" },
+    { url = "https://files.pythonhosted.org/packages/4e/d5/ba05ed95c6b8ec30bd468dfeab20589f2cf709b5c940483e31d991f2ca58/zstandard-0.25.0-cp313-cp313-musllinux_1_1_aarch64.whl", hash = "sha256:3a39c94ad7866160a4a46d772e43311a743c316942037671beb264e395bdd611", size = 5046736, upload-time = "2025-09-14T22:17:37.891Z" },
+    { url = "https://files.pythonhosted.org/packages/50/d5/870aa06b3a76c73eced65c044b92286a3c4e00554005ff51962deef28e28/zstandard-0.25.0-cp313-cp313-musllinux_1_1_x86_64.whl", hash = "sha256:172de1f06947577d3a3005416977cce6168f2261284c02080e7ad0185faeced3", size = 5576368, upload-time = "2025-09-14T22:17:40.206Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/35/398dc2ffc89d304d59bc12f0fdd931b4ce455bddf7038a0a67733a25f550/zstandard-0.25.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:3c83b0188c852a47cd13ef3bf9209fb0a77fa5374958b8c53aaa699398c6bd7b", size = 4954022, upload-time = "2025-09-14T22:17:41.879Z" },
+    { url = "https://files.pythonhosted.org/packages/9a/5c/36ba1e5507d56d2213202ec2b05e8541734af5f2ce378c5d1ceaf4d88dc4/zstandard-0.25.0-cp313-cp313-musllinux_1_2_i686.whl", hash = "sha256:1673b7199bbe763365b81a4f3252b8e80f44c9e323fc42940dc8843bfeaf9851", size = 5267889, upload-time = "2025-09-14T22:17:43.577Z" },
+    { url = "https://files.pythonhosted.org/packages/70/e8/2ec6b6fb7358b2ec0113ae202647ca7c0e9d15b61c005ae5225ad0995df5/zstandard-0.25.0-cp313-cp313-musllinux_1_2_ppc64le.whl", hash = "sha256:0be7622c37c183406f3dbf0cba104118eb16a4ea7359eeb5752f0794882fc250", size = 5433952, upload-time = "2025-09-14T22:17:45.271Z" },
+    { url = "https://files.pythonhosted.org/packages/7b/01/b5f4d4dbc59ef193e870495c6f1275f5b2928e01ff5a81fecb22a06e22fb/zstandard-0.25.0-cp313-cp313-musllinux_1_2_s390x.whl", hash = "sha256:5f5e4c2a23ca271c218ac025bd7d635597048b366d6f31f420aaeb715239fc98", size = 5814054, upload-time = "2025-09-14T22:17:47.08Z" },
+    { url = "https://files.pythonhosted.org/packages/b2/e5/fbd822d5c6f427cf158316d012c5a12f233473c2f9c5fe5ab1ae5d21f3d8/zstandard-0.25.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:4f187a0bb61b35119d1926aee039524d1f93aaf38a9916b8c4b78ac8514a0aaf", size = 5360113, upload-time = "2025-09-14T22:17:48.893Z" },
+    { url = "https://files.pythonhosted.org/packages/8e/e0/69a553d2047f9a2c7347caa225bb3a63b6d7704ad74610cb7823baa08ed7/zstandard-0.25.0-cp313-cp313-win32.whl", hash = "sha256:7030defa83eef3e51ff26f0b7bfb229f0204b66fe18e04359ce3474ac33cbc09", size = 436936, upload-time = "2025-09-14T22:17:52.658Z" },
+    { url = "https://files.pythonhosted.org/packages/d9/82/b9c06c870f3bd8767c201f1edbdf9e8dc34be5b0fbc5682c4f80fe948475/zstandard-0.25.0-cp313-cp313-win_amd64.whl", hash = "sha256:1f830a0dac88719af0ae43b8b2d6aef487d437036468ef3c2ea59c51f9d55fd5", size = 506232, upload-time = "2025-09-14T22:17:50.402Z" },
+    { url = "https://files.pythonhosted.org/packages/d4/57/60c3c01243bb81d381c9916e2a6d9e149ab8627c0c7d7abb2d73384b3c0c/zstandard-0.25.0-cp313-cp313-win_arm64.whl", hash = "sha256:85304a43f4d513f5464ceb938aa02c1e78c2943b29f44a750b48b25ac999a049", size = 462671, upload-time = "2025-09-14T22:17:51.533Z" },
+    { url = "https://files.pythonhosted.org/packages/3d/5c/f8923b595b55fe49e30612987ad8bf053aef555c14f05bb659dd5dbe3e8a/zstandard-0.25.0-cp314-cp314-macosx_10_13_x86_64.whl", hash = "sha256:e29f0cf06974c899b2c188ef7f783607dbef36da4c242eb6c82dcd8b512855e3", size = 795887, upload-time = "2025-09-14T22:17:54.198Z" },
+    { url = "https://files.pythonhosted.org/packages/8d/09/d0a2a14fc3439c5f874042dca72a79c70a532090b7ba0003be73fee37ae2/zstandard-0.25.0-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:05df5136bc5a011f33cd25bc9f506e7426c0c9b3f9954f056831ce68f3b6689f", size = 640658, upload-time = "2025-09-14T22:17:55.423Z" },
+    { url = "https://files.pythonhosted.org/packages/5d/7c/8b6b71b1ddd517f68ffb55e10834388d4f793c49c6b83effaaa05785b0b4/zstandard-0.25.0-cp314-cp314-manylinux2010_i686.manylinux_2_12_i686.manylinux_2_28_i686.whl", hash = "sha256:f604efd28f239cc21b3adb53eb061e2a205dc164be408e553b41ba2ffe0ca15c", size = 5379849, upload-time = "2025-09-14T22:17:57.372Z" },
+    { url = "https://files.pythonhosted.org/packages/a4/86/a48e56320d0a17189ab7a42645387334fba2200e904ee47fc5a26c1fd8ca/zstandard-0.25.0-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:223415140608d0f0da010499eaa8ccdb9af210a543fac54bce15babbcfc78439", size = 5058095, upload-time = "2025-09-14T22:17:59.498Z" },
+    { url = "https://files.pythonhosted.org/packages/f8/ad/eb659984ee2c0a779f9d06dbfe45e2dc39d99ff40a319895df2d3d9a48e5/zstandard-0.25.0-cp314-cp314-manylinux2014_ppc64le.manylinux_2_17_ppc64le.manylinux_2_28_ppc64le.whl", hash = "sha256:2e54296a283f3ab5a26fc9b8b5d4978ea0532f37b231644f367aa588930aa043", size = 5551751, upload-time = "2025-09-14T22:18:01.618Z" },
+    { url = "https://files.pythonhosted.org/packages/61/b3/b637faea43677eb7bd42ab204dfb7053bd5c4582bfe6b1baefa80ac0c47b/zstandard-0.25.0-cp314-cp314-manylinux2014_s390x.manylinux_2_17_s390x.manylinux_2_28_s390x.whl", hash = "sha256:ca54090275939dc8ec5dea2d2afb400e0f83444b2fc24e07df7fdef677110859", size = 6364818, upload-time = "2025-09-14T22:18:03.769Z" },
+    { url = "https://files.pythonhosted.org/packages/31/dc/cc50210e11e465c975462439a492516a73300ab8caa8f5e0902544fd748b/zstandard-0.25.0-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:e09bb6252b6476d8d56100e8147b803befa9a12cea144bbe629dd508800d1ad0", size = 5560402, upload-time = "2025-09-14T22:18:05.954Z" },
+    { url = "https://files.pythonhosted.org/packages/c9/ae/56523ae9c142f0c08efd5e868a6da613ae76614eca1305259c3bf6a0ed43/zstandard-0.25.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:a9ec8c642d1ec73287ae3e726792dd86c96f5681eb8df274a757bf62b750eae7", size = 4955108, upload-time = "2025-09-14T22:18:07.68Z" },
+    { url = "https://files.pythonhosted.org/packages/98/cf/c899f2d6df0840d5e384cf4c4121458c72802e8bda19691f3b16619f51e9/zstandard-0.25.0-cp314-cp314-musllinux_1_2_i686.whl", hash = "sha256:a4089a10e598eae6393756b036e0f419e8c1d60f44a831520f9af41c14216cf2", size = 5269248, upload-time = "2025-09-14T22:18:09.753Z" },
+    { url = "https://files.pythonhosted.org/packages/1b/c0/59e912a531d91e1c192d3085fc0f6fb2852753c301a812d856d857ea03c6/zstandard-0.25.0-cp314-cp314-musllinux_1_2_ppc64le.whl", hash = "sha256:f67e8f1a324a900e75b5e28ffb152bcac9fbed1cc7b43f99cd90f395c4375344", size = 5430330, upload-time = "2025-09-14T22:18:11.966Z" },
+    { url = "https://files.pythonhosted.org/packages/a0/1d/7e31db1240de2df22a58e2ea9a93fc6e38cc29353e660c0272b6735d6669/zstandard-0.25.0-cp314-cp314-musllinux_1_2_s390x.whl", hash = "sha256:9654dbc012d8b06fc3d19cc825af3f7bf8ae242226df5f83936cb39f5fdc846c", size = 5811123, upload-time = "2025-09-14T22:18:13.907Z" },
+    { url = "https://files.pythonhosted.org/packages/f6/49/fac46df5ad353d50535e118d6983069df68ca5908d4d65b8c466150a4ff1/zstandard-0.25.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:4203ce3b31aec23012d3a4cf4a2ed64d12fea5269c49aed5e4c3611b938e4088", size = 5359591, upload-time = "2025-09-14T22:18:16.465Z" },
+    { url = "https://files.pythonhosted.org/packages/c2/38/f249a2050ad1eea0bb364046153942e34abba95dd5520af199aed86fbb49/zstandard-0.25.0-cp314-cp314-win32.whl", hash = "sha256:da469dc041701583e34de852d8634703550348d5822e66a0c827d39b05365b12", size = 444513, upload-time = "2025-09-14T22:18:20.61Z" },
+    { url = "https://files.pythonhosted.org/packages/3a/43/241f9615bcf8ba8903b3f0432da069e857fc4fd1783bd26183db53c4804b/zstandard-0.25.0-cp314-cp314-win_amd64.whl", hash = "sha256:c19bcdd826e95671065f8692b5a4aa95c52dc7a02a4c5a0cac46deb879a017a2", size = 516118, upload-time = "2025-09-14T22:18:17.849Z" },
+    { url = "https://files.pythonhosted.org/packages/f0/ef/da163ce2450ed4febf6467d77ccb4cd52c4c30ab45624bad26ca0a27260c/zstandard-0.25.0-cp314-cp314-win_arm64.whl", hash = "sha256:d7541afd73985c630bafcd6338d2518ae96060075f9463d7dc14cfb33514383d", size = 476940, upload-time = "2025-09-14T22:18:19.088Z" },
+]
diff --git a/mcp_server/README.md b/mcp_server/README.md
new file mode 100644
index 0000000..9a7e4b8
--- /dev/null
+++ b/mcp_server/README.md
@@ -0,0 +1,5 @@
+# PaddleOCR MCP 服务器
+
+中文 | [English](./README_en.md)
+
+请查看 [文档](../docs/version3.x/integrations/mcp_server.md)。
diff --git a/mcp_server/README_en.md b/mcp_server/README_en.md
new file mode 100644
index 0000000..d8c32e2
--- /dev/null
+++ b/mcp_server/README_en.md
@@ -0,0 +1,5 @@
+# PaddleOCR MCP Server
+
+[中文](./README.md)| English 
+
+Please refer to the [documentation](../docs/version3.x/integrations/mcp_server.en.md)。
diff --git a/mcp_server/paddleocr_mcp/__init__.py b/mcp_server/paddleocr_mcp/__init__.py
new file mode 100644
index 0000000..6247046
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/__init__.py
@@ -0,0 +1,15 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+__version__ = "0.1.0"
diff --git a/mcp_server/paddleocr_mcp/__main__.py b/mcp_server/paddleocr_mcp/__main__.py
new file mode 100644
index 0000000..4269dd0
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/__main__.py
@@ -0,0 +1,278 @@
+#!/usr/bin/env python3
+
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import annotations
+
+import argparse
+import asyncio
+import os
+import sys
+
+from fastmcp import FastMCP
+
+from .inference import create_inference
+from .selection import DEFAULT_MODEL, resolve_model
+from .providers import InferenceProvider, provider_choices
+from .tasks import create_task
+
+
+def _parse_args() -> argparse.Namespace:
+    """Parse command line arguments."""
+    parser = argparse.ArgumentParser(description="PaddleOCR MCP server.")
+
+    parser.add_argument(
+        "--model",
+        default=os.getenv("PADDLEOCR_MCP_MODEL", DEFAULT_MODEL),
+        help="Model to run. Env: PADDLEOCR_MCP_MODEL.",
+    )
+    parser.add_argument(
+        "--ppocr_source",
+        choices=provider_choices(),
+        default=os.getenv("PADDLEOCR_MCP_PPOCR_SOURCE", "local"),
+        help="Inference provider. Env: PADDLEOCR_MCP_PPOCR_SOURCE.",
+    )
+
+    parser.add_argument(
+        "--http",
+        action="store_true",
+        help="Use Streamable HTTP instead of stdio.",
+    )
+    parser.add_argument(
+        "--host",
+        default="127.0.0.1",
+        help="HTTP bind host (with --http). Default: 127.0.0.1.",
+    )
+    parser.add_argument(
+        "--port",
+        type=int,
+        default=8000,
+        help="HTTP bind port (with --http). Default: 8000.",
+    )
+    parser.add_argument(
+        "--verbose",
+        action="store_true",
+        help="Enable verbose logging.",
+    )
+
+    parser.add_argument(
+        "--pipeline_config",
+        default=os.getenv("PADDLEOCR_MCP_PIPELINE_CONFIG"),
+        help="Pipeline config file path (local). Env: PADDLEOCR_MCP_PIPELINE_CONFIG.",
+    )
+    parser.add_argument(
+        "--device",
+        default=os.getenv("PADDLEOCR_MCP_DEVICE"),
+        help="Inference device (local). Env: PADDLEOCR_MCP_DEVICE.",
+    )
+
+    parser.add_argument(
+        "--aistudio-base-url",
+        dest="aistudio_base_url",
+        default=os.getenv("PADDLEOCR_MCP_AISTUDIO_BASE_URL"),
+        help="AI Studio API base URL (aistudio). Env: PADDLEOCR_MCP_AISTUDIO_BASE_URL.",
+    )
+    parser.add_argument(
+        "--qianfan-base-url",
+        dest="qianfan_base_url",
+        default=os.getenv("PADDLEOCR_MCP_QIANFAN_BASE_URL")
+        or "https://qianfan.baidubce.com/v2/ocr",
+        help="Qianfan API base URL (qianfan). Env: PADDLEOCR_MCP_QIANFAN_BASE_URL.",
+    )
+    parser.add_argument(
+        "--self-hosted-base-url",
+        dest="self_hosted_base_url",
+        default=os.getenv("PADDLEOCR_MCP_SELF_HOSTED_BASE_URL"),
+        help="Self-hosted service base URL (self_hosted). Env: PADDLEOCR_MCP_SELF_HOSTED_BASE_URL.",
+    )
+    parser.add_argument(
+        "--aistudio_access_token",
+        default=os.getenv("PADDLEOCR_MCP_AISTUDIO_ACCESS_TOKEN"),
+        help="AI Studio access token (aistudio). Env: PADDLEOCR_MCP_AISTUDIO_ACCESS_TOKEN.",
+    )
+    parser.add_argument(
+        "--qianfan_api_key",
+        default=os.getenv("PADDLEOCR_MCP_QIANFAN_API_KEY"),
+        help="Qianfan API key (qianfan). Env: PADDLEOCR_MCP_QIANFAN_API_KEY.",
+    )
+    parser.add_argument(
+        "--http-timeout",
+        dest="http_timeout",
+        type=int,
+        default=int(os.getenv("PADDLEOCR_MCP_HTTP_TIMEOUT", "600")),
+        help=(
+            "HTTP read timeout in seconds for synchronous APIs. "
+            "Env: PADDLEOCR_MCP_HTTP_TIMEOUT."
+        ),
+    )
+    parser.add_argument(
+        "--aistudio-request-timeout",
+        dest="aistudio_request_timeout",
+        type=int,
+        default=int(os.getenv("PADDLEOCR_MCP_AISTUDIO_REQUEST_TIMEOUT", "120")),
+        help=(
+            "Per-request HTTP timeout in seconds for AI Studio API calls. "
+            "Env: PADDLEOCR_MCP_AISTUDIO_REQUEST_TIMEOUT."
+        ),
+    )
+    parser.add_argument(
+        "--aistudio-poll-timeout",
+        dest="aistudio_poll_timeout",
+        type=int,
+        default=int(os.getenv("PADDLEOCR_MCP_AISTUDIO_POLL_TIMEOUT", "600")),
+        help=(
+            "Total job polling timeout in seconds for AI Studio. "
+            "Env: PADDLEOCR_MCP_AISTUDIO_POLL_TIMEOUT."
+        ),
+    )
+
+    return parser.parse_args()
+
+
+def _validate_args(args: argparse.Namespace) -> None:
+    """Validate command line arguments."""
+    if not args.http and (args.host != "127.0.0.1" or args.port != 8000):
+        print(
+            "Host and port arguments are only valid when using HTTP transport (see: `--http`).",
+            file=sys.stderr,
+        )
+        sys.exit(2)
+
+    if args.ppocr_source == InferenceProvider.AISTUDIO.value:
+        if not args.aistudio_access_token:
+            print("Error: The AI Studio access token is required.", file=sys.stderr)
+            print(
+                "Please either set `--aistudio_access_token` or set the environment variable "
+                "`PADDLEOCR_MCP_AISTUDIO_ACCESS_TOKEN`.",
+                file=sys.stderr,
+            )
+            sys.exit(2)
+    elif args.ppocr_source == InferenceProvider.QIANFAN.value:
+        if not args.qianfan_api_key:
+            print("Error: The Qianfan API key is required.", file=sys.stderr)
+            print(
+                "Please either set `--qianfan_api_key` or set the environment variable "
+                "`PADDLEOCR_MCP_QIANFAN_API_KEY`.",
+                file=sys.stderr,
+            )
+            sys.exit(2)
+    elif args.ppocr_source == InferenceProvider.SELF_HOSTED.value:
+        if not args.self_hosted_base_url:
+            print(
+                "Error: The self-hosted service base URL is required.", file=sys.stderr
+            )
+            print(
+                f"Please set `--self-hosted-base-url` or the environment variable "
+                "`PADDLEOCR_MCP_SELF_HOSTED_BASE_URL`.",
+                file=sys.stderr,
+            )
+            sys.exit(2)
+
+
+def _create_inference_from_args(args: argparse.Namespace, model: str):
+    provider = args.ppocr_source
+
+    if provider == InferenceProvider.LOCAL.value:
+        return create_inference(
+            model=model,
+            provider=provider,
+            config=args.pipeline_config,
+            device=args.device,
+        )
+    elif provider == InferenceProvider.AISTUDIO.value:
+        return create_inference(
+            model=model,
+            provider=provider,
+            token=args.aistudio_access_token,
+            base_url=args.aistudio_base_url,
+            request_timeout=float(args.aistudio_request_timeout),
+            poll_timeout=float(args.aistudio_poll_timeout),
+        )
+    elif provider == InferenceProvider.QIANFAN.value:
+        return create_inference(
+            model=model,
+            provider=provider,
+            base_url=args.qianfan_base_url,
+            api_key=args.qianfan_api_key,
+            http_timeout=args.http_timeout,
+        )
+    elif provider == InferenceProvider.SELF_HOSTED.value:
+        return create_inference(
+            model=model,
+            provider=provider,
+            base_url=args.self_hosted_base_url,
+            http_timeout=args.http_timeout,
+        )
+    else:
+        raise ValueError(f"Unknown provider: {provider}")
+
+
+async def async_main() -> None:
+    """Asynchronous main entry point."""
+    args = _parse_args()
+    _validate_args(args)
+
+    try:
+        model = resolve_model(args.model, args.ppocr_source)
+    except ValueError as e:
+        print(f"Error: {e}", file=sys.stderr)
+        sys.exit(2)
+
+    inference = _create_inference_from_args(args, model)
+
+    try:
+        await inference.start()
+
+        task = create_task(model, inference)
+
+        server_name = f"PaddleOCR {model} MCP server"
+        mcp = FastMCP(
+            name=server_name,
+            mask_error_details=True,
+        )
+
+        task.register_tools(mcp)
+
+        log_level = "INFO" if args.verbose else "WARNING"
+
+        if args.http:
+            await mcp.run_async(
+                transport="streamable-http",
+                host=args.host,
+                port=args.port,
+                log_level=log_level,
+            )
+        else:
+            await mcp.run_async(log_level=log_level)
+
+    except Exception as e:
+        print(f"Failed to start the server: {e}", file=sys.stderr)
+        if args.verbose:
+            import traceback
+
+            traceback.print_exc(file=sys.stderr)
+        sys.exit(1)
+
+    finally:
+        await inference.stop()
+
+
+def main() -> None:
+    """Main entry point."""
+    asyncio.run(async_main())
+
+
+if __name__ == "__main__":
+    main()
diff --git a/mcp_server/paddleocr_mcp/inference/__init__.py b/mcp_server/paddleocr_mcp/inference/__init__.py
new file mode 100644
index 0000000..e3cd4ce
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/__init__.py
@@ -0,0 +1,22 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .base import Inference
+from .factory import InferenceFactory, create_inference
+
+__all__ = [
+    "Inference",
+    "InferenceFactory",
+    "create_inference",
+]
diff --git a/mcp_server/paddleocr_mcp/inference/base.py b/mcp_server/paddleocr_mcp/inference/base.py
new file mode 100644
index 0000000..7d8a315
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/base.py
@@ -0,0 +1,93 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import abc
+from typing import Any, Dict
+
+from .shared.input_adapters import InputAdapter
+from .types import InferenceRequest, InferenceResult
+
+
+class Inference(abc.ABC):
+    @property
+    @abc.abstractmethod
+    def input_adapter(self) -> InputAdapter:
+        """Input adapter that validates and prepares user input for this source."""
+        pass
+
+    @abc.abstractmethod
+    async def start(self) -> None:
+        """Initialize inference resources."""
+        pass
+
+    @abc.abstractmethod
+    async def stop(self) -> None:
+        """Clean up inference resources."""
+        pass
+
+    @abc.abstractmethod
+    async def predict(
+        self,
+        request: InferenceRequest,
+    ) -> InferenceResult:
+        """Run inference for the given request."""
+        pass
+
+    @abc.abstractmethod
+    def get_valid_params(self) -> set[str]:
+        """Get valid runtime parameter names for this inference.
+
+        Returns:
+            Set of valid parameter names.
+        """
+        pass
+
+    @abc.abstractmethod
+    def get_default_params(self) -> Dict[str, Any]:
+        """Get default runtime parameters for this inference.
+
+        Returns:
+            Dictionary of default parameters.
+        """
+        pass
+
+    def validate_params(self, params: Dict[str, Any]) -> None:
+        """Validate runtime parameters against allowed set.
+
+        Args:
+            params: Runtime parameters dictionary.
+
+        Raises:
+            ValueError: If any parameter is not in the valid set.
+        """
+        valid_params = self.get_valid_params()
+        invalid_params = set(params.keys()) - valid_params
+        if invalid_params:
+            raise ValueError(
+                f"Invalid runtime parameters: {invalid_params}. "
+                f"Valid parameters are: {sorted(valid_params)}."
+            )
+
+    def get_final_params(self, runtime_params: Dict[str, Any]) -> Dict[str, Any]:
+        """Get final parameters with defaults merged.
+
+        Args:
+            runtime_params: User-provided runtime parameters.
+
+        Returns:
+            Dictionary with defaults overridden by user params.
+        """
+        final_params = self.get_default_params()
+        final_params.update(runtime_params)
+        return final_params
diff --git a/mcp_server/paddleocr_mcp/inference/errors.py b/mcp_server/paddleocr_mcp/inference/errors.py
new file mode 100644
index 0000000..1997ac9
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/errors.py
@@ -0,0 +1,31 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Inference-related exceptions."""
+
+
+class InferenceError(RuntimeError):
+    """Base class for inference errors."""
+
+
+class AuthenticationError(InferenceError):
+    """Authentication failed."""
+
+
+class ResourceUnavailableError(InferenceError):
+    """Service unavailable."""
+
+
+class ExecutionTimeoutError(InferenceError):
+    """Request timeout."""
diff --git a/mcp_server/paddleocr_mcp/inference/factory.py b/mcp_server/paddleocr_mcp/inference/factory.py
new file mode 100644
index 0000000..3e39020
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/factory.py
@@ -0,0 +1,200 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Any, Callable
+
+from .base import Inference
+from .ocr.aistudio import OCRAIStudioInference
+from .ocr.local import OCRLocalInference
+from .ocr.self_hosted import OCRSelfHostedInference
+from .paddleocr_vl.aistudio import PaddleOCRVLAIStudioInference
+from .paddleocr_vl.local import PaddleOCRVLLocalInference
+from .paddleocr_vl.qianfan import PaddleOCRVLQianfanInference
+from .paddleocr_vl.self_hosted import PaddleOCRVLSelfHostedInference
+from .pp_structurev3.aistudio import PPStructureV3AIStudioInference
+from .pp_structurev3.local import PPStructureV3LocalInference
+from .pp_structurev3.qianfan import PPStructureV3QianfanInference
+from .pp_structurev3.self_hosted import PPStructureV3SelfHostedInference
+from ..selection import tool_for_model
+from ..providers import InferenceProvider, normalize_provider
+
+
+class InferenceFactory:
+    _registry: dict[tuple[str, InferenceProvider], Callable[..., Inference]] = {}
+
+    @classmethod
+    def register(
+        cls,
+        tool: str,
+        provider: InferenceProvider | str,
+        factory_fn: Callable[..., Inference],
+    ) -> None:
+        cls._registry[(tool, normalize_provider(provider))] = factory_fn
+
+    @classmethod
+    def create(
+        cls,
+        model: str,
+        provider: InferenceProvider | str,
+        **kwargs,
+    ) -> Inference:
+        tool = tool_for_model(model)
+        normalized_provider = normalize_provider(provider)
+        key = (tool, normalized_provider)
+        if key not in cls._registry:
+            supported = [
+                (tool, provider.value)
+                for tool, provider in sorted(
+                    cls._registry.keys(), key=lambda item: (item[0], item[1].value)
+                )
+            ]
+            raise ValueError(
+                f"Unsupported inference combination: model={model!r}, "
+                f"provider={normalized_provider.value!r}. Supported combinations: "
+                f"{supported}"
+            )
+        factory_fn = cls._registry[key]
+        return factory_fn(model=model, **kwargs)
+
+    @classmethod
+    def list_supported(cls) -> set[tuple[str, InferenceProvider]]:
+        return set(cls._registry.keys())
+
+
+def _create_ocr_local(model: str, **kwargs: Any) -> Inference:
+    return OCRLocalInference(
+        config=kwargs.get("config"),
+        device=kwargs.get("device"),
+        model=model,
+    )
+
+
+def _create_ocr_aistudio(model: str, **kwargs: Any) -> Inference:
+    return OCRAIStudioInference(
+        token=kwargs["token"],
+        base_url=kwargs.get("base_url"),
+        request_timeout=kwargs.get("request_timeout", 120.0),
+        poll_timeout=kwargs.get("poll_timeout", 600.0),
+        model=model,
+    )
+
+
+def _create_ocr_self_hosted(model: str, **kwargs: Any) -> Inference:
+    return OCRSelfHostedInference(
+        base_url=kwargs["base_url"],
+        http_timeout=kwargs.get("http_timeout", 600),
+    )
+
+
+def _create_pp_structurev3_local(model: str, **kwargs: Any) -> Inference:
+    return PPStructureV3LocalInference(
+        config=kwargs.get("config"),
+        device=kwargs.get("device"),
+    )
+
+
+def _create_pp_structurev3_aistudio(model: str, **kwargs: Any) -> Inference:
+    return PPStructureV3AIStudioInference(
+        token=kwargs["token"],
+        base_url=kwargs.get("base_url"),
+        request_timeout=kwargs.get("request_timeout", 120.0),
+        poll_timeout=kwargs.get("poll_timeout", 600.0),
+        model=model,
+    )
+
+
+def _create_pp_structurev3_self_hosted(model: str, **kwargs: Any) -> Inference:
+    return PPStructureV3SelfHostedInference(
+        base_url=kwargs["base_url"],
+        http_timeout=kwargs.get("http_timeout", 600),
+    )
+
+
+def _create_paddleocr_vl_local(model: str, **kwargs: Any) -> Inference:
+    return PaddleOCRVLLocalInference(
+        config=kwargs.get("config"),
+        device=kwargs.get("device"),
+        model=model,
+    )
+
+
+def _create_paddleocr_vl_aistudio(model: str, **kwargs: Any) -> Inference:
+    return PaddleOCRVLAIStudioInference(
+        token=kwargs["token"],
+        base_url=kwargs.get("base_url"),
+        request_timeout=kwargs.get("request_timeout", 120.0),
+        poll_timeout=kwargs.get("poll_timeout", 600.0),
+        model=model,
+    )
+
+
+def _create_paddleocr_vl_self_hosted(model: str, **kwargs: Any) -> Inference:
+    return PaddleOCRVLSelfHostedInference(
+        base_url=kwargs["base_url"],
+        http_timeout=kwargs.get("http_timeout", 600),
+    )
+
+
+def _create_pp_structurev3_qianfan(model: str, **kwargs: Any) -> Inference:
+    return PPStructureV3QianfanInference(
+        base_url=kwargs["base_url"],
+        api_key=kwargs["api_key"],
+        http_timeout=kwargs.get("http_timeout", 600),
+    )
+
+
+def _create_paddleocr_vl_qianfan(model: str, **kwargs: Any) -> Inference:
+    return PaddleOCRVLQianfanInference(
+        base_url=kwargs["base_url"],
+        api_key=kwargs["api_key"],
+        http_timeout=kwargs.get("http_timeout", 600),
+    )
+
+
+InferenceFactory.register("ocr", InferenceProvider.LOCAL, _create_ocr_local)
+InferenceFactory.register("ocr", InferenceProvider.AISTUDIO, _create_ocr_aistudio)
+InferenceFactory.register("ocr", InferenceProvider.SELF_HOSTED, _create_ocr_self_hosted)
+
+InferenceFactory.register(
+    "pp_structurev3", InferenceProvider.LOCAL, _create_pp_structurev3_local
+)
+InferenceFactory.register(
+    "pp_structurev3", InferenceProvider.AISTUDIO, _create_pp_structurev3_aistudio
+)
+InferenceFactory.register(
+    "pp_structurev3", InferenceProvider.QIANFAN, _create_pp_structurev3_qianfan
+)
+InferenceFactory.register(
+    "pp_structurev3", InferenceProvider.SELF_HOSTED, _create_pp_structurev3_self_hosted
+)
+
+InferenceFactory.register(
+    "paddleocr_vl", InferenceProvider.LOCAL, _create_paddleocr_vl_local
+)
+InferenceFactory.register(
+    "paddleocr_vl", InferenceProvider.AISTUDIO, _create_paddleocr_vl_aistudio
+)
+InferenceFactory.register(
+    "paddleocr_vl", InferenceProvider.QIANFAN, _create_paddleocr_vl_qianfan
+)
+InferenceFactory.register(
+    "paddleocr_vl", InferenceProvider.SELF_HOSTED, _create_paddleocr_vl_self_hosted
+)
+
+
+def create_inference(
+    model: str,
+    provider: InferenceProvider | str,
+    **kwargs,
+) -> Inference:
+    return InferenceFactory.create(model, provider, **kwargs)
diff --git a/mcp_server/paddleocr_mcp/inference/ocr/aistudio.py b/mcp_server/paddleocr_mcp/inference/ocr/aistudio.py
new file mode 100644
index 0000000..dd9433b
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/ocr/aistudio.py
@@ -0,0 +1,132 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Optional
+
+from ..shared.paddleocr_api_sdk import (
+    AsyncPaddleOCRClient,
+    APIError,
+    AuthError,
+    JobFailedError,
+    RequestTimeoutError,
+    ResponseFormatError,
+    ResultParseError,
+    ServiceUnavailableError,
+    Model,
+    OCROptions,
+    resolve_ocr_model,
+)
+
+from ..base import Inference
+from ..errors import (
+    AuthenticationError,
+    ExecutionTimeoutError,
+    InferenceError,
+    ResourceUnavailableError,
+)
+from ..shared.input_adapters import AISTUDIO_INPUT_ADAPTER, InputAdapter
+from ..types import InferenceRequest, OCRResult, TextLine
+from .params import OCR_DEFAULT_PARAMS, OCR_RUNTIME_PARAMS
+
+
+class OCRAIStudioInference(Inference):
+    def __init__(
+        self,
+        token: str,
+        base_url: Optional[str] = None,
+        request_timeout: float = 120.0,
+        poll_timeout: float = 600.0,
+        model: str = Model.PP_OCRV5.value,
+    ):
+        self._token = token
+        self._base_url = base_url
+        self._request_timeout = request_timeout
+        self._poll_timeout = poll_timeout
+        self._model = resolve_ocr_model(model)
+        self._client = None
+
+    @property
+    def input_adapter(self) -> InputAdapter:
+        return AISTUDIO_INPUT_ADAPTER
+
+    async def start(self) -> None:
+        try:
+            self._client = AsyncPaddleOCRClient(
+                token=self._token,
+                base_url=self._base_url,
+                request_timeout=self._request_timeout,
+                poll_timeout=self._poll_timeout,
+            )
+        except AuthError as e:
+            raise AuthenticationError(f"Authentication failed: {e}")
+
+    async def stop(self) -> None:
+        if self._client:
+            await self._client.close()
+            self._client = None
+
+    async def predict(self, request: InferenceRequest) -> OCRResult:
+        if not self._client:
+            raise RuntimeError("Inference not started")
+
+        try:
+            with self.input_adapter.prepare(request.input_data) as input_payload:
+                options = OCROptions(**request.runtime_params, visualize=False)
+
+                result = await self._client.ocr(
+                    model=self._model, **input_payload, options=options
+                )
+
+            return self._parse_result(result)
+
+        except AuthError as e:
+            raise AuthenticationError(f"Authentication failed: {e}")
+        except ServiceUnavailableError as e:
+            raise ResourceUnavailableError(f"Service unavailable: {e}")
+        except (JobFailedError, APIError, ResponseFormatError, ResultParseError) as e:
+            raise InferenceError(f"Execution failed: {e}")
+        except RequestTimeoutError as e:
+            raise ExecutionTimeoutError(f"Request timeout: {e}")
+
+    def get_valid_params(self) -> set[str]:
+        return set(OCR_RUNTIME_PARAMS.keys())
+
+    def get_default_params(self) -> dict[str, object]:
+        return OCR_DEFAULT_PARAMS.copy()
+
+    def _parse_result(self, result) -> OCRResult:
+        clean_texts, confidences, text_lines = [], [], []
+
+        for page_result in result.pages:
+            pruned = page_result.pruned_result
+            texts = pruned.get("rec_texts", [])
+            scores = pruned.get("rec_scores", [])
+            boxes = pruned.get("rec_boxes", [])
+
+            for i, text in enumerate(texts):
+                if text and text.strip():
+                    conf = scores[i] if i < len(scores) else 0
+                    clean_texts.append(text.strip())
+                    confidences.append(conf)
+                    text_lines.append(
+                        TextLine(
+                            text=text.strip(), confidence=round(conf, 3), bbox=boxes[i]
+                        )
+                    )
+
+        return OCRResult(
+            text="\n".join(clean_texts),
+            confidence=sum(confidences) / len(confidences) if confidences else 0,
+            text_lines=text_lines,
+        )
diff --git a/mcp_server/paddleocr_mcp/inference/ocr/local.py b/mcp_server/paddleocr_mcp/inference/ocr/local.py
new file mode 100644
index 0000000..07f609b
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/ocr/local.py
@@ -0,0 +1,131 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Any, Optional
+
+from paddleocr import PaddleOCR
+
+from ..base import Inference
+from ..shared.input_adapters import LOCAL_INPUT_ADAPTER, InputAdapter
+from ..shared.local_sync_runner import LocalSyncRunner
+from ..types import InferenceRequest, OCRResult, TextLine
+from .params import OCR_DEFAULT_PARAMS, OCR_RUNTIME_PARAMS
+
+_LOCAL_OCR_INIT_BY_MODEL: dict[str, dict[str, str]] = {
+    "PP-OCRv5": {"ocr_version": "PP-OCRv5"},
+    "PP-OCRv6": {"ocr_version": "PP-OCRv6"},
+    "PP-OCRv5-latin": {
+        "text_detection_model_name": "PP-OCRv5_server_det",
+        "text_recognition_model_name": "latin_PP-OCRv5_mobile_rec",
+    },
+}
+
+
+def _build_local_ocr_init_kwargs(
+    *,
+    config: Optional[str],
+    device: Optional[str],
+    model: Optional[str],
+) -> dict[str, Any]:
+    init_kwargs: dict[str, Any] = {
+        "paddlex_config": config,
+        "device": device,
+    }
+    if config is None and model is not None:
+        model_kwargs = _LOCAL_OCR_INIT_BY_MODEL.get(model)
+        if model_kwargs is None:
+            raise ValueError(f"Unsupported local OCR model: {model!r}")
+        init_kwargs.update(model_kwargs)
+    return init_kwargs
+
+
+class OCRLocalInference(Inference):
+    def __init__(
+        self,
+        config: Optional[str] = None,
+        device: Optional[str] = None,
+        model: Optional[str] = None,
+    ):
+        self._config = config
+        self._device = device
+        self._model = model
+        self._inference: Optional[Any] = None
+        self._wrapper: Optional[LocalSyncRunner] = None
+
+    @property
+    def input_adapter(self) -> InputAdapter:
+        return LOCAL_INPUT_ADAPTER
+
+    async def start(self) -> None:
+        try:
+            init_kwargs = _build_local_ocr_init_kwargs(
+                config=self._config,
+                device=self._device,
+                model=self._model,
+            )
+            self._inference = PaddleOCR(**init_kwargs)
+            self._wrapper = LocalSyncRunner(self._inference)
+        except Exception as e:
+            raise RuntimeError(f"Failed to create PaddleOCR inference: {str(e)}") from e
+
+    async def stop(self) -> None:
+        if self._wrapper:
+            await self._wrapper.close()
+            self._wrapper = None
+
+    async def predict(self, request: InferenceRequest) -> OCRResult:
+        if not self._wrapper:
+            raise RuntimeError("Inference not started")
+
+        with self.input_adapter.prepare(request.input_data) as processed_input:
+            result = await self._wrapper.call(
+                self._wrapper.inference.predict,
+                processed_input,
+                **request.runtime_params,
+            )
+
+        return self._parse_result(result)
+
+    def get_valid_params(self) -> set[str]:
+        return set(OCR_RUNTIME_PARAMS.keys())
+
+    def get_default_params(self) -> dict[str, Any]:
+        return OCR_DEFAULT_PARAMS.copy()
+
+    def _parse_result(self, result: Any) -> OCRResult:
+        clean_texts, confidences, text_lines = [], [], []
+
+        for res in result:
+            texts = res["rec_texts"]
+            scores = res["rec_scores"]
+            boxes = res["rec_boxes"]
+
+            for i, text in enumerate(texts):
+                if text and text.strip():
+                    conf = scores[i] if i < len(scores) else 0
+                    clean_texts.append(text.strip())
+                    confidences.append(conf)
+                    text_lines.append(
+                        TextLine(
+                            text=text.strip(),
+                            confidence=round(conf, 3),
+                            bbox=boxes[i].tolist(),
+                        )
+                    )
+
+        return OCRResult(
+            text="\n".join(clean_texts),
+            confidence=sum(confidences) / len(confidences) if confidences else 0,
+            text_lines=text_lines,
+        )
diff --git a/mcp_server/paddleocr_mcp/inference/ocr/params.py b/mcp_server/paddleocr_mcp/inference/ocr/params.py
new file mode 100644
index 0000000..556b8ec
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/ocr/params.py
@@ -0,0 +1,33 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Any
+
+
+OCR_RUNTIME_PARAMS: dict[str, type] = {
+    "use_doc_orientation_classify": bool,
+    "use_doc_unwarping": bool,
+    "use_textline_orientation": bool,
+    "text_det_limit_side_len": int,
+    "text_det_limit_type": str,
+    "text_det_thresh": float,
+    "text_det_box_thresh": float,
+    "text_det_unclip_ratio": float,
+    "text_rec_score_thresh": float,
+}
+
+OCR_DEFAULT_PARAMS: dict[str, Any] = {
+    "use_doc_orientation_classify": False,
+    "use_doc_unwarping": False,
+}
diff --git a/mcp_server/paddleocr_mcp/inference/ocr/self_hosted.py b/mcp_server/paddleocr_mcp/inference/ocr/self_hosted.py
new file mode 100644
index 0000000..924e41b
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/ocr/self_hosted.py
@@ -0,0 +1,45 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Any
+from typing_extensions import override
+
+from ...providers import InferenceProvider
+from ..shared.http_base import HTTPInferenceBase
+from ..types import InferenceResult
+from ..shared.http_result_parsers import parse_ocr_result
+from .params import OCR_DEFAULT_PARAMS, OCR_RUNTIME_PARAMS
+
+
+class OCRSelfHostedInference(HTTPInferenceBase):
+    def __init__(self, base_url: str, http_timeout: int = 600):
+        super().__init__(
+            base_url,
+            http_timeout,
+            api_key=None,
+            provider=InferenceProvider.SELF_HOSTED,
+        )
+
+    @override
+    def _get_endpoint(self) -> str:
+        return "ocr"
+
+    @override
+    def _get_params(self) -> tuple[set[str], dict[str, Any]]:
+        return set(OCR_RUNTIME_PARAMS.keys()), OCR_DEFAULT_PARAMS.copy()
+
+    @override
+    def _parse_result(self, response: dict[str, Any]) -> InferenceResult:
+        result_data = response.get("result", response)
+        return parse_ocr_result(result_data)
diff --git a/mcp_server/paddleocr_mcp/inference/paddleocr_vl/aistudio.py b/mcp_server/paddleocr_mcp/inference/paddleocr_vl/aistudio.py
new file mode 100644
index 0000000..7f29faa
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/paddleocr_vl/aistudio.py
@@ -0,0 +1,111 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Optional
+
+from ..shared.paddleocr_api_sdk import (
+    AsyncPaddleOCRClient,
+    APIError,
+    AuthError,
+    JobFailedError,
+    RequestTimeoutError,
+    ResponseFormatError,
+    ResultParseError,
+    ServiceUnavailableError,
+    Model,
+    PaddleOCRVLOptions,
+    resolve_document_model,
+)
+
+from ..base import Inference
+from ..errors import (
+    AuthenticationError,
+    ExecutionTimeoutError,
+    InferenceError,
+    ResourceUnavailableError,
+)
+from ..shared.doc_parsing_result_adapters import parse_aistudio_doc_parsing_result
+from ..shared.input_adapters import AISTUDIO_INPUT_ADAPTER, InputAdapter
+from ..types import DocParsingResult, InferenceRequest
+from .params import PADDLEOCR_VL_DEFAULT_PARAMS, PADDLEOCR_VL_RUNTIME_PARAMS
+
+
+class PaddleOCRVLAIStudioInference(Inference):
+    def __init__(
+        self,
+        token: str,
+        base_url: Optional[str] = None,
+        request_timeout: float = 120.0,
+        poll_timeout: float = 600.0,
+        model: str = Model.PADDLE_OCR_VL.value,
+    ):
+        self._token = token
+        self._base_url = base_url
+        self._request_timeout = request_timeout
+        self._poll_timeout = poll_timeout
+        self._model = resolve_document_model(model)
+        self._client = None
+
+    @property
+    def input_adapter(self) -> InputAdapter:
+        return AISTUDIO_INPUT_ADAPTER
+
+    async def start(self) -> None:
+        try:
+            self._client = AsyncPaddleOCRClient(
+                token=self._token,
+                base_url=self._base_url,
+                request_timeout=self._request_timeout,
+                poll_timeout=self._poll_timeout,
+            )
+        except AuthError as e:
+            raise AuthenticationError(f"Authentication failed: {e}")
+
+    async def stop(self) -> None:
+        if self._client:
+            await self._client.close()
+            self._client = None
+
+    async def predict(self, request: InferenceRequest) -> DocParsingResult:
+        if not self._client:
+            raise RuntimeError("Inference not started")
+
+        try:
+            with self.input_adapter.prepare(request.input_data) as input_payload:
+                options = PaddleOCRVLOptions(**request.runtime_params)
+
+                model = self._model
+                result = await self._client.parse_document(
+                    model=model, **input_payload, options=options
+                )
+
+            return self._parse_result(result)
+
+        except AuthError as e:
+            raise AuthenticationError(f"Authentication failed: {e}")
+        except ServiceUnavailableError as e:
+            raise ResourceUnavailableError(f"Service unavailable: {e}")
+        except (JobFailedError, APIError, ResponseFormatError, ResultParseError) as e:
+            raise InferenceError(f"Execution failed: {e}")
+        except RequestTimeoutError as e:
+            raise ExecutionTimeoutError(f"Request timeout: {e}")
+
+    def get_valid_params(self) -> set[str]:
+        return set(PADDLEOCR_VL_RUNTIME_PARAMS.keys())
+
+    def get_default_params(self) -> dict[str, object]:
+        return PADDLEOCR_VL_DEFAULT_PARAMS.copy()
+
+    def _parse_result(self, result) -> DocParsingResult:
+        return parse_aistudio_doc_parsing_result(result)
diff --git a/mcp_server/paddleocr_mcp/inference/paddleocr_vl/local.py b/mcp_server/paddleocr_mcp/inference/paddleocr_vl/local.py
new file mode 100644
index 0000000..edb74b8
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/paddleocr_vl/local.py
@@ -0,0 +1,96 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Any, Optional
+
+from ..base import Inference
+from ..shared.doc_parsing_result_adapters import parse_local_doc_parsing_result
+from ..shared.input_adapters import LOCAL_INPUT_ADAPTER, InputAdapter
+from ..shared.local_sync_runner import LocalSyncRunner
+from ..types import DocParsingResult, InferenceRequest
+from .params import PADDLEOCR_VL_DEFAULT_PARAMS, PADDLEOCR_VL_RUNTIME_PARAMS
+
+try:
+    from paddleocr import PaddleOCRVL
+
+    LOCAL_VL_AVAILABLE = True
+except ImportError:
+    LOCAL_VL_AVAILABLE = False
+
+_PIPELINE_VERSION_BY_MODEL = {
+    "PaddleOCR-VL": "v1",
+    "PaddleOCR-VL-1.5": "v1.5",
+    "PaddleOCR-VL-1.6": "v1.6",
+}
+
+
+class PaddleOCRVLLocalInference(Inference):
+    def __init__(
+        self,
+        config: Optional[str] = None,
+        device: Optional[str] = None,
+        model: str = "PaddleOCR-VL",
+    ):
+        self._config = config
+        self._device = device
+        self._model = model
+        self._inference: Optional[Any] = None
+        self._wrapper: Optional[LocalSyncRunner] = None
+
+    @property
+    def input_adapter(self) -> InputAdapter:
+        return LOCAL_INPUT_ADAPTER
+
+    async def start(self) -> None:
+        if not LOCAL_VL_AVAILABLE:
+            raise RuntimeError("PaddleOCRVL is not locally available")
+        try:
+            pipeline_version = _PIPELINE_VERSION_BY_MODEL[self._model]
+            self._inference = PaddleOCRVL(
+                pipeline_version=pipeline_version,
+                paddlex_config=self._config,
+                device=self._device,
+            )
+            self._wrapper = LocalSyncRunner(self._inference)
+        except Exception as e:
+            raise RuntimeError(
+                f"Failed to create PaddleOCRVL inference: {str(e)}"
+            ) from e
+
+    async def stop(self) -> None:
+        if self._wrapper:
+            await self._wrapper.close()
+            self._wrapper = None
+
+    async def predict(self, request: InferenceRequest) -> DocParsingResult:
+        if not self._wrapper:
+            raise RuntimeError("Inference not started")
+
+        with self.input_adapter.prepare(request.input_data) as processed_input:
+            result = await self._wrapper.call(
+                self._wrapper.inference.predict,
+                processed_input,
+                **request.runtime_params,
+            )
+
+        return self._parse_result(result)
+
+    def get_valid_params(self) -> set[str]:
+        return set(PADDLEOCR_VL_RUNTIME_PARAMS.keys())
+
+    def get_default_params(self) -> dict[str, Any]:
+        return PADDLEOCR_VL_DEFAULT_PARAMS.copy()
+
+    def _parse_result(self, result: Any) -> DocParsingResult:
+        return parse_local_doc_parsing_result(result)
diff --git a/mcp_server/paddleocr_mcp/inference/paddleocr_vl/params.py b/mcp_server/paddleocr_mcp/inference/paddleocr_vl/params.py
new file mode 100644
index 0000000..67e6226
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/paddleocr_vl/params.py
@@ -0,0 +1,48 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Any
+
+
+PADDLEOCR_VL_RUNTIME_PARAMS: dict[str, type] = {
+    "use_doc_orientation_classify": bool,
+    "use_doc_unwarping": bool,
+    "use_layout_detection": bool,
+    "use_chart_recognition": bool,
+    "use_seal_recognition": bool,
+    "use_ocr_for_image_block": bool,
+    "layout_threshold": float,
+    "layout_nms": bool,
+    "layout_unclip_ratio": float,
+    "layout_merge_bboxes_mode": int,
+    "layout_shape_mode": str,
+    "prompt_label": str,
+    "format_block_content": bool,
+    "repetition_penalty": float,
+    "temperature": float,
+    "top_p": float,
+    "min_pixels": int,
+    "max_pixels": int,
+    "max_new_tokens": int,
+    "vlm_extra_args": dict,
+    "merge_layout_blocks": bool,
+    "markdown_ignore_labels": list,
+}
+
+PADDLEOCR_VL_DEFAULT_PARAMS: dict[str, Any] = {
+    "use_doc_orientation_classify": False,
+    "use_doc_unwarping": False,
+    "use_chart_recognition": True,
+    "use_seal_recognition": True,
+}
diff --git a/mcp_server/paddleocr_mcp/inference/paddleocr_vl/qianfan.py b/mcp_server/paddleocr_mcp/inference/paddleocr_vl/qianfan.py
new file mode 100644
index 0000000..81fa6a8
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/paddleocr_vl/qianfan.py
@@ -0,0 +1,45 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Any
+from typing_extensions import override
+
+from ...providers import InferenceProvider
+from ..shared.http_base import HTTPInferenceBase
+from ..types import InferenceResult
+from ..shared.http_result_parsers import parse_doc_parsing_result
+from .params import PADDLEOCR_VL_DEFAULT_PARAMS, PADDLEOCR_VL_RUNTIME_PARAMS
+
+
+class PaddleOCRVLQianfanInference(HTTPInferenceBase):
+    def __init__(self, base_url: str, api_key: str, http_timeout: int = 600):
+        super().__init__(
+            base_url, http_timeout, api_key, provider=InferenceProvider.QIANFAN
+        )
+
+    @override
+    def _get_endpoint(self) -> str:
+        return "paddleocr"
+
+    @override
+    def _get_params(self) -> tuple[set[str], dict[str, Any]]:
+        return (
+            set(PADDLEOCR_VL_RUNTIME_PARAMS.keys()),
+            PADDLEOCR_VL_DEFAULT_PARAMS.copy(),
+        )
+
+    @override
+    def _parse_result(self, response: dict[str, Any]) -> InferenceResult:
+        result_data = response.get("result", response)
+        return parse_doc_parsing_result(result_data)
diff --git a/mcp_server/paddleocr_mcp/inference/paddleocr_vl/self_hosted.py b/mcp_server/paddleocr_mcp/inference/paddleocr_vl/self_hosted.py
new file mode 100644
index 0000000..2b4ebb6
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/paddleocr_vl/self_hosted.py
@@ -0,0 +1,48 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Any
+from typing_extensions import override
+
+from ...providers import InferenceProvider
+from ..shared.http_base import HTTPInferenceBase
+from ..types import InferenceResult
+from ..shared.http_result_parsers import parse_doc_parsing_result
+from .params import PADDLEOCR_VL_DEFAULT_PARAMS, PADDLEOCR_VL_RUNTIME_PARAMS
+
+
+class PaddleOCRVLSelfHostedInference(HTTPInferenceBase):
+    def __init__(self, base_url: str, http_timeout: int = 600):
+        super().__init__(
+            base_url,
+            http_timeout,
+            api_key=None,
+            provider=InferenceProvider.SELF_HOSTED,
+        )
+
+    @override
+    def _get_endpoint(self) -> str:
+        return "layout-parsing"
+
+    @override
+    def _get_params(self) -> tuple[set[str], dict[str, Any]]:
+        return (
+            set(PADDLEOCR_VL_RUNTIME_PARAMS.keys()),
+            PADDLEOCR_VL_DEFAULT_PARAMS.copy(),
+        )
+
+    @override
+    def _parse_result(self, response: dict[str, Any]) -> InferenceResult:
+        result_data = response.get("result", response)
+        return parse_doc_parsing_result(result_data)
diff --git a/mcp_server/paddleocr_mcp/inference/pp_structurev3/aistudio.py b/mcp_server/paddleocr_mcp/inference/pp_structurev3/aistudio.py
new file mode 100644
index 0000000..c1b413c
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/pp_structurev3/aistudio.py
@@ -0,0 +1,112 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Optional
+
+from ..shared.paddleocr_api_sdk import (
+    AsyncPaddleOCRClient,
+    APIError,
+    AuthError,
+    JobFailedError,
+    RequestTimeoutError,
+    ResponseFormatError,
+    ResultParseError,
+    ServiceUnavailableError,
+    Model,
+    PPStructureV3Options,
+    resolve_document_model,
+)
+
+from ..base import Inference
+from ..errors import (
+    AuthenticationError,
+    ExecutionTimeoutError,
+    InferenceError,
+    ResourceUnavailableError,
+)
+from ..shared.doc_parsing_result_adapters import parse_aistudio_doc_parsing_result
+from ..shared.input_adapters import AISTUDIO_INPUT_ADAPTER, InputAdapter
+from ..types import DocParsingResult, InferenceRequest
+from .params import PP_STRUCTUREV3_DEFAULT_PARAMS, PP_STRUCTUREV3_RUNTIME_PARAMS
+
+
+class PPStructureV3AIStudioInference(Inference):
+    def __init__(
+        self,
+        token: str,
+        base_url: Optional[str] = None,
+        request_timeout: float = 120.0,
+        poll_timeout: float = 600.0,
+        model: str = Model.PP_STRUCTURE_V3.value,
+    ):
+        self._token = token
+        self._base_url = base_url
+        self._request_timeout = request_timeout
+        self._poll_timeout = poll_timeout
+        self._model = resolve_document_model(model)
+        self._client = None
+
+    @property
+    def input_adapter(self) -> InputAdapter:
+        return AISTUDIO_INPUT_ADAPTER
+
+    async def start(self) -> None:
+        try:
+            self._client = AsyncPaddleOCRClient(
+                token=self._token,
+                base_url=self._base_url,
+                request_timeout=self._request_timeout,
+                poll_timeout=self._poll_timeout,
+            )
+        except AuthError as e:
+            raise AuthenticationError(f"Authentication failed: {e}")
+
+    async def stop(self) -> None:
+        if self._client:
+            await self._client.close()
+            self._client = None
+
+    async def predict(self, request: InferenceRequest) -> DocParsingResult:
+        if not self._client:
+            raise RuntimeError("Inference not started")
+
+        try:
+            with self.input_adapter.prepare(request.input_data) as input_payload:
+                options = PPStructureV3Options(
+                    **request.runtime_params, visualize=False
+                )
+
+                result = await self._client.parse_document(
+                    model=self._model, **input_payload, options=options
+                )
+
+            return self._parse_result(result)
+
+        except AuthError as e:
+            raise AuthenticationError(f"Authentication failed: {e}")
+        except ServiceUnavailableError as e:
+            raise ResourceUnavailableError(f"Service unavailable: {e}")
+        except (JobFailedError, APIError, ResponseFormatError, ResultParseError) as e:
+            raise InferenceError(f"Execution failed: {e}")
+        except RequestTimeoutError as e:
+            raise ExecutionTimeoutError(f"Request timeout: {e}")
+
+    def get_valid_params(self) -> set[str]:
+        return set(PP_STRUCTUREV3_RUNTIME_PARAMS.keys())
+
+    def get_default_params(self) -> dict[str, object]:
+        return PP_STRUCTUREV3_DEFAULT_PARAMS.copy()
+
+    def _parse_result(self, result) -> DocParsingResult:
+        return parse_aistudio_doc_parsing_result(result)
diff --git a/mcp_server/paddleocr_mcp/inference/pp_structurev3/local.py b/mcp_server/paddleocr_mcp/inference/pp_structurev3/local.py
new file mode 100644
index 0000000..dff4c3f
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/pp_structurev3/local.py
@@ -0,0 +1,81 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Any, Optional
+
+from ..base import Inference
+from ..shared.doc_parsing_result_adapters import parse_local_doc_parsing_result
+from ..shared.input_adapters import LOCAL_INPUT_ADAPTER, InputAdapter
+from ..shared.local_sync_runner import LocalSyncRunner
+from ..types import DocParsingResult, InferenceRequest
+from .params import PP_STRUCTUREV3_DEFAULT_PARAMS, PP_STRUCTUREV3_RUNTIME_PARAMS
+
+try:
+    from paddleocr import PPStructureV3
+
+    LOCAL_PPSTRUCTURE_AVAILABLE = True
+except ImportError:
+    LOCAL_PPSTRUCTURE_AVAILABLE = False
+
+
+class PPStructureV3LocalInference(Inference):
+    def __init__(self, config: Optional[str] = None, device: Optional[str] = None):
+        self._config = config
+        self._device = device
+        self._inference: Optional[Any] = None
+        self._wrapper: Optional[LocalSyncRunner] = None
+
+    @property
+    def input_adapter(self) -> InputAdapter:
+        return LOCAL_INPUT_ADAPTER
+
+    async def start(self) -> None:
+        if not LOCAL_PPSTRUCTURE_AVAILABLE:
+            raise RuntimeError("PPStructureV3 is not locally available")
+        try:
+            self._inference = PPStructureV3(
+                paddlex_config=self._config, device=self._device
+            )
+            self._wrapper = LocalSyncRunner(self._inference)
+        except Exception as e:
+            raise RuntimeError(
+                f"Failed to create PPStructureV3 inference: {str(e)}"
+            ) from e
+
+    async def stop(self) -> None:
+        if self._wrapper:
+            await self._wrapper.close()
+            self._wrapper = None
+
+    async def predict(self, request: InferenceRequest) -> DocParsingResult:
+        if not self._wrapper:
+            raise RuntimeError("Inference not started")
+
+        with self.input_adapter.prepare(request.input_data) as processed_input:
+            result = await self._wrapper.call(
+                self._wrapper.inference.predict,
+                processed_input,
+                **request.runtime_params,
+            )
+
+        return self._parse_result(result)
+
+    def get_valid_params(self) -> set[str]:
+        return set(PP_STRUCTUREV3_RUNTIME_PARAMS.keys())
+
+    def get_default_params(self) -> dict[str, Any]:
+        return PP_STRUCTUREV3_DEFAULT_PARAMS.copy()
+
+    def _parse_result(self, result: Any) -> DocParsingResult:
+        return parse_local_doc_parsing_result(result)
diff --git a/mcp_server/paddleocr_mcp/inference/pp_structurev3/params.py b/mcp_server/paddleocr_mcp/inference/pp_structurev3/params.py
new file mode 100644
index 0000000..67132d5
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/pp_structurev3/params.py
@@ -0,0 +1,57 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Any
+
+
+PP_STRUCTUREV3_RUNTIME_PARAMS: dict[str, type] = {
+    "use_doc_orientation_classify": bool,
+    "use_doc_unwarping": bool,
+    "use_textline_orientation": bool,
+    "use_seal_recognition": bool,
+    "use_table_recognition": bool,
+    "use_formula_recognition": bool,
+    "use_chart_recognition": bool,
+    "use_region_detection": bool,
+    "layout_threshold": float,
+    "layout_nms": bool,
+    "layout_unclip_ratio": float,
+    "layout_merge_bboxes_mode": int,
+    "format_block_content": bool,
+    "text_det_limit_side_len": int,
+    "text_det_limit_type": str,
+    "text_det_thresh": float,
+    "text_det_box_thresh": float,
+    "text_det_unclip_ratio": float,
+    "text_rec_score_thresh": float,
+    "seal_det_limit_side_len": int,
+    "seal_det_limit_type": str,
+    "seal_det_thresh": float,
+    "seal_det_box_thresh": float,
+    "seal_det_unclip_ratio": float,
+    "seal_rec_score_thresh": float,
+    "use_wired_table_cells_trans_to_html": bool,
+    "use_wireless_table_cells_trans_to_html": bool,
+    "use_table_orientation_classify": bool,
+    "use_ocr_results_with_table_cells": bool,
+    "use_e2e_wired_table_rec_model": bool,
+    "use_e2e_wireless_table_rec_model": bool,
+    "markdown_ignore_labels": list,
+}
+
+PP_STRUCTUREV3_DEFAULT_PARAMS: dict[str, Any] = {
+    "use_doc_orientation_classify": False,
+    "use_doc_unwarping": False,
+    "use_chart_recognition": True,
+}
diff --git a/mcp_server/paddleocr_mcp/inference/pp_structurev3/qianfan.py b/mcp_server/paddleocr_mcp/inference/pp_structurev3/qianfan.py
new file mode 100644
index 0000000..3b66a1e
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/pp_structurev3/qianfan.py
@@ -0,0 +1,45 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Any
+from typing_extensions import override
+
+from ...providers import InferenceProvider
+from ..shared.http_base import HTTPInferenceBase
+from ..types import InferenceResult
+from ..shared.http_result_parsers import parse_doc_parsing_result
+from .params import PP_STRUCTUREV3_DEFAULT_PARAMS, PP_STRUCTUREV3_RUNTIME_PARAMS
+
+
+class PPStructureV3QianfanInference(HTTPInferenceBase):
+    def __init__(self, base_url: str, api_key: str, http_timeout: int = 600):
+        super().__init__(
+            base_url, http_timeout, api_key, provider=InferenceProvider.QIANFAN
+        )
+
+    @override
+    def _get_endpoint(self) -> str:
+        return "paddleocr"
+
+    @override
+    def _get_params(self) -> tuple[set[str], dict[str, Any]]:
+        return (
+            set(PP_STRUCTUREV3_RUNTIME_PARAMS.keys()),
+            PP_STRUCTUREV3_DEFAULT_PARAMS.copy(),
+        )
+
+    @override
+    def _parse_result(self, response: dict[str, Any]) -> InferenceResult:
+        result_data = response.get("result", response)
+        return parse_doc_parsing_result(result_data)
diff --git a/mcp_server/paddleocr_mcp/inference/pp_structurev3/self_hosted.py b/mcp_server/paddleocr_mcp/inference/pp_structurev3/self_hosted.py
new file mode 100644
index 0000000..09a82ca
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/pp_structurev3/self_hosted.py
@@ -0,0 +1,48 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Any
+from typing_extensions import override
+
+from ...providers import InferenceProvider
+from ..shared.http_base import HTTPInferenceBase
+from ..types import InferenceResult
+from ..shared.http_result_parsers import parse_doc_parsing_result
+from .params import PP_STRUCTUREV3_DEFAULT_PARAMS, PP_STRUCTUREV3_RUNTIME_PARAMS
+
+
+class PPStructureV3SelfHostedInference(HTTPInferenceBase):
+    def __init__(self, base_url: str, http_timeout: int = 600):
+        super().__init__(
+            base_url,
+            http_timeout,
+            api_key=None,
+            provider=InferenceProvider.SELF_HOSTED,
+        )
+
+    @override
+    def _get_endpoint(self) -> str:
+        return "layout-parsing"
+
+    @override
+    def _get_params(self) -> tuple[set[str], dict[str, Any]]:
+        return (
+            set(PP_STRUCTUREV3_RUNTIME_PARAMS.keys()),
+            PP_STRUCTUREV3_DEFAULT_PARAMS.copy(),
+        )
+
+    @override
+    def _parse_result(self, response: dict[str, Any]) -> InferenceResult:
+        result_data = response.get("result", response)
+        return parse_doc_parsing_result(result_data)
diff --git a/mcp_server/paddleocr_mcp/inference/shared/async_http_client.py b/mcp_server/paddleocr_mcp/inference/shared/async_http_client.py
new file mode 100644
index 0000000..f8a3f05
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/shared/async_http_client.py
@@ -0,0 +1,61 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Any, Optional
+
+import httpx
+
+
+class AsyncHTTPClient:
+    def __init__(
+        self,
+        base_url: str,
+        http_timeout: int = 600,
+        headers: Optional[dict[str, str]] = None,
+    ):
+        self._base_url = base_url
+        self._http_timeout = http_timeout
+        self._headers = headers or {}
+        self._client: Optional[httpx.AsyncClient] = None
+
+    async def start(self) -> None:
+        write_timeout = min(float(self._http_timeout), 120.0)
+        timeout = httpx.Timeout(
+            connect=30.0,
+            read=float(self._http_timeout),
+            write=write_timeout,
+            pool=30.0,
+        )
+        self._client = httpx.AsyncClient(timeout=timeout)
+
+    async def stop(self) -> None:
+        if self._client:
+            await self._client.aclose()
+            self._client = None
+
+    async def post(
+        self,
+        endpoint: str,
+        payload: dict[str, Any],
+        headers: Optional[dict[str, str]] = None,
+    ) -> dict[str, Any]:
+        if not self._client:
+            raise RuntimeError("HTTP client not started")
+
+        url = f"{self._base_url.rstrip('/')}/{endpoint.lstrip('/')}"
+        merged_headers = {**self._headers, **(headers or {})}
+
+        response = await self._client.post(url, json=payload, headers=merged_headers)
+        response.raise_for_status()
+        return response.json()
diff --git a/mcp_server/paddleocr_mcp/inference/shared/doc_parsing_result_adapters.py b/mcp_server/paddleocr_mcp/inference/shared/doc_parsing_result_adapters.py
new file mode 100644
index 0000000..3b2d606
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/shared/doc_parsing_result_adapters.py
@@ -0,0 +1,57 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import base64
+import io
+from typing import Any
+
+from ..types import DocParsingResult
+
+
+def parse_local_doc_parsing_result(result: Any) -> DocParsingResult:
+    markdown_parts: list[str] = []
+    all_images_mapping: dict[str, str] = {}
+
+    for res in result:
+        markdown = res.markdown
+        markdown_parts.append(markdown["markdown_texts"])
+        processed_images = {}
+        for img_key, img_data in markdown["markdown_images"].items():
+            with io.BytesIO() as buffer:
+                img_data.save(buffer, format="JPEG")
+                processed_images[img_key] = base64.b64encode(buffer.getvalue()).decode(
+                    "ascii"
+                )
+        all_images_mapping.update(processed_images)
+
+    return DocParsingResult(
+        markdown="\n".join(markdown_parts),
+        pages=len(result),
+        images_mapping=all_images_mapping,
+    )
+
+
+def parse_aistudio_doc_parsing_result(result: Any) -> DocParsingResult:
+    markdown_parts: list[str] = []
+    all_images_mapping: dict[str, str] = {}
+
+    for page in result.pages:
+        markdown_parts.append(page.markdown_text)
+        all_images_mapping.update(page.markdown_images)
+
+    return DocParsingResult(
+        markdown="\n".join(markdown_parts),
+        pages=len(result.pages),
+        images_mapping=all_images_mapping,
+    )
diff --git a/mcp_server/paddleocr_mcp/inference/shared/http_base.py b/mcp_server/paddleocr_mcp/inference/shared/http_base.py
new file mode 100644
index 0000000..94cdfeb
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/shared/http_base.py
@@ -0,0 +1,123 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from abc import abstractmethod
+from typing import Any, Optional
+
+import httpx
+
+from ..base import Inference
+from ..errors import InferenceError
+from ..types import InferenceRequest, InferenceResult
+from .async_http_client import AsyncHTTPClient
+from .input_adapters import HTTPInputAdapter, InputAdapter
+from .param_mapping import convert_params_to_camel
+from ...providers import InferenceProvider
+
+
+class HTTPInferenceBase(Inference):
+    def __init__(
+        self,
+        base_url: str,
+        http_timeout: int = 600,
+        api_key: Optional[str] = None,
+        provider: InferenceProvider | str = InferenceProvider.SELF_HOSTED,
+    ):
+        self._base_url = base_url
+        self._http_timeout = http_timeout
+        self._api_key = api_key
+        self._input_adapter = HTTPInputAdapter(provider)
+        self._client: Optional[AsyncHTTPClient] = None
+
+    @property
+    def input_adapter(self) -> InputAdapter:
+        return self._input_adapter
+
+    async def start(self) -> None:
+        headers = {}
+        if self._api_key:
+            headers["Authorization"] = f"Bearer {self._api_key}"
+        self._client = AsyncHTTPClient(self._base_url, self._http_timeout, headers)
+        await self._client.start()
+
+    async def stop(self) -> None:
+        if self._client:
+            await self._client.stop()
+            self._client = None
+
+    async def predict(self, request: InferenceRequest) -> InferenceResult:
+        if not self._client:
+            raise RuntimeError("Inference not started")
+
+        payload = self._prepare_payload(
+            request.input_data,
+            request.file_type,
+            **request.runtime_params,
+        )
+
+        endpoint = self._get_endpoint()
+        provider = self._input_adapter.provider.value
+        try:
+            response = await self._client.post(endpoint, payload)
+            return self._parse_result(response)
+        except httpx.ReadTimeout as e:
+            raise InferenceError(
+                f"HTTP read timeout after {self._http_timeout}s "
+                f"({provider}/{endpoint}): {e}"
+            ) from e
+        except httpx.HTTPStatusError as e:
+            raise InferenceError(
+                f"HTTP {e.response.status_code} ({provider}/{endpoint}): {e}"
+            ) from e
+        except httpx.HTTPError as e:
+            raise InferenceError(f"HTTP error ({provider}/{endpoint}): {e}") from e
+        except Exception as e:
+            raise InferenceError(
+                f"HTTP request failed ({provider}/{endpoint}): {e}"
+            ) from e
+
+    def _prepare_payload(
+        self, input_data: str, file_type: Optional[str], **params
+    ) -> dict[str, Any]:
+        http_adapter = self._input_adapter
+        payload: dict[str, Any] = {
+            "file": http_adapter.prepare_http_file_field(input_data)
+        }
+        if file_type == "image":
+            payload["fileType"] = 1
+        elif file_type == "pdf":
+            payload["fileType"] = 0
+
+        payload.update(convert_params_to_camel({**params, "visualize": False}))
+        return payload
+
+    @abstractmethod
+    def _get_endpoint(self) -> str:
+        pass
+
+    @abstractmethod
+    def _get_params(self) -> tuple[set[str], dict[str, Any]]:
+        pass
+
+    @abstractmethod
+    def _parse_result(self, response: dict[str, Any]) -> InferenceResult:
+        pass
+
+    def get_valid_params(self) -> set[str]:
+        valid_params, _ = self._get_params()
+        return valid_params
+
+    def get_default_params(self) -> dict[str, Any]:
+        _, defaults = self._get_params()
+        return defaults
diff --git a/mcp_server/paddleocr_mcp/inference/shared/http_result_parsers.py b/mcp_server/paddleocr_mcp/inference/shared/http_result_parsers.py
new file mode 100644
index 0000000..1035546
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/shared/http_result_parsers.py
@@ -0,0 +1,66 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Any
+
+from ..types import DocParsingResult, OCRResult, TextLine
+
+
+def parse_ocr_result(result_data: dict[str, Any]) -> OCRResult:
+    """Parse OCR response payload."""
+    ocr_results = result_data.get("ocrResults", [])
+    all_texts, all_confidences, text_lines = [], [], []
+
+    for ocr_result in ocr_results:
+        pruned = ocr_result["prunedResult"]
+        texts = pruned["rec_texts"]
+        scores = pruned["rec_scores"]
+        boxes = pruned["rec_boxes"]
+
+        for i, text in enumerate(texts):
+            if text and text.strip():
+                conf = scores[i] if i < len(scores) else 0
+                all_texts.append(text.strip())
+                all_confidences.append(conf)
+                text_lines.append(
+                    TextLine(
+                        text=text.strip(), confidence=round(conf, 3), bbox=boxes[i]
+                    )
+                )
+
+    return OCRResult(
+        text="\n".join(all_texts),
+        confidence=(
+            sum(all_confidences) / len(all_confidences) if all_confidences else 0
+        ),
+        text_lines=text_lines,
+    )
+
+
+def parse_doc_parsing_result(result_data: dict[str, Any]) -> DocParsingResult:
+    """Parse document parsing response payload."""
+    doc_parsing_results = result_data.get("layoutParsingResults", [])
+    markdown_parts = []
+    all_images_mapping = {}
+
+    for res in doc_parsing_results:
+        markdown_parts.append(res["markdown"]["text"])
+        images = res["markdown"]["images"]
+        all_images_mapping.update(images)
+
+    return DocParsingResult(
+        markdown="\n".join(markdown_parts),
+        pages=len(doc_parsing_results),
+        images_mapping=all_images_mapping,
+    )
diff --git a/mcp_server/paddleocr_mcp/inference/shared/input_adapters.py b/mcp_server/paddleocr_mcp/inference/shared/input_adapters.py
new file mode 100644
index 0000000..a873e7b
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/shared/input_adapters.py
@@ -0,0 +1,182 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Provider-specific input adapters for the unified MCP input contract."""
+
+from __future__ import annotations
+
+import abc
+import base64
+import io
+from contextlib import contextmanager
+from typing import Any, Generator, Union
+
+from .input_contract import (
+    COMMON_INPUT_DESCRIPTION,
+    InputKind,
+    classify_input,
+    decode_input_bytes,
+    extract_base64_payload,
+    infer_file_type_from_bytes,
+    materialize_bytes_as_temp_file,
+    normalize,
+    resolve_absolute_path,
+    validate_external_contract,
+)
+from ...providers import (
+    InferenceProvider,
+    get_provider_spec,
+    is_http_provider,
+    normalize_provider,
+)
+
+
+class InputAdapter(abc.ABC):
+    @property
+    @abc.abstractmethod
+    def provider(self) -> InferenceProvider:
+        pass
+
+    @property
+    def description(self) -> str:
+        spec = get_provider_spec(self.provider)
+        return (
+            f"Inference provider: {spec.provider.value} ({spec.display_name}). "
+            f"{COMMON_INPUT_DESCRIPTION}"
+        )
+
+    def normalize(self, input_data: str) -> str:
+        return normalize(input_data)
+
+    def validate(self, input_data: str) -> None:
+        validate_external_contract(input_data)
+
+    @abc.abstractmethod
+    @contextmanager
+    def prepare(self, input_data: str) -> Generator[Any, None, None]:
+        """Convert user input into the inference provider's native form."""
+        raise NotImplementedError
+
+
+class LocalInputAdapter(InputAdapter):
+    @property
+    def provider(self) -> InferenceProvider:
+        return InferenceProvider.LOCAL
+
+    @contextmanager
+    def prepare(self, input_data: str) -> Generator[Union[str, Any], None, None]:
+        value = normalize(input_data)
+        kind = classify_input(value)
+
+        if kind is InputKind.URL:
+            yield value
+            return
+
+        if kind is InputKind.ABSOLUTE_PATH:
+            yield str(resolve_absolute_path(value))
+            return
+
+        data = decode_input_bytes(value, kind)
+        file_type = infer_file_type_from_bytes(data)
+        if file_type == "image":
+            import numpy as np
+            from PIL import Image as PILImage
+
+            try:
+                image_pil = PILImage.open(io.BytesIO(data))
+                image_arr = np.array(image_pil.convert("RGB"))
+                yield np.ascontiguousarray(image_arr[..., ::-1])
+            except Exception as e:
+                raise ValueError(f"Failed to decode Base64 image: {e}") from e
+            return
+
+        if file_type == "pdf":
+            with materialize_bytes_as_temp_file(data, file_type="pdf") as temp_path:
+                yield str(temp_path)
+            return
+
+        raise ValueError(
+            "Unsupported Base64 content type for local inference. "
+            "Only image and PDF inputs are supported."
+        )
+
+
+class AIStudioInputAdapter(InputAdapter):
+    @property
+    def provider(self) -> InferenceProvider:
+        return InferenceProvider.AISTUDIO
+
+    @contextmanager
+    def prepare(self, input_data: str) -> Generator[dict[str, str], None, None]:
+        value = normalize(input_data)
+        kind = classify_input(value)
+
+        if kind is InputKind.URL:
+            yield {"file_url": value}
+            return
+
+        if kind is InputKind.ABSOLUTE_PATH:
+            yield {"file_path": str(resolve_absolute_path(value))}
+            return
+
+        data = decode_input_bytes(value, kind)
+        file_type = infer_file_type_from_bytes(data)
+        if file_type not in ("image", "pdf"):
+            raise ValueError(
+                "Unsupported Base64 content type for AI Studio. "
+                "Only image and PDF inputs are supported."
+            )
+
+        with materialize_bytes_as_temp_file(data, file_type=file_type) as temp_path:
+            yield {"file_path": str(temp_path)}
+
+
+class HTTPInputAdapter(InputAdapter):
+    def __init__(self, provider: InferenceProvider | str) -> None:
+        normalized_provider = normalize_provider(provider)
+        if not is_http_provider(normalized_provider):
+            raise ValueError(
+                f"HTTPInputAdapter requires an HTTP transport provider, "
+                f"got {normalized_provider.value!r}."
+            )
+        self._provider = normalized_provider
+
+    @property
+    def provider(self) -> InferenceProvider:
+        return self._provider
+
+    @contextmanager
+    def prepare(self, input_data: str) -> Generator[str, None, None]:
+        yield self.prepare_http_file_field(input_data)
+
+    def prepare_http_file_field(self, input_data: str) -> str:
+        value = normalize(input_data)
+        kind = classify_input(value)
+
+        if kind is InputKind.URL:
+            return value
+
+        if kind is InputKind.ABSOLUTE_PATH:
+            path = resolve_absolute_path(value)
+            return base64.b64encode(path.read_bytes()).decode("ascii")
+
+        payload = extract_base64_payload(value)
+        decode_input_bytes(value, kind)
+        return payload
+
+
+LOCAL_INPUT_ADAPTER = LocalInputAdapter()
+AISTUDIO_INPUT_ADAPTER = AIStudioInputAdapter()
+SELF_HOSTED_INPUT_ADAPTER = HTTPInputAdapter(InferenceProvider.SELF_HOSTED)
+QIANFAN_INPUT_ADAPTER = HTTPInputAdapter(InferenceProvider.QIANFAN)
diff --git a/mcp_server/paddleocr_mcp/inference/shared/input_contract.py b/mcp_server/paddleocr_mcp/inference/shared/input_contract.py
new file mode 100644
index 0000000..58b2bd5
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/shared/input_contract.py
@@ -0,0 +1,153 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Shared user-facing input contract for all MCP inference providers."""
+
+from __future__ import annotations
+
+import tempfile
+from contextlib import contextmanager
+from enum import Enum
+from pathlib import Path
+from typing import Generator, Optional
+
+from ...utils import (
+    decode_base64_payload,
+    extract_base64_payload,
+    infer_file_type_from_bytes,
+    is_base64,
+    is_url,
+)
+
+COMMON_INPUT_DESCRIPTION = (
+    "Supported input_data: absolute file path accessible to the MCP server process "
+    "(~ is expanded), HTTP(S) URL, raw Base64, or data URL. Relative paths are "
+    "rejected. Backend-specific conversion is handled by the MCP server."
+)
+
+_INPUT_EXTENSIONS = frozenset(
+    {
+        ".jpg",
+        ".jpeg",
+        ".png",
+        ".pdf",
+        ".bmp",
+        ".webp",
+        ".tiff",
+        ".tif",
+        ".gif",
+    }
+)
+
+_SUFFIX_BY_FILE_TYPE = {
+    "image": ".png",
+    "pdf": ".pdf",
+}
+
+
+class InputKind(str, Enum):
+    URL = "url"
+    ABSOLUTE_PATH = "absolute_path"
+    BASE64 = "base64"
+    DATA_URL = "data_url"
+
+
+def normalize(input_data: str) -> str:
+    return input_data.strip()
+
+
+def looks_like_file_path(value: str) -> bool:
+    if is_url(value) or value.startswith("data:"):
+        return False
+    if value.startswith("~") or "/" in value or "\\" in value:
+        return True
+    return Path(value).suffix.lower() in _INPUT_EXTENSIONS
+
+
+def classify_input(value: str) -> InputKind:
+    if is_url(value):
+        return InputKind.URL
+    if value.startswith("data:"):
+        payload = extract_base64_payload(value)
+        if not is_base64(payload):
+            raise ValueError("Invalid data URL: payload is not valid Base64.")
+        return InputKind.DATA_URL
+    if is_base64(value):
+        return InputKind.BASE64
+    if looks_like_file_path(value):
+        return InputKind.ABSOLUTE_PATH
+    raise ValueError(f"Invalid input_data. {COMMON_INPUT_DESCRIPTION}")
+
+
+def resolve_absolute_path(value: str) -> Path:
+    path = Path(value).expanduser()
+    if not path.is_absolute():
+        raise ValueError(
+            f"Relative paths are not supported: {value!r}. "
+            "Use an absolute path accessible to the MCP server process."
+        )
+    if not path.exists():
+        raise ValueError(
+            f"File not found: {value!r} (resolved to {path}). "
+            f"{COMMON_INPUT_DESCRIPTION}"
+        )
+    return path.resolve()
+
+
+def validate_external_contract(input_data: str) -> InputKind:
+    value = normalize(input_data)
+    if not value:
+        raise ValueError("input_data must not be empty.")
+
+    kind = classify_input(value)
+
+    if kind is InputKind.ABSOLUTE_PATH:
+        resolve_absolute_path(value)
+        return kind
+
+    if kind in (InputKind.BASE64, InputKind.DATA_URL):
+        payload = extract_base64_payload(value)
+        data = decode_base64_payload(payload)
+        file_type = infer_file_type_from_bytes(data)
+        if file_type not in ("image", "pdf"):
+            raise ValueError(
+                "Unsupported Base64 content type. Only image and PDF inputs are "
+                "supported."
+            )
+
+    return kind
+
+
+def decode_input_bytes(value: str, kind: InputKind) -> bytes:
+    if kind is InputKind.ABSOLUTE_PATH:
+        return resolve_absolute_path(value).read_bytes()
+    payload = extract_base64_payload(value)
+    return decode_base64_payload(payload)
+
+
+@contextmanager
+def materialize_bytes_as_temp_file(
+    data: bytes,
+    *,
+    file_type: str,
+) -> Generator[Path, None, None]:
+    suffix = _SUFFIX_BY_FILE_TYPE.get(file_type, ".bin")
+    temp_file = tempfile.NamedTemporaryFile(delete=False, suffix=suffix)
+    temp_path = Path(temp_file.name)
+    try:
+        temp_file.write(data)
+        temp_file.close()
+        yield temp_path
+    finally:
+        temp_path.unlink(missing_ok=True)
diff --git a/mcp_server/paddleocr_mcp/inference/shared/local_sync_runner.py b/mcp_server/paddleocr_mcp/inference/shared/local_sync_runner.py
new file mode 100644
index 0000000..49aacc6
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/shared/local_sync_runner.py
@@ -0,0 +1,59 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import asyncio
+from queue import Queue
+from threading import Thread
+from typing import Any, Callable
+
+
+class LocalSyncRunner:
+    def __init__(self, inference: Any) -> None:
+        self._inference = inference
+        self._queue: Queue = Queue()
+        self._closed = False
+        self._loop = asyncio.get_running_loop()
+        self._thread = Thread(target=self._worker)
+        self._thread.start()
+
+    @property
+    def inference(self) -> Any:
+        return self._inference
+
+    async def call(self, func: Callable, *args: Any, **kwargs: Any) -> Any:
+        if self._closed:
+            raise RuntimeError("Local sync runner has already been closed")
+        fut = self._loop.create_future()
+        self._queue.put((func, args, kwargs, fut))
+        return await fut
+
+    async def close(self) -> None:
+        if not self._closed:
+            self._queue.put(None)
+            await self._loop.run_in_executor(None, self._thread.join)
+            self._closed = True
+
+    def _worker(self) -> None:
+        while not self._closed:
+            item = self._queue.get()
+            if item is None:
+                break
+            func, args, kwargs, fut = item
+            try:
+                result = func(*args, **kwargs)
+                self._loop.call_soon_threadsafe(fut.set_result, result)
+            except Exception as e:
+                self._loop.call_soon_threadsafe(fut.set_exception, e)
+            finally:
+                self._queue.task_done()
diff --git a/mcp_server/paddleocr_mcp/inference/shared/paddleocr_api_sdk.py b/mcp_server/paddleocr_mcp/inference/shared/paddleocr_api_sdk.py
new file mode 100644
index 0000000..d45c29c
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/shared/paddleocr_api_sdk.py
@@ -0,0 +1,187 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import importlib
+import importlib.machinery
+import importlib.util
+import sys
+from dataclasses import dataclass
+from pathlib import Path
+from types import ModuleType
+from typing import Any, Callable, TypeAlias
+
+
+SDKObject: TypeAlias = Any
+SDKException: TypeAlias = type[BaseException]
+ImportModule: TypeAlias = Callable[[str], ModuleType]
+
+
+@dataclass(frozen=True)
+class PaddleOCRAPISDK:
+    AsyncPaddleOCRClient: SDKObject
+    Model: SDKObject
+    OCROptions: SDKObject
+    PPStructureV3Options: SDKObject
+    PaddleOCRVLOptions: SDKObject
+    APIError: SDKException
+    AuthError: SDKException
+    JobFailedError: SDKException
+    RequestTimeoutError: SDKException
+    ResponseFormatError: SDKException
+    ResultParseError: SDKException
+    ServiceUnavailableError: SDKException
+
+
+_NAMES = (
+    "AsyncPaddleOCRClient",
+    "Model",
+    "OCROptions",
+    "PPStructureV3Options",
+    "PaddleOCRVLOptions",
+    "APIError",
+    "AuthError",
+    "JobFailedError",
+    "RequestTimeoutError",
+    "ResponseFormatError",
+    "ResultParseError",
+    "ServiceUnavailableError",
+)
+
+
+def _sdk_object(module: ModuleType, name: str) -> SDKObject:
+    return getattr(module, name)
+
+
+def _sdk_exception(module: ModuleType, name: str) -> SDKException:
+    value = getattr(module, name)
+    if not isinstance(value, type) or not issubclass(value, BaseException):
+        raise TypeError(f"{module.__name__}.{name} must be an exception class")
+    return value
+
+
+def _build_sdk_from_top_level(paddleocr: ModuleType) -> PaddleOCRAPISDK:
+    missing = [name for name in _NAMES if not hasattr(paddleocr, name)]
+    if missing:
+        raise AttributeError(f"paddleocr is missing API SDK exports: {missing}")
+    return PaddleOCRAPISDK(
+        AsyncPaddleOCRClient=_sdk_object(paddleocr, "AsyncPaddleOCRClient"),
+        Model=_sdk_object(paddleocr, "Model"),
+        OCROptions=_sdk_object(paddleocr, "OCROptions"),
+        PPStructureV3Options=_sdk_object(paddleocr, "PPStructureV3Options"),
+        PaddleOCRVLOptions=_sdk_object(paddleocr, "PaddleOCRVLOptions"),
+        APIError=_sdk_exception(paddleocr, "APIError"),
+        AuthError=_sdk_exception(paddleocr, "AuthError"),
+        JobFailedError=_sdk_exception(paddleocr, "JobFailedError"),
+        RequestTimeoutError=_sdk_exception(paddleocr, "RequestTimeoutError"),
+        ResponseFormatError=_sdk_exception(paddleocr, "ResponseFormatError"),
+        ResultParseError=_sdk_exception(paddleocr, "ResultParseError"),
+        ServiceUnavailableError=_sdk_exception(paddleocr, "ServiceUnavailableError"),
+    )
+
+
+def _build_sdk_from_private_modules(import_module: ImportModule) -> PaddleOCRAPISDK:
+    async_client = import_module("paddleocr._api_client.async_client")
+    models = import_module("paddleocr._api_client.models")
+    errors = import_module("paddleocr._api_client.errors")
+    return PaddleOCRAPISDK(
+        AsyncPaddleOCRClient=_sdk_object(async_client, "AsyncPaddleOCRClient"),
+        Model=_sdk_object(models, "Model"),
+        OCROptions=_sdk_object(models, "OCROptions"),
+        PPStructureV3Options=_sdk_object(models, "PPStructureV3Options"),
+        PaddleOCRVLOptions=_sdk_object(models, "PaddleOCRVLOptions"),
+        APIError=_sdk_exception(errors, "APIError"),
+        AuthError=_sdk_exception(errors, "AuthError"),
+        JobFailedError=_sdk_exception(errors, "JobFailedError"),
+        RequestTimeoutError=_sdk_exception(errors, "RequestTimeoutError"),
+        ResponseFormatError=_sdk_exception(errors, "ResponseFormatError"),
+        ResultParseError=_sdk_exception(errors, "ResultParseError"),
+        ServiceUnavailableError=_sdk_exception(errors, "ServiceUnavailableError"),
+    )
+
+
+def _find_paddleocr_package_dir() -> Path:
+    spec = importlib.machinery.PathFinder.find_spec("paddleocr")
+    if spec is None or spec.submodule_search_locations is None:
+        raise ImportError("Cannot find paddleocr package")
+    return Path(next(iter(spec.submodule_search_locations)))
+
+
+def _register_api_client_package_from_files() -> None:
+    package_dir = _find_paddleocr_package_dir()
+    api_client_dir = package_dir / "_api_client"
+    init_file = api_client_dir / "__init__.py"
+    if not init_file.exists():
+        raise ImportError(f"Cannot find PaddleOCR API client at {init_file}")
+
+    if "paddleocr" not in sys.modules:
+        parent = ModuleType("paddleocr")
+        parent.__path__ = [str(package_dir)]  # type: ignore[attr-defined]
+        parent.__package__ = "paddleocr"
+        parent.__spec__ = importlib.machinery.ModuleSpec(
+            "paddleocr", loader=None, is_package=True
+        )
+        sys.modules["paddleocr"] = parent
+
+    if "paddleocr._api_client" in sys.modules:
+        return
+
+    spec = importlib.util.spec_from_file_location(
+        "paddleocr._api_client",
+        init_file,
+        submodule_search_locations=[str(api_client_dir)],
+    )
+    if spec is None or spec.loader is None:
+        raise ImportError(f"Cannot load PaddleOCR API client from {init_file}")
+
+    module = importlib.util.module_from_spec(spec)
+    sys.modules["paddleocr._api_client"] = module
+    spec.loader.exec_module(module)
+
+
+def _build_sdk_from_private_module_files() -> PaddleOCRAPISDK:
+    _register_api_client_package_from_files()
+    return _build_sdk_from_private_modules(importlib.import_module)
+
+
+def load_paddleocr_api_sdk(
+    import_module: ImportModule = importlib.import_module,
+) -> PaddleOCRAPISDK:
+    try:
+        return _build_sdk_from_top_level(import_module("paddleocr"))
+    except (AttributeError, ImportError):
+        try:
+            return _build_sdk_from_private_modules(import_module)
+        except ImportError:
+            return _build_sdk_from_private_module_files()
+
+
+_SDK = load_paddleocr_api_sdk()
+
+AsyncPaddleOCRClient = _SDK.AsyncPaddleOCRClient
+Model = _SDK.Model
+OCROptions = _SDK.OCROptions
+PPStructureV3Options = _SDK.PPStructureV3Options
+PaddleOCRVLOptions = _SDK.PaddleOCRVLOptions
+APIError = _SDK.APIError
+AuthError = _SDK.AuthError
+JobFailedError = _SDK.JobFailedError
+RequestTimeoutError = _SDK.RequestTimeoutError
+ResponseFormatError = _SDK.ResponseFormatError
+ResultParseError = _SDK.ResultParseError
+ServiceUnavailableError = _SDK.ServiceUnavailableError
+
+from paddleocr._api_client._core import (  # noqa: E402
+    resolve_document_model,
+    resolve_ocr_model,
+)
diff --git a/mcp_server/paddleocr_mcp/inference/shared/param_mapping.py b/mcp_server/paddleocr_mcp/inference/shared/param_mapping.py
new file mode 100644
index 0000000..50ebd6f
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/shared/param_mapping.py
@@ -0,0 +1,26 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Any
+
+
+def snake_to_camel(name: str) -> str:
+    components = name.split("_")
+    return components[0] + "".join(
+        component.capitalize() for component in components[1:]
+    )
+
+
+def convert_params_to_camel(params: dict[str, Any]) -> dict[str, Any]:
+    return {snake_to_camel(key): value for key, value in params.items()}
diff --git a/mcp_server/paddleocr_mcp/inference/types.py b/mcp_server/paddleocr_mcp/inference/types.py
new file mode 100644
index 0000000..6fabe08
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/inference/types.py
@@ -0,0 +1,61 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from dataclasses import asdict, dataclass, field
+from typing import Any, Mapping, Optional, Union
+
+
+@dataclass(frozen=True)
+class InferenceRequest:
+    """Inference request payload."""
+
+    input_data: str
+    file_type: Optional[str] = None
+    runtime_params: Mapping[str, Any] = field(default_factory=dict)
+
+
+@dataclass(frozen=True)
+class TextLine:
+    """A single text line with its bounding box and confidence."""
+
+    text: str
+    confidence: float
+    bbox: Any
+
+
+@dataclass(frozen=True)
+class OCRResult:
+    """OCR result."""
+
+    text: str
+    confidence: float
+    text_lines: list[TextLine]
+
+    def to_dict(self) -> dict[str, Any]:
+        return asdict(self)
+
+
+@dataclass(frozen=True)
+class DocParsingResult:
+    """Document parsing result."""
+
+    markdown: str
+    pages: int
+    images_mapping: dict[str, str]
+
+    def to_dict(self) -> dict[str, Any]:
+        return asdict(self)
+
+
+InferenceResult = Union[OCRResult, DocParsingResult]
diff --git a/mcp_server/paddleocr_mcp/providers.py b/mcp_server/paddleocr_mcp/providers.py
new file mode 100644
index 0000000..9c84d70
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/providers.py
@@ -0,0 +1,88 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from enum import Enum
+
+
+class InferenceProvider(str, Enum):
+    LOCAL = "local"
+    AISTUDIO = "aistudio"
+    QIANFAN = "qianfan"
+    SELF_HOSTED = "self_hosted"
+
+
+class ProviderTransport(str, Enum):
+    LOCAL = "local"
+    AISTUDIO_API = "aistudio_api"
+    HTTP = "http"
+
+
+@dataclass(frozen=True)
+class ProviderSpec:
+    provider: InferenceProvider
+    transport: ProviderTransport
+    display_name: str
+
+
+_PROVIDER_SPECS: dict[InferenceProvider, ProviderSpec] = {
+    InferenceProvider.LOCAL: ProviderSpec(
+        provider=InferenceProvider.LOCAL,
+        transport=ProviderTransport.LOCAL,
+        display_name="local",
+    ),
+    InferenceProvider.AISTUDIO: ProviderSpec(
+        provider=InferenceProvider.AISTUDIO,
+        transport=ProviderTransport.AISTUDIO_API,
+        display_name="AI Studio",
+    ),
+    InferenceProvider.QIANFAN: ProviderSpec(
+        provider=InferenceProvider.QIANFAN,
+        transport=ProviderTransport.HTTP,
+        display_name="Qianfan",
+    ),
+    InferenceProvider.SELF_HOSTED: ProviderSpec(
+        provider=InferenceProvider.SELF_HOSTED,
+        transport=ProviderTransport.HTTP,
+        display_name="self-hosted",
+    ),
+}
+
+
+def normalize_provider(provider: InferenceProvider | str) -> InferenceProvider:
+    if isinstance(provider, InferenceProvider):
+        return provider
+    return InferenceProvider(provider)
+
+
+def get_provider_spec(provider: InferenceProvider | str) -> ProviderSpec:
+    return _PROVIDER_SPECS[normalize_provider(provider)]
+
+
+def provider_choices() -> list[str]:
+    return [provider.value for provider in InferenceProvider]
+
+
+def http_providers() -> set[InferenceProvider]:
+    return {
+        spec.provider
+        for spec in _PROVIDER_SPECS.values()
+        if spec.transport is ProviderTransport.HTTP
+    }
+
+
+def is_http_provider(provider: InferenceProvider | str) -> bool:
+    return get_provider_spec(provider).transport is ProviderTransport.HTTP
diff --git a/mcp_server/paddleocr_mcp/py.typed b/mcp_server/paddleocr_mcp/py.typed
new file mode 100644
index 0000000..e69de29
diff --git a/mcp_server/paddleocr_mcp/selection.py b/mcp_server/paddleocr_mcp/selection.py
new file mode 100644
index 0000000..02a979d
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/selection.py
@@ -0,0 +1,77 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import annotations
+
+from typing import Optional
+
+from .providers import InferenceProvider, normalize_provider
+
+DEFAULT_MODEL = "PP-OCRv6"
+
+QIANFAN_SUPPORTED_MODELS = frozenset(
+    {
+        "PP-StructureV3",
+        "PaddleOCR-VL",
+    }
+)
+
+SUPPORTED_MODELS = frozenset(
+    {
+        "PP-OCRv5",
+        "PP-OCRv5-latin",
+        "PP-OCRv6",
+        "PP-StructureV3",
+        "PaddleOCR-VL",
+        "PaddleOCR-VL-1.5",
+        "PaddleOCR-VL-1.6",
+    }
+)
+
+_MODEL_TOOLS: dict[str, str] = {
+    "PP-OCRv5": "ocr",
+    "PP-OCRv5-latin": "ocr",
+    "PP-OCRv6": "ocr",
+    "PP-StructureV3": "pp_structurev3",
+    "PaddleOCR-VL": "paddleocr_vl",
+    "PaddleOCR-VL-1.5": "paddleocr_vl",
+    "PaddleOCR-VL-1.6": "paddleocr_vl",
+}
+
+
+def tool_for_model(model: str) -> str:
+    """Return the MCP tool name for a validated model."""
+    return _MODEL_TOOLS[model]
+
+
+def resolve_model(model: Optional[str], provider: str) -> str:
+    """Validate and normalize the user-facing model name."""
+    normalized = (model or DEFAULT_MODEL).strip()
+    normalized_provider = normalize_provider(provider)
+    if normalized not in SUPPORTED_MODELS:
+        supported = ", ".join(sorted(SUPPORTED_MODELS))
+        raise ValueError(
+            f"Unsupported model: {normalized!r}. Supported models: {supported}."
+        )
+
+    if (
+        normalized_provider is InferenceProvider.QIANFAN
+        and normalized not in QIANFAN_SUPPORTED_MODELS
+    ):
+        supported = ", ".join(sorted(QIANFAN_SUPPORTED_MODELS))
+        raise ValueError(
+            f"Model {normalized!r} is not supported with qianfan source. "
+            f"Supported models: {supported}."
+        )
+
+    return normalized
diff --git a/mcp_server/paddleocr_mcp/tasks/__init__.py b/mcp_server/paddleocr_mcp/tasks/__init__.py
new file mode 100644
index 0000000..b47b955
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/tasks/__init__.py
@@ -0,0 +1,18 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .base import Task
+from .factory import TaskFactory, create_task
+
+__all__ = ["Task", "TaskFactory", "create_task"]
diff --git a/mcp_server/paddleocr_mcp/tasks/base.py b/mcp_server/paddleocr_mcp/tasks/base.py
new file mode 100644
index 0000000..bcd41dd
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/tasks/base.py
@@ -0,0 +1,106 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import abc
+from typing import Any, List, Optional, Union
+
+from fastmcp import FastMCP
+from fastmcp.server.dependencies import get_context
+from mcp.types import ImageContent, TextContent
+
+from ..inference.base import Inference
+from ..inference.types import InferenceRequest, InferenceResult
+from ..providers import is_http_provider
+
+ToolReturn = Union[str, List[Union[TextContent, ImageContent]]]
+
+
+class Task(abc.ABC):
+    def __init__(self, inference: Inference):
+        self._inference = inference
+
+    @property
+    @abc.abstractmethod
+    def tool_name(self) -> str:
+        pass
+
+    @abc.abstractmethod
+    def _format_result(
+        self, result: InferenceResult, detailed: bool, **kwargs
+    ) -> ToolReturn:
+        pass
+
+    def _tool_description(self) -> str:
+        adapter = self._inference.input_adapter
+        valid_params = sorted(self._inference.get_valid_params())
+        params_hint = ", ".join(valid_params)
+        file_type_hint = ""
+        if is_http_provider(adapter.provider):
+            file_type_hint = (
+                " For HTTP inference providers, set file_type to 'image' or 'pdf' "
+                "when the service cannot infer the type from input_data."
+            )
+        return (
+            f"PaddleOCR {self.tool_name} tool. {adapter.description}"
+            f"{file_type_hint} "
+            f"Optional runtime_params keys: {params_hint}."
+        )
+
+    async def _invoke_tool(
+        self,
+        input_data: str,
+        output_mode: str = "simple",
+        file_type: Optional[str] = None,
+        return_images: bool = True,
+        runtime_params: Optional[dict[str, Any]] = None,
+    ) -> ToolReturn:
+        """Run PaddleOCR on an image or PDF.
+
+        Args:
+            input_data: File input. Format depends on the configured inference
+                provider; see the tool description for supported forms.
+            output_mode: ``simple`` for plain text, ``detailed`` for JSON.
+            file_type: ``image`` or ``pdf`` when required by the HTTP API.
+            return_images: Whether to include images in document parsing output.
+            runtime_params: Optional pipeline parameters as a JSON object.
+        """
+        adapter = self._inference.input_adapter
+        normalized_input = adapter.normalize(input_data)
+        adapter.validate(normalized_input)
+
+        ctx = get_context()
+        await ctx.info(
+            f"--- {self.tool_name} tool received input_data: {normalized_input[:50]} ---"
+        )
+
+        final_params = self._inference.get_final_params(runtime_params or {})
+        self._inference.validate_params(final_params)
+
+        request = InferenceRequest(
+            input_data=normalized_input,
+            file_type=file_type,
+            runtime_params=final_params,
+        )
+        result = await self._inference.predict(request)
+
+        return self._format_result(
+            result,
+            detailed=(output_mode == "detailed"),
+            return_images=return_images,
+        )
+
+    def register_tools(self, mcp: FastMCP) -> None:
+        mcp.tool(self.tool_name, description=self._tool_description())(
+            self._invoke_tool
+        )
diff --git a/mcp_server/paddleocr_mcp/tasks/doc_parsing.py b/mcp_server/paddleocr_mcp/tasks/doc_parsing.py
new file mode 100644
index 0000000..e672c5f
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/tasks/doc_parsing.py
@@ -0,0 +1,119 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import json
+import re
+from typing import Dict, List, Union
+from typing_extensions import override
+
+from mcp.types import ImageContent, TextContent
+
+from ..inference.types import DocParsingResult, InferenceResult
+from .base import Task
+from .mcp_image import normalize_mcp_image_payload
+
+
+class DocParsingTask(Task):
+    def _format_result(
+        self,
+        result: InferenceResult,
+        detailed: bool,
+        return_images: bool = True,
+        **kwargs,
+    ) -> Union[str, List[Union[TextContent, ImageContent]]]:
+        if not isinstance(result, DocParsingResult):
+            raise TypeError(
+                f"DocParsingTask expected DocParsingResult, got {type(result).__name__}"
+            )
+
+        if not result.markdown.strip():
+            return (
+                "No document content detected"
+                if not detailed
+                else json.dumps({"error": "No content detected"}, ensure_ascii=False)
+            )
+
+        markdown_text = result.markdown
+        images_mapping = result.images_mapping
+
+        if return_images and images_mapping:
+            content_list = self._parse_markdown_with_images(
+                markdown_text, images_mapping
+            )
+        else:
+            content_list = [TextContent(type="text", text=markdown_text)]
+
+        if detailed:
+            content_list.append(
+                TextContent(
+                    type="text",
+                    text=f"Pages: {result.pages}",
+                )
+            )
+
+        if len(content_list) == 1:
+            return content_list[0]
+        return content_list
+
+    def _parse_markdown_with_images(
+        self, markdown_text: str, images_mapping: Dict[str, str]
+    ) -> List[Union[TextContent, ImageContent]]:
+        if not images_mapping:
+            return [TextContent(type="text", text=markdown_text)]
+
+        content_list = []
+        img_pattern = r'<img[^>]+src="([^"]+)"[^>]*>'
+        last_pos = 0
+
+        for match in re.finditer(img_pattern, markdown_text):
+            text_before = markdown_text[last_pos : match.start()]
+            if text_before.strip():
+                content_list.append(TextContent(type="text", text=text_before))
+
+            img_src = match.group(1)
+            if img_src in images_mapping:
+                payload = normalize_mcp_image_payload(images_mapping[img_src])
+                if payload is not None:
+                    image_data, mime_type = payload
+                    content_list.append(
+                        ImageContent(
+                            type="image",
+                            data=image_data,
+                            mimeType=mime_type,
+                        )
+                    )
+                else:
+                    content_list.append(TextContent(type="text", text=match.group(0)))
+
+            last_pos = match.end()
+
+        remaining_text = markdown_text[last_pos:]
+        if remaining_text.strip():
+            content_list.append(TextContent(type="text", text=remaining_text))
+
+        return content_list or [TextContent(type="text", text=markdown_text)]
+
+
+class PPStructureV3Task(DocParsingTask):
+    @property
+    @override
+    def tool_name(self) -> str:
+        return "pp_structurev3"
+
+
+class PaddleOCRVLTask(DocParsingTask):
+    @property
+    @override
+    def tool_name(self) -> str:
+        return "paddleocr_vl"
diff --git a/mcp_server/paddleocr_mcp/tasks/factory.py b/mcp_server/paddleocr_mcp/tasks/factory.py
new file mode 100644
index 0000000..846f33e
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/tasks/factory.py
@@ -0,0 +1,53 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Dict
+
+from ..inference.base import Inference
+from ..selection import tool_for_model
+from .base import Task
+
+
+class TaskFactory:
+    _registry: Dict[str, type[Task]] = {}
+
+    @classmethod
+    def register(cls, tool: str, task_class: type[Task]) -> None:
+        cls._registry[tool] = task_class
+
+    @classmethod
+    def create(cls, model: str, inference: Inference) -> Task:
+        tool = tool_for_model(model)
+        if tool not in cls._registry:
+            raise ValueError(
+                f"Unknown tool for model {model!r}: {tool}. "
+                f"Supported: {sorted(cls._registry.keys())}"
+            )
+        task_class = cls._registry[tool]
+        return task_class(inference)
+
+    @classmethod
+    def list_supported(cls) -> set[str]:
+        return set(cls._registry.keys())
+
+
+from .ocr import OCRTask
+from .doc_parsing import PPStructureV3Task, PaddleOCRVLTask
+
+TaskFactory.register("ocr", OCRTask)
+TaskFactory.register("pp_structurev3", PPStructureV3Task)
+TaskFactory.register("paddleocr_vl", PaddleOCRVLTask)
+
+
+def create_task(model: str, inference: Inference) -> Task:
+    return TaskFactory.create(model, inference)
diff --git a/mcp_server/paddleocr_mcp/tasks/mcp_image.py b/mcp_server/paddleocr_mcp/tasks/mcp_image.py
new file mode 100644
index 0000000..7bc9d91
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/tasks/mcp_image.py
@@ -0,0 +1,81 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""MCP ImageContent formatting helpers for tool output."""
+
+from __future__ import annotations
+
+import base64
+from typing import Optional, Tuple
+
+import httpx
+
+from ..utils import (
+    decode_base64_payload,
+    extract_base64_payload,
+    infer_file_type_from_bytes,
+    is_url,
+)
+
+
+def _mime_type_from_image_bytes(data: bytes) -> str:
+    if infer_file_type_from_bytes(data) != "image":
+        return "image/jpeg"
+    import puremagic
+
+    mime = puremagic.from_string(data, mime=True)
+    if mime.startswith("image/"):
+        return mime
+    return "image/jpeg"
+
+
+def normalize_mcp_image_payload(raw: str) -> Optional[Tuple[str, str]]:
+    """Convert provider image payloads to MCP ImageContent (base64 + mimeType)."""
+    value = raw.strip()
+    if not value:
+        return None
+
+    if is_url(value):
+        try:
+            with httpx.Client(timeout=30.0) as client:
+                response = client.get(value)
+                response.raise_for_status()
+            data = response.content
+            return (
+                base64.b64encode(data).decode("ascii"),
+                _mime_type_from_image_bytes(data),
+            )
+        except Exception:
+            return None
+
+    if value.startswith("data:"):
+        try:
+            data = decode_base64_payload(extract_base64_payload(value))
+            mime = "image/jpeg"
+            header = value[5:].split(",", 1)[0]
+            if ";" in header:
+                declared = header.split(";", 1)[0]
+                if declared.startswith("image/"):
+                    mime = declared
+            if mime == "image/jpeg":
+                mime = _mime_type_from_image_bytes(data)
+            return base64.b64encode(data).decode("ascii"), mime
+        except ValueError:
+            return None
+
+    try:
+        data = decode_base64_payload(value)
+        return value, _mime_type_from_image_bytes(data)
+    except ValueError:
+        return None
diff --git a/mcp_server/paddleocr_mcp/tasks/ocr.py b/mcp_server/paddleocr_mcp/tasks/ocr.py
new file mode 100644
index 0000000..5626e08
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/tasks/ocr.py
@@ -0,0 +1,55 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import json
+from typing import List, Union
+from typing_extensions import override
+
+from mcp.types import ImageContent, TextContent
+
+from ..inference.types import InferenceResult, OCRResult
+from .base import Task
+
+
+class OCRTask(Task):
+    @property
+    @override
+    def tool_name(self) -> str:
+        return "ocr"
+
+    @override
+    def _format_result(
+        self, result: InferenceResult, detailed: bool, **kwargs
+    ) -> Union[str, List[Union[TextContent, ImageContent]]]:
+        if not isinstance(result, OCRResult):
+            raise TypeError(f"OCRTask expected OCRResult, got {type(result).__name__}")
+
+        if not result.text.strip():
+            return (
+                "No text detected"
+                if not detailed
+                else json.dumps({"error": "No text detected"}, ensure_ascii=False)
+            )
+
+        if detailed:
+            return json.dumps(result.to_dict(), ensure_ascii=False, indent=2)
+
+        confidence = result.confidence
+        text_line_count = len(result.text_lines)
+
+        output = result.text
+        if confidence > 0:
+            output += f"\n\nConfidence: {(confidence * 100):.1f}% | {text_line_count} text lines"
+
+        return output
diff --git a/mcp_server/paddleocr_mcp/utils.py b/mcp_server/paddleocr_mcp/utils.py
new file mode 100644
index 0000000..619e8a7
--- /dev/null
+++ b/mcp_server/paddleocr_mcp/utils.py
@@ -0,0 +1,64 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Encoding helpers shared by input contract and inference adapters."""
+
+from __future__ import annotations
+
+import base64
+import re
+from typing import Optional
+from urllib.parse import urlparse
+
+
+def is_url(value: str) -> bool:
+    if not (value.startswith("http://") or value.startswith("https://")):
+        return False
+
+    result = urlparse(value)
+    return all([result.scheme, result.netloc]) and result.scheme in ("http", "https")
+
+
+def is_base64(value: str) -> bool:
+    pattern = r"^[A-Za-z0-9+/]+={0,2}$"
+    return bool(re.fullmatch(pattern, value))
+
+
+def extract_base64_payload(input_data: str) -> str:
+    if input_data.startswith("data:"):
+        if "," not in input_data:
+            raise ValueError("Invalid data URL: expected a comma after the MIME type.")
+        return input_data.split(",", 1)[1]
+    return input_data
+
+
+def decode_base64_payload(payload: str) -> bytes:
+    try:
+        return base64.b64decode(payload, validate=True)
+    except Exception as e:
+        raise ValueError(
+            f"Invalid Base64 input: {e}. "
+            "Ensure the string is complete and correctly padded."
+        ) from e
+
+
+def infer_file_type_from_bytes(data: bytes) -> Optional[str]:
+    import puremagic
+
+    mime = puremagic.from_string(data, mime=True)
+    if mime.startswith("image/"):
+        return "image"
+    if mime == "application/pdf":
+        return "pdf"
+    return None
diff --git a/mcp_server/pyproject.toml b/mcp_server/pyproject.toml
new file mode 100644
index 0000000..be33b0c
--- /dev/null
+++ b/mcp_server/pyproject.toml
@@ -0,0 +1,30 @@
+[build-system]
+requires = ["setuptools>=69"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "paddleocr_mcp"
+version = "0.8.5"
+requires-python = ">=3.10"
+dependencies = [
+    "mcp>=1.5.0",
+    "fastmcp>=2.0.0",
+    "httpx>=0.24.0",
+    "numpy>=1.24.0",
+    "paddleocr>=3.7.0",
+    "pillow>=9.0.0",
+    "puremagic>=1.30.0",
+    "typing-extensions>=4.0.0",
+]
+
+[project.optional-dependencies]
+local = [
+    "paddleocr[doc-parser]>=3.7.0",
+]
+local-cpu = [
+    "paddleocr[doc-parser]>=3.7.0",
+    "paddlepaddle>=3.2.1",
+]
+
+[project.scripts]
+paddleocr_mcp = "paddleocr_mcp.__main__:main"
diff --git a/mkdocs-ci.yml b/mkdocs-ci.yml
new file mode 100644
index 0000000..28782e9
--- /dev/null
+++ b/mkdocs-ci.yml
@@ -0,0 +1,10 @@
+INHERIT: mkdocs.yml
+
+strict: true
+
+validation:
+  links:
+    anchors: warn
+    not_found: info
+    absolute_links: info
+    unrecognized_links: info
diff --git a/mkdocs.yml b/mkdocs.yml
new file mode 100644
index 0000000..7ee1601
--- /dev/null
+++ b/mkdocs.yml
@@ -0,0 +1,438 @@
+site_name: PaddleOCR 文档
+site_url: https://paddlepaddle.github.io/PaddleOCR/
+site_author: PaddleOCR PMC
+site_description:
+  Awesome multilingual OCR toolkits based on PaddlePaddle (practical ultra lightweight OCR system, support 80+ languages recognition, provide data annotation and synthesis tools, support training and deployment among server, mobile, embedded and IoT devices)
+
+repo_name: PaddlePaddle/PaddleOCR
+repo_url: https://github.com/PaddlePaddle/PaddleOCR
+
+copyright: Copyright &copy; 2024 Maintained by PaddleOCR PMC.
+
+edit_uri: edit/main/docs/
+
+theme:
+  name: material
+  logo: version3.x/logo.jpg
+  favicon: version3.x/logo.jpg
+  custom_dir: overrides
+  features:
+    - announce.dismiss
+    - content.tooltips
+    - content.code.copy
+    - content.tabs.link
+    - content.footnote.tooltips
+    - content.action.edit
+    - content.action.view
+    - navigation.expand  # 默认打开所有的字节
+    - navigation.sections # 嵌套section渲染为分组标题（不可收起）
+    - navigation.tabs # 顶级索引被作为tab
+    - navigation.tabs.sticky # tab始终可见
+    - navigation.top # 开启顶部导航栏
+    - navigation.tracking # 导航栏跟踪
+    - navigation.footer
+    - navigation.indexes
+    - search.highlight # 搜索高亮
+    - search.share # 搜索分享
+    - search.suggest # 搜索建议
+    - toc.follow # 目录跟踪-页面右侧的小目录
+
+  palette:
+    - media: "(prefers-color-scheme: light)" # 浅色
+      scheme: default
+      primary: indigo
+      accent: indigo
+      toggle:
+        icon: material/brightness-7
+        name: Switch to dark mode
+    - media: "(prefers-color-scheme: dark)" # 深色
+      scheme: slate
+      primary: black
+      accent: indigo
+      toggle:
+        icon: material/brightness-4
+        name: Switch to system preference
+
+  icon:
+    logo: logo
+    previous: fontawesome/solid/angle-left
+    next: fontawesome/solid/angle-right
+    repo: fontawesome/brands/github
+    edit: material/pencil
+    view: material/eye
+    tag:
+      default-tag: fontawesome/solid/tag
+      hardware-tag: fontawesome/solid/microchip
+      software-tag: fontawesome/solid/laptop-code
+    admonition:
+      note: octicons/tag-16
+      abstract: octicons/checklist-16
+      info: octicons/info-16
+      tip: octicons/squirrel-16
+      success: octicons/check-16
+      question: octicons/question-16
+      warning: octicons/alert-16
+      failure: octicons/x-circle-16
+      danger: octicons/zap-16
+      bug: octicons/bug-16
+      example: octicons/beaker-16
+      quote: octicons/quote-16
+
+plugins:
+  - tags
+  - offline
+  - search:
+      separator: '[\s\u200b\-_,:!=\[\: )"`/]+|\.(?!\d)|&[lg]t;|(?!\b)(?=[A-Z][a-z])'
+  - i18n:
+      docs_structure: suffix
+      fallback_to_default: true
+      reconfigure_material: true
+      reconfigure_search: true
+      languages:
+        - locale: zh
+          name: 简体中文
+          default: true
+          build: true
+        - locale: en
+          name: English
+          site_name: PaddleOCR Documentation
+          link: /en/
+          nav_translations:
+            Home: Home
+            安装: Installation
+            快速开始: Quick Start
+            PP-OCRv6: PP-OCRv6
+            使用教程: Usage Tutorial
+            PP-OCRv6简介: PP-OCRv6 Introduction
+            PP-StructureV3: PP-StructureV3
+            PP-StructureV3简介: PP-StructureV3 Introduction
+            PP-ChatOCRv4: PP-ChatOCRv4
+            PP-ChatOCRv4简介: PP-ChatOCRv4 Introduction
+            PaddleOCR-VL: PaddleOCR-VL
+            PaddleOCR-VL-1.5简介: PaddleOCR-VL-1.5 Introduction
+            PaddleOCR-VL-1.6简介: PaddleOCR-VL-1.6 Introduction
+            PaddleOCR-VL简介: PaddleOCR-VL Introduction
+            PaddleOCR-VL NVIDIA Blackwell 架构 GPU 使用教程: PaddleOCR-VL NVIDIA Blackwell-Architecture GPUs Usage Tutorial
+            PaddleOCR-VL 昆仑芯 XPU 使用教程: PaddleOCR-VL Kunlunxin XPU Usage Tutorial
+            PaddleOCR-VL 海光 DCU 使用教程: PaddleOCR-VL Hygon DCU Usage Tutorial
+            PaddleOCR-VL 沐曦 GPU 使用教程: PaddleOCR-VL MetaX GPU Usage Tutorial
+            PaddleOCR-VL 天数 GPU 使用教程: PaddleOCR-VL Iluvatar GPU Usage Tutorial
+            PaddleOCR-VL 华为昇腾 NPU 使用教程: PaddleOCR-VL Huawei Ascend NPU Usage Tutorial
+            PaddleOCR-VL Apple Silicon 使用教程: PaddleOCR-VL Apple Silicon Usage Tutorial
+            PaddleOCR-VL AMD GPU 使用教程: PaddleOCR-VL AMD GPU Usage Tutorial
+            PaddleOCR-VL Intel Arc GPU 使用教程: PaddleOCR-VL Intel Arc GPU Usage Tutorial
+            推理与部署: Inference and Deployment
+            总览: Overview
+            本地推理: Local Inference
+            服务化: Serving
+            自部署服务化: Self-hosted Serving
+            PaddleOCR 官方 API: PaddleOCR official API
+            Python SDK: Python SDK
+            TypeScript SDK: TypeScript SDK
+            Go SDK: Go SDK
+            CLI: CLI
+            跨端部署: Cross-Platform Deployment
+            其它: Other
+            生态集成: Ecosystem Integrations
+            高性能推理: High-Performance Inference
+            打包 PaddleOCR 项目: Package PaddleOCR Projects
+            获取ONNX模型: Obtaining ONNX Models
+            Android 部署: Android Deployment
+            iOS 部署: iOS Deployment
+            服务化部署: Serving
+            浏览器端部署: Browser Deployment
+            模块列表: Module List
+            文档图像方向分类模块: Document Image Orientation Classification Module
+            文档类视觉语言模型模块: Document Visual Language Model Module
+            公式识别模块: Formula Recognition Module
+            版面区域检测模块: Layout Detection
+            版面分析模块: Layout Analysis
+            印章文本检测模块: Seal Text Detection
+            表格单元格检测模块: Table Cell Detection Module
+            表格分类模块: Table Classification Module
+            表格结构识别模块: Table Structure Recognition Module
+            文本检测模块: Text Detection Module
+            文本图像矫正模块: Text Image Rectification Module
+            文本行方向分类模块: Text Line Orientation Classification Module
+            文本识别模块: Text Recognition Module
+            产线列表: Pipeline List
+            公式识别产线: Formula Recognition Pipeline
+            文档图像预处理产线: Document Image Preprocessing Pipeline
+            文档理解产线: Document Understanding Pipeline
+            印章文本识别产线: Seal Text Recognition Pipeline
+            通用表格识别v2产线: General Table Recognition v2 Pipeline
+            图表解析模块: Chart Parsing Module
+            PaddleOCR-VL产线: PaddleOCR-VL Pipeline
+            PP-DocTranslation产线: PP-DocTranslation Pipeline
+            推理引擎与配置说明: Inference Engine and Configuration
+            数据标注与合成&数据集: Data Annotation and Synthesis & Datasets
+            数据标注: Data Annotation
+            X-AnyLabeling 文档解析: Document Parsing with X-AnyLabeling
+            其它数据标注工具: Other Data Annotation Tools
+            数据合成: Data Synthesis
+            数据合成工具: Data Synthesis Tools
+            数据集: Datasets
+            多硬件使用: Multi-Devices Usage
+            PaddleOCR 多硬件使用指南: PaddleOCR Multi-Devices Usage Guide
+            昇腾 NPU 飞桨安装教程: Ascend NPU PaddlePaddle Installation Tutorial
+            昆仑 XPU 飞桨安装教程: Kunlun XPU PaddlePaddle Installation Tutorial
+            其他说明: Other Notes
+            PaddleOCR 与 PaddleX: PaddleOCR and PaddleX
+            PaddleOCR 3.x 升级说明: PaddleOCR 3.x Upgrade Notes
+            低代码全流程开发: All-in-One Development
+            概述: Overview
+            通用中英文OCR数据集: General Chinese and English OCR dataset
+            手写中文OCR数据集: Handwritten Chinese OCR Dataset
+            垂类多语言OCR数据集: Vertical multi-language OCR dataset
+            版面分析数据集: Layout Analysis Dataset
+            表格识别数据集: Table recognition dataset
+            关键信息提取数据集: Key Information Extraction Dataset
+            近期更新: Recently Update
+            版本2.x: Version 2.x
+            模型列表: Model List
+            基于Python预测引擎推理: Python Inference
+            基于C++预测引擎推理: CPP Inference
+            Visual Studio 2019 Community CMake 编译指南: Visual Studio 2019 Community CMake Compilation Guide
+            服务化部署: Server Deployment
+            Android部署: Android Deployment
+            Paddle2ONNX模型转化与预测: Paddle2ONNX
+            云上飞桨部署工具: Paddle Cloud
+            Benchmark: Benchmark
+            博客: Blog
+            常见问题: FAQ
+            社区: Community
+            社区贡献: Community Contribution
+            附录: Appendix
+            配置 paddleocr 包日志系统: Configure logging for the paddleocr Python package
+            模块概述: Module Overview
+            产线概述: Pipeline Overview
+            通用 OCR 产线 C++ 本地部署 - Linux: C++ Local Deployment for General OCR Pipeline - Linux
+            通用 OCR 产线 C++ 本地部署 - Windows: C++ Local Deployment for General OCR Pipeline - Windows
+            C++ 本地部署: C++ Local Deployment
+            MCP 服务器: MCP Server
+            Agent Skills: Agent Skills
+            产线并行推理: Parallel Inference for Pipelines
+        # - locale: ja
+        #   name: 日本語
+        #   site_name: PaddleOCR ドキュメント
+        #   link: /ja/
+        #   nav_translations:
+        #     Home: トップページ
+        # - locale: ru
+        #   name: Pу́сский язы́к
+        #   site_name: Документация PaddleOCR
+        #   link: /ru/
+        #   nav_translations:
+        #     Home: Главная страница
+        # - locale: hi
+        #   name: हिन्दी
+        #   site_name: पैडलओसीआर दस्तावेज़ीकरण
+        #   link: /hi/
+        #   nav_translations:
+        #     Home: घर पृष्ठ
+        # - locale: ko
+        #   name: 한국인
+        #   site_name: PaddleOCR 문서
+        #   link: /ko/
+        #   nav_translations:
+        #     Home: 첫 페이지
+        # - locale: "null"
+        #   name: Help translating
+        #   build: false
+        #   fixed_link: "https://github.com/PaddlePaddle/PaddleOCR/discussions/13374"
+  - git-committers:
+      enabled: !ENV [ENABLE_GIT_PLUGINS, true]
+      repository: PaddlePaddle/PaddleOCR
+      branch: main
+      token: !!python/object/apply:os.getenv ["MKDOCS_GIT_COMMITTERS_APIKEY"]
+  - git-revision-date-localized:
+      enabled: !ENV [ENABLE_GIT_PLUGINS, true]
+      enable_creation_date: true
+      timezone: Asia/Shanghai
+      type: iso_datetime
+      exclude:
+        - "index*.md"
+        - "quick_start*.md"
+        - "version3.x/**"
+
+markdown_extensions:
+  - abbr
+  - attr_list
+  - callouts
+  - pymdownx.snippets
+  - pymdownx.critic
+  - pymdownx.caret
+  - pymdownx.keys
+  - pymdownx.mark
+  - pymdownx.tilde
+  - footnotes
+  - def_list
+  - md_in_html
+  - pymdownx.tasklist:
+      custom_checkbox: true
+  - toc:
+      permalink: true
+  - pymdownx.superfences:
+      custom_fences:
+        - name: mermaid
+          class: mermaid
+          format: !!python/name:pymdownx.superfences.fence_code_format
+  - pymdownx.emoji:
+      emoji_index: !!python/name:material.extensions.emoji.twemoji
+      emoji_generator: !!python/name:material.extensions.emoji.to_svg
+  - pymdownx.highlight:
+        anchor_linenums: true
+        line_spans: __span
+        pygments_lang_class: true
+  - pymdownx.inlinehilite
+  - pymdownx.tabbed:
+      alternate_style: true
+  - admonition
+  - pymdownx.details
+  - pymdownx.superfences
+  - pymdownx.arithmatex:
+      generic: true
+
+hooks:
+  - overrides/hooks/expiry.py
+
+extra:
+  social:
+    - icon: fontawesome/brands/github
+      link: https://github.com/PaddlePaddle/PaddleOCR
+    - icon: fontawesome/brands/python
+      link: https://pypi.org/project/paddleocr/
+  version:
+    provider: mike
+  expiry_days: 365
+
+extra_javascript:
+  - javascripts/katex.js
+  - https://unpkg.com/katex@0/dist/katex.min.js
+  - https://unpkg.com/katex@0/dist/contrib/auto-render.min.js
+
+extra_css:
+  - https://unpkg.com/katex@0/dist/katex.min.css
+
+nav:
+  - Home: index.md
+  - 安装: version3.x/installation.md
+  - 快速开始: quick_start.md
+  - PP-OCRv6:
+    - 使用教程: version3.x/pipeline_usage/OCR.md
+    - PP-OCRv6简介: version3.x/algorithm/PP-OCRv6/PP-OCRv6.md
+  - PP-StructureV3:
+    - 使用教程: version3.x/pipeline_usage/PP-StructureV3.md
+    - PP-StructureV3简介: version3.x/algorithm/PP-StructureV3/PP-StructureV3.md
+  - PP-ChatOCRv4:
+    - 使用教程: version3.x/pipeline_usage/PP-ChatOCRv4.md
+    - PP-ChatOCRv4简介: version3.x/algorithm/PP-ChatOCRv4/PP-ChatOCRv4.md
+  - PaddleOCR-VL:
+    - 使用教程: version3.x/pipeline_usage/PaddleOCR-VL.md
+    - PaddleOCR-VL-1.5简介: version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL-1.5.md
+    - PaddleOCR-VL-1.6简介: version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL-1.6.md
+    - PaddleOCR-VL简介: version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL.md
+    - PaddleOCR-VL NVIDIA Blackwell 架构 GPU 使用教程: version3.x/pipeline_usage/PaddleOCR-VL-NVIDIA-Blackwell.md
+    - PaddleOCR-VL 昆仑芯 XPU 使用教程: version3.x/pipeline_usage/PaddleOCR-VL-Kunlunxin-XPU.md
+    - PaddleOCR-VL 海光 DCU 使用教程: version3.x/pipeline_usage/PaddleOCR-VL-Hygon-DCU.md
+    - PaddleOCR-VL 沐曦 GPU 使用教程: version3.x/pipeline_usage/PaddleOCR-VL-MetaX-GPU.md
+    - PaddleOCR-VL 天数 GPU 使用教程: version3.x/pipeline_usage/PaddleOCR-VL-Iluvatar-GPU.md
+    - PaddleOCR-VL 华为昇腾 NPU 使用教程: version3.x/pipeline_usage/PaddleOCR-VL-Huawei-Ascend-NPU.md
+    - PaddleOCR-VL Apple Silicon 使用教程: version3.x/pipeline_usage/PaddleOCR-VL-Apple-Silicon.md
+    - PaddleOCR-VL AMD GPU 使用教程: version3.x/pipeline_usage/PaddleOCR-VL-AMD-GPU.md
+    - PaddleOCR-VL Intel Arc GPU 使用教程: version3.x/pipeline_usage/PaddleOCR-VL-Intel-Arc-GPU.md
+  - 推理与部署:
+    - 本地推理:
+      - 高性能推理: version3.x/inference_deployment/local_inference/high_performance_inference.md
+      - 推理引擎与配置说明: version3.x/inference_deployment/local_inference/inference_engine.md
+      - 产线并行推理: version3.x/inference_deployment/local_inference/parallel_inference.md
+      - C++ 本地部署:
+        - 概述: version3.x/inference_deployment/local_inference/cpp/cpp_local_deployment.md
+        - 通用 OCR 产线 C++ 本地部署 - Linux: version3.x/inference_deployment/local_inference/cpp/OCR.md
+        - 通用 OCR 产线 C++ 本地部署 - Windows: version3.x/inference_deployment/local_inference/cpp/OCR_windows.md
+    - 服务化:
+      - 自部署服务化: version3.x/inference_deployment/serving/serving.md
+      - PaddleOCR 官方 API:
+        - 总览: version3.x/inference_deployment/serving/paddleocr_official_api/overview.md
+        - Python SDK: version3.x/inference_deployment/serving/paddleocr_official_api/python.md
+        - TypeScript SDK: version3.x/inference_deployment/serving/paddleocr_official_api/typescript.md
+        - Go SDK: version3.x/inference_deployment/serving/paddleocr_official_api/go.md
+        - CLI: version3.x/inference_deployment/serving/paddleocr_official_api/cli.md
+    - 跨端部署:
+      - Android 部署: version3.x/inference_deployment/cross_platform/android_deployment.md
+      - iOS 部署: version3.x/inference_deployment/cross_platform/ios_deployment.md
+      - 浏览器端部署: version3.x/inference_deployment/cross_platform/browser.md
+    - 其它:
+      - 获取ONNX模型: version3.x/inference_deployment/others/obtaining_onnx_models.md
+      - 打包 PaddleOCR 项目: version3.x/inference_deployment/others/packaging.md
+      - Benchmark: version3.x/inference_deployment/others/benchmark.md
+  - 生态集成:
+    - MCP 服务器: version3.x/integrations/mcp_server.md
+    - Agent Skills: version3.x/integrations/skills.md
+
+  - 模块列表:
+    - 模块概述: version3.x/module_usage/module_overview.md
+    - 文档图像方向分类模块: version3.x/module_usage/doc_img_orientation_classification.md
+    - 文档类视觉语言模型模块: version3.x/module_usage/doc_vlm.md
+    - 公式识别模块: version3.x/module_usage/formula_recognition.md
+    - 版面区域检测模块: version3.x/module_usage/layout_detection.md
+    - 版面分析模块: version3.x/module_usage/layout_analysis.md
+    - 印章文本检测模块: version3.x/module_usage/seal_text_detection.md
+    - 表格单元格检测模块: version3.x/module_usage/table_cells_detection.md
+    - 表格分类模块: version3.x/module_usage/table_classification.md
+    - 表格结构识别模块: version3.x/module_usage/table_structure_recognition.md
+    - 文本检测模块: version3.x/module_usage/text_detection.md
+    - 文本图像矫正模块: version3.x/module_usage/text_image_unwarping.md
+    - 文本行方向分类模块: version3.x/module_usage/textline_orientation_classification.md
+    - 文本识别模块: version3.x/module_usage/text_recognition.md
+    - 图表解析模块: version3.x/module_usage/chart_parsing.md
+  - 产线列表:
+    - 产线概述: version3.x/pipeline_usage/pipeline_overview.md
+    - 公式识别产线: version3.x/pipeline_usage/formula_recognition.md
+    - 文档图像预处理产线: version3.x/pipeline_usage/doc_preprocessor.md
+    - 文档理解产线: version3.x/pipeline_usage/doc_understanding.md
+    - 印章文本识别产线: version3.x/pipeline_usage/seal_recognition.md
+    - 通用表格识别v2产线: version3.x/pipeline_usage/table_recognition_v2.md
+    - PP-DocTranslation产线: version3.x/pipeline_usage/PP-DocTranslation.md
+  - 数据标注与合成&数据集:
+    - 数据标注:
+      - X-AnyLabeling 文档解析: data_anno_synth/x_anylabeling.md
+      - 其它数据标注工具: data_anno_synth/data_annotation.md
+    - 数据合成:
+      - 数据合成工具: data_anno_synth/data_synthesis.md
+    - 数据集:
+      - 通用中英文OCR数据集: datasets/datasets.md
+      - 手写中文OCR数据集: datasets/handwritten_datasets.md
+      - 垂类多语言OCR数据集: datasets/vertical_and_multilingual_datasets.md
+      - 版面分析数据集: datasets/layout_datasets.md
+      - 表格识别数据集: datasets/table_datasets.md
+      - 关键信息提取数据集: datasets/kie_datasets.md
+  - 多硬件使用:
+    - PaddleOCR 多硬件使用指南: version3.x/other_devices_support/multi_devices_use_guide.md
+    - 昇腾 NPU 飞桨安装教程: version3.x/other_devices_support/paddlepaddle_install_NPU.md
+    - 昆仑 XPU 飞桨安装教程: version3.x/other_devices_support/paddlepaddle_install_XPU.md
+  - 其他说明:
+    - PaddleOCR 与 PaddleX: version3.x/paddleocr_and_paddlex.md
+    - PaddleOCR 3.x 升级说明: update/upgrade_notes.md
+    - 配置 paddleocr 包日志系统: version3.x/logging.md
+
+  - 低代码全流程开发:
+      - 概述: version3.x/paddlex/overview.md
+      - 快速开始: version3.x/paddlex/quick_start.md
+  - 近期更新: update/update.md
+  - version2.x:
+    - 概述: version2.x/legacy/index.md
+    - 模型列表: version2.x/legacy/model_list_2.x.md
+    - 基于Python预测引擎推理: version2.x/legacy/python_infer.md
+    - 基于C++预测引擎推理: version2.x/legacy/cpp_infer.md
+    - Visual Studio 2019 Community CMake 编译指南: version2.x/legacy/windows_vs2019_build.md
+    - 服务化部署: version2.x/legacy/paddle_server.md
+    - Android部署: version2.x/legacy/android_demo.md
+    - 端侧部署: version2.x/legacy/lite.md
+    - Paddle2ONNX模型转化与预测: version2.x/legacy/paddle2onnx.md
+    - 云上飞桨部署工具: version2.x/legacy/paddle_cloud.md
+    - Benchmark: version2.x/legacy/benchmark.md
+  - FAQ: FAQ.md
+  - 社区:
+    - 社区贡献: community/community_contribution.md
+    - 附录: community/code_and_doc.md
diff --git a/overrides/hooks/expiry.py b/overrides/hooks/expiry.py
new file mode 100644
index 0000000..f3dd749
--- /dev/null
+++ b/overrides/hooks/expiry.py
@@ -0,0 +1,34 @@
+import re
+from datetime import datetime
+
+
+def on_page_context(context, page, config, nav):
+    expiry_days = config.get("extra", {}).get("expiry_days", 365)
+
+    def compute_expiry(meta):
+        revision = (
+            meta.get("git_revision_date_localized")
+            or meta.get("git_creation_date_localized")
+            or meta.get("revision_date")
+        )
+        is_expired = False
+        last_update = None
+        if revision:
+            m = re.search(r"(\d{4}-\d{2}-\d{2})", str(revision))
+            if m:
+                last_update = m.group(1)
+                try:
+                    dt = datetime.strptime(last_update, "%Y-%m-%d")
+                    if (datetime.now() - dt).days > expiry_days:
+                        is_expired = True
+                except Exception:
+                    # 无法解析日期时，保持不显示过期提示
+                    pass
+        return is_expired, last_update
+
+    page.is_expired, page.last_update = compute_expiry(page.meta)
+    context["is_expired"] = page.is_expired
+    context["last_update"] = page.last_update
+    context["expiry_days"] = expiry_days
+
+    return context
diff --git a/overrides/main.html b/overrides/main.html
new file mode 100644
index 0000000..033fa3c
--- /dev/null
+++ b/overrides/main.html
@@ -0,0 +1,8 @@
+{% extends "base.html" %}
+
+{% block outdated %}
+You're not viewing the latest version.
+<a href="{{ '../' ~ base_url }}">
+    <strong>Click here to go to latest.</strong>
+</a>
+{% endblock %}
diff --git a/overrides/partials/comments.html b/overrides/partials/comments.html
new file mode 100644
index 0000000..8907194
--- /dev/null
+++ b/overrides/partials/comments.html
@@ -0,0 +1,46 @@
+{% if page.meta.comments %}
+<h2 id="__comments">{{ lang.t("meta.comments") }}</h2>
+<!-- Insert generated snippet here -->
+
+<script src="https://giscus.app/client.js" data-repo="PaddlePaddle/PaddleOCR"
+    data-repo-id="MDEwOlJlcG9zaXRvcnkyNjIyOTYxMjI=" data-category="Q&A" data-category-id="DIC_kwDOD6JSOs4COrbO"
+    data-mapping="pathname" data-strict="0" data-reactions-enabled="1" data-emit-metadata="0" data-input-position="top"
+    data-theme="preferred_color_scheme" data-lang="en" data-loading="lazy" crossorigin="anonymous" async>
+</script>
+
+<!-- Synchronize Giscus theme with palette -->
+<script>
+    var giscus = document.querySelector("script[src*=giscus]")
+
+    // Set palette on initial load
+    var palette = __md_get("__palette")
+    if (palette && typeof palette.color === "object") {
+        var theme = palette.color.scheme === "slate"
+            ? "transparent_dark"
+            : "light"
+
+        // Instruct Giscus to set theme
+        giscus.setAttribute("data-theme", theme)
+    }
+
+    // Register event handlers after documented loaded
+    document.addEventListener("DOMContentLoaded", function () {
+        var ref = document.querySelector("[data-md-component=palette]")
+        ref.addEventListener("change", function () {
+            var palette = __md_get("__palette")
+            if (palette && typeof palette.color === "object") {
+                var theme = palette.color.scheme === "slate"
+                    ? "transparent_dark"
+                    : "light"
+
+                // Instruct Giscus to change theme
+                var frame = document.querySelector(".giscus-frame")
+                frame.contentWindow.postMessage(
+                    { giscus: { setConfig: { theme } } },
+                    "https://giscus.app"
+                )
+            }
+        })
+    })
+</script>
+{% endif %}
diff --git a/overrides/partials/content.html b/overrides/partials/content.html
new file mode 100644
index 0000000..2735741
--- /dev/null
+++ b/overrides/partials/content.html
@@ -0,0 +1,49 @@
+<!--
+  Copyright (c) 2016-2025 Martin Donath <martin.donath@squidfunk.com>
+
+  Permission is hereby granted, free of charge, to any person obtaining a copy
+  of this software and associated documentation files (the "Software"), to
+  deal in the Software without restriction, including without limitation the
+  rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+  sell copies of the Software, and to permit persons to whom the Software is
+  furnished to do so, subject to the following conditions:
+
+  The above copyright notice and this permission notice shall be included in
+  all copies or substantial portions of the Software.
+
+  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+  IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+  FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL THE
+  AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+  LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+  FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+  IN THE SOFTWARE.
+-->
+
+<!-- Tags -->
+{% include "partials/tags.html" %}
+
+<!-- Actions -->
+{% include "partials/actions.html" %}
+
+<!--
+  Hack: check whether the content contains a h1 headline. If it doesn't, the
+  page title (or respectively site name) is used as the main headline.
+-->
+{% if "\u003ch1" not in page.content %}
+  <h1>{{ page.title | d(config.site_name, true)}}</h1>
+{% endif %}
+
+{% include "partials/expired_notice.html" %}
+
+<!-- Page content -->
+{{ page.content }}
+
+<!-- Source file information -->
+{% include "partials/source-file.html" %}
+
+<!-- Was this page helpful? -->
+{% include "partials/feedback.html" %}
+
+<!-- Comment system -->
+{% include "partials/comments.html" %}
diff --git a/overrides/partials/expired_notice.html b/overrides/partials/expired_notice.html
new file mode 100644
index 0000000..f977cb1
--- /dev/null
+++ b/overrides/partials/expired_notice.html
@@ -0,0 +1,9 @@
+{% if is_expired %}
+    <div class="admonition warning">
+        <p class="admonition-title">Warning</p>
+        <p>
+            本文档最后更新于 {{ last_update }}，
+            已超过 {{ expiry_days }} 天未更新，内容可能已经过时，阅读注意甄别。
+        </p>
+    </div>
+{% endif %}
diff --git a/paddleocr-js/.gitignore b/paddleocr-js/.gitignore
new file mode 100644
index 0000000..7e611c6
--- /dev/null
+++ b/paddleocr-js/.gitignore
@@ -0,0 +1,28 @@
+# OS / editor
+.DS_Store
+.idea/
+.vscode/
+
+# Dependencies
+node_modules/
+
+# Build output
+dist/
+build/
+coverage/
+.cache/
+
+# Tool caches
+.eslintcache
+*.tsbuildinfo
+
+# Logs
+*.log
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+pnpm-debug.log*
+
+# Local env files
+.env
+.env.*
diff --git a/paddleocr-js/.prettierignore b/paddleocr-js/.prettierignore
new file mode 100644
index 0000000..1c1b803
--- /dev/null
+++ b/paddleocr-js/.prettierignore
@@ -0,0 +1,5 @@
+node_modules/
+dist/
+build/
+coverage/
+.cache/
diff --git a/paddleocr-js/.prettierrc.json b/paddleocr-js/.prettierrc.json
new file mode 100644
index 0000000..2a9870e
--- /dev/null
+++ b/paddleocr-js/.prettierrc.json
@@ -0,0 +1,6 @@
+{
+  "semi": true,
+  "singleQuote": false,
+  "printWidth": 100,
+  "trailingComma": "none"
+}
diff --git a/paddleocr-js/README.md b/paddleocr-js/README.md
new file mode 100644
index 0000000..758e09c
--- /dev/null
+++ b/paddleocr-js/README.md
@@ -0,0 +1,49 @@
+# PaddleOCR.js
+
+English | [简体中文](README_cn.md)
+
+Official browser OCR SDK and demo for PaddleOCR.
+
+## Project structure
+
+| Path             | Role                                                                   |
+| ---------------- | ---------------------------------------------------------------------- |
+| `packages/core/` | Browser SDK sources; published to npm as **`@paddleocr/paddleocr-js`** |
+| `apps/demo/`     | Vite demo app that depends on the SDK                                  |
+
+## Local development and demo
+
+```bash
+npm install
+npm run dev:demo
+```
+
+Other common commands:
+
+```bash
+npm run build
+npm run test
+npm run typecheck
+npm run check
+```
+
+Workspace-scoped examples:
+
+```bash
+npm run build --workspace packages/core
+npm run build --workspace apps/demo
+```
+
+## Documentation
+
+| Topic | Link |
+|-------|------|
+| Architecture | [architecture.md](docs/architecture.md) |
+| Development | [development.md](docs/development.md) |
+| Monorepo conventions | [monorepo.md](docs/monorepo.md) |
+| SDK package README | [packages/core/README.md](packages/core/README.md) |
+
+## Acknowledgements
+
+- [ONNX Runtime Web](https://onnxruntime.ai/) for browser-side ONNX inference
+- [OpenCV.js](https://opencv.org/) for browser-side image processing
diff --git a/paddleocr-js/README_cn.md b/paddleocr-js/README_cn.md
new file mode 100644
index 0000000..84ba5e3
--- /dev/null
+++ b/paddleocr-js/README_cn.md
@@ -0,0 +1,49 @@
+# PaddleOCR.js
+
+[English](README.md) | 简体中文
+
+PaddleOCR 官方浏览器 OCR SDK 与演示应用。
+
+## 项目结构
+
+| 路径             | 作用                                                                 |
+| ---------------- | -------------------------------------------------------------------- |
+| `packages/core/` | 浏览器 SDK 源码；发布到 npm 时的包名为 **`@paddleocr/paddleocr-js`** |
+| `apps/demo/`     | 依赖该 SDK 的 Vite 演示应用                                          |
+
+## 本地开发与演示
+
+```bash
+npm install
+npm run dev:demo
+```
+
+其他常用命令：
+
+```bash
+npm run build
+npm run test
+npm run typecheck
+npm run check
+```
+
+按 workspace 作用域：
+
+```bash
+npm run build --workspace packages/core
+npm run build --workspace apps/demo
+```
+
+## 文档
+
+| 说明 | 链接 |
+|------|------|
+| 架构说明 | [architecture_cn.md](docs/architecture_cn.md) |
+| 开发指南 | [development_cn.md](docs/development_cn.md) |
+| Monorepo 约定 | [monorepo_cn.md](docs/monorepo_cn.md) |
+| SDK 包 README | [packages/core/README_cn.md](packages/core/README_cn.md) |
+
+## 致谢
+
+- [ONNX Runtime Web](https://onnxruntime.ai/)：提供浏览器端 ONNX 推理能力
+- [OpenCV.js](https://opencv.org/)：提供浏览器端图像处理能力
diff --git a/paddleocr-js/apps/demo/index.html b/paddleocr-js/apps/demo/index.html
new file mode 100644
index 0000000..c91b9fa
--- /dev/null
+++ b/paddleocr-js/apps/demo/index.html
@@ -0,0 +1,228 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="utf-8" />
+    <meta name="viewport" content="width=device-width,initial-scale=1" />
+    <title>PaddleOCR.js</title>
+    <style>
+      :root {
+        color-scheme: light dark;
+        font-family: "Helvetica Neue", Helvetica, Arial, sans-serif;
+      }
+
+      body {
+        margin: 0;
+        padding: 20px;
+        max-width: 1440px;
+        margin-inline: auto;
+        background: color-mix(in srgb, Canvas 96%, #2563eb 4%);
+      }
+
+      h1,
+      h2,
+      h3,
+      p {
+        margin: 0;
+      }
+
+      .layout {
+        display: grid;
+        gap: 16px;
+      }
+
+      .hero {
+        display: grid;
+        gap: 8px;
+      }
+
+      .hero p {
+        max-width: 880px;
+        opacity: 0.78;
+      }
+
+      .panel {
+        border: 1px solid color-mix(in srgb, CanvasText 12%, transparent);
+        border-radius: 16px;
+        background: color-mix(in srgb, Canvas 93%, #fff 7%);
+        padding: 16px;
+        box-shadow: 0 10px 30px #00000010;
+      }
+
+      .panel-grid {
+        display: grid;
+        grid-template-columns: repeat(auto-fit, minmax(220px, 1fr));
+        gap: 12px;
+      }
+
+      label {
+        display: grid;
+        gap: 6px;
+        font-size: 14px;
+      }
+
+      input,
+      select,
+      button,
+      textarea {
+        border-radius: 10px;
+        border: 1px solid #6665;
+        padding: 9px 12px;
+        font: inherit;
+        background: color-mix(in srgb, Canvas 94%, #000 6%);
+      }
+
+      textarea {
+        min-height: 88px;
+        resize: vertical;
+      }
+
+      button {
+        cursor: pointer;
+        background: color-mix(in srgb, Canvas 86%, #2563eb 14%);
+      }
+
+      .button-row {
+        display: flex;
+        flex-wrap: wrap;
+        gap: 8px;
+      }
+
+      #status {
+        min-height: 1.4em;
+        padding: 10px 12px;
+        border-radius: 12px;
+        border: 1px solid color-mix(in srgb, CanvasText 10%, transparent);
+        background: color-mix(in srgb, Canvas 94%, #22c55e 6%);
+      }
+
+      .content-grid {
+        display: grid;
+        grid-template-columns: minmax(0, 1.5fr) minmax(320px, 420px);
+        gap: 16px;
+        align-items: start;
+      }
+
+      canvas {
+        display: none;
+      }
+
+      img#vizImage {
+        width: 100%;
+        height: auto;
+        display: block;
+      }
+
+      img#vizImage[hidden] {
+        display: none;
+      }
+
+      pre,
+      code {
+        font-family: ui-monospace, SFMono-Regular, Menlo, monospace;
+      }
+
+      #metrics {
+        white-space: pre-wrap;
+        line-height: 1.55;
+        font-size: 13px;
+      }
+
+      #results {
+        margin: 0;
+        padding-left: 18px;
+        display: grid;
+        gap: 8px;
+      }
+
+      .muted {
+        opacity: 0.72;
+        font-size: 13px;
+      }
+
+      @media (max-width: 980px) {
+        .content-grid {
+          grid-template-columns: 1fr;
+        }
+      }
+    </style>
+  </head>
+  <body>
+    <main class="layout">
+      <header class="hero">
+        <h1>PaddleOCR.js</h1>
+      </header>
+
+      <section class="panel">
+        <div class="panel-grid">
+          <label>
+            Model
+            <select id="modelPreset">
+              <option value="PP-OCRv5_mobile">PP-OCRv5_mobile</option>
+              <option value="PP-OCRv6_small">PP-OCRv6_small</option>
+              <option value="PP-OCRv6_tiny">PP-OCRv6_tiny</option>
+            </select>
+          </label>
+          <label>
+            Runtime backend
+            <select id="runtimeBackend">
+              <option value="auto">auto</option>
+              <option value="webgpu">webgpu</option>
+              <option value="wasm">wasm</option>
+            </select>
+          </label>
+          <label>
+            Detection threshold
+            <input id="detThresh" type="number" step="0.05" min="0" max="1" value="0.3" />
+          </label>
+          <label>
+            Box threshold
+            <input id="boxThresh" type="number" step="0.05" min="0" max="1" value="0.6" />
+          </label>
+          <label>
+            Unclip ratio
+            <input id="unclipRatio" type="number" step="0.1" min="0" value="1.5" />
+          </label>
+          <label>
+            Recognition threshold
+            <input id="recScoreThresh" type="number" step="0.05" min="0" max="1" value="0.1" />
+          </label>
+        </div>
+        <div class="button-row" style="margin-top: 12px">
+          <input id="imageInput" type="file" accept="image/*" hidden />
+          <button id="chooseImageBtn">Choose image</button>
+          <button id="reinitializeBtn">Reinitialize</button>
+          <button id="runBtn" disabled>Run OCR</button>
+        </div>
+      </section>
+
+      <div id="status">Initializing...</div>
+
+      <section class="content-grid">
+        <section class="panel">
+          <div style="display: grid; gap: 10px">
+            <h2>Visualization</h2>
+            <img
+              id="vizImage"
+              alt="OCR visualization"
+              hidden
+              style="width: 100%; border-radius: 12px; border: 1px solid #6665; background: #0001"
+            />
+          </div>
+        </section>
+
+        <div style="display: grid; gap: 16px">
+          <section class="panel">
+            <h3>Metrics</h3>
+            <pre id="metrics">No runs yet</pre>
+          </section>
+          <section class="panel">
+            <h3>Recognition results</h3>
+            <ol id="results"></ol>
+          </section>
+        </div>
+      </section>
+    </main>
+
+    <script type="module" src="./src/main.ts"></script>
+  </body>
+</html>
diff --git a/paddleocr-js/apps/demo/package.json b/paddleocr-js/apps/demo/package.json
new file mode 100644
index 0000000..ef75dd2
--- /dev/null
+++ b/paddleocr-js/apps/demo/package.json
@@ -0,0 +1,28 @@
+{
+  "name": "demo",
+  "private": true,
+  "version": "0.1.0",
+  "type": "module",
+  "keywords": [
+    "paddleocr",
+    "ocr-demo",
+    "browser-ocr",
+    "vite"
+  ],
+  "engines": {
+    "node": ">=20.11"
+  },
+  "scripts": {
+    "dev": "vite",
+    "build": "vite build",
+    "preview": "vite preview",
+    "typecheck": "tsc --noEmit"
+  },
+  "dependencies": {
+    "@paddleocr/paddleocr-js": "*"
+  },
+  "devDependencies": {
+    "typescript": "^5.9.3",
+    "vite": "^6.4.1"
+  }
+}
diff --git a/paddleocr-js/apps/demo/src/main.ts b/paddleocr-js/apps/demo/src/main.ts
new file mode 100644
index 0000000..69b6247
--- /dev/null
+++ b/paddleocr-js/apps/demo/src/main.ts
@@ -0,0 +1,254 @@
+import { PaddleOCR } from "@paddleocr/paddleocr-js";
+import type { OcrResult, OcrResultItem } from "@paddleocr/paddleocr-js";
+import { OcrVisualizer } from "@paddleocr/paddleocr-js/viz";
+
+type OcrEngine = Awaited<ReturnType<typeof PaddleOCR.create>>;
+
+const ORT_WASM_PATHS = "https://cdn.jsdelivr.net/npm/onnxruntime-web/dist/";
+const DEFAULT_RUNTIME_PARAMS = Object.freeze({
+  textDetThresh: 0.3,
+  textDetBoxThresh: 0.6,
+  textDetUnclipRatio: 1.5,
+  textRecScoreThresh: 0.1
+});
+
+function getDemoThreadCount(): number {
+  return self.crossOriginIsolated
+    ? Math.min(4, Math.max(1, (navigator.hardwareConcurrency || 2) - 1))
+    : 1;
+}
+
+const ui = {
+  modelPreset: document.getElementById("modelPreset") as HTMLSelectElement,
+  runtimeBackend: document.getElementById("runtimeBackend") as HTMLSelectElement,
+  detThresh: document.getElementById("detThresh") as HTMLInputElement,
+  boxThresh: document.getElementById("boxThresh") as HTMLInputElement,
+  unclipRatio: document.getElementById("unclipRatio") as HTMLInputElement,
+  recScoreThresh: document.getElementById("recScoreThresh") as HTMLInputElement,
+  imageInput: document.getElementById("imageInput") as HTMLInputElement,
+  chooseImageBtn: document.getElementById("chooseImageBtn") as HTMLButtonElement,
+  reinitializeBtn: document.getElementById("reinitializeBtn") as HTMLButtonElement,
+  runBtn: document.getElementById("runBtn") as HTMLButtonElement,
+  status: document.getElementById("status") as HTMLElement,
+  metrics: document.getElementById("metrics") as HTMLPreElement,
+  results: document.getElementById("results") as HTMLOListElement,
+  vizImage: document.getElementById("vizImage") as HTMLImageElement
+};
+
+interface AppState {
+  imageFile: File | null;
+  previewBitmap: ImageBitmap | null;
+  lastResult: OcrResult | null;
+  ocr: OcrEngine | null;
+  ocrReady: boolean;
+  vizObjectUrl: string | null;
+}
+
+const state: AppState = {
+  imageFile: null,
+  previewBitmap: null,
+  lastResult: null,
+  ocr: null,
+  ocrReady: false,
+  vizObjectUrl: null
+};
+
+function updateRunButtonState(): void {
+  ui.runBtn.disabled = !state.imageFile || !state.ocrReady;
+}
+
+const visualizer = new OcrVisualizer({
+  font: {
+    family: "PingFang SC",
+    source:
+      "https://paddle-model-ecology.bj.bcebos.com/paddlex/PaddleX3.0/fonts/PingFang-SC-Regular.ttf"
+  }
+});
+
+function setStatus(text: string, isError = false): void {
+  ui.status.textContent = text;
+  ui.status.style.color = isError ? "#b91c1c" : "";
+}
+
+function formatMs(value: number): string {
+  return `${value.toFixed(1)} ms`;
+}
+
+function showVizImage(blob: Blob): void {
+  if (state.vizObjectUrl) {
+    URL.revokeObjectURL(state.vizObjectUrl);
+  }
+  state.vizObjectUrl = URL.createObjectURL(blob);
+  ui.vizImage.src = state.vizObjectUrl;
+  ui.vizImage.hidden = false;
+}
+
+function showPreviewImage(bitmap: ImageBitmap): void {
+  // For pre-OCR preview, draw to an offscreen canvas and display as image
+  const canvas = document.createElement("canvas");
+  canvas.width = bitmap.width;
+  canvas.height = bitmap.height;
+  const ctx = canvas.getContext("2d");
+  if (!ctx) return;
+  ctx.drawImage(bitmap, 0, 0);
+  canvas.toBlob((blob) => {
+    if (blob) showVizImage(blob);
+  });
+}
+
+function renderResults(items: OcrResultItem[]): void {
+  ui.results.innerHTML = "";
+  items.forEach((item) => {
+    const li = document.createElement("li");
+    li.textContent = `${item.text} | score=${item.score.toFixed(3)}`;
+    ui.results.appendChild(li);
+  });
+}
+
+function getRuntimeOptions() {
+  return {
+    backend: ui.runtimeBackend.value as "auto" | "webgpu" | "wasm",
+    wasmPaths: ORT_WASM_PATHS,
+    numThreads: getDemoThreadCount(),
+    simd: true
+  };
+}
+
+async function initializeOcrEngine(): Promise<void> {
+  state.ocrReady = false;
+  updateRunButtonState();
+
+  if (state.ocr) {
+    await state.ocr.dispose();
+  }
+
+  const preset = ui.modelPreset.value;
+
+  state.ocr = await PaddleOCR.create({
+    initialize: false,
+    worker: false,
+    textDetectionModelName: `${preset}_det`,
+    textRecognitionModelName: `${preset}_rec`,
+    ortOptions: getRuntimeOptions()
+  });
+
+  const summary = await state.ocr.initialize();
+  state.ocrReady = true;
+  ui.metrics.textContent = [
+    `model: ${preset}`,
+    `initialize: ${formatMs(summary.elapsedMs)}`,
+    `backend(requested): ${summary.backend}`,
+    `webgpu available: ${summary.webgpuAvailable ? "yes" : "no"}`,
+    `provider(det): ${summary.detProvider}`,
+    `provider(rec): ${summary.recProvider}`,
+    `assets: ${String(summary.assets.length)}`
+  ].join("\n");
+  updateRunButtonState();
+}
+
+async function handleImageSelection(file: File | undefined): Promise<void> {
+  if (!file) return;
+  state.imageFile = file;
+  state.previewBitmap?.close();
+  state.previewBitmap = await createImageBitmap(file);
+  showPreviewImage(state.previewBitmap);
+  updateRunButtonState();
+  setStatus(`Image selected: ${file.name}`);
+}
+
+async function runOcr(): Promise<void> {
+  if (!state.ocrReady || !state.ocr || !state.imageFile) {
+    setStatus("Wait for OCR engine initialization to finish, then choose an image.", true);
+    return;
+  }
+
+  try {
+    setStatus("Running OCR...");
+    const result: OcrResult = (
+      await state.ocr.predict(state.imageFile, {
+        textDetThresh: Number(ui.detThresh.value),
+        textDetBoxThresh: Number(ui.boxThresh.value),
+        textDetUnclipRatio: Number(ui.unclipRatio.value),
+        textRecScoreThresh: Number(ui.recScoreThresh.value)
+      })
+    )[0];
+
+    if (!state.previewBitmap) {
+      state.previewBitmap = await createImageBitmap(state.imageFile);
+    }
+
+    // Render side-by-side visualization using viz module
+    const blob = await visualizer.toBlob(state.previewBitmap, result);
+    showVizImage(blob);
+
+    renderResults(result.items);
+    state.lastResult = result;
+    ui.metrics.textContent = [
+      ui.metrics.textContent,
+      "",
+      `det: ${formatMs(result.metrics.detMs)}`,
+      `rec: ${formatMs(result.metrics.recMs)}`,
+      `total: ${formatMs(result.metrics.totalMs)}`,
+      `detected boxes: ${String(result.metrics.detectedBoxes)}`,
+      `recognized lines: ${String(result.metrics.recognizedCount)}`
+    ].join("\n");
+    setStatus(`OCR complete: ${String(result.metrics.recognizedCount)} text lines recognized.`);
+  } catch (err: unknown) {
+    console.error(err);
+    const message = err instanceof Error ? err.message : String(err);
+    setStatus(`OCR failed: ${message}`, true);
+  }
+}
+
+ui.imageInput.addEventListener("change", (event: Event) => {
+  const target = event.target as HTMLInputElement;
+  void handleImageSelection(target.files?.[0]);
+});
+
+ui.chooseImageBtn.addEventListener("click", () => {
+  ui.imageInput.click();
+});
+
+async function initialize(): Promise<void> {
+  try {
+    ui.reinitializeBtn.disabled = true;
+    state.ocrReady = false;
+    updateRunButtonState();
+
+    setStatus("Initializing...");
+    await initializeOcrEngine();
+
+    setStatus("Loading visualization font...");
+    try {
+      await visualizer.loadFont();
+    } catch (fontErr: unknown) {
+      console.warn("Font load failed, using system font:", fontErr);
+      setStatus("Ready (visualization will use system font).");
+      updateRunButtonState();
+      return;
+    }
+
+    setStatus("Ready.");
+    updateRunButtonState();
+  } catch (err: unknown) {
+    console.error(err);
+    const message = err instanceof Error ? err.message : String(err);
+    setStatus(`Initialization failed: ${message}`, true);
+    state.ocrReady = false;
+    updateRunButtonState();
+  } finally {
+    ui.reinitializeBtn.disabled = false;
+  }
+}
+
+ui.detThresh.value = String(DEFAULT_RUNTIME_PARAMS.textDetThresh);
+ui.boxThresh.value = String(DEFAULT_RUNTIME_PARAMS.textDetBoxThresh);
+ui.unclipRatio.value = String(DEFAULT_RUNTIME_PARAMS.textDetUnclipRatio);
+ui.recScoreThresh.value = String(DEFAULT_RUNTIME_PARAMS.textRecScoreThresh);
+ui.reinitializeBtn.addEventListener("click", () => void initialize());
+ui.modelPreset.addEventListener("change", () => void initialize());
+ui.runtimeBackend.addEventListener("change", () => void initialize());
+
+ui.runBtn.addEventListener("click", () => void runOcr());
+
+void initialize();
diff --git a/paddleocr-js/apps/demo/tsconfig.json b/paddleocr-js/apps/demo/tsconfig.json
new file mode 100644
index 0000000..b7e5673
--- /dev/null
+++ b/paddleocr-js/apps/demo/tsconfig.json
@@ -0,0 +1,21 @@
+{
+  "compilerOptions": {
+    "target": "ES2022",
+    "module": "ESNext",
+    "moduleResolution": "bundler",
+    "strict": true,
+    "noEmit": true,
+    "skipLibCheck": true,
+    "isolatedModules": true,
+    "verbatimModuleSyntax": true,
+    "esModuleInterop": true,
+    "forceConsistentCasingInFileNames": true,
+    "lib": ["ES2022", "DOM", "DOM.Iterable"],
+    "baseUrl": ".",
+    "paths": {
+      "@paddleocr/paddleocr-js": ["../../packages/core/src/index.ts"],
+      "@paddleocr/paddleocr-js/viz": ["../../packages/core/src/viz/index.ts"]
+    }
+  },
+  "include": ["src", "../../packages/core/src/types/*.d.ts"]
+}
diff --git a/paddleocr-js/apps/demo/vite.config.js b/paddleocr-js/apps/demo/vite.config.js
new file mode 100644
index 0000000..fecc3e2
--- /dev/null
+++ b/paddleocr-js/apps/demo/vite.config.js
@@ -0,0 +1,32 @@
+import { resolve } from "node:path";
+import { defineConfig } from "vite";
+
+export default defineConfig(({ command }) => ({
+  resolve: {
+    alias:
+      command === "serve"
+        ? {
+            "@paddleocr/paddleocr-js/viz": resolve(
+              __dirname,
+              "../../packages/core/src/viz/index.ts"
+            ),
+            "@paddleocr/paddleocr-js": resolve(__dirname, "../../packages/core/src/index.ts")
+          }
+        : {}
+  },
+  worker: {
+    format: "es"
+  },
+  server: {
+    headers: {
+      "Cross-Origin-Opener-Policy": "same-origin",
+      "Cross-Origin-Embedder-Policy": "credentialless"
+    }
+  },
+  preview: {
+    headers: {
+      "Cross-Origin-Opener-Policy": "same-origin",
+      "Cross-Origin-Embedder-Policy": "credentialless"
+    }
+  }
+}));
diff --git a/paddleocr-js/docs/architecture.md b/paddleocr-js/docs/architecture.md
new file mode 100644
index 0000000..dc7560d
--- /dev/null
+++ b/paddleocr-js/docs/architecture.md
@@ -0,0 +1,82 @@
+# Architecture
+
+English | [简体中文](architecture_cn.md)
+
+## Project structure
+
+The `paddleocr-js` folder has two main parts:
+
+- `packages/core`: the browser PaddleOCR SDK (published on npm as `@paddleocr/paddleocr-js`)
+- `apps/demo`: a demo application for PP-OCR that consumes the SDK
+
+## SDK package layout (`packages/core`)
+
+```
+src/
+├── runtime/       — inference runtime setup
+├── resources/     — model & asset management
+├── models/        — model wiring
+├── platform/      — browser/worker input adaptation
+├── worker/        — worker transport layer
+├── pipelines/     — pipeline implementations
+├── viz/           — visualization (optional)
+├── types/         — external type declarations
+└── utils/         — shared utilities
+```
+
+The current high-level pipeline entry point is `PaddleOCR.create()`. It coordinates:
+
+1. runtime initialization
+2. execution backend selection
+3. model download
+4. inference session creation
+5. OCR pipeline execution
+
+## Worker execution model
+
+`PaddleOCR.create()` supports 2 execution modes:
+
+- main-thread mode: returns `PaddleOCR`, which runs OCR directly on the calling thread
+- worker-backed mode: returns `WorkerBackedPaddleOCR`, which forwards OCR lifecycle calls to a dedicated worker
+
+The runtime flow for worker mode is:
+
+1. `PaddleOCR.create({ worker: true })` resolves OCR options and creates a `WorkerBackedPaddleOCR`
+2. `WorkerBackedPaddleOCR` sends `init/predict/dispose` requests through `WorkerTransportClient`
+3. the OCR pipeline layer owns the default worker factory and points it at `src/pipelines/ocr/worker-entry.ts`
+4. `src/pipelines/ocr/worker-entry.ts` binds the generic worker bootstrap in `src/worker/entry.ts` to the OCR-specific worker handler
+5. `OcrPipelineRunner` runs OpenCV.js, ONNX Runtime Web, model loading, detection, and recognition inside the worker
+6. results and errors are serialized back to the main thread
+
+Input handling is split by environment:
+
+- main thread: browser inputs are normalized into transferable payloads
+- worker: payloads are reconstructed into runtime inputs such as `cv.Mat`
+
+Worker mode uses the package worker path and explicitly disables ONNX Runtime Web wasm proxy internally. This avoids stacking two worker layers and keeps the package responsible for the concurrency model.
+
+ONNX Runtime Web requires WASM binaries at runtime. `ortOptions.wasmPaths` is a unified configuration that applies to both execution modes — setting it once controls where WASM is loaded in both main-thread and worker contexts:
+
+```ts
+PaddleOCR.create({
+  ortOptions: { wasmPaths: "/assets/" }
+});
+```
+
+When `wasmPaths` is set, both modes fetch WASM from the specified path. When it is not set, each mode falls back differently:
+
+- main-thread mode: ORT resolves WASM through the consumer's bundler (the bundler copies `.wasm` files from `node_modules/onnxruntime-web/dist/` into the build output and rewrites the URLs automatically)
+- worker mode: the SDK falls back to a CDN URL pinned to the ORT version installed at SDK build time, and emits a console warning recommending the consumer set `ortOptions.wasmPaths`
+
+Setting `ortOptions.wasmPaths` explicitly is recommended for worker mode to ensure version consistency between the two modes.
+
+## Application responsibilities
+
+The SDK owns OCR runtime setup and inference orchestration. The host application still owns:
+
+- deployment headers required by the runtime environment
+- static asset hosting and model URL configuration
+- worker-capable bundler/runtime support when `worker: true` is used
+- application UI, status messaging, and visualization
+
+Here, the `apps/` directory contains such host applications.
diff --git a/paddleocr-js/docs/architecture_cn.md b/paddleocr-js/docs/architecture_cn.md
new file mode 100644
index 0000000..f14bdd8
--- /dev/null
+++ b/paddleocr-js/docs/architecture_cn.md
@@ -0,0 +1,82 @@
+# 架构说明
+
+[English](architecture.md) | 简体中文
+
+## 项目结构
+
+`paddleocr-js` 目录下主要有两部分：
+
+- `packages/core`：浏览器 PaddleOCR SDK，发布到 npm 时包名为 `@paddleocr/paddleocr-js`
+- `apps/demo`：依赖该 SDK 的 PP-OCR 演示应用
+
+## SDK 包布局（`packages/core`）
+
+```
+src/
+├── runtime/       — 推理运行时初始化
+├── resources/     — 模型与资源管理
+├── models/        — 模型接线
+├── platform/      — 浏览器 / Worker 输入适配
+├── worker/        — Worker 传输层
+├── pipelines/     — 产线实现
+├── viz/           — 可视化（可选）
+├── types/         — 外部库类型声明
+└── utils/         — 共享工具
+```
+
+当前高层产线入口为 `PaddleOCR.create()`，它负责协调：
+
+1. 运行时初始化
+2. 执行后端选择
+3. 模型下载
+4. 推理会话创建
+5. OCR 产线执行
+
+## Worker 执行模型
+
+`PaddleOCR.create()` 支持两种执行模式：
+
+- 主线程模式：返回 `PaddleOCR`，直接在调用线程上执行 OCR
+- Worker 模式：返回 `WorkerBackedPaddleOCR`，将 OCR 生命周期调用转发到独立 Worker
+
+Worker 模式下的运行流程如下：
+
+1. `PaddleOCR.create({ worker: true })` 解析 OCR 选项并创建 `WorkerBackedPaddleOCR`
+2. `WorkerBackedPaddleOCR` 通过 `WorkerTransportClient` 发送 `init` / `predict` / `dispose` 请求
+3. OCR 产线层持有默认 Worker 工厂，并将其指向 `src/pipelines/ocr/worker-entry.ts`
+4. `src/pipelines/ocr/worker-entry.ts` 将 `src/worker/entry.ts` 中的通用 Worker 引导逻辑与 OCR 专用处理逻辑绑定
+5. `OcrPipelineRunner` 在 Worker 内运行 OpenCV.js、ONNX Runtime Web、模型加载、检测与识别
+6. 结果和错误会被序列化后传回主线程
+
+输入处理按环境拆分：
+
+- 主线程：将浏览器输入标准化为可传输的负载
+- Worker：将负载还原为 `cv.Mat` 等运行时输入
+
+Worker 模式使用包内 Worker 路径，并在内部显式关闭 ONNX Runtime Web 的 wasm proxy。这样可以避免双层 Worker 叠加，并让包本身负责并发模型。
+
+ONNX Runtime Web 在运行时需要 WASM 二进制。`ortOptions.wasmPaths` 对两种执行模式统一生效，设置一次即可控制主线程和 Worker 两侧的 WASM 加载位置：
+
+```ts
+PaddleOCR.create({
+  ortOptions: { wasmPaths: "/assets/" }
+});
+```
+
+设置了 `wasmPaths` 时，两种模式都会从指定路径拉取 WASM。未设置时，两种模式的回退行为不同：
+
+- 主线程模式：ORT 通过使用方的打包工具解析 WASM（通常由打包工具把 `node_modules/onnxruntime-web/dist/` 下的 `.wasm` 文件拷贝到构建产物并自动改写 URL）
+- Worker 模式：SDK 会回退到与构建时安装的 ORT 版本绑定的 CDN，并在控制台提示建议显式设置 `ortOptions.wasmPaths`
+
+因此，在 Worker 模式下建议显式设置 `ortOptions.wasmPaths`，以保证两种模式使用同一套 WASM 版本。
+
+## 应用侧职责
+
+SDK 负责 OCR 运行时初始化与推理编排；宿主应用仍需负责：
+
+- 运行环境所需的部署响应头
+- 静态资源托管与模型 URL 配置
+- `worker: true` 场景下支持产出并加载 Worker 的打包工具或运行时
+- 应用界面、状态提示与可视化
+
+这里的 `apps/` 目录就承载了这类宿主应用。
diff --git a/paddleocr-js/docs/development.md b/paddleocr-js/docs/development.md
new file mode 100644
index 0000000..aa96aab
--- /dev/null
+++ b/paddleocr-js/docs/development.md
@@ -0,0 +1,66 @@
+# Development
+
+English | [简体中文](development_cn.md)
+
+## Install
+
+```bash
+npm install
+```
+
+## Common commands
+
+Commands from the `paddleocr-js/` root:
+
+```bash
+npm run build          # build SDK then demo (explicit topological order)
+npm run build:sdk      # build only the SDK (packages/core)
+npm run build:demo     # build only the demo app (apps/demo)
+npm run lint
+npm run test
+npm run typecheck      # typecheck all workspaces (core + demo)
+npm run check          # format:check → lint → build:sdk → typecheck → test → build:demo
+npm run clean          # remove all dist/ directories
+```
+
+Demo app (Vite dev server):
+
+```bash
+npm run dev:demo
+```
+
+Single-workspace examples:
+
+```bash
+npm run build --workspace packages/core
+npm run build --workspace apps/demo
+```
+
+## TypeScript
+
+Both the SDK (`packages/core`) and the demo app (`apps/demo`) are written in TypeScript with strict mode enabled. ESLint uses `typescript-eslint` with `strictTypeChecked` for source files under `packages/**/src/` and `apps/**/src/`. Test files under `packages/**/test/` use the lighter `recommendedTypeChecked` preset with relaxed rules (e.g. `no-unsafe-*` and `no-explicit-any` are disabled).
+
+`npm run typecheck` runs `tsc --noEmit` across all workspaces. The demo typechecks directly against the SDK's source using `paths` mapping in its `tsconfig.json`, so it does not strictly require `build:sdk` to run first for typechecking.
+
+## Build
+
+The SDK builds with Vite library mode (`npm run build` in `packages/core`). Output in `dist/`:
+
+- `index.mjs` — ESM entry
+- `index.d.ts` — type declarations
+- `viz.mjs` — ESM (viz subpath)
+- `assets/worker-entry-*.js` — self-contained worker bundle (OpenCV.js + ORT JS runtime)
+
+A custom Vite plugin (`libraryWorkerPlugin`) post-processes the build output for npm compatibility:
+
+1. Rewrites absolute worker asset paths to relative, so the file resolves from the SDK module's location rather than the web origin.
+2. Splits the `new Worker(new URL(STRING, import.meta.url))` pattern into a URL variable + Worker construction. This lets downstream bundlers' asset-URL plugins copy the worker file, while preventing their worker-detection plugins from trying to re-bundle it.
+3. Strips base64-encoded WASM binaries that Vite inlines into the worker asset. In worker mode, ORT loads WASM at runtime via `ort.env.wasm.wasmPaths` (set by the consumer, or falling back to a CDN URL pinned to the installed ORT version). This significantly reduces the size of the worker file.
+
+The demo app uses a Vite alias during development (`npm run dev`) to build directly from core's TypeScript source, enabling instant HMR. During production builds (`npm run build`), it consumes the SDK's pre-built `dist/` via workspace linking — the downstream-compatible worker URL pattern allows Vite to correctly copy the worker asset into the demo's output.
+
+## Testing strategy
+
+- unit tests for config parsing and registry behavior
+- lightweight jsdom checks for browser platform helpers
+- no large real-model inference in CI by default
diff --git a/paddleocr-js/docs/development_cn.md b/paddleocr-js/docs/development_cn.md
new file mode 100644
index 0000000..be2da6e
--- /dev/null
+++ b/paddleocr-js/docs/development_cn.md
@@ -0,0 +1,66 @@
+# 开发指南
+
+[English](development.md) | 简体中文
+
+## 安装依赖
+
+```bash
+npm install
+```
+
+## 常用命令
+
+在 `paddleocr-js/` 根目录执行：
+
+```bash
+npm run build          # 先构建 SDK 再构建 demo（显式拓扑顺序）
+npm run build:sdk      # 仅构建 SDK（packages/core）
+npm run build:demo     # 仅构建 demo 应用（apps/demo）
+npm run lint
+npm run test
+npm run typecheck      # 对所有 workspace 执行类型检查（core + demo）
+npm run check          # format:check → lint → build:sdk → typecheck → test → build:demo
+npm run clean          # 删除所有 dist/ 目录
+```
+
+Demo 应用（Vite 开发服务器）：
+
+```bash
+npm run dev:demo
+```
+
+单个 workspace 的示例：
+
+```bash
+npm run build --workspace packages/core
+npm run build --workspace apps/demo
+```
+
+## TypeScript
+
+SDK（`packages/core`）与 demo 应用（`apps/demo`）都使用 TypeScript 严格模式。`packages/**/src/` 与 `apps/**/src/` 下的源码使用 `typescript-eslint` 的 `strictTypeChecked` 规则集；`packages/**/test/` 下的测试使用较轻的 `recommendedTypeChecked`，并放宽了部分规则（例如 `no-unsafe-*` 与 `no-explicit-any`）。
+
+`npm run typecheck` 会在所有 workspace 上执行 `tsc --noEmit`。demo 在 `tsconfig.json` 里通过 `paths` 直接引用 SDK 源码，因此类型检查并不严格依赖先执行 `build:sdk`。
+
+## 构建
+
+SDK 使用 Vite 的库模式构建（在 `packages/core` 中执行 `npm run build`）。`dist/` 下的构建产物包括：
+
+- `index.mjs` — ESM 入口
+- `index.d.ts` — 类型声明
+- `viz.mjs` — 可视化子路径的 ESM 入口
+- `assets/worker-entry-*.js` — 自包含的 Worker bundle（OpenCV.js + ORT JS 运行时）
+
+自定义 Vite 插件 `libraryWorkerPlugin` 会对构建产物做后处理，以兼容 npm 分发场景：
+
+1. 将 Worker 资源的绝对路径改写为相对路径，使文件相对于 SDK 模块路径解析，而不是站点根路径
+2. 将 `new Worker(new URL(STRING, import.meta.url))` 拆成 URL 变量与 Worker 构造，便于下游打包工具复制 Worker 文件，并避免被当作二次打包目标
+3. 移除 Vite 内联到 Worker 产物中的 base64 WASM 二进制；在 Worker 模式下，ORT 会在运行时通过 `ort.env.wasm.wasmPaths` 加载 WASM（由使用方配置，或回退到与安装 ORT 版本绑定的 CDN），这样可以显著减小 Worker 文件体积
+
+开发阶段，demo 通过 Vite alias 直接引用 core 的 TypeScript 源码，从而获得更好的 HMR 体验；生产构建（`npm run build`）时，则通过 workspace 链接使用 SDK 预构建的 `dist/`，并依靠兼容下游打包器的 Worker URL 形式将 Worker 文件正确复制到 demo 产物中。
+
+## 测试策略
+
+- 配置解析与注册表行为的单元测试
+- 面向浏览器平台辅助函数的轻量级 jsdom 测试
+- CI 默认不运行大规模真实模型推理
diff --git a/paddleocr-js/docs/monorepo.md b/paddleocr-js/docs/monorepo.md
new file mode 100644
index 0000000..035dee2
--- /dev/null
+++ b/paddleocr-js/docs/monorepo.md
@@ -0,0 +1,35 @@
+# Monorepo conventions
+
+English | [简体中文](monorepo_cn.md)
+
+## Command execution
+
+When you only want one workspace, use root-level workspace commands with explicit paths:
+
+```bash
+npm run build --workspace packages/core
+npm run dev --workspace apps/demo
+```
+
+(You can also use workspace package names where unambiguous, e.g. `npm run dev --workspace demo`.)
+
+## Workspace roles
+
+- `packages/*`: reusable packages; the SDK lives under `packages/core` but keeps the **npm package name** `@paddleocr/paddleocr-js`
+- `apps/*`: private applications such as demos (`apps/demo`); not published to npm as products
+
+## Versioning and release
+
+- **Directory:** `packages/core` — SDK source and publish manifest for the public package
+- **npm package name:** `@paddleocr/paddleocr-js` — what consumers `npm install` and import in code
+- **Directory:** `apps/demo` — private demo, not an npm release target
+- Changesets manages versioning; the demo package is ignored in `.changeset/config.json`
+- `npm run release` builds the SDK and publishes via `changeset publish`
+- `packages/core` has a `prepublishOnly` script that auto-builds before `npm publish` / `npm pack`
+
+## Linting and tests
+
+- `packages/**/src/**/*.ts` is linted with `strictTypeChecked` TypeScript rules and browser-oriented globals
+- `packages/**/test/**/*.ts` is linted with the lighter `recommendedTypeChecked` preset (browser + Node globals, with relaxed rules such as `no-unsafe-*` and `no-explicit-any` disabled)
+- `apps/**/src/**/*.ts` also uses `strictTypeChecked` TypeScript rules with browser-oriented globals
+- `apps/**/*.js`, root config files (`*.config.{js,ts}`), and package config files (`packages/**/*.config.*`) are linted with basic ESLint rules and both Node and browser globals
diff --git a/paddleocr-js/docs/monorepo_cn.md b/paddleocr-js/docs/monorepo_cn.md
new file mode 100644
index 0000000..8c3ce11
--- /dev/null
+++ b/paddleocr-js/docs/monorepo_cn.md
@@ -0,0 +1,35 @@
+# Monorepo 约定
+
+[English](monorepo.md) | 简体中文
+
+## 命令与 workspace
+
+当您只想操作一个 workspace 时，请在仓库根目录使用带显式路径的 workspace 命令：
+
+```bash
+npm run build --workspace packages/core
+npm run dev --workspace apps/demo
+```
+
+（在包名没有歧义时，也可以直接使用 workspace 的包名，例如 `npm run dev --workspace demo`。）
+
+## Workspace 角色
+
+- `packages/*`：可复用包；SDK 位于 `packages/core`，但其 **npm 包名** 仍为 `@paddleocr/paddleocr-js`
+- `apps/*`：私有应用（如 `apps/demo`），不作为 npm 产品发布
+
+## 版本与发布
+
+- **目录：** `packages/core` — 对外 npm 包的源码与发布清单
+- **npm 包名：** `@paddleocr/paddleocr-js` — 用户 `npm install` 后在代码中实际导入的名称
+- **目录：** `apps/demo` — 私有 demo，不是 npm 发布目标
+- 版本由 Changesets 管理；demo 包在 `.changeset/config.json` 中被忽略
+- `npm run release` 会构建 SDK 并通过 `changeset publish` 发布
+- `packages/core` 的 `prepublishOnly` 会在 `npm publish` / `npm pack` 前自动执行构建
+
+## Lint 与测试
+
+- `packages/**/src/**/*.ts` 使用 `strictTypeChecked` TypeScript 规则，并启用面向浏览器的全局变量
+- `packages/**/test/**/*.ts` 使用较轻的 `recommendedTypeChecked` 规则集（浏览器 + Node 全局），并放宽部分规则，例如 `no-unsafe-*` 和 `no-explicit-any`
+- `apps/**/src/**/*.ts` 同样使用 `strictTypeChecked` TypeScript 规则与浏览器全局变量
+- `apps/**/*.js`、仓库根目录配置文件（`*.config.{js,ts}`）以及包内配置文件（`packages/**/*.config.*`）使用基础 ESLint 规则，同时启用 Node 与浏览器全局
diff --git a/paddleocr-js/eslint.config.js b/paddleocr-js/eslint.config.js
new file mode 100644
index 0000000..dc9a4a9
--- /dev/null
+++ b/paddleocr-js/eslint.config.js
@@ -0,0 +1,49 @@
+import eslint from "@eslint/js";
+import tseslint from "typescript-eslint";
+import globals from "globals";
+
+export default tseslint.config(
+  {
+    ignores: ["**/dist", "**/node_modules", "**/coverage", "**/.cache"]
+  },
+  eslint.configs.recommended,
+  {
+    files: ["packages/**/src/**/*.ts", "apps/**/src/**/*.ts"],
+    extends: [...tseslint.configs.strictTypeChecked],
+    languageOptions: {
+      globals: { ...globals.browser },
+      parserOptions: {
+        project: "./tsconfig.eslint.json",
+        tsconfigRootDir: import.meta.dirname
+      }
+    }
+  },
+  {
+    files: ["packages/**/test/**/*.ts"],
+    extends: [...tseslint.configs.recommendedTypeChecked],
+    languageOptions: {
+      globals: { ...globals.browser },
+      parserOptions: {
+        project: "./tsconfig.eslint.json",
+        tsconfigRootDir: import.meta.dirname
+      }
+    },
+    rules: {
+      "@typescript-eslint/no-unsafe-assignment": "off",
+      "@typescript-eslint/no-unsafe-argument": "off",
+      "@typescript-eslint/no-unsafe-member-access": "off",
+      "@typescript-eslint/no-unsafe-call": "off",
+      "@typescript-eslint/no-unsafe-return": "off",
+      "@typescript-eslint/no-explicit-any": "off",
+      "@typescript-eslint/require-await": "off",
+      "@typescript-eslint/no-extraneous-class": "off",
+      "@typescript-eslint/unbound-method": "off"
+    }
+  },
+  {
+    files: ["apps/**/*.js", "*.config.{js,ts}", "packages/**/*.config.*"],
+    languageOptions: {
+      globals: { ...globals.browser, ...globals.node }
+    }
+  }
+);
diff --git a/paddleocr-js/package-lock.json b/paddleocr-js/package-lock.json
new file mode 100644
index 0000000..58c0c5d
--- /dev/null
+++ b/paddleocr-js/package-lock.json
@@ -0,0 +1,5684 @@
+{
+  "name": "paddleocr-js-workspace",
+  "version": "0.1.0",
+  "lockfileVersion": 3,
+  "requires": true,
+  "packages": {
+    "": {
+      "name": "paddleocr-js-workspace",
+      "version": "0.1.0",
+      "workspaces": [
+        "packages/*",
+        "apps/*"
+      ],
+      "devDependencies": {
+        "@eslint/js": "^10.0.1",
+        "@vitest/coverage-v8": "^3.2.4",
+        "eslint": "^10.0.2",
+        "globals": "^17.4.0",
+        "husky": "^9.1.7",
+        "jsdom": "^26.1.0",
+        "lint-staged": "^16.3.2",
+        "prettier": "^3.8.1",
+        "typescript-eslint": "^8.57.2",
+        "vitest": "^3.2.4"
+      },
+      "engines": {
+        "node": ">=20.11"
+      }
+    },
+    "apps/demo": {
+      "version": "0.1.0",
+      "dependencies": {
+        "@paddleocr/paddleocr-js": "*"
+      },
+      "devDependencies": {
+        "typescript": "^5.9.3",
+        "vite": "^6.4.1"
+      },
+      "engines": {
+        "node": ">=20.11"
+      }
+    },
+    "apps/demo/node_modules/vite": {
+      "version": "6.4.2",
+      "resolved": "https://registry.npmjs.org/vite/-/vite-6.4.2.tgz",
+      "integrity": "sha512-2N/55r4JDJ4gdrCvGgINMy+HH3iRpNIz8K6SFwVsA+JbQScLiC+clmAxBgwiSPgcG9U15QmvqCGWzMbqda5zGQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "esbuild": "^0.25.0",
+        "fdir": "^6.4.4",
+        "picomatch": "^4.0.2",
+        "postcss": "^8.5.3",
+        "rollup": "^4.34.9",
+        "tinyglobby": "^0.2.13"
+      },
+      "bin": {
+        "vite": "bin/vite.js"
+      },
+      "engines": {
+        "node": "^18.0.0 || ^20.0.0 || >=22.0.0"
+      },
+      "funding": {
+        "url": "https://github.com/vitejs/vite?sponsor=1"
+      },
+      "optionalDependencies": {
+        "fsevents": "~2.3.3"
+      },
+      "peerDependencies": {
+        "@types/node": "^18.0.0 || ^20.0.0 || >=22.0.0",
+        "jiti": ">=1.21.0",
+        "less": "*",
+        "lightningcss": "^1.21.0",
+        "sass": "*",
+        "sass-embedded": "*",
+        "stylus": "*",
+        "sugarss": "*",
+        "terser": "^5.16.0",
+        "tsx": "^4.8.1",
+        "yaml": "^2.4.2"
+      },
+      "peerDependenciesMeta": {
+        "@types/node": {
+          "optional": true
+        },
+        "jiti": {
+          "optional": true
+        },
+        "less": {
+          "optional": true
+        },
+        "lightningcss": {
+          "optional": true
+        },
+        "sass": {
+          "optional": true
+        },
+        "sass-embedded": {
+          "optional": true
+        },
+        "stylus": {
+          "optional": true
+        },
+        "sugarss": {
+          "optional": true
+        },
+        "terser": {
+          "optional": true
+        },
+        "tsx": {
+          "optional": true
+        },
+        "yaml": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/@ampproject/remapping": {
+      "version": "2.3.0",
+      "resolved": "https://registry.npmjs.org/@ampproject/remapping/-/remapping-2.3.0.tgz",
+      "integrity": "sha512-30iZtAPgz+LTIYoeivqYo853f02jBYSd5uGnGpkFV0M3xOt9aN73erkgYAmZU43x4VfqcnLxW9Kpg3R5LC4YYw==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "dependencies": {
+        "@jridgewell/gen-mapping": "^0.3.5",
+        "@jridgewell/trace-mapping": "^0.3.24"
+      },
+      "engines": {
+        "node": ">=6.0.0"
+      }
+    },
+    "node_modules/@asamuzakjp/css-color": {
+      "version": "3.2.0",
+      "resolved": "https://registry.npmjs.org/@asamuzakjp/css-color/-/css-color-3.2.0.tgz",
+      "integrity": "sha512-K1A6z8tS3XsmCMM86xoWdn7Fkdn9m6RSVtocUrJYIwZnFVkng/PvkEoWtOWmP+Scc6saYWHWZYbndEEXxl24jw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@csstools/css-calc": "^2.1.3",
+        "@csstools/css-color-parser": "^3.0.9",
+        "@csstools/css-parser-algorithms": "^3.0.4",
+        "@csstools/css-tokenizer": "^3.0.3",
+        "lru-cache": "^10.4.3"
+      }
+    },
+    "node_modules/@babel/helper-string-parser": {
+      "version": "7.27.1",
+      "resolved": "https://registry.npmjs.org/@babel/helper-string-parser/-/helper-string-parser-7.27.1.tgz",
+      "integrity": "sha512-qMlSxKbpRlAridDExk92nSobyDdpPijUq2DW6oDnUqd0iOGxmQjyqhMIihI9+zv4LPyZdRje2cavWPbCbWm3eA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=6.9.0"
+      }
+    },
+    "node_modules/@babel/helper-validator-identifier": {
+      "version": "7.28.5",
+      "resolved": "https://registry.npmjs.org/@babel/helper-validator-identifier/-/helper-validator-identifier-7.28.5.tgz",
+      "integrity": "sha512-qSs4ifwzKJSV39ucNjsvc6WVHs6b7S03sOh2OcHF9UHfVPqWWALUsNUVzhSBiItjRZoLHx7nIarVjqKVusUZ1Q==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=6.9.0"
+      }
+    },
+    "node_modules/@babel/parser": {
+      "version": "7.29.2",
+      "resolved": "https://registry.npmjs.org/@babel/parser/-/parser-7.29.2.tgz",
+      "integrity": "sha512-4GgRzy/+fsBa72/RZVJmGKPmZu9Byn8o4MoLpmNe1m8ZfYnz5emHLQz3U4gLud6Zwl0RZIcgiLD7Uq7ySFuDLA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@babel/types": "^7.29.0"
+      },
+      "bin": {
+        "parser": "bin/babel-parser.js"
+      },
+      "engines": {
+        "node": ">=6.0.0"
+      }
+    },
+    "node_modules/@babel/types": {
+      "version": "7.29.0",
+      "resolved": "https://registry.npmjs.org/@babel/types/-/types-7.29.0.tgz",
+      "integrity": "sha512-LwdZHpScM4Qz8Xw2iKSzS+cfglZzJGvofQICy7W7v4caru4EaAmyUuO6BGrbyQ2mYV11W0U8j5mBhd14dd3B0A==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@babel/helper-string-parser": "^7.27.1",
+        "@babel/helper-validator-identifier": "^7.28.5"
+      },
+      "engines": {
+        "node": ">=6.9.0"
+      }
+    },
+    "node_modules/@bcoe/v8-coverage": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@bcoe/v8-coverage/-/v8-coverage-1.0.2.tgz",
+      "integrity": "sha512-6zABk/ECA/QYSCQ1NGiVwwbQerUCZ+TQbp64Q3AgmfNvurHH0j8TtXa1qbShXA6qqkpAj4V5W8pP6mLe1mcMqA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@csstools/color-helpers": {
+      "version": "5.1.0",
+      "resolved": "https://registry.npmjs.org/@csstools/color-helpers/-/color-helpers-5.1.0.tgz",
+      "integrity": "sha512-S11EXWJyy0Mz5SYvRmY8nJYTFFd1LCNV+7cXyAgQtOOuzb4EsgfqDufL+9esx72/eLhsRdGZwaldu/h+E4t4BA==",
+      "dev": true,
+      "funding": [
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/csstools"
+        },
+        {
+          "type": "opencollective",
+          "url": "https://opencollective.com/csstools"
+        }
+      ],
+      "license": "MIT-0",
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@csstools/css-calc": {
+      "version": "2.1.4",
+      "resolved": "https://registry.npmjs.org/@csstools/css-calc/-/css-calc-2.1.4.tgz",
+      "integrity": "sha512-3N8oaj+0juUw/1H3YwmDDJXCgTB1gKU6Hc/bB502u9zR0q2vd786XJH9QfrKIEgFlZmhZiq6epXl4rHqhzsIgQ==",
+      "dev": true,
+      "funding": [
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/csstools"
+        },
+        {
+          "type": "opencollective",
+          "url": "https://opencollective.com/csstools"
+        }
+      ],
+      "license": "MIT",
+      "engines": {
+        "node": ">=18"
+      },
+      "peerDependencies": {
+        "@csstools/css-parser-algorithms": "^3.0.5",
+        "@csstools/css-tokenizer": "^3.0.4"
+      }
+    },
+    "node_modules/@csstools/css-color-parser": {
+      "version": "3.1.0",
+      "resolved": "https://registry.npmjs.org/@csstools/css-color-parser/-/css-color-parser-3.1.0.tgz",
+      "integrity": "sha512-nbtKwh3a6xNVIp/VRuXV64yTKnb1IjTAEEh3irzS+HkKjAOYLTGNb9pmVNntZ8iVBHcWDA2Dof0QtPgFI1BaTA==",
+      "dev": true,
+      "funding": [
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/csstools"
+        },
+        {
+          "type": "opencollective",
+          "url": "https://opencollective.com/csstools"
+        }
+      ],
+      "license": "MIT",
+      "dependencies": {
+        "@csstools/color-helpers": "^5.1.0",
+        "@csstools/css-calc": "^2.1.4"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "peerDependencies": {
+        "@csstools/css-parser-algorithms": "^3.0.5",
+        "@csstools/css-tokenizer": "^3.0.4"
+      }
+    },
+    "node_modules/@csstools/css-parser-algorithms": {
+      "version": "3.0.5",
+      "resolved": "https://registry.npmjs.org/@csstools/css-parser-algorithms/-/css-parser-algorithms-3.0.5.tgz",
+      "integrity": "sha512-DaDeUkXZKjdGhgYaHNJTV9pV7Y9B3b644jCLs9Upc3VeNGg6LWARAT6O+Q+/COo+2gg/bM5rhpMAtf70WqfBdQ==",
+      "dev": true,
+      "funding": [
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/csstools"
+        },
+        {
+          "type": "opencollective",
+          "url": "https://opencollective.com/csstools"
+        }
+      ],
+      "license": "MIT",
+      "engines": {
+        "node": ">=18"
+      },
+      "peerDependencies": {
+        "@csstools/css-tokenizer": "^3.0.4"
+      }
+    },
+    "node_modules/@csstools/css-tokenizer": {
+      "version": "3.0.4",
+      "resolved": "https://registry.npmjs.org/@csstools/css-tokenizer/-/css-tokenizer-3.0.4.tgz",
+      "integrity": "sha512-Vd/9EVDiu6PPJt9yAh6roZP6El1xHrdvIVGjyBsHR0RYwNHgL7FJPyIIW4fANJNG6FtyZfvlRPpFI4ZM/lubvw==",
+      "dev": true,
+      "funding": [
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/csstools"
+        },
+        {
+          "type": "opencollective",
+          "url": "https://opencollective.com/csstools"
+        }
+      ],
+      "license": "MIT",
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/aix-ppc64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/aix-ppc64/-/aix-ppc64-0.25.12.tgz",
+      "integrity": "sha512-Hhmwd6CInZ3dwpuGTF8fJG6yoWmsToE+vYgD4nytZVxcu1ulHpUQRAB1UJ8+N1Am3Mz4+xOByoQoSZf4D+CpkA==",
+      "cpu": [
+        "ppc64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "aix"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/android-arm": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/android-arm/-/android-arm-0.25.12.tgz",
+      "integrity": "sha512-VJ+sKvNA/GE7Ccacc9Cha7bpS8nyzVv0jdVgwNDaR4gDMC/2TTRc33Ip8qrNYUcpkOHUT5OZ0bUcNNVZQ9RLlg==",
+      "cpu": [
+        "arm"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "android"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/android-arm64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/android-arm64/-/android-arm64-0.25.12.tgz",
+      "integrity": "sha512-6AAmLG7zwD1Z159jCKPvAxZd4y/VTO0VkprYy+3N2FtJ8+BQWFXU+OxARIwA46c5tdD9SsKGZ/1ocqBS/gAKHg==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "android"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/android-x64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/android-x64/-/android-x64-0.25.12.tgz",
+      "integrity": "sha512-5jbb+2hhDHx5phYR2By8GTWEzn6I9UqR11Kwf22iKbNpYrsmRB18aX/9ivc5cabcUiAT/wM+YIZ6SG9QO6a8kg==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "android"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/darwin-arm64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/darwin-arm64/-/darwin-arm64-0.25.12.tgz",
+      "integrity": "sha512-N3zl+lxHCifgIlcMUP5016ESkeQjLj/959RxxNYIthIg+CQHInujFuXeWbWMgnTo4cp5XVHqFPmpyu9J65C1Yg==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "darwin"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/darwin-x64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/darwin-x64/-/darwin-x64-0.25.12.tgz",
+      "integrity": "sha512-HQ9ka4Kx21qHXwtlTUVbKJOAnmG1ipXhdWTmNXiPzPfWKpXqASVcWdnf2bnL73wgjNrFXAa3yYvBSd9pzfEIpA==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "darwin"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/freebsd-arm64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/freebsd-arm64/-/freebsd-arm64-0.25.12.tgz",
+      "integrity": "sha512-gA0Bx759+7Jve03K1S0vkOu5Lg/85dou3EseOGUes8flVOGxbhDDh/iZaoek11Y8mtyKPGF3vP8XhnkDEAmzeg==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "freebsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/freebsd-x64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/freebsd-x64/-/freebsd-x64-0.25.12.tgz",
+      "integrity": "sha512-TGbO26Yw2xsHzxtbVFGEXBFH0FRAP7gtcPE7P5yP7wGy7cXK2oO7RyOhL5NLiqTlBh47XhmIUXuGciXEqYFfBQ==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "freebsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-arm": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-arm/-/linux-arm-0.25.12.tgz",
+      "integrity": "sha512-lPDGyC1JPDou8kGcywY0YILzWlhhnRjdof3UlcoqYmS9El818LLfJJc3PXXgZHrHCAKs/Z2SeZtDJr5MrkxtOw==",
+      "cpu": [
+        "arm"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-arm64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-arm64/-/linux-arm64-0.25.12.tgz",
+      "integrity": "sha512-8bwX7a8FghIgrupcxb4aUmYDLp8pX06rGh5HqDT7bB+8Rdells6mHvrFHHW2JAOPZUbnjUpKTLg6ECyzvas2AQ==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-ia32": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-ia32/-/linux-ia32-0.25.12.tgz",
+      "integrity": "sha512-0y9KrdVnbMM2/vG8KfU0byhUN+EFCny9+8g202gYqSSVMonbsCfLjUO+rCci7pM0WBEtz+oK/PIwHkzxkyharA==",
+      "cpu": [
+        "ia32"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-loong64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-loong64/-/linux-loong64-0.25.12.tgz",
+      "integrity": "sha512-h///Lr5a9rib/v1GGqXVGzjL4TMvVTv+s1DPoxQdz7l/AYv6LDSxdIwzxkrPW438oUXiDtwM10o9PmwS/6Z0Ng==",
+      "cpu": [
+        "loong64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-mips64el": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-mips64el/-/linux-mips64el-0.25.12.tgz",
+      "integrity": "sha512-iyRrM1Pzy9GFMDLsXn1iHUm18nhKnNMWscjmp4+hpafcZjrr2WbT//d20xaGljXDBYHqRcl8HnxbX6uaA/eGVw==",
+      "cpu": [
+        "mips64el"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-ppc64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-ppc64/-/linux-ppc64-0.25.12.tgz",
+      "integrity": "sha512-9meM/lRXxMi5PSUqEXRCtVjEZBGwB7P/D4yT8UG/mwIdze2aV4Vo6U5gD3+RsoHXKkHCfSxZKzmDssVlRj1QQA==",
+      "cpu": [
+        "ppc64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-riscv64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-riscv64/-/linux-riscv64-0.25.12.tgz",
+      "integrity": "sha512-Zr7KR4hgKUpWAwb1f3o5ygT04MzqVrGEGXGLnj15YQDJErYu/BGg+wmFlIDOdJp0PmB0lLvxFIOXZgFRrdjR0w==",
+      "cpu": [
+        "riscv64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-s390x": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-s390x/-/linux-s390x-0.25.12.tgz",
+      "integrity": "sha512-MsKncOcgTNvdtiISc/jZs/Zf8d0cl/t3gYWX8J9ubBnVOwlk65UIEEvgBORTiljloIWnBzLs4qhzPkJcitIzIg==",
+      "cpu": [
+        "s390x"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/linux-x64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-x64/-/linux-x64-0.25.12.tgz",
+      "integrity": "sha512-uqZMTLr/zR/ed4jIGnwSLkaHmPjOjJvnm6TVVitAa08SLS9Z0VM8wIRx7gWbJB5/J54YuIMInDquWyYvQLZkgw==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/netbsd-arm64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/netbsd-arm64/-/netbsd-arm64-0.25.12.tgz",
+      "integrity": "sha512-xXwcTq4GhRM7J9A8Gv5boanHhRa/Q9KLVmcyXHCTaM4wKfIpWkdXiMog/KsnxzJ0A1+nD+zoecuzqPmCRyBGjg==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "netbsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/netbsd-x64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/netbsd-x64/-/netbsd-x64-0.25.12.tgz",
+      "integrity": "sha512-Ld5pTlzPy3YwGec4OuHh1aCVCRvOXdH8DgRjfDy/oumVovmuSzWfnSJg+VtakB9Cm0gxNO9BzWkj6mtO1FMXkQ==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "netbsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/openbsd-arm64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/openbsd-arm64/-/openbsd-arm64-0.25.12.tgz",
+      "integrity": "sha512-fF96T6KsBo/pkQI950FARU9apGNTSlZGsv1jZBAlcLL1MLjLNIWPBkj5NlSz8aAzYKg+eNqknrUJ24QBybeR5A==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "openbsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/openbsd-x64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/openbsd-x64/-/openbsd-x64-0.25.12.tgz",
+      "integrity": "sha512-MZyXUkZHjQxUvzK7rN8DJ3SRmrVrke8ZyRusHlP+kuwqTcfWLyqMOE3sScPPyeIXN/mDJIfGXvcMqCgYKekoQw==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "openbsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/openharmony-arm64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/openharmony-arm64/-/openharmony-arm64-0.25.12.tgz",
+      "integrity": "sha512-rm0YWsqUSRrjncSXGA7Zv78Nbnw4XL6/dzr20cyrQf7ZmRcsovpcRBdhD43Nuk3y7XIoW2OxMVvwuRvk9XdASg==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "openharmony"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/sunos-x64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/sunos-x64/-/sunos-x64-0.25.12.tgz",
+      "integrity": "sha512-3wGSCDyuTHQUzt0nV7bocDy72r2lI33QL3gkDNGkod22EsYl04sMf0qLb8luNKTOmgF/eDEDP5BFNwoBKH441w==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "sunos"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/win32-arm64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/win32-arm64/-/win32-arm64-0.25.12.tgz",
+      "integrity": "sha512-rMmLrur64A7+DKlnSuwqUdRKyd3UE7oPJZmnljqEptesKM8wx9J8gx5u0+9Pq0fQQW8vqeKebwNXdfOyP+8Bsg==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/win32-ia32": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/win32-ia32/-/win32-ia32-0.25.12.tgz",
+      "integrity": "sha512-HkqnmmBoCbCwxUKKNPBixiWDGCpQGVsrQfJoVGYLPT41XWF8lHuE5N6WhVia2n4o5QK5M4tYr21827fNhi4byQ==",
+      "cpu": [
+        "ia32"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@esbuild/win32-x64": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/@esbuild/win32-x64/-/win32-x64-0.25.12.tgz",
+      "integrity": "sha512-alJC0uCZpTFrSL0CCDjcgleBXPnCrEAhTBILpeAp7M/OFgoqtAetfBzX0xM00MUsVVPpVjlPuMbREqnZCXaTnA==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/@eslint-community/eslint-utils": {
+      "version": "4.9.1",
+      "resolved": "https://registry.npmjs.org/@eslint-community/eslint-utils/-/eslint-utils-4.9.1.tgz",
+      "integrity": "sha512-phrYmNiYppR7znFEdqgfWHXR6NCkZEK7hwWDHZUjit/2/U0r6XvkDl0SYnoM51Hq7FhCGdLDT6zxCCOY1hexsQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "eslint-visitor-keys": "^3.4.3"
+      },
+      "engines": {
+        "node": "^12.22.0 || ^14.17.0 || >=16.0.0"
+      },
+      "funding": {
+        "url": "https://opencollective.com/eslint"
+      },
+      "peerDependencies": {
+        "eslint": "^6.0.0 || ^7.0.0 || >=8.0.0"
+      }
+    },
+    "node_modules/@eslint-community/eslint-utils/node_modules/eslint-visitor-keys": {
+      "version": "3.4.3",
+      "resolved": "https://registry.npmjs.org/eslint-visitor-keys/-/eslint-visitor-keys-3.4.3.tgz",
+      "integrity": "sha512-wpc+LXeiyiisxPlEkUzU6svyS1frIO3Mgxj1fdy7Pm8Ygzguax2N3Fa/D/ag1WqbOprdI+uY6wMUl8/a2G+iag==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "engines": {
+        "node": "^12.22.0 || ^14.17.0 || >=16.0.0"
+      },
+      "funding": {
+        "url": "https://opencollective.com/eslint"
+      }
+    },
+    "node_modules/@eslint-community/regexpp": {
+      "version": "4.12.2",
+      "resolved": "https://registry.npmjs.org/@eslint-community/regexpp/-/regexpp-4.12.2.tgz",
+      "integrity": "sha512-EriSTlt5OC9/7SXkRSCAhfSxxoSUgBm33OH+IkwbdpgoqsSsUg7y3uh+IICI/Qg4BBWr3U2i39RpmycbxMq4ew==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": "^12.0.0 || ^14.0.0 || >=16.0.0"
+      }
+    },
+    "node_modules/@eslint/config-array": {
+      "version": "0.23.3",
+      "resolved": "https://registry.npmjs.org/@eslint/config-array/-/config-array-0.23.3.tgz",
+      "integrity": "sha512-j+eEWmB6YYLwcNOdlwQ6L2OsptI/LO6lNBuLIqe5R7RetD658HLoF+Mn7LzYmAWWNNzdC6cqP+L6r8ujeYXWLw==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "dependencies": {
+        "@eslint/object-schema": "^3.0.3",
+        "debug": "^4.3.1",
+        "minimatch": "^10.2.4"
+      },
+      "engines": {
+        "node": "^20.19.0 || ^22.13.0 || >=24"
+      }
+    },
+    "node_modules/@eslint/config-helpers": {
+      "version": "0.5.3",
+      "resolved": "https://registry.npmjs.org/@eslint/config-helpers/-/config-helpers-0.5.3.tgz",
+      "integrity": "sha512-lzGN0onllOZCGroKJmRwY6QcEHxbjBw1gwB8SgRSqK8YbbtEXMvKynsXc3553ckIEBxsbMBU7oOZXKIPGZNeZw==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "dependencies": {
+        "@eslint/core": "^1.1.1"
+      },
+      "engines": {
+        "node": "^20.19.0 || ^22.13.0 || >=24"
+      }
+    },
+    "node_modules/@eslint/core": {
+      "version": "1.1.1",
+      "resolved": "https://registry.npmjs.org/@eslint/core/-/core-1.1.1.tgz",
+      "integrity": "sha512-QUPblTtE51/7/Zhfv8BDwO0qkkzQL7P/aWWbqcf4xWLEYn1oKjdO0gglQBB4GAsu7u6wjijbCmzsUTy6mnk6oQ==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "dependencies": {
+        "@types/json-schema": "^7.0.15"
+      },
+      "engines": {
+        "node": "^20.19.0 || ^22.13.0 || >=24"
+      }
+    },
+    "node_modules/@eslint/js": {
+      "version": "10.0.1",
+      "resolved": "https://registry.npmjs.org/@eslint/js/-/js-10.0.1.tgz",
+      "integrity": "sha512-zeR9k5pd4gxjZ0abRoIaxdc7I3nDktoXZk2qOv9gCNWx3mVwEn32VRhyLaRsDiJjTs0xq/T8mfPtyuXu7GWBcA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": "^20.19.0 || ^22.13.0 || >=24"
+      },
+      "funding": {
+        "url": "https://eslint.org/donate"
+      },
+      "peerDependencies": {
+        "eslint": "^10.0.0"
+      },
+      "peerDependenciesMeta": {
+        "eslint": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/@eslint/object-schema": {
+      "version": "3.0.3",
+      "resolved": "https://registry.npmjs.org/@eslint/object-schema/-/object-schema-3.0.3.tgz",
+      "integrity": "sha512-iM869Pugn9Nsxbh/YHRqYiqd23AmIbxJOcpUMOuWCVNdoQJ5ZtwL6h3t0bcZzJUlC3Dq9jCFCESBZnX0GTv7iQ==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "engines": {
+        "node": "^20.19.0 || ^22.13.0 || >=24"
+      }
+    },
+    "node_modules/@eslint/plugin-kit": {
+      "version": "0.6.1",
+      "resolved": "https://registry.npmjs.org/@eslint/plugin-kit/-/plugin-kit-0.6.1.tgz",
+      "integrity": "sha512-iH1B076HoAshH1mLpHMgwdGeTs0CYwL0SPMkGuSebZrwBp16v415e9NZXg2jtrqPVQjf6IANe2Vtlr5KswtcZQ==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "dependencies": {
+        "@eslint/core": "^1.1.1",
+        "levn": "^0.4.1"
+      },
+      "engines": {
+        "node": "^20.19.0 || ^22.13.0 || >=24"
+      }
+    },
+    "node_modules/@humanfs/core": {
+      "version": "0.19.1",
+      "resolved": "https://registry.npmjs.org/@humanfs/core/-/core-0.19.1.tgz",
+      "integrity": "sha512-5DyQ4+1JEUzejeK1JGICcideyfUbGixgS9jNgex5nqkW+cY7WZhxBigmieN5Qnw9ZosSNVC9KQKyb+GUaGyKUA==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "engines": {
+        "node": ">=18.18.0"
+      }
+    },
+    "node_modules/@humanfs/node": {
+      "version": "0.16.7",
+      "resolved": "https://registry.npmjs.org/@humanfs/node/-/node-0.16.7.tgz",
+      "integrity": "sha512-/zUx+yOsIrG4Y43Eh2peDeKCxlRt/gET6aHfaKpuq267qXdYDFViVHfMaLyygZOnl0kGWxFIgsBy8QFuTLUXEQ==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "dependencies": {
+        "@humanfs/core": "^0.19.1",
+        "@humanwhocodes/retry": "^0.4.0"
+      },
+      "engines": {
+        "node": ">=18.18.0"
+      }
+    },
+    "node_modules/@humanwhocodes/module-importer": {
+      "version": "1.0.1",
+      "resolved": "https://registry.npmjs.org/@humanwhocodes/module-importer/-/module-importer-1.0.1.tgz",
+      "integrity": "sha512-bxveV4V8v5Yb4ncFTT3rPSgZBOpCkjfK0y4oVVVJwIuDVBRMDXrPyXRL988i5ap9m9bnyEEjWfm5WkBmtffLfA==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "engines": {
+        "node": ">=12.22"
+      },
+      "funding": {
+        "type": "github",
+        "url": "https://github.com/sponsors/nzakas"
+      }
+    },
+    "node_modules/@humanwhocodes/retry": {
+      "version": "0.4.3",
+      "resolved": "https://registry.npmjs.org/@humanwhocodes/retry/-/retry-0.4.3.tgz",
+      "integrity": "sha512-bV0Tgo9K4hfPCek+aMAn81RppFKv2ySDQeMoSZuvTASywNTnVJCArCZE2FWqpvIatKu7VMRLWlR1EazvVhDyhQ==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "engines": {
+        "node": ">=18.18"
+      },
+      "funding": {
+        "type": "github",
+        "url": "https://github.com/sponsors/nzakas"
+      }
+    },
+    "node_modules/@isaacs/cliui": {
+      "version": "8.0.2",
+      "resolved": "https://registry.npmjs.org/@isaacs/cliui/-/cliui-8.0.2.tgz",
+      "integrity": "sha512-O8jcjabXaleOG9DQ0+ARXWZBTfnP4WNAqzuiJK7ll44AmxGKv/J2M4TPjxjY3znBCfvBXFzucm1twdyFybFqEA==",
+      "dev": true,
+      "license": "ISC",
+      "dependencies": {
+        "string-width": "^5.1.2",
+        "string-width-cjs": "npm:string-width@^4.2.0",
+        "strip-ansi": "^7.0.1",
+        "strip-ansi-cjs": "npm:strip-ansi@^6.0.1",
+        "wrap-ansi": "^8.1.0",
+        "wrap-ansi-cjs": "npm:wrap-ansi@^7.0.0"
+      },
+      "engines": {
+        "node": ">=12"
+      }
+    },
+    "node_modules/@isaacs/cliui/node_modules/ansi-regex": {
+      "version": "6.2.2",
+      "resolved": "https://registry.npmjs.org/ansi-regex/-/ansi-regex-6.2.2.tgz",
+      "integrity": "sha512-Bq3SmSpyFHaWjPk8If9yc6svM8c56dB5BAtW4Qbw5jHTwwXXcTLoRMkpDJp6VL0XzlWaCHTXrkFURMYmD0sLqg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=12"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/ansi-regex?sponsor=1"
+      }
+    },
+    "node_modules/@isaacs/cliui/node_modules/emoji-regex": {
+      "version": "9.2.2",
+      "resolved": "https://registry.npmjs.org/emoji-regex/-/emoji-regex-9.2.2.tgz",
+      "integrity": "sha512-L18DaJsXSUk2+42pv8mLs5jJT2hqFkFE4j21wOmgbUqsZ2hL72NsUU785g9RXgo3s0ZNgVl42TiHp3ZtOv/Vyg==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/@isaacs/cliui/node_modules/string-width": {
+      "version": "5.1.2",
+      "resolved": "https://registry.npmjs.org/string-width/-/string-width-5.1.2.tgz",
+      "integrity": "sha512-HnLOCR3vjcY8beoNLtcjZ5/nxn2afmME6lhrDrebokqMap+XbeW8n9TXpPDOqdGK5qcI3oT0GKTW6wC7EMiVqA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "eastasianwidth": "^0.2.0",
+        "emoji-regex": "^9.2.2",
+        "strip-ansi": "^7.0.1"
+      },
+      "engines": {
+        "node": ">=12"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/@isaacs/cliui/node_modules/strip-ansi": {
+      "version": "7.2.0",
+      "resolved": "https://registry.npmjs.org/strip-ansi/-/strip-ansi-7.2.0.tgz",
+      "integrity": "sha512-yDPMNjp4WyfYBkHnjIRLfca1i6KMyGCtsVgoKe/z1+6vukgaENdgGBZt+ZmKPc4gavvEZ5OgHfHdrazhgNyG7w==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "ansi-regex": "^6.2.2"
+      },
+      "engines": {
+        "node": ">=12"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/strip-ansi?sponsor=1"
+      }
+    },
+    "node_modules/@isaacs/cliui/node_modules/wrap-ansi": {
+      "version": "8.1.0",
+      "resolved": "https://registry.npmjs.org/wrap-ansi/-/wrap-ansi-8.1.0.tgz",
+      "integrity": "sha512-si7QWI6zUMq56bESFvagtmzMdGOtoxfR+Sez11Mobfc7tm+VkUckk9bW2UeffTGVUbOksxmSw0AA2gs8g71NCQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "ansi-styles": "^6.1.0",
+        "string-width": "^5.0.1",
+        "strip-ansi": "^7.0.1"
+      },
+      "engines": {
+        "node": ">=12"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/wrap-ansi?sponsor=1"
+      }
+    },
+    "node_modules/@istanbuljs/schema": {
+      "version": "0.1.3",
+      "resolved": "https://registry.npmjs.org/@istanbuljs/schema/-/schema-0.1.3.tgz",
+      "integrity": "sha512-ZXRY4jNvVgSVQ8DL3LTcakaAtXwTVUxE81hslsyD2AtoXW/wVob10HkOJ1X/pAlcI7D+2YoZKg5do8G/w6RYgA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/@jridgewell/gen-mapping": {
+      "version": "0.3.13",
+      "resolved": "https://registry.npmjs.org/@jridgewell/gen-mapping/-/gen-mapping-0.3.13.tgz",
+      "integrity": "sha512-2kkt/7niJ6MgEPxF0bYdQ6etZaA+fQvDcLKckhy1yIQOzaoKjBBjSj63/aLVjYE3qhRt5dvM+uUyfCg6UKCBbA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@jridgewell/sourcemap-codec": "^1.5.0",
+        "@jridgewell/trace-mapping": "^0.3.24"
+      }
+    },
+    "node_modules/@jridgewell/resolve-uri": {
+      "version": "3.1.2",
+      "resolved": "https://registry.npmjs.org/@jridgewell/resolve-uri/-/resolve-uri-3.1.2.tgz",
+      "integrity": "sha512-bRISgCIjP20/tbWSPWMEi54QVPRZExkuD9lJL+UIxUKtwVJA8wW1Trb1jMs1RFXo1CBTNZ/5hpC9QvmKWdopKw==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=6.0.0"
+      }
+    },
+    "node_modules/@jridgewell/sourcemap-codec": {
+      "version": "1.5.5",
+      "resolved": "https://registry.npmjs.org/@jridgewell/sourcemap-codec/-/sourcemap-codec-1.5.5.tgz",
+      "integrity": "sha512-cYQ9310grqxueWbl+WuIUIaiUaDcj7WOq5fVhEljNVgRfOUhY9fy2zTvfoqWsnebh8Sl70VScFbICvJnLKB0Og==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/@jridgewell/trace-mapping": {
+      "version": "0.3.31",
+      "resolved": "https://registry.npmjs.org/@jridgewell/trace-mapping/-/trace-mapping-0.3.31.tgz",
+      "integrity": "sha512-zzNR+SdQSDJzc8joaeP8QQoCQr8NuYx2dIIytl1QeBEZHJ9uW6hebsrYgbz8hJwUQao3TWCMtmfV8Nu1twOLAw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@jridgewell/resolve-uri": "^3.1.0",
+        "@jridgewell/sourcemap-codec": "^1.4.14"
+      }
+    },
+    "node_modules/@nodelib/fs.scandir": {
+      "version": "2.1.5",
+      "resolved": "https://registry.npmjs.org/@nodelib/fs.scandir/-/fs.scandir-2.1.5.tgz",
+      "integrity": "sha512-vq24Bq3ym5HEQm2NKCr3yXDwjc7vTsEThRDnkp2DK9p1uqLR+DHurm/NOTo0KG7HYHU7eppKZj3MyqYuMBf62g==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@nodelib/fs.stat": "2.0.5",
+        "run-parallel": "^1.1.9"
+      },
+      "engines": {
+        "node": ">= 8"
+      }
+    },
+    "node_modules/@nodelib/fs.stat": {
+      "version": "2.0.5",
+      "resolved": "https://registry.npmjs.org/@nodelib/fs.stat/-/fs.stat-2.0.5.tgz",
+      "integrity": "sha512-RkhPPp2zrqDAQA/2jNhnztcPAlv64XdhIp7a7454A5ovI7Bukxgt7MX7udwAu3zg1DcpPU0rz3VV1SeaqvY4+A==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">= 8"
+      }
+    },
+    "node_modules/@nodelib/fs.walk": {
+      "version": "1.2.8",
+      "resolved": "https://registry.npmjs.org/@nodelib/fs.walk/-/fs.walk-1.2.8.tgz",
+      "integrity": "sha512-oGB+UxlgWcgQkgwo8GcEGwemoTFt3FIO9ababBmaGwXIoBKZ+GTy0pP185beGg7Llih/NSHSV2XAs1lnznocSg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@nodelib/fs.scandir": "2.1.5",
+        "fastq": "^1.6.0"
+      },
+      "engines": {
+        "node": ">= 8"
+      }
+    },
+    "node_modules/@paddleocr/paddleocr-js": {
+      "resolved": "packages/core",
+      "link": true
+    },
+    "node_modules/@pkgjs/parseargs": {
+      "version": "0.11.0",
+      "resolved": "https://registry.npmjs.org/@pkgjs/parseargs/-/parseargs-0.11.0.tgz",
+      "integrity": "sha512-+1VkjdD0QBLPodGrJUeqarH8VAIvQODIbwh9XpP5Syisf7YoQgsJKPNFoqqLQlu+VQ/tVSshMR6loPMn8U+dPg==",
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "engines": {
+        "node": ">=14"
+      }
+    },
+    "node_modules/@protobufjs/aspromise": {
+      "version": "1.1.2",
+      "resolved": "https://registry.npmjs.org/@protobufjs/aspromise/-/aspromise-1.1.2.tgz",
+      "integrity": "sha512-j+gKExEuLmKwvz3OgROXtrJ2UG2x8Ch2YZUxahh+s1F2HZ+wAceUNLkvy6zKCPVRkU++ZWQrdxsUeQXmcg4uoQ==",
+      "license": "BSD-3-Clause"
+    },
+    "node_modules/@protobufjs/base64": {
+      "version": "1.1.2",
+      "resolved": "https://registry.npmjs.org/@protobufjs/base64/-/base64-1.1.2.tgz",
+      "integrity": "sha512-AZkcAA5vnN/v4PDqKyMR5lx7hZttPDgClv83E//FMNhR2TMcLUhfRUBHCmSl0oi9zMgDDqRUJkSxO3wm85+XLg==",
+      "license": "BSD-3-Clause"
+    },
+    "node_modules/@protobufjs/codegen": {
+      "version": "2.0.4",
+      "resolved": "https://registry.npmjs.org/@protobufjs/codegen/-/codegen-2.0.4.tgz",
+      "integrity": "sha512-YyFaikqM5sH0ziFZCN3xDC7zeGaB/d0IUb9CATugHWbd1FRFwWwt4ld4OYMPWu5a3Xe01mGAULCdqhMlPl29Jg==",
+      "license": "BSD-3-Clause"
+    },
+    "node_modules/@protobufjs/eventemitter": {
+      "version": "1.1.0",
+      "resolved": "https://registry.npmjs.org/@protobufjs/eventemitter/-/eventemitter-1.1.0.tgz",
+      "integrity": "sha512-j9ednRT81vYJ9OfVuXG6ERSTdEL1xVsNgqpkxMsbIabzSo3goCjDIveeGv5d03om39ML71RdmrGNjG5SReBP/Q==",
+      "license": "BSD-3-Clause"
+    },
+    "node_modules/@protobufjs/fetch": {
+      "version": "1.1.0",
+      "resolved": "https://registry.npmjs.org/@protobufjs/fetch/-/fetch-1.1.0.tgz",
+      "integrity": "sha512-lljVXpqXebpsijW71PZaCYeIcE5on1w5DlQy5WH6GLbFryLUrBD4932W/E2BSpfRJWseIL4v/KPgBFxDOIdKpQ==",
+      "license": "BSD-3-Clause",
+      "dependencies": {
+        "@protobufjs/aspromise": "^1.1.1",
+        "@protobufjs/inquire": "^1.1.0"
+      }
+    },
+    "node_modules/@protobufjs/float": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/@protobufjs/float/-/float-1.0.2.tgz",
+      "integrity": "sha512-Ddb+kVXlXst9d+R9PfTIxh1EdNkgoRe5tOX6t01f1lYWOvJnSPDBlG241QLzcyPdoNTsblLUdujGSE4RzrTZGQ==",
+      "license": "BSD-3-Clause"
+    },
+    "node_modules/@protobufjs/inquire": {
+      "version": "1.1.0",
+      "resolved": "https://registry.npmjs.org/@protobufjs/inquire/-/inquire-1.1.0.tgz",
+      "integrity": "sha512-kdSefcPdruJiFMVSbn801t4vFK7KB/5gd2fYvrxhuJYg8ILrmn9SKSX2tZdV6V+ksulWqS7aXjBcRXl3wHoD9Q==",
+      "license": "BSD-3-Clause"
+    },
+    "node_modules/@protobufjs/path": {
+      "version": "1.1.2",
+      "resolved": "https://registry.npmjs.org/@protobufjs/path/-/path-1.1.2.tgz",
+      "integrity": "sha512-6JOcJ5Tm08dOHAbdR3GrvP+yUUfkjG5ePsHYczMFLq3ZmMkAD98cDgcT2iA1lJ9NVwFd4tH/iSSoe44YWkltEA==",
+      "license": "BSD-3-Clause"
+    },
+    "node_modules/@protobufjs/pool": {
+      "version": "1.1.0",
+      "resolved": "https://registry.npmjs.org/@protobufjs/pool/-/pool-1.1.0.tgz",
+      "integrity": "sha512-0kELaGSIDBKvcgS4zkjz1PeddatrjYcmMWOlAuAPwAeccUrPHdUqo/J6LiymHHEiJT5NrF1UVwxY14f+fy4WQw==",
+      "license": "BSD-3-Clause"
+    },
+    "node_modules/@protobufjs/utf8": {
+      "version": "1.1.0",
+      "resolved": "https://registry.npmjs.org/@protobufjs/utf8/-/utf8-1.1.0.tgz",
+      "integrity": "sha512-Vvn3zZrhQZkkBE8LSuW3em98c0FwgO4nxzv6OdSxPKJIEKY2bGbHn+mhGIPerzI4twdxaP8/0+06HBpwf345Lw==",
+      "license": "BSD-3-Clause"
+    },
+    "node_modules/@rollup/rollup-android-arm-eabi": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-android-arm-eabi/-/rollup-android-arm-eabi-4.59.0.tgz",
+      "integrity": "sha512-upnNBkA6ZH2VKGcBj9Fyl9IGNPULcjXRlg0LLeaioQWueH30p6IXtJEbKAgvyv+mJaMxSm1l6xwDXYjpEMiLMg==",
+      "cpu": [
+        "arm"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "android"
+      ]
+    },
+    "node_modules/@rollup/rollup-android-arm64": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-android-arm64/-/rollup-android-arm64-4.59.0.tgz",
+      "integrity": "sha512-hZ+Zxj3SySm4A/DylsDKZAeVg0mvi++0PYVceVyX7hemkw7OreKdCvW2oQ3T1FMZvCaQXqOTHb8qmBShoqk69Q==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "android"
+      ]
+    },
+    "node_modules/@rollup/rollup-darwin-arm64": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-darwin-arm64/-/rollup-darwin-arm64-4.59.0.tgz",
+      "integrity": "sha512-W2Psnbh1J8ZJw0xKAd8zdNgF9HRLkdWwwdWqubSVk0pUuQkoHnv7rx4GiF9rT4t5DIZGAsConRE3AxCdJ4m8rg==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "darwin"
+      ]
+    },
+    "node_modules/@rollup/rollup-darwin-x64": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-darwin-x64/-/rollup-darwin-x64-4.59.0.tgz",
+      "integrity": "sha512-ZW2KkwlS4lwTv7ZVsYDiARfFCnSGhzYPdiOU4IM2fDbL+QGlyAbjgSFuqNRbSthybLbIJ915UtZBtmuLrQAT/w==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "darwin"
+      ]
+    },
+    "node_modules/@rollup/rollup-freebsd-arm64": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-freebsd-arm64/-/rollup-freebsd-arm64-4.59.0.tgz",
+      "integrity": "sha512-EsKaJ5ytAu9jI3lonzn3BgG8iRBjV4LxZexygcQbpiU0wU0ATxhNVEpXKfUa0pS05gTcSDMKpn3Sx+QB9RlTTA==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "freebsd"
+      ]
+    },
+    "node_modules/@rollup/rollup-freebsd-x64": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-freebsd-x64/-/rollup-freebsd-x64-4.59.0.tgz",
+      "integrity": "sha512-d3DuZi2KzTMjImrxoHIAODUZYoUUMsuUiY4SRRcJy6NJoZ6iIqWnJu9IScV9jXysyGMVuW+KNzZvBLOcpdl3Vg==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "freebsd"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-arm-gnueabihf": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-arm-gnueabihf/-/rollup-linux-arm-gnueabihf-4.59.0.tgz",
+      "integrity": "sha512-t4ONHboXi/3E0rT6OZl1pKbl2Vgxf9vJfWgmUoCEVQVxhW6Cw/c8I6hbbu7DAvgp82RKiH7TpLwxnJeKv2pbsw==",
+      "cpu": [
+        "arm"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-arm-musleabihf": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-arm-musleabihf/-/rollup-linux-arm-musleabihf-4.59.0.tgz",
+      "integrity": "sha512-CikFT7aYPA2ufMD086cVORBYGHffBo4K8MQ4uPS/ZnY54GKj36i196u8U+aDVT2LX4eSMbyHtyOh7D7Zvk2VvA==",
+      "cpu": [
+        "arm"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-arm64-gnu": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-arm64-gnu/-/rollup-linux-arm64-gnu-4.59.0.tgz",
+      "integrity": "sha512-jYgUGk5aLd1nUb1CtQ8E+t5JhLc9x5WdBKew9ZgAXg7DBk0ZHErLHdXM24rfX+bKrFe+Xp5YuJo54I5HFjGDAA==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-arm64-musl": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-arm64-musl/-/rollup-linux-arm64-musl-4.59.0.tgz",
+      "integrity": "sha512-peZRVEdnFWZ5Bh2KeumKG9ty7aCXzzEsHShOZEFiCQlDEepP1dpUl/SrUNXNg13UmZl+gzVDPsiCwnV1uI0RUA==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-loong64-gnu": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-loong64-gnu/-/rollup-linux-loong64-gnu-4.59.0.tgz",
+      "integrity": "sha512-gbUSW/97f7+r4gHy3Jlup8zDG190AuodsWnNiXErp9mT90iCy9NKKU0Xwx5k8VlRAIV2uU9CsMnEFg/xXaOfXg==",
+      "cpu": [
+        "loong64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-loong64-musl": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-loong64-musl/-/rollup-linux-loong64-musl-4.59.0.tgz",
+      "integrity": "sha512-yTRONe79E+o0FWFijasoTjtzG9EBedFXJMl888NBEDCDV9I2wGbFFfJQQe63OijbFCUZqxpHz1GzpbtSFikJ4Q==",
+      "cpu": [
+        "loong64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-ppc64-gnu": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-ppc64-gnu/-/rollup-linux-ppc64-gnu-4.59.0.tgz",
+      "integrity": "sha512-sw1o3tfyk12k3OEpRddF68a1unZ5VCN7zoTNtSn2KndUE+ea3m3ROOKRCZxEpmT9nsGnogpFP9x6mnLTCaoLkA==",
+      "cpu": [
+        "ppc64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-ppc64-musl": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-ppc64-musl/-/rollup-linux-ppc64-musl-4.59.0.tgz",
+      "integrity": "sha512-+2kLtQ4xT3AiIxkzFVFXfsmlZiG5FXYW7ZyIIvGA7Bdeuh9Z0aN4hVyXS/G1E9bTP/vqszNIN/pUKCk/BTHsKA==",
+      "cpu": [
+        "ppc64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-riscv64-gnu": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-riscv64-gnu/-/rollup-linux-riscv64-gnu-4.59.0.tgz",
+      "integrity": "sha512-NDYMpsXYJJaj+I7UdwIuHHNxXZ/b/N2hR15NyH3m2qAtb/hHPA4g4SuuvrdxetTdndfj9b1WOmy73kcPRoERUg==",
+      "cpu": [
+        "riscv64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-riscv64-musl": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-riscv64-musl/-/rollup-linux-riscv64-musl-4.59.0.tgz",
+      "integrity": "sha512-nLckB8WOqHIf1bhymk+oHxvM9D3tyPndZH8i8+35p/1YiVoVswPid2yLzgX7ZJP0KQvnkhM4H6QZ5m0LzbyIAg==",
+      "cpu": [
+        "riscv64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-s390x-gnu": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-s390x-gnu/-/rollup-linux-s390x-gnu-4.59.0.tgz",
+      "integrity": "sha512-oF87Ie3uAIvORFBpwnCvUzdeYUqi2wY6jRFWJAy1qus/udHFYIkplYRW+wo+GRUP4sKzYdmE1Y3+rY5Gc4ZO+w==",
+      "cpu": [
+        "s390x"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-x64-gnu": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-x64-gnu/-/rollup-linux-x64-gnu-4.59.0.tgz",
+      "integrity": "sha512-3AHmtQq/ppNuUspKAlvA8HtLybkDflkMuLK4DPo77DfthRb71V84/c4MlWJXixZz4uruIH4uaa07IqoAkG64fg==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-linux-x64-musl": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-linux-x64-musl/-/rollup-linux-x64-musl-4.59.0.tgz",
+      "integrity": "sha512-2UdiwS/9cTAx7qIUZB/fWtToJwvt0Vbo0zmnYt7ED35KPg13Q0ym1g442THLC7VyI6JfYTP4PiSOWyoMdV2/xg==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ]
+    },
+    "node_modules/@rollup/rollup-openbsd-x64": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-openbsd-x64/-/rollup-openbsd-x64-4.59.0.tgz",
+      "integrity": "sha512-M3bLRAVk6GOwFlPTIxVBSYKUaqfLrn8l0psKinkCFxl4lQvOSz8ZrKDz2gxcBwHFpci0B6rttydI4IpS4IS/jQ==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "openbsd"
+      ]
+    },
+    "node_modules/@rollup/rollup-openharmony-arm64": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-openharmony-arm64/-/rollup-openharmony-arm64-4.59.0.tgz",
+      "integrity": "sha512-tt9KBJqaqp5i5HUZzoafHZX8b5Q2Fe7UjYERADll83O4fGqJ49O1FsL6LpdzVFQcpwvnyd0i+K/VSwu/o/nWlA==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "openharmony"
+      ]
+    },
+    "node_modules/@rollup/rollup-win32-arm64-msvc": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-win32-arm64-msvc/-/rollup-win32-arm64-msvc-4.59.0.tgz",
+      "integrity": "sha512-V5B6mG7OrGTwnxaNUzZTDTjDS7F75PO1ae6MJYdiMu60sq0CqN5CVeVsbhPxalupvTX8gXVSU9gq+Rx1/hvu6A==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ]
+    },
+    "node_modules/@rollup/rollup-win32-ia32-msvc": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-win32-ia32-msvc/-/rollup-win32-ia32-msvc-4.59.0.tgz",
+      "integrity": "sha512-UKFMHPuM9R0iBegwzKF4y0C4J9u8C6MEJgFuXTBerMk7EJ92GFVFYBfOZaSGLu6COf7FxpQNqhNS4c4icUPqxA==",
+      "cpu": [
+        "ia32"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ]
+    },
+    "node_modules/@rollup/rollup-win32-x64-gnu": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-win32-x64-gnu/-/rollup-win32-x64-gnu-4.59.0.tgz",
+      "integrity": "sha512-laBkYlSS1n2L8fSo1thDNGrCTQMmxjYY5G0WFWjFFYZkKPjsMBsgJfGf4TLxXrF6RyhI60L8TMOjBMvXiTcxeA==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ]
+    },
+    "node_modules/@rollup/rollup-win32-x64-msvc": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/@rollup/rollup-win32-x64-msvc/-/rollup-win32-x64-msvc-4.59.0.tgz",
+      "integrity": "sha512-2HRCml6OztYXyJXAvdDXPKcawukWY2GpR5/nxKp4iBgiO3wcoEGkAaqctIbZcNB6KlUQBIqt8VYkNSj2397EfA==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ]
+    },
+    "node_modules/@techstark/opencv-js": {
+      "version": "4.10.0-release.1",
+      "resolved": "https://registry.npmjs.org/@techstark/opencv-js/-/opencv-js-4.10.0-release.1.tgz",
+      "integrity": "sha512-S4XELidRiQeA0q1s9VQLo540wCxUo24r1O4C+LqZ6llX+sPCXvZCPv3Ice8dEIr0uavyZ8YZeKXSBdDgMXSXjw==",
+      "license": "Apache-2.0"
+    },
+    "node_modules/@ts-morph/common": {
+      "version": "0.13.0",
+      "resolved": "https://registry.npmjs.org/@ts-morph/common/-/common-0.13.0.tgz",
+      "integrity": "sha512-fEJ6j7Cu8yiWjA4UmybOBH9Efgb/64ZTWuvCF4KysGu4xz8ettfyaqFt8WZ1btCxXsGZJjZ2/3svOF6rL+UFdQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "fast-glob": "^3.2.11",
+        "minimatch": "^5.0.1",
+        "mkdirp": "^1.0.4",
+        "path-browserify": "^1.0.1"
+      }
+    },
+    "node_modules/@ts-morph/common/node_modules/balanced-match": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/balanced-match/-/balanced-match-1.0.2.tgz",
+      "integrity": "sha512-3oSeUO0TMV67hN1AmbXsK4yaqU7tjiHlbxRDZOpH0KW9+CeX4bRAaX0Anxt0tx2MrpRpWwQaPwIlISEJhYU5Pw==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/@ts-morph/common/node_modules/brace-expansion": {
+      "version": "2.0.3",
+      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-2.0.3.tgz",
+      "integrity": "sha512-MCV/fYJEbqx68aE58kv2cA/kiky1G8vux3OR6/jbS+jIMe/6fJWa0DTzJU7dqijOWYwHi1t29FlfYI9uytqlpA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "balanced-match": "^1.0.0"
+      }
+    },
+    "node_modules/@ts-morph/common/node_modules/minimatch": {
+      "version": "5.1.9",
+      "resolved": "https://registry.npmjs.org/minimatch/-/minimatch-5.1.9.tgz",
+      "integrity": "sha512-7o1wEA2RyMP7Iu7GNba9vc0RWWGACJOCZBJX2GJWip0ikV+wcOsgVuY9uE8CPiyQhkGFSlhuSkZPavN7u1c2Fw==",
+      "dev": true,
+      "license": "ISC",
+      "dependencies": {
+        "brace-expansion": "^2.0.1"
+      },
+      "engines": {
+        "node": ">=10"
+      }
+    },
+    "node_modules/@types/chai": {
+      "version": "5.2.3",
+      "resolved": "https://registry.npmjs.org/@types/chai/-/chai-5.2.3.tgz",
+      "integrity": "sha512-Mw558oeA9fFbv65/y4mHtXDs9bPnFMZAL/jxdPFUpOHHIXX91mcgEHbS5Lahr+pwZFR8A7GQleRWeI6cGFC2UA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@types/deep-eql": "*",
+        "assertion-error": "^2.0.1"
+      }
+    },
+    "node_modules/@types/deep-eql": {
+      "version": "4.0.2",
+      "resolved": "https://registry.npmjs.org/@types/deep-eql/-/deep-eql-4.0.2.tgz",
+      "integrity": "sha512-c9h9dVVMigMPc4bwTvC5dxqtqJZwQPePsWjPlpSOnojbor6pGqdk541lfA7AqFQr5pB1BRdq0juY9db81BwyFw==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/@types/esrecurse": {
+      "version": "4.3.1",
+      "resolved": "https://registry.npmjs.org/@types/esrecurse/-/esrecurse-4.3.1.tgz",
+      "integrity": "sha512-xJBAbDifo5hpffDBuHl0Y8ywswbiAp/Wi7Y/GtAgSlZyIABppyurxVueOPE8LUQOxdlgi6Zqce7uoEpqNTeiUw==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/@types/estree": {
+      "version": "1.0.8",
+      "resolved": "https://registry.npmjs.org/@types/estree/-/estree-1.0.8.tgz",
+      "integrity": "sha512-dWHzHa2WqEXI/O1E9OjrocMTKJl2mSrEolh1Iomrv6U+JuNwaHXsXx9bLu5gG7BUWFIN0skIQJQ/L1rIex4X6w==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/@types/js-yaml": {
+      "version": "4.0.9",
+      "resolved": "https://registry.npmjs.org/@types/js-yaml/-/js-yaml-4.0.9.tgz",
+      "integrity": "sha512-k4MGaQl5TGo/iipqb2UDG2UwjXziSWkh0uysQelTlJpX1qGlpUZYm8PnO4DxG1qBomtJUdYJ6qR6xdIah10JLg==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/@types/json-schema": {
+      "version": "7.0.15",
+      "resolved": "https://registry.npmjs.org/@types/json-schema/-/json-schema-7.0.15.tgz",
+      "integrity": "sha512-5+fP8P8MFNC+AyZCDxrB2pkZFPGzqQWUzpSeuuVLvm8VMcorNYavBqoFcxK8bQz4Qsbn4oUEEem4wDLfcysGHA==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/@types/node": {
+      "version": "25.5.0",
+      "resolved": "https://registry.npmjs.org/@types/node/-/node-25.5.0.tgz",
+      "integrity": "sha512-jp2P3tQMSxWugkCUKLRPVUpGaL5MVFwF8RDuSRztfwgN1wmqJeMSbKlnEtQqU8UrhTmzEmZdu2I6v2dpp7XIxw==",
+      "license": "MIT",
+      "dependencies": {
+        "undici-types": "~7.18.0"
+      }
+    },
+    "node_modules/@typescript-eslint/eslint-plugin": {
+      "version": "8.57.2",
+      "resolved": "https://registry.npmjs.org/@typescript-eslint/eslint-plugin/-/eslint-plugin-8.57.2.tgz",
+      "integrity": "sha512-NZZgp0Fm2IkD+La5PR81sd+g+8oS6JwJje+aRWsDocxHkjyRw0J5L5ZTlN3LI1LlOcGL7ph3eaIUmTXMIjLk0w==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@eslint-community/regexpp": "^4.12.2",
+        "@typescript-eslint/scope-manager": "8.57.2",
+        "@typescript-eslint/type-utils": "8.57.2",
+        "@typescript-eslint/utils": "8.57.2",
+        "@typescript-eslint/visitor-keys": "8.57.2",
+        "ignore": "^7.0.5",
+        "natural-compare": "^1.4.0",
+        "ts-api-utils": "^2.4.0"
+      },
+      "engines": {
+        "node": "^18.18.0 || ^20.9.0 || >=21.1.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/typescript-eslint"
+      },
+      "peerDependencies": {
+        "@typescript-eslint/parser": "^8.57.2",
+        "eslint": "^8.57.0 || ^9.0.0 || ^10.0.0",
+        "typescript": ">=4.8.4 <6.0.0"
+      }
+    },
+    "node_modules/@typescript-eslint/eslint-plugin/node_modules/ignore": {
+      "version": "7.0.5",
+      "resolved": "https://registry.npmjs.org/ignore/-/ignore-7.0.5.tgz",
+      "integrity": "sha512-Hs59xBNfUIunMFgWAbGX5cq6893IbWg4KnrjbYwX3tx0ztorVgTDA6B2sxf8ejHJ4wz8BqGUMYlnzNBer5NvGg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">= 4"
+      }
+    },
+    "node_modules/@typescript-eslint/parser": {
+      "version": "8.57.2",
+      "resolved": "https://registry.npmjs.org/@typescript-eslint/parser/-/parser-8.57.2.tgz",
+      "integrity": "sha512-30ScMRHIAD33JJQkgfGW1t8CURZtjc2JpTrq5n2HFhOefbAhb7ucc7xJwdWcrEtqUIYJ73Nybpsggii6GtAHjA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@typescript-eslint/scope-manager": "8.57.2",
+        "@typescript-eslint/types": "8.57.2",
+        "@typescript-eslint/typescript-estree": "8.57.2",
+        "@typescript-eslint/visitor-keys": "8.57.2",
+        "debug": "^4.4.3"
+      },
+      "engines": {
+        "node": "^18.18.0 || ^20.9.0 || >=21.1.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/typescript-eslint"
+      },
+      "peerDependencies": {
+        "eslint": "^8.57.0 || ^9.0.0 || ^10.0.0",
+        "typescript": ">=4.8.4 <6.0.0"
+      }
+    },
+    "node_modules/@typescript-eslint/project-service": {
+      "version": "8.57.2",
+      "resolved": "https://registry.npmjs.org/@typescript-eslint/project-service/-/project-service-8.57.2.tgz",
+      "integrity": "sha512-FuH0wipFywXRTHf+bTTjNyuNQQsQC3qh/dYzaM4I4W0jrCqjCVuUh99+xd9KamUfmCGPvbO8NDngo/vsnNVqgw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@typescript-eslint/tsconfig-utils": "^8.57.2",
+        "@typescript-eslint/types": "^8.57.2",
+        "debug": "^4.4.3"
+      },
+      "engines": {
+        "node": "^18.18.0 || ^20.9.0 || >=21.1.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/typescript-eslint"
+      },
+      "peerDependencies": {
+        "typescript": ">=4.8.4 <6.0.0"
+      }
+    },
+    "node_modules/@typescript-eslint/scope-manager": {
+      "version": "8.57.2",
+      "resolved": "https://registry.npmjs.org/@typescript-eslint/scope-manager/-/scope-manager-8.57.2.tgz",
+      "integrity": "sha512-snZKH+W4WbWkrBqj4gUNRIGb/jipDW3qMqVJ4C9rzdFc+wLwruxk+2a5D+uoFcKPAqyqEnSb4l2ULuZf95eSkw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@typescript-eslint/types": "8.57.2",
+        "@typescript-eslint/visitor-keys": "8.57.2"
+      },
+      "engines": {
+        "node": "^18.18.0 || ^20.9.0 || >=21.1.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/typescript-eslint"
+      }
+    },
+    "node_modules/@typescript-eslint/tsconfig-utils": {
+      "version": "8.57.2",
+      "resolved": "https://registry.npmjs.org/@typescript-eslint/tsconfig-utils/-/tsconfig-utils-8.57.2.tgz",
+      "integrity": "sha512-3Lm5DSM+DCowsUOJC+YqHHnKEfFh5CoGkj5Z31NQSNF4l5wdOwqGn99wmwN/LImhfY3KJnmordBq/4+VDe2eKw==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": "^18.18.0 || ^20.9.0 || >=21.1.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/typescript-eslint"
+      },
+      "peerDependencies": {
+        "typescript": ">=4.8.4 <6.0.0"
+      }
+    },
+    "node_modules/@typescript-eslint/type-utils": {
+      "version": "8.57.2",
+      "resolved": "https://registry.npmjs.org/@typescript-eslint/type-utils/-/type-utils-8.57.2.tgz",
+      "integrity": "sha512-Co6ZCShm6kIbAM/s+oYVpKFfW7LBc6FXoPXjTRQ449PPNBY8U0KZXuevz5IFuuUj2H9ss40atTaf9dlGLzbWZg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@typescript-eslint/types": "8.57.2",
+        "@typescript-eslint/typescript-estree": "8.57.2",
+        "@typescript-eslint/utils": "8.57.2",
+        "debug": "^4.4.3",
+        "ts-api-utils": "^2.4.0"
+      },
+      "engines": {
+        "node": "^18.18.0 || ^20.9.0 || >=21.1.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/typescript-eslint"
+      },
+      "peerDependencies": {
+        "eslint": "^8.57.0 || ^9.0.0 || ^10.0.0",
+        "typescript": ">=4.8.4 <6.0.0"
+      }
+    },
+    "node_modules/@typescript-eslint/types": {
+      "version": "8.57.2",
+      "resolved": "https://registry.npmjs.org/@typescript-eslint/types/-/types-8.57.2.tgz",
+      "integrity": "sha512-/iZM6FnM4tnx9csuTxspMW4BOSegshwX5oBDznJ7S4WggL7Vczz5d2W11ecc4vRrQMQHXRSxzrCsyG5EsPPTbA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": "^18.18.0 || ^20.9.0 || >=21.1.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/typescript-eslint"
+      }
+    },
+    "node_modules/@typescript-eslint/typescript-estree": {
+      "version": "8.57.2",
+      "resolved": "https://registry.npmjs.org/@typescript-eslint/typescript-estree/-/typescript-estree-8.57.2.tgz",
+      "integrity": "sha512-2MKM+I6g8tJxfSmFKOnHv2t8Sk3T6rF20A1Puk0svLK+uVapDZB/4pfAeB7nE83uAZrU6OxW+HmOd5wHVdXwXA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@typescript-eslint/project-service": "8.57.2",
+        "@typescript-eslint/tsconfig-utils": "8.57.2",
+        "@typescript-eslint/types": "8.57.2",
+        "@typescript-eslint/visitor-keys": "8.57.2",
+        "debug": "^4.4.3",
+        "minimatch": "^10.2.2",
+        "semver": "^7.7.3",
+        "tinyglobby": "^0.2.15",
+        "ts-api-utils": "^2.4.0"
+      },
+      "engines": {
+        "node": "^18.18.0 || ^20.9.0 || >=21.1.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/typescript-eslint"
+      },
+      "peerDependencies": {
+        "typescript": ">=4.8.4 <6.0.0"
+      }
+    },
+    "node_modules/@typescript-eslint/utils": {
+      "version": "8.57.2",
+      "resolved": "https://registry.npmjs.org/@typescript-eslint/utils/-/utils-8.57.2.tgz",
+      "integrity": "sha512-krRIbvPK1ju1WBKIefiX+bngPs+odIQUtR7kymzPfo1POVw3jlF+nLkmexdSSd4UCbDcQn+wMBATOOmpBbqgKg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@eslint-community/eslint-utils": "^4.9.1",
+        "@typescript-eslint/scope-manager": "8.57.2",
+        "@typescript-eslint/types": "8.57.2",
+        "@typescript-eslint/typescript-estree": "8.57.2"
+      },
+      "engines": {
+        "node": "^18.18.0 || ^20.9.0 || >=21.1.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/typescript-eslint"
+      },
+      "peerDependencies": {
+        "eslint": "^8.57.0 || ^9.0.0 || ^10.0.0",
+        "typescript": ">=4.8.4 <6.0.0"
+      }
+    },
+    "node_modules/@typescript-eslint/visitor-keys": {
+      "version": "8.57.2",
+      "resolved": "https://registry.npmjs.org/@typescript-eslint/visitor-keys/-/visitor-keys-8.57.2.tgz",
+      "integrity": "sha512-zhahknjobV2FiD6Ee9iLbS7OV9zi10rG26odsQdfBO/hjSzUQbkIYgda+iNKK1zNiW2ey+Lf8MU5btN17V3dUw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@typescript-eslint/types": "8.57.2",
+        "eslint-visitor-keys": "^5.0.0"
+      },
+      "engines": {
+        "node": "^18.18.0 || ^20.9.0 || >=21.1.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/typescript-eslint"
+      }
+    },
+    "node_modules/@vitest/coverage-v8": {
+      "version": "3.2.4",
+      "resolved": "https://registry.npmjs.org/@vitest/coverage-v8/-/coverage-v8-3.2.4.tgz",
+      "integrity": "sha512-EyF9SXU6kS5Ku/U82E259WSnvg6c8KTjppUncuNdm5QHpe17mwREHnjDzozC8x9MZ0xfBUFSaLkRv4TMA75ALQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@ampproject/remapping": "^2.3.0",
+        "@bcoe/v8-coverage": "^1.0.2",
+        "ast-v8-to-istanbul": "^0.3.3",
+        "debug": "^4.4.1",
+        "istanbul-lib-coverage": "^3.2.2",
+        "istanbul-lib-report": "^3.0.1",
+        "istanbul-lib-source-maps": "^5.0.6",
+        "istanbul-reports": "^3.1.7",
+        "magic-string": "^0.30.17",
+        "magicast": "^0.3.5",
+        "std-env": "^3.9.0",
+        "test-exclude": "^7.0.1",
+        "tinyrainbow": "^2.0.0"
+      },
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      },
+      "peerDependencies": {
+        "@vitest/browser": "3.2.4",
+        "vitest": "3.2.4"
+      },
+      "peerDependenciesMeta": {
+        "@vitest/browser": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/@vitest/expect": {
+      "version": "3.2.4",
+      "resolved": "https://registry.npmjs.org/@vitest/expect/-/expect-3.2.4.tgz",
+      "integrity": "sha512-Io0yyORnB6sikFlt8QW5K7slY4OjqNX9jmJQ02QDda8lyM6B5oNgVWoSoKPac8/kgnCUzuHQKrSLtu/uOqqrig==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@types/chai": "^5.2.2",
+        "@vitest/spy": "3.2.4",
+        "@vitest/utils": "3.2.4",
+        "chai": "^5.2.0",
+        "tinyrainbow": "^2.0.0"
+      },
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      }
+    },
+    "node_modules/@vitest/mocker": {
+      "version": "3.2.4",
+      "resolved": "https://registry.npmjs.org/@vitest/mocker/-/mocker-3.2.4.tgz",
+      "integrity": "sha512-46ryTE9RZO/rfDd7pEqFl7etuyzekzEhUbTW3BvmeO/BcCMEgq59BKhek3dXDWgAj4oMK6OZi+vRr1wPW6qjEQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@vitest/spy": "3.2.4",
+        "estree-walker": "^3.0.3",
+        "magic-string": "^0.30.17"
+      },
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      },
+      "peerDependencies": {
+        "msw": "^2.4.9",
+        "vite": "^5.0.0 || ^6.0.0 || ^7.0.0-0"
+      },
+      "peerDependenciesMeta": {
+        "msw": {
+          "optional": true
+        },
+        "vite": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/@vitest/pretty-format": {
+      "version": "3.2.4",
+      "resolved": "https://registry.npmjs.org/@vitest/pretty-format/-/pretty-format-3.2.4.tgz",
+      "integrity": "sha512-IVNZik8IVRJRTr9fxlitMKeJeXFFFN0JaB9PHPGQ8NKQbGpfjlTx9zO4RefN8gp7eqjNy8nyK3NZmBzOPeIxtA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "tinyrainbow": "^2.0.0"
+      },
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      }
+    },
+    "node_modules/@vitest/runner": {
+      "version": "3.2.4",
+      "resolved": "https://registry.npmjs.org/@vitest/runner/-/runner-3.2.4.tgz",
+      "integrity": "sha512-oukfKT9Mk41LreEW09vt45f8wx7DordoWUZMYdY/cyAk7w5TWkTRCNZYF7sX7n2wB7jyGAl74OxgwhPgKaqDMQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@vitest/utils": "3.2.4",
+        "pathe": "^2.0.3",
+        "strip-literal": "^3.0.0"
+      },
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      }
+    },
+    "node_modules/@vitest/snapshot": {
+      "version": "3.2.4",
+      "resolved": "https://registry.npmjs.org/@vitest/snapshot/-/snapshot-3.2.4.tgz",
+      "integrity": "sha512-dEYtS7qQP2CjU27QBC5oUOxLE/v5eLkGqPE0ZKEIDGMs4vKWe7IjgLOeauHsR0D5YuuycGRO5oSRXnwnmA78fQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@vitest/pretty-format": "3.2.4",
+        "magic-string": "^0.30.17",
+        "pathe": "^2.0.3"
+      },
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      }
+    },
+    "node_modules/@vitest/spy": {
+      "version": "3.2.4",
+      "resolved": "https://registry.npmjs.org/@vitest/spy/-/spy-3.2.4.tgz",
+      "integrity": "sha512-vAfasCOe6AIK70iP5UD11Ac4siNUNJ9i/9PZ3NKx07sG6sUxeag1LWdNrMWeKKYBLlzuK+Gn65Yd5nyL6ds+nw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "tinyspy": "^4.0.3"
+      },
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      }
+    },
+    "node_modules/@vitest/utils": {
+      "version": "3.2.4",
+      "resolved": "https://registry.npmjs.org/@vitest/utils/-/utils-3.2.4.tgz",
+      "integrity": "sha512-fB2V0JFrQSMsCo9HiSq3Ezpdv4iYaXRG1Sx8edX3MwxfyNn83mKiGzOcH+Fkxt4MHxr3y42fQi1oeAInqgX2QA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@vitest/pretty-format": "3.2.4",
+        "loupe": "^3.1.4",
+        "tinyrainbow": "^2.0.0"
+      },
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      }
+    },
+    "node_modules/acorn": {
+      "version": "8.16.0",
+      "resolved": "https://registry.npmjs.org/acorn/-/acorn-8.16.0.tgz",
+      "integrity": "sha512-UVJyE9MttOsBQIDKw1skb9nAwQuR5wuGD3+82K6JgJlm/Y+KI92oNsMNGZCYdDsVtRHSak0pcV5Dno5+4jh9sw==",
+      "dev": true,
+      "license": "MIT",
+      "bin": {
+        "acorn": "bin/acorn"
+      },
+      "engines": {
+        "node": ">=0.4.0"
+      }
+    },
+    "node_modules/acorn-jsx": {
+      "version": "5.3.2",
+      "resolved": "https://registry.npmjs.org/acorn-jsx/-/acorn-jsx-5.3.2.tgz",
+      "integrity": "sha512-rq9s+JNhf0IChjtDXxllJ7g41oZk5SlXtp0LHwyA5cejwn7vKmKp4pPri6YEePv2PU65sAsegbXtIinmDFDXgQ==",
+      "dev": true,
+      "license": "MIT",
+      "peerDependencies": {
+        "acorn": "^6.0.0 || ^7.0.0 || ^8.0.0"
+      }
+    },
+    "node_modules/agent-base": {
+      "version": "7.1.4",
+      "resolved": "https://registry.npmjs.org/agent-base/-/agent-base-7.1.4.tgz",
+      "integrity": "sha512-MnA+YT8fwfJPgBx3m60MNqakm30XOkyIoH1y6huTQvC0PwZG7ki8NacLBcrPbNoo8vEZy7Jpuk7+jMO+CUovTQ==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">= 14"
+      }
+    },
+    "node_modules/ajv": {
+      "version": "6.14.0",
+      "resolved": "https://registry.npmjs.org/ajv/-/ajv-6.14.0.tgz",
+      "integrity": "sha512-IWrosm/yrn43eiKqkfkHis7QioDleaXQHdDVPKg0FSwwd/DuvyX79TZnFOnYpB7dcsFAMmtFztZuXPDvSePkFw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "fast-deep-equal": "^3.1.1",
+        "fast-json-stable-stringify": "^2.0.0",
+        "json-schema-traverse": "^0.4.1",
+        "uri-js": "^4.2.2"
+      },
+      "funding": {
+        "type": "github",
+        "url": "https://github.com/sponsors/epoberezkin"
+      }
+    },
+    "node_modules/ansi-escapes": {
+      "version": "7.3.0",
+      "resolved": "https://registry.npmjs.org/ansi-escapes/-/ansi-escapes-7.3.0.tgz",
+      "integrity": "sha512-BvU8nYgGQBxcmMuEeUEmNTvrMVjJNSH7RgW24vXexN4Ven6qCvy4TntnvlnwnMLTVlcRQQdbRY8NKnaIoeWDNg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "environment": "^1.0.0"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/ansi-regex": {
+      "version": "5.0.1",
+      "resolved": "https://registry.npmjs.org/ansi-regex/-/ansi-regex-5.0.1.tgz",
+      "integrity": "sha512-quJQXlTSUGL2LH9SUXo8VwsY4soanhgo6LNSm84E1LBcE8s3O0wpdiRzyR9z/ZZJMlMWv37qOOb9pdJlMUEKFQ==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/ansi-styles": {
+      "version": "6.2.3",
+      "resolved": "https://registry.npmjs.org/ansi-styles/-/ansi-styles-6.2.3.tgz",
+      "integrity": "sha512-4Dj6M28JB+oAH8kFkTLUo+a2jwOFkuqb3yucU0CANcRRUbxS0cP0nZYCGjcc3BNXwRIsUVmDGgzawme7zvJHvg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=12"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/ansi-styles?sponsor=1"
+      }
+    },
+    "node_modules/argparse": {
+      "version": "2.0.1",
+      "resolved": "https://registry.npmjs.org/argparse/-/argparse-2.0.1.tgz",
+      "integrity": "sha512-8+9WqebbFzpX9OR+Wa6O29asIogeRMzcGtAINdpMHHyAg10f05aSFVBbcEqGf/PXw1EjAZ+q2/bEBg3DvurK3Q==",
+      "license": "Python-2.0"
+    },
+    "node_modules/assertion-error": {
+      "version": "2.0.1",
+      "resolved": "https://registry.npmjs.org/assertion-error/-/assertion-error-2.0.1.tgz",
+      "integrity": "sha512-Izi8RQcffqCeNVgFigKli1ssklIbpHnCYc6AknXGYoB6grJqyeby7jv12JUQgmTAnIDnbck1uxksT4dzN3PWBA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=12"
+      }
+    },
+    "node_modules/ast-v8-to-istanbul": {
+      "version": "0.3.12",
+      "resolved": "https://registry.npmjs.org/ast-v8-to-istanbul/-/ast-v8-to-istanbul-0.3.12.tgz",
+      "integrity": "sha512-BRRC8VRZY2R4Z4lFIL35MwNXmwVqBityvOIwETtsCSwvjl0IdgFsy9NhdaA6j74nUdtJJlIypeRhpDam19Wq3g==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@jridgewell/trace-mapping": "^0.3.31",
+        "estree-walker": "^3.0.3",
+        "js-tokens": "^10.0.0"
+      }
+    },
+    "node_modules/ast-v8-to-istanbul/node_modules/js-tokens": {
+      "version": "10.0.0",
+      "resolved": "https://registry.npmjs.org/js-tokens/-/js-tokens-10.0.0.tgz",
+      "integrity": "sha512-lM/UBzQmfJRo9ABXbPWemivdCW8V2G8FHaHdypQaIy523snUjog0W71ayWXTjiR+ixeMyVHN2XcpnTd/liPg/Q==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/balanced-match": {
+      "version": "4.0.4",
+      "resolved": "https://registry.npmjs.org/balanced-match/-/balanced-match-4.0.4.tgz",
+      "integrity": "sha512-BLrgEcRTwX2o6gGxGOCNyMvGSp35YofuYzw9h1IMTRmKqttAZZVU67bdb9Pr2vUHA8+j3i2tJfjO6C6+4myGTA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": "18 || 20 || >=22"
+      }
+    },
+    "node_modules/brace-expansion": {
+      "version": "5.0.5",
+      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-5.0.5.tgz",
+      "integrity": "sha512-VZznLgtwhn+Mact9tfiwx64fA9erHH/MCXEUfB/0bX/6Fz6ny5EGTXYltMocqg4xFAQZtnO3DHWWXi8RiuN7cQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "balanced-match": "^4.0.2"
+      },
+      "engines": {
+        "node": "18 || 20 || >=22"
+      }
+    },
+    "node_modules/braces": {
+      "version": "3.0.3",
+      "resolved": "https://registry.npmjs.org/braces/-/braces-3.0.3.tgz",
+      "integrity": "sha512-yQbXgO/OSZVD2IsiLlro+7Hf6Q18EJrKSEsdoMzKePKXct3gvD8oLcOQdIzGupr5Fj+EDe8gO/lxc1BzfMpxvA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "fill-range": "^7.1.1"
+      },
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/cac": {
+      "version": "6.7.14",
+      "resolved": "https://registry.npmjs.org/cac/-/cac-6.7.14.tgz",
+      "integrity": "sha512-b6Ilus+c3RrdDk+JhLKUAQfzzgLEPy6wcXqS7f/xe1EETvsDP6GORG7SFuOs6cID5YkqchW/LXZbX5bc8j7ZcQ==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/chai": {
+      "version": "5.3.3",
+      "resolved": "https://registry.npmjs.org/chai/-/chai-5.3.3.tgz",
+      "integrity": "sha512-4zNhdJD/iOjSH0A05ea+Ke6MU5mmpQcbQsSOkgdaUMJ9zTlDTD/GYlwohmIE2u0gaxHYiVHEn1Fw9mZ/ktJWgw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "assertion-error": "^2.0.1",
+        "check-error": "^2.1.1",
+        "deep-eql": "^5.0.1",
+        "loupe": "^3.1.0",
+        "pathval": "^2.0.0"
+      },
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/chalk": {
+      "version": "4.1.2",
+      "resolved": "https://registry.npmjs.org/chalk/-/chalk-4.1.2.tgz",
+      "integrity": "sha512-oKnbhFyRIXpUuez8iBMmyEa4nbj4IOQyuhc/wy9kY7/WVPcwIO9VA668Pu8RkO7+0G76SLROeyw9CpQ061i4mA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "ansi-styles": "^4.1.0",
+        "supports-color": "^7.1.0"
+      },
+      "engines": {
+        "node": ">=10"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/chalk?sponsor=1"
+      }
+    },
+    "node_modules/chalk/node_modules/ansi-styles": {
+      "version": "4.3.0",
+      "resolved": "https://registry.npmjs.org/ansi-styles/-/ansi-styles-4.3.0.tgz",
+      "integrity": "sha512-zbB9rCJAT1rbjiVDb2hqKFHNYLxgtk8NURxZ3IZwD3F6NtxbXZQCnnSi1Lkx+IDohdPlFp222wVALIheZJQSEg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "color-convert": "^2.0.1"
+      },
+      "engines": {
+        "node": ">=8"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/ansi-styles?sponsor=1"
+      }
+    },
+    "node_modules/check-error": {
+      "version": "2.1.3",
+      "resolved": "https://registry.npmjs.org/check-error/-/check-error-2.1.3.tgz",
+      "integrity": "sha512-PAJdDJusoxnwm1VwW07VWwUN1sl7smmC3OKggvndJFadxxDRyFJBX/ggnu/KE4kQAB7a3Dp8f/YXC1FlUprWmA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">= 16"
+      }
+    },
+    "node_modules/cli-cursor": {
+      "version": "5.0.0",
+      "resolved": "https://registry.npmjs.org/cli-cursor/-/cli-cursor-5.0.0.tgz",
+      "integrity": "sha512-aCj4O5wKyszjMmDT4tZj93kxyydN/K5zPWSCe6/0AV/AA1pqe5ZBIw0a2ZfPQV7lL5/yb5HsUreJ6UFAF1tEQw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "restore-cursor": "^5.0.0"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/cli-truncate": {
+      "version": "5.2.0",
+      "resolved": "https://registry.npmjs.org/cli-truncate/-/cli-truncate-5.2.0.tgz",
+      "integrity": "sha512-xRwvIOMGrfOAnM1JYtqQImuaNtDEv9v6oIYAs4LIHwTiKee8uwvIi363igssOC0O5U04i4AlENs79LQLu9tEMw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "slice-ansi": "^8.0.0",
+        "string-width": "^8.2.0"
+      },
+      "engines": {
+        "node": ">=20"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/clipper-lib": {
+      "version": "6.4.2",
+      "resolved": "https://registry.npmjs.org/clipper-lib/-/clipper-lib-6.4.2.tgz",
+      "integrity": "sha512-knglhjQX5ihNj/XCIs6zCHrTemdvHY3LPZP9XB2nq2/3igyYMFueFXtfp84baJvEE+f8pO1ZS4UVeEgmLnAprQ==",
+      "license": "BSL"
+    },
+    "node_modules/code-block-writer": {
+      "version": "11.0.3",
+      "resolved": "https://registry.npmjs.org/code-block-writer/-/code-block-writer-11.0.3.tgz",
+      "integrity": "sha512-NiujjUFB4SwScJq2bwbYUtXbZhBSlY6vYzm++3Q6oC+U+injTqfPYFK8wS9COOmb2lueqp0ZRB4nK1VYeHgNyw==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/color-convert": {
+      "version": "2.0.1",
+      "resolved": "https://registry.npmjs.org/color-convert/-/color-convert-2.0.1.tgz",
+      "integrity": "sha512-RRECPsj7iu/xb5oKYcsFHSppFNnsj/52OVTRKb4zP5onXwVF3zVmmToNcOfGC+CRDpfK/U584fMg38ZHCaElKQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "color-name": "~1.1.4"
+      },
+      "engines": {
+        "node": ">=7.0.0"
+      }
+    },
+    "node_modules/color-name": {
+      "version": "1.1.4",
+      "resolved": "https://registry.npmjs.org/color-name/-/color-name-1.1.4.tgz",
+      "integrity": "sha512-dOy+3AuW3a2wNbZHIuMZpTcgjGuLU/uBL/ubcZF9OXbDo8ff4O8yVp5Bf0efS8uEoYo5q4Fx7dY9OgQGXgAsQA==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/colorette": {
+      "version": "2.0.20",
+      "resolved": "https://registry.npmjs.org/colorette/-/colorette-2.0.20.tgz",
+      "integrity": "sha512-IfEDxwoWIjkeXL1eXcDiow4UbKjhLdq6/EuSVR9GMN7KVH3r9gQ83e73hsz1Nd1T3ijd5xv1wcWRYO+D6kCI2w==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/commander": {
+      "version": "14.0.3",
+      "resolved": "https://registry.npmjs.org/commander/-/commander-14.0.3.tgz",
+      "integrity": "sha512-H+y0Jo/T1RZ9qPP4Eh1pkcQcLRglraJaSLoyOtHxu6AapkjWVCy2Sit1QQ4x3Dng8qDlSsZEet7g5Pq06MvTgw==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=20"
+      }
+    },
+    "node_modules/cross-spawn": {
+      "version": "7.0.6",
+      "resolved": "https://registry.npmjs.org/cross-spawn/-/cross-spawn-7.0.6.tgz",
+      "integrity": "sha512-uV2QOWP2nWzsy2aMp8aRibhi9dlzF5Hgh5SHaB9OiTGEyDTiJJyx0uy51QXdyWbtAHNua4XJzUKca3OzKUd3vA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "path-key": "^3.1.0",
+        "shebang-command": "^2.0.0",
+        "which": "^2.0.1"
+      },
+      "engines": {
+        "node": ">= 8"
+      }
+    },
+    "node_modules/cssstyle": {
+      "version": "4.6.0",
+      "resolved": "https://registry.npmjs.org/cssstyle/-/cssstyle-4.6.0.tgz",
+      "integrity": "sha512-2z+rWdzbbSZv6/rhtvzvqeZQHrBaqgogqt85sqFNbabZOuFbCVFb8kPeEtZjiKkbrm395irpNKiYeFeLiQnFPg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@asamuzakjp/css-color": "^3.2.0",
+        "rrweb-cssom": "^0.8.0"
+      },
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/data-urls": {
+      "version": "5.0.0",
+      "resolved": "https://registry.npmjs.org/data-urls/-/data-urls-5.0.0.tgz",
+      "integrity": "sha512-ZYP5VBHshaDAiVZxjbRVcFJpc+4xGgT0bK3vzy1HLN8jTO975HEbuYzZJcHoQEY5K1a0z8YayJkyVETa08eNTg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "whatwg-mimetype": "^4.0.0",
+        "whatwg-url": "^14.0.0"
+      },
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/debug": {
+      "version": "4.4.3",
+      "resolved": "https://registry.npmjs.org/debug/-/debug-4.4.3.tgz",
+      "integrity": "sha512-RGwwWnwQvkVfavKVt22FGLw+xYSdzARwm0ru6DhTVA3umU5hZc28V3kO4stgYryrTlLpuvgI9GiijltAjNbcqA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "ms": "^2.1.3"
+      },
+      "engines": {
+        "node": ">=6.0"
+      },
+      "peerDependenciesMeta": {
+        "supports-color": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/decimal.js": {
+      "version": "10.6.0",
+      "resolved": "https://registry.npmjs.org/decimal.js/-/decimal.js-10.6.0.tgz",
+      "integrity": "sha512-YpgQiITW3JXGntzdUmyUR1V812Hn8T1YVXhCu+wO3OpS4eU9l4YdD3qjyiKdV6mvV29zapkMeD390UVEf2lkUg==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/deep-eql": {
+      "version": "5.0.2",
+      "resolved": "https://registry.npmjs.org/deep-eql/-/deep-eql-5.0.2.tgz",
+      "integrity": "sha512-h5k/5U50IJJFpzfL6nO9jaaumfjO/f2NjK/oYB2Djzm4p9L+3T9qWpZqZ2hAbLPuuYq9wrU08WQyBTL5GbPk5Q==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=6"
+      }
+    },
+    "node_modules/deep-is": {
+      "version": "0.1.4",
+      "resolved": "https://registry.npmjs.org/deep-is/-/deep-is-0.1.4.tgz",
+      "integrity": "sha512-oIPzksmTg4/MriiaYGO+okXDT7ztn/w3Eptv/+gSIdMdKsJo0u4CfYNFJPy+4SKMuCqGw2wxnA+URMg3t8a/bQ==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/demo": {
+      "resolved": "apps/demo",
+      "link": true
+    },
+    "node_modules/eastasianwidth": {
+      "version": "0.2.0",
+      "resolved": "https://registry.npmjs.org/eastasianwidth/-/eastasianwidth-0.2.0.tgz",
+      "integrity": "sha512-I88TYZWc9XiYHRQ4/3c5rjjfgkjhLyW2luGIheGERbNQ6OY7yTybanSpDXZa8y7VUP9YmDcYa+eyq4ca7iLqWA==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/emoji-regex": {
+      "version": "10.6.0",
+      "resolved": "https://registry.npmjs.org/emoji-regex/-/emoji-regex-10.6.0.tgz",
+      "integrity": "sha512-toUI84YS5YmxW219erniWD0CIVOo46xGKColeNQRgOzDorgBi1v4D71/OFzgD9GO2UGKIv1C3Sp8DAn0+j5w7A==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/entities": {
+      "version": "6.0.1",
+      "resolved": "https://registry.npmjs.org/entities/-/entities-6.0.1.tgz",
+      "integrity": "sha512-aN97NXWF6AWBTahfVOIrB/NShkzi5H7F9r1s9mD3cDj4Ko5f2qhhVoYMibXF7GlLveb/D2ioWay8lxI97Ven3g==",
+      "dev": true,
+      "license": "BSD-2-Clause",
+      "engines": {
+        "node": ">=0.12"
+      },
+      "funding": {
+        "url": "https://github.com/fb55/entities?sponsor=1"
+      }
+    },
+    "node_modules/environment": {
+      "version": "1.1.0",
+      "resolved": "https://registry.npmjs.org/environment/-/environment-1.1.0.tgz",
+      "integrity": "sha512-xUtoPkMggbz0MPyPiIWr1Kp4aeWJjDZ6SMvURhimjdZgsRuDplF5/s9hcgGhyXMhs+6vpnuoiZ2kFiu3FMnS8Q==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=18"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/es-module-lexer": {
+      "version": "1.7.0",
+      "resolved": "https://registry.npmjs.org/es-module-lexer/-/es-module-lexer-1.7.0.tgz",
+      "integrity": "sha512-jEQoCwk8hyb2AZziIOLhDqpm5+2ww5uIE6lkO/6jcOCusfk6LhMHpXXfBLXTZ7Ydyt0j4VoUQv6uGNYbdW+kBA==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/esbuild": {
+      "version": "0.25.12",
+      "resolved": "https://registry.npmjs.org/esbuild/-/esbuild-0.25.12.tgz",
+      "integrity": "sha512-bbPBYYrtZbkt6Os6FiTLCTFxvq4tt3JKall1vRwshA3fdVztsLAatFaZobhkBC8/BrPetoa0oksYoKXoG4ryJg==",
+      "dev": true,
+      "hasInstallScript": true,
+      "license": "MIT",
+      "bin": {
+        "esbuild": "bin/esbuild"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "optionalDependencies": {
+        "@esbuild/aix-ppc64": "0.25.12",
+        "@esbuild/android-arm": "0.25.12",
+        "@esbuild/android-arm64": "0.25.12",
+        "@esbuild/android-x64": "0.25.12",
+        "@esbuild/darwin-arm64": "0.25.12",
+        "@esbuild/darwin-x64": "0.25.12",
+        "@esbuild/freebsd-arm64": "0.25.12",
+        "@esbuild/freebsd-x64": "0.25.12",
+        "@esbuild/linux-arm": "0.25.12",
+        "@esbuild/linux-arm64": "0.25.12",
+        "@esbuild/linux-ia32": "0.25.12",
+        "@esbuild/linux-loong64": "0.25.12",
+        "@esbuild/linux-mips64el": "0.25.12",
+        "@esbuild/linux-ppc64": "0.25.12",
+        "@esbuild/linux-riscv64": "0.25.12",
+        "@esbuild/linux-s390x": "0.25.12",
+        "@esbuild/linux-x64": "0.25.12",
+        "@esbuild/netbsd-arm64": "0.25.12",
+        "@esbuild/netbsd-x64": "0.25.12",
+        "@esbuild/openbsd-arm64": "0.25.12",
+        "@esbuild/openbsd-x64": "0.25.12",
+        "@esbuild/openharmony-arm64": "0.25.12",
+        "@esbuild/sunos-x64": "0.25.12",
+        "@esbuild/win32-arm64": "0.25.12",
+        "@esbuild/win32-ia32": "0.25.12",
+        "@esbuild/win32-x64": "0.25.12"
+      }
+    },
+    "node_modules/escape-string-regexp": {
+      "version": "4.0.0",
+      "resolved": "https://registry.npmjs.org/escape-string-regexp/-/escape-string-regexp-4.0.0.tgz",
+      "integrity": "sha512-TtpcNJ3XAzx3Gq8sWRzJaVajRs0uVxA2YAkdb1jm2YkPz4G6egUFAyA3n5vtEIZefPk5Wa4UXbKuS5fKkJWdgA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=10"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/eslint": {
+      "version": "10.0.3",
+      "resolved": "https://registry.npmjs.org/eslint/-/eslint-10.0.3.tgz",
+      "integrity": "sha512-COV33RzXZkqhG9P2rZCFl9ZmJ7WL+gQSCRzE7RhkbclbQPtLAWReL7ysA0Sh4c8Im2U9ynybdR56PV0XcKvqaQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@eslint-community/eslint-utils": "^4.8.0",
+        "@eslint-community/regexpp": "^4.12.2",
+        "@eslint/config-array": "^0.23.3",
+        "@eslint/config-helpers": "^0.5.2",
+        "@eslint/core": "^1.1.1",
+        "@eslint/plugin-kit": "^0.6.1",
+        "@humanfs/node": "^0.16.6",
+        "@humanwhocodes/module-importer": "^1.0.1",
+        "@humanwhocodes/retry": "^0.4.2",
+        "@types/estree": "^1.0.6",
+        "ajv": "^6.14.0",
+        "cross-spawn": "^7.0.6",
+        "debug": "^4.3.2",
+        "escape-string-regexp": "^4.0.0",
+        "eslint-scope": "^9.1.2",
+        "eslint-visitor-keys": "^5.0.1",
+        "espree": "^11.1.1",
+        "esquery": "^1.7.0",
+        "esutils": "^2.0.2",
+        "fast-deep-equal": "^3.1.3",
+        "file-entry-cache": "^8.0.0",
+        "find-up": "^5.0.0",
+        "glob-parent": "^6.0.2",
+        "ignore": "^5.2.0",
+        "imurmurhash": "^0.1.4",
+        "is-glob": "^4.0.0",
+        "json-stable-stringify-without-jsonify": "^1.0.1",
+        "minimatch": "^10.2.4",
+        "natural-compare": "^1.4.0",
+        "optionator": "^0.9.3"
+      },
+      "bin": {
+        "eslint": "bin/eslint.js"
+      },
+      "engines": {
+        "node": "^20.19.0 || ^22.13.0 || >=24"
+      },
+      "funding": {
+        "url": "https://eslint.org/donate"
+      },
+      "peerDependencies": {
+        "jiti": "*"
+      },
+      "peerDependenciesMeta": {
+        "jiti": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/eslint-scope": {
+      "version": "9.1.2",
+      "resolved": "https://registry.npmjs.org/eslint-scope/-/eslint-scope-9.1.2.tgz",
+      "integrity": "sha512-xS90H51cKw0jltxmvmHy2Iai1LIqrfbw57b79w/J7MfvDfkIkFZ+kj6zC3BjtUwh150HsSSdxXZcsuv72miDFQ==",
+      "dev": true,
+      "license": "BSD-2-Clause",
+      "dependencies": {
+        "@types/esrecurse": "^4.3.1",
+        "@types/estree": "^1.0.8",
+        "esrecurse": "^4.3.0",
+        "estraverse": "^5.2.0"
+      },
+      "engines": {
+        "node": "^20.19.0 || ^22.13.0 || >=24"
+      },
+      "funding": {
+        "url": "https://opencollective.com/eslint"
+      }
+    },
+    "node_modules/eslint-visitor-keys": {
+      "version": "5.0.1",
+      "resolved": "https://registry.npmjs.org/eslint-visitor-keys/-/eslint-visitor-keys-5.0.1.tgz",
+      "integrity": "sha512-tD40eHxA35h0PEIZNeIjkHoDR4YjjJp34biM0mDvplBe//mB+IHCqHDGV7pxF+7MklTvighcCPPZC7ynWyjdTA==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "engines": {
+        "node": "^20.19.0 || ^22.13.0 || >=24"
+      },
+      "funding": {
+        "url": "https://opencollective.com/eslint"
+      }
+    },
+    "node_modules/eslint/node_modules/find-up": {
+      "version": "5.0.0",
+      "resolved": "https://registry.npmjs.org/find-up/-/find-up-5.0.0.tgz",
+      "integrity": "sha512-78/PXT1wlLLDgTzDs7sjq9hzz0vXD+zn+7wypEe4fXQxCmdmqfGsEPQxmiCSQI3ajFV91bVSsvNtrJRiW6nGng==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "locate-path": "^6.0.0",
+        "path-exists": "^4.0.0"
+      },
+      "engines": {
+        "node": ">=10"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/eslint/node_modules/locate-path": {
+      "version": "6.0.0",
+      "resolved": "https://registry.npmjs.org/locate-path/-/locate-path-6.0.0.tgz",
+      "integrity": "sha512-iPZK6eYjbxRu3uB4/WZ3EsEIMJFMqAoopl3R+zuq0UjcAm/MO6KCweDgPfP3elTztoKP3KtnVHxTn2NHBSDVUw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "p-locate": "^5.0.0"
+      },
+      "engines": {
+        "node": ">=10"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/eslint/node_modules/p-limit": {
+      "version": "3.1.0",
+      "resolved": "https://registry.npmjs.org/p-limit/-/p-limit-3.1.0.tgz",
+      "integrity": "sha512-TYOanM3wGwNGsZN2cVTYPArw454xnXj5qmWF1bEoAc4+cU/ol7GVh7odevjp1FNHduHc3KZMcFduxU5Xc6uJRQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "yocto-queue": "^0.1.0"
+      },
+      "engines": {
+        "node": ">=10"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/eslint/node_modules/p-locate": {
+      "version": "5.0.0",
+      "resolved": "https://registry.npmjs.org/p-locate/-/p-locate-5.0.0.tgz",
+      "integrity": "sha512-LaNjtRWUBY++zB5nE/NwcaoMylSPk+S+ZHNB1TzdbMJMny6dynpAGt7X/tl/QYq3TIeE6nxHppbo2LGymrG5Pw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "p-limit": "^3.0.2"
+      },
+      "engines": {
+        "node": ">=10"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/espree": {
+      "version": "11.2.0",
+      "resolved": "https://registry.npmjs.org/espree/-/espree-11.2.0.tgz",
+      "integrity": "sha512-7p3DrVEIopW1B1avAGLuCSh1jubc01H2JHc8B4qqGblmg5gI9yumBgACjWo4JlIc04ufug4xJ3SQI8HkS/Rgzw==",
+      "dev": true,
+      "license": "BSD-2-Clause",
+      "dependencies": {
+        "acorn": "^8.16.0",
+        "acorn-jsx": "^5.3.2",
+        "eslint-visitor-keys": "^5.0.1"
+      },
+      "engines": {
+        "node": "^20.19.0 || ^22.13.0 || >=24"
+      },
+      "funding": {
+        "url": "https://opencollective.com/eslint"
+      }
+    },
+    "node_modules/esquery": {
+      "version": "1.7.0",
+      "resolved": "https://registry.npmjs.org/esquery/-/esquery-1.7.0.tgz",
+      "integrity": "sha512-Ap6G0WQwcU/LHsvLwON1fAQX9Zp0A2Y6Y/cJBl9r/JbW90Zyg4/zbG6zzKa2OTALELarYHmKu0GhpM5EO+7T0g==",
+      "dev": true,
+      "license": "BSD-3-Clause",
+      "dependencies": {
+        "estraverse": "^5.1.0"
+      },
+      "engines": {
+        "node": ">=0.10"
+      }
+    },
+    "node_modules/esrecurse": {
+      "version": "4.3.0",
+      "resolved": "https://registry.npmjs.org/esrecurse/-/esrecurse-4.3.0.tgz",
+      "integrity": "sha512-KmfKL3b6G+RXvP8N1vr3Tq1kL/oCFgn2NYXEtqP8/L3pKapUA4G8cFVaoF3SU323CD4XypR/ffioHmkti6/Tag==",
+      "dev": true,
+      "license": "BSD-2-Clause",
+      "dependencies": {
+        "estraverse": "^5.2.0"
+      },
+      "engines": {
+        "node": ">=4.0"
+      }
+    },
+    "node_modules/estraverse": {
+      "version": "5.3.0",
+      "resolved": "https://registry.npmjs.org/estraverse/-/estraverse-5.3.0.tgz",
+      "integrity": "sha512-MMdARuVEQziNTeJD8DgMqmhwR11BRQ/cBP+pLtYdSTnf3MIO8fFeiINEbX36ZdNlfU/7A9f3gUw49B3oQsvwBA==",
+      "dev": true,
+      "license": "BSD-2-Clause",
+      "engines": {
+        "node": ">=4.0"
+      }
+    },
+    "node_modules/estree-walker": {
+      "version": "3.0.3",
+      "resolved": "https://registry.npmjs.org/estree-walker/-/estree-walker-3.0.3.tgz",
+      "integrity": "sha512-7RUKfXgSMMkzt6ZuXmqapOurLGPPfgj6l9uRZ7lRGolvk0y2yocc35LdcxKC5PQZdn2DMqioAQ2NoWcrTKmm6g==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@types/estree": "^1.0.0"
+      }
+    },
+    "node_modules/esutils": {
+      "version": "2.0.3",
+      "resolved": "https://registry.npmjs.org/esutils/-/esutils-2.0.3.tgz",
+      "integrity": "sha512-kVscqXk4OCp68SZ0dkgEKVi6/8ij300KBWTJq32P/dYeWTSwK41WyTxalN1eRmA5Z9UU/LX9D7FWSmV9SAYx6g==",
+      "dev": true,
+      "license": "BSD-2-Clause",
+      "engines": {
+        "node": ">=0.10.0"
+      }
+    },
+    "node_modules/eventemitter3": {
+      "version": "5.0.4",
+      "resolved": "https://registry.npmjs.org/eventemitter3/-/eventemitter3-5.0.4.tgz",
+      "integrity": "sha512-mlsTRyGaPBjPedk6Bvw+aqbsXDtoAyAzm5MO7JgU+yVRyMQ5O8bD4Kcci7BS85f93veegeCPkL8R4GLClnjLFw==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/expect-type": {
+      "version": "1.3.0",
+      "resolved": "https://registry.npmjs.org/expect-type/-/expect-type-1.3.0.tgz",
+      "integrity": "sha512-knvyeauYhqjOYvQ66MznSMs83wmHrCycNEN6Ao+2AeYEfxUIkuiVxdEa1qlGEPK+We3n0THiDciYSsCcgW/DoA==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "engines": {
+        "node": ">=12.0.0"
+      }
+    },
+    "node_modules/fast-deep-equal": {
+      "version": "3.1.3",
+      "resolved": "https://registry.npmjs.org/fast-deep-equal/-/fast-deep-equal-3.1.3.tgz",
+      "integrity": "sha512-f3qQ9oQy9j2AhBe/H9VC91wLmKBCCU/gDOnKNAYG5hswO7BLKj09Hc5HYNz9cGI++xlpDCIgDaitVs03ATR84Q==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/fast-glob": {
+      "version": "3.3.3",
+      "resolved": "https://registry.npmjs.org/fast-glob/-/fast-glob-3.3.3.tgz",
+      "integrity": "sha512-7MptL8U0cqcFdzIzwOTHoilX9x5BrNqye7Z/LuC7kCMRio1EMSyqRK3BEAUD7sXRq4iT4AzTVuZdhgQ2TCvYLg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@nodelib/fs.stat": "^2.0.2",
+        "@nodelib/fs.walk": "^1.2.3",
+        "glob-parent": "^5.1.2",
+        "merge2": "^1.3.0",
+        "micromatch": "^4.0.8"
+      },
+      "engines": {
+        "node": ">=8.6.0"
+      }
+    },
+    "node_modules/fast-glob/node_modules/glob-parent": {
+      "version": "5.1.2",
+      "resolved": "https://registry.npmjs.org/glob-parent/-/glob-parent-5.1.2.tgz",
+      "integrity": "sha512-AOIgSQCepiJYwP3ARnGx+5VnTu2HBYdzbGP45eLw1vr3zB3vZLeyed1sC9hnbcOc9/SrMyM5RPQrkGz4aS9Zow==",
+      "dev": true,
+      "license": "ISC",
+      "dependencies": {
+        "is-glob": "^4.0.1"
+      },
+      "engines": {
+        "node": ">= 6"
+      }
+    },
+    "node_modules/fast-json-stable-stringify": {
+      "version": "2.1.0",
+      "resolved": "https://registry.npmjs.org/fast-json-stable-stringify/-/fast-json-stable-stringify-2.1.0.tgz",
+      "integrity": "sha512-lhd/wF+Lk98HZoTCtlVraHtfh5XYijIjalXck7saUtuanSDyLMxnHhSXEDJqHxD7msR8D0uCmqlkwjCV8xvwHw==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/fast-levenshtein": {
+      "version": "2.0.6",
+      "resolved": "https://registry.npmjs.org/fast-levenshtein/-/fast-levenshtein-2.0.6.tgz",
+      "integrity": "sha512-DCXu6Ifhqcks7TZKY3Hxp3y6qphY5SJZmrWMDrKcERSOXWQdMhU9Ig/PYrzyw/ul9jOIyh0N4M0tbC5hodg8dw==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/fastq": {
+      "version": "1.20.1",
+      "resolved": "https://registry.npmjs.org/fastq/-/fastq-1.20.1.tgz",
+      "integrity": "sha512-GGToxJ/w1x32s/D2EKND7kTil4n8OVk/9mycTc4VDza13lOvpUZTGX3mFSCtV9ksdGBVzvsyAVLM6mHFThxXxw==",
+      "dev": true,
+      "license": "ISC",
+      "dependencies": {
+        "reusify": "^1.0.4"
+      }
+    },
+    "node_modules/fdir": {
+      "version": "6.5.0",
+      "resolved": "https://registry.npmjs.org/fdir/-/fdir-6.5.0.tgz",
+      "integrity": "sha512-tIbYtZbucOs0BRGqPJkshJUYdL+SDH7dVM8gjy+ERp3WAUjLEFJE+02kanyHtwjWOnwrKYBiwAmM0p4kLJAnXg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=12.0.0"
+      },
+      "peerDependencies": {
+        "picomatch": "^3 || ^4"
+      },
+      "peerDependenciesMeta": {
+        "picomatch": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/file-entry-cache": {
+      "version": "8.0.0",
+      "resolved": "https://registry.npmjs.org/file-entry-cache/-/file-entry-cache-8.0.0.tgz",
+      "integrity": "sha512-XXTUwCvisa5oacNGRP9SfNtYBNAMi+RPwBFmblZEF7N7swHYQS6/Zfk7SRwx4D5j3CH211YNRco1DEMNVfZCnQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "flat-cache": "^4.0.0"
+      },
+      "engines": {
+        "node": ">=16.0.0"
+      }
+    },
+    "node_modules/fill-range": {
+      "version": "7.1.1",
+      "resolved": "https://registry.npmjs.org/fill-range/-/fill-range-7.1.1.tgz",
+      "integrity": "sha512-YsGpe3WHLK8ZYi4tWDg2Jy3ebRz2rXowDxnld4bkQB00cc/1Zw9AWnC0i9ztDJitivtQvaI9KaLyKrc+hBW0yg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "to-regex-range": "^5.0.1"
+      },
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/flat-cache": {
+      "version": "4.0.1",
+      "resolved": "https://registry.npmjs.org/flat-cache/-/flat-cache-4.0.1.tgz",
+      "integrity": "sha512-f7ccFPK3SXFHpx15UIGyRJ/FJQctuKZ0zVuN3frBo4HnK3cay9VEW0R6yPYFHC0AgqhukPzKjq22t5DmAyqGyw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "flatted": "^3.2.9",
+        "keyv": "^4.5.4"
+      },
+      "engines": {
+        "node": ">=16"
+      }
+    },
+    "node_modules/flatbuffers": {
+      "version": "25.9.23",
+      "resolved": "https://registry.npmjs.org/flatbuffers/-/flatbuffers-25.9.23.tgz",
+      "integrity": "sha512-MI1qs7Lo4Syw0EOzUl0xjs2lsoeqFku44KpngfIduHBYvzm8h2+7K8YMQh1JtVVVrUvhLpNwqVi4DERegUJhPQ==",
+      "license": "Apache-2.0"
+    },
+    "node_modules/flatted": {
+      "version": "3.4.2",
+      "resolved": "https://registry.npmjs.org/flatted/-/flatted-3.4.2.tgz",
+      "integrity": "sha512-PjDse7RzhcPkIJwy5t7KPWQSZ9cAbzQXcafsetQoD7sOJRQlGikNbx7yZp2OotDnJyrDcbyRq3Ttb18iYOqkxA==",
+      "dev": true,
+      "license": "ISC"
+    },
+    "node_modules/foreground-child": {
+      "version": "3.3.1",
+      "resolved": "https://registry.npmjs.org/foreground-child/-/foreground-child-3.3.1.tgz",
+      "integrity": "sha512-gIXjKqtFuWEgzFRJA9WCQeSJLZDjgJUOMCMzxtvFq/37KojM1BFGufqsCy0r4qSQmYLsZYMeyRqzIWOMup03sw==",
+      "dev": true,
+      "license": "ISC",
+      "dependencies": {
+        "cross-spawn": "^7.0.6",
+        "signal-exit": "^4.0.1"
+      },
+      "engines": {
+        "node": ">=14"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/isaacs"
+      }
+    },
+    "node_modules/fsevents": {
+      "version": "2.3.3",
+      "resolved": "https://registry.npmjs.org/fsevents/-/fsevents-2.3.3.tgz",
+      "integrity": "sha512-5xoDfX+fL7faATnagmWPpbFtwh/R77WmMMqqHGS65C3vvB0YHrgF+B1YmZ3441tMj5n63k0212XNoJwzlhffQw==",
+      "dev": true,
+      "hasInstallScript": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "darwin"
+      ],
+      "engines": {
+        "node": "^8.16.0 || ^10.6.0 || >=11.0.0"
+      }
+    },
+    "node_modules/get-east-asian-width": {
+      "version": "1.5.0",
+      "resolved": "https://registry.npmjs.org/get-east-asian-width/-/get-east-asian-width-1.5.0.tgz",
+      "integrity": "sha512-CQ+bEO+Tva/qlmw24dCejulK5pMzVnUOFOijVogd3KQs07HnRIgp8TGipvCCRT06xeYEbpbgwaCxglFyiuIcmA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=18"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/glob": {
+      "version": "10.5.0",
+      "resolved": "https://registry.npmjs.org/glob/-/glob-10.5.0.tgz",
+      "integrity": "sha512-DfXN8DfhJ7NH3Oe7cFmu3NCu1wKbkReJ8TorzSAFbSKrlNaQSKfIzqYqVY8zlbs2NLBbWpRiU52GX2PbaBVNkg==",
+      "deprecated": "Old versions of glob are not supported, and contain widely publicized security vulnerabilities, which have been fixed in the current version. Please update. Support for old versions may be purchased (at exorbitant rates) by contacting i@izs.me",
+      "dev": true,
+      "license": "ISC",
+      "dependencies": {
+        "foreground-child": "^3.1.0",
+        "jackspeak": "^3.1.2",
+        "minimatch": "^9.0.4",
+        "minipass": "^7.1.2",
+        "package-json-from-dist": "^1.0.0",
+        "path-scurry": "^1.11.1"
+      },
+      "bin": {
+        "glob": "dist/esm/bin.mjs"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/isaacs"
+      }
+    },
+    "node_modules/glob-parent": {
+      "version": "6.0.2",
+      "resolved": "https://registry.npmjs.org/glob-parent/-/glob-parent-6.0.2.tgz",
+      "integrity": "sha512-XxwI8EOhVQgWp6iDL+3b0r86f4d6AX6zSU55HfB4ydCEuXLXc5FcYeOu+nnGftS4TEju/11rt4KJPTMgbfmv4A==",
+      "dev": true,
+      "license": "ISC",
+      "dependencies": {
+        "is-glob": "^4.0.3"
+      },
+      "engines": {
+        "node": ">=10.13.0"
+      }
+    },
+    "node_modules/glob/node_modules/balanced-match": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/balanced-match/-/balanced-match-1.0.2.tgz",
+      "integrity": "sha512-3oSeUO0TMV67hN1AmbXsK4yaqU7tjiHlbxRDZOpH0KW9+CeX4bRAaX0Anxt0tx2MrpRpWwQaPwIlISEJhYU5Pw==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/glob/node_modules/brace-expansion": {
+      "version": "2.0.3",
+      "resolved": "https://registry.npmjs.org/brace-expansion/-/brace-expansion-2.0.3.tgz",
+      "integrity": "sha512-MCV/fYJEbqx68aE58kv2cA/kiky1G8vux3OR6/jbS+jIMe/6fJWa0DTzJU7dqijOWYwHi1t29FlfYI9uytqlpA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "balanced-match": "^1.0.0"
+      }
+    },
+    "node_modules/glob/node_modules/minimatch": {
+      "version": "9.0.9",
+      "resolved": "https://registry.npmjs.org/minimatch/-/minimatch-9.0.9.tgz",
+      "integrity": "sha512-OBwBN9AL4dqmETlpS2zasx+vTeWclWzkblfZk7KTA5j3jeOONz/tRCnZomUyvNg83wL5Zv9Ss6HMJXAgL8R2Yg==",
+      "dev": true,
+      "license": "ISC",
+      "dependencies": {
+        "brace-expansion": "^2.0.2"
+      },
+      "engines": {
+        "node": ">=16 || 14 >=14.17"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/isaacs"
+      }
+    },
+    "node_modules/globals": {
+      "version": "17.4.0",
+      "resolved": "https://registry.npmjs.org/globals/-/globals-17.4.0.tgz",
+      "integrity": "sha512-hjrNztw/VajQwOLsMNT1cbJiH2muO3OROCHnbehc8eY5JyD2gqz4AcMHPqgaOR59DjgUjYAYLeH699g/eWi2jw==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=18"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/graceful-fs": {
+      "version": "4.2.11",
+      "resolved": "https://registry.npmjs.org/graceful-fs/-/graceful-fs-4.2.11.tgz",
+      "integrity": "sha512-RbJ5/jmFcNNCcDV5o9eTnBLJ/HszWV0P73bc+Ff4nS/rJj+YaS6IGyiOL0VoBYX+l1Wrl3k63h/KrH+nhJ0XvQ==",
+      "dev": true,
+      "license": "ISC"
+    },
+    "node_modules/guid-typescript": {
+      "version": "1.0.9",
+      "resolved": "https://registry.npmjs.org/guid-typescript/-/guid-typescript-1.0.9.tgz",
+      "integrity": "sha512-Y8T4vYhEfwJOTbouREvG+3XDsjr8E3kIr7uf+JZ0BYloFsttiHU0WfvANVsR7TxNUJa/WpCnw/Ino/p+DeBhBQ==",
+      "license": "ISC"
+    },
+    "node_modules/has-flag": {
+      "version": "4.0.0",
+      "resolved": "https://registry.npmjs.org/has-flag/-/has-flag-4.0.0.tgz",
+      "integrity": "sha512-EykJT/Q1KjTWctppgIAgfSO0tKVuZUjhgMr17kqTumMl6Afv3EISleU7qZUzoXDFTAHTDC4NOoG/ZxU3EvlMPQ==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/html-encoding-sniffer": {
+      "version": "4.0.0",
+      "resolved": "https://registry.npmjs.org/html-encoding-sniffer/-/html-encoding-sniffer-4.0.0.tgz",
+      "integrity": "sha512-Y22oTqIU4uuPgEemfz7NDJz6OeKf12Lsu+QC+s3BVpda64lTiMYCyGwg5ki4vFxkMwQdeZDl2adZoqUgdFuTgQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "whatwg-encoding": "^3.1.1"
+      },
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/html-escaper": {
+      "version": "2.0.2",
+      "resolved": "https://registry.npmjs.org/html-escaper/-/html-escaper-2.0.2.tgz",
+      "integrity": "sha512-H2iMtd0I4Mt5eYiapRdIDjp+XzelXQ0tFE4JS7YFwFevXXMmOp9myNrUvCg0D6ws8iqkRPBfKHgbwig1SmlLfg==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/http-proxy-agent": {
+      "version": "7.0.2",
+      "resolved": "https://registry.npmjs.org/http-proxy-agent/-/http-proxy-agent-7.0.2.tgz",
+      "integrity": "sha512-T1gkAiYYDWYx3V5Bmyu7HcfcvL7mUrTWiM6yOfa3PIphViJ/gFPbvidQ+veqSOHci/PxBcDabeUNCzpOODJZig==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "agent-base": "^7.1.0",
+        "debug": "^4.3.4"
+      },
+      "engines": {
+        "node": ">= 14"
+      }
+    },
+    "node_modules/https-proxy-agent": {
+      "version": "7.0.6",
+      "resolved": "https://registry.npmjs.org/https-proxy-agent/-/https-proxy-agent-7.0.6.tgz",
+      "integrity": "sha512-vK9P5/iUfdl95AI+JVyUuIcVtd4ofvtrOr3HNtM2yxC9bnMbEdp3x01OhQNnjb8IJYi38VlTE3mBXwcfvywuSw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "agent-base": "^7.1.2",
+        "debug": "4"
+      },
+      "engines": {
+        "node": ">= 14"
+      }
+    },
+    "node_modules/husky": {
+      "version": "9.1.7",
+      "resolved": "https://registry.npmjs.org/husky/-/husky-9.1.7.tgz",
+      "integrity": "sha512-5gs5ytaNjBrh5Ow3zrvdUUY+0VxIuWVL4i9irt6friV+BqdCfmV11CQTWMiBYWHbXhco+J1kHfTOUkePhCDvMA==",
+      "dev": true,
+      "license": "MIT",
+      "bin": {
+        "husky": "bin.js"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/typicode"
+      }
+    },
+    "node_modules/ignore": {
+      "version": "5.3.2",
+      "resolved": "https://registry.npmjs.org/ignore/-/ignore-5.3.2.tgz",
+      "integrity": "sha512-hsBTNUqQTDwkWtcdYI2i06Y/nUBEsNEDJKjWdigLvegy8kDuJAS8uRlpkkcQpyEXL0Z/pjDy5HBmMjRCJ2gq+g==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">= 4"
+      }
+    },
+    "node_modules/imurmurhash": {
+      "version": "0.1.4",
+      "resolved": "https://registry.npmjs.org/imurmurhash/-/imurmurhash-0.1.4.tgz",
+      "integrity": "sha512-JmXMZ6wuvDmLiHEml9ykzqO6lwFbof0GG4IkcGaENdCRDDmMVnny7s5HsIgHCbaq0w2MyPhDqkhTUgS2LU2PHA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=0.8.19"
+      }
+    },
+    "node_modules/is-extglob": {
+      "version": "2.1.1",
+      "resolved": "https://registry.npmjs.org/is-extglob/-/is-extglob-2.1.1.tgz",
+      "integrity": "sha512-SbKbANkN603Vi4jEZv49LeVJMn4yGwsbzZworEoyEiutsN3nJYdbO36zfhGJ6QEDpOZIFkDtnq5JRxmvl3jsoQ==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=0.10.0"
+      }
+    },
+    "node_modules/is-fullwidth-code-point": {
+      "version": "5.1.0",
+      "resolved": "https://registry.npmjs.org/is-fullwidth-code-point/-/is-fullwidth-code-point-5.1.0.tgz",
+      "integrity": "sha512-5XHYaSyiqADb4RnZ1Bdad6cPp8Toise4TzEjcOYDHZkTCbKgiUl7WTUCpNWHuxmDt91wnsZBc9xinNzopv3JMQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "get-east-asian-width": "^1.3.1"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/is-glob": {
+      "version": "4.0.3",
+      "resolved": "https://registry.npmjs.org/is-glob/-/is-glob-4.0.3.tgz",
+      "integrity": "sha512-xelSayHH36ZgE7ZWhli7pW34hNbNl8Ojv5KVmkJD4hBdD3th8Tfk9vYasLM+mXWOZhFkgZfxhLSnrwRr4elSSg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "is-extglob": "^2.1.1"
+      },
+      "engines": {
+        "node": ">=0.10.0"
+      }
+    },
+    "node_modules/is-number": {
+      "version": "7.0.0",
+      "resolved": "https://registry.npmjs.org/is-number/-/is-number-7.0.0.tgz",
+      "integrity": "sha512-41Cifkg6e8TylSpdtTpeLVMqvSBEVzTttHvERD741+pnZ8ANv0004MRL43QKPDlK9cGvNp6NZWZUBlbGXYxxng==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=0.12.0"
+      }
+    },
+    "node_modules/is-potential-custom-element-name": {
+      "version": "1.0.1",
+      "resolved": "https://registry.npmjs.org/is-potential-custom-element-name/-/is-potential-custom-element-name-1.0.1.tgz",
+      "integrity": "sha512-bCYeRA2rVibKZd+s2625gGnGF/t7DSqDs4dP7CrLA1m7jKWz6pps0LpYLJN8Q64HtmPKJ1hrN3nzPNKFEKOUiQ==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/isexe": {
+      "version": "2.0.0",
+      "resolved": "https://registry.npmjs.org/isexe/-/isexe-2.0.0.tgz",
+      "integrity": "sha512-RHxMLp9lnKHGHRng9QFhRCMbYAcVpn69smSGcq3f36xjgVVWThj4qqLbTLlq7Ssj8B+fIQ1EuCEGI2lKsyQeIw==",
+      "dev": true,
+      "license": "ISC"
+    },
+    "node_modules/istanbul-lib-coverage": {
+      "version": "3.2.2",
+      "resolved": "https://registry.npmjs.org/istanbul-lib-coverage/-/istanbul-lib-coverage-3.2.2.tgz",
+      "integrity": "sha512-O8dpsF+r0WV/8MNRKfnmrtCWhuKjxrq2w+jpzBL5UZKTi2LeVWnWOmWRxFlesJONmc+wLAGvKQZEOanko0LFTg==",
+      "dev": true,
+      "license": "BSD-3-Clause",
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/istanbul-lib-report": {
+      "version": "3.0.1",
+      "resolved": "https://registry.npmjs.org/istanbul-lib-report/-/istanbul-lib-report-3.0.1.tgz",
+      "integrity": "sha512-GCfE1mtsHGOELCU8e/Z7YWzpmybrx/+dSTfLrvY8qRmaY6zXTKWn6WQIjaAFw069icm6GVMNkgu0NzI4iPZUNw==",
+      "dev": true,
+      "license": "BSD-3-Clause",
+      "dependencies": {
+        "istanbul-lib-coverage": "^3.0.0",
+        "make-dir": "^4.0.0",
+        "supports-color": "^7.1.0"
+      },
+      "engines": {
+        "node": ">=10"
+      }
+    },
+    "node_modules/istanbul-lib-source-maps": {
+      "version": "5.0.6",
+      "resolved": "https://registry.npmjs.org/istanbul-lib-source-maps/-/istanbul-lib-source-maps-5.0.6.tgz",
+      "integrity": "sha512-yg2d+Em4KizZC5niWhQaIomgf5WlL4vOOjZ5xGCmF8SnPE/mDWWXgvRExdcpCgh9lLRRa1/fSYp2ymmbJ1pI+A==",
+      "dev": true,
+      "license": "BSD-3-Clause",
+      "dependencies": {
+        "@jridgewell/trace-mapping": "^0.3.23",
+        "debug": "^4.1.1",
+        "istanbul-lib-coverage": "^3.0.0"
+      },
+      "engines": {
+        "node": ">=10"
+      }
+    },
+    "node_modules/istanbul-reports": {
+      "version": "3.2.0",
+      "resolved": "https://registry.npmjs.org/istanbul-reports/-/istanbul-reports-3.2.0.tgz",
+      "integrity": "sha512-HGYWWS/ehqTV3xN10i23tkPkpH46MLCIMFNCaaKNavAXTF1RkqxawEPtnjnGZ6XKSInBKkiOA5BKS+aZiY3AvA==",
+      "dev": true,
+      "license": "BSD-3-Clause",
+      "dependencies": {
+        "html-escaper": "^2.0.0",
+        "istanbul-lib-report": "^3.0.0"
+      },
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/jackspeak": {
+      "version": "3.4.3",
+      "resolved": "https://registry.npmjs.org/jackspeak/-/jackspeak-3.4.3.tgz",
+      "integrity": "sha512-OGlZQpz2yfahA/Rd1Y8Cd9SIEsqvXkLVoSw/cgwhnhFMDbsQFeZYoJJ7bIZBS9BcamUW96asq/npPWugM+RQBw==",
+      "dev": true,
+      "license": "BlueOak-1.0.0",
+      "dependencies": {
+        "@isaacs/cliui": "^8.0.2"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/isaacs"
+      },
+      "optionalDependencies": {
+        "@pkgjs/parseargs": "^0.11.0"
+      }
+    },
+    "node_modules/js-tokens": {
+      "version": "9.0.1",
+      "resolved": "https://registry.npmjs.org/js-tokens/-/js-tokens-9.0.1.tgz",
+      "integrity": "sha512-mxa9E9ITFOt0ban3j6L5MpjwegGz6lBQmM1IJkWeBZGcMxto50+eWdjC/52xDbS2vy0k7vIMK0Fe2wfL9OQSpQ==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/js-yaml": {
+      "version": "4.1.1",
+      "resolved": "https://registry.npmjs.org/js-yaml/-/js-yaml-4.1.1.tgz",
+      "integrity": "sha512-qQKT4zQxXl8lLwBtHMWwaTcGfFOZviOJet3Oy/xmGk2gZH677CJM9EvtfdSkgWcATZhj/55JZ0rmy3myCT5lsA==",
+      "license": "MIT",
+      "dependencies": {
+        "argparse": "^2.0.1"
+      },
+      "bin": {
+        "js-yaml": "bin/js-yaml.js"
+      }
+    },
+    "node_modules/jsdom": {
+      "version": "26.1.0",
+      "resolved": "https://registry.npmjs.org/jsdom/-/jsdom-26.1.0.tgz",
+      "integrity": "sha512-Cvc9WUhxSMEo4McES3P7oK3QaXldCfNWp7pl2NNeiIFlCoLr3kfq9kb1fxftiwk1FLV7CvpvDfonxtzUDeSOPg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "cssstyle": "^4.2.1",
+        "data-urls": "^5.0.0",
+        "decimal.js": "^10.5.0",
+        "html-encoding-sniffer": "^4.0.0",
+        "http-proxy-agent": "^7.0.2",
+        "https-proxy-agent": "^7.0.6",
+        "is-potential-custom-element-name": "^1.0.1",
+        "nwsapi": "^2.2.16",
+        "parse5": "^7.2.1",
+        "rrweb-cssom": "^0.8.0",
+        "saxes": "^6.0.0",
+        "symbol-tree": "^3.2.4",
+        "tough-cookie": "^5.1.1",
+        "w3c-xmlserializer": "^5.0.0",
+        "webidl-conversions": "^7.0.0",
+        "whatwg-encoding": "^3.1.1",
+        "whatwg-mimetype": "^4.0.0",
+        "whatwg-url": "^14.1.1",
+        "ws": "^8.18.0",
+        "xml-name-validator": "^5.0.0"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "peerDependencies": {
+        "canvas": "^3.0.0"
+      },
+      "peerDependenciesMeta": {
+        "canvas": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/json-buffer": {
+      "version": "3.0.1",
+      "resolved": "https://registry.npmjs.org/json-buffer/-/json-buffer-3.0.1.tgz",
+      "integrity": "sha512-4bV5BfR2mqfQTJm+V5tPPdf+ZpuhiIvTuAB5g8kcrXOZpTT/QwwVRWBywX1ozr6lEuPdbHxwaJlm9G6mI2sfSQ==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/json-schema-traverse": {
+      "version": "0.4.1",
+      "resolved": "https://registry.npmjs.org/json-schema-traverse/-/json-schema-traverse-0.4.1.tgz",
+      "integrity": "sha512-xbbCH5dCYU5T8LcEhhuh7HJ88HXuW3qsI3Y0zOZFKfZEHcpWiHU/Jxzk629Brsab/mMiHQti9wMP+845RPe3Vg==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/json-stable-stringify-without-jsonify": {
+      "version": "1.0.1",
+      "resolved": "https://registry.npmjs.org/json-stable-stringify-without-jsonify/-/json-stable-stringify-without-jsonify-1.0.1.tgz",
+      "integrity": "sha512-Bdboy+l7tA3OGW6FjyFHWkP5LuByj1Tk33Ljyq0axyzdk9//JSi2u3fP1QSmd1KNwq6VOKYGlAu87CisVir6Pw==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/keyv": {
+      "version": "4.5.4",
+      "resolved": "https://registry.npmjs.org/keyv/-/keyv-4.5.4.tgz",
+      "integrity": "sha512-oxVHkHR/EJf2CNXnWxRLW6mg7JyCCUcG0DtEGmL2ctUo1PNTin1PUil+r/+4r5MpVgC/fn1kjsx7mjSujKqIpw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "json-buffer": "3.0.1"
+      }
+    },
+    "node_modules/levn": {
+      "version": "0.4.1",
+      "resolved": "https://registry.npmjs.org/levn/-/levn-0.4.1.tgz",
+      "integrity": "sha512-+bT2uH4E5LGE7h/n3evcS/sQlJXCpIp6ym8OWJ5eV6+67Dsql/LaaT7qJBAt2rzfoa/5QBGBhxDix1dMt2kQKQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "prelude-ls": "^1.2.1",
+        "type-check": "~0.4.0"
+      },
+      "engines": {
+        "node": ">= 0.8.0"
+      }
+    },
+    "node_modules/lint-staged": {
+      "version": "16.4.0",
+      "resolved": "https://registry.npmjs.org/lint-staged/-/lint-staged-16.4.0.tgz",
+      "integrity": "sha512-lBWt8hujh/Cjysw5GYVmZpFHXDCgZzhrOm8vbcUdobADZNOK/bRshr2kM3DfgrrtR1DQhfupW9gnIXOfiFi+bw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "commander": "^14.0.3",
+        "listr2": "^9.0.5",
+        "picomatch": "^4.0.3",
+        "string-argv": "^0.3.2",
+        "tinyexec": "^1.0.4",
+        "yaml": "^2.8.2"
+      },
+      "bin": {
+        "lint-staged": "bin/lint-staged.js"
+      },
+      "engines": {
+        "node": ">=20.17"
+      },
+      "funding": {
+        "url": "https://opencollective.com/lint-staged"
+      }
+    },
+    "node_modules/listr2": {
+      "version": "9.0.5",
+      "resolved": "https://registry.npmjs.org/listr2/-/listr2-9.0.5.tgz",
+      "integrity": "sha512-ME4Fb83LgEgwNw96RKNvKV4VTLuXfoKudAmm2lP8Kk87KaMK0/Xrx/aAkMWmT8mDb+3MlFDspfbCs7adjRxA2g==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "cli-truncate": "^5.0.0",
+        "colorette": "^2.0.20",
+        "eventemitter3": "^5.0.1",
+        "log-update": "^6.1.0",
+        "rfdc": "^1.4.1",
+        "wrap-ansi": "^9.0.0"
+      },
+      "engines": {
+        "node": ">=20.0.0"
+      }
+    },
+    "node_modules/log-update": {
+      "version": "6.1.0",
+      "resolved": "https://registry.npmjs.org/log-update/-/log-update-6.1.0.tgz",
+      "integrity": "sha512-9ie8ItPR6tjY5uYJh8K/Zrv/RMZ5VOlOWvtZdEHYSTFKZfIBPQa9tOAEeAWhd+AnIneLJ22w5fjOYtoutpWq5w==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "ansi-escapes": "^7.0.0",
+        "cli-cursor": "^5.0.0",
+        "slice-ansi": "^7.1.0",
+        "strip-ansi": "^7.1.0",
+        "wrap-ansi": "^9.0.0"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/log-update/node_modules/ansi-regex": {
+      "version": "6.2.2",
+      "resolved": "https://registry.npmjs.org/ansi-regex/-/ansi-regex-6.2.2.tgz",
+      "integrity": "sha512-Bq3SmSpyFHaWjPk8If9yc6svM8c56dB5BAtW4Qbw5jHTwwXXcTLoRMkpDJp6VL0XzlWaCHTXrkFURMYmD0sLqg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=12"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/ansi-regex?sponsor=1"
+      }
+    },
+    "node_modules/log-update/node_modules/slice-ansi": {
+      "version": "7.1.2",
+      "resolved": "https://registry.npmjs.org/slice-ansi/-/slice-ansi-7.1.2.tgz",
+      "integrity": "sha512-iOBWFgUX7caIZiuutICxVgX1SdxwAVFFKwt1EvMYYec/NWO5meOJ6K5uQxhrYBdQJne4KxiqZc+KptFOWFSI9w==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "ansi-styles": "^6.2.1",
+        "is-fullwidth-code-point": "^5.0.0"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/slice-ansi?sponsor=1"
+      }
+    },
+    "node_modules/log-update/node_modules/strip-ansi": {
+      "version": "7.2.0",
+      "resolved": "https://registry.npmjs.org/strip-ansi/-/strip-ansi-7.2.0.tgz",
+      "integrity": "sha512-yDPMNjp4WyfYBkHnjIRLfca1i6KMyGCtsVgoKe/z1+6vukgaENdgGBZt+ZmKPc4gavvEZ5OgHfHdrazhgNyG7w==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "ansi-regex": "^6.2.2"
+      },
+      "engines": {
+        "node": ">=12"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/strip-ansi?sponsor=1"
+      }
+    },
+    "node_modules/long": {
+      "version": "5.3.2",
+      "resolved": "https://registry.npmjs.org/long/-/long-5.3.2.tgz",
+      "integrity": "sha512-mNAgZ1GmyNhD7AuqnTG3/VQ26o760+ZYBPKjPvugO8+nLbYfX6TVpJPseBvopbdY+qpZ/lKUnmEc1LeZYS3QAA==",
+      "license": "Apache-2.0"
+    },
+    "node_modules/loupe": {
+      "version": "3.2.1",
+      "resolved": "https://registry.npmjs.org/loupe/-/loupe-3.2.1.tgz",
+      "integrity": "sha512-CdzqowRJCeLU72bHvWqwRBBlLcMEtIvGrlvef74kMnV2AolS9Y8xUv1I0U/MNAWMhBlKIoyuEgoJ0t/bbwHbLQ==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/lru-cache": {
+      "version": "10.4.3",
+      "resolved": "https://registry.npmjs.org/lru-cache/-/lru-cache-10.4.3.tgz",
+      "integrity": "sha512-JNAzZcXrCt42VGLuYz0zfAzDfAvJWW6AfYlDBQyDV5DClI2m5sAmK+OIO7s59XfsRsWHp02jAJrRadPRGTt6SQ==",
+      "dev": true,
+      "license": "ISC"
+    },
+    "node_modules/magic-string": {
+      "version": "0.30.21",
+      "resolved": "https://registry.npmjs.org/magic-string/-/magic-string-0.30.21.tgz",
+      "integrity": "sha512-vd2F4YUyEXKGcLHoq+TEyCjxueSeHnFxyyjNp80yg0XV4vUhnDer/lvvlqM/arB5bXQN5K2/3oinyCRyx8T2CQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@jridgewell/sourcemap-codec": "^1.5.5"
+      }
+    },
+    "node_modules/magicast": {
+      "version": "0.3.5",
+      "resolved": "https://registry.npmjs.org/magicast/-/magicast-0.3.5.tgz",
+      "integrity": "sha512-L0WhttDl+2BOsybvEOLK7fW3UA0OQ0IQ2d6Zl2x/a6vVRs3bAY0ECOSHHeL5jD+SbOpOCUEi0y1DgHEn9Qn1AQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@babel/parser": "^7.25.4",
+        "@babel/types": "^7.25.4",
+        "source-map-js": "^1.2.0"
+      }
+    },
+    "node_modules/make-dir": {
+      "version": "4.0.0",
+      "resolved": "https://registry.npmjs.org/make-dir/-/make-dir-4.0.0.tgz",
+      "integrity": "sha512-hXdUTZYIVOt1Ex//jAQi+wTZZpUpwBj/0QsOzqegb3rGMMeJiSEu5xLHnYfBrRV4RH2+OCSOO95Is/7x1WJ4bw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "semver": "^7.5.3"
+      },
+      "engines": {
+        "node": ">=10"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/merge2": {
+      "version": "1.4.1",
+      "resolved": "https://registry.npmjs.org/merge2/-/merge2-1.4.1.tgz",
+      "integrity": "sha512-8q7VEgMJW4J8tcfVPy8g09NcQwZdbwFEqhe/WZkoIzjn/3TGDwtOCYtXGxA3O8tPzpczCCDgv+P2P5y00ZJOOg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">= 8"
+      }
+    },
+    "node_modules/micromatch": {
+      "version": "4.0.8",
+      "resolved": "https://registry.npmjs.org/micromatch/-/micromatch-4.0.8.tgz",
+      "integrity": "sha512-PXwfBhYu0hBCPw8Dn0E+WDYb7af3dSLVWKi3HGv84IdF4TyFoC0ysxFd0Goxw7nSv4T/PzEJQxsYsEiFCKo2BA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "braces": "^3.0.3",
+        "picomatch": "^2.3.1"
+      },
+      "engines": {
+        "node": ">=8.6"
+      }
+    },
+    "node_modules/micromatch/node_modules/picomatch": {
+      "version": "2.3.2",
+      "resolved": "https://registry.npmjs.org/picomatch/-/picomatch-2.3.2.tgz",
+      "integrity": "sha512-V7+vQEJ06Z+c5tSye8S+nHUfI51xoXIXjHQ99cQtKUkQqqO1kO/KCJUfZXuB47h/YBlDhah2H3hdUGXn8ie0oA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=8.6"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/jonschlinkert"
+      }
+    },
+    "node_modules/mimic-function": {
+      "version": "5.0.1",
+      "resolved": "https://registry.npmjs.org/mimic-function/-/mimic-function-5.0.1.tgz",
+      "integrity": "sha512-VP79XUPxV2CigYP3jWwAUFSku2aKqBH7uTAapFWCBqutsbmDo96KY5o8uh6U+/YSIn5OxJnXp73beVkpqMIGhA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=18"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/minimatch": {
+      "version": "10.2.4",
+      "resolved": "https://registry.npmjs.org/minimatch/-/minimatch-10.2.4.tgz",
+      "integrity": "sha512-oRjTw/97aTBN0RHbYCdtF1MQfvusSIBQM0IZEgzl6426+8jSC0nF1a/GmnVLpfB9yyr6g6FTqWqiZVbxrtaCIg==",
+      "dev": true,
+      "license": "BlueOak-1.0.0",
+      "dependencies": {
+        "brace-expansion": "^5.0.2"
+      },
+      "engines": {
+        "node": "18 || 20 || >=22"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/isaacs"
+      }
+    },
+    "node_modules/minipass": {
+      "version": "7.1.3",
+      "resolved": "https://registry.npmjs.org/minipass/-/minipass-7.1.3.tgz",
+      "integrity": "sha512-tEBHqDnIoM/1rXME1zgka9g6Q2lcoCkxHLuc7ODJ5BxbP5d4c2Z5cGgtXAku59200Cx7diuHTOYfSBD8n6mm8A==",
+      "dev": true,
+      "license": "BlueOak-1.0.0",
+      "engines": {
+        "node": ">=16 || 14 >=14.17"
+      }
+    },
+    "node_modules/mkdirp": {
+      "version": "1.0.4",
+      "resolved": "https://registry.npmjs.org/mkdirp/-/mkdirp-1.0.4.tgz",
+      "integrity": "sha512-vVqVZQyf3WLx2Shd0qJ9xuvqgAyKPLAiqITEtqW0oIUjzo3PePDd6fW9iFz30ef7Ysp/oiWqbhszeGWW2T6Gzw==",
+      "dev": true,
+      "license": "MIT",
+      "bin": {
+        "mkdirp": "bin/cmd.js"
+      },
+      "engines": {
+        "node": ">=10"
+      }
+    },
+    "node_modules/ms": {
+      "version": "2.1.3",
+      "resolved": "https://registry.npmjs.org/ms/-/ms-2.1.3.tgz",
+      "integrity": "sha512-6FlzubTLZG3J2a/NVCAleEhjzq5oxgHyaCU9yYXvcLsvoVaHJq/s5xXI6/XXP6tz7R9xAOtHnSO/tXtF3WRTlA==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/nanoid": {
+      "version": "3.3.11",
+      "resolved": "https://registry.npmjs.org/nanoid/-/nanoid-3.3.11.tgz",
+      "integrity": "sha512-N8SpfPUnUp1bK+PMYW8qSWdl9U+wwNWI4QKxOYDy9JAro3WMX7p2OeVRF9v+347pnakNevPmiHhNmZ2HbFA76w==",
+      "dev": true,
+      "funding": [
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/ai"
+        }
+      ],
+      "license": "MIT",
+      "bin": {
+        "nanoid": "bin/nanoid.cjs"
+      },
+      "engines": {
+        "node": "^10 || ^12 || ^13.7 || ^14 || >=15.0.1"
+      }
+    },
+    "node_modules/natural-compare": {
+      "version": "1.4.0",
+      "resolved": "https://registry.npmjs.org/natural-compare/-/natural-compare-1.4.0.tgz",
+      "integrity": "sha512-OWND8ei3VtNC9h7V60qff3SVobHr996CTwgxubgyQYEpg290h9J0buyECNNJexkFm5sOajh5G116RYA1c8ZMSw==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/nwsapi": {
+      "version": "2.2.23",
+      "resolved": "https://registry.npmjs.org/nwsapi/-/nwsapi-2.2.23.tgz",
+      "integrity": "sha512-7wfH4sLbt4M0gCDzGE6vzQBo0bfTKjU7Sfpqy/7gs1qBfYz2vEJH6vXcBKpO3+6Yu1telwd0t9HpyOoLEQQbIQ==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/onetime": {
+      "version": "7.0.0",
+      "resolved": "https://registry.npmjs.org/onetime/-/onetime-7.0.0.tgz",
+      "integrity": "sha512-VXJjc87FScF88uafS3JllDgvAm+c/Slfz06lorj2uAY34rlUu0Nt+v8wreiImcrgAjjIHp1rXpTDlLOGw29WwQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "mimic-function": "^5.0.0"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/onnxruntime-common": {
+      "version": "1.24.3",
+      "resolved": "https://registry.npmjs.org/onnxruntime-common/-/onnxruntime-common-1.24.3.tgz",
+      "integrity": "sha512-GeuPZO6U/LBJXvwdaqHbuUmoXiEdeCjWi/EG7Y1HNnDwJYuk6WUbNXpF6luSUY8yASul3cmUlLGrCCL1ZgVXqA==",
+      "license": "MIT"
+    },
+    "node_modules/onnxruntime-web": {
+      "version": "1.24.3",
+      "resolved": "https://registry.npmjs.org/onnxruntime-web/-/onnxruntime-web-1.24.3.tgz",
+      "integrity": "sha512-41dDq7fxtTm0XzGE7N0d6m8FcOY8EWtUA65GkOixJPB/G7DGzBmiDAnVVXHznRw9bgUZpb+4/1lQK/PNxGpbrQ==",
+      "license": "MIT",
+      "dependencies": {
+        "flatbuffers": "^25.1.24",
+        "guid-typescript": "^1.0.9",
+        "long": "^5.2.3",
+        "onnxruntime-common": "1.24.3",
+        "platform": "^1.3.6",
+        "protobufjs": "^7.2.4"
+      }
+    },
+    "node_modules/optionator": {
+      "version": "0.9.4",
+      "resolved": "https://registry.npmjs.org/optionator/-/optionator-0.9.4.tgz",
+      "integrity": "sha512-6IpQ7mKUxRcZNLIObR0hz7lxsapSSIYNZJwXPGeF0mTVqGKFIXj1DQcMoT22S3ROcLyY/rz0PWaWZ9ayWmad9g==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "deep-is": "^0.1.3",
+        "fast-levenshtein": "^2.0.6",
+        "levn": "^0.4.1",
+        "prelude-ls": "^1.2.1",
+        "type-check": "^0.4.0",
+        "word-wrap": "^1.2.5"
+      },
+      "engines": {
+        "node": ">= 0.8.0"
+      }
+    },
+    "node_modules/package-json-from-dist": {
+      "version": "1.0.1",
+      "resolved": "https://registry.npmjs.org/package-json-from-dist/-/package-json-from-dist-1.0.1.tgz",
+      "integrity": "sha512-UEZIS3/by4OC8vL3P2dTXRETpebLI2NiI5vIrjaD/5UtrkFX/tNbwjTSRAGC/+7CAo2pIcBaRgWmcBBHcsaCIw==",
+      "dev": true,
+      "license": "BlueOak-1.0.0"
+    },
+    "node_modules/parse5": {
+      "version": "7.3.0",
+      "resolved": "https://registry.npmjs.org/parse5/-/parse5-7.3.0.tgz",
+      "integrity": "sha512-IInvU7fabl34qmi9gY8XOVxhYyMyuH2xUNpb2q8/Y+7552KlejkRvqvD19nMoUW/uQGGbqNpA6Tufu5FL5BZgw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "entities": "^6.0.0"
+      },
+      "funding": {
+        "url": "https://github.com/inikulin/parse5?sponsor=1"
+      }
+    },
+    "node_modules/path-browserify": {
+      "version": "1.0.1",
+      "resolved": "https://registry.npmjs.org/path-browserify/-/path-browserify-1.0.1.tgz",
+      "integrity": "sha512-b7uo2UCUOYZcnF/3ID0lulOJi/bafxa1xPe7ZPsammBSpjSWQkjNxlt635YGS2MiR9GjvuXCtz2emr3jbsz98g==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/path-exists": {
+      "version": "4.0.0",
+      "resolved": "https://registry.npmjs.org/path-exists/-/path-exists-4.0.0.tgz",
+      "integrity": "sha512-ak9Qy5Q7jYb2Wwcey5Fpvg2KoAc/ZIhLSLOSBmRmygPsGwkVVt0fZa0qrtMz+m6tJTAHfZQ8FnmB4MG4LWy7/w==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/path-key": {
+      "version": "3.1.1",
+      "resolved": "https://registry.npmjs.org/path-key/-/path-key-3.1.1.tgz",
+      "integrity": "sha512-ojmeN0qd+y0jszEtoY48r0Peq5dwMEkIlCOu6Q5f41lfkswXuKtYrhgoTpLnyIcHm24Uhqx+5Tqm2InSwLhE6Q==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/path-scurry": {
+      "version": "1.11.1",
+      "resolved": "https://registry.npmjs.org/path-scurry/-/path-scurry-1.11.1.tgz",
+      "integrity": "sha512-Xa4Nw17FS9ApQFJ9umLiJS4orGjm7ZzwUrwamcGQuHSzDyth9boKDaycYdDcZDuqYATXw4HFXgaqWTctW/v1HA==",
+      "dev": true,
+      "license": "BlueOak-1.0.0",
+      "dependencies": {
+        "lru-cache": "^10.2.0",
+        "minipass": "^5.0.0 || ^6.0.2 || ^7.0.0"
+      },
+      "engines": {
+        "node": ">=16 || 14 >=14.18"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/isaacs"
+      }
+    },
+    "node_modules/pathe": {
+      "version": "2.0.3",
+      "resolved": "https://registry.npmjs.org/pathe/-/pathe-2.0.3.tgz",
+      "integrity": "sha512-WUjGcAqP1gQacoQe+OBJsFA7Ld4DyXuUIjZ5cc75cLHvJ7dtNsTugphxIADwspS+AraAUePCKrSVtPLFj/F88w==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/pathval": {
+      "version": "2.0.1",
+      "resolved": "https://registry.npmjs.org/pathval/-/pathval-2.0.1.tgz",
+      "integrity": "sha512-//nshmD55c46FuFw26xV/xFAaB5HF9Xdap7HJBBnrKdAd6/GxDBaNA1870O79+9ueg61cZLSVc+OaFlfmObYVQ==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">= 14.16"
+      }
+    },
+    "node_modules/picocolors": {
+      "version": "1.1.1",
+      "resolved": "https://registry.npmjs.org/picocolors/-/picocolors-1.1.1.tgz",
+      "integrity": "sha512-xceH2snhtb5M9liqDsmEw56le376mTZkEX/jEb/RxNFyegNul7eNslCXP9FDj/Lcu0X8KEyMceP2ntpaHrDEVA==",
+      "dev": true,
+      "license": "ISC"
+    },
+    "node_modules/picomatch": {
+      "version": "4.0.4",
+      "resolved": "https://registry.npmjs.org/picomatch/-/picomatch-4.0.4.tgz",
+      "integrity": "sha512-QP88BAKvMam/3NxH6vj2o21R6MjxZUAd6nlwAS/pnGvN9IVLocLHxGYIzFhg6fUQ+5th6P4dv4eW9jX3DSIj7A==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=12"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/jonschlinkert"
+      }
+    },
+    "node_modules/platform": {
+      "version": "1.3.6",
+      "resolved": "https://registry.npmjs.org/platform/-/platform-1.3.6.tgz",
+      "integrity": "sha512-fnWVljUchTro6RiCFvCXBbNhJc2NijN7oIQxbwsyL0buWJPG85v81ehlHI9fXrJsMNgTofEoWIQeClKpgxFLrg==",
+      "license": "MIT"
+    },
+    "node_modules/postcss": {
+      "version": "8.5.8",
+      "resolved": "https://registry.npmjs.org/postcss/-/postcss-8.5.8.tgz",
+      "integrity": "sha512-OW/rX8O/jXnm82Ey1k44pObPtdblfiuWnrd8X7GJ7emImCOstunGbXUpp7HdBrFQX6rJzn3sPT397Wp5aCwCHg==",
+      "dev": true,
+      "funding": [
+        {
+          "type": "opencollective",
+          "url": "https://opencollective.com/postcss/"
+        },
+        {
+          "type": "tidelift",
+          "url": "https://tidelift.com/funding/github/npm/postcss"
+        },
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/ai"
+        }
+      ],
+      "license": "MIT",
+      "dependencies": {
+        "nanoid": "^3.3.11",
+        "picocolors": "^1.1.1",
+        "source-map-js": "^1.2.1"
+      },
+      "engines": {
+        "node": "^10 || ^12 || >=14"
+      }
+    },
+    "node_modules/prelude-ls": {
+      "version": "1.2.1",
+      "resolved": "https://registry.npmjs.org/prelude-ls/-/prelude-ls-1.2.1.tgz",
+      "integrity": "sha512-vkcDPrRZo1QZLbn5RLGPpg/WmIQ65qoWWhcGKf/b5eplkkarX0m9z8ppCat4mlOqUsWpyNuYgO3VRyrYHSzX5g==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">= 0.8.0"
+      }
+    },
+    "node_modules/prettier": {
+      "version": "3.8.1",
+      "resolved": "https://registry.npmjs.org/prettier/-/prettier-3.8.1.tgz",
+      "integrity": "sha512-UOnG6LftzbdaHZcKoPFtOcCKztrQ57WkHDeRD9t/PTQtmT0NHSeWWepj6pS0z/N7+08BHFDQVUrfmfMRcZwbMg==",
+      "dev": true,
+      "license": "MIT",
+      "bin": {
+        "prettier": "bin/prettier.cjs"
+      },
+      "engines": {
+        "node": ">=14"
+      },
+      "funding": {
+        "url": "https://github.com/prettier/prettier?sponsor=1"
+      }
+    },
+    "node_modules/protobufjs": {
+      "version": "7.5.4",
+      "resolved": "https://registry.npmjs.org/protobufjs/-/protobufjs-7.5.4.tgz",
+      "integrity": "sha512-CvexbZtbov6jW2eXAvLukXjXUW1TzFaivC46BpWc/3BpcCysb5Vffu+B3XHMm8lVEuy2Mm4XGex8hBSg1yapPg==",
+      "hasInstallScript": true,
+      "license": "BSD-3-Clause",
+      "dependencies": {
+        "@protobufjs/aspromise": "^1.1.2",
+        "@protobufjs/base64": "^1.1.2",
+        "@protobufjs/codegen": "^2.0.4",
+        "@protobufjs/eventemitter": "^1.1.0",
+        "@protobufjs/fetch": "^1.1.0",
+        "@protobufjs/float": "^1.0.2",
+        "@protobufjs/inquire": "^1.1.0",
+        "@protobufjs/path": "^1.1.2",
+        "@protobufjs/pool": "^1.1.0",
+        "@protobufjs/utf8": "^1.1.0",
+        "@types/node": ">=13.7.0",
+        "long": "^5.0.0"
+      },
+      "engines": {
+        "node": ">=12.0.0"
+      }
+    },
+    "node_modules/punycode": {
+      "version": "2.3.1",
+      "resolved": "https://registry.npmjs.org/punycode/-/punycode-2.3.1.tgz",
+      "integrity": "sha512-vYt7UD1U9Wg6138shLtLOvdAu+8DsC/ilFtEVHcH+wydcSpNE20AfSOduf6MkRFahL5FY7X1oU7nKVZFtfq8Fg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=6"
+      }
+    },
+    "node_modules/queue-microtask": {
+      "version": "1.2.3",
+      "resolved": "https://registry.npmjs.org/queue-microtask/-/queue-microtask-1.2.3.tgz",
+      "integrity": "sha512-NuaNSa6flKT5JaSYQzJok04JzTL1CA6aGhv5rfLW3PgqA+M2ChpZQnAC8h8i4ZFkBS8X5RqkDBHA7r4hej3K9A==",
+      "dev": true,
+      "funding": [
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/feross"
+        },
+        {
+          "type": "patreon",
+          "url": "https://www.patreon.com/feross"
+        },
+        {
+          "type": "consulting",
+          "url": "https://feross.org/support"
+        }
+      ],
+      "license": "MIT"
+    },
+    "node_modules/restore-cursor": {
+      "version": "5.1.0",
+      "resolved": "https://registry.npmjs.org/restore-cursor/-/restore-cursor-5.1.0.tgz",
+      "integrity": "sha512-oMA2dcrw6u0YfxJQXm342bFKX/E4sG9rbTzO9ptUcR/e8A33cHuvStiYOwH7fszkZlZ1z/ta9AAoPk2F4qIOHA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "onetime": "^7.0.0",
+        "signal-exit": "^4.1.0"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/reusify": {
+      "version": "1.1.0",
+      "resolved": "https://registry.npmjs.org/reusify/-/reusify-1.1.0.tgz",
+      "integrity": "sha512-g6QUff04oZpHs0eG5p83rFLhHeV00ug/Yf9nZM6fLeUrPguBTkTQOdpAWWspMh55TZfVQDPaN3NQJfbVRAxdIw==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "iojs": ">=1.0.0",
+        "node": ">=0.10.0"
+      }
+    },
+    "node_modules/rfdc": {
+      "version": "1.4.1",
+      "resolved": "https://registry.npmjs.org/rfdc/-/rfdc-1.4.1.tgz",
+      "integrity": "sha512-q1b3N5QkRUWUl7iyylaaj3kOpIT0N2i9MqIEQXP73GVsN9cw3fdx8X63cEmWhJGi2PPCF23Ijp7ktmd39rawIA==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/rollup": {
+      "version": "4.59.0",
+      "resolved": "https://registry.npmjs.org/rollup/-/rollup-4.59.0.tgz",
+      "integrity": "sha512-2oMpl67a3zCH9H79LeMcbDhXW/UmWG/y2zuqnF2jQq5uq9TbM9TVyXvA4+t+ne2IIkBdrLpAaRQAvo7YI/Yyeg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@types/estree": "1.0.8"
+      },
+      "bin": {
+        "rollup": "dist/bin/rollup"
+      },
+      "engines": {
+        "node": ">=18.0.0",
+        "npm": ">=8.0.0"
+      },
+      "optionalDependencies": {
+        "@rollup/rollup-android-arm-eabi": "4.59.0",
+        "@rollup/rollup-android-arm64": "4.59.0",
+        "@rollup/rollup-darwin-arm64": "4.59.0",
+        "@rollup/rollup-darwin-x64": "4.59.0",
+        "@rollup/rollup-freebsd-arm64": "4.59.0",
+        "@rollup/rollup-freebsd-x64": "4.59.0",
+        "@rollup/rollup-linux-arm-gnueabihf": "4.59.0",
+        "@rollup/rollup-linux-arm-musleabihf": "4.59.0",
+        "@rollup/rollup-linux-arm64-gnu": "4.59.0",
+        "@rollup/rollup-linux-arm64-musl": "4.59.0",
+        "@rollup/rollup-linux-loong64-gnu": "4.59.0",
+        "@rollup/rollup-linux-loong64-musl": "4.59.0",
+        "@rollup/rollup-linux-ppc64-gnu": "4.59.0",
+        "@rollup/rollup-linux-ppc64-musl": "4.59.0",
+        "@rollup/rollup-linux-riscv64-gnu": "4.59.0",
+        "@rollup/rollup-linux-riscv64-musl": "4.59.0",
+        "@rollup/rollup-linux-s390x-gnu": "4.59.0",
+        "@rollup/rollup-linux-x64-gnu": "4.59.0",
+        "@rollup/rollup-linux-x64-musl": "4.59.0",
+        "@rollup/rollup-openbsd-x64": "4.59.0",
+        "@rollup/rollup-openharmony-arm64": "4.59.0",
+        "@rollup/rollup-win32-arm64-msvc": "4.59.0",
+        "@rollup/rollup-win32-ia32-msvc": "4.59.0",
+        "@rollup/rollup-win32-x64-gnu": "4.59.0",
+        "@rollup/rollup-win32-x64-msvc": "4.59.0",
+        "fsevents": "~2.3.2"
+      }
+    },
+    "node_modules/rrweb-cssom": {
+      "version": "0.8.0",
+      "resolved": "https://registry.npmjs.org/rrweb-cssom/-/rrweb-cssom-0.8.0.tgz",
+      "integrity": "sha512-guoltQEx+9aMf2gDZ0s62EcV8lsXR+0w8915TC3ITdn2YueuNjdAYh/levpU9nFaoChh9RUS5ZdQMrKfVEN9tw==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/run-parallel": {
+      "version": "1.2.0",
+      "resolved": "https://registry.npmjs.org/run-parallel/-/run-parallel-1.2.0.tgz",
+      "integrity": "sha512-5l4VyZR86LZ/lDxZTR6jqL8AFE2S0IFLMP26AbjsLVADxHdhB/c0GUsH+y39UfCi3dzz8OlQuPmnaJOMoDHQBA==",
+      "dev": true,
+      "funding": [
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/feross"
+        },
+        {
+          "type": "patreon",
+          "url": "https://www.patreon.com/feross"
+        },
+        {
+          "type": "consulting",
+          "url": "https://feross.org/support"
+        }
+      ],
+      "license": "MIT",
+      "dependencies": {
+        "queue-microtask": "^1.2.2"
+      }
+    },
+    "node_modules/safer-buffer": {
+      "version": "2.1.2",
+      "resolved": "https://registry.npmjs.org/safer-buffer/-/safer-buffer-2.1.2.tgz",
+      "integrity": "sha512-YZo3K82SD7Riyi0E1EQPojLz7kpepnSQI9IyPbHHg1XXXevb5dJI7tpyN2ADxGcQbHG7vcyRHk0cbwqcQriUtg==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/saxes": {
+      "version": "6.0.0",
+      "resolved": "https://registry.npmjs.org/saxes/-/saxes-6.0.0.tgz",
+      "integrity": "sha512-xAg7SOnEhrm5zI3puOOKyy1OMcMlIJZYNJY7xLBwSze0UjhPLnWfj2GF2EpT0jmzaJKIWKHLsaSSajf35bcYnA==",
+      "dev": true,
+      "license": "ISC",
+      "dependencies": {
+        "xmlchars": "^2.2.0"
+      },
+      "engines": {
+        "node": ">=v12.22.7"
+      }
+    },
+    "node_modules/semver": {
+      "version": "7.7.4",
+      "resolved": "https://registry.npmjs.org/semver/-/semver-7.7.4.tgz",
+      "integrity": "sha512-vFKC2IEtQnVhpT78h1Yp8wzwrf8CM+MzKMHGJZfBtzhZNycRFnXsHk6E5TxIkkMsgNS7mdX3AGB7x2QM2di4lA==",
+      "dev": true,
+      "license": "ISC",
+      "bin": {
+        "semver": "bin/semver.js"
+      },
+      "engines": {
+        "node": ">=10"
+      }
+    },
+    "node_modules/shebang-command": {
+      "version": "2.0.0",
+      "resolved": "https://registry.npmjs.org/shebang-command/-/shebang-command-2.0.0.tgz",
+      "integrity": "sha512-kHxr2zZpYtdmrN1qDjrrX/Z1rR1kG8Dx+gkpK1G4eXmvXswmcE1hTWBWYUzlraYw1/yZp6YuDY77YtvbN0dmDA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "shebang-regex": "^3.0.0"
+      },
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/shebang-regex": {
+      "version": "3.0.0",
+      "resolved": "https://registry.npmjs.org/shebang-regex/-/shebang-regex-3.0.0.tgz",
+      "integrity": "sha512-7++dFhtcx3353uBaq8DDR4NuxBetBzC7ZQOhmTQInHEd6bSrXdiEyzCvG07Z44UYdLShWUyXt5M/yhz8ekcb1A==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/siginfo": {
+      "version": "2.0.0",
+      "resolved": "https://registry.npmjs.org/siginfo/-/siginfo-2.0.0.tgz",
+      "integrity": "sha512-ybx0WO1/8bSBLEWXZvEd7gMW3Sn3JFlW3TvX1nREbDLRNQNaeNN8WK0meBwPdAaOI7TtRRRJn/Es1zhrrCHu7g==",
+      "dev": true,
+      "license": "ISC"
+    },
+    "node_modules/signal-exit": {
+      "version": "4.1.0",
+      "resolved": "https://registry.npmjs.org/signal-exit/-/signal-exit-4.1.0.tgz",
+      "integrity": "sha512-bzyZ1e88w9O1iNJbKnOlvYTrWPDl46O1bG0D3XInv+9tkPrxrN8jUUTiFlDkkmKWgn1M6CfIA13SuGqOa9Korw==",
+      "dev": true,
+      "license": "ISC",
+      "engines": {
+        "node": ">=14"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/isaacs"
+      }
+    },
+    "node_modules/slice-ansi": {
+      "version": "8.0.0",
+      "resolved": "https://registry.npmjs.org/slice-ansi/-/slice-ansi-8.0.0.tgz",
+      "integrity": "sha512-stxByr12oeeOyY2BlviTNQlYV5xOj47GirPr4yA1hE9JCtxfQN0+tVbkxwCtYDQWhEKWFHsEK48ORg5jrouCAg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "ansi-styles": "^6.2.3",
+        "is-fullwidth-code-point": "^5.1.0"
+      },
+      "engines": {
+        "node": ">=20"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/slice-ansi?sponsor=1"
+      }
+    },
+    "node_modules/source-map-js": {
+      "version": "1.2.1",
+      "resolved": "https://registry.npmjs.org/source-map-js/-/source-map-js-1.2.1.tgz",
+      "integrity": "sha512-UXWMKhLOwVKb728IUtQPXxfYU+usdybtUrK/8uGE8CQMvrhOpwvzDBwj0QhSL7MQc7vIsISBG8VQ8+IDQxpfQA==",
+      "dev": true,
+      "license": "BSD-3-Clause",
+      "engines": {
+        "node": ">=0.10.0"
+      }
+    },
+    "node_modules/stackback": {
+      "version": "0.0.2",
+      "resolved": "https://registry.npmjs.org/stackback/-/stackback-0.0.2.tgz",
+      "integrity": "sha512-1XMJE5fQo1jGH6Y/7ebnwPOBEkIEnT4QF32d5R1+VXdXveM0IBMJt8zfaxX1P3QhVwrYe+576+jkANtSS2mBbw==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/std-env": {
+      "version": "3.10.0",
+      "resolved": "https://registry.npmjs.org/std-env/-/std-env-3.10.0.tgz",
+      "integrity": "sha512-5GS12FdOZNliM5mAOxFRg7Ir0pWz8MdpYm6AY6VPkGpbA7ZzmbzNcBJQ0GPvvyWgcY7QAhCgf9Uy89I03faLkg==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/string-argv": {
+      "version": "0.3.2",
+      "resolved": "https://registry.npmjs.org/string-argv/-/string-argv-0.3.2.tgz",
+      "integrity": "sha512-aqD2Q0144Z+/RqG52NeHEkZauTAUWJO8c6yTftGJKO3Tja5tUgIfmIl6kExvhtxSDP7fXB6DvzkfMpCd/F3G+Q==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=0.6.19"
+      }
+    },
+    "node_modules/string-width": {
+      "version": "8.2.0",
+      "resolved": "https://registry.npmjs.org/string-width/-/string-width-8.2.0.tgz",
+      "integrity": "sha512-6hJPQ8N0V0P3SNmP6h2J99RLuzrWz2gvT7VnK5tKvrNqJoyS9W4/Fb8mo31UiPvy00z7DQXkP2hnKBVav76thw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "get-east-asian-width": "^1.5.0",
+        "strip-ansi": "^7.1.2"
+      },
+      "engines": {
+        "node": ">=20"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/string-width-cjs": {
+      "name": "string-width",
+      "version": "4.2.3",
+      "resolved": "https://registry.npmjs.org/string-width/-/string-width-4.2.3.tgz",
+      "integrity": "sha512-wKyQRQpjJ0sIp62ErSZdGsjMJWsap5oRNihHhu6G7JVO/9jIB6UyevL+tXuOqrng8j/cxKTWyWUwvSTriiZz/g==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "emoji-regex": "^8.0.0",
+        "is-fullwidth-code-point": "^3.0.0",
+        "strip-ansi": "^6.0.1"
+      },
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/string-width-cjs/node_modules/emoji-regex": {
+      "version": "8.0.0",
+      "resolved": "https://registry.npmjs.org/emoji-regex/-/emoji-regex-8.0.0.tgz",
+      "integrity": "sha512-MSjYzcWNOA0ewAHpz0MxpYFvwg6yjy1NG3xteoqz644VCo/RPgnr1/GGt+ic3iJTzQ8Eu3TdM14SawnVUmGE6A==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/string-width-cjs/node_modules/is-fullwidth-code-point": {
+      "version": "3.0.0",
+      "resolved": "https://registry.npmjs.org/is-fullwidth-code-point/-/is-fullwidth-code-point-3.0.0.tgz",
+      "integrity": "sha512-zymm5+u+sCsSWyD9qNaejV3DFvhCKclKdizYaJUuHA83RLjb7nSuGnddCHGv0hk+KY7BMAlsWeK4Ueg6EV6XQg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/string-width/node_modules/ansi-regex": {
+      "version": "6.2.2",
+      "resolved": "https://registry.npmjs.org/ansi-regex/-/ansi-regex-6.2.2.tgz",
+      "integrity": "sha512-Bq3SmSpyFHaWjPk8If9yc6svM8c56dB5BAtW4Qbw5jHTwwXXcTLoRMkpDJp6VL0XzlWaCHTXrkFURMYmD0sLqg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=12"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/ansi-regex?sponsor=1"
+      }
+    },
+    "node_modules/string-width/node_modules/strip-ansi": {
+      "version": "7.2.0",
+      "resolved": "https://registry.npmjs.org/strip-ansi/-/strip-ansi-7.2.0.tgz",
+      "integrity": "sha512-yDPMNjp4WyfYBkHnjIRLfca1i6KMyGCtsVgoKe/z1+6vukgaENdgGBZt+ZmKPc4gavvEZ5OgHfHdrazhgNyG7w==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "ansi-regex": "^6.2.2"
+      },
+      "engines": {
+        "node": ">=12"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/strip-ansi?sponsor=1"
+      }
+    },
+    "node_modules/strip-ansi": {
+      "version": "6.0.1",
+      "resolved": "https://registry.npmjs.org/strip-ansi/-/strip-ansi-6.0.1.tgz",
+      "integrity": "sha512-Y38VPSHcqkFrCpFnQ9vuSXmquuv5oXOKpGeT6aGrr3o3Gc9AlVa6JBfUSOCnbxGGZF+/0ooI7KrPuUSztUdU5A==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "ansi-regex": "^5.0.1"
+      },
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/strip-ansi-cjs": {
+      "name": "strip-ansi",
+      "version": "6.0.1",
+      "resolved": "https://registry.npmjs.org/strip-ansi/-/strip-ansi-6.0.1.tgz",
+      "integrity": "sha512-Y38VPSHcqkFrCpFnQ9vuSXmquuv5oXOKpGeT6aGrr3o3Gc9AlVa6JBfUSOCnbxGGZF+/0ooI7KrPuUSztUdU5A==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "ansi-regex": "^5.0.1"
+      },
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/strip-literal": {
+      "version": "3.1.0",
+      "resolved": "https://registry.npmjs.org/strip-literal/-/strip-literal-3.1.0.tgz",
+      "integrity": "sha512-8r3mkIM/2+PpjHoOtiAW8Rg3jJLHaV7xPwG+YRGrv6FP0wwk/toTpATxWYOW0BKdWwl82VT2tFYi5DlROa0Mxg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "js-tokens": "^9.0.1"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/antfu"
+      }
+    },
+    "node_modules/supports-color": {
+      "version": "7.2.0",
+      "resolved": "https://registry.npmjs.org/supports-color/-/supports-color-7.2.0.tgz",
+      "integrity": "sha512-qpCAvRl9stuOHveKsn7HncJRvv501qIacKzQlO/+Lwxc9+0q2wLyv4Dfvt80/DPn2pqOBsJdDiogXGR9+OvwRw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "has-flag": "^4.0.0"
+      },
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/symbol-tree": {
+      "version": "3.2.4",
+      "resolved": "https://registry.npmjs.org/symbol-tree/-/symbol-tree-3.2.4.tgz",
+      "integrity": "sha512-9QNk5KwDF+Bvz+PyObkmSYjI5ksVUYtjW7AU22r2NKcfLJcXp96hkDWU3+XndOsUb+AQ9QhfzfCT2O+CNWT5Tw==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/test-exclude": {
+      "version": "7.0.2",
+      "resolved": "https://registry.npmjs.org/test-exclude/-/test-exclude-7.0.2.tgz",
+      "integrity": "sha512-u9E6A+ZDYdp7a4WnarkXPZOx8Ilz46+kby6p1yZ8zsGTz9gYa6FIS7lj2oezzNKmtdyyJNNmmXDppga5GB7kSw==",
+      "dev": true,
+      "license": "ISC",
+      "dependencies": {
+        "@istanbuljs/schema": "^0.1.2",
+        "glob": "^10.4.1",
+        "minimatch": "^10.2.2"
+      },
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/tinybench": {
+      "version": "2.9.0",
+      "resolved": "https://registry.npmjs.org/tinybench/-/tinybench-2.9.0.tgz",
+      "integrity": "sha512-0+DUvqWMValLmha6lr4kD8iAMK1HzV0/aKnCtWb9v9641TnP/MFb7Pc2bxoxQjTXAErryXVgUOfv2YqNllqGeg==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/tinyexec": {
+      "version": "1.0.4",
+      "resolved": "https://registry.npmjs.org/tinyexec/-/tinyexec-1.0.4.tgz",
+      "integrity": "sha512-u9r3uZC0bdpGOXtlxUIdwf9pkmvhqJdrVCH9fapQtgy/OeTTMZ1nqH7agtvEfmGui6e1XxjcdrlxvxJvc3sMqw==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/tinyglobby": {
+      "version": "0.2.15",
+      "resolved": "https://registry.npmjs.org/tinyglobby/-/tinyglobby-0.2.15.tgz",
+      "integrity": "sha512-j2Zq4NyQYG5XMST4cbs02Ak8iJUdxRM0XI5QyxXuZOzKOINmWurp3smXu3y5wDcJrptwpSjgXHzIQxR0omXljQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "fdir": "^6.5.0",
+        "picomatch": "^4.0.3"
+      },
+      "engines": {
+        "node": ">=12.0.0"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/SuperchupuDev"
+      }
+    },
+    "node_modules/tinypool": {
+      "version": "1.1.1",
+      "resolved": "https://registry.npmjs.org/tinypool/-/tinypool-1.1.1.tgz",
+      "integrity": "sha512-Zba82s87IFq9A9XmjiX5uZA/ARWDrB03OHlq+Vw1fSdt0I+4/Kutwy8BP4Y/y/aORMo61FQ0vIb5j44vSo5Pkg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": "^18.0.0 || >=20.0.0"
+      }
+    },
+    "node_modules/tinyrainbow": {
+      "version": "2.0.0",
+      "resolved": "https://registry.npmjs.org/tinyrainbow/-/tinyrainbow-2.0.0.tgz",
+      "integrity": "sha512-op4nsTR47R6p0vMUUoYl/a+ljLFVtlfaXkLQmqfLR1qHma1h/ysYk4hEXZ880bf2CYgTskvTa/e196Vd5dDQXw==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=14.0.0"
+      }
+    },
+    "node_modules/tinyspy": {
+      "version": "4.0.4",
+      "resolved": "https://registry.npmjs.org/tinyspy/-/tinyspy-4.0.4.tgz",
+      "integrity": "sha512-azl+t0z7pw/z958Gy9svOTuzqIk6xq+NSheJzn5MMWtWTFywIacg2wUlzKFGtt3cthx0r2SxMK0yzJOR0IES7Q==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=14.0.0"
+      }
+    },
+    "node_modules/tldts": {
+      "version": "6.1.86",
+      "resolved": "https://registry.npmjs.org/tldts/-/tldts-6.1.86.tgz",
+      "integrity": "sha512-WMi/OQ2axVTf/ykqCQgXiIct+mSQDFdH2fkwhPwgEwvJ1kSzZRiinb0zF2Xb8u4+OqPChmyI6MEu4EezNJz+FQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "tldts-core": "^6.1.86"
+      },
+      "bin": {
+        "tldts": "bin/cli.js"
+      }
+    },
+    "node_modules/tldts-core": {
+      "version": "6.1.86",
+      "resolved": "https://registry.npmjs.org/tldts-core/-/tldts-core-6.1.86.tgz",
+      "integrity": "sha512-Je6p7pkk+KMzMv2XXKmAE3McmolOQFdxkKw0R8EYNr7sELW46JqnNeTX8ybPiQgvg1ymCoF8LXs5fzFaZvJPTA==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/to-regex-range": {
+      "version": "5.0.1",
+      "resolved": "https://registry.npmjs.org/to-regex-range/-/to-regex-range-5.0.1.tgz",
+      "integrity": "sha512-65P7iz6X5yEr1cwcgvQxbbIw7Uk3gOy5dIdtZ4rDveLqhrdJP+Li/Hx6tyK0NEb+2GCyneCMJiGqrADCSNk8sQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "is-number": "^7.0.0"
+      },
+      "engines": {
+        "node": ">=8.0"
+      }
+    },
+    "node_modules/tough-cookie": {
+      "version": "5.1.2",
+      "resolved": "https://registry.npmjs.org/tough-cookie/-/tough-cookie-5.1.2.tgz",
+      "integrity": "sha512-FVDYdxtnj0G6Qm/DhNPSb8Ju59ULcup3tuJxkFb5K8Bv2pUXILbf0xZWU8PX8Ov19OXljbUyveOFwRMwkXzO+A==",
+      "dev": true,
+      "license": "BSD-3-Clause",
+      "dependencies": {
+        "tldts": "^6.1.32"
+      },
+      "engines": {
+        "node": ">=16"
+      }
+    },
+    "node_modules/tr46": {
+      "version": "5.1.1",
+      "resolved": "https://registry.npmjs.org/tr46/-/tr46-5.1.1.tgz",
+      "integrity": "sha512-hdF5ZgjTqgAntKkklYw0R03MG2x/bSzTtkxmIRw/sTNV8YXsCJ1tfLAX23lhxhHJlEf3CRCOCGGWw3vI3GaSPw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "punycode": "^2.3.1"
+      },
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/ts-api-utils": {
+      "version": "2.5.0",
+      "resolved": "https://registry.npmjs.org/ts-api-utils/-/ts-api-utils-2.5.0.tgz",
+      "integrity": "sha512-OJ/ibxhPlqrMM0UiNHJ/0CKQkoKF243/AEmplt3qpRgkW8VG7IfOS41h7V8TjITqdByHzrjcS/2si+y4lIh8NA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=18.12"
+      },
+      "peerDependencies": {
+        "typescript": ">=4.8.4"
+      }
+    },
+    "node_modules/ts-morph": {
+      "version": "14.0.0",
+      "resolved": "https://registry.npmjs.org/ts-morph/-/ts-morph-14.0.0.tgz",
+      "integrity": "sha512-tO8YQ1dP41fw8GVmeQAdNsD8roZi1JMqB7YwZrqU856DvmG5/710e41q2XauzTYrygH9XmMryaFeLo+kdCziyA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@ts-morph/common": "~0.13.0",
+        "code-block-writer": "^11.0.0"
+      }
+    },
+    "node_modules/type-check": {
+      "version": "0.4.0",
+      "resolved": "https://registry.npmjs.org/type-check/-/type-check-0.4.0.tgz",
+      "integrity": "sha512-XleUoc9uwGXqjWwXaUTZAmzMcFZ5858QA2vvx1Ur5xIcixXIP+8LnFDgRplU30us6teqdlskFfu+ae4K79Ooew==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "prelude-ls": "^1.2.1"
+      },
+      "engines": {
+        "node": ">= 0.8.0"
+      }
+    },
+    "node_modules/typescript": {
+      "version": "5.9.3",
+      "resolved": "https://registry.npmjs.org/typescript/-/typescript-5.9.3.tgz",
+      "integrity": "sha512-jl1vZzPDinLr9eUt3J/t7V6FgNEw9QjvBPdysz9KfQDD41fQrC2Y4vKQdiaUpFT4bXlb1RHhLpp8wtm6M5TgSw==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "bin": {
+        "tsc": "bin/tsc",
+        "tsserver": "bin/tsserver"
+      },
+      "engines": {
+        "node": ">=14.17"
+      }
+    },
+    "node_modules/typescript-eslint": {
+      "version": "8.57.2",
+      "resolved": "https://registry.npmjs.org/typescript-eslint/-/typescript-eslint-8.57.2.tgz",
+      "integrity": "sha512-VEPQ0iPgWO/sBaZOU1xo4nuNdODVOajPnTIbog2GKYr31nIlZ0fWPoCQgGfF3ETyBl1vn63F/p50Um9Z4J8O8A==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@typescript-eslint/eslint-plugin": "8.57.2",
+        "@typescript-eslint/parser": "8.57.2",
+        "@typescript-eslint/typescript-estree": "8.57.2",
+        "@typescript-eslint/utils": "8.57.2"
+      },
+      "engines": {
+        "node": "^18.18.0 || ^20.9.0 || >=21.1.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/typescript-eslint"
+      },
+      "peerDependencies": {
+        "eslint": "^8.57.0 || ^9.0.0 || ^10.0.0",
+        "typescript": ">=4.8.4 <6.0.0"
+      }
+    },
+    "node_modules/undici-types": {
+      "version": "7.18.2",
+      "resolved": "https://registry.npmjs.org/undici-types/-/undici-types-7.18.2.tgz",
+      "integrity": "sha512-AsuCzffGHJybSaRrmr5eHr81mwJU3kjw6M+uprWvCXiNeN9SOGwQ3Jn8jb8m3Z6izVgknn1R0FTCEAP2QrLY/w==",
+      "license": "MIT"
+    },
+    "node_modules/uri-js": {
+      "version": "4.4.1",
+      "resolved": "https://registry.npmjs.org/uri-js/-/uri-js-4.4.1.tgz",
+      "integrity": "sha512-7rKUyy33Q1yc98pQ1DAmLtwX109F7TIfWlW1Ydo8Wl1ii1SeHieeh0HHfPeL2fMXK6z0s8ecKs9frCuLJvndBg==",
+      "dev": true,
+      "license": "BSD-2-Clause",
+      "dependencies": {
+        "punycode": "^2.1.0"
+      }
+    },
+    "node_modules/vite": {
+      "version": "7.3.2",
+      "resolved": "https://registry.npmjs.org/vite/-/vite-7.3.2.tgz",
+      "integrity": "sha512-Bby3NOsna2jsjfLVOHKes8sGwgl4TT0E6vvpYgnAYDIF/tie7MRaFthmKuHx1NSXjiTueXH3do80FMQgvEktRg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "esbuild": "^0.27.0",
+        "fdir": "^6.5.0",
+        "picomatch": "^4.0.3",
+        "postcss": "^8.5.6",
+        "rollup": "^4.43.0",
+        "tinyglobby": "^0.2.15"
+      },
+      "bin": {
+        "vite": "bin/vite.js"
+      },
+      "engines": {
+        "node": "^20.19.0 || >=22.12.0"
+      },
+      "funding": {
+        "url": "https://github.com/vitejs/vite?sponsor=1"
+      },
+      "optionalDependencies": {
+        "fsevents": "~2.3.3"
+      },
+      "peerDependencies": {
+        "@types/node": "^20.19.0 || >=22.12.0",
+        "jiti": ">=1.21.0",
+        "less": "^4.0.0",
+        "lightningcss": "^1.21.0",
+        "sass": "^1.70.0",
+        "sass-embedded": "^1.70.0",
+        "stylus": ">=0.54.8",
+        "sugarss": "^5.0.0",
+        "terser": "^5.16.0",
+        "tsx": "^4.8.1",
+        "yaml": "^2.4.2"
+      },
+      "peerDependenciesMeta": {
+        "@types/node": {
+          "optional": true
+        },
+        "jiti": {
+          "optional": true
+        },
+        "less": {
+          "optional": true
+        },
+        "lightningcss": {
+          "optional": true
+        },
+        "sass": {
+          "optional": true
+        },
+        "sass-embedded": {
+          "optional": true
+        },
+        "stylus": {
+          "optional": true
+        },
+        "sugarss": {
+          "optional": true
+        },
+        "terser": {
+          "optional": true
+        },
+        "tsx": {
+          "optional": true
+        },
+        "yaml": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/vite-node": {
+      "version": "3.2.4",
+      "resolved": "https://registry.npmjs.org/vite-node/-/vite-node-3.2.4.tgz",
+      "integrity": "sha512-EbKSKh+bh1E1IFxeO0pg1n4dvoOTt0UDiXMd/qn++r98+jPO1xtJilvXldeuQ8giIB5IkpjCgMleHMNEsGH6pg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "cac": "^6.7.14",
+        "debug": "^4.4.1",
+        "es-module-lexer": "^1.7.0",
+        "pathe": "^2.0.3",
+        "vite": "^5.0.0 || ^6.0.0 || ^7.0.0-0"
+      },
+      "bin": {
+        "vite-node": "vite-node.mjs"
+      },
+      "engines": {
+        "node": "^18.0.0 || ^20.0.0 || >=22.0.0"
+      },
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/aix-ppc64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/aix-ppc64/-/aix-ppc64-0.27.4.tgz",
+      "integrity": "sha512-cQPwL2mp2nSmHHJlCyoXgHGhbEPMrEEU5xhkcy3Hs/O7nGZqEpZ2sUtLaL9MORLtDfRvVl2/3PAuEkYZH0Ty8Q==",
+      "cpu": [
+        "ppc64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "aix"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/android-arm": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/android-arm/-/android-arm-0.27.4.tgz",
+      "integrity": "sha512-X9bUgvxiC8CHAGKYufLIHGXPJWnr0OCdR0anD2e21vdvgCI8lIfqFbnoeOz7lBjdrAGUhqLZLcQo6MLhTO2DKQ==",
+      "cpu": [
+        "arm"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "android"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/android-arm64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/android-arm64/-/android-arm64-0.27.4.tgz",
+      "integrity": "sha512-gdLscB7v75wRfu7QSm/zg6Rx29VLdy9eTr2t44sfTW7CxwAtQghZ4ZnqHk3/ogz7xao0QAgrkradbBzcqFPasw==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "android"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/android-x64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/android-x64/-/android-x64-0.27.4.tgz",
+      "integrity": "sha512-PzPFnBNVF292sfpfhiyiXCGSn9HZg5BcAz+ivBuSsl6Rk4ga1oEXAamhOXRFyMcjwr2DVtm40G65N3GLeH1Lvw==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "android"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/darwin-arm64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/darwin-arm64/-/darwin-arm64-0.27.4.tgz",
+      "integrity": "sha512-b7xaGIwdJlht8ZFCvMkpDN6uiSmnxxK56N2GDTMYPr2/gzvfdQN8rTfBsvVKmIVY/X7EM+/hJKEIbbHs9oA4tQ==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "darwin"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/darwin-x64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/darwin-x64/-/darwin-x64-0.27.4.tgz",
+      "integrity": "sha512-sR+OiKLwd15nmCdqpXMnuJ9W2kpy0KigzqScqHI3Hqwr7IXxBp3Yva+yJwoqh7rE8V77tdoheRYataNKL4QrPw==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "darwin"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/freebsd-arm64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/freebsd-arm64/-/freebsd-arm64-0.27.4.tgz",
+      "integrity": "sha512-jnfpKe+p79tCnm4GVav68A7tUFeKQwQyLgESwEAUzyxk/TJr4QdGog9sqWNcUbr/bZt/O/HXouspuQDd9JxFSw==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "freebsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/freebsd-x64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/freebsd-x64/-/freebsd-x64-0.27.4.tgz",
+      "integrity": "sha512-2kb4ceA/CpfUrIcTUl1wrP/9ad9Atrp5J94Lq69w7UwOMolPIGrfLSvAKJp0RTvkPPyn6CIWrNy13kyLikZRZQ==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "freebsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/linux-arm": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-arm/-/linux-arm-0.27.4.tgz",
+      "integrity": "sha512-aBYgcIxX/wd5n2ys0yESGeYMGF+pv6g0DhZr3G1ZG4jMfruU9Tl1i2Z+Wnj9/KjGz1lTLCcorqE2viePZqj4Eg==",
+      "cpu": [
+        "arm"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/linux-arm64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-arm64/-/linux-arm64-0.27.4.tgz",
+      "integrity": "sha512-7nQOttdzVGth1iz57kxg9uCz57dxQLHWxopL6mYuYthohPKEK0vU0C3O21CcBK6KDlkYVcnDXY099HcCDXd9dA==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/linux-ia32": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-ia32/-/linux-ia32-0.27.4.tgz",
+      "integrity": "sha512-oPtixtAIzgvzYcKBQM/qZ3R+9TEUd1aNJQu0HhGyqtx6oS7qTpvjheIWBbes4+qu1bNlo2V4cbkISr8q6gRBFA==",
+      "cpu": [
+        "ia32"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/linux-loong64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-loong64/-/linux-loong64-0.27.4.tgz",
+      "integrity": "sha512-8mL/vh8qeCoRcFH2nM8wm5uJP+ZcVYGGayMavi8GmRJjuI3g1v6Z7Ni0JJKAJW+m0EtUuARb6Lmp4hMjzCBWzA==",
+      "cpu": [
+        "loong64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/linux-mips64el": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-mips64el/-/linux-mips64el-0.27.4.tgz",
+      "integrity": "sha512-1RdrWFFiiLIW7LQq9Q2NES+HiD4NyT8Itj9AUeCl0IVCA459WnPhREKgwrpaIfTOe+/2rdntisegiPWn/r/aAw==",
+      "cpu": [
+        "mips64el"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/linux-ppc64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-ppc64/-/linux-ppc64-0.27.4.tgz",
+      "integrity": "sha512-tLCwNG47l3sd9lpfyx9LAGEGItCUeRCWeAx6x2Jmbav65nAwoPXfewtAdtbtit/pJFLUWOhpv0FpS6GQAmPrHA==",
+      "cpu": [
+        "ppc64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/linux-riscv64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-riscv64/-/linux-riscv64-0.27.4.tgz",
+      "integrity": "sha512-BnASypppbUWyqjd1KIpU4AUBiIhVr6YlHx/cnPgqEkNoVOhHg+YiSVxM1RLfiy4t9cAulbRGTNCKOcqHrEQLIw==",
+      "cpu": [
+        "riscv64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/linux-s390x": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-s390x/-/linux-s390x-0.27.4.tgz",
+      "integrity": "sha512-+eUqgb/Z7vxVLezG8bVB9SfBie89gMueS+I0xYh2tJdw3vqA/0ImZJ2ROeWwVJN59ihBeZ7Tu92dF/5dy5FttA==",
+      "cpu": [
+        "s390x"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/linux-x64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/linux-x64/-/linux-x64-0.27.4.tgz",
+      "integrity": "sha512-S5qOXrKV8BQEzJPVxAwnryi2+Iq5pB40gTEIT69BQONqR7JH1EPIcQ/Uiv9mCnn05jff9umq/5nqzxlqTOg9NA==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "linux"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/netbsd-arm64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/netbsd-arm64/-/netbsd-arm64-0.27.4.tgz",
+      "integrity": "sha512-xHT8X4sb0GS8qTqiwzHqpY00C95DPAq7nAwX35Ie/s+LO9830hrMd3oX0ZMKLvy7vsonee73x0lmcdOVXFzd6Q==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "netbsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/netbsd-x64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/netbsd-x64/-/netbsd-x64-0.27.4.tgz",
+      "integrity": "sha512-RugOvOdXfdyi5Tyv40kgQnI0byv66BFgAqjdgtAKqHoZTbTF2QqfQrFwa7cHEORJf6X2ht+l9ABLMP0dnKYsgg==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "netbsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/openbsd-arm64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/openbsd-arm64/-/openbsd-arm64-0.27.4.tgz",
+      "integrity": "sha512-2MyL3IAaTX+1/qP0O1SwskwcwCoOI4kV2IBX1xYnDDqthmq5ArrW94qSIKCAuRraMgPOmG0RDTA74mzYNQA9ow==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "openbsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/openbsd-x64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/openbsd-x64/-/openbsd-x64-0.27.4.tgz",
+      "integrity": "sha512-u8fg/jQ5aQDfsnIV6+KwLOf1CmJnfu1ShpwqdwC0uA7ZPwFws55Ngc12vBdeUdnuWoQYx/SOQLGDcdlfXhYmXQ==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "openbsd"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/openharmony-arm64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/openharmony-arm64/-/openharmony-arm64-0.27.4.tgz",
+      "integrity": "sha512-JkTZrl6VbyO8lDQO3yv26nNr2RM2yZzNrNHEsj9bm6dOwwu9OYN28CjzZkH57bh4w0I2F7IodpQvUAEd1mbWXg==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "openharmony"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/sunos-x64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/sunos-x64/-/sunos-x64-0.27.4.tgz",
+      "integrity": "sha512-/gOzgaewZJfeJTlsWhvUEmUG4tWEY2Spp5M20INYRg2ZKl9QPO3QEEgPeRtLjEWSW8FilRNacPOg8R1uaYkA6g==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "sunos"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/win32-arm64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/win32-arm64/-/win32-arm64-0.27.4.tgz",
+      "integrity": "sha512-Z9SExBg2y32smoDQdf1HRwHRt6vAHLXcxD2uGgO/v2jK7Y718Ix4ndsbNMU/+1Qiem9OiOdaqitioZwxivhXYg==",
+      "cpu": [
+        "arm64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/win32-ia32": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/win32-ia32/-/win32-ia32-0.27.4.tgz",
+      "integrity": "sha512-DAyGLS0Jz5G5iixEbMHi5KdiApqHBWMGzTtMiJ72ZOLhbu/bzxgAe8Ue8CTS3n3HbIUHQz/L51yMdGMeoxXNJw==",
+      "cpu": [
+        "ia32"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/@esbuild/win32-x64": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/@esbuild/win32-x64/-/win32-x64-0.27.4.tgz",
+      "integrity": "sha512-+knoa0BDoeXgkNvvV1vvbZX4+hizelrkwmGJBdT17t8FNPwG2lKemmuMZlmaNQ3ws3DKKCxpb4zRZEIp3UxFCg==",
+      "cpu": [
+        "x64"
+      ],
+      "dev": true,
+      "license": "MIT",
+      "optional": true,
+      "os": [
+        "win32"
+      ],
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/vite/node_modules/esbuild": {
+      "version": "0.27.4",
+      "resolved": "https://registry.npmjs.org/esbuild/-/esbuild-0.27.4.tgz",
+      "integrity": "sha512-Rq4vbHnYkK5fws5NF7MYTU68FPRE1ajX7heQ/8QXXWqNgqqJ/GkmmyxIzUnf2Sr/bakf8l54716CcMGHYhMrrQ==",
+      "dev": true,
+      "hasInstallScript": true,
+      "license": "MIT",
+      "bin": {
+        "esbuild": "bin/esbuild"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "optionalDependencies": {
+        "@esbuild/aix-ppc64": "0.27.4",
+        "@esbuild/android-arm": "0.27.4",
+        "@esbuild/android-arm64": "0.27.4",
+        "@esbuild/android-x64": "0.27.4",
+        "@esbuild/darwin-arm64": "0.27.4",
+        "@esbuild/darwin-x64": "0.27.4",
+        "@esbuild/freebsd-arm64": "0.27.4",
+        "@esbuild/freebsd-x64": "0.27.4",
+        "@esbuild/linux-arm": "0.27.4",
+        "@esbuild/linux-arm64": "0.27.4",
+        "@esbuild/linux-ia32": "0.27.4",
+        "@esbuild/linux-loong64": "0.27.4",
+        "@esbuild/linux-mips64el": "0.27.4",
+        "@esbuild/linux-ppc64": "0.27.4",
+        "@esbuild/linux-riscv64": "0.27.4",
+        "@esbuild/linux-s390x": "0.27.4",
+        "@esbuild/linux-x64": "0.27.4",
+        "@esbuild/netbsd-arm64": "0.27.4",
+        "@esbuild/netbsd-x64": "0.27.4",
+        "@esbuild/openbsd-arm64": "0.27.4",
+        "@esbuild/openbsd-x64": "0.27.4",
+        "@esbuild/openharmony-arm64": "0.27.4",
+        "@esbuild/sunos-x64": "0.27.4",
+        "@esbuild/win32-arm64": "0.27.4",
+        "@esbuild/win32-ia32": "0.27.4",
+        "@esbuild/win32-x64": "0.27.4"
+      }
+    },
+    "node_modules/vitest": {
+      "version": "3.2.4",
+      "resolved": "https://registry.npmjs.org/vitest/-/vitest-3.2.4.tgz",
+      "integrity": "sha512-LUCP5ev3GURDysTWiP47wRRUpLKMOfPh+yKTx3kVIEiu5KOMeqzpnYNsKyOoVrULivR8tLcks4+lga33Whn90A==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "@types/chai": "^5.2.2",
+        "@vitest/expect": "3.2.4",
+        "@vitest/mocker": "3.2.4",
+        "@vitest/pretty-format": "^3.2.4",
+        "@vitest/runner": "3.2.4",
+        "@vitest/snapshot": "3.2.4",
+        "@vitest/spy": "3.2.4",
+        "@vitest/utils": "3.2.4",
+        "chai": "^5.2.0",
+        "debug": "^4.4.1",
+        "expect-type": "^1.2.1",
+        "magic-string": "^0.30.17",
+        "pathe": "^2.0.3",
+        "picomatch": "^4.0.2",
+        "std-env": "^3.9.0",
+        "tinybench": "^2.9.0",
+        "tinyexec": "^0.3.2",
+        "tinyglobby": "^0.2.14",
+        "tinypool": "^1.1.1",
+        "tinyrainbow": "^2.0.0",
+        "vite": "^5.0.0 || ^6.0.0 || ^7.0.0-0",
+        "vite-node": "3.2.4",
+        "why-is-node-running": "^2.3.0"
+      },
+      "bin": {
+        "vitest": "vitest.mjs"
+      },
+      "engines": {
+        "node": "^18.0.0 || ^20.0.0 || >=22.0.0"
+      },
+      "funding": {
+        "url": "https://opencollective.com/vitest"
+      },
+      "peerDependencies": {
+        "@edge-runtime/vm": "*",
+        "@types/debug": "^4.1.12",
+        "@types/node": "^18.0.0 || ^20.0.0 || >=22.0.0",
+        "@vitest/browser": "3.2.4",
+        "@vitest/ui": "3.2.4",
+        "happy-dom": "*",
+        "jsdom": "*"
+      },
+      "peerDependenciesMeta": {
+        "@edge-runtime/vm": {
+          "optional": true
+        },
+        "@types/debug": {
+          "optional": true
+        },
+        "@types/node": {
+          "optional": true
+        },
+        "@vitest/browser": {
+          "optional": true
+        },
+        "@vitest/ui": {
+          "optional": true
+        },
+        "happy-dom": {
+          "optional": true
+        },
+        "jsdom": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/vitest/node_modules/tinyexec": {
+      "version": "0.3.2",
+      "resolved": "https://registry.npmjs.org/tinyexec/-/tinyexec-0.3.2.tgz",
+      "integrity": "sha512-KQQR9yN7R5+OSwaK0XQoj22pwHoTlgYqmUscPYoknOoWCWfj/5/ABTMRi69FrKU5ffPVh5QcFikpWJI/P1ocHA==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/w3c-xmlserializer": {
+      "version": "5.0.0",
+      "resolved": "https://registry.npmjs.org/w3c-xmlserializer/-/w3c-xmlserializer-5.0.0.tgz",
+      "integrity": "sha512-o8qghlI8NZHU1lLPrpi2+Uq7abh4GGPpYANlalzWxyWteJOCsr/P+oPBA49TOLu5FTZO4d3F9MnWJfiMo4BkmA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "xml-name-validator": "^5.0.0"
+      },
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/webidl-conversions": {
+      "version": "7.0.0",
+      "resolved": "https://registry.npmjs.org/webidl-conversions/-/webidl-conversions-7.0.0.tgz",
+      "integrity": "sha512-VwddBukDzu71offAQR975unBIGqfKZpM+8ZX6ySk8nYhVoo5CYaZyzt3YBvYtRtO+aoGlqxPg/B87NGVZ/fu6g==",
+      "dev": true,
+      "license": "BSD-2-Clause",
+      "engines": {
+        "node": ">=12"
+      }
+    },
+    "node_modules/whatwg-encoding": {
+      "version": "3.1.1",
+      "resolved": "https://registry.npmjs.org/whatwg-encoding/-/whatwg-encoding-3.1.1.tgz",
+      "integrity": "sha512-6qN4hJdMwfYBtE3YBTTHhoeuUrDBPZmbQaxWAqSALV/MeEnR5z1xd8UKud2RAkFoPkmB+hli1TZSnyi84xz1vQ==",
+      "deprecated": "Use @exodus/bytes instead for a more spec-conformant and faster implementation",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "iconv-lite": "0.6.3"
+      },
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/whatwg-encoding/node_modules/iconv-lite": {
+      "version": "0.6.3",
+      "resolved": "https://registry.npmjs.org/iconv-lite/-/iconv-lite-0.6.3.tgz",
+      "integrity": "sha512-4fCk79wshMdzMp2rH06qWrJE4iolqLhCUH+OiuIgU++RB0+94NlDL81atO7GX55uUKueo0txHNtvEyI6D7WdMw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "safer-buffer": ">= 2.1.2 < 3.0.0"
+      },
+      "engines": {
+        "node": ">=0.10.0"
+      }
+    },
+    "node_modules/whatwg-mimetype": {
+      "version": "4.0.0",
+      "resolved": "https://registry.npmjs.org/whatwg-mimetype/-/whatwg-mimetype-4.0.0.tgz",
+      "integrity": "sha512-QaKxh0eNIi2mE9p2vEdzfagOKHCcj1pJ56EEHGQOVxp8r9/iszLUUV7v89x9O1p/T+NlTM5W7jW6+cz4Fq1YVg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/whatwg-url": {
+      "version": "14.2.0",
+      "resolved": "https://registry.npmjs.org/whatwg-url/-/whatwg-url-14.2.0.tgz",
+      "integrity": "sha512-De72GdQZzNTUBBChsXueQUnPKDkg/5A5zp7pFDuQAj5UFoENpiACU0wlCvzpAGnTkj++ihpKwKyYewn/XNUbKw==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "tr46": "^5.1.0",
+        "webidl-conversions": "^7.0.0"
+      },
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/which": {
+      "version": "2.0.2",
+      "resolved": "https://registry.npmjs.org/which/-/which-2.0.2.tgz",
+      "integrity": "sha512-BLI3Tl1TW3Pvl70l3yq3Y64i+awpwXqsGBYWkkqMtnbXgrMD+yj7rhW0kuEDxzJaYXGjEW5ogapKNMEKNMjibA==",
+      "dev": true,
+      "license": "ISC",
+      "dependencies": {
+        "isexe": "^2.0.0"
+      },
+      "bin": {
+        "node-which": "bin/node-which"
+      },
+      "engines": {
+        "node": ">= 8"
+      }
+    },
+    "node_modules/why-is-node-running": {
+      "version": "2.3.0",
+      "resolved": "https://registry.npmjs.org/why-is-node-running/-/why-is-node-running-2.3.0.tgz",
+      "integrity": "sha512-hUrmaWBdVDcxvYqnyh09zunKzROWjbZTiNy8dBEjkS7ehEDQibXJ7XvlmtbwuTclUiIyN+CyXQD4Vmko8fNm8w==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "siginfo": "^2.0.0",
+        "stackback": "0.0.2"
+      },
+      "bin": {
+        "why-is-node-running": "cli.js"
+      },
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/word-wrap": {
+      "version": "1.2.5",
+      "resolved": "https://registry.npmjs.org/word-wrap/-/word-wrap-1.2.5.tgz",
+      "integrity": "sha512-BN22B5eaMMI9UMtjrGd5g5eCYPpCPDUy0FJXbYsaT5zYxjFOckS53SQDE3pWkVoWpHXVb3BrYcEN4Twa55B5cA==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=0.10.0"
+      }
+    },
+    "node_modules/wrap-ansi": {
+      "version": "9.0.2",
+      "resolved": "https://registry.npmjs.org/wrap-ansi/-/wrap-ansi-9.0.2.tgz",
+      "integrity": "sha512-42AtmgqjV+X1VpdOfyTGOYRi0/zsoLqtXQckTmqTeybT+BDIbM/Guxo7x3pE2vtpr1ok6xRqM9OpBe+Jyoqyww==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "ansi-styles": "^6.2.1",
+        "string-width": "^7.0.0",
+        "strip-ansi": "^7.1.0"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/wrap-ansi?sponsor=1"
+      }
+    },
+    "node_modules/wrap-ansi-cjs": {
+      "name": "wrap-ansi",
+      "version": "7.0.0",
+      "resolved": "https://registry.npmjs.org/wrap-ansi/-/wrap-ansi-7.0.0.tgz",
+      "integrity": "sha512-YVGIj2kamLSTxw6NsZjoBxfSwsn0ycdesmc4p+Q21c5zPuZ1pl+NfxVdxPtdHvmNVOQ6XSYG4AUtyt/Fi7D16Q==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "ansi-styles": "^4.0.0",
+        "string-width": "^4.1.0",
+        "strip-ansi": "^6.0.0"
+      },
+      "engines": {
+        "node": ">=10"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/wrap-ansi?sponsor=1"
+      }
+    },
+    "node_modules/wrap-ansi-cjs/node_modules/ansi-styles": {
+      "version": "4.3.0",
+      "resolved": "https://registry.npmjs.org/ansi-styles/-/ansi-styles-4.3.0.tgz",
+      "integrity": "sha512-zbB9rCJAT1rbjiVDb2hqKFHNYLxgtk8NURxZ3IZwD3F6NtxbXZQCnnSi1Lkx+IDohdPlFp222wVALIheZJQSEg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "color-convert": "^2.0.1"
+      },
+      "engines": {
+        "node": ">=8"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/ansi-styles?sponsor=1"
+      }
+    },
+    "node_modules/wrap-ansi-cjs/node_modules/emoji-regex": {
+      "version": "8.0.0",
+      "resolved": "https://registry.npmjs.org/emoji-regex/-/emoji-regex-8.0.0.tgz",
+      "integrity": "sha512-MSjYzcWNOA0ewAHpz0MxpYFvwg6yjy1NG3xteoqz644VCo/RPgnr1/GGt+ic3iJTzQ8Eu3TdM14SawnVUmGE6A==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/wrap-ansi-cjs/node_modules/is-fullwidth-code-point": {
+      "version": "3.0.0",
+      "resolved": "https://registry.npmjs.org/is-fullwidth-code-point/-/is-fullwidth-code-point-3.0.0.tgz",
+      "integrity": "sha512-zymm5+u+sCsSWyD9qNaejV3DFvhCKclKdizYaJUuHA83RLjb7nSuGnddCHGv0hk+KY7BMAlsWeK4Ueg6EV6XQg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/wrap-ansi-cjs/node_modules/string-width": {
+      "version": "4.2.3",
+      "resolved": "https://registry.npmjs.org/string-width/-/string-width-4.2.3.tgz",
+      "integrity": "sha512-wKyQRQpjJ0sIp62ErSZdGsjMJWsap5oRNihHhu6G7JVO/9jIB6UyevL+tXuOqrng8j/cxKTWyWUwvSTriiZz/g==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "emoji-regex": "^8.0.0",
+        "is-fullwidth-code-point": "^3.0.0",
+        "strip-ansi": "^6.0.1"
+      },
+      "engines": {
+        "node": ">=8"
+      }
+    },
+    "node_modules/wrap-ansi/node_modules/ansi-regex": {
+      "version": "6.2.2",
+      "resolved": "https://registry.npmjs.org/ansi-regex/-/ansi-regex-6.2.2.tgz",
+      "integrity": "sha512-Bq3SmSpyFHaWjPk8If9yc6svM8c56dB5BAtW4Qbw5jHTwwXXcTLoRMkpDJp6VL0XzlWaCHTXrkFURMYmD0sLqg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=12"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/ansi-regex?sponsor=1"
+      }
+    },
+    "node_modules/wrap-ansi/node_modules/string-width": {
+      "version": "7.2.0",
+      "resolved": "https://registry.npmjs.org/string-width/-/string-width-7.2.0.tgz",
+      "integrity": "sha512-tsaTIkKW9b4N+AEj+SVA+WhJzV7/zMhcSu78mLKWSk7cXMOSHsBKFWUs0fWwq8QyK3MgJBQRX6Gbi4kYbdvGkQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "emoji-regex": "^10.3.0",
+        "get-east-asian-width": "^1.0.0",
+        "strip-ansi": "^7.1.0"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "node_modules/wrap-ansi/node_modules/strip-ansi": {
+      "version": "7.2.0",
+      "resolved": "https://registry.npmjs.org/strip-ansi/-/strip-ansi-7.2.0.tgz",
+      "integrity": "sha512-yDPMNjp4WyfYBkHnjIRLfca1i6KMyGCtsVgoKe/z1+6vukgaENdgGBZt+ZmKPc4gavvEZ5OgHfHdrazhgNyG7w==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "ansi-regex": "^6.2.2"
+      },
+      "engines": {
+        "node": ">=12"
+      },
+      "funding": {
+        "url": "https://github.com/chalk/strip-ansi?sponsor=1"
+      }
+    },
+    "node_modules/ws": {
+      "version": "8.19.0",
+      "resolved": "https://registry.npmjs.org/ws/-/ws-8.19.0.tgz",
+      "integrity": "sha512-blAT2mjOEIi0ZzruJfIhb3nps74PRWTCz1IjglWEEpQl5XS/UNama6u2/rjFkDDouqr4L67ry+1aGIALViWjDg==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=10.0.0"
+      },
+      "peerDependencies": {
+        "bufferutil": "^4.0.1",
+        "utf-8-validate": ">=5.0.2"
+      },
+      "peerDependenciesMeta": {
+        "bufferutil": {
+          "optional": true
+        },
+        "utf-8-validate": {
+          "optional": true
+        }
+      }
+    },
+    "node_modules/xml-name-validator": {
+      "version": "5.0.0",
+      "resolved": "https://registry.npmjs.org/xml-name-validator/-/xml-name-validator-5.0.0.tgz",
+      "integrity": "sha512-EvGK8EJ3DhaHfbRlETOWAS5pO9MZITeauHKJyb8wyajUfQUenkIg2MvLDTZ4T/TgIcm3HU0TFBgWWboAZ30UHg==",
+      "dev": true,
+      "license": "Apache-2.0",
+      "engines": {
+        "node": ">=18"
+      }
+    },
+    "node_modules/xmlchars": {
+      "version": "2.2.0",
+      "resolved": "https://registry.npmjs.org/xmlchars/-/xmlchars-2.2.0.tgz",
+      "integrity": "sha512-JZnDKK8B0RCDw84FNdDAIpZK+JuJw+s7Lz8nksI7SIuU3UXJJslUthsi+uWBUYOwPFwW7W7PRLRfUKpxjtjFCw==",
+      "dev": true,
+      "license": "MIT"
+    },
+    "node_modules/yaml": {
+      "version": "2.8.3",
+      "resolved": "https://registry.npmjs.org/yaml/-/yaml-2.8.3.tgz",
+      "integrity": "sha512-AvbaCLOO2Otw/lW5bmh9d/WEdcDFdQp2Z2ZUH3pX9U2ihyUY0nvLv7J6TrWowklRGPYbB/IuIMfYgxaCPg5Bpg==",
+      "dev": true,
+      "license": "ISC",
+      "bin": {
+        "yaml": "bin.mjs"
+      },
+      "engines": {
+        "node": ">= 14.6"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/eemeli"
+      }
+    },
+    "node_modules/yocto-queue": {
+      "version": "0.1.0",
+      "resolved": "https://registry.npmjs.org/yocto-queue/-/yocto-queue-0.1.0.tgz",
+      "integrity": "sha512-rVksvsnNCdJ/ohGc6xgPwyN8eheCxsiLM8mxuE/t/mOVqJewPuO1miLpTHQiRgTKCLexL4MeAFVagts7HmNZ2Q==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">=10"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
+    "packages/core": {
+      "name": "@paddleocr/paddleocr-js",
+      "version": "0.3.3",
+      "license": "Apache-2.0",
+      "dependencies": {
+        "@techstark/opencv-js": "^4.10.0-release.1",
+        "clipper-lib": "^6.4.2",
+        "js-yaml": "^4.1.0",
+        "onnxruntime-web": "^1.22.0"
+      },
+      "devDependencies": {
+        "@types/js-yaml": "^4.0.9",
+        "typescript": "^5.9.3",
+        "vite": "^6.4.1",
+        "vite-plugin-dts": "^1.0.5"
+      }
+    },
+    "packages/core/node_modules/fs-extra": {
+      "version": "10.1.0",
+      "resolved": "https://registry.npmjs.org/fs-extra/-/fs-extra-10.1.0.tgz",
+      "integrity": "sha512-oRXApq54ETRj4eMiFzGnHWGy+zo5raudjuxN0b8H7s/RU2oW0Wvsx9O0ACRN/kRq9E8Vu/ReskGB5o3ji+FzHQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "graceful-fs": "^4.2.0",
+        "jsonfile": "^6.0.1",
+        "universalify": "^2.0.0"
+      },
+      "engines": {
+        "node": ">=12"
+      }
+    },
+    "packages/core/node_modules/jsonfile": {
+      "version": "6.2.0",
+      "resolved": "https://registry.npmjs.org/jsonfile/-/jsonfile-6.2.0.tgz",
+      "integrity": "sha512-FGuPw30AdOIUTRMC2OMRtQV+jkVj2cfPqSeWXv1NEAJ1qZ5zb1X6z1mFhbfOB/iy3ssJCD+3KuZ8r8C3uVFlAg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "universalify": "^2.0.0"
+      },
+      "optionalDependencies": {
+        "graceful-fs": "^4.1.6"
+      }
+    },
+    "packages/core/node_modules/universalify": {
+      "version": "2.0.1",
+      "resolved": "https://registry.npmjs.org/universalify/-/universalify-2.0.1.tgz",
+      "integrity": "sha512-gptHNQghINnc/vTGIk0SOFGFNXw7JVrlRUtConJRlvaw6DuX0wO5Jeko9sWrMBhh+PsYAZ7oXAiOnf/UKogyiw==",
+      "dev": true,
+      "license": "MIT",
+      "engines": {
+        "node": ">= 10.0.0"
+      }
+    },
+    "packages/core/node_modules/vite": {
+      "version": "6.4.2",
+      "resolved": "https://registry.npmjs.org/vite/-/vite-6.4.2.tgz",
+      "integrity": "sha512-2N/55r4JDJ4gdrCvGgINMy+HH3iRpNIz8K6SFwVsA+JbQScLiC+clmAxBgwiSPgcG9U15QmvqCGWzMbqda5zGQ==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "esbuild": "^0.25.0",
+        "fdir": "^6.4.4",
+        "picomatch": "^4.0.2",
+        "postcss": "^8.5.3",
+        "rollup": "^4.34.9",
+        "tinyglobby": "^0.2.13"
+      },
+      "bin": {
+        "vite": "bin/vite.js"
+      },
+      "engines": {
+        "node": "^18.0.0 || ^20.0.0 || >=22.0.0"
+      },
+      "funding": {
+        "url": "https://github.com/vitejs/vite?sponsor=1"
+      },
+      "optionalDependencies": {
+        "fsevents": "~2.3.3"
+      },
+      "peerDependencies": {
+        "@types/node": "^18.0.0 || ^20.0.0 || >=22.0.0",
+        "jiti": ">=1.21.0",
+        "less": "*",
+        "lightningcss": "^1.21.0",
+        "sass": "*",
+        "sass-embedded": "*",
+        "stylus": "*",
+        "sugarss": "*",
+        "terser": "^5.16.0",
+        "tsx": "^4.8.1",
+        "yaml": "^2.4.2"
+      },
+      "peerDependenciesMeta": {
+        "@types/node": {
+          "optional": true
+        },
+        "jiti": {
+          "optional": true
+        },
+        "less": {
+          "optional": true
+        },
+        "lightningcss": {
+          "optional": true
+        },
+        "sass": {
+          "optional": true
+        },
+        "sass-embedded": {
+          "optional": true
+        },
+        "stylus": {
+          "optional": true
+        },
+        "sugarss": {
+          "optional": true
+        },
+        "terser": {
+          "optional": true
+        },
+        "tsx": {
+          "optional": true
+        },
+        "yaml": {
+          "optional": true
+        }
+      }
+    },
+    "packages/core/node_modules/vite-plugin-dts": {
+      "version": "1.0.5",
+      "resolved": "https://registry.npmjs.org/vite-plugin-dts/-/vite-plugin-dts-1.0.5.tgz",
+      "integrity": "sha512-gTm9f+yUK88fNngpxdQIDpGk7oeWnD01fnKOmS2Z9sTRKTmDfMowXGE78g1mxfNMApnt99RNS0CNq9qc+RMCQA==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "chalk": "^4.1.2",
+        "debug": "^4.3.4",
+        "fast-glob": "^3.2.11",
+        "fs-extra": "^10.0.1",
+        "ts-morph": "^14.0.0"
+      },
+      "engines": {
+        "node": ">=12.0.0"
+      },
+      "peerDependencies": {
+        "vite": ">=2.4.4"
+      }
+    }
+  }
+}
diff --git a/paddleocr-js/package.json b/paddleocr-js/package.json
new file mode 100644
index 0000000..b35634b
--- /dev/null
+++ b/paddleocr-js/package.json
@@ -0,0 +1,68 @@
+{
+  "name": "paddleocr-js-workspace",
+  "version": "0.1.0",
+  "private": true,
+  "type": "module",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/PaddlePaddle/PaddleOCR.git",
+    "directory": "paddleocr-js"
+  },
+  "homepage": "https://github.com/PaddlePaddle/PaddleOCR/tree/main/paddleocr-js",
+  "bugs": {
+    "url": "https://github.com/PaddlePaddle/PaddleOCR/issues"
+  },
+  "keywords": [
+    "paddleocr",
+    "ocr",
+    "monorepo",
+    "browser-ocr",
+    "typescript"
+  ],
+  "engines": {
+    "node": ">=20.11"
+  },
+  "workspaces": [
+    "packages/*",
+    "apps/*"
+  ],
+  "scripts": {
+    "dev:demo": "npm run dev --workspace apps/demo --",
+    "build": "npm run build:sdk && npm run build:demo",
+    "build:demo": "npm run build --workspace apps/demo",
+    "build:sdk": "npm run build --workspace packages/core",
+    "preview:demo": "npm run preview --workspace apps/demo --",
+    "lint": "eslint .",
+    "lint:fix": "eslint . --fix",
+    "typecheck": "npm run typecheck --workspaces --if-present",
+    "test": "vitest run",
+    "test:coverage": "vitest run --coverage",
+    "test:watch": "vitest",
+    "format": "prettier . --write",
+    "format:check": "prettier . --check",
+    "clean": "rm -rf packages/*/dist apps/*/dist",
+    "check": "npm run format:check && npm run lint && npm run build:sdk && npm run typecheck && npm run test && npm run build:demo",
+    "release": "npm run build:sdk && npm publish --workspace packages/core"
+  },
+  "devDependencies": {
+    "@eslint/js": "^10.0.1",
+    "@vitest/coverage-v8": "^3.2.4",
+    "eslint": "^10.0.2",
+    "globals": "^17.4.0",
+    "husky": "^9.1.7",
+    "jsdom": "^26.1.0",
+    "lint-staged": "^16.3.2",
+    "prettier": "^3.8.1",
+    "typescript-eslint": "^8.57.2",
+    "vitest": "^3.2.4"
+  },
+  "lint-staged": {
+    "*.{js,ts}": [
+      "eslint --fix",
+      "prettier --write"
+    ],
+    "*.{json,md,html,css,yaml,yml}": [
+      "prettier --write"
+    ]
+  }
+}
diff --git a/paddleocr-js/packages/core/README.md b/paddleocr-js/packages/core/README.md
new file mode 100644
index 0000000..c966b2b
--- /dev/null
+++ b/paddleocr-js/packages/core/README.md
@@ -0,0 +1,254 @@
+# PaddleOCR.js SDK
+
+English | [简体中文](README_cn.md)
+
+`@paddleocr/paddleocr-js` is the browser SDK package for running PaddleOCR pipelines in the frontend.
+
+## Install
+
+```bash
+npm install @paddleocr/paddleocr-js
+```
+
+## Quick Start
+
+```js
+import { PaddleOCR } from "@paddleocr/paddleocr-js";
+
+const ocr = await PaddleOCR.create({
+  lang: "ch",
+  ocrVersion: "PP-OCRv5",
+  ortOptions: {
+    backend: "auto"
+  }
+});
+
+const [result] = await ocr.predict(fileOrBlob);
+console.log(result.items);
+```
+
+`predict` resolves to an **array** of `OcrResult` (one per input image). A single `Blob` / `File` still produces a one-element array—use destructuring or `results[0]`.
+
+## Construction Options
+
+There are two main construction styles:
+
+### 1. Direct parameters
+
+With direct parameters, you can specify models and configure inference batch sizes, ORT options, and other runtime settings.
+
+**Model selection** — `lang` + `ocrVersion`:
+
+```js
+await PaddleOCR.create({
+  lang: "ch",
+  ocrVersion: "PP-OCRv5"
+});
+```
+
+`ocrVersion: "PP-OCRv6"` maps supported `lang` values to the built-in **PP-OCRv6_small** det/rec pair. For **PP-OCRv6_tiny**, pass explicit model names:
+
+```js
+await PaddleOCR.create({
+  textDetectionModelName: "PP-OCRv6_tiny_det",
+  textRecognitionModelName: "PP-OCRv6_tiny_rec"
+});
+```
+
+**Model selection** — explicit model names:
+
+```js
+await PaddleOCR.create({
+  textDetectionModelName: "PP-OCRv5_mobile_det",
+  textRecognitionModelName: "PP-OCRv5_mobile_rec"
+});
+```
+
+**Custom models** — provide model names and asset URLs for detection and recognition:
+
+```js
+await PaddleOCR.create({
+  textDetectionModelName: "my_det_model",
+  textDetectionModelAsset: {
+    url: "https://example.com/models/my_det_model.tar"
+  },
+  textRecognitionModelName: "my_rec_model",
+  textRecognitionModelAsset: {
+    url: "https://example.com/models/my_rec_model.tar"
+  }
+});
+```
+
+**Custom model archive format and validation:**
+
+- The downloaded bytes must be an **uncompressed ustar `.tar`**. The SDK does **not** gunzip **`.tar.gz`**; gzip-compressed payloads will usually fail to parse.
+- The tar must contain **`inference.onnx`** and **`inference.yml`** (optionally under a subdirectory; matched by basename).
+- **`inference.yml`** must define **`model_name`**, and it must match `textDetectionModelName` / `textRecognitionModelName`. This is checked during initialization after load.
+
+Failures surface as **`Error`** during initialization (HTTP errors, missing tar entries, empty resources, missing/mismatched `model_name`, incomplete model config, ONNX session errors)—not silent fallbacks.
+
+**Batch sizes, ORT options, and other runtime settings:**
+
+```js
+await PaddleOCR.create({
+  lang: "ch",
+  ocrVersion: "PP-OCRv5",
+  textDetectionBatchSize: 2,
+  textRecognitionBatchSize: 8,
+  ortOptions: {
+    backend: "wasm",
+    wasmPaths: "/assets/"
+  }
+});
+```
+
+### 2. Pipeline config
+
+```js
+import { PaddleOCR } from "@paddleocr/paddleocr-js";
+
+const pipelineConfig = `
+pipeline_name: OCR
+SubModules:
+  TextDetection:
+    model_name: PP-OCRv5_mobile_det
+    batch_size: 2
+  TextRecognition:
+    model_name: PP-OCRv5_mobile_rec
+    batch_size: 6
+`;
+
+const ocr = await PaddleOCR.create({ pipelineConfig });
+```
+
+`pipelineConfig` can be either YAML text or a parsed object.
+
+If direct parameters and `pipelineConfig` are both provided, direct parameters take precedence.
+
+## Prediction
+
+### Params
+
+`ocr.predict(image | images[], params?)` accepts both camelCase names and PaddleOCR-style snake_case names:
+
+- `textDetLimitSideLen` or `text_det_limit_side_len`
+- `textDetLimitType` or `text_det_limit_type`
+- `textDetMaxSideLimit` or `text_det_max_side_limit`
+- `textDetThresh` or `text_det_thresh`
+- `textDetBoxThresh` or `text_det_box_thresh`
+- `textDetUnclipRatio` or `text_det_unclip_ratio`
+- `textRecScoreThresh` or `text_rec_score_thresh`
+
+Supported `image` inputs include `Blob`, `ImageBitmap`, `ImageData`, `HTMLCanvasElement`, `HTMLImageElement`, and `cv.Mat`. Pass an array of these to run detection and recognition on multiple images in one call.
+
+In worker mode (see next section), `cv.Mat` is not transferable and is therefore not supported as a worker input.
+
+### Return value
+
+Resolves to `Promise<OcrResult[]>`. Each `OcrResult` contains:
+
+- `image`: `{ width, height }` for that source
+- `items`: recognized lines (`poly`, `text`, `score`)
+- `metrics`: `detMs`, `recMs`, `totalMs`, `detectedBoxes`, `recognizedCount` — box and line counts are per image; `detMs`, `recMs`, and `totalMs` cover the **entire** `predict()` call (so they are identical on every element when you pass multiple images)
+- `runtime`: requested backend and provider metadata
+
+## Worker Mode
+
+You can run the OCR pipeline inside a dedicated Worker while keeping the same high-level API:
+
+```js
+import { PaddleOCR } from "@paddleocr/paddleocr-js";
+
+const ocr = await PaddleOCR.create({
+  lang: "ch",
+  ocrVersion: "PP-OCRv5",
+  worker: true,
+  ortOptions: {
+    backend: "wasm",
+    wasmPaths: "https://cdn.jsdelivr.net/npm/onnxruntime-web/dist/",
+    numThreads: 2,
+    simd: true
+  }
+});
+```
+
+Worker behavior:
+
+- Worker mode uses the package worker path, not ONNX Runtime Web `env.wasm.proxy`.
+- When `worker: true` is enabled, the package forces ORT wasm proxy off internally.
+- Browser inputs are normalized on the main thread and transferred into the worker before inference runs.
+- `cv.Mat` is only supported in the direct main-thread pipeline path.
+
+## Visualization
+
+The optional `@paddleocr/paddleocr-js/viz` subpath provides visualization utilities for rendering OCR results as images.
+
+```js
+import { OcrVisualizer } from "@paddleocr/paddleocr-js/viz";
+
+const viz = new OcrVisualizer({
+  font: { family: "Noto Sans SC", source: "/fonts/NotoSansSC-Regular.ttf" }
+});
+
+const blob = await viz.toBlob(imageBitmap, result);
+
+// Trigger browser download
+const url = URL.createObjectURL(blob);
+const a = document.createElement("a");
+a.href = url;
+a.download = "ocr_result.png";
+a.click();
+URL.revokeObjectURL(url);
+
+viz.dispose();
+```
+
+A one-shot convenience function is also available:
+
+```js
+import { renderOcrToBlob } from "@paddleocr/paddleocr-js/viz";
+
+const blob = await renderOcrToBlob(imageBitmap, result, {
+  font: { family: "Noto Sans SC", source: "/fonts/NotoSansSC-Regular.ttf" }
+});
+```
+
+The viz module renders a side-by-side composite image: the original image with detection box overlays on the left, and recognized text on the right. Custom fonts can be loaded for CJK text rendering. Visualization requires a **single** `OcrResult` (for one image, take the first element of the array returned by `predict`, e.g. `const [result] = await ocr.predict(image)`).
+
+`deterministicColor(index)` is also exported from the viz subpath. It maps a numeric index to a stable RGB color and is used internally as the default color function for detection boxes and text labels. You can call it directly when building custom visualizations that need colors consistent with the built-in renderer.
+
+## API
+
+- `PaddleOCR.create(options)`
+- `ocr.initialize()`
+- `ocr.getInitializationSummary()`
+- `ocr.predict(image | images[], params?)` → `Promise<OcrResult[]>`
+- `ocr.dispose()`
+- `parseOcrPipelineConfigText(text)`
+- `normalizeOcrPipelineConfig(config)`
+- `OcrVisualizer` (from `@paddleocr/paddleocr-js/viz`)
+- `renderOcrToBlob` (from `@paddleocr/paddleocr-js/viz`)
+- `deterministicColor` (from `@paddleocr/paddleocr-js/viz`)
+
+## Package Layout
+
+```
+src/
+├── runtime/       — inference runtime setup
+├── resources/     — model & asset management
+├── models/        — model wiring
+├── platform/      — browser/worker input adaptation
+├── worker/        — worker transport layer
+├── pipelines/     — pipeline implementations
+├── viz/           — visualization (optional)
+├── types/         — external type declarations
+└── utils/         — shared utilities
+```
+
+## Runtime Responsibilities
+
+The SDK manages OpenCV.js and ONNX Runtime internally. The host application is still responsible for runtime environment concerns, including:
+
+- COOP/COEP headers when enabling threaded WASM or WebGPU
+- ONNX Runtime Web environment options such as wasm asset hosting paths, thread counts, and SIMD flags
+- a bundler/runtime setup that can emit and load module workers when `worker: true` is used
diff --git a/paddleocr-js/packages/core/README_cn.md b/paddleocr-js/packages/core/README_cn.md
new file mode 100644
index 0000000..25518eb
--- /dev/null
+++ b/paddleocr-js/packages/core/README_cn.md
@@ -0,0 +1,254 @@
+# PaddleOCR.js SDK
+
+[English](README.md) | 简体中文
+
+`@paddleocr/paddleocr-js` 是在前端运行 PaddleOCR 产线的浏览器 SDK 包。
+
+## 安装
+
+```bash
+npm install @paddleocr/paddleocr-js
+```
+
+## 快速开始
+
+```js
+import { PaddleOCR } from "@paddleocr/paddleocr-js";
+
+const ocr = await PaddleOCR.create({
+  lang: "ch",
+  ocrVersion: "PP-OCRv5",
+  ortOptions: {
+    backend: "auto"
+  }
+});
+
+const [result] = await ocr.predict(fileOrBlob);
+console.log(result.items);
+```
+
+`predict` 返回 **`OcrResult` 组成的数组**（每张输入图像对应一项）。传入单个 `Blob` / `File` 时也会得到长度为 1 的数组，请使用解构或 `results[0]` 取值。
+
+## 构造方式
+
+主要有两种构造方式：
+
+### 1. 直接参数
+
+可通过直接参数指定模型，也可配置推理 batch size、ORT 选项等运行参数。
+
+**模型选择 — `lang` + `ocrVersion`：**
+
+```js
+await PaddleOCR.create({
+  lang: "ch",
+  ocrVersion: "PP-OCRv5"
+});
+```
+
+`ocrVersion: "PP-OCRv6"` 会将受支持的 `lang` 映射到内置的 **PP-OCRv6_small** 检测/识别模型对。若需 **PP-OCRv6_tiny**，请显式指定模型名：
+
+```js
+await PaddleOCR.create({
+  textDetectionModelName: "PP-OCRv6_tiny_det",
+  textRecognitionModelName: "PP-OCRv6_tiny_rec"
+});
+```
+
+**模型选择 — 显式模型名：**
+
+```js
+await PaddleOCR.create({
+  textDetectionModelName: "PP-OCRv5_mobile_det",
+  textRecognitionModelName: "PP-OCRv5_mobile_rec"
+});
+```
+
+**自定义模型** — 为检测 / 识别分别指定模型名与资源地址：
+
+```js
+await PaddleOCR.create({
+  textDetectionModelName: "my_det_model",
+  textDetectionModelAsset: {
+    url: "https://example.com/models/my_det_model.tar"
+  },
+  textRecognitionModelName: "my_rec_model",
+  textRecognitionModelAsset: {
+    url: "https://example.com/models/my_rec_model.tar"
+  }
+});
+```
+
+**自定义模型包格式与校验行为：**
+
+- 资源需为 **未压缩的标准 tar**（`.tar`）。SDK 按字节解析 tar，**不对 gzip 压缩包解压**；若 URL 指向 `.tar.gz` 等，通常会解析失败并报错。
+- tar 内必须包含 **`inference.onnx`** 与 **`inference.yml`**（可在子目录中，按文件名匹配）。
+- `inference.yml` 必须能解析出 **`model_name`**，且须与 `textDetectionModelName` / `textRecognitionModelName` **一致**；在 `initialize` 加载模型后会校验。
+
+不满足时通常在初始化阶段以 **`Error`** 失败（下载失败、tar 中缺少条目、空资源、`model_name` 缺失或不匹配、模型配置不完整、ONNX 会话创建失败等），不会静默忽略。
+
+**batch size、ORT 选项等运行参数：**
+
+```js
+await PaddleOCR.create({
+  lang: "ch",
+  ocrVersion: "PP-OCRv5",
+  textDetectionBatchSize: 2,
+  textRecognitionBatchSize: 8,
+  ortOptions: {
+    backend: "wasm",
+    wasmPaths: "/assets/"
+  }
+});
+```
+
+### 2. 产线配置
+
+```js
+import { PaddleOCR } from "@paddleocr/paddleocr-js";
+
+const pipelineConfig = `
+pipeline_name: OCR
+SubModules:
+  TextDetection:
+    model_name: PP-OCRv5_mobile_det
+    batch_size: 2
+  TextRecognition:
+    model_name: PP-OCRv5_mobile_rec
+    batch_size: 6
+`;
+
+const ocr = await PaddleOCR.create({ pipelineConfig });
+```
+
+`pipelineConfig` 可以是 YAML 文本，也可以是解析后的对象。
+
+如果同时提供直接参数和 `pipelineConfig`，则以直接参数为准。
+
+## 预测
+
+### 参数
+
+`ocr.predict(image | images[], params?)` 同时接受 camelCase 命名和 PaddleOCR 风格的 snake_case 命名：
+
+- `textDetLimitSideLen` 或 `text_det_limit_side_len`
+- `textDetLimitType` 或 `text_det_limit_type`
+- `textDetMaxSideLimit` 或 `text_det_max_side_limit`
+- `textDetThresh` 或 `text_det_thresh`
+- `textDetBoxThresh` 或 `text_det_box_thresh`
+- `textDetUnclipRatio` 或 `text_det_unclip_ratio`
+- `textRecScoreThresh` 或 `text_rec_score_thresh`
+
+支持的 `image` 输入包括 `Blob`、`ImageBitmap`、`ImageData`、`HTMLCanvasElement`、`HTMLImageElement` 和 `cv.Mat`。传入上述类型的数组可在一次调用中对多张图像做检测与识别。
+
+在 worker 模式下（见下一节），`cv.Mat` 无法传输，因此不能作为 worker 输入。
+
+### 返回值
+
+返回 `Promise<OcrResult[]>`。每个 `OcrResult` 包含：
+
+- `image`：该图源的尺寸 `{ width, height }`
+- `items`：识别行（`poly`、`text`、`score`）
+- `metrics`：`detMs`、`recMs`、`totalMs`、`detectedBoxes`、`recognizedCount` — 检测框与文本行数为**每张图**统计；`detMs`、`recMs`、`totalMs` 表示**整次** `predict()` 调用的耗时（传入多图时，数组中每一项上的这三个值相同）
+- `runtime`：请求的后端与各阶段 Provider 等元数据
+
+## Worker 模式
+
+你可以在专用 Worker 中运行 OCR 产线，同时保持相同的高层 API：
+
+```js
+import { PaddleOCR } from "@paddleocr/paddleocr-js";
+
+const ocr = await PaddleOCR.create({
+  lang: "ch",
+  ocrVersion: "PP-OCRv5",
+  worker: true,
+  ortOptions: {
+    backend: "wasm",
+    wasmPaths: "https://cdn.jsdelivr.net/npm/onnxruntime-web/dist/",
+    numThreads: 2,
+    simd: true
+  }
+});
+```
+
+Worker 模式的行为：
+
+- Worker 模式使用包内的 worker 路径，而不是 ONNX Runtime Web 的 `env.wasm.proxy`
+- 启用 `worker: true` 时，包内部会强制关闭 ORT 的 wasm proxy
+- 浏览器输入会先在主线程标准化，再传入 worker 执行推理
+- `cv.Mat` 仅支持直接在主线程产线路径中使用
+
+## 可视化
+
+可选的 `@paddleocr/paddleocr-js/viz` 子路径提供了将 OCR 结果渲染为图像的可视化工具。
+
+```js
+import { OcrVisualizer } from "@paddleocr/paddleocr-js/viz";
+
+const viz = new OcrVisualizer({
+  font: { family: "Noto Sans SC", source: "/fonts/NotoSansSC-Regular.ttf" }
+});
+
+const blob = await viz.toBlob(imageBitmap, result);
+
+// 触发浏览器下载
+const url = URL.createObjectURL(blob);
+const a = document.createElement("a");
+a.href = url;
+a.download = "ocr_result.png";
+a.click();
+URL.revokeObjectURL(url);
+
+viz.dispose();
+```
+
+也提供了一次性便捷函数：
+
+```js
+import { renderOcrToBlob } from "@paddleocr/paddleocr-js/viz";
+
+const blob = await renderOcrToBlob(imageBitmap, result, {
+  font: { family: "Noto Sans SC", source: "/fonts/NotoSansSC-Regular.ttf" }
+});
+```
+
+viz 模块会渲染一张左右对比的合成图像：左侧为带有检测框叠加的原始图像，右侧为识别出的文字。支持加载自定义字体以正确渲染中日韩等文字。可视化需传入**单个** `OcrResult`（单张图时取 `predict` 返回数组的首项，例如 `const [result] = await ocr.predict(image)`）。
+
+`deterministicColor(index)` 同样从 viz 子路径导出。它根据数字索引生成稳定的 RGB 颜色，内部用作检测框和文字标签的默认配色函数。当你构建自定义可视化并需要与内置渲染器保持一致的配色时，可以直接调用该函数。
+
+## API
+
+- `PaddleOCR.create(options)`
+- `ocr.initialize()`
+- `ocr.getInitializationSummary()`
+- `ocr.predict(image | images[], params?)` → `Promise<OcrResult[]>`
+- `ocr.dispose()`
+- `parseOcrPipelineConfigText(text)`
+- `normalizeOcrPipelineConfig(config)`
+- `OcrVisualizer`（来自 `@paddleocr/paddleocr-js/viz`）
+- `renderOcrToBlob`（来自 `@paddleocr/paddleocr-js/viz`）
+- `deterministicColor`（来自 `@paddleocr/paddleocr-js/viz`）
+
+## 包结构
+
+```
+src/
+├── runtime/       — 推理运行时初始化
+├── resources/     — 模型与资源管理
+├── models/        — 模型接线
+├── platform/      — 浏览器/worker 输入适配
+├── worker/        — worker 传输层
+├── pipelines/     — 产线实现
+├── viz/           — 可视化（可选）
+├── types/         — 外部库类型声明
+└── utils/         — 共享工具
+```
+
+## 运行时职责边界
+
+SDK 内部负责管理 OpenCV.js 和 ONNX Runtime。宿主应用仍需负责运行时环境相关事项，包括：
+
+- 启用多线程 WASM 或 WebGPU 时所需的 COOP/COEP 响应头
+- ONNX Runtime Web 的环境选项，例如 wasm 资源托管路径、线程数和 SIMD 开关
+- 当使用 `worker: true` 时，能够产出并加载 module worker 的构建工具或运行时配置
diff --git a/paddleocr-js/packages/core/package.json b/paddleocr-js/packages/core/package.json
new file mode 100644
index 0000000..cbd416d
--- /dev/null
+++ b/paddleocr-js/packages/core/package.json
@@ -0,0 +1,73 @@
+{
+  "name": "@paddleocr/paddleocr-js",
+  "version": "0.4.2",
+  "description": "Browser-based OCR SDK powered by PaddleOCR, ONNX Runtime Web and OpenCV.js",
+  "license": "Apache-2.0",
+  "author": "PaddlePaddle Authors",
+  "type": "module",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/PaddlePaddle/PaddleOCR.git",
+    "directory": "paddleocr-js/packages/core"
+  },
+  "homepage": "https://github.com/PaddlePaddle/PaddleOCR/tree/main/paddleocr-js",
+  "bugs": {
+    "url": "https://github.com/PaddlePaddle/PaddleOCR/issues"
+  },
+  "publishConfig": {
+    "access": "public"
+  },
+  "exports": {
+    ".": {
+      "import": {
+        "types": "./dist/index.d.ts",
+        "default": "./dist/index.mjs"
+      }
+    },
+    "./viz": {
+      "import": {
+        "types": "./dist/viz/index.d.ts",
+        "default": "./dist/viz.mjs"
+      }
+    }
+  },
+  "files": [
+    "dist",
+    "README.md",
+    "README_cn.md"
+  ],
+  "sideEffects": false,
+  "keywords": [
+    "ocr",
+    "paddleocr",
+    "paddle",
+    "text-recognition",
+    "text-detection",
+    "onnx",
+    "onnxruntime",
+    "opencv",
+    "browser-ocr",
+    "wasm",
+    "webgpu",
+    "deep-learning",
+    "machine-learning",
+    "computer-vision"
+  ],
+  "scripts": {
+    "build": "vite build",
+    "typecheck": "tsc --noEmit",
+    "prepublishOnly": "npm run build"
+  },
+  "dependencies": {
+    "@techstark/opencv-js": "^4.10.0-release.1",
+    "clipper-lib": "^6.4.2",
+    "js-yaml": "^4.1.0",
+    "onnxruntime-web": "^1.22.0"
+  },
+  "devDependencies": {
+    "@types/js-yaml": "^4.0.9",
+    "typescript": "^5.9.3",
+    "vite": "^6.4.1",
+    "vite-plugin-dts": "^1.0.5"
+  }
+}
diff --git a/paddleocr-js/packages/core/src/index.ts b/paddleocr-js/packages/core/src/index.ts
new file mode 100644
index 0000000..bfa8479
--- /dev/null
+++ b/paddleocr-js/packages/core/src/index.ts
@@ -0,0 +1,59 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+export {
+  PaddleOCR,
+  normalizeOcrPipelineConfig,
+  parseOcrPipelineConfigText
+} from "./pipelines/ocr/index";
+
+export type { Point2D, NormalizeConfig, DetBox } from "./models/common";
+
+export type {
+  DetModelConfig,
+  DetPostprocessConfig,
+  DetModel,
+  DetResult,
+  DetRuntimeOverrides,
+  LimitType
+} from "./models/det";
+
+export type { RecModelConfig, RecModel, RecResult, RecRuntimeOverrides } from "./models/rec";
+
+export type {
+  OcrRuntimeParamsInput,
+  OcrModelConfig,
+  ResolvedOcrParams
+} from "./pipelines/ocr/runtime-params";
+
+export type {
+  OcrResult,
+  OcrResultItem,
+  OcrResultMetrics,
+  OcrResultRuntime,
+  InitializationSummary,
+  OcrPipelineRunnerOptions
+} from "./pipelines/ocr/core";
+
+export type {
+  NormalizedPipelineConfig,
+  PipelineModelSelection,
+  PipelineRuntimeDefaults
+} from "./pipelines/ocr/config";
+
+export type {
+  ResolvedBackend,
+  ResolvedOcrOptions,
+  NormalizedOrtOptions,
+  WorkerResolvedOptions
+} from "./pipelines/ocr/shared";
+
+export type { ModelAsset, ModelAssetsMap } from "./resources/model-asset";
+
+export type { WebGpuState, OrtOptions } from "./runtime/ort";
+
+export type { ImageSource, SourceMatResult } from "./platform/browser";
+
+export type { PaddleOCRCreateOptions } from "./pipelines/ocr/index";
diff --git a/paddleocr-js/packages/core/src/models/common.ts b/paddleocr-js/packages/core/src/models/common.ts
new file mode 100644
index 0000000..411bce6
--- /dev/null
+++ b/paddleocr-js/packages/core/src/models/common.ts
@@ -0,0 +1,260 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import yaml from "js-yaml";
+import ClipperLib from "clipper-lib";
+import type { OpenCv, Mat } from "@techstark/opencv-js";
+
+import { clamp, distance2 } from "../utils/common";
+
+export type Point2D = [number, number];
+
+export interface NormalizeConfig {
+  mean: number[];
+  std: number[];
+  scale: number;
+}
+
+export interface MiniBox {
+  box: Point2D[];
+  side: number;
+}
+
+export interface DetBox {
+  poly: Point2D[];
+  score: number;
+}
+
+type YamlValue = string | number | boolean | null | YamlValue[] | { [key: string]: YamlValue };
+type YamlObject = Record<string, YamlValue>;
+
+function isPlainObject(value: unknown): value is YamlObject {
+  return Boolean(value) && typeof value === "object" && !Array.isArray(value);
+}
+
+export function parseInferenceConfigText(text: string): YamlObject {
+  const parsed = yaml.load(text);
+  return isPlainObject(parsed) ? parsed : {};
+}
+
+export function parseScaleValue(rawScale: unknown, fallback: number): number {
+  if (typeof rawScale === "number") return rawScale;
+  if (typeof rawScale !== "string") return fallback;
+  const normalized = rawScale.replace(/\s/g, "");
+  const direct = Number(normalized);
+  if (!Number.isNaN(direct)) return direct;
+  const divParts = normalized.split("/");
+  if (divParts.length === 2) {
+    const numerator = Number(divParts[0].replace(/\.+$/, ""));
+    const denominator = Number(divParts[1].replace(/\.+$/, ""));
+    if (!Number.isNaN(numerator) && !Number.isNaN(denominator) && denominator !== 0) {
+      return numerator / denominator;
+    }
+  }
+  return fallback;
+}
+
+export function getTransformOp(
+  transformOps: Array<Record<string, unknown>> | undefined,
+  opName: string
+): Record<string, unknown> | null {
+  for (const op of transformOps || []) {
+    if (Object.prototype.hasOwnProperty.call(op, opName)) {
+      return op[opName] as YamlObject;
+    }
+  }
+  return null;
+}
+
+function findModelNameInYamlNode(value: YamlValue): string | null {
+  if (Array.isArray(value)) {
+    for (const item of value) {
+      const match = findModelNameInYamlNode(item);
+      if (match) return match;
+    }
+    return null;
+  }
+
+  if (!isPlainObject(value)) {
+    return null;
+  }
+
+  for (const [key, childValue] of Object.entries(value)) {
+    if (key === "model_name" && typeof childValue === "string" && childValue.trim()) {
+      return childValue;
+    }
+    const match = findModelNameInYamlNode(childValue);
+    if (match) return match;
+  }
+
+  return null;
+}
+
+export function extractInferenceModelName(configText: string): string | null {
+  const parsed = parseInferenceConfigText(configText);
+  const preferredCandidates = [
+    (parsed.Global as YamlObject | undefined)?.model_name,
+    parsed.model_name
+  ];
+  for (const candidate of preferredCandidates) {
+    if (typeof candidate === "string" && candidate.trim()) {
+      return candidate;
+    }
+  }
+
+  return findModelNameInYamlNode(parsed);
+}
+
+export function toBgrFloatCHWFromBgr(
+  bgr: Uint8Array,
+  width: number,
+  height: number,
+  normalizeConfig: NormalizeConfig
+): Float32Array {
+  const data = new Float32Array(3 * width * height);
+  const hw = width * height;
+  const mean = normalizeConfig.mean;
+  const std = normalizeConfig.std;
+  const scale = normalizeConfig.scale;
+  for (let y = 0; y < height; y += 1) {
+    for (let x = 0; x < width; x += 1) {
+      const idx = y * width + x;
+      const p = idx * 3;
+      const b = bgr[p];
+      const g = bgr[p + 1];
+      const r = bgr[p + 2];
+      data[idx] = (b * scale - mean[0]) / std[0];
+      data[idx + hw] = (g * scale - mean[1]) / std[1];
+      data[idx + 2 * hw] = (r * scale - mean[2]) / std[2];
+    }
+  }
+  return data;
+}
+
+function orderQuad(pts: Point2D[]): Point2D[] {
+  const points = pts.slice().sort((a, b) => a[0] - b[0]);
+  let indexA: number;
+  let indexB: number;
+  let indexC: number;
+  let indexD: number;
+  if (points[1][1] > points[0][1]) {
+    indexA = 0;
+    indexD = 1;
+  } else {
+    indexA = 1;
+    indexD = 0;
+  }
+  if (points[3][1] > points[2][1]) {
+    indexB = 2;
+    indexC = 3;
+  } else {
+    indexB = 3;
+    indexC = 2;
+  }
+  return [points[indexA], points[indexB], points[indexC], points[indexD]];
+}
+
+function polygonArea(poly: Point2D[]): number {
+  let area = 0;
+  for (let i = 0; i < poly.length; i += 1) {
+    const j = (i + 1) % poly.length;
+    area += poly[i][0] * poly[j][1] - poly[j][0] * poly[i][1];
+  }
+  return Math.abs(area) * 0.5;
+}
+
+function polygonPerimeter(poly: Point2D[]): number {
+  let peri = 0;
+  for (let i = 0; i < poly.length; i += 1) {
+    const j = (i + 1) % poly.length;
+    peri += distance2(poly[i], poly[j]);
+  }
+  return peri;
+}
+
+interface ClipperPoint {
+  X: number;
+  Y: number;
+}
+
+function chooseMaxAreaPath(paths: ClipperPoint[][]): ClipperPoint[] | null {
+  let best: ClipperPoint[] | null = null;
+  let bestArea = 0;
+  for (const path of paths) {
+    if (path.length < 4) continue;
+    const poly: Point2D[] = path.map((pt) => [pt.X, pt.Y]);
+    const area = polygonArea(poly);
+    if (area > bestArea) {
+      bestArea = area;
+      best = path;
+    }
+  }
+  return best;
+}
+
+export function unclip(poly: Point2D[], unclipRatio: number): Point2D[] | null {
+  const area = polygonArea(poly);
+  const perimeter = polygonPerimeter(poly);
+  if (perimeter <= 0) return null;
+  const distance = (area * unclipRatio) / perimeter;
+  const path: ClipperPoint[] = poly.map((p) => ({ X: Math.trunc(p[0]), Y: Math.trunc(p[1]) }));
+  const offset = new ClipperLib.ClipperOffset();
+  offset.AddPath(path, ClipperLib.JoinType.jtRound, ClipperLib.EndType.etClosedPolygon);
+  const expanded = new ClipperLib.Paths();
+  offset.Execute(expanded, distance);
+  const best = chooseMaxAreaPath(expanded);
+  if (!best) return null;
+  return best.map((pt) => [pt.X, pt.Y]);
+}
+
+export function getMiniBoxFromPoints(cv: OpenCv, points: Point2D[]): MiniBox {
+  const flat: number[] = [];
+  for (const p of points) flat.push(p[0], p[1]);
+  const contour = cv.matFromArray(points.length, 1, cv.CV_32FC2, flat);
+  const rect = cv.minAreaRect(contour);
+  const vertices = cv.RotatedRect.points(rect);
+  const box: Point2D[] = [];
+  for (let i = 0; i < 4; i += 1) box.push([vertices[i].x, vertices[i].y]);
+  contour.delete();
+  const ordered = orderQuad(box);
+  const side = Math.min(distance2(ordered[0], ordered[1]), distance2(ordered[1], ordered[2]));
+  return { box: ordered, side };
+}
+
+export function boxScoreFast(cv: OpenCv, predMat: Mat, box: Point2D[]): number {
+  const h = predMat.rows;
+  const w = predMat.cols;
+  let minX = w - 1;
+  let maxX = 0;
+  let minY = h - 1;
+  let maxY = 0;
+  for (const p of box) {
+    minX = Math.min(minX, p[0]);
+    maxX = Math.max(maxX, p[0]);
+    minY = Math.min(minY, p[1]);
+    maxY = Math.max(maxY, p[1]);
+  }
+  minX = clamp(Math.floor(minX), 0, w - 1);
+  maxX = clamp(Math.ceil(maxX), 0, w - 1);
+  minY = clamp(Math.floor(minY), 0, h - 1);
+  maxY = clamp(Math.ceil(maxY), 0, h - 1);
+  const rw = Math.max(1, maxX - minX + 1);
+  const rh = Math.max(1, maxY - minY + 1);
+  const roi = predMat.roi(new cv.Rect(minX, minY, rw, rh));
+  const mask = cv.Mat.zeros(rh, rw, cv.CV_8UC1);
+  const shifted = box.map((p) => [Math.trunc(p[0] - minX), Math.trunc(p[1] - minY)]);
+  const flat: number[] = [];
+  for (const p of shifted) flat.push(p[0], p[1]);
+  const pts = cv.matFromArray(shifted.length, 1, cv.CV_32SC2, flat);
+  const ptsVec = new cv.MatVector();
+  ptsVec.push_back(pts);
+  cv.fillPoly(mask, ptsVec, new cv.Scalar(1));
+  const mean = cv.mean(roi, mask)[0];
+  roi.delete();
+  mask.delete();
+  pts.delete();
+  ptsVec.delete();
+  return mean;
+}
diff --git a/paddleocr-js/packages/core/src/models/det.ts b/paddleocr-js/packages/core/src/models/det.ts
new file mode 100644
index 0000000..7acf6d9
--- /dev/null
+++ b/paddleocr-js/packages/core/src/models/det.ts
@@ -0,0 +1,551 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { OpenCv, Mat } from "@techstark/opencv-js";
+import type { InferenceSession, Tensor } from "onnxruntime-web";
+
+import { assertModelResources } from "../resources/model-asset";
+import { createSession, getProviderCandidates, releaseSessions } from "../runtime/ort";
+import type { OrtModule, WebGpuState, SessionState } from "../runtime/ort";
+import { chunkArray, clamp, resolveRuntimeBatchSize, withTimeout } from "../utils/common";
+import { runInference } from "./infer";
+import {
+  boxScoreFast,
+  getMiniBoxFromPoints,
+  getTransformOp,
+  parseInferenceConfigText,
+  parseScaleValue,
+  toBgrFloatCHWFromBgr,
+  unclip
+} from "./common";
+import type { Point2D, NormalizeConfig, DetBox } from "./common";
+
+export type LimitType = "min" | "max";
+
+export interface DetRuntimeOverrides {
+  batchSize?: number;
+  thresh?: number;
+  boxThresh?: number;
+  unclipRatio?: number;
+  limitSideLen?: number;
+  limitType?: LimitType;
+  maxSideLimit?: number;
+}
+
+export interface DetPostprocessConfig {
+  thresh: number;
+  boxThresh: number;
+  maxCandidates: number;
+  unclipRatio: number;
+}
+
+export interface DetModelConfig {
+  resizeLong: number;
+  limitType: LimitType;
+  maxSideLimit: number;
+  normalize: NormalizeConfig;
+  postprocess: DetPostprocessConfig;
+}
+
+export interface DetModel {
+  readonly kind: "det";
+  readonly config: DetModelConfig;
+  readonly provider: string;
+  predict(cv: OpenCv, mats: Mat[], overrides?: DetRuntimeOverrides): Promise<DetResult[]>;
+  dispose(): Promise<void>;
+}
+
+interface DetPreprocessResult {
+  tensor: Tensor;
+  srcW: number;
+  srcH: number;
+  dstW: number;
+  dstH: number;
+}
+
+export interface DetResult {
+  boxes: DetBox[];
+  srcW: number;
+  srcH: number;
+}
+
+interface InternalDetParams {
+  limitSideLen: number;
+  limitType: LimitType;
+  maxSideLimit: number;
+  thresh: number;
+  boxThresh: number;
+  unclipRatio: number;
+}
+
+interface InternalDetBatchItem {
+  prep: DetPreprocessResult;
+  boxes: DetBox[];
+}
+
+const DET_BOX_MIN_SIZE = 3;
+
+export const DEFAULT_DET_MODEL_PARSE_FALLBACKS: Readonly<DetModelConfig> = Object.freeze({
+  resizeLong: 960,
+  limitType: "max",
+  maxSideLimit: 4000,
+  normalize: {
+    mean: [0.485, 0.456, 0.406],
+    std: [0.229, 0.224, 0.225],
+    scale: 1 / 255
+  },
+  postprocess: {
+    thresh: 0.3,
+    boxThresh: 0.6,
+    maxCandidates: 1000,
+    unclipRatio: 2.0
+  }
+});
+
+export const DEFAULT_DET_MODEL_CONFIG: Readonly<DetModelConfig> = Object.freeze({
+  ...DEFAULT_DET_MODEL_PARSE_FALLBACKS
+});
+
+function parseDetLimitType(raw: unknown): LimitType {
+  const v = typeof raw === "string" ? raw.trim().toLowerCase() : "";
+  if (v === "min" || v === "max") {
+    return v;
+  }
+  return DEFAULT_DET_MODEL_PARSE_FALLBACKS.limitType;
+}
+
+export function parseDetModelConfigText(text: string): DetModelConfig {
+  const parsed = parseInferenceConfigText(text);
+  const preProcess = parsed.PreProcess as Record<string, unknown> | undefined;
+  const transformOps = preProcess?.transform_ops as Array<Record<string, unknown>> | undefined;
+  const resize = getTransformOp(transformOps, "DetResizeForTest");
+  const normalize = getTransformOp(transformOps, "NormalizeImage");
+  const postprocess = (parsed.PostProcess || {}) as Record<string, unknown>;
+
+  const maxSideRaw = resize?.max_side_limit;
+  const maxSideLimit = Number(maxSideRaw);
+  const maxSide =
+    Number.isFinite(maxSideLimit) && maxSideLimit > 0
+      ? maxSideLimit
+      : DEFAULT_DET_MODEL_PARSE_FALLBACKS.maxSideLimit;
+
+  return {
+    resizeLong: Number(resize?.resize_long ?? DEFAULT_DET_MODEL_PARSE_FALLBACKS.resizeLong),
+    limitType: parseDetLimitType(resize?.limit_type),
+    maxSideLimit: maxSide,
+    normalize: {
+      mean:
+        (normalize?.mean as number[] | undefined) ??
+        DEFAULT_DET_MODEL_PARSE_FALLBACKS.normalize.mean,
+      std:
+        (normalize?.std as number[] | undefined) ?? DEFAULT_DET_MODEL_PARSE_FALLBACKS.normalize.std,
+      scale: parseScaleValue(normalize?.scale, DEFAULT_DET_MODEL_PARSE_FALLBACKS.normalize.scale)
+    },
+    postprocess: {
+      thresh: Number(postprocess.thresh ?? DEFAULT_DET_MODEL_PARSE_FALLBACKS.postprocess.thresh),
+      boxThresh: Number(
+        postprocess.box_thresh ?? DEFAULT_DET_MODEL_PARSE_FALLBACKS.postprocess.boxThresh
+      ),
+      maxCandidates: Number(
+        postprocess.max_candidates ?? DEFAULT_DET_MODEL_PARSE_FALLBACKS.postprocess.maxCandidates
+      ),
+      unclipRatio: Number(
+        postprocess.unclip_ratio ?? DEFAULT_DET_MODEL_PARSE_FALLBACKS.postprocess.unclipRatio
+      )
+    }
+  };
+}
+
+interface CreateDetModelArgs {
+  ort: OrtModule;
+  modelBytes: Uint8Array;
+  configText: string;
+  backend: string;
+  webgpuState: WebGpuState;
+  batchSize?: number;
+}
+
+function resolveDetParams(
+  defaults: InternalDetParams,
+  overrides?: DetRuntimeOverrides
+): InternalDetParams {
+  return {
+    limitSideLen: overrides?.limitSideLen ?? defaults.limitSideLen,
+    limitType: overrides?.limitType ?? defaults.limitType,
+    maxSideLimit: overrides?.maxSideLimit ?? defaults.maxSideLimit,
+    thresh: overrides?.thresh ?? defaults.thresh,
+    boxThresh: overrides?.boxThresh ?? defaults.boxThresh,
+    unclipRatio: overrides?.unclipRatio ?? defaults.unclipRatio
+  };
+}
+
+export async function createDetModel({
+  ort,
+  modelBytes,
+  configText,
+  backend,
+  webgpuState,
+  batchSize: batchSizeArg
+}: CreateDetModelArgs): Promise<DetModel> {
+  assertModelResources("Detection", {
+    model: modelBytes,
+    config: configText
+  });
+  const config = parseDetModelConfigText(configText);
+  const defaultBatchSize = Math.max(1, batchSizeArg ?? 1);
+  const defaultParams: InternalDetParams = {
+    limitSideLen: config.resizeLong,
+    limitType: config.limitType,
+    maxSideLimit: config.maxSideLimit,
+    thresh: config.postprocess.thresh,
+    boxThresh: config.postprocess.boxThresh,
+    unclipRatio: config.postprocess.unclipRatio
+  };
+  let sessionState: SessionState | null = await createDetModelSession(
+    ort,
+    modelBytes,
+    backend,
+    webgpuState
+  );
+
+  return {
+    kind: "det",
+    config,
+    get provider() {
+      return sessionState?.provider || "";
+    },
+    async predict(cv, mats, overrides) {
+      if (!sessionState?.session) {
+        throw new Error("Detection model session is not initialized.");
+      }
+      const params = resolveDetParams(defaultParams, overrides);
+      const batchSize = resolveRuntimeBatchSize(overrides?.batchSize, defaultBatchSize);
+      const results: DetResult[] = [];
+      const runCtx: DetRunContext = {
+        cv,
+        ort,
+        config,
+        session: sessionState.session
+      };
+      for (const chunk of chunkArray(mats, batchSize)) {
+        const preps = preprocess({ cv, ort, config }, chunk, params);
+        const inputTensor = packDetBatchTensor(ort, preps);
+        const fullOutput = await runInference(sessionState.session, inputTensor);
+        const internals = postprocess(runCtx, fullOutput, preps, params);
+        for (const internal of internals) {
+          results.push({
+            boxes: internal.boxes,
+            srcW: internal.prep.srcW,
+            srcH: internal.prep.srcH
+          });
+        }
+      }
+      return results;
+    },
+    async dispose() {
+      await releaseSessions(sessionState?.session);
+      sessionState = null;
+    }
+  };
+}
+
+export async function createDetModelSession(
+  ort: OrtModule,
+  modelBytes: Uint8Array,
+  backend: string,
+  webgpuState: WebGpuState
+): Promise<SessionState> {
+  const providerCandidates = getProviderCandidates(backend, webgpuState);
+  return withTimeout(createSession(ort, modelBytes, providerCandidates), 60000, "Detection model");
+}
+
+interface DetContext {
+  cv: OpenCv;
+  ort: OrtModule;
+  config: DetModelConfig;
+}
+
+interface DetRunContext extends DetContext {
+  session: InferenceSession;
+}
+
+function preprocess(
+  context: DetContext,
+  mats: Mat[],
+  params: InternalDetParams
+): DetPreprocessResult[] {
+  return mats.map((mat) => preprocessSample(context, mat, params));
+}
+
+function preprocessSample(
+  context: DetContext,
+  sourceMat: Mat,
+  params: InternalDetParams
+): DetPreprocessResult {
+  const { cv, ort, config } = context;
+  const srcW = sourceMat.cols;
+  const srcH = sourceMat.rows;
+  const limitSideLen = Math.max(32, params.limitSideLen);
+  const limitType: LimitType = params.limitType;
+  const maxSideLimit = Math.max(32, params.maxSideLimit);
+  let scale = 1.0;
+  if (limitType === "max") {
+    const maxSide = Math.max(srcW, srcH);
+    if (maxSide > limitSideLen) {
+      scale = limitSideLen / Math.max(1, maxSide);
+    }
+  } else {
+    const minSide = Math.min(srcW, srcH);
+    if (minSide < limitSideLen) {
+      scale = limitSideLen / Math.max(1, minSide);
+    }
+  }
+  let dstW = Math.max(32, Math.round((srcW * scale) / 32) * 32);
+  let dstH = Math.max(32, Math.round((srcH * scale) / 32) * 32);
+  if (Math.max(dstW, dstH) > maxSideLimit) {
+    const limitScale = maxSideLimit / Math.max(dstW, dstH);
+    dstW = Math.max(32, Math.floor(dstW * limitScale));
+    dstH = Math.max(32, Math.floor(dstH * limitScale));
+  }
+  dstW = clamp(dstW, 32, maxSideLimit);
+  dstH = clamp(dstH, 32, maxSideLimit);
+  dstW = Math.max(32, Math.round(dstW / 32) * 32);
+  dstH = Math.max(32, Math.round(dstH / 32) * 32);
+
+  const resized = new cv.Mat();
+  const bgr = new cv.Mat();
+  cv.resize(sourceMat, resized, new cv.Size(dstW, dstH), 0, 0, cv.INTER_LINEAR);
+  if (resized.channels() === 4) {
+    cv.cvtColor(resized, bgr, cv.COLOR_RGBA2BGR);
+  } else if (resized.channels() === 1) {
+    cv.cvtColor(resized, bgr, cv.COLOR_GRAY2BGR);
+  } else {
+    resized.copyTo(bgr);
+  }
+  const chw = toBgrFloatCHWFromBgr(bgr.data, dstW, dstH, config.normalize);
+  resized.delete();
+  bgr.delete();
+
+  return {
+    tensor: new ort.Tensor("float32", chw, [1, 3, dstH, dstW]),
+    srcW,
+    srcH,
+    dstW,
+    dstH
+  };
+}
+
+function getDetMap(outputTensor: Tensor): { data: Float32Array; h: number; w: number } {
+  const dims = outputTensor.dims;
+  const data = outputTensor.data as Float32Array;
+  if (dims.length === 4) return { data, h: dims[2], w: dims[3] };
+  if (dims.length === 3) return { data, h: dims[1], w: dims[2] };
+  throw new Error(`Unexpected det output dims: [${dims.join(", ")}]`);
+}
+
+function createBatchDetTensor(
+  ort: OrtModule,
+  preps: DetPreprocessResult[],
+  maxH: number,
+  maxW: number
+): Tensor {
+  const batch = preps.length;
+  const plane = 3 * maxH * maxW;
+  const out = new Float32Array(batch * plane);
+  for (let i = 0; i < batch; i += 1) {
+    const prep = preps[i];
+    const chw = prep.tensor.data as Float32Array;
+    const { dstH, dstW } = prep;
+    const base = i * plane;
+    for (let c = 0; c < 3; c += 1) {
+      const srcChannelBase = c * dstH * dstW;
+      const dstChannelBase = base + c * maxH * maxW;
+      for (let y = 0; y < dstH; y += 1) {
+        const srcRow = srcChannelBase + y * dstW;
+        const dstRow = dstChannelBase + y * maxW;
+        out.set(chw.subarray(srcRow, srcRow + dstW), dstRow);
+      }
+    }
+  }
+  return new ort.Tensor("float32", out, [batch, 3, maxH, maxW]);
+}
+
+function packDetBatchTensor(ort: OrtModule, preps: DetPreprocessResult[]): Tensor {
+  const maxH = Math.max(...preps.map((p) => p.dstH));
+  const maxW = Math.max(...preps.map((p) => p.dstW));
+  return createBatchDetTensor(ort, preps, maxH, maxW);
+}
+
+function batchDetOutputPlaneOffset(dims: readonly number[], batchIndex: number): number {
+  const tail = dims.slice(1).reduce((a, b) => a * b, 1);
+  return batchIndex * tail;
+}
+
+function detFeatureCropDims(
+  dstH: number,
+  dstW: number,
+  maxH: number,
+  maxW: number,
+  ohFull: number,
+  owFull: number
+): { cropOh: number; cropOw: number } {
+  const cropOh = Math.max(1, Math.min(ohFull, Math.round((ohFull * dstH) / maxH)));
+  const cropOw = Math.max(1, Math.min(owFull, Math.round((owFull * dstW) / maxW)));
+  return { cropOh, cropOw };
+}
+
+function sliceBatchedDetOutputPlane(
+  ort: OrtModule,
+  fullOutput: Tensor,
+  batchIndex: number,
+  cropOh: number,
+  cropOw: number,
+  ohFull: number,
+  owFull: number
+): Tensor {
+  const data = fullOutput.data as Float32Array;
+  const dims = fullOutput.dims;
+  const base = batchDetOutputPlaneOffset(dims, batchIndex);
+  const out = new Float32Array(cropOh * cropOw);
+  for (let r = 0; r < cropOh; r += 1) {
+    const rowStart = base + r * owFull;
+    out.set(data.subarray(rowStart, rowStart + cropOw), r * cropOw);
+  }
+  return new ort.Tensor("float32", out, [1, 1, cropOh, cropOw]);
+}
+
+function postprocess(
+  context: DetRunContext,
+  fullOutput: Tensor,
+  preps: DetPreprocessResult[],
+  params: InternalDetParams
+): InternalDetBatchItem[] {
+  const { cv, ort, config } = context;
+  const od = fullOutput.dims;
+  if (od.length !== 3 && od.length !== 4) {
+    throw new Error(`Unexpected det output dims: [${od.join(", ")}]`);
+  }
+  const ohFull = od.length === 4 ? od[2] : od[1];
+  const owFull = od.length === 4 ? od[3] : od[2];
+  const nOut = od.length === 4 ? od[0] : preps.length === 1 ? 1 : od[0];
+  if (nOut !== preps.length) {
+    throw new Error(
+      `Detection batch output N=${String(nOut)} does not match input batch ${String(preps.length)}`
+    );
+  }
+
+  const maxH = Math.max(...preps.map((p) => p.dstH));
+  const maxW = Math.max(...preps.map((p) => p.dstW));
+
+  const items: InternalDetBatchItem[] = [];
+  for (let i = 0; i < preps.length; i += 1) {
+    const prep = preps[i];
+    const { cropOh, cropOw } = detFeatureCropDims(prep.dstH, prep.dstW, maxH, maxW, ohFull, owFull);
+    const planeTensor = sliceBatchedDetOutputPlane(
+      ort,
+      fullOutput,
+      i,
+      cropOh,
+      cropOw,
+      ohFull,
+      owFull
+    );
+    const boxes = decodeDetOutput(
+      { cv, config },
+      planeTensor,
+      prep,
+      params.thresh,
+      params.boxThresh,
+      params.unclipRatio
+    );
+    items.push({ prep, boxes });
+  }
+  return items;
+}
+
+function decodeDetOutput(
+  context: { cv: OpenCv; config: DetModelConfig },
+  detOutput: Tensor,
+  meta: DetPreprocessResult,
+  detThresh: number,
+  boxThresh: number,
+  unclipRatio: number
+): DetBox[] {
+  const { cv, config } = context;
+  const { data, h, w } = getDetMap(detOutput);
+  const pred = cv.matFromArray(h, w, cv.CV_32FC1, data);
+  const maskData = new Uint8Array(h * w);
+  for (let i = 0; i < data.length; i += 1) {
+    maskData[i] = data[i] > detThresh ? 255 : 0;
+  }
+  const bitmap = cv.matFromArray(h, w, cv.CV_8UC1, maskData);
+  const contours = new cv.MatVector();
+  const hierarchy = new cv.Mat();
+  cv.findContours(bitmap, contours, hierarchy, cv.RETR_LIST, cv.CHAIN_APPROX_SIMPLE);
+
+  const boxes: DetBox[] = [];
+  const candidateCount = Math.min(config.postprocess.maxCandidates, contours.size());
+  for (let i = 0; i < candidateCount; i += 1) {
+    const contour = contours.get(i);
+    if (contour.rows < 4) {
+      contour.delete();
+      continue;
+    }
+    const points: Point2D[] = [];
+    for (let row = 0; row < contour.rows; row += 1) {
+      points.push([contour.data32S[row * 2], contour.data32S[row * 2 + 1]]);
+    }
+    const mini = getMiniBoxFromPoints(cv, points);
+    if (mini.side < DET_BOX_MIN_SIZE) {
+      contour.delete();
+      continue;
+    }
+    const score = boxScoreFast(cv, pred, mini.box);
+    if (score < boxThresh) {
+      contour.delete();
+      continue;
+    }
+    const expanded = unclip(mini.box, unclipRatio);
+    if (!expanded || expanded.length < 4) {
+      contour.delete();
+      continue;
+    }
+    const miniUnclip = getMiniBoxFromPoints(cv, expanded);
+    if (miniUnclip.side < DET_BOX_MIN_SIZE + 2) {
+      contour.delete();
+      continue;
+    }
+
+    const poly: Point2D[] = miniUnclip.box.map((point) => [
+      clamp(Math.round((point[0] * meta.srcW) / Math.max(1, w)), 0, meta.srcW),
+      clamp(Math.round((point[1] * meta.srcH) / Math.max(1, h)), 0, meta.srcH)
+    ]);
+    boxes.push({ poly, score });
+    contour.delete();
+  }
+
+  pred.delete();
+  bitmap.delete();
+  contours.delete();
+  hierarchy.delete();
+
+  boxes.sort((a, b) => a.poly[0][1] - b.poly[0][1] || a.poly[0][0] - b.poly[0][0]);
+  for (let i = 0; i < boxes.length - 1; i += 1) {
+    for (let j = i; j >= 0; j -= 1) {
+      if (
+        Math.abs(boxes[j + 1].poly[0][1] - boxes[j].poly[0][1]) < 10 &&
+        boxes[j + 1].poly[0][0] < boxes[j].poly[0][0]
+      ) {
+        const tmp = boxes[j];
+        boxes[j] = boxes[j + 1];
+        boxes[j + 1] = tmp;
+      } else {
+        break;
+      }
+    }
+  }
+
+  return boxes;
+}
diff --git a/paddleocr-js/packages/core/src/models/index.ts b/paddleocr-js/packages/core/src/models/index.ts
new file mode 100644
index 0000000..86d0d58
--- /dev/null
+++ b/paddleocr-js/packages/core/src/models/index.ts
@@ -0,0 +1,21 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+export {
+  DEFAULT_DET_MODEL_PARSE_FALLBACKS,
+  DEFAULT_DET_MODEL_CONFIG,
+  createDetModel,
+  createDetModelSession,
+  parseDetModelConfigText
+} from "./det";
+export {
+  DEFAULT_REC_MODEL_PARSE_FALLBACKS,
+  DEFAULT_REC_RUNTIME_LIMITS,
+  DEFAULT_REC_MODEL_CONFIG,
+  createRecModel,
+  createRecModelSession,
+  parseRecModelConfigText
+} from "./rec";
+export type { RecRuntimeOverrides } from "./rec";
diff --git a/paddleocr-js/packages/core/src/models/infer.ts b/paddleocr-js/packages/core/src/models/infer.ts
new file mode 100644
index 0000000..7b16b9f
--- /dev/null
+++ b/paddleocr-js/packages/core/src/models/infer.ts
@@ -0,0 +1,15 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { InferenceSession, Tensor } from "onnxruntime-web";
+
+export async function runInference(
+  session: InferenceSession,
+  inputTensor: Tensor
+): Promise<Tensor> {
+  const inputName = session.inputNames[0];
+  const outputMap = await session.run({ [inputName]: inputTensor });
+  return outputMap[session.outputNames[0]];
+}
diff --git a/paddleocr-js/packages/core/src/models/rec.ts b/paddleocr-js/packages/core/src/models/rec.ts
new file mode 100644
index 0000000..b85c72b
--- /dev/null
+++ b/paddleocr-js/packages/core/src/models/rec.ts
@@ -0,0 +1,306 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { OpenCv, Mat } from "@techstark/opencv-js";
+import type { Tensor } from "onnxruntime-web";
+
+import { assertModelResources } from "../resources/model-asset";
+import { createSession, getProviderCandidates, releaseSessions } from "../runtime/ort";
+import type { OrtModule, WebGpuState, SessionState } from "../runtime/ort";
+import { chunkArray, clamp, resolveRuntimeBatchSize, withTimeout } from "../utils/common";
+import { getTransformOp, parseInferenceConfigText, toBgrFloatCHWFromBgr } from "./common";
+import { runInference } from "./infer";
+import type { NormalizeConfig } from "./common";
+
+export interface RecModelConfig {
+  imageShape: number[];
+  charDict: string[];
+}
+
+export interface RecResult {
+  text: string;
+  score: number;
+}
+
+export interface RecRuntimeOverrides {
+  batchSize?: number;
+}
+
+export interface RecModel {
+  readonly kind: "rec";
+  readonly config: RecModelConfig;
+  readonly provider: string;
+  predict(cv: OpenCv, mats: Mat[], overrides?: RecRuntimeOverrides): Promise<RecResult[]>;
+  dispose(): Promise<void>;
+}
+
+interface RecSample {
+  inputIndex: number;
+  width: number;
+  chw: Float32Array;
+}
+
+const DEFAULT_REC_ALPHANUMERIC_DICT = "0123456789abcdefghijklmnopqrstuvwxyz".split("");
+
+const REC_NORMALIZE: Readonly<NormalizeConfig> = Object.freeze({
+  mean: [0.5, 0.5, 0.5],
+  std: [0.5, 0.5, 0.5],
+  scale: 1 / 255
+});
+
+export const DEFAULT_REC_MODEL_PARSE_FALLBACKS: Readonly<
+  Pick<RecModelConfig, "imageShape" | "charDict">
+> = Object.freeze({
+  imageShape: [3, 48, 320],
+  charDict: []
+});
+
+export const DEFAULT_REC_RUNTIME_LIMITS = Object.freeze({});
+
+const MAX_REC_WIDTH = 3200;
+
+export const DEFAULT_REC_MODEL_CONFIG: Readonly<RecModelConfig> = Object.freeze({
+  ...DEFAULT_REC_MODEL_PARSE_FALLBACKS
+});
+
+export function parseRecModelConfigText(text: string): RecModelConfig {
+  const parsed = parseInferenceConfigText(text);
+  const preProcess = parsed.PreProcess as Record<string, unknown> | undefined;
+  const transformOps = preProcess?.transform_ops as Array<Record<string, unknown>> | undefined;
+  const resize = getTransformOp(transformOps, "RecResizeImg");
+  const postprocess = (parsed.PostProcess || {}) as Record<string, unknown>;
+  const baseCharDict = postprocess.character_dict;
+
+  const imageShape = resize?.image_shape as number[] | undefined;
+  if (!imageShape || !Array.isArray(imageShape) || imageShape.length < 3) {
+    throw new Error("RecResizeImg.image_shape is required in rec inference.yml");
+  }
+
+  const charDict =
+    Array.isArray(baseCharDict) && baseCharDict.length > 0
+      ? [...(baseCharDict as string[]), " "]
+      : [...DEFAULT_REC_ALPHANUMERIC_DICT, " "];
+
+  return {
+    imageShape,
+    charDict
+  };
+}
+
+interface CreateRecModelArgs {
+  ort: OrtModule;
+  modelBytes: Uint8Array;
+  configText: string;
+  backend: string;
+  webgpuState: WebGpuState;
+  batchSize?: number;
+}
+
+export async function createRecModel({
+  ort,
+  modelBytes,
+  configText,
+  backend,
+  webgpuState,
+  batchSize: batchSizeArg
+}: CreateRecModelArgs): Promise<RecModel> {
+  assertModelResources("Recognition", {
+    model: modelBytes,
+    config: configText
+  });
+  const config = parseRecModelConfigText(configText);
+  const defaultBatchSize = Math.max(1, batchSizeArg ?? 1);
+  let sessionState: SessionState | null = await createRecModelSession(
+    ort,
+    modelBytes,
+    backend,
+    webgpuState
+  );
+
+  return {
+    kind: "rec",
+    config,
+    get provider() {
+      return sessionState?.provider || "";
+    },
+    async predict(cv, mats, overrides) {
+      if (!sessionState?.session) {
+        throw new Error("Recognition model session is not initialized.");
+      }
+      const batchSize = resolveRuntimeBatchSize(overrides?.batchSize, defaultBatchSize);
+      const ctx = { cv, config };
+      const samples = preprocess(ctx, mats);
+      const charDict = config.charDict;
+      const ordered = samples.slice().sort((a, b) => a.width - b.width);
+      const decoded: Array<{ inputIndex: number; text: string; score: number }> = [];
+      const targetH = config.imageShape[1];
+
+      for (const batch of chunkArray(ordered, batchSize)) {
+        const inputTensor = packRecBatchTensor(ort, batch, targetH);
+        const output = await runInference(sessionState.session, inputTensor);
+        const batchResults = postprocess(output, charDict);
+        for (let index = 0; index < batchResults.length; index += 1) {
+          decoded.push({
+            inputIndex: batch[index].inputIndex,
+            ...batchResults[index]
+          });
+        }
+      }
+
+      decoded.sort((a, b) => a.inputIndex - b.inputIndex);
+      return decoded.map(({ text, score }) => ({ text, score }));
+    },
+    async dispose() {
+      await releaseSessions(sessionState?.session);
+      sessionState = null;
+    }
+  };
+}
+
+export async function createRecModelSession(
+  ort: OrtModule,
+  modelBytes: Uint8Array,
+  backend: string,
+  webgpuState: WebGpuState
+): Promise<SessionState> {
+  const providerCandidates = getProviderCandidates(backend, webgpuState);
+  return withTimeout(
+    createSession(ort, modelBytes, providerCandidates),
+    60000,
+    "Recognition model"
+  );
+}
+
+function preprocess(context: { cv: OpenCv; config: RecModelConfig }, mats: Mat[]): RecSample[] {
+  const samples: RecSample[] = [];
+  for (let i = 0; i < mats.length; i += 1) {
+    samples.push(preprocessSample(context, mats[i], i));
+  }
+  return samples;
+}
+
+function preprocessSample(
+  context: { cv: OpenCv; config: RecModelConfig },
+  cropMat: Mat,
+  inputIndex: number
+): RecSample {
+  const { cv, config } = context;
+  const [channels, targetH, baseW] = config.imageShape;
+  const srcW = cropMat.cols;
+  const srcH = cropMat.rows;
+  if (channels !== 3) {
+    throw new Error(`Unexpected recognition channels: ${String(channels)}`);
+  }
+  const ratio = srcW / Math.max(1, srcH);
+  const maxWhRatio = Math.max(baseW / Math.max(1, targetH), ratio);
+  const recW = clamp(Math.trunc(targetH * maxWhRatio), 1, MAX_REC_WIDTH);
+  const resizedW = Math.min(recW, Math.ceil(targetH * ratio));
+  const resized = new cv.Mat();
+  const bgr = new cv.Mat();
+  cv.resize(cropMat, resized, new cv.Size(resizedW, targetH), 0, 0, cv.INTER_LINEAR);
+  if (resized.channels() === 4) {
+    cv.cvtColor(resized, bgr, cv.COLOR_RGBA2BGR);
+  } else if (resized.channels() === 1) {
+    cv.cvtColor(resized, bgr, cv.COLOR_GRAY2BGR);
+  } else {
+    resized.copyTo(bgr);
+  }
+  const resizedChw = toBgrFloatCHWFromBgr(bgr.data, resizedW, targetH, REC_NORMALIZE);
+  const chw = new Float32Array(3 * targetH * recW);
+  const dstPerChannel = targetH * recW;
+  const srcPerChannel = targetH * resizedW;
+  for (let channel = 0; channel < 3; channel += 1) {
+    for (let row = 0; row < targetH; row += 1) {
+      const srcStart = channel * srcPerChannel + row * resizedW;
+      const dstStart = channel * dstPerChannel + row * recW;
+      chw.set(resizedChw.subarray(srcStart, srcStart + resizedW), dstStart);
+    }
+  }
+  bgr.delete();
+  resized.delete();
+  return { inputIndex, width: recW, chw };
+}
+
+function createBatchTensor(
+  ort: OrtModule,
+  samples: RecSample[],
+  maxW: number,
+  targetH: number
+): Tensor {
+  const batch = samples.length;
+  const out = new Float32Array(batch * 3 * targetH * maxW);
+  const dstPerChannel = targetH * maxW;
+  for (let index = 0; index < batch; index += 1) {
+    const sample = samples[index];
+    const srcW = sample.width;
+    const srcPerChannel = targetH * srcW;
+    for (let channel = 0; channel < 3; channel += 1) {
+      const srcBase = channel * srcPerChannel;
+      const dstBase = index * (3 * dstPerChannel) + channel * dstPerChannel;
+      for (let row = 0; row < targetH; row += 1) {
+        const srcStart = srcBase + row * srcW;
+        const dstStart = dstBase + row * maxW;
+        out.set(sample.chw.subarray(srcStart, srcStart + srcW), dstStart);
+      }
+    }
+  }
+  return new ort.Tensor("float32", out, [batch, 3, targetH, maxW]);
+}
+
+function packRecBatchTensor(ort: OrtModule, samples: RecSample[], targetH: number): Tensor {
+  const maxW = samples.reduce((acc, sample) => Math.max(acc, sample.width), 1);
+  return createBatchTensor(ort, samples, maxW, targetH);
+}
+
+function decodeCTCSample(
+  data: Float32Array,
+  offset: number,
+  timeSteps: number,
+  classes: number,
+  charDict: string[]
+): { text: string; score: number } {
+  let prevIdx = -1;
+  let text = "";
+  const probs: number[] = [];
+  for (let step = 0; step < timeSteps; step += 1) {
+    let maxIdx = 0;
+    let maxVal = -Infinity;
+    const stepOffset = offset + step * classes;
+    for (let cls = 0; cls < classes; cls += 1) {
+      const value = data[stepOffset + cls];
+      if (value > maxVal) {
+        maxVal = value;
+        maxIdx = cls;
+      }
+    }
+    if (maxIdx > 0 && maxIdx !== prevIdx) {
+      const dictIdx = maxIdx - 1;
+      if (dictIdx >= 0 && dictIdx < charDict.length) {
+        text += charDict[dictIdx];
+        probs.push(maxVal);
+      }
+    }
+    prevIdx = maxIdx;
+  }
+  const score = probs.length ? probs.reduce((a, b) => a + b, 0) / probs.length : 0;
+  return { text, score };
+}
+
+function postprocess(output: Tensor, charDict: string[]): Array<{ text: string; score: number }> {
+  const dims = output.dims;
+  if (dims.length !== 3) {
+    throw new Error(`Unexpected rec output dims: [${dims.join(", ")}]`);
+  }
+  const sampleCount = dims[0];
+  const timeSteps = dims[1];
+  const classes = dims[2];
+  const data = output.data as Float32Array;
+  const stride = timeSteps * classes;
+  const results: Array<{ text: string; score: number }> = [];
+  for (let index = 0; index < sampleCount; index += 1) {
+    results.push(decodeCTCSample(data, index * stride, timeSteps, classes, charDict));
+  }
+  return results;
+}
diff --git a/paddleocr-js/packages/core/src/pipelines/index.ts b/paddleocr-js/packages/core/src/pipelines/index.ts
new file mode 100644
index 0000000..617b83f
--- /dev/null
+++ b/paddleocr-js/packages/core/src/pipelines/index.ts
@@ -0,0 +1,6 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+export { PaddleOCR, parseOcrPipelineConfigText, normalizeOcrPipelineConfig } from "./ocr/index";
diff --git a/paddleocr-js/packages/core/src/pipelines/ocr/config.ts b/paddleocr-js/packages/core/src/pipelines/ocr/config.ts
new file mode 100644
index 0000000..c2c5a09
--- /dev/null
+++ b/paddleocr-js/packages/core/src/pipelines/ocr/config.ts
@@ -0,0 +1,217 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import yaml from "js-yaml";
+
+import type { ModelAsset } from "../../resources/model-asset";
+import { normalizeModelAsset } from "../../resources/model-asset";
+import type { LimitType } from "./runtime-params";
+
+const SUPPORTED_PIPELINE_NAME = "OCR";
+
+export interface NormalizedPipelineConfig {
+  pipelineName: string;
+  raw: Record<string, unknown>;
+  warnings: string[];
+  unsupportedFeatures: string[];
+  modelSelection: PipelineModelSelection;
+  assets: Partial<Record<string, ModelAsset>>;
+  runtimeDefaults: PipelineRuntimeDefaults;
+  pipelineBatchSize: number;
+  textDetectionBatchSize: number;
+  textRecognitionBatchSize: number;
+}
+
+export interface PipelineModelSelection {
+  textDetectionModelName: string | null;
+  textRecognitionModelName: string | null;
+}
+
+export interface PipelineRuntimeDefaults {
+  text_det_limit_side_len?: number;
+  text_det_limit_type?: LimitType;
+  text_det_max_side_limit?: number;
+  text_det_thresh?: number;
+  text_det_box_thresh?: number;
+  text_det_unclip_ratio?: number;
+  text_rec_score_thresh?: number;
+}
+
+type YamlObject = Record<string, unknown>;
+
+function isPlainObject(value: unknown): value is YamlObject {
+  return Boolean(value) && typeof value === "object" && !Array.isArray(value);
+}
+
+export function toFiniteNumber(value: unknown): number | undefined {
+  if (value === null || value === undefined || value === "") {
+    return undefined;
+  }
+  const normalized = Number(value);
+  return Number.isFinite(normalized) ? normalized : undefined;
+}
+
+function batchSizeOrOne(value: unknown): number {
+  const n = toFiniteNumber(value);
+  return n !== undefined && n >= 1 ? n : 1;
+}
+
+function applyGeneralPipelineRuntimeDefaults(
+  textType: string,
+  runtimeDefaults: PipelineRuntimeDefaults
+): PipelineRuntimeDefaults {
+  if (textType !== "general") {
+    return runtimeDefaults;
+  }
+  return {
+    text_det_limit_side_len: runtimeDefaults.text_det_limit_side_len ?? 960,
+    text_det_limit_type: runtimeDefaults.text_det_limit_type ?? "max",
+    text_det_max_side_limit: runtimeDefaults.text_det_max_side_limit ?? 4000,
+    text_det_thresh: runtimeDefaults.text_det_thresh ?? 0.3,
+    text_det_box_thresh: runtimeDefaults.text_det_box_thresh ?? 0.6,
+    text_det_unclip_ratio: runtimeDefaults.text_det_unclip_ratio ?? 2.0,
+    text_rec_score_thresh: runtimeDefaults.text_rec_score_thresh ?? 0
+  };
+}
+
+function parsePipelineConfigInput(input: unknown): YamlObject {
+  if (typeof input === "string") {
+    const parsed = yaml.load(input);
+    if (!isPlainObject(parsed)) {
+      throw new Error("OCR pipeline config text must decode to an object.");
+    }
+    return parsed;
+  }
+  if (!isPlainObject(input)) {
+    throw new Error("OCR pipeline config must be an object or YAML text.");
+  }
+  return input;
+}
+
+function addFeatureWarning(warnings: string[], featureName: string, reason?: string): void {
+  warnings.push(
+    `${featureName} is not yet supported in PaddleOCR.js${reason ? `: ${reason}` : ""}.`
+  );
+}
+
+function getModuleModelName(moduleConfig: YamlObject | null): string | null {
+  return typeof moduleConfig?.model_name === "string" ? moduleConfig.model_name : null;
+}
+
+function validateModuleAsset(modulePath: string, modelName: string | null): void {
+  if (!modelName) {
+    throw new Error(
+      `${modulePath}.model_name must be provided when ${modulePath}.model_dir is set.`
+    );
+  }
+}
+
+function getModuleAsset(
+  assetName: string,
+  modulePath: string,
+  moduleConfig: YamlObject | null
+): ModelAsset | null {
+  if (moduleConfig?.model_dir == null) {
+    return null;
+  }
+  if (isPlainObject(moduleConfig.model_dir)) {
+    const asset = normalizeModelAsset(assetName, moduleConfig.model_dir);
+    validateModuleAsset(modulePath, getModuleModelName(moduleConfig));
+    return asset;
+  }
+  throw new Error(
+    `${modulePath}.model_dir must be null or an asset descriptor object in browser usage.`
+  );
+}
+
+export function parseOcrPipelineConfigText(text: string): YamlObject {
+  return parsePipelineConfigInput(text);
+}
+
+export function normalizeOcrPipelineConfig(input: unknown): NormalizedPipelineConfig {
+  const config = parsePipelineConfigInput(input);
+  const pipelineName = (config.pipeline_name as string | undefined) ?? SUPPORTED_PIPELINE_NAME;
+
+  if (pipelineName !== SUPPORTED_PIPELINE_NAME) {
+    throw new Error(
+      `Unsupported pipeline_name "${pipelineName}". PaddleOCR.js currently supports only "${SUPPORTED_PIPELINE_NAME}".`
+    );
+  }
+
+  const warnings: string[] = [];
+  const subModules = isPlainObject(config.SubModules) ? config.SubModules : {};
+  const textDetection = isPlainObject(subModules.TextDetection) ? subModules.TextDetection : null;
+  const textRecognition = isPlainObject(subModules.TextRecognition)
+    ? subModules.TextRecognition
+    : null;
+
+  if (!textDetection || !textRecognition) {
+    throw new Error(
+      'OCR pipeline config must define both "SubModules.TextDetection" and "SubModules.TextRecognition".'
+    );
+  }
+
+  const useDocPreprocessor = Boolean(config.use_doc_preprocessor);
+  const useTextlineOrientation = Boolean(config.use_textline_orientation);
+  const subPipelines = config.SubPipelines as YamlObject | undefined;
+  const docPreprocessor = isPlainObject(subPipelines?.DocPreprocessor)
+    ? subPipelines.DocPreprocessor
+    : null;
+  const textLineOrientation = isPlainObject(subModules.TextLineOrientation)
+    ? subModules.TextLineOrientation
+    : null;
+
+  if (useDocPreprocessor || docPreprocessor) {
+    addFeatureWarning(warnings, "DocPreprocessor", "config will be ignored for now");
+  }
+  if (useTextlineOrientation || textLineOrientation) {
+    addFeatureWarning(warnings, "TextLineOrientation", "config will be ignored for now");
+  }
+  const textType =
+    typeof config.text_type === "string" && config.text_type.length > 0
+      ? config.text_type
+      : "general";
+
+  if (config.text_type && config.text_type !== "general") {
+    warnings.push(`text_type ${JSON.stringify(config.text_type)} is not used by PaddleOCR.js yet.`);
+  }
+
+  const detAsset = getModuleAsset("det", "SubModules.TextDetection", textDetection);
+  const recAsset = getModuleAsset("rec", "SubModules.TextRecognition", textRecognition);
+
+  const pipelineBatchSize = batchSizeOrOne(config.batch_size);
+  const textDetectionBatchSize = batchSizeOrOne(textDetection.batch_size);
+  const textRecognitionBatchSizeFromModule = batchSizeOrOne(textRecognition.batch_size);
+
+  return {
+    pipelineName,
+    raw: config,
+    warnings,
+    unsupportedFeatures: [
+      ...(useDocPreprocessor || docPreprocessor ? ["DocPreprocessor"] : []),
+      ...(useTextlineOrientation || textLineOrientation ? ["TextLineOrientation"] : [])
+    ],
+    modelSelection: {
+      textDetectionModelName: getModuleModelName(textDetection),
+      textRecognitionModelName: getModuleModelName(textRecognition)
+    },
+    assets: {
+      ...(detAsset ? { det: detAsset } : {}),
+      ...(recAsset ? { rec: recAsset } : {})
+    },
+    runtimeDefaults: applyGeneralPipelineRuntimeDefaults(textType, {
+      text_det_limit_side_len: toFiniteNumber(textDetection.limit_side_len),
+      text_det_limit_type: (textDetection.limit_type as LimitType | undefined) || undefined,
+      text_det_max_side_limit: toFiniteNumber(textDetection.max_side_limit),
+      text_det_thresh: toFiniteNumber(textDetection.thresh),
+      text_det_box_thresh: toFiniteNumber(textDetection.box_thresh),
+      text_det_unclip_ratio: toFiniteNumber(textDetection.unclip_ratio),
+      text_rec_score_thresh: toFiniteNumber(textRecognition.score_thresh)
+    }),
+    pipelineBatchSize,
+    textDetectionBatchSize,
+    textRecognitionBatchSize: textRecognitionBatchSizeFromModule
+  };
+}
diff --git a/paddleocr-js/packages/core/src/pipelines/ocr/core.ts b/paddleocr-js/packages/core/src/pipelines/ocr/core.ts
new file mode 100644
index 0000000..280f014
--- /dev/null
+++ b/paddleocr-js/packages/core/src/pipelines/ocr/core.ts
@@ -0,0 +1,332 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { OpenCv, Mat } from "@techstark/opencv-js";
+import type { ModelAsset, ModelLoadSummary } from "../../resources/model-asset";
+import { loadModelAsset } from "../../resources/index";
+import { createDetModel, createRecModel } from "../../models/index";
+import type { DetModel } from "../../models/det";
+import type { RecModel } from "../../models/rec";
+import type { Point2D } from "../../models/common";
+import { cropByPoly } from "./crop";
+import { initOpenCvRuntime } from "../../runtime/opencv";
+import { initOrtRuntime } from "../../runtime/ort";
+import type { OrtModule, WebGpuState, OrtOptions } from "../../runtime/ort";
+import { chunkArray, nowMs } from "../../utils/common";
+import type { OcrModelConfig, OcrRuntimeParamsInput } from "./runtime-params";
+import { getOcrRuntimeParams } from "./runtime-params";
+import type { NormalizedPipelineConfig } from "./config";
+import { cloneDefaultOcrConfig, validateLoadedModelName } from "./shared";
+import type { NormalizedOrtOptions } from "./shared";
+import type { SourceMatResult } from "../../platform/browser";
+
+export interface OcrResultItem {
+  poly: Point2D[];
+  text: string;
+  score: number;
+}
+
+export interface OcrResultMetrics {
+  detMs: number;
+  recMs: number;
+  totalMs: number;
+  detectedBoxes: number;
+  recognizedCount: number;
+}
+
+export interface OcrResultRuntime {
+  requestedBackend: string;
+  detProvider: string;
+  recProvider: string;
+  webgpuAvailable: boolean;
+}
+
+export interface OcrResult {
+  image: { width: number; height: number };
+  items: OcrResultItem[];
+  metrics: OcrResultMetrics;
+  runtime: OcrResultRuntime;
+}
+
+export interface InitializationSummary {
+  backend: string;
+  webgpuAvailable: boolean;
+  detProvider: string;
+  recProvider: string;
+  assets: ModelLoadSummary[];
+  elapsedMs: number;
+  pipelineConfigWarnings: string[];
+}
+
+export type SourceToMatFn = (
+  cv: OpenCv,
+  source: unknown
+) => SourceMatResult | Promise<SourceMatResult>;
+type EnsureServedFromHttpFn = () => void;
+
+export interface OcrPipelineRunnerOptions {
+  pipelineConfig: NormalizedPipelineConfig;
+  ortOptions?: OrtOptions | NormalizedOrtOptions;
+  fetch?: typeof fetch;
+  ensureServedFromHttp?: EnsureServedFromHttpFn;
+  sourceToMat?: SourceToMatFn;
+}
+
+function noopEnsureServedFromHttp(): void {}
+
+function getResolvedAssets(assets: Partial<Record<string, ModelAsset>> | undefined): {
+  det: ModelAsset;
+  rec: ModelAsset;
+} {
+  const det = assets?.det;
+  const rec = assets?.rec;
+  if (!det || typeof det !== "object" || !rec || typeof rec !== "object") {
+    throw new Error(
+      "PaddleOCRCore requires pre-resolved detection and recognition asset descriptors."
+    );
+  }
+  return { det, rec };
+}
+
+export class OcrPipelineRunner {
+  protected options: OcrPipelineRunnerOptions;
+  protected modelConfig: OcrModelConfig;
+  protected runtimeDefaults: Partial<OcrRuntimeParamsInput>;
+  protected cv: OpenCv | null;
+  protected ort: OrtModule | null;
+  protected detModel: DetModel | null;
+  protected recModel: RecModel | null;
+  protected webgpuState: WebGpuState;
+  protected pipelineConfig: NormalizedPipelineConfig;
+  protected lastInitializationSummary: InitializationSummary | null;
+  private ensureServedFromHttp: EnsureServedFromHttpFn;
+  private sourceToMat: SourceToMatFn | undefined;
+
+  constructor(options: OcrPipelineRunnerOptions) {
+    this.options = options;
+    this.modelConfig = cloneDefaultOcrConfig();
+    this.pipelineConfig = options.pipelineConfig;
+    this.runtimeDefaults = { ...options.pipelineConfig.runtimeDefaults };
+    this.cv = null;
+    this.ort = null;
+    this.detModel = null;
+    this.recModel = null;
+    this.webgpuState = { available: false, reason: "" };
+    this.lastInitializationSummary = null;
+    this.ensureServedFromHttp = options.ensureServedFromHttp || noopEnsureServedFromHttp;
+    this.sourceToMat = options.sourceToMat;
+  }
+
+  async initialize(): Promise<InitializationSummary> {
+    this.ensureServedFromHttp();
+    const start = nowMs();
+    const { cv } = await initOpenCvRuntime();
+    this.cv = cv;
+    const { ort, webgpuState, backend } = await initOrtRuntime(this.options.ortOptions || {});
+    this.ort = ort;
+    this.webgpuState = webgpuState;
+
+    const assets = getResolvedAssets(this.pipelineConfig.assets);
+    const fetchImpl = this.options.fetch || fetch;
+    const loadedAssets = await Promise.all([
+      loadModelAsset(assets.det, fetchImpl),
+      loadModelAsset(assets.rec, fetchImpl)
+    ]);
+    validateLoadedModelName(
+      "TextDetection",
+      this.pipelineConfig.modelSelection.textDetectionModelName,
+      loadedAssets[0].configText
+    );
+    validateLoadedModelName(
+      "TextRecognition",
+      this.pipelineConfig.modelSelection.textRecognitionModelName,
+      loadedAssets[1].configText
+    );
+    await this.disposeModelsOnly();
+    const detBatchSize = this.pipelineConfig.textDetectionBatchSize;
+    const recBatchSize = this.pipelineConfig.textRecognitionBatchSize;
+
+    const [detModel, recModel] = await Promise.all([
+      createDetModel({
+        ort: this.ort,
+        modelBytes: loadedAssets[0].modelBytes,
+        configText: loadedAssets[0].configText,
+        backend,
+        webgpuState,
+        batchSize: detBatchSize
+      }),
+      createRecModel({
+        ort: this.ort,
+        modelBytes: loadedAssets[1].modelBytes,
+        configText: loadedAssets[1].configText,
+        backend,
+        webgpuState,
+        batchSize: recBatchSize
+      })
+    ]);
+    this.detModel = detModel;
+    this.recModel = recModel;
+    this.modelConfig = {
+      det: this.detModel.config,
+      rec: this.recModel.config
+    };
+
+    const elapsed = nowMs() - start;
+    this.lastInitializationSummary = {
+      backend,
+      webgpuAvailable: webgpuState.available,
+      detProvider: this.detModel.provider,
+      recProvider: this.recModel.provider,
+      assets: loadedAssets.map((asset) => asset.download),
+      elapsedMs: elapsed,
+      pipelineConfigWarnings: this.pipelineConfig.warnings
+    };
+    return this.lastInitializationSummary;
+  }
+
+  getInitializationSummary(): InitializationSummary | null {
+    return this.lastInitializationSummary;
+  }
+
+  getModelConfig(): OcrModelConfig {
+    return this.modelConfig;
+  }
+
+  async predict(input: unknown, params: OcrRuntimeParamsInput = {}): Promise<OcrResult[]> {
+    if (!this.sourceToMat) {
+      throw new Error("PaddleOCR source adapter is not configured.");
+    }
+    if (!this.detModel || !this.recModel || !this.cv || !this.ort) {
+      await this.initialize();
+    }
+
+    const cv = this.cv;
+    const detModel = this.detModel;
+    const recModel = this.recModel;
+    if (!cv || !detModel || !recModel) {
+      throw new Error("Initialization did not complete. Call initialize() first.");
+    }
+
+    const sources = Array.isArray(input) ? input : [input];
+    const sourceToMat = this.sourceToMat;
+    const pipelineBatchSize = Math.max(1, Math.floor(this.pipelineConfig.pipelineBatchSize) || 1);
+    const sourceBatches = chunkArray(sources, pipelineBatchSize);
+
+    const totalStart = nowMs();
+    const resolved = getOcrRuntimeParams(this.modelConfig, this.runtimeDefaults, params);
+
+    let sumDetMs = 0;
+    let sumRecMs = 0;
+    const partials: Array<{
+      image: { width: number; height: number };
+      items: OcrResultItem[];
+      detectedBoxes: number;
+      recognizedCount: number;
+    }> = [];
+
+    for (const batchSources of sourceBatches) {
+      const sourceImages = await Promise.all(
+        batchSources.map((source) => Promise.resolve(sourceToMat(cv, source)))
+      );
+      try {
+        const detStart = nowMs();
+        const detResults = await detModel.predict(
+          cv,
+          sourceImages.map((s) => s.mat),
+          resolved.det
+        );
+        sumDetMs += nowMs() - detStart;
+
+        const recStart = nowMs();
+        const perImageItems: OcrResultItem[][] = [];
+
+        for (let imgIdx = 0; imgIdx < detResults.length; imgIdx += 1) {
+          const detBoxes = detResults[imgIdx]?.boxes ?? [];
+          const cropMats: Mat[] = [];
+          for (let boxIdx = 0; boxIdx < detBoxes.length; boxIdx += 1) {
+            cropMats.push(cropByPoly(cv, sourceImages[imgIdx].mat, detBoxes[boxIdx].poly));
+          }
+
+          try {
+            const recResults = cropMats.length ? await recModel.predict(cv, cropMats) : [];
+            const items: OcrResultItem[] = [];
+            for (let boxIdx = 0; boxIdx < recResults.length; boxIdx += 1) {
+              const rec = recResults[boxIdx];
+              if (rec.text && rec.score >= resolved.pipeline.scoreThresh) {
+                items.push({
+                  poly: detBoxes[boxIdx].poly,
+                  text: rec.text,
+                  score: rec.score
+                });
+              }
+            }
+            perImageItems.push(items);
+          } finally {
+            for (const mat of cropMats) {
+              mat.delete();
+            }
+          }
+        }
+
+        sumRecMs += nowMs() - recStart;
+
+        for (let i = 0; i < sourceImages.length; i += 1) {
+          const sourceImage = sourceImages[i];
+          const detBoxes = detResults[i]?.boxes ?? [];
+          const items = perImageItems[i] ?? [];
+          partials.push({
+            image: {
+              width: sourceImage.width,
+              height: sourceImage.height
+            },
+            items,
+            detectedBoxes: detBoxes.length,
+            recognizedCount: items.length
+          });
+        }
+      } finally {
+        for (const sourceImage of sourceImages) {
+          sourceImage.dispose();
+        }
+      }
+    }
+
+    const totalElapsed = nowMs() - totalStart;
+    const requestedBackend =
+      (this.options.ortOptions as NormalizedOrtOptions | undefined)?.backend ?? "auto";
+
+    return partials.map(
+      (p): OcrResult => ({
+        image: p.image,
+        items: p.items,
+        metrics: {
+          detMs: sumDetMs,
+          recMs: sumRecMs,
+          totalMs: totalElapsed,
+          detectedBoxes: p.detectedBoxes,
+          recognizedCount: p.recognizedCount
+        },
+        runtime: {
+          requestedBackend,
+          detProvider: detModel.provider,
+          recProvider: recModel.provider,
+          webgpuAvailable: this.webgpuState.available
+        }
+      })
+    );
+  }
+
+  async disposeModelsOnly(): Promise<void> {
+    await Promise.all([this.detModel?.dispose(), this.recModel?.dispose()]);
+    this.detModel = null;
+    this.recModel = null;
+  }
+
+  async dispose(): Promise<void> {
+    await this.disposeModelsOnly();
+  }
+}
+
+export { OcrPipelineRunner as PaddleOCRCore };
diff --git a/paddleocr-js/packages/core/src/pipelines/ocr/crop.ts b/paddleocr-js/packages/core/src/pipelines/ocr/crop.ts
new file mode 100644
index 0000000..97d8c0b
--- /dev/null
+++ b/paddleocr-js/packages/core/src/pipelines/ocr/crop.ts
@@ -0,0 +1,53 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { OpenCv, Mat } from "@techstark/opencv-js";
+
+import { getMiniBoxFromPoints } from "../../models/common";
+import type { Point2D } from "../../models/common";
+
+export function cropByPoly(cv: OpenCv, srcMat: Mat, poly: Point2D[]): Mat {
+  const ordered = getMiniBoxFromPoints(cv, poly).box;
+  const widthTop = Math.hypot(ordered[1][0] - ordered[0][0], ordered[1][1] - ordered[0][1]);
+  const widthBottom = Math.hypot(ordered[2][0] - ordered[3][0], ordered[2][1] - ordered[3][1]);
+  const heightLeft = Math.hypot(ordered[3][0] - ordered[0][0], ordered[3][1] - ordered[0][1]);
+  const heightRight = Math.hypot(ordered[2][0] - ordered[1][0], ordered[2][1] - ordered[1][1]);
+  const cropW = Math.max(1, Math.floor(Math.max(widthTop, widthBottom)));
+  const cropH = Math.max(1, Math.floor(Math.max(heightLeft, heightRight)));
+
+  const srcTri = cv.matFromArray(4, 1, cv.CV_32FC2, [
+    ordered[0][0],
+    ordered[0][1],
+    ordered[1][0],
+    ordered[1][1],
+    ordered[2][0],
+    ordered[2][1],
+    ordered[3][0],
+    ordered[3][1]
+  ]);
+  const dstTri = cv.matFromArray(4, 1, cv.CV_32FC2, [0, 0, cropW, 0, cropW, cropH, 0, cropH]);
+  const transform = cv.getPerspectiveTransform(srcTri, dstTri);
+  const warped = new cv.Mat();
+  cv.warpPerspective(
+    srcMat,
+    warped,
+    transform,
+    new cv.Size(cropW, cropH),
+    cv.INTER_CUBIC,
+    cv.BORDER_REPLICATE,
+    new cv.Scalar()
+  );
+  srcTri.delete();
+  dstTri.delete();
+  transform.delete();
+
+  if (warped.rows / Math.max(1, warped.cols) >= 1.5) {
+    const rotated = new cv.Mat();
+    cv.rotate(warped, rotated, cv.ROTATE_90_COUNTERCLOCKWISE);
+    warped.delete();
+    return rotated;
+  }
+  return warped;
+}
diff --git a/paddleocr-js/packages/core/src/pipelines/ocr/default-config.ts b/paddleocr-js/packages/core/src/pipelines/ocr/default-config.ts
new file mode 100644
index 0000000..130c1fb
--- /dev/null
+++ b/paddleocr-js/packages/core/src/pipelines/ocr/default-config.ts
@@ -0,0 +1,51 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+export const DEFAULT_OCR_PIPELINE_CONFIG_TEXT = `
+pipeline_name: OCR
+
+text_type: general
+
+use_doc_preprocessor: False
+use_textline_orientation: False
+
+SubPipelines:
+  DocPreprocessor:
+    pipeline_name: doc_preprocessor
+    use_doc_orientation_classify: False
+    use_doc_unwarping: False
+    SubModules:
+      DocOrientationClassify:
+        module_name: doc_text_orientation
+        model_name: PP-LCNet_x1_0_doc_ori
+        model_dir: null
+      DocUnwarping:
+        module_name: image_unwarping
+        model_name: UVDoc
+        model_dir: null
+
+SubModules:
+  TextDetection:
+    module_name: text_detection
+    model_name: PP-OCRv5_mobile_det
+    model_dir: null
+    limit_side_len: 64
+    limit_type: min
+    max_side_limit: 4000
+    thresh: 0.3
+    box_thresh: 0.6
+    unclip_ratio: 1.5
+  TextLineOrientation:
+    module_name: textline_orientation
+    model_name: PP-LCNet_x1_0_textline_ori
+    model_dir: null
+    batch_size: 6
+  TextRecognition:
+    module_name: text_recognition
+    model_name: PP-OCRv5_mobile_rec
+    model_dir: null
+    batch_size: 6
+    score_thresh: 0.0
+`.trimStart();
diff --git a/paddleocr-js/packages/core/src/pipelines/ocr/index.ts b/paddleocr-js/packages/core/src/pipelines/ocr/index.ts
new file mode 100644
index 0000000..10bbafa
--- /dev/null
+++ b/paddleocr-js/packages/core/src/pipelines/ocr/index.ts
@@ -0,0 +1,100 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import { normalizeOcrPipelineConfig, parseOcrPipelineConfigText } from "./config";
+import { ensureServedFromHttp, sourceToMat } from "../../platform/browser";
+import type { OcrPipelineRunnerOptions } from "./core";
+import { OcrPipelineRunner } from "./core";
+import { resolvePaddleOCROptions, resolveWorkerOptions } from "./shared";
+import { createWorkerBackedPaddleOCR } from "./worker-backed";
+import type { WorkerBackedPaddleOCR } from "./worker-backed";
+import type { OrtOptions } from "../../runtime/ort";
+import type { ModelAsset } from "../../resources/model-asset";
+import type { LimitType } from "./runtime-params";
+
+export interface PaddleOCRCreateOptions {
+  worker?: boolean | { createWorker?: () => Worker };
+  fetch?: typeof fetch;
+  initialize?: boolean;
+  ortOptions?: OrtOptions;
+
+  pipelineConfig?: unknown;
+  unsupportedBehavior?: "warn" | "ignore" | "error";
+
+  lang?: string;
+  ocrVersion?: string;
+  ocr_version?: string;
+
+  textDetectionModelName?: string;
+  text_detection_model_name?: string;
+  textRecognitionModelName?: string;
+  text_recognition_model_name?: string;
+
+  textDetectionModelAsset?: ModelAsset;
+  textDetectionModelDir?: ModelAsset;
+  text_detection_model_dir?: ModelAsset;
+  textRecognitionModelAsset?: ModelAsset;
+  textRecognitionModelDir?: ModelAsset;
+  text_recognition_model_dir?: ModelAsset;
+
+  textDetectionBatchSize?: number;
+  text_detection_batch_size?: number;
+  textRecognitionBatchSize?: number;
+  text_recognition_batch_size?: number;
+  batch_size?: number;
+
+  textDetLimitSideLen?: number;
+  text_det_limit_side_len?: number;
+  textDetLimitType?: LimitType;
+  text_det_limit_type?: LimitType;
+  textDetMaxSideLimit?: number;
+  text_det_max_side_limit?: number;
+  textDetThresh?: number;
+  text_det_thresh?: number;
+  textDetBoxThresh?: number;
+  text_det_box_thresh?: number;
+  textDetUnclipRatio?: number;
+  text_det_unclip_ratio?: number;
+  textRecScoreThresh?: number;
+  text_rec_score_thresh?: number;
+
+  [key: string]: unknown;
+}
+
+export class PaddleOCR extends OcrPipelineRunner {
+  constructor(options: OcrPipelineRunnerOptions) {
+    super({
+      ...options,
+      ensureServedFromHttp,
+      sourceToMat
+    });
+  }
+
+  static async create(
+    options: PaddleOCRCreateOptions = {}
+  ): Promise<PaddleOCR | WorkerBackedPaddleOCR> {
+    const workerOptions = resolveWorkerOptions(options.worker);
+    if (workerOptions.enabled && options.fetch) {
+      throw new Error("worker mode does not support a custom fetch implementation.");
+    }
+
+    const resolvedOptions = resolvePaddleOCROptions(options);
+    const instance = workerOptions.enabled
+      ? createWorkerBackedPaddleOCR(resolvedOptions, {
+          createWorker: workerOptions.createWorker ?? undefined
+        })
+      : new PaddleOCR({
+          ...resolvedOptions,
+          fetch: options.fetch
+        });
+
+    if (options.initialize !== false) {
+      await instance.initialize();
+    }
+    return instance;
+  }
+}
+
+export { normalizeOcrPipelineConfig, parseOcrPipelineConfigText };
diff --git a/paddleocr-js/packages/core/src/pipelines/ocr/runtime-params.ts b/paddleocr-js/packages/core/src/pipelines/ocr/runtime-params.ts
new file mode 100644
index 0000000..d038c92
--- /dev/null
+++ b/paddleocr-js/packages/core/src/pipelines/ocr/runtime-params.ts
@@ -0,0 +1,125 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { DetModelConfig, DetRuntimeOverrides, LimitType } from "../../models/det";
+import type { RecModelConfig } from "../../models/rec";
+
+export type { LimitType };
+
+export interface OcrModelConfig {
+  det: DetModelConfig;
+  rec: RecModelConfig;
+}
+
+export interface ResolvedOcrParams {
+  det: DetRuntimeOverrides;
+  pipeline: { scoreThresh: number };
+}
+
+export interface OcrRuntimeParamsInput {
+  text_det_limit_side_len?: number;
+  textDetLimitSideLen?: number;
+  text_det_limit_type?: LimitType;
+  textDetLimitType?: LimitType;
+  text_det_max_side_limit?: number;
+  textDetMaxSideLimit?: number;
+  text_det_thresh?: number;
+  textDetThresh?: number;
+  text_det_box_thresh?: number;
+  textDetBoxThresh?: number;
+  text_det_unclip_ratio?: number;
+  textDetUnclipRatio?: number;
+  text_rec_score_thresh?: number;
+  textRecScoreThresh?: number;
+}
+
+function firstDefined<T>(...values: Array<T | undefined | null>): T | undefined {
+  for (const value of values) {
+    if (value !== undefined && value !== null) {
+      return value;
+    }
+  }
+  return undefined;
+}
+
+function toNumberOrUndefined(value: unknown): number | undefined {
+  if (value === undefined || value === null) return undefined;
+  const num = Number(value);
+  return Number.isFinite(num) ? num : undefined;
+}
+
+export function getOcrRuntimeParams(
+  config: OcrModelConfig,
+  defaults: Partial<OcrRuntimeParamsInput> = {},
+  params: OcrRuntimeParamsInput = {}
+): ResolvedOcrParams {
+  return {
+    det: {
+      limitSideLen: toNumberOrUndefined(
+        firstDefined(
+          params.text_det_limit_side_len,
+          params.textDetLimitSideLen,
+          defaults.text_det_limit_side_len,
+          defaults.textDetLimitSideLen,
+          config.det.resizeLong
+        )
+      ),
+      limitType: firstDefined(
+        params.text_det_limit_type,
+        params.textDetLimitType,
+        defaults.text_det_limit_type,
+        defaults.textDetLimitType,
+        config.det.limitType
+      ),
+      maxSideLimit: toNumberOrUndefined(
+        firstDefined(
+          params.text_det_max_side_limit,
+          params.textDetMaxSideLimit,
+          defaults.text_det_max_side_limit,
+          defaults.textDetMaxSideLimit,
+          config.det.maxSideLimit
+        )
+      ),
+      thresh: toNumberOrUndefined(
+        firstDefined(
+          params.text_det_thresh,
+          params.textDetThresh,
+          defaults.text_det_thresh,
+          defaults.textDetThresh,
+          config.det.postprocess.thresh
+        )
+      ),
+      boxThresh: toNumberOrUndefined(
+        firstDefined(
+          params.text_det_box_thresh,
+          params.textDetBoxThresh,
+          defaults.text_det_box_thresh,
+          defaults.textDetBoxThresh,
+          config.det.postprocess.boxThresh
+        )
+      ),
+      unclipRatio: toNumberOrUndefined(
+        firstDefined(
+          params.text_det_unclip_ratio,
+          params.textDetUnclipRatio,
+          defaults.text_det_unclip_ratio,
+          defaults.textDetUnclipRatio,
+          config.det.postprocess.unclipRatio
+        )
+      )
+    },
+    pipeline: {
+      scoreThresh: Number(
+        firstDefined(
+          params.text_rec_score_thresh,
+          params.textRecScoreThresh,
+          defaults.text_rec_score_thresh,
+          defaults.textRecScoreThresh,
+          0
+        )
+      )
+    }
+  };
+}
diff --git a/paddleocr-js/packages/core/src/pipelines/ocr/shared.ts b/paddleocr-js/packages/core/src/pipelines/ocr/shared.ts
new file mode 100644
index 0000000..d83dd96
--- /dev/null
+++ b/paddleocr-js/packages/core/src/pipelines/ocr/shared.ts
@@ -0,0 +1,637 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { ModelAsset } from "../../resources/model-asset";
+import { DEFAULT_MODEL_ASSETS } from "../../resources/model-asset";
+import { DEFAULT_DET_MODEL_CONFIG } from "../../models/det";
+import { DEFAULT_REC_MODEL_CONFIG } from "../../models/rec";
+import { extractInferenceModelName } from "../../models/common";
+import { deepClone } from "../../utils/common";
+import type {
+  NormalizedPipelineConfig,
+  PipelineModelSelection,
+  PipelineRuntimeDefaults
+} from "./config";
+import { normalizeOcrPipelineConfig, toFiniteNumber } from "./config";
+import type { LimitType } from "./runtime-params";
+import { DEFAULT_OCR_PIPELINE_CONFIG_TEXT } from "./default-config";
+import type { OcrModelConfig } from "./runtime-params";
+import type { OrtOptions } from "../../runtime/ort";
+
+export interface ResolvedOcrOptions {
+  pipelineConfig: NormalizedPipelineConfig;
+  ortOptions: NormalizedOrtOptions;
+}
+
+export type ResolvedBackend = "webgpu" | "wasm" | "auto";
+
+export interface NormalizedOrtOptions {
+  backend: ResolvedBackend;
+  wasmPaths?: string;
+  numThreads?: number;
+  simd?: boolean;
+  proxy?: boolean;
+}
+
+export interface WorkerResolvedOptions {
+  enabled: boolean;
+  createWorker: (() => Worker) | null;
+}
+
+export const DEFAULT_OCR_CONFIG: OcrModelConfig = {
+  det: DEFAULT_DET_MODEL_CONFIG,
+  rec: DEFAULT_REC_MODEL_CONFIG
+};
+
+interface ModelRole {
+  assetKey: string;
+  modelRole: string;
+  selectionKey: keyof PipelineModelSelection;
+  nameAliases: string[];
+  assetAliases: string[];
+  nameLabel: string;
+  assetLabel: string;
+  assetRequirementError: string;
+}
+
+const DEFAULT_NORMALIZED_PIPELINE_CONFIG = normalizeOcrPipelineConfig(
+  DEFAULT_OCR_PIPELINE_CONFIG_TEXT
+);
+const DEFAULT_MODEL_SELECTION: Readonly<PipelineModelSelection> = Object.freeze({
+  ...DEFAULT_NORMALIZED_PIPELINE_CONFIG.modelSelection
+});
+const DEFAULT_LANG_VERSION_MODEL_SELECTION: Readonly<PipelineModelSelection> = Object.freeze({
+  ...DEFAULT_MODEL_SELECTION
+});
+const PP_OCRV6_LANG_VERSION_MODEL_SELECTION: Readonly<PipelineModelSelection> = Object.freeze({
+  textDetectionModelName: "PP-OCRv6_small_det",
+  textRecognitionModelName: "PP-OCRv6_small_rec"
+});
+const _LATIN_LANGS = new Set([
+  "af",
+  "az",
+  "bs",
+  "cs",
+  "cy",
+  "da",
+  "de",
+  "es",
+  "et",
+  "fr",
+  "ga",
+  "hr",
+  "hu",
+  "id",
+  "is",
+  "it",
+  "ku",
+  "la",
+  "lt",
+  "lv",
+  "mi",
+  "ms",
+  "mt",
+  "nl",
+  "no",
+  "oc",
+  "pi",
+  "pl",
+  "pt",
+  "ro",
+  "rs_latin",
+  "sk",
+  "sl",
+  "sq",
+  "sv",
+  "sw",
+  "tl",
+  "tr",
+  "uz",
+  "vi",
+  "french",
+  "german",
+  "fi",
+  "eu",
+  "gl",
+  "lb",
+  "rm",
+  "ca",
+  "qu"
+]);
+const _PPOCRV6_UNSUPPORTED_LATIN_LANGS = new Set(["pi"]);
+const _PPOCRV6_LANGS = new Set([
+  "ch",
+  "chinese_cht",
+  "en",
+  "japan",
+  ...[..._LATIN_LANGS].filter((lang) => !_PPOCRV6_UNSUPPORTED_LATIN_LANGS.has(lang))
+]);
+
+function isPpOcrV6Lang(lang: string): boolean {
+  return _PPOCRV6_LANGS.has(lang);
+}
+const OCR_MODEL_ROLES: Readonly<ModelRole[]> = Object.freeze([
+  {
+    assetKey: "det",
+    modelRole: "TextDetection",
+    selectionKey: "textDetectionModelName",
+    nameAliases: ["text_detection_model_name", "textDetectionModelName"],
+    assetAliases: ["textDetectionModelAsset", "text_detection_model_dir", "textDetectionModelDir"],
+    nameLabel: "text detection model name",
+    assetLabel: "text detection model asset",
+    assetRequirementError: "text_detection_model_dir requires text_detection_model_name."
+  },
+  {
+    assetKey: "rec",
+    modelRole: "TextRecognition",
+    selectionKey: "textRecognitionModelName",
+    nameAliases: ["text_recognition_model_name", "textRecognitionModelName"],
+    assetAliases: [
+      "textRecognitionModelAsset",
+      "text_recognition_model_dir",
+      "textRecognitionModelDir"
+    ],
+    nameLabel: "text recognition model name",
+    assetLabel: "text recognition model asset",
+    assetRequirementError: "text_recognition_model_dir requires text_recognition_model_name."
+  }
+]);
+
+const SUPPORTED_LANG_VERSION_MODELS = new Map<string, Readonly<PipelineModelSelection>>([
+  ["ch::PP-OCRv5", DEFAULT_LANG_VERSION_MODEL_SELECTION],
+  ["chinese_cht::PP-OCRv5", DEFAULT_LANG_VERSION_MODEL_SELECTION],
+  ["en::PP-OCRv5", DEFAULT_LANG_VERSION_MODEL_SELECTION],
+  ["japan::PP-OCRv5", DEFAULT_LANG_VERSION_MODEL_SELECTION]
+]);
+
+function readAliasedOption(
+  options: Record<string, unknown>,
+  aliases: string[],
+  label: string
+): unknown {
+  let resolved: unknown;
+  let hasResolvedValue = false;
+
+  for (const alias of aliases) {
+    if (!(alias in options)) continue;
+    const value = options[alias];
+    if (!hasResolvedValue) {
+      resolved = value;
+      hasResolvedValue = true;
+      continue;
+    }
+    if (value !== resolved) {
+      throw new Error(`Conflicting values provided for ${label}: ${aliases.join(", ")}.`);
+    }
+  }
+
+  return hasResolvedValue ? resolved : undefined;
+}
+
+function isLimitType(value: unknown): value is LimitType {
+  return value === "min" || value === "max";
+}
+
+function overlayPipelineRuntimeDefaults(
+  base: PipelineRuntimeDefaults,
+  explicit: Partial<PipelineRuntimeDefaults>
+): PipelineRuntimeDefaults {
+  const next: Record<string, unknown> = { ...base };
+  for (const key of Object.keys(explicit) as Array<keyof PipelineRuntimeDefaults>) {
+    const value = explicit[key];
+    if (value === undefined) continue;
+    next[key as string] = value as unknown;
+  }
+  return next as PipelineRuntimeDefaults;
+}
+
+function readExplicitPipelineRuntimeDefaults(
+  options: Record<string, unknown>
+): Partial<PipelineRuntimeDefaults> {
+  const out: Partial<PipelineRuntimeDefaults> = {};
+
+  const limitSide = readAliasedOption(
+    options,
+    ["text_det_limit_side_len", "textDetLimitSideLen"],
+    "text_det_limit_side_len"
+  );
+  if (limitSide !== undefined) {
+    const n = toFiniteNumber(limitSide);
+    if (n !== undefined) out.text_det_limit_side_len = n;
+  }
+
+  const limitType = readAliasedOption(
+    options,
+    ["text_det_limit_type", "textDetLimitType"],
+    "text_det_limit_type"
+  );
+  if (limitType !== undefined && isLimitType(limitType)) {
+    out.text_det_limit_type = limitType;
+  }
+
+  const maxSide = readAliasedOption(
+    options,
+    ["text_det_max_side_limit", "textDetMaxSideLimit"],
+    "text_det_max_side_limit"
+  );
+  if (maxSide !== undefined) {
+    const n = toFiniteNumber(maxSide);
+    if (n !== undefined) out.text_det_max_side_limit = n;
+  }
+
+  const detThresh = readAliasedOption(
+    options,
+    ["text_det_thresh", "textDetThresh"],
+    "text_det_thresh"
+  );
+  if (detThresh !== undefined) {
+    const n = toFiniteNumber(detThresh);
+    if (n !== undefined) out.text_det_thresh = n;
+  }
+
+  const boxThresh = readAliasedOption(
+    options,
+    ["text_det_box_thresh", "textDetBoxThresh"],
+    "text_det_box_thresh"
+  );
+  if (boxThresh !== undefined) {
+    const n = toFiniteNumber(boxThresh);
+    if (n !== undefined) out.text_det_box_thresh = n;
+  }
+
+  const unclip = readAliasedOption(
+    options,
+    ["text_det_unclip_ratio", "textDetUnclipRatio"],
+    "text_det_unclip_ratio"
+  );
+  if (unclip !== undefined) {
+    const n = toFiniteNumber(unclip);
+    if (n !== undefined) out.text_det_unclip_ratio = n;
+  }
+
+  const recScore = readAliasedOption(
+    options,
+    ["text_rec_score_thresh", "textRecScoreThresh"],
+    "text_rec_score_thresh"
+  );
+  if (recScore !== undefined) {
+    const n = toFiniteNumber(recScore);
+    if (n !== undefined) out.text_rec_score_thresh = n;
+  }
+
+  return out;
+}
+
+function toBatchSizeOption(value: unknown): number | undefined {
+  const n = toFiniteNumber(value);
+  return n !== undefined && n >= 1 ? Math.floor(n) : undefined;
+}
+
+function readExplicitBatchSizes(options: Record<string, unknown>): {
+  det: number | undefined;
+  rec: number | undefined;
+  pipeline: number | undefined;
+} {
+  return {
+    det: toBatchSizeOption(
+      readAliasedOption(
+        options,
+        ["textDetectionBatchSize", "text_detection_batch_size"],
+        "textDetectionBatchSize"
+      )
+    ),
+    rec: toBatchSizeOption(
+      readAliasedOption(
+        options,
+        ["textRecognitionBatchSize", "text_recognition_batch_size"],
+        "textRecognitionBatchSize"
+      )
+    ),
+    pipeline: toBatchSizeOption(
+      readAliasedOption(
+        options,
+        ["pipelineBatchSize", "pipeline_batch_size", "batch_size"],
+        "pipelineBatchSize"
+      )
+    )
+  };
+}
+
+function mergeNormalizedPipelineConfigWithExplicit(
+  normalized: NormalizedPipelineConfig,
+  options: Record<string, unknown>
+): NormalizedPipelineConfig {
+  const explicitRuntime = readExplicitPipelineRuntimeDefaults(options);
+  const explicitBatch = readExplicitBatchSizes(options);
+  const merged = deepClone(normalized);
+  merged.runtimeDefaults = overlayPipelineRuntimeDefaults(merged.runtimeDefaults, explicitRuntime);
+  if (explicitBatch.det !== undefined) {
+    merged.textDetectionBatchSize = explicitBatch.det;
+  }
+  if (explicitBatch.rec !== undefined) {
+    merged.textRecognitionBatchSize = explicitBatch.rec;
+  }
+  if (explicitBatch.pipeline !== undefined) {
+    merged.pipelineBatchSize = explicitBatch.pipeline;
+  }
+  return merged;
+}
+
+function resolveWarningBehavior(value: unknown): "warn" | "ignore" | "error" {
+  if (value === "ignore" || value === "error") return value;
+  return "warn";
+}
+
+function emitPipelineWarnings(warnings: string[], behavior: "warn" | "ignore" | "error"): void {
+  if (!warnings.length || behavior === "ignore") return;
+  if (behavior === "error") {
+    throw new Error(warnings.join(" "));
+  }
+  for (const warning of warnings) {
+    console.warn(`[PaddleOCR.js] ${warning}`);
+  }
+}
+
+function resolveModelAssetByName(_modelRole: string, modelName: string): ModelAsset {
+  const asset = DEFAULT_MODEL_ASSETS[modelName];
+  // eslint-disable-next-line @typescript-eslint/no-unnecessary-condition -- runtime guard for missing Record key
+  if (!asset) {
+    throw new Error(`Unknown model asset "${modelName}".`);
+  }
+  return { url: asset.url };
+}
+
+function getSelectedModelName(
+  baseSelection: PipelineModelSelection | null,
+  configSelection: PipelineModelSelection | null,
+  explicitSelection: Record<string, string | null> | null,
+  selectionKey: keyof PipelineModelSelection
+): string | null {
+  return (
+    explicitSelection?.[selectionKey] ??
+    configSelection?.[selectionKey] ??
+    baseSelection?.[selectionKey] ??
+    null
+  );
+}
+
+function createResolvedModelSelection(
+  baseSelection: PipelineModelSelection | null,
+  configSelection: PipelineModelSelection | null,
+  explicitSelection: Record<string, string | null> | null
+): PipelineModelSelection {
+  return Object.fromEntries(
+    OCR_MODEL_ROLES.map((role) => [
+      role.selectionKey,
+      getSelectedModelName(baseSelection, configSelection, explicitSelection, role.selectionKey)
+    ])
+  ) as unknown as PipelineModelSelection;
+}
+
+export function validateLoadedModelName(
+  modelRole: string,
+  expectedModelName: string | null | undefined,
+  configText: string
+): void {
+  if (!expectedModelName) {
+    throw new Error(`${modelRole} model selection must define model_name.`);
+  }
+  const declaredModelName = extractInferenceModelName(configText);
+  if (!declaredModelName) {
+    throw new Error(`${modelRole} in inference.yml must define model_name.`);
+  }
+  if (declaredModelName !== expectedModelName) {
+    throw new Error(
+      `${modelRole} in inference.yml declares model_name "${declaredModelName}" but requested model_name is "${expectedModelName}".`
+    );
+  }
+}
+
+function resolveSelectedAsset(
+  assetRole: string,
+  modelRole: string,
+  selectionKey: keyof PipelineModelSelection,
+  baseSelection: PipelineModelSelection | null,
+  configSelection: PipelineModelSelection | null,
+  explicitSelection: Record<string, string | null> | null,
+  configAssets: Partial<Record<string, ModelAsset>> | null,
+  explicitAssets: Record<string, ModelAsset> | null
+): ModelAsset | null {
+  const explicitAsset = explicitAssets?.[assetRole];
+  if (explicitAsset) {
+    return explicitAsset;
+  }
+  const explicitModelName = explicitSelection?.[selectionKey];
+  if (explicitModelName) {
+    return resolveModelAssetByName(modelRole, explicitModelName);
+  }
+  const configAsset = configAssets?.[assetRole];
+  if (configAsset) {
+    return configAsset;
+  }
+  const configModelName = configSelection?.[selectionKey];
+  if (configModelName) {
+    return resolveModelAssetByName(modelRole, configModelName);
+  }
+  const baseModelName = baseSelection?.[selectionKey];
+  if (baseModelName) {
+    return resolveModelAssetByName(modelRole, baseModelName);
+  }
+  return null;
+}
+
+function createOcrAssets(
+  baseSelection: PipelineModelSelection | null,
+  configSelection: PipelineModelSelection | null,
+  explicitSelection: Record<string, string | null> | null,
+  configAssets: Partial<Record<string, ModelAsset>> | null,
+  explicitAssets: Record<string, ModelAsset> | null
+): Record<string, ModelAsset> {
+  const assets = Object.fromEntries(
+    OCR_MODEL_ROLES.map((role) => [
+      role.assetKey,
+      resolveSelectedAsset(
+        role.assetKey,
+        role.modelRole,
+        role.selectionKey,
+        baseSelection,
+        configSelection,
+        explicitSelection,
+        configAssets,
+        explicitAssets
+      )
+    ])
+  );
+
+  if (Object.values(assets).some((asset) => !asset)) {
+    throw new Error("OCR model selection must define both detection and recognition models.");
+  }
+
+  return assets as Record<string, ModelAsset>;
+}
+
+function getExplicitModelSelection(options: Record<string, unknown>): {
+  modelSelection: Record<string, string | null>;
+  assets: Record<string, ModelAsset>;
+} | null {
+  const modelSelection: Record<string, string | null> = {};
+  const assets: Record<string, ModelAsset> = {};
+  let hasAnyOption = false;
+
+  for (const role of OCR_MODEL_ROLES) {
+    const modelName = readAliasedOption(options, role.nameAliases, role.nameLabel) as
+      | string
+      | undefined;
+    const asset = readAliasedOption(options, role.assetAliases, role.assetLabel) as
+      | ModelAsset
+      | undefined;
+
+    if (modelName !== undefined) {
+      modelSelection[role.selectionKey] = modelName;
+      hasAnyOption = true;
+    }
+    if (asset !== undefined) {
+      if (modelName === undefined) {
+        throw new Error(role.assetRequirementError);
+      }
+      assets[role.assetKey] = asset;
+      hasAnyOption = true;
+    }
+  }
+
+  if (!hasAnyOption) {
+    return null;
+  }
+
+  return {
+    modelSelection,
+    assets
+  };
+}
+
+function resolveBaseModelSelection(
+  options: Record<string, unknown>,
+  includeDefaultBase = false
+): Readonly<PipelineModelSelection> | null {
+  const ocrVersion = readAliasedOption(options, ["ocrVersion", "ocr_version"], "ocrVersion") as
+    | string
+    | undefined;
+  if (!options.lang && !ocrVersion) {
+    return includeDefaultBase ? DEFAULT_MODEL_SELECTION : null;
+  }
+
+  const lang = (options.lang as string) || "ch";
+  const resolvedOcrVersion = ocrVersion || "PP-OCRv5";
+
+  if (resolvedOcrVersion === "PP-OCRv6") {
+    if (!isPpOcrV6Lang(lang)) {
+      throw new Error(
+        `Unsupported lang/ocrVersion combination: lang="${lang}", ocrVersion="${resolvedOcrVersion}".`
+      );
+    }
+    return PP_OCRV6_LANG_VERSION_MODEL_SELECTION;
+  }
+
+  const modelSelection = SUPPORTED_LANG_VERSION_MODELS.get(`${lang}::${resolvedOcrVersion}`);
+
+  if (!modelSelection) {
+    throw new Error(
+      `Unsupported lang/ocrVersion combination: lang="${lang}", ocrVersion="${resolvedOcrVersion}".`
+    );
+  }
+  return modelSelection;
+}
+
+function resolveConstructionOptions(
+  options: Record<string, unknown> = {}
+): NormalizedPipelineConfig {
+  const pipelineInput = options.pipelineConfig;
+  const userPipelineConfig =
+    pipelineInput != null ? normalizeOcrPipelineConfig(pipelineInput) : null;
+  const warningBehavior = resolveWarningBehavior(options.unsupportedBehavior);
+  const warnings = userPipelineConfig?.warnings || [];
+  const baseSelection = resolveBaseModelSelection(options, !userPipelineConfig);
+  const configSelection = userPipelineConfig?.modelSelection || null;
+  const configAssets = userPipelineConfig?.assets || null;
+  const explicitOptions = getExplicitModelSelection(options);
+  const explicitSelection = explicitOptions?.modelSelection || null;
+  const explicitAssets = explicitOptions?.assets || null;
+  const resolvedModelSelection = createResolvedModelSelection(
+    baseSelection,
+    configSelection,
+    explicitSelection
+  );
+  const assets = createOcrAssets(
+    baseSelection,
+    configSelection,
+    explicitSelection,
+    configAssets,
+    explicitAssets
+  );
+
+  const baseNormalized = userPipelineConfig ?? DEFAULT_NORMALIZED_PIPELINE_CONFIG;
+  if (userPipelineConfig) {
+    emitPipelineWarnings(warnings, warningBehavior);
+  }
+  const merged = mergeNormalizedPipelineConfigWithExplicit(baseNormalized, options);
+  merged.modelSelection = resolvedModelSelection;
+  merged.assets = { ...assets };
+  return merged;
+}
+
+function resolveBackend(raw: string | undefined): ResolvedBackend {
+  if (raw === "webgpu" || raw === "wasm") return raw;
+  return "auto";
+}
+
+export function normalizeOrtOptions(ortOptions: OrtOptions = {}): NormalizedOrtOptions {
+  const backend = resolveBackend(ortOptions.backend);
+
+  return {
+    backend,
+    ...(ortOptions.wasmPaths !== undefined ? { wasmPaths: ortOptions.wasmPaths } : {}),
+    ...(ortOptions.numThreads !== undefined ? { numThreads: ortOptions.numThreads } : {}),
+    ...(ortOptions.simd !== undefined ? { simd: ortOptions.simd } : {}),
+    ...(ortOptions.proxy !== undefined ? { proxy: ortOptions.proxy } : {})
+  };
+}
+
+export function resolveWorkerOptions(workerOption: unknown): WorkerResolvedOptions {
+  if (!workerOption) {
+    return {
+      enabled: false,
+      createWorker: null
+    };
+  }
+
+  if (workerOption === true) {
+    return {
+      enabled: true,
+      createWorker: null
+    };
+  }
+
+  if (typeof workerOption === "object") {
+    const opts = workerOption as Record<string, unknown>;
+    return {
+      enabled: true,
+      createWorker:
+        typeof opts.createWorker === "function" ? (opts.createWorker as () => Worker) : null
+    };
+  }
+
+  throw new Error("worker must be a boolean or an options object.");
+}
+
+export function resolvePaddleOCROptions(options: Record<string, unknown> = {}): ResolvedOcrOptions {
+  return {
+    pipelineConfig: resolveConstructionOptions(options),
+    ortOptions: normalizeOrtOptions((options.ortOptions || {}) as OrtOptions)
+  };
+}
+
+export function cloneDefaultOcrConfig(): OcrModelConfig {
+  return deepClone(DEFAULT_OCR_CONFIG);
+}
diff --git a/paddleocr-js/packages/core/src/pipelines/ocr/worker-backed.ts b/paddleocr-js/packages/core/src/pipelines/ocr/worker-backed.ts
new file mode 100644
index 0000000..4750695
--- /dev/null
+++ b/paddleocr-js/packages/core/src/pipelines/ocr/worker-backed.ts
@@ -0,0 +1,148 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import { sourceToWorkerPayload } from "../../platform/browser";
+import { createWorkerTransportClient } from "../../worker/client";
+import type { WorkerTransportClient, WorkerOptions } from "../../worker/client";
+import type { OcrModelConfig, OcrRuntimeParamsInput } from "./runtime-params";
+import type { InitializationSummary, OcrResult, OcrPipelineRunnerOptions } from "./core";
+import { cloneDefaultOcrConfig } from "./shared";
+
+declare const __ORT_WASM_CDN_PREFIX__: string | undefined;
+
+function createDefaultWorker(): Worker {
+  if (typeof Worker !== "function") {
+    throw new Error("worker mode requires Web Worker support in this environment.");
+  }
+  return new Worker(new URL("./worker-entry.ts", import.meta.url), {
+    type: "module"
+  });
+}
+
+export class WorkerBackedPaddleOCR {
+  private options: OcrPipelineRunnerOptions;
+  private lastInitializationSummary: InitializationSummary | null;
+  private modelConfig: OcrModelConfig;
+  private transportClient: WorkerTransportClient;
+  private initPromise: Promise<InitializationSummary> | null;
+  private disposed: boolean;
+
+  constructor(options: OcrPipelineRunnerOptions, transportClient: WorkerTransportClient) {
+    this.options = options;
+    this.lastInitializationSummary = null;
+    this.modelConfig = cloneDefaultOcrConfig();
+    this.transportClient = transportClient;
+    this.initPromise = null;
+    this.disposed = false;
+  }
+
+  ensureActive(): void {
+    if (this.disposed) {
+      throw new Error("PaddleOCR worker instance has been disposed.");
+    }
+  }
+
+  async initialize(): Promise<InitializationSummary> {
+    this.ensureActive();
+    if (this.lastInitializationSummary) {
+      return this.lastInitializationSummary;
+    }
+    if (!this.initPromise) {
+      const ortOpts = (this.options.ortOptions || {}) as Record<string, unknown>;
+      if (ortOpts["wasmPaths"] === undefined && typeof __ORT_WASM_CDN_PREFIX__ === "string") {
+        console.warn(
+          "[PaddleOCR.js] Worker mode: ortOptions.wasmPaths is not set — falling back to CDN (%s). " +
+            "For version consistency between main thread and worker, set ortOptions.wasmPaths " +
+            "to the path where your bundler outputs the onnxruntime-web WASM files " +
+            '(e.g. ortOptions: { wasmPaths: "/assets/" }).',
+          __ORT_WASM_CDN_PREFIX__
+        );
+      }
+      const wasmCdnFallback =
+        ortOpts["wasmPaths"] === undefined && typeof __ORT_WASM_CDN_PREFIX__ === "string"
+          ? { wasmPaths: __ORT_WASM_CDN_PREFIX__ }
+          : {};
+      this.initPromise = this.transportClient
+        .request("init", {
+          options: {
+            ...this.options,
+            ortOptions: {
+              ...ortOpts,
+              ...wasmCdnFallback,
+              disableWasmProxy: true
+            }
+          }
+        })
+        .then((rawPayload) => {
+          const payload = rawPayload as {
+            summary: InitializationSummary;
+            modelConfig: OcrModelConfig;
+          };
+          this.lastInitializationSummary = payload.summary;
+          this.modelConfig = payload.modelConfig;
+          return this.lastInitializationSummary;
+        })
+        .catch((error: unknown) => {
+          this.initPromise = null;
+          this.transportClient.dispose();
+          throw error;
+        });
+    }
+    return this.initPromise;
+  }
+
+  getInitializationSummary(): InitializationSummary | null {
+    return this.lastInitializationSummary;
+  }
+
+  getModelConfig(): OcrModelConfig {
+    return this.modelConfig;
+  }
+
+  async predict(input: unknown, params: OcrRuntimeParamsInput = {}): Promise<OcrResult[]> {
+    this.ensureActive();
+    await this.initialize();
+    const sources: unknown[] = Array.isArray(input) ? input : [input];
+    const payloads: Array<{ payload: unknown; transferables: Transferable[] }> = await Promise.all(
+      sources.map((source) =>
+        sourceToWorkerPayload(source as Parameters<typeof sourceToWorkerPayload>[0])
+      )
+    );
+    const combinedPayloads = payloads.map((p) => p.payload);
+    const combinedTransferables = payloads.flatMap((p) => p.transferables);
+    return this.transportClient.request(
+      "predict",
+      {
+        sources: combinedPayloads,
+        params
+      },
+      combinedTransferables
+    ) as Promise<OcrResult[]>;
+  }
+
+  async dispose(): Promise<void> {
+    if (this.disposed) {
+      return;
+    }
+    this.disposed = true;
+    try {
+      await this.transportClient.request("dispose", {});
+    } catch {
+      // Transport disposal is authoritative even if the worker cannot respond.
+    }
+    this.transportClient.dispose();
+  }
+}
+
+export function createWorkerBackedPaddleOCR(
+  options: OcrPipelineRunnerOptions,
+  workerOptions: WorkerOptions = {}
+): WorkerBackedPaddleOCR {
+  const transportClient = createWorkerTransportClient({
+    ...workerOptions,
+    createWorker: workerOptions.createWorker || createDefaultWorker
+  });
+  return new WorkerBackedPaddleOCR(options, transportClient);
+}
diff --git a/paddleocr-js/packages/core/src/pipelines/ocr/worker-entry.ts b/paddleocr-js/packages/core/src/pipelines/ocr/worker-entry.ts
new file mode 100644
index 0000000..f92c8a7
--- /dev/null
+++ b/paddleocr-js/packages/core/src/pipelines/ocr/worker-entry.ts
@@ -0,0 +1,57 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import { attachWorkerMessageHandler } from "../../worker/entry";
+import { sourcePayloadToMat, ensureServedFromHttp } from "../../platform/worker";
+import type { OcrPipelineRunnerOptions } from "./core";
+import { OcrPipelineRunner } from "./core";
+import type { OcrRuntimeParamsInput } from "./runtime-params";
+
+function createPaddleOCRWorkerMessageHandler() {
+  let ocr: OcrPipelineRunner | null = null;
+
+  async function handleInit(payload: Record<string, unknown>) {
+    await ocr?.dispose();
+    ocr = new OcrPipelineRunner({
+      ...(payload.options as OcrPipelineRunnerOptions),
+      ensureServedFromHttp,
+      sourceToMat: sourcePayloadToMat
+    });
+    const summary = await ocr.initialize();
+    return {
+      summary,
+      modelConfig: ocr.getModelConfig()
+    };
+  }
+
+  async function handlePredict(payload: Record<string, unknown>) {
+    if (!ocr) {
+      throw new Error("OCR worker is not initialized.");
+    }
+    const sources = payload.sources;
+    return ocr.predict(sources, (payload.params || {}) as OcrRuntimeParamsInput);
+  }
+
+  async function handleDispose() {
+    await ocr?.dispose();
+    ocr = null;
+    return {};
+  }
+
+  return async function handleMessage(type: string, payload: Record<string, unknown>) {
+    switch (type) {
+      case "init":
+        return handleInit(payload);
+      case "predict":
+        return handlePredict(payload);
+      case "dispose":
+        return handleDispose();
+      default:
+        throw new Error(`Unsupported worker request type "${type}".`);
+    }
+  };
+}
+
+attachWorkerMessageHandler(createPaddleOCRWorkerMessageHandler());
diff --git a/paddleocr-js/packages/core/src/platform/browser.ts b/paddleocr-js/packages/core/src/platform/browser.ts
new file mode 100644
index 0000000..6a37380
--- /dev/null
+++ b/paddleocr-js/packages/core/src/platform/browser.ts
@@ -0,0 +1,119 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { OpenCv, Mat } from "@techstark/opencv-js";
+
+export type ImageSource = ImageBitmap | Blob | HTMLCanvasElement | ImageData | HTMLImageElement;
+
+export interface SourceMatResult {
+  width: number;
+  height: number;
+  mat: Mat;
+  dispose(): void;
+}
+
+export interface WorkerPayload {
+  kind: "imageBitmap";
+  imageBitmap: ImageBitmap;
+}
+
+export interface WorkerPayloadResult {
+  payload: WorkerPayload;
+  transferables: Transferable[];
+}
+
+export function ensureServedFromHttp(): void {
+  if (globalThis.location.protocol === "file:") {
+    throw new Error("PaddleOCR.js requires an HTTP(S) origin so model assets can be fetched.");
+  }
+}
+
+function hasDomConstructor(name: string): boolean {
+  return typeof (globalThis as Record<string, unknown>)[name] !== "undefined";
+}
+
+export async function sourceToImageBitmap(source: ImageSource): Promise<ImageBitmap> {
+  if (typeof ImageBitmap !== "undefined" && source instanceof ImageBitmap) return source;
+  if (source instanceof Blob) return createImageBitmap(source);
+  if (hasDomConstructor("HTMLCanvasElement") && source instanceof HTMLCanvasElement) {
+    return createImageBitmap(source);
+  }
+  if (source instanceof ImageData) {
+    const canvas = document.createElement("canvas");
+    canvas.width = source.width;
+    canvas.height = source.height;
+    const ctx = canvas.getContext("2d");
+    if (!ctx) throw new Error("Failed to create a 2D canvas context.");
+    ctx.putImageData(source, 0, 0);
+    return createImageBitmap(canvas);
+  }
+  if (hasDomConstructor("HTMLImageElement") && source instanceof HTMLImageElement) {
+    return createImageBitmap(source);
+  }
+  throw new Error("Unsupported image source. Use a Blob, ImageBitmap, ImageData, canvas, or img.");
+}
+
+async function sourceToClonedImageBitmap(source: ImageSource): Promise<ImageBitmap> {
+  if (typeof ImageBitmap !== "undefined" && source instanceof ImageBitmap) {
+    return createImageBitmap(source);
+  }
+  return sourceToImageBitmap(source);
+}
+
+export function bitmapToSourceMat(
+  cv: OpenCv,
+  imageBitmap: ImageBitmap
+): { canvas: HTMLCanvasElement; mat: Mat } {
+  const canvas = document.createElement("canvas");
+  canvas.width = imageBitmap.width;
+  canvas.height = imageBitmap.height;
+  const ctx = canvas.getContext("2d", { willReadFrequently: true });
+  if (!ctx) throw new Error("Failed to create a 2D canvas context.");
+  ctx.drawImage(imageBitmap, 0, 0);
+  return {
+    canvas,
+    mat: cv.imread(canvas)
+  };
+}
+
+export async function sourceToMat(cv: OpenCv, source: unknown): Promise<SourceMatResult> {
+  if (typeof cv.Mat === "function" && source instanceof cv.Mat) {
+    const cloned = source.clone();
+    return {
+      width: source.cols,
+      height: source.rows,
+      mat: cloned,
+      dispose() {
+        cloned.delete();
+      }
+    };
+  }
+
+  const imageBitmap = await sourceToImageBitmap(source as ImageSource);
+  const sourceImage = bitmapToSourceMat(cv, imageBitmap);
+  return {
+    width: imageBitmap.width,
+    height: imageBitmap.height,
+    mat: sourceImage.mat,
+    dispose() {
+      sourceImage.mat.delete();
+      imageBitmap.close();
+    }
+  };
+}
+
+export async function sourceToWorkerPayload(source: ImageSource): Promise<WorkerPayloadResult> {
+  if (typeof ImageBitmap === "undefined" || typeof createImageBitmap !== "function") {
+    throw new Error("Worker mode requires ImageBitmap support in this browser.");
+  }
+  const imageBitmap = await sourceToClonedImageBitmap(source);
+  return {
+    payload: {
+      kind: "imageBitmap",
+      imageBitmap
+    },
+    transferables: [imageBitmap]
+  };
+}
diff --git a/paddleocr-js/packages/core/src/platform/worker.ts b/paddleocr-js/packages/core/src/platform/worker.ts
new file mode 100644
index 0000000..6c87881
--- /dev/null
+++ b/paddleocr-js/packages/core/src/platform/worker.ts
@@ -0,0 +1,72 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { OpenCv, Mat } from "@techstark/opencv-js";
+import type { SourceMatResult } from "./browser";
+import { ensureServedFromHttp } from "./browser";
+
+export interface WorkerSourcePayload {
+  kind: "imageBitmap";
+  imageBitmap: ImageBitmap;
+}
+
+function imageBitmapToImageData(imageBitmap: ImageBitmap): ImageData {
+  if (typeof OffscreenCanvas !== "function") {
+    throw new Error("Worker mode requires OffscreenCanvas support in this browser.");
+  }
+  const canvas = new OffscreenCanvas(imageBitmap.width, imageBitmap.height);
+  const ctx = canvas.getContext("2d", { willReadFrequently: true });
+  if (!ctx) {
+    throw new Error("Failed to create a 2D canvas context in the OCR worker.");
+  }
+  ctx.drawImage(imageBitmap, 0, 0);
+  return ctx.getImageData(0, 0, imageBitmap.width, imageBitmap.height);
+}
+
+function imageDataToMat(cv: OpenCv, imageData: ImageData): Mat {
+  return cv.matFromArray(imageData.height, imageData.width, cv.CV_8UC4, imageData.data);
+}
+
+function isWorkerSourcePayload(source: unknown): source is WorkerSourcePayload {
+  if (typeof source !== "object" || source === null) return false;
+  const candidate = source as Record<string, unknown>;
+  return (
+    candidate["kind"] === "imageBitmap" &&
+    typeof ImageBitmap !== "undefined" &&
+    candidate["imageBitmap"] instanceof ImageBitmap
+  );
+}
+
+export function sourcePayloadToMat(cv: OpenCv, source: unknown): SourceMatResult {
+  if (typeof cv.Mat === "function" && source instanceof cv.Mat) {
+    const cloned = source.clone();
+    return {
+      width: source.cols,
+      height: source.rows,
+      mat: cloned,
+      dispose() {
+        cloned.delete();
+      }
+    };
+  }
+
+  if (isWorkerSourcePayload(source)) {
+    const imageData = imageBitmapToImageData(source.imageBitmap);
+    const mat = imageDataToMat(cv, imageData);
+    return {
+      width: imageData.width,
+      height: imageData.height,
+      mat,
+      dispose() {
+        mat.delete();
+        source.imageBitmap.close();
+      }
+    };
+  }
+
+  throw new Error("Unsupported worker image source payload.");
+}
+
+export { ensureServedFromHttp };
diff --git a/paddleocr-js/packages/core/src/resources/index.ts b/paddleocr-js/packages/core/src/resources/index.ts
new file mode 100644
index 0000000..f98d5d0
--- /dev/null
+++ b/paddleocr-js/packages/core/src/resources/index.ts
@@ -0,0 +1,17 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+export type { ModelAsset, ModelAssetsMap, ModelLoadResult, ModelLoadSummary } from "./model-asset";
+export {
+  DEFAULT_MODEL_ASSETS,
+  MODEL_ENTRY_PATHS,
+  assertModelResourceSlot,
+  assertModelResources,
+  getModelEntryPath,
+  loadModelAsset,
+  normalizeAssets,
+  normalizeModelAsset
+} from "./model-asset";
+export { extractTarEntries, pickTarEntry } from "./tar";
diff --git a/paddleocr-js/packages/core/src/resources/model-asset.ts b/paddleocr-js/packages/core/src/resources/model-asset.ts
new file mode 100644
index 0000000..765cfc1
--- /dev/null
+++ b/paddleocr-js/packages/core/src/resources/model-asset.ts
@@ -0,0 +1,150 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+export interface ModelAsset {
+  url: string;
+}
+
+export type ModelAssetsMap = Record<string, ModelAsset>;
+
+export const DEFAULT_MODEL_ASSETS: ModelAssetsMap = {
+  "PP-OCRv5_mobile_det": {
+    url: "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_det_onnx_infer.tar"
+  },
+  "PP-OCRv5_mobile_rec": {
+    url: "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv5_mobile_rec_onnx_infer.tar"
+  },
+  "PP-OCRv6_small_det": {
+    url: "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_det_onnx_infer.tar"
+  },
+  "PP-OCRv6_small_rec": {
+    url: "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_small_rec_onnx_infer.tar"
+  },
+  "PP-OCRv6_tiny_det": {
+    url: "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_det_onnx_infer.tar"
+  },
+  "PP-OCRv6_tiny_rec": {
+    url: "https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv6_tiny_rec_onnx_infer.tar"
+  }
+};
+
+export const MODEL_ENTRY_PATHS: Readonly<Record<string, string>> = Object.freeze({
+  model: "inference.onnx",
+  config: "inference.yml"
+});
+
+export interface ModelLoadResult {
+  modelBytes: Uint8Array;
+  configText: string;
+  download: ModelLoadSummary;
+}
+
+export interface ModelLoadSummary {
+  url: string;
+  bytes: number;
+}
+
+// --- Validation helpers ---
+
+function isNonEmptyString(value: unknown): value is string {
+  return typeof value === "string" && value.length > 0;
+}
+
+function isObject(value: unknown): value is Record<string, unknown> {
+  return Boolean(value && typeof value === "object" && !Array.isArray(value));
+}
+
+export function normalizeModelAsset(assetName: string, asset: unknown): ModelAsset {
+  if (isNonEmptyString(asset)) {
+    const resolvedAsset = DEFAULT_MODEL_ASSETS[asset];
+    // eslint-disable-next-line @typescript-eslint/no-unnecessary-condition -- runtime guard for missing Record key
+    if (!resolvedAsset) {
+      throw new Error(`Asset "${assetName}" references unknown model asset "${asset}".`);
+    }
+    return { url: resolvedAsset.url };
+  }
+
+  if (!isObject(asset)) {
+    throw new Error(`Asset "${assetName}" must be an object.`);
+  }
+  if (!isNonEmptyString(asset.url)) {
+    throw new Error(`Asset "${assetName}" must define url.`);
+  }
+
+  return {
+    url: asset.url
+  };
+}
+
+export function normalizeAssets(
+  assets: Record<string, unknown> | undefined
+): Record<string, ModelAsset> {
+  const assetEntries = Object.entries(assets || {});
+
+  if (assetEntries.length === 0) {
+    throw new Error("Assets must define at least one asset.");
+  }
+
+  return Object.fromEntries(
+    assetEntries.map(([assetName, asset]) => [assetName, normalizeModelAsset(assetName, asset)])
+  );
+}
+
+// --- Model loading ---
+
+export function getModelEntryPath(slot: string): string | null {
+  return MODEL_ENTRY_PATHS[slot] || null;
+}
+
+export function assertModelResourceSlot(kind: string, slot: string, value: unknown): void {
+  if (slot === "model") {
+    if (!(value instanceof Uint8Array) || value.byteLength === 0) {
+      throw new Error(`${kind} model requires a non-empty ${MODEL_ENTRY_PATHS.model} resource.`);
+    }
+    return;
+  }
+
+  if (slot === "config") {
+    if (typeof value !== "string" || value.trim().length === 0) {
+      throw new Error(`${kind} model requires a non-empty ${MODEL_ENTRY_PATHS.config} resource.`);
+    }
+    return;
+  }
+
+  throw new Error(`Unsupported model resource slot "${slot}".`);
+}
+
+export function assertModelResources(kind: string, resources: Record<string, unknown>): void {
+  for (const [slot, value] of Object.entries(resources)) {
+    assertModelResourceSlot(kind, slot, value);
+  }
+}
+
+// --- Model loading (fetch + tar extraction) ---
+
+import { extractTarEntries, pickTarEntry } from "./tar";
+
+export async function loadModelAsset(
+  asset: ModelAsset,
+  fetchImpl: typeof fetch = fetch
+): Promise<ModelLoadResult> {
+  const response = await fetchImpl(asset.url);
+  if (!response.ok) {
+    throw new Error(`Failed to download ${asset.url}: HTTP ${String(response.status)}`);
+  }
+  const buffer = await response.arrayBuffer();
+  const entries = extractTarEntries(buffer);
+  const modelBytes = pickTarEntry(entries, MODEL_ENTRY_PATHS.model);
+  const configBytes = pickTarEntry(entries, MODEL_ENTRY_PATHS.config);
+
+  return {
+    modelBytes,
+    configText: new TextDecoder().decode(configBytes),
+    download: {
+      url: asset.url,
+      bytes: buffer.byteLength
+    }
+  };
+}
diff --git a/paddleocr-js/packages/core/src/resources/tar.ts b/paddleocr-js/packages/core/src/resources/tar.ts
new file mode 100644
index 0000000..90946c8
--- /dev/null
+++ b/paddleocr-js/packages/core/src/resources/tar.ts
@@ -0,0 +1,80 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+function readString(bytes: Uint8Array, start: number, length: number): string {
+  let output = "";
+  for (let index = start; index < start + length; index += 1) {
+    const value = bytes[index];
+    if (value === 0) break;
+    output += String.fromCharCode(value);
+  }
+  return output.replace(/\0.*$/, "").trim();
+}
+
+function readOctal(bytes: Uint8Array, start: number, length: number): number {
+  const raw = readString(bytes, start, length).replace(/\0/g, "").trim();
+  return raw ? Number.parseInt(raw, 8) : 0;
+}
+
+function isEmptyBlock(bytes: Uint8Array, offset: number): boolean {
+  for (let index = offset; index < offset + 512; index += 1) {
+    if (bytes[index] !== 0) return false;
+  }
+  return true;
+}
+
+function normalizeEntryName(name: string): string {
+  return name.replace(/^\.?\//, "");
+}
+
+function isMetadataEntry(name: string): boolean {
+  const segments = normalizeEntryName(name).split("/");
+  const baseName = segments[segments.length - 1] || "";
+  return (
+    baseName.startsWith("._") || segments.includes("PaxHeader") || segments.includes("__MACOSX")
+  );
+}
+
+export function extractTarEntries(buffer: ArrayBuffer | Uint8Array): Map<string, Uint8Array> {
+  const bytes = buffer instanceof Uint8Array ? buffer : new Uint8Array(buffer);
+  const entries = new Map<string, Uint8Array>();
+  let offset = 0;
+
+  while (offset + 512 <= bytes.length) {
+    if (isEmptyBlock(bytes, offset)) {
+      break;
+    }
+
+    const name = normalizeEntryName(readString(bytes, offset, 100));
+    const size = readOctal(bytes, offset + 124, 12);
+    const type = bytes[offset + 156];
+    const dataStart = offset + 512;
+    const dataEnd = dataStart + size;
+
+    if (type !== 53 && type !== 120 && name && !isMetadataEntry(name)) {
+      entries.set(name, bytes.slice(dataStart, dataEnd));
+    }
+
+    offset = dataStart + Math.ceil(size / 512) * 512;
+  }
+
+  return entries;
+}
+
+export function pickTarEntry(entries: Map<string, Uint8Array>, targetName: string): Uint8Array {
+  const normalizedTarget = normalizeEntryName(targetName);
+  const entry = entries.get(normalizedTarget);
+  if (entry) {
+    return entry;
+  }
+
+  for (const [name, value] of entries) {
+    if (name.endsWith(`/${normalizedTarget}`) || name === normalizedTarget) {
+      return value;
+    }
+  }
+
+  throw new Error(`Entry "${targetName}" was not found in the tar archive.`);
+}
diff --git a/paddleocr-js/packages/core/src/runtime/index.ts b/paddleocr-js/packages/core/src/runtime/index.ts
new file mode 100644
index 0000000..6e1abe8
--- /dev/null
+++ b/paddleocr-js/packages/core/src/runtime/index.ts
@@ -0,0 +1,6 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+export { initOpenCvRuntime } from "./opencv";
diff --git a/paddleocr-js/packages/core/src/runtime/opencv.ts b/paddleocr-js/packages/core/src/runtime/opencv.ts
new file mode 100644
index 0000000..afe39c9
--- /dev/null
+++ b/paddleocr-js/packages/core/src/runtime/opencv.ts
@@ -0,0 +1,39 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { OpenCv } from "@techstark/opencv-js";
+import cvModule from "@techstark/opencv-js";
+
+let cachedCvPromise: Promise<{ cv: OpenCv }> | null = null;
+
+async function getOpenCv(): Promise<{ cv: OpenCv }> {
+  let cv: OpenCv;
+  if (cvModule instanceof Promise) {
+    cv = await cvModule;
+  } else {
+    const mod = cvModule as { Mat?: unknown; onRuntimeInitialized?: () => void };
+    if (mod.Mat) {
+      cv = cvModule as OpenCv;
+    } else {
+      await new Promise<void>((resolve) => {
+        mod.onRuntimeInitialized = () => {
+          resolve();
+        };
+      });
+      cv = cvModule as OpenCv;
+    }
+  }
+  return { cv };
+}
+
+export async function initOpenCvRuntime(): Promise<{ cv: OpenCv }> {
+  if (!cachedCvPromise) {
+    cachedCvPromise = getOpenCv().catch((error: unknown) => {
+      cachedCvPromise = null;
+      throw error;
+    });
+  }
+  return cachedCvPromise;
+}
diff --git a/paddleocr-js/packages/core/src/runtime/ort.ts b/paddleocr-js/packages/core/src/runtime/ort.ts
new file mode 100644
index 0000000..9412557
--- /dev/null
+++ b/paddleocr-js/packages/core/src/runtime/ort.ts
@@ -0,0 +1,158 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+export type OrtModule = typeof import("onnxruntime-web");
+
+export interface WebGpuState {
+  available: boolean;
+  reason: string;
+}
+
+export interface OrtOptions {
+  backend?: "webgpu" | "wasm" | "auto" | (string & {});
+  wasmPaths?: string;
+  numThreads?: number;
+  simd?: boolean;
+  proxy?: boolean;
+  disableWasmProxy?: boolean;
+}
+
+export interface OrtRuntimeResult {
+  ort: OrtModule;
+  webgpuState: WebGpuState;
+  backend: string;
+}
+
+export interface SessionState {
+  session: import("onnxruntime-web").InferenceSession;
+  provider: string;
+}
+
+let ortModulePromise: Promise<OrtModule> | null = null;
+
+async function loadOrtModule(): Promise<OrtModule> {
+  if (ortModulePromise) {
+    return ortModulePromise;
+  }
+  ortModulePromise = import("onnxruntime-web");
+  return ortModulePromise;
+}
+
+interface GpuLike {
+  requestAdapter(): Promise<unknown>;
+}
+
+export async function detectWebGpuAvailability(): Promise<WebGpuState> {
+  const gpu = (globalThis.navigator as (Navigator & { gpu?: GpuLike }) | undefined)?.gpu;
+  if (!gpu?.requestAdapter) {
+    return {
+      available: false,
+      reason: "navigator.gpu is unavailable in this browser."
+    };
+  }
+  try {
+    const adapter = await gpu.requestAdapter();
+    if (!adapter) {
+      return {
+        available: false,
+        reason: "The browser did not return a WebGPU adapter."
+      };
+    }
+    return {
+      available: true,
+      reason: ""
+    };
+  } catch (err: unknown) {
+    return {
+      available: false,
+      reason: err instanceof Error ? err.message : "Failed to request a WebGPU adapter."
+    };
+  }
+}
+
+export function getProviderCandidates(backend: string, webgpuState: WebGpuState): string[][] {
+  if (backend === "webgpu") {
+    if (!webgpuState.available) {
+      throw new Error(`WebGPU is unavailable: ${webgpuState.reason}`);
+    }
+    return [["webgpu"]];
+  }
+  if (backend === "wasm") {
+    return [["wasm"]];
+  }
+  return webgpuState.available ? [["webgpu"], ["wasm"]] : [["wasm"]];
+}
+
+function applyOrtEnvironmentOptions(ort: OrtModule, ortOptions: OrtOptions): void {
+  const wasmOptions = ort.env.wasm;
+
+  if (ortOptions.wasmPaths !== undefined) {
+    wasmOptions.wasmPaths = ortOptions.wasmPaths;
+  }
+  if (ortOptions.numThreads !== undefined) {
+    wasmOptions.numThreads = ortOptions.numThreads;
+  }
+  if (ortOptions.simd !== undefined) {
+    wasmOptions.simd = ortOptions.simd;
+  }
+  if (ortOptions.proxy !== undefined) {
+    wasmOptions.proxy = ortOptions.proxy;
+  }
+  if (ortOptions.disableWasmProxy) {
+    wasmOptions.proxy = false;
+  }
+}
+
+export async function initOrtRuntime(
+  ortOptions: OrtOptions | string = {}
+): Promise<OrtRuntimeResult> {
+  const backend =
+    typeof ortOptions === "string"
+      ? ortOptions
+      : ortOptions.backend === "webgpu" || ortOptions.backend === "wasm"
+        ? ortOptions.backend
+        : "auto";
+  const webgpuState = await detectWebGpuAvailability();
+  const ort = await loadOrtModule();
+  if (typeof ortOptions !== "string") {
+    applyOrtEnvironmentOptions(ort, ortOptions);
+  }
+  return {
+    ort,
+    webgpuState,
+    backend
+  };
+}
+
+export async function createSession(
+  ort: OrtModule,
+  modelBytes: Uint8Array,
+  providerCandidates: string[][]
+): Promise<SessionState> {
+  let lastErr: unknown = null;
+  for (const executionProviders of providerCandidates) {
+    try {
+      const session = await ort.InferenceSession.create(modelBytes, {
+        executionProviders,
+        graphOptimizationLevel: "all"
+      });
+      return { session, provider: executionProviders[0] };
+    } catch (err: unknown) {
+      lastErr = err;
+    }
+  }
+  throw lastErr instanceof Error ? lastErr : new Error("Failed to create ONNX session.");
+}
+
+export async function releaseSessions(
+  ...sessions: Array<import("onnxruntime-web").InferenceSession | null | undefined>
+): Promise<void> {
+  await Promise.all(
+    sessions.map(async (session) => {
+      if (!session?.release) return;
+      await session.release();
+    })
+  );
+}
diff --git a/paddleocr-js/packages/core/src/types/clipper-lib.d.ts b/paddleocr-js/packages/core/src/types/clipper-lib.d.ts
new file mode 100644
index 0000000..cbaddee
--- /dev/null
+++ b/paddleocr-js/packages/core/src/types/clipper-lib.d.ts
@@ -0,0 +1,37 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+/* eslint-disable @typescript-eslint/no-unused-vars */
+declare module "clipper-lib" {
+  interface IntPoint {
+    X: number;
+    Y: number;
+  }
+
+  type Path = IntPoint[];
+  type Paths = Path[];
+
+  class ClipperOffset {
+    AddPath(path: Path, joinType: number, endType: number): void;
+    Execute(result: Paths, delta: number): void;
+  }
+
+  const JoinType: {
+    readonly jtRound: number;
+  };
+
+  const EndType: {
+    readonly etClosedPolygon: number;
+  };
+
+  const ClipperLib: {
+    ClipperOffset: typeof ClipperOffset;
+    Paths: { new (): Paths };
+    JoinType: typeof JoinType;
+    EndType: typeof EndType;
+  };
+
+  export default ClipperLib;
+}
diff --git a/paddleocr-js/packages/core/src/types/index.ts b/paddleocr-js/packages/core/src/types/index.ts
new file mode 100644
index 0000000..89e717d
--- /dev/null
+++ b/paddleocr-js/packages/core/src/types/index.ts
@@ -0,0 +1,78 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+export type { OpenCv, Mat, MatVector, Size, Rect, Scalar, RotatedRect } from "@techstark/opencv-js";
+
+export type { Point2D, NormalizeConfig, DetBox, MiniBox } from "../models/common";
+
+export type {
+  DetModelConfig,
+  DetPostprocessConfig,
+  DetModel,
+  DetResult,
+  DetRuntimeOverrides,
+  LimitType
+} from "../models/det";
+
+export type { RecModelConfig, RecModel, RecResult, RecRuntimeOverrides } from "../models/rec";
+
+export type {
+  OcrRuntimeParamsInput,
+  OcrModelConfig,
+  ResolvedOcrParams
+} from "../pipelines/ocr/runtime-params";
+
+export type {
+  OcrResult,
+  OcrResultItem,
+  OcrResultMetrics,
+  OcrResultRuntime,
+  InitializationSummary,
+  OcrPipelineRunnerOptions,
+  SourceToMatFn
+} from "../pipelines/ocr/core";
+
+export type {
+  NormalizedPipelineConfig,
+  PipelineModelSelection,
+  PipelineRuntimeDefaults
+} from "../pipelines/ocr/config";
+
+export type {
+  ResolvedBackend,
+  ResolvedOcrOptions,
+  NormalizedOrtOptions,
+  WorkerResolvedOptions
+} from "../pipelines/ocr/shared";
+
+export type { PaddleOCRCreateOptions } from "../pipelines/ocr/index";
+
+export type { ModelAsset, ModelAssetsMap } from "../resources/model-asset";
+
+export type {
+  OrtModule,
+  WebGpuState,
+  OrtOptions,
+  OrtRuntimeResult,
+  SessionState
+} from "../runtime/ort";
+
+export type {
+  ImageSource,
+  SourceMatResult,
+  WorkerPayload,
+  WorkerPayloadResult
+} from "../platform/browser";
+
+export type {
+  TransportRequest,
+  TransportResponse,
+  TransportSuccessResponse,
+  TransportErrorResponse,
+  SerializedError
+} from "../worker/protocol";
+
+export type { WorkerOptions } from "../worker/client";
+export type { MessageHandler } from "../worker/entry";
diff --git a/paddleocr-js/packages/core/src/types/opencv.d.ts b/paddleocr-js/packages/core/src/types/opencv.d.ts
new file mode 100644
index 0000000..944292b
--- /dev/null
+++ b/paddleocr-js/packages/core/src/types/opencv.d.ts
@@ -0,0 +1,119 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+declare module "@techstark/opencv-js" {
+  export interface Mat {
+    rows: number;
+    cols: number;
+    data: Uint8Array;
+    data32S: Int32Array;
+    channels(): number;
+    copyTo(dst: Mat): void;
+    clone(): Mat;
+    delete(): void;
+    roi(rect: Rect): Mat;
+  }
+
+  export interface MatVector {
+    push_back(mat: Mat): void;
+    size(): number;
+    get(index: number): Mat;
+    delete(): void;
+  }
+
+  export interface Size {
+    width: number;
+    height: number;
+  }
+
+  export interface Rect {
+    x: number;
+    y: number;
+    width: number;
+    height: number;
+  }
+
+  export interface Scalar {
+    [index: number]: number;
+  }
+
+  export interface RotatedRect {
+    center: { x: number; y: number };
+    size: Size;
+    angle: number;
+  }
+
+  export interface OpenCv {
+    Mat: {
+      new (): Mat;
+      zeros(rows: number, cols: number, type: number): Mat;
+    };
+    MatVector: { new (): MatVector };
+    Size: { new (width: number, height: number): Size };
+    Rect: { new (x: number, y: number, width: number, height: number): Rect };
+    Scalar: { new (v0?: number, v1?: number, v2?: number, v3?: number): Scalar };
+    RotatedRect: {
+      points(rect: RotatedRect): Array<{ x: number; y: number }>;
+    };
+
+    matFromArray(rows: number, cols: number, type: number, data: ArrayLike<number>): Mat;
+    minAreaRect(points: Mat): RotatedRect;
+    imread(canvas: HTMLCanvasElement): Mat;
+    resize(src: Mat, dst: Mat, dsize: Size, fx?: number, fy?: number, interpolation?: number): void;
+    cvtColor(src: Mat, dst: Mat, code: number): void;
+    findContours(
+      image: Mat,
+      contours: MatVector,
+      hierarchy: Mat,
+      mode: number,
+      method: number
+    ): void;
+    fillPoly(img: Mat, pts: MatVector, color: Scalar): void;
+    mean(src: Mat, mask?: Mat): number[];
+    getPerspectiveTransform(src: Mat, dst: Mat): Mat;
+    warpPerspective(
+      src: Mat,
+      dst: Mat,
+      M: Mat,
+      dsize: Size,
+      flags?: number,
+      borderMode?: number,
+      borderValue?: Scalar
+    ): void;
+    rotate(src: Mat, dst: Mat, rotateCode: number): void;
+
+    // Mat type constants
+    CV_32FC1: number;
+    CV_32FC2: number;
+    CV_32SC2: number;
+    CV_8UC1: number;
+    CV_8UC4: number;
+
+    // Color conversion codes
+    COLOR_RGBA2BGR: number;
+    COLOR_GRAY2BGR: number;
+
+    // Interpolation flags
+    INTER_LINEAR: number;
+    INTER_CUBIC: number;
+
+    // Border types
+    BORDER_REPLICATE: number;
+
+    // Contour retrieval modes
+    RETR_LIST: number;
+
+    // Contour approximation methods
+    CHAIN_APPROX_SIMPLE: number;
+
+    // Rotation codes
+    ROTATE_90_COUNTERCLOCKWISE: number;
+  }
+
+  export type CvModule = OpenCv & { onRuntimeInitialized?: () => void };
+
+  const cv: CvModule | Promise<CvModule>;
+  export default cv;
+}
diff --git a/paddleocr-js/packages/core/src/utils/common.ts b/paddleocr-js/packages/core/src/utils/common.ts
new file mode 100644
index 0000000..db2cb36
--- /dev/null
+++ b/paddleocr-js/packages/core/src/utils/common.ts
@@ -0,0 +1,71 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+export function nowMs(): number {
+  return performance.now();
+}
+
+export function clamp(value: number, min: number, max: number): number {
+  return Math.max(min, Math.min(max, value));
+}
+
+export function distance2(p0: [number, number], p1: [number, number]): number {
+  const dx = p0[0] - p1[0];
+  const dy = p0[1] - p1[1];
+  return Math.sqrt(dx * dx + dy * dy);
+}
+
+export function formatMs(value: number): string {
+  return `${value.toFixed(1)} ms`;
+}
+
+export function withTimeout<T>(promise: Promise<T>, ms: number, label: string): Promise<T> {
+  let settled = false;
+  return new Promise<T>((resolve, reject) => {
+    const timer = setTimeout(() => {
+      if (settled) return;
+      settled = true;
+      reject(new Error(`${label} timed out after ${String(ms / 1000)}s`));
+    }, ms);
+
+    promise
+      .then((result) => {
+        if (settled) return;
+        settled = true;
+        clearTimeout(timer);
+        resolve(result);
+      })
+      .catch((err: unknown) => {
+        if (settled) return;
+        settled = true;
+        clearTimeout(timer);
+        // eslint-disable-next-line @typescript-eslint/prefer-promise-reject-errors -- propagating upstream rejection
+        reject(err);
+      });
+  });
+}
+
+export function resolveRuntimeBatchSize(override: unknown, defaultBatchSize: number): number {
+  const rawBatch = override ?? defaultBatchSize;
+  const coercedBatch =
+    typeof rawBatch === "number"
+      ? rawBatch
+      : typeof rawBatch === "string"
+        ? Number.parseInt(rawBatch, 10)
+        : Number.NaN;
+  return Math.max(1, Number.isFinite(coercedBatch) ? coercedBatch : 1);
+}
+
+export function chunkArray<T>(items: T[], size: number): T[][] {
+  const chunks: T[][] = [];
+  for (let i = 0; i < items.length; i += size) {
+    chunks.push(items.slice(i, i + size));
+  }
+  return chunks;
+}
+
+export function deepClone<T>(value: T): T {
+  return structuredClone(value);
+}
diff --git a/paddleocr-js/packages/core/src/viz/canvas-factory.ts b/paddleocr-js/packages/core/src/viz/canvas-factory.ts
new file mode 100644
index 0000000..0856198
--- /dev/null
+++ b/paddleocr-js/packages/core/src/viz/canvas-factory.ts
@@ -0,0 +1,45 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+type AnyCanvas = OffscreenCanvas | HTMLCanvasElement;
+
+export function createCanvas(width: number, height: number): AnyCanvas {
+  if (typeof OffscreenCanvas !== "undefined") {
+    return new OffscreenCanvas(width, height);
+  }
+  const canvas = document.createElement("canvas");
+  canvas.width = width;
+  canvas.height = height;
+  return canvas;
+}
+
+export function getContext2D(
+  canvas: AnyCanvas
+): CanvasRenderingContext2D | OffscreenCanvasRenderingContext2D {
+  const ctx = canvas.getContext("2d");
+  if (!ctx) {
+    throw new Error("Failed to create 2D rendering context.");
+  }
+  return ctx;
+}
+
+export function canvasToBlob(canvas: AnyCanvas, type: string, quality: number): Promise<Blob> {
+  if (canvas instanceof OffscreenCanvas) {
+    return canvas.convertToBlob({ type, quality });
+  }
+  return new Promise<Blob>((resolve, reject) => {
+    canvas.toBlob(
+      (blob) => {
+        if (blob) {
+          resolve(blob);
+        } else {
+          reject(new Error("canvas.toBlob() returned null."));
+        }
+      },
+      type,
+      quality
+    );
+  });
+}
diff --git a/paddleocr-js/packages/core/src/viz/color.ts b/paddleocr-js/packages/core/src/viz/color.ts
new file mode 100644
index 0000000..0db7acb
--- /dev/null
+++ b/paddleocr-js/packages/core/src/viz/color.ts
@@ -0,0 +1,17 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { RgbColor } from "./types";
+
+export function deterministicColor(index: number): RgbColor {
+  let seed = (index + 1) * 1103515245 + 12345;
+  seed >>>= 0;
+  const r = (seed >> 16) & 0xff;
+  seed = (seed * 1103515245 + 12345) >>> 0;
+  const g = (seed >> 16) & 0xff;
+  seed = (seed * 1103515245 + 12345) >>> 0;
+  const b = (seed >> 16) & 0xff;
+  return [r, g, b];
+}
diff --git a/paddleocr-js/packages/core/src/viz/font.ts b/paddleocr-js/packages/core/src/viz/font.ts
new file mode 100644
index 0000000..58fcf3c
--- /dev/null
+++ b/paddleocr-js/packages/core/src/viz/font.ts
@@ -0,0 +1,19 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { FontConfig } from "./types";
+
+export async function loadFontFace(config: FontConfig): Promise<FontFace> {
+  const source = typeof config.source === "string" ? `url(${config.source})` : config.source;
+
+  const face = new FontFace(config.family, source, config.descriptors);
+  await face.load();
+  document.fonts.add(face);
+  return face;
+}
+
+export function removeFontFace(face: FontFace): void {
+  document.fonts.delete(face);
+}
diff --git a/paddleocr-js/packages/core/src/viz/index.ts b/paddleocr-js/packages/core/src/viz/index.ts
new file mode 100644
index 0000000..bf15d8f
--- /dev/null
+++ b/paddleocr-js/packages/core/src/viz/index.ts
@@ -0,0 +1,10 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+export { OcrVisualizer, renderOcrToBlob } from "./ocr/renderer";
+export { deterministicColor } from "./color";
+
+export type { RgbColor, FontConfig } from "./types";
+export type { BoxStyleOptions, OcrVisualizerOptions } from "./ocr/types";
diff --git a/paddleocr-js/packages/core/src/viz/ocr/draw-boxes.ts b/paddleocr-js/packages/core/src/viz/ocr/draw-boxes.ts
new file mode 100644
index 0000000..74196e0
--- /dev/null
+++ b/paddleocr-js/packages/core/src/viz/ocr/draw-boxes.ts
@@ -0,0 +1,51 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { OcrResultItem } from "../../pipelines/ocr/core";
+import type { Point2D } from "../../models/common";
+import type { RgbColor } from "../types";
+import type { BoxStyleOptions } from "./types";
+import { deterministicColor } from "../color";
+
+const DEFAULT_FILL_OPACITY = 0.5;
+
+type DrawableImage = ImageBitmap | HTMLImageElement;
+
+function drawPolygonPath(
+  ctx: CanvasRenderingContext2D | OffscreenCanvasRenderingContext2D,
+  poly: Point2D[]
+): void {
+  ctx.beginPath();
+  ctx.moveTo(poly[0][0], poly[0][1]);
+  for (let i = 1; i < poly.length; i += 1) {
+    ctx.lineTo(poly[i][0], poly[i][1]);
+  }
+  ctx.closePath();
+}
+
+export function drawBoxesPanel(
+  ctx: CanvasRenderingContext2D | OffscreenCanvasRenderingContext2D,
+  image: DrawableImage,
+  items: OcrResultItem[],
+  style: BoxStyleOptions
+): void {
+  const fillOpacity = style.fillOpacity ?? DEFAULT_FILL_OPACITY;
+  const getColor = style.colorFn ?? deterministicColor;
+
+  // Draw original image
+  ctx.drawImage(image, 0, 0);
+
+  // Draw filled polygons with the specified opacity.
+  // Canvas source-over compositing with alpha gives:
+  //   result = color * alpha + original * (1 - alpha)
+  for (let i = 0; i < items.length; i += 1) {
+    const [r, g, b]: RgbColor = getColor(i);
+    ctx.save();
+    ctx.fillStyle = `rgba(${String(r)}, ${String(g)}, ${String(b)}, ${String(fillOpacity)})`;
+    drawPolygonPath(ctx, items[i].poly);
+    ctx.fill();
+    ctx.restore();
+  }
+}
diff --git a/paddleocr-js/packages/core/src/viz/ocr/draw-text.ts b/paddleocr-js/packages/core/src/viz/ocr/draw-text.ts
new file mode 100644
index 0000000..9dca752
--- /dev/null
+++ b/paddleocr-js/packages/core/src/viz/ocr/draw-text.ts
@@ -0,0 +1,166 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { OcrResultItem } from "../../pipelines/ocr/core";
+import type { Point2D } from "../../models/common";
+import type { RgbColor } from "../types";
+import type { BoxStyleOptions } from "./types";
+import { deterministicColor } from "../color";
+
+const DEFAULT_BG = "#ffffff";
+const OUTLINE_LINE_WIDTH = 1;
+const TEXT_COLOR = "#000000";
+const ROTATION_THRESHOLD_DEG = 5;
+const VERTICAL_LINE_SPACING = 2;
+
+function topEdgeAngle(poly: Point2D[]): number {
+  const dx = poly[1][0] - poly[0][0];
+  const dy = poly[1][1] - poly[0][1];
+  return Math.atan2(dy, dx);
+}
+
+function polyBounds(poly: Point2D[]): {
+  minX: number;
+  minY: number;
+  maxX: number;
+  maxY: number;
+  width: number;
+  height: number;
+} {
+  let minX = Infinity;
+  let minY = Infinity;
+  let maxX = -Infinity;
+  let maxY = -Infinity;
+  for (const [x, y] of poly) {
+    if (x < minX) minX = x;
+    if (y < minY) minY = y;
+    if (x > maxX) maxX = x;
+    if (y > maxY) maxY = y;
+  }
+  return { minX, minY, maxX, maxY, width: maxX - minX, height: maxY - minY };
+}
+
+function drawPolygonPath(
+  ctx: CanvasRenderingContext2D | OffscreenCanvasRenderingContext2D,
+  poly: Point2D[],
+  offsetX: number
+): void {
+  ctx.beginPath();
+  ctx.moveTo(poly[0][0] + offsetX, poly[0][1]);
+  for (let i = 1; i < poly.length; i += 1) {
+    ctx.lineTo(poly[i][0] + offsetX, poly[i][1]);
+  }
+  ctx.closePath();
+}
+
+function drawVerticalText(
+  ctx: CanvasRenderingContext2D | OffscreenCanvasRenderingContext2D,
+  text: string,
+  x: number,
+  startY: number,
+  fontSize: number,
+  fontFamily: string
+): void {
+  ctx.font = `${String(fontSize)}px "${fontFamily}"`;
+  let y = startY;
+  for (const char of text) {
+    ctx.fillText(char, x, y);
+    y += fontSize + VERTICAL_LINE_SPACING;
+  }
+}
+
+export function drawTextPanel(
+  ctx: CanvasRenderingContext2D | OffscreenCanvasRenderingContext2D,
+  offsetX: number,
+  height: number,
+  items: OcrResultItem[],
+  style: BoxStyleOptions,
+  fontFamily: string,
+  background?: string
+): void {
+  const getColor = style.colorFn ?? deterministicColor;
+  const bg = background ?? DEFAULT_BG;
+
+  // Fill background
+  ctx.save();
+  ctx.fillStyle = bg;
+  ctx.fillRect(offsetX, 0, offsetX, height);
+  ctx.restore();
+
+  for (let i = 0; i < items.length; i += 1) {
+    const item = items[i];
+    const [r, g, b]: RgbColor = getColor(i);
+    const bounds = polyBounds(item.poly);
+    const angle = topEdgeAngle(item.poly);
+    const absDeg = Math.abs(angle * (180 / Math.PI));
+    const needsRotation = absDeg > ROTATION_THRESHOLD_DEG && absDeg < 180 - ROTATION_THRESHOLD_DEG;
+
+    // Detect vertical text: height > 2 * width and height > 30px
+    const isVertical = bounds.height > 2 * bounds.width && bounds.height > 30;
+
+    // Draw box outline
+    ctx.save();
+    ctx.lineWidth = OUTLINE_LINE_WIDTH;
+    ctx.strokeStyle = `rgb(${String(r)}, ${String(g)}, ${String(b)})`;
+    drawPolygonPath(ctx, item.poly, offsetX);
+    ctx.stroke();
+    ctx.restore();
+
+    // Draw text
+    ctx.save();
+    ctx.fillStyle = TEXT_COLOR;
+
+    if (isVertical) {
+      // Vertical text: render characters one-by-one, stacked vertically
+      ctx.textBaseline = "top";
+      const chars = Array.from(item.text);
+      const charCount = Math.max(1, chars.length);
+      let fontSize = Math.max(8, Math.floor(bounds.width * 0.8));
+
+      // Scale down if total character height exceeds box height
+      const totalHeight = charCount * (fontSize + VERTICAL_LINE_SPACING);
+      if (totalHeight > bounds.height) {
+        fontSize = Math.max(8, Math.floor((bounds.height / charCount) * 0.8));
+      }
+
+      // Ensure each character fits within the box width
+      ctx.font = `${String(fontSize)}px "${fontFamily}"`;
+      const maxCharWidth = Math.max(...chars.map((c) => ctx.measureText(c).width));
+      if (maxCharWidth > bounds.width) {
+        fontSize = Math.max(8, Math.floor(fontSize * (bounds.width / maxCharWidth)));
+      }
+
+      const x = bounds.minX + offsetX + (bounds.width - fontSize) / 2;
+      const y = bounds.minY + 2;
+      drawVerticalText(ctx, item.text, x, y, fontSize, fontFamily);
+    } else {
+      // Horizontal text
+      ctx.textBaseline = "middle";
+      let fontSize = Math.max(12, Math.floor(bounds.height * 0.8));
+      ctx.font = `${String(fontSize)}px "${fontFamily}"`;
+
+      // Shrink font if text is wider than the box
+      const measured = ctx.measureText(item.text);
+      if (measured.width > bounds.width && bounds.width > 0) {
+        fontSize = Math.max(8, Math.floor(fontSize * (bounds.width / measured.width)));
+        ctx.font = `${String(fontSize)}px "${fontFamily}"`;
+      }
+
+      if (needsRotation) {
+        const cx = bounds.minX + bounds.width / 2 + offsetX;
+        const cy = bounds.minY + bounds.height / 2;
+        ctx.translate(cx, cy);
+        ctx.rotate(angle);
+        ctx.fillText(item.text, -bounds.width / 2, 0);
+      } else {
+        const x = bounds.minX + offsetX + 2;
+        const y = bounds.minY + bounds.height / 2;
+        ctx.fillText(item.text, x, y);
+      }
+    }
+
+    ctx.restore();
+  }
+}
diff --git a/paddleocr-js/packages/core/src/viz/ocr/renderer.ts b/paddleocr-js/packages/core/src/viz/ocr/renderer.ts
new file mode 100644
index 0000000..02325ee
--- /dev/null
+++ b/paddleocr-js/packages/core/src/viz/ocr/renderer.ts
@@ -0,0 +1,86 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { OcrResult } from "../../pipelines/ocr/core";
+import type { OcrVisualizerOptions } from "./types";
+import { loadFontFace, removeFontFace } from "../font";
+import { renderSideBySideToImageBitmap, renderSideBySideToBlob } from "./side-by-side";
+import type { SideBySideOptions } from "./side-by-side";
+
+type DrawableImage = ImageBitmap | HTMLImageElement;
+
+const DEFAULT_FONT_FAMILY = "sans-serif";
+const DEFAULT_OUTPUT_FORMAT = "png";
+const DEFAULT_OUTPUT_QUALITY = 0.92;
+const DEFAULT_TEXT_PANEL_BG = "#ffffff";
+
+function resolveOptions(
+  base: OcrVisualizerOptions,
+  overrides?: Partial<OcrVisualizerOptions>
+): SideBySideOptions {
+  const merged = overrides ? { ...base, ...overrides } : base;
+  return {
+    boxStyle: merged.boxStyle ?? {},
+    fontFamily: merged.font?.family ?? DEFAULT_FONT_FAMILY,
+    textPanelBackground: merged.textPanelBackground ?? DEFAULT_TEXT_PANEL_BG,
+    outputFormat: merged.outputFormat ?? DEFAULT_OUTPUT_FORMAT,
+    outputQuality: merged.outputQuality ?? DEFAULT_OUTPUT_QUALITY
+  };
+}
+
+export class OcrVisualizer {
+  private options: OcrVisualizerOptions;
+  private loadedFace: FontFace | null = null;
+
+  constructor(options?: OcrVisualizerOptions) {
+    this.options = options ?? {};
+  }
+
+  async loadFont(): Promise<void> {
+    if (!this.options.font) return;
+    if (this.loadedFace) return;
+    this.loadedFace = await loadFontFace(this.options.font);
+  }
+
+  async renderSideBySide(
+    image: DrawableImage,
+    result: OcrResult,
+    overrides?: Partial<OcrVisualizerOptions>
+  ): Promise<ImageBitmap> {
+    await this.loadFont();
+    const opts = resolveOptions(this.options, overrides);
+    return renderSideBySideToImageBitmap(image, result, opts);
+  }
+
+  async toBlob(
+    image: DrawableImage,
+    result: OcrResult,
+    overrides?: Partial<OcrVisualizerOptions>
+  ): Promise<Blob> {
+    await this.loadFont();
+    const opts = resolveOptions(this.options, overrides);
+    return renderSideBySideToBlob(image, result, opts);
+  }
+
+  dispose(): void {
+    if (this.loadedFace) {
+      removeFontFace(this.loadedFace);
+      this.loadedFace = null;
+    }
+  }
+}
+
+export async function renderOcrToBlob(
+  image: DrawableImage,
+  result: OcrResult,
+  options?: OcrVisualizerOptions
+): Promise<Blob> {
+  const viz = new OcrVisualizer(options);
+  try {
+    return await viz.toBlob(image, result);
+  } finally {
+    viz.dispose();
+  }
+}
diff --git a/paddleocr-js/packages/core/src/viz/ocr/side-by-side.ts b/paddleocr-js/packages/core/src/viz/ocr/side-by-side.ts
new file mode 100644
index 0000000..3e7b24d
--- /dev/null
+++ b/paddleocr-js/packages/core/src/viz/ocr/side-by-side.ts
@@ -0,0 +1,70 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { OcrResult } from "../../pipelines/ocr/core";
+import type { BoxStyleOptions } from "./types";
+import { drawBoxesPanel } from "./draw-boxes";
+import { drawTextPanel } from "./draw-text";
+import { createCanvas, getContext2D, canvasToBlob } from "../canvas-factory";
+
+type DrawableImage = ImageBitmap | HTMLImageElement;
+
+function imageWidth(image: DrawableImage): number {
+  return image instanceof HTMLImageElement ? image.naturalWidth : image.width;
+}
+
+function imageHeight(image: DrawableImage): number {
+  return image instanceof HTMLImageElement ? image.naturalHeight : image.height;
+}
+
+export interface SideBySideOptions {
+  boxStyle: BoxStyleOptions;
+  fontFamily: string;
+  textPanelBackground: string;
+  outputFormat: string;
+  outputQuality: number;
+}
+
+export function renderSideBySideToCanvas(
+  image: DrawableImage,
+  result: OcrResult,
+  options: SideBySideOptions
+): OffscreenCanvas | HTMLCanvasElement {
+  const w = imageWidth(image);
+  const h = imageHeight(image);
+  const canvas = createCanvas(w * 2, h);
+  const ctx = getContext2D(canvas);
+
+  drawBoxesPanel(ctx, image, result.items, options.boxStyle);
+  drawTextPanel(
+    ctx,
+    w,
+    h,
+    result.items,
+    options.boxStyle,
+    options.fontFamily,
+    options.textPanelBackground
+  );
+
+  return canvas;
+}
+
+export async function renderSideBySideToImageBitmap(
+  image: DrawableImage,
+  result: OcrResult,
+  options: SideBySideOptions
+): Promise<ImageBitmap> {
+  const canvas = renderSideBySideToCanvas(image, result, options);
+  return createImageBitmap(canvas as ImageBitmapSource);
+}
+
+export async function renderSideBySideToBlob(
+  image: DrawableImage,
+  result: OcrResult,
+  options: SideBySideOptions
+): Promise<Blob> {
+  const canvas = renderSideBySideToCanvas(image, result, options);
+  return canvasToBlob(canvas, `image/${options.outputFormat}`, options.outputQuality);
+}
diff --git a/paddleocr-js/packages/core/src/viz/ocr/types.ts b/paddleocr-js/packages/core/src/viz/ocr/types.ts
new file mode 100644
index 0000000..f1ea65e
--- /dev/null
+++ b/paddleocr-js/packages/core/src/viz/ocr/types.ts
@@ -0,0 +1,26 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import type { RgbColor, FontConfig } from "../types";
+
+export interface BoxStyleOptions {
+  /** Fill opacity 0-1. Default: 0.5. */
+  fillOpacity?: number;
+  /** Custom color function. Default: deterministic LCG-based colors. */
+  colorFn?: (index: number) => RgbColor;
+}
+
+export interface OcrVisualizerOptions {
+  /** Custom font configuration. Falls back to system sans-serif if omitted. */
+  font?: FontConfig;
+  /** Detection box style overrides. */
+  boxStyle?: BoxStyleOptions;
+  /** Right panel background color. Default: "#ffffff". */
+  textPanelBackground?: string;
+  /** Output image format. Default: "png". */
+  outputFormat?: "png" | "jpeg" | "webp";
+  /** JPEG/WebP quality 0-1. Default: 0.92. */
+  outputQuality?: number;
+}
diff --git a/paddleocr-js/packages/core/src/viz/types.ts b/paddleocr-js/packages/core/src/viz/types.ts
new file mode 100644
index 0000000..db45177
--- /dev/null
+++ b/paddleocr-js/packages/core/src/viz/types.ts
@@ -0,0 +1,16 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+/** An RGB color as a 3-element tuple of 0-255 integers. */
+export type RgbColor = [number, number, number];
+
+export interface FontConfig {
+  /** CSS font-family name. */
+  family: string;
+  /** Font source: URL string or ArrayBuffer. */
+  source: string | ArrayBuffer;
+  /** FontFace descriptors (weight, style, etc.). */
+  descriptors?: FontFaceDescriptors;
+}
diff --git a/paddleocr-js/packages/core/src/worker/client.ts b/paddleocr-js/packages/core/src/worker/client.ts
new file mode 100644
index 0000000..edf13cc
--- /dev/null
+++ b/paddleocr-js/packages/core/src/worker/client.ts
@@ -0,0 +1,106 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import { createTransportRequest, deserializeError, isTransportResponse } from "./protocol";
+
+interface PendingRequest {
+  resolve: (value: unknown) => void;
+  reject: (reason: unknown) => void;
+}
+
+export interface WorkerOptions {
+  createWorker?: () => Worker;
+}
+
+export class WorkerTransportClient {
+  private workerOptions: WorkerOptions;
+  private worker: Worker | null;
+  private pending: Map<number, PendingRequest>;
+  private nextRequestId: number;
+  private disposed: boolean;
+
+  constructor(workerOptions: WorkerOptions = {}) {
+    this.workerOptions = workerOptions;
+    this.worker = null;
+    this.pending = new Map();
+    this.nextRequestId = 1;
+    this.disposed = false;
+  }
+
+  ensureActive(): void {
+    if (this.disposed) {
+      throw new Error("Worker transport client has been disposed.");
+    }
+  }
+
+  ensureWorker(): Worker {
+    this.ensureActive();
+    if (this.worker) {
+      return this.worker;
+    }
+
+    const workerFactory = this.workerOptions.createWorker;
+    if (typeof workerFactory !== "function") {
+      throw new Error("Worker transport client requires a createWorker() factory.");
+    }
+    const worker = workerFactory();
+    worker.onmessage = (event: MessageEvent) => {
+      const message = event.data as unknown;
+      if (!isTransportResponse(message)) return;
+      const pending = this.pending.get(message.requestId);
+      if (!pending) return;
+      this.pending.delete(message.requestId);
+      if (message.status === "success") {
+        pending.resolve(message.payload);
+      } else {
+        pending.reject(deserializeError(message.error));
+      }
+    };
+    worker.onerror = (event: ErrorEvent) => {
+      const error = new Error(event.message || "OCR worker failed.");
+      for (const pending of this.pending.values()) {
+        pending.reject(error);
+      }
+      this.pending.clear();
+    };
+    this.worker = worker;
+    return worker;
+  }
+
+  request(type: string, payload: unknown, transferables: Transferable[] = []): Promise<unknown> {
+    const worker = this.ensureWorker();
+    const requestId = this.nextRequestId;
+    this.nextRequestId += 1;
+
+    return new Promise((resolve, reject) => {
+      this.pending.set(requestId, { resolve, reject });
+      worker.postMessage(createTransportRequest(type, payload, requestId), transferables);
+    });
+  }
+
+  disposeWorker(): void {
+    if (!this.worker) {
+      return;
+    }
+    this.worker.terminate();
+    this.worker = null;
+  }
+
+  dispose(): void {
+    if (this.disposed) {
+      return;
+    }
+    this.disposed = true;
+    for (const pending of this.pending.values()) {
+      pending.reject(new Error("Worker transport client has been disposed."));
+    }
+    this.pending.clear();
+    this.disposeWorker();
+  }
+}
+
+export function createWorkerTransportClient(workerOptions: WorkerOptions): WorkerTransportClient {
+  return new WorkerTransportClient(workerOptions);
+}
diff --git a/paddleocr-js/packages/core/src/worker/entry.ts b/paddleocr-js/packages/core/src/worker/entry.ts
new file mode 100644
index 0000000..096c173
--- /dev/null
+++ b/paddleocr-js/packages/core/src/worker/entry.ts
@@ -0,0 +1,37 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import { createTransportError, createTransportSuccess, isTransportRequest } from "./protocol";
+
+export type MessageHandler = (type: string, payload: Record<string, unknown>) => Promise<unknown>;
+
+interface WorkerLikeScope {
+  onmessage: ((event: MessageEvent) => void) | null;
+  postMessage(message: unknown): void;
+}
+
+export function attachWorkerMessageHandler(
+  handleMessage: MessageHandler,
+  workerScope: WorkerLikeScope = self as unknown as WorkerLikeScope
+): void {
+  workerScope.onmessage = (event: MessageEvent) => {
+    const message = event.data as unknown;
+    if (!isTransportRequest(message)) {
+      return;
+    }
+
+    void (async () => {
+      try {
+        const payload = await handleMessage(
+          message.type,
+          (message.payload || {}) as Record<string, unknown>
+        );
+        workerScope.postMessage(createTransportSuccess(message.requestId, payload));
+      } catch (error: unknown) {
+        workerScope.postMessage(createTransportError(message.requestId, error));
+      }
+    })();
+  };
+}
diff --git a/paddleocr-js/packages/core/src/worker/protocol.ts b/paddleocr-js/packages/core/src/worker/protocol.ts
new file mode 100644
index 0000000..fb9c7f8
--- /dev/null
+++ b/paddleocr-js/packages/core/src/worker/protocol.ts
@@ -0,0 +1,107 @@
+/*
+ * Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+const REQUEST_KIND = "worker-transport-request";
+const RESPONSE_KIND = "worker-transport-response";
+
+export interface SerializedError {
+  name: string;
+  message: string;
+  stack: string;
+}
+
+export interface TransportRequest {
+  kind: typeof REQUEST_KIND;
+  type: string;
+  payload: unknown;
+  requestId: number;
+}
+
+export interface TransportSuccessResponse {
+  kind: typeof RESPONSE_KIND;
+  status: "success";
+  requestId: number;
+  payload: unknown;
+}
+
+export interface TransportErrorResponse {
+  kind: typeof RESPONSE_KIND;
+  status: "error";
+  requestId: number;
+  error: SerializedError;
+}
+
+export type TransportResponse = TransportSuccessResponse | TransportErrorResponse;
+
+export function createTransportRequest(
+  type: string,
+  payload: unknown,
+  requestId: number
+): TransportRequest {
+  return {
+    kind: REQUEST_KIND,
+    type,
+    payload,
+    requestId
+  };
+}
+
+export function createTransportSuccess(
+  requestId: number,
+  payload: unknown
+): TransportSuccessResponse {
+  return {
+    kind: RESPONSE_KIND,
+    status: "success",
+    requestId,
+    payload
+  };
+}
+
+export function createTransportError(requestId: number, error: unknown): TransportErrorResponse {
+  return {
+    kind: RESPONSE_KIND,
+    status: "error",
+    requestId,
+    error: serializeError(error)
+  };
+}
+
+export function isTransportRequest(message: unknown): message is TransportRequest {
+  return (
+    typeof message === "object" &&
+    message !== null &&
+    "kind" in message &&
+    message.kind === REQUEST_KIND
+  );
+}
+
+export function isTransportResponse(message: unknown): message is TransportResponse {
+  return (
+    typeof message === "object" &&
+    message !== null &&
+    "kind" in message &&
+    message.kind === RESPONSE_KIND
+  );
+}
+
+export function serializeError(error: unknown): SerializedError {
+  const err = error as Partial<Error> | undefined;
+  return {
+    name: err?.name || "Error",
+    message: err?.message || "Unknown worker error.",
+    stack: err?.stack || ""
+  };
+}
+
+export function deserializeError(error: unknown): Error {
+  const normalized = (error || {}) as SerializedError;
+  const instance = new Error(normalized.message || "Unknown worker error.");
+  instance.name = normalized.name || "Error";
+  if (normalized.stack) {
+    instance.stack = normalized.stack;
+  }
+  return instance;
+}
diff --git a/paddleocr-js/packages/core/test/browser-source.test.ts b/paddleocr-js/packages/core/test/browser-source.test.ts
new file mode 100644
index 0000000..618ac16
--- /dev/null
+++ b/paddleocr-js/packages/core/test/browser-source.test.ts
@@ -0,0 +1,35 @@
+import { describe, expect, it } from "vitest";
+
+import { sourceToMat } from "../src/platform/browser";
+
+class FakeMat {
+  constructor(rows, cols) {
+    this.rows = rows;
+    this.cols = cols;
+    this.deleted = false;
+  }
+
+  clone() {
+    return new FakeMat(this.rows, this.cols);
+  }
+
+  delete() {
+    this.deleted = true;
+  }
+}
+
+describe("browser source helpers", () => {
+  it("accepts cv.Mat inputs by cloning them", async () => {
+    const source = new FakeMat(32, 64);
+    const loaded = await sourceToMat({ Mat: FakeMat }, source);
+
+    expect(loaded.width).toBe(64);
+    expect(loaded.height).toBe(32);
+    expect(loaded.mat).toBeInstanceOf(FakeMat);
+    expect(loaded.mat).not.toBe(source);
+
+    loaded.dispose();
+    expect(loaded.mat.deleted).toBe(true);
+    expect(source.deleted).toBe(false);
+  });
+});
diff --git a/paddleocr-js/packages/core/test/det-model.test.ts b/paddleocr-js/packages/core/test/det-model.test.ts
new file mode 100644
index 0000000..727b788
--- /dev/null
+++ b/paddleocr-js/packages/core/test/det-model.test.ts
@@ -0,0 +1,608 @@
+import { afterEach, describe, expect, it, vi } from "vitest";
+import { createMockOrtTensorClass } from "./helpers/mock-ort-tensor";
+
+const assertModelResources = vi.fn();
+const createSession = vi.fn();
+const getProviderCandidates = vi.fn();
+const releaseSessions = vi.fn();
+const clamp = vi.fn((value, min, max) => Math.max(min, Math.min(max, value)));
+const withTimeout = vi.fn((promise) => promise);
+const boxScoreFast = vi.fn();
+const getMiniBoxFromPoints = vi.fn();
+const getTransformOp = vi.fn();
+const parseInferenceConfigText = vi.fn();
+const parseScaleValue = vi.fn();
+const toBgrFloatCHWFromBgr = vi.fn();
+const unclip = vi.fn();
+
+vi.mock("../src/resources/model-asset", () => ({
+  assertModelResources
+}));
+
+vi.mock("../src/runtime/ort", () => ({
+  createSession,
+  getProviderCandidates,
+  releaseSessions
+}));
+
+vi.mock("../src/utils/common", async (importOriginal) => {
+  const actual = await importOriginal<typeof import("../src/utils/common")>();
+  return {
+    ...actual,
+    clamp,
+    withTimeout
+  };
+});
+
+vi.mock("../src/models/common", () => ({
+  boxScoreFast,
+  getMiniBoxFromPoints,
+  getTransformOp,
+  parseInferenceConfigText,
+  parseScaleValue,
+  toBgrFloatCHWFromBgr,
+  unclip
+}));
+
+afterEach(() => {
+  vi.resetModules();
+  vi.clearAllMocks();
+});
+
+async function loadDetModule() {
+  return import("../src/models/det");
+}
+
+/** CV facade for `createDetModel().predict()` integration-style test (preprocess → infer → postprocess). */
+function createDetModelIntegrationCv() {
+  return {
+    Mat: class Mat {
+      constructor() {
+        this.data = new Uint8Array(1);
+      }
+      channels() {
+        return 3;
+      }
+      copyTo() {}
+      delete() {}
+    },
+    Size: class Size {
+      constructor(width, height) {
+        this.width = width;
+        this.height = height;
+      }
+    },
+    INTER_LINEAR: "linear",
+    CV_32FC1: "float1",
+    CV_8UC1: "mask1",
+    RETR_LIST: "list",
+    CHAIN_APPROX_SIMPLE: "chain",
+    resize: vi.fn((src, dst, size) => {
+      dst.data = new Uint8Array(size.width * size.height * 3);
+      dst.channels = () => 3;
+      dst.copyTo = vi.fn();
+      dst.delete = vi.fn();
+    }),
+    cvtColor: vi.fn(),
+    matFromArray: vi
+      .fn()
+      .mockImplementationOnce(() => ({ delete: vi.fn() }))
+      .mockImplementationOnce(() => ({ delete: vi.fn() })),
+    MatVector: class MatVector {
+      size() {
+        return 1;
+      }
+      get() {
+        return {
+          rows: 4,
+          data32S: [0, 0, 4, 0, 4, 2, 0, 2],
+          delete: vi.fn()
+        };
+      }
+      delete() {}
+    },
+    findContours: vi.fn()
+  };
+}
+
+describe("detection model", () => {
+  it("parses detection configs with explicit values and fallbacks", async () => {
+    parseInferenceConfigText.mockReturnValue({
+      PreProcess: {
+        transform_ops: [{ id: "resize" }, { id: "normalize" }]
+      },
+      PostProcess: {
+        thresh: "0.22",
+        box_thresh: "0.55",
+        max_candidates: "200",
+        unclip_ratio: "1.8"
+      }
+    });
+    getTransformOp
+      .mockReturnValueOnce({ resize_long: 736 })
+      .mockReturnValueOnce({ mean: [0.1], std: [0.9], scale: "1./2." });
+    parseScaleValue.mockReturnValue(0.5);
+
+    const { DEFAULT_DET_MODEL_PARSE_FALLBACKS, parseDetModelConfigText } = await loadDetModule();
+    expect(parseDetModelConfigText("config")).toEqual({
+      resizeLong: 736,
+      limitType: "max",
+      maxSideLimit: 4000,
+      normalize: {
+        mean: [0.1],
+        std: [0.9],
+        scale: 0.5
+      },
+      postprocess: {
+        thresh: 0.22,
+        boxThresh: 0.55,
+        maxCandidates: 200,
+        unclipRatio: 1.8
+      }
+    });
+
+    parseInferenceConfigText.mockReturnValue({});
+    getTransformOp.mockReturnValue(undefined);
+    parseScaleValue.mockReturnValue(1 / 255);
+
+    expect(parseDetModelConfigText("fallback")).toEqual({
+      resizeLong: DEFAULT_DET_MODEL_PARSE_FALLBACKS.resizeLong,
+      limitType: DEFAULT_DET_MODEL_PARSE_FALLBACKS.limitType,
+      maxSideLimit: DEFAULT_DET_MODEL_PARSE_FALLBACKS.maxSideLimit,
+      normalize: {
+        mean: DEFAULT_DET_MODEL_PARSE_FALLBACKS.normalize.mean,
+        std: DEFAULT_DET_MODEL_PARSE_FALLBACKS.normalize.std,
+        scale: 1 / 255
+      },
+      postprocess: {
+        thresh: DEFAULT_DET_MODEL_PARSE_FALLBACKS.postprocess.thresh,
+        boxThresh: DEFAULT_DET_MODEL_PARSE_FALLBACKS.postprocess.boxThresh,
+        maxCandidates: DEFAULT_DET_MODEL_PARSE_FALLBACKS.postprocess.maxCandidates,
+        unclipRatio: DEFAULT_DET_MODEL_PARSE_FALLBACKS.postprocess.unclipRatio
+      }
+    });
+  });
+
+  it("runs detection models and crops rotated boxes", async () => {
+    const { cropByPoly } = await import("../src/pipelines/ocr/crop");
+    parseInferenceConfigText.mockReturnValue({
+      PreProcess: { transform_ops: [] },
+      PostProcess: { max_candidates: "10" }
+    });
+    getTransformOp.mockImplementation((_ops, id) => {
+      if (id === "DetResizeForTest") return { resize_long: 64 };
+      return null;
+    });
+    parseScaleValue.mockReturnValue(1 / 255);
+    clamp.mockImplementation((value, min, max) => Math.max(min, Math.min(max, value)));
+    getProviderCandidates.mockReturnValue([["wasm"]]);
+
+    const tensorCalls = [];
+    const ort = {
+      Tensor: createMockOrtTensorClass(tensorCalls)
+    };
+    const sessionRun = vi.fn().mockResolvedValue({
+      output: {
+        dims: [1, 1, 4, 8],
+        data: new Float32Array(32).fill(0.9)
+      }
+    });
+    const session = {
+      inputNames: ["input"],
+      outputNames: ["output"],
+      run: sessionRun
+    };
+    createSession.mockResolvedValue({
+      session,
+      provider: "wasm"
+    });
+    toBgrFloatCHWFromBgr.mockReturnValue(new Float32Array(3 * 32 * 64).fill(1));
+
+    const makeCv = () => {
+      const pred = { delete: vi.fn() };
+      const bitmap = { delete: vi.fn() };
+      const contour = {
+        rows: 4,
+        data32S: [0, 0, 4, 0, 4, 2, 0, 2],
+        delete: vi.fn()
+      };
+      const warped = {
+        rows: 20,
+        cols: 10,
+        delete: vi.fn()
+      };
+      const rotated = {
+        rows: 10,
+        cols: 20,
+        delete: vi.fn()
+      };
+      return {
+        warped,
+        rotated,
+        Mat: class Mat {
+          constructor() {
+            return warped;
+          }
+        },
+        Size: class Size {
+          constructor(width, height) {
+            this.width = width;
+            this.height = height;
+          }
+        },
+        Scalar: class Scalar {},
+        INTER_LINEAR: "linear",
+        INTER_CUBIC: "cubic",
+        BORDER_REPLICATE: "replicate",
+        COLOR_RGBA2BGR: "rgba",
+        COLOR_GRAY2BGR: "gray",
+        ROTATE_90_COUNTERCLOCKWISE: "ccw",
+        CV_32FC1: "float1",
+        CV_8UC1: "mask1",
+        CV_32FC2: "float",
+        RETR_LIST: "list",
+        CHAIN_APPROX_SIMPLE: "chain",
+        resize: vi.fn((src, dst, size) => {
+          dst.data = new Uint8Array(size.width * size.height * 3);
+          dst.channels = () => 3;
+          dst.copyTo = vi.fn();
+          dst.delete = vi.fn();
+        }),
+        cvtColor: vi.fn(),
+        matFromArray: vi
+          .fn()
+          .mockImplementationOnce(() => pred)
+          .mockImplementationOnce(() => bitmap)
+          .mockImplementationOnce(() => ({ delete: vi.fn() }))
+          .mockImplementationOnce(() => ({ delete: vi.fn() })),
+        MatVector: class MatVector {
+          size() {
+            return 1;
+          }
+          get() {
+            return contour;
+          }
+          delete() {}
+        },
+        findContours: vi.fn(),
+        getPerspectiveTransform: vi.fn(() => ({ delete: vi.fn() })),
+        warpPerspective: vi.fn(),
+        rotate: vi.fn()
+      };
+    };
+    const cv = makeCv();
+
+    const { createDetModel } = await loadDetModule();
+    getMiniBoxFromPoints
+      .mockReturnValueOnce({
+        side: 4,
+        box: [
+          [0, 0],
+          [4, 0],
+          [4, 2],
+          [0, 2]
+        ]
+      })
+      .mockReturnValueOnce({
+        side: 6,
+        box: [
+          [0, 0],
+          [5, 0],
+          [5, 3],
+          [0, 3]
+        ]
+      });
+    boxScoreFast.mockReturnValue(0.9);
+    unclip.mockReturnValue([
+      [0, 0],
+      [5, 0],
+      [5, 3],
+      [0, 3]
+    ]);
+
+    const model = await createDetModel({
+      ort,
+      modelBytes: new Uint8Array([1]),
+      configText: "det-crop",
+      backend: "auto",
+      webgpuState: { available: false, reason: "" }
+    });
+    const [detResult] = await model.predict(
+      cv,
+      [
+        {
+          cols: 64,
+          rows: 32,
+          channels: () => 3
+        }
+      ],
+      {
+        thresh: 0.3,
+        boxThresh: 0.5,
+        unclipRatio: 1.5,
+        limitSideLen: 64,
+        limitType: "max",
+        maxSideLimit: 96
+      }
+    );
+
+    expect(sessionRun).toHaveBeenCalledTimes(1);
+    expect(tensorCalls[0]).toEqual({ type: "float32", dims: [1, 3, 32, 64], size: 6144 });
+    expect(detResult.boxes).toEqual([
+      {
+        poly: [
+          [0, 0],
+          [40, 0],
+          [40, 24],
+          [0, 24]
+        ],
+        score: 0.9
+      }
+    ]);
+
+    const cropWarped = {
+      rows: 20,
+      cols: 10,
+      delete: vi.fn()
+    };
+    const cropRotated = {
+      rows: 10,
+      cols: 20,
+      delete: vi.fn()
+    };
+    let cropMatCount = 0;
+    const cropCv = {
+      Size: cv.Size,
+      Scalar: cv.Scalar,
+      INTER_CUBIC: cv.INTER_CUBIC,
+      BORDER_REPLICATE: cv.BORDER_REPLICATE,
+      ROTATE_90_COUNTERCLOCKWISE: cv.ROTATE_90_COUNTERCLOCKWISE,
+      CV_32FC2: cv.CV_32FC2,
+      Mat: class Mat {
+        constructor() {
+          cropMatCount += 1;
+          return cropMatCount === 1 ? cropWarped : cropRotated;
+        }
+      },
+      matFromArray: vi
+        .fn()
+        .mockImplementationOnce(() => ({ delete: vi.fn() }))
+        .mockImplementationOnce(() => ({ delete: vi.fn() })),
+      getPerspectiveTransform: vi.fn(() => ({ delete: vi.fn() })),
+      warpPerspective: vi.fn(),
+      rotate: vi.fn()
+    };
+
+    getMiniBoxFromPoints.mockReturnValue({
+      box: [
+        [0, 0],
+        [10, 0],
+        [10, 20],
+        [0, 20]
+      ]
+    });
+    const rotatedCrop = cropByPoly(cropCv, { id: "src" }, [[0, 0]]);
+    expect(cropCv.rotate).toHaveBeenCalled();
+    expect(rotatedCrop).toBe(cropRotated);
+  });
+
+  it("runs batched detection when batchSize > 1 (one session.run per chunk)", async () => {
+    parseInferenceConfigText.mockReturnValue({
+      PreProcess: { transform_ops: [] },
+      PostProcess: { max_candidates: "10" }
+    });
+    getTransformOp.mockImplementation((_ops, id) => {
+      if (id === "DetResizeForTest") return { resize_long: 64 };
+      return null;
+    });
+    parseScaleValue.mockReturnValue(1 / 255);
+    clamp.mockImplementation((value, min, max) => Math.max(min, Math.min(max, value)));
+    getProviderCandidates.mockReturnValue([["wasm"]]);
+
+    const tensorCalls = [];
+    const ort = {
+      Tensor: createMockOrtTensorClass(tensorCalls)
+    };
+    const sessionRun = vi.fn().mockResolvedValue({
+      output: {
+        dims: [2, 1, 4, 8],
+        data: new Float32Array(64).fill(0.1)
+      }
+    });
+    const session = {
+      inputNames: ["input"],
+      outputNames: ["output"],
+      run: sessionRun
+    };
+    createSession.mockResolvedValue({
+      session,
+      provider: "wasm"
+    });
+    toBgrFloatCHWFromBgr.mockReturnValue(new Float32Array(3 * 32 * 64).fill(1));
+
+    const pred = { delete: vi.fn() };
+    const bitmap = { delete: vi.fn() };
+    const warped = {
+      rows: 20,
+      cols: 10,
+      delete: vi.fn()
+    };
+    const cv = {
+      warped,
+      Mat: class Mat {
+        constructor() {
+          return warped;
+        }
+      },
+      Size: class Size {
+        constructor(width, height) {
+          this.width = width;
+          this.height = height;
+        }
+      },
+      Scalar: class Scalar {},
+      INTER_LINEAR: "linear",
+      INTER_CUBIC: "cubic",
+      BORDER_REPLICATE: "replicate",
+      COLOR_RGBA2BGR: "rgba",
+      COLOR_GRAY2BGR: "gray",
+      ROTATE_90_COUNTERCLOCKWISE: "ccw",
+      CV_32FC1: "float1",
+      CV_8UC1: "mask1",
+      CV_32FC2: "float",
+      RETR_LIST: "list",
+      CHAIN_APPROX_SIMPLE: "chain",
+      resize: vi.fn((src, dst, size) => {
+        dst.data = new Uint8Array(size.width * size.height * 3);
+        dst.channels = () => 3;
+        dst.copyTo = vi.fn();
+        dst.delete = vi.fn();
+      }),
+      cvtColor: vi.fn(),
+      matFromArray: vi
+        .fn()
+        .mockImplementationOnce(() => pred)
+        .mockImplementationOnce(() => bitmap)
+        .mockImplementationOnce(() => pred)
+        .mockImplementationOnce(() => bitmap),
+      MatVector: class MatVector {
+        size() {
+          return 0;
+        }
+        delete() {}
+      },
+      findContours: vi.fn(),
+      getPerspectiveTransform: vi.fn(() => ({ delete: vi.fn() })),
+      warpPerspective: vi.fn(),
+      rotate: vi.fn()
+    };
+
+    const { createDetModel } = await loadDetModule();
+    const model = await createDetModel({
+      ort,
+      modelBytes: new Uint8Array([1]),
+      configText: "det-batch",
+      backend: "auto",
+      webgpuState: { available: false, reason: "" },
+      batchSize: 2
+    });
+
+    const mat = { cols: 64, rows: 32, channels: () => 3 };
+    const results = await model.predict(cv, [mat, mat], {
+      thresh: 0.3,
+      boxThresh: 0.5,
+      unclipRatio: 1.5,
+      limitSideLen: 64,
+      limitType: "max",
+      maxSideLimit: 96
+    });
+
+    expect(sessionRun).toHaveBeenCalledTimes(1);
+    const batchInput = tensorCalls.find((t) => t.dims[0] === 2);
+    expect(batchInput).toEqual({ type: "float32", dims: [2, 3, 32, 64], size: 12288 });
+    expect(results).toHaveLength(2);
+    expect(results[0].srcW).toBe(64);
+    expect(results[1].srcW).toBe(64);
+  });
+
+  it("creates, uses, and disposes detection models through runtime wrappers", async () => {
+    parseInferenceConfigText.mockReturnValue({
+      PreProcess: {
+        transform_ops: []
+      },
+      PostProcess: {}
+    });
+    getTransformOp.mockReturnValue(undefined);
+    parseScaleValue.mockReturnValue(1 / 255);
+    getProviderCandidates.mockReturnValue([["wasm"]]);
+    createSession.mockResolvedValue({
+      session: {
+        inputNames: ["input"],
+        outputNames: ["output"],
+        run: vi.fn().mockResolvedValue({
+          output: {
+            dims: [1, 1, 4, 8],
+            data: new Float32Array(32).fill(0.9)
+          }
+        })
+      },
+      provider: "wasm"
+    });
+    releaseSessions.mockResolvedValue(undefined);
+    toBgrFloatCHWFromBgr.mockReturnValue(new Float32Array(3 * 32 * 64).fill(1));
+    getMiniBoxFromPoints
+      .mockReturnValueOnce({
+        side: 4,
+        box: [
+          [0, 0],
+          [4, 0],
+          [4, 2],
+          [0, 2]
+        ]
+      })
+      .mockReturnValueOnce({
+        side: 6,
+        box: [
+          [0, 0],
+          [5, 0],
+          [5, 3],
+          [0, 3]
+        ]
+      });
+    boxScoreFast.mockReturnValue(0.9);
+    unclip.mockReturnValue([
+      [0, 0],
+      [5, 0],
+      [5, 3],
+      [0, 3]
+    ]);
+
+    const { createDetModel, createDetModelSession } = await loadDetModule();
+    const sessionState = await createDetModelSession({}, new Uint8Array([1]), "auto", {
+      available: false,
+      reason: ""
+    });
+    expect(getProviderCandidates).toHaveBeenCalledWith("auto", { available: false, reason: "" });
+    expect(withTimeout).toHaveBeenCalled();
+    expect(sessionState.provider).toBe("wasm");
+
+    const model = await createDetModel({
+      ort: {
+        Tensor: createMockOrtTensorClass()
+      },
+      modelBytes: new Uint8Array([1]),
+      configText: "config",
+      backend: "auto",
+      webgpuState: { available: false, reason: "" }
+    });
+
+    expect(assertModelResources).toHaveBeenCalled();
+    expect(model.kind).toBe("det");
+    expect(model.provider).toBe("wasm");
+    await expect(
+      model.predict(
+        createDetModelIntegrationCv(),
+        [
+          {
+            cols: 64,
+            rows: 32,
+            channels: () => 3
+          }
+        ],
+        {
+          thresh: 0.3,
+          boxThresh: 0.5,
+          unclipRatio: 1.5
+        }
+      )
+    ).resolves.toMatchObject([
+      {
+        boxes: expect.any(Array),
+        srcW: 64,
+        srcH: 32
+      }
+    ]);
+    await expect(model.dispose()).resolves.toBeUndefined();
+    await expect(model.predict({}, [{}], {})).rejects.toThrow(/session is not initialized/i);
+  });
+});
diff --git a/paddleocr-js/packages/core/test/helpers/mock-ort-tensor.ts b/paddleocr-js/packages/core/test/helpers/mock-ort-tensor.ts
new file mode 100644
index 0000000..1b83b8f
--- /dev/null
+++ b/paddleocr-js/packages/core/test/helpers/mock-ort-tensor.ts
@@ -0,0 +1,11 @@
+/** Minimal ORT `Tensor` mock; optional `tensorCalls` records each `{ type, dims, size }`. */
+export function createMockOrtTensorClass(tensorCalls = null) {
+  return class Tensor {
+    constructor(type, data, dims) {
+      this.type = type;
+      this.data = data;
+      this.dims = dims;
+      tensorCalls?.push({ type, dims, size: data.length });
+    }
+  };
+}
diff --git a/paddleocr-js/packages/core/test/model-asset.test.ts b/paddleocr-js/packages/core/test/model-asset.test.ts
new file mode 100644
index 0000000..dc9f8a9
--- /dev/null
+++ b/paddleocr-js/packages/core/test/model-asset.test.ts
@@ -0,0 +1,101 @@
+import { describe, expect, it } from "vitest";
+
+import {
+  normalizeModelAsset,
+  normalizeAssets,
+  getModelEntryPath,
+  assertModelResourceSlot,
+  assertModelResources
+} from "../src/resources/model-asset";
+
+describe("model asset normalization", () => {
+  it("resolves built-in model asset references", () => {
+    const assets = normalizeAssets({
+      det: "PP-OCRv5_mobile_det",
+      rec: "PP-OCRv5_mobile_rec"
+    });
+    expect(assets.det.url).toMatch(/\.tar$/);
+  });
+
+  it("resolves PP-OCRv6 tiny and small model asset references", () => {
+    const tiny = normalizeAssets({
+      det: "PP-OCRv6_tiny_det",
+      rec: "PP-OCRv6_tiny_rec"
+    });
+    expect(tiny.det.url).toMatch(/PP-OCRv6_tiny_det_onnx_infer\.tar$/);
+    expect(tiny.rec.url).toMatch(/PP-OCRv6_tiny_rec_onnx_infer\.tar$/);
+
+    const small = normalizeAssets({
+      det: "PP-OCRv6_small_det",
+      rec: "PP-OCRv6_small_rec"
+    });
+    expect(small.det.url).toMatch(/PP-OCRv6_small_det_onnx_infer\.tar$/);
+    expect(small.rec.url).toMatch(/PP-OCRv6_small_rec_onnx_infer\.tar$/);
+  });
+
+  it("normalizes a single model asset directly", () => {
+    const asset = normalizeModelAsset("det", {
+      url: "/det.tar"
+    });
+
+    expect(asset.url).toBe("/det.tar");
+  });
+
+  it("rejects invalid assets", () => {
+    expect(() =>
+      normalizeAssets({
+        encoder: {}
+      })
+    ).toThrow(/must define url/i);
+  });
+
+  it("rejects non-object asset descriptors", () => {
+    expect(() => normalizeModelAsset("det", null)).toThrow(/must be an object/i);
+  });
+
+  it("rejects unknown model asset references", () => {
+    expect(() =>
+      normalizeAssets({
+        det: "missing_model"
+      })
+    ).toThrow(/unknown model asset/i);
+  });
+});
+
+describe("model resource validation", () => {
+  it("provides standard entry names", () => {
+    expect(getModelEntryPath("model")).toBe("inference.onnx");
+    expect(getModelEntryPath("config")).toBe("inference.yml");
+    expect(getModelEntryPath("other")).toBe(null);
+  });
+
+  it("rejects missing model binary resources", () => {
+    expect(() => assertModelResourceSlot("Detection", "model", new Uint8Array())).toThrow(
+      /inference\.onnx/i
+    );
+  });
+
+  it("rejects missing model config resources", () => {
+    expect(() => assertModelResourceSlot("Recognition", "config", "")).toThrow(/inference\.yml/i);
+  });
+
+  it("supports validating multiple model resources together", () => {
+    expect(() =>
+      assertModelResources("Detection", {
+        model: new Uint8Array([1]),
+        config: "Global:\n  model_name: det"
+      })
+    ).not.toThrow();
+  });
+
+  it("rejects unsupported model resource slots", () => {
+    expect(() => assertModelResourceSlot("Detection", "labels", "abc")).toThrow(
+      /Unsupported model resource slot/i
+    );
+    expect(() =>
+      assertModelResources("Detection", {
+        labels: "abc"
+      })
+    ).toThrow(/Unsupported model resource slot/i);
+  });
+});
diff --git a/paddleocr-js/packages/core/test/model-common.test.ts b/paddleocr-js/packages/core/test/model-common.test.ts
new file mode 100644
index 0000000..b442696
--- /dev/null
+++ b/paddleocr-js/packages/core/test/model-common.test.ts
@@ -0,0 +1,174 @@
+import { describe, expect, it } from "vitest";
+
+import {
+  boxScoreFast,
+  extractInferenceModelName,
+  getMiniBoxFromPoints,
+  getTransformOp,
+  parseInferenceConfigText,
+  parseScaleValue,
+  toBgrFloatCHWFromBgr,
+  unclip
+} from "../src/models/common";
+
+describe("model common helpers", () => {
+  it("parses inference config text into plain objects", () => {
+    expect(parseInferenceConfigText("Global:\n  model_name: det")).toEqual({
+      Global: {
+        model_name: "det"
+      }
+    });
+    expect(parseInferenceConfigText("- item")).toEqual({});
+  });
+
+  it("parses numeric and fractional scale values with fallback behavior", () => {
+    expect(parseScaleValue(0.5, 1)).toBe(0.5);
+    expect(parseScaleValue("1./255.", 1)).toBeCloseTo(1 / 255);
+    expect(parseScaleValue("2/0", 3)).toBe(3);
+    expect(parseScaleValue({}, 4)).toBe(4);
+  });
+
+  it("finds transform ops and nested model names", () => {
+    expect(
+      getTransformOp([{ NormalizeImage: { mean: [0.5] } }, { Other: {} }], "NormalizeImage")
+    ).toEqual({ mean: [0.5] });
+    expect(getTransformOp([{ Other: {} }], "NormalizeImage")).toBeNull();
+
+    expect(extractInferenceModelName("Global:\n  model_name: det")).toBe("det");
+    expect(
+      extractInferenceModelName(
+        "Outer:\n  nested:\n    - item: 1\n    - model_name: rec\narray:\n  - another: thing"
+      )
+    ).toBe("rec");
+  });
+
+  it("converts BGR image data into CHW float tensors", () => {
+    const bgr = new Uint8Array([10, 20, 30, 40, 50, 60]);
+    const result = toBgrFloatCHWFromBgr(bgr, 2, 1, {
+      mean: [0, 0, 0],
+      std: [1, 1, 1],
+      scale: 1
+    });
+
+    expect(Array.from(result)).toEqual([10, 40, 20, 50, 30, 60]);
+  });
+
+  it("returns null when unclipping a degenerate polygon and expands valid polygons", () => {
+    expect(
+      unclip(
+        [
+          [0, 0],
+          [0, 0],
+          [0, 0]
+        ],
+        2
+      )
+    ).toBeNull();
+
+    const expanded = unclip(
+      [
+        [0, 0],
+        [10, 0],
+        [10, 10],
+        [0, 10]
+      ],
+      2
+    );
+
+    expect(expanded).not.toBeNull();
+    expect(expanded.length).toBeGreaterThanOrEqual(4);
+  });
+
+  it("computes mini boxes from points with a cv facade", () => {
+    const contour = {
+      deleteCalled: false,
+      delete() {
+        this.deleteCalled = true;
+      }
+    };
+    const cv = {
+      CV_32FC2: "float",
+      matFromArray: () => contour,
+      minAreaRect: () => ({ id: "rect" }),
+      RotatedRect: {
+        points: () => [
+          { x: 0, y: 0 },
+          { x: 10, y: 0 },
+          { x: 10, y: 5 },
+          { x: 0, y: 5 }
+        ]
+      }
+    };
+
+    const result = getMiniBoxFromPoints(cv, [
+      [0, 0],
+      [10, 0],
+      [10, 5],
+      [0, 5]
+    ]);
+
+    expect(result.box).toEqual([
+      [0, 0],
+      [10, 0],
+      [10, 5],
+      [0, 5]
+    ]);
+    expect(result.side).toBe(5);
+    expect(contour.deleteCalled).toBe(true);
+  });
+
+  it("computes masked box scores with a cv facade", () => {
+    const roi = {
+      delete: () => {}
+    };
+    const mask = {
+      delete: () => {}
+    };
+    const pts = {
+      delete: () => {}
+    };
+    const predMat = {
+      rows: 20,
+      cols: 30,
+      roi: () => roi
+    };
+    const cv = {
+      Rect: class Rect {
+        constructor(x, y, width, height) {
+          this.x = x;
+          this.y = y;
+          this.width = width;
+          this.height = height;
+        }
+      },
+      Mat: {
+        zeros: function zeros() {
+          return mask;
+        }
+      },
+      CV_8UC1: "mask",
+      CV_32SC2: "int",
+      MatVector: class MatVector {
+        push_back() {}
+        delete() {}
+      },
+      Scalar: class Scalar {
+        constructor(value) {
+          this.value = value;
+        }
+      },
+      matFromArray: () => pts,
+      fillPoly: () => {},
+      mean: () => [0.75]
+    };
+
+    expect(
+      boxScoreFast(cv, predMat, [
+        [1.2, 2.1],
+        [8.5, 2.4],
+        [8.1, 9.2],
+        [1.4, 9.8]
+      ])
+    ).toBe(0.75);
+  });
+});
diff --git a/paddleocr-js/packages/core/test/model-config.test.ts b/paddleocr-js/packages/core/test/model-config.test.ts
new file mode 100644
index 0000000..1827255
--- /dev/null
+++ b/paddleocr-js/packages/core/test/model-config.test.ts
@@ -0,0 +1,57 @@
+import { describe, expect, it } from "vitest";
+
+import { parseDetModelConfigText, parseRecModelConfigText } from "../src/models/index";
+
+const detConfig = `
+PreProcess:
+  transform_ops:
+    - DetResizeForTest:
+        resize_long: 736
+        limit_type: max
+        max_side_limit: 4500
+    - NormalizeImage:
+        mean: [0.1, 0.2, 0.3]
+        std: [0.9, 0.8, 0.7]
+        scale: 1./255.
+PostProcess:
+  thresh: 0.22
+  box_thresh: 0.55
+  max_candidates: 200
+  unclip_ratio: 1.8
+`;
+
+const recConfig = `
+PreProcess:
+  transform_ops:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - NormalizeImage:
+        mean: [0.5, 0.5, 0.5]
+        std: [0.5, 0.5, 0.5]
+        scale: 1./255.
+PostProcess:
+  character_dict:
+    - a
+    - b
+    - c
+`;
+
+describe("model config parsers", () => {
+  it("parses detection model config text", () => {
+    const config = parseDetModelConfigText(detConfig);
+
+    expect(config.resizeLong).toBe(736);
+    expect(config.limitType).toBe("max");
+    expect(config.maxSideLimit).toBe(4500);
+    expect(config.postprocess.thresh).toBe(0.22);
+    expect(config.postprocess.boxThresh).toBe(0.55);
+  });
+
+  it("parses recognition model config text", () => {
+    const config = parseRecModelConfigText(recConfig);
+
+    expect(config.imageShape).toEqual([3, 48, 320]);
+    expect(config.charDict.slice(0, 3)).toEqual(["a", "b", "c"]);
+    expect(config.charDict.at(-1)).toBe(" ");
+  });
+});
diff --git a/paddleocr-js/packages/core/test/ocr-api.test.ts b/paddleocr-js/packages/core/test/ocr-api.test.ts
new file mode 100644
index 0000000..4d933e1
--- /dev/null
+++ b/paddleocr-js/packages/core/test/ocr-api.test.ts
@@ -0,0 +1,564 @@
+import { describe, expect, it, vi } from "vitest";
+
+vi.mock("@techstark/opencv-js", () => ({
+  default: {
+    Mat() {}
+  }
+}));
+
+import { PaddleOCR, normalizeOcrPipelineConfig, parseOcrPipelineConfigText } from "../src/index";
+import { extractInferenceModelName } from "../src/models/common";
+import { DEFAULT_OCR_PIPELINE_CONFIG_TEXT } from "../src/pipelines/ocr/default-config";
+import { normalizeOrtOptions } from "../src/pipelines/ocr/shared";
+import { getOcrRuntimeParams } from "../src/pipelines/ocr/runtime-params";
+
+const CREATE_WITHOUT_INIT = Object.freeze({
+  initialize: false
+});
+
+const IGNORE_UNSUPPORTED = Object.freeze({
+  initialize: false,
+  unsupportedBehavior: "ignore"
+});
+
+const pipelineConfigText = `
+pipeline_name: OCR
+use_doc_preprocessor: true
+use_textline_orientation: true
+SubModules:
+  TextDetection:
+    model_name: PP-OCRv5_mobile_det
+    limit_side_len: 64
+    limit_type: min
+    max_side_limit: 4000
+    thresh: 0.3
+    box_thresh: 0.6
+    unclip_ratio: 1.5
+  TextLineOrientation:
+    model_name: PP-LCNet_x1_0_textline_ori
+  TextRecognition:
+    model_name: PP-OCRv5_mobile_rec
+    score_thresh: 0.0
+`;
+
+const customDetAsset = {
+  url: "https://example.com/custom-det.tar"
+};
+
+const customRecAsset = {
+  url: "https://example.com/custom-rec.tar"
+};
+
+class MockWorker {
+  constructor(responder) {
+    this.responder = responder;
+    this.messages = [];
+    this.terminated = false;
+    this.onmessage = null;
+    this.onerror = null;
+  }
+
+  postMessage(message) {
+    this.messages.push(message);
+    queueMicrotask(() => {
+      if (this.terminated) return;
+      const response = this.responder(message);
+      if (!response) return;
+      this.onmessage?.({ data: response });
+    });
+  }
+
+  terminate() {
+    this.terminated = true;
+  }
+}
+
+function expectDefaultModelAssets(ocr) {
+  expect(ocr.options.pipelineConfig.assets.det?.url).toMatch(/PP-OCRv5_mobile_det/);
+  expect(ocr.options.pipelineConfig.assets.rec?.url).toMatch(/PP-OCRv5_mobile_rec/);
+}
+
+describe("PaddleOCR high-level API", () => {
+  it("parses and normalizes OCR pipeline configs", () => {
+    const parsed = parseOcrPipelineConfigText(pipelineConfigText);
+    const normalized = normalizeOcrPipelineConfig(parsed);
+
+    expect(normalized.pipelineName).toBe("OCR");
+    expect(normalized.modelSelection.textDetectionModelName).toBe("PP-OCRv5_mobile_det");
+    expect(normalized.modelSelection.textRecognitionModelName).toBe("PP-OCRv5_mobile_rec");
+    expect(normalized.modelSelection).not.toHaveProperty("detAsset");
+    expect(normalized.modelSelection).not.toHaveProperty("recAsset");
+    expect(normalized.runtimeDefaults.text_det_limit_side_len).toBe(64);
+    expect(normalized.runtimeDefaults.text_det_limit_type).toBe("min");
+    expect(normalized.warnings).toHaveLength(2);
+  });
+
+  it("keeps pipeline-declared custom assets separate from model selection", () => {
+    const normalized = normalizeOcrPipelineConfig({
+      pipeline_name: "OCR",
+      SubModules: {
+        TextDetection: {
+          model_name: "custom_det",
+          model_dir: customDetAsset
+        },
+        TextRecognition: {
+          model_name: "custom_rec",
+          model_dir: customRecAsset
+        }
+      }
+    });
+
+    expect(normalized.modelSelection.textDetectionModelName).toBe("custom_det");
+    expect(normalized.modelSelection.textRecognitionModelName).toBe("custom_rec");
+    expect(normalized.assets.det.url).toBe("https://example.com/custom-det.tar");
+    expect(normalized.assets.rec.url).toBe("https://example.com/custom-rec.tar");
+  });
+
+  it("creates an OCR instance from lang and ocrVersion", async () => {
+    const ocr = await PaddleOCR.create({
+      lang: "ch",
+      ocrVersion: "PP-OCRv5",
+      ...CREATE_WITHOUT_INIT
+    });
+
+    expect(ocr).toBeInstanceOf(PaddleOCR);
+    expectDefaultModelAssets(ocr);
+  });
+
+  it("keeps the same create API when worker mode is enabled", async () => {
+    const defaultOrt = normalizeOrtOptions();
+    const ocr = await PaddleOCR.create({
+      lang: "ch",
+      ocrVersion: "PP-OCRv5",
+      worker: true,
+      ...CREATE_WITHOUT_INIT
+    });
+
+    expect(typeof ocr.initialize).toBe("function");
+    expect(typeof ocr.predict).toBe("function");
+    expect(typeof ocr.dispose).toBe("function");
+    expect(ocr.options.pipelineConfig.assets.det?.url).toMatch(/PP-OCRv5_mobile_det/);
+    expect(ocr.options.ortOptions.backend).toBe(defaultOrt.backend);
+  });
+
+  it("uses the package OCR.yaml defaults when no pipeline config is passed", async () => {
+    const defaultPipeline = normalizeOcrPipelineConfig(DEFAULT_OCR_PIPELINE_CONFIG_TEXT);
+    const ocr = await PaddleOCR.create(CREATE_WITHOUT_INIT);
+
+    expect(ocr.options.pipelineConfig.assets.det?.url).toMatch(
+      new RegExp(defaultPipeline.modelSelection.textDetectionModelName)
+    );
+    expect(ocr.options.pipelineConfig.assets.rec?.url).toMatch(
+      new RegExp(defaultPipeline.modelSelection.textRecognitionModelName)
+    );
+    expect(ocr.pipelineConfig.runtimeDefaults).toMatchObject(defaultPipeline.runtimeDefaults);
+  });
+
+  it("maps English PP-OCRv5 selection to the mobile model set", async () => {
+    const ocr = await PaddleOCR.create({
+      lang: "en",
+      ocrVersion: "PP-OCRv5",
+      ...CREATE_WITHOUT_INIT
+    });
+
+    expectDefaultModelAssets(ocr);
+  });
+
+  it("maps PP-OCRv6 lang selection to the small model set", async () => {
+    const ocr = await PaddleOCR.create({
+      lang: "ch",
+      ocrVersion: "PP-OCRv6",
+      ...CREATE_WITHOUT_INIT
+    });
+
+    expect(ocr.options.pipelineConfig.assets.det?.url).toMatch(/PP-OCRv6_small_det/);
+    expect(ocr.options.pipelineConfig.assets.rec?.url).toMatch(/PP-OCRv6_small_rec/);
+    expect(ocr.options.pipelineConfig.modelSelection.textDetectionModelName).toBe(
+      "PP-OCRv6_small_det"
+    );
+    expect(ocr.options.pipelineConfig.modelSelection.textRecognitionModelName).toBe(
+      "PP-OCRv6_small_rec"
+    );
+  });
+
+  it("maps PP-OCRv6 Latin languages to the small model set", async () => {
+    const ocr = await PaddleOCR.create({
+      lang: "fr",
+      ocrVersion: "PP-OCRv6",
+      ...CREATE_WITHOUT_INIT
+    });
+
+    expect(ocr.options.pipelineConfig.assets.det?.url).toMatch(/PP-OCRv6_small_det/);
+    expect(ocr.options.pipelineConfig.assets.rec?.url).toMatch(/PP-OCRv6_small_rec/);
+  });
+
+  it("rejects PP-OCRv6 with languages outside the v6 coverage", async () => {
+    for (const lang of ["ru", "pi"]) {
+      await expect(
+        PaddleOCR.create({
+          lang,
+          ocrVersion: "PP-OCRv6",
+          ...CREATE_WITHOUT_INIT
+        })
+      ).rejects.toThrow(/Unsupported lang\/ocrVersion combination/);
+    }
+  });
+
+  it("maps PP-OCRv6 az and ku lang selection to the small model set", async () => {
+    for (const lang of ["az", "ku"]) {
+      const ocr = await PaddleOCR.create({
+        lang,
+        ocrVersion: "PP-OCRv6",
+        ...CREATE_WITHOUT_INIT
+      });
+
+      expect(ocr.options.pipelineConfig.assets.det?.url).toMatch(/PP-OCRv6_small_det/);
+      expect(ocr.options.pipelineConfig.assets.rec?.url).toMatch(/PP-OCRv6_small_rec/);
+    }
+  });
+
+  it("resolves PP-OCRv6 tiny models via explicit model names", async () => {
+    const ocr = await PaddleOCR.create({
+      text_detection_model_name: "PP-OCRv6_tiny_det",
+      text_recognition_model_name: "PP-OCRv6_tiny_rec",
+      ...CREATE_WITHOUT_INIT
+    });
+
+    expect(ocr.options.pipelineConfig.assets.det?.url).toMatch(/PP-OCRv6_tiny_det_onnx_infer\.tar$/);
+    expect(ocr.options.pipelineConfig.assets.rec?.url).toMatch(/PP-OCRv6_tiny_rec_onnx_infer\.tar$/);
+  });
+
+  it("allows overriding model selection via model_name options", async () => {
+    const ocr = await PaddleOCR.create({
+      text_detection_model_name: "PP-OCRv5_mobile_det",
+      text_recognition_model_name: "PP-OCRv5_mobile_rec",
+      ...CREATE_WITHOUT_INIT
+    });
+
+    expectDefaultModelAssets(ocr);
+  });
+
+  it("creates an OCR instance from pipeline config model names", async () => {
+    const ocr = await PaddleOCR.create({
+      pipelineConfig: pipelineConfigText,
+      ...IGNORE_UNSUPPORTED
+    });
+
+    expectDefaultModelAssets(ocr);
+    expect(ocr.pipelineConfig.runtimeDefaults.text_det_limit_type).toBe("min");
+    expect(ocr.pipelineConfig.runtimeDefaults.text_rec_score_thresh).toBe(0);
+  });
+
+  it("lets explicit model assets override pipeline config model names", async () => {
+    const ocr = await PaddleOCR.create({
+      pipelineConfig: pipelineConfigText,
+      text_detection_model_name: "custom_det",
+      textDetectionModelAsset: customDetAsset,
+      text_recognition_model_name: "custom_rec",
+      textRecognitionModelAsset: customRecAsset,
+      ...IGNORE_UNSUPPORTED
+    });
+
+    expect(ocr.options.pipelineConfig.assets.det?.url).toBe("https://example.com/custom-det.tar");
+    expect(ocr.options.pipelineConfig.assets.rec?.url).toBe("https://example.com/custom-rec.tar");
+  });
+
+  it("allows overriding only one side with a custom model asset", async () => {
+    const ocr = await PaddleOCR.create({
+      pipelineConfig: pipelineConfigText,
+      text_detection_model_name: "custom_det",
+      textDetectionModelAsset: customDetAsset,
+      ...IGNORE_UNSUPPORTED
+    });
+
+    expect(ocr.options.pipelineConfig.assets.det?.url).toBe("https://example.com/custom-det.tar");
+    expect(ocr.options.pipelineConfig.assets.rec?.url).toMatch(/PP-OCRv5_mobile_rec/);
+  });
+
+  it("lets explicit model names override pipeline config model assets", async () => {
+    const ocr = await PaddleOCR.create({
+      pipelineConfig: {
+        pipeline_name: "OCR",
+        SubModules: {
+          TextDetection: {
+            model_name: "custom_det",
+            model_dir: customDetAsset
+          },
+          TextRecognition: {
+            model_name: "custom_rec",
+            model_dir: customRecAsset
+          }
+        }
+      },
+      text_detection_model_name: "PP-OCRv5_mobile_det",
+      text_recognition_model_name: "PP-OCRv5_mobile_rec",
+      ...CREATE_WITHOUT_INIT
+    });
+
+    expectDefaultModelAssets(ocr);
+  });
+
+  it("rejects unsupported lang/ocrVersion combinations", async () => {
+    await expect(
+      PaddleOCR.create({
+        lang: "kl",
+        ocrVersion: "PP-OCRv5",
+        ...CREATE_WITHOUT_INIT
+      })
+    ).rejects.toThrow(/Unsupported lang\/ocrVersion combination/i);
+  });
+
+  it("warns about unsupported pipeline features by default", async () => {
+    const warn = vi.spyOn(console, "warn").mockImplementation(() => {});
+    await PaddleOCR.create({ pipelineConfig: pipelineConfigText, ...CREATE_WITHOUT_INIT });
+
+    expect(warn).toHaveBeenCalled();
+    warn.mockRestore();
+  });
+
+  it("lets direct pipeline runtime parameters override pipelineConfig defaults", async () => {
+    const ocr = await PaddleOCR.create({
+      pipelineConfig: pipelineConfigText,
+      text_det_limit_side_len: 200,
+      text_rec_score_thresh: 0.42,
+      ...IGNORE_UNSUPPORTED
+    });
+
+    expect(ocr.pipelineConfig.runtimeDefaults.text_det_limit_side_len).toBe(200);
+    expect(ocr.pipelineConfig.runtimeDefaults.text_rec_score_thresh).toBe(0.42);
+  });
+
+  it("lets direct batch sizes override pipelineConfig", async () => {
+    const ocr = await PaddleOCR.create({
+      pipelineConfig: pipelineConfigText,
+      textRecognitionBatchSize: 2,
+      text_detection_batch_size: 3,
+      ...IGNORE_UNSUPPORTED
+    });
+
+    expect(ocr.pipelineConfig.textRecognitionBatchSize).toBe(2);
+    expect(ocr.pipelineConfig.textDetectionBatchSize).toBe(3);
+  });
+
+  it("applies textRecognitionBatchSize without user pipelineConfig", async () => {
+    const ocr = await PaddleOCR.create({
+      lang: "ch",
+      ocrVersion: "PP-OCRv5",
+      text_recognition_batch_size: 4,
+      ...CREATE_WITHOUT_INIT
+    });
+
+    expect(ocr.pipelineConfig.textRecognitionBatchSize).toBe(4);
+  });
+
+  it("can turn unsupported pipeline warnings into errors", async () => {
+    await expect(
+      PaddleOCR.create({
+        pipelineConfig: pipelineConfigText,
+        ...CREATE_WITHOUT_INIT,
+        unsupportedBehavior: "error"
+      })
+    ).rejects.toThrow(/not yet supported/i);
+  });
+
+  it("parses the generated default OCR pipeline config text", () => {
+    const parsed = parseOcrPipelineConfigText(DEFAULT_OCR_PIPELINE_CONFIG_TEXT);
+    const normalized = normalizeOcrPipelineConfig(DEFAULT_OCR_PIPELINE_CONFIG_TEXT);
+
+    expect(normalized).toMatchObject(normalizeOcrPipelineConfig(parsed));
+  });
+
+  it("rejects custom model_dir without matching model_name", async () => {
+    await expect(
+      PaddleOCR.create({
+        textDetectionModelAsset: customDetAsset,
+        ...CREATE_WITHOUT_INIT
+      })
+    ).rejects.toThrow(/text_detection_model_dir requires text_detection_model_name/i);
+  });
+
+  it("rejects conflicting model asset aliases", async () => {
+    await expect(
+      PaddleOCR.create({
+        text_detection_model_name: "custom_det",
+        textDetectionModelAsset: customDetAsset,
+        text_detection_model_dir: customRecAsset,
+        ...CREATE_WITHOUT_INIT
+      })
+    ).rejects.toThrow(/Conflicting values provided for text detection model asset/i);
+  });
+
+  it("ignores unsupported asset metadata before inference.yml is loaded", async () => {
+    const ocr = await PaddleOCR.create({
+      text_detection_model_name: "custom_det",
+      textDetectionModelAsset: {
+        ...customDetAsset,
+        model_name: "other_det"
+      },
+      text_recognition_model_name: "custom_rec",
+      textRecognitionModelAsset: customRecAsset,
+      ...CREATE_WITHOUT_INIT
+    });
+
+    expect(ocr.options.pipelineConfig.assets.det?.url).toBe("https://example.com/custom-det.tar");
+  });
+
+  it("initializes worker mode through the same API surface", async () => {
+    const worker = new MockWorker((message) => {
+      if (message.type === "init") {
+        return {
+          kind: "worker-transport-response",
+          status: "success",
+          requestId: message.requestId,
+          payload: {
+            summary: {
+              backend: "wasm",
+              webgpuAvailable: false,
+              detProvider: "wasm",
+              recProvider: "wasm",
+              assets: [],
+              elapsedMs: 12,
+              pipelineConfigWarnings: []
+            },
+            modelConfig: {
+              det: { resizeLong: 960 },
+              rec: { imageShape: [3, 48, 320] }
+            }
+          }
+        };
+      }
+      if (message.type === "dispose") {
+        return {
+          kind: "worker-transport-response",
+          status: "success",
+          requestId: message.requestId,
+          payload: {}
+        };
+      }
+      return null;
+    });
+    const ocr = await PaddleOCR.create({
+      worker: {
+        createWorker: () => worker
+      },
+      ortOptions: {
+        backend: "wasm",
+        proxy: true
+      },
+      ...CREATE_WITHOUT_INIT
+    });
+
+    const summary = await ocr.initialize();
+
+    expect(summary.backend).toBe("wasm");
+    expect(worker.messages[0].type).toBe("init");
+    expect(worker.messages[0].payload.options.ortOptions.backend).toBe("wasm");
+    expect(worker.messages[0].payload.options.ortOptions.disableWasmProxy).toBe(true);
+    expect(ocr.getModelConfig().det.resizeLong).toBe(960);
+
+    await ocr.dispose();
+    expect(worker.messages[1].type).toBe("dispose");
+    expect(worker.terminated).toBe(true);
+  });
+
+  it("surfaces worker initialization failures", async () => {
+    const worker = new MockWorker((message) => {
+      if (message.type === "init") {
+        return {
+          kind: "worker-transport-response",
+          status: "error",
+          requestId: message.requestId,
+          error: {
+            name: "Error",
+            message: "worker init failed"
+          }
+        };
+      }
+      return null;
+    });
+    const ocr = await PaddleOCR.create({
+      worker: {
+        createWorker: () => worker
+      },
+      ...CREATE_WITHOUT_INIT
+    });
+
+    await expect(ocr.initialize()).rejects.toThrow(/worker init failed/i);
+    expect(worker.terminated).toBe(true);
+  });
+});
+
+describe("OCR runtime parameter normalization", () => {
+  it("accepts camelCase aliases while preserving PaddleOCR names", () => {
+    const params = getOcrRuntimeParams(
+      {
+        det: {
+          resizeLong: 960,
+          limitType: "max",
+          maxSideLimit: 4000,
+          normalize: {
+            mean: [0.485, 0.456, 0.406],
+            std: [0.229, 0.224, 0.225],
+            scale: 1 / 255
+          },
+          postprocess: {
+            thresh: 0.3,
+            boxThresh: 0.6,
+            unclipRatio: 1.5
+          }
+        },
+        rec: {
+          charDict: [],
+          imageShape: [3, 48, 320]
+        }
+      },
+      {
+        text_det_limit_type: "min"
+      },
+      {
+        textDetThresh: 0.4,
+        textDetBoxThresh: 0.7,
+        textDetUnclipRatio: 2,
+        textRecScoreThresh: 0.2
+      }
+    );
+
+    expect(params.det.limitType).toBe("min");
+    expect(params.det.thresh).toBe(0.4);
+    expect(params.det.boxThresh).toBe(0.7);
+    expect(params.det.unclipRatio).toBe(2);
+    expect(params.pipeline.scoreThresh).toBe(0.2);
+  });
+});
+
+describe("inference.yml model_name extraction", () => {
+  it("prefers Global.model_name when present", () => {
+    expect(
+      extractInferenceModelName(`
+Global:
+  model_name: custom_det
+PostProcess:
+  name: DBPostProcess
+`)
+    ).toBe("custom_det");
+  });
+
+  it("falls back to nested model_name fields", () => {
+    expect(
+      extractInferenceModelName(`
+Deploy:
+  metadata:
+    model_name: custom_rec
+`)
+    ).toBe("custom_rec");
+  });
+
+  it("returns null when inference.yml has no model_name", () => {
+    expect(
+      extractInferenceModelName(`
+PostProcess:
+  name: DBPostProcess
+`)
+    ).toBeNull();
+  });
+});
diff --git a/paddleocr-js/packages/core/test/ocr-config-branches.test.ts b/paddleocr-js/packages/core/test/ocr-config-branches.test.ts
new file mode 100644
index 0000000..937d7e1
--- /dev/null
+++ b/paddleocr-js/packages/core/test/ocr-config-branches.test.ts
@@ -0,0 +1,98 @@
+import { describe, expect, it } from "vitest";
+
+import {
+  normalizeOcrPipelineConfig,
+  parseOcrPipelineConfigText
+} from "../src/pipelines/ocr/config";
+
+describe("OCR pipeline config branches", () => {
+  it("rejects non-object YAML payloads and invalid input types", () => {
+    expect(() => parseOcrPipelineConfigText("- item")).toThrow(/must decode to an object/i);
+    expect(() => normalizeOcrPipelineConfig(123)).toThrow(/must be an object or YAML text/i);
+  });
+
+  it("rejects unsupported pipeline names", () => {
+    expect(() =>
+      normalizeOcrPipelineConfig({
+        pipeline_name: "DET_ONLY",
+        SubModules: {
+          TextDetection: { model_name: "PP-OCRv5_mobile_det" },
+          TextRecognition: { model_name: "PP-OCRv5_mobile_rec" }
+        }
+      })
+    ).toThrow(/Unsupported pipeline_name/i);
+  });
+
+  it("accepts default pipeline names and suppresses general text_type warnings", () => {
+    const normalized = normalizeOcrPipelineConfig({
+      SubModules: {
+        TextDetection: { model_name: "PP-OCRv5_mobile_det" },
+        TextRecognition: { model_name: "PP-OCRv5_mobile_rec" }
+      },
+      text_type: "general"
+    });
+
+    expect(normalized.pipelineName).toBe("OCR");
+    expect(normalized.warnings).toEqual([]);
+    expect(normalized.unsupportedFeatures).toEqual([]);
+  });
+
+  it("supports object model_dir assets and warns on unsupported text_type values", () => {
+    const normalized = normalizeOcrPipelineConfig({
+      SubModules: {
+        TextDetection: {
+          model_name: "custom_det",
+          model_dir: {
+            url: "/det.tar"
+          },
+          limit_side_len: "",
+          max_side_limit: null
+        },
+        TextRecognition: {
+          model_name: "custom_rec",
+          score_thresh: "0.7"
+        }
+      },
+      text_type: "seal"
+    });
+
+    expect(normalized.assets.det).toMatchObject({
+      url: "/det.tar"
+    });
+    expect(normalized.runtimeDefaults.text_det_limit_side_len).toBeUndefined();
+    expect(normalized.runtimeDefaults.text_det_max_side_limit).toBeUndefined();
+    expect(normalized.runtimeDefaults.text_rec_score_thresh).toBe(0.7);
+    expect(normalized.warnings).toContain('text_type "seal" is not used by PaddleOCR.js yet.');
+  });
+
+  it("rejects invalid model_dir values and missing model names", () => {
+    expect(() =>
+      normalizeOcrPipelineConfig({
+        SubModules: {
+          TextDetection: {
+            model_dir: {
+              url: "/det.tar"
+            }
+          },
+          TextRecognition: {
+            model_name: "PP-OCRv5_mobile_rec"
+          }
+        }
+      })
+    ).toThrow(/model_name must be provided/i);
+
+    expect(() =>
+      normalizeOcrPipelineConfig({
+        SubModules: {
+          TextDetection: {
+            model_name: "PP-OCRv5_mobile_det",
+            model_dir: "/det"
+          },
+          TextRecognition: {
+            model_name: "PP-OCRv5_mobile_rec"
+          }
+        }
+      })
+    ).toThrow(/model_dir must be null or an asset descriptor object/i);
+  });
+});
diff --git a/paddleocr-js/packages/core/test/ocr-core.test.ts b/paddleocr-js/packages/core/test/ocr-core.test.ts
new file mode 100644
index 0000000..e44356f
--- /dev/null
+++ b/paddleocr-js/packages/core/test/ocr-core.test.ts
@@ -0,0 +1,518 @@
+import { afterEach, describe, expect, it, vi } from "vitest";
+
+const loadModelAsset = vi.fn();
+const createDetModel = vi.fn();
+const createRecModel = vi.fn();
+const cropByPoly = vi.fn();
+const initOpenCvRuntime = vi.fn();
+const initOrtRuntime = vi.fn();
+const nowMs = vi.fn();
+const getOcrRuntimeParams = vi.fn();
+const cloneDefaultOcrConfig = vi.fn();
+const validateLoadedModelName = vi.fn();
+
+vi.mock("../src/resources/index", () => ({
+  loadModelAsset
+}));
+
+vi.mock("../src/models/index", () => ({
+  createDetModel,
+  createRecModel
+}));
+
+vi.mock("../src/pipelines/ocr/crop", () => ({
+  cropByPoly
+}));
+
+vi.mock("../src/runtime/opencv", () => ({
+  initOpenCvRuntime
+}));
+
+vi.mock("../src/runtime/ort", () => ({
+  initOrtRuntime
+}));
+
+vi.mock("../src/utils/common", async (importOriginal) => {
+  const actual = await importOriginal<typeof import("../src/utils/common")>();
+  return {
+    ...actual,
+    nowMs
+  };
+});
+
+vi.mock("../src/pipelines/ocr/runtime-params", () => ({
+  getOcrRuntimeParams
+}));
+
+vi.mock("../src/pipelines/ocr/shared", () => ({
+  cloneDefaultOcrConfig,
+  validateLoadedModelName
+}));
+
+afterEach(() => {
+  vi.resetModules();
+  vi.clearAllMocks();
+});
+
+const AUTO_ORT_OPTIONS = Object.freeze({
+  backend: "auto"
+});
+
+function createResolvedAssets() {
+  return {
+    det: { url: "/det.tar" },
+    rec: { url: "/rec.tar" }
+  };
+}
+
+function minimalPipelineConfig(overrides: Record<string, unknown> = {}) {
+  return {
+    pipelineName: "OCR",
+    raw: {},
+    warnings: [] as string[],
+    unsupportedFeatures: [] as string[],
+    modelSelection: {
+      textDetectionModelName: "det-name",
+      textRecognitionModelName: "rec-name"
+    },
+    assets: createResolvedAssets(),
+    runtimeDefaults: {} as Record<string, unknown>,
+    pipelineBatchSize: 1,
+    textDetectionBatchSize: 1,
+    textRecognitionBatchSize: 1,
+    ...overrides
+  };
+}
+
+function mockEmptyDefaultOcrConfig() {
+  cloneDefaultOcrConfig.mockReturnValue({ det: {}, rec: {} });
+}
+
+async function loadCoreModule() {
+  return import("../src/pipelines/ocr/core");
+}
+
+describe("OCR pipeline core", () => {
+  it("initializes OpenCV and ORT, loads assets, and creates models", async () => {
+    const cv = { name: "cv" };
+    const ort = { name: "ort" };
+    const detModel = { config: { det: true }, provider: "wasm", dispose: vi.fn() };
+    const recModel = { config: { rec: true }, provider: "webgpu", dispose: vi.fn() };
+
+    cloneDefaultOcrConfig.mockReturnValue({
+      det: { marker: "default-det-config" },
+      rec: { marker: "default-rec-config" }
+    });
+    nowMs.mockReturnValueOnce(100).mockReturnValueOnce(145);
+    initOpenCvRuntime.mockResolvedValue({ cv });
+    initOrtRuntime.mockResolvedValue({
+      ort,
+      webgpuState: { available: true, reason: "" },
+      backend: "auto"
+    });
+    loadModelAsset
+      .mockResolvedValueOnce({
+        modelBytes: new Uint8Array([1]),
+        configText: "det-config",
+        download: { url: "/det.tar", bytes: 100 }
+      })
+      .mockResolvedValueOnce({
+        modelBytes: new Uint8Array([2]),
+        configText: "rec-config",
+        download: { url: "/rec.tar", bytes: 200 }
+      });
+    createDetModel.mockResolvedValue(detModel);
+    createRecModel.mockResolvedValue(recModel);
+
+    const { OcrPipelineRunner } = await loadCoreModule();
+    const ensureServedFromHttp = vi.fn();
+    const runner = new OcrPipelineRunner({
+      pipelineConfig: minimalPipelineConfig({
+        warnings: ["warning"]
+      }),
+      ortOptions: AUTO_ORT_OPTIONS,
+      ensureServedFromHttp
+    });
+
+    const summary = await runner.initialize();
+
+    expect(ensureServedFromHttp).toHaveBeenCalledTimes(1);
+    expect(initOpenCvRuntime).toHaveBeenCalledTimes(1);
+    expect(initOrtRuntime).toHaveBeenCalledWith(AUTO_ORT_OPTIONS);
+    expect(loadModelAsset).toHaveBeenCalledTimes(2);
+    expect(validateLoadedModelName).toHaveBeenNthCalledWith(
+      1,
+      "TextDetection",
+      "det-name",
+      "det-config"
+    );
+    expect(validateLoadedModelName).toHaveBeenNthCalledWith(
+      2,
+      "TextRecognition",
+      "rec-name",
+      "rec-config"
+    );
+    expect(createDetModel).toHaveBeenCalledWith({
+      ort,
+      modelBytes: new Uint8Array([1]),
+      configText: "det-config",
+      backend: AUTO_ORT_OPTIONS.backend,
+      webgpuState: { available: true, reason: "" },
+      batchSize: 1
+    });
+    expect(createRecModel).toHaveBeenCalledWith({
+      ort,
+      modelBytes: new Uint8Array([2]),
+      configText: "rec-config",
+      backend: AUTO_ORT_OPTIONS.backend,
+      webgpuState: { available: true, reason: "" },
+      batchSize: 1
+    });
+    expect(summary).toEqual({
+      backend: AUTO_ORT_OPTIONS.backend,
+      webgpuAvailable: true,
+      detProvider: "wasm",
+      recProvider: "webgpu",
+      assets: [
+        { url: "/det.tar", bytes: 100 },
+        { url: "/rec.tar", bytes: 200 }
+      ],
+      elapsedMs: 45,
+      pipelineConfigWarnings: ["warning"]
+    });
+    expect(runner.getInitializationSummary()).toEqual(summary);
+    expect(runner.getModelConfig()).toEqual({
+      det: { det: true },
+      rec: { rec: true }
+    });
+  });
+
+  it("rejects initialization when assets are not pre-resolved", async () => {
+    mockEmptyDefaultOcrConfig();
+    initOpenCvRuntime.mockResolvedValue({ cv: {} });
+    initOrtRuntime.mockResolvedValue({
+      ort: {},
+      webgpuState: { available: false, reason: "" },
+      backend: "wasm"
+    });
+
+    const { OcrPipelineRunner } = await loadCoreModule();
+    const runner = new OcrPipelineRunner({
+      pipelineConfig: minimalPipelineConfig({
+        assets: {
+          det: null,
+          rec: { id: "rec" }
+        }
+      })
+    });
+
+    await expect(runner.initialize()).rejects.toThrow(
+      /requires pre-resolved detection and recognition asset/i
+    );
+  });
+
+  it("predicts OCR results and filters by score threshold", async () => {
+    const cv = { name: "cv" };
+    const sourceMat = { delete: vi.fn() };
+    const sourceImage = {
+      width: 640,
+      height: 480,
+      mat: sourceMat,
+      dispose: vi.fn()
+    };
+    const cropA = { delete: vi.fn() };
+    const cropB = { delete: vi.fn() };
+    const detModel = {
+      provider: "wasm",
+      predict: vi
+        .fn()
+        .mockResolvedValue([
+          { boxes: [{ poly: [[1, 1]] }, { poly: [[2, 2]] }], srcW: 640, srcH: 480 }
+        ]),
+      dispose: vi.fn()
+    };
+    const recModel = {
+      provider: "wasm",
+      predict: vi.fn().mockResolvedValue([
+        { text: "high", score: 0.95 },
+        { text: "low", score: 0.4 }
+      ]),
+      dispose: vi.fn()
+    };
+
+    mockEmptyDefaultOcrConfig();
+    getOcrRuntimeParams.mockReturnValue({
+      det: {},
+      pipeline: { scoreThresh: 0.5 }
+    });
+    cropByPoly.mockReturnValueOnce(cropA).mockReturnValueOnce(cropB);
+    nowMs
+      .mockReturnValueOnce(10)
+      .mockReturnValueOnce(20)
+      .mockReturnValueOnce(30)
+      .mockReturnValueOnce(40)
+      .mockReturnValueOnce(60)
+      .mockReturnValueOnce(70);
+
+    const { OcrPipelineRunner } = await loadCoreModule();
+    const runner = new OcrPipelineRunner({
+      pipelineConfig: minimalPipelineConfig({
+        runtimeDefaults: { text_det_limit_side_len: 64 }
+      }),
+      ortOptions: AUTO_ORT_OPTIONS,
+      sourceToMat: vi.fn().mockResolvedValue(sourceImage)
+    });
+    runner.cv = cv;
+    runner.ort = { name: "ort" };
+    runner.detModel = detModel;
+    runner.recModel = recModel;
+    runner.webgpuState = { available: false, reason: "" };
+    runner.modelConfig = { det: { conf: true }, rec: { conf: true } };
+
+    const result = await runner.predict({ kind: "blob" }, { text_rec_score_thresh: 0.8 });
+
+    expect(getOcrRuntimeParams).toHaveBeenCalledWith(
+      { det: { conf: true }, rec: { conf: true } },
+      { text_det_limit_side_len: 64 },
+      { text_rec_score_thresh: 0.8 }
+    );
+    expect(detModel.predict).toHaveBeenCalledWith(cv, [sourceMat], {});
+    expect(cropByPoly).toHaveBeenNthCalledWith(1, cv, sourceMat, [[1, 1]]);
+    expect(cropByPoly).toHaveBeenNthCalledWith(2, cv, sourceMat, [[2, 2]]);
+    expect(recModel.predict).toHaveBeenCalledWith(cv, [cropA, cropB]);
+    expect(cropA.delete).toHaveBeenCalledTimes(1);
+    expect(cropB.delete).toHaveBeenCalledTimes(1);
+    expect(sourceImage.dispose).toHaveBeenCalledTimes(1);
+    expect(result).toEqual([
+      {
+        image: { width: 640, height: 480 },
+        items: [{ poly: [[1, 1]], text: "high", score: 0.95 }],
+        metrics: {
+          detMs: 10,
+          recMs: 20,
+          totalMs: 60,
+          detectedBoxes: 2,
+          recognizedCount: 1
+        },
+        runtime: {
+          requestedBackend: AUTO_ORT_OPTIONS.backend,
+          detProvider: "wasm",
+          recProvider: "wasm",
+          webgpuAvailable: false
+        }
+      }
+    ]);
+  });
+
+  it("returns one OCR result per source when predict receives an array of inputs", async () => {
+    const cv = { name: "cv" };
+    const mat1 = { delete: vi.fn() };
+    const mat2 = { delete: vi.fn() };
+    const dispose1 = vi.fn();
+    const dispose2 = vi.fn();
+    const sourceImage1 = { width: 100, height: 100, mat: mat1, dispose: dispose1 };
+    const sourceImage2 = { width: 200, height: 200, mat: mat2, dispose: dispose2 };
+    const crop1 = { delete: vi.fn() };
+    const crop2 = { delete: vi.fn() };
+    const detModel = {
+      provider: "wasm",
+      predict: vi
+        .fn()
+        .mockResolvedValueOnce([{ boxes: [{ poly: [[1, 1]] }], srcW: 100, srcH: 100 }])
+        .mockResolvedValueOnce([{ boxes: [{ poly: [[2, 2]] }], srcW: 200, srcH: 200 }]),
+      dispose: vi.fn()
+    };
+    const recModel = {
+      provider: "wasm",
+      predict: vi
+        .fn()
+        .mockResolvedValueOnce([{ text: "a", score: 1 }])
+        .mockResolvedValueOnce([{ text: "b", score: 1 }]),
+      dispose: vi.fn()
+    };
+
+    mockEmptyDefaultOcrConfig();
+    getOcrRuntimeParams.mockReturnValue({
+      det: {},
+      pipeline: { scoreThresh: 0 }
+    });
+    cropByPoly.mockReturnValueOnce(crop1).mockReturnValueOnce(crop2);
+    nowMs.mockReturnValue(0);
+
+    const { OcrPipelineRunner } = await loadCoreModule();
+    const runner = new OcrPipelineRunner({
+      pipelineConfig: minimalPipelineConfig(),
+      ortOptions: AUTO_ORT_OPTIONS,
+      sourceToMat: vi.fn().mockResolvedValueOnce(sourceImage1).mockResolvedValueOnce(sourceImage2)
+    });
+    runner.cv = cv;
+    runner.ort = { name: "ort" };
+    runner.detModel = detModel;
+    runner.recModel = recModel;
+    runner.webgpuState = { available: false, reason: "" };
+    runner.modelConfig = { det: { conf: true }, rec: { conf: true } };
+
+    const results = await runner.predict([{ kind: "a" }, { kind: "b" }], {});
+
+    expect(detModel.predict).toHaveBeenCalledTimes(2);
+    expect(detModel.predict).toHaveBeenNthCalledWith(1, cv, [mat1], {});
+    expect(detModel.predict).toHaveBeenNthCalledWith(2, cv, [mat2], {});
+    expect(recModel.predict).toHaveBeenNthCalledWith(1, cv, [crop1]);
+    expect(recModel.predict).toHaveBeenNthCalledWith(2, cv, [crop2]);
+    expect(results).toHaveLength(2);
+    expect(results[0].image).toEqual({ width: 100, height: 100 });
+    expect(results[0].items[0].text).toBe("a");
+    expect(results[1].image).toEqual({ width: 200, height: 200 });
+    expect(results[1].items[0].text).toBe("b");
+    expect(dispose1).toHaveBeenCalledTimes(1);
+    expect(dispose2).toHaveBeenCalledTimes(1);
+  });
+
+  it("passes multiple sources to det in one pipeline batch when pipelineBatchSize > 1", async () => {
+    const cv = { name: "cv" };
+    const mat1 = { delete: vi.fn() };
+    const mat2 = { delete: vi.fn() };
+    const dispose1 = vi.fn();
+    const dispose2 = vi.fn();
+    const sourceImage1 = { width: 100, height: 100, mat: mat1, dispose: dispose1 };
+    const sourceImage2 = { width: 200, height: 200, mat: mat2, dispose: dispose2 };
+    const crop1 = { delete: vi.fn() };
+    const crop2 = { delete: vi.fn() };
+    const detModel = {
+      provider: "wasm",
+      predict: vi.fn().mockResolvedValue([
+        { boxes: [{ poly: [[1, 1]] }], srcW: 100, srcH: 100 },
+        { boxes: [{ poly: [[2, 2]] }], srcW: 200, srcH: 200 }
+      ]),
+      dispose: vi.fn()
+    };
+    const recModel = {
+      provider: "wasm",
+      predict: vi
+        .fn()
+        .mockResolvedValueOnce([{ text: "a", score: 1 }])
+        .mockResolvedValueOnce([{ text: "b", score: 1 }]),
+      dispose: vi.fn()
+    };
+
+    mockEmptyDefaultOcrConfig();
+    getOcrRuntimeParams.mockReturnValue({
+      det: {},
+      pipeline: { scoreThresh: 0 }
+    });
+    cropByPoly.mockReturnValueOnce(crop1).mockReturnValueOnce(crop2);
+    nowMs.mockReturnValue(0);
+
+    const { OcrPipelineRunner } = await loadCoreModule();
+    const runner = new OcrPipelineRunner({
+      pipelineConfig: minimalPipelineConfig({ pipelineBatchSize: 2 }),
+      ortOptions: AUTO_ORT_OPTIONS,
+      sourceToMat: vi.fn().mockResolvedValueOnce(sourceImage1).mockResolvedValueOnce(sourceImage2)
+    });
+    runner.cv = cv;
+    runner.ort = { name: "ort" };
+    runner.detModel = detModel;
+    runner.recModel = recModel;
+    runner.webgpuState = { available: false, reason: "" };
+    runner.modelConfig = { det: { conf: true }, rec: { conf: true } };
+
+    const results = await runner.predict([{ kind: "a" }, { kind: "b" }], {});
+
+    expect(detModel.predict).toHaveBeenCalledTimes(1);
+    expect(detModel.predict).toHaveBeenCalledWith(cv, [mat1, mat2], {});
+    expect(recModel.predict).toHaveBeenNthCalledWith(1, cv, [crop1]);
+    expect(recModel.predict).toHaveBeenNthCalledWith(2, cv, [crop2]);
+    expect(results).toHaveLength(2);
+    expect(dispose1).toHaveBeenCalledTimes(1);
+    expect(dispose2).toHaveBeenCalledTimes(1);
+  });
+
+  it("auto-initializes on predict and rejects when source adapter is missing", async () => {
+    const detModel = {
+      provider: "wasm",
+      predict: vi.fn().mockResolvedValue([{ boxes: [], srcW: 1, srcH: 1 }]),
+      dispose: vi.fn()
+    };
+    const recModel = {
+      provider: "wasm",
+      predict: vi.fn().mockResolvedValue([]),
+      dispose: vi.fn()
+    };
+
+    mockEmptyDefaultOcrConfig();
+    nowMs.mockReturnValue(0);
+    initOpenCvRuntime.mockResolvedValue({ cv: {} });
+    initOrtRuntime.mockResolvedValue({
+      ort: {},
+      webgpuState: { available: false, reason: "" },
+      backend: "wasm"
+    });
+    loadModelAsset
+      .mockResolvedValueOnce({
+        modelBytes: new Uint8Array([1]),
+        configText: "det-config",
+        download: { url: "/det.tar", bytes: 100 }
+      })
+      .mockResolvedValueOnce({
+        modelBytes: new Uint8Array([2]),
+        configText: "rec-config",
+        download: { url: "/rec.tar", bytes: 200 }
+      });
+    createDetModel.mockResolvedValue(detModel);
+    createRecModel.mockResolvedValue(recModel);
+    getOcrRuntimeParams.mockReturnValue({
+      det: {},
+      pipeline: { scoreThresh: 0 }
+    });
+
+    const { OcrPipelineRunner } = await loadCoreModule();
+    const noSourceRunner = new OcrPipelineRunner({
+      pipelineConfig: minimalPipelineConfig()
+    });
+    await expect(noSourceRunner.predict({}, {})).rejects.toThrow(
+      /source adapter is not configured/i
+    );
+
+    const sourceImage = {
+      width: 1,
+      height: 1,
+      mat: {},
+      dispose: vi.fn()
+    };
+    const runner = new OcrPipelineRunner({
+      pipelineConfig: minimalPipelineConfig(),
+      sourceToMat: vi.fn().mockResolvedValue(sourceImage)
+    });
+
+    const result = await runner.predict({}, {});
+
+    expect(initOpenCvRuntime).toHaveBeenCalled();
+    expect(result[0].items).toEqual([]);
+    expect(sourceImage.dispose).toHaveBeenCalledTimes(1);
+  });
+
+  it("disposes models and clears references", async () => {
+    mockEmptyDefaultOcrConfig();
+    const detDispose = vi.fn().mockResolvedValue(undefined);
+    const recDispose = vi.fn().mockResolvedValue(undefined);
+
+    const { OcrPipelineRunner } = await loadCoreModule();
+    const runner = new OcrPipelineRunner({
+      pipelineConfig: minimalPipelineConfig()
+    });
+    runner.detModel = { dispose: detDispose };
+    runner.recModel = { dispose: recDispose };
+
+    await runner.disposeModelsOnly();
+    expect(detDispose).toHaveBeenCalledTimes(1);
+    expect(recDispose).toHaveBeenCalledTimes(1);
+    expect(runner.detModel).toBeNull();
+    expect(runner.recModel).toBeNull();
+
+    runner.detModel = { dispose: detDispose };
+    runner.recModel = { dispose: recDispose };
+    await runner.dispose();
+    expect(runner.detModel).toBeNull();
+    expect(runner.recModel).toBeNull();
+  });
+});
diff --git a/paddleocr-js/packages/core/test/ocr-shared.test.ts b/paddleocr-js/packages/core/test/ocr-shared.test.ts
new file mode 100644
index 0000000..bcd87b1
--- /dev/null
+++ b/paddleocr-js/packages/core/test/ocr-shared.test.ts
@@ -0,0 +1,130 @@
+import { describe, expect, it } from "vitest";
+
+import {
+  cloneDefaultOcrConfig,
+  normalizeOrtOptions,
+  resolvePaddleOCROptions,
+  resolveWorkerOptions,
+  validateLoadedModelName
+} from "../src/pipelines/ocr/shared";
+
+describe("OCR shared option resolution", () => {
+  it("normalizes ORT options and reuses the same backend fallback", () => {
+    const defaultOrt = normalizeOrtOptions();
+
+    expect(defaultOrt).toMatchObject({
+      backend: expect.any(String)
+    });
+    expect(normalizeOrtOptions({ backend: "invalid", proxy: true })).toEqual({
+      backend: defaultOrt.backend,
+      proxy: true
+    });
+    expect(
+      normalizeOrtOptions({
+        backend: "wasm",
+        wasmPaths: "/wasm/",
+        numThreads: 2,
+        simd: true,
+        proxy: false
+      })
+    ).toEqual({
+      backend: "wasm",
+      wasmPaths: "/wasm/",
+      numThreads: 2,
+      simd: true,
+      proxy: false
+    });
+  });
+
+  it("resolves worker options from booleans and custom factories", () => {
+    const createWorker = () => ({});
+
+    expect(resolveWorkerOptions(false)).toEqual({
+      enabled: false,
+      createWorker: null
+    });
+    expect(resolveWorkerOptions(true)).toEqual({
+      enabled: true,
+      createWorker: null
+    });
+    expect(resolveWorkerOptions({ createWorker })).toEqual({
+      enabled: true,
+      createWorker
+    });
+    expect(resolveWorkerOptions({})).toEqual({
+      enabled: true,
+      createWorker: null
+    });
+  });
+
+  it("rejects unsupported worker option types", () => {
+    expect(() => resolveWorkerOptions("yes")).toThrow(
+      /worker must be a boolean or an options object/i
+    );
+  });
+
+  it("returns ortOptions, assets, and model selection for explicit model names", () => {
+    const options = resolvePaddleOCROptions({
+      text_detection_model_name: "PP-OCRv5_mobile_det",
+      text_recognition_model_name: "PP-OCRv5_mobile_rec",
+      ortOptions: {
+        backend: "webgpu",
+        proxy: true
+      }
+    });
+
+    expect(options.ortOptions).toEqual({
+      backend: "webgpu",
+      proxy: true
+    });
+    expect(options.pipelineConfig.assets.det?.url).toMatch(/PP-OCRv5_mobile_det.*\.tar$/);
+    expect(options.pipelineConfig.assets.rec?.url).toMatch(/PP-OCRv5_mobile_rec.*\.tar$/);
+    expect(options.pipelineConfig.modelSelection).toEqual({
+      textDetectionModelName: "PP-OCRv5_mobile_det",
+      textRecognitionModelName: "PP-OCRv5_mobile_rec"
+    });
+  });
+
+  it("rejects incomplete pipeline model selection", () => {
+    expect(() =>
+      resolvePaddleOCROptions({
+        pipelineConfig: {
+          pipeline_name: "OCR",
+          SubModules: {
+            TextDetection: {
+              model_name: "PP-OCRv5_mobile_det"
+            }
+          }
+        }
+      })
+    ).toThrow(/must define both "SubModules.TextDetection" and "SubModules.TextRecognition"/i);
+  });
+
+  it("clones the default OCR config deeply", () => {
+    const cloned = cloneDefaultOcrConfig();
+    cloned.det.postprocess.thresh = 0.99;
+
+    expect(cloned.det.postprocess.thresh).toBe(0.99);
+
+    const freshClone = cloneDefaultOcrConfig();
+    expect(freshClone.det.postprocess.thresh).not.toBe(0.99);
+  });
+
+  it("validates loaded model names against inference.yml (roles match pipeline initialize)", () => {
+    expect(() =>
+      validateLoadedModelName(
+        "TextDetection",
+        "PP-OCRv5_mobile_det",
+        "Global:\n  model_name: PP-OCRv5_mobile_det"
+      )
+    ).not.toThrow();
+
+    expect(() =>
+      validateLoadedModelName(
+        "TextDetection",
+        "PP-OCRv5_mobile_det",
+        "Global:\n  model_name: other"
+      )
+    ).toThrow(/requested model_name is "PP-OCRv5_mobile_det"/i);
+  });
+});
diff --git a/paddleocr-js/packages/core/test/ocr-worker-entry.test.ts b/paddleocr-js/packages/core/test/ocr-worker-entry.test.ts
new file mode 100644
index 0000000..b6c972f
--- /dev/null
+++ b/paddleocr-js/packages/core/test/ocr-worker-entry.test.ts
@@ -0,0 +1,163 @@
+import { afterEach, describe, expect, it, vi } from "vitest";
+
+let capturedHandler = null;
+const attachWorkerMessageHandler = vi.fn((handler) => {
+  capturedHandler = handler;
+});
+
+const sourcePayloadToMat = vi.fn();
+const ensureServedFromHttp = vi.fn();
+const initialize = vi.fn();
+const getModelConfig = vi.fn();
+const predict = vi.fn();
+const dispose = vi.fn();
+
+const OcrPipelineRunner = vi.fn(function MockOcrPipelineRunner(options) {
+  this.options = options;
+  this.initialize = initialize;
+  this.getModelConfig = getModelConfig;
+  this.predict = predict;
+  this.dispose = dispose;
+});
+
+vi.mock("../src/worker/entry", () => ({
+  attachWorkerMessageHandler
+}));
+
+vi.mock("../src/platform/worker", () => ({
+  sourcePayloadToMat,
+  ensureServedFromHttp
+}));
+
+vi.mock("../src/pipelines/ocr/core", () => ({
+  OcrPipelineRunner
+}));
+
+afterEach(() => {
+  capturedHandler = null;
+  vi.resetModules();
+  vi.clearAllMocks();
+});
+
+async function loadWorkerEntry() {
+  await import("../src/pipelines/ocr/worker-entry");
+  expect(typeof capturedHandler).toBe("function");
+}
+
+const WASM_INIT_SUMMARY = Object.freeze({ backend: "wasm" });
+const EMPTY_MODEL_CONFIG = Object.freeze({ det: {}, rec: {} });
+
+const WORKER_TEST_PIPELINE = Object.freeze({
+  pipelineName: "OCR",
+  raw: {},
+  warnings: [] as string[],
+  unsupportedFeatures: [] as string[],
+  modelSelection: {
+    textDetectionModelName: "d",
+    textRecognitionModelName: "r"
+  },
+  assets: {
+    det: { url: "/d" },
+    rec: { url: "/r" }
+  },
+  runtimeDefaults: {},
+  pipelineBatchSize: 1,
+  textDetectionBatchSize: 1,
+  textRecognitionBatchSize: 1
+});
+
+function setupResolvedInitAndModelConfig() {
+  initialize.mockResolvedValue(WASM_INIT_SUMMARY);
+  getModelConfig.mockReturnValue(EMPTY_MODEL_CONFIG);
+}
+
+describe("OCR worker entry bootstrap", () => {
+  it("registers a worker message handler on module load", async () => {
+    await loadWorkerEntry();
+
+    expect(attachWorkerMessageHandler).toHaveBeenCalledTimes(1);
+    expect(typeof capturedHandler).toBe("function");
+  });
+
+  it("initializes a runner and returns summary + model config", async () => {
+    setupResolvedInitAndModelConfig();
+
+    await loadWorkerEntry();
+    const result = await capturedHandler("init", {
+      options: {
+        pipelineConfig: WORKER_TEST_PIPELINE,
+        ortOptions: { backend: "wasm" }
+      }
+    });
+
+    expect(OcrPipelineRunner).toHaveBeenCalledWith({
+      pipelineConfig: WORKER_TEST_PIPELINE,
+      ortOptions: { backend: "wasm" },
+      ensureServedFromHttp,
+      sourceToMat: sourcePayloadToMat
+    });
+    expect(result).toEqual({
+      summary: WASM_INIT_SUMMARY,
+      modelConfig: EMPTY_MODEL_CONFIG
+    });
+  });
+
+  it("disposes an existing runner before re-initializing", async () => {
+    setupResolvedInitAndModelConfig();
+
+    await loadWorkerEntry();
+    await capturedHandler("init", { options: { pipelineConfig: WORKER_TEST_PIPELINE, id: 1 } });
+    await capturedHandler("init", { options: { pipelineConfig: WORKER_TEST_PIPELINE, id: 2 } });
+
+    expect(dispose).toHaveBeenCalledTimes(1);
+    expect(OcrPipelineRunner).toHaveBeenCalledTimes(2);
+  });
+
+  it("routes predict and dispose requests through the active runner", async () => {
+    setupResolvedInitAndModelConfig();
+    const predictPayload = [
+      {
+        image: { width: 1, height: 1 },
+        items: [],
+        metrics: {
+          detMs: 0,
+          recMs: 0,
+          totalMs: 0,
+          detectedBoxes: 0,
+          recognizedCount: 0
+        },
+        runtime: {
+          requestedBackend: "auto",
+          detProvider: "wasm",
+          recProvider: "wasm",
+          webgpuAvailable: false
+        }
+      }
+    ];
+    predict.mockResolvedValue(predictPayload);
+    dispose.mockResolvedValue(undefined);
+
+    await loadWorkerEntry();
+    await capturedHandler("init", { options: { pipelineConfig: WORKER_TEST_PIPELINE } });
+
+    await expect(
+      capturedHandler("predict", {
+        sources: [{ kind: "imageBitmap" }],
+        params: { limit: 1 }
+      })
+    ).resolves.toEqual(predictPayload);
+    expect(predict).toHaveBeenCalledWith([{ kind: "imageBitmap" }], { limit: 1 });
+
+    await expect(capturedHandler("dispose", {})).resolves.toEqual({});
+    expect(dispose).toHaveBeenCalledTimes(1);
+  });
+
+  it("rejects predict before initialization and unknown request types", async () => {
+    await loadWorkerEntry();
+
+    await expect(capturedHandler("predict", { sources: [{}], params: {} })).rejects.toThrow(
+      /not initialized/i
+    );
+    await expect(capturedHandler("other", {})).rejects.toThrow(/Unsupported worker request type/i);
+  });
+});
diff --git a/paddleocr-js/packages/core/test/pipelines-index.test.ts b/paddleocr-js/packages/core/test/pipelines-index.test.ts
new file mode 100644
index 0000000..08ede2a
--- /dev/null
+++ b/paddleocr-js/packages/core/test/pipelines-index.test.ts
@@ -0,0 +1,24 @@
+import { describe, expect, it, vi } from "vitest";
+
+vi.mock("@techstark/opencv-js", () => ({
+  default: {
+    Mat() {}
+  }
+}));
+
+import * as publicPipelines from "../src/pipelines/index";
+import * as ocrPipeline from "../src/pipelines/ocr/index";
+
+describe("pipeline entrypoints", () => {
+  it("re-exports OCR pipeline APIs from the pipelines index", () => {
+    expect(publicPipelines.PaddleOCR).toBe(ocrPipeline.PaddleOCR);
+    expect(publicPipelines.parseOcrPipelineConfigText).toBe(ocrPipeline.parseOcrPipelineConfigText);
+    expect(publicPipelines.normalizeOcrPipelineConfig).toBe(ocrPipeline.normalizeOcrPipelineConfig);
+  });
+
+  it("exposes only the barrel surface re-exported from pipelines/ocr (no drift)", () => {
+    expect(Object.keys(publicPipelines).sort()).toEqual(
+      ["normalizeOcrPipelineConfig", "parseOcrPipelineConfigText", "PaddleOCR"].sort()
+    );
+  });
+});
diff --git a/paddleocr-js/packages/core/test/platform-browser.test.ts b/paddleocr-js/packages/core/test/platform-browser.test.ts
new file mode 100644
index 0000000..b583221
--- /dev/null
+++ b/paddleocr-js/packages/core/test/platform-browser.test.ts
@@ -0,0 +1,172 @@
+import { afterEach, describe, expect, it, vi } from "vitest";
+
+import {
+  ensureServedFromHttp,
+  sourceToImageBitmap,
+  sourceToMat,
+  sourceToWorkerPayload
+} from "../src/platform/browser";
+
+class FakeImageBitmap {
+  constructor(width = 10, height = 20) {
+    this.width = width;
+    this.height = height;
+    this.close = vi.fn();
+  }
+}
+
+class FakeImageData {
+  constructor(data, width, height) {
+    this.data = data;
+    this.width = width;
+    this.height = height;
+  }
+}
+
+class FakeCanvasElement {}
+class FakeImageElement {}
+
+function stubCanvasDocument() {
+  const drawImage = vi.fn();
+  const putImageData = vi.fn();
+  const canvas = {
+    width: 0,
+    height: 0,
+    getContext: vi.fn(() => ({
+      drawImage,
+      putImageData
+    }))
+  };
+
+  vi.stubGlobal("document", {
+    createElement: vi.fn(() => canvas)
+  });
+
+  return {
+    canvas,
+    drawImage,
+    putImageData
+  };
+}
+
+afterEach(() => {
+  vi.unstubAllGlobals();
+  vi.restoreAllMocks();
+});
+
+describe("platform/browser", () => {
+  it("rejects file:// origins", () => {
+    vi.stubGlobal("location", { protocol: "file:" });
+
+    expect(() => ensureServedFromHttp()).toThrow(/requires an HTTP\(S\) origin/i);
+  });
+
+  it("allows http origins", () => {
+    vi.stubGlobal("location", { protocol: "https:" });
+
+    expect(() => ensureServedFromHttp()).not.toThrow();
+  });
+
+  it("reuses existing ImageBitmap sources", async () => {
+    vi.stubGlobal("ImageBitmap", FakeImageBitmap);
+    const bitmap = new FakeImageBitmap();
+
+    await expect(sourceToImageBitmap(bitmap)).resolves.toBe(bitmap);
+  });
+
+  it("converts Blob and canvas sources through createImageBitmap", async () => {
+    const bitmap = new FakeImageBitmap();
+    const createImageBitmap = vi.fn().mockResolvedValue(bitmap);
+
+    vi.stubGlobal("ImageBitmap", FakeImageBitmap);
+    vi.stubGlobal("HTMLCanvasElement", FakeCanvasElement);
+    vi.stubGlobal("createImageBitmap", createImageBitmap);
+
+    const blob = new Blob(["hello"], { type: "text/plain" });
+    await expect(sourceToImageBitmap(blob)).resolves.toBe(bitmap);
+
+    const canvas = new FakeCanvasElement();
+    await expect(sourceToImageBitmap(canvas)).resolves.toBe(bitmap);
+    expect(createImageBitmap).toHaveBeenCalledTimes(2);
+  });
+
+  it("converts ImageData and img sources through createImageBitmap", async () => {
+    const bitmap = new FakeImageBitmap();
+    const createImageBitmap = vi.fn().mockResolvedValue(bitmap);
+    const { canvas, putImageData } = stubCanvasDocument();
+
+    vi.stubGlobal("ImageBitmap", FakeImageBitmap);
+    vi.stubGlobal("ImageData", FakeImageData);
+    vi.stubGlobal("HTMLImageElement", FakeImageElement);
+    vi.stubGlobal("createImageBitmap", createImageBitmap);
+
+    const imageData = new FakeImageData(new Uint8ClampedArray(16), 2, 2);
+    await expect(sourceToImageBitmap(imageData)).resolves.toBe(bitmap);
+    expect(canvas.width).toBe(2);
+    expect(canvas.height).toBe(2);
+    expect(putImageData).toHaveBeenCalledWith(imageData, 0, 0);
+
+    const image = new FakeImageElement();
+    await expect(sourceToImageBitmap(image)).resolves.toBe(bitmap);
+  });
+
+  it("rejects unsupported image sources", async () => {
+    vi.stubGlobal("ImageData", FakeImageData);
+
+    await expect(sourceToImageBitmap({})).rejects.toThrow(/Unsupported image source/i);
+  });
+
+  it("converts bitmap-backed sources into cv.Mat wrappers", async () => {
+    const bitmap = new FakeImageBitmap(64, 32);
+    const { drawImage } = stubCanvasDocument();
+    const mat = {
+      deleted: false,
+      delete() {
+        this.deleted = true;
+      }
+    };
+
+    vi.stubGlobal("ImageBitmap", FakeImageBitmap);
+
+    const loaded = await sourceToMat(
+      {
+        imread: vi.fn(() => mat)
+      },
+      bitmap
+    );
+
+    expect(loaded.width).toBe(64);
+    expect(loaded.height).toBe(32);
+    expect(drawImage).toHaveBeenCalledWith(bitmap, 0, 0);
+
+    loaded.dispose();
+    expect(mat.deleted).toBe(true);
+    expect(bitmap.close).toHaveBeenCalledTimes(1);
+  });
+
+  it("creates transferable worker payloads from ImageBitmap sources", async () => {
+    const sourceBitmap = new FakeImageBitmap(10, 20);
+    const clonedBitmap = new FakeImageBitmap(10, 20);
+    const createImageBitmap = vi.fn().mockResolvedValue(clonedBitmap);
+
+    vi.stubGlobal("ImageBitmap", FakeImageBitmap);
+    vi.stubGlobal("createImageBitmap", createImageBitmap);
+
+    const result = await sourceToWorkerPayload(sourceBitmap);
+
+    expect(createImageBitmap).toHaveBeenCalledWith(sourceBitmap);
+    expect(result).toEqual({
+      payload: {
+        kind: "imageBitmap",
+        imageBitmap: clonedBitmap
+      },
+      transferables: [clonedBitmap]
+    });
+  });
+
+  it("rejects worker payload creation when ImageBitmap support is missing", async () => {
+    await expect(sourceToWorkerPayload(new Blob(["x"]))).rejects.toThrow(
+      /Worker mode requires ImageBitmap support/i
+    );
+  });
+});
diff --git a/paddleocr-js/packages/core/test/platform-worker.test.ts b/paddleocr-js/packages/core/test/platform-worker.test.ts
new file mode 100644
index 0000000..97082ab
--- /dev/null
+++ b/paddleocr-js/packages/core/test/platform-worker.test.ts
@@ -0,0 +1,143 @@
+import { afterEach, describe, expect, it, vi } from "vitest";
+
+import { ensureServedFromHttp, sourcePayloadToMat } from "../src/platform/worker";
+
+class FakeMat {
+  constructor(rows, cols) {
+    this.rows = rows;
+    this.cols = cols;
+    this.deleted = false;
+  }
+
+  clone() {
+    return new FakeMat(this.rows, this.cols);
+  }
+
+  delete() {
+    this.deleted = true;
+  }
+}
+
+class FakeImageBitmap {
+  constructor(width = 8, height = 6) {
+    this.width = width;
+    this.height = height;
+    this.close = vi.fn();
+  }
+}
+
+afterEach(() => {
+  vi.unstubAllGlobals();
+});
+
+describe("platform/worker", () => {
+  it("re-exports the HTTP origin guard", () => {
+    vi.stubGlobal("location", { protocol: "file:" });
+
+    expect(() => ensureServedFromHttp()).toThrow(/requires an HTTP\(S\) origin/i);
+  });
+
+  it("accepts cv.Mat payloads by cloning them", () => {
+    const source = new FakeMat(32, 64);
+    const loaded = sourcePayloadToMat({ Mat: FakeMat } as any, source);
+
+    expect(loaded.width).toBe(64);
+    expect(loaded.height).toBe(32);
+    expect(loaded.mat).toBeInstanceOf(FakeMat);
+    expect(loaded.mat).not.toBe(source);
+
+    loaded.dispose();
+    expect(loaded.mat.deleted).toBe(true);
+    expect(source.deleted).toBe(false);
+  });
+
+  it("converts imageBitmap payloads into cv mats", () => {
+    const imageBitmap = new FakeImageBitmap(20, 10);
+    const getImageData = vi.fn(() => ({
+      width: 20,
+      height: 10,
+      data: new Uint8ClampedArray(20 * 10 * 4)
+    }));
+    const drawImage = vi.fn();
+    const mat = {
+      deleted: false,
+      delete() {
+        this.deleted = true;
+      }
+    };
+
+    vi.stubGlobal("ImageBitmap", FakeImageBitmap);
+    vi.stubGlobal(
+      "OffscreenCanvas",
+      class {
+        constructor(width, height) {
+          this.width = width;
+          this.height = height;
+        }
+
+        getContext() {
+          return {
+            drawImage,
+            getImageData
+          };
+        }
+      }
+    );
+
+    const loaded = sourcePayloadToMat(
+      {
+        Mat: FakeMat,
+        CV_8UC4: "rgba",
+        matFromArray: vi.fn(() => mat)
+      } as any,
+      {
+        kind: "imageBitmap",
+        imageBitmap
+      }
+    );
+
+    expect(drawImage).toHaveBeenCalledWith(imageBitmap, 0, 0);
+    expect(loaded.width).toBe(20);
+    expect(loaded.height).toBe(10);
+
+    loaded.dispose();
+    expect(mat.deleted).toBe(true);
+    expect(imageBitmap.close).toHaveBeenCalledTimes(1);
+  });
+
+  it("rejects imageBitmap payloads when OffscreenCanvas is unavailable", () => {
+    vi.stubGlobal("ImageBitmap", FakeImageBitmap);
+
+    expect(() =>
+      sourcePayloadToMat({ Mat: FakeMat } as any, {
+        kind: "imageBitmap",
+        imageBitmap: new FakeImageBitmap()
+      })
+    ).toThrow(/requires OffscreenCanvas support/i);
+  });
+
+  it("rejects imageBitmap payloads when the canvas context cannot be created", () => {
+    vi.stubGlobal("ImageBitmap", FakeImageBitmap);
+    vi.stubGlobal(
+      "OffscreenCanvas",
+      class {
+        getContext() {
+          return null;
+        }
+      }
+    );
+
+    expect(() =>
+      sourcePayloadToMat({ Mat: FakeMat } as any, {
+        kind: "imageBitmap",
+        imageBitmap: new FakeImageBitmap()
+      })
+    ).toThrow(/Failed to create a 2D canvas context/i);
+  });
+
+  it("rejects unsupported worker payloads", () => {
+    expect(() => sourcePayloadToMat({ Mat: FakeMat } as any, { kind: "other" })).toThrow(
+      /Unsupported worker image source payload/i
+    );
+  });
+});
diff --git a/paddleocr-js/packages/core/test/public-api.test.ts b/paddleocr-js/packages/core/test/public-api.test.ts
new file mode 100644
index 0000000..b4b3d7d
--- /dev/null
+++ b/paddleocr-js/packages/core/test/public-api.test.ts
@@ -0,0 +1,43 @@
+import { describe, expect, it, vi } from "vitest";
+
+vi.mock("@techstark/opencv-js", () => ({
+  default: {
+    Mat() {}
+  }
+}));
+
+import { PaddleOCR, normalizeOcrPipelineConfig, parseOcrPipelineConfigText } from "../src/index";
+
+describe("public pipeline exports", () => {
+  it("exports PaddleOCR", () => {
+    expect(typeof PaddleOCR).toBe("function");
+  });
+
+  it("exports OCR pipeline config helpers", () => {
+    expect(typeof parseOcrPipelineConfigText).toBe("function");
+    expect(typeof normalizeOcrPipelineConfig).toBe("function");
+  });
+
+  it("supports worker mode through PaddleOCR.create()", async () => {
+    const ocr = await PaddleOCR.create({
+      worker: true,
+      initialize: false
+    });
+
+    expect(typeof ocr.initialize).toBe("function");
+    expect(typeof ocr.predict).toBe("function");
+    expect(typeof ocr.dispose).toBe("function");
+  });
+
+  it("resolves ortOptions on the instance (create-time ORT settings)", async () => {
+    const ocr = await PaddleOCR.create({
+      lang: "ch",
+      ocrVersion: "PP-OCRv5",
+      initialize: false,
+      ortOptions: { backend: "wasm", wasmPaths: "/dist/" }
+    });
+
+    expect(ocr.options.ortOptions.backend).toBe("wasm");
+    expect(ocr.options.ortOptions.wasmPaths).toBe("/dist/");
+  });
+});
diff --git a/paddleocr-js/packages/core/test/rec-model.test.ts b/paddleocr-js/packages/core/test/rec-model.test.ts
new file mode 100644
index 0000000..83ad3ef
--- /dev/null
+++ b/paddleocr-js/packages/core/test/rec-model.test.ts
@@ -0,0 +1,381 @@
+import { afterEach, describe, expect, it, vi } from "vitest";
+import { createMockOrtTensorClass } from "./helpers/mock-ort-tensor";
+
+const assertModelResources = vi.fn();
+const createSession = vi.fn();
+const getProviderCandidates = vi.fn();
+const releaseSessions = vi.fn();
+const clamp = vi.fn((value, min, max) => Math.max(min, Math.min(max, value)));
+const withTimeout = vi.fn((promise) => promise);
+const chunkArray = vi.fn((items, size) => {
+  const chunks = [];
+  for (let i = 0; i < items.length; i += size) {
+    chunks.push(items.slice(i, i + size));
+  }
+  return chunks;
+});
+const getTransformOp = vi.fn();
+const parseInferenceConfigText = vi.fn();
+const toBgrFloatCHWFromBgr = vi.fn();
+
+vi.mock("../src/resources/model-asset", () => ({
+  assertModelResources
+}));
+
+vi.mock("../src/runtime/ort", () => ({
+  createSession,
+  getProviderCandidates,
+  releaseSessions
+}));
+
+vi.mock("../src/utils/common", async (importOriginal) => {
+  const actual = await importOriginal<typeof import("../src/utils/common")>();
+  return {
+    ...actual,
+    clamp,
+    withTimeout,
+    chunkArray
+  };
+});
+
+vi.mock("../src/models/common", () => ({
+  getTransformOp,
+  parseInferenceConfigText,
+  toBgrFloatCHWFromBgr
+}));
+
+afterEach(() => {
+  vi.resetModules();
+  vi.clearAllMocks();
+});
+
+async function loadRecModule() {
+  return import("../src/models/rec");
+}
+
+function createMat(channels, cols = 20, rows = 10) {
+  return {
+    cols,
+    rows,
+    data: new Uint8Array(cols * rows * 3).fill(1),
+    channels: () => channels,
+    copyTo: vi.fn(),
+    delete: vi.fn()
+  };
+}
+
+/** Minimal OpenCV-like `cv` used by `createRecModel().predict()` → internal `preprocessSample`. */
+function createRecPredictCvStub() {
+  return {
+    Mat: class Mat {
+      constructor() {
+        this.deleted = false;
+        this.data = new Uint8Array(8);
+        this._channels = 3;
+      }
+      channels() {
+        return this._channels;
+      }
+      copyTo(target) {
+        target.data = this.data;
+        target._channels = 3;
+      }
+      delete() {
+        this.deleted = true;
+      }
+    },
+    Size: class Size {
+      constructor(width, height) {
+        this.width = width;
+        this.height = height;
+      }
+    },
+    INTER_LINEAR: "linear",
+    COLOR_RGBA2BGR: "rgba",
+    COLOR_GRAY2BGR: "gray",
+    resize: vi.fn((src, dst, size) => {
+      dst.data = new Uint8Array(size.width * size.height * 3);
+      dst._channels = src.channels();
+    }),
+    cvtColor: vi.fn((src, dst) => {
+      dst.data = src.data;
+      dst._channels = 3;
+    })
+  };
+}
+
+describe("recognition model", () => {
+  it("parses recognition configs and validates character dictionaries", async () => {
+    parseInferenceConfigText.mockReturnValue({
+      PreProcess: {
+        transform_ops: [{ id: "resize" }, { id: "normalize" }]
+      },
+      PostProcess: {
+        character_dict: ["a", "b"]
+      }
+    });
+    getTransformOp.mockReturnValueOnce({ image_shape: [3, 32, 160] });
+
+    const { DEFAULT_REC_MODEL_PARSE_FALLBACKS, parseRecModelConfigText } = await loadRecModule();
+    expect(parseRecModelConfigText("config")).toEqual({
+      imageShape: [3, 32, 160],
+      charDict: ["a", "b", " "]
+    });
+
+    parseInferenceConfigText.mockReturnValue({
+      PreProcess: {},
+      PostProcess: {}
+    });
+    getTransformOp.mockReturnValue(undefined);
+
+    expect(() => parseRecModelConfigText("invalid")).toThrow(
+      /RecResizeImg\.image_shape is required/i
+    );
+  });
+
+  it("runs recognition batches through predict and decodes CTC output", async () => {
+    parseInferenceConfigText.mockReturnValue({
+      PreProcess: { transform_ops: [] },
+      PostProcess: { character_dict: ["A", "B", "C"] }
+    });
+    getTransformOp.mockImplementation((_ops, id) => {
+      if (id === "RecResizeImg") return { image_shape: [3, 4, 8] };
+      return null;
+    });
+    clamp.mockImplementation((value, min, max) => Math.max(min, Math.min(max, value)));
+
+    const tensorCalls = [];
+    const ort = {
+      Tensor: createMockOrtTensorClass(tensorCalls)
+    };
+    const ctcRow = new Float32Array([0.1, 0.9, 0.2, 0.1, 0.2, 0.1, 0.8, 0.1, 0.8, 0.1, 0.1, 0.0]);
+    const sessionRun = vi
+      .fn()
+      .mockResolvedValueOnce({
+        output: {
+          dims: [2, 3, 4],
+          data: new Float32Array([...ctcRow, ...ctcRow])
+        }
+      })
+      .mockResolvedValueOnce({
+        output: {
+          dims: [1, 3, 4],
+          data: ctcRow
+        }
+      });
+    const session = {
+      inputNames: ["input"],
+      outputNames: ["output"],
+      run: sessionRun
+    };
+    getProviderCandidates.mockReturnValue([["wasm"]]);
+    createSession.mockResolvedValue({
+      session,
+      provider: "wasm"
+    });
+
+    const { createRecModel } = await loadRecModule();
+    const model = await createRecModel({
+      ort,
+      modelBytes: new Uint8Array([1]),
+      configText: "rec-batch",
+      backend: "auto",
+      webgpuState: { available: false, reason: "" },
+      batchSize: 2
+    });
+
+    const cvFixture = createRecPredictCvStub();
+    toBgrFloatCHWFromBgr.mockImplementation((data, width, height) => {
+      const out = new Float32Array(3 * width * height);
+      for (let i = 0; i < out.length; i += 1) out[i] = i + 1;
+      return out;
+    });
+    const mat = createMat(3, 8, 4);
+    const results = await model.predict(cvFixture, [mat, mat, mat]);
+
+    expect(sessionRun).toHaveBeenCalledTimes(2);
+    expect(tensorCalls).toEqual([
+      { type: "float32", dims: [2, 3, 4, 8], size: 192 },
+      { type: "float32", dims: [1, 3, 4, 8], size: 96 }
+    ]);
+    expect(results).toHaveLength(3);
+    expect(results[0]).toMatchObject({ text: "AB" });
+    expect(results[0].score).toBeCloseTo(0.85, 5);
+    expect(results[1]).toMatchObject({ text: "AB" });
+    expect(results[1].score).toBeCloseTo(0.85, 5);
+    expect(results[2]).toMatchObject({ text: "AB" });
+    expect(results[2].score).toBeCloseTo(0.85, 5);
+  });
+
+  it("allows per-predict batch size override on the recognition model", async () => {
+    parseInferenceConfigText.mockReturnValue({
+      PreProcess: { transform_ops: [] },
+      PostProcess: { character_dict: ["A", "B", "C"] }
+    });
+    getTransformOp.mockImplementation((_ops, id) => {
+      if (id === "RecResizeImg") return { image_shape: [3, 4, 8] };
+      return null;
+    });
+    clamp.mockImplementation((value, min, max) => Math.max(min, Math.min(max, value)));
+
+    const ctcRow = new Float32Array([0.1, 0.9, 0.2, 0.1, 0.2, 0.1, 0.8, 0.1, 0.8, 0.1, 0.1, 0.0]);
+    const sessionRun = vi.fn().mockResolvedValue({
+      output: {
+        dims: [1, 3, 4],
+        data: ctcRow
+      }
+    });
+    const session = {
+      inputNames: ["input"],
+      outputNames: ["output"],
+      run: sessionRun
+    };
+    getProviderCandidates.mockReturnValue([["wasm"]]);
+    createSession.mockResolvedValue({
+      session,
+      provider: "wasm"
+    });
+
+    const { createRecModel } = await loadRecModule();
+    const model = await createRecModel({
+      ort: { Tensor: createMockOrtTensorClass() },
+      modelBytes: new Uint8Array([1]),
+      configText: "rec-override-batch",
+      backend: "auto",
+      webgpuState: { available: false, reason: "" },
+      batchSize: 6
+    });
+
+    const cvFixture = createRecPredictCvStub();
+    toBgrFloatCHWFromBgr.mockImplementation((data, width, height) => {
+      const out = new Float32Array(3 * width * height);
+      for (let i = 0; i < out.length; i += 1) out[i] = i + 1;
+      return out;
+    });
+    const mat = createMat(3, 8, 4);
+    await model.predict(cvFixture, [mat, mat, mat], { batchSize: 1 });
+
+    expect(sessionRun).toHaveBeenCalledTimes(3);
+  });
+
+  it("rejects unexpected recognition output dimensions", async () => {
+    parseInferenceConfigText.mockReturnValue({
+      PreProcess: { transform_ops: [] },
+      PostProcess: { character_dict: ["A"] }
+    });
+    getTransformOp.mockImplementation((_ops, id) => {
+      if (id === "RecResizeImg") return { image_shape: [3, 4, 8] };
+      return null;
+    });
+    getProviderCandidates.mockReturnValue([["wasm"]]);
+    createSession.mockResolvedValue({
+      session: {
+        inputNames: ["input"],
+        outputNames: ["output"],
+        run: vi.fn().mockResolvedValue({
+          output: {
+            dims: [1, 4],
+            data: new Float32Array([1, 2, 3, 4])
+          }
+        })
+      },
+      provider: "wasm"
+    });
+
+    const { createRecModel } = await loadRecModule();
+    const model = await createRecModel({
+      ort: {
+        Tensor: createMockOrtTensorClass()
+      },
+      modelBytes: new Uint8Array([1]),
+      configText: "rec-bad-out",
+      backend: "auto",
+      webgpuState: { available: false, reason: "" }
+    });
+    const cvFixture = createRecPredictCvStub();
+    toBgrFloatCHWFromBgr.mockImplementation((data, width, height) => {
+      const out = new Float32Array(3 * width * height);
+      out.fill(1);
+      return out;
+    });
+
+    await expect(model.predict(cvFixture, [createMat(3, 8, 4)])).rejects.toThrow(
+      /Unexpected rec output dims/i
+    );
+  });
+
+  it("creates, uses, and disposes recognition models through runtime wrappers", async () => {
+    parseInferenceConfigText.mockReturnValue({
+      PreProcess: {
+        transform_ops: []
+      },
+      PostProcess: {
+        character_dict: ["A"]
+      }
+    });
+    getTransformOp.mockImplementation((_ops, id) => {
+      if (id === "RecResizeImg") return { image_shape: [3, 4, 8] };
+      return undefined;
+    });
+    getProviderCandidates.mockReturnValue([["wasm"]]);
+    createSession.mockResolvedValue({
+      session: {
+        inputNames: ["input"],
+        outputNames: ["output"],
+        run: vi.fn().mockResolvedValue({
+          output: {
+            dims: [1, 2, 2],
+            data: new Float32Array([0.1, 0.9, 0.9, 0.1])
+          }
+        })
+      },
+      provider: "wasm"
+    });
+    const released = [];
+    releaseSessions.mockImplementation(async (session) => {
+      released.push(session);
+    });
+
+    const { createRecModel, createRecModelSession } = await loadRecModule();
+    const sessionState = await createRecModelSession({}, new Uint8Array([1]), "auto", {
+      available: false,
+      reason: ""
+    });
+    expect(getProviderCandidates).toHaveBeenCalledWith("auto", { available: false, reason: "" });
+    expect(withTimeout).toHaveBeenCalled();
+    expect(sessionState.provider).toBe("wasm");
+
+    const model = await createRecModel({
+      ort: {
+        Tensor: createMockOrtTensorClass()
+      },
+      modelBytes: new Uint8Array([1]),
+      configText: "config",
+      backend: "auto",
+      webgpuState: { available: false, reason: "" }
+    });
+
+    expect(assertModelResources).toHaveBeenCalled();
+    expect(model.kind).toBe("rec");
+    expect(model.provider).toBe("wasm");
+    expect(model.config.charDict).toEqual(["A", " "]);
+
+    const cvFixture = createRecPredictCvStub();
+    toBgrFloatCHWFromBgr.mockImplementation((data, width, height) => {
+      const out = new Float32Array(3 * width * height);
+      for (let i = 0; i < out.length; i += 1) out[i] = 1;
+      return out;
+    });
+    await expect(model.predict(cvFixture, [createMat(3, 8, 4)])).resolves.toSatisfy((results) => {
+      expect(results).toHaveLength(1);
+      expect(results[0]).toMatchObject({ text: "A" });
+      expect(results[0].score).toBeCloseTo(0.9, 5);
+      return true;
+    });
+    await expect(model.dispose()).resolves.toBeUndefined();
+    expect(released.at(-1)).toBeTruthy();
+    await expect(model.predict(cvFixture, [createMat(3, 8, 4)])).rejects.toThrow(
+      /session is not initialized/i
+    );
+  });
+});
diff --git a/paddleocr-js/packages/core/test/resolve.test.ts b/paddleocr-js/packages/core/test/resolve.test.ts
new file mode 100644
index 0000000..f03dc3d
--- /dev/null
+++ b/paddleocr-js/packages/core/test/resolve.test.ts
@@ -0,0 +1,56 @@
+import { describe, expect, it, vi } from "vitest";
+
+import { loadModelAsset } from "../src/resources/model-asset";
+import { createTar } from "./tar-fixture";
+
+describe("model asset loading", () => {
+  it("resolves model tar assets", async () => {
+    const asset = {
+      url: "/det.tar"
+    };
+
+    const fetchImpl = vi.fn(async (url) => {
+      if (url.endsWith("det.tar")) {
+        return new Response(
+          createTar([
+            { name: "nested/inference.onnx", content: new Uint8Array([57, 56, 55]) },
+            { name: "nested/inference.yml", content: "name: det" }
+          ])
+        );
+      }
+      return new Response(
+        createTar([
+          { name: "nested/inference.onnx", content: new Uint8Array([49, 50, 51]) },
+          { name: "nested/inference.yml", content: "name: det" }
+        ])
+      );
+    });
+
+    const resolved = await loadModelAsset(asset, fetchImpl);
+
+    expect(Array.from(resolved.modelBytes)).toEqual([57, 56, 55]);
+    expect(resolved.configText).toBe("name: det");
+    expect(resolved.download.url).toBe("/det.tar");
+  });
+
+  it("uses standard inference entry names", async () => {
+    const asset = {
+      url: "/det-default.tar"
+    };
+
+    const fetchImpl = vi.fn(
+      async () =>
+        new Response(
+          createTar([
+            { name: "nested/inference.onnx", content: new Uint8Array([1, 2, 3]) },
+            { name: "nested/inference.yml", content: "name: config" }
+          ])
+        )
+    );
+
+    const resolved = await loadModelAsset(asset, fetchImpl);
+
+    expect(Array.from(resolved.modelBytes)).toEqual([1, 2, 3]);
+    expect(resolved.configText).toBe("name: config");
+  });
+});
diff --git a/paddleocr-js/packages/core/test/runtime-opencv.test.ts b/paddleocr-js/packages/core/test/runtime-opencv.test.ts
new file mode 100644
index 0000000..deac34e
--- /dev/null
+++ b/paddleocr-js/packages/core/test/runtime-opencv.test.ts
@@ -0,0 +1,61 @@
+import { afterEach, describe, expect, it, vi } from "vitest";
+
+async function loadOpenCvModule(mockedDefault) {
+  vi.resetModules();
+  vi.doMock("@techstark/opencv-js", () => ({
+    __esModule: true,
+    default: mockedDefault
+  }));
+  return import("../src/runtime/opencv");
+}
+
+afterEach(() => {
+  vi.resetModules();
+  vi.doUnmock("@techstark/opencv-js");
+});
+
+describe("runtime/opencv", () => {
+  it("returns OpenCV immediately when the module already exposes Mat", async () => {
+    const cv = { Mat() {} };
+    const { initOpenCvRuntime } = await loadOpenCvModule(cv);
+
+    await expect(initOpenCvRuntime()).resolves.toEqual({ cv });
+  });
+
+  it("awaits promised OpenCV modules", async () => {
+    const cv = { Mat() {} };
+    const { initOpenCvRuntime } = await loadOpenCvModule(Promise.resolve(cv));
+
+    await expect(initOpenCvRuntime()).resolves.toEqual({ cv });
+  });
+
+  it("awaits callback-based OpenCV initialization", async () => {
+    const cv = {};
+    Object.defineProperty(cv, "onRuntimeInitialized", {
+      set(callback) {
+        callback();
+      }
+    });
+    const { initOpenCvRuntime } = await loadOpenCvModule(cv);
+
+    await expect(initOpenCvRuntime()).resolves.toEqual({ cv });
+  });
+
+  it("caches successful initialization results", async () => {
+    const cv = { Mat() {} };
+    const { initOpenCvRuntime } = await loadOpenCvModule(cv);
+
+    const first = await initOpenCvRuntime();
+    const second = await initOpenCvRuntime();
+
+    expect(first).toBe(second);
+  });
+
+  it("re-exports initOpenCvRuntime from the runtime index", async () => {
+    const cv = { Mat() {} };
+    const runtimeModule = await loadOpenCvModule(cv);
+    const runtimeIndex = await import("../src/runtime/index");
+
+    expect(runtimeIndex.initOpenCvRuntime).toBe(runtimeModule.initOpenCvRuntime);
+  });
+});
diff --git a/paddleocr-js/packages/core/test/runtime-ort.test.ts b/paddleocr-js/packages/core/test/runtime-ort.test.ts
new file mode 100644
index 0000000..ac4b56c
--- /dev/null
+++ b/paddleocr-js/packages/core/test/runtime-ort.test.ts
@@ -0,0 +1,174 @@
+import { afterEach, describe, expect, it, vi } from "vitest";
+
+async function loadOrtModule(mockedOrt = { env: { wasm: {} } }) {
+  vi.resetModules();
+  vi.doMock("onnxruntime-web", () => ({
+    __esModule: true,
+    ...mockedOrt
+  }));
+  return import("../src/runtime/ort");
+}
+
+afterEach(() => {
+  vi.resetModules();
+  vi.doUnmock("onnxruntime-web");
+  vi.unstubAllGlobals();
+});
+
+describe("runtime/ort", () => {
+  it("reports when WebGPU is unavailable", async () => {
+    vi.stubGlobal("navigator", {});
+    const { detectWebGpuAvailability } = await loadOrtModule();
+
+    await expect(detectWebGpuAvailability()).resolves.toEqual({
+      available: false,
+      reason: "navigator.gpu is unavailable in this browser."
+    });
+  });
+
+  it("reports WebGPU available when the browser returns an adapter", async () => {
+    vi.stubGlobal("navigator", {
+      gpu: {
+        requestAdapter: vi.fn().mockResolvedValue({})
+      }
+    });
+    const { detectWebGpuAvailability } = await loadOrtModule();
+    await expect(detectWebGpuAvailability()).resolves.toEqual({
+      available: true,
+      reason: ""
+    });
+  });
+
+  it("reports WebGPU unavailable when the browser returns no adapter", async () => {
+    vi.stubGlobal("navigator", {
+      gpu: {
+        requestAdapter: vi.fn().mockResolvedValue(null)
+      }
+    });
+    const { detectWebGpuAvailability } = await loadOrtModule();
+    await expect(detectWebGpuAvailability()).resolves.toEqual({
+      available: false,
+      reason: "The browser did not return a WebGPU adapter."
+    });
+  });
+
+  it("reports WebGPU unavailable when adapter lookup throws", async () => {
+    vi.stubGlobal("navigator", {
+      gpu: {
+        requestAdapter: vi.fn().mockRejectedValue(new Error("GPU blocked"))
+      }
+    });
+    const { detectWebGpuAvailability } = await loadOrtModule();
+    await expect(detectWebGpuAvailability()).resolves.toEqual({
+      available: false,
+      reason: "GPU blocked"
+    });
+  });
+
+  it("selects provider candidates from backend preference and WebGPU availability", async () => {
+    const { getProviderCandidates } = await loadOrtModule();
+
+    expect(getProviderCandidates("webgpu", { available: true, reason: "" })).toEqual([["webgpu"]]);
+    expect(getProviderCandidates("wasm", { available: false, reason: "n/a" })).toEqual([["wasm"]]);
+    expect(getProviderCandidates("auto", { available: true, reason: "" })).toEqual([
+      ["webgpu"],
+      ["wasm"]
+    ]);
+    expect(getProviderCandidates("auto", { available: false, reason: "n/a" })).toEqual([["wasm"]]);
+    expect(() => getProviderCandidates("webgpu", { available: false, reason: "missing" })).toThrow(
+      /WebGPU is unavailable: missing/
+    );
+  });
+
+  it("initializes ORT runtime and applies wasm environment options", async () => {
+    vi.stubGlobal("navigator", {});
+    const mockedOrt = {
+      env: {
+        wasm: {}
+      }
+    };
+    const { initOrtRuntime } = await loadOrtModule(mockedOrt);
+
+    const runtime = await initOrtRuntime({
+      wasmPaths: "/wasm/",
+      numThreads: 2,
+      simd: true,
+      proxy: true,
+      disableWasmProxy: true
+    });
+
+    expect(runtime.backend).toBe("auto");
+    expect(runtime.webgpuState.available).toBe(false);
+    expect(runtime.ort.env.wasm).toEqual({
+      wasmPaths: "/wasm/",
+      numThreads: 2,
+      simd: true,
+      proxy: false
+    });
+    expect(mockedOrt.env.wasm).toEqual({
+      wasmPaths: "/wasm/",
+      numThreads: 2,
+      simd: true,
+      proxy: false
+    });
+  });
+
+  it("creates sessions by trying providers in order", async () => {
+    const create = vi
+      .fn()
+      .mockRejectedValueOnce(new Error("webgpu failed"))
+      .mockResolvedValueOnce({ id: "session" });
+    const mockedOrt = {
+      InferenceSession: { create }
+    };
+    const { createSession } = await loadOrtModule({
+      env: { wasm: {} },
+      InferenceSession: { create }
+    });
+    const modelBytes = new Uint8Array([1, 2, 3]);
+
+    const result = await createSession(mockedOrt, modelBytes, [["webgpu"], ["wasm"]]);
+
+    expect(create).toHaveBeenNthCalledWith(1, modelBytes, {
+      executionProviders: ["webgpu"],
+      graphOptimizationLevel: "all"
+    });
+    expect(create).toHaveBeenNthCalledWith(2, modelBytes, {
+      executionProviders: ["wasm"],
+      graphOptimizationLevel: "all"
+    });
+    expect(result).toEqual({
+      session: { id: "session" },
+      provider: "wasm"
+    });
+  });
+
+  it("throws the last session creation error when all providers fail", async () => {
+    const create = vi
+      .fn()
+      .mockRejectedValueOnce(new Error("webgpu failed"))
+      .mockRejectedValueOnce(new Error("wasm failed"));
+    const mockedOrt = {
+      InferenceSession: { create }
+    };
+    const { createSession } = await loadOrtModule({
+      env: { wasm: {} },
+      InferenceSession: { create }
+    });
+
+    await expect(
+      createSession(mockedOrt, new Uint8Array([1]), [["webgpu"], ["wasm"]])
+    ).rejects.toThrow("wasm failed");
+  });
+
+  it("releases every session that exposes a release method", async () => {
+    const releaseA = vi.fn().mockResolvedValue(undefined);
+    const releaseB = vi.fn().mockResolvedValue(undefined);
+    const { releaseSessions } = await loadOrtModule();
+
+    await releaseSessions({ release: releaseA }, null, { release: releaseB }, {});
+
+    expect(releaseA).toHaveBeenCalledTimes(1);
+    expect(releaseB).toHaveBeenCalledTimes(1);
+  });
+});
diff --git a/paddleocr-js/packages/core/test/runtime-params.test.ts b/paddleocr-js/packages/core/test/runtime-params.test.ts
new file mode 100644
index 0000000..6bdf99e
--- /dev/null
+++ b/paddleocr-js/packages/core/test/runtime-params.test.ts
@@ -0,0 +1,103 @@
+import { describe, expect, it } from "vitest";
+
+import { DEFAULT_DET_MODEL_CONFIG, DEFAULT_REC_MODEL_CONFIG } from "../src/models/index";
+import type { OcrModelConfig } from "../src/pipelines/ocr/runtime-params";
+import { getOcrRuntimeParams } from "../src/pipelines/ocr/runtime-params";
+
+/** Fixture: full `OcrModelConfig` shapes (same as pipeline `getModelConfig()` after init). */
+const SAMPLE_MODEL_CONFIG: OcrModelConfig = {
+  det: {
+    ...DEFAULT_DET_MODEL_CONFIG,
+    resizeLong: 960,
+    maxSideLimit: 3200
+  },
+  rec: {
+    ...DEFAULT_REC_MODEL_CONFIG
+  }
+};
+
+describe("OCR runtime params", () => {
+  it("falls back from params to defaults and config values", () => {
+    expect(
+      getOcrRuntimeParams(
+        SAMPLE_MODEL_CONFIG,
+        {
+          text_det_limit_side_len: 736,
+          text_det_limit_type: "min",
+          text_det_max_side_limit: 4096,
+          text_det_thresh: 0.25,
+          text_det_box_thresh: 0.55,
+          text_det_unclip_ratio: 1.8,
+          text_rec_score_thresh: 0.4
+        },
+        {}
+      )
+    ).toEqual({
+      det: {
+        limitSideLen: 736,
+        limitType: "min",
+        maxSideLimit: 4096,
+        thresh: 0.25,
+        boxThresh: 0.55,
+        unclipRatio: 1.8
+      },
+      pipeline: { scoreThresh: 0.4 }
+    });
+  });
+
+  it("prefers camelCase params; limit type falls back to model config when not specified", () => {
+    expect(
+      getOcrRuntimeParams(
+        SAMPLE_MODEL_CONFIG,
+        {},
+        {
+          textDetLimitSideLen: 512,
+          textDetMaxSideLimit: 2048,
+          textDetThresh: 0.22,
+          textDetBoxThresh: 0.44,
+          textDetUnclipRatio: 2.2,
+          textRecScoreThresh: 0.9
+        }
+      )
+    ).toEqual({
+      det: {
+        limitSideLen: 512,
+        limitType: DEFAULT_DET_MODEL_CONFIG.limitType,
+        maxSideLimit: 2048,
+        thresh: 0.22,
+        boxThresh: 0.44,
+        unclipRatio: 2.2
+      },
+      pipeline: { scoreThresh: 0.9 }
+    });
+  });
+
+  it("treats nulls as missing values when resolving fallbacks", () => {
+    expect(
+      getOcrRuntimeParams(
+        SAMPLE_MODEL_CONFIG,
+        {
+          text_det_limit_type: null
+        },
+        {
+          text_det_limit_side_len: null,
+          text_det_max_side_limit: null,
+          text_det_thresh: null,
+          text_det_box_thresh: null,
+          text_det_unclip_ratio: null,
+          text_rec_score_thresh: null
+        }
+      )
+    ).toEqual({
+      det: {
+        limitSideLen: SAMPLE_MODEL_CONFIG.det.resizeLong,
+        limitType: SAMPLE_MODEL_CONFIG.det.limitType,
+        maxSideLimit: SAMPLE_MODEL_CONFIG.det.maxSideLimit,
+        thresh: SAMPLE_MODEL_CONFIG.det.postprocess.thresh,
+        boxThresh: SAMPLE_MODEL_CONFIG.det.postprocess.boxThresh,
+        unclipRatio: SAMPLE_MODEL_CONFIG.det.postprocess.unclipRatio
+      },
+      pipeline: { scoreThresh: 0 }
+    });
+  });
+});
diff --git a/paddleocr-js/packages/core/test/tar-fixture.ts b/paddleocr-js/packages/core/test/tar-fixture.ts
new file mode 100644
index 0000000..07d00e4
--- /dev/null
+++ b/paddleocr-js/packages/core/test/tar-fixture.ts
@@ -0,0 +1,47 @@
+function writeString(target, offset, length, value) {
+  const encoded = new TextEncoder().encode(value);
+  target.set(encoded.slice(0, length), offset);
+}
+
+function writeOctal(target, offset, length, value) {
+  const octal = value.toString(8).padStart(length - 1, "0");
+  writeString(target, offset, length, `${octal}\0`);
+}
+
+export function createTar(entries) {
+  const chunks = [];
+  for (const entry of entries) {
+    const data =
+      typeof entry.content === "string" ? new TextEncoder().encode(entry.content) : entry.content;
+    const header = new Uint8Array(512);
+    writeString(header, 0, 100, entry.name);
+    writeString(header, 100, 8, "0000777");
+    writeString(header, 108, 8, "0000000");
+    writeString(header, 116, 8, "0000000");
+    writeOctal(header, 124, 12, data.length);
+    writeString(header, 136, 12, "00000000000");
+    header[156] = "0".charCodeAt(0);
+    writeString(header, 257, 6, "ustar");
+    writeString(header, 263, 2, "00");
+    for (let index = 148; index < 156; index += 1) {
+      header[index] = 32;
+    }
+    const checksum = header.reduce((sum, value) => sum + value, 0);
+    writeString(header, 148, 8, `${checksum.toString(8).padStart(6, "0")}\0 `);
+    chunks.push(header, data);
+    const remainder = data.length % 512;
+    if (remainder !== 0) {
+      chunks.push(new Uint8Array(512 - remainder));
+    }
+  }
+  chunks.push(new Uint8Array(1024));
+
+  const total = chunks.reduce((sum, chunk) => sum + chunk.length, 0);
+  const out = new Uint8Array(total);
+  let offset = 0;
+  for (const chunk of chunks) {
+    out.set(chunk, offset);
+    offset += chunk.length;
+  }
+  return out.buffer;
+}
diff --git a/paddleocr-js/packages/core/test/tar.test.ts b/paddleocr-js/packages/core/test/tar.test.ts
new file mode 100644
index 0000000..e58414f
--- /dev/null
+++ b/paddleocr-js/packages/core/test/tar.test.ts
@@ -0,0 +1,37 @@
+import { describe, expect, it } from "vitest";
+
+import { extractTarEntries, pickTarEntry } from "../src/resources/tar";
+import { createTar } from "./tar-fixture";
+
+describe("tar helpers", () => {
+  it("extracts nested inference files from a tar archive", () => {
+    const tarBuffer = createTar([
+      { name: "det/inference.onnx", content: "onnx" },
+      { name: "det/inference.yml", content: "yaml" }
+    ]);
+
+    const entries = extractTarEntries(tarBuffer);
+
+    expect(new TextDecoder().decode(pickTarEntry(entries, "inference.onnx"))).toBe("onnx");
+    expect(new TextDecoder().decode(pickTarEntry(entries, "inference.yml"))).toBe("yaml");
+  });
+
+  it("ignores AppleDouble and PaxHeader metadata entries", () => {
+    const tarBuffer = createTar([
+      { name: "._det", content: "metadata" },
+      { name: "PaxHeader/det", content: "metadata" },
+      { name: "det/._inference.onnx", content: "metadata" },
+      { name: "det/PaxHeader/inference.onnx", content: "metadata" },
+      { name: "det/inference.onnx", content: "onnx" },
+      { name: "det/._inference.yml", content: "metadata" },
+      { name: "det/PaxHeader/inference.yml", content: "metadata" },
+      { name: "det/inference.yml", content: "yaml" }
+    ]);
+
+    const entries = extractTarEntries(tarBuffer);
+
+    expect([...entries.keys()]).toEqual(["det/inference.onnx", "det/inference.yml"]);
+    expect(new TextDecoder().decode(pickTarEntry(entries, "inference.onnx"))).toBe("onnx");
+    expect(new TextDecoder().decode(pickTarEntry(entries, "inference.yml"))).toBe("yaml");
+  });
+});
diff --git a/paddleocr-js/packages/core/test/utils-common.test.ts b/paddleocr-js/packages/core/test/utils-common.test.ts
new file mode 100644
index 0000000..ad44782
--- /dev/null
+++ b/paddleocr-js/packages/core/test/utils-common.test.ts
@@ -0,0 +1,84 @@
+import { describe, expect, it, vi } from "vitest";
+
+import {
+  clamp,
+  deepClone,
+  distance2,
+  formatMs,
+  nowMs,
+  resolveRuntimeBatchSize,
+  withTimeout
+} from "../src/utils/common";
+
+describe("utils/common", () => {
+  it("reads the current high-resolution timestamp", () => {
+    const spy = vi.spyOn(performance, "now").mockReturnValue(123.45);
+
+    expect(nowMs()).toBe(123.45);
+
+    spy.mockRestore();
+  });
+
+  it("clamps values into the requested range", () => {
+    expect(clamp(-1, 0, 10)).toBe(0);
+    expect(clamp(5, 0, 10)).toBe(5);
+    expect(clamp(15, 0, 10)).toBe(10);
+  });
+
+  it("computes Euclidean distance between two points", () => {
+    expect(distance2([0, 0], [3, 4])).toBe(5);
+  });
+
+  it("formats millisecond durations with one decimal place", () => {
+    expect(formatMs(12.34)).toBe("12.3 ms");
+  });
+
+  it("resolveRuntimeBatchSize uses default when override is undefined", () => {
+    expect(resolveRuntimeBatchSize(undefined, 4)).toBe(4);
+  });
+
+  it("resolveRuntimeBatchSize coerces numeric strings and enforces a minimum of 1", () => {
+    expect(resolveRuntimeBatchSize("3", 1)).toBe(3);
+    expect(resolveRuntimeBatchSize(0, 2)).toBe(1);
+    expect(resolveRuntimeBatchSize(-1, 2)).toBe(1);
+    expect(resolveRuntimeBatchSize(Number.NaN, 2)).toBe(1);
+  });
+
+  it("resolves before the timeout when the promise settles in time", async () => {
+    await expect(withTimeout(Promise.resolve("ok"), 50, "load")).resolves.toBe("ok");
+  });
+
+  it("rejects with the original error before the timeout elapses", async () => {
+    const error = new Error("boom");
+
+    await expect(withTimeout(Promise.reject(error), 50, "load")).rejects.toBe(error);
+  });
+
+  it("rejects when the timeout elapses first", async () => {
+    await expect(withTimeout(new Promise(() => {}), 1, "load")).rejects.toThrow(
+      "load timed out after 0.001s"
+    );
+  });
+
+  it("deep clones structured data", () => {
+    const original = {
+      nested: {
+        values: [1, 2, 3]
+      }
+    };
+
+    const cloned = deepClone(original);
+    cloned.nested.values.push(4);
+
+    expect(cloned).toEqual({
+      nested: {
+        values: [1, 2, 3, 4]
+      }
+    });
+    expect(original).toEqual({
+      nested: {
+        values: [1, 2, 3]
+      }
+    });
+  });
+});
diff --git a/paddleocr-js/packages/core/test/viz-canvas-factory.test.ts b/paddleocr-js/packages/core/test/viz-canvas-factory.test.ts
new file mode 100644
index 0000000..bfc49d4
--- /dev/null
+++ b/paddleocr-js/packages/core/test/viz-canvas-factory.test.ts
@@ -0,0 +1,44 @@
+import { describe, expect, it, vi, afterEach } from "vitest";
+import { createCanvas } from "../src/viz/canvas-factory";
+
+describe("viz/canvas-factory", () => {
+  const origOffscreen = globalThis.OffscreenCanvas;
+
+  afterEach(() => {
+    if (origOffscreen) {
+      globalThis.OffscreenCanvas = origOffscreen;
+    } else {
+      delete (globalThis as Record<string, unknown>).OffscreenCanvas;
+    }
+  });
+
+  it("returns an OffscreenCanvas when available", () => {
+    const mockCanvas = { width: 0, height: 0, getContext: vi.fn() };
+    (globalThis as Record<string, unknown>).OffscreenCanvas = vi.fn((w: number, h: number) => {
+      mockCanvas.width = w;
+      mockCanvas.height = h;
+      return mockCanvas;
+    });
+
+    const result = createCanvas(100, 200);
+    expect(result.width).toBe(100);
+    expect(result.height).toBe(200);
+  });
+
+  it("falls back to document.createElement when OffscreenCanvas is unavailable", () => {
+    delete (globalThis as Record<string, unknown>).OffscreenCanvas;
+
+    const mockCtx = {};
+    const mockCanvas = {
+      width: 0,
+      height: 0,
+      getContext: vi.fn(() => mockCtx)
+    };
+    vi.spyOn(document, "createElement").mockReturnValue(mockCanvas as unknown as HTMLElement);
+
+    const result = createCanvas(300, 400);
+    expect(result.width).toBe(300);
+    expect(result.height).toBe(400);
+    expect(document.createElement).toHaveBeenCalledWith("canvas");
+  });
+});
diff --git a/paddleocr-js/packages/core/test/viz-color.test.ts b/paddleocr-js/packages/core/test/viz-color.test.ts
new file mode 100644
index 0000000..8d7cb13
--- /dev/null
+++ b/paddleocr-js/packages/core/test/viz-color.test.ts
@@ -0,0 +1,51 @@
+import { describe, expect, it } from "vitest";
+import { deterministicColor } from "../src/viz/color";
+
+describe("viz/color", () => {
+  it("returns an RGB tuple of three integers 0-255", () => {
+    const [r, g, b] = deterministicColor(0);
+    expect(Number.isInteger(r)).toBe(true);
+    expect(Number.isInteger(g)).toBe(true);
+    expect(Number.isInteger(b)).toBe(true);
+    expect(r).toBeGreaterThanOrEqual(0);
+    expect(r).toBeLessThanOrEqual(255);
+    expect(g).toBeGreaterThanOrEqual(0);
+    expect(g).toBeLessThanOrEqual(255);
+    expect(b).toBeGreaterThanOrEqual(0);
+    expect(b).toBeLessThanOrEqual(255);
+  });
+
+  it("produces the same color for the same index", () => {
+    expect(deterministicColor(5)).toEqual(deterministicColor(5));
+    expect(deterministicColor(42)).toEqual(deterministicColor(42));
+  });
+
+  it("produces different colors for different indices", () => {
+    const c0 = deterministicColor(0);
+    const c1 = deterministicColor(1);
+    const c2 = deterministicColor(2);
+    const allSame =
+      JSON.stringify(c0) === JSON.stringify(c1) && JSON.stringify(c1) === JSON.stringify(c2);
+    expect(allSame).toBe(false);
+  });
+
+  it("matches the exact values from the demo app's LCG", () => {
+    const c0 = deterministicColor(0);
+    const c1 = deterministicColor(1);
+    // Verify determinism by snapshot
+    expect(c0).toMatchInlineSnapshot(`
+      [
+        198,
+        126,
+        223,
+      ]
+    `);
+    expect(c1).toMatchInlineSnapshot(`
+      [
+        140,
+        33,
+        85,
+      ]
+    `);
+  });
+});
diff --git a/paddleocr-js/packages/core/test/viz-draw-boxes.test.ts b/paddleocr-js/packages/core/test/viz-draw-boxes.test.ts
new file mode 100644
index 0000000..316e990
--- /dev/null
+++ b/paddleocr-js/packages/core/test/viz-draw-boxes.test.ts
@@ -0,0 +1,108 @@
+import { describe, expect, it, vi } from "vitest";
+import { drawBoxesPanel } from "../src/viz/ocr/draw-boxes";
+import type { OcrResultItem } from "../src/pipelines/ocr/core";
+import type { Point2D } from "../src/models/common";
+
+function createMockCtx() {
+  const calls: string[] = [];
+  return {
+    calls,
+    save: vi.fn(() => calls.push("save")),
+    restore: vi.fn(() => calls.push("restore")),
+    drawImage: vi.fn(() => calls.push("drawImage")),
+    beginPath: vi.fn(() => calls.push("beginPath")),
+    moveTo: vi.fn(() => calls.push("moveTo")),
+    lineTo: vi.fn(() => calls.push("lineTo")),
+    closePath: vi.fn(() => calls.push("closePath")),
+    fill: vi.fn(() => calls.push("fill")),
+    stroke: vi.fn(() => calls.push("stroke")),
+    lineWidth: 0,
+    strokeStyle: "",
+    fillStyle: ""
+  } as unknown as CanvasRenderingContext2D & { calls: string[] };
+}
+
+function makeItem(poly: Point2D[], text: string): OcrResultItem {
+  return { originalIndex: 0, poly, text, score: 0.95 };
+}
+
+describe("viz/draw-boxes", () => {
+  it("draws the source image at (0,0)", () => {
+    const ctx = createMockCtx();
+    const image = { width: 100, height: 50 } as ImageBitmap;
+
+    drawBoxesPanel(ctx, image, [], {});
+
+    expect(ctx.drawImage).toHaveBeenCalledWith(image, 0, 0);
+  });
+
+  it("draws a polygon fill and stroke for each item", () => {
+    const ctx = createMockCtx();
+    const image = { width: 100, height: 50 } as ImageBitmap;
+    const items: OcrResultItem[] = [
+      makeItem(
+        [
+          [10, 10],
+          [90, 10],
+          [90, 40],
+          [10, 40]
+        ],
+        "hello"
+      )
+    ];
+
+    drawBoxesPanel(ctx, image, items, {});
+
+    expect(ctx.beginPath).toHaveBeenCalled();
+    expect(ctx.moveTo).toHaveBeenCalledWith(10, 10);
+    expect(ctx.lineTo).toHaveBeenCalledWith(90, 10);
+    expect(ctx.lineTo).toHaveBeenCalledWith(90, 40);
+    expect(ctx.lineTo).toHaveBeenCalledWith(10, 40);
+    expect(ctx.closePath).toHaveBeenCalled();
+    expect(ctx.fill).toHaveBeenCalled();
+  });
+
+  it("wraps each item draw in save/restore", () => {
+    const ctx = createMockCtx();
+    const image = { width: 100, height: 50 } as ImageBitmap;
+    const items: OcrResultItem[] = [
+      makeItem(
+        [
+          [0, 0],
+          [10, 0],
+          [10, 10],
+          [0, 10]
+        ],
+        "a"
+      )
+    ];
+
+    drawBoxesPanel(ctx, image, items, {});
+
+    const saveIdx = ctx.calls.indexOf("save");
+    const restoreIdx = ctx.calls.lastIndexOf("restore");
+    expect(saveIdx).toBeGreaterThan(-1);
+    expect(restoreIdx).toBeGreaterThan(saveIdx);
+  });
+
+  it("uses custom colorFn when provided", () => {
+    const ctx = createMockCtx();
+    const image = { width: 100, height: 50 } as ImageBitmap;
+    const items: OcrResultItem[] = [
+      makeItem(
+        [
+          [0, 0],
+          [10, 0],
+          [10, 10],
+          [0, 10]
+        ],
+        "a"
+      )
+    ];
+    const colorFn = vi.fn().mockReturnValue([255, 0, 0]);
+
+    drawBoxesPanel(ctx, image, items, { colorFn });
+
+    expect(colorFn).toHaveBeenCalledWith(0);
+  });
+});
diff --git a/paddleocr-js/packages/core/test/viz-draw-text.test.ts b/paddleocr-js/packages/core/test/viz-draw-text.test.ts
new file mode 100644
index 0000000..91f2f22
--- /dev/null
+++ b/paddleocr-js/packages/core/test/viz-draw-text.test.ts
@@ -0,0 +1,90 @@
+import { describe, expect, it, vi } from "vitest";
+import { drawTextPanel } from "../src/viz/ocr/draw-text";
+import type { OcrResultItem } from "../src/pipelines/ocr/core";
+import type { Point2D } from "../src/models/common";
+
+function createMockCtx() {
+  return {
+    save: vi.fn(),
+    restore: vi.fn(),
+    fillRect: vi.fn(),
+    beginPath: vi.fn(),
+    moveTo: vi.fn(),
+    lineTo: vi.fn(),
+    closePath: vi.fn(),
+    stroke: vi.fn(),
+    fill: vi.fn(),
+    measureText: vi.fn().mockReturnValue({ width: 50 }),
+    fillText: vi.fn(),
+    translate: vi.fn(),
+    rotate: vi.fn(),
+    lineWidth: 0,
+    strokeStyle: "",
+    fillStyle: "",
+    font: "",
+    textBaseline: "" as CanvasTextBaseline
+  } as unknown as CanvasRenderingContext2D;
+}
+
+function makeItem(poly: Point2D[], text: string): OcrResultItem {
+  return { originalIndex: 0, poly, text, score: 0.95 };
+}
+
+describe("viz/draw-text", () => {
+  it("fills the panel background", () => {
+    const ctx = createMockCtx();
+    drawTextPanel(ctx, 200, 100, [], {}, "sans-serif");
+
+    expect(ctx.fillRect).toHaveBeenCalledWith(200, 0, 200, 100);
+  });
+
+  it("draws text for each item", () => {
+    const ctx = createMockCtx();
+    const items: OcrResultItem[] = [
+      makeItem(
+        [
+          [10, 10],
+          [90, 10],
+          [90, 40],
+          [10, 40]
+        ],
+        "hello"
+      )
+    ];
+
+    drawTextPanel(ctx, 200, 100, items, {}, "sans-serif");
+
+    expect(ctx.fillText).toHaveBeenCalled();
+    // Verify the text content is "hello"
+    const textCall = (ctx.fillText as ReturnType<typeof vi.fn>).mock.calls[0];
+    expect(textCall[0]).toBe("hello");
+  });
+
+  it("draws box outlines on the right panel for each item", () => {
+    const ctx = createMockCtx();
+    const items: OcrResultItem[] = [
+      makeItem(
+        [
+          [10, 10],
+          [90, 10],
+          [90, 40],
+          [10, 40]
+        ],
+        "hello"
+      )
+    ];
+
+    drawTextPanel(ctx, 200, 100, items, {}, "sans-serif");
+
+    expect(ctx.beginPath).toHaveBeenCalled();
+    expect(ctx.stroke).toHaveBeenCalled();
+  });
+
+  it("uses custom textPanelBackground", () => {
+    const ctx = createMockCtx();
+    drawTextPanel(ctx, 200, 100, [], {}, "sans-serif", "#f0f0f0");
+
+    // fillRect call for the background should happen
+    expect(ctx.fillRect).toHaveBeenCalledWith(200, 0, 200, 100);
+  });
+});
diff --git a/paddleocr-js/packages/core/test/viz-font.test.ts b/paddleocr-js/packages/core/test/viz-font.test.ts
new file mode 100644
index 0000000..003215e
--- /dev/null
+++ b/paddleocr-js/packages/core/test/viz-font.test.ts
@@ -0,0 +1,74 @@
+import { describe, expect, it, vi, beforeEach } from "vitest";
+import { loadFontFace, removeFontFace } from "../src/viz/font";
+import type { FontConfig } from "../src/viz/types";
+
+describe("viz/font", () => {
+  let mockFontFace: { load: ReturnType<typeof vi.fn>; family: string };
+
+  beforeEach(() => {
+    mockFontFace = {
+      load: vi.fn().mockResolvedValue(undefined),
+      family: ""
+    };
+    (globalThis as Record<string, unknown>).FontFace = vi.fn((family: string) => {
+      mockFontFace.family = family;
+      return mockFontFace;
+    });
+    if (!document.fonts) {
+      Object.defineProperty(document, "fonts", {
+        value: { add: vi.fn(), delete: vi.fn() },
+        configurable: true
+      });
+    } else {
+      vi.spyOn(document.fonts, "add").mockImplementation(() => {});
+      vi.spyOn(document.fonts, "delete").mockImplementation(() => true);
+    }
+  });
+
+  it("loads a font from a URL string and adds to document.fonts", async () => {
+    const config: FontConfig = {
+      family: "TestFont",
+      source: "https://example.com/font.woff2"
+    };
+
+    const face = await loadFontFace(config);
+    expect(globalThis.FontFace).toHaveBeenCalledWith(
+      "TestFont",
+      "url(https://example.com/font.woff2)",
+      undefined
+    );
+    expect(mockFontFace.load).toHaveBeenCalled();
+    expect(document.fonts.add).toHaveBeenCalledWith(face);
+  });
+
+  it("loads a font from an ArrayBuffer source", async () => {
+    const buffer = new ArrayBuffer(8);
+    const config: FontConfig = {
+      family: "BufFont",
+      source: buffer
+    };
+
+    await loadFontFace(config);
+    expect(globalThis.FontFace).toHaveBeenCalledWith("BufFont", buffer, undefined);
+  });
+
+  it("passes descriptors to FontFace constructor", async () => {
+    const config: FontConfig = {
+      family: "DescFont",
+      source: "https://example.com/font.woff2",
+      descriptors: { weight: "bold" }
+    };
+
+    await loadFontFace(config);
+    expect(globalThis.FontFace).toHaveBeenCalledWith(
+      "DescFont",
+      "url(https://example.com/font.woff2)",
+      { weight: "bold" }
+    );
+  });
+
+  it("removes a font face from document.fonts", () => {
+    removeFontFace(mockFontFace as unknown as FontFace);
+    expect(document.fonts.delete).toHaveBeenCalledWith(mockFontFace);
+  });
+});
diff --git a/paddleocr-js/packages/core/test/viz-public-api.test.ts b/paddleocr-js/packages/core/test/viz-public-api.test.ts
new file mode 100644
index 0000000..31bde91
--- /dev/null
+++ b/paddleocr-js/packages/core/test/viz-public-api.test.ts
@@ -0,0 +1,28 @@
+import { describe, expect, it, vi } from "vitest";
+
+// Mock browser APIs needed by the module
+(globalThis as Record<string, unknown>).FontFace = vi.fn(() => ({
+  load: vi.fn().mockResolvedValue(undefined)
+}));
+if (!document.fonts) {
+  Object.defineProperty(document, "fonts", {
+    value: { add: vi.fn(), delete: vi.fn() },
+    configurable: true
+  });
+}
+
+import { OcrVisualizer, renderOcrToBlob, deterministicColor } from "../src/viz/index";
+
+describe("viz public API", () => {
+  it("exports OcrVisualizer as a class", () => {
+    expect(typeof OcrVisualizer).toBe("function");
+  });
+
+  it("exports renderOcrToBlob as a function", () => {
+    expect(typeof renderOcrToBlob).toBe("function");
+  });
+
+  it("exports deterministicColor as a function", () => {
+    expect(typeof deterministicColor).toBe("function");
+  });
+});
diff --git a/paddleocr-js/packages/core/test/viz-renderer.test.ts b/paddleocr-js/packages/core/test/viz-renderer.test.ts
new file mode 100644
index 0000000..64310c1
--- /dev/null
+++ b/paddleocr-js/packages/core/test/viz-renderer.test.ts
@@ -0,0 +1,55 @@
+import { describe, expect, it, vi } from "vitest";
+
+// Mock FontFace and document.fonts before importing
+const mockFontFace = {
+  load: vi.fn().mockResolvedValue(undefined),
+  family: ""
+};
+(globalThis as Record<string, unknown>).FontFace = vi.fn((family: string) => {
+  mockFontFace.family = family;
+  return mockFontFace;
+});
+if (!document.fonts) {
+  Object.defineProperty(document, "fonts", {
+    value: { add: vi.fn(), delete: vi.fn() },
+    configurable: true
+  });
+} else {
+  vi.spyOn(document.fonts, "add").mockImplementation(() => {});
+  vi.spyOn(document.fonts, "delete").mockImplementation(() => true);
+}
+
+// Mock createImageBitmap
+(globalThis as Record<string, unknown>).createImageBitmap = vi
+  .fn()
+  .mockResolvedValue({ width: 200, height: 100, close: vi.fn() });
+
+import { OcrVisualizer, renderOcrToBlob } from "../src/viz/ocr/renderer";
+
+describe("OcrVisualizer", () => {
+  it("can be constructed with no options", () => {
+    const viz = new OcrVisualizer();
+    expect(viz).toBeDefined();
+    viz.dispose();
+  });
+
+  it("can be constructed with font config", () => {
+    const viz = new OcrVisualizer({
+      font: { family: "TestFont", source: "https://example.com/f.woff2" }
+    });
+    expect(viz).toBeDefined();
+    viz.dispose();
+  });
+
+  it("dispose() is safe to call multiple times", () => {
+    const viz = new OcrVisualizer();
+    viz.dispose();
+    viz.dispose(); // should not throw
+  });
+});
+
+describe("renderOcrToBlob", () => {
+  it("is exported as a function", () => {
+    expect(typeof renderOcrToBlob).toBe("function");
+  });
+});
diff --git a/paddleocr-js/packages/core/test/viz-types.test.ts b/paddleocr-js/packages/core/test/viz-types.test.ts
new file mode 100644
index 0000000..3244b09
--- /dev/null
+++ b/paddleocr-js/packages/core/test/viz-types.test.ts
@@ -0,0 +1,32 @@
+import { describe, expect, it } from "vitest";
+import type { FontConfig, RgbColor } from "../src/viz/types";
+import type { BoxStyleOptions, OcrVisualizerOptions } from "../src/viz/ocr/types";
+
+describe("viz/types", () => {
+  it("allows constructing a minimal OcrVisualizerOptions", () => {
+    const opts: OcrVisualizerOptions = {};
+    expect(opts).toEqual({});
+  });
+
+  it("allows constructing a full OcrVisualizerOptions", () => {
+    const font: FontConfig = {
+      family: "Test",
+      source: "https://example.com/font.woff2"
+    };
+    const boxStyle: BoxStyleOptions = {
+      lineWidth: 3,
+      fillOpacity: 0.5,
+      colorFn: (i: number): RgbColor => [i, i, i]
+    };
+    const opts: OcrVisualizerOptions = {
+      font,
+      boxStyle,
+      textPanelBackground: "#f0f0f0",
+      outputFormat: "jpeg",
+      outputQuality: 0.8
+    };
+    expect(opts.font?.family).toBe("Test");
+    expect(opts.boxStyle?.lineWidth).toBe(3);
+    expect(opts.outputFormat).toBe("jpeg");
+  });
+});
diff --git a/paddleocr-js/packages/core/test/worker-backed.test.ts b/paddleocr-js/packages/core/test/worker-backed.test.ts
new file mode 100644
index 0000000..98d6b25
--- /dev/null
+++ b/paddleocr-js/packages/core/test/worker-backed.test.ts
@@ -0,0 +1,233 @@
+import { afterEach, describe, expect, it, vi } from "vitest";
+
+const sourceToWorkerPayload = vi.fn();
+
+vi.mock("../src/platform/browser", () => ({
+  sourceToWorkerPayload
+}));
+
+afterEach(() => {
+  vi.clearAllMocks();
+});
+
+function basePipelineConfig() {
+  return {
+    pipelineName: "OCR",
+    raw: {},
+    warnings: [] as string[],
+    unsupportedFeatures: [] as string[],
+    modelSelection: {
+      textDetectionModelName: "PP-OCRv5_mobile_det",
+      textRecognitionModelName: "PP-OCRv5_mobile_rec"
+    },
+    assets: {
+      det: { id: "det" },
+      rec: { id: "rec" }
+    },
+    runtimeDefaults: {} as Record<string, unknown>,
+    pipelineBatchSize: 1,
+    textDetectionBatchSize: 1,
+    textRecognitionBatchSize: 1
+  };
+}
+
+function createWorkerBackedOptions(overrides: Record<string, unknown> = {}) {
+  const { pipelineConfig: pipelineOverrides, ...rest } = overrides;
+  return {
+    pipelineConfig: {
+      ...basePipelineConfig(),
+      ...(pipelineOverrides as object)
+    },
+    ortOptions: {},
+    ...rest
+  };
+}
+
+describe("worker-backed OCR adapter", () => {
+  it("initializes once and forces wasm proxy off in worker mode", async () => {
+    const { WorkerBackedPaddleOCR } = await import("../src/pipelines/ocr/worker-backed");
+    const transportClient = {
+      request: vi.fn().mockResolvedValue({
+        summary: { backend: "wasm" },
+        modelConfig: { det: { name: "det" }, rec: { name: "rec" } }
+      }),
+      dispose: vi.fn()
+    };
+
+    const ocr = new WorkerBackedPaddleOCR(
+      createWorkerBackedOptions({
+        pipelineConfig: { runtimeDefaults: { text_det_limit_side_len: 64 } },
+        ortOptions: { proxy: true }
+      }),
+      transportClient
+    );
+
+    const first = await ocr.initialize();
+    const second = await ocr.initialize();
+
+    expect(first).toEqual({ backend: "wasm" });
+    expect(second).toEqual({ backend: "wasm" });
+    expect(transportClient.request).toHaveBeenCalledTimes(1);
+    expect(transportClient.request).toHaveBeenCalledWith("init", {
+      options: expect.objectContaining({
+        ortOptions: {
+          proxy: true,
+          disableWasmProxy: true
+        }
+      })
+    });
+    expect(ocr.getInitializationSummary()).toEqual({ backend: "wasm" });
+    expect(ocr.getModelConfig()).toEqual({ det: { name: "det" }, rec: { name: "rec" } });
+  });
+
+  it("predicts through the worker transport using transferable payloads", async () => {
+    const { WorkerBackedPaddleOCR } = await import("../src/pipelines/ocr/worker-backed");
+    const transferables = [{ id: "bitmap" }];
+    const mockOcrResult = {
+      image: { width: 1, height: 1 },
+      items: [{ poly: [[0, 0]], text: "hello", score: 1 }],
+      metrics: {
+        detMs: 1,
+        recMs: 1,
+        totalMs: 2,
+        detectedBoxes: 1,
+        recognizedCount: 1
+      },
+      runtime: {
+        requestedBackend: "auto",
+        detProvider: "wasm",
+        recProvider: "wasm",
+        webgpuAvailable: false
+      }
+    };
+    const transportClient = {
+      request: vi
+        .fn()
+        .mockResolvedValueOnce({
+          summary: { backend: "wasm" },
+          modelConfig: { det: {}, rec: {} }
+        })
+        .mockResolvedValueOnce([mockOcrResult]),
+      dispose: vi.fn()
+    };
+    sourceToWorkerPayload.mockResolvedValue({
+      payload: {
+        kind: "imageBitmap",
+        imageBitmap: transferables[0]
+      },
+      transferables
+    });
+
+    const ocr = new WorkerBackedPaddleOCR(createWorkerBackedOptions(), transportClient);
+
+    const result = await ocr.predict({ kind: "source" }, { text_rec_score_thresh: 0.5 });
+
+    expect(sourceToWorkerPayload).toHaveBeenCalledWith({ kind: "source" });
+    expect(transportClient.request).toHaveBeenNthCalledWith(
+      2,
+      "predict",
+      {
+        sources: [
+          {
+            kind: "imageBitmap",
+            imageBitmap: transferables[0]
+          }
+        ],
+        params: { text_rec_score_thresh: 0.5 }
+      },
+      transferables
+    );
+    expect(result).toEqual([mockOcrResult]);
+  });
+
+  it("predicts multiple sources when input is an array", async () => {
+    const { WorkerBackedPaddleOCR } = await import("../src/pipelines/ocr/worker-backed");
+    const mockOcrResult = {
+      image: { width: 10, height: 10 },
+      items: [],
+      metrics: {
+        detMs: 0,
+        recMs: 0,
+        totalMs: 0,
+        detectedBoxes: 0,
+        recognizedCount: 0
+      },
+      runtime: {
+        requestedBackend: "auto",
+        detProvider: "wasm",
+        recProvider: "wasm",
+        webgpuAvailable: false
+      }
+    };
+    const transportClient = {
+      request: vi
+        .fn()
+        .mockResolvedValueOnce({
+          summary: { backend: "wasm" },
+          modelConfig: { det: {}, rec: {} }
+        })
+        .mockResolvedValueOnce([
+          mockOcrResult,
+          { ...mockOcrResult, image: { width: 20, height: 20 } }
+        ]),
+      dispose: vi.fn()
+    };
+    sourceToWorkerPayload.mockResolvedValue({
+      payload: { kind: "imageBitmap", imageBitmap: {} },
+      transferables: []
+    });
+
+    const ocr = new WorkerBackedPaddleOCR(createWorkerBackedOptions(), transportClient);
+
+    await ocr.predict([{ a: 1 }, { b: 2 }], {});
+
+    expect(sourceToWorkerPayload).toHaveBeenCalledTimes(2);
+    expect(transportClient.request).toHaveBeenNthCalledWith(
+      2,
+      "predict",
+      {
+        sources: [
+          { kind: "imageBitmap", imageBitmap: {} },
+          { kind: "imageBitmap", imageBitmap: {} }
+        ],
+        params: {}
+      },
+      []
+    );
+  });
+
+  it("disposes the transport after initialization failures and allows retrying", async () => {
+    const { WorkerBackedPaddleOCR } = await import("../src/pipelines/ocr/worker-backed");
+    const transportClient = {
+      request: vi
+        .fn()
+        .mockRejectedValueOnce(new Error("init failed"))
+        .mockResolvedValueOnce({
+          summary: { backend: "wasm" },
+          modelConfig: { det: {}, rec: {} }
+        }),
+      dispose: vi.fn()
+    };
+
+    const ocr = new WorkerBackedPaddleOCR(createWorkerBackedOptions(), transportClient);
+
+    await expect(ocr.initialize()).rejects.toThrow("init failed");
+    expect(transportClient.dispose).toHaveBeenCalledTimes(1);
+    await expect(ocr.initialize()).resolves.toEqual({ backend: "wasm" });
+  });
+
+  it("swallows dispose request failures and rejects use after disposal", async () => {
+    const { WorkerBackedPaddleOCR } = await import("../src/pipelines/ocr/worker-backed");
+    const transportClient = {
+      request: vi.fn().mockRejectedValue(new Error("worker already gone")),
+      dispose: vi.fn()
+    };
+
+    const ocr = new WorkerBackedPaddleOCR(createWorkerBackedOptions(), transportClient);
+
+    await expect(ocr.dispose()).resolves.toBeUndefined();
+    expect(transportClient.request).toHaveBeenCalledWith("dispose", {});
+    expect(transportClient.dispose).toHaveBeenCalledTimes(1);
+    await expect(ocr.predict({}, {})).rejects.toThrow(/worker instance has been disposed/i);
+  });
+});
diff --git a/paddleocr-js/packages/core/test/worker-client.test.ts b/paddleocr-js/packages/core/test/worker-client.test.ts
new file mode 100644
index 0000000..577aa09
--- /dev/null
+++ b/paddleocr-js/packages/core/test/worker-client.test.ts
@@ -0,0 +1,132 @@
+import { describe, expect, it } from "vitest";
+
+import { WorkerTransportClient, createWorkerTransportClient } from "../src/worker/client";
+import { createTransportSuccess } from "../src/worker/protocol";
+
+class MockWorker {
+  constructor() {
+    this.messages = [];
+    this.terminated = false;
+    this.onmessage = null;
+    this.onerror = null;
+  }
+
+  postMessage(message, transferables) {
+    this.messages.push({ message, transferables });
+  }
+
+  terminate() {
+    this.terminated = true;
+  }
+}
+
+function createClientWithWorker() {
+  const worker = new MockWorker();
+  const client = new WorkerTransportClient({
+    createWorker: () => worker
+  });
+  return { worker, client };
+}
+
+describe("worker transport client", () => {
+  it("creates a worker lazily and resolves request responses", async () => {
+    const { worker, client } = createClientWithWorker();
+
+    const responsePromise = client.request("predict", { value: 1 }, ["xfer"]);
+    expect(worker.messages).toHaveLength(1);
+    expect(worker.messages[0]).toMatchObject({
+      message: {
+        kind: "worker-transport-request",
+        type: "predict",
+        payload: { value: 1 },
+        requestId: 1
+      },
+      transferables: ["xfer"]
+    });
+
+    worker.onmessage({
+      data: createTransportSuccess(1, { ok: true })
+    });
+
+    await expect(responsePromise).resolves.toEqual({ ok: true });
+  });
+
+  it("rejects when the worker reports an error response", async () => {
+    const { worker, client } = createClientWithWorker();
+
+    const responsePromise = client.request("predict", {});
+    worker.onmessage({
+      data: {
+        kind: "worker-transport-response",
+        status: "error",
+        requestId: 1,
+        error: {
+          name: "TypeError",
+          message: "bad payload",
+          stack: "stack"
+        }
+      }
+    });
+
+    await expect(responsePromise).rejects.toMatchObject({
+      name: "TypeError",
+      message: "bad payload",
+      stack: "stack"
+    });
+  });
+
+  it("ignores messages that are not transport responses", async () => {
+    const { worker, client } = createClientWithWorker();
+
+    const responsePromise = client.request("predict", {});
+    worker.onmessage({
+      data: { kind: "other" }
+    });
+    worker.onmessage({
+      data: createTransportSuccess(1, { ok: true })
+    });
+
+    await expect(responsePromise).resolves.toEqual({ ok: true });
+  });
+
+  it("rejects all pending requests when the worker errors", async () => {
+    const { worker, client } = createClientWithWorker();
+
+    const first = client.request("init", {});
+    const second = client.request("predict", {});
+    worker.onerror({ message: "worker crashed" });
+
+    await expect(first).rejects.toThrow("worker crashed");
+    await expect(second).rejects.toThrow("worker crashed");
+  });
+
+  it("rejects requests after disposal and terminates the worker", async () => {
+    const worker = new MockWorker();
+    const client = createWorkerTransportClient({
+      createWorker: () => worker
+    });
+
+    const pending = client.request("predict", {});
+    client.dispose();
+
+    await expect(pending).rejects.toThrow("Worker transport client has been disposed.");
+    expect(worker.terminated).toBe(true);
+    expect(() => client.dispose()).not.toThrow();
+    expect(() => client.ensureActive()).toThrow(/has been disposed/i);
+  });
+
+  it("requires a createWorker factory when the worker is first needed", () => {
+    const client = new WorkerTransportClient();
+
+    expect(() => client.ensureWorker()).toThrow(/requires a createWorker\(\) factory/i);
+  });
+
+  it("allows disposing before a worker has ever been created", () => {
+    const client = new WorkerTransportClient({
+      createWorker: () => new MockWorker()
+    });
+
+    expect(() => client.disposeWorker()).not.toThrow();
+    expect(() => client.dispose()).not.toThrow();
+  });
+});
diff --git a/paddleocr-js/packages/core/test/worker-entry.test.ts b/paddleocr-js/packages/core/test/worker-entry.test.ts
new file mode 100644
index 0000000..0bb6b94
--- /dev/null
+++ b/paddleocr-js/packages/core/test/worker-entry.test.ts
@@ -0,0 +1,76 @@
+import { describe, expect, it, vi } from "vitest";
+
+import { attachWorkerMessageHandler } from "../src/worker/entry";
+
+function createWorkerScope() {
+  return {
+    postMessage: vi.fn()
+  };
+}
+
+function transportRequest(overrides) {
+  return {
+    kind: "worker-transport-request",
+    type: "predict",
+    payload: {},
+    requestId: 4,
+    ...overrides
+  };
+}
+
+describe("worker entry message handler", () => {
+  it("ignores messages that do not use the transport protocol", async () => {
+    const scope = createWorkerScope();
+
+    attachWorkerMessageHandler(async () => ({}), scope);
+    await scope.onmessage({ data: { kind: "other" } });
+
+    expect(scope.postMessage).not.toHaveBeenCalled();
+  });
+
+  it("posts success responses for handled messages", async () => {
+    const scope = createWorkerScope();
+
+    attachWorkerMessageHandler(async (type, payload) => ({ type, payload }), scope);
+    await scope.onmessage({
+      data: transportRequest({
+        payload: { value: 1 },
+        requestId: 3
+      })
+    });
+
+    expect(scope.postMessage).toHaveBeenCalledWith({
+      kind: "worker-transport-response",
+      status: "success",
+      requestId: 3,
+      payload: {
+        type: "predict",
+        payload: { value: 1 }
+      }
+    });
+  });
+
+  it("posts error responses when the handler throws", async () => {
+    const scope = createWorkerScope();
+
+    attachWorkerMessageHandler(async () => {
+      throw new Error("boom");
+    }, scope);
+
+    await scope.onmessage({
+      data: transportRequest()
+    });
+
+    expect(scope.postMessage).toHaveBeenCalledWith(
+      expect.objectContaining({
+        kind: "worker-transport-response",
+        status: "error",
+        requestId: 4,
+        error: expect.objectContaining({
+          name: "Error",
+          message: "boom"
+        })
+      })
+    );
+  });
+});
diff --git a/paddleocr-js/packages/core/test/worker-protocol.test.ts b/paddleocr-js/packages/core/test/worker-protocol.test.ts
new file mode 100644
index 0000000..406c91f
--- /dev/null
+++ b/paddleocr-js/packages/core/test/worker-protocol.test.ts
@@ -0,0 +1,74 @@
+import { describe, expect, it } from "vitest";
+
+import {
+  createTransportError,
+  createTransportRequest,
+  createTransportSuccess,
+  deserializeError,
+  isTransportRequest,
+  isTransportResponse,
+  serializeError
+} from "../src/worker/protocol";
+
+describe("worker transport protocol", () => {
+  it("creates transport requests with the expected wire format", () => {
+    const request = createTransportRequest("predict", { value: 1 }, 7);
+
+    expect(request).toEqual({
+      kind: "worker-transport-request",
+      type: "predict",
+      payload: { value: 1 },
+      requestId: 7
+    });
+    expect(isTransportRequest(request)).toBe(true);
+    expect(isTransportResponse(request)).toBe(false);
+  });
+
+  it("creates success responses", () => {
+    const response = createTransportSuccess(7, { ok: true });
+
+    expect(response).toEqual({
+      kind: "worker-transport-response",
+      status: "success",
+      requestId: 7,
+      payload: { ok: true }
+    });
+    expect(isTransportResponse(response)).toBe(true);
+    expect(isTransportRequest(response)).toBe(false);
+  });
+
+  it("creates error responses by serializing the error", () => {
+    const response = createTransportError(9, new TypeError("bad payload"));
+
+    expect(response).toMatchObject({
+      kind: "worker-transport-response",
+      status: "error",
+      requestId: 9,
+      error: {
+        name: "TypeError",
+        message: "bad payload"
+      }
+    });
+  });
+
+  it("serializes unknown errors with defaults", () => {
+    expect(serializeError(null)).toEqual({
+      name: "Error",
+      message: "Unknown worker error.",
+      stack: ""
+    });
+  });
+
+  it("deserializes error payloads back into Error instances", () => {
+    const error = deserializeError({
+      name: "RangeError",
+      message: "out of range",
+      stack: "stack-trace"
+    });
+
+    expect(error).toBeInstanceOf(Error);
+    expect(error.name).toBe("RangeError");
+    expect(error.message).toBe("out of range");
+    expect(error.stack).toBe("stack-trace");
+  });
+});
diff --git a/paddleocr-js/packages/core/tsconfig.json b/paddleocr-js/packages/core/tsconfig.json
new file mode 100644
index 0000000..fa3db3e
--- /dev/null
+++ b/paddleocr-js/packages/core/tsconfig.json
@@ -0,0 +1,21 @@
+{
+  "compilerOptions": {
+    "target": "ES2022",
+    "module": "ESNext",
+    "moduleResolution": "bundler",
+    "strict": true,
+    "esModuleInterop": true,
+    "skipLibCheck": true,
+    "forceConsistentCasingInFileNames": true,
+    "declaration": true,
+    "declarationMap": true,
+    "sourceMap": true,
+    "isolatedModules": true,
+    "verbatimModuleSyntax": true,
+    "lib": ["ES2022", "DOM", "DOM.Iterable", "WebWorker"],
+    "outDir": "./dist",
+    "rootDir": "./src"
+  },
+  "include": ["src"],
+  "exclude": ["node_modules", "dist", "test"]
+}
diff --git a/paddleocr-js/packages/core/tsconfig.test.json b/paddleocr-js/packages/core/tsconfig.test.json
new file mode 100644
index 0000000..bba93f7
--- /dev/null
+++ b/paddleocr-js/packages/core/tsconfig.test.json
@@ -0,0 +1,10 @@
+{
+  "extends": "./tsconfig.json",
+  "compilerOptions": {
+    "rootDir": ".",
+    "declaration": false,
+    "declarationMap": false
+  },
+  "include": ["src", "test"],
+  "exclude": ["node_modules", "dist"]
+}
diff --git a/paddleocr-js/packages/core/vite.config.ts b/paddleocr-js/packages/core/vite.config.ts
new file mode 100644
index 0000000..451c6bd
--- /dev/null
+++ b/paddleocr-js/packages/core/vite.config.ts
@@ -0,0 +1,105 @@
+import { resolve, dirname, join } from "node:path";
+import { readFileSync } from "node:fs";
+import { createRequire } from "node:module";
+import { defineConfig } from "vite";
+import dts from "vite-plugin-dts";
+
+const require = createRequire(import.meta.url);
+const ortPackageDir = dirname(require.resolve("onnxruntime-web"));
+const ortVersion = JSON.parse(
+  readFileSync(join(ortPackageDir, "..", "package.json"), "utf-8")
+).version;
+
+// Post-process library output for npm compatibility.
+//
+// 1. Rewrite Vite's worker URL references so downstream bundlers handle them
+//    correctly (relative paths + split new Worker(new URL()) pattern).
+//
+// 2. Strip ORT's base64-encoded WASM data URIs from the worker asset.
+//    Vite converts ORT's `new URL('./ort-wasm-*.wasm', import.meta.url)`
+//    references into `data:application/wasm;base64,...` data URIs, inflating
+//    the worker by ~50 MB.  The SDK provides a CDN-based fallback for
+//    `ort.env.wasm.wasmPaths`, so the embedded binaries are never loaded.
+//    Only `data:application/wasm` URIs are stripped — OpenCV.js embeds its
+//    WASM via Emscripten as `data:application/octet-stream` and must be kept
+//    since there is no equivalent fallback for it.
+//
+//    The stripping must run here (main build generateBundle), not in
+//    worker.rollupOptions.plugins, because Vite injects the data URIs as a
+//    post-processing step after the worker's own Rollup pipeline finishes.
+function libraryWorkerPlugin() {
+  const ortWasmDataUriPattern = /data:application\/wasm;base64,[A-Za-z0-9+/=]+/g;
+
+  return {
+    name: "paddleocr-library-worker",
+    apply: "build",
+    generateBundle(_, bundle) {
+      for (const item of Object.values(bundle)) {
+        if (item.type === "asset" && typeof item.source === "string") {
+          item.source = item.source.replace(ortWasmDataUriPattern, "data:application/wasm;base64,");
+          continue;
+        }
+        if (item.type !== "chunk") continue;
+        if (!item.code.includes("/assets/")) continue;
+
+        item.code = item.code.replace(
+          /\/\*\s*@vite-ignore\s*\*\/\s*"\/(assets\/[^"]+)"/g,
+          '"./$1"'
+        );
+
+        item.code = item.code.replace(
+          /new\s+Worker\(\s*new\s+URL\(\s*"(\.\/(assets\/[^"]+))"\s*,\s*(import\.meta\.url)\s*\)\s*,\s*(\{[^}]*\})\s*\)/g,
+          (_, fullPath, _assetPath, urlBase, workerOpts) => {
+            return `(() => { const _w = new URL("${fullPath}", ${urlBase}); return new Worker(_w, ${workerOpts}); })()`;
+          }
+        );
+      }
+    }
+  };
+}
+
+export default defineConfig({
+  plugins: [
+    dts({
+      rollupTypes: false
+    }),
+    libraryWorkerPlugin()
+  ],
+  define: {
+    __ORT_WASM_CDN_PREFIX__: JSON.stringify(
+      `https://cdn.jsdelivr.net/npm/onnxruntime-web@${ortVersion}/dist/`
+    )
+  },
+  worker: {
+    format: "es",
+    rollupOptions: {
+      output: {
+        inlineDynamicImports: true
+      }
+    }
+  },
+  build: {
+    lib: {
+      entry: {
+        index: resolve(__dirname, "src/index.ts"),
+        viz: resolve(__dirname, "src/viz/index.ts")
+      },
+      name: "paddleocr",
+      formats: ["es"],
+      fileName: (_format, entryName) => `${entryName}.mjs`
+    },
+    rollupOptions: {
+      external: ["onnxruntime-web", "@techstark/opencv-js", "clipper-lib", "js-yaml"],
+      output: {
+        globals: {
+          "onnxruntime-web": "ort",
+          "@techstark/opencv-js": "cv",
+          "clipper-lib": "ClipperLib",
+          "js-yaml": "jsyaml"
+        }
+      }
+    },
+    sourcemap: true,
+    minify: false
+  }
+});
diff --git a/paddleocr-js/tsconfig.eslint.json b/paddleocr-js/tsconfig.eslint.json
new file mode 100644
index 0000000..969664a
--- /dev/null
+++ b/paddleocr-js/tsconfig.eslint.json
@@ -0,0 +1,14 @@
+{
+  "extends": "./packages/core/tsconfig.json",
+  "compilerOptions": {
+    "rootDir": ".",
+    "noEmit": true,
+    "baseUrl": ".",
+    "paths": {
+      "@paddleocr/paddleocr-js": ["./packages/core/src/index.ts"],
+      "@paddleocr/paddleocr-js/viz": ["./packages/core/src/viz/index.ts"]
+    }
+  },
+  "include": ["packages/**/src/**/*.ts", "packages/**/test/**/*.ts", "apps/**/src/**/*.ts"],
+  "exclude": ["node_modules", "dist"]
+}
diff --git a/paddleocr-js/tsconfig.json b/paddleocr-js/tsconfig.json
new file mode 100644
index 0000000..dfab9f5
--- /dev/null
+++ b/paddleocr-js/tsconfig.json
@@ -0,0 +1,4 @@
+{
+  "files": [],
+  "references": [{ "path": "packages/core" }, { "path": "apps/demo" }]
+}
diff --git a/paddleocr-js/vitest.config.js b/paddleocr-js/vitest.config.js
new file mode 100644
index 0000000..ac2909e
--- /dev/null
+++ b/paddleocr-js/vitest.config.js
@@ -0,0 +1,11 @@
+import { defineConfig } from "vitest/config";
+
+export default defineConfig({
+  test: {
+    environment: "jsdom",
+    coverage: {
+      provider: "v8",
+      include: ["packages/core/src/**/*.ts"]
+    }
+  }
+});
diff --git a/paddleocr/__init__.py b/paddleocr/__init__.py
new file mode 100644
index 0000000..a09deb1
--- /dev/null
+++ b/paddleocr/__init__.py
@@ -0,0 +1,131 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddlex.inference.utils.benchmark import benchmark
+
+from ._models import (
+    ChartParsing,
+    DocImgOrientationClassification,
+    DocVLM,
+    FormulaRecognition,
+    LayoutDetection,
+    SealTextDetection,
+    TableCellsDetection,
+    TableClassification,
+    TableStructureRecognition,
+    TextDetection,
+    TextImageUnwarping,
+    TextLineOrientationClassification,
+    TextRecognition,
+)
+from ._pipelines import (
+    DocPreprocessor,
+    DocUnderstanding,
+    FormulaRecognitionPipeline,
+    PaddleOCR,
+    PaddleOCRVL,
+    PPChatOCRv4Doc,
+    PPDocTranslation,
+    PPStructureV3,
+    SealRecognition,
+    TableRecognitionPipelineV2,
+)
+from ._api_client.async_client import AsyncPaddleOCRClient
+from ._api_client.client import PaddleOCRClient
+from ._api_client.errors import (
+    APIError,
+    AuthError,
+    InvalidRequestError,
+    JobFailedError,
+    NetworkError,
+    PaddleOCRAPIError,
+    PollTimeoutError,
+    RateLimitError,
+    RequestTimeoutError,
+    ResponseFormatError,
+    ResultParseError,
+    ServiceUnavailableError,
+)
+from ._api_client.models import (
+    Model,
+    OCROptions,
+    PaddleOCRVLOptions,
+    PPStructureV3Options,
+)
+from ._utils.logging import logger
+from ._version import version as __version__
+
+
+def doc2md_convert(source, **kwargs):
+    """Convert an office document to Markdown. See paddleocr._doc2md.convert."""
+    from ._doc2md import convert
+
+    return convert(source, **kwargs)
+
+
+def doc2md_supported_formats():
+    """Return supported file extensions. See paddleocr._doc2md.supported_formats."""
+    from ._doc2md import supported_formats
+
+    return supported_formats()
+
+
+__all__ = [
+    "benchmark",
+    "PaddleOCRClient",
+    "AsyncPaddleOCRClient",
+    "Model",
+    "OCROptions",
+    "PPStructureV3Options",
+    "PaddleOCRVLOptions",
+    "ChartParsing",
+    "DocImgOrientationClassification",
+    "DocVLM",
+    "FormulaRecognition",
+    "SealTextDetection",
+    "LayoutDetection",
+    "TableCellsDetection",
+    "TableClassification",
+    "TableStructureRecognition",
+    "TextDetection",
+    "TextImageUnwarping",
+    "TextLineOrientationClassification",
+    "TextRecognition",
+    "DocPreprocessor",
+    "DocUnderstanding",
+    "FormulaRecognitionPipeline",
+    "PaddleOCR",
+    "PaddleOCRVL",
+    "PPChatOCRv4Doc",
+    "PPDocTranslation",
+    "PPStructureV3",
+    "SealRecognition",
+    "TableRecognitionPipelineV2",
+    "doc2md_convert",
+    "doc2md_supported_formats",
+    "PaddleOCRAPIError",
+    "AuthError",
+    "InvalidRequestError",
+    "APIError",
+    "JobFailedError",
+    "RateLimitError",
+    "RequestTimeoutError",
+    "PollTimeoutError",
+    "ResponseFormatError",
+    "ResultParseError",
+    "ServiceUnavailableError",
+    "NetworkError",
+    "logger",
+    "__version__",
+]
diff --git a/paddleocr/__main__.py b/paddleocr/__main__.py
new file mode 100644
index 0000000..abc3c1b
--- /dev/null
+++ b/paddleocr/__main__.py
@@ -0,0 +1,39 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import sys
+
+from ._cli import main
+
+
+def console_entry() -> int:
+    # See https://docs.python.org/3/library/signal.html#note-on-sigpipe
+    try:
+        # Flush output here to force SIGPIPE to be triggered while inside this
+        # try block.
+        main()
+        sys.stdout.flush()
+        sys.stderr.flush()
+    except BrokenPipeError:
+        # Python flushes standard streams on exit;
+        # redirect remaining output to devnull to avoid another BrokenPipeError
+        # at shutdown.
+        devnull = os.open(os.devnull, os.O_WRONLY)
+        os.dup2(devnull, sys.stdout.fileno())
+        sys.exit(1)
+
+
+if __name__ == "__main__":
+    console_entry()
diff --git a/paddleocr/_abstract.py b/paddleocr/_abstract.py
new file mode 100644
index 0000000..773e4fb
--- /dev/null
+++ b/paddleocr/_abstract.py
@@ -0,0 +1,25 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import abc
+
+
+class CLISubcommandExecutor(metaclass=abc.ABCMeta):
+    @abc.abstractmethod
+    def add_subparser(self, subparsers):
+        raise NotImplementedError
+
+    @abc.abstractmethod
+    def execute_with_args(self, args):
+        raise NotImplementedError
diff --git a/paddleocr/_api_client/__init__.py b/paddleocr/_api_client/__init__.py
new file mode 100644
index 0000000..290f972
--- /dev/null
+++ b/paddleocr/_api_client/__init__.py
@@ -0,0 +1,13 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/paddleocr/_api_client/_async_http.py b/paddleocr/_api_client/_async_http.py
new file mode 100644
index 0000000..7d02bed
--- /dev/null
+++ b/paddleocr/_api_client/_async_http.py
@@ -0,0 +1,188 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import json
+import os
+from typing import Any, Dict, Optional
+
+import aiohttp
+
+from ._core import (
+    extract_api_message_from_payload,
+    extract_job_id,
+    raise_for_status,
+    unwrap_api_response,
+)
+from .errors import (
+    NetworkError,
+    RequestTimeoutError,
+    ResponseFormatError,
+    ResultParseError,
+)
+from ._http import API_PATH, DEFAULT_BASE_URL
+
+
+class AsyncHTTPClient:
+    def __init__(
+        self,
+        token: str,
+        base_url: str,
+        timeout: float,
+        client_platform: Optional[str] = None,
+    ):
+        self._token = token
+        self._base_url = base_url.rstrip("/")
+        self._jobs_url = f"{self._base_url}{API_PATH}"
+        self._timeout = timeout
+        self._client_platform = client_platform
+        self._session = None
+
+    @property
+    def timeout(self) -> float:
+        return self._timeout
+
+    async def __aenter__(self):
+        await self._ensure_session()
+        return self
+
+    async def __aexit__(self, *args):
+        await self.close()
+
+    async def _ensure_session(self):
+        if self._session is None:
+            self._session = aiohttp.ClientSession(
+                headers=self._api_headers(),
+                timeout=aiohttp.ClientTimeout(total=self._timeout),
+            )
+
+    async def close(self):
+        if self._session:
+            await self._session.close()
+            self._session = None
+
+    def _api_headers(self) -> dict:
+        headers = {"Authorization": f"Bearer {self._token}"}
+        if self._client_platform:
+            headers["Client-Platform"] = self._client_platform
+        return headers
+
+    async def submit_url(
+        self,
+        model: str,
+        file_url: str,
+        optional_payload: dict,
+        page_ranges: Optional[str] = None,
+        batch_id: Optional[str] = None,
+    ) -> str:
+        body = {
+            "fileUrl": file_url,
+            "model": model,
+            "optionalPayload": optional_payload,
+        }
+        if page_ranges is not None:
+            body["pageRanges"] = page_ranges
+        if batch_id is not None:
+            body["batchId"] = batch_id
+
+        await self._ensure_session()
+        async with self._session.post(
+            self._jobs_url,
+            json=body,
+            headers={"Content-Type": "application/json"},
+        ) as resp:
+            await self._raise_for_response(resp)
+            data = await self._response_data(resp)
+            return extract_job_id(data)
+
+    async def submit_file(
+        self,
+        model: str,
+        file_path: str,
+        optional_payload: dict,
+        page_ranges: Optional[str] = None,
+        batch_id: Optional[str] = None,
+    ) -> str:
+        if not os.path.exists(file_path):
+            raise FileNotFoundError(file_path)
+
+        form = aiohttp.FormData()
+        form.add_field("model", model)
+        form.add_field("optionalPayload", json.dumps(optional_payload))
+        if page_ranges is not None:
+            form.add_field("pageRanges", page_ranges)
+        if batch_id is not None:
+            form.add_field("batchId", batch_id)
+
+        with open(file_path, "rb") as f:
+            file_data = f.read()
+        form.add_field(
+            "file",
+            file_data,
+            filename=os.path.basename(file_path),
+        )
+
+        await self._ensure_session()
+        async with self._session.post(self._jobs_url, data=form) as resp:
+            await self._raise_for_response(resp)
+            data = await self._response_data(resp)
+            return extract_job_id(data)
+
+    async def get_job_status(self, job_id: str) -> Dict[str, Any]:
+        await self._ensure_session()
+        async with self._session.get(f"{self._jobs_url}/{job_id}") as resp:
+            await self._raise_for_response(resp)
+            return await self._response_data(resp)
+
+    async def get_batch_status(self, batch_id: str) -> Dict[str, Any]:
+        await self._ensure_session()
+        async with self._session.get(f"{self._jobs_url}/batch/{batch_id}") as resp:
+            await self._raise_for_response(resp)
+            return await self._response_data(resp)
+
+    async def fetch_jsonl(self, url: str) -> list:
+        timeout = aiohttp.ClientTimeout(total=self._timeout)
+        async with aiohttp.ClientSession(timeout=timeout) as bare_session:
+            async with bare_session.get(url) as resp:
+                await self._raise_for_response(resp)
+                text = await resp.text()
+                try:
+                    lines = text.strip().split("\n")
+                    return [json.loads(line) for line in lines if line.strip()]
+                except json.JSONDecodeError as e:
+                    raise ResultParseError(
+                        f"Malformed JSONL result payload: {e}"
+                    ) from e
+
+    async def _raise_for_response(self, resp) -> None:
+        if 200 <= resp.status < 300:
+            return
+        try:
+            body = await resp.json()
+            msg = (
+                extract_api_message_from_payload(body)
+                if isinstance(body, dict)
+                else None
+            )
+            if not msg:
+                msg = await resp.text()
+        except Exception:
+            msg = await resp.text()
+        raise_for_status(resp.status, msg)
+
+    async def _response_data(self, resp) -> Dict[str, Any]:
+        try:
+            payload = await resp.json()
+        except Exception as e:
+            raise ResponseFormatError(f"Response body is not valid JSON: {e}") from e
+        return unwrap_api_response(payload, resp.status)
diff --git a/paddleocr/_api_client/_async_poller.py b/paddleocr/_api_client/_async_poller.py
new file mode 100644
index 0000000..9195d04
--- /dev/null
+++ b/paddleocr/_api_client/_async_poller.py
@@ -0,0 +1,86 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import asyncio
+from typing import Any
+
+from ._core import (
+    job_status_from_data,
+    parse_batch_status,
+    validate_result_json_url,
+    validate_state,
+)
+from .errors import (
+    JobFailedError,
+    PollTimeoutError,
+)
+from .results import BatchStatus, JobStatus
+
+DEFAULT_INITIAL_INTERVAL = 3.0
+DEFAULT_MULTIPLIER = 1.5
+DEFAULT_MAX_INTERVAL = 15.0
+DEFAULT_MAX_WAIT_TIME = 600.0
+
+
+class AsyncPoller:
+    def __init__(
+        self,
+        http_client,
+        initial_interval: float = DEFAULT_INITIAL_INTERVAL,
+        multiplier: float = DEFAULT_MULTIPLIER,
+        max_interval: float = DEFAULT_MAX_INTERVAL,
+        max_wait_time: float = DEFAULT_MAX_WAIT_TIME,
+    ):
+        self._http = http_client
+        self._initial_interval = initial_interval
+        self._multiplier = multiplier
+        self._max_interval = max_interval
+        self._max_wait_time = max_wait_time
+
+    async def poll_until_done(self, job_id: str) -> Any:
+        interval = self._initial_interval
+        loop = asyncio.get_running_loop()
+        start = loop.time()
+        deadline = start + self._max_wait_time
+
+        while True:
+            now = loop.time()
+            if now >= deadline:
+                raise PollTimeoutError(job_id, now - start)
+
+            data = await self._http.get_job_status(job_id)
+            state = validate_state(data)
+
+            if state == "done":
+                json_url = validate_result_json_url(data)
+                jsonl_data = await self._http.fetch_jsonl(json_url)
+                return jsonl_data, data
+
+            if state == "failed":
+                error_msg = data.get("errorMsg", "Unknown error")
+                raise JobFailedError(job_id, error_msg)
+
+            remaining = deadline - loop.time()
+            if remaining <= 0:
+                raise PollTimeoutError(job_id, loop.time() - start)
+            await asyncio.sleep(min(interval, remaining))
+            interval = min(interval * self._multiplier, self._max_interval)
+
+    async def get_status(self, job_id: str) -> JobStatus:
+        data = await self._http.get_job_status(job_id)
+        return job_status_from_data(job_id, data)
+
+    async def get_batch_status(self, batch_id: str) -> BatchStatus:
+        data = await self._http.get_batch_status(batch_id)
+        return parse_batch_status(batch_id, data)
diff --git a/paddleocr/_api_client/_core.py b/paddleocr/_api_client/_core.py
new file mode 100644
index 0000000..77f0308
--- /dev/null
+++ b/paddleocr/_api_client/_core.py
@@ -0,0 +1,211 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Optional, Union
+
+from .errors import (
+    APIError,
+    AuthError,
+    InvalidRequestError,
+    RateLimitError,
+    ResponseFormatError,
+    ServiceUnavailableError,
+)
+from .models import (
+    DocParsingOptions,
+    Model,
+    OCROptions,
+    PaddleOCRVLOptions,
+    PPStructureV3Options,
+    is_document_parsing_model,
+    is_ocr_model,
+    is_vl_model,
+)
+from .results import BatchStatus, Job, JobStatus, Progress
+
+
+def validate_input_source(file_url: Optional[str], file_path: Optional[str]) -> None:
+    if not file_url and not file_path:
+        raise InvalidRequestError("Either file_url or file_path is required.")
+    if file_url and file_path:
+        raise InvalidRequestError("file_url and file_path are mutually exclusive.")
+
+
+def default_payload(model: Model) -> dict:
+    if is_ocr_model(model):
+        return OCROptions().to_payload()
+    return resolve_document_options(model, None).to_payload()
+
+
+def resolve_ocr_model(model: Union[Model, str]) -> Model:
+    resolved = resolve_model(model)
+    if not is_ocr_model(resolved):
+        raise InvalidRequestError(f"Unsupported OCR model: {model}")
+    return resolved
+
+
+def resolve_document_model(model: Union[Model, str]) -> Model:
+    resolved = resolve_model(model)
+    if not is_document_parsing_model(resolved):
+        raise InvalidRequestError(f"Unsupported document parsing model: {model}")
+    return resolved
+
+
+def resolve_model(model: Union[Model, str]) -> Model:
+    if isinstance(model, Model):
+        return model
+    try:
+        return Model(model)
+    except ValueError as e:
+        raise InvalidRequestError(f"Unsupported model: {model}") from e
+
+
+def resolve_document_options(
+    model: Model, options: Optional[DocParsingOptions]
+) -> DocParsingOptions:
+    if options is not None:
+        if model == Model.PP_STRUCTURE_V3 and not isinstance(
+            options, PPStructureV3Options
+        ):
+            raise InvalidRequestError("PP-StructureV3 requires PPStructureV3Options.")
+        if is_vl_model(model) and not isinstance(options, PaddleOCRVLOptions):
+            raise InvalidRequestError("PaddleOCR-VL models require PaddleOCRVLOptions.")
+        return options
+    if model == Model.PP_STRUCTURE_V3:
+        return PPStructureV3Options()
+    return PaddleOCRVLOptions()
+
+
+def job_id_for_task(job: Union[Job, str], task: str) -> str:
+    if isinstance(job, str):
+        return job
+    if job.task != task:
+        raise InvalidRequestError(
+            f"Job task mismatch: expected {task}, got {job.task}."
+        )
+    if task == "ocr" and not is_ocr_model(job.model):
+        raise InvalidRequestError(f"Job model is not an OCR model: {job.model}.")
+    if task == "document_parsing" and not is_document_parsing_model(job.model):
+        raise InvalidRequestError(
+            f"Job model is not a document parsing model: {job.model}."
+        )
+    return job.job_id
+
+
+def extract_api_message_from_payload(payload: dict) -> Optional[str]:
+    for key in ("msg", "errorMsg", "message"):
+        value = payload.get(key)
+        if value:
+            return str(value)
+    data = payload.get("data")
+    if isinstance(data, dict):
+        value = data.get("errorMsg")
+        if value:
+            return str(value)
+    return None
+
+
+def validate_state(data: dict) -> str:
+    state = data.get("state")
+    if state not in {"pending", "running", "done", "failed"}:
+        raise ResponseFormatError(f"Unknown or missing job state: {state}")
+    return state
+
+
+def job_status_from_data(job_id: str, data: dict) -> JobStatus:
+    state = validate_state(data)
+    progress = None
+    ep = data.get("extractProgress")
+    if ep:
+        if not isinstance(ep, dict):
+            raise ResponseFormatError("'extractProgress' must be an object.")
+        progress = Progress(
+            total_pages=ep.get("totalPages", 0),
+            extracted_pages=ep.get("extractedPages", 0),
+            start_time=ep.get("startTime"),
+            end_time=ep.get("endTime"),
+        )
+    return JobStatus(
+        job_id=job_id,
+        state=state,
+        progress=progress,
+        result=data.get("resultUrl"),
+        error_msg=data.get("errorMsg"),
+    )
+
+
+def raise_for_status(status_code: int, msg: str) -> None:
+    if 200 <= status_code < 300:
+        return
+    if status_code in (401, 403):
+        raise AuthError(f"Authentication failed: {msg}")
+    if status_code == 400:
+        raise InvalidRequestError(f"Bad request: {msg}")
+    if status_code == 429:
+        raise RateLimitError(f"Rate limit exceeded: {msg}")
+    if status_code in (503, 504):
+        raise ServiceUnavailableError(status_code, f"Service unavailable: {msg}")
+    raise APIError(status_code, msg)
+
+
+def unwrap_api_response(payload: dict, status_code: int) -> dict:
+    if not isinstance(payload, dict):
+        raise ResponseFormatError("Response body must be a JSON object.")
+    code = payload.get("code", 0)
+    if code not in (0, None):
+        raise APIError(status_code, extract_api_message_from_payload(payload) or "")
+    data = payload.get("data")
+    if not isinstance(data, dict):
+        raise ResponseFormatError("Response JSON must contain object field 'data'.")
+    return data
+
+
+def extract_job_id(data: dict) -> str:
+    job_id = data.get("jobId")
+    if not isinstance(job_id, str) or not job_id:
+        raise ResponseFormatError(
+            "Response data must contain non-empty string 'jobId'."
+        )
+    return job_id
+
+
+def validate_result_json_url(data: dict) -> str:
+    result_url = data.get("resultUrl")
+    if not isinstance(result_url, dict):
+        raise ResponseFormatError("Done job response must contain object 'resultUrl'.")
+    json_url = result_url.get("jsonUrl")
+    if not isinstance(json_url, str) or not json_url:
+        raise ResponseFormatError(
+            "Done job response resultUrl must contain non-empty string 'jsonUrl'."
+        )
+    return json_url
+
+
+def parse_batch_status(batch_id: str, data: dict) -> BatchStatus:
+    result = data.get("extractResult")
+    if not isinstance(result, list):
+        raise ResponseFormatError(
+            "Batch response data must contain list 'extractResult'."
+        )
+    jobs = []
+    for item in result:
+        if not isinstance(item, dict):
+            raise ResponseFormatError("Batch extractResult items must be objects.")
+        job_id = item.get("jobId")
+        if not isinstance(job_id, str) or not job_id:
+            raise ResponseFormatError(
+                "Batch extractResult items must contain non-empty string 'jobId'."
+            )
+        jobs.append(job_status_from_data(job_id, item))
+    return BatchStatus(batch_id=batch_id, jobs=jobs)
diff --git a/paddleocr/_api_client/_http.py b/paddleocr/_api_client/_http.py
new file mode 100644
index 0000000..26a1e32
--- /dev/null
+++ b/paddleocr/_api_client/_http.py
@@ -0,0 +1,206 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import json
+import os
+from typing import Any, Dict, Optional
+
+import requests
+
+from ._core import (
+    extract_api_message_from_payload,
+    extract_job_id,
+    raise_for_status,
+    unwrap_api_response,
+)
+from .errors import (
+    NetworkError,
+    RequestTimeoutError,
+    ResponseFormatError,
+    ResultParseError,
+)
+
+DEFAULT_BASE_URL = "https://paddleocr.aistudio-app.com"
+API_PATH = "/api/v2/ocr/jobs"
+
+
+def _raise_for_response(response: requests.Response) -> None:
+    if 200 <= response.status_code < 300:
+        return
+    raise_for_status(response.status_code, _extract_api_message(response))
+
+
+def _extract_api_message(response: requests.Response) -> str:
+    try:
+        payload = response.json()
+    except ValueError:
+        return response.text
+    if isinstance(payload, dict):
+        msg = extract_api_message_from_payload(payload)
+        if msg:
+            return msg
+    return response.text
+
+
+def _response_json(response: requests.Response) -> Dict[str, Any]:
+    try:
+        payload = response.json()
+    except ValueError as e:
+        raise ResponseFormatError(f"Response body is not valid JSON: {e}") from e
+    if not isinstance(payload, dict):
+        raise ResponseFormatError("Response body must be a JSON object.")
+    return payload
+
+
+def _response_data(response: requests.Response) -> Dict[str, Any]:
+    payload = _response_json(response)
+    return unwrap_api_response(payload, response.status_code)
+
+
+def _job_id_from_response(response: requests.Response) -> str:
+    return extract_job_id(_response_data(response))
+
+
+class HTTPClient:
+    def __init__(
+        self,
+        token: str,
+        base_url: str,
+        timeout: float,
+        client_platform: Optional[str] = None,
+    ):
+        self._token = token
+        self._base_url = base_url.rstrip("/")
+        self._jobs_url = f"{self._base_url}{API_PATH}"
+        self._timeout = timeout
+        self._session = requests.Session()
+        self._session.headers["Authorization"] = f"Bearer {token}"
+        if client_platform:
+            self._session.headers["Client-Platform"] = client_platform
+
+    @property
+    def timeout(self) -> float:
+        return self._timeout
+
+    def submit_url(
+        self,
+        model: str,
+        file_url: str,
+        optional_payload: dict,
+        page_ranges: Optional[str] = None,
+        batch_id: Optional[str] = None,
+    ) -> str:
+        body = {
+            "fileUrl": file_url,
+            "model": model,
+            "optionalPayload": optional_payload,
+        }
+        if page_ranges is not None:
+            body["pageRanges"] = page_ranges
+        if batch_id is not None:
+            body["batchId"] = batch_id
+        try:
+            resp = self._session.post(
+                self._jobs_url,
+                json=body,
+                timeout=self._timeout,
+            )
+        except requests.Timeout as e:
+            raise RequestTimeoutError(f"Request timed out: {e}") from e
+        except requests.ConnectionError as e:
+            raise NetworkError(f"Connection failed: {e}") from e
+        _raise_for_response(resp)
+        return _job_id_from_response(resp)
+
+    def submit_file(
+        self,
+        model: str,
+        file_path: str,
+        optional_payload: dict,
+        page_ranges: Optional[str] = None,
+        batch_id: Optional[str] = None,
+    ) -> str:
+        if not os.path.exists(file_path):
+            raise FileNotFoundError(file_path)
+        data = {
+            "model": model,
+            "optionalPayload": json.dumps(optional_payload),
+        }
+        if page_ranges is not None:
+            data["pageRanges"] = page_ranges
+        if batch_id is not None:
+            data["batchId"] = batch_id
+        try:
+            with open(file_path, "rb") as f:
+                resp = self._session.post(
+                    self._jobs_url,
+                    data=data,
+                    files={"file": f},
+                    timeout=self._timeout,
+                )
+        except requests.Timeout as e:
+            raise RequestTimeoutError(f"Request timed out: {e}") from e
+        except requests.ConnectionError as e:
+            raise NetworkError(f"Connection failed: {e}") from e
+        _raise_for_response(resp)
+        return _job_id_from_response(resp)
+
+    def get_job_status(self, job_id: str) -> Dict[str, Any]:
+        try:
+            resp = self._session.get(
+                f"{self._jobs_url}/{job_id}",
+                timeout=self._timeout,
+            )
+        except requests.Timeout as e:
+            raise RequestTimeoutError(f"Request timed out: {e}") from e
+        except requests.ConnectionError as e:
+            raise NetworkError(f"Connection failed: {e}") from e
+        _raise_for_response(resp)
+        return _response_data(resp)
+
+    def get_batch_status(self, batch_id: str) -> Dict[str, Any]:
+        try:
+            resp = self._session.get(
+                f"{self._jobs_url}/batch/{batch_id}",
+                timeout=self._timeout,
+            )
+        except requests.Timeout as e:
+            raise RequestTimeoutError(f"Request timed out: {e}") from e
+        except requests.ConnectionError as e:
+            raise NetworkError(f"Connection failed: {e}") from e
+        _raise_for_response(resp)
+        return _response_data(resp)
+
+    def fetch_jsonl(self, url: str) -> list:
+        # Result URLs are often pre-signed object storage links.
+        try:
+            resp = requests.get(url, timeout=self._timeout)
+        except requests.Timeout as e:
+            raise RequestTimeoutError(f"Request timed out: {e}") from e
+        except requests.ConnectionError as e:
+            raise NetworkError(f"Connection failed: {e}") from e
+        try:
+            resp.raise_for_status()
+            lines = resp.text.strip().split("\n")
+            results = []
+            for line in lines:
+                line = line.strip()
+                if line:
+                    results.append(json.loads(line))
+            return results
+        except json.JSONDecodeError as e:
+            raise ResultParseError(f"Malformed JSONL result payload: {e}") from e
+
+    def close(self):
+        self._session.close()
diff --git a/paddleocr/_api_client/_naming.py b/paddleocr/_api_client/_naming.py
new file mode 100644
index 0000000..63cc03b
--- /dev/null
+++ b/paddleocr/_api_client/_naming.py
@@ -0,0 +1,17 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddleocr._utils.naming import snake_keys_to_camel, snake_to_camel
+
+__all__ = ["snake_keys_to_camel", "snake_to_camel"]
diff --git a/paddleocr/_api_client/_poller.py b/paddleocr/_api_client/_poller.py
new file mode 100644
index 0000000..b85b5ac
--- /dev/null
+++ b/paddleocr/_api_client/_poller.py
@@ -0,0 +1,152 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import time
+from typing import Any
+
+from ._core import (
+    job_status_from_data,
+    parse_batch_status,
+    validate_result_json_url,
+    validate_state,
+)
+from .errors import (
+    JobFailedError,
+    PollTimeoutError,
+    ResultParseError,
+)
+from .results import (
+    BatchStatus,
+    DocParsingPage,
+    DocParsingResult,
+    JobStatus,
+    OCRPage,
+    OCRResult,
+)
+
+DEFAULT_INITIAL_INTERVAL = 3.0
+DEFAULT_MULTIPLIER = 1.5
+DEFAULT_MAX_INTERVAL = 15.0
+DEFAULT_MAX_WAIT_TIME = 600.0
+
+
+class Poller:
+    def __init__(
+        self,
+        http_client,
+        initial_interval: float = DEFAULT_INITIAL_INTERVAL,
+        multiplier: float = DEFAULT_MULTIPLIER,
+        max_interval: float = DEFAULT_MAX_INTERVAL,
+        max_wait_time: float = DEFAULT_MAX_WAIT_TIME,
+    ):
+        self._http = http_client
+        self._initial_interval = initial_interval
+        self._multiplier = multiplier
+        self._max_interval = max_interval
+        self._max_wait_time = max_wait_time
+
+    def poll_until_done(self, job_id: str) -> Any:
+        interval = self._initial_interval
+        start = time.monotonic()
+        deadline = start + self._max_wait_time
+
+        while True:
+            now = time.monotonic()
+            if now >= deadline:
+                raise PollTimeoutError(job_id, now - start)
+
+            data = self._http.get_job_status(job_id)
+            state = validate_state(data)
+
+            if state == "done":
+                json_url = validate_result_json_url(data)
+                jsonl_data = self._http.fetch_jsonl(json_url)
+                return jsonl_data, data
+
+            if state == "failed":
+                error_msg = data.get("errorMsg", "Unknown error")
+                raise JobFailedError(job_id, error_msg)
+
+            remaining = deadline - time.monotonic()
+            if remaining <= 0:
+                raise PollTimeoutError(job_id, time.monotonic() - start)
+
+            time.sleep(min(interval, remaining))
+            interval = min(interval * self._multiplier, self._max_interval)
+
+    def get_status(self, job_id: str) -> JobStatus:
+        data = self._http.get_job_status(job_id)
+        return job_status_from_data(job_id, data)
+
+    def get_batch_status(self, batch_id: str) -> BatchStatus:
+        data = self._http.get_batch_status(batch_id)
+        return parse_batch_status(batch_id, data)
+
+
+def parse_ocr_result(job_id: str, jsonl_data: list) -> OCRResult:
+    try:
+        pages = []
+        data_info = {}
+        for line_obj in jsonl_data:
+            result = line_obj["result"]
+            if isinstance(result.get("dataInfo"), dict):
+                data_info.update(result["dataInfo"])
+            for item in result["ocrResults"]:
+                pages.append(
+                    OCRPage(
+                        pruned_result=item["prunedResult"],
+                        ocr_image_url=item.get("ocrImage"),
+                        doc_preprocessing_image_url=item.get("docPreprocessingImage"),
+                        input_image_url=item.get("inputImage"),
+                        raw=item,
+                    )
+                )
+        return OCRResult(
+            job_id=job_id,
+            pages=pages,
+            data_info=data_info,
+        )
+    except (KeyError, TypeError) as e:
+        raise ResultParseError(f"Malformed OCR result payload: {e}") from e
+
+
+def parse_doc_parsing_result(job_id: str, jsonl_data: list) -> DocParsingResult:
+    try:
+        pages = []
+        data_info = {}
+        for line_obj in jsonl_data:
+            result = line_obj["result"]
+            if isinstance(result.get("dataInfo"), dict):
+                data_info.update(result["dataInfo"])
+            for item in result["layoutParsingResults"]:
+                markdown = item["markdown"]
+                pages.append(
+                    DocParsingPage(
+                        markdown_text=markdown["text"],
+                        markdown_images=markdown.get("images", {}),
+                        output_images=item.get("outputImages", {}),
+                        pruned_result=item.get("prunedResult"),
+                        input_image_url=item.get("inputImage"),
+                        exports=item.get("exports", {}),
+                        markdown=markdown,
+                        raw=item,
+                    )
+                )
+        return DocParsingResult(
+            job_id=job_id,
+            pages=pages,
+            data_info=data_info,
+        )
+    except (KeyError, TypeError) as e:
+        raise ResultParseError(f"Malformed document parsing result payload: {e}") from e
diff --git a/paddleocr/_api_client/_resources.py b/paddleocr/_api_client/_resources.py
new file mode 100644
index 0000000..0f4caf0
--- /dev/null
+++ b/paddleocr/_api_client/_resources.py
@@ -0,0 +1,212 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import tempfile
+from pathlib import Path
+from typing import Dict, Iterable, List, Optional, Tuple
+from urllib.parse import unquote, urlparse
+
+import requests
+
+from .errors import InvalidRequestError, NetworkError, RequestTimeoutError
+from .results import DocParsingResult, OCRResult
+
+
+def save_resource(
+    resource_url: str,
+    destination: str,
+    *,
+    overwrite: bool = False,
+    filename: Optional[str] = None,
+    timeout: float = 300.0,
+) -> str:
+    if not resource_url:
+        raise InvalidRequestError("resource_url is required.")
+    if not destination:
+        raise InvalidRequestError("destination is required.")
+
+    parsed_url = urlparse(resource_url)
+    if parsed_url.scheme not in ("http", "https") or not parsed_url.netloc:
+        raise InvalidRequestError(f"Invalid resource URL: {resource_url}")
+
+    target = _resolve_destination(parsed_url.path, destination, filename)
+    _require_writable_target(target, overwrite)
+
+    try:
+        response = requests.get(resource_url, timeout=timeout)
+    except requests.Timeout as e:
+        raise RequestTimeoutError(f"Request timed out: {e}") from e
+    except requests.ConnectionError as e:
+        raise NetworkError(f"Connection failed: {e}") from e
+
+    try:
+        response.raise_for_status()
+    except requests.RequestException as e:
+        raise NetworkError(f"Failed to download resource: {e}") from e
+
+    _atomic_write(target, response.content, overwrite)
+    return str(target)
+
+
+def save_ocr_result_resources(
+    result: OCRResult,
+    destination: str,
+    *,
+    overwrite: bool = False,
+    timeout: float = 300.0,
+) -> List[str]:
+    if result is None:
+        raise InvalidRequestError("OCR result is required.")
+    dest_dir = _require_existing_directory(destination)
+    saved_paths = []
+    for index, page in enumerate(result.pages):
+        if not page.ocr_image_url:
+            continue
+        filename = f"ocr-page-{index + 1}{_safe_resource_extension(page.ocr_image_url)}"
+        saved_paths.append(
+            save_resource(
+                page.ocr_image_url,
+                str(dest_dir / filename),
+                overwrite=overwrite,
+                timeout=timeout,
+            )
+        )
+    return saved_paths
+
+
+def save_document_parsing_result_resources(
+    result: DocParsingResult,
+    destination: str,
+    *,
+    overwrite: bool = False,
+    timeout: float = 300.0,
+) -> List[str]:
+    if result is None:
+        raise InvalidRequestError("document parsing result is required.")
+    dest_dir = _require_existing_directory(destination)
+    saved_paths = []
+    for page in result.pages:
+        for filename, resource_url in _iter_named_resources(page.markdown_images):
+            saved_paths.append(
+                save_resource(
+                    resource_url,
+                    str(dest_dir / filename),
+                    overwrite=overwrite,
+                    timeout=timeout,
+                )
+            )
+        for filename, resource_url in _iter_named_resources(page.output_images):
+            saved_paths.append(
+                save_resource(
+                    resource_url,
+                    str(dest_dir / filename),
+                    overwrite=overwrite,
+                    timeout=timeout,
+                )
+            )
+    return saved_paths
+
+
+def _resolve_destination(
+    url_path: str, destination: str, filename: Optional[str]
+) -> Path:
+    destination_path = Path(destination)
+    if filename is not None:
+        _validate_result_resource_filename(filename)
+        target = destination_path / filename
+    elif destination_path.exists() and destination_path.is_dir():
+        target = destination_path / _safe_url_basename(url_path)
+    else:
+        target = destination_path
+
+    parent = target.parent
+    if not parent.exists():
+        raise FileNotFoundError(str(parent))
+    if not parent.is_dir():
+        raise InvalidRequestError(f"Destination parent must be a directory: {parent}")
+    return target
+
+
+def _require_existing_directory(destination: str) -> Path:
+    dest_dir = Path(destination)
+    if not dest_dir.exists():
+        raise FileNotFoundError(destination)
+    if not dest_dir.is_dir():
+        raise InvalidRequestError(
+            f"Destination must be an existing directory: {destination}"
+        )
+    return dest_dir
+
+
+def _require_writable_target(target: Path, overwrite: bool) -> None:
+    if target.exists() and not overwrite:
+        raise InvalidRequestError(f"Destination already exists: {target}")
+
+
+def _atomic_write(target: Path, content: bytes, overwrite: bool) -> None:
+    fd, temp_path = tempfile.mkstemp(
+        prefix=f".{target.name}.tmp-",
+        dir=str(target.parent),
+    )
+    try:
+        with os.fdopen(fd, "wb") as temp_file:
+            temp_file.write(content)
+        if overwrite:
+            os.replace(temp_path, target)
+        else:
+            os.link(temp_path, target)
+            os.remove(temp_path)
+    except FileExistsError as e:
+        raise InvalidRequestError(f"Destination already exists: {target}") from e
+    finally:
+        if os.path.exists(temp_path):
+            os.remove(temp_path)
+
+
+def _iter_named_resources(resources: Dict[str, str]) -> Iterable[Tuple[str, str]]:
+    for key in sorted(resources):
+        resource_url = resources[key]
+        if not resource_url:
+            continue
+        _validate_result_resource_filename(key)
+        yield key, resource_url
+
+
+def _safe_url_basename(url_path: str) -> str:
+    name = Path(unquote(url_path)).name
+    if not name or name in (".", ".."):
+        return "resource"
+    _validate_result_resource_filename(name)
+    return name
+
+
+def _safe_resource_extension(resource_url: str) -> str:
+    parsed = urlparse(resource_url)
+    suffix = Path(unquote(parsed.path)).suffix
+    if not suffix:
+        return ""
+    try:
+        _validate_result_resource_filename(f"resource{suffix}")
+    except InvalidRequestError:
+        return ""
+    return suffix
+
+
+def _validate_result_resource_filename(name: str) -> None:
+    if not name:
+        raise InvalidRequestError("Resource filename must not be empty.")
+    path = Path(name)
+    if path.name != name or "/" in name or "\\" in name or name in (".", ".."):
+        raise InvalidRequestError(f"Unsafe resource filename: {name}")
diff --git a/paddleocr/_api_client/async_client.py b/paddleocr/_api_client/async_client.py
new file mode 100644
index 0000000..74812c7
--- /dev/null
+++ b/paddleocr/_api_client/async_client.py
@@ -0,0 +1,253 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import asyncio
+import os
+from typing import Optional, Union
+
+from ._core import (
+    default_payload,
+    job_id_for_task,
+    resolve_document_model,
+    resolve_document_options,
+    resolve_ocr_model,
+    validate_input_source,
+)
+from ._async_http import AsyncHTTPClient
+from ._http import DEFAULT_BASE_URL
+from ._async_poller import AsyncPoller
+from ._poller import parse_doc_parsing_result, parse_ocr_result
+from ._resources import (
+    save_document_parsing_result_resources,
+    save_ocr_result_resources,
+    save_resource,
+)
+from .errors import AuthError
+from .models import (
+    DocParsingOptions,
+    Model,
+    OCROptions,
+)
+from .results import BatchStatus, DocParsingResult, Job, JobStatus, OCRResult
+
+
+class AsyncPaddleOCRClient:
+    """Async client for PaddleOCR API.
+
+    Supports asyncio.gather for concurrent job submission and polling.
+    """
+
+    def __init__(
+        self,
+        token: Optional[str] = None,
+        base_url: Optional[str] = None,
+        request_timeout: float = 300.0,
+        poll_timeout: float = 600.0,
+        timeout: Optional[float] = None,
+        client_platform: Optional[str] = None,
+    ):
+        self._token = token or os.environ.get("PADDLEOCR_ACCESS_TOKEN", "")
+        if not self._token:
+            raise AuthError(
+                "Token is required. Set PADDLEOCR_ACCESS_TOKEN or pass token=."
+            )
+        resolved_base_url = (
+            base_url or os.environ.get("PADDLEOCR_BASE_URL") or DEFAULT_BASE_URL
+        )
+        if timeout is not None:
+            request_timeout = timeout
+            poll_timeout = timeout
+        self._http = AsyncHTTPClient(
+            self._token,
+            resolved_base_url,
+            request_timeout,
+            client_platform=client_platform,
+        )
+        self._poller = AsyncPoller(self._http, max_wait_time=poll_timeout)
+
+    async def __aenter__(self):
+        await self._http.__aenter__()
+        return self
+
+    async def __aexit__(self, *args):
+        await self._http.close()
+
+    async def close(self):
+        await self._http.close()
+
+    async def ocr(
+        self,
+        file_url: Optional[str] = None,
+        file_path: Optional[str] = None,
+        options: Optional[OCROptions] = None,
+        page_ranges: Optional[str] = None,
+        batch_id: Optional[str] = None,
+        model: Union[Model, str] = Model.PP_OCRV6,
+    ) -> OCRResult:
+        model = resolve_ocr_model(model)
+        job_id = await self._submit(
+            model,
+            file_url,
+            file_path,
+            options,
+            page_ranges,
+            batch_id,
+        )
+        jsonl_data, _ = await self._poller.poll_until_done(job_id)
+        return parse_ocr_result(job_id, jsonl_data)
+
+    async def parse_document(
+        self,
+        model: Union[Model, str] = Model.PADDLE_OCR_VL_16,
+        file_url: Optional[str] = None,
+        file_path: Optional[str] = None,
+        options: Optional[DocParsingOptions] = None,
+        page_ranges: Optional[str] = None,
+        batch_id: Optional[str] = None,
+    ) -> DocParsingResult:
+        model = resolve_document_model(model)
+        options = resolve_document_options(model, options)
+        job_id = await self._submit(
+            model, file_url, file_path, options, page_ranges, batch_id
+        )
+        jsonl_data, _ = await self._poller.poll_until_done(job_id)
+        return parse_doc_parsing_result(job_id, jsonl_data)
+
+    async def submit_ocr(
+        self,
+        file_url: Optional[str] = None,
+        file_path: Optional[str] = None,
+        options: Optional[OCROptions] = None,
+        page_ranges: Optional[str] = None,
+        batch_id: Optional[str] = None,
+        model: Union[Model, str] = Model.PP_OCRV6,
+    ) -> Job:
+        model = resolve_ocr_model(model)
+        job_id = await self._submit(
+            model,
+            file_url,
+            file_path,
+            options,
+            page_ranges,
+            batch_id,
+        )
+        return Job(job_id=job_id, model=model.value, task="ocr")
+
+    async def submit_document_parsing(
+        self,
+        model: Union[Model, str] = Model.PADDLE_OCR_VL_16,
+        file_url: Optional[str] = None,
+        file_path: Optional[str] = None,
+        options: Optional[DocParsingOptions] = None,
+        page_ranges: Optional[str] = None,
+        batch_id: Optional[str] = None,
+    ) -> Job:
+        model = resolve_document_model(model)
+        options = resolve_document_options(model, options)
+        job_id = await self._submit(
+            model, file_url, file_path, options, page_ranges, batch_id
+        )
+        return Job(job_id=job_id, model=model.value, task="document_parsing")
+
+    async def wait_ocr_result(self, job: Union[Job, str]) -> OCRResult:
+        job_id = job_id_for_task(job, "ocr")
+        jsonl_data, _ = await self._poller.poll_until_done(job_id)
+        return parse_ocr_result(job_id, jsonl_data)
+
+    async def wait_document_parsing_result(
+        self, job: Union[Job, str]
+    ) -> DocParsingResult:
+        job_id = job_id_for_task(job, "document_parsing")
+        jsonl_data, _ = await self._poller.poll_until_done(job_id)
+        return parse_doc_parsing_result(job_id, jsonl_data)
+
+    async def get_status(self, job_id: str) -> JobStatus:
+        return await self._poller.get_status(job_id)
+
+    async def get_batch_status(self, batch_id: str) -> BatchStatus:
+        return await self._poller.get_batch_status(batch_id)
+
+    async def save_resource(
+        self,
+        resource_url: str,
+        destination: str,
+        *,
+        overwrite: bool = False,
+        filename: Optional[str] = None,
+    ) -> str:
+        return await asyncio.to_thread(
+            save_resource,
+            resource_url,
+            destination,
+            overwrite=overwrite,
+            filename=filename,
+            timeout=self._http.timeout,
+        )
+
+    async def save_ocr_result_resources(
+        self,
+        result: OCRResult,
+        destination: str,
+        *,
+        overwrite: bool = False,
+    ) -> list:
+        return await asyncio.to_thread(
+            save_ocr_result_resources,
+            result,
+            destination,
+            overwrite=overwrite,
+            timeout=self._http.timeout,
+        )
+
+    async def save_document_parsing_result_resources(
+        self,
+        result: DocParsingResult,
+        destination: str,
+        *,
+        overwrite: bool = False,
+    ) -> list:
+        return await asyncio.to_thread(
+            save_document_parsing_result_resources,
+            result,
+            destination,
+            overwrite=overwrite,
+            timeout=self._http.timeout,
+        )
+
+    async def _submit(
+        self,
+        model: Model,
+        file_url: Optional[str],
+        file_path: Optional[str],
+        options,
+        page_ranges: Optional[str],
+        batch_id: Optional[str],
+    ) -> str:
+        validate_input_source(file_url, file_path)
+        payload = options.to_payload() if options else default_payload(model)
+        if file_url:
+            return await self._http.submit_url(
+                model.value,
+                file_url,
+                payload,
+                page_ranges=page_ranges,
+                batch_id=batch_id,
+            )
+        return await self._http.submit_file(
+            model.value,
+            file_path,
+            payload,
+            page_ranges=page_ranges,
+            batch_id=batch_id,
+        )
diff --git a/paddleocr/_api_client/cli.py b/paddleocr/_api_client/cli.py
new file mode 100644
index 0000000..5e76a86
--- /dev/null
+++ b/paddleocr/_api_client/cli.py
@@ -0,0 +1,357 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import json
+import sys
+
+from .._utils.cli import str2bool
+from .client import PaddleOCRClient
+from .models import (
+    Model,
+    OCROptions,
+    PaddleOCRVLOptions,
+    PPStructureV3Options,
+    is_ocr_model,
+    is_vl_model,
+)
+
+
+def register_api_command(subparsers):
+    """Register the 'api' subcommand into paddleocr CLI."""
+    subparser = subparsers.add_parser(
+        "api",
+        help="Call PaddleOCR cloud API for OCR or document parsing",
+    )
+    subparser.add_argument(
+        "--model_type",
+        type=str,
+        required=True,
+        choices=["ocr", "doc_parsing"],
+        help="Task type: ocr or doc_parsing",
+    )
+    subparser.add_argument(
+        "--model",
+        type=str,
+        default=None,
+        choices=[m.value for m in Model],
+        help="Model name.",
+    )
+    subparser.add_argument(
+        "--file_url",
+        type=str,
+        default=None,
+        help="URL of the file to process",
+    )
+    subparser.add_argument(
+        "--file_path",
+        type=str,
+        default=None,
+        help="Local file path to process",
+    )
+    subparser.add_argument(
+        "--base_url",
+        type=str,
+        default=None,
+        help="Base URL of the PaddleOCR API service",
+    )
+    subparser.add_argument(
+        "--token",
+        type=str,
+        default=None,
+        help="Access token (or set PADDLEOCR_ACCESS_TOKEN env variable)",
+    )
+    subparser.add_argument(
+        "--client_platform",
+        type=str,
+        default=None,
+        help="Value for the Client-Platform request header",
+    )
+    subparser.add_argument(
+        "--output",
+        type=str,
+        default=None,
+        help="Output JSON file path (prints to stdout if omitted)",
+    )
+    subparser.add_argument(
+        "--request_timeout",
+        type=float,
+        default=300.0,
+        help="Timeout in seconds for one HTTP request",
+    )
+    subparser.add_argument(
+        "--poll_timeout",
+        type=float,
+        default=600.0,
+        help="Total timeout in seconds while waiting for the remote job",
+    )
+    subparser.add_argument(
+        "--save_resources",
+        type=str,
+        default=None,
+        help="Directory for saving resources referenced by the result",
+    )
+    subparser.add_argument(
+        "--overwrite_resources",
+        action="store_true",
+        help="Overwrite existing files when saving resources",
+    )
+    subparser.add_argument(
+        "--page_ranges",
+        type=str,
+        default=None,
+        help='Page ranges to parse, for example "2,4-6"',
+    )
+    subparser.add_argument(
+        "--batch_id",
+        type=str,
+        default=None,
+        help="Optional batch identifier for querying related jobs",
+    )
+    # --- Preprocessing ---
+    subparser.add_argument(
+        "--use_doc_orientation_classify",
+        type=str2bool,
+        default=None,
+        help="Enable document orientation classification (True/False)",
+    )
+    subparser.add_argument(
+        "--use_doc_unwarping",
+        type=str2bool,
+        default=None,
+        help="Enable document unwarping (True/False)",
+    )
+    # --- Text detection ---
+    subparser.add_argument(
+        "--use_textline_orientation",
+        type=str2bool,
+        default=None,
+        help="Enable textline orientation detection for OCR (True/False)",
+    )
+    subparser.add_argument(
+        "--text_det_limit_side_len",
+        type=int,
+        default=None,
+        help="Image side length limit for text detection",
+    )
+    subparser.add_argument(
+        "--text_det_limit_type",
+        type=str,
+        default=None,
+        choices=["min", "max"],
+        help="Side length limit type: min or max",
+    )
+    # --- Text recognition ---
+    subparser.add_argument(
+        "--text_rec_score_thresh",
+        type=float,
+        default=None,
+        help="Score threshold for text recognition results",
+    )
+    # --- Layout and feature toggles (doc_parsing only) ---
+    subparser.add_argument(
+        "--use_layout_detection",
+        type=str2bool,
+        default=None,
+        help="Enable layout detection for document parsing (True/False)",
+    )
+    subparser.add_argument(
+        "--use_seal_recognition",
+        type=str2bool,
+        default=None,
+        help="Enable seal recognition for document parsing (True/False)",
+    )
+    subparser.add_argument(
+        "--use_table_recognition",
+        type=str2bool,
+        default=None,
+        help="Enable table recognition for PP-StructureV3 (True/False)",
+    )
+    subparser.add_argument(
+        "--use_formula_recognition",
+        type=str2bool,
+        default=None,
+        help="Enable formula recognition for PP-StructureV3 (True/False)",
+    )
+    subparser.add_argument(
+        "--use_chart_recognition",
+        type=str2bool,
+        default=None,
+        help="Enable chart recognition for document parsing (True/False)",
+    )
+    # --- Output ---
+    subparser.add_argument(
+        "--visualize",
+        type=str2bool,
+        default=None,
+        help="Enable result visualization images (True/False)",
+    )
+    subparser.add_argument(
+        "--prettify_markdown",
+        type=str2bool,
+        default=None,
+        help="Enable markdown prettification for document parsing (True/False)",
+    )
+    subparser.set_defaults(executor=_execute_api)
+
+
+def _execute_api(args):
+    kwargs = {}
+    if args.token:
+        kwargs["token"] = args.token
+    if args.base_url:
+        kwargs["base_url"] = args.base_url
+    kwargs["request_timeout"] = args.request_timeout
+    kwargs["poll_timeout"] = args.poll_timeout
+    if args.client_platform:
+        kwargs["client_platform"] = args.client_platform
+
+    try:
+        client = PaddleOCRClient(**kwargs)
+    except Exception as e:
+        print(f"Error: {e}", file=sys.stderr)
+        sys.exit(1)
+
+    try:
+        model = _resolve_model(args.model) if args.model else None
+
+        if args.model_type == "ocr":
+            if model is not None and not is_ocr_model(model):
+                print(
+                    f"Error: OCR task does not support {model.value}.",
+                    file=sys.stderr,
+                )
+                sys.exit(2)
+            options = OCROptions(
+                use_doc_orientation_classify=args.use_doc_orientation_classify,
+                use_doc_unwarping=args.use_doc_unwarping,
+                use_textline_orientation=args.use_textline_orientation,
+                text_det_limit_side_len=args.text_det_limit_side_len,
+                text_det_limit_type=args.text_det_limit_type,
+                text_rec_score_thresh=args.text_rec_score_thresh,
+                visualize=args.visualize,
+            )
+            result = client.ocr(
+                file_url=args.file_url,
+                file_path=args.file_path,
+                options=options,
+                page_ranges=args.page_ranges,
+                batch_id=args.batch_id,
+                model=model or Model.PP_OCRV6,
+            )
+            output = _ocr_result_to_dict(result)
+            save_resources = client.save_ocr_result_resources
+        else:
+            if model is None:
+                model = Model.PADDLE_OCR_VL_16
+            if is_vl_model(model):
+                options = PaddleOCRVLOptions(
+                    use_doc_orientation_classify=args.use_doc_orientation_classify,
+                    use_doc_unwarping=args.use_doc_unwarping,
+                    use_chart_recognition=args.use_chart_recognition,
+                    use_seal_recognition=args.use_seal_recognition,
+                    use_layout_detection=args.use_layout_detection,
+                    prettify_markdown=args.prettify_markdown,
+                    visualize=args.visualize,
+                )
+            else:
+                options = PPStructureV3Options(
+                    use_doc_orientation_classify=args.use_doc_orientation_classify,
+                    use_doc_unwarping=args.use_doc_unwarping,
+                    use_textline_orientation=args.use_textline_orientation,
+                    use_chart_recognition=args.use_chart_recognition,
+                    use_seal_recognition=args.use_seal_recognition,
+                    use_table_recognition=args.use_table_recognition,
+                    use_formula_recognition=args.use_formula_recognition,
+                    use_layout_detection=args.use_layout_detection,
+                    text_det_limit_side_len=args.text_det_limit_side_len,
+                    text_det_limit_type=args.text_det_limit_type,
+                    text_rec_score_thresh=args.text_rec_score_thresh,
+                    prettify_markdown=args.prettify_markdown,
+                    visualize=args.visualize,
+                )
+            result = client.parse_document(
+                model=model,
+                file_url=args.file_url,
+                file_path=args.file_path,
+                options=options,
+                page_ranges=args.page_ranges,
+                batch_id=args.batch_id,
+            )
+            output = _doc_parsing_result_to_dict(result)
+            save_resources = client.save_document_parsing_result_resources
+
+        json_str = json.dumps(output, ensure_ascii=False, indent=2)
+
+        if args.save_resources:
+            saved_paths = save_resources(
+                result,
+                args.save_resources,
+                overwrite=args.overwrite_resources,
+            )
+            print(
+                f"Resources saved to: {args.save_resources} ({len(saved_paths)} files)",
+                file=sys.stderr,
+            )
+
+        if args.output:
+            with open(args.output, "w", encoding="utf-8") as f:
+                f.write(json_str)
+            print(f"Result saved to: {args.output}")
+        else:
+            print(json_str)
+
+    except Exception as e:
+        print(f"Error: {e}", file=sys.stderr)
+        sys.exit(1)
+    finally:
+        client.close()
+
+
+def _resolve_model(model_str: str) -> Model:
+    try:
+        return Model(model_str)
+    except ValueError:
+        print(
+            f"Error: Unknown model '{model_str}'. "
+            f"Choose from: {', '.join(m.value for m in Model)}",
+            file=sys.stderr,
+        )
+        sys.exit(1)
+
+
+def _ocr_result_to_dict(result) -> dict:
+    return {
+        "jobId": result.job_id,
+        "pages": [
+            {
+                "prunedResult": page.pruned_result,
+                "ocrImageUrl": page.ocr_image_url,
+            }
+            for page in result.pages
+        ],
+    }
+
+
+def _doc_parsing_result_to_dict(result) -> dict:
+    return {
+        "jobId": result.job_id,
+        "pages": [
+            {
+                "markdownText": page.markdown_text,
+                "markdownImages": page.markdown_images,
+                "outputImages": page.output_images,
+            }
+            for page in result.pages
+        ],
+    }
diff --git a/paddleocr/_api_client/client.py b/paddleocr/_api_client/client.py
new file mode 100644
index 0000000..b3ed57f
--- /dev/null
+++ b/paddleocr/_api_client/client.py
@@ -0,0 +1,240 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+from typing import Optional, Union
+
+from ._core import (
+    default_payload,
+    job_id_for_task,
+    resolve_document_model,
+    resolve_document_options,
+    resolve_ocr_model,
+    validate_input_source,
+)
+from ._http import DEFAULT_BASE_URL, HTTPClient
+from ._poller import Poller, parse_doc_parsing_result, parse_ocr_result
+from ._resources import (
+    save_document_parsing_result_resources,
+    save_ocr_result_resources,
+    save_resource,
+)
+from .errors import AuthError
+from .models import (
+    DocParsingOptions,
+    Model,
+    OCROptions,
+)
+from .results import BatchStatus, DocParsingResult, Job, JobStatus, OCRResult
+
+
+class PaddleOCRClient:
+    """Synchronous blocking client for PaddleOCR official API.
+
+    Wraps the async job API internally: submit → poll → fetch result.
+    """
+
+    def __init__(
+        self,
+        token: Optional[str] = None,
+        base_url: Optional[str] = None,
+        request_timeout: float = 300.0,
+        poll_timeout: float = 600.0,
+        client_platform: Optional[str] = None,
+    ):
+        self._token = token or os.environ.get("PADDLEOCR_ACCESS_TOKEN", "")
+        if not self._token:
+            raise AuthError(
+                "Token is required. Set PADDLEOCR_ACCESS_TOKEN or pass token=."
+            )
+        resolved_base_url = (
+            base_url or os.environ.get("PADDLEOCR_BASE_URL") or DEFAULT_BASE_URL
+        )
+        self._http = HTTPClient(
+            self._token,
+            resolved_base_url,
+            request_timeout,
+            client_platform=client_platform,
+        )
+        self._poller = Poller(self._http, max_wait_time=poll_timeout)
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, *args):
+        self.close()
+
+    def close(self):
+        self._http.close()
+
+    def ocr(
+        self,
+        file_url: Optional[str] = None,
+        file_path: Optional[str] = None,
+        options: Optional[OCROptions] = None,
+        page_ranges: Optional[str] = None,
+        batch_id: Optional[str] = None,
+        model: Union[Model, str] = Model.PP_OCRV6,
+    ) -> OCRResult:
+        model = resolve_ocr_model(model)
+        job_id = self._submit(
+            model,
+            file_url,
+            file_path,
+            options,
+            page_ranges,
+            batch_id,
+        )
+        jsonl_data, _ = self._poller.poll_until_done(job_id)
+        return parse_ocr_result(job_id, jsonl_data)
+
+    def parse_document(
+        self,
+        model: Union[Model, str] = Model.PADDLE_OCR_VL_16,
+        file_url: Optional[str] = None,
+        file_path: Optional[str] = None,
+        options: Optional[DocParsingOptions] = None,
+        page_ranges: Optional[str] = None,
+        batch_id: Optional[str] = None,
+    ) -> DocParsingResult:
+        model = resolve_document_model(model)
+        options = resolve_document_options(model, options)
+        job_id = self._submit(
+            model, file_url, file_path, options, page_ranges, batch_id
+        )
+        jsonl_data, _ = self._poller.poll_until_done(job_id)
+        return parse_doc_parsing_result(job_id, jsonl_data)
+
+    def submit_ocr(
+        self,
+        file_url: Optional[str] = None,
+        file_path: Optional[str] = None,
+        options: Optional[OCROptions] = None,
+        page_ranges: Optional[str] = None,
+        batch_id: Optional[str] = None,
+        model: Union[Model, str] = Model.PP_OCRV6,
+    ) -> Job:
+        model = resolve_ocr_model(model)
+        job_id = self._submit(
+            model,
+            file_url,
+            file_path,
+            options,
+            page_ranges,
+            batch_id,
+        )
+        return Job(job_id=job_id, model=model.value, task="ocr")
+
+    def submit_document_parsing(
+        self,
+        model: Union[Model, str] = Model.PADDLE_OCR_VL_16,
+        file_url: Optional[str] = None,
+        file_path: Optional[str] = None,
+        options: Optional[DocParsingOptions] = None,
+        page_ranges: Optional[str] = None,
+        batch_id: Optional[str] = None,
+    ) -> Job:
+        model = resolve_document_model(model)
+        options = resolve_document_options(model, options)
+        job_id = self._submit(
+            model, file_url, file_path, options, page_ranges, batch_id
+        )
+        return Job(job_id=job_id, model=model.value, task="document_parsing")
+
+    def wait_ocr_result(self, job: Union[Job, str]) -> OCRResult:
+        job_id = job_id_for_task(job, "ocr")
+        jsonl_data, _ = self._poller.poll_until_done(job_id)
+        return parse_ocr_result(job_id, jsonl_data)
+
+    def wait_document_parsing_result(self, job: Union[Job, str]) -> DocParsingResult:
+        job_id = job_id_for_task(job, "document_parsing")
+        jsonl_data, _ = self._poller.poll_until_done(job_id)
+        return parse_doc_parsing_result(job_id, jsonl_data)
+
+    def get_status(self, job_id: str) -> JobStatus:
+        return self._poller.get_status(job_id)
+
+    def get_batch_status(self, batch_id: str) -> BatchStatus:
+        return self._poller.get_batch_status(batch_id)
+
+    def save_resource(
+        self,
+        resource_url: str,
+        destination: str,
+        *,
+        overwrite: bool = False,
+        filename: Optional[str] = None,
+    ) -> str:
+        return save_resource(
+            resource_url,
+            destination,
+            overwrite=overwrite,
+            filename=filename,
+            timeout=self._http.timeout,
+        )
+
+    def save_ocr_result_resources(
+        self,
+        result: OCRResult,
+        destination: str,
+        *,
+        overwrite: bool = False,
+    ) -> list:
+        return save_ocr_result_resources(
+            result,
+            destination,
+            overwrite=overwrite,
+            timeout=self._http.timeout,
+        )
+
+    def save_document_parsing_result_resources(
+        self,
+        result: DocParsingResult,
+        destination: str,
+        *,
+        overwrite: bool = False,
+    ) -> list:
+        return save_document_parsing_result_resources(
+            result,
+            destination,
+            overwrite=overwrite,
+            timeout=self._http.timeout,
+        )
+
+    def _submit(
+        self,
+        model: Model,
+        file_url: Optional[str],
+        file_path: Optional[str],
+        options,
+        page_ranges: Optional[str],
+        batch_id: Optional[str],
+    ) -> str:
+        validate_input_source(file_url, file_path)
+        payload = options.to_payload() if options else default_payload(model)
+        if file_url:
+            return self._http.submit_url(
+                model.value,
+                file_url,
+                payload,
+                page_ranges=page_ranges,
+                batch_id=batch_id,
+            )
+        return self._http.submit_file(
+            model.value,
+            file_path,
+            payload,
+            page_ranges=page_ranges,
+            batch_id=batch_id,
+        )
diff --git a/paddleocr/_api_client/errors.py b/paddleocr/_api_client/errors.py
new file mode 100644
index 0000000..c679304
--- /dev/null
+++ b/paddleocr/_api_client/errors.py
@@ -0,0 +1,85 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+class PaddleOCRAPIError(Exception):
+    """Base exception for PaddleOCR API SDK."""
+
+    def __init__(self, message: str):
+        self.message = message
+        super().__init__(message)
+
+
+class AuthError(PaddleOCRAPIError):
+    """Token missing, invalid, or expired (HTTP 401/403)."""
+
+
+class InvalidRequestError(PaddleOCRAPIError):
+    """Invalid parameters (HTTP 400)."""
+
+
+class APIError(PaddleOCRAPIError):
+    """Non-2xx response from the API server."""
+
+    def __init__(self, status_code: int, message: str):
+        self.status_code = status_code
+        super().__init__(f"HTTP {status_code}: {message}")
+
+
+class RateLimitError(APIError):
+    """Daily quota exceeded (HTTP 429)."""
+
+    def __init__(self, message: str):
+        super().__init__(429, message)
+
+
+class ServiceUnavailableError(APIError):
+    """Server overloaded or gateway timeout (HTTP 503/504)."""
+
+    def __init__(self, status_code: int, message: str):
+        super().__init__(status_code, message)
+
+
+class JobFailedError(PaddleOCRAPIError):
+    """Job execution failed on the server side."""
+
+    def __init__(self, job_id: str, error_msg: str):
+        self.job_id = job_id
+        self.error_msg = error_msg
+        super().__init__(f"Job {job_id} failed: {error_msg}")
+
+
+class RequestTimeoutError(PaddleOCRAPIError):
+    """A single HTTP request exceeded the configured timeout."""
+
+
+class PollTimeoutError(PaddleOCRAPIError):
+    """Polling timed out waiting for job completion."""
+
+    def __init__(self, job_id: str, elapsed: float):
+        self.job_id = job_id
+        self.elapsed = elapsed
+        super().__init__(f"Timed out after {elapsed:.1f}s waiting for job {job_id}")
+
+
+class ResponseFormatError(PaddleOCRAPIError):
+    """A successful API response did not match the documented schema."""
+
+
+class ResultParseError(PaddleOCRAPIError):
+    """A result JSONL payload could not be parsed as the expected result type."""
+
+
+class NetworkError(PaddleOCRAPIError):
+    """Network connection failure."""
diff --git a/paddleocr/_api_client/models.py b/paddleocr/_api_client/models.py
new file mode 100644
index 0000000..d3a2f3f
--- /dev/null
+++ b/paddleocr/_api_client/models.py
@@ -0,0 +1,202 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from dataclasses import dataclass, fields
+from enum import Enum
+from typing import Optional, Union
+
+from ._naming import snake_to_camel
+from .errors import InvalidRequestError
+
+
+class Model(str, Enum):
+    PP_OCRV5 = "PP-OCRv5"
+    PP_OCRV5_LATIN = "PP-OCRv5-latin"
+    PP_OCRV6 = "PP-OCRv6"
+    PP_STRUCTURE_V3 = "PP-StructureV3"
+    PADDLE_OCR_VL = "PaddleOCR-VL"
+    PADDLE_OCR_VL_15 = "PaddleOCR-VL-1.5"
+    PADDLE_OCR_VL_16 = "PaddleOCR-VL-1.6"
+
+
+_OCR_MODELS = frozenset({Model.PP_OCRV5, Model.PP_OCRV5_LATIN, Model.PP_OCRV6})
+_DOCUMENT_PARSING_MODELS = frozenset(
+    {
+        Model.PP_STRUCTURE_V3,
+        Model.PADDLE_OCR_VL,
+        Model.PADDLE_OCR_VL_15,
+        Model.PADDLE_OCR_VL_16,
+    }
+)
+_VL_MODELS = frozenset(
+    {
+        Model.PADDLE_OCR_VL,
+        Model.PADDLE_OCR_VL_15,
+        Model.PADDLE_OCR_VL_16,
+    }
+)
+
+
+def _coerce_model(model: Union[Model, str]) -> Optional[Model]:
+    if isinstance(model, Model):
+        return model
+    try:
+        return Model(model)
+    except ValueError:
+        return None
+
+
+def is_ocr_model(model: Union[Model, str]) -> bool:
+    return _coerce_model(model) in _OCR_MODELS
+
+
+def is_document_parsing_model(model: Union[Model, str]) -> bool:
+    return _coerce_model(model) in _DOCUMENT_PARSING_MODELS
+
+
+def is_vl_model(model: Union[Model, str]) -> bool:
+    return _coerce_model(model) in _VL_MODELS
+
+
+@dataclass
+class OCROptions:
+    use_doc_orientation_classify: Optional[bool] = None
+    use_doc_unwarping: Optional[bool] = None
+    use_textline_orientation: Optional[bool] = None
+    text_det_limit_side_len: Optional[int] = None
+    text_det_limit_type: Optional[str] = None
+    text_det_thresh: Optional[float] = None
+    text_det_box_thresh: Optional[float] = None
+    text_det_unclip_ratio: Optional[float] = None
+    text_rec_score_thresh: Optional[float] = None
+    visualize: Optional[bool] = None
+    extra_options: Optional[dict] = None
+
+    def to_payload(self) -> dict:
+        return _build_payload(self)
+
+
+@dataclass
+class PPStructureV3Options:
+    use_doc_orientation_classify: Optional[bool] = None
+    use_doc_unwarping: Optional[bool] = None
+    use_textline_orientation: Optional[bool] = None
+    use_seal_recognition: Optional[bool] = None
+    use_table_recognition: Optional[bool] = None
+    use_formula_recognition: Optional[bool] = None
+    use_chart_recognition: Optional[bool] = None
+    use_region_detection: Optional[bool] = None
+    layout_threshold: Optional[Union[float, dict]] = None
+    layout_nms: Optional[bool] = None
+    layout_unclip_ratio: Optional[Union[float, list, dict]] = None
+    layout_merge_bboxes_mode: Optional[Union[str, dict]] = None
+    format_block_content: Optional[bool] = None
+    text_det_limit_side_len: Optional[int] = None
+    text_det_limit_type: Optional[str] = None
+    text_det_thresh: Optional[float] = None
+    text_det_box_thresh: Optional[float] = None
+    text_det_unclip_ratio: Optional[float] = None
+    text_rec_score_thresh: Optional[float] = None
+    use_wired_table_cells_trans_to_html: Optional[bool] = None
+    use_wireless_table_cells_trans_to_html: Optional[bool] = None
+    use_table_orientation_classify: Optional[bool] = None
+    use_ocr_results_with_table_cells: Optional[bool] = None
+    use_e2e_wired_table_rec_model: Optional[bool] = None
+    use_e2e_wireless_table_rec_model: Optional[bool] = None
+    markdown_ignore_labels: Optional[list] = None
+    prettify_markdown: Optional[bool] = None
+    show_formula_number: Optional[bool] = None
+    return_markdown_images: Optional[bool] = None
+    output_formats: Optional[list] = None
+    visualize: Optional[bool] = None
+    extra_options: Optional[dict] = None
+
+    def to_payload(self) -> dict:
+        return _build_payload(self)
+
+
+@dataclass
+class PaddleOCRVLOptions:
+    use_doc_orientation_classify: Optional[bool] = None
+    use_doc_unwarping: Optional[bool] = None
+    use_layout_detection: Optional[bool] = None
+    use_chart_recognition: Optional[bool] = None
+    use_seal_recognition: Optional[bool] = None
+    use_ocr_for_image_block: Optional[bool] = None
+    layout_threshold: Optional[Union[float, dict]] = None
+    layout_nms: Optional[bool] = None
+    layout_unclip_ratio: Optional[Union[float, list, dict]] = None
+    layout_merge_bboxes_mode: Optional[Union[str, dict]] = None
+    layout_shape_mode: Optional[str] = None
+    prompt_label: Optional[str] = None
+    format_block_content: Optional[bool] = None
+    repetition_penalty: Optional[float] = None
+    temperature: Optional[float] = None
+    top_p: Optional[float] = None
+    min_pixels: Optional[int] = None
+    max_pixels: Optional[int] = None
+    max_new_tokens: Optional[int] = None
+    vlm_extra_args: Optional[dict] = None
+    merge_layout_blocks: Optional[bool] = None
+    markdown_ignore_labels: Optional[list] = None
+    prettify_markdown: Optional[bool] = None
+    show_formula_number: Optional[bool] = None
+    restructure_pages: Optional[bool] = None
+    merge_tables: Optional[bool] = None
+    relevel_titles: Optional[bool] = None
+    return_markdown_images: Optional[bool] = None
+    output_formats: Optional[list] = None
+    visualize: Optional[bool] = None
+    extra_options: Optional[dict] = None
+
+    def to_payload(self) -> dict:
+        _validate_vl_options(self)
+        return _build_payload(self)
+
+
+DocParsingOptions = Union[PPStructureV3Options, PaddleOCRVLOptions]
+
+
+def _build_payload(options) -> dict:
+    payload = {}
+    for field in fields(options):
+        value = getattr(options, field.name)
+        if value is None:
+            continue
+        if field.name == "extra_options":
+            payload.update(value)
+        else:
+            payload[snake_to_camel(field.name)] = value
+    return payload
+
+
+def _validate_vl_options(options: PaddleOCRVLOptions) -> None:
+    if options.top_p is not None and not (0 < options.top_p <= 1):
+        raise InvalidRequestError(
+            "top_p must be greater than 0 and less than or equal to 1."
+        )
+    if options.temperature is not None and options.temperature < 0:
+        raise InvalidRequestError("temperature must be greater than or equal to 0.")
+    if options.repetition_penalty is not None and options.repetition_penalty <= 0:
+        raise InvalidRequestError("repetition_penalty must be greater than 0.")
+    if options.min_pixels is not None and options.min_pixels <= 0:
+        raise InvalidRequestError("min_pixels must be greater than 0.")
+    if options.max_pixels is not None and options.max_pixels <= 0:
+        raise InvalidRequestError("max_pixels must be greater than 0.")
+    if (
+        options.min_pixels is not None
+        and options.max_pixels is not None
+        and options.min_pixels > options.max_pixels
+    ):
+        raise InvalidRequestError("min_pixels cannot be greater than max_pixels.")
diff --git a/paddleocr/_api_client/results.py b/paddleocr/_api_client/results.py
new file mode 100644
index 0000000..dc69b23
--- /dev/null
+++ b/paddleocr/_api_client/results.py
@@ -0,0 +1,81 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from dataclasses import dataclass, field
+from typing import Any, Dict, List, Literal, Optional
+
+
+@dataclass
+class OCRPage:
+    pruned_result: Any
+    ocr_image_url: Optional[str] = None
+    doc_preprocessing_image_url: Optional[str] = None
+    input_image_url: Optional[str] = None
+    raw: Dict[str, Any] = field(default_factory=dict)
+
+
+@dataclass
+class DocParsingPage:
+    markdown_text: str
+    markdown_images: Dict[str, str] = field(default_factory=dict)
+    output_images: Dict[str, str] = field(default_factory=dict)
+    pruned_result: Any = None
+    input_image_url: Optional[str] = None
+    exports: Dict[str, Any] = field(default_factory=dict)
+    markdown: Dict[str, Any] = field(default_factory=dict)
+    raw: Dict[str, Any] = field(default_factory=dict)
+
+
+@dataclass
+class OCRResult:
+    job_id: str
+    pages: List[OCRPage] = field(default_factory=list)
+    data_info: Dict[str, Any] = field(default_factory=dict)
+
+
+@dataclass
+class DocParsingResult:
+    job_id: str
+    pages: List[DocParsingPage] = field(default_factory=list)
+    data_info: Dict[str, Any] = field(default_factory=dict)
+
+
+@dataclass
+class Progress:
+    total_pages: int = 0
+    extracted_pages: int = 0
+    start_time: Optional[str] = None
+    end_time: Optional[str] = None
+
+
+@dataclass
+class Job:
+    job_id: str
+    model: str
+    task: Literal["ocr", "document_parsing"]
+
+
+@dataclass
+class JobStatus:
+    job_id: str
+    state: str
+    progress: Optional[Progress] = None
+    result: Any = None
+    error_msg: Optional[str] = None
+
+
+@dataclass
+class BatchStatus:
+    batch_id: str
+    jobs: List[JobStatus] = field(default_factory=list)
diff --git a/paddleocr/_cli.py b/paddleocr/_cli.py
new file mode 100644
index 0000000..157df1b
--- /dev/null
+++ b/paddleocr/_cli.py
@@ -0,0 +1,309 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import argparse
+import logging
+import subprocess
+import sys
+import time
+import warnings
+from threading import Thread
+
+import requests
+
+from ._models import (
+    ChartParsing,
+    DocImgOrientationClassification,
+    DocVLM,
+    FormulaRecognition,
+    LayoutDetection,
+    SealTextDetection,
+    TableCellsDetection,
+    TableClassification,
+    TableStructureRecognition,
+    TextDetection,
+    TextImageUnwarping,
+    TextLineOrientationClassification,
+    TextRecognition,
+)
+from ._pipelines import (
+    DocPreprocessor,
+    DocUnderstanding,
+    FormulaRecognitionPipeline,
+    PaddleOCR,
+    PaddleOCRVL,
+    PPChatOCRv4Doc,
+    PPDocTranslation,
+    PPStructureV3,
+    SealRecognition,
+    TableRecognitionPipelineV2,
+)
+from ._version import version
+from ._utils.deprecation import CLIDeprecationWarning
+from ._utils.logging import logger
+
+
+def _register_pipelines(subparsers):
+    for cls in [
+        DocPreprocessor,
+        DocUnderstanding,
+        FormulaRecognitionPipeline,
+        PaddleOCR,
+        PaddleOCRVL,
+        PPChatOCRv4Doc,
+        PPDocTranslation,
+        PPStructureV3,
+        SealRecognition,
+        TableRecognitionPipelineV2,
+    ]:
+        subcommand_executor = cls.get_cli_subcommand_executor()
+        subparser = subcommand_executor.add_subparser(subparsers)
+        subparser.set_defaults(executor=subcommand_executor.execute_with_args)
+
+
+def _register_models(subparsers):
+    for cls in [
+        ChartParsing,
+        DocImgOrientationClassification,
+        DocVLM,
+        FormulaRecognition,
+        LayoutDetection,
+        SealTextDetection,
+        TableCellsDetection,
+        TableClassification,
+        TableStructureRecognition,
+        TextDetection,
+        TextImageUnwarping,
+        TextLineOrientationClassification,
+        TextRecognition,
+    ]:
+        subcommand_executor = cls.get_cli_subcommand_executor()
+        subparser = subcommand_executor.add_subparser(subparsers)
+        subparser.set_defaults(executor=subcommand_executor.execute_with_args)
+
+
+def _register_install_hpi_deps_command(subparsers):
+    def _install_hpi_deps(args):
+        hpip = f"hpi-{args.variant}"
+        try:
+            subprocess.check_call(["paddlex", "--install", hpip])
+            subprocess.check_call(["paddlex", "--install", "paddle2onnx"])
+        except subprocess.CalledProcessError:
+            sys.exit("Failed to install dependencies")
+
+    subparser = subparsers.add_parser("install_hpi_deps")
+    subparser.add_argument("variant", type=str, choices=["cpu", "gpu", "npu"])
+    subparser.set_defaults(executor=_install_hpi_deps)
+
+
+def _register_install_genai_server_deps_command(subparsers):
+    def _install_genai_server_deps(args):
+        try:
+            subprocess.check_call(
+                ["paddlex", "--install", f"genai-{args.variant}-server"]
+            )
+        except subprocess.CalledProcessError:
+            sys.exit("Failed to install dependencies")
+
+    subparser = subparsers.add_parser("install_genai_server_deps")
+    subparser.add_argument(
+        "variant", type=str, choices=["vllm", "sglang", "fastdeploy"]
+    )
+    subparser.set_defaults(executor=_install_genai_server_deps)
+
+
+def _register_genai_server_command(subparsers):
+    # TODO: Register the subparser whether the plugin is installed or not
+    try:
+        from paddlex.inference.genai.server import get_arg_parser, run_genai_server
+    except RuntimeError:
+        return
+
+    def _show_prompt_when_server_is_running(host, port, backend):
+        if host == "0.0.0.0":
+            host = "localhost"
+        while True:
+            try:
+                resp = requests.get(f"http://{host}:{port}/health", timeout=1)
+                if resp.status_code == 200:
+                    break
+            except (requests.exceptions.ConnectionError, requests.exceptions.Timeout):
+                pass
+            time.sleep(1)
+        prompt = f"""The PaddleOCR GenAI server has been started. You can either:
+    1. Set the server URL in the module or pipeline configuration and call the PaddleOCR CLI or Python API. For example:
+        paddleocr doc_parser --input demo.png --vl_rec_backend {backend}-server --vl_rec_server_url http://{host}:{port}/v1
+    2. Make HTTP requests directly, or using the OpenAI client library."""
+        logger.info(prompt)
+
+    def _run_genai_server(args):
+        Thread(
+            target=_show_prompt_when_server_is_running,
+            args=(args.host, args.port, args.backend),
+            daemon=True,
+        ).start()
+        try:
+            run_genai_server(args)
+        except subprocess.CalledProcessError:
+            sys.exit("Failed to run the server")
+
+    paddlex_parser = get_arg_parser()
+    subparser = subparsers.add_parser(
+        "genai_server", parents=[paddlex_parser], conflict_handler="resolve"
+    )
+    subparser.set_defaults(executor=_run_genai_server)
+
+
+def _register_doc2md_command(subparsers):
+    """Register the doc2md subcommand."""
+
+    def _execute_doc2md(args):
+        if args.formats:
+            from ._doc2md import supported_formats
+
+            fmts = supported_formats()
+            print("Supported formats: " + ", ".join(f".{f}" for f in fmts))
+            return
+
+        if not args.input:
+            logger.error("--input is required when --formats is not set")
+            sys.exit(2)
+
+        from ._doc2md import convert
+        from pathlib import Path
+
+        output = args.output
+        quiet = args.quiet
+
+        # Build converter kwargs from CLI args
+        converter_kwargs = {}
+        if args.no_drawings:
+            converter_kwargs["extract_drawings"] = False
+        if args.no_headers_footers:
+            converter_kwargs["extract_headers_footers"] = False
+        if args.sheet_name is not None:
+            converter_kwargs["sheet_name"] = args.sheet_name
+        if args.max_rows is not None:
+            converter_kwargs["max_rows"] = args.max_rows
+
+        t1 = time.time()
+        try:
+            result = convert(args.input, output=output, **converter_kwargs)
+        except Exception as e:
+            logger.error(f"Conversion failed: {e}")
+            sys.exit(1)
+
+        elapsed = (time.time() - t1) * 1000
+        if not quiet:
+            logger.info(f"Conversion done in {elapsed:.0f} ms")
+
+        if output:
+            if not quiet:
+                logger.info(f"Saved to: {output}")
+                if result.images:
+                    logger.info(f"Images saved to: {Path(output).parent / 'images'}/")
+        else:
+            print(result.markdown)
+
+    subparser = subparsers.add_parser(
+        "doc2md",
+        help="Convert office documents (docx/xlsx/pptx) to Markdown",
+    )
+    subparser.add_argument(
+        "-i",
+        "--input",
+        type=str,
+        default=None,
+        help="Input file path (.docx/.xlsx/.pptx)",
+    )
+    subparser.add_argument(
+        "-o",
+        "--output",
+        type=str,
+        default=None,
+        help="Output Markdown file path (prints to stdout if omitted)",
+    )
+    subparser.add_argument(
+        "-q",
+        "--quiet",
+        action="store_true",
+        help="Suppress informational output",
+    )
+    subparser.add_argument(
+        "--formats",
+        action="store_true",
+        help="List supported formats and exit",
+    )
+    # docx options
+    subparser.add_argument(
+        "--no-drawings",
+        action="store_true",
+        help="[docx/xlsx] Skip text box / drawing layer content extraction",
+    )
+    subparser.add_argument(
+        "--no-headers-footers",
+        action="store_true",
+        help="[docx] Skip header and footer content extraction",
+    )
+    # xlsx options
+    subparser.add_argument(
+        "--sheet-name",
+        type=str,
+        default=None,
+        help="[xlsx] Convert only the specified sheet (by name)",
+    )
+    subparser.add_argument(
+        "--max-rows",
+        type=int,
+        default=None,
+        help="[xlsx] Maximum number of rows to convert per sheet",
+    )
+    subparser.set_defaults(executor=_execute_doc2md)
+
+
+def _register_api_command(subparsers):
+    from ._api_client.cli import register_api_command
+
+    register_api_command(subparsers)
+
+
+def _get_parser():
+    parser = argparse.ArgumentParser(prog="paddleocr")
+    parser.add_argument(
+        "-v", "--version", action="version", version=f"%(prog)s {version}"
+    )
+    subparsers = parser.add_subparsers(dest="subcommand", metavar="COMMAND")
+    _register_pipelines(subparsers)
+    _register_models(subparsers)
+    _register_install_hpi_deps_command(subparsers)
+    _register_install_genai_server_deps_command(subparsers)
+    _register_genai_server_command(subparsers)
+    _register_doc2md_command(subparsers)
+    _register_api_command(subparsers)
+    return parser
+
+
+def _execute(args):
+    args.executor(args)
+
+
+def main():
+    logger.setLevel(logging.INFO)
+    warnings.filterwarnings("default", category=CLIDeprecationWarning)
+    parser = _get_parser()
+    args = parser.parse_args()
+    if args.subcommand is None:
+        parser.print_usage(sys.stderr)
+        sys.exit(2)
+    _execute(args)
diff --git a/paddleocr/_common_args.py b/paddleocr/_common_args.py
new file mode 100644
index 0000000..2a99bb6
--- /dev/null
+++ b/paddleocr/_common_args.py
@@ -0,0 +1,188 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddlex.utils.device import get_default_device, parse_device
+
+from ._constants import (
+    DEFAULT_CPU_THREADS,
+    DEFAULT_DEVICE,
+    DEFAULT_ENABLE_MKLDNN,
+    DEFAULT_MKLDNN_CACHE_CAPACITY,
+    DEFAULT_PRECISION,
+    DEFAULT_USE_TENSORRT,
+    SUPPORTED_PRECISION_LIST,
+    DEFAULT_USE_CINN,
+)
+from ._utils.cli import str2bool
+
+SUPPORTED_INFERENCE_ENGINE_LIST = [
+    "paddle",
+    "paddle_static",
+    "paddle_dynamic",
+    "transformers",
+    "onnxruntime",
+]
+
+
+def parse_common_args(kwargs, *, default_enable_hpi):
+    default_vals = {
+        "device": DEFAULT_DEVICE,
+        "engine": None,
+        "engine_config": None,
+        "enable_hpi": default_enable_hpi,
+        "use_tensorrt": DEFAULT_USE_TENSORRT,
+        "precision": DEFAULT_PRECISION,
+        "enable_mkldnn": DEFAULT_ENABLE_MKLDNN,
+        "mkldnn_cache_capacity": DEFAULT_MKLDNN_CACHE_CAPACITY,
+        "cpu_threads": DEFAULT_CPU_THREADS,
+        "enable_cinn": DEFAULT_USE_CINN,
+    }
+
+    unknown_names = kwargs.keys() - default_vals.keys()
+    for name in unknown_names:
+        raise ValueError(f"Unknown argument: {name}")
+
+    kwargs = {**default_vals, **kwargs}
+
+    if (
+        kwargs["engine"] is not None
+        and kwargs["engine"] not in SUPPORTED_INFERENCE_ENGINE_LIST
+    ):
+        raise ValueError(
+            f"Invalid engine: {kwargs['engine']}. Supported values are: {SUPPORTED_INFERENCE_ENGINE_LIST}."
+        )
+
+    if kwargs["precision"] not in SUPPORTED_PRECISION_LIST:
+        raise ValueError(
+            f"Invalid precision: {kwargs['precision']}. Supported values are: {SUPPORTED_PRECISION_LIST}."
+        )
+
+    kwargs["use_pptrt"] = kwargs.pop("use_tensorrt")
+    kwargs["pptrt_precision"] = kwargs.pop("precision")
+
+    return kwargs
+
+
+def _build_paddle_static_engine_config(common_args, device_type):
+    cfg = {}
+    if device_type == "gpu":
+        if common_args["use_pptrt"]:
+            if common_args["pptrt_precision"] == "fp32":
+                cfg["run_mode"] = "trt_fp32"
+            else:
+                assert common_args["pptrt_precision"] == "fp16", common_args[
+                    "pptrt_precision"
+                ]
+                cfg["run_mode"] = "trt_fp16"
+        else:
+            cfg["run_mode"] = "paddle"
+    elif device_type == "cpu":
+        if common_args["enable_mkldnn"]:
+            cfg["mkldnn_cache_capacity"] = common_args["mkldnn_cache_capacity"]
+        else:
+            cfg["run_mode"] = "paddle"
+        cfg["cpu_threads"] = common_args["cpu_threads"]
+    else:
+        cfg["run_mode"] = "paddle"
+    cfg["enable_cinn"] = common_args["enable_cinn"]
+    return cfg
+
+
+def prepare_common_init_args(model_name, common_args):
+    device = common_args["device"]
+    if device is None:
+        device = get_default_device()
+    device_type, _ = parse_device(device)
+
+    init_kwargs = {}
+    init_kwargs["device"] = device
+    init_kwargs["engine"] = common_args["engine"]
+    init_kwargs["use_hpip"] = common_args["enable_hpi"]
+
+    user_engine_config = common_args["engine_config"]
+    engine = common_args["engine"]
+    built = _build_paddle_static_engine_config(common_args, device_type)
+
+    if user_engine_config is not None:
+        init_kwargs["engine_config"] = user_engine_config
+    elif engine == "paddle_static":
+        init_kwargs["engine_config"] = built
+    elif engine in (None, "paddle"):
+        init_kwargs["engine_config"] = {"paddle_static": built}
+    else:
+        init_kwargs["engine_config"] = None
+
+    return init_kwargs
+
+
+def add_common_cli_opts(parser, *, default_enable_hpi, allow_multiple_devices):
+    if allow_multiple_devices:
+        help_ = "Device(s) to use for inference, e.g., `cpu`, `gpu`, `npu`, `gpu:0`, `gpu:0,1`. If multiple devices are specified, inference will be performed in parallel. Note that parallel inference is not always supported. By default, GPU 0 will be used if available; otherwise, the CPU will be used."
+    else:
+        help_ = "Device to use for inference, e.g., `cpu`, `gpu`, `npu`, `gpu:0`. By default, GPU 0 will be used if available; otherwise, the CPU will be used."
+    parser.add_argument(
+        "--device",
+        type=str,
+        default=DEFAULT_DEVICE,
+        help=help_,
+    )
+    parser.add_argument(
+        "--engine",
+        type=str,
+        choices=SUPPORTED_INFERENCE_ENGINE_LIST,
+        help="Inference engine to use. For CLI, engine-specific configuration should be set in the PaddleX YAML config file.",
+    )
+    parser.add_argument(
+        "--enable_hpi",
+        type=str2bool,
+        default=default_enable_hpi,
+        help="Enable the high performance inference.",
+    )
+    parser.add_argument(
+        "--use_tensorrt",
+        type=str2bool,
+        default=DEFAULT_USE_TENSORRT,
+        help="Whether to use the Paddle Inference TensorRT subgraph engine. If the model does not support TensorRT acceleration, even if this flag is set, acceleration will not be used.",
+    )
+    parser.add_argument(
+        "--precision",
+        type=str,
+        default=DEFAULT_PRECISION,
+        choices=SUPPORTED_PRECISION_LIST,
+        help="Precision for TensorRT when using the Paddle Inference TensorRT subgraph engine.",
+    )
+    parser.add_argument(
+        "--enable_mkldnn",
+        type=str2bool,
+        default=DEFAULT_ENABLE_MKLDNN,
+        help="Enable MKL-DNN acceleration for inference. If MKL-DNN is unavailable or the model does not support it, acceleration will not be used even if this flag is set.",
+    )
+    parser.add_argument(
+        "--mkldnn_cache_capacity",
+        type=int,
+        default=DEFAULT_MKLDNN_CACHE_CAPACITY,
+        help="MKL-DNN cache capacity.",
+    )
+    parser.add_argument(
+        "--cpu_threads",
+        type=int,
+        default=DEFAULT_CPU_THREADS,
+        help="Number of threads to use for inference on CPUs.",
+    )
+    parser.add_argument(
+        "--enable_cinn",
+        type=str2bool,
+        default=DEFAULT_USE_CINN,
+        help="Whether to use the CINN compiler.",
+    )
diff --git a/paddleocr/_constants.py b/paddleocr/_constants.py
new file mode 100644
index 0000000..d96ba78
--- /dev/null
+++ b/paddleocr/_constants.py
@@ -0,0 +1,22 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+DEFAULT_DEVICE = None
+DEFAULT_USE_TENSORRT = False
+DEFAULT_PRECISION = "fp32"
+DEFAULT_ENABLE_MKLDNN = True
+DEFAULT_MKLDNN_CACHE_CAPACITY = 10
+DEFAULT_CPU_THREADS = 10
+SUPPORTED_PRECISION_LIST = ["fp32", "fp16"]
+DEFAULT_USE_CINN = False
diff --git a/paddleocr/_doc2md/__init__.py b/paddleocr/_doc2md/__init__.py
new file mode 100644
index 0000000..d055fad
--- /dev/null
+++ b/paddleocr/_doc2md/__init__.py
@@ -0,0 +1,26 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""paddleocr._doc2md - Convert office documents to Markdown."""
+
+from .core import convert, supported_formats
+from .base import ConvertResult, BaseConverter
+from .registry import default_registry
+
+__all__ = [
+    "convert",
+    "supported_formats",
+    "ConvertResult",
+    "BaseConverter",
+    "default_registry",
+]
diff --git a/paddleocr/_doc2md/base.py b/paddleocr/_doc2md/base.py
new file mode 100644
index 0000000..755a59c
--- /dev/null
+++ b/paddleocr/_doc2md/base.py
@@ -0,0 +1,39 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from abc import ABC, abstractmethod
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Optional
+
+
+@dataclass
+class ConvertResult:
+    """Conversion result."""
+
+    markdown: str
+    title: Optional[str] = None
+    metadata: dict = field(default_factory=dict)
+    images: dict = field(default_factory=dict)  # {relative_path: image_bytes}
+
+
+class BaseConverter(ABC):
+    """Abstract base class for all format converters."""
+
+    supported_extensions: list[str] = []
+    supported_mimetypes: list[str] = []
+
+    @abstractmethod
+    def convert_file(self, file_path: Path, **kwargs) -> ConvertResult:
+        """Convert a file to Markdown."""
+        ...
diff --git a/paddleocr/_doc2md/converters/__init__.py b/paddleocr/_doc2md/converters/__init__.py
new file mode 100644
index 0000000..d48c02e
--- /dev/null
+++ b/paddleocr/_doc2md/converters/__init__.py
@@ -0,0 +1,29 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Built-in converters - importing this package triggers registration of all converters."""
+
+import importlib
+
+_converter_modules = [
+    "docx",
+    "pptx",
+    "xlsx",
+]
+
+for _mod in _converter_modules:
+    try:
+        importlib.import_module(f".{_mod}", package=__name__)
+    except ImportError:
+        # Silently skip missing optional dependencies; users will get a clear error on actual use
+        pass
diff --git a/paddleocr/_doc2md/converters/docx.py b/paddleocr/_doc2md/converters/docx.py
new file mode 100644
index 0000000..0661223
--- /dev/null
+++ b/paddleocr/_doc2md/converters/docx.py
@@ -0,0 +1,1630 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import re
+from collections import Counter
+from pathlib import Path
+
+from ..base import BaseConverter, ConvertResult
+from ..math import convert_omath as _convert_omath
+from ..math import _M
+from ..registry import default_registry
+
+# Regex patterns for Chinese numbered headings
+_RE_H2 = re.compile(r"^[一二三四五六七八九十百千]+[、．.]")
+_RE_H3 = re.compile(r"^（[一二三四五六七八九十百千]+）")
+
+# Regex for field-code hyperlink instruction
+_RE_FIELD_HYPERLINK = re.compile(r'HYPERLINK\s+"([^"]+)"')
+
+# Regex for page-number-only footer/header text (e.g. "第  页", "共 页", "- 3 -", "Page of")
+_RE_PAGE_ONLY = re.compile(r"^[\s第页共of\d\-/|·]*$", re.IGNORECASE)
+
+# Word XML namespace
+_W_NS = "http://schemas.openxmlformats.org/wordprocessingml/2006/main"
+_W = "{" + _W_NS + "}"
+
+# Markup Compatibility namespace (mc:AlternateContent)
+_MC_NS = "http://schemas.openxmlformats.org/markup-compatibility/2006"
+_MC = "{" + _MC_NS + "}"
+
+# WordprocessingShape namespace (wps:txbx)
+_WPS_NS = "http://schemas.microsoft.com/office/word/2010/wordprocessingShape"
+_WPS = "{" + _WPS_NS + "}"
+
+# Chart namespace
+_C_NS = "http://schemas.openxmlformats.org/drawingml/2006/chart"
+_C = "{" + _C_NS + "}"
+
+# WordprocessingDrawing namespace (wp:inline, wp:anchor)
+_WPD = "{http://schemas.openxmlformats.org/drawingml/2006/wordprocessingDrawing}"
+# DrawingML main namespace
+_A = "{http://schemas.openxmlformats.org/drawingml/2006/main}"
+# OPC relationships namespace
+_R = "{http://schemas.openxmlformats.org/officeDocument/2006/relationships}"
+
+
+def _escape_md_url(url: str) -> str:
+    """Escape parentheses in URL for Markdown link syntax."""
+    return url.replace("(", "%28").replace(")", "%29")
+
+
+def _effective_bold(run, para) -> bool:
+    """Resolve effective bold: run-level > character style > paragraph style."""
+    b = run.bold
+    if b is not None:
+        return b
+    try:
+        if run.style and run.style.font.bold is not None:
+            return run.style.font.bold
+    except Exception:
+        pass
+    try:
+        if para.style and para.style.font.bold is not None:
+            return para.style.font.bold
+    except Exception:
+        pass
+    return False
+
+
+def _effective_italic(run, para) -> bool:
+    """Resolve effective italic: run-level > character style > paragraph style."""
+    i = run.italic
+    if i is not None:
+        return i
+    try:
+        if run.style and run.style.font.italic is not None:
+            return run.style.font.italic
+    except Exception:
+        pass
+    try:
+        if para.style and para.style.font.italic is not None:
+            return para.style.font.italic
+    except Exception:
+        pass
+    return False
+
+
+def _effective_underline(run, para) -> bool:
+    """Resolve effective underline: run-level > character style > paragraph style."""
+    u = run.underline
+    if u is not None:
+        return bool(u)
+    try:
+        if run.style and run.style.font.underline is not None:
+            return bool(run.style.font.underline)
+    except Exception:
+        pass
+    try:
+        if para.style and para.style.font.underline is not None:
+            return bool(para.style.font.underline)
+    except Exception:
+        pass
+    return False
+
+
+def _effective_superscript(run) -> bool:
+    """Return True if run has superscript set (w:vertAlign w:val='superscript')."""
+    try:
+        return bool(run.font.superscript)
+    except Exception:
+        return False
+
+
+def _effective_subscript(run) -> bool:
+    """Return True if run has subscript set (w:vertAlign w:val='subscript')."""
+    try:
+        return bool(run.font.subscript)
+    except Exception:
+        return False
+
+
+def _paragraph_has_math(para) -> bool:
+    """Check if paragraph XML contains OMML math elements."""
+    return para._element.find(f".//{_M}oMath") is not None
+
+
+def _iter_math_paragraph_parts(para) -> list:
+    """Parse a paragraph with mixed text/math content into a list of parts.
+
+    Returns a list where each element is either:
+    - ("text", items_list) — a group of text runs to be formatted
+    - ("display_math", latex_str) — a display math block ($$...$$)
+    - ("inline_math", latex_str) — an inline math expression ($...$)
+    """
+    from docx.text.run import Run
+    from docx.text.hyperlink import Hyperlink
+
+    text_items = []
+    parts = []
+
+    def flush_text():
+        if text_items:
+            parts.append(("text", list(text_items)))
+            text_items.clear()
+
+    for child in para._element:
+        tag = child.tag
+        local = tag.split("}")[-1] if "}" in tag else tag
+
+        if tag == f"{_M}oMathPara":
+            flush_text()
+            for omath in child.findall(f"{_M}oMath"):
+                latex = _convert_omath(omath)
+                if latex:
+                    parts.append(("display_math", latex))
+        elif tag == f"{_M}oMath":
+            flush_text()
+            latex = _convert_omath(child)
+            if latex:
+                parts.append(("inline_math", latex))
+        elif local == "r":
+            try:
+                run = Run(child, para)
+                if run.text:
+                    text_items.append(
+                        (
+                            _effective_bold(run, para),
+                            _effective_italic(run, para),
+                            _effective_underline(run, para),
+                            bool(run.font.strike),
+                            _effective_superscript(run),
+                            _effective_subscript(run),
+                            run.text,
+                            "",
+                        )
+                    )
+            except Exception:
+                pass
+        elif local == "hyperlink":
+            try:
+                hl = Hyperlink(child, para)
+                try:
+                    url = hl.url or ""
+                except (KeyError, AttributeError):
+                    url = ""
+                for run in hl.runs:
+                    if run.text:
+                        text_items.append(
+                            (
+                                _effective_bold(run, para),
+                                _effective_italic(run, para),
+                                False,
+                                bool(run.font.strike),
+                                _effective_superscript(run),
+                                _effective_subscript(run),
+                                run.text,
+                                url,
+                            )
+                        )
+            except Exception:
+                pass
+
+    flush_text()
+    return parts
+
+
+def _paragraph_math_to_markdown(para) -> str:
+    """Convert a paragraph containing OMML math to Markdown."""
+    result = []
+    for kind, data in _iter_math_paragraph_parts(para):
+        if kind == "text":
+            md = _runs_to_markdown(data)
+            if md:
+                result.append(md)
+        elif kind == "display_math":
+            result.append(f"$$\n{data}\n$$")
+        elif kind == "inline_math":
+            result.append(f"${data}$")
+    return "".join(result)
+
+
+def _paragraph_math_to_html(para) -> str:
+    """Convert a paragraph containing OMML math to HTML inline text."""
+    result = []
+    for kind, data in _iter_math_paragraph_parts(para):
+        if kind == "text":
+            html = _runs_to_html(data)
+            if html:
+                result.append(html)
+        elif kind == "display_math":
+            result.append(f"$$\n{data}\n$$")
+        elif kind == "inline_math":
+            result.append(f"${data}$")
+    return "".join(result)
+
+
+def _get_body_font_size(doc) -> float:
+    """Return the most common font size in the document (used as body size). Defaults to 16.0."""
+    sizes: Counter = Counter()
+    for p in doc.paragraphs:
+        if not p.text.strip():
+            continue
+        for run in p.runs:
+            if run.font.size:
+                sizes[run.font.size.pt] += 1
+                break  # only check the first run with an explicit size per paragraph
+    return sizes.most_common(1)[0][0] if sizes else 16.0
+
+
+def _detect_heading_level(para, body_font_size: float) -> int:
+    """Return heading level (0 = not a heading, 1-6 = heading level)."""
+    # Prefer Word built-in Heading styles
+    style_name = para.style.name if para.style else ""
+    if style_name.startswith("Heading"):
+        try:
+            return int(style_name.split()[-1])
+        except ValueError:
+            return 1
+    if style_name == "Title":
+        return 1
+    if style_name == "Subtitle":
+        return 2
+
+    text = para.text.strip()
+    if not text:
+        return 0
+
+    # Use font size of the first run that has an explicit size
+    font_size = None
+    for run in para.runs:
+        if run.font.size:
+            font_size = run.font.size.pt
+            break
+
+    # Significantly larger than body font -> treat as heading (threshold: 1.5x, short paragraphs only)
+    if font_size and font_size > body_font_size * 1.5:
+        from docx.enum.text import WD_ALIGN_PARAGRAPH
+
+        if para.alignment == WD_ALIGN_PARAGRAPH.CENTER:
+            return 1
+        if len(text) <= 60:
+            return 2
+
+    # Chinese numbered heading patterns
+    if _RE_H2.match(text):
+        return 2
+    if _RE_H3.match(text):
+        return 3
+
+    return 0
+
+
+class _FieldState:
+    """Mutable state for tracking w:fldChar field codes across paragraphs."""
+
+    __slots__ = ("active", "phase", "nest_depth", "url")
+    active: bool
+    phase: object  # None | "instr" | "result"
+    nest_depth: int
+    url: object  # None | str
+
+    def __init__(self):
+        self.active = False
+        self.phase = None  # None | "instr" | "result"
+        self.nest_depth = 0
+        self.url = None
+
+
+def _update_field_state_for_paragraph(para_element, field_state):
+    """Update field_state by scanning a paragraph element's runs.
+
+    Used for early-exit paragraphs (TOC / math / empty / code) to keep
+    cross-paragraph field tracking accurate without building item lists.
+    """
+    for child in para_element:
+        tag = child.tag.split("}")[-1] if "}" in child.tag else child.tag
+        if tag != "r":
+            continue
+        fld_char = child.find(_W + "fldChar")
+        if fld_char is not None:
+            fld_type = fld_char.get(_W + "fldCharType")
+            if fld_type == "begin":
+                if field_state.phase == "result":
+                    field_state.nest_depth += 1
+                else:
+                    field_state.active = True
+                    field_state.phase = "instr"
+                    field_state.url = None
+            elif fld_type == "separate":
+                if field_state.nest_depth == 0:
+                    field_state.phase = "result"
+            elif fld_type == "end":
+                if field_state.nest_depth > 0:
+                    field_state.nest_depth -= 1
+                else:
+                    field_state.active = False
+                    field_state.phase = None
+                    field_state.url = None
+            continue
+        instr_elem = child.find(_W + "instrText")
+        if instr_elem is not None and field_state.phase == "instr":
+            if instr_elem.text:
+                m = _RE_FIELD_HYPERLINK.search(instr_elem.text)
+                if m:
+                    field_state.url = m.group(1)
+
+
+def _iter_paragraph_items(para, field_state=None) -> list:
+    """Extract (bold, italic, underline, strikethrough, superscript, subscript, text, url) tuples from a paragraph in document order.
+
+    Handles python-docx Hyperlink objects and w:fldChar field-code hyperlinks.
+    Silently degrades to plain text on error.
+    Note: underline is forced to False inside Hyperlink/field-hyperlink runs to avoid Word's default hyperlink underline style.
+
+    field_state: optional _FieldState instance for cross-paragraph field tracking.
+                 If None, a fresh _FieldState is created (single-paragraph mode).
+    """
+    if field_state is None:
+        field_state = _FieldState()
+
+    def _split_breaks(items):
+        """Expand items containing \\n (from <w:br/> soft line breaks) into per-line items with <br> separators."""
+        expanded = []
+        for (
+            bold,
+            italic,
+            underline,
+            strikethrough,
+            superscript,
+            subscript,
+            text,
+            url,
+        ) in items:
+            if "\n" not in text:
+                expanded.append(
+                    (
+                        bold,
+                        italic,
+                        underline,
+                        strikethrough,
+                        superscript,
+                        subscript,
+                        text,
+                        url,
+                    )
+                )
+                continue
+            segments = text.split("\n")
+            for j, seg in enumerate(segments):
+                if seg:
+                    expanded.append(
+                        (
+                            bold,
+                            italic,
+                            underline,
+                            strikethrough,
+                            superscript,
+                            subscript,
+                            seg,
+                            url,
+                        )
+                    )
+                if j < len(segments) - 1:
+                    expanded.append(
+                        (False, False, False, False, False, False, "<br>\n", "")
+                    )
+        return expanded
+
+    try:
+        from docx.text.hyperlink import Hyperlink
+    except ImportError:
+        return _split_breaks(
+            [
+                (
+                    _effective_bold(r, para),
+                    _effective_italic(r, para),
+                    _effective_underline(r, para),
+                    bool(r.font.strike),
+                    _effective_superscript(r),
+                    _effective_subscript(r),
+                    r.text,
+                    "",
+                )
+                for r in para.runs
+                if r.text
+            ]
+        )
+
+    items = []
+    try:
+        content_iter = para.iter_inner_content()
+    except Exception:
+        return _split_breaks(
+            [
+                (
+                    _effective_bold(r, para),
+                    _effective_italic(r, para),
+                    _effective_underline(r, para),
+                    bool(r.font.strike),
+                    _effective_superscript(r),
+                    _effective_subscript(r),
+                    r.text,
+                    "",
+                )
+                for r in para.runs
+                if r.text
+            ]
+        )
+
+    for element in content_iter:
+        try:
+            if isinstance(element, Hyperlink):
+                try:
+                    url = element.url or ""
+                except (KeyError, AttributeError):
+                    url = ""
+                for run in element.runs:
+                    if not run.text:
+                        continue
+                    # Force underline=False: Word's Hyperlink style adds underline by default
+                    items.append(
+                        (
+                            _effective_bold(run, para),
+                            _effective_italic(run, para),
+                            False,
+                            bool(run.font.strike),
+                            _effective_superscript(run),
+                            _effective_subscript(run),
+                            run.text,
+                            url,
+                        )
+                    )
+                # Fallback: hyperlink with no runs but has text
+                if not element.runs and element.text:
+                    items.append(
+                        (False, False, False, False, False, False, element.text, url)
+                    )
+            else:
+                # Plain Run — check for fldChar control elements first
+                fld_char = element._element.find(_W + "fldChar")
+                if fld_char is not None:
+                    fld_type = fld_char.get(_W + "fldCharType")
+                    if fld_type == "begin":
+                        if field_state.phase == "result":
+                            field_state.nest_depth += 1
+                        else:
+                            field_state.active = True
+                            field_state.phase = "instr"
+                            field_state.url = None
+                    elif fld_type == "separate":
+                        if field_state.nest_depth == 0:
+                            field_state.phase = "result"
+                    elif fld_type == "end":
+                        if field_state.nest_depth > 0:
+                            field_state.nest_depth -= 1
+                        else:
+                            field_state.active = False
+                            field_state.phase = None
+                            field_state.url = None
+                    continue
+
+                instr_elem = element._element.find(_W + "instrText")
+                if instr_elem is not None:
+                    if field_state.active and field_state.phase == "instr":
+                        # Extract HYPERLINK url from instrText
+                        if instr_elem.text:
+                            m = _RE_FIELD_HYPERLINK.search(instr_elem.text)
+                            if m:
+                                field_state.url = m.group(1)
+                    continue  # Never emit instrText run as content
+
+                # Plain Run
+                if not element.text:
+                    continue
+
+                # If we are in the result phase of a field-code hyperlink, apply URL
+                # and suppress underline (same as w:hyperlink element handling above).
+                if (
+                    field_state.phase == "result"
+                    and field_state.nest_depth == 0
+                    and field_state.url
+                ):
+                    items.append(
+                        (
+                            _effective_bold(element, para),
+                            _effective_italic(element, para),
+                            False,  # suppress underline for field hyperlinks
+                            bool(element.font.strike),
+                            _effective_superscript(element),
+                            _effective_subscript(element),
+                            element.text,
+                            field_state.url,
+                        )
+                    )
+                else:
+                    items.append(
+                        (
+                            _effective_bold(element, para),
+                            _effective_italic(element, para),
+                            _effective_underline(element, para),
+                            bool(element.font.strike),
+                            _effective_superscript(element),
+                            _effective_subscript(element),
+                            element.text,
+                            "",
+                        )
+                    )
+        except Exception:
+            continue
+
+    return _split_breaks(items)
+
+
+def _merge_runs(items) -> list:
+    """Merge adjacent items with identical (bold, italic, underline, strikethrough, superscript, subscript, url).
+
+    Returns [(bold, italic, underline, strikethrough, superscript, subscript, text, url)].
+    """
+    merged: list[tuple[bool, bool, bool, bool, bool, bool, str, str]] = []
+    for (
+        bold,
+        italic,
+        underline,
+        strikethrough,
+        superscript,
+        subscript,
+        text,
+        url,
+    ) in items:
+        if not text:
+            continue
+        if (
+            merged
+            and merged[-1][0] == bold
+            and merged[-1][1] == italic
+            and merged[-1][2] == underline
+            and merged[-1][3] == strikethrough
+            and merged[-1][4] == superscript
+            and merged[-1][5] == subscript
+            and merged[-1][7] == url
+        ):
+            merged[-1] = (
+                bold,
+                italic,
+                underline,
+                strikethrough,
+                superscript,
+                subscript,
+                merged[-1][6] + text,
+                url,
+            )
+        else:
+            merged.append(
+                (
+                    bold,
+                    italic,
+                    underline,
+                    strikethrough,
+                    superscript,
+                    subscript,
+                    text,
+                    url,
+                )
+            )
+    return merged
+
+
+def _runs_to_markdown(items) -> str:
+    """Convert paragraph items to Markdown inline text, merging adjacent items with identical formatting."""
+    parts = []
+    for (
+        bold,
+        italic,
+        underline,
+        strikethrough,
+        superscript,
+        subscript,
+        text,
+        url,
+    ) in _merge_runs(items):
+        if bold or italic or underline or strikethrough or superscript or subscript:
+            # CommonMark: marker characters must not be surrounded by spaces
+            leading = len(text) - len(text.lstrip())
+            trailing = len(text) - len(text.rstrip())
+            prefix = text[:leading] if leading else ""
+            suffix = text[len(text) - trailing :] if trailing else ""
+            inner = text.strip()
+            if inner:
+                # Apply strikethrough first (innermost)
+                if strikethrough:
+                    inner = f"~~{inner}~~"
+                # Apply bold/italic
+                if bold and italic:
+                    inner = f"***{inner}***"
+                elif bold:
+                    inner = f"**{inner}**"
+                elif italic:
+                    inner = f"*{inner}*"
+                # Apply underline
+                if underline:
+                    inner = f"<u>{inner}</u>"
+                # Apply superscript/subscript (outermost)
+                if superscript:
+                    inner = f"<sup>{inner}</sup>"
+                elif subscript:
+                    inner = f"<sub>{inner}</sub>"
+                text = prefix + inner + suffix
+            elif underline and text:
+                # Pure whitespace + underline = fill-in line (e.g. "作者姓名：___")
+                # Replace spaces with NBSP so Markdown renderers preserve width
+                text = "<u>" + "\u00a0" * len(text) + "</u>"
+        if url:
+            text = f"[{text}]({_escape_md_url(url)})"
+        parts.append(text)
+    # Prevent bold/italic/strikethrough markers from merging with adjacent alphanumeric text (CommonMark requirement)
+    result = []
+    for i, part in enumerate(parts):
+        if i > 0 and result:
+            prev = result[-1]
+            # Previous part ends with closing marker and current part starts with alphanumeric
+            if prev.endswith(("**", "*", "~~")) and part and part[0].isalnum():
+                result.append("\u200b")
+        result.append(part)
+    return "".join(result)
+
+
+def _runs_to_html(items) -> str:
+    """Convert paragraph items to HTML inline text."""
+    parts = []
+    for (
+        bold,
+        italic,
+        underline,
+        strikethrough,
+        superscript,
+        subscript,
+        text,
+        url,
+    ) in _merge_runs(items):
+        if bold:
+            text = f"<b>{text}</b>"
+        if italic:
+            text = f"<i>{text}</i>"
+        if underline:
+            text = f"<u>{text}</u>"
+        if strikethrough:
+            text = f"<del>{text}</del>"
+        if superscript:
+            text = f"<sup>{text}</sup>"
+        if subscript:
+            text = f"<sub>{text}</sub>"
+        if url:
+            text = f'<a href="{url}">{text}</a>'
+        parts.append(text)
+    return "".join(parts)
+
+
+# Regex for TOC style names (e.g. "toc 1", "TOC 2", "TOC3")
+_RE_TOC_STYLE = re.compile(r"(?i)^toc\s*(\d+)$")
+
+# Flat TOC styles with no level concept (e.g. figure/table of contents)
+_TOC_FLAT_STYLES = {"table of figures"}
+
+# Regex for PAGEREF anchor in field instructions
+_RE_PAGEREF = re.compile(r"PAGEREF\s+(_Toc\w+)")
+
+
+def _is_toc_paragraph(para):
+    """Return TOC level (1-9) if paragraph uses a TOC style, else None."""
+    style_name = (para.style.name if para.style else "").strip()
+    m = _RE_TOC_STYLE.match(style_name)
+    if m:
+        return int(m.group(1))
+    if style_name.lower() in _TOC_FLAT_STYLES:
+        return 1
+    return None
+
+
+def _extract_toc_text(para) -> str:
+    """Extract display text from a TOC paragraph, stripping trailing page numbers."""
+    text = para.text
+    # Remove trailing page number: split on last tab, discard if it's a pure number
+    if "\t" in text:
+        before_tab, _, after_tab = text.rpartition("\t")
+        if after_tab.strip().isdigit():
+            text = before_tab
+    return text.strip()
+
+
+def _extract_toc_anchor(para_element):
+    """Extract anchor name from a TOC paragraph element.
+
+    Priority:
+    1. w:hyperlink[@w:anchor] attribute
+    2. PAGEREF _TocXXXX in w:instrText
+    """
+    # Check for w:hyperlink with anchor attribute
+    for hl in para_element.findall(f".//{_W}hyperlink"):
+        anchor = hl.get(f"{_W}anchor")
+        if anchor and anchor.startswith("_Toc"):
+            return anchor
+
+    # Fall back to PAGEREF field instruction
+    for instr in para_element.findall(f".//{_W}instrText"):
+        if instr.text:
+            m = _RE_PAGEREF.search(instr.text)
+            if m:
+                return m.group(1)
+    return None
+
+
+def _toc_entries_to_markdown(entries: list) -> str:
+    """Convert list of (text, anchor, level) TOC entries to Markdown list."""
+    lines = []
+    for text, anchor, level in entries:
+        indent = "  " * (level - 1)
+        if anchor:
+            lines.append(f"{indent}- [{text}](#{anchor})")
+        else:
+            lines.append(f"{indent}- {text}")
+    return "\n".join(lines)
+
+
+def _extract_heading_toc_bookmarks(para_element):
+    """Return all _Toc bookmark names found in a paragraph element.
+
+    A heading may carry multiple _Toc bookmarks from repeated TOC updates
+    (each update inserts a new bookmark without removing old ones).  Returning
+    all of them ensures that TOC entries from any generation can link here.
+    """
+    names = []
+    for bm in para_element.findall(f".//{_W}bookmarkStart"):
+        name = bm.get(f"{_W}name", "")
+        if name.startswith("_Toc"):
+            names.append(name)
+    return names
+
+
+def _flatten_body(body):
+    """Yield body children, expanding sdt elements into their sdtContent children."""
+    for child in body:
+        tag = child.tag.split("}")[-1] if "}" in child.tag else child.tag
+        if tag == "sdt":
+            sdt_content = child.find(f"{_W}sdtContent")
+            if sdt_content is not None:
+                yield from _flatten_body(sdt_content)
+        else:
+            yield child
+
+
+_CODE_FONTS = {
+    "Courier New",
+    "Courier",
+    "Consolas",
+    "Monaco",
+    "Menlo",
+    "Source Code Pro",
+    "Fira Code",
+    "DejaVu Sans Mono",
+    "monospace",
+}
+
+
+def _is_code_paragraph(para) -> bool:
+    """Return True if all text-bearing runs in the paragraph use a monospace font."""
+    runs_with_text = [r for r in para.runs if r.text.strip()]
+    if not runs_with_text:
+        return False
+    runs_with_font = [r for r in runs_with_text if r.font.name]
+    # At least one run must have an explicit font, and all such runs must be monospace
+    if not runs_with_font:
+        return False
+    return all(r.font.name in _CODE_FONTS for r in runs_with_font)
+
+
+def _get_content_width(doc) -> int:
+    """Return the content area width of the document in EMU."""
+    section = doc.sections[0]
+    return section.page_width - section.left_margin - section.right_margin
+
+
+def _table_to_html(
+    table, doc, image_counter: list, images: dict, content_width: int = 0
+) -> str:
+    """Convert a python-docx Table to an HTML table, handling merged cells and inline images."""
+    grid = [[cell for cell in row.cells] for row in table.rows]
+    nrows = len(grid)
+    if nrows == 0:
+        return ""
+    ncols = len(grid[0])
+
+    visited: set[tuple[int, int]] = set()
+    html_parts = ["<table>"]
+
+    for i in range(nrows):
+        html_parts.append("<tr>")
+        for j in range(ncols):
+            if (i, j) in visited:
+                continue
+            tc = grid[i][j]._tc
+            # Compute colspan
+            colspan = 1
+            while j + colspan < ncols and grid[i][j + colspan]._tc is tc:
+                visited.add((i, j + colspan))
+                colspan += 1
+            # Compute rowspan
+            rowspan = 1
+            while i + rowspan < nrows and grid[i + rowspan][j]._tc is tc:
+                for k in range(colspan):
+                    visited.add((i + rowspan, j + k))
+                rowspan += 1
+
+            cell = grid[i][j]
+            content_parts = []
+            for para in cell.paragraphs:
+                img_list = _extract_images_from_paragraph(para, doc, image_counter)
+                for filename, img_bytes, cx_emu in img_list:
+                    rel_path = f"images/{filename}"
+                    images[rel_path] = img_bytes
+                    if cx_emu and content_width:
+                        pct = min(round(cx_emu / content_width * 100), 100)
+                        content_parts.append(
+                            f'<img src="images/{filename}" width="{pct}%">'
+                        )
+                    else:
+                        content_parts.append(f'<img src="images/{filename}">')
+                para_html = (
+                    _paragraph_math_to_html(para).strip()
+                    if _paragraph_has_math(para)
+                    else _runs_to_html(_iter_paragraph_items(para)).strip()
+                    or para.text.strip()
+                )
+                if para_html:
+                    content_parts.append(para_html)
+                # Extract text box content from table cell paragraph
+                try:
+                    tb_groups = _extract_textbox_paragraphs(para._element)
+                    for tb_paras in tb_groups:
+                        for tb_p in tb_paras:
+                            tb_t_elems = tb_p.findall(f".//{_W}t")
+                            tb_text = "".join(
+                                t.text for t in tb_t_elems if t.text
+                            ).strip()
+                            if tb_text:
+                                content_parts.append(f"[{tb_text}]")
+                except Exception:
+                    pass
+            cell_html = "<br>".join(content_parts) if content_parts else ""
+
+            is_header = False
+            trPr = table.rows[i]._tr.find(f"{_W}trPr")
+            if trPr is not None:
+                tbl_header = trPr.find(f"{_W}tblHeader")
+                if tbl_header is not None:
+                    val = tbl_header.get(f"{_W}val")
+                    is_header = val is None or val.lower() not in ("false", "0", "off")
+            if not is_header and i == 0 and nrows > 1:
+                is_header = (
+                    True  # fallback: treat first row as header in multi-row tables
+                )
+            tag = "th" if is_header else "td"
+            attrs = ""
+            if colspan > 1:
+                attrs += f' colspan="{colspan}"'
+            if rowspan > 1:
+                attrs += f' rowspan="{rowspan}"'
+            html_parts.append(f"<{tag}{attrs}>{cell_html}</{tag}>")
+        html_parts.append("</tr>")
+
+    html_parts.append("</table>")
+    return "\n".join(html_parts)
+
+
+_MIME_TO_EXT = {
+    "image/jpeg": "jpeg",
+    "image/jpg": "jpg",
+    "image/png": "png",
+    "image/gif": "gif",
+    "image/bmp": "bmp",
+    "image/tiff": "tiff",
+    "image/webp": "webp",
+    "image/svg+xml": "svg",
+    "image/x-emf": "emf",
+    "image/x-wmf": "wmf",
+}
+
+
+def _extract_images_from_paragraph(para, doc, image_counter: list) -> list:
+    """Extract images from a paragraph. Returns [(filename, bytes, cx_emu)] where cx_emu is 0 if unknown."""
+    results = []
+    containers = para._element.findall(f".//{_WPD}inline") + para._element.findall(
+        f".//{_WPD}anchor"
+    )
+    for container in containers:
+        extent = container.find(f"{_WPD}extent")
+        cx_emu = int(extent.get("cx", 0)) if extent is not None else 0
+
+        blip = container.find(f".//{_A}blip")
+        if blip is None:
+            continue
+        r_embed = blip.get(f"{_R}embed")
+        if not r_embed:
+            continue
+        try:
+            rel = doc.part.rels[r_embed]
+            img_bytes = rel.target_part.blob
+            content_type = rel.target_part.content_type
+            ext = _MIME_TO_EXT.get(content_type, "png")
+            image_counter[0] += 1
+            filename = f"image{image_counter[0]}.{ext}"
+            results.append((filename, img_bytes, cx_emu))
+        except (KeyError, AttributeError):
+            pass
+    return results
+
+
+def _extract_chart_tables(para, doc) -> list:
+    """Extract chart data as HTML tables from drawings in a paragraph."""
+    try:
+        from datetime import datetime, timedelta
+
+        results = []
+        containers = para._element.findall(f".//{_WPD}inline") + para._element.findall(
+            f".//{_WPD}anchor"
+        )
+
+        for container in containers:
+            chart_ref = container.find(f".//{_C}chart")
+            if chart_ref is None:
+                continue
+            r_id = chart_ref.get(f"{_R}id")
+            if not r_id:
+                continue
+            try:
+                rel = doc.part.rels[r_id]
+                chart_part = rel.target_part
+            except (KeyError, AttributeError):
+                continue
+
+            try:
+                import lxml.etree as etree
+
+                chart_root = etree.fromstring(chart_part.blob)
+            except Exception:
+                continue
+
+            # Extract chart title
+            title_text = ""
+            title_el = chart_root.find(f".//{_C}title")
+            if title_el is not None:
+                a_t_els = title_el.findall(f".//{_A}t")
+                title_text = "".join((el.text or "") for el in a_t_els).strip()
+
+            # Extract axis titles
+            cat_ax_title = ""
+            for ax_tag in (f"{_C}catAx", f"{_C}dateAx"):
+                ax_el = chart_root.find(f".//{ax_tag}")
+                if ax_el is not None:
+                    t_el = ax_el.find(f"{_C}title")
+                    if t_el is not None:
+                        texts = t_el.findall(f".//{_A}t")
+                        cat_ax_title = "".join(el.text or "" for el in texts).strip()
+                        break
+
+            val_ax_title = ""
+            val_ax_el = chart_root.find(f".//{_C}valAx")
+            if val_ax_el is not None:
+                t_el = val_ax_el.find(f"{_C}title")
+                if t_el is not None:
+                    texts = t_el.findall(f".//{_A}t")
+                    val_ax_title = "".join(el.text or "" for el in texts).strip()
+
+            # Extract series data
+            series_list = chart_root.findall(f".//{_C}ser")
+            if not series_list:
+                continue
+
+            # Collect categories from first series
+            categories = []
+            cat_is_date = False
+            date_format_code = ""
+            first_ser = series_list[0]
+            cat_el = first_ser.find(f"{_C}cat")
+            if cat_el is not None:
+                str_cache = cat_el.find(f".//{_C}strCache")
+                num_cache = cat_el.find(f".//{_C}numCache")
+                if str_cache is not None:
+                    for pt in str_cache.findall(f"{_C}pt"):
+                        v = pt.find(f"{_C}v")
+                        categories.append(v.text if v is not None else "")
+                elif num_cache is not None:
+                    fmt_el = num_cache.find(f"{_C}formatCode")
+                    if fmt_el is not None:
+                        fc = fmt_el.text or ""
+                        date_format_code = fc
+                        cat_is_date = any(k in fc.lower() for k in ("y", "m", "d"))
+                    for pt in num_cache.findall(f"{_C}pt"):
+                        v = pt.find(f"{_C}v")
+                        if v is not None and v.text:
+                            try:
+                                serial = float(v.text)
+                                if cat_is_date:
+                                    dt = datetime(1899, 12, 30) + timedelta(days=serial)
+                                    categories.append(dt.strftime("%Y-%m-%d"))
+                                else:
+                                    categories.append(v.text)
+                            except (ValueError, OverflowError):
+                                categories.append(v.text or "")
+                        else:
+                            categories.append("")
+
+            # Collect series names and values
+            series_names = []
+            series_values = []
+            for ser in series_list:
+                # Series name: val axis title > <c:tx> > ""
+                name = val_ax_title
+                if not name:
+                    tx_el = ser.find(f"{_C}tx")
+                    if tx_el is not None:
+                        str_ref = tx_el.find(f".//{_C}strCache")
+                        if str_ref is not None:
+                            pt = str_ref.find(f"{_C}pt")
+                            if pt is not None:
+                                v = pt.find(f"{_C}v")
+                                name = v.text if v is not None else ""
+                        else:
+                            v = tx_el.find(f"{_C}v")
+                            if v is not None:
+                                name = v.text or ""
+                series_names.append(name)
+
+                # Values
+                vals = []
+                val_el = ser.find(f"{_C}val")
+                if val_el is not None:
+                    num_cache = val_el.find(f".//{_C}numCache")
+                    if num_cache is not None:
+                        pts = {
+                            int(pt.get("idx", 0)): pt
+                            for pt in num_cache.findall(f"{_C}pt")
+                        }
+                        max_idx = max(pts.keys()) if pts else -1
+                        for idx in range(max_idx + 1):
+                            pt = pts.get(idx)
+                            if pt is not None:
+                                v = pt.find(f"{_C}v")
+                                vals.append(v.text if v is not None else "")
+                            else:
+                                vals.append("")
+                series_values.append(vals)
+
+            if not series_names:
+                continue
+
+            # Build HTML table
+            n_rows = max(
+                len(categories),
+                max((len(v) for v in series_values), default=0),
+            )
+            html_parts = ["<table>"]
+            if title_text:
+                html_parts.append(f"<caption>{title_text}</caption>")
+            # Header row — omit <thead> entirely if no header info
+            has_header = cat_ax_title or any(name for name in series_names)
+            if has_header:
+                html_parts.append("<thead><tr>")
+                html_parts.append(f"<th>{cat_ax_title}</th>")
+                for name in series_names:
+                    html_parts.append(f"<th>{name}</th>")
+                html_parts.append("</tr></thead>")
+            # Data rows
+            html_parts.append("<tbody>")
+            for i in range(n_rows):
+                cat = categories[i] if i < len(categories) else ""
+                html_parts.append(f"<tr><td>{cat}</td>")
+                for vals in series_values:
+                    val = vals[i] if i < len(vals) else ""
+                    html_parts.append(f"<td>{val}</td>")
+                html_parts.append("</tr>")
+            html_parts.append("</tbody></table>")
+            results.append("".join(html_parts))
+
+        return results
+    except Exception:
+        return []
+
+
+def _extract_textbox_paragraphs(element) -> list:
+    """Extract paragraphs from text boxes embedded in a body element.
+
+    Text boxes appear as mc:AlternateContent > mc:Choice > wps:txbx > w:txbxContent > w:p.
+    Only mc:Choice is used to avoid duplicating VML fallback content.
+
+    Returns list of lists of w:p elements, one list per non-empty text box.
+    """
+    try:
+        result = []
+        for alt_content in element.iter(f"{_MC}AlternateContent"):
+            choice = alt_content.find(f"{_MC}Choice")
+            if choice is None:
+                continue
+            for txbx in choice.iter(f"{_WPS}txbx"):
+                txbx_content = txbx.find(f"{_W}txbxContent")
+                if txbx_content is None:
+                    continue
+                paras = txbx_content.findall(f"{_W}p")
+                if not paras:
+                    continue
+                # Skip boxes where all paragraphs are empty
+                has_text = any(
+                    p.find(f".//{_W}t") is not None
+                    and any(t.text for t in p.findall(f".//{_W}t") if t.text)
+                    for p in paras
+                )
+                if has_text:
+                    result.append(paras)
+        return result
+    except Exception:
+        return []
+
+
+def _textbox_paragraphs_to_markdown(
+    textbox_groups, doc, body_font_size, image_counter, images, content_width
+) -> list:
+    """Convert text box paragraph groups to blockquote Markdown lines.
+
+    Returns a list of strings (lines) to append to the main output.
+    """
+    try:
+        from docx.text.paragraph import Paragraph
+    except ImportError:
+        return []
+
+    output_lines = []
+    for group_idx, para_elements in enumerate(textbox_groups):
+        group_lines = []
+        for p_elem in para_elements:
+            try:
+                para = Paragraph(p_elem, doc)
+
+                # Handle images in text box
+                img_list = _extract_images_from_paragraph(para, doc, image_counter)
+                for filename, img_bytes, cx_emu in img_list:
+                    rel_path = f"images/{filename}"
+                    images[rel_path] = img_bytes
+                    if cx_emu and content_width:
+                        pct = min(round(cx_emu / content_width * 100), 100)
+                        group_lines.append(
+                            f'> <img src="images/{filename}" width="{pct}%">'
+                        )
+                    else:
+                        group_lines.append(f'> <img src="images/{filename}">')
+
+                # Handle math formulas
+                if _paragraph_has_math(para):
+                    math_md = _paragraph_math_to_markdown(para)
+                    if math_md:
+                        group_lines.append(f"> {math_md}")
+                    continue
+
+                # Plain inline text (no heading/list/code detection for text boxes)
+                inline = _runs_to_markdown(_iter_paragraph_items(para))
+                if not inline:
+                    inline = para.text.strip()
+                if inline:
+                    group_lines.append(f"> {inline}")
+            except Exception:
+                continue
+
+        if group_lines:
+            if output_lines:
+                output_lines.append("")
+            output_lines.extend(group_lines)
+
+    return output_lines
+
+
+def _build_numbering_map(doc) -> dict:
+    """Parse numbering.xml and return {numId: {ilvl: numFmt}} mapping."""
+    numbering_map = {}
+    try:
+        numbering_part = doc.part.numbering_part
+    except Exception:
+        return numbering_map
+    numbering_elem = numbering_part._element
+
+    # Build abstractNumId -> {ilvl: numFmt}
+    abstract = {}
+    for abs_num in numbering_elem.findall(f"{_W}abstractNum"):
+        abs_id = abs_num.get(f"{_W}abstractNumId")
+        levels = {}
+        for lvl in abs_num.findall(f"{_W}lvl"):
+            ilvl = int(lvl.get(f"{_W}ilvl", "0"))
+            fmt_elem = lvl.find(f"{_W}numFmt")
+            fmt = (
+                fmt_elem.get(f"{_W}val", "bullet") if fmt_elem is not None else "bullet"
+            )
+            levels[ilvl] = fmt
+        abstract[abs_id] = levels
+
+    # Map numId -> abstractNumId
+    for num in numbering_elem.findall(f"{_W}num"):
+        num_id = num.get(f"{_W}numId")
+        abs_ref = num.find(f"{_W}abstractNumId")
+        if abs_ref is not None:
+            abs_id = abs_ref.get(f"{_W}val")
+            if abs_id in abstract:
+                numbering_map[num_id] = abstract[abs_id]
+    return numbering_map
+
+
+def _get_list_info(para, numbering_map) -> tuple | None:
+    """Return (list_type, ilvl, num_id) or None. list_type: 'bullet' | 'ordered'"""
+    pPr = para._element.find(f"{_W}pPr")
+    if pPr is None:
+        return None
+    numPr = pPr.find(f"{_W}numPr")
+    if numPr is None:
+        return None
+    numId_elem = numPr.find(f"{_W}numId")
+    ilvl_elem = numPr.find(f"{_W}ilvl")
+    if numId_elem is None:
+        return None
+    num_id = numId_elem.get(f"{_W}val")
+    ilvl = int(ilvl_elem.get(f"{_W}val", "0")) if ilvl_elem is not None else 0
+    if num_id not in numbering_map:
+        return None
+    fmt = numbering_map[num_id].get(ilvl, "bullet")
+    list_type = (
+        "ordered"
+        if fmt in ("decimal", "lowerLetter", "upperLetter", "lowerRoman", "upperRoman")
+        else "bullet"
+    )
+    return (list_type, ilvl, num_id)
+
+
+def _convert_body(doc, *, extract_drawings=True) -> tuple:
+    """Traverse body elements in order and produce Markdown. Returns (markdown_str, images_dict)."""
+    try:
+        from docx.table import Table
+        from docx.text.paragraph import Paragraph
+    except ImportError:
+        raise RuntimeError(
+            "DOCX conversion requires python-docx: pip install paddleocr[doc2md]"
+        )
+
+    body_font_size = _get_body_font_size(doc)
+    content_width = _get_content_width(doc)
+    numbering_map = _build_numbering_map(doc)
+    lines: list[str] = []
+    images: dict = {}
+    image_counter = [0]  # wrapped in list so inner functions can mutate it
+    code_buf: list[str] = []  # buffer for consecutive code paragraphs
+    toc_buf: list[tuple] = []  # buffer for consecutive TOC paragraphs
+    ol_counters: dict[str, int] = {}  # key = "{numId}-{ilvl}", value = current index
+    prev_was_list = False
+
+    def flush_code_buf():
+        """Flush the code buffer as a fenced code block."""
+        if code_buf:
+            lines.append("```")
+            lines.extend(code_buf)
+            lines.append("```")
+            lines.append("")
+            code_buf.clear()
+
+    def flush_toc_buf():
+        """Flush the TOC buffer as a Markdown list."""
+        if toc_buf:
+            if lines and lines[-1] != "":
+                lines.append("")
+            lines.append(_toc_entries_to_markdown(toc_buf))
+            lines.append("")
+            toc_buf.clear()
+
+    field_state = _FieldState()
+    for child in _flatten_body(doc.element.body):
+        tag = child.tag.split("}")[-1]
+
+        if tag == "p":
+            para = Paragraph(child, doc)
+
+            # TOC paragraph detection — before image extraction and other processing
+            toc_level = _is_toc_paragraph(para)
+            if toc_level is not None:
+                toc_text = _extract_toc_text(para)
+                if toc_text:
+                    toc_anchor = _extract_toc_anchor(child)
+                    toc_buf.append((toc_text, toc_anchor, toc_level))
+                _update_field_state_for_paragraph(child, field_state)
+                continue
+
+            # Non-TOC paragraph: flush any buffered TOC entries
+            flush_toc_buf()
+
+            # Extract text box content for this paragraph element
+            pending_textbox_lines = []
+            if extract_drawings:
+                tb_groups = _extract_textbox_paragraphs(child)
+                if tb_groups:
+                    pending_textbox_lines = _textbox_paragraphs_to_markdown(
+                        tb_groups,
+                        doc,
+                        body_font_size,
+                        image_counter,
+                        images,
+                        content_width,
+                    )
+
+            def _flush_textbox():
+                if pending_textbox_lines:
+                    if lines and lines[-1] != "":
+                        lines.append("")
+                    lines.extend(pending_textbox_lines)
+                    lines.append("")
+
+            # Extract images first
+            img_list = _extract_images_from_paragraph(para, doc, image_counter)
+            for filename, img_bytes, cx_emu in img_list:
+                flush_code_buf()
+                rel_path = f"images/{filename}"
+                images[rel_path] = img_bytes
+                if cx_emu and content_width:
+                    pct = min(round(cx_emu / content_width * 100), 100)
+                    lines.append(f'<img src="images/{filename}" width="{pct}%">')
+                else:
+                    lines.append(f'<img src="images/{filename}">')
+                lines.append("")
+
+            # Extract chart data tables
+            chart_tables = _extract_chart_tables(para, doc)
+            for chart_html in chart_tables:
+                flush_code_buf()
+                lines.append(chart_html)
+                lines.append("")
+
+            text = para.text.strip()
+
+            # Math formula detection — must check before skipping empty-text paragraphs
+            # (pure formula paragraphs have para.text == "")
+            if _paragraph_has_math(para):
+                flush_code_buf()
+                if prev_was_list:
+                    lines.append("")
+                prev_was_list = False
+                math_md = _paragraph_math_to_markdown(para)
+                if math_md:
+                    lines.append(math_md)
+                    lines.append("")
+                _flush_textbox()
+                _update_field_state_for_paragraph(child, field_state)
+                continue
+
+            if not text:
+                # Emit any _Toc bookmark anchors even for empty paragraphs.
+                # Headings whose visible text comes from list numbering (not w:t
+                # runs) have para.text == "" but still carry _Toc bookmarks that
+                # TOC entries link to.
+                for toc_bm in _extract_heading_toc_bookmarks(child):
+                    lines.append(f'<a id="{toc_bm}"></a>')
+                _flush_textbox()
+                if not img_list and not pending_textbox_lines:
+                    if code_buf:
+                        code_buf.append("")  # preserve blank lines inside code blocks
+                    elif lines and lines[-1] != "":
+                        lines.append("")
+                _update_field_state_for_paragraph(child, field_state)
+                continue
+
+            # Code paragraph: buffer it without heading/inline formatting
+            if _is_code_paragraph(para):
+                code_buf.append(para.text)
+                _flush_textbox()
+                _update_field_state_for_paragraph(child, field_state)
+                continue
+
+            # Non-code paragraph: flush any buffered code first
+            flush_code_buf()
+
+            level = _detect_heading_level(para, body_font_size)
+            inline = _runs_to_markdown(_iter_paragraph_items(para, field_state)) or text
+
+            if level > 0:
+                # Strip outer **...** wrapping that headings may have inherited
+                clean = inline.strip()
+                if clean.startswith("**") and clean.endswith("**"):
+                    clean = clean[2:-2]
+                # Heading lines cannot span multiple source lines; revert <br>\n → <br>
+                clean = clean.replace("<br>\n", "<br>")
+                if prev_was_list:
+                    lines.append("")
+                prev_was_list = False
+                # Add _Toc bookmark anchors if present (makes TOC links jumpable).
+                # A heading may have multiple _Toc bookmarks from repeated TOC updates.
+                for toc_bm in _extract_heading_toc_bookmarks(child):
+                    lines.append(f'<a id="{toc_bm}"></a>')
+                lines.append(f"{'#' * level} {clean}")
+                lines.append("")
+            else:
+                list_info = _get_list_info(para, numbering_map)
+                if list_info:
+                    list_type, ilvl, num_id = list_info
+                    indent = "    " * ilvl
+                    if list_type == "ordered":
+                        counter_key = f"{num_id}-{ilvl}"
+                        ol_counters[counter_key] = ol_counters.get(counter_key, 0) + 1
+                        prefix = f"{indent}{ol_counters[counter_key]}. "
+                    else:
+                        prefix = f"{indent}- "
+                    if not prev_was_list and lines and lines[-1] != "":
+                        lines.append("")
+                    prev_was_list = True
+                    # List item continuation lines need indentation; revert <br>\n → <br>
+                    lines.append(prefix + inline.replace("<br>\n", "<br>"))
+                else:
+                    if prev_was_list:
+                        lines.append("")
+                    prev_was_list = False
+                    # Reset ordered list counters when a list is interrupted
+                    ol_counters.clear()
+                    # Add _Toc bookmark anchors for non-heading paragraphs (e.g. Caption)
+                    for toc_bm in _extract_heading_toc_bookmarks(child):
+                        lines.append(f'<a id="{toc_bm}"></a>')
+                    lines.append(inline)
+                    lines.append("")
+
+            _flush_textbox()
+
+        elif tag == "tbl":
+            flush_code_buf()
+            flush_toc_buf()
+            if prev_was_list:
+                lines.append("")
+            prev_was_list = False
+            ol_counters.clear()
+            table = Table(child, doc)
+            if lines and lines[-1] != "":
+                lines.append("")
+            lines.append(
+                _table_to_html(table, doc, image_counter, images, content_width)
+            )
+            lines.append("")
+
+    # Strip trailing blank lines
+    flush_code_buf()
+    flush_toc_buf()
+    while lines and lines[-1] == "":
+        lines.pop()
+
+    return "\n".join(lines), images
+
+
+def _extract_headers_footers(doc) -> tuple:
+    """Extract header and footer text from all document sections.
+
+    Returns (header_lines: list[str], footer_lines: list[str]).
+    Deduplicates content and filters out page-number-only text.
+    """
+
+    def _collect_from_parts(parts, seen: set) -> list:
+        results = []
+        for part in parts:
+            try:
+                if part.is_linked_to_previous:
+                    continue
+                texts = []
+                for para in part.paragraphs:
+                    try:
+                        items = list(_iter_paragraph_items(para))
+                        inline = _runs_to_markdown(items)
+                        if not inline:
+                            inline = para.text.strip()
+                        if inline:
+                            texts.append(inline)
+                    except Exception:
+                        pass
+                text = " ".join(texts)
+                # Filter: skip empty text, pure digits (page numbers),
+                # page-number patterns (e.g. "第  页", "第6页", "共 页"), and duplicates
+                if (
+                    text
+                    and not text.strip().isdigit()
+                    and not _RE_PAGE_ONLY.match(text.strip())
+                    and text not in seen
+                ):
+                    seen.add(text)
+                    results.append(text)
+            except Exception:
+                pass
+        return results
+
+    header_lines = []
+    footer_lines = []
+    seen_headers: set = set()
+    seen_footers: set = set()
+
+    # Check if the document uses different odd/even page headers/footers
+    try:
+        odd_even = doc.settings.odd_and_even_pages_header_footer
+    except Exception:
+        odd_even = False
+
+    for section in doc.sections:
+        try:
+            # Collect headers
+            hdrs = [section.header]
+            if odd_even:
+                try:
+                    hdrs.append(section.even_page_header)
+                except Exception:
+                    pass
+            try:
+                if section.different_first_page_header_footer:
+                    hdrs.append(section.first_page_header)
+            except Exception:
+                pass
+            header_lines.extend(_collect_from_parts(hdrs, seen_headers))
+
+            # Collect footers
+            ftrs = [section.footer]
+            if odd_even:
+                try:
+                    ftrs.append(section.even_page_footer)
+                except Exception:
+                    pass
+            try:
+                if section.different_first_page_header_footer:
+                    ftrs.append(section.first_page_footer)
+            except Exception:
+                pass
+            footer_lines.extend(_collect_from_parts(ftrs, seen_footers))
+        except Exception:
+            pass
+
+    return header_lines, footer_lines
+
+
+@default_registry.register
+class DocxConverter(BaseConverter):
+    supported_extensions = ["docx"]
+    supported_mimetypes = [
+        "application/vnd.openxmlformats-officedocument.wordprocessingml.document"
+    ]
+
+    def convert_file(self, file_path: Path, **kwargs) -> ConvertResult:
+        try:
+            from docx import Document
+        except ImportError:
+            raise RuntimeError(
+                "DOCX conversion requires python-docx: pip install paddleocr[doc2md]"
+            )
+
+        extract_drawings = kwargs.pop("extract_drawings", True)
+        extract_headers_footers = kwargs.pop("extract_headers_footers", True)
+        doc = Document(str(file_path))
+        md_text, images = _convert_body(doc, extract_drawings=extract_drawings)
+
+        if extract_headers_footers:
+            header_lines, footer_lines = _extract_headers_footers(doc)
+            parts = []
+            if header_lines:
+                parts.append("\n\n".join(header_lines))
+                parts.append("")
+            parts.append(md_text)
+            if footer_lines:
+                parts.append("")
+                parts.append("---")
+                parts.append("\n\n".join(footer_lines))
+            md_text = "\n".join(parts)
+
+        return ConvertResult(
+            markdown=md_text,
+            images=images,
+            metadata={
+                "format": "DOCX",
+                "paragraph_count": len(doc.paragraphs),
+                "table_count": len(doc.tables),
+            },
+        )
diff --git a/paddleocr/_doc2md/converters/pptx.py b/paddleocr/_doc2md/converters/pptx.py
new file mode 100644
index 0000000..910bd9e
--- /dev/null
+++ b/paddleocr/_doc2md/converters/pptx.py
@@ -0,0 +1,563 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from pathlib import Path
+
+from ..base import BaseConverter, ConvertResult
+from ..math import (
+    extract_math_from_paragraph as _extract_math_from_paragraph,
+    paragraph_has_math as _paragraph_has_math,
+)
+from ..registry import default_registry
+
+# pptx XML namespace for DrawingML run properties
+_A_NS = "http://schemas.openxmlformats.org/drawingml/2006/main"
+_A = "{" + _A_NS + "}"
+_A_STRIKE = f"{_A}strike"
+_A_RPR = f"{_A}rPr"
+_A_P = f"{_A}p"
+
+# Markup Compatibility namespace
+_MC_NS = "http://schemas.openxmlformats.org/markup-compatibility/2006"
+_MC = "{" + _MC_NS + "}"
+_MC_CHOICE = f"{_MC}Choice"
+_MC_ALT = f"{_MC}AlternateContent"
+
+# Chart namespace
+_C_NS = "http://schemas.openxmlformats.org/drawingml/2006/chart"
+_C = "{" + _C_NS + "}"
+
+# Chart type name lookup (used in _chart_to_html)
+_CHART_TYPE_NAMES = {
+    1: "Area Chart",  # AREA
+    2: "Area Chart",  # AREA_STACKED
+    4: "Line Chart",  # LINE
+    5: "Pie Chart",  # PIE
+    15: "Bubble Chart",  # BUBBLE
+    51: "Column Chart",  # COLUMN_CLUSTERED
+    52: "Column Chart",  # COLUMN_STACKED
+    53: "Column Chart",  # COLUMN_STACKED_100
+    57: "Bar Chart",  # BAR_CLUSTERED
+    58: "Bar Chart",  # BAR_STACKED
+    65: "Line Chart",  # LINE_MARKERS
+    -4120: "Doughnut Chart",  # DOUGHNUT
+    -4151: "Radar Chart",  # RADAR
+    -4169: "Scatter Chart",  # XY_SCATTER
+}
+
+
+def _pptx_run_strike(run) -> bool:
+    """Return True if the run has strikethrough (sngStrike or dblStrike) set in XML."""
+    try:
+        rPr = run._r.find(_A_RPR)
+        if rPr is not None:
+            val = rPr.get(_A_STRIKE)
+            return val in ("sngStrike", "dblStrike")
+    except Exception:
+        pass
+    return False
+
+
+def _pptx_run_script(run) -> str:
+    """Return 'super', 'sub', or '' based on DrawingML baseline attribute.
+
+    baseline > 0  -> superscript, baseline < 0 -> subscript.
+    baseline attribute is on <a:rPr> with no namespace prefix.
+    """
+    try:
+        rPr = run._r.find(_A_RPR)
+        if rPr is not None:
+            val = rPr.get("baseline")
+            if val is not None:
+                n = int(val)
+                if n > 0:
+                    return "super"
+                if n < 0:
+                    return "sub"
+    except Exception:
+        pass
+    return ""
+
+
+def _escape_md_url(url: str) -> str:
+    """Escape parentheses in URL for Markdown link syntax."""
+    return url.replace("(", "%28").replace(")", "%29")
+
+
+def _classify_part(part: str) -> str:
+    """Classify a slide content part for grouping (heading/html/list/blockquote/other)."""
+    s = part.lstrip()
+    if s.startswith("##"):
+        return "heading"
+    if s.startswith("<"):
+        return "html"
+    if s.startswith("- "):
+        return "list"
+    if s.startswith(">"):
+        return "blockquote"
+    return "other"
+
+
+def _format_run_segment(
+    seg: str,
+    bold: bool,
+    italic: bool,
+    underline: bool,
+    strikethrough: bool,
+    script: str,
+    url: str,
+) -> str:
+    """Apply Markdown/HTML inline formatting to a text segment."""
+    t = seg
+    if bold or italic or underline or strikethrough or script:
+        leading = len(t) - len(t.lstrip())
+        trailing = len(t) - len(t.rstrip())
+        prefix = t[:leading] if leading else ""
+        suffix = t[len(t) - trailing :] if trailing else ""
+        inner = t.strip()
+        if inner:
+            if strikethrough:
+                inner = f"~~{inner}~~"
+            if bold and italic:
+                inner = f"***{inner}***"
+            elif bold:
+                inner = f"**{inner}**"
+            elif italic:
+                inner = f"*{inner}*"
+            if underline:
+                inner = f"<u>{inner}</u>"
+            if script == "super":
+                inner = f"<sup>{inner}</sup>"
+            elif script == "sub":
+                inner = f"<sub>{inner}</sub>"
+            t = prefix + inner + suffix
+        elif underline and t:
+            # Pure whitespace + underline = fill-in line
+            # Replace spaces with NBSP so Markdown renderers preserve width
+            t = "<u>" + "\u00a0" * len(t) + "</u>"
+    if url:
+        escaped_url = _escape_md_url(url)
+        t = f"[{t}]({escaped_url})"
+    return t
+
+
+@default_registry.register
+class PptxConverter(BaseConverter):
+    supported_extensions = ["pptx"]
+    supported_mimetypes = [
+        "application/vnd.openxmlformats-officedocument.presentationml.presentation"
+    ]
+
+    def convert_file(self, file_path: Path, **kwargs) -> ConvertResult:
+        try:
+            from pptx import Presentation
+            from pptx.shapes.picture import Picture
+        except ImportError:
+            raise RuntimeError(
+                "PPTX conversion requires python-pptx: pip install paddleocr[doc2md]"
+            )
+        try:
+            from pptx.enum.shapes import MSO_SHAPE_TYPE
+        except ImportError:
+            MSO_SHAPE_TYPE = None  # type: ignore[assignment]
+
+        prs = Presentation(str(file_path))
+        slides_md = []
+        images: dict = {}
+        image_counter = [0]
+        slide_width = prs.slide_width
+        self._Picture = Picture
+        self._MSO_SHAPE_TYPE = MSO_SHAPE_TYPE
+
+        for slide in prs.slides:
+            slide_parts = []
+
+            # Process all shapes
+            for shape in slide.shapes:
+                self._process_shape(
+                    shape, slide_parts, images, image_counter, slide_width, slide.part
+                )
+
+            # Handle math formulas inside mc:AlternateContent elements
+            # (python-pptx doesn't expose these as Shape objects)
+            for alt_content in slide._element.iter(_MC_ALT):
+                # Only look at mc:Choice (the preferred rendering path)
+                choice = alt_content.find(_MC_CHOICE)
+                if choice is None:
+                    continue
+                for para_elem in choice.iter(_A_P):
+                    if _paragraph_has_math(para_elem):
+                        math_items = _extract_math_from_paragraph(para_elem)
+                        for latex in math_items:
+                            slide_parts.append(f"$$\n{latex}\n$$")
+
+            # Speaker notes
+            if slide.has_notes_slide:
+                notes_text = slide.notes_slide.notes_text_frame.text.strip()
+                if notes_text:
+                    slide_parts.append(f"\n> **Notes**: {notes_text}")
+
+            # Group parts by content type and separate groups with blank lines
+            # to prevent HTML blocks from consuming adjacent list items
+            groups: list[list[str]] = []
+            for part in slide_parts:
+                kind = _classify_part(part)
+                if groups and _classify_part(groups[-1][0]) == kind:
+                    groups[-1].append(part)
+                else:
+                    groups.append([part])
+
+            slides_md.append("\n\n".join("\n".join(g) for g in groups))
+
+        md_text = "\n\n---\n\n".join(slides_md)
+
+        return ConvertResult(
+            markdown=md_text,
+            images=images,
+            metadata={
+                "format": "PPTX",
+                "slide_count": len(prs.slides),
+            },
+        )
+
+    def _process_shape(
+        self, shape, slide_parts, images, image_counter, slide_width, slide_part
+    ):
+        """Recursively process a shape: Picture, GroupShape, Chart, Table, or TextFrame."""
+        Picture = self._Picture
+        MSO_SHAPE_TYPE = self._MSO_SHAPE_TYPE
+
+        # 1. Picture
+        if isinstance(shape, Picture):
+            try:
+                img = shape.image
+                image_counter[0] += 1
+                filename = f"image{image_counter[0]}.{img.ext}"
+                rel_path = f"images/{filename}"
+                images[rel_path] = img.blob
+                if shape.width and slide_width:
+                    pct = min(round(shape.width / slide_width * 100), 100)
+                    slide_parts.append(f'<img src="images/{filename}" width="{pct}%">')
+                else:
+                    slide_parts.append(f'<img src="images/{filename}">')
+            except (ValueError, AttributeError):
+                pass
+            return
+
+        # 2. GroupShape - recurse into child shapes
+        if MSO_SHAPE_TYPE and shape.shape_type == MSO_SHAPE_TYPE.GROUP:
+            try:
+                for child in shape.shapes:
+                    self._process_shape(
+                        child,
+                        slide_parts,
+                        images,
+                        image_counter,
+                        slide_width,
+                        slide_part,
+                    )
+            except AttributeError:
+                pass
+            return
+
+        # 3. Chart
+        if shape.has_chart:
+            slide_parts.append(self._chart_to_html(shape.chart))
+            return
+
+        # 4. Table
+        if shape.has_table:
+            slide_parts.append(
+                self._table_to_html(shape.table, slide_part, image_counter, images)
+            )
+            return
+
+        # 5. TextFrame
+        if shape.has_text_frame:
+            for paragraph in shape.text_frame.paragraphs:
+                # Check for math elements first
+                para_xml = paragraph._p
+                if _paragraph_has_math(para_xml):
+                    math_items = _extract_math_from_paragraph(para_xml)
+                    for latex in math_items:
+                        slide_parts.append(f"$$\n{latex}\n$$")
+                    continue
+
+                parts = []
+                for run in paragraph.runs:
+                    t = run.text
+                    if not t:
+                        continue
+                    try:
+                        url = run.hyperlink.address
+                    except Exception:
+                        url = None
+
+                    bold = bool(run.font.bold)
+                    italic = bool(run.font.italic)
+                    underline = bool(run.font.underline) and not url
+                    strikethrough = _pptx_run_strike(run)
+                    script = _pptx_run_script(run)
+
+                    if "\n" in t:
+                        segments = t.split("\n")
+                        for j, seg in enumerate(segments):
+                            if seg:
+                                parts.append(
+                                    _format_run_segment(
+                                        seg,
+                                        bold,
+                                        italic,
+                                        underline,
+                                        strikethrough,
+                                        script,
+                                        url,
+                                    )
+                                )
+                            if j < len(segments) - 1:
+                                parts.append("<br>\n")
+                    else:
+                        parts.append(
+                            _format_run_segment(
+                                t, bold, italic, underline, strikethrough, script, url
+                            )
+                        )
+                text = "".join(parts).strip()
+                if not text:
+                    continue
+                level = paragraph.level
+                indent = "  " * level
+                # List item continuation lines need indentation; revert <br>\n → <br>
+                text = text.replace("<br>\n", "<br>")
+                slide_parts.append(f"{indent}- {text}")
+
+    def _chart_to_html(self, chart) -> str:
+        """Extract chart data as an HTML table."""
+        try:
+            chart_type_val = chart.chart_type.value if chart.chart_type else 0
+            chart_type_name = _CHART_TYPE_NAMES.get(chart_type_val, "Chart")
+        except Exception:
+            chart_type_name = "Chart"
+
+        try:
+            title_text = ""
+            try:
+                title_text = chart.chart_title.text_frame.text.strip()
+            except Exception:
+                pass
+
+            # Extract axis info from OOXML
+            chart_root = chart._element
+
+            cat_ax_title = ""
+            for ax_tag in (f"{_C}catAx", f"{_C}dateAx"):
+                ax_el = chart_root.find(f".//{ax_tag}")
+                if ax_el is not None:
+                    t_el = ax_el.find(f"{_C}title")
+                    if t_el is not None:
+                        texts = t_el.findall(f".//{_A}t")
+                        cat_ax_title = "".join(el.text or "" for el in texts).strip()
+                    break
+
+            has_date_ax = chart_root.find(f".//{_C}dateAx") is not None
+
+            val_ax_title = ""
+            val_ax_el = chart_root.find(f".//{_C}valAx")
+            if val_ax_el is not None:
+                t_el = val_ax_el.find(f"{_C}title")
+                if t_el is not None:
+                    texts = t_el.findall(f".//{_A}t")
+                    val_ax_title = "".join(el.text or "" for el in texts).strip()
+
+            plot = chart.plots[0]
+            categories = list(plot.categories) if plot.categories else []
+            series_list = list(plot.series)
+
+            if not series_list:
+                return f"[{chart_type_name}]"
+
+            # Convert Excel date serials to YYYY-MM-DD for date axes
+            if has_date_ax and categories:
+                from datetime import datetime, timedelta
+
+                converted = []
+                for c in categories:
+                    try:
+                        dt = datetime(1899, 12, 30) + timedelta(days=float(c))
+                        converted.append(dt.strftime("%Y-%m-%d"))
+                    except (ValueError, TypeError):
+                        converted.append(str(c) if c is not None else "")
+                categories = converted
+
+            # Collect series names and values
+            series_names = []
+            series_values = []
+            for idx, series in enumerate(series_list):
+                try:
+                    name = (
+                        (series.tx.text if series.tx else "")
+                        or val_ax_title
+                        or f"Series{idx+1}"
+                    )
+                except Exception:
+                    name = val_ax_title or f"Series{idx+1}"
+                series_names.append(name)
+                try:
+                    vals = [
+                        str(round(v, 4)) if v is not None else "" for v in series.values
+                    ]
+                except Exception:
+                    vals = []
+                series_values.append(vals)
+
+            # Build HTML table
+            html_parts = ["<table>"]
+            if title_text:
+                html_parts.append(f"<caption>{title_text}</caption>")
+
+            has_header = cat_ax_title or any(name for name in series_names)
+            if has_header:
+                html_parts.append("<thead><tr>")
+                html_parts.append(f"<th>{cat_ax_title}</th>")
+                for name in series_names:
+                    html_parts.append(f"<th>{name}</th>")
+                html_parts.append("</tr></thead>")
+
+            html_parts.append("<tbody>")
+            if categories:
+                for i, cat in enumerate(categories):
+                    html_parts.append(f"<tr><td>{cat}</td>")
+                    for vals in series_values:
+                        v = vals[i] if i < len(vals) else ""
+                        html_parts.append(f"<td>{v}</td>")
+                    html_parts.append("</tr>")
+            else:
+                max_len = max((len(v) for v in series_values), default=0)
+                for i in range(max_len):
+                    html_parts.append(f"<tr><td>Item{i+1}</td>")
+                    for vals in series_values:
+                        v = vals[i] if i < len(vals) else ""
+                        html_parts.append(f"<td>{v}</td>")
+                    html_parts.append("</tr>")
+            html_parts.append("</tbody></table>")
+
+            return "\n".join(html_parts)
+        except Exception:
+            return f"[{chart_type_name}]"
+
+    @staticmethod
+    def _table_to_html(
+        table, slide_part, image_counter_list: list, images: dict
+    ) -> str:
+        """Convert a PPTX table to an HTML table, handling merged cells and cell background images."""
+        _BLIP_NS = "{http://schemas.openxmlformats.org/drawingml/2006/main}"
+        _REL_NS = (
+            "{http://schemas.openxmlformats.org/officeDocument/2006/relationships}"
+        )
+
+        visited: set[tuple[int, int]] = set()
+        html_parts = ["<table>"]
+
+        for i, row in enumerate(table.rows):
+            html_parts.append("<tr>")
+            for j, cell in enumerate(row.cells):
+                if (i, j) in visited:
+                    continue
+                tag = "th" if i == 0 else "td"
+                attrs = ""
+                if cell.is_merge_origin:
+                    rs = cell.span_height
+                    cs = cell.span_width
+                    if cs > 1:
+                        attrs += f' colspan="{cs}"'
+                    if rs > 1:
+                        attrs += f' rowspan="{rs}"'
+                    for di in range(rs):
+                        for dj in range(cs):
+                            if (di, dj) != (0, 0):
+                                visited.add((i + di, j + dj))
+
+                content_parts = []
+
+                # Extract cell background blip images
+                blips = cell._tc.findall(f".//{_BLIP_NS}blip")
+                for blip in blips:
+                    r_embed = blip.get(f"{_REL_NS}embed")
+                    if r_embed:
+                        try:
+                            image_part = slide_part.related_parts[r_embed]
+                            image_counter_list[0] += 1
+                            ext = image_part.content_type.split("/")[-1]
+                            filename = f"image{image_counter_list[0]}.{ext}"
+                            rel_path = f"images/{filename}"
+                            images[rel_path] = image_part.blob
+                            content_parts.append(
+                                f'<img src="images/{filename}" width="100%">'
+                            )
+                        except (KeyError, AttributeError):
+                            pass
+
+                cell_text_parts = []
+                for para in cell.text_frame.paragraphs:
+                    # Check for math elements first
+                    para_xml = para._p
+                    if _paragraph_has_math(para_xml):
+                        math_items = _extract_math_from_paragraph(para_xml)
+                        for latex in math_items:
+                            cell_text_parts.append(f"${latex}$")
+                        continue
+
+                    run_parts = []
+                    for run in para.runs:
+                        t = run.text or ""
+                        if not t:
+                            continue
+                        try:
+                            url = run.hyperlink.address
+                        except Exception:
+                            url = None
+
+                        bold = bool(run.font.bold)
+                        italic = bool(run.font.italic)
+                        underline = bool(run.font.underline) and not url
+                        strikethrough = bool(run.font.strike)
+                        script = _pptx_run_script(run)
+
+                        if bold:
+                            t = f"<b>{t}</b>"
+                        if italic:
+                            t = f"<i>{t}</i>"
+                        if underline:
+                            t = f"<u>{t}</u>"
+                        if strikethrough:
+                            t = f"<del>{t}</del>"
+                        if script == "super":
+                            t = f"<sup>{t}</sup>"
+                        elif script == "sub":
+                            t = f"<sub>{t}</sub>"
+
+                        if url:
+                            run_parts.append(f'<a href="{url}">{t}</a>')
+                        else:
+                            run_parts.append(t)
+                    cell_text_parts.append("".join(run_parts))
+                text = "<br>".join(p for p in cell_text_parts if p.strip())
+                if text:
+                    content_parts.append(text)
+                cell_html = "<br>".join(content_parts) if content_parts else ""
+
+                html_parts.append(f"<{tag}{attrs}>{cell_html}</{tag}>")
+            html_parts.append("</tr>")
+
+        html_parts.append("</table>")
+        return "\n".join(html_parts)
diff --git a/paddleocr/_doc2md/converters/xlsx.py b/paddleocr/_doc2md/converters/xlsx.py
new file mode 100644
index 0000000..9b917bc
--- /dev/null
+++ b/paddleocr/_doc2md/converters/xlsx.py
@@ -0,0 +1,339 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from io import BytesIO
+from pathlib import Path
+from typing import Optional
+
+from ..base import BaseConverter, ConvertResult
+from ..math import (
+    extract_math_from_paragraph as _extract_math_from_paragraph,
+    paragraph_has_math as _paragraph_has_math,
+)
+from ..registry import default_registry
+
+# DrawingML main namespace
+_A = "{http://schemas.openxmlformats.org/drawingml/2006/main}"
+# Markup Compatibility namespace
+_MC = "{http://schemas.openxmlformats.org/markup-compatibility/2006}"
+# OPC relationship type for drawings
+_REL_DRAWING = (
+    "http://schemas.openxmlformats.org/officeDocument/2006/relationships/drawing"
+)
+_REL_NS = "http://schemas.openxmlformats.org/package/2006/relationships"
+
+
+def _extract_drawing_math(zf, sheet_index: int) -> list:
+    """Extract LaTeX formulas from drawing layer of an xlsx sheet.
+
+    Args:
+        zf: An already-opened zipfile.ZipFile object.
+        sheet_index: Zero-based sheet index.
+    """
+    from lxml import etree
+
+    results = []
+    rels_path = f"xl/worksheets/_rels/sheet{sheet_index + 1}.xml.rels"
+
+    # Read rels (file may not exist if sheet has no drawing)
+    try:
+        rels_data = zf.read(rels_path)
+    except KeyError:
+        return results
+
+    # Find drawing relationship targets
+    rels_root = etree.fromstring(rels_data)
+    drawing_targets = []
+    for rel in rels_root.findall(f"{{{_REL_NS}}}Relationship"):
+        if rel.get("Type") == _REL_DRAWING:
+            target = rel.get("Target", "")
+            # "../drawings/drawingX.xml" → "xl/drawings/drawingX.xml"
+            if target.startswith("../"):
+                target = "xl/" + target[3:]
+            elif not target.startswith("xl/"):
+                target = "xl/worksheets/" + target
+            drawing_targets.append(target)
+
+    for drawing_path in drawing_targets:
+        try:
+            drawing_data = zf.read(drawing_path)
+            drawing_root = etree.fromstring(drawing_data)
+        except Exception:
+            continue  # silently skip corrupted or missing drawing
+
+        # Iterate over a:p paragraphs under mc:AlternateContent/mc:Choice
+        for alt in drawing_root.iter(f"{_MC}AlternateContent"):
+            choice = alt.find(f"{_MC}Choice")
+            if choice is None:
+                continue
+            for para in choice.iter(f"{_A}p"):
+                if _paragraph_has_math(para):
+                    results.extend(_extract_math_from_paragraph(para))
+
+    return results
+
+
+def _get_sheet_width_emu(ws, openpyxl_mod) -> int:
+    """Return the total column width of the worksheet in EMU. 1 char width ~ 7px, 1px = 9525 EMU."""
+    CHAR_TO_EMU = 7 * 9525
+    total = 0
+    for col_idx in range(1, (ws.max_column or 1) + 1):
+        col_letter = openpyxl_mod.utils.get_column_letter(col_idx)
+        col_dim = ws.column_dimensions.get(col_letter)
+        if col_dim and col_dim.width is not None:
+            total += col_dim.width * CHAR_TO_EMU
+        else:
+            total += 8.43 * CHAR_TO_EMU  # default column width
+    return int(total)
+
+
+def _get_image_cx(anchor) -> Optional[int]:
+    """Return image display width in EMU for OneCellAnchor; return None for TwoCellAnchor."""
+    try:
+        from openpyxl.drawing.spreadsheet_drawing import OneCellAnchor
+
+        if isinstance(anchor, OneCellAnchor):
+            return anchor.ext.cx
+    except (AttributeError, ImportError):
+        pass
+    return None
+
+
+def _find_data_bounds(ws, image_map, max_rows=None):
+    """Return (min_row, max_row, min_col, max_col) 1-based, or None if the sheet is empty."""
+    min_r = min_c = float("inf")
+    max_r = max_c = 0
+
+    # Non-empty cells
+    for cell in ws._cells.values():
+        if cell.value is not None:
+            r, c = cell.row, cell.column
+            if max_rows is not None and r > max_rows:
+                continue
+            min_r = min(min_r, r)
+            max_r = max(max_r, r)
+            min_c = min(min_c, c)
+            max_c = max(max_c, c)
+
+    # Merged cell ranges
+    for mr in ws.merged_cells.ranges:
+        r1, r2 = mr.min_row, mr.max_row
+        if max_rows is not None:
+            r2 = min(r2, max_rows)
+        if r1 > r2:
+            continue
+        min_r = min(min_r, r1)
+        max_r = max(max_r, r2)
+        min_c = min(min_c, mr.min_col)
+        max_c = max(max_c, mr.max_col)
+
+    # Image anchors (0-based -> 1-based)
+    for img_r0, img_c0 in image_map:
+        r, c = img_r0 + 1, img_c0 + 1
+        if max_rows is not None and r > max_rows:
+            continue
+        min_r = min(min_r, r)
+        max_r = max(max_r, r)
+        min_c = min(min_c, c)
+        max_c = max(max_c, c)
+
+    if max_r == 0:
+        return None
+    return (int(min_r), int(max_r), int(min_c), int(max_c))
+
+
+@default_registry.register
+class XlsxConverter(BaseConverter):
+    supported_extensions = ["xlsx"]
+    supported_mimetypes = [
+        "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
+    ]
+
+    def convert_file(self, file_path: Path, **kwargs) -> ConvertResult:
+        try:
+            import openpyxl
+            from openpyxl.drawing.spreadsheet_drawing import (
+                OneCellAnchor,
+                TwoCellAnchor,
+            )
+        except ImportError:
+            raise RuntimeError(
+                "XLSX conversion requires openpyxl: pip install paddleocr[doc2md]"
+            )
+
+        sheet_name: Optional[str] = kwargs.get("sheet_name", None)
+        max_rows: Optional[int] = kwargs.get("max_rows", None)
+        extract_drawings: bool = kwargs.get("extract_drawings", True)
+
+        import zipfile
+
+        # read_only=False is required to access merged_cells
+        wb = openpyxl.load_workbook(str(file_path), read_only=False, data_only=True)
+        sheets_md = []
+        images: dict = {}
+        image_counter = 0
+
+        target_sheets = [sheet_name] if sheet_name else wb.sheetnames
+
+        with zipfile.ZipFile(str(file_path), "r") as _zf:
+            for sname in target_sheets:
+                if sname not in wb.sheetnames:
+                    continue
+                ws = wb[sname]
+
+                # Total sheet width in EMU, used for image percentage calculation
+                sheet_width_emu = _get_sheet_width_emu(ws, openpyxl)
+
+                # Floating image map: (0-based row, 0-based col) -> [Image, ...]
+                image_map: dict = {}
+                for img in getattr(ws, "_images", []):
+                    anchor = img.anchor
+                    if isinstance(anchor, (OneCellAnchor, TwoCellAnchor)):
+                        r, c = anchor._from.row, anchor._from.col
+                        image_map.setdefault((r, c), []).append(img)
+                    elif isinstance(anchor, str):
+                        try:
+                            from openpyxl.utils import coordinate_to_tuple
+
+                            r, c = coordinate_to_tuple(anchor)
+                            image_map.setdefault((r - 1, c - 1), []).append(img)
+                        except Exception:
+                            pass
+
+                # Merged cell map: (row, col) -> MergedCellRange
+                merge_map = {}
+                for mr in ws.merged_cells.ranges:
+                    for r in range(mr.min_row, mr.max_row + 1):
+                        for c in range(mr.min_col, mr.max_col + 1):
+                            merge_map[(r, c)] = mr
+
+                # Trim surrounding empty rows/columns
+                bounds = _find_data_bounds(ws, image_map, max_rows)
+                if bounds is None:
+                    continue
+                data_min_row, data_max_row, data_min_col, data_max_col = bounds
+
+                html_parts = ["<table>"]
+                for row_idx in range(data_min_row, data_max_row + 1):
+                    html_parts.append("<tr>")
+                    for col_idx in range(data_min_col, data_max_col + 1):
+                        cell = ws.cell(row_idx, col_idx)
+                        mr = merge_map.get((row_idx, col_idx))
+                        # Skip non-origin cells in a merged range
+                        if mr and (row_idx, col_idx) != (mr.min_row, mr.min_col):
+                            continue
+                        tag = "th" if row_idx == data_min_row else "td"
+                        attrs = ""
+                        if mr:
+                            cs = (
+                                min(mr.max_col, data_max_col)
+                                - max(mr.min_col, data_min_col)
+                                + 1
+                            )
+                            rs = (
+                                min(mr.max_row, data_max_row)
+                                - max(mr.min_row, data_min_row)
+                                + 1
+                            )
+                            if cs > 1:
+                                attrs += f' colspan="{cs}"'
+                            if rs > 1:
+                                attrs += f' rowspan="{rs}"'
+
+                        # Cell text
+                        value = cell.value
+                        text = str(value) if value is not None else ""
+                        # Cell-level font formatting (bold/italic/underline/strikethrough)
+                        if text:
+                            try:
+                                font = cell.font
+                                if font.bold:
+                                    text = f"<b>{text}</b>"
+                                if font.italic:
+                                    text = f"<i>{text}</i>"
+                                if font.underline:
+                                    text = f"<u>{text}</u>"
+                                if font.strike:
+                                    text = f"<del>{text}</del>"
+                                vert_align = font.vertAlign
+                                if vert_align == "superscript":
+                                    text = f"<sup>{text}</sup>"
+                                elif vert_align == "subscript":
+                                    text = f"<sub>{text}</sub>"
+                            except Exception:
+                                pass
+                        # Hyperlink wrapping
+                        if text:
+                            try:
+                                hl = cell.hyperlink
+                                if hl and hl.target:
+                                    text = f'<a href="{hl.target}">{text}</a>'
+                            except Exception:
+                                pass
+
+                        # Floating images
+                        cell_images = image_map.get((row_idx - 1, col_idx - 1), [])
+                        img_html = ""
+                        for img_obj in cell_images:
+                            image_counter += 1
+                            ext = (img_obj.format or "png").lower()
+                            filename = f"image{image_counter}.{ext}"
+                            rel_path = f"images/{filename}"
+                            try:
+                                ref = img_obj.ref
+                                if isinstance(ref, BytesIO):
+                                    ref.seek(0)
+                                    images[rel_path] = ref.read()
+                                else:
+                                    images[rel_path] = img_obj._data()
+                                cx_emu = _get_image_cx(img_obj.anchor)
+                                if cx_emu and sheet_width_emu:
+                                    pct = min(
+                                        round(cx_emu / sheet_width_emu * 100), 100
+                                    )
+                                    img_html += (
+                                        f'<img src="images/{filename}" width="{pct}%">'
+                                    )
+                                else:
+                                    img_html += f'<img src="images/{filename}">'
+                            except Exception:
+                                pass
+
+                        cell_content = img_html + text if img_html else text
+                        html_parts.append(f"<{tag}{attrs}>{cell_content}</{tag}>")
+                    html_parts.append("</tr>")
+                html_parts.append("</table>")
+
+                table_html = "\n".join(html_parts)
+
+                sheet_parts = [f"## {sname}\n\n{table_html}"]
+                if extract_drawings:
+                    sheet_idx = list(wb.sheetnames).index(sname)
+                    for latex in _extract_drawing_math(_zf, sheet_idx):
+                        sheet_parts.append(f"\n$$\n{latex}\n$$")
+                sheets_md.append("\n".join(sheet_parts))
+
+        sheet_names = list(wb.sheetnames)
+        wb.close()
+
+        md_text = "\n\n".join(sheets_md)
+
+        return ConvertResult(
+            markdown=md_text,
+            images=images,
+            metadata={
+                "format": "XLSX",
+                "sheet_count": len(sheet_names),
+                "sheet_names": sheet_names,
+            },
+        )
diff --git a/paddleocr/_doc2md/core.py b/paddleocr/_doc2md/core.py
new file mode 100644
index 0000000..f0ff2b0
--- /dev/null
+++ b/paddleocr/_doc2md/core.py
@@ -0,0 +1,76 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from pathlib import Path
+from typing import Union, Optional
+
+from .base import ConvertResult
+from .registry import default_registry
+
+# Trigger registration of all built-in converters
+from . import converters  # noqa: F401
+
+
+def convert(
+    source: Union[str, Path],
+    *,
+    output: Optional[Union[str, Path]] = None,
+    **kwargs,
+) -> ConvertResult:
+    """
+    Convert an office document to Markdown.
+
+    Args:
+        source: Path to the source file.
+        output: Optional output file path. If provided, Markdown is written there.
+        **kwargs: Extra arguments forwarded to the specific converter.
+
+    Returns:
+        ConvertResult object.
+
+    Examples:
+        >>> from paddleocr import doc2md_convert
+        >>> result = doc2md_convert("report.docx")
+        >>> print(result.markdown)
+    """
+    file_path = Path(source)
+
+    if not file_path.exists():
+        raise FileNotFoundError(f"File not found: {file_path}")
+
+    converter = default_registry.get_converter(file_path)
+
+    try:
+        result = converter.convert_file(file_path, **kwargs)
+    except Exception as e:
+        if isinstance(e, (FileNotFoundError, ValueError, RuntimeError)):
+            raise
+        raise RuntimeError(f"Failed to convert {file_path.name}: {e}") from e
+
+    if output:
+        output_path = Path(output)
+        output_path.parent.mkdir(parents=True, exist_ok=True)
+        output_path.write_text(result.markdown, encoding="utf-8")
+        if result.images:
+            images_dir = output_path.parent / "images"
+            images_dir.mkdir(exist_ok=True)
+            for rel_path, img_bytes in result.images.items():
+                img_file = output_path.parent / rel_path
+                img_file.write_bytes(img_bytes)
+
+    return result
+
+
+def supported_formats() -> list[str]:
+    """Return a list of supported file extensions."""
+    return default_registry.supported_extensions()
diff --git a/paddleocr/_doc2md/math/__init__.py b/paddleocr/_doc2md/math/__init__.py
new file mode 100644
index 0000000..4fd73ec
--- /dev/null
+++ b/paddleocr/_doc2md/math/__init__.py
@@ -0,0 +1,84 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from .omml import OMML_NS, oMath2Latex
+
+# XML namespace constants shared by converters for DrawingML math extraction
+_M = OMML_NS  # already includes braces: "{http://...}"
+_A14 = "{http://schemas.microsoft.com/office/drawing/2010/main}"
+
+
+def convert_omath(omath_element) -> str:
+    """Convert an m:oMath lxml element to LaTeX string. Returns empty string on failure."""
+    try:
+        return str(oMath2Latex(omath_element)).strip()
+    except Exception:
+        return ""
+
+
+def paragraph_has_math(para_element) -> bool:
+    """Check if an XML element contains OMML math (a14:m or m:oMath)."""
+    return (
+        para_element.find(f".//{_A14}m") is not None
+        or para_element.find(f".//{_M}oMath") is not None
+    )
+
+
+def extract_math_from_paragraph(para_element) -> list:
+    """Extract LaTeX strings from math elements in a DrawingML paragraph XML element.
+
+    Handles three nesting patterns:
+    1. a14:m → m:oMath (or m:oMathPara → m:oMath)
+    2. Direct m:oMathPara → m:oMath (not wrapped in a14:m)
+    3. Direct m:oMath (not inside a14:m or m:oMathPara)
+    """
+    results = []
+    # a14:m wraps m:oMathPara or m:oMath
+    for a14m in para_element.findall(f".//{_A14}m"):
+        found_omath = False
+        for omath in a14m.findall(f".//{_M}oMath"):
+            latex = convert_omath(omath)
+            if latex:
+                results.append(latex)
+                found_omath = True
+        # No oMath inside this a14:m? Try the a14:m element itself
+        if not found_omath:
+            latex = convert_omath(a14m)
+            if latex:
+                results.append(latex)
+    # Direct m:oMathPara / m:oMath not wrapped in a14:m
+    for omath_para in para_element.findall(f".//{_M}oMathPara"):
+        parent = omath_para.getparent()
+        if parent is not None and parent.tag == f"{_A14}m":
+            continue  # already handled above (oMathPara is inside a14:m)
+        for omath in omath_para.findall(f"{_M}oMath"):
+            latex = convert_omath(omath)
+            if latex:
+                results.append(latex)
+    for omath in para_element.findall(f".//{_M}oMath"):
+        parent = omath.getparent()
+        if parent is not None and parent.tag in (f"{_A14}m", f"{_M}oMathPara"):
+            continue  # already handled
+        latex = convert_omath(omath)
+        if latex:
+            results.append(latex)
+    return results
+
+
+__all__ = [
+    "oMath2Latex",
+    "OMML_NS",
+    "convert_omath",
+    "paragraph_has_math",
+    "extract_math_from_paragraph",
+]
diff --git a/paddleocr/_doc2md/math/latex_dict.py b/paddleocr/_doc2md/math/latex_dict.py
new file mode 100644
index 0000000..cd65132
--- /dev/null
+++ b/paddleocr/_doc2md/math/latex_dict.py
@@ -0,0 +1,329 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+LaTeX symbol dictionaries for OMML conversion.
+
+Adapted from https://github.com/xiilei/dwml/blob/master/dwml/latex_dict.py
+via MinerU 3.0.0 (MIT License)
+"""
+
+CHARS = ("{", "}", "_", "^", "#", "&", "$", "%")
+
+BLANK = ""
+BACKSLASH = "\\"
+ALN = "&"
+
+CHR = {
+    # Unicode : Latex Math Symbols
+    # Top accents
+    "\u0300": "\\grave{{{0}}}",
+    "\u0301": "\\acute{{{0}}}",
+    "\u0302": "\\hat{{{0}}}",
+    "\u0303": "\\tilde{{{0}}}",
+    "\u0304": "\\bar{{{0}}}",
+    "\u0305": "\\overbar{{{0}}}",
+    "\u0306": "\\breve{{{0}}}",
+    "\u0307": "\\dot{{{0}}}",
+    "\u0308": "\\ddot{{{0}}}",
+    "\u0309": "\\ovhook{{{0}}}",
+    "\u030a": "\\ocirc{{{0}}}",
+    "\u030c": "\\check{{{0}}}",
+    "\u0310": "\\candra{{{0}}}",
+    "\u0312": "\\oturnedcomma{{{0}}}",
+    "\u0315": "\\ocommatopright{{{0}}}",
+    "\u031a": "\\droang{{{0}}}",
+    "\u0338": "\\not{{{0}}}",
+    "\u20d0": "\\leftharpoonaccent{{{0}}}",
+    "\u20d1": "\\rightharpoonaccent{{{0}}}",
+    "\u20d2": "\\vertoverlay{{{0}}}",
+    "\u20d6": "\\overleftarrow{{{0}}}",
+    "\u20d7": "\\vec{{{0}}}",
+    "\u20db": "\\dddot{{{0}}}",
+    "\u20dc": "\\ddddot{{{0}}}",
+    "\u20e1": "\\overleftrightarrow{{{0}}}",
+    "\u20e7": "\\annuity{{{0}}}",
+    "\u20e9": "\\widebridgeabove{{{0}}}",
+    "\u20f0": "\\asteraccent{{{0}}}",
+    # Bottom accents
+    "\u0330": "\\wideutilde{{{0}}}",
+    "\u0331": "\\underbar{{{0}}}",
+    "\u20e8": "\\threeunderdot{{{0}}}",
+    "\u20ec": "\\underrightharpoondown{{{0}}}",
+    "\u20ed": "\\underleftharpoondown{{{0}}}",
+    "\u20ee": "\\underleftarrow{{{0}}}",
+    "\u20ef": "\\underrightarrow{{{0}}}",
+    # Over | group
+    "\u23b4": "\\overbracket{{{0}}}",
+    "\u23dc": "\\overparen{{{0}}}",
+    "\u23de": "\\overbrace{{{0}}}",
+    # Under| group
+    "\u23b5": "\\underbracket{{{0}}}",
+    "\u23dd": "\\underparen{{{0}}}",
+    "\u23df": "\\underbrace{{{0}}}",
+}
+
+CHR_BO = {
+    # Big operators,
+    "\u2140": "\\Bbbsum",
+    "\u220f": "\\prod",
+    "\u2210": "\\coprod",
+    "\u2211": "\\sum",
+    "\u222b": "\\int",
+    "\u222c": "\\iint",
+    "\u222d": "\\iiint",
+    "\u222e": "\\oint",
+    "\u222f": "\\oiint",
+    "\u2230": "\\oiiint",
+    "\u22c0": "\\bigwedge",
+    "\u22c1": "\\bigvee",
+    "\u22c2": "\\bigcap",
+    "\u22c3": "\\bigcup",
+    "\u2a00": "\\bigodot",
+    "\u2a01": "\\bigoplus",
+    "\u2a02": "\\bigotimes",
+}
+
+T = {
+    # Whitespace characters
+    " ": " ",  # NON-BREAKING SPACE (U+00A0) — pylatexenc maps this to "~" (text-mode),
+    # which escape_latex would mangle to "\~" (invalid in math mode);
+    # use a plain space instead.
+    # Greek letters
+    "\U0001d6fc": "\\alpha ",
+    "\U0001d6fd": "\\beta ",
+    "\U0001d6fe": "\\gamma ",
+    "\U0001d6ff": "\\delta ",
+    "\U0001d700": "\\epsilon ",
+    "\U0001d701": "\\zeta ",
+    "\U0001d702": "\\eta ",
+    "\U0001d703": "\\theta ",
+    "\U0001d704": "\\iota ",
+    "\U0001d705": "\\kappa ",
+    "\U0001d706": "\\lambda ",
+    "\U0001d707": "\\mu ",
+    "\U0001d708": "\\nu ",
+    "\U0001d709": "\\xi ",
+    "\U0001d70a": "\\omicron ",
+    "\U0001d70b": "\\pi ",
+    "\U0001d70c": "\\rho ",
+    "\U0001d70d": "\\varsigma ",
+    "\U0001d70e": "\\sigma ",
+    "\U0001d70f": "\\tau ",
+    "\U0001d710": "\\upsilon ",
+    "\U0001d711": "\\phi ",
+    "\U0001d712": "\\chi ",
+    "\U0001d713": "\\psi ",
+    "\U0001d714": "\\omega ",
+    "\U0001d715": "\\partial ",
+    "\U0001d716": "\\varepsilon ",
+    "\U0001d717": "\\vartheta ",
+    "\U0001d718": "\\varkappa ",
+    "\U0001d719": "\\varphi ",
+    "\U0001d71a": "\\varrho ",
+    "\U0001d71b": "\\varpi ",
+    # Relation symbols
+    "\u2190": "\\leftarrow ",
+    "\u2191": "\\uparrow ",
+    "\u2192": "\\rightarrow ",
+    "\u2193": "\\downarrow ",
+    "\u2194": "\\leftrightarrow ",
+    "\u2195": "\\updownarrow ",
+    "\u2196": "\\nwarrow ",
+    "\u2197": "\\nearrow ",
+    "\u2198": "\\searrow ",
+    "\u2199": "\\swarrow ",
+    "\u2026": "\\ldots ",  # HORIZONTAL ELLIPSIS (…)
+    "\u22ee": "\\vdots ",
+    "\u22ef": "\\cdots ",
+    "\u22f0": "\\adots ",
+    "\u22f1": "\\ddots ",
+    "\u2260": "\\ne ",
+    "\u2264": "\\leq ",
+    "\u2265": "\\geq ",
+    "\u2266": "\\leqq ",
+    "\u2267": "\\geqq ",
+    "\u2268": "\\lneqq ",
+    "\u2269": "\\gneqq ",
+    "\u226a": "\\ll ",
+    "\u226b": "\\gg ",
+    "\u2208": "\\in ",
+    "\u2209": "\\notin ",
+    "\u220b": "\\ni ",
+    "\u220c": "\\nni ",
+    # Ordinary symbols
+    "\u221e": "\\infty ",
+    # Binary relations
+    "\u00b1": "\\pm ",
+    "\u2213": "\\mp ",
+    # Characters whose pylatexenc text-mode mappings are invalid in math environments
+    "\u00f0": "\\eth ",
+    "\u0131": "\\imath ",
+    "\u2127": "\\mho ",
+    "\u212e": "e",
+    "\u00c5": "\\mathring{A} ",
+    # Multiplication/division operators
+    "\u00b7": "\\cdot ",
+    "\u22c5": "\\cdot ",
+    "\u2219": "\\bullet ",
+    "\u00d7": "\\times ",
+    "\u00f7": "\\div ",
+    "\u2212": "-",
+    # Degree / prime
+    "\u00b0": "\\circ ",
+    "\u2032": "'",
+    "\u2033": "''",
+    # Superscript digits
+    "\u00b2": "2",
+    "\u00b3": "3",
+    "\u00b9": "1",
+    # Big operators as plain text characters
+    "\u222f": "\\oiint ",
+    "\u2230": "\\oiiint ",
+    "\u2231": "\u2231",
+    "\u2232": "\u2232",
+    "\u2233": "\u2233",
+    "\u2a00": "\\bigodot ",
+    "\u2a01": "\\bigoplus ",
+    "\u2a02": "\\bigotimes ",
+    "\u2a03": "\u2a03",
+    "\u2a04": "\u2a04",
+    # Wave arrows
+    "\u219c": "\u219c",
+    "\u219d": "\u219d",
+    # Italic, Latin, uppercase
+    "\U0001d434": "A",
+    "\U0001d435": "B",
+    "\U0001d436": "C",
+    "\U0001d437": "D",
+    "\U0001d438": "E",
+    "\U0001d439": "F",
+    "\U0001d43a": "G",
+    "\U0001d43b": "H",
+    "\U0001d43c": "I",
+    "\U0001d43d": "J",
+    "\U0001d43e": "K",
+    "\U0001d43f": "L",
+    "\U0001d440": "M",
+    "\U0001d441": "N",
+    "\U0001d442": "O",
+    "\U0001d443": "P",
+    "\U0001d444": "Q",
+    "\U0001d445": "R",
+    "\U0001d446": "S",
+    "\U0001d447": "T",
+    "\U0001d448": "U",
+    "\U0001d449": "V",
+    "\U0001d44a": "W",
+    "\U0001d44b": "X",
+    "\U0001d44c": "Y",
+    "\U0001d44d": "Z",
+    # Italic, Latin, lowercase
+    "\U0001d44e": "a",
+    "\U0001d44f": "b",
+    "\U0001d450": "c",
+    "\U0001d451": "d",
+    "\U0001d452": "e",
+    "\U0001d453": "f",
+    "\U0001d454": "g",
+    "\U0001d456": "i",
+    "\U0001d457": "j",
+    "\U0001d458": "k",
+    "\U0001d459": "l",
+    "\U0001d45a": "m",
+    "\U0001d45b": "n",
+    "\U0001d45c": "o",
+    "\U0001d45d": "p",
+    "\U0001d45e": "q",
+    "\U0001d45f": "r",
+    "\U0001d460": "s",
+    "\U0001d461": "t",
+    "\U0001d462": "u",
+    "\U0001d463": "v",
+    "\U0001d464": "w",
+    "\U0001d465": "x",
+    "\U0001d466": "y",
+    "\U0001d467": "z",
+}
+
+FUNC = {
+    "sin": "\\sin({fe})",
+    "cos": "\\cos({fe})",
+    "tan": "\\tan({fe})",
+    "arcsin": "\\arcsin({fe})",
+    "arccos": "\\arccos({fe})",
+    "arctan": "\\arctan({fe})",
+    "arccot": "\\arccot({fe})",
+    "sinh": "\\sinh({fe})",
+    "cosh": "\\cosh({fe})",
+    "tanh": "\\tanh({fe})",
+    "coth": "\\coth({fe})",
+    "sec": "\\sec({fe})",
+    "csc": "\\csc({fe})",
+    "mod": "\\mod {fe}",
+    "max": "\\max({fe})",
+    "min": "\\min({fe})",
+}
+
+FUNC_PLACE = "{fe}"
+
+BRK = "\\\\"
+
+CHR_DEFAULT = {
+    "ACC_VAL": "\\hat{{{0}}}",
+}
+
+POS = {
+    "top": "\\overline{{{0}}}",  # not sure
+    "bot": "\\underline{{{0}}}",
+}
+
+POS_DEFAULT = {
+    "BAR_VAL": "\\overline{{{0}}}",
+}
+
+SUB = "_{{{0}}}"
+
+SUP = "^{{{0}}}"
+
+F = {
+    "bar": "\\frac{{{num}}}{{{den}}}",
+    "skw": r"^{{{num}}}/_{{{den}}}",
+    "noBar": "\\genfrac{{}}{{}}{{0pt}}{{}}{{{num}}}{{{den}}}",
+    "lin": "{{{num}}}/{{{den}}}",
+}
+F_DEFAULT = "\\frac{{{num}}}{{{den}}}"
+
+D = "\\left{left}{text}\\right{right}"
+
+D_DEFAULT = {
+    "left": "(",
+    "right": ")",
+    "null": ".",
+}
+
+RAD = "\\sqrt[{deg}]{{{text}}}"
+RAD_DEFAULT = "\\sqrt{{{text}}}"
+ARR = "\\begin{{array}}{{c}}{text}\\end{{array}}"
+
+LIM_FUNC = {
+    "lim": "\\lim_{{{lim}}}",
+    "max": "\\max_{{{lim}}}",
+    "min": "\\min_{{{lim}}}",
+}
+
+LIM_TO = ("\\rightarrow", "\\to")
+
+LIM_UPP = "\\overset{{{lim}}}{{{text}}}"
+
+M = "\\begin{{matrix}}{text}\\end{{matrix}}"
diff --git a/paddleocr/_doc2md/math/omml.py b/paddleocr/_doc2md/math/omml.py
new file mode 100644
index 0000000..8fadc1c
--- /dev/null
+++ b/paddleocr/_doc2md/math/omml.py
@@ -0,0 +1,482 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+Office Math Markup Language (OMML) to LaTeX converter.
+
+Adapted from https://github.com/xiilei/dwml/blob/master/dwml/omml.py
+via MinerU 3.0.0 (MIT License)
+"""
+
+import logging
+import re
+
+import lxml.etree as ET
+
+from .latex_dict import (
+    ALN,
+    ARR,
+    BACKSLASH,
+    BLANK,
+    BRK,
+    CHARS,
+    CHR,
+    CHR_BO,
+    CHR_DEFAULT,
+    D_DEFAULT,
+    F_DEFAULT,
+    FUNC,
+    FUNC_PLACE,
+    LIM_FUNC,
+    LIM_TO,
+    LIM_UPP,
+    POS,
+    POS_DEFAULT,
+    RAD,
+    RAD_DEFAULT,
+    SUB,
+    SUP,
+    D,
+    F,
+    M,
+    T,
+)
+
+logger = logging.getLogger(__name__)
+
+OMML_NS = "{http://schemas.openxmlformats.org/officeDocument/2006/math}"
+
+# Mapping from OMML <m:scr> values to LaTeX math font commands.
+SCR_TO_LATEX = {
+    "script": "\\mathscr{{{0}}}",
+    "fraktur": "\\mathfrak{{{0}}}",
+    "double-struck": "\\mathbb{{{0}}}",
+    "sans-serif": "\\mathsf{{{0}}}",
+    "monospace": "\\mathtt{{{0}}}",
+}
+
+
+def escape_latex(strs):
+    last = None
+    new_chr = []
+    strs = strs.replace(r"\\", "\\")
+    for c in strs:
+        if (c in CHARS) and (last != BACKSLASH):
+            new_chr.append(BACKSLASH + c)
+        else:
+            new_chr.append(c)
+        last = c
+    return BLANK.join(new_chr)
+
+
+def get_val(key, default=None, store=CHR):
+    if key is not None:
+        return key if not store else store.get(key, key)
+    else:
+        return default
+
+
+class Tag2Method:
+    def call_method(self, elm, stag=None):
+        getmethod = self.tag2meth.get
+        if stag is None:
+            stag = elm.tag.replace(OMML_NS, "")
+        method = getmethod(stag)
+        if method:
+            return method(self, elm)
+        else:
+            return None
+
+    def process_children_list(self, elm, include=None):
+        """
+        process children of the elm,return iterable
+        """
+        for _e in list(elm):
+            if OMML_NS not in _e.tag:
+                continue
+            stag = _e.tag.replace(OMML_NS, "")
+            if include and (stag not in include):
+                continue
+            t = self.call_method(_e, stag=stag)
+            if t is None:
+                t = self.process_unknow(_e, stag)
+                if t is None:
+                    continue
+            yield (stag, t, _e)
+
+    def process_children_dict(self, elm, include=None):
+        """
+        process children of the elm,return dict
+        """
+        latex_chars = dict()
+        for stag, t, e in self.process_children_list(elm, include):
+            latex_chars[stag] = t
+        return latex_chars
+
+    def process_children(self, elm, include=None):
+        """
+        process children of the elm,return string
+        """
+        return BLANK.join(
+            (
+                t if not isinstance(t, Tag2Method) else str(t)
+                for stag, t, e in self.process_children_list(elm, include)
+            )
+        )
+
+    def process_unknow(self, elm, stag):
+        return None
+
+
+class Pr(Tag2Method):
+    text = ""
+
+    __val_tags = ("chr", "pos", "begChr", "endChr", "type")
+
+    __innerdict = None  # can't use the __dict__
+
+    """ common properties of element"""
+
+    def __init__(self, elm):
+        self.__innerdict = {}
+        self.text = self.process_children(elm)
+
+    def __str__(self):
+        return self.text
+
+    def __getattr__(self, name):
+        return self.__innerdict.get(name, None)
+
+    def do_brk(self, elm):
+        self.__innerdict["brk"] = BRK
+        return BRK
+
+    def do_common(self, elm):
+        stag = elm.tag.replace(OMML_NS, "")
+        if stag in self.__val_tags:
+            t = elm.get(f"{OMML_NS}val")
+            self.__innerdict[stag] = t
+        return None
+
+    tag2meth = {
+        "brk": do_brk,
+        "chr": do_common,
+        "pos": do_common,
+        "begChr": do_common,
+        "endChr": do_common,
+        "type": do_common,
+    }
+
+
+class oMath2Latex(Tag2Method):
+    """
+    Convert oMath element of omml to latex
+    """
+
+    _t_dict = T
+
+    __direct_tags = ("box", "sSub", "sSup", "sSubSup", "num", "den", "deg", "e")
+
+    _encoder = None  # cached UnicodeToLatexEncoder instance (class-level)
+
+    def __init__(self, element):
+        if oMath2Latex._encoder is None:
+            # Delayed import: pylatexenc is an optional dependency
+            try:
+                from pylatexenc.latexencode import UnicodeToLatexEncoder
+            except ImportError:
+                raise RuntimeError(
+                    "pylatexenc is required for math formula conversion. "
+                    "Install it with: pip install pylatexenc"
+                )
+            oMath2Latex._encoder = UnicodeToLatexEncoder(
+                replacement_latex_protection="braces-all",
+                unknown_char_policy="keep",
+                unknown_char_warning=False,
+            )
+        self.u = oMath2Latex._encoder
+        self._latex = self.process_children(element)
+
+    def __str__(self):
+        return self.latex.replace("  ", " ")
+
+    def process_unknow(self, elm, stag):
+        if stag in self.__direct_tags:
+            return self.process_children(elm)
+        elif stag[-2:] == "Pr":
+            return Pr(elm)
+        else:
+            return None
+
+    @property
+    def latex(self):
+        return self._latex
+
+    def do_acc(self, elm):
+        """
+        the accent function
+        """
+        c_dict = self.process_children_dict(elm)
+        latex_s = get_val(
+            c_dict["accPr"].chr, default=CHR_DEFAULT.get("ACC_VAL"), store=CHR
+        )
+        return latex_s.format(c_dict["e"])
+
+    def do_bar(self, elm):
+        """
+        the bar function
+        """
+        c_dict = self.process_children_dict(elm)
+        pr = c_dict["barPr"]
+        latex_s = get_val(pr.pos, default=POS_DEFAULT.get("BAR_VAL"), store=POS)
+        return pr.text + latex_s.format(c_dict["e"])
+
+    def do_d(self, elm):
+        """
+        the delimiter object
+        """
+        c_dict = self.process_children_dict(elm)
+        pr = c_dict["dPr"]
+        null = D_DEFAULT.get("null")
+
+        s_val = get_val(pr.begChr, default=D_DEFAULT.get("left"), store=T)
+        e_val = get_val(pr.endChr, default=D_DEFAULT.get("right"), store=T)
+        delim = pr.text + D.format(
+            left=null if not s_val else escape_latex(s_val),
+            text=c_dict["e"],
+            right=null if not e_val else escape_latex(e_val),
+        )
+        return delim
+
+    def do_sub(self, elm):
+        text = self.process_children(elm)
+        return SUB.format(text)
+
+    def do_sup(self, elm):
+        text = self.process_children(elm)
+        return SUP.format(text)
+
+    def do_f(self, elm):
+        """
+        the fraction object
+        """
+        c_dict = self.process_children_dict(elm)
+        pr = c_dict.get("fPr")
+        if pr is None:
+            logger.debug("Missing fPr element in fraction, using default formatting")
+            latex_s = F_DEFAULT
+            return latex_s.format(
+                num=c_dict.get("num"),
+                den=c_dict.get("den"),
+            )
+        latex_s = get_val(pr.type, default=F_DEFAULT, store=F)
+        return pr.text + latex_s.format(num=c_dict.get("num"), den=c_dict.get("den"))
+
+    def do_func(self, elm):
+        """
+        the Function-Apply object (Examples:sin cos)
+        """
+        c_dict = self.process_children_dict(elm)
+        func_name = c_dict.get("fName")
+        return func_name.replace(FUNC_PLACE, c_dict.get("e"))
+
+    def do_fname(self, elm):
+        """
+        the func name
+        """
+        latex_chars = []
+        for stag, t, e in self.process_children_list(elm):
+            if stag == "r":
+                if FUNC.get(t):
+                    latex_chars.append(FUNC[t])
+                else:
+                    logger.warning(
+                        "Function not supported, will default to text: %s", t
+                    )
+                    if isinstance(t, str):
+                        latex_chars.append(t)
+            elif isinstance(t, str):
+                latex_chars.append(t)
+        t = BLANK.join(latex_chars)
+        return t if FUNC_PLACE in t else t + FUNC_PLACE  # do_func will replace this
+
+    def do_groupchr(self, elm):
+        """
+        the Group-Character object
+        """
+        c_dict = self.process_children_dict(elm)
+        pr = c_dict["groupChrPr"]
+        latex_s = get_val(pr.chr)
+        return pr.text + latex_s.format(c_dict["e"])
+
+    def do_rad(self, elm):
+        """
+        the radical object
+        """
+        c_dict = self.process_children_dict(elm)
+        text = c_dict.get("e")
+        deg_text = c_dict.get("deg")
+        if deg_text:
+            return RAD.format(deg=deg_text, text=text)
+        else:
+            return RAD_DEFAULT.format(text=text)
+
+    def do_eqarr(self, elm):
+        """
+        the Array object.
+        """
+        rows = [t for stag, t, e in self.process_children_list(elm, include=("e",))]
+
+        if len(rows) == 1:
+            row = rows[0]
+            tag_match = re.search(r"\\#\s*\(([^)]*)\)\s*$", row)
+            if tag_match:
+                formula = row[: tag_match.start()].rstrip()
+                tag_content = tag_match.group(1)
+                return f"{formula}\\tag{{{tag_content}}}"
+            return row
+
+        return ARR.format(text=BRK.join(rows))
+
+    def do_limlow(self, elm):
+        """
+        the Lower-Limit object
+        """
+        t_dict = self.process_children_dict(elm, include=("e", "lim"))
+        latex_s = LIM_FUNC.get(t_dict["e"])
+        if not latex_s:
+            raise RuntimeError("Not support lim {}".format(t_dict["e"]))
+        else:
+            return latex_s.format(lim=t_dict.get("lim"))
+
+    def do_limupp(self, elm):
+        """
+        the Upper-Limit object
+        """
+        t_dict = self.process_children_dict(elm, include=("e", "lim"))
+        return LIM_UPP.format(lim=t_dict.get("lim"), text=t_dict.get("e"))
+
+    def do_lim(self, elm):
+        """
+        the lower limit of the limLow object and the upper limit of the limUpp function
+        """
+        return self.process_children(elm).replace(LIM_TO[0], LIM_TO[1])
+
+    def do_m(self, elm):
+        """
+        the Matrix object
+        """
+        rows = []
+        for stag, t, e in self.process_children_list(elm):
+            if stag == "mPr":
+                pass
+            elif stag == "mr":
+                rows.append(t)
+        return M.format(text=BRK.join(rows))
+
+    def do_mr(self, elm):
+        """
+        a single row of the matrix m
+        """
+        return ALN.join(
+            [t for stag, t, e in self.process_children_list(elm, include=("e",))]
+        )
+
+    def do_nary(self, elm):
+        """
+        the n-ary object
+        """
+        res = []
+        bo = ""
+        for stag, t, e in self.process_children_list(elm):
+            if stag == "naryPr":
+                bo = get_val(t.chr, default="\\int", store=CHR_BO)
+            else:
+                res.append(t)
+        return bo + BLANK.join(res)
+
+    def process_unicode(self, s):
+        t_result = self._t_dict.get(s)
+        if t_result is not None:
+            return t_result
+
+        out_latex_str = self.u.unicode_to_latex(s)
+
+        if (
+            s.startswith("{") is False
+            and out_latex_str.startswith("{")
+            and s.endswith("}") is False
+            and out_latex_str.endswith("}")
+        ):
+            out_latex_str = f" {out_latex_str[1:-1]} "
+
+        if "ensuremath" in out_latex_str:
+            out_latex_str = out_latex_str.replace("\\ensuremath{", " ")
+            out_latex_str = out_latex_str.replace("}", " ")
+
+        return out_latex_str
+
+    def do_r(self, elm):
+        """
+        Get text from 'r' element,And try convert them to latex symbols
+        """
+        _str = []
+        _base_str = []
+        found_text = elm.findtext(f"./{OMML_NS}t")
+        if found_text:
+            for s in found_text:
+                out_latex_str = self.process_unicode(s)
+                _str.append(out_latex_str)
+                _base_str.append(s)
+
+        proc_str = escape_latex(BLANK.join(_str))
+        base_proc_str = BLANK.join(_base_str)
+
+        if "{" not in base_proc_str and "\\{" in proc_str:
+            proc_str = proc_str.replace("\\{", "{")
+
+        if "}" not in base_proc_str and "\\}" in proc_str:
+            proc_str = proc_str.replace("\\}", "}")
+
+        # Handle <m:scr> math font style
+        rPr = elm.find(f"{OMML_NS}rPr")
+        if rPr is not None:
+            scr_elem = rPr.find(f"{OMML_NS}scr")
+            if scr_elem is not None:
+                scr_val = scr_elem.get(f"{OMML_NS}val")
+                latex_template = SCR_TO_LATEX.get(scr_val)
+                if latex_template and proc_str.strip():
+                    proc_str = latex_template.format(proc_str.strip())
+
+        return proc_str
+
+    tag2meth = {
+        "acc": do_acc,
+        "r": do_r,
+        "bar": do_bar,
+        "sub": do_sub,
+        "sup": do_sup,
+        "f": do_f,
+        "func": do_func,
+        "fName": do_fname,
+        "groupChr": do_groupchr,
+        "d": do_d,
+        "rad": do_rad,
+        "eqArr": do_eqarr,
+        "limLow": do_limlow,
+        "limUpp": do_limupp,
+        "lim": do_lim,
+        "m": do_m,
+        "mr": do_mr,
+        "nary": do_nary,
+    }
diff --git a/paddleocr/_doc2md/registry.py b/paddleocr/_doc2md/registry.py
new file mode 100644
index 0000000..829d725
--- /dev/null
+++ b/paddleocr/_doc2md/registry.py
@@ -0,0 +1,54 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import mimetypes
+from pathlib import Path
+from typing import Type
+
+from .base import BaseConverter
+
+
+class ConverterRegistry:
+    """Registry mapping file extensions and MIME types to converter classes."""
+
+    def __init__(self):
+        self._ext_map: dict[str, Type[BaseConverter]] = {}
+        self._mime_map: dict[str, Type[BaseConverter]] = {}
+
+    def register(self, converter_cls: Type[BaseConverter]) -> Type[BaseConverter]:
+        """Register a converter class; can be used as a decorator."""
+        for ext in converter_cls.supported_extensions:
+            self._ext_map[ext.lower().lstrip(".")] = converter_cls
+        for mime in converter_cls.supported_mimetypes:
+            self._mime_map[mime] = converter_cls
+        return converter_cls
+
+    def get_converter(self, file_path: Path) -> BaseConverter:
+        """Return an appropriate converter instance for the given file path."""
+        ext = file_path.suffix.lower().lstrip(".")
+        if ext in self._ext_map:
+            return self._ext_map[ext]()
+
+        mime_type, _ = mimetypes.guess_type(str(file_path))
+        if mime_type and mime_type in self._mime_map:
+            return self._mime_map[mime_type]()
+
+        supported = ", ".join(f".{e}" for e in sorted(self._ext_map.keys()))
+        raise ValueError(f"Unsupported format: .{ext}\nSupported formats: {supported}")
+
+    def supported_extensions(self) -> list[str]:
+        return sorted(self._ext_map.keys())
+
+
+# Global singleton registry
+default_registry = ConverterRegistry()
diff --git a/paddleocr/_env.py b/paddleocr/_env.py
new file mode 100644
index 0000000..9f90e17
--- /dev/null
+++ b/paddleocr/_env.py
@@ -0,0 +1,19 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+
+DISABLE_AUTO_LOGGING_CONFIG = (
+    os.getenv("PADDLEOCR_DISABLE_AUTO_LOGGING_CONFIG", "0") == "1"
+)
diff --git a/paddleocr/_models/__init__.py b/paddleocr/_models/__init__.py
new file mode 100644
index 0000000..3e969ff
--- /dev/null
+++ b/paddleocr/_models/__init__.py
@@ -0,0 +1,43 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .chart_parsing import ChartParsing
+from .doc_img_orientation_classification import DocImgOrientationClassification
+from .doc_vlm import DocVLM
+from .formula_recognition import FormulaRecognition
+from .layout_detection import LayoutDetection
+from .seal_text_detection import SealTextDetection
+from .table_cells_detection import TableCellsDetection
+from .table_classification import TableClassification
+from .table_structure_recognition import TableStructureRecognition
+from .text_detection import TextDetection
+from .text_image_unwarping import TextImageUnwarping
+from .textline_orientation_classification import TextLineOrientationClassification
+from .text_recognition import TextRecognition
+
+__all__ = [
+    "ChartParsing",
+    "DocImgOrientationClassification",
+    "DocVLM",
+    "FormulaRecognition",
+    "LayoutDetection",
+    "SealTextDetection",
+    "TableCellsDetection",
+    "TableClassification",
+    "TableStructureRecognition",
+    "TextDetection",
+    "TextImageUnwarping",
+    "TextLineOrientationClassification",
+    "TextRecognition",
+]
diff --git a/paddleocr/_models/_doc_vlm.py b/paddleocr/_models/_doc_vlm.py
new file mode 100644
index 0000000..18134fd
--- /dev/null
+++ b/paddleocr/_models/_doc_vlm.py
@@ -0,0 +1,49 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import abc
+
+from .._utils.cli import (
+    get_subcommand_args,
+    perform_simple_inference,
+)
+from .base import PaddleXPredictorWrapper, PredictorCLISubcommandExecutor
+from paddlex.utils.pipeline_arguments import custom_type
+
+
+class BaseDocVLM(PaddleXPredictorWrapper):
+    def __init__(
+        self,
+        *args,
+        **kwargs,
+    ):
+        self._extra_init_args = {}
+        super().__init__(*args, **kwargs)
+
+    def _get_extra_paddlex_predictor_init_args(self):
+        return self._extra_init_args
+
+
+class BaseDocVLMSubcommandExecutor(PredictorCLISubcommandExecutor):
+    input_validator = staticmethod(custom_type(dict))
+
+    @property
+    @abc.abstractmethod
+    def wrapper_cls(self):
+        raise NotImplementedError
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+        params["input"] = self.input_validator(params["input"])
+        perform_simple_inference(self.wrapper_cls, params)
diff --git a/paddleocr/_models/_image_classification.py b/paddleocr/_models/_image_classification.py
new file mode 100644
index 0000000..2bad088
--- /dev/null
+++ b/paddleocr/_models/_image_classification.py
@@ -0,0 +1,58 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import abc
+
+from .._utils.cli import (
+    add_simple_inference_args,
+    get_subcommand_args,
+    perform_simple_inference,
+)
+from .base import PaddleXPredictorWrapper, PredictorCLISubcommandExecutor
+
+
+class ImageClassification(PaddleXPredictorWrapper):
+    def __init__(
+        self,
+        *,
+        topk=None,
+        **kwargs,
+    ):
+        self._extra_init_args = {
+            "topk": topk,
+        }
+        super().__init__(**kwargs)
+
+    def _get_extra_paddlex_predictor_init_args(self):
+        return self._extra_init_args
+
+
+class ImageClassificationSubcommandExecutor(PredictorCLISubcommandExecutor):
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(subparser)
+
+        subparser.add_argument(
+            "--topk",
+            type=int,
+            help="Top-k value for prediction results.",
+        )
+
+    @property
+    @abc.abstractmethod
+    def wrapper_cls(self):
+        raise NotImplementedError
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+        perform_simple_inference(self.wrapper_cls, params)
diff --git a/paddleocr/_models/_object_detection.py b/paddleocr/_models/_object_detection.py
new file mode 100644
index 0000000..f7615d9
--- /dev/null
+++ b/paddleocr/_models/_object_detection.py
@@ -0,0 +1,87 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import abc
+
+from .._utils.cli import (
+    add_simple_inference_args,
+    get_subcommand_args,
+    perform_simple_inference,
+    str2bool,
+)
+from .base import PaddleXPredictorWrapper, PredictorCLISubcommandExecutor
+
+
+class ObjectDetection(PaddleXPredictorWrapper):
+    def __init__(
+        self,
+        *,
+        img_size=None,
+        threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        **kwargs,
+    ):
+        self._extra_init_args = {
+            "img_size": img_size,
+            "threshold": threshold,
+            "layout_nms": layout_nms,
+            "layout_unclip_ratio": layout_unclip_ratio,
+            "layout_merge_bboxes_mode": layout_merge_bboxes_mode,
+        }
+        super().__init__(**kwargs)
+
+    def _get_extra_paddlex_predictor_init_args(self):
+        return self._extra_init_args
+
+
+class ObjectDetectionSubcommandExecutor(PredictorCLISubcommandExecutor):
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(subparser)
+
+        subparser.add_argument(
+            "--img_size",
+            type=int,
+            help="Input image size (w, h).",
+        )
+        subparser.add_argument(
+            "--threshold",
+            type=float,
+            help="Threshold for filtering out low-confidence predictions.",
+        )
+        subparser.add_argument(
+            "--layout_nms",
+            type=str2bool,
+            help="Whether to use layout-aware NMS.",
+        )
+        subparser.add_argument(
+            "--layout_unclip_ratio",
+            type=float,
+            help="Ratio of unclipping the bounding box.",
+        )
+        subparser.add_argument(
+            "--layout_merge_bboxes_mode",
+            type=str,
+            help="Mode for merging bounding boxes.",
+        )
+
+    @property
+    @abc.abstractmethod
+    def wrapper_cls(self):
+        raise NotImplementedError
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+        perform_simple_inference(self.wrapper_cls, params)
diff --git a/paddleocr/_models/_text_detection.py b/paddleocr/_models/_text_detection.py
new file mode 100644
index 0000000..22122d3
--- /dev/null
+++ b/paddleocr/_models/_text_detection.py
@@ -0,0 +1,75 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+class TextDetectionMixin:
+    def __init__(
+        self,
+        *,
+        limit_side_len=None,
+        limit_type=None,
+        thresh=None,
+        box_thresh=None,
+        unclip_ratio=None,
+        input_shape=None,
+        **kwargs,
+    ):
+        self._extra_init_args = {
+            "limit_side_len": limit_side_len,
+            "limit_type": limit_type,
+            "thresh": thresh,
+            "box_thresh": box_thresh,
+            "unclip_ratio": unclip_ratio,
+            "input_shape": input_shape,
+        }
+        super().__init__(**kwargs)
+
+    def _get_extra_paddlex_predictor_init_args(self):
+        return self._extra_init_args
+
+
+class TextDetectionSubcommandExecutorMixin:
+    def _add_text_detection_args(self, subparser):
+        subparser.add_argument(
+            "--limit_side_len",
+            type=int,
+            help="This sets a limit on the side length of the input image for the model.",
+        )
+        subparser.add_argument(
+            "--limit_type",
+            type=str,
+            help="This determines how the side length limit is applied to the input image before feeding it into the model.",
+        )
+        subparser.add_argument(
+            "--thresh",
+            type=float,
+            help="Detection pixel threshold for the model. Pixels with scores greater than this threshold in the output probability map are considered text pixels.",
+        )
+        subparser.add_argument(
+            "--box_thresh",
+            type=float,
+            help="Detection box threshold for the model. A detection result is considered a text region if the average score of all pixels within the border of the result is greater than this threshold.",
+        )
+        subparser.add_argument(
+            "--unclip_ratio",
+            type=float,
+            help="Expansion coefficient, which expands the text region using this method. The larger the value, the larger the expansion area.",
+        )
+        subparser.add_argument(
+            "--input_shape",
+            nargs=3,
+            type=int,
+            metavar=("C", "H", "W"),
+            help="Input shape of the model.",
+        )
diff --git a/paddleocr/_models/base.py b/paddleocr/_models/base.py
new file mode 100644
index 0000000..2dd88bc
--- /dev/null
+++ b/paddleocr/_models/base.py
@@ -0,0 +1,107 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import abc
+
+from paddlex import create_predictor
+from paddlex.utils.deps import DependencyError
+
+from .._abstract import CLISubcommandExecutor
+from .._common_args import (
+    add_common_cli_opts,
+    parse_common_args,
+    prepare_common_init_args,
+)
+
+_DEFAULT_ENABLE_HPI = False
+
+
+class PaddleXPredictorWrapper(metaclass=abc.ABCMeta):
+    def __init__(
+        self,
+        *,
+        model_name=None,
+        model_dir=None,
+        **common_args,
+    ):
+        super().__init__()
+        self._model_name = (
+            model_name if model_name is not None else self.default_model_name
+        )
+        self._model_dir = model_dir
+        self._common_args = parse_common_args(
+            common_args, default_enable_hpi=_DEFAULT_ENABLE_HPI
+        )
+        self.paddlex_predictor = self._create_paddlex_predictor()
+
+    @property
+    @abc.abstractmethod
+    def default_model_name(self):
+        raise NotImplementedError
+
+    def predict_iter(self, *args, **kwargs):
+        return self.paddlex_predictor.predict(*args, **kwargs)
+
+    def predict(self, *args, **kwargs):
+        result = list(self.predict_iter(*args, **kwargs))
+        return result
+
+    def close(self):
+        self.paddlex_predictor.close()
+
+    @classmethod
+    @abc.abstractmethod
+    def get_cli_subcommand_executor(cls):
+        raise NotImplementedError
+
+    def _get_extra_paddlex_predictor_init_args(self):
+        return {}
+
+    def _create_paddlex_predictor(self):
+        kwargs = prepare_common_init_args(self._model_name, self._common_args)
+        kwargs = {**self._get_extra_paddlex_predictor_init_args(), **kwargs}
+        # Should we check model names?
+        try:
+            return create_predictor(
+                model_name=self._model_name, model_dir=self._model_dir, **kwargs
+            )
+        except DependencyError as e:
+            raise RuntimeError(
+                "A dependency error occurred during predictor creation. Please refer to the installation documentation to ensure all required dependencies are installed."
+            ) from e
+
+
+class PredictorCLISubcommandExecutor(CLISubcommandExecutor):
+    @property
+    @abc.abstractmethod
+    def subparser_name(self):
+        raise NotImplementedError
+
+    def add_subparser(self, subparsers):
+        subparser = subparsers.add_parser(name=self.subparser_name)
+        self._update_subparser(subparser)
+        subparser.add_argument("--model_name", type=str, help="Name of the model.")
+        subparser.add_argument(
+            "--model_dir", type=str, help="Directory where the model is stored."
+        )
+        add_common_cli_opts(
+            subparser,
+            default_enable_hpi=_DEFAULT_ENABLE_HPI,
+            allow_multiple_devices=False,
+        )
+        return subparser
+
+    @abc.abstractmethod
+    def _update_subparser(self, subparser):
+        raise NotImplementedError
diff --git a/paddleocr/_models/chart_parsing.py b/paddleocr/_models/chart_parsing.py
new file mode 100644
index 0000000..0078bff
--- /dev/null
+++ b/paddleocr/_models/chart_parsing.py
@@ -0,0 +1,45 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .._utils.cli import add_simple_inference_args
+from ._doc_vlm import (
+    BaseDocVLM,
+    BaseDocVLMSubcommandExecutor,
+)
+
+
+class ChartParsing(BaseDocVLM):
+    @property
+    def default_model_name(self):
+        return "PP-Chart2Table"
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return ChartParsingSubcommandExecutor()
+
+
+class ChartParsingSubcommandExecutor(BaseDocVLMSubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "chart_parsing"
+
+    @property
+    def wrapper_cls(self):
+        return ChartParsing
+
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(
+            subparser,
+            input_help='Input dict, e.g. `{"image": "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/chart_parsing_02.png"}`.',
+        )
diff --git a/paddleocr/_models/doc_img_orientation_classification.py b/paddleocr/_models/doc_img_orientation_classification.py
new file mode 100644
index 0000000..7339dfe
--- /dev/null
+++ b/paddleocr/_models/doc_img_orientation_classification.py
@@ -0,0 +1,40 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from ._image_classification import (
+    ImageClassification,
+    ImageClassificationSubcommandExecutor,
+)
+
+
+class DocImgOrientationClassification(ImageClassification):
+    @property
+    def default_model_name(self):
+        return "PP-LCNet_x1_0_doc_ori"
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return DocImgOrientationClassificationSubcommandExecutor()
+
+
+class DocImgOrientationClassificationSubcommandExecutor(
+    ImageClassificationSubcommandExecutor
+):
+    @property
+    def subparser_name(self):
+        return "doc_img_orientation_classification"
+
+    @property
+    def wrapper_cls(self):
+        return DocImgOrientationClassification
diff --git a/paddleocr/_models/doc_vlm.py b/paddleocr/_models/doc_vlm.py
new file mode 100644
index 0000000..b1452a3
--- /dev/null
+++ b/paddleocr/_models/doc_vlm.py
@@ -0,0 +1,45 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .._utils.cli import add_simple_inference_args
+from ._doc_vlm import (
+    BaseDocVLM,
+    BaseDocVLMSubcommandExecutor,
+)
+
+
+class DocVLM(BaseDocVLM):
+    @property
+    def default_model_name(self):
+        return "PP-DocBee2-3B"
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return DocVLMSubcommandExecutor()
+
+
+class DocVLMSubcommandExecutor(BaseDocVLMSubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "doc_vlm"
+
+    @property
+    def wrapper_cls(self):
+        return DocVLM
+
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(
+            subparser,
+            input_help='Input dict, e.g. `{"image": "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/medal_table.png", "query": "Recognize this table"}`.',
+        )
diff --git a/paddleocr/_models/formula_recognition.py b/paddleocr/_models/formula_recognition.py
new file mode 100644
index 0000000..99d2ba7
--- /dev/null
+++ b/paddleocr/_models/formula_recognition.py
@@ -0,0 +1,54 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .._utils.cli import (
+    add_simple_inference_args,
+    get_subcommand_args,
+    perform_simple_inference,
+)
+from .base import PaddleXPredictorWrapper, PredictorCLISubcommandExecutor
+
+
+class FormulaRecognition(PaddleXPredictorWrapper):
+    def __init__(
+        self,
+        *args,
+        **kwargs,
+    ):
+        self._extra_init_args = {}
+        super().__init__(*args, **kwargs)
+
+    @property
+    def default_model_name(self):
+        return "PP-FormulaNet_plus-M"
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return FormulaRecognitionSubcommandExecutor()
+
+    def _get_extra_paddlex_predictor_init_args(self):
+        return self._extra_init_args
+
+
+class FormulaRecognitionSubcommandExecutor(PredictorCLISubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "formula_recognition"
+
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(subparser)
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+        perform_simple_inference(FormulaRecognition, params)
diff --git a/paddleocr/_models/layout_detection.py b/paddleocr/_models/layout_detection.py
new file mode 100644
index 0000000..14427fb
--- /dev/null
+++ b/paddleocr/_models/layout_detection.py
@@ -0,0 +1,38 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from ._object_detection import (
+    ObjectDetection,
+    ObjectDetectionSubcommandExecutor,
+)
+
+
+class LayoutDetection(ObjectDetection):
+    @property
+    def default_model_name(self):
+        return "PP-DocLayout_plus-L"
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return LayoutDetectionSubcommandExecutor()
+
+
+class LayoutDetectionSubcommandExecutor(ObjectDetectionSubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "layout_detection"
+
+    @property
+    def wrapper_cls(self):
+        return LayoutDetection
diff --git a/paddleocr/_models/seal_text_detection.py b/paddleocr/_models/seal_text_detection.py
new file mode 100644
index 0000000..adfff70
--- /dev/null
+++ b/paddleocr/_models/seal_text_detection.py
@@ -0,0 +1,47 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .._utils.cli import (
+    add_simple_inference_args,
+    get_subcommand_args,
+    perform_simple_inference,
+)
+from .base import PaddleXPredictorWrapper, PredictorCLISubcommandExecutor
+from ._text_detection import TextDetectionMixin, TextDetectionSubcommandExecutorMixin
+
+
+class SealTextDetection(TextDetectionMixin, PaddleXPredictorWrapper):
+    @property
+    def default_model_name(self):
+        return "PP-OCRv4_mobile_seal_det"
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return SealTextDetectionSubcommandExecutor()
+
+
+class SealTextDetectionSubcommandExecutor(
+    TextDetectionSubcommandExecutorMixin, PredictorCLISubcommandExecutor
+):
+    @property
+    def subparser_name(self):
+        return "seal_text_detection"
+
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(subparser)
+        self._add_text_detection_args(subparser)
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+        perform_simple_inference(SealTextDetection, params)
diff --git a/paddleocr/_models/table_cells_detection.py b/paddleocr/_models/table_cells_detection.py
new file mode 100644
index 0000000..afca6ed
--- /dev/null
+++ b/paddleocr/_models/table_cells_detection.py
@@ -0,0 +1,38 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from ._object_detection import (
+    ObjectDetection,
+    ObjectDetectionSubcommandExecutor,
+)
+
+
+class TableCellsDetection(ObjectDetection):
+    @property
+    def default_model_name(self):
+        return "RT-DETR-L_wired_table_cell_det"
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return TableCellsDetectionSubcommandExecutor()
+
+
+class TableCellsDetectionSubcommandExecutor(ObjectDetectionSubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "table_cells_detection"
+
+    @property
+    def wrapper_cls(self):
+        return TableCellsDetection
diff --git a/paddleocr/_models/table_classification.py b/paddleocr/_models/table_classification.py
new file mode 100644
index 0000000..028e8d8
--- /dev/null
+++ b/paddleocr/_models/table_classification.py
@@ -0,0 +1,38 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from ._image_classification import (
+    ImageClassification,
+    ImageClassificationSubcommandExecutor,
+)
+
+
+class TableClassification(ImageClassification):
+    @property
+    def default_model_name(self):
+        return "PP-LCNet_x1_0_table_cls"
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return TableClassificationSubcommandExecutor()
+
+
+class TableClassificationSubcommandExecutor(ImageClassificationSubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "table_classification"
+
+    @property
+    def wrapper_cls(self):
+        return TableClassification
diff --git a/paddleocr/_models/table_structure_recognition.py b/paddleocr/_models/table_structure_recognition.py
new file mode 100644
index 0000000..270bd7b
--- /dev/null
+++ b/paddleocr/_models/table_structure_recognition.py
@@ -0,0 +1,54 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .._utils.cli import (
+    add_simple_inference_args,
+    get_subcommand_args,
+    perform_simple_inference,
+)
+from .base import PaddleXPredictorWrapper, PredictorCLISubcommandExecutor
+
+
+class TableStructureRecognition(PaddleXPredictorWrapper):
+    def __init__(
+        self,
+        *args,
+        **kwargs,
+    ):
+        self._extra_init_args = {}
+        super().__init__(*args, **kwargs)
+
+    @property
+    def default_model_name(self):
+        return "SLANet"
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return TableStructureRecognitionSubcommandExecutor()
+
+    def _get_extra_paddlex_predictor_init_args(self):
+        return self._extra_init_args
+
+
+class TableStructureRecognitionSubcommandExecutor(PredictorCLISubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "table_structure_recognition"
+
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(subparser)
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+        perform_simple_inference(TableStructureRecognition, params)
diff --git a/paddleocr/_models/text_detection.py b/paddleocr/_models/text_detection.py
new file mode 100644
index 0000000..9f86316
--- /dev/null
+++ b/paddleocr/_models/text_detection.py
@@ -0,0 +1,47 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .._utils.cli import (
+    add_simple_inference_args,
+    get_subcommand_args,
+    perform_simple_inference,
+)
+from .base import PaddleXPredictorWrapper, PredictorCLISubcommandExecutor
+from ._text_detection import TextDetectionMixin, TextDetectionSubcommandExecutorMixin
+
+
+class TextDetection(TextDetectionMixin, PaddleXPredictorWrapper):
+    @property
+    def default_model_name(self):
+        return "PP-OCRv6_medium_det"
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return TextDetectionSubcommandExecutor()
+
+
+class TextDetectionSubcommandExecutor(
+    TextDetectionSubcommandExecutorMixin, PredictorCLISubcommandExecutor
+):
+    @property
+    def subparser_name(self):
+        return "text_detection"
+
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(subparser)
+        self._add_text_detection_args(subparser)
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+        perform_simple_inference(TextDetection, params)
diff --git a/paddleocr/_models/text_image_unwarping.py b/paddleocr/_models/text_image_unwarping.py
new file mode 100644
index 0000000..6dbbe56
--- /dev/null
+++ b/paddleocr/_models/text_image_unwarping.py
@@ -0,0 +1,54 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .._utils.cli import (
+    add_simple_inference_args,
+    get_subcommand_args,
+    perform_simple_inference,
+)
+from .base import PaddleXPredictorWrapper, PredictorCLISubcommandExecutor
+
+
+class TextImageUnwarping(PaddleXPredictorWrapper):
+    def __init__(
+        self,
+        *args,
+        **kwargs,
+    ):
+        self._extra_init_args = {}
+        super().__init__(*args, **kwargs)
+
+    @property
+    def default_model_name(self):
+        return "UVDoc"
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return TextImageUnwarpingSubcommandExecutor()
+
+    def _get_extra_paddlex_predictor_init_args(self):
+        return self._extra_init_args
+
+
+class TextImageUnwarpingSubcommandExecutor(PredictorCLISubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "text_image_unwarping"
+
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(subparser)
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+        perform_simple_inference(TextImageUnwarping, params)
diff --git a/paddleocr/_models/text_recognition.py b/paddleocr/_models/text_recognition.py
new file mode 100644
index 0000000..e53002a
--- /dev/null
+++ b/paddleocr/_models/text_recognition.py
@@ -0,0 +1,64 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .._utils.cli import (
+    add_simple_inference_args,
+    get_subcommand_args,
+    perform_simple_inference,
+)
+from .base import PaddleXPredictorWrapper, PredictorCLISubcommandExecutor
+
+
+class TextRecognition(PaddleXPredictorWrapper):
+    def __init__(
+        self,
+        *,
+        input_shape=None,
+        **kwargs,
+    ):
+        self._extra_init_args = {
+            "input_shape": input_shape,
+        }
+        super().__init__(**kwargs)
+
+    @property
+    def default_model_name(self):
+        return "PP-OCRv6_medium_rec"
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return TextRecognitionSubcommandExecutor()
+
+    def _get_extra_paddlex_predictor_init_args(self):
+        return self._extra_init_args
+
+
+class TextRecognitionSubcommandExecutor(PredictorCLISubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "text_recognition"
+
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(subparser)
+        subparser.add_argument(
+            "--input_shape",
+            nargs=3,
+            type=int,
+            metavar=("C", "H", "W"),
+            help="Input shape of the model.",
+        )
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+        perform_simple_inference(TextRecognition, params)
diff --git a/paddleocr/_models/textline_orientation_classification.py b/paddleocr/_models/textline_orientation_classification.py
new file mode 100644
index 0000000..908c0dd
--- /dev/null
+++ b/paddleocr/_models/textline_orientation_classification.py
@@ -0,0 +1,40 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from ._image_classification import (
+    ImageClassification,
+    ImageClassificationSubcommandExecutor,
+)
+
+
+class TextLineOrientationClassification(ImageClassification):
+    @property
+    def default_model_name(self):
+        return "PP-LCNet_x0_25_textline_ori"
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return TextLineOrientationClassificationSubcommandExecutor()
+
+
+class TextLineOrientationClassificationSubcommandExecutor(
+    ImageClassificationSubcommandExecutor
+):
+    @property
+    def subparser_name(self):
+        return "textline_orientation_classification"
+
+    @property
+    def wrapper_cls(self):
+        return TextLineOrientationClassification
diff --git a/paddleocr/_pipelines/__init__.py b/paddleocr/_pipelines/__init__.py
new file mode 100644
index 0000000..7315e65
--- /dev/null
+++ b/paddleocr/_pipelines/__init__.py
@@ -0,0 +1,37 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .doc_preprocessor import DocPreprocessor
+from .doc_understanding import DocUnderstanding
+from .formula_recognition import FormulaRecognitionPipeline
+from .ocr import PaddleOCR
+from .paddleocr_vl import PaddleOCRVL
+from .pp_chatocrv4_doc import PPChatOCRv4Doc
+from .pp_doctranslation import PPDocTranslation
+from .pp_structurev3 import PPStructureV3
+from .seal_recognition import SealRecognition
+from .table_recognition_v2 import TableRecognitionPipelineV2
+
+__all__ = [
+    "DocPreprocessor",
+    "DocUnderstanding",
+    "FormulaRecognitionPipeline",
+    "PaddleOCR",
+    "PaddleOCRVL",
+    "PPChatOCRv4Doc",
+    "PPDocTranslation",
+    "PPStructureV3",
+    "SealRecognition",
+    "TableRecognitionPipelineV2",
+]
diff --git a/paddleocr/_pipelines/_patch_layout_parsing.py b/paddleocr/_pipelines/_patch_layout_parsing.py
new file mode 100644
index 0000000..6b73400
--- /dev/null
+++ b/paddleocr/_pipelines/_patch_layout_parsing.py
@@ -0,0 +1,133 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+Patches for paddlex layout parsing utilities to fix:
+- Integer overflow in calculate_overlap_ratio when bounding box coordinates
+  are large (e.g. after doc_unwarping), causing RuntimeWarning and incorrect
+  overlap calculations.
+- ValueError in calculate_minimum_enclosing_bbox when the bounding box list
+  is empty, which can happen when overflow causes all overlap matches to fail.
+
+See: https://github.com/PaddlePaddle/PaddleOCR/issues/17503
+"""
+
+import logging
+
+import numpy as np
+
+logger = logging.getLogger(__name__)
+
+_patched = False
+
+
+def _fixed_calculate_overlap_ratio(bbox1, bbox2, mode="union"):
+    """
+    Calculate the overlap ratio between two bounding boxes.
+
+    This version casts coordinates to float64 before computing areas to
+    prevent integer overflow when bounding box coordinates are large
+    (e.g. after document unwarping).
+    """
+    bbox1 = np.array(bbox1, dtype=np.float64)
+    bbox2 = np.array(bbox2, dtype=np.float64)
+
+    x_min_inter = np.maximum(bbox1[0], bbox2[0])
+    y_min_inter = np.maximum(bbox1[1], bbox2[1])
+    x_max_inter = np.minimum(bbox1[2], bbox2[2])
+    y_max_inter = np.minimum(bbox1[3], bbox2[3])
+
+    inter_width = np.maximum(0, x_max_inter - x_min_inter)
+    inter_height = np.maximum(0, y_max_inter - y_min_inter)
+
+    inter_area = inter_width * inter_height
+
+    bbox1_area = abs((bbox1[2] - bbox1[0]) * (bbox1[3] - bbox1[1]))
+    bbox2_area = abs((bbox2[2] - bbox2[0]) * (bbox2[3] - bbox2[1]))
+
+    if mode == "union":
+        ref_area = bbox1_area + bbox2_area - inter_area
+    elif mode == "small":
+        ref_area = np.minimum(bbox1_area, bbox2_area)
+    elif mode == "large":
+        ref_area = np.maximum(bbox1_area, bbox2_area)
+    else:
+        raise ValueError(
+            f"Invalid mode {mode}, must be one of ['union', 'small', 'large']."
+        )
+
+    if ref_area == 0:
+        return 0.0
+
+    return inter_area / ref_area
+
+
+def _fixed_calculate_minimum_enclosing_bbox(bboxes):
+    """
+    Calculate the minimum enclosing bounding box for a list of bounding boxes.
+
+    This version returns a zero-area bounding box at the origin instead of
+    raising ValueError when the list is empty, allowing the caller to
+    continue without crashing.
+    """
+    if not bboxes:
+        logger.debug(
+            "calculate_minimum_enclosing_bbox received an empty list; "
+            "returning a degenerate bounding box"
+        )
+        return np.array([0, 0, 0, 0])
+
+    bboxes_array = np.array(bboxes)
+
+    min_x = np.min(bboxes_array[:, 0])
+    min_y = np.min(bboxes_array[:, 1])
+    max_x = np.max(bboxes_array[:, 2])
+    max_y = np.max(bboxes_array[:, 3])
+
+    return np.array([min_x, min_y, max_x, max_y])
+
+
+def apply_patches():
+    """
+    Apply patches to paddlex layout parsing utilities to fix integer overflow
+    and empty bounding box errors.
+
+    This function is idempotent and safe to call multiple times.
+    """
+    global _patched
+    if _patched:
+        return
+
+    try:
+        import paddlex.inference.pipelines.layout_parsing.utils as lp_utils
+        import paddlex.inference.pipelines.layout_parsing.pipeline_v2 as lp_pipeline
+    except ImportError:
+        logger.debug("paddlex layout parsing modules not available; skipping patches")
+        return
+
+    # Patch the utils module
+    lp_utils.calculate_overlap_ratio = _fixed_calculate_overlap_ratio
+    lp_utils.calculate_minimum_enclosing_bbox = _fixed_calculate_minimum_enclosing_bbox
+
+    # Also patch the references imported directly into pipeline_v2, since
+    # Python binds names at import time
+    if hasattr(lp_pipeline, "calculate_overlap_ratio"):
+        lp_pipeline.calculate_overlap_ratio = _fixed_calculate_overlap_ratio
+    if hasattr(lp_pipeline, "calculate_minimum_enclosing_bbox"):
+        lp_pipeline.calculate_minimum_enclosing_bbox = (
+            _fixed_calculate_minimum_enclosing_bbox
+        )
+
+    _patched = True
+    logger.debug("Applied layout parsing patches for issue #17503")
diff --git a/paddleocr/_pipelines/base.py b/paddleocr/_pipelines/base.py
new file mode 100644
index 0000000..f400cab
--- /dev/null
+++ b/paddleocr/_pipelines/base.py
@@ -0,0 +1,135 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import abc
+
+import yaml
+from paddlex import create_pipeline
+from paddlex.inference import load_pipeline_config
+from paddlex.utils.config import AttrDict
+from paddlex.utils.deps import DependencyError
+
+from .._abstract import CLISubcommandExecutor
+from .._common_args import (
+    add_common_cli_opts,
+    parse_common_args,
+    prepare_common_init_args,
+)
+
+_DEFAULT_ENABLE_HPI = None
+
+
+def _merge_dicts(d1, d2):
+    res = d1.copy()
+    for k, v in d2.items():
+        if k in res and isinstance(res[k], dict) and isinstance(v, dict):
+            res[k] = _merge_dicts(res[k], v)
+        else:
+            res[k] = v
+    return res
+
+
+def _to_builtin(obj):
+    if isinstance(obj, AttrDict):
+        return {k: _to_builtin(v) for k, v in obj.items()}
+    elif isinstance(obj, dict):
+        return {k: _to_builtin(v) for k, v in obj.items()}
+    elif isinstance(obj, list):
+        return [_to_builtin(item) for item in obj]
+    else:
+        return obj
+
+
+class PaddleXPipelineWrapper(metaclass=abc.ABCMeta):
+    def __init__(
+        self,
+        *,
+        paddlex_config=None,
+        **common_args,
+    ):
+        super().__init__()
+        self._paddlex_config = paddlex_config
+        self._common_args = parse_common_args(
+            common_args, default_enable_hpi=_DEFAULT_ENABLE_HPI
+        )
+        self._merged_paddlex_config = self._get_merged_paddlex_config()
+        self.paddlex_pipeline = self._create_paddlex_pipeline()
+
+    @property
+    @abc.abstractmethod
+    def _paddlex_pipeline_name(self):
+        raise NotImplementedError
+
+    def export_paddlex_config_to_yaml(self, yaml_path):
+        with open(yaml_path, "w", encoding="utf-8") as f:
+            config = _to_builtin(self._merged_paddlex_config)
+            yaml.safe_dump(config, f)
+
+    def close(self):
+        self.paddlex_pipeline.close()
+
+    @classmethod
+    @abc.abstractmethod
+    def get_cli_subcommand_executor(cls):
+        raise NotImplementedError
+
+    def _get_paddlex_config_overrides(self):
+        return {}
+
+    def _get_merged_paddlex_config(self):
+        if self._paddlex_config is None:
+            config = load_pipeline_config(self._paddlex_pipeline_name)
+        elif isinstance(self._paddlex_config, str):
+            config = load_pipeline_config(self._paddlex_config)
+        else:
+            config = self._paddlex_config
+
+        overrides = self._get_paddlex_config_overrides()
+
+        return _merge_dicts(config, overrides)
+
+    def _create_paddlex_pipeline(self):
+        kwargs = prepare_common_init_args(None, self._common_args)
+        try:
+            return create_pipeline(config=self._merged_paddlex_config, **kwargs)
+        except DependencyError as e:
+            raise RuntimeError(
+                "A dependency error occurred during pipeline creation. Please refer to the installation documentation to ensure all required dependencies are installed."
+            ) from e
+
+
+class PipelineCLISubcommandExecutor(CLISubcommandExecutor):
+    @property
+    @abc.abstractmethod
+    def subparser_name(self):
+        raise NotImplementedError
+
+    def add_subparser(self, subparsers):
+        subparser = subparsers.add_parser(name=self.subparser_name)
+        self._update_subparser(subparser)
+        add_common_cli_opts(
+            subparser,
+            default_enable_hpi=_DEFAULT_ENABLE_HPI,
+            allow_multiple_devices=True,
+        )
+        subparser.add_argument(
+            "--paddlex_config",
+            type=str,
+            help="Path to PaddleX pipeline configuration file.",
+        )
+        return subparser
+
+    @abc.abstractmethod
+    def _update_subparser(self, subparser):
+        raise NotImplementedError
diff --git a/paddleocr/_pipelines/doc_preprocessor.py b/paddleocr/_pipelines/doc_preprocessor.py
new file mode 100644
index 0000000..b8c34df
--- /dev/null
+++ b/paddleocr/_pipelines/doc_preprocessor.py
@@ -0,0 +1,147 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .._utils.cli import (
+    add_simple_inference_args,
+    get_subcommand_args,
+    perform_simple_inference,
+    str2bool,
+)
+from .base import PaddleXPipelineWrapper, PipelineCLISubcommandExecutor
+from .utils import create_config_from_structure
+
+
+class DocPreprocessor(PaddleXPipelineWrapper):
+    def __init__(
+        self,
+        doc_orientation_classify_model_name=None,
+        doc_orientation_classify_model_dir=None,
+        doc_unwarping_model_name=None,
+        doc_unwarping_model_dir=None,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        **kwargs,
+    ):
+
+        self._params = {
+            "doc_orientation_classify_model_name": doc_orientation_classify_model_name,
+            "doc_orientation_classify_model_dir": doc_orientation_classify_model_dir,
+            "doc_unwarping_model_name": doc_unwarping_model_name,
+            "doc_unwarping_model_dir": doc_unwarping_model_dir,
+            "use_doc_orientation_classify": use_doc_orientation_classify,
+            "use_doc_unwarping": use_doc_unwarping,
+        }
+        super().__init__(**kwargs)
+
+    @property
+    def _paddlex_pipeline_name(self):
+        return "doc_preprocessor"
+
+    def predict_iter(
+        self,
+        input,
+        *,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+    ):
+        return self.paddlex_pipeline.predict(
+            input,
+            use_doc_orientation_classify=use_doc_orientation_classify,
+            use_doc_unwarping=use_doc_unwarping,
+        )
+
+    def predict(
+        self,
+        input,
+        *,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+    ):
+        return list(
+            self.predict_iter(
+                input,
+                use_doc_orientation_classify=use_doc_orientation_classify,
+                use_doc_unwarping=use_doc_unwarping,
+            )
+        )
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return DocPreprocessorCLISubcommandExecutor()
+
+    def _get_paddlex_config_overrides(self):
+        STRUCTURE = {
+            "SubModules.DocOrientationClassify.model_name": self._params[
+                "doc_orientation_classify_model_name"
+            ],
+            "SubModules.DocOrientationClassify.model_dir": self._params[
+                "doc_orientation_classify_model_dir"
+            ],
+            "SubModules.DocUnwarping.model_name": self._params[
+                "doc_unwarping_model_name"
+            ],
+            "SubModules.DocUnwarping.model_dir": self._params[
+                "doc_unwarping_model_dir"
+            ],
+            "use_doc_orientation_classify": self._params[
+                "use_doc_orientation_classify"
+            ],
+            "use_doc_unwarping": self._params["use_doc_unwarping"],
+        }
+        return create_config_from_structure(STRUCTURE)
+
+
+class DocPreprocessorCLISubcommandExecutor(PipelineCLISubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "doc_preprocessor"
+
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(subparser)
+
+        subparser.add_argument(
+            "--doc_orientation_classify_model_name",
+            type=str,
+            help="Name of the document image orientation classification model.",
+        )
+        subparser.add_argument(
+            "--doc_orientation_classify_model_dir",
+            type=str,
+            help="Path to the document image orientation classification model directory.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_name",
+            type=str,
+            help="Name of the document image unwarping model.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_dir",
+            type=str,
+            help="Path to the document image unwarping model directory.",
+        )
+        subparser.add_argument(
+            "--use_doc_orientation_classify",
+            type=str2bool,
+            help="Whether to use document image orientation classification.",
+        )
+        subparser.add_argument(
+            "--use_doc_unwarping",
+            type=str2bool,
+            help="Whether to use text image unwarping.",
+        )
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+
+        perform_simple_inference(DocPreprocessor, params)
diff --git a/paddleocr/_pipelines/doc_understanding.py b/paddleocr/_pipelines/doc_understanding.py
new file mode 100644
index 0000000..0cde72f
--- /dev/null
+++ b/paddleocr/_pipelines/doc_understanding.py
@@ -0,0 +1,107 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddlex.utils.pipeline_arguments import custom_type
+
+from .._utils.cli import (
+    add_simple_inference_args,
+    get_subcommand_args,
+    perform_simple_inference,
+)
+from .base import PaddleXPipelineWrapper, PipelineCLISubcommandExecutor
+from .utils import create_config_from_structure
+
+
+class DocUnderstanding(PaddleXPipelineWrapper):
+    def __init__(
+        self,
+        doc_understanding_model_name=None,
+        doc_understanding_model_dir=None,
+        doc_understanding_batch_size=None,
+        **kwargs,
+    ):
+
+        self._params = {
+            "doc_understanding_model_name": doc_understanding_model_name,
+            "doc_understanding_model_dir": doc_understanding_model_dir,
+            "doc_understanding_batch_size": doc_understanding_batch_size,
+        }
+        super().__init__(**kwargs)
+
+    @property
+    def _paddlex_pipeline_name(self):
+        return "doc_understanding"
+
+    def predict_iter(self, input, **kwargs):
+        return self.paddlex_pipeline.predict(input, **kwargs)
+
+    def predict(
+        self,
+        input,
+        **kwargs,
+    ):
+        return list(self.predict_iter(input, **kwargs))
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return DocUnderstandingCLISubcommandExecutor()
+
+    def _get_paddlex_config_overrides(self):
+        STRUCTURE = {
+            "SubModules.DocUnderstanding.model_name": self._params[
+                "doc_understanding_model_name"
+            ],
+            "SubModules.DocUnderstanding.model_dir": self._params[
+                "doc_understanding_model_dir"
+            ],
+            "SubModules.DocUnderstanding.batch_size": self._params[
+                "doc_understanding_batch_size"
+            ],
+        }
+        return create_config_from_structure(STRUCTURE)
+
+
+class DocUnderstandingCLISubcommandExecutor(PipelineCLISubcommandExecutor):
+    input_validator = staticmethod(custom_type(dict))
+
+    @property
+    def subparser_name(self):
+        return "doc_understanding"
+
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(
+            subparser,
+            input_help='Input dict, e.g. `{"image": "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/medal_table.png", "query": "Recognize this table"}`.',
+        )
+
+        subparser.add_argument(
+            "--doc_understanding_model_name",
+            type=str,
+            help="Name of the document understanding model.",
+        )
+        subparser.add_argument(
+            "--doc_understanding_model_dir",
+            type=str,
+            help="Path to the document understanding model directory.",
+        )
+        subparser.add_argument(
+            "--doc_understanding_batch_size",
+            type=str,
+            help="Batch size for the document understanding model.",
+        )
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+        params["input"] = self.input_validator(params["input"])
+        perform_simple_inference(DocUnderstanding, params)
diff --git a/paddleocr/_pipelines/formula_recognition.py b/paddleocr/_pipelines/formula_recognition.py
new file mode 100644
index 0000000..7588f81
--- /dev/null
+++ b/paddleocr/_pipelines/formula_recognition.py
@@ -0,0 +1,285 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .._utils.cli import (
+    add_simple_inference_args,
+    get_subcommand_args,
+    perform_simple_inference,
+    str2bool,
+)
+from .base import PaddleXPipelineWrapper, PipelineCLISubcommandExecutor
+from .utils import create_config_from_structure
+
+
+class FormulaRecognitionPipeline(PaddleXPipelineWrapper):
+    def __init__(
+        self,
+        doc_orientation_classify_model_name=None,
+        doc_orientation_classify_model_dir=None,
+        doc_orientation_classify_batch_size=None,
+        doc_unwarping_model_name=None,
+        doc_unwarping_model_dir=None,
+        doc_unwarping_batch_size=None,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        layout_detection_model_name=None,
+        layout_detection_model_dir=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        layout_detection_batch_size=None,
+        use_layout_detection=None,
+        formula_recognition_model_name=None,
+        formula_recognition_model_dir=None,
+        formula_recognition_batch_size=None,
+        **kwargs,
+    ):
+        params = locals().copy()
+        params.pop("self")
+        params.pop("kwargs")
+        self._params = params
+
+        super().__init__(**kwargs)
+
+    @property
+    def _paddlex_pipeline_name(self):
+        return "formula_recognition"
+
+    def predict_iter(
+        self,
+        input,
+        *,
+        use_layout_detection=None,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        layout_det_res=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        **kwargs,
+    ):
+        return self.paddlex_pipeline.predict(
+            input,
+            use_layout_detection=use_layout_detection,
+            use_doc_orientation_classify=use_doc_orientation_classify,
+            use_doc_unwarping=use_doc_unwarping,
+            layout_det_res=layout_det_res,
+            layout_threshold=layout_threshold,
+            layout_nms=layout_nms,
+            layout_unclip_ratio=layout_unclip_ratio,
+            layout_merge_bboxes_mode=layout_merge_bboxes_mode,
+            **kwargs,
+        )
+
+    def predict(
+        self,
+        input,
+        *,
+        use_layout_detection=None,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        layout_det_res=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        **kwargs,
+    ):
+        return list(
+            self.predict_iter(
+                input,
+                use_layout_detection=use_layout_detection,
+                use_doc_orientation_classify=use_doc_orientation_classify,
+                use_doc_unwarping=use_doc_unwarping,
+                layout_det_res=layout_det_res,
+                layout_threshold=layout_threshold,
+                layout_nms=layout_nms,
+                layout_unclip_ratio=layout_unclip_ratio,
+                layout_merge_bboxes_mode=layout_merge_bboxes_mode,
+                **kwargs,
+            )
+        )
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return FormulaRecognitionPipelineCLISubcommandExecutor()
+
+    def _get_paddlex_config_overrides(self):
+        STRUCTURE = {
+            "use_layout_detection": self._params["use_layout_detection"],
+            "SubModules.LayoutDetection.model_name": self._params[
+                "layout_detection_model_name"
+            ],
+            "SubModules.LayoutDetection.model_dir": self._params[
+                "layout_detection_model_dir"
+            ],
+            "SubModules.LayoutDetection.threshold": self._params["layout_threshold"],
+            "SubModules.LayoutDetection.layout_nms": self._params["layout_nms"],
+            "SubModules.LayoutDetection.layout_unclip_ratio": self._params[
+                "layout_unclip_ratio"
+            ],
+            "SubModules.LayoutDetection.layout_merge_bboxes_mode": self._params[
+                "layout_merge_bboxes_mode"
+            ],
+            "SubModules.LayoutDetection.batch_size": self._params[
+                "layout_detection_batch_size"
+            ],
+            "SubModules.FormulaRecognition.model_name": self._params[
+                "formula_recognition_model_name"
+            ],
+            "SubModules.FormulaRecognition.model_dir": self._params[
+                "formula_recognition_model_dir"
+            ],
+            "SubModules.FormulaRecognition.batch_size": self._params[
+                "formula_recognition_batch_size"
+            ],
+            "SubPipelines.DocPreprocessor.use_doc_orientation_classify": self._params[
+                "use_doc_orientation_classify"
+            ],
+            "SubPipelines.DocPreprocessor.use_doc_unwarping": self._params[
+                "use_doc_unwarping"
+            ],
+            "use_doc_preprocessor": self._params["use_doc_orientation_classify"]
+            or self._params["use_doc_unwarping"],
+            "SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify.model_name": self._params[
+                "doc_orientation_classify_model_name"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify.model_dir": self._params[
+                "doc_orientation_classify_model_dir"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify.batch_size": self._params[
+                "doc_orientation_classify_batch_size"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_name": self._params[
+                "doc_unwarping_model_name"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_dir": self._params[
+                "doc_unwarping_model_dir"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocUnwarping.batch_size": self._params[
+                "doc_unwarping_batch_size"
+            ],
+        }
+        return create_config_from_structure(STRUCTURE)
+
+
+class FormulaRecognitionPipelineCLISubcommandExecutor(PipelineCLISubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "formula_recognition_pipeline"
+
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(subparser)
+
+        subparser.add_argument(
+            "--doc_orientation_classify_model_name",
+            type=str,
+            help="Name of the document image orientation classification model.",
+        )
+        subparser.add_argument(
+            "--doc_orientation_classify_model_dir",
+            type=str,
+            help="Directory of the document image orientation classification model.",
+        )
+        subparser.add_argument(
+            "--doc_orientation_classify_batch_size",
+            type=int,
+            help="Batch size for document image orientation classification.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_name",
+            type=str,
+            help="Name of the document unwarping model.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_dir",
+            type=str,
+            help="Directory of the document unwarping model.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_batch_size",
+            type=int,
+            help="Batch size for document unwarping.",
+        )
+        subparser.add_argument(
+            "--use_doc_orientation_classify",
+            type=str2bool,
+            help="Use document image orientation classification.",
+        )
+        subparser.add_argument(
+            "--use_doc_unwarping",
+            type=str2bool,
+            help="Use document unwarping.",
+        )
+        subparser.add_argument(
+            "--layout_detection_model_name",
+            type=str,
+            help="Name of the layout detection model.",
+        )
+        subparser.add_argument(
+            "--layout_detection_model_dir",
+            type=str,
+            help="Directory of the layout detection model.",
+        )
+        subparser.add_argument(
+            "--layout_threshold",
+            type=float,
+            help="Threshold for layout detection.",
+        )
+        subparser.add_argument(
+            "--layout_nms",
+            type=str2bool,
+            help="Non-maximum suppression for layout detection.",
+        )
+        subparser.add_argument(
+            "--layout_unclip_ratio",
+            type=float,
+            help="Unclip ratio for layout detection.",
+        )
+        subparser.add_argument(
+            "--layout_merge_bboxes_mode",
+            type=str,
+            help="Mode for merging bounding boxes in layout detection.",
+        )
+        subparser.add_argument(
+            "--layout_detection_batch_size",
+            type=int,
+            help="Batch size for layout detection.",
+        )
+        subparser.add_argument(
+            "--use_layout_detection",
+            type=str2bool,
+            help="Use layout detection.",
+        )
+        subparser.add_argument(
+            "--formula_recognition_model_name",
+            type=str,
+            help="Name of the formula recognition model.",
+        )
+        subparser.add_argument(
+            "--formula_recognition_model_dir",
+            type=str,
+            help="Directory of the formula recognition model.",
+        )
+        subparser.add_argument(
+            "--formula_recognition_batch_size",
+            type=int,
+            help="Batch size for formula recognition.",
+        )
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+        perform_simple_inference(FormulaRecognitionPipeline, params)
diff --git a/paddleocr/_pipelines/ocr.py b/paddleocr/_pipelines/ocr.py
new file mode 100644
index 0000000..a708a0e
--- /dev/null
+++ b/paddleocr/_pipelines/ocr.py
@@ -0,0 +1,604 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# TODO: Should we use a third-party CLI library to auto-generate command-line
+# arguments from the pipeline class, to reduce boilerplate and improve
+# maintainability?
+
+import sys
+import warnings
+
+from .._utils.cli import (
+    add_simple_inference_args,
+    get_subcommand_args,
+    perform_simple_inference,
+    str2bool,
+)
+from .._utils.deprecation import (
+    DeprecatedOptionAction,
+    deprecated,
+    warn_deprecated_param,
+)
+from .._utils.langs import (
+    ARABIC_LANGS,
+    CYRILLIC_LANGS,
+    DEVANAGARI_LANGS,
+    ESLAV_LANGS,
+    LATIN_LANGS,
+)
+from .._utils.logging import logger
+from .base import PaddleXPipelineWrapper, PipelineCLISubcommandExecutor
+from .utils import create_config_from_structure
+
+_DEPRECATED_PARAM_NAME_MAPPING = {
+    "det_model_dir": "text_detection_model_dir",
+    "det_limit_side_len": "text_det_limit_side_len",
+    "det_limit_type": "text_det_limit_type",
+    "det_db_thresh": "text_det_thresh",
+    "det_db_box_thresh": "text_det_box_thresh",
+    "det_db_unclip_ratio": "text_det_unclip_ratio",
+    "rec_model_dir": "text_recognition_model_dir",
+    "rec_batch_num": "text_recognition_batch_size",
+    "use_angle_cls": "use_textline_orientation",
+    "cls_model_dir": "textline_orientation_model_dir",
+    "cls_batch_num": "textline_orientation_batch_size",
+}
+
+_SUPPORTED_OCR_VERSIONS = ["PP-OCRv3", "PP-OCRv4", "PP-OCRv5", "PP-OCRv6"]
+_PPOCRV6_UNSUPPORTED_LATIN_LANGS = frozenset({"pi"})
+_PPOCRV6_LANGS = frozenset({"ch", "chinese_cht", "en", "japan"}) | (
+    LATIN_LANGS - _PPOCRV6_UNSUPPORTED_LATIN_LANGS
+)
+
+
+# Be comptable with PaddleOCR 2.x interfaces
+class PaddleOCR(PaddleXPipelineWrapper):
+    def __init__(
+        self,
+        doc_orientation_classify_model_name=None,
+        doc_orientation_classify_model_dir=None,
+        doc_unwarping_model_name=None,
+        doc_unwarping_model_dir=None,
+        text_detection_model_name=None,
+        text_detection_model_dir=None,
+        textline_orientation_model_name=None,
+        textline_orientation_model_dir=None,
+        textline_orientation_batch_size=None,
+        text_recognition_model_name=None,
+        text_recognition_model_dir=None,
+        text_recognition_batch_size=None,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_textline_orientation=None,
+        text_det_limit_side_len=None,
+        text_det_limit_type=None,
+        text_det_thresh=None,
+        text_det_box_thresh=None,
+        text_det_unclip_ratio=None,
+        text_det_input_shape=None,
+        text_rec_score_thresh=None,
+        return_word_box=None,
+        text_rec_input_shape=None,
+        lang=None,
+        ocr_version=None,
+        **kwargs,
+    ):
+        if ocr_version is not None and ocr_version not in _SUPPORTED_OCR_VERSIONS:
+            raise ValueError(
+                f"Invalid OCR version: {ocr_version}. Supported values are {_SUPPORTED_OCR_VERSIONS}."
+            )
+
+        if all(
+            map(
+                lambda p: p is None,
+                (
+                    text_detection_model_name,
+                    text_detection_model_dir,
+                    text_recognition_model_name,
+                    text_recognition_model_dir,
+                ),
+            )
+        ):
+            det_model_name, rec_model_name = self._get_ocr_model_names(
+                lang, ocr_version
+            )
+            if det_model_name is None or rec_model_name is None:
+                raise ValueError(
+                    f"No models are available for lang={repr(lang)} and ocr_version={repr(ocr_version)}."
+                )
+            text_detection_model_name = det_model_name
+            text_recognition_model_name = rec_model_name
+        else:
+            if lang is not None or ocr_version is not None:
+                warnings.warn(
+                    "`lang` and `ocr_version` will be ignored when model names or model directories are not `None`.",
+                    stacklevel=2,
+                )
+
+        params = {
+            "doc_orientation_classify_model_name": doc_orientation_classify_model_name,
+            "doc_orientation_classify_model_dir": doc_orientation_classify_model_dir,
+            "doc_unwarping_model_name": doc_unwarping_model_name,
+            "doc_unwarping_model_dir": doc_unwarping_model_dir,
+            "text_detection_model_name": text_detection_model_name,
+            "text_detection_model_dir": text_detection_model_dir,
+            "textline_orientation_model_name": textline_orientation_model_name,
+            "textline_orientation_model_dir": textline_orientation_model_dir,
+            "textline_orientation_batch_size": textline_orientation_batch_size,
+            "text_recognition_model_name": text_recognition_model_name,
+            "text_recognition_model_dir": text_recognition_model_dir,
+            "text_recognition_batch_size": text_recognition_batch_size,
+            "use_doc_orientation_classify": use_doc_orientation_classify,
+            "use_doc_unwarping": use_doc_unwarping,
+            "use_textline_orientation": use_textline_orientation,
+            "text_det_limit_side_len": text_det_limit_side_len,
+            "text_det_limit_type": text_det_limit_type,
+            "text_det_thresh": text_det_thresh,
+            "text_det_box_thresh": text_det_box_thresh,
+            "text_det_unclip_ratio": text_det_unclip_ratio,
+            "text_det_input_shape": text_det_input_shape,
+            "text_rec_score_thresh": text_rec_score_thresh,
+            "return_word_box": return_word_box,
+            "text_rec_input_shape": text_rec_input_shape,
+        }
+        base_params = {}
+        for name, val in kwargs.items():
+            if name in _DEPRECATED_PARAM_NAME_MAPPING:
+                new_name = _DEPRECATED_PARAM_NAME_MAPPING[name]
+                warn_deprecated_param(name, new_name)
+                assert (
+                    new_name in params
+                ), f"{repr(new_name)} is not a valid parameter name."
+                if params[new_name] is not None:
+                    raise ValueError(
+                        f"`{name}` and `{new_name}` are mutually exclusive."
+                    )
+                params[new_name] = val
+            else:
+                base_params[name] = val
+
+        self._params = params
+
+        super().__init__(**base_params)
+
+    @property
+    def _paddlex_pipeline_name(self):
+        return "OCR"
+
+    def predict_iter(
+        self,
+        input,
+        *,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_textline_orientation=None,
+        text_det_limit_side_len=None,
+        text_det_limit_type=None,
+        text_det_thresh=None,
+        text_det_box_thresh=None,
+        text_det_unclip_ratio=None,
+        text_rec_score_thresh=None,
+        return_word_box=None,
+    ):
+        return self.paddlex_pipeline.predict(
+            input,
+            use_doc_orientation_classify=use_doc_orientation_classify,
+            use_doc_unwarping=use_doc_unwarping,
+            use_textline_orientation=use_textline_orientation,
+            text_det_limit_side_len=text_det_limit_side_len,
+            text_det_limit_type=text_det_limit_type,
+            text_det_thresh=text_det_thresh,
+            text_det_box_thresh=text_det_box_thresh,
+            text_det_unclip_ratio=text_det_unclip_ratio,
+            text_rec_score_thresh=text_rec_score_thresh,
+            return_word_box=return_word_box,
+        )
+
+    def predict(
+        self,
+        input,
+        *,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_textline_orientation=None,
+        text_det_limit_side_len=None,
+        text_det_limit_type=None,
+        text_det_thresh=None,
+        text_det_box_thresh=None,
+        text_det_unclip_ratio=None,
+        text_rec_score_thresh=None,
+        return_word_box=None,
+    ):
+        return list(
+            self.predict_iter(
+                input,
+                use_doc_orientation_classify=use_doc_orientation_classify,
+                use_doc_unwarping=use_doc_unwarping,
+                use_textline_orientation=use_textline_orientation,
+                text_det_limit_side_len=text_det_limit_side_len,
+                text_det_limit_type=text_det_limit_type,
+                text_det_thresh=text_det_thresh,
+                text_det_box_thresh=text_det_box_thresh,
+                text_det_unclip_ratio=text_det_unclip_ratio,
+                text_rec_score_thresh=text_rec_score_thresh,
+                return_word_box=return_word_box,
+            )
+        )
+
+    @deprecated("Please use `predict` instead.")
+    def ocr(self, img, **kwargs):
+        return self.predict(img, **kwargs)
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return PaddleOCRCLISubcommandExecutor()
+
+    def _get_paddlex_config_overrides(self):
+        STRUCTURE = {
+            "SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify.model_name": self._params[
+                "doc_orientation_classify_model_name"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify.model_dir": self._params[
+                "doc_orientation_classify_model_dir"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_name": self._params[
+                "doc_unwarping_model_name"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_dir": self._params[
+                "doc_unwarping_model_dir"
+            ],
+            "SubModules.TextDetection.model_name": self._params[
+                "text_detection_model_name"
+            ],
+            "SubModules.TextDetection.model_dir": self._params[
+                "text_detection_model_dir"
+            ],
+            "SubModules.TextLineOrientation.model_name": self._params[
+                "textline_orientation_model_name"
+            ],
+            "SubModules.TextLineOrientation.model_dir": self._params[
+                "textline_orientation_model_dir"
+            ],
+            "SubModules.TextLineOrientation.batch_size": self._params[
+                "textline_orientation_batch_size"
+            ],
+            "SubModules.TextRecognition.model_name": self._params[
+                "text_recognition_model_name"
+            ],
+            "SubModules.TextRecognition.model_dir": self._params[
+                "text_recognition_model_dir"
+            ],
+            "SubModules.TextRecognition.batch_size": self._params[
+                "text_recognition_batch_size"
+            ],
+            "SubPipelines.DocPreprocessor.use_doc_orientation_classify": self._params[
+                "use_doc_orientation_classify"
+            ],
+            "SubPipelines.DocPreprocessor.use_doc_unwarping": self._params[
+                "use_doc_unwarping"
+            ],
+            "use_doc_preprocessor": self._params["use_doc_orientation_classify"]
+            or self._params["use_doc_unwarping"],
+            "use_textline_orientation": self._params["use_textline_orientation"],
+            "SubModules.TextDetection.limit_side_len": self._params[
+                "text_det_limit_side_len"
+            ],
+            "SubModules.TextDetection.limit_type": self._params["text_det_limit_type"],
+            "SubModules.TextDetection.thresh": self._params["text_det_thresh"],
+            "SubModules.TextDetection.box_thresh": self._params["text_det_box_thresh"],
+            "SubModules.TextDetection.unclip_ratio": self._params[
+                "text_det_unclip_ratio"
+            ],
+            "SubModules.TextDetection.input_shape": self._params[
+                "text_det_input_shape"
+            ],
+            "SubModules.TextRecognition.score_thresh": self._params[
+                "text_rec_score_thresh"
+            ],
+            "SubModules.TextRecognition.return_word_box": self._params[
+                "return_word_box"
+            ],
+            "SubModules.TextRecognition.input_shape": self._params[
+                "text_rec_input_shape"
+            ],
+        }
+        return create_config_from_structure(STRUCTURE)
+
+    def _get_ocr_model_names(self, lang, ppocr_version):
+        SPECIFIC_LANGS = [
+            "ch",
+            "en",
+            "korean",
+            "japan",
+            "chinese_cht",
+            "te",
+            "ka",
+            "ta",
+        ]
+
+        if lang is None and ppocr_version is None:
+            return "PP-OCRv6_medium_det", "PP-OCRv6_medium_rec"
+
+        if lang is None:
+            lang = "ch"
+
+        if ppocr_version is None:
+            if lang in _PPOCRV6_LANGS:
+                ppocr_version = "PP-OCRv6"
+            elif lang in {
+                "korean",
+                "th",
+                "el",
+                "te",
+                "ta",
+            } | ESLAV_LANGS | ARABIC_LANGS | CYRILLIC_LANGS | DEVANAGARI_LANGS | (
+                LATIN_LANGS - _PPOCRV6_LANGS
+            ):
+                ppocr_version = "PP-OCRv5"
+            elif lang == "ka":
+                ppocr_version = "PP-OCRv3"
+            else:
+                # Unknown language specified
+                return None, None
+
+        if ppocr_version == "PP-OCRv6":
+            if lang in _PPOCRV6_LANGS:
+                return "PP-OCRv6_medium_det", "PP-OCRv6_medium_rec"
+            return None, None
+
+        elif ppocr_version == "PP-OCRv5":
+            rec_lang, rec_model_name = None, None
+            if lang in ("ch", "chinese_cht", "japan"):
+                rec_model_name = "PP-OCRv5_server_rec"
+            elif lang == "en":
+                rec_model_name = "en_PP-OCRv5_mobile_rec"
+            elif lang in LATIN_LANGS:
+                rec_lang = "latin"
+            elif lang in ESLAV_LANGS:
+                rec_lang = "eslav"
+            elif lang in ARABIC_LANGS:
+                rec_lang = "arabic"
+            elif lang in CYRILLIC_LANGS:
+                rec_lang = "cyrillic"
+            elif lang in DEVANAGARI_LANGS:
+                rec_lang = "devanagari"
+            elif lang == "korean":
+                rec_lang = "korean"
+            elif lang == "th":
+                rec_lang = "th"
+            elif lang == "el":
+                rec_lang = "el"
+            elif lang == "te":
+                rec_lang = "te"
+            elif lang == "ta":
+                rec_lang = "ta"
+
+            if rec_lang is not None:
+                rec_model_name = f"{rec_lang}_PP-OCRv5_mobile_rec"
+            if rec_model_name is None:
+                return None, None
+            return "PP-OCRv5_server_det", rec_model_name
+
+        elif ppocr_version == "PP-OCRv4":
+            if lang == "ch":
+                return "PP-OCRv4_mobile_det", "PP-OCRv4_mobile_rec"
+            elif lang == "en":
+                return "PP-OCRv4_mobile_det", "en_PP-OCRv4_mobile_rec"
+            else:
+                return None, None
+        else:
+            # PP-OCRv3
+            rec_lang = None
+            if lang in LATIN_LANGS:
+                rec_lang = "latin"
+            elif lang in ARABIC_LANGS:
+                rec_lang = "arabic"
+            elif lang in CYRILLIC_LANGS:
+                rec_lang = "cyrillic"
+            elif lang in DEVANAGARI_LANGS:
+                rec_lang = "devanagari"
+            else:
+                if lang in SPECIFIC_LANGS:
+                    rec_lang = lang
+
+            rec_model_name = None
+            if rec_lang == "ch":
+                rec_model_name = "PP-OCRv3_mobile_rec"
+            elif rec_lang is not None:
+                rec_model_name = f"{rec_lang}_PP-OCRv3_mobile_rec"
+            if rec_model_name is None:
+                return None, None
+            return "PP-OCRv3_mobile_det", rec_model_name
+
+
+class PaddleOCRCLISubcommandExecutor(PipelineCLISubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "ocr"
+
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(subparser)
+
+        subparser.add_argument(
+            "--doc_orientation_classify_model_name",
+            type=str,
+            help="Name of the document image orientation classification model.",
+        )
+        subparser.add_argument(
+            "--doc_orientation_classify_model_dir",
+            type=str,
+            help="Path to the document image orientation classification model directory.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_name",
+            type=str,
+            help="Name of the text image unwarping model.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_dir",
+            type=str,
+            help="Path to the image unwarping model directory.",
+        )
+        subparser.add_argument(
+            "--text_detection_model_name",
+            type=str,
+            help="Name of the text detection model.",
+        )
+        subparser.add_argument(
+            "--text_detection_model_dir",
+            type=str,
+            help="Path to the text detection model directory.",
+        )
+        subparser.add_argument(
+            "--textline_orientation_model_name",
+            type=str,
+            help="Name of the text line orientation classification model.",
+        )
+        subparser.add_argument(
+            "--textline_orientation_model_dir",
+            type=str,
+            help="Path to the text line orientation classification model directory.",
+        )
+        subparser.add_argument(
+            "--textline_orientation_batch_size",
+            type=int,
+            help="Batch size for the text line orientation classification model.",
+        )
+        subparser.add_argument(
+            "--text_recognition_model_name",
+            type=str,
+            help="Name of the text recognition model.",
+        )
+        subparser.add_argument(
+            "--text_recognition_model_dir",
+            type=str,
+            help="Path to the text recognition model directory.",
+        )
+        subparser.add_argument(
+            "--text_recognition_batch_size",
+            type=int,
+            help="Batch size for the text recognition model.",
+        )
+        subparser.add_argument(
+            "--use_doc_orientation_classify",
+            type=str2bool,
+            help="Whether to use document image orientation classification.",
+        )
+        subparser.add_argument(
+            "--use_doc_unwarping",
+            type=str2bool,
+            help="Whether to use text image unwarping.",
+        )
+        subparser.add_argument(
+            "--use_textline_orientation",
+            type=str2bool,
+            help="Whether to use text line orientation classification.",
+        )
+        subparser.add_argument(
+            "--text_det_limit_side_len",
+            type=int,
+            help="This sets a limit on the side length of the input image for the text detection model.",
+        )
+        subparser.add_argument(
+            "--text_det_limit_type",
+            type=str,
+            help="This determines how the side length limit is applied to the input image before feeding it into the text deteciton model.",
+        )
+        subparser.add_argument(
+            "--text_det_thresh",
+            type=float,
+            help="Detection pixel threshold for the text detection model. Pixels with scores greater than this threshold in the output probability map are considered text pixels.",
+        )
+        subparser.add_argument(
+            "--text_det_box_thresh",
+            type=float,
+            help="Detection box threshold for the text detection model. A detection result is considered a text region if the average score of all pixels within the border of the result is greater than this threshold.",
+        )
+        subparser.add_argument(
+            "--text_det_unclip_ratio",
+            type=float,
+            help="Text detection expansion coefficient, which expands the text region using this method. The larger the value, the larger the expansion area.",
+        )
+        subparser.add_argument(
+            "--text_det_input_shape",
+            nargs=3,
+            type=int,
+            metavar=("C", "H", "W"),
+            help="Input shape of the text detection model.",
+        )
+        subparser.add_argument(
+            "--text_rec_score_thresh",
+            type=float,
+            help="Text recognition threshold. Text results with scores greater than this threshold are retained.",
+        )
+        subparser.add_argument(
+            "--return_word_box",
+            type=str2bool,
+            help="Whether to return the coordinates of the recognition result.",
+        )
+        subparser.add_argument(
+            "--text_rec_input_shape",
+            nargs=3,
+            type=int,
+            metavar=("C", "H", "W"),
+            help="Input shape of the text recognition model.",
+        )
+        subparser.add_argument(
+            "--lang", type=str, help="Language in the input image for OCR processing."
+        )
+        subparser.add_argument(
+            "--ocr_version",
+            type=str,
+            choices=_SUPPORTED_OCR_VERSIONS,
+            help="PP-OCR version to use.",
+        )
+
+        deprecated_arg_types = {
+            "det_model_dir": str,
+            "det_limit_side_len": int,
+            "det_limit_type": str,
+            "det_db_thresh": float,
+            "det_db_box_thresh": float,
+            "det_db_unclip_ratio": float,
+            "rec_model_dir": str,
+            "rec_batch_num": int,
+            "use_angle_cls": str2bool,
+            "cls_model_dir": str,
+            "cls_batch_num": int,
+        }
+
+        for name, new_name in _DEPRECATED_PARAM_NAME_MAPPING.items():
+            assert name in deprecated_arg_types, name
+            subparser.add_argument(
+                "--" + name,
+                action=DeprecatedOptionAction,
+                type=deprecated_arg_types[name],
+                help=f"[Deprecated] Please use `--{new_name}` instead.",
+            )
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+        for name, new_name in _DEPRECATED_PARAM_NAME_MAPPING.items():
+            assert name in params
+            val = params[name]
+            new_val = params[new_name]
+            if val is not None and new_val is not None:
+                logger.error(
+                    "`--%s` and `--%s` are mutually exclusive.", name, new_name
+                )
+                sys.exit(2)
+            if val is None:
+                params.pop(name)
+
+        perform_simple_inference(PaddleOCR, params)
diff --git a/paddleocr/_pipelines/paddleocr_vl.py b/paddleocr/_pipelines/paddleocr_vl.py
new file mode 100644
index 0000000..8edad89
--- /dev/null
+++ b/paddleocr/_pipelines/paddleocr_vl.py
@@ -0,0 +1,514 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .._utils.cli import (
+    add_simple_inference_args,
+    get_subcommand_args,
+    perform_simple_inference,
+    str2bool,
+)
+from .base import PaddleXPipelineWrapper, PipelineCLISubcommandExecutor
+from .utils import create_config_from_structure
+
+
+_AVAILABLE_PIPELINE_VERSIONS = ["v1", "v1.5", "v1.6"]
+_DEFAULT_PIPELINE_VERSION = "v1.6"
+_SUPPORTED_VL_BACKENDS = [
+    "native",
+    "vllm-server",
+    "sglang-server",
+    "fastdeploy-server",
+    "mlx-vlm-server",
+    "llama-cpp-server",
+]
+
+
+class PaddleOCRVL(PaddleXPipelineWrapper):
+    def __init__(
+        self,
+        pipeline_version=_DEFAULT_PIPELINE_VERSION,
+        layout_detection_model_name=None,
+        layout_detection_model_dir=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        vl_rec_model_name=None,
+        vl_rec_model_dir=None,
+        vl_rec_backend=None,
+        vl_rec_server_url=None,
+        vl_rec_max_concurrency=None,
+        vl_rec_api_model_name=None,
+        vl_rec_api_key=None,
+        doc_orientation_classify_model_name=None,
+        doc_orientation_classify_model_dir=None,
+        doc_unwarping_model_name=None,
+        doc_unwarping_model_dir=None,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_layout_detection=None,
+        use_chart_recognition=None,
+        use_seal_recognition=None,
+        use_ocr_for_image_block=None,
+        format_block_content=None,
+        merge_layout_blocks=None,
+        markdown_ignore_labels=None,
+        use_queues=None,
+        **kwargs,
+    ):
+        if pipeline_version not in _AVAILABLE_PIPELINE_VERSIONS:
+            raise ValueError(
+                f"Invalid pipeline version: {pipeline_version}. Supported versions are {_AVAILABLE_PIPELINE_VERSIONS}."
+            )
+
+        if vl_rec_backend is not None and vl_rec_backend not in _SUPPORTED_VL_BACKENDS:
+            raise ValueError(
+                f"Invalid backend for the VL recognition module: {vl_rec_backend}. Supported values are {_SUPPORTED_VL_BACKENDS}."
+            )
+
+        params = locals().copy()
+        params.pop("self")
+        params.pop("pipeline_version")
+        params.pop("kwargs")
+        self._params = params
+        self.pipeline_version = pipeline_version
+
+        super().__init__(**kwargs)
+
+    @property
+    def _paddlex_pipeline_name(self):
+        if self.pipeline_version == "v1":
+            return "PaddleOCR-VL"
+        elif self.pipeline_version == "v1.5":
+            return "PaddleOCR-VL-1.5"
+        elif self.pipeline_version == "v1.6":
+            return "PaddleOCR-VL-1.6"
+        else:
+            raise AssertionError(f"Unknown pipeline version: {self.pipeline_version}")
+
+    def predict_iter(
+        self,
+        input,
+        *,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_layout_detection=None,
+        use_chart_recognition=None,
+        use_seal_recognition=None,
+        use_ocr_for_image_block=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        layout_shape_mode="auto",
+        use_queues=None,
+        prompt_label=None,
+        format_block_content=None,
+        repetition_penalty=None,
+        temperature=None,
+        top_p=None,
+        min_pixels=None,
+        max_pixels=None,
+        max_new_tokens=None,
+        merge_layout_blocks=None,
+        markdown_ignore_labels=None,
+        vlm_extra_args=None,
+        **kwargs,
+    ):
+        return self.paddlex_pipeline.predict(
+            input,
+            use_doc_orientation_classify=use_doc_orientation_classify,
+            use_doc_unwarping=use_doc_unwarping,
+            use_layout_detection=use_layout_detection,
+            use_chart_recognition=use_chart_recognition,
+            use_seal_recognition=use_seal_recognition,
+            use_ocr_for_image_block=use_ocr_for_image_block,
+            layout_threshold=layout_threshold,
+            layout_nms=layout_nms,
+            layout_unclip_ratio=layout_unclip_ratio,
+            layout_merge_bboxes_mode=layout_merge_bboxes_mode,
+            layout_shape_mode=layout_shape_mode,
+            use_queues=use_queues,
+            prompt_label=prompt_label,
+            format_block_content=format_block_content,
+            repetition_penalty=repetition_penalty,
+            temperature=temperature,
+            top_p=top_p,
+            min_pixels=min_pixels,
+            max_pixels=max_pixels,
+            max_new_tokens=max_new_tokens,
+            merge_layout_blocks=merge_layout_blocks,
+            markdown_ignore_labels=markdown_ignore_labels,
+            vlm_extra_args=vlm_extra_args,
+            **kwargs,
+        )
+
+    def predict(
+        self,
+        input,
+        *,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_layout_detection=None,
+        use_chart_recognition=None,
+        use_seal_recognition=None,
+        use_ocr_for_image_block=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        layout_shape_mode="auto",
+        use_queues=None,
+        prompt_label=None,
+        format_block_content=None,
+        repetition_penalty=None,
+        temperature=None,
+        top_p=None,
+        min_pixels=None,
+        max_pixels=None,
+        max_new_tokens=None,
+        merge_layout_blocks=None,
+        markdown_ignore_labels=None,
+        vlm_extra_args=None,
+        **kwargs,
+    ):
+        return list(
+            self.predict_iter(
+                input,
+                use_doc_orientation_classify=use_doc_orientation_classify,
+                use_doc_unwarping=use_doc_unwarping,
+                use_layout_detection=use_layout_detection,
+                use_chart_recognition=use_chart_recognition,
+                use_seal_recognition=use_seal_recognition,
+                use_ocr_for_image_block=use_ocr_for_image_block,
+                layout_threshold=layout_threshold,
+                layout_nms=layout_nms,
+                layout_unclip_ratio=layout_unclip_ratio,
+                layout_merge_bboxes_mode=layout_merge_bboxes_mode,
+                layout_shape_mode=layout_shape_mode,
+                use_queues=use_queues,
+                prompt_label=prompt_label,
+                format_block_content=format_block_content,
+                repetition_penalty=repetition_penalty,
+                temperature=temperature,
+                top_p=top_p,
+                min_pixels=min_pixels,
+                max_pixels=max_pixels,
+                max_new_tokens=max_new_tokens,
+                merge_layout_blocks=merge_layout_blocks,
+                markdown_ignore_labels=markdown_ignore_labels,
+                vlm_extra_args=vlm_extra_args,
+                **kwargs,
+            )
+        )
+
+    def concatenate_markdown_pages(self, markdown_list):
+        return self.paddlex_pipeline.concatenate_markdown_pages(markdown_list)
+
+    def restructure_pages(
+        self, res_list, merge_tables=True, relevel_titles=True, concatenate_pages=False
+    ):
+        return list(
+            self.paddlex_pipeline.restructure_pages(
+                res_list,
+                merge_tables=merge_tables,
+                relevel_titles=relevel_titles,
+                concatenate_pages=concatenate_pages,
+            )
+        )
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return PaddleOCRVLCLISubcommandExecutor()
+
+    def _get_paddlex_config_overrides(self):
+        STRUCTURE = {
+            "SubPipelines.DocPreprocessor.use_doc_orientation_classify": self._params[
+                "use_doc_orientation_classify"
+            ],
+            "SubPipelines.DocPreprocessor.use_doc_unwarping": self._params[
+                "use_doc_unwarping"
+            ],
+            "use_doc_preprocessor": self._params["use_doc_orientation_classify"]
+            or self._params["use_doc_unwarping"],
+            "use_layout_detection": self._params["use_layout_detection"],
+            "use_chart_recognition": self._params["use_chart_recognition"],
+            "format_block_content": self._params["format_block_content"],
+            "merge_layout_blocks": self._params["merge_layout_blocks"],
+            "markdown_ignore_labels": self._params["markdown_ignore_labels"],
+            "use_queues": self._params["use_queues"],
+            "SubModules.LayoutDetection.model_name": self._params[
+                "layout_detection_model_name"
+            ],
+            "SubModules.LayoutDetection.model_dir": self._params[
+                "layout_detection_model_dir"
+            ],
+            "SubModules.LayoutDetection.threshold": self._params["layout_threshold"],
+            "SubModules.LayoutDetection.layout_nms": self._params["layout_nms"],
+            "SubModules.LayoutDetection.layout_unclip_ratio": self._params[
+                "layout_unclip_ratio"
+            ],
+            "SubModules.LayoutDetection.layout_merge_bboxes_mode": self._params[
+                "layout_merge_bboxes_mode"
+            ],
+            "SubModules.VLRecognition.model_name": self._params["vl_rec_model_name"],
+            "SubModules.VLRecognition.model_dir": self._params["vl_rec_model_dir"],
+            "SubModules.VLRecognition.genai_config.backend": self._params[
+                "vl_rec_backend"
+            ],
+            "SubModules.VLRecognition.genai_config.server_url": self._params[
+                "vl_rec_server_url"
+            ],
+            "SubModules.VLRecognition.genai_config.max_concurrency": self._params[
+                "vl_rec_max_concurrency"
+            ],
+            "SubModules.VLRecognition.genai_config.client_kwargs.model_name": self._params[
+                "vl_rec_api_model_name"
+            ],
+            "SubModules.VLRecognition.genai_config.client_kwargs.api_key": self._params[
+                "vl_rec_api_key"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify.model_name": self._params[
+                "doc_orientation_classify_model_name"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify.model_dir": self._params[
+                "doc_orientation_classify_model_dir"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_name": self._params[
+                "doc_unwarping_model_name"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_dir": self._params[
+                "doc_unwarping_model_dir"
+            ],
+            "use_seal_recognition": self._params["use_seal_recognition"],
+            "use_ocr_for_image_block": self._params["use_ocr_for_image_block"],
+        }
+        return create_config_from_structure(STRUCTURE)
+
+
+class PaddleOCRVLCLISubcommandExecutor(PipelineCLISubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "doc_parser"
+
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(subparser)
+
+        subparser.add_argument(
+            "--pipeline_version",
+            type=str,
+            default=_DEFAULT_PIPELINE_VERSION,
+            choices=_AVAILABLE_PIPELINE_VERSIONS,
+        )
+
+        subparser.add_argument(
+            "--layout_detection_model_name",
+            type=str,
+            help="Name of the layout analysis model.",
+        )
+        subparser.add_argument(
+            "--layout_detection_model_dir",
+            type=str,
+            help="Path to the layout analysis model directory.",
+        )
+        subparser.add_argument(
+            "--layout_threshold",
+            type=float,
+            help="Score threshold for the layout analysis model.",
+        )
+        subparser.add_argument(
+            "--layout_nms",
+            type=str2bool,
+            help="Whether to use NMS in layout analysis.",
+        )
+        subparser.add_argument(
+            "--layout_unclip_ratio",
+            type=float,
+            help="Expansion coefficient for layout analysis.",
+        )
+        subparser.add_argument(
+            "--layout_merge_bboxes_mode",
+            type=str,
+            help="Overlapping box filtering method.",
+        )
+
+        subparser.add_argument(
+            "--vl_rec_model_name",
+            type=str,
+            help="Name of the VL recognition model.",
+        )
+        subparser.add_argument(
+            "--vl_rec_model_dir",
+            type=str,
+            help="Path to the VL recognition model directory.",
+        )
+        subparser.add_argument(
+            "--vl_rec_backend",
+            type=str,
+            help="Backend used by the VL recognition module.",
+            choices=_SUPPORTED_VL_BACKENDS,
+        )
+        subparser.add_argument(
+            "--vl_rec_server_url",
+            type=str,
+            help="Server URL used by the VL recognition module.",
+        )
+        subparser.add_argument(
+            "--vl_rec_max_concurrency",
+            type=int,
+            help="Maximum concurrency for making VLM requests.",
+        )
+        subparser.add_argument(
+            "--vl_rec_api_model_name",
+            type=str,
+            help="Model name for the VLM server.",
+        )
+        subparser.add_argument(
+            "--vl_rec_api_key",
+            type=str,
+            help="API key for the VLM server.",
+        )
+
+        subparser.add_argument(
+            "--doc_orientation_classify_model_name",
+            type=str,
+            help="Name of the document image orientation classification model.",
+        )
+        subparser.add_argument(
+            "--doc_orientation_classify_model_dir",
+            type=str,
+            help="Path to the document image orientation classification model directory.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_name",
+            type=str,
+            help="Name of the text image unwarping model.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_dir",
+            type=str,
+            help="Path to the image unwarping model directory.",
+        )
+
+        subparser.add_argument(
+            "--use_doc_orientation_classify",
+            type=str2bool,
+            help="Whether to use document image orientation classification.",
+        )
+        subparser.add_argument(
+            "--use_doc_unwarping",
+            type=str2bool,
+            help="Whether to use text image unwarping.",
+        )
+        subparser.add_argument(
+            "--use_layout_detection",
+            type=str2bool,
+            help="Whether to use layout analysis.",
+        )
+        subparser.add_argument(
+            "--use_chart_recognition",
+            type=str2bool,
+            help="Whether to use chart recognition.",
+        )
+        subparser.add_argument(
+            "--use_seal_recognition",
+            type=str2bool,
+            help="Whether to use seal recognition.",
+        )
+        subparser.add_argument(
+            "--use_ocr_for_image_block",
+            type=str2bool,
+            help="Whether to use OCR for image blocks.",
+        )
+        subparser.add_argument(
+            "--format_block_content",
+            type=str2bool,
+            help="Whether to format block content to Markdown.",
+        )
+        subparser.add_argument(
+            "--merge_layout_blocks",
+            type=str2bool,
+            help="Whether to merge layout blocks.",
+        )
+        subparser.add_argument(
+            "--markdown_ignore_labels",
+            type=str,
+            nargs="+",
+            help="List of layout labels to ignore in Markdown output.",
+        )
+
+        subparser.add_argument(
+            "--use_queues",
+            type=str2bool,
+            help="Whether to use queues for asynchronous processing.",
+        )
+
+        subparser.add_argument(
+            "--layout_shape_mode",
+            type=str,
+            default="auto",
+            help="Mode for layout shape.",
+        )
+        subparser.add_argument(
+            "--prompt_label",
+            type=str,
+            help="Prompt label for the VLM.",
+        )
+        subparser.add_argument(
+            "--repetition_penalty",
+            type=float,
+            help="Repetition penalty used in sampling for the VLM.",
+        )
+        subparser.add_argument(
+            "--temperature",
+            type=float,
+            help="Temperature parameter used in sampling for the VLM.",
+        )
+        subparser.add_argument(
+            "--top_p",
+            type=float,
+            help="Top-p parameter used in sampling for the VLM.",
+        )
+        subparser.add_argument(
+            "--min_pixels",
+            type=int,
+            help="Minimum pixels for image preprocessing for the VLM.",
+        )
+        subparser.add_argument(
+            "--max_pixels",
+            type=int,
+            help="Maximum pixels for image preprocessing for the VLM.",
+        )
+        subparser.add_argument(
+            "--max_new_tokens",
+            type=int,
+            help="Maximum number of tokens generated by the VLM.",
+        )
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+        perform_simple_inference(
+            PaddleOCRVL,
+            params,
+            predict_param_names={
+                "layout_shape_mode",
+                "prompt_label",
+                "repetition_penalty",
+                "temperature",
+                "top_p",
+                "min_pixels",
+                "max_pixels",
+                "max_new_tokens",
+            },
+        )
diff --git a/paddleocr/_pipelines/pp_chatocrv4_doc.py b/paddleocr/_pipelines/pp_chatocrv4_doc.py
new file mode 100644
index 0000000..c9f0f92
--- /dev/null
+++ b/paddleocr/_pipelines/pp_chatocrv4_doc.py
@@ -0,0 +1,745 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .._utils.cli import (
+    get_subcommand_args,
+    str2bool,
+)
+from .base import PaddleXPipelineWrapper, PipelineCLISubcommandExecutor
+from .utils import create_config_from_structure
+
+
+class PPChatOCRv4Doc(PaddleXPipelineWrapper):
+    def __init__(
+        self,
+        layout_detection_model_name=None,
+        layout_detection_model_dir=None,
+        doc_orientation_classify_model_name=None,
+        doc_orientation_classify_model_dir=None,
+        doc_unwarping_model_name=None,
+        doc_unwarping_model_dir=None,
+        text_detection_model_name=None,
+        text_detection_model_dir=None,
+        textline_orientation_model_name=None,
+        textline_orientation_model_dir=None,
+        textline_orientation_batch_size=None,
+        text_recognition_model_name=None,
+        text_recognition_model_dir=None,
+        text_recognition_batch_size=None,
+        table_structure_recognition_model_name=None,
+        table_structure_recognition_model_dir=None,
+        seal_text_detection_model_name=None,
+        seal_text_detection_model_dir=None,
+        seal_text_recognition_model_name=None,
+        seal_text_recognition_model_dir=None,
+        seal_text_recognition_batch_size=None,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_textline_orientation=None,
+        use_seal_recognition=None,
+        use_table_recognition=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        text_det_limit_side_len=None,
+        text_det_limit_type=None,
+        text_det_thresh=None,
+        text_det_box_thresh=None,
+        text_det_unclip_ratio=None,
+        text_rec_score_thresh=None,
+        seal_det_limit_side_len=None,
+        seal_det_limit_type=None,
+        seal_det_thresh=None,
+        seal_det_box_thresh=None,
+        seal_det_unclip_ratio=None,
+        seal_rec_score_thresh=None,
+        retriever_config=None,
+        mllm_chat_bot_config=None,
+        chat_bot_config=None,
+        **kwargs,
+    ):
+        params = locals().copy()
+        params.pop("self")
+        params.pop("kwargs")
+        self._params = params
+
+        super().__init__(**kwargs)
+
+    @property
+    def _paddlex_pipeline_name(self):
+        return "PP-ChatOCRv4-doc"
+
+    def save_vector(self, vector_info, save_path, retriever_config=None):
+        return self.paddlex_pipeline.save_vector(
+            vector_info=vector_info,
+            save_path=save_path,
+            retriever_config=retriever_config,
+        )
+
+    def load_vector(self, data_path, retriever_config=None):
+        return self.paddlex_pipeline.load_vector(
+            data_path=data_path, retriever_config=retriever_config
+        )
+
+    def load_visual_info_list(self, data_path):
+        return self.paddlex_pipeline.load_visual_info_list(data_path=data_path)
+
+    def save_visual_info_list(self, visual_info, save_path):
+        return self.paddlex_pipeline.save_visual_info_list(
+            visual_info=visual_info, save_path=save_path
+        )
+
+    def visual_predict_iter(
+        self,
+        input,
+        *,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_textline_orientation=None,
+        use_seal_recognition=None,
+        use_table_recognition=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        text_det_limit_side_len=None,
+        text_det_limit_type=None,
+        text_det_thresh=None,
+        text_det_box_thresh=None,
+        text_det_unclip_ratio=None,
+        text_rec_score_thresh=None,
+        seal_det_limit_side_len=None,
+        seal_det_limit_type=None,
+        seal_det_thresh=None,
+        seal_det_box_thresh=None,
+        seal_det_unclip_ratio=None,
+        seal_rec_score_thresh=None,
+        **kwargs,
+    ):
+        return self.paddlex_pipeline.visual_predict(
+            input,
+            use_doc_orientation_classify=use_doc_orientation_classify,
+            use_doc_unwarping=use_doc_unwarping,
+            use_textline_orientation=use_textline_orientation,
+            use_seal_recognition=use_seal_recognition,
+            use_table_recognition=use_table_recognition,
+            layout_threshold=layout_threshold,
+            layout_nms=layout_nms,
+            layout_unclip_ratio=layout_unclip_ratio,
+            layout_merge_bboxes_mode=layout_merge_bboxes_mode,
+            text_det_limit_side_len=text_det_limit_side_len,
+            text_det_limit_type=text_det_limit_type,
+            text_det_thresh=text_det_thresh,
+            text_det_box_thresh=text_det_box_thresh,
+            text_det_unclip_ratio=text_det_unclip_ratio,
+            text_rec_score_thresh=text_rec_score_thresh,
+            seal_det_limit_side_len=seal_det_limit_side_len,
+            seal_det_limit_type=seal_det_limit_type,
+            seal_det_thresh=seal_det_thresh,
+            seal_det_box_thresh=seal_det_box_thresh,
+            seal_det_unclip_ratio=seal_det_unclip_ratio,
+            seal_rec_score_thresh=seal_rec_score_thresh,
+            **kwargs,
+        )
+
+    def visual_predict(
+        self,
+        input,
+        *,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_textline_orientation=None,
+        use_seal_recognition=None,
+        use_table_recognition=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        text_det_limit_side_len=None,
+        text_det_limit_type=None,
+        text_det_thresh=None,
+        text_det_box_thresh=None,
+        text_det_unclip_ratio=None,
+        text_rec_score_thresh=None,
+        seal_det_limit_side_len=None,
+        seal_det_limit_type=None,
+        seal_det_thresh=None,
+        seal_det_box_thresh=None,
+        seal_det_unclip_ratio=None,
+        seal_rec_score_thresh=None,
+        **kwargs,
+    ):
+        return list(
+            self.visual_predict_iter(
+                input,
+                use_doc_orientation_classify=use_doc_orientation_classify,
+                use_doc_unwarping=use_doc_unwarping,
+                use_textline_orientation=use_textline_orientation,
+                use_seal_recognition=use_seal_recognition,
+                use_table_recognition=use_table_recognition,
+                layout_threshold=layout_threshold,
+                layout_nms=layout_nms,
+                layout_unclip_ratio=layout_unclip_ratio,
+                layout_merge_bboxes_mode=layout_merge_bboxes_mode,
+                text_det_limit_side_len=text_det_limit_side_len,
+                text_det_limit_type=text_det_limit_type,
+                text_det_thresh=text_det_thresh,
+                text_det_box_thresh=text_det_box_thresh,
+                text_det_unclip_ratio=text_det_unclip_ratio,
+                text_rec_score_thresh=text_rec_score_thresh,
+                seal_det_limit_side_len=seal_det_limit_side_len,
+                seal_det_limit_type=seal_det_limit_type,
+                seal_det_thresh=seal_det_thresh,
+                seal_det_box_thresh=seal_det_box_thresh,
+                seal_det_unclip_ratio=seal_det_unclip_ratio,
+                seal_rec_score_thresh=seal_rec_score_thresh,
+                **kwargs,
+            )
+        )
+
+    def build_vector(
+        self,
+        visual_info,
+        *,
+        min_characters=3500,
+        block_size=300,
+        flag_save_bytes_vector=False,
+        retriever_config=None,
+    ):
+        return self.paddlex_pipeline.build_vector(
+            visual_info,
+            min_characters=min_characters,
+            block_size=block_size,
+            flag_save_bytes_vector=flag_save_bytes_vector,
+            retriever_config=retriever_config,
+        )
+
+    def mllm_pred(self, input, key_list, *, mllm_chat_bot_config=None):
+        return self.paddlex_pipeline.mllm_pred(
+            input,
+            key_list,
+            mllm_chat_bot_config=mllm_chat_bot_config,
+        )
+
+    def chat(
+        self,
+        key_list,
+        visual_info,
+        *,
+        use_vector_retrieval=True,
+        vector_info=None,
+        min_characters=3500,
+        text_task_description=None,
+        text_output_format=None,
+        text_rules_str=None,
+        text_few_shot_demo_text_content=None,
+        text_few_shot_demo_key_value_list=None,
+        table_task_description=None,
+        table_output_format=None,
+        table_rules_str=None,
+        table_few_shot_demo_text_content=None,
+        table_few_shot_demo_key_value_list=None,
+        mllm_predict_info=None,
+        mllm_integration_strategy="integration",
+        chat_bot_config=None,
+        retriever_config=None,
+    ):
+        return self.paddlex_pipeline.chat(
+            key_list,
+            visual_info,
+            use_vector_retrieval=use_vector_retrieval,
+            vector_info=vector_info,
+            min_characters=min_characters,
+            text_task_description=text_task_description,
+            text_output_format=text_output_format,
+            text_rules_str=text_rules_str,
+            text_few_shot_demo_text_content=text_few_shot_demo_text_content,
+            text_few_shot_demo_key_value_list=text_few_shot_demo_key_value_list,
+            table_task_description=table_task_description,
+            table_output_format=table_output_format,
+            table_rules_str=table_rules_str,
+            table_few_shot_demo_text_content=table_few_shot_demo_text_content,
+            table_few_shot_demo_key_value_list=table_few_shot_demo_key_value_list,
+            mllm_predict_info=mllm_predict_info,
+            mllm_integration_strategy=mllm_integration_strategy,
+            chat_bot_config=chat_bot_config,
+            retriever_config=retriever_config,
+        )
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return PPChatOCRv4DocCLISubcommandExecutor()
+
+    def _get_paddlex_config_overrides(self):
+        STRUCTURE = {
+            "SubPipelines.LayoutParser.SubModules.LayoutDetection.model_name": self._params[
+                "layout_detection_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubModules.LayoutDetection.model_dir": self._params[
+                "layout_detection_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify.model_name": self._params[
+                "doc_orientation_classify_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify.model_dir": self._params[
+                "doc_orientation_classify_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_name": self._params[
+                "doc_unwarping_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_dir": self._params[
+                "doc_unwarping_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextDetection.model_name": self._params[
+                "text_detection_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextDetection.model_dir": self._params[
+                "text_detection_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextLineOrientation.model_name": self._params[
+                "textline_orientation_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextLineOrientation.model_dir": self._params[
+                "textline_orientation_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextLineOrientation.batch_size": self._params[
+                "textline_orientation_batch_size"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextRecognition.model_name": self._params[
+                "text_recognition_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextRecognition.model_dir": self._params[
+                "text_recognition_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextRecognition.batch_size": self._params[
+                "text_recognition_batch_size"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubModules.TableStructureRecognition.model_name": self._params[
+                "table_structure_recognition_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubModules.TableStructureRecognition.model_dir": self._params[
+                "table_structure_recognition_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.model_name": self._params[
+                "seal_text_detection_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.model_dir": self._params[
+                "seal_text_detection_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextRecognition.model_name": self._params[
+                "seal_text_recognition_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextRecognition.model_dir": self._params[
+                "seal_text_recognition_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextRecognition.batch_size": self._params[
+                "seal_text_recognition_batch_size"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.DocPreprocessor.use_doc_orientation_classify": self._params[
+                "use_doc_orientation_classify"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.DocPreprocessor.use_doc_unwarping": self._params[
+                "use_doc_unwarping"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.use_textline_orientation": self._params[
+                "use_textline_orientation"
+            ],
+            "SubPipelines.LayoutParser.use_doc_preprocessor": self._params[
+                "use_doc_orientation_classify"
+            ]
+            or self._params["use_doc_unwarping"],
+            "SubPipelines.LayoutParser.use_seal_recognition": self._params[
+                "use_seal_recognition"
+            ],
+            "SubPipelines.LayoutParser.use_table_recognition": self._params[
+                "use_table_recognition"
+            ],
+            "SubPipelines.LayoutParser.SubModules.LayoutDetection.threshold": self._params[
+                "layout_threshold"
+            ],
+            "SubPipelines.LayoutParser.SubModules.LayoutDetection.nms": self._params[
+                "layout_nms"
+            ],
+            "SubPipelines.LayoutParser.SubModules.LayoutDetection.unclip_ratio": self._params[
+                "layout_unclip_ratio"
+            ],
+            "SubPipelines.LayoutParser.SubModules.LayoutDetection.merge_bboxes_mode": self._params[
+                "layout_merge_bboxes_mode"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextDetection.limit_side_len": self._params[
+                "text_det_limit_side_len"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextDetection.limit_type": self._params[
+                "text_det_limit_type"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextDetection.thresh": self._params[
+                "text_det_thresh"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextDetection.box_thresh": self._params[
+                "text_det_box_thresh"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextDetection.unclip_ratio": self._params[
+                "text_det_unclip_ratio"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextRecognition.score_thresh": self._params[
+                "text_rec_score_thresh"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.limit_side_len": self._params[
+                "text_det_limit_side_len"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.limit_type": self._params[
+                "seal_det_limit_type"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.thresh": self._params[
+                "seal_det_thresh"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.box_thresh": self._params[
+                "seal_det_box_thresh"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.unclip_ratio": self._params[
+                "seal_det_unclip_ratio"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextRecognition.score_thresh": self._params[
+                "seal_rec_score_thresh"
+            ],
+            "SubModules.LLM_Retriever": self._params["retriever_config"],
+            "SubModules.MLLM_Chat": self._params["mllm_chat_bot_config"],
+            "SubModules.LLM_Chat": self._params["chat_bot_config"],
+        }
+        return create_config_from_structure(STRUCTURE)
+
+
+class PPChatOCRv4DocCLISubcommandExecutor(PipelineCLISubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "pp_chatocrv4_doc"
+
+    def _update_subparser(self, subparser):
+        subparser.add_argument(
+            "-i",
+            "--input",
+            type=str,
+            required=True,
+            help="Input path or URL.",
+        )
+        subparser.add_argument(
+            "-k",
+            "--keys",
+            type=str,
+            nargs="+",
+            required=True,
+            metavar="KEY",
+            help="Keys use for information extraction.",
+        )
+        subparser.add_argument(
+            "--save_path",
+            type=str,
+            help="Path to the output directory.",
+        )
+
+        subparser.add_argument(
+            "--invoke_mllm",
+            type=str2bool,
+            default=False,
+            help="Whether to invoke the multimodal large language model.",
+        )
+
+        subparser.add_argument(
+            "--layout_detection_model_name",
+            type=str,
+            help="Name of the layout detection model.",
+        )
+        subparser.add_argument(
+            "--layout_detection_model_dir",
+            type=str,
+            help="Path to the layout detection model directory.",
+        )
+        subparser.add_argument(
+            "--doc_orientation_classify_model_name",
+            type=str,
+            help="Name of the document image orientation classification model.",
+        )
+        subparser.add_argument(
+            "--doc_orientation_classify_model_dir",
+            type=str,
+            help="Path to the document image orientation classification model directory.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_name",
+            type=str,
+            help="Name of the text image unwarping model.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_dir",
+            type=str,
+            help="Path to the image unwarping model directory.",
+        )
+        subparser.add_argument(
+            "--text_detection_model_name",
+            type=str,
+            help="Name of the text detection model.",
+        )
+        subparser.add_argument(
+            "--text_detection_model_dir",
+            type=str,
+            help="Path to the text detection model directory.",
+        )
+        subparser.add_argument(
+            "--textline_orientation_model_name",
+            type=str,
+            help="Name of the text line orientation classification model.",
+        )
+        subparser.add_argument(
+            "--textline_orientation_model_dir",
+            type=str,
+            help="Path to the text line orientation classification model directory.",
+        )
+        subparser.add_argument(
+            "--textline_orientation_batch_size",
+            type=int,
+            help="Batch size for the text line orientation classification model.",
+        )
+        subparser.add_argument(
+            "--text_recognition_model_name",
+            type=str,
+            help="Name of the text recognition model.",
+        )
+        subparser.add_argument(
+            "--text_recognition_model_dir",
+            type=str,
+            help="Path to the text recognition model directory.",
+        )
+        subparser.add_argument(
+            "--text_recognition_batch_size",
+            type=int,
+            help="Batch size for the text recognition model.",
+        )
+        subparser.add_argument(
+            "--table_structure_recognition_model_name",
+            type=str,
+            help="Name of the table structure recognition model.",
+        )
+        subparser.add_argument(
+            "--table_structure_recognition_model_dir",
+            type=str,
+            help="Path to the table structure recognition model directory.",
+        )
+        subparser.add_argument(
+            "--seal_text_detection_model_name",
+            type=str,
+            help="Name of the seal text detection model.",
+        )
+        subparser.add_argument(
+            "--seal_text_detection_model_dir",
+            type=str,
+            help="Path to the seal text detection model directory.",
+        )
+        subparser.add_argument(
+            "--seal_text_recognition_model_name",
+            type=str,
+            help="Name of the seal text recognition model.",
+        )
+        subparser.add_argument(
+            "--seal_text_recognition_model_dir",
+            type=str,
+            help="Path to the seal text recognition model directory.",
+        )
+        subparser.add_argument(
+            "--seal_text_recognition_batch_size",
+            type=int,
+            help="Batch size for the seal text recognition model.",
+        )
+        subparser.add_argument(
+            "--use_doc_orientation_classify",
+            type=str2bool,
+            help="Whether to use document image orientation classification.",
+        )
+        subparser.add_argument(
+            "--use_doc_unwarping",
+            type=str2bool,
+            help="Whether to use text image unwarping.",
+        )
+        subparser.add_argument(
+            "--use_textline_orientation",
+            type=str2bool,
+            help="Whether to use text line orientation classification.",
+        )
+        subparser.add_argument(
+            "--use_seal_recognition",
+            type=str2bool,
+            help="Whether to use seal recognition.",
+        )
+        subparser.add_argument(
+            "--use_table_recognition",
+            type=str2bool,
+            help="Whether to use table recognition.",
+        )
+        # TODO: Support dict and list types
+        subparser.add_argument(
+            "--layout_threshold",
+            type=float,
+            help="Score threshold for the layout detection model.",
+        )
+        subparser.add_argument(
+            "--layout_nms",
+            type=str2bool,
+            help="Whether to use NMS in layout detection.",
+        )
+        subparser.add_argument(
+            "--layout_unclip_ratio",
+            type=float,
+            help="Expansion coefficient for layout detection.",
+        )
+        subparser.add_argument(
+            "--layout_merge_bboxes_mode",
+            type=str,
+            help="Overlapping box filtering method.",
+        )
+        subparser.add_argument(
+            "--text_det_limit_side_len",
+            type=int,
+            help="This sets a limit on the side length of the input image for the text detection model.",
+        )
+        subparser.add_argument(
+            "--text_det_limit_type",
+            type=str,
+            help="This determines how the side length limit is applied to the input image before feeding it into the text deteciton model.",
+        )
+        subparser.add_argument(
+            "--text_det_thresh",
+            type=float,
+            help="Detection pixel threshold for the text detection model. Pixels with scores greater than this threshold in the output probability map are considered text pixels.",
+        )
+        subparser.add_argument(
+            "--text_det_box_thresh",
+            type=float,
+            help="Detection box threshold for the text detection model. A detection result is considered a text region if the average score of all pixels within the border of the result is greater than this threshold.",
+        )
+        subparser.add_argument(
+            "--text_det_unclip_ratio",
+            type=float,
+            help="Text detection expansion coefficient, which expands the text region using this method. The larger the value, the larger the expansion area.",
+        )
+        subparser.add_argument(
+            "--text_rec_score_thresh",
+            type=float,
+            help="Text recognition threshold used in general OCR. Text results with scores greater than this threshold are retained.",
+        )
+        subparser.add_argument(
+            "--seal_det_limit_side_len",
+            type=int,
+            help="This sets a limit on the side length of the input image for the seal text detection model.",
+        )
+        subparser.add_argument(
+            "--seal_det_limit_type",
+            type=str,
+            help="This determines how the side length limit is applied to the input image before feeding it into the seal text deteciton model.",
+        )
+        subparser.add_argument(
+            "--seal_det_thresh",
+            type=float,
+            help="Detection pixel threshold for the seal text detection model. Pixels with scores greater than this threshold in the output probability map are considered text pixels.",
+        )
+        subparser.add_argument(
+            "--seal_det_box_thresh",
+            type=float,
+            help="Detection box threshold for the seal text detection model. A detection result is considered a text region if the average score of all pixels within the border of the result is greater than this threshold.",
+        )
+        subparser.add_argument(
+            "--seal_det_unclip_ratio",
+            type=float,
+            help="Seal text detection expansion coefficient, which expands the text region using this method. The larger the value, the larger the expansion area.",
+        )
+        subparser.add_argument(
+            "--seal_rec_score_thresh",
+            type=float,
+            help="Seal text recognition threshold. Text results with scores greater than this threshold are retained.",
+        )
+
+        # FIXME: Passing API key through CLI is not secure; consider using
+        # environment variables.
+        subparser.add_argument(
+            "--qianfan_api_key",
+            type=str,
+            help="Configuration for the embedding model.",
+        )
+        subparser.add_argument(
+            "--pp_docbee_base_url",
+            type=str,
+            help="Configuration for the multimodal large language model.",
+        )
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+        input = params.pop("input")
+        keys = params.pop("keys")
+        save_path = params.pop("save_path")
+        invoke_mllm = params.pop("invoke_mllm")
+        qianfan_api_key = params.pop("qianfan_api_key")
+        if qianfan_api_key is not None:
+            params["retriever_config"] = {
+                "module_name": "retriever",
+                "model_name": "embedding-v1",
+                "base_url": "https://qianfan.baidubce.com/v2",
+                "api_type": "qianfan",
+                "api_key": qianfan_api_key,
+            }
+            params["chat_bot_config"] = {
+                "module_name": "chat_bot",
+                "model_name": "ernie-3.5-8k",
+                "base_url": "https://qianfan.baidubce.com/v2",
+                "api_type": "openai",
+                "api_key": qianfan_api_key,
+            }
+        pp_docbee_base_url = params.pop("pp_docbee_base_url")
+        if pp_docbee_base_url is not None:
+            params["mllm_chat_bot_config"] = {
+                "module_name": "chat_bot",
+                "model_name": "PP-DocBee",
+                # PaddleX requires endpoints such as ".../chat/completions",
+                # which, as the parameter name suggests, are not base URLs.
+                "base_url": pp_docbee_base_url,
+                "api_type": "openai",
+                "api_key": "fake_key",
+            }
+
+        chatocr = PPChatOCRv4Doc(**params)
+
+        result_visual = chatocr.visual_predict_iter(input)
+
+        visual_info_list = []
+        for res in result_visual:
+            visual_info_list.append(res["visual_info"])
+            if save_path:
+                res["layout_parsing_result"].save_all(save_path)
+
+        vector_info = chatocr.build_vector(visual_info_list)
+
+        if invoke_mllm:
+            result_mllm = chatocr.mllm_pred(input, keys)
+            mllm_predict_info = result_mllm["mllm_res"]
+        else:
+            mllm_predict_info = None
+
+        result_chat = chatocr.chat(
+            keys,
+            visual_info_list,
+            vector_info=vector_info,
+            mllm_predict_info=mllm_predict_info,
+        )
+
+        # Print the result to stdout
+        for k, v in result_chat["chat_res"].items():
+            print(f"{k} {v}")
diff --git a/paddleocr/_pipelines/pp_doctranslation.py b/paddleocr/_pipelines/pp_doctranslation.py
new file mode 100644
index 0000000..22b907c
--- /dev/null
+++ b/paddleocr/_pipelines/pp_doctranslation.py
@@ -0,0 +1,945 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .._utils.cli import (
+    get_subcommand_args,
+    str2bool,
+)
+from .._utils.logging import logger
+from .base import PaddleXPipelineWrapper, PipelineCLISubcommandExecutor
+from .utils import create_config_from_structure
+
+
+class PPDocTranslation(PaddleXPipelineWrapper):
+    def __init__(
+        self,
+        layout_detection_model_name=None,
+        layout_detection_model_dir=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        chart_recognition_model_name=None,
+        chart_recognition_model_dir=None,
+        chart_recognition_batch_size=None,
+        region_detection_model_name=None,
+        region_detection_model_dir=None,
+        doc_orientation_classify_model_name=None,
+        doc_orientation_classify_model_dir=None,
+        doc_unwarping_model_name=None,
+        doc_unwarping_model_dir=None,
+        text_detection_model_name=None,
+        text_detection_model_dir=None,
+        text_det_limit_side_len=None,
+        text_det_limit_type=None,
+        text_det_thresh=None,
+        text_det_box_thresh=None,
+        text_det_unclip_ratio=None,
+        textline_orientation_model_name=None,
+        textline_orientation_model_dir=None,
+        textline_orientation_batch_size=None,
+        text_recognition_model_name=None,
+        text_recognition_model_dir=None,
+        text_recognition_batch_size=None,
+        text_rec_score_thresh=None,
+        table_classification_model_name=None,
+        table_classification_model_dir=None,
+        wired_table_structure_recognition_model_name=None,
+        wired_table_structure_recognition_model_dir=None,
+        wireless_table_structure_recognition_model_name=None,
+        wireless_table_structure_recognition_model_dir=None,
+        wired_table_cells_detection_model_name=None,
+        wired_table_cells_detection_model_dir=None,
+        wireless_table_cells_detection_model_name=None,
+        wireless_table_cells_detection_model_dir=None,
+        table_orientation_classify_model_name=None,
+        table_orientation_classify_model_dir=None,
+        seal_text_detection_model_name=None,
+        seal_text_detection_model_dir=None,
+        seal_det_limit_side_len=None,
+        seal_det_limit_type=None,
+        seal_det_thresh=None,
+        seal_det_box_thresh=None,
+        seal_det_unclip_ratio=None,
+        seal_text_recognition_model_name=None,
+        seal_text_recognition_model_dir=None,
+        seal_text_recognition_batch_size=None,
+        seal_rec_score_thresh=None,
+        formula_recognition_model_name=None,
+        formula_recognition_model_dir=None,
+        formula_recognition_batch_size=None,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_textline_orientation=None,
+        use_seal_recognition=None,
+        use_table_recognition=None,
+        use_formula_recognition=None,
+        use_chart_recognition=None,
+        use_region_detection=None,
+        chat_bot_config=None,
+        **kwargs,
+    ):
+        params = locals().copy()
+        params.pop("self")
+        params.pop("kwargs")
+        self._params = params
+
+        super().__init__(**kwargs)
+
+    @property
+    def _paddlex_pipeline_name(self):
+        return "PP-DocTranslation"
+
+    def visual_predict_iter(
+        self,
+        input,
+        *,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_textline_orientation=None,
+        use_seal_recognition=None,
+        use_table_recognition=None,
+        use_formula_recognition=None,
+        use_chart_recognition=None,
+        use_region_detection=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        text_det_limit_side_len=None,
+        text_det_limit_type=None,
+        text_det_thresh=None,
+        text_det_box_thresh=None,
+        text_det_unclip_ratio=None,
+        text_rec_score_thresh=None,
+        seal_det_limit_side_len=None,
+        seal_det_limit_type=None,
+        seal_det_thresh=None,
+        seal_det_box_thresh=None,
+        seal_det_unclip_ratio=None,
+        seal_rec_score_thresh=None,
+        use_wired_table_cells_trans_to_html=False,
+        use_wireless_table_cells_trans_to_html=False,
+        use_table_orientation_classify=True,
+        use_ocr_results_with_table_cells=True,
+        use_e2e_wired_table_rec_model=False,
+        use_e2e_wireless_table_rec_model=True,
+        **kwargs,
+    ):
+        return self.paddlex_pipeline.visual_predict(
+            input,
+            use_doc_orientation_classify=use_doc_orientation_classify,
+            use_doc_unwarping=use_doc_unwarping,
+            use_textline_orientation=use_textline_orientation,
+            use_seal_recognition=use_seal_recognition,
+            use_table_recognition=use_table_recognition,
+            use_formula_recognition=use_formula_recognition,
+            use_chart_recognition=use_chart_recognition,
+            use_region_detection=use_region_detection,
+            layout_threshold=layout_threshold,
+            layout_nms=layout_nms,
+            layout_unclip_ratio=layout_unclip_ratio,
+            layout_merge_bboxes_mode=layout_merge_bboxes_mode,
+            text_det_limit_side_len=text_det_limit_side_len,
+            text_det_limit_type=text_det_limit_type,
+            text_det_thresh=text_det_thresh,
+            text_det_box_thresh=text_det_box_thresh,
+            text_det_unclip_ratio=text_det_unclip_ratio,
+            text_rec_score_thresh=text_rec_score_thresh,
+            seal_det_limit_side_len=seal_det_limit_side_len,
+            seal_det_limit_type=seal_det_limit_type,
+            seal_det_thresh=seal_det_thresh,
+            seal_det_box_thresh=seal_det_box_thresh,
+            seal_det_unclip_ratio=seal_det_unclip_ratio,
+            seal_rec_score_thresh=seal_rec_score_thresh,
+            use_wired_table_cells_trans_to_html=use_wired_table_cells_trans_to_html,
+            use_wireless_table_cells_trans_to_html=use_wireless_table_cells_trans_to_html,
+            use_table_orientation_classify=use_table_orientation_classify,
+            use_ocr_results_with_table_cells=use_ocr_results_with_table_cells,
+            use_e2e_wired_table_rec_model=use_e2e_wired_table_rec_model,
+            use_e2e_wireless_table_rec_model=use_e2e_wireless_table_rec_model,
+            **kwargs,
+        )
+
+    def visual_predict(
+        self,
+        input,
+        *,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_textline_orientation=None,
+        use_seal_recognition=None,
+        use_table_recognition=None,
+        use_formula_recognition=None,
+        use_chart_recognition=None,
+        use_region_detection=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        text_det_limit_side_len=None,
+        text_det_limit_type=None,
+        text_det_thresh=None,
+        text_det_box_thresh=None,
+        text_det_unclip_ratio=None,
+        text_rec_score_thresh=None,
+        seal_det_limit_side_len=None,
+        seal_det_limit_type=None,
+        seal_det_thresh=None,
+        seal_det_box_thresh=None,
+        seal_det_unclip_ratio=None,
+        seal_rec_score_thresh=None,
+        use_wired_table_cells_trans_to_html=False,
+        use_wireless_table_cells_trans_to_html=False,
+        use_table_orientation_classify=True,
+        use_ocr_results_with_table_cells=True,
+        use_e2e_wired_table_rec_model=False,
+        use_e2e_wireless_table_rec_model=True,
+        **kwargs,
+    ):
+        return list(
+            self.visual_predict_iter(
+                input,
+                use_doc_orientation_classify=use_doc_orientation_classify,
+                use_doc_unwarping=use_doc_unwarping,
+                use_textline_orientation=use_textline_orientation,
+                use_seal_recognition=use_seal_recognition,
+                use_table_recognition=use_table_recognition,
+                use_formula_recognition=use_formula_recognition,
+                use_chart_recognition=use_chart_recognition,
+                use_region_detection=use_region_detection,
+                layout_threshold=layout_threshold,
+                layout_nms=layout_nms,
+                layout_unclip_ratio=layout_unclip_ratio,
+                layout_merge_bboxes_mode=layout_merge_bboxes_mode,
+                text_det_limit_side_len=text_det_limit_side_len,
+                text_det_limit_type=text_det_limit_type,
+                text_det_thresh=text_det_thresh,
+                text_det_box_thresh=text_det_box_thresh,
+                text_det_unclip_ratio=text_det_unclip_ratio,
+                text_rec_score_thresh=text_rec_score_thresh,
+                seal_det_limit_side_len=seal_det_limit_side_len,
+                seal_det_limit_type=seal_det_limit_type,
+                seal_det_thresh=seal_det_thresh,
+                seal_det_box_thresh=seal_det_box_thresh,
+                seal_det_unclip_ratio=seal_det_unclip_ratio,
+                seal_rec_score_thresh=seal_rec_score_thresh,
+                use_wired_table_cells_trans_to_html=use_wired_table_cells_trans_to_html,
+                use_wireless_table_cells_trans_to_html=use_wireless_table_cells_trans_to_html,
+                use_table_orientation_classify=use_table_orientation_classify,
+                use_ocr_results_with_table_cells=use_ocr_results_with_table_cells,
+                use_e2e_wired_table_rec_model=use_e2e_wired_table_rec_model,
+                use_e2e_wireless_table_rec_model=use_e2e_wireless_table_rec_model,
+                **kwargs,
+            )
+        )
+
+    def translate_iter(
+        self,
+        ori_md_info_list,
+        *,
+        target_language="zh",
+        chunk_size=5000,
+        task_description=None,
+        output_format=None,
+        rules_str=None,
+        few_shot_demo_text_content=None,
+        few_shot_demo_key_value_list=None,
+        glossary=None,
+        llm_request_interval=0.0,
+        chat_bot_config=None,
+        **kwargs,
+    ):
+        return self.paddlex_pipeline.translate(
+            ori_md_info_list,
+            target_language=target_language,
+            chunk_size=chunk_size,
+            task_description=task_description,
+            output_format=output_format,
+            rules_str=rules_str,
+            few_shot_demo_text_content=few_shot_demo_text_content,
+            few_shot_demo_key_value_list=few_shot_demo_key_value_list,
+            glossary=glossary,
+            llm_request_interval=llm_request_interval,
+            chat_bot_config=chat_bot_config,
+            **kwargs,
+        )
+
+    def translate(
+        self,
+        ori_md_info_list,
+        *,
+        target_language="zh",
+        chunk_size=5000,
+        task_description=None,
+        output_format=None,
+        rules_str=None,
+        few_shot_demo_text_content=None,
+        few_shot_demo_key_value_list=None,
+        glossary=None,
+        llm_request_interval=0.0,
+        chat_bot_config=None,
+        **kwargs,
+    ):
+        return list(
+            self.translate_iter(
+                ori_md_info_list,
+                target_language=target_language,
+                chunk_size=chunk_size,
+                task_description=task_description,
+                output_format=output_format,
+                rules_str=rules_str,
+                few_shot_demo_text_content=few_shot_demo_text_content,
+                few_shot_demo_key_value_list=few_shot_demo_key_value_list,
+                glossary=glossary,
+                llm_request_interval=llm_request_interval,
+                chat_bot_config=chat_bot_config,
+                **kwargs,
+            )
+        )
+
+    def load_from_markdown(self, input):
+        return self.paddlex_pipeline.load_from_markdown(input)
+
+    def concatenate_markdown_pages(self, markdown_list):
+        return self.paddlex_pipeline.concatenate_markdown_pages(markdown_list)
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return PPDocTranslationCLISubcommandExecutor()
+
+    def _get_paddlex_config_overrides(self):
+        # HACK: We should consider reducing duplication.
+        STRUCTURE = {
+            "SubPipelines.LayoutParser.SubPipelines.DocPreprocessor.use_doc_orientation_classify": self._params[
+                "use_doc_orientation_classify"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.DocPreprocessor.use_doc_unwarping": self._params[
+                "use_doc_unwarping"
+            ],
+            "SubPipelines.LayoutParser.use_doc_preprocessor": self._params[
+                "use_doc_orientation_classify"
+            ]
+            or self._params["use_doc_unwarping"],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.use_textline_orientation": self._params[
+                "use_textline_orientation"
+            ],
+            "SubPipelines.LayoutParser.use_seal_recognition": self._params[
+                "use_seal_recognition"
+            ],
+            "SubPipelines.LayoutParser.use_table_recognition": self._params[
+                "use_table_recognition"
+            ],
+            "SubPipelines.LayoutParser.use_formula_recognition": self._params[
+                "use_formula_recognition"
+            ],
+            "SubPipelines.LayoutParser.use_chart_recognition": self._params[
+                "use_chart_recognition"
+            ],
+            "SubPipelines.LayoutParser.use_region_detection": self._params[
+                "use_region_detection"
+            ],
+            "SubPipelines.LayoutParser.SubModules.LayoutDetection.model_name": self._params[
+                "layout_detection_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubModules.LayoutDetection.model_dir": self._params[
+                "layout_detection_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubModules.LayoutDetection.threshold": self._params[
+                "layout_threshold"
+            ],
+            "SubPipelines.LayoutParser.SubModules.LayoutDetection.layout_nms": self._params[
+                "layout_nms"
+            ],
+            "SubPipelines.LayoutParser.SubModules.LayoutDetection.layout_unclip_ratio": self._params[
+                "layout_unclip_ratio"
+            ],
+            "SubPipelines.LayoutParser.SubModules.LayoutDetection.layout_merge_bboxes_mode": self._params[
+                "layout_merge_bboxes_mode"
+            ],
+            "SubPipelines.LayoutParser.SubModules.ChartRecognition.model_name": self._params[
+                "chart_recognition_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubModules.ChartRecognition.model_dir": self._params[
+                "chart_recognition_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubModules.ChartRecognition.batch_size": self._params[
+                "chart_recognition_batch_size"
+            ],
+            "SubPipelines.LayoutParser.SubModules.RegionDetection.model_name": self._params[
+                "region_detection_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubModules.RegionDetection.model_dir": self._params[
+                "region_detection_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify.model_name": self._params[
+                "doc_orientation_classify_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify.model_dir": self._params[
+                "doc_orientation_classify_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_name": self._params[
+                "doc_unwarping_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_dir": self._params[
+                "doc_unwarping_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextDetection.model_name": self._params[
+                "text_detection_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextDetection.model_dir": self._params[
+                "text_detection_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextDetection.limit_side_len": self._params[
+                "text_det_limit_side_len"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextDetection.limit_type": self._params[
+                "text_det_limit_type"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextDetection.thresh": self._params[
+                "text_det_thresh"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextDetection.box_thresh": self._params[
+                "text_det_box_thresh"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextDetection.unclip_ratio": self._params[
+                "text_det_unclip_ratio"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextLineOrientation.model_name": self._params[
+                "textline_orientation_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextLineOrientation.model_dir": self._params[
+                "textline_orientation_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextLineOrientation.batch_size": self._params[
+                "textline_orientation_batch_size"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextRecognition.model_name": self._params[
+                "text_recognition_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextRecognition.model_dir": self._params[
+                "text_recognition_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextRecognition.batch_size": self._params[
+                "text_recognition_batch_size"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.GeneralOCR.SubModules.TextRecognition.score_thresh": self._params[
+                "text_rec_score_thresh"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubModules.TableClassification.model_name": self._params[
+                "table_classification_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubModules.TableClassification.model_dir": self._params[
+                "table_classification_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubModules.WiredTableStructureRecognition.model_name": self._params[
+                "wired_table_structure_recognition_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubModules.WiredTableStructureRecognition.model_dir": self._params[
+                "wired_table_structure_recognition_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubModules.WirelessTableStructureRecognition.model_name": self._params[
+                "wireless_table_structure_recognition_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubModules.WirelessTableStructureRecognition.model_dir": self._params[
+                "wireless_table_structure_recognition_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubModules.WiredTableCellsDetection.model_name": self._params[
+                "wired_table_cells_detection_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubModules.WiredTableCellsDetection.model_dir": self._params[
+                "wired_table_cells_detection_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubModules.WirelessTableCellsDetection.model_name": self._params[
+                "wireless_table_cells_detection_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubModules.WirelessTableCellsDetection.model_dir": self._params[
+                "wireless_table_cells_detection_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubModules.TableOrientationClassify.model_name": self._params[
+                "table_orientation_classify_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubModules.TableOrientationClassify.model_dir": self._params[
+                "table_orientation_classify_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextDetection.model_name": self._params[
+                "text_detection_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextDetection.model_dir": self._params[
+                "text_detection_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextDetection.limit_side_len": self._params[
+                "text_det_limit_side_len"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextDetection.limit_type": self._params[
+                "text_det_limit_type"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextDetection.thresh": self._params[
+                "text_det_thresh"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextDetection.box_thresh": self._params[
+                "text_det_box_thresh"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextDetection.unclip_ratio": self._params[
+                "text_det_unclip_ratio"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextLineOrientation.model_name": self._params[
+                "textline_orientation_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextLineOrientation.model_dir": self._params[
+                "textline_orientation_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextLineOrientation.batch_size": self._params[
+                "textline_orientation_batch_size"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextRecognition.model_name": self._params[
+                "text_recognition_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextRecognition.model_dir": self._params[
+                "text_recognition_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextRecognition.batch_size": self._params[
+                "text_recognition_batch_size"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextRecognition.score_thresh": self._params[
+                "text_rec_score_thresh"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.model_name": self._params[
+                "seal_text_detection_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.model_dir": self._params[
+                "seal_text_detection_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.limit_side_len": self._params[
+                "text_det_limit_side_len"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.limit_type": self._params[
+                "seal_det_limit_type"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.thresh": self._params[
+                "seal_det_thresh"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.box_thresh": self._params[
+                "seal_det_box_thresh"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.unclip_ratio": self._params[
+                "seal_det_unclip_ratio"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextRecognition.model_name": self._params[
+                "seal_text_recognition_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextRecognition.model_dir": self._params[
+                "seal_text_recognition_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextRecognition.batch_size": self._params[
+                "seal_text_recognition_batch_size"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.FormulaRecognition.SubModules.FormulaRecognition.model_name": self._params[
+                "formula_recognition_model_name"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.FormulaRecognition.SubModules.FormulaRecognition.model_dir": self._params[
+                "formula_recognition_model_dir"
+            ],
+            "SubPipelines.LayoutParser.SubPipelines.FormulaRecognition.SubModules.FormulaRecognition.batch_size": self._params[
+                "formula_recognition_batch_size"
+            ],
+            "SubModules.LLM_Chat": self._params["chat_bot_config"],
+        }
+        return create_config_from_structure(STRUCTURE)
+
+
+class PPDocTranslationCLISubcommandExecutor(PipelineCLISubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "pp_doctranslation"
+
+    def _update_subparser(self, subparser):
+        subparser.add_argument(
+            "-i",
+            "--input",
+            type=str,
+            required=True,
+            help="Input path or URL.",
+        )
+        subparser.add_argument(
+            "--save_path",
+            type=str,
+            help="Path to the output directory.",
+        )
+
+        subparser.add_argument(
+            "--target_language",
+            type=str,
+            default="zh",
+            help="Target language.",
+        )
+
+        subparser.add_argument(
+            "--layout_detection_model_name",
+            type=str,
+            help="Name of the layout detection model.",
+        )
+        subparser.add_argument(
+            "--layout_detection_model_dir",
+            type=str,
+            help="Path to the layout detection model directory.",
+        )
+        subparser.add_argument(
+            "--layout_threshold",
+            type=float,
+            help="Score threshold for the layout detection model.",
+        )
+        subparser.add_argument(
+            "--layout_nms",
+            type=str2bool,
+            help="Whether to use NMS in layout detection.",
+        )
+        subparser.add_argument(
+            "--layout_unclip_ratio",
+            type=float,
+            help="Expansion coefficient for layout detection.",
+        )
+        subparser.add_argument(
+            "--layout_merge_bboxes_mode",
+            type=str,
+            help="Overlapping box filtering method.",
+        )
+
+        subparser.add_argument(
+            "--chart_recognition_model_name",
+            type=str,
+            help="Name of the chart recognition model.",
+        )
+        subparser.add_argument(
+            "--chart_recognition_model_dir",
+            type=str,
+            help="Path to the chart recognition model directory.",
+        )
+        subparser.add_argument(
+            "--chart_recognition_batch_size",
+            type=int,
+            help="Batch size for the chart recognition model.",
+        )
+
+        subparser.add_argument(
+            "--region_detection_model_name",
+            type=str,
+            help="Name of the region detection model.",
+        )
+        subparser.add_argument(
+            "--region_detection_model_dir",
+            type=str,
+            help="Path to the region detection model directory.",
+        )
+
+        subparser.add_argument(
+            "--doc_orientation_classify_model_name",
+            type=str,
+            help="Name of the document image orientation classification model.",
+        )
+        subparser.add_argument(
+            "--doc_orientation_classify_model_dir",
+            type=str,
+            help="Path to the document image orientation classification model directory.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_name",
+            type=str,
+            help="Name of the text image unwarping model.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_dir",
+            type=str,
+            help="Path to the image unwarping model directory.",
+        )
+
+        subparser.add_argument(
+            "--text_detection_model_name",
+            type=str,
+            help="Name of the text detection model.",
+        )
+        subparser.add_argument(
+            "--text_detection_model_dir",
+            type=str,
+            help="Path to the text detection model directory.",
+        )
+        subparser.add_argument(
+            "--text_det_limit_side_len",
+            type=int,
+            help="This sets a limit on the side length of the input image for the text detection model.",
+        )
+        subparser.add_argument(
+            "--text_det_limit_type",
+            type=str,
+            help="This determines how the side length limit is applied to the input image before feeding it into the text deteciton model.",
+        )
+        subparser.add_argument(
+            "--text_det_thresh",
+            type=float,
+            help="Detection pixel threshold for the text detection model. Pixels with scores greater than this threshold in the output probability map are considered text pixels.",
+        )
+        subparser.add_argument(
+            "--text_det_box_thresh",
+            type=float,
+            help="Detection box threshold for the text detection model. A detection result is considered a text region if the average score of all pixels within the border of the result is greater than this threshold.",
+        )
+        subparser.add_argument(
+            "--text_det_unclip_ratio",
+            type=float,
+            help="Text detection expansion coefficient, which expands the text region using this method. The larger the value, the larger the expansion area.",
+        )
+        subparser.add_argument(
+            "--textline_orientation_model_name",
+            type=str,
+            help="Name of the text line orientation classification model.",
+        )
+        subparser.add_argument(
+            "--textline_orientation_model_dir",
+            type=str,
+            help="Path to the text line orientation classification directory.",
+        )
+        subparser.add_argument(
+            "--textline_orientation_batch_size",
+            type=int,
+            help="Batch size for the text line orientation classification model.",
+        )
+        subparser.add_argument(
+            "--text_recognition_model_name",
+            type=str,
+            help="Name of the text recognition model.",
+        )
+        subparser.add_argument(
+            "--text_recognition_model_dir",
+            type=str,
+            help="Path to the text recognition model directory.",
+        )
+        subparser.add_argument(
+            "--text_recognition_batch_size",
+            type=int,
+            help="Batch size for the text recognition model.",
+        )
+        subparser.add_argument(
+            "--text_rec_score_thresh",
+            type=float,
+            help="Text recognition threshold used in general OCR. Text results with scores greater than this threshold are retained.",
+        )
+
+        subparser.add_argument(
+            "--table_classification_model_name",
+            type=str,
+            help="Name of the table classification model.",
+        )
+        subparser.add_argument(
+            "--table_classification_model_dir",
+            type=str,
+            help="Path to the table classification model directory.",
+        )
+        subparser.add_argument(
+            "--wired_table_structure_recognition_model_name",
+            type=str,
+            help="Name of the wired table structure recognition model.",
+        )
+        subparser.add_argument(
+            "--wired_table_structure_recognition_model_dir",
+            type=str,
+            help="Path to the wired table structure recognition model directory.",
+        )
+        subparser.add_argument(
+            "--wireless_table_structure_recognition_model_name",
+            type=str,
+            help="Name of the wireless table structure recognition model.",
+        )
+        subparser.add_argument(
+            "--wireless_table_structure_recognition_model_dir",
+            type=str,
+            help="Path to the wired table structure recognition model directory.",
+        )
+        subparser.add_argument(
+            "--wired_table_cells_detection_model_name",
+            type=str,
+            help="Name of the wired table cells detection model.",
+        )
+        subparser.add_argument(
+            "--wired_table_cells_detection_model_dir",
+            type=str,
+            help="Path to the wired table cells detection model directory.",
+        )
+        subparser.add_argument(
+            "--wireless_table_cells_detection_model_name",
+            type=str,
+            help="Name of the wireless table cells detection model.",
+        )
+        subparser.add_argument(
+            "--wireless_table_cells_detection_model_dir",
+            type=str,
+            help="Path to the wireless table cells detection model directory.",
+        )
+
+        subparser.add_argument(
+            "--seal_text_detection_model_name",
+            type=str,
+            help="Name of the seal text detection model.",
+        )
+        subparser.add_argument(
+            "--seal_text_detection_model_dir",
+            type=str,
+            help="Path to the seal text detection model directory.",
+        )
+        subparser.add_argument(
+            "--seal_det_limit_side_len",
+            type=int,
+            help="This sets a limit on the side length of the input image for the seal text detection model.",
+        )
+        subparser.add_argument(
+            "--seal_det_limit_type",
+            type=str,
+            help="This determines how the side length limit is applied to the input image before feeding it into the seal text deteciton model.",
+        )
+        subparser.add_argument(
+            "--seal_det_thresh",
+            type=float,
+            help="Detection pixel threshold for the seal text detection model. Pixels with scores greater than this threshold in the output probability map are considered text pixels.",
+        )
+        subparser.add_argument(
+            "--seal_det_box_thresh",
+            type=float,
+            help="Detection box threshold for the seal text detection model. A detection result is considered a text region if the average score of all pixels within the border of the result is greater than this threshold.",
+        )
+        subparser.add_argument(
+            "--seal_det_unclip_ratio",
+            type=float,
+            help="Seal text detection expansion coefficient, which expands the text region using this method. The larger the value, the larger the expansion area.",
+        )
+        subparser.add_argument(
+            "--seal_text_recognition_model_name",
+            type=str,
+            help="Name of the seal text recognition model.",
+        )
+        subparser.add_argument(
+            "--seal_text_recognition_model_dir",
+            type=str,
+            help="Path to the seal text recognition model directory.",
+        )
+        subparser.add_argument(
+            "--seal_text_recognition_batch_size",
+            type=int,
+            help="Batch size for the seal text recognition model.",
+        )
+        subparser.add_argument(
+            "--seal_rec_score_thresh",
+            type=float,
+            help="Seal text recognition threshold. Text results with scores greater than this threshold are retained.",
+        )
+
+        subparser.add_argument(
+            "--formula_recognition_model_name",
+            type=str,
+            help="Name of the formula recognition model.",
+        )
+        subparser.add_argument(
+            "--formula_recognition_model_dir",
+            type=str,
+            help="Path to the formula recognition model directory.",
+        )
+        subparser.add_argument(
+            "--formula_recognition_batch_size",
+            type=int,
+            help="Batch size for the formula recognition model.",
+        )
+
+        subparser.add_argument(
+            "--use_doc_orientation_classify",
+            type=str2bool,
+            help="Whether to use document image orientation classification.",
+        )
+        subparser.add_argument(
+            "--use_doc_unwarping",
+            type=str2bool,
+            help="Whether to use text image unwarping.",
+        )
+        subparser.add_argument(
+            "--use_textline_orientation",
+            type=str2bool,
+            help="Whether to use text line orientation classification.",
+        )
+        subparser.add_argument(
+            "--use_seal_recognition",
+            type=str2bool,
+            help="Whether to use seal recognition.",
+        )
+        subparser.add_argument(
+            "--use_table_recognition",
+            type=str2bool,
+            help="Whether to use table recognition.",
+        )
+        subparser.add_argument(
+            "--use_formula_recognition",
+            type=str2bool,
+            help="Whether to use formula recognition.",
+        )
+        subparser.add_argument(
+            "--use_chart_recognition",
+            type=str2bool,
+            help="Whether to use chart recognition.",
+        )
+        subparser.add_argument(
+            "--use_region_detection",
+            type=str2bool,
+            help="Whether to use region detection.",
+        )
+
+        # FIXME: Passing API key through CLI is not secure; consider using
+        # environment variables.
+        subparser.add_argument(
+            "--qianfan_api_key",
+            type=str,
+            help="Configuration for the embedding model.",
+        )
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+        input = params.pop("input")
+        target_language = params.pop("target_language")
+        save_path = params.pop("save_path")
+        qianfan_api_key = params.pop("qianfan_api_key")
+        if qianfan_api_key is not None:
+            params["chat_bot_config"] = {
+                "module_name": "chat_bot",
+                "model_name": "ernie-3.5-8k",
+                "base_url": "https://qianfan.baidubce.com/v2",
+                "api_type": "openai",
+                "api_key": qianfan_api_key,
+            }
+
+        chatocr = PPDocTranslation(**params)
+
+        logger.info("Start analyzing images")
+        result_visual = chatocr.visual_predict_iter(input)
+
+        ori_md_info_list = []
+        for res in result_visual:
+            ori_md_info_list.append(res["layout_parsing_result"].markdown)
+            if save_path:
+                res["layout_parsing_result"].save_all(save_path)
+
+        logger.info("Start translation")
+        result_translate = chatocr.translate_iter(
+            ori_md_info_list,
+            target_language=target_language,
+        )
+
+        for res in result_translate:
+            res.print()
+            if save_path:
+                res.save_to_markdown(save_path)
diff --git a/paddleocr/_pipelines/pp_structurev3.py b/paddleocr/_pipelines/pp_structurev3.py
new file mode 100644
index 0000000..eefe7d1
--- /dev/null
+++ b/paddleocr/_pipelines/pp_structurev3.py
@@ -0,0 +1,1032 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import warnings
+from .._utils.cli import (
+    add_simple_inference_args,
+    get_subcommand_args,
+    perform_simple_inference,
+    str2bool,
+)
+from .base import PaddleXPipelineWrapper, PipelineCLISubcommandExecutor
+from .utils import create_config_from_structure
+from ._patch_layout_parsing import apply_patches as _apply_layout_parsing_patches
+
+_apply_layout_parsing_patches()
+
+_SUPPORTED_OCR_VERSIONS = ["PP-OCRv3", "PP-OCRv4", "PP-OCRv5"]
+
+
+class PPStructureV3(PaddleXPipelineWrapper):
+    def __init__(
+        self,
+        layout_detection_model_name=None,
+        layout_detection_model_dir=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        chart_recognition_model_name=None,
+        chart_recognition_model_dir=None,
+        chart_recognition_batch_size=None,
+        region_detection_model_name=None,
+        region_detection_model_dir=None,
+        doc_orientation_classify_model_name=None,
+        doc_orientation_classify_model_dir=None,
+        doc_unwarping_model_name=None,
+        doc_unwarping_model_dir=None,
+        text_detection_model_name=None,
+        text_detection_model_dir=None,
+        text_det_limit_side_len=None,
+        text_det_limit_type=None,
+        text_det_thresh=None,
+        text_det_box_thresh=None,
+        text_det_unclip_ratio=None,
+        textline_orientation_model_name=None,
+        textline_orientation_model_dir=None,
+        textline_orientation_batch_size=None,
+        text_recognition_model_name=None,
+        text_recognition_model_dir=None,
+        text_recognition_batch_size=None,
+        text_rec_score_thresh=None,
+        table_classification_model_name=None,
+        table_classification_model_dir=None,
+        wired_table_structure_recognition_model_name=None,
+        wired_table_structure_recognition_model_dir=None,
+        wireless_table_structure_recognition_model_name=None,
+        wireless_table_structure_recognition_model_dir=None,
+        wired_table_cells_detection_model_name=None,
+        wired_table_cells_detection_model_dir=None,
+        wireless_table_cells_detection_model_name=None,
+        wireless_table_cells_detection_model_dir=None,
+        table_orientation_classify_model_name=None,
+        table_orientation_classify_model_dir=None,
+        seal_text_detection_model_name=None,
+        seal_text_detection_model_dir=None,
+        seal_det_limit_side_len=None,
+        seal_det_limit_type=None,
+        seal_det_thresh=None,
+        seal_det_box_thresh=None,
+        seal_det_unclip_ratio=None,
+        seal_text_recognition_model_name=None,
+        seal_text_recognition_model_dir=None,
+        seal_text_recognition_batch_size=None,
+        seal_rec_score_thresh=None,
+        formula_recognition_model_name=None,
+        formula_recognition_model_dir=None,
+        formula_recognition_batch_size=None,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_textline_orientation=None,
+        use_seal_recognition=None,
+        use_table_recognition=None,
+        use_formula_recognition=None,
+        use_chart_recognition=None,
+        use_region_detection=None,
+        format_block_content=None,
+        markdown_ignore_labels=None,
+        lang=None,
+        ocr_version=None,
+        **kwargs,
+    ):
+        if ocr_version is not None and ocr_version not in _SUPPORTED_OCR_VERSIONS:
+            raise ValueError(
+                f"Invalid OCR version: {ocr_version}. Supported values are {_SUPPORTED_OCR_VERSIONS}."
+            )
+
+        if all(
+            map(
+                lambda p: p is None,
+                (
+                    text_detection_model_name,
+                    text_detection_model_dir,
+                    text_recognition_model_name,
+                    text_recognition_model_dir,
+                ),
+            )
+        ):
+            if lang is not None or ocr_version is not None:
+                det_model_name, rec_model_name = self._get_ocr_model_names(
+                    lang, ocr_version
+                )
+                if det_model_name is None or rec_model_name is None:
+                    raise ValueError(
+                        f"No models are available for the language {repr(lang)} and OCR version {repr(ocr_version)}."
+                    )
+                text_detection_model_name = det_model_name
+                text_recognition_model_name = rec_model_name
+        else:
+            if lang is not None or ocr_version is not None:
+                warnings.warn(
+                    "`lang` and `ocr_version` will be ignored when model names or model directories are not `None`.",
+                    stacklevel=2,
+                )
+        params = locals().copy()
+        params["text_detection_model_name"] = text_detection_model_name
+        params["text_recognition_model_name"] = text_recognition_model_name
+        params.pop("self")
+        params.pop("kwargs")
+        self._params = params
+
+        super().__init__(**kwargs)
+
+    @property
+    def _paddlex_pipeline_name(self):
+        return "PP-StructureV3"
+
+    def predict_iter(
+        self,
+        input,
+        *,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_textline_orientation=None,
+        use_seal_recognition=None,
+        use_table_recognition=None,
+        use_formula_recognition=None,
+        use_chart_recognition=None,
+        use_region_detection=None,
+        format_block_content=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        text_det_limit_side_len=None,
+        text_det_limit_type=None,
+        text_det_thresh=None,
+        text_det_box_thresh=None,
+        text_det_unclip_ratio=None,
+        text_rec_score_thresh=None,
+        seal_det_limit_side_len=None,
+        seal_det_limit_type=None,
+        seal_det_thresh=None,
+        seal_det_box_thresh=None,
+        seal_det_unclip_ratio=None,
+        seal_rec_score_thresh=None,
+        use_wired_table_cells_trans_to_html=False,
+        use_wireless_table_cells_trans_to_html=False,
+        use_table_orientation_classify=True,
+        use_ocr_results_with_table_cells=True,
+        use_e2e_wired_table_rec_model=False,
+        use_e2e_wireless_table_rec_model=True,
+        markdown_ignore_labels=None,
+        **kwargs,
+    ):
+        return self.paddlex_pipeline.predict(
+            input,
+            use_doc_orientation_classify=use_doc_orientation_classify,
+            use_doc_unwarping=use_doc_unwarping,
+            use_textline_orientation=use_textline_orientation,
+            use_seal_recognition=use_seal_recognition,
+            use_table_recognition=use_table_recognition,
+            use_formula_recognition=use_formula_recognition,
+            use_chart_recognition=use_chart_recognition,
+            use_region_detection=use_region_detection,
+            format_block_content=format_block_content,
+            layout_threshold=layout_threshold,
+            layout_nms=layout_nms,
+            layout_unclip_ratio=layout_unclip_ratio,
+            layout_merge_bboxes_mode=layout_merge_bboxes_mode,
+            text_det_limit_side_len=text_det_limit_side_len,
+            text_det_limit_type=text_det_limit_type,
+            text_det_thresh=text_det_thresh,
+            text_det_box_thresh=text_det_box_thresh,
+            text_det_unclip_ratio=text_det_unclip_ratio,
+            text_rec_score_thresh=text_rec_score_thresh,
+            seal_det_limit_side_len=seal_det_limit_side_len,
+            seal_det_limit_type=seal_det_limit_type,
+            seal_det_thresh=seal_det_thresh,
+            seal_det_box_thresh=seal_det_box_thresh,
+            seal_det_unclip_ratio=seal_det_unclip_ratio,
+            seal_rec_score_thresh=seal_rec_score_thresh,
+            use_wired_table_cells_trans_to_html=use_wired_table_cells_trans_to_html,
+            use_wireless_table_cells_trans_to_html=use_wireless_table_cells_trans_to_html,
+            use_table_orientation_classify=use_table_orientation_classify,
+            use_ocr_results_with_table_cells=use_ocr_results_with_table_cells,
+            use_e2e_wired_table_rec_model=use_e2e_wired_table_rec_model,
+            use_e2e_wireless_table_rec_model=use_e2e_wireless_table_rec_model,
+            markdown_ignore_labels=markdown_ignore_labels,
+            **kwargs,
+        )
+
+    def predict(
+        self,
+        input,
+        *,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_textline_orientation=None,
+        use_seal_recognition=None,
+        use_table_recognition=None,
+        use_formula_recognition=None,
+        use_chart_recognition=None,
+        use_region_detection=None,
+        format_block_content=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        text_det_limit_side_len=None,
+        text_det_limit_type=None,
+        text_det_thresh=None,
+        text_det_box_thresh=None,
+        text_det_unclip_ratio=None,
+        text_rec_score_thresh=None,
+        seal_det_limit_side_len=None,
+        seal_det_limit_type=None,
+        seal_det_thresh=None,
+        seal_det_box_thresh=None,
+        seal_det_unclip_ratio=None,
+        seal_rec_score_thresh=None,
+        use_wired_table_cells_trans_to_html=False,
+        use_wireless_table_cells_trans_to_html=False,
+        use_table_orientation_classify=True,
+        use_ocr_results_with_table_cells=True,
+        use_e2e_wired_table_rec_model=False,
+        use_e2e_wireless_table_rec_model=True,
+        markdown_ignore_labels=None,
+        **kwargs,
+    ):
+        return list(
+            self.predict_iter(
+                input,
+                use_doc_orientation_classify=use_doc_orientation_classify,
+                use_doc_unwarping=use_doc_unwarping,
+                use_textline_orientation=use_textline_orientation,
+                use_seal_recognition=use_seal_recognition,
+                use_table_recognition=use_table_recognition,
+                use_formula_recognition=use_formula_recognition,
+                use_chart_recognition=use_chart_recognition,
+                use_region_detection=use_region_detection,
+                format_block_content=format_block_content,
+                layout_threshold=layout_threshold,
+                layout_nms=layout_nms,
+                layout_unclip_ratio=layout_unclip_ratio,
+                layout_merge_bboxes_mode=layout_merge_bboxes_mode,
+                text_det_limit_side_len=text_det_limit_side_len,
+                text_det_limit_type=text_det_limit_type,
+                text_det_thresh=text_det_thresh,
+                text_det_box_thresh=text_det_box_thresh,
+                text_det_unclip_ratio=text_det_unclip_ratio,
+                text_rec_score_thresh=text_rec_score_thresh,
+                seal_det_limit_side_len=seal_det_limit_side_len,
+                seal_det_limit_type=seal_det_limit_type,
+                seal_det_thresh=seal_det_thresh,
+                seal_det_box_thresh=seal_det_box_thresh,
+                seal_det_unclip_ratio=seal_det_unclip_ratio,
+                seal_rec_score_thresh=seal_rec_score_thresh,
+                use_wired_table_cells_trans_to_html=use_wired_table_cells_trans_to_html,
+                use_wireless_table_cells_trans_to_html=use_wireless_table_cells_trans_to_html,
+                use_table_orientation_classify=use_table_orientation_classify,
+                use_ocr_results_with_table_cells=use_ocr_results_with_table_cells,
+                use_e2e_wired_table_rec_model=use_e2e_wired_table_rec_model,
+                use_e2e_wireless_table_rec_model=use_e2e_wireless_table_rec_model,
+                markdown_ignore_labels=markdown_ignore_labels,
+                **kwargs,
+            )
+        )
+
+    def concatenate_markdown_pages(self, markdown_list):
+        return self.paddlex_pipeline.concatenate_markdown_pages(markdown_list)
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return PPStructureV3CLISubcommandExecutor()
+
+    def _get_paddlex_config_overrides(self):
+        STRUCTURE = {
+            "SubPipelines.DocPreprocessor.use_doc_orientation_classify": self._params[
+                "use_doc_orientation_classify"
+            ],
+            "SubPipelines.DocPreprocessor.use_doc_unwarping": self._params[
+                "use_doc_unwarping"
+            ],
+            "use_doc_preprocessor": self._params["use_doc_orientation_classify"]
+            or self._params["use_doc_unwarping"],
+            "SubPipelines.GeneralOCR.use_textline_orientation": self._params[
+                "use_textline_orientation"
+            ],
+            "use_seal_recognition": self._params["use_seal_recognition"],
+            "use_table_recognition": self._params["use_table_recognition"],
+            "use_formula_recognition": self._params["use_formula_recognition"],
+            "use_chart_recognition": self._params["use_chart_recognition"],
+            "use_region_detection": self._params["use_region_detection"],
+            "format_block_content": self._params["format_block_content"],
+            "markdown_ignore_labels": self._params["markdown_ignore_labels"],
+            "SubModules.LayoutDetection.model_name": self._params[
+                "layout_detection_model_name"
+            ],
+            "SubModules.LayoutDetection.model_dir": self._params[
+                "layout_detection_model_dir"
+            ],
+            "SubModules.LayoutDetection.threshold": self._params["layout_threshold"],
+            "SubModules.LayoutDetection.layout_nms": self._params["layout_nms"],
+            "SubModules.LayoutDetection.layout_unclip_ratio": self._params[
+                "layout_unclip_ratio"
+            ],
+            "SubModules.LayoutDetection.layout_merge_bboxes_mode": self._params[
+                "layout_merge_bboxes_mode"
+            ],
+            "SubModules.ChartRecognition.model_name": self._params[
+                "chart_recognition_model_name"
+            ],
+            "SubModules.ChartRecognition.model_dir": self._params[
+                "chart_recognition_model_dir"
+            ],
+            "SubModules.ChartRecognition.batch_size": self._params[
+                "chart_recognition_batch_size"
+            ],
+            "SubModules.RegionDetection.model_name": self._params[
+                "region_detection_model_name"
+            ],
+            "SubModules.RegionDetection.model_dir": self._params[
+                "region_detection_model_dir"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify.model_name": self._params[
+                "doc_orientation_classify_model_name"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify.model_dir": self._params[
+                "doc_orientation_classify_model_dir"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_name": self._params[
+                "doc_unwarping_model_name"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_dir": self._params[
+                "doc_unwarping_model_dir"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextDetection.model_name": self._params[
+                "text_detection_model_name"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextDetection.model_dir": self._params[
+                "text_detection_model_dir"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextDetection.limit_side_len": self._params[
+                "text_det_limit_side_len"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextDetection.limit_type": self._params[
+                "text_det_limit_type"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextDetection.thresh": self._params[
+                "text_det_thresh"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextDetection.box_thresh": self._params[
+                "text_det_box_thresh"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextDetection.unclip_ratio": self._params[
+                "text_det_unclip_ratio"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextLineOrientation.model_name": self._params[
+                "textline_orientation_model_name"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextLineOrientation.model_dir": self._params[
+                "textline_orientation_model_dir"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextLineOrientation.batch_size": self._params[
+                "textline_orientation_batch_size"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextRecognition.model_name": self._params[
+                "text_recognition_model_name"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextRecognition.model_dir": self._params[
+                "text_recognition_model_dir"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextRecognition.batch_size": self._params[
+                "text_recognition_batch_size"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextRecognition.score_thresh": self._params[
+                "text_rec_score_thresh"
+            ],
+            "SubPipelines.TableRecognition.SubModules.TableClassification.model_name": self._params[
+                "table_classification_model_name"
+            ],
+            "SubPipelines.TableRecognition.SubModules.TableClassification.model_dir": self._params[
+                "table_classification_model_dir"
+            ],
+            "SubPipelines.TableRecognition.SubModules.WiredTableStructureRecognition.model_name": self._params[
+                "wired_table_structure_recognition_model_name"
+            ],
+            "SubPipelines.TableRecognition.SubModules.WiredTableStructureRecognition.model_dir": self._params[
+                "wired_table_structure_recognition_model_dir"
+            ],
+            "SubPipelines.TableRecognition.SubModules.WirelessTableStructureRecognition.model_name": self._params[
+                "wireless_table_structure_recognition_model_name"
+            ],
+            "SubPipelines.TableRecognition.SubModules.WirelessTableStructureRecognition.model_dir": self._params[
+                "wireless_table_structure_recognition_model_dir"
+            ],
+            "SubPipelines.TableRecognition.SubModules.WiredTableCellsDetection.model_name": self._params[
+                "wired_table_cells_detection_model_name"
+            ],
+            "SubPipelines.TableRecognition.SubModules.WiredTableCellsDetection.model_dir": self._params[
+                "wired_table_cells_detection_model_dir"
+            ],
+            "SubPipelines.TableRecognition.SubModules.WirelessTableCellsDetection.model_name": self._params[
+                "wireless_table_cells_detection_model_name"
+            ],
+            "SubPipelines.TableRecognition.SubModules.WirelessTableCellsDetection.model_dir": self._params[
+                "wireless_table_cells_detection_model_dir"
+            ],
+            "SubPipelines.TableRecognition.SubModules.TableOrientationClassify.model_name": self._params[
+                "table_orientation_classify_model_name"
+            ],
+            "SubPipelines.TableRecognition.SubModules.TableOrientationClassify.model_dir": self._params[
+                "table_orientation_classify_model_dir"
+            ],
+            "SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextDetection.model_name": self._params[
+                "text_detection_model_name"
+            ],
+            "SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextDetection.model_dir": self._params[
+                "text_detection_model_dir"
+            ],
+            "SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextDetection.limit_side_len": self._params[
+                "text_det_limit_side_len"
+            ],
+            "SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextDetection.limit_type": self._params[
+                "text_det_limit_type"
+            ],
+            "SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextDetection.thresh": self._params[
+                "text_det_thresh"
+            ],
+            "SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextDetection.box_thresh": self._params[
+                "text_det_box_thresh"
+            ],
+            "SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextDetection.unclip_ratio": self._params[
+                "text_det_unclip_ratio"
+            ],
+            "SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextLineOrientation.model_name": self._params[
+                "textline_orientation_model_name"
+            ],
+            "SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextLineOrientation.model_dir": self._params[
+                "textline_orientation_model_dir"
+            ],
+            "SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextLineOrientation.batch_size": self._params[
+                "textline_orientation_batch_size"
+            ],
+            "SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextRecognition.model_name": self._params[
+                "text_recognition_model_name"
+            ],
+            "SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextRecognition.model_dir": self._params[
+                "text_recognition_model_dir"
+            ],
+            "SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextRecognition.batch_size": self._params[
+                "text_recognition_batch_size"
+            ],
+            "SubPipelines.TableRecognition.SubPipelines.GeneralOCR.SubModules.TextRecognition.score_thresh": self._params[
+                "text_rec_score_thresh"
+            ],
+            "SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.model_name": self._params[
+                "seal_text_detection_model_name"
+            ],
+            "SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.model_dir": self._params[
+                "seal_text_detection_model_dir"
+            ],
+            "SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.limit_side_len": self._params[
+                "text_det_limit_side_len"
+            ],
+            "SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.limit_type": self._params[
+                "seal_det_limit_type"
+            ],
+            "SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.thresh": self._params[
+                "seal_det_thresh"
+            ],
+            "SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.box_thresh": self._params[
+                "seal_det_box_thresh"
+            ],
+            "SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextDetection.unclip_ratio": self._params[
+                "seal_det_unclip_ratio"
+            ],
+            "SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextRecognition.model_name": self._params[
+                "seal_text_recognition_model_name"
+            ],
+            "SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextRecognition.model_dir": self._params[
+                "seal_text_recognition_model_dir"
+            ],
+            "SubPipelines.SealRecognition.SubPipelines.SealOCR.SubModules.TextRecognition.batch_size": self._params[
+                "seal_text_recognition_batch_size"
+            ],
+            "SubPipelines.FormulaRecognition.SubModules.FormulaRecognition.model_name": self._params[
+                "formula_recognition_model_name"
+            ],
+            "SubPipelines.FormulaRecognition.SubModules.FormulaRecognition.model_dir": self._params[
+                "formula_recognition_model_dir"
+            ],
+            "SubPipelines.FormulaRecognition.SubModules.FormulaRecognition.batch_size": self._params[
+                "formula_recognition_batch_size"
+            ],
+        }
+        return create_config_from_structure(STRUCTURE)
+
+    def _get_ocr_model_names(self, lang, ppocr_version):
+        LATIN_LANGS = [
+            "af",
+            "az",
+            "bs",
+            "cs",
+            "cy",
+            "da",
+            "de",
+            "es",
+            "et",
+            "fr",
+            "ga",
+            "hr",
+            "hu",
+            "id",
+            "is",
+            "it",
+            "ku",
+            "la",
+            "lt",
+            "lv",
+            "mi",
+            "ms",
+            "mt",
+            "nl",
+            "no",
+            "oc",
+            "pi",
+            "pl",
+            "pt",
+            "ro",
+            "rs_latin",
+            "sk",
+            "sl",
+            "sq",
+            "sv",
+            "sw",
+            "tl",
+            "tr",
+            "uz",
+            "vi",
+            "french",
+            "german",
+        ]
+        ARABIC_LANGS = ["ar", "fa", "ug", "ur"]
+        ESLAV_LANGS = ["ru", "be", "uk"]
+        CYRILLIC_LANGS = [
+            "ru",
+            "rs_cyrillic",
+            "be",
+            "bg",
+            "uk",
+            "mn",
+            "abq",
+            "ady",
+            "kbd",
+            "ava",
+            "dar",
+            "inh",
+            "che",
+            "lbe",
+            "lez",
+            "tab",
+        ]
+        DEVANAGARI_LANGS = [
+            "hi",
+            "mr",
+            "ne",
+            "bh",
+            "mai",
+            "ang",
+            "bho",
+            "mah",
+            "sck",
+            "new",
+            "gom",
+            "sa",
+            "bgc",
+        ]
+        SPECIFIC_LANGS = [
+            "ch",
+            "en",
+            "korean",
+            "japan",
+            "chinese_cht",
+            "te",
+            "ka",
+            "ta",
+        ]
+
+        if lang is None:
+            lang = "ch"
+
+        if ppocr_version is None:
+            if (
+                lang
+                in ["ch", "chinese_cht", "en", "japan", "korean", "th", "el"]
+                + LATIN_LANGS
+                + ESLAV_LANGS
+            ):
+                ppocr_version = "PP-OCRv5"
+            elif lang in (
+                LATIN_LANGS
+                + ARABIC_LANGS
+                + CYRILLIC_LANGS
+                + DEVANAGARI_LANGS
+                + SPECIFIC_LANGS
+            ):
+                ppocr_version = "PP-OCRv3"
+            else:
+                # Unknown language specified
+                return None, None
+
+        if ppocr_version == "PP-OCRv5":
+            rec_lang, rec_model_name = None, None
+            if lang in ("ch", "chinese_cht", "en", "japan"):
+                rec_model_name = "PP-OCRv5_server_rec"
+            elif lang in LATIN_LANGS:
+                rec_lang = "latin"
+            elif lang in ESLAV_LANGS:
+                rec_lang = "eslav"
+            elif lang == "korean":
+                rec_lang = "korean"
+            elif lang == "th":
+                rec_lang = "th"
+            elif lang == "el":
+                rec_lang = "el"
+
+            if rec_lang is not None:
+                rec_model_name = f"{rec_lang}_PP-OCRv5_mobile_rec"
+            return "PP-OCRv5_server_det", rec_model_name
+
+        elif ppocr_version == "PP-OCRv4":
+            if lang == "ch":
+                return "PP-OCRv4_mobile_det", "PP-OCRv4_mobile_rec"
+            elif lang == "en":
+                return "PP-OCRv4_mobile_det", "en_PP-OCRv4_mobile_rec"
+            else:
+                return None, None
+        else:
+            # PP-OCRv3
+            rec_lang = None
+            if lang in LATIN_LANGS:
+                rec_lang = "latin"
+            elif lang in ARABIC_LANGS:
+                rec_lang = "arabic"
+            elif lang in CYRILLIC_LANGS:
+                rec_lang = "cyrillic"
+            elif lang in DEVANAGARI_LANGS:
+                rec_lang = "devanagari"
+            else:
+                if lang in SPECIFIC_LANGS:
+                    rec_lang = lang
+
+            rec_model_name = None
+            if rec_lang == "ch":
+                rec_model_name = "PP-OCRv3_mobile_rec"
+            elif rec_lang is not None:
+                rec_model_name = f"{rec_lang}_PP-OCRv3_mobile_rec"
+            return "PP-OCRv3_mobile_det", rec_model_name
+
+
+class PPStructureV3CLISubcommandExecutor(PipelineCLISubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "pp_structurev3"
+
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(subparser)
+
+        subparser.add_argument(
+            "--layout_detection_model_name",
+            type=str,
+            help="Name of the layout detection model.",
+        )
+        subparser.add_argument(
+            "--layout_detection_model_dir",
+            type=str,
+            help="Path to the layout detection model directory.",
+        )
+        subparser.add_argument(
+            "--layout_threshold",
+            type=float,
+            help="Score threshold for the layout detection model.",
+        )
+        subparser.add_argument(
+            "--layout_nms",
+            type=str2bool,
+            help="Whether to use NMS in layout detection.",
+        )
+        subparser.add_argument(
+            "--layout_unclip_ratio",
+            type=float,
+            help="Expansion coefficient for layout detection.",
+        )
+        subparser.add_argument(
+            "--layout_merge_bboxes_mode",
+            type=str,
+            help="Overlapping box filtering method.",
+        )
+
+        subparser.add_argument(
+            "--chart_recognition_model_name",
+            type=str,
+            help="Name of the chart recognition model.",
+        )
+        subparser.add_argument(
+            "--chart_recognition_model_dir",
+            type=str,
+            help="Path to the chart recognition model directory.",
+        )
+        subparser.add_argument(
+            "--chart_recognition_batch_size",
+            type=int,
+            help="Batch size for the chart recognition model.",
+        )
+
+        subparser.add_argument(
+            "--region_detection_model_name",
+            type=str,
+            help="Name of the region detection model.",
+        )
+        subparser.add_argument(
+            "--region_detection_model_dir",
+            type=str,
+            help="Path to the region detection model directory.",
+        )
+
+        subparser.add_argument(
+            "--doc_orientation_classify_model_name",
+            type=str,
+            help="Name of the document image orientation classification model.",
+        )
+        subparser.add_argument(
+            "--doc_orientation_classify_model_dir",
+            type=str,
+            help="Path to the document image orientation classification model directory.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_name",
+            type=str,
+            help="Name of the text image unwarping model.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_dir",
+            type=str,
+            help="Path to the image unwarping model directory.",
+        )
+
+        subparser.add_argument(
+            "--text_detection_model_name",
+            type=str,
+            help="Name of the text detection model.",
+        )
+        subparser.add_argument(
+            "--text_detection_model_dir",
+            type=str,
+            help="Path to the text detection model directory.",
+        )
+        subparser.add_argument(
+            "--text_det_limit_side_len",
+            type=int,
+            help="This sets a limit on the side length of the input image for the text detection model.",
+        )
+        subparser.add_argument(
+            "--text_det_limit_type",
+            type=str,
+            help="This determines how the side length limit is applied to the input image before feeding it into the text deteciton model.",
+        )
+        subparser.add_argument(
+            "--text_det_thresh",
+            type=float,
+            help="Detection pixel threshold for the text detection model. Pixels with scores greater than this threshold in the output probability map are considered text pixels.",
+        )
+        subparser.add_argument(
+            "--text_det_box_thresh",
+            type=float,
+            help="Detection box threshold for the text detection model. A detection result is considered a text region if the average score of all pixels within the border of the result is greater than this threshold.",
+        )
+        subparser.add_argument(
+            "--text_det_unclip_ratio",
+            type=float,
+            help="Text detection expansion coefficient, which expands the text region using this method. The larger the value, the larger the expansion area.",
+        )
+        subparser.add_argument(
+            "--textline_orientation_model_name",
+            type=str,
+            help="Name of the text line orientation classification model.",
+        )
+        subparser.add_argument(
+            "--textline_orientation_model_dir",
+            type=str,
+            help="Path to the text line orientation classification directory.",
+        )
+        subparser.add_argument(
+            "--textline_orientation_batch_size",
+            type=int,
+            help="Batch size for the text line orientation classification model.",
+        )
+        subparser.add_argument(
+            "--text_recognition_model_name",
+            type=str,
+            help="Name of the text recognition model.",
+        )
+        subparser.add_argument(
+            "--text_recognition_model_dir",
+            type=str,
+            help="Path to the text recognition model directory.",
+        )
+        subparser.add_argument(
+            "--text_recognition_batch_size",
+            type=int,
+            help="Batch size for the text recognition model.",
+        )
+        subparser.add_argument(
+            "--text_rec_score_thresh",
+            type=float,
+            help="Text recognition threshold used in general OCR. Text results with scores greater than this threshold are retained.",
+        )
+
+        subparser.add_argument(
+            "--table_classification_model_name",
+            type=str,
+            help="Name of the table classification model.",
+        )
+        subparser.add_argument(
+            "--table_classification_model_dir",
+            type=str,
+            help="Path to the table classification model directory.",
+        )
+        subparser.add_argument(
+            "--wired_table_structure_recognition_model_name",
+            type=str,
+            help="Name of the wired table structure recognition model.",
+        )
+        subparser.add_argument(
+            "--wired_table_structure_recognition_model_dir",
+            type=str,
+            help="Path to the wired table structure recognition model directory.",
+        )
+        subparser.add_argument(
+            "--wireless_table_structure_recognition_model_name",
+            type=str,
+            help="Name of the wireless table structure recognition model.",
+        )
+        subparser.add_argument(
+            "--wireless_table_structure_recognition_model_dir",
+            type=str,
+            help="Path to the wired table structure recognition model directory.",
+        )
+        subparser.add_argument(
+            "--wired_table_cells_detection_model_name",
+            type=str,
+            help="Name of the wired table cells detection model.",
+        )
+        subparser.add_argument(
+            "--wired_table_cells_detection_model_dir",
+            type=str,
+            help="Path to the wired table cells detection model directory.",
+        )
+        subparser.add_argument(
+            "--wireless_table_cells_detection_model_name",
+            type=str,
+            help="Name of the wireless table cells detection model.",
+        )
+        subparser.add_argument(
+            "--wireless_table_cells_detection_model_dir",
+            type=str,
+            help="Path to the wireless table cells detection model directory.",
+        )
+
+        subparser.add_argument(
+            "--seal_text_detection_model_name",
+            type=str,
+            help="Name of the seal text detection model.",
+        )
+        subparser.add_argument(
+            "--seal_text_detection_model_dir",
+            type=str,
+            help="Path to the seal text detection model directory.",
+        )
+        subparser.add_argument(
+            "--seal_det_limit_side_len",
+            type=int,
+            help="This sets a limit on the side length of the input image for the seal text detection model.",
+        )
+        subparser.add_argument(
+            "--seal_det_limit_type",
+            type=str,
+            help="This determines how the side length limit is applied to the input image before feeding it into the seal text deteciton model.",
+        )
+        subparser.add_argument(
+            "--seal_det_thresh",
+            type=float,
+            help="Detection pixel threshold for the seal text detection model. Pixels with scores greater than this threshold in the output probability map are considered text pixels.",
+        )
+        subparser.add_argument(
+            "--seal_det_box_thresh",
+            type=float,
+            help="Detection box threshold for the seal text detection model. A detection result is considered a text region if the average score of all pixels within the border of the result is greater than this threshold.",
+        )
+        subparser.add_argument(
+            "--seal_det_unclip_ratio",
+            type=float,
+            help="Seal text detection expansion coefficient, which expands the text region using this method. The larger the value, the larger the expansion area.",
+        )
+        subparser.add_argument(
+            "--seal_text_recognition_model_name",
+            type=str,
+            help="Name of the seal text recognition model.",
+        )
+        subparser.add_argument(
+            "--seal_text_recognition_model_dir",
+            type=str,
+            help="Path to the seal text recognition model directory.",
+        )
+        subparser.add_argument(
+            "--seal_text_recognition_batch_size",
+            type=int,
+            help="Batch size for the seal text recognition model.",
+        )
+        subparser.add_argument(
+            "--seal_rec_score_thresh",
+            type=float,
+            help="Seal text recognition threshold. Text results with scores greater than this threshold are retained.",
+        )
+
+        subparser.add_argument(
+            "--formula_recognition_model_name",
+            type=str,
+            help="Name of the formula recognition model.",
+        )
+        subparser.add_argument(
+            "--formula_recognition_model_dir",
+            type=str,
+            help="Path to the formula recognition model directory.",
+        )
+        subparser.add_argument(
+            "--formula_recognition_batch_size",
+            type=int,
+            help="Batch size for the formula recognition model.",
+        )
+
+        subparser.add_argument(
+            "--use_doc_orientation_classify",
+            type=str2bool,
+            help="Whether to use document image orientation classification.",
+        )
+        subparser.add_argument(
+            "--use_doc_unwarping",
+            type=str2bool,
+            help="Whether to use text image unwarping.",
+        )
+        subparser.add_argument(
+            "--use_textline_orientation",
+            type=str2bool,
+            help="Whether to use text line orientation classification.",
+        )
+        subparser.add_argument(
+            "--use_seal_recognition",
+            type=str2bool,
+            help="Whether to use seal recognition.",
+        )
+        subparser.add_argument(
+            "--use_table_recognition",
+            type=str2bool,
+            help="Whether to use table recognition.",
+        )
+        subparser.add_argument(
+            "--use_formula_recognition",
+            type=str2bool,
+            help="Whether to use formula recognition.",
+        )
+        subparser.add_argument(
+            "--use_chart_recognition",
+            type=str2bool,
+            help="Whether to use chart recognition.",
+        )
+        subparser.add_argument(
+            "--use_region_detection",
+            type=str2bool,
+            help="Whether to use region detection.",
+        )
+
+        subparser.add_argument(
+            "--format_block_content",
+            type=str2bool,
+            help="Whether to format block content to Markdown.",
+        )
+        subparser.add_argument(
+            "--markdown_ignore_labels",
+            type=str,
+            nargs="+",
+            help="List of layout labels to ignore in Markdown output.",
+        )
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+        perform_simple_inference(
+            PPStructureV3,
+            params,
+        )
diff --git a/paddleocr/_pipelines/seal_recognition.py b/paddleocr/_pipelines/seal_recognition.py
new file mode 100644
index 0000000..b018531
--- /dev/null
+++ b/paddleocr/_pipelines/seal_recognition.py
@@ -0,0 +1,377 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .._utils.cli import (
+    add_simple_inference_args,
+    get_subcommand_args,
+    perform_simple_inference,
+    str2bool,
+)
+from .base import PaddleXPipelineWrapper, PipelineCLISubcommandExecutor
+from .utils import create_config_from_structure
+
+
+class SealRecognition(PaddleXPipelineWrapper):
+    def __init__(
+        self,
+        doc_orientation_classify_model_name=None,
+        doc_orientation_classify_model_dir=None,
+        doc_unwarping_model_name=None,
+        doc_unwarping_model_dir=None,
+        layout_detection_model_name=None,
+        layout_detection_model_dir=None,
+        seal_text_detection_model_name=None,
+        seal_text_detection_model_dir=None,
+        text_recognition_model_name=None,
+        text_recognition_model_dir=None,
+        text_recognition_batch_size=None,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_layout_detection=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        seal_det_limit_side_len=None,
+        seal_det_limit_type=None,
+        seal_det_thresh=None,
+        seal_det_box_thresh=None,
+        seal_det_unclip_ratio=None,
+        seal_rec_score_thresh=None,
+        **kwargs,
+    ):
+
+        self._params = {
+            "doc_orientation_classify_model_name": doc_orientation_classify_model_name,
+            "doc_orientation_classify_model_dir": doc_orientation_classify_model_dir,
+            "doc_unwarping_model_name": doc_unwarping_model_name,
+            "doc_unwarping_model_dir": doc_unwarping_model_dir,
+            "layout_detection_model_name": layout_detection_model_name,
+            "layout_detection_model_dir": layout_detection_model_dir,
+            "seal_text_detection_model_name": seal_text_detection_model_name,
+            "seal_text_detection_model_dir": seal_text_detection_model_dir,
+            "text_recognition_model_name": text_recognition_model_name,
+            "text_recognition_model_dir": text_recognition_model_dir,
+            "text_recognition_batch_size": text_recognition_batch_size,
+            "use_doc_orientation_classify": use_doc_orientation_classify,
+            "use_doc_unwarping": use_doc_unwarping,
+            "use_layout_detection": use_layout_detection,
+            "layout_threshold": layout_threshold,
+            "layout_nms": layout_nms,
+            "layout_unclip_ratio": layout_unclip_ratio,
+            "layout_merge_bboxes_mode": layout_merge_bboxes_mode,
+            "seal_det_limit_side_len": seal_det_limit_side_len,
+            "seal_det_limit_type": seal_det_limit_type,
+            "seal_det_thresh": seal_det_thresh,
+            "seal_det_box_thresh": seal_det_box_thresh,
+            "seal_det_unclip_ratio": seal_det_unclip_ratio,
+            "seal_rec_score_thresh": seal_rec_score_thresh,
+        }
+        super().__init__(**kwargs)
+
+    @property
+    def _paddlex_pipeline_name(self):
+        return "seal_recognition"
+
+    def predict_iter(
+        self,
+        input,
+        *,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_layout_detection=None,
+        layout_det_res=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        seal_det_limit_side_len=None,
+        seal_det_limit_type=None,
+        seal_det_thresh=None,
+        seal_det_box_thresh=None,
+        seal_det_unclip_ratio=None,
+        seal_rec_score_thresh=None,
+        **kwargs,
+    ):
+        return self.paddlex_pipeline.predict(
+            input,
+            use_doc_orientation_classify=use_doc_orientation_classify,
+            use_doc_unwarping=use_doc_unwarping,
+            use_layout_detection=use_layout_detection,
+            layout_det_res=layout_det_res,
+            layout_threshold=layout_threshold,
+            layout_nms=layout_nms,
+            layout_unclip_ratio=layout_unclip_ratio,
+            layout_merge_bboxes_mode=layout_merge_bboxes_mode,
+            seal_det_limit_side_len=seal_det_limit_side_len,
+            seal_det_limit_type=seal_det_limit_type,
+            seal_det_thresh=seal_det_thresh,
+            seal_det_box_thresh=seal_det_box_thresh,
+            seal_det_unclip_ratio=seal_det_unclip_ratio,
+            seal_rec_score_thresh=seal_rec_score_thresh,
+            **kwargs,
+        )
+
+    def predict(
+        self,
+        input,
+        *,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_layout_detection=None,
+        layout_det_res=None,
+        layout_threshold=None,
+        layout_nms=None,
+        layout_unclip_ratio=None,
+        layout_merge_bboxes_mode=None,
+        seal_det_limit_side_len=None,
+        seal_det_limit_type=None,
+        seal_det_thresh=None,
+        seal_det_box_thresh=None,
+        seal_det_unclip_ratio=None,
+        seal_rec_score_thresh=None,
+        **kwargs,
+    ):
+        return list(
+            self.predict_iter(
+                input,
+                use_doc_orientation_classify=use_doc_orientation_classify,
+                use_doc_unwarping=use_doc_unwarping,
+                use_layout_detection=use_layout_detection,
+                layout_det_res=layout_det_res,
+                layout_threshold=layout_threshold,
+                layout_nms=layout_nms,
+                layout_unclip_ratio=layout_unclip_ratio,
+                layout_merge_bboxes_mode=layout_merge_bboxes_mode,
+                seal_det_limit_side_len=seal_det_limit_side_len,
+                seal_det_limit_type=seal_det_limit_type,
+                seal_det_thresh=seal_det_thresh,
+                seal_det_box_thresh=seal_det_box_thresh,
+                seal_det_unclip_ratio=seal_det_unclip_ratio,
+                seal_rec_score_thresh=seal_rec_score_thresh,
+                **kwargs,
+            )
+        )
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return SealRecognitionCLISubcommandExecutor()
+
+    def _get_paddlex_config_overrides(self):
+        STRUCTURE = {
+            "SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify.model_name": self._params[
+                "doc_orientation_classify_model_name"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify.model_dir": self._params[
+                "doc_orientation_classify_model_dir"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_name": self._params[
+                "doc_unwarping_model_name"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_dir": self._params[
+                "doc_unwarping_model_dir"
+            ],
+            "SubModules.LayoutDetection.model_name": self._params[
+                "layout_detection_model_name"
+            ],
+            "SubModules.LayoutDetection.model_dir": self._params[
+                "layout_detection_model_dir"
+            ],
+            "SubModules.LayoutDetection.threshold": self._params["layout_threshold"],
+            "SubModules.LayoutDetection.layout_nms": self._params["layout_nms"],
+            "SubModules.LayoutDetection.layout_unclip_ratio": self._params[
+                "layout_unclip_ratio"
+            ],
+            "SubModules.LayoutDetection.layout_merge_bboxes_mode": self._params[
+                "layout_merge_bboxes_mode"
+            ],
+            "SubPipelines.DocPreprocessor.use_doc_orientation_classify": self._params[
+                "use_doc_orientation_classify"
+            ],
+            "SubPipelines.DocPreprocessor.use_doc_unwarping": self._params[
+                "use_doc_unwarping"
+            ],
+            "use_doc_preprocessor": self._params["use_doc_orientation_classify"]
+            or self._params["use_doc_unwarping"],
+            "SubPipelines.SealOCR.SubModules.TextDetection.model_name": self._params[
+                "seal_text_detection_model_name"
+            ],
+            "SubPipelines.SealOCR.SubModules.TextDetection.model_dir": self._params[
+                "seal_text_detection_model_dir"
+            ],
+            "SubPipelines.SealOCR.SubModules.TextDetection.limit_side_len": self._params[
+                "seal_det_limit_side_len"
+            ],
+            "SubPipelines.SealOCR.SubModules.TextDetection.limit_type": self._params[
+                "seal_det_limit_type"
+            ],
+            "SubPipelines.SealOCR.SubModules.TextDetection.thresh": self._params[
+                "seal_det_thresh"
+            ],
+            "SubPipelines.SealOCR.SubModules.TextDetection.box_thresh": self._params[
+                "seal_det_box_thresh"
+            ],
+            "SubPipelines.SealOCR.SubModules.TextDetection.unclip_ratio": self._params[
+                "seal_det_unclip_ratio"
+            ],
+            "SubPipelines.SealOCR.SubModules.TextRecognition.model_name": self._params[
+                "text_recognition_model_name"
+            ],
+            "SubPipelines.SealOCR.SubModules.TextRecognition.model_dir": self._params[
+                "text_recognition_model_dir"
+            ],
+            "SubPipelines.SealOCR.SubModules.TextRecognition.batch_size": self._params[
+                "text_recognition_batch_size"
+            ],
+            "SubPipelines.SealOCR.SubModules.TextRecognition.score_thresh": self._params[
+                "seal_rec_score_thresh"
+            ],
+            "use_layout_detection": self._params["use_layout_detection"],
+        }
+        return create_config_from_structure(STRUCTURE)
+
+
+class SealRecognitionCLISubcommandExecutor(PipelineCLISubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "seal_recognition"
+
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(subparser)
+
+        subparser.add_argument(
+            "--doc_orientation_classify_model_name",
+            type=str,
+            help="Name of the document image orientation classification model.",
+        )
+        subparser.add_argument(
+            "--doc_orientation_classify_model_dir",
+            type=str,
+            help="Path to the document image orientation classification model directory.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_name",
+            type=str,
+            help="Name of the document image unwarping model.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_dir",
+            type=str,
+            help="Path to the document image unwarping model directory.",
+        )
+        subparser.add_argument(
+            "--layout_detection_model_name",
+            type=str,
+            help="Name of the layout detection model.",
+        )
+        subparser.add_argument(
+            "--layout_detection_model_dir",
+            type=str,
+            help="Path to the layout detection model directory.",
+        )
+        subparser.add_argument(
+            "--seal_text_detection_model_name",
+            type=str,
+            help="Name of the seal text detection model.",
+        )
+        subparser.add_argument(
+            "--seal_text_detection_model_dir",
+            type=str,
+            help="Path to the seal text detection model directory.",
+        )
+        subparser.add_argument(
+            "--text_recognition_model_name",
+            type=str,
+            help="Name of the text recognition model.",
+        )
+        subparser.add_argument(
+            "--text_recognition_model_dir",
+            type=str,
+            help="Path to the text recognition model directory.",
+        )
+        subparser.add_argument(
+            "--text_recognition_batch_size",
+            type=int,
+            help="Batch size for the text recognition model.",
+        )
+        subparser.add_argument(
+            "--use_doc_orientation_classify",
+            type=str2bool,
+            help="Whether to use document image orientation classification.",
+        )
+        subparser.add_argument(
+            "--use_doc_unwarping",
+            type=str2bool,
+            help="Whether to use document image unwarping.",
+        )
+        subparser.add_argument(
+            "--use_layout_detection",
+            type=str2bool,
+            help="Whether to use layout detection.",
+        )
+        subparser.add_argument(
+            "--layout_threshold",
+            type=float,
+            help="Threshold for layout detection model.",
+        )
+        subparser.add_argument(
+            "--layout_nms",
+            type=str2bool,
+            help="Non-Maximum Suppression threshold for layout detection.",
+        )
+        subparser.add_argument(
+            "--layout_unclip_ratio",
+            type=float,
+            help="Layout detection expansion coefficient.",
+        )
+        subparser.add_argument(
+            "--layout_merge_bboxes_mode",
+            type=str,
+            help="Mode for merging bounding boxes in layout detection.",
+        )
+        subparser.add_argument(
+            "--seal_det_limit_side_len",
+            type=int,
+            help="This sets a limit on the side length of the input image for the seal text detection model.",
+        )
+        subparser.add_argument(
+            "--seal_det_limit_type",
+            type=str,
+            help="This determines how the side length limit is applied to the input image before feeding it into the seal text detection model.",
+        )
+        subparser.add_argument(
+            "--seal_det_thresh",
+            type=float,
+            help="Detection pixel threshold for the seal text detection model. Pixels with scores greater than this threshold in the output probability map are considered text pixels.",
+        )
+        subparser.add_argument(
+            "--seal_det_box_thresh",
+            type=float,
+            help="Detection box threshold for the seal text detection model. A detection result is considered a text region if the average score of all pixels within the border of the result is greater than this threshold.",
+        )
+        subparser.add_argument(
+            "--seal_det_unclip_ratio",
+            type=float,
+            help="Seal text detection expansion coefficient, which expands the text region using this method. The larger the value, the larger the expansion area.",
+        )
+        subparser.add_argument(
+            "--seal_rec_score_thresh",
+            type=float,
+            help="Text recognition threshold. Text results with scores greater than this threshold are retained.",
+        )
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+
+        perform_simple_inference(SealRecognition, params)
diff --git a/paddleocr/_pipelines/table_recognition_v2.py b/paddleocr/_pipelines/table_recognition_v2.py
new file mode 100644
index 0000000..5cf4868
--- /dev/null
+++ b/paddleocr/_pipelines/table_recognition_v2.py
@@ -0,0 +1,438 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .._utils.cli import (
+    add_simple_inference_args,
+    get_subcommand_args,
+    perform_simple_inference,
+    str2bool,
+)
+from .base import PaddleXPipelineWrapper, PipelineCLISubcommandExecutor
+from .utils import create_config_from_structure
+
+
+class TableRecognitionPipelineV2(PaddleXPipelineWrapper):
+    def __init__(
+        self,
+        layout_detection_model_name=None,
+        layout_detection_model_dir=None,
+        table_classification_model_name=None,
+        table_classification_model_dir=None,
+        wired_table_structure_recognition_model_name=None,
+        wired_table_structure_recognition_model_dir=None,
+        wireless_table_structure_recognition_model_name=None,
+        wireless_table_structure_recognition_model_dir=None,
+        wired_table_cells_detection_model_name=None,
+        wired_table_cells_detection_model_dir=None,
+        wireless_table_cells_detection_model_name=None,
+        wireless_table_cells_detection_model_dir=None,
+        doc_orientation_classify_model_name=None,
+        doc_orientation_classify_model_dir=None,
+        doc_unwarping_model_name=None,
+        doc_unwarping_model_dir=None,
+        text_detection_model_name=None,
+        text_detection_model_dir=None,
+        text_det_limit_side_len=None,
+        text_det_limit_type=None,
+        text_det_thresh=None,
+        text_det_box_thresh=None,
+        text_det_unclip_ratio=None,
+        text_recognition_model_name=None,
+        text_recognition_model_dir=None,
+        text_recognition_batch_size=None,
+        text_rec_score_thresh=None,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_layout_detection=None,
+        use_ocr_model=None,
+        **kwargs,
+    ):
+        params = locals().copy()
+        params.pop("self")
+        params.pop("kwargs")
+        self._params = params
+
+        super().__init__(**kwargs)
+
+    @property
+    def _paddlex_pipeline_name(self):
+        return "table_recognition_v2"
+
+    def predict_iter(
+        self,
+        input,
+        *,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_layout_detection=None,
+        use_ocr_model=None,
+        overall_ocr_res=None,
+        layout_det_res=None,
+        text_det_limit_side_len=None,
+        text_det_limit_type=None,
+        text_det_thresh=None,
+        text_det_box_thresh=None,
+        text_det_unclip_ratio=None,
+        text_rec_score_thresh=None,
+        use_e2e_wired_table_rec_model=False,
+        use_e2e_wireless_table_rec_model=False,
+        use_wired_table_cells_trans_to_html=False,
+        use_wireless_table_cells_trans_to_html=False,
+        use_table_orientation_classify=True,
+        use_ocr_results_with_table_cells=True,
+        **kwargs,
+    ):
+        return self.paddlex_pipeline.predict(
+            input,
+            use_doc_orientation_classify=use_doc_orientation_classify,
+            use_doc_unwarping=use_doc_unwarping,
+            use_layout_detection=use_layout_detection,
+            use_ocr_model=use_ocr_model,
+            overall_ocr_res=overall_ocr_res,
+            layout_det_res=layout_det_res,
+            text_det_limit_side_len=text_det_limit_side_len,
+            text_det_limit_type=text_det_limit_type,
+            text_det_thresh=text_det_thresh,
+            text_det_box_thresh=text_det_box_thresh,
+            text_det_unclip_ratio=text_det_unclip_ratio,
+            text_rec_score_thresh=text_rec_score_thresh,
+            use_e2e_wired_table_rec_model=use_e2e_wired_table_rec_model,
+            use_e2e_wireless_table_rec_model=use_e2e_wireless_table_rec_model,
+            use_wired_table_cells_trans_to_html=use_wired_table_cells_trans_to_html,
+            use_wireless_table_cells_trans_to_html=use_wireless_table_cells_trans_to_html,
+            use_table_orientation_classify=use_table_orientation_classify,
+            use_ocr_results_with_table_cells=use_ocr_results_with_table_cells,
+            **kwargs,
+        )
+
+    def predict(
+        self,
+        input,
+        *,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_layout_detection=None,
+        use_ocr_model=None,
+        overall_ocr_res=None,
+        layout_det_res=None,
+        text_det_limit_side_len=None,
+        text_det_limit_type=None,
+        text_det_thresh=None,
+        text_det_box_thresh=None,
+        text_det_unclip_ratio=None,
+        text_rec_score_thresh=None,
+        use_e2e_wired_table_rec_model=False,
+        use_e2e_wireless_table_rec_model=False,
+        use_wired_table_cells_trans_to_html=False,
+        use_wireless_table_cells_trans_to_html=False,
+        use_table_orientation_classify=True,
+        use_ocr_results_with_table_cells=True,
+        **kwargs,
+    ):
+        return list(
+            self.predict_iter(
+                input,
+                use_doc_orientation_classify=use_doc_orientation_classify,
+                use_doc_unwarping=use_doc_unwarping,
+                use_layout_detection=use_layout_detection,
+                use_ocr_model=use_ocr_model,
+                overall_ocr_res=overall_ocr_res,
+                layout_det_res=layout_det_res,
+                text_det_limit_side_len=text_det_limit_side_len,
+                text_det_limit_type=text_det_limit_type,
+                text_det_thresh=text_det_thresh,
+                text_det_box_thresh=text_det_box_thresh,
+                text_det_unclip_ratio=text_det_unclip_ratio,
+                text_rec_score_thresh=text_rec_score_thresh,
+                use_e2e_wired_table_rec_model=use_e2e_wired_table_rec_model,
+                use_e2e_wireless_table_rec_model=use_e2e_wireless_table_rec_model,
+                use_wired_table_cells_trans_to_html=use_wired_table_cells_trans_to_html,
+                use_wireless_table_cells_trans_to_html=use_wireless_table_cells_trans_to_html,
+                use_table_orientation_classify=use_table_orientation_classify,
+                use_ocr_results_with_table_cells=use_ocr_results_with_table_cells,
+                **kwargs,
+            )
+        )
+
+    @classmethod
+    def get_cli_subcommand_executor(cls):
+        return TableRecognitionPipelineV2CLISubcommandExecutor()
+
+    def _get_paddlex_config_overrides(self):
+        STRUCTURE = {
+            "SubPipelines.DocPreprocessor.use_doc_orientation_classify": self._params[
+                "use_doc_orientation_classify"
+            ],
+            "SubPipelines.DocPreprocessor.use_doc_unwarping": self._params[
+                "use_doc_unwarping"
+            ],
+            "use_doc_preprocessor": self._params["use_doc_orientation_classify"]
+            or self._params["use_doc_unwarping"],
+            "use_layout_detection": self._params["use_layout_detection"],
+            "use_ocr_model": self._params["use_ocr_model"],
+            "SubModules.LayoutDetection.model_name": self._params[
+                "layout_detection_model_name"
+            ],
+            "SubModules.LayoutDetection.model_dir": self._params[
+                "layout_detection_model_dir"
+            ],
+            "SubModules.TableClassification.model_name": self._params[
+                "table_classification_model_name"
+            ],
+            "SubModules.TableClassification.model_dir": self._params[
+                "table_classification_model_dir"
+            ],
+            "SubModules.WiredTableStructureRecognition.model_name": self._params[
+                "wired_table_structure_recognition_model_name"
+            ],
+            "SubModules.WiredTableStructureRecognition.model_dir": self._params[
+                "wired_table_structure_recognition_model_dir"
+            ],
+            "SubModules.WirelessTableStructureRecognition.model_name": self._params[
+                "wireless_table_structure_recognition_model_name"
+            ],
+            "SubModules.WirelessTableStructureRecognition.model_dir": self._params[
+                "wireless_table_structure_recognition_model_dir"
+            ],
+            "SubModules.WiredTableCellsDetection.model_name": self._params[
+                "wired_table_cells_detection_model_name"
+            ],
+            "SubModules.WiredTableCellsDetection.model_dir": self._params[
+                "wired_table_cells_detection_model_dir"
+            ],
+            "SubModules.WirelessTableCellsDetection.model_name": self._params[
+                "wireless_table_cells_detection_model_name"
+            ],
+            "SubModules.WirelessTableCellsDetection.model_dir": self._params[
+                "wireless_table_cells_detection_model_dir"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify.model_name": self._params[
+                "doc_orientation_classify_model_name"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocOrientationClassify.model_dir": self._params[
+                "doc_orientation_classify_model_dir"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_name": self._params[
+                "doc_unwarping_model_name"
+            ],
+            "SubPipelines.DocPreprocessor.SubModules.DocUnwarping.model_dir": self._params[
+                "doc_unwarping_model_dir"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextDetection.model_name": self._params[
+                "text_detection_model_name"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextDetection.model_dir": self._params[
+                "text_detection_model_dir"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextDetection.limit_side_len": self._params[
+                "text_det_limit_side_len"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextDetection.limit_type": self._params[
+                "text_det_limit_type"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextDetection.thresh": self._params[
+                "text_det_thresh"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextDetection.box_thresh": self._params[
+                "text_det_box_thresh"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextDetection.unclip_ratio": self._params[
+                "text_det_unclip_ratio"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextRecognition.model_name": self._params[
+                "text_recognition_model_name"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextRecognition.model_dir": self._params[
+                "text_recognition_model_dir"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextRecognition.batch_size": self._params[
+                "text_recognition_batch_size"
+            ],
+            "SubPipelines.GeneralOCR.SubModules.TextRecognition.score_thresh": self._params[
+                "text_rec_score_thresh"
+            ],
+        }
+        return create_config_from_structure(STRUCTURE)
+
+
+class TableRecognitionPipelineV2CLISubcommandExecutor(PipelineCLISubcommandExecutor):
+    @property
+    def subparser_name(self):
+        return "table_recognition_v2"
+
+    def _update_subparser(self, subparser):
+        add_simple_inference_args(subparser)
+
+        subparser.add_argument(
+            "--layout_detection_model_name",
+            type=str,
+            help="Name of the layout detection model.",
+        )
+        subparser.add_argument(
+            "--layout_detection_model_dir",
+            type=str,
+            help="Path to the layout detection model directory.",
+        )
+        subparser.add_argument(
+            "--table_classification_model_name",
+            type=str,
+            help="Name of the table classification model.",
+        )
+        subparser.add_argument(
+            "--table_classification_model_dir",
+            type=str,
+            help="Path to the table classification model directory.",
+        )
+        subparser.add_argument(
+            "--wired_table_structure_recognition_model_name",
+            type=str,
+            help="Name of the wired table structure recognition model.",
+        )
+        subparser.add_argument(
+            "--wired_table_structure_recognition_model_dir",
+            type=str,
+            help="Path to the wired table structure recognition model directory.",
+        )
+        subparser.add_argument(
+            "--wireless_table_structure_recognition_model_name",
+            type=str,
+            help="Name of the wireless table structure recognition model.",
+        )
+        subparser.add_argument(
+            "--wireless_table_structure_recognition_model_dir",
+            type=str,
+            help="Path to the wired table structure recognition model directory.",
+        )
+        subparser.add_argument(
+            "--wired_table_cells_detection_model_name",
+            type=str,
+            help="Name of the wired table cells detection model.",
+        )
+        subparser.add_argument(
+            "--wired_table_cells_detection_model_dir",
+            type=str,
+            help="Path to the wired table cells detection model directory.",
+        )
+        subparser.add_argument(
+            "--wireless_table_cells_detection_model_name",
+            type=str,
+            help="Name of the wireless table cells detection model.",
+        )
+        subparser.add_argument(
+            "--wireless_table_cells_detection_model_dir",
+            type=str,
+            help="Path to the wireless table cells detection model directory.",
+        )
+
+        subparser.add_argument(
+            "--doc_orientation_classify_model_name",
+            type=str,
+            help="Name of the document image orientation classification model.",
+        )
+        subparser.add_argument(
+            "--doc_orientation_classify_model_dir",
+            type=str,
+            help="Path to the document image orientation classification model directory.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_name",
+            type=str,
+            help="Name of the text image unwarping model.",
+        )
+        subparser.add_argument(
+            "--doc_unwarping_model_dir",
+            type=str,
+            help="Path to the image unwarping model directory.",
+        )
+
+        subparser.add_argument(
+            "--text_detection_model_name",
+            type=str,
+            help="Name of the text detection model.",
+        )
+        subparser.add_argument(
+            "--text_detection_model_dir",
+            type=str,
+            help="Path to the text detection model directory.",
+        )
+        subparser.add_argument(
+            "--text_det_limit_side_len",
+            type=int,
+            help="This sets a limit on the side length of the input image for the text detection model.",
+        )
+        subparser.add_argument(
+            "--text_det_limit_type",
+            type=str,
+            help="This determines how the side length limit is applied to the input image before feeding it into the text deteciton model.",
+        )
+        subparser.add_argument(
+            "--text_det_thresh",
+            type=float,
+            help="Detection pixel threshold for the text detection model. Pixels with scores greater than this threshold in the output probability map are considered text pixels.",
+        )
+        subparser.add_argument(
+            "--text_det_box_thresh",
+            type=float,
+            help="Detection box threshold for the text detection model. A detection result is considered a text region if the average score of all pixels within the border of the result is greater than this threshold.",
+        )
+        subparser.add_argument(
+            "--text_det_unclip_ratio",
+            type=float,
+            help="Text detection expansion coefficient, which expands the text region using this method. The larger the value, the larger the expansion area.",
+        )
+        subparser.add_argument(
+            "--text_recognition_model_name",
+            type=str,
+            help="Name of the text recognition model.",
+        )
+        subparser.add_argument(
+            "--text_recognition_model_dir",
+            type=str,
+            help="Path to the text recognition model directory.",
+        )
+        subparser.add_argument(
+            "--text_recognition_batch_size",
+            type=int,
+            help="Batch size for the text recognition model.",
+        )
+        subparser.add_argument(
+            "--text_rec_score_thresh",
+            type=float,
+            help="Text recognition threshold used in general OCR. Text results with scores greater than this threshold are retained.",
+        )
+
+        subparser.add_argument(
+            "--use_doc_orientation_classify",
+            type=str2bool,
+            help="Whether to use document image orientation classification.",
+        )
+        subparser.add_argument(
+            "--use_doc_unwarping",
+            type=str2bool,
+            help="Whether to use text image unwarping.",
+        )
+        subparser.add_argument(
+            "--use_layout_detection",
+            type=str2bool,
+            help="Whether to use layout detection.",
+        )
+        subparser.add_argument(
+            "--use_ocr_model",
+            type=str2bool,
+            help="Whether to use OCR models.",
+        )
+
+    def execute_with_args(self, args):
+        params = get_subcommand_args(args)
+        perform_simple_inference(TableRecognitionPipelineV2, params)
diff --git a/paddleocr/_pipelines/utils.py b/paddleocr/_pipelines/utils.py
new file mode 100644
index 0000000..9a7b4ba
--- /dev/null
+++ b/paddleocr/_pipelines/utils.py
@@ -0,0 +1,30 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+def create_config_from_structure(structure, *, unset=None, config=None):
+    if config is None:
+        config = {}
+    for k, v in structure.items():
+        if v is unset:
+            continue
+        idx = k.find(".")
+        if idx == -1:
+            config[k] = v
+        else:
+            sk = k[:idx]
+            if sk not in config:
+                config[sk] = {}
+            create_config_from_structure({k[idx + 1 :]: v}, config=config[sk])
+    return config
diff --git a/paddleocr/_utils/__init__.py b/paddleocr/_utils/__init__.py
new file mode 100644
index 0000000..a9cc79c
--- /dev/null
+++ b/paddleocr/_utils/__init__.py
@@ -0,0 +1,13 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/paddleocr/_utils/cli.py b/paddleocr/_utils/cli.py
new file mode 100644
index 0000000..6d21875
--- /dev/null
+++ b/paddleocr/_utils/cli.py
@@ -0,0 +1,75 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import time
+
+from .logging import logger
+
+
+def str2bool(v, /):
+    return v.lower() in ("true", "yes", "t", "y", "1")
+
+
+def get_subcommand_args(args):
+    args = vars(args).copy()
+    args.pop("subcommand")
+    args.pop("executor")
+    return args
+
+
+def add_simple_inference_args(subparser, *, input_help=None):
+    if input_help is None:
+        input_help = "Input path or URL."
+    subparser.add_argument(
+        "-i",
+        "--input",
+        type=str,
+        required=True,
+        help=input_help,
+    )
+    subparser.add_argument(
+        "--save_path",
+        type=str,
+        help="Path to the output directory.",
+    )
+
+
+def perform_simple_inference(wrapper_cls, params, predict_param_names=None):
+    params = params.copy()
+
+    input_ = params.pop("input")
+    save_path = params.pop("save_path")
+
+    if predict_param_names is not None:
+        predict_params = {}
+        for name in predict_param_names:
+            predict_params[name] = params.pop(name)
+    else:
+        predict_params = {}
+    init_params = params
+
+    wrapper = wrapper_cls(**init_params)
+
+    try:
+        result = wrapper.predict_iter(input_, **predict_params)
+
+        t1 = time.time()
+        for i, res in enumerate(result):
+            logger.info(f"Processed item {i} in {(time.time()-t1) * 1000} ms")
+            t1 = time.time()
+            res.print()
+            if save_path:
+                res.save_all(save_path)
+    finally:
+        wrapper.close()
diff --git a/paddleocr/_utils/deprecation.py b/paddleocr/_utils/deprecation.py
new file mode 100644
index 0000000..b30419d
--- /dev/null
+++ b/paddleocr/_utils/deprecation.py
@@ -0,0 +1,42 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import argparse
+import sys
+import warnings
+
+from typing_extensions import deprecated as deprecated
+
+
+class CLIDeprecationWarning(DeprecationWarning):
+    pass
+
+
+class DeprecatedOptionAction(argparse.Action):
+    def __call__(self, parser, namespace, values, option_string=None):
+        assert option_string
+        warnings.warn(
+            f"The option `{option_string}` has been deprecated and will be removed in the future. Please refer to the documentation for more details.",
+            CLIDeprecationWarning,
+        )
+        setattr(namespace, self.dest, values)
+
+
+def warn_deprecated_param(name, new_name=None):
+    msg = (
+        f"The parameter `{name}` has been deprecated and will be removed in the future."
+    )
+    if new_name is not None:
+        msg += f" Please use `{new_name}` instead."
+    warnings.warn(msg, DeprecationWarning, stacklevel=3)
diff --git a/paddleocr/_utils/langs.py b/paddleocr/_utils/langs.py
new file mode 100644
index 0000000..77e7d00
--- /dev/null
+++ b/paddleocr/_utils/langs.py
@@ -0,0 +1,126 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Language code groupings shared across OCR pipelines."""
+
+LATIN_LANGS = frozenset(
+    {
+        "af",
+        "az",
+        "bs",
+        "cs",
+        "cy",
+        "da",
+        "de",
+        "es",
+        "et",
+        "fr",
+        "ga",
+        "hr",
+        "hu",
+        "id",
+        "is",
+        "it",
+        "ku",
+        "la",
+        "lt",
+        "lv",
+        "mi",
+        "ms",
+        "mt",
+        "nl",
+        "no",
+        "oc",
+        "pi",
+        "pl",
+        "pt",
+        "ro",
+        "rs_latin",
+        "sk",
+        "sl",
+        "sq",
+        "sv",
+        "sw",
+        "tl",
+        "tr",
+        "uz",
+        "vi",
+        "french",
+        "german",
+        "fi",
+        "eu",
+        "gl",
+        "lb",
+        "rm",
+        "ca",
+        "qu",
+    }
+)
+
+ARABIC_LANGS = frozenset({"ar", "fa", "ug", "ur", "ps", "ku", "sd", "bal"})
+ESLAV_LANGS = frozenset({"ru", "be", "uk"})
+CYRILLIC_LANGS = frozenset(
+    {
+        "ru",
+        "rs_cyrillic",
+        "be",
+        "bg",
+        "uk",
+        "mn",
+        "abq",
+        "ady",
+        "kbd",
+        "ava",
+        "dar",
+        "inh",
+        "che",
+        "lbe",
+        "lez",
+        "tab",
+        "kk",
+        "ky",
+        "tg",
+        "mk",
+        "tt",
+        "cv",
+        "ba",
+        "mhr",
+        "mo",
+        "udm",
+        "kv",
+        "os",
+        "bua",
+        "xal",
+        "tyv",
+        "sah",
+        "kaa",
+    }
+)
+DEVANAGARI_LANGS = frozenset(
+    {
+        "hi",
+        "mr",
+        "ne",
+        "bh",
+        "mai",
+        "ang",
+        "bho",
+        "mah",
+        "sck",
+        "new",
+        "gom",
+        "sa",
+        "bgc",
+    }
+)
diff --git a/paddleocr/_utils/logging.py b/paddleocr/_utils/logging.py
new file mode 100644
index 0000000..0ef8939
--- /dev/null
+++ b/paddleocr/_utils/logging.py
@@ -0,0 +1,39 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+
+from .._env import DISABLE_AUTO_LOGGING_CONFIG
+
+LOGGER_NAME = "paddleocr"
+
+logger = logging.getLogger(LOGGER_NAME)
+
+
+def _set_up_logger():
+    if DISABLE_AUTO_LOGGING_CONFIG:
+        return
+
+    # Basically compatible with PaddleOCR 2.x, except for logging to stderr
+    formatter = logging.Formatter(
+        "[%(asctime)s] %(name)s %(levelname)s: %(message)s", datefmt="%Y/%m/%d %H:%M:%S"
+    )
+    stream_handler = logging.StreamHandler()
+    stream_handler.setFormatter(formatter)
+    logger.addHandler(stream_handler)
+    logger.setLevel(logging.ERROR)
+    logger.propagate = False
+
+
+_set_up_logger()
diff --git a/paddleocr/_utils/naming.py b/paddleocr/_utils/naming.py
new file mode 100644
index 0000000..94817f6
--- /dev/null
+++ b/paddleocr/_utils/naming.py
@@ -0,0 +1,30 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import Any, Dict
+
+
+def snake_to_camel(name: str) -> str:
+    if not name:
+        return name
+    components = name.split("_")
+    if len(components) == 1:
+        return name
+    return components[0] + "".join(
+        component.capitalize() for component in components[1:]
+    )
+
+
+def snake_keys_to_camel(params: Dict[str, Any]) -> Dict[str, Any]:
+    return {snake_to_camel(key): value for key, value in params.items()}
diff --git a/paddleocr/_version.py b/paddleocr/_version.py
new file mode 100644
index 0000000..91e943c
--- /dev/null
+++ b/paddleocr/_version.py
@@ -0,0 +1,20 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import importlib.metadata
+
+try:
+    version = importlib.metadata.version(__package__)
+except importlib.metadata.PackageNotFoundError:
+    version = "0.0.0"
diff --git a/ppocr/__init__.py b/ppocr/__init__.py
new file mode 100755
index 0000000..d0c32e2
--- /dev/null
+++ b/ppocr/__init__.py
@@ -0,0 +1,13 @@
+# Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/ppocr/data/__init__.py b/ppocr/data/__init__.py
new file mode 100644
index 0000000..f288989
--- /dev/null
+++ b/ppocr/data/__init__.py
@@ -0,0 +1,154 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+import os
+import sys
+import numpy as np
+import skimage
+import paddle
+import signal
+import random
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(os.path.abspath(os.path.join(__dir__, "../..")))
+
+import copy
+from paddle.io import Dataset, DataLoader, BatchSampler, DistributedBatchSampler
+import paddle.distributed as dist
+
+from ppocr.data.imaug import transform, create_operators
+from ppocr.data.simple_dataset import SimpleDataSet, MultiScaleDataSet
+from ppocr.data.lmdb_dataset import LMDBDataSet, LMDBDataSetSR, LMDBDataSetTableMaster
+from ppocr.data.pgnet_dataset import PGDataSet
+from ppocr.data.pubtab_dataset import PubTabDataSet
+from ppocr.data.multi_scale_sampler import MultiScaleSampler
+from ppocr.data.latexocr_dataset import LaTeXOCRDataSet
+
+# for PaddleX dataset_type
+TextDetDataset = SimpleDataSet
+TextRecDataset = SimpleDataSet
+MSTextRecDataset = MultiScaleDataSet
+PubTabTableRecDataset = PubTabDataSet
+KieDataset = SimpleDataSet
+LaTeXOCRDataSet = LaTeXOCRDataSet
+
+__all__ = ["build_dataloader", "transform", "create_operators", "set_signal_handlers"]
+
+
+def term_mp(sig_num, frame):
+    """kill all child processes"""
+    pid = os.getpid()
+    pgid = os.getpgid(os.getpid())
+    print("main proc {} exit, kill process group " "{}".format(pid, pgid))
+    os.killpg(pgid, signal.SIGKILL)
+
+
+def set_signal_handlers():
+    pid = os.getpid()
+    try:
+        pgid = os.getpgid(pid)
+    except AttributeError:
+        # In case `os.getpgid` is not available, no signal handler will be set,
+        # because we cannot do safe cleanup.
+        pass
+    else:
+        # XXX: `term_mp` kills all processes in the process group, which in
+        # some cases includes the parent process of current process and may
+        # cause unexpected results. To solve this problem, we set signal
+        # handlers only when current process is the group leader. In the
+        # future, it would be better to consider killing only descendants of
+        # the current process.
+        if pid == pgid:
+            # support exit using ctrl+c
+            signal.signal(signal.SIGINT, term_mp)
+            signal.signal(signal.SIGTERM, term_mp)
+
+
+def build_dataloader(config, mode, device, logger, seed=None):
+    config = copy.deepcopy(config)
+
+    support_dict = [
+        "SimpleDataSet",
+        "LMDBDataSet",
+        "PGDataSet",
+        "PubTabDataSet",
+        "LMDBDataSetSR",
+        "LMDBDataSetTableMaster",
+        "MultiScaleDataSet",
+        "TextDetDataset",
+        "TextRecDataset",
+        "MSTextRecDataset",
+        "PubTabTableRecDataset",
+        "KieDataset",
+        "LaTeXOCRDataSet",
+    ]
+    module_name = config[mode]["dataset"]["name"]
+    assert module_name in support_dict, Exception(
+        "DataSet only support {}".format(support_dict)
+    )
+    assert mode in ["Train", "Eval", "Test"], "Mode should be Train, Eval or Test."
+
+    dataset = eval(module_name)(config, mode, logger, seed)
+    loader_config = config[mode]["loader"]
+    batch_size = loader_config["batch_size_per_card"]
+    drop_last = loader_config["drop_last"]
+    shuffle = loader_config["shuffle"]
+    num_workers = loader_config["num_workers"]
+    if "use_shared_memory" in loader_config.keys():
+        use_shared_memory = loader_config["use_shared_memory"]
+    else:
+        use_shared_memory = True
+
+    if mode == "Train":
+        # Distribute data to multiple cards
+        if "sampler" in config[mode]:
+            config_sampler = config[mode]["sampler"]
+            sampler_name = config_sampler.pop("name")
+            batch_sampler = eval(sampler_name)(dataset, **config_sampler)
+        else:
+            batch_sampler = DistributedBatchSampler(
+                dataset=dataset,
+                batch_size=batch_size,
+                shuffle=shuffle,
+                drop_last=drop_last,
+            )
+    else:
+        # Distribute data to single card
+        batch_sampler = BatchSampler(
+            dataset=dataset, batch_size=batch_size, shuffle=shuffle, drop_last=drop_last
+        )
+
+    if "collate_fn" in loader_config:
+        from . import collate_fn
+
+        collate_fn = getattr(collate_fn, loader_config["collate_fn"])()
+    else:
+        collate_fn = None
+    data_loader = DataLoader(
+        dataset=dataset,
+        batch_sampler=batch_sampler,
+        places=device,
+        num_workers=num_workers,
+        return_list=True,
+        use_shared_memory=use_shared_memory,
+        collate_fn=collate_fn,
+        persistent_workers=True if num_workers > 0 and mode == "Train" else False,
+    )
+
+    return data_loader
diff --git a/ppocr/data/collate_fn.py b/ppocr/data/collate_fn.py
new file mode 100644
index 0000000..309f60a
--- /dev/null
+++ b/ppocr/data/collate_fn.py
@@ -0,0 +1,173 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle
+import numbers
+import numpy as np
+from collections import defaultdict
+
+
+class DictCollator(object):
+    """
+    data batch
+    """
+
+    def __call__(self, batch):
+        # todo：support batch operators
+        data_dict = defaultdict(list)
+        to_tensor_keys = []
+        for sample in batch:
+            for k, v in sample.items():
+                if isinstance(v, (np.ndarray, paddle.Tensor, numbers.Number)):
+                    if k not in to_tensor_keys:
+                        to_tensor_keys.append(k)
+                data_dict[k].append(v)
+        for k in to_tensor_keys:
+            data_dict[k] = paddle.to_tensor(data_dict[k])
+        return data_dict
+
+
+class ListCollator(object):
+    """
+    data batch
+    """
+
+    def __call__(self, batch):
+        # todo：support batch operators
+        data_dict = defaultdict(list)
+        to_tensor_idxs = []
+        for sample in batch:
+            for idx, v in enumerate(sample):
+                if isinstance(v, (np.ndarray, paddle.Tensor, numbers.Number)):
+                    if idx not in to_tensor_idxs:
+                        to_tensor_idxs.append(idx)
+                data_dict[idx].append(v)
+        for idx in to_tensor_idxs:
+            data_dict[idx] = paddle.to_tensor(data_dict[idx])
+        return list(data_dict.values())
+
+
+class SSLRotateCollate(object):
+    """
+    bach: [
+        [(4*3xH*W), (4,)]
+        [(4*3xH*W), (4,)]
+        ...
+    ]
+    """
+
+    def __call__(self, batch):
+        output = [np.concatenate(d, axis=0) for d in zip(*batch)]
+        return output
+
+
+class DyMaskCollator(object):
+    """
+    batch: [
+        image [batch_size, channel, maxHinbatch, maxWinbatch]
+        image_mask [batch_size, channel, maxHinbatch, maxWinbatch]
+        label [batch_size, maxLabelLen]
+        label_mask [batch_size, maxLabelLen]
+        ...
+    ]
+    """
+
+    def __call__(self, batch):
+        max_width, max_height, max_length = 0, 0, 0
+        bs, channel = len(batch), batch[0][0].shape[0]
+        proper_items = []
+        for item in batch:
+            if (
+                item[0].shape[1] * max_width > 1600 * 320
+                or item[0].shape[2] * max_height > 1600 * 320
+            ):
+                continue
+            max_height = (
+                item[0].shape[1] if item[0].shape[1] > max_height else max_height
+            )
+            max_width = item[0].shape[2] if item[0].shape[2] > max_width else max_width
+            max_length = len(item[1]) if len(item[1]) > max_length else max_length
+            proper_items.append(item)
+
+        images, image_masks = np.zeros(
+            (len(proper_items), channel, max_height, max_width), dtype="float32"
+        ), np.zeros((len(proper_items), 1, max_height, max_width), dtype="float32")
+        labels, label_masks = np.zeros(
+            (len(proper_items), max_length), dtype="int64"
+        ), np.zeros((len(proper_items), max_length), dtype="int64")
+
+        for i in range(len(proper_items)):
+            _, h, w = proper_items[i][0].shape
+            images[i][:, :h, :w] = proper_items[i][0]
+            image_masks[i][:, :h, :w] = 1
+            l = len(proper_items[i][1])
+            labels[i][:l] = proper_items[i][1]
+            label_masks[i][:l] = 1
+
+        return images, image_masks, labels, label_masks
+
+
+class LaTeXOCRCollator(object):
+    """
+    batch: [
+        image [batch_size, channel, maxHinbatch, maxWinbatch]
+        label [batch_size, maxLabelLen]
+        label_mask [batch_size, maxLabelLen]
+        ...
+    ]
+    """
+
+    def __call__(self, batch):
+        images, labels, attention_mask = batch[0]
+        return images, labels, attention_mask
+
+
+class UniMERNetCollator(object):
+    """
+    batch: [
+        image [batch_size, channel, maxHinbatch, maxWinbatch]
+        image_mask [batch_size, channel, maxHinbatch, maxWinbatch]
+        label [batch_size, maxLabelLen]
+        label_mask [batch_size, maxLabelLen]
+        ...
+    ]
+    """
+
+    def __call__(self, batch):
+
+        max_width, max_height, max_length = 0, 0, 0
+        bs, channel = len(batch), batch[0][0].shape[0]
+        proper_items = []
+        for item in batch:
+            max_height = (
+                item[0].shape[1] if item[0].shape[1] > max_height else max_height
+            )
+            max_width = item[0].shape[2] if item[0].shape[2] > max_width else max_width
+            max_length = len(item[1]) if len(item[1]) > max_length else max_length
+            proper_items.append(item)
+
+        images = np.ones(
+            (len(proper_items), channel, max_height, max_width), dtype="float32"
+        )
+
+        labels, label_masks = np.ones(
+            (len(proper_items), max_length), dtype="int64"
+        ), np.zeros((len(proper_items), max_length), dtype="int64")
+        for i in range(len(proper_items)):
+            _, h, w = proper_items[i][0].shape
+            images[i][:, :h, :w] = proper_items[i][0]
+            l = len(proper_items[i][1])
+            labels[i][:l] = proper_items[i][1]
+            label_masks[i][:l] = proper_items[i][2]
+        return images, labels, label_masks
diff --git a/ppocr/data/imaug/ColorJitter.py b/ppocr/data/imaug/ColorJitter.py
new file mode 100644
index 0000000..46c1955
--- /dev/null
+++ b/ppocr/data/imaug/ColorJitter.py
@@ -0,0 +1,27 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from paddle.vision.transforms import ColorJitter as pp_ColorJitter
+
+__all__ = ["ColorJitter"]
+
+
+class ColorJitter(object):
+    def __init__(self, brightness=0, contrast=0, saturation=0, hue=0, **kwargs):
+        self.aug = pp_ColorJitter(brightness, contrast, saturation, hue)
+
+    def __call__(self, data):
+        image = data["image"]
+        image = self.aug(image)
+        data["image"] = image
+        return data
diff --git a/ppocr/data/imaug/__init__.py b/ppocr/data/imaug/__init__.py
new file mode 100644
index 0000000..06ac51a
--- /dev/null
+++ b/ppocr/data/imaug/__init__.py
@@ -0,0 +1,96 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+from .iaa_augment import IaaAugment
+from .make_border_map import MakeBorderMap
+from .make_shrink_map import MakeShrinkMap
+from .random_crop_data import EastRandomCropData, RandomCrop, RandomCropImgMask
+from .make_pse_gt import MakePseGt
+
+
+from .rec_img_aug import (
+    BaseDataAugmentation,
+    RecAug,
+    RecConAug,
+    RecResizeImg,
+    ClsResizeImg,
+    SRNRecResizeImg,
+    GrayRecResizeImg,
+    SARRecResizeImg,
+    PRENResizeImg,
+    ABINetRecResizeImg,
+    SVTRRecResizeImg,
+    ABINetRecAug,
+    VLRecResizeImg,
+    SPINRecResizeImg,
+    RobustScannerRecResizeImg,
+    RFLRecResizeImg,
+    SVTRRecAug,
+    ParseQRecAug,
+)
+from .ssl_img_aug import SSLRotateResize
+from .randaugment import RandAugment
+from .copy_paste import CopyPaste
+from .ColorJitter import ColorJitter
+from .operators import *
+from .label_ops import *
+
+from .east_process import *
+from .sast_process import *
+from .pg_process import *
+from .table_ops import *
+
+from .vqa import *
+
+from .fce_aug import *
+from .fce_targets import FCENetTargets
+from .ct_process import *
+from .drrg_targets import DRRGTargets
+from .latex_ocr_aug import *
+from .unimernet_aug import *
+
+
+def transform(data, ops=None):
+    """transform"""
+    if ops is None:
+        ops = []
+    for op in ops:
+        data = op(data)
+        if data is None:
+            return None
+    return data
+
+
+def create_operators(op_param_list, global_config=None):
+    """
+    create operators based on the config
+
+    Args:
+        params(list): a dict list, used to create some operators
+    """
+    assert isinstance(op_param_list, list), "operator config should be a list"
+    ops = []
+    for operator in op_param_list:
+        assert isinstance(operator, dict) and len(operator) == 1, "yaml format error"
+        op_name = list(operator)[0]
+        param = {} if operator[op_name] is None else operator[op_name]
+        if global_config is not None:
+            param.update(global_config)
+        op = eval(op_name)(**param)
+        ops.append(op)
+    return ops
diff --git a/ppocr/data/imaug/abinet_aug.py b/ppocr/data/imaug/abinet_aug.py
new file mode 100644
index 0000000..3df255e
--- /dev/null
+++ b/ppocr/data/imaug/abinet_aug.py
@@ -0,0 +1,512 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/FangShancheng/ABINet/blob/main/transforms.py
+"""
+import math
+import numbers
+import random
+
+import cv2
+import numpy as np
+from paddle.vision.transforms import Compose, ColorJitter
+
+
+def sample_asym(magnitude, size=None):
+    return np.random.beta(1, 4, size) * magnitude
+
+
+def sample_sym(magnitude, size=None):
+    return (np.random.beta(4, 4, size=size) - 0.5) * 2 * magnitude
+
+
+def sample_uniform(low, high, size=None):
+    return np.random.uniform(low, high, size=size)
+
+
+def get_interpolation(type="random"):
+    if type == "random":
+        choice = [cv2.INTER_NEAREST, cv2.INTER_LINEAR, cv2.INTER_CUBIC, cv2.INTER_AREA]
+        interpolation = choice[random.randint(0, len(choice) - 1)]
+    elif type == "nearest":
+        interpolation = cv2.INTER_NEAREST
+    elif type == "linear":
+        interpolation = cv2.INTER_LINEAR
+    elif type == "cubic":
+        interpolation = cv2.INTER_CUBIC
+    elif type == "area":
+        interpolation = cv2.INTER_AREA
+    else:
+        raise TypeError(
+            "Interpolation types only nearest, linear, cubic, area are supported!"
+        )
+    return interpolation
+
+
+class CVRandomRotation(object):
+    def __init__(self, degrees=15):
+        assert isinstance(degrees, numbers.Number), "degree should be a single number."
+        assert degrees >= 0, "degree must be positive."
+        self.degrees = degrees
+
+    @staticmethod
+    def get_params(degrees):
+        return sample_sym(degrees)
+
+    def __call__(self, img):
+        angle = self.get_params(self.degrees)
+        src_h, src_w = img.shape[:2]
+        M = cv2.getRotationMatrix2D(
+            center=(src_w / 2, src_h / 2), angle=angle, scale=1.0
+        )
+        abs_cos, abs_sin = abs(M[0, 0]), abs(M[0, 1])
+        dst_w = int(src_h * abs_sin + src_w * abs_cos)
+        dst_h = int(src_h * abs_cos + src_w * abs_sin)
+        M[0, 2] += (dst_w - src_w) / 2
+        M[1, 2] += (dst_h - src_h) / 2
+
+        flags = get_interpolation()
+        return cv2.warpAffine(
+            img, M, (dst_w, dst_h), flags=flags, borderMode=cv2.BORDER_REPLICATE
+        )
+
+
+class CVRandomAffine(object):
+    def __init__(self, degrees, translate=None, scale=None, shear=None):
+        assert isinstance(degrees, numbers.Number), "degree should be a single number."
+        assert degrees >= 0, "degree must be positive."
+        self.degrees = degrees
+
+        if translate is not None:
+            assert (
+                isinstance(translate, (tuple, list)) and len(translate) == 2
+            ), "translate should be a list or tuple and it must be of length 2."
+            for t in translate:
+                if not (0.0 <= t <= 1.0):
+                    raise ValueError("translation values should be between 0 and 1")
+        self.translate = translate
+
+        if scale is not None:
+            assert (
+                isinstance(scale, (tuple, list)) and len(scale) == 2
+            ), "scale should be a list or tuple and it must be of length 2."
+            for s in scale:
+                if s <= 0:
+                    raise ValueError("scale values should be positive")
+        self.scale = scale
+
+        if shear is not None:
+            if isinstance(shear, numbers.Number):
+                if shear < 0:
+                    raise ValueError(
+                        "If shear is a single number, it must be positive."
+                    )
+                self.shear = [shear]
+            else:
+                assert isinstance(shear, (tuple, list)) and (
+                    len(shear) == 2
+                ), "shear should be a list or tuple and it must be of length 2."
+                self.shear = shear
+        else:
+            self.shear = shear
+
+    def _get_inverse_affine_matrix(self, center, angle, translate, scale, shear):
+        # https://github.com/pytorch/vision/blob/v0.4.0/torchvision/transforms/functional.py#L717
+        from numpy import sin, cos, tan
+
+        if isinstance(shear, numbers.Number):
+            shear = [shear, 0]
+
+        if not isinstance(shear, (tuple, list)) and len(shear) == 2:
+            raise ValueError(
+                "Shear should be a single value or a tuple/list containing "
+                + "two values. Got {}".format(shear)
+            )
+
+        rot = math.radians(angle)
+        sx, sy = [math.radians(s) for s in shear]
+
+        cx, cy = center
+        tx, ty = translate
+
+        # RSS without scaling
+        a = cos(rot - sy) / cos(sy)
+        b = -cos(rot - sy) * tan(sx) / cos(sy) - sin(rot)
+        c = sin(rot - sy) / cos(sy)
+        d = -sin(rot - sy) * tan(sx) / cos(sy) + cos(rot)
+
+        # Inverted rotation matrix with scale and shear
+        # det([[a, b], [c, d]]) == 1, since det(rotation) = 1 and det(shear) = 1
+        M = [d, -b, 0, -c, a, 0]
+        M = [x / scale for x in M]
+
+        # Apply inverse of translation and of center translation: RSS^-1 * C^-1 * T^-1
+        M[2] += M[0] * (-cx - tx) + M[1] * (-cy - ty)
+        M[5] += M[3] * (-cx - tx) + M[4] * (-cy - ty)
+
+        # Apply center translation: C * RSS^-1 * C^-1 * T^-1
+        M[2] += cx
+        M[5] += cy
+        return M
+
+    @staticmethod
+    def get_params(degrees, translate, scale_ranges, shears, height):
+        angle = sample_sym(degrees)
+        if translate is not None:
+            max_dx = translate[0] * height
+            max_dy = translate[1] * height
+            translations = (np.round(sample_sym(max_dx)), np.round(sample_sym(max_dy)))
+        else:
+            translations = (0, 0)
+
+        if scale_ranges is not None:
+            scale = sample_uniform(scale_ranges[0], scale_ranges[1])
+        else:
+            scale = 1.0
+
+        if shears is not None:
+            if len(shears) == 1:
+                shear = [sample_sym(shears[0]), 0.0]
+            elif len(shears) == 2:
+                shear = [sample_sym(shears[0]), sample_sym(shears[1])]
+        else:
+            shear = 0.0
+
+        return angle, translations, scale, shear
+
+    def __call__(self, img):
+        src_h, src_w = img.shape[:2]
+        angle, translate, scale, shear = self.get_params(
+            self.degrees, self.translate, self.scale, self.shear, src_h
+        )
+
+        M = self._get_inverse_affine_matrix(
+            (src_w / 2, src_h / 2), angle, (0, 0), scale, shear
+        )
+        M = np.array(M).reshape(2, 3)
+
+        startpoints = [(0, 0), (src_w - 1, 0), (src_w - 1, src_h - 1), (0, src_h - 1)]
+        project = lambda x, y, a, b, c: int(a * x + b * y + c)
+        endpoints = [
+            (project(x, y, *M[0]), project(x, y, *M[1])) for x, y in startpoints
+        ]
+
+        rect = cv2.minAreaRect(np.array(endpoints))
+        bbox = cv2.boxPoints(rect).astype(dtype=np.int32)
+        max_x, max_y = bbox[:, 0].max(), bbox[:, 1].max()
+        min_x, min_y = bbox[:, 0].min(), bbox[:, 1].min()
+
+        dst_w = int(max_x - min_x)
+        dst_h = int(max_y - min_y)
+        M[0, 2] += (dst_w - src_w) / 2
+        M[1, 2] += (dst_h - src_h) / 2
+
+        # add translate
+        dst_w += int(abs(translate[0]))
+        dst_h += int(abs(translate[1]))
+        if translate[0] < 0:
+            M[0, 2] += abs(translate[0])
+        if translate[1] < 0:
+            M[1, 2] += abs(translate[1])
+
+        flags = get_interpolation()
+        return cv2.warpAffine(
+            img, M, (dst_w, dst_h), flags=flags, borderMode=cv2.BORDER_REPLICATE
+        )
+
+
+class CVRandomPerspective(object):
+    def __init__(self, distortion=0.5):
+        self.distortion = distortion
+
+    def get_params(self, width, height, distortion):
+        offset_h = sample_asym(distortion * height / 2, size=4).astype(dtype=np.int32)
+        offset_w = sample_asym(distortion * width / 2, size=4).astype(dtype=np.int32)
+        topleft = (offset_w[0], offset_h[0])
+        topright = (width - 1 - offset_w[1], offset_h[1])
+        botright = (width - 1 - offset_w[2], height - 1 - offset_h[2])
+        botleft = (offset_w[3], height - 1 - offset_h[3])
+
+        startpoints = [(0, 0), (width - 1, 0), (width - 1, height - 1), (0, height - 1)]
+        endpoints = [topleft, topright, botright, botleft]
+        return np.array(startpoints, dtype=np.float32), np.array(
+            endpoints, dtype=np.float32
+        )
+
+    def __call__(self, img):
+        height, width = img.shape[:2]
+        startpoints, endpoints = self.get_params(width, height, self.distortion)
+        M = cv2.getPerspectiveTransform(startpoints, endpoints)
+
+        # TODO: more robust way to crop image
+        rect = cv2.minAreaRect(endpoints)
+        bbox = cv2.boxPoints(rect).astype(dtype=np.int32)
+        max_x, max_y = bbox[:, 0].max(), bbox[:, 1].max()
+        min_x, min_y = bbox[:, 0].min(), bbox[:, 1].min()
+        min_x, min_y = max(min_x, 0), max(min_y, 0)
+
+        flags = get_interpolation()
+        img = cv2.warpPerspective(
+            img, M, (max_x, max_y), flags=flags, borderMode=cv2.BORDER_REPLICATE
+        )
+        img = img[min_y:, min_x:]
+        return img
+
+
+class CVRescale(object):
+    def __init__(self, factor=4, base_size=(128, 512)):
+        """Define image scales using gaussian pyramid and rescale image to target scale.
+
+        Args:
+            factor: the decayed factor from base size, factor=4 keeps target scale by default.
+            base_size: base size the build the bottom layer of pyramid
+        """
+        if isinstance(factor, numbers.Number):
+            self.factor = round(sample_uniform(0, factor))
+        elif isinstance(factor, (tuple, list)) and len(factor) == 2:
+            self.factor = round(sample_uniform(factor[0], factor[1]))
+        else:
+            raise Exception("factor must be number or list with length 2")
+        # assert factor is valid
+        self.base_h, self.base_w = base_size[:2]
+
+    def __call__(self, img):
+        if self.factor == 0:
+            return img
+        src_h, src_w = img.shape[:2]
+        cur_w, cur_h = self.base_w, self.base_h
+        scale_img = cv2.resize(img, (cur_w, cur_h), interpolation=get_interpolation())
+        for _ in range(self.factor):
+            scale_img = cv2.pyrDown(scale_img)
+        scale_img = cv2.resize(
+            scale_img, (src_w, src_h), interpolation=get_interpolation()
+        )
+        return scale_img
+
+
+class CVGaussianNoise(object):
+    def __init__(self, mean=0, var=20):
+        self.mean = mean
+        if isinstance(var, numbers.Number):
+            self.var = max(int(sample_asym(var)), 1)
+        elif isinstance(var, (tuple, list)) and len(var) == 2:
+            self.var = int(sample_uniform(var[0], var[1]))
+        else:
+            raise Exception("degree must be number or list with length 2")
+
+    def __call__(self, img):
+        noise = np.random.normal(self.mean, self.var**0.5, img.shape)
+        img = np.clip(img + noise, 0, 255).astype(np.uint8)
+        return img
+
+
+class CVPossionNoise(object):
+    def __init__(self, lam=20):
+        self.lam = lam
+        if isinstance(lam, numbers.Number):
+            self.lam = max(int(sample_asym(lam)), 1)
+        elif isinstance(lam, (tuple, list)) and len(lam) == 2:
+            self.lam = int(sample_uniform(lam[0], lam[1]))
+        else:
+            raise Exception("lam must be number or list with length 2")
+
+    def __call__(self, img):
+        noise = np.random.poisson(lam=self.lam, size=img.shape)
+        img = np.clip(img + noise, 0, 255).astype(np.uint8)
+        return img
+
+
+class CVGaussionBlur(object):
+    def __init__(self, radius):
+        self.radius = radius
+        if isinstance(radius, numbers.Number):
+            self.radius = max(int(sample_asym(radius)), 1)
+        elif isinstance(radius, (tuple, list)) and len(radius) == 2:
+            self.radius = int(sample_uniform(radius[0], radius[1]))
+        else:
+            raise Exception("radius must be number or list with length 2")
+
+    def __call__(self, img):
+        fil = cv2.getGaussianKernel(ksize=self.radius, sigma=1, ktype=cv2.CV_32F)
+        img = cv2.sepFilter2D(img, -1, fil, fil)
+        return img
+
+
+class CVMotionBlur(object):
+    def __init__(self, degrees=12, angle=90):
+        if isinstance(degrees, numbers.Number):
+            self.degree = max(int(sample_asym(degrees)), 1)
+        elif isinstance(degrees, (tuple, list)) and len(degrees) == 2:
+            self.degree = int(sample_uniform(degrees[0], degrees[1]))
+        else:
+            raise Exception("degree must be number or list with length 2")
+        self.angle = sample_uniform(-angle, angle)
+
+    def __call__(self, img):
+        M = cv2.getRotationMatrix2D((self.degree // 2, self.degree // 2), self.angle, 1)
+        motion_blur_kernel = np.zeros((self.degree, self.degree))
+        motion_blur_kernel[self.degree // 2, :] = 1
+        motion_blur_kernel = cv2.warpAffine(
+            motion_blur_kernel, M, (self.degree, self.degree)
+        )
+        motion_blur_kernel = motion_blur_kernel / self.degree
+        img = cv2.filter2D(img, -1, motion_blur_kernel)
+        img = np.clip(img, 0, 255).astype(np.uint8)
+        return img
+
+
+class CVGeometry(object):
+    def __init__(
+        self,
+        degrees=15,
+        translate=(0.3, 0.3),
+        scale=(0.5, 2.0),
+        shear=(45, 15),
+        distortion=0.5,
+        p=0.5,
+    ):
+        self.p = p
+        type_p = random.random()
+        if type_p < 0.33:
+            self.transforms = CVRandomRotation(degrees=degrees)
+        elif type_p < 0.66:
+            self.transforms = CVRandomAffine(
+                degrees=degrees, translate=translate, scale=scale, shear=shear
+            )
+        else:
+            self.transforms = CVRandomPerspective(distortion=distortion)
+
+    def __call__(self, img):
+        if random.random() < self.p:
+            return self.transforms(img)
+        else:
+            return img
+
+
+class CVDeterioration(object):
+    def __init__(self, var, degrees, factor, p=0.5):
+        self.p = p
+        transforms = []
+        if var is not None:
+            transforms.append(CVGaussianNoise(var=var))
+        if degrees is not None:
+            transforms.append(CVMotionBlur(degrees=degrees))
+        if factor is not None:
+            transforms.append(CVRescale(factor=factor))
+
+        random.shuffle(transforms)
+        transforms = Compose(transforms)
+        self.transforms = transforms
+
+    def __call__(self, img):
+        if random.random() < self.p:
+            return self.transforms(img)
+        else:
+            return img
+
+
+class CVColorJitter(object):
+    def __init__(self, brightness=0.5, contrast=0.5, saturation=0.5, hue=0.1, p=0.5):
+        self.p = p
+        self.transforms = ColorJitter(
+            brightness=brightness, contrast=contrast, saturation=saturation, hue=hue
+        )
+
+    def __call__(self, img):
+        if random.random() < self.p:
+            return self.transforms(img)
+        else:
+            return img
+
+
+class SVTRDeterioration(object):
+    def __init__(self, var, degrees, factor, p=0.5):
+        self.p = p
+        transforms = []
+        if var is not None:
+            transforms.append(CVGaussianNoise(var=var))
+        if degrees is not None:
+            transforms.append(CVMotionBlur(degrees=degrees))
+        if factor is not None:
+            transforms.append(CVRescale(factor=factor))
+        self.transforms = transforms
+
+    def __call__(self, img):
+        if random.random() < self.p:
+            random.shuffle(self.transforms)
+            transforms = Compose(self.transforms)
+            return transforms(img)
+        else:
+            return img
+
+
+class ParseQDeterioration(object):
+    def __init__(self, var, degrees, lam, radius, factor, p=0.5):
+        self.p = p
+        transforms = []
+        if var is not None:
+            transforms.append(CVGaussianNoise(var=var))
+        if degrees is not None:
+            transforms.append(CVMotionBlur(degrees=degrees))
+        if lam is not None:
+            transforms.append(CVPossionNoise(lam=lam))
+        if radius is not None:
+            transforms.append(CVGaussionBlur(radius=radius))
+        if factor is not None:
+            transforms.append(CVRescale(factor=factor))
+        self.transforms = transforms
+
+    def __call__(self, img):
+        if random.random() < self.p:
+            random.shuffle(self.transforms)
+            transforms = Compose(self.transforms)
+            return transforms(img)
+        else:
+            return img
+
+
+class SVTRGeometry(object):
+    def __init__(
+        self,
+        aug_type=0,
+        degrees=15,
+        translate=(0.3, 0.3),
+        scale=(0.5, 2.0),
+        shear=(45, 15),
+        distortion=0.5,
+        p=0.5,
+    ):
+        self.aug_type = aug_type
+        self.p = p
+        self.transforms = []
+        self.transforms.append(CVRandomRotation(degrees=degrees))
+        self.transforms.append(
+            CVRandomAffine(
+                degrees=degrees, translate=translate, scale=scale, shear=shear
+            )
+        )
+        self.transforms.append(CVRandomPerspective(distortion=distortion))
+
+    def __call__(self, img):
+        if random.random() < self.p:
+            if self.aug_type:
+                random.shuffle(self.transforms)
+                transforms = Compose(self.transforms[: random.randint(1, 3)])
+                img = transforms(img)
+            else:
+                img = self.transforms[random.randint(0, 2)](img)
+            return img
+        else:
+            return img
diff --git a/ppocr/data/imaug/copy_paste.py b/ppocr/data/imaug/copy_paste.py
new file mode 100644
index 0000000..4d627d0
--- /dev/null
+++ b/ppocr/data/imaug/copy_paste.py
@@ -0,0 +1,179 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import copy
+import cv2
+import random
+import numpy as np
+from PIL import Image
+from shapely.geometry import Polygon
+
+from ppocr.data.imaug.iaa_augment import IaaAugment
+from ppocr.data.imaug.random_crop_data import is_poly_outside_rect
+from tools.infer.utility import get_rotate_crop_image
+
+
+class CopyPaste(object):
+    def __init__(self, objects_paste_ratio=0.2, limit_paste=True, **kwargs):
+        self.ext_data_num = 1
+        self.objects_paste_ratio = objects_paste_ratio
+        self.limit_paste = limit_paste
+        augmenter_args = [{"type": "Resize", "args": {"size": [0.5, 3]}}]
+        self.aug = IaaAugment(augmenter_args)
+
+    def __call__(self, data):
+        point_num = data["polys"].shape[1]
+        src_img = data["image"]
+        src_polys = data["polys"].tolist()
+        src_texts = data["texts"]
+        src_ignores = data["ignore_tags"].tolist()
+        ext_data = data["ext_data"][0]
+        ext_image = ext_data["image"]
+        ext_polys = ext_data["polys"]
+        ext_texts = ext_data["texts"]
+        ext_ignores = ext_data["ignore_tags"]
+
+        indexes = [i for i in range(len(ext_ignores)) if not ext_ignores[i]]
+        select_num = max(1, min(int(self.objects_paste_ratio * len(ext_polys)), 30))
+
+        random.shuffle(indexes)
+        select_idxs = indexes[:select_num]
+        select_polys = ext_polys[select_idxs]
+        select_ignores = ext_ignores[select_idxs]
+
+        src_img = cv2.cvtColor(src_img, cv2.COLOR_BGR2RGB)
+        ext_image = cv2.cvtColor(ext_image, cv2.COLOR_BGR2RGB)
+        src_img = Image.fromarray(src_img).convert("RGBA")
+        for idx, poly, tag in zip(select_idxs, select_polys, select_ignores):
+            box_img = get_rotate_crop_image(ext_image, poly)
+
+            src_img, box = self.paste_img(src_img, box_img, src_polys)
+            if box is not None:
+                box = box.tolist()
+                for _ in range(len(box), point_num):
+                    box.append(box[-1])
+                src_polys.append(box)
+                src_texts.append(ext_texts[idx])
+                src_ignores.append(tag)
+        src_img = cv2.cvtColor(np.array(src_img), cv2.COLOR_RGB2BGR)
+        h, w = src_img.shape[:2]
+        src_polys = np.array(src_polys)
+        src_polys[:, :, 0] = np.clip(src_polys[:, :, 0], 0, w)
+        src_polys[:, :, 1] = np.clip(src_polys[:, :, 1], 0, h)
+        data["image"] = src_img
+        data["polys"] = src_polys
+        data["texts"] = src_texts
+        data["ignore_tags"] = np.array(src_ignores)
+        return data
+
+    def paste_img(self, src_img, box_img, src_polys):
+        box_img_pil = Image.fromarray(box_img).convert("RGBA")
+        src_w, src_h = src_img.size
+        box_w, box_h = box_img_pil.size
+
+        angle = np.random.randint(0, 360)
+        box = np.array([[[0, 0], [box_w, 0], [box_w, box_h], [0, box_h]]])
+        box = rotate_bbox(box_img, box, angle)[0]
+        box_img_pil = box_img_pil.rotate(angle, expand=1)
+        box_w, box_h = box_img_pil.width, box_img_pil.height
+        if src_w - box_w < 0 or src_h - box_h < 0:
+            return src_img, None
+
+        paste_x, paste_y = self.select_coord(
+            src_polys, box, src_w - box_w, src_h - box_h
+        )
+        if paste_x is None:
+            return src_img, None
+        box[:, 0] += paste_x
+        box[:, 1] += paste_y
+        r, g, b, A = box_img_pil.split()
+        src_img.paste(box_img_pil, (paste_x, paste_y), mask=A)
+
+        return src_img, box
+
+    def select_coord(self, src_polys, box, endx, endy):
+        if self.limit_paste:
+            xmin, ymin, xmax, ymax = (
+                box[:, 0].min(),
+                box[:, 1].min(),
+                box[:, 0].max(),
+                box[:, 1].max(),
+            )
+            for _ in range(50):
+                paste_x = random.randint(0, endx)
+                paste_y = random.randint(0, endy)
+                xmin1 = xmin + paste_x
+                xmax1 = xmax + paste_x
+                ymin1 = ymin + paste_y
+                ymax1 = ymax + paste_y
+
+                num_poly_in_rect = 0
+                for poly in src_polys:
+                    if not is_poly_outside_rect(
+                        poly, xmin1, ymin1, xmax1 - xmin1, ymax1 - ymin1
+                    ):
+                        num_poly_in_rect += 1
+                        break
+                if num_poly_in_rect == 0:
+                    return paste_x, paste_y
+            return None, None
+        else:
+            paste_x = random.randint(0, endx)
+            paste_y = random.randint(0, endy)
+            return paste_x, paste_y
+
+
+def get_union(pD, pG):
+    return Polygon(pD).union(Polygon(pG)).area
+
+
+def get_intersection_over_union(pD, pG):
+    return get_intersection(pD, pG) / get_union(pD, pG)
+
+
+def get_intersection(pD, pG):
+    return Polygon(pD).intersection(Polygon(pG)).area
+
+
+def rotate_bbox(img, text_polys, angle, scale=1):
+    """
+    from https://github.com/WenmuZhou/DBNet.pytorch/blob/master/data_loader/modules/augment.py
+    Args:
+        img: np.ndarray
+        text_polys: np.ndarray N*4*2
+        angle: int
+        scale: int
+
+    Returns:
+
+    """
+    w = img.shape[1]
+    h = img.shape[0]
+
+    rangle = np.deg2rad(angle)
+    nw = abs(np.sin(rangle) * h) + abs(np.cos(rangle) * w)
+    nh = abs(np.cos(rangle) * h) + abs(np.sin(rangle) * w)
+    rot_mat = cv2.getRotationMatrix2D((nw * 0.5, nh * 0.5), angle, scale)
+    rot_move = np.dot(rot_mat, np.array([(nw - w) * 0.5, (nh - h) * 0.5, 0]))
+    rot_mat[0, 2] += rot_move[0]
+    rot_mat[1, 2] += rot_move[1]
+
+    # ---------------------- rotate box ----------------------
+    rot_text_polys = list()
+    for bbox in text_polys:
+        point1 = np.dot(rot_mat, np.array([bbox[0, 0], bbox[0, 1], 1]))
+        point2 = np.dot(rot_mat, np.array([bbox[1, 0], bbox[1, 1], 1]))
+        point3 = np.dot(rot_mat, np.array([bbox[2, 0], bbox[2, 1], 1]))
+        point4 = np.dot(rot_mat, np.array([bbox[3, 0], bbox[3, 1], 1]))
+        rot_text_polys.append([point1, point2, point3, point4])
+    return np.array(rot_text_polys, dtype=np.float32)
diff --git a/ppocr/data/imaug/ct_process.py b/ppocr/data/imaug/ct_process.py
new file mode 100644
index 0000000..26e111d
--- /dev/null
+++ b/ppocr/data/imaug/ct_process.py
@@ -0,0 +1,376 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import cv2
+import paddle
+import random
+import pyclipper
+import numpy as np
+from PIL import Image
+
+import paddle.vision.transforms as transforms
+
+from ppocr.utils.utility import check_install
+
+
+class RandomScale:
+    def __init__(self, short_size=640, **kwargs):
+        self.short_size = short_size
+
+    def scale_aligned(self, img, scale):
+        oh, ow = img.shape[0:2]
+        h = int(oh * scale + 0.5)
+        w = int(ow * scale + 0.5)
+        if h % 32 != 0:
+            h = h + (32 - h % 32)
+        if w % 32 != 0:
+            w = w + (32 - w % 32)
+        img = cv2.resize(img, dsize=(w, h))
+        factor_h = h / oh
+        factor_w = w / ow
+        return img, factor_h, factor_w
+
+    def __call__(self, data):
+        img = data["image"]
+
+        h, w = img.shape[0:2]
+        random_scale = np.array([0.7, 0.8, 0.9, 1.0, 1.1, 1.2, 1.3])
+        scale = (np.random.choice(random_scale) * self.short_size) / min(h, w)
+        img, factor_h, factor_w = self.scale_aligned(img, scale)
+
+        data["scale_factor"] = (factor_w, factor_h)
+        data["image"] = img
+        return data
+
+
+class MakeShrink:
+    def __init__(self, kernel_scale=0.7, **kwargs):
+        self.kernel_scale = kernel_scale
+
+    def dist(self, a, b):
+        return np.linalg.norm((a - b), ord=2, axis=0)
+
+    def perimeter(self, bbox):
+        peri = 0.0
+        for i in range(bbox.shape[0]):
+            peri += self.dist(bbox[i], bbox[(i + 1) % bbox.shape[0]])
+        return peri
+
+    def shrink(self, bboxes, rate, max_shr=20):
+        check_install("Polygon", "Polygon3")
+        import Polygon as plg
+
+        rate = rate * rate
+        shrinked_bboxes = []
+        for bbox in bboxes:
+            area = plg.Polygon(bbox).area()
+            peri = self.perimeter(bbox)
+
+            try:
+                pco = pyclipper.PyclipperOffset()
+                pco.AddPath(bbox, pyclipper.JT_ROUND, pyclipper.ET_CLOSEDPOLYGON)
+                offset = min(int(area * (1 - rate) / (peri + 0.001) + 0.5), max_shr)
+
+                shrinked_bbox = pco.Execute(-offset)
+                if len(shrinked_bbox) == 0:
+                    shrinked_bboxes.append(bbox)
+                    continue
+
+                shrinked_bbox = np.array(shrinked_bbox[0])
+                if shrinked_bbox.shape[0] <= 2:
+                    shrinked_bboxes.append(bbox)
+                    continue
+
+                shrinked_bboxes.append(shrinked_bbox)
+            except Exception as e:
+                shrinked_bboxes.append(bbox)
+
+        return shrinked_bboxes
+
+    def __call__(self, data):
+        img = data["image"]
+        bboxes = data["polys"]
+        words = data["texts"]
+        scale_factor = data["scale_factor"]
+
+        gt_instance = np.zeros(img.shape[0:2], dtype="uint8")  # h,w
+        training_mask = np.ones(img.shape[0:2], dtype="uint8")
+        training_mask_distance = np.ones(img.shape[0:2], dtype="uint8")
+
+        for i in range(len(bboxes)):
+            bboxes[i] = np.reshape(
+                bboxes[i]
+                * ([scale_factor[0], scale_factor[1]] * (bboxes[i].shape[0] // 2)),
+                (bboxes[i].shape[0] // 2, 2),
+            ).astype("int32")
+
+        for i in range(len(bboxes)):
+            # different value for different bbox
+            cv2.drawContours(gt_instance, [bboxes[i]], -1, i + 1, -1)
+
+            # set training mask to 0
+            cv2.drawContours(training_mask, [bboxes[i]], -1, 0, -1)
+
+            # for not accurate annotation, use training_mask_distance
+            if words[i] == "###" or words[i] == "???":
+                cv2.drawContours(training_mask_distance, [bboxes[i]], -1, 0, -1)
+
+        # make shrink
+        gt_kernel_instance = np.zeros(img.shape[0:2], dtype="uint8")
+        kernel_bboxes = self.shrink(bboxes, self.kernel_scale)
+        for i in range(len(bboxes)):
+            cv2.drawContours(gt_kernel_instance, [kernel_bboxes[i]], -1, i + 1, -1)
+
+            # for training mask, kernel and background= 1, box region=0
+            if words[i] != "###" and words[i] != "???":
+                cv2.drawContours(training_mask, [kernel_bboxes[i]], -1, 1, -1)
+
+        gt_kernel = gt_kernel_instance.copy()
+        # for gt_kernel, kernel = 1
+        gt_kernel[gt_kernel > 0] = 1
+
+        # shrink 2 times
+        tmp1 = gt_kernel_instance.copy()
+        erode_kernel = np.ones((3, 3), np.uint8)
+        tmp1 = cv2.erode(tmp1, erode_kernel, iterations=1)
+        tmp2 = tmp1.copy()
+        tmp2 = cv2.erode(tmp2, erode_kernel, iterations=1)
+
+        # compute text region
+        gt_kernel_inner = tmp1 - tmp2
+
+        # gt_instance: text instance, bg=0, diff word use diff value
+        # training_mask: text instance mask, word=0，kernel and bg=1
+        # gt_kernel_instance: text kernel instance, bg=0, diff word use diff value
+        # gt_kernel: text_kernel, bg=0，diff word use same value
+        # gt_kernel_inner: text kernel reference
+        # training_mask_distance: word without anno = 0, else 1
+
+        data["image"] = [
+            img,
+            gt_instance,
+            training_mask,
+            gt_kernel_instance,
+            gt_kernel,
+            gt_kernel_inner,
+            training_mask_distance,
+        ]
+        return data
+
+
+class GroupRandomHorizontalFlip:
+    def __init__(self, p=0.5, **kwargs):
+        self.p = p
+
+    def __call__(self, data):
+        imgs = data["image"]
+
+        if random.random() < self.p:
+            for i in range(len(imgs)):
+                imgs[i] = np.flip(imgs[i], axis=1).copy()
+        data["image"] = imgs
+        return data
+
+
+class GroupRandomRotate:
+    def __init__(self, **kwargs):
+        pass
+
+    def __call__(self, data):
+        imgs = data["image"]
+
+        max_angle = 10
+        angle = random.random() * 2 * max_angle - max_angle
+        for i in range(len(imgs)):
+            img = imgs[i]
+            w, h = img.shape[:2]
+            rotation_matrix = cv2.getRotationMatrix2D((h / 2, w / 2), angle, 1)
+            img_rotation = cv2.warpAffine(
+                img, rotation_matrix, (h, w), flags=cv2.INTER_NEAREST
+            )
+            imgs[i] = img_rotation
+
+        data["image"] = imgs
+        return data
+
+
+class GroupRandomCropPadding:
+    def __init__(self, target_size=(640, 640), **kwargs):
+        self.target_size = target_size
+
+    def __call__(self, data):
+        imgs = data["image"]
+
+        h, w = imgs[0].shape[0:2]
+        t_w, t_h = self.target_size
+        p_w, p_h = self.target_size
+        if w == t_w and h == t_h:
+            return data
+
+        t_h = t_h if t_h < h else h
+        t_w = t_w if t_w < w else w
+
+        if random.random() > 3.0 / 8.0 and np.max(imgs[1]) > 0:
+            # make sure to crop the text region
+            tl = np.min(np.where(imgs[1] > 0), axis=1) - (t_h, t_w)
+            tl[tl < 0] = 0
+            br = np.max(np.where(imgs[1] > 0), axis=1) - (t_h, t_w)
+            br[br < 0] = 0
+            br[0] = min(br[0], h - t_h)
+            br[1] = min(br[1], w - t_w)
+
+            i = random.randint(tl[0], br[0]) if tl[0] < br[0] else 0
+            j = random.randint(tl[1], br[1]) if tl[1] < br[1] else 0
+        else:
+            i = random.randint(0, h - t_h) if h - t_h > 0 else 0
+            j = random.randint(0, w - t_w) if w - t_w > 0 else 0
+
+        n_imgs = []
+        for idx in range(len(imgs)):
+            if len(imgs[idx].shape) == 3:
+                s3_length = int(imgs[idx].shape[-1])
+                img = imgs[idx][i : i + t_h, j : j + t_w, :]
+                img_p = cv2.copyMakeBorder(
+                    img,
+                    0,
+                    p_h - t_h,
+                    0,
+                    p_w - t_w,
+                    borderType=cv2.BORDER_CONSTANT,
+                    value=tuple(0 for i in range(s3_length)),
+                )
+            else:
+                img = imgs[idx][i : i + t_h, j : j + t_w]
+                img_p = cv2.copyMakeBorder(
+                    img,
+                    0,
+                    p_h - t_h,
+                    0,
+                    p_w - t_w,
+                    borderType=cv2.BORDER_CONSTANT,
+                    value=(0,),
+                )
+            n_imgs.append(img_p)
+
+        data["image"] = n_imgs
+        return data
+
+
+class MakeCentripetalShift:
+    def __init__(self, **kwargs):
+        pass
+
+    def jaccard(self, As, Bs):
+        A = As.shape[0]  # small
+        B = Bs.shape[0]  # large
+
+        dis = np.sqrt(
+            np.sum(
+                (
+                    As[:, np.newaxis, :].repeat(B, axis=1)
+                    - Bs[np.newaxis, :, :].repeat(A, axis=0)
+                )
+                ** 2,
+                axis=-1,
+            )
+        )
+
+        ind = np.argmin(dis, axis=-1)
+
+        return ind
+
+    def __call__(self, data):
+        imgs = data["image"]
+
+        (
+            img,
+            gt_instance,
+            training_mask,
+            gt_kernel_instance,
+            gt_kernel,
+            gt_kernel_inner,
+            training_mask_distance,
+        ) = (imgs[0], imgs[1], imgs[2], imgs[3], imgs[4], imgs[5], imgs[6])
+
+        max_instance = np.max(gt_instance)  # num bbox
+
+        # make centripetal shift
+        gt_distance = np.zeros((2, *img.shape[0:2]), dtype=np.float32)
+        for i in range(1, max_instance + 1):
+            # kernel_reference
+            ind = gt_kernel_inner == i
+
+            if np.sum(ind) == 0:
+                training_mask[gt_instance == i] = 0
+                training_mask_distance[gt_instance == i] = 0
+                continue
+
+            kpoints = (
+                np.array(np.where(ind)).transpose((1, 0))[:, ::-1].astype("float32")
+            )
+
+            ind = (gt_instance == i) * (gt_kernel_instance == 0)
+            if np.sum(ind) == 0:
+                continue
+            pixels = np.where(ind)
+
+            points = np.array(pixels).transpose((1, 0))[:, ::-1].astype("float32")
+
+            bbox_ind = self.jaccard(points, kpoints)
+
+            offset_gt = kpoints[bbox_ind] - points
+
+            gt_distance[:, pixels[0], pixels[1]] = offset_gt.T * 0.1
+
+        img = Image.fromarray(img)
+        img = img.convert("RGB")
+
+        data["image"] = img
+        data["gt_kernel"] = gt_kernel.astype("int64")
+        data["training_mask"] = training_mask.astype("int64")
+        data["gt_instance"] = gt_instance.astype("int64")
+        data["gt_kernel_instance"] = gt_kernel_instance.astype("int64")
+        data["training_mask_distance"] = training_mask_distance.astype("int64")
+        data["gt_distance"] = gt_distance.astype("float32")
+
+        return data
+
+
+class ScaleAlignedShort:
+    def __init__(self, short_size=640, **kwargs):
+        self.short_size = short_size
+
+    def __call__(self, data):
+        img = data["image"]
+
+        org_img_shape = img.shape
+
+        h, w = img.shape[0:2]
+        scale = self.short_size * 1.0 / min(h, w)
+        h = int(h * scale + 0.5)
+        w = int(w * scale + 0.5)
+        if h % 32 != 0:
+            h = h + (32 - h % 32)
+        if w % 32 != 0:
+            w = w + (32 - w % 32)
+        img = cv2.resize(img, dsize=(w, h))
+
+        new_img_shape = img.shape
+        img_shape = np.array(org_img_shape + new_img_shape)
+
+        data["shape"] = img_shape
+        data["image"] = img
+
+        return data
diff --git a/ppocr/data/imaug/drrg_targets.py b/ppocr/data/imaug/drrg_targets.py
new file mode 100644
index 0000000..8eeef26
--- /dev/null
+++ b/ppocr/data/imaug/drrg_targets.py
@@ -0,0 +1,770 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/datasets/pipelines/textdet_targets/drrg_targets.py
+"""
+
+import cv2
+import numpy as np
+from ppocr.utils.utility import check_install
+from numpy.linalg import norm
+
+
+class DRRGTargets(object):
+    def __init__(
+        self,
+        orientation_thr=2.0,
+        resample_step=8.0,
+        num_min_comps=9,
+        num_max_comps=600,
+        min_width=8.0,
+        max_width=24.0,
+        center_region_shrink_ratio=0.3,
+        comp_shrink_ratio=1.0,
+        comp_w_h_ratio=0.3,
+        text_comp_nms_thr=0.25,
+        min_rand_half_height=8.0,
+        max_rand_half_height=24.0,
+        jitter_level=0.2,
+        **kwargs,
+    ):
+        super().__init__()
+        self.orientation_thr = orientation_thr
+        self.resample_step = resample_step
+        self.num_max_comps = num_max_comps
+        self.num_min_comps = num_min_comps
+        self.min_width = min_width
+        self.max_width = max_width
+        self.center_region_shrink_ratio = center_region_shrink_ratio
+        self.comp_shrink_ratio = comp_shrink_ratio
+        self.comp_w_h_ratio = comp_w_h_ratio
+        self.text_comp_nms_thr = text_comp_nms_thr
+        self.min_rand_half_height = min_rand_half_height
+        self.max_rand_half_height = max_rand_half_height
+        self.jitter_level = jitter_level
+        self.eps = 1e-8
+
+    def vector_angle(self, vec1, vec2):
+        if vec1.ndim > 1:
+            unit_vec1 = vec1 / (norm(vec1, axis=-1) + self.eps).reshape((-1, 1))
+        else:
+            unit_vec1 = vec1 / (norm(vec1, axis=-1) + self.eps)
+        if vec2.ndim > 1:
+            unit_vec2 = vec2 / (norm(vec2, axis=-1) + self.eps).reshape((-1, 1))
+        else:
+            unit_vec2 = vec2 / (norm(vec2, axis=-1) + self.eps)
+        return np.arccos(np.clip(np.sum(unit_vec1 * unit_vec2, axis=-1), -1.0, 1.0))
+
+    def vector_slope(self, vec):
+        assert len(vec) == 2
+        return abs(vec[1] / (vec[0] + self.eps))
+
+    def vector_sin(self, vec):
+        assert len(vec) == 2
+        return vec[1] / (norm(vec) + self.eps)
+
+    def vector_cos(self, vec):
+        assert len(vec) == 2
+        return vec[0] / (norm(vec) + self.eps)
+
+    def find_head_tail(self, points, orientation_thr):
+        assert points.ndim == 2
+        assert points.shape[0] >= 4
+        assert points.shape[1] == 2
+        assert isinstance(orientation_thr, float)
+
+        if len(points) > 4:
+            pad_points = np.vstack([points, points[0]])
+            edge_vec = pad_points[1:] - pad_points[:-1]
+
+            theta_sum = []
+            adjacent_vec_theta = []
+            for i, edge_vec1 in enumerate(edge_vec):
+                adjacent_ind = [x % len(edge_vec) for x in [i - 1, i + 1]]
+                adjacent_edge_vec = edge_vec[adjacent_ind]
+                temp_theta_sum = np.sum(self.vector_angle(edge_vec1, adjacent_edge_vec))
+                temp_adjacent_theta = self.vector_angle(
+                    adjacent_edge_vec[0], adjacent_edge_vec[1]
+                )
+                theta_sum.append(temp_theta_sum)
+                adjacent_vec_theta.append(temp_adjacent_theta)
+            theta_sum_score = np.array(theta_sum) / np.pi
+            adjacent_theta_score = np.array(adjacent_vec_theta) / np.pi
+            poly_center = np.mean(points, axis=0)
+            edge_dist = np.maximum(
+                norm(pad_points[1:] - poly_center, axis=-1),
+                norm(pad_points[:-1] - poly_center, axis=-1),
+            )
+            dist_score = edge_dist / (np.max(edge_dist) + self.eps)
+            position_score = np.zeros(len(edge_vec))
+            score = 0.5 * theta_sum_score + 0.15 * adjacent_theta_score
+            score += 0.35 * dist_score
+            if len(points) % 2 == 0:
+                position_score[(len(score) // 2 - 1)] += 1
+                position_score[-1] += 1
+            score += 0.1 * position_score
+            pad_score = np.concatenate([score, score])
+            score_matrix = np.zeros((len(score), len(score) - 3))
+            x = np.arange(len(score) - 3) / float(len(score) - 4)
+            gaussian = (
+                1.0
+                / (np.sqrt(2.0 * np.pi) * 0.5)
+                * np.exp(-np.power((x - 0.5) / 0.5, 2.0) / 2)
+            )
+            gaussian = gaussian / np.max(gaussian)
+            for i in range(len(score)):
+                score_matrix[i, :] = (
+                    score[i]
+                    + pad_score[(i + 2) : (i + len(score) - 1)] * gaussian * 0.3
+                )
+
+            head_start, tail_increment = np.unravel_index(
+                score_matrix.argmax(), score_matrix.shape
+            )
+            tail_start = (head_start + tail_increment + 2) % len(points)
+            head_end = (head_start + 1) % len(points)
+            tail_end = (tail_start + 1) % len(points)
+
+            if head_end > tail_end:
+                head_start, tail_start = tail_start, head_start
+                head_end, tail_end = tail_end, head_end
+            head_inds = [head_start, head_end]
+            tail_inds = [tail_start, tail_end]
+        else:
+            if self.vector_slope(points[1] - points[0]) + self.vector_slope(
+                points[3] - points[2]
+            ) < self.vector_slope(points[2] - points[1]) + self.vector_slope(
+                points[0] - points[3]
+            ):
+                horizontal_edge_inds = [[0, 1], [2, 3]]
+                vertical_edge_inds = [[3, 0], [1, 2]]
+            else:
+                horizontal_edge_inds = [[3, 0], [1, 2]]
+                vertical_edge_inds = [[0, 1], [2, 3]]
+
+            vertical_len_sum = norm(
+                points[vertical_edge_inds[0][0]] - points[vertical_edge_inds[0][1]]
+            ) + norm(
+                points[vertical_edge_inds[1][0]] - points[vertical_edge_inds[1][1]]
+            )
+            horizontal_len_sum = norm(
+                points[horizontal_edge_inds[0][0]] - points[horizontal_edge_inds[0][1]]
+            ) + norm(
+                points[horizontal_edge_inds[1][0]] - points[horizontal_edge_inds[1][1]]
+            )
+
+            if vertical_len_sum > horizontal_len_sum * orientation_thr:
+                head_inds = horizontal_edge_inds[0]
+                tail_inds = horizontal_edge_inds[1]
+            else:
+                head_inds = vertical_edge_inds[0]
+                tail_inds = vertical_edge_inds[1]
+
+        return head_inds, tail_inds
+
+    def reorder_poly_edge(self, points):
+        assert points.ndim == 2
+        assert points.shape[0] >= 4
+        assert points.shape[1] == 2
+
+        head_inds, tail_inds = self.find_head_tail(points, self.orientation_thr)
+        head_edge, tail_edge = points[head_inds], points[tail_inds]
+
+        pad_points = np.vstack([points, points])
+        if tail_inds[1] < 1:
+            tail_inds[1] = len(points)
+        sideline1 = pad_points[head_inds[1] : tail_inds[1]]
+        sideline2 = pad_points[tail_inds[1] : (head_inds[1] + len(points))]
+        sideline_mean_shift = np.mean(sideline1, axis=0) - np.mean(sideline2, axis=0)
+
+        if sideline_mean_shift[1] > 0:
+            top_sideline, bot_sideline = sideline2, sideline1
+        else:
+            top_sideline, bot_sideline = sideline1, sideline2
+
+        return head_edge, tail_edge, top_sideline, bot_sideline
+
+    def cal_curve_length(self, line):
+        assert line.ndim == 2
+        assert len(line) >= 2
+
+        edges_length = np.sqrt(
+            (line[1:, 0] - line[:-1, 0]) ** 2 + (line[1:, 1] - line[:-1, 1]) ** 2
+        )
+        total_length = np.sum(edges_length)
+        return edges_length, total_length
+
+    def resample_line(self, line, n):
+        assert line.ndim == 2
+        assert line.shape[0] >= 2
+        assert line.shape[1] == 2
+        assert isinstance(n, int)
+        assert n > 2
+
+        edges_length, total_length = self.cal_curve_length(line)
+        t_org = np.insert(np.cumsum(edges_length), 0, 0)
+        unit_t = total_length / (n - 1)
+        t_equidistant = np.arange(1, n - 1, dtype=np.float32) * unit_t
+        edge_ind = 0
+        points = [line[0]]
+        for t in t_equidistant:
+            while edge_ind < len(edges_length) - 1 and t > t_org[edge_ind + 1]:
+                edge_ind += 1
+            t_l, t_r = t_org[edge_ind], t_org[edge_ind + 1]
+            weight = np.array([t_r - t, t - t_l], dtype=np.float32) / (
+                t_r - t_l + self.eps
+            )
+            p_coords = np.dot(weight, line[[edge_ind, edge_ind + 1]])
+            points.append(p_coords)
+        points.append(line[-1])
+        resampled_line = np.vstack(points)
+
+        return resampled_line
+
+    def resample_sidelines(self, sideline1, sideline2, resample_step):
+        assert sideline1.ndim == sideline2.ndim == 2
+        assert sideline1.shape[1] == sideline2.shape[1] == 2
+        assert sideline1.shape[0] >= 2
+        assert sideline2.shape[0] >= 2
+        assert isinstance(resample_step, float)
+
+        _, length1 = self.cal_curve_length(sideline1)
+        _, length2 = self.cal_curve_length(sideline2)
+
+        avg_length = (length1 + length2) / 2
+        resample_point_num = max(int(float(avg_length) / resample_step) + 1, 3)
+
+        resampled_line1 = self.resample_line(sideline1, resample_point_num)
+        resampled_line2 = self.resample_line(sideline2, resample_point_num)
+
+        return resampled_line1, resampled_line2
+
+    def dist_point2line(self, point, line):
+        assert isinstance(line, tuple)
+        point1, point2 = line
+        d = abs(np.cross(point2 - point1, point - point1)) / (
+            norm(point2 - point1) + 1e-8
+        )
+        return d
+
+    def draw_center_region_maps(
+        self,
+        top_line,
+        bot_line,
+        center_line,
+        center_region_mask,
+        top_height_map,
+        bot_height_map,
+        sin_map,
+        cos_map,
+        region_shrink_ratio,
+    ):
+        assert top_line.shape == bot_line.shape == center_line.shape
+        assert (
+            center_region_mask.shape
+            == top_height_map.shape
+            == bot_height_map.shape
+            == sin_map.shape
+            == cos_map.shape
+        )
+        assert isinstance(region_shrink_ratio, float)
+
+        h, w = center_region_mask.shape
+        for i in range(0, len(center_line) - 1):
+            top_mid_point = (top_line[i] + top_line[i + 1]) / 2
+            bot_mid_point = (bot_line[i] + bot_line[i + 1]) / 2
+
+            sin_theta = self.vector_sin(top_mid_point - bot_mid_point)
+            cos_theta = self.vector_cos(top_mid_point - bot_mid_point)
+
+            tl = center_line[i] + (top_line[i] - center_line[i]) * region_shrink_ratio
+            tr = (
+                center_line[i + 1]
+                + (top_line[i + 1] - center_line[i + 1]) * region_shrink_ratio
+            )
+            br = (
+                center_line[i + 1]
+                + (bot_line[i + 1] - center_line[i + 1]) * region_shrink_ratio
+            )
+            bl = center_line[i] + (bot_line[i] - center_line[i]) * region_shrink_ratio
+            current_center_box = np.vstack([tl, tr, br, bl]).astype(np.int32)
+
+            cv2.fillPoly(center_region_mask, [current_center_box], color=1)
+            cv2.fillPoly(sin_map, [current_center_box], color=sin_theta)
+            cv2.fillPoly(cos_map, [current_center_box], color=cos_theta)
+
+            current_center_box[:, 0] = np.clip(current_center_box[:, 0], 0, w - 1)
+            current_center_box[:, 1] = np.clip(current_center_box[:, 1], 0, h - 1)
+            min_coord = np.min(current_center_box, axis=0).astype(np.int32)
+            max_coord = np.max(current_center_box, axis=0).astype(np.int32)
+            current_center_box = current_center_box - min_coord
+            box_sz = max_coord - min_coord + 1
+
+            center_box_mask = np.zeros((box_sz[1], box_sz[0]), dtype=np.uint8)
+            cv2.fillPoly(center_box_mask, [current_center_box], color=1)
+
+            inds = np.argwhere(center_box_mask > 0)
+            inds = inds + (min_coord[1], min_coord[0])
+            inds_xy = np.fliplr(inds)
+            top_height_map[(inds[:, 0], inds[:, 1])] = self.dist_point2line(
+                inds_xy, (top_line[i], top_line[i + 1])
+            )
+            bot_height_map[(inds[:, 0], inds[:, 1])] = self.dist_point2line(
+                inds_xy, (bot_line[i], bot_line[i + 1])
+            )
+
+    def generate_center_mask_attrib_maps(self, img_size, text_polys):
+        assert isinstance(img_size, tuple)
+
+        h, w = img_size
+
+        center_lines = []
+        center_region_mask = np.zeros((h, w), np.uint8)
+        top_height_map = np.zeros((h, w), dtype=np.float32)
+        bot_height_map = np.zeros((h, w), dtype=np.float32)
+        sin_map = np.zeros((h, w), dtype=np.float32)
+        cos_map = np.zeros((h, w), dtype=np.float32)
+
+        for poly in text_polys:
+            polygon_points = poly
+            _, _, top_line, bot_line = self.reorder_poly_edge(polygon_points)
+            resampled_top_line, resampled_bot_line = self.resample_sidelines(
+                top_line, bot_line, self.resample_step
+            )
+            resampled_bot_line = resampled_bot_line[::-1]
+            center_line = (resampled_top_line + resampled_bot_line) / 2
+
+            if self.vector_slope(center_line[-1] - center_line[0]) > 2:
+                if (center_line[-1] - center_line[0])[1] < 0:
+                    center_line = center_line[::-1]
+                    resampled_top_line = resampled_top_line[::-1]
+                    resampled_bot_line = resampled_bot_line[::-1]
+            else:
+                if (center_line[-1] - center_line[0])[0] < 0:
+                    center_line = center_line[::-1]
+                    resampled_top_line = resampled_top_line[::-1]
+                    resampled_bot_line = resampled_bot_line[::-1]
+
+            line_head_shrink_len = (
+                np.clip(
+                    (norm(top_line[0] - bot_line[0]) * self.comp_w_h_ratio),
+                    self.min_width,
+                    self.max_width,
+                )
+                / 2
+            )
+            line_tail_shrink_len = (
+                np.clip(
+                    (norm(top_line[-1] - bot_line[-1]) * self.comp_w_h_ratio),
+                    self.min_width,
+                    self.max_width,
+                )
+                / 2
+            )
+            num_head_shrink = int(line_head_shrink_len // self.resample_step)
+            num_tail_shrink = int(line_tail_shrink_len // self.resample_step)
+            if len(center_line) > num_head_shrink + num_tail_shrink + 2:
+                center_line = center_line[
+                    num_head_shrink : len(center_line) - num_tail_shrink
+                ]
+                resampled_top_line = resampled_top_line[
+                    num_head_shrink : len(resampled_top_line) - num_tail_shrink
+                ]
+                resampled_bot_line = resampled_bot_line[
+                    num_head_shrink : len(resampled_bot_line) - num_tail_shrink
+                ]
+            center_lines.append(center_line.astype(np.int32))
+
+            self.draw_center_region_maps(
+                resampled_top_line,
+                resampled_bot_line,
+                center_line,
+                center_region_mask,
+                top_height_map,
+                bot_height_map,
+                sin_map,
+                cos_map,
+                self.center_region_shrink_ratio,
+            )
+
+        return (
+            center_lines,
+            center_region_mask,
+            top_height_map,
+            bot_height_map,
+            sin_map,
+            cos_map,
+        )
+
+    def generate_rand_comp_attribs(self, num_rand_comps, center_sample_mask):
+        assert isinstance(num_rand_comps, int)
+        assert num_rand_comps > 0
+        assert center_sample_mask.ndim == 2
+
+        h, w = center_sample_mask.shape
+
+        max_rand_half_height = self.max_rand_half_height
+        min_rand_half_height = self.min_rand_half_height
+        max_rand_height = max_rand_half_height * 2
+        max_rand_width = np.clip(
+            max_rand_height * self.comp_w_h_ratio, self.min_width, self.max_width
+        )
+        margin = (
+            int(np.sqrt((max_rand_height / 2) ** 2 + (max_rand_width / 2) ** 2)) + 1
+        )
+
+        if 2 * margin + 1 > min(h, w):
+            assert min(h, w) > (np.sqrt(2) * (self.min_width + 1))
+            max_rand_half_height = max(min(h, w) / 4, self.min_width / 2 + 1)
+            min_rand_half_height = max(max_rand_half_height / 4, self.min_width / 2)
+
+            max_rand_height = max_rand_half_height * 2
+            max_rand_width = np.clip(
+                max_rand_height * self.comp_w_h_ratio, self.min_width, self.max_width
+            )
+            margin = (
+                int(np.sqrt((max_rand_height / 2) ** 2 + (max_rand_width / 2) ** 2)) + 1
+            )
+
+        inner_center_sample_mask = np.zeros_like(center_sample_mask)
+        inner_center_sample_mask[margin : h - margin, margin : w - margin] = (
+            center_sample_mask[margin : h - margin, margin : w - margin]
+        )
+        kernel_size = int(np.clip(max_rand_half_height, 7, 21))
+        inner_center_sample_mask = cv2.erode(
+            inner_center_sample_mask, np.ones((kernel_size, kernel_size), np.uint8)
+        )
+
+        center_candidates = np.argwhere(inner_center_sample_mask > 0)
+        num_center_candidates = len(center_candidates)
+        sample_inds = np.random.choice(num_center_candidates, num_rand_comps)
+        rand_centers = center_candidates[sample_inds]
+
+        rand_top_height = np.random.randint(
+            min_rand_half_height, max_rand_half_height, size=(len(rand_centers), 1)
+        )
+        rand_bot_height = np.random.randint(
+            min_rand_half_height, max_rand_half_height, size=(len(rand_centers), 1)
+        )
+
+        rand_cos = 2 * np.random.random(size=(len(rand_centers), 1)) - 1
+        rand_sin = 2 * np.random.random(size=(len(rand_centers), 1)) - 1
+        scale = np.sqrt(1.0 / (rand_cos**2 + rand_sin**2 + 1e-8))
+        rand_cos = rand_cos * scale
+        rand_sin = rand_sin * scale
+
+        height = rand_top_height + rand_bot_height
+        width = np.clip(height * self.comp_w_h_ratio, self.min_width, self.max_width)
+
+        rand_comp_attribs = np.hstack(
+            [
+                rand_centers[:, ::-1],
+                height,
+                width,
+                rand_cos,
+                rand_sin,
+                np.zeros_like(rand_sin),
+            ]
+        ).astype(np.float32)
+
+        return rand_comp_attribs
+
+    def jitter_comp_attribs(self, comp_attribs, jitter_level):
+        """Jitter text components attributes.
+
+        Args:
+            comp_attribs (ndarray): The text component attributes.
+            jitter_level (float): The jitter level of text components
+                attributes.
+
+        Returns:
+            jittered_comp_attribs (ndarray): The jittered text component
+                attributes (x, y, h, w, cos, sin, comp_label).
+        """
+
+        assert comp_attribs.shape[1] == 7
+        assert comp_attribs.shape[0] > 0
+        assert isinstance(jitter_level, float)
+
+        x = comp_attribs[:, 0].reshape((-1, 1))
+        y = comp_attribs[:, 1].reshape((-1, 1))
+        h = comp_attribs[:, 2].reshape((-1, 1))
+        w = comp_attribs[:, 3].reshape((-1, 1))
+        cos = comp_attribs[:, 4].reshape((-1, 1))
+        sin = comp_attribs[:, 5].reshape((-1, 1))
+        comp_labels = comp_attribs[:, 6].reshape((-1, 1))
+
+        x += (
+            (np.random.random(size=(len(comp_attribs), 1)) - 0.5)
+            * (h * np.abs(cos) + w * np.abs(sin))
+            * jitter_level
+        )
+        y += (
+            (np.random.random(size=(len(comp_attribs), 1)) - 0.5)
+            * (h * np.abs(sin) + w * np.abs(cos))
+            * jitter_level
+        )
+
+        h += (np.random.random(size=(len(comp_attribs), 1)) - 0.5) * h * jitter_level
+        w += (np.random.random(size=(len(comp_attribs), 1)) - 0.5) * w * jitter_level
+
+        cos += (np.random.random(size=(len(comp_attribs), 1)) - 0.5) * 2 * jitter_level
+        sin += (np.random.random(size=(len(comp_attribs), 1)) - 0.5) * 2 * jitter_level
+
+        scale = np.sqrt(1.0 / (cos**2 + sin**2 + 1e-8))
+        cos = cos * scale
+        sin = sin * scale
+
+        jittered_comp_attribs = np.hstack([x, y, h, w, cos, sin, comp_labels])
+
+        return jittered_comp_attribs
+
+    def generate_comp_attribs(
+        self,
+        center_lines,
+        text_mask,
+        center_region_mask,
+        top_height_map,
+        bot_height_map,
+        sin_map,
+        cos_map,
+    ):
+        """Generate text component attributes.
+
+        Args:
+            center_lines (list[ndarray]): The list of text center lines .
+            text_mask (ndarray): The text region mask.
+            center_region_mask (ndarray): The text center region mask.
+            top_height_map (ndarray): The map on which the distance from points
+                to top side lines will be drawn for each pixel in text center
+                regions.
+            bot_height_map (ndarray): The map on which the distance from points
+                to bottom side lines will be drawn for each pixel in text
+                center regions.
+            sin_map (ndarray): The sin(theta) map where theta is the angle
+                between vector (top point - bottom point) and vector (1, 0).
+            cos_map (ndarray): The cos(theta) map where theta is the angle
+                between vector (top point - bottom point) and vector (1, 0).
+
+        Returns:
+            pad_comp_attribs (ndarray): The padded text component attributes
+                of a fixed size.
+        """
+
+        assert isinstance(center_lines, list)
+        assert (
+            text_mask.shape
+            == center_region_mask.shape
+            == top_height_map.shape
+            == bot_height_map.shape
+            == sin_map.shape
+            == cos_map.shape
+        )
+
+        center_lines_mask = np.zeros_like(center_region_mask)
+        cv2.polylines(center_lines_mask, center_lines, 0, 1, 1)
+        center_lines_mask = center_lines_mask * center_region_mask
+        comp_centers = np.argwhere(center_lines_mask > 0)
+
+        y = comp_centers[:, 0]
+        x = comp_centers[:, 1]
+
+        top_height = top_height_map[y, x].reshape((-1, 1)) * self.comp_shrink_ratio
+        bot_height = bot_height_map[y, x].reshape((-1, 1)) * self.comp_shrink_ratio
+        sin = sin_map[y, x].reshape((-1, 1))
+        cos = cos_map[y, x].reshape((-1, 1))
+
+        top_mid_points = comp_centers + np.hstack([top_height * sin, top_height * cos])
+        bot_mid_points = comp_centers - np.hstack([bot_height * sin, bot_height * cos])
+
+        width = (top_height + bot_height) * self.comp_w_h_ratio
+        width = np.clip(width, self.min_width, self.max_width)
+        r = width / 2
+
+        tl = top_mid_points[:, ::-1] - np.hstack([-r * sin, r * cos])
+        tr = top_mid_points[:, ::-1] + np.hstack([-r * sin, r * cos])
+        br = bot_mid_points[:, ::-1] + np.hstack([-r * sin, r * cos])
+        bl = bot_mid_points[:, ::-1] - np.hstack([-r * sin, r * cos])
+        text_comps = np.hstack([tl, tr, br, bl]).astype(np.float32)
+
+        score = np.ones((text_comps.shape[0], 1), dtype=np.float32)
+        text_comps = np.hstack([text_comps, score])
+        check_install("lanms", "lanms-neo")
+        from lanms import merge_quadrangle_n9 as la_nms
+
+        text_comps = la_nms(text_comps, self.text_comp_nms_thr)
+
+        if text_comps.shape[0] >= 1:
+            img_h, img_w = center_region_mask.shape
+            text_comps[:, 0:8:2] = np.clip(text_comps[:, 0:8:2], 0, img_w - 1)
+            text_comps[:, 1:8:2] = np.clip(text_comps[:, 1:8:2], 0, img_h - 1)
+
+            comp_centers = np.mean(
+                text_comps[:, 0:8].reshape((-1, 4, 2)), axis=1
+            ).astype(np.int32)
+            x = comp_centers[:, 0]
+            y = comp_centers[:, 1]
+
+            height = (top_height_map[y, x] + bot_height_map[y, x]).reshape((-1, 1))
+            width = np.clip(
+                height * self.comp_w_h_ratio, self.min_width, self.max_width
+            )
+
+            cos = cos_map[y, x].reshape((-1, 1))
+            sin = sin_map[y, x].reshape((-1, 1))
+
+            _, comp_label_mask = cv2.connectedComponents(
+                center_region_mask, connectivity=8
+            )
+            comp_labels = comp_label_mask[y, x].reshape((-1, 1)).astype(np.float32)
+
+            x = x.reshape((-1, 1)).astype(np.float32)
+            y = y.reshape((-1, 1)).astype(np.float32)
+            comp_attribs = np.hstack([x, y, height, width, cos, sin, comp_labels])
+            comp_attribs = self.jitter_comp_attribs(comp_attribs, self.jitter_level)
+
+            if comp_attribs.shape[0] < self.num_min_comps:
+                num_rand_comps = self.num_min_comps - comp_attribs.shape[0]
+                rand_comp_attribs = self.generate_rand_comp_attribs(
+                    num_rand_comps, 1 - text_mask
+                )
+                comp_attribs = np.vstack([comp_attribs, rand_comp_attribs])
+        else:
+            comp_attribs = self.generate_rand_comp_attribs(
+                self.num_min_comps, 1 - text_mask
+            )
+
+        num_comps = (
+            np.ones((comp_attribs.shape[0], 1), dtype=np.float32)
+            * comp_attribs.shape[0]
+        )
+        comp_attribs = np.hstack([num_comps, comp_attribs])
+
+        if comp_attribs.shape[0] > self.num_max_comps:
+            comp_attribs = comp_attribs[: self.num_max_comps, :]
+            comp_attribs[:, 0] = self.num_max_comps
+
+        pad_comp_attribs = np.zeros(
+            (self.num_max_comps, comp_attribs.shape[1]), dtype=np.float32
+        )
+        pad_comp_attribs[: comp_attribs.shape[0], :] = comp_attribs
+
+        return pad_comp_attribs
+
+    def generate_text_region_mask(self, img_size, text_polys):
+        """Generate text center region mask and geometry attribute maps.
+
+        Args:
+            img_size (tuple): The image size (height, width).
+            text_polys (list[list[ndarray]]): The list of text polygons.
+
+        Returns:
+            text_region_mask (ndarray): The text region mask.
+        """
+
+        assert isinstance(img_size, tuple)
+
+        h, w = img_size
+        text_region_mask = np.zeros((h, w), dtype=np.uint8)
+
+        for poly in text_polys:
+            polygon = np.array(poly, dtype=np.int32).reshape((1, -1, 2))
+            cv2.fillPoly(text_region_mask, polygon, 1)
+
+        return text_region_mask
+
+    def generate_effective_mask(self, mask_size: tuple, polygons_ignore):
+        """Generate effective mask by setting the ineffective regions to 0 and
+        effective regions to 1.
+
+        Args:
+            mask_size (tuple): The mask size.
+            polygons_ignore (list[[ndarray]]: The list of ignored text
+                polygons.
+
+        Returns:
+            mask (ndarray): The effective mask of (height, width).
+        """
+        mask = np.ones(mask_size, dtype=np.uint8)
+
+        for poly in polygons_ignore:
+            instance = poly.astype(np.int32).reshape(1, -1, 2)
+            cv2.fillPoly(mask, instance, 0)
+
+        return mask
+
+    def generate_targets(self, data):
+        """Generate the gt targets for DRRG.
+
+        Args:
+            data (dict): The input result dictionary.
+
+        Returns:
+            data (dict): The output result dictionary.
+        """
+
+        assert isinstance(data, dict)
+
+        image = data["image"]
+        polygons = data["polys"]
+        ignore_tags = data["ignore_tags"]
+        h, w, _ = image.shape
+
+        polygon_masks = []
+        polygon_masks_ignore = []
+        for tag, polygon in zip(ignore_tags, polygons):
+            if tag is True:
+                polygon_masks_ignore.append(polygon)
+            else:
+                polygon_masks.append(polygon)
+
+        gt_text_mask = self.generate_text_region_mask((h, w), polygon_masks)
+        gt_mask = self.generate_effective_mask((h, w), polygon_masks_ignore)
+        (
+            center_lines,
+            gt_center_region_mask,
+            gt_top_height_map,
+            gt_bot_height_map,
+            gt_sin_map,
+            gt_cos_map,
+        ) = self.generate_center_mask_attrib_maps((h, w), polygon_masks)
+
+        gt_comp_attribs = self.generate_comp_attribs(
+            center_lines,
+            gt_text_mask,
+            gt_center_region_mask,
+            gt_top_height_map,
+            gt_bot_height_map,
+            gt_sin_map,
+            gt_cos_map,
+        )
+
+        mapping = {
+            "gt_text_mask": gt_text_mask,
+            "gt_center_region_mask": gt_center_region_mask,
+            "gt_mask": gt_mask,
+            "gt_top_height_map": gt_top_height_map,
+            "gt_bot_height_map": gt_bot_height_map,
+            "gt_sin_map": gt_sin_map,
+            "gt_cos_map": gt_cos_map,
+        }
+
+        data.update(mapping)
+        data["gt_comp_attribs"] = gt_comp_attribs
+        return data
+
+    def __call__(self, data):
+        data = self.generate_targets(data)
+        return data
diff --git a/ppocr/data/imaug/east_process.py b/ppocr/data/imaug/east_process.py
new file mode 100644
index 0000000..7e5bfa0
--- /dev/null
+++ b/ppocr/data/imaug/east_process.py
@@ -0,0 +1,446 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is referred from:
+https://github.com/songdejia/EAST/blob/master/data_utils.py
+"""
+import math
+import cv2
+import numpy as np
+import json
+import sys
+import os
+
+__all__ = ["EASTProcessTrain"]
+
+
+class EASTProcessTrain(object):
+    def __init__(
+        self,
+        image_shape=[512, 512],
+        background_ratio=0.125,
+        min_crop_side_ratio=0.1,
+        min_text_size=10,
+        **kwargs,
+    ):
+        self.input_size = image_shape[1]
+        self.random_scale = np.array([0.5, 1, 2.0, 3.0])
+        self.background_ratio = background_ratio
+        self.min_crop_side_ratio = min_crop_side_ratio
+        self.min_text_size = min_text_size
+
+    def preprocess(self, im):
+        input_size = self.input_size
+        im_shape = im.shape
+        im_size_min = np.min(im_shape[0:2])
+        im_size_max = np.max(im_shape[0:2])
+        im_scale = float(input_size) / float(im_size_max)
+        im = cv2.resize(im, None, None, fx=im_scale, fy=im_scale)
+        img_mean = [0.485, 0.456, 0.406]
+        img_std = [0.229, 0.224, 0.225]
+        # im = im[:, :, ::-1].astype(np.float32)
+        im = im / 255
+        im -= img_mean
+        im /= img_std
+        new_h, new_w, _ = im.shape
+        im_padded = np.zeros((input_size, input_size, 3), dtype=np.float32)
+        im_padded[:new_h, :new_w, :] = im
+        im_padded = im_padded.transpose((2, 0, 1))
+        im_padded = im_padded[np.newaxis, :]
+        return im_padded, im_scale
+
+    def rotate_im_poly(self, im, text_polys):
+        """
+        rotate image with 90 / 180 / 270 degre
+        """
+        im_w, im_h = im.shape[1], im.shape[0]
+        dst_im = im.copy()
+        dst_polys = []
+        rand_degree_ratio = np.random.rand()
+        rand_degree_cnt = 1
+        if 0.333 < rand_degree_ratio < 0.666:
+            rand_degree_cnt = 2
+        elif rand_degree_ratio > 0.666:
+            rand_degree_cnt = 3
+        for i in range(rand_degree_cnt):
+            dst_im = np.rot90(dst_im)
+        rot_degree = -90 * rand_degree_cnt
+        rot_angle = rot_degree * math.pi / 180.0
+        n_poly = text_polys.shape[0]
+        cx, cy = 0.5 * im_w, 0.5 * im_h
+        ncx, ncy = 0.5 * dst_im.shape[1], 0.5 * dst_im.shape[0]
+        for i in range(n_poly):
+            wordBB = text_polys[i]
+            poly = []
+            for j in range(4):
+                sx, sy = wordBB[j][0], wordBB[j][1]
+                dx = (
+                    math.cos(rot_angle) * (sx - cx)
+                    - math.sin(rot_angle) * (sy - cy)
+                    + ncx
+                )
+                dy = (
+                    math.sin(rot_angle) * (sx - cx)
+                    + math.cos(rot_angle) * (sy - cy)
+                    + ncy
+                )
+                poly.append([dx, dy])
+            dst_polys.append(poly)
+        dst_polys = np.array(dst_polys, dtype=np.float32)
+        return dst_im, dst_polys
+
+    def polygon_area(self, poly):
+        """
+        compute area of a polygon
+        :param poly:
+        :return:
+        """
+        edge = [
+            (poly[1][0] - poly[0][0]) * (poly[1][1] + poly[0][1]),
+            (poly[2][0] - poly[1][0]) * (poly[2][1] + poly[1][1]),
+            (poly[3][0] - poly[2][0]) * (poly[3][1] + poly[2][1]),
+            (poly[0][0] - poly[3][0]) * (poly[0][1] + poly[3][1]),
+        ]
+        return np.sum(edge) / 2.0
+
+    def check_and_validate_polys(self, polys, tags, img_height, img_width):
+        """
+        check so that the text poly is in the same direction,
+        and also filter some invalid polygons
+        :param polys:
+        :param tags:
+        :return:
+        """
+        h, w = img_height, img_width
+        if polys.shape[0] == 0:
+            return polys
+        polys[:, :, 0] = np.clip(polys[:, :, 0], 0, w - 1)
+        polys[:, :, 1] = np.clip(polys[:, :, 1], 0, h - 1)
+
+        validated_polys = []
+        validated_tags = []
+        for poly, tag in zip(polys, tags):
+            p_area = self.polygon_area(poly)
+            # invalid poly
+            if abs(p_area) < 1:
+                continue
+            if p_area > 0:
+                #'poly in wrong direction'
+                if not tag:
+                    tag = True  # reversed cases should be ignore
+                poly = poly[(0, 3, 2, 1), :]
+            validated_polys.append(poly)
+            validated_tags.append(tag)
+        return np.array(validated_polys), np.array(validated_tags)
+
+    def draw_img_polys(self, img, polys):
+        if len(img.shape) == 4:
+            img = np.squeeze(img, axis=0)
+        if img.shape[0] == 3:
+            img = img.transpose((1, 2, 0))
+            img[:, :, 2] += 123.68
+            img[:, :, 1] += 116.78
+            img[:, :, 0] += 103.94
+        cv2.imwrite("tmp.jpg", img)
+        img = cv2.imread("tmp.jpg")
+        for box in polys:
+            box = box.astype(np.int32).reshape((-1, 1, 2))
+            cv2.polylines(img, [box], True, color=(255, 255, 0), thickness=2)
+        import random
+
+        ino = random.randint(0, 100)
+        cv2.imwrite("tmp_%d.jpg" % ino, img)
+        return
+
+    def shrink_poly(self, poly, r):
+        """
+        fit a poly inside the origin poly, maybe bugs here...
+        used for generate the score map
+        :param poly: the text poly
+        :param r: r in the paper
+        :return: the shrunk poly
+        """
+        # shrink ratio
+        R = 0.3
+        # find the longer pair
+        dist0 = np.linalg.norm(poly[0] - poly[1])
+        dist1 = np.linalg.norm(poly[2] - poly[3])
+        dist2 = np.linalg.norm(poly[0] - poly[3])
+        dist3 = np.linalg.norm(poly[1] - poly[2])
+        if dist0 + dist1 > dist2 + dist3:
+            # first move (p0, p1), (p2, p3), then (p0, p3), (p1, p2)
+            ## p0, p1
+            theta = np.arctan2((poly[1][1] - poly[0][1]), (poly[1][0] - poly[0][0]))
+            poly[0][0] += R * r[0] * np.cos(theta)
+            poly[0][1] += R * r[0] * np.sin(theta)
+            poly[1][0] -= R * r[1] * np.cos(theta)
+            poly[1][1] -= R * r[1] * np.sin(theta)
+            ## p2, p3
+            theta = np.arctan2((poly[2][1] - poly[3][1]), (poly[2][0] - poly[3][0]))
+            poly[3][0] += R * r[3] * np.cos(theta)
+            poly[3][1] += R * r[3] * np.sin(theta)
+            poly[2][0] -= R * r[2] * np.cos(theta)
+            poly[2][1] -= R * r[2] * np.sin(theta)
+            ## p0, p3
+            theta = np.arctan2((poly[3][0] - poly[0][0]), (poly[3][1] - poly[0][1]))
+            poly[0][0] += R * r[0] * np.sin(theta)
+            poly[0][1] += R * r[0] * np.cos(theta)
+            poly[3][0] -= R * r[3] * np.sin(theta)
+            poly[3][1] -= R * r[3] * np.cos(theta)
+            ## p1, p2
+            theta = np.arctan2((poly[2][0] - poly[1][0]), (poly[2][1] - poly[1][1]))
+            poly[1][0] += R * r[1] * np.sin(theta)
+            poly[1][1] += R * r[1] * np.cos(theta)
+            poly[2][0] -= R * r[2] * np.sin(theta)
+            poly[2][1] -= R * r[2] * np.cos(theta)
+        else:
+            ## p0, p3
+            # print poly
+            theta = np.arctan2((poly[3][0] - poly[0][0]), (poly[3][1] - poly[0][1]))
+            poly[0][0] += R * r[0] * np.sin(theta)
+            poly[0][1] += R * r[0] * np.cos(theta)
+            poly[3][0] -= R * r[3] * np.sin(theta)
+            poly[3][1] -= R * r[3] * np.cos(theta)
+            ## p1, p2
+            theta = np.arctan2((poly[2][0] - poly[1][0]), (poly[2][1] - poly[1][1]))
+            poly[1][0] += R * r[1] * np.sin(theta)
+            poly[1][1] += R * r[1] * np.cos(theta)
+            poly[2][0] -= R * r[2] * np.sin(theta)
+            poly[2][1] -= R * r[2] * np.cos(theta)
+            ## p0, p1
+            theta = np.arctan2((poly[1][1] - poly[0][1]), (poly[1][0] - poly[0][0]))
+            poly[0][0] += R * r[0] * np.cos(theta)
+            poly[0][1] += R * r[0] * np.sin(theta)
+            poly[1][0] -= R * r[1] * np.cos(theta)
+            poly[1][1] -= R * r[1] * np.sin(theta)
+            ## p2, p3
+            theta = np.arctan2((poly[2][1] - poly[3][1]), (poly[2][0] - poly[3][0]))
+            poly[3][0] += R * r[3] * np.cos(theta)
+            poly[3][1] += R * r[3] * np.sin(theta)
+            poly[2][0] -= R * r[2] * np.cos(theta)
+            poly[2][1] -= R * r[2] * np.sin(theta)
+        return poly
+
+    def generate_quad(self, im_size, polys, tags):
+        """
+        Generate quadrangle.
+        """
+        h, w = im_size
+        poly_mask = np.zeros((h, w), dtype=np.uint8)
+        score_map = np.zeros((h, w), dtype=np.uint8)
+        # (x1, y1, ..., x4, y4, short_edge_norm)
+        geo_map = np.zeros((h, w, 9), dtype=np.float32)
+        # mask used during training, to ignore some hard areas
+        training_mask = np.ones((h, w), dtype=np.uint8)
+        for poly_idx, poly_tag in enumerate(zip(polys, tags)):
+            poly = poly_tag[0]
+            tag = poly_tag[1]
+
+            r = [None, None, None, None]
+            for i in range(4):
+                dist1 = np.linalg.norm(poly[i] - poly[(i + 1) % 4])
+                dist2 = np.linalg.norm(poly[i] - poly[(i - 1) % 4])
+                r[i] = min(dist1, dist2)
+            # score map
+            shrinked_poly = self.shrink_poly(poly.copy(), r).astype(np.int32)[
+                np.newaxis, :, :
+            ]
+            cv2.fillPoly(score_map, shrinked_poly, 1)
+            cv2.fillPoly(poly_mask, shrinked_poly, poly_idx + 1)
+            # if the poly is too small, then ignore it during training
+            poly_h = min(
+                np.linalg.norm(poly[0] - poly[3]), np.linalg.norm(poly[1] - poly[2])
+            )
+            poly_w = min(
+                np.linalg.norm(poly[0] - poly[1]), np.linalg.norm(poly[2] - poly[3])
+            )
+            if min(poly_h, poly_w) < self.min_text_size:
+                cv2.fillPoly(training_mask, poly.astype(np.int32)[np.newaxis, :, :], 0)
+
+            if tag:
+                cv2.fillPoly(training_mask, poly.astype(np.int32)[np.newaxis, :, :], 0)
+
+            xy_in_poly = np.argwhere(poly_mask == (poly_idx + 1))
+            # geo map.
+            y_in_poly = xy_in_poly[:, 0]
+            x_in_poly = xy_in_poly[:, 1]
+            poly[:, 0] = np.minimum(np.maximum(poly[:, 0], 0), w)
+            poly[:, 1] = np.minimum(np.maximum(poly[:, 1], 0), h)
+            for pno in range(4):
+                geo_channel_beg = pno * 2
+                geo_map[y_in_poly, x_in_poly, geo_channel_beg] = (
+                    x_in_poly - poly[pno, 0]
+                )
+                geo_map[y_in_poly, x_in_poly, geo_channel_beg + 1] = (
+                    y_in_poly - poly[pno, 1]
+                )
+            geo_map[y_in_poly, x_in_poly, 8] = 1.0 / max(min(poly_h, poly_w), 1.0)
+        return score_map, geo_map, training_mask
+
+    def crop_area(self, im, polys, tags, crop_background=False, max_tries=50):
+        """
+        make random crop from the input image
+        :param im:
+        :param polys:
+        :param tags:
+        :param crop_background:
+        :param max_tries:
+        :return:
+        """
+        h, w, _ = im.shape
+        pad_h = h // 10
+        pad_w = w // 10
+        h_array = np.zeros((h + pad_h * 2), dtype=np.int32)
+        w_array = np.zeros((w + pad_w * 2), dtype=np.int32)
+        for poly in polys:
+            poly = np.round(poly, decimals=0).astype(np.int32)
+            minx = np.min(poly[:, 0])
+            maxx = np.max(poly[:, 0])
+            w_array[minx + pad_w : maxx + pad_w] = 1
+            miny = np.min(poly[:, 1])
+            maxy = np.max(poly[:, 1])
+            h_array[miny + pad_h : maxy + pad_h] = 1
+        # ensure the cropped area not across a text
+        h_axis = np.where(h_array == 0)[0]
+        w_axis = np.where(w_array == 0)[0]
+        if len(h_axis) == 0 or len(w_axis) == 0:
+            return im, polys, tags
+
+        for i in range(max_tries):
+            xx = np.random.choice(w_axis, size=2)
+            xmin = np.min(xx) - pad_w
+            xmax = np.max(xx) - pad_w
+            xmin = np.clip(xmin, 0, w - 1)
+            xmax = np.clip(xmax, 0, w - 1)
+            yy = np.random.choice(h_axis, size=2)
+            ymin = np.min(yy) - pad_h
+            ymax = np.max(yy) - pad_h
+            ymin = np.clip(ymin, 0, h - 1)
+            ymax = np.clip(ymax, 0, h - 1)
+            if (
+                xmax - xmin < self.min_crop_side_ratio * w
+                or ymax - ymin < self.min_crop_side_ratio * h
+            ):
+                # area too small
+                continue
+            if polys.shape[0] != 0:
+                poly_axis_in_area = (
+                    (polys[:, :, 0] >= xmin)
+                    & (polys[:, :, 0] <= xmax)
+                    & (polys[:, :, 1] >= ymin)
+                    & (polys[:, :, 1] <= ymax)
+                )
+                selected_polys = np.where(np.sum(poly_axis_in_area, axis=1) == 4)[0]
+            else:
+                selected_polys = []
+
+            if len(selected_polys) == 0:
+                # no text in this area
+                if crop_background:
+                    im = im[ymin : ymax + 1, xmin : xmax + 1, :]
+                    polys = []
+                    tags = []
+                    return im, polys, tags
+                else:
+                    continue
+
+            im = im[ymin : ymax + 1, xmin : xmax + 1, :]
+            polys = polys[selected_polys]
+            tags = tags[selected_polys]
+            polys[:, :, 0] -= xmin
+            polys[:, :, 1] -= ymin
+            return im, polys, tags
+        return im, polys, tags
+
+    def crop_background_infor(self, im, text_polys, text_tags):
+        im, text_polys, text_tags = self.crop_area(
+            im, text_polys, text_tags, crop_background=True
+        )
+
+        if len(text_polys) > 0:
+            return None
+        # pad and resize image
+        input_size = self.input_size
+        im, ratio = self.preprocess(im)
+        score_map = np.zeros((input_size, input_size), dtype=np.float32)
+        geo_map = np.zeros((input_size, input_size, 9), dtype=np.float32)
+        training_mask = np.ones((input_size, input_size), dtype=np.float32)
+        return im, score_map, geo_map, training_mask
+
+    def crop_foreground_infor(self, im, text_polys, text_tags):
+        im, text_polys, text_tags = self.crop_area(
+            im, text_polys, text_tags, crop_background=False
+        )
+
+        if text_polys.shape[0] == 0:
+            return None
+        # continue for all ignore case
+        if np.sum((text_tags * 1.0)) >= text_tags.size:
+            return None
+        # pad and resize image
+        input_size = self.input_size
+        im, ratio = self.preprocess(im)
+        text_polys[:, :, 0] *= ratio
+        text_polys[:, :, 1] *= ratio
+        _, _, new_h, new_w = im.shape
+        #         print(im.shape)
+        #         self.draw_img_polys(im, text_polys)
+        score_map, geo_map, training_mask = self.generate_quad(
+            (new_h, new_w), text_polys, text_tags
+        )
+        return im, score_map, geo_map, training_mask
+
+    def __call__(self, data):
+        im = data["image"]
+        text_polys = data["polys"]
+        text_tags = data["ignore_tags"]
+        if im is None:
+            return None
+        if text_polys.shape[0] == 0:
+            return None
+
+        # add rotate cases
+        if np.random.rand() < 0.5:
+            im, text_polys = self.rotate_im_poly(im, text_polys)
+        h, w, _ = im.shape
+        text_polys, text_tags = self.check_and_validate_polys(
+            text_polys, text_tags, h, w
+        )
+        if text_polys.shape[0] == 0:
+            return None
+
+        # random scale this image
+        rd_scale = np.random.choice(self.random_scale)
+        im = cv2.resize(im, dsize=None, fx=rd_scale, fy=rd_scale)
+        text_polys *= rd_scale
+        if np.random.rand() < self.background_ratio:
+            outs = self.crop_background_infor(im, text_polys, text_tags)
+        else:
+            outs = self.crop_foreground_infor(im, text_polys, text_tags)
+
+        if outs is None:
+            return None
+        im, score_map, geo_map, training_mask = outs
+        score_map = score_map[np.newaxis, ::4, ::4].astype(np.float32)
+        geo_map = np.swapaxes(geo_map, 1, 2)
+        geo_map = np.swapaxes(geo_map, 1, 0)
+        geo_map = geo_map[:, ::4, ::4].astype(np.float32)
+        training_mask = training_mask[np.newaxis, ::4, ::4]
+        training_mask = training_mask.astype(np.float32)
+
+        data["image"] = im[0]
+        data["score_map"] = score_map
+        data["geo_map"] = geo_map
+        data["training_mask"] = training_mask
+        return data
diff --git a/ppocr/data/imaug/fce_aug.py b/ppocr/data/imaug/fce_aug.py
new file mode 100644
index 0000000..2697344
--- /dev/null
+++ b/ppocr/data/imaug/fce_aug.py
@@ -0,0 +1,575 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/datasets/pipelines/transforms.py
+"""
+import numpy as np
+from PIL import Image, ImageDraw
+import cv2
+from shapely.geometry import Polygon
+import math
+from ppocr.utils.poly_nms import poly_intersection
+
+
+class RandomScaling:
+    def __init__(self, size=800, scale=(3.0 / 4, 5.0 / 2), **kwargs):
+        """Random scale the image while keeping aspect.
+
+        Args:
+            size (int) : Base size before scaling.
+            scale (tuple(float)) : The range of scaling.
+        """
+        assert isinstance(size, int)
+        assert isinstance(scale, float) or isinstance(scale, tuple)
+        self.size = size
+        self.scale = scale if isinstance(scale, tuple) else (1 - scale, 1 + scale)
+
+    def __call__(self, data):
+        image = data["image"]
+        text_polys = data["polys"]
+        h, w, _ = image.shape
+
+        aspect_ratio = np.random.uniform(min(self.scale), max(self.scale))
+        scales = self.size * 1.0 / max(h, w) * aspect_ratio
+        scales = np.array([scales, scales])
+        out_size = (int(h * scales[1]), int(w * scales[0]))
+        image = cv2.resize(image, out_size[::-1])
+
+        data["image"] = image
+        text_polys[:, :, 0::2] = text_polys[:, :, 0::2] * scales[1]
+        text_polys[:, :, 1::2] = text_polys[:, :, 1::2] * scales[0]
+        data["polys"] = text_polys
+
+        return data
+
+
+class RandomCropFlip:
+    def __init__(
+        self, pad_ratio=0.1, crop_ratio=0.5, iter_num=1, min_area_ratio=0.2, **kwargs
+    ):
+        """Random crop and flip a patch of the image.
+
+        Args:
+            crop_ratio (float): The ratio of cropping.
+            iter_num (int): Number of operations.
+            min_area_ratio (float): Minimal area ratio between cropped patch
+                and original image.
+        """
+        assert isinstance(crop_ratio, float)
+        assert isinstance(iter_num, int)
+        assert isinstance(min_area_ratio, float)
+
+        self.pad_ratio = pad_ratio
+        self.epsilon = 1e-2
+        self.crop_ratio = crop_ratio
+        self.iter_num = iter_num
+        self.min_area_ratio = min_area_ratio
+
+    def __call__(self, results):
+        for i in range(self.iter_num):
+            results = self.random_crop_flip(results)
+
+        return results
+
+    def random_crop_flip(self, results):
+        image = results["image"]
+        polygons = results["polys"]
+        ignore_tags = results["ignore_tags"]
+        if len(polygons) == 0:
+            return results
+
+        if np.random.random() >= self.crop_ratio:
+            return results
+
+        h, w, _ = image.shape
+        area = h * w
+        pad_h = int(h * self.pad_ratio)
+        pad_w = int(w * self.pad_ratio)
+        h_axis, w_axis = self.generate_crop_target(image, polygons, pad_h, pad_w)
+        if len(h_axis) == 0 or len(w_axis) == 0:
+            return results
+
+        attempt = 0
+        while attempt < 50:
+            attempt += 1
+            polys_keep = []
+            polys_new = []
+            ignore_tags_keep = []
+            ignore_tags_new = []
+            xx = np.random.choice(w_axis, size=2)
+            xmin = np.min(xx) - pad_w
+            xmax = np.max(xx) - pad_w
+            xmin = np.clip(xmin, 0, w - 1)
+            xmax = np.clip(xmax, 0, w - 1)
+            yy = np.random.choice(h_axis, size=2)
+            ymin = np.min(yy) - pad_h
+            ymax = np.max(yy) - pad_h
+            ymin = np.clip(ymin, 0, h - 1)
+            ymax = np.clip(ymax, 0, h - 1)
+            if (xmax - xmin) * (ymax - ymin) < area * self.min_area_ratio:
+                # area too small
+                continue
+
+            pts = np.stack(
+                [[xmin, xmax, xmax, xmin], [ymin, ymin, ymax, ymax]]
+            ).T.astype(np.int32)
+            pp = Polygon(pts)
+            fail_flag = False
+            for polygon, ignore_tag in zip(polygons, ignore_tags):
+                ppi = Polygon(polygon.reshape(-1, 2))
+                ppiou, _ = poly_intersection(ppi, pp, buffer=0)
+                if (
+                    np.abs(ppiou - float(ppi.area)) > self.epsilon
+                    and np.abs(ppiou) > self.epsilon
+                ):
+                    fail_flag = True
+                    break
+                elif np.abs(ppiou - float(ppi.area)) < self.epsilon:
+                    polys_new.append(polygon)
+                    ignore_tags_new.append(ignore_tag)
+                else:
+                    polys_keep.append(polygon)
+                    ignore_tags_keep.append(ignore_tag)
+
+            if fail_flag:
+                continue
+            else:
+                break
+
+        cropped = image[ymin:ymax, xmin:xmax, :]
+        select_type = np.random.randint(3)
+        if select_type == 0:
+            img = np.ascontiguousarray(cropped[:, ::-1])
+        elif select_type == 1:
+            img = np.ascontiguousarray(cropped[::-1, :])
+        else:
+            img = np.ascontiguousarray(cropped[::-1, ::-1])
+        image[ymin:ymax, xmin:xmax, :] = img
+        results["img"] = image
+
+        if len(polys_new) != 0:
+            height, width, _ = cropped.shape
+            if select_type == 0:
+                for idx, polygon in enumerate(polys_new):
+                    poly = polygon.reshape(-1, 2)
+                    poly[:, 0] = width - poly[:, 0] + 2 * xmin
+                    polys_new[idx] = poly
+            elif select_type == 1:
+                for idx, polygon in enumerate(polys_new):
+                    poly = polygon.reshape(-1, 2)
+                    poly[:, 1] = height - poly[:, 1] + 2 * ymin
+                    polys_new[idx] = poly
+            else:
+                for idx, polygon in enumerate(polys_new):
+                    poly = polygon.reshape(-1, 2)
+                    poly[:, 0] = width - poly[:, 0] + 2 * xmin
+                    poly[:, 1] = height - poly[:, 1] + 2 * ymin
+                    polys_new[idx] = poly
+            polygons = polys_keep + polys_new
+            ignore_tags = ignore_tags_keep + ignore_tags_new
+            results["polys"] = np.array(polygons)
+            results["ignore_tags"] = ignore_tags
+
+        return results
+
+    def generate_crop_target(self, image, all_polys, pad_h, pad_w):
+        """Generate crop target and make sure not to crop the polygon
+        instances.
+
+        Args:
+            image (ndarray): The image waited to be crop.
+            all_polys (list[list[ndarray]]): All polygons including ground
+                truth polygons and ground truth ignored polygons.
+            pad_h (int): Padding length of height.
+            pad_w (int): Padding length of width.
+        Returns:
+            h_axis (ndarray): Vertical cropping range.
+            w_axis (ndarray): Horizontal cropping range.
+        """
+        h, w, _ = image.shape
+        h_array = np.zeros((h + pad_h * 2), dtype=np.int32)
+        w_array = np.zeros((w + pad_w * 2), dtype=np.int32)
+
+        text_polys = []
+        for polygon in all_polys:
+            rect = cv2.minAreaRect(polygon.astype(np.int32).reshape(-1, 2))
+            box = cv2.boxPoints(rect)
+            box = np.int64(box)
+            text_polys.append([box[0], box[1], box[2], box[3]])
+
+        polys = np.array(text_polys, dtype=np.int32)
+        for poly in polys:
+            poly = np.round(poly, decimals=0).astype(np.int32)
+            minx = np.min(poly[:, 0])
+            maxx = np.max(poly[:, 0])
+            w_array[minx + pad_w : maxx + pad_w] = 1
+            miny = np.min(poly[:, 1])
+            maxy = np.max(poly[:, 1])
+            h_array[miny + pad_h : maxy + pad_h] = 1
+
+        h_axis = np.where(h_array == 0)[0]
+        w_axis = np.where(w_array == 0)[0]
+        return h_axis, w_axis
+
+
+class RandomCropPolyInstances:
+    """Randomly crop images and make sure to contain at least one intact
+    instance."""
+
+    def __init__(self, crop_ratio=5.0 / 8.0, min_side_ratio=0.4, **kwargs):
+        super().__init__()
+        self.crop_ratio = crop_ratio
+        self.min_side_ratio = min_side_ratio
+
+    def sample_valid_start_end(self, valid_array, min_len, max_start, min_end):
+        assert isinstance(min_len, int)
+        assert len(valid_array) > min_len
+
+        start_array = valid_array.copy()
+        max_start = min(len(start_array) - min_len, max_start)
+        start_array[max_start:] = 0
+        start_array[0] = 1
+        diff_array = np.hstack([0, start_array]) - np.hstack([start_array, 0])
+        region_starts = np.where(diff_array < 0)[0]
+        region_ends = np.where(diff_array > 0)[0]
+        region_ind = np.random.randint(0, len(region_starts))
+        start = np.random.randint(region_starts[region_ind], region_ends[region_ind])
+
+        end_array = valid_array.copy()
+        min_end = max(start + min_len, min_end)
+        end_array[:min_end] = 0
+        end_array[-1] = 1
+        diff_array = np.hstack([0, end_array]) - np.hstack([end_array, 0])
+        region_starts = np.where(diff_array < 0)[0]
+        region_ends = np.where(diff_array > 0)[0]
+        region_ind = np.random.randint(0, len(region_starts))
+        end = np.random.randint(region_starts[region_ind], region_ends[region_ind])
+        return start, end
+
+    def sample_crop_box(self, img_size, results):
+        """Generate crop box and make sure not to crop the polygon instances.
+
+        Args:
+            img_size (tuple(int)): The image size (h, w).
+            results (dict): The results dict.
+        """
+
+        assert isinstance(img_size, tuple)
+        h, w = img_size[:2]
+
+        key_masks = results["polys"]
+
+        x_valid_array = np.ones(w, dtype=np.int32)
+        y_valid_array = np.ones(h, dtype=np.int32)
+
+        selected_mask = key_masks[np.random.randint(0, len(key_masks))]
+        selected_mask = selected_mask.reshape((-1, 2)).astype(np.int32)
+        max_x_start = max(np.min(selected_mask[:, 0]) - 2, 0)
+        min_x_end = min(np.max(selected_mask[:, 0]) + 3, w - 1)
+        max_y_start = max(np.min(selected_mask[:, 1]) - 2, 0)
+        min_y_end = min(np.max(selected_mask[:, 1]) + 3, h - 1)
+
+        for mask in key_masks:
+            mask = mask.reshape((-1, 2)).astype(np.int32)
+            clip_x = np.clip(mask[:, 0], 0, w - 1)
+            clip_y = np.clip(mask[:, 1], 0, h - 1)
+            min_x, max_x = np.min(clip_x), np.max(clip_x)
+            min_y, max_y = np.min(clip_y), np.max(clip_y)
+
+            x_valid_array[min_x - 2 : max_x + 3] = 0
+            y_valid_array[min_y - 2 : max_y + 3] = 0
+
+        min_w = int(w * self.min_side_ratio)
+        min_h = int(h * self.min_side_ratio)
+
+        x1, x2 = self.sample_valid_start_end(
+            x_valid_array, min_w, max_x_start, min_x_end
+        )
+        y1, y2 = self.sample_valid_start_end(
+            y_valid_array, min_h, max_y_start, min_y_end
+        )
+
+        return np.array([x1, y1, x2, y2])
+
+    def crop_img(self, img, bbox):
+        assert img.ndim == 3
+        h, w, _ = img.shape
+        assert 0 <= bbox[1] < bbox[3] <= h
+        assert 0 <= bbox[0] < bbox[2] <= w
+        return img[bbox[1] : bbox[3], bbox[0] : bbox[2]]
+
+    def __call__(self, results):
+        image = results["image"]
+        polygons = results["polys"]
+        ignore_tags = results["ignore_tags"]
+        if len(polygons) < 1:
+            return results
+
+        if np.random.random_sample() < self.crop_ratio:
+            crop_box = self.sample_crop_box(image.shape, results)
+            img = self.crop_img(image, crop_box)
+            results["image"] = img
+            # crop and filter masks
+            x1, y1, x2, y2 = crop_box
+            w = max(x2 - x1, 1)
+            h = max(y2 - y1, 1)
+            polygons[:, :, 0::2] = polygons[:, :, 0::2] - x1
+            polygons[:, :, 1::2] = polygons[:, :, 1::2] - y1
+
+            valid_masks_list = []
+            valid_tags_list = []
+            for ind, polygon in enumerate(polygons):
+                if (
+                    (polygon[:, ::2] > -4).all()
+                    and (polygon[:, ::2] < w + 4).all()
+                    and (polygon[:, 1::2] > -4).all()
+                    and (polygon[:, 1::2] < h + 4).all()
+                ):
+                    polygon[:, ::2] = np.clip(polygon[:, ::2], 0, w)
+                    polygon[:, 1::2] = np.clip(polygon[:, 1::2], 0, h)
+                    valid_masks_list.append(polygon)
+                    valid_tags_list.append(ignore_tags[ind])
+
+            results["polys"] = np.array(valid_masks_list)
+            results["ignore_tags"] = valid_tags_list
+
+        return results
+
+    def __repr__(self):
+        repr_str = self.__class__.__name__
+        return repr_str
+
+
+class RandomRotatePolyInstances:
+    def __init__(
+        self,
+        rotate_ratio=0.5,
+        max_angle=10,
+        pad_with_fixed_color=False,
+        pad_value=(0, 0, 0),
+        **kwargs,
+    ):
+        """Randomly rotate images and polygon masks.
+
+        Args:
+            rotate_ratio (float): The ratio of samples to operate rotation.
+            max_angle (int): The maximum rotation angle.
+            pad_with_fixed_color (bool): The flag for whether to pad rotated
+               image with fixed value. If set to False, the rotated image will
+               be padded onto cropped image.
+            pad_value (tuple(int)): The color value for padding rotated image.
+        """
+        self.rotate_ratio = rotate_ratio
+        self.max_angle = max_angle
+        self.pad_with_fixed_color = pad_with_fixed_color
+        self.pad_value = pad_value
+
+    def rotate(self, center, points, theta, center_shift=(0, 0)):
+        # rotate points.
+        (center_x, center_y) = center
+        center_y = -center_y
+        x, y = points[:, ::2], points[:, 1::2]
+        y = -y
+
+        theta = theta / 180 * math.pi
+        cos = math.cos(theta)
+        sin = math.sin(theta)
+
+        x = x - center_x
+        y = y - center_y
+
+        _x = center_x + x * cos - y * sin + center_shift[0]
+        _y = -(center_y + x * sin + y * cos) + center_shift[1]
+
+        points[:, ::2], points[:, 1::2] = _x, _y
+        return points
+
+    def cal_canvas_size(self, ori_size, degree):
+        assert isinstance(ori_size, tuple)
+        angle = degree * math.pi / 180.0
+        h, w = ori_size[:2]
+
+        cos = math.cos(angle)
+        sin = math.sin(angle)
+        canvas_h = int(w * math.fabs(sin) + h * math.fabs(cos))
+        canvas_w = int(w * math.fabs(cos) + h * math.fabs(sin))
+
+        canvas_size = (canvas_h, canvas_w)
+        return canvas_size
+
+    def sample_angle(self, max_angle):
+        angle = np.random.random_sample() * 2 * max_angle - max_angle
+        return angle
+
+    def rotate_img(self, img, angle, canvas_size):
+        h, w = img.shape[:2]
+        rotation_matrix = cv2.getRotationMatrix2D((w / 2, h / 2), angle, 1)
+        rotation_matrix[0, 2] += int((canvas_size[1] - w) / 2)
+        rotation_matrix[1, 2] += int((canvas_size[0] - h) / 2)
+
+        if self.pad_with_fixed_color:
+            target_img = cv2.warpAffine(
+                img,
+                rotation_matrix,
+                (canvas_size[1], canvas_size[0]),
+                flags=cv2.INTER_NEAREST,
+                borderValue=self.pad_value,
+            )
+        else:
+            mask = np.zeros_like(img)
+            (h_ind, w_ind) = (
+                np.random.randint(0, h * 7 // 8),
+                np.random.randint(0, w * 7 // 8),
+            )
+            img_cut = img[h_ind : (h_ind + h // 9), w_ind : (w_ind + w // 9)]
+            img_cut = cv2.resize(img_cut, (canvas_size[1], canvas_size[0]))
+
+            mask = cv2.warpAffine(
+                mask,
+                rotation_matrix,
+                (canvas_size[1], canvas_size[0]),
+                borderValue=[1, 1, 1],
+            )
+            target_img = cv2.warpAffine(
+                img,
+                rotation_matrix,
+                (canvas_size[1], canvas_size[0]),
+                borderValue=[0, 0, 0],
+            )
+            target_img = target_img + img_cut * mask
+
+        return target_img
+
+    def __call__(self, results):
+        if np.random.random_sample() < self.rotate_ratio:
+            image = results["image"]
+            polygons = results["polys"]
+            h, w = image.shape[:2]
+
+            angle = self.sample_angle(self.max_angle)
+            canvas_size = self.cal_canvas_size((h, w), angle)
+            center_shift = (
+                int((canvas_size[1] - w) / 2),
+                int((canvas_size[0] - h) / 2),
+            )
+            image = self.rotate_img(image, angle, canvas_size)
+            results["image"] = image
+            # rotate polygons
+            rotated_masks = []
+            for mask in polygons:
+                rotated_mask = self.rotate((w / 2, h / 2), mask, angle, center_shift)
+                rotated_masks.append(rotated_mask)
+            results["polys"] = np.array(rotated_masks)
+
+        return results
+
+    def __repr__(self):
+        repr_str = self.__class__.__name__
+        return repr_str
+
+
+class SquareResizePad:
+    def __init__(
+        self,
+        target_size,
+        pad_ratio=0.6,
+        pad_with_fixed_color=False,
+        pad_value=(0, 0, 0),
+        **kwargs,
+    ):
+        """Resize or pad images to be square shape.
+
+        Args:
+            target_size (int): The target size of square shaped image.
+            pad_with_fixed_color (bool): The flag for whether to pad rotated
+               image with fixed value. If set to False, the rescales image will
+               be padded onto cropped image.
+            pad_value (tuple(int)): The color value for padding rotated image.
+        """
+        assert isinstance(target_size, int)
+        assert isinstance(pad_ratio, float)
+        assert isinstance(pad_with_fixed_color, bool)
+        assert isinstance(pad_value, tuple)
+
+        self.target_size = target_size
+        self.pad_ratio = pad_ratio
+        self.pad_with_fixed_color = pad_with_fixed_color
+        self.pad_value = pad_value
+
+    def resize_img(self, img, keep_ratio=True):
+        h, w, _ = img.shape
+        if keep_ratio:
+            t_h = self.target_size if h >= w else int(h * self.target_size / w)
+            t_w = self.target_size if h <= w else int(w * self.target_size / h)
+        else:
+            t_h = t_w = self.target_size
+        img = cv2.resize(img, (t_w, t_h))
+        return img, (t_h, t_w)
+
+    def square_pad(self, img):
+        h, w = img.shape[:2]
+        if h == w:
+            return img, (0, 0)
+        pad_size = max(h, w)
+        if self.pad_with_fixed_color:
+            expand_img = np.ones((pad_size, pad_size, 3), dtype=np.uint8)
+            expand_img[:] = self.pad_value
+        else:
+            (h_ind, w_ind) = (
+                np.random.randint(0, h * 7 // 8),
+                np.random.randint(0, w * 7 // 8),
+            )
+            img_cut = img[h_ind : (h_ind + h // 9), w_ind : (w_ind + w // 9)]
+            expand_img = cv2.resize(img_cut, (pad_size, pad_size))
+        if h > w:
+            y0, x0 = 0, (h - w) // 2
+        else:
+            y0, x0 = (w - h) // 2, 0
+        expand_img[y0 : y0 + h, x0 : x0 + w] = img
+        offset = (x0, y0)
+
+        return expand_img, offset
+
+    def square_pad_mask(self, points, offset):
+        x0, y0 = offset
+        pad_points = points.copy()
+        pad_points[::2] = pad_points[::2] + x0
+        pad_points[1::2] = pad_points[1::2] + y0
+        return pad_points
+
+    def __call__(self, results):
+        image = results["image"]
+        polygons = results["polys"]
+        h, w = image.shape[:2]
+
+        if np.random.random_sample() < self.pad_ratio:
+            image, out_size = self.resize_img(image, keep_ratio=True)
+            image, offset = self.square_pad(image)
+        else:
+            image, out_size = self.resize_img(image, keep_ratio=False)
+            offset = (0, 0)
+        results["image"] = image
+        try:
+            polygons[:, :, 0::2] = polygons[:, :, 0::2] * out_size[1] / w + offset[0]
+            polygons[:, :, 1::2] = polygons[:, :, 1::2] * out_size[0] / h + offset[1]
+        except:
+            pass
+        results["polys"] = polygons
+
+        return results
+
+    def __repr__(self):
+        repr_str = self.__class__.__name__
+        return repr_str
diff --git a/ppocr/data/imaug/fce_targets.py b/ppocr/data/imaug/fce_targets.py
new file mode 100644
index 0000000..9955980
--- /dev/null
+++ b/ppocr/data/imaug/fce_targets.py
@@ -0,0 +1,697 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/datasets/pipelines/textdet_targets/fcenet_targets.py
+"""
+
+import cv2
+import numpy as np
+from numpy.fft import fft
+from numpy.linalg import norm
+import sys
+
+
+def vector_slope(vec):
+    assert len(vec) == 2
+    return abs(vec[1] / (vec[0] + 1e-8))
+
+
+class FCENetTargets:
+    """Generate the ground truth targets of FCENet: Fourier Contour Embedding
+    for Arbitrary-Shaped Text Detection.
+
+    [https://arxiv.org/abs/2104.10442]
+
+    Args:
+        fourier_degree (int): The maximum Fourier transform degree k.
+        resample_step (float): The step size for resampling the text center
+            line (TCL). It's better not to exceed half of the minimum width.
+        center_region_shrink_ratio (float): The shrink ratio of text center
+            region.
+        level_size_divisors (tuple(int)): The downsample ratio on each level.
+        level_proportion_range (tuple(tuple(int))): The range of text sizes
+            assigned to each level.
+    """
+
+    def __init__(
+        self,
+        fourier_degree=5,
+        resample_step=4.0,
+        center_region_shrink_ratio=0.3,
+        level_size_divisors=(8, 16, 32),
+        level_proportion_range=((0, 0.25), (0.2, 0.65), (0.55, 1.0)),
+        orientation_thr=2.0,
+        **kwargs,
+    ):
+        super().__init__()
+        assert isinstance(level_size_divisors, tuple)
+        assert isinstance(level_proportion_range, tuple)
+        assert len(level_size_divisors) == len(level_proportion_range)
+        self.fourier_degree = fourier_degree
+        self.resample_step = resample_step
+        self.center_region_shrink_ratio = center_region_shrink_ratio
+        self.level_size_divisors = level_size_divisors
+        self.level_proportion_range = level_proportion_range
+
+        self.orientation_thr = orientation_thr
+
+    def vector_angle(self, vec1, vec2):
+        if vec1.ndim > 1:
+            unit_vec1 = vec1 / (norm(vec1, axis=-1) + 1e-8).reshape((-1, 1))
+        else:
+            unit_vec1 = vec1 / (norm(vec1, axis=-1) + 1e-8)
+        if vec2.ndim > 1:
+            unit_vec2 = vec2 / (norm(vec2, axis=-1) + 1e-8).reshape((-1, 1))
+        else:
+            unit_vec2 = vec2 / (norm(vec2, axis=-1) + 1e-8)
+        return np.arccos(np.clip(np.sum(unit_vec1 * unit_vec2, axis=-1), -1.0, 1.0))
+
+    def resample_line(self, line, n):
+        """Resample n points on a line.
+
+        Args:
+            line (ndarray): The points composing a line.
+            n (int): The resampled points number.
+
+        Returns:
+            resampled_line (ndarray): The points composing the resampled line.
+        """
+
+        assert line.ndim == 2
+        assert line.shape[0] >= 2
+        assert line.shape[1] == 2
+        assert isinstance(n, int)
+        assert n > 0
+
+        length_list = [norm(line[i + 1] - line[i]) for i in range(len(line) - 1)]
+        total_length = sum(length_list)
+        length_cumsum = np.cumsum([0.0] + length_list)
+        delta_length = total_length / (float(n) + 1e-8)
+
+        current_edge_ind = 0
+        resampled_line = [line[0]]
+
+        for i in range(1, n):
+            current_line_len = i * delta_length
+
+            while (
+                current_edge_ind + 1 < len(length_cumsum)
+                and current_line_len >= length_cumsum[current_edge_ind + 1]
+            ):
+                current_edge_ind += 1
+
+            current_edge_end_shift = current_line_len - length_cumsum[current_edge_ind]
+
+            if current_edge_ind >= len(length_list):
+                break
+            end_shift_ratio = current_edge_end_shift / length_list[current_edge_ind]
+            current_point = (
+                line[current_edge_ind]
+                + (line[current_edge_ind + 1] - line[current_edge_ind])
+                * end_shift_ratio
+            )
+            resampled_line.append(current_point)
+        resampled_line.append(line[-1])
+        resampled_line = np.array(resampled_line)
+
+        return resampled_line
+
+    def reorder_poly_edge(self, points):
+        """Get the respective points composing head edge, tail edge, top
+        sideline and bottom sideline.
+
+        Args:
+            points (ndarray): The points composing a text polygon.
+
+        Returns:
+            head_edge (ndarray): The two points composing the head edge of text
+                polygon.
+            tail_edge (ndarray): The two points composing the tail edge of text
+                polygon.
+            top_sideline (ndarray): The points composing top curved sideline of
+                text polygon.
+            bot_sideline (ndarray): The points composing bottom curved sideline
+                of text polygon.
+        """
+
+        assert points.ndim == 2
+        assert points.shape[0] >= 4
+        assert points.shape[1] == 2
+
+        head_inds, tail_inds = self.find_head_tail(points, self.orientation_thr)
+        head_edge, tail_edge = points[head_inds], points[tail_inds]
+
+        pad_points = np.vstack([points, points])
+        if tail_inds[1] < 1:
+            tail_inds[1] = len(points)
+        sideline1 = pad_points[head_inds[1] : tail_inds[1]]
+        sideline2 = pad_points[tail_inds[1] : (head_inds[1] + len(points))]
+        sideline_mean_shift = np.mean(sideline1, axis=0) - np.mean(sideline2, axis=0)
+
+        if sideline_mean_shift[1] > 0:
+            top_sideline, bot_sideline = sideline2, sideline1
+        else:
+            top_sideline, bot_sideline = sideline1, sideline2
+
+        return head_edge, tail_edge, top_sideline, bot_sideline
+
+    def find_head_tail(self, points, orientation_thr):
+        """Find the head edge and tail edge of a text polygon.
+
+        Args:
+            points (ndarray): The points composing a text polygon.
+            orientation_thr (float): The threshold for distinguishing between
+                head edge and tail edge among the horizontal and vertical edges
+                of a quadrangle.
+
+        Returns:
+            head_inds (list): The indexes of two points composing head edge.
+            tail_inds (list): The indexes of two points composing tail edge.
+        """
+
+        assert points.ndim == 2
+        assert points.shape[0] >= 4
+        assert points.shape[1] == 2
+        assert isinstance(orientation_thr, float)
+
+        if len(points) > 4:
+            pad_points = np.vstack([points, points[0]])
+            edge_vec = pad_points[1:] - pad_points[:-1]
+
+            theta_sum = []
+            adjacent_vec_theta = []
+            for i, edge_vec1 in enumerate(edge_vec):
+                adjacent_ind = [x % len(edge_vec) for x in [i - 1, i + 1]]
+                adjacent_edge_vec = edge_vec[adjacent_ind]
+                temp_theta_sum = np.sum(self.vector_angle(edge_vec1, adjacent_edge_vec))
+                temp_adjacent_theta = self.vector_angle(
+                    adjacent_edge_vec[0], adjacent_edge_vec[1]
+                )
+                theta_sum.append(temp_theta_sum)
+                adjacent_vec_theta.append(temp_adjacent_theta)
+            theta_sum_score = np.array(theta_sum) / np.pi
+            adjacent_theta_score = np.array(adjacent_vec_theta) / np.pi
+            poly_center = np.mean(points, axis=0)
+            edge_dist = np.maximum(
+                norm(pad_points[1:] - poly_center, axis=-1),
+                norm(pad_points[:-1] - poly_center, axis=-1),
+            )
+            dist_score = edge_dist / np.max(edge_dist)
+            position_score = np.zeros(len(edge_vec))
+            score = 0.5 * theta_sum_score + 0.15 * adjacent_theta_score
+            score += 0.35 * dist_score
+            if len(points) % 2 == 0:
+                position_score[(len(score) // 2 - 1)] += 1
+                position_score[-1] += 1
+            score += 0.1 * position_score
+            pad_score = np.concatenate([score, score])
+            score_matrix = np.zeros((len(score), len(score) - 3))
+            x = np.arange(len(score) - 3) / float(len(score) - 4)
+            gaussian = (
+                1.0
+                / (np.sqrt(2.0 * np.pi) * 0.5)
+                * np.exp(-np.power((x - 0.5) / 0.5, 2.0) / 2)
+            )
+            gaussian = gaussian / np.max(gaussian)
+            for i in range(len(score)):
+                score_matrix[i, :] = (
+                    score[i]
+                    + pad_score[(i + 2) : (i + len(score) - 1)] * gaussian * 0.3
+                )
+
+            head_start, tail_increment = np.unravel_index(
+                score_matrix.argmax(), score_matrix.shape
+            )
+            tail_start = (head_start + tail_increment + 2) % len(points)
+            head_end = (head_start + 1) % len(points)
+            tail_end = (tail_start + 1) % len(points)
+
+            if head_end > tail_end:
+                head_start, tail_start = tail_start, head_start
+                head_end, tail_end = tail_end, head_end
+            head_inds = [head_start, head_end]
+            tail_inds = [tail_start, tail_end]
+        else:
+            if vector_slope(points[1] - points[0]) + vector_slope(
+                points[3] - points[2]
+            ) < vector_slope(points[2] - points[1]) + vector_slope(
+                points[0] - points[3]
+            ):
+                horizontal_edge_inds = [[0, 1], [2, 3]]
+                vertical_edge_inds = [[3, 0], [1, 2]]
+            else:
+                horizontal_edge_inds = [[3, 0], [1, 2]]
+                vertical_edge_inds = [[0, 1], [2, 3]]
+
+            vertical_len_sum = norm(
+                points[vertical_edge_inds[0][0]] - points[vertical_edge_inds[0][1]]
+            ) + norm(
+                points[vertical_edge_inds[1][0]] - points[vertical_edge_inds[1][1]]
+            )
+            horizontal_len_sum = norm(
+                points[horizontal_edge_inds[0][0]] - points[horizontal_edge_inds[0][1]]
+            ) + norm(
+                points[horizontal_edge_inds[1][0]] - points[horizontal_edge_inds[1][1]]
+            )
+
+            if vertical_len_sum > horizontal_len_sum * orientation_thr:
+                head_inds = horizontal_edge_inds[0]
+                tail_inds = horizontal_edge_inds[1]
+            else:
+                head_inds = vertical_edge_inds[0]
+                tail_inds = vertical_edge_inds[1]
+
+        return head_inds, tail_inds
+
+    def resample_sidelines(self, sideline1, sideline2, resample_step):
+        """Resample two sidelines to be of the same points number according to
+        step size.
+
+        Args:
+            sideline1 (ndarray): The points composing a sideline of a text
+                polygon.
+            sideline2 (ndarray): The points composing another sideline of a
+                text polygon.
+            resample_step (float): The resampled step size.
+
+        Returns:
+            resampled_line1 (ndarray): The resampled line 1.
+            resampled_line2 (ndarray): The resampled line 2.
+        """
+
+        assert sideline1.ndim == sideline2.ndim == 2
+        assert sideline1.shape[1] == sideline2.shape[1] == 2
+        assert sideline1.shape[0] >= 2
+        assert sideline2.shape[0] >= 2
+        assert isinstance(resample_step, float)
+
+        length1 = sum(
+            [norm(sideline1[i + 1] - sideline1[i]) for i in range(len(sideline1) - 1)]
+        )
+        length2 = sum(
+            [norm(sideline2[i + 1] - sideline2[i]) for i in range(len(sideline2) - 1)]
+        )
+
+        total_length = (length1 + length2) / 2
+        resample_point_num = max(int(float(total_length) / resample_step), 1)
+
+        resampled_line1 = self.resample_line(sideline1, resample_point_num)
+        resampled_line2 = self.resample_line(sideline2, resample_point_num)
+
+        return resampled_line1, resampled_line2
+
+    def generate_center_region_mask(self, img_size, text_polys):
+        """Generate text center region mask.
+
+        Args:
+            img_size (tuple): The image size of (height, width).
+            text_polys (list[list[ndarray]]): The list of text polygons.
+
+        Returns:
+            center_region_mask (ndarray): The text center region mask.
+        """
+
+        assert isinstance(img_size, tuple)
+        # assert check_argument.is_2dlist(text_polys)
+
+        h, w = img_size
+
+        center_region_mask = np.zeros((h, w), np.uint8)
+
+        center_region_boxes = []
+        for poly in text_polys:
+            # assert len(poly) == 1
+            polygon_points = poly.reshape(-1, 2)
+            _, _, top_line, bot_line = self.reorder_poly_edge(polygon_points)
+            resampled_top_line, resampled_bot_line = self.resample_sidelines(
+                top_line, bot_line, self.resample_step
+            )
+            resampled_bot_line = resampled_bot_line[::-1]
+            if len(resampled_top_line) != len(resampled_bot_line):
+                continue
+            center_line = (resampled_top_line + resampled_bot_line) / 2
+
+            line_head_shrink_len = (
+                norm(resampled_top_line[0] - resampled_bot_line[0]) / 4.0
+            )
+            line_tail_shrink_len = (
+                norm(resampled_top_line[-1] - resampled_bot_line[-1]) / 4.0
+            )
+            head_shrink_num = int(line_head_shrink_len // self.resample_step)
+            tail_shrink_num = int(line_tail_shrink_len // self.resample_step)
+            if len(center_line) > head_shrink_num + tail_shrink_num + 2:
+                center_line = center_line[
+                    head_shrink_num : len(center_line) - tail_shrink_num
+                ]
+                resampled_top_line = resampled_top_line[
+                    head_shrink_num : len(resampled_top_line) - tail_shrink_num
+                ]
+                resampled_bot_line = resampled_bot_line[
+                    head_shrink_num : len(resampled_bot_line) - tail_shrink_num
+                ]
+
+            for i in range(0, len(center_line) - 1):
+                tl = (
+                    center_line[i]
+                    + (resampled_top_line[i] - center_line[i])
+                    * self.center_region_shrink_ratio
+                )
+                tr = (
+                    center_line[i + 1]
+                    + (resampled_top_line[i + 1] - center_line[i + 1])
+                    * self.center_region_shrink_ratio
+                )
+                br = (
+                    center_line[i + 1]
+                    + (resampled_bot_line[i + 1] - center_line[i + 1])
+                    * self.center_region_shrink_ratio
+                )
+                bl = (
+                    center_line[i]
+                    + (resampled_bot_line[i] - center_line[i])
+                    * self.center_region_shrink_ratio
+                )
+                current_center_box = np.vstack([tl, tr, br, bl]).astype(np.int32)
+                center_region_boxes.append(current_center_box)
+
+        cv2.fillPoly(center_region_mask, center_region_boxes, 1)
+        return center_region_mask
+
+    def resample_polygon(self, polygon, n=400):
+        """Resample one polygon with n points on its boundary.
+
+        Args:
+            polygon (list[float]): The input polygon.
+            n (int): The number of resampled points.
+        Returns:
+            resampled_polygon (list[float]): The resampled polygon.
+        """
+        length = []
+
+        for i in range(len(polygon)):
+            p1 = polygon[i]
+            if i == len(polygon) - 1:
+                p2 = polygon[0]
+            else:
+                p2 = polygon[i + 1]
+            length.append(((p1[0] - p2[0]) ** 2 + (p1[1] - p2[1]) ** 2) ** 0.5)
+
+        total_length = sum(length)
+        n_on_each_line = (np.array(length) / (total_length + 1e-8)) * n
+        n_on_each_line = n_on_each_line.astype(np.int32)
+        new_polygon = []
+
+        for i in range(len(polygon)):
+            num = n_on_each_line[i]
+            p1 = polygon[i]
+            if i == len(polygon) - 1:
+                p2 = polygon[0]
+            else:
+                p2 = polygon[i + 1]
+
+            if num == 0:
+                continue
+
+            dxdy = (p2 - p1) / num
+            for j in range(num):
+                point = p1 + dxdy * j
+                new_polygon.append(point)
+
+        return np.array(new_polygon)
+
+    def normalize_polygon(self, polygon):
+        """Normalize one polygon so that its start point is at right most.
+
+        Args:
+            polygon (list[float]): The origin polygon.
+        Returns:
+            new_polygon (lost[float]): The polygon with start point at right.
+        """
+        temp_polygon = polygon - polygon.mean(axis=0)
+        x = np.abs(temp_polygon[:, 0])
+        y = temp_polygon[:, 1]
+        index_x = np.argsort(x)
+        index_y = np.argmin(y[index_x[:8]])
+        index = index_x[index_y]
+        new_polygon = np.concatenate([polygon[index:], polygon[:index]])
+        return new_polygon
+
+    def poly2fourier(self, polygon, fourier_degree):
+        """Perform Fourier transformation to generate Fourier coefficients ck
+        from polygon.
+
+        Args:
+            polygon (ndarray): An input polygon.
+            fourier_degree (int): The maximum Fourier degree K.
+        Returns:
+            c (ndarray(complex)): Fourier coefficients.
+        """
+        points = polygon[:, 0] + polygon[:, 1] * 1j
+        c_fft = fft(points) / len(points)
+        c = np.hstack((c_fft[-fourier_degree:], c_fft[: fourier_degree + 1]))
+        return c
+
+    def clockwise(self, c, fourier_degree):
+        """Make sure the polygon reconstructed from Fourier coefficients c in
+        the clockwise direction.
+
+        Args:
+            polygon (list[float]): The origin polygon.
+        Returns:
+            new_polygon (lost[float]): The polygon in clockwise point order.
+        """
+        if np.abs(c[fourier_degree + 1]) > np.abs(c[fourier_degree - 1]):
+            return c
+        elif np.abs(c[fourier_degree + 1]) < np.abs(c[fourier_degree - 1]):
+            return c[::-1]
+        else:
+            if np.abs(c[fourier_degree + 2]) > np.abs(c[fourier_degree - 2]):
+                return c
+            else:
+                return c[::-1]
+
+    def cal_fourier_signature(self, polygon, fourier_degree):
+        """Calculate Fourier signature from input polygon.
+
+        Args:
+              polygon (ndarray): The input polygon.
+              fourier_degree (int): The maximum Fourier degree K.
+        Returns:
+              fourier_signature (ndarray): An array shaped (2k+1, 2) containing
+                  real part and image part of 2k+1 Fourier coefficients.
+        """
+        resampled_polygon = self.resample_polygon(polygon)
+        resampled_polygon = self.normalize_polygon(resampled_polygon)
+
+        fourier_coeff = self.poly2fourier(resampled_polygon, fourier_degree)
+        fourier_coeff = self.clockwise(fourier_coeff, fourier_degree)
+
+        real_part = np.real(fourier_coeff).reshape((-1, 1))
+        image_part = np.imag(fourier_coeff).reshape((-1, 1))
+        fourier_signature = np.hstack([real_part, image_part])
+
+        return fourier_signature
+
+    def generate_fourier_maps(self, img_size, text_polys):
+        """Generate Fourier coefficient maps.
+
+        Args:
+            img_size (tuple): The image size of (height, width).
+            text_polys (list[list[ndarray]]): The list of text polygons.
+
+        Returns:
+            fourier_real_map (ndarray): The Fourier coefficient real part maps.
+            fourier_image_map (ndarray): The Fourier coefficient image part
+                maps.
+        """
+
+        assert isinstance(img_size, tuple)
+
+        h, w = img_size
+        k = self.fourier_degree
+        real_map = np.zeros((k * 2 + 1, h, w), dtype=np.float32)
+        imag_map = np.zeros((k * 2 + 1, h, w), dtype=np.float32)
+
+        for poly in text_polys:
+            mask = np.zeros((h, w), dtype=np.uint8)
+            polygon = np.array(poly).reshape((1, -1, 2))
+            cv2.fillPoly(mask, polygon.astype(np.int32), 1)
+            fourier_coeff = self.cal_fourier_signature(polygon[0], k)
+            for i in range(-k, k + 1):
+                if i != 0:
+                    real_map[i + k, :, :] = (
+                        mask * fourier_coeff[i + k, 0]
+                        + (1 - mask) * real_map[i + k, :, :]
+                    )
+                    imag_map[i + k, :, :] = (
+                        mask * fourier_coeff[i + k, 1]
+                        + (1 - mask) * imag_map[i + k, :, :]
+                    )
+                else:
+                    yx = np.argwhere(mask > 0.5)
+                    k_ind = np.ones((len(yx)), dtype=np.int64) * k
+                    y, x = yx[:, 0], yx[:, 1]
+                    real_map[k_ind, y, x] = fourier_coeff[k, 0] - x
+                    imag_map[k_ind, y, x] = fourier_coeff[k, 1] - y
+
+        return real_map, imag_map
+
+    def generate_text_region_mask(self, img_size, text_polys):
+        """Generate text center region mask and geometry attribute maps.
+
+        Args:
+            img_size (tuple): The image size (height, width).
+            text_polys (list[list[ndarray]]): The list of text polygons.
+
+        Returns:
+            text_region_mask (ndarray): The text region mask.
+        """
+
+        assert isinstance(img_size, tuple)
+
+        h, w = img_size
+        text_region_mask = np.zeros((h, w), dtype=np.uint8)
+
+        for poly in text_polys:
+            polygon = np.array(poly, dtype=np.int32).reshape((1, -1, 2))
+            cv2.fillPoly(text_region_mask, polygon, 1)
+
+        return text_region_mask
+
+    def generate_effective_mask(self, mask_size: tuple, polygons_ignore):
+        """Generate effective mask by setting the ineffective regions to 0 and
+        effective regions to 1.
+
+        Args:
+            mask_size (tuple): The mask size.
+            polygons_ignore (list[[ndarray]]: The list of ignored text
+                polygons.
+
+        Returns:
+            mask (ndarray): The effective mask of (height, width).
+        """
+
+        mask = np.ones(mask_size, dtype=np.uint8)
+
+        for poly in polygons_ignore:
+            instance = poly.reshape(-1, 2).astype(np.int32).reshape(1, -1, 2)
+            cv2.fillPoly(mask, instance, 0)
+
+        return mask
+
+    def generate_level_targets(self, img_size, text_polys, ignore_polys):
+        """Generate ground truth target on each level.
+
+        Args:
+            img_size (list[int]): Shape of input image.
+            text_polys (list[list[ndarray]]): A list of ground truth polygons.
+            ignore_polys (list[list[ndarray]]): A list of ignored polygons.
+        Returns:
+            level_maps (list(ndarray)): A list of ground target on each level.
+        """
+        h, w = img_size
+        lv_size_divs = self.level_size_divisors
+        lv_proportion_range = self.level_proportion_range
+        lv_text_polys = [[] for i in range(len(lv_size_divs))]
+        lv_ignore_polys = [[] for i in range(len(lv_size_divs))]
+        level_maps = []
+        for poly in text_polys:
+            polygon = np.array(poly, dtype=np.int32).reshape((1, -1, 2))
+            _, _, box_w, box_h = cv2.boundingRect(polygon)
+            proportion = max(box_h, box_w) / (h + 1e-8)
+
+            for ind, proportion_range in enumerate(lv_proportion_range):
+                if proportion_range[0] < proportion < proportion_range[1]:
+                    lv_text_polys[ind].append(poly / lv_size_divs[ind])
+
+        for ignore_poly in ignore_polys:
+            polygon = np.array(ignore_poly, dtype=np.int32).reshape((1, -1, 2))
+            _, _, box_w, box_h = cv2.boundingRect(polygon)
+            proportion = max(box_h, box_w) / (h + 1e-8)
+
+            for ind, proportion_range in enumerate(lv_proportion_range):
+                if proportion_range[0] < proportion < proportion_range[1]:
+                    lv_ignore_polys[ind].append(ignore_poly / lv_size_divs[ind])
+
+        for ind, size_divisor in enumerate(lv_size_divs):
+            current_level_maps = []
+            level_img_size = (h // size_divisor, w // size_divisor)
+
+            text_region = self.generate_text_region_mask(
+                level_img_size, lv_text_polys[ind]
+            )[None]
+            current_level_maps.append(text_region)
+
+            center_region = self.generate_center_region_mask(
+                level_img_size, lv_text_polys[ind]
+            )[None]
+            current_level_maps.append(center_region)
+
+            effective_mask = self.generate_effective_mask(
+                level_img_size, lv_ignore_polys[ind]
+            )[None]
+            current_level_maps.append(effective_mask)
+
+            fourier_real_map, fourier_image_maps = self.generate_fourier_maps(
+                level_img_size, lv_text_polys[ind]
+            )
+            current_level_maps.append(fourier_real_map)
+            current_level_maps.append(fourier_image_maps)
+
+            level_maps.append(np.concatenate(current_level_maps))
+
+        return level_maps
+
+    def generate_targets(self, results):
+        """Generate the ground truth targets for FCENet.
+
+        Args:
+            results (dict): The input result dictionary.
+
+        Returns:
+            results (dict): The output result dictionary.
+        """
+
+        assert isinstance(results, dict)
+        image = results["image"]
+        polygons = results["polys"]
+        ignore_tags = results["ignore_tags"]
+        h, w, _ = image.shape
+
+        polygon_masks = []
+        polygon_masks_ignore = []
+        for tag, polygon in zip(ignore_tags, polygons):
+            if tag is True:
+                polygon_masks_ignore.append(polygon)
+            else:
+                polygon_masks.append(polygon)
+
+        level_maps = self.generate_level_targets(
+            (h, w), polygon_masks, polygon_masks_ignore
+        )
+
+        mapping = {
+            "p3_maps": level_maps[0],
+            "p4_maps": level_maps[1],
+            "p5_maps": level_maps[2],
+        }
+        for key, value in mapping.items():
+            results[key] = value
+
+        return results
+
+    def __call__(self, results):
+        results = self.generate_targets(results)
+        return results
diff --git a/ppocr/data/imaug/frost_img/frost1.jpg b/ppocr/data/imaug/frost_img/frost1.jpg
new file mode 100644
index 0000000..8705320
Binary files /dev/null and b/ppocr/data/imaug/frost_img/frost1.jpg differ
diff --git a/ppocr/data/imaug/frost_img/frost2.png b/ppocr/data/imaug/frost_img/frost2.png
new file mode 100644
index 0000000..48f7a86
Binary files /dev/null and b/ppocr/data/imaug/frost_img/frost2.png differ
diff --git a/ppocr/data/imaug/frost_img/frost3.png b/ppocr/data/imaug/frost_img/frost3.png
new file mode 100644
index 0000000..d47f9d2
Binary files /dev/null and b/ppocr/data/imaug/frost_img/frost3.png differ
diff --git a/ppocr/data/imaug/frost_img/frost4.jpg b/ppocr/data/imaug/frost_img/frost4.jpg
new file mode 100644
index 0000000..f8b0c41
Binary files /dev/null and b/ppocr/data/imaug/frost_img/frost4.jpg differ
diff --git a/ppocr/data/imaug/frost_img/frost5.jpg b/ppocr/data/imaug/frost_img/frost5.jpg
new file mode 100644
index 0000000..95dc905
Binary files /dev/null and b/ppocr/data/imaug/frost_img/frost5.jpg differ
diff --git a/ppocr/data/imaug/frost_img/frost6.jpg b/ppocr/data/imaug/frost_img/frost6.jpg
new file mode 100644
index 0000000..14e5d58
Binary files /dev/null and b/ppocr/data/imaug/frost_img/frost6.jpg differ
diff --git a/ppocr/data/imaug/iaa_augment.py b/ppocr/data/imaug/iaa_augment.py
new file mode 100644
index 0000000..4b2c67f
--- /dev/null
+++ b/ppocr/data/imaug/iaa_augment.py
@@ -0,0 +1,220 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/WenmuZhou/DBNet.pytorch/blob/master/data_loader/modules/iaa_augment.py
+"""
+import os
+
+# Prevent automatic updates in Albumentations for stability in augmentation behavior
+os.environ["NO_ALBUMENTATIONS_UPDATE"] = "1"
+
+import numpy as np
+import albumentations as A
+from albumentations.core.transforms_interface import DualTransform
+from albumentations.augmentations.geometric import functional as fgeometric
+from packaging import version
+
+ALBU_VERSION = version.parse(A.__version__)
+IS_ALBU_NEW_VERSION = ALBU_VERSION >= version.parse("1.4.15")
+
+
+# Custom resize transformation mimicking Imgaug's behavior with scaling
+class ImgaugLikeResize(DualTransform):
+    def __init__(self, scale_range=(0.5, 3.0), interpolation=1, p=1.0):
+        super(ImgaugLikeResize, self).__init__(p)
+        self.scale_range = scale_range
+        self.interpolation = interpolation
+
+    # Resize the image based on a randomly chosen scale within the scale range
+    def apply(self, img, scale=1.0, **params):
+        height, width = img.shape[:2]
+        new_height = int(height * scale)
+        new_width = int(width * scale)
+
+        if IS_ALBU_NEW_VERSION:
+            return fgeometric.resize(
+                img, (new_height, new_width), interpolation=self.interpolation
+            )
+        return fgeometric.resize(
+            img, new_height, new_width, interpolation=self.interpolation
+        )
+
+    # Apply the same scaling transformation to keypoints (e.g., polygon points)
+    def apply_to_keypoints(self, keypoints, scale=1.0, **params):
+        return np.array(
+            [(x * scale, y * scale) + tuple(rest) for x, y, *rest in keypoints]
+        )
+
+    # Get random scale parameter within the specified range
+    def get_params(self):
+        scale = np.random.uniform(self.scale_range[0], self.scale_range[1])
+        return {"scale": scale}
+
+
+# Builder class to translate custom augmenter arguments into Albumentations-compatible format
+class AugmenterBuilder(object):
+    def __init__(self):
+        # Map common Imgaug transformations to equivalent Albumentations transforms
+        self.imgaug_to_albu = {
+            "Fliplr": "HorizontalFlip",
+            "Flipud": "VerticalFlip",
+            "Affine": "Affine",
+            # Additional mappings can be added here if needed
+        }
+
+    # Recursive method to construct augmentation pipeline based on provided arguments
+    def build(self, args, root=True):
+        if args is None or len(args) == 0:
+            return None
+        elif isinstance(args, list):
+            # Build the full augmentation sequence if it's a root-level call
+            if root:
+                sequence = [self.build(value, root=False) for value in args]
+                return A.Compose(
+                    sequence,
+                    keypoint_params=A.KeypointParams(
+                        format="xy", remove_invisible=False
+                    ),
+                )
+            else:
+                # Build individual augmenters for nested arguments
+                augmenter_type = args[0]
+                augmenter_args = args[1] if len(args) > 1 else {}
+                augmenter_args_mapped = self.map_arguments(
+                    augmenter_type, augmenter_args
+                )
+                augmenter_type_mapped = self.imgaug_to_albu.get(
+                    augmenter_type, augmenter_type
+                )
+                if augmenter_type_mapped == "Resize":
+                    return ImgaugLikeResize(**augmenter_args_mapped)
+                else:
+                    cls = getattr(A, augmenter_type_mapped)
+                    return cls(
+                        **{
+                            k: self.to_tuple_if_list(v)
+                            for k, v in augmenter_args_mapped.items()
+                        }
+                    )
+        elif isinstance(args, dict):
+            # Process individual transformation specified as dictionary
+            augmenter_type = args["type"]
+            augmenter_args = args.get("args", {})
+            augmenter_args_mapped = self.map_arguments(augmenter_type, augmenter_args)
+            augmenter_type_mapped = self.imgaug_to_albu.get(
+                augmenter_type, augmenter_type
+            )
+            if augmenter_type_mapped == "Resize":
+                return ImgaugLikeResize(**augmenter_args_mapped)
+            else:
+                cls = getattr(A, augmenter_type_mapped)
+                return cls(
+                    **{
+                        k: self.to_tuple_if_list(v)
+                        for k, v in augmenter_args_mapped.items()
+                    }
+                )
+        else:
+            raise RuntimeError("Unknown augmenter arg: " + str(args))
+
+    # Map arguments to expected format for each augmenter type
+    def map_arguments(self, augmenter_type, augmenter_args):
+        augmenter_args = augmenter_args.copy()  # Avoid modifying the original arguments
+        if augmenter_type == "Resize":
+            # Ensure size is a valid 2-element list or tuple
+            size = augmenter_args.get("size")
+            if size:
+                if not isinstance(size, (list, tuple)) or len(size) != 2:
+                    raise ValueError(
+                        f"'size' must be a list or tuple of two numbers, but got {size}"
+                    )
+                min_scale, max_scale = size
+                return {
+                    "scale_range": (min_scale, max_scale),
+                    "interpolation": 1,  # Linear interpolation
+                    "p": 1.0,
+                }
+            else:
+                return {"scale_range": (1.0, 1.0), "interpolation": 1, "p": 1.0}
+        elif augmenter_type == "Affine":
+            # Map rotation to a tuple and ensure p=1.0 to apply transformation
+            rotate = augmenter_args.get("rotate", 0)
+            if isinstance(rotate, list):
+                rotate = tuple(rotate)
+            elif isinstance(rotate, (int, float)):
+                rotate = (float(rotate), float(rotate))
+            augmenter_args["rotate"] = rotate
+            augmenter_args["p"] = augmenter_args.get("p", 1.0)
+            if augmenter_args.get("fit_output") is True:
+                if "border_mode" not in augmenter_args:
+                    import cv2
+
+                    augmenter_args["border_mode"] = cv2.BORDER_CONSTANT
+                if "fill" not in augmenter_args:
+                    augmenter_args["fill"] = 0
+            return augmenter_args
+        else:
+            # For other augmenters, ensure 'p' probability is specified
+            p = augmenter_args.get("p", 1.0)
+            augmenter_args["p"] = p
+            return augmenter_args
+
+    # Convert lists to tuples for Albumentations compatibility
+    def to_tuple_if_list(self, obj):
+        if isinstance(obj, list):
+            return tuple(obj)
+        return obj
+
+
+# Wrapper class for image and polygon transformations using Imgaug-style augmentation
+class IaaAugment:
+    def __init__(self, augmenter_args=None, **kwargs):
+        if augmenter_args is None:
+            # Default augmenters if none are specified
+            augmenter_args = [
+                {"type": "Fliplr", "args": {"p": 0.5}},
+                {"type": "Affine", "args": {"rotate": [-10, 10]}},
+                {"type": "Resize", "args": {"size": [0.5, 3]}},
+            ]
+        self.augmenter = AugmenterBuilder().build(augmenter_args)
+
+    # Apply the augmentations to image and polygon data
+    def __call__(self, data):
+        image = data["image"]
+
+        if self.augmenter:
+            # Flatten polygons to individual keypoints for transformation
+            keypoints = []
+            keypoints_lengths = []
+            for poly in data["polys"]:
+                keypoints.extend([tuple(point) for point in poly])
+                keypoints_lengths.append(len(poly))
+
+            # Apply the augmentation pipeline to image and keypoints
+            transformed = self.augmenter(image=image, keypoints=keypoints)
+            data["image"] = transformed["image"]
+
+            # Extract transformed keypoints and reconstruct polygon structures
+            transformed_keypoints = transformed["keypoints"]
+
+            # Reassemble polygons from transformed keypoints
+            new_polys = []
+            idx = 0
+            for length in keypoints_lengths:
+                new_poly = transformed_keypoints[idx : idx + length]
+                new_polys.append(np.array([kp[:2] for kp in new_poly]))
+                idx += length
+            data["polys"] = np.array(new_polys)
+        return data
diff --git a/ppocr/data/imaug/label_ops.py b/ppocr/data/imaug/label_ops.py
new file mode 100644
index 0000000..7988902
--- /dev/null
+++ b/ppocr/data/imaug/label_ops.py
@@ -0,0 +1,2205 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+import os
+from enum import Enum
+import copy
+import numpy as np
+import string
+from shapely.geometry import LineString, Point, Polygon
+import json
+import copy
+import random
+from random import sample
+from collections import defaultdict
+
+from ppocr.utils.logging import get_logger
+from ppocr.data.imaug.vqa.augment import order_by_tbyx
+
+
+class ClsLabelEncode(object):
+    def __init__(self, label_list, **kwargs):
+        self.label_list = label_list
+
+    def __call__(self, data):
+        label = data["label"]
+        if label not in self.label_list:
+            return None
+        label = self.label_list.index(label)
+        data["label"] = label
+        return data
+
+
+class DetLabelEncode(object):
+    def __init__(self, **kwargs):
+        pass
+
+    def __call__(self, data):
+        label = data["label"]
+        label = json.loads(label)
+        nBox = len(label)
+        boxes, txts, txt_tags = [], [], []
+        for bno in range(0, nBox):
+            box = label[bno]["points"]
+            txt = label[bno]["transcription"]
+            boxes.append(box)
+            txts.append(txt)
+            if txt in ["*", "###"]:
+                txt_tags.append(True)
+            else:
+                txt_tags.append(False)
+        if len(boxes) == 0:
+            return None
+        boxes = self.expand_points_num(boxes)
+        boxes = np.array(boxes, dtype=np.float32)
+        txt_tags = np.array(txt_tags, dtype=np.bool_)
+
+        # Mark polygons with NaN coordinates as ignore to avoid downstream errors
+        nan_mask = np.isnan(boxes).any(axis=(1, 2))
+        txt_tags[nan_mask] = True
+
+        data["polys"] = boxes
+        data["texts"] = txts
+        data["ignore_tags"] = txt_tags
+        return data
+
+    def order_points_clockwise(self, pts):
+        rect = np.zeros((4, 2), dtype="float32")
+        s = pts.sum(axis=1)
+        rect[0] = pts[np.argmin(s)]
+        rect[2] = pts[np.argmax(s)]
+        tmp = np.delete(pts, (np.argmin(s), np.argmax(s)), axis=0)
+        diff = np.diff(np.array(tmp), axis=1)
+        rect[1] = tmp[np.argmin(diff)]
+        rect[3] = tmp[np.argmax(diff)]
+        return rect
+
+    def expand_points_num(self, boxes):
+        max_points_num = 0
+        for box in boxes:
+            if len(box) > max_points_num:
+                max_points_num = len(box)
+        ex_boxes = []
+        for box in boxes:
+            ex_box = box + [box[-1]] * (max_points_num - len(box))
+            ex_boxes.append(ex_box)
+        return ex_boxes
+
+
+class BaseRecLabelEncode(object):
+    """Convert between text-label and text-index"""
+
+    def __init__(
+        self,
+        max_text_length,
+        character_dict_path=None,
+        use_space_char=False,
+        lower=False,
+    ):
+        self.max_text_len = max_text_length
+        self.beg_str = "sos"
+        self.end_str = "eos"
+        self.lower = lower
+
+        if character_dict_path is None:
+            logger = get_logger()
+            logger.warning(
+                "The character_dict_path is None, model can only recognize number and lower letters"
+            )
+            self.character_str = "0123456789abcdefghijklmnopqrstuvwxyz"
+            dict_character = list(self.character_str)
+            self.lower = True
+        else:
+            self.character_str = []
+            with open(character_dict_path, "rb") as fin:
+                lines = fin.readlines()
+                for line in lines:
+                    line = line.decode("utf-8").strip("\n").strip("\r\n")
+                    self.character_str.append(line)
+            if use_space_char:
+                self.character_str.append(" ")
+            dict_character = list(self.character_str)
+        dict_character = self.add_special_char(dict_character)
+        self.dict = {}
+        for i, char in enumerate(dict_character):
+            self.dict[char] = i
+        self.character = dict_character
+
+    def add_special_char(self, dict_character):
+        return dict_character
+
+    def encode(self, text):
+        """convert text-label into text-index.
+        input:
+            text: text labels of each image. [batch_size]
+
+        output:
+            text: concatenated text index for CTCLoss.
+                    [sum(text_lengths)] = [text_index_0 + text_index_1 + ... + text_index_(n - 1)]
+            length: length of each text. [batch_size]
+        """
+        if len(text) == 0 or len(text) > self.max_text_len:
+            return None
+        if self.lower:
+            text = text.lower()
+        text_list = []
+        for char in text:
+            if char not in self.dict:
+                # logger = get_logger()
+                # logger.warning('{} is not in dict'.format(char))
+                continue
+            text_list.append(self.dict[char])
+        if len(text_list) == 0:
+            return None
+        return text_list
+
+
+class CTCLabelEncode(BaseRecLabelEncode):
+    """Convert between text-label and text-index"""
+
+    def __init__(
+        self, max_text_length, character_dict_path=None, use_space_char=False, **kwargs
+    ):
+        super(CTCLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+
+    def __call__(self, data):
+        text = data["label"]
+        text = self.encode(text)
+        if text is None:
+            return None
+        data["length"] = np.array(len(text))
+        text = text + [0] * (self.max_text_len - len(text))
+        data["label"] = np.array(text)
+
+        label = [0] * len(self.character)
+        for x in text:
+            label[x] += 1
+        data["label_ace"] = np.array(label)
+        return data
+
+    def add_special_char(self, dict_character):
+        dict_character = ["blank"] + dict_character
+        return dict_character
+
+
+class E2ELabelEncodeTest(BaseRecLabelEncode):
+    def __init__(
+        self, max_text_length, character_dict_path=None, use_space_char=False, **kwargs
+    ):
+        super(E2ELabelEncodeTest, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+
+    def __call__(self, data):
+        import json
+
+        padnum = len(self.dict)
+        label = data["label"]
+        label = json.loads(label)
+        nBox = len(label)
+        boxes, txts, txt_tags = [], [], []
+        for bno in range(0, nBox):
+            box = label[bno]["points"]
+            txt = label[bno]["transcription"]
+            boxes.append(box)
+            txts.append(txt)
+            if txt in ["*", "###"]:
+                txt_tags.append(True)
+            else:
+                txt_tags.append(False)
+        boxes = np.array(boxes, dtype=np.float32)
+        txt_tags = np.array(txt_tags, dtype=np.bool_)
+        data["polys"] = boxes
+        data["ignore_tags"] = txt_tags
+        temp_texts = []
+        for text in txts:
+            text = text.lower()
+            text = self.encode(text)
+            if text is None:
+                return None
+            text = text + [padnum] * (self.max_text_len - len(text))  # use 36 to pad
+            temp_texts.append(text)
+        data["texts"] = np.array(temp_texts)
+        return data
+
+
+class E2ELabelEncodeTrain(object):
+    def __init__(self, **kwargs):
+        pass
+
+    def __call__(self, data):
+        import json
+
+        label = data["label"]
+        label = json.loads(label)
+        nBox = len(label)
+        boxes, txts, txt_tags = [], [], []
+        for bno in range(0, nBox):
+            box = label[bno]["points"]
+            txt = label[bno]["transcription"]
+            boxes.append(box)
+            txts.append(txt)
+            if txt in ["*", "###"]:
+                txt_tags.append(True)
+            else:
+                txt_tags.append(False)
+        boxes = np.array(boxes, dtype=np.float32)
+        txt_tags = np.array(txt_tags, dtype=np.bool_)
+
+        data["polys"] = boxes
+        data["texts"] = txts
+        data["ignore_tags"] = txt_tags
+        return data
+
+
+class KieLabelEncode(object):
+    def __init__(
+        self, character_dict_path, class_path, norm=10, directed=False, **kwargs
+    ):
+        super(KieLabelEncode, self).__init__()
+        self.dict = dict({"": 0})
+        self.label2classid_map = dict()
+        with open(character_dict_path, "r", encoding="utf-8") as fr:
+            idx = 1
+            for line in fr:
+                char = line.strip()
+                self.dict[char] = idx
+                idx += 1
+        with open(class_path, "r") as fin:
+            lines = fin.readlines()
+            for idx, line in enumerate(lines):
+                line = line.strip("\n")
+                self.label2classid_map[line] = idx
+        self.norm = norm
+        self.directed = directed
+
+    def compute_relation(self, boxes):
+        """Compute relation between every two boxes."""
+        x1s, y1s = boxes[:, 0:1], boxes[:, 1:2]
+        x2s, y2s = boxes[:, 4:5], boxes[:, 5:6]
+        ws, hs = x2s - x1s + 1, np.maximum(y2s - y1s + 1, 1)
+        dxs = (x1s[:, 0][None] - x1s) / self.norm
+        dys = (y1s[:, 0][None] - y1s) / self.norm
+        xhhs, xwhs = hs[:, 0][None] / hs, ws[:, 0][None] / hs
+        whs = ws / hs + np.zeros_like(xhhs)
+        relations = np.stack([dxs, dys, whs, xhhs, xwhs], -1)
+        bboxes = np.concatenate([x1s, y1s, x2s, y2s], -1).astype(np.float32)
+        return relations, bboxes
+
+    def pad_text_indices(self, text_inds):
+        """Pad text index to same length."""
+        max_len = 300
+        recoder_len = max([len(text_ind) for text_ind in text_inds])
+        padded_text_inds = -np.ones((len(text_inds), max_len), np.int32)
+        for idx, text_ind in enumerate(text_inds):
+            padded_text_inds[idx, : len(text_ind)] = np.array(text_ind)
+        return padded_text_inds, recoder_len
+
+    def list_to_numpy(self, ann_infos):
+        """Convert bboxes, relations, texts and labels to ndarray."""
+        boxes, text_inds = ann_infos["points"], ann_infos["text_inds"]
+        boxes = np.array(boxes, np.int32)
+        relations, bboxes = self.compute_relation(boxes)
+
+        labels = ann_infos.get("labels", None)
+        if labels is not None:
+            labels = np.array(labels, np.int32)
+            edges = ann_infos.get("edges", None)
+            if edges is not None:
+                labels = labels[:, None]
+                edges = np.array(edges)
+                edges = (edges[:, None] == edges[None, :]).astype(np.int32)
+                if self.directed:
+                    edges = (edges & labels == 1).astype(np.int32)
+                np.fill_diagonal(edges, -1)
+                labels = np.concatenate([labels, edges], -1)
+        padded_text_inds, recoder_len = self.pad_text_indices(text_inds)
+        max_num = 300
+        temp_bboxes = np.zeros([max_num, 4])
+        h, _ = bboxes.shape
+        temp_bboxes[:h, :] = bboxes
+
+        temp_relations = np.zeros([max_num, max_num, 5])
+        temp_relations[:h, :h, :] = relations
+
+        temp_padded_text_inds = np.zeros([max_num, max_num])
+        temp_padded_text_inds[:h, :] = padded_text_inds
+
+        temp_labels = np.zeros([max_num, max_num])
+        temp_labels[:h, : h + 1] = labels
+
+        tag = np.array([h, recoder_len])
+        return dict(
+            image=ann_infos["image"],
+            points=temp_bboxes,
+            relations=temp_relations,
+            texts=temp_padded_text_inds,
+            labels=temp_labels,
+            tag=tag,
+        )
+
+    def convert_canonical(self, points_x, points_y):
+        assert len(points_x) == 4
+        assert len(points_y) == 4
+
+        points = [Point(points_x[i], points_y[i]) for i in range(4)]
+
+        polygon = Polygon([(p.x, p.y) for p in points])
+        min_x, min_y, _, _ = polygon.bounds
+        points_to_lefttop = [
+            LineString([points[i], Point(min_x, min_y)]) for i in range(4)
+        ]
+        distances = np.array([line.length for line in points_to_lefttop])
+        sort_dist_idx = np.argsort(distances)
+        lefttop_idx = sort_dist_idx[0]
+
+        if lefttop_idx == 0:
+            point_orders = [0, 1, 2, 3]
+        elif lefttop_idx == 1:
+            point_orders = [1, 2, 3, 0]
+        elif lefttop_idx == 2:
+            point_orders = [2, 3, 0, 1]
+        else:
+            point_orders = [3, 0, 1, 2]
+
+        sorted_points_x = [points_x[i] for i in point_orders]
+        sorted_points_y = [points_y[j] for j in point_orders]
+
+        return sorted_points_x, sorted_points_y
+
+    def sort_vertex(self, points_x, points_y):
+        assert len(points_x) == 4
+        assert len(points_y) == 4
+
+        x = np.array(points_x)
+        y = np.array(points_y)
+        center_x = np.sum(x) * 0.25
+        center_y = np.sum(y) * 0.25
+
+        x_arr = np.array(x - center_x)
+        y_arr = np.array(y - center_y)
+
+        angle = np.arctan2(y_arr, x_arr) * 180.0 / np.pi
+        sort_idx = np.argsort(angle)
+
+        sorted_points_x, sorted_points_y = [], []
+        for i in range(4):
+            sorted_points_x.append(points_x[sort_idx[i]])
+            sorted_points_y.append(points_y[sort_idx[i]])
+
+        return self.convert_canonical(sorted_points_x, sorted_points_y)
+
+    def __call__(self, data):
+        import json
+
+        label = data["label"]
+        annotations = json.loads(label)
+        boxes, texts, text_inds, labels, edges = [], [], [], [], []
+        for ann in annotations:
+            box = ann["points"]
+            x_list = [box[i][0] for i in range(4)]
+            y_list = [box[i][1] for i in range(4)]
+            sorted_x_list, sorted_y_list = self.sort_vertex(x_list, y_list)
+            sorted_box = []
+            for x, y in zip(sorted_x_list, sorted_y_list):
+                sorted_box.append(x)
+                sorted_box.append(y)
+            boxes.append(sorted_box)
+            text = ann["transcription"]
+            texts.append(ann["transcription"])
+            text_ind = [self.dict[c] for c in text if c in self.dict]
+            text_inds.append(text_ind)
+            if "label" in ann.keys():
+                labels.append(self.label2classid_map[ann["label"]])
+            elif "key_cls" in ann.keys():
+                labels.append(ann["key_cls"])
+            else:
+                raise ValueError(
+                    "Cannot found 'key_cls' in ann.keys(), please check your training annotation."
+                )
+            edges.append(ann.get("edge", 0))
+        ann_infos = dict(
+            image=data["image"],
+            points=boxes,
+            texts=texts,
+            text_inds=text_inds,
+            edges=edges,
+            labels=labels,
+        )
+
+        return self.list_to_numpy(ann_infos)
+
+
+class AttnLabelEncode(BaseRecLabelEncode):
+    """Convert between text-label and text-index"""
+
+    def __init__(
+        self, max_text_length, character_dict_path=None, use_space_char=False, **kwargs
+    ):
+        super(AttnLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+
+    def add_special_char(self, dict_character):
+        self.beg_str = "sos"
+        self.end_str = "eos"
+        dict_character = [self.beg_str] + dict_character + [self.end_str]
+        return dict_character
+
+    def __call__(self, data):
+        text = data["label"]
+        text = self.encode(text)
+        if text is None:
+            return None
+        if len(text) >= self.max_text_len:
+            return None
+        data["length"] = np.array(len(text))
+        text = (
+            [0]
+            + text
+            + [len(self.character) - 1]
+            + [0] * (self.max_text_len - len(text) - 2)
+        )
+        data["label"] = np.array(text)
+        return data
+
+    def get_ignored_tokens(self):
+        beg_idx = self.get_beg_end_flag_idx("beg")
+        end_idx = self.get_beg_end_flag_idx("end")
+        return [beg_idx, end_idx]
+
+    def get_beg_end_flag_idx(self, beg_or_end):
+        if beg_or_end == "beg":
+            idx = np.array(self.dict[self.beg_str])
+        elif beg_or_end == "end":
+            idx = np.array(self.dict[self.end_str])
+        else:
+            assert False, "Unsupported type %s in get_beg_end_flag_idx" % beg_or_end
+        return idx
+
+
+class RFLLabelEncode(BaseRecLabelEncode):
+    """Convert between text-label and text-index"""
+
+    def __init__(
+        self, max_text_length, character_dict_path=None, use_space_char=False, **kwargs
+    ):
+        super(RFLLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+
+    def add_special_char(self, dict_character):
+        self.beg_str = "sos"
+        self.end_str = "eos"
+        dict_character = [self.beg_str] + dict_character + [self.end_str]
+        return dict_character
+
+    def encode_cnt(self, text):
+        cnt_label = [0.0] * len(self.character)
+        for char_ in text:
+            cnt_label[char_] += 1
+        return np.array(cnt_label)
+
+    def __call__(self, data):
+        text = data["label"]
+        text = self.encode(text)
+        if text is None:
+            return None
+        if len(text) >= self.max_text_len:
+            return None
+        cnt_label = self.encode_cnt(text)
+        data["length"] = np.array(len(text))
+        text = (
+            [0]
+            + text
+            + [len(self.character) - 1]
+            + [0] * (self.max_text_len - len(text) - 2)
+        )
+        if len(text) != self.max_text_len:
+            return None
+        data["label"] = np.array(text)
+        data["cnt_label"] = cnt_label
+        return data
+
+    def get_ignored_tokens(self):
+        beg_idx = self.get_beg_end_flag_idx("beg")
+        end_idx = self.get_beg_end_flag_idx("end")
+        return [beg_idx, end_idx]
+
+    def get_beg_end_flag_idx(self, beg_or_end):
+        if beg_or_end == "beg":
+            idx = np.array(self.dict[self.beg_str])
+        elif beg_or_end == "end":
+            idx = np.array(self.dict[self.end_str])
+        else:
+            assert False, "Unsupported type %s in get_beg_end_flag_idx" % beg_or_end
+        return idx
+
+
+class SEEDLabelEncode(BaseRecLabelEncode):
+    """Convert between text-label and text-index"""
+
+    def __init__(
+        self, max_text_length, character_dict_path=None, use_space_char=False, **kwargs
+    ):
+        super(SEEDLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+
+    def add_special_char(self, dict_character):
+        self.padding = "padding"
+        self.end_str = "eos"
+        self.unknown = "unknown"
+        dict_character = dict_character + [self.end_str, self.padding, self.unknown]
+        return dict_character
+
+    def __call__(self, data):
+        text = data["label"]
+        text = self.encode(text)
+        if text is None:
+            return None
+        if len(text) >= self.max_text_len:
+            return None
+        data["length"] = np.array(len(text)) + 1  # conclude eos
+        text = (
+            text
+            + [len(self.character) - 3]
+            + [len(self.character) - 2] * (self.max_text_len - len(text) - 1)
+        )
+        data["label"] = np.array(text)
+        return data
+
+
+class SRNLabelEncode(BaseRecLabelEncode):
+    """Convert between text-label and text-index"""
+
+    def __init__(
+        self,
+        max_text_length=25,
+        character_dict_path=None,
+        use_space_char=False,
+        **kwargs,
+    ):
+        super(SRNLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+
+    def add_special_char(self, dict_character):
+        dict_character = dict_character + [self.beg_str, self.end_str]
+        return dict_character
+
+    def __call__(self, data):
+        text = data["label"]
+        text = self.encode(text)
+        char_num = len(self.character)
+        if text is None:
+            return None
+        if len(text) > self.max_text_len:
+            return None
+        data["length"] = np.array(len(text))
+        text = text + [char_num - 1] * (self.max_text_len - len(text))
+        data["label"] = np.array(text)
+        return data
+
+    def get_ignored_tokens(self):
+        beg_idx = self.get_beg_end_flag_idx("beg")
+        end_idx = self.get_beg_end_flag_idx("end")
+        return [beg_idx, end_idx]
+
+    def get_beg_end_flag_idx(self, beg_or_end):
+        if beg_or_end == "beg":
+            idx = np.array(self.dict[self.beg_str])
+        elif beg_or_end == "end":
+            idx = np.array(self.dict[self.end_str])
+        else:
+            assert False, "Unsupported type %s in get_beg_end_flag_idx" % beg_or_end
+        return idx
+
+
+class TableLabelEncode(AttnLabelEncode):
+    """Convert between text-label and text-index"""
+
+    def __init__(
+        self,
+        max_text_length,
+        character_dict_path,
+        replace_empty_cell_token=False,
+        merge_no_span_structure=False,
+        learn_empty_box=False,
+        loc_reg_num=4,
+        **kwargs,
+    ):
+        self.max_text_len = max_text_length
+        self.lower = False
+        self.learn_empty_box = learn_empty_box
+        self.merge_no_span_structure = merge_no_span_structure
+        self.replace_empty_cell_token = replace_empty_cell_token
+
+        dict_character = []
+        with open(character_dict_path, "rb") as fin:
+            lines = fin.readlines()
+            for line in lines:
+                line = line.decode("utf-8").strip("\n").strip("\r\n")
+                dict_character.append(line)
+
+        if self.merge_no_span_structure:
+            if "<td></td>" not in dict_character:
+                dict_character.append("<td></td>")
+            if "<td>" in dict_character:
+                dict_character.remove("<td>")
+
+        dict_character = self.add_special_char(dict_character)
+        self.dict = {}
+        for i, char in enumerate(dict_character):
+            self.dict[char] = i
+        self.idx2char = {v: k for k, v in self.dict.items()}
+
+        self.character = dict_character
+        self.loc_reg_num = loc_reg_num
+        self.pad_idx = self.dict[self.beg_str]
+        self.start_idx = self.dict[self.beg_str]
+        self.end_idx = self.dict[self.end_str]
+
+        self.td_token = ["<td>", "<td", "<eb></eb>", "<td></td>"]
+        self.empty_bbox_token_dict = {
+            "[]": "<eb></eb>",
+            "[' ']": "<eb1></eb1>",
+            "['<b>', ' ', '</b>']": "<eb2></eb2>",
+            "['\\u2028', '\\u2028']": "<eb3></eb3>",
+            "['<sup>', ' ', '</sup>']": "<eb4></eb4>",
+            "['<b>', '</b>']": "<eb5></eb5>",
+            "['<i>', ' ', '</i>']": "<eb6></eb6>",
+            "['<b>', '<i>', '</i>', '</b>']": "<eb7></eb7>",
+            "['<b>', '<i>', ' ', '</i>', '</b>']": "<eb8></eb8>",
+            "['<i>', '</i>']": "<eb9></eb9>",
+            "['<b>', ' ', '\\u2028', ' ', '\\u2028', ' ', '</b>']": "<eb10></eb10>",
+        }
+
+    @property
+    def _max_text_len(self):
+        return self.max_text_len + 2
+
+    def __call__(self, data):
+        cells = data["cells"]
+        structure = data["structure"]
+        if self.merge_no_span_structure:
+            structure = self._merge_no_span_structure(structure)
+        if self.replace_empty_cell_token:
+            structure = self._replace_empty_cell_token(structure, cells)
+        # remove empty token and add " " to span token
+        new_structure = []
+        for token in structure:
+            if token != "":
+                if "span" in token and token[0] != " ":
+                    token = " " + token
+                new_structure.append(token)
+        # encode structure
+        structure = self.encode(new_structure)
+        if structure is None:
+            return None
+        data["length"] = len(structure)
+        structure = [self.start_idx] + structure + [self.end_idx]  # add sos abd eos
+        structure = structure + [self.pad_idx] * (
+            self._max_text_len - len(structure)
+        )  # pad
+        structure = np.array(structure)
+        data["structure"] = structure
+
+        if len(structure) > self._max_text_len:
+            return None
+
+        # encode box
+        bboxes = np.zeros((self._max_text_len, self.loc_reg_num), dtype=np.float32)
+        bbox_masks = np.zeros((self._max_text_len, 1), dtype=np.float32)
+
+        bbox_idx = 0
+
+        for i, token in enumerate(structure):
+            if self.idx2char[token] in self.td_token:
+                if "bbox" in cells[bbox_idx] and len(cells[bbox_idx]["tokens"]) > 0:
+                    bbox = cells[bbox_idx]["bbox"].copy()
+                    bbox = np.array(bbox, dtype=np.float32).reshape(-1)
+                    bboxes[i] = bbox
+                    bbox_masks[i] = 1.0
+                if self.learn_empty_box:
+                    bbox_masks[i] = 1.0
+                bbox_idx += 1
+        data["bboxes"] = bboxes
+        data["bbox_masks"] = bbox_masks
+        return data
+
+    def _merge_no_span_structure(self, structure):
+        """
+        This code is refer from:
+        https://github.com/JiaquanYe/TableMASTER-mmocr/blob/master/table_recognition/data_preprocess.py
+        """
+        new_structure = []
+        i = 0
+        while i < len(structure):
+            token = structure[i]
+            if token == "<td>":
+                token = "<td></td>"
+                i += 1
+            new_structure.append(token)
+            i += 1
+        return new_structure
+
+    def _replace_empty_cell_token(self, token_list, cells):
+        """
+        This fun code is refer from:
+        https://github.com/JiaquanYe/TableMASTER-mmocr/blob/master/table_recognition/data_preprocess.py
+        """
+
+        bbox_idx = 0
+        add_empty_bbox_token_list = []
+        for token in token_list:
+            if token in ["<td></td>", "<td", "<td>"]:
+                if "bbox" not in cells[bbox_idx].keys():
+                    content = str(cells[bbox_idx]["tokens"])
+                    token = self.empty_bbox_token_dict[content]
+                add_empty_bbox_token_list.append(token)
+                bbox_idx += 1
+            else:
+                add_empty_bbox_token_list.append(token)
+        return add_empty_bbox_token_list
+
+
+class TableMasterLabelEncode(TableLabelEncode):
+    """Convert between text-label and text-index"""
+
+    def __init__(
+        self,
+        max_text_length,
+        character_dict_path,
+        replace_empty_cell_token=False,
+        merge_no_span_structure=False,
+        learn_empty_box=False,
+        loc_reg_num=4,
+        **kwargs,
+    ):
+        super(TableMasterLabelEncode, self).__init__(
+            max_text_length,
+            character_dict_path,
+            replace_empty_cell_token,
+            merge_no_span_structure,
+            learn_empty_box,
+            loc_reg_num,
+            **kwargs,
+        )
+        self.pad_idx = self.dict[self.pad_str]
+        self.unknown_idx = self.dict[self.unknown_str]
+
+    @property
+    def _max_text_len(self):
+        return self.max_text_len
+
+    def add_special_char(self, dict_character):
+        self.beg_str = "<SOS>"
+        self.end_str = "<EOS>"
+        self.unknown_str = "<UKN>"
+        self.pad_str = "<PAD>"
+        dict_character = dict_character
+        dict_character = dict_character + [
+            self.unknown_str,
+            self.beg_str,
+            self.end_str,
+            self.pad_str,
+        ]
+        return dict_character
+
+
+class TableBoxEncode(object):
+    def __init__(self, in_box_format="xyxy", out_box_format="xyxy", **kwargs):
+        assert out_box_format in ["xywh", "xyxy", "xyxyxyxy"]
+        self.in_box_format = in_box_format
+        self.out_box_format = out_box_format
+
+    def __call__(self, data):
+        img_height, img_width = data["image"].shape[:2]
+        bboxes = data["bboxes"]
+        if self.in_box_format != self.out_box_format:
+            if self.out_box_format == "xywh":
+                if self.in_box_format == "xyxyxyxy":
+                    bboxes = self.xyxyxyxy2xywh(bboxes)
+                elif self.in_box_format == "xyxy":
+                    bboxes = self.xyxy2xywh(bboxes)
+
+        bboxes[:, 0::2] /= img_width
+        bboxes[:, 1::2] /= img_height
+        data["bboxes"] = bboxes
+        return data
+
+    def xyxyxyxy2xywh(self, boxes):
+        new_bboxes = np.zeros([len(boxes), 4])
+        new_bboxes[:, 0] = boxes[:, 0::2].min()  # x1
+        new_bboxes[:, 1] = boxes[:, 1::2].min()  # y1
+        new_bboxes[:, 2] = boxes[:, 0::2].max() - new_bboxes[:, 0]  # w
+        new_bboxes[:, 3] = boxes[:, 1::2].max() - new_bboxes[:, 1]  # h
+        return new_bboxes
+
+    def xyxy2xywh(self, bboxes):
+        new_bboxes = np.empty_like(bboxes)
+        new_bboxes[:, 0] = (bboxes[:, 0] + bboxes[:, 2]) / 2  # x center
+        new_bboxes[:, 1] = (bboxes[:, 1] + bboxes[:, 3]) / 2  # y center
+        new_bboxes[:, 2] = bboxes[:, 2] - bboxes[:, 0]  # width
+        new_bboxes[:, 3] = bboxes[:, 3] - bboxes[:, 1]  # height
+        return new_bboxes
+
+
+class SARLabelEncode(BaseRecLabelEncode):
+    """Convert between text-label and text-index"""
+
+    def __init__(
+        self, max_text_length, character_dict_path=None, use_space_char=False, **kwargs
+    ):
+        super(SARLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+
+    def add_special_char(self, dict_character):
+        beg_end_str = "<BOS/EOS>"
+        unknown_str = "<UKN>"
+        padding_str = "<PAD>"
+        dict_character = dict_character + [unknown_str]
+        self.unknown_idx = len(dict_character) - 1
+        dict_character = dict_character + [beg_end_str]
+        self.start_idx = len(dict_character) - 1
+        self.end_idx = len(dict_character) - 1
+        dict_character = dict_character + [padding_str]
+        self.padding_idx = len(dict_character) - 1
+
+        return dict_character
+
+    def __call__(self, data):
+        text = data["label"]
+        text = self.encode(text)
+        if text is None:
+            return None
+        if len(text) >= self.max_text_len - 1:
+            return None
+        data["length"] = np.array(len(text))
+        target = [self.start_idx] + text + [self.end_idx]
+        padded_text = [self.padding_idx for _ in range(self.max_text_len)]
+
+        padded_text[: len(target)] = target
+        data["label"] = np.array(padded_text)
+        return data
+
+    def get_ignored_tokens(self):
+        return [self.padding_idx]
+
+
+class SATRNLabelEncode(BaseRecLabelEncode):
+    """Convert between text-label and text-index"""
+
+    def __init__(
+        self,
+        max_text_length,
+        character_dict_path=None,
+        use_space_char=False,
+        lower=False,
+        **kwargs,
+    ):
+        super(SATRNLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+        self.lower = lower
+
+    def add_special_char(self, dict_character):
+        beg_end_str = "<BOS/EOS>"
+        unknown_str = "<UKN>"
+        padding_str = "<PAD>"
+        dict_character = dict_character + [unknown_str]
+        self.unknown_idx = len(dict_character) - 1
+        dict_character = dict_character + [beg_end_str]
+        self.start_idx = len(dict_character) - 1
+        self.end_idx = len(dict_character) - 1
+        dict_character = dict_character + [padding_str]
+        self.padding_idx = len(dict_character) - 1
+
+        return dict_character
+
+    def encode(self, text):
+        if self.lower:
+            text = text.lower()
+        text_list = []
+        for char in text:
+            text_list.append(self.dict.get(char, self.unknown_idx))
+        if len(text_list) == 0:
+            return None
+        return text_list
+
+    def __call__(self, data):
+        text = data["label"]
+        text = self.encode(text)
+        if text is None:
+            return None
+        data["length"] = np.array(len(text))
+        target = [self.start_idx] + text + [self.end_idx]
+        padded_text = [self.padding_idx for _ in range(self.max_text_len)]
+        if len(target) > self.max_text_len:
+            padded_text = target[: self.max_text_len]
+        else:
+            padded_text[: len(target)] = target
+        data["label"] = np.array(padded_text)
+        return data
+
+    def get_ignored_tokens(self):
+        return [self.padding_idx]
+
+
+class PRENLabelEncode(BaseRecLabelEncode):
+    def __init__(
+        self, max_text_length, character_dict_path, use_space_char=False, **kwargs
+    ):
+        super(PRENLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+
+    def add_special_char(self, dict_character):
+        padding_str = "<PAD>"  # 0
+        end_str = "<EOS>"  # 1
+        unknown_str = "<UNK>"  # 2
+
+        dict_character = [padding_str, end_str, unknown_str] + dict_character
+        self.padding_idx = 0
+        self.end_idx = 1
+        self.unknown_idx = 2
+
+        return dict_character
+
+    def encode(self, text):
+        if len(text) == 0 or len(text) >= self.max_text_len:
+            return None
+        if self.lower:
+            text = text.lower()
+        text_list = []
+        for char in text:
+            if char not in self.dict:
+                text_list.append(self.unknown_idx)
+            else:
+                text_list.append(self.dict[char])
+        text_list.append(self.end_idx)
+        if len(text_list) < self.max_text_len:
+            text_list += [self.padding_idx] * (self.max_text_len - len(text_list))
+        return text_list
+
+    def __call__(self, data):
+        text = data["label"]
+        encoded_text = self.encode(text)
+        if encoded_text is None:
+            return None
+        data["label"] = np.array(encoded_text)
+        return data
+
+
+class VQATokenLabelEncode(object):
+    """
+    Label encode for NLP VQA methods
+    """
+
+    def __init__(
+        self,
+        class_path,
+        contains_re=False,
+        add_special_ids=False,
+        algorithm="LayoutXLM",
+        use_textline_bbox_info=True,
+        order_method=None,
+        infer_mode=False,
+        ocr_engine=None,
+        **kwargs,
+    ):
+        super(VQATokenLabelEncode, self).__init__()
+        from paddlenlp.transformers import (
+            LayoutXLMTokenizer,
+            LayoutLMTokenizer,
+            LayoutLMv2Tokenizer,
+        )
+        from ppocr.utils.utility import load_vqa_bio_label_maps
+
+        tokenizer_dict = {
+            "LayoutXLM": {
+                "class": LayoutXLMTokenizer,
+                "pretrained_model": "layoutxlm-base-uncased",
+            },
+            "LayoutLM": {
+                "class": LayoutLMTokenizer,
+                "pretrained_model": "layoutlm-base-uncased",
+            },
+            "LayoutLMv2": {
+                "class": LayoutLMv2Tokenizer,
+                "pretrained_model": "layoutlmv2-base-uncased",
+            },
+        }
+        self.contains_re = contains_re
+        tokenizer_config = tokenizer_dict[algorithm]
+        self.tokenizer = tokenizer_config["class"].from_pretrained(
+            tokenizer_config["pretrained_model"]
+        )
+        self.label2id_map, id2label_map = load_vqa_bio_label_maps(class_path)
+        self.add_special_ids = add_special_ids
+        self.infer_mode = infer_mode
+        self.ocr_engine = ocr_engine
+        self.use_textline_bbox_info = use_textline_bbox_info
+        self.order_method = order_method
+        assert self.order_method in [None, "tb-yx"]
+
+    def split_bbox(self, bbox, text, tokenizer):
+        words = text.split()
+        token_bboxes = []
+        curr_word_idx = 0
+        x1, y1, x2, y2 = bbox
+        unit_w = (x2 - x1) / len(text)
+        for idx, word in enumerate(words):
+            curr_w = len(word) * unit_w
+            word_bbox = [x1, y1, x1 + curr_w, y2]
+            token_bboxes.extend([word_bbox] * len(tokenizer.tokenize(word)))
+            x1 += (len(word) + 1) * unit_w
+        return token_bboxes
+
+    def filter_empty_contents(self, ocr_info):
+        """
+        find out the empty texts and remove the links
+        """
+        new_ocr_info = []
+        empty_index = []
+        for idx, info in enumerate(ocr_info):
+            if len(info["transcription"]) > 0:
+                new_ocr_info.append(copy.deepcopy(info))
+            else:
+                empty_index.append(info["id"])
+
+        for idx, info in enumerate(new_ocr_info):
+            new_link = []
+            for link in info["linking"]:
+                if link[0] in empty_index or link[1] in empty_index:
+                    continue
+                new_link.append(link)
+            new_ocr_info[idx]["linking"] = new_link
+        return new_ocr_info
+
+    def __call__(self, data):
+        # load bbox and label info
+        ocr_info = self._load_ocr_info(data)
+
+        for idx in range(len(ocr_info)):
+            if "bbox" not in ocr_info[idx]:
+                ocr_info[idx]["bbox"] = self.trans_poly_to_bbox(ocr_info[idx]["points"])
+
+        if self.order_method == "tb-yx":
+            ocr_info = order_by_tbyx(ocr_info)
+
+        # for re
+        train_re = self.contains_re and not self.infer_mode
+        if train_re:
+            ocr_info = self.filter_empty_contents(ocr_info)
+
+        height, width, _ = data["image"].shape
+
+        words_list = []
+        bbox_list = []
+        input_ids_list = []
+        token_type_ids_list = []
+        segment_offset_id = []
+        gt_label_list = []
+
+        entities = []
+
+        if train_re:
+            relations = []
+            id2label = {}
+            entity_id_to_index_map = {}
+            empty_entity = set()
+
+        data["ocr_info"] = copy.deepcopy(ocr_info)
+
+        for info in ocr_info:
+            text = info["transcription"]
+            if len(text) <= 0:
+                continue
+            if train_re:
+                # for re
+                if len(text) == 0:
+                    empty_entity.add(info["id"])
+                    continue
+                id2label[info["id"]] = info["label"]
+                relations.extend([tuple(sorted(l)) for l in info["linking"]])
+            # smooth_box
+            info["bbox"] = self.trans_poly_to_bbox(info["points"])
+
+            encode_res = self.tokenizer.encode(
+                text,
+                pad_to_max_seq_len=False,
+                return_attention_mask=True,
+                return_token_type_ids=True,
+            )
+
+            if not self.add_special_ids:
+                # TODO: use tok.all_special_ids to remove
+                encode_res["input_ids"] = encode_res["input_ids"][1:-1]
+                encode_res["token_type_ids"] = encode_res["token_type_ids"][1:-1]
+                encode_res["attention_mask"] = encode_res["attention_mask"][1:-1]
+
+            if self.use_textline_bbox_info:
+                bbox = [info["bbox"]] * len(encode_res["input_ids"])
+            else:
+                bbox = self.split_bbox(
+                    info["bbox"], info["transcription"], self.tokenizer
+                )
+            if len(bbox) <= 0:
+                continue
+            bbox = self._smooth_box(bbox, height, width)
+            if self.add_special_ids:
+                bbox.insert(0, [0, 0, 0, 0])
+                bbox.append([0, 0, 0, 0])
+
+            # parse label
+            if not self.infer_mode:
+                label = info["label"]
+                gt_label = self._parse_label(label, encode_res)
+
+            # construct entities for re
+            if train_re:
+                if gt_label[0] != self.label2id_map["O"]:
+                    entity_id_to_index_map[info["id"]] = len(entities)
+                    label = label.upper()
+                    entities.append(
+                        {
+                            "start": len(input_ids_list),
+                            "end": len(input_ids_list) + len(encode_res["input_ids"]),
+                            "label": label.upper(),
+                        }
+                    )
+            else:
+                entities.append(
+                    {
+                        "start": len(input_ids_list),
+                        "end": len(input_ids_list) + len(encode_res["input_ids"]),
+                        "label": "O",
+                    }
+                )
+            input_ids_list.extend(encode_res["input_ids"])
+            token_type_ids_list.extend(encode_res["token_type_ids"])
+            bbox_list.extend(bbox)
+            words_list.append(text)
+            segment_offset_id.append(len(input_ids_list))
+            if not self.infer_mode:
+                gt_label_list.extend(gt_label)
+
+        data["input_ids"] = input_ids_list
+        data["token_type_ids"] = token_type_ids_list
+        data["bbox"] = bbox_list
+        data["attention_mask"] = [1] * len(input_ids_list)
+        data["labels"] = gt_label_list
+        data["segment_offset_id"] = segment_offset_id
+        data["tokenizer_params"] = dict(
+            padding_side=self.tokenizer.padding_side,
+            pad_token_type_id=self.tokenizer.pad_token_type_id,
+            pad_token_id=self.tokenizer.pad_token_id,
+        )
+        data["entities"] = entities
+
+        if train_re:
+            data["relations"] = relations
+            data["id2label"] = id2label
+            data["empty_entity"] = empty_entity
+            data["entity_id_to_index_map"] = entity_id_to_index_map
+        return data
+
+    def trans_poly_to_bbox(self, poly):
+        x1 = int(np.min([p[0] for p in poly]))
+        x2 = int(np.max([p[0] for p in poly]))
+        y1 = int(np.min([p[1] for p in poly]))
+        y2 = int(np.max([p[1] for p in poly]))
+        return [x1, y1, x2, y2]
+
+    def _load_ocr_info(self, data):
+        if self.infer_mode:
+            ocr_result = self.ocr_engine.ocr(data["image"], cls=False)[0]
+            ocr_info = []
+            for res in ocr_result:
+                ocr_info.append(
+                    {
+                        "transcription": res[1][0],
+                        "bbox": self.trans_poly_to_bbox(res[0]),
+                        "points": res[0],
+                    }
+                )
+            return ocr_info
+        else:
+            info = data["label"]
+            # read text info
+            info_dict = json.loads(info)
+            return info_dict
+
+    def _smooth_box(self, bboxes, height, width):
+        bboxes = np.array(bboxes)
+        bboxes[:, 0] = bboxes[:, 0] * 1000 / width
+        bboxes[:, 2] = bboxes[:, 2] * 1000 / width
+        bboxes[:, 1] = bboxes[:, 1] * 1000 / height
+        bboxes[:, 3] = bboxes[:, 3] * 1000 / height
+        bboxes = bboxes.astype("int64").tolist()
+        return bboxes
+
+    def _parse_label(self, label, encode_res):
+        gt_label = []
+        if label.lower() in ["other", "others", "ignore"]:
+            gt_label.extend([0] * len(encode_res["input_ids"]))
+        else:
+            gt_label.append(self.label2id_map[("b-" + label).upper()])
+            gt_label.extend(
+                [self.label2id_map[("i-" + label).upper()]]
+                * (len(encode_res["input_ids"]) - 1)
+            )
+        return gt_label
+
+
+class MultiLabelEncode(BaseRecLabelEncode):
+    def __init__(
+        self,
+        max_text_length,
+        character_dict_path=None,
+        use_space_char=False,
+        gtc_encode=None,
+        **kwargs,
+    ):
+        super(MultiLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+
+        self.ctc_encode = CTCLabelEncode(
+            max_text_length, character_dict_path, use_space_char, **kwargs
+        )
+        self.gtc_encode_type = gtc_encode
+        if gtc_encode is None:
+            self.gtc_encode = SARLabelEncode(
+                max_text_length, character_dict_path, use_space_char, **kwargs
+            )
+        else:
+            self.gtc_encode = eval(gtc_encode)(
+                max_text_length, character_dict_path, use_space_char, **kwargs
+            )
+
+    def __call__(self, data):
+        data_ctc = copy.deepcopy(data)
+        data_gtc = copy.deepcopy(data)
+        data_out = dict()
+        data_out["img_path"] = data.get("img_path", None)
+        data_out["image"] = data["image"]
+        ctc = self.ctc_encode.__call__(data_ctc)
+        gtc = self.gtc_encode.__call__(data_gtc)
+        if ctc is None or gtc is None:
+            return None
+        data_out["label_ctc"] = ctc["label"]
+        if self.gtc_encode_type is not None:
+            data_out["label_gtc"] = gtc["label"]
+        else:
+            data_out["label_sar"] = gtc["label"]
+        data_out["length"] = ctc["length"]
+        return data_out
+
+
+class NRTRLabelEncode(BaseRecLabelEncode):
+    """Convert between text-label and text-index"""
+
+    def __init__(
+        self, max_text_length, character_dict_path=None, use_space_char=False, **kwargs
+    ):
+        super(NRTRLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+
+    def __call__(self, data):
+        text = data["label"]
+        text = self.encode(text)
+        if text is None:
+            return None
+        if len(text) >= self.max_text_len - 1:
+            return None
+        data["length"] = np.array(len(text))
+        text.insert(0, 2)
+        text.append(3)
+        text = text + [0] * (self.max_text_len - len(text))
+        data["label"] = np.array(text)
+        return data
+
+    def add_special_char(self, dict_character):
+        dict_character = ["blank", "<unk>", "<s>", "</s>"] + dict_character
+        return dict_character
+
+
+class ParseQLabelEncode(BaseRecLabelEncode):
+    """Convert between text-label and text-index"""
+
+    BOS = "[B]"
+    EOS = "[E]"
+    PAD = "[P]"
+
+    def __init__(
+        self, max_text_length, character_dict_path=None, use_space_char=False, **kwargs
+    ):
+        super(ParseQLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+
+    def __call__(self, data):
+        text = data["label"]
+        text = self.encode(text)
+        if text is None:
+            return None
+        if len(text) >= self.max_text_len - 2:
+            return None
+        data["length"] = np.array(len(text))
+        text = [self.dict[self.BOS]] + text + [self.dict[self.EOS]]
+        text = text + [self.dict[self.PAD]] * (self.max_text_len - len(text))
+        data["label"] = np.array(text)
+        return data
+
+    def add_special_char(self, dict_character):
+        dict_character = [self.EOS] + dict_character + [self.BOS, self.PAD]
+        return dict_character
+
+
+class ViTSTRLabelEncode(BaseRecLabelEncode):
+    """Convert between text-label and text-index"""
+
+    def __init__(
+        self,
+        max_text_length,
+        character_dict_path=None,
+        use_space_char=False,
+        ignore_index=0,
+        **kwargs,
+    ):
+        super(ViTSTRLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+        self.ignore_index = ignore_index
+
+    def __call__(self, data):
+        text = data["label"]
+        text = self.encode(text)
+        if text is None:
+            return None
+        if len(text) >= self.max_text_len:
+            return None
+        data["length"] = np.array(len(text))
+        text.insert(0, self.ignore_index)
+        text.append(1)
+        text = text + [self.ignore_index] * (self.max_text_len + 2 - len(text))
+        data["label"] = np.array(text)
+        return data
+
+    def add_special_char(self, dict_character):
+        dict_character = ["<s>", "</s>"] + dict_character
+        return dict_character
+
+
+class ABINetLabelEncode(BaseRecLabelEncode):
+    """Convert between text-label and text-index"""
+
+    def __init__(
+        self,
+        max_text_length,
+        character_dict_path=None,
+        use_space_char=False,
+        ignore_index=100,
+        **kwargs,
+    ):
+        super(ABINetLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+        self.ignore_index = ignore_index
+
+    def __call__(self, data):
+        text = data["label"]
+        text = self.encode(text)
+        if text is None:
+            return None
+        if len(text) >= self.max_text_len:
+            return None
+        data["length"] = np.array(len(text))
+        text.append(0)
+        text = text + [self.ignore_index] * (self.max_text_len + 1 - len(text))
+        data["label"] = np.array(text)
+        return data
+
+    def add_special_char(self, dict_character):
+        dict_character = ["</s>"] + dict_character
+        return dict_character
+
+
+class SRLabelEncode(BaseRecLabelEncode):
+    def __init__(
+        self, max_text_length, character_dict_path=None, use_space_char=False, **kwargs
+    ):
+        super(SRLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+        self.dic = {}
+        with open(character_dict_path, "r") as fin:
+            for line in fin.readlines():
+                line = line.strip()
+                character, sequence = line.split()
+                self.dic[character] = sequence
+        english_stroke_alphabet = "0123456789"
+        self.english_stroke_dict = {}
+        for index in range(len(english_stroke_alphabet)):
+            self.english_stroke_dict[english_stroke_alphabet[index]] = index
+
+    def encode(self, label):
+        stroke_sequence = ""
+        for character in label:
+            if character not in self.dic:
+                continue
+            else:
+                stroke_sequence += self.dic[character]
+        stroke_sequence += "0"
+        label = stroke_sequence
+
+        length = len(label)
+
+        input_tensor = np.zeros(self.max_text_len).astype("int64")
+        for j in range(length - 1):
+            input_tensor[j + 1] = self.english_stroke_dict[label[j]]
+
+        return length, input_tensor
+
+    def __call__(self, data):
+        text = data["label"]
+        length, input_tensor = self.encode(text)
+
+        data["length"] = length
+        data["input_tensor"] = input_tensor
+        if text is None:
+            return None
+        return data
+
+
+class SPINLabelEncode(AttnLabelEncode):
+    """Convert between text-label and text-index"""
+
+    def __init__(
+        self,
+        max_text_length,
+        character_dict_path=None,
+        use_space_char=False,
+        lower=True,
+        **kwargs,
+    ):
+        super(SPINLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+        self.lower = lower
+
+    def add_special_char(self, dict_character):
+        self.beg_str = "sos"
+        self.end_str = "eos"
+        dict_character = [self.beg_str] + [self.end_str] + dict_character
+        return dict_character
+
+    def __call__(self, data):
+        text = data["label"]
+        text = self.encode(text)
+        if text is None:
+            return None
+        if len(text) > self.max_text_len:
+            return None
+        data["length"] = np.array(len(text))
+        target = [0] + text + [1]
+        padded_text = [0 for _ in range(self.max_text_len + 2)]
+
+        padded_text[: len(target)] = target
+        data["label"] = np.array(padded_text)
+        return data
+
+
+class VLLabelEncode(BaseRecLabelEncode):
+    """Convert between text-label and text-index"""
+
+    def __init__(
+        self, max_text_length, character_dict_path=None, use_space_char=False, **kwargs
+    ):
+        super(VLLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+        self.dict = {}
+        for i, char in enumerate(self.character):
+            self.dict[char] = i
+
+    def __call__(self, data):
+        text = data["label"]  # original string
+        # generate occluded text
+        len_str = len(text)
+        if len_str <= 0:
+            return None
+        change_num = 1
+        order = list(range(len_str))
+        change_id = sample(order, change_num)[0]
+        label_sub = text[change_id]
+        if change_id == (len_str - 1):
+            label_res = text[:change_id]
+        elif change_id == 0:
+            label_res = text[1:]
+        else:
+            label_res = text[:change_id] + text[change_id + 1 :]
+
+        data["label_res"] = label_res  # remaining string
+        data["label_sub"] = label_sub  # occluded character
+        data["label_id"] = change_id  # character index
+        # encode label
+        text = self.encode(text)
+        if text is None:
+            return None
+        text = [i + 1 for i in text]
+        data["length"] = np.array(len(text))
+        text = text + [0] * (self.max_text_len - len(text))
+        data["label"] = np.array(text)
+        label_res = self.encode(label_res)
+        label_sub = self.encode(label_sub)
+        if label_res is None:
+            label_res = []
+        else:
+            label_res = [i + 1 for i in label_res]
+        if label_sub is None:
+            label_sub = []
+        else:
+            label_sub = [i + 1 for i in label_sub]
+        data["length_res"] = np.array(len(label_res))
+        data["length_sub"] = np.array(len(label_sub))
+        label_res = label_res + [0] * (self.max_text_len - len(label_res))
+        label_sub = label_sub + [0] * (self.max_text_len - len(label_sub))
+        data["label_res"] = np.array(label_res)
+        data["label_sub"] = np.array(label_sub)
+        return data
+
+
+class CTLabelEncode(object):
+    def __init__(self, **kwargs):
+        pass
+
+    def __call__(self, data):
+        label = data["label"]
+
+        label = json.loads(label)
+        nBox = len(label)
+        boxes, txts = [], []
+        for bno in range(0, nBox):
+            box = label[bno]["points"]
+            box = np.array(box)
+
+            boxes.append(box)
+            txt = label[bno]["transcription"]
+            txts.append(txt)
+
+        if len(boxes) == 0:
+            return None
+
+        data["polys"] = boxes
+        data["texts"] = txts
+        return data
+
+
+class CANLabelEncode(BaseRecLabelEncode):
+    def __init__(
+        self,
+        character_dict_path,
+        max_text_length=100,
+        use_space_char=False,
+        lower=True,
+        **kwargs,
+    ):
+        super(CANLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char, lower
+        )
+
+    def encode(self, text_seq):
+        text_seq_encoded = []
+        for text in text_seq:
+            if text not in self.character:
+                continue
+            text_seq_encoded.append(self.dict.get(text))
+        if len(text_seq_encoded) == 0:
+            return None
+        return text_seq_encoded
+
+    def __call__(self, data):
+        label = data["label"]
+        if isinstance(label, str):
+            label = label.strip().split()
+        label.append(self.end_str)
+        data["label"] = self.encode(label)
+        return data
+
+
+class CPPDLabelEncode(BaseRecLabelEncode):
+    """Convert between text-label and text-index"""
+
+    def __init__(
+        self,
+        max_text_length,
+        character_dict_path=None,
+        use_space_char=False,
+        ch=False,
+        ignore_index=100,
+        **kwargs,
+    ):
+        super(CPPDLabelEncode, self).__init__(
+            max_text_length, character_dict_path, use_space_char
+        )
+        self.ch = ch
+        self.ignore_index = ignore_index
+
+    def __call__(self, data):
+        text = data["label"]
+        if self.ch:
+            text, text_node_index, text_node_num = self.encodech(text)
+            if text is None:
+                return None
+            if len(text) > self.max_text_len:
+                return None
+            data["length"] = np.array(len(text))
+
+            text_pos_node = [1] * (len(text) + 1) + [0] * (
+                self.max_text_len - len(text)
+            )
+
+            text.append(0)  # eos
+            text = text + [self.ignore_index] * (self.max_text_len + 1 - len(text))
+
+            data["label"] = np.array(text)
+            data["label_node"] = np.array(text_node_num + text_pos_node)
+            data["label_index"] = np.array(text_node_index)
+            return data
+        else:
+            text, text_char_node, ch_order = self.encode(text)
+            if text is None:
+                return None
+            if len(text) >= self.max_text_len:
+                return None
+            data["length"] = np.array(len(text))
+
+            text_pos_node = [1] * (len(text) + 1) + [0] * (
+                self.max_text_len - len(text)
+            )
+
+            text.append(0)  # eos
+
+            text = text + [self.ignore_index] * (self.max_text_len + 1 - len(text))
+            data["label"] = np.array(text)
+            data["label_node"] = np.array(text_char_node + text_pos_node)
+            data["label_order"] = np.array(ch_order)
+
+            return data
+
+    def add_special_char(self, dict_character):
+        dict_character = ["</s>"] + dict_character
+        self.num_character = len(dict_character)
+        return dict_character
+
+    def encode(self, text):
+        """ """
+        if len(text) == 0 or len(text) > self.max_text_len:
+            return None, None, None
+        if self.lower:
+            text = text.lower()
+        text_node = [0 for _ in range(self.num_character)]
+        text_node[0] = 1
+        text_list = []
+        ch_order = []
+        order = 1
+        for char in text:
+            if char not in self.dict:
+                continue
+            text_list.append(self.dict[char])
+            text_node[self.dict[char]] += 1
+            ch_order.append([self.dict[char], text_node[self.dict[char]], order])
+            order += 1
+
+        no_ch_order = []
+        for char in self.character:
+            if char not in text:
+                no_ch_order.append([self.dict[char], 1, 0])
+        random.shuffle(no_ch_order)
+        ch_order = ch_order + no_ch_order
+        ch_order = ch_order[: self.max_text_len + 1]
+
+        if len(text_list) == 0:
+            return None, None, None
+        return text_list, text_node, ch_order.sort()
+
+    def encodech(self, text):
+        """ """
+        if len(text) == 0 or len(text) > self.max_text_len:
+            return None, None, None
+        if self.lower:
+            text = text.lower()
+        text_node_dict = {}
+        text_node_dict.update({0: 1})
+        character_index = [_ for _ in range(self.num_character)]
+        text_list = []
+        for char in text:
+            if char not in self.dict:
+                continue
+            i_c = self.dict[char]
+            text_list.append(i_c)
+            if i_c in text_node_dict.keys():
+                text_node_dict[i_c] += 1
+            else:
+                text_node_dict.update({i_c: 1})
+        for ic in list(text_node_dict.keys()):
+            character_index.remove(ic)
+        none_char_index = sample(character_index, 37 - len(list(text_node_dict.keys())))
+        for ic in none_char_index:
+            text_node_dict[ic] = 0
+
+        text_node_index = sorted(text_node_dict)
+        text_node_num = [text_node_dict[k] for k in text_node_index]
+        if len(text_list) == 0:
+            return None, None, None
+        return text_list, text_node_index, text_node_num
+
+
+class LatexOCRLabelEncode(object):
+    def __init__(
+        self,
+        rec_char_dict_path,
+        **kwargs,
+    ):
+        # Set the TOKENIZERS_PARALLELISM environment variable to 'false' to suppress
+        # the warning: "The current process just got forked, Disabling parallelism to avoid deadlocks..
+        #  To disable this warning, please explicitly set TOKENIZERS_PARALLELISM=(true | false)" from tokenizers
+        os.environ["TOKENIZERS_PARALLELISM"] = "false"
+        from tokenizers import Tokenizer as TokenizerFast
+
+        self.tokenizer = TokenizerFast.from_file(rec_char_dict_path)
+        self.model_input_names = ["input_ids", "token_type_ids", "attention_mask"]
+        self.pad_token_id = 0
+        self.bos_token_id = 1
+        self.eos_token_id = 2
+
+    def _convert_encoding(
+        self,
+        encoding,
+        return_token_type_ids=None,
+        return_attention_mask=None,
+        return_overflowing_tokens=False,
+        return_special_tokens_mask=False,
+        return_offsets_mapping=False,
+        return_length=False,
+        verbose=True,
+    ):
+
+        if return_token_type_ids is None:
+            return_token_type_ids = "token_type_ids" in self.model_input_names
+        if return_attention_mask is None:
+            return_attention_mask = "attention_mask" in self.model_input_names
+
+        if return_overflowing_tokens and encoding.overflowing is not None:
+            encodings = [encoding] + encoding.overflowing
+        else:
+            encodings = [encoding]
+
+        encoding_dict = defaultdict(list)
+        for e in encodings:
+            encoding_dict["input_ids"].append(e.ids)
+
+            if return_token_type_ids:
+                encoding_dict["token_type_ids"].append(e.type_ids)
+            if return_attention_mask:
+                encoding_dict["attention_mask"].append(e.attention_mask)
+            if return_special_tokens_mask:
+                encoding_dict["special_tokens_mask"].append(e.special_tokens_mask)
+            if return_offsets_mapping:
+                encoding_dict["offset_mapping"].append(e.offsets)
+            if return_length:
+                encoding_dict["length"].append(len(e.ids))
+
+        return encoding_dict, encodings
+
+    def encode(
+        self,
+        text,
+        text_pair=None,
+        return_token_type_ids=False,
+        add_special_tokens=True,
+        is_split_into_words=False,
+    ):
+        batched_input = text
+        encodings = self.tokenizer.encode_batch(
+            batched_input,
+            add_special_tokens=add_special_tokens,
+            is_pretokenized=is_split_into_words,
+        )
+        tokens_and_encodings = [
+            self._convert_encoding(
+                encoding=encoding,
+                return_token_type_ids=False,
+                return_attention_mask=None,
+                return_overflowing_tokens=False,
+                return_special_tokens_mask=False,
+                return_offsets_mapping=False,
+                return_length=False,
+                verbose=True,
+            )
+            for encoding in encodings
+        ]
+        sanitized_tokens = {}
+        for key in tokens_and_encodings[0][0].keys():
+            stack = [e for item, _ in tokens_and_encodings for e in item[key]]
+            sanitized_tokens[key] = stack
+        return sanitized_tokens
+
+    def __call__(self, eqs):
+        topk = self.encode(eqs)
+        for k, p in zip(topk, [[self.bos_token_id, self.eos_token_id], [1, 1]]):
+            process_seq = [[p[0]] + x + [p[1]] for x in topk[k]]
+            max_length = 0
+            for seq in process_seq:
+                max_length = max(max_length, len(seq))
+            labels = np.zeros((len(process_seq), max_length), dtype="int64")
+            for idx, seq in enumerate(process_seq):
+                l = len(seq)
+                labels[idx][:l] = seq
+            topk[k] = labels
+        return (
+            np.array(topk["input_ids"]).astype(np.int64),
+            np.array(topk["attention_mask"]).astype(np.int64),
+            max_length,
+        )
+
+
+class ExplicitEnum(str, Enum):
+    """
+    Enum with more explicit error message for missing values.
+    """
+
+    @classmethod
+    def _missing_(cls, value):
+        raise ValueError(
+            f"{value} is not a valid {cls.__name__}, please select one of {list(cls._value2member_map_.keys())}"
+        )
+
+
+class TruncationStrategy(ExplicitEnum):
+    """
+    Possible values for the `truncation` argument in [`PreTrainedTokenizerBase.__call__`]. Useful for tab-completion in
+    an IDE.
+    """
+
+    ONLY_FIRST = "only_first"
+    ONLY_SECOND = "only_second"
+    LONGEST_FIRST = "longest_first"
+    DO_NOT_TRUNCATE = "do_not_truncate"
+
+
+class PaddingStrategy(ExplicitEnum):
+    """
+    Possible values for the `padding` argument in [`PreTrainedTokenizerBase.__call__`]. Useful for tab-completion in an
+    IDE.
+    """
+
+    LONGEST = "longest"
+    MAX_LENGTH = "max_length"
+    DO_NOT_PAD = "do_not_pad"
+
+
+class UniMERNetLabelEncode(object):
+
+    SPECIAL_TOKENS_ATTRIBUTES = [
+        "bos_token",
+        "eos_token",
+        "unk_token",
+        "sep_token",
+        "pad_token",
+        "cls_token",
+        "mask_token",
+        "additional_special_tokens",
+    ]
+
+    def __init__(
+        self,
+        rec_char_dict_path,
+        max_seq_len,
+        **kwargs,
+    ):
+        # Set the TOKENIZERS_PARALLELISM environment variable to 'false' to suppress
+        # the warning: "The current process just got forked, Disabling parallelism to avoid deadlocks..
+        #  To disable this warning, please explicitly set TOKENIZERS_PARALLELISM=(true | false)" from tokenizers
+        os.environ["TOKENIZERS_PARALLELISM"] = "false"
+        from tokenizers import Tokenizer as TokenizerFast
+        from tokenizers import AddedToken
+
+        self._unk_token = "<unk>"
+        self._bos_token = "<s>"
+        self._eos_token = "</s>"
+        self._pad_token = "<pad>"
+        self._sep_token = None
+        self._cls_token = None
+        self._mask_token = None
+        self._additional_special_tokens = []
+        self.model_input_names = ["input_ids", "token_type_ids", "attention_mask"]
+        self.max_seq_len = max_seq_len
+        self.pad_token_id = 1
+        self.bos_token_id = 0
+        self.eos_token_id = 2
+        self.padding_side = "right"
+        self.pad_token = "<pad>"
+        self.pad_token_type_id = 0
+        self.pad_to_multiple_of = None
+        fast_tokenizer_file = os.path.join(rec_char_dict_path, "tokenizer.json")
+        tokenizer_config_file = os.path.join(
+            rec_char_dict_path, "tokenizer_config.json"
+        )
+        self.tokenizer = TokenizerFast.from_file(fast_tokenizer_file)
+        added_tokens_decoder = {}
+        added_tokens_map = {}
+
+        if tokenizer_config_file is not None:
+            with open(
+                tokenizer_config_file, encoding="utf-8"
+            ) as tokenizer_config_handle:
+                init_kwargs = json.load(tokenizer_config_handle)
+                if "added_tokens_decoder" in init_kwargs:
+                    for idx, token in init_kwargs["added_tokens_decoder"].items():
+                        if isinstance(token, dict):
+                            token = AddedToken(**token)
+                        if isinstance(token, AddedToken):
+                            added_tokens_decoder[int(idx)] = token
+                            added_tokens_map[str(token)] = token
+                        else:
+                            raise ValueError(
+                                f"Found a {token.__class__} in the saved `added_tokens_decoder`, should be a dictionary or an AddedToken instance"
+                            )
+                init_kwargs["added_tokens_decoder"] = added_tokens_decoder
+                added_tokens_decoder = init_kwargs.pop("added_tokens_decoder", {})
+                tokens_to_add = [
+                    token
+                    for index, token in sorted(
+                        added_tokens_decoder.items(), key=lambda x: x[0]
+                    )
+                    if token not in added_tokens_decoder
+                ]
+                added_tokens_encoder = self.added_tokens_encoder(added_tokens_decoder)
+                encoder = list(added_tokens_encoder.keys()) + [
+                    str(token) for token in tokens_to_add
+                ]
+                tokens_to_add += [
+                    token
+                    for token in self.all_special_tokens_extended
+                    if token not in encoder and token not in tokens_to_add
+                ]
+                if len(tokens_to_add) > 0:
+                    is_last_special = None
+                    tokens = []
+                    special_tokens = self.all_special_tokens
+                    for token in tokens_to_add:
+                        is_special = (
+                            (token.special or str(token) in special_tokens)
+                            if isinstance(token, AddedToken)
+                            else str(token) in special_tokens
+                        )
+                        if is_last_special is None or is_last_special == is_special:
+                            tokens.append(token)
+                        else:
+                            self._add_tokens(tokens, special_tokens=is_last_special)
+                            tokens = [token]
+                        is_last_special = is_special
+                    if tokens:
+                        self._add_tokens(tokens, special_tokens=is_last_special)
+
+    def _add_tokens(self, new_tokens, special_tokens=False) -> int:
+        if special_tokens:
+            return self.tokenizer.add_special_tokens(new_tokens)
+
+        return self.tokenizer.add_tokens(new_tokens)
+
+    def added_tokens_encoder(self, added_tokens_decoder):
+        return {
+            k.content: v
+            for v, k in sorted(added_tokens_decoder.items(), key=lambda item: item[0])
+        }
+
+    @property
+    def all_special_tokens(self):
+        all_toks = [str(s) for s in self.all_special_tokens_extended]
+        return all_toks
+
+    @property
+    def all_special_tokens_extended(self):
+        all_tokens = []
+        seen = set()
+        for value in self.special_tokens_map_extended.values():
+            if isinstance(value, (list, tuple)):
+                tokens_to_add = [token for token in value if str(token) not in seen]
+            else:
+                tokens_to_add = [value] if str(value) not in seen else []
+            seen.update(map(str, tokens_to_add))
+            all_tokens.extend(tokens_to_add)
+        return all_tokens
+
+    @property
+    def special_tokens_map_extended(self):
+        set_attr = {}
+        for attr in self.SPECIAL_TOKENS_ATTRIBUTES:
+            attr_value = getattr(self, "_" + attr)
+            if attr_value:
+                set_attr[attr] = attr_value
+        return set_attr
+
+    def set_truncation_and_padding(
+        self,
+        padding_strategy,
+        truncation_strategy,
+        max_length,
+        stride,
+        pad_to_multiple_of,
+    ):
+        _truncation = self.tokenizer.truncation
+        _padding = self.tokenizer.padding
+        # Set truncation and padding on the backend tokenizer
+        if truncation_strategy == TruncationStrategy.DO_NOT_TRUNCATE:
+            if _truncation is not None:
+                self._tokenizer.no_truncation()
+        else:
+            target = {
+                "max_length": max_length,
+                "stride": stride,
+                "strategy": truncation_strategy.value,
+                "direction": "right",
+            }
+
+            if _truncation is None:
+                current = None
+            else:
+                current = {k: _truncation.get(k, None) for k in target}
+
+            if current != target:
+                self.tokenizer.enable_truncation(**target)
+        if padding_strategy == PaddingStrategy.DO_NOT_PAD:
+            if _padding is not None:
+                self.tokenizer.no_padding()
+        else:
+            length = (
+                max_length if padding_strategy == PaddingStrategy.MAX_LENGTH else None
+            )
+            target = {
+                "length": length,
+                "direction": self.padding_side,
+                "pad_id": self.pad_token_id,
+                "pad_token": self.pad_token,
+                "pad_type_id": self.pad_token_type_id,
+                "pad_to_multiple_of": pad_to_multiple_of,
+            }
+            if _padding != target:
+                self.tokenizer.enable_padding(**target)
+
+    def _convert_encoding(
+        self,
+        encoding,
+        return_token_type_ids=None,
+        return_attention_mask=None,
+        return_overflowing_tokens=False,
+        return_special_tokens_mask=False,
+        return_offsets_mapping=False,
+        return_length=False,
+        verbose=True,
+    ):
+
+        if return_token_type_ids is None:
+            return_token_type_ids = "token_type_ids" in self.model_input_names
+        if return_attention_mask is None:
+            return_attention_mask = "attention_mask" in self.model_input_names
+
+        if return_overflowing_tokens and encoding.overflowing is not None:
+            encodings = [encoding] + encoding.overflowing
+        else:
+            encodings = [encoding]
+
+        encoding_dict = defaultdict(list)
+        for e in encodings:
+            encoding_dict["input_ids"].append(e.ids)
+            if return_token_type_ids:
+                encoding_dict["token_type_ids"].append(e.type_ids)
+            if return_attention_mask:
+                encoding_dict["attention_mask"].append(e.attention_mask)
+            if return_special_tokens_mask:
+                encoding_dict["special_tokens_mask"].append(e.special_tokens_mask)
+            if return_offsets_mapping:
+                encoding_dict["offset_mapping"].append(e.offsets)
+            if return_length:
+                encoding_dict["length"].append(len(e.ids))
+
+        return encoding_dict, encodings
+
+    def encode(
+        self,
+        text,
+        text_pair=None,
+        return_token_type_ids=False,
+        add_special_tokens=True,
+        is_split_into_words=False,
+    ):
+        batched_input = text
+        self.set_truncation_and_padding(
+            padding_strategy=PaddingStrategy.LONGEST,
+            truncation_strategy=TruncationStrategy.LONGEST_FIRST,
+            max_length=self.max_seq_len,
+            stride=0,
+            pad_to_multiple_of=None,
+        )
+        encodings = self.tokenizer.encode_batch(
+            batched_input,
+            add_special_tokens=add_special_tokens,
+            is_pretokenized=is_split_into_words,
+        )
+
+        tokens_and_encodings = [
+            self._convert_encoding(
+                encoding=encoding,
+                return_token_type_ids=False,
+                return_attention_mask=None,
+                return_overflowing_tokens=False,
+                return_special_tokens_mask=False,
+                return_offsets_mapping=False,
+                return_length=False,
+                verbose=True,
+            )
+            for encoding in encodings
+        ]
+        sanitized_tokens = {}
+        for key in tokens_and_encodings[0][0].keys():
+            stack = [e for item, _ in tokens_and_encodings for e in item[key]]
+            sanitized_tokens[key] = stack
+        return sanitized_tokens
+
+    def __call__(self, data):
+        eqs = data["label"]
+        topk = self.encode([eqs])
+        for k, p in zip(topk, [[self.bos_token_id, self.eos_token_id], [1, 1]]):
+            process_seq = [x for x in topk[k]]
+            max_length = 0
+            for seq in process_seq:
+                max_length = max(max_length, len(seq))
+        data["label"] = np.array(topk["input_ids"]).astype(np.int64)[0]
+        data["attention_mask"] = np.array(topk["attention_mask"]).astype(np.int64)[0]
+        return data
diff --git a/ppocr/data/imaug/latex_ocr_aug.py b/ppocr/data/imaug/latex_ocr_aug.py
new file mode 100644
index 0000000..6417dab
--- /dev/null
+++ b/ppocr/data/imaug/latex_ocr_aug.py
@@ -0,0 +1,183 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+This code is refer from:
+https://github.com/lukas-blecher/LaTeX-OCR/blob/main/pix2tex/dataset/transforms.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+import os
+
+os.environ["NO_ALBUMENTATIONS_UPDATE"] = "1"
+
+import math
+import cv2
+import numpy as np
+import albumentations as A
+from PIL import Image
+
+
+class LatexTrainTransform:
+    def __init__(self, bitmap_prob=0.04, **kwargs):
+        # your init code
+        self.bitmap_prob = bitmap_prob
+        self.train_transform = A.Compose(
+            [
+                A.Compose(
+                    [
+                        A.ShiftScaleRotate(
+                            shift_limit=0,
+                            scale_limit=(-0.15, 0),
+                            rotate_limit=1,
+                            border_mode=0,
+                            interpolation=3,
+                            value=[255, 255, 255],
+                            p=1,
+                        ),
+                        A.GridDistortion(
+                            distort_limit=0.1,
+                            border_mode=0,
+                            interpolation=3,
+                            value=[255, 255, 255],
+                            p=0.5,
+                        ),
+                    ],
+                    p=0.15,
+                ),
+                A.RGBShift(r_shift_limit=15, g_shift_limit=15, b_shift_limit=15, p=0.3),
+                A.GaussNoise(10, p=0.2),
+                A.RandomBrightnessContrast(0.05, (-0.2, 0), True, p=0.2),
+                A.ImageCompression(95, p=0.3),
+                A.ToGray(always_apply=True),
+            ]
+        )
+
+    def __call__(self, data):
+        img = data["image"]
+        if np.random.random() < self.bitmap_prob:
+            img[img != 255] = 0
+        img = self.train_transform(image=img)["image"]
+        data["image"] = img
+        return data
+
+
+class LatexTestTransform:
+    def __init__(self, **kwargs):
+        # your init code
+        self.test_transform = A.Compose(
+            [
+                A.ToGray(always_apply=True),
+            ]
+        )
+
+    def __call__(self, data):
+        img = data["image"]
+        img = self.test_transform(image=img)["image"]
+        data["image"] = img
+        return data
+
+
+class MinMaxResize:
+    def __init__(self, min_dimensions=[32, 32], max_dimensions=[672, 192], **kwargs):
+        # your init code
+        self.min_dimensions = min_dimensions
+        self.max_dimensions = max_dimensions
+        # pass
+
+    def pad_(self, img, divable=32):
+        threshold = 128
+        data = np.array(img.convert("LA"))
+        if data[..., -1].var() == 0:
+            data = (data[..., 0]).astype(np.uint8)
+        else:
+            data = (255 - data[..., -1]).astype(np.uint8)
+        data = (data - data.min()) / (data.max() - data.min()) * 255
+        if data.mean() > threshold:
+            # To invert the text to white
+            gray = 255 * (data < threshold).astype(np.uint8)
+        else:
+            gray = 255 * (data > threshold).astype(np.uint8)
+            data = 255 - data
+
+        coords = cv2.findNonZero(gray)  # Find all non-zero points (text)
+        a, b, w, h = cv2.boundingRect(coords)  # Find minimum spanning bounding box
+        rect = data[b : b + h, a : a + w]
+        im = Image.fromarray(rect).convert("L")
+        dims = []
+        for x in [w, h]:
+            div, mod = divmod(x, divable)
+            dims.append(divable * (div + (1 if mod > 0 else 0)))
+        padded = Image.new("L", dims, 255)
+        padded.paste(im, (0, 0, im.size[0], im.size[1]))
+        return padded
+
+    def minmax_size_(self, img, max_dimensions, min_dimensions):
+        if max_dimensions is not None:
+            ratios = [a / b for a, b in zip(img.size, max_dimensions)]
+            if any([r > 1 for r in ratios]):
+                size = np.array(img.size) // max(ratios)
+                img = img.resize(tuple(size.astype(int)), Image.BILINEAR)
+        if min_dimensions is not None:
+            # hypothesis: there is a dim in img smaller than min_dimensions, and return a proper dim >= min_dimensions
+            padded_size = [
+                max(img_dim, min_dim)
+                for img_dim, min_dim in zip(img.size, min_dimensions)
+            ]
+            if padded_size != list(img.size):  # assert hypothesis
+                padded_im = Image.new("L", padded_size, 255)
+                padded_im.paste(img, img.getbbox())
+                img = padded_im
+        return img
+
+    def __call__(self, data):
+        img = data["image"]
+        h, w = img.shape[:2]
+        if (
+            self.min_dimensions[0] <= w <= self.max_dimensions[0]
+            and self.min_dimensions[1] <= h <= self.max_dimensions[1]
+        ):
+            return data
+        else:
+            im = Image.fromarray(np.uint8(img))
+            im = self.minmax_size_(
+                self.pad_(im), self.max_dimensions, self.min_dimensions
+            )
+            im = np.array(im)
+            im = np.dstack((im, im, im))
+            data["image"] = im
+            return data
+
+
+class LatexImageFormat:
+    def __init__(self, **kwargs):
+        # your init code
+        pass
+
+    def __call__(self, data):
+        img = data["image"]
+        im_h, im_w = img.shape[:2]
+        divide_h = math.ceil(im_h / 16) * 16
+        divide_w = math.ceil(im_w / 16) * 16
+        img = img[:, :, 0]
+        img = np.pad(
+            img, ((0, divide_h - im_h), (0, divide_w - im_w)), constant_values=(1, 1)
+        )
+        img_expanded = img[:, :, np.newaxis].transpose(2, 0, 1)
+        data["image"] = img_expanded
+        return data
diff --git a/ppocr/data/imaug/make_border_map.py b/ppocr/data/imaug/make_border_map.py
new file mode 100644
index 0000000..537c8b4
--- /dev/null
+++ b/ppocr/data/imaug/make_border_map.py
@@ -0,0 +1,191 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/WenmuZhou/DBNet.pytorch/blob/master/data_loader/modules/make_border_map.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+import numpy as np
+import cv2
+
+np.seterr(divide="ignore", invalid="ignore")
+import pyclipper
+from shapely.geometry import Polygon
+import sys
+import warnings
+
+warnings.simplefilter("ignore")
+
+__all__ = ["MakeBorderMap"]
+
+
+class MakeBorderMap(object):
+    def __init__(self, shrink_ratio=0.4, thresh_min=0.3, thresh_max=0.7, **kwargs):
+        self.shrink_ratio = shrink_ratio
+        self.thresh_min = thresh_min
+        self.thresh_max = thresh_max
+        if "total_epoch" in kwargs:
+            self._base_shrink_ratio = shrink_ratio
+            self._total_epoch = kwargs["total_epoch"]
+
+    def __call__(self, data):
+        img = data["image"]
+        text_polys = data["polys"]
+        ignore_tags = data["ignore_tags"]
+
+        shrink_ratio = self.shrink_ratio
+        if "epoch" in data and hasattr(self, "_base_shrink_ratio"):
+            shrink_ratio = self._base_shrink_ratio + 0.2 * data["epoch"] / float(
+                self._total_epoch
+            )
+
+        canvas = np.zeros(img.shape[:2], dtype=np.float32)
+        mask = np.zeros(img.shape[:2], dtype=np.float32)
+
+        for i in range(len(text_polys)):
+            if ignore_tags[i]:
+                continue
+            self.draw_border_map(
+                text_polys[i], canvas, mask=mask, shrink_ratio=shrink_ratio
+            )
+        canvas = canvas * (self.thresh_max - self.thresh_min) + self.thresh_min
+
+        data["threshold_map"] = canvas
+        data["threshold_mask"] = mask
+        return data
+
+    def draw_border_map(self, polygon, canvas, mask, shrink_ratio=None):
+        if shrink_ratio is None:
+            shrink_ratio = self.shrink_ratio
+        polygon = np.array(polygon)
+        assert polygon.ndim == 2
+        assert polygon.shape[1] == 2
+
+        if np.isnan(polygon).any():
+            return
+        polygon_shape = Polygon(polygon)
+        if polygon_shape.area <= 0:
+            return
+        distance = (
+            polygon_shape.area * (1 - np.power(shrink_ratio, 2)) / polygon_shape.length
+        )
+        subject = [tuple(l) for l in polygon]
+        padding = pyclipper.PyclipperOffset()
+        padding.AddPath(subject, pyclipper.JT_ROUND, pyclipper.ET_CLOSEDPOLYGON)
+
+        padded_polygon = padding.Execute(distance)
+        if not padded_polygon:
+            return
+        padded_polygon = np.array(padded_polygon[0])
+        cv2.fillPoly(mask, [padded_polygon.astype(np.int32)], 1.0)
+
+        xmin = padded_polygon[:, 0].min()
+        xmax = padded_polygon[:, 0].max()
+        ymin = padded_polygon[:, 1].min()
+        ymax = padded_polygon[:, 1].max()
+        width = xmax - xmin + 1
+        height = ymax - ymin + 1
+
+        polygon[:, 0] = polygon[:, 0] - xmin
+        polygon[:, 1] = polygon[:, 1] - ymin
+
+        xs = np.broadcast_to(
+            np.linspace(0, width - 1, num=width).reshape(1, width), (height, width)
+        )
+        ys = np.broadcast_to(
+            np.linspace(0, height - 1, num=height).reshape(height, 1), (height, width)
+        )
+
+        distance_map = np.zeros((polygon.shape[0], height, width), dtype=np.float32)
+        for i in range(polygon.shape[0]):
+            j = (i + 1) % polygon.shape[0]
+            absolute_distance = self._distance(xs, ys, polygon[i], polygon[j])
+            distance_map[i] = np.clip(absolute_distance / distance, 0, 1)
+        distance_map = distance_map.min(axis=0)
+
+        xmin_valid = min(max(0, xmin), canvas.shape[1] - 1)
+        xmax_valid = min(max(0, xmax), canvas.shape[1] - 1)
+        ymin_valid = min(max(0, ymin), canvas.shape[0] - 1)
+        ymax_valid = min(max(0, ymax), canvas.shape[0] - 1)
+        canvas[ymin_valid : ymax_valid + 1, xmin_valid : xmax_valid + 1] = np.fmax(
+            1
+            - distance_map[
+                ymin_valid - ymin : ymax_valid - ymax + height,
+                xmin_valid - xmin : xmax_valid - xmax + width,
+            ],
+            canvas[ymin_valid : ymax_valid + 1, xmin_valid : xmax_valid + 1],
+        )
+
+    def _distance(self, xs, ys, point_1, point_2):
+        """
+        compute the distance from point to a line
+        ys: coordinates in the first axis
+        xs: coordinates in the second axis
+        point_1, point_2: (x, y), the end of the line
+        """
+        height, width = xs.shape[:2]
+        square_distance_1 = np.square(xs - point_1[0]) + np.square(ys - point_1[1])
+        square_distance_2 = np.square(xs - point_2[0]) + np.square(ys - point_2[1])
+        square_distance = np.square(point_1[0] - point_2[0]) + np.square(
+            point_1[1] - point_2[1]
+        )
+
+        cosin = (square_distance - square_distance_1 - square_distance_2) / (
+            2 * np.sqrt(square_distance_1 * square_distance_2)
+        )
+        square_sin = 1 - np.square(cosin)
+        square_sin = np.nan_to_num(square_sin)
+        result = np.sqrt(
+            square_distance_1 * square_distance_2 * square_sin / square_distance
+        )
+
+        result[cosin < 0] = np.sqrt(np.fmin(square_distance_1, square_distance_2))[
+            cosin < 0
+        ]
+        # self.extend_line(point_1, point_2, result)
+        return result
+
+    def extend_line(self, point_1, point_2, result, shrink_ratio):
+        ex_point_1 = (
+            int(round(point_1[0] + (point_1[0] - point_2[0]) * (1 + shrink_ratio))),
+            int(round(point_1[1] + (point_1[1] - point_2[1]) * (1 + shrink_ratio))),
+        )
+        cv2.line(
+            result,
+            tuple(ex_point_1),
+            tuple(point_1),
+            4096.0,
+            1,
+            lineType=cv2.LINE_AA,
+            shift=0,
+        )
+        ex_point_2 = (
+            int(round(point_2[0] + (point_2[0] - point_1[0]) * (1 + shrink_ratio))),
+            int(round(point_2[1] + (point_2[1] - point_1[1]) * (1 + shrink_ratio))),
+        )
+        cv2.line(
+            result,
+            tuple(ex_point_2),
+            tuple(point_2),
+            4096.0,
+            1,
+            lineType=cv2.LINE_AA,
+            shift=0,
+        )
+        return ex_point_1, ex_point_2
diff --git a/ppocr/data/imaug/make_pse_gt.py b/ppocr/data/imaug/make_pse_gt.py
new file mode 100644
index 0000000..7be1bb8
--- /dev/null
+++ b/ppocr/data/imaug/make_pse_gt.py
@@ -0,0 +1,104 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+import cv2
+import numpy as np
+import pyclipper
+from shapely.geometry import Polygon
+
+__all__ = ["MakePseGt"]
+
+
+class MakePseGt(object):
+    def __init__(self, kernel_num=7, size=640, min_shrink_ratio=0.4, **kwargs):
+        self.kernel_num = kernel_num
+        self.min_shrink_ratio = min_shrink_ratio
+        self.size = size
+
+    def __call__(self, data):
+        image = data["image"]
+        text_polys = data["polys"]
+        ignore_tags = data["ignore_tags"]
+
+        h, w, _ = image.shape
+        short_edge = min(h, w)
+        if short_edge < self.size:
+            # keep short_size >= self.size
+            scale = self.size / short_edge
+            image = cv2.resize(image, dsize=None, fx=scale, fy=scale)
+            text_polys *= scale
+
+        gt_kernels = []
+        for i in range(1, self.kernel_num + 1):
+            # s1->sn, from big to small
+            rate = 1.0 - (1.0 - self.min_shrink_ratio) / (self.kernel_num - 1) * i
+            text_kernel, ignore_tags = self.generate_kernel(
+                image.shape[0:2], rate, text_polys, ignore_tags
+            )
+            gt_kernels.append(text_kernel)
+
+        training_mask = np.ones(image.shape[0:2], dtype="uint8")
+        for i in range(text_polys.shape[0]):
+            if ignore_tags[i]:
+                cv2.fillPoly(
+                    training_mask, text_polys[i].astype(np.int32)[np.newaxis, :, :], 0
+                )
+
+        gt_kernels = np.array(gt_kernels)
+        gt_kernels[gt_kernels > 0] = 1
+
+        data["image"] = image
+        data["polys"] = text_polys
+        data["gt_kernels"] = gt_kernels[0:]
+        data["gt_text"] = gt_kernels[0]
+        data["mask"] = training_mask.astype("float32")
+        return data
+
+    def generate_kernel(self, img_size, shrink_ratio, text_polys, ignore_tags=None):
+        """
+        Refer to part of the code:
+        https://github.com/open-mmlab/mmocr/blob/main/mmocr/datasets/pipelines/textdet_targets/base_textdet_targets.py
+        """
+
+        h, w = img_size
+        text_kernel = np.zeros((h, w), dtype=np.float32)
+        for i, poly in enumerate(text_polys):
+            polygon = Polygon(poly)
+            distance = (
+                polygon.area
+                * (1 - shrink_ratio * shrink_ratio)
+                / (polygon.length + 1e-6)
+            )
+            subject = [tuple(l) for l in poly]
+            pco = pyclipper.PyclipperOffset()
+            pco.AddPath(subject, pyclipper.JT_ROUND, pyclipper.ET_CLOSEDPOLYGON)
+            shrunk = np.array(pco.Execute(-distance))
+
+            if len(shrunk) == 0 or shrunk.size == 0:
+                if ignore_tags is not None:
+                    ignore_tags[i] = True
+                continue
+            try:
+                shrunk = np.array(shrunk[0]).reshape(-1, 2)
+            except:
+                if ignore_tags is not None:
+                    ignore_tags[i] = True
+                continue
+            cv2.fillPoly(text_kernel, [shrunk.astype(np.int32)], i + 1)
+        return text_kernel, ignore_tags
diff --git a/ppocr/data/imaug/make_shrink_map.py b/ppocr/data/imaug/make_shrink_map.py
new file mode 100644
index 0000000..c7bcc87
--- /dev/null
+++ b/ppocr/data/imaug/make_shrink_map.py
@@ -0,0 +1,132 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/WenmuZhou/DBNet.pytorch/blob/master/data_loader/modules/make_shrink_map.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+import numpy as np
+import cv2
+from shapely.geometry import Polygon
+import pyclipper
+
+__all__ = ["MakeShrinkMap"]
+
+
+class MakeShrinkMap(object):
+    r"""
+    Making binary mask from detection data with ICDAR format.
+    Typically following the process of class `MakeICDARData`.
+    """
+
+    def __init__(self, min_text_size=8, shrink_ratio=0.4, **kwargs):
+        self.min_text_size = min_text_size
+        self.shrink_ratio = shrink_ratio
+        if "total_epoch" in kwargs:
+            self._base_shrink_ratio = shrink_ratio
+            self._total_epoch = kwargs["total_epoch"]
+
+    def __call__(self, data):
+        image = data["image"]
+        text_polys = data["polys"]
+        ignore_tags = data["ignore_tags"]
+
+        h, w = image.shape[:2]
+        text_polys, ignore_tags = self.validate_polygons(text_polys, ignore_tags, h, w)
+        gt = np.zeros((h, w), dtype=np.float32)
+        mask = np.ones((h, w), dtype=np.float32)
+        for i in range(len(text_polys)):
+            polygon = text_polys[i]
+            height = max(polygon[:, 1]) - min(polygon[:, 1])
+            width = max(polygon[:, 0]) - min(polygon[:, 0])
+            if ignore_tags[i] or min(height, width) < self.min_text_size:
+                cv2.fillPoly(mask, polygon.astype(np.int32)[np.newaxis, :, :], 0)
+                ignore_tags[i] = True
+            else:
+                if np.isnan(polygon).any():
+                    ignore_tags[i] = True
+                    continue
+                polygon_shape = Polygon(polygon)
+                subject = [tuple(l) for l in polygon]
+                padding = pyclipper.PyclipperOffset()
+                padding.AddPath(subject, pyclipper.JT_ROUND, pyclipper.ET_CLOSEDPOLYGON)
+                shrunk = []
+
+                # Increase the shrink ratio every time we get multiple polygon returned back
+                shrink_ratio = self.shrink_ratio
+                if "epoch" in data and hasattr(self, "_base_shrink_ratio"):
+                    shrink_ratio = self._base_shrink_ratio + 0.2 * data[
+                        "epoch"
+                    ] / float(self._total_epoch)
+                possible_ratios = np.arange(shrink_ratio, 1, shrink_ratio)
+                np.append(possible_ratios, 1)
+                # print(possible_ratios)
+                for ratio in possible_ratios:
+                    # print(f"Change shrink ratio to {ratio}")
+                    distance = (
+                        polygon_shape.area
+                        * (1 - np.power(ratio, 2))
+                        / polygon_shape.length
+                    )
+                    shrunk = padding.Execute(-distance)
+                    if len(shrunk) == 1:
+                        break
+
+                if shrunk == []:
+                    cv2.fillPoly(mask, polygon.astype(np.int32)[np.newaxis, :, :], 0)
+                    ignore_tags[i] = True
+                    continue
+
+                for each_shrink in shrunk:
+                    shrink = np.array(each_shrink).reshape(-1, 2)
+                    cv2.fillPoly(gt, [shrink.astype(np.int32)], 1)
+
+        data["shrink_map"] = gt
+        data["shrink_mask"] = mask
+        return data
+
+    def validate_polygons(self, polygons, ignore_tags, h, w):
+        """
+        polygons (numpy.array, required): of shape (num_instances, num_points, 2)
+        """
+        if len(polygons) == 0:
+            return polygons, ignore_tags
+        assert len(polygons) == len(ignore_tags)
+        for polygon in polygons:
+            polygon[:, 0] = np.clip(polygon[:, 0], 0, w - 1)
+            polygon[:, 1] = np.clip(polygon[:, 1], 0, h - 1)
+
+        for i in range(len(polygons)):
+            area = self.polygon_area(polygons[i])
+            if abs(area) < 1:
+                ignore_tags[i] = True
+            if area > 0:
+                polygons[i] = polygons[i][::-1, :]
+        return polygons, ignore_tags
+
+    def polygon_area(self, polygon):
+        """
+        compute polygon area
+        """
+        area = 0
+        q = polygon[-1]
+        for p in polygon:
+            area += p[0] * q[1] - p[1] * q[0]
+            q = p
+        return area / 2.0
diff --git a/ppocr/data/imaug/operators.py b/ppocr/data/imaug/operators.py
new file mode 100644
index 0000000..79b7a86
--- /dev/null
+++ b/ppocr/data/imaug/operators.py
@@ -0,0 +1,692 @@
+"""
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+import sys
+import cv2
+import numpy as np
+import math
+import random
+from PIL import Image
+from paddle import get_device
+
+
+class DecodeImage(object):
+    """decode image"""
+
+    def __init__(
+        self, img_mode="RGB", channel_first=False, ignore_orientation=False, **kwargs
+    ):
+        self.img_mode = img_mode
+        self.channel_first = channel_first
+        self.ignore_orientation = ignore_orientation
+
+    def __call__(self, data):
+        img = data["image"]
+        assert type(img) is bytes and len(img) > 0, "invalid input 'img' in DecodeImage"
+        img = np.frombuffer(img, dtype="uint8")
+        if self.img_mode == "GRAY":
+            # For GRAY mode, decode directly to a single-channel grayscale image.
+            decode_flag = cv2.IMREAD_GRAYSCALE
+        else:
+            # For RGB mode, decode to a 3-channel color image.
+            decode_flag = cv2.IMREAD_COLOR
+
+        if self.ignore_orientation:
+            decode_flag |= cv2.IMREAD_IGNORE_ORIENTATION
+
+        img = cv2.imdecode(img, decode_flag)
+
+        if img is None:
+            return None
+        if self.img_mode == "GRAY":
+            img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
+        elif self.img_mode == "RGB":
+            assert img.shape[2] == 3, "invalid shape of image[%s]" % (img.shape)
+            img = img[:, :, ::-1]
+
+        if self.channel_first:
+            img = img.transpose((2, 0, 1))
+
+        data["image"] = img
+        return data
+
+
+class NormalizeImage(object):
+    """normalize image such as subtract mean, divide std"""
+
+    def __init__(self, scale=None, mean=None, std=None, order="chw", **kwargs):
+        if isinstance(scale, str):
+            scale = eval(scale)
+        self.scale = np.float32(scale if scale is not None else 1.0 / 255.0)
+        mean = mean if mean is not None else [0.485, 0.456, 0.406]
+        std = std if std is not None else [0.229, 0.224, 0.225]
+
+        shape = (3, 1, 1) if order == "chw" else (1, 1, 3)
+        self.mean = np.array(mean).reshape(shape).astype("float32")
+        self.std = np.array(std).reshape(shape).astype("float32")
+
+    def __call__(self, data):
+        img = data["image"]
+        from PIL import Image
+
+        if isinstance(img, Image.Image):
+            img = np.array(img)
+        assert isinstance(img, np.ndarray), "invalid input 'img' in NormalizeImage"
+        data["image"] = (img.astype("float32") * self.scale - self.mean) / self.std
+        return data
+
+
+class ToCHWImage(object):
+    """convert hwc image to chw image"""
+
+    def __init__(self, **kwargs):
+        pass
+
+    def __call__(self, data):
+        img = data["image"]
+        from PIL import Image
+
+        if isinstance(img, Image.Image):
+            img = np.array(img)
+        data["image"] = img.transpose((2, 0, 1))
+        return data
+
+
+class Fasttext(object):
+    def __init__(self, path="None", **kwargs):
+        import fasttext
+
+        self.fast_model = fasttext.load_model(path)
+
+    def __call__(self, data):
+        label = data["label"]
+        fast_label = self.fast_model[label]
+        data["fast_label"] = fast_label
+        return data
+
+
+class KeepKeys(object):
+    def __init__(self, keep_keys, **kwargs):
+        self.keep_keys = keep_keys
+
+    def __call__(self, data):
+        data_list = []
+        for key in self.keep_keys:
+            data_list.append(data[key])
+        return data_list
+
+
+class Pad(object):
+    def __init__(self, size=None, size_div=32, **kwargs):
+        if size is not None and not isinstance(size, (int, list, tuple)):
+            raise TypeError(
+                "Type of target_size is invalid. Now is {}".format(type(size))
+            )
+        if isinstance(size, int):
+            size = [size, size]
+        self.size = size
+        self.size_div = size_div
+
+    def __call__(self, data):
+        img = data["image"]
+        img_h, img_w = img.shape[0], img.shape[1]
+        if self.size:
+            resize_h2, resize_w2 = self.size
+            assert (
+                img_h < resize_h2 and img_w < resize_w2
+            ), "(h, w) of target size should be greater than (img_h, img_w)"
+        else:
+            resize_h2 = max(
+                int(math.ceil(img.shape[0] / self.size_div) * self.size_div),
+                self.size_div,
+            )
+            resize_w2 = max(
+                int(math.ceil(img.shape[1] / self.size_div) * self.size_div),
+                self.size_div,
+            )
+        img = cv2.copyMakeBorder(
+            img,
+            0,
+            resize_h2 - img_h,
+            0,
+            resize_w2 - img_w,
+            cv2.BORDER_CONSTANT,
+            value=0,
+        )
+        data["image"] = img
+        return data
+
+
+class Resize(object):
+    def __init__(self, size=(640, 640), **kwargs):
+        self.size = size
+
+    def resize_image(self, img):
+        resize_h, resize_w = self.size
+        ori_h, ori_w = img.shape[:2]  # (h, w, c)
+        ratio_h = float(resize_h) / ori_h
+        ratio_w = float(resize_w) / ori_w
+        img = cv2.resize(img, (int(resize_w), int(resize_h)))
+        return img, [ratio_h, ratio_w]
+
+    def __call__(self, data):
+        img = data["image"]
+        if "polys" in data:
+            text_polys = data["polys"]
+
+        img_resize, [ratio_h, ratio_w] = self.resize_image(img)
+        if "polys" in data:
+            new_boxes = []
+            for box in text_polys:
+                new_box = []
+                for cord in box:
+                    new_box.append([cord[0] * ratio_w, cord[1] * ratio_h])
+                new_boxes.append(new_box)
+            data["polys"] = np.array(new_boxes, dtype=np.float32)
+        data["image"] = img_resize
+        return data
+
+
+class DetResizeForTest(object):
+    def __init__(self, **kwargs):
+        super(DetResizeForTest, self).__init__()
+        self.resize_type = 0
+        self.keep_ratio = False
+        self.max_side_limit = kwargs.get("max_side_limit", 4000)
+        if "image_shape" in kwargs:
+            self.image_shape = kwargs["image_shape"]
+            self.resize_type = 1
+            if "keep_ratio" in kwargs:
+                self.keep_ratio = kwargs["keep_ratio"]
+        elif "limit_side_len" in kwargs:
+            self.limit_side_len = kwargs["limit_side_len"]
+            self.limit_type = kwargs.get("limit_type", "min")
+        elif "resize_long" in kwargs:
+            self.resize_type = 2
+            self.resize_long = kwargs.get("resize_long", 960)
+        else:
+            self.limit_side_len = 736
+            self.limit_type = "min"
+
+    def __call__(self, data):
+        img = data["image"]
+        src_h, src_w, _ = img.shape
+        if sum([src_h, src_w]) < 64:
+            img = self.image_padding(img)
+
+        if self.resize_type == 0:
+            # img, shape = self.resize_image_type0(img)
+            img, [ratio_h, ratio_w] = self.resize_image_type0(img)
+        elif self.resize_type == 2:
+            img, [ratio_h, ratio_w] = self.resize_image_type2(img)
+        else:
+            # img, shape = self.resize_image_type1(img)
+            img, [ratio_h, ratio_w] = self.resize_image_type1(img)
+        data["image"] = img
+        data["shape"] = np.array([src_h, src_w, ratio_h, ratio_w])
+        if "iluvatar_gpu" in get_device():
+            data["shape"] = np.array([src_h, src_w, ratio_h, ratio_w]).astype(
+                np.float32
+            )
+        return data
+
+    def image_padding(self, im, value=0):
+        h, w, c = im.shape
+        im_pad = np.zeros((max(32, h), max(32, w), c), np.uint8) + value
+        im_pad[:h, :w, :] = im
+        return im_pad
+
+    def resize_image_type1(self, img):
+        resize_h, resize_w = self.image_shape
+        ori_h, ori_w = img.shape[:2]  # (h, w, c)
+        if self.keep_ratio is True:
+            resize_w = ori_w * resize_h / ori_h
+            N = math.ceil(resize_w / 32)
+            resize_w = N * 32
+        ratio_h = float(resize_h) / ori_h
+        ratio_w = float(resize_w) / ori_w
+        img = cv2.resize(img, (int(resize_w), int(resize_h)))
+        # return img, np.array([ori_h, ori_w])
+        return img, [ratio_h, ratio_w]
+
+    def resize_image_type0(self, img):
+        """
+        resize image to a size multiple of 32 which is required by the network
+        args:
+            img(array): array with shape [h, w, c]
+        return(tuple):
+            img, (ratio_h, ratio_w)
+        """
+        limit_side_len = self.limit_side_len
+        h, w, c = img.shape
+
+        # limit the max side
+        if self.limit_type == "max":
+            if max(h, w) > limit_side_len:
+                if h > w:
+                    ratio = float(limit_side_len) / h
+                else:
+                    ratio = float(limit_side_len) / w
+            else:
+                ratio = 1.0
+        elif self.limit_type == "min":
+            if min(h, w) < limit_side_len:
+                if h < w:
+                    ratio = float(limit_side_len) / h
+                else:
+                    ratio = float(limit_side_len) / w
+            else:
+                ratio = 1.0
+        elif self.limit_type == "resize_long":
+            ratio = float(limit_side_len) / max(h, w)
+        else:
+            raise Exception("not support limit type, image ")
+        resize_h = int(h * ratio)
+        resize_w = int(w * ratio)
+        if max(resize_h, resize_w) > self.max_side_limit:
+            print(
+                f"Resized image size ({resize_h}x{resize_w}) exceeds max_side_limit of {self.max_side_limit}. "
+                f"Resizing to fit within limit."
+            )
+            ratio = float(self.max_side_limit) / max(resize_h, resize_w)
+            resize_h, resize_w = int(resize_h * ratio), int(resize_w * ratio)
+
+        resize_h = max(int(round(resize_h / 32) * 32), 32)
+        resize_w = max(int(round(resize_w / 32) * 32), 32)
+
+        try:
+            if int(resize_w) <= 0 or int(resize_h) <= 0:
+                return None, (None, None)
+            img = cv2.resize(img, (int(resize_w), int(resize_h)))
+        except:
+            print(img.shape, resize_w, resize_h)
+            sys.exit(0)
+        ratio_h = resize_h / float(h)
+        ratio_w = resize_w / float(w)
+        return img, [ratio_h, ratio_w]
+
+    def resize_image_type2(self, img):
+        h, w, _ = img.shape
+
+        resize_w = w
+        resize_h = h
+
+        if resize_h > resize_w:
+            ratio = float(self.resize_long) / resize_h
+        else:
+            ratio = float(self.resize_long) / resize_w
+
+        resize_h = int(resize_h * ratio)
+        resize_w = int(resize_w * ratio)
+
+        max_stride = 128
+        resize_h = (resize_h + max_stride - 1) // max_stride * max_stride
+        resize_w = (resize_w + max_stride - 1) // max_stride * max_stride
+        img = cv2.resize(img, (int(resize_w), int(resize_h)))
+        ratio_h = resize_h / float(h)
+        ratio_w = resize_w / float(w)
+
+        return img, [ratio_h, ratio_w]
+
+
+class E2EResizeForTest(object):
+    def __init__(self, **kwargs):
+        super(E2EResizeForTest, self).__init__()
+        self.max_side_len = kwargs["max_side_len"]
+        self.valid_set = kwargs["valid_set"]
+
+    def __call__(self, data):
+        img = data["image"]
+        src_h, src_w, _ = img.shape
+        if self.valid_set == "totaltext":
+            im_resized, [ratio_h, ratio_w] = self.resize_image_for_totaltext(
+                img, max_side_len=self.max_side_len
+            )
+        else:
+            im_resized, (ratio_h, ratio_w) = self.resize_image(
+                img, max_side_len=self.max_side_len
+            )
+        data["image"] = im_resized
+        data["shape"] = np.array([src_h, src_w, ratio_h, ratio_w])
+        return data
+
+    def resize_image_for_totaltext(self, im, max_side_len=512):
+        h, w, _ = im.shape
+        resize_w = w
+        resize_h = h
+        ratio = 1.25
+        if h * ratio > max_side_len:
+            ratio = float(max_side_len) / resize_h
+        resize_h = int(resize_h * ratio)
+        resize_w = int(resize_w * ratio)
+
+        max_stride = 128
+        resize_h = (resize_h + max_stride - 1) // max_stride * max_stride
+        resize_w = (resize_w + max_stride - 1) // max_stride * max_stride
+        im = cv2.resize(im, (int(resize_w), int(resize_h)))
+        ratio_h = resize_h / float(h)
+        ratio_w = resize_w / float(w)
+        return im, (ratio_h, ratio_w)
+
+    def resize_image(self, im, max_side_len=512):
+        """
+        resize image to a size multiple of max_stride which is required by the network
+        :param im: the resized image
+        :param max_side_len: limit of max image size to avoid out of memory in gpu
+        :return: the resized image and the resize ratio
+        """
+        h, w, _ = im.shape
+
+        resize_w = w
+        resize_h = h
+
+        # Fix the longer side
+        if resize_h > resize_w:
+            ratio = float(max_side_len) / resize_h
+        else:
+            ratio = float(max_side_len) / resize_w
+
+        resize_h = int(resize_h * ratio)
+        resize_w = int(resize_w * ratio)
+
+        max_stride = 128
+        resize_h = (resize_h + max_stride - 1) // max_stride * max_stride
+        resize_w = (resize_w + max_stride - 1) // max_stride * max_stride
+        im = cv2.resize(im, (int(resize_w), int(resize_h)))
+        ratio_h = resize_h / float(h)
+        ratio_w = resize_w / float(w)
+
+        return im, (ratio_h, ratio_w)
+
+
+class KieResize(object):
+    def __init__(self, **kwargs):
+        super(KieResize, self).__init__()
+        self.max_side, self.min_side = kwargs["img_scale"][0], kwargs["img_scale"][1]
+
+    def __call__(self, data):
+        img = data["image"]
+        points = data["points"]
+        src_h, src_w, _ = img.shape
+        (
+            im_resized,
+            scale_factor,
+            [ratio_h, ratio_w],
+            [new_h, new_w],
+        ) = self.resize_image(img)
+        resize_points = self.resize_boxes(img, points, scale_factor)
+        data["ori_image"] = img
+        data["ori_boxes"] = points
+        data["points"] = resize_points
+        data["image"] = im_resized
+        data["shape"] = np.array([new_h, new_w])
+        return data
+
+    def resize_image(self, img):
+        norm_img = np.zeros([1024, 1024, 3], dtype="float32")
+        scale = [512, 1024]
+        h, w = img.shape[:2]
+        max_long_edge = max(scale)
+        max_short_edge = min(scale)
+        scale_factor = min(max_long_edge / max(h, w), max_short_edge / min(h, w))
+        resize_w, resize_h = int(w * float(scale_factor) + 0.5), int(
+            h * float(scale_factor) + 0.5
+        )
+        max_stride = 32
+        resize_h = (resize_h + max_stride - 1) // max_stride * max_stride
+        resize_w = (resize_w + max_stride - 1) // max_stride * max_stride
+        im = cv2.resize(img, (resize_w, resize_h))
+        new_h, new_w = im.shape[:2]
+        w_scale = new_w / w
+        h_scale = new_h / h
+        scale_factor = np.array([w_scale, h_scale, w_scale, h_scale], dtype=np.float32)
+        norm_img[:new_h, :new_w, :] = im
+        return norm_img, scale_factor, [h_scale, w_scale], [new_h, new_w]
+
+    def resize_boxes(self, im, points, scale_factor):
+        points = points * scale_factor
+        img_shape = im.shape[:2]
+        points[:, 0::2] = np.clip(points[:, 0::2], 0, img_shape[1])
+        points[:, 1::2] = np.clip(points[:, 1::2], 0, img_shape[0])
+        return points
+
+
+class SRResize(object):
+    def __init__(
+        self,
+        imgH=32,
+        imgW=128,
+        down_sample_scale=4,
+        keep_ratio=False,
+        min_ratio=1,
+        mask=False,
+        infer_mode=False,
+        **kwargs,
+    ):
+        self.imgH = imgH
+        self.imgW = imgW
+        self.keep_ratio = keep_ratio
+        self.min_ratio = min_ratio
+        self.down_sample_scale = down_sample_scale
+        self.mask = mask
+        self.infer_mode = infer_mode
+
+    def __call__(self, data):
+        imgH = self.imgH
+        imgW = self.imgW
+        images_lr = data["image_lr"]
+        transform2 = ResizeNormalize(
+            (imgW // self.down_sample_scale, imgH // self.down_sample_scale)
+        )
+        images_lr = transform2(images_lr)
+        data["img_lr"] = images_lr
+        if self.infer_mode:
+            return data
+
+        images_HR = data["image_hr"]
+        label_strs = data["label"]
+        transform = ResizeNormalize((imgW, imgH))
+        images_HR = transform(images_HR)
+        data["img_hr"] = images_HR
+        return data
+
+
+class ResizeNormalize(object):
+    def __init__(self, size, interpolation=Image.BICUBIC):
+        self.size = size
+        self.interpolation = interpolation
+
+    def __call__(self, img):
+        img = img.resize(self.size, self.interpolation)
+        img_numpy = np.array(img).astype("float32")
+        img_numpy = img_numpy.transpose((2, 0, 1)) / 255
+        return img_numpy
+
+
+class GrayImageChannelFormat(object):
+    """
+    format gray scale image's channel: (3,h,w) -> (1,h,w)
+    Args:
+        inverse: inverse gray image
+    """
+
+    def __init__(self, inverse=False, **kwargs):
+        self.inverse = inverse
+
+    def __call__(self, data):
+        img = data["image"]
+        img_single_channel = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+        img_expanded = np.expand_dims(img_single_channel, 0)
+
+        if self.inverse:
+            data["image"] = np.abs(img_expanded - 1)
+        else:
+            data["image"] = img_expanded
+
+        data["src_image"] = img
+        return data
+
+
+class RandomPerspective(object):
+    """Random perspective transform for OCR detection training.
+
+    Operates on data["image"] (H,W,C) and data["polys"] (N, P, 2).
+    Should be placed after IaaAugment and before EastRandomCropData.
+    """
+
+    def __init__(
+        self,
+        prob=0.3,
+        degrees=0.0,
+        scale=0.2,
+        shear=5.0,
+        perspective=0.0002,
+        fit_output=True,
+        fill_value=(123.675, 116.28, 103.53),
+        min_area_ratio=0.1,
+        **kwargs,
+    ):
+        self.prob = prob
+        self.degrees = degrees
+        self.scale = scale
+        self.shear = shear
+        self.perspective = perspective
+        self.fit_output = fit_output
+        self.min_area_ratio = min_area_ratio
+        if isinstance(fill_value, (int, float)):
+            fill_value = (fill_value,) * 3
+        self.fill_value = tuple(fill_value)
+
+    def __call__(self, data):
+        if random.random() > self.prob:
+            return data
+
+        im = data["image"]
+        h, w = im.shape[:2]
+
+        # Build perspective matrix
+        M_core = self._get_core_matrix(h, w)
+
+        # Compute output bounds
+        corners = np.array([[0, 0], [w, 0], [w, h], [0, h]], dtype=np.float32).reshape(
+            -1, 1, 2
+        )
+        warped_corners = cv2.perspectiveTransform(corners, M_core)
+        x_min, y_min = warped_corners.min(axis=0).ravel()
+        x_max, y_max = warped_corners.max(axis=0).ravel()
+
+        if self.fit_output:
+            new_w = int(np.ceil(x_max) - np.floor(x_min))
+            new_h = int(np.ceil(y_max) - np.floor(y_min))
+            T_fit = np.eye(3, dtype=np.float32)
+            T_fit[0, 2] = -np.floor(x_min)
+            T_fit[1, 2] = -np.floor(y_min)
+            M = T_fit @ M_core
+            target_size = (new_w, new_h)
+        else:
+            T_orig = np.eye(3, dtype=np.float32)
+            T_orig[0, 2] = w / 2.0
+            T_orig[1, 2] = h / 2.0
+            M = T_orig @ M_core
+            target_size = (w, h)
+
+        # Warp image
+        transformed_im = cv2.warpPerspective(
+            im,
+            M,
+            target_size,
+            flags=cv2.INTER_LINEAR,
+            borderMode=cv2.BORDER_CONSTANT,
+            borderValue=self.fill_value,
+        )
+        data["image"] = transformed_im
+
+        # Transform polys
+        polys = data["polys"]
+        ignore_tags = data["ignore_tags"]
+        texts = data["texts"]
+
+        if len(polys) > 0:
+            # polys: (N, P, 2) — flatten to (N*P, 1, 2) for perspectiveTransform
+            n = len(polys)
+            points_per_poly = (
+                polys[0].shape[0] if hasattr(polys[0], "shape") else len(polys[0])
+            )
+            all_points = np.array(polys, dtype=np.float32).reshape(-1, 1, 2)
+
+            warped_pts = cv2.perspectiveTransform(all_points, M)
+            warped_pts = warped_pts.reshape(n, points_per_poly, 2)
+
+            # Compute original areas for filtering
+            orig_areas = np.array(
+                [
+                    cv2.contourArea(p.astype(np.float32))
+                    for p in np.array(polys, dtype=np.float32)
+                ]
+            )
+            new_areas = np.array(
+                [cv2.contourArea(p.astype(np.float32)) for p in warped_pts]
+            )
+
+            tw, th = target_size
+            # Filter: area ratio + center within bounds
+            centers_x = warped_pts[:, :, 0].mean(axis=1)
+            centers_y = warped_pts[:, :, 1].mean(axis=1)
+            valid = (
+                (new_areas > orig_areas * self.min_area_ratio)
+                & (centers_x > 0)
+                & (centers_x < tw)
+                & (centers_y > 0)
+                & (centers_y < th)
+            )
+
+            valid_ids = np.where(valid)[0]
+            data["polys"] = warped_pts[valid_ids]
+            data["ignore_tags"] = [ignore_tags[i] for i in valid_ids]
+            data["texts"] = [texts[i] for i in valid_ids]
+
+        return data
+
+    def _get_core_matrix(self, h, w):
+        C = np.eye(3, dtype=np.float32)
+        C[0, 2] = -w / 2
+        C[1, 2] = -h / 2
+
+        # Normalize perspective coefficients using 640px as reference size
+        # to ensure consistent distortion ratio across different image sizes
+        ref_size = 640.0
+        max_dim = max(h, w)
+        p_normalized = self.perspective * (ref_size / max_dim)
+
+        P = np.eye(3, dtype=np.float32)
+        P[2, 0] = random.uniform(-p_normalized, p_normalized)
+        P[2, 1] = random.uniform(-p_normalized, p_normalized)
+
+        s = random.uniform(1 - self.scale, 1 + self.scale)
+        a = random.uniform(-self.degrees, self.degrees)
+        R = np.eye(3, dtype=np.float32)
+        R[:2] = cv2.getRotationMatrix2D(angle=a, center=(0, 0), scale=s)
+
+        S = np.eye(3, dtype=np.float32)
+        S[0, 1] = math.tan(random.uniform(-self.shear, self.shear) * math.pi / 180)
+        S[1, 0] = math.tan(random.uniform(-self.shear, self.shear) * math.pi / 180)
+
+        return S @ R @ P @ C
diff --git a/ppocr/data/imaug/pg_process.py b/ppocr/data/imaug/pg_process.py
new file mode 100644
index 0000000..b86590d
--- /dev/null
+++ b/ppocr/data/imaug/pg_process.py
@@ -0,0 +1,1116 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import math
+import cv2
+import numpy as np
+from skimage.morphology._skeletonize import thin
+from ppocr.utils.e2e_utils.extract_textpoint_fast import (
+    sort_and_expand_with_direction_v2,
+)
+
+__all__ = ["PGProcessTrain"]
+
+
+class PGProcessTrain(object):
+    def __init__(
+        self,
+        character_dict_path,
+        max_text_length,
+        max_text_nums,
+        tcl_len,
+        batch_size=14,
+        use_resize=True,
+        use_random_crop=False,
+        min_crop_size=24,
+        min_text_size=4,
+        max_text_size=512,
+        point_gather_mode=None,
+        **kwargs,
+    ):
+        self.tcl_len = tcl_len
+        self.max_text_length = max_text_length
+        self.max_text_nums = max_text_nums
+        self.batch_size = batch_size
+        if use_random_crop is True:
+            self.min_crop_size = min_crop_size
+            self.use_random_crop = use_random_crop
+        self.min_text_size = min_text_size
+        self.max_text_size = max_text_size
+        self.use_resize = use_resize
+        self.point_gather_mode = point_gather_mode
+        self.Lexicon_Table = self.get_dict(character_dict_path)
+        self.pad_num = len(self.Lexicon_Table)
+        self.img_id = 0
+
+    def get_dict(self, character_dict_path):
+        character_str = ""
+        with open(character_dict_path, "rb") as fin:
+            lines = fin.readlines()
+            for line in lines:
+                line = line.decode("utf-8").strip("\n").strip("\r\n")
+                character_str += line
+            dict_character = list(character_str)
+        return dict_character
+
+    def quad_area(self, poly):
+        """
+        compute area of a polygon
+        :param poly:
+        :return:
+        """
+        edge = [
+            (poly[1][0] - poly[0][0]) * (poly[1][1] + poly[0][1]),
+            (poly[2][0] - poly[1][0]) * (poly[2][1] + poly[1][1]),
+            (poly[3][0] - poly[2][0]) * (poly[3][1] + poly[2][1]),
+            (poly[0][0] - poly[3][0]) * (poly[0][1] + poly[3][1]),
+        ]
+        return np.sum(edge) / 2.0
+
+    def gen_quad_from_poly(self, poly):
+        """
+        Generate min area quad from poly.
+        """
+        point_num = poly.shape[0]
+        min_area_quad = np.zeros((4, 2), dtype=np.float32)
+        rect = cv2.minAreaRect(
+            poly.astype(np.int32)
+        )  # (center (x,y), (width, height), angle of rotation)
+        box = np.array(cv2.boxPoints(rect))
+
+        first_point_idx = 0
+        min_dist = 1e4
+        for i in range(4):
+            dist = (
+                np.linalg.norm(box[(i + 0) % 4] - poly[0])
+                + np.linalg.norm(box[(i + 1) % 4] - poly[point_num // 2 - 1])
+                + np.linalg.norm(box[(i + 2) % 4] - poly[point_num // 2])
+                + np.linalg.norm(box[(i + 3) % 4] - poly[-1])
+            )
+            if dist < min_dist:
+                min_dist = dist
+                first_point_idx = i
+        for i in range(4):
+            min_area_quad[i] = box[(first_point_idx + i) % 4]
+
+        return min_area_quad
+
+    def check_and_validate_polys(self, polys, tags, im_size):
+        """
+        check so that the text poly is in the same direction,
+        and also filter some invalid polygons
+        :param polys:
+        :param tags:
+        :return:
+        """
+        (h, w) = im_size
+        if polys.shape[0] == 0:
+            return polys, np.array([]), np.array([])
+        polys[:, :, 0] = np.clip(polys[:, :, 0], 0, w - 1)
+        polys[:, :, 1] = np.clip(polys[:, :, 1], 0, h - 1)
+
+        validated_polys = []
+        validated_tags = []
+        hv_tags = []
+        for poly, tag in zip(polys, tags):
+            quad = self.gen_quad_from_poly(poly)
+            p_area = self.quad_area(quad)
+            if abs(p_area) < 1:
+                print("invalid poly")
+                continue
+            if p_area > 0:
+                if tag == False:
+                    print("poly in wrong direction")
+                    tag = True  # reversed cases should be ignore
+                poly = poly[(0, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1), :]
+                quad = quad[(0, 3, 2, 1), :]
+
+            len_w = np.linalg.norm(quad[0] - quad[1]) + np.linalg.norm(
+                quad[3] - quad[2]
+            )
+            len_h = np.linalg.norm(quad[0] - quad[3]) + np.linalg.norm(
+                quad[1] - quad[2]
+            )
+            hv_tag = 1
+
+            if len_w * 2.0 < len_h:
+                hv_tag = 0
+
+            validated_polys.append(poly)
+            validated_tags.append(tag)
+            hv_tags.append(hv_tag)
+        return np.array(validated_polys), np.array(validated_tags), np.array(hv_tags)
+
+    def crop_area(
+        self, im, polys, tags, hv_tags, txts, crop_background=False, max_tries=25
+    ):
+        """
+        make random crop from the input image
+        :param im:
+        :param polys:  [b,4,2]
+        :param tags:
+        :param crop_background:
+        :param max_tries: 50 -> 25
+        :return:
+        """
+        h, w, _ = im.shape
+        pad_h = h // 10
+        pad_w = w // 10
+        h_array = np.zeros((h + pad_h * 2), dtype=np.int32)
+        w_array = np.zeros((w + pad_w * 2), dtype=np.int32)
+        for poly in polys:
+            poly = np.round(poly, decimals=0).astype(np.int32)
+            minx = np.min(poly[:, 0])
+            maxx = np.max(poly[:, 0])
+            w_array[minx + pad_w : maxx + pad_w] = 1
+            miny = np.min(poly[:, 1])
+            maxy = np.max(poly[:, 1])
+            h_array[miny + pad_h : maxy + pad_h] = 1
+        # ensure the cropped area not across a text
+        h_axis = np.where(h_array == 0)[0]
+        w_axis = np.where(w_array == 0)[0]
+        if len(h_axis) == 0 or len(w_axis) == 0:
+            return im, polys, tags, hv_tags, txts
+        for i in range(max_tries):
+            xx = np.random.choice(w_axis, size=2)
+            xmin = np.min(xx) - pad_w
+            xmax = np.max(xx) - pad_w
+            xmin = np.clip(xmin, 0, w - 1)
+            xmax = np.clip(xmax, 0, w - 1)
+            yy = np.random.choice(h_axis, size=2)
+            ymin = np.min(yy) - pad_h
+            ymax = np.max(yy) - pad_h
+            ymin = np.clip(ymin, 0, h - 1)
+            ymax = np.clip(ymax, 0, h - 1)
+            if xmax - xmin < self.min_crop_size or ymax - ymin < self.min_crop_size:
+                continue
+            if polys.shape[0] != 0:
+                poly_axis_in_area = (
+                    (polys[:, :, 0] >= xmin)
+                    & (polys[:, :, 0] <= xmax)
+                    & (polys[:, :, 1] >= ymin)
+                    & (polys[:, :, 1] <= ymax)
+                )
+                selected_polys = np.where(np.sum(poly_axis_in_area, axis=1) == 4)[0]
+            else:
+                selected_polys = []
+            if len(selected_polys) == 0:
+                # no text in this area
+                if crop_background:
+                    txts_tmp = []
+                    for selected_poly in selected_polys:
+                        txts_tmp.append(txts[selected_poly])
+                    txts = txts_tmp
+                    return (
+                        im[ymin : ymax + 1, xmin : xmax + 1, :],
+                        polys[selected_polys],
+                        tags[selected_polys],
+                        hv_tags[selected_polys],
+                        txts,
+                    )
+                else:
+                    continue
+            im = im[ymin : ymax + 1, xmin : xmax + 1, :]
+            polys = polys[selected_polys]
+            tags = tags[selected_polys]
+            hv_tags = hv_tags[selected_polys]
+            txts_tmp = []
+            for selected_poly in selected_polys:
+                txts_tmp.append(txts[selected_poly])
+            txts = txts_tmp
+            polys[:, :, 0] -= xmin
+            polys[:, :, 1] -= ymin
+            return im, polys, tags, hv_tags, txts
+
+        return im, polys, tags, hv_tags, txts
+
+    def fit_and_gather_tcl_points_v2(
+        self,
+        min_area_quad,
+        poly,
+        max_h,
+        max_w,
+        fixed_point_num=64,
+        img_id=0,
+        reference_height=3,
+    ):
+        """
+        Find the center point of poly as key_points, then fit and gather.
+        """
+        key_point_xys = []
+        point_num = poly.shape[0]
+        for idx in range(point_num // 2):
+            center_point = (poly[idx] + poly[point_num - 1 - idx]) / 2.0
+            key_point_xys.append(center_point)
+
+        tmp_image = np.zeros(
+            shape=(
+                max_h,
+                max_w,
+            ),
+            dtype="float32",
+        )
+        cv2.polylines(tmp_image, [np.array(key_point_xys).astype("int32")], False, 1.0)
+        ys, xs = np.where(tmp_image > 0)
+        xy_text = np.array(list(zip(xs, ys)), dtype="float32")
+
+        left_center_pt = ((min_area_quad[0] + min_area_quad[3]) / 2.0).reshape(1, 2)
+        right_center_pt = ((min_area_quad[1] + min_area_quad[2]) / 2.0).reshape(1, 2)
+        proj_unit_vec = (right_center_pt - left_center_pt) / (
+            np.linalg.norm(right_center_pt - left_center_pt) + 1e-6
+        )
+        proj_unit_vec_tile = np.tile(proj_unit_vec, (xy_text.shape[0], 1))  # (n, 2)
+        left_center_pt_tile = np.tile(left_center_pt, (xy_text.shape[0], 1))  # (n, 2)
+        xy_text_to_left_center = xy_text - left_center_pt_tile
+        proj_value = np.sum(xy_text_to_left_center * proj_unit_vec_tile, axis=1)
+        xy_text = xy_text[np.argsort(proj_value)]
+
+        # convert to np and keep the num of point not greater then fixed_point_num
+        pos_info = np.array(xy_text).reshape(-1, 2)[:, ::-1]  # xy-> yx
+        point_num = len(pos_info)
+        if point_num > fixed_point_num:
+            keep_ids = [
+                int((point_num * 1.0 / fixed_point_num) * x)
+                for x in range(fixed_point_num)
+            ]
+            pos_info = pos_info[keep_ids, :]
+
+        keep = int(min(len(pos_info), fixed_point_num))
+        if np.random.rand() < 0.2 and reference_height >= 3:
+            dl = (np.random.rand(keep) - 0.5) * reference_height * 0.3
+            random_float = np.array([1, 0]).reshape([1, 2]) * dl.reshape([keep, 1])
+            pos_info += random_float
+            pos_info[:, 0] = np.clip(pos_info[:, 0], 0, max_h - 1)
+            pos_info[:, 1] = np.clip(pos_info[:, 1], 0, max_w - 1)
+
+        # padding to fixed length
+        pos_l = np.zeros((self.tcl_len, 3), dtype=np.int32)
+        pos_l[:, 0] = np.ones((self.tcl_len,)) * img_id
+        pos_m = np.zeros((self.tcl_len, 1), dtype=np.float32)
+        pos_l[:keep, 1:] = np.round(pos_info).astype(np.int32)
+        pos_m[:keep] = 1.0
+        return pos_l, pos_m
+
+    def fit_and_gather_tcl_points_v3(
+        self,
+        min_area_quad,
+        poly,
+        max_h,
+        max_w,
+        fixed_point_num=64,
+        img_id=0,
+        reference_height=3,
+    ):
+        """
+        Find the center point of poly as key_points, then fit and gather.
+        """
+        det_mask = np.zeros(
+            (int(max_h / self.ds_ratio), int(max_w / self.ds_ratio))
+        ).astype(np.float32)
+
+        # score_big_map
+        cv2.fillPoly(det_mask, np.round(poly / self.ds_ratio).astype(np.int32), 1.0)
+        det_mask = cv2.resize(det_mask, dsize=None, fx=self.ds_ratio, fy=self.ds_ratio)
+        det_mask = np.array(det_mask > 1e-3, dtype="float32")
+
+        f_direction = self.f_direction
+        skeleton_map = thin(det_mask.astype(np.uint8))
+        instance_count, instance_label_map = cv2.connectedComponents(
+            skeleton_map.astype(np.uint8), connectivity=8
+        )
+
+        ys, xs = np.where(instance_label_map == 1)
+        pos_list = list(zip(ys, xs))
+        if len(pos_list) < 3:
+            return None
+        pos_list_sorted = sort_and_expand_with_direction_v2(
+            pos_list, f_direction, det_mask
+        )
+
+        pos_list_sorted = np.array(pos_list_sorted)
+        length = len(pos_list_sorted) - 1
+        insert_num = 0
+        for index in range(length):
+            stride_y = np.abs(
+                pos_list_sorted[index + insert_num][0]
+                - pos_list_sorted[index + 1 + insert_num][0]
+            )
+            stride_x = np.abs(
+                pos_list_sorted[index + insert_num][1]
+                - pos_list_sorted[index + 1 + insert_num][1]
+            )
+            max_points = int(max(stride_x, stride_y))
+
+            stride = (
+                pos_list_sorted[index + insert_num]
+                - pos_list_sorted[index + 1 + insert_num]
+            ) / (max_points)
+            insert_num_temp = max_points - 1
+
+            for i in range(int(insert_num_temp)):
+                insert_value = pos_list_sorted[index + insert_num] - (i + 1) * stride
+                insert_index = index + i + 1 + insert_num
+                pos_list_sorted = np.insert(
+                    pos_list_sorted, insert_index, insert_value, axis=0
+                )
+            insert_num += insert_num_temp
+
+        pos_info = (
+            np.array(pos_list_sorted).reshape(-1, 2).astype(np.float32)
+        )  # xy-> yx
+
+        point_num = len(pos_info)
+        if point_num > fixed_point_num:
+            keep_ids = [
+                int((point_num * 1.0 / fixed_point_num) * x)
+                for x in range(fixed_point_num)
+            ]
+            pos_info = pos_info[keep_ids, :]
+
+        keep = int(min(len(pos_info), fixed_point_num))
+        reference_width = (
+            np.abs(poly[0, 0, 0] - poly[-1, 1, 0])
+            + np.abs(poly[0, 3, 0] - poly[-1, 2, 0])
+        ) // 2
+        if np.random.rand() < 1:
+            dh = (np.random.rand(keep) - 0.5) * reference_height
+            offset = np.random.rand() - 0.5
+            dw = np.array([[0, offset * reference_width * 0.2]])
+            random_float_h = np.array([1, 0]).reshape([1, 2]) * dh.reshape([keep, 1])
+            random_float_w = dw.repeat(keep, axis=0)
+            pos_info += random_float_h
+            pos_info += random_float_w
+            pos_info[:, 0] = np.clip(pos_info[:, 0], 0, max_h - 1)
+            pos_info[:, 1] = np.clip(pos_info[:, 1], 0, max_w - 1)
+
+        # padding to fixed length
+        pos_l = np.zeros((self.tcl_len, 3), dtype=np.int32)
+        pos_l[:, 0] = np.ones((self.tcl_len,)) * img_id
+        pos_m = np.zeros((self.tcl_len, 1), dtype=np.float32)
+        pos_l[:keep, 1:] = np.round(pos_info).astype(np.int32)
+        pos_m[:keep] = 1.0
+        return pos_l, pos_m
+
+    def generate_direction_map(self, poly_quads, n_char, direction_map):
+        """ """
+        width_list = []
+        height_list = []
+        for quad in poly_quads:
+            quad_w = (
+                np.linalg.norm(quad[0] - quad[1]) + np.linalg.norm(quad[2] - quad[3])
+            ) / 2.0
+            quad_h = (
+                np.linalg.norm(quad[0] - quad[3]) + np.linalg.norm(quad[2] - quad[1])
+            ) / 2.0
+            width_list.append(quad_w)
+            height_list.append(quad_h)
+        norm_width = max(sum(width_list) / n_char, 1.0)
+        average_height = max(sum(height_list) / len(height_list), 1.0)
+        k = 1
+        for quad in poly_quads:
+            direct_vector_full = ((quad[1] + quad[2]) - (quad[0] + quad[3])) / 2.0
+            direct_vector = (
+                direct_vector_full
+                / (np.linalg.norm(direct_vector_full) + 1e-6)
+                * norm_width
+            )
+            direction_label = tuple(
+                map(float, [direct_vector[0], direct_vector[1], 1.0 / average_height])
+            )
+            cv2.fillPoly(
+                direction_map,
+                quad.round().astype(np.int32)[np.newaxis, :, :],
+                direction_label,
+            )
+            k += 1
+        return direction_map
+
+    def calculate_average_height(self, poly_quads):
+        """ """
+        height_list = []
+        for quad in poly_quads:
+            quad_h = (
+                np.linalg.norm(quad[0] - quad[3]) + np.linalg.norm(quad[2] - quad[1])
+            ) / 2.0
+            height_list.append(quad_h)
+        average_height = max(sum(height_list) / len(height_list), 1.0)
+        return average_height
+
+    def generate_tcl_ctc_label(
+        self,
+        h,
+        w,
+        polys,
+        tags,
+        text_strs,
+        ds_ratio,
+        tcl_ratio=0.3,
+        shrink_ratio_of_width=0.15,
+    ):
+        """
+        Generate polygon.
+        """
+        self.ds_ratio = ds_ratio
+        score_map_big = np.zeros(
+            (
+                h,
+                w,
+            ),
+            dtype=np.float32,
+        )
+        h, w = int(h * ds_ratio), int(w * ds_ratio)
+        polys = polys * ds_ratio
+
+        score_map = np.zeros(
+            (
+                h,
+                w,
+            ),
+            dtype=np.float32,
+        )
+        score_label_map = np.zeros(
+            (
+                h,
+                w,
+            ),
+            dtype=np.float32,
+        )
+        tbo_map = np.zeros((h, w, 5), dtype=np.float32)
+        training_mask = np.ones(
+            (
+                h,
+                w,
+            ),
+            dtype=np.float32,
+        )
+        direction_map = np.ones((h, w, 3)) * np.array([0, 0, 1]).reshape(
+            [1, 1, 3]
+        ).astype(np.float32)
+
+        label_idx = 0
+        score_label_map_text_label_list = []
+        pos_list, pos_mask, label_list = [], [], []
+        for poly_idx, poly_tag in enumerate(zip(polys, tags)):
+            poly = poly_tag[0]
+            tag = poly_tag[1]
+
+            # generate min_area_quad
+            min_area_quad, center_point = self.gen_min_area_quad_from_poly(poly)
+            min_area_quad_h = 0.5 * (
+                np.linalg.norm(min_area_quad[0] - min_area_quad[3])
+                + np.linalg.norm(min_area_quad[1] - min_area_quad[2])
+            )
+            min_area_quad_w = 0.5 * (
+                np.linalg.norm(min_area_quad[0] - min_area_quad[1])
+                + np.linalg.norm(min_area_quad[2] - min_area_quad[3])
+            )
+
+            if (
+                min(min_area_quad_h, min_area_quad_w) < self.min_text_size * ds_ratio
+                or min(min_area_quad_h, min_area_quad_w) > self.max_text_size * ds_ratio
+            ):
+                continue
+
+            if tag:
+                cv2.fillPoly(
+                    training_mask, poly.astype(np.int32)[np.newaxis, :, :], 0.15
+                )
+            else:
+                text_label = text_strs[poly_idx]
+                text_label = self.prepare_text_label(text_label, self.Lexicon_Table)
+                text_label_index_list = [
+                    [self.Lexicon_Table.index(c_)]
+                    for c_ in text_label
+                    if c_ in self.Lexicon_Table
+                ]
+                if len(text_label_index_list) < 1:
+                    continue
+
+                tcl_poly = self.poly2tcl(poly, tcl_ratio)
+                tcl_quads = self.poly2quads(tcl_poly)
+                poly_quads = self.poly2quads(poly)
+
+                stcl_quads, quad_index = self.shrink_poly_along_width(
+                    tcl_quads,
+                    shrink_ratio_of_width=shrink_ratio_of_width,
+                    expand_height_ratio=1.0 / tcl_ratio,
+                )
+
+                cv2.fillPoly(score_map, np.round(stcl_quads).astype(np.int32), 1.0)
+                cv2.fillPoly(
+                    score_map_big, np.round(stcl_quads / ds_ratio).astype(np.int32), 1.0
+                )
+
+                for idx, quad in enumerate(stcl_quads):
+                    quad_mask = np.zeros((h, w), dtype=np.float32)
+                    quad_mask = cv2.fillPoly(
+                        quad_mask,
+                        np.round(quad[np.newaxis, :, :]).astype(np.int32),
+                        1.0,
+                    )
+                    tbo_map = self.gen_quad_tbo(
+                        poly_quads[quad_index[idx]], quad_mask, tbo_map
+                    )
+
+                # score label map and score_label_map_text_label_list for refine
+                if label_idx == 0:
+                    text_pos_list_ = [
+                        [len(self.Lexicon_Table)],
+                    ]
+                    score_label_map_text_label_list.append(text_pos_list_)
+
+                label_idx += 1
+                cv2.fillPoly(
+                    score_label_map, np.round(poly_quads).astype(np.int32), label_idx
+                )
+                score_label_map_text_label_list.append(text_label_index_list)
+
+                # direction info, fix-me
+                n_char = len(text_label_index_list)
+                direction_map = self.generate_direction_map(
+                    poly_quads, n_char, direction_map
+                )
+
+                # pos info
+                average_shrink_height = self.calculate_average_height(stcl_quads)
+
+                if self.point_gather_mode == "align":
+                    self.f_direction = direction_map[:, :, :-1].copy()
+                    pos_res = self.fit_and_gather_tcl_points_v3(
+                        min_area_quad,
+                        stcl_quads,
+                        max_h=h,
+                        max_w=w,
+                        fixed_point_num=64,
+                        img_id=self.img_id,
+                        reference_height=average_shrink_height,
+                    )
+                    if pos_res is None:
+                        continue
+                    pos_l, pos_m = pos_res[0], pos_res[1]
+
+                else:
+                    pos_l, pos_m = self.fit_and_gather_tcl_points_v2(
+                        min_area_quad,
+                        poly,
+                        max_h=h,
+                        max_w=w,
+                        fixed_point_num=64,
+                        img_id=self.img_id,
+                        reference_height=average_shrink_height,
+                    )
+
+                label_l = text_label_index_list
+                if len(text_label_index_list) < 2:
+                    continue
+
+                pos_list.append(pos_l)
+                pos_mask.append(pos_m)
+                label_list.append(label_l)
+
+        # use big score_map for smooth tcl lines
+        score_map_big_resized = cv2.resize(
+            score_map_big, dsize=None, fx=ds_ratio, fy=ds_ratio
+        )
+        score_map = np.array(score_map_big_resized > 1e-3, dtype="float32")
+
+        return (
+            score_map,
+            score_label_map,
+            tbo_map,
+            direction_map,
+            training_mask,
+            pos_list,
+            pos_mask,
+            label_list,
+            score_label_map_text_label_list,
+        )
+
+    def adjust_point(self, poly):
+        """
+        adjust point order.
+        """
+        point_num = poly.shape[0]
+        if point_num == 4:
+            len_1 = np.linalg.norm(poly[0] - poly[1])
+            len_2 = np.linalg.norm(poly[1] - poly[2])
+            len_3 = np.linalg.norm(poly[2] - poly[3])
+            len_4 = np.linalg.norm(poly[3] - poly[0])
+
+            if (len_1 + len_3) * 1.5 < (len_2 + len_4):
+                poly = poly[[1, 2, 3, 0], :]
+
+        elif point_num > 4:
+            vector_1 = poly[0] - poly[1]
+            vector_2 = poly[1] - poly[2]
+            cos_theta = np.dot(vector_1, vector_2) / (
+                np.linalg.norm(vector_1) * np.linalg.norm(vector_2) + 1e-6
+            )
+            theta = np.arccos(np.round(cos_theta, decimals=4))
+
+            if abs(theta) > (70 / 180 * math.pi):
+                index = list(range(1, point_num)) + [0]
+                poly = poly[np.array(index), :]
+        return poly
+
+    def gen_min_area_quad_from_poly(self, poly):
+        """
+        Generate min area quad from poly.
+        """
+        point_num = poly.shape[0]
+        min_area_quad = np.zeros((4, 2), dtype=np.float32)
+        if point_num == 4:
+            min_area_quad = poly
+            center_point = np.sum(poly, axis=0) / 4
+        else:
+            rect = cv2.minAreaRect(
+                poly.astype(np.int32)
+            )  # (center (x,y), (width, height), angle of rotation)
+            center_point = rect[0]
+            box = np.array(cv2.boxPoints(rect))
+
+            first_point_idx = 0
+            min_dist = 1e4
+            for i in range(4):
+                dist = (
+                    np.linalg.norm(box[(i + 0) % 4] - poly[0])
+                    + np.linalg.norm(box[(i + 1) % 4] - poly[point_num // 2 - 1])
+                    + np.linalg.norm(box[(i + 2) % 4] - poly[point_num // 2])
+                    + np.linalg.norm(box[(i + 3) % 4] - poly[-1])
+                )
+                if dist < min_dist:
+                    min_dist = dist
+                    first_point_idx = i
+
+            for i in range(4):
+                min_area_quad[i] = box[(first_point_idx + i) % 4]
+
+        return min_area_quad, center_point
+
+    def shrink_quad_along_width(self, quad, begin_width_ratio=0.0, end_width_ratio=1.0):
+        """
+        Generate shrink_quad_along_width.
+        """
+        ratio_pair = np.array(
+            [[begin_width_ratio], [end_width_ratio]], dtype=np.float32
+        )
+        p0_1 = quad[0] + (quad[1] - quad[0]) * ratio_pair
+        p3_2 = quad[3] + (quad[2] - quad[3]) * ratio_pair
+        return np.array([p0_1[0], p0_1[1], p3_2[1], p3_2[0]])
+
+    def shrink_poly_along_width(
+        self, quads, shrink_ratio_of_width, expand_height_ratio=1.0
+    ):
+        """
+        shrink poly with given length.
+        """
+        upper_edge_list = []
+
+        def get_cut_info(edge_len_list, cut_len):
+            for idx, edge_len in enumerate(edge_len_list):
+                cut_len -= edge_len
+                if cut_len <= 0.000001:
+                    ratio = (cut_len + edge_len_list[idx]) / edge_len_list[idx]
+                    return idx, ratio
+
+        for quad in quads:
+            upper_edge_len = np.linalg.norm(quad[0] - quad[1])
+            upper_edge_list.append(upper_edge_len)
+
+        # length of left edge and right edge.
+        left_length = np.linalg.norm(quads[0][0] - quads[0][3]) * expand_height_ratio
+        right_length = np.linalg.norm(quads[-1][1] - quads[-1][2]) * expand_height_ratio
+
+        shrink_length = (
+            min(left_length, right_length, sum(upper_edge_list)) * shrink_ratio_of_width
+        )
+        # shrinking length
+        upper_len_left = shrink_length
+        upper_len_right = sum(upper_edge_list) - shrink_length
+
+        left_idx, left_ratio = get_cut_info(upper_edge_list, upper_len_left)
+        left_quad = self.shrink_quad_along_width(
+            quads[left_idx], begin_width_ratio=left_ratio, end_width_ratio=1
+        )
+        right_idx, right_ratio = get_cut_info(upper_edge_list, upper_len_right)
+        right_quad = self.shrink_quad_along_width(
+            quads[right_idx], begin_width_ratio=0, end_width_ratio=right_ratio
+        )
+
+        out_quad_list = []
+        if left_idx == right_idx:
+            out_quad_list.append(
+                [left_quad[0], right_quad[1], right_quad[2], left_quad[3]]
+            )
+        else:
+            out_quad_list.append(left_quad)
+            for idx in range(left_idx + 1, right_idx):
+                out_quad_list.append(quads[idx])
+            out_quad_list.append(right_quad)
+
+        return np.array(out_quad_list), list(range(left_idx, right_idx + 1))
+
+    def prepare_text_label(self, label_str, Lexicon_Table):
+        """
+        Prepare text label by given Lexicon_Table.
+        """
+        if len(Lexicon_Table) == 36:
+            return label_str.lower()
+        else:
+            return label_str
+
+    def vector_angle(self, A, B):
+        """
+        Calculate the angle between vector AB and x-axis positive direction.
+        """
+        AB = np.array([B[1] - A[1], B[0] - A[0]])
+        return np.arctan2(*AB)
+
+    def theta_line_cross_point(self, theta, point):
+        """
+        Calculate the line through given point and angle in ax + by + c =0 form.
+        """
+        x, y = point
+        cos = np.cos(theta)
+        sin = np.sin(theta)
+        return [sin, -cos, cos * y - sin * x]
+
+    def line_cross_two_point(self, A, B):
+        """
+        Calculate the line through given point A and B in ax + by + c =0 form.
+        """
+        angle = self.vector_angle(A, B)
+        return self.theta_line_cross_point(angle, A)
+
+    def average_angle(self, poly):
+        """
+        Calculate the average angle between left and right edge in given poly.
+        """
+        p0, p1, p2, p3 = poly
+        angle30 = self.vector_angle(p3, p0)
+        angle21 = self.vector_angle(p2, p1)
+        return (angle30 + angle21) / 2
+
+    def line_cross_point(self, line1, line2):
+        """
+        line1 and line2 in  0=ax+by+c form, compute the cross point of line1 and line2
+        """
+        a1, b1, c1 = line1
+        a2, b2, c2 = line2
+        d = a1 * b2 - a2 * b1
+
+        if d == 0:
+            print("Cross point does not exist")
+            return np.array([0, 0], dtype=np.float32)
+        else:
+            x = (b1 * c2 - b2 * c1) / d
+            y = (a2 * c1 - a1 * c2) / d
+
+        return np.array([x, y], dtype=np.float32)
+
+    def quad2tcl(self, poly, ratio):
+        """
+        Generate center line by poly clock-wise point. (4, 2)
+        """
+        ratio_pair = np.array([[0.5 - ratio / 2], [0.5 + ratio / 2]], dtype=np.float32)
+        p0_3 = poly[0] + (poly[3] - poly[0]) * ratio_pair
+        p1_2 = poly[1] + (poly[2] - poly[1]) * ratio_pair
+        return np.array([p0_3[0], p1_2[0], p1_2[1], p0_3[1]])
+
+    def poly2tcl(self, poly, ratio):
+        """
+        Generate center line by poly clock-wise point.
+        """
+        ratio_pair = np.array([[0.5 - ratio / 2], [0.5 + ratio / 2]], dtype=np.float32)
+        tcl_poly = np.zeros_like(poly)
+        point_num = poly.shape[0]
+
+        for idx in range(point_num // 2):
+            point_pair = (
+                poly[idx] + (poly[point_num - 1 - idx] - poly[idx]) * ratio_pair
+            )
+            tcl_poly[idx] = point_pair[0]
+            tcl_poly[point_num - 1 - idx] = point_pair[1]
+        return tcl_poly
+
+    def gen_quad_tbo(self, quad, tcl_mask, tbo_map):
+        """
+        Generate tbo_map for give quad.
+        """
+        # upper and lower line function: ax + by + c = 0;
+        up_line = self.line_cross_two_point(quad[0], quad[1])
+        lower_line = self.line_cross_two_point(quad[3], quad[2])
+
+        quad_h = 0.5 * (
+            np.linalg.norm(quad[0] - quad[3]) + np.linalg.norm(quad[1] - quad[2])
+        )
+        quad_w = 0.5 * (
+            np.linalg.norm(quad[0] - quad[1]) + np.linalg.norm(quad[2] - quad[3])
+        )
+
+        # average angle of left and right line.
+        angle = self.average_angle(quad)
+
+        xy_in_poly = np.argwhere(tcl_mask == 1)
+        for y, x in xy_in_poly:
+            point = (x, y)
+            line = self.theta_line_cross_point(angle, point)
+            cross_point_upper = self.line_cross_point(up_line, line)
+            cross_point_lower = self.line_cross_point(lower_line, line)
+            ##FIX, offset reverse
+            upper_offset_x, upper_offset_y = cross_point_upper - point
+            lower_offset_x, lower_offset_y = cross_point_lower - point
+            tbo_map[y, x, 0] = upper_offset_y
+            tbo_map[y, x, 1] = upper_offset_x
+            tbo_map[y, x, 2] = lower_offset_y
+            tbo_map[y, x, 3] = lower_offset_x
+            tbo_map[y, x, 4] = 1.0 / max(min(quad_h, quad_w), 1.0) * 2
+        return tbo_map
+
+    def poly2quads(self, poly):
+        """
+        Split poly into quads.
+        """
+        quad_list = []
+        point_num = poly.shape[0]
+
+        # point pair
+        point_pair_list = []
+        for idx in range(point_num // 2):
+            point_pair = [poly[idx], poly[point_num - 1 - idx]]
+            point_pair_list.append(point_pair)
+
+        quad_num = point_num // 2 - 1
+        for idx in range(quad_num):
+            # reshape and adjust to clock-wise
+            quad_list.append(
+                (np.array(point_pair_list)[[idx, idx + 1]]).reshape(4, 2)[[0, 2, 3, 1]]
+            )
+
+        return np.array(quad_list)
+
+    def rotate_im_poly(self, im, text_polys):
+        """
+        rotate image with 90 / 180 / 270 degre
+        """
+        im_w, im_h = im.shape[1], im.shape[0]
+        dst_im = im.copy()
+        dst_polys = []
+        rand_degree_ratio = np.random.rand()
+        rand_degree_cnt = 1
+        if rand_degree_ratio > 0.5:
+            rand_degree_cnt = 3
+        for i in range(rand_degree_cnt):
+            dst_im = np.rot90(dst_im)
+        rot_degree = -90 * rand_degree_cnt
+        rot_angle = rot_degree * math.pi / 180.0
+        n_poly = text_polys.shape[0]
+        cx, cy = 0.5 * im_w, 0.5 * im_h
+        ncx, ncy = 0.5 * dst_im.shape[1], 0.5 * dst_im.shape[0]
+        for i in range(n_poly):
+            wordBB = text_polys[i]
+            poly = []
+            for j in range(4):  # 16->4
+                sx, sy = wordBB[j][0], wordBB[j][1]
+                dx = (
+                    math.cos(rot_angle) * (sx - cx)
+                    - math.sin(rot_angle) * (sy - cy)
+                    + ncx
+                )
+                dy = (
+                    math.sin(rot_angle) * (sx - cx)
+                    + math.cos(rot_angle) * (sy - cy)
+                    + ncy
+                )
+                poly.append([dx, dy])
+            dst_polys.append(poly)
+        return dst_im, np.array(dst_polys, dtype=np.float32)
+
+    def __call__(self, data):
+        input_size = 512
+        im = data["image"]
+        text_polys = data["polys"]
+        text_tags = data["ignore_tags"]
+        text_strs = data["texts"]
+        h, w, _ = im.shape
+        text_polys, text_tags, hv_tags = self.check_and_validate_polys(
+            text_polys, text_tags, (h, w)
+        )
+        if text_polys.shape[0] <= 0:
+            return None
+        # set aspect ratio and keep area fix
+        asp_scales = np.arange(1.0, 1.55, 0.1)
+        asp_scale = np.random.choice(asp_scales)
+        if np.random.rand() < 0.5:
+            asp_scale = 1.0 / asp_scale
+        asp_scale = math.sqrt(asp_scale)
+
+        asp_wx = asp_scale
+        asp_hy = 1.0 / asp_scale
+        im = cv2.resize(im, dsize=None, fx=asp_wx, fy=asp_hy)
+        text_polys[:, :, 0] *= asp_wx
+        text_polys[:, :, 1] *= asp_hy
+
+        if self.use_resize is True:
+            ori_h, ori_w, _ = im.shape
+            if max(ori_h, ori_w) < 200:
+                ratio = 200 / max(ori_h, ori_w)
+                im = cv2.resize(im, (int(ori_w * ratio), int(ori_h * ratio)))
+                text_polys[:, :, 0] *= ratio
+                text_polys[:, :, 1] *= ratio
+
+            if max(ori_h, ori_w) > 512:
+                ratio = 512 / max(ori_h, ori_w)
+                im = cv2.resize(im, (int(ori_w * ratio), int(ori_h * ratio)))
+                text_polys[:, :, 0] *= ratio
+                text_polys[:, :, 1] *= ratio
+        elif self.use_random_crop is True:
+            h, w, _ = im.shape
+            if max(h, w) > 2048:
+                rd_scale = 2048.0 / max(h, w)
+                im = cv2.resize(im, dsize=None, fx=rd_scale, fy=rd_scale)
+                text_polys *= rd_scale
+            h, w, _ = im.shape
+            if min(h, w) < 16:
+                return None
+
+            # no background
+            im, text_polys, text_tags, hv_tags, text_strs = self.crop_area(
+                im, text_polys, text_tags, hv_tags, text_strs, crop_background=False
+            )
+
+        if text_polys.shape[0] == 0:
+            return None
+        # continue for all ignore case
+        if np.sum((text_tags * 1.0)) >= text_tags.size:
+            return None
+        new_h, new_w, _ = im.shape
+        if (new_h is None) or (new_w is None):
+            return None
+        # resize image
+        std_ratio = float(input_size) / max(new_w, new_h)
+        rand_scales = np.array(
+            [0.25, 0.375, 0.5, 0.625, 0.75, 0.875, 1.0, 1.0, 1.0, 1.0, 1.0]
+        )
+        rz_scale = std_ratio * np.random.choice(rand_scales)
+        im = cv2.resize(im, dsize=None, fx=rz_scale, fy=rz_scale)
+        text_polys[:, :, 0] *= rz_scale
+        text_polys[:, :, 1] *= rz_scale
+
+        # add gaussian blur
+        if np.random.rand() < 0.1 * 0.5:
+            ks = np.random.permutation(5)[0] + 1
+            ks = int(ks / 2) * 2 + 1
+            im = cv2.GaussianBlur(im, ksize=(ks, ks), sigmaX=0, sigmaY=0)
+        # add brighter
+        if np.random.rand() < 0.1 * 0.5:
+            im = im * (1.0 + np.random.rand() * 0.5)
+            im = np.clip(im, 0.0, 255.0)
+        # add darker
+        if np.random.rand() < 0.1 * 0.5:
+            im = im * (1.0 - np.random.rand() * 0.5)
+            im = np.clip(im, 0.0, 255.0)
+
+        # Padding the im to [input_size, input_size]
+        new_h, new_w, _ = im.shape
+        if min(new_w, new_h) < input_size * 0.5:
+            return None
+        im_padded = np.ones((input_size, input_size, 3), dtype=np.float32)
+        im_padded[:, :, 2] = 0.485 * 255
+        im_padded[:, :, 1] = 0.456 * 255
+        im_padded[:, :, 0] = 0.406 * 255
+
+        # Random the start position
+        del_h = input_size - new_h
+        del_w = input_size - new_w
+        sh, sw = 0, 0
+        if del_h > 1:
+            sh = int(np.random.rand() * del_h)
+        if del_w > 1:
+            sw = int(np.random.rand() * del_w)
+
+        # Padding
+        im_padded[sh : sh + new_h, sw : sw + new_w, :] = im.copy()
+        text_polys[:, :, 0] += sw
+        text_polys[:, :, 1] += sh
+
+        (
+            score_map,
+            score_label_map,
+            border_map,
+            direction_map,
+            training_mask,
+            pos_list,
+            pos_mask,
+            label_list,
+            score_label_map_text_label,
+        ) = self.generate_tcl_ctc_label(
+            input_size, input_size, text_polys, text_tags, text_strs, 0.25
+        )
+        if len(label_list) <= 0:  # eliminate negative samples
+            return None
+        pos_list_temp = np.zeros([64, 3])
+        pos_mask_temp = np.zeros([64, 1])
+        label_list_temp = np.zeros([self.max_text_length, 1]) + self.pad_num
+
+        for i, label in enumerate(label_list):
+            n = len(label)
+            if n > self.max_text_length:
+                label_list[i] = label[: self.max_text_length]
+                continue
+            while n < self.max_text_length:
+                label.append([self.pad_num])
+                n += 1
+
+        for i in range(len(label_list)):
+            label_list[i] = np.array(label_list[i])
+
+        if len(pos_list) <= 0 or len(pos_list) > self.max_text_nums:
+            return None
+        for __ in range(self.max_text_nums - len(pos_list), 0, -1):
+            pos_list.append(pos_list_temp)
+            pos_mask.append(pos_mask_temp)
+            label_list.append(label_list_temp)
+
+        if self.img_id == self.batch_size - 1:
+            self.img_id = 0
+        else:
+            self.img_id += 1
+
+        im_padded[:, :, 2] -= 0.485 * 255
+        im_padded[:, :, 1] -= 0.456 * 255
+        im_padded[:, :, 0] -= 0.406 * 255
+        im_padded[:, :, 2] /= 255.0 * 0.229
+        im_padded[:, :, 1] /= 255.0 * 0.224
+        im_padded[:, :, 0] /= 255.0 * 0.225
+        im_padded = im_padded.transpose((2, 0, 1))
+        images = im_padded[::-1, :, :]
+        tcl_maps = score_map[np.newaxis, :, :]
+        tcl_label_maps = score_label_map[np.newaxis, :, :]
+        border_maps = border_map.transpose((2, 0, 1))
+        direction_maps = direction_map.transpose((2, 0, 1))
+        training_masks = training_mask[np.newaxis, :, :]
+        pos_list = np.array(pos_list)
+        pos_mask = np.array(pos_mask)
+        label_list = np.array(label_list)
+        data["images"] = images
+        data["tcl_maps"] = tcl_maps
+        data["tcl_label_maps"] = tcl_label_maps
+        data["border_maps"] = border_maps
+        data["direction_maps"] = direction_maps
+        data["training_masks"] = training_masks
+        data["label_list"] = label_list
+        data["pos_list"] = pos_list
+        data["pos_mask"] = pos_mask
+        return data
diff --git a/ppocr/data/imaug/randaugment.py b/ppocr/data/imaug/randaugment.py
new file mode 100644
index 0000000..588fb7b
--- /dev/null
+++ b/ppocr/data/imaug/randaugment.py
@@ -0,0 +1,134 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+from PIL import Image, ImageEnhance, ImageOps
+import numpy as np
+import random
+
+
+class RawRandAugment(object):
+    def __init__(self, num_layers=2, magnitude=5, fillcolor=(128, 128, 128), **kwargs):
+        self.num_layers = num_layers
+        self.magnitude = magnitude
+        self.max_level = 10
+
+        abso_level = self.magnitude / self.max_level
+        self.level_map = {
+            "shearX": 0.3 * abso_level,
+            "shearY": 0.3 * abso_level,
+            "translateX": 150.0 / 331 * abso_level,
+            "translateY": 150.0 / 331 * abso_level,
+            "rotate": 30 * abso_level,
+            "color": 0.9 * abso_level,
+            "posterize": int(4.0 * abso_level),
+            "solarize": 256.0 * abso_level,
+            "contrast": 0.9 * abso_level,
+            "sharpness": 0.9 * abso_level,
+            "brightness": 0.9 * abso_level,
+            "autocontrast": 0,
+            "equalize": 0,
+            "invert": 0,
+        }
+
+        # from https://stackoverflow.com/questions/5252170/
+        # specify-image-filling-color-when-rotating-in-python-with-pil-and-setting-expand
+        def rotate_with_fill(img, magnitude):
+            rot = img.convert("RGBA").rotate(magnitude)
+            return Image.composite(
+                rot, Image.new("RGBA", rot.size, (128,) * 4), rot
+            ).convert(img.mode)
+
+        rnd_ch_op = random.choice
+
+        self.func = {
+            "shearX": lambda img, magnitude: img.transform(
+                img.size,
+                Image.AFFINE,
+                (1, magnitude * rnd_ch_op([-1, 1]), 0, 0, 1, 0),
+                Image.BICUBIC,
+                fillcolor=fillcolor,
+            ),
+            "shearY": lambda img, magnitude: img.transform(
+                img.size,
+                Image.AFFINE,
+                (1, 0, 0, magnitude * rnd_ch_op([-1, 1]), 1, 0),
+                Image.BICUBIC,
+                fillcolor=fillcolor,
+            ),
+            "translateX": lambda img, magnitude: img.transform(
+                img.size,
+                Image.AFFINE,
+                (1, 0, magnitude * img.size[0] * rnd_ch_op([-1, 1]), 0, 1, 0),
+                fillcolor=fillcolor,
+            ),
+            "translateY": lambda img, magnitude: img.transform(
+                img.size,
+                Image.AFFINE,
+                (1, 0, 0, 0, 1, magnitude * img.size[1] * rnd_ch_op([-1, 1])),
+                fillcolor=fillcolor,
+            ),
+            "rotate": lambda img, magnitude: rotate_with_fill(img, magnitude),
+            "color": lambda img, magnitude: ImageEnhance.Color(img).enhance(
+                1 + magnitude * rnd_ch_op([-1, 1])
+            ),
+            "posterize": lambda img, magnitude: ImageOps.posterize(img, magnitude),
+            "solarize": lambda img, magnitude: ImageOps.solarize(img, magnitude),
+            "contrast": lambda img, magnitude: ImageEnhance.Contrast(img).enhance(
+                1 + magnitude * rnd_ch_op([-1, 1])
+            ),
+            "sharpness": lambda img, magnitude: ImageEnhance.Sharpness(img).enhance(
+                1 + magnitude * rnd_ch_op([-1, 1])
+            ),
+            "brightness": lambda img, magnitude: ImageEnhance.Brightness(img).enhance(
+                1 + magnitude * rnd_ch_op([-1, 1])
+            ),
+            "autocontrast": lambda img, magnitude: ImageOps.autocontrast(img),
+            "equalize": lambda img, magnitude: ImageOps.equalize(img),
+            "invert": lambda img, magnitude: ImageOps.invert(img),
+        }
+
+    def __call__(self, img):
+        avaiable_op_names = list(self.level_map.keys())
+        for layer_num in range(self.num_layers):
+            op_name = np.random.choice(avaiable_op_names)
+            img = self.func[op_name](img, self.level_map[op_name])
+        return img
+
+
+class RandAugment(RawRandAugment):
+    """RandAugment wrapper to auto fit different img types"""
+
+    def __init__(self, prob=0.5, *args, **kwargs):
+        self.prob = prob
+        super().__init__(*args, **kwargs)
+
+    def __call__(self, data):
+        if np.random.rand() > self.prob:
+            return data
+        img = data["image"]
+        if not isinstance(img, Image.Image):
+            img = np.ascontiguousarray(img)
+            img = Image.fromarray(img)
+
+        img = super().__call__(img)
+
+        if isinstance(img, Image.Image):
+            img = np.asarray(img)
+        data["image"] = img
+        return data
diff --git a/ppocr/data/imaug/random_crop_data.py b/ppocr/data/imaug/random_crop_data.py
new file mode 100644
index 0000000..bc61db7
--- /dev/null
+++ b/ppocr/data/imaug/random_crop_data.py
@@ -0,0 +1,577 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/WenmuZhou/DBNet.pytorch/blob/master/data_loader/modules/random_crop_data.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+import numpy as np
+import cv2
+import random
+
+from paddle import get_device
+
+from shapely.geometry import Polygon, box as shapely_box
+from shapely import intersection
+
+
+def is_poly_in_rect(poly, x, y, w, h):
+    poly = np.array(poly)
+    if poly[:, 0].min() < x or poly[:, 0].max() > x + w:
+        return False
+    if poly[:, 1].min() < y or poly[:, 1].max() > y + h:
+        return False
+    return True
+
+
+def is_poly_outside_rect(poly, x, y, w, h):
+    poly = np.array(poly)
+    if poly[:, 0].max() < x or poly[:, 0].min() > x + w:
+        return True
+    if poly[:, 1].max() < y or poly[:, 1].min() > y + h:
+        return True
+    return False
+
+
+def split_regions(axis):
+    regions = []
+    min_axis = 0
+    for i in range(1, axis.shape[0]):
+        if axis[i] != axis[i - 1] + 1:
+            region = axis[min_axis:i]
+            min_axis = i
+            regions.append(region)
+    return regions
+
+
+def random_select(axis, max_size):
+    xx = np.random.choice(axis, size=2)
+    xmin = np.min(xx)
+    xmax = np.max(xx)
+    xmin = np.clip(xmin, 0, max_size - 1)
+    xmax = np.clip(xmax, 0, max_size - 1)
+    return xmin, xmax
+
+
+def region_wise_random_select(regions, max_size):
+    selected_index = list(np.random.choice(len(regions), 2))
+    selected_values = []
+    for index in selected_index:
+        axis = regions[index]
+        xx = int(np.random.choice(axis, size=1))
+        selected_values.append(xx)
+    xmin = min(selected_values)
+    xmax = max(selected_values)
+    return xmin, xmax
+
+
+def get_min_rotated_rect_side(poly):
+    """
+    Compute the minimum side length of the minimum bounding rotated rectangle of a polygon.
+    """
+    poly = np.array(poly).astype(np.float32)
+    if len(poly) < 3:
+        return 0
+    rect = cv2.minAreaRect(poly)
+    width, height = rect[1]
+    return min(width, height)
+
+
+def get_min_quad_side(quad):
+    """
+    Compute the minimum side length of a quadrilateral.
+    """
+    if len(quad) != 4:
+        return 0
+    quad = np.array(quad)
+    sides = []
+    for i in range(4):
+        side = np.linalg.norm(quad[i] - quad[(i + 1) % 4])
+        sides.append(side)
+    return min(sides) if sides else 0
+
+
+def clip_poly_to_rect(poly, x, y, w, h):
+    """
+    Clip a polygon to a rectangular region and return the clipped quadrilateral.
+
+    Args:
+        poly: Original polygon vertices [[x1, y1], [x2, y2], ...]
+        x, y, w, h: Position and size of the clipping rectangle
+
+    Returns:
+        Clipped quadrilateral vertices, or None if the result is invalid.
+    """
+    try:
+        # Create polygon and clipping rectangle
+        poly_shape = Polygon(poly)
+        crop_rect = shapely_box(x, y, x + w, y + h)
+
+        # Compute intersection
+        clipped = intersection(poly_shape, crop_rect)
+
+        # No intersection or empty
+        if clipped.is_empty:
+            return None
+
+        # Get intersection coordinates
+        if clipped.geom_type == "Polygon":
+            coords = list(clipped.exterior.coords[:-1])  # Remove duplicate last point
+        elif clipped.geom_type == "MultiPolygon":
+            # Multiple polygons, select the largest by area
+            largest = max(clipped.geoms, key=lambda p: p.area)
+            coords = list(largest.exterior.coords[:-1])
+        elif clipped.geom_type == "GeometryCollection":
+            # Extract polygons from geometry collection
+            polygons = [g for g in clipped.geoms if g.geom_type == "Polygon"]
+            if not polygons:
+                return None
+            largest = max(polygons, key=lambda p: p.area)
+            coords = list(largest.exterior.coords[:-1])
+        else:
+            return None
+
+        # Less than 3 points, invalid
+        if len(coords) <= 3:
+            return None
+
+        # Convert to numpy array
+        coords = np.array(coords)
+
+        # Exactly 4 points, return directly
+        if len(coords) == 4:
+            return coords
+
+        # More than 4 points, use Douglas-Peucker to simplify to quadrilateral
+        # Output points are a subset of original coords (no out-of-bounds), IoU~0.99
+        if len(coords) > 4:
+            poly_cv = coords.reshape(-1, 1, 2).astype(np.float32)
+            peri = cv2.arcLength(poly_cv, True)
+            if peri < 1e-6:
+                return None
+            lo, hi = 0.0, 0.5
+            best = None
+            for _ in range(50):
+                mid = (lo + hi) / 2
+                approx = cv2.approxPolyDP(poly_cv, mid * peri, True)
+                if len(approx) <= 4:
+                    best = approx
+                    hi = mid
+                else:
+                    lo = mid
+            if best is not None and len(best) >= 3:
+                return best.reshape(-1, 2)
+            return None
+
+        return coords
+    except Exception as e:
+        return None
+
+
+class RandomCrop(object):
+    def __init__(
+        self,
+        size=(640, 640),
+        max_tries=10,
+        min_crop_side_ratio=0.1,
+        keep_ratio=True,
+        **kwargs,
+    ):
+        self.size = size
+        self.max_tries = max_tries
+        self.min_crop_side_ratio = min_crop_side_ratio
+        self.keep_ratio = keep_ratio
+
+    def __call__(self, data):
+        img = data["image"]
+        text_polys = data["polys"]
+        ignore_tags = data["ignore_tags"]
+        texts = data["texts"]
+
+        # Separate care and ignore text boxes
+        care_indices = [i for i, tag in enumerate(ignore_tags) if not tag]
+        all_care_polys = [text_polys[i] for i in care_indices]
+
+        h, w, _ = img.shape
+
+        # If no valid text boxes, still need to resize and pad the image
+        if len(all_care_polys) == 0:
+            # Use entire image as crop region, skip crop loop
+            crop_x, crop_y, crop_w, crop_h = 0, 0, w, h
+            valid_care_data = []
+        else:
+            # Pre-compute char heights (min side of min bounding rotated rect) for all care boxes
+            char_heights = np.array(
+                [get_min_rotated_rect_side(poly) for poly in all_care_polys]
+            )
+
+            # Try to find a suitable crop region
+            valid_care_data = []
+            for attempt in range(self.max_tries):
+                # Randomly determine crop region width and height
+                crop_w_min = min(int(w * self.min_crop_side_ratio), self.size[0])
+                crop_w_max = int(self.size[0] * 3)
+                crop_w = (
+                    w
+                    if crop_w_min >= crop_w_max
+                    else min(random.randint(crop_w_min, crop_w_max), w)
+                )
+
+                crop_h_min = min(int(h * self.min_crop_side_ratio), self.size[1])
+                crop_h_max = int(self.size[1] * 3)
+                crop_h = (
+                    h
+                    if crop_h_min >= crop_h_max
+                    else min(random.randint(crop_h_min, crop_h_max), h)
+                )
+
+                # Randomly determine crop region start position
+                crop_x = 0 if crop_w >= w else random.randint(0, w - crop_w)
+                crop_y = 0 if crop_h >= h else random.randint(0, h - crop_h)
+
+                # Check each care text box, clip and validate simultaneously (computed once)
+                valid_care_data = []
+                for care_idx, (poly, char_height) in enumerate(
+                    zip(all_care_polys, char_heights)
+                ):
+                    # Quick check: completely outside, skip
+                    if is_poly_outside_rect(poly, crop_x, crop_y, crop_w, crop_h):
+                        continue
+
+                    # Completely inside, no clipping needed
+                    if is_poly_in_rect(poly, crop_x, crop_y, crop_w, crop_h):
+                        valid_care_data.append(
+                            (care_idx, None)
+                        )  # None means no clipping needed
+                        continue
+
+                    # Truncated box, clip and validate (executed once)
+                    clipped_poly = clip_poly_to_rect(
+                        poly, crop_x, crop_y, crop_w, crop_h
+                    )
+                    if clipped_poly is None:
+                        continue
+
+                    # Validate clipped polygon - area check
+                    clipped_area = cv2.contourArea(clipped_poly.astype(np.float32))
+                    if clipped_area < 80:
+                        continue
+
+                    # Validate - char height check
+                    clipped_char_height = get_min_rotated_rect_side(clipped_poly)
+                    if clipped_char_height < char_height * 0.35:
+                        continue
+
+                    # Validate - min side length check (quadrilaterals only)
+                    if len(clipped_poly) == 4:
+                        min_side = get_min_quad_side(clipped_poly)
+                        if min_side < char_height * 0.35:
+                            continue
+
+                    # All validations passed, save the clipped polygon
+                    valid_care_data.append((care_idx, clipped_poly))
+
+                # At least one valid text box, use this crop region
+                if len(valid_care_data) >= 1:
+                    break
+            else:
+                # All attempts failed, use original region
+                crop_x, crop_y, crop_w, crop_h = 0, 0, w, h
+                valid_care_data = [(i, None) for i in range(len(all_care_polys))]
+
+        # Crop and scale image
+        # Only shrink when crop region is larger than target size, otherwise just pad
+        need_resize = crop_w > self.size[0] or crop_h > self.size[1]
+
+        if need_resize:
+            # Crop region larger than target, need to shrink
+            scale_w = self.size[0] / crop_w
+            scale_h = self.size[1] / crop_h
+            scale = min(scale_w, scale_h)
+            h_resized = int(crop_h * scale)
+            w_resized = int(crop_w * scale)
+        else:
+            # Crop region smaller than or equal to target, no upscaling
+            scale = 1.0
+            h_resized = crop_h
+            w_resized = crop_w
+
+        if self.keep_ratio:
+            # Random padding - compute padding size
+            pad_h = self.size[1] - h_resized
+            pad_w = self.size[0] - w_resized
+
+            # Randomly distribute padding to each side
+            pad_top = random.randint(0, pad_h) if pad_h > 0 else 0
+            pad_left = random.randint(0, pad_w) if pad_w > 0 else 0
+
+            # Resize cropped image (only when shrinking is needed)
+            cropped_img = img[crop_y : crop_y + crop_h, crop_x : crop_x + crop_w]
+            if need_resize:
+                resized_img = cv2.resize(cropped_img, (w_resized, h_resized))
+            else:
+                resized_img = cropped_img
+
+            # Create padded image
+            padimg = np.zeros((self.size[1], self.size[0], img.shape[2]), img.dtype)
+            padimg[pad_top : pad_top + h_resized, pad_left : pad_left + w_resized] = (
+                resized_img
+            )
+            img = padimg
+        else:
+            img = cv2.resize(
+                img[crop_y : crop_y + crop_h, crop_x : crop_x + crop_w],
+                tuple(self.size),
+            )
+            pad_left = 0
+            pad_top = 0
+
+        # Build fast lookup set of valid care indices
+        valid_care_indices_set = {care_idx for care_idx, _ in valid_care_data}
+
+        # Build mapping from care_idx to clipped polygon
+        care_idx_to_clipped = {
+            care_idx: clipped for care_idx, clipped in valid_care_data
+        }
+
+        # Build output text box list
+        text_polys_crop = []
+        ignore_tags_crop = []
+        texts_crop = []
+
+        for all_idx, (poly, text, tag) in enumerate(
+            zip(text_polys, texts, ignore_tags)
+        ):
+            if tag:
+                # Ignore text box, simple processing
+                if not is_poly_outside_rect(poly, crop_x, crop_y, crop_w, crop_h):
+                    adjusted_poly = (poly - (crop_x, crop_y)) * scale + (
+                        pad_left,
+                        pad_top,
+                    )
+                    adjusted_poly[:, 0] = np.clip(adjusted_poly[:, 0], 0, self.size[0])
+                    adjusted_poly[:, 1] = np.clip(adjusted_poly[:, 1], 0, self.size[1])
+                    text_polys_crop.append(adjusted_poly.tolist())
+                    ignore_tags_crop.append(tag)
+                    texts_crop.append(text)
+            else:
+                # Care text box, find corresponding care_idx
+                try:
+                    care_idx = care_indices.index(all_idx)
+                except ValueError:
+                    continue
+
+                # Check if this is a valid text box
+                if care_idx not in valid_care_indices_set:
+                    continue
+
+                # Get clipped polygon (if any)
+                clipped_poly = care_idx_to_clipped[care_idx]
+
+                if clipped_poly is None:
+                    # Completely inside, use original polygon
+                    adjusted_poly = (poly - (crop_x, crop_y)) * scale + (
+                        pad_left,
+                        pad_top,
+                    )
+                else:
+                    # Use clipped polygon
+                    adjusted_poly = (clipped_poly - (crop_x, crop_y)) * scale + (
+                        pad_left,
+                        pad_top,
+                    )
+
+                text_polys_crop.append(adjusted_poly.tolist())
+                ignore_tags_crop.append(tag)
+                texts_crop.append(text)
+        data["image"] = img
+        # Pad polygons to uniform point count to avoid inhomogeneous array error
+        if text_polys_crop:
+            max_points = max(len(p) for p in text_polys_crop)
+            for i, poly in enumerate(text_polys_crop):
+                if len(poly) < max_points:
+                    text_polys_crop[i] = poly + [poly[-1]] * (max_points - len(poly))
+        data["polys"] = np.array(text_polys_crop, dtype=np.float32)
+        data["ignore_tags"] = ignore_tags_crop
+        data["texts"] = texts_crop
+        return data
+
+
+def crop_area(im, text_polys, min_crop_side_ratio, max_tries):
+    h, w, _ = im.shape
+    h_array = np.zeros(h, dtype=np.int32)
+    w_array = np.zeros(w, dtype=np.int32)
+    for points in text_polys:
+        points = np.round(points, decimals=0).astype(np.int32)
+        minx = np.min(points[:, 0])
+        maxx = np.max(points[:, 0])
+        w_array[minx:maxx] = 1
+        miny = np.min(points[:, 1])
+        maxy = np.max(points[:, 1])
+        h_array[miny:maxy] = 1
+    # ensure the cropped area not across a text
+    h_axis = np.where(h_array == 0)[0]
+    w_axis = np.where(w_array == 0)[0]
+
+    if len(h_axis) == 0 or len(w_axis) == 0:
+        return 0, 0, w, h
+
+    h_regions = split_regions(h_axis)
+    w_regions = split_regions(w_axis)
+
+    for i in range(max_tries):
+        if len(w_regions) > 1:
+            xmin, xmax = region_wise_random_select(w_regions, w)
+        else:
+            xmin, xmax = random_select(w_axis, w)
+        if len(h_regions) > 1:
+            ymin, ymax = region_wise_random_select(h_regions, h)
+        else:
+            ymin, ymax = random_select(h_axis, h)
+
+        if (
+            xmax - xmin < min_crop_side_ratio * w
+            or ymax - ymin < min_crop_side_ratio * h
+        ):
+            # area too small
+            continue
+        num_poly_in_rect = 0
+        for poly in text_polys:
+            if not is_poly_outside_rect(poly, xmin, ymin, xmax - xmin, ymax - ymin):
+                num_poly_in_rect += 1
+                break
+
+        if num_poly_in_rect > 0:
+            return xmin, ymin, xmax - xmin, ymax - ymin
+
+    return 0, 0, w, h
+
+
+class EastRandomCropData(object):
+    def __init__(
+        self,
+        size=(640, 640),
+        max_tries=10,
+        min_crop_side_ratio=0.1,
+        keep_ratio=True,
+        **kwargs,
+    ):
+        self.size = size
+        self.max_tries = max_tries
+        self.min_crop_side_ratio = min_crop_side_ratio
+        self.keep_ratio = keep_ratio
+
+    def __call__(self, data):
+        img = data["image"]
+        text_polys = data["polys"]
+        ignore_tags = data["ignore_tags"]
+        texts = data["texts"]
+        all_care_polys = [text_polys[i] for i, tag in enumerate(ignore_tags) if not tag]
+        # 计算crop区域
+        crop_x, crop_y, crop_w, crop_h = crop_area(
+            img, all_care_polys, self.min_crop_side_ratio, self.max_tries
+        )
+        # crop 图片 保持比例填充
+        scale_w = self.size[0] / crop_w
+        scale_h = self.size[1] / crop_h
+        scale = min(scale_w, scale_h)
+        h = int(crop_h * scale)
+        w = int(crop_w * scale)
+        if self.keep_ratio:
+            padimg = np.zeros((self.size[1], self.size[0], img.shape[2]), img.dtype)
+            padimg[:h, :w] = cv2.resize(
+                img[crop_y : crop_y + crop_h, crop_x : crop_x + crop_w], (w, h)
+            )
+            img = padimg
+        else:
+            img = cv2.resize(
+                img[crop_y : crop_y + crop_h, crop_x : crop_x + crop_w],
+                tuple(self.size),
+            )
+        # crop 文本框
+        text_polys_crop = []
+        ignore_tags_crop = []
+        texts_crop = []
+        for poly, text, tag in zip(text_polys, texts, ignore_tags):
+            poly = ((poly - (crop_x, crop_y)) * scale).tolist()
+            if not is_poly_outside_rect(poly, 0, 0, w, h):
+                text_polys_crop.append(poly)
+                ignore_tags_crop.append(tag)
+                texts_crop.append(text)
+        data["image"] = img
+        data["polys"] = np.array(text_polys_crop)
+        if "iluvatar_gpu" in get_device():
+            data["polys"] = np.array(text_polys_crop).astype(np.float32)
+        data["ignore_tags"] = ignore_tags_crop
+        data["texts"] = texts_crop
+        return data
+
+
+class RandomCropImgMask(object):
+    def __init__(self, size, main_key, crop_keys, p=3 / 8, **kwargs):
+        self.size = size
+        self.main_key = main_key
+        self.crop_keys = crop_keys
+        self.p = p
+
+    def __call__(self, data):
+        image = data["image"]
+
+        h, w = image.shape[0:2]
+        th, tw = self.size
+        if w == tw and h == th:
+            return data
+
+        mask = data[self.main_key]
+        if np.max(mask) > 0 and random.random() > self.p:
+            # make sure to crop the text region
+            tl = np.min(np.where(mask > 0), axis=1) - (th, tw)
+            tl[tl < 0] = 0
+            br = np.max(np.where(mask > 0), axis=1) - (th, tw)
+            br[br < 0] = 0
+
+            br[0] = min(br[0], h - th)
+            br[1] = min(br[1], w - tw)
+
+            i = random.randint(tl[0], br[0]) if tl[0] < br[0] else 0
+            j = random.randint(tl[1], br[1]) if tl[1] < br[1] else 0
+        else:
+            i = random.randint(0, h - th) if h - th > 0 else 0
+            j = random.randint(0, w - tw) if w - tw > 0 else 0
+
+        # return i, j, th, tw
+        for k in data:
+            if k in self.crop_keys:
+                if len(data[k].shape) == 3:
+                    if np.argmin(data[k].shape) == 0:
+                        img = data[k][:, i : i + th, j : j + tw]
+                        if img.shape[1] != img.shape[2]:
+                            a = 1
+                    elif np.argmin(data[k].shape) == 2:
+                        img = data[k][i : i + th, j : j + tw, :]
+                        if img.shape[1] != img.shape[0]:
+                            a = 1
+                    else:
+                        img = data[k]
+                else:
+                    img = data[k][i : i + th, j : j + tw]
+                    if img.shape[0] != img.shape[1]:
+                        a = 1
+                data[k] = img
+        return data
diff --git a/ppocr/data/imaug/rec_img_aug.py b/ppocr/data/imaug/rec_img_aug.py
new file mode 100644
index 0000000..ad9b7d9
--- /dev/null
+++ b/ppocr/data/imaug/rec_img_aug.py
@@ -0,0 +1,947 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import math
+import cv2
+import numpy as np
+import random
+import copy
+from PIL import Image
+import PIL
+from .text_image_aug import tia_perspective, tia_stretch, tia_distort
+from .abinet_aug import (
+    CVGeometry,
+    CVDeterioration,
+    CVColorJitter,
+    SVTRGeometry,
+    SVTRDeterioration,
+    ParseQDeterioration,
+)
+from paddle import get_device
+from paddle.vision.transforms import Compose
+
+
+class RecAug(object):
+    def __init__(
+        self,
+        tia_prob=0.4,
+        crop_prob=0.4,
+        reverse_prob=0.4,
+        noise_prob=0.4,
+        jitter_prob=0.4,
+        blur_prob=0.4,
+        hsv_aug_prob=0.4,
+        **kwargs,
+    ):
+        self.tia_prob = tia_prob
+        self.bda = BaseDataAugmentation(
+            crop_prob, reverse_prob, noise_prob, jitter_prob, blur_prob, hsv_aug_prob
+        )
+
+    def __call__(self, data):
+        img = data["image"]
+        h, w, _ = img.shape
+
+        # tia
+        if random.random() <= self.tia_prob:
+            if h >= 20 and w >= 20:
+                img = tia_distort(img, random.randint(3, 6))
+                img = tia_stretch(img, random.randint(3, 6))
+                img = tia_perspective(img)
+
+        # bda
+        data["image"] = img
+        data = self.bda(data)
+        return data
+
+
+class BaseDataAugmentation(object):
+    def __init__(
+        self,
+        crop_prob=0.4,
+        reverse_prob=0.4,
+        noise_prob=0.4,
+        jitter_prob=0.4,
+        blur_prob=0.4,
+        hsv_aug_prob=0.4,
+        **kwargs,
+    ):
+        self.crop_prob = crop_prob
+        self.reverse_prob = reverse_prob
+        self.noise_prob = noise_prob
+        self.jitter_prob = jitter_prob
+        self.blur_prob = blur_prob
+        self.hsv_aug_prob = hsv_aug_prob
+        # for GaussianBlur
+        self.fil = cv2.getGaussianKernel(ksize=5, sigma=1, ktype=cv2.CV_32F)
+
+    def __call__(self, data):
+        img = data["image"]
+        h, w, _ = img.shape
+
+        if random.random() <= self.crop_prob and h >= 20 and w >= 20:
+            img = get_crop(img)
+
+        if random.random() <= self.blur_prob:
+            # GaussianBlur
+            img = cv2.sepFilter2D(img, -1, self.fil, self.fil)
+
+        if random.random() <= self.hsv_aug_prob:
+            img = hsv_aug(img)
+
+        if random.random() <= self.jitter_prob:
+            img = jitter(img)
+
+        if random.random() <= self.noise_prob:
+            img = add_gaussian_noise(img)
+
+        if random.random() <= self.reverse_prob:
+            img = 255 - img
+
+        data["image"] = img
+        return data
+
+
+class ABINetRecAug(object):
+    def __init__(
+        self, geometry_p=0.5, deterioration_p=0.25, colorjitter_p=0.25, **kwargs
+    ):
+        self.transforms = Compose(
+            [
+                CVGeometry(
+                    degrees=45,
+                    translate=(0.0, 0.0),
+                    scale=(0.5, 2.0),
+                    shear=(45, 15),
+                    distortion=0.5,
+                    p=geometry_p,
+                ),
+                CVDeterioration(var=20, degrees=6, factor=4, p=deterioration_p),
+                CVColorJitter(
+                    brightness=0.5,
+                    contrast=0.5,
+                    saturation=0.5,
+                    hue=0.1,
+                    p=colorjitter_p,
+                ),
+            ]
+        )
+
+    def __call__(self, data):
+        img = data["image"]
+        img = self.transforms(img)
+        data["image"] = img
+        return data
+
+
+class RecConAug(object):
+    def __init__(
+        self,
+        prob=0.5,
+        image_shape=(32, 320, 3),
+        max_text_length=25,
+        ext_data_num=1,
+        **kwargs,
+    ):
+        self.ext_data_num = ext_data_num
+        self.prob = prob
+        self.max_text_length = max_text_length
+        self.image_shape = image_shape
+        self.max_wh_ratio = self.image_shape[1] / self.image_shape[0]
+
+    def merge_ext_data(self, data, ext_data):
+        ori_w = round(
+            data["image"].shape[1] / data["image"].shape[0] * self.image_shape[0]
+        )
+        ext_w = round(
+            ext_data["image"].shape[1]
+            / ext_data["image"].shape[0]
+            * self.image_shape[0]
+        )
+        data["image"] = cv2.resize(data["image"], (ori_w, self.image_shape[0]))
+        ext_data["image"] = cv2.resize(ext_data["image"], (ext_w, self.image_shape[0]))
+        data["image"] = np.concatenate([data["image"], ext_data["image"]], axis=1)
+        data["label"] += ext_data["label"]
+        return data
+
+    def __call__(self, data):
+        rnd_num = random.random()
+        if rnd_num > self.prob:
+            return data
+        for idx, ext_data in enumerate(data["ext_data"]):
+            if len(data["label"]) + len(ext_data["label"]) > self.max_text_length:
+                break
+            concat_ratio = (
+                data["image"].shape[1] / data["image"].shape[0]
+                + ext_data["image"].shape[1] / ext_data["image"].shape[0]
+            )
+            if concat_ratio > self.max_wh_ratio:
+                break
+            data = self.merge_ext_data(data, ext_data)
+        data.pop("ext_data")
+        return data
+
+
+class SVTRRecAug(object):
+    def __init__(
+        self,
+        aug_type=0,
+        geometry_p=0.5,
+        deterioration_p=0.25,
+        colorjitter_p=0.25,
+        **kwargs,
+    ):
+        self.transforms = Compose(
+            [
+                SVTRGeometry(
+                    aug_type=aug_type,
+                    degrees=45,
+                    translate=(0.0, 0.0),
+                    scale=(0.5, 2.0),
+                    shear=(45, 15),
+                    distortion=0.5,
+                    p=geometry_p,
+                ),
+                SVTRDeterioration(var=20, degrees=6, factor=4, p=deterioration_p),
+                CVColorJitter(
+                    brightness=0.5,
+                    contrast=0.5,
+                    saturation=0.5,
+                    hue=0.1,
+                    p=colorjitter_p,
+                ),
+            ]
+        )
+
+    def __call__(self, data):
+        img = data["image"]
+        img = self.transforms(img)
+        data["image"] = img
+        return data
+
+
+class ParseQRecAug(object):
+    def __init__(
+        self,
+        aug_type=0,
+        geometry_p=0.5,
+        deterioration_p=0.25,
+        colorjitter_p=0.25,
+        **kwargs,
+    ):
+        self.transforms = Compose(
+            [
+                SVTRGeometry(
+                    aug_type=aug_type,
+                    degrees=45,
+                    translate=(0.0, 0.0),
+                    scale=(0.5, 2.0),
+                    shear=(45, 15),
+                    distortion=0.5,
+                    p=geometry_p,
+                ),
+                ParseQDeterioration(
+                    var=20, degrees=6, lam=20, radius=2.0, factor=4, p=deterioration_p
+                ),
+                CVColorJitter(
+                    brightness=0.5,
+                    contrast=0.5,
+                    saturation=0.5,
+                    hue=0.1,
+                    p=colorjitter_p,
+                ),
+            ]
+        )
+
+    def __call__(self, data):
+        img = data["image"]
+        img = self.transforms(img)
+        data["image"] = img
+        return data
+
+
+class ClsResizeImg(object):
+    def __init__(self, image_shape, **kwargs):
+        self.image_shape = image_shape
+
+    def __call__(self, data):
+        img = data["image"]
+        norm_img, _ = resize_norm_img(img, self.image_shape)
+        data["image"] = norm_img
+        return data
+
+
+class RecResizeImg(object):
+    def __init__(
+        self,
+        image_shape,
+        infer_mode=False,
+        eval_mode=False,
+        character_dict_path="./ppocr/utils/ppocr_keys_v1.txt",
+        padding=True,
+        **kwargs,
+    ):
+        self.image_shape = image_shape
+        self.infer_mode = infer_mode
+        self.eval_mode = eval_mode
+        self.character_dict_path = character_dict_path
+        self.padding = padding
+
+    def __call__(self, data):
+        img = data["image"]
+        if self.eval_mode or (self.infer_mode and self.character_dict_path is not None):
+            norm_img, valid_ratio = resize_norm_img_chinese(img, self.image_shape)
+        else:
+            norm_img, valid_ratio = resize_norm_img(img, self.image_shape, self.padding)
+        data["image"] = norm_img
+        data["valid_ratio"] = valid_ratio
+        if "iluvatar_gpu" in get_device():
+            data["valid_ratio"] = np.float32(valid_ratio)
+        return data
+
+
+class VLRecResizeImg(object):
+    def __init__(
+        self,
+        image_shape,
+        infer_mode=False,
+        character_dict_path="./ppocr/utils/ppocr_keys_v1.txt",
+        padding=True,
+        **kwargs,
+    ):
+        self.image_shape = image_shape
+        self.infer_mode = infer_mode
+        self.character_dict_path = character_dict_path
+        self.padding = padding
+
+    def __call__(self, data):
+        img = data["image"]
+
+        imgC, imgH, imgW = self.image_shape
+        resized_image = cv2.resize(img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
+        resized_w = imgW
+        resized_image = resized_image.astype("float32")
+        if self.image_shape[0] == 1:
+            resized_image = resized_image / 255
+            norm_img = resized_image[np.newaxis, :]
+        else:
+            norm_img = resized_image.transpose((2, 0, 1)) / 255
+        valid_ratio = min(1.0, float(resized_w / imgW))
+
+        data["image"] = norm_img
+        data["valid_ratio"] = valid_ratio
+        if "iluvatar_gpu" in get_device():
+            data["valid_ratio"] = np.float32(valid_ratio)
+        return data
+
+
+class RFLRecResizeImg(object):
+    def __init__(self, image_shape, padding=True, interpolation=1, **kwargs):
+        self.image_shape = image_shape
+        self.padding = padding
+
+        self.interpolation = interpolation
+        if self.interpolation == 0:
+            self.interpolation = cv2.INTER_NEAREST
+        elif self.interpolation == 1:
+            self.interpolation = cv2.INTER_LINEAR
+        elif self.interpolation == 2:
+            self.interpolation = cv2.INTER_CUBIC
+        elif self.interpolation == 3:
+            self.interpolation = cv2.INTER_AREA
+        else:
+            raise Exception("Unsupported interpolation type !!!")
+
+    def __call__(self, data):
+        img = data["image"]
+        img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+        norm_img, valid_ratio = resize_norm_img(
+            img, self.image_shape, self.padding, self.interpolation
+        )
+        data["image"] = norm_img
+        data["valid_ratio"] = valid_ratio
+        if "iluvatar_gpu" in get_device():
+            data["valid_ratio"] = np.float32(valid_ratio)
+        return data
+
+
+class SRNRecResizeImg(object):
+    def __init__(self, image_shape, num_heads, max_text_length, **kwargs):
+        self.image_shape = image_shape
+        self.num_heads = num_heads
+        self.max_text_length = max_text_length
+
+    def __call__(self, data):
+        img = data["image"]
+        norm_img = resize_norm_img_srn(img, self.image_shape)
+        data["image"] = norm_img
+        [
+            encoder_word_pos,
+            gsrm_word_pos,
+            gsrm_slf_attn_bias1,
+            gsrm_slf_attn_bias2,
+        ] = srn_other_inputs(self.image_shape, self.num_heads, self.max_text_length)
+
+        data["encoder_word_pos"] = encoder_word_pos
+        data["gsrm_word_pos"] = gsrm_word_pos
+        data["gsrm_slf_attn_bias1"] = gsrm_slf_attn_bias1
+        data["gsrm_slf_attn_bias2"] = gsrm_slf_attn_bias2
+        return data
+
+
+class SARRecResizeImg(object):
+    def __init__(self, image_shape, width_downsample_ratio=0.25, **kwargs):
+        self.image_shape = image_shape
+        self.width_downsample_ratio = width_downsample_ratio
+
+    def __call__(self, data):
+        img = data["image"]
+        norm_img, resize_shape, pad_shape, valid_ratio = resize_norm_img_sar(
+            img, self.image_shape, self.width_downsample_ratio
+        )
+        data["image"] = norm_img
+        data["resized_shape"] = resize_shape
+        data["pad_shape"] = pad_shape
+        data["valid_ratio"] = valid_ratio
+        if "iluvatar_gpu" in get_device():
+            data["valid_ratio"] = np.float32(valid_ratio)
+        return data
+
+
+class PRENResizeImg(object):
+    def __init__(self, image_shape, **kwargs):
+        """
+        According to original paper's realization, it's a hard resize method here.
+        So maybe you should optimize it to fit for your task better.
+        """
+        self.dst_h, self.dst_w = image_shape
+
+    def __call__(self, data):
+        img = data["image"]
+        resized_img = cv2.resize(
+            img, (self.dst_w, self.dst_h), interpolation=cv2.INTER_LINEAR
+        )
+        resized_img = resized_img.transpose((2, 0, 1)) / 255
+        resized_img -= 0.5
+        resized_img /= 0.5
+        data["image"] = resized_img.astype(np.float32)
+        return data
+
+
+class SPINRecResizeImg(object):
+    def __init__(
+        self,
+        image_shape,
+        interpolation=2,
+        mean=(127.5, 127.5, 127.5),
+        std=(127.5, 127.5, 127.5),
+        **kwargs,
+    ):
+        self.image_shape = image_shape
+
+        self.mean = np.array(mean, dtype=np.float32)
+        self.std = np.array(std, dtype=np.float32)
+        self.interpolation = interpolation
+
+    def __call__(self, data):
+        img = data["image"]
+        img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+        # different interpolation type corresponding the OpenCV
+        if self.interpolation == 0:
+            interpolation = cv2.INTER_NEAREST
+        elif self.interpolation == 1:
+            interpolation = cv2.INTER_LINEAR
+        elif self.interpolation == 2:
+            interpolation = cv2.INTER_CUBIC
+        elif self.interpolation == 3:
+            interpolation = cv2.INTER_AREA
+        else:
+            raise Exception("Unsupported interpolation type !!!")
+        # Deal with the image error during image loading
+        if img is None:
+            return None
+
+        img = cv2.resize(img, tuple(self.image_shape), interpolation)
+        img = np.array(img, np.float32)
+        img = np.expand_dims(img, -1)
+        img = img.transpose((2, 0, 1))
+        # normalize the image
+        img = img.copy().astype(np.float32)
+        mean = np.float64(self.mean.reshape(1, -1))
+        stdinv = 1 / np.float64(self.std.reshape(1, -1))
+        img -= mean
+        img *= stdinv
+        data["image"] = img
+        return data
+
+
+class GrayRecResizeImg(object):
+    def __init__(
+        self,
+        image_shape,
+        resize_type,
+        inter_type="Image.Resampling.LANCZOS",
+        scale=True,
+        padding=False,
+        **kwargs,
+    ):
+        self.image_shape = image_shape
+        self.resize_type = resize_type
+        self.padding = padding
+        self.inter_type = eval(inter_type)
+        self.scale = scale
+
+    def __call__(self, data):
+        img = data["image"]
+        img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+        image_shape = self.image_shape
+        if self.padding:
+            imgC, imgH, imgW = image_shape
+            # todo: change to 0 and modified image shape
+            h = img.shape[0]
+            w = img.shape[1]
+            ratio = w / float(h)
+            if math.ceil(imgH * ratio) > imgW:
+                resized_w = imgW
+            else:
+                resized_w = int(math.ceil(imgH * ratio))
+            resized_image = cv2.resize(img, (resized_w, imgH))
+            norm_img = np.expand_dims(resized_image, -1)
+            norm_img = norm_img.transpose((2, 0, 1))
+            resized_image = norm_img.astype(np.float32) / 128.0 - 1.0
+            padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
+            padding_im[:, :, 0:resized_w] = resized_image
+            data["image"] = padding_im
+            return data
+        if self.resize_type == "PIL":
+            image_pil = Image.fromarray(np.uint8(img))
+            img = image_pil.resize(self.image_shape, self.inter_type)
+            img = np.array(img)
+        if self.resize_type == "OpenCV":
+            img = cv2.resize(img, self.image_shape)
+        norm_img = np.expand_dims(img, -1)
+        norm_img = norm_img.transpose((2, 0, 1))
+        if self.scale:
+            data["image"] = norm_img.astype(np.float32) / 128.0 - 1.0
+        else:
+            data["image"] = norm_img.astype(np.float32) / 255.0
+        return data
+
+
+class ABINetRecResizeImg(object):
+    def __init__(self, image_shape, **kwargs):
+        self.image_shape = image_shape
+
+    def __call__(self, data):
+        img = data["image"]
+        norm_img, valid_ratio = resize_norm_img_abinet(img, self.image_shape)
+        data["image"] = norm_img
+        data["valid_ratio"] = valid_ratio
+        if "iluvatar_gpu" in get_device():
+            data["valid_ratio"] = np.float32(valid_ratio)
+        return data
+
+
+class SVTRRecResizeImg(object):
+    def __init__(self, image_shape, padding=True, **kwargs):
+        self.image_shape = image_shape
+        self.padding = padding
+
+    def __call__(self, data):
+        img = data["image"]
+
+        norm_img, valid_ratio = resize_norm_img(img, self.image_shape, self.padding)
+        data["image"] = norm_img
+        data["valid_ratio"] = valid_ratio
+        if "iluvatar_gpu" in get_device():
+            data["valid_ratio"] = np.float32(valid_ratio)
+        return data
+
+
+class RobustScannerRecResizeImg(object):
+    def __init__(
+        self, image_shape, max_text_length, width_downsample_ratio=0.25, **kwargs
+    ):
+        self.image_shape = image_shape
+        self.width_downsample_ratio = width_downsample_ratio
+        self.max_text_length = max_text_length
+
+    def __call__(self, data):
+        img = data["image"]
+        norm_img, resize_shape, pad_shape, valid_ratio = resize_norm_img_sar(
+            img, self.image_shape, self.width_downsample_ratio
+        )
+        word_positons = np.array(range(0, self.max_text_length)).astype("int64")
+        data["image"] = norm_img
+        data["resized_shape"] = resize_shape
+        data["pad_shape"] = pad_shape
+        data["valid_ratio"] = valid_ratio
+        if "iluvatar_gpu" in get_device():
+            data["valid_ratio"] = np.float32(valid_ratio)
+        data["word_positons"] = word_positons
+        return data
+
+
+def resize_norm_img_sar(img, image_shape, width_downsample_ratio=0.25):
+    imgC, imgH, imgW_min, imgW_max = image_shape
+    h = img.shape[0]
+    w = img.shape[1]
+    valid_ratio = 1.0
+    # make sure new_width is an integral multiple of width_divisor.
+    width_divisor = int(1 / width_downsample_ratio)
+    # resize
+    ratio = w / float(h)
+    resize_w = math.ceil(imgH * ratio)
+    if resize_w % width_divisor != 0:
+        resize_w = round(resize_w / width_divisor) * width_divisor
+    if imgW_min is not None:
+        resize_w = max(imgW_min, resize_w)
+    if imgW_max is not None:
+        valid_ratio = min(1.0, 1.0 * resize_w / imgW_max)
+        resize_w = min(imgW_max, resize_w)
+    resized_image = cv2.resize(img, (resize_w, imgH))
+    resized_image = resized_image.astype("float32")
+    # norm
+    if image_shape[0] == 1:
+        resized_image = resized_image / 255
+        resized_image = resized_image[np.newaxis, :]
+    else:
+        resized_image = resized_image.transpose((2, 0, 1)) / 255
+    resized_image -= 0.5
+    resized_image /= 0.5
+    resize_shape = resized_image.shape
+    padding_im = -1.0 * np.ones((imgC, imgH, imgW_max), dtype=np.float32)
+    padding_im[:, :, 0:resize_w] = resized_image
+    pad_shape = padding_im.shape
+
+    return padding_im, resize_shape, pad_shape, valid_ratio
+
+
+def resize_norm_img(img, image_shape, padding=True, interpolation=cv2.INTER_LINEAR):
+    imgC, imgH, imgW = image_shape
+    h = img.shape[0]
+    w = img.shape[1]
+    if not padding:
+        resized_image = cv2.resize(img, (imgW, imgH), interpolation=interpolation)
+        resized_w = imgW
+    else:
+        ratio = w / float(h)
+        if math.ceil(imgH * ratio) > imgW:
+            resized_w = imgW
+        else:
+            resized_w = int(math.ceil(imgH * ratio))
+        resized_image = cv2.resize(img, (resized_w, imgH))
+    resized_image = resized_image.astype("float32")
+    if image_shape[0] == 1:
+        resized_image = resized_image / 255
+        resized_image = resized_image[np.newaxis, :]
+    else:
+        resized_image = resized_image.transpose((2, 0, 1)) / 255
+    resized_image -= 0.5
+    resized_image /= 0.5
+    padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
+    padding_im[:, :, 0:resized_w] = resized_image
+    valid_ratio = min(1.0, float(resized_w / imgW))
+    return padding_im, valid_ratio
+
+
+def resize_norm_img_chinese(img, image_shape):
+    imgC, imgH, imgW = image_shape
+    # todo: change to 0 and modified image shape
+    max_wh_ratio = imgW * 1.0 / imgH
+    h, w = img.shape[0], img.shape[1]
+    ratio = w * 1.0 / h
+    max_wh_ratio = max(max_wh_ratio, ratio)
+    imgW = int(imgH * max_wh_ratio)
+    if math.ceil(imgH * ratio) > imgW:
+        resized_w = imgW
+    else:
+        resized_w = int(math.ceil(imgH * ratio))
+    resized_image = cv2.resize(img, (resized_w, imgH))
+    resized_image = resized_image.astype("float32")
+    if image_shape[0] == 1:
+        resized_image = resized_image / 255
+        resized_image = resized_image[np.newaxis, :]
+    else:
+        resized_image = resized_image.transpose((2, 0, 1)) / 255
+    resized_image -= 0.5
+    resized_image /= 0.5
+    padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
+    padding_im[:, :, 0:resized_w] = resized_image
+    valid_ratio = min(1.0, float(resized_w / imgW))
+    return padding_im, valid_ratio
+
+
+def resize_norm_img_srn(img, image_shape):
+    imgC, imgH, imgW = image_shape
+
+    img_black = np.zeros((imgH, imgW))
+    im_hei = img.shape[0]
+    im_wid = img.shape[1]
+
+    if im_wid <= im_hei * 1:
+        img_new = cv2.resize(img, (imgH * 1, imgH))
+    elif im_wid <= im_hei * 2:
+        img_new = cv2.resize(img, (imgH * 2, imgH))
+    elif im_wid <= im_hei * 3:
+        img_new = cv2.resize(img, (imgH * 3, imgH))
+    else:
+        img_new = cv2.resize(img, (imgW, imgH))
+
+    img_np = np.asarray(img_new)
+    img_np = cv2.cvtColor(img_np, cv2.COLOR_BGR2GRAY)
+    img_black[:, 0 : img_np.shape[1]] = img_np
+    img_black = img_black[:, :, np.newaxis]
+
+    row, col, c = img_black.shape
+    c = 1
+
+    return np.reshape(img_black, (c, row, col)).astype(np.float32)
+
+
+def resize_norm_img_abinet(img, image_shape):
+    imgC, imgH, imgW = image_shape
+
+    resized_image = cv2.resize(img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
+    resized_w = imgW
+    resized_image = resized_image.astype("float32")
+    resized_image = resized_image / 255.0
+
+    mean = np.array([0.485, 0.456, 0.406])
+    std = np.array([0.229, 0.224, 0.225])
+    resized_image = (resized_image - mean[None, None, ...]) / std[None, None, ...]
+    resized_image = resized_image.transpose((2, 0, 1))
+    resized_image = resized_image.astype("float32")
+
+    valid_ratio = min(1.0, float(resized_w / imgW))
+    return resized_image, valid_ratio
+
+
+def srn_other_inputs(image_shape, num_heads, max_text_length):
+    imgC, imgH, imgW = image_shape
+    feature_dim = int((imgH / 8) * (imgW / 8))
+
+    encoder_word_pos = (
+        np.array(range(0, feature_dim)).reshape((feature_dim, 1)).astype("int64")
+    )
+    gsrm_word_pos = (
+        np.array(range(0, max_text_length))
+        .reshape((max_text_length, 1))
+        .astype("int64")
+    )
+
+    gsrm_attn_bias_data = np.ones((1, max_text_length, max_text_length))
+    gsrm_slf_attn_bias1 = np.triu(gsrm_attn_bias_data, 1).reshape(
+        [1, max_text_length, max_text_length]
+    )
+    gsrm_slf_attn_bias1 = np.tile(gsrm_slf_attn_bias1, [num_heads, 1, 1]) * [-1e9]
+
+    gsrm_slf_attn_bias2 = np.tril(gsrm_attn_bias_data, -1).reshape(
+        [1, max_text_length, max_text_length]
+    )
+    gsrm_slf_attn_bias2 = np.tile(gsrm_slf_attn_bias2, [num_heads, 1, 1]) * [-1e9]
+
+    return [encoder_word_pos, gsrm_word_pos, gsrm_slf_attn_bias1, gsrm_slf_attn_bias2]
+
+
+def flag():
+    """
+    flag
+    """
+    return 1 if random.random() > 0.5000001 else -1
+
+
+def hsv_aug(img):
+    """
+    cvtColor
+    """
+    hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
+    delta = 0.001 * random.random() * flag()
+    hsv[:, :, 2] = hsv[:, :, 2] * (1 + delta)
+    new_img = cv2.cvtColor(hsv, cv2.COLOR_HSV2BGR)
+    return new_img
+
+
+def blur(img):
+    """
+    blur
+    """
+    h, w, _ = img.shape
+    if h > 10 and w > 10:
+        return cv2.GaussianBlur(img, (5, 5), 1)
+    else:
+        return img
+
+
+def jitter(img):
+    """
+    jitter
+    """
+    w, h, _ = img.shape
+    if h > 10 and w > 10:
+        thres = min(w, h)
+        s = int(random.random() * thres * 0.01)
+        src_img = img.copy()
+        for i in range(s):
+            img[i:, i:, :] = src_img[: w - i, : h - i, :]
+        return img
+    else:
+        return img
+
+
+def add_gaussian_noise(image, mean=0, var=0.1):
+    """
+    Gaussian noise
+    """
+
+    noise = np.random.normal(mean, var**0.5, image.shape)
+    out = image + 0.5 * noise
+    out = np.clip(out, 0, 255)
+    out = np.uint8(out)
+    return out
+
+
+def get_crop(image):
+    """
+    random crop
+    """
+    h, w, _ = image.shape
+    top_min = 1
+    top_max = 8
+    top_crop = int(random.randint(top_min, top_max))
+    top_crop = min(top_crop, h - 1)
+    crop_img = image.copy()
+    ratio = random.randint(0, 1)
+    if ratio:
+        crop_img = crop_img[top_crop:h, :, :]
+    else:
+        crop_img = crop_img[0 : h - top_crop, :, :]
+    return crop_img
+
+
+def rad(x):
+    """
+    rad
+    """
+    return x * np.pi / 180
+
+
+def get_warpR(config):
+    """
+    get_warpR
+    """
+    anglex, angley, anglez, fov, w, h, r = (
+        config.anglex,
+        config.angley,
+        config.anglez,
+        config.fov,
+        config.w,
+        config.h,
+        config.r,
+    )
+    if w > 69 and w < 112:
+        anglex = anglex * 1.5
+
+    z = np.sqrt(w**2 + h**2) / 2 / np.tan(rad(fov / 2))
+    # Homogeneous coordinate transformation matrix
+    rx = np.array(
+        [
+            [1, 0, 0, 0],
+            [0, np.cos(rad(anglex)), -np.sin(rad(anglex)), 0],
+            [
+                0,
+                -np.sin(rad(anglex)),
+                np.cos(rad(anglex)),
+                0,
+            ],
+            [0, 0, 0, 1],
+        ],
+        np.float32,
+    )
+    ry = np.array(
+        [
+            [np.cos(rad(angley)), 0, np.sin(rad(angley)), 0],
+            [0, 1, 0, 0],
+            [
+                -np.sin(rad(angley)),
+                0,
+                np.cos(rad(angley)),
+                0,
+            ],
+            [0, 0, 0, 1],
+        ],
+        np.float32,
+    )
+    rz = np.array(
+        [
+            [np.cos(rad(anglez)), np.sin(rad(anglez)), 0, 0],
+            [-np.sin(rad(anglez)), np.cos(rad(anglez)), 0, 0],
+            [0, 0, 1, 0],
+            [0, 0, 0, 1],
+        ],
+        np.float32,
+    )
+    r = rx.dot(ry).dot(rz)
+    # generate 4 points
+    pcenter = np.array([h / 2, w / 2, 0, 0], np.float32)
+    p1 = np.array([0, 0, 0, 0], np.float32) - pcenter
+    p2 = np.array([w, 0, 0, 0], np.float32) - pcenter
+    p3 = np.array([0, h, 0, 0], np.float32) - pcenter
+    p4 = np.array([w, h, 0, 0], np.float32) - pcenter
+    dst1 = r.dot(p1)
+    dst2 = r.dot(p2)
+    dst3 = r.dot(p3)
+    dst4 = r.dot(p4)
+    list_dst = np.array([dst1, dst2, dst3, dst4])
+    org = np.array([[0, 0], [w, 0], [0, h], [w, h]], np.float32)
+    dst = np.zeros((4, 2), np.float32)
+    # Project onto the image plane
+    dst[:, 0] = list_dst[:, 0] * z / (z - list_dst[:, 2]) + pcenter[0]
+    dst[:, 1] = list_dst[:, 1] * z / (z - list_dst[:, 2]) + pcenter[1]
+
+    warpR = cv2.getPerspectiveTransform(org, dst)
+
+    dst1, dst2, dst3, dst4 = dst
+    r1 = int(min(dst1[1], dst2[1]))
+    r2 = int(max(dst3[1], dst4[1]))
+    c1 = int(min(dst1[0], dst3[0]))
+    c2 = int(max(dst2[0], dst4[0]))
+
+    try:
+        ratio = min(1.0 * h / (r2 - r1), 1.0 * w / (c2 - c1))
+
+        dx = -c1
+        dy = -r1
+        T1 = np.float32([[1.0, 0, dx], [0, 1.0, dy], [0, 0, 1.0 / ratio]])
+        ret = T1.dot(warpR)
+    except:
+        ratio = 1.0
+        T1 = np.float32([[1.0, 0, 0], [0, 1.0, 0], [0, 0, 1.0]])
+        ret = T1
+    return ret, (-r1, -c1), ratio, dst
+
+
+def get_warpAffine(config):
+    """
+    get_warpAffine
+    """
+    anglez = config.anglez
+    rz = np.array(
+        [
+            [np.cos(rad(anglez)), np.sin(rad(anglez)), 0],
+            [-np.sin(rad(anglez)), np.cos(rad(anglez)), 0],
+        ],
+        np.float32,
+    )
+    return rz
diff --git a/ppocr/data/imaug/sast_process.py b/ppocr/data/imaug/sast_process.py
new file mode 100644
index 0000000..caa061b
--- /dev/null
+++ b/ppocr/data/imaug/sast_process.py
@@ -0,0 +1,810 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This part code is referred from:
+https://github.com/songdejia/EAST/blob/master/data_utils.py
+"""
+import math
+import cv2
+import numpy as np
+import json
+import sys
+import os
+
+__all__ = ["SASTProcessTrain"]
+
+
+class SASTProcessTrain(object):
+    def __init__(
+        self,
+        image_shape=[512, 512],
+        min_crop_size=24,
+        min_crop_side_ratio=0.3,
+        min_text_size=10,
+        max_text_size=512,
+        **kwargs,
+    ):
+        self.input_size = image_shape[1]
+        self.min_crop_size = min_crop_size
+        self.min_crop_side_ratio = min_crop_side_ratio
+        self.min_text_size = min_text_size
+        self.max_text_size = max_text_size
+
+    def quad_area(self, poly):
+        """
+        compute area of a polygon
+        :param poly:
+        :return:
+        """
+        edge = [
+            (poly[1][0] - poly[0][0]) * (poly[1][1] + poly[0][1]),
+            (poly[2][0] - poly[1][0]) * (poly[2][1] + poly[1][1]),
+            (poly[3][0] - poly[2][0]) * (poly[3][1] + poly[2][1]),
+            (poly[0][0] - poly[3][0]) * (poly[0][1] + poly[3][1]),
+        ]
+        return np.sum(edge) / 2.0
+
+    def gen_quad_from_poly(self, poly):
+        """
+        Generate min area quad from poly.
+        """
+        point_num = poly.shape[0]
+        min_area_quad = np.zeros((4, 2), dtype=np.float32)
+        if True:
+            rect = cv2.minAreaRect(
+                poly.astype(np.int32)
+            )  # (center (x,y), (width, height), angle of rotation)
+            center_point = rect[0]
+            box = np.array(cv2.boxPoints(rect))
+
+            first_point_idx = 0
+            min_dist = 1e4
+            for i in range(4):
+                dist = (
+                    np.linalg.norm(box[(i + 0) % 4] - poly[0])
+                    + np.linalg.norm(box[(i + 1) % 4] - poly[point_num // 2 - 1])
+                    + np.linalg.norm(box[(i + 2) % 4] - poly[point_num // 2])
+                    + np.linalg.norm(box[(i + 3) % 4] - poly[-1])
+                )
+                if dist < min_dist:
+                    min_dist = dist
+                    first_point_idx = i
+            for i in range(4):
+                min_area_quad[i] = box[(first_point_idx + i) % 4]
+
+        return min_area_quad
+
+    def check_and_validate_polys(self, polys, tags, xxx_todo_changeme):
+        """
+        check so that the text poly is in the same direction,
+        and also filter some invalid polygons
+        :param polys:
+        :param tags:
+        :return:
+        """
+        (h, w) = xxx_todo_changeme
+        if polys.shape[0] == 0:
+            return polys, np.array([]), np.array([])
+        polys[:, :, 0] = np.clip(polys[:, :, 0], 0, w - 1)
+        polys[:, :, 1] = np.clip(polys[:, :, 1], 0, h - 1)
+
+        validated_polys = []
+        validated_tags = []
+        hv_tags = []
+        for poly, tag in zip(polys, tags):
+            quad = self.gen_quad_from_poly(poly)
+            p_area = self.quad_area(quad)
+            if abs(p_area) < 1:
+                print("invalid poly")
+                continue
+            if p_area > 0:
+                if tag == False:
+                    print("poly in wrong direction")
+                    tag = True  # reversed cases should be ignore
+                poly = poly[(0, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1), :]
+                quad = quad[(0, 3, 2, 1), :]
+
+            len_w = np.linalg.norm(quad[0] - quad[1]) + np.linalg.norm(
+                quad[3] - quad[2]
+            )
+            len_h = np.linalg.norm(quad[0] - quad[3]) + np.linalg.norm(
+                quad[1] - quad[2]
+            )
+            hv_tag = 1
+
+            if len_w * 2.0 < len_h:
+                hv_tag = 0
+
+            validated_polys.append(poly)
+            validated_tags.append(tag)
+            hv_tags.append(hv_tag)
+        return np.array(validated_polys), np.array(validated_tags), np.array(hv_tags)
+
+    def crop_area(self, im, polys, tags, hv_tags, crop_background=False, max_tries=25):
+        """
+        make random crop from the input image
+        :param im:
+        :param polys:
+        :param tags:
+        :param crop_background:
+        :param max_tries: 50 -> 25
+        :return:
+        """
+        h, w, _ = im.shape
+        pad_h = h // 10
+        pad_w = w // 10
+        h_array = np.zeros((h + pad_h * 2), dtype=np.int32)
+        w_array = np.zeros((w + pad_w * 2), dtype=np.int32)
+        for poly in polys:
+            poly = np.round(poly, decimals=0).astype(np.int32)
+            minx = np.min(poly[:, 0])
+            maxx = np.max(poly[:, 0])
+            w_array[minx + pad_w : maxx + pad_w] = 1
+            miny = np.min(poly[:, 1])
+            maxy = np.max(poly[:, 1])
+            h_array[miny + pad_h : maxy + pad_h] = 1
+        # ensure the cropped area not across a text
+        h_axis = np.where(h_array == 0)[0]
+        w_axis = np.where(w_array == 0)[0]
+        if len(h_axis) == 0 or len(w_axis) == 0:
+            return im, polys, tags, hv_tags
+        for i in range(max_tries):
+            xx = np.random.choice(w_axis, size=2)
+            xmin = np.min(xx) - pad_w
+            xmax = np.max(xx) - pad_w
+            xmin = np.clip(xmin, 0, w - 1)
+            xmax = np.clip(xmax, 0, w - 1)
+            yy = np.random.choice(h_axis, size=2)
+            ymin = np.min(yy) - pad_h
+            ymax = np.max(yy) - pad_h
+            ymin = np.clip(ymin, 0, h - 1)
+            ymax = np.clip(ymax, 0, h - 1)
+            # if xmax - xmin < ARGS.min_crop_side_ratio * w or \
+            #   ymax - ymin < ARGS.min_crop_side_ratio * h:
+            if xmax - xmin < self.min_crop_size or ymax - ymin < self.min_crop_size:
+                # area too small
+                continue
+            if polys.shape[0] != 0:
+                poly_axis_in_area = (
+                    (polys[:, :, 0] >= xmin)
+                    & (polys[:, :, 0] <= xmax)
+                    & (polys[:, :, 1] >= ymin)
+                    & (polys[:, :, 1] <= ymax)
+                )
+                selected_polys = np.where(np.sum(poly_axis_in_area, axis=1) == 4)[0]
+            else:
+                selected_polys = []
+            if len(selected_polys) == 0:
+                # no text in this area
+                if crop_background:
+                    return (
+                        im[ymin : ymax + 1, xmin : xmax + 1, :],
+                        polys[selected_polys],
+                        tags[selected_polys],
+                        hv_tags[selected_polys],
+                    )
+                else:
+                    continue
+            im = im[ymin : ymax + 1, xmin : xmax + 1, :]
+            polys = polys[selected_polys]
+            tags = tags[selected_polys]
+            hv_tags = hv_tags[selected_polys]
+            polys[:, :, 0] -= xmin
+            polys[:, :, 1] -= ymin
+            return im, polys, tags, hv_tags
+
+        return im, polys, tags, hv_tags
+
+    def generate_direction_map(self, poly_quads, direction_map):
+        """ """
+        width_list = []
+        height_list = []
+        for quad in poly_quads:
+            quad_w = (
+                np.linalg.norm(quad[0] - quad[1]) + np.linalg.norm(quad[2] - quad[3])
+            ) / 2.0
+            quad_h = (
+                np.linalg.norm(quad[0] - quad[3]) + np.linalg.norm(quad[2] - quad[1])
+            ) / 2.0
+            width_list.append(quad_w)
+            height_list.append(quad_h)
+        norm_width = max(sum(width_list) / (len(width_list) + 1e-6), 1.0)
+        average_height = max(sum(height_list) / (len(height_list) + 1e-6), 1.0)
+
+        for quad in poly_quads:
+            direct_vector_full = ((quad[1] + quad[2]) - (quad[0] + quad[3])) / 2.0
+            direct_vector = (
+                direct_vector_full
+                / (np.linalg.norm(direct_vector_full) + 1e-6)
+                * norm_width
+            )
+            direction_label = tuple(
+                map(
+                    float,
+                    [direct_vector[0], direct_vector[1], 1.0 / (average_height + 1e-6)],
+                )
+            )
+            cv2.fillPoly(
+                direction_map,
+                quad.round().astype(np.int32)[np.newaxis, :, :],
+                direction_label,
+            )
+        return direction_map
+
+    def calculate_average_height(self, poly_quads):
+        """ """
+        height_list = []
+        for quad in poly_quads:
+            quad_h = (
+                np.linalg.norm(quad[0] - quad[3]) + np.linalg.norm(quad[2] - quad[1])
+            ) / 2.0
+            height_list.append(quad_h)
+        average_height = max(sum(height_list) / len(height_list), 1.0)
+        return average_height
+
+    def generate_tcl_label(
+        self, hw, polys, tags, ds_ratio, tcl_ratio=0.3, shrink_ratio_of_width=0.15
+    ):
+        """
+        Generate polygon.
+        """
+        h, w = hw
+        h, w = int(h * ds_ratio), int(w * ds_ratio)
+        polys = polys * ds_ratio
+
+        score_map = np.zeros(
+            (
+                h,
+                w,
+            ),
+            dtype=np.float32,
+        )
+        tbo_map = np.zeros((h, w, 5), dtype=np.float32)
+        training_mask = np.ones(
+            (
+                h,
+                w,
+            ),
+            dtype=np.float32,
+        )
+        direction_map = np.ones((h, w, 3)) * np.array([0, 0, 1]).reshape(
+            [1, 1, 3]
+        ).astype(np.float32)
+
+        for poly_idx, poly_tag in enumerate(zip(polys, tags)):
+            poly = poly_tag[0]
+            tag = poly_tag[1]
+
+            # generate min_area_quad
+            min_area_quad, center_point = self.gen_min_area_quad_from_poly(poly)
+            min_area_quad_h = 0.5 * (
+                np.linalg.norm(min_area_quad[0] - min_area_quad[3])
+                + np.linalg.norm(min_area_quad[1] - min_area_quad[2])
+            )
+            min_area_quad_w = 0.5 * (
+                np.linalg.norm(min_area_quad[0] - min_area_quad[1])
+                + np.linalg.norm(min_area_quad[2] - min_area_quad[3])
+            )
+
+            if (
+                min(min_area_quad_h, min_area_quad_w) < self.min_text_size * ds_ratio
+                or min(min_area_quad_h, min_area_quad_w) > self.max_text_size * ds_ratio
+            ):
+                continue
+
+            if tag:
+                # continue
+                cv2.fillPoly(
+                    training_mask, poly.astype(np.int32)[np.newaxis, :, :], 0.15
+                )
+            else:
+                tcl_poly = self.poly2tcl(poly, tcl_ratio)
+                tcl_quads = self.poly2quads(tcl_poly)
+                poly_quads = self.poly2quads(poly)
+                # stcl map
+                stcl_quads, quad_index = self.shrink_poly_along_width(
+                    tcl_quads,
+                    shrink_ratio_of_width=shrink_ratio_of_width,
+                    expand_height_ratio=1.0 / tcl_ratio,
+                )
+                # generate tcl map
+                cv2.fillPoly(score_map, np.round(stcl_quads).astype(np.int32), 1.0)
+
+                # generate tbo map
+                for idx, quad in enumerate(stcl_quads):
+                    quad_mask = np.zeros((h, w), dtype=np.float32)
+                    quad_mask = cv2.fillPoly(
+                        quad_mask,
+                        np.round(quad[np.newaxis, :, :]).astype(np.int32),
+                        1.0,
+                    )
+                    tbo_map = self.gen_quad_tbo(
+                        poly_quads[quad_index[idx]], quad_mask, tbo_map
+                    )
+        return score_map, tbo_map, training_mask
+
+    def generate_tvo_and_tco(self, hw, polys, tags, tcl_ratio=0.3, ds_ratio=0.25):
+        """
+        Generate tcl map, tvo map and tbo map.
+        """
+        h, w = hw
+        h, w = int(h * ds_ratio), int(w * ds_ratio)
+        polys = polys * ds_ratio
+        poly_mask = np.zeros((h, w), dtype=np.float32)
+
+        tvo_map = np.ones((9, h, w), dtype=np.float32)
+        tvo_map[0:-1:2] = np.tile(np.arange(0, w), (h, 1))
+        tvo_map[1:-1:2] = np.tile(np.arange(0, w), (h, 1)).T
+        poly_tv_xy_map = np.zeros((8, h, w), dtype=np.float32)
+
+        # tco map
+        tco_map = np.ones((3, h, w), dtype=np.float32)
+        tco_map[0] = np.tile(np.arange(0, w), (h, 1))
+        tco_map[1] = np.tile(np.arange(0, w), (h, 1)).T
+        poly_tc_xy_map = np.zeros((2, h, w), dtype=np.float32)
+
+        poly_short_edge_map = np.ones((h, w), dtype=np.float32)
+
+        for poly, poly_tag in zip(polys, tags):
+            if poly_tag == True:
+                continue
+
+            # adjust point order for vertical poly
+            poly = self.adjust_point(poly)
+
+            # generate min_area_quad
+            min_area_quad, center_point = self.gen_min_area_quad_from_poly(poly)
+            min_area_quad_h = 0.5 * (
+                np.linalg.norm(min_area_quad[0] - min_area_quad[3])
+                + np.linalg.norm(min_area_quad[1] - min_area_quad[2])
+            )
+            min_area_quad_w = 0.5 * (
+                np.linalg.norm(min_area_quad[0] - min_area_quad[1])
+                + np.linalg.norm(min_area_quad[2] - min_area_quad[3])
+            )
+
+            # generate tcl map and text, 128 * 128
+            tcl_poly = self.poly2tcl(poly, tcl_ratio)
+
+            # generate poly_tv_xy_map
+            for idx in range(4):
+                cv2.fillPoly(
+                    poly_tv_xy_map[2 * idx],
+                    np.round(tcl_poly[np.newaxis, :, :]).astype(np.int32),
+                    float(min(max(min_area_quad[idx, 0], 0), w)),
+                )
+                cv2.fillPoly(
+                    poly_tv_xy_map[2 * idx + 1],
+                    np.round(tcl_poly[np.newaxis, :, :]).astype(np.int32),
+                    float(min(max(min_area_quad[idx, 1], 0), h)),
+                )
+
+            # generate poly_tc_xy_map
+            for idx in range(2):
+                cv2.fillPoly(
+                    poly_tc_xy_map[idx],
+                    np.round(tcl_poly[np.newaxis, :, :]).astype(np.int32),
+                    float(center_point[idx]),
+                )
+
+            # generate poly_short_edge_map
+            cv2.fillPoly(
+                poly_short_edge_map,
+                np.round(tcl_poly[np.newaxis, :, :]).astype(np.int32),
+                float(max(min(min_area_quad_h, min_area_quad_w), 1.0)),
+            )
+
+            # generate poly_mask and training_mask
+            cv2.fillPoly(
+                poly_mask, np.round(tcl_poly[np.newaxis, :, :]).astype(np.int32), 1
+            )
+
+        tvo_map *= poly_mask
+        tvo_map[:8] -= poly_tv_xy_map
+        tvo_map[-1] /= poly_short_edge_map
+        tvo_map = tvo_map.transpose((1, 2, 0))
+
+        tco_map *= poly_mask
+        tco_map[:2] -= poly_tc_xy_map
+        tco_map[-1] /= poly_short_edge_map
+        tco_map = tco_map.transpose((1, 2, 0))
+
+        return tvo_map, tco_map
+
+    def adjust_point(self, poly):
+        """
+        adjust point order.
+        """
+        point_num = poly.shape[0]
+        if point_num == 4:
+            len_1 = np.linalg.norm(poly[0] - poly[1])
+            len_2 = np.linalg.norm(poly[1] - poly[2])
+            len_3 = np.linalg.norm(poly[2] - poly[3])
+            len_4 = np.linalg.norm(poly[3] - poly[0])
+
+            if (len_1 + len_3) * 1.5 < (len_2 + len_4):
+                poly = poly[[1, 2, 3, 0], :]
+
+        elif point_num > 4:
+            vector_1 = poly[0] - poly[1]
+            vector_2 = poly[1] - poly[2]
+            cos_theta = np.dot(vector_1, vector_2) / (
+                np.linalg.norm(vector_1) * np.linalg.norm(vector_2) + 1e-6
+            )
+            theta = np.arccos(np.round(cos_theta, decimals=4))
+
+            if abs(theta) > (70 / 180 * math.pi):
+                index = list(range(1, point_num)) + [0]
+                poly = poly[np.array(index), :]
+        return poly
+
+    def gen_min_area_quad_from_poly(self, poly):
+        """
+        Generate min area quad from poly.
+        """
+        point_num = poly.shape[0]
+        min_area_quad = np.zeros((4, 2), dtype=np.float32)
+        if point_num == 4:
+            min_area_quad = poly
+            center_point = np.sum(poly, axis=0) / 4
+        else:
+            rect = cv2.minAreaRect(
+                poly.astype(np.int32)
+            )  # (center (x,y), (width, height), angle of rotation)
+            center_point = rect[0]
+            box = np.array(cv2.boxPoints(rect))
+
+            first_point_idx = 0
+            min_dist = 1e4
+            for i in range(4):
+                dist = (
+                    np.linalg.norm(box[(i + 0) % 4] - poly[0])
+                    + np.linalg.norm(box[(i + 1) % 4] - poly[point_num // 2 - 1])
+                    + np.linalg.norm(box[(i + 2) % 4] - poly[point_num // 2])
+                    + np.linalg.norm(box[(i + 3) % 4] - poly[-1])
+                )
+                if dist < min_dist:
+                    min_dist = dist
+                    first_point_idx = i
+
+            for i in range(4):
+                min_area_quad[i] = box[(first_point_idx + i) % 4]
+
+        return min_area_quad, center_point
+
+    def shrink_quad_along_width(self, quad, begin_width_ratio=0.0, end_width_ratio=1.0):
+        """
+        Generate shrink_quad_along_width.
+        """
+        ratio_pair = np.array(
+            [[begin_width_ratio], [end_width_ratio]], dtype=np.float32
+        )
+        p0_1 = quad[0] + (quad[1] - quad[0]) * ratio_pair
+        p3_2 = quad[3] + (quad[2] - quad[3]) * ratio_pair
+        return np.array([p0_1[0], p0_1[1], p3_2[1], p3_2[0]])
+
+    def shrink_poly_along_width(
+        self, quads, shrink_ratio_of_width, expand_height_ratio=1.0
+    ):
+        """
+        shrink poly with given length.
+        """
+        upper_edge_list = []
+
+        def get_cut_info(edge_len_list, cut_len):
+            for idx, edge_len in enumerate(edge_len_list):
+                cut_len -= edge_len
+                if cut_len <= 0.000001:
+                    ratio = (cut_len + edge_len_list[idx]) / edge_len_list[idx]
+                    return idx, ratio
+
+        for quad in quads:
+            upper_edge_len = np.linalg.norm(quad[0] - quad[1])
+            upper_edge_list.append(upper_edge_len)
+
+        # length of left edge and right edge.
+        left_length = np.linalg.norm(quads[0][0] - quads[0][3]) * expand_height_ratio
+        right_length = np.linalg.norm(quads[-1][1] - quads[-1][2]) * expand_height_ratio
+
+        shrink_length = (
+            min(left_length, right_length, sum(upper_edge_list)) * shrink_ratio_of_width
+        )
+        # shrinking length
+        upper_len_left = shrink_length
+        upper_len_right = sum(upper_edge_list) - shrink_length
+
+        left_idx, left_ratio = get_cut_info(upper_edge_list, upper_len_left)
+        left_quad = self.shrink_quad_along_width(
+            quads[left_idx], begin_width_ratio=left_ratio, end_width_ratio=1
+        )
+        right_idx, right_ratio = get_cut_info(upper_edge_list, upper_len_right)
+        right_quad = self.shrink_quad_along_width(
+            quads[right_idx], begin_width_ratio=0, end_width_ratio=right_ratio
+        )
+
+        out_quad_list = []
+        if left_idx == right_idx:
+            out_quad_list.append(
+                [left_quad[0], right_quad[1], right_quad[2], left_quad[3]]
+            )
+        else:
+            out_quad_list.append(left_quad)
+            for idx in range(left_idx + 1, right_idx):
+                out_quad_list.append(quads[idx])
+            out_quad_list.append(right_quad)
+
+        return np.array(out_quad_list), list(range(left_idx, right_idx + 1))
+
+    def vector_angle(self, A, B):
+        """
+        Calculate the angle between vector AB and x-axis positive direction.
+        """
+        AB = np.array([B[1] - A[1], B[0] - A[0]])
+        return np.arctan2(*AB)
+
+    def theta_line_cross_point(self, theta, point):
+        """
+        Calculate the line through given point and angle in ax + by + c =0 form.
+        """
+        x, y = point
+        cos = np.cos(theta)
+        sin = np.sin(theta)
+        return [sin, -cos, cos * y - sin * x]
+
+    def line_cross_two_point(self, A, B):
+        """
+        Calculate the line through given point A and B in ax + by + c =0 form.
+        """
+        angle = self.vector_angle(A, B)
+        return self.theta_line_cross_point(angle, A)
+
+    def average_angle(self, poly):
+        """
+        Calculate the average angle between left and right edge in given poly.
+        """
+        p0, p1, p2, p3 = poly
+        angle30 = self.vector_angle(p3, p0)
+        angle21 = self.vector_angle(p2, p1)
+        return (angle30 + angle21) / 2
+
+    def line_cross_point(self, line1, line2):
+        """
+        line1 and line2 in  0=ax+by+c form, compute the cross point of line1 and line2
+        """
+        a1, b1, c1 = line1
+        a2, b2, c2 = line2
+        d = a1 * b2 - a2 * b1
+
+        if d == 0:
+            # print("line1", line1)
+            # print("line2", line2)
+            print("Cross point does not exist")
+            return np.array([0, 0], dtype=np.float32)
+        else:
+            x = (b1 * c2 - b2 * c1) / d
+            y = (a2 * c1 - a1 * c2) / d
+
+        return np.array([x, y], dtype=np.float32)
+
+    def quad2tcl(self, poly, ratio):
+        """
+        Generate center line by poly clock-wise point. (4, 2)
+        """
+        ratio_pair = np.array([[0.5 - ratio / 2], [0.5 + ratio / 2]], dtype=np.float32)
+        p0_3 = poly[0] + (poly[3] - poly[0]) * ratio_pair
+        p1_2 = poly[1] + (poly[2] - poly[1]) * ratio_pair
+        return np.array([p0_3[0], p1_2[0], p1_2[1], p0_3[1]])
+
+    def poly2tcl(self, poly, ratio):
+        """
+        Generate center line by poly clock-wise point.
+        """
+        ratio_pair = np.array([[0.5 - ratio / 2], [0.5 + ratio / 2]], dtype=np.float32)
+        tcl_poly = np.zeros_like(poly)
+        point_num = poly.shape[0]
+
+        for idx in range(point_num // 2):
+            point_pair = (
+                poly[idx] + (poly[point_num - 1 - idx] - poly[idx]) * ratio_pair
+            )
+            tcl_poly[idx] = point_pair[0]
+            tcl_poly[point_num - 1 - idx] = point_pair[1]
+        return tcl_poly
+
+    def gen_quad_tbo(self, quad, tcl_mask, tbo_map):
+        """
+        Generate tbo_map for give quad.
+        """
+        # upper and lower line function: ax + by + c = 0;
+        up_line = self.line_cross_two_point(quad[0], quad[1])
+        lower_line = self.line_cross_two_point(quad[3], quad[2])
+
+        quad_h = 0.5 * (
+            np.linalg.norm(quad[0] - quad[3]) + np.linalg.norm(quad[1] - quad[2])
+        )
+        quad_w = 0.5 * (
+            np.linalg.norm(quad[0] - quad[1]) + np.linalg.norm(quad[2] - quad[3])
+        )
+
+        # average angle of left and right line.
+        angle = self.average_angle(quad)
+
+        xy_in_poly = np.argwhere(tcl_mask == 1)
+        for y, x in xy_in_poly:
+            point = (x, y)
+            line = self.theta_line_cross_point(angle, point)
+            cross_point_upper = self.line_cross_point(up_line, line)
+            cross_point_lower = self.line_cross_point(lower_line, line)
+            ##FIX, offset reverse
+            upper_offset_x, upper_offset_y = cross_point_upper - point
+            lower_offset_x, lower_offset_y = cross_point_lower - point
+            tbo_map[y, x, 0] = upper_offset_y
+            tbo_map[y, x, 1] = upper_offset_x
+            tbo_map[y, x, 2] = lower_offset_y
+            tbo_map[y, x, 3] = lower_offset_x
+            tbo_map[y, x, 4] = 1.0 / max(min(quad_h, quad_w), 1.0) * 2
+        return tbo_map
+
+    def poly2quads(self, poly):
+        """
+        Split poly into quads.
+        """
+        quad_list = []
+        point_num = poly.shape[0]
+
+        # point pair
+        point_pair_list = []
+        for idx in range(point_num // 2):
+            point_pair = [poly[idx], poly[point_num - 1 - idx]]
+            point_pair_list.append(point_pair)
+
+        quad_num = point_num // 2 - 1
+        for idx in range(quad_num):
+            # reshape and adjust to clock-wise
+            quad_list.append(
+                (np.array(point_pair_list)[[idx, idx + 1]]).reshape(4, 2)[[0, 2, 3, 1]]
+            )
+
+        return np.array(quad_list)
+
+    def __call__(self, data):
+        im = data["image"]
+        text_polys = data["polys"]
+        text_tags = data["ignore_tags"]
+        if im is None:
+            return None
+        if text_polys.shape[0] == 0:
+            return None
+
+        h, w, _ = im.shape
+        text_polys, text_tags, hv_tags = self.check_and_validate_polys(
+            text_polys, text_tags, (h, w)
+        )
+
+        if text_polys.shape[0] == 0:
+            return None
+
+        # set aspect ratio and keep area fix
+        asp_scales = np.arange(1.0, 1.55, 0.1)
+        asp_scale = np.random.choice(asp_scales)
+
+        if np.random.rand() < 0.5:
+            asp_scale = 1.0 / asp_scale
+        asp_scale = math.sqrt(asp_scale)
+
+        asp_wx = asp_scale
+        asp_hy = 1.0 / asp_scale
+        im = cv2.resize(im, dsize=None, fx=asp_wx, fy=asp_hy)
+        text_polys[:, :, 0] *= asp_wx
+        text_polys[:, :, 1] *= asp_hy
+
+        h, w, _ = im.shape
+        if max(h, w) > 2048:
+            rd_scale = 2048.0 / max(h, w)
+            im = cv2.resize(im, dsize=None, fx=rd_scale, fy=rd_scale)
+            text_polys *= rd_scale
+        h, w, _ = im.shape
+        if min(h, w) < 16:
+            return None
+
+        # no background
+        im, text_polys, text_tags, hv_tags = self.crop_area(
+            im, text_polys, text_tags, hv_tags, crop_background=False
+        )
+
+        if text_polys.shape[0] == 0:
+            return None
+        # continue for all ignore case
+        if np.sum((text_tags * 1.0)) >= text_tags.size:
+            return None
+        new_h, new_w, _ = im.shape
+        if (new_h is None) or (new_w is None):
+            return None
+        # resize image
+        std_ratio = float(self.input_size) / max(new_w, new_h)
+        rand_scales = np.array(
+            [0.25, 0.375, 0.5, 0.625, 0.75, 0.875, 1.0, 1.0, 1.0, 1.0, 1.0]
+        )
+        rz_scale = std_ratio * np.random.choice(rand_scales)
+        im = cv2.resize(im, dsize=None, fx=rz_scale, fy=rz_scale)
+        text_polys[:, :, 0] *= rz_scale
+        text_polys[:, :, 1] *= rz_scale
+
+        # add gaussian blur
+        if np.random.rand() < 0.1 * 0.5:
+            ks = np.random.permutation(5)[0] + 1
+            ks = int(ks / 2) * 2 + 1
+            im = cv2.GaussianBlur(im, ksize=(ks, ks), sigmaX=0, sigmaY=0)
+        # add brighter
+        if np.random.rand() < 0.1 * 0.5:
+            im = im * (1.0 + np.random.rand() * 0.5)
+            im = np.clip(im, 0.0, 255.0)
+        # add darker
+        if np.random.rand() < 0.1 * 0.5:
+            im = im * (1.0 - np.random.rand() * 0.5)
+            im = np.clip(im, 0.0, 255.0)
+
+        # Padding the im to [input_size, input_size]
+        new_h, new_w, _ = im.shape
+        if min(new_w, new_h) < self.input_size * 0.5:
+            return None
+
+        im_padded = np.ones((self.input_size, self.input_size, 3), dtype=np.float32)
+        im_padded[:, :, 2] = 0.485 * 255
+        im_padded[:, :, 1] = 0.456 * 255
+        im_padded[:, :, 0] = 0.406 * 255
+
+        # Random the start position
+        del_h = self.input_size - new_h
+        del_w = self.input_size - new_w
+        sh, sw = 0, 0
+        if del_h > 1:
+            sh = int(np.random.rand() * del_h)
+        if del_w > 1:
+            sw = int(np.random.rand() * del_w)
+
+        # Padding
+        im_padded[sh : sh + new_h, sw : sw + new_w, :] = im.copy()
+        text_polys[:, :, 0] += sw
+        text_polys[:, :, 1] += sh
+
+        score_map, border_map, training_mask = self.generate_tcl_label(
+            (self.input_size, self.input_size), text_polys, text_tags, 0.25
+        )
+
+        # SAST head
+        tvo_map, tco_map = self.generate_tvo_and_tco(
+            (self.input_size, self.input_size),
+            text_polys,
+            text_tags,
+            tcl_ratio=0.3,
+            ds_ratio=0.25,
+        )
+        # print("test--------tvo_map shape:", tvo_map.shape)
+
+        im_padded[:, :, 2] -= 0.485 * 255
+        im_padded[:, :, 1] -= 0.456 * 255
+        im_padded[:, :, 0] -= 0.406 * 255
+        im_padded[:, :, 2] /= 255.0 * 0.229
+        im_padded[:, :, 1] /= 255.0 * 0.224
+        im_padded[:, :, 0] /= 255.0 * 0.225
+        im_padded = im_padded.transpose((2, 0, 1))
+
+        data["image"] = im_padded[::-1, :, :]
+        data["score_map"] = score_map[np.newaxis, :, :]
+        data["border_map"] = border_map.transpose((2, 0, 1))
+        data["training_mask"] = training_mask[np.newaxis, :, :]
+        data["tvo_map"] = tvo_map.transpose((2, 0, 1))
+        data["tco_map"] = tco_map.transpose((2, 0, 1))
+        return data
diff --git a/ppocr/data/imaug/ssl_img_aug.py b/ppocr/data/imaug/ssl_img_aug.py
new file mode 100644
index 0000000..8162087
--- /dev/null
+++ b/ppocr/data/imaug/ssl_img_aug.py
@@ -0,0 +1,55 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import math
+import cv2
+import numpy as np
+import random
+from PIL import Image
+
+from .rec_img_aug import resize_norm_img
+
+
+class SSLRotateResize(object):
+    def __init__(
+        self, image_shape, padding=False, select_all=True, mode="train", **kwargs
+    ):
+        self.image_shape = image_shape
+        self.padding = padding
+        self.select_all = select_all
+        self.mode = mode
+
+    def __call__(self, data):
+        img = data["image"]
+
+        data["image_r90"] = cv2.rotate(img, cv2.ROTATE_90_CLOCKWISE)
+        data["image_r180"] = cv2.rotate(data["image_r90"], cv2.ROTATE_90_CLOCKWISE)
+        data["image_r270"] = cv2.rotate(data["image_r180"], cv2.ROTATE_90_CLOCKWISE)
+
+        images = []
+        for key in ["image", "image_r90", "image_r180", "image_r270"]:
+            images.append(
+                resize_norm_img(
+                    data.pop(key), image_shape=self.image_shape, padding=self.padding
+                )[0]
+            )
+        data["image"] = np.stack(images, axis=0)
+        data["label"] = np.array(list(range(4)))
+        if not self.select_all:
+            data["image"] = data["image"][0::2]  # just choose 0 and 180
+            data["label"] = data["label"][0:2]  # label needs to be continuous
+        if self.mode == "test":
+            data["image"] = data["image"][0]
+            data["label"] = data["label"][0]
+        return data
diff --git a/ppocr/data/imaug/table_ops.py b/ppocr/data/imaug/table_ops.py
new file mode 100644
index 0000000..cc244dd
--- /dev/null
+++ b/ppocr/data/imaug/table_ops.py
@@ -0,0 +1,232 @@
+"""
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+import sys
+import cv2
+import numpy as np
+
+
+class GenTableMask(object):
+    """gen table mask"""
+
+    def __init__(self, shrink_h_max, shrink_w_max, mask_type=0, **kwargs):
+        self.shrink_h_max = 5
+        self.shrink_w_max = 5
+        self.mask_type = mask_type
+
+    def projection(self, erosion, h, w, spilt_threshold=0):
+        # 水平投影
+        projection_map = np.ones_like(erosion)
+        project_val_array = [0 for _ in range(0, h)]
+
+        for j in range(0, h):
+            for i in range(0, w):
+                if erosion[j, i] == 255:
+                    project_val_array[j] += 1
+        # 根据数组，获取切割点
+        start_idx = 0  # 记录进入字符区的索引
+        end_idx = 0  # 记录进入空白区域的索引
+        in_text = False  # 是否遍历到了字符区内
+        box_list = []
+        for i in range(len(project_val_array)):
+            if (
+                in_text == False and project_val_array[i] > spilt_threshold
+            ):  # 进入字符区了
+                in_text = True
+                start_idx = i
+            elif (
+                project_val_array[i] <= spilt_threshold and in_text == True
+            ):  # 进入空白区了
+                end_idx = i
+                in_text = False
+                if end_idx - start_idx <= 2:
+                    continue
+                box_list.append((start_idx, end_idx + 1))
+
+        if in_text:
+            box_list.append((start_idx, h - 1))
+        # 绘制投影直方图
+        for j in range(0, h):
+            for i in range(0, project_val_array[j]):
+                projection_map[j, i] = 0
+        return box_list, projection_map
+
+    def projection_cx(self, box_img):
+        box_gray_img = cv2.cvtColor(box_img, cv2.COLOR_BGR2GRAY)
+        h, w = box_gray_img.shape
+        # 灰度图片进行二值化处理
+        ret, thresh1 = cv2.threshold(box_gray_img, 200, 255, cv2.THRESH_BINARY_INV)
+        # 纵向腐蚀
+        if h < w:
+            kernel = np.ones((2, 1), np.uint8)
+            erode = cv2.erode(thresh1, kernel, iterations=1)
+        else:
+            erode = thresh1
+        # 水平膨胀
+        kernel = np.ones((1, 5), np.uint8)
+        erosion = cv2.dilate(erode, kernel, iterations=1)
+        # 水平投影
+        projection_map = np.ones_like(erosion)
+        project_val_array = [0 for _ in range(0, h)]
+
+        for j in range(0, h):
+            for i in range(0, w):
+                if erosion[j, i] == 255:
+                    project_val_array[j] += 1
+        # 根据数组，获取切割点
+        start_idx = 0  # 记录进入字符区的索引
+        end_idx = 0  # 记录进入空白区域的索引
+        in_text = False  # 是否遍历到了字符区内
+        box_list = []
+        spilt_threshold = 0
+        for i in range(len(project_val_array)):
+            if (
+                in_text == False and project_val_array[i] > spilt_threshold
+            ):  # 进入字符区了
+                in_text = True
+                start_idx = i
+            elif (
+                project_val_array[i] <= spilt_threshold and in_text == True
+            ):  # 进入空白区了
+                end_idx = i
+                in_text = False
+                if end_idx - start_idx <= 2:
+                    continue
+                box_list.append((start_idx, end_idx + 1))
+
+        if in_text:
+            box_list.append((start_idx, h - 1))
+        # 绘制投影直方图
+        for j in range(0, h):
+            for i in range(0, project_val_array[j]):
+                projection_map[j, i] = 0
+        split_bbox_list = []
+        if len(box_list) > 1:
+            for i, (h_start, h_end) in enumerate(box_list):
+                if i == 0:
+                    h_start = 0
+                if i == len(box_list):
+                    h_end = h
+                word_img = erosion[h_start : h_end + 1, :]
+                word_h, word_w = word_img.shape
+                w_split_list, w_projection_map = self.projection(
+                    word_img.T, word_w, word_h
+                )
+                w_start, w_end = w_split_list[0][0], w_split_list[-1][1]
+                if h_start > 0:
+                    h_start -= 1
+                h_end += 1
+                word_img = box_img[h_start : h_end + 1 :, w_start : w_end + 1, :]
+                split_bbox_list.append([w_start, h_start, w_end, h_end])
+        else:
+            split_bbox_list.append([0, 0, w, h])
+        return split_bbox_list
+
+    def shrink_bbox(self, bbox):
+        left, top, right, bottom = bbox
+        sh_h = min(max(int((bottom - top) * 0.1), 1), self.shrink_h_max)
+        sh_w = min(max(int((right - left) * 0.1), 1), self.shrink_w_max)
+        left_new = left + sh_w
+        right_new = right - sh_w
+        top_new = top + sh_h
+        bottom_new = bottom - sh_h
+        if left_new >= right_new:
+            left_new = left
+            right_new = right
+        if top_new >= bottom_new:
+            top_new = top
+            bottom_new = bottom
+        return [left_new, top_new, right_new, bottom_new]
+
+    def __call__(self, data):
+        img = data["image"]
+        cells = data["cells"]
+        height, width = img.shape[0:2]
+        if self.mask_type == 1:
+            mask_img = np.zeros((height, width), dtype=np.float32)
+        else:
+            mask_img = np.zeros((height, width, 3), dtype=np.float32)
+        cell_num = len(cells)
+        for cno in range(cell_num):
+            if "bbox" in cells[cno]:
+                bbox = cells[cno]["bbox"]
+                left, top, right, bottom = bbox
+                box_img = img[top:bottom, left:right, :].copy()
+                split_bbox_list = self.projection_cx(box_img)
+                for sno in range(len(split_bbox_list)):
+                    split_bbox_list[sno][0] += left
+                    split_bbox_list[sno][1] += top
+                    split_bbox_list[sno][2] += left
+                    split_bbox_list[sno][3] += top
+
+                for sno in range(len(split_bbox_list)):
+                    left, top, right, bottom = split_bbox_list[sno]
+                    left, top, right, bottom = self.shrink_bbox(
+                        [left, top, right, bottom]
+                    )
+                    if self.mask_type == 1:
+                        mask_img[top:bottom, left:right] = 1.0
+                        data["mask_img"] = mask_img
+                    else:
+                        mask_img[top:bottom, left:right, :] = (255, 255, 255)
+                        data["image"] = mask_img
+        return data
+
+
+class ResizeTableImage(object):
+    def __init__(self, max_len, resize_bboxes=False, infer_mode=False, **kwargs):
+        super(ResizeTableImage, self).__init__()
+        self.max_len = max_len
+        self.resize_bboxes = resize_bboxes
+        self.infer_mode = infer_mode
+
+    def __call__(self, data):
+        img = data["image"]
+        height, width = img.shape[0:2]
+        ratio = self.max_len / (max(height, width) * 1.0)
+        resize_h = int(height * ratio)
+        resize_w = int(width * ratio)
+        resize_img = cv2.resize(img, (resize_w, resize_h))
+        if self.resize_bboxes and not self.infer_mode:
+            data["bboxes"] = data["bboxes"] * ratio
+        data["image"] = resize_img
+        data["src_img"] = img
+        data["shape"] = np.array([height, width, ratio, ratio])
+        data["max_len"] = self.max_len
+        return data
+
+
+class PaddingTableImage(object):
+    def __init__(self, size, **kwargs):
+        super(PaddingTableImage, self).__init__()
+        self.size = size
+
+    def __call__(self, data):
+        img = data["image"]
+        pad_h, pad_w = self.size
+        padding_img = np.zeros((pad_h, pad_w, 3), dtype=np.float32)
+        height, width = img.shape[0:2]
+        padding_img[0:height, 0:width, :] = img.copy()
+        data["image"] = padding_img
+        shape = data["shape"].tolist()
+        shape.extend([pad_h, pad_w])
+        data["shape"] = np.array(shape)
+        return data
diff --git a/ppocr/data/imaug/text_image_aug/__init__.py b/ppocr/data/imaug/text_image_aug/__init__.py
new file mode 100644
index 0000000..16f179f
--- /dev/null
+++ b/ppocr/data/imaug/text_image_aug/__init__.py
@@ -0,0 +1,17 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .augment import tia_perspective, tia_distort, tia_stretch
+
+__all__ = ["tia_distort", "tia_stretch", "tia_perspective"]
diff --git a/ppocr/data/imaug/text_image_aug/augment.py b/ppocr/data/imaug/text_image_aug/augment.py
new file mode 100644
index 0000000..1044abd
--- /dev/null
+++ b/ppocr/data/imaug/text_image_aug/augment.py
@@ -0,0 +1,123 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/RubanSeven/Text-Image-Augmentation-python/blob/master/augment.py
+"""
+
+import numpy as np
+from .warp_mls import WarpMLS
+
+
+def tia_distort(src, segment=4):
+    img_h, img_w = src.shape[:2]
+
+    cut = img_w // segment
+    thresh = cut // 3
+
+    src_pts = list()
+    dst_pts = list()
+
+    src_pts.append([0, 0])
+    src_pts.append([img_w, 0])
+    src_pts.append([img_w, img_h])
+    src_pts.append([0, img_h])
+
+    dst_pts.append([np.random.randint(thresh), np.random.randint(thresh)])
+    dst_pts.append([img_w - np.random.randint(thresh), np.random.randint(thresh)])
+    dst_pts.append(
+        [img_w - np.random.randint(thresh), img_h - np.random.randint(thresh)]
+    )
+    dst_pts.append([np.random.randint(thresh), img_h - np.random.randint(thresh)])
+
+    half_thresh = thresh * 0.5
+
+    for cut_idx in np.arange(1, segment, 1):
+        src_pts.append([cut * cut_idx, 0])
+        src_pts.append([cut * cut_idx, img_h])
+        dst_pts.append(
+            [
+                cut * cut_idx + np.random.randint(thresh) - half_thresh,
+                np.random.randint(thresh) - half_thresh,
+            ]
+        )
+        dst_pts.append(
+            [
+                cut * cut_idx + np.random.randint(thresh) - half_thresh,
+                img_h + np.random.randint(thresh) - half_thresh,
+            ]
+        )
+
+    trans = WarpMLS(src, src_pts, dst_pts, img_w, img_h)
+    dst = trans.generate()
+
+    return dst
+
+
+def tia_stretch(src, segment=4):
+    img_h, img_w = src.shape[:2]
+
+    cut = img_w // segment
+    thresh = cut * 4 // 5
+
+    src_pts = list()
+    dst_pts = list()
+
+    src_pts.append([0, 0])
+    src_pts.append([img_w, 0])
+    src_pts.append([img_w, img_h])
+    src_pts.append([0, img_h])
+
+    dst_pts.append([0, 0])
+    dst_pts.append([img_w, 0])
+    dst_pts.append([img_w, img_h])
+    dst_pts.append([0, img_h])
+
+    half_thresh = thresh * 0.5
+
+    for cut_idx in np.arange(1, segment, 1):
+        move = np.random.randint(thresh) - half_thresh
+        src_pts.append([cut * cut_idx, 0])
+        src_pts.append([cut * cut_idx, img_h])
+        dst_pts.append([cut * cut_idx + move, 0])
+        dst_pts.append([cut * cut_idx + move, img_h])
+
+    trans = WarpMLS(src, src_pts, dst_pts, img_w, img_h)
+    dst = trans.generate()
+
+    return dst
+
+
+def tia_perspective(src):
+    img_h, img_w = src.shape[:2]
+
+    thresh = img_h // 2
+
+    src_pts = list()
+    dst_pts = list()
+
+    src_pts.append([0, 0])
+    src_pts.append([img_w, 0])
+    src_pts.append([img_w, img_h])
+    src_pts.append([0, img_h])
+
+    dst_pts.append([0, np.random.randint(thresh)])
+    dst_pts.append([img_w, np.random.randint(thresh)])
+    dst_pts.append([img_w, img_h - np.random.randint(thresh)])
+    dst_pts.append([0, img_h - np.random.randint(thresh)])
+
+    trans = WarpMLS(src, src_pts, dst_pts, img_w, img_h)
+    dst = trans.generate()
+
+    return dst
diff --git a/ppocr/data/imaug/text_image_aug/warp_mls.py b/ppocr/data/imaug/text_image_aug/warp_mls.py
new file mode 100644
index 0000000..2d349a6
--- /dev/null
+++ b/ppocr/data/imaug/text_image_aug/warp_mls.py
@@ -0,0 +1,187 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/RubanSeven/Text-Image-Augmentation-python/blob/master/warp_mls.py
+"""
+
+import numpy as np
+
+
+class WarpMLS:
+    def __init__(self, src, src_pts, dst_pts, dst_w, dst_h, trans_ratio=1.0):
+        self.src = src
+        self.src_pts = src_pts
+        self.dst_pts = dst_pts
+        self.pt_count = len(self.dst_pts)
+        self.dst_w = dst_w
+        self.dst_h = dst_h
+        self.trans_ratio = trans_ratio
+        self.grid_size = 100
+        self.rdx = np.zeros((self.dst_h, self.dst_w))
+        self.rdy = np.zeros((self.dst_h, self.dst_w))
+
+    @staticmethod
+    def __bilinear_interp(x, y, v11, v12, v21, v22):
+        return (v11 * (1 - y) + v12 * y) * (1 - x) + (v21 * (1 - y) + v22 * y) * x
+
+    def generate(self):
+        self.calc_delta()
+        return self.gen_img()
+
+    def calc_delta(self):
+        w = np.zeros(self.pt_count, dtype=np.float32)
+
+        if self.pt_count < 2:
+            return
+
+        i = 0
+        while 1:
+            if self.dst_w <= i < self.dst_w + self.grid_size - 1:
+                i = self.dst_w - 1
+            elif i >= self.dst_w:
+                break
+
+            j = 0
+            while 1:
+                if self.dst_h <= j < self.dst_h + self.grid_size - 1:
+                    j = self.dst_h - 1
+                elif j >= self.dst_h:
+                    break
+
+                sw = 0
+                swp = np.zeros(2, dtype=np.float32)
+                swq = np.zeros(2, dtype=np.float32)
+                new_pt = np.zeros(2, dtype=np.float32)
+                cur_pt = np.array([i, j], dtype=np.float32)
+
+                k = 0
+                for k in range(self.pt_count):
+                    if i == self.dst_pts[k][0] and j == self.dst_pts[k][1]:
+                        break
+
+                    w[k] = 1.0 / (
+                        (i - self.dst_pts[k][0]) * (i - self.dst_pts[k][0])
+                        + (j - self.dst_pts[k][1]) * (j - self.dst_pts[k][1])
+                    )
+
+                    sw += w[k]
+                    swp = swp + w[k] * np.array(self.dst_pts[k])
+                    swq = swq + w[k] * np.array(self.src_pts[k])
+
+                if k == self.pt_count - 1:
+                    pstar = 1 / sw * swp
+                    qstar = 1 / sw * swq
+
+                    miu_s = 0
+                    for k in range(self.pt_count):
+                        if i == self.dst_pts[k][0] and j == self.dst_pts[k][1]:
+                            continue
+                        pt_i = self.dst_pts[k] - pstar
+                        miu_s += w[k] * np.sum(pt_i * pt_i)
+
+                    cur_pt -= pstar
+                    cur_pt_j = np.array([-cur_pt[1], cur_pt[0]])
+
+                    for k in range(self.pt_count):
+                        if i == self.dst_pts[k][0] and j == self.dst_pts[k][1]:
+                            continue
+
+                        pt_i = self.dst_pts[k] - pstar
+                        pt_j = np.array([-pt_i[1], pt_i[0]])
+
+                        tmp_pt = np.zeros(2, dtype=np.float32)
+                        tmp_pt[0] = (
+                            np.sum(pt_i * cur_pt) * self.src_pts[k][0]
+                            - np.sum(pt_j * cur_pt) * self.src_pts[k][1]
+                        )
+                        tmp_pt[1] = (
+                            -np.sum(pt_i * cur_pt_j) * self.src_pts[k][0]
+                            + np.sum(pt_j * cur_pt_j) * self.src_pts[k][1]
+                        )
+                        tmp_pt *= w[k] / miu_s
+                        new_pt += tmp_pt
+
+                    new_pt += qstar
+                else:
+                    new_pt = self.src_pts[k]
+
+                self.rdx[j, i] = new_pt[0] - i
+                self.rdy[j, i] = new_pt[1] - j
+
+                j += self.grid_size
+            i += self.grid_size
+
+    def gen_img(self):
+        src_h, src_w = self.src.shape[:2]
+        dst = np.zeros_like(self.src, dtype=np.float32)
+
+        for i in np.arange(0, self.dst_h, self.grid_size):
+            for j in np.arange(0, self.dst_w, self.grid_size):
+                ni = i + self.grid_size
+                nj = j + self.grid_size
+                w = h = self.grid_size
+                if ni >= self.dst_h:
+                    ni = self.dst_h - 1
+                    h = ni - i + 1
+                if nj >= self.dst_w:
+                    nj = self.dst_w - 1
+                    w = nj - j + 1
+
+                di = np.reshape(np.arange(h), (-1, 1))
+                dj = np.reshape(np.arange(w), (1, -1))
+                delta_x = self.__bilinear_interp(
+                    di / h,
+                    dj / w,
+                    self.rdx[i, j],
+                    self.rdx[i, nj],
+                    self.rdx[ni, j],
+                    self.rdx[ni, nj],
+                )
+                delta_y = self.__bilinear_interp(
+                    di / h,
+                    dj / w,
+                    self.rdy[i, j],
+                    self.rdy[i, nj],
+                    self.rdy[ni, j],
+                    self.rdy[ni, nj],
+                )
+                nx = j + dj + delta_x * self.trans_ratio
+                ny = i + di + delta_y * self.trans_ratio
+                nx = np.clip(nx, 0, src_w - 1)
+                ny = np.clip(ny, 0, src_h - 1)
+                nxi = np.array(np.floor(nx), dtype=np.int32)
+                nyi = np.array(np.floor(ny), dtype=np.int32)
+                nxi1 = np.array(np.ceil(nx), dtype=np.int32)
+                nyi1 = np.array(np.ceil(ny), dtype=np.int32)
+
+                if len(self.src.shape) == 3:
+                    x = np.tile(np.expand_dims(ny - nyi, axis=-1), (1, 1, 3))
+                    y = np.tile(np.expand_dims(nx - nxi, axis=-1), (1, 1, 3))
+                else:
+                    x = ny - nyi
+                    y = nx - nxi
+                dst[i : i + h, j : j + w] = self.__bilinear_interp(
+                    x,
+                    y,
+                    self.src[nyi, nxi],
+                    self.src[nyi, nxi1],
+                    self.src[nyi1, nxi],
+                    self.src[nyi1, nxi1],
+                )
+
+        dst = np.clip(dst, 0, 255)
+        dst = np.array(dst, dtype=np.uint8)
+
+        return dst
diff --git a/ppocr/data/imaug/unimernet_aug.py b/ppocr/data/imaug/unimernet_aug.py
new file mode 100644
index 0000000..a9fe8f8
--- /dev/null
+++ b/ppocr/data/imaug/unimernet_aug.py
@@ -0,0 +1,834 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+import os
+
+os.environ["NO_ALBUMENTATIONS_UPDATE"] = "1"
+
+import cv2
+import math
+import numpy as np
+from io import BytesIO
+import albumentations as A
+from PIL import Image, ImageOps, ImageDraw
+from scipy.ndimage import zoom as scizoom
+
+
+class Erosion(A.ImageOnlyTransform):
+    def __init__(self, scale, always_apply=False, p=0.5):
+        super().__init__(always_apply=always_apply, p=p)
+        if type(scale) is tuple or type(scale) is list:
+            assert len(scale) == 2
+            self.scale = scale
+        else:
+            self.scale = (scale, scale)
+
+    def apply(self, img, **params):
+        kernel = cv2.getStructuringElement(
+            cv2.MORPH_ELLIPSE, tuple(np.random.randint(self.scale[0], self.scale[1], 2))
+        )
+        img = cv2.erode(img, kernel, iterations=1)
+        return img
+
+
+class Dilation(A.ImageOnlyTransform):
+    def __init__(self, scale, always_apply=False, p=0.5):
+        super().__init__(always_apply=always_apply, p=p)
+        if type(scale) is tuple or type(scale) is list:
+            assert len(scale) == 2
+            self.scale = scale
+        else:
+            self.scale = (scale, scale)
+
+    def apply(self, img, **params):
+        kernel = cv2.getStructuringElement(
+            cv2.MORPH_ELLIPSE, tuple(np.random.randint(self.scale[0], self.scale[1], 2))
+        )
+        img = cv2.dilate(img, kernel, iterations=1)
+        return img
+
+
+class Bitmap(A.ImageOnlyTransform):
+    def __init__(self, value=0, lower=200, always_apply=False, p=0.5):
+        super().__init__(always_apply=always_apply, p=p)
+        self.lower = lower
+        self.value = value
+
+    def apply(self, img, **params):
+        img = img.copy()
+        img[img < self.lower] = self.value
+        return img
+
+
+def clipped_zoom(img, zoom_factor):
+    h = img.shape[1]
+    ch = int(np.ceil(h / float(zoom_factor)))
+    top = (h - ch) // 2
+    img = scizoom(
+        img[top : top + ch, top : top + ch], (zoom_factor, zoom_factor, 1), order=1
+    )
+    trim_top = (img.shape[0] - h) // 2
+
+    return img[trim_top : trim_top + h, trim_top : trim_top + h]
+
+
+def disk(radius, alias_blur=0.1, dtype=np.float32):
+    if radius <= 8:
+        coords = np.arange(-8, 8 + 1)
+        ksize = (3, 3)
+    else:
+        coords = np.arange(-radius, radius + 1)
+        ksize = (5, 5)
+    x, y = np.meshgrid(coords, coords)
+    aliased_disk = np.asarray((x**2 + y**2) <= radius**2, dtype=dtype)
+    aliased_disk /= np.sum(aliased_disk)
+
+    return cv2.GaussianBlur(aliased_disk, ksize=ksize, sigmaX=alias_blur)
+
+
+def plasma_fractal(mapsize=256, wibbledecay=3, rng=None):
+    """
+    Generate a heightmap using diamond-square algorithm.
+    Return square 2d array, side length 'mapsize', of floats in range 0-255.
+    'mapsize' must be a power of two.
+    """
+    assert mapsize & (mapsize - 1) == 0
+    maparray = np.empty((mapsize, mapsize), dtype=np.float64)
+    maparray[0, 0] = 0
+    stepsize = mapsize
+    wibble = 100
+    if rng is None:
+        rng = np.random.default_rng()
+
+    def wibbledmean(array):
+        return array / 4 + wibble * rng.uniform(-wibble, wibble, array.shape)
+
+    def fillsquares():
+        """For each square of points stepsize apart,
+        calculate middle value as mean of points + wibble"""
+        cornerref = maparray[0:mapsize:stepsize, 0:mapsize:stepsize]
+        squareaccum = cornerref + np.roll(cornerref, shift=-1, axis=0)
+        squareaccum += np.roll(squareaccum, shift=-1, axis=1)
+        maparray[
+            stepsize // 2 : mapsize : stepsize, stepsize // 2 : mapsize : stepsize
+        ] = wibbledmean(squareaccum)
+
+    def filldiamonds():
+        """For each diamond of points stepsize apart,
+        calculate middle value as mean of points + wibble"""
+        drgrid = maparray[
+            stepsize // 2 : mapsize : stepsize, stepsize // 2 : mapsize : stepsize
+        ]
+        ulgrid = maparray[0:mapsize:stepsize, 0:mapsize:stepsize]
+        ldrsum = drgrid + np.roll(drgrid, 1, axis=0)
+        lulsum = ulgrid + np.roll(ulgrid, -1, axis=1)
+        ltsum = ldrsum + lulsum
+        maparray[0:mapsize:stepsize, stepsize // 2 : mapsize : stepsize] = wibbledmean(
+            ltsum
+        )
+        tdrsum = drgrid + np.roll(drgrid, 1, axis=1)
+        tulsum = ulgrid + np.roll(ulgrid, -1, axis=0)
+        ttsum = tdrsum + tulsum
+        maparray[stepsize // 2 : mapsize : stepsize, 0:mapsize:stepsize] = wibbledmean(
+            ttsum
+        )
+
+    while stepsize >= 2:
+        fillsquares()
+        filldiamonds()
+        stepsize //= 2
+        wibble /= wibbledecay
+
+    maparray -= maparray.min()
+    return maparray / maparray.max()
+
+
+class Fog(A.ImageOnlyTransform):
+    def __init__(self, mag=-1, always_apply=False, p=1.0):
+        super().__init__(always_apply=always_apply, p=p)
+        self.rng = np.random.default_rng()
+        self.mag = mag
+
+    def apply(self, img, **params):
+        img = Image.fromarray(img.astype(np.uint8))
+        w, h = img.size
+        c = [(1.5, 2), (2.0, 2), (2.5, 1.7)]
+        if self.mag < 0 or self.mag >= len(c):
+            index = self.rng.integers(0, len(c))
+        else:
+            index = self.mag
+        c = c[index]
+
+        n_channels = len(img.getbands())
+        isgray = n_channels == 1
+
+        img = np.asarray(img) / 255.0
+        max_val = img.max()
+        max_size = 2 ** math.ceil(math.log2(max(w, h)) + 1)
+        fog = (
+            c[0]
+            * plasma_fractal(mapsize=max_size, wibbledecay=c[1], rng=self.rng)[:h, :w][
+                ..., np.newaxis
+            ]
+        )
+        if isgray:
+            fog = np.squeeze(fog)
+        else:
+            fog = np.repeat(fog, 3, axis=2)
+
+        img += fog
+        img = np.clip(img * max_val / (max_val + c[0]), 0, 1) * 255
+        return img.astype(np.uint8)
+
+
+class Frost(A.ImageOnlyTransform):
+    def __init__(self, mag=-1, always_apply=False, p=1.0):
+        super().__init__(always_apply=always_apply, p=p)
+        self.rng = np.random.default_rng()
+        self.mag = mag
+
+    def apply(self, img, **params):
+        img = Image.fromarray(img.astype(np.uint8))
+        w, h = img.size
+        c = [(0.78, 0.22), (0.64, 0.36), (0.5, 0.5)]
+        if self.mag < 0 or self.mag >= len(c):
+            index = self.rng.integers(0, len(c))
+        else:
+            index = self.mag
+        c = c[index]
+
+        file_dir = os.path.dirname(__file__)
+        filename = [
+            os.path.join(file_dir, "frost_img", "frost1.jpg"),
+            os.path.join(file_dir, "frost_img", "frost2.png"),
+            os.path.join(file_dir, "frost_img", "frost3.png"),
+            os.path.join(file_dir, "frost_img", "frost4.jpg"),
+            os.path.join(file_dir, "frost_img", "frost5.jpg"),
+            os.path.join(file_dir, "frost_img", "frost6.jpg"),
+        ]
+        index = self.rng.integers(0, len(filename))
+        filename = filename[index]
+        frost = Image.open(filename).convert("RGB")
+
+        f_w, f_h = frost.size
+        if w / h > f_w / f_h:
+            f_h = round(f_h * w / f_w)
+            f_w = w
+        else:
+            f_w = round(f_w * h / f_h)
+            f_h = h
+        frost = np.asarray(frost.resize((f_w, f_h)))
+
+        # randomly crop
+        y_start, x_start = self.rng.integers(0, f_h - h + 1), self.rng.integers(
+            0, f_w - w + 1
+        )
+        frost = frost[y_start : y_start + h, x_start : x_start + w]
+
+        n_channels = len(img.getbands())
+        isgray = n_channels == 1
+
+        img = np.asarray(img)
+
+        if isgray:
+            img = np.expand_dims(img, axis=2)
+            img = np.repeat(img, 3, axis=2)
+
+        img = np.clip(np.round(c[0] * img + c[1] * frost), 0, 255)
+        img = img.astype(np.uint8)
+        if isgray:
+            img = np.squeeze(img)
+        return img
+
+
+class Snow(A.ImageOnlyTransform):
+    def __init__(self, mag=-1, always_apply=False, p=1.0):
+        super().__init__(always_apply=always_apply, p=p)
+        self.rng = np.random.default_rng()
+        self.mag = mag
+
+    def apply(self, img, **params):
+        from wand.image import Image as WandImage
+
+        img = Image.fromarray(img.astype(np.uint8))
+        w, h = img.size
+        c = [
+            (0.1, 0.3, 3, 0.5, 10, 4, 0.8),
+            (0.2, 0.3, 2, 0.5, 12, 4, 0.7),
+            (0.55, 0.3, 4, 0.9, 12, 8, 0.7),
+        ]
+        if self.mag < 0 or self.mag >= len(c):
+            index = self.rng.integers(0, len(c))
+        else:
+            index = self.mag
+        c = c[index]
+
+        n_channels = len(img.getbands())
+        isgray = n_channels == 1
+
+        img = np.asarray(img, dtype=np.float32) / 255.0
+        if isgray:
+            img = np.expand_dims(img, axis=2)
+            img = np.repeat(img, 3, axis=2)
+
+        snow_layer = self.rng.normal(size=img.shape[:2], loc=c[0], scale=c[1])
+
+        snow_layer[snow_layer < c[3]] = 0
+
+        snow_layer = Image.fromarray(
+            (np.clip(snow_layer.squeeze(), 0, 1) * 255).astype(np.uint8), mode="L"
+        )
+        output = BytesIO()
+        snow_layer.save(output, format="PNG")
+        snow_layer = WandImage(blob=output.getvalue())
+
+        snow_layer.motion_blur(
+            radius=c[4], sigma=c[5], angle=self.rng.uniform(-135, -45)
+        )
+
+        snow_layer = (
+            cv2.imdecode(
+                np.frombuffer(snow_layer.make_blob(), np.uint8), cv2.IMREAD_UNCHANGED
+            )
+            / 255.0
+        )
+
+        snow_layer = snow_layer[..., np.newaxis]
+
+        img = c[6] * img
+        gray_img = (1 - c[6]) * np.maximum(
+            img, cv2.cvtColor(img, cv2.COLOR_RGB2GRAY).reshape(h, w, 1) * 1.5 + 0.5
+        )
+        img += gray_img
+        img = np.clip(img + snow_layer + np.rot90(snow_layer, k=2), 0, 1) * 255
+        img = img.astype(np.uint8)
+        if isgray:
+            img = np.squeeze(img)
+        return img
+
+
+class Rain(A.ImageOnlyTransform):
+    def __init__(self, mag=-1, always_apply=False, p=1.0):
+        super().__init__(always_apply=always_apply, p=p)
+        self.rng = np.random.default_rng()
+        self.mag = mag
+
+    def apply(self, img, **params):
+        img = Image.fromarray(img.astype(np.uint8))
+        img = img.copy()
+        w, h = img.size
+        n_channels = len(img.getbands())
+        isgray = n_channels == 1
+        line_width = self.rng.integers(1, 2)
+
+        c = [50, 70, 90]
+        if self.mag < 0 or self.mag >= len(c):
+            index = 0
+        else:
+            index = self.mag
+        c = c[index]
+
+        n_rains = self.rng.integers(c, c + 20)
+        slant = self.rng.integers(-60, 60)
+        fillcolor = 200 if isgray else (200, 200, 200)
+
+        draw = ImageDraw.Draw(img)
+        max_length = min(w, h, 10)
+        for i in range(1, n_rains):
+            length = self.rng.integers(5, max_length)
+            x1 = self.rng.integers(0, w - length)
+            y1 = self.rng.integers(0, h - length)
+            x2 = x1 + length * math.sin(slant * math.pi / 180.0)
+            y2 = y1 + length * math.cos(slant * math.pi / 180.0)
+            x2 = int(x2)
+            y2 = int(y2)
+            draw.line([(x1, y1), (x2, y2)], width=line_width, fill=fillcolor)
+        img = np.asarray(img).astype(np.uint8)
+        return img
+
+
+class Shadow(A.ImageOnlyTransform):
+    def __init__(self, mag=-1, always_apply=False, p=1.0):
+        super().__init__(always_apply=always_apply, p=p)
+        self.rng = np.random.default_rng()
+        self.mag = mag
+
+    def apply(self, img, **params):
+        img = Image.fromarray(img.astype(np.uint8))
+        w, h = img.size
+        n_channels = len(img.getbands())
+        isgray = n_channels == 1
+
+        c = [64, 96, 128]
+        if self.mag < 0 or self.mag >= len(c):
+            index = 0
+        else:
+            index = self.mag
+        c = c[index]
+
+        img = img.convert("RGBA")
+        overlay = Image.new("RGBA", img.size, (255, 255, 255, 0))
+        draw = ImageDraw.Draw(overlay)
+        transparency = self.rng.integers(c, c + 32)
+        x1 = self.rng.integers(0, w // 2)
+        y1 = 0
+
+        x2 = self.rng.integers(w // 2, w)
+        y2 = 0
+
+        x3 = self.rng.integers(w // 2, w)
+        y3 = h - 1
+
+        x4 = self.rng.integers(0, w // 2)
+        y4 = h - 1
+
+        draw.polygon(
+            [(x1, y1), (x2, y2), (x3, y3), (x4, y4)], fill=(0, 0, 0, transparency)
+        )
+
+        img = Image.alpha_composite(img, overlay)
+        img = img.convert("RGB")
+        if isgray:
+            img = ImageOps.grayscale(img)
+        img = np.asarray(img).astype(np.uint8)
+        return img
+
+
+class UniMERNetTrainTransform:
+    def __init__(self, bitmap_prob=0.04, **kwargs):
+        self.bitmap_prob = bitmap_prob
+        if tuple(map(int, A.__version__.split("."))) >= (2, 0, 0):
+            new_val = (0, (10 / 255) ** 0.5)
+            GaussNoise = A.GaussNoise(new_val, p=0.2)
+            ImageCompression = A.ImageCompression(quality_range=(95, 100), p=0.3)
+        else:
+            GaussNoise = A.GaussNoise(10, p=0.2)
+            ImageCompression = A.ImageCompression(95, p=0.3)
+        self.train_transform = A.Compose(
+            [
+                A.Compose(
+                    [
+                        Bitmap(p=0.05),
+                        A.OneOf([Fog(), Frost(), Snow(), Rain(), Shadow()], p=0.2),
+                        A.OneOf([Erosion((2, 3)), Dilation((2, 3))], p=0.2),
+                        A.ShiftScaleRotate(
+                            shift_limit=0,
+                            scale_limit=(-0.15, 0),
+                            rotate_limit=1,
+                            border_mode=0,
+                            interpolation=3,
+                            value=[255, 255, 255],
+                            p=1,
+                        ),
+                        A.GridDistortion(
+                            distort_limit=0.1,
+                            border_mode=0,
+                            interpolation=3,
+                            value=[255, 255, 255],
+                            p=0.5,
+                        ),
+                    ],
+                    p=0.15,
+                ),
+                A.RGBShift(r_shift_limit=15, g_shift_limit=15, b_shift_limit=15, p=0.3),
+                GaussNoise,
+                A.RandomBrightnessContrast(0.05, (-0.2, 0), True, p=0.2),
+                ImageCompression,
+                A.ToGray(always_apply=True),
+                A.Normalize((0.7931, 0.7931, 0.7931), (0.1738, 0.1738, 0.1738)),
+            ]
+        )
+
+    def __call__(self, data):
+        img = data["image"]
+        if np.random.random() < self.bitmap_prob:
+            img[img != 255] = 0
+        img = self.train_transform(image=img)["image"]
+        data["image"] = img
+        return data
+
+
+class UniMERNetTestTransform:
+    def __init__(self, **kwargs):
+        self.test_transform = A.Compose(
+            [
+                A.ToGray(always_apply=True),
+                A.Normalize((0.7931, 0.7931, 0.7931), (0.1738, 0.1738, 0.1738)),
+            ]
+        )
+
+    def __call__(self, data):
+        img = data["image"]
+        img = self.test_transform(image=img)["image"]
+        data["image"] = img
+        return data
+
+
+class GoTImgDecode:
+    def __init__(self, input_size, random_padding=False, **kwargs):
+        self.input_size = input_size
+        self.random_padding = random_padding
+
+    def crop_margin(self, img):
+        data = np.array(img.convert("L"))
+        data = data.astype(np.uint8)
+        max_val = data.max()
+        min_val = data.min()
+        if max_val == min_val:
+            return img
+        data = (data - min_val) / (max_val - min_val) * 255
+        gray = 255 * (data < 200).astype(np.uint8)
+        coords = cv2.findNonZero(gray)  # Find all non-zero points (text)
+        if coords is None:
+            return img
+        a, b, w, h = cv2.boundingRect(coords)  # Find minimum spanning bounding box
+        if w == 0 or h == 0:
+            return img
+        # Avoid extreme aspect ratios that cause errors in downstream processing
+        if max(w, h) / min(w, h) > 200:
+            return img
+        return img.crop((a, b, w + a, h + b))
+
+    def get_dimensions(self, img):
+        if hasattr(img, "getbands"):
+            channels = len(img.getbands())
+        else:
+            channels = img.channels
+        width, height = img.size
+        return [channels, height, width]
+
+    def _compute_resized_output_size(self, image_size, size, max_size=None):
+        if len(size) == 1:  # specified size only for the smallest edge
+            h, w = image_size
+            short, long = (w, h) if w <= h else (h, w)
+            requested_new_short = size if isinstance(size, int) else size[0]
+
+            new_short, new_long = requested_new_short, int(
+                requested_new_short * long / short
+            )
+
+            if max_size is not None:
+                if max_size <= requested_new_short:
+                    raise ValueError(
+                        f"max_size = {max_size} must be strictly greater than the requested "
+                        f"size for the smaller edge size = {size}"
+                    )
+                if new_long > max_size:
+                    new_short, new_long = int(max_size * new_short / new_long), max_size
+
+            new_w, new_h = (new_short, new_long) if w <= h else (new_long, new_short)
+        else:  # specified both h and w
+            new_w, new_h = size[1], size[0]
+        return [new_h, new_w]
+
+    def resize(self, img, size):
+        _, image_height, image_width = self.get_dimensions(img)
+        if isinstance(size, int):
+            size = [size]
+        max_size = None
+        output_size = self._compute_resized_output_size(
+            (image_height, image_width), size, max_size
+        )
+        img = img.resize(tuple(output_size[::-1]), resample=2)
+        return img
+
+    def __call__(self, data):
+        filename = data["filename"]
+        img = Image.open(filename)
+        try:
+            img = self.crop_margin(img.convert("RGB"))
+        except OSError:
+            return
+        if img.height == 0 or img.width == 0:
+            return
+        img = self.resize(img, min(self.input_size))
+        img.thumbnail((self.input_size[1], self.input_size[0]))
+        delta_width = self.input_size[1] - img.width
+        delta_height = self.input_size[0] - img.height
+        if self.random_padding:
+            pad_width = np.random.randint(low=0, high=delta_width + 1)
+            pad_height = np.random.randint(low=0, high=delta_height + 1)
+        else:
+            pad_width = delta_width // 2
+            pad_height = delta_height // 2
+        padding = (
+            pad_width,
+            pad_height,
+            delta_width - pad_width,
+            delta_height - pad_height,
+        )
+
+        data["image"] = np.array(ImageOps.expand(img, padding))
+        return data
+
+
+class UniMERNetImgDecode:
+    def __init__(
+        self,
+        input_size,
+        random_padding=False,
+        random_resize=False,
+        random_crop=False,
+        **kwargs,
+    ):
+        self.input_size = input_size
+        self.is_random_padding = random_padding
+        self.is_random_resize = random_resize
+        self.is_random_crop = random_crop
+
+    def crop_margin(self, img):
+        data = np.array(img.convert("L"))
+        data = data.astype(np.uint8)
+        max_val = data.max()
+        min_val = data.min()
+        if max_val == min_val:
+            return img
+        data = (data - min_val) / (max_val - min_val) * 255
+        gray = 255 * (data < 200).astype(np.uint8)
+        coords = cv2.findNonZero(gray)  # Find all non-zero points (text)
+        if coords is None:
+            return img
+        a, b, w, h = cv2.boundingRect(coords)  # Find minimum spanning bounding box
+        if w == 0 or h == 0:
+            return img
+        # Avoid extreme aspect ratios that cause errors in downstream processing
+        if max(w, h) / min(w, h) > 200:
+            return img
+        return img.crop((a, b, w + a, h + b))
+
+    def get_dimensions(self, img):
+        if hasattr(img, "getbands"):
+            channels = len(img.getbands())
+        else:
+            channels = img.channels
+        width, height = img.size
+        return [channels, height, width]
+
+    def _compute_resized_output_size(self, image_size, size, max_size=None):
+        if len(size) == 1:  # specified size only for the smallest edge
+            h, w = image_size
+            short, long = (w, h) if w <= h else (h, w)
+            requested_new_short = size if isinstance(size, int) else size[0]
+
+            new_short, new_long = requested_new_short, int(
+                requested_new_short * long / short
+            )
+
+            if max_size is not None:
+                if max_size <= requested_new_short:
+                    raise ValueError(
+                        f"max_size = {max_size} must be strictly greater than the requested "
+                        f"size for the smaller edge size = {size}"
+                    )
+                if new_long > max_size:
+                    new_short, new_long = int(max_size * new_short / new_long), max_size
+
+            new_w, new_h = (new_short, new_long) if w <= h else (new_long, new_short)
+        else:  # specified both h and w
+            new_w, new_h = size[1], size[0]
+        return [new_h, new_w]
+
+    def resize(self, img, size):
+        _, image_height, image_width = self.get_dimensions(img)
+        if isinstance(size, int):
+            size = [size]
+        max_size = None
+        output_size = self._compute_resized_output_size(
+            (image_height, image_width), size, max_size
+        )
+        img = img.resize(tuple(output_size[::-1]), resample=2)
+        return img
+
+    def random_resize(self, img):
+        scale = np.random.uniform(0.5, 1)
+        img = img.resize([int(scale * s) for s in img.size])
+        return img
+
+    def random_crop(self, img, crop_ratio):
+        width, height = img.width, img.height
+        max_crop_pixel = min(width, height) * crop_ratio
+        crop_left = np.random.uniform(0, max_crop_pixel)
+        crop_right = np.random.uniform(0, max_crop_pixel)
+        crop_top = np.random.uniform(0, max_crop_pixel)
+        crop_bottom = np.random.uniform(0, max_crop_pixel)
+        # 计算裁剪后的边界
+        left = crop_left
+        top = crop_top
+        right = width - crop_right
+        bottom = height - crop_bottom
+        # 裁剪图像
+        img = img.crop((left, top, right, bottom))
+
+        return img
+
+    def __call__(self, data):
+        filename = data["filename"]
+        img = Image.open(filename)
+        try:
+            if self.is_random_resize:
+                img = self.random_resize(img)
+            img = self.crop_margin(img.convert("RGB"))
+            if "label" in data and self.is_random_crop:
+                label = data["label"]
+                equation_length = len(label)
+                if equation_length < 256:
+                    img = self.random_crop(img, crop_ratio=0.1)
+                elif 256 < equation_length <= 512:
+                    img = self.random_crop(img, crop_ratio=0.05)
+                else:
+                    img = self.random_crop(img, crop_ratio=0.03)
+        except OSError:
+            return
+        if img.height == 0 or img.width == 0:
+            return
+        img = self.resize(img, min(self.input_size))
+        img.thumbnail((self.input_size[1], self.input_size[0]))
+        delta_width = self.input_size[1] - img.width
+        delta_height = self.input_size[0] - img.height
+        if self.is_random_padding:
+            pad_width = np.random.randint(low=0, high=delta_width + 1)
+            pad_height = np.random.randint(low=0, high=delta_height + 1)
+        else:
+            pad_width = delta_width // 2
+            pad_height = delta_height // 2
+        padding = (
+            pad_width,
+            pad_height,
+            delta_width - pad_width,
+            delta_height - pad_height,
+        )
+
+        data["image"] = np.array(ImageOps.expand(img, padding))
+        return data
+
+
+class UniMERNetResize:
+    def __init__(self, input_size, random_padding=False, **kwargs):
+        self.input_size = input_size
+        self.random_padding = random_padding
+
+    def crop_margin(self, img):
+        data = np.array(img.convert("L"))
+        data = data.astype(np.uint8)
+        max_val = data.max()
+        min_val = data.min()
+        if max_val == min_val:
+            return img
+        data = (data - min_val) / (max_val - min_val) * 255
+        gray = 255 * (data < 200).astype(np.uint8)
+
+        coords = cv2.findNonZero(gray)  # Find all non-zero points (text)
+        if coords is None:
+            return img
+        a, b, w, h = cv2.boundingRect(coords)  # Find minimum spanning bounding box
+        if w == 0 or h == 0:
+            return img
+        # Avoid extreme aspect ratios that cause errors in downstream processing
+        if max(w, h) / min(w, h) > 200:
+            return img
+        return img.crop((a, b, w + a, h + b))
+
+    def get_dimensions(self, img):
+        if hasattr(img, "getbands"):
+            channels = len(img.getbands())
+        else:
+            channels = img.channels
+        width, height = img.size
+        return [channels, height, width]
+
+    def _compute_resized_output_size(self, image_size, size, max_size=None):
+        if len(size) == 1:  # specified size only for the smallest edge
+            h, w = image_size
+            short, long = (w, h) if w <= h else (h, w)
+            requested_new_short = size if isinstance(size, int) else size[0]
+
+            new_short, new_long = requested_new_short, int(
+                requested_new_short * long / short
+            )
+
+            if max_size is not None:
+                if max_size <= requested_new_short:
+                    raise ValueError(
+                        f"max_size = {max_size} must be strictly greater than the requested "
+                        f"size for the smaller edge size = {size}"
+                    )
+                if new_long > max_size:
+                    new_short, new_long = int(max_size * new_short / new_long), max_size
+
+            new_w, new_h = (new_short, new_long) if w <= h else (new_long, new_short)
+        else:  # specified both h and w
+            new_w, new_h = size[1], size[0]
+        return [new_h, new_w]
+
+    def resize(self, img, size):
+        _, image_height, image_width = self.get_dimensions(img)
+        if isinstance(size, int):
+            size = [size]
+        max_size = None
+        output_size = self._compute_resized_output_size(
+            (image_height, image_width), size, max_size
+        )
+        img.resize(tuple(output_size[::-1]), resample=2)
+        return img
+
+    def __call__(self, data):
+        img = data["image"]
+        img = Image.fromarray(img)
+        try:
+            img = self.crop_margin(img)
+        except OSError:
+            return
+        if img.height == 0 or img.width == 0:
+            return
+        img = self.resize(img, min(self.input_size))
+        img.thumbnail((self.input_size[1], self.input_size[0]))
+        delta_width = self.input_size[1] - img.width
+        delta_height = self.input_size[0] - img.height
+        if self.random_padding:
+            pad_width = np.random.randint(low=0, high=delta_width + 1)
+            pad_height = np.random.randint(low=0, high=delta_height + 1)
+        else:
+            pad_width = delta_width // 2
+            pad_height = delta_height // 2
+        padding = (
+            pad_width,
+            pad_height,
+            delta_width - pad_width,
+            delta_height - pad_height,
+        )
+        data["image"] = np.array(ImageOps.expand(img, padding))
+        return data
+
+
+class UniMERNetImageFormat:
+    def __init__(self, **kwargs):
+        pass
+
+    def __call__(self, data):
+        img = data["image"]
+        im_h, im_w = img.shape[:2]
+        divide_h = math.ceil(im_h / 32) * 32
+        divide_w = math.ceil(im_w / 32) * 32
+        img = img[:, :, 0]
+        img = np.pad(
+            img, ((0, divide_h - im_h), (0, divide_w - im_w)), constant_values=(1, 1)
+        )
+        img_expanded = img[:, :, np.newaxis].transpose(2, 0, 1)
+        data["image"] = img_expanded
+        return data
diff --git a/ppocr/data/imaug/vqa/__init__.py b/ppocr/data/imaug/vqa/__init__.py
new file mode 100644
index 0000000..812f20b
--- /dev/null
+++ b/ppocr/data/imaug/vqa/__init__.py
@@ -0,0 +1,29 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .token import (
+    VQATokenPad,
+    VQASerTokenChunk,
+    VQAReTokenChunk,
+    VQAReTokenRelation,
+    TensorizeEntitiesRelations,
+)
+
+__all__ = [
+    "VQATokenPad",
+    "VQASerTokenChunk",
+    "VQAReTokenChunk",
+    "VQAReTokenRelation",
+    "TensorizeEntitiesRelations",
+]
diff --git a/ppocr/data/imaug/vqa/augment.py b/ppocr/data/imaug/vqa/augment.py
new file mode 100644
index 0000000..d4f4cf2
--- /dev/null
+++ b/ppocr/data/imaug/vqa/augment.py
@@ -0,0 +1,34 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import sys
+import numpy as np
+import random
+from copy import deepcopy
+
+
+def order_by_tbyx(ocr_info):
+    res = sorted(ocr_info, key=lambda r: (r["bbox"][1], r["bbox"][0]))
+    for i in range(len(res) - 1):
+        for j in range(i, 0, -1):
+            if abs(res[j + 1]["bbox"][1] - res[j]["bbox"][1]) < 20 and (
+                res[j + 1]["bbox"][0] < res[j]["bbox"][0]
+            ):
+                tmp = deepcopy(res[j])
+                res[j] = deepcopy(res[j + 1])
+                res[j + 1] = deepcopy(tmp)
+            else:
+                break
+    return res
diff --git a/ppocr/data/imaug/vqa/token/__init__.py b/ppocr/data/imaug/vqa/token/__init__.py
new file mode 100644
index 0000000..e349dd7
--- /dev/null
+++ b/ppocr/data/imaug/vqa/token/__init__.py
@@ -0,0 +1,18 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .vqa_token_chunk import VQASerTokenChunk, VQAReTokenChunk
+from .vqa_token_pad import VQATokenPad
+from .vqa_token_relation import VQAReTokenRelation
+from .vqa_re_convert import TensorizeEntitiesRelations
diff --git a/ppocr/data/imaug/vqa/token/vqa_re_convert.py b/ppocr/data/imaug/vqa/token/vqa_re_convert.py
new file mode 100644
index 0000000..fa14915
--- /dev/null
+++ b/ppocr/data/imaug/vqa/token/vqa_re_convert.py
@@ -0,0 +1,49 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import numpy as np
+
+
+class TensorizeEntitiesRelations(object):
+    def __init__(self, max_seq_len=512, infer_mode=False, **kwargs):
+        self.max_seq_len = max_seq_len
+        self.infer_mode = infer_mode
+
+    def __call__(self, data):
+        entities = data["entities"]
+        relations = data["relations"]
+
+        entities_new = np.full(
+            shape=[self.max_seq_len + 1, 3], fill_value=-1, dtype="int64"
+        )
+        entities_new[0, 0] = len(entities["start"])
+        entities_new[0, 1] = len(entities["end"])
+        entities_new[0, 2] = len(entities["label"])
+        entities_new[1 : len(entities["start"]) + 1, 0] = np.array(entities["start"])
+        entities_new[1 : len(entities["end"]) + 1, 1] = np.array(entities["end"])
+        entities_new[1 : len(entities["label"]) + 1, 2] = np.array(entities["label"])
+
+        relations_new = np.full(
+            shape=[self.max_seq_len * self.max_seq_len + 1, 2],
+            fill_value=-1,
+            dtype="int64",
+        )
+        relations_new[0, 0] = len(relations["head"])
+        relations_new[0, 1] = len(relations["tail"])
+        relations_new[1 : len(relations["head"]) + 1, 0] = np.array(relations["head"])
+        relations_new[1 : len(relations["tail"]) + 1, 1] = np.array(relations["tail"])
+
+        data["entities"] = entities_new
+        data["relations"] = relations_new
+        return data
diff --git a/ppocr/data/imaug/vqa/token/vqa_token_chunk.py b/ppocr/data/imaug/vqa/token/vqa_token_chunk.py
new file mode 100644
index 0000000..d46a47d
--- /dev/null
+++ b/ppocr/data/imaug/vqa/token/vqa_token_chunk.py
@@ -0,0 +1,134 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from collections import defaultdict
+
+
+class VQASerTokenChunk(object):
+    def __init__(self, max_seq_len=512, infer_mode=False, **kwargs):
+        self.max_seq_len = max_seq_len
+        self.infer_mode = infer_mode
+
+    def __call__(self, data):
+        encoded_inputs_all = []
+        seq_len = len(data["input_ids"])
+        for index in range(0, seq_len, self.max_seq_len):
+            chunk_beg = index
+            chunk_end = min(index + self.max_seq_len, seq_len)
+            encoded_inputs_example = {}
+            for key in data:
+                if key in [
+                    "label",
+                    "input_ids",
+                    "labels",
+                    "token_type_ids",
+                    "bbox",
+                    "attention_mask",
+                ]:
+                    if self.infer_mode and key == "labels":
+                        encoded_inputs_example[key] = data[key]
+                    else:
+                        encoded_inputs_example[key] = data[key][chunk_beg:chunk_end]
+                else:
+                    encoded_inputs_example[key] = data[key]
+
+            encoded_inputs_all.append(encoded_inputs_example)
+        if len(encoded_inputs_all) == 0:
+            return None
+        return encoded_inputs_all[0]
+
+
+class VQAReTokenChunk(object):
+    def __init__(
+        self, max_seq_len=512, entities_labels=None, infer_mode=False, **kwargs
+    ):
+        self.max_seq_len = max_seq_len
+        self.entities_labels = (
+            {"HEADER": 0, "QUESTION": 1, "ANSWER": 2}
+            if entities_labels is None
+            else entities_labels
+        )
+        self.infer_mode = infer_mode
+
+    def __call__(self, data):
+        # prepare data
+        entities = data.pop("entities")
+        relations = data.pop("relations")
+        encoded_inputs_all = []
+        for index in range(0, len(data["input_ids"]), self.max_seq_len):
+            item = {}
+            for key in data:
+                if key in [
+                    "label",
+                    "input_ids",
+                    "labels",
+                    "token_type_ids",
+                    "bbox",
+                    "attention_mask",
+                ]:
+                    if self.infer_mode and key == "labels":
+                        item[key] = data[key]
+                    else:
+                        item[key] = data[key][index : index + self.max_seq_len]
+                else:
+                    item[key] = data[key]
+            # select entity in current chunk
+            entities_in_this_span = []
+            global_to_local_map = {}  #
+            for entity_id, entity in enumerate(entities):
+                if (
+                    index <= entity["start"] < index + self.max_seq_len
+                    and index <= entity["end"] < index + self.max_seq_len
+                ):
+                    entity["start"] = entity["start"] - index
+                    entity["end"] = entity["end"] - index
+                    global_to_local_map[entity_id] = len(entities_in_this_span)
+                    entities_in_this_span.append(entity)
+
+            # select relations in current chunk
+            relations_in_this_span = []
+            for relation in relations:
+                if (
+                    index <= relation["start_index"] < index + self.max_seq_len
+                    and index <= relation["end_index"] < index + self.max_seq_len
+                ):
+                    relations_in_this_span.append(
+                        {
+                            "head": global_to_local_map[relation["head"]],
+                            "tail": global_to_local_map[relation["tail"]],
+                            "start_index": relation["start_index"] - index,
+                            "end_index": relation["end_index"] - index,
+                        }
+                    )
+            item.update(
+                {
+                    "entities": self.reformat(entities_in_this_span),
+                    "relations": self.reformat(relations_in_this_span),
+                }
+            )
+            if len(item["entities"]) > 0:
+                item["entities"]["label"] = [
+                    self.entities_labels[x] for x in item["entities"]["label"]
+                ]
+                encoded_inputs_all.append(item)
+        if len(encoded_inputs_all) == 0:
+            return None
+        return encoded_inputs_all[0]
+
+    def reformat(self, data):
+        new_data = defaultdict(list)
+        for item in data:
+            for k, v in item.items():
+                new_data[k].append(v)
+        return new_data
diff --git a/ppocr/data/imaug/vqa/token/vqa_token_pad.py b/ppocr/data/imaug/vqa/token/vqa_token_pad.py
new file mode 100644
index 0000000..9466e37
--- /dev/null
+++ b/ppocr/data/imaug/vqa/token/vqa_token_pad.py
@@ -0,0 +1,117 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import paddle
+import numpy as np
+
+
+class VQATokenPad(object):
+    def __init__(
+        self,
+        max_seq_len=512,
+        pad_to_max_seq_len=True,
+        return_attention_mask=True,
+        return_token_type_ids=True,
+        truncation_strategy="longest_first",
+        return_overflowing_tokens=False,
+        return_special_tokens_mask=False,
+        infer_mode=False,
+        **kwargs,
+    ):
+        self.max_seq_len = max_seq_len
+        self.pad_to_max_seq_len = max_seq_len
+        self.return_attention_mask = return_attention_mask
+        self.return_token_type_ids = return_token_type_ids
+        self.truncation_strategy = truncation_strategy
+        self.return_overflowing_tokens = return_overflowing_tokens
+        self.return_special_tokens_mask = return_special_tokens_mask
+        self.pad_token_label_id = paddle.nn.CrossEntropyLoss().ignore_index
+        self.infer_mode = infer_mode
+
+    def __call__(self, data):
+        needs_to_be_padded = (
+            self.pad_to_max_seq_len and len(data["input_ids"]) < self.max_seq_len
+        )
+
+        if needs_to_be_padded:
+            if "tokenizer_params" in data:
+                tokenizer_params = data.pop("tokenizer_params")
+            else:
+                tokenizer_params = dict(
+                    padding_side="right", pad_token_type_id=0, pad_token_id=1
+                )
+
+            difference = self.max_seq_len - len(data["input_ids"])
+            if tokenizer_params["padding_side"] == "right":
+                if self.return_attention_mask:
+                    data["attention_mask"] = [1] * len(data["input_ids"]) + [
+                        0
+                    ] * difference
+                if self.return_token_type_ids:
+                    data["token_type_ids"] = (
+                        data["token_type_ids"]
+                        + [tokenizer_params["pad_token_type_id"]] * difference
+                    )
+                if self.return_special_tokens_mask:
+                    data["special_tokens_mask"] = (
+                        data["special_tokens_mask"] + [1] * difference
+                    )
+                data["input_ids"] = (
+                    data["input_ids"] + [tokenizer_params["pad_token_id"]] * difference
+                )
+                if not self.infer_mode:
+                    data["labels"] = (
+                        data["labels"] + [self.pad_token_label_id] * difference
+                    )
+                data["bbox"] = data["bbox"] + [[0, 0, 0, 0]] * difference
+            elif tokenizer_params["padding_side"] == "left":
+                if self.return_attention_mask:
+                    data["attention_mask"] = [0] * difference + [1] * len(
+                        data["input_ids"]
+                    )
+                if self.return_token_type_ids:
+                    data["token_type_ids"] = [
+                        tokenizer_params["pad_token_type_id"]
+                    ] * difference + data["token_type_ids"]
+                if self.return_special_tokens_mask:
+                    data["special_tokens_mask"] = [1] * difference + data[
+                        "special_tokens_mask"
+                    ]
+                data["input_ids"] = [
+                    tokenizer_params["pad_token_id"]
+                ] * difference + data["input_ids"]
+                if not self.infer_mode:
+                    data["labels"] = [self.pad_token_label_id] * difference + data[
+                        "labels"
+                    ]
+                data["bbox"] = [[0, 0, 0, 0]] * difference + data["bbox"]
+        else:
+            if self.return_attention_mask:
+                data["attention_mask"] = [1] * len(data["input_ids"])
+
+        for key in data:
+            if key in [
+                "input_ids",
+                "labels",
+                "token_type_ids",
+                "bbox",
+                "attention_mask",
+            ]:
+                if self.infer_mode:
+                    if key != "labels":
+                        length = min(len(data[key]), self.max_seq_len)
+                        data[key] = data[key][:length]
+                    else:
+                        continue
+                data[key] = np.array(data[key], dtype="int64")
+        return data
diff --git a/ppocr/data/imaug/vqa/token/vqa_token_relation.py b/ppocr/data/imaug/vqa/token/vqa_token_relation.py
new file mode 100644
index 0000000..1946c58
--- /dev/null
+++ b/ppocr/data/imaug/vqa/token/vqa_token_relation.py
@@ -0,0 +1,76 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+class VQAReTokenRelation(object):
+    def __init__(self, **kwargs):
+        pass
+
+    def __call__(self, data):
+        """
+        build relations
+        """
+        entities = data["entities"]
+        relations = data["relations"]
+        id2label = data.pop("id2label")
+        empty_entity = data.pop("empty_entity")
+        entity_id_to_index_map = data.pop("entity_id_to_index_map")
+
+        relations = list(set(relations))
+        relations = [
+            rel
+            for rel in relations
+            if rel[0] not in empty_entity and rel[1] not in empty_entity
+        ]
+        kv_relations = []
+        for rel in relations:
+            pair = [id2label[rel[0]], id2label[rel[1]]]
+            if pair == ["question", "answer"]:
+                kv_relations.append(
+                    {
+                        "head": entity_id_to_index_map[rel[0]],
+                        "tail": entity_id_to_index_map[rel[1]],
+                    }
+                )
+            elif pair == ["answer", "question"]:
+                kv_relations.append(
+                    {
+                        "head": entity_id_to_index_map[rel[1]],
+                        "tail": entity_id_to_index_map[rel[0]],
+                    }
+                )
+            else:
+                continue
+        relations = sorted(
+            [
+                {
+                    "head": rel["head"],
+                    "tail": rel["tail"],
+                    "start_index": self.get_relation_span(rel, entities)[0],
+                    "end_index": self.get_relation_span(rel, entities)[1],
+                }
+                for rel in kv_relations
+            ],
+            key=lambda x: x["head"],
+        )
+
+        data["relations"] = relations
+        return data
+
+    def get_relation_span(self, rel, entities):
+        bound = []
+        for entity_index in [rel["head"], rel["tail"]]:
+            bound.append(entities[entity_index]["start"])
+            bound.append(entities[entity_index]["end"])
+        return min(bound), max(bound)
diff --git a/ppocr/data/latexocr_dataset.py b/ppocr/data/latexocr_dataset.py
new file mode 100644
index 0000000..d748e08
--- /dev/null
+++ b/ppocr/data/latexocr_dataset.py
@@ -0,0 +1,206 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+This code is refer from:
+https://github.com/lukas-blecher/LaTeX-OCR/blob/main/pix2tex/dataset/dataset.py
+"""
+
+import numpy as np
+import cv2
+import math
+import os
+import json
+import pickle
+import random
+import traceback
+import paddle
+from paddle.io import Dataset
+from .imaug.label_ops import LatexOCRLabelEncode
+from .imaug import transform, create_operators
+
+
+_ALLOWED_PICKLE_GLOBALS = {
+    ("builtins", "dict"): dict,
+    ("builtins", "list"): list,
+    ("builtins", "tuple"): tuple,
+    ("builtins", "set"): set,
+    ("builtins", "frozenset"): frozenset,
+    ("builtins", "str"): str,
+    ("builtins", "int"): int,
+    ("builtins", "float"): float,
+    ("builtins", "bool"): bool,
+    ("builtins", "bytes"): bytes,
+}
+
+
+class _RestrictedDatasetUnpickler(pickle.Unpickler):
+    def find_class(self, module, name):
+        allowed = _ALLOWED_PICKLE_GLOBALS.get((module, name))
+        if allowed is None:
+            raise pickle.UnpicklingError(
+                "Unsupported pickle global in LaTeXOCR dataset payload"
+            )
+        return allowed
+
+
+def _restricted_pickle_load(file_obj):
+    return _RestrictedDatasetUnpickler(file_obj).load()
+
+
+class LaTeXOCRDataSet(Dataset):
+    def __init__(self, config, mode, logger, seed=None):
+        super(LaTeXOCRDataSet, self).__init__()
+        self.logger = logger
+        self.mode = mode.lower()
+
+        global_config = config["Global"]
+        dataset_config = config[mode]["dataset"]
+        loader_config = config[mode]["loader"]
+
+        pkl_path = dataset_config.pop("data")
+        self.data_dir = dataset_config["data_dir"]
+        self.min_dimensions = dataset_config.pop("min_dimensions")
+        self.max_dimensions = dataset_config.pop("max_dimensions")
+        self.batchsize = dataset_config.pop("batch_size_per_pair")
+        self.keep_smaller_batches = dataset_config.pop("keep_smaller_batches")
+        self.max_seq_len = global_config.pop("max_seq_len")
+        self.rec_char_dict_path = global_config.pop("rec_char_dict_path")
+        self.tokenizer = LatexOCRLabelEncode(self.rec_char_dict_path)
+
+        with open(pkl_path, "rb") as file:
+            data = _restricted_pickle_load(file)
+        if not isinstance(data, dict):
+            raise pickle.UnpicklingError(
+                "LaTeXOCR dataset payload must deserialize to a dict"
+            )
+        temp = {}
+        for k in data:
+            if (
+                self.min_dimensions[0] <= k[0] <= self.max_dimensions[0]
+                and self.min_dimensions[1] <= k[1] <= self.max_dimensions[1]
+            ):
+                temp[k] = data[k]
+        self.data = temp
+        self.do_shuffle = loader_config["shuffle"]
+        self.seed = seed
+
+        if self.mode == "train" and self.do_shuffle:
+            random.seed(self.seed)
+        self.pairs = []
+        for k in self.data:
+            info = np.array(self.data[k], dtype=object)
+            p = (
+                paddle.randperm(len(info))
+                if self.mode == "train" and self.do_shuffle
+                else paddle.arange(len(info))
+            )
+            for i in range(0, len(info), self.batchsize):
+                batch = info[p[i : i + self.batchsize]]
+                if len(batch.shape) == 1:
+                    batch = batch[None, :]
+                if len(batch) < self.batchsize and not self.keep_smaller_batches:
+                    continue
+                self.pairs.append(batch)
+        if self.do_shuffle:
+            self.pairs = np.random.permutation(np.array(self.pairs, dtype=object))
+        else:
+            self.pairs = np.array(self.pairs, dtype=object)
+
+        self.size = len(self.pairs)
+        self.set_epoch_as_seed(self.seed, dataset_config)
+
+        self.ops = create_operators(dataset_config["transforms"], global_config)
+        self.ext_op_transform_idx = dataset_config.get("ext_op_transform_idx", 2)
+        self.need_reset = True
+
+    def set_epoch_as_seed(self, seed, dataset_config):
+        if self.mode == "train":
+            try:
+                border_map_id = [
+                    index
+                    for index, dictionary in enumerate(dataset_config["transforms"])
+                    if "MakeBorderMap" in dictionary
+                ][0]
+                shrink_map_id = [
+                    index
+                    for index, dictionary in enumerate(dataset_config["transforms"])
+                    if "MakeShrinkMap" in dictionary
+                ][0]
+                dataset_config["transforms"][border_map_id]["MakeBorderMap"][
+                    "epoch"
+                ] = (seed if seed is not None else 0)
+                dataset_config["transforms"][shrink_map_id]["MakeShrinkMap"][
+                    "epoch"
+                ] = (seed if seed is not None else 0)
+            except Exception as E:
+                print(E)
+                return
+
+    def shuffle_data_random(self):
+        random.seed(self.seed)
+        random.shuffle(self.data_lines)
+        return
+
+    def __getitem__(self, idx):
+        batch = self.pairs[idx]
+        eqs, ims = batch.T
+        try:
+            max_width, max_height, max_length = 0, 0, 0
+
+            images_transform = []
+
+            for file_name in ims:
+                img_path = os.path.join(self.data_dir, file_name)
+                data = {
+                    "img_path": img_path,
+                }
+                with open(data["img_path"], "rb") as f:
+                    img = f.read()
+                    data["image"] = img
+                    item = transform(data, self.ops)
+                    images_transform.append(np.array(item[0]))
+            image_concat = np.concatenate(images_transform, axis=0)[:, np.newaxis, :, :]
+            images_transform = image_concat.astype(np.float32)
+            labels, attention_mask, max_length = self.tokenizer(list(eqs))
+            if self.max_seq_len < max_length:
+                rnd_idx = (
+                    np.random.randint(self.__len__())
+                    if self.mode == "train"
+                    else (idx + 1) % self.__len__()
+                )
+                return self.__getitem__(rnd_idx)
+            return (images_transform, labels, attention_mask)
+
+        except:
+
+            self.logger.error(
+                "When parsing line {}, error happened with msg: {}".format(
+                    data["img_path"], traceback.format_exc()
+                )
+            )
+            outs = None
+
+        if outs is None:
+            # during evaluation, we should fix the idx to get same results for many times of evaluation.
+            rnd_idx = (
+                np.random.randint(self.__len__())
+                if self.mode == "train"
+                else (idx + 1) % self.__len__()
+            )
+            return self.__getitem__(rnd_idx)
+        return outs
+
+    def __len__(self):
+        return self.size
diff --git a/ppocr/data/lmdb_dataset.py b/ppocr/data/lmdb_dataset.py
new file mode 100644
index 0000000..6de2802
--- /dev/null
+++ b/ppocr/data/lmdb_dataset.py
@@ -0,0 +1,334 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import numpy as np
+import io
+import os
+from paddle.io import Dataset
+import lmdb
+import cv2
+import string
+import pickle
+from PIL import Image
+
+from .imaug import transform, create_operators
+
+_ALLOWED_PICKLE_GLOBALS = {
+    ("builtins", "bytes"),
+    ("builtins", "bytearray"),
+    ("builtins", "str"),
+    ("builtins", "tuple"),
+    ("builtins", "list"),
+    ("builtins", "dict"),
+    ("builtins", "set"),
+    ("builtins", "frozenset"),
+    ("builtins", "int"),
+    ("builtins", "float"),
+    ("builtins", "bool"),
+}
+
+
+class _RestrictedDatasetUnpickler(pickle.Unpickler):
+    def find_class(self, module, name):
+        if (module, name) in _ALLOWED_PICKLE_GLOBALS:
+            return super().find_class(module, name)
+        raise pickle.UnpicklingError(
+            "Unsupported pickle payload in LMDBDataSetTableMaster dataset"
+        )
+
+
+def _restricted_pickle_loads(data):
+    if data is None:
+        raise ValueError("Missing LMDB dataset value")
+    return _RestrictedDatasetUnpickler(io.BytesIO(data)).load()
+
+
+class LMDBDataSet(Dataset):
+    def __init__(self, config, mode, logger, seed=None):
+        super(LMDBDataSet, self).__init__()
+
+        global_config = config["Global"]
+        dataset_config = config[mode]["dataset"]
+        loader_config = config[mode]["loader"]
+        batch_size = loader_config["batch_size_per_card"]
+        data_dir = dataset_config["data_dir"]
+        self.do_shuffle = loader_config["shuffle"]
+
+        self.lmdb_sets = self.load_hierarchical_lmdb_dataset(data_dir)
+        logger.info("Initialize indexes of datasets:%s" % data_dir)
+        self.data_idx_order_list = self.dataset_traversal()
+        if self.do_shuffle:
+            np.random.shuffle(self.data_idx_order_list)
+        self.ops = create_operators(dataset_config["transforms"], global_config)
+        self.ext_op_transform_idx = dataset_config.get("ext_op_transform_idx", 1)
+
+        ratio_list = dataset_config.get("ratio_list", [1.0])
+        self.need_reset = True in [x < 1 for x in ratio_list]
+
+    def load_hierarchical_lmdb_dataset(self, data_dir):
+        lmdb_sets = {}
+        dataset_idx = 0
+        for dirpath, dirnames, filenames in os.walk(data_dir + "/"):
+            if not dirnames:
+                env = lmdb.open(
+                    dirpath,
+                    max_readers=32,
+                    readonly=True,
+                    lock=False,
+                    readahead=False,
+                    meminit=False,
+                )
+                txn = env.begin(write=False)
+                num_samples = int(txn.get("num-samples".encode()))
+                lmdb_sets[dataset_idx] = {
+                    "dirpath": dirpath,
+                    "env": env,
+                    "txn": txn,
+                    "num_samples": num_samples,
+                }
+                dataset_idx += 1
+        return lmdb_sets
+
+    def dataset_traversal(self):
+        lmdb_num = len(self.lmdb_sets)
+        total_sample_num = 0
+        for lno in range(lmdb_num):
+            total_sample_num += self.lmdb_sets[lno]["num_samples"]
+        data_idx_order_list = np.zeros((total_sample_num, 2))
+        beg_idx = 0
+        for lno in range(lmdb_num):
+            tmp_sample_num = self.lmdb_sets[lno]["num_samples"]
+            end_idx = beg_idx + tmp_sample_num
+            data_idx_order_list[beg_idx:end_idx, 0] = lno
+            data_idx_order_list[beg_idx:end_idx, 1] = list(range(tmp_sample_num))
+            data_idx_order_list[beg_idx:end_idx, 1] += 1
+            beg_idx = beg_idx + tmp_sample_num
+        return data_idx_order_list
+
+    def get_img_data(self, value):
+        """get_img_data"""
+        if not value:
+            return None
+        imgdata = np.frombuffer(value, dtype="uint8")
+        if imgdata is None:
+            return None
+        imgori = cv2.imdecode(imgdata, 1)
+        if imgori is None:
+            return None
+        return imgori
+
+    def get_ext_data(self):
+        ext_data_num = 0
+        for op in self.ops:
+            if hasattr(op, "ext_data_num"):
+                ext_data_num = getattr(op, "ext_data_num")
+                break
+        load_data_ops = self.ops[: self.ext_op_transform_idx]
+        ext_data = []
+
+        while len(ext_data) < ext_data_num:
+            lmdb_idx, file_idx = self.data_idx_order_list[np.random.randint(len(self))]
+            lmdb_idx = int(lmdb_idx)
+            file_idx = int(file_idx)
+            sample_info = self.get_lmdb_sample_info(
+                self.lmdb_sets[lmdb_idx]["txn"], file_idx
+            )
+            if sample_info is None:
+                continue
+            img, label = sample_info
+            data = {"image": img, "label": label}
+            data = transform(data, load_data_ops)
+            if data is None:
+                continue
+            ext_data.append(data)
+        return ext_data
+
+    def get_lmdb_sample_info(self, txn, index):
+        label_key = "label-%09d".encode() % index
+        label = txn.get(label_key)
+        if label is None:
+            return None
+        label = label.decode("utf-8")
+        img_key = "image-%09d".encode() % index
+        imgbuf = txn.get(img_key)
+        return imgbuf, label
+
+    def __getitem__(self, idx):
+        lmdb_idx, file_idx = self.data_idx_order_list[idx]
+        lmdb_idx = int(lmdb_idx)
+        file_idx = int(file_idx)
+        sample_info = self.get_lmdb_sample_info(
+            self.lmdb_sets[lmdb_idx]["txn"], file_idx
+        )
+        if sample_info is None:
+            return self.__getitem__(np.random.randint(self.__len__()))
+        img, label = sample_info
+        data = {"image": img, "label": label}
+        data["ext_data"] = self.get_ext_data()
+        outs = transform(data, self.ops)
+        if outs is None:
+            return self.__getitem__(np.random.randint(self.__len__()))
+        return outs
+
+    def __len__(self):
+        return self.data_idx_order_list.shape[0]
+
+
+class LMDBDataSetSR(LMDBDataSet):
+    def buf2PIL(self, txn, key, type="RGB"):
+        imgbuf = txn.get(key)
+        buf = io.BytesIO()
+        buf.write(imgbuf)
+        buf.seek(0)
+        im = Image.open(buf).convert(type)
+        return im
+
+    def str_filt(self, str_, voc_type):
+        alpha_dict = {
+            "digit": string.digits,
+            "lower": string.digits + string.ascii_lowercase,
+            "upper": string.digits + string.ascii_letters,
+            "all": string.digits + string.ascii_letters + string.punctuation,
+        }
+        if voc_type == "lower":
+            str_ = str_.lower()
+        for char in str_:
+            if char not in alpha_dict[voc_type]:
+                str_ = str_.replace(char, "")
+        return str_
+
+    def get_lmdb_sample_info(self, txn, index):
+        self.voc_type = "upper"
+        self.max_len = 100
+        self.test = False
+        label_key = b"label-%09d" % index
+        word = str(txn.get(label_key).decode())
+        img_HR_key = b"image_hr-%09d" % index  # 128*32
+        img_lr_key = b"image_lr-%09d" % index  # 64*16
+        try:
+            img_HR = self.buf2PIL(txn, img_HR_key, "RGB")
+            img_lr = self.buf2PIL(txn, img_lr_key, "RGB")
+        except IOError or len(word) > self.max_len:
+            return self[index + 1]
+        label_str = self.str_filt(word, self.voc_type)
+        return img_HR, img_lr, label_str
+
+    def __getitem__(self, idx):
+        lmdb_idx, file_idx = self.data_idx_order_list[idx]
+        lmdb_idx = int(lmdb_idx)
+        file_idx = int(file_idx)
+        sample_info = self.get_lmdb_sample_info(
+            self.lmdb_sets[lmdb_idx]["txn"], file_idx
+        )
+        if sample_info is None:
+            return self.__getitem__(np.random.randint(self.__len__()))
+        img_HR, img_lr, label_str = sample_info
+        data = {"image_hr": img_HR, "image_lr": img_lr, "label": label_str}
+        outs = transform(data, self.ops)
+        if outs is None:
+            return self.__getitem__(np.random.randint(self.__len__()))
+        return outs
+
+
+class LMDBDataSetTableMaster(LMDBDataSet):
+    def load_hierarchical_lmdb_dataset(self, data_dir):
+        lmdb_sets = {}
+        dataset_idx = 0
+        env = lmdb.open(
+            data_dir,
+            max_readers=32,
+            readonly=True,
+            lock=False,
+            readahead=False,
+            meminit=False,
+        )
+        txn = env.begin(write=False)
+        num_samples = _restricted_pickle_loads(txn.get(b"__len__"))
+        if not isinstance(num_samples, int):
+            raise ValueError("Invalid LMDB dataset length metadata")
+        lmdb_sets[dataset_idx] = {
+            "dirpath": data_dir,
+            "env": env,
+            "txn": txn,
+            "num_samples": num_samples,
+        }
+        return lmdb_sets
+
+    def get_img_data(self, value):
+        """get_img_data"""
+        if not value:
+            return None
+        imgdata = np.frombuffer(value, dtype="uint8")
+        if imgdata is None:
+            return None
+        imgori = cv2.imdecode(imgdata, 1)
+        if imgori is None:
+            return None
+        return imgori
+
+    def get_lmdb_sample_info(self, txn, index):
+        def convert_bbox(bbox_str_list):
+            bbox_list = []
+            for bbox_str in bbox_str_list:
+                bbox_list.append(int(bbox_str))
+            return bbox_list
+
+        try:
+            data = _restricted_pickle_loads(txn.get(str(index).encode("utf8")))
+        except (pickle.UnpicklingError, ValueError, TypeError, EOFError):
+            return None
+        if not isinstance(data, (list, tuple)) or len(data) < 3:
+            return None
+
+        # img_name, img, info_lines
+        file_name = data[0]
+        bytes = data[1]
+        info_lines = data[2]  # raw data from TableMASTER annotation file.
+        # parse info_lines
+        raw_data = info_lines.strip().split("\n")
+        raw_name, text = (
+            raw_data[0],
+            raw_data[1],
+        )  # don't filter the samples's length over max_seq_len.
+        text = text.split(",")
+        bbox_str_list = raw_data[2:]
+        bbox_split = ","
+        bboxes = [
+            {"bbox": convert_bbox(bsl.strip().split(bbox_split)), "tokens": ["1", "2"]}
+            for bsl in bbox_str_list
+        ]
+
+        # advance parse bbox
+        # import pdb;pdb.set_trace()
+
+        line_info = {}
+        line_info["file_name"] = file_name
+        line_info["structure"] = text
+        line_info["cells"] = bboxes
+        line_info["image"] = bytes
+        return line_info
+
+    def __getitem__(self, idx):
+        lmdb_idx, file_idx = self.data_idx_order_list[idx]
+        lmdb_idx = int(lmdb_idx)
+        file_idx = int(file_idx)
+        data = self.get_lmdb_sample_info(self.lmdb_sets[lmdb_idx]["txn"], file_idx)
+        if data is None:
+            return self.__getitem__(np.random.randint(self.__len__()))
+        outs = transform(data, self.ops)
+        if outs is None:
+            return self.__getitem__(np.random.randint(self.__len__()))
+        return outs
+
+    def __len__(self):
+        return self.data_idx_order_list.shape[0]
diff --git a/ppocr/data/multi_scale_sampler.py b/ppocr/data/multi_scale_sampler.py
new file mode 100644
index 0000000..4ab38fc
--- /dev/null
+++ b/ppocr/data/multi_scale_sampler.py
@@ -0,0 +1,171 @@
+from paddle.io import Sampler
+import paddle.distributed as dist
+
+import numpy as np
+import random
+import math
+
+
+class MultiScaleSampler(Sampler):
+    def __init__(
+        self,
+        data_source,
+        scales,
+        first_bs=128,
+        fix_bs=True,
+        divided_factor=[8, 16],
+        is_training=True,
+        ratio_wh=0.8,
+        max_w=480.0,
+        seed=None,
+    ):
+        """
+        multi scale samper
+        Args:
+            data_source(dataset)
+            scales(list): several scales for image resolution
+            first_bs(int): batch size for the first scale in scales
+            divided_factor(list[w, h]): ImageNet models down-sample images by a factor, ensure that width and height dimensions are multiples are multiple of devided_factor.
+            is_training(boolean): mode
+        """
+        # min. and max. spatial dimensions
+        self.data_source = data_source
+        self.data_idx_order_list = np.array(data_source.data_idx_order_list)
+        self.ds_width = data_source.ds_width
+        self.seed = data_source.seed
+        if self.ds_width:
+            self.wh_ratio = data_source.wh_ratio
+            self.wh_ratio_sort = data_source.wh_ratio_sort
+        self.n_data_samples = len(self.data_source)
+        self.ratio_wh = ratio_wh
+        self.max_w = max_w
+
+        if isinstance(scales[0], list):
+            width_dims = [i[0] for i in scales]
+            height_dims = [i[1] for i in scales]
+        elif isinstance(scales[0], int):
+            width_dims = scales
+            height_dims = scales
+        base_im_w = width_dims[0]
+        base_im_h = height_dims[0]
+        base_batch_size = first_bs
+
+        # Get the GPU and node related information
+        num_replicas = dist.get_world_size()
+        rank = dist.get_rank()
+        # adjust the total samples to avoid batch dropping
+        num_samples_per_replica = int(self.n_data_samples * 1.0 / num_replicas)
+
+        img_indices = [idx for idx in range(self.n_data_samples)]
+
+        self.shuffle = False
+        if is_training:
+            # compute the spatial dimensions and corresponding batch size
+            # ImageNet models down-sample images by a factor of 32.
+            # Ensure that width and height dimensions are multiples are multiple of 32.
+            width_dims = [
+                int((w // divided_factor[0]) * divided_factor[0]) for w in width_dims
+            ]
+            height_dims = [
+                int((h // divided_factor[1]) * divided_factor[1]) for h in height_dims
+            ]
+
+            img_batch_pairs = list()
+            base_elements = base_im_w * base_im_h * base_batch_size
+            for h, w in zip(height_dims, width_dims):
+                if fix_bs:
+                    batch_size = base_batch_size
+                else:
+                    batch_size = int(max(1, (base_elements / (h * w))))
+                img_batch_pairs.append((w, h, batch_size))
+            self.img_batch_pairs = img_batch_pairs
+            self.shuffle = True
+        else:
+            self.img_batch_pairs = [(base_im_w, base_im_h, base_batch_size)]
+
+        self.img_indices = img_indices
+        self.n_samples_per_replica = num_samples_per_replica
+        self.epoch = 0
+        self.rank = rank
+        self.num_replicas = num_replicas
+
+        self.batch_list = []
+        self.current = 0
+        last_index = num_samples_per_replica * num_replicas
+        indices_rank_i = self.img_indices[self.rank : last_index : self.num_replicas]
+        while self.current < self.n_samples_per_replica:
+            for curr_w, curr_h, curr_bsz in self.img_batch_pairs:
+                end_index = min(self.current + curr_bsz, self.n_samples_per_replica)
+                batch_ids = indices_rank_i[self.current : end_index]
+                n_batch_samples = len(batch_ids)
+                if n_batch_samples != curr_bsz:
+                    batch_ids += indices_rank_i[: (curr_bsz - n_batch_samples)]
+                self.current += curr_bsz
+
+                if len(batch_ids) > 0:
+                    batch = [curr_w, curr_h, len(batch_ids)]
+                    self.batch_list.append(batch)
+        random.shuffle(self.batch_list)
+        self.length = len(self.batch_list)
+        self.batchs_in_one_epoch = self.iter()
+        self.batchs_in_one_epoch_id = [i for i in range(len(self.batchs_in_one_epoch))]
+
+    def __iter__(self):
+        if self.seed is None:
+            random.seed(self.epoch)
+            self.epoch += 1
+        else:
+            random.seed(self.seed)
+        random.shuffle(self.batchs_in_one_epoch_id)
+        for batch_tuple_id in self.batchs_in_one_epoch_id:
+            yield self.batchs_in_one_epoch[batch_tuple_id]
+
+    def iter(self):
+        if self.shuffle:
+            if self.seed is not None:
+                random.seed(self.seed)
+            else:
+                random.seed(self.epoch)
+            if not self.ds_width:
+                random.shuffle(self.img_indices)
+            random.shuffle(self.img_batch_pairs)
+            indices_rank_i = self.img_indices[
+                self.rank : len(self.img_indices) : self.num_replicas
+            ]
+        else:
+            indices_rank_i = self.img_indices[
+                self.rank : len(self.img_indices) : self.num_replicas
+            ]
+
+        start_index = 0
+        batchs_in_one_epoch = []
+        for batch_tuple in self.batch_list:
+            curr_w, curr_h, curr_bsz = batch_tuple
+            end_index = min(start_index + curr_bsz, self.n_samples_per_replica)
+            batch_ids = indices_rank_i[start_index:end_index]
+            n_batch_samples = len(batch_ids)
+            if n_batch_samples != curr_bsz:
+                batch_ids += indices_rank_i[: (curr_bsz - n_batch_samples)]
+            start_index += curr_bsz
+
+            if len(batch_ids) > 0:
+                if self.ds_width:
+                    wh_ratio_current = self.wh_ratio[self.wh_ratio_sort[batch_ids]]
+                    ratio_current = wh_ratio_current.mean()
+                    ratio_current = (
+                        ratio_current
+                        if ratio_current * curr_h < self.max_w
+                        else self.max_w / curr_h
+                    )
+                else:
+                    ratio_current = None
+                batch = [(curr_w, curr_h, b_id, ratio_current) for b_id in batch_ids]
+                # yield batch
+                batchs_in_one_epoch.append(batch)
+        return batchs_in_one_epoch
+
+    def set_epoch(self, epoch: int):
+        self.epoch = epoch
+
+    def __len__(self):
+        return self.length
diff --git a/ppocr/data/pgnet_dataset.py b/ppocr/data/pgnet_dataset.py
new file mode 100644
index 0000000..fd7a3d1
--- /dev/null
+++ b/ppocr/data/pgnet_dataset.py
@@ -0,0 +1,107 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import numpy as np
+import os
+from paddle.io import Dataset
+from .imaug import transform, create_operators
+import random
+
+
+class PGDataSet(Dataset):
+    def __init__(self, config, mode, logger, seed=None):
+        super(PGDataSet, self).__init__()
+
+        self.logger = logger
+        self.seed = seed
+        self.mode = mode
+        global_config = config["Global"]
+        dataset_config = config[mode]["dataset"]
+        loader_config = config[mode]["loader"]
+
+        self.delimiter = dataset_config.get("delimiter", "\t")
+        label_file_list = dataset_config.pop("label_file_list")
+        data_source_num = len(label_file_list)
+        ratio_list = dataset_config.get("ratio_list", [1.0])
+        if isinstance(ratio_list, (float, int)):
+            ratio_list = [float(ratio_list)] * int(data_source_num)
+        assert (
+            len(ratio_list) == data_source_num
+        ), "The length of ratio_list should be the same as the file_list."
+        self.data_dir = dataset_config["data_dir"]
+        self.do_shuffle = loader_config["shuffle"]
+
+        logger.info("Initialize indexes of datasets:%s" % label_file_list)
+        self.data_lines = self.get_image_info_list(label_file_list, ratio_list)
+        self.data_idx_order_list = list(range(len(self.data_lines)))
+        if mode.lower() == "train":
+            self.shuffle_data_random()
+
+        self.ops = create_operators(dataset_config["transforms"], global_config)
+
+        self.need_reset = True in [x < 1 for x in ratio_list]
+
+    def shuffle_data_random(self):
+        if self.do_shuffle:
+            random.seed(self.seed)
+            random.shuffle(self.data_lines)
+        return
+
+    def get_image_info_list(self, file_list, ratio_list):
+        if isinstance(file_list, str):
+            file_list = [file_list]
+        data_lines = []
+        for idx, file in enumerate(file_list):
+            with open(file, "rb") as f:
+                lines = f.readlines()
+                if self.mode == "train" or ratio_list[idx] < 1.0:
+                    random.seed(self.seed)
+                    lines = random.sample(lines, round(len(lines) * ratio_list[idx]))
+                data_lines.extend(lines)
+        return data_lines
+
+    def __getitem__(self, idx):
+        file_idx = self.data_idx_order_list[idx]
+        data_line = self.data_lines[file_idx]
+        img_id = 0
+        try:
+            data_line = data_line.decode("utf-8")
+            substr = data_line.strip("\n").split(self.delimiter)
+            file_name = substr[0]
+            label = substr[1]
+            img_path = os.path.join(self.data_dir, file_name)
+            if self.mode.lower() == "eval":
+                try:
+                    img_id = int(data_line.split(".")[0][7:])
+                except:
+                    img_id = 0
+            data = {"img_path": img_path, "label": label, "img_id": img_id}
+            if not os.path.exists(img_path):
+                raise Exception("{} does not exist!".format(img_path))
+            with open(data["img_path"], "rb") as f:
+                img = f.read()
+                data["image"] = img
+            outs = transform(data, self.ops)
+        except Exception as e:
+            self.logger.error(
+                "When parsing line {}, error happened with msg: {}".format(
+                    self.data_idx_order_list[idx], e
+                )
+            )
+            outs = None
+        if outs is None:
+            return self.__getitem__(np.random.randint(self.__len__()))
+        return outs
+
+    def __len__(self):
+        return len(self.data_idx_order_list)
diff --git a/ppocr/data/pubtab_dataset.py b/ppocr/data/pubtab_dataset.py
new file mode 100644
index 0000000..f0f9a04
--- /dev/null
+++ b/ppocr/data/pubtab_dataset.py
@@ -0,0 +1,138 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import numpy as np
+import os
+import random
+from paddle.io import Dataset
+import json
+from copy import deepcopy
+
+from .imaug import transform, create_operators
+
+
+class PubTabDataSet(Dataset):
+    def __init__(self, config, mode, logger, seed=None):
+        super(PubTabDataSet, self).__init__()
+        self.logger = logger
+
+        global_config = config["Global"]
+        dataset_config = config[mode]["dataset"]
+        loader_config = config[mode]["loader"]
+
+        label_file_list = dataset_config.pop("label_file_list")
+        data_source_num = len(label_file_list)
+        ratio_list = dataset_config.get("ratio_list", [1.0])
+        if isinstance(ratio_list, (float, int)):
+            ratio_list = [float(ratio_list)] * int(data_source_num)
+
+        assert (
+            len(ratio_list) == data_source_num
+        ), "The length of ratio_list should be the same as the file_list."
+
+        self.data_dir = dataset_config["data_dir"]
+        self.do_shuffle = loader_config["shuffle"]
+
+        self.seed = seed
+        self.mode = mode.lower()
+        logger.info("Initialize indexes of datasets:%s" % label_file_list)
+        self.data_lines = self.get_image_info_list(label_file_list, ratio_list)
+        # self.check(config['Global']['max_text_length'])
+
+        if mode.lower() == "train" and self.do_shuffle:
+            self.shuffle_data_random()
+        self.ops = create_operators(dataset_config["transforms"], global_config)
+        self.need_reset = True in [x < 1 for x in ratio_list]
+
+    def get_image_info_list(self, file_list, ratio_list):
+        if isinstance(file_list, str):
+            file_list = [file_list]
+        data_lines = []
+        for idx, file in enumerate(file_list):
+            with open(file, "rb") as f:
+                lines = f.readlines()
+                if self.mode == "train" or ratio_list[idx] < 1.0:
+                    random.seed(self.seed)
+                    lines = random.sample(lines, round(len(lines) * ratio_list[idx]))
+                data_lines.extend(lines)
+        return data_lines
+
+    def check(self, max_text_length):
+        data_lines = []
+        for line in self.data_lines:
+            data_line = line.decode("utf-8").strip("\n")
+            info = json.loads(data_line)
+            file_name = info["filename"]
+            cells = info["html"]["cells"].copy()
+            structure = info["html"]["structure"]["tokens"].copy()
+
+            img_path = os.path.join(self.data_dir, file_name)
+            if not os.path.exists(img_path):
+                self.logger.warning("{} does not exist!".format(img_path))
+                continue
+            if len(structure) == 0 or len(structure) > max_text_length:
+                continue
+            # data = {'img_path': img_path, 'cells': cells, 'structure':structure,'file_name':file_name}
+            data_lines.append(line)
+        self.data_lines = data_lines
+
+    def shuffle_data_random(self):
+        if self.do_shuffle:
+            random.seed(self.seed)
+            random.shuffle(self.data_lines)
+        return
+
+    def __getitem__(self, idx):
+        try:
+            data_line = self.data_lines[idx]
+            data_line = data_line.decode("utf-8").strip("\n")
+            info = json.loads(data_line)
+            file_name = info["filename"]
+            cells = info["html"]["cells"].copy()
+            structure = info["html"]["structure"]["tokens"].copy()
+
+            img_path = os.path.join(self.data_dir, file_name)
+            if not os.path.exists(img_path):
+                raise Exception("{} does not exist!".format(img_path))
+            data = {
+                "img_path": img_path,
+                "cells": cells,
+                "structure": structure,
+                "file_name": file_name,
+            }
+
+            with open(data["img_path"], "rb") as f:
+                img = f.read()
+                data["image"] = img
+            outs = transform(data, self.ops)
+        except:
+            import traceback
+
+            err = traceback.format_exc()
+            self.logger.error(
+                "When parsing line {}, error happened with msg: {}".format(
+                    data_line, err
+                )
+            )
+            outs = None
+        if outs is None:
+            rnd_idx = (
+                np.random.randint(self.__len__())
+                if self.mode == "train"
+                else (idx + 1) % self.__len__()
+            )
+            return self.__getitem__(rnd_idx)
+        return outs
+
+    def __len__(self):
+        return len(self.data_lines)
diff --git a/ppocr/data/simple_dataset.py b/ppocr/data/simple_dataset.py
new file mode 100644
index 0000000..57e9b43
--- /dev/null
+++ b/ppocr/data/simple_dataset.py
@@ -0,0 +1,544 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import numpy as np
+import cv2
+import math
+import os
+import json
+import random
+import traceback
+import multiprocessing
+import urllib.request
+import urllib.parse
+import threading
+import concurrent.futures
+from collections import OrderedDict
+from paddle.io import Dataset
+from .imaug import transform, create_operators
+from paddle import get_device
+
+# ------------------------------------------------------------------ #
+#  Per-worker-process URL prefetch cache
+#
+#  Each DataLoader worker is a forked process with its own copy of
+#  these globals.  The thread pool and LRU cache are therefore
+#  completely independent across workers — no cross-process locking
+#  needed, and memory usage is bounded per worker.
+#
+#  Memory budget (per worker):
+#    _URL_CACHE_MAX × avg_image_size  ≈  200 × 270 KB  ≈  54 MB
+#  With num_workers=4 the total extra footprint is ~216 MB.
+#
+#  How prefetch works:
+#    _ensure_index_map() fires inside the worker when an epoch changes.
+#    It calls _prefetch_epoch_urls(), which scans the new _index_map,
+#    picks the first _URL_PREFETCH_SUBMIT URL entries (epoch order ≈
+#    access order), and submits them to the background thread pool.
+#    _load_image_bytes() checks the cache / in-flight future before
+#    falling back to a synchronous download.
+# ------------------------------------------------------------------ #
+
+_URL_CACHE_MAX = 200  # max cached images per worker
+_URL_PREFETCH_SUBMIT = 200  # URL items submitted to thread pool per epoch
+
+# LRU cache: url -> bytes
+_url_cache: "OrderedDict[str, bytes]" = OrderedDict()
+_url_cache_lock = threading.Lock()
+
+# In-flight futures: url -> Future
+_url_futures: "dict[str, concurrent.futures.Future]" = {}
+_url_futures_lock = threading.Lock()
+
+# Lazily created per-process thread pool
+_url_executor: "concurrent.futures.ThreadPoolExecutor | None" = None
+_url_executor_lock = threading.Lock()
+
+
+def _get_url_executor():
+    global _url_executor
+    if _url_executor is None:
+        with _url_executor_lock:
+            if _url_executor is None:
+                _url_executor = concurrent.futures.ThreadPoolExecutor(
+                    max_workers=4,
+                    thread_name_prefix="url_prefetch",
+                )
+    return _url_executor
+
+
+def _encode_url(url):
+    """Percent-encode non-ASCII characters in the URL path so that
+    urllib can handle URLs containing CJK or other non-ASCII filenames.
+    Scheme, netloc, query and fragment are left untouched.
+    """
+    parts = urllib.parse.urlparse(url)
+    encoded_path = urllib.parse.quote(parts.path, safe="/:@!$&'()*+,;=")
+    return urllib.parse.urlunparse(parts._replace(path=encoded_path))
+
+
+def _download_url_bytes(url):
+    """Download *url*, store in LRU cache, remove from futures dict.
+    The futures entry is always cleaned up (success or failure) so that
+    a failed URL can be retried on the next access.
+    """
+    encoded = _encode_url(url)
+    try:
+        with urllib.request.urlopen(encoded, timeout=30) as resp:
+            data = resp.read()
+    except Exception:
+        with _url_futures_lock:
+            _url_futures.pop(url, None)
+        raise
+    with _url_cache_lock:
+        if url not in _url_cache:
+            if len(_url_cache) >= _URL_CACHE_MAX:
+                _url_cache.popitem(last=False)  # evict LRU entry
+            _url_cache[url] = data
+        else:
+            _url_cache.move_to_end(url)
+    with _url_futures_lock:
+        _url_futures.pop(url, None)
+    return data
+
+
+def _submit_url_prefetch(url):
+    """Submit a background download for *url* if not already cached/in-flight."""
+    with _url_cache_lock:
+        if url in _url_cache:
+            return
+    with _url_futures_lock:
+        if url in _url_futures:
+            return
+        future = _get_url_executor().submit(_download_url_bytes, url)
+        _url_futures[url] = future
+
+
+def _prefetch_epoch_urls(index_map, all_lines, delimiter):
+    """Scan *index_map* and submit the first _URL_PREFETCH_SUBMIT URL
+    items for background download.  Called inside worker processes."""
+    submitted = 0
+    for file_idx in index_map:
+        if submitted >= _URL_PREFETCH_SUBMIT:
+            break
+        try:
+            line = all_lines[file_idx].decode("utf-8")
+            fname = line.strip("\n").split(delimiter)[0]
+            if fname and fname[0] == "[":  # JSON list — skip
+                continue
+            if fname.startswith("http://") or fname.startswith("https://"):
+                _submit_url_prefetch(fname)
+                submitted += 1
+        except Exception:
+            pass
+
+
+def _load_image_bytes(img_path):
+    """Return raw image bytes.  For URLs checks prefetch cache/future first."""
+    if img_path.startswith("http://") or img_path.startswith("https://"):
+        # 1. Cache hit — return immediately
+        with _url_cache_lock:
+            if img_path in _url_cache:
+                _url_cache.move_to_end(img_path)
+                return _url_cache[img_path]
+        # 2. In-flight future — wait for background download to finish
+        with _url_futures_lock:
+            future = _url_futures.get(img_path)
+        if future is not None:
+            return future.result(timeout=60)
+        # 3. Cold miss — download synchronously (also fills cache)
+        return _download_url_bytes(img_path)
+    with open(img_path, "rb") as f:
+        return f.read()
+
+
+def _img_path_exists(img_path):
+    """Return True if the image source is accessible (local file exists or URL)."""
+    if img_path.startswith("http://") or img_path.startswith("https://"):
+        return True
+    return os.path.exists(img_path)
+
+
+class SimpleDataSet(Dataset):
+    def __init__(self, config, mode, logger, seed=None):
+        super(SimpleDataSet, self).__init__()
+        self.logger = logger
+        self.mode = mode.lower()
+
+        global_config = config["Global"]
+        dataset_config = config[mode]["dataset"]
+        loader_config = config[mode]["loader"]
+
+        self.delimiter = dataset_config.get("delimiter", "\t")
+        label_file_list = dataset_config.pop("label_file_list")
+        data_source_num = len(label_file_list)
+        ratio_list = dataset_config.get("ratio_list", 1.0)
+        if isinstance(ratio_list, (float, int)):
+            ratio_list = [float(ratio_list)] * int(data_source_num)
+        self.label_file_list = label_file_list
+        self.ratio_list = ratio_list
+
+        assert (
+            len(ratio_list) == data_source_num
+        ), "The length of ratio_list should be the same as the file_list."
+        self.data_dir = dataset_config["data_dir"]
+        self.do_shuffle = loader_config["shuffle"]
+        self.seed = seed
+        self.need_reset = True in [x < 1 for x in ratio_list]
+
+        logger.info("Initialize indexs of datasets:%s" % label_file_list)
+
+        if self.need_reset:
+            # Pre-load all lines once (immutable, never re-read from disk).
+            # Per-epoch ratio sampling is done via _index_map (virtual idx -> global idx).
+            self._all_lines, self.file_boundaries = self._load_all_lines(
+                label_file_list
+            )
+            self._index_map = self._generate_index_map(seed)
+            self._cached_epoch = seed if seed is not None else 0
+            # data_lines / data_idx_order_list kept for API compat but NOT used in __getitem__
+            self.data_lines = self._all_lines
+            self.data_idx_order_list = list(range(len(self._index_map)))
+        else:
+            self._all_lines = None
+            self._index_map = None
+            self._cached_epoch = None
+            self.file_boundaries = None
+            self.data_lines = self.get_image_info_list(label_file_list, ratio_list)
+            self.data_idx_order_list = list(range(len(self.data_lines)))
+            if self.mode == "train" and self.do_shuffle:
+                self.shuffle_data_random()
+
+        # Shared epoch value: workers read this via shared memory to detect epoch changes
+        self._shared_epoch = multiprocessing.Value("i", seed if seed is not None else 0)
+
+        self.ops = create_operators(dataset_config["transforms"], global_config)
+        self.ext_op_transform_idx = dataset_config.get("ext_op_transform_idx", 2)
+
+    # ------------------------------------------------------------------ #
+    #  Data loading helpers
+    # ------------------------------------------------------------------ #
+
+    def _load_all_lines(self, file_list):
+        """Read all label files once. Returns (all_lines, file_boundaries)."""
+        if isinstance(file_list, str):
+            file_list = [file_list]
+        all_lines = []
+        boundaries = [0]
+        for f in file_list:
+            with open(f, "rb") as fh:
+                lines = fh.readlines()
+                all_lines.extend(lines)
+                boundaries.append(len(all_lines))
+        return all_lines, boundaries
+
+    def _generate_index_map(self, seed):
+        """Generate virtual-index -> global-index mapping.
+
+        Replicates the EXACT sampling logic of original get_image_info_list +
+        shuffle_data_random: for each file, random.seed(seed) then
+        random.sample to pick indices, then random.seed(seed) + shuffle.
+
+        Since random.sample(population, k) with the same seed selects the
+        same POSITIONS regardless of population type, sampling from
+        range(start, end) yields the same positions as from lines[start:end].
+        """
+        sampled = []
+        for i in range(len(self.ratio_list)):
+            start = self.file_boundaries[i]
+            end = self.file_boundaries[i + 1]
+            file_size = end - start
+            count = round(file_size * self.ratio_list[i])
+            if self.mode == "train" or self.ratio_list[i] < 1.0:
+                random.seed(seed)
+                sampled.extend(random.sample(range(start, end), count))
+            else:
+                sampled.extend(range(start, end))
+        if self.mode == "train" and self.do_shuffle:
+            random.seed(seed)
+            random.shuffle(sampled)
+        return sampled
+
+    def _ensure_index_map(self):
+        """Lazily rebuild _index_map when worker detects epoch change via shared memory.
+        Also triggers URL prefetch on first call (epoch 0) and on every epoch change.
+        """
+        if self._all_lines is None:
+            return
+        current_epoch = self._shared_epoch.value
+        epoch_changed = current_epoch != self._cached_epoch
+        first_call = not getattr(self, "_url_prefetch_initialized", False)
+
+        if epoch_changed:
+            self._index_map = self._generate_index_map(current_epoch)
+            self._cached_epoch = current_epoch
+
+        if epoch_changed or first_call:
+            self._url_prefetch_initialized = True
+            _prefetch_epoch_urls(self._index_map, self._all_lines, self.delimiter)
+
+    def get_image_info_list(self, file_list, ratio_list):
+        if isinstance(file_list, str):
+            file_list = [file_list]
+        data_lines = []
+        for idx, file in enumerate(file_list):
+            with open(file, "rb") as f:
+                lines = f.readlines()
+                if self.mode == "train" or ratio_list[idx] < 1.0:
+                    random.seed(self.seed)
+                    lines = random.sample(lines, round(len(lines) * ratio_list[idx]))
+                data_lines.extend(lines)
+        return data_lines
+
+    def shuffle_data_random(self):
+        random.seed(self.seed)
+        random.shuffle(self.data_lines)
+        return
+
+    # ------------------------------------------------------------------ #
+    #  Epoch update (called from main process each epoch)
+    # ------------------------------------------------------------------ #
+
+    def reset_data_lines(self, seed=None, epoch=None):
+        """Signal new epoch to persistent workers via shared memory.
+
+        Workers lazily rebuild their _index_map on next __getitem__ call.
+        No disk I/O, no dataloader reconstruction.
+        """
+        self.seed = seed
+        epoch_val = epoch if epoch is not None else (seed if seed is not None else 0)
+        self._shared_epoch.value = int(epoch_val)
+
+        if self._all_lines is not None:
+            # Update main-process index_map (used by len() and batch_sampler)
+            self._index_map = self._generate_index_map(seed)
+            self._cached_epoch = int(epoch_val)
+            self.data_idx_order_list = list(range(len(self._index_map)))
+        else:
+            # Fallback for non-ratio cases
+            self.data_lines = self.get_image_info_list(
+                self.label_file_list, self.ratio_list
+            )
+            self.data_idx_order_list = list(range(len(self.data_lines)))
+            if self.mode == "train" and self.do_shuffle:
+                self.shuffle_data_random()
+
+    # ------------------------------------------------------------------ #
+    #  Data access
+    # ------------------------------------------------------------------ #
+
+    def _try_parse_filename_list(self, file_name):
+        # multiple images -> one gt label
+        if len(file_name) > 0 and file_name[0] == "[":
+            try:
+                info = json.loads(file_name)
+                file_name = random.choice(info)
+            except:
+                pass
+        return file_name
+
+    def get_ext_data(self):
+        ext_data_num = 0
+        for op in self.ops:
+            if hasattr(op, "ext_data_num"):
+                ext_data_num = getattr(op, "ext_data_num")
+                break
+        load_data_ops = self.ops[: self.ext_op_transform_idx]
+        ext_data = []
+
+        while len(ext_data) < ext_data_num:
+            if self._index_map is not None:
+                # Sample from current epoch's subset (same as original)
+                self._ensure_index_map()
+                rand_virtual = np.random.randint(len(self._index_map))
+                file_idx = self._index_map[rand_virtual]
+                data_line = self._all_lines[file_idx]
+            else:
+                file_idx = self.data_idx_order_list[np.random.randint(self.__len__())]
+                data_line = self.data_lines[file_idx]
+            data_line = data_line.decode("utf-8")
+            substr = data_line.strip("\n").split(self.delimiter)
+            file_name = substr[0]
+            file_name = self._try_parse_filename_list(file_name)
+            label = substr[1]
+            img_path = (
+                file_name
+                if file_name.startswith("http://") or file_name.startswith("https://")
+                else os.path.join(self.data_dir, file_name)
+            )
+            data = {"img_path": img_path, "label": label}
+            if not _img_path_exists(img_path):
+                continue
+            try:
+                data["image"] = _load_image_bytes(img_path)
+            except Exception:
+                continue
+            data = transform(data, load_data_ops)
+
+            if data is None:
+                continue
+            if "polys" in data.keys():
+                if data["polys"].shape[1] != 4:
+                    continue
+            ext_data.append(data)
+        return ext_data
+
+    def __getitem__(self, idx):
+        if self._index_map is not None:
+            self._ensure_index_map()
+            file_idx = self._index_map[idx]
+            data_line = self._all_lines[file_idx]
+        else:
+            file_idx = self.data_idx_order_list[idx]
+            data_line = self.data_lines[file_idx]
+        try:
+            data_line = data_line.decode("utf-8")
+            substr = data_line.strip("\n").split(self.delimiter)
+            file_name = substr[0]
+            file_name = self._try_parse_filename_list(file_name)
+            label = substr[1]
+            img_path = (
+                file_name
+                if file_name.startswith("http://") or file_name.startswith("https://")
+                else os.path.join(self.data_dir, file_name)
+            )
+            data = {"img_path": img_path, "label": label}
+            if not _img_path_exists(img_path):
+                raise Exception("{} does not exist!".format(img_path))
+            data["image"] = _load_image_bytes(img_path)
+            data["ext_data"] = self.get_ext_data()
+            data["filename"] = data["img_path"]
+            data["epoch"] = self._shared_epoch.value
+            outs = transform(data, self.ops)
+        except:
+            self.logger.error(
+                "When parsing line {}, error happened with msg: {}".format(
+                    data_line, traceback.format_exc()
+                )
+            )
+            outs = None
+        if outs is None:
+            # during evaluation, we should fix the idx to get same results for many times of evaluation.
+            rnd_idx = (
+                np.random.randint(self.__len__())
+                if self.mode == "train"
+                else (idx + 1) % self.__len__()
+            )
+            return self.__getitem__(rnd_idx)
+        return outs
+
+    def __len__(self):
+        if self._index_map is not None:
+            return len(self._index_map)
+        return len(self.data_idx_order_list)
+
+
+class MultiScaleDataSet(SimpleDataSet):
+    def __init__(self, config, mode, logger, seed=None):
+        super(MultiScaleDataSet, self).__init__(config, mode, logger, seed)
+        self.ds_width = config[mode]["dataset"].get("ds_width", False)
+        if self.ds_width:
+            self.wh_aware()
+
+    def wh_aware(self):
+        data_line_new = []
+        wh_ratio = []
+        for line in self.data_lines:
+            data_line_new.append(line)
+            line = line.decode("utf-8")
+            name, label, w, h = line.strip("\n").split(self.delimiter)
+            wh_ratio.append(float(w) / float(h))
+
+        self.data_lines = data_line_new
+        self.wh_ratio = np.array(wh_ratio)
+        self.wh_ratio_sort = np.argsort(self.wh_ratio)
+        self.data_idx_order_list = list(range(len(self.data_lines)))
+
+    def resize_norm_img(self, data, imgW, imgH, padding=True):
+        img = data["image"]
+        h = img.shape[0]
+        w = img.shape[1]
+        if not padding:
+            resized_image = cv2.resize(
+                img, (imgW, imgH), interpolation=cv2.INTER_LINEAR
+            )
+            resized_w = imgW
+        else:
+            ratio = w / float(h)
+            if math.ceil(imgH * ratio) > imgW:
+                resized_w = imgW
+            else:
+                resized_w = int(math.ceil(imgH * ratio))
+            resized_image = cv2.resize(img, (resized_w, imgH))
+        resized_image = resized_image.astype("float32")
+
+        resized_image = resized_image.transpose((2, 0, 1)) / 255
+        resized_image -= 0.5
+        resized_image /= 0.5
+        padding_im = np.zeros((3, imgH, imgW), dtype=np.float32)
+        padding_im[:, :, :resized_w] = resized_image
+        valid_ratio = min(1.0, float(resized_w / imgW))
+        data["image"] = padding_im
+        data["valid_ratio"] = valid_ratio
+        if "iluvatar_gpu" in get_device():
+            data["valid_ratio"] = np.float32(valid_ratio)
+        return data
+
+    def __getitem__(self, properties):
+        # properties is a tuple, contains (width, height, index)
+        img_height = properties[1]
+        idx = properties[2]
+        if self.ds_width and properties[3] is not None:
+            wh_ratio = properties[3]
+            img_width = img_height * (
+                1 if int(round(wh_ratio)) == 0 else int(round(wh_ratio))
+            )
+            file_idx = self.wh_ratio_sort[idx]
+        else:
+            file_idx = self.data_idx_order_list[idx]
+            img_width = properties[0]
+            wh_ratio = None
+
+        data_line = self.data_lines[file_idx]
+        try:
+            data_line = data_line.decode("utf-8")
+            substr = data_line.strip("\n").split(self.delimiter)
+            file_name = substr[0]
+            file_name = self._try_parse_filename_list(file_name)
+            label = substr[1]
+            img_path = (
+                file_name
+                if file_name.startswith("http://") or file_name.startswith("https://")
+                else os.path.join(self.data_dir, file_name)
+            )
+            data = {"img_path": img_path, "label": label}
+            if not _img_path_exists(img_path):
+                raise Exception("{} does not exist!".format(img_path))
+            data["image"] = _load_image_bytes(img_path)
+            data["ext_data"] = self.get_ext_data()
+            outs = transform(data, self.ops[:-1])
+            if outs is not None:
+                outs = self.resize_norm_img(outs, img_width, img_height)
+                outs = transform(outs, self.ops[-1:])
+        except:
+            self.logger.error(
+                "When parsing line {}, error happened with msg: {}".format(
+                    data_line, traceback.format_exc()
+                )
+            )
+            outs = None
+        if outs is None:
+            # during evaluation, we should fix the idx to get same results for many times of evaluation.
+            rnd_idx = (idx + 1) % self.__len__()
+            return self.__getitem__([img_width, img_height, rnd_idx, wh_ratio])
+        return outs
diff --git a/ppocr/ext_op/__init__.py b/ppocr/ext_op/__init__.py
new file mode 100644
index 0000000..b0be1ad
--- /dev/null
+++ b/ppocr/ext_op/__init__.py
@@ -0,0 +1 @@
+from .roi_align_rotated.roi_align_rotated import RoIAlignRotated
diff --git a/ppocr/ext_op/roi_align_rotated/roi_align_rotated.cc b/ppocr/ext_op/roi_align_rotated/roi_align_rotated.cc
new file mode 100644
index 0000000..f6b6217
--- /dev/null
+++ b/ppocr/ext_op/roi_align_rotated/roi_align_rotated.cc
@@ -0,0 +1,528 @@
+
+// This code is refer from:
+// https://github.com/open-mmlab/mmcv/blob/master/mmcv/ops/csrc/pytorch/cpu/roi_align_rotated.cpp
+
+#include <cassert>
+#include <cmath>
+#include <vector>
+
+#include "paddle/extension.h"
+
+#define PADDLE_WITH_CUDA
+#define CHECK_INPUT_SAME(x1, x2)                                               \
+  PD_CHECK(x1.place() == x2.place(), "input must be same place.")
+#define CHECK_INPUT_CPU(x) PD_CHECK(x.is_cpu(), #x " must be a CPU Tensor.")
+
+template <typename T> struct PreCalc {
+  int pos1;
+  int pos2;
+  int pos3;
+  int pos4;
+  T w1;
+  T w2;
+  T w3;
+  T w4;
+};
+
+template <typename T>
+void pre_calc_for_bilinear_interpolate(
+    const int height, const int width, const int pooled_height,
+    const int pooled_width, const int iy_upper, const int ix_upper,
+    T roi_start_h, T roi_start_w, T bin_size_h, T bin_size_w,
+    int roi_bin_grid_h, int roi_bin_grid_w, T roi_center_h, T roi_center_w,
+    T cos_theta, T sin_theta, std::vector<PreCalc<T>> &pre_calc) {
+  int pre_calc_index = 0;
+  for (int ph = 0; ph < pooled_height; ph++) {
+    for (int pw = 0; pw < pooled_width; pw++) {
+      for (int iy = 0; iy < iy_upper; iy++) {
+        const T yy = roi_start_h + ph * bin_size_h +
+                     static_cast<T>(iy + .5f) * bin_size_h /
+                         static_cast<T>(roi_bin_grid_h); // e.g., 0.5, 1.5
+        for (int ix = 0; ix < ix_upper; ix++) {
+          const T xx = roi_start_w + pw * bin_size_w +
+                       static_cast<T>(ix + .5f) * bin_size_w /
+                           static_cast<T>(roi_bin_grid_w);
+
+          // Rotate by theta around the center and translate
+          // In image space, (y, x) is the order for Right Handed System,
+          // and this is essentially multiplying the point by a rotation matrix
+          // to rotate it counterclockwise through angle theta.
+          T y = yy * cos_theta - xx * sin_theta + roi_center_h;
+          T x = yy * sin_theta + xx * cos_theta + roi_center_w;
+          // deal with: inverse elements are out of feature map boundary
+          if (y < -1.0 || y > height || x < -1.0 || x > width) {
+            // empty
+            PreCalc<T> pc;
+            pc.pos1 = 0;
+            pc.pos2 = 0;
+            pc.pos3 = 0;
+            pc.pos4 = 0;
+            pc.w1 = 0;
+            pc.w2 = 0;
+            pc.w3 = 0;
+            pc.w4 = 0;
+            pre_calc[pre_calc_index] = pc;
+            pre_calc_index += 1;
+            continue;
+          }
+
+          if (y < 0) {
+            y = 0;
+          }
+          if (x < 0) {
+            x = 0;
+          }
+
+          int y_low = (int)y;
+          int x_low = (int)x;
+          int y_high;
+          int x_high;
+
+          if (y_low >= height - 1) {
+            y_high = y_low = height - 1;
+            y = (T)y_low;
+          } else {
+            y_high = y_low + 1;
+          }
+
+          if (x_low >= width - 1) {
+            x_high = x_low = width - 1;
+            x = (T)x_low;
+          } else {
+            x_high = x_low + 1;
+          }
+
+          T ly = y - y_low;
+          T lx = x - x_low;
+          T hy = 1. - ly, hx = 1. - lx;
+          T w1 = hy * hx, w2 = hy * lx, w3 = ly * hx, w4 = ly * lx;
+
+          // save weights and indices
+          PreCalc<T> pc;
+          pc.pos1 = y_low * width + x_low;
+          pc.pos2 = y_low * width + x_high;
+          pc.pos3 = y_high * width + x_low;
+          pc.pos4 = y_high * width + x_high;
+          pc.w1 = w1;
+          pc.w2 = w2;
+          pc.w3 = w3;
+          pc.w4 = w4;
+          pre_calc[pre_calc_index] = pc;
+
+          pre_calc_index += 1;
+        }
+      }
+    }
+  }
+}
+
+template <typename T>
+void roi_align_rotated_cpu_forward(const int nthreads, const T *input,
+                                   const T &spatial_scale, const bool aligned,
+                                   const bool clockwise, const int channels,
+                                   const int height, const int width,
+                                   const int pooled_height,
+                                   const int pooled_width,
+                                   const int sampling_ratio, const T *rois,
+                                   T *output) {
+  int n_rois = nthreads / channels / pooled_width / pooled_height;
+  // (n, c, ph, pw) is an element in the pooled output
+  // can be parallelized using omp
+  // #pragma omp parallel for num_threads(32)
+  for (int n = 0; n < n_rois; n++) {
+    int index_n = n * channels * pooled_width * pooled_height;
+
+    const T *current_roi = rois + n * 6;
+    int roi_batch_ind = current_roi[0];
+
+    // Do not use rounding; this implementation detail is critical
+    T offset = aligned ? (T)0.5 : (T)0.0;
+    T roi_center_w = current_roi[1] * spatial_scale - offset;
+    T roi_center_h = current_roi[2] * spatial_scale - offset;
+    T roi_width = current_roi[3] * spatial_scale;
+    T roi_height = current_roi[4] * spatial_scale;
+    T theta = current_roi[5];
+    if (clockwise) {
+      theta = -theta; // If clockwise, the angle needs to be reversed.
+    }
+    T cos_theta = cos(theta);
+    T sin_theta = sin(theta);
+
+    if (aligned) {
+      assert(roi_width >= 0 && roi_height >= 0);
+    } else { // for backward-compatibility only
+      roi_width = std::max(roi_width, (T)1.);
+      roi_height = std::max(roi_height, (T)1.);
+    }
+
+    T bin_size_h = static_cast<T>(roi_height) / static_cast<T>(pooled_height);
+    T bin_size_w = static_cast<T>(roi_width) / static_cast<T>(pooled_width);
+
+    // We use roi_bin_grid to sample the grid and mimic integral
+    int roi_bin_grid_h = (sampling_ratio > 0)
+                             ? sampling_ratio
+                             : ceilf(roi_height / pooled_height); // e.g., = 2
+    int roi_bin_grid_w =
+        (sampling_ratio > 0) ? sampling_ratio : ceilf(roi_width / pooled_width);
+
+    // We do average (integral) pooling inside a bin
+    const T count = std::max(roi_bin_grid_h * roi_bin_grid_w, 1); // e.g. = 4
+
+    // we want to precalculate indices and weights shared by all channels,
+    // this is the key point of optimization
+    std::vector<PreCalc<T>> pre_calc(roi_bin_grid_h * roi_bin_grid_w *
+                                     pooled_width * pooled_height);
+
+    // roi_start_h and roi_start_w are computed wrt the center of RoI (x, y).
+    // Appropriate translation needs to be applied after.
+    T roi_start_h = -roi_height / 2.0;
+    T roi_start_w = -roi_width / 2.0;
+
+    pre_calc_for_bilinear_interpolate(
+        height, width, pooled_height, pooled_width, roi_bin_grid_h,
+        roi_bin_grid_w, roi_start_h, roi_start_w, bin_size_h, bin_size_w,
+        roi_bin_grid_h, roi_bin_grid_w, roi_center_h, roi_center_w, cos_theta,
+        sin_theta, pre_calc);
+
+    for (int c = 0; c < channels; c++) {
+      int index_n_c = index_n + c * pooled_width * pooled_height;
+      const T *offset_input =
+          input + (roi_batch_ind * channels + c) * height * width;
+      int pre_calc_index = 0;
+
+      for (int ph = 0; ph < pooled_height; ph++) {
+        for (int pw = 0; pw < pooled_width; pw++) {
+          int index = index_n_c + ph * pooled_width + pw;
+
+          T output_val = 0.;
+          for (int iy = 0; iy < roi_bin_grid_h; iy++) {
+            for (int ix = 0; ix < roi_bin_grid_w; ix++) {
+              PreCalc<T> pc = pre_calc[pre_calc_index];
+              output_val += pc.w1 * offset_input[pc.pos1] +
+                            pc.w2 * offset_input[pc.pos2] +
+                            pc.w3 * offset_input[pc.pos3] +
+                            pc.w4 * offset_input[pc.pos4];
+
+              pre_calc_index += 1;
+            }
+          }
+          output_val /= count;
+
+          output[index] = output_val;
+        } // for pw
+      }   // for ph
+    }     // for c
+  }       // for n
+}
+
+template <typename T>
+void bilinear_interpolate_gradient(const int height, const int width, T y, T x,
+                                   T &w1, T &w2, T &w3, T &w4, int &x_low,
+                                   int &x_high, int &y_low, int &y_high) {
+  // deal with cases that inverse elements are out of feature map boundary
+  if (y < -1.0 || y > height || x < -1.0 || x > width) {
+    // empty
+    w1 = w2 = w3 = w4 = 0.;
+    x_low = x_high = y_low = y_high = -1;
+    return;
+  }
+
+  if (y < 0) {
+    y = 0;
+  }
+
+  if (x < 0) {
+    x = 0;
+  }
+
+  y_low = (int)y;
+  x_low = (int)x;
+
+  if (y_low >= height - 1) {
+    y_high = y_low = height - 1;
+    y = (T)y_low;
+  } else {
+    y_high = y_low + 1;
+  }
+
+  if (x_low >= width - 1) {
+    x_high = x_low = width - 1;
+    x = (T)x_low;
+  } else {
+    x_high = x_low + 1;
+  }
+
+  T ly = y - y_low;
+  T lx = x - x_low;
+  T hy = 1. - ly, hx = 1. - lx;
+
+  // reference in forward
+  // T v1 = input[y_low * width + x_low];
+  // T v2 = input[y_low * width + x_high];
+  // T v3 = input[y_high * width + x_low];
+  // T v4 = input[y_high * width + x_high];
+  // T val = (w1 * v1 + w2 * v2 + w3 * v3 + w4 * v4);
+
+  w1 = hy * hx, w2 = hy * lx, w3 = ly * hx, w4 = ly * lx;
+
+  return;
+}
+
+template <class T> inline void add(T *address, const T &val) {
+  *address += val;
+}
+
+template <typename T>
+void roi_align_rotated_cpu_backward(
+    const int nthreads,
+    // may not be contiguous. should index using n_stride, etc
+    const T *grad_output, const T &spatial_scale, const bool aligned,
+    const bool clockwise, const int channels, const int height, const int width,
+    const int pooled_height, const int pooled_width, const int sampling_ratio,
+    T *grad_input, const T *rois, const int n_stride, const int c_stride,
+    const int h_stride, const int w_stride) {
+  for (int index = 0; index < nthreads; index++) {
+    // (n, c, ph, pw) is an element in the pooled output
+    int pw = index % pooled_width;
+    int ph = (index / pooled_width) % pooled_height;
+    int c = (index / pooled_width / pooled_height) % channels;
+    int n = index / pooled_width / pooled_height / channels;
+
+    const T *current_roi = rois + n * 6;
+    int roi_batch_ind = current_roi[0];
+
+    // Do not use rounding; this implementation detail is critical
+    T offset = aligned ? (T)0.5 : (T)0.0;
+    T roi_center_w = current_roi[1] * spatial_scale - offset;
+    T roi_center_h = current_roi[2] * spatial_scale - offset;
+    T roi_width = current_roi[3] * spatial_scale;
+    T roi_height = current_roi[4] * spatial_scale;
+    T theta = current_roi[5];
+    if (clockwise) {
+      theta = -theta; // If clockwise, the angle needs to be reversed.
+    }
+    T cos_theta = cos(theta);
+    T sin_theta = sin(theta);
+
+    if (aligned) {
+      assert(roi_width >= 0 && roi_height >= 0);
+    } else { // for backward-compatibility only
+      roi_width = std::max(roi_width, (T)1.);
+      roi_height = std::max(roi_height, (T)1.);
+    }
+
+    T bin_size_h = static_cast<T>(roi_height) / static_cast<T>(pooled_height);
+    T bin_size_w = static_cast<T>(roi_width) / static_cast<T>(pooled_width);
+
+    T *offset_grad_input =
+        grad_input + ((roi_batch_ind * channels + c) * height * width);
+
+    int output_offset = n * n_stride + c * c_stride;
+    const T *offset_grad_output = grad_output + output_offset;
+    const T grad_output_this_bin =
+        offset_grad_output[ph * h_stride + pw * w_stride];
+
+    // We use roi_bin_grid to sample the grid and mimic integral
+    int roi_bin_grid_h = (sampling_ratio > 0)
+                             ? sampling_ratio
+                             : ceilf(roi_height / pooled_height); // e.g., = 2
+    int roi_bin_grid_w =
+        (sampling_ratio > 0) ? sampling_ratio : ceilf(roi_width / pooled_width);
+
+    // roi_start_h and roi_start_w are computed wrt the center of RoI (x, y).
+    // Appropriate translation needs to be applied after.
+    T roi_start_h = -roi_height / 2.0;
+    T roi_start_w = -roi_width / 2.0;
+
+    // We do average (integral) pooling inside a bin
+    const T count = roi_bin_grid_h * roi_bin_grid_w; // e.g. = 4
+
+    for (int iy = 0; iy < roi_bin_grid_h; iy++) {
+      const T yy = roi_start_h + ph * bin_size_h +
+                   static_cast<T>(iy + .5f) * bin_size_h /
+                       static_cast<T>(roi_bin_grid_h); // e.g., 0.5, 1.5
+      for (int ix = 0; ix < roi_bin_grid_w; ix++) {
+        const T xx = roi_start_w + pw * bin_size_w +
+                     static_cast<T>(ix + .5f) * bin_size_w /
+                         static_cast<T>(roi_bin_grid_w);
+
+        // Rotate by theta around the center and translate
+        T y = yy * cos_theta - xx * sin_theta + roi_center_h;
+        T x = yy * sin_theta + xx * cos_theta + roi_center_w;
+
+        T w1, w2, w3, w4;
+        int x_low, x_high, y_low, y_high;
+
+        bilinear_interpolate_gradient(height, width, y, x, w1, w2, w3, w4,
+                                      x_low, x_high, y_low, y_high);
+
+        T g1 = grad_output_this_bin * w1 / count;
+        T g2 = grad_output_this_bin * w2 / count;
+        T g3 = grad_output_this_bin * w3 / count;
+        T g4 = grad_output_this_bin * w4 / count;
+
+        if (x_low >= 0 && x_high >= 0 && y_low >= 0 && y_high >= 0) {
+          // atomic add is not needed for now since it is single threaded
+          add(offset_grad_input + y_low * width + x_low, static_cast<T>(g1));
+          add(offset_grad_input + y_low * width + x_high, static_cast<T>(g2));
+          add(offset_grad_input + y_high * width + x_low, static_cast<T>(g3));
+          add(offset_grad_input + y_high * width + x_high, static_cast<T>(g4));
+        } // if
+      }   // ix
+    }     // iy
+  }       // for
+} // ROIAlignRotatedBackward
+
+std::vector<paddle::Tensor>
+RoIAlignRotatedCPUForward(const paddle::Tensor &input,
+                          const paddle::Tensor &rois, int aligned_height,
+                          int aligned_width, float spatial_scale,
+                          int sampling_ratio, bool aligned, bool clockwise) {
+  CHECK_INPUT_CPU(input);
+  CHECK_INPUT_CPU(rois);
+
+  auto num_rois = rois.shape()[0];
+
+  auto channels = input.shape()[1];
+  auto height = input.shape()[2];
+  auto width = input.shape()[3];
+
+  auto output =
+      paddle::empty({num_rois, channels, aligned_height, aligned_width},
+                    input.type(), paddle::CPUPlace());
+  auto output_size = output.numel();
+
+  PD_DISPATCH_FLOATING_TYPES(
+      input.type(), "roi_align_rotated_cpu_forward", ([&] {
+        roi_align_rotated_cpu_forward<data_t>(
+            output_size, input.data<data_t>(),
+            static_cast<data_t>(spatial_scale), aligned, clockwise, channels,
+            height, width, aligned_height, aligned_width, sampling_ratio,
+            rois.data<data_t>(), output.data<data_t>());
+      }));
+
+  return {output};
+}
+
+std::vector<paddle::Tensor> RoIAlignRotatedCPUBackward(
+    const paddle::Tensor &input, const paddle::Tensor &rois,
+    const paddle::Tensor &grad_output, int aligned_height, int aligned_width,
+    float spatial_scale, int sampling_ratio, bool aligned, bool clockwise) {
+
+  auto batch_size = input.shape()[0];
+  auto channels = input.shape()[1];
+  auto height = input.shape()[2];
+  auto width = input.shape()[3];
+
+  auto grad_input = paddle::full({batch_size, channels, height, width}, 0.0,
+                                 input.type(), paddle::CPUPlace());
+
+  // get stride values to ensure indexing into gradients is correct.
+  int n_stride = grad_output.shape()[0];
+  int c_stride = grad_output.shape()[1];
+  int h_stride = grad_output.shape()[2];
+  int w_stride = grad_output.shape()[3];
+
+  PD_DISPATCH_FLOATING_TYPES(
+      grad_output.type(), "roi_align_rotated_cpu_backward", [&] {
+        roi_align_rotated_cpu_backward<data_t>(
+            grad_output.numel(), grad_output.data<data_t>(),
+            static_cast<data_t>(spatial_scale), aligned, clockwise, channels,
+            height, width, aligned_height, aligned_width, sampling_ratio,
+            grad_input.data<data_t>(), rois.data<data_t>(), n_stride, c_stride,
+            h_stride, w_stride);
+      });
+  return {grad_input};
+}
+
+#ifdef PADDLE_WITH_CUDA
+std::vector<paddle::Tensor>
+RoIAlignRotatedCUDAForward(const paddle::Tensor &input,
+                           const paddle::Tensor &rois, int aligned_height,
+                           int aligned_width, float spatial_scale,
+                           int sampling_ratio, bool aligned, bool clockwise);
+#endif
+
+#ifdef PADDLE_WITH_CUDA
+std::vector<paddle::Tensor> RoIAlignRotatedCUDABackward(
+    const paddle::Tensor &input, const paddle::Tensor &rois,
+    const paddle::Tensor &grad_output, int aligned_height, int aligned_width,
+    float spatial_scale, int sampling_ratio, bool aligned, bool clockwise);
+#endif
+
+std::vector<paddle::Tensor>
+RoIAlignRotatedForward(const paddle::Tensor &input, const paddle::Tensor &rois,
+                       int aligned_height, int aligned_width,
+                       float spatial_scale, int sampling_ratio, bool aligned,
+                       bool clockwise) {
+  CHECK_INPUT_SAME(input, rois);
+  if (input.is_cpu()) {
+    return RoIAlignRotatedCPUForward(input, rois, aligned_height, aligned_width,
+                                     spatial_scale, sampling_ratio, aligned,
+                                     clockwise);
+#ifdef PADDLE_WITH_CUDA
+  } else if (input.is_gpu()) {
+    return RoIAlignRotatedCUDAForward(input, rois, aligned_height,
+                                      aligned_width, spatial_scale,
+                                      sampling_ratio, aligned, clockwise);
+#endif
+  } else {
+    PD_THROW("Unsupported device type for forward function of roi align "
+             "rotated operator.");
+  }
+}
+
+std::vector<paddle::Tensor>
+RoIAlignRotatedBackward(const paddle::Tensor &input, const paddle::Tensor &rois,
+                        const paddle::Tensor &grad_output, int aligned_height,
+                        int aligned_width, float spatial_scale,
+                        int sampling_ratio, bool aligned, bool clockwise) {
+  CHECK_INPUT_SAME(input, rois);
+  if (input.is_cpu()) {
+    return RoIAlignRotatedCPUBackward(input, rois, grad_output, aligned_height,
+                                      aligned_width, spatial_scale,
+                                      sampling_ratio, aligned, clockwise);
+#ifdef PADDLE_WITH_CUDA
+  } else if (input.is_gpu()) {
+    return RoIAlignRotatedCUDABackward(input, rois, grad_output, aligned_height,
+                                       aligned_width, spatial_scale,
+                                       sampling_ratio, aligned, clockwise);
+#endif
+  } else {
+    PD_THROW("Unsupported device type for forward function of roi align "
+             "rotated operator.");
+  }
+}
+
+std::vector<std::vector<int64_t>> InferShape(std::vector<int64_t> input_shape,
+                                             std::vector<int64_t> rois_shape) {
+  return {{rois_shape[0], input_shape[1], input_shape[2], input_shape[3]}};
+}
+
+std::vector<std::vector<int64_t>>
+InferBackShape(std::vector<int64_t> input_shape,
+               std::vector<int64_t> rois_shape) {
+  return {input_shape};
+}
+
+std::vector<paddle::DataType> InferDtype(paddle::DataType input_dtype,
+                                         paddle::DataType rois_dtype) {
+  return {input_dtype};
+}
+
+PD_BUILD_OP(roi_align_rotated)
+    .Inputs({"Input", "Rois"})
+    .Outputs({"Output"})
+    .Attrs({"aligned_height: int", "aligned_width: int", "spatial_scale: float",
+            "sampling_ratio: int", "aligned: bool", "clockwise: bool"})
+    .SetKernelFn(PD_KERNEL(RoIAlignRotatedForward))
+    .SetInferShapeFn(PD_INFER_SHAPE(InferShape))
+    .SetInferDtypeFn(PD_INFER_DTYPE(InferDtype));
+
+PD_BUILD_GRAD_OP(roi_align_rotated)
+    .Inputs({"Input", "Rois", paddle::Grad("Output")})
+    .Attrs({"aligned_height: int", "aligned_width: int", "spatial_scale: float",
+            "sampling_ratio: int", "aligned: bool", "clockwise: bool"})
+    .Outputs({paddle::Grad("Input")})
+    .SetKernelFn(PD_KERNEL(RoIAlignRotatedBackward))
+    .SetInferShapeFn(PD_INFER_SHAPE(InferBackShape));
diff --git a/ppocr/ext_op/roi_align_rotated/roi_align_rotated.cu b/ppocr/ext_op/roi_align_rotated/roi_align_rotated.cu
new file mode 100644
index 0000000..d9baefb
--- /dev/null
+++ b/ppocr/ext_op/roi_align_rotated/roi_align_rotated.cu
@@ -0,0 +1,381 @@
+
+// This code is refer from:
+// https://github.com/open-mmlab/mmcv/blob/master/mmcv/ops/csrc/common/cuda/roi_align_rotated_cuda_kernel.cuh
+
+#include <cassert>
+#include <cmath>
+#include <vector>
+
+#include "paddle/extension.h"
+#include <cuda.h>
+
+#define CUDA_1D_KERNEL_LOOP(i, n)                                              \
+  for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < (n);                 \
+       i += blockDim.x * gridDim.x)
+
+#define THREADS_PER_BLOCK 512
+
+inline int GET_BLOCKS(const int N) {
+  int optimal_block_num = (N + THREADS_PER_BLOCK - 1) / THREADS_PER_BLOCK;
+  int max_block_num = 4096;
+  return min(optimal_block_num, max_block_num);
+}
+
+#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 600
+
+static __inline__ __device__ double atomicAdd(double *address, double val) {
+  unsigned long long int *address_as_ull = (unsigned long long int *)address;
+  unsigned long long int old = *address_as_ull, assumed;
+  if (val == 0.0)
+    return __longlong_as_double(old);
+  do {
+    assumed = old;
+    old = atomicCAS(address_as_ull, assumed,
+                    __double_as_longlong(val + __longlong_as_double(assumed)));
+  } while (assumed != old);
+  return __longlong_as_double(old);
+}
+
+#endif
+
+template <typename T>
+__device__ T bilinear_interpolate(const T *input, const int height,
+                                  const int width, T y, T x,
+                                  const int index /* index for debug only*/) {
+  // deal with cases that inverse elements are out of feature map boundary
+  if (y < -1.0 || y > height || x < -1.0 || x > width)
+    return 0;
+
+  if (y <= 0)
+    y = 0;
+  if (x <= 0)
+    x = 0;
+
+  int y_low = (int)y;
+  int x_low = (int)x;
+  int y_high;
+  int x_high;
+
+  if (y_low >= height - 1) {
+    y_high = y_low = height - 1;
+    y = (T)y_low;
+  } else {
+    y_high = y_low + 1;
+  }
+
+  if (x_low >= width - 1) {
+    x_high = x_low = width - 1;
+    x = (T)x_low;
+  } else {
+    x_high = x_low + 1;
+  }
+
+  T ly = y - y_low;
+  T lx = x - x_low;
+  T hy = 1. - ly, hx = 1. - lx;
+  // do bilinear interpolation
+  T v1 = input[y_low * width + x_low];
+  T v2 = input[y_low * width + x_high];
+  T v3 = input[y_high * width + x_low];
+  T v4 = input[y_high * width + x_high];
+  T w1 = hy * hx, w2 = hy * lx, w3 = ly * hx, w4 = ly * lx;
+
+  T val = (w1 * v1 + w2 * v2 + w3 * v3 + w4 * v4);
+
+  return val;
+}
+
+template <typename T>
+__device__ void
+bilinear_interpolate_gradient(const int height, const int width, T y, T x,
+                              T &w1, T &w2, T &w3, T &w4, int &x_low,
+                              int &x_high, int &y_low, int &y_high,
+                              const int index /* index for debug only*/) {
+  // deal with cases that inverse elements are out of feature map boundary
+  if (y < -1.0 || y > height || x < -1.0 || x > width) {
+    // empty
+    w1 = w2 = w3 = w4 = 0.;
+    x_low = x_high = y_low = y_high = -1;
+    return;
+  }
+
+  if (y <= 0)
+    y = 0;
+  if (x <= 0)
+    x = 0;
+
+  y_low = (int)y;
+  x_low = (int)x;
+
+  if (y_low >= height - 1) {
+    y_high = y_low = height - 1;
+    y = (T)y_low;
+  } else {
+    y_high = y_low + 1;
+  }
+
+  if (x_low >= width - 1) {
+    x_high = x_low = width - 1;
+    x = (T)x_low;
+  } else {
+    x_high = x_low + 1;
+  }
+
+  T ly = y - y_low;
+  T lx = x - x_low;
+  T hy = 1. - ly, hx = 1. - lx;
+
+  // reference in forward
+  // T v1 = input[y_low * width + x_low];
+  // T v2 = input[y_low * width + x_high];
+  // T v3 = input[y_high * width + x_low];
+  // T v4 = input[y_high * width + x_high];
+  // T val = (w1 * v1 + w2 * v2 + w3 * v3 + w4 * v4);
+
+  w1 = hy * hx, w2 = hy * lx, w3 = ly * hx, w4 = ly * lx;
+
+  return;
+}
+
+/*** Forward ***/
+template <typename scalar_t>
+__global__ void roi_align_rotated_cuda_forward_kernel(
+    const int nthreads, const scalar_t *bottom_data,
+    const scalar_t *bottom_rois, const scalar_t spatial_scale,
+    const int sample_num, const bool aligned, const bool clockwise,
+    const int channels, const int height, const int width,
+    const int pooled_height, const int pooled_width, scalar_t *top_data) {
+  CUDA_1D_KERNEL_LOOP(index, nthreads) {
+    // (n, c, ph, pw) is an element in the pooled output
+    int pw = index % pooled_width;
+    int ph = (index / pooled_width) % pooled_height;
+    int c = (index / pooled_width / pooled_height) % channels;
+    int n = index / pooled_width / pooled_height / channels;
+
+    const scalar_t *offset_bottom_rois = bottom_rois + n * 6;
+    int roi_batch_ind = offset_bottom_rois[0];
+
+    // Do not using rounding; this implementation detail is critical
+    scalar_t offset = aligned ? (scalar_t)0.5 : (scalar_t)0.0;
+    scalar_t roi_center_w = offset_bottom_rois[1] * spatial_scale - offset;
+    scalar_t roi_center_h = offset_bottom_rois[2] * spatial_scale - offset;
+    scalar_t roi_width = offset_bottom_rois[3] * spatial_scale;
+    scalar_t roi_height = offset_bottom_rois[4] * spatial_scale;
+    // scalar_t theta = offset_bottom_rois[5] * M_PI / 180.0;
+    scalar_t theta = offset_bottom_rois[5];
+    if (clockwise) {
+      theta = -theta; // If clockwise, the angle needs to be reversed.
+    }
+    if (!aligned) { // for backward-compatibility only
+      // Force malformed ROIs to be 1x1
+      roi_width = max(roi_width, (scalar_t)1.);
+      roi_height = max(roi_height, (scalar_t)1.);
+    }
+    scalar_t bin_size_h = static_cast<scalar_t>(roi_height) /
+                          static_cast<scalar_t>(pooled_height);
+    scalar_t bin_size_w =
+        static_cast<scalar_t>(roi_width) / static_cast<scalar_t>(pooled_width);
+
+    const scalar_t *offset_bottom_data =
+        bottom_data + (roi_batch_ind * channels + c) * height * width;
+
+    // We use roi_bin_grid to sample the grid and mimic integral
+    int roi_bin_grid_h = (sample_num > 0)
+                             ? sample_num
+                             : ceilf(roi_height / pooled_height); // e.g., = 2
+    int roi_bin_grid_w =
+        (sample_num > 0) ? sample_num : ceilf(roi_width / pooled_width);
+
+    // roi_start_h and roi_start_w are computed wrt the center of RoI (x, y).
+    // Appropriate translation needs to be applied after.
+    scalar_t roi_start_h = -roi_height / 2.0;
+    scalar_t roi_start_w = -roi_width / 2.0;
+    scalar_t cosscalar_theta = cos(theta);
+    scalar_t sinscalar_theta = sin(theta);
+
+    // We do average (integral) pooling inside a bin
+    const scalar_t count = max(roi_bin_grid_h * roi_bin_grid_w, 1); // e.g. = 4
+
+    scalar_t output_val = 0.;
+    for (int iy = 0; iy < roi_bin_grid_h; iy++) { // e.g., iy = 0, 1
+      const scalar_t yy =
+          roi_start_h + ph * bin_size_h +
+          static_cast<scalar_t>(iy + .5f) * bin_size_h /
+              static_cast<scalar_t>(roi_bin_grid_h); // e.g., 0.5, 1.5
+      for (int ix = 0; ix < roi_bin_grid_w; ix++) {
+        const scalar_t xx = roi_start_w + pw * bin_size_w +
+                            static_cast<scalar_t>(ix + .5f) * bin_size_w /
+                                static_cast<scalar_t>(roi_bin_grid_w);
+
+        // Rotate by theta (counterclockwise) around the center and translate
+        scalar_t y = yy * cosscalar_theta - xx * sinscalar_theta + roi_center_h;
+        scalar_t x = yy * sinscalar_theta + xx * cosscalar_theta + roi_center_w;
+
+        scalar_t val = bilinear_interpolate<scalar_t>(
+            offset_bottom_data, height, width, y, x, index);
+        output_val += val;
+      }
+    }
+    output_val /= count;
+
+    top_data[index] = output_val;
+  }
+}
+
+/*** Backward ***/
+template <typename scalar_t>
+__global__ void roi_align_rotated_backward_cuda_kernel(
+    const int nthreads, const scalar_t *top_diff, const scalar_t *bottom_rois,
+    const scalar_t spatial_scale, const int sample_num, const bool aligned,
+    const bool clockwise, const int channels, const int height, const int width,
+    const int pooled_height, const int pooled_width, scalar_t *bottom_diff) {
+  CUDA_1D_KERNEL_LOOP(index, nthreads) {
+    // (n, c, ph, pw) is an element in the pooled output
+    int pw = index % pooled_width;
+    int ph = (index / pooled_width) % pooled_height;
+    int c = (index / pooled_width / pooled_height) % channels;
+    int n = index / pooled_width / pooled_height / channels;
+
+    const scalar_t *offset_bottom_rois = bottom_rois + n * 6;
+    int roi_batch_ind = offset_bottom_rois[0];
+
+    // Do not round
+    scalar_t offset = aligned ? (scalar_t)0.5 : (scalar_t)0.0;
+    scalar_t roi_center_w = offset_bottom_rois[1] * spatial_scale - offset;
+    scalar_t roi_center_h = offset_bottom_rois[2] * spatial_scale - offset;
+    scalar_t roi_width = offset_bottom_rois[3] * spatial_scale;
+    scalar_t roi_height = offset_bottom_rois[4] * spatial_scale;
+    // scalar_t theta = offset_bottom_rois[5] * M_PI / 180.0;
+    scalar_t theta = offset_bottom_rois[5];
+    if (clockwise) {
+      theta = -theta; // If clockwise, the angle needs to be reversed.
+    }
+    if (!aligned) { // for backward-compatibility only
+      // Force malformed ROIs to be 1x1
+      roi_width = max(roi_width, (scalar_t)1.);
+      roi_height = max(roi_height, (scalar_t)1.);
+    }
+    scalar_t bin_size_h = static_cast<scalar_t>(roi_height) /
+                          static_cast<scalar_t>(pooled_height);
+    scalar_t bin_size_w =
+        static_cast<scalar_t>(roi_width) / static_cast<scalar_t>(pooled_width);
+
+    scalar_t *offset_bottom_diff =
+        bottom_diff + (roi_batch_ind * channels + c) * height * width;
+
+    int top_offset = (n * channels + c) * pooled_height * pooled_width;
+    const scalar_t *offset_top_diff = top_diff + top_offset;
+    const scalar_t top_diff_this_bin = offset_top_diff[ph * pooled_width + pw];
+
+    // We use roi_bin_grid to sample the grid and mimic integral
+    int roi_bin_grid_h = (sample_num > 0)
+                             ? sample_num
+                             : ceilf(roi_height / pooled_height); // e.g., = 2
+    int roi_bin_grid_w =
+        (sample_num > 0) ? sample_num : ceilf(roi_width / pooled_width);
+
+    // roi_start_h and roi_start_w are computed wrt the center of RoI (x, y).
+    // Appropriate translation needs to be applied after.
+    scalar_t roi_start_h = -roi_height / 2.0;
+    scalar_t roi_start_w = -roi_width / 2.0;
+    scalar_t cosTheta = cos(theta);
+    scalar_t sinTheta = sin(theta);
+
+    // We do average (integral) pooling inside a bin
+    const scalar_t count = roi_bin_grid_h * roi_bin_grid_w; // e.g. = 4
+
+    for (int iy = 0; iy < roi_bin_grid_h; iy++) { // e.g., iy = 0, 1
+      const scalar_t yy =
+          roi_start_h + ph * bin_size_h +
+          static_cast<scalar_t>(iy + .5f) * bin_size_h /
+              static_cast<scalar_t>(roi_bin_grid_h); // e.g., 0.5, 1.5
+      for (int ix = 0; ix < roi_bin_grid_w; ix++) {
+        const scalar_t xx = roi_start_w + pw * bin_size_w +
+                            static_cast<scalar_t>(ix + .5f) * bin_size_w /
+                                static_cast<scalar_t>(roi_bin_grid_w);
+
+        // Rotate by theta around the center and translate
+        scalar_t y = yy * cosTheta - xx * sinTheta + roi_center_h;
+        scalar_t x = yy * sinTheta + xx * cosTheta + roi_center_w;
+
+        scalar_t w1, w2, w3, w4;
+        int x_low, x_high, y_low, y_high;
+
+        bilinear_interpolate_gradient<scalar_t>(height, width, y, x, w1, w2, w3,
+                                                w4, x_low, x_high, y_low,
+                                                y_high, index);
+
+        scalar_t g1 = top_diff_this_bin * w1 / count;
+        scalar_t g2 = top_diff_this_bin * w2 / count;
+        scalar_t g3 = top_diff_this_bin * w3 / count;
+        scalar_t g4 = top_diff_this_bin * w4 / count;
+
+        if (x_low >= 0 && x_high >= 0 && y_low >= 0 && y_high >= 0) {
+          atomicAdd(offset_bottom_diff + y_low * width + x_low, g1);
+          atomicAdd(offset_bottom_diff + y_low * width + x_high, g2);
+          atomicAdd(offset_bottom_diff + y_high * width + x_low, g3);
+          atomicAdd(offset_bottom_diff + y_high * width + x_high, g4);
+        } // if
+      }   // ix
+    }     // iy
+  }       // CUDA_1D_KERNEL_LOOP
+} // RoIAlignBackward
+
+std::vector<paddle::Tensor>
+RoIAlignRotatedCUDAForward(const paddle::Tensor &input,
+                           const paddle::Tensor &rois, int aligned_height,
+                           int aligned_width, float spatial_scale,
+                           int sampling_ratio, bool aligned, bool clockwise) {
+
+  auto num_rois = rois.shape()[0];
+
+  auto channels = input.shape()[1];
+  auto height = input.shape()[2];
+  auto width = input.shape()[3];
+
+  auto output =
+      paddle::empty({num_rois, channels, aligned_height, aligned_width},
+                    input.type(), paddle::GPUPlace());
+  auto output_size = output.numel();
+
+  PD_DISPATCH_FLOATING_TYPES(
+      input.type(), "roi_align_rotated_cuda_forward_kernel", ([&] {
+        roi_align_rotated_cuda_forward_kernel<data_t>
+            <<<GET_BLOCKS(output_size), THREADS_PER_BLOCK>>>(
+                output_size, input.data<data_t>(), rois.data<data_t>(),
+                static_cast<data_t>(spatial_scale), sampling_ratio, aligned,
+                clockwise, channels, height, width, aligned_height,
+                aligned_width, output.data<data_t>());
+      }));
+
+  return {output};
+}
+
+std::vector<paddle::Tensor> RoIAlignRotatedCUDABackward(
+    const paddle::Tensor &input, const paddle::Tensor &rois,
+    const paddle::Tensor &grad_output, int aligned_height, int aligned_width,
+    float spatial_scale, int sampling_ratio, bool aligned, bool clockwise) {
+
+  auto num_rois = rois.shape()[0];
+
+  auto batch_size = input.shape()[0];
+  auto channels = input.shape()[1];
+  auto height = input.shape()[2];
+  auto width = input.shape()[3];
+
+  auto grad_input = paddle::full({batch_size, channels, height, width}, 0.0,
+                                 input.type(), paddle::GPUPlace());
+
+  const int output_size = num_rois * aligned_height * aligned_width * channels;
+
+  PD_DISPATCH_FLOATING_TYPES(
+      grad_output.type(), "roi_align_rotated_backward_cuda_kernel", ([&] {
+        roi_align_rotated_backward_cuda_kernel<data_t>
+            <<<GET_BLOCKS(output_size), THREADS_PER_BLOCK>>>(
+                output_size, grad_output.data<data_t>(), rois.data<data_t>(),
+                spatial_scale, sampling_ratio, aligned, clockwise, channels,
+                height, width, aligned_height, aligned_width,
+                grad_input.data<data_t>());
+      }));
+  return {grad_input};
+}
diff --git a/ppocr/ext_op/roi_align_rotated/roi_align_rotated.py b/ppocr/ext_op/roi_align_rotated/roi_align_rotated.py
new file mode 100644
index 0000000..4cc1bf6
--- /dev/null
+++ b/ppocr/ext_op/roi_align_rotated/roi_align_rotated.py
@@ -0,0 +1,69 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmcv/blob/master/mmcv/ops/roi_align_rotated.py
+"""
+
+import paddle
+import paddle.nn as nn
+from paddle.utils.cpp_extension import load
+
+custom_ops = load(
+    name="custom_jit_ops",
+    sources=[
+        "ppocr/ext_op/roi_align_rotated/roi_align_rotated.cc",
+        "ppocr/ext_op/roi_align_rotated/roi_align_rotated.cu",
+    ],
+)
+
+roi_align_rotated = custom_ops.roi_align_rotated
+
+
+class RoIAlignRotated(nn.Layer):
+    """RoI align pooling layer for rotated proposals."""
+
+    def __init__(
+        self, out_size, spatial_scale, sample_num=0, aligned=True, clockwise=False
+    ):
+        super(RoIAlignRotated, self).__init__()
+
+        if isinstance(out_size, int):
+            self.out_h = out_size
+            self.out_w = out_size
+        elif isinstance(out_size, tuple):
+            assert len(out_size) == 2
+            assert isinstance(out_size[0], int)
+            assert isinstance(out_size[1], int)
+            self.out_h, self.out_w = out_size
+        else:
+            raise TypeError('"out_size" must be an integer or tuple of integers')
+
+        self.spatial_scale = float(spatial_scale)
+        self.sample_num = int(sample_num)
+        self.aligned = aligned
+        self.clockwise = clockwise
+
+    def forward(self, feats, rois):
+        output = roi_align_rotated(
+            feats,
+            rois,
+            self.out_h,
+            self.out_w,
+            self.spatial_scale,
+            self.sample_num,
+            self.aligned,
+            self.clockwise,
+        )
+        return output
diff --git a/ppocr/losses/__init__.py b/ppocr/losses/__init__.py
new file mode 100644
index 0000000..59b7ecf
--- /dev/null
+++ b/ppocr/losses/__init__.py
@@ -0,0 +1,124 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import copy
+import paddle
+import paddle.nn as nn
+
+# basic_loss
+from .basic_loss import LossFromOutput
+
+# det loss
+from .det_db_loss import DBLoss
+from .det_east_loss import EASTLoss
+from .det_sast_loss import SASTLoss
+from .det_pse_loss import PSELoss
+from .det_fce_loss import FCELoss
+from .det_ct_loss import CTLoss
+from .det_drrg_loss import DRRGLoss
+
+# rec loss
+from .rec_ctc_loss import CTCLoss
+from .rec_att_loss import AttentionLoss
+from .rec_srn_loss import SRNLoss
+from .rec_ce_loss import CELoss
+from .rec_sar_loss import SARLoss
+from .rec_aster_loss import AsterLoss
+from .rec_pren_loss import PRENLoss
+from .rec_multi_loss import MultiLoss
+from .rec_vl_loss import VLLoss
+from .rec_spin_att_loss import SPINAttentionLoss
+from .rec_rfl_loss import RFLLoss
+from .rec_can_loss import CANLoss
+from .rec_satrn_loss import SATRNLoss
+from .rec_nrtr_loss import NRTRLoss
+from .rec_parseq_loss import ParseQLoss
+from .rec_cppd_loss import CPPDLoss
+from .rec_latexocr_loss import LaTeXOCRLoss
+from .rec_unimernet_loss import UniMERNetLoss
+from .rec_ppformulanet_loss import PPFormulaNet_S_Loss, PPFormulaNet_L_Loss
+
+# cls loss
+from .cls_loss import ClsLoss
+
+# e2e loss
+from .e2e_pg_loss import PGLoss
+from .kie_sdmgr_loss import SDMGRLoss
+
+# basic loss function
+from .basic_loss import DistanceLoss
+
+# combined loss function
+from .combined_loss import CombinedLoss
+
+# table loss
+from .table_att_loss import TableAttentionLoss, SLALoss
+from .table_master_loss import TableMasterLoss
+
+# vqa token loss
+from .vqa_token_layoutlm_loss import VQASerTokenLayoutLMLoss
+
+# sr loss
+from .stroke_focus_loss import StrokeFocusLoss
+from .text_focus_loss import TelescopeLoss
+
+
+def build_loss(config):
+    support_dict = [
+        "DBLoss",
+        "PSELoss",
+        "EASTLoss",
+        "SASTLoss",
+        "FCELoss",
+        "CTCLoss",
+        "ClsLoss",
+        "AttentionLoss",
+        "SRNLoss",
+        "PGLoss",
+        "CombinedLoss",
+        "CELoss",
+        "TableAttentionLoss",
+        "SARLoss",
+        "AsterLoss",
+        "SDMGRLoss",
+        "VQASerTokenLayoutLMLoss",
+        "LossFromOutput",
+        "PRENLoss",
+        "MultiLoss",
+        "TableMasterLoss",
+        "SPINAttentionLoss",
+        "VLLoss",
+        "StrokeFocusLoss",
+        "SLALoss",
+        "CTLoss",
+        "RFLLoss",
+        "DRRGLoss",
+        "CANLoss",
+        "TelescopeLoss",
+        "SATRNLoss",
+        "NRTRLoss",
+        "ParseQLoss",
+        "CPPDLoss",
+        "LaTeXOCRLoss",
+        "UniMERNetLoss",
+        "PPFormulaNet_S_Loss",
+        "PPFormulaNet_L_Loss",
+    ]
+    config = copy.deepcopy(config)
+    module_name = config.pop("name")
+    assert module_name in support_dict, Exception(
+        "loss only support {}".format(support_dict)
+    )
+    module_class = eval(module_name)(**config)
+    return module_class
diff --git a/ppocr/losses/ace_loss.py b/ppocr/losses/ace_loss.py
new file mode 100644
index 0000000..961528d
--- /dev/null
+++ b/ppocr/losses/ace_loss.py
@@ -0,0 +1,49 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This code is refer from: https://github.com/viig99/LS-ACELoss
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+import paddle.nn as nn
+
+
+class ACELoss(nn.Layer):
+    def __init__(self, **kwargs):
+        super().__init__()
+        self.loss_func = nn.CrossEntropyLoss(
+            weight=None, ignore_index=0, reduction="none", soft_label=True, axis=-1
+        )
+
+    def __call__(self, predicts, batch):
+        if isinstance(predicts, (list, tuple)):
+            predicts = predicts[-1]
+
+        B, N = predicts.shape[:2]
+        div = paddle.to_tensor([N]).astype("float32")
+
+        predicts = nn.functional.softmax(predicts, axis=-1)
+        aggregation_preds = paddle.sum(predicts, axis=1)
+        aggregation_preds = paddle.divide(aggregation_preds, div)
+
+        length = batch[2].astype("float32")
+        batch = batch[3].astype("float32")
+        batch[:, 0] = paddle.subtract(div, length)
+        batch = paddle.divide(batch, div)
+
+        loss = self.loss_func(aggregation_preds, batch)
+        return {"loss_ace": loss}
diff --git a/ppocr/losses/basic_loss.py b/ppocr/losses/basic_loss.py
new file mode 100644
index 0000000..e85911a
--- /dev/null
+++ b/ppocr/losses/basic_loss.py
@@ -0,0 +1,247 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+from paddle.nn import L1Loss
+from paddle.nn import MSELoss as L2Loss
+from paddle.nn import SmoothL1Loss
+
+
+class CELoss(nn.Layer):
+    def __init__(self, epsilon=None):
+        super().__init__()
+        if epsilon is not None and (epsilon <= 0 or epsilon >= 1):
+            epsilon = None
+        self.epsilon = epsilon
+
+    def _labelsmoothing(self, target, class_num):
+        if target.shape[-1] != class_num:
+            one_hot_target = F.one_hot(target, class_num)
+        else:
+            one_hot_target = target
+        soft_target = F.label_smooth(one_hot_target, epsilon=self.epsilon)
+        soft_target = paddle.reshape(soft_target, shape=[-1, class_num])
+        return soft_target
+
+    def forward(self, x, label):
+        loss_dict = {}
+        if self.epsilon is not None:
+            class_num = x.shape[-1]
+            label = self._labelsmoothing(label, class_num)
+            x = -F.log_softmax(x, axis=-1)
+            loss = paddle.sum(x * label, axis=-1)
+        else:
+            if label.shape[-1] == x.shape[-1]:
+                label = F.softmax(label, axis=-1)
+                soft_label = True
+            else:
+                soft_label = False
+            loss = F.cross_entropy(x, label=label, soft_label=soft_label)
+        return loss
+
+
+class KLJSLoss(object):
+    def __init__(self, mode="kl"):
+        assert mode in [
+            "kl",
+            "js",
+            "KL",
+            "JS",
+        ], "mode can only be one of ['kl', 'KL', 'js', 'JS']"
+        self.mode = mode
+
+    def __call__(self, p1, p2, reduction="mean", eps=1e-5):
+        if self.mode.lower() == "kl":
+            loss = paddle.multiply(p2, paddle.log((p2 + eps) / (p1 + eps) + eps))
+            loss += paddle.multiply(p1, paddle.log((p1 + eps) / (p2 + eps) + eps))
+            loss *= 0.5
+        elif self.mode.lower() == "js":
+            loss = paddle.multiply(
+                p2, paddle.log((2 * p2 + eps) / (p1 + p2 + eps) + eps)
+            )
+            loss += paddle.multiply(
+                p1, paddle.log((2 * p1 + eps) / (p1 + p2 + eps) + eps)
+            )
+            loss *= 0.5
+        else:
+            raise ValueError(
+                "The mode.lower() if KLJSLoss should be one of ['kl', 'js']"
+            )
+
+        if reduction == "mean":
+            loss = paddle.mean(loss, axis=[1, 2])
+        elif reduction == "none" or reduction is None:
+            return loss
+        else:
+            loss = paddle.sum(loss, axis=[1, 2])
+
+        return loss
+
+
+class DMLLoss(nn.Layer):
+    """
+    DMLLoss
+    """
+
+    def __init__(self, act=None, use_log=False):
+        super().__init__()
+        if act is not None:
+            assert act in ["softmax", "sigmoid"]
+        if act == "softmax":
+            self.act = nn.Softmax(axis=-1)
+        elif act == "sigmoid":
+            self.act = nn.Sigmoid()
+        else:
+            self.act = None
+
+        self.use_log = use_log
+        self.jskl_loss = KLJSLoss(mode="kl")
+
+    def _kldiv(self, x, target):
+        eps = 1.0e-10
+        loss = target * (paddle.log(target + eps) - x)
+        # batch mean loss
+        loss = paddle.sum(loss) / loss.shape[0]
+        return loss
+
+    def forward(self, out1, out2):
+        if self.act is not None:
+            out1 = self.act(out1) + 1e-10
+            out2 = self.act(out2) + 1e-10
+        if self.use_log:
+            # for recognition distillation, log is needed for feature map
+            log_out1 = paddle.log(out1)
+            log_out2 = paddle.log(out2)
+            loss = (self._kldiv(log_out1, out2) + self._kldiv(log_out2, out1)) / 2.0
+        else:
+            # for detection distillation log is not needed
+            loss = self.jskl_loss(out1, out2)
+        return loss
+
+
+class DistanceLoss(nn.Layer):
+    """
+    DistanceLoss:
+        mode: loss mode
+    """
+
+    def __init__(self, mode="l2", **kargs):
+        super().__init__()
+        assert mode in ["l1", "l2", "smooth_l1"]
+        if mode == "l1":
+            self.loss_func = nn.L1Loss(**kargs)
+        elif mode == "l2":
+            self.loss_func = nn.MSELoss(**kargs)
+        elif mode == "smooth_l1":
+            self.loss_func = nn.SmoothL1Loss(**kargs)
+
+    def forward(self, x, y):
+        return self.loss_func(x, y)
+
+
+class LossFromOutput(nn.Layer):
+    def __init__(self, key="loss", reduction="none"):
+        super().__init__()
+        self.key = key
+        self.reduction = reduction
+
+    def forward(self, predicts, batch):
+        loss = predicts
+        if self.key is not None and isinstance(predicts, dict):
+            loss = loss[self.key]
+        if self.reduction == "mean":
+            loss = paddle.mean(loss)
+        elif self.reduction == "sum":
+            loss = paddle.sum(loss)
+        return {"loss": loss}
+
+
+class KLDivLoss(nn.Layer):
+    """
+    KLDivLoss
+    """
+
+    def __init__(self):
+        super().__init__()
+
+    def _kldiv(self, x, target, mask=None):
+        eps = 1.0e-10
+        loss = target * (paddle.log(target + eps) - x)
+        if mask is not None:
+            loss = loss.flatten(0, 1).sum(axis=1)
+            loss = loss.masked_select(mask).mean()
+        else:
+            # batch mean loss
+            loss = paddle.sum(loss) / loss.shape[0]
+        return loss
+
+    def forward(self, logits_s, logits_t, mask=None):
+        log_out_s = F.log_softmax(logits_s, axis=-1)
+        out_t = F.softmax(logits_t, axis=-1)
+        loss = self._kldiv(log_out_s, out_t, mask)
+        return loss
+
+
+class DKDLoss(nn.Layer):
+    """
+    KLDivLoss
+    """
+
+    def __init__(self, temperature=1.0, alpha=1.0, beta=1.0):
+        super().__init__()
+        self.temperature = temperature
+        self.alpha = alpha
+        self.beta = beta
+
+    def _cat_mask(self, t, mask1, mask2):
+        t1 = (t * mask1).sum(axis=1, keepdim=True)
+        t2 = (t * mask2).sum(axis=1, keepdim=True)
+        rt = paddle.concat([t1, t2], axis=1)
+        return rt
+
+    def _kl_div(self, x, label, mask=None):
+        y = (label * (paddle.log(label + 1e-10) - x)).sum(axis=1)
+        if mask is not None:
+            y = y.masked_select(mask).mean()
+        else:
+            y = y.mean()
+        return y
+
+    def forward(self, logits_student, logits_teacher, target, mask=None):
+        gt_mask = F.one_hot(target.reshape([-1]), num_classes=logits_student.shape[-1])
+        other_mask = 1 - gt_mask
+        logits_student = logits_student.flatten(0, 1)
+        logits_teacher = logits_teacher.flatten(0, 1)
+        pred_student = F.softmax(logits_student / self.temperature, axis=1)
+        pred_teacher = F.softmax(logits_teacher / self.temperature, axis=1)
+        pred_student = self._cat_mask(pred_student, gt_mask, other_mask)
+        pred_teacher = self._cat_mask(pred_teacher, gt_mask, other_mask)
+        log_pred_student = paddle.log(pred_student)
+        tckd_loss = self._kl_div(log_pred_student, pred_teacher) * (self.temperature**2)
+        pred_teacher_part2 = F.softmax(
+            logits_teacher / self.temperature - 1000.0 * gt_mask, axis=1
+        )
+        log_pred_student_part2 = F.log_softmax(
+            logits_student / self.temperature - 1000.0 * gt_mask, axis=1
+        )
+        nckd_loss = self._kl_div(log_pred_student_part2, pred_teacher_part2) * (
+            self.temperature**2
+        )
+
+        loss = self.alpha * tckd_loss + self.beta * nckd_loss
+
+        return loss
diff --git a/ppocr/losses/center_loss.py b/ppocr/losses/center_loss.py
new file mode 100644
index 0000000..87d4d44
--- /dev/null
+++ b/ppocr/losses/center_loss.py
@@ -0,0 +1,89 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This code is refer from: https://github.com/KaiyangZhou/pytorch-center-loss
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+import os
+import pickle
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+
+class CenterLoss(nn.Layer):
+    """
+    Reference: Wen et al. A Discriminative Feature Learning Approach for Deep Face Recognition. ECCV 2016.
+    """
+
+    def __init__(self, num_classes=6625, feat_dim=96, center_file_path=None):
+        super().__init__()
+        self.num_classes = num_classes
+        self.feat_dim = feat_dim
+        self.centers = paddle.randn(shape=[self.num_classes, self.feat_dim]).astype(
+            "float64"
+        )
+
+        if center_file_path is not None:
+            assert os.path.exists(
+                center_file_path
+            ), f"center path({center_file_path}) must exist when it is not None."
+            with open(center_file_path, "rb") as f:
+                char_dict = pickle.load(f)
+                for key in char_dict.keys():
+                    self.centers[key] = paddle.to_tensor(char_dict[key])
+
+    def __call__(self, predicts, batch):
+        assert isinstance(predicts, (list, tuple))
+        features, predicts = predicts
+
+        feats_reshape = paddle.reshape(features, [-1, features.shape[-1]]).astype(
+            "float64"
+        )
+        label = paddle.argmax(predicts, axis=2)
+        label = paddle.reshape(label, [label.shape[0] * label.shape[1]])
+
+        batch_size = feats_reshape.shape[0]
+
+        # calc l2 distance between feats and centers
+        square_feat = paddle.sum(paddle.square(feats_reshape), axis=1, keepdim=True)
+        square_feat = paddle.expand(square_feat, [batch_size, self.num_classes])
+
+        square_center = paddle.sum(paddle.square(self.centers), axis=1, keepdim=True)
+        square_center = paddle.expand(
+            square_center, [self.num_classes, batch_size]
+        ).astype("float64")
+        square_center = paddle.transpose(square_center, [1, 0])
+
+        distmat = paddle.add(square_feat, square_center)
+        feat_dot_center = paddle.matmul(
+            feats_reshape, paddle.transpose(self.centers, [1, 0])
+        )
+        distmat = distmat - 2.0 * feat_dot_center
+
+        # generate the mask
+        classes = paddle.arange(self.num_classes).astype("int64")
+        label = paddle.expand(
+            paddle.unsqueeze(label, 1), (batch_size, self.num_classes)
+        )
+        mask = paddle.equal(
+            paddle.expand(classes, [batch_size, self.num_classes]), label
+        ).astype("float64")
+        dist = paddle.multiply(distmat, mask)
+
+        loss = paddle.sum(paddle.clip(dist, min=1e-12, max=1e12)) / batch_size
+        return {"loss_center": loss}
diff --git a/ppocr/losses/cls_loss.py b/ppocr/losses/cls_loss.py
new file mode 100755
index 0000000..eb3b17e
--- /dev/null
+++ b/ppocr/losses/cls_loss.py
@@ -0,0 +1,30 @@
+# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from paddle import nn
+
+
+class ClsLoss(nn.Layer):
+    def __init__(self, **kwargs):
+        super(ClsLoss, self).__init__()
+        self.loss_func = nn.CrossEntropyLoss(reduction="mean")
+
+    def forward(self, predicts, batch):
+        label = batch[1].astype("int64")
+        loss = self.loss_func(input=predicts, label=label)
+        return {"loss": loss}
diff --git a/ppocr/losses/combined_loss.py b/ppocr/losses/combined_loss.py
new file mode 100644
index 0000000..10cdb01
--- /dev/null
+++ b/ppocr/losses/combined_loss.py
@@ -0,0 +1,84 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle
+import paddle.nn as nn
+
+from .rec_ctc_loss import CTCLoss
+from .center_loss import CenterLoss
+from .ace_loss import ACELoss
+from .rec_sar_loss import SARLoss
+
+from .distillation_loss import DistillationCTCLoss, DistillCTCLogits
+from .distillation_loss import DistillationSARLoss, DistillationNRTRLoss
+from .distillation_loss import (
+    DistillationDMLLoss,
+    DistillationKLDivLoss,
+    DistillationDKDLoss,
+)
+from .distillation_loss import (
+    DistillationDistanceLoss,
+    DistillationDBLoss,
+    DistillationDilaDBLoss,
+)
+from .distillation_loss import (
+    DistillationVQASerTokenLayoutLMLoss,
+    DistillationSERDMLLoss,
+)
+from .distillation_loss import DistillationLossFromOutput
+from .distillation_loss import DistillationVQADistanceLoss
+
+
+class CombinedLoss(nn.Layer):
+    """
+    CombinedLoss:
+        a combionation of loss function
+    """
+
+    def __init__(self, loss_config_list=None):
+        super().__init__()
+        self.loss_func = []
+        self.loss_weight = []
+        assert isinstance(loss_config_list, list), "operator config should be a list"
+        for config in loss_config_list:
+            assert isinstance(config, dict) and len(config) == 1, "yaml format error"
+            name = list(config)[0]
+            param = config[name]
+            assert (
+                "weight" in param
+            ), "weight must be in param, but param just contains {}".format(
+                param.keys()
+            )
+            self.loss_weight.append(param.pop("weight"))
+            self.loss_func.append(eval(name)(**param))
+
+    def forward(self, input, batch, **kargs):
+        loss_dict = {}
+        loss_all = 0.0
+        for idx, loss_func in enumerate(self.loss_func):
+            loss = loss_func(input, batch, **kargs)
+            if isinstance(loss, paddle.Tensor):
+                loss = {"loss_{}_{}".format(str(loss), idx): loss}
+
+            weight = self.loss_weight[idx]
+
+            loss = {key: loss[key] * weight for key in loss}
+
+            if "loss" in loss:
+                loss_all += loss["loss"]
+            else:
+                loss_all += paddle.add_n(list(loss.values()))
+            loss_dict.update(loss)
+        loss_dict["loss"] = loss_all
+        return loss_dict
diff --git a/ppocr/losses/det_basic_loss.py b/ppocr/losses/det_basic_loss.py
new file mode 100644
index 0000000..bee82b7
--- /dev/null
+++ b/ppocr/losses/det_basic_loss.py
@@ -0,0 +1,284 @@
+# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/WenmuZhou/DBNet.pytorch/blob/master/models/losses/basic_loss.py
+"""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+
+
+class BalanceLoss(nn.Layer):
+    def __init__(
+        self,
+        balance_loss=True,
+        main_loss_type="DiceLoss",
+        negative_ratio=3,
+        return_origin=False,
+        eps=1e-6,
+        **kwargs,
+    ):
+        """
+        The BalanceLoss for Differentiable Binarization text detection
+        args:
+            balance_loss (bool): whether balance loss or not, default is True
+            main_loss_type (str): can only be one of ['CrossEntropy','DiceLoss',
+                'Euclidean','BCELoss', 'MaskL1Loss'], default is  'DiceLoss'.
+            negative_ratio (int|float): float, default is 3.
+            return_origin (bool): whether return unbalanced loss or not, default is False.
+            eps (float): default is 1e-6.
+        """
+        super(BalanceLoss, self).__init__()
+        self.balance_loss = balance_loss
+        self.main_loss_type = main_loss_type
+        self.negative_ratio = negative_ratio
+        self.return_origin = return_origin
+        self.eps = eps
+
+        if self.main_loss_type == "CrossEntropy":
+            self.loss = nn.CrossEntropyLoss()
+        elif self.main_loss_type == "Euclidean":
+            self.loss = nn.MSELoss()
+        elif self.main_loss_type == "DiceLoss":
+            self.loss = DiceLoss(self.eps)
+        elif self.main_loss_type == "BCELoss":
+            self.loss = BCELoss(reduction="none")
+        elif self.main_loss_type == "MaskL1Loss":
+            self.loss = MaskL1Loss(self.eps)
+        else:
+            loss_type = [
+                "CrossEntropy",
+                "DiceLoss",
+                "Euclidean",
+                "BCELoss",
+                "MaskL1Loss",
+            ]
+            raise Exception(
+                "main_loss_type in BalanceLoss() can only be one of {}".format(
+                    loss_type
+                )
+            )
+
+    def forward(self, pred, gt, mask=None):
+        """
+        The BalanceLoss for Differentiable Binarization text detection
+        args:
+            pred (variable): predicted feature maps.
+            gt (variable): ground truth feature maps.
+            mask (variable): masked maps.
+        return: (variable) balanced loss
+        """
+        positive = gt * mask
+        negative = (1 - gt) * mask
+
+        positive_count = int(positive.sum())
+        negative_count = int(min(negative.sum(), positive_count * self.negative_ratio))
+        loss = self.loss(pred, gt, mask=mask)
+
+        if not self.balance_loss:
+            return loss
+
+        positive_loss = positive * loss
+        negative_loss = negative * loss
+        negative_loss = paddle.reshape(negative_loss, shape=[-1])
+        if negative_count > 0:
+            sort_loss = negative_loss.sort(descending=True)
+            negative_loss = sort_loss[:negative_count]
+            # negative_loss, _ = paddle.topk(negative_loss, k=negative_count_int)
+            balance_loss = (positive_loss.sum() + negative_loss.sum()) / (
+                positive_count + negative_count + self.eps
+            )
+        else:
+            balance_loss = positive_loss.sum() / (positive_count + self.eps)
+        if self.return_origin:
+            return balance_loss, loss
+
+        return balance_loss
+
+
+class DiceLoss(nn.Layer):
+    def __init__(self, eps=1e-6):
+        super(DiceLoss, self).__init__()
+        self.eps = eps
+
+    def forward(self, pred, gt, mask, weights=None):
+        """
+        DiceLoss function.
+        """
+
+        assert pred.shape == gt.shape
+        assert pred.shape == mask.shape
+        if weights is not None:
+            assert weights.shape == mask.shape
+            mask = weights * mask
+        intersection = paddle.sum(pred * gt * mask)
+
+        union = paddle.sum(pred * mask) + paddle.sum(gt * mask) + self.eps
+        loss = 1 - 2.0 * intersection / union
+        assert loss <= 1
+        return loss
+
+
+class MaskL1Loss(nn.Layer):
+    def __init__(self, eps=1e-6):
+        super(MaskL1Loss, self).__init__()
+        self.eps = eps
+
+    def forward(self, pred, gt, mask):
+        """
+        Mask L1 Loss
+        """
+        loss = (paddle.abs(pred - gt) * mask).sum() / (mask.sum() + self.eps)
+        loss = paddle.mean(loss)
+        return loss
+
+
+class BCELoss(nn.Layer):
+    def __init__(self, reduction="mean"):
+        super(BCELoss, self).__init__()
+        self.reduction = reduction
+
+    def forward(self, input, label, mask=None, weight=None, name=None):
+        loss = F.binary_cross_entropy(input, label, reduction=self.reduction)
+        return loss
+
+
+class MaskedFocalLoss(nn.Layer):
+    """
+    Binary Focal Loss with mask support, designed for text segmentation tasks.
+
+    Focal Loss addresses class imbalance by down-weighting easy examples and
+    focusing training on hard examples:
+        FL(p_t) = -alpha_t * (1 - p_t)^gamma * log(p_t)
+
+    Compared to OHEM (which hard-selects a fixed ratio of negatives), Focal Loss
+    applies a continuous per-pixel weight that gracefully scales with difficulty,
+    making it a strictly superior drop-in for the OHEM + DiceLoss pattern when
+    DiceLoss returns a scalar and OHEM has no discriminating effect.
+
+    Args:
+        alpha (float): Balancing factor for the positive (text) class.
+            Since text pixels are a small minority, alpha > 0.5 gives them
+            higher weight. Default: 0.75.
+        gamma (float): Focusing parameter. gamma=0 reduces to masked BCE.
+            gamma=2 is the standard value from the original Focal Loss paper.
+            Default: 2.0.
+        eps (float): Small constant for numerical stability. Default: 1e-6.
+    """
+
+    def __init__(self, alpha=0.25, gamma=2.0, eps=1e-6):
+        super(MaskedFocalLoss, self).__init__()
+        self.alpha = alpha
+        self.gamma = gamma
+        self.eps = eps
+
+    def forward(self, pred, gt, mask):
+        """
+        Args:
+            pred  (Tensor): Predicted probability map, shape (B, H, W), in [0, 1].
+                            (i.e. after sigmoid — the direct output of DBHead.binarize)
+            gt    (Tensor): Binary ground-truth map, shape (B, H, W), values 0 or 1.
+            mask  (Tensor): Valid-pixel mask, shape (B, H, W), values 0 or 1.
+                            Pixels with mask=0 are ignored regions (e.g. too-small text).
+        Returns:
+            Tensor: Scalar focal loss averaged over valid (mask=1) pixels.
+        """
+        # F.sigmoid_focal_loss expects a logit (pre-sigmoid) input and applies
+        # sigmoid internally using the numerically stable log-sum-exp form:
+        #   log(σ(x)) = -softplus(-x),  log(1-σ(x)) = -softplus(x)
+        # This avoids the log(0) issue of the manual implementation.
+        # Since pred is already a probability (post-sigmoid from DBHead), we
+        # invert it: logit = log(p / (1-p)).  The round-trip is numerically safe
+        # after clamping, and the stable path inside paddle takes over from there.
+        pred = paddle.clip(pred, self.eps, 1.0 - self.eps)
+        logit = paddle.log(pred / (1.0 - pred))
+
+        # Per-pixel focal loss, shape (B, H, W)
+        # reduction='none' so we can apply the mask ourselves
+        loss = F.sigmoid_focal_loss(
+            logit,
+            gt,
+            normalizer=None,
+            alpha=self.alpha,
+            gamma=self.gamma,
+            reduction="none",
+        )
+
+        # Average over valid (mask=1) pixels only
+        return (loss * mask).sum() / (mask.sum() + self.eps)
+
+
+class DiceFocalLoss(nn.Layer):
+    """
+    Combined DiceLoss + MaskedFocalLoss for binary text segmentation.
+
+    Rationale for the combination:
+    - DiceLoss optimizes the global F1 / region overlap between prediction and GT.
+      It is naturally robust to class imbalance (text vs background) because it
+      normalizes by the sum of both sets, not by pixel count.
+    - MaskedFocalLoss provides per-pixel supervision with adaptive hard-example
+      weighting. It compensates for DiceLoss being a global metric that cannot
+      distinguish which specific pixels are mispredicted.
+    Together they provide complementary supervision: DiceLoss for global shape
+    quality, FocalLoss for pixel-level precision on ambiguous boundaries.
+
+    This design follows the Dice + Focal combination used in mmsegmentation and
+    segmentation_models_pytorch for binary segmentation with class imbalance.
+
+    This class is a drop-in replacement for BalanceLoss when main_loss_type is
+    'DiceLoss' — both share the same forward(pred, gt, mask) signature and
+    return a scalar.
+
+    Args:
+        dice_weight  (float): Weight for the DiceLoss term. Default: 1.0.
+        focal_weight (float): Weight for the MaskedFocalLoss term. Default: 1.0.
+        focal_alpha  (float): Positive-class balancing factor for FocalLoss.
+            Default: 0.75.
+        focal_gamma  (float): Focusing exponent for FocalLoss. Default: 2.0.
+        eps          (float): Numerical stability constant. Default: 1e-6.
+    """
+
+    def __init__(
+        self,
+        dice_weight=1.0,
+        focal_weight=1.0,
+        focal_alpha=0.75,
+        focal_gamma=2.0,
+        eps=1e-6,
+    ):
+        super(DiceFocalLoss, self).__init__()
+        self.dice_weight = dice_weight
+        self.focal_weight = focal_weight
+        self.dice_loss = DiceLoss(eps=eps)
+        self.focal_loss = MaskedFocalLoss(alpha=focal_alpha, gamma=focal_gamma, eps=eps)
+
+    def forward(self, pred, gt, mask=None):
+        """
+        Args:
+            pred (Tensor): Predicted probability map, shape (B, H, W), in [0, 1].
+            gt   (Tensor): Binary ground-truth shrink map, shape (B, H, W).
+            mask (Tensor): Valid-pixel mask, shape (B, H, W).
+        Returns:
+            Tensor: Scalar combined loss.
+        """
+        loss_dice = self.dice_loss(pred, gt, mask)
+        loss_focal = self.focal_loss(pred, gt, mask)
+        return self.dice_weight * loss_dice + self.focal_weight * loss_focal
diff --git a/ppocr/losses/det_ct_loss.py b/ppocr/losses/det_ct_loss.py
new file mode 100755
index 0000000..4655eff
--- /dev/null
+++ b/ppocr/losses/det_ct_loss.py
@@ -0,0 +1,302 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/shengtao96/CentripetalText/tree/main/models/loss
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+import numpy as np
+
+
+def ohem_single(score, gt_text, training_mask):
+    # online hard example mining
+
+    pos_num = int(paddle.sum(gt_text > 0.5)) - int(
+        paddle.sum((gt_text > 0.5) & (training_mask <= 0.5))
+    )
+
+    if pos_num == 0:
+        # selected_mask = gt_text.copy() * 0 # may be not good
+        selected_mask = training_mask
+        selected_mask = paddle.cast(
+            selected_mask.reshape((1, selected_mask.shape[0], selected_mask.shape[1])),
+            "float32",
+        )
+        return selected_mask
+
+    neg_num = int(paddle.sum((gt_text <= 0.5) & (training_mask > 0.5)))
+    neg_num = int(min(pos_num * 3, neg_num))
+
+    if neg_num == 0:
+        selected_mask = training_mask
+        selected_mask = paddle.cast(
+            selected_mask.reshape((1, selected_mask.shape[0], selected_mask.shape[1])),
+            "float32",
+        )
+        return selected_mask
+
+    # hard example
+    neg_score = score[(gt_text <= 0.5) & (training_mask > 0.5)]
+    neg_score_sorted = paddle.sort(-neg_score)
+    threshold = -neg_score_sorted[neg_num - 1]
+
+    selected_mask = ((score >= threshold) | (gt_text > 0.5)) & (training_mask > 0.5)
+    selected_mask = paddle.cast(
+        selected_mask.reshape((1, selected_mask.shape[0], selected_mask.shape[1])),
+        "float32",
+    )
+    return selected_mask
+
+
+def ohem_batch(scores, gt_texts, training_masks):
+    selected_masks = []
+    for i in range(scores.shape[0]):
+        selected_masks.append(
+            ohem_single(scores[i, :, :], gt_texts[i, :, :], training_masks[i, :, :])
+        )
+
+    selected_masks = paddle.cast(paddle.concat(selected_masks, 0), "float32")
+    return selected_masks
+
+
+def iou_single(a, b, mask, n_class):
+    EPS = 1e-6
+    valid = mask == 1
+    a = a[valid]
+    b = b[valid]
+    miou = []
+
+    # iou of each class
+    for i in range(n_class):
+        inter = paddle.cast(((a == i) & (b == i)), "float32")
+        union = paddle.cast(((a == i) | (b == i)), "float32")
+
+        miou.append(paddle.sum(inter) / (paddle.sum(union) + EPS))
+    miou = sum(miou) / len(miou)
+    return miou
+
+
+def iou(a, b, mask, n_class=2, reduce=True):
+    batch_size = a.shape[0]
+
+    a = a.reshape((batch_size, -1))
+    b = b.reshape((batch_size, -1))
+    mask = mask.reshape((batch_size, -1))
+
+    iou = paddle.zeros((batch_size,), dtype="float32")
+    for i in range(batch_size):
+        iou[i] = iou_single(a[i], b[i], mask[i], n_class)
+
+    if reduce:
+        iou = paddle.mean(iou)
+    return iou
+
+
+class DiceLoss(nn.Layer):
+    def __init__(self, loss_weight=1.0):
+        super(DiceLoss, self).__init__()
+        self.loss_weight = loss_weight
+
+    def forward(self, input, target, mask, reduce=True):
+        batch_size = input.shape[0]
+        input = F.sigmoid(input)  # scale to 0-1
+
+        input = input.reshape((batch_size, -1))
+        target = paddle.cast(target.reshape((batch_size, -1)), "float32")
+        mask = paddle.cast(mask.reshape((batch_size, -1)), "float32")
+
+        input = input * mask
+        target = target * mask
+
+        a = paddle.sum(input * target, axis=1)
+        b = paddle.sum(input * input, axis=1) + 0.001
+        c = paddle.sum(target * target, axis=1) + 0.001
+        d = (2 * a) / (b + c)
+        loss = 1 - d
+
+        loss = self.loss_weight * loss
+
+        if reduce:
+            loss = paddle.mean(loss)
+
+        return loss
+
+
+class SmoothL1Loss(nn.Layer):
+    def __init__(self, beta=1.0, loss_weight=1.0):
+        super(SmoothL1Loss, self).__init__()
+        self.beta = beta
+        self.loss_weight = loss_weight
+
+        np_coord = np.zeros(shape=[640, 640, 2], dtype=np.int64)
+        for i in range(640):
+            for j in range(640):
+                np_coord[i, j, 0] = j
+                np_coord[i, j, 1] = i
+        np_coord = np_coord.reshape((-1, 2))
+
+        self.coord = self.create_parameter(
+            shape=[640 * 640, 2],
+            dtype="int32",  # NOTE: not support "int64" before paddle 2.3.1
+            default_initializer=nn.initializer.Assign(value=np_coord),
+        )
+        self.coord.stop_gradient = True
+
+    def forward_single(self, input, target, mask, beta=1.0, eps=1e-6):
+        batch_size = input.shape[0]
+
+        diff = paddle.abs(input - target) * mask.unsqueeze(1)
+        loss = paddle.where(diff < beta, 0.5 * diff * diff / beta, diff - 0.5 * beta)
+        loss = paddle.cast(loss.reshape((batch_size, -1)), "float32")
+        mask = paddle.cast(mask.reshape((batch_size, -1)), "float32")
+        loss = paddle.sum(loss, axis=-1)
+        loss = loss / (mask.sum(axis=-1) + eps)
+
+        return loss
+
+    def select_single(self, distance, gt_instance, gt_kernel_instance, training_mask):
+        with paddle.no_grad():
+            # paddle 2.3.1, paddle.slice not support:
+            # distance[:, self.coord[:, 1], self.coord[:, 0]]
+            select_distance_list = []
+            for i in range(2):
+                tmp1 = distance[i, :]
+                tmp2 = tmp1[self.coord[:, 1], self.coord[:, 0]]
+                select_distance_list.append(tmp2.unsqueeze(0))
+            select_distance = paddle.concat(select_distance_list, axis=0)
+
+            off_points = paddle.cast(
+                self.coord, "float32"
+            ) + 10 * select_distance.transpose((1, 0))
+
+            off_points = paddle.cast(off_points, "int64")
+            off_points = paddle.clip(off_points, 0, distance.shape[-1] - 1)
+
+            selected_mask = (
+                gt_instance[self.coord[:, 1], self.coord[:, 0]]
+                != gt_kernel_instance[off_points[:, 1], off_points[:, 0]]
+            )
+            selected_mask = paddle.cast(
+                selected_mask.reshape((1, -1, distance.shape[-1])), "int64"
+            )
+            selected_training_mask = selected_mask * training_mask
+
+            return selected_training_mask
+
+    def forward(
+        self,
+        distances,
+        gt_instances,
+        gt_kernel_instances,
+        training_masks,
+        gt_distances,
+        reduce=True,
+    ):
+        selected_training_masks = []
+        for i in range(distances.shape[0]):
+            selected_training_masks.append(
+                self.select_single(
+                    distances[i, :, :, :],
+                    gt_instances[i, :, :],
+                    gt_kernel_instances[i, :, :],
+                    training_masks[i, :, :],
+                )
+            )
+        selected_training_masks = paddle.cast(
+            paddle.concat(selected_training_masks, 0), "float32"
+        )
+
+        loss = self.forward_single(
+            distances, gt_distances, selected_training_masks, self.beta
+        )
+        loss = self.loss_weight * loss
+
+        with paddle.no_grad():
+            batch_size = distances.shape[0]
+            false_num = selected_training_masks.reshape((batch_size, -1))
+            false_num = false_num.sum(axis=-1)
+            total_num = paddle.cast(training_masks.reshape((batch_size, -1)), "float32")
+            total_num = total_num.sum(axis=-1)
+            iou_text = (total_num - false_num) / (total_num + 1e-6)
+
+        if reduce:
+            loss = paddle.mean(loss)
+
+        return loss, iou_text
+
+
+class CTLoss(nn.Layer):
+    def __init__(self):
+        super(CTLoss, self).__init__()
+        self.kernel_loss = DiceLoss()
+        self.loc_loss = SmoothL1Loss(beta=0.1, loss_weight=0.05)
+
+    def forward(self, preds, batch):
+        imgs = batch[0]
+        out = preds["maps"]
+        (
+            gt_kernels,
+            training_masks,
+            gt_instances,
+            gt_kernel_instances,
+            training_mask_distances,
+            gt_distances,
+        ) = batch[1:]
+
+        kernels = out[:, 0, :, :]
+        distances = out[:, 1:, :, :]
+
+        # kernel loss
+        selected_masks = ohem_batch(kernels, gt_kernels, training_masks)
+
+        loss_kernel = self.kernel_loss(
+            kernels, gt_kernels, selected_masks, reduce=False
+        )
+
+        iou_kernel = iou(
+            paddle.cast((kernels > 0), "int64"),
+            gt_kernels,
+            training_masks,
+            reduce=False,
+        )
+        losses = dict(
+            loss_kernels=loss_kernel,
+        )
+
+        # loc loss
+        loss_loc, iou_text = self.loc_loss(
+            distances,
+            gt_instances,
+            gt_kernel_instances,
+            training_mask_distances,
+            gt_distances,
+            reduce=False,
+        )
+        losses.update(
+            dict(
+                loss_loc=loss_loc,
+            )
+        )
+
+        loss_all = loss_kernel + loss_loc
+        losses = {"loss": loss_all}
+
+        return losses
diff --git a/ppocr/losses/det_db_loss.py b/ppocr/losses/det_db_loss.py
new file mode 100755
index 0000000..ff0c6ee
--- /dev/null
+++ b/ppocr/losses/det_db_loss.py
@@ -0,0 +1,146 @@
+# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/WenmuZhou/DBNet.pytorch/blob/master/models/losses/DB_loss.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+
+from .det_basic_loss import BalanceLoss, MaskL1Loss, DiceLoss, DiceFocalLoss
+
+
+class DBLoss(nn.Layer):
+    """
+    Differentiable Binarization (DB) Loss Function
+    args:
+        param (dict): the super parameter for DB Loss
+    """
+
+    def __init__(
+        self,
+        balance_loss=True,
+        main_loss_type="DiceLoss",
+        alpha=5,
+        beta=10,
+        ohem_ratio=3,
+        eps=1e-6,
+        aux_weight_p4=0.0,
+        aux_weight_p3=0.0,
+        aux_weight_p2=0.0,
+        focal_alpha=0.25,
+        focal_gamma=2.0,
+        dice_weight=1.0,
+        focal_weight=1.0,
+        **kwargs,
+    ):
+        super(DBLoss, self).__init__()
+        self.alpha = alpha
+        self.beta = beta
+        self.l1_loss = MaskL1Loss(eps=eps)
+        if main_loss_type == "DiceFocalLoss":
+            self.bce_loss = DiceFocalLoss(
+                dice_weight=dice_weight,
+                focal_weight=focal_weight,
+                focal_alpha=focal_alpha,
+                focal_gamma=focal_gamma,
+                eps=eps,
+            )
+            self.dice_loss = self.bce_loss
+        elif main_loss_type == "DiceLoss":
+            self.bce_loss = BalanceLoss(
+                balance_loss=balance_loss,
+                main_loss_type=main_loss_type,
+                negative_ratio=ohem_ratio,
+            )
+            self.dice_loss = DiceLoss(eps=eps)
+        else:
+            raise Exception("[DBLoss]: Unrecognized main loss type!")
+        self.aux_weight_p4 = aux_weight_p4
+        self.aux_weight_p3 = aux_weight_p3
+        self.aux_weight_p2 = aux_weight_p2
+
+    def forward(self, predicts, labels):
+        predict_maps = predicts["maps"]
+        (
+            label_threshold_map,
+            label_threshold_mask,
+            label_shrink_map,
+            label_shrink_mask,
+        ) = labels[1:]
+        shrink_maps = predict_maps[:, 0, :, :]
+        threshold_maps = predict_maps[:, 1, :, :]
+        binary_maps = predict_maps[:, 2, :, :]
+
+        loss_shrink_maps = self.bce_loss(
+            shrink_maps, label_shrink_map, label_shrink_mask
+        )
+        loss_threshold_maps = self.l1_loss(
+            threshold_maps, label_threshold_map, label_threshold_mask
+        )
+        loss_binary_maps = self.dice_loss(
+            binary_maps, label_shrink_map, label_shrink_mask
+        )
+        loss_shrink_maps = self.alpha * loss_shrink_maps
+        loss_threshold_maps = self.beta * loss_threshold_maps
+        # CBN loss
+        if "distance_maps" in predicts.keys():
+            distance_maps = predicts["distance_maps"]
+            cbn_maps = predicts["cbn_maps"]
+            cbn_loss = self.bce_loss(
+                cbn_maps[:, 0, :, :], label_shrink_map, label_shrink_mask
+            )
+        else:
+            dis_loss = paddle.to_tensor([0.0])
+            cbn_loss = paddle.to_tensor([0.0])
+
+        loss_all = loss_shrink_maps + loss_threshold_maps + loss_binary_maps
+        losses = {
+            "loss": loss_all + cbn_loss,
+            "loss_shrink_maps": loss_shrink_maps,
+            "loss_threshold_maps": loss_threshold_maps,
+            "loss_binary_maps": loss_binary_maps,
+            "loss_cbn": cbn_loss,
+        }
+
+        # Auxiliary loss
+        for aux_key, aux_w in [
+            ("aux_maps_p4", self.aux_weight_p4),
+            ("aux_maps_p3", self.aux_weight_p3),
+            ("aux_maps_p2", self.aux_weight_p2),
+        ]:
+            if aux_w > 0 and aux_key in predicts:
+                aux_maps = predicts[aux_key]
+                aux_shrink = aux_maps[:, 0, :, :]
+                aux_threshold = aux_maps[:, 1, :, :]
+                aux_binary = aux_maps[:, 2, :, :]
+                l_shrink = self.alpha * self.bce_loss(
+                    aux_shrink, label_shrink_map, label_shrink_mask
+                )
+                l_threshold = self.beta * self.l1_loss(
+                    aux_threshold, label_threshold_map, label_threshold_mask
+                )
+                l_binary = self.dice_loss(
+                    aux_binary, label_shrink_map, label_shrink_mask
+                )
+                aux_loss = l_shrink + l_threshold + l_binary
+                losses["loss_{}".format(aux_key)] = aux_loss
+                losses["loss"] = losses["loss"] + aux_w * aux_loss
+
+        return losses
diff --git a/ppocr/losses/det_drrg_loss.py b/ppocr/losses/det_drrg_loss.py
new file mode 100644
index 0000000..6e990ce
--- /dev/null
+++ b/ppocr/losses/det_drrg_loss.py
@@ -0,0 +1,234 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textdet/losses/drrg_loss.py
+"""
+
+import paddle
+import paddle.nn.functional as F
+from paddle import nn
+
+
+class DRRGLoss(nn.Layer):
+    def __init__(self, ohem_ratio=3.0):
+        super().__init__()
+        self.ohem_ratio = ohem_ratio
+        self.downsample_ratio = 1.0
+
+    def balance_bce_loss(self, pred, gt, mask):
+        """Balanced Binary-CrossEntropy Loss.
+
+        Args:
+            pred (Tensor): Shape of :math:`(1, H, W)`.
+            gt (Tensor): Shape of :math:`(1, H, W)`.
+            mask (Tensor): Shape of :math:`(1, H, W)`.
+
+        Returns:
+            Tensor: Balanced bce loss.
+        """
+        assert pred.shape == gt.shape == mask.shape
+        assert paddle.all(pred >= 0) and paddle.all(pred <= 1)
+        assert paddle.all(gt >= 0) and paddle.all(gt <= 1)
+        positive = gt * mask
+        negative = (1 - gt) * mask
+        positive_count = int(positive.sum())
+
+        if positive_count > 0:
+            loss = F.binary_cross_entropy(pred, gt, reduction="none")
+            positive_loss = paddle.sum(loss * positive)
+            negative_loss = loss * negative
+            negative_count = min(
+                int(negative.sum()), int(positive_count * self.ohem_ratio)
+            )
+        else:
+            positive_loss = paddle.to_tensor(0.0)
+            loss = F.binary_cross_entropy(pred, gt, reduction="none")
+            negative_loss = loss * negative
+            negative_count = 100
+        negative_loss, _ = paddle.topk(negative_loss.reshape([-1]), negative_count)
+
+        balance_loss = (positive_loss + paddle.sum(negative_loss)) / (
+            float(positive_count + negative_count) + 1e-5
+        )
+
+        return balance_loss
+
+    def gcn_loss(self, gcn_data):
+        """CrossEntropy Loss from gcn module.
+
+        Args:
+            gcn_data (tuple(Tensor, Tensor)): The first is the
+                prediction with shape :math:`(N, 2)` and the
+                second is the gt label with shape :math:`(m, n)`
+                where :math:`m * n = N`.
+
+        Returns:
+            Tensor: CrossEntropy loss.
+        """
+        gcn_pred, gt_labels = gcn_data
+        gt_labels = gt_labels.reshape([-1])
+        loss = F.cross_entropy(gcn_pred, gt_labels)
+
+        return loss
+
+    def bitmasks2tensor(self, bitmasks, target_sz):
+        """Convert Bitmasks to tensor.
+
+        Args:
+            bitmasks (list[BitmapMasks]): The BitmapMasks list. Each item is
+                for one img.
+            target_sz (tuple(int, int)): The target tensor of size
+                :math:`(H, W)`.
+
+        Returns:
+            list[Tensor]: The list of kernel tensors. Each element stands for
+            one kernel level.
+        """
+        batch_size = len(bitmasks)
+        results = []
+
+        kernel = []
+        for batch_inx in range(batch_size):
+            mask = bitmasks[batch_inx]
+            # hxw
+            mask_sz = mask.shape
+            # left, right, top, bottom
+            pad = [0, target_sz[1] - mask_sz[1], 0, target_sz[0] - mask_sz[0]]
+            mask = F.pad(mask, pad, mode="constant", value=0)
+            kernel.append(mask)
+        kernel = paddle.stack(kernel)
+        results.append(kernel)
+
+        return results
+
+    def forward(self, preds, labels):
+        """Compute Drrg loss."""
+
+        assert isinstance(preds, tuple)
+        (
+            gt_text_mask,
+            gt_center_region_mask,
+            gt_mask,
+            gt_top_height_map,
+            gt_bot_height_map,
+            gt_sin_map,
+            gt_cos_map,
+        ) = labels[1:8]
+
+        downsample_ratio = self.downsample_ratio
+
+        pred_maps, gcn_data = preds
+        pred_text_region = pred_maps[:, 0, :, :]
+        pred_center_region = pred_maps[:, 1, :, :]
+        pred_sin_map = pred_maps[:, 2, :, :]
+        pred_cos_map = pred_maps[:, 3, :, :]
+        pred_top_height_map = pred_maps[:, 4, :, :]
+        pred_bot_height_map = pred_maps[:, 5, :, :]
+        feature_sz = pred_maps.shape
+
+        # bitmask 2 tensor
+        mapping = {
+            "gt_text_mask": paddle.cast(gt_text_mask, "float32"),
+            "gt_center_region_mask": paddle.cast(gt_center_region_mask, "float32"),
+            "gt_mask": paddle.cast(gt_mask, "float32"),
+            "gt_top_height_map": paddle.cast(gt_top_height_map, "float32"),
+            "gt_bot_height_map": paddle.cast(gt_bot_height_map, "float32"),
+            "gt_sin_map": paddle.cast(gt_sin_map, "float32"),
+            "gt_cos_map": paddle.cast(gt_cos_map, "float32"),
+        }
+        gt = {}
+        for key, value in mapping.items():
+            gt[key] = value
+            if abs(downsample_ratio - 1.0) < 1e-2:
+                gt[key] = self.bitmasks2tensor(gt[key], feature_sz[2:])
+            else:
+                gt[key] = [item.rescale(downsample_ratio) for item in gt[key]]
+                gt[key] = self.bitmasks2tensor(gt[key], feature_sz[2:])
+                if key in ["gt_top_height_map", "gt_bot_height_map"]:
+                    gt[key] = [item * downsample_ratio for item in gt[key]]
+            gt[key] = [item for item in gt[key]]
+
+        scale = paddle.sqrt(1.0 / (pred_sin_map**2 + pred_cos_map**2 + 1e-8))
+        pred_sin_map = pred_sin_map * scale
+        pred_cos_map = pred_cos_map * scale
+
+        loss_text = self.balance_bce_loss(
+            F.sigmoid(pred_text_region), gt["gt_text_mask"][0], gt["gt_mask"][0]
+        )
+
+        text_mask = gt["gt_text_mask"][0] * gt["gt_mask"][0]
+        negative_text_mask = (1 - gt["gt_text_mask"][0]) * gt["gt_mask"][0]
+        loss_center_map = F.binary_cross_entropy(
+            F.sigmoid(pred_center_region),
+            gt["gt_center_region_mask"][0],
+            reduction="none",
+        )
+        if int(text_mask.sum()) > 0:
+            loss_center_positive = paddle.sum(loss_center_map * text_mask) / paddle.sum(
+                text_mask
+            )
+        else:
+            loss_center_positive = paddle.to_tensor(0.0)
+        loss_center_negative = paddle.sum(
+            loss_center_map * negative_text_mask
+        ) / paddle.sum(negative_text_mask)
+        loss_center = loss_center_positive + 0.5 * loss_center_negative
+
+        center_mask = gt["gt_center_region_mask"][0] * gt["gt_mask"][0]
+        if int(center_mask.sum()) > 0:
+            map_sz = pred_top_height_map.shape
+            ones = paddle.ones(map_sz, dtype="float32")
+            loss_top = F.smooth_l1_loss(
+                pred_top_height_map / (gt["gt_top_height_map"][0] + 1e-2),
+                ones,
+                reduction="none",
+            )
+            loss_bot = F.smooth_l1_loss(
+                pred_bot_height_map / (gt["gt_bot_height_map"][0] + 1e-2),
+                ones,
+                reduction="none",
+            )
+            gt_height = gt["gt_top_height_map"][0] + gt["gt_bot_height_map"][0]
+            loss_height = paddle.sum(
+                (paddle.log(gt_height + 1) * (loss_top + loss_bot)) * center_mask
+            ) / paddle.sum(center_mask)
+
+            loss_sin = paddle.sum(
+                F.smooth_l1_loss(pred_sin_map, gt["gt_sin_map"][0], reduction="none")
+                * center_mask
+            ) / paddle.sum(center_mask)
+            loss_cos = paddle.sum(
+                F.smooth_l1_loss(pred_cos_map, gt["gt_cos_map"][0], reduction="none")
+                * center_mask
+            ) / paddle.sum(center_mask)
+        else:
+            loss_height = paddle.to_tensor(0.0)
+            loss_sin = paddle.to_tensor(0.0)
+            loss_cos = paddle.to_tensor(0.0)
+
+        loss_gcn = self.gcn_loss(gcn_data)
+
+        loss = loss_text + loss_center + loss_height + loss_sin + loss_cos + loss_gcn
+        results = dict(
+            loss=loss,
+            loss_text=loss_text,
+            loss_center=loss_center,
+            loss_height=loss_height,
+            loss_sin=loss_sin,
+            loss_cos=loss_cos,
+            loss_gcn=loss_gcn,
+        )
+
+        return results
diff --git a/ppocr/losses/det_east_loss.py b/ppocr/losses/det_east_loss.py
new file mode 100644
index 0000000..e5725d0
--- /dev/null
+++ b/ppocr/losses/det_east_loss.py
@@ -0,0 +1,62 @@
+# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+from .det_basic_loss import DiceLoss
+
+
+class EASTLoss(nn.Layer):
+    """ """
+
+    def __init__(self, eps=1e-6, **kwargs):
+        super(EASTLoss, self).__init__()
+        self.dice_loss = DiceLoss(eps=eps)
+
+    def forward(self, predicts, labels):
+        l_score, l_geo, l_mask = labels[1:]
+        f_score = predicts["f_score"]
+        f_geo = predicts["f_geo"]
+
+        dice_loss = self.dice_loss(f_score, l_score, l_mask)
+
+        # smoooth_l1_loss
+        channels = 8
+        l_geo_split = paddle.split(l_geo, num_or_sections=channels + 1, axis=1)
+        f_geo_split = paddle.split(f_geo, num_or_sections=channels, axis=1)
+        smooth_l1 = 0
+        for i in range(0, channels):
+            geo_diff = l_geo_split[i] - f_geo_split[i]
+            abs_geo_diff = paddle.abs(geo_diff)
+            smooth_l1_sign = paddle.less_than(abs_geo_diff, l_score)
+            smooth_l1_sign = paddle.cast(smooth_l1_sign, dtype="float32")
+            in_loss = abs_geo_diff * abs_geo_diff * smooth_l1_sign + (
+                abs_geo_diff - 0.5
+            ) * (1.0 - smooth_l1_sign)
+            out_loss = l_geo_split[-1] / channels * in_loss * l_score
+            smooth_l1 += out_loss
+        smooth_l1_loss = paddle.mean(smooth_l1 * l_score)
+
+        dice_loss = dice_loss * 0.01
+        total_loss = dice_loss + smooth_l1_loss
+        losses = {
+            "loss": total_loss,
+            "dice_loss": dice_loss,
+            "smooth_l1_loss": smooth_l1_loss,
+        }
+        return losses
diff --git a/ppocr/losses/det_fce_loss.py b/ppocr/losses/det_fce_loss.py
new file mode 100644
index 0000000..86b82a8
--- /dev/null
+++ b/ppocr/losses/det_fce_loss.py
@@ -0,0 +1,240 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textdet/losses/fce_loss.py
+"""
+
+import numpy as np
+from paddle import nn
+import paddle
+import paddle.nn.functional as F
+from functools import partial
+
+
+def multi_apply(func, *args, **kwargs):
+    pfunc = partial(func, **kwargs) if kwargs else func
+    map_results = map(pfunc, *args)
+    return tuple(map(list, zip(*map_results)))
+
+
+class FCELoss(nn.Layer):
+    """The class for implementing FCENet loss
+    FCENet(CVPR2021): Fourier Contour Embedding for Arbitrary-shaped
+        Text Detection
+
+    [https://arxiv.org/abs/2104.10442]
+
+    Args:
+        fourier_degree (int) : The maximum Fourier transform degree k.
+        num_sample (int) : The sampling points number of regression
+            loss. If it is too small, fcenet tends to be overfitting.
+        ohem_ratio (float): the negative/positive ratio in OHEM.
+    """
+
+    def __init__(self, fourier_degree, num_sample, ohem_ratio=3.0):
+        super().__init__()
+        self.fourier_degree = fourier_degree
+        self.num_sample = num_sample
+        self.ohem_ratio = ohem_ratio
+
+    def forward(self, preds, labels):
+        assert isinstance(preds, dict)
+        preds = preds["levels"]
+
+        p3_maps, p4_maps, p5_maps = labels[1:]
+        assert (
+            p3_maps[0].shape[0] == 4 * self.fourier_degree + 5
+        ), "fourier degree not equal in FCEhead and FCEtarget"
+
+        # to tensor
+        gts = [p3_maps, p4_maps, p5_maps]
+        for idx, maps in enumerate(gts):
+            gts[idx] = paddle.to_tensor(np.stack(maps))
+
+        losses = multi_apply(self.forward_single, preds, gts)
+
+        loss_tr = paddle.to_tensor(0.0).astype("float32")
+        loss_tcl = paddle.to_tensor(0.0).astype("float32")
+        loss_reg_x = paddle.to_tensor(0.0).astype("float32")
+        loss_reg_y = paddle.to_tensor(0.0).astype("float32")
+        loss_all = paddle.to_tensor(0.0).astype("float32")
+
+        for idx, loss in enumerate(losses):
+            loss_all += sum(loss)
+            if idx == 0:
+                loss_tr += sum(loss)
+            elif idx == 1:
+                loss_tcl += sum(loss)
+            elif idx == 2:
+                loss_reg_x += sum(loss)
+            else:
+                loss_reg_y += sum(loss)
+
+        results = dict(
+            loss=loss_all,
+            loss_text=loss_tr,
+            loss_center=loss_tcl,
+            loss_reg_x=loss_reg_x,
+            loss_reg_y=loss_reg_y,
+        )
+        return results
+
+    def forward_single(self, pred, gt):
+        cls_pred = paddle.transpose(pred[0], (0, 2, 3, 1))
+        reg_pred = paddle.transpose(pred[1], (0, 2, 3, 1))
+        gt = paddle.transpose(gt, (0, 2, 3, 1))
+
+        k = 2 * self.fourier_degree + 1
+        tr_pred = paddle.reshape(cls_pred[:, :, :, :2], (-1, 2))
+        tcl_pred = paddle.reshape(cls_pred[:, :, :, 2:], (-1, 2))
+        x_pred = paddle.reshape(reg_pred[:, :, :, 0:k], (-1, k))
+        y_pred = paddle.reshape(reg_pred[:, :, :, k : 2 * k], (-1, k))
+
+        tr_mask = gt[:, :, :, :1].reshape([-1])
+        tcl_mask = gt[:, :, :, 1:2].reshape([-1])
+        train_mask = gt[:, :, :, 2:3].reshape([-1])
+        x_map = paddle.reshape(gt[:, :, :, 3 : 3 + k], (-1, k))
+        y_map = paddle.reshape(gt[:, :, :, 3 + k :], (-1, k))
+
+        tr_train_mask = (train_mask * tr_mask).astype("bool")
+        tr_train_mask2 = paddle.concat(
+            [tr_train_mask.unsqueeze(1), tr_train_mask.unsqueeze(1)], axis=1
+        )
+        # tr loss
+        loss_tr = self.ohem(tr_pred, tr_mask, train_mask)
+        # tcl loss
+        loss_tcl = paddle.to_tensor(0.0).astype("float32")
+        tr_neg_mask = tr_train_mask.logical_not()
+        tr_neg_mask2 = paddle.concat(
+            [tr_neg_mask.unsqueeze(1), tr_neg_mask.unsqueeze(1)], axis=1
+        )
+        if tr_train_mask.sum().item() > 0:
+            loss_tcl_pos = F.cross_entropy(
+                tcl_pred.masked_select(tr_train_mask2).reshape([-1, 2]),
+                tcl_mask.masked_select(tr_train_mask).astype("int64"),
+            )
+            loss_tcl_neg = F.cross_entropy(
+                tcl_pred.masked_select(tr_neg_mask2).reshape([-1, 2]),
+                tcl_mask.masked_select(tr_neg_mask).astype("int64"),
+            )
+            loss_tcl = loss_tcl_pos + 0.5 * loss_tcl_neg
+
+        # regression loss
+        loss_reg_x = paddle.to_tensor(0.0).astype("float32")
+        loss_reg_y = paddle.to_tensor(0.0).astype("float32")
+        if tr_train_mask.sum().item() > 0:
+            weight = (
+                tr_mask.masked_select(tr_train_mask.astype("bool")).astype("float32")
+                + tcl_mask.masked_select(tr_train_mask.astype("bool")).astype("float32")
+            ) / 2
+            weight = weight.reshape([-1, 1])
+
+            ft_x, ft_y = self.fourier2poly(x_map, y_map)
+            ft_x_pre, ft_y_pre = self.fourier2poly(x_pred, y_pred)
+
+            dim = ft_x.shape[1]
+
+            tr_train_mask3 = paddle.concat(
+                [tr_train_mask.unsqueeze(1) for i in range(dim)], axis=1
+            )
+
+            loss_reg_x = paddle.mean(
+                weight
+                * F.smooth_l1_loss(
+                    ft_x_pre.masked_select(tr_train_mask3).reshape([-1, dim]),
+                    ft_x.masked_select(tr_train_mask3).reshape([-1, dim]),
+                    reduction="none",
+                )
+            )
+            loss_reg_y = paddle.mean(
+                weight
+                * F.smooth_l1_loss(
+                    ft_y_pre.masked_select(tr_train_mask3).reshape([-1, dim]),
+                    ft_y.masked_select(tr_train_mask3).reshape([-1, dim]),
+                    reduction="none",
+                )
+            )
+
+        return loss_tr, loss_tcl, loss_reg_x, loss_reg_y
+
+    def ohem(self, predict, target, train_mask):
+        pos = (target * train_mask).astype("bool")
+        neg = ((1 - target) * train_mask).astype("bool")
+
+        pos2 = paddle.concat([pos.unsqueeze(1), pos.unsqueeze(1)], axis=1)
+        neg2 = paddle.concat([neg.unsqueeze(1), neg.unsqueeze(1)], axis=1)
+
+        n_pos = pos.astype("float32").sum()
+
+        if n_pos.item() > 0:
+            loss_pos = F.cross_entropy(
+                predict.masked_select(pos2).reshape([-1, 2]),
+                target.masked_select(pos).astype("int64"),
+                reduction="sum",
+            )
+            loss_neg = F.cross_entropy(
+                predict.masked_select(neg2).reshape([-1, 2]),
+                target.masked_select(neg).astype("int64"),
+                reduction="none",
+            )
+            n_neg = min(
+                int(neg.astype("float32").sum().item()),
+                int(self.ohem_ratio * n_pos.astype("float32")),
+            )
+        else:
+            loss_pos = paddle.to_tensor(0.0)
+            loss_neg = F.cross_entropy(
+                predict.masked_select(neg2).reshape([-1, 2]),
+                target.masked_select(neg).astype("int64"),
+                reduction="none",
+            )
+            n_neg = 100
+        if len(loss_neg) > n_neg:
+            loss_neg, _ = paddle.topk(loss_neg, n_neg)
+
+        return (loss_pos + loss_neg.sum()) / (n_pos + n_neg).astype("float32")
+
+    def fourier2poly(self, real_maps, imag_maps):
+        """Transform Fourier coefficient maps to polygon maps.
+
+        Args:
+            real_maps (tensor): A map composed of the real parts of the
+                Fourier coefficients, whose shape is (-1, 2k+1)
+            imag_maps (tensor):A map composed of the imag parts of the
+                Fourier coefficients, whose shape is (-1, 2k+1)
+
+        Returns
+            x_maps (tensor): A map composed of the x value of the polygon
+                represented by n sample points (xn, yn), whose shape is (-1, n)
+            y_maps (tensor): A map composed of the y value of the polygon
+                represented by n sample points (xn, yn), whose shape is (-1, n)
+        """
+
+        k_vect = paddle.arange(
+            -self.fourier_degree, self.fourier_degree + 1, dtype="float32"
+        ).reshape([-1, 1])
+        i_vect = paddle.arange(0, self.num_sample, dtype="float32").reshape([1, -1])
+
+        transform_matrix = 2 * np.pi / self.num_sample * paddle.matmul(k_vect, i_vect)
+
+        x1 = paddle.einsum("ak, kn-> an", real_maps, paddle.cos(transform_matrix))
+        x2 = paddle.einsum("ak, kn-> an", imag_maps, paddle.sin(transform_matrix))
+        y1 = paddle.einsum("ak, kn-> an", real_maps, paddle.sin(transform_matrix))
+        y2 = paddle.einsum("ak, kn-> an", imag_maps, paddle.cos(transform_matrix))
+
+        x_maps = x1 - x2
+        y_maps = y1 + y2
+
+        return x_maps, y_maps
diff --git a/ppocr/losses/det_pse_loss.py b/ppocr/losses/det_pse_loss.py
new file mode 100644
index 0000000..962339b
--- /dev/null
+++ b/ppocr/losses/det_pse_loss.py
@@ -0,0 +1,158 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/whai362/PSENet/blob/python3/models/head/psenet_head.py
+"""
+
+import paddle
+from paddle import nn
+from paddle.nn import functional as F
+import numpy as np
+from ppocr.utils.iou import iou
+
+
+class PSELoss(nn.Layer):
+    def __init__(
+        self,
+        alpha,
+        ohem_ratio=3,
+        kernel_sample_mask="pred",
+        reduction="sum",
+        eps=1e-6,
+        **kwargs,
+    ):
+        """Implement PSE Loss."""
+        super(PSELoss, self).__init__()
+        assert reduction in ["sum", "mean", "none"]
+        self.alpha = alpha
+        self.ohem_ratio = ohem_ratio
+        self.kernel_sample_mask = kernel_sample_mask
+        self.reduction = reduction
+        self.eps = eps
+
+    def forward(self, outputs, labels):
+        predicts = outputs["maps"]
+        predicts = F.interpolate(predicts, scale_factor=4)
+
+        texts = predicts[:, 0, :, :]
+        kernels = predicts[:, 1:, :, :]
+        gt_texts, gt_kernels, training_masks = labels[1:]
+
+        # text loss
+        selected_masks = self.ohem_batch(texts, gt_texts, training_masks)
+
+        loss_text = self.dice_loss(texts, gt_texts, selected_masks)
+        iou_text = iou(
+            (texts > 0).astype("int64"), gt_texts, training_masks, reduce=False
+        )
+        losses = dict(loss_text=loss_text, iou_text=iou_text)
+
+        # kernel loss
+        loss_kernels = []
+        if self.kernel_sample_mask == "gt":
+            selected_masks = gt_texts * training_masks
+        elif self.kernel_sample_mask == "pred":
+            selected_masks = (F.sigmoid(texts) > 0.5).astype("float32") * training_masks
+
+        for i in range(kernels.shape[1]):
+            kernel_i = kernels[:, i, :, :]
+            gt_kernel_i = gt_kernels[:, i, :, :]
+            loss_kernel_i = self.dice_loss(kernel_i, gt_kernel_i, selected_masks)
+            loss_kernels.append(loss_kernel_i)
+        loss_kernels = paddle.mean(paddle.stack(loss_kernels, axis=1), axis=1)
+        iou_kernel = iou(
+            (kernels[:, -1, :, :] > 0).astype("int64"),
+            gt_kernels[:, -1, :, :],
+            training_masks * gt_texts,
+            reduce=False,
+        )
+        losses.update(dict(loss_kernels=loss_kernels, iou_kernel=iou_kernel))
+        loss = self.alpha * loss_text + (1 - self.alpha) * loss_kernels
+        losses["loss"] = loss
+        if self.reduction == "sum":
+            losses = {x: paddle.sum(v) for x, v in losses.items()}
+        elif self.reduction == "mean":
+            losses = {x: paddle.mean(v) for x, v in losses.items()}
+        return losses
+
+    def dice_loss(self, input, target, mask):
+        input = F.sigmoid(input)
+
+        input = input.reshape([input.shape[0], -1])
+        target = target.reshape([target.shape[0], -1])
+        mask = mask.reshape([mask.shape[0], -1])
+
+        input = input * mask
+        target = target * mask
+
+        a = paddle.sum(input * target, 1)
+        b = paddle.sum(input * input, 1) + self.eps
+        c = paddle.sum(target * target, 1) + self.eps
+        d = (2 * a) / (b + c)
+        return 1 - d
+
+    def ohem_single(self, score, gt_text, training_mask, ohem_ratio=3):
+        pos_num = int(paddle.sum((gt_text > 0.5).astype("float32"))) - int(
+            paddle.sum(
+                paddle.logical_and((gt_text > 0.5), (training_mask <= 0.5)).astype(
+                    "float32"
+                )
+            )
+        )
+
+        if pos_num == 0:
+            selected_mask = training_mask
+            selected_mask = selected_mask.reshape(
+                [1, selected_mask.shape[0], selected_mask.shape[1]]
+            ).astype("float32")
+            return selected_mask
+
+        neg_num = int(paddle.sum((gt_text <= 0.5).astype("float32")))
+        neg_num = int(min(pos_num * ohem_ratio, neg_num))
+
+        if neg_num == 0:
+            selected_mask = training_mask
+            selected_mask = selected_mask.reshape(
+                [1, selected_mask.shape[0], selected_mask.shape[1]]
+            ).astype("float32")
+            return selected_mask
+
+        neg_score = paddle.masked_select(score, gt_text <= 0.5)
+        neg_score_sorted = paddle.sort(-neg_score)
+        threshold = -neg_score_sorted[neg_num - 1]
+
+        selected_mask = paddle.logical_and(
+            paddle.logical_or((score >= threshold), (gt_text > 0.5)),
+            (training_mask > 0.5),
+        )
+        selected_mask = selected_mask.reshape(
+            [1, selected_mask.shape[0], selected_mask.shape[1]]
+        ).astype("float32")
+        return selected_mask
+
+    def ohem_batch(self, scores, gt_texts, training_masks, ohem_ratio=3):
+        selected_masks = []
+        for i in range(scores.shape[0]):
+            selected_masks.append(
+                self.ohem_single(
+                    scores[i, :, :],
+                    gt_texts[i, :, :],
+                    training_masks[i, :, :],
+                    ohem_ratio,
+                )
+            )
+
+        selected_masks = paddle.concat(selected_masks, 0).astype("float32")
+        return selected_masks
diff --git a/ppocr/losses/det_sast_loss.py b/ppocr/losses/det_sast_loss.py
new file mode 100644
index 0000000..160eb06
--- /dev/null
+++ b/ppocr/losses/det_sast_loss.py
@@ -0,0 +1,133 @@
+# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+from .det_basic_loss import DiceLoss
+import numpy as np
+
+
+class SASTLoss(nn.Layer):
+    """ """
+
+    def __init__(self, eps=1e-6, **kwargs):
+        super(SASTLoss, self).__init__()
+        self.dice_loss = DiceLoss(eps=eps)
+
+    def forward(self, predicts, labels):
+        """
+        tcl_pos: N x 128 x 3
+        tcl_mask: N x 128 x 1
+        tcl_label: N x X list or LoDTensor
+        """
+
+        f_score = predicts["f_score"]
+        f_border = predicts["f_border"]
+        f_tvo = predicts["f_tvo"]
+        f_tco = predicts["f_tco"]
+
+        l_score, l_border, l_mask, l_tvo, l_tco = labels[1:]
+
+        # score_loss
+        intersection = paddle.sum(f_score * l_score * l_mask)
+        union = paddle.sum(f_score * l_mask) + paddle.sum(l_score * l_mask)
+        score_loss = 1.0 - 2 * intersection / (union + 1e-5)
+
+        # border loss
+        l_border_split, l_border_norm = paddle.split(
+            l_border, num_or_sections=[4, 1], axis=1
+        )
+        f_border_split = f_border
+        border_ex_shape = l_border_norm.shape * np.array([1, 4, 1, 1])
+        l_border_norm_split = paddle.expand(x=l_border_norm, shape=border_ex_shape)
+        l_border_score = paddle.expand(x=l_score, shape=border_ex_shape)
+        l_border_mask = paddle.expand(x=l_mask, shape=border_ex_shape)
+
+        border_diff = l_border_split - f_border_split
+        abs_border_diff = paddle.abs(border_diff)
+        border_sign = abs_border_diff < 1.0
+        border_sign = paddle.cast(border_sign, dtype="float32")
+        border_sign.stop_gradient = True
+        border_in_loss = 0.5 * abs_border_diff * abs_border_diff * border_sign + (
+            abs_border_diff - 0.5
+        ) * (1.0 - border_sign)
+        border_out_loss = l_border_norm_split * border_in_loss
+        border_loss = paddle.sum(border_out_loss * l_border_score * l_border_mask) / (
+            paddle.sum(l_border_score * l_border_mask) + 1e-5
+        )
+
+        # tvo_loss
+        l_tvo_split, l_tvo_norm = paddle.split(l_tvo, num_or_sections=[8, 1], axis=1)
+        f_tvo_split = f_tvo
+        tvo_ex_shape = l_tvo_norm.shape * np.array([1, 8, 1, 1])
+        l_tvo_norm_split = paddle.expand(x=l_tvo_norm, shape=tvo_ex_shape)
+        l_tvo_score = paddle.expand(x=l_score, shape=tvo_ex_shape)
+        l_tvo_mask = paddle.expand(x=l_mask, shape=tvo_ex_shape)
+        #
+        tvo_geo_diff = l_tvo_split - f_tvo_split
+        abs_tvo_geo_diff = paddle.abs(tvo_geo_diff)
+        tvo_sign = abs_tvo_geo_diff < 1.0
+        tvo_sign = paddle.cast(tvo_sign, dtype="float32")
+        tvo_sign.stop_gradient = True
+        tvo_in_loss = 0.5 * abs_tvo_geo_diff * abs_tvo_geo_diff * tvo_sign + (
+            abs_tvo_geo_diff - 0.5
+        ) * (1.0 - tvo_sign)
+        tvo_out_loss = l_tvo_norm_split * tvo_in_loss
+        tvo_loss = paddle.sum(tvo_out_loss * l_tvo_score * l_tvo_mask) / (
+            paddle.sum(l_tvo_score * l_tvo_mask) + 1e-5
+        )
+
+        # tco_loss
+        l_tco_split, l_tco_norm = paddle.split(l_tco, num_or_sections=[2, 1], axis=1)
+        f_tco_split = f_tco
+        tco_ex_shape = l_tco_norm.shape * np.array([1, 2, 1, 1])
+        l_tco_norm_split = paddle.expand(x=l_tco_norm, shape=tco_ex_shape)
+        l_tco_score = paddle.expand(x=l_score, shape=tco_ex_shape)
+        l_tco_mask = paddle.expand(x=l_mask, shape=tco_ex_shape)
+
+        tco_geo_diff = l_tco_split - f_tco_split
+        abs_tco_geo_diff = paddle.abs(tco_geo_diff)
+        tco_sign = abs_tco_geo_diff < 1.0
+        tco_sign = paddle.cast(tco_sign, dtype="float32")
+        tco_sign.stop_gradient = True
+        tco_in_loss = 0.5 * abs_tco_geo_diff * abs_tco_geo_diff * tco_sign + (
+            abs_tco_geo_diff - 0.5
+        ) * (1.0 - tco_sign)
+        tco_out_loss = l_tco_norm_split * tco_in_loss
+        tco_loss = paddle.sum(tco_out_loss * l_tco_score * l_tco_mask) / (
+            paddle.sum(l_tco_score * l_tco_mask) + 1e-5
+        )
+
+        # total loss
+        tvo_lw, tco_lw = 1.5, 1.5
+        score_lw, border_lw = 1.0, 1.0
+        total_loss = (
+            score_loss * score_lw
+            + border_loss * border_lw
+            + tvo_loss * tvo_lw
+            + tco_loss * tco_lw
+        )
+
+        losses = {
+            "loss": total_loss,
+            "score_loss": score_loss,
+            "border_loss": border_loss,
+            "tvo_loss": tvo_loss,
+            "tco_loss": tco_loss,
+        }
+        return losses
diff --git a/ppocr/losses/distillation_loss.py b/ppocr/losses/distillation_loss.py
new file mode 100644
index 0000000..5b69e1d
--- /dev/null
+++ b/ppocr/losses/distillation_loss.py
@@ -0,0 +1,1192 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+import numpy as np
+import cv2
+
+from .rec_ctc_loss import CTCLoss
+from .rec_sar_loss import SARLoss
+from .rec_ce_loss import CELoss
+from .basic_loss import DMLLoss, KLDivLoss, DKDLoss
+from .basic_loss import DistanceLoss
+from .basic_loss import LossFromOutput
+from .det_db_loss import DBLoss
+from .det_basic_loss import BalanceLoss, MaskL1Loss, DiceLoss
+from .vqa_token_layoutlm_loss import VQASerTokenLayoutLMLoss
+
+
+def _sum_loss(loss_dict):
+    if "loss" in loss_dict.keys():
+        return loss_dict
+    else:
+        loss_dict["loss"] = 0.0
+        for k, value in loss_dict.items():
+            if k == "loss":
+                continue
+            else:
+                loss_dict["loss"] += value
+        return loss_dict
+
+
+class DistillationDMLLoss(DMLLoss):
+    """ """
+
+    def __init__(
+        self,
+        model_name_pairs=[],
+        act=None,
+        use_log=False,
+        key=None,
+        multi_head=False,
+        dis_head="ctc",
+        maps_name=None,
+        name="dml",
+    ):
+        super().__init__(act=act, use_log=use_log)
+        assert isinstance(model_name_pairs, list)
+        self.key = key
+        self.multi_head = multi_head
+        self.dis_head = dis_head
+        self.model_name_pairs = self._check_model_name_pairs(model_name_pairs)
+        self.name = name
+        self.maps_name = self._check_maps_name(maps_name)
+
+    def _check_model_name_pairs(self, model_name_pairs):
+        if not isinstance(model_name_pairs, list):
+            return []
+        elif isinstance(model_name_pairs[0], list) and isinstance(
+            model_name_pairs[0][0], str
+        ):
+            return model_name_pairs
+        else:
+            return [model_name_pairs]
+
+    def _check_maps_name(self, maps_name):
+        if maps_name is None:
+            return None
+        elif isinstance(maps_name, str):
+            return [maps_name]
+        elif isinstance(maps_name, list):
+            return [maps_name]
+        else:
+            return None
+
+    def _slice_out(self, outs):
+        new_outs = {}
+        for k in self.maps_name:
+            if k == "thrink_maps":
+                new_outs[k] = outs[:, 0, :, :]
+            elif k == "threshold_maps":
+                new_outs[k] = outs[:, 1, :, :]
+            elif k == "binary_maps":
+                new_outs[k] = outs[:, 2, :, :]
+            else:
+                continue
+        return new_outs
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, pair in enumerate(self.model_name_pairs):
+            out1 = predicts[pair[0]]
+            out2 = predicts[pair[1]]
+            if self.key is not None:
+                out1 = out1[self.key]
+                out2 = out2[self.key]
+            if self.maps_name is None:
+                if self.multi_head:
+                    loss = super().forward(out1[self.dis_head], out2[self.dis_head])
+                else:
+                    loss = super().forward(out1, out2)
+                if isinstance(loss, dict):
+                    for key in loss:
+                        loss_dict["{}_{}_{}_{}".format(key, pair[0], pair[1], idx)] = (
+                            loss[key]
+                        )
+                else:
+                    loss_dict["{}_{}".format(self.name, idx)] = loss
+            else:
+                outs1 = self._slice_out(out1)
+                outs2 = self._slice_out(out2)
+                for _c, k in enumerate(outs1.keys()):
+                    loss = super().forward(outs1[k], outs2[k])
+                    if isinstance(loss, dict):
+                        for key in loss:
+                            loss_dict[
+                                "{}_{}_{}_{}_{}".format(
+                                    key, pair[0], pair[1], self.maps_name, idx
+                                )
+                            ] = loss[key]
+                    else:
+                        loss_dict[
+                            "{}_{}_{}".format(self.name, self.maps_name[_c], idx)
+                        ] = loss
+
+        loss_dict = _sum_loss(loss_dict)
+
+        return loss_dict
+
+
+class DistillationKLDivLoss(KLDivLoss):
+    """ """
+
+    def __init__(
+        self,
+        model_name_pairs=[],
+        key=None,
+        multi_head=False,
+        dis_head="ctc",
+        maps_name=None,
+        name="kl_div",
+    ):
+        super().__init__()
+        assert isinstance(model_name_pairs, list)
+        self.key = key
+        self.multi_head = multi_head
+        self.dis_head = dis_head
+        self.model_name_pairs = self._check_model_name_pairs(model_name_pairs)
+        self.name = name
+        self.maps_name = self._check_maps_name(maps_name)
+
+    def _check_model_name_pairs(self, model_name_pairs):
+        if not isinstance(model_name_pairs, list):
+            return []
+        elif isinstance(model_name_pairs[0], list) and isinstance(
+            model_name_pairs[0][0], str
+        ):
+            return model_name_pairs
+        else:
+            return [model_name_pairs]
+
+    def _check_maps_name(self, maps_name):
+        if maps_name is None:
+            return None
+        elif isinstance(maps_name, str):
+            return [maps_name]
+        elif isinstance(maps_name, list):
+            return [maps_name]
+        else:
+            return None
+
+    def _slice_out(self, outs):
+        new_outs = {}
+        for k in self.maps_name:
+            if k == "thrink_maps":
+                new_outs[k] = outs[:, 0, :, :]
+            elif k == "threshold_maps":
+                new_outs[k] = outs[:, 1, :, :]
+            elif k == "binary_maps":
+                new_outs[k] = outs[:, 2, :, :]
+            else:
+                continue
+        return new_outs
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, pair in enumerate(self.model_name_pairs):
+            out1 = predicts[pair[0]]
+            out2 = predicts[pair[1]]
+            if self.key is not None:
+                out1 = out1[self.key]
+                out2 = out2[self.key]
+            if self.maps_name is None:
+                if self.multi_head:
+                    # for nrtr dml loss
+                    max_len = batch[3].max()
+                    tgt = batch[2][:, 1 : 2 + max_len]
+                    tgt = tgt.reshape([-1])
+                    non_pad_mask = paddle.not_equal(
+                        tgt, paddle.zeros(tgt.shape, dtype=tgt.dtype)
+                    )
+                    loss = super().forward(
+                        out1[self.dis_head], out2[self.dis_head], non_pad_mask
+                    )
+                else:
+                    loss = super().forward(out1, out2)
+                if isinstance(loss, dict):
+                    for key in loss:
+                        loss_dict["{}_{}_{}_{}".format(key, pair[0], pair[1], idx)] = (
+                            loss[key]
+                        )
+                else:
+                    loss_dict["{}_{}".format(self.name, idx)] = loss
+            else:
+                outs1 = self._slice_out(out1)
+                outs2 = self._slice_out(out2)
+                for _c, k in enumerate(outs1.keys()):
+                    loss = super().forward(outs1[k], outs2[k])
+                    if isinstance(loss, dict):
+                        for key in loss:
+                            loss_dict[
+                                "{}_{}_{}_{}_{}".format(
+                                    key, pair[0], pair[1], self.maps_name, idx
+                                )
+                            ] = loss[key]
+                    else:
+                        loss_dict[
+                            "{}_{}_{}".format(self.name, self.maps_name[_c], idx)
+                        ] = loss
+
+        loss_dict = _sum_loss(loss_dict)
+
+        return loss_dict
+
+
+class DistillationDKDLoss(DKDLoss):
+    """ """
+
+    def __init__(
+        self,
+        model_name_pairs=[],
+        key=None,
+        multi_head=False,
+        dis_head="ctc",
+        maps_name=None,
+        name="dkd",
+        temperature=1.0,
+        alpha=1.0,
+        beta=1.0,
+    ):
+        super().__init__(temperature, alpha, beta)
+        assert isinstance(model_name_pairs, list)
+        self.key = key
+        self.multi_head = multi_head
+        self.dis_head = dis_head
+        self.model_name_pairs = self._check_model_name_pairs(model_name_pairs)
+        self.name = name
+        self.maps_name = self._check_maps_name(maps_name)
+
+    def _check_model_name_pairs(self, model_name_pairs):
+        if not isinstance(model_name_pairs, list):
+            return []
+        elif isinstance(model_name_pairs[0], list) and isinstance(
+            model_name_pairs[0][0], str
+        ):
+            return model_name_pairs
+        else:
+            return [model_name_pairs]
+
+    def _check_maps_name(self, maps_name):
+        if maps_name is None:
+            return None
+        elif isinstance(maps_name, str):
+            return [maps_name]
+        elif isinstance(maps_name, list):
+            return [maps_name]
+        else:
+            return None
+
+    def _slice_out(self, outs):
+        new_outs = {}
+        for k in self.maps_name:
+            if k == "thrink_maps":
+                new_outs[k] = outs[:, 0, :, :]
+            elif k == "threshold_maps":
+                new_outs[k] = outs[:, 1, :, :]
+            elif k == "binary_maps":
+                new_outs[k] = outs[:, 2, :, :]
+            else:
+                continue
+        return new_outs
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+
+        for idx, pair in enumerate(self.model_name_pairs):
+            out1 = predicts[pair[0]]
+            out2 = predicts[pair[1]]
+            if self.key is not None:
+                out1 = out1[self.key]
+                out2 = out2[self.key]
+            if self.maps_name is None:
+                if self.multi_head:
+                    # for nrtr dml loss
+                    max_len = batch[3].max()
+                    tgt = batch[2][:, 1 : 2 + max_len]  # [batch_size, max_len + 1]
+
+                    tgt = tgt.reshape([-1])  # batch_size * (max_len + 1)
+                    non_pad_mask = paddle.not_equal(
+                        tgt, paddle.zeros(tgt.shape, dtype=tgt.dtype)
+                    )  # batch_size * (max_len + 1)
+
+                    loss = super().forward(
+                        out1[self.dis_head], out2[self.dis_head], tgt, non_pad_mask
+                    )  # [batch_size, max_len + 1, num_char]
+                else:
+                    loss = super().forward(out1, out2)
+                if isinstance(loss, dict):
+                    for key in loss:
+                        loss_dict["{}_{}_{}_{}".format(key, pair[0], pair[1], idx)] = (
+                            loss[key]
+                        )
+                else:
+                    loss_dict["{}_{}".format(self.name, idx)] = loss
+            else:
+                outs1 = self._slice_out(out1)
+                outs2 = self._slice_out(out2)
+                for _c, k in enumerate(outs1.keys()):
+                    loss = super().forward(outs1[k], outs2[k])
+                    if isinstance(loss, dict):
+                        for key in loss:
+                            loss_dict[
+                                "{}_{}_{}_{}_{}".format(
+                                    key, pair[0], pair[1], self.maps_name, idx
+                                )
+                            ] = loss[key]
+                    else:
+                        loss_dict[
+                            "{}_{}_{}".format(self.name, self.maps_name[_c], idx)
+                        ] = loss
+
+        loss_dict = _sum_loss(loss_dict)
+
+        return loss_dict
+
+
+class DistillationNRTRDMLLoss(DistillationDMLLoss):
+    """ """
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, pair in enumerate(self.model_name_pairs):
+            out1 = predicts[pair[0]]
+            out2 = predicts[pair[1]]
+            if self.key is not None:
+                out1 = out1[self.key]
+                out2 = out2[self.key]
+
+            if self.multi_head:
+                # for nrtr dml loss
+                max_len = batch[3].max()
+                tgt = batch[2][:, 1 : 2 + max_len]
+                tgt = tgt.reshape([-1])
+                non_pad_mask = paddle.not_equal(
+                    tgt, paddle.zeros(tgt.shape, dtype=tgt.dtype)
+                )
+                loss = super().forward(
+                    out1[self.dis_head], out2[self.dis_head], non_pad_mask
+                )
+            else:
+                loss = super().forward(out1, out2)
+            if isinstance(loss, dict):
+                for key in loss:
+                    loss_dict["{}_{}_{}_{}".format(key, pair[0], pair[1], idx)] = loss[
+                        key
+                    ]
+            else:
+                loss_dict["{}_{}".format(self.name, idx)] = loss
+
+        loss_dict = _sum_loss(loss_dict)
+
+        return loss_dict
+
+
+class DistillationKLDivLoss(KLDivLoss):
+    """ """
+
+    def __init__(
+        self,
+        model_name_pairs=[],
+        key=None,
+        multi_head=False,
+        dis_head="ctc",
+        maps_name=None,
+        name="kl_div",
+    ):
+        super().__init__()
+        assert isinstance(model_name_pairs, list)
+        self.key = key
+        self.multi_head = multi_head
+        self.dis_head = dis_head
+        self.model_name_pairs = self._check_model_name_pairs(model_name_pairs)
+        self.name = name
+        self.maps_name = self._check_maps_name(maps_name)
+
+    def _check_model_name_pairs(self, model_name_pairs):
+        if not isinstance(model_name_pairs, list):
+            return []
+        elif isinstance(model_name_pairs[0], list) and isinstance(
+            model_name_pairs[0][0], str
+        ):
+            return model_name_pairs
+        else:
+            return [model_name_pairs]
+
+    def _check_maps_name(self, maps_name):
+        if maps_name is None:
+            return None
+        elif isinstance(maps_name, str):
+            return [maps_name]
+        elif isinstance(maps_name, list):
+            return [maps_name]
+        else:
+            return None
+
+    def _slice_out(self, outs):
+        new_outs = {}
+        for k in self.maps_name:
+            if k == "thrink_maps":
+                new_outs[k] = outs[:, 0, :, :]
+            elif k == "threshold_maps":
+                new_outs[k] = outs[:, 1, :, :]
+            elif k == "binary_maps":
+                new_outs[k] = outs[:, 2, :, :]
+            else:
+                continue
+        return new_outs
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, pair in enumerate(self.model_name_pairs):
+            out1 = predicts[pair[0]]
+            out2 = predicts[pair[1]]
+            if self.key is not None:
+                out1 = out1[self.key]
+                out2 = out2[self.key]
+            if self.maps_name is None:
+                if self.multi_head:
+                    # for nrtr dml loss
+                    max_len = batch[3].max()
+                    tgt = batch[2][:, 1 : 2 + max_len]
+                    tgt = tgt.reshape([-1])
+                    non_pad_mask = paddle.not_equal(
+                        tgt, paddle.zeros(tgt.shape, dtype=tgt.dtype)
+                    )
+                    loss = super().forward(
+                        out1[self.dis_head], out2[self.dis_head], non_pad_mask
+                    )
+                else:
+                    loss = super().forward(out1, out2)
+                if isinstance(loss, dict):
+                    for key in loss:
+                        loss_dict["{}_{}_{}_{}".format(key, pair[0], pair[1], idx)] = (
+                            loss[key]
+                        )
+                else:
+                    loss_dict["{}_{}".format(self.name, idx)] = loss
+            else:
+                outs1 = self._slice_out(out1)
+                outs2 = self._slice_out(out2)
+                for _c, k in enumerate(outs1.keys()):
+                    loss = super().forward(outs1[k], outs2[k])
+                    if isinstance(loss, dict):
+                        for key in loss:
+                            loss_dict[
+                                "{}_{}_{}_{}_{}".format(
+                                    key, pair[0], pair[1], self.maps_name, idx
+                                )
+                            ] = loss[key]
+                    else:
+                        loss_dict[
+                            "{}_{}_{}".format(self.name, self.maps_name[_c], idx)
+                        ] = loss
+
+        loss_dict = _sum_loss(loss_dict)
+
+        return loss_dict
+
+
+class DistillationDKDLoss(DKDLoss):
+    """ """
+
+    def __init__(
+        self,
+        model_name_pairs=[],
+        key=None,
+        multi_head=False,
+        dis_head="ctc",
+        maps_name=None,
+        name="dkd",
+        temperature=1.0,
+        alpha=1.0,
+        beta=1.0,
+    ):
+        super().__init__(temperature, alpha, beta)
+        assert isinstance(model_name_pairs, list)
+        self.key = key
+        self.multi_head = multi_head
+        self.dis_head = dis_head
+        self.model_name_pairs = self._check_model_name_pairs(model_name_pairs)
+        self.name = name
+        self.maps_name = self._check_maps_name(maps_name)
+
+    def _check_model_name_pairs(self, model_name_pairs):
+        if not isinstance(model_name_pairs, list):
+            return []
+        elif isinstance(model_name_pairs[0], list) and isinstance(
+            model_name_pairs[0][0], str
+        ):
+            return model_name_pairs
+        else:
+            return [model_name_pairs]
+
+    def _check_maps_name(self, maps_name):
+        if maps_name is None:
+            return None
+        elif isinstance(maps_name, str):
+            return [maps_name]
+        elif isinstance(maps_name, list):
+            return [maps_name]
+        else:
+            return None
+
+    def _slice_out(self, outs):
+        new_outs = {}
+        for k in self.maps_name:
+            if k == "thrink_maps":
+                new_outs[k] = outs[:, 0, :, :]
+            elif k == "threshold_maps":
+                new_outs[k] = outs[:, 1, :, :]
+            elif k == "binary_maps":
+                new_outs[k] = outs[:, 2, :, :]
+            else:
+                continue
+        return new_outs
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+
+        for idx, pair in enumerate(self.model_name_pairs):
+            out1 = predicts[pair[0]]
+            out2 = predicts[pair[1]]
+            if self.key is not None:
+                out1 = out1[self.key]
+                out2 = out2[self.key]
+            if self.maps_name is None:
+                if self.multi_head:
+                    # for nrtr dml loss
+                    max_len = batch[3].max()
+                    tgt = batch[2][:, 1 : 2 + max_len]  # [batch_size, max_len + 1]
+
+                    tgt = tgt.reshape([-1])  # batch_size * (max_len + 1)
+                    non_pad_mask = paddle.not_equal(
+                        tgt, paddle.zeros(tgt.shape, dtype=tgt.dtype)
+                    )  # batch_size * (max_len + 1)
+
+                    loss = super().forward(
+                        out1[self.dis_head], out2[self.dis_head], tgt, non_pad_mask
+                    )  # [batch_size, max_len + 1, num_char]
+                else:
+                    loss = super().forward(out1, out2)
+                if isinstance(loss, dict):
+                    for key in loss:
+                        loss_dict["{}_{}_{}_{}".format(key, pair[0], pair[1], idx)] = (
+                            loss[key]
+                        )
+                else:
+                    loss_dict["{}_{}".format(self.name, idx)] = loss
+            else:
+                outs1 = self._slice_out(out1)
+                outs2 = self._slice_out(out2)
+                for _c, k in enumerate(outs1.keys()):
+                    loss = super().forward(outs1[k], outs2[k])
+                    if isinstance(loss, dict):
+                        for key in loss:
+                            loss_dict[
+                                "{}_{}_{}_{}_{}".format(
+                                    key, pair[0], pair[1], self.maps_name, idx
+                                )
+                            ] = loss[key]
+                    else:
+                        loss_dict[
+                            "{}_{}_{}".format(self.name, self.maps_name[_c], idx)
+                        ] = loss
+
+        loss_dict = _sum_loss(loss_dict)
+
+        return loss_dict
+
+
+class DistillationCTCLoss(CTCLoss):
+    def __init__(self, model_name_list=[], key=None, multi_head=False, name="loss_ctc"):
+        super().__init__()
+        self.model_name_list = model_name_list
+        self.key = key
+        self.name = name
+        self.multi_head = multi_head
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, model_name in enumerate(self.model_name_list):
+            out = predicts[model_name]
+            if self.key is not None:
+                out = out[self.key]
+            if self.multi_head:
+                assert "ctc" in out, "multi head has multi out"
+                loss = super().forward(out["ctc"], batch[:2] + batch[3:])
+            else:
+                loss = super().forward(out, batch)
+            if isinstance(loss, dict):
+                for key in loss:
+                    loss_dict["{}_{}_{}".format(self.name, model_name, idx)] = loss[key]
+            else:
+                loss_dict["{}_{}".format(self.name, model_name)] = loss
+        return loss_dict
+
+
+class DistillationSARLoss(SARLoss):
+    def __init__(
+        self, model_name_list=[], key=None, multi_head=False, name="loss_sar", **kwargs
+    ):
+        ignore_index = kwargs.get("ignore_index", 92)
+        super().__init__(ignore_index=ignore_index)
+        self.model_name_list = model_name_list
+        self.key = key
+        self.name = name
+        self.multi_head = multi_head
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, model_name in enumerate(self.model_name_list):
+            out = predicts[model_name]
+            if self.key is not None:
+                out = out[self.key]
+            if self.multi_head:
+                assert "sar" in out, "multi head has multi out"
+                loss = super().forward(out["sar"], batch[:1] + batch[2:])
+            else:
+                loss = super().forward(out, batch)
+            if isinstance(loss, dict):
+                for key in loss:
+                    loss_dict["{}_{}_{}".format(self.name, model_name, idx)] = loss[key]
+            else:
+                loss_dict["{}_{}".format(self.name, model_name)] = loss
+        return loss_dict
+
+
+class DistillationNRTRLoss(CELoss):
+    def __init__(
+        self,
+        model_name_list=[],
+        key=None,
+        multi_head=False,
+        smoothing=True,
+        name="loss_nrtr",
+        **kwargs,
+    ):
+        super().__init__(smoothing=smoothing)
+        self.model_name_list = model_name_list
+        self.key = key
+        self.name = name
+        self.multi_head = multi_head
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, model_name in enumerate(self.model_name_list):
+            out = predicts[model_name]
+            if self.key is not None:
+                out = out[self.key]
+            if self.multi_head:
+                assert "gtc" in out, "multi head has multi out"
+                loss = super().forward(out["gtc"], batch[:1] + batch[2:])
+            else:
+                loss = super().forward(out, batch)
+            if isinstance(loss, dict):
+                for key in loss:
+                    loss_dict["{}_{}_{}".format(self.name, model_name, idx)] = loss[key]
+            else:
+                loss_dict["{}_{}".format(self.name, model_name)] = loss
+        return loss_dict
+
+
+class DistillationDBLoss(DBLoss):
+    def __init__(
+        self,
+        model_name_list=[],
+        balance_loss=True,
+        main_loss_type="DiceLoss",
+        alpha=5,
+        beta=10,
+        ohem_ratio=3,
+        eps=1e-6,
+        name="db",
+        **kwargs,
+    ):
+        super().__init__()
+        self.model_name_list = model_name_list
+        self.name = name
+        self.key = None
+
+    def forward(self, predicts, batch):
+        loss_dict = {}
+        for idx, model_name in enumerate(self.model_name_list):
+            out = predicts[model_name]
+            if self.key is not None:
+                out = out[self.key]
+            loss = super().forward(out, batch)
+
+            if isinstance(loss, dict):
+                for key in loss.keys():
+                    if key == "loss":
+                        continue
+                    name = "{}_{}_{}".format(self.name, model_name, key)
+                    loss_dict[name] = loss[key]
+            else:
+                loss_dict["{}_{}".format(self.name, model_name)] = loss
+
+        loss_dict = _sum_loss(loss_dict)
+        return loss_dict
+
+
+class DistillationDilaDBLoss(DBLoss):
+    def __init__(
+        self,
+        model_name_pairs=[],
+        key=None,
+        balance_loss=True,
+        main_loss_type="DiceLoss",
+        alpha=5,
+        beta=10,
+        ohem_ratio=3,
+        eps=1e-6,
+        name="dila_dbloss",
+    ):
+        super().__init__()
+        self.model_name_pairs = model_name_pairs
+        self.name = name
+        self.key = key
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, pair in enumerate(self.model_name_pairs):
+            stu_outs = predicts[pair[0]]
+            tch_outs = predicts[pair[1]]
+            if self.key is not None:
+                stu_preds = stu_outs[self.key]
+                tch_preds = tch_outs[self.key]
+
+            stu_shrink_maps = stu_preds[:, 0, :, :]
+            stu_binary_maps = stu_preds[:, 2, :, :]
+
+            # dilation to teacher prediction
+            dilation_w = np.array([[1, 1], [1, 1]])
+            th_shrink_maps = tch_preds[:, 0, :, :]
+            if hasattr(paddle.Tensor, "contiguous"):
+                th_shrink_maps = th_shrink_maps.contiguous()
+            th_shrink_maps = th_shrink_maps.numpy() > 0.3  # thresh = 0.3
+            dilate_maps = np.zeros_like(th_shrink_maps).astype(np.float32)
+            for i in range(th_shrink_maps.shape[0]):
+                dilate_maps[i] = cv2.dilate(
+                    th_shrink_maps[i, :, :].astype(np.uint8), dilation_w
+                )
+            th_shrink_maps = paddle.to_tensor(dilate_maps)
+
+            (
+                label_threshold_map,
+                label_threshold_mask,
+                label_shrink_map,
+                label_shrink_mask,
+            ) = batch[1:]
+
+            # calculate the shrink map loss
+            bce_loss = self.alpha * self.bce_loss(
+                stu_shrink_maps, th_shrink_maps, label_shrink_mask
+            )
+            loss_binary_maps = self.dice_loss(
+                stu_binary_maps, th_shrink_maps, label_shrink_mask
+            )
+
+            # k = f"{self.name}_{pair[0]}_{pair[1]}"
+            k = "{}_{}_{}".format(self.name, pair[0], pair[1])
+            loss_dict[k] = bce_loss + loss_binary_maps
+
+        loss_dict = _sum_loss(loss_dict)
+        return loss_dict
+
+
+class DistillationDistanceLoss(DistanceLoss):
+    """ """
+
+    def __init__(
+        self, mode="l2", model_name_pairs=[], key=None, name="loss_distance", **kargs
+    ):
+        super().__init__(mode=mode, **kargs)
+        assert isinstance(model_name_pairs, list)
+        self.key = key
+        self.model_name_pairs = model_name_pairs
+        self.name = name + "_l2"
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, pair in enumerate(self.model_name_pairs):
+            out1 = predicts[pair[0]]
+            out2 = predicts[pair[1]]
+            if self.key is not None:
+                out1 = out1[self.key]
+                out2 = out2[self.key]
+            loss = super().forward(out1, out2)
+            if isinstance(loss, dict):
+                for key in loss:
+                    loss_dict["{}_{}_{}".format(self.name, key, idx)] = loss[key]
+            else:
+                loss_dict["{}_{}_{}_{}".format(self.name, pair[0], pair[1], idx)] = loss
+        return loss_dict
+
+
+class DistillationVQASerTokenLayoutLMLoss(VQASerTokenLayoutLMLoss):
+    def __init__(self, num_classes, model_name_list=[], key=None, name="loss_ser"):
+        super().__init__(num_classes=num_classes)
+        self.model_name_list = model_name_list
+        self.key = key
+        self.name = name
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, model_name in enumerate(self.model_name_list):
+            out = predicts[model_name]
+            if self.key is not None:
+                out = out[self.key]
+            loss = super().forward(out, batch)
+            loss_dict["{}_{}".format(self.name, model_name)] = loss["loss"]
+        return loss_dict
+
+
+class DistillationLossFromOutput(LossFromOutput):
+    def __init__(
+        self,
+        reduction="none",
+        model_name_list=[],
+        dist_key=None,
+        key="loss",
+        name="loss_re",
+    ):
+        super().__init__(key=key, reduction=reduction)
+        self.model_name_list = model_name_list
+        self.name = name
+        self.dist_key = dist_key
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, model_name in enumerate(self.model_name_list):
+            out = predicts[model_name]
+            if self.dist_key is not None:
+                out = out[self.dist_key]
+            loss = super().forward(out, batch)
+            loss_dict["{}_{}".format(self.name, model_name)] = loss["loss"]
+        return loss_dict
+
+
+class DistillationSERDMLLoss(DMLLoss):
+    """ """
+
+    def __init__(
+        self,
+        act="softmax",
+        use_log=True,
+        num_classes=7,
+        model_name_pairs=[],
+        key=None,
+        name="loss_dml_ser",
+    ):
+        super().__init__(act=act, use_log=use_log)
+        assert isinstance(model_name_pairs, list)
+        self.key = key
+        self.name = name
+        self.num_classes = num_classes
+        self.model_name_pairs = model_name_pairs
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, pair in enumerate(self.model_name_pairs):
+            out1 = predicts[pair[0]]
+            out2 = predicts[pair[1]]
+            if self.key is not None:
+                out1 = out1[self.key]
+                out2 = out2[self.key]
+            out1 = out1.reshape([-1, out1.shape[-1]])
+            out2 = out2.reshape([-1, out2.shape[-1]])
+
+            attention_mask = batch[2]
+            if attention_mask is not None:
+                active_output = (
+                    attention_mask.reshape(
+                        [
+                            -1,
+                        ]
+                    )
+                    == 1
+                )
+                out1 = out1[active_output]
+                out2 = out2[active_output]
+
+            loss_dict["{}_{}".format(self.name, idx)] = super().forward(out1, out2)
+
+        return loss_dict
+
+
+class DistillationVQADistanceLoss(DistanceLoss):
+    def __init__(
+        self,
+        mode="l2",
+        model_name_pairs=[],
+        key=None,
+        index=None,
+        name="loss_distance",
+        **kargs,
+    ):
+        super().__init__(mode=mode, **kargs)
+        assert isinstance(model_name_pairs, list)
+        self.key = key
+        self.index = index
+        self.model_name_pairs = model_name_pairs
+        self.name = name + "_l2"
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, pair in enumerate(self.model_name_pairs):
+            out1 = predicts[pair[0]]
+            out2 = predicts[pair[1]]
+            attention_mask = batch[2]
+            if self.key is not None:
+                out1 = out1[self.key]
+                out2 = out2[self.key]
+                if self.index is not None:
+                    out1 = out1[:, self.index, :, :]
+                    out2 = out2[:, self.index, :, :]
+                if attention_mask is not None:
+                    max_len = attention_mask.shape[-1]
+                    out1 = out1[:, :max_len]
+                    out2 = out2[:, :max_len]
+                out1 = out1.reshape([-1, out1.shape[-1]])
+                out2 = out2.reshape([-1, out2.shape[-1]])
+            if attention_mask is not None:
+                active_output = (
+                    attention_mask.reshape(
+                        [
+                            -1,
+                        ]
+                    )
+                    == 1
+                )
+                out1 = out1[active_output]
+                out2 = out2[active_output]
+
+            loss = super().forward(out1, out2)
+            if isinstance(loss, dict):
+                for key in loss:
+                    loss_dict["{}_{}nohu_{}".format(self.name, key, idx)] = loss[key]
+            else:
+                loss_dict["{}_{}_{}_{}".format(self.name, pair[0], pair[1], idx)] = loss
+        return loss_dict
+
+
+class CTCDKDLoss(nn.Layer):
+    """
+    KLDivLoss
+    """
+
+    def __init__(self, temperature=0.5, alpha=1.0, beta=1.0):
+        super().__init__()
+        self.temperature = temperature
+        self.alpha = alpha
+        self.beta = beta
+        self.eps = 1e-6
+        self.t = temperature
+        self.act = nn.Softmax(axis=-1)
+        self.use_log = True
+
+    def kl_loss(self, p1, p2):  # predict, label
+        loss = paddle.multiply(
+            p2, paddle.log((p2 + self.eps) / (p1 + self.eps) + self.eps)
+        )
+        bs = loss.shape[0]
+        loss = paddle.sum(loss) / bs
+        return loss
+
+    def _cat_mask(self, t, mask1, mask2):
+        t1 = (t * mask1).sum(axis=1, keepdim=True)
+        t2 = (t * mask2).sum(axis=1, keepdim=True)
+        rt = paddle.concat([t1, t2], axis=1)
+        return rt
+
+    def multi_label_mask(self, targets):
+        targets = targets.astype("int32")
+        res = F.one_hot(targets, num_classes=11465)
+        mask = paddle.clip(paddle.sum(res, axis=1), 0, 1)
+        mask[:, 0] = 0  # ignore ctc blank label
+        return mask
+
+    def forward(self, logits_student, logits_teacher, targets, mask=None):
+        gt_mask = self.multi_label_mask(targets)
+        other_mask = paddle.ones_like(gt_mask) - gt_mask
+
+        pred_student = F.softmax(logits_student / self.temperature, axis=-1)
+        pred_teacher = F.softmax(logits_teacher / self.temperature, axis=-1)
+
+        # differences with dkd
+        pred_student = paddle.mean(pred_student, axis=1)
+        pred_teacher = paddle.mean(pred_teacher, axis=1)
+
+        pred_student = self._cat_mask(pred_student, gt_mask, other_mask)
+        pred_teacher = self._cat_mask(pred_teacher, gt_mask, other_mask)
+
+        # differences with dkd
+        tckd_loss = self.kl_loss(pred_student, pred_teacher)
+
+        gt_mask_ex = paddle.expand_as(gt_mask.unsqueeze(axis=1), logits_teacher)
+        pred_teacher_part2 = F.softmax(
+            logits_teacher / self.temperature - 1000.0 * gt_mask_ex, axis=-1
+        )
+        pred_student_part2 = F.softmax(
+            logits_student / self.temperature - 1000.0 * gt_mask_ex, axis=-1
+        )
+        # differences with dkd
+        pred_teacher_part2 = paddle.mean(pred_teacher_part2, axis=1)
+        pred_student_part2 = paddle.mean(pred_student_part2, axis=1)
+
+        # differences with dkd
+        nckd_loss = self.kl_loss(pred_student_part2, pred_teacher_part2)
+        loss = self.alpha * tckd_loss + self.beta * nckd_loss
+        return loss
+
+
+class KLCTCLogits(nn.Layer):
+    def __init__(self, weight=1.0, reduction="mean", mode="mean"):
+        super().__init__()
+        self.weight = weight
+        self.reduction = reduction
+        self.eps = 1e-6
+        self.t = 0.5
+        self.act = nn.Softmax(axis=-1)
+        self.use_log = True
+        self.mode = mode
+        self.ctc_dkd_loss = CTCDKDLoss()
+
+    def kl_loss(self, p1, p2):  # predict, label
+        loss = paddle.multiply(
+            p2, paddle.log((p2 + self.eps) / (p1 + self.eps) + self.eps)
+        )
+        bs = loss.shape[0]
+        loss = paddle.sum(loss) / bs
+        return loss
+
+    def forward_meanmax(self, stu_out, tea_out):
+        stu_out = paddle.mean(F.softmax(stu_out / self.t, axis=-1), axis=1)
+        tea_out = paddle.mean(F.softmax(tea_out / self.t, axis=-1), axis=1)
+        loss = self.kl_loss(stu_out, tea_out)
+
+        return loss
+
+    def forward_meanlog(self, stu_out, tea_out):
+        stu_out = paddle.mean(F.softmax(stu_out / self.t, axis=-1), axis=1)
+        tea_out = paddle.mean(F.softmax(tea_out / self.t, axis=-1), axis=1)
+        if self.use_log is True:
+            # for recognition distillation, log is needed for feature map
+            log_out1 = paddle.log(stu_out)
+            log_out2 = paddle.log(tea_out)
+            loss = (
+                self._kldiv(log_out1, tea_out) + self._kldiv(log_out2, stu_out)
+            ) / 2.0
+
+        return loss
+
+    def forward_sum(self, stu_out, tea_out):
+        stu_out = paddle.sum(F.softmax(stu_out / self.t, axis=-1), axis=1)
+        tea_out = paddle.sum(F.softmax(tea_out / self.t, axis=-1), axis=1)
+        stu_out = paddle.log(stu_out)
+        bs = stu_out.shape[0]
+        loss = tea_out * (paddle.log(tea_out + self.eps) - stu_out)
+        loss = paddle.sum(loss, axis=1) / loss.shape[0]
+        return loss
+
+    def _kldiv(self, x, target):
+        eps = 1.0e-10
+        loss = target * (paddle.log(target + eps) - x)
+        loss = paddle.sum(paddle.mean(loss, axis=1)) / loss.shape[0]
+        return loss
+
+    def forward(self, stu_out, tea_out, targets=None):
+        if self.mode == "log":
+            return self.forward_log(stu_out, tea_out)
+        elif self.mode == "mean":
+            blank_mask = paddle.ones_like(stu_out)
+            blank_mask.stop_gradient = True
+            blank_mask[:, :, 0] = -1
+            stu_out *= blank_mask
+            tea_out *= blank_mask
+            return self.forward_meanmax(stu_out, tea_out)
+        elif self.mode == "sum":
+            return self.forward_sum(stu_out, tea_out)
+        elif self.mode == "meanlog":
+            blank_mask = paddle.ones_like(stu_out)
+            blank_mask.stop_gradient = True
+            blank_mask[:, :, 0] = -1
+            stu_out *= blank_mask
+            tea_out *= blank_mask
+            return self.forward_meanlog(stu_out, tea_out)
+        elif self.mode == "ctcdkd":
+            # ignore ctc blank logits
+            blank_mask = paddle.ones_like(stu_out)
+            blank_mask.stop_gradient = True
+            blank_mask[:, :, 0] = -1
+            stu_out *= blank_mask
+            tea_out *= blank_mask
+            return self.ctc_dkd_loss(stu_out, tea_out, targets)
+        else:
+            raise ValueError("error!!!!!!")
+
+    def forward_log(self, out1, out2):
+        if self.act is not None:
+            out1 = self.act(out1) + 1e-10
+            out2 = self.act(out2) + 1e-10
+        if self.use_log is True:
+            # for recognition distillation, log is needed for feature map
+            log_out1 = paddle.log(out1)
+            log_out2 = paddle.log(out2)
+            loss = (self._kldiv(log_out1, out2) + self._kldiv(log_out2, out1)) / 2.0
+
+        return loss
+
+
+class DistillCTCLogits(KLCTCLogits):
+    def __init__(
+        self, model_name_pairs=[], key=None, name="ctc_logits", reduction="mean"
+    ):
+        super().__init__(reduction=reduction)
+        self.model_name_pairs = self._check_model_name_pairs(model_name_pairs)
+        self.key = key
+        self.name = name
+
+    def _check_model_name_pairs(self, model_name_pairs):
+        if not isinstance(model_name_pairs, list):
+            return []
+        elif isinstance(model_name_pairs[0], list) and isinstance(
+            model_name_pairs[0][0], str
+        ):
+            return model_name_pairs
+        else:
+            return [model_name_pairs]
+
+    def forward(self, predicts, batch):
+        loss_dict = dict()
+        for idx, pair in enumerate(self.model_name_pairs):
+            out1 = predicts[pair[0]]
+            out2 = predicts[pair[1]]
+
+            if self.key is not None:
+                out1 = out1[self.key]["ctc"]
+                out2 = out2[self.key]["ctc"]
+
+            ctc_label = batch[1]
+            loss = super().forward(out1, out2, ctc_label)
+            if isinstance(loss, dict):
+                for key in loss:
+                    loss_dict[
+                        "{}_{}_{}".format(self.name, self.model_name_pairs, idx)
+                    ] = loss[key]
+            else:
+                loss_dict["{}_{}".format(self.name, idx)] = loss
+        return loss_dict
diff --git a/ppocr/losses/e2e_pg_loss.py b/ppocr/losses/e2e_pg_loss.py
new file mode 100644
index 0000000..8b667f0
--- /dev/null
+++ b/ppocr/losses/e2e_pg_loss.py
@@ -0,0 +1,165 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from paddle import nn
+import paddle
+
+from .det_basic_loss import DiceLoss
+from ppocr.utils.e2e_utils.extract_batchsize import pre_process
+
+
+class PGLoss(nn.Layer):
+    def __init__(
+        self, tcl_bs, max_text_length, max_text_nums, pad_num, eps=1e-6, **kwargs
+    ):
+        super(PGLoss, self).__init__()
+        self.tcl_bs = tcl_bs
+        self.max_text_nums = max_text_nums
+        self.max_text_length = max_text_length
+        self.pad_num = pad_num
+        self.dice_loss = DiceLoss(eps=eps)
+
+    def border_loss(self, f_border, l_border, l_score, l_mask):
+        l_border_split, l_border_norm = paddle.tensor.split(
+            l_border, num_or_sections=[4, 1], axis=1
+        )
+        f_border_split = f_border
+        b, c, h, w = l_border_norm.shape
+        l_border_norm_split = paddle.expand(x=l_border_norm, shape=[b, 4 * c, h, w])
+        b, c, h, w = l_score.shape
+        l_border_score = paddle.expand(x=l_score, shape=[b, 4 * c, h, w])
+        b, c, h, w = l_mask.shape
+        l_border_mask = paddle.expand(x=l_mask, shape=[b, 4 * c, h, w])
+        border_diff = l_border_split - f_border_split
+        abs_border_diff = paddle.abs(border_diff)
+        border_sign = abs_border_diff < 1.0
+        border_sign = paddle.cast(border_sign, dtype="float32")
+        border_sign.stop_gradient = True
+        border_in_loss = 0.5 * abs_border_diff * abs_border_diff * border_sign + (
+            abs_border_diff - 0.5
+        ) * (1.0 - border_sign)
+        border_out_loss = l_border_norm_split * border_in_loss
+        border_loss = paddle.sum(border_out_loss * l_border_score * l_border_mask) / (
+            paddle.sum(l_border_score * l_border_mask) + 1e-5
+        )
+        return border_loss
+
+    def direction_loss(self, f_direction, l_direction, l_score, l_mask):
+        l_direction_split, l_direction_norm = paddle.tensor.split(
+            l_direction, num_or_sections=[2, 1], axis=1
+        )
+        f_direction_split = f_direction
+        b, c, h, w = l_direction_norm.shape
+        l_direction_norm_split = paddle.expand(
+            x=l_direction_norm, shape=[b, 2 * c, h, w]
+        )
+        b, c, h, w = l_score.shape
+        l_direction_score = paddle.expand(x=l_score, shape=[b, 2 * c, h, w])
+        b, c, h, w = l_mask.shape
+        l_direction_mask = paddle.expand(x=l_mask, shape=[b, 2 * c, h, w])
+        direction_diff = l_direction_split - f_direction_split
+        abs_direction_diff = paddle.abs(direction_diff)
+        direction_sign = abs_direction_diff < 1.0
+        direction_sign = paddle.cast(direction_sign, dtype="float32")
+        direction_sign.stop_gradient = True
+        direction_in_loss = (
+            0.5 * abs_direction_diff * abs_direction_diff * direction_sign
+            + (abs_direction_diff - 0.5) * (1.0 - direction_sign)
+        )
+        direction_out_loss = l_direction_norm_split * direction_in_loss
+        direction_loss = paddle.sum(
+            direction_out_loss * l_direction_score * l_direction_mask
+        ) / (paddle.sum(l_direction_score * l_direction_mask) + 1e-5)
+        return direction_loss
+
+    def ctcloss(self, f_char, tcl_pos, tcl_mask, tcl_label, label_t):
+        f_char = paddle.transpose(f_char, [0, 2, 3, 1])
+        tcl_pos = paddle.reshape(tcl_pos, [-1, 3])
+        tcl_pos = paddle.cast(tcl_pos, dtype=int)
+        f_tcl_char = paddle.gather_nd(f_char, tcl_pos)
+        f_tcl_char = paddle.reshape(
+            f_tcl_char, [-1, 64, self.pad_num + 1]
+        )  # len(Lexicon_Table)+1
+        f_tcl_char_fg, f_tcl_char_bg = paddle.split(
+            f_tcl_char, [self.pad_num, 1], axis=2
+        )
+        f_tcl_char_bg = f_tcl_char_bg * tcl_mask + (1.0 - tcl_mask) * 20.0
+        b, c, l = tcl_mask.shape
+        tcl_mask_fg = paddle.expand(x=tcl_mask, shape=[b, c, self.pad_num * l])
+        tcl_mask_fg.stop_gradient = True
+        f_tcl_char_fg = f_tcl_char_fg * tcl_mask_fg + (1.0 - tcl_mask_fg) * (-20.0)
+        f_tcl_char_mask = paddle.concat([f_tcl_char_fg, f_tcl_char_bg], axis=2)
+        f_tcl_char_ld = paddle.transpose(f_tcl_char_mask, (1, 0, 2))
+        N, B, _ = f_tcl_char_ld.shape
+        input_lengths = paddle.to_tensor([N] * B, dtype="int64")
+        cost = paddle.nn.functional.ctc_loss(
+            log_probs=f_tcl_char_ld,
+            labels=tcl_label,
+            input_lengths=input_lengths,
+            label_lengths=label_t,
+            blank=self.pad_num,
+            reduction="none",
+        )
+        cost = cost.mean()
+        return cost
+
+    def forward(self, predicts, labels):
+        (
+            images,
+            tcl_maps,
+            tcl_label_maps,
+            border_maps,
+            direction_maps,
+            training_masks,
+            label_list,
+            pos_list,
+            pos_mask,
+        ) = labels
+        # for all the batch_size
+        pos_list, pos_mask, label_list, label_t = pre_process(
+            label_list,
+            pos_list,
+            pos_mask,
+            self.max_text_length,
+            self.max_text_nums,
+            self.pad_num,
+            self.tcl_bs,
+        )
+
+        f_score, f_border, f_direction, f_char = (
+            predicts["f_score"],
+            predicts["f_border"],
+            predicts["f_direction"],
+            predicts["f_char"],
+        )
+        score_loss = self.dice_loss(f_score, tcl_maps, training_masks)
+        border_loss = self.border_loss(f_border, border_maps, tcl_maps, training_masks)
+        direction_loss = self.direction_loss(
+            f_direction, direction_maps, tcl_maps, training_masks
+        )
+        ctc_loss = self.ctcloss(f_char, pos_list, pos_mask, label_list, label_t)
+        loss_all = score_loss + border_loss + direction_loss + 5 * ctc_loss
+
+        losses = {
+            "loss": loss_all,
+            "score_loss": score_loss,
+            "border_loss": border_loss,
+            "direction_loss": direction_loss,
+            "ctc_loss": ctc_loss,
+        }
+        return losses
diff --git a/ppocr/losses/kie_sdmgr_loss.py b/ppocr/losses/kie_sdmgr_loss.py
new file mode 100644
index 0000000..685a141
--- /dev/null
+++ b/ppocr/losses/kie_sdmgr_loss.py
@@ -0,0 +1,116 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# reference from : https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/kie/losses/sdmgr_loss.py
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from paddle import nn
+import paddle
+
+
+class SDMGRLoss(nn.Layer):
+    def __init__(self, node_weight=1.0, edge_weight=1.0, ignore=0):
+        super().__init__()
+        self.loss_node = nn.CrossEntropyLoss(ignore_index=ignore)
+        self.loss_edge = nn.CrossEntropyLoss(ignore_index=-1)
+        self.node_weight = node_weight
+        self.edge_weight = edge_weight
+        self.ignore = ignore
+
+    def pre_process(self, gts, tag):
+        gts, tag = gts.numpy(), tag.numpy().tolist()
+        temp_gts = []
+        batch = len(tag)
+        for i in range(batch):
+            num, recoder_len = tag[i][0], tag[i][1]
+            temp_gts.append(paddle.to_tensor(gts[i, :num, : num + 1], dtype="int64"))
+        return temp_gts
+
+    def accuracy(self, pred, target, topk=1, thresh=None):
+        """Calculate accuracy according to the prediction and target.
+
+        Args:
+            pred (torch.Tensor): The model prediction, shape (N, num_class)
+            target (torch.Tensor): The target of each prediction, shape (N, )
+            topk (int | tuple[int], optional): If the predictions in ``topk``
+                matches the target, the predictions will be regarded as
+                correct ones. Defaults to 1.
+            thresh (float, optional): If not None, predictions with scores under
+                this threshold are considered incorrect. Default to None.
+
+        Returns:
+            float | tuple[float]: If the input ``topk`` is a single integer,
+                the function will return a single float as accuracy. If
+                ``topk`` is a tuple containing multiple integers, the
+                function will return a tuple containing accuracies of
+                each ``topk`` number.
+        """
+        assert isinstance(topk, (int, tuple))
+        if isinstance(topk, int):
+            topk = (topk,)
+            return_single = True
+        else:
+            return_single = False
+
+        maxk = max(topk)
+        if pred.shape[0] == 0:
+            accu = [pred.new_tensor(0.0) for i in range(len(topk))]
+            return accu[0] if return_single else accu
+        pred_value, pred_label = paddle.topk(pred, maxk, axis=1)
+        pred_label = pred_label.transpose([1, 0])  # transpose to shape (maxk, N)
+        correct = paddle.equal(
+            pred_label, (target.reshape([1, -1]).expand_as(pred_label))
+        )
+        res = []
+        for k in topk:
+            correct_k = paddle.sum(
+                correct[:k].reshape([-1]).astype("float32"), axis=0, keepdim=True
+            )
+            res.append(
+                paddle.multiply(correct_k, paddle.to_tensor(100.0 / pred.shape[0]))
+            )
+        return res[0] if return_single else res
+
+    def forward(self, pred, batch):
+        node_preds, edge_preds = pred
+        gts, tag = batch[4], batch[5]
+        gts = self.pre_process(gts, tag)
+        node_gts, edge_gts = [], []
+        for gt in gts:
+            node_gts.append(gt[:, 0])
+            edge_gts.append(gt[:, 1:].reshape([-1]))
+        node_gts = paddle.concat(node_gts)
+        edge_gts = paddle.concat(edge_gts)
+
+        node_valids = paddle.nonzero(node_gts != self.ignore).reshape([-1])
+        edge_valids = paddle.nonzero(edge_gts != -1).reshape([-1])
+        loss_node = self.loss_node(node_preds, node_gts)
+        loss_edge = self.loss_edge(edge_preds, edge_gts)
+        loss = self.node_weight * loss_node + self.edge_weight * loss_edge
+        return dict(
+            loss=loss,
+            loss_node=loss_node,
+            loss_edge=loss_edge,
+            acc_node=self.accuracy(
+                paddle.gather(node_preds, node_valids),
+                paddle.gather(node_gts, node_valids),
+            ),
+            acc_edge=self.accuracy(
+                paddle.gather(edge_preds, edge_valids),
+                paddle.gather(edge_gts, edge_valids),
+            ),
+        )
diff --git a/ppocr/losses/rec_aster_loss.py b/ppocr/losses/rec_aster_loss.py
new file mode 100644
index 0000000..8a8faa1
--- /dev/null
+++ b/ppocr/losses/rec_aster_loss.py
@@ -0,0 +1,103 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+
+
+class CosineEmbeddingLoss(nn.Layer):
+    def __init__(self, margin=0.0):
+        super(CosineEmbeddingLoss, self).__init__()
+        self.margin = margin
+        self.epsilon = 1e-12
+
+    def forward(self, x1, x2, target):
+        similarity = paddle.sum(x1 * x2, axis=-1) / (
+            paddle.norm(x1, axis=-1) * paddle.norm(x2, axis=-1) + self.epsilon
+        )
+        one_list = paddle.full_like(target, fill_value=1)
+        out = paddle.mean(
+            paddle.where(
+                paddle.equal(target, one_list),
+                1.0 - similarity,
+                paddle.maximum(paddle.zeros_like(similarity), similarity - self.margin),
+            )
+        )
+
+        return out
+
+
+class AsterLoss(nn.Layer):
+    def __init__(
+        self,
+        weight=None,
+        size_average=True,
+        ignore_index=-100,
+        sequence_normalize=False,
+        sample_normalize=True,
+        **kwargs,
+    ):
+        super(AsterLoss, self).__init__()
+        self.weight = weight
+        self.size_average = size_average
+        self.ignore_index = ignore_index
+        self.sequence_normalize = sequence_normalize
+        self.sample_normalize = sample_normalize
+        self.loss_sem = CosineEmbeddingLoss()
+        self.is_cosin_loss = True
+        self.loss_func_rec = nn.CrossEntropyLoss(weight=None, reduction="none")
+
+    def forward(self, predicts, batch):
+        targets = batch[1].astype("int64")
+        label_lengths = batch[2].astype("int64")
+        sem_target = batch[3].astype("float32")
+        embedding_vectors = predicts["embedding_vectors"]
+        rec_pred = predicts["rec_pred"]
+
+        if not self.is_cosin_loss:
+            sem_loss = paddle.sum(self.loss_sem(embedding_vectors, sem_target))
+        else:
+            label_target = paddle.ones([embedding_vectors.shape[0]])
+            sem_loss = paddle.sum(
+                self.loss_sem(embedding_vectors, sem_target, label_target)
+            )
+
+        # rec loss
+        batch_size, def_max_length = targets.shape[0], targets.shape[1]
+
+        mask = paddle.zeros([batch_size, def_max_length])
+        for i in range(batch_size):
+            mask[i, : label_lengths[i]] = 1
+        mask = paddle.cast(mask, "float32")
+        max_length = max(label_lengths)
+        assert max_length == rec_pred.shape[1]
+        targets = targets[:, :max_length]
+        mask = mask[:, :max_length]
+        rec_pred = paddle.reshape(rec_pred, [-1, rec_pred.shape[2]])
+        input = nn.functional.log_softmax(rec_pred, axis=1)
+        targets = paddle.reshape(targets, [-1, 1])
+        mask = paddle.reshape(mask, [-1, 1])
+        output = -paddle.index_sample(input, index=targets) * mask
+        output = paddle.sum(output)
+        if self.sequence_normalize:
+            output = output / paddle.sum(mask)
+        if self.sample_normalize:
+            output = output / batch_size
+
+        loss = output + sem_loss * 0.1
+        return {"loss": loss}
diff --git a/ppocr/losses/rec_att_loss.py b/ppocr/losses/rec_att_loss.py
new file mode 100644
index 0000000..e0f65d9
--- /dev/null
+++ b/ppocr/losses/rec_att_loss.py
@@ -0,0 +1,43 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+
+
+class AttentionLoss(nn.Layer):
+    def __init__(self, **kwargs):
+        super(AttentionLoss, self).__init__()
+        self.loss_func = nn.CrossEntropyLoss(weight=None, reduction="none")
+
+    def forward(self, predicts, batch):
+        targets = batch[1].astype("int64")
+        label_lengths = batch[2].astype("int64")
+        batch_size, num_steps, num_classes = (
+            predicts.shape[0],
+            predicts.shape[1],
+            predicts.shape[2],
+        )
+        assert (
+            len(targets.shape) == len(list(predicts.shape)) - 1
+        ), "The target's shape and inputs's shape is [N, d] and [N, num_steps]"
+
+        inputs = paddle.reshape(predicts, [-1, predicts.shape[-1]])
+        targets = paddle.reshape(targets, [-1])
+
+        return {"loss": paddle.sum(self.loss_func(inputs, targets))}
diff --git a/ppocr/losses/rec_can_loss.py b/ppocr/losses/rec_can_loss.py
new file mode 100644
index 0000000..1bc58f5
--- /dev/null
+++ b/ppocr/losses/rec_can_loss.py
@@ -0,0 +1,88 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/LBH1024/CAN/models/can.py
+"""
+
+import paddle
+import paddle.nn as nn
+import numpy as np
+
+
+class CANLoss(nn.Layer):
+    """
+    CANLoss is consist of two part:
+        word_average_loss: average accuracy of the symbol
+        counting_loss: counting loss of every symbol
+    """
+
+    def __init__(self):
+        super(CANLoss, self).__init__()
+
+        self.use_label_mask = False
+        self.out_channel = 111
+        self.cross = (
+            nn.CrossEntropyLoss(reduction="none")
+            if self.use_label_mask
+            else nn.CrossEntropyLoss()
+        )
+        self.counting_loss = nn.SmoothL1Loss(reduction="mean")
+        self.ratio = 16
+
+    def forward(self, preds, batch):
+        word_probs = preds[0]
+        counting_preds = preds[1]
+        counting_preds1 = preds[2]
+        counting_preds2 = preds[3]
+        labels = batch[2]
+        labels_mask = batch[3]
+        counting_labels = gen_counting_label(labels, self.out_channel, True)
+        counting_loss = (
+            self.counting_loss(counting_preds1, counting_labels)
+            + self.counting_loss(counting_preds2, counting_labels)
+            + self.counting_loss(counting_preds, counting_labels)
+        )
+
+        word_loss = self.cross(
+            paddle.reshape(word_probs, [-1, word_probs.shape[-1]]),
+            paddle.reshape(labels, [-1]),
+        )
+        word_average_loss = (
+            paddle.sum(paddle.reshape(word_loss * labels_mask, [-1]))
+            / (paddle.sum(labels_mask) + 1e-10)
+            if self.use_label_mask
+            else word_loss
+        )
+        loss = word_average_loss + counting_loss
+        return {"loss": loss}
+
+
+def gen_counting_label(labels, channel, tag):
+    b, t = labels.shape
+    counting_labels = np.zeros([b, channel])
+
+    if tag:
+        ignore = [0, 1, 107, 108, 109, 110]
+    else:
+        ignore = []
+    for i in range(b):
+        for j in range(t):
+            k = labels[i][j]
+            if k in ignore:
+                continue
+            else:
+                counting_labels[i][k] += 1
+    counting_labels = paddle.to_tensor(counting_labels, dtype="float32")
+    return counting_labels
diff --git a/ppocr/losses/rec_ce_loss.py b/ppocr/losses/rec_ce_loss.py
new file mode 100644
index 0000000..45906fd
--- /dev/null
+++ b/ppocr/losses/rec_ce_loss.py
@@ -0,0 +1,61 @@
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+
+
+class CELoss(nn.Layer):
+    def __init__(self, smoothing=False, with_all=False, ignore_index=-1, **kwargs):
+        super(CELoss, self).__init__()
+        if ignore_index >= 0:
+            self.loss_func = nn.CrossEntropyLoss(
+                reduction="mean", ignore_index=ignore_index
+            )
+        else:
+            self.loss_func = nn.CrossEntropyLoss(reduction="mean")
+        self.smoothing = smoothing
+        self.with_all = with_all
+
+    def forward(self, pred, batch):
+        if isinstance(pred, dict):  # for ABINet
+            loss = {}
+            loss_sum = []
+            for name, logits in pred.items():
+                if isinstance(logits, list):
+                    logit_num = len(logits)
+                    all_tgt = paddle.concat([batch[1]] * logit_num, 0)
+                    all_logits = paddle.concat(logits, 0)
+                    flt_logtis = all_logits.reshape([-1, all_logits.shape[2]])
+                    flt_tgt = all_tgt.reshape([-1])
+                else:
+                    flt_logtis = logits.reshape([-1, logits.shape[2]])
+                    flt_tgt = batch[1].reshape([-1])
+                loss[name + "_loss"] = self.loss_func(flt_logtis, flt_tgt)
+                loss_sum.append(loss[name + "_loss"])
+            loss["loss"] = sum(loss_sum)
+            return loss
+        else:
+            if self.with_all:  # for ViTSTR
+                tgt = batch[1]
+                pred = pred.reshape([-1, pred.shape[2]])
+                tgt = tgt.reshape([-1])
+                loss = self.loss_func(pred, tgt)
+                return {"loss": loss}
+            else:  # for NRTR
+                max_len = batch[2].max()
+                tgt = batch[1][:, 1 : 2 + max_len]
+                pred = pred.reshape([-1, pred.shape[2]])
+                tgt = tgt.reshape([-1])
+                if self.smoothing:
+                    eps = 0.1
+                    n_class = pred.shape[1]
+                    one_hot = F.one_hot(tgt, pred.shape[1])
+                    one_hot = one_hot * (1 - eps) + (1 - one_hot) * eps / (n_class - 1)
+                    log_prb = F.log_softmax(pred, axis=1)
+                    non_pad_mask = paddle.not_equal(
+                        tgt, paddle.zeros(tgt.shape, dtype=tgt.dtype)
+                    )
+                    loss = -(one_hot * log_prb).sum(axis=1)
+                    loss = loss.masked_select(non_pad_mask).mean()
+                else:
+                    loss = self.loss_func(pred, tgt)
+                return {"loss": loss}
diff --git a/ppocr/losses/rec_cppd_loss.py b/ppocr/losses/rec_cppd_loss.py
new file mode 100755
index 0000000..1e884ce
--- /dev/null
+++ b/ppocr/losses/rec_cppd_loss.py
@@ -0,0 +1,76 @@
+# copyright (c) 2023 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+
+
+class CPPDLoss(nn.Layer):
+    def __init__(
+        self, smoothing=False, ignore_index=100, sideloss_weight=1.0, **kwargs
+    ):
+        super(CPPDLoss, self).__init__()
+        self.edge_ce = nn.CrossEntropyLoss(reduction="mean", ignore_index=ignore_index)
+        self.char_node_ce = nn.CrossEntropyLoss(reduction="mean")
+        self.pos_node_ce = nn.BCEWithLogitsLoss(reduction="mean")
+        self.smoothing = smoothing
+        self.ignore_index = ignore_index
+        self.sideloss_weight = sideloss_weight
+
+    def label_smoothing_ce(self, preds, targets):
+        non_pad_mask = paddle.not_equal(
+            targets,
+            paddle.zeros(targets.shape, dtype=targets.dtype) + self.ignore_index,
+        )
+        tgts = paddle.where(
+            targets
+            == (paddle.zeros(targets.shape, dtype=targets.dtype) + self.ignore_index),
+            paddle.zeros(targets.shape, dtype=targets.dtype),
+            targets,
+        )
+        eps = 0.1
+        n_class = preds.shape[1]
+        one_hot = F.one_hot(tgts, preds.shape[1])
+        one_hot = one_hot * (1 - eps) + (1 - one_hot) * eps / (n_class - 1)
+        log_prb = F.log_softmax(preds, axis=1)
+        loss = -(one_hot * log_prb).sum(axis=1)
+        loss = loss.masked_select(non_pad_mask).mean()
+        return loss
+
+    def forward(self, pred, batch):
+        node_feats, edge_feats = pred
+        node_tgt = batch[2]
+        char_tgt = batch[1]
+
+        loss_char_node = self.char_node_ce(
+            node_feats[0].flatten(0, 1), node_tgt[:, :-26].flatten(0, 1)
+        )
+        loss_pos_node = self.pos_node_ce(
+            node_feats[1].flatten(0, 1), node_tgt[:, -26:].flatten(0, 1).cast("float32")
+        )
+        loss_node = loss_char_node + loss_pos_node
+
+        edge_feats = edge_feats.flatten(0, 1)
+        char_tgt = char_tgt.flatten(0, 1)
+        if self.smoothing:
+            loss_edge = self.label_smoothing_ce(edge_feats, char_tgt)
+        else:
+            loss_edge = self.edge_ce(edge_feats, char_tgt)
+
+        return {
+            "loss": self.sideloss_weight * loss_node + loss_edge,
+            "loss_node": self.sideloss_weight * loss_node,
+            "loss_edge": loss_edge,
+        }
diff --git a/ppocr/losses/rec_ctc_loss.py b/ppocr/losses/rec_ctc_loss.py
new file mode 100755
index 0000000..c701fef
--- /dev/null
+++ b/ppocr/losses/rec_ctc_loss.py
@@ -0,0 +1,46 @@
+# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+
+
+class CTCLoss(nn.Layer):
+    def __init__(self, use_focal_loss=False, **kwargs):
+        super(CTCLoss, self).__init__()
+        self.loss_func = nn.CTCLoss(blank=0, reduction="none")
+        self.use_focal_loss = use_focal_loss
+
+    def forward(self, predicts, batch):
+        if isinstance(predicts, (list, tuple)):
+            predicts = predicts[-1]
+        predicts = predicts.transpose((1, 0, 2))
+        N, B, _ = predicts.shape
+        preds_lengths = paddle.to_tensor(
+            [N] * B, dtype="int64", place=paddle.CPUPlace()
+        )
+        labels = batch[1].astype("int32")
+        label_lengths = batch[2].astype("int64")
+        loss = self.loss_func(predicts, labels, preds_lengths, label_lengths)
+        if self.use_focal_loss:
+            weight = paddle.exp(-loss)
+            weight = paddle.subtract(paddle.to_tensor([1.0]), weight)
+            weight = paddle.square(weight)
+            loss = paddle.multiply(loss, weight)
+        loss = loss.mean()
+        return {"loss": loss}
diff --git a/ppocr/losses/rec_enhanced_ctc_loss.py b/ppocr/losses/rec_enhanced_ctc_loss.py
new file mode 100644
index 0000000..e7e0696
--- /dev/null
+++ b/ppocr/losses/rec_enhanced_ctc_loss.py
@@ -0,0 +1,76 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+from .ace_loss import ACELoss
+from .center_loss import CenterLoss
+from .rec_ctc_loss import CTCLoss
+
+
+class EnhancedCTCLoss(nn.Layer):
+    def __init__(
+        self,
+        use_focal_loss=False,
+        use_ace_loss=False,
+        ace_loss_weight=0.1,
+        use_center_loss=False,
+        center_loss_weight=0.05,
+        num_classes=6625,
+        feat_dim=96,
+        init_center=False,
+        center_file_path=None,
+        **kwargs,
+    ):
+        super(EnhancedCTCLoss, self).__init__()
+        self.ctc_loss_func = CTCLoss(use_focal_loss=use_focal_loss)
+
+        self.use_ace_loss = False
+        if use_ace_loss:
+            self.use_ace_loss = use_ace_loss
+            self.ace_loss_func = ACELoss()
+            self.ace_loss_weight = ace_loss_weight
+
+        self.use_center_loss = False
+        if use_center_loss:
+            self.use_center_loss = use_center_loss
+            self.center_loss_func = CenterLoss(
+                num_classes=num_classes,
+                feat_dim=feat_dim,
+                init_center=init_center,
+                center_file_path=center_file_path,
+            )
+            self.center_loss_weight = center_loss_weight
+
+    def __call__(self, predicts, batch):
+        loss = self.ctc_loss_func(predicts, batch)["loss"]
+
+        if self.use_center_loss:
+            center_loss = (
+                self.center_loss_func(predicts, batch)["loss_center"]
+                * self.center_loss_weight
+            )
+            loss = loss + center_loss
+
+        if self.use_ace_loss:
+            ace_loss = (
+                self.ace_loss_func(predicts, batch)["loss_ace"] * self.ace_loss_weight
+            )
+            loss = loss + ace_loss
+
+        return {"enhanced_ctc_loss": loss}
diff --git a/ppocr/losses/rec_latexocr_loss.py b/ppocr/losses/rec_latexocr_loss.py
new file mode 100644
index 0000000..d209c04
--- /dev/null
+++ b/ppocr/losses/rec_latexocr_loss.py
@@ -0,0 +1,47 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+This code is refer from:
+https://github.com/lucidrains/x-transformers/blob/main/x_transformers/autoregressive_wrapper.py
+"""
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+import numpy as np
+
+
+class LaTeXOCRLoss(nn.Layer):
+    """
+    LaTeXOCR adopt CrossEntropyLoss for network training.
+    """
+
+    def __init__(self):
+        super(LaTeXOCRLoss, self).__init__()
+        self.ignore_index = -100
+        self.cross = nn.CrossEntropyLoss(
+            reduction="mean", ignore_index=self.ignore_index
+        )
+
+    def forward(self, preds, batch):
+        word_probs = preds
+        labels = batch[1][:, 1:]
+        word_loss = self.cross(
+            paddle.reshape(word_probs, [-1, word_probs.shape[-1]]),
+            paddle.reshape(labels, [-1]),
+        )
+
+        loss = word_loss
+        return {"loss": loss}
diff --git a/ppocr/losses/rec_multi_loss.py b/ppocr/losses/rec_multi_loss.py
new file mode 100644
index 0000000..74be385
--- /dev/null
+++ b/ppocr/losses/rec_multi_loss.py
@@ -0,0 +1,68 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+
+from .rec_ctc_loss import CTCLoss
+from .rec_sar_loss import SARLoss
+from .rec_nrtr_loss import NRTRLoss
+
+
+class MultiLoss(nn.Layer):
+    def __init__(self, **kwargs):
+        super().__init__()
+        self.loss_funcs = {}
+        self.loss_list = kwargs.pop("loss_config_list")
+        self.weight_1 = kwargs.get("weight_1", 1.0)
+        self.weight_2 = kwargs.get("weight_2", 1.0)
+        for loss_info in self.loss_list:
+            for name, param in loss_info.items():
+                if param is not None:
+                    kwargs.update(param)
+                loss = eval(name)(**kwargs)
+                self.loss_funcs[name] = loss
+
+    def forward(self, predicts, batch):
+        self.total_loss = {}
+        total_loss = 0.0
+        # batch [image, label_ctc, label_sar, length, valid_ratio]
+        for name, loss_func in self.loss_funcs.items():
+            if name == "CTCLoss":
+                loss = (
+                    loss_func(predicts["ctc"], batch[:2] + batch[3:])["loss"]
+                    * self.weight_1
+                )
+            elif name == "SARLoss":
+                loss = (
+                    loss_func(predicts["sar"], batch[:1] + batch[2:])["loss"]
+                    * self.weight_2
+                )
+            elif name == "NRTRLoss":
+                loss = (
+                    loss_func(predicts["gtc"], batch[:1] + batch[2:])["loss"]
+                    * self.weight_2
+                )
+            else:
+                raise NotImplementedError(
+                    "{} is not supported in MultiLoss yet".format(name)
+                )
+            self.total_loss[name] = loss
+            total_loss += loss
+        self.total_loss["loss"] = total_loss
+        return self.total_loss
diff --git a/ppocr/losses/rec_nrtr_loss.py b/ppocr/losses/rec_nrtr_loss.py
new file mode 100644
index 0000000..a4f56de
--- /dev/null
+++ b/ppocr/losses/rec_nrtr_loss.py
@@ -0,0 +1,33 @@
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+
+
+class NRTRLoss(nn.Layer):
+    def __init__(self, smoothing=True, ignore_index=0, **kwargs):
+        super(NRTRLoss, self).__init__()
+        if ignore_index >= 0 and not smoothing:
+            self.loss_func = nn.CrossEntropyLoss(
+                reduction="mean", ignore_index=ignore_index
+            )
+        self.smoothing = smoothing
+
+    def forward(self, pred, batch):
+        max_len = batch[2].max()
+        tgt = batch[1][:, 1 : 2 + max_len]
+        pred = pred.reshape([-1, pred.shape[2]])
+        tgt = tgt.reshape([-1])
+        if self.smoothing:
+            eps = 0.1
+            n_class = pred.shape[1]
+            one_hot = F.one_hot(tgt, pred.shape[1])
+            one_hot = one_hot * (1 - eps) + (1 - one_hot) * eps / (n_class - 1)
+            log_prb = F.log_softmax(pred, axis=1)
+            non_pad_mask = paddle.not_equal(
+                tgt, paddle.zeros(tgt.shape, dtype=tgt.dtype)
+            )
+            loss = -(one_hot * log_prb).sum(axis=1)
+            loss = loss.masked_select(non_pad_mask).mean()
+        else:
+            loss = self.loss_func(pred, tgt)
+        return {"loss": loss}
diff --git a/ppocr/losses/rec_parseq_loss.py b/ppocr/losses/rec_parseq_loss.py
new file mode 100644
index 0000000..a1731e8
--- /dev/null
+++ b/ppocr/losses/rec_parseq_loss.py
@@ -0,0 +1,52 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+
+
+class ParseQLoss(nn.Layer):
+    def __init__(self, **kwargs):
+        super(ParseQLoss, self).__init__()
+
+    def forward(self, predicts, targets):
+        label = targets[1]  # label
+        label_len = targets[2]
+        max_step = paddle.max(label_len).cpu().numpy()[0] + 2
+        tgt = label[:, :max_step]
+
+        logits_list = predicts["logits_list"]
+        pad_id = predicts["pad_id"]
+        eos_id = predicts["eos_id"]
+
+        tgt_out = tgt[:, 1:]
+        loss = 0
+        loss_numel = 0
+        n = (tgt_out != pad_id).sum().item()
+
+        for i, logits in enumerate(logits_list):
+            loss += n * paddle.nn.functional.cross_entropy(
+                input=logits, label=tgt_out.flatten(), ignore_index=pad_id
+            )
+            loss_numel += n
+            if i == 1:
+                tgt_out = paddle.where(condition=tgt_out == eos_id, x=pad_id, y=tgt_out)
+                n = (tgt_out != pad_id).sum().item()
+        loss /= loss_numel
+
+        return {"loss": loss}
diff --git a/ppocr/losses/rec_ppformulanet_loss.py b/ppocr/losses/rec_ppformulanet_loss.py
new file mode 100644
index 0000000..95af575
--- /dev/null
+++ b/ppocr/losses/rec_ppformulanet_loss.py
@@ -0,0 +1,74 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import paddle
+import paddle.nn as nn
+
+
+class PPFormulaNet_S_Loss(nn.Layer):
+    """
+    PP=FormulaNet-S adopt CrossEntropyLoss for network training.
+    """
+
+    def __init__(self, vocab_size=50000, parallel_step=1):
+        super(PPFormulaNet_S_Loss, self).__init__()
+        self.ignore_index = -100
+        self.vocab_size = vocab_size
+        self.parallel_step = int(parallel_step)
+        self.pad_token_id = 1
+        # ignore padding characters during training
+        self.cross = nn.CrossEntropyLoss(
+            reduction="mean", ignore_index=self.ignore_index
+        )
+
+    def forward(self, preds, batch):
+        logits, masked_label = preds
+
+        word_loss = self.cross(
+            paddle.reshape(logits, [-1, logits.shape[-1]]),
+            paddle.reshape(masked_label[:, self.parallel_step :], [-1]),
+        )
+        loss = word_loss
+        return {
+            "loss": loss,
+            "word_loss": word_loss,
+        }
+
+
+class PPFormulaNet_L_Loss(nn.Layer):
+    """
+    PPFormulaNet_L adopt CrossEntropyLoss for network training.
+    """
+
+    def __init__(self, vocab_size=50000):
+        super(PPFormulaNet_L_Loss, self).__init__()
+        self.ignore_index = -100
+        self.vocab_size = vocab_size
+        self.pad_token_id = 1
+        # ignore padding characters during training
+        self.cross = nn.CrossEntropyLoss(
+            reduction="mean", ignore_index=self.ignore_index
+        )
+
+    def forward(self, preds, batch):
+        logits, masked_label = preds
+
+        word_loss = self.cross(
+            paddle.reshape(logits, [-1, logits.shape[-1]]),
+            paddle.reshape(masked_label[:, 1:], [-1]),
+        )
+        loss = word_loss
+        return {
+            "loss": loss,
+            "word_loss": word_loss,
+        }
diff --git a/ppocr/losses/rec_pren_loss.py b/ppocr/losses/rec_pren_loss.py
new file mode 100644
index 0000000..d08b27a
--- /dev/null
+++ b/ppocr/losses/rec_pren_loss.py
@@ -0,0 +1,30 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from paddle import nn
+
+
+class PRENLoss(nn.Layer):
+    def __init__(self, **kwargs):
+        super(PRENLoss, self).__init__()
+        # note: 0 is padding idx
+        self.loss_func = nn.CrossEntropyLoss(reduction="mean", ignore_index=0)
+
+    def forward(self, predicts, batch):
+        loss = self.loss_func(predicts, batch[1].astype("int64"))
+        return {"loss": loss}
diff --git a/ppocr/losses/rec_rfl_loss.py b/ppocr/losses/rec_rfl_loss.py
new file mode 100644
index 0000000..b2b1856
--- /dev/null
+++ b/ppocr/losses/rec_rfl_loss.py
@@ -0,0 +1,70 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/hikopensource/DAVAR-Lab-OCR/blob/main/davarocr/davar_common/models/loss/cross_entropy_loss.py
+"""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+
+from .basic_loss import CELoss, DistanceLoss
+
+
+class RFLLoss(nn.Layer):
+    def __init__(self, ignore_index=-100, **kwargs):
+        super().__init__()
+
+        self.cnt_loss = nn.MSELoss(**kwargs)
+        self.seq_loss = nn.CrossEntropyLoss(ignore_index=ignore_index)
+
+    def forward(self, predicts, batch):
+        self.total_loss = {}
+        total_loss = 0.0
+        if isinstance(predicts, tuple) or isinstance(predicts, list):
+            cnt_outputs, seq_outputs = predicts
+        else:
+            cnt_outputs, seq_outputs = predicts, None
+        # batch [image, label, length, cnt_label]
+        if cnt_outputs is not None:
+            cnt_loss = self.cnt_loss(cnt_outputs, paddle.cast(batch[3], paddle.float32))
+            self.total_loss["cnt_loss"] = cnt_loss
+            total_loss += cnt_loss
+
+        if seq_outputs is not None:
+            targets = batch[1].astype("int64")
+            label_lengths = batch[2].astype("int64")
+            batch_size, num_steps, num_classes = (
+                seq_outputs.shape[0],
+                seq_outputs.shape[1],
+                seq_outputs.shape[2],
+            )
+            assert (
+                len(targets.shape) == len(list(seq_outputs.shape)) - 1
+            ), "The target's shape and inputs's shape is [N, d] and [N, num_steps]"
+
+            inputs = seq_outputs[:, :-1, :]
+            targets = targets[:, 1:]
+
+            inputs = paddle.reshape(inputs, [-1, inputs.shape[-1]])
+            targets = paddle.reshape(targets, [-1])
+            seq_loss = self.seq_loss(inputs, targets)
+            self.total_loss["seq_loss"] = seq_loss
+            total_loss += seq_loss
+
+        self.total_loss["loss"] = total_loss
+        return self.total_loss
diff --git a/ppocr/losses/rec_sar_loss.py b/ppocr/losses/rec_sar_loss.py
new file mode 100644
index 0000000..a2cba69
--- /dev/null
+++ b/ppocr/losses/rec_sar_loss.py
@@ -0,0 +1,36 @@
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+
+
+class SARLoss(nn.Layer):
+    def __init__(self, **kwargs):
+        super(SARLoss, self).__init__()
+        ignore_index = kwargs.get("ignore_index", 92)  # 6626
+        self.loss_func = paddle.nn.loss.CrossEntropyLoss(
+            reduction="mean", ignore_index=ignore_index
+        )
+
+    def forward(self, predicts, batch):
+        predict = predicts[
+            :, :-1, :
+        ]  # ignore last index of outputs to be in same seq_len with targets
+        label = batch[1].astype("int64")[
+            :, 1:
+        ]  # ignore first index of target in loss calculation
+        batch_size, num_steps, num_classes = (
+            predict.shape[0],
+            predict.shape[1],
+            predict.shape[2],
+        )
+        assert (
+            len(label.shape) == len(list(predict.shape)) - 1
+        ), "The target's shape and inputs's shape is [N, d] and [N, num_steps]"
+
+        inputs = paddle.reshape(predict, [-1, num_classes])
+        targets = paddle.reshape(label, [-1])
+        loss = self.loss_func(inputs, targets)
+        return {"loss": loss}
diff --git a/ppocr/losses/rec_satrn_loss.py b/ppocr/losses/rec_satrn_loss.py
new file mode 100644
index 0000000..16f1b6a
--- /dev/null
+++ b/ppocr/losses/rec_satrn_loss.py
@@ -0,0 +1,53 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/1.x/mmocr/models/textrecog/module_losses/ce_module_loss.py
+"""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+
+
+class SATRNLoss(nn.Layer):
+    def __init__(self, **kwargs):
+        super(SATRNLoss, self).__init__()
+        ignore_index = kwargs.get("ignore_index", 92)  # 6626
+        self.loss_func = paddle.nn.loss.CrossEntropyLoss(
+            reduction="none", ignore_index=ignore_index
+        )
+
+    def forward(self, predicts, batch):
+        predict = predicts[
+            :, :-1, :
+        ]  # ignore last index of outputs to be in same seq_len with targets
+        label = batch[1].astype("int64")[
+            :, 1:
+        ]  # ignore first index of target in loss calculation
+        batch_size, num_steps, num_classes = (
+            predict.shape[0],
+            predict.shape[1],
+            predict.shape[2],
+        )
+        assert (
+            len(label.shape) == len(list(predict.shape)) - 1
+        ), "The target's shape and inputs's shape is [N, d] and [N, num_steps]"
+
+        inputs = paddle.reshape(predict, [-1, num_classes])
+        targets = paddle.reshape(label, [-1])
+        loss = self.loss_func(inputs, targets)
+        return {"loss": loss.mean()}
diff --git a/ppocr/losses/rec_spin_att_loss.py b/ppocr/losses/rec_spin_att_loss.py
new file mode 100644
index 0000000..591ae08
--- /dev/null
+++ b/ppocr/losses/rec_spin_att_loss.py
@@ -0,0 +1,52 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+
+"""This code is refer from:
+https://github.com/hikopensource/DAVAR-Lab-OCR
+"""
+
+
+class SPINAttentionLoss(nn.Layer):
+    def __init__(self, reduction="mean", ignore_index=-100, **kwargs):
+        super(SPINAttentionLoss, self).__init__()
+        self.loss_func = nn.CrossEntropyLoss(
+            weight=None, reduction=reduction, ignore_index=ignore_index
+        )
+
+    def forward(self, predicts, batch):
+        targets = batch[1].astype("int64")
+        targets = targets[:, 1:]  # remove [eos] in label
+
+        label_lengths = batch[2].astype("int64")
+        batch_size, num_steps, num_classes = (
+            predicts.shape[0],
+            predicts.shape[1],
+            predicts.shape[2],
+        )
+        assert (
+            len(targets.shape) == len(list(predicts.shape)) - 1
+        ), "The target's shape and inputs's shape is [N, d] and [N, num_steps]"
+
+        inputs = paddle.reshape(predicts, [-1, predicts.shape[-1]])
+        targets = paddle.reshape(targets, [-1])
+
+        return {"loss": self.loss_func(inputs, targets)}
diff --git a/ppocr/losses/rec_srn_loss.py b/ppocr/losses/rec_srn_loss.py
new file mode 100644
index 0000000..cb034f3
--- /dev/null
+++ b/ppocr/losses/rec_srn_loss.py
@@ -0,0 +1,47 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+
+
+class SRNLoss(nn.Layer):
+    def __init__(self, **kwargs):
+        super(SRNLoss, self).__init__()
+        self.loss_func = paddle.nn.loss.CrossEntropyLoss(reduction="sum")
+
+    def forward(self, predicts, batch):
+        predict = predicts["predict"]
+        word_predict = predicts["word_out"]
+        gsrm_predict = predicts["gsrm_out"]
+        label = batch[1]
+
+        casted_label = paddle.cast(x=label, dtype="int64")
+        casted_label = paddle.reshape(x=casted_label, shape=[-1, 1])
+
+        cost_word = self.loss_func(word_predict, label=casted_label)
+        cost_gsrm = self.loss_func(gsrm_predict, label=casted_label)
+        cost_vsfd = self.loss_func(predict, label=casted_label)
+
+        cost_word = paddle.reshape(x=paddle.sum(cost_word), shape=[1])
+        cost_gsrm = paddle.reshape(x=paddle.sum(cost_gsrm), shape=[1])
+        cost_vsfd = paddle.reshape(x=paddle.sum(cost_vsfd), shape=[1])
+
+        sum_cost = cost_word * 3.0 + cost_vsfd + cost_gsrm * 0.15
+
+        return {"loss": sum_cost, "word_loss": cost_word, "img_loss": cost_vsfd}
diff --git a/ppocr/losses/rec_unimernet_loss.py b/ppocr/losses/rec_unimernet_loss.py
new file mode 100644
index 0000000..2611cf8
--- /dev/null
+++ b/ppocr/losses/rec_unimernet_loss.py
@@ -0,0 +1,54 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+import numpy as np
+
+
+class UniMERNetLoss(nn.Layer):
+    def __init__(self, length_aware=True, vocab_size=50000):
+        super(UniMERNetLoss, self).__init__()
+        self.ignore_index = -100
+        self.vocab_size = vocab_size
+        self.pad_token_id = 1
+        self.length_aware = length_aware
+        self.cross = nn.CrossEntropyLoss(
+            reduction="mean", ignore_index=self.ignore_index
+        )
+        self.counting_loss_fct = nn.SmoothL1Loss()
+
+    def _get_count_gt(self, labels):
+        mask = (labels != self.pad_token_id).cast("float32")
+        one_hot_labels = F.one_hot(
+            labels, num_classes=self.vocab_size
+        ) * mask.unsqueeze(-1)
+        count_gt = paddle.sum(one_hot_labels, axis=1)
+        return count_gt
+
+    def forward(self, preds, batch):
+        logits, count_pred, masked_label = preds
+        labels = batch[1][:, 1:]
+        word_loss = self.cross(
+            paddle.reshape(logits, [-1, logits.shape[-1]]),
+            paddle.reshape(masked_label[:, 1:], [-1]),
+        )
+        loss = word_loss
+        if self.length_aware:
+            count_gt = self._get_count_gt(labels)
+            count_gt = paddle.log(count_gt.cast(paddle.float32) + 1)
+            count_loss = self.counting_loss_fct(count_pred, count_gt)
+            loss += 0.5 * count_loss
+        return {"loss": loss, "word_loss": word_loss, "count_loss": count_loss}
diff --git a/ppocr/losses/rec_vl_loss.py b/ppocr/losses/rec_vl_loss.py
new file mode 100644
index 0000000..6c41177
--- /dev/null
+++ b/ppocr/losses/rec_vl_loss.py
@@ -0,0 +1,70 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/wangyuxin87/VisionLAN
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+
+
+class VLLoss(nn.Layer):
+    def __init__(self, mode="LF_1", weight_res=0.5, weight_mas=0.5, **kwargs):
+        super(VLLoss, self).__init__()
+        self.loss_func = paddle.nn.loss.CrossEntropyLoss(reduction="mean")
+        assert mode in ["LF_1", "LF_2", "LA"]
+        self.mode = mode
+        self.weight_res = weight_res
+        self.weight_mas = weight_mas
+
+    def flatten_label(self, target):
+        label_flatten = []
+        label_length = []
+        for i in range(0, target.shape[0]):
+            cur_label = target[i].tolist()
+            label_flatten += cur_label[: cur_label.index(0) + 1]
+            label_length.append(cur_label.index(0) + 1)
+        label_flatten = paddle.to_tensor(label_flatten, dtype="int64")
+        label_length = paddle.to_tensor(label_length, dtype="int32")
+        return (label_flatten, label_length)
+
+    def _flatten(self, sources, lengths):
+        return paddle.concat([t[:l] for t, l in zip(sources, lengths)])
+
+    def forward(self, predicts, batch):
+        text_pre = predicts[0]
+        target = batch[1].astype("int64")
+        label_flatten, length = self.flatten_label(target)
+        text_pre = self._flatten(text_pre, length)
+        if self.mode == "LF_1":
+            loss = self.loss_func(text_pre, label_flatten)
+        else:
+            text_rem = predicts[1]
+            text_mas = predicts[2]
+            target_res = batch[2].astype("int64")
+            target_sub = batch[3].astype("int64")
+            label_flatten_res, length_res = self.flatten_label(target_res)
+            label_flatten_sub, length_sub = self.flatten_label(target_sub)
+            text_rem = self._flatten(text_rem, length_res)
+            text_mas = self._flatten(text_mas, length_sub)
+            loss_ori = self.loss_func(text_pre, label_flatten)
+            loss_res = self.loss_func(text_rem, label_flatten_res)
+            loss_mas = self.loss_func(text_mas, label_flatten_sub)
+            loss = loss_ori + loss_res * self.weight_res + loss_mas * self.weight_mas
+        return {"loss": loss}
diff --git a/ppocr/losses/stroke_focus_loss.py b/ppocr/losses/stroke_focus_loss.py
new file mode 100644
index 0000000..9b7850c
--- /dev/null
+++ b/ppocr/losses/stroke_focus_loss.py
@@ -0,0 +1,63 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/FudanVI/FudanOCR/blob/main/text-gestalt/loss/stroke_focus_loss.py
+"""
+import cv2
+import sys
+import time
+import string
+import random
+import numpy as np
+import paddle.nn as nn
+import paddle
+
+
+class StrokeFocusLoss(nn.Layer):
+    def __init__(self, character_dict_path=None, **kwargs):
+        super(StrokeFocusLoss, self).__init__(character_dict_path)
+        self.mse_loss = nn.MSELoss()
+        self.ce_loss = nn.CrossEntropyLoss()
+        self.l1_loss = nn.L1Loss()
+        self.english_stroke_alphabet = "0123456789"
+        self.english_stroke_dict = {}
+        for index in range(len(self.english_stroke_alphabet)):
+            self.english_stroke_dict[self.english_stroke_alphabet[index]] = index
+
+        stroke_decompose_lines = open(character_dict_path, "r").readlines()
+        self.dic = {}
+        for line in stroke_decompose_lines:
+            line = line.strip()
+            character, sequence = line.split()
+            self.dic[character] = sequence
+
+    def forward(self, pred, data):
+        sr_img = pred["sr_img"]
+        hr_img = pred["hr_img"]
+
+        mse_loss = self.mse_loss(sr_img, hr_img)
+        word_attention_map_gt = pred["word_attention_map_gt"]
+        word_attention_map_pred = pred["word_attention_map_pred"]
+
+        hr_pred = pred["hr_pred"]
+        sr_pred = pred["sr_pred"]
+
+        attention_loss = paddle.nn.functional.l1_loss(
+            word_attention_map_gt, word_attention_map_pred
+        )
+
+        loss = (mse_loss + attention_loss * 50) * 100
+
+        return {"mse_loss": mse_loss, "attention_loss": attention_loss, "loss": loss}
diff --git a/ppocr/losses/table_att_loss.py b/ppocr/losses/table_att_loss.py
new file mode 100644
index 0000000..72b737d
--- /dev/null
+++ b/ppocr/losses/table_att_loss.py
@@ -0,0 +1,100 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+from paddle.nn import functional as F
+
+
+class TableAttentionLoss(nn.Layer):
+    def __init__(self, structure_weight=1.0, loc_weight=0.0, **kwargs):
+        super(TableAttentionLoss, self).__init__()
+        self.loss_func = nn.CrossEntropyLoss(weight=None, reduction="none")
+        self.structure_weight = structure_weight
+        self.loc_weight = loc_weight
+
+    def forward(self, predicts, batch):
+        structure_probs = predicts["structure_probs"]
+        structure_targets = batch[1].astype("int64")
+        structure_targets = structure_targets[:, 1:]
+        structure_probs = paddle.reshape(
+            structure_probs, [-1, structure_probs.shape[-1]]
+        )
+        structure_targets = paddle.reshape(structure_targets, [-1])
+        structure_loss = self.loss_func(structure_probs, structure_targets)
+
+        structure_loss = paddle.mean(structure_loss) * self.structure_weight
+
+        loc_preds = predicts["loc_preds"]
+        loc_targets = batch[2].astype("float32")
+        loc_targets_mask = batch[3].astype("float32")
+        loc_targets = loc_targets[:, 1:, :]
+        loc_targets_mask = loc_targets_mask[:, 1:, :]
+        loc_loss = (
+            F.mse_loss(loc_preds * loc_targets_mask, loc_targets) * self.loc_weight
+        )
+
+        total_loss = structure_loss + loc_loss
+        return {
+            "loss": total_loss,
+            "structure_loss": structure_loss,
+            "loc_loss": loc_loss,
+        }
+
+
+class SLALoss(nn.Layer):
+    def __init__(self, structure_weight=1.0, loc_weight=0.0, loc_loss="mse", **kwargs):
+        super(SLALoss, self).__init__()
+        self.loss_func = nn.CrossEntropyLoss(weight=None, reduction="mean")
+        self.structure_weight = structure_weight
+        self.loc_weight = loc_weight
+        self.loc_loss = loc_loss
+        self.eps = 1e-12
+
+    def forward(self, predicts, batch):
+        structure_probs = predicts["structure_probs"]
+        structure_targets = batch[1].astype("int64")
+        max_len = batch[-2].max().astype("int32")
+        structure_targets = structure_targets[:, 1 : max_len + 2]
+
+        structure_loss = self.loss_func(structure_probs, structure_targets)
+
+        structure_loss = paddle.mean(structure_loss) * self.structure_weight
+
+        loc_preds = predicts["loc_preds"]
+        loc_targets = batch[2].astype("float32")
+        loc_targets_mask = batch[3].astype("float32")
+        loc_targets = loc_targets[:, 1 : max_len + 2]
+        loc_targets_mask = loc_targets_mask[:, 1 : max_len + 2]
+
+        loc_loss = (
+            F.smooth_l1_loss(
+                loc_preds * loc_targets_mask,
+                loc_targets * loc_targets_mask,
+                reduction="sum",
+            )
+            * self.loc_weight
+        )
+
+        loc_loss = loc_loss / (loc_targets_mask.sum() + self.eps)
+        total_loss = structure_loss + loc_loss
+        return {
+            "loss": total_loss,
+            "structure_loss": structure_loss,
+            "loc_loss": loc_loss,
+        }
diff --git a/ppocr/losses/table_master_loss.py b/ppocr/losses/table_master_loss.py
new file mode 100644
index 0000000..08b1c29
--- /dev/null
+++ b/ppocr/losses/table_master_loss.py
@@ -0,0 +1,74 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/JiaquanYe/TableMASTER-mmocr/tree/master/mmocr/models/textrecog/losses
+"""
+
+import paddle
+from paddle import nn
+
+
+class TableMasterLoss(nn.Layer):
+    def __init__(self, ignore_index=-1):
+        super(TableMasterLoss, self).__init__()
+        self.structure_loss = nn.CrossEntropyLoss(
+            ignore_index=ignore_index, reduction="mean"
+        )
+        self.box_loss = nn.L1Loss(reduction="sum")
+        self.eps = 1e-12
+
+    def forward(self, predicts, batch):
+        # structure_loss
+        structure_probs = predicts["structure_probs"]
+        structure_targets = batch[1]
+        structure_targets = structure_targets[:, 1:]
+        structure_probs = structure_probs.reshape([-1, structure_probs.shape[-1]])
+        structure_targets = structure_targets.reshape([-1])
+
+        structure_loss = self.structure_loss(structure_probs, structure_targets)
+        structure_loss = structure_loss.mean()
+        losses = dict(structure_loss=structure_loss)
+
+        # box loss
+        bboxes_preds = predicts["loc_preds"]
+        bboxes_targets = batch[2][:, 1:, :]
+        bbox_masks = batch[3][:, 1:]
+        # mask empty-bbox or non-bbox structure token's bbox.
+
+        masked_bboxes_preds = bboxes_preds * bbox_masks
+        masked_bboxes_targets = bboxes_targets * bbox_masks
+
+        # horizon loss (x and width)
+        horizon_sum_loss = self.box_loss(
+            masked_bboxes_preds[:, :, 0::2], masked_bboxes_targets[:, :, 0::2]
+        )
+        horizon_loss = horizon_sum_loss / (bbox_masks.sum() + self.eps)
+        # vertical loss (y and height)
+        vertical_sum_loss = self.box_loss(
+            masked_bboxes_preds[:, :, 1::2], masked_bboxes_targets[:, :, 1::2]
+        )
+        vertical_loss = vertical_sum_loss / (bbox_masks.sum() + self.eps)
+
+        horizon_loss = horizon_loss.mean()
+        vertical_loss = vertical_loss.mean()
+        all_loss = structure_loss + horizon_loss + vertical_loss
+        losses.update(
+            {
+                "loss": all_loss,
+                "horizon_bbox_loss": horizon_loss,
+                "vertical_bbox_loss": vertical_loss,
+            }
+        )
+        return losses
diff --git a/ppocr/losses/text_focus_loss.py b/ppocr/losses/text_focus_loss.py
new file mode 100644
index 0000000..310140b
--- /dev/null
+++ b/ppocr/losses/text_focus_loss.py
@@ -0,0 +1,91 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/FudanVI/FudanOCR/blob/main/scene-text-telescope/loss/text_focus_loss.py
+"""
+
+import paddle.nn as nn
+import paddle
+import numpy as np
+import pickle as pkl
+
+standard_alphebet = "-0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ"
+standard_dict = {}
+for index in range(len(standard_alphebet)):
+    standard_dict[standard_alphebet[index]] = index
+
+
+def load_confuse_matrix(confuse_dict_path):
+    f = open(confuse_dict_path, "rb")
+    data = pkl.load(f)
+    f.close()
+    number = data[:10]
+    upper = data[10:36]
+    lower = data[36:]
+    end = np.ones((1, 62))
+    pad = np.ones((63, 1))
+    rearrange_data = np.concatenate((end, number, lower, upper), axis=0)
+    rearrange_data = np.concatenate((pad, rearrange_data), axis=1)
+    rearrange_data = 1 / rearrange_data
+    rearrange_data[rearrange_data == np.inf] = 1
+    rearrange_data = paddle.to_tensor(rearrange_data)
+
+    lower_alpha = "abcdefghijklmnopqrstuvwxyz"
+    # upper_alpha = 'ABCDEFGHIJKLMNOPQRSTUVWXYZ'
+    for i in range(63):
+        for j in range(63):
+            if i != j and standard_alphebet[j] in lower_alpha:
+                rearrange_data[i][j] = max(
+                    rearrange_data[i][j], rearrange_data[i][j + 26]
+                )
+    rearrange_data = rearrange_data[:37, :37]
+
+    return rearrange_data
+
+
+def weight_cross_entropy(pred, gt, weight_table):
+    batch = gt.shape[0]
+    weight = weight_table[gt]
+    pred_exp = paddle.exp(pred)
+    pred_exp_weight = weight * pred_exp
+    loss = 0
+    for i in range(len(gt)):
+        loss -= paddle.log(
+            pred_exp_weight[i][gt[i]] / paddle.sum(pred_exp_weight, 1)[i]
+        )
+    return loss / batch
+
+
+class TelescopeLoss(nn.Layer):
+    def __init__(self, confuse_dict_path):
+        super(TelescopeLoss, self).__init__()
+        self.weight_table = load_confuse_matrix(confuse_dict_path)
+        self.mse_loss = nn.MSELoss()
+        self.ce_loss = nn.CrossEntropyLoss()
+        self.l1_loss = nn.L1Loss()
+
+    def forward(self, pred, data):
+        sr_img = pred["sr_img"]
+        hr_img = pred["hr_img"]
+        sr_pred = pred["sr_pred"]
+        text_gt = pred["text_gt"]
+
+        word_attention_map_gt = pred["word_attention_map_gt"]
+        word_attention_map_pred = pred["word_attention_map_pred"]
+        mse_loss = self.mse_loss(sr_img, hr_img)
+        attention_loss = self.l1_loss(word_attention_map_gt, word_attention_map_pred)
+        recognition_loss = weight_cross_entropy(sr_pred, text_gt, self.weight_table)
+        loss = mse_loss + attention_loss * 10 + recognition_loss * 0.0005
+        return {"mse_loss": mse_loss, "attention_loss": attention_loss, "loss": loss}
diff --git a/ppocr/losses/vqa_token_layoutlm_loss.py b/ppocr/losses/vqa_token_layoutlm_loss.py
new file mode 100755
index 0000000..d01c091
--- /dev/null
+++ b/ppocr/losses/vqa_token_layoutlm_loss.py
@@ -0,0 +1,61 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from paddle import nn
+from ppocr.losses.basic_loss import DMLLoss
+
+
+class VQASerTokenLayoutLMLoss(nn.Layer):
+    def __init__(self, num_classes, key=None):
+        super().__init__()
+        self.loss_class = nn.CrossEntropyLoss()
+        self.num_classes = num_classes
+        self.ignore_index = self.loss_class.ignore_index
+        self.key = key
+
+    def forward(self, predicts, batch):
+        if isinstance(predicts, dict) and self.key is not None:
+            predicts = predicts[self.key]
+        labels = batch[5]
+        attention_mask = batch[2]
+        if attention_mask is not None:
+            active_loss = (
+                attention_mask.reshape(
+                    [
+                        -1,
+                    ]
+                )
+                == 1
+            )
+            active_output = predicts.reshape([-1, self.num_classes])[active_loss]
+            active_label = labels.reshape(
+                [
+                    -1,
+                ]
+            )[active_loss]
+            loss = self.loss_class(active_output, active_label)
+        else:
+            loss = self.loss_class(
+                predicts.reshape([-1, self.num_classes]),
+                labels.reshape(
+                    [
+                        -1,
+                    ]
+                ),
+            )
+        return {"loss": loss}
diff --git a/ppocr/metrics/__init__.py b/ppocr/metrics/__init__.py
new file mode 100644
index 0000000..dd28d73
--- /dev/null
+++ b/ppocr/metrics/__init__.py
@@ -0,0 +1,62 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+import copy
+
+__all__ = ["build_metric"]
+
+from .det_metric import DetMetric, DetFCEMetric
+from .rec_metric import RecMetric, CNTMetric, CANMetric, LaTeXOCRMetric
+from .cls_metric import ClsMetric
+from .e2e_metric import E2EMetric
+from .distillation_metric import DistillationMetric
+from .table_metric import TableMetric
+from .kie_metric import KIEMetric
+from .vqa_token_ser_metric import VQASerTokenMetric
+from .vqa_token_re_metric import VQAReTokenMetric
+from .sr_metric import SRMetric
+from .ct_metric import CTMetric
+
+
+def build_metric(config):
+    support_dict = [
+        "DetMetric",
+        "DetFCEMetric",
+        "RecMetric",
+        "ClsMetric",
+        "E2EMetric",
+        "DistillationMetric",
+        "TableMetric",
+        "KIEMetric",
+        "VQASerTokenMetric",
+        "VQAReTokenMetric",
+        "SRMetric",
+        "CTMetric",
+        "CNTMetric",
+        "CANMetric",
+        "LaTeXOCRMetric",
+    ]
+
+    config = copy.deepcopy(config)
+    module_name = config.pop("name")
+    assert module_name in support_dict, Exception(
+        "metric only support {}".format(support_dict)
+    )
+    module_class = eval(module_name)(**config)
+    return module_class
diff --git a/ppocr/metrics/bleu.py b/ppocr/metrics/bleu.py
new file mode 100644
index 0000000..c68e624
--- /dev/null
+++ b/ppocr/metrics/bleu.py
@@ -0,0 +1,243 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+This code is refer from:
+https://github.com/tensorflow/nmt/blob/master/nmt/scripts/bleu.py
+"""
+
+import re
+import math
+import collections
+from functools import lru_cache
+
+
+def _get_ngrams(segment, max_order):
+    """Extracts all n-grams upto a given maximum order from an input segment.
+
+    Args:
+      segment: text segment from which n-grams will be extracted.
+      max_order: maximum length in tokens of the n-grams returned by this
+          methods.
+
+    Returns:
+      The Counter containing all n-grams upto max_order in segment
+      with a count of how many times each n-gram occurred.
+    """
+    ngram_counts = collections.Counter()
+    for order in range(1, max_order + 1):
+        for i in range(0, len(segment) - order + 1):
+            ngram = tuple(segment[i : i + order])
+            ngram_counts[ngram] += 1
+    return ngram_counts
+
+
+def compute_bleu(reference_corpus, translation_corpus, max_order=4, smooth=False):
+    """Computes BLEU score of translated segments against one or more references.
+
+    Args:
+      reference_corpus: list of lists of references for each translation. Each
+          reference should be tokenized into a list of tokens.
+      translation_corpus: list of translations to score. Each translation
+          should be tokenized into a list of tokens.
+      max_order: Maximum n-gram order to use when computing BLEU score.
+      smooth: Whether or not to apply Lin et al. 2004 smoothing.
+
+    Returns:
+      3-Tuple with the BLEU score, n-gram precisions, geometric mean of n-gram
+      precisions and brevity penalty.
+    """
+    matches_by_order = [0] * max_order
+    possible_matches_by_order = [0] * max_order
+    reference_length = 0
+    translation_length = 0
+    for references, translation in zip(reference_corpus, translation_corpus):
+        reference_length += min(len(r) for r in references)
+        translation_length += len(translation)
+
+        merged_ref_ngram_counts = collections.Counter()
+        for reference in references:
+            merged_ref_ngram_counts |= _get_ngrams(reference, max_order)
+        translation_ngram_counts = _get_ngrams(translation, max_order)
+        overlap = translation_ngram_counts & merged_ref_ngram_counts
+        for ngram in overlap:
+            matches_by_order[len(ngram) - 1] += overlap[ngram]
+        for order in range(1, max_order + 1):
+            possible_matches = len(translation) - order + 1
+            if possible_matches > 0:
+                possible_matches_by_order[order - 1] += possible_matches
+
+    precisions = [0] * max_order
+    for i in range(0, max_order):
+        if smooth:
+            precisions[i] = (matches_by_order[i] + 1.0) / (
+                possible_matches_by_order[i] + 1.0
+            )
+        else:
+            if possible_matches_by_order[i] > 0:
+                precisions[i] = (
+                    float(matches_by_order[i]) / possible_matches_by_order[i]
+                )
+            else:
+                precisions[i] = 0.0
+
+    if min(precisions) > 0:
+        p_log_sum = sum((1.0 / max_order) * math.log(p) for p in precisions)
+        geo_mean = math.exp(p_log_sum)
+    else:
+        geo_mean = 0
+
+    if float(translation_length) == 0 or float(reference_length) == 0:
+        ratio = 1e-5
+    else:
+        ratio = float(translation_length) / reference_length
+
+    if ratio > 1.0:
+        bp = 1.0
+    else:
+        bp = math.exp(1 - 1.0 / ratio)
+
+    bleu = geo_mean * bp
+
+    return (bleu, precisions, bp, ratio, translation_length, reference_length)
+
+
+class BaseTokenizer:
+    """A base dummy tokenizer to derive from."""
+
+    def signature(self):
+        """
+        Returns a signature for the tokenizer.
+        :return: signature string
+        """
+        return "none"
+
+    def __call__(self, line):
+        """
+        Tokenizes an input line with the tokenizer.
+        :param line: a segment to tokenize
+        :return: the tokenized line
+        """
+        return line
+
+
+class TokenizerRegexp(BaseTokenizer):
+    def signature(self):
+        return "re"
+
+    def __init__(self):
+        self._re = [
+            # language-dependent part (assuming Western languages)
+            (re.compile(r"([\{-\~\[-\` -\&\(-\+\:-\@\/])"), r" \1 "),
+            # tokenize period and comma unless preceded by a digit
+            (re.compile(r"([^0-9])([\.,])"), r"\1 \2 "),
+            # tokenize period and comma unless followed by a digit
+            (re.compile(r"([\.,])([^0-9])"), r" \1 \2"),
+            # tokenize dash when preceded by a digit
+            (re.compile(r"([0-9])(-)"), r"\1 \2 "),
+            # one space only between words
+            # NOTE: Doing this in Python (below) is faster
+            # (re.compile(r'\s+'), r' '),
+        ]
+
+    @lru_cache(maxsize=2**16)
+    def __call__(self, line):
+        """Common post-processing tokenizer for `13a` and `zh` tokenizers.
+        :param line: a segment to tokenize
+        :return: the tokenized line
+        """
+        for _re, repl in self._re:
+            line = _re.sub(repl, line)
+
+        # no leading or trailing spaces, single space within words
+        # return ' '.join(line.split())
+        # This line is changed with regards to the original tokenizer (seen above) to return individual words
+        return line.split()
+
+
+class Tokenizer13a(BaseTokenizer):
+    def signature(self):
+        return "13a"
+
+    def __init__(self):
+        self._post_tokenizer = TokenizerRegexp()
+
+    @lru_cache(maxsize=2**16)
+    def __call__(self, line):
+        """Tokenizes an input line using a relatively minimal tokenization
+        that is however equivalent to mteval-v13a, used by WMT.
+
+        :param line: a segment to tokenize
+        :return: the tokenized line
+        """
+
+        # language-independent part:
+        line = line.replace("<skipped>", "")
+        line = line.replace("-\n", "")
+        line = line.replace("\n", " ")
+
+        if "&" in line:
+            line = line.replace("&quot;", '"')
+            line = line.replace("&amp;", "&")
+            line = line.replace("&lt;", "<")
+            line = line.replace("&gt;", ">")
+
+        return self._post_tokenizer(f" {line} ")
+
+
+def compute_bleu_score(
+    predictions, references, tokenizer=Tokenizer13a(), max_order=4, smooth=False
+):
+    # if only one reference is provided make sure we still use list of lists
+    if isinstance(references[0], str):
+        references = [[ref] for ref in references]
+
+    references = [[tokenizer(r) for r in ref] for ref in references]
+    predictions = [tokenizer(p) for p in predictions]
+    score = compute_bleu(
+        reference_corpus=references,
+        translation_corpus=predictions,
+        max_order=max_order,
+        smooth=smooth,
+    )
+    (bleu, precisions, bp, ratio, translation_length, reference_length) = score
+    return bleu
+
+
+def cal_distance(word1, word2):
+    m = len(word1)
+    n = len(word2)
+    if m * n == 0:
+        return m + n
+    dp = [[0] * (n + 1) for _ in range(m + 1)]
+    for i in range(m + 1):
+        dp[i][0] = i
+    for j in range(n + 1):
+        dp[0][j] = j
+    for i in range(1, m + 1):
+        for j in range(1, n + 1):
+            a = dp[i - 1][j] + 1
+            b = dp[i][j - 1] + 1
+            c = dp[i - 1][j - 1]
+            if word1[i - 1] != word2[j - 1]:
+                c += 1
+            dp[i][j] = min(a, b, c)
+    return dp[m][n]
+
+
+def compute_edit_distance(prediction, label):
+    prediction = prediction.strip().split(" ")
+    label = label.strip().split(" ")
+    distance = cal_distance(prediction, label)
+    return distance
diff --git a/ppocr/metrics/cls_metric.py b/ppocr/metrics/cls_metric.py
new file mode 100644
index 0000000..820fe75
--- /dev/null
+++ b/ppocr/metrics/cls_metric.py
@@ -0,0 +1,48 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+class ClsMetric(object):
+    def __init__(self, main_indicator="acc", **kwargs):
+        self.main_indicator = main_indicator
+        self.eps = 1e-5
+        self.reset()
+
+    def __call__(self, pred_label, *args, **kwargs):
+        preds, labels = pred_label
+        correct_num = 0
+        all_num = 0
+        for (pred, pred_conf), (target, _) in zip(preds, labels):
+            if pred == target:
+                correct_num += 1
+            all_num += 1
+        self.correct_num += correct_num
+        self.all_num += all_num
+        return {
+            "acc": correct_num / (all_num + self.eps),
+        }
+
+    def get_metric(self):
+        """
+        return metrics {
+                 'acc': 0
+            }
+        """
+        acc = self.correct_num / (self.all_num + self.eps)
+        self.reset()
+        return {"acc": acc}
+
+    def reset(self):
+        self.correct_num = 0
+        self.all_num = 0
diff --git a/ppocr/metrics/ct_metric.py b/ppocr/metrics/ct_metric.py
new file mode 100644
index 0000000..bec3506
--- /dev/null
+++ b/ppocr/metrics/ct_metric.py
@@ -0,0 +1,51 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+from scipy import io
+import numpy as np
+
+from ppocr.utils.e2e_metric.Deteval import combine_results, get_score_C
+
+
+class CTMetric(object):
+    def __init__(self, main_indicator, delimiter="\t", **kwargs):
+        self.delimiter = delimiter
+        self.main_indicator = main_indicator
+        self.reset()
+
+    def reset(self):
+        self.results = []  # clear results
+
+    def __call__(self, preds, batch, **kwargs):
+        # NOTE: only support bs=1 now, as the label length of different sample is Unequal
+        assert len(preds) == 1, "CentripetalText test now only support batch_size=1."
+        label = batch[2]
+        text = batch[3]
+        pred = preds[0]["points"]
+        result = get_score_C(label, text, pred)
+
+        self.results.append(result)
+
+    def get_metric(self):
+        """
+        Input format: y0,x0, ..... yn,xn. Each detection is separated by the end of line token ('\n')'
+        """
+        metrics = combine_results(self.results, rec_flag=False)
+        self.reset()
+        return metrics
diff --git a/ppocr/metrics/det_metric.py b/ppocr/metrics/det_metric.py
new file mode 100644
index 0000000..be95ec3
--- /dev/null
+++ b/ppocr/metrics/det_metric.py
@@ -0,0 +1,153 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+__all__ = ["DetMetric", "DetFCEMetric"]
+
+from .eval_det_iou import DetectionIoUEvaluator
+
+
+class DetMetric(object):
+    def __init__(self, main_indicator="hmean", **kwargs):
+        self.evaluator = DetectionIoUEvaluator()
+        self.main_indicator = main_indicator
+        self.reset()
+
+    def __call__(self, preds, batch, **kwargs):
+        """
+        batch: a list produced by dataloaders.
+            image: np.ndarray  of shape (N, C, H, W).
+            ratio_list: np.ndarray  of shape(N,2)
+            polygons: np.ndarray  of shape (N, K, 4, 2), the polygons of objective regions.
+            ignore_tags: np.ndarray  of shape (N, K), indicates whether a region is ignorable or not.
+        preds: a list of dict produced by post process
+             points: np.ndarray of shape (N, K, 4, 2), the polygons of objective regions.
+        """
+        gt_polyons_batch = batch[2]
+        ignore_tags_batch = batch[3]
+        for pred, gt_polyons, ignore_tags in zip(
+            preds, gt_polyons_batch, ignore_tags_batch
+        ):
+            # prepare gt
+            gt_info_list = [
+                {"points": gt_polyon, "text": "", "ignore": ignore_tag}
+                for gt_polyon, ignore_tag in zip(gt_polyons, ignore_tags)
+            ]
+            # prepare det
+            det_info_list = [
+                {"points": det_polyon, "text": ""} for det_polyon in pred["points"]
+            ]
+            result = self.evaluator.evaluate_image(gt_info_list, det_info_list)
+            self.results.append(result)
+
+    def get_metric(self):
+        """
+        return metrics {
+                 'precision': 0,
+                 'recall': 0,
+                 'hmean': 0
+            }
+        """
+
+        metrics = self.evaluator.combine_results(self.results)
+        self.reset()
+        return metrics
+
+    def reset(self):
+        self.results = []  # clear results
+
+
+class DetFCEMetric(object):
+    def __init__(self, main_indicator="hmean", **kwargs):
+        self.evaluator = DetectionIoUEvaluator()
+        self.main_indicator = main_indicator
+        self.reset()
+
+    def __call__(self, preds, batch, **kwargs):
+        """
+        batch: a list produced by dataloaders.
+            image: np.ndarray  of shape (N, C, H, W).
+            ratio_list: np.ndarray  of shape(N,2)
+            polygons: np.ndarray  of shape (N, K, 4, 2), the polygons of objective regions.
+            ignore_tags: np.ndarray  of shape (N, K), indicates whether a region is ignorable or not.
+        preds: a list of dict produced by post process
+             points: np.ndarray of shape (N, K, 4, 2), the polygons of objective regions.
+        """
+        gt_polyons_batch = batch[2]
+        ignore_tags_batch = batch[3]
+
+        for pred, gt_polyons, ignore_tags in zip(
+            preds, gt_polyons_batch, ignore_tags_batch
+        ):
+            # prepare gt
+            gt_info_list = [
+                {"points": gt_polyon, "text": "", "ignore": ignore_tag}
+                for gt_polyon, ignore_tag in zip(gt_polyons, ignore_tags)
+            ]
+            # prepare det
+            det_info_list = [
+                {"points": det_polyon, "text": "", "score": score}
+                for det_polyon, score in zip(pred["points"], pred["scores"])
+            ]
+
+            for score_thr in self.results.keys():
+                det_info_list_thr = [
+                    det_info
+                    for det_info in det_info_list
+                    if det_info["score"] >= score_thr
+                ]
+                result = self.evaluator.evaluate_image(gt_info_list, det_info_list_thr)
+                self.results[score_thr].append(result)
+
+    def get_metric(self):
+        """
+        return metrics {'heman':0,
+            'thr 0.3':'precision: 0 recall: 0 hmean: 0',
+            'thr 0.4':'precision: 0 recall: 0 hmean: 0',
+            'thr 0.5':'precision: 0 recall: 0 hmean: 0',
+            'thr 0.6':'precision: 0 recall: 0 hmean: 0',
+            'thr 0.7':'precision: 0 recall: 0 hmean: 0',
+            'thr 0.8':'precision: 0 recall: 0 hmean: 0',
+            'thr 0.9':'precision: 0 recall: 0 hmean: 0',
+            }
+        """
+        metrics = {}
+        hmean = 0
+        for score_thr in self.results.keys():
+            metric = self.evaluator.combine_results(self.results[score_thr])
+            # for key, value in metric.items():
+            #     metrics['{}_{}'.format(key, score_thr)] = value
+            metric_str = "precision:{:.5f} recall:{:.5f} hmean:{:.5f}".format(
+                metric["precision"], metric["recall"], metric["hmean"]
+            )
+            metrics["thr {}".format(score_thr)] = metric_str
+            hmean = max(hmean, metric["hmean"])
+        metrics["hmean"] = hmean
+
+        self.reset()
+        return metrics
+
+    def reset(self):
+        self.results = {
+            0.3: [],
+            0.4: [],
+            0.5: [],
+            0.6: [],
+            0.7: [],
+            0.8: [],
+            0.9: [],
+        }  # clear results
diff --git a/ppocr/metrics/distillation_metric.py b/ppocr/metrics/distillation_metric.py
new file mode 100644
index 0000000..8e0bcf1
--- /dev/null
+++ b/ppocr/metrics/distillation_metric.py
@@ -0,0 +1,72 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import importlib
+import copy
+
+from .rec_metric import RecMetric
+from .det_metric import DetMetric
+from .e2e_metric import E2EMetric
+from .cls_metric import ClsMetric
+from .vqa_token_ser_metric import VQASerTokenMetric
+from .vqa_token_re_metric import VQAReTokenMetric
+
+
+class DistillationMetric(object):
+    def __init__(self, key=None, base_metric_name=None, main_indicator=None, **kwargs):
+        self.main_indicator = main_indicator
+        self.key = key
+        self.main_indicator = main_indicator
+        self.base_metric_name = base_metric_name
+        self.kwargs = kwargs
+        self.metrics = None
+
+    def _init_metrcis(self, preds):
+        self.metrics = dict()
+        mod = importlib.import_module(__name__)
+        for key in preds:
+            self.metrics[key] = getattr(mod, self.base_metric_name)(
+                main_indicator=self.main_indicator, **self.kwargs
+            )
+            self.metrics[key].reset()
+
+    def __call__(self, preds, batch, **kwargs):
+        assert isinstance(preds, dict)
+        if self.metrics is None:
+            self._init_metrcis(preds)
+        output = dict()
+        for key in preds:
+            self.metrics[key].__call__(preds[key], batch, **kwargs)
+
+    def get_metric(self):
+        """
+        return metrics {
+                 'acc': 0,
+                 'norm_edit_dis': 0,
+            }
+        """
+        output = dict()
+        for key in self.metrics:
+            metric = self.metrics[key].get_metric()
+            # main indicator
+            if key == self.key:
+                output.update(metric)
+            else:
+                for sub_key in metric:
+                    output["{}_{}".format(key, sub_key)] = metric[sub_key]
+        return output
+
+    def reset(self):
+        for key in self.metrics:
+            self.metrics[key].reset()
diff --git a/ppocr/metrics/e2e_metric.py b/ppocr/metrics/e2e_metric.py
new file mode 100644
index 0000000..97796c4
--- /dev/null
+++ b/ppocr/metrics/e2e_metric.py
@@ -0,0 +1,88 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+__all__ = ["E2EMetric"]
+
+from ppocr.utils.e2e_metric.Deteval import get_socre_A, get_socre_B, combine_results
+from ppocr.utils.e2e_utils.extract_textpoint_slow import get_dict
+
+
+class E2EMetric(object):
+    def __init__(
+        self,
+        mode,
+        gt_mat_dir,
+        character_dict_path,
+        main_indicator="f_score_e2e",
+        **kwargs,
+    ):
+        self.mode = mode
+        self.gt_mat_dir = gt_mat_dir
+        self.label_list = get_dict(character_dict_path)
+        self.max_index = len(self.label_list)
+        self.main_indicator = main_indicator
+        self.reset()
+
+    def __call__(self, preds, batch, **kwargs):
+        if self.mode == "A":
+            gt_polyons_batch = batch[2]
+            temp_gt_strs_batch = batch[3][0]
+            ignore_tags_batch = batch[4]
+            gt_strs_batch = []
+
+            for temp_list in temp_gt_strs_batch:
+                t = ""
+                for index in temp_list:
+                    if index < self.max_index:
+                        t += self.label_list[index]
+                gt_strs_batch.append(t)
+
+            for pred, gt_polyons, gt_strs, ignore_tags in zip(
+                [preds], gt_polyons_batch, [gt_strs_batch], ignore_tags_batch
+            ):
+                # prepare gt
+                gt_info_list = [
+                    {"points": gt_polyon, "text": gt_str, "ignore": ignore_tag}
+                    for gt_polyon, gt_str, ignore_tag in zip(
+                        gt_polyons, gt_strs, ignore_tags
+                    )
+                ]
+                # prepare det
+                e2e_info_list = [
+                    {"points": det_polyon, "texts": pred_str}
+                    for det_polyon, pred_str in zip(pred["points"], pred["texts"])
+                ]
+
+                result = get_socre_A(gt_info_list, e2e_info_list)
+                self.results.append(result)
+        else:
+            img_id = batch[5][0]
+            e2e_info_list = [
+                {"points": det_polyon, "texts": pred_str}
+                for det_polyon, pred_str in zip(preds["points"], preds["texts"])
+            ]
+            result = get_socre_B(self.gt_mat_dir, img_id, e2e_info_list)
+            self.results.append(result)
+
+    def get_metric(self):
+        metrics = combine_results(self.results)
+        self.reset()
+        return metrics
+
+    def reset(self):
+        self.results = []  # clear results
diff --git a/ppocr/metrics/eval_det_iou.py b/ppocr/metrics/eval_det_iou.py
new file mode 100644
index 0000000..21d5c18
--- /dev/null
+++ b/ppocr/metrics/eval_det_iou.py
@@ -0,0 +1,257 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+from collections import namedtuple
+import numpy as np
+from shapely.geometry import Polygon
+
+"""
+reference from :
+https://github.com/MhLiao/DB/blob/3c32b808d4412680310d3d28eeb6a2d5bf1566c5/concern/icdar2015_eval/detection/iou.py#L8
+"""
+
+
+class DetectionIoUEvaluator(object):
+    def __init__(self, iou_constraint=0.5, area_precision_constraint=0.5):
+        self.iou_constraint = iou_constraint
+        self.area_precision_constraint = area_precision_constraint
+
+    def evaluate_image(self, gt, pred):
+        def get_union(pD, pG):
+            return Polygon(pD).union(Polygon(pG)).area
+
+        def get_intersection_over_union(pD, pG):
+            return get_intersection(pD, pG) / get_union(pD, pG)
+
+        def get_intersection(pD, pG):
+            return Polygon(pD).intersection(Polygon(pG)).area
+
+        def compute_ap(confList, matchList, numGtCare):
+            correct = 0
+            AP = 0
+            if len(confList) > 0:
+                confList = np.array(confList)
+                matchList = np.array(matchList)
+                sorted_ind = np.argsort(-confList)
+                confList = confList[sorted_ind]
+                matchList = matchList[sorted_ind]
+                for n in range(len(confList)):
+                    match = matchList[n]
+                    if match:
+                        correct += 1
+                        AP += float(correct) / (n + 1)
+
+                if numGtCare > 0:
+                    AP /= numGtCare
+
+            return AP
+
+        perSampleMetrics = {}
+
+        matchedSum = 0
+
+        Rectangle = namedtuple("Rectangle", "xmin ymin xmax ymax")
+
+        numGlobalCareGt = 0
+        numGlobalCareDet = 0
+
+        arrGlobalConfidences = []
+        arrGlobalMatches = []
+
+        recall = 0
+        precision = 0
+        hmean = 0
+
+        detMatched = 0
+
+        iouMat = np.empty([1, 1])
+
+        gtPols = []
+        detPols = []
+
+        gtPolPoints = []
+        detPolPoints = []
+
+        # Array of Ground Truth Polygons' keys marked as don't Care
+        gtDontCarePolsNum = []
+        # Array of Detected Polygons' matched with a don't Care GT
+        detDontCarePolsNum = []
+
+        pairs = []
+        detMatchedNums = []
+
+        arrSampleConfidences = []
+        arrSampleMatch = []
+
+        evaluationLog = ""
+
+        for n in range(len(gt)):
+            points = gt[n]["points"]
+            dontCare = gt[n]["ignore"]
+            if not Polygon(points).is_valid:
+                continue
+
+            gtPol = points
+            gtPols.append(gtPol)
+            gtPolPoints.append(points)
+            if dontCare:
+                gtDontCarePolsNum.append(len(gtPols) - 1)
+
+        evaluationLog += (
+            "GT polygons: "
+            + str(len(gtPols))
+            + (
+                " (" + str(len(gtDontCarePolsNum)) + " don't care)\n"
+                if len(gtDontCarePolsNum) > 0
+                else "\n"
+            )
+        )
+
+        for n in range(len(pred)):
+            points = pred[n]["points"]
+            if not Polygon(points).is_valid:
+                continue
+
+            detPol = points
+            detPols.append(detPol)
+            detPolPoints.append(points)
+            if len(gtDontCarePolsNum) > 0:
+                for dontCarePol in gtDontCarePolsNum:
+                    dontCarePol = gtPols[dontCarePol]
+                    intersected_area = get_intersection(dontCarePol, detPol)
+                    pdDimensions = Polygon(detPol).area
+                    precision = (
+                        0 if pdDimensions == 0 else intersected_area / pdDimensions
+                    )
+                    if precision > self.area_precision_constraint:
+                        detDontCarePolsNum.append(len(detPols) - 1)
+                        break
+
+        evaluationLog += (
+            "DET polygons: "
+            + str(len(detPols))
+            + (
+                " (" + str(len(detDontCarePolsNum)) + " don't care)\n"
+                if len(detDontCarePolsNum) > 0
+                else "\n"
+            )
+        )
+
+        if len(gtPols) > 0 and len(detPols) > 0:
+            # Calculate IoU and precision matrixs
+            outputShape = [len(gtPols), len(detPols)]
+            iouMat = np.empty(outputShape)
+            gtRectMat = np.zeros(len(gtPols), np.int8)
+            detRectMat = np.zeros(len(detPols), np.int8)
+            for gtNum in range(len(gtPols)):
+                for detNum in range(len(detPols)):
+                    pG = gtPols[gtNum]
+                    pD = detPols[detNum]
+                    iouMat[gtNum, detNum] = get_intersection_over_union(pD, pG)
+
+            for gtNum in range(len(gtPols)):
+                for detNum in range(len(detPols)):
+                    if (
+                        gtRectMat[gtNum] == 0
+                        and detRectMat[detNum] == 0
+                        and gtNum not in gtDontCarePolsNum
+                        and detNum not in detDontCarePolsNum
+                    ):
+                        if iouMat[gtNum, detNum] > self.iou_constraint:
+                            gtRectMat[gtNum] = 1
+                            detRectMat[detNum] = 1
+                            detMatched += 1
+                            pairs.append({"gt": gtNum, "det": detNum})
+                            detMatchedNums.append(detNum)
+                            evaluationLog += (
+                                "Match GT #"
+                                + str(gtNum)
+                                + " with Det #"
+                                + str(detNum)
+                                + "\n"
+                            )
+
+        numGtCare = len(gtPols) - len(gtDontCarePolsNum)
+        numDetCare = len(detPols) - len(detDontCarePolsNum)
+        if numGtCare == 0:
+            recall = float(1)
+            precision = float(0) if numDetCare > 0 else float(1)
+        else:
+            recall = float(detMatched) / numGtCare
+            precision = 0 if numDetCare == 0 else float(detMatched) / numDetCare
+
+        hmean = (
+            0
+            if (precision + recall) == 0
+            else 2.0 * precision * recall / (precision + recall)
+        )
+
+        matchedSum += detMatched
+        numGlobalCareGt += numGtCare
+        numGlobalCareDet += numDetCare
+
+        perSampleMetrics = {
+            "gtCare": numGtCare,
+            "detCare": numDetCare,
+            "detMatched": detMatched,
+        }
+        return perSampleMetrics
+
+    def combine_results(self, results):
+        numGlobalCareGt = 0
+        numGlobalCareDet = 0
+        matchedSum = 0
+        for result in results:
+            numGlobalCareGt += result["gtCare"]
+            numGlobalCareDet += result["detCare"]
+            matchedSum += result["detMatched"]
+
+        methodRecall = (
+            0 if numGlobalCareGt == 0 else float(matchedSum) / numGlobalCareGt
+        )
+        methodPrecision = (
+            0 if numGlobalCareDet == 0 else float(matchedSum) / numGlobalCareDet
+        )
+        methodHmean = (
+            0
+            if methodRecall + methodPrecision == 0
+            else 2 * methodRecall * methodPrecision / (methodRecall + methodPrecision)
+        )
+        methodMetrics = {
+            "precision": methodPrecision,
+            "recall": methodRecall,
+            "hmean": methodHmean,
+        }
+
+        return methodMetrics
+
+
+if __name__ == "__main__":
+    evaluator = DetectionIoUEvaluator()
+    gts = [
+        [
+            {
+                "points": [(0, 0), (1, 0), (1, 1), (0, 1)],
+                "text": 1234,
+                "ignore": False,
+            },
+            {
+                "points": [(2, 2), (3, 2), (3, 3), (2, 3)],
+                "text": 5678,
+                "ignore": False,
+            },
+        ]
+    ]
+    preds = [
+        [
+            {
+                "points": [(0.1, 0.1), (1, 0), (1, 1), (0, 1)],
+                "text": 123,
+                "ignore": False,
+            }
+        ]
+    ]
+    results = []
+    for gt, pred in zip(gts, preds):
+        results.append(evaluator.evaluate_image(gt, pred))
+    metrics = evaluator.combine_results(results)
+    print(metrics)
diff --git a/ppocr/metrics/kie_metric.py b/ppocr/metrics/kie_metric.py
new file mode 100644
index 0000000..0c83756
--- /dev/null
+++ b/ppocr/metrics/kie_metric.py
@@ -0,0 +1,72 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# The code is refer from: https://github.com/open-mmlab/mmocr/blob/main/mmocr/core/evaluation/kie_metric.py
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+import paddle
+
+__all__ = ["KIEMetric"]
+
+
+class KIEMetric(object):
+    def __init__(self, main_indicator="hmean", **kwargs):
+        self.main_indicator = main_indicator
+        self.reset()
+        self.node = []
+        self.gt = []
+
+    def __call__(self, preds, batch, **kwargs):
+        nodes, _ = preds
+        gts, tag = batch[4].squeeze(0), batch[5].tolist()[0]
+        gts = gts[: tag[0], :1].reshape([-1])
+        self.node.append(nodes.numpy())
+        self.gt.append(gts)
+        # result = self.compute_f1_score(nodes, gts)
+        # self.results.append(result)
+
+    def compute_f1_score(self, preds, gts):
+        ignores = [0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 25]
+        C = preds.shape[1]
+        classes = np.array(sorted(set(range(C)) - set(ignores)))
+        hist = (
+            np.bincount((gts * C).astype("int64") + preds.argmax(1), minlength=C**2)
+            .reshape([C, C])
+            .astype("float32")
+        )
+        diag = np.diag(hist)
+        recalls = diag / hist.sum(1).clip(min=1)
+        precisions = diag / hist.sum(0).clip(min=1)
+        f1 = 2 * recalls * precisions / (recalls + precisions).clip(min=1e-8)
+        return f1[classes]
+
+    def combine_results(self, results):
+        node = np.concatenate(self.node, 0)
+        gts = np.concatenate(self.gt, 0)
+        results = self.compute_f1_score(node, gts)
+        data = {"hmean": results.mean()}
+        return data
+
+    def get_metric(self):
+        metrics = self.combine_results(self.results)
+        self.reset()
+        return metrics
+
+    def reset(self):
+        self.results = []  # clear results
+        self.node = []
+        self.gt = []
diff --git a/ppocr/metrics/rec_metric.py b/ppocr/metrics/rec_metric.py
new file mode 100644
index 0000000..eef0a9a
--- /dev/null
+++ b/ppocr/metrics/rec_metric.py
@@ -0,0 +1,297 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from rapidfuzz.distance import Levenshtein
+from difflib import SequenceMatcher
+
+import numpy as np
+import string
+from .bleu import compute_bleu_score, compute_edit_distance
+
+
+class RecMetric(object):
+    def __init__(
+        self, main_indicator="acc", is_filter=False, ignore_space=True, **kwargs
+    ):
+        self.main_indicator = main_indicator
+        self.is_filter = is_filter
+        self.ignore_space = ignore_space
+        self.eps = 1e-5
+        self.reset()
+
+    def _normalize_text(self, text):
+        text = "".join(
+            filter(lambda x: x in (string.digits + string.ascii_letters), text)
+        )
+        return text.lower()
+
+    def __call__(self, pred_label, *args, **kwargs):
+        preds, labels = pred_label
+        correct_num = 0
+        all_num = 0
+        norm_edit_dis = 0.0
+        for (pred, pred_conf), (target, _) in zip(preds, labels):
+            if self.ignore_space:
+                pred = pred.replace(" ", "")
+                target = target.replace(" ", "")
+            if self.is_filter:
+                pred = self._normalize_text(pred)
+                target = self._normalize_text(target)
+            norm_edit_dis += Levenshtein.normalized_distance(pred, target)
+            if pred == target:
+                correct_num += 1
+            all_num += 1
+        self.correct_num += correct_num
+        self.all_num += all_num
+        self.norm_edit_dis += norm_edit_dis
+        return {
+            "acc": correct_num / (all_num + self.eps),
+            "norm_edit_dis": 1 - norm_edit_dis / (all_num + self.eps),
+        }
+
+    def get_metric(self):
+        """
+        return metrics {
+                 'acc': 0,
+                 'norm_edit_dis': 0,
+            }
+        """
+        acc = 1.0 * self.correct_num / (self.all_num + self.eps)
+        norm_edit_dis = 1 - self.norm_edit_dis / (self.all_num + self.eps)
+        self.reset()
+        return {"acc": acc, "norm_edit_dis": norm_edit_dis}
+
+    def reset(self):
+        self.correct_num = 0
+        self.all_num = 0
+        self.norm_edit_dis = 0
+
+
+class CNTMetric(object):
+    def __init__(self, main_indicator="acc", **kwargs):
+        self.main_indicator = main_indicator
+        self.eps = 1e-5
+        self.reset()
+
+    def __call__(self, pred_label, *args, **kwargs):
+        preds, labels = pred_label
+        correct_num = 0
+        all_num = 0
+        for pred, target in zip(preds, labels):
+            if pred == target:
+                correct_num += 1
+            all_num += 1
+        self.correct_num += correct_num
+        self.all_num += all_num
+        return {
+            "acc": correct_num / (all_num + self.eps),
+        }
+
+    def get_metric(self):
+        """
+        return metrics {
+                 'acc': 0,
+            }
+        """
+        acc = 1.0 * self.correct_num / (self.all_num + self.eps)
+        self.reset()
+        return {"acc": acc}
+
+    def reset(self):
+        self.correct_num = 0
+        self.all_num = 0
+
+
+class CANMetric(object):
+    def __init__(self, main_indicator="exp_rate", **kwargs):
+        self.main_indicator = main_indicator
+        self.word_right = []
+        self.exp_right = []
+        self.word_total_length = 0
+        self.exp_total_num = 0
+        self.word_rate = 0
+        self.exp_rate = 0
+        self.reset()
+        self.epoch_reset()
+
+    def __call__(self, preds, batch, **kwargs):
+        for k, v in kwargs.items():
+            epoch_reset = v
+            if epoch_reset:
+                self.epoch_reset()
+        word_probs = preds
+        word_label, word_label_mask = batch
+        line_right = 0
+        if word_probs is not None:
+            word_pred = word_probs.argmax(2)
+        word_pred = word_pred.cpu().detach().numpy()
+        word_scores = [
+            SequenceMatcher(
+                None, s1[: int(np.sum(s3))], s2[: int(np.sum(s3))], autojunk=False
+            ).ratio()
+            * (len(s1[: int(np.sum(s3))]) + len(s2[: int(np.sum(s3))]))
+            / len(s1[: int(np.sum(s3))])
+            / 2
+            for s1, s2, s3 in zip(word_label, word_pred, word_label_mask)
+        ]
+        batch_size = len(word_scores)
+        for i in range(batch_size):
+            if word_scores[i] == 1:
+                line_right += 1
+        self.word_rate = np.mean(word_scores)  # float
+        self.exp_rate = line_right / batch_size  # float
+        exp_length, word_length = word_label.shape[:2]
+        self.word_right.append(self.word_rate * word_length)
+        self.exp_right.append(self.exp_rate * exp_length)
+        self.word_total_length = self.word_total_length + word_length
+        self.exp_total_num = self.exp_total_num + exp_length
+
+    def get_metric(self):
+        """
+        return {
+            'word_rate': 0,
+            "exp_rate": 0,
+        }
+        """
+        cur_word_rate = sum(self.word_right) / self.word_total_length
+        cur_exp_rate = sum(self.exp_right) / self.exp_total_num
+        self.reset()
+        return {"word_rate": cur_word_rate, "exp_rate": cur_exp_rate}
+
+    def reset(self):
+        self.word_rate = 0
+        self.exp_rate = 0
+
+    def epoch_reset(self):
+        self.word_right = []
+        self.exp_right = []
+        self.word_total_length = 0
+        self.exp_total_num = 0
+
+
+class LaTeXOCRMetric(object):
+    def __init__(self, main_indicator="exp_rate", cal_bleu_score=False, **kwargs):
+        self.main_indicator = main_indicator
+        self.cal_bleu_score = cal_bleu_score
+        self.edit_right = []
+        self.exp_right = []
+        self.bleu_right = []
+        self.e1_right = []
+        self.e2_right = []
+        self.e3_right = []
+        self.exp_total_num = 0
+        self.edit_dist = 0
+        self.exp_rate = 0
+        if self.cal_bleu_score:
+            self.bleu_score = 0
+        self.e1 = 0
+        self.e2 = 0
+        self.e3 = 0
+        self.reset()
+        self.epoch_reset()
+
+    def __call__(self, preds, batch, **kwargs):
+        for k, v in kwargs.items():
+            epoch_reset = v
+            if epoch_reset:
+                self.epoch_reset()
+        word_pred = preds
+        word_label = batch
+        line_right, e1, e2, e3 = 0, 0, 0, 0
+        bleu_list, lev_dist = [], []
+        for labels, prediction in zip(word_label, word_pred):
+            if prediction == labels:
+                line_right += 1
+            distance = compute_edit_distance(prediction, labels)
+            bleu_list.append(compute_bleu_score([prediction], [labels]))
+            lev_dist.append(Levenshtein.normalized_distance(prediction, labels))
+            if distance <= 1:
+                e1 += 1
+            if distance <= 2:
+                e2 += 1
+            if distance <= 3:
+                e3 += 1
+
+        batch_size = len(lev_dist)
+
+        self.edit_dist = sum(lev_dist)  # float
+        self.exp_rate = line_right  # float
+        if self.cal_bleu_score:
+            self.bleu_score = sum(bleu_list)
+            self.bleu_right.append(self.bleu_score)
+        self.e1 = e1
+        self.e2 = e2
+        self.e3 = e3
+        exp_length = len(word_label)
+        self.edit_right.append(self.edit_dist)
+        self.exp_right.append(self.exp_rate)
+        self.e1_right.append(self.e1)
+        self.e2_right.append(self.e2)
+        self.e3_right.append(self.e3)
+        self.exp_total_num = self.exp_total_num + exp_length
+
+    def get_metric(self):
+        """
+        return {
+            'edit distance': 0,
+            "bleu_score": 0,
+            "exp_rate": 0,
+        }
+        """
+        cur_edit_distance = sum(self.edit_right) / self.exp_total_num
+        cur_exp_rate = sum(self.exp_right) / self.exp_total_num
+        if self.cal_bleu_score:
+            cur_bleu_score = sum(self.bleu_right) / self.exp_total_num
+        cur_exp_1 = sum(self.e1_right) / self.exp_total_num
+        cur_exp_2 = sum(self.e2_right) / self.exp_total_num
+        cur_exp_3 = sum(self.e3_right) / self.exp_total_num
+        self.reset()
+        if self.cal_bleu_score:
+            return {
+                "bleu_score": cur_bleu_score,
+                "edit distance": cur_edit_distance,
+                "exp_rate": cur_exp_rate,
+                "exp_rate<=1 ": cur_exp_1,
+                "exp_rate<=2 ": cur_exp_2,
+                "exp_rate<=3 ": cur_exp_3,
+            }
+        else:
+
+            return {
+                "edit distance": cur_edit_distance,
+                "exp_rate": cur_exp_rate,
+                "exp_rate<=1 ": cur_exp_1,
+                "exp_rate<=2 ": cur_exp_2,
+                "exp_rate<=3 ": cur_exp_3,
+            }
+
+    def reset(self):
+        self.edit_dist = 0
+        self.exp_rate = 0
+        if self.cal_bleu_score:
+            self.bleu_score = 0
+        self.e1 = 0
+        self.e2 = 0
+        self.e3 = 0
+
+    def epoch_reset(self):
+        self.edit_right = []
+        self.exp_right = []
+        if self.cal_bleu_score:
+            self.bleu_right = []
+        self.e1_right = []
+        self.e2_right = []
+        self.e3_right = []
+        self.editdistance_total_length = 0
+        self.exp_total_num = 0
diff --git a/ppocr/metrics/sr_metric.py b/ppocr/metrics/sr_metric.py
new file mode 100644
index 0000000..ef9ef96
--- /dev/null
+++ b/ppocr/metrics/sr_metric.py
@@ -0,0 +1,161 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+https://github.com/FudanVI/FudanOCR/blob/main/text-gestalt/utils/ssim_psnr.py
+"""
+
+from math import exp
+
+import paddle
+import paddle.nn.functional as F
+import paddle.nn as nn
+import string
+
+
+class SSIM(nn.Layer):
+    def __init__(self, window_size=11, size_average=True):
+        super(SSIM, self).__init__()
+        self.window_size = window_size
+        self.size_average = size_average
+        self.channel = 1
+        self.window = self.create_window(window_size, self.channel)
+
+    def gaussian(self, window_size, sigma):
+        gauss = paddle.to_tensor(
+            [
+                exp(-((x - window_size // 2) ** 2) / float(2 * sigma**2))
+                for x in range(window_size)
+            ]
+        )
+        return gauss / gauss.sum()
+
+    def create_window(self, window_size, channel):
+        _1D_window = self.gaussian(window_size, 1.5).unsqueeze(1)
+        _2D_window = _1D_window.mm(_1D_window.t()).unsqueeze(0).unsqueeze(0)
+        window = _2D_window.expand([channel, 1, window_size, window_size])
+        return window
+
+    def _ssim(self, img1, img2, window, window_size, channel, size_average=True):
+        mu1 = F.conv2d(img1, window, padding=window_size // 2, groups=channel)
+        mu2 = F.conv2d(img2, window, padding=window_size // 2, groups=channel)
+
+        mu1_sq = mu1.pow(2)
+        mu2_sq = mu2.pow(2)
+        mu1_mu2 = mu1 * mu2
+
+        sigma1_sq = (
+            F.conv2d(img1 * img1, window, padding=window_size // 2, groups=channel)
+            - mu1_sq
+        )
+        sigma2_sq = (
+            F.conv2d(img2 * img2, window, padding=window_size // 2, groups=channel)
+            - mu2_sq
+        )
+        sigma12 = (
+            F.conv2d(img1 * img2, window, padding=window_size // 2, groups=channel)
+            - mu1_mu2
+        )
+
+        C1 = 0.01**2
+        C2 = 0.03**2
+
+        ssim_map = ((2 * mu1_mu2 + C1) * (2 * sigma12 + C2)) / (
+            (mu1_sq + mu2_sq + C1) * (sigma1_sq + sigma2_sq + C2)
+        )
+
+        if size_average:
+            return ssim_map.mean()
+        else:
+            return ssim_map.mean([1, 2, 3])
+
+    def ssim(self, img1, img2, window_size=11, size_average=True):
+        (_, channel, _, _) = img1.shape
+        window = self.create_window(window_size, channel)
+
+        return self._ssim(img1, img2, window, window_size, channel, size_average)
+
+    def forward(self, img1, img2):
+        (_, channel, _, _) = img1.shape
+
+        if channel == self.channel and self.window.dtype == img1.dtype:
+            window = self.window
+        else:
+            window = self.create_window(self.window_size, channel)
+
+            self.window = window
+            self.channel = channel
+
+        return self._ssim(
+            img1, img2, window, self.window_size, channel, self.size_average
+        )
+
+
+class SRMetric(object):
+    def __init__(self, main_indicator="all", **kwargs):
+        self.main_indicator = main_indicator
+        self.eps = 1e-5
+        self.psnr_result = []
+        self.ssim_result = []
+        self.calculate_ssim = SSIM()
+        self.reset()
+
+    def reset(self):
+        self.correct_num = 0
+        self.all_num = 0
+        self.norm_edit_dis = 0
+        self.psnr_result = []
+        self.ssim_result = []
+
+    def calculate_psnr(self, img1, img2):
+        # img1 and img2 have range [0, 1]
+        mse = ((img1 * 255 - img2 * 255) ** 2).mean()
+        if mse == 0:
+            return float("inf")
+        return 20 * paddle.log10(255.0 / paddle.sqrt(mse))
+
+    def _normalize_text(self, text):
+        text = "".join(
+            filter(lambda x: x in (string.digits + string.ascii_letters), text)
+        )
+        return text.lower()
+
+    def __call__(self, pred_label, *args, **kwargs):
+        metric = {}
+        images_sr = pred_label["sr_img"]
+        images_hr = pred_label["hr_img"]
+        psnr = self.calculate_psnr(images_sr, images_hr)
+        ssim = self.calculate_ssim(images_sr, images_hr)
+        self.psnr_result.append(psnr)
+        self.ssim_result.append(ssim)
+
+    def get_metric(self):
+        """
+        return metrics {
+                 'acc': 0,
+                 'norm_edit_dis': 0,
+            }
+        """
+        self.psnr_avg = sum(self.psnr_result) / len(self.psnr_result)
+        self.psnr_avg = round(self.psnr_avg.item(), 6)
+        self.ssim_avg = sum(self.ssim_result) / len(self.ssim_result)
+        self.ssim_avg = round(self.ssim_avg.item(), 6)
+
+        self.all_avg = self.psnr_avg + self.ssim_avg
+
+        self.reset()
+        return {
+            "psnr_avg": self.psnr_avg,
+            "ssim_avg": self.ssim_avg,
+            "all": self.all_avg,
+        }
diff --git a/ppocr/metrics/table_metric.py b/ppocr/metrics/table_metric.py
new file mode 100644
index 0000000..98847e1
--- /dev/null
+++ b/ppocr/metrics/table_metric.py
@@ -0,0 +1,161 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import numpy as np
+from ppocr.metrics.det_metric import DetMetric
+
+
+class TableStructureMetric(object):
+    def __init__(self, main_indicator="acc", eps=1e-6, del_thead_tbody=False, **kwargs):
+        self.main_indicator = main_indicator
+        self.eps = eps
+        self.del_thead_tbody = del_thead_tbody
+        self.reset()
+
+    def __call__(self, pred_label, batch=None, *args, **kwargs):
+        preds, labels = pred_label
+        pred_structure_batch_list = preds["structure_batch_list"]
+        gt_structure_batch_list = labels["structure_batch_list"]
+        correct_num = 0
+        all_num = 0
+        for (pred, pred_conf), target in zip(
+            pred_structure_batch_list, gt_structure_batch_list
+        ):
+            pred_str = "".join(pred)
+            target_str = "".join(target)
+            if self.del_thead_tbody:
+                pred_str = (
+                    pred_str.replace("<thead>", "")
+                    .replace("</thead>", "")
+                    .replace("<tbody>", "")
+                    .replace("</tbody>", "")
+                )
+                target_str = (
+                    target_str.replace("<thead>", "")
+                    .replace("</thead>", "")
+                    .replace("<tbody>", "")
+                    .replace("</tbody>", "")
+                )
+            if pred_str == target_str:
+                correct_num += 1
+            all_num += 1
+        self.correct_num += correct_num
+        self.all_num += all_num
+
+    def get_metric(self):
+        """
+        return metrics {
+                 'acc': 0,
+            }
+        """
+        acc = 1.0 * self.correct_num / (self.all_num + self.eps)
+        self.reset()
+        return {"acc": acc}
+
+    def reset(self):
+        self.correct_num = 0
+        self.all_num = 0
+        self.len_acc_num = 0
+        self.token_nums = 0
+        self.anys_dict = dict()
+
+
+class TableMetric(object):
+    def __init__(
+        self,
+        main_indicator="acc",
+        compute_bbox_metric=False,
+        box_format="xyxy",
+        del_thead_tbody=False,
+        **kwargs,
+    ):
+        """
+
+        @param sub_metrics: configs of sub_metric
+        @param main_matric: main_matric for save best_model
+        @param kwargs:
+        """
+        self.structure_metric = TableStructureMetric(del_thead_tbody=del_thead_tbody)
+        self.bbox_metric = DetMetric() if compute_bbox_metric else None
+        self.main_indicator = main_indicator
+        self.box_format = box_format
+        self.reset()
+
+    def __call__(self, pred_label, batch=None, *args, **kwargs):
+        self.structure_metric(pred_label)
+        if self.bbox_metric is not None:
+            self.bbox_metric(*self.prepare_bbox_metric_input(pred_label))
+
+    def prepare_bbox_metric_input(self, pred_label):
+        pred_bbox_batch_list = []
+        gt_ignore_tags_batch_list = []
+        gt_bbox_batch_list = []
+        preds, labels = pred_label
+
+        batch_num = len(preds["bbox_batch_list"])
+        for batch_idx in range(batch_num):
+            # pred
+            pred_bbox_list = [
+                self.format_box(pred_box)
+                for pred_box in preds["bbox_batch_list"][batch_idx]
+            ]
+            pred_bbox_batch_list.append({"points": pred_bbox_list})
+
+            # gt
+            gt_bbox_list = []
+            gt_ignore_tags_list = []
+            for gt_box in labels["bbox_batch_list"][batch_idx]:
+                gt_bbox_list.append(self.format_box(gt_box))
+                gt_ignore_tags_list.append(0)
+            gt_bbox_batch_list.append(gt_bbox_list)
+            gt_ignore_tags_batch_list.append(gt_ignore_tags_list)
+
+        return [
+            pred_bbox_batch_list,
+            [0, 0, gt_bbox_batch_list, gt_ignore_tags_batch_list],
+        ]
+
+    def get_metric(self):
+        structure_metric = self.structure_metric.get_metric()
+        if self.bbox_metric is None:
+            return structure_metric
+        bbox_metric = self.bbox_metric.get_metric()
+        if self.main_indicator == self.bbox_metric.main_indicator:
+            output = bbox_metric
+            for sub_key in structure_metric:
+                output["structure_metric_{}".format(sub_key)] = structure_metric[
+                    sub_key
+                ]
+        else:
+            output = structure_metric
+            for sub_key in bbox_metric:
+                output["bbox_metric_{}".format(sub_key)] = bbox_metric[sub_key]
+        return output
+
+    def reset(self):
+        self.structure_metric.reset()
+        if self.bbox_metric is not None:
+            self.bbox_metric.reset()
+
+    def format_box(self, box):
+        if self.box_format == "xyxy":
+            x1, y1, x2, y2 = box
+            box = [[x1, y1], [x2, y1], [x2, y2], [x1, y2]]
+        elif self.box_format == "xywh":
+            x, y, w, h = box
+            x1, y1, x2, y2 = x - w // 2, y - h // 2, x + w // 2, y + h // 2
+            box = [[x1, y1], [x2, y1], [x2, y2], [x1, y2]]
+        elif self.box_format == "xyxyxyxy":
+            x1, y1, x2, y2, x3, y3, x4, y4 = box
+            box = [[x1, y1], [x2, y2], [x3, y3], [x4, y4]]
+        return box
diff --git a/ppocr/metrics/vqa_token_re_metric.py b/ppocr/metrics/vqa_token_re_metric.py
new file mode 100644
index 0000000..8c85be5
--- /dev/null
+++ b/ppocr/metrics/vqa_token_re_metric.py
@@ -0,0 +1,191 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+import paddle
+
+__all__ = ["VQAReTokenMetric"]
+
+
+class VQAReTokenMetric(object):
+    def __init__(self, main_indicator="hmean", **kwargs):
+        self.main_indicator = main_indicator
+        self.reset()
+
+    def __call__(self, preds, batch, **kwargs):
+        pred_relations, relations, entities = preds
+        self.pred_relations_list.extend(pred_relations)
+        self.relations_list.extend(relations)
+        self.entities_list.extend(entities)
+
+    def get_metric(self):
+        gt_relations = []
+        for b in range(len(self.relations_list)):
+            rel_sent = []
+            relation_list = self.relations_list[b]
+            entitie_list = self.entities_list[b]
+            head_len = relation_list[0, 0]
+            if head_len > 0:
+                entitie_start_list = entitie_list[1 : entitie_list[0, 0] + 1, 0]
+                entitie_end_list = entitie_list[1 : entitie_list[0, 1] + 1, 1]
+                entitie_label_list = entitie_list[1 : entitie_list[0, 2] + 1, 2]
+                for head, tail in zip(
+                    relation_list[1 : head_len + 1, 0],
+                    relation_list[1 : head_len + 1, 1],
+                ):
+                    rel = {}
+                    rel["head_id"] = head
+                    rel["head"] = (entitie_start_list[head], entitie_end_list[head])
+                    rel["head_type"] = entitie_label_list[head]
+
+                    rel["tail_id"] = tail
+                    rel["tail"] = (entitie_start_list[tail], entitie_end_list[tail])
+                    rel["tail_type"] = entitie_label_list[tail]
+
+                    rel["type"] = 1
+                    rel_sent.append(rel)
+            gt_relations.append(rel_sent)
+        re_metrics = self.re_score(
+            self.pred_relations_list, gt_relations, mode="boundaries"
+        )
+        metrics = {
+            "precision": re_metrics["ALL"]["p"],
+            "recall": re_metrics["ALL"]["r"],
+            "hmean": re_metrics["ALL"]["f1"],
+        }
+        self.reset()
+        return metrics
+
+    def reset(self):
+        self.pred_relations_list = []
+        self.relations_list = []
+        self.entities_list = []
+
+    def re_score(self, pred_relations, gt_relations, mode="strict"):
+        """Evaluate RE predictions
+
+        Args:
+            pred_relations (list) :  list of list of predicted relations (several relations in each sentence)
+            gt_relations (list) :    list of list of ground truth relations
+
+                rel = { "head": (start_idx (inclusive), end_idx (exclusive)),
+                        "tail": (start_idx (inclusive), end_idx (exclusive)),
+                        "head_type": ent_type,
+                        "tail_type": ent_type,
+                        "type": rel_type}
+
+            vocab (Vocab) :         dataset vocabulary
+            mode (str) :            in 'strict' or 'boundaries'"""
+
+        assert mode in ["strict", "boundaries"]
+
+        relation_types = [v for v in [0, 1] if not v == 0]
+        scores = {rel: {"tp": 0, "fp": 0, "fn": 0} for rel in relation_types + ["ALL"]}
+
+        # Count GT relations and Predicted relations
+        n_sents = len(gt_relations)
+        n_rels = sum([len([rel for rel in sent]) for sent in gt_relations])
+        n_found = sum([len([rel for rel in sent]) for sent in pred_relations])
+
+        # Count TP, FP and FN per type
+        for pred_sent, gt_sent in zip(pred_relations, gt_relations):
+            for rel_type in relation_types:
+                # strict mode takes argument types into account
+                if mode == "strict":
+                    pred_rels = {
+                        (rel["head"], rel["head_type"], rel["tail"], rel["tail_type"])
+                        for rel in pred_sent
+                        if rel["type"] == rel_type
+                    }
+                    gt_rels = {
+                        (rel["head"], rel["head_type"], rel["tail"], rel["tail_type"])
+                        for rel in gt_sent
+                        if rel["type"] == rel_type
+                    }
+
+                # boundaries mode only takes argument spans into account
+                elif mode == "boundaries":
+                    pred_rels = {
+                        (rel["head"], rel["tail"])
+                        for rel in pred_sent
+                        if rel["type"] == rel_type
+                    }
+                    gt_rels = {
+                        (rel["head"], rel["tail"])
+                        for rel in gt_sent
+                        if rel["type"] == rel_type
+                    }
+
+                scores[rel_type]["tp"] += len(pred_rels & gt_rels)
+                scores[rel_type]["fp"] += len(pred_rels - gt_rels)
+                scores[rel_type]["fn"] += len(gt_rels - pred_rels)
+
+        # Compute per entity Precision / Recall / F1
+        for rel_type in scores.keys():
+            if scores[rel_type]["tp"]:
+                scores[rel_type]["p"] = scores[rel_type]["tp"] / (
+                    scores[rel_type]["fp"] + scores[rel_type]["tp"]
+                )
+                scores[rel_type]["r"] = scores[rel_type]["tp"] / (
+                    scores[rel_type]["fn"] + scores[rel_type]["tp"]
+                )
+            else:
+                scores[rel_type]["p"], scores[rel_type]["r"] = 0, 0
+
+            if not scores[rel_type]["p"] + scores[rel_type]["r"] == 0:
+                scores[rel_type]["f1"] = (
+                    2
+                    * scores[rel_type]["p"]
+                    * scores[rel_type]["r"]
+                    / (scores[rel_type]["p"] + scores[rel_type]["r"])
+                )
+            else:
+                scores[rel_type]["f1"] = 0
+
+        # Compute micro F1 Scores
+        tp = sum([scores[rel_type]["tp"] for rel_type in relation_types])
+        fp = sum([scores[rel_type]["fp"] for rel_type in relation_types])
+        fn = sum([scores[rel_type]["fn"] for rel_type in relation_types])
+
+        if tp:
+            precision = tp / (tp + fp)
+            recall = tp / (tp + fn)
+            f1 = 2 * precision * recall / (precision + recall)
+
+        else:
+            precision, recall, f1 = 0, 0, 0
+
+        scores["ALL"]["p"] = precision
+        scores["ALL"]["r"] = recall
+        scores["ALL"]["f1"] = f1
+        scores["ALL"]["tp"] = tp
+        scores["ALL"]["fp"] = fp
+        scores["ALL"]["fn"] = fn
+
+        # Compute Macro F1 Scores
+        scores["ALL"]["Macro_f1"] = np.mean(
+            [scores[ent_type]["f1"] for ent_type in relation_types]
+        )
+        scores["ALL"]["Macro_p"] = np.mean(
+            [scores[ent_type]["p"] for ent_type in relation_types]
+        )
+        scores["ALL"]["Macro_r"] = np.mean(
+            [scores[ent_type]["r"] for ent_type in relation_types]
+        )
+
+        return scores
diff --git a/ppocr/metrics/vqa_token_ser_metric.py b/ppocr/metrics/vqa_token_ser_metric.py
new file mode 100644
index 0000000..3afcb05
--- /dev/null
+++ b/ppocr/metrics/vqa_token_ser_metric.py
@@ -0,0 +1,48 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+import paddle
+
+__all__ = ["VQASerTokenMetric"]
+
+
+class VQASerTokenMetric(object):
+    def __init__(self, main_indicator="hmean", **kwargs):
+        self.main_indicator = main_indicator
+        self.reset()
+
+    def __call__(self, preds, batch, **kwargs):
+        preds, labels = preds
+        self.pred_list.extend(preds)
+        self.gt_list.extend(labels)
+
+    def get_metric(self):
+        from seqeval.metrics import f1_score, precision_score, recall_score
+
+        metrics = {
+            "precision": precision_score(self.gt_list, self.pred_list),
+            "recall": recall_score(self.gt_list, self.pred_list),
+            "hmean": f1_score(self.gt_list, self.pred_list),
+        }
+        self.reset()
+        return metrics
+
+    def reset(self):
+        self.pred_list = []
+        self.gt_list = []
diff --git a/ppocr/modeling/architectures/__init__.py b/ppocr/modeling/architectures/__init__.py
new file mode 100755
index 0000000..65b6318
--- /dev/null
+++ b/ppocr/modeling/architectures/__init__.py
@@ -0,0 +1,145 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import copy
+import importlib
+
+from paddle.jit import to_static
+from paddle.static import InputSpec
+
+from .base_model import BaseModel
+from .distillation_model import DistillationModel
+
+__all__ = ["build_model", "apply_to_static"]
+
+
+def build_model(config):
+    config = copy.deepcopy(config)
+    if not "name" in config:
+        arch = BaseModel(config)
+    else:
+        name = config.pop("name")
+        mod = importlib.import_module(__name__)
+        arch = getattr(mod, name)(config)
+    return arch
+
+
+def apply_to_static(model, config, logger):
+    if config["Global"].get("to_static", False) is not True:
+        return model
+    assert (
+        "d2s_train_image_shape" in config["Global"]
+    ), "d2s_train_image_shape must be assigned for static training mode..."
+    supported_list = [
+        "DB",
+        "SVTR_LCNet",
+        "TableMaster",
+        "LayoutXLM",
+        "SLANet",
+        "SVTR",
+        "SVTR_HGNet",
+        "LaTeXOCR",
+        "UniMERNet",
+        "PP-FormulaNet-S",
+        "PP-FormulaNet-L",
+    ]
+    if config["Architecture"]["algorithm"] in ["Distillation"]:
+        algo = list(config["Architecture"]["Models"].values())[0]["algorithm"]
+    else:
+        algo = config["Architecture"]["algorithm"]
+    assert (
+        algo in supported_list
+    ), f"algorithms that supports static training must in in {supported_list} but got {algo}"
+
+    specs = [
+        InputSpec([None] + config["Global"]["d2s_train_image_shape"], dtype="float32")
+    ]
+
+    if algo == "SVTR_LCNet":
+        specs.append(
+            [
+                InputSpec([None, config["Global"]["max_text_length"]], dtype="int64"),
+                InputSpec([None, config["Global"]["max_text_length"]], dtype="int64"),
+                InputSpec([None], dtype="int64"),
+                InputSpec([None], dtype="float64"),
+            ]
+        )
+    elif algo == "TableMaster":
+        specs.append(
+            [
+                InputSpec([None, config["Global"]["max_text_length"]], dtype="int64"),
+                InputSpec(
+                    [None, config["Global"]["max_text_length"], 4], dtype="float32"
+                ),
+                InputSpec(
+                    [None, config["Global"]["max_text_length"], 1], dtype="float32"
+                ),
+                InputSpec([None, 6], dtype="float32"),
+            ]
+        )
+    elif algo == "LayoutXLM":
+        specs = [
+            [
+                InputSpec(shape=[None, 512], dtype="int64"),  # input_ids
+                InputSpec(shape=[None, 512, 4], dtype="int64"),  # bbox
+                InputSpec(shape=[None, 512], dtype="int64"),  # attention_mask
+                InputSpec(shape=[None, 512], dtype="int64"),  # token_type_ids
+                InputSpec(shape=[None, 3, 224, 224], dtype="float32"),  # image
+                InputSpec(shape=[None, 512], dtype="int64"),  # label
+            ]
+        ]
+    elif algo == "SLANet":
+        specs.append(
+            [
+                InputSpec(
+                    [None, config["Global"]["max_text_length"] + 2], dtype="int64"
+                ),
+                InputSpec(
+                    [None, config["Global"]["max_text_length"] + 2, 4], dtype="float32"
+                ),
+                InputSpec(
+                    [None, config["Global"]["max_text_length"] + 2, 1], dtype="float32"
+                ),
+                InputSpec([None], dtype="int64"),
+                InputSpec([None, 6], dtype="float64"),
+            ]
+        )
+    elif algo == "SVTR":
+        specs.append(
+            [
+                InputSpec([None, config["Global"]["max_text_length"]], dtype="int64"),
+                InputSpec([None], dtype="int64"),
+            ]
+        )
+    elif algo == "LaTeXOCR":
+        specs = [
+            [
+                InputSpec(shape=[None, 1, None, None], dtype="float32"),
+                InputSpec(shape=[None, None], dtype="float32"),
+                InputSpec(shape=[None, None], dtype="float32"),
+            ]
+        ]
+    elif algo in ["UniMERNet", "PP-FormulaNet-S", "PP-FormulaNet-L"]:
+        specs = [
+            [
+                InputSpec(
+                    [None] + config["Global"]["d2s_train_image_shape"], dtype="float32"
+                ),
+                InputSpec(shape=[None, None], dtype="float32"),
+                InputSpec(shape=[None, None], dtype="float32"),
+            ]
+        ]
+    model = to_static(model, input_spec=specs)
+    logger.info("Successfully to apply @to_static with specs: {}".format(specs))
+    return model
diff --git a/ppocr/modeling/architectures/base_model.py b/ppocr/modeling/architectures/base_model.py
new file mode 100644
index 0000000..036f90c
--- /dev/null
+++ b/ppocr/modeling/architectures/base_model.py
@@ -0,0 +1,117 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from paddle import nn
+from ppocr.modeling.transforms import build_transform
+from ppocr.modeling.backbones import build_backbone
+from ppocr.modeling.necks import build_neck
+from ppocr.modeling.heads import build_head
+
+__all__ = ["BaseModel"]
+
+
+class BaseModel(nn.Layer):
+    def __init__(self, config):
+        """
+        the module for OCR.
+        args:
+            config (dict): the super parameters for module.
+        """
+        super(BaseModel, self).__init__()
+        in_channels = config.get("in_channels", 3)
+        model_type = config["model_type"]
+        # build transform,
+        # for rec, transform can be TPS,None
+        # for det and cls, transform should to be None,
+        # if you make model differently, you can use transform in det and cls
+        if "Transform" not in config or config["Transform"] is None:
+            self.use_transform = False
+        else:
+            self.use_transform = True
+            config["Transform"]["in_channels"] = in_channels
+            self.transform = build_transform(config["Transform"])
+            in_channels = self.transform.out_channels
+
+        # build backbone, backbone is need for del, rec and cls
+        if "Backbone" not in config or config["Backbone"] is None:
+            self.use_backbone = False
+        else:
+            self.use_backbone = True
+            config["Backbone"]["in_channels"] = in_channels
+            self.backbone = build_backbone(config["Backbone"], model_type)
+            in_channels = self.backbone.out_channels
+
+        # build neck
+        # for rec, neck can be cnn,rnn or reshape(None)
+        # for det, neck can be FPN, BIFPN and so on.
+        # for cls, neck should be none
+        if "Neck" not in config or config["Neck"] is None:
+            self.use_neck = False
+        else:
+            self.use_neck = True
+            config["Neck"]["in_channels"] = in_channels
+            self.neck = build_neck(config["Neck"])
+            in_channels = self.neck.out_channels
+
+        # # build head, head is need for det, rec and cls
+        if "Head" not in config or config["Head"] is None:
+            self.use_head = False
+        else:
+            self.use_head = True
+            config["Head"]["in_channels"] = in_channels
+            self.head = build_head(config["Head"])
+
+        self.return_all_feats = config.get("return_all_feats", False)
+
+    def forward(self, x, data=None):
+        y = dict()
+        if self.use_transform:
+            x = self.transform(x)
+        if self.use_backbone:
+            x = self.backbone(x)
+        if isinstance(x, dict):
+            y.update(x)
+        else:
+            y["backbone_out"] = x
+        final_name = "backbone_out"
+        if self.use_neck:
+            x = self.neck(x)
+            if isinstance(x, dict):
+                y.update(x)
+            else:
+                y["neck_out"] = x
+            final_name = "neck_out"
+        if self.use_head:
+            x = self.head(x, targets=data)
+            # for multi head, save ctc neck out for udml
+            if isinstance(x, dict) and "ctc_neck" in x.keys():
+                y["neck_out"] = x["ctc_neck"]
+                y["head_out"] = x
+            elif isinstance(x, dict):
+                y.update(x)
+            else:
+                y["head_out"] = x
+            final_name = "head_out"
+        if self.return_all_feats:
+            if self.training:
+                return y
+            elif isinstance(x, dict):
+                return x
+            else:
+                return {final_name: x}
+        else:
+            return x
diff --git a/ppocr/modeling/architectures/distillation_model.py b/ppocr/modeling/architectures/distillation_model.py
new file mode 100644
index 0000000..2309d2e
--- /dev/null
+++ b/ppocr/modeling/architectures/distillation_model.py
@@ -0,0 +1,60 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from paddle import nn
+from ppocr.modeling.transforms import build_transform
+from ppocr.modeling.backbones import build_backbone
+from ppocr.modeling.necks import build_neck
+from ppocr.modeling.heads import build_head
+from .base_model import BaseModel
+from ppocr.utils.save_load import load_pretrained_params
+
+__all__ = ["DistillationModel"]
+
+
+class DistillationModel(nn.Layer):
+    def __init__(self, config):
+        """
+        the module for OCR distillation.
+        args:
+            config (dict): the super parameters for module.
+        """
+        super().__init__()
+        self.model_list = []
+        self.model_name_list = []
+        for key in config["Models"]:
+            model_config = config["Models"][key]
+            freeze_params = False
+            pretrained = None
+            if "freeze_params" in model_config:
+                freeze_params = model_config.pop("freeze_params")
+            if "pretrained" in model_config:
+                pretrained = model_config.pop("pretrained")
+            model = BaseModel(model_config)
+            if pretrained is not None:
+                load_pretrained_params(model, pretrained)
+            if freeze_params:
+                for param in model.parameters():
+                    param.trainable = False
+            self.model_list.append(self.add_sublayer(key, model))
+            self.model_name_list.append(key)
+
+    def forward(self, x, data=None):
+        result_dict = dict()
+        for idx, model_name in enumerate(self.model_name_list):
+            result_dict[model_name] = self.model_list[idx](x, data)
+        return result_dict
diff --git a/ppocr/modeling/backbones/__init__.py b/ppocr/modeling/backbones/__init__.py
new file mode 100755
index 0000000..69da404
--- /dev/null
+++ b/ppocr/modeling/backbones/__init__.py
@@ -0,0 +1,156 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+__all__ = ["build_backbone"]
+
+
+def build_backbone(config, model_type):
+    if model_type == "det" or model_type == "table":
+        from .det_mobilenet_v3 import MobileNetV3
+        from .det_resnet import ResNet
+        from .det_resnet_vd import ResNet_vd
+        from .det_resnet_vd_sast import ResNet_SAST
+        from .det_pp_lcnet import PPLCNet
+        from .rec_lcnetv3 import PPLCNetV3
+        from .rec_lcnetv4 import PPLCNetV4
+        from .rec_hgnet import PPHGNet_small
+        from .rec_vit import ViT
+        from .det_pp_lcnet_v2 import PPLCNetV2_base
+        from .rec_repvit import RepSVTR_det
+        from .rec_vary_vit import Vary_VIT_B
+        from .rec_pphgnetv2 import PPHGNetV2_B4
+
+        support_dict = [
+            "MobileNetV3",
+            "ResNet",
+            "ResNet_vd",
+            "ResNet_SAST",
+            "PPLCNet",
+            "PPLCNetV3",
+            "PPLCNetV4",
+            "PPHGNet_small",
+            "PPLCNetV2_base",
+            "RepSVTR_det",
+            "Vary_VIT_B",
+            "PPHGNetV2_B4",
+        ]
+        if model_type == "table":
+            from .table_master_resnet import TableResNetExtra
+
+            support_dict.append("TableResNetExtra")
+    elif model_type == "rec" or model_type == "cls":
+        from .rec_mobilenet_v3 import MobileNetV3
+        from .rec_resnet_vd import ResNet
+        from .rec_resnet_fpn import ResNetFPN
+        from .rec_mv1_enhance import MobileNetV1Enhance
+        from .rec_nrtr_mtb import MTB
+        from .rec_resnet_31 import ResNet31
+        from .rec_resnet_32 import ResNet32
+        from .rec_resnet_45 import ResNet45
+        from .rec_resnet_aster import ResNet_ASTER
+        from .rec_micronet import MicroNet
+        from .rec_efficientb3_pren import EfficientNetb3_PREN
+        from .rec_svtrnet import SVTRNet
+        from .rec_vitstr import ViTSTR
+        from .rec_resnet_rfl import ResNetRFL
+        from .rec_densenet import DenseNet
+        from .rec_resnetv2 import ResNetV2
+        from .rec_hybridvit import HybridTransformer
+        from .rec_donut_swin import DonutSwinModel
+        from .rec_shallow_cnn import ShallowCNN
+        from .rec_lcnetv3 import PPLCNetV3
+        from .rec_lcnetv4 import PPLCNetV4
+        from .rec_hgnet import PPHGNet_small
+        from .rec_vit_parseq import ViTParseQ
+        from .rec_repvit import RepSVTR
+        from .rec_svtrv2 import SVTRv2
+        from .rec_vary_vit import Vary_VIT_B, Vary_VIT_B_Formula
+        from .rec_pphgnetv2 import (
+            PPHGNetV2_B4,
+            PPHGNetV2_B4_Formula,
+            PPHGNetV2_B6_Formula,
+        )
+
+        support_dict = [
+            "MobileNetV1Enhance",
+            "MobileNetV3",
+            "ResNet",
+            "ResNetFPN",
+            "MTB",
+            "ResNet31",
+            "ResNet45",
+            "ResNet_ASTER",
+            "MicroNet",
+            "EfficientNetb3_PREN",
+            "SVTRNet",
+            "ViTSTR",
+            "ResNet32",
+            "ResNetRFL",
+            "DenseNet",
+            "ShallowCNN",
+            "PPLCNetV3",
+            "PPLCNetV4",
+            "PPHGNet_small",
+            "ViTParseQ",
+            "ViT",
+            "RepSVTR",
+            "SVTRv2",
+            "ResNetV2",
+            "HybridTransformer",
+            "DonutSwinModel",
+            "Vary_VIT_B",
+            "PPHGNetV2_B4",
+            "PPHGNetV2_B4_Formula",
+            "PPHGNetV2_B6_Formula",
+            "Vary_VIT_B_Formula",
+        ]
+    elif model_type == "e2e":
+        from .e2e_resnet_vd_pg import ResNet
+
+        support_dict = ["ResNet"]
+    elif model_type == "kie":
+        from .kie_unet_sdmgr import Kie_backbone
+        from .vqa_layoutlm import (
+            LayoutLMForSer,
+            LayoutLMv2ForSer,
+            LayoutLMv2ForRe,
+            LayoutXLMForSer,
+            LayoutXLMForRe,
+        )
+
+        support_dict = [
+            "Kie_backbone",
+            "LayoutLMForSer",
+            "LayoutLMv2ForSer",
+            "LayoutLMv2ForRe",
+            "LayoutXLMForSer",
+            "LayoutXLMForRe",
+        ]
+    elif model_type == "table":
+        from .table_resnet_vd import ResNet
+        from .table_mobilenet_v3 import MobileNetV3
+        from .rec_vary_vit import Vary_VIT_B
+
+        support_dict = ["ResNet", "MobileNetV3", "Vary_VIT_B"]
+    else:
+        raise NotImplementedError
+
+    module_name = config.pop("name")
+    assert module_name in support_dict, Exception(
+        "when model typs is {}, backbone only support {}".format(
+            model_type, support_dict
+        )
+    )
+    module_class = eval(module_name)(**config)
+    return module_class
diff --git a/ppocr/modeling/backbones/det_mobilenet_v3.py b/ppocr/modeling/backbones/det_mobilenet_v3.py
new file mode 100755
index 0000000..5ff08c6
--- /dev/null
+++ b/ppocr/modeling/backbones/det_mobilenet_v3.py
@@ -0,0 +1,289 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+from ppocr.modeling.backbones.rec_hgnet import MeanPool2D
+
+__all__ = ["MobileNetV3"]
+
+
+def make_divisible(v, divisor=8, min_value=None):
+    if min_value is None:
+        min_value = divisor
+    new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
+    if new_v < 0.9 * v:
+        new_v += divisor
+    return new_v
+
+
+class MobileNetV3(nn.Layer):
+    def __init__(
+        self, in_channels=3, model_name="large", scale=0.5, disable_se=False, **kwargs
+    ):
+        """
+        the MobilenetV3 backbone network for detection module.
+        Args:
+            params(dict): the super parameters for build network
+        """
+        super(MobileNetV3, self).__init__()
+
+        self.disable_se = disable_se
+
+        if model_name == "large":
+            cfg = [
+                # k, exp, c,  se,     nl,  s,
+                [3, 16, 16, False, "relu", 1],
+                [3, 64, 24, False, "relu", 2],
+                [3, 72, 24, False, "relu", 1],
+                [5, 72, 40, True, "relu", 2],
+                [5, 120, 40, True, "relu", 1],
+                [5, 120, 40, True, "relu", 1],
+                [3, 240, 80, False, "hardswish", 2],
+                [3, 200, 80, False, "hardswish", 1],
+                [3, 184, 80, False, "hardswish", 1],
+                [3, 184, 80, False, "hardswish", 1],
+                [3, 480, 112, True, "hardswish", 1],
+                [3, 672, 112, True, "hardswish", 1],
+                [5, 672, 160, True, "hardswish", 2],
+                [5, 960, 160, True, "hardswish", 1],
+                [5, 960, 160, True, "hardswish", 1],
+            ]
+            cls_ch_squeeze = 960
+        elif model_name == "small":
+            cfg = [
+                # k, exp, c,  se,     nl,  s,
+                [3, 16, 16, True, "relu", 2],
+                [3, 72, 24, False, "relu", 2],
+                [3, 88, 24, False, "relu", 1],
+                [5, 96, 40, True, "hardswish", 2],
+                [5, 240, 40, True, "hardswish", 1],
+                [5, 240, 40, True, "hardswish", 1],
+                [5, 120, 48, True, "hardswish", 1],
+                [5, 144, 48, True, "hardswish", 1],
+                [5, 288, 96, True, "hardswish", 2],
+                [5, 576, 96, True, "hardswish", 1],
+                [5, 576, 96, True, "hardswish", 1],
+            ]
+            cls_ch_squeeze = 576
+        else:
+            raise NotImplementedError(
+                "mode[" + model_name + "_model] is not implemented!"
+            )
+
+        supported_scale = [0.35, 0.5, 0.75, 1.0, 1.25]
+        assert (
+            scale in supported_scale
+        ), "supported scale are {} but input scale is {}".format(supported_scale, scale)
+        inplanes = 16
+        # conv1
+        self.conv = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=make_divisible(inplanes * scale),
+            kernel_size=3,
+            stride=2,
+            padding=1,
+            groups=1,
+            if_act=True,
+            act="hardswish",
+        )
+
+        self.stages = []
+        self.out_channels = []
+        block_list = []
+        i = 0
+        inplanes = make_divisible(inplanes * scale)
+        for k, exp, c, se, nl, s in cfg:
+            se = se and not self.disable_se
+            start_idx = 2 if model_name == "large" else 0
+            if s == 2 and i > start_idx:
+                self.out_channels.append(inplanes)
+                self.stages.append(nn.Sequential(*block_list))
+                block_list = []
+            block_list.append(
+                ResidualUnit(
+                    in_channels=inplanes,
+                    mid_channels=make_divisible(scale * exp),
+                    out_channels=make_divisible(scale * c),
+                    kernel_size=k,
+                    stride=s,
+                    use_se=se,
+                    act=nl,
+                )
+            )
+            inplanes = make_divisible(scale * c)
+            i += 1
+        block_list.append(
+            ConvBNLayer(
+                in_channels=inplanes,
+                out_channels=make_divisible(scale * cls_ch_squeeze),
+                kernel_size=1,
+                stride=1,
+                padding=0,
+                groups=1,
+                if_act=True,
+                act="hardswish",
+            )
+        )
+        self.stages.append(nn.Sequential(*block_list))
+        self.out_channels.append(make_divisible(scale * cls_ch_squeeze))
+        for i, stage in enumerate(self.stages):
+            self.add_sublayer(sublayer=stage, name="stage{}".format(i))
+
+    def forward(self, x):
+        x = self.conv(x)
+        out_list = []
+        for stage in self.stages:
+            x = stage(x)
+            out_list.append(x)
+        return out_list
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride,
+        padding,
+        groups=1,
+        if_act=True,
+        act=None,
+    ):
+        super(ConvBNLayer, self).__init__()
+        self.if_act = if_act
+        self.act = act
+        self.conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            groups=groups,
+            bias_attr=False,
+        )
+
+        self.bn = nn.BatchNorm(num_channels=out_channels, act=None)
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        if self.if_act:
+            if self.act == "relu":
+                x = F.relu(x)
+            elif self.act == "hardswish":
+                x = F.hardswish(x)
+            else:
+                print(
+                    "The activation function({}) is selected incorrectly.".format(
+                        self.act
+                    )
+                )
+                exit()
+        return x
+
+
+class ResidualUnit(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        mid_channels,
+        out_channels,
+        kernel_size,
+        stride,
+        use_se,
+        act=None,
+    ):
+        super(ResidualUnit, self).__init__()
+        self.if_shortcut = stride == 1 and in_channels == out_channels
+        self.if_se = use_se
+
+        self.expand_conv = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=mid_channels,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            if_act=True,
+            act=act,
+        )
+        self.bottleneck_conv = ConvBNLayer(
+            in_channels=mid_channels,
+            out_channels=mid_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=int((kernel_size - 1) // 2),
+            groups=mid_channels,
+            if_act=True,
+            act=act,
+        )
+        if self.if_se:
+            self.mid_se = SEModule(mid_channels)
+        self.linear_conv = ConvBNLayer(
+            in_channels=mid_channels,
+            out_channels=out_channels,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            if_act=False,
+            act=None,
+        )
+
+    def forward(self, inputs):
+        x = self.expand_conv(inputs)
+        x = self.bottleneck_conv(x)
+        if self.if_se:
+            x = self.mid_se(x)
+        x = self.linear_conv(x)
+        if self.if_shortcut:
+            x = paddle.add(inputs, x)
+        return x
+
+
+class SEModule(nn.Layer):
+    def __init__(self, in_channels, reduction=4):
+        super(SEModule, self).__init__()
+        if "npu" in paddle.device.get_device():
+            self.avg_pool = MeanPool2D(1, 1)
+        else:
+            self.avg_pool = nn.AdaptiveAvgPool2D(1)
+        self.conv1 = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=in_channels // reduction,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        )
+        self.conv2 = nn.Conv2D(
+            in_channels=in_channels // reduction,
+            out_channels=in_channels,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        )
+
+    def forward(self, inputs):
+        outputs = self.avg_pool(inputs)
+        outputs = self.conv1(outputs)
+        outputs = F.relu(outputs)
+        outputs = self.conv2(outputs)
+        outputs = F.hardsigmoid(outputs, slope=0.2, offset=0.5)
+        return inputs * outputs
diff --git a/ppocr/modeling/backbones/det_pp_lcnet.py b/ppocr/modeling/backbones/det_pp_lcnet.py
new file mode 100644
index 0000000..bf557a4
--- /dev/null
+++ b/ppocr/modeling/backbones/det_pp_lcnet.py
@@ -0,0 +1,274 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import, division, print_function
+
+import os
+import paddle
+import paddle.nn as nn
+from paddle import ParamAttr
+from paddle.nn import AdaptiveAvgPool2D, BatchNorm, Conv2D, Dropout, Linear
+from paddle.regularizer import L2Decay
+from paddle.nn.initializer import KaimingNormal
+from paddle.utils.download import get_path_from_url
+
+MODEL_URLS = {
+    "PPLCNet_x0.25": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPLCNet_x0_25_pretrained.pdparams",
+    "PPLCNet_x0.35": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPLCNet_x0_35_pretrained.pdparams",
+    "PPLCNet_x0.5": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPLCNet_x0_5_pretrained.pdparams",
+    "PPLCNet_x0.75": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPLCNet_x0_75_pretrained.pdparams",
+    "PPLCNet_x1.0": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPLCNet_x1_0_pretrained.pdparams",
+    "PPLCNet_x1.5": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPLCNet_x1_5_pretrained.pdparams",
+    "PPLCNet_x2.0": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPLCNet_x2_0_pretrained.pdparams",
+    "PPLCNet_x2.5": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPLCNet_x2_5_pretrained.pdparams",
+}
+
+MODEL_STAGES_PATTERN = {
+    "PPLCNet": ["blocks2", "blocks3", "blocks4", "blocks5", "blocks6"]
+}
+
+__all__ = list(MODEL_URLS.keys())
+
+# Each element(list) represents a depthwise block, which is composed of k, in_c, out_c, s, use_se.
+# k: kernel_size
+# in_c: input channel number in depthwise block
+# out_c: output channel number in depthwise block
+# s: stride in depthwise block
+# use_se: whether to use SE block
+
+NET_CONFIG = {
+    "blocks2":
+    # k, in_c, out_c, s, use_se
+    [[3, 16, 32, 1, False]],
+    "blocks3": [[3, 32, 64, 2, False], [3, 64, 64, 1, False]],
+    "blocks4": [[3, 64, 128, 2, False], [3, 128, 128, 1, False]],
+    "blocks5": [
+        [3, 128, 256, 2, False],
+        [5, 256, 256, 1, False],
+        [5, 256, 256, 1, False],
+        [5, 256, 256, 1, False],
+        [5, 256, 256, 1, False],
+        [5, 256, 256, 1, False],
+    ],
+    "blocks6": [[5, 256, 512, 2, True], [5, 512, 512, 1, True]],
+}
+
+
+def make_divisible(v, divisor=8, min_value=None):
+    if min_value is None:
+        min_value = divisor
+    new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
+    if new_v < 0.9 * v:
+        new_v += divisor
+    return new_v
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(self, num_channels, filter_size, num_filters, stride, num_groups=1):
+        super().__init__()
+
+        self.conv = Conv2D(
+            in_channels=num_channels,
+            out_channels=num_filters,
+            kernel_size=filter_size,
+            stride=stride,
+            padding=(filter_size - 1) // 2,
+            groups=num_groups,
+            weight_attr=ParamAttr(initializer=KaimingNormal()),
+            bias_attr=False,
+        )
+
+        self.bn = BatchNorm(
+            num_filters,
+            param_attr=ParamAttr(regularizer=L2Decay(0.0)),
+            bias_attr=ParamAttr(regularizer=L2Decay(0.0)),
+        )
+        self.hardswish = nn.Hardswish()
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        x = self.hardswish(x)
+        return x
+
+
+class DepthwiseSeparable(nn.Layer):
+    def __init__(self, num_channels, num_filters, stride, dw_size=3, use_se=False):
+        super().__init__()
+        self.use_se = use_se
+        self.dw_conv = ConvBNLayer(
+            num_channels=num_channels,
+            num_filters=num_channels,
+            filter_size=dw_size,
+            stride=stride,
+            num_groups=num_channels,
+        )
+        if use_se:
+            self.se = SEModule(num_channels)
+        self.pw_conv = ConvBNLayer(
+            num_channels=num_channels, filter_size=1, num_filters=num_filters, stride=1
+        )
+
+    def forward(self, x):
+        x = self.dw_conv(x)
+        if self.use_se:
+            x = self.se(x)
+        x = self.pw_conv(x)
+        return x
+
+
+class SEModule(nn.Layer):
+    def __init__(self, channel, reduction=4):
+        super().__init__()
+        self.avg_pool = AdaptiveAvgPool2D(1)
+        self.conv1 = Conv2D(
+            in_channels=channel,
+            out_channels=channel // reduction,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        )
+        self.relu = nn.ReLU()
+        self.conv2 = Conv2D(
+            in_channels=channel // reduction,
+            out_channels=channel,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        )
+        self.hardsigmoid = nn.Hardsigmoid()
+
+    def forward(self, x):
+        identity = x
+        x = self.avg_pool(x)
+        x = self.conv1(x)
+        x = self.relu(x)
+        x = self.conv2(x)
+        x = self.hardsigmoid(x)
+        x = paddle.multiply(x=identity, y=x)
+        return x
+
+
+class PPLCNet(nn.Layer):
+    def __init__(self, in_channels=3, scale=1.0, pretrained=False, use_ssld=False):
+        super().__init__()
+        self.out_channels = [
+            int(NET_CONFIG["blocks3"][-1][2] * scale),
+            int(NET_CONFIG["blocks4"][-1][2] * scale),
+            int(NET_CONFIG["blocks5"][-1][2] * scale),
+            int(NET_CONFIG["blocks6"][-1][2] * scale),
+        ]
+        self.scale = scale
+
+        self.conv1 = ConvBNLayer(
+            num_channels=in_channels,
+            filter_size=3,
+            num_filters=make_divisible(16 * scale),
+            stride=2,
+        )
+
+        self.blocks2 = nn.Sequential(
+            *[
+                DepthwiseSeparable(
+                    num_channels=make_divisible(in_c * scale),
+                    num_filters=make_divisible(out_c * scale),
+                    dw_size=k,
+                    stride=s,
+                    use_se=se,
+                )
+                for i, (k, in_c, out_c, s, se) in enumerate(NET_CONFIG["blocks2"])
+            ]
+        )
+
+        self.blocks3 = nn.Sequential(
+            *[
+                DepthwiseSeparable(
+                    num_channels=make_divisible(in_c * scale),
+                    num_filters=make_divisible(out_c * scale),
+                    dw_size=k,
+                    stride=s,
+                    use_se=se,
+                )
+                for i, (k, in_c, out_c, s, se) in enumerate(NET_CONFIG["blocks3"])
+            ]
+        )
+
+        self.blocks4 = nn.Sequential(
+            *[
+                DepthwiseSeparable(
+                    num_channels=make_divisible(in_c * scale),
+                    num_filters=make_divisible(out_c * scale),
+                    dw_size=k,
+                    stride=s,
+                    use_se=se,
+                )
+                for i, (k, in_c, out_c, s, se) in enumerate(NET_CONFIG["blocks4"])
+            ]
+        )
+
+        self.blocks5 = nn.Sequential(
+            *[
+                DepthwiseSeparable(
+                    num_channels=make_divisible(in_c * scale),
+                    num_filters=make_divisible(out_c * scale),
+                    dw_size=k,
+                    stride=s,
+                    use_se=se,
+                )
+                for i, (k, in_c, out_c, s, se) in enumerate(NET_CONFIG["blocks5"])
+            ]
+        )
+
+        self.blocks6 = nn.Sequential(
+            *[
+                DepthwiseSeparable(
+                    num_channels=make_divisible(in_c * scale),
+                    num_filters=make_divisible(out_c * scale),
+                    dw_size=k,
+                    stride=s,
+                    use_se=se,
+                )
+                for i, (k, in_c, out_c, s, se) in enumerate(NET_CONFIG["blocks6"])
+            ]
+        )
+
+        if pretrained:
+            self._load_pretrained(
+                MODEL_URLS["PPLCNet_x{}".format(scale)], use_ssld=use_ssld
+            )
+
+    def forward(self, x):
+        outs = []
+        x = self.conv1(x)
+        x = self.blocks2(x)
+        x = self.blocks3(x)
+        outs.append(x)
+        x = self.blocks4(x)
+        outs.append(x)
+        x = self.blocks5(x)
+        outs.append(x)
+        x = self.blocks6(x)
+        outs.append(x)
+        return outs
+
+    def _load_pretrained(self, pretrained_url, use_ssld=False):
+        if use_ssld:
+            pretrained_url = pretrained_url.replace("_pretrained", "_ssld_pretrained")
+        print(pretrained_url)
+        local_weight_path = get_path_from_url(
+            pretrained_url, os.path.expanduser("~/.paddleclas/weights")
+        )
+        param_state_dict = paddle.load(local_weight_path)
+        self.set_dict(param_state_dict)
+        return
diff --git a/ppocr/modeling/backbones/det_pp_lcnet_v2.py b/ppocr/modeling/backbones/det_pp_lcnet_v2.py
new file mode 100644
index 0000000..5b5a568
--- /dev/null
+++ b/ppocr/modeling/backbones/det_pp_lcnet_v2.py
@@ -0,0 +1,358 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import, division, print_function
+import os
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+from paddle.nn import AdaptiveAvgPool2D, BatchNorm2D, Conv2D, Dropout, Linear
+from paddle.regularizer import L2Decay
+from paddle.nn.initializer import KaimingNormal
+from paddle.utils.download import get_path_from_url
+
+MODEL_URLS = {
+    "PPLCNetV2_small": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPLCNetV2_small_ssld_pretrained.pdparams",
+    "PPLCNetV2_base": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPLCNetV2_base_ssld_pretrained.pdparams",
+    "PPLCNetV2_large": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPLCNetV2_large_ssld_pretrained.pdparams",
+}
+
+__all__ = list(MODEL_URLS.keys())
+
+NET_CONFIG = {
+    # in_channels, kernel_size, split_pw, use_rep, use_se, use_shortcut
+    "stage1": [64, 3, False, False, False, False],
+    "stage2": [128, 3, False, False, False, False],
+    "stage3": [256, 5, True, True, True, False],
+    "stage4": [512, 5, False, True, False, True],
+}
+
+
+def make_divisible(v, divisor=8, min_value=None):
+    if min_value is None:
+        min_value = divisor
+    new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
+    if new_v < 0.9 * v:
+        new_v += divisor
+    return new_v
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self, in_channels, out_channels, kernel_size, stride, groups=1, use_act=True
+    ):
+        super().__init__()
+        self.use_act = use_act
+        self.conv = Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=(kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(initializer=KaimingNormal()),
+            bias_attr=False,
+        )
+
+        self.bn = BatchNorm2D(
+            out_channels,
+            weight_attr=ParamAttr(regularizer=L2Decay(0.0)),
+            bias_attr=ParamAttr(regularizer=L2Decay(0.0)),
+        )
+        if self.use_act:
+            self.act = nn.ReLU()
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        if self.use_act:
+            x = self.act(x)
+        return x
+
+
+class SEModule(nn.Layer):
+    def __init__(self, channel, reduction=4):
+        super().__init__()
+        self.avg_pool = AdaptiveAvgPool2D(1)
+        self.conv1 = Conv2D(
+            in_channels=channel,
+            out_channels=channel // reduction,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        )
+        self.relu = nn.ReLU()
+        self.conv2 = Conv2D(
+            in_channels=channel // reduction,
+            out_channels=channel,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        )
+        self.hardsigmoid = nn.Sigmoid()
+
+    def forward(self, x):
+        identity = x
+        x = self.avg_pool(x)
+        x = self.conv1(x)
+        x = self.relu(x)
+        x = self.conv2(x)
+        x = self.hardsigmoid(x)
+        x = paddle.multiply(x=identity, y=x)
+        return x
+
+
+class RepDepthwiseSeparable(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        stride,
+        dw_size=3,
+        split_pw=False,
+        use_rep=False,
+        use_se=False,
+        use_shortcut=False,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.is_repped = False
+
+        self.dw_size = dw_size
+        self.split_pw = split_pw
+        self.use_rep = use_rep
+        self.use_se = use_se
+        self.use_shortcut = (
+            True
+            if use_shortcut and stride == 1 and in_channels == out_channels
+            else False
+        )
+
+        if self.use_rep:
+            self.dw_conv_list = nn.LayerList()
+            for kernel_size in range(self.dw_size, 0, -2):
+                if kernel_size == 1 and stride != 1:
+                    continue
+                dw_conv = ConvBNLayer(
+                    in_channels=in_channels,
+                    out_channels=in_channels,
+                    kernel_size=kernel_size,
+                    stride=stride,
+                    groups=in_channels,
+                    use_act=False,
+                )
+                self.dw_conv_list.append(dw_conv)
+            self.dw_conv = nn.Conv2D(
+                in_channels=in_channels,
+                out_channels=in_channels,
+                kernel_size=dw_size,
+                stride=stride,
+                padding=(dw_size - 1) // 2,
+                groups=in_channels,
+            )
+        else:
+            self.dw_conv = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=in_channels,
+                kernel_size=dw_size,
+                stride=stride,
+                groups=in_channels,
+            )
+
+        self.act = nn.ReLU()
+
+        if use_se:
+            self.se = SEModule(in_channels)
+
+        if self.split_pw:
+            pw_ratio = 0.5
+            self.pw_conv_1 = ConvBNLayer(
+                in_channels=in_channels,
+                kernel_size=1,
+                out_channels=int(out_channels * pw_ratio),
+                stride=1,
+            )
+            self.pw_conv_2 = ConvBNLayer(
+                in_channels=int(out_channels * pw_ratio),
+                kernel_size=1,
+                out_channels=out_channels,
+                stride=1,
+            )
+        else:
+            self.pw_conv = ConvBNLayer(
+                in_channels=in_channels,
+                kernel_size=1,
+                out_channels=out_channels,
+                stride=1,
+            )
+
+    def forward(self, x):
+        if self.use_rep:
+            input_x = x
+            if self.is_repped:
+                x = self.act(self.dw_conv(x))
+            else:
+                y = self.dw_conv_list[0](x)
+                for dw_conv in self.dw_conv_list[1:]:
+                    y += dw_conv(x)
+                x = self.act(y)
+        else:
+            x = self.dw_conv(x)
+
+        if self.use_se:
+            x = self.se(x)
+        if self.split_pw:
+            x = self.pw_conv_1(x)
+            x = self.pw_conv_2(x)
+        else:
+            x = self.pw_conv(x)
+        if self.use_shortcut:
+            x = x + input_x
+        return x
+
+    def re_parameterize(self):
+        if self.use_rep:
+            self.is_repped = True
+            kernel, bias = self._get_equivalent_kernel_bias()
+            self.dw_conv.weight.set_value(kernel)
+            self.dw_conv.bias.set_value(bias)
+
+    def _get_equivalent_kernel_bias(self):
+        kernel_sum = 0
+        bias_sum = 0
+        for dw_conv in self.dw_conv_list:
+            kernel, bias = self._fuse_bn_tensor(dw_conv)
+            kernel = self._pad_tensor(kernel, to_size=self.dw_size)
+            kernel_sum += kernel
+            bias_sum += bias
+        return kernel_sum, bias_sum
+
+    def _fuse_bn_tensor(self, branch):
+        kernel = branch.conv.weight
+        running_mean = branch.bn._mean
+        running_var = branch.bn._variance
+        gamma = branch.bn.weight
+        beta = branch.bn.bias
+        eps = branch.bn._epsilon
+        std = (running_var + eps).sqrt()
+        t = (gamma / std).reshape((-1, 1, 1, 1))
+        return kernel * t, beta - running_mean * gamma / std
+
+    def _pad_tensor(self, tensor, to_size):
+        from_size = tensor.shape[-1]
+        if from_size == to_size:
+            return tensor
+        pad = (to_size - from_size) // 2
+        return F.pad(tensor, [pad, pad, pad, pad])
+
+
+class PPLCNetV2(nn.Layer):
+    def __init__(self, scale, depths, out_indx=[1, 2, 3, 4], **kwargs):
+        super().__init__(**kwargs)
+        self.scale = scale
+        self.out_channels = [
+            # int(NET_CONFIG["blocks3"][-1][2] * scale),
+            int(NET_CONFIG["stage1"][0] * scale * 2),
+            int(NET_CONFIG["stage2"][0] * scale * 2),
+            int(NET_CONFIG["stage3"][0] * scale * 2),
+            int(NET_CONFIG["stage4"][0] * scale * 2),
+        ]
+        self.stem = nn.Sequential(
+            *[
+                ConvBNLayer(
+                    in_channels=3,
+                    kernel_size=3,
+                    out_channels=make_divisible(32 * scale),
+                    stride=2,
+                ),
+                RepDepthwiseSeparable(
+                    in_channels=make_divisible(32 * scale),
+                    out_channels=make_divisible(64 * scale),
+                    stride=1,
+                    dw_size=3,
+                ),
+            ]
+        )
+        self.out_indx = out_indx
+        # stages
+        self.stages = nn.LayerList()
+        for depth_idx, k in enumerate(NET_CONFIG):
+            (
+                in_channels,
+                kernel_size,
+                split_pw,
+                use_rep,
+                use_se,
+                use_shortcut,
+            ) = NET_CONFIG[k]
+            self.stages.append(
+                nn.Sequential(
+                    *[
+                        RepDepthwiseSeparable(
+                            in_channels=make_divisible(
+                                (in_channels if i == 0 else in_channels * 2) * scale
+                            ),
+                            out_channels=make_divisible(in_channels * 2 * scale),
+                            stride=2 if i == 0 else 1,
+                            dw_size=kernel_size,
+                            split_pw=split_pw,
+                            use_rep=use_rep,
+                            use_se=use_se,
+                            use_shortcut=use_shortcut,
+                        )
+                        for i in range(depths[depth_idx])
+                    ]
+                )
+            )
+
+        # if pretrained:
+        self._load_pretrained(MODEL_URLS["PPLCNetV2_base"], use_ssld=True)
+
+    def forward(self, x):
+        x = self.stem(x)
+        i = 1
+        outs = []
+        for stage in self.stages:
+            x = stage(x)
+            if i in self.out_indx:
+                outs.append(x)
+            i += 1
+        return outs
+
+    def _load_pretrained(self, pretrained_url, use_ssld=False):
+        print(pretrained_url)
+        local_weight_path = get_path_from_url(
+            pretrained_url, os.path.expanduser("~/.paddleclas/weights")
+        )
+        param_state_dict = paddle.load(local_weight_path)
+        self.set_dict(param_state_dict)
+        print("load pretrain ssd success!")
+        return
+
+
+def PPLCNetV2_base(in_channels=3, **kwargs):
+    """
+    PPLCNetV2_base
+    Args:
+        pretrained: bool=False or str. If `True` load pretrained parameters, `False` otherwise.
+                    If str, means the path of the pretrained model.
+        use_ssld: bool=False. Whether using distillation pretrained model when pretrained=True.
+    Returns:
+        model: nn.Layer. Specific `PPLCNetV2_base` model depends on args.
+    """
+    model = PPLCNetV2(scale=1.0, depths=[2, 2, 6, 2], **kwargs)
+    return model
diff --git a/ppocr/modeling/backbones/det_resnet.py b/ppocr/modeling/backbones/det_resnet.py
new file mode 100644
index 0000000..3842829
--- /dev/null
+++ b/ppocr/modeling/backbones/det_resnet.py
@@ -0,0 +1,235 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+import paddle
+from paddle import ParamAttr
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle.nn import Conv2D, BatchNorm, Linear, Dropout
+from paddle.nn import AdaptiveAvgPool2D, MaxPool2D, AvgPool2D
+from paddle.nn.initializer import Uniform
+
+import math
+
+from paddle.vision.ops import DeformConv2D
+from paddle.regularizer import L2Decay
+from paddle.nn.initializer import Normal, Constant, XavierUniform
+from .det_resnet_vd import DeformableConvV2, ConvBNLayer
+
+
+class BottleneckBlock(nn.Layer):
+    def __init__(self, num_channels, num_filters, stride, shortcut=True, is_dcn=False):
+        super(BottleneckBlock, self).__init__()
+
+        self.conv0 = ConvBNLayer(
+            in_channels=num_channels,
+            out_channels=num_filters,
+            kernel_size=1,
+            act="relu",
+        )
+        self.conv1 = ConvBNLayer(
+            in_channels=num_filters,
+            out_channels=num_filters,
+            kernel_size=3,
+            stride=stride,
+            act="relu",
+            is_dcn=is_dcn,
+            dcn_groups=1,
+        )
+        self.conv2 = ConvBNLayer(
+            in_channels=num_filters,
+            out_channels=num_filters * 4,
+            kernel_size=1,
+            act=None,
+        )
+
+        if not shortcut:
+            self.short = ConvBNLayer(
+                in_channels=num_channels,
+                out_channels=num_filters * 4,
+                kernel_size=1,
+                stride=stride,
+            )
+
+        self.shortcut = shortcut
+
+        self._num_channels_out = num_filters * 4
+
+    def forward(self, inputs):
+        y = self.conv0(inputs)
+        conv1 = self.conv1(y)
+        conv2 = self.conv2(conv1)
+
+        if self.shortcut:
+            short = inputs
+        else:
+            short = self.short(inputs)
+
+        y = paddle.add(x=short, y=conv2)
+        y = F.relu(y)
+        return y
+
+
+class BasicBlock(nn.Layer):
+    def __init__(self, num_channels, num_filters, stride, shortcut=True, name=None):
+        super(BasicBlock, self).__init__()
+        self.stride = stride
+        self.conv0 = ConvBNLayer(
+            in_channels=num_channels,
+            out_channels=num_filters,
+            kernel_size=3,
+            stride=stride,
+            act="relu",
+        )
+        self.conv1 = ConvBNLayer(
+            in_channels=num_filters, out_channels=num_filters, kernel_size=3, act=None
+        )
+
+        if not shortcut:
+            self.short = ConvBNLayer(
+                in_channels=num_channels,
+                out_channels=num_filters,
+                kernel_size=1,
+                stride=stride,
+            )
+
+        self.shortcut = shortcut
+
+    def forward(self, inputs):
+        y = self.conv0(inputs)
+        conv1 = self.conv1(y)
+
+        if self.shortcut:
+            short = inputs
+        else:
+            short = self.short(inputs)
+        y = paddle.add(x=short, y=conv1)
+        y = F.relu(y)
+        return y
+
+
+class ResNet(nn.Layer):
+    def __init__(self, in_channels=3, layers=50, out_indices=None, dcn_stage=None):
+        super(ResNet, self).__init__()
+
+        self.layers = layers
+        self.input_image_channel = in_channels
+
+        supported_layers = [18, 34, 50, 101, 152]
+        assert (
+            layers in supported_layers
+        ), "supported layers are {} but input layer is {}".format(
+            supported_layers, layers
+        )
+
+        if layers == 18:
+            depth = [2, 2, 2, 2]
+        elif layers == 34 or layers == 50:
+            depth = [3, 4, 6, 3]
+        elif layers == 101:
+            depth = [3, 4, 23, 3]
+        elif layers == 152:
+            depth = [3, 8, 36, 3]
+        num_channels = [64, 256, 512, 1024] if layers >= 50 else [64, 64, 128, 256]
+        num_filters = [64, 128, 256, 512]
+
+        self.dcn_stage = (
+            dcn_stage if dcn_stage is not None else [False, False, False, False]
+        )
+        self.out_indices = out_indices if out_indices is not None else [0, 1, 2, 3]
+
+        self.conv = ConvBNLayer(
+            in_channels=self.input_image_channel,
+            out_channels=64,
+            kernel_size=7,
+            stride=2,
+            act="relu",
+        )
+        self.pool2d_max = MaxPool2D(
+            kernel_size=3,
+            stride=2,
+            padding=1,
+        )
+
+        self.stages = []
+        self.out_channels = []
+        if layers >= 50:
+            for block in range(len(depth)):
+                shortcut = False
+                block_list = []
+                is_dcn = self.dcn_stage[block]
+                for i in range(depth[block]):
+                    if layers in [101, 152] and block == 2:
+                        if i == 0:
+                            conv_name = "res" + str(block + 2) + "a"
+                        else:
+                            conv_name = "res" + str(block + 2) + "b" + str(i)
+                    else:
+                        conv_name = "res" + str(block + 2) + chr(97 + i)
+                    bottleneck_block = self.add_sublayer(
+                        conv_name,
+                        BottleneckBlock(
+                            num_channels=(
+                                num_channels[block]
+                                if i == 0
+                                else num_filters[block] * 4
+                            ),
+                            num_filters=num_filters[block],
+                            stride=2 if i == 0 and block != 0 else 1,
+                            shortcut=shortcut,
+                            is_dcn=is_dcn,
+                        ),
+                    )
+                    block_list.append(bottleneck_block)
+                    shortcut = True
+                if block in self.out_indices:
+                    self.out_channels.append(num_filters[block] * 4)
+                self.stages.append(nn.Sequential(*block_list))
+        else:
+            for block in range(len(depth)):
+                shortcut = False
+                block_list = []
+                for i in range(depth[block]):
+                    conv_name = "res" + str(block + 2) + chr(97 + i)
+                    basic_block = self.add_sublayer(
+                        conv_name,
+                        BasicBlock(
+                            num_channels=(
+                                num_channels[block] if i == 0 else num_filters[block]
+                            ),
+                            num_filters=num_filters[block],
+                            stride=2 if i == 0 and block != 0 else 1,
+                            shortcut=shortcut,
+                        ),
+                    )
+                    block_list.append(basic_block)
+                    shortcut = True
+                if block in self.out_indices:
+                    self.out_channels.append(num_filters[block])
+                self.stages.append(nn.Sequential(*block_list))
+
+    def forward(self, inputs):
+        y = self.conv(inputs)
+        y = self.pool2d_max(y)
+        out = []
+        for i, block in enumerate(self.stages):
+            y = block(y)
+            if i in self.out_indices:
+                out.append(y)
+        return out
diff --git a/ppocr/modeling/backbones/det_resnet_vd.py b/ppocr/modeling/backbones/det_resnet_vd.py
new file mode 100644
index 0000000..070ba3c
--- /dev/null
+++ b/ppocr/modeling/backbones/det_resnet_vd.py
@@ -0,0 +1,369 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import ParamAttr
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+from paddle.vision.ops import DeformConv2D
+from paddle.regularizer import L2Decay
+from paddle.nn.initializer import Normal, Constant, XavierUniform
+
+__all__ = ["ResNet_vd", "ConvBNLayer", "DeformableConvV2"]
+
+
+class DeformableConvV2(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride=1,
+        padding=0,
+        dilation=1,
+        groups=1,
+        weight_attr=None,
+        bias_attr=None,
+        lr_scale=1,
+        regularizer=None,
+        skip_quant=False,
+        dcn_bias_regularizer=L2Decay(0.0),
+        dcn_bias_lr_scale=2.0,
+    ):
+        super(DeformableConvV2, self).__init__()
+        self.offset_channel = 2 * kernel_size**2 * groups
+        self.mask_channel = kernel_size**2 * groups
+
+        if bias_attr:
+            # in FCOS-DCN head, specifically need learning_rate and regularizer
+            dcn_bias_attr = ParamAttr(
+                initializer=Constant(value=0),
+                regularizer=dcn_bias_regularizer,
+                learning_rate=dcn_bias_lr_scale,
+            )
+        else:
+            # in ResNet backbone, do not need bias
+            dcn_bias_attr = False
+        self.conv_dcn = DeformConv2D(
+            in_channels,
+            out_channels,
+            kernel_size,
+            stride=stride,
+            padding=(kernel_size - 1) // 2 * dilation,
+            dilation=dilation,
+            deformable_groups=groups,
+            weight_attr=weight_attr,
+            bias_attr=dcn_bias_attr,
+        )
+
+        if lr_scale == 1 and regularizer is None:
+            offset_bias_attr = ParamAttr(initializer=Constant(0.0))
+        else:
+            offset_bias_attr = ParamAttr(
+                initializer=Constant(0.0),
+                learning_rate=lr_scale,
+                regularizer=regularizer,
+            )
+        self.conv_offset = nn.Conv2D(
+            in_channels,
+            groups * 3 * kernel_size**2,
+            kernel_size,
+            stride=stride,
+            padding=(kernel_size - 1) // 2,
+            weight_attr=ParamAttr(initializer=Constant(0.0)),
+            bias_attr=offset_bias_attr,
+        )
+        if skip_quant:
+            self.conv_offset.skip_quant = True
+
+    def forward(self, x):
+        offset_mask = self.conv_offset(x)
+        offset, mask = paddle.split(
+            offset_mask,
+            num_or_sections=[self.offset_channel, self.mask_channel],
+            axis=1,
+        )
+        mask = F.sigmoid(mask)
+        y = self.conv_dcn(x, offset, mask=mask)
+        return y
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride=1,
+        groups=1,
+        dcn_groups=1,
+        is_vd_mode=False,
+        act=None,
+        is_dcn=False,
+    ):
+        super(ConvBNLayer, self).__init__()
+
+        self.is_vd_mode = is_vd_mode
+        self._pool2d_avg = nn.AvgPool2D(
+            kernel_size=2, stride=2, padding=0, ceil_mode=True
+        )
+        if not is_dcn:
+            self._conv = nn.Conv2D(
+                in_channels=in_channels,
+                out_channels=out_channels,
+                kernel_size=kernel_size,
+                stride=stride,
+                padding=(kernel_size - 1) // 2,
+                groups=groups,
+                bias_attr=False,
+            )
+        else:
+            self._conv = DeformableConvV2(
+                in_channels=in_channels,
+                out_channels=out_channels,
+                kernel_size=kernel_size,
+                stride=stride,
+                padding=(kernel_size - 1) // 2,
+                groups=dcn_groups,  # groups,
+                bias_attr=False,
+            )
+        self._batch_norm = nn.BatchNorm(out_channels, act=act)
+
+    def forward(self, inputs):
+        if self.is_vd_mode:
+            inputs = self._pool2d_avg(inputs)
+        y = self._conv(inputs)
+        y = self._batch_norm(y)
+        return y
+
+
+class BottleneckBlock(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        stride,
+        shortcut=True,
+        if_first=False,
+        is_dcn=False,
+    ):
+        super(BottleneckBlock, self).__init__()
+
+        self.conv0 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=1,
+            act="relu",
+        )
+        self.conv1 = ConvBNLayer(
+            in_channels=out_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            stride=stride,
+            act="relu",
+            is_dcn=is_dcn,
+            dcn_groups=2,
+        )
+        self.conv2 = ConvBNLayer(
+            in_channels=out_channels,
+            out_channels=out_channels * 4,
+            kernel_size=1,
+            act=None,
+        )
+
+        if not shortcut:
+            self.short = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=out_channels * 4,
+                kernel_size=1,
+                stride=1,
+                is_vd_mode=False if if_first else True,
+            )
+
+        self.shortcut = shortcut
+
+    def forward(self, inputs):
+        y = self.conv0(inputs)
+        conv1 = self.conv1(y)
+        conv2 = self.conv2(conv1)
+
+        if self.shortcut:
+            short = inputs
+        else:
+            short = self.short(inputs)
+        y = paddle.add(x=short, y=conv2)
+        y = F.relu(y)
+        return y
+
+
+class BasicBlock(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        stride,
+        shortcut=True,
+        if_first=False,
+    ):
+        super(BasicBlock, self).__init__()
+        self.stride = stride
+        self.conv0 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            stride=stride,
+            act="relu",
+        )
+        self.conv1 = ConvBNLayer(
+            in_channels=out_channels, out_channels=out_channels, kernel_size=3, act=None
+        )
+
+        if not shortcut:
+            self.short = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=out_channels,
+                kernel_size=1,
+                stride=1,
+                is_vd_mode=False if if_first else True,
+            )
+
+        self.shortcut = shortcut
+
+    def forward(self, inputs):
+        y = self.conv0(inputs)
+        conv1 = self.conv1(y)
+
+        if self.shortcut:
+            short = inputs
+        else:
+            short = self.short(inputs)
+        y = paddle.add(x=short, y=conv1)
+        y = F.relu(y)
+        return y
+
+
+class ResNet_vd(nn.Layer):
+    def __init__(
+        self, in_channels=3, layers=50, dcn_stage=None, out_indices=None, **kwargs
+    ):
+        super(ResNet_vd, self).__init__()
+
+        self.layers = layers
+        supported_layers = [18, 34, 50, 101, 152, 200]
+        assert (
+            layers in supported_layers
+        ), "supported layers are {} but input layer is {}".format(
+            supported_layers, layers
+        )
+
+        if layers == 18:
+            depth = [2, 2, 2, 2]
+        elif layers == 34 or layers == 50:
+            depth = [3, 4, 6, 3]
+        elif layers == 101:
+            depth = [3, 4, 23, 3]
+        elif layers == 152:
+            depth = [3, 8, 36, 3]
+        elif layers == 200:
+            depth = [3, 12, 48, 3]
+        num_channels = [64, 256, 512, 1024] if layers >= 50 else [64, 64, 128, 256]
+        num_filters = [64, 128, 256, 512]
+
+        self.dcn_stage = (
+            dcn_stage if dcn_stage is not None else [False, False, False, False]
+        )
+        self.out_indices = out_indices if out_indices is not None else [0, 1, 2, 3]
+
+        self.conv1_1 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=32,
+            kernel_size=3,
+            stride=2,
+            act="relu",
+        )
+        self.conv1_2 = ConvBNLayer(
+            in_channels=32, out_channels=32, kernel_size=3, stride=1, act="relu"
+        )
+        self.conv1_3 = ConvBNLayer(
+            in_channels=32, out_channels=64, kernel_size=3, stride=1, act="relu"
+        )
+        self.pool2d_max = nn.MaxPool2D(kernel_size=3, stride=2, padding=1)
+
+        self.stages = []
+        self.out_channels = []
+        if layers >= 50:
+            for block in range(len(depth)):
+                block_list = []
+                shortcut = False
+                is_dcn = self.dcn_stage[block]
+                for i in range(depth[block]):
+                    bottleneck_block = self.add_sublayer(
+                        "bb_%d_%d" % (block, i),
+                        BottleneckBlock(
+                            in_channels=(
+                                num_channels[block]
+                                if i == 0
+                                else num_filters[block] * 4
+                            ),
+                            out_channels=num_filters[block],
+                            stride=2 if i == 0 and block != 0 else 1,
+                            shortcut=shortcut,
+                            if_first=block == i == 0,
+                            is_dcn=is_dcn,
+                        ),
+                    )
+                    shortcut = True
+                    block_list.append(bottleneck_block)
+                if block in self.out_indices:
+                    self.out_channels.append(num_filters[block] * 4)
+                self.stages.append(nn.Sequential(*block_list))
+        else:
+            for block in range(len(depth)):
+                block_list = []
+                shortcut = False
+                for i in range(depth[block]):
+                    basic_block = self.add_sublayer(
+                        "bb_%d_%d" % (block, i),
+                        BasicBlock(
+                            in_channels=(
+                                num_channels[block] if i == 0 else num_filters[block]
+                            ),
+                            out_channels=num_filters[block],
+                            stride=2 if i == 0 and block != 0 else 1,
+                            shortcut=shortcut,
+                            if_first=block == i == 0,
+                        ),
+                    )
+                    shortcut = True
+                    block_list.append(basic_block)
+                if block in self.out_indices:
+                    self.out_channels.append(num_filters[block])
+                self.stages.append(nn.Sequential(*block_list))
+
+    def forward(self, inputs):
+        y = self.conv1_1(inputs)
+        y = self.conv1_2(y)
+        y = self.conv1_3(y)
+        y = self.pool2d_max(y)
+        out = []
+        for i, block in enumerate(self.stages):
+            y = block(y)
+            if i in self.out_indices:
+                out.append(y)
+        return out
diff --git a/ppocr/modeling/backbones/det_resnet_vd_sast.py b/ppocr/modeling/backbones/det_resnet_vd_sast.py
new file mode 100644
index 0000000..6769dd9
--- /dev/null
+++ b/ppocr/modeling/backbones/det_resnet_vd_sast.py
@@ -0,0 +1,314 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import ParamAttr
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+__all__ = ["ResNet_SAST"]
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride=1,
+        groups=1,
+        is_vd_mode=False,
+        act=None,
+        name=None,
+    ):
+        super(ConvBNLayer, self).__init__()
+
+        self.is_vd_mode = is_vd_mode
+        self._pool2d_avg = nn.AvgPool2D(
+            kernel_size=2, stride=2, padding=0, ceil_mode=True
+        )
+        self._conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=(kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(name=name + "_weights"),
+            bias_attr=False,
+        )
+        if name == "conv1":
+            bn_name = "bn_" + name
+        else:
+            bn_name = "bn" + name[3:]
+        self._batch_norm = nn.BatchNorm(
+            out_channels,
+            act=act,
+            param_attr=ParamAttr(name=bn_name + "_scale"),
+            bias_attr=ParamAttr(bn_name + "_offset"),
+            moving_mean_name=bn_name + "_mean",
+            moving_variance_name=bn_name + "_variance",
+        )
+
+    def forward(self, inputs):
+        if self.is_vd_mode:
+            inputs = self._pool2d_avg(inputs)
+        y = self._conv(inputs)
+        y = self._batch_norm(y)
+        return y
+
+
+class BottleneckBlock(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        stride,
+        shortcut=True,
+        if_first=False,
+        name=None,
+    ):
+        super(BottleneckBlock, self).__init__()
+
+        self.conv0 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=1,
+            act="relu",
+            name=name + "_branch2a",
+        )
+        self.conv1 = ConvBNLayer(
+            in_channels=out_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            stride=stride,
+            act="relu",
+            name=name + "_branch2b",
+        )
+        self.conv2 = ConvBNLayer(
+            in_channels=out_channels,
+            out_channels=out_channels * 4,
+            kernel_size=1,
+            act=None,
+            name=name + "_branch2c",
+        )
+
+        if not shortcut:
+            self.short = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=out_channels * 4,
+                kernel_size=1,
+                stride=1,
+                is_vd_mode=False if if_first else True,
+                name=name + "_branch1",
+            )
+
+        self.shortcut = shortcut
+
+    def forward(self, inputs):
+        y = self.conv0(inputs)
+        conv1 = self.conv1(y)
+        conv2 = self.conv2(conv1)
+
+        if self.shortcut:
+            short = inputs
+        else:
+            short = self.short(inputs)
+        y = paddle.add(x=short, y=conv2)
+        y = F.relu(y)
+        return y
+
+
+class BasicBlock(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        stride,
+        shortcut=True,
+        if_first=False,
+        name=None,
+    ):
+        super(BasicBlock, self).__init__()
+        self.stride = stride
+        self.conv0 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            stride=stride,
+            act="relu",
+            name=name + "_branch2a",
+        )
+        self.conv1 = ConvBNLayer(
+            in_channels=out_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            act=None,
+            name=name + "_branch2b",
+        )
+
+        if not shortcut:
+            self.short = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=out_channels,
+                kernel_size=1,
+                stride=1,
+                is_vd_mode=False if if_first else True,
+                name=name + "_branch1",
+            )
+
+        self.shortcut = shortcut
+
+    def forward(self, inputs):
+        y = self.conv0(inputs)
+        conv1 = self.conv1(y)
+
+        if self.shortcut:
+            short = inputs
+        else:
+            short = self.short(inputs)
+        y = paddle.add(x=short, y=conv1)
+        y = F.relu(y)
+        return y
+
+
+class ResNet_SAST(nn.Layer):
+    def __init__(self, in_channels=3, layers=50, **kwargs):
+        super(ResNet_SAST, self).__init__()
+
+        self.layers = layers
+        supported_layers = [18, 34, 50, 101, 152, 200]
+        assert (
+            layers in supported_layers
+        ), "supported layers are {} but input layer is {}".format(
+            supported_layers, layers
+        )
+
+        if layers == 18:
+            depth = [2, 2, 2, 2]
+        elif layers == 34 or layers == 50:
+            # depth = [3, 4, 6, 3]
+            depth = [3, 4, 6, 3, 3]
+        elif layers == 101:
+            depth = [3, 4, 23, 3]
+        elif layers == 152:
+            depth = [3, 8, 36, 3]
+        elif layers == 200:
+            depth = [3, 12, 48, 3]
+        # num_channels = [64, 256, 512,
+        #                 1024] if layers >= 50 else [64, 64, 128, 256]
+        # num_filters = [64, 128, 256, 512]
+        num_channels = (
+            [64, 256, 512, 1024, 2048] if layers >= 50 else [64, 64, 128, 256]
+        )
+        num_filters = [64, 128, 256, 512, 512]
+
+        self.conv1_1 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=32,
+            kernel_size=3,
+            stride=2,
+            act="relu",
+            name="conv1_1",
+        )
+        self.conv1_2 = ConvBNLayer(
+            in_channels=32,
+            out_channels=32,
+            kernel_size=3,
+            stride=1,
+            act="relu",
+            name="conv1_2",
+        )
+        self.conv1_3 = ConvBNLayer(
+            in_channels=32,
+            out_channels=64,
+            kernel_size=3,
+            stride=1,
+            act="relu",
+            name="conv1_3",
+        )
+        self.pool2d_max = nn.MaxPool2D(kernel_size=3, stride=2, padding=1)
+
+        self.stages = []
+        self.out_channels = [3, 64]
+        if layers >= 50:
+            for block in range(len(depth)):
+                block_list = []
+                shortcut = False
+                for i in range(depth[block]):
+                    if layers in [101, 152] and block == 2:
+                        if i == 0:
+                            conv_name = "res" + str(block + 2) + "a"
+                        else:
+                            conv_name = "res" + str(block + 2) + "b" + str(i)
+                    else:
+                        conv_name = "res" + str(block + 2) + chr(97 + i)
+                    bottleneck_block = self.add_sublayer(
+                        "bb_%d_%d" % (block, i),
+                        BottleneckBlock(
+                            in_channels=(
+                                num_channels[block]
+                                if i == 0
+                                else num_filters[block] * 4
+                            ),
+                            out_channels=num_filters[block],
+                            stride=2 if i == 0 and block != 0 else 1,
+                            shortcut=shortcut,
+                            if_first=block == i == 0,
+                            name=conv_name,
+                        ),
+                    )
+                    shortcut = True
+                    block_list.append(bottleneck_block)
+                self.out_channels.append(num_filters[block] * 4)
+                self.stages.append(nn.Sequential(*block_list))
+        else:
+            for block in range(len(depth)):
+                block_list = []
+                shortcut = False
+                for i in range(depth[block]):
+                    conv_name = "res" + str(block + 2) + chr(97 + i)
+                    basic_block = self.add_sublayer(
+                        "bb_%d_%d" % (block, i),
+                        BasicBlock(
+                            in_channels=(
+                                num_channels[block] if i == 0 else num_filters[block]
+                            ),
+                            out_channels=num_filters[block],
+                            stride=2 if i == 0 and block != 0 else 1,
+                            shortcut=shortcut,
+                            if_first=block == i == 0,
+                            name=conv_name,
+                        ),
+                    )
+                    shortcut = True
+                    block_list.append(basic_block)
+                self.out_channels.append(num_filters[block])
+                self.stages.append(nn.Sequential(*block_list))
+
+    def forward(self, inputs):
+        out = [inputs]
+        y = self.conv1_1(inputs)
+        y = self.conv1_2(y)
+        y = self.conv1_3(y)
+        out.append(y)
+        y = self.pool2d_max(y)
+        for block in self.stages:
+            y = block(y)
+            out.append(y)
+        return out
diff --git a/ppocr/modeling/backbones/e2e_resnet_vd_pg.py b/ppocr/modeling/backbones/e2e_resnet_vd_pg.py
new file mode 100644
index 0000000..25738cd
--- /dev/null
+++ b/ppocr/modeling/backbones/e2e_resnet_vd_pg.py
@@ -0,0 +1,292 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import ParamAttr
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+__all__ = ["ResNet"]
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride=1,
+        groups=1,
+        is_vd_mode=False,
+        act=None,
+        name=None,
+    ):
+        super(ConvBNLayer, self).__init__()
+
+        self.is_vd_mode = is_vd_mode
+        self._pool2d_avg = nn.AvgPool2D(
+            kernel_size=2, stride=2, padding=0, ceil_mode=True
+        )
+        self._conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=(kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(name=name + "_weights"),
+            bias_attr=False,
+        )
+        if name == "conv1":
+            bn_name = "bn_" + name
+        else:
+            bn_name = "bn" + name[3:]
+        self._batch_norm = nn.BatchNorm(
+            out_channels,
+            act=act,
+            param_attr=ParamAttr(name=bn_name + "_scale"),
+            bias_attr=ParamAttr(bn_name + "_offset"),
+            moving_mean_name=bn_name + "_mean",
+            moving_variance_name=bn_name + "_variance",
+        )
+
+    def forward(self, inputs):
+        y = self._conv(inputs)
+        y = self._batch_norm(y)
+        return y
+
+
+class BottleneckBlock(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        stride,
+        shortcut=True,
+        if_first=False,
+        name=None,
+    ):
+        super(BottleneckBlock, self).__init__()
+
+        self.conv0 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=1,
+            act="relu",
+            name=name + "_branch2a",
+        )
+        self.conv1 = ConvBNLayer(
+            in_channels=out_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            stride=stride,
+            act="relu",
+            name=name + "_branch2b",
+        )
+        self.conv2 = ConvBNLayer(
+            in_channels=out_channels,
+            out_channels=out_channels * 4,
+            kernel_size=1,
+            act=None,
+            name=name + "_branch2c",
+        )
+
+        if not shortcut:
+            self.short = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=out_channels * 4,
+                kernel_size=1,
+                stride=stride,
+                is_vd_mode=False if if_first else True,
+                name=name + "_branch1",
+            )
+
+        self.shortcut = shortcut
+
+    def forward(self, inputs):
+        y = self.conv0(inputs)
+        conv1 = self.conv1(y)
+        conv2 = self.conv2(conv1)
+
+        if self.shortcut:
+            short = inputs
+        else:
+            short = self.short(inputs)
+        y = paddle.add(x=short, y=conv2)
+        y = F.relu(y)
+        return y
+
+
+class BasicBlock(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        stride,
+        shortcut=True,
+        if_first=False,
+        name=None,
+    ):
+        super(BasicBlock, self).__init__()
+        self.stride = stride
+        self.conv0 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            stride=stride,
+            act="relu",
+            name=name + "_branch2a",
+        )
+        self.conv1 = ConvBNLayer(
+            in_channels=out_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            act=None,
+            name=name + "_branch2b",
+        )
+
+        if not shortcut:
+            self.short = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=out_channels,
+                kernel_size=1,
+                stride=1,
+                is_vd_mode=False if if_first else True,
+                name=name + "_branch1",
+            )
+
+        self.shortcut = shortcut
+
+    def forward(self, inputs):
+        y = self.conv0(inputs)
+        conv1 = self.conv1(y)
+
+        if self.shortcut:
+            short = inputs
+        else:
+            short = self.short(inputs)
+        y = paddle.add(x=short, y=conv1)
+        y = F.relu(y)
+        return y
+
+
+class ResNet(nn.Layer):
+    def __init__(self, in_channels=3, layers=50, **kwargs):
+        super(ResNet, self).__init__()
+
+        self.layers = layers
+        supported_layers = [18, 34, 50, 101, 152, 200]
+        assert (
+            layers in supported_layers
+        ), "supported layers are {} but input layer is {}".format(
+            supported_layers, layers
+        )
+
+        if layers == 18:
+            depth = [2, 2, 2, 2]
+        elif layers == 34 or layers == 50:
+            # depth = [3, 4, 6, 3]
+            depth = [3, 4, 6, 3, 3]
+        elif layers == 101:
+            depth = [3, 4, 23, 3]
+        elif layers == 152:
+            depth = [3, 8, 36, 3]
+        elif layers == 200:
+            depth = [3, 12, 48, 3]
+        num_channels = (
+            [64, 256, 512, 1024, 2048] if layers >= 50 else [64, 64, 128, 256]
+        )
+        num_filters = [64, 128, 256, 512, 512]
+
+        self.conv1_1 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=64,
+            kernel_size=7,
+            stride=2,
+            act="relu",
+            name="conv1_1",
+        )
+        self.pool2d_max = nn.MaxPool2D(kernel_size=3, stride=2, padding=1)
+
+        self.stages = []
+        self.out_channels = [3, 64]
+        # num_filters = [64, 128, 256, 512, 512]
+        if layers >= 50:
+            for block in range(len(depth)):
+                block_list = []
+                shortcut = False
+                for i in range(depth[block]):
+                    if layers in [101, 152] and block == 2:
+                        if i == 0:
+                            conv_name = "res" + str(block + 2) + "a"
+                        else:
+                            conv_name = "res" + str(block + 2) + "b" + str(i)
+                    else:
+                        conv_name = "res" + str(block + 2) + chr(97 + i)
+                    bottleneck_block = self.add_sublayer(
+                        "bb_%d_%d" % (block, i),
+                        BottleneckBlock(
+                            in_channels=(
+                                num_channels[block]
+                                if i == 0
+                                else num_filters[block] * 4
+                            ),
+                            out_channels=num_filters[block],
+                            stride=2 if i == 0 and block != 0 else 1,
+                            shortcut=shortcut,
+                            if_first=block == i == 0,
+                            name=conv_name,
+                        ),
+                    )
+                    shortcut = True
+                    block_list.append(bottleneck_block)
+                self.out_channels.append(num_filters[block] * 4)
+                self.stages.append(nn.Sequential(*block_list))
+        else:
+            for block in range(len(depth)):
+                block_list = []
+                shortcut = False
+                for i in range(depth[block]):
+                    conv_name = "res" + str(block + 2) + chr(97 + i)
+                    basic_block = self.add_sublayer(
+                        "bb_%d_%d" % (block, i),
+                        BasicBlock(
+                            in_channels=(
+                                num_channels[block] if i == 0 else num_filters[block]
+                            ),
+                            out_channels=num_filters[block],
+                            stride=2 if i == 0 and block != 0 else 1,
+                            shortcut=shortcut,
+                            if_first=block == i == 0,
+                            name=conv_name,
+                        ),
+                    )
+                    shortcut = True
+                    block_list.append(basic_block)
+                self.out_channels.append(num_filters[block])
+                self.stages.append(nn.Sequential(*block_list))
+
+    def forward(self, inputs):
+        out = [inputs]
+        y = self.conv1_1(inputs)
+        out.append(y)
+        y = self.pool2d_max(y)
+        for block in self.stages:
+            y = block(y)
+            out.append(y)
+        return out
diff --git a/ppocr/modeling/backbones/kie_unet_sdmgr.py b/ppocr/modeling/backbones/kie_unet_sdmgr.py
new file mode 100644
index 0000000..7b0c2b5
--- /dev/null
+++ b/ppocr/modeling/backbones/kie_unet_sdmgr.py
@@ -0,0 +1,199 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+import numpy as np
+import cv2
+
+__all__ = ["Kie_backbone"]
+
+
+class Encoder(nn.Layer):
+    def __init__(self, num_channels, num_filters):
+        super(Encoder, self).__init__()
+        self.conv1 = nn.Conv2D(
+            num_channels,
+            num_filters,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            bias_attr=False,
+        )
+        self.bn1 = nn.BatchNorm(num_filters, act="relu")
+
+        self.conv2 = nn.Conv2D(
+            num_filters,
+            num_filters,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            bias_attr=False,
+        )
+        self.bn2 = nn.BatchNorm(num_filters, act="relu")
+
+        self.pool = nn.MaxPool2D(kernel_size=3, stride=2, padding=1)
+
+    def forward(self, inputs):
+        x = self.conv1(inputs)
+        x = self.bn1(x)
+        x = self.conv2(x)
+        x = self.bn2(x)
+        x_pooled = self.pool(x)
+        return x, x_pooled
+
+
+class Decoder(nn.Layer):
+    def __init__(self, num_channels, num_filters):
+        super(Decoder, self).__init__()
+
+        self.conv1 = nn.Conv2D(
+            num_channels,
+            num_filters,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            bias_attr=False,
+        )
+        self.bn1 = nn.BatchNorm(num_filters, act="relu")
+
+        self.conv2 = nn.Conv2D(
+            num_filters,
+            num_filters,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            bias_attr=False,
+        )
+        self.bn2 = nn.BatchNorm(num_filters, act="relu")
+
+        self.conv0 = nn.Conv2D(
+            num_channels,
+            num_filters,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            bias_attr=False,
+        )
+        self.bn0 = nn.BatchNorm(num_filters, act="relu")
+
+    def forward(self, inputs_prev, inputs):
+        x = self.conv0(inputs)
+        x = self.bn0(x)
+        x = paddle.nn.functional.interpolate(
+            x, scale_factor=2, mode="bilinear", align_corners=False
+        )
+        x = paddle.concat([inputs_prev, x], axis=1)
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = self.conv2(x)
+        x = self.bn2(x)
+        return x
+
+
+class UNet(nn.Layer):
+    def __init__(self):
+        super(UNet, self).__init__()
+        self.down1 = Encoder(num_channels=3, num_filters=16)
+        self.down2 = Encoder(num_channels=16, num_filters=32)
+        self.down3 = Encoder(num_channels=32, num_filters=64)
+        self.down4 = Encoder(num_channels=64, num_filters=128)
+        self.down5 = Encoder(num_channels=128, num_filters=256)
+
+        self.up1 = Decoder(32, 16)
+        self.up2 = Decoder(64, 32)
+        self.up3 = Decoder(128, 64)
+        self.up4 = Decoder(256, 128)
+        self.out_channels = 16
+
+    def forward(self, inputs):
+        x1, _ = self.down1(inputs)
+        _, x2 = self.down2(x1)
+        _, x3 = self.down3(x2)
+        _, x4 = self.down4(x3)
+        _, x5 = self.down5(x4)
+
+        x = self.up4(x4, x5)
+        x = self.up3(x3, x)
+        x = self.up2(x2, x)
+        x = self.up1(x1, x)
+        return x
+
+
+class Kie_backbone(nn.Layer):
+    def __init__(self, in_channels, **kwargs):
+        super(Kie_backbone, self).__init__()
+        self.out_channels = 16
+        self.img_feat = UNet()
+        self.maxpool = nn.MaxPool2D(kernel_size=7)
+
+    def bbox2roi(self, bbox_list):
+        rois_list = []
+        rois_num = []
+        for img_id, bboxes in enumerate(bbox_list):
+            rois_num.append(bboxes.shape[0])
+            rois_list.append(bboxes)
+        rois = paddle.concat(rois_list, 0)
+        rois_num = paddle.to_tensor(rois_num, dtype="int32")
+        return rois, rois_num
+
+    def pre_process(self, img, relations, texts, gt_bboxes, tag, img_size):
+        img, relations, texts, gt_bboxes, tag, img_size = (
+            img.numpy(),
+            relations.numpy(),
+            texts.numpy(),
+            gt_bboxes.numpy(),
+            tag.numpy().tolist(),
+            img_size.numpy(),
+        )
+        temp_relations, temp_texts, temp_gt_bboxes = [], [], []
+        h, w = int(np.max(img_size[:, 0])), int(np.max(img_size[:, 1]))
+        img = paddle.to_tensor(img[:, :, :h, :w])
+        batch = len(tag)
+        for i in range(batch):
+            num, recoder_len = tag[i][0], tag[i][1]
+            temp_relations.append(
+                paddle.to_tensor(relations[i, :num, :num, :], dtype="float32")
+            )
+            temp_texts.append(
+                paddle.to_tensor(texts[i, :num, :recoder_len], dtype="float32")
+            )
+            temp_gt_bboxes.append(
+                paddle.to_tensor(gt_bboxes[i, :num, ...], dtype="float32")
+            )
+        return img, temp_relations, temp_texts, temp_gt_bboxes
+
+    def forward(self, inputs):
+        img = inputs[0]
+        relations, texts, gt_bboxes, tag, img_size = (
+            inputs[1],
+            inputs[2],
+            inputs[3],
+            inputs[5],
+            inputs[-1],
+        )
+        img, relations, texts, gt_bboxes = self.pre_process(
+            img, relations, texts, gt_bboxes, tag, img_size
+        )
+        x = self.img_feat(img)
+        boxes, rois_num = self.bbox2roi(gt_bboxes)
+        feats = paddle.vision.ops.roi_align(
+            x, boxes, spatial_scale=1.0, output_size=7, boxes_num=rois_num
+        )
+        feats = self.maxpool(feats).squeeze(-1).squeeze(-1)
+        return [relations, texts, feats]
diff --git a/ppocr/modeling/backbones/rec_densenet.py b/ppocr/modeling/backbones/rec_densenet.py
new file mode 100644
index 0000000..3e64584
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_densenet.py
@@ -0,0 +1,150 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/LBH1024/CAN/models/densenet.py
+
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+
+class Bottleneck(nn.Layer):
+    def __init__(self, nChannels, growthRate, use_dropout):
+        super(Bottleneck, self).__init__()
+        interChannels = 4 * growthRate
+        self.bn1 = nn.BatchNorm2D(interChannels)
+        self.conv1 = nn.Conv2D(
+            nChannels, interChannels, kernel_size=1, bias_attr=None
+        )  # Xavier initialization
+        self.bn2 = nn.BatchNorm2D(growthRate)
+        self.conv2 = nn.Conv2D(
+            interChannels, growthRate, kernel_size=3, padding=1, bias_attr=None
+        )  # Xavier initialization
+        self.use_dropout = use_dropout
+        self.dropout = nn.Dropout(p=0.2)
+
+    def forward(self, x):
+        out = F.relu(self.bn1(self.conv1(x)))
+        if self.use_dropout:
+            out = self.dropout(out)
+        out = F.relu(self.bn2(self.conv2(out)))
+        if self.use_dropout:
+            out = self.dropout(out)
+        out = paddle.concat([x, out], 1)
+        return out
+
+
+class SingleLayer(nn.Layer):
+    def __init__(self, nChannels, growthRate, use_dropout):
+        super(SingleLayer, self).__init__()
+        self.bn1 = nn.BatchNorm2D(nChannels)
+        self.conv1 = nn.Conv2D(
+            nChannels, growthRate, kernel_size=3, padding=1, bias_attr=False
+        )
+
+        self.use_dropout = use_dropout
+        self.dropout = nn.Dropout(p=0.2)
+
+    def forward(self, x):
+        out = self.conv1(F.relu(x))
+        if self.use_dropout:
+            out = self.dropout(out)
+
+        out = paddle.concat([x, out], 1)
+        return out
+
+
+class Transition(nn.Layer):
+    def __init__(self, nChannels, out_channels, use_dropout):
+        super(Transition, self).__init__()
+        self.bn1 = nn.BatchNorm2D(out_channels)
+        self.conv1 = nn.Conv2D(nChannels, out_channels, kernel_size=1, bias_attr=False)
+        self.use_dropout = use_dropout
+        self.dropout = nn.Dropout(p=0.2)
+
+    def forward(self, x):
+        out = F.relu(self.bn1(self.conv1(x)))
+        if self.use_dropout:
+            out = self.dropout(out)
+        out = F.avg_pool2d(out, 2, ceil_mode=True, exclusive=False)
+        return out
+
+
+class DenseNet(nn.Layer):
+    def __init__(
+        self, growthRate, reduction, bottleneck, use_dropout, input_channel, **kwargs
+    ):
+        super(DenseNet, self).__init__()
+
+        nDenseBlocks = 16
+        nChannels = 2 * growthRate
+
+        self.conv1 = nn.Conv2D(
+            input_channel,
+            nChannels,
+            kernel_size=7,
+            padding=3,
+            stride=2,
+            bias_attr=False,
+        )
+        self.dense1 = self._make_dense(
+            nChannels, growthRate, nDenseBlocks, bottleneck, use_dropout
+        )
+        nChannels += nDenseBlocks * growthRate
+        out_channels = int(math.floor(nChannels * reduction))
+        self.trans1 = Transition(nChannels, out_channels, use_dropout)
+
+        nChannels = out_channels
+        self.dense2 = self._make_dense(
+            nChannels, growthRate, nDenseBlocks, bottleneck, use_dropout
+        )
+        nChannels += nDenseBlocks * growthRate
+        out_channels = int(math.floor(nChannels * reduction))
+        self.trans2 = Transition(nChannels, out_channels, use_dropout)
+
+        nChannels = out_channels
+        self.dense3 = self._make_dense(
+            nChannels, growthRate, nDenseBlocks, bottleneck, use_dropout
+        )
+        self.out_channels = out_channels
+
+    def _make_dense(self, nChannels, growthRate, nDenseBlocks, bottleneck, use_dropout):
+        layers = []
+        for i in range(int(nDenseBlocks)):
+            if bottleneck:
+                layers.append(Bottleneck(nChannels, growthRate, use_dropout))
+            else:
+                layers.append(SingleLayer(nChannels, growthRate, use_dropout))
+            nChannels += growthRate
+        return nn.Sequential(*layers)
+
+    def forward(self, inputs):
+        x, x_m, y = inputs
+        out = self.conv1(x)
+        out = F.relu(out)
+        out = F.max_pool2d(out, 2, ceil_mode=True)
+        out = self.dense1(out)
+        out = self.trans1(out)
+        out = self.dense2(out)
+        out = self.trans2(out)
+        out = self.dense3(out)
+        return out, x_m, y
diff --git a/ppocr/modeling/backbones/rec_donut_swin.py b/ppocr/modeling/backbones/rec_donut_swin.py
new file mode 100644
index 0000000..accad00
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_donut_swin.py
@@ -0,0 +1,1296 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/huggingface/transformers/blob/main/src/transformers/models/donut/modeling_donut_swin.py
+
+"""
+
+import collections.abc
+from collections import OrderedDict
+import math
+import numpy as np
+from dataclasses import dataclass
+from typing import Optional, Tuple, Union
+
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+
+from paddle.nn.initializer import (
+    TruncatedNormal,
+    Constant,
+    Normal,
+    KaimingUniform,
+    XavierUniform,
+)
+
+zeros_ = Constant(value=0.0)
+ones_ = Constant(value=1.0)
+kaiming_normal_ = KaimingUniform(nonlinearity="relu")
+trunc_normal_ = TruncatedNormal(std=0.02)
+xavier_uniform_ = XavierUniform()
+
+# General docstring
+_CONFIG_FOR_DOC = "DonutSwinConfig"
+
+# Base docstring
+_CHECKPOINT_FOR_DOC = "https://huggingface.co/naver-clova-ix/donut-base"
+_EXPECTED_OUTPUT_SHAPE = [1, 49, 768]
+
+
+class DonutSwinConfig(object):
+    model_type = "donut-swin"
+
+    attribute_map = {
+        "num_attention_heads": "num_heads",
+        "num_hidden_layers": "num_layers",
+    }
+
+    def __init__(
+        self,
+        image_size=224,
+        patch_size=4,
+        num_channels=3,
+        embed_dim=96,
+        depths=[2, 2, 6, 2],
+        num_heads=[3, 6, 12, 24],
+        window_size=7,
+        mlp_ratio=4.0,
+        qkv_bias=True,
+        hidden_dropout_prob=0.0,
+        attention_probs_dropout_prob=0.0,
+        drop_path_rate=0.1,
+        hidden_act="gelu",
+        use_absolute_embeddings=False,
+        initializer_range=0.02,
+        layer_norm_eps=1e-5,
+        **kwargs,
+    ):
+        super().__init__()
+
+        self.image_size = image_size
+        self.patch_size = patch_size
+        self.num_channels = num_channels
+        self.embed_dim = embed_dim
+        self.depths = depths
+        self.num_layers = len(depths)
+        self.num_heads = num_heads
+        self.window_size = window_size
+        self.mlp_ratio = mlp_ratio
+        self.qkv_bias = qkv_bias
+        self.hidden_dropout_prob = hidden_dropout_prob
+        self.attention_probs_dropout_prob = attention_probs_dropout_prob
+        self.drop_path_rate = drop_path_rate
+        self.hidden_act = hidden_act
+        self.use_absolute_embeddings = use_absolute_embeddings
+        self.layer_norm_eps = layer_norm_eps
+        self.initializer_range = initializer_range
+        self.hidden_size = int(embed_dim * 2 ** (len(depths) - 1))
+
+        for key, value in kwargs.items():
+            try:
+                setattr(self, key, value)
+            except AttributeError as err:
+                print(f"Can't set {key} with value {value} for {self}")
+                raise err
+
+
+@dataclass
+# Copied from transformers.models.swin.modeling_swin.SwinEncoderOutput with Swin->DonutSwin
+class DonutSwinEncoderOutput(OrderedDict):
+    last_hidden_state = None
+    hidden_states = None
+    attentions = None
+    reshaped_hidden_states = None
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+    def __getitem__(self, k):
+        if isinstance(k, str):
+            inner_dict = dict(self.items())
+            return inner_dict[k]
+        else:
+            return self.to_tuple()[k]
+
+    def __setattr__(self, name, value):
+        if name in self.keys() and value is not None:
+            super().__setitem__(name, value)
+        super().__setattr__(name, value)
+
+    def __setitem__(self, key, value):
+        super().__setitem__(key, value)
+        super().__setattr__(key, value)
+
+    def to_tuple(self):
+        """
+        Convert self to a tuple containing all the attributes/keys that are not `None`.
+        """
+        return tuple(self[k] for k in self.keys())
+
+
+@dataclass
+# Copied from transformers.models.swin.modeling_swin.SwinModelOutput with Swin->DonutSwin
+class DonutSwinModelOutput(OrderedDict):
+    last_hidden_state = None
+    pooler_output = None
+    hidden_states = None
+    attentions = None
+    reshaped_hidden_states = None
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+    def __getitem__(self, k):
+        if isinstance(k, str):
+            inner_dict = dict(self.items())
+            return inner_dict[k]
+        else:
+            return self.to_tuple()[k]
+
+    def __setattr__(self, name, value):
+        if name in self.keys() and value is not None:
+            super().__setitem__(name, value)
+        super().__setattr__(name, value)
+
+    def __setitem__(self, key, value):
+        super().__setitem__(key, value)
+        super().__setattr__(key, value)
+
+    def to_tuple(self):
+        """
+        Convert self to a tuple containing all the attributes/keys that are not `None`.
+        """
+        return tuple(self[k] for k in self.keys())
+
+
+# Copied from transformers.models.swin.modeling_swin.window_partition
+def window_partition(input_feature, window_size):
+    """
+    Partitions the given input into windows.
+    """
+    batch_size, height, width, num_channels = input_feature.shape
+    input_feature = input_feature.reshape(
+        [
+            batch_size,
+            height // window_size,
+            window_size,
+            width // window_size,
+            window_size,
+            num_channels,
+        ]
+    )
+    windows = input_feature.transpose([0, 1, 3, 2, 4, 5]).reshape(
+        [-1, window_size, window_size, num_channels]
+    )
+    return windows
+
+
+# Copied from transformers.models.swin.modeling_swin.window_reverse
+def window_reverse(windows, window_size, height, width):
+    """
+    Merges windows to produce higher resolution features.
+    """
+    num_channels = windows.shape[-1]
+    windows = windows.reshape(
+        [
+            -1,
+            height // window_size,
+            width // window_size,
+            window_size,
+            window_size,
+            num_channels,
+        ]
+    )
+    windows = windows.transpose([0, 1, 3, 2, 4, 5]).reshape(
+        [-1, height, width, num_channels]
+    )
+    return windows
+
+
+# Copied from transformers.models.swin.modeling_swin.SwinEmbeddings with Swin->DonutSwin
+class DonutSwinEmbeddings(nn.Layer):
+    """
+    Construct the patch and position embeddings. Optionally, also the mask token.
+    """
+
+    def __init__(self, config, use_mask_token=False):
+        super().__init__()
+
+        self.patch_embeddings = DonutSwinPatchEmbeddings(config)
+        num_patches = self.patch_embeddings.num_patches
+        self.patch_grid = self.patch_embeddings.grid_size
+        if use_mask_token:
+            self.mask_token = paddle.create_parameter(
+                [1, 1, config.embed_dim], dtype="float32"
+            )
+            zeros_(self.mask_token)
+        else:
+            self.mask_token = None
+        if config.use_absolute_embeddings:
+            self.position_embeddings = paddle.create_parameter(
+                [1, num_patches + 1, config.embed_dim], dtype="float32"
+            )
+            zeros_(self.position_embedding)
+        else:
+            self.position_embeddings = None
+
+        self.norm = nn.LayerNorm(config.embed_dim)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+    def forward(self, pixel_values, bool_masked_pos=None):
+
+        embeddings, output_dimensions = self.patch_embeddings(pixel_values)
+        embeddings = self.norm(embeddings)
+
+        batch_size, seq_len, _ = embeddings.shape
+
+        if bool_masked_pos is not None:
+            mask_tokens = self.mask_token.expand(batch_size, seq_len, -1)
+            mask = bool_masked_pos.unsqueeze(-1).type_as(mask_tokens)
+            embeddings = embeddings * (1.0 - mask) + mask_tokens * mask
+
+        if self.position_embeddings is not None:
+            embeddings = embeddings + self.position_embeddings
+        embeddings = self.dropout(embeddings)
+        return embeddings, output_dimensions
+
+
+class MyConv2d(nn.Conv2D):
+    def __init__(
+        self,
+        in_channel,
+        out_channels,
+        kernel_size,
+        stride=1,
+        padding="SAME",
+        dilation=1,
+        groups=1,
+        bias_attr=False,
+        eps=1e-6,
+    ):
+        super().__init__(
+            in_channel,
+            out_channels,
+            kernel_size,
+            stride=stride,
+            padding=padding,
+            dilation=dilation,
+            groups=groups,
+            bias_attr=bias_attr,
+        )
+        self.weight = paddle.create_parameter(
+            [out_channels, in_channel, kernel_size[0], kernel_size[1]], dtype="float32"
+        )
+        self.bias = paddle.create_parameter([out_channels], dtype="float32")
+        ones_(self.weight)
+        zeros_(self.bias)
+
+    def forward(self, x):
+        x = F.conv2d(
+            x,
+            self.weight,
+            self.bias,
+            self._stride,
+            self._padding,
+            self._dilation,
+            self._groups,
+        )
+        return x
+
+
+# Copied from transformers.models.swin.modeling_swin.SwinPatchEmbeddings
+class DonutSwinPatchEmbeddings(nn.Layer):
+    """
+    This class turns `pixel_values` of shape `(batch_size, num_channels, height, width)` into the initial
+    `hidden_states` (patch embeddings) of shape `(batch_size, seq_length, hidden_size)` to be consumed by a
+    Transformer.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        image_size, patch_size = config.image_size, config.patch_size
+        num_channels, hidden_size = config.num_channels, config.embed_dim
+        image_size = (
+            image_size
+            if isinstance(image_size, collections.abc.Iterable)
+            else (image_size, image_size)
+        )
+        patch_size = (
+            patch_size
+            if isinstance(patch_size, collections.abc.Iterable)
+            else (patch_size, patch_size)
+        )
+        num_patches = (image_size[1] // patch_size[1]) * (
+            image_size[0] // patch_size[0]
+        )
+        self.image_size = image_size
+        self.patch_size = patch_size
+        self.num_channels = num_channels
+        self.num_patches = num_patches
+        self.is_export = config.is_export
+        self.grid_size = (
+            image_size[0] // patch_size[0],
+            image_size[1] // patch_size[1],
+        )
+        self.projection = nn.Conv2D(
+            num_channels, hidden_size, kernel_size=patch_size, stride=patch_size
+        )
+
+    def maybe_pad(self, pixel_values, height, width):
+        if width % self.patch_size[1] != 0:
+            pad_values = (0, self.patch_size[1] - width % self.patch_size[1])
+            if self.is_export:
+                pad_values = paddle.to_tensor(pad_values, dtype="int32")
+            pixel_values = nn.functional.pad(pixel_values, pad_values)
+        if height % self.patch_size[0] != 0:
+            pad_values = (0, 0, 0, self.patch_size[0] - height % self.patch_size[0])
+            if self.is_export:
+                pad_values = paddle.to_tensor(pad_values, dtype="int32")
+            pixel_values = nn.functional.pad(pixel_values, pad_values)
+        return pixel_values
+
+    def forward(self, pixel_values) -> Tuple[paddle.Tensor, Tuple[int]]:
+        _, num_channels, height, width = pixel_values.shape
+        if num_channels != self.num_channels:
+            raise ValueError(
+                "Make sure that the channel dimension of the pixel values match with the one set in the configuration."
+            )
+        pixel_values = self.maybe_pad(pixel_values, height, width)
+        embeddings = self.projection(pixel_values)
+
+        _, _, height, width = embeddings.shape
+        output_dimensions = (height, width)
+        embeddings = embeddings.flatten(2).transpose([0, 2, 1])
+
+        return embeddings, output_dimensions
+
+
+# Copied from transformers.models.swin.modeling_swin.SwinPatchMerging
+class DonutSwinPatchMerging(nn.Layer):
+    """
+    Patch Merging Layer.
+
+    Args:
+        input_resolution (`Tuple[int]`):
+            Resolution of input feature.
+        dim (`int`):
+            Number of input channels.
+        norm_layer (`nn.Layer`, *optional*, defaults to `nn.LayerNorm`):
+            Normalization layer class.
+    """
+
+    def __init__(
+        self,
+        input_resolution: Tuple[int],
+        dim: int,
+        norm_layer: nn.Layer = nn.LayerNorm,
+        is_export=False,
+    ):
+        super().__init__()
+        self.input_resolution = input_resolution
+        self.dim = dim
+        self.reduction = nn.Linear(4 * dim, 2 * dim, bias_attr=False)
+        self.norm = norm_layer(4 * dim)
+        self.is_export = is_export
+
+    def maybe_pad(self, input_feature, height, width):
+        should_pad = (height % 2 == 1) or (width % 2 == 1)
+        if should_pad:
+            pad_values = (0, 0, 0, width % 2, 0, height % 2)
+            if self.is_export:
+                pad_values = paddle.to_tensor(pad_values, dtype="int32")
+            input_feature = nn.functional.pad(input_feature, pad_values)
+
+        return input_feature
+
+    def forward(
+        self, input_feature: paddle.Tensor, input_dimensions: Tuple[int, int]
+    ) -> paddle.Tensor:
+        height, width = input_dimensions
+        batch_size, dim, num_channels = input_feature.shape
+
+        input_feature = input_feature.reshape([batch_size, height, width, num_channels])
+
+        input_feature = self.maybe_pad(input_feature, height, width)
+        input_feature_0 = input_feature[:, 0::2, 0::2, :]
+        input_feature_1 = input_feature[:, 1::2, 0::2, :]
+        input_feature_2 = input_feature[:, 0::2, 1::2, :]
+        input_feature_3 = input_feature[:, 1::2, 1::2, :]
+        input_feature = paddle.concat(
+            [input_feature_0, input_feature_1, input_feature_2, input_feature_3], -1
+        )
+        input_feature = input_feature.reshape(
+            [batch_size, -1, 4 * num_channels]
+        )  # batch_size height/2*width/2 4*C
+
+        input_feature = self.norm(input_feature)
+        input_feature = self.reduction(input_feature)
+
+        return input_feature
+
+
+# Copied from transformers.models.beit.modeling_beit.drop_path
+def drop_path(
+    input: paddle.Tensor, drop_prob: float = 0.0, training: bool = False
+) -> paddle.Tensor:
+    if drop_prob == 0.0 or not training:
+        return input
+    keep_prob = 1 - drop_prob
+    shape = (input.shape[0],) + (1,) * (
+        input.ndim - 1
+    )  # work with diff dim tensors, not just 2D ConvNets
+    random_tensor = keep_prob + paddle.rand(
+        shape,
+        dtype=input.dtype,
+    )
+    random_tensor.floor_()  # binarize
+    output = input / keep_prob * random_tensor
+    return output
+
+
+# Copied from transformers.models.swin.modeling_swin.SwinDropPath
+class DonutSwinDropPath(nn.Layer):
+    """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks)."""
+
+    def __init__(self, drop_prob: Optional[float] = None) -> None:
+        super().__init__()
+        self.drop_prob = drop_prob
+
+    def forward(self, hidden_states: paddle.Tensor) -> paddle.Tensor:
+        return drop_path(hidden_states, self.drop_prob, self.training)
+
+    def extra_repr(self) -> str:
+        return "p={}".format(self.drop_prob)
+
+
+class DonutSwinSelfAttention(nn.Layer):
+    def __init__(self, config, dim, num_heads, window_size):
+        super().__init__()
+        if dim % num_heads != 0:
+            raise ValueError(
+                f"The hidden size ({dim}) is not a multiple of the number of attention heads ({num_heads})"
+            )
+
+        self.num_attention_heads = num_heads
+        self.attention_head_size = int(dim / num_heads)
+        self.all_head_size = self.num_attention_heads * self.attention_head_size
+        self.window_size = (
+            window_size
+            if isinstance(window_size, collections.abc.Iterable)
+            else (window_size, window_size)
+        )
+        self.relative_position_bias_table = paddle.create_parameter(
+            [(2 * self.window_size[0] - 1) * (2 * self.window_size[1] - 1), num_heads],
+            dtype="float32",
+        )
+        zeros_(self.relative_position_bias_table)
+
+        # get pair-wise relative position index for each token inside the window
+        coords_h = paddle.arange(self.window_size[0])
+        coords_w = paddle.arange(self.window_size[1])
+        coords = paddle.stack(paddle.meshgrid(coords_h, coords_w, indexing="ij"))
+        coords_flatten = paddle.flatten(coords, 1)
+        relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]
+        relative_coords = relative_coords.transpose([1, 2, 0])
+        relative_coords[:, :, 0] += self.window_size[0] - 1
+        relative_coords[:, :, 1] += self.window_size[1] - 1
+        relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1
+        relative_position_index = relative_coords.sum(-1)
+        self.register_buffer("relative_position_index", relative_position_index)
+
+        self.query = nn.Linear(
+            self.all_head_size, self.all_head_size, bias_attr=config.qkv_bias
+        )
+        self.key = nn.Linear(
+            self.all_head_size, self.all_head_size, bias_attr=config.qkv_bias
+        )
+        self.value = nn.Linear(
+            self.all_head_size, self.all_head_size, bias_attr=config.qkv_bias
+        )
+
+        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
+
+    def transpose_for_scores(self, x):
+        new_x_shape = x.shape[:-1] + [
+            self.num_attention_heads,
+            self.attention_head_size,
+        ]
+        x = x.reshape(new_x_shape)
+        return x.transpose([0, 2, 1, 3])
+
+    def forward(
+        self,
+        hidden_states: paddle.Tensor,
+        attention_mask=None,
+        head_mask=None,
+        output_attentions=False,
+    ) -> Tuple[paddle.Tensor]:
+        batch_size, dim, num_channels = hidden_states.shape
+        mixed_query_layer = self.query(hidden_states)
+        key_layer = self.transpose_for_scores(self.key(hidden_states))
+        value_layer = self.transpose_for_scores(self.value(hidden_states))
+        query_layer = self.transpose_for_scores(mixed_query_layer)
+
+        # Take the dot product between "query" and "key" to get the raw attention scores.
+        attention_scores = paddle.matmul(query_layer, key_layer.transpose([0, 1, 3, 2]))
+
+        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
+
+        relative_position_bias = self.relative_position_bias_table[
+            self.relative_position_index.reshape([-1])
+        ]
+        relative_position_bias = relative_position_bias.reshape(
+            [
+                self.window_size[0] * self.window_size[1],
+                self.window_size[0] * self.window_size[1],
+                -1,
+            ]
+        )
+
+        relative_position_bias = relative_position_bias.transpose([2, 0, 1])
+        attention_scores = attention_scores + relative_position_bias.unsqueeze(0)
+
+        if attention_mask is not None:
+            # Apply the attention mask is (precomputed for all layers in DonutSwinModel forward() function)
+            mask_shape = attention_mask.shape[0]
+            attention_scores = attention_scores.reshape(
+                [
+                    batch_size // mask_shape,
+                    mask_shape,
+                    self.num_attention_heads,
+                    dim,
+                    dim,
+                ]
+            )
+            attention_scores = attention_scores + attention_mask.unsqueeze(1).unsqueeze(
+                0
+            )
+            attention_scores = attention_scores.reshape(
+                [-1, self.num_attention_heads, dim, dim]
+            )
+
+        # Normalize the attention scores to probabilities.
+        attention_probs = nn.functional.softmax(attention_scores, axis=-1)
+
+        # This is actually dropping out entire tokens to attend to, which might
+        # seem a bit unusual, but is taken from the original Transformer paper.
+        attention_probs = self.dropout(attention_probs)
+
+        # Mask heads if we want to
+        if head_mask is not None:
+            attention_probs = attention_probs * head_mask
+
+        context_layer = paddle.matmul(attention_probs, value_layer)
+        context_layer = context_layer.transpose([0, 2, 1, 3])
+        new_context_layer_shape = tuple(context_layer.shape[:-2]) + (
+            self.all_head_size,
+        )
+        context_layer = context_layer.reshape(new_context_layer_shape)
+        outputs = (
+            (context_layer, attention_probs) if output_attentions else (context_layer,)
+        )
+        return outputs
+
+
+# Copied from transformers.models.swin.modeling_swin.SwinSelfOutput
+class DonutSwinSelfOutput(nn.Layer):
+    def __init__(self, config, dim):
+        super().__init__()
+        self.dense = nn.Linear(dim, dim)
+        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)
+
+    def forward(
+        self, hidden_states: paddle.Tensor, input_tensor: paddle.Tensor
+    ) -> paddle.Tensor:
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+
+        return hidden_states
+
+
+# Copied from transformers.models.swin.modeling_swin.SwinAttention with Swin->DonutSwin
+class DonutSwinAttention(nn.Layer):
+    def __init__(self, config, dim, num_heads, window_size):
+        super().__init__()
+        self.self = DonutSwinSelfAttention(config, dim, num_heads, window_size)
+        self.output = DonutSwinSelfOutput(config, dim)
+        self.pruned_heads = set()
+
+    def forward(
+        self,
+        hidden_states: paddle.Tensor,
+        attention_mask=None,
+        head_mask=None,
+        output_attentions=False,
+    ) -> Tuple[paddle.Tensor]:
+        self_outputs = self.self(
+            hidden_states, attention_mask, head_mask, output_attentions
+        )
+        attention_output = self.output(self_outputs[0], hidden_states)
+        outputs = (attention_output,) + self_outputs[
+            1:
+        ]  # add attentions if we output them
+        return outputs
+
+
+# Copied from transformers.models.swin.modeling_swin.SwinIntermediate
+class DonutSwinIntermediate(nn.Layer):
+    def __init__(self, config, dim):
+        super().__init__()
+        self.dense = nn.Linear(dim, int(config.mlp_ratio * dim))
+        self.intermediate_act_fn = F.gelu
+
+    def forward(self, hidden_states: paddle.Tensor) -> paddle.Tensor:
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.intermediate_act_fn(hidden_states)
+        return hidden_states
+
+
+# Copied from transformers.models.swin.modeling_swin.SwinOutput
+class DonutSwinOutput(nn.Layer):
+    def __init__(self, config, dim):
+        super().__init__()
+        self.dense = nn.Linear(int(config.mlp_ratio * dim), dim)
+        self.dropout = nn.Dropout(config.hidden_dropout_prob)
+
+    def forward(self, hidden_states: paddle.Tensor) -> paddle.Tensor:
+        hidden_states = self.dense(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        return hidden_states
+
+
+# Copied from transformers.models.swin.modeling_swin.SwinLayer with Swin->DonutSwin
+class DonutSwinLayer(nn.Layer):
+    def __init__(self, config, dim, input_resolution, num_heads, shift_size=0):
+        super().__init__()
+        self.chunk_size_feed_forward = config.chunk_size_feed_forward
+        self.shift_size = shift_size
+        self.window_size = config.window_size
+        self.input_resolution = input_resolution
+        self.layernorm_before = nn.LayerNorm(dim, epsilon=config.layer_norm_eps)
+        self.attention = DonutSwinAttention(
+            config, dim, num_heads, window_size=self.window_size
+        )
+        self.drop_path = (
+            DonutSwinDropPath(config.drop_path_rate)
+            if config.drop_path_rate > 0.0
+            else nn.Identity()
+        )
+        self.layernorm_after = nn.LayerNorm(dim, epsilon=config.layer_norm_eps)
+        self.intermediate = DonutSwinIntermediate(config, dim)
+        self.output = DonutSwinOutput(config, dim)
+        self.is_export = config.is_export
+
+    def set_shift_and_window_size(self, input_resolution):
+        if min(input_resolution) <= self.window_size:
+            # if window size is larger than input resolution, we don't partition windows
+            self.shift_size = 0
+            self.window_size = min(input_resolution)
+
+    def get_attn_mask_export(self, height, width, dtype):
+
+        attn_mask = None
+        height_slices = (
+            slice(0, -self.window_size),
+            slice(-self.window_size, -self.shift_size),
+            slice(-self.shift_size, None),
+        )
+        width_slices = (
+            slice(0, -self.window_size),
+            slice(-self.window_size, -self.shift_size),
+            slice(-self.shift_size, None),
+        )
+        img_mask = paddle.zeros((1, height, width, 1), dtype=dtype)
+        count = 0
+        for height_slice in height_slices:
+            for width_slice in width_slices:
+                if self.shift_size > 0:
+                    img_mask[:, height_slice, width_slice, :] = count
+                    count += 1
+        if paddle.to_tensor(self.shift_size > 0).cast(paddle.bool):
+            # calculate attention mask for SW-MSA
+            mask_windows = window_partition(img_mask, self.window_size)
+            mask_windows = mask_windows.reshape(
+                [-1, self.window_size * self.window_size]
+            )
+            attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
+            attn_mask = attn_mask.masked_fill(
+                attn_mask != 0, float(-100.0)
+            ).masked_fill(attn_mask == 0, float(0.0))
+
+        return attn_mask
+
+    def get_attn_mask(self, height, width, dtype):
+        if self.shift_size > 0:
+            # calculate attention mask for SW-MSA
+            img_mask = paddle.zeros((1, height, width, 1), dtype=dtype)
+            height_slices = (
+                slice(0, -self.window_size),
+                slice(-self.window_size, -self.shift_size),
+                slice(-self.shift_size, None),
+            )
+            width_slices = (
+                slice(0, -self.window_size),
+                slice(-self.window_size, -self.shift_size),
+                slice(-self.shift_size, None),
+            )
+
+            count = 0
+            for height_slice in height_slices:
+                for width_slice in width_slices:
+                    img_mask[:, height_slice, width_slice, :] = count
+                    count += 1
+
+            mask_windows = window_partition(img_mask, self.window_size)
+            mask_windows = mask_windows.reshape(
+                [-1, self.window_size * self.window_size]
+            )
+            attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
+            attn_mask = attn_mask.masked_fill(
+                attn_mask != 0, float(-100.0)
+            ).masked_fill(attn_mask == 0, float(0.0))
+        else:
+            attn_mask = None
+        return attn_mask
+
+    def maybe_pad(self, hidden_states, height, width):
+        pad_right = (self.window_size - width % self.window_size) % self.window_size
+        pad_bottom = (self.window_size - height % self.window_size) % self.window_size
+        pad_values = (0, 0, 0, pad_bottom, 0, pad_right, 0, 0)
+        hidden_states = nn.functional.pad(hidden_states, pad_values)
+        return hidden_states, pad_values
+
+    def forward(
+        self,
+        hidden_states: paddle.Tensor,
+        input_dimensions: Tuple[int, int],
+        head_mask=None,
+        output_attentions=False,
+        always_partition=False,
+    ) -> Tuple[paddle.Tensor, paddle.Tensor]:
+        if not always_partition:
+            self.set_shift_and_window_size(input_dimensions)
+        else:
+            pass
+        height, width = input_dimensions
+        batch_size, _, channels = hidden_states.shape
+        shortcut = hidden_states
+
+        hidden_states = self.layernorm_before(hidden_states)
+
+        hidden_states = hidden_states.reshape([batch_size, height, width, channels])
+
+        # pad hidden_states to multiples of window size
+        hidden_states, pad_values = self.maybe_pad(hidden_states, height, width)
+
+        _, height_pad, width_pad, _ = hidden_states.shape
+
+        # cyclic shift
+        if self.shift_size > 0:
+            shift_value = (-self.shift_size, -self.shift_size)
+            if self.is_export:
+                shift_value = paddle.to_tensor(shift_value, dtype="int32")
+            shifted_hidden_states = paddle.roll(
+                hidden_states, shifts=shift_value, axis=(1, 2)
+            )
+        else:
+            shifted_hidden_states = hidden_states
+
+        # partition windows
+        hidden_states_windows = window_partition(
+            shifted_hidden_states, self.window_size
+        )
+        hidden_states_windows = hidden_states_windows.reshape(
+            [-1, self.window_size * self.window_size, channels]
+        )
+        attn_mask = self.get_attn_mask(height_pad, width_pad, dtype=hidden_states.dtype)
+
+        attention_outputs = self.attention(
+            hidden_states_windows,
+            attn_mask,
+            head_mask,
+            output_attentions=output_attentions,
+        )
+        attention_output = attention_outputs[0]
+
+        attention_windows = attention_output.reshape(
+            [-1, self.window_size, self.window_size, channels]
+        )
+        shifted_windows = window_reverse(
+            attention_windows, self.window_size, height_pad, width_pad
+        )
+        # reverse cyclic shift
+        if self.shift_size > 0:
+            shift_value = (self.shift_size, self.shift_size)
+            if self.is_export:
+                shift_value = paddle.to_tensor(shift_value, dtype="int32")
+            attention_windows = paddle.roll(
+                shifted_windows, shifts=shift_value, axis=(1, 2)
+            )
+        else:
+            attention_windows = shifted_windows
+
+        was_padded = pad_values[3] > 0 or pad_values[5] > 0
+        if was_padded:
+            attention_windows = attention_windows[:, :height, :width, :].contiguous()
+
+        attention_windows = attention_windows.reshape(
+            [batch_size, height * width, channels]
+        )
+        hidden_states = shortcut + self.drop_path(attention_windows)
+        layer_output = self.layernorm_after(hidden_states)
+        layer_output = self.intermediate(layer_output)
+        layer_output = hidden_states + self.output(layer_output)
+        layer_outputs = (
+            (layer_output, attention_outputs[1])
+            if output_attentions
+            else (layer_output,)
+        )
+        return layer_outputs
+
+
+# Copied from transformers.models.swin.modeling_swin.SwinStage with Swin->DonutSwin
+class DonutSwinStage(nn.Layer):
+    def __init__(
+        self, config, dim, input_resolution, depth, num_heads, drop_path, downsample
+    ):
+        super().__init__()
+        self.config = config
+        self.dim = dim
+        self.blocks = nn.LayerList(
+            [
+                DonutSwinLayer(
+                    config=config,
+                    dim=dim,
+                    input_resolution=input_resolution,
+                    num_heads=num_heads,
+                    shift_size=0 if (i % 2 == 0) else config.window_size // 2,
+                )
+                for i in range(depth)
+            ]
+        )
+        self.is_export = config.is_export
+
+        # patch merging layer
+        if downsample is not None:
+            self.downsample = downsample(
+                input_resolution,
+                dim=dim,
+                norm_layer=nn.LayerNorm,
+                is_export=self.is_export,
+            )
+        else:
+            self.downsample = None
+
+        self.pointing = False
+
+    def forward(
+        self,
+        hidden_states: paddle.Tensor,
+        input_dimensions: Tuple[int, int],
+        head_mask=None,
+        output_attentions=False,
+        always_partition=False,
+    ) -> Tuple[paddle.Tensor]:
+        height, width = input_dimensions
+
+        for i, layer_module in enumerate(self.blocks):
+            layer_head_mask = head_mask[i] if head_mask is not None else None
+
+            layer_outputs = layer_module(
+                hidden_states,
+                input_dimensions,
+                layer_head_mask,
+                output_attentions,
+                always_partition,
+            )
+
+            hidden_states = layer_outputs[0]
+
+        hidden_states_before_downsampling = hidden_states
+        if self.downsample is not None:
+            height_downsampled, width_downsampled = (height + 1) // 2, (width + 1) // 2
+            output_dimensions = (height, width, height_downsampled, width_downsampled)
+            hidden_states = self.downsample(
+                hidden_states_before_downsampling, input_dimensions
+            )
+        else:
+            output_dimensions = (height, width, height, width)
+
+        stage_outputs = (
+            hidden_states,
+            hidden_states_before_downsampling,
+            output_dimensions,
+        )
+
+        if output_attentions:
+            stage_outputs += layer_outputs[1:]
+        return stage_outputs
+
+
+# Copied from transformers.models.swin.modeling_swin.SwinEncoder with Swin->DonutSwin
+class DonutSwinEncoder(nn.Layer):
+    def __init__(self, config, grid_size):
+        super().__init__()
+        self.num_layers = len(config.depths)
+        self.config = config
+        dpr = [
+            x.item()
+            for x in paddle.linspace(0, config.drop_path_rate, sum(config.depths))
+        ]
+        self.layers = nn.LayerList(
+            [
+                DonutSwinStage(
+                    config=config,
+                    dim=int(config.embed_dim * 2**i_layer),
+                    input_resolution=(
+                        grid_size[0] // (2**i_layer),
+                        grid_size[1] // (2**i_layer),
+                    ),
+                    depth=config.depths[i_layer],
+                    num_heads=config.num_heads[i_layer],
+                    drop_path=dpr[
+                        sum(config.depths[:i_layer]) : sum(config.depths[: i_layer + 1])
+                    ],
+                    downsample=(
+                        DonutSwinPatchMerging
+                        if (i_layer < self.num_layers - 1)
+                        else None
+                    ),
+                )
+                for i_layer in range(self.num_layers)
+            ]
+        )
+
+        self.gradient_checkpointing = False
+
+    def forward(
+        self,
+        hidden_states: paddle.Tensor,
+        input_dimensions: Tuple[int, int],
+        head_mask=None,
+        output_attentions=False,
+        output_hidden_states=False,
+        output_hidden_states_before_downsampling=False,
+        always_partition=False,
+        return_dict=True,
+    ):
+        all_hidden_states = () if output_hidden_states else None
+        all_reshaped_hidden_states = () if output_hidden_states else None
+        all_self_attentions = () if output_attentions else None
+
+        if output_hidden_states:
+            batch_size, _, hidden_size = hidden_states.shape
+            reshaped_hidden_state = hidden_states.view(
+                batch_size, *input_dimensions, hidden_size
+            )
+            reshaped_hidden_state = reshaped_hidden_state.permute(0, 3, 1, 2)
+            all_hidden_states += (hidden_states,)
+            all_reshaped_hidden_states += (reshaped_hidden_state,)
+
+        for i, layer_module in enumerate(self.layers):
+            layer_head_mask = head_mask[i] if head_mask is not None else None
+
+            if self.gradient_checkpointing and self.training:
+                layer_outputs = self._gradient_checkpointing_func(
+                    layer_module.__call__,
+                    hidden_states,
+                    input_dimensions,
+                    layer_head_mask,
+                    output_attentions,
+                    always_partition,
+                )
+            else:
+                layer_outputs = layer_module(
+                    hidden_states,
+                    input_dimensions,
+                    layer_head_mask,
+                    output_attentions,
+                    always_partition,
+                )
+
+            hidden_states = layer_outputs[0]
+
+            hidden_states_before_downsampling = layer_outputs[1]
+            output_dimensions = layer_outputs[2]
+
+            input_dimensions = (output_dimensions[-2], output_dimensions[-1])
+
+            if output_hidden_states and output_hidden_states_before_downsampling:
+                batch_size, _, hidden_size = hidden_states_before_downsampling.shape
+                reshaped_hidden_state = hidden_states_before_downsampling.reshape(
+                    [
+                        batch_size,
+                        *(output_dimensions[0], output_dimensions[1]),
+                        hidden_size,
+                    ]
+                )
+                reshaped_hidden_state = reshaped_hidden_state.transpose([0, 3, 1, 2])
+                all_hidden_states += (hidden_states_before_downsampling,)
+                all_reshaped_hidden_states += (reshaped_hidden_state,)
+            elif output_hidden_states and not output_hidden_states_before_downsampling:
+                batch_size, _, hidden_size = hidden_states.shape
+                reshaped_hidden_state = hidden_states.reshape(
+                    [batch_size, *input_dimensions, hidden_size]
+                )
+                reshaped_hidden_state = reshaped_hidden_state.transpose([0, 3, 1, 2])
+                all_hidden_states += (hidden_states,)
+                all_reshaped_hidden_states += (reshaped_hidden_state,)
+
+            if output_attentions:
+                all_self_attentions += layer_outputs[3:]
+
+        if not return_dict:
+            return tuple(
+                v
+                for v in [hidden_states, all_hidden_states, all_self_attentions]
+                if v is not None
+            )
+
+        return DonutSwinEncoderOutput(
+            last_hidden_state=hidden_states,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+            reshaped_hidden_states=all_reshaped_hidden_states,
+        )
+
+
+class DonutSwinPreTrainedModel(nn.Layer):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = DonutSwinConfig
+    base_model_prefix = "swin"
+    main_input_name = "pixel_values"
+    supports_gradient_checkpointing = True
+
+    def _init_weights(self, module):
+        """Initialize the weights"""
+        if isinstance(module, (nn.Linear, nn.Conv2D)):
+            normal_ = Normal(mean=0.0, std=self.config.initializer_range)
+            normal_(module.weight)
+            if module.bias is not None:
+                zeros_(module.bias)
+        elif isinstance(module, nn.LayerNorm):
+            zeros_(module.bias)
+            ones_(module.weight)
+
+    def _initialize_weights(self, module):
+        """
+        Initialize the weights if they are not already initialized.
+        """
+        if getattr(module, "_is_hf_initialized", False):
+            return
+        self._init_weights(module)
+
+    def post_init(self):
+        self.apply(self._initialize_weights)
+
+    def get_head_mask(self, head_mask, num_hidden_layers, is_attention_chunked=False):
+        if head_mask is not None:
+            head_mask = self._convert_head_mask_to_5d(head_mask, num_hidden_layers)
+            if is_attention_chunked is True:
+                head_mask = head_mask.unsqueeze(-1)
+        else:
+            head_mask = [None] * num_hidden_layers
+
+        return head_mask
+
+
+class DonutSwinModel(DonutSwinPreTrainedModel):
+    def __init__(
+        self,
+        in_channels=3,
+        hidden_size=1024,
+        num_layers=4,
+        num_heads=[4, 8, 16, 32],
+        add_pooling_layer=True,
+        use_mask_token=False,
+        is_export=False,
+    ):
+        super().__init__()
+        donut_swin_config = {
+            "return_dict": True,
+            "output_hidden_states": False,
+            "output_attentions": False,
+            "use_bfloat16": False,
+            "tf_legacy_loss": False,
+            "pruned_heads": {},
+            "tie_word_embeddings": True,
+            "chunk_size_feed_forward": 0,
+            "is_encoder_decoder": False,
+            "is_decoder": False,
+            "cross_attention_hidden_size": None,
+            "add_cross_attention": False,
+            "tie_encoder_decoder": False,
+            "max_length": 20,
+            "min_length": 0,
+            "do_sample": False,
+            "early_stopping": False,
+            "num_beams": 1,
+            "num_beam_groups": 1,
+            "diversity_penalty": 0.0,
+            "temperature": 1.0,
+            "top_k": 50,
+            "top_p": 1.0,
+            "typical_p": 1.0,
+            "repetition_penalty": 1.0,
+            "length_penalty": 1.0,
+            "no_repeat_ngram_size": 0,
+            "encoder_no_repeat_ngram_size": 0,
+            "bad_words_ids": None,
+            "num_return_sequences": 1,
+            "output_scores": False,
+            "return_dict_in_generate": False,
+            "forced_bos_token_id": None,
+            "forced_eos_token_id": None,
+            "remove_invalid_values": False,
+            "exponential_decay_length_penalty": None,
+            "suppress_tokens": None,
+            "begin_suppress_tokens": None,
+            "architectures": None,
+            "finetuning_task": None,
+            "id2label": {0: "LABEL_0", 1: "LABEL_1"},
+            "label2id": {"LABEL_0": 0, "LABEL_1": 1},
+            "tokenizer_class": None,
+            "prefix": None,
+            "bos_token_id": None,
+            "pad_token_id": None,
+            "eos_token_id": None,
+            "sep_token_id": None,
+            "decoder_start_token_id": None,
+            "task_specific_params": None,
+            "problem_type": None,
+            "_name_or_path": "",
+            "_commit_hash": None,
+            "_attn_implementation_internal": None,
+            "transformers_version": None,
+            "hidden_size": hidden_size,
+            "num_layers": num_layers,
+            "path_norm": True,
+            "use_2d_embeddings": False,
+            "image_size": [420, 420],
+            "patch_size": 4,
+            "num_channels": in_channels,
+            "embed_dim": 128,
+            "depths": [2, 2, 14, 2],
+            "num_heads": num_heads,
+            "window_size": 5,
+            "mlp_ratio": 4.0,
+            "qkv_bias": True,
+            "hidden_dropout_prob": 0.0,
+            "attention_probs_dropout_prob": 0.0,
+            "drop_path_rate": 0.1,
+            "hidden_act": "gelu",
+            "use_absolute_embeddings": False,
+            "layer_norm_eps": 1e-05,
+            "initializer_range": 0.02,
+            "is_export": is_export,
+        }
+
+        config = DonutSwinConfig(**donut_swin_config)
+        self.config = config
+        self.num_layers = len(config.depths)
+        self.num_features = int(config.embed_dim * 2 ** (self.num_layers - 1))
+
+        self.embeddings = DonutSwinEmbeddings(config, use_mask_token=use_mask_token)
+        self.encoder = DonutSwinEncoder(config, self.embeddings.patch_grid)
+
+        self.pooler = nn.AdaptiveAvgPool1D(1) if add_pooling_layer else None
+        self.out_channels = hidden_size
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embeddings.patch_embeddings
+
+    def forward(
+        self,
+        input_data=None,
+        bool_masked_pos=None,
+        head_mask=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ) -> Union[Tuple, DonutSwinModelOutput]:
+        r"""
+        bool_masked_pos (`paddle.BoolTensor` of shape `(batch_size, num_patches)`):
+            Boolean masked positions. Indicates which patches are masked (1) and which aren't (0).
+        """
+        if self.training:
+            pixel_values, label, attention_mask = input_data
+        else:
+            if isinstance(input_data, list):
+                pixel_values = input_data[0]
+            else:
+                pixel_values = input_data
+        output_attentions = (
+            output_attentions
+            if output_attentions is not None
+            else self.config.output_attentions
+        )
+        output_hidden_states = (
+            output_hidden_states
+            if output_hidden_states is not None
+            else self.config.output_hidden_states
+        )
+        return_dict = (
+            return_dict if return_dict is not None else self.config.return_dict
+        )
+
+        if pixel_values is None:
+            raise ValueError("You have to specify pixel_values")
+        num_channels = pixel_values.shape[1]
+        if num_channels == 1:
+            pixel_values = paddle.repeat_interleave(pixel_values, repeats=3, axis=1)
+
+        head_mask = self.get_head_mask(head_mask, len(self.config.depths))
+
+        embedding_output, input_dimensions = self.embeddings(
+            pixel_values, bool_masked_pos=bool_masked_pos
+        )
+
+        encoder_outputs = self.encoder(
+            embedding_output,
+            input_dimensions,
+            head_mask=head_mask,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        sequence_output = encoder_outputs[0]
+
+        pooled_output = None
+        if self.pooler is not None:
+            pooled_output = self.pooler(sequence_output.transpose([0, 2, 1]))
+            pooled_output = paddle.flatten(pooled_output, 1)
+
+        if not return_dict:
+            output = (sequence_output, pooled_output) + encoder_outputs[1:]
+            return output
+
+        donut_swin_output = DonutSwinModelOutput(
+            last_hidden_state=sequence_output,
+            pooler_output=pooled_output,
+            hidden_states=encoder_outputs.hidden_states,
+            attentions=encoder_outputs.attentions,
+            reshaped_hidden_states=encoder_outputs.reshaped_hidden_states,
+        )
+        if self.training:
+            return donut_swin_output, label, attention_mask
+        else:
+            return donut_swin_output
diff --git a/ppocr/modeling/backbones/rec_efficientb3_pren.py b/ppocr/modeling/backbones/rec_efficientb3_pren.py
new file mode 100644
index 0000000..5334cc3
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_efficientb3_pren.py
@@ -0,0 +1,305 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+Code is refer from:
+https://github.com/RuijieJ/pren/blob/main/Nets/EfficientNet.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import re
+import collections
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+__all__ = ["EfficientNetb3_PREN"]
+
+GlobalParams = collections.namedtuple(
+    "GlobalParams",
+    [
+        "batch_norm_momentum",
+        "batch_norm_epsilon",
+        "dropout_rate",
+        "num_classes",
+        "width_coefficient",
+        "depth_coefficient",
+        "depth_divisor",
+        "min_depth",
+        "drop_connect_rate",
+        "image_size",
+    ],
+)
+
+BlockArgs = collections.namedtuple(
+    "BlockArgs",
+    [
+        "kernel_size",
+        "num_repeat",
+        "input_filters",
+        "output_filters",
+        "expand_ratio",
+        "id_skip",
+        "stride",
+        "se_ratio",
+    ],
+)
+
+
+class BlockDecoder:
+    @staticmethod
+    def _decode_block_string(block_string):
+        assert isinstance(block_string, str)
+
+        ops = block_string.split("_")
+        options = {}
+        for op in ops:
+            splits = re.split(r"(\d.*)", op)
+            if len(splits) >= 2:
+                key, value = splits[:2]
+                options[key] = value
+
+        assert ("s" in options and len(options["s"]) == 1) or (
+            len(options["s"]) == 2 and options["s"][0] == options["s"][1]
+        )
+
+        return BlockArgs(
+            kernel_size=int(options["k"]),
+            num_repeat=int(options["r"]),
+            input_filters=int(options["i"]),
+            output_filters=int(options["o"]),
+            expand_ratio=int(options["e"]),
+            id_skip=("noskip" not in block_string),
+            se_ratio=float(options["se"]) if "se" in options else None,
+            stride=[int(options["s"][0])],
+        )
+
+    @staticmethod
+    def decode(string_list):
+        assert isinstance(string_list, list)
+        blocks_args = []
+        for block_string in string_list:
+            blocks_args.append(BlockDecoder._decode_block_string(block_string))
+        return blocks_args
+
+
+def efficientnet(
+    width_coefficient=None,
+    depth_coefficient=None,
+    dropout_rate=0.2,
+    drop_connect_rate=0.2,
+    image_size=None,
+    num_classes=1000,
+):
+    blocks_args = [
+        "r1_k3_s11_e1_i32_o16_se0.25",
+        "r2_k3_s22_e6_i16_o24_se0.25",
+        "r2_k5_s22_e6_i24_o40_se0.25",
+        "r3_k3_s22_e6_i40_o80_se0.25",
+        "r3_k5_s11_e6_i80_o112_se0.25",
+        "r4_k5_s22_e6_i112_o192_se0.25",
+        "r1_k3_s11_e6_i192_o320_se0.25",
+    ]
+    blocks_args = BlockDecoder.decode(blocks_args)
+
+    global_params = GlobalParams(
+        batch_norm_momentum=0.99,
+        batch_norm_epsilon=1e-3,
+        dropout_rate=dropout_rate,
+        drop_connect_rate=drop_connect_rate,
+        num_classes=num_classes,
+        width_coefficient=width_coefficient,
+        depth_coefficient=depth_coefficient,
+        depth_divisor=8,
+        min_depth=None,
+        image_size=image_size,
+    )
+    return blocks_args, global_params
+
+
+class EffUtils:
+    @staticmethod
+    def round_filters(filters, global_params):
+        """Calculate and round number of filters based on depth multiplier."""
+        multiplier = global_params.width_coefficient
+        if not multiplier:
+            return filters
+        divisor = global_params.depth_divisor
+        min_depth = global_params.min_depth
+        filters *= multiplier
+        min_depth = min_depth or divisor
+        new_filters = max(min_depth, int(filters + divisor / 2) // divisor * divisor)
+        if new_filters < 0.9 * filters:
+            new_filters += divisor
+        return int(new_filters)
+
+    @staticmethod
+    def round_repeats(repeats, global_params):
+        """Round number of filters based on depth multiplier."""
+        multiplier = global_params.depth_coefficient
+        if not multiplier:
+            return repeats
+        return int(math.ceil(multiplier * repeats))
+
+
+class MbConvBlock(nn.Layer):
+    def __init__(self, block_args):
+        super(MbConvBlock, self).__init__()
+        self._block_args = block_args
+        self.has_se = (self._block_args.se_ratio is not None) and (
+            0 < self._block_args.se_ratio <= 1
+        )
+        self.id_skip = block_args.id_skip
+
+        # expansion phase
+        self.inp = self._block_args.input_filters
+        oup = self._block_args.input_filters * self._block_args.expand_ratio
+        if self._block_args.expand_ratio != 1:
+            self._expand_conv = nn.Conv2D(self.inp, oup, 1, bias_attr=False)
+            self._bn0 = nn.BatchNorm(oup)
+
+        # depthwise conv phase
+        k = self._block_args.kernel_size
+        s = self._block_args.stride
+        if isinstance(s, list):
+            s = s[0]
+        self._depthwise_conv = nn.Conv2D(
+            oup,
+            oup,
+            groups=oup,
+            kernel_size=k,
+            stride=s,
+            padding="same",
+            bias_attr=False,
+        )
+        self._bn1 = nn.BatchNorm(oup)
+
+        # squeeze and excitation layer, if desired
+        if self.has_se:
+            num_squeezed_channels = max(
+                1, int(self._block_args.input_filters * self._block_args.se_ratio)
+            )
+            self._se_reduce = nn.Conv2D(oup, num_squeezed_channels, 1)
+            self._se_expand = nn.Conv2D(num_squeezed_channels, oup, 1)
+
+        # output phase and some util class
+        self.final_oup = self._block_args.output_filters
+        self._project_conv = nn.Conv2D(oup, self.final_oup, 1, bias_attr=False)
+        self._bn2 = nn.BatchNorm(self.final_oup)
+        self._swish = nn.Swish()
+
+    def _drop_connect(self, inputs, p, training):
+        if not training:
+            return inputs
+        batch_size = inputs.shape[0]
+        keep_prob = 1 - p
+        random_tensor = keep_prob
+        random_tensor += paddle.rand([batch_size, 1, 1, 1], dtype=inputs.dtype)
+        random_tensor = paddle.to_tensor(random_tensor, place=inputs.place)
+        binary_tensor = paddle.floor(random_tensor)
+        output = inputs / keep_prob * binary_tensor
+        return output
+
+    def forward(self, inputs, drop_connect_rate=None):
+        # expansion and depthwise conv
+        x = inputs
+        if self._block_args.expand_ratio != 1:
+            x = self._swish(self._bn0(self._expand_conv(inputs)))
+        x = self._swish(self._bn1(self._depthwise_conv(x)))
+
+        # squeeze and excitation
+        if self.has_se:
+            x_squeezed = F.adaptive_avg_pool2d(x, 1)
+            x_squeezed = self._se_expand(self._swish(self._se_reduce(x_squeezed)))
+            x = F.sigmoid(x_squeezed) * x
+        x = self._bn2(self._project_conv(x))
+
+        # skip connection and drop connect
+        if self.id_skip and self._block_args.stride == 1 and self.inp == self.final_oup:
+            if drop_connect_rate:
+                x = self._drop_connect(x, p=drop_connect_rate, training=self.training)
+            x = x + inputs
+        return x
+
+
+class EfficientNetb3_PREN(nn.Layer):
+    def __init__(self, in_channels):
+        super(EfficientNetb3_PREN, self).__init__()
+        """
+        the fllowing are efficientnetb3's superparams,
+        they means efficientnetb3 network's width, depth, resolution and
+        dropout respectively, to fit for text recognition task, the resolution
+        here is changed from 300 to 64.
+        """
+        w, d, s, p = 1.2, 1.4, 64, 0.3
+        self._blocks_args, self._global_params = efficientnet(
+            width_coefficient=w, depth_coefficient=d, dropout_rate=p, image_size=s
+        )
+        self.out_channels = []
+        # stem
+        out_channels = EffUtils.round_filters(32, self._global_params)
+        self._conv_stem = nn.Conv2D(
+            in_channels, out_channels, 3, 2, padding="same", bias_attr=False
+        )
+        self._bn0 = nn.BatchNorm(out_channels)
+
+        # build blocks
+        self._blocks = []
+        # to extract three feature maps for fpn based on efficientnetb3 backbone
+        self._concerned_block_idxes = [7, 17, 25]
+        _concerned_idx = 0
+        for i, block_args in enumerate(self._blocks_args):
+            block_args = block_args._replace(
+                input_filters=EffUtils.round_filters(
+                    block_args.input_filters, self._global_params
+                ),
+                output_filters=EffUtils.round_filters(
+                    block_args.output_filters, self._global_params
+                ),
+                num_repeat=EffUtils.round_repeats(
+                    block_args.num_repeat, self._global_params
+                ),
+            )
+            self._blocks.append(self.add_sublayer(f"{i}-0", MbConvBlock(block_args)))
+            _concerned_idx += 1
+            if _concerned_idx in self._concerned_block_idxes:
+                self.out_channels.append(block_args.output_filters)
+            if block_args.num_repeat > 1:
+                block_args = block_args._replace(
+                    input_filters=block_args.output_filters, stride=1
+                )
+            for j in range(block_args.num_repeat - 1):
+                self._blocks.append(
+                    self.add_sublayer(f"{i}-{j+1}", MbConvBlock(block_args))
+                )
+                _concerned_idx += 1
+                if _concerned_idx in self._concerned_block_idxes:
+                    self.out_channels.append(block_args.output_filters)
+
+        self._swish = nn.Swish()
+
+    def forward(self, inputs):
+        outs = []
+        x = self._swish(self._bn0(self._conv_stem(inputs)))
+        for idx, block in enumerate(self._blocks):
+            drop_connect_rate = self._global_params.drop_connect_rate
+            if drop_connect_rate:
+                drop_connect_rate *= float(idx) / len(self._blocks)
+            x = block(x, drop_connect_rate=drop_connect_rate)
+            if idx in self._concerned_block_idxes:
+                outs.append(x)
+        return outs
diff --git a/ppocr/modeling/backbones/rec_hgnet.py b/ppocr/modeling/backbones/rec_hgnet.py
new file mode 100644
index 0000000..d33e1dc
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_hgnet.py
@@ -0,0 +1,385 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle.nn.initializer import KaimingNormal, Constant
+from paddle.nn import Conv2D, BatchNorm2D, ReLU, AdaptiveAvgPool2D, MaxPool2D
+from paddle.regularizer import L2Decay
+from paddle import ParamAttr
+
+kaiming_normal_ = KaimingNormal()
+zeros_ = Constant(value=0.0)
+ones_ = Constant(value=1.0)
+
+
+class MeanPool2D(nn.Layer):
+    def __init__(self, w, h):
+        super().__init__()
+        self.w = w
+        self.h = h
+
+    def forward(self, feat):
+        batch_size, channels, _, _ = feat.shape
+        feat_flat = paddle.reshape(feat, [batch_size, channels, -1])
+        feat_mean = paddle.mean(feat_flat, axis=2)
+        feat_mean = paddle.reshape(feat_mean, [batch_size, channels, self.w, self.h])
+        return feat_mean
+
+
+class ConvBNAct(nn.Layer):
+    def __init__(
+        self, in_channels, out_channels, kernel_size, stride, groups=1, use_act=True
+    ):
+        super().__init__()
+        self.use_act = use_act
+        self.conv = Conv2D(
+            in_channels,
+            out_channels,
+            kernel_size,
+            stride,
+            padding=(kernel_size - 1) // 2,
+            groups=groups,
+            bias_attr=False,
+        )
+        self.bn = BatchNorm2D(
+            out_channels,
+            weight_attr=ParamAttr(regularizer=L2Decay(0.0)),
+            bias_attr=ParamAttr(regularizer=L2Decay(0.0)),
+        )
+        if self.use_act:
+            self.act = ReLU()
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        if self.use_act:
+            x = self.act(x)
+        return x
+
+
+class ESEModule(nn.Layer):
+    def __init__(self, channels):
+        super().__init__()
+        if "npu" in paddle.device.get_device():
+            self.avg_pool = MeanPool2D(1, 1)
+        else:
+            self.avg_pool = AdaptiveAvgPool2D(1)
+        self.conv = Conv2D(
+            in_channels=channels,
+            out_channels=channels,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        )
+        self.sigmoid = nn.Sigmoid()
+
+    def forward(self, x):
+        identity = x
+        x = self.avg_pool(x)
+        x = self.conv(x)
+        x = self.sigmoid(x)
+        return paddle.multiply(x=identity, y=x)
+
+
+class HG_Block(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        mid_channels,
+        out_channels,
+        layer_num,
+        identity=False,
+    ):
+        super().__init__()
+        self.identity = identity
+
+        self.layers = nn.LayerList()
+        self.layers.append(
+            ConvBNAct(
+                in_channels=in_channels,
+                out_channels=mid_channels,
+                kernel_size=3,
+                stride=1,
+            )
+        )
+        for _ in range(layer_num - 1):
+            self.layers.append(
+                ConvBNAct(
+                    in_channels=mid_channels,
+                    out_channels=mid_channels,
+                    kernel_size=3,
+                    stride=1,
+                )
+            )
+
+        # feature aggregation
+        total_channels = in_channels + layer_num * mid_channels
+        self.aggregation_conv = ConvBNAct(
+            in_channels=total_channels,
+            out_channels=out_channels,
+            kernel_size=1,
+            stride=1,
+        )
+        self.att = ESEModule(out_channels)
+
+    def forward(self, x):
+        identity = x
+        output = []
+        output.append(x)
+        for layer in self.layers:
+            x = layer(x)
+            output.append(x)
+        x = paddle.concat(output, axis=1)
+        x = self.aggregation_conv(x)
+        x = self.att(x)
+        if self.identity:
+            x += identity
+        return x
+
+
+class HG_Stage(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        mid_channels,
+        out_channels,
+        block_num,
+        layer_num,
+        downsample=True,
+        stride=[2, 1],
+    ):
+        super().__init__()
+        self.downsample = downsample
+        if downsample:
+            self.downsample = ConvBNAct(
+                in_channels=in_channels,
+                out_channels=in_channels,
+                kernel_size=3,
+                stride=stride,
+                groups=in_channels,
+                use_act=False,
+            )
+
+        blocks_list = []
+        blocks_list.append(
+            HG_Block(in_channels, mid_channels, out_channels, layer_num, identity=False)
+        )
+        for _ in range(block_num - 1):
+            blocks_list.append(
+                HG_Block(
+                    out_channels, mid_channels, out_channels, layer_num, identity=True
+                )
+            )
+        self.blocks = nn.Sequential(*blocks_list)
+
+    def forward(self, x):
+        if self.downsample:
+            x = self.downsample(x)
+        x = self.blocks(x)
+        return x
+
+
+class PPHGNet(nn.Layer):
+    """
+    PPHGNet
+    Args:
+        stem_channels: list. Stem channel list of PPHGNet.
+        stage_config: dict. The configuration of each stage of PPHGNet. such as the number of channels, stride, etc.
+        layer_num: int. Number of layers of HG_Block.
+        use_last_conv: boolean. Whether to use a 1x1 convolutional layer before the classification layer.
+        class_expand: int=2048. Number of channels for the last 1x1 convolutional layer.
+        dropout_prob: float. Parameters of dropout, 0.0 means dropout is not used.
+        class_num: int=1000. The number of classes.
+    Returns:
+        model: nn.Layer. Specific PPHGNet model depends on args.
+    """
+
+    def __init__(
+        self,
+        stem_channels,
+        stage_config,
+        layer_num,
+        in_channels=3,
+        det=False,
+        out_indices=None,
+    ):
+        super().__init__()
+        self.det = det
+        self.out_indices = out_indices if out_indices is not None else [0, 1, 2, 3]
+
+        # stem
+        stem_channels.insert(0, in_channels)
+        self.stem = nn.Sequential(
+            *[
+                ConvBNAct(
+                    in_channels=stem_channels[i],
+                    out_channels=stem_channels[i + 1],
+                    kernel_size=3,
+                    stride=2 if i == 0 else 1,
+                )
+                for i in range(len(stem_channels) - 1)
+            ]
+        )
+
+        if self.det:
+            self.pool = nn.MaxPool2D(kernel_size=3, stride=2, padding=1)
+        # stages
+        self.stages = nn.LayerList()
+        self.out_channels = []
+        for block_id, k in enumerate(stage_config):
+            (
+                in_channels,
+                mid_channels,
+                out_channels,
+                block_num,
+                downsample,
+                stride,
+            ) = stage_config[k]
+            self.stages.append(
+                HG_Stage(
+                    in_channels,
+                    mid_channels,
+                    out_channels,
+                    block_num,
+                    layer_num,
+                    downsample,
+                    stride,
+                )
+            )
+            if block_id in self.out_indices:
+                self.out_channels.append(out_channels)
+
+        if not self.det:
+            self.out_channels = stage_config["stage4"][2]
+
+        self._init_weights()
+
+    def _init_weights(self):
+        for m in self.sublayers():
+            if isinstance(m, nn.Conv2D):
+                kaiming_normal_(m.weight)
+            elif isinstance(m, (nn.BatchNorm2D)):
+                ones_(m.weight)
+                zeros_(m.bias)
+            elif isinstance(m, nn.Linear):
+                zeros_(m.bias)
+
+    def forward(self, x):
+        x = self.stem(x)
+        if self.det:
+            x = self.pool(x)
+
+        out = []
+        for i, stage in enumerate(self.stages):
+            x = stage(x)
+            if self.det and i in self.out_indices:
+                out.append(x)
+        if self.det:
+            return out
+
+        if self.training:
+            x = F.adaptive_avg_pool2d(x, [1, 40])
+        else:
+            x = F.avg_pool2d(x, [3, 2])
+        return x
+
+
+def PPHGNet_tiny(pretrained=False, use_ssld=False, **kwargs):
+    """
+    PPHGNet_tiny
+    Args:
+        pretrained: bool=False or str. If `True` load pretrained parameters, `False` otherwise.
+                    If str, means the path of the pretrained model.
+        use_ssld: bool=False. Whether using distillation pretrained model when pretrained=True.
+    Returns:
+        model: nn.Layer. Specific `PPHGNet_tiny` model depends on args.
+    """
+    stage_config = {
+        # in_channels, mid_channels, out_channels, blocks, downsample
+        "stage1": [96, 96, 224, 1, False, [2, 1]],
+        "stage2": [224, 128, 448, 1, True, [1, 2]],
+        "stage3": [448, 160, 512, 2, True, [2, 1]],
+        "stage4": [512, 192, 768, 1, True, [2, 1]],
+    }
+
+    model = PPHGNet(
+        stem_channels=[48, 48, 96], stage_config=stage_config, layer_num=5, **kwargs
+    )
+    return model
+
+
+def PPHGNet_small(pretrained=False, use_ssld=False, det=False, **kwargs):
+    """
+    PPHGNet_small
+    Args:
+        pretrained: bool=False or str. If `True` load pretrained parameters, `False` otherwise.
+                    If str, means the path of the pretrained model.
+        use_ssld: bool=False. Whether using distillation pretrained model when pretrained=True.
+    Returns:
+        model: nn.Layer. Specific `PPHGNet_small` model depends on args.
+    """
+    stage_config_det = {
+        # in_channels, mid_channels, out_channels, blocks, downsample
+        "stage1": [128, 128, 256, 1, False, 2],
+        "stage2": [256, 160, 512, 1, True, 2],
+        "stage3": [512, 192, 768, 2, True, 2],
+        "stage4": [768, 224, 1024, 1, True, 2],
+    }
+
+    stage_config_rec = {
+        # in_channels, mid_channels, out_channels, blocks, downsample
+        "stage1": [128, 128, 256, 1, True, [2, 1]],
+        "stage2": [256, 160, 512, 1, True, [1, 2]],
+        "stage3": [512, 192, 768, 2, True, [2, 1]],
+        "stage4": [768, 224, 1024, 1, True, [2, 1]],
+    }
+
+    model = PPHGNet(
+        stem_channels=[64, 64, 128],
+        stage_config=stage_config_det if det else stage_config_rec,
+        layer_num=6,
+        det=det,
+        **kwargs,
+    )
+    return model
+
+
+def PPHGNet_base(pretrained=False, use_ssld=True, **kwargs):
+    """
+    PPHGNet_base
+    Args:
+        pretrained: bool=False or str. If `True` load pretrained parameters, `False` otherwise.
+                    If str, means the path of the pretrained model.
+        use_ssld: bool=False. Whether using distillation pretrained model when pretrained=True.
+    Returns:
+        model: nn.Layer. Specific `PPHGNet_base` model depends on args.
+    """
+    stage_config = {
+        # in_channels, mid_channels, out_channels, blocks, downsample
+        "stage1": [160, 192, 320, 1, False, [2, 1]],
+        "stage2": [320, 224, 640, 2, True, [1, 2]],
+        "stage3": [640, 256, 960, 3, True, [2, 1]],
+        "stage4": [960, 288, 1280, 2, True, [2, 1]],
+    }
+
+    model = PPHGNet(
+        stem_channels=[96, 96, 160],
+        stage_config=stage_config,
+        layer_num=7,
+        dropout_prob=0.2,
+        **kwargs,
+    )
+    return model
diff --git a/ppocr/modeling/backbones/rec_hybridvit.py b/ppocr/modeling/backbones/rec_hybridvit.py
new file mode 100644
index 0000000..02aa305
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_hybridvit.py
@@ -0,0 +1,528 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+This code is refer from:
+https://github.com/huggingface/pytorch-image-models/blob/main/timm/models/vision_transformer_hybrid.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from itertools import repeat
+import collections
+import math
+from functools import partial
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from ppocr.modeling.backbones.rec_resnetv2 import (
+    ResNetV2,
+    StdConv2dSame,
+    DropPath,
+    get_padding,
+)
+from paddle.nn.initializer import (
+    TruncatedNormal,
+    Constant,
+    Normal,
+    KaimingUniform,
+    XavierUniform,
+)
+
+normal_ = Normal(mean=0.0, std=1e-6)
+zeros_ = Constant(value=0.0)
+ones_ = Constant(value=1.0)
+kaiming_normal_ = KaimingUniform(nonlinearity="relu")
+trunc_normal_ = TruncatedNormal(std=0.02)
+xavier_uniform_ = XavierUniform()
+
+
+def _ntuple(n):
+    def parse(x):
+        if isinstance(x, collections.abc.Iterable):
+            return x
+        return tuple(repeat(x, n))
+
+    return parse
+
+
+to_1tuple = _ntuple(1)
+to_2tuple = _ntuple(2)
+to_3tuple = _ntuple(3)
+to_4tuple = _ntuple(4)
+to_ntuple = _ntuple
+
+
+class Conv2dAlign(nn.Conv2D):
+    """Conv2d with Weight Standardization. Used for BiT ResNet-V2 models.
+
+    Paper: `Micro-Batch Training with Batch-Channel Normalization and Weight Standardization` -
+        https://arxiv.org/abs/1903.10520v2
+    """
+
+    def __init__(
+        self,
+        in_channel,
+        out_channels,
+        kernel_size,
+        stride=1,
+        padding=0,
+        dilation=1,
+        groups=1,
+        bias=True,
+        eps=1e-6,
+    ):
+
+        super().__init__(
+            in_channel,
+            out_channels,
+            kernel_size,
+            stride=stride,
+            padding=padding,
+            dilation=dilation,
+            groups=groups,
+            bias_attr=bias,
+            weight_attr=True,
+        )
+        self.eps = eps
+
+    def forward(self, x):
+        x = F.conv2d(
+            x,
+            self.weight,
+            self.bias,
+            self._stride,
+            self._padding,
+            self._dilation,
+            self._groups,
+        )
+        return x
+
+
+class HybridEmbed(nn.Layer):
+    """CNN Feature Map Embedding
+    Extract feature map from CNN, flatten, project to embedding dim.
+    """
+
+    def __init__(
+        self,
+        backbone,
+        img_size=224,
+        patch_size=1,
+        feature_size=None,
+        in_chans=3,
+        embed_dim=768,
+    ):
+        super().__init__()
+        assert isinstance(backbone, nn.Layer)
+        img_size = to_2tuple(img_size)
+        patch_size = to_2tuple(patch_size)
+        self.img_size = img_size
+        self.patch_size = patch_size
+        self.backbone = backbone
+        feature_dim = 1024
+        feature_size = (42, 12)
+        patch_size = (1, 1)
+        assert (
+            feature_size[0] % patch_size[0] == 0
+            and feature_size[1] % patch_size[1] == 0
+        )
+
+        self.grid_size = (
+            feature_size[0] // patch_size[0],
+            feature_size[1] // patch_size[1],
+        )
+        self.num_patches = self.grid_size[0] * self.grid_size[1]
+        self.proj = nn.Conv2D(
+            feature_dim,
+            embed_dim,
+            kernel_size=patch_size,
+            stride=patch_size,
+            weight_attr=True,
+            bias_attr=True,
+        )
+
+    def forward(self, x):
+
+        x = self.backbone(x)
+        if isinstance(x, (list, tuple)):
+            x = x[-1]  # last feature if backbone outputs list/tuple of features
+        x = self.proj(x).flatten(2).transpose([0, 2, 1])
+
+        return x
+
+
+class myLinear(nn.Linear):
+    def __init__(self, in_channel, out_channels, weight_attr=True, bias_attr=True):
+        super().__init__(
+            in_channel, out_channels, weight_attr=weight_attr, bias_attr=bias_attr
+        )
+
+    def forward(self, x):
+        return paddle.matmul(x, self.weight, transpose_y=True) + self.bias
+
+
+class Attention(nn.Layer):
+    def __init__(self, dim, num_heads=8, qkv_bias=False, attn_drop=0.0, proj_drop=0.0):
+        super().__init__()
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        self.scale = head_dim**-0.5
+
+        self.qkv = nn.Linear(dim, dim * 3, bias_attr=qkv_bias)
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.proj = myLinear(dim, dim, weight_attr=True, bias_attr=True)
+        self.proj_drop = nn.Dropout(proj_drop)
+
+    def forward(self, x):
+        B, N, C = x.shape
+        qkv = (
+            self.qkv(x)
+            .reshape([B, N, 3, self.num_heads, C // self.num_heads])
+            .transpose([2, 0, 3, 1, 4])
+        )
+        q, k, v = qkv.unbind(0)  # make torchscript happy (cannot use tensor as tuple)
+
+        attn = (q @ k.transpose([0, 1, 3, 2])) * self.scale
+
+        attn = F.softmax(attn, axis=-1)
+        attn = self.attn_drop(attn)
+
+        x = (attn @ v).transpose([0, 2, 1, 3]).reshape([B, N, C])
+
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+
+
+class Mlp(nn.Layer):
+    """MLP as used in Vision Transformer, MLP-Mixer and related networks"""
+
+    def __init__(
+        self,
+        in_features,
+        hidden_features=None,
+        out_features=None,
+        act_layer=nn.GELU,
+        drop=0.0,
+    ):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        drop_probs = to_2tuple(drop)
+
+        self.fc1 = nn.Linear(in_features, hidden_features)
+        self.act = act_layer()
+        self.drop1 = nn.Dropout(drop_probs[0])
+        self.fc2 = nn.Linear(hidden_features, out_features)
+        self.drop2 = nn.Dropout(drop_probs[1])
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.act(x)
+        x = self.drop1(x)
+        x = self.fc2(x)
+        x = self.drop2(x)
+        return x
+
+
+class Block(nn.Layer):
+    def __init__(
+        self,
+        dim,
+        num_heads,
+        mlp_ratio=4.0,
+        qkv_bias=False,
+        drop=0.0,
+        attn_drop=0.0,
+        drop_path=0.0,
+        act_layer=nn.GELU,
+        norm_layer=nn.LayerNorm,
+    ):
+        super().__init__()
+        self.norm1 = norm_layer(dim)
+        self.attn = Attention(
+            dim,
+            num_heads=num_heads,
+            qkv_bias=qkv_bias,
+            attn_drop=attn_drop,
+            proj_drop=drop,
+        )
+        # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
+        self.drop_path = DropPath(drop_path) if drop_path > 0.0 else nn.Identity()
+        self.norm2 = norm_layer(dim)
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp = Mlp(
+            in_features=dim,
+            hidden_features=mlp_hidden_dim,
+            act_layer=act_layer,
+            drop=drop,
+        )
+
+    def forward(self, x):
+
+        x = x + self.drop_path(self.attn(self.norm1(x)))
+        x = x + self.drop_path(self.mlp(self.norm2(x)))
+        return x
+
+
+class HybridTransformer(nn.Layer):
+    """Implementation of HybridTransformer.
+
+    Args:
+      x: input images with shape [N, 1, H, W]
+      label: LaTeX-OCR labels with shape [N, L] , L is the max sequence length
+      attention_mask: LaTeX-OCR attention mask with shape [N, L]  , L is the max sequence length
+
+    Returns:
+      The encoded features with shape [N, 1, H//16, W//16]
+    """
+
+    def __init__(
+        self,
+        backbone_layers=[2, 3, 7],
+        input_channel=1,
+        is_predict=False,
+        is_export=False,
+        img_size=(224, 224),
+        patch_size=16,
+        num_classes=1000,
+        embed_dim=768,
+        depth=12,
+        num_heads=12,
+        mlp_ratio=4.0,
+        qkv_bias=True,
+        representation_size=None,
+        distilled=False,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.0,
+        embed_layer=None,
+        norm_layer=None,
+        act_layer=None,
+        weight_init="",
+        **kwargs,
+    ):
+        super(HybridTransformer, self).__init__()
+        self.num_classes = num_classes
+        self.num_features = self.embed_dim = (
+            embed_dim  # num_features for consistency with other models
+        )
+        self.num_tokens = 2 if distilled else 1
+        norm_layer = norm_layer or partial(nn.LayerNorm, epsilon=1e-6)
+        act_layer = act_layer or nn.GELU
+        self.height, self.width = img_size
+        self.patch_size = patch_size
+        backbone = ResNetV2(
+            layers=backbone_layers,
+            num_classes=0,
+            global_pool="",
+            in_chans=input_channel,
+            preact=False,
+            stem_type="same",
+            conv_layer=StdConv2dSame,
+            is_export=is_export,
+        )
+        min_patch_size = 2 ** (len(backbone_layers) + 1)
+        self.patch_embed = HybridEmbed(
+            img_size=img_size,
+            patch_size=patch_size // min_patch_size,
+            in_chans=input_channel,
+            embed_dim=embed_dim,
+            backbone=backbone,
+        )
+        num_patches = self.patch_embed.num_patches
+
+        self.cls_token = paddle.create_parameter([1, 1, embed_dim], dtype="float32")
+        self.dist_token = (
+            paddle.create_parameter(
+                [1, 1, embed_dim],
+                dtype="float32",
+            )
+            if distilled
+            else None
+        )
+        self.pos_embed = paddle.create_parameter(
+            [1, num_patches + self.num_tokens, embed_dim], dtype="float32"
+        )
+        self.pos_drop = nn.Dropout(p=drop_rate)
+        zeros_(self.cls_token)
+        if self.dist_token is not None:
+            zeros_(self.dist_token)
+        zeros_(self.pos_embed)
+
+        dpr = [
+            x.item() for x in paddle.linspace(0, drop_path_rate, depth)
+        ]  # stochastic depth decay rule
+        self.blocks = nn.Sequential(
+            *[
+                Block(
+                    dim=embed_dim,
+                    num_heads=num_heads,
+                    mlp_ratio=mlp_ratio,
+                    qkv_bias=qkv_bias,
+                    drop=drop_rate,
+                    attn_drop=attn_drop_rate,
+                    drop_path=dpr[i],
+                    norm_layer=norm_layer,
+                    act_layer=act_layer,
+                )
+                for i in range(depth)
+            ]
+        )
+        self.norm = norm_layer(embed_dim)
+
+        # Representation layer
+        if representation_size and not distilled:
+            self.num_features = representation_size
+            self.pre_logits = nn.Sequential(
+                ("fc", nn.Linear(embed_dim, representation_size)), ("act", nn.Tanh())
+            )
+        else:
+            self.pre_logits = nn.Identity()
+
+        # Classifier head(s)
+        self.head = (
+            nn.Linear(self.num_features, num_classes)
+            if num_classes > 0
+            else nn.Identity()
+        )
+        self.head_dist = None
+        if distilled:
+            self.head_dist = (
+                nn.Linear(self.embed_dim, self.num_classes)
+                if num_classes > 0
+                else nn.Identity()
+            )
+        self.init_weights(weight_init)
+        self.out_channels = embed_dim
+        self.is_predict = is_predict
+        self.is_export = is_export
+
+    def init_weights(self, mode=""):
+        assert mode in ("jax", "jax_nlhb", "nlhb", "")
+        head_bias = -math.log(self.num_classes) if "nlhb" in mode else 0.0
+        trunc_normal_(self.pos_embed)
+        trunc_normal_(self.cls_token)
+        self.apply(_init_vit_weights)
+
+    def _init_weights(self, m):
+        # this fn left here for compat with downstream users
+        _init_vit_weights(m)
+
+    def load_pretrained(self, checkpoint_path, prefix=""):
+        raise NotImplementedError
+
+    def no_weight_decay(self):
+        return {"pos_embed", "cls_token", "dist_token"}
+
+    def get_classifier(self):
+        if self.dist_token is None:
+            return self.head
+        else:
+            return self.head, self.head_dist
+
+    def reset_classifier(self, num_classes, global_pool=""):
+        self.num_classes = num_classes
+        self.head = (
+            nn.Linear(self.embed_dim, num_classes) if num_classes > 0 else nn.Identity()
+        )
+        if self.num_tokens == 2:
+            self.head_dist = (
+                nn.Linear(self.embed_dim, self.num_classes)
+                if num_classes > 0
+                else nn.Identity()
+            )
+
+    def forward_features(self, x):
+        B, c, h, w = x.shape
+        x = self.patch_embed(x)
+        cls_tokens = self.cls_token.expand(
+            [B, -1, -1]
+        )  # stole cls_tokens impl from Phil Wang, thanks
+        x = paddle.concat((cls_tokens, x), axis=1)
+        h, w = h // self.patch_size, w // self.patch_size
+        repeat_tensor = (
+            paddle.arange(h) * (self.width // self.patch_size - w)
+        ).reshape([-1, 1])
+        repeat_tensor = paddle.repeat_interleave(
+            repeat_tensor, paddle.to_tensor(w), axis=1
+        ).reshape([-1])
+        pos_emb_ind = repeat_tensor + paddle.arange(h * w)
+        pos_emb_ind = paddle.concat(
+            (paddle.zeros([1], dtype="int64"), pos_emb_ind + 1), axis=0
+        ).cast(paddle.int64)
+        x += self.pos_embed[:, pos_emb_ind]
+        x = self.pos_drop(x)
+
+        for blk in self.blocks:
+            x = blk(x)
+
+        x = self.norm(x)
+        return x
+
+    def forward(self, input_data):
+
+        if self.training:
+            x, label, attention_mask = input_data
+        else:
+            if isinstance(input_data, list):
+                x = input_data[0]
+            else:
+                x = input_data
+        x = self.forward_features(x)
+        x = self.head(x)
+        if self.training:
+            return x, label, attention_mask
+        else:
+            return x
+
+
+def _init_vit_weights(
+    module: nn.Layer, name: str = "", head_bias: float = 0.0, jax_impl: bool = False
+):
+    """ViT weight initialization
+    * When called without n, head_bias, jax_impl args it will behave exactly the same
+      as my original init for compatibility with prev hparam / downstream use cases (ie DeiT).
+    * When called w/ valid n (module name) and jax_impl=True, will (hopefully) match JAX impl
+    """
+    if isinstance(module, nn.Linear):
+        if name.startswith("head"):
+            zeros_(module.weight)
+            constant_ = Constant(value=head_bias)
+            constant_(module.bias, head_bias)
+        elif name.startswith("pre_logits"):
+            zeros_(module.bias)
+        else:
+            if jax_impl:
+                xavier_uniform_(module.weight)
+                if module.bias is not None:
+                    if "mlp" in name:
+                        normal_(module.bias)
+                    else:
+                        zeros_(module.bias)
+            else:
+                trunc_normal_(module.weight)
+                if module.bias is not None:
+                    zeros_(module.bias)
+    elif jax_impl and isinstance(module, nn.Conv2D):
+        # NOTE conv was left to pytorch default in my original init
+        if module.bias is not None:
+            zeros_(module.bias)
+    elif isinstance(module, (nn.LayerNorm, nn.GroupNorm, nn.BatchNorm2D)):
+        zeros_(module.bias)
+        ones_(module.weight)
diff --git a/ppocr/modeling/backbones/rec_lcnetv3.py b/ppocr/modeling/backbones/rec_lcnetv3.py
new file mode 100644
index 0000000..387c761
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_lcnetv3.py
@@ -0,0 +1,567 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+from paddle.nn.initializer import Constant, KaimingNormal
+from paddle.nn import (
+    AdaptiveAvgPool2D,
+    BatchNorm2D,
+    Conv2D,
+    Dropout,
+    Hardsigmoid,
+    Hardswish,
+    Identity,
+    Linear,
+    ReLU,
+)
+from paddle.regularizer import L2Decay
+from ppocr.modeling.backbones.rec_hgnet import MeanPool2D
+
+NET_CONFIG_det = {
+    "blocks2":
+    # k, in_c, out_c, s, use_se
+    [[3, 16, 32, 1, False]],
+    "blocks3": [[3, 32, 64, 2, False], [3, 64, 64, 1, False]],
+    "blocks4": [[3, 64, 128, 2, False], [3, 128, 128, 1, False]],
+    "blocks5": [
+        [3, 128, 256, 2, False],
+        [5, 256, 256, 1, False],
+        [5, 256, 256, 1, False],
+        [5, 256, 256, 1, False],
+        [5, 256, 256, 1, False],
+    ],
+    "blocks6": [
+        [5, 256, 512, 2, True],
+        [5, 512, 512, 1, True],
+        [5, 512, 512, 1, False],
+        [5, 512, 512, 1, False],
+    ],
+}
+
+NET_CONFIG_rec = {
+    "blocks2":
+    # k, in_c, out_c, s, use_se
+    [[3, 16, 32, 1, False]],
+    "blocks3": [[3, 32, 64, 1, False], [3, 64, 64, 1, False]],
+    "blocks4": [[3, 64, 128, (2, 1), False], [3, 128, 128, 1, False]],
+    "blocks5": [
+        [3, 128, 256, (1, 2), False],
+        [5, 256, 256, 1, False],
+        [5, 256, 256, 1, False],
+        [5, 256, 256, 1, False],
+        [5, 256, 256, 1, False],
+    ],
+    "blocks6": [
+        [5, 256, 512, (2, 1), True],
+        [5, 512, 512, 1, True],
+        [5, 512, 512, (2, 1), False],
+        [5, 512, 512, 1, False],
+    ],
+}
+
+
+def make_divisible(v, divisor=16, min_value=None):
+    if min_value is None:
+        min_value = divisor
+    new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
+    if new_v < 0.9 * v:
+        new_v += divisor
+    return new_v
+
+
+class LearnableAffineBlock(nn.Layer):
+    def __init__(self, scale_value=1.0, bias_value=0.0, lr_mult=1.0, lab_lr=0.1):
+        super().__init__()
+        self.scale = self.create_parameter(
+            shape=[
+                1,
+            ],
+            default_initializer=Constant(value=scale_value),
+            attr=ParamAttr(learning_rate=lr_mult * lab_lr),
+        )
+        self.add_parameter("scale", self.scale)
+        self.bias = self.create_parameter(
+            shape=[
+                1,
+            ],
+            default_initializer=Constant(value=bias_value),
+            attr=ParamAttr(learning_rate=lr_mult * lab_lr),
+        )
+        self.add_parameter("bias", self.bias)
+
+    def forward(self, x):
+        return self.scale * x + self.bias
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self, in_channels, out_channels, kernel_size, stride, groups=1, lr_mult=1.0
+    ):
+        super().__init__()
+        self.conv = Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=(kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(initializer=KaimingNormal(), learning_rate=lr_mult),
+            bias_attr=False,
+        )
+
+        self.bn = BatchNorm2D(
+            out_channels,
+            weight_attr=ParamAttr(regularizer=L2Decay(0.0), learning_rate=lr_mult),
+            bias_attr=ParamAttr(regularizer=L2Decay(0.0), learning_rate=lr_mult),
+        )
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        return x
+
+
+class Act(nn.Layer):
+    def __init__(self, act="hswish", lr_mult=1.0, lab_lr=0.1):
+        super().__init__()
+        if act == "hswish":
+            self.act = Hardswish()
+        else:
+            assert act == "relu"
+            self.act = ReLU()
+        self.lab = LearnableAffineBlock(lr_mult=lr_mult, lab_lr=lab_lr)
+
+    def forward(self, x):
+        return self.lab(self.act(x))
+
+
+class LearnableRepLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride=1,
+        groups=1,
+        num_conv_branches=1,
+        lr_mult=1.0,
+        lab_lr=0.1,
+    ):
+        super().__init__()
+        self.is_repped = False
+        self.groups = groups
+        self.stride = stride
+        self.kernel_size = kernel_size
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.num_conv_branches = num_conv_branches
+        self.padding = (kernel_size - 1) // 2
+
+        self.identity = (
+            BatchNorm2D(
+                num_features=in_channels,
+                weight_attr=ParamAttr(learning_rate=lr_mult),
+                bias_attr=ParamAttr(learning_rate=lr_mult),
+            )
+            if out_channels == in_channels and stride == 1
+            else None
+        )
+
+        self.conv_kxk = nn.LayerList(
+            [
+                ConvBNLayer(
+                    in_channels,
+                    out_channels,
+                    kernel_size,
+                    stride,
+                    groups=groups,
+                    lr_mult=lr_mult,
+                )
+                for _ in range(self.num_conv_branches)
+            ]
+        )
+
+        self.conv_1x1 = (
+            ConvBNLayer(
+                in_channels, out_channels, 1, stride, groups=groups, lr_mult=lr_mult
+            )
+            if kernel_size > 1
+            else None
+        )
+
+        self.lab = LearnableAffineBlock(lr_mult=lr_mult, lab_lr=lab_lr)
+        self.act = Act(lr_mult=lr_mult, lab_lr=lab_lr)
+
+    def forward(self, x):
+        # for export
+        if self.is_repped:
+            out = self.lab(self.reparam_conv(x))
+            if self.stride != 2:
+                out = self.act(out)
+            return out
+
+        out = 0
+        if self.identity is not None:
+            out += self.identity(x)
+
+        if self.conv_1x1 is not None:
+            out += self.conv_1x1(x)
+
+        for conv in self.conv_kxk:
+            out += conv(x)
+
+        out = self.lab(out)
+        if self.stride != 2:
+            out = self.act(out)
+        return out
+
+    @paddle.no_grad()
+    def rep(self):
+        if self.is_repped:
+            return
+        kernel, bias = self._get_kernel_bias()
+        self.reparam_conv = Conv2D(
+            in_channels=self.in_channels,
+            out_channels=self.out_channels,
+            kernel_size=self.kernel_size,
+            stride=self.stride,
+            padding=self.padding,
+            groups=self.groups,
+        )
+        self.reparam_conv.weight.set_value(kernel)
+        self.reparam_conv.bias.set_value(bias)
+        # Remove training branches to avoid duplicate parameter counting
+        del self.conv_kxk
+        if self.conv_1x1 is not None:
+            del self.conv_1x1
+        if self.identity is not None:
+            del self.identity
+        if hasattr(self, "id_tensor"):
+            del self.id_tensor
+        self.is_repped = True
+
+    def _pad_kernel_1x1_to_kxk(self, kernel1x1, pad):
+        if not isinstance(kernel1x1, paddle.Tensor):
+            return 0
+        else:
+            return nn.functional.pad(kernel1x1, [pad, pad, pad, pad])
+
+    def _get_kernel_bias(self):
+        kernel_conv_1x1, bias_conv_1x1 = self._fuse_bn_tensor(self.conv_1x1)
+        kernel_conv_1x1 = self._pad_kernel_1x1_to_kxk(
+            kernel_conv_1x1, self.kernel_size // 2
+        )
+
+        kernel_identity, bias_identity = self._fuse_bn_tensor(self.identity)
+
+        kernel_conv_kxk = 0
+        bias_conv_kxk = 0
+        for conv in self.conv_kxk:
+            kernel, bias = self._fuse_bn_tensor(conv)
+            kernel_conv_kxk += kernel
+            bias_conv_kxk += bias
+
+        kernel_reparam = kernel_conv_kxk + kernel_conv_1x1 + kernel_identity
+        bias_reparam = bias_conv_kxk + bias_conv_1x1 + bias_identity
+        return kernel_reparam, bias_reparam
+
+    def _fuse_bn_tensor(self, branch):
+        if not branch:
+            return 0, 0
+        elif isinstance(branch, ConvBNLayer):
+            kernel = branch.conv.weight
+            running_mean = branch.bn._mean
+            running_var = branch.bn._variance
+            gamma = branch.bn.weight
+            beta = branch.bn.bias
+            eps = branch.bn._epsilon
+        else:
+            assert isinstance(branch, BatchNorm2D)
+            if not hasattr(self, "id_tensor"):
+                input_dim = self.in_channels // self.groups
+                kernel_value = paddle.zeros(
+                    (self.in_channels, input_dim, self.kernel_size, self.kernel_size),
+                    dtype=branch.weight.dtype,
+                )
+                for i in range(self.in_channels):
+                    kernel_value[
+                        i, i % input_dim, self.kernel_size // 2, self.kernel_size // 2
+                    ] = 1
+                self.id_tensor = kernel_value
+            kernel = self.id_tensor
+            running_mean = branch._mean
+            running_var = branch._variance
+            gamma = branch.weight
+            beta = branch.bias
+            eps = branch._epsilon
+        std = (running_var + eps).sqrt()
+        t = (gamma / std).reshape((-1, 1, 1, 1))
+        return kernel * t, beta - running_mean * gamma / std
+
+
+class SELayer(nn.Layer):
+    def __init__(self, channel, reduction=4, lr_mult=1.0):
+        super().__init__()
+        if "npu" in paddle.device.get_device():
+            self.avg_pool = MeanPool2D(1, 1)
+        else:
+            self.avg_pool = AdaptiveAvgPool2D(1)
+        self.conv1 = Conv2D(
+            in_channels=channel,
+            out_channels=channel // reduction,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            weight_attr=ParamAttr(learning_rate=lr_mult),
+            bias_attr=ParamAttr(learning_rate=lr_mult),
+        )
+        self.relu = ReLU()
+        self.conv2 = Conv2D(
+            in_channels=channel // reduction,
+            out_channels=channel,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            weight_attr=ParamAttr(learning_rate=lr_mult),
+            bias_attr=ParamAttr(learning_rate=lr_mult),
+        )
+        self.hardsigmoid = Hardsigmoid()
+
+    def forward(self, x):
+        identity = x
+        x = self.avg_pool(x)
+        x = self.conv1(x)
+        x = self.relu(x)
+        x = self.conv2(x)
+        x = self.hardsigmoid(x)
+        x = paddle.multiply(x=identity, y=x)
+        return x
+
+
+class LCNetV3Block(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        stride,
+        dw_size,
+        use_se=False,
+        conv_kxk_num=4,
+        lr_mult=1.0,
+        lab_lr=0.1,
+    ):
+        super().__init__()
+        self.use_se = use_se
+        self.dw_conv = LearnableRepLayer(
+            in_channels=in_channels,
+            out_channels=in_channels,
+            kernel_size=dw_size,
+            stride=stride,
+            groups=in_channels,
+            num_conv_branches=conv_kxk_num,
+            lr_mult=lr_mult,
+            lab_lr=lab_lr,
+        )
+        if use_se:
+            self.se = SELayer(in_channels, lr_mult=lr_mult)
+        self.pw_conv = LearnableRepLayer(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=1,
+            stride=1,
+            num_conv_branches=conv_kxk_num,
+            lr_mult=lr_mult,
+            lab_lr=lab_lr,
+        )
+
+    def forward(self, x):
+        x = self.dw_conv(x)
+        if self.use_se:
+            x = self.se(x)
+        x = self.pw_conv(x)
+        return x
+
+
+class PPLCNetV3(nn.Layer):
+    def __init__(
+        self,
+        scale=1.0,
+        conv_kxk_num=4,
+        lr_mult_list=[1.0, 1.0, 1.0, 1.0, 1.0, 1.0],
+        lab_lr=0.1,
+        det=False,
+        **kwargs,
+    ):
+        super().__init__()
+        self.scale = scale
+        self.lr_mult_list = lr_mult_list
+        self.det = det
+
+        self.net_config = NET_CONFIG_det if self.det else NET_CONFIG_rec
+
+        assert isinstance(
+            self.lr_mult_list, (list, tuple)
+        ), "lr_mult_list should be in (list, tuple) but got {}".format(
+            type(self.lr_mult_list)
+        )
+        assert (
+            len(self.lr_mult_list) == 6
+        ), "lr_mult_list length should be 6 but got {}".format(len(self.lr_mult_list))
+
+        self.conv1 = ConvBNLayer(
+            in_channels=3,
+            out_channels=make_divisible(16 * scale),
+            kernel_size=3,
+            stride=2,
+            lr_mult=self.lr_mult_list[0],
+        )
+
+        self.blocks2 = nn.Sequential(
+            *[
+                LCNetV3Block(
+                    in_channels=make_divisible(in_c * scale),
+                    out_channels=make_divisible(out_c * scale),
+                    dw_size=k,
+                    stride=s,
+                    use_se=se,
+                    conv_kxk_num=conv_kxk_num,
+                    lr_mult=self.lr_mult_list[1],
+                    lab_lr=lab_lr,
+                )
+                for i, (k, in_c, out_c, s, se) in enumerate(self.net_config["blocks2"])
+            ]
+        )
+
+        self.blocks3 = nn.Sequential(
+            *[
+                LCNetV3Block(
+                    in_channels=make_divisible(in_c * scale),
+                    out_channels=make_divisible(out_c * scale),
+                    dw_size=k,
+                    stride=s,
+                    use_se=se,
+                    conv_kxk_num=conv_kxk_num,
+                    lr_mult=self.lr_mult_list[2],
+                    lab_lr=lab_lr,
+                )
+                for i, (k, in_c, out_c, s, se) in enumerate(self.net_config["blocks3"])
+            ]
+        )
+
+        self.blocks4 = nn.Sequential(
+            *[
+                LCNetV3Block(
+                    in_channels=make_divisible(in_c * scale),
+                    out_channels=make_divisible(out_c * scale),
+                    dw_size=k,
+                    stride=s,
+                    use_se=se,
+                    conv_kxk_num=conv_kxk_num,
+                    lr_mult=self.lr_mult_list[3],
+                    lab_lr=lab_lr,
+                )
+                for i, (k, in_c, out_c, s, se) in enumerate(self.net_config["blocks4"])
+            ]
+        )
+
+        self.blocks5 = nn.Sequential(
+            *[
+                LCNetV3Block(
+                    in_channels=make_divisible(in_c * scale),
+                    out_channels=make_divisible(out_c * scale),
+                    dw_size=k,
+                    stride=s,
+                    use_se=se,
+                    conv_kxk_num=conv_kxk_num,
+                    lr_mult=self.lr_mult_list[4],
+                    lab_lr=lab_lr,
+                )
+                for i, (k, in_c, out_c, s, se) in enumerate(self.net_config["blocks5"])
+            ]
+        )
+
+        self.blocks6 = nn.Sequential(
+            *[
+                LCNetV3Block(
+                    in_channels=make_divisible(in_c * scale),
+                    out_channels=make_divisible(out_c * scale),
+                    dw_size=k,
+                    stride=s,
+                    use_se=se,
+                    conv_kxk_num=conv_kxk_num,
+                    lr_mult=self.lr_mult_list[5],
+                    lab_lr=lab_lr,
+                )
+                for i, (k, in_c, out_c, s, se) in enumerate(self.net_config["blocks6"])
+            ]
+        )
+        self.out_channels = make_divisible(512 * scale)
+
+        if self.det:
+            mv_c = [16, 24, 56, 480]
+            self.out_channels = [
+                make_divisible(self.net_config["blocks3"][-1][2] * scale),
+                make_divisible(self.net_config["blocks4"][-1][2] * scale),
+                make_divisible(self.net_config["blocks5"][-1][2] * scale),
+                make_divisible(self.net_config["blocks6"][-1][2] * scale),
+            ]
+
+            self.layer_list = nn.LayerList(
+                [
+                    nn.Conv2D(self.out_channels[0], int(mv_c[0] * scale), 1, 1, 0),
+                    nn.Conv2D(self.out_channels[1], int(mv_c[1] * scale), 1, 1, 0),
+                    nn.Conv2D(self.out_channels[2], int(mv_c[2] * scale), 1, 1, 0),
+                    nn.Conv2D(self.out_channels[3], int(mv_c[3] * scale), 1, 1, 0),
+                ]
+            )
+            self.out_channels = [
+                int(mv_c[0] * scale),
+                int(mv_c[1] * scale),
+                int(mv_c[2] * scale),
+                int(mv_c[3] * scale),
+            ]
+
+    def forward(self, x):
+        out_list = []
+        x = self.conv1(x)
+
+        x = self.blocks2(x)
+        x = self.blocks3(x)
+        out_list.append(x)
+        x = self.blocks4(x)
+        out_list.append(x)
+        x = self.blocks5(x)
+        out_list.append(x)
+        x = self.blocks6(x)
+        out_list.append(x)
+
+        if self.det:
+            out_list[0] = self.layer_list[0](out_list[0])
+            out_list[1] = self.layer_list[1](out_list[1])
+            out_list[2] = self.layer_list[2](out_list[2])
+            out_list[3] = self.layer_list[3](out_list[3])
+            return out_list
+
+        if self.training:
+            x = F.adaptive_avg_pool2d(x, [1, 40])
+        else:
+            x = F.avg_pool2d(x, [3, 2])
+        return x
diff --git a/ppocr/modeling/backbones/rec_lcnetv4.py b/ppocr/modeling/backbones/rec_lcnetv4.py
new file mode 100644
index 0000000..405da04
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_lcnetv4.py
@@ -0,0 +1,669 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+from paddle.nn.initializer import Constant
+from paddle.nn import (
+    BatchNorm2D,
+    Conv2D,
+    GELU,
+    Hardsigmoid,
+    Hardswish,
+    ReLU,
+)
+from paddle.regularizer import L2Decay
+
+
+NET_CONFIG_DET = {
+    "tiny": {
+        # stem(mid=16, out=32)  channels: 32 → 48 → 64 → 160
+        "stem": (16, 32),
+        "blocks_s1": [[3, 32, 32, 1, True], [3, 32, 32, 1, False]],
+        "blocks_s2": [
+            [3, 32, 48, 2, False],
+            [3, 48, 48, 1, True],
+            [3, 48, 48, 1, False],
+        ],
+        "blocks_s3": [
+            [3, 48, 64, 2, False],
+            [3, 64, 64, 1, True],
+            [3, 64, 64, 1, False],
+            [3, 64, 64, 1, True],
+            [3, 64, 64, 1, False],
+        ],
+        "blocks_s4": [
+            [3, 64, 160, 2, False],
+            [3, 160, 160, 1, True],
+            [3, 160, 160, 1, False],
+        ],
+    },
+    "small": {
+        # stem(mid=24, out=48)  channels: 48 → 96 → 192 → 384
+        "stem": (24, 48),
+        "blocks_s1": [[3, 48, 48, 1, True], [3, 48, 48, 1, False]],
+        "blocks_s2": [
+            [3, 48, 96, 2, False],
+            [3, 96, 96, 1, True],
+            [3, 96, 96, 1, False],
+        ],
+        "blocks_s3": [
+            [3, 96, 192, 2, False],
+            [3, 192, 192, 1, True],
+            [3, 192, 192, 1, False],
+            [3, 192, 192, 1, True],
+            [3, 192, 192, 1, False],
+        ],
+        "blocks_s4": [
+            [3, 192, 384, 2, False],
+            [3, 384, 384, 1, True],
+            [3, 384, 384, 1, False],
+        ],
+    },
+    "medium": {
+        # stem(mid=64, out=128)  channels: 128 → 256 → 512 → 896
+        "stem": (64, 128),
+        "blocks_s1": [[3, 128, 128, 1, True], [3, 128, 128, 1, False]],
+        "blocks_s2": [
+            [3, 128, 256, 2, False],
+            [3, 256, 256, 1, True],
+            [3, 256, 256, 1, False],
+        ],
+        "blocks_s3": [
+            [3, 256, 512, 2, False],
+            [3, 512, 512, 1, True],
+            [3, 512, 512, 1, False],
+            [3, 512, 512, 1, True],
+            [3, 512, 512, 1, False],
+        ],
+        "blocks_s4": [
+            [3, 512, 896, 2, False],
+            [3, 896, 896, 1, True],
+            [3, 896, 896, 1, False],
+        ],
+    },
+}
+
+
+NET_CONFIG_REC = {
+    "tiny": {
+        # stem: simple (2×Conv2D_BN+GELU, mid=24, out=48)  channels: 48 → 96 → 160
+        "stem": (24, 48),
+        "stem_type": "simple",
+        "blocks2": [[3, 48, 48, 1, True]],
+        "blocks3": [[3, 48, 48, 1, False]],
+        "blocks4": [
+            [3, 48, 96, (2, 1), False],
+            [3, 96, 96, 1, True],
+            [3, 96, 96, 1, False],
+        ],
+        "blocks5": [
+            [3, 96, 160, (2, 1), False],
+            [3, 160, 160, 1, True],
+            [3, 160, 160, 1, False],
+            [3, 160, 160, 1, False],
+        ],
+        "blocks6": [],
+    },
+    "small": {
+        # stem: branch StemBlock (mid=48, out=96)  channels: 96 → 192 → 384
+        "stem": (48, 96),
+        "stem_type": "branch",
+        "blocks2": [[3, 96, 96, 1, True]],
+        "blocks3": [[3, 96, 96, 1, False], [3, 96, 96, 1, False]],
+        "blocks4": [
+            [3, 96, 192, (2, 1), False],
+            [3, 192, 192, 1, True],
+            [3, 192, 192, 1, False],
+            [3, 192, 192, 1, True],
+            [3, 192, 192, 1, False],
+            [3, 192, 192, 1, True],
+            [3, 192, 192, 1, False],
+        ],
+        "blocks5": [
+            [3, 192, 384, (2, 1), False],
+            [3, 384, 384, 1, True],
+            [3, 384, 384, 1, False],
+        ],
+        "blocks6": [],
+    },
+    "medium": {
+        # stem: branch StemBlock (mid=64, out=128)  channels: 128 → 256 → 512 → 768
+        "stem": (64, 128),
+        "stem_type": "branch",
+        "blocks2": [[3, 128, 128, 1, True]],
+        "blocks3": [
+            [3, 128, 256, 1, False],
+            [3, 256, 256, 1, False],
+            [3, 256, 256, 1, True],
+        ],
+        "blocks4": [
+            [3, 256, 512, (2, 1), False],
+            [3, 512, 512, 1, True],
+            [3, 512, 512, 1, False],
+            [3, 512, 512, 1, True],
+            [3, 512, 512, 1, False],
+            [3, 512, 512, 1, True],
+            [3, 512, 512, 1, False],
+        ],
+        "blocks5": [
+            [3, 512, 768, (2, 1), False],
+            [3, 768, 768, 1, True],
+            [3, 768, 768, 1, False],
+        ],
+        "blocks6": [],
+    },
+}
+
+
+class Conv2D_BN(nn.Sequential):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size=1,
+        stride=1,
+        padding=0,
+        groups=1,
+        bn_weight_init=1.0,
+    ):
+        super().__init__()
+        self.add_sublayer(
+            "conv",
+            Conv2D(
+                in_channels,
+                out_channels,
+                kernel_size,
+                stride,
+                padding,
+                groups=groups,
+                bias_attr=False,
+            ),
+        )
+        bn = BatchNorm2D(out_channels)
+        Constant(1.0 if bn_weight_init == 1.0 else 0.0)(bn.weight)
+        Constant(0.0)(bn.bias)
+        self.add_sublayer("bn", bn)
+
+    @paddle.no_grad()
+    def fuse(self):
+        c, bn = self.conv, self.bn
+        w = bn.weight / (bn._variance + bn._epsilon) ** 0.5
+        w = c.weight * w[:, None, None, None]
+        b = bn.bias - bn._mean * bn.weight / (bn._variance + bn._epsilon) ** 0.5
+        m = Conv2D(
+            w.shape[1] * c._groups,
+            w.shape[0],
+            w.shape[2:],
+            stride=c._stride,
+            padding=c._padding,
+            groups=c._groups,
+        )
+        m.weight.set_value(w)
+        m.bias.set_value(b)
+        return m
+
+
+class ConvBNAct(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size=3,
+        stride=1,
+        padding=1,
+        groups=1,
+        use_act=True,
+        lr_mult=1.0,
+    ):
+        super().__init__()
+        self.use_act = use_act
+        self.is_repped = False
+        self.conv = Conv2D(
+            in_channels,
+            out_channels,
+            kernel_size,
+            stride,
+            padding=padding if isinstance(padding, str) else (kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(learning_rate=lr_mult),
+            bias_attr=False,
+        )
+        self.bn = BatchNorm2D(
+            out_channels,
+            weight_attr=ParamAttr(regularizer=L2Decay(0.0), learning_rate=lr_mult),
+            bias_attr=ParamAttr(regularizer=L2Decay(0.0), learning_rate=lr_mult),
+        )
+        if self.use_act:
+            self.act = ReLU()
+
+    def forward(self, x):
+        x = self.conv(x)
+        if not self.is_repped:
+            x = self.bn(x)
+        if self.use_act:
+            x = self.act(x)
+        return x
+
+    @paddle.no_grad()
+    def rep(self):
+        if self.is_repped:
+            return
+        c, bn = self.conv, self.bn
+        w = bn.weight / (bn._variance + bn._epsilon) ** 0.5
+        fused_w = c.weight * w[:, None, None, None]
+        fused_b = bn.bias - bn._mean * bn.weight / (bn._variance + bn._epsilon) ** 0.5
+        m = Conv2D(
+            c._in_channels,
+            c._out_channels,
+            c._kernel_size,
+            stride=c._stride,
+            padding=c._padding,
+            dilation=c._dilation,
+            groups=c._groups,
+        )
+        m.weight.set_value(fused_w)
+        m.bias.set_value(fused_b)
+        self.conv = m
+        del self.bn
+        self.is_repped = True
+
+
+class StemBlock(nn.Layer):
+    """Multi-branch stem with total stride 4 (stem1 stride=2 + stem3 stride=2)."""
+
+    def __init__(self, in_channels=3, mid_channels=48, out_channels=96, lr_mult=1.0):
+        super().__init__()
+        self.is_repped = False
+        self.stem1 = ConvBNAct(
+            in_channels, mid_channels, 3, 2, use_act=True, lr_mult=lr_mult
+        )
+        self.stem2a = ConvBNAct(
+            mid_channels,
+            mid_channels // 2,
+            2,
+            1,
+            padding="SAME",
+            use_act=True,
+            lr_mult=lr_mult,
+        )
+        self.stem2b = ConvBNAct(
+            mid_channels // 2,
+            mid_channels,
+            2,
+            1,
+            padding="SAME",
+            use_act=True,
+            lr_mult=lr_mult,
+        )
+        self.stem3 = ConvBNAct(
+            mid_channels * 2, mid_channels, 3, 2, use_act=True, lr_mult=lr_mult
+        )
+        self.stem4 = ConvBNAct(
+            mid_channels, out_channels, 1, 1, use_act=True, lr_mult=lr_mult
+        )
+        self.pool = nn.MaxPool2D(kernel_size=2, stride=1, padding="SAME")
+
+    def forward(self, x):
+        x = self.stem1(x)
+        x2 = self.stem2b(self.stem2a(x))
+        x1 = self.pool(x)
+        x = self.stem4(self.stem3(paddle.concat([x1, x2], axis=1)))
+        return x
+
+    def rep(self, fuse_lab=None):
+        if self.is_repped:
+            return
+        for attr in ("stem1", "stem2a", "stem2b", "stem3", "stem4"):
+            getattr(self, attr).rep()
+        self.is_repped = True
+
+
+class SELayer(nn.Layer):
+    def __init__(self, channel, reduction=4, lr_mult=1.0):
+        super().__init__()
+        self.conv1 = Conv2D(
+            channel,
+            channel // reduction,
+            1,
+            weight_attr=ParamAttr(learning_rate=lr_mult),
+            bias_attr=ParamAttr(learning_rate=lr_mult),
+        )
+        self.relu = ReLU()
+        self.conv2 = Conv2D(
+            channel // reduction,
+            channel,
+            1,
+            weight_attr=ParamAttr(learning_rate=lr_mult),
+            bias_attr=ParamAttr(learning_rate=lr_mult),
+        )
+        self.hardsigmoid = Hardsigmoid()
+
+    def forward(self, x):
+        identity = x
+        x = x.mean(axis=[2, 3], keepdim=True)
+        x = self.relu(self.conv1(x))
+        x = self.hardsigmoid(self.conv2(x))
+        return paddle.multiply(x=identity, y=x)
+
+
+class RepDWConv(nn.Layer):
+    """Reparameterizable depthwise convolution.
+
+    Training: 3-branch (3×3 DW + 1×1 DW + identity BN)
+    Inference: fused into a single 3×3 DW Conv
+    """
+
+    def __init__(self, channels, kernel_size=3):
+        super().__init__()
+        self.channels = channels
+        self.kernel_size = kernel_size
+        padding = (kernel_size - 1) // 2
+
+        self.conv = Conv2D_BN(
+            channels, channels, kernel_size, 1, padding, groups=channels
+        )
+        self.conv1 = Conv2D(
+            channels, channels, 1, 1, 0, groups=channels, bias_attr=False
+        )
+        self.bn = BatchNorm2D(channels)
+        Constant(1.0)(self.bn.weight)
+        Constant(0.0)(self.bn.bias)
+
+        self.is_repped = False
+        self.reparam_conv = None
+
+    def forward(self, x):
+        if self.is_repped:
+            return self.reparam_conv(x)
+        return self.bn(self.conv(x) + self.conv1(x) + x)
+
+    def rep(self, fuse_lab=None):
+        if self.is_repped:
+            return
+        fused = self._fuse_conv()
+        padding = (self.kernel_size - 1) // 2
+        self.reparam_conv = Conv2D(
+            self.channels,
+            self.channels,
+            self.kernel_size,
+            1,
+            padding,
+            groups=self.channels,
+        )
+        self.reparam_conv.weight.set_value(fused.weight)
+        self.reparam_conv.bias.set_value(fused.bias)
+        self.__delattr__("conv")
+        self.__delattr__("conv1")
+        self.__delattr__("bn")
+        self.is_repped = True
+
+    @paddle.no_grad()
+    def _fuse_conv(self):
+        conv = self.conv.fuse()
+        pad_size = self.kernel_size // 2
+        conv1_w = F.pad(self.conv1.weight, [pad_size, pad_size, pad_size, pad_size])
+        identity = F.pad(
+            paddle.ones([self.conv1.weight.shape[0], self.conv1.weight.shape[1], 1, 1]),
+            [pad_size, pad_size, pad_size, pad_size],
+        )
+        w = conv.weight + conv1_w + identity
+        conv.weight.set_value(w)
+        bn = self.bn
+        scale = bn.weight / (bn._variance + bn._epsilon) ** 0.5
+        conv.weight.set_value(conv.weight * scale[:, None, None, None])
+        conv.bias.set_value(bn.bias + (conv.bias - bn._mean) * scale)
+        return conv
+
+    def fuse(self):
+        return self._fuse_conv()
+
+
+class LCNetV4Block(nn.Layer):
+    """LCNetV4 block for detection and recognition.
+
+    Token mixer: RepDWConv when stride=1 and in==out, else plain Conv2D_BN DW conv.
+    Channel mixer: expand → act → compress (+ residual when stride=1 and in==out)
+    rep() fuses all Conv2D_BN layers (mathematically exact, no accuracy change).
+    """
+
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        stride,
+        dw_size,
+        use_se=False,
+        lr_mult=1.0,
+        expand_ratio=2,
+        act_type="gelu",
+    ):
+        super().__init__()
+        self.is_repped = False
+        self.has_residual = in_channels == out_channels and stride == 1
+        self.use_rep_dw = stride == 1 and in_channels == out_channels
+
+        self.token_mixer = nn.Sequential()
+        if self.use_rep_dw:
+            self.token_mixer.add_sublayer("rep_dw", RepDWConv(in_channels, dw_size))
+        else:
+            padding = (dw_size - 1) // 2
+            self.token_mixer.add_sublayer(
+                "dw_conv",
+                Conv2D_BN(
+                    in_channels,
+                    in_channels,
+                    dw_size,
+                    stride,
+                    padding,
+                    groups=in_channels,
+                ),
+            )
+        if use_se:
+            self.token_mixer.add_sublayer("se", SELayer(in_channels, lr_mult=lr_mult))
+
+        hidden_channels = int(in_channels * expand_ratio)
+        compress_bn_init = 0.0 if self.has_residual else 1.0
+        self.channel_mixer = nn.Sequential()
+        self.channel_mixer.add_sublayer(
+            "expand", Conv2D_BN(in_channels, hidden_channels, 1, 1, 0)
+        )
+        if act_type == "gelu":
+            self.channel_mixer.add_sublayer("act", GELU())
+        elif act_type == "hswish":
+            self.channel_mixer.add_sublayer("act", Hardswish())
+        elif act_type == "relu":
+            self.channel_mixer.add_sublayer("act", ReLU())
+        self.channel_mixer.add_sublayer(
+            "compress",
+            Conv2D_BN(
+                hidden_channels, out_channels, 1, 1, 0, bn_weight_init=compress_bn_init
+            ),
+        )
+
+    def forward(self, x):
+        x = self.token_mixer(x)
+        if self.has_residual:
+            return x + self.channel_mixer(x)
+        return self.channel_mixer(x)
+
+    def rep(self, fuse_lab=None):
+        if self.is_repped:
+            return
+        if self.use_rep_dw:
+            self.token_mixer.rep_dw.rep(fuse_lab=fuse_lab)
+        else:
+            self.token_mixer.dw_conv = self.token_mixer.dw_conv.fuse()
+        for name in ("expand", "compress"):
+            m = getattr(self.channel_mixer, name, None)
+            if isinstance(m, Conv2D_BN):
+                setattr(self.channel_mixer, name, m.fuse())
+        self.is_repped = True
+
+
+class PPLCNetV4(nn.Layer):
+    """Unified PPLCNetV4 backbone for text detection and recognition.
+
+    Detection (det=True):
+        model_size in {'tiny', 'small', 'medium'} — see NET_CONFIG_DET.
+        Returns 4-level feature list [s1_out, s2_out, s3_out, s4_out].
+
+    Recognition (det=False):
+        model_size in {'tiny', 'small', 'medium'} — see NET_CONFIG_REC.
+        Returns pooled feature tensor [B, C, 1, W].
+    """
+
+    def __init__(
+        self,
+        det=False,
+        model_size="small",
+        in_channels=3,
+        lr_mult_list=[1.0, 1.0, 1.0, 1.0, 1.0, 1.0],
+        **kwargs,
+    ):
+        super().__init__()
+        self.det = det
+        self.is_repped = False
+
+        if det:
+            assert (
+                model_size in NET_CONFIG_DET
+            ), "det model_size must be one of {} but got '{}'".format(
+                list(NET_CONFIG_DET.keys()), model_size
+            )
+            cfg = NET_CONFIG_DET[model_size]
+            stem_mid, stem_out = cfg["stem"]
+            self.stem = StemBlock(in_channels, stem_mid, stem_out)
+
+            def make_stage(key):
+                return nn.Sequential(
+                    *[
+                        LCNetV4Block(in_c, out_c, s, k, se, expand_ratio=2)
+                        for k, in_c, out_c, s, se in cfg[key]
+                    ]
+                )
+
+            self.blocks_s1 = make_stage("blocks_s1")
+            self.blocks_s2 = make_stage("blocks_s2")
+            self.blocks_s3 = make_stage("blocks_s3")
+            self.blocks_s4 = make_stage("blocks_s4")
+            self.out_channels = [
+                cfg["blocks_s1"][-1][2],
+                cfg["blocks_s2"][-1][2],
+                cfg["blocks_s3"][-1][2],
+                cfg["blocks_s4"][-1][2],
+            ]
+
+        else:
+            assert isinstance(lr_mult_list, (list, tuple)) and len(lr_mult_list) == 6
+            assert (
+                model_size in NET_CONFIG_REC
+            ), "rec model_size must be one of {} but got '{}'".format(
+                list(NET_CONFIG_REC.keys()), model_size
+            )
+            self.lr_mult_list = lr_mult_list
+            cfg = NET_CONFIG_REC[model_size]
+            stem_mid, stem_out = cfg["stem"]
+            if cfg["stem_type"] == "branch":
+                self.conv1 = StemBlock(3, stem_mid, stem_out, lr_mult=lr_mult_list[0])
+            else:
+                self.conv1 = nn.Sequential(
+                    Conv2D_BN(3, stem_mid, 3, 2, 1),
+                    GELU(),
+                    Conv2D_BN(stem_mid, stem_out, 3, 2, 1),
+                )
+
+            def make_stage(stage_name, lr_idx):
+                return nn.Sequential(
+                    *[
+                        LCNetV4Block(
+                            in_c,
+                            out_c,
+                            s,
+                            k,
+                            se,
+                            lr_mult=lr_mult_list[lr_idx],
+                            expand_ratio=2,
+                        )
+                        for k, in_c, out_c, s, se in cfg.get(stage_name, [])
+                    ]
+                )
+
+            self.blocks2 = make_stage("blocks2", 1)
+            self.blocks3 = make_stage("blocks3", 2)
+            self.blocks4 = make_stage("blocks4", 3)
+            self.blocks5 = make_stage("blocks5", 4)
+            self.blocks6 = make_stage("blocks6", 5)
+
+            for sname in reversed(
+                ["blocks2", "blocks3", "blocks4", "blocks5", "blocks6"]
+            ):
+                if cfg.get(sname):
+                    self.out_channels = cfg[sname][-1][2]
+                    break
+
+    def forward(self, x):
+        if self.det:
+            x = self.stem(x)
+            o1 = self.blocks_s1(x)
+            o2 = self.blocks_s2(o1)
+            o3 = self.blocks_s3(o2)
+            o4 = self.blocks_s4(o3)
+            return [o1, o2, o3, o4]
+        else:
+            x = self.conv1(x)
+            x = self.blocks2(x)
+            x = self.blocks3(x)
+            x = self.blocks4(x)
+            x = self.blocks5(x)
+            x = self.blocks6(x)
+            if self.training:
+                x = F.adaptive_avg_pool2d(x, [1, 40])
+            else:
+                assert x.shape[2] >= 3, f"Feature height {x.shape[2]} < pool kernel 3."
+                x = F.avg_pool2d(x, [3, 2])
+            return x
+
+    def rep(self, fuse_lab=None):
+        if self.is_repped:
+            return
+        if self.det:
+            self.stem.rep()
+            for stage in [
+                self.blocks_s1,
+                self.blocks_s2,
+                self.blocks_s3,
+                self.blocks_s4,
+            ]:
+                for block in stage:
+                    block.rep(fuse_lab=fuse_lab)
+        else:
+            if hasattr(self.conv1, "rep"):
+                self.conv1.rep()
+            for stage in [
+                self.blocks2,
+                self.blocks3,
+                self.blocks4,
+                self.blocks5,
+                self.blocks6,
+            ]:
+                for block in stage:
+                    block.rep()
+        self.is_repped = True
diff --git a/ppocr/modeling/backbones/rec_micronet.py b/ppocr/modeling/backbones/rec_micronet.py
new file mode 100644
index 0000000..f554a7d
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_micronet.py
@@ -0,0 +1,605 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/liyunsheng13/micronet/blob/main/backbone/micronet.py
+https://github.com/liyunsheng13/micronet/blob/main/backbone/activation.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+import paddle.nn as nn
+
+from ppocr.modeling.backbones.det_mobilenet_v3 import make_divisible
+
+M0_cfgs = [
+    # s, n, c, ks, c1, c2, g1, g2, c3, g3, g4, y1, y2, y3, r
+    [2, 1, 8, 3, 2, 2, 0, 4, 8, 2, 2, 2, 0, 1, 1],
+    [2, 1, 12, 3, 2, 2, 0, 8, 12, 4, 4, 2, 2, 1, 1],
+    [2, 1, 16, 5, 2, 2, 0, 12, 16, 4, 4, 2, 2, 1, 1],
+    [1, 1, 32, 5, 1, 4, 4, 4, 32, 4, 4, 2, 2, 1, 1],
+    [2, 1, 64, 5, 1, 4, 8, 8, 64, 8, 8, 2, 2, 1, 1],
+    [1, 1, 96, 3, 1, 4, 8, 8, 96, 8, 8, 2, 2, 1, 2],
+    [1, 1, 384, 3, 1, 4, 12, 12, 0, 0, 0, 2, 2, 1, 2],
+]
+M1_cfgs = [
+    # s, n, c, ks, c1, c2, g1, g2, c3, g3, g4
+    [2, 1, 8, 3, 2, 2, 0, 6, 8, 2, 2, 2, 0, 1, 1],
+    [2, 1, 16, 3, 2, 2, 0, 8, 16, 4, 4, 2, 2, 1, 1],
+    [2, 1, 16, 5, 2, 2, 0, 16, 16, 4, 4, 2, 2, 1, 1],
+    [1, 1, 32, 5, 1, 6, 4, 4, 32, 4, 4, 2, 2, 1, 1],
+    [2, 1, 64, 5, 1, 6, 8, 8, 64, 8, 8, 2, 2, 1, 1],
+    [1, 1, 96, 3, 1, 6, 8, 8, 96, 8, 8, 2, 2, 1, 2],
+    [1, 1, 576, 3, 1, 6, 12, 12, 0, 0, 0, 2, 2, 1, 2],
+]
+M2_cfgs = [
+    # s, n, c, ks, c1, c2, g1, g2, c3, g3, g4
+    [2, 1, 12, 3, 2, 2, 0, 8, 12, 4, 4, 2, 0, 1, 1],
+    [2, 1, 16, 3, 2, 2, 0, 12, 16, 4, 4, 2, 2, 1, 1],
+    [1, 1, 24, 3, 2, 2, 0, 16, 24, 4, 4, 2, 2, 1, 1],
+    [2, 1, 32, 5, 1, 6, 6, 6, 32, 4, 4, 2, 2, 1, 1],
+    [1, 1, 32, 5, 1, 6, 8, 8, 32, 4, 4, 2, 2, 1, 2],
+    [1, 1, 64, 5, 1, 6, 8, 8, 64, 8, 8, 2, 2, 1, 2],
+    [2, 1, 96, 5, 1, 6, 8, 8, 96, 8, 8, 2, 2, 1, 2],
+    [1, 1, 128, 3, 1, 6, 12, 12, 128, 8, 8, 2, 2, 1, 2],
+    [1, 1, 768, 3, 1, 6, 16, 16, 0, 0, 0, 2, 2, 1, 2],
+]
+M3_cfgs = [
+    # s, n, c, ks, c1, c2, g1, g2, c3, g3, g4
+    [2, 1, 16, 3, 2, 2, 0, 12, 16, 4, 4, 0, 2, 0, 1],
+    [2, 1, 24, 3, 2, 2, 0, 16, 24, 4, 4, 0, 2, 0, 1],
+    [1, 1, 24, 3, 2, 2, 0, 24, 24, 4, 4, 0, 2, 0, 1],
+    [2, 1, 32, 5, 1, 6, 6, 6, 32, 4, 4, 0, 2, 0, 1],
+    [1, 1, 32, 5, 1, 6, 8, 8, 32, 4, 4, 0, 2, 0, 2],
+    [1, 1, 64, 5, 1, 6, 8, 8, 48, 8, 8, 0, 2, 0, 2],
+    [1, 1, 80, 5, 1, 6, 8, 8, 80, 8, 8, 0, 2, 0, 2],
+    [1, 1, 80, 5, 1, 6, 10, 10, 80, 8, 8, 0, 2, 0, 2],
+    [1, 1, 120, 5, 1, 6, 10, 10, 120, 10, 10, 0, 2, 0, 2],
+    [1, 1, 120, 5, 1, 6, 12, 12, 120, 10, 10, 0, 2, 0, 2],
+    [1, 1, 144, 3, 1, 6, 12, 12, 144, 12, 12, 0, 2, 0, 2],
+    [1, 1, 432, 3, 1, 3, 12, 12, 0, 0, 0, 0, 2, 0, 2],
+]
+
+
+def get_micronet_config(mode):
+    return eval(mode + "_cfgs")
+
+
+class MaxGroupPooling(nn.Layer):
+    def __init__(self, channel_per_group=2):
+        super(MaxGroupPooling, self).__init__()
+        self.channel_per_group = channel_per_group
+
+    def forward(self, x):
+        if self.channel_per_group == 1:
+            return x
+        # max op
+        b, c, h, w = x.shape
+
+        # reshape
+        y = paddle.reshape(x, [b, c // self.channel_per_group, -1, h, w])
+        out = paddle.max(y, axis=2)
+        return out
+
+
+class SpatialSepConvSF(nn.Layer):
+    def __init__(self, inp, oups, kernel_size, stride):
+        super(SpatialSepConvSF, self).__init__()
+
+        oup1, oup2 = oups
+        self.conv = nn.Sequential(
+            nn.Conv2D(
+                inp,
+                oup1,
+                (kernel_size, 1),
+                (stride, 1),
+                (kernel_size // 2, 0),
+                bias_attr=False,
+                groups=1,
+            ),
+            nn.BatchNorm2D(oup1),
+            nn.Conv2D(
+                oup1,
+                oup1 * oup2,
+                (1, kernel_size),
+                (1, stride),
+                (0, kernel_size // 2),
+                bias_attr=False,
+                groups=oup1,
+            ),
+            nn.BatchNorm2D(oup1 * oup2),
+            ChannelShuffle(oup1),
+        )
+
+    def forward(self, x):
+        out = self.conv(x)
+        return out
+
+
+class ChannelShuffle(nn.Layer):
+    def __init__(self, groups):
+        super(ChannelShuffle, self).__init__()
+        self.groups = groups
+
+    def forward(self, x):
+        b, c, h, w = x.shape
+
+        channels_per_group = c // self.groups
+
+        # reshape
+        x = paddle.reshape(x, [b, self.groups, channels_per_group, h, w])
+
+        x = paddle.transpose(x, (0, 2, 1, 3, 4))
+        out = paddle.reshape(x, [b, -1, h, w])
+
+        return out
+
+
+class StemLayer(nn.Layer):
+    def __init__(self, inp, oup, stride, groups=(4, 4)):
+        super(StemLayer, self).__init__()
+
+        g1, g2 = groups
+        self.stem = nn.Sequential(
+            SpatialSepConvSF(inp, groups, 3, stride),
+            MaxGroupPooling(2) if g1 * g2 == 2 * oup else nn.ReLU6(),
+        )
+
+    def forward(self, x):
+        out = self.stem(x)
+        return out
+
+
+class DepthSpatialSepConv(nn.Layer):
+    def __init__(self, inp, expand, kernel_size, stride):
+        super(DepthSpatialSepConv, self).__init__()
+
+        exp1, exp2 = expand
+
+        hidden_dim = inp * exp1
+        oup = inp * exp1 * exp2
+
+        self.conv = nn.Sequential(
+            nn.Conv2D(
+                inp,
+                inp * exp1,
+                (kernel_size, 1),
+                (stride, 1),
+                (kernel_size // 2, 0),
+                bias_attr=False,
+                groups=inp,
+            ),
+            nn.BatchNorm2D(inp * exp1),
+            nn.Conv2D(
+                hidden_dim,
+                oup,
+                (1, kernel_size),
+                1,
+                (0, kernel_size // 2),
+                bias_attr=False,
+                groups=hidden_dim,
+            ),
+            nn.BatchNorm2D(oup),
+        )
+
+    def forward(self, x):
+        x = self.conv(x)
+        return x
+
+
+class GroupConv(nn.Layer):
+    def __init__(self, inp, oup, groups=2):
+        super(GroupConv, self).__init__()
+        self.inp = inp
+        self.oup = oup
+        self.groups = groups
+        self.conv = nn.Sequential(
+            nn.Conv2D(inp, oup, 1, 1, 0, bias_attr=False, groups=self.groups[0]),
+            nn.BatchNorm2D(oup),
+        )
+
+    def forward(self, x):
+        x = self.conv(x)
+        return x
+
+
+class DepthConv(nn.Layer):
+    def __init__(self, inp, oup, kernel_size, stride):
+        super(DepthConv, self).__init__()
+        self.conv = nn.Sequential(
+            nn.Conv2D(
+                inp,
+                oup,
+                kernel_size,
+                stride,
+                kernel_size // 2,
+                bias_attr=False,
+                groups=inp,
+            ),
+            nn.BatchNorm2D(oup),
+        )
+
+    def forward(self, x):
+        out = self.conv(x)
+        return out
+
+
+class DYShiftMax(nn.Layer):
+    def __init__(
+        self,
+        inp,
+        oup,
+        reduction=4,
+        act_max=1.0,
+        act_relu=True,
+        init_a=[0.0, 0.0],
+        init_b=[0.0, 0.0],
+        relu_before_pool=False,
+        g=None,
+        expansion=False,
+    ):
+        super(DYShiftMax, self).__init__()
+        self.oup = oup
+        self.act_max = act_max * 2
+        self.act_relu = act_relu
+        self.avg_pool = nn.Sequential(
+            nn.ReLU() if relu_before_pool == True else nn.Sequential(),
+            nn.AdaptiveAvgPool2D(1),
+        )
+
+        self.exp = 4 if act_relu else 2
+        self.init_a = init_a
+        self.init_b = init_b
+
+        # determine squeeze
+        squeeze = make_divisible(inp // reduction, 4)
+        if squeeze < 4:
+            squeeze = 4
+
+        self.fc = nn.Sequential(
+            nn.Linear(inp, squeeze),
+            nn.ReLU(),
+            nn.Linear(squeeze, oup * self.exp),
+            nn.Hardsigmoid(),
+        )
+
+        if g is None:
+            g = 1
+        self.g = g[1]
+        if self.g != 1 and expansion:
+            self.g = inp // self.g
+
+        self.gc = inp // self.g
+        index = paddle.to_tensor([range(inp)])
+        index = paddle.reshape(index, [1, inp, 1, 1])
+        index = paddle.reshape(index, [1, self.g, self.gc, 1, 1])
+        indexgs = paddle.split(index, [1, self.g - 1], axis=1)
+        indexgs = paddle.concat((indexgs[1], indexgs[0]), axis=1)
+        indexes = paddle.split(indexgs, [1, self.gc - 1], axis=2)
+        indexes = paddle.concat((indexes[1], indexes[0]), axis=2)
+        self.index = paddle.reshape(indexes, [inp])
+        self.expansion = expansion
+
+    def forward(self, x):
+        x_in = x
+        x_out = x
+
+        b, c, _, _ = x_in.shape
+        y = self.avg_pool(x_in)
+        y = paddle.reshape(y, [b, c])
+        y = self.fc(y)
+        y = paddle.reshape(y, [b, self.oup * self.exp, 1, 1])
+        y = (y - 0.5) * self.act_max
+
+        n2, c2, h2, w2 = x_out.shape
+        x2 = paddle.to_tensor(x_out.numpy()[:, self.index.numpy(), :, :])
+
+        if self.exp == 4:
+            temp = y.shape
+            a1, b1, a2, b2 = paddle.split(y, temp[1] // self.oup, axis=1)
+
+            a1 = a1 + self.init_a[0]
+            a2 = a2 + self.init_a[1]
+
+            b1 = b1 + self.init_b[0]
+            b2 = b2 + self.init_b[1]
+
+            z1 = x_out * a1 + x2 * b1
+            z2 = x_out * a2 + x2 * b2
+
+            out = paddle.maximum(z1, z2)
+
+        elif self.exp == 2:
+            temp = y.shape
+            a1, b1 = paddle.split(y, temp[1] // self.oup, axis=1)
+            a1 = a1 + self.init_a[0]
+            b1 = b1 + self.init_b[0]
+            out = x_out * a1 + x2 * b1
+
+        return out
+
+
+class DYMicroBlock(nn.Layer):
+    def __init__(
+        self,
+        inp,
+        oup,
+        kernel_size=3,
+        stride=1,
+        ch_exp=(2, 2),
+        ch_per_group=4,
+        groups_1x1=(1, 1),
+        depthsep=True,
+        shuffle=False,
+        activation_cfg=None,
+    ):
+        super(DYMicroBlock, self).__init__()
+
+        self.identity = stride == 1 and inp == oup
+
+        y1, y2, y3 = activation_cfg["dy"]
+        act_reduction = 8 * activation_cfg["ratio"]
+        init_a = activation_cfg["init_a"]
+        init_b = activation_cfg["init_b"]
+
+        t1 = ch_exp
+        gs1 = ch_per_group
+        hidden_fft, g1, g2 = groups_1x1
+        hidden_dim2 = inp * t1[0] * t1[1]
+
+        if gs1[0] == 0:
+            self.layers = nn.Sequential(
+                DepthSpatialSepConv(inp, t1, kernel_size, stride),
+                (
+                    DYShiftMax(
+                        hidden_dim2,
+                        hidden_dim2,
+                        act_max=2.0,
+                        act_relu=True if y2 == 2 else False,
+                        init_a=init_a,
+                        reduction=act_reduction,
+                        init_b=init_b,
+                        g=gs1,
+                        expansion=False,
+                    )
+                    if y2 > 0
+                    else nn.ReLU6()
+                ),
+                ChannelShuffle(gs1[1]) if shuffle else nn.Sequential(),
+                (
+                    ChannelShuffle(hidden_dim2 // 2)
+                    if shuffle and y2 != 0
+                    else nn.Sequential()
+                ),
+                GroupConv(hidden_dim2, oup, (g1, g2)),
+                (
+                    DYShiftMax(
+                        oup,
+                        oup,
+                        act_max=2.0,
+                        act_relu=False,
+                        init_a=[1.0, 0.0],
+                        reduction=act_reduction // 2,
+                        init_b=[0.0, 0.0],
+                        g=(g1, g2),
+                        expansion=False,
+                    )
+                    if y3 > 0
+                    else nn.Sequential()
+                ),
+                ChannelShuffle(g2) if shuffle else nn.Sequential(),
+                (
+                    ChannelShuffle(oup // 2)
+                    if shuffle and oup % 2 == 0 and y3 != 0
+                    else nn.Sequential()
+                ),
+            )
+        elif g2 == 0:
+            self.layers = nn.Sequential(
+                GroupConv(inp, hidden_dim2, gs1),
+                (
+                    DYShiftMax(
+                        hidden_dim2,
+                        hidden_dim2,
+                        act_max=2.0,
+                        act_relu=False,
+                        init_a=[1.0, 0.0],
+                        reduction=act_reduction,
+                        init_b=[0.0, 0.0],
+                        g=gs1,
+                        expansion=False,
+                    )
+                    if y3 > 0
+                    else nn.Sequential()
+                ),
+            )
+        else:
+            self.layers = nn.Sequential(
+                GroupConv(inp, hidden_dim2, gs1),
+                (
+                    DYShiftMax(
+                        hidden_dim2,
+                        hidden_dim2,
+                        act_max=2.0,
+                        act_relu=True if y1 == 2 else False,
+                        init_a=init_a,
+                        reduction=act_reduction,
+                        init_b=init_b,
+                        g=gs1,
+                        expansion=False,
+                    )
+                    if y1 > 0
+                    else nn.ReLU6()
+                ),
+                ChannelShuffle(gs1[1]) if shuffle else nn.Sequential(),
+                (
+                    DepthSpatialSepConv(hidden_dim2, (1, 1), kernel_size, stride)
+                    if depthsep
+                    else DepthConv(hidden_dim2, hidden_dim2, kernel_size, stride)
+                ),
+                nn.Sequential(),
+                (
+                    DYShiftMax(
+                        hidden_dim2,
+                        hidden_dim2,
+                        act_max=2.0,
+                        act_relu=True if y2 == 2 else False,
+                        init_a=init_a,
+                        reduction=act_reduction,
+                        init_b=init_b,
+                        g=gs1,
+                        expansion=True,
+                    )
+                    if y2 > 0
+                    else nn.ReLU6()
+                ),
+                (
+                    ChannelShuffle(hidden_dim2 // 4)
+                    if shuffle and y1 != 0 and y2 != 0
+                    else (
+                        nn.Sequential()
+                        if y1 == 0 and y2 == 0
+                        else ChannelShuffle(hidden_dim2 // 2)
+                    )
+                ),
+                GroupConv(hidden_dim2, oup, (g1, g2)),
+                (
+                    DYShiftMax(
+                        oup,
+                        oup,
+                        act_max=2.0,
+                        act_relu=False,
+                        init_a=[1.0, 0.0],
+                        reduction=(
+                            act_reduction // 2 if oup < hidden_dim2 else act_reduction
+                        ),
+                        init_b=[0.0, 0.0],
+                        g=(g1, g2),
+                        expansion=False,
+                    )
+                    if y3 > 0
+                    else nn.Sequential()
+                ),
+                ChannelShuffle(g2) if shuffle else nn.Sequential(),
+                ChannelShuffle(oup // 2) if shuffle and y3 != 0 else nn.Sequential(),
+            )
+
+    def forward(self, x):
+        identity = x
+        out = self.layers(x)
+
+        if self.identity:
+            out = out + identity
+
+        return out
+
+
+class MicroNet(nn.Layer):
+    """
+    the MicroNet backbone network for recognition module.
+    Args:
+        mode(str): {'M0', 'M1', 'M2', 'M3'}
+            Four models are proposed based on four different computational costs (4M, 6M, 12M, 21M MAdds)
+            Default: 'M3'.
+    """
+
+    def __init__(self, mode="M3", **kwargs):
+        super(MicroNet, self).__init__()
+
+        self.cfgs = get_micronet_config(mode)
+
+        activation_cfg = {}
+        if mode == "M0":
+            input_channel = 4
+            stem_groups = 2, 2
+            out_ch = 384
+            activation_cfg["init_a"] = 1.0, 1.0
+            activation_cfg["init_b"] = 0.0, 0.0
+        elif mode == "M1":
+            input_channel = 6
+            stem_groups = 3, 2
+            out_ch = 576
+            activation_cfg["init_a"] = 1.0, 1.0
+            activation_cfg["init_b"] = 0.0, 0.0
+        elif mode == "M2":
+            input_channel = 8
+            stem_groups = 4, 2
+            out_ch = 768
+            activation_cfg["init_a"] = 1.0, 1.0
+            activation_cfg["init_b"] = 0.0, 0.0
+        elif mode == "M3":
+            input_channel = 12
+            stem_groups = 4, 3
+            out_ch = 432
+            activation_cfg["init_a"] = 1.0, 0.5
+            activation_cfg["init_b"] = 0.0, 0.5
+        else:
+            raise NotImplementedError("mode[" + mode + "_model] is not implemented!")
+
+        layers = [StemLayer(3, input_channel, stride=2, groups=stem_groups)]
+
+        for idx, val in enumerate(self.cfgs):
+            s, n, c, ks, c1, c2, g1, g2, c3, g3, g4, y1, y2, y3, r = val
+
+            t1 = (c1, c2)
+            gs1 = (g1, g2)
+            gs2 = (c3, g3, g4)
+            activation_cfg["dy"] = [y1, y2, y3]
+            activation_cfg["ratio"] = r
+
+            output_channel = c
+            layers.append(
+                DYMicroBlock(
+                    input_channel,
+                    output_channel,
+                    kernel_size=ks,
+                    stride=s,
+                    ch_exp=t1,
+                    ch_per_group=gs1,
+                    groups_1x1=gs2,
+                    depthsep=True,
+                    shuffle=True,
+                    activation_cfg=activation_cfg,
+                )
+            )
+            input_channel = output_channel
+            for i in range(1, n):
+                layers.append(
+                    DYMicroBlock(
+                        input_channel,
+                        output_channel,
+                        kernel_size=ks,
+                        stride=1,
+                        ch_exp=t1,
+                        ch_per_group=gs1,
+                        groups_1x1=gs2,
+                        depthsep=True,
+                        shuffle=True,
+                        activation_cfg=activation_cfg,
+                    )
+                )
+                input_channel = output_channel
+        self.features = nn.Sequential(*layers)
+
+        self.pool = nn.MaxPool2D(kernel_size=2, stride=2, padding=0)
+
+        self.out_channels = make_divisible(out_ch)
+
+    def forward(self, x):
+        x = self.features(x)
+        x = self.pool(x)
+        return x
diff --git a/ppocr/modeling/backbones/rec_mobilenet_v3.py b/ppocr/modeling/backbones/rec_mobilenet_v3.py
new file mode 100644
index 0000000..00ee5a3
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_mobilenet_v3.py
@@ -0,0 +1,156 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddle import nn
+
+from ppocr.modeling.backbones.det_mobilenet_v3 import (
+    ResidualUnit,
+    ConvBNLayer,
+    make_divisible,
+)
+
+__all__ = ["MobileNetV3"]
+
+
+class MobileNetV3(nn.Layer):
+    def __init__(
+        self,
+        in_channels=3,
+        model_name="small",
+        scale=0.5,
+        large_stride=None,
+        small_stride=None,
+        disable_se=False,
+        **kwargs,
+    ):
+        super(MobileNetV3, self).__init__()
+        self.disable_se = disable_se
+        if small_stride is None:
+            small_stride = [2, 2, 2, 2]
+        if large_stride is None:
+            large_stride = [1, 2, 2, 2]
+
+        assert isinstance(
+            large_stride, list
+        ), "large_stride type must " "be list but got {}".format(type(large_stride))
+        assert isinstance(
+            small_stride, list
+        ), "small_stride type must " "be list but got {}".format(type(small_stride))
+        assert (
+            len(large_stride) == 4
+        ), "large_stride length must be " "4 but got {}".format(len(large_stride))
+        assert (
+            len(small_stride) == 4
+        ), "small_stride length must be " "4 but got {}".format(len(small_stride))
+
+        if model_name == "large":
+            cfg = [
+                # k, exp, c,  se,     nl,  s,
+                [3, 16, 16, False, "relu", large_stride[0]],
+                [3, 64, 24, False, "relu", (large_stride[1], 1)],
+                [3, 72, 24, False, "relu", 1],
+                [5, 72, 40, True, "relu", (large_stride[2], 1)],
+                [5, 120, 40, True, "relu", 1],
+                [5, 120, 40, True, "relu", 1],
+                [3, 240, 80, False, "hardswish", 1],
+                [3, 200, 80, False, "hardswish", 1],
+                [3, 184, 80, False, "hardswish", 1],
+                [3, 184, 80, False, "hardswish", 1],
+                [3, 480, 112, True, "hardswish", 1],
+                [3, 672, 112, True, "hardswish", 1],
+                [5, 672, 160, True, "hardswish", (large_stride[3], 1)],
+                [5, 960, 160, True, "hardswish", 1],
+                [5, 960, 160, True, "hardswish", 1],
+            ]
+            cls_ch_squeeze = 960
+        elif model_name == "small":
+            cfg = [
+                # k, exp, c,  se,     nl,  s,
+                [3, 16, 16, True, "relu", (small_stride[0], 1)],
+                [3, 72, 24, False, "relu", (small_stride[1], 1)],
+                [3, 88, 24, False, "relu", 1],
+                [5, 96, 40, True, "hardswish", (small_stride[2], 1)],
+                [5, 240, 40, True, "hardswish", 1],
+                [5, 240, 40, True, "hardswish", 1],
+                [5, 120, 48, True, "hardswish", 1],
+                [5, 144, 48, True, "hardswish", 1],
+                [5, 288, 96, True, "hardswish", (small_stride[3], 1)],
+                [5, 576, 96, True, "hardswish", 1],
+                [5, 576, 96, True, "hardswish", 1],
+            ]
+            cls_ch_squeeze = 576
+        else:
+            raise NotImplementedError(
+                "mode[" + model_name + "_model] is not implemented!"
+            )
+
+        supported_scale = [0.35, 0.5, 0.75, 1.0, 1.25]
+        assert (
+            scale in supported_scale
+        ), "supported scales are {} but input scale is {}".format(
+            supported_scale, scale
+        )
+
+        inplanes = 16
+        # conv1
+        self.conv1 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=make_divisible(inplanes * scale),
+            kernel_size=3,
+            stride=2,
+            padding=1,
+            groups=1,
+            if_act=True,
+            act="hardswish",
+        )
+        i = 0
+        block_list = []
+        inplanes = make_divisible(inplanes * scale)
+        for k, exp, c, se, nl, s in cfg:
+            se = se and not self.disable_se
+            block_list.append(
+                ResidualUnit(
+                    in_channels=inplanes,
+                    mid_channels=make_divisible(scale * exp),
+                    out_channels=make_divisible(scale * c),
+                    kernel_size=k,
+                    stride=s,
+                    use_se=se,
+                    act=nl,
+                )
+            )
+            inplanes = make_divisible(scale * c)
+            i += 1
+        self.blocks = nn.Sequential(*block_list)
+
+        self.conv2 = ConvBNLayer(
+            in_channels=inplanes,
+            out_channels=make_divisible(scale * cls_ch_squeeze),
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            groups=1,
+            if_act=True,
+            act="hardswish",
+        )
+
+        self.pool = nn.MaxPool2D(kernel_size=2, stride=2, padding=0)
+        self.out_channels = make_divisible(scale * cls_ch_squeeze)
+
+    def forward(self, x):
+        x = self.conv1(x)
+        x = self.blocks(x)
+        x = self.conv2(x)
+        x = self.pool(x)
+        return x
diff --git a/ppocr/modeling/backbones/rec_mv1_enhance.py b/ppocr/modeling/backbones/rec_mv1_enhance.py
new file mode 100644
index 0000000..f20fa4c
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_mv1_enhance.py
@@ -0,0 +1,283 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This code is refer from: https://github.com/PaddlePaddle/PaddleClas/blob/develop/ppcls/arch/backbone/legendary_models/pp_lcnet.py
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import numpy as np
+import paddle
+from paddle import ParamAttr, reshape, transpose
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle.nn import Conv2D, BatchNorm, Linear, Dropout
+from paddle.nn import AdaptiveAvgPool2D, MaxPool2D, AvgPool2D
+from paddle.nn.initializer import KaimingNormal
+from paddle.regularizer import L2Decay
+from paddle.nn.functional import hardswish, hardsigmoid
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        num_channels,
+        filter_size,
+        num_filters,
+        stride,
+        padding,
+        channels=None,
+        num_groups=1,
+        act="hard_swish",
+    ):
+        super(ConvBNLayer, self).__init__()
+
+        self._conv = Conv2D(
+            in_channels=num_channels,
+            out_channels=num_filters,
+            kernel_size=filter_size,
+            stride=stride,
+            padding=padding,
+            groups=num_groups,
+            weight_attr=ParamAttr(initializer=KaimingNormal()),
+            bias_attr=False,
+        )
+
+        self._batch_norm = BatchNorm(
+            num_filters,
+            act=act,
+            param_attr=ParamAttr(regularizer=L2Decay(0.0)),
+            bias_attr=ParamAttr(regularizer=L2Decay(0.0)),
+        )
+
+    def forward(self, inputs):
+        y = self._conv(inputs)
+        y = self._batch_norm(y)
+        return y
+
+
+class DepthwiseSeparable(nn.Layer):
+    def __init__(
+        self,
+        num_channels,
+        num_filters1,
+        num_filters2,
+        num_groups,
+        stride,
+        scale,
+        dw_size=3,
+        padding=1,
+        use_se=False,
+    ):
+        super(DepthwiseSeparable, self).__init__()
+        self.use_se = use_se
+        self._depthwise_conv = ConvBNLayer(
+            num_channels=num_channels,
+            num_filters=int(num_filters1 * scale),
+            filter_size=dw_size,
+            stride=stride,
+            padding=padding,
+            num_groups=int(num_groups * scale),
+        )
+        if use_se:
+            self._se = SEModule(int(num_filters1 * scale))
+        self._pointwise_conv = ConvBNLayer(
+            num_channels=int(num_filters1 * scale),
+            filter_size=1,
+            num_filters=int(num_filters2 * scale),
+            stride=1,
+            padding=0,
+        )
+
+    def forward(self, inputs):
+        y = self._depthwise_conv(inputs)
+        if self.use_se:
+            y = self._se(y)
+        y = self._pointwise_conv(y)
+        return y
+
+
+class MobileNetV1Enhance(nn.Layer):
+    def __init__(
+        self,
+        in_channels=3,
+        scale=0.5,
+        last_conv_stride=1,
+        last_pool_type="max",
+        last_pool_kernel_size=[3, 2],
+        **kwargs,
+    ):
+        super().__init__()
+        self.scale = scale
+        self.block_list = []
+
+        self.conv1 = ConvBNLayer(
+            num_channels=3,
+            filter_size=3,
+            channels=3,
+            num_filters=int(32 * scale),
+            stride=2,
+            padding=1,
+        )
+
+        conv2_1 = DepthwiseSeparable(
+            num_channels=int(32 * scale),
+            num_filters1=32,
+            num_filters2=64,
+            num_groups=32,
+            stride=1,
+            scale=scale,
+        )
+        self.block_list.append(conv2_1)
+
+        conv2_2 = DepthwiseSeparable(
+            num_channels=int(64 * scale),
+            num_filters1=64,
+            num_filters2=128,
+            num_groups=64,
+            stride=1,
+            scale=scale,
+        )
+        self.block_list.append(conv2_2)
+
+        conv3_1 = DepthwiseSeparable(
+            num_channels=int(128 * scale),
+            num_filters1=128,
+            num_filters2=128,
+            num_groups=128,
+            stride=1,
+            scale=scale,
+        )
+        self.block_list.append(conv3_1)
+
+        conv3_2 = DepthwiseSeparable(
+            num_channels=int(128 * scale),
+            num_filters1=128,
+            num_filters2=256,
+            num_groups=128,
+            stride=(2, 1),
+            scale=scale,
+        )
+        self.block_list.append(conv3_2)
+
+        conv4_1 = DepthwiseSeparable(
+            num_channels=int(256 * scale),
+            num_filters1=256,
+            num_filters2=256,
+            num_groups=256,
+            stride=1,
+            scale=scale,
+        )
+        self.block_list.append(conv4_1)
+
+        conv4_2 = DepthwiseSeparable(
+            num_channels=int(256 * scale),
+            num_filters1=256,
+            num_filters2=512,
+            num_groups=256,
+            stride=(2, 1),
+            scale=scale,
+        )
+        self.block_list.append(conv4_2)
+
+        for _ in range(5):
+            conv5 = DepthwiseSeparable(
+                num_channels=int(512 * scale),
+                num_filters1=512,
+                num_filters2=512,
+                num_groups=512,
+                stride=1,
+                dw_size=5,
+                padding=2,
+                scale=scale,
+                use_se=False,
+            )
+            self.block_list.append(conv5)
+
+        conv5_6 = DepthwiseSeparable(
+            num_channels=int(512 * scale),
+            num_filters1=512,
+            num_filters2=1024,
+            num_groups=512,
+            stride=(2, 1),
+            dw_size=5,
+            padding=2,
+            scale=scale,
+            use_se=True,
+        )
+        self.block_list.append(conv5_6)
+
+        conv6 = DepthwiseSeparable(
+            num_channels=int(1024 * scale),
+            num_filters1=1024,
+            num_filters2=1024,
+            num_groups=1024,
+            stride=last_conv_stride,
+            dw_size=5,
+            padding=2,
+            use_se=True,
+            scale=scale,
+        )
+        self.block_list.append(conv6)
+
+        self.block_list = nn.Sequential(*self.block_list)
+        if last_pool_type == "avg":
+            self.pool = nn.AvgPool2D(
+                kernel_size=last_pool_kernel_size,
+                stride=last_pool_kernel_size,
+                padding=0,
+            )
+        else:
+            self.pool = nn.MaxPool2D(kernel_size=2, stride=2, padding=0)
+        self.out_channels = int(1024 * scale)
+
+    def forward(self, inputs):
+        y = self.conv1(inputs)
+        y = self.block_list(y)
+        y = self.pool(y)
+        return y
+
+
+class SEModule(nn.Layer):
+    def __init__(self, channel, reduction=4):
+        super(SEModule, self).__init__()
+        self.avg_pool = AdaptiveAvgPool2D(1)
+        self.conv1 = Conv2D(
+            in_channels=channel,
+            out_channels=channel // reduction,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            weight_attr=ParamAttr(),
+            bias_attr=ParamAttr(),
+        )
+        self.conv2 = Conv2D(
+            in_channels=channel // reduction,
+            out_channels=channel,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            weight_attr=ParamAttr(),
+            bias_attr=ParamAttr(),
+        )
+
+    def forward(self, inputs):
+        outputs = self.avg_pool(inputs)
+        outputs = self.conv1(outputs)
+        outputs = F.relu(outputs)
+        outputs = self.conv2(outputs)
+        outputs = hardsigmoid(outputs)
+        return paddle.multiply(x=inputs, y=outputs)
diff --git a/ppocr/modeling/backbones/rec_nrtr_mtb.py b/ppocr/modeling/backbones/rec_nrtr_mtb.py
new file mode 100644
index 0000000..608a8a7
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_nrtr_mtb.py
@@ -0,0 +1,47 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddle import nn
+import paddle
+
+
+class MTB(nn.Layer):
+    def __init__(self, cnn_num, in_channels):
+        super(MTB, self).__init__()
+        self.block = nn.Sequential()
+        self.out_channels = in_channels
+        self.cnn_num = cnn_num
+        if self.cnn_num == 2:
+            for i in range(self.cnn_num):
+                self.block.add_sublayer(
+                    "conv_{}".format(i),
+                    nn.Conv2D(
+                        in_channels=in_channels if i == 0 else 32 * (2 ** (i - 1)),
+                        out_channels=32 * (2**i),
+                        kernel_size=3,
+                        stride=2,
+                        padding=1,
+                    ),
+                )
+                self.block.add_sublayer("relu_{}".format(i), nn.ReLU())
+                self.block.add_sublayer("bn_{}".format(i), nn.BatchNorm2D(32 * (2**i)))
+
+    def forward(self, images):
+        x = self.block(images)
+        if self.cnn_num == 2:
+            # (b, w, h, c)
+            x = paddle.transpose(x, [0, 3, 2, 1])
+            x_shape = x.shape
+            x = paddle.reshape(x, [x_shape[0], x_shape[1], x_shape[2] * x_shape[3]])
+        return x
diff --git a/ppocr/modeling/backbones/rec_pphgnetv2.py b/ppocr/modeling/backbones/rec_pphgnetv2.py
new file mode 100644
index 0000000..a969f52
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_pphgnetv2.py
@@ -0,0 +1,1713 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/PaddlePaddle/PaddleClas/blob/2f36cab604e439b59d1a854df34ece3b10d888e3/ppcls/arch/backbone/legendary_models/pp_hgnet_v2.py
+"""
+
+from __future__ import absolute_import, division, print_function
+
+import math
+import numpy as np
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+from paddle.nn import Conv2D, BatchNorm, Linear, BatchNorm2D, MaxPool2D, AvgPool2D
+from paddle.nn.initializer import Uniform
+from paddle.regularizer import L2Decay
+
+from typing import Tuple, List, Dict, Union, Callable, Any
+from ppocr.modeling.backbones.rec_donut_swin import DonutSwinModelOutput
+
+
+class IdentityBasedConv1x1(nn.Conv2D):
+    def __init__(self, channels, groups=1):
+        super(IdentityBasedConv1x1, self).__init__(
+            in_channels=channels,
+            out_channels=channels,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            groups=groups,
+            bias_attr=False,
+        )
+
+        assert channels % groups == 0
+        input_dim = channels // groups
+        id_value = np.zeros((channels, input_dim, 1, 1))
+        for i in range(channels):
+            id_value[i, i % input_dim, 0, 0] = 1
+        self.id_tensor = paddle.to_tensor(id_value)
+        self.weight.set_value(paddle.zeros_like(self.weight))
+
+    def forward(self, input):
+        kernel = self.weight + self.id_tensor
+        result = F.conv2d(
+            input,
+            kernel,
+            None,
+            stride=1,
+            padding=0,
+            dilation=self._dilation,
+            groups=self._groups,
+        )
+        return result
+
+    def get_actual_kernel(self):
+        return self.weight + self.id_tensor
+
+
+class BNAndPad(nn.Layer):
+    def __init__(
+        self,
+        pad_pixels,
+        num_features,
+        epsilon=1e-5,
+        momentum=0.1,
+        last_conv_bias=None,
+        bn=nn.BatchNorm2D,
+    ):
+        super().__init__()
+        self.bn = bn(num_features, momentum=momentum, epsilon=epsilon)
+        self.pad_pixels = pad_pixels
+        self.last_conv_bias = last_conv_bias
+
+    def forward(self, input):
+        output = self.bn(input)
+        if self.pad_pixels > 0:
+            bias = -self.bn._mean
+            if self.last_conv_bias is not None:
+                bias += self.last_conv_bias
+            pad_values = self.bn.bias + self.bn.weight * (
+                bias / paddle.sqrt(self.bn._variance + self.bn._epsilon)
+            )
+            """ pad """
+            # TODO: n,h,w,c format is not supported yet
+            n, c, h, w = output.shape
+            values = pad_values.reshape([1, -1, 1, 1])
+            w_values = values.expand([n, -1, self.pad_pixels, w])
+            x = paddle.concat([w_values, output, w_values], axis=2)
+            h = h + self.pad_pixels * 2
+            h_values = values.expand([n, -1, h, self.pad_pixels])
+            x = paddle.concat([h_values, x, h_values], axis=3)
+            output = x
+        return output
+
+    @property
+    def weight(self):
+        return self.bn.weight
+
+    @property
+    def bias(self):
+        return self.bn.bias
+
+    @property
+    def _mean(self):
+        return self.bn._mean
+
+    @property
+    def _variance(self):
+        return self.bn._variance
+
+    @property
+    def _epsilon(self):
+        return self.bn._epsilon
+
+
+def conv_bn(
+    in_channels,
+    out_channels,
+    kernel_size,
+    stride=1,
+    padding=0,
+    dilation=1,
+    groups=1,
+    padding_mode="zeros",
+):
+    conv_layer = nn.Conv2D(
+        in_channels=in_channels,
+        out_channels=out_channels,
+        kernel_size=kernel_size,
+        stride=stride,
+        padding=padding,
+        dilation=dilation,
+        groups=groups,
+        bias_attr=False,
+        padding_mode=padding_mode,
+    )
+    bn_layer = nn.BatchNorm2D(num_features=out_channels)
+    se = nn.Sequential()
+    se.add_sublayer("conv", conv_layer)
+    se.add_sublayer("bn", bn_layer)
+    return se
+
+
+def transI_fusebn(kernel, bn):
+    gamma = bn.weight
+    std = (bn._variance + bn._epsilon).sqrt()
+    return (
+        kernel * ((gamma / std).reshape([-1, 1, 1, 1])),
+        bn.bias - bn._mean * gamma / std,
+    )
+
+
+def transII_addbranch(kernels, biases):
+    return sum(kernels), sum(biases)
+
+
+def transIII_1x1_kxk(k1, b1, k2, b2, groups):
+    if groups == 1:
+        k = F.conv2d(k2, k1.transpose([1, 0, 2, 3]))
+        b_hat = (k2 * b1.reshape([1, -1, 1, 1])).sum((1, 2, 3))
+    else:
+        k_slices = []
+        b_slices = []
+        k1_T = k1.transpose([1, 0, 2, 3])
+        k1_group_width = k1.shape[0] // groups
+        k2_group_width = k2.shape[0] // groups
+        for g in range(groups):
+            k1_T_slice = k1_T[:, g * k1_group_width : (g + 1) * k1_group_width, :, :]
+            k2_slice = k2[g * k2_group_width : (g + 1) * k2_group_width, :, :, :]
+            k_slices.append(F.conv2d(k2_slice, k1_T_slice))
+            b_slices.append(
+                (
+                    k2_slice
+                    * b1[g * k1_group_width : (g + 1) * k1_group_width].reshape(
+                        [1, -1, 1, 1]
+                    )
+                ).sum((1, 2, 3))
+            )
+        k, b_hat = transIV_depthconcat(k_slices, b_slices)
+    return k, b_hat + b2
+
+
+def transIV_depthconcat(kernels, biases):
+    return paddle.cat(kernels, axis=0), paddle.cat(biases)
+
+
+def transV_avg(channels, kernel_size, groups):
+    input_dim = channels // groups
+    k = paddle.zeros((channels, input_dim, kernel_size, kernel_size))
+    k[np.arange(channels), np.tile(np.arange(input_dim), groups), :, :] = (
+        1.0 / kernel_size**2
+    )
+    return k
+
+
+def transVI_multiscale(kernel, target_kernel_size):
+    H_pixels_to_pad = (target_kernel_size - kernel.shape[2]) // 2
+    W_pixels_to_pad = (target_kernel_size - kernel.shape[3]) // 2
+    return F.pad(
+        kernel, [H_pixels_to_pad, H_pixels_to_pad, W_pixels_to_pad, W_pixels_to_pad]
+    )
+
+
+class DiverseBranchBlock(nn.Layer):
+    def __init__(
+        self,
+        num_channels,
+        num_filters,
+        filter_size,
+        stride=1,
+        groups=1,
+        act=None,
+        is_repped=False,
+        single_init=False,
+        **kwargs,
+    ):
+        super().__init__()
+
+        padding = (filter_size - 1) // 2
+        dilation = 1
+
+        in_channels = num_channels
+        out_channels = num_filters
+        kernel_size = filter_size
+        internal_channels_1x1_3x3 = None
+        nonlinear = act
+
+        self.is_repped = is_repped
+
+        if nonlinear is None:
+            self.nonlinear = nn.Identity()
+        else:
+            self.nonlinear = nn.ReLU()
+
+        self.kernel_size = kernel_size
+        self.out_channels = out_channels
+        self.groups = groups
+        assert padding == kernel_size // 2
+
+        if is_repped:
+            self.dbb_reparam = nn.Conv2D(
+                in_channels=in_channels,
+                out_channels=out_channels,
+                kernel_size=kernel_size,
+                stride=stride,
+                padding=padding,
+                dilation=dilation,
+                groups=groups,
+                bias_attr=True,
+            )
+        else:
+            self.dbb_origin = conv_bn(
+                in_channels=in_channels,
+                out_channels=out_channels,
+                kernel_size=kernel_size,
+                stride=stride,
+                padding=padding,
+                dilation=dilation,
+                groups=groups,
+            )
+
+            self.dbb_avg = nn.Sequential()
+            if groups < out_channels:
+                self.dbb_avg.add_sublayer(
+                    "conv",
+                    nn.Conv2D(
+                        in_channels=in_channels,
+                        out_channels=out_channels,
+                        kernel_size=1,
+                        stride=1,
+                        padding=0,
+                        groups=groups,
+                        bias_attr=False,
+                    ),
+                )
+                self.dbb_avg.add_sublayer(
+                    "bn", BNAndPad(pad_pixels=padding, num_features=out_channels)
+                )
+                self.dbb_avg.add_sublayer(
+                    "avg",
+                    nn.AvgPool2D(kernel_size=kernel_size, stride=stride, padding=0),
+                )
+                self.dbb_1x1 = conv_bn(
+                    in_channels=in_channels,
+                    out_channels=out_channels,
+                    kernel_size=1,
+                    stride=stride,
+                    padding=0,
+                    groups=groups,
+                )
+            else:
+                self.dbb_avg.add_sublayer(
+                    "avg",
+                    nn.AvgPool2D(
+                        kernel_size=kernel_size, stride=stride, padding=padding
+                    ),
+                )
+
+            self.dbb_avg.add_sublayer("avgbn", nn.BatchNorm2D(out_channels))
+
+            if internal_channels_1x1_3x3 is None:
+                internal_channels_1x1_3x3 = (
+                    in_channels if groups < out_channels else 2 * in_channels
+                )  # For mobilenet, it is better to have 2X internal channels
+
+            self.dbb_1x1_kxk = nn.Sequential()
+            if internal_channels_1x1_3x3 == in_channels:
+                self.dbb_1x1_kxk.add_sublayer(
+                    "idconv1", IdentityBasedConv1x1(channels=in_channels, groups=groups)
+                )
+            else:
+                self.dbb_1x1_kxk.add_sublayer(
+                    "conv1",
+                    nn.Conv2D(
+                        in_channels=in_channels,
+                        out_channels=internal_channels_1x1_3x3,
+                        kernel_size=1,
+                        stride=1,
+                        padding=0,
+                        groups=groups,
+                        bias_attr=False,
+                    ),
+                )
+            self.dbb_1x1_kxk.add_sublayer(
+                "bn1",
+                BNAndPad(pad_pixels=padding, num_features=internal_channels_1x1_3x3),
+            )
+            self.dbb_1x1_kxk.add_sublayer(
+                "conv2",
+                nn.Conv2D(
+                    in_channels=internal_channels_1x1_3x3,
+                    out_channels=out_channels,
+                    kernel_size=kernel_size,
+                    stride=stride,
+                    padding=0,
+                    groups=groups,
+                    bias_attr=False,
+                ),
+            )
+            self.dbb_1x1_kxk.add_sublayer("bn2", nn.BatchNorm2D(out_channels))
+
+        #   The experiments reported in the paper used the default initialization of bn.weight (all as 1). But changing the initialization may be useful in some cases.
+        if single_init:
+            #   Initialize the bn.weight of dbb_origin as 1 and others as 0. This is not the default setting.
+            self.single_init()
+
+    def forward(self, inputs):
+        if self.is_repped:
+            return self.nonlinear(self.dbb_reparam(inputs))
+
+        out = self.dbb_origin(inputs)
+        if hasattr(self, "dbb_1x1"):
+            out += self.dbb_1x1(inputs)
+        out += self.dbb_avg(inputs)
+        out += self.dbb_1x1_kxk(inputs)
+        return self.nonlinear(out)
+
+    def init_gamma(self, gamma_value):
+        if hasattr(self, "dbb_origin"):
+            paddle.nn.init.constant_(self.dbb_origin.bn.weight, gamma_value)
+        if hasattr(self, "dbb_1x1"):
+            paddle.nn.init.constant_(self.dbb_1x1.bn.weight, gamma_value)
+        if hasattr(self, "dbb_avg"):
+            paddle.nn.init.constant_(self.dbb_avg.avgbn.weight, gamma_value)
+        if hasattr(self, "dbb_1x1_kxk"):
+            paddle.nn.init.constant_(self.dbb_1x1_kxk.bn2.weight, gamma_value)
+
+    def single_init(self):
+        self.init_gamma(0.0)
+        if hasattr(self, "dbb_origin"):
+            paddle.nn.init.constant_(self.dbb_origin.bn.weight, 1.0)
+
+    def get_equivalent_kernel_bias(self):
+        k_origin, b_origin = transI_fusebn(
+            self.dbb_origin.conv.weight, self.dbb_origin.bn
+        )
+
+        if hasattr(self, "dbb_1x1"):
+            k_1x1, b_1x1 = transI_fusebn(self.dbb_1x1.conv.weight, self.dbb_1x1.bn)
+            k_1x1 = transVI_multiscale(k_1x1, self.kernel_size)
+        else:
+            k_1x1, b_1x1 = 0, 0
+
+        if hasattr(self.dbb_1x1_kxk, "idconv1"):
+            k_1x1_kxk_first = self.dbb_1x1_kxk.idconv1.get_actual_kernel()
+        else:
+            k_1x1_kxk_first = self.dbb_1x1_kxk.conv1.weight
+        k_1x1_kxk_first, b_1x1_kxk_first = transI_fusebn(
+            k_1x1_kxk_first, self.dbb_1x1_kxk.bn1
+        )
+        k_1x1_kxk_second, b_1x1_kxk_second = transI_fusebn(
+            self.dbb_1x1_kxk.conv2.weight, self.dbb_1x1_kxk.bn2
+        )
+        k_1x1_kxk_merged, b_1x1_kxk_merged = transIII_1x1_kxk(
+            k_1x1_kxk_first,
+            b_1x1_kxk_first,
+            k_1x1_kxk_second,
+            b_1x1_kxk_second,
+            groups=self.groups,
+        )
+
+        k_avg = transV_avg(self.out_channels, self.kernel_size, self.groups)
+        k_1x1_avg_second, b_1x1_avg_second = transI_fusebn(k_avg, self.dbb_avg.avgbn)
+        if hasattr(self.dbb_avg, "conv"):
+            k_1x1_avg_first, b_1x1_avg_first = transI_fusebn(
+                self.dbb_avg.conv.weight, self.dbb_avg.bn
+            )
+            k_1x1_avg_merged, b_1x1_avg_merged = transIII_1x1_kxk(
+                k_1x1_avg_first,
+                b_1x1_avg_first,
+                k_1x1_avg_second,
+                b_1x1_avg_second,
+                groups=self.groups,
+            )
+        else:
+            k_1x1_avg_merged, b_1x1_avg_merged = k_1x1_avg_second, b_1x1_avg_second
+
+        return transII_addbranch(
+            (k_origin, k_1x1, k_1x1_kxk_merged, k_1x1_avg_merged),
+            (b_origin, b_1x1, b_1x1_kxk_merged, b_1x1_avg_merged),
+        )
+
+    def re_parameterize(self):
+        if self.is_repped:
+            return
+
+        kernel, bias = self.get_equivalent_kernel_bias()
+        self.dbb_reparam = nn.Conv2D(
+            in_channels=self.dbb_origin.conv._in_channels,
+            out_channels=self.dbb_origin.conv._out_channels,
+            kernel_size=self.dbb_origin.conv._kernel_size,
+            stride=self.dbb_origin.conv._stride,
+            padding=self.dbb_origin.conv._padding,
+            dilation=self.dbb_origin.conv._dilation,
+            groups=self.dbb_origin.conv._groups,
+            bias_attr=True,
+        )
+
+        self.dbb_reparam.weight.set_value(kernel)
+        self.dbb_reparam.bias.set_value(bias)
+
+        self.__delattr__("dbb_origin")
+        self.__delattr__("dbb_avg")
+        if hasattr(self, "dbb_1x1"):
+            self.__delattr__("dbb_1x1")
+        self.__delattr__("dbb_1x1_kxk")
+        self.is_repped = True
+
+
+class Identity(nn.Layer):
+    def __init__(self):
+        super(Identity, self).__init__()
+
+    def forward(self, inputs):
+        return inputs
+
+
+class TheseusLayer(nn.Layer):
+    def __init__(self, *args, **kwargs):
+        super().__init__()
+        self.res_dict = {}
+        self.res_name = self.full_name()
+        self.pruner = None
+        self.quanter = None
+
+        self.init_net(*args, **kwargs)
+
+    def _return_dict_hook(self, layer, input, output):
+        res_dict = {"logits": output}
+        # 'list' is needed to avoid error raised by popping self.res_dict
+        for res_key in list(self.res_dict):
+            # clear the res_dict because the forward process may change according to input
+            res_dict[res_key] = self.res_dict.pop(res_key)
+        return res_dict
+
+    def init_net(
+        self,
+        stages_pattern=None,
+        return_patterns=None,
+        return_stages=None,
+        freeze_befor=None,
+        stop_after=None,
+        *args,
+        **kwargs,
+    ):
+        # init the output of net
+        if return_patterns or return_stages:
+            if return_patterns and return_stages:
+                msg = f"The 'return_patterns' would be ignored when 'return_stages' is set."
+
+                return_stages = None
+
+            if return_stages is True:
+                return_patterns = stages_pattern
+
+            # return_stages is int or bool
+            if type(return_stages) is int:
+                return_stages = [return_stages]
+            if isinstance(return_stages, list):
+                if max(return_stages) > len(stages_pattern) or min(return_stages) < 0:
+                    msg = f"The 'return_stages' set error. Illegal value(s) have been ignored. The stages' pattern list is {stages_pattern}."
+
+                    return_stages = [
+                        val
+                        for val in return_stages
+                        if val >= 0 and val < len(stages_pattern)
+                    ]
+                return_patterns = [stages_pattern[i] for i in return_stages]
+
+            if return_patterns:
+                # call update_res function after the __init__ of the object has completed execution, that is, the constructing of layer or model has been completed.
+                def update_res_hook(layer, input):
+                    self.update_res(return_patterns)
+
+                self.register_forward_pre_hook(update_res_hook)
+
+        # freeze subnet
+        if freeze_befor is not None:
+            self.freeze_befor(freeze_befor)
+
+        # set subnet to Identity
+        if stop_after is not None:
+            self.stop_after(stop_after)
+
+    def init_res(self, stages_pattern, return_patterns=None, return_stages=None):
+
+        if return_patterns and return_stages:
+            return_stages = None
+
+        if return_stages is True:
+            return_patterns = stages_pattern
+        # return_stages is int or bool
+        if type(return_stages) is int:
+            return_stages = [return_stages]
+        if isinstance(return_stages, list):
+            if max(return_stages) > len(stages_pattern) or min(return_stages) < 0:
+                return_stages = [
+                    val
+                    for val in return_stages
+                    if val >= 0 and val < len(stages_pattern)
+                ]
+            return_patterns = [stages_pattern[i] for i in return_stages]
+
+        if return_patterns:
+            self.update_res(return_patterns)
+
+    def replace_sub(self, *args, **kwargs) -> None:
+        msg = "The function 'replace_sub()' is deprecated, please use 'upgrade_sublayer()' instead."
+        raise DeprecationWarning(msg)
+
+    def upgrade_sublayer(
+        self,
+        layer_name_pattern: Union[str, List[str]],
+        handle_func: Callable[[nn.Layer, str], nn.Layer],
+    ) -> Dict[str, nn.Layer]:
+        """use 'handle_func' to modify the sub-layer(s) specified by 'layer_name_pattern'.
+
+        Args:
+            layer_name_pattern (Union[str, List[str]]): The name of layer to be modified by 'handle_func'.
+            handle_func (Callable[[nn.Layer, str], nn.Layer]): The function to modify target layer specified by 'layer_name_pattern'. The formal params are the layer(nn.Layer) and pattern(str) that is (a member of) layer_name_pattern (when layer_name_pattern is List type). And the return is the layer processed.
+
+        Returns:
+            Dict[str, nn.Layer]: The key is the pattern and corresponding value is the result returned by 'handle_func()'.
+
+        Examples:
+
+            from paddle import nn
+            import paddleclas
+
+            def rep_func(layer: nn.Layer, pattern: str):
+                new_layer = nn.Conv2D(
+                    in_channels=layer._in_channels,
+                    out_channels=layer._out_channels,
+                    kernel_size=5,
+                    padding=2
+                )
+                return new_layer
+
+            net = paddleclas.MobileNetV1()
+            res = net.upgrade_sublayer(layer_name_pattern=["blocks[11].depthwise_conv.conv", "blocks[12].depthwise_conv.conv"], handle_func=rep_func)
+            print(res)
+            # {'blocks[11].depthwise_conv.conv': the corresponding new_layer, 'blocks[12].depthwise_conv.conv': the corresponding new_layer}
+        """
+
+        if not isinstance(layer_name_pattern, list):
+            layer_name_pattern = [layer_name_pattern]
+
+        hit_layer_pattern_list = []
+        for pattern in layer_name_pattern:
+            # parse pattern to find target layer and its parent
+            layer_list = parse_pattern_str(pattern=pattern, parent_layer=self)
+            if not layer_list:
+                continue
+
+            sub_layer_parent = layer_list[-2]["layer"] if len(layer_list) > 1 else self
+            sub_layer = layer_list[-1]["layer"]
+            sub_layer_name = layer_list[-1]["name"]
+            sub_layer_index_list = layer_list[-1]["index_list"]
+
+            new_sub_layer = handle_func(sub_layer, pattern)
+
+            if sub_layer_index_list:
+                if len(sub_layer_index_list) > 1:
+                    sub_layer_parent = getattr(sub_layer_parent, sub_layer_name)[
+                        sub_layer_index_list[0]
+                    ]
+                    for sub_layer_index in sub_layer_index_list[1:-1]:
+                        sub_layer_parent = sub_layer_parent[sub_layer_index]
+                    sub_layer_parent[sub_layer_index_list[-1]] = new_sub_layer
+                else:
+                    getattr(sub_layer_parent, sub_layer_name)[
+                        sub_layer_index_list[0]
+                    ] = new_sub_layer
+            else:
+                setattr(sub_layer_parent, sub_layer_name, new_sub_layer)
+
+            hit_layer_pattern_list.append(pattern)
+        return hit_layer_pattern_list
+
+    def stop_after(self, stop_layer_name: str) -> bool:
+        """stop forward and backward after 'stop_layer_name'.
+
+        Args:
+            stop_layer_name (str): The name of layer that stop forward and backward after this layer.
+
+        Returns:
+            bool: 'True' if successful, 'False' otherwise.
+        """
+
+        layer_list = parse_pattern_str(stop_layer_name, self)
+        if not layer_list:
+            return False
+
+        parent_layer = self
+        for layer_dict in layer_list:
+            name, index_list = layer_dict["name"], layer_dict["index_list"]
+            if not set_identity(parent_layer, name, index_list):
+                msg = f"Failed to set the layers that after stop_layer_name('{stop_layer_name}') to IdentityLayer. The error layer's name is '{name}'."
+                return False
+            parent_layer = layer_dict["layer"]
+
+        return True
+
+    def freeze_befor(self, layer_name: str) -> bool:
+        """freeze the layer named layer_name and its previous layer.
+
+        Args:
+            layer_name (str): The name of layer that would be freezed.
+
+        Returns:
+            bool: 'True' if successful, 'False' otherwise.
+        """
+
+        def stop_grad(layer, pattern):
+            class StopGradLayer(nn.Layer):
+                def __init__(self):
+                    super().__init__()
+                    self.layer = layer
+
+                def forward(self, x):
+                    x = self.layer(x)
+                    x.stop_gradient = True
+                    return x
+
+            new_layer = StopGradLayer()
+            return new_layer
+
+        res = self.upgrade_sublayer(layer_name, stop_grad)
+        if len(res) == 0:
+            msg = "Failed to stop the gradient before the layer named '{layer_name}'"
+            return False
+        return True
+
+    def update_res(self, return_patterns: Union[str, List[str]]) -> Dict[str, nn.Layer]:
+        """update the result(s) to be returned.
+
+        Args:
+            return_patterns (Union[str, List[str]]): The name of layer to return output.
+
+        Returns:
+            Dict[str, nn.Layer]: The pattern(str) and corresponding layer(nn.Layer) that have been set successfully.
+        """
+
+        # clear res_dict that could have been set
+        self.res_dict = {}
+
+        class Handler(object):
+            def __init__(self, res_dict):
+                # res_dict is a reference
+                self.res_dict = res_dict
+
+            def __call__(self, layer, pattern):
+                layer.res_dict = self.res_dict
+                layer.res_name = pattern
+                if hasattr(layer, "hook_remove_helper"):
+                    layer.hook_remove_helper.remove()
+                layer.hook_remove_helper = layer.register_forward_post_hook(
+                    save_sub_res_hook
+                )
+                return layer
+
+        handle_func = Handler(self.res_dict)
+
+        hit_layer_pattern_list = self.upgrade_sublayer(
+            return_patterns, handle_func=handle_func
+        )
+
+        if hasattr(self, "hook_remove_helper"):
+            self.hook_remove_helper.remove()
+        self.hook_remove_helper = self.register_forward_post_hook(
+            self._return_dict_hook
+        )
+
+        return hit_layer_pattern_list
+
+
+def save_sub_res_hook(layer, input, output):
+    layer.res_dict[layer.res_name] = output
+
+
+def set_identity(
+    parent_layer: nn.Layer, layer_name: str, layer_index_list: str = None
+) -> bool:
+    """set the layer specified by layer_name and layer_index_list to Identity.
+
+    Args:
+        parent_layer (nn.Layer): The parent layer of target layer specified by layer_name and layer_index_list.
+        layer_name (str): The name of target layer to be set to Identity.
+        layer_index_list (str, optional): The index of target layer to be set to Identity in parent_layer. Defaults to None.
+
+    Returns:
+        bool: True if successfully, False otherwise.
+    """
+
+    stop_after = False
+    for sub_layer_name in parent_layer._sub_layers:
+        if stop_after:
+            parent_layer._sub_layers[sub_layer_name] = Identity()
+            continue
+        if sub_layer_name == layer_name:
+            stop_after = True
+
+    if layer_index_list and stop_after:
+        layer_container = parent_layer._sub_layers[layer_name]
+        for num, layer_index in enumerate(layer_index_list):
+            stop_after = False
+            for i in range(num):
+                layer_container = layer_container[layer_index_list[i]]
+            for sub_layer_index in layer_container._sub_layers:
+                if stop_after:
+                    parent_layer._sub_layers[layer_name][sub_layer_index] = Identity()
+                    continue
+                if layer_index == sub_layer_index:
+                    stop_after = True
+
+    return stop_after
+
+
+def parse_pattern_str(
+    pattern: str, parent_layer: nn.Layer
+) -> Union[None, List[Dict[str, Union[nn.Layer, str, None]]]]:
+    """parse the string type pattern.
+
+    Args:
+        pattern (str): The pattern to describe layer.
+        parent_layer (nn.Layer): The root layer relative to the pattern.
+
+    Returns:
+        Union[None, List[Dict[str, Union[nn.Layer, str, None]]]]: None if failed. If successfully, the members are layers parsed in order:
+                                                                [
+                                                                    {"layer": first layer, "name": first layer's name parsed, "index": first layer's index parsed if exist},
+                                                                    {"layer": second layer, "name": second layer's name parsed, "index": second layer's index parsed if exist},
+                                                                    ...
+                                                                ]
+    """
+
+    pattern_list = pattern.split(".")
+    if not pattern_list:
+        msg = f"The pattern('{pattern}') is illegal. Please check and retry."
+        return None
+
+    layer_list = []
+    while len(pattern_list) > 0:
+        if "[" in pattern_list[0]:
+            target_layer_name = pattern_list[0].split("[")[0]
+            target_layer_index_list = list(
+                index.split("]")[0] for index in pattern_list[0].split("[")[1:]
+            )
+        else:
+            target_layer_name = pattern_list[0]
+            target_layer_index_list = None
+
+        target_layer = getattr(parent_layer, target_layer_name, None)
+
+        if target_layer is None:
+            msg = f"Not found layer named('{target_layer_name}') specified in pattern('{pattern}')."
+            return None
+
+        if target_layer_index_list:
+            for target_layer_index in target_layer_index_list:
+                if int(target_layer_index) < 0 or int(target_layer_index) >= len(
+                    target_layer
+                ):
+                    msg = f"Not found layer by index('{target_layer_index}') specified in pattern('{pattern}'). The index should < {len(target_layer)} and > 0."
+                    return None
+                target_layer = target_layer[target_layer_index]
+
+        layer_list.append(
+            {
+                "layer": target_layer,
+                "name": target_layer_name,
+                "index_list": target_layer_index_list,
+            }
+        )
+
+        pattern_list = pattern_list[1:]
+        parent_layer = target_layer
+
+    return layer_list
+
+
+class AdaptiveAvgPool2D(nn.AdaptiveAvgPool2D):
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        if paddle.device.get_device().startswith("npu"):
+            self.device = "npu"
+        else:
+            self.device = None
+
+        if isinstance(self._output_size, int) and self._output_size == 1:
+            self._gap = True
+        elif (
+            isinstance(self._output_size, tuple)
+            and self._output_size[0] == 1
+            and self._output_size[1] == 1
+        ):
+            self._gap = True
+        else:
+            self._gap = False
+
+    def forward(self, x):
+        if self.device == "npu" and self._gap:
+            # Global Average Pooling
+            N, C, _, _ = x.shape
+            x_mean = paddle.mean(x, axis=[2, 3])
+            x_mean = paddle.reshape(x_mean, [N, C, 1, 1])
+            return x_mean
+        else:
+            return F.adaptive_avg_pool2d(
+                x,
+                output_size=self._output_size,
+                data_format=self._data_format,
+                name=self._name,
+            )
+
+
+# copyright (c) 2023 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle.nn.initializer import KaimingNormal, Constant
+from paddle.nn import Conv2D, BatchNorm2D, ReLU, AdaptiveAvgPool2D, MaxPool2D
+from paddle.regularizer import L2Decay
+from paddle import ParamAttr
+
+MODEL_URLS = {
+    "PPHGNetV2_B0": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPHGNetV2_B0_ssld_pretrained.pdparams",
+    "PPHGNetV2_B1": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPHGNetV2_B1_ssld_pretrained.pdparams",
+    "PPHGNetV2_B2": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPHGNetV2_B2_ssld_pretrained.pdparams",
+    "PPHGNetV2_B3": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPHGNetV2_B3_ssld_pretrained.pdparams",
+    "PPHGNetV2_B4": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPHGNetV2_B4_ssld_pretrained.pdparams",
+    "PPHGNetV2_B5": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPHGNetV2_B5_ssld_pretrained.pdparams",
+    "PPHGNetV2_B6": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/legendary_models/PPHGNetV2_B6_ssld_pretrained.pdparams",
+}
+
+__all__ = list(MODEL_URLS.keys())
+
+kaiming_normal_ = KaimingNormal()
+zeros_ = Constant(value=0.0)
+ones_ = Constant(value=1.0)
+
+
+class LearnableAffineBlock(TheseusLayer):
+    """
+    Create a learnable affine block module. This module can significantly improve accuracy on smaller models.
+
+    Args:
+        scale_value (float): The initial value of the scale parameter, default is 1.0.
+        bias_value (float): The initial value of the bias parameter, default is 0.0.
+        lr_mult (float): The learning rate multiplier, default is 1.0.
+        lab_lr (float): The learning rate, default is 0.01.
+    """
+
+    def __init__(self, scale_value=1.0, bias_value=0.0, lr_mult=1.0, lab_lr=0.01):
+        super().__init__()
+        self.scale = self.create_parameter(
+            shape=[
+                1,
+            ],
+            default_initializer=Constant(value=scale_value),
+            attr=ParamAttr(learning_rate=lr_mult * lab_lr),
+        )
+        self.add_parameter("scale", self.scale)
+        self.bias = self.create_parameter(
+            shape=[
+                1,
+            ],
+            default_initializer=Constant(value=bias_value),
+            attr=ParamAttr(learning_rate=lr_mult * lab_lr),
+        )
+        self.add_parameter("bias", self.bias)
+
+    def forward(self, x):
+        return self.scale * x + self.bias
+
+
+class ConvBNAct(TheseusLayer):
+    """
+    ConvBNAct is a combination of convolution and batchnorm layers.
+
+    Args:
+        in_channels (int): Number of input channels.
+        out_channels (int): Number of output channels.
+        kernel_size (int): Size of the convolution kernel. Defaults to 3.
+        stride (int): Stride of the convolution. Defaults to 1.
+        padding (int/str): Padding or padding type for the convolution. Defaults to 1.
+        groups (int): Number of groups for the convolution. Defaults to 1.
+        use_act: (bool): Whether to use activation function. Defaults to True.
+        use_lab (bool): Whether to use the LAB operation. Defaults to False.
+        lr_mult (float): Learning rate multiplier for the layer. Defaults to 1.0.
+    """
+
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size=3,
+        stride=1,
+        padding=1,
+        groups=1,
+        use_act=True,
+        use_lab=False,
+        lr_mult=1.0,
+    ):
+        super().__init__()
+        self.use_act = use_act
+        self.use_lab = use_lab
+        self.conv = Conv2D(
+            in_channels,
+            out_channels,
+            kernel_size,
+            stride,
+            padding=padding if isinstance(padding, str) else (kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(learning_rate=lr_mult),
+            bias_attr=False,
+        )
+        self.bn = BatchNorm2D(
+            out_channels,
+            weight_attr=ParamAttr(regularizer=L2Decay(0.0), learning_rate=lr_mult),
+            bias_attr=ParamAttr(regularizer=L2Decay(0.0), learning_rate=lr_mult),
+        )
+        if self.use_act:
+            self.act = ReLU()
+            if self.use_lab:
+                self.lab = LearnableAffineBlock(lr_mult=lr_mult)
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        if self.use_act:
+            x = self.act(x)
+            if self.use_lab:
+                x = self.lab(x)
+        return x
+
+
+class LightConvBNAct(TheseusLayer):
+    """
+    LightConvBNAct is a combination of pw and dw layers.
+
+    Args:
+        in_channels (int): Number of input channels.
+        out_channels (int): Number of output channels.
+        kernel_size (int): Size of the depth-wise convolution kernel.
+        use_lab (bool): Whether to use the LAB operation. Defaults to False.
+        lr_mult (float): Learning rate multiplier for the layer. Defaults to 1.0.
+    """
+
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        use_lab=False,
+        lr_mult=1.0,
+        **kwargs,
+    ):
+        super().__init__()
+        self.conv1 = ConvBNAct(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=1,
+            use_act=False,
+            use_lab=use_lab,
+            lr_mult=lr_mult,
+        )
+        self.conv2 = ConvBNAct(
+            in_channels=out_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            groups=out_channels,
+            use_act=True,
+            use_lab=use_lab,
+            lr_mult=lr_mult,
+        )
+
+    def forward(self, x):
+        x = self.conv1(x)
+        x = self.conv2(x)
+        return x
+
+
+class StemBlock(TheseusLayer):
+    """
+    StemBlock for PP-HGNetV2.
+
+    Args:
+        in_channels (int): Number of input channels.
+        mid_channels (int): Number of middle channels.
+        out_channels (int): Number of output channels.
+        use_lab (bool): Whether to use the LAB operation. Defaults to False.
+        lr_mult (float): Learning rate multiplier for the layer. Defaults to 1.0.
+    """
+
+    def __init__(
+        self,
+        in_channels,
+        mid_channels,
+        out_channels,
+        use_lab=False,
+        lr_mult=1.0,
+        text_rec=False,
+    ):
+        super().__init__()
+        self.stem1 = ConvBNAct(
+            in_channels=in_channels,
+            out_channels=mid_channels,
+            kernel_size=3,
+            stride=2,
+            use_lab=use_lab,
+            lr_mult=lr_mult,
+        )
+        self.stem2a = ConvBNAct(
+            in_channels=mid_channels,
+            out_channels=mid_channels // 2,
+            kernel_size=2,
+            stride=1,
+            padding="SAME",
+            use_lab=use_lab,
+            lr_mult=lr_mult,
+        )
+        self.stem2b = ConvBNAct(
+            in_channels=mid_channels // 2,
+            out_channels=mid_channels,
+            kernel_size=2,
+            stride=1,
+            padding="SAME",
+            use_lab=use_lab,
+            lr_mult=lr_mult,
+        )
+        self.stem3 = ConvBNAct(
+            in_channels=mid_channels * 2,
+            out_channels=mid_channels,
+            kernel_size=3,
+            stride=1 if text_rec else 2,
+            use_lab=use_lab,
+            lr_mult=lr_mult,
+        )
+        self.stem4 = ConvBNAct(
+            in_channels=mid_channels,
+            out_channels=out_channels,
+            kernel_size=1,
+            stride=1,
+            use_lab=use_lab,
+            lr_mult=lr_mult,
+        )
+        self.pool = nn.MaxPool2D(
+            kernel_size=2, stride=1, ceil_mode=True, padding="SAME"
+        )
+
+    def forward(self, x):
+        x = self.stem1(x)
+        x2 = self.stem2a(x)
+        x2 = self.stem2b(x2)
+        x1 = self.pool(x)
+        x = paddle.concat([x1, x2], 1)
+        x = self.stem3(x)
+        x = self.stem4(x)
+
+        return x
+
+
+class HGV2_Block(TheseusLayer):
+    """
+    HGV2_Block, the basic unit that constitutes the HGV2_Stage.
+
+    Args:
+        in_channels (int): Number of input channels.
+        mid_channels (int): Number of middle channels.
+        out_channels (int): Number of output channels.
+        kernel_size (int): Size of the convolution kernel. Defaults to 3.
+        layer_num (int): Number of layers in the HGV2 block. Defaults to 6.
+        stride (int): Stride of the convolution. Defaults to 1.
+        padding (int/str): Padding or padding type for the convolution. Defaults to 1.
+        groups (int): Number of groups for the convolution. Defaults to 1.
+        use_act (bool): Whether to use activation function. Defaults to True.
+        use_lab (bool): Whether to use the LAB operation. Defaults to False.
+        lr_mult (float): Learning rate multiplier for the layer. Defaults to 1.0.
+    """
+
+    def __init__(
+        self,
+        in_channels,
+        mid_channels,
+        out_channels,
+        kernel_size=3,
+        layer_num=6,
+        identity=False,
+        light_block=True,
+        use_lab=False,
+        lr_mult=1.0,
+    ):
+        super().__init__()
+        self.identity = identity
+
+        self.layers = nn.LayerList()
+        block_type = "LightConvBNAct" if light_block else "ConvBNAct"
+        for i in range(layer_num):
+            self.layers.append(
+                eval(block_type)(
+                    in_channels=in_channels if i == 0 else mid_channels,
+                    out_channels=mid_channels,
+                    stride=1,
+                    kernel_size=kernel_size,
+                    use_lab=use_lab,
+                    lr_mult=lr_mult,
+                )
+            )
+        # feature aggregation
+        total_channels = in_channels + layer_num * mid_channels
+        self.aggregation_squeeze_conv = ConvBNAct(
+            in_channels=total_channels,
+            out_channels=out_channels // 2,
+            kernel_size=1,
+            stride=1,
+            use_lab=use_lab,
+            lr_mult=lr_mult,
+        )
+        self.aggregation_excitation_conv = ConvBNAct(
+            in_channels=out_channels // 2,
+            out_channels=out_channels,
+            kernel_size=1,
+            stride=1,
+            use_lab=use_lab,
+            lr_mult=lr_mult,
+        )
+
+    def forward(self, x):
+        identity = x
+        output = []
+        output.append(x)
+        for layer in self.layers:
+            x = layer(x)
+            output.append(x)
+        x = paddle.concat(output, axis=1)
+        x = self.aggregation_squeeze_conv(x)
+        x = self.aggregation_excitation_conv(x)
+        if self.identity:
+            x += identity
+        return x
+
+
+class HGV2_Stage(TheseusLayer):
+    """
+    HGV2_Stage, the basic unit that constitutes the PPHGNetV2.
+
+    Args:
+        in_channels (int): Number of input channels.
+        mid_channels (int): Number of middle channels.
+        out_channels (int): Number of output channels.
+        block_num (int): Number of blocks in the HGV2 stage.
+        layer_num (int): Number of layers in the HGV2 block. Defaults to 6.
+        is_downsample (bool): Whether to use downsampling operation. Defaults to False.
+        light_block (bool): Whether to use light block. Defaults to True.
+        kernel_size (int): Size of the convolution kernel. Defaults to 3.
+        use_lab (bool, optional): Whether to use the LAB operation. Defaults to False.
+        lr_mult (float, optional): Learning rate multiplier for the layer. Defaults to 1.0.
+    """
+
+    def __init__(
+        self,
+        in_channels,
+        mid_channels,
+        out_channels,
+        block_num,
+        layer_num=6,
+        is_downsample=True,
+        light_block=True,
+        kernel_size=3,
+        use_lab=False,
+        stride=2,
+        lr_mult=1.0,
+    ):
+
+        super().__init__()
+        self.is_downsample = is_downsample
+        if self.is_downsample:
+            self.downsample = ConvBNAct(
+                in_channels=in_channels,
+                out_channels=in_channels,
+                kernel_size=3,
+                stride=stride,
+                groups=in_channels,
+                use_act=False,
+                use_lab=use_lab,
+                lr_mult=lr_mult,
+            )
+
+        blocks_list = []
+        for i in range(block_num):
+            blocks_list.append(
+                HGV2_Block(
+                    in_channels=in_channels if i == 0 else out_channels,
+                    mid_channels=mid_channels,
+                    out_channels=out_channels,
+                    kernel_size=kernel_size,
+                    layer_num=layer_num,
+                    identity=False if i == 0 else True,
+                    light_block=light_block,
+                    use_lab=use_lab,
+                    lr_mult=lr_mult,
+                )
+            )
+        self.blocks = nn.Sequential(*blocks_list)
+
+    def forward(self, x):
+        if self.is_downsample:
+            x = self.downsample(x)
+        x = self.blocks(x)
+        return x
+
+
+class PPHGNetV2(TheseusLayer):
+    """
+    PPHGNetV2
+
+    Args:
+        stage_config (dict): Config for PPHGNetV2 stages. such as the number of channels, stride, etc.
+        stem_channels: (list): Number of channels of the stem of the PPHGNetV2.
+        use_lab (bool): Whether to use the LAB operation. Defaults to False.
+        use_last_conv (bool): Whether to use the last conv layer as the output channel. Defaults to True.
+        class_expand (int): Number of channels for the last 1x1 convolutional layer.
+        drop_prob (float): Dropout probability for the last 1x1 convolutional layer. Defaults to 0.0.
+        class_num (int): The number of classes for the classification layer. Defaults to 1000.
+        lr_mult_list (list): Learning rate multiplier for the stages. Defaults to [1.0, 1.0, 1.0, 1.0, 1.0].
+    Returns:
+        model: nn.Layer. Specific PPHGNetV2 model depends on args.
+    """
+
+    def __init__(
+        self,
+        stage_config,
+        stem_channels=[3, 32, 64],
+        use_lab=False,
+        use_last_conv=True,
+        class_expand=2048,
+        dropout_prob=0.0,
+        class_num=1000,
+        lr_mult_list=[1.0, 1.0, 1.0, 1.0, 1.0],
+        det=False,
+        text_rec=False,
+        out_indices=None,
+        **kwargs,
+    ):
+        super().__init__()
+        self.det = det
+        self.text_rec = text_rec
+        self.use_lab = use_lab
+        self.use_last_conv = use_last_conv
+        self.class_expand = class_expand
+        self.class_num = class_num
+        self.out_indices = out_indices if out_indices is not None else [0, 1, 2, 3]
+        self.out_channels = []
+
+        # stem
+        self.stem = StemBlock(
+            in_channels=stem_channels[0],
+            mid_channels=stem_channels[1],
+            out_channels=stem_channels[2],
+            use_lab=use_lab,
+            lr_mult=lr_mult_list[0],
+            text_rec=text_rec,
+        )
+
+        # stages
+        self.stages = nn.LayerList()
+        for i, k in enumerate(stage_config):
+            (
+                in_channels,
+                mid_channels,
+                out_channels,
+                block_num,
+                is_downsample,
+                light_block,
+                kernel_size,
+                layer_num,
+                stride,
+            ) = stage_config[k]
+            self.stages.append(
+                HGV2_Stage(
+                    in_channels,
+                    mid_channels,
+                    out_channels,
+                    block_num,
+                    layer_num,
+                    is_downsample,
+                    light_block,
+                    kernel_size,
+                    use_lab,
+                    stride,
+                    lr_mult=lr_mult_list[i + 1],
+                )
+            )
+            if i in self.out_indices:
+                self.out_channels.append(out_channels)
+        if not self.det:
+            self.out_channels = stage_config["stage4"][2]
+
+        self.avg_pool = AdaptiveAvgPool2D(1)
+
+        if self.use_last_conv:
+            self.last_conv = Conv2D(
+                in_channels=out_channels,
+                out_channels=self.class_expand,
+                kernel_size=1,
+                stride=1,
+                padding=0,
+                bias_attr=False,
+            )
+            self.act = ReLU()
+            if self.use_lab:
+                self.lab = LearnableAffineBlock()
+            self.dropout = nn.Dropout(p=dropout_prob, mode="downscale_in_infer")
+
+        self.flatten = nn.Flatten(start_axis=1, stop_axis=-1)
+        if not self.det:
+            self.fc = nn.Linear(
+                self.class_expand if self.use_last_conv else out_channels,
+                self.class_num,
+            )
+
+        self._init_weights()
+
+    def _init_weights(self):
+        for m in self.sublayers():
+            if isinstance(m, nn.Conv2D):
+                kaiming_normal_(m.weight)
+            elif isinstance(m, (nn.BatchNorm2D)):
+                ones_(m.weight)
+                zeros_(m.bias)
+            elif isinstance(m, nn.Linear):
+                zeros_(m.bias)
+
+    def forward(self, x):
+        x = self.stem(x)
+        out = []
+        for i, stage in enumerate(self.stages):
+            x = stage(x)
+            if self.det and i in self.out_indices:
+                out.append(x)
+        if self.det:
+            return out
+
+        if self.text_rec:
+            if self.training:
+                x = F.adaptive_avg_pool2d(x, [1, 40])
+            else:
+                x = F.avg_pool2d(x, [3, 2])
+        return x
+
+
+def PPHGNetV2_B0(pretrained=False, use_ssld=False, **kwargs):
+    """
+    PPHGNetV2_B0
+    Args:
+        pretrained (bool/str): If `True` load pretrained parameters, `False` otherwise.
+                    If str, means the path of the pretrained model.
+        use_ssld (bool) Whether using ssld pretrained model when pretrained is True.
+    Returns:
+        model: nn.Layer. Specific `PPHGNetV2_B0` model depends on args.
+    """
+    stage_config = {
+        # in_channels, mid_channels, out_channels, num_blocks, is_downsample, light_block, kernel_size, layer_num
+        "stage1": [16, 16, 64, 1, False, False, 3, 3],
+        "stage2": [64, 32, 256, 1, True, False, 3, 3],
+        "stage3": [256, 64, 512, 2, True, True, 5, 3],
+        "stage4": [512, 128, 1024, 1, True, True, 5, 3],
+    }
+
+    model = PPHGNetV2(
+        stem_channels=[3, 16, 16], stage_config=stage_config, use_lab=True, **kwargs
+    )
+    return model
+
+
+def PPHGNetV2_B1(pretrained=False, use_ssld=False, **kwargs):
+    """
+    PPHGNetV2_B1
+    Args:
+        pretrained (bool/str): If `True` load pretrained parameters, `False` otherwise.
+                    If str, means the path of the pretrained model.
+        use_ssld (bool) Whether using ssld pretrained model when pretrained is True.
+    Returns:
+        model: nn.Layer. Specific `PPHGNetV2_B1` model depends on args.
+    """
+    stage_config = {
+        # in_channels, mid_channels, out_channels, num_blocks, is_downsample, light_block, kernel_size, layer_num
+        "stage1": [32, 32, 64, 1, False, False, 3, 3],
+        "stage2": [64, 48, 256, 1, True, False, 3, 3],
+        "stage3": [256, 96, 512, 2, True, True, 5, 3],
+        "stage4": [512, 192, 1024, 1, True, True, 5, 3],
+    }
+
+    model = PPHGNetV2(
+        stem_channels=[3, 24, 32], stage_config=stage_config, use_lab=True, **kwargs
+    )
+    return model
+
+
+def PPHGNetV2_B2(pretrained=False, use_ssld=False, **kwargs):
+    """
+    PPHGNetV2_B2
+    Args:
+        pretrained (bool/str): If `True` load pretrained parameters, `False` otherwise.
+                    If str, means the path of the pretrained model.
+        use_ssld (bool) Whether using ssld pretrained model when pretrained is True.
+    Returns:
+        model: nn.Layer. Specific `PPHGNetV2_B2` model depends on args.
+    """
+    stage_config = {
+        # in_channels, mid_channels, out_channels, num_blocks, is_downsample, light_block, kernel_size, layer_num
+        "stage1": [32, 32, 96, 1, False, False, 3, 4],
+        "stage2": [96, 64, 384, 1, True, False, 3, 4],
+        "stage3": [384, 128, 768, 3, True, True, 5, 4],
+        "stage4": [768, 256, 1536, 1, True, True, 5, 4],
+    }
+
+    model = PPHGNetV2(
+        stem_channels=[3, 24, 32], stage_config=stage_config, use_lab=True, **kwargs
+    )
+    return model
+
+
+def PPHGNetV2_B3(pretrained=False, use_ssld=False, **kwargs):
+    """
+    PPHGNetV2_B3
+    Args:
+        pretrained (bool/str): If `True` load pretrained parameters, `False` otherwise.
+                    If str, means the path of the pretrained model.
+        use_ssld (bool) Whether using ssld pretrained model when pretrained is True.
+    Returns:
+        model: nn.Layer. Specific `PPHGNetV2_B3` model depends on args.
+    """
+    stage_config = {
+        # in_channels, mid_channels, out_channels, num_blocks, is_downsample, light_block, kernel_size, layer_num
+        "stage1": [32, 32, 128, 1, False, False, 3, 5],
+        "stage2": [128, 64, 512, 1, True, False, 3, 5],
+        "stage3": [512, 128, 1024, 3, True, True, 5, 5],
+        "stage4": [1024, 256, 2048, 1, True, True, 5, 5],
+    }
+
+    model = PPHGNetV2(
+        stem_channels=[3, 24, 32], stage_config=stage_config, use_lab=True, **kwargs
+    )
+    return model
+
+
+def PPHGNetV2_B4(pretrained=False, use_ssld=False, det=False, text_rec=False, **kwargs):
+    """
+    PPHGNetV2_B4
+    Args:
+        pretrained (bool/str): If `True` load pretrained parameters, `False` otherwise.
+                    If str, means the path of the pretrained model.
+        use_ssld (bool) Whether using ssld pretrained model when pretrained is True.
+    Returns:
+        model: nn.Layer. Specific `PPHGNetV2_B4` model depends on args.
+    """
+    stage_config_rec = {
+        # in_channels, mid_channels, out_channels, num_blocks, is_downsample, light_block, kernel_size, layer_num, stride
+        "stage1": [48, 48, 128, 1, True, False, 3, 6, [2, 1]],
+        "stage2": [128, 96, 512, 1, True, False, 3, 6, [1, 2]],
+        "stage3": [512, 192, 1024, 3, True, True, 5, 6, [2, 1]],
+        "stage4": [1024, 384, 2048, 1, True, True, 5, 6, [2, 1]],
+    }
+
+    stage_config_det = {
+        # in_channels, mid_channels, out_channels, num_blocks, is_downsample, light_block, kernel_size, layer_num
+        "stage1": [48, 48, 128, 1, False, False, 3, 6, 2],
+        "stage2": [128, 96, 512, 1, True, False, 3, 6, 2],
+        "stage3": [512, 192, 1024, 3, True, True, 5, 6, 2],
+        "stage4": [1024, 384, 2048, 1, True, True, 5, 6, 2],
+    }
+    model = PPHGNetV2(
+        stem_channels=[3, 32, 48],
+        stage_config=stage_config_det if det else stage_config_rec,
+        use_lab=False,
+        det=det,
+        text_rec=text_rec,
+        **kwargs,
+    )
+    return model
+
+
+def PPHGNetV2_B5(pretrained=False, use_ssld=False, **kwargs):
+    """
+    PPHGNetV2_B5
+    Args:
+        pretrained (bool/str): If `True` load pretrained parameters, `False` otherwise.
+                    If str, means the path of the pretrained model.
+        use_ssld (bool) Whether using ssld pretrained model when pretrained is True.
+    Returns:
+        model: nn.Layer. Specific `PPHGNetV2_B5` model depends on args.
+    """
+    stage_config = {
+        # in_channels, mid_channels, out_channels, num_blocks, is_downsample, light_block, kernel_size, layer_num
+        "stage1": [64, 64, 128, 1, False, False, 3, 6],
+        "stage2": [128, 128, 512, 2, True, False, 3, 6],
+        "stage3": [512, 256, 1024, 5, True, True, 5, 6],
+        "stage4": [1024, 512, 2048, 2, True, True, 5, 6],
+    }
+
+    model = PPHGNetV2(
+        stem_channels=[3, 32, 64], stage_config=stage_config, use_lab=False, **kwargs
+    )
+    return model
+
+
+def PPHGNetV2_B6(pretrained=False, use_ssld=False, **kwargs):
+    """
+    PPHGNetV2_B6
+    Args:
+        pretrained (bool/str): If `True` load pretrained parameters, `False` otherwise.
+                    If str, means the path of the pretrained model.
+        use_ssld (bool) Whether using ssld pretrained model when pretrained is True.
+    Returns:
+        model: nn.Layer. Specific `PPHGNetV2_B6` model depends on args.
+    """
+    stage_config = {
+        # in_channels, mid_channels, out_channels, num_blocks, is_downsample, light_block, kernel_size, layer_num
+        "stage1": [96, 96, 192, 2, False, False, 3, 6],
+        "stage2": [192, 192, 512, 3, True, False, 3, 6],
+        "stage3": [512, 384, 1024, 6, True, True, 5, 6],
+        "stage4": [1024, 768, 2048, 3, True, True, 5, 6],
+    }
+
+    model = PPHGNetV2(
+        stem_channels=[3, 48, 96], stage_config=stage_config, use_lab=False, **kwargs
+    )
+    return model
+
+
+class PPHGNetV2_B4_Formula(nn.Layer):
+    """
+    PPHGNetV2_B4_Formula
+    Args:
+        in_channels (int): Number of input channels. Default is 3 (for RGB images).
+        class_num (int): Number of classes for classification. Default is 1000.
+    Returns:
+        model: nn.Layer. Specific `PPHGNetV2_B4` model with defined architecture.
+    """
+
+    def __init__(self, in_channels=3, class_num=1000):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = 2048
+        stage_config = {
+            # in_channels, mid_channels, out_channels, num_blocks, is_downsample, light_block, kernel_size, layer_num
+            "stage1": [48, 48, 128, 1, False, False, 3, 6, 2],
+            "stage2": [128, 96, 512, 1, True, False, 3, 6, 2],
+            "stage3": [512, 192, 1024, 3, True, True, 5, 6, 2],
+            "stage4": [1024, 384, 2048, 1, True, True, 5, 6, 2],
+        }
+
+        self.pphgnet_b4 = PPHGNetV2(
+            stem_channels=[3, 32, 48],
+            stage_config=stage_config,
+            class_num=class_num,
+            use_lab=False,
+        )
+
+    def forward(self, input_data):
+        if self.training:
+            pixel_values, label, attention_mask = input_data
+        else:
+            if isinstance(input_data, list):
+                pixel_values = input_data[0]
+            else:
+                pixel_values = input_data
+        num_channels = pixel_values.shape[1]
+        if num_channels == 1:
+            pixel_values = paddle.repeat_interleave(pixel_values, repeats=3, axis=1)
+        pphgnet_b4_output = self.pphgnet_b4(pixel_values)
+        b, c, h, w = pphgnet_b4_output.shape
+        pphgnet_b4_output = pphgnet_b4_output.reshape([b, c, h * w]).transpose(
+            [0, 2, 1]
+        )
+        pphgnet_b4_output = DonutSwinModelOutput(
+            last_hidden_state=pphgnet_b4_output,
+            pooler_output=None,
+            hidden_states=None,
+            attentions=False,
+            reshaped_hidden_states=None,
+        )
+        if self.training:
+            return pphgnet_b4_output, label, attention_mask
+        else:
+            return pphgnet_b4_output
+
+
+class PPHGNetV2_B6_Formula(nn.Layer):
+    """
+    PPHGNetV2_B6_Formula
+    Args:
+        in_channels (int): Number of input channels. Default is 3 (for RGB images).
+        class_num (int): Number of classes for classification. Default is 1000.
+    Returns:
+        model: nn.Layer. Specific `PPHGNetV2_B6` model with defined architecture.
+    """
+
+    def __init__(self, in_channels=3, class_num=1000):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = 2048
+        stage_config = {
+            # in_channels, mid_channels, out_channels, num_blocks, is_downsample, light_block, kernel_size, layer_num
+            "stage1": [96, 96, 192, 2, False, False, 3, 6, 2],
+            "stage2": [192, 192, 512, 3, True, False, 3, 6, 2],
+            "stage3": [512, 384, 1024, 6, True, True, 5, 6, 2],
+            "stage4": [1024, 768, 2048, 3, True, True, 5, 6, 2],
+        }
+
+        self.pphgnet_b6 = PPHGNetV2(
+            stem_channels=[3, 48, 96],
+            class_num=class_num,
+            stage_config=stage_config,
+            use_lab=False,
+        )
+
+    def forward(self, input_data):
+        if self.training:
+            pixel_values, label, attention_mask = input_data
+        else:
+            if isinstance(input_data, list):
+                pixel_values = input_data[0]
+            else:
+                pixel_values = input_data
+        num_channels = pixel_values.shape[1]
+        if num_channels == 1:
+            pixel_values = paddle.repeat_interleave(pixel_values, repeats=3, axis=1)
+        pphgnet_b6_output = self.pphgnet_b6(pixel_values)
+        b, c, h, w = pphgnet_b6_output.shape
+        pphgnet_b6_output = pphgnet_b6_output.reshape([b, c, h * w]).transpose(
+            [0, 2, 1]
+        )
+        pphgnet_b6_output = DonutSwinModelOutput(
+            last_hidden_state=pphgnet_b6_output,
+            pooler_output=None,
+            hidden_states=None,
+            attentions=False,
+            reshaped_hidden_states=None,
+        )
+        if self.training:
+            return pphgnet_b6_output, label, attention_mask
+        else:
+            return pphgnet_b6_output
diff --git a/ppocr/modeling/backbones/rec_repvit.py b/ppocr/modeling/backbones/rec_repvit.py
new file mode 100644
index 0000000..0ef66ea
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_repvit.py
@@ -0,0 +1,392 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+This code is refer from:
+https://github.com/THU-MIG/RepViT
+"""
+
+import paddle.nn as nn
+import paddle
+from paddle.nn.initializer import TruncatedNormal, Constant, Normal
+
+trunc_normal_ = TruncatedNormal(std=0.02)
+normal_ = Normal
+zeros_ = Constant(value=0.0)
+ones_ = Constant(value=1.0)
+
+
+def _make_divisible(v, divisor, min_value=None):
+    """
+    This function is taken from the original tf repo.
+    It ensures that all layers have a channel number that is divisible by 8
+    It can be seen here:
+    https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/mobilenet.py
+    :param v:
+    :param divisor:
+    :param min_value:
+    :return:
+    """
+    if min_value is None:
+        min_value = divisor
+    new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
+    # Make sure that round down does not go down by more than 10%.
+    if new_v < 0.9 * v:
+        new_v += divisor
+    return new_v
+
+
+# from timm.models.layers import SqueezeExcite
+
+
+def make_divisible(v, divisor=8, min_value=None, round_limit=0.9):
+    min_value = min_value or divisor
+    new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
+    # Make sure that round down does not go down by more than 10%.
+    if new_v < round_limit * v:
+        new_v += divisor
+    return new_v
+
+
+class SEModule(nn.Layer):
+    """SE Module as defined in original SE-Nets with a few additions
+    Additions include:
+        * divisor can be specified to keep channels % div == 0 (default: 8)
+        * reduction channels can be specified directly by arg (if rd_channels is set)
+        * reduction channels can be specified by float rd_ratio (default: 1/16)
+        * global max pooling can be added to the squeeze aggregation
+        * customizable activation, normalization, and gate layer
+    """
+
+    def __init__(
+        self,
+        channels,
+        rd_ratio=1.0 / 16,
+        rd_channels=None,
+        rd_divisor=8,
+        act_layer=nn.ReLU,
+    ):
+        super(SEModule, self).__init__()
+        if not rd_channels:
+            rd_channels = make_divisible(
+                channels * rd_ratio, rd_divisor, round_limit=0.0
+            )
+        self.fc1 = nn.Conv2D(channels, rd_channels, kernel_size=1, bias_attr=True)
+        self.act = act_layer()
+        self.fc2 = nn.Conv2D(rd_channels, channels, kernel_size=1, bias_attr=True)
+
+    def forward(self, x):
+        x_se = x.mean((2, 3), keepdim=True)
+        x_se = self.fc1(x_se)
+        x_se = self.act(x_se)
+        x_se = self.fc2(x_se)
+        return x * nn.functional.sigmoid(x_se)
+
+
+class Conv2D_BN(nn.Sequential):
+    def __init__(
+        self,
+        a,
+        b,
+        ks=1,
+        stride=1,
+        pad=0,
+        dilation=1,
+        groups=1,
+        bn_weight_init=1,
+        resolution=-10000,
+    ):
+        super().__init__()
+        self.add_sublayer(
+            "c", nn.Conv2D(a, b, ks, stride, pad, dilation, groups, bias_attr=False)
+        )
+        self.add_sublayer("bn", nn.BatchNorm2D(b))
+        if bn_weight_init == 1:
+            ones_(self.bn.weight)
+        else:
+            zeros_(self.bn.weight)
+        zeros_(self.bn.bias)
+
+    @paddle.no_grad()
+    def fuse(self):
+        c, bn = self.c, self.bn
+        w = bn.weight / (bn._variance + bn._epsilon) ** 0.5
+        w = c.weight * w[:, None, None, None]
+        b = bn.bias - bn._mean * bn.weight / (bn._variance + bn._epsilon) ** 0.5
+        m = nn.Conv2D(
+            w.shape[1] * self.c._groups,
+            w.shape[0],
+            w.shape[2:],
+            stride=self.c._stride,
+            padding=self.c._padding,
+            dilation=self.c._dilation,
+            groups=self.c._groups,
+        )
+        m.weight.set_value(w)
+        m.bias.set_value(b)
+        return m
+
+
+class Residual(nn.Layer):
+    def __init__(self, m, drop=0.0):
+        super().__init__()
+        self.m = m
+        self.drop = drop
+
+    def forward(self, x):
+        if self.training and self.drop > 0:
+            return (
+                x
+                + self.m(x)
+                * paddle.rand(x.size(0), 1, 1, 1)
+                .ge_(self.drop)
+                .div(1 - self.drop)
+                .detach()
+            )
+        else:
+            return x + self.m(x)
+
+    @paddle.no_grad()
+    def fuse(self):
+        if isinstance(self.m, Conv2D_BN):
+            m = self.m.fuse()
+            assert m._groups == m.in_channels
+            identity = paddle.ones([m.weight.shape[0], m.weight.shape[1], 1, 1])
+            identity = nn.functional.pad(identity, [1, 1, 1, 1])
+            m.weight += identity
+            return m
+        elif isinstance(self.m, nn.Conv2D):
+            m = self.m
+            assert m._groups != m.in_channels
+            identity = paddle.ones([m.weight.shape[0], m.weight.shape[1], 1, 1])
+            identity = nn.functional.pad(identity, [1, 1, 1, 1])
+            m.weight += identity
+            return m
+        else:
+            return self
+
+
+class RepVGGDW(nn.Layer):
+    def __init__(self, ed) -> None:
+        super().__init__()
+        self.conv = Conv2D_BN(ed, ed, 3, 1, 1, groups=ed)
+        self.conv1 = nn.Conv2D(ed, ed, 1, 1, 0, groups=ed)
+        self.dim = ed
+        self.bn = nn.BatchNorm2D(ed)
+
+    def forward(self, x):
+        return self.bn((self.conv(x) + self.conv1(x)) + x)
+
+    @paddle.no_grad()
+    def fuse(self):
+        conv = self.conv.fuse()
+        conv1 = self.conv1
+
+        conv_w = conv.weight
+        conv_b = conv.bias
+        conv1_w = conv1.weight
+        conv1_b = conv1.bias
+
+        conv1_w = nn.functional.pad(conv1_w, [1, 1, 1, 1])
+
+        identity = nn.functional.pad(
+            paddle.ones([conv1_w.shape[0], conv1_w.shape[1], 1, 1]), [1, 1, 1, 1]
+        )
+
+        final_conv_w = conv_w + conv1_w + identity
+        final_conv_b = conv_b + conv1_b
+
+        conv.weight.set_value(final_conv_w)
+        conv.bias.set_value(final_conv_b)
+
+        bn = self.bn
+        w = bn.weight / (bn._variance + bn._epsilon) ** 0.5
+        w = conv.weight * w[:, None, None, None]
+        b = (
+            bn.bias
+            + (conv.bias - bn._mean) * bn.weight / (bn._variance + bn._epsilon) ** 0.5
+        )
+        conv.weight.set_value(w)
+        conv.bias.set_value(b)
+        return conv
+
+
+class RepViTBlock(nn.Layer):
+    def __init__(self, inp, hidden_dim, oup, kernel_size, stride, use_se, use_hs):
+        super(RepViTBlock, self).__init__()
+
+        self.identity = stride == 1 and inp == oup
+        assert hidden_dim == 2 * inp
+
+        if stride != 1:
+            self.token_mixer = nn.Sequential(
+                Conv2D_BN(
+                    inp, inp, kernel_size, stride, (kernel_size - 1) // 2, groups=inp
+                ),
+                SEModule(inp, 0.25) if use_se else nn.Identity(),
+                Conv2D_BN(inp, oup, ks=1, stride=1, pad=0),
+            )
+            self.channel_mixer = Residual(
+                nn.Sequential(
+                    # pw
+                    Conv2D_BN(oup, 2 * oup, 1, 1, 0),
+                    nn.GELU() if use_hs else nn.GELU(),
+                    # pw-linear
+                    Conv2D_BN(2 * oup, oup, 1, 1, 0, bn_weight_init=0),
+                )
+            )
+        else:
+            assert self.identity
+            self.token_mixer = nn.Sequential(
+                RepVGGDW(inp),
+                SEModule(inp, 0.25) if use_se else nn.Identity(),
+            )
+            self.channel_mixer = Residual(
+                nn.Sequential(
+                    # pw
+                    Conv2D_BN(inp, hidden_dim, 1, 1, 0),
+                    nn.GELU() if use_hs else nn.GELU(),
+                    # pw-linear
+                    Conv2D_BN(hidden_dim, oup, 1, 1, 0, bn_weight_init=0),
+                )
+            )
+
+    def forward(self, x):
+        return self.channel_mixer(self.token_mixer(x))
+
+
+class RepViT(nn.Layer):
+    def __init__(self, cfgs, in_channels=3, out_indices=None):
+        super(RepViT, self).__init__()
+        # setting of inverted residual blocks
+        self.cfgs = cfgs
+
+        # building first layer
+        input_channel = self.cfgs[0][2]
+        patch_embed = nn.Sequential(
+            Conv2D_BN(in_channels, input_channel // 2, 3, 2, 1),
+            nn.GELU(),
+            Conv2D_BN(input_channel // 2, input_channel, 3, 2, 1),
+        )
+        layers = [patch_embed]
+        # building inverted residual blocks
+        block = RepViTBlock
+        for k, t, c, use_se, use_hs, s in self.cfgs:
+            output_channel = _make_divisible(c, 8)
+            exp_size = _make_divisible(input_channel * t, 8)
+            layers.append(
+                block(input_channel, exp_size, output_channel, k, s, use_se, use_hs)
+            )
+            input_channel = output_channel
+        self.features = nn.LayerList(layers)
+        self.out_indices = out_indices
+        if out_indices is not None:
+            self.out_channels = [self.cfgs[ids - 1][2] for ids in out_indices]
+        else:
+            self.out_channels = self.cfgs[-1][2]
+        self.is_repped = False
+
+    def forward(self, x):
+        if self.out_indices is not None:
+            return self.forward_det(x)
+        return self.forward_rec(x)
+
+    def forward_det(self, x):
+        outs = []
+        for i, f in enumerate(self.features):
+            x = f(x)
+            if i in self.out_indices:
+                outs.append(x)
+        return outs
+
+    def forward_rec(self, x):
+        for f in self.features:
+            x = f(x)
+        h = x.shape[2]
+        x = nn.functional.avg_pool2d(x, [h, 2])
+        return x
+
+    @paddle.no_grad()
+    def rep(self):
+        """Fuse all Conv2D_BN, RepVGGDW, Residual branches for deployment."""
+        if self.is_repped:
+            return
+        # Snapshot the list first; track replaced prefixes to skip orphaned
+        # children (e.g. Conv2D_BN inside a RepVGGDW that was already fused).
+        fusable = [
+            (name, sublayer)
+            for name, sublayer in self.named_sublayers()
+            if isinstance(sublayer, (Conv2D_BN, RepVGGDW, Residual))
+        ]
+        replaced_prefixes = []
+        for name, sublayer in fusable:
+            if any(name.startswith(pfx + ".") for pfx in replaced_prefixes):
+                continue
+            parts = name.rsplit(".", 1)
+            if len(parts) == 2:
+                parent = self
+                for p in parts[0].split("."):
+                    parent = getattr(parent, p)
+                fused = sublayer.fuse()
+                setattr(parent, parts[1], fused)
+                # Only skip children when the module was truly replaced
+                if fused is not sublayer:
+                    replaced_prefixes.append(name)
+        self.is_repped = True
+
+
+def RepSVTR(in_channels=3):
+    """
+    Constructs a MobileNetV3-Large model
+    """
+    # k, t, c, SE, HS, s
+    cfgs = [
+        [3, 2, 96, 1, 0, 1],
+        [3, 2, 96, 0, 0, 1],
+        [3, 2, 96, 0, 0, 1],
+        [3, 2, 192, 0, 1, (2, 1)],
+        [3, 2, 192, 1, 1, 1],
+        [3, 2, 192, 0, 1, 1],
+        [3, 2, 192, 1, 1, 1],
+        [3, 2, 192, 0, 1, 1],
+        [3, 2, 192, 1, 1, 1],
+        [3, 2, 192, 0, 1, 1],
+        [3, 2, 384, 0, 1, (2, 1)],
+        [3, 2, 384, 1, 1, 1],
+        [3, 2, 384, 0, 1, 1],
+    ]
+    return RepViT(cfgs, in_channels=in_channels)
+
+
+def RepSVTR_det(in_channels=3, out_indices=[2, 5, 10, 13]):
+    """
+    Constructs a MobileNetV3-Large model
+    """
+    # k, t, c, SE, HS, s
+    cfgs = [
+        [3, 2, 48, 1, 0, 1],
+        [3, 2, 48, 0, 0, 1],
+        [3, 2, 96, 0, 0, 2],
+        [3, 2, 96, 1, 0, 1],
+        [3, 2, 96, 0, 0, 1],
+        [3, 2, 192, 0, 1, 2],
+        [3, 2, 192, 1, 1, 1],
+        [3, 2, 192, 0, 1, 1],
+        [3, 2, 192, 1, 1, 1],
+        [3, 2, 192, 0, 1, 1],
+        [3, 2, 384, 0, 1, 2],
+        [3, 2, 384, 1, 1, 1],
+        [3, 2, 384, 0, 1, 1],
+    ]
+    return RepViT(cfgs, in_channels=in_channels, out_indices=out_indices)
diff --git a/ppocr/modeling/backbones/rec_resnet_31.py b/ppocr/modeling/backbones/rec_resnet_31.py
new file mode 100644
index 0000000..c4aa4fc
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_resnet_31.py
@@ -0,0 +1,318 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textrecog/layers/conv_layer.py
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textrecog/backbones/resnet31_ocr.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import ParamAttr
+import paddle.nn as nn
+import paddle.nn.functional as F
+import numpy as np
+
+__all__ = ["ResNet31"]
+
+
+def conv3x3(in_channel, out_channel, stride=1, conv_weight_attr=None):
+    return nn.Conv2D(
+        in_channel,
+        out_channel,
+        kernel_size=3,
+        stride=stride,
+        padding=1,
+        weight_attr=conv_weight_attr,
+        bias_attr=False,
+    )
+
+
+class BasicBlock(nn.Layer):
+    expansion = 1
+
+    def __init__(
+        self,
+        in_channels,
+        channels,
+        stride=1,
+        downsample=False,
+        conv_weight_attr=None,
+        bn_weight_attr=None,
+    ):
+        super().__init__()
+        self.conv1 = conv3x3(
+            in_channels, channels, stride, conv_weight_attr=conv_weight_attr
+        )
+        self.bn1 = nn.BatchNorm2D(channels, weight_attr=bn_weight_attr)
+        self.relu = nn.ReLU()
+        self.conv2 = conv3x3(channels, channels, conv_weight_attr=conv_weight_attr)
+        self.bn2 = nn.BatchNorm2D(channels, weight_attr=bn_weight_attr)
+        self.downsample = downsample
+        if downsample:
+            self.downsample = nn.Sequential(
+                nn.Conv2D(
+                    in_channels,
+                    channels * self.expansion,
+                    1,
+                    stride,
+                    weight_attr=conv_weight_attr,
+                    bias_attr=False,
+                ),
+                nn.BatchNorm2D(channels * self.expansion, weight_attr=bn_weight_attr),
+            )
+        else:
+            self.downsample = nn.Sequential()
+        self.stride = stride
+
+    def forward(self, x):
+        residual = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+
+        if self.downsample:
+            residual = self.downsample(x)
+
+        out += residual
+        out = self.relu(out)
+
+        return out
+
+
+class ResNet31(nn.Layer):
+    """
+    Args:
+        in_channels (int): Number of channels of input image tensor.
+        layers (list[int]): List of BasicBlock number for each stage.
+        channels (list[int]): List of out_channels of Conv2d layer.
+        out_indices (None | Sequence[int]): Indices of output stages.
+        last_stage_pool (bool): If True, add `MaxPool2d` layer to last stage.
+        init_type (None | str): the config to control the initialization.
+    """
+
+    def __init__(
+        self,
+        in_channels=3,
+        layers=[1, 2, 5, 3],
+        channels=[64, 128, 256, 256, 512, 512, 512],
+        out_indices=None,
+        last_stage_pool=False,
+        init_type=None,
+    ):
+        super(ResNet31, self).__init__()
+        assert isinstance(in_channels, int)
+        assert isinstance(last_stage_pool, bool)
+
+        self.out_indices = out_indices
+        self.last_stage_pool = last_stage_pool
+
+        conv_weight_attr = None
+        bn_weight_attr = None
+
+        if init_type is not None:
+            support_dict = ["KaimingNormal"]
+            assert init_type in support_dict, Exception(
+                "resnet31 only support {}".format(support_dict)
+            )
+            conv_weight_attr = nn.initializer.KaimingNormal()
+            bn_weight_attr = ParamAttr(
+                initializer=nn.initializer.Uniform(), learning_rate=1
+            )
+
+        # conv 1 (Conv Conv)
+        self.conv1_1 = nn.Conv2D(
+            in_channels,
+            channels[0],
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            weight_attr=conv_weight_attr,
+        )
+        self.bn1_1 = nn.BatchNorm2D(channels[0], weight_attr=bn_weight_attr)
+        self.relu1_1 = nn.ReLU()
+
+        self.conv1_2 = nn.Conv2D(
+            channels[0],
+            channels[1],
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            weight_attr=conv_weight_attr,
+        )
+        self.bn1_2 = nn.BatchNorm2D(channels[1], weight_attr=bn_weight_attr)
+        self.relu1_2 = nn.ReLU()
+
+        # conv 2 (Max-pooling, Residual block, Conv)
+        self.pool2 = nn.MaxPool2D(kernel_size=2, stride=2, padding=0, ceil_mode=True)
+        self.block2 = self._make_layer(
+            channels[1],
+            channels[2],
+            layers[0],
+            conv_weight_attr=conv_weight_attr,
+            bn_weight_attr=bn_weight_attr,
+        )
+        self.conv2 = nn.Conv2D(
+            channels[2],
+            channels[2],
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            weight_attr=conv_weight_attr,
+        )
+        self.bn2 = nn.BatchNorm2D(channels[2], weight_attr=bn_weight_attr)
+        self.relu2 = nn.ReLU()
+
+        # conv 3 (Max-pooling, Residual block, Conv)
+        self.pool3 = nn.MaxPool2D(kernel_size=2, stride=2, padding=0, ceil_mode=True)
+        self.block3 = self._make_layer(
+            channels[2],
+            channels[3],
+            layers[1],
+            conv_weight_attr=conv_weight_attr,
+            bn_weight_attr=bn_weight_attr,
+        )
+        self.conv3 = nn.Conv2D(
+            channels[3],
+            channels[3],
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            weight_attr=conv_weight_attr,
+        )
+        self.bn3 = nn.BatchNorm2D(channels[3], weight_attr=bn_weight_attr)
+        self.relu3 = nn.ReLU()
+
+        # conv 4 (Max-pooling, Residual block, Conv)
+        self.pool4 = nn.MaxPool2D(
+            kernel_size=(2, 1), stride=(2, 1), padding=0, ceil_mode=True
+        )
+        self.block4 = self._make_layer(
+            channels[3],
+            channels[4],
+            layers[2],
+            conv_weight_attr=conv_weight_attr,
+            bn_weight_attr=bn_weight_attr,
+        )
+        self.conv4 = nn.Conv2D(
+            channels[4],
+            channels[4],
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            weight_attr=conv_weight_attr,
+        )
+        self.bn4 = nn.BatchNorm2D(channels[4], weight_attr=bn_weight_attr)
+        self.relu4 = nn.ReLU()
+
+        # conv 5 ((Max-pooling), Residual block, Conv)
+        self.pool5 = None
+        if self.last_stage_pool:
+            self.pool5 = nn.MaxPool2D(
+                kernel_size=2, stride=2, padding=0, ceil_mode=True
+            )
+        self.block5 = self._make_layer(
+            channels[4],
+            channels[5],
+            layers[3],
+            conv_weight_attr=conv_weight_attr,
+            bn_weight_attr=bn_weight_attr,
+        )
+        self.conv5 = nn.Conv2D(
+            channels[5],
+            channels[5],
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            weight_attr=conv_weight_attr,
+        )
+        self.bn5 = nn.BatchNorm2D(channels[5], weight_attr=bn_weight_attr)
+        self.relu5 = nn.ReLU()
+
+        self.out_channels = channels[-1]
+
+    def _make_layer(
+        self,
+        input_channels,
+        output_channels,
+        blocks,
+        conv_weight_attr=None,
+        bn_weight_attr=None,
+    ):
+        layers = []
+        for _ in range(blocks):
+            downsample = None
+            if input_channels != output_channels:
+                downsample = nn.Sequential(
+                    nn.Conv2D(
+                        input_channels,
+                        output_channels,
+                        kernel_size=1,
+                        stride=1,
+                        weight_attr=conv_weight_attr,
+                        bias_attr=False,
+                    ),
+                    nn.BatchNorm2D(output_channels, weight_attr=bn_weight_attr),
+                )
+
+            layers.append(
+                BasicBlock(
+                    input_channels,
+                    output_channels,
+                    downsample=downsample,
+                    conv_weight_attr=conv_weight_attr,
+                    bn_weight_attr=bn_weight_attr,
+                )
+            )
+            input_channels = output_channels
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        x = self.conv1_1(x)
+        x = self.bn1_1(x)
+        x = self.relu1_1(x)
+
+        x = self.conv1_2(x)
+        x = self.bn1_2(x)
+        x = self.relu1_2(x)
+
+        outs = []
+        for i in range(4):
+            layer_index = i + 2
+            pool_layer = getattr(self, f"pool{layer_index}")
+            block_layer = getattr(self, f"block{layer_index}")
+            conv_layer = getattr(self, f"conv{layer_index}")
+            bn_layer = getattr(self, f"bn{layer_index}")
+            relu_layer = getattr(self, f"relu{layer_index}")
+
+            if pool_layer is not None:
+                x = pool_layer(x)
+            x = block_layer(x)
+            x = conv_layer(x)
+            x = bn_layer(x)
+            x = relu_layer(x)
+
+            outs.append(x)
+
+        if self.out_indices is not None:
+            return tuple([outs[i] for i in self.out_indices])
+
+        return x
diff --git a/ppocr/modeling/backbones/rec_resnet_32.py b/ppocr/modeling/backbones/rec_resnet_32.py
new file mode 100644
index 0000000..63d78d3
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_resnet_32.py
@@ -0,0 +1,305 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/hikopensource/DAVAR-Lab-OCR/davarocr/davar_rcg/models/backbones/ResNet32.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle.nn as nn
+
+__all__ = ["ResNet32"]
+
+conv_weight_attr = nn.initializer.KaimingNormal()
+
+
+class ResNet32(nn.Layer):
+    """
+    Feature Extractor is proposed in  FAN Ref [1]
+
+    Ref [1]: Focusing Attention: Towards Accurate Text Recognition in Neural Images ICCV-2017
+    """
+
+    def __init__(self, in_channels, out_channels=512):
+        """
+
+        Args:
+            in_channels (int): input channel
+            output_channel (int): output channel
+        """
+        super(ResNet32, self).__init__()
+        self.out_channels = out_channels
+        self.ConvNet = ResNet(in_channels, out_channels, BasicBlock, [1, 2, 5, 3])
+
+    def forward(self, inputs):
+        """
+        Args:
+            inputs: input feature
+
+        Returns:
+            output feature
+
+        """
+        return self.ConvNet(inputs)
+
+
+class BasicBlock(nn.Layer):
+    """Res-net Basic Block"""
+
+    expansion = 1
+
+    def __init__(
+        self, inplanes, planes, stride=1, downsample=None, norm_type="BN", **kwargs
+    ):
+        """
+        Args:
+            inplanes (int): input channel
+            planes (int): channels of the middle feature
+            stride (int): stride of the convolution
+            downsample (int): type of the down_sample
+            norm_type (str): type of the normalization
+            **kwargs (None): backup parameter
+        """
+        super(BasicBlock, self).__init__()
+        self.conv1 = self._conv3x3(inplanes, planes)
+        self.bn1 = nn.BatchNorm2D(planes)
+        self.conv2 = self._conv3x3(planes, planes)
+        self.bn2 = nn.BatchNorm2D(planes)
+        self.relu = nn.ReLU()
+        self.downsample = downsample
+        self.stride = stride
+
+    def _conv3x3(self, in_planes, out_planes, stride=1):
+        """
+
+        Args:
+            in_planes (int): input channel
+            out_planes (int): channels of the middle feature
+            stride (int): stride of the convolution
+        Returns:
+            nn.Layer: Conv2D with kernel = 3
+
+        """
+
+        return nn.Conv2D(
+            in_planes,
+            out_planes,
+            kernel_size=3,
+            stride=stride,
+            padding=1,
+            weight_attr=conv_weight_attr,
+            bias_attr=False,
+        )
+
+    def forward(self, x):
+        residual = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+
+        if self.downsample is not None:
+            residual = self.downsample(x)
+        out += residual
+        out = self.relu(out)
+
+        return out
+
+
+class ResNet(nn.Layer):
+    """Res-Net network structure"""
+
+    def __init__(self, input_channel, output_channel, block, layers):
+        """
+
+        Args:
+            input_channel (int): input channel
+            output_channel (int): output channel
+            block (BasicBlock): convolution block
+            layers (list): layers of the block
+        """
+        super(ResNet, self).__init__()
+
+        self.output_channel_block = [
+            int(output_channel / 4),
+            int(output_channel / 2),
+            output_channel,
+            output_channel,
+        ]
+
+        self.inplanes = int(output_channel / 8)
+        self.conv0_1 = nn.Conv2D(
+            input_channel,
+            int(output_channel / 16),
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            weight_attr=conv_weight_attr,
+            bias_attr=False,
+        )
+        self.bn0_1 = nn.BatchNorm2D(int(output_channel / 16))
+        self.conv0_2 = nn.Conv2D(
+            int(output_channel / 16),
+            self.inplanes,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            weight_attr=conv_weight_attr,
+            bias_attr=False,
+        )
+        self.bn0_2 = nn.BatchNorm2D(self.inplanes)
+        self.relu = nn.ReLU()
+
+        self.maxpool1 = nn.MaxPool2D(kernel_size=2, stride=2, padding=0)
+        self.layer1 = self._make_layer(block, self.output_channel_block[0], layers[0])
+        self.conv1 = nn.Conv2D(
+            self.output_channel_block[0],
+            self.output_channel_block[0],
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            weight_attr=conv_weight_attr,
+            bias_attr=False,
+        )
+        self.bn1 = nn.BatchNorm2D(self.output_channel_block[0])
+
+        self.maxpool2 = nn.MaxPool2D(kernel_size=2, stride=2, padding=0)
+        self.layer2 = self._make_layer(
+            block, self.output_channel_block[1], layers[1], stride=1
+        )
+        self.conv2 = nn.Conv2D(
+            self.output_channel_block[1],
+            self.output_channel_block[1],
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            weight_attr=conv_weight_attr,
+            bias_attr=False,
+        )
+        self.bn2 = nn.BatchNorm2D(self.output_channel_block[1])
+
+        self.maxpool3 = nn.MaxPool2D(kernel_size=2, stride=(2, 1), padding=(0, 1))
+        self.layer3 = self._make_layer(
+            block, self.output_channel_block[2], layers[2], stride=1
+        )
+        self.conv3 = nn.Conv2D(
+            self.output_channel_block[2],
+            self.output_channel_block[2],
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            weight_attr=conv_weight_attr,
+            bias_attr=False,
+        )
+        self.bn3 = nn.BatchNorm2D(self.output_channel_block[2])
+
+        self.layer4 = self._make_layer(
+            block, self.output_channel_block[3], layers[3], stride=1
+        )
+        self.conv4_1 = nn.Conv2D(
+            self.output_channel_block[3],
+            self.output_channel_block[3],
+            kernel_size=2,
+            stride=(2, 1),
+            padding=(0, 1),
+            weight_attr=conv_weight_attr,
+            bias_attr=False,
+        )
+        self.bn4_1 = nn.BatchNorm2D(self.output_channel_block[3])
+        self.conv4_2 = nn.Conv2D(
+            self.output_channel_block[3],
+            self.output_channel_block[3],
+            kernel_size=2,
+            stride=1,
+            padding=0,
+            weight_attr=conv_weight_attr,
+            bias_attr=False,
+        )
+        self.bn4_2 = nn.BatchNorm2D(self.output_channel_block[3])
+
+    def _make_layer(self, block, planes, blocks, stride=1):
+        """
+
+        Args:
+            block (block): convolution block
+            planes (int): input channels
+            blocks (list): layers of the block
+            stride (int): stride of the convolution
+
+        Returns:
+            nn.Sequential: the combination of the convolution block
+
+        """
+        downsample = None
+        if stride != 1 or self.inplanes != planes * block.expansion:
+            downsample = nn.Sequential(
+                nn.Conv2D(
+                    self.inplanes,
+                    planes * block.expansion,
+                    kernel_size=1,
+                    stride=stride,
+                    weight_attr=conv_weight_attr,
+                    bias_attr=False,
+                ),
+                nn.BatchNorm2D(planes * block.expansion),
+            )
+
+        layers = list()
+        layers.append(block(self.inplanes, planes, stride, downsample))
+        self.inplanes = planes * block.expansion
+        for _ in range(1, blocks):
+            layers.append(block(self.inplanes, planes))
+
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        x = self.conv0_1(x)
+        x = self.bn0_1(x)
+        x = self.relu(x)
+        x = self.conv0_2(x)
+        x = self.bn0_2(x)
+        x = self.relu(x)
+
+        x = self.maxpool1(x)
+        x = self.layer1(x)
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = self.relu(x)
+
+        x = self.maxpool2(x)
+        x = self.layer2(x)
+        x = self.conv2(x)
+        x = self.bn2(x)
+        x = self.relu(x)
+
+        x = self.maxpool3(x)
+        x = self.layer3(x)
+        x = self.conv3(x)
+        x = self.bn3(x)
+        x = self.relu(x)
+
+        x = self.layer4(x)
+        x = self.conv4_1(x)
+        x = self.bn4_1(x)
+        x = self.relu(x)
+        x = self.conv4_2(x)
+        x = self.bn4_2(x)
+        x = self.relu(x)
+        return x
diff --git a/ppocr/modeling/backbones/rec_resnet_45.py b/ppocr/modeling/backbones/rec_resnet_45.py
new file mode 100644
index 0000000..914d972
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_resnet_45.py
@@ -0,0 +1,150 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/FangShancheng/ABINet/tree/main/modules
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import ParamAttr
+from paddle.nn.initializer import KaimingNormal
+import paddle.nn as nn
+import paddle.nn.functional as F
+import numpy as np
+import math
+
+__all__ = ["ResNet45"]
+
+
+def conv1x1(in_planes, out_planes, stride=1):
+    return nn.Conv2D(
+        in_planes,
+        out_planes,
+        kernel_size=1,
+        stride=1,
+        weight_attr=ParamAttr(initializer=KaimingNormal()),
+        bias_attr=False,
+    )
+
+
+def conv3x3(in_channel, out_channel, stride=1):
+    return nn.Conv2D(
+        in_channel,
+        out_channel,
+        kernel_size=3,
+        stride=stride,
+        padding=1,
+        weight_attr=ParamAttr(initializer=KaimingNormal()),
+        bias_attr=False,
+    )
+
+
+class BasicBlock(nn.Layer):
+    expansion = 1
+
+    def __init__(self, in_channels, channels, stride=1, downsample=None):
+        super().__init__()
+        self.conv1 = conv1x1(in_channels, channels)
+        self.bn1 = nn.BatchNorm2D(channels)
+        self.relu = nn.ReLU()
+        self.conv2 = conv3x3(channels, channels, stride)
+        self.bn2 = nn.BatchNorm2D(channels)
+        self.downsample = downsample
+        self.stride = stride
+
+    def forward(self, x):
+        residual = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+
+        if self.downsample is not None:
+            residual = self.downsample(x)
+        out += residual
+        out = self.relu(out)
+
+        return out
+
+
+class ResNet45(nn.Layer):
+    def __init__(
+        self,
+        in_channels=3,
+        block=BasicBlock,
+        layers=[3, 4, 6, 6, 3],
+        strides=[2, 1, 2, 1, 1],
+    ):
+        self.inplanes = 32
+        super(ResNet45, self).__init__()
+        self.conv1 = nn.Conv2D(
+            in_channels,
+            32,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            weight_attr=ParamAttr(initializer=KaimingNormal()),
+            bias_attr=False,
+        )
+        self.bn1 = nn.BatchNorm2D(32)
+        self.relu = nn.ReLU()
+
+        self.layer1 = self._make_layer(block, 32, layers[0], stride=strides[0])
+        self.layer2 = self._make_layer(block, 64, layers[1], stride=strides[1])
+        self.layer3 = self._make_layer(block, 128, layers[2], stride=strides[2])
+        self.layer4 = self._make_layer(block, 256, layers[3], stride=strides[3])
+        self.layer5 = self._make_layer(block, 512, layers[4], stride=strides[4])
+        self.out_channels = 512
+
+    def _make_layer(self, block, planes, blocks, stride=1):
+        downsample = None
+        if stride != 1 or self.inplanes != planes * block.expansion:
+            # downsample = True
+            downsample = nn.Sequential(
+                nn.Conv2D(
+                    self.inplanes,
+                    planes * block.expansion,
+                    kernel_size=1,
+                    stride=stride,
+                    weight_attr=ParamAttr(initializer=KaimingNormal()),
+                    bias_attr=False,
+                ),
+                nn.BatchNorm2D(planes * block.expansion),
+            )
+
+        layers = []
+        layers.append(block(self.inplanes, planes, stride, downsample))
+        self.inplanes = planes * block.expansion
+        for i in range(1, blocks):
+            layers.append(block(self.inplanes, planes))
+
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = self.relu(x)
+        x = self.layer1(x)
+        x = self.layer2(x)
+        x = self.layer3(x)
+        x = self.layer4(x)
+        x = self.layer5(x)
+        return x
diff --git a/ppocr/modeling/backbones/rec_resnet_aster.py b/ppocr/modeling/backbones/rec_resnet_aster.py
new file mode 100644
index 0000000..9b5a15e
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_resnet_aster.py
@@ -0,0 +1,141 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/ayumiymk/aster.pytorch/blob/master/lib/models/resnet_aster.py
+"""
+import paddle
+import paddle.nn as nn
+
+import sys
+import math
+
+
+def conv3x3(in_planes, out_planes, stride=1):
+    """3x3 convolution with padding"""
+    return nn.Conv2D(
+        in_planes, out_planes, kernel_size=3, stride=stride, padding=1, bias_attr=False
+    )
+
+
+def conv1x1(in_planes, out_planes, stride=1):
+    """1x1 convolution"""
+    return nn.Conv2D(
+        in_planes, out_planes, kernel_size=1, stride=stride, bias_attr=False
+    )
+
+
+def get_sinusoid_encoding(n_position, feat_dim, wave_length=10000):
+    # [n_position]
+    positions = paddle.arange(0, n_position)
+    # [feat_dim]
+    dim_range = paddle.arange(0, feat_dim)
+    dim_range = paddle.pow(wave_length, 2 * (dim_range // 2) / feat_dim)
+    # [n_position, feat_dim]
+    angles = paddle.unsqueeze(positions, axis=1) / paddle.unsqueeze(dim_range, axis=0)
+    angles = paddle.cast(angles, "float32")
+    angles[:, 0::2] = paddle.sin(angles[:, 0::2])
+    angles[:, 1::2] = paddle.cos(angles[:, 1::2])
+    return angles
+
+
+class AsterBlock(nn.Layer):
+    def __init__(self, inplanes, planes, stride=1, downsample=None):
+        super(AsterBlock, self).__init__()
+        self.conv1 = conv1x1(inplanes, planes, stride)
+        self.bn1 = nn.BatchNorm2D(planes)
+        self.relu = nn.ReLU()
+        self.conv2 = conv3x3(planes, planes)
+        self.bn2 = nn.BatchNorm2D(planes)
+        self.downsample = downsample
+        self.stride = stride
+
+    def forward(self, x):
+        residual = x
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+        out = self.conv2(out)
+        out = self.bn2(out)
+
+        if self.downsample is not None:
+            residual = self.downsample(x)
+        out += residual
+        out = self.relu(out)
+        return out
+
+
+class ResNet_ASTER(nn.Layer):
+    """For aster or crnn"""
+
+    def __init__(self, with_lstm=True, n_group=1, in_channels=3):
+        super(ResNet_ASTER, self).__init__()
+        self.with_lstm = with_lstm
+        self.n_group = n_group
+
+        self.layer0 = nn.Sequential(
+            nn.Conv2D(
+                in_channels,
+                32,
+                kernel_size=(3, 3),
+                stride=1,
+                padding=1,
+                bias_attr=False,
+            ),
+            nn.BatchNorm2D(32),
+            nn.ReLU(),
+        )
+
+        self.inplanes = 32
+        self.layer1 = self._make_layer(32, 3, [2, 2])  # [16, 50]
+        self.layer2 = self._make_layer(64, 4, [2, 2])  # [8, 25]
+        self.layer3 = self._make_layer(128, 6, [2, 1])  # [4, 25]
+        self.layer4 = self._make_layer(256, 6, [2, 1])  # [2, 25]
+        self.layer5 = self._make_layer(512, 3, [2, 1])  # [1, 25]
+
+        if with_lstm:
+            self.rnn = nn.LSTM(512, 256, direction="bidirect", num_layers=2)
+            self.out_channels = 2 * 256
+        else:
+            self.out_channels = 512
+
+    def _make_layer(self, planes, blocks, stride):
+        downsample = None
+        if stride != [1, 1] or self.inplanes != planes:
+            downsample = nn.Sequential(
+                conv1x1(self.inplanes, planes, stride), nn.BatchNorm2D(planes)
+            )
+
+        layers = []
+        layers.append(AsterBlock(self.inplanes, planes, stride, downsample))
+        self.inplanes = planes
+        for _ in range(1, blocks):
+            layers.append(AsterBlock(self.inplanes, planes))
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        x0 = self.layer0(x)
+        x1 = self.layer1(x0)
+        x2 = self.layer2(x1)
+        x3 = self.layer3(x2)
+        x4 = self.layer4(x3)
+        x5 = self.layer5(x4)
+
+        cnn_feat = x5.squeeze(2)  # [N, c, w]
+        cnn_feat = paddle.transpose(cnn_feat, perm=[0, 2, 1])
+        if self.with_lstm:
+            rnn_feat, _ = self.rnn(cnn_feat)
+            return rnn_feat
+        else:
+            return cnn_feat
diff --git a/ppocr/modeling/backbones/rec_resnet_fpn.py b/ppocr/modeling/backbones/rec_resnet_fpn.py
new file mode 100644
index 0000000..d259f1d
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_resnet_fpn.py
@@ -0,0 +1,317 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from paddle import nn, ParamAttr
+from paddle.nn import functional as F
+import paddle
+import numpy as np
+
+__all__ = ["ResNetFPN"]
+
+
+class ResNetFPN(nn.Layer):
+    def __init__(self, in_channels=1, layers=50, **kwargs):
+        super(ResNetFPN, self).__init__()
+        supported_layers = {
+            18: {"depth": [2, 2, 2, 2], "block_class": BasicBlock},
+            34: {"depth": [3, 4, 6, 3], "block_class": BasicBlock},
+            50: {"depth": [3, 4, 6, 3], "block_class": BottleneckBlock},
+            101: {"depth": [3, 4, 23, 3], "block_class": BottleneckBlock},
+            152: {"depth": [3, 8, 36, 3], "block_class": BottleneckBlock},
+        }
+        stride_list = [(2, 2), (2, 2), (1, 1), (1, 1)]
+        num_filters = [64, 128, 256, 512]
+        self.depth = supported_layers[layers]["depth"]
+        self.F = []
+        self.conv = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=64,
+            kernel_size=7,
+            stride=2,
+            act="relu",
+            name="conv1",
+        )
+        self.block_list = []
+        in_ch = 64
+        if layers >= 50:
+            for block in range(len(self.depth)):
+                for i in range(self.depth[block]):
+                    if layers in [101, 152] and block == 2:
+                        if i == 0:
+                            conv_name = "res" + str(block + 2) + "a"
+                        else:
+                            conv_name = "res" + str(block + 2) + "b" + str(i)
+                    else:
+                        conv_name = "res" + str(block + 2) + chr(97 + i)
+                    block_list = self.add_sublayer(
+                        "bottleneckBlock_{}_{}".format(block, i),
+                        BottleneckBlock(
+                            in_channels=in_ch,
+                            out_channels=num_filters[block],
+                            stride=stride_list[block] if i == 0 else 1,
+                            name=conv_name,
+                        ),
+                    )
+                    in_ch = num_filters[block] * 4
+                    self.block_list.append(block_list)
+                self.F.append(block_list)
+        else:
+            for block in range(len(self.depth)):
+                for i in range(self.depth[block]):
+                    conv_name = "res" + str(block + 2) + chr(97 + i)
+                    if i == 0 and block != 0:
+                        stride = (2, 1)
+                    else:
+                        stride = (1, 1)
+                    basic_block = self.add_sublayer(
+                        conv_name,
+                        BasicBlock(
+                            in_channels=in_ch,
+                            out_channels=num_filters[block],
+                            stride=stride_list[block] if i == 0 else 1,
+                            is_first=block == i == 0,
+                            name=conv_name,
+                        ),
+                    )
+                    in_ch = basic_block.out_channels
+                    self.block_list.append(basic_block)
+        out_ch_list = [in_ch // 4, in_ch // 2, in_ch]
+        self.base_block = []
+        self.conv_trans = []
+        self.bn_block = []
+        for i in [-2, -3]:
+            in_channels = out_ch_list[i + 1] + out_ch_list[i]
+
+            self.base_block.append(
+                self.add_sublayer(
+                    "F_{}_base_block_0".format(i),
+                    nn.Conv2D(
+                        in_channels=in_channels,
+                        out_channels=out_ch_list[i],
+                        kernel_size=1,
+                        weight_attr=ParamAttr(trainable=True),
+                        bias_attr=ParamAttr(trainable=True),
+                    ),
+                )
+            )
+            self.base_block.append(
+                self.add_sublayer(
+                    "F_{}_base_block_1".format(i),
+                    nn.Conv2D(
+                        in_channels=out_ch_list[i],
+                        out_channels=out_ch_list[i],
+                        kernel_size=3,
+                        padding=1,
+                        weight_attr=ParamAttr(trainable=True),
+                        bias_attr=ParamAttr(trainable=True),
+                    ),
+                )
+            )
+            self.base_block.append(
+                self.add_sublayer(
+                    "F_{}_base_block_2".format(i),
+                    nn.BatchNorm(
+                        num_channels=out_ch_list[i],
+                        act="relu",
+                        param_attr=ParamAttr(trainable=True),
+                        bias_attr=ParamAttr(trainable=True),
+                    ),
+                )
+            )
+        self.base_block.append(
+            self.add_sublayer(
+                "F_{}_base_block_3".format(i),
+                nn.Conv2D(
+                    in_channels=out_ch_list[i],
+                    out_channels=512,
+                    kernel_size=1,
+                    bias_attr=ParamAttr(trainable=True),
+                    weight_attr=ParamAttr(trainable=True),
+                ),
+            )
+        )
+        self.out_channels = 512
+
+    def __call__(self, x):
+        x = self.conv(x)
+        fpn_list = []
+        F = []
+        for i in range(len(self.depth)):
+            fpn_list.append(np.sum(self.depth[: i + 1]))
+
+        for i, block in enumerate(self.block_list):
+            x = block(x)
+            for number in fpn_list:
+                if i + 1 == number:
+                    F.append(x)
+        base = F[-1]
+
+        j = 0
+        for i, block in enumerate(self.base_block):
+            if i % 3 == 0 and i < 6:
+                j = j + 1
+                b, c, w, h = F[-j - 1].shape
+                if [w, h] == list(base.shape[2:]):
+                    base = base
+                else:
+                    base = self.conv_trans[j - 1](base)
+                    base = self.bn_block[j - 1](base)
+                base = paddle.concat([base, F[-j - 1]], axis=1)
+            base = block(base)
+        return base
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride=1,
+        groups=1,
+        act=None,
+        name=None,
+    ):
+        super(ConvBNLayer, self).__init__()
+        self.conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=2 if stride == (1, 1) else kernel_size,
+            dilation=2 if stride == (1, 1) else 1,
+            stride=stride,
+            padding=(kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(name=name + ".conv2d.output.1.w_0"),
+            bias_attr=False,
+        )
+
+        if name == "conv1":
+            bn_name = "bn_" + name
+        else:
+            bn_name = "bn" + name[3:]
+        self.bn = nn.BatchNorm(
+            num_channels=out_channels,
+            act=act,
+            param_attr=ParamAttr(name=name + ".output.1.w_0"),
+            bias_attr=ParamAttr(name=name + ".output.1.b_0"),
+            moving_mean_name=bn_name + "_mean",
+            moving_variance_name=bn_name + "_variance",
+        )
+
+    def __call__(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        return x
+
+
+class ShortCut(nn.Layer):
+    def __init__(self, in_channels, out_channels, stride, name, is_first=False):
+        super(ShortCut, self).__init__()
+        self.use_conv = True
+
+        if in_channels != out_channels or stride != 1 or is_first == True:
+            if stride == (1, 1):
+                self.conv = ConvBNLayer(in_channels, out_channels, 1, 1, name=name)
+            else:  # stride==(2,2)
+                self.conv = ConvBNLayer(in_channels, out_channels, 1, stride, name=name)
+        else:
+            self.use_conv = False
+
+    def forward(self, x):
+        if self.use_conv:
+            x = self.conv(x)
+        return x
+
+
+class BottleneckBlock(nn.Layer):
+    def __init__(self, in_channels, out_channels, stride, name):
+        super(BottleneckBlock, self).__init__()
+        self.conv0 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=1,
+            act="relu",
+            name=name + "_branch2a",
+        )
+        self.conv1 = ConvBNLayer(
+            in_channels=out_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            stride=stride,
+            act="relu",
+            name=name + "_branch2b",
+        )
+
+        self.conv2 = ConvBNLayer(
+            in_channels=out_channels,
+            out_channels=out_channels * 4,
+            kernel_size=1,
+            act=None,
+            name=name + "_branch2c",
+        )
+
+        self.short = ShortCut(
+            in_channels=in_channels,
+            out_channels=out_channels * 4,
+            stride=stride,
+            is_first=False,
+            name=name + "_branch1",
+        )
+        self.out_channels = out_channels * 4
+
+    def forward(self, x):
+        y = self.conv0(x)
+        y = self.conv1(y)
+        y = self.conv2(y)
+        y = y + self.short(x)
+        y = F.relu(y)
+        return y
+
+
+class BasicBlock(nn.Layer):
+    def __init__(self, in_channels, out_channels, stride, name, is_first):
+        super(BasicBlock, self).__init__()
+        self.conv0 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            act="relu",
+            stride=stride,
+            name=name + "_branch2a",
+        )
+        self.conv1 = ConvBNLayer(
+            in_channels=out_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            act=None,
+            name=name + "_branch2b",
+        )
+        self.short = ShortCut(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            stride=stride,
+            is_first=is_first,
+            name=name + "_branch1",
+        )
+        self.out_channels = out_channels
+
+    def forward(self, x):
+        y = self.conv0(x)
+        y = self.conv1(y)
+        y = y + self.short(x)
+        return F.relu(y)
diff --git a/ppocr/modeling/backbones/rec_resnet_rfl.py b/ppocr/modeling/backbones/rec_resnet_rfl.py
new file mode 100644
index 0000000..4742ef8
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_resnet_rfl.py
@@ -0,0 +1,359 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/hikopensource/DAVAR-Lab-OCR/blob/main/davarocr/davar_rcg/models/backbones/ResNetRFL.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+import paddle.nn as nn
+
+from paddle.nn.initializer import TruncatedNormal, Constant, Normal, KaimingNormal
+
+kaiming_init_ = KaimingNormal()
+zeros_ = Constant(value=0.0)
+ones_ = Constant(value=1.0)
+
+
+class BasicBlock(nn.Layer):
+    """Res-net Basic Block"""
+
+    expansion = 1
+
+    def __init__(
+        self, inplanes, planes, stride=1, downsample=None, norm_type="BN", **kwargs
+    ):
+        """
+        Args:
+            inplanes (int): input channel
+            planes (int): channels of the middle feature
+            stride (int): stride of the convolution
+            downsample (int): type of the down_sample
+            norm_type (str): type of the normalization
+            **kwargs (None): backup parameter
+        """
+        super(BasicBlock, self).__init__()
+        self.conv1 = self._conv3x3(inplanes, planes)
+        self.bn1 = nn.BatchNorm(planes)
+        self.conv2 = self._conv3x3(planes, planes)
+        self.bn2 = nn.BatchNorm(planes)
+        self.relu = nn.ReLU()
+        self.downsample = downsample
+        self.stride = stride
+
+    def _conv3x3(self, in_planes, out_planes, stride=1):
+        return nn.Conv2D(
+            in_planes,
+            out_planes,
+            kernel_size=3,
+            stride=stride,
+            padding=1,
+            bias_attr=False,
+        )
+
+    def forward(self, x):
+        residual = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+
+        if self.downsample is not None:
+            residual = self.downsample(x)
+        out += residual
+        out = self.relu(out)
+
+        return out
+
+
+class ResNetRFL(nn.Layer):
+    def __init__(self, in_channels, out_channels=512, use_cnt=True, use_seq=True):
+        """
+
+        Args:
+            in_channels (int): input channel
+            out_channels (int): output channel
+        """
+        super(ResNetRFL, self).__init__()
+        assert use_cnt or use_seq
+        self.use_cnt, self.use_seq = use_cnt, use_seq
+        self.backbone = RFLBase(in_channels)
+
+        self.out_channels = out_channels
+        self.out_channels_block = [
+            int(self.out_channels / 4),
+            int(self.out_channels / 2),
+            self.out_channels,
+            self.out_channels,
+        ]
+        block = BasicBlock
+        layers = [1, 2, 5, 3]
+        self.inplanes = int(self.out_channels // 2)
+
+        self.relu = nn.ReLU()
+        if self.use_seq:
+            self.maxpool3 = nn.MaxPool2D(kernel_size=2, stride=(2, 1), padding=(0, 1))
+            self.layer3 = self._make_layer(
+                block, self.out_channels_block[2], layers[2], stride=1
+            )
+            self.conv3 = nn.Conv2D(
+                self.out_channels_block[2],
+                self.out_channels_block[2],
+                kernel_size=3,
+                stride=1,
+                padding=1,
+                bias_attr=False,
+            )
+            self.bn3 = nn.BatchNorm(self.out_channels_block[2])
+
+            self.layer4 = self._make_layer(
+                block, self.out_channels_block[3], layers[3], stride=1
+            )
+            self.conv4_1 = nn.Conv2D(
+                self.out_channels_block[3],
+                self.out_channels_block[3],
+                kernel_size=2,
+                stride=(2, 1),
+                padding=(0, 1),
+                bias_attr=False,
+            )
+            self.bn4_1 = nn.BatchNorm(self.out_channels_block[3])
+            self.conv4_2 = nn.Conv2D(
+                self.out_channels_block[3],
+                self.out_channels_block[3],
+                kernel_size=2,
+                stride=1,
+                padding=0,
+                bias_attr=False,
+            )
+            self.bn4_2 = nn.BatchNorm(self.out_channels_block[3])
+
+        if self.use_cnt:
+            self.inplanes = int(self.out_channels // 2)
+            self.v_maxpool3 = nn.MaxPool2D(kernel_size=2, stride=(2, 1), padding=(0, 1))
+            self.v_layer3 = self._make_layer(
+                block, self.out_channels_block[2], layers[2], stride=1
+            )
+            self.v_conv3 = nn.Conv2D(
+                self.out_channels_block[2],
+                self.out_channels_block[2],
+                kernel_size=3,
+                stride=1,
+                padding=1,
+                bias_attr=False,
+            )
+            self.v_bn3 = nn.BatchNorm(self.out_channels_block[2])
+
+            self.v_layer4 = self._make_layer(
+                block, self.out_channels_block[3], layers[3], stride=1
+            )
+            self.v_conv4_1 = nn.Conv2D(
+                self.out_channels_block[3],
+                self.out_channels_block[3],
+                kernel_size=2,
+                stride=(2, 1),
+                padding=(0, 1),
+                bias_attr=False,
+            )
+            self.v_bn4_1 = nn.BatchNorm(self.out_channels_block[3])
+            self.v_conv4_2 = nn.Conv2D(
+                self.out_channels_block[3],
+                self.out_channels_block[3],
+                kernel_size=2,
+                stride=1,
+                padding=0,
+                bias_attr=False,
+            )
+            self.v_bn4_2 = nn.BatchNorm(self.out_channels_block[3])
+
+    def _make_layer(self, block, planes, blocks, stride=1):
+        downsample = None
+        if stride != 1 or self.inplanes != planes * block.expansion:
+            downsample = nn.Sequential(
+                nn.Conv2D(
+                    self.inplanes,
+                    planes * block.expansion,
+                    kernel_size=1,
+                    stride=stride,
+                    bias_attr=False,
+                ),
+                nn.BatchNorm(planes * block.expansion),
+            )
+
+        layers = list()
+        layers.append(block(self.inplanes, planes, stride, downsample))
+        self.inplanes = planes * block.expansion
+        for _ in range(1, blocks):
+            layers.append(block(self.inplanes, planes))
+
+        return nn.Sequential(*layers)
+
+    def forward(self, inputs):
+        x_1 = self.backbone(inputs)
+
+        if self.use_cnt:
+            v_x = self.v_maxpool3(x_1)
+            v_x = self.v_layer3(v_x)
+            v_x = self.v_conv3(v_x)
+            v_x = self.v_bn3(v_x)
+            visual_feature_2 = self.relu(v_x)
+
+            v_x = self.v_layer4(visual_feature_2)
+            v_x = self.v_conv4_1(v_x)
+            v_x = self.v_bn4_1(v_x)
+            v_x = self.relu(v_x)
+            v_x = self.v_conv4_2(v_x)
+            v_x = self.v_bn4_2(v_x)
+            visual_feature_3 = self.relu(v_x)
+        else:
+            visual_feature_3 = None
+        if self.use_seq:
+            x = self.maxpool3(x_1)
+            x = self.layer3(x)
+            x = self.conv3(x)
+            x = self.bn3(x)
+            x_2 = self.relu(x)
+
+            x = self.layer4(x_2)
+            x = self.conv4_1(x)
+            x = self.bn4_1(x)
+            x = self.relu(x)
+            x = self.conv4_2(x)
+            x = self.bn4_2(x)
+            x_3 = self.relu(x)
+        else:
+            x_3 = None
+
+        return [visual_feature_3, x_3]
+
+
+class ResNetBase(nn.Layer):
+    def __init__(self, in_channels, out_channels, block, layers):
+        super(ResNetBase, self).__init__()
+
+        self.out_channels_block = [
+            int(out_channels / 4),
+            int(out_channels / 2),
+            out_channels,
+            out_channels,
+        ]
+
+        self.inplanes = int(out_channels / 8)
+        self.conv0_1 = nn.Conv2D(
+            in_channels,
+            int(out_channels / 16),
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            bias_attr=False,
+        )
+        self.bn0_1 = nn.BatchNorm(int(out_channels / 16))
+        self.conv0_2 = nn.Conv2D(
+            int(out_channels / 16),
+            self.inplanes,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            bias_attr=False,
+        )
+        self.bn0_2 = nn.BatchNorm(self.inplanes)
+        self.relu = nn.ReLU()
+
+        self.maxpool1 = nn.MaxPool2D(kernel_size=2, stride=2, padding=0)
+        self.layer1 = self._make_layer(block, self.out_channels_block[0], layers[0])
+        self.conv1 = nn.Conv2D(
+            self.out_channels_block[0],
+            self.out_channels_block[0],
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            bias_attr=False,
+        )
+        self.bn1 = nn.BatchNorm(self.out_channels_block[0])
+
+        self.maxpool2 = nn.MaxPool2D(kernel_size=2, stride=2, padding=0)
+        self.layer2 = self._make_layer(
+            block, self.out_channels_block[1], layers[1], stride=1
+        )
+        self.conv2 = nn.Conv2D(
+            self.out_channels_block[1],
+            self.out_channels_block[1],
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            bias_attr=False,
+        )
+        self.bn2 = nn.BatchNorm(self.out_channels_block[1])
+
+    def _make_layer(self, block, planes, blocks, stride=1):
+        downsample = None
+        if stride != 1 or self.inplanes != planes * block.expansion:
+            downsample = nn.Sequential(
+                nn.Conv2D(
+                    self.inplanes,
+                    planes * block.expansion,
+                    kernel_size=1,
+                    stride=stride,
+                    bias_attr=False,
+                ),
+                nn.BatchNorm(planes * block.expansion),
+            )
+
+        layers = list()
+        layers.append(block(self.inplanes, planes, stride, downsample))
+        self.inplanes = planes * block.expansion
+        for _ in range(1, blocks):
+            layers.append(block(self.inplanes, planes))
+
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        x = self.conv0_1(x)
+        x = self.bn0_1(x)
+        x = self.relu(x)
+        x = self.conv0_2(x)
+        x = self.bn0_2(x)
+        x = self.relu(x)
+
+        x = self.maxpool1(x)
+        x = self.layer1(x)
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = self.relu(x)
+
+        x = self.maxpool2(x)
+        x = self.layer2(x)
+        x = self.conv2(x)
+        x = self.bn2(x)
+        x = self.relu(x)
+
+        return x
+
+
+class RFLBase(nn.Layer):
+    """Reciprocal feature learning share backbone network"""
+
+    def __init__(self, in_channels, out_channels=512):
+        super(RFLBase, self).__init__()
+        self.ConvNet = ResNetBase(in_channels, out_channels, BasicBlock, [1, 2, 5, 3])
+
+    def forward(self, inputs):
+        return self.ConvNet(inputs)
diff --git a/ppocr/modeling/backbones/rec_resnet_vd.py b/ppocr/modeling/backbones/rec_resnet_vd.py
new file mode 100644
index 0000000..343d5ed
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_resnet_vd.py
@@ -0,0 +1,313 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import ParamAttr
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+__all__ = ["ResNet"]
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride=1,
+        groups=1,
+        is_vd_mode=False,
+        act=None,
+        name=None,
+    ):
+        super(ConvBNLayer, self).__init__()
+
+        self.is_vd_mode = is_vd_mode
+        self._pool2d_avg = nn.AvgPool2D(
+            kernel_size=stride, stride=stride, padding=0, ceil_mode=True
+        )
+        self._conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=1 if is_vd_mode else stride,
+            padding=(kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(name=name + "_weights"),
+            bias_attr=False,
+        )
+        if name == "conv1":
+            bn_name = "bn_" + name
+        else:
+            bn_name = "bn" + name[3:]
+        self._batch_norm = nn.BatchNorm(
+            out_channels,
+            act=act,
+            param_attr=ParamAttr(name=bn_name + "_scale"),
+            bias_attr=ParamAttr(bn_name + "_offset"),
+            moving_mean_name=bn_name + "_mean",
+            moving_variance_name=bn_name + "_variance",
+        )
+
+    def forward(self, inputs):
+        if self.is_vd_mode:
+            inputs = self._pool2d_avg(inputs)
+        y = self._conv(inputs)
+        y = self._batch_norm(y)
+        return y
+
+
+class BottleneckBlock(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        stride,
+        shortcut=True,
+        if_first=False,
+        name=None,
+    ):
+        super(BottleneckBlock, self).__init__()
+
+        self.conv0 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=1,
+            act="relu",
+            name=name + "_branch2a",
+        )
+        self.conv1 = ConvBNLayer(
+            in_channels=out_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            stride=stride,
+            act="relu",
+            name=name + "_branch2b",
+        )
+        self.conv2 = ConvBNLayer(
+            in_channels=out_channels,
+            out_channels=out_channels * 4,
+            kernel_size=1,
+            act=None,
+            name=name + "_branch2c",
+        )
+
+        if not shortcut:
+            self.short = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=out_channels * 4,
+                kernel_size=1,
+                stride=stride,
+                is_vd_mode=not if_first and stride[0] != 1,
+                name=name + "_branch1",
+            )
+
+        self.shortcut = shortcut
+
+    def forward(self, inputs):
+        y = self.conv0(inputs)
+
+        conv1 = self.conv1(y)
+        conv2 = self.conv2(conv1)
+
+        if self.shortcut:
+            short = inputs
+        else:
+            short = self.short(inputs)
+        y = paddle.add(x=short, y=conv2)
+        y = F.relu(y)
+        return y
+
+
+class BasicBlock(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        stride,
+        shortcut=True,
+        if_first=False,
+        name=None,
+    ):
+        super(BasicBlock, self).__init__()
+        self.stride = stride
+        self.conv0 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            stride=stride,
+            act="relu",
+            name=name + "_branch2a",
+        )
+        self.conv1 = ConvBNLayer(
+            in_channels=out_channels,
+            out_channels=out_channels,
+            kernel_size=3,
+            act=None,
+            name=name + "_branch2b",
+        )
+
+        if not shortcut:
+            self.short = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=out_channels,
+                kernel_size=1,
+                stride=stride,
+                is_vd_mode=not if_first and stride[0] != 1,
+                name=name + "_branch1",
+            )
+
+        self.shortcut = shortcut
+
+    def forward(self, inputs):
+        y = self.conv0(inputs)
+        conv1 = self.conv1(y)
+
+        if self.shortcut:
+            short = inputs
+        else:
+            short = self.short(inputs)
+        y = paddle.add(x=short, y=conv1)
+        y = F.relu(y)
+        return y
+
+
+class ResNet(nn.Layer):
+    def __init__(self, in_channels=3, layers=50, **kwargs):
+        super(ResNet, self).__init__()
+
+        self.layers = layers
+        supported_layers = [18, 34, 50, 101, 152, 200]
+        assert (
+            layers in supported_layers
+        ), "supported layers are {} but input layer is {}".format(
+            supported_layers, layers
+        )
+
+        if layers == 18:
+            depth = [2, 2, 2, 2]
+        elif layers == 34 or layers == 50:
+            depth = [3, 4, 6, 3]
+        elif layers == 101:
+            depth = [3, 4, 23, 3]
+        elif layers == 152:
+            depth = [3, 8, 36, 3]
+        elif layers == 200:
+            depth = [3, 12, 48, 3]
+        num_channels = [64, 256, 512, 1024] if layers >= 50 else [64, 64, 128, 256]
+        num_filters = [64, 128, 256, 512]
+
+        self.conv1_1 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=32,
+            kernel_size=3,
+            stride=1,
+            act="relu",
+            name="conv1_1",
+        )
+        self.conv1_2 = ConvBNLayer(
+            in_channels=32,
+            out_channels=32,
+            kernel_size=3,
+            stride=1,
+            act="relu",
+            name="conv1_2",
+        )
+        self.conv1_3 = ConvBNLayer(
+            in_channels=32,
+            out_channels=64,
+            kernel_size=3,
+            stride=1,
+            act="relu",
+            name="conv1_3",
+        )
+        self.pool2d_max = nn.MaxPool2D(kernel_size=3, stride=2, padding=1)
+
+        self.block_list = []
+        if layers >= 50:
+            for block in range(len(depth)):
+                shortcut = False
+                for i in range(depth[block]):
+                    if layers in [101, 152, 200] and block == 2:
+                        if i == 0:
+                            conv_name = "res" + str(block + 2) + "a"
+                        else:
+                            conv_name = "res" + str(block + 2) + "b" + str(i)
+                    else:
+                        conv_name = "res" + str(block + 2) + chr(97 + i)
+
+                    if i == 0 and block != 0:
+                        stride = (2, 1)
+                    else:
+                        stride = (1, 1)
+                    bottleneck_block = self.add_sublayer(
+                        "bb_%d_%d" % (block, i),
+                        BottleneckBlock(
+                            in_channels=(
+                                num_channels[block]
+                                if i == 0
+                                else num_filters[block] * 4
+                            ),
+                            out_channels=num_filters[block],
+                            stride=stride,
+                            shortcut=shortcut,
+                            if_first=block == i == 0,
+                            name=conv_name,
+                        ),
+                    )
+                    shortcut = True
+                    self.block_list.append(bottleneck_block)
+                self.out_channels = num_filters[block] * 4
+        else:
+            for block in range(len(depth)):
+                shortcut = False
+                for i in range(depth[block]):
+                    conv_name = "res" + str(block + 2) + chr(97 + i)
+                    if i == 0 and block != 0:
+                        stride = (2, 1)
+                    else:
+                        stride = (1, 1)
+
+                    basic_block = self.add_sublayer(
+                        "bb_%d_%d" % (block, i),
+                        BasicBlock(
+                            in_channels=(
+                                num_channels[block] if i == 0 else num_filters[block]
+                            ),
+                            out_channels=num_filters[block],
+                            stride=stride,
+                            shortcut=shortcut,
+                            if_first=block == i == 0,
+                            name=conv_name,
+                        ),
+                    )
+                    shortcut = True
+                    self.block_list.append(basic_block)
+                self.out_channels = num_filters[block]
+        self.out_pool = nn.MaxPool2D(kernel_size=2, stride=2, padding=0)
+
+    def forward(self, inputs):
+        y = self.conv1_1(inputs)
+        y = self.conv1_2(y)
+        y = self.conv1_3(y)
+        y = self.pool2d_max(y)
+        for block in self.block_list:
+            y = block(y)
+        y = self.out_pool(y)
+        return y
diff --git a/ppocr/modeling/backbones/rec_resnetv2.py b/ppocr/modeling/backbones/rec_resnetv2.py
new file mode 100644
index 0000000..c121b6f
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_resnetv2.py
@@ -0,0 +1,1227 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+This code is refer from:
+https://github.com/huggingface/pytorch-image-models/blob/main/timm/models/resnetv2.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import collections.abc
+from itertools import repeat
+from collections import OrderedDict  # pylint: disable=g-importing-member
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle.nn.initializer import TruncatedNormal, Constant, Normal, KaimingUniform
+from functools import partial
+from typing import Union, Callable, Type, List, Tuple
+
+IMAGENET_INCEPTION_MEAN = (0.5, 0.5, 0.5)
+IMAGENET_INCEPTION_STD = (0.5, 0.5, 0.5)
+normal_ = Normal(mean=0.0, std=0.01)
+zeros_ = Constant(value=0.0)
+ones_ = Constant(value=1.0)
+kaiming_normal_ = KaimingUniform(nonlinearity="relu")
+
+
+def _ntuple(n):
+    def parse(x):
+        if isinstance(x, collections.abc.Iterable):
+            return x
+        return tuple(repeat(x, n))
+
+    return parse
+
+
+to_1tuple = _ntuple(1)
+to_2tuple = _ntuple(2)
+to_3tuple = _ntuple(3)
+to_4tuple = _ntuple(4)
+to_ntuple = _ntuple
+
+
+class StdConv2dSame(nn.Conv2D):
+    def __init__(
+        self,
+        in_channel,
+        out_channels,
+        kernel_size,
+        stride=1,
+        padding="SAME",
+        dilation=1,
+        groups=1,
+        bias_attr=False,
+        eps=1e-6,
+        is_export=False,
+    ):
+        padding, is_dynamic = get_padding_value(
+            padding, kernel_size, stride=stride, dilation=dilation
+        )
+        super().__init__(
+            in_channel,
+            out_channels,
+            kernel_size,
+            stride=stride,
+            padding=padding,
+            dilation=dilation,
+            groups=groups,
+            bias_attr=bias_attr,
+        )
+        self.same_pad = is_dynamic
+        self.export = is_export
+        self.eps = eps
+
+        self.running_mean = paddle.zeros([self._out_channels], dtype="float32")
+        self.running_variance = paddle.ones([self._out_channels], dtype="float32")
+        self.batch_norm = paddle.nn.BatchNorm1D(
+            self._out_channels, use_global_stats=False
+        )
+
+    def forward(self, x):
+        if not self.training:
+            self.export = True
+        if self.same_pad:
+            if self.export:
+                x = pad_same_export(x, self._kernel_size, self._stride, self._dilation)
+            else:
+                x = pad_same(x, self._kernel_size, self._stride, self._dilation)
+        if self.export:
+            weight = paddle.reshape(
+                self.batch_norm(
+                    self.weight.reshape([1, self._out_channels, -1]).cast(
+                        paddle.float32
+                    ),
+                ),
+                self.weight.shape,
+            )
+        else:
+            weight = paddle.reshape(
+                F.batch_norm(
+                    self.weight.reshape([1, self._out_channels, -1]),
+                    self.running_mean,
+                    self.running_variance,
+                    training=True,
+                    momentum=0.0,
+                    epsilon=self.eps,
+                ),
+                self.weight.shape,
+            )
+        x = F.conv2d(
+            x,
+            weight,
+            self.bias,
+            self._stride,
+            self._padding,
+            self._dilation,
+            self._groups,
+        )
+        return x
+
+
+class StdConv2d(nn.Conv2D):
+    """Conv2d with Weight Standardization. Used for BiT ResNet-V2 models.
+
+    Paper: `Micro-Batch Training with Batch-Channel Normalization and Weight Standardization` -
+        https://arxiv.org/abs/1903.10520v2
+    """
+
+    def __init__(
+        self,
+        in_channel,
+        out_channels,
+        kernel_size,
+        stride=1,
+        padding=None,
+        dilation=1,
+        groups=1,
+        bias=False,
+        eps=1e-6,
+    ):
+        if padding is None:
+            padding = get_padding(kernel_size, stride, dilation)
+        super().__init__(
+            in_channel,
+            out_channels,
+            kernel_size,
+            stride=stride,
+            padding=padding,
+            dilation=dilation,
+            groups=groups,
+            bias_attr=bias,
+        )
+        self.eps = eps
+
+    def forward(self, x):
+        weight = F.batch_norm(
+            self.weight.reshape(1, self.out_channels, -1),
+            None,
+            None,
+            training=True,
+            momentum=0.0,
+            epsilon=self.eps,
+        ).reshape_as(self.weight)
+        x = F.conv2d(
+            x, weight, self.bias, self.stride, self.padding, self.dilation, self.groups
+        )
+        return x
+
+
+class MaxPool2dSame(nn.MaxPool2D):
+    """Tensorflow like 'SAME' wrapper for 2D max pooling"""
+
+    def __init__(
+        self,
+        kernel_size: int,
+        stride=None,
+        padding=0,
+        dilation=1,
+        ceil_mode=False,
+        is_export=False,
+    ):
+        kernel_size = to_2tuple(kernel_size)
+        stride = to_2tuple(stride)
+        dilation = to_2tuple(dilation)
+        self.export = is_export
+        super(MaxPool2dSame, self).__init__(
+            kernel_size, stride, (0, 0), dilation, ceil_mode
+        )
+
+    def forward(self, x):
+        if not self.training:
+            self.export = True
+        if self.export:
+            x = pad_same_export(x, self.ksize, self.stride, value=-float("inf"))
+        else:
+            x = pad_same(x, self.ksize, self.stride, value=-float("inf"))
+        return F.max_pool2d(x, self.ksize, self.stride, (0, 0), self.ceil_mode)
+
+
+def get_padding(kernel_size: int, stride: int = 1, dilation: int = 1, **_) -> int:
+    padding = ((stride - 1) + dilation * (kernel_size - 1)) // 2
+    return padding
+
+
+def is_static_pad(kernel_size: int, stride: int = 1, dilation: int = 1, **_):
+    return stride == 1 and (dilation * (kernel_size - 1)) % 2 == 0
+
+
+def get_padding_value(padding, kernel_size, **kwargs) -> Tuple[Tuple, bool]:
+    dynamic = False
+    if isinstance(padding, str):
+        # for any string padding, the padding will be calculated for you, one of three ways
+        padding = padding.lower()
+        if padding == "same":
+            # TF compatible 'SAME' padding, has a performance and GPU memory allocation impact
+            if is_static_pad(kernel_size, **kwargs):
+                # static case, no extra overhead
+                padding = get_padding(kernel_size, **kwargs)
+            else:
+                # dynamic 'SAME' padding, has runtime/GPU memory overhead
+                padding = 0
+                dynamic = True
+        elif padding == "valid":
+            # 'VALID' padding, same as padding=0
+            padding = 0
+        else:
+            # Default to PyTorch style 'same'-ish symmetric padding
+            padding = get_padding(kernel_size, **kwargs)
+    return padding, dynamic
+
+
+def create_pool2d(pool_type, kernel_size, stride=None, is_export=False, **kwargs):
+    stride = stride or kernel_size
+    padding = kwargs.pop("padding", "")
+    padding, is_dynamic = get_padding_value(
+        padding, kernel_size, stride=stride, **kwargs
+    )
+    if is_dynamic:
+        if pool_type == "avg":
+            return AvgPool2dSame(
+                kernel_size, stride=stride, is_export=is_export, **kwargs
+            )
+        elif pool_type == "max":
+            return MaxPool2dSame(
+                kernel_size, stride=stride, is_export=is_export, **kwargs
+            )
+        else:
+            assert False, f"Unsupported pool type {pool_type}"
+
+
+def get_same_padding(x, k, s, d):
+    return max((math.ceil(x / s) - 1) * s + (k - 1) * d + 1 - x, 0)
+
+
+def get_same_padding_export(x, k, s, d):
+    x = paddle.to_tensor(x)
+    k = paddle.to_tensor(k)
+    s = paddle.to_tensor(s)
+    d = paddle.to_tensor(d)
+    return paddle.max((paddle.ceil(x / s) - 1) * s + (k - 1) * d + 1 - x, 0)
+
+
+def pad_same_export(x, k, s, d=(1, 1), value=0):
+    ih, iw = x.shape[-2:]
+    pad_h, pad_w = get_same_padding_export(
+        ih, k[0], s[0], d[0]
+    ), get_same_padding_export(iw, k[1], s[1], d[1])
+    pad_h = pad_h.cast(paddle.int32)
+    pad_w = pad_w.cast(paddle.int32)
+    pad_list = paddle.to_tensor(
+        [
+            (pad_w // 2),
+            (pad_w - pad_w // 2).cast(paddle.int32),
+            (pad_h // 2).cast(paddle.int32),
+            (pad_h - pad_h // 2).cast(paddle.int32),
+        ]
+    )
+
+    if pad_h > 0 or pad_w > 0:
+        if len(pad_list.shape) == 2:
+            pad_list = pad_list.squeeze(1)
+        x = F.pad(x, pad_list.cast(paddle.int32), value=value)
+    return x
+
+
+def pad_same(x, k, s, d=(1, 1), value=0, pad_h=None, pad_w=None):
+    ih, iw = x.shape[-2:]
+
+    pad_h, pad_w = get_same_padding(ih, k[0], s[0], d[0]), get_same_padding(
+        iw, k[1], s[1], d[1]
+    )
+    if pad_h > 0 or pad_w > 0:
+        x = F.pad(
+            x,
+            [pad_w // 2, pad_w - pad_w // 2, pad_h // 2, pad_h - pad_h // 2],
+            value=value,
+        )
+    return x
+
+
+class AvgPool2dSame(nn.AvgPool2D):
+    """Tensorflow like 'SAME' wrapper for 2D average pooling"""
+
+    def __init__(
+        self,
+        kernel_size: int,
+        stride=None,
+        padding=0,
+        ceil_mode=False,
+        count_include_pad=True,
+    ):
+        kernel_size = to_2tuple(kernel_size)
+        stride = to_2tuple(stride)
+        super(AvgPool2dSame, self).__init__(
+            kernel_size, stride, (0, 0), ceil_mode, count_include_pad
+        )
+
+    def forward(self, x):
+        x = pad_same(x, self.kernel_size, self.stride)
+        return F.avg_pool2d(
+            x,
+            self.kernel_size,
+            self.stride,
+            self.padding,
+            self.ceil_mode,
+            self.count_include_pad,
+        )
+
+
+def drop_path(
+    x, drop_prob: float = 0.0, training: bool = False, scale_by_keep: bool = True
+):
+    if drop_prob == 0.0 or not training:
+        return x
+    keep_prob = 1 - drop_prob
+    shape = (x.shape[0],) + (1,) * (
+        x.ndim - 1
+    )  # work with diff dim tensors, not just 2D ConvNets
+    random_tensor = x.new_empty(shape).bernoulli_(keep_prob)
+    if keep_prob > 0.0 and scale_by_keep:
+        random_tensor.div_(keep_prob)
+    return x * random_tensor
+
+
+class DropPath(nn.Layer):
+    """Drop paths (Stochastic Depth) per sample  (when applied in main path of residual blocks)."""
+
+    def __init__(self, drop_prob=None, scale_by_keep=True):
+        super(DropPath, self).__init__()
+        self.drop_prob = drop_prob
+        self.scale_by_keep = scale_by_keep
+
+    def forward(self, x):
+        return drop_path(x, self.drop_prob, self.training, self.scale_by_keep)
+
+
+def adaptive_pool_feat_mult(pool_type="avg"):
+    if pool_type == "catavgmax":
+        return 2
+    else:
+        return 1
+
+
+class SelectAdaptivePool2d(nn.Layer):
+    """Selectable global pooling layer with dynamic input kernel size"""
+
+    def __init__(self, output_size=1, pool_type="fast", flatten=False):
+        super(SelectAdaptivePool2d, self).__init__()
+        self.pool_type = (
+            pool_type or ""
+        )  # convert other falsy values to empty string for consistent TS typing
+        self.flatten = nn.Flatten(1) if flatten else nn.Identity()
+        if pool_type == "":
+            self.pool = nn.Identity()  # pass through
+
+    def is_identity(self):
+        return not self.pool_type
+
+    def forward(self, x):
+        x = self.pool(x)
+        x = self.flatten(x)
+        return x
+
+    def feat_mult(self):
+        return adaptive_pool_feat_mult(self.pool_type)
+
+    def __repr__(self):
+        return (
+            self.__class__.__name__
+            + " ("
+            + "pool_type="
+            + self.pool_type
+            + ", flatten="
+            + str(self.flatten)
+            + ")"
+        )
+
+
+def _create_pool(num_features, num_classes, pool_type="avg", use_conv=False):
+    flatten_in_pool = not use_conv  # flatten when we use a Linear layer after pooling
+    if not pool_type:
+        assert (
+            num_classes == 0 or use_conv
+        ), "Pooling can only be disabled if classifier is also removed or conv classifier is used"
+        flatten_in_pool = (
+            False  # disable flattening if pooling is pass-through (no pooling)
+        )
+    global_pool = SelectAdaptivePool2d(pool_type=pool_type, flatten=flatten_in_pool)
+    num_pooled_features = num_features * global_pool.feat_mult()
+    return global_pool, num_pooled_features
+
+
+def _create_fc(num_features, num_classes, use_conv=False):
+    if num_classes <= 0:
+        fc = nn.Identity()  # pass-through (no classifier)
+    elif use_conv:
+        fc = nn.Conv2D(num_features, num_classes, 1, bias_attr=True)
+    else:
+        fc = nn.Linear(num_features, num_classes, bias_attr=True)
+    return fc
+
+
+class ClassifierHead(nn.Layer):
+    """Classifier head w/ configurable global pooling and dropout."""
+
+    def __init__(
+        self, in_chs, num_classes, pool_type="avg", drop_rate=0.0, use_conv=False
+    ):
+        super(ClassifierHead, self).__init__()
+        self.drop_rate = drop_rate
+        self.global_pool, num_pooled_features = _create_pool(
+            in_chs, num_classes, pool_type, use_conv=use_conv
+        )
+        self.fc = _create_fc(num_pooled_features, num_classes, use_conv=use_conv)
+        self.flatten = nn.Flatten(1) if use_conv and pool_type else nn.Identity()
+
+    def forward(self, x):
+        x = self.global_pool(x)
+        if self.drop_rate:
+            x = F.dropout(x, p=float(self.drop_rate), training=self.training)
+        x = self.fc(x)
+        x = self.flatten(x)
+        return x
+
+
+class EvoNormBatch2d(nn.Layer):
+    def __init__(
+        self, num_features, apply_act=True, momentum=0.1, eps=1e-5, drop_block=None
+    ):
+        super(EvoNormBatch2d, self).__init__()
+        self.apply_act = apply_act  # apply activation (non-linearity)
+        self.momentum = momentum
+        self.eps = eps
+        self.weight = paddle.create_parameter(
+            paddle.ones(num_features), dtype="float32"
+        )
+        self.bias = paddle.create_parameter(paddle.zeros(num_features), dtype="float32")
+        self.v = (
+            paddle.create_parameter(paddle.ones(num_features), dtype="float32")
+            if apply_act
+            else None
+        )
+        self.register_buffer("running_var", paddle.ones([num_features]))
+        self.reset_parameters()
+
+    def reset_parameters(self):
+        ones_(self.weight)
+        zeros_(self.bias)
+        if self.apply_act:
+            ones_(self.v)
+
+    def forward(self, x):
+        x_type = x.dtype
+        if self.v is not None:
+            running_var = self.running_var.view(1, -1, 1, 1)
+            if self.training:
+                var = x.var(dim=(0, 2, 3), unbiased=False, keepdim=True)
+                n = x.numel() / x.shape[1]
+                running_var = var.detach() * self.momentum * (
+                    n / (n - 1)
+                ) + running_var * (1 - self.momentum)
+                self.running_var.copy_(running_var.view(self.running_var.shape))
+            else:
+                var = running_var
+            v = self.v.to(dtype=x_type).reshape(1, -1, 1, 1)
+            d = x * v + (
+                x.var(dim=(2, 3), unbiased=False, keepdim=True) + self.eps
+            ).sqrt().to(dtype=x_type)
+            d = d.max((var + self.eps).sqrt().to(dtype=x_type))
+            x = x / d
+        return x * self.weight.view(1, -1, 1, 1) + self.bias.view(1, -1, 1, 1)
+
+
+class EvoNormSample2d(nn.Layer):
+    def __init__(
+        self, num_features, apply_act=True, groups=32, eps=1e-5, drop_block=None
+    ):
+        super(EvoNormSample2d, self).__init__()
+        self.apply_act = apply_act
+        self.groups = groups
+        self.eps = eps
+        self.weight = paddle.create_parameter(
+            paddle.ones(num_features), dtype="float32"
+        )
+        self.bias = paddle.create_parameter(paddle.zeros(num_features), dtype="float32")
+        self.v = (
+            paddle.create_parameter(paddle.ones(num_features), dtype="float32")
+            if apply_act
+            else None
+        )
+        self.reset_parameters()
+
+    def reset_parameters(self):
+        ones_(self.weight)
+        zeros_(self.bias)
+        if self.apply_act:
+            ones_(self.v)
+
+    def forward(self, x):
+        B, C, H, W = x.shape
+        if self.v is not None:
+            n = x * (x * self.v.view(1, -1, 1, 1)).sigmoid()
+            x = x.reshape(B, self.groups, -1)
+            x = (
+                n.reshape(B, self.groups, -1)
+                / (x.var(dim=-1, unbiased=False, keepdim=True) + self.eps).sqrt()
+            )
+            x = x.reshape(B, C, H, W)
+        return x * self.weight.reshape([1, -1, 1, 1]) + self.bias.reshape([1, -1, 1, 1])
+
+
+class GroupNormAct(nn.GroupNorm):
+    # NOTE num_channel and num_groups order flipped for easier layer swaps / binding of fixed args
+    def __init__(
+        self,
+        num_channels,
+        num_groups=32,
+        eps=1e-5,
+        affine=True,
+        apply_act=True,
+        act_layer=nn.ReLU,
+        drop_block=None,
+    ):
+        super(GroupNormAct, self).__init__(num_groups, num_channels, epsilon=eps)
+        if affine:
+            self.weight = paddle.create_parameter([num_channels], dtype="float32")
+            self.bias = paddle.create_parameter([num_channels], dtype="float32")
+            ones_(self.weight)
+            zeros_(self.bias)
+        if act_layer is not None and apply_act:
+            act_args = {}
+            self.act = act_layer(**act_args)
+        else:
+            self.act = nn.Identity()
+
+    def forward(self, x):
+        x = F.group_norm(
+            x,
+            num_groups=self._num_groups,
+            epsilon=self._epsilon,
+            weight=self.weight,
+            bias=self.bias,
+        )
+        x = self.act(x)
+        return x
+
+
+class BatchNormAct2d(nn.BatchNorm2D):
+    def __init__(
+        self,
+        num_features,
+        eps=1e-5,
+        momentum=0.1,
+        affine=True,
+        track_running_stats=True,
+        apply_act=True,
+        act_layer=nn.ReLU,
+        drop_block=None,
+    ):
+        super(BatchNormAct2d, self).__init__(
+            num_features,
+            epsilon=eps,
+            momentum=momentum,
+            use_global_stats=track_running_stats,
+        )
+        if act_layer is not None and apply_act:
+            act_args = dict()
+            self.act = act_layer(**act_args)
+        else:
+            self.act = nn.Identity()
+
+    def _forward_python(self, x):
+        return super(BatchNormAct2d, self).forward(x)
+
+    def forward(self, x):
+        x = self._forward_python(x)
+        x = self.act(x)
+        return x
+
+
+def adapt_input_conv(in_chans, conv_weight):
+    conv_type = conv_weight.dtype
+    conv_weight = (
+        conv_weight.float()
+    )  # Some weights are in torch.half, ensure it's float for sum on CPU
+    O, I, J, K = conv_weight.shape
+    if in_chans == 1:
+        if I > 3:
+            assert conv_weight.shape[1] % 3 == 0
+            # For models with space2depth stems
+            conv_weight = conv_weight.reshape(O, I // 3, 3, J, K)
+            conv_weight = conv_weight.sum(dim=2, keepdim=False)
+        else:
+            conv_weight = conv_weight.sum(dim=1, keepdim=True)
+    elif in_chans != 3:
+        if I != 3:
+            raise NotImplementedError("Weight format not supported by conversion.")
+        else:
+            # NOTE this strategy should be better than random init, but there could be other combinations of
+            # the original RGB input layer weights that'd work better for specific cases.
+            repeat = int(math.ceil(in_chans / 3))
+            conv_weight = conv_weight.repeat(1, repeat, 1, 1)[:, :in_chans, :, :]
+            conv_weight *= 3 / float(in_chans)
+    conv_weight = conv_weight.to(conv_type)
+    return conv_weight
+
+
+def named_apply(
+    fn: Callable, module: nn.Layer, name="", depth_first=True, include_root=False
+) -> nn.Layer:
+    if not depth_first and include_root:
+        fn(module=module, name=name)
+    for child_name, child_module in module.named_children():
+        child_name = ".".join((name, child_name)) if name else child_name
+        named_apply(
+            fn=fn,
+            module=child_module,
+            name=child_name,
+            depth_first=depth_first,
+            include_root=True,
+        )
+    if depth_first and include_root:
+        fn(module=module, name=name)
+    return module
+
+
+def _cfg(url="", **kwargs):
+    return {
+        "url": url,
+        "num_classes": 1000,
+        "input_size": (3, 224, 224),
+        "pool_size": (7, 7),
+        "crop_pct": 0.875,
+        "interpolation": "bilinear",
+        "mean": IMAGENET_INCEPTION_MEAN,
+        "std": IMAGENET_INCEPTION_STD,
+        "first_conv": "stem.conv",
+        "classifier": "head.fc",
+        **kwargs,
+    }
+
+
+def make_div(v, divisor=8):
+    min_value = divisor
+    new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
+    if new_v < 0.9 * v:
+        new_v += divisor
+    return new_v
+
+
+class PreActBottleneck(nn.Layer):
+    """Pre-activation (v2) bottleneck block.
+
+    Follows the implementation of "Identity Mappings in Deep Residual Networks":
+    https://github.com/KaimingHe/resnet-1k-layers/blob/master/resnet-pre-act.lua
+
+    Except it puts the stride on 3x3 conv when available.
+    """
+
+    def __init__(
+        self,
+        in_chs,
+        out_chs=None,
+        bottle_ratio=0.25,
+        stride=1,
+        dilation=1,
+        first_dilation=None,
+        groups=1,
+        act_layer=None,
+        conv_layer=None,
+        norm_layer=None,
+        proj_layer=None,
+        drop_path_rate=0.0,
+        is_export=False,
+    ):
+        super().__init__()
+        first_dilation = first_dilation or dilation
+        conv_layer = conv_layer or StdConv2d
+        norm_layer = norm_layer or partial(GroupNormAct, num_groups=32)
+        out_chs = out_chs or in_chs
+        mid_chs = make_div(out_chs * bottle_ratio)
+
+        if proj_layer is not None:
+            self.downsample = proj_layer(
+                in_chs,
+                out_chs,
+                stride=stride,
+                dilation=dilation,
+                first_dilation=first_dilation,
+                preact=True,
+                conv_layer=conv_layer,
+                norm_layer=norm_layer,
+            )
+        else:
+            self.downsample = None
+
+        self.norm1 = norm_layer(in_chs)
+        self.conv1 = conv_layer(in_chs, mid_chs, 1, is_export=is_export)
+        self.norm2 = norm_layer(mid_chs)
+        self.conv2 = conv_layer(
+            mid_chs,
+            mid_chs,
+            3,
+            stride=stride,
+            dilation=first_dilation,
+            groups=groups,
+            is_export=is_export,
+        )
+        self.norm3 = norm_layer(mid_chs)
+        self.conv3 = conv_layer(mid_chs, out_chs, 1, is_export=is_export)
+        self.drop_path = (
+            DropPath(drop_path_rate) if drop_path_rate > 0 else nn.Identity()
+        )
+
+    def zero_init_last(self):
+        zeros_(self.conv3.weight)
+
+    def forward(self, x):
+        x_preact = self.norm1(x)
+
+        # shortcut branch
+        shortcut = x
+        if self.downsample is not None:
+            shortcut = self.downsample(x_preact)
+
+        # residual branch
+        x = self.conv1(x_preact)
+        x = self.conv2(self.norm2(x))
+        x = self.conv3(self.norm3(x))
+        x = self.drop_path(x)
+        return x + shortcut
+
+
+class Bottleneck(nn.Layer):
+    """Non Pre-activation bottleneck block, equiv to V1.5/V1b Bottleneck. Used for ViT."""
+
+    def __init__(
+        self,
+        in_chs,
+        out_chs=None,
+        bottle_ratio=0.25,
+        stride=1,
+        dilation=1,
+        first_dilation=None,
+        groups=1,
+        act_layer=None,
+        conv_layer=None,
+        norm_layer=None,
+        proj_layer=None,
+        drop_path_rate=0.0,
+        is_export=False,
+    ):
+        super().__init__()
+        first_dilation = first_dilation or dilation
+        act_layer = act_layer or nn.ReLU
+        conv_layer = conv_layer or StdConv2d
+        norm_layer = norm_layer or partial(GroupNormAct, num_groups=32)
+        out_chs = out_chs or in_chs
+        mid_chs = make_div(out_chs * bottle_ratio)
+
+        if proj_layer is not None:
+            self.downsample = proj_layer(
+                in_chs,
+                out_chs,
+                stride=stride,
+                dilation=dilation,
+                preact=False,
+                conv_layer=conv_layer,
+                norm_layer=norm_layer,
+                is_export=is_export,
+            )
+        else:
+            self.downsample = None
+
+        self.conv1 = conv_layer(in_chs, mid_chs, 1, is_export=is_export)
+        self.norm1 = norm_layer(mid_chs)
+        self.conv2 = conv_layer(
+            mid_chs,
+            mid_chs,
+            3,
+            stride=stride,
+            dilation=first_dilation,
+            groups=groups,
+            is_export=is_export,
+        )
+        self.norm2 = norm_layer(mid_chs)
+        self.conv3 = conv_layer(mid_chs, out_chs, 1, is_export=is_export)
+        self.norm3 = norm_layer(out_chs, apply_act=False)
+        self.drop_path = (
+            DropPath(drop_path_rate) if drop_path_rate > 0 else nn.Identity()
+        )
+        self.act3 = act_layer()
+
+    def zero_init_last(self):
+        zeros_(self.norm3.weight)
+
+    def forward(self, x):
+        # shortcut branch
+        shortcut = x
+        if self.downsample is not None:
+            shortcut = self.downsample(x)
+
+        # residual
+        x = self.conv1(x)
+        x = self.norm1(x)
+        x = self.conv2(x)
+        x = self.norm2(x)
+        x = self.conv3(x)
+        x = self.norm3(x)
+        x = self.drop_path(x)
+        x = self.act3(x + shortcut)
+        return x
+
+
+class DownsampleConv(nn.Layer):
+    def __init__(
+        self,
+        in_chs,
+        out_chs,
+        stride=1,
+        dilation=1,
+        first_dilation=None,
+        preact=True,
+        conv_layer=None,
+        norm_layer=None,
+        is_export=False,
+    ):
+        super(DownsampleConv, self).__init__()
+        self.conv = conv_layer(in_chs, out_chs, 1, stride=stride, is_export=is_export)
+        self.norm = nn.Identity() if preact else norm_layer(out_chs, apply_act=False)
+
+    def forward(self, x):
+        return self.norm(self.conv(x))
+
+
+class DownsampleAvg(nn.Layer):
+    def __init__(
+        self,
+        in_chs,
+        out_chs,
+        stride=1,
+        dilation=1,
+        first_dilation=None,
+        preact=True,
+        conv_layer=None,
+        norm_layer=None,
+        is_export=False,
+    ):
+        """AvgPool Downsampling as in 'D' ResNet variants. This is not in RegNet space but I might experiment."""
+        super(DownsampleAvg, self).__init__()
+        avg_stride = stride if dilation == 1 else 1
+        if stride > 1 or dilation > 1:
+            avg_pool_fn = (
+                AvgPool2dSame if avg_stride == 1 and dilation > 1 else nn.AvgPool2D
+            )
+            self.pool = avg_pool_fn(2, avg_stride, ceil_mode=True, exclusive=False)
+        else:
+            self.pool = nn.Identity()
+        self.conv = conv_layer(in_chs, out_chs, 1, stride=1, is_export=is_export)
+        self.norm = nn.Identity() if preact else norm_layer(out_chs, apply_act=False)
+
+    def forward(self, x):
+        return self.norm(self.conv(self.pool(x)))
+
+
+class ResNetStage(nn.Layer):
+    """ResNet Stage."""
+
+    def __init__(
+        self,
+        in_chs,
+        out_chs,
+        stride,
+        dilation,
+        depth,
+        bottle_ratio=0.25,
+        groups=1,
+        avg_down=False,
+        block_dpr=None,
+        block_fn=PreActBottleneck,
+        is_export=False,
+        act_layer=None,
+        conv_layer=None,
+        norm_layer=None,
+        **block_kwargs,
+    ):
+        super(ResNetStage, self).__init__()
+        first_dilation = 1 if dilation in (1, 2) else 2
+        layer_kwargs = dict(
+            act_layer=act_layer, conv_layer=conv_layer, norm_layer=norm_layer
+        )
+        proj_layer = DownsampleAvg if avg_down else DownsampleConv
+        prev_chs = in_chs
+        self.blocks = nn.Sequential()
+        for block_idx in range(depth):
+            drop_path_rate = block_dpr[block_idx] if block_dpr else 0.0
+            stride = stride if block_idx == 0 else 1
+            self.blocks.add_sublayer(
+                str(block_idx),
+                block_fn(
+                    prev_chs,
+                    out_chs,
+                    stride=stride,
+                    dilation=dilation,
+                    bottle_ratio=bottle_ratio,
+                    groups=groups,
+                    first_dilation=first_dilation,
+                    proj_layer=proj_layer,
+                    drop_path_rate=drop_path_rate,
+                    is_export=is_export,
+                    **layer_kwargs,
+                    **block_kwargs,
+                ),
+            )
+            prev_chs = out_chs
+            first_dilation = dilation
+            proj_layer = None
+
+    def forward(self, x):
+        x = self.blocks(x)
+        return x
+
+
+def is_stem_deep(stem_type):
+    return any([s in stem_type for s in ("deep", "tiered")])
+
+
+def create_resnetv2_stem(
+    in_chs,
+    out_chs=64,
+    stem_type="",
+    preact=True,
+    conv_layer=StdConv2d,
+    norm_layer=partial(GroupNormAct, num_groups=32),
+    is_export=False,
+):
+    stem = OrderedDict()
+    assert stem_type in (
+        "",
+        "fixed",
+        "same",
+        "deep",
+        "deep_fixed",
+        "deep_same",
+        "tiered",
+    )
+
+    # NOTE conv padding mode can be changed by overriding the conv_layer def
+    if is_stem_deep(stem_type):
+        # A 3 deep 3x3  conv stack as in ResNet V1D models
+        if "tiered" in stem_type:
+            stem_chs = (3 * out_chs // 8, out_chs // 2)  # 'T' resnets in resnet.py
+        else:
+            stem_chs = (out_chs // 2, out_chs // 2)  # 'D' ResNets
+        stem["conv1"] = conv_layer(
+            in_chs, stem_chs[0], kernel_size=3, stride=2, is_export=is_export
+        )
+        stem["norm1"] = norm_layer(stem_chs[0])
+        stem["conv2"] = conv_layer(
+            stem_chs[0], stem_chs[1], kernel_size=3, stride=1, is_export=is_export
+        )
+        stem["norm2"] = norm_layer(stem_chs[1])
+        stem["conv3"] = conv_layer(
+            stem_chs[1], out_chs, kernel_size=3, stride=1, is_export=is_export
+        )
+        if not preact:
+            stem["norm3"] = norm_layer(out_chs)
+    else:
+        # The usual 7x7 stem conv
+        stem["conv"] = conv_layer(
+            in_chs, out_chs, kernel_size=7, stride=2, is_export=is_export
+        )
+        if not preact:
+            stem["norm"] = norm_layer(out_chs)
+
+    if "fixed" in stem_type:
+        # 'fixed' SAME padding approximation that is used in BiT models
+        stem["pad"] = paddle.nn.Pad2D(
+            1, mode="constant", value=0.0, data_format="NCHW", name=None
+        )
+        stem["pool"] = nn.MaxPool2D(kernel_size=3, stride=2, padding=0)
+    elif "same" in stem_type:
+        # full, input size based 'SAME' padding, used in ViT Hybrid model
+        stem["pool"] = create_pool2d(
+            "max", kernel_size=3, stride=2, padding="same", is_export=is_export
+        )
+    else:
+        # the usual Pypaddle symmetric padding
+        stem["pool"] = nn.MaxPool2D(kernel_size=3, stride=2, padding=1)
+    stem_seq = nn.Sequential()
+    for key, value in stem.items():
+        stem_seq.add_sublayer(key, value)
+
+    return stem_seq
+
+
+class ResNetV2(nn.Layer):
+    """Implementation of Pre-activation (v2) ResNet mode.
+
+    Args:
+      x: input images with shape [N, 1, H, W]
+
+    Returns:
+      The extracted features [N, 1, H//16, W//16]
+    """
+
+    def __init__(
+        self,
+        layers,
+        channels=(256, 512, 1024, 2048),
+        num_classes=1000,
+        in_chans=3,
+        global_pool="avg",
+        output_stride=32,
+        width_factor=1,
+        stem_chs=64,
+        stem_type="",
+        avg_down=False,
+        preact=True,
+        act_layer=nn.ReLU,
+        conv_layer=StdConv2d,
+        norm_layer=partial(GroupNormAct, num_groups=32),
+        drop_rate=0.0,
+        drop_path_rate=0.0,
+        zero_init_last=False,
+        is_export=False,
+    ):
+        super().__init__()
+        self.num_classes = num_classes
+        self.drop_rate = drop_rate
+        self.is_export = is_export
+        wf = width_factor
+        self.feature_info = []
+        stem_chs = make_div(stem_chs * wf)
+        self.stem = create_resnetv2_stem(
+            in_chans,
+            stem_chs,
+            stem_type,
+            preact,
+            conv_layer=conv_layer,
+            norm_layer=norm_layer,
+            is_export=is_export,
+        )
+        stem_feat = (
+            ("stem.conv3" if is_stem_deep(stem_type) else "stem.conv")
+            if preact
+            else "stem.norm"
+        )
+        self.feature_info.append(dict(num_chs=stem_chs, reduction=2, module=stem_feat))
+
+        prev_chs = stem_chs
+        curr_stride = 4
+        dilation = 1
+        block_dprs = [
+            x.tolist()
+            for x in paddle.linspace(0, drop_path_rate, sum(layers)).split(layers)
+        ]
+        block_fn = PreActBottleneck if preact else Bottleneck
+        self.stages = nn.Sequential()
+        for stage_idx, (d, c, bdpr) in enumerate(zip(layers, channels, block_dprs)):
+            out_chs = make_div(c * wf)
+            stride = 1 if stage_idx == 0 else 2
+            if curr_stride >= output_stride:
+                dilation *= stride
+                stride = 1
+            stage = ResNetStage(
+                prev_chs,
+                out_chs,
+                stride=stride,
+                dilation=dilation,
+                depth=d,
+                avg_down=avg_down,
+                act_layer=act_layer,
+                conv_layer=conv_layer,
+                norm_layer=norm_layer,
+                block_dpr=bdpr,
+                block_fn=block_fn,
+                is_export=is_export,
+            )
+            prev_chs = out_chs
+            curr_stride *= stride
+            self.feature_info += [
+                dict(
+                    num_chs=prev_chs,
+                    reduction=curr_stride,
+                    module=f"stages.{stage_idx}",
+                )
+            ]
+            self.stages.add_sublayer(str(stage_idx), stage)
+
+        self.num_features = prev_chs
+        self.norm = norm_layer(self.num_features) if preact else nn.Identity()
+        self.head = ClassifierHead(
+            self.num_features,
+            num_classes,
+            pool_type=global_pool,
+            drop_rate=self.drop_rate,
+            use_conv=True,
+        )
+
+        self.init_weights(zero_init_last=zero_init_last)
+
+    def init_weights(self, zero_init_last=True):
+        named_apply(partial(_init_weights, zero_init_last=zero_init_last), self)
+
+    def load_pretrained(self, checkpoint_path, prefix="resnet/"):
+        _load_weights(self, checkpoint_path, prefix)
+
+    def get_classifier(self):
+        return self.head.fc
+
+    def reset_classifier(self, num_classes, global_pool="avg"):
+        self.num_classes = num_classes
+        self.head = ClassifierHead(
+            self.num_features,
+            num_classes,
+            pool_type=global_pool,
+            drop_rate=self.drop_rate,
+            use_conv=True,
+        )
+
+    def forward_features(self, x):
+        x = self.stem(x)
+        x = self.stages(x)
+        x = self.norm(x)
+        return x
+
+    def forward(self, x):
+        x = self.forward_features(x)
+        x = self.head(x)
+        return x
+
+
+def _init_weights(module: nn.Layer, name: str = "", zero_init_last=True):
+    if isinstance(module, nn.Linear) or (
+        "head.fc" in name and isinstance(module, nn.Conv2D)
+    ):
+        normal_(module.weight)
+        zeros_(module.bias)
+    elif isinstance(module, nn.Conv2D):
+        kaiming_normal_(module.weight)
+        if module.bias is not None:
+            zeros_(module.bias)
+    elif isinstance(module, (nn.BatchNorm2D, nn.LayerNorm, nn.GroupNorm)):
+        ones_(module.weight)
+        zeros_(module.bias)
+    elif zero_init_last and hasattr(module, "zero_init_last"):
+        module.zero_init_last()
+
+
+@paddle.no_grad()
+def _load_weights(model: nn.Layer, checkpoint_path: str, prefix: str = "resnet/"):
+    import numpy as np
+
+    def t2p(conv_weights):
+        """Possibly convert HWIO to OIHW."""
+        if conv_weights.ndim == 4:
+            conv_weights = conv_weights.transpose([3, 2, 0, 1])
+        return paddle.to_tensor(conv_weights)
+
+    weights = np.load(checkpoint_path)
+    stem_conv_w = adapt_input_conv(
+        model.stem.conv.weight.shape[1],
+        t2p(weights[f"{prefix}root_block/standardized_conv2d/kernel"]),
+    )
+    model.stem.conv.weight.copy_(stem_conv_w)
+    model.norm.weight.copy_(t2p(weights[f"{prefix}group_norm/gamma"]))
+    model.norm.bias.copy_(t2p(weights[f"{prefix}group_norm/beta"]))
+    if (
+        isinstance(getattr(model.head, "fc", None), nn.Conv2D)
+        and model.head.fc.weight.shape[0]
+        == weights[f"{prefix}head/conv2d/kernel"].shape[-1]
+    ):
+        model.head.fc.weight.copy_(t2p(weights[f"{prefix}head/conv2d/kernel"]))
+        model.head.fc.bias.copy_(t2p(weights[f"{prefix}head/conv2d/bias"]))
+    for i, (sname, stage) in enumerate(model.stages.named_children()):
+        for j, (bname, block) in enumerate(stage.blocks.named_children()):
+            cname = "standardized_conv2d"
+            block_prefix = f"{prefix}block{i + 1}/unit{j + 1:02d}/"
+            block.conv1.weight.copy_(t2p(weights[f"{block_prefix}a/{cname}/kernel"]))
+            block.conv2.weight.copy_(t2p(weights[f"{block_prefix}b/{cname}/kernel"]))
+            block.conv3.weight.copy_(t2p(weights[f"{block_prefix}c/{cname}/kernel"]))
+            block.norm1.weight.copy_(t2p(weights[f"{block_prefix}a/group_norm/gamma"]))
+            block.norm2.weight.copy_(t2p(weights[f"{block_prefix}b/group_norm/gamma"]))
+            block.norm3.weight.copy_(t2p(weights[f"{block_prefix}c/group_norm/gamma"]))
+            block.norm1.bias.copy_(t2p(weights[f"{block_prefix}a/group_norm/beta"]))
+            block.norm2.bias.copy_(t2p(weights[f"{block_prefix}b/group_norm/beta"]))
+            block.norm3.bias.copy_(t2p(weights[f"{block_prefix}c/group_norm/beta"]))
+            if block.downsample is not None:
+                w = weights[f"{block_prefix}a/proj/{cname}/kernel"]
+                block.downsample.conv.weight.copy_(t2p(w))
diff --git a/ppocr/modeling/backbones/rec_shallow_cnn.py b/ppocr/modeling/backbones/rec_shallow_cnn.py
new file mode 100644
index 0000000..e5a8b65
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_shallow_cnn.py
@@ -0,0 +1,82 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/1.x/mmocr/models/textrecog/backbones/shallow_cnn.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import numpy as np
+import paddle
+from paddle import ParamAttr
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle.nn import MaxPool2D
+from paddle.nn.initializer import KaimingNormal, Uniform, Constant
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self, num_channels, filter_size, num_filters, stride, padding, num_groups=1
+    ):
+        super(ConvBNLayer, self).__init__()
+
+        self.conv = nn.Conv2D(
+            in_channels=num_channels,
+            out_channels=num_filters,
+            kernel_size=filter_size,
+            stride=stride,
+            padding=padding,
+            groups=num_groups,
+            weight_attr=ParamAttr(initializer=KaimingNormal()),
+            bias_attr=False,
+        )
+
+        self.bn = nn.BatchNorm2D(
+            num_filters,
+            weight_attr=ParamAttr(initializer=Uniform(0, 1)),
+            bias_attr=ParamAttr(initializer=Constant(0)),
+        )
+        self.relu = nn.ReLU()
+
+    def forward(self, inputs):
+        y = self.conv(inputs)
+        y = self.bn(y)
+        y = self.relu(y)
+        return y
+
+
+class ShallowCNN(nn.Layer):
+    def __init__(self, in_channels=1, hidden_dim=512):
+        super().__init__()
+        assert isinstance(in_channels, int)
+        assert isinstance(hidden_dim, int)
+
+        self.conv1 = ConvBNLayer(in_channels, 3, hidden_dim // 2, stride=1, padding=1)
+        self.conv2 = ConvBNLayer(hidden_dim // 2, 3, hidden_dim, stride=1, padding=1)
+        self.pool = nn.MaxPool2D(kernel_size=2, stride=2, padding=0)
+        self.out_channels = hidden_dim
+
+    def forward(self, x):
+        x = self.conv1(x)
+        x = self.pool(x)
+
+        x = self.conv2(x)
+        x = self.pool(x)
+
+        return x
diff --git a/ppocr/modeling/backbones/rec_svtrnet.py b/ppocr/modeling/backbones/rec_svtrnet.py
new file mode 100644
index 0000000..427c87b
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_svtrnet.py
@@ -0,0 +1,642 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddle import ParamAttr
+from paddle.nn.initializer import KaimingNormal
+import numpy as np
+import paddle
+import paddle.nn as nn
+from paddle.nn.initializer import TruncatedNormal, Constant, Normal
+
+trunc_normal_ = TruncatedNormal(std=0.02)
+normal_ = Normal
+zeros_ = Constant(value=0.0)
+ones_ = Constant(value=1.0)
+
+
+def drop_path(x, drop_prob=0.0, training=False):
+    """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
+    the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
+    See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ...
+    """
+    if drop_prob == 0.0 or not training:
+        return x
+    keep_prob = paddle.to_tensor(1 - drop_prob, dtype=x.dtype)
+    shape = (x.shape[0],) + (1,) * (x.ndim - 1)
+    random_tensor = keep_prob + paddle.rand(shape, dtype=x.dtype)
+    random_tensor = paddle.floor(random_tensor)  # binarize
+    output = x.divide(keep_prob) * random_tensor
+    return output
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size=3,
+        stride=1,
+        padding=0,
+        bias_attr=False,
+        groups=1,
+        act=nn.GELU,
+    ):
+        super().__init__()
+        self.conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            groups=groups,
+            weight_attr=paddle.ParamAttr(initializer=nn.initializer.KaimingUniform()),
+            bias_attr=bias_attr,
+        )
+        self.norm = nn.BatchNorm2D(out_channels)
+        self.act = act()
+
+    def forward(self, inputs):
+        out = self.conv(inputs)
+        out = self.norm(out)
+        out = self.act(out)
+        return out
+
+
+class DropPath(nn.Layer):
+    """Drop paths (Stochastic Depth) per sample  (when applied in main path of residual blocks)."""
+
+    def __init__(self, drop_prob=None):
+        super(DropPath, self).__init__()
+        self.drop_prob = drop_prob
+
+    def forward(self, x):
+        return drop_path(x, self.drop_prob, self.training)
+
+
+class Identity(nn.Layer):
+    def __init__(self):
+        super(Identity, self).__init__()
+
+    def forward(self, input):
+        return input
+
+
+class Mlp(nn.Layer):
+    def __init__(
+        self,
+        in_features,
+        hidden_features=None,
+        out_features=None,
+        act_layer=nn.GELU,
+        drop=0.0,
+    ):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.fc1 = nn.Linear(in_features, hidden_features)
+        self.act = act_layer()
+        self.fc2 = nn.Linear(hidden_features, out_features)
+        self.drop = nn.Dropout(drop)
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.act(x)
+        x = self.drop(x)
+        x = self.fc2(x)
+        x = self.drop(x)
+        return x
+
+
+class ConvMixer(nn.Layer):
+    def __init__(
+        self,
+        dim,
+        num_heads=8,
+        HW=[8, 25],
+        local_k=[3, 3],
+    ):
+        super().__init__()
+        self.HW = HW
+        self.dim = dim
+        self.local_mixer = nn.Conv2D(
+            dim,
+            dim,
+            local_k,
+            1,
+            [local_k[0] // 2, local_k[1] // 2],
+            groups=num_heads,
+            weight_attr=ParamAttr(initializer=KaimingNormal()),
+        )
+
+    def forward(self, x):
+        h = self.HW[0]
+        w = self.HW[1]
+        x = x.transpose([0, 2, 1]).reshape([0, self.dim, h, w])
+        x = self.local_mixer(x)
+        x = x.flatten(2).transpose([0, 2, 1])
+        return x
+
+
+class Attention(nn.Layer):
+    def __init__(
+        self,
+        dim,
+        num_heads=8,
+        mixer="Global",
+        HW=None,
+        local_k=[7, 11],
+        qkv_bias=False,
+        qk_scale=None,
+        attn_drop=0.0,
+        proj_drop=0.0,
+    ):
+        super().__init__()
+        self.num_heads = num_heads
+        self.dim = dim
+        self.head_dim = dim // num_heads
+        self.scale = qk_scale or self.head_dim**-0.5
+
+        self.qkv = nn.Linear(dim, dim * 3, bias_attr=qkv_bias)
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.proj = nn.Linear(dim, dim)
+        self.proj_drop = nn.Dropout(proj_drop)
+        self.HW = HW
+        if HW is not None:
+            H = HW[0]
+            W = HW[1]
+            self.N = H * W
+            self.C = dim
+        if mixer == "Local" and HW is not None:
+            hk = local_k[0]
+            wk = local_k[1]
+            mask = paddle.ones([H * W, H + hk - 1, W + wk - 1], dtype="float32")
+            for h in range(0, H):
+                for w in range(0, W):
+                    mask[h * W + w, h : h + hk, w : w + wk] = 0.0
+            mask_paddle = mask[:, hk // 2 : H + hk // 2, wk // 2 : W + wk // 2].flatten(
+                1
+            )
+            mask_inf = paddle.full([H * W, H * W], "-inf", dtype="float32")
+            mask = paddle.where(mask_paddle < 1, mask_paddle, mask_inf)
+            self.mask = mask.unsqueeze([0, 1])
+        self.mixer = mixer
+
+    def forward(self, x):
+        qkv = (
+            self.qkv(x)
+            .reshape((0, -1, 3, self.num_heads, self.head_dim))
+            .transpose((2, 0, 3, 1, 4))
+        )
+        q, k, v = qkv[0] * self.scale, qkv[1], qkv[2]
+
+        attn = q.matmul(k.transpose((0, 1, 3, 2)))
+        if self.mixer == "Local":
+            attn += self.mask
+        attn = nn.functional.softmax(attn, axis=-1)
+        attn = self.attn_drop(attn)
+
+        x = (attn.matmul(v)).transpose((0, 2, 1, 3)).reshape((0, -1, self.dim))
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+
+
+class Block(nn.Layer):
+    def __init__(
+        self,
+        dim,
+        num_heads,
+        mixer="Global",
+        local_mixer=[7, 11],
+        HW=None,
+        mlp_ratio=4.0,
+        qkv_bias=False,
+        qk_scale=None,
+        drop=0.0,
+        attn_drop=0.0,
+        drop_path=0.0,
+        act_layer=nn.GELU,
+        norm_layer="nn.LayerNorm",
+        epsilon=1e-6,
+        prenorm=True,
+    ):
+        super().__init__()
+        if isinstance(norm_layer, str):
+            self.norm1 = eval(norm_layer)(dim, epsilon=epsilon)
+        else:
+            self.norm1 = norm_layer(dim)
+        if mixer == "Global" or mixer == "Local":
+            self.mixer = Attention(
+                dim,
+                num_heads=num_heads,
+                mixer=mixer,
+                HW=HW,
+                local_k=local_mixer,
+                qkv_bias=qkv_bias,
+                qk_scale=qk_scale,
+                attn_drop=attn_drop,
+                proj_drop=drop,
+            )
+        elif mixer == "Conv":
+            self.mixer = ConvMixer(dim, num_heads=num_heads, HW=HW, local_k=local_mixer)
+        else:
+            raise TypeError("The mixer must be one of [Global, Local, Conv]")
+
+        self.drop_path = DropPath(drop_path) if drop_path > 0.0 else Identity()
+        if isinstance(norm_layer, str):
+            self.norm2 = eval(norm_layer)(dim, epsilon=epsilon)
+        else:
+            self.norm2 = norm_layer(dim)
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp_ratio = mlp_ratio
+        self.mlp = Mlp(
+            in_features=dim,
+            hidden_features=mlp_hidden_dim,
+            act_layer=act_layer,
+            drop=drop,
+        )
+        self.prenorm = prenorm
+
+    def forward(self, x):
+        if self.prenorm:
+            x = self.norm1(x + self.drop_path(self.mixer(x)))
+            x = self.norm2(x + self.drop_path(self.mlp(x)))
+        else:
+            x = x + self.drop_path(self.mixer(self.norm1(x)))
+            x = x + self.drop_path(self.mlp(self.norm2(x)))
+        return x
+
+
+class PatchEmbed(nn.Layer):
+    """Image to Patch Embedding"""
+
+    def __init__(
+        self,
+        img_size=[32, 100],
+        in_channels=3,
+        embed_dim=768,
+        sub_num=2,
+        patch_size=[4, 4],
+        mode="pope",
+    ):
+        super().__init__()
+        num_patches = (img_size[1] // (2**sub_num)) * (img_size[0] // (2**sub_num))
+        self.img_size = img_size
+        self.num_patches = num_patches
+        self.embed_dim = embed_dim
+        self.norm = None
+        if mode == "pope":
+            if sub_num == 2:
+                self.proj = nn.Sequential(
+                    ConvBNLayer(
+                        in_channels=in_channels,
+                        out_channels=embed_dim // 2,
+                        kernel_size=3,
+                        stride=2,
+                        padding=1,
+                        act=nn.GELU,
+                        bias_attr=None,
+                    ),
+                    ConvBNLayer(
+                        in_channels=embed_dim // 2,
+                        out_channels=embed_dim,
+                        kernel_size=3,
+                        stride=2,
+                        padding=1,
+                        act=nn.GELU,
+                        bias_attr=None,
+                    ),
+                )
+            if sub_num == 3:
+                self.proj = nn.Sequential(
+                    ConvBNLayer(
+                        in_channels=in_channels,
+                        out_channels=embed_dim // 4,
+                        kernel_size=3,
+                        stride=2,
+                        padding=1,
+                        act=nn.GELU,
+                        bias_attr=None,
+                    ),
+                    ConvBNLayer(
+                        in_channels=embed_dim // 4,
+                        out_channels=embed_dim // 2,
+                        kernel_size=3,
+                        stride=2,
+                        padding=1,
+                        act=nn.GELU,
+                        bias_attr=None,
+                    ),
+                    ConvBNLayer(
+                        in_channels=embed_dim // 2,
+                        out_channels=embed_dim,
+                        kernel_size=3,
+                        stride=2,
+                        padding=1,
+                        act=nn.GELU,
+                        bias_attr=None,
+                    ),
+                )
+        elif mode == "linear":
+            self.proj = nn.Conv2D(
+                1, embed_dim, kernel_size=patch_size, stride=patch_size
+            )
+            self.num_patches = (
+                img_size[0] // patch_size[0] * img_size[1] // patch_size[1]
+            )
+
+    def forward(self, x):
+        B, C, H, W = x.shape
+        assert (
+            H == self.img_size[0] and W == self.img_size[1]
+        ), f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})."
+        x = self.proj(x).flatten(2).transpose((0, 2, 1))
+        return x
+
+
+class SubSample(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        types="Pool",
+        stride=[2, 1],
+        sub_norm="nn.LayerNorm",
+        act=None,
+    ):
+        super().__init__()
+        self.types = types
+        if types == "Pool":
+            self.avgpool = nn.AvgPool2D(
+                kernel_size=[3, 5], stride=stride, padding=[1, 2]
+            )
+            self.maxpool = nn.MaxPool2D(
+                kernel_size=[3, 5], stride=stride, padding=[1, 2]
+            )
+            self.proj = nn.Linear(in_channels, out_channels)
+        else:
+            self.conv = nn.Conv2D(
+                in_channels,
+                out_channels,
+                kernel_size=3,
+                stride=stride,
+                padding=1,
+                weight_attr=ParamAttr(initializer=KaimingNormal()),
+            )
+        self.norm = eval(sub_norm)(out_channels)
+        if act is not None:
+            self.act = act()
+        else:
+            self.act = None
+
+    def forward(self, x):
+        if self.types == "Pool":
+            x1 = self.avgpool(x)
+            x2 = self.maxpool(x)
+            x = (x1 + x2) * 0.5
+            out = self.proj(x.flatten(2).transpose((0, 2, 1)))
+        else:
+            x = self.conv(x)
+            out = x.flatten(2).transpose((0, 2, 1))
+        out = self.norm(out)
+        if self.act is not None:
+            out = self.act(out)
+
+        return out
+
+
+class SVTRNet(nn.Layer):
+    def __init__(
+        self,
+        img_size=[32, 100],
+        in_channels=3,
+        embed_dim=[64, 128, 256],
+        depth=[3, 6, 3],
+        num_heads=[2, 4, 8],
+        mixer=["Local"] * 6 + ["Global"] * 6,  # Local atten, Global atten, Conv
+        local_mixer=[[7, 11], [7, 11], [7, 11]],
+        patch_merging="Conv",  # Conv, Pool, None
+        mlp_ratio=4,
+        qkv_bias=True,
+        qk_scale=None,
+        drop_rate=0.0,
+        last_drop=0.1,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.1,
+        norm_layer="nn.LayerNorm",
+        sub_norm="nn.LayerNorm",
+        epsilon=1e-6,
+        out_channels=192,
+        out_char_num=25,
+        block_unit="Block",
+        act="nn.GELU",
+        last_stage=True,
+        sub_num=2,
+        prenorm=True,
+        use_lenhead=False,
+        **kwargs,
+    ):
+        super().__init__()
+        self.img_size = img_size
+        self.embed_dim = embed_dim
+        self.out_channels = out_channels
+        self.prenorm = prenorm
+        patch_merging = (
+            None
+            if patch_merging != "Conv" and patch_merging != "Pool"
+            else patch_merging
+        )
+        self.patch_embed = PatchEmbed(
+            img_size=img_size,
+            in_channels=in_channels,
+            embed_dim=embed_dim[0],
+            sub_num=sub_num,
+        )
+        num_patches = self.patch_embed.num_patches
+        self.HW = [img_size[0] // (2**sub_num), img_size[1] // (2**sub_num)]
+        self.pos_embed = self.create_parameter(
+            shape=[1, num_patches, embed_dim[0]], default_initializer=zeros_
+        )
+        self.add_parameter("pos_embed", self.pos_embed)
+        self.pos_drop = nn.Dropout(p=drop_rate)
+        Block_unit = eval(block_unit)
+
+        dpr = np.linspace(0, drop_path_rate, sum(depth))
+        self.blocks1 = nn.LayerList(
+            [
+                Block_unit(
+                    dim=embed_dim[0],
+                    num_heads=num_heads[0],
+                    mixer=mixer[0 : depth[0]][i],
+                    HW=self.HW,
+                    local_mixer=local_mixer[0],
+                    mlp_ratio=mlp_ratio,
+                    qkv_bias=qkv_bias,
+                    qk_scale=qk_scale,
+                    drop=drop_rate,
+                    act_layer=eval(act),
+                    attn_drop=attn_drop_rate,
+                    drop_path=dpr[0 : depth[0]][i],
+                    norm_layer=norm_layer,
+                    epsilon=epsilon,
+                    prenorm=prenorm,
+                )
+                for i in range(depth[0])
+            ]
+        )
+        if patch_merging is not None:
+            self.sub_sample1 = SubSample(
+                embed_dim[0],
+                embed_dim[1],
+                sub_norm=sub_norm,
+                stride=[2, 1],
+                types=patch_merging,
+            )
+            HW = [self.HW[0] // 2, self.HW[1]]
+        else:
+            HW = self.HW
+        self.patch_merging = patch_merging
+        self.blocks2 = nn.LayerList(
+            [
+                Block_unit(
+                    dim=embed_dim[1],
+                    num_heads=num_heads[1],
+                    mixer=mixer[depth[0] : depth[0] + depth[1]][i],
+                    HW=HW,
+                    local_mixer=local_mixer[1],
+                    mlp_ratio=mlp_ratio,
+                    qkv_bias=qkv_bias,
+                    qk_scale=qk_scale,
+                    drop=drop_rate,
+                    act_layer=eval(act),
+                    attn_drop=attn_drop_rate,
+                    drop_path=dpr[depth[0] : depth[0] + depth[1]][i],
+                    norm_layer=norm_layer,
+                    epsilon=epsilon,
+                    prenorm=prenorm,
+                )
+                for i in range(depth[1])
+            ]
+        )
+        if patch_merging is not None:
+            self.sub_sample2 = SubSample(
+                embed_dim[1],
+                embed_dim[2],
+                sub_norm=sub_norm,
+                stride=[2, 1],
+                types=patch_merging,
+            )
+            HW = [self.HW[0] // 4, self.HW[1]]
+        else:
+            HW = self.HW
+        self.blocks3 = nn.LayerList(
+            [
+                Block_unit(
+                    dim=embed_dim[2],
+                    num_heads=num_heads[2],
+                    mixer=mixer[depth[0] + depth[1] :][i],
+                    HW=HW,
+                    local_mixer=local_mixer[2],
+                    mlp_ratio=mlp_ratio,
+                    qkv_bias=qkv_bias,
+                    qk_scale=qk_scale,
+                    drop=drop_rate,
+                    act_layer=eval(act),
+                    attn_drop=attn_drop_rate,
+                    drop_path=dpr[depth[0] + depth[1] :][i],
+                    norm_layer=norm_layer,
+                    epsilon=epsilon,
+                    prenorm=prenorm,
+                )
+                for i in range(depth[2])
+            ]
+        )
+        self.last_stage = last_stage
+        if last_stage:
+            self.avg_pool = nn.AdaptiveAvgPool2D([1, out_char_num])
+            self.last_conv = nn.Conv2D(
+                in_channels=embed_dim[2],
+                out_channels=self.out_channels,
+                kernel_size=1,
+                stride=1,
+                padding=0,
+                bias_attr=False,
+            )
+            self.hardswish = nn.Hardswish()
+            self.dropout = nn.Dropout(p=last_drop, mode="downscale_in_infer")
+        if not prenorm:
+            self.norm = eval(norm_layer)(embed_dim[-1], epsilon=epsilon)
+        self.use_lenhead = use_lenhead
+        if use_lenhead:
+            self.len_conv = nn.Linear(embed_dim[2], self.out_channels)
+            self.hardswish_len = nn.Hardswish()
+            self.dropout_len = nn.Dropout(p=last_drop, mode="downscale_in_infer")
+
+        trunc_normal_(self.pos_embed)
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight)
+            if isinstance(m, nn.Linear) and m.bias is not None:
+                zeros_(m.bias)
+        elif isinstance(m, nn.LayerNorm):
+            zeros_(m.bias)
+            ones_(m.weight)
+
+    def forward_features(self, x):
+        x = self.patch_embed(x)
+        x = x + self.pos_embed
+        x = self.pos_drop(x)
+        for blk in self.blocks1:
+            x = blk(x)
+        if self.patch_merging is not None:
+            x = self.sub_sample1(
+                x.transpose([0, 2, 1]).reshape(
+                    [0, self.embed_dim[0], self.HW[0], self.HW[1]]
+                )
+            )
+        for blk in self.blocks2:
+            x = blk(x)
+        if self.patch_merging is not None:
+            x = self.sub_sample2(
+                x.transpose([0, 2, 1]).reshape(
+                    [0, self.embed_dim[1], self.HW[0] // 2, self.HW[1]]
+                )
+            )
+        for blk in self.blocks3:
+            x = blk(x)
+        if not self.prenorm:
+            x = self.norm(x)
+        return x
+
+    def forward(self, x):
+        x = self.forward_features(x)
+        if self.use_lenhead:
+            len_x = self.len_conv(x.mean(1))
+            len_x = self.dropout_len(self.hardswish_len(len_x))
+        if self.last_stage:
+            if self.patch_merging is not None:
+                h = self.HW[0] // 4
+            else:
+                h = self.HW[0]
+            x = self.avg_pool(
+                x.transpose([0, 2, 1]).reshape([0, self.embed_dim[2], h, self.HW[1]])
+            )
+            x = self.last_conv(x)
+            x = self.hardswish(x)
+            x = self.dropout(x)
+        if self.use_lenhead:
+            return x, len_x
+        return x
diff --git a/ppocr/modeling/backbones/rec_svtrv2.py b/ppocr/modeling/backbones/rec_svtrv2.py
new file mode 100644
index 0000000..03281b3
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_svtrv2.py
@@ -0,0 +1,575 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddle import ParamAttr
+from paddle.nn.initializer import KaimingNormal
+import numpy as np
+import paddle
+import paddle.nn as nn
+from paddle.nn.initializer import TruncatedNormal, Constant, Normal
+
+trunc_normal_ = TruncatedNormal(std=0.02)
+normal_ = Normal
+zeros_ = Constant(value=0.0)
+ones_ = Constant(value=1.0)
+
+
+def drop_path(x, drop_prob=0.0, training=False):
+    """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
+    the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
+    See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ...
+    """
+    if drop_prob == 0.0 or not training:
+        return x
+    keep_prob = paddle.to_tensor(1 - drop_prob, dtype=x.dtype)
+    shape = (paddle.shape(x)[0],) + (1,) * (x.ndim - 1)
+    random_tensor = keep_prob + paddle.rand(shape, dtype=x.dtype)
+    random_tensor = paddle.floor(random_tensor)  # binarize
+    output = x.divide(keep_prob) * random_tensor
+    return output
+
+
+class DropPath(nn.Layer):
+    """Drop paths (Stochastic Depth) per sample  (when applied in main path of residual blocks)."""
+
+    def __init__(self, drop_prob=None):
+        super(DropPath, self).__init__()
+        self.drop_prob = drop_prob
+
+    def forward(self, x):
+        return drop_path(x, self.drop_prob, self.training)
+
+
+class Identity(nn.Layer):
+    def __init__(self):
+        super(Identity, self).__init__()
+
+    def forward(self, input):
+        return input
+
+
+class Mlp(nn.Layer):
+    def __init__(
+        self,
+        in_features,
+        hidden_features=None,
+        out_features=None,
+        act_layer=nn.GELU,
+        drop=0.0,
+    ):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.fc1 = nn.Linear(in_features, hidden_features)
+        self.act = act_layer()
+        self.fc2 = nn.Linear(hidden_features, out_features)
+        self.drop = nn.Dropout(drop)
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.act(x)
+        x = self.drop(x)
+        x = self.fc2(x)
+        x = self.drop(x)
+        return x
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size=3,
+        stride=1,
+        padding=0,
+        bias_attr=False,
+        groups=1,
+        act=nn.GELU,
+    ):
+        super().__init__()
+        self.conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            groups=groups,
+            weight_attr=paddle.ParamAttr(initializer=nn.initializer.KaimingUniform()),
+            bias_attr=bias_attr,
+        )
+        self.norm = nn.BatchNorm2D(out_channels)
+        self.act = act()
+
+    def forward(self, inputs):
+        out = self.conv(inputs)
+        out = self.norm(out)
+        out = self.act(out)
+        return out
+
+
+class Attention(nn.Layer):
+    def __init__(
+        self,
+        dim,
+        num_heads=8,
+        qkv_bias=False,
+        qk_scale=None,
+        attn_drop=0.0,
+        proj_drop=0.0,
+    ):
+        super().__init__()
+        self.num_heads = num_heads
+        self.dim = dim
+        self.head_dim = dim // num_heads
+        self.scale = qk_scale or self.head_dim**-0.5
+
+        self.qkv = nn.Linear(dim, dim * 3, bias_attr=qkv_bias)
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.proj = nn.Linear(dim, dim)
+        self.proj_drop = nn.Dropout(proj_drop)
+
+    def forward(self, x):
+        qkv = (
+            self.qkv(x)
+            .reshape((0, -1, 3, self.num_heads, self.head_dim))
+            .transpose((2, 0, 3, 1, 4))
+        )
+        q, k, v = qkv[0], qkv[1], qkv[2]
+
+        attn = (q.matmul(k.transpose((0, 1, 3, 2)))) * self.scale
+        attn = nn.functional.softmax(attn, axis=-1)
+        attn = self.attn_drop(attn)
+        x = (attn.matmul(v)).transpose((0, 2, 1, 3)).reshape((0, -1, self.dim))
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+
+
+class Block(nn.Layer):
+    def __init__(
+        self,
+        dim,
+        num_heads,
+        mlp_ratio=4.0,
+        qkv_bias=False,
+        qk_scale=None,
+        drop=0.0,
+        attn_drop=0.0,
+        drop_path=0.0,
+        act_layer=nn.GELU,
+        norm_layer=nn.LayerNorm,
+        epsilon=1e-6,
+    ):
+        super().__init__()
+        self.norm1 = norm_layer(dim, epsilon=epsilon)
+        self.mixer = Attention(
+            dim,
+            num_heads=num_heads,
+            qkv_bias=qkv_bias,
+            qk_scale=qk_scale,
+            attn_drop=attn_drop,
+            proj_drop=drop,
+        )
+        self.drop_path = DropPath(drop_path) if drop_path > 0.0 else Identity()
+        self.norm2 = norm_layer(dim, epsilon=epsilon)
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp_ratio = mlp_ratio
+        self.mlp = Mlp(
+            in_features=dim,
+            hidden_features=mlp_hidden_dim,
+            act_layer=act_layer,
+            drop=drop,
+        )
+
+    def forward(self, x):
+        x = self.norm1(x + self.drop_path(self.mixer(x)))
+        x = self.norm2(x + self.drop_path(self.mlp(x)))
+        return x
+
+
+class ConvBlock(nn.Layer):
+    def __init__(
+        self,
+        dim,
+        num_heads,
+        mlp_ratio=4.0,
+        drop=0.0,
+        drop_path=0.0,
+        act_layer=nn.GELU,
+        norm_layer=nn.LayerNorm,
+        epsilon=1e-6,
+    ):
+        super().__init__()
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.norm1 = norm_layer(dim, epsilon=epsilon)
+        self.mixer = nn.Conv2D(
+            dim,
+            dim,
+            5,
+            1,
+            2,
+            groups=num_heads,
+            weight_attr=ParamAttr(initializer=KaimingNormal()),
+        )
+        self.drop_path = DropPath(drop_path) if drop_path > 0.0 else Identity()
+        self.norm2 = norm_layer(dim, epsilon=epsilon)
+        self.mlp = Mlp(
+            in_features=dim,
+            hidden_features=mlp_hidden_dim,
+            act_layer=act_layer,
+            drop=drop,
+        )
+
+    def forward(self, x):
+        C, H, W = x.shape[1:]
+        x = x + self.drop_path(self.mixer(x))
+        x = self.norm1(x.flatten(2).transpose([0, 2, 1]))
+        x = self.norm2(x + self.drop_path(self.mlp(x)))
+        x = x.transpose([0, 2, 1]).reshape([0, C, H, W])
+        return x
+
+
+class FlattenTranspose(nn.Layer):
+    def forward(self, x):
+        return x.flatten(2).transpose([0, 2, 1])
+
+
+class SubSample2D(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        stride=[2, 1],
+    ):
+        super().__init__()
+        self.conv = nn.Conv2D(
+            in_channels,
+            out_channels,
+            kernel_size=3,
+            stride=stride,
+            padding=1,
+            weight_attr=ParamAttr(initializer=KaimingNormal()),
+        )
+        self.norm = nn.LayerNorm(out_channels)
+
+    def forward(self, x, sz):
+        # print(x.shape)
+        x = self.conv(x)
+        C, H, W = x.shape[1:]
+        x = self.norm(x.flatten(2).transpose([0, 2, 1]))
+        x = x.transpose([0, 2, 1]).reshape([0, C, H, W])
+        return x, [H, W]
+
+
+class SubSample1D(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        stride=[2, 1],
+    ):
+        super().__init__()
+        self.conv = nn.Conv2D(
+            in_channels,
+            out_channels,
+            kernel_size=3,
+            stride=stride,
+            padding=1,
+            weight_attr=ParamAttr(initializer=KaimingNormal()),
+        )
+        self.norm = nn.LayerNorm(out_channels)
+
+    def forward(self, x, sz):
+        C = x.shape[-1]
+        x = x.transpose([0, 2, 1]).reshape([0, C, sz[0], sz[1]])
+        x = self.conv(x)
+        C, H, W = x.shape[1:]
+        x = self.norm(x.flatten(2).transpose([0, 2, 1]))
+        return x, [H, W]
+
+
+class IdentitySize(nn.Layer):
+    def forward(self, x, sz):
+        return x, sz
+
+
+class SVTRStage(nn.Layer):
+    def __init__(
+        self,
+        dim=64,
+        out_dim=256,
+        depth=3,
+        mixer=["Local"] * 3,
+        sub_k=[2, 1],
+        num_heads=2,
+        mlp_ratio=4,
+        qkv_bias=True,
+        qk_scale=None,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path=[0.1] * 3,
+        norm_layer=nn.LayerNorm,
+        act=nn.GELU,
+        eps=1e-6,
+        downsample=None,
+        **kwargs,
+    ):
+        super().__init__()
+        self.dim = dim
+
+        conv_block_num = sum([1 if mix == "Conv" else 0 for mix in mixer])
+        blocks = []
+        for i in range(depth):
+            if mixer[i] == "Conv":
+                blocks.append(
+                    ConvBlock(
+                        dim=dim,
+                        num_heads=num_heads,
+                        mlp_ratio=mlp_ratio,
+                        drop=drop_rate,
+                        act_layer=act,
+                        drop_path=drop_path[i],
+                        norm_layer=norm_layer,
+                        epsilon=eps,
+                    )
+                )
+            else:
+                blocks.append(
+                    Block(
+                        dim=dim,
+                        num_heads=num_heads,
+                        mlp_ratio=mlp_ratio,
+                        qkv_bias=qkv_bias,
+                        qk_scale=qk_scale,
+                        drop=drop_rate,
+                        act_layer=act,
+                        attn_drop=attn_drop_rate,
+                        drop_path=drop_path[i],
+                        norm_layer=norm_layer,
+                        epsilon=eps,
+                    )
+                )
+            if i == conv_block_num - 1 and mixer[-1] != "Conv":
+                blocks.append(FlattenTranspose())
+        self.blocks = nn.Sequential(*blocks)
+        if downsample:
+            if mixer[-1] == "Conv":
+                self.downsample = SubSample2D(dim, out_dim, stride=sub_k)
+            elif mixer[-1] == "Global":
+                self.downsample = SubSample1D(dim, out_dim, stride=sub_k)
+        else:
+            self.downsample = IdentitySize()
+
+    def forward(self, x, sz):
+        x = self.blocks(x)
+        x, sz = self.downsample(x, sz)
+        return x, sz
+
+
+class ADDPosEmbed(nn.Layer):
+    def __init__(self, feat_max_size=[8, 32], embed_dim=768):
+        super().__init__()
+        pos_embed = paddle.zeros(
+            [1, feat_max_size[0] * feat_max_size[1], embed_dim], dtype=paddle.float32
+        )
+        trunc_normal_(pos_embed)
+        pos_embed = pos_embed.transpose([0, 2, 1]).reshape(
+            [1, embed_dim, feat_max_size[0], feat_max_size[1]]
+        )
+        self.pos_embed = self.create_parameter(
+            [1, embed_dim, feat_max_size[0], feat_max_size[1]]
+        )
+        self.add_parameter("pos_embed", self.pos_embed)
+        self.pos_embed.set_value(pos_embed)
+
+    def forward(self, x):
+        sz = x.shape[2:]
+        x = x + self.pos_embed[:, :, : sz[0], : sz[1]]
+        return x
+
+
+class POPatchEmbed(nn.Layer):
+    """Image to Patch Embedding"""
+
+    def __init__(
+        self,
+        in_channels=3,
+        feat_max_size=[8, 32],
+        embed_dim=768,
+        use_pos_embed=False,
+        flatten=False,
+    ):
+        super().__init__()
+        patch_embed = [
+            ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=embed_dim // 2,
+                kernel_size=3,
+                stride=2,
+                padding=1,
+                act=nn.GELU,
+                bias_attr=None,
+            ),
+            ConvBNLayer(
+                in_channels=embed_dim // 2,
+                out_channels=embed_dim,
+                kernel_size=3,
+                stride=2,
+                padding=1,
+                act=nn.GELU,
+                bias_attr=None,
+            ),
+        ]
+        if use_pos_embed:
+            patch_embed.append(ADDPosEmbed(feat_max_size, embed_dim))
+        if flatten:
+            patch_embed.append(FlattenTranspose())
+        self.patch_embed = nn.Sequential(*patch_embed)
+
+    def forward(self, x):
+        sz = x.shape[2:]
+        x = self.patch_embed(x)
+        return x, [sz[0] // 4, sz[1] // 4]
+
+
+class LastStage(nn.Layer):
+    def __init__(self, in_channels, out_channels, last_drop, out_char_num):
+        super().__init__()
+        self.last_conv = nn.Linear(in_channels, out_channels, bias_attr=False)
+        self.hardswish = nn.Hardswish()
+        self.dropout = nn.Dropout(p=last_drop, mode="downscale_in_infer")
+
+    def forward(self, x, sz):
+        x = x.reshape([0, sz[0], sz[1], x.shape[-1]])
+        x = x.mean(1)
+        x = self.last_conv(x)
+        x = self.hardswish(x)
+        x = self.dropout(x)
+        return x, [1, sz[1]]
+
+
+class OutPool(nn.Layer):
+    def __init__(self):
+        super().__init__()
+
+    def forward(self, x, sz):
+        C = x.shape[-1]
+        x = x.transpose([0, 2, 1]).reshape([0, C, sz[0], sz[1]])
+        x = nn.functional.avg_pool2d(x, [sz[0], 2])
+        return x, [1, sz[1] // 2]
+
+
+class Feat2D(nn.Layer):
+    def __init__(self):
+        super().__init__()
+
+    def forward(self, x, sz):
+        C = x.shape[-1]
+        x = x.transpose([0, 2, 1]).reshape([0, C, sz[0], sz[1]])
+        return x, sz
+
+
+class SVTRv2(nn.Layer):
+    def __init__(
+        self,
+        max_sz=[32, 128],
+        in_channels=3,
+        out_channels=192,
+        out_char_num=25,
+        depths=[3, 6, 3],
+        dims=[64, 128, 256],
+        mixer=[["Conv"] * 3, ["Conv"] * 3 + ["Global"] * 3, ["Global"] * 3],
+        use_pos_embed=False,
+        sub_k=[[1, 1], [2, 1], [1, 1]],
+        num_heads=[2, 4, 8],
+        mlp_ratio=4,
+        qkv_bias=True,
+        qk_scale=None,
+        drop_rate=0.0,
+        last_drop=0.1,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.1,
+        norm_layer=nn.LayerNorm,
+        act=nn.GELU,
+        last_stage=False,
+        eps=1e-6,
+        use_pool=False,
+        feat2d=False,
+        **kwargs,
+    ):
+        super().__init__()
+        num_stages = len(depths)
+        self.num_features = dims[-1]
+
+        feat_max_size = [max_sz[0] // 4, max_sz[1] // 4]
+        self.pope = POPatchEmbed(
+            in_channels=in_channels,
+            feat_max_size=feat_max_size,
+            embed_dim=dims[0],
+            use_pos_embed=use_pos_embed,
+            flatten=mixer[0][0] != "Conv",
+        )
+
+        dpr = np.linspace(0, drop_path_rate, sum(depths))  # stochastic depth decay rule
+
+        self.stages = nn.LayerList()
+        for i_stage in range(num_stages):
+            stage = SVTRStage(
+                dim=dims[i_stage],
+                out_dim=dims[i_stage + 1] if i_stage < num_stages - 1 else 0,
+                depth=depths[i_stage],
+                mixer=mixer[i_stage],
+                sub_k=sub_k[i_stage],
+                num_heads=num_heads[i_stage],
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                qk_scale=qk_scale,
+                drop=drop_rate,
+                attn_drop=attn_drop_rate,
+                drop_path=dpr[sum(depths[:i_stage]) : sum(depths[: i_stage + 1])],
+                norm_layer=norm_layer,
+                act=act,
+                downsample=False if i_stage == num_stages - 1 else True,
+                eps=eps,
+            )
+            self.stages.append(stage)
+
+        self.out_channels = self.num_features
+        self.last_stage = last_stage
+        if last_stage:
+            self.out_channels = out_channels
+            self.stages.append(
+                LastStage(self.num_features, out_channels, last_drop, out_char_num)
+            )
+        if use_pool:
+            self.stages.append(OutPool())
+
+        if feat2d:
+            self.stages.append(Feat2D())
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight)
+            if isinstance(m, nn.Linear) and m.bias is not None:
+                zeros_(m.bias)
+        elif isinstance(m, nn.LayerNorm):
+            zeros_(m.bias)
+            ones_(m.weight)
+
+    def forward(self, x):
+        x, sz = self.pope(x)
+        for stage in self.stages:
+            x, sz = stage(x, sz)
+        return x
diff --git a/ppocr/modeling/backbones/rec_vary_vit.py b/ppocr/modeling/backbones/rec_vary_vit.py
new file mode 100644
index 0000000..d912ecf
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_vary_vit.py
@@ -0,0 +1,616 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import math
+from functools import partial
+from typing import Optional, Tuple, Type
+
+import numpy as np
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle.nn.initializer import (
+    Constant,
+    KaimingUniform,
+    Normal,
+    TruncatedNormal,
+    XavierUniform,
+)
+from ppocr.modeling.backbones.rec_donut_swin import DonutSwinModelOutput
+
+zeros_ = Constant(value=0.0)
+ones_ = Constant(value=1.0)
+kaiming_normal_ = KaimingUniform(nonlinearity="relu")
+trunc_normal_ = TruncatedNormal(std=0.02)
+xavier_uniform_ = XavierUniform()
+
+
+class MLPBlock(nn.Layer):
+    def __init__(
+        self,
+        embedding_dim: int,
+        mlp_dim: int,
+        act: Type[nn.Layer] = nn.GELU,
+    ) -> None:
+        super().__init__()
+        self.lin1 = nn.Linear(embedding_dim, mlp_dim)
+        self.lin2 = nn.Linear(mlp_dim, embedding_dim)
+        self.act = act()
+
+    def forward(self, x):
+        return self.lin2(self.act(self.lin1(x)))
+
+
+# From https://github.com/facebookresearch/detectron2/blob/main/detectron2/layers/batch_norm.py # noqa
+# Itself from https://github.com/facebookresearch/ConvNeXt/blob/d1fa8f6fef0a165b27399986cc2bdacc92777e40/models/convnext.py#L119  # noqa
+class LayerNorm2d(nn.Layer):
+    def __init__(self, num_channels: int, epsilon: float = 1e-6) -> None:
+        super().__init__()
+        self.weight = paddle.create_parameter([num_channels], dtype="float32")
+        ones_(self.weight)
+        self.bias = paddle.create_parameter([num_channels], dtype="float32")
+        zeros_(self.bias)
+        self.epsilon = epsilon
+
+    def forward(self, x):
+        u = x.mean(1, keepdim=True)
+        s = (x - u).pow(2).mean(1, keepdim=True)
+        x = (x - u) / paddle.sqrt(s + self.epsilon)
+        x = self.weight[:, None, None] * x + self.bias[:, None, None]
+        return x
+
+
+# This class and its supporting functions below lightly adapted from the ViTDet backbone available at: https://github.com/facebookresearch/detectron2/blob/main/detectron2/modeling/backbone/vit.py # noqa
+class ImageEncoderViT(nn.Layer):
+    def __init__(
+        self,
+        img_size: int = 1024,
+        patch_size: int = 16,
+        in_chans: int = 3,
+        embed_dim: int = 768,
+        depth: int = 12,
+        num_heads: int = 12,
+        mlp_ratio: float = 4.0,
+        out_chans: int = 256,
+        qkv_bias: bool = True,
+        norm_layer: Type[nn.Layer] = nn.LayerNorm,
+        act_layer: Type[nn.Layer] = nn.GELU,
+        use_abs_pos: bool = True,
+        use_rel_pos: bool = False,
+        rel_pos_zero_init: bool = True,
+        window_size: int = 0,
+        global_attn_indexes: Tuple[int, ...] = (),
+        is_formula: bool = False,
+    ) -> None:
+        """
+        Args:
+            img_size (int): Input image size.
+            patch_size (int): Patch size.
+            in_chans (int): Number of input image channels.
+            embed_dim (int): Patch embedding dimension.
+            depth (int): Depth of ViT.
+            num_heads (int): Number of attention heads in each ViT block.
+            mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
+            qkv_bias (bool): If True, add a learnable bias to query, key, value.
+            norm_layer (nn.Layer): Normalization layer.
+            act_layer (nn.Layer): Activation layer.
+            use_abs_pos (bool): If True, use absolute positional embeddings.
+            use_rel_pos (bool): If True, add relative positional embeddings to the attention map.
+            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
+            window_size (int): Window size for window attention blocks.
+            global_attn_indexes (list): Indexes for blocks using global attention.
+        """
+        super().__init__()
+        self.img_size = img_size
+
+        self.patch_embed = PatchEmbed(
+            kernel_size=(patch_size, patch_size),
+            stride=(patch_size, patch_size),
+            in_chans=in_chans,
+            embed_dim=embed_dim,
+        )
+
+        self.pos_embed = None
+        if use_abs_pos:
+            # Initialize absolute positional embedding with pretrain image size.
+            self.pos_embed = paddle.create_parameter(
+                shape=(1, img_size // patch_size, img_size // patch_size, embed_dim),
+                dtype="float32",
+            )
+            zeros_(self.pos_embed)
+
+        self.blocks = nn.LayerList()
+        for i in range(depth):
+            block = Block(
+                dim=embed_dim,
+                num_heads=num_heads,
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                norm_layer=norm_layer,
+                act_layer=act_layer,
+                use_rel_pos=use_rel_pos,
+                rel_pos_zero_init=rel_pos_zero_init,
+                window_size=window_size if i not in global_attn_indexes else 0,
+                input_size=(img_size // patch_size, img_size // patch_size),
+            )
+            self.blocks.append(block)
+
+        self.neck = nn.Sequential(
+            nn.Conv2D(
+                embed_dim,
+                out_chans,
+                kernel_size=1,
+                bias_attr=False,
+            ),
+            LayerNorm2d(out_chans),
+            nn.Conv2D(
+                out_chans,
+                out_chans,
+                kernel_size=3,
+                padding=1,
+                bias_attr=False,
+            ),
+            LayerNorm2d(out_chans),
+        )
+
+        self.net_2 = nn.Conv2D(
+            256, 512, kernel_size=3, stride=2, padding=1, bias_attr=False
+        )
+        self.net_3 = nn.Conv2D(
+            512, 1024, kernel_size=3, stride=2, padding=1, bias_attr=False
+        )
+        self.is_formula = is_formula
+
+    def forward(self, x):
+        x = self.patch_embed(x)
+        if self.pos_embed is not None:
+            x = x + self.pos_embed
+        for blk in self.blocks:
+            x = blk(x)
+        x = self.neck(x.transpose([0, 3, 1, 2]))
+        x = self.net_2(x)
+        if self.is_formula:
+            x = self.net_3(x)
+        return x
+
+
+class Block(nn.Layer):
+    """Transformer blocks with support of window attention and residual propagation blocks"""
+
+    def __init__(
+        self,
+        dim: int,
+        num_heads: int,
+        mlp_ratio: float = 4.0,
+        qkv_bias: bool = True,
+        norm_layer: Type[nn.Layer] = nn.LayerNorm,
+        act_layer: Type[nn.Layer] = nn.GELU,
+        use_rel_pos: bool = False,
+        rel_pos_zero_init: bool = True,
+        window_size: int = 0,
+        input_size: Optional[Tuple[int, int]] = None,
+    ) -> None:
+        """
+        Args:
+            dim (int): Number of input channels.
+            num_heads (int): Number of attention heads in each ViT block.
+            mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
+            qkv_bias (bool): If True, add a learnable bias to query, key, value.
+            norm_layer (nn.Layer): Normalization layer.
+            act_layer (nn.Layer): Activation layer.
+            use_rel_pos (bool): If True, add relative positional embeddings to the attention map.
+            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
+            window_size (int): Window size for window attention blocks. If it equals 0, then
+                use global attention.
+            input_size (tuple(int, int) or None): Input resolution for calculating the relative
+                positional parameter size.
+        """
+        super().__init__()
+        self.norm1 = norm_layer(dim)
+        self.attn = Attention(
+            dim,
+            num_heads=num_heads,
+            qkv_bias=qkv_bias,
+            use_rel_pos=use_rel_pos,
+            rel_pos_zero_init=rel_pos_zero_init,
+            input_size=input_size if window_size == 0 else (window_size, window_size),
+        )
+
+        self.norm2 = norm_layer(dim)
+        self.mlp = MLPBlock(
+            embedding_dim=dim, mlp_dim=int(dim * mlp_ratio), act=act_layer
+        )
+
+        self.window_size = window_size
+
+    def forward(self, x):
+        shortcut = x
+
+        x = self.norm1(x)
+        # Window partition
+        if self.window_size > 0:
+            H, W = x.shape[1], x.shape[2]
+            x, pad_hw = window_partition(x, self.window_size)
+        x = self.attn(x)
+        # Reverse window partition
+        if self.window_size > 0:
+            x = window_unpartition(x, self.window_size, pad_hw, (H, W))
+        x = shortcut + x
+        x = x + self.mlp(self.norm2(x))
+
+        return x
+
+
+class Attention(nn.Layer):
+    """Multi-head Attention block with relative position embeddings."""
+
+    def __init__(
+        self,
+        dim: int,
+        num_heads: int = 8,
+        qkv_bias: bool = True,
+        use_rel_pos: bool = False,
+        rel_pos_zero_init: bool = True,
+        input_size: Optional[Tuple[int, int]] = None,
+    ) -> None:
+        """
+        Args:
+            dim (int): Number of input channels.
+            num_heads (int): Number of attention heads.
+            qkv_bias (bool):  If True, add a learnable bias to query, key, value.
+            rel_pos (bool): If True, add relative positional embeddings to the attention map.
+            rel_pos_zero_init (bool): If True, zero initialize relative positional parameters.
+            input_size (tuple(int, int) or None): Input resolution for calculating the relative
+                positional parameter size.
+        """
+        super().__init__()
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        self.scale = head_dim**-0.5
+
+        self.qkv = nn.Linear(dim, dim * 3, bias_attr=qkv_bias)
+        self.proj = nn.Linear(dim, dim)
+
+        self.use_rel_pos = use_rel_pos
+        if self.use_rel_pos:
+            assert (
+                input_size is not None
+            ), "Input size must be provided if using relative positional encoding."
+            # initialize relative positional embeddings
+            self.rel_pos_h = paddle.create_parameter(
+                [2 * input_size[0] - 1, head_dim], dtype="float32"
+            )
+            zeros_(self.rel_pos_h)
+            self.rel_pos_w = paddle.create_parameter(
+                [2 * input_size[1] - 1, head_dim], dtype="float32"
+            )
+            zeros_(self.rel_pos_w)
+
+    def forward(self, x):
+
+        B, H, W, _ = x.shape
+        qkv = (
+            self.qkv(x)
+            .reshape([B, H * W, 3, self.num_heads, -1])
+            .transpose([2, 0, 3, 1, 4])
+        )
+        q, k, v = qkv.reshape([3, B * self.num_heads, H * W, -1]).unbind(0)
+        attn = (q * self.scale) @ k.transpose([0, 2, 1])
+
+        if self.use_rel_pos:
+            attn = add_decomposed_rel_pos(
+                attn, q, self.rel_pos_h, self.rel_pos_w, (H, W), (H, W)
+            )
+        attn = F.softmax(attn, axis=-1)
+        x = (
+            (attn @ v)
+            .reshape([B, self.num_heads, H, W, -1])
+            .transpose([0, 2, 3, 1, 4])
+            .reshape([B, H, W, -1])
+        )
+        x = self.proj(x)
+
+        return x
+
+
+def window_partition(x, window_size: int):
+    """
+    Partition into non-overlapping windows with padding if needed.
+    Args:
+        x (tensor): input tokens with [B, H, W, C].
+        window_size (int): window size.
+
+    Returns:
+        windows: windows after partition with [B * num_windows, window_size, window_size, C].
+        (Hp, Wp): padded height and width before partition
+    """
+    B, H, W, C = x.shape
+
+    pad_h = (window_size - H % window_size) % window_size
+    pad_w = (window_size - W % window_size) % window_size
+    if pad_h > 0 or pad_w > 0:
+        x = F.pad(x, (0, 0, 0, pad_w, 0, pad_h, 0, 0))
+    Hp, Wp = H + pad_h, W + pad_w
+
+    x = x.reshape(
+        [B, Hp // window_size, window_size, Wp // window_size, window_size, C]
+    )
+    windows = x.transpose([0, 1, 3, 2, 4, 5]).reshape([-1, window_size, window_size, C])
+    return windows, (Hp, Wp)
+
+
+def window_unpartition(
+    windows, window_size: int, pad_hw: Tuple[int, int], hw: Tuple[int, int]
+):
+    """
+    Window unpartition into original sequences and removing padding.
+    Args:
+        windows (tensor): input tokens with [B * num_windows, window_size, window_size, C].
+        window_size (int): window size.
+        pad_hw (Tuple): padded height and width (Hp, Wp).
+        hw (Tuple): original height and width (H, W) before padding.
+
+    Returns:
+        x: unpartitioned sequences with [B, H, W, C].
+    """
+    Hp, Wp = pad_hw
+    H, W = hw
+    B = windows.shape[0] // (Hp * Wp // window_size // window_size)
+    x = windows.reshape(
+        [B, Hp // window_size, Wp // window_size, window_size, window_size, -1]
+    )
+    x = x.transpose([0, 1, 3, 2, 4, 5]).contiguous().reshape([B, Hp, Wp, -1])
+
+    if Hp > H or Wp > W:
+        x = x[:, :H, :W, :].contiguous()
+    return x
+
+
+def get_rel_pos(q_size: int, k_size: int, rel_pos):
+    """
+    Get relative positional embeddings according to the relative positions of
+        query and key sizes.
+    Args:
+        q_size (int): size of query q.
+        k_size (int): size of key k.
+        rel_pos (Tensor): relative position embeddings (L, C).
+
+    Returns:
+        Extracted positional embeddings according to relative positions.
+    """
+    max_rel_dist = int(2 * max(q_size, k_size) - 1)
+    # Interpolate rel pos if needed.
+    if rel_pos.shape[0] != max_rel_dist:
+        # Interpolate rel pos.
+        rel_pos_resized = F.interpolate(
+            rel_pos.reshape(1, rel_pos.shape[0], -1).transpose(0, 2, 1),
+            size=max_rel_dist,
+            mode="linear",
+        )
+        rel_pos_resized = rel_pos_resized.reshape(-1, max_rel_dist).transpose(1, 0)
+    else:
+        rel_pos_resized = rel_pos
+
+    # Scale the coords with short length if shapes for q and k are different.
+    q_coords = paddle.arange(q_size)[:, None] * max(k_size / q_size, 1.0)
+    k_coords = paddle.arange(k_size)[None, :] * max(q_size / k_size, 1.0)
+    relative_coords = (q_coords - k_coords) + (k_size - 1) * max(q_size / k_size, 1.0)
+
+    return rel_pos_resized[relative_coords.cast(paddle.int64)]
+
+
+def add_decomposed_rel_pos(
+    attn,
+    q,
+    rel_pos_h,
+    rel_pos_w,
+    q_size: Tuple[int, int],
+    k_size: Tuple[int, int],
+):
+    """
+    Calculate decomposed Relative Positional Embeddings from :paper:`mvitv2`.
+    https://github.com/facebookresearch/mvit/blob/19786631e330df9f3622e5402b4a419a263a2c80/mvit/models/attention.py   # noqa B950
+    Args:
+        attn (Tensor): attention map.
+        q (Tensor): query q in the attention layer with shape (B, q_h * q_w, C).
+        rel_pos_h (Tensor): relative position embeddings (Lh, C) for height axis.
+        rel_pos_w (Tensor): relative position embeddings (Lw, C) for width axis.
+        q_size (Tuple): spatial sequence size of query q with (q_h, q_w).
+        k_size (Tuple): spatial sequence size of key k with (k_h, k_w).
+
+    Returns:
+        attn (Tensor): attention map with added relative positional embeddings.
+    """
+    q_h, q_w = q_size
+    k_h, k_w = k_size
+    Rh = get_rel_pos(q_h, k_h, rel_pos_h)
+    Rw = get_rel_pos(q_w, k_w, rel_pos_w)
+
+    B, _, dim = q.shape
+    r_q = q.reshape([B, q_h, q_w, dim])
+    rel_h = paddle.einsum("bhwc,hkc->bhwk", r_q, Rh)
+    rel_w = paddle.einsum("bhwc,wkc->bhwk", r_q, Rw)
+
+    attn = (
+        attn.reshape([B, q_h, q_w, k_h, k_w])
+        + rel_h[:, :, :, :, None]
+        + rel_w[:, :, :, None, :]
+    ).reshape([B, q_h * q_w, k_h * k_w])
+
+    return attn
+
+
+class PatchEmbed(nn.Layer):
+    """
+    Image to Patch Embedding.
+    """
+
+    def __init__(
+        self,
+        kernel_size: Tuple[int, int] = (16, 16),
+        stride: Tuple[int, int] = (16, 16),
+        padding: Tuple[int, int] = (0, 0),
+        in_chans: int = 3,
+        embed_dim: int = 768,
+    ) -> None:
+        """
+        Args:
+            kernel_size (Tuple): kernel size of the projection layer.
+            stride (Tuple): stride of the projection layer.
+            padding (Tuple): padding size of the projection layer.
+            in_chans (int): Number of input image channels.
+            embed_dim (int): Patch embedding dimension.
+        """
+        super().__init__()
+
+        self.proj = nn.Conv2D(
+            in_chans,
+            embed_dim,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            weight_attr=True,
+            bias_attr=True,
+        )
+
+    def forward(self, x):
+        x = self.proj(x)
+        # B C H W -> B H W C
+        x = x.transpose([0, 2, 3, 1])
+        return x
+
+
+def _build_vary(
+    encoder_embed_dim,
+    encoder_depth,
+    encoder_num_heads,
+    encoder_global_attn_indexes,
+    image_size,
+    is_formula=False,
+):
+    prompt_embed_dim = 256
+    vit_patch_size = 16
+    image_embedding_size = image_size // vit_patch_size
+    image_encoder = ImageEncoderViT(
+        depth=encoder_depth,
+        embed_dim=encoder_embed_dim,
+        img_size=image_size,
+        mlp_ratio=4,
+        norm_layer=partial(paddle.nn.LayerNorm, epsilon=1e-6),
+        num_heads=encoder_num_heads,
+        patch_size=vit_patch_size,
+        qkv_bias=True,
+        use_rel_pos=True,
+        global_attn_indexes=encoder_global_attn_indexes,
+        window_size=14,
+        out_chans=prompt_embed_dim,
+        is_formula=is_formula,
+    )
+    return image_encoder
+
+
+class Vary_VIT_B(nn.Layer):
+    def __init__(
+        self,
+        in_channels=3,
+        image_size=768,
+        encoder_embed_dim=768,
+        encoder_depth=12,
+        encoder_num_heads=12,
+        encoder_global_attn_indexes=[2, 5, 8, 11],
+    ):
+        super().__init__()
+
+        self.vision_tower_high = _build_vary(
+            encoder_embed_dim=768,
+            encoder_depth=12,
+            encoder_num_heads=12,
+            encoder_global_attn_indexes=[2, 5, 8, 11],
+            image_size=image_size,
+        )
+
+        self.out_channels = 1024
+
+    def forward(self, input_data):
+        pixel_values = input_data
+        num_channels = pixel_values.shape[1]
+        if num_channels == 1:
+            pixel_values = paddle.repeat_interleave(pixel_values, repeats=3, axis=1)
+        cnn_feature = self.vision_tower_high(pixel_values)
+        cnn_feature = cnn_feature.flatten(2).transpose([0, 2, 1])
+        return cnn_feature
+
+
+class Vary_VIT_B_Formula(nn.Layer):
+    def __init__(
+        self,
+        in_channels=3,
+        image_size=768,
+        encoder_embed_dim=768,
+        encoder_depth=12,
+        encoder_num_heads=12,
+        encoder_global_attn_indexes=[2, 5, 8, 11],
+    ):
+        """
+        Vary_VIT_B_Formula
+        Args:
+            in_channels (int): Number of input channels. Default is 3 (for RGB images).
+            image_size (int): Size of the input image. Default is 768.
+            encoder_embed_dim (int): Dimension of the encoder's embedding. Default is 768.
+            encoder_depth (int): Number of layers (depth) in the encoder. Default is 12.
+            encoder_num_heads (int): Number of attention heads in the encoder. Default is 12.
+            encoder_global_attn_indexes (list): List of indices specifying which encoder layers use global attention. Default is [2, 5, 8, 11].
+        Returns:
+            model: nn.Layer. Specific `Vary_VIT_B_Formula` model with defined architecture.
+        """
+        super(Vary_VIT_B_Formula, self).__init__()
+
+        self.vision_tower_high = _build_vary(
+            encoder_embed_dim=encoder_embed_dim,
+            encoder_depth=encoder_depth,
+            encoder_num_heads=encoder_num_heads,
+            encoder_global_attn_indexes=[2, 5, 8, 11],
+            image_size=image_size,
+            is_formula=True,
+        )
+        self.mm_projector_vary = nn.Linear(1024, 1024)
+        self.out_channels = 1024
+
+    def forward(self, input_data):
+        if self.training:
+            pixel_values, label, attention_mask = input_data
+        else:
+            if isinstance(input_data, list):
+                pixel_values = input_data[0]
+            else:
+                pixel_values = input_data
+        num_channels = pixel_values.shape[1]
+        if num_channels == 1:
+            pixel_values = paddle.repeat_interleave(pixel_values, repeats=3, axis=1)
+
+        cnn_feature = self.vision_tower_high(pixel_values)
+        cnn_feature = cnn_feature.flatten(2).transpose([0, 2, 1])
+
+        cnn_feature = self.mm_projector_vary(cnn_feature)
+        donut_swin_output = DonutSwinModelOutput(
+            last_hidden_state=cnn_feature,
+            pooler_output=None,
+            hidden_states=None,
+            attentions=None,
+            reshaped_hidden_states=None,
+        )
+        if self.training:
+            return donut_swin_output, label, attention_mask
+        else:
+            return donut_swin_output
diff --git a/ppocr/modeling/backbones/rec_vit.py b/ppocr/modeling/backbones/rec_vit.py
new file mode 100644
index 0000000..165b868
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_vit.py
@@ -0,0 +1,273 @@
+# copyright (c) 2023 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddle import ParamAttr
+from paddle.nn.initializer import KaimingNormal
+import numpy as np
+import paddle
+import paddle.nn as nn
+from paddle.nn.initializer import TruncatedNormal, Constant, Normal
+
+trunc_normal_ = TruncatedNormal(std=0.02)
+normal_ = Normal
+zeros_ = Constant(value=0.0)
+ones_ = Constant(value=1.0)
+
+
+def drop_path(x, drop_prob=0.0, training=False):
+    """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
+    the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
+    See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ...
+    """
+    if drop_prob == 0.0 or not training:
+        return x
+    keep_prob = paddle.to_tensor(1 - drop_prob)
+    shape = (x.shape[0],) + (1,) * (x.ndim - 1)
+    random_tensor = keep_prob + paddle.rand(shape, dtype=x.dtype)
+    random_tensor = paddle.floor(random_tensor)  # binarize
+    output = x.divide(keep_prob) * random_tensor
+    return output
+
+
+class DropPath(nn.Layer):
+    """Drop paths (Stochastic Depth) per sample  (when applied in main path of residual blocks)."""
+
+    def __init__(self, drop_prob=None):
+        super(DropPath, self).__init__()
+        self.drop_prob = drop_prob
+
+    def forward(self, x):
+        return drop_path(x, self.drop_prob, self.training)
+
+
+class Identity(nn.Layer):
+    def __init__(self):
+        super(Identity, self).__init__()
+
+    def forward(self, input):
+        return input
+
+
+class Mlp(nn.Layer):
+    def __init__(
+        self,
+        in_features,
+        hidden_features=None,
+        out_features=None,
+        act_layer=nn.GELU,
+        drop=0.0,
+    ):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.fc1 = nn.Linear(in_features, hidden_features)
+        self.act = act_layer()
+        self.fc2 = nn.Linear(hidden_features, out_features)
+        self.drop = nn.Dropout(drop)
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.act(x)
+        x = self.drop(x)
+        x = self.fc2(x)
+        x = self.drop(x)
+        return x
+
+
+class Attention(nn.Layer):
+    def __init__(
+        self,
+        dim,
+        num_heads=8,
+        qkv_bias=False,
+        qk_scale=None,
+        attn_drop=0.0,
+        proj_drop=0.0,
+    ):
+        super().__init__()
+        self.num_heads = num_heads
+        self.dim = dim
+        head_dim = dim // num_heads
+        self.scale = qk_scale or head_dim**-0.5
+
+        self.qkv = nn.Linear(dim, dim * 3, bias_attr=qkv_bias)
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.proj = nn.Linear(dim, dim)
+        self.proj_drop = nn.Dropout(proj_drop)
+
+    def forward(self, x):
+        qkv = paddle.reshape(
+            self.qkv(x), (0, -1, 3, self.num_heads, self.dim // self.num_heads)
+        ).transpose((2, 0, 3, 1, 4))
+        q, k, v = qkv[0] * self.scale, qkv[1], qkv[2]
+
+        attn = q.matmul(k.transpose((0, 1, 3, 2)))
+        attn = nn.functional.softmax(attn, axis=-1)
+        attn = self.attn_drop(attn)
+
+        x = (attn.matmul(v)).transpose((0, 2, 1, 3)).reshape((0, -1, self.dim))
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+
+
+class Block(nn.Layer):
+    def __init__(
+        self,
+        dim,
+        num_heads,
+        mlp_ratio=4.0,
+        qkv_bias=False,
+        qk_scale=None,
+        drop=0.0,
+        attn_drop=0.0,
+        drop_path=0.0,
+        act_layer=nn.GELU,
+        norm_layer="nn.LayerNorm",
+        epsilon=1e-6,
+        prenorm=True,
+    ):
+        super().__init__()
+        if isinstance(norm_layer, str):
+            self.norm1 = eval(norm_layer)(dim, epsilon=epsilon)
+        else:
+            self.norm1 = norm_layer(dim)
+        self.mixer = Attention(
+            dim,
+            num_heads=num_heads,
+            qkv_bias=qkv_bias,
+            qk_scale=qk_scale,
+            attn_drop=attn_drop,
+            proj_drop=drop,
+        )
+
+        self.drop_path = DropPath(drop_path) if drop_path > 0.0 else Identity()
+        if isinstance(norm_layer, str):
+            self.norm2 = eval(norm_layer)(dim, epsilon=epsilon)
+        else:
+            self.norm2 = norm_layer(dim)
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp_ratio = mlp_ratio
+        self.mlp = Mlp(
+            in_features=dim,
+            hidden_features=mlp_hidden_dim,
+            act_layer=act_layer,
+            drop=drop,
+        )
+        self.prenorm = prenorm
+
+    def forward(self, x):
+        if self.prenorm:
+            x = self.norm1(x + self.drop_path(self.mixer(x)))
+            x = self.norm2(x + self.drop_path(self.mlp(x)))
+        else:
+            x = x + self.drop_path(self.mixer(self.norm1(x)))
+            x = x + self.drop_path(self.mlp(self.norm2(x)))
+        return x
+
+
+class ViT(nn.Layer):
+    def __init__(
+        self,
+        img_size=[32, 128],
+        patch_size=[4, 4],
+        in_channels=3,
+        embed_dim=384,
+        depth=12,
+        num_heads=6,
+        mlp_ratio=4,
+        qkv_bias=False,
+        qk_scale=None,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.1,
+        norm_layer="nn.LayerNorm",
+        epsilon=1e-6,
+        act="nn.GELU",
+        prenorm=False,
+        **kwargs,
+    ):
+        super().__init__()
+        self.embed_dim = embed_dim
+        self.out_channels = embed_dim
+        self.prenorm = prenorm
+        self.patch_embed = nn.Conv2D(
+            in_channels, embed_dim, patch_size, patch_size, padding=(0, 0)
+        )
+        self.pos_embed = self.create_parameter(
+            shape=[1, 257, embed_dim], default_initializer=zeros_
+        )
+        self.add_parameter("pos_embed", self.pos_embed)
+        self.pos_drop = nn.Dropout(p=drop_rate)
+        dpr = np.linspace(0, drop_path_rate, depth)
+        self.blocks1 = nn.LayerList(
+            [
+                Block(
+                    dim=embed_dim,
+                    num_heads=num_heads,
+                    mlp_ratio=mlp_ratio,
+                    qkv_bias=qkv_bias,
+                    qk_scale=qk_scale,
+                    drop=drop_rate,
+                    act_layer=eval(act),
+                    attn_drop=attn_drop_rate,
+                    drop_path=dpr[i],
+                    norm_layer=norm_layer,
+                    epsilon=epsilon,
+                    prenorm=prenorm,
+                )
+                for i in range(depth)
+            ]
+        )
+        if not prenorm:
+            self.norm = eval(norm_layer)(embed_dim, epsilon=epsilon)
+
+        self.avg_pool = nn.AdaptiveAvgPool2D([1, 25])
+        self.last_conv = nn.Conv2D(
+            in_channels=embed_dim,
+            out_channels=self.out_channels,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            bias_attr=False,
+        )
+        self.hardswish = nn.Hardswish()
+        self.dropout = nn.Dropout(p=0.1, mode="downscale_in_infer")
+
+        trunc_normal_(self.pos_embed)
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight)
+            if isinstance(m, nn.Linear) and m.bias is not None:
+                zeros_(m.bias)
+        elif isinstance(m, nn.LayerNorm):
+            zeros_(m.bias)
+            ones_(m.weight)
+
+    def forward(self, x):
+        x = self.patch_embed(x).flatten(2).transpose((0, 2, 1))
+        x = x + self.pos_embed[:, 1:, :]  # [:, :x.shape[1], :]
+        x = self.pos_drop(x)
+        for blk in self.blocks1:
+            x = blk(x)
+        if not self.prenorm:
+            x = self.norm(x)
+
+        x = self.avg_pool(x.transpose([0, 2, 1]).reshape([0, self.embed_dim, -1, 25]))
+        x = self.last_conv(x)
+        x = self.hardswish(x)
+        x = self.dropout(x)
+        return x
diff --git a/ppocr/modeling/backbones/rec_vit_parseq.py b/ppocr/modeling/backbones/rec_vit_parseq.py
new file mode 100644
index 0000000..049733f
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_vit_parseq.py
@@ -0,0 +1,348 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+This code is refer from:
+https://github.com/PaddlePaddle/PaddleClas/blob/release%2F2.5/ppcls/arch/backbone/model_zoo/vision_transformer.py
+"""
+
+from collections.abc import Callable
+
+import numpy as np
+import paddle
+import paddle.nn as nn
+from paddle.nn.initializer import TruncatedNormal, Constant, Normal
+
+
+trunc_normal_ = TruncatedNormal(std=0.02)
+normal_ = Normal
+zeros_ = Constant(value=0.0)
+ones_ = Constant(value=1.0)
+
+
+def to_2tuple(x):
+    return tuple([x] * 2)
+
+
+def drop_path(x, drop_prob=0.0, training=False):
+    """Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
+    the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper...
+    See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ...
+    """
+    if drop_prob == 0.0 or not training:
+        return x
+    keep_prob = paddle.to_tensor(1 - drop_prob, dtype=x.dtype)
+    shape = (x.shape[0],) + (1,) * (x.ndim - 1)
+    random_tensor = keep_prob + paddle.rand(shape).astype(x.dtype)
+    random_tensor = paddle.floor(random_tensor)  # binarize
+    output = x.divide(keep_prob) * random_tensor
+    return output
+
+
+class DropPath(nn.Layer):
+    """Drop paths (Stochastic Depth) per sample  (when applied in main path of residual blocks)."""
+
+    def __init__(self, drop_prob=None):
+        super(DropPath, self).__init__()
+        self.drop_prob = drop_prob
+
+    def forward(self, x):
+        return drop_path(x, self.drop_prob, self.training)
+
+
+class Identity(nn.Layer):
+    def __init__(self):
+        super(Identity, self).__init__()
+
+    def forward(self, input):
+        return input
+
+
+class Mlp(nn.Layer):
+    def __init__(
+        self,
+        in_features,
+        hidden_features=None,
+        out_features=None,
+        act_layer=nn.GELU,
+        drop=0.0,
+    ):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.fc1 = nn.Linear(in_features, hidden_features)
+        self.act = act_layer()
+        self.fc2 = nn.Linear(hidden_features, out_features)
+        self.drop = nn.Dropout(drop)
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.act(x)
+        x = self.drop(x)
+        x = self.fc2(x)
+        x = self.drop(x)
+        return x
+
+
+class Attention(nn.Layer):
+    def __init__(
+        self,
+        dim,
+        num_heads=8,
+        qkv_bias=False,
+        qk_scale=None,
+        attn_drop=0.0,
+        proj_drop=0.0,
+    ):
+        super().__init__()
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        self.scale = qk_scale or head_dim**-0.5
+
+        self.qkv = nn.Linear(dim, dim * 3, bias_attr=qkv_bias)
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.proj = nn.Linear(dim, dim)
+        self.proj_drop = nn.Dropout(proj_drop)
+
+    def forward(self, x):
+        # B= x.shape[0]
+        N, C = x.shape[1:]
+        qkv = (
+            self.qkv(x)
+            .reshape((-1, N, 3, self.num_heads, C // self.num_heads))
+            .transpose((2, 0, 3, 1, 4))
+        )
+        q, k, v = qkv[0], qkv[1], qkv[2]
+
+        attn = (q.matmul(k.transpose((0, 1, 3, 2)))) * self.scale
+        attn = nn.functional.softmax(attn, axis=-1)
+        attn = self.attn_drop(attn)
+
+        x = (attn.matmul(v)).transpose((0, 2, 1, 3)).reshape((-1, N, C))
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+
+
+class Block(nn.Layer):
+    def __init__(
+        self,
+        dim,
+        num_heads,
+        mlp_ratio=4.0,
+        qkv_bias=False,
+        qk_scale=None,
+        drop=0.0,
+        attn_drop=0.0,
+        drop_path=0.0,
+        act_layer=nn.GELU,
+        norm_layer="nn.LayerNorm",
+        epsilon=1e-5,
+    ):
+        super().__init__()
+        if isinstance(norm_layer, str):
+            self.norm1 = eval(norm_layer)(dim, epsilon=epsilon)
+        elif isinstance(norm_layer, Callable):
+            self.norm1 = norm_layer(dim)
+        else:
+            raise TypeError("The norm_layer must be str or paddle.nn.layer.Layer class")
+        self.attn = Attention(
+            dim,
+            num_heads=num_heads,
+            qkv_bias=qkv_bias,
+            qk_scale=qk_scale,
+            attn_drop=attn_drop,
+            proj_drop=drop,
+        )
+        # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
+        self.drop_path = DropPath(drop_path) if drop_path > 0.0 else Identity()
+        if isinstance(norm_layer, str):
+            self.norm2 = eval(norm_layer)(dim, epsilon=epsilon)
+        elif isinstance(norm_layer, Callable):
+            self.norm2 = norm_layer(dim)
+        else:
+            raise TypeError("The norm_layer must be str or paddle.nn.layer.Layer class")
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp = Mlp(
+            in_features=dim,
+            hidden_features=mlp_hidden_dim,
+            act_layer=act_layer,
+            drop=drop,
+        )
+
+    def forward(self, x):
+        x = x + self.drop_path(self.attn(self.norm1(x)))
+        x = x + self.drop_path(self.mlp(self.norm2(x)))
+        return x
+
+
+class PatchEmbed(nn.Layer):
+    """Image to Patch Embedding"""
+
+    def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768):
+        super().__init__()
+        if isinstance(img_size, int):
+            img_size = to_2tuple(img_size)
+        if isinstance(patch_size, int):
+            patch_size = to_2tuple(patch_size)
+        num_patches = (img_size[1] // patch_size[1]) * (img_size[0] // patch_size[0])
+        self.img_size = img_size
+        self.patch_size = patch_size
+        self.num_patches = num_patches
+
+        self.proj = nn.Conv2D(
+            in_chans, embed_dim, kernel_size=patch_size, stride=patch_size
+        )
+
+    def forward(self, x):
+        B, C, H, W = x.shape
+        assert (
+            H == self.img_size[0] and W == self.img_size[1]
+        ), f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})."
+
+        x = self.proj(x).flatten(2).transpose((0, 2, 1))
+        return x
+
+
+class VisionTransformer(nn.Layer):
+    """Vision Transformer with support for patch input"""
+
+    def __init__(
+        self,
+        img_size=224,
+        patch_size=16,
+        in_channels=3,
+        class_num=1000,
+        embed_dim=768,
+        depth=12,
+        num_heads=12,
+        mlp_ratio=4,
+        qkv_bias=False,
+        qk_scale=None,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.0,
+        norm_layer="nn.LayerNorm",
+        epsilon=1e-5,
+        **kwargs,
+    ):
+        super().__init__()
+        self.class_num = class_num
+
+        self.num_features = self.embed_dim = embed_dim
+
+        self.patch_embed = PatchEmbed(
+            img_size=img_size,
+            patch_size=patch_size,
+            in_chans=in_channels,
+            embed_dim=embed_dim,
+        )
+        num_patches = self.patch_embed.num_patches
+
+        self.pos_embed = self.create_parameter(
+            shape=(1, num_patches, embed_dim), default_initializer=zeros_
+        )
+        self.add_parameter("pos_embed", self.pos_embed)
+        self.cls_token = self.create_parameter(
+            shape=(1, 1, embed_dim), default_initializer=zeros_
+        )
+        self.add_parameter("cls_token", self.cls_token)
+        self.pos_drop = nn.Dropout(p=drop_rate)
+
+        dpr = np.linspace(0, drop_path_rate, depth)
+
+        self.blocks = nn.LayerList(
+            [
+                Block(
+                    dim=embed_dim,
+                    num_heads=num_heads,
+                    mlp_ratio=mlp_ratio,
+                    qkv_bias=qkv_bias,
+                    qk_scale=qk_scale,
+                    drop=drop_rate,
+                    attn_drop=attn_drop_rate,
+                    drop_path=dpr[i],
+                    norm_layer=norm_layer,
+                    epsilon=epsilon,
+                )
+                for i in range(depth)
+            ]
+        )
+
+        self.norm = eval(norm_layer)(embed_dim, epsilon=epsilon)
+
+        # Classifier head
+        self.head = nn.Linear(embed_dim, class_num) if class_num > 0 else Identity()
+
+        trunc_normal_(self.pos_embed)
+        self.out_channels = embed_dim
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight)
+            if isinstance(m, nn.Linear) and m.bias is not None:
+                zeros_(m.bias)
+        elif isinstance(m, nn.LayerNorm):
+            zeros_(m.bias)
+            ones_(m.weight)
+
+    def forward_features(self, x):
+        B = x.shape[0]
+        x = self.patch_embed(x)
+        x = x + self.pos_embed
+        x = self.pos_drop(x)
+        for blk in self.blocks:
+            x = blk(x)
+        x = self.norm(x)
+        return x
+
+    def forward(self, x):
+        x = self.forward_features(x)
+        x = self.head(x)
+        return x
+
+
+class ViTParseQ(VisionTransformer):
+    def __init__(
+        self,
+        img_size=[224, 224],
+        patch_size=[16, 16],
+        in_channels=3,
+        embed_dim=768,
+        depth=12,
+        num_heads=12,
+        mlp_ratio=4.0,
+        qkv_bias=True,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.0,
+    ):
+        super().__init__(
+            img_size,
+            patch_size,
+            in_channels,
+            embed_dim=embed_dim,
+            depth=depth,
+            num_heads=num_heads,
+            mlp_ratio=mlp_ratio,
+            qkv_bias=qkv_bias,
+            drop_rate=drop_rate,
+            attn_drop_rate=attn_drop_rate,
+            drop_path_rate=drop_path_rate,
+            class_num=0,
+        )
+
+    def forward(self, x):
+        return self.forward_features(x)
diff --git a/ppocr/modeling/backbones/rec_vitstr.py b/ppocr/modeling/backbones/rec_vitstr.py
new file mode 100644
index 0000000..9a273da
--- /dev/null
+++ b/ppocr/modeling/backbones/rec_vitstr.py
@@ -0,0 +1,133 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/roatienza/deep-text-recognition-benchmark/blob/master/modules/vitstr.py
+"""
+
+import numpy as np
+import paddle
+import paddle.nn as nn
+from ppocr.modeling.backbones.rec_svtrnet import (
+    Block,
+    PatchEmbed,
+    zeros_,
+    trunc_normal_,
+    ones_,
+)
+
+scale_dim_heads = {"tiny": [192, 3], "small": [384, 6], "base": [768, 12]}
+
+
+class ViTSTR(nn.Layer):
+    def __init__(
+        self,
+        img_size=[224, 224],
+        in_channels=1,
+        scale="tiny",
+        seqlen=27,
+        patch_size=[16, 16],
+        embed_dim=None,
+        depth=12,
+        num_heads=None,
+        mlp_ratio=4,
+        qkv_bias=True,
+        qk_scale=None,
+        drop_path_rate=0.0,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        norm_layer="nn.LayerNorm",
+        act_layer="nn.GELU",
+        epsilon=1e-6,
+        out_channels=None,
+        **kwargs,
+    ):
+        super().__init__()
+        self.seqlen = seqlen
+        embed_dim = embed_dim if embed_dim is not None else scale_dim_heads[scale][0]
+        num_heads = num_heads if num_heads is not None else scale_dim_heads[scale][1]
+        out_channels = out_channels if out_channels is not None else embed_dim
+        self.patch_embed = PatchEmbed(
+            img_size=img_size,
+            in_channels=in_channels,
+            embed_dim=embed_dim,
+            patch_size=patch_size,
+            mode="linear",
+        )
+        num_patches = self.patch_embed.num_patches
+
+        self.pos_embed = self.create_parameter(
+            shape=[1, num_patches + 1, embed_dim], default_initializer=zeros_
+        )
+        self.add_parameter("pos_embed", self.pos_embed)
+        self.cls_token = self.create_parameter(
+            shape=[1, 1, embed_dim], default_initializer=zeros_
+        )
+        self.add_parameter("cls_token", self.cls_token)
+
+        self.pos_drop = nn.Dropout(p=drop_rate)
+
+        dpr = np.linspace(0, drop_path_rate, depth)
+        self.blocks = nn.LayerList(
+            [
+                Block(
+                    dim=embed_dim,
+                    num_heads=num_heads,
+                    mlp_ratio=mlp_ratio,
+                    qkv_bias=qkv_bias,
+                    qk_scale=qk_scale,
+                    drop=drop_rate,
+                    attn_drop=attn_drop_rate,
+                    drop_path=dpr[i],
+                    norm_layer=norm_layer,
+                    act_layer=eval(act_layer),
+                    epsilon=epsilon,
+                    prenorm=False,
+                )
+                for i in range(depth)
+            ]
+        )
+        self.norm = eval(norm_layer)(embed_dim, epsilon=epsilon)
+
+        self.out_channels = out_channels
+
+        trunc_normal_(self.pos_embed)
+        trunc_normal_(self.cls_token)
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight)
+            if isinstance(m, nn.Linear) and m.bias is not None:
+                zeros_(m.bias)
+        elif isinstance(m, nn.LayerNorm):
+            zeros_(m.bias)
+            ones_(m.weight)
+
+    def forward_features(self, x):
+        B = x.shape[0]
+        x = self.patch_embed(x)
+        cls_tokens = paddle.tile(self.cls_token, repeat_times=[B, 1, 1])
+        x = paddle.concat((cls_tokens, x), axis=1)
+        x = x + self.pos_embed
+        x = self.pos_drop(x)
+        for blk in self.blocks:
+            x = blk(x)
+        x = self.norm(x)
+        return x
+
+    def forward(self, x):
+        x = self.forward_features(x)
+        x = x[:, : self.seqlen]
+        return x.transpose([0, 2, 1]).unsqueeze(2)
diff --git a/ppocr/modeling/backbones/table_master_resnet.py b/ppocr/modeling/backbones/table_master_resnet.py
new file mode 100644
index 0000000..6880ae3
--- /dev/null
+++ b/ppocr/modeling/backbones/table_master_resnet.py
@@ -0,0 +1,365 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/JiaquanYe/TableMASTER-mmocr/blob/master/mmocr/models/textrecog/backbones/table_resnet_extra.py
+"""
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+
+class BasicBlock(nn.Layer):
+    expansion = 1
+
+    def __init__(self, inplanes, planes, stride=1, downsample=None, gcb_config=None):
+        super(BasicBlock, self).__init__()
+        self.conv1 = nn.Conv2D(
+            inplanes, planes, kernel_size=3, stride=stride, padding=1, bias_attr=False
+        )
+        self.bn1 = nn.BatchNorm2D(planes, momentum=0.9)
+        self.relu = nn.ReLU()
+        self.conv2 = nn.Conv2D(
+            planes, planes, kernel_size=3, stride=1, padding=1, bias_attr=False
+        )
+        self.bn2 = nn.BatchNorm2D(planes, momentum=0.9)
+        self.downsample = downsample
+        self.stride = stride
+        self.gcb_config = gcb_config
+
+        if self.gcb_config is not None:
+            gcb_ratio = gcb_config["ratio"]
+            gcb_headers = gcb_config["headers"]
+            att_scale = gcb_config["att_scale"]
+            fusion_type = gcb_config["fusion_type"]
+            self.context_block = MultiAspectGCAttention(
+                inplanes=planes,
+                ratio=gcb_ratio,
+                headers=gcb_headers,
+                att_scale=att_scale,
+                fusion_type=fusion_type,
+            )
+
+    def forward(self, x):
+        residual = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+
+        if self.gcb_config is not None:
+            out = self.context_block(out)
+
+        if self.downsample is not None:
+            residual = self.downsample(x)
+
+        out += residual
+        out = self.relu(out)
+
+        return out
+
+
+def get_gcb_config(gcb_config, layer):
+    if gcb_config is None or not gcb_config["layers"][layer]:
+        return None
+    else:
+        return gcb_config
+
+
+class TableResNetExtra(nn.Layer):
+    def __init__(self, layers, in_channels=3, gcb_config=None):
+        assert len(layers) >= 4
+
+        super(TableResNetExtra, self).__init__()
+        self.inplanes = 128
+        self.conv1 = nn.Conv2D(
+            in_channels, 64, kernel_size=3, stride=1, padding=1, bias_attr=False
+        )
+        self.bn1 = nn.BatchNorm2D(64)
+        self.relu1 = nn.ReLU()
+
+        self.conv2 = nn.Conv2D(
+            64, 128, kernel_size=3, stride=1, padding=1, bias_attr=False
+        )
+        self.bn2 = nn.BatchNorm2D(128)
+        self.relu2 = nn.ReLU()
+
+        self.maxpool1 = nn.MaxPool2D(kernel_size=2, stride=2)
+
+        self.layer1 = self._make_layer(
+            BasicBlock,
+            256,
+            layers[0],
+            stride=1,
+            gcb_config=get_gcb_config(gcb_config, 0),
+        )
+
+        self.conv3 = nn.Conv2D(
+            256, 256, kernel_size=3, stride=1, padding=1, bias_attr=False
+        )
+        self.bn3 = nn.BatchNorm2D(256)
+        self.relu3 = nn.ReLU()
+
+        self.maxpool2 = nn.MaxPool2D(kernel_size=2, stride=2)
+
+        self.layer2 = self._make_layer(
+            BasicBlock,
+            256,
+            layers[1],
+            stride=1,
+            gcb_config=get_gcb_config(gcb_config, 1),
+        )
+
+        self.conv4 = nn.Conv2D(
+            256, 256, kernel_size=3, stride=1, padding=1, bias_attr=False
+        )
+        self.bn4 = nn.BatchNorm2D(256)
+        self.relu4 = nn.ReLU()
+
+        self.maxpool3 = nn.MaxPool2D(kernel_size=2, stride=2)
+
+        self.layer3 = self._make_layer(
+            BasicBlock,
+            512,
+            layers[2],
+            stride=1,
+            gcb_config=get_gcb_config(gcb_config, 2),
+        )
+
+        self.conv5 = nn.Conv2D(
+            512, 512, kernel_size=3, stride=1, padding=1, bias_attr=False
+        )
+        self.bn5 = nn.BatchNorm2D(512)
+        self.relu5 = nn.ReLU()
+
+        self.layer4 = self._make_layer(
+            BasicBlock,
+            512,
+            layers[3],
+            stride=1,
+            gcb_config=get_gcb_config(gcb_config, 3),
+        )
+
+        self.conv6 = nn.Conv2D(
+            512, 512, kernel_size=3, stride=1, padding=1, bias_attr=False
+        )
+        self.bn6 = nn.BatchNorm2D(512)
+        self.relu6 = nn.ReLU()
+
+        self.out_channels = [256, 256, 512]
+
+    def _make_layer(self, block, planes, blocks, stride=1, gcb_config=None):
+        downsample = None
+        if stride != 1 or self.inplanes != planes * block.expansion:
+            downsample = nn.Sequential(
+                nn.Conv2D(
+                    self.inplanes,
+                    planes * block.expansion,
+                    kernel_size=1,
+                    stride=stride,
+                    bias_attr=False,
+                ),
+                nn.BatchNorm2D(planes * block.expansion),
+            )
+
+        layers = []
+        layers.append(
+            block(self.inplanes, planes, stride, downsample, gcb_config=gcb_config)
+        )
+        self.inplanes = planes * block.expansion
+        for _ in range(1, blocks):
+            layers.append(block(self.inplanes, planes))
+
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        f = []
+        x = self.conv1(x)
+
+        x = self.bn1(x)
+        x = self.relu1(x)
+
+        x = self.conv2(x)
+        x = self.bn2(x)
+        x = self.relu2(x)
+
+        x = self.maxpool1(x)
+        x = self.layer1(x)
+
+        x = self.conv3(x)
+        x = self.bn3(x)
+        x = self.relu3(x)
+        f.append(x)
+
+        x = self.maxpool2(x)
+        x = self.layer2(x)
+
+        x = self.conv4(x)
+        x = self.bn4(x)
+        x = self.relu4(x)
+        f.append(x)
+
+        x = self.maxpool3(x)
+
+        x = self.layer3(x)
+        x = self.conv5(x)
+        x = self.bn5(x)
+        x = self.relu5(x)
+
+        x = self.layer4(x)
+        x = self.conv6(x)
+        x = self.bn6(x)
+        x = self.relu6(x)
+        f.append(x)
+        return f
+
+
+class MultiAspectGCAttention(nn.Layer):
+    def __init__(
+        self,
+        inplanes,
+        ratio,
+        headers,
+        pooling_type="att",
+        att_scale=False,
+        fusion_type="channel_add",
+    ):
+        super(MultiAspectGCAttention, self).__init__()
+        assert pooling_type in ["avg", "att"]
+
+        assert fusion_type in ["channel_add", "channel_mul", "channel_concat"]
+        assert (
+            inplanes % headers == 0 and inplanes >= 8
+        )  # inplanes must be divided by headers evenly
+
+        self.headers = headers
+        self.inplanes = inplanes
+        self.ratio = ratio
+        self.planes = int(inplanes * ratio)
+        self.pooling_type = pooling_type
+        self.fusion_type = fusion_type
+        self.att_scale = False
+
+        self.single_header_inplanes = int(inplanes / headers)
+
+        if pooling_type == "att":
+            self.conv_mask = nn.Conv2D(self.single_header_inplanes, 1, kernel_size=1)
+            self.softmax = nn.Softmax(axis=2)
+        else:
+            self.avg_pool = nn.AdaptiveAvgPool2D(1)
+
+        if fusion_type == "channel_add":
+            self.channel_add_conv = nn.Sequential(
+                nn.Conv2D(self.inplanes, self.planes, kernel_size=1),
+                nn.LayerNorm([self.planes, 1, 1]),
+                nn.ReLU(),
+                nn.Conv2D(self.planes, self.inplanes, kernel_size=1),
+            )
+        elif fusion_type == "channel_concat":
+            self.channel_concat_conv = nn.Sequential(
+                nn.Conv2D(self.inplanes, self.planes, kernel_size=1),
+                nn.LayerNorm([self.planes, 1, 1]),
+                nn.ReLU(),
+                nn.Conv2D(self.planes, self.inplanes, kernel_size=1),
+            )
+            # for concat
+            self.cat_conv = nn.Conv2D(2 * self.inplanes, self.inplanes, kernel_size=1)
+        elif fusion_type == "channel_mul":
+            self.channel_mul_conv = nn.Sequential(
+                nn.Conv2D(self.inplanes, self.planes, kernel_size=1),
+                nn.LayerNorm([self.planes, 1, 1]),
+                nn.ReLU(),
+                nn.Conv2D(self.planes, self.inplanes, kernel_size=1),
+            )
+
+    def spatial_pool(self, x):
+        batch, channel, height, width = x.shape
+        if self.pooling_type == "att":
+            # [N*headers, C', H , W] C = headers * C'
+            x = x.reshape(
+                [batch * self.headers, self.single_header_inplanes, height, width]
+            )
+            input_x = x
+
+            # [N*headers, C', H * W] C = headers * C'
+            # input_x = input_x.view(batch, channel, height * width)
+            input_x = input_x.reshape(
+                [batch * self.headers, self.single_header_inplanes, height * width]
+            )
+
+            # [N*headers, 1, C', H * W]
+            input_x = input_x.unsqueeze(1)
+            # [N*headers, 1, H, W]
+            context_mask = self.conv_mask(x)
+            # [N*headers, 1, H * W]
+            context_mask = context_mask.reshape(
+                [batch * self.headers, 1, height * width]
+            )
+
+            # scale variance
+            if self.att_scale and self.headers > 1:
+                context_mask = context_mask / paddle.sqrt(self.single_header_inplanes)
+
+            # [N*headers, 1, H * W]
+            context_mask = self.softmax(context_mask)
+
+            # [N*headers, 1, H * W, 1]
+            context_mask = context_mask.unsqueeze(-1)
+            # [N*headers, 1, C', 1] = [N*headers, 1, C', H * W] * [N*headers, 1, H * W, 1]
+            context = paddle.matmul(input_x, context_mask)
+
+            # [N, headers * C', 1, 1]
+            context = context.reshape(
+                [batch, self.headers * self.single_header_inplanes, 1, 1]
+            )
+        else:
+            # [N, C, 1, 1]
+            context = self.avg_pool(x)
+
+        return context
+
+    def forward(self, x):
+        # [N, C, 1, 1]
+        context = self.spatial_pool(x)
+
+        out = x
+
+        if self.fusion_type == "channel_mul":
+            # [N, C, 1, 1]
+            channel_mul_term = F.sigmoid(self.channel_mul_conv(context))
+            out = out * channel_mul_term
+        elif self.fusion_type == "channel_add":
+            # [N, C, 1, 1]
+            channel_add_term = self.channel_add_conv(context)
+            out = out + channel_add_term
+        else:
+            # [N, C, 1, 1]
+            channel_concat_term = self.channel_concat_conv(context)
+
+            # use concat
+            _, C1, _, _ = channel_concat_term.shape
+            N, C2, H, W = out.shape
+
+            out = paddle.concat(
+                [out, channel_concat_term.expand([-1, -1, H, W])], axis=1
+            )
+            out = self.cat_conv(out)
+            out = F.layer_norm(out, [self.inplanes, H, W])
+            out = F.relu(out)
+
+        return out
diff --git a/ppocr/modeling/backbones/vqa_layoutlm.py b/ppocr/modeling/backbones/vqa_layoutlm.py
new file mode 100644
index 0000000..c80cb14
--- /dev/null
+++ b/ppocr/modeling/backbones/vqa_layoutlm.py
@@ -0,0 +1,260 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+from paddle import nn
+
+from paddlenlp.transformers import (
+    LayoutXLMModel,
+    LayoutXLMForTokenClassification,
+    LayoutXLMForRelationExtraction,
+)
+from paddlenlp.transformers import LayoutLMModel, LayoutLMForTokenClassification
+from paddlenlp.transformers import (
+    LayoutLMv2Model,
+    LayoutLMv2ForTokenClassification,
+    LayoutLMv2ForRelationExtraction,
+)
+from paddlenlp.transformers import AutoModel
+
+__all__ = ["LayoutXLMForSer", "LayoutLMForSer"]
+
+pretrained_model_dict = {
+    LayoutXLMModel: {
+        "base": "layoutxlm-base-uncased",
+        "vi": "vi-layoutxlm-base-uncased",
+    },
+    LayoutLMModel: {
+        "base": "layoutlm-base-uncased",
+    },
+    LayoutLMv2Model: {
+        "base": "layoutlmv2-base-uncased",
+        "vi": "vi-layoutlmv2-base-uncased",
+    },
+}
+
+
+class NLPBaseModel(nn.Layer):
+    def __init__(
+        self,
+        base_model_class,
+        model_class,
+        mode="base",
+        type="ser",
+        pretrained=True,
+        checkpoints=None,
+        **kwargs,
+    ):
+        super(NLPBaseModel, self).__init__()
+        if checkpoints is not None:  # load the trained model
+            self.model = model_class.from_pretrained(checkpoints)
+        else:  # load the pretrained-model
+            pretrained_model_name = pretrained_model_dict[base_model_class][mode]
+            if type == "ser":
+                self.model = model_class.from_pretrained(
+                    pretrained_model_name, num_classes=kwargs["num_classes"], dropout=0
+                )
+            else:
+                self.model = model_class.from_pretrained(
+                    pretrained_model_name, dropout=0
+                )
+        self.out_channels = 1
+        self.use_visual_backbone = True
+
+
+class LayoutLMForSer(NLPBaseModel):
+    def __init__(
+        self, num_classes, pretrained=True, checkpoints=None, mode="base", **kwargs
+    ):
+        super(LayoutLMForSer, self).__init__(
+            LayoutLMModel,
+            LayoutLMForTokenClassification,
+            mode,
+            "ser",
+            pretrained,
+            checkpoints,
+            num_classes=num_classes,
+        )
+        self.use_visual_backbone = False
+
+    def forward(self, x):
+        x = self.model(
+            input_ids=x[0],
+            bbox=x[1],
+            attention_mask=x[2],
+            token_type_ids=x[3],
+            position_ids=None,
+            output_hidden_states=False,
+        )
+        return x
+
+
+class LayoutLMv2ForSer(NLPBaseModel):
+    def __init__(
+        self, num_classes, pretrained=True, checkpoints=None, mode="base", **kwargs
+    ):
+        super(LayoutLMv2ForSer, self).__init__(
+            LayoutLMv2Model,
+            LayoutLMv2ForTokenClassification,
+            mode,
+            "ser",
+            pretrained,
+            checkpoints,
+            num_classes=num_classes,
+        )
+        if (
+            hasattr(self.model.layoutlmv2, "use_visual_backbone")
+            and self.model.layoutlmv2.use_visual_backbone is False
+        ):
+            self.use_visual_backbone = False
+
+    def forward(self, x):
+        if self.use_visual_backbone is True:
+            image = x[4]
+        else:
+            image = None
+        x = self.model(
+            input_ids=x[0],
+            bbox=x[1],
+            attention_mask=x[2],
+            token_type_ids=x[3],
+            image=image,
+            position_ids=None,
+            head_mask=None,
+            labels=None,
+        )
+        if self.training:
+            res = {"backbone_out": x[0]}
+            res.update(x[1])
+            return res
+        else:
+            return x
+
+
+class LayoutXLMForSer(NLPBaseModel):
+    def __init__(
+        self, num_classes, pretrained=True, checkpoints=None, mode="base", **kwargs
+    ):
+        super(LayoutXLMForSer, self).__init__(
+            LayoutXLMModel,
+            LayoutXLMForTokenClassification,
+            mode,
+            "ser",
+            pretrained,
+            checkpoints,
+            num_classes=num_classes,
+        )
+        if (
+            hasattr(self.model.layoutxlm, "use_visual_backbone")
+            and self.model.layoutxlm.use_visual_backbone is False
+        ):
+            self.use_visual_backbone = False
+
+    def forward(self, x):
+        if self.use_visual_backbone is True:
+            image = x[4]
+        else:
+            image = None
+        x = self.model(
+            input_ids=x[0],
+            bbox=x[1],
+            attention_mask=x[2],
+            token_type_ids=x[3],
+            image=image,
+            position_ids=None,
+            head_mask=None,
+            labels=None,
+        )
+        if self.training:
+            res = {"backbone_out": x[0]}
+            res.update(x[1])
+            return res
+        else:
+            return x
+
+
+class LayoutLMv2ForRe(NLPBaseModel):
+    def __init__(self, pretrained=True, checkpoints=None, mode="base", **kwargs):
+        super(LayoutLMv2ForRe, self).__init__(
+            LayoutLMv2Model,
+            LayoutLMv2ForRelationExtraction,
+            mode,
+            "re",
+            pretrained,
+            checkpoints,
+        )
+        if (
+            hasattr(self.model.layoutlmv2, "use_visual_backbone")
+            and self.model.layoutlmv2.use_visual_backbone is False
+        ):
+            self.use_visual_backbone = False
+
+    def forward(self, x):
+        x = self.model(
+            input_ids=x[0],
+            bbox=x[1],
+            attention_mask=x[2],
+            token_type_ids=x[3],
+            image=x[4],
+            position_ids=None,
+            head_mask=None,
+            labels=None,
+            entities=x[5],
+            relations=x[6],
+        )
+        return x
+
+
+class LayoutXLMForRe(NLPBaseModel):
+    def __init__(self, pretrained=True, checkpoints=None, mode="base", **kwargs):
+        super(LayoutXLMForRe, self).__init__(
+            LayoutXLMModel,
+            LayoutXLMForRelationExtraction,
+            mode,
+            "re",
+            pretrained,
+            checkpoints,
+        )
+        if (
+            hasattr(self.model.layoutxlm, "use_visual_backbone")
+            and self.model.layoutxlm.use_visual_backbone is False
+        ):
+            self.use_visual_backbone = False
+
+    def forward(self, x):
+        if self.use_visual_backbone is True:
+            image = x[4]
+            entities = x[5]
+            relations = x[6]
+        else:
+            image = None
+            entities = x[4]
+            relations = x[5]
+        x = self.model(
+            input_ids=x[0],
+            bbox=x[1],
+            attention_mask=x[2],
+            token_type_ids=x[3],
+            image=image,
+            position_ids=None,
+            head_mask=None,
+            labels=None,
+            entities=entities,
+            relations=relations,
+        )
+        return x
diff --git a/ppocr/modeling/heads/__init__.py b/ppocr/modeling/heads/__init__.py
new file mode 100755
index 0000000..c410ccc
--- /dev/null
+++ b/ppocr/modeling/heads/__init__.py
@@ -0,0 +1,108 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+__all__ = ["build_head"]
+
+
+def build_head(config):
+    # det head
+    from .det_db_head import DBHead, PFHeadLocal
+    from .det_east_head import EASTHead
+    from .det_sast_head import SASTHead
+    from .det_pse_head import PSEHead
+    from .det_fce_head import FCEHead
+    from .e2e_pg_head import PGHead
+    from .det_ct_head import CT_Head
+
+    # rec head
+    from .rec_ctc_head import CTCHead
+    from .rec_att_head import AttentionHead
+    from .rec_srn_head import SRNHead
+    from .rec_nrtr_head import Transformer
+    from .rec_sar_head import SARHead
+    from .rec_aster_head import AsterHead
+    from .rec_pren_head import PRENHead
+    from .rec_multi_head import MultiHead
+    from .rec_spin_att_head import SPINAttentionHead
+    from .rec_abinet_head import ABINetHead
+    from .rec_robustscanner_head import RobustScannerHead
+    from .rec_visionlan_head import VLHead
+    from .rec_rfl_head import RFLHead
+    from .rec_can_head import CANHead
+    from .rec_latexocr_head import LaTeXOCRHead
+    from .rec_satrn_head import SATRNHead
+    from .rec_parseq_head import ParseQHead
+    from .rec_cppd_head import CPPDHead
+    from .rec_unimernet_head import UniMERNetHead
+    from .rec_ppformulanet_head import PPFormulaNet_Head
+
+    # cls head
+    from .cls_head import ClsHead
+
+    # kie head
+    from .kie_sdmgr_head import SDMGRHead
+
+    from .table_att_head import TableAttentionHead, SLAHead
+    from .table_master_head import TableMasterHead
+
+    support_dict = [
+        "DBHead",
+        "PSEHead",
+        "FCEHead",
+        "EASTHead",
+        "SASTHead",
+        "CTCHead",
+        "ClsHead",
+        "AttentionHead",
+        "SRNHead",
+        "PGHead",
+        "Transformer",
+        "TableAttentionHead",
+        "SARHead",
+        "AsterHead",
+        "SDMGRHead",
+        "PRENHead",
+        "MultiHead",
+        "ABINetHead",
+        "TableMasterHead",
+        "SPINAttentionHead",
+        "VLHead",
+        "SLAHead",
+        "RobustScannerHead",
+        "CT_Head",
+        "RFLHead",
+        "DRRGHead",
+        "CANHead",
+        "LaTeXOCRHead",
+        "SATRNHead",
+        "PFHeadLocal",
+        "ParseQHead",
+        "CPPDHead",
+        "UniMERNetHead",
+        "PPFormulaNet_Head",
+    ]
+
+    if config["name"] == "DRRGHead":
+        from .det_drrg_head import DRRGHead
+
+        support_dict.append("DRRGHead")
+
+    # table head
+
+    module_name = config.pop("name")
+    assert module_name in support_dict, Exception(
+        "head only support {}".format(support_dict)
+    )
+    module_class = eval(module_name)(**config)
+    return module_class
diff --git a/ppocr/modeling/heads/cls_head.py b/ppocr/modeling/heads/cls_head.py
new file mode 100644
index 0000000..867e960
--- /dev/null
+++ b/ppocr/modeling/heads/cls_head.py
@@ -0,0 +1,53 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+from paddle import nn, ParamAttr
+import paddle.nn.functional as F
+
+
+class ClsHead(nn.Layer):
+    """
+    Class orientation
+
+    Args:
+
+        params(dict): super parameters for build Class network
+    """
+
+    def __init__(self, in_channels, class_dim, **kwargs):
+        super(ClsHead, self).__init__()
+        self.pool = nn.AdaptiveAvgPool2D(1)
+        stdv = 1.0 / math.sqrt(in_channels * 1.0)
+        self.fc = nn.Linear(
+            in_channels,
+            class_dim,
+            weight_attr=ParamAttr(
+                name="fc_0.w_0", initializer=nn.initializer.Uniform(-stdv, stdv)
+            ),
+            bias_attr=ParamAttr(name="fc_0.b_0"),
+        )
+
+    def forward(self, x, targets=None):
+        x = self.pool(x)
+        x = paddle.reshape(x, shape=[x.shape[0], x.shape[1]])
+        x = self.fc(x)
+        if not self.training:
+            x = F.softmax(x, axis=1)
+        return x
diff --git a/ppocr/modeling/heads/det_ct_head.py b/ppocr/modeling/heads/det_ct_head.py
new file mode 100644
index 0000000..cd050fc
--- /dev/null
+++ b/ppocr/modeling/heads/det_ct_head.py
@@ -0,0 +1,69 @@
+# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+
+import math
+from paddle.nn.initializer import TruncatedNormal, Constant, Normal
+
+ones_ = Constant(value=1.0)
+zeros_ = Constant(value=0.0)
+
+
+class CT_Head(nn.Layer):
+    def __init__(
+        self, in_channels, hidden_dim, num_classes, loss_kernel=None, loss_loc=None
+    ):
+        super(CT_Head, self).__init__()
+        self.conv1 = nn.Conv2D(
+            in_channels, hidden_dim, kernel_size=3, stride=1, padding=1
+        )
+        self.bn1 = nn.BatchNorm2D(hidden_dim)
+        self.relu1 = nn.ReLU()
+
+        self.conv2 = nn.Conv2D(
+            hidden_dim, num_classes, kernel_size=1, stride=1, padding=0
+        )
+
+        for m in self.sublayers():
+            if isinstance(m, nn.Conv2D):
+                n = m._kernel_size[0] * m._kernel_size[1] * m._out_channels
+                normal_ = Normal(mean=0.0, std=math.sqrt(2.0 / n))
+                normal_(m.weight)
+            elif isinstance(m, nn.BatchNorm2D):
+                zeros_(m.bias)
+                ones_(m.weight)
+
+    def _upsample(self, x, scale=1):
+        return F.upsample(x, scale_factor=scale, mode="bilinear")
+
+    def forward(self, f, targets=None):
+        out = self.conv1(f)
+        out = self.relu1(self.bn1(out))
+        out = self.conv2(out)
+
+        if self.training:
+            out = self._upsample(out, scale=4)
+            return {"maps": out}
+        else:
+            score = F.sigmoid(out[:, 0, :, :])
+            return {"maps": out, "score": score}
diff --git a/ppocr/modeling/heads/det_db_head.py b/ppocr/modeling/heads/det_db_head.py
new file mode 100644
index 0000000..51aa256
--- /dev/null
+++ b/ppocr/modeling/heads/det_db_head.py
@@ -0,0 +1,283 @@
+# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+from ppocr.modeling.backbones.det_mobilenet_v3 import ConvBNLayer
+
+
+def get_bias_attr(k):
+    stdv = 1.0 / math.sqrt(k * 1.0)
+    initializer = paddle.nn.initializer.Uniform(-stdv, stdv)
+    bias_attr = ParamAttr(initializer=initializer)
+    return bias_attr
+
+
+class Head(nn.Layer):
+    def __init__(self, in_channels, kernel_list=[3, 2, 2], fix_nan=False, **kwargs):
+        super(Head, self).__init__()
+
+        self.conv1 = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=in_channels // 4,
+            kernel_size=kernel_list[0],
+            padding=int(kernel_list[0] // 2),
+            weight_attr=ParamAttr(),
+            bias_attr=False,
+        )
+        self.conv_bn1 = nn.BatchNorm(
+            num_channels=in_channels // 4,
+            param_attr=ParamAttr(initializer=paddle.nn.initializer.Constant(value=1.0)),
+            bias_attr=ParamAttr(initializer=paddle.nn.initializer.Constant(value=1e-4)),
+            act="relu",
+        )
+
+        self.conv2 = nn.Conv2DTranspose(
+            in_channels=in_channels // 4,
+            out_channels=in_channels // 4,
+            kernel_size=kernel_list[1],
+            stride=2,
+            weight_attr=ParamAttr(initializer=paddle.nn.initializer.KaimingUniform()),
+            bias_attr=get_bias_attr(in_channels // 4),
+        )
+        self.conv_bn2 = nn.BatchNorm(
+            num_channels=in_channels // 4,
+            param_attr=ParamAttr(initializer=paddle.nn.initializer.Constant(value=1.0)),
+            bias_attr=ParamAttr(initializer=paddle.nn.initializer.Constant(value=1e-4)),
+            act="relu",
+        )
+        self.conv3 = nn.Conv2DTranspose(
+            in_channels=in_channels // 4,
+            out_channels=1,
+            kernel_size=kernel_list[2],
+            stride=2,
+            weight_attr=ParamAttr(initializer=paddle.nn.initializer.KaimingUniform()),
+            bias_attr=get_bias_attr(in_channels // 4),
+        )
+
+        self.fix_nan = fix_nan
+        self.is_repped = False
+
+    def forward(self, x, return_f=False):
+        x = self.conv1(x)
+        x = self.conv_bn1(x)
+        if self.fix_nan and self.training:
+            x = paddle.where(paddle.isnan(x), paddle.zeros_like(x), x)
+        x = self.conv2(x)
+        x = self.conv_bn2(x)
+        if self.fix_nan and self.training:
+            x = paddle.where(paddle.isnan(x), paddle.zeros_like(x), x)
+        if return_f is True:
+            f = x
+        x = self.conv3(x)
+        x = F.sigmoid(x)
+        if return_f is True:
+            return x, f
+        return x
+
+    @paddle.no_grad()
+    def rep(self):
+        """Fuse Conv+BN and ConvTranspose+BN pairs for deployment."""
+        if self.is_repped:
+            return
+
+        # conv1 (Conv2D, no bias) + conv_bn1 (BatchNorm, act=relu)
+        self.conv1 = self._fuse_conv_bn(self.conv1, self.conv_bn1)
+        self.conv_bn1 = nn.ReLU()
+
+        # conv2 (Conv2DTranspose, has bias) + conv_bn2 (BatchNorm, act=relu)
+        self.conv2 = self._fuse_convtranspose_bn(self.conv2, self.conv_bn2)
+        self.conv_bn2 = nn.ReLU()
+
+        self.is_repped = True
+
+    @staticmethod
+    @paddle.no_grad()
+    def _fuse_conv_bn(conv, bn):
+        """Fuse Conv2D + BatchNorm into Conv2D with bias."""
+        gamma = bn.weight
+        std = paddle.sqrt(bn._variance + bn._epsilon)
+        scale = gamma / std
+
+        w = conv.weight * scale[:, None, None, None]
+        b = bn.bias - bn._mean * scale
+
+        fused = nn.Conv2D(
+            conv._in_channels,
+            conv._out_channels,
+            conv._kernel_size,
+            stride=conv._stride,
+            padding=conv._padding,
+            dilation=conv._dilation,
+            groups=conv._groups,
+        )
+        fused.weight.set_value(w)
+        fused.bias.set_value(b)
+        return fused
+
+    @staticmethod
+    @paddle.no_grad()
+    def _fuse_convtranspose_bn(conv, bn):
+        """Fuse Conv2DTranspose + BatchNorm into Conv2DTranspose with bias.
+
+        Conv2DTranspose weight shape: [in_ch, out_ch/groups, kH, kW]
+        BN scale applies on out_ch, i.e. axis=1.
+        """
+        gamma = bn.weight
+        std = paddle.sqrt(bn._variance + bn._epsilon)
+        scale = gamma / std
+
+        # axis=1 for ConvTranspose (output channel dimension)
+        w = conv.weight * scale[None, :, None, None]
+        b = bn.bias - bn._mean * scale
+        if conv.bias is not None:
+            b = b + conv.bias * scale
+
+        fused = nn.Conv2DTranspose(
+            conv._in_channels,
+            conv._out_channels,
+            conv._kernel_size,
+            stride=conv._stride,
+            padding=conv._padding,
+            dilation=conv._dilation,
+            groups=conv._groups,
+        )
+        fused.weight.set_value(w)
+        fused.bias.set_value(b)
+        return fused
+
+
+class DBHead(nn.Layer):
+    """
+    Differentiable Binarization (DB) for text detection:
+        see https://arxiv.org/abs/1911.08947
+    args:
+        params(dict): super parameters for build DB network
+    """
+
+    def __init__(self, in_channels, k=50, aux_in_channels=0, **kwargs):
+        super(DBHead, self).__init__()
+        self.k = k
+        self.is_repped = False
+        self.binarize = Head(in_channels, **kwargs)
+        self.thresh = Head(in_channels, **kwargs)
+        self.aux_in_channels = aux_in_channels
+
+        if aux_in_channels > 0:
+            self._aux_upsample_scale = {
+                "aux_p4": 4,  # 1/16 -> 1/4
+                "aux_p3": 2,  # 1/8  -> 1/4
+                "aux_p2": 1,  # 1/4  -> 1/4 (no-op)
+            }
+            # Create independent binarize + thresh Head pairs for each scale
+            self.aux_binarize_p4 = Head(aux_in_channels, **kwargs)
+            self.aux_thresh_p4 = Head(aux_in_channels, **kwargs)
+            self.aux_binarize_p3 = Head(aux_in_channels, **kwargs)
+            self.aux_thresh_p3 = Head(aux_in_channels, **kwargs)
+            self.aux_binarize_p2 = Head(aux_in_channels, **kwargs)
+            self.aux_thresh_p2 = Head(aux_in_channels, **kwargs)
+
+    def step_function(self, x, y):
+        return paddle.reciprocal(1 + paddle.exp(-self.k * (x - y)))
+
+    def forward(self, x, targets=None):
+        # Compatible with neck returning dict (training) or tensor (inference)
+        if isinstance(x, dict):
+            fuse = x["fuse"]
+            aux_feats = {k: x[k] for k in ("aux_p4", "aux_p3", "aux_p2") if k in x}
+        else:
+            fuse = x
+            aux_feats = {}
+
+        shrink_maps = self.binarize(fuse)
+        if not self.training:
+            return {"maps": shrink_maps}
+
+        threshold_maps = self.thresh(fuse)
+        binary_maps = self.step_function(shrink_maps, threshold_maps)
+        y = paddle.concat([shrink_maps, threshold_maps, binary_maps], axis=1)
+        result = {"maps": y}
+
+        if self.aux_in_channels > 0 and aux_feats:
+            for key, feat in aux_feats.items():
+                scale = self._aux_upsample_scale[key]
+                if scale > 1:
+                    feat = F.interpolate(
+                        feat, scale_factor=scale, mode="bilinear", align_corners=False
+                    )
+                level = key[4:]  # 'p4', 'p3', 'p2'
+                aux_binarize = getattr(self, "aux_binarize_" + level)
+                aux_thresh_head = getattr(self, "aux_thresh_" + level)
+                aux_shrink = aux_binarize(feat)
+                aux_thresh = aux_thresh_head(feat)
+                aux_binary = self.step_function(aux_shrink, aux_thresh)
+                result["aux_maps_" + level] = paddle.concat(
+                    [aux_shrink, aux_thresh, aux_binary], axis=1
+                )
+
+        return result
+
+    def rep(self):
+        """Fuse reparam structures in all sub-modules for deployment."""
+        if self.is_repped:
+            return
+        for layer in self.sublayers():
+            if isinstance(layer, Head):
+                layer.rep()
+        self.is_repped = True
+
+
+class LocalModule(nn.Layer):
+    def __init__(self, in_c, mid_c, use_distance=True):
+        super(self.__class__, self).__init__()
+        self.last_3 = ConvBNLayer(in_c + 1, mid_c, 3, 1, 1, act="relu")
+        self.last_1 = nn.Conv2D(mid_c, 1, 1, 1, 0)
+
+    def forward(self, x, init_map, distance_map):
+        outf = paddle.concat([init_map, x], axis=1)
+        # last Conv
+        out = self.last_1(self.last_3(outf))
+        return out
+
+
+class PFHeadLocal(DBHead):
+    def __init__(self, in_channels, k=50, mode="small", **kwargs):
+        super(PFHeadLocal, self).__init__(in_channels, k, **kwargs)
+        self.mode = mode
+
+        self.up_conv = nn.Upsample(scale_factor=2, mode="nearest", align_mode=1)
+        if self.mode == "large":
+            self.cbn_layer = LocalModule(in_channels // 4, in_channels // 4)
+        elif self.mode == "small":
+            self.cbn_layer = LocalModule(in_channels // 4, in_channels // 8)
+
+    def forward(self, x, targets=None):
+        shrink_maps, f = self.binarize(x, return_f=True)
+        base_maps = shrink_maps
+        cbn_maps = self.cbn_layer(self.up_conv(f), shrink_maps, None)
+        cbn_maps = F.sigmoid(cbn_maps)
+        if not self.training:
+            return {"maps": 0.5 * (base_maps + cbn_maps)}
+
+        threshold_maps = self.thresh(x)
+        binary_maps = self.step_function(shrink_maps, threshold_maps)
+        y = paddle.concat([cbn_maps, threshold_maps, binary_maps], axis=1)
+        return {"maps": y, "distance_maps": cbn_maps, "cbn_maps": binary_maps}
diff --git a/ppocr/modeling/heads/det_drrg_head.py b/ppocr/modeling/heads/det_drrg_head.py
new file mode 100644
index 0000000..4d6f875
--- /dev/null
+++ b/ppocr/modeling/heads/det_drrg_head.py
@@ -0,0 +1,216 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textdet/dense_heads/drrg_head.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import warnings
+import cv2
+import numpy as np
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from .gcn import GCN
+from .local_graph import LocalGraphs
+from .proposal_local_graph import ProposalLocalGraphs
+
+
+class DRRGHead(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        k_at_hops=(8, 4),
+        num_adjacent_linkages=3,
+        node_geo_feat_len=120,
+        pooling_scale=1.0,
+        pooling_output_size=(4, 3),
+        nms_thr=0.3,
+        min_width=8.0,
+        max_width=24.0,
+        comp_shrink_ratio=1.03,
+        comp_ratio=0.4,
+        comp_score_thr=0.3,
+        text_region_thr=0.2,
+        center_region_thr=0.2,
+        center_region_area_thr=50,
+        local_graph_thr=0.7,
+        **kwargs,
+    ):
+        super().__init__()
+
+        assert isinstance(in_channels, int)
+        assert isinstance(k_at_hops, tuple)
+        assert isinstance(num_adjacent_linkages, int)
+        assert isinstance(node_geo_feat_len, int)
+        assert isinstance(pooling_scale, float)
+        assert isinstance(pooling_output_size, tuple)
+        assert isinstance(comp_shrink_ratio, float)
+        assert isinstance(nms_thr, float)
+        assert isinstance(min_width, float)
+        assert isinstance(max_width, float)
+        assert isinstance(comp_ratio, float)
+        assert isinstance(comp_score_thr, float)
+        assert isinstance(text_region_thr, float)
+        assert isinstance(center_region_thr, float)
+        assert isinstance(center_region_area_thr, int)
+        assert isinstance(local_graph_thr, float)
+
+        self.in_channels = in_channels
+        self.out_channels = 6
+        self.downsample_ratio = 1.0
+        self.k_at_hops = k_at_hops
+        self.num_adjacent_linkages = num_adjacent_linkages
+        self.node_geo_feat_len = node_geo_feat_len
+        self.pooling_scale = pooling_scale
+        self.pooling_output_size = pooling_output_size
+        self.comp_shrink_ratio = comp_shrink_ratio
+        self.nms_thr = nms_thr
+        self.min_width = min_width
+        self.max_width = max_width
+        self.comp_ratio = comp_ratio
+        self.comp_score_thr = comp_score_thr
+        self.text_region_thr = text_region_thr
+        self.center_region_thr = center_region_thr
+        self.center_region_area_thr = center_region_area_thr
+        self.local_graph_thr = local_graph_thr
+
+        self.out_conv = nn.Conv2D(
+            in_channels=self.in_channels,
+            out_channels=self.out_channels,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        )
+
+        self.graph_train = LocalGraphs(
+            self.k_at_hops,
+            self.num_adjacent_linkages,
+            self.node_geo_feat_len,
+            self.pooling_scale,
+            self.pooling_output_size,
+            self.local_graph_thr,
+        )
+
+        self.graph_test = ProposalLocalGraphs(
+            self.k_at_hops,
+            self.num_adjacent_linkages,
+            self.node_geo_feat_len,
+            self.pooling_scale,
+            self.pooling_output_size,
+            self.nms_thr,
+            self.min_width,
+            self.max_width,
+            self.comp_shrink_ratio,
+            self.comp_ratio,
+            self.comp_score_thr,
+            self.text_region_thr,
+            self.center_region_thr,
+            self.center_region_area_thr,
+        )
+
+        pool_w, pool_h = self.pooling_output_size
+        node_feat_len = (pool_w * pool_h) * (
+            self.in_channels + self.out_channels
+        ) + self.node_geo_feat_len
+        self.gcn = GCN(node_feat_len)
+
+    def forward(self, inputs, targets=None):
+        """
+        Args:
+            inputs (Tensor): Shape of :math:`(N, C, H, W)`.
+            gt_comp_attribs (list[ndarray]): The padded text component
+                attributes. Shape: (num_component, 8).
+
+        Returns:
+            tuple: Returns (pred_maps, (gcn_pred, gt_labels)).
+
+                - | pred_maps (Tensor): Prediction map with shape
+                    :math:`(N, C_{out}, H, W)`.
+                - | gcn_pred (Tensor): Prediction from GCN module, with
+                    shape :math:`(N, 2)`.
+                - | gt_labels (Tensor): Ground-truth label with shape
+                    :math:`(N, 8)`.
+        """
+        if self.training:
+            assert targets is not None
+            gt_comp_attribs = targets[7]
+            pred_maps = self.out_conv(inputs)
+            feat_maps = paddle.concat([inputs, pred_maps], axis=1)
+            node_feats, adjacent_matrices, knn_inds, gt_labels = self.graph_train(
+                feat_maps, np.stack(gt_comp_attribs)
+            )
+
+            gcn_pred = self.gcn(node_feats, adjacent_matrices, knn_inds)
+
+            return pred_maps, (gcn_pred, gt_labels)
+        else:
+            return self.single_test(inputs)
+
+    def single_test(self, feat_maps):
+        r"""
+        Args:
+            feat_maps (Tensor): Shape of :math:`(N, C, H, W)`.
+
+        Returns:
+            tuple: Returns (edge, score, text_comps).
+
+                - | edge (ndarray): The edge array of shape :math:`(N, 2)`
+                    where each row is a pair of text component indices
+                    that makes up an edge in graph.
+                - | score (ndarray): The score array of shape :math:`(N,)`,
+                    corresponding to the edge above.
+                - | text_comps (ndarray): The text components of shape
+                    :math:`(N, 9)` where each row corresponds to one box and
+                    its score: (x1, y1, x2, y2, x3, y3, x4, y4, score).
+        """
+        pred_maps = self.out_conv(feat_maps)
+        feat_maps = paddle.concat([feat_maps, pred_maps], axis=1)
+
+        none_flag, graph_data = self.graph_test(pred_maps, feat_maps)
+
+        (
+            local_graphs_node_feat,
+            adjacent_matrices,
+            pivots_knn_inds,
+            pivot_local_graphs,
+            text_comps,
+        ) = graph_data
+
+        if none_flag:
+            return None, None, None
+        gcn_pred = self.gcn(local_graphs_node_feat, adjacent_matrices, pivots_knn_inds)
+        pred_labels = F.softmax(gcn_pred, axis=1)
+
+        edges = []
+        scores = []
+        pivot_local_graphs = pivot_local_graphs.squeeze().numpy()
+
+        for pivot_ind, pivot_local_graph in enumerate(pivot_local_graphs):
+            pivot = pivot_local_graph[0]
+            for k_ind, neighbor_ind in enumerate(pivots_knn_inds[pivot_ind]):
+                neighbor = pivot_local_graph[neighbor_ind.item()]
+                edges.append([pivot, neighbor])
+                scores.append(
+                    pred_labels[pivot_ind * pivots_knn_inds.shape[1] + k_ind, 1].item()
+                )
+
+        edges = np.asarray(edges)
+        scores = np.asarray(scores)
+
+        return edges, scores, text_comps
diff --git a/ppocr/modeling/heads/det_east_head.py b/ppocr/modeling/heads/det_east_head.py
new file mode 100644
index 0000000..c0ad6e8
--- /dev/null
+++ b/ppocr/modeling/heads/det_east_head.py
@@ -0,0 +1,129 @@
+# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride,
+        padding,
+        groups=1,
+        if_act=True,
+        act=None,
+        name=None,
+    ):
+        super(ConvBNLayer, self).__init__()
+        self.if_act = if_act
+        self.act = act
+        self.conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            groups=groups,
+            weight_attr=ParamAttr(name=name + "_weights"),
+            bias_attr=False,
+        )
+
+        self.bn = nn.BatchNorm(
+            num_channels=out_channels,
+            act=act,
+            param_attr=ParamAttr(name="bn_" + name + "_scale"),
+            bias_attr=ParamAttr(name="bn_" + name + "_offset"),
+            moving_mean_name="bn_" + name + "_mean",
+            moving_variance_name="bn_" + name + "_variance",
+        )
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        return x
+
+
+class EASTHead(nn.Layer):
+    """ """
+
+    def __init__(self, in_channels, model_name, **kwargs):
+        super(EASTHead, self).__init__()
+        self.model_name = model_name
+        if self.model_name == "large":
+            num_outputs = [128, 64, 1, 8]
+        else:
+            num_outputs = [64, 32, 1, 8]
+
+        self.det_conv1 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=num_outputs[0],
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            if_act=True,
+            act="relu",
+            name="det_head1",
+        )
+        self.det_conv2 = ConvBNLayer(
+            in_channels=num_outputs[0],
+            out_channels=num_outputs[1],
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            if_act=True,
+            act="relu",
+            name="det_head2",
+        )
+        self.score_conv = ConvBNLayer(
+            in_channels=num_outputs[1],
+            out_channels=num_outputs[2],
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            if_act=False,
+            act=None,
+            name="f_score",
+        )
+        self.geo_conv = ConvBNLayer(
+            in_channels=num_outputs[1],
+            out_channels=num_outputs[3],
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            if_act=False,
+            act=None,
+            name="f_geo",
+        )
+
+    def forward(self, x, targets=None):
+        f_det = self.det_conv1(x)
+        f_det = self.det_conv2(f_det)
+        f_score = self.score_conv(f_det)
+        f_score = F.sigmoid(f_score)
+        f_geo = self.geo_conv(f_det)
+        f_geo = (F.sigmoid(f_geo) - 0.5) * 2 * 800
+
+        pred = {"f_score": f_score, "f_geo": f_geo}
+        return pred
diff --git a/ppocr/modeling/heads/det_fce_head.py b/ppocr/modeling/heads/det_fce_head.py
new file mode 100644
index 0000000..1a90a9a
--- /dev/null
+++ b/ppocr/modeling/heads/det_fce_head.py
@@ -0,0 +1,100 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textdet/dense_heads/fce_head.py
+"""
+
+from paddle import nn
+from paddle import ParamAttr
+import paddle.nn.functional as F
+from paddle.nn.initializer import Normal
+import paddle
+from functools import partial
+
+
+def multi_apply(func, *args, **kwargs):
+    pfunc = partial(func, **kwargs) if kwargs else func
+    map_results = map(pfunc, *args)
+    return tuple(map(list, zip(*map_results)))
+
+
+class FCEHead(nn.Layer):
+    """The class for implementing FCENet head.
+    FCENet(CVPR2021): Fourier Contour Embedding for Arbitrary-shaped Text
+    Detection.
+
+    [https://arxiv.org/abs/2104.10442]
+
+    Args:
+        in_channels (int): The number of input channels.
+        scales (list[int]) : The scale of each layer.
+        fourier_degree (int) : The maximum Fourier transform degree k.
+    """
+
+    def __init__(self, in_channels, fourier_degree=5):
+        super().__init__()
+        assert isinstance(in_channels, int)
+
+        self.downsample_ratio = 1.0
+        self.in_channels = in_channels
+        self.fourier_degree = fourier_degree
+        self.out_channels_cls = 4
+        self.out_channels_reg = (2 * self.fourier_degree + 1) * 2
+
+        self.out_conv_cls = nn.Conv2D(
+            in_channels=self.in_channels,
+            out_channels=self.out_channels_cls,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            groups=1,
+            weight_attr=ParamAttr(
+                name="cls_weights", initializer=Normal(mean=0.0, std=0.01)
+            ),
+            bias_attr=True,
+        )
+        self.out_conv_reg = nn.Conv2D(
+            in_channels=self.in_channels,
+            out_channels=self.out_channels_reg,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            groups=1,
+            weight_attr=ParamAttr(
+                name="reg_weights", initializer=Normal(mean=0.0, std=0.01)
+            ),
+            bias_attr=True,
+        )
+
+    def forward(self, feats, targets=None):
+        cls_res, reg_res = multi_apply(self.forward_single, feats)
+        level_num = len(cls_res)
+        outs = {}
+        if not self.training:
+            for i in range(level_num):
+                tr_pred = F.softmax(cls_res[i][:, 0:2, :, :], axis=1)
+                tcl_pred = F.softmax(cls_res[i][:, 2:, :, :], axis=1)
+                outs["level_{}".format(i)] = paddle.concat(
+                    [tr_pred, tcl_pred, reg_res[i]], axis=1
+                )
+        else:
+            preds = [[cls_res[i], reg_res[i]] for i in range(level_num)]
+            outs["levels"] = preds
+        return outs
+
+    def forward_single(self, x):
+        cls_predict = self.out_conv_cls(x)
+        reg_predict = self.out_conv_reg(x)
+        return cls_predict, reg_predict
diff --git a/ppocr/modeling/heads/det_pse_head.py b/ppocr/modeling/heads/det_pse_head.py
new file mode 100644
index 0000000..2f51621
--- /dev/null
+++ b/ppocr/modeling/heads/det_pse_head.py
@@ -0,0 +1,39 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/whai362/PSENet/blob/python3/models/head/psenet_head.py
+"""
+
+from paddle import nn
+
+
+class PSEHead(nn.Layer):
+    def __init__(self, in_channels, hidden_dim=256, out_channels=7, **kwargs):
+        super(PSEHead, self).__init__()
+        self.conv1 = nn.Conv2D(
+            in_channels, hidden_dim, kernel_size=3, stride=1, padding=1
+        )
+        self.bn1 = nn.BatchNorm2D(hidden_dim)
+        self.relu1 = nn.ReLU()
+
+        self.conv2 = nn.Conv2D(
+            hidden_dim, out_channels, kernel_size=1, stride=1, padding=0
+        )
+
+    def forward(self, x, **kwargs):
+        out = self.conv1(x)
+        out = self.relu1(self.bn1(out))
+        out = self.conv2(out)
+        return {"maps": out}
diff --git a/ppocr/modeling/heads/det_sast_head.py b/ppocr/modeling/heads/det_sast_head.py
new file mode 100644
index 0000000..9246355
--- /dev/null
+++ b/ppocr/modeling/heads/det_sast_head.py
@@ -0,0 +1,152 @@
+# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride,
+        groups=1,
+        if_act=True,
+        act=None,
+        name=None,
+    ):
+        super(ConvBNLayer, self).__init__()
+        self.if_act = if_act
+        self.act = act
+        self.conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=(kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(name=name + "_weights"),
+            bias_attr=False,
+        )
+
+        self.bn = nn.BatchNorm(
+            num_channels=out_channels,
+            act=act,
+            param_attr=ParamAttr(name="bn_" + name + "_scale"),
+            bias_attr=ParamAttr(name="bn_" + name + "_offset"),
+            moving_mean_name="bn_" + name + "_mean",
+            moving_variance_name="bn_" + name + "_variance",
+        )
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        return x
+
+
+class SAST_Header1(nn.Layer):
+    def __init__(self, in_channels, **kwargs):
+        super(SAST_Header1, self).__init__()
+        out_channels = [64, 64, 128]
+        self.score_conv = nn.Sequential(
+            ConvBNLayer(
+                in_channels, out_channels[0], 1, 1, act="relu", name="f_score1"
+            ),
+            ConvBNLayer(
+                out_channels[0], out_channels[1], 3, 1, act="relu", name="f_score2"
+            ),
+            ConvBNLayer(
+                out_channels[1], out_channels[2], 1, 1, act="relu", name="f_score3"
+            ),
+            ConvBNLayer(out_channels[2], 1, 3, 1, act=None, name="f_score4"),
+        )
+        self.border_conv = nn.Sequential(
+            ConvBNLayer(
+                in_channels, out_channels[0], 1, 1, act="relu", name="f_border1"
+            ),
+            ConvBNLayer(
+                out_channels[0], out_channels[1], 3, 1, act="relu", name="f_border2"
+            ),
+            ConvBNLayer(
+                out_channels[1], out_channels[2], 1, 1, act="relu", name="f_border3"
+            ),
+            ConvBNLayer(out_channels[2], 4, 3, 1, act=None, name="f_border4"),
+        )
+
+    def forward(self, x):
+        f_score = self.score_conv(x)
+        f_score = F.sigmoid(f_score)
+        f_border = self.border_conv(x)
+        return f_score, f_border
+
+
+class SAST_Header2(nn.Layer):
+    def __init__(self, in_channels, **kwargs):
+        super(SAST_Header2, self).__init__()
+        out_channels = [64, 64, 128]
+        self.tvo_conv = nn.Sequential(
+            ConvBNLayer(in_channels, out_channels[0], 1, 1, act="relu", name="f_tvo1"),
+            ConvBNLayer(
+                out_channels[0], out_channels[1], 3, 1, act="relu", name="f_tvo2"
+            ),
+            ConvBNLayer(
+                out_channels[1], out_channels[2], 1, 1, act="relu", name="f_tvo3"
+            ),
+            ConvBNLayer(out_channels[2], 8, 3, 1, act=None, name="f_tvo4"),
+        )
+        self.tco_conv = nn.Sequential(
+            ConvBNLayer(in_channels, out_channels[0], 1, 1, act="relu", name="f_tco1"),
+            ConvBNLayer(
+                out_channels[0], out_channels[1], 3, 1, act="relu", name="f_tco2"
+            ),
+            ConvBNLayer(
+                out_channels[1], out_channels[2], 1, 1, act="relu", name="f_tco3"
+            ),
+            ConvBNLayer(out_channels[2], 2, 3, 1, act=None, name="f_tco4"),
+        )
+
+    def forward(self, x):
+        f_tvo = self.tvo_conv(x)
+        f_tco = self.tco_conv(x)
+        return f_tvo, f_tco
+
+
+class SASTHead(nn.Layer):
+    """ """
+
+    def __init__(self, in_channels, **kwargs):
+        super(SASTHead, self).__init__()
+
+        self.head1 = SAST_Header1(in_channels)
+        self.head2 = SAST_Header2(in_channels)
+
+    def forward(self, x, targets=None):
+        f_score, f_border = self.head1(x)
+        f_tvo, f_tco = self.head2(x)
+
+        predicts = {}
+        predicts["f_score"] = f_score
+        predicts["f_border"] = f_border
+        predicts["f_tvo"] = f_tvo
+        predicts["f_tco"] = f_tco
+        return predicts
diff --git a/ppocr/modeling/heads/e2e_pg_head.py b/ppocr/modeling/heads/e2e_pg_head.py
new file mode 100644
index 0000000..27b0472
--- /dev/null
+++ b/ppocr/modeling/heads/e2e_pg_head.py
@@ -0,0 +1,282 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride,
+        padding,
+        groups=1,
+        if_act=True,
+        act=None,
+        name=None,
+    ):
+        super(ConvBNLayer, self).__init__()
+        self.if_act = if_act
+        self.act = act
+        self.conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            groups=groups,
+            weight_attr=ParamAttr(name=name + "_weights"),
+            bias_attr=False,
+        )
+
+        self.bn = nn.BatchNorm(
+            num_channels=out_channels,
+            act=act,
+            param_attr=ParamAttr(name="bn_" + name + "_scale"),
+            bias_attr=ParamAttr(name="bn_" + name + "_offset"),
+            moving_mean_name="bn_" + name + "_mean",
+            moving_variance_name="bn_" + name + "_variance",
+            use_global_stats=False,
+        )
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        return x
+
+
+class PGHead(nn.Layer):
+    """ """
+
+    def __init__(
+        self, in_channels, character_dict_path="ppocr/utils/ic15_dict.txt", **kwargs
+    ):
+        super(PGHead, self).__init__()
+
+        # get character_length
+        with open(character_dict_path, "rb") as fin:
+            lines = fin.readlines()
+            character_length = len(lines) + 1
+
+        self.conv_f_score1 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=64,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            act="relu",
+            name="conv_f_score{}".format(1),
+        )
+        self.conv_f_score2 = ConvBNLayer(
+            in_channels=64,
+            out_channels=64,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            act="relu",
+            name="conv_f_score{}".format(2),
+        )
+        self.conv_f_score3 = ConvBNLayer(
+            in_channels=64,
+            out_channels=128,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            act="relu",
+            name="conv_f_score{}".format(3),
+        )
+
+        self.conv1 = nn.Conv2D(
+            in_channels=128,
+            out_channels=1,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            groups=1,
+            weight_attr=ParamAttr(name="conv_f_score{}".format(4)),
+            bias_attr=False,
+        )
+
+        self.conv_f_boder1 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=64,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            act="relu",
+            name="conv_f_boder{}".format(1),
+        )
+        self.conv_f_boder2 = ConvBNLayer(
+            in_channels=64,
+            out_channels=64,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            act="relu",
+            name="conv_f_boder{}".format(2),
+        )
+        self.conv_f_boder3 = ConvBNLayer(
+            in_channels=64,
+            out_channels=128,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            act="relu",
+            name="conv_f_boder{}".format(3),
+        )
+        self.conv2 = nn.Conv2D(
+            in_channels=128,
+            out_channels=4,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            groups=1,
+            weight_attr=ParamAttr(name="conv_f_boder{}".format(4)),
+            bias_attr=False,
+        )
+        self.conv_f_char1 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=128,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            act="relu",
+            name="conv_f_char{}".format(1),
+        )
+        self.conv_f_char2 = ConvBNLayer(
+            in_channels=128,
+            out_channels=128,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            act="relu",
+            name="conv_f_char{}".format(2),
+        )
+        self.conv_f_char3 = ConvBNLayer(
+            in_channels=128,
+            out_channels=256,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            act="relu",
+            name="conv_f_char{}".format(3),
+        )
+        self.conv_f_char4 = ConvBNLayer(
+            in_channels=256,
+            out_channels=256,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            act="relu",
+            name="conv_f_char{}".format(4),
+        )
+        self.conv_f_char5 = ConvBNLayer(
+            in_channels=256,
+            out_channels=256,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            act="relu",
+            name="conv_f_char{}".format(5),
+        )
+        self.conv3 = nn.Conv2D(
+            in_channels=256,
+            out_channels=character_length,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            groups=1,
+            weight_attr=ParamAttr(name="conv_f_char{}".format(6)),
+            bias_attr=False,
+        )
+
+        self.conv_f_direc1 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=64,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            act="relu",
+            name="conv_f_direc{}".format(1),
+        )
+        self.conv_f_direc2 = ConvBNLayer(
+            in_channels=64,
+            out_channels=64,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            act="relu",
+            name="conv_f_direc{}".format(2),
+        )
+        self.conv_f_direc3 = ConvBNLayer(
+            in_channels=64,
+            out_channels=128,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            act="relu",
+            name="conv_f_direc{}".format(3),
+        )
+        self.conv4 = nn.Conv2D(
+            in_channels=128,
+            out_channels=2,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            groups=1,
+            weight_attr=ParamAttr(name="conv_f_direc{}".format(4)),
+            bias_attr=False,
+        )
+
+    def forward(self, x, targets=None):
+        f_score = self.conv_f_score1(x)
+        f_score = self.conv_f_score2(f_score)
+        f_score = self.conv_f_score3(f_score)
+        f_score = self.conv1(f_score)
+        f_score = F.sigmoid(f_score)
+
+        # f_border
+        f_border = self.conv_f_boder1(x)
+        f_border = self.conv_f_boder2(f_border)
+        f_border = self.conv_f_boder3(f_border)
+        f_border = self.conv2(f_border)
+
+        f_char = self.conv_f_char1(x)
+        f_char = self.conv_f_char2(f_char)
+        f_char = self.conv_f_char3(f_char)
+        f_char = self.conv_f_char4(f_char)
+        f_char = self.conv_f_char5(f_char)
+        f_char = self.conv3(f_char)
+
+        f_direction = self.conv_f_direc1(x)
+        f_direction = self.conv_f_direc2(f_direction)
+        f_direction = self.conv_f_direc3(f_direction)
+        f_direction = self.conv4(f_direction)
+
+        predicts = {}
+        predicts["f_score"] = f_score
+        predicts["f_border"] = f_border
+        predicts["f_char"] = f_char
+        predicts["f_direction"] = f_direction
+        return predicts
diff --git a/ppocr/modeling/heads/gcn.py b/ppocr/modeling/heads/gcn.py
new file mode 100644
index 0000000..6e6e2ea
--- /dev/null
+++ b/ppocr/modeling/heads/gcn.py
@@ -0,0 +1,118 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textdet/modules/gcn.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+
+class BatchNorm1D(nn.BatchNorm1D):
+    def __init__(
+        self,
+        num_features,
+        eps=1e-05,
+        momentum=0.1,
+        affine=True,
+        track_running_stats=True,
+    ):
+        momentum = 1 - momentum
+        weight_attr = None
+        bias_attr = None
+        if not affine:
+            weight_attr = paddle.ParamAttr(learning_rate=0.0)
+            bias_attr = paddle.ParamAttr(learning_rate=0.0)
+        super().__init__(
+            num_features,
+            momentum=momentum,
+            epsilon=eps,
+            weight_attr=weight_attr,
+            bias_attr=bias_attr,
+            use_global_stats=track_running_stats,
+        )
+
+
+class MeanAggregator(nn.Layer):
+    def forward(self, features, A):
+        x = paddle.bmm(A, features)
+        return x
+
+
+class GraphConv(nn.Layer):
+    def __init__(self, in_dim, out_dim):
+        super().__init__()
+        self.in_dim = in_dim
+        self.out_dim = out_dim
+        self.weight = self.create_parameter(
+            [in_dim * 2, out_dim], default_initializer=nn.initializer.XavierUniform()
+        )
+        self.bias = self.create_parameter(
+            [out_dim],
+            is_bias=True,
+            default_initializer=nn.initializer.Assign([0] * out_dim),
+        )
+
+        self.aggregator = MeanAggregator()
+
+    def forward(self, features, A):
+        b, n, d = features.shape
+        assert d == self.in_dim
+        agg_feats = self.aggregator(features, A)
+        cat_feats = paddle.concat([features, agg_feats], axis=2)
+        out = paddle.einsum("bnd,df->bnf", cat_feats, self.weight)
+        out = F.relu(out + self.bias)
+        return out
+
+
+class GCN(nn.Layer):
+    def __init__(self, feat_len):
+        super(GCN, self).__init__()
+        self.bn0 = BatchNorm1D(feat_len, affine=False)
+        self.conv1 = GraphConv(feat_len, 512)
+        self.conv2 = GraphConv(512, 256)
+        self.conv3 = GraphConv(256, 128)
+        self.conv4 = GraphConv(128, 64)
+        self.classifier = nn.Sequential(
+            nn.Linear(64, 32), nn.PReLU(32), nn.Linear(32, 2)
+        )
+
+    def forward(self, x, A, knn_inds):
+        num_local_graphs, num_max_nodes, feat_len = x.shape
+
+        x = x.reshape([-1, feat_len])
+        x = self.bn0(x)
+        x = x.reshape([num_local_graphs, num_max_nodes, feat_len])
+
+        x = self.conv1(x, A)
+        x = self.conv2(x, A)
+        x = self.conv3(x, A)
+        x = self.conv4(x, A)
+        k = knn_inds.shape[-1]
+        mid_feat_len = x.shape[-1]
+        edge_feat = paddle.zeros([num_local_graphs, k, mid_feat_len])
+        for graph_ind in range(num_local_graphs):
+            edge_feat[graph_ind, :, :] = x[graph_ind][
+                paddle.to_tensor(knn_inds[graph_ind])
+            ]
+        edge_feat = edge_feat.reshape([-1, mid_feat_len])
+        pred = self.classifier(edge_feat)
+
+        return pred
diff --git a/ppocr/modeling/heads/kie_sdmgr_head.py b/ppocr/modeling/heads/kie_sdmgr_head.py
new file mode 100644
index 0000000..eca8efd
--- /dev/null
+++ b/ppocr/modeling/heads/kie_sdmgr_head.py
@@ -0,0 +1,223 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# reference from : https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/kie/heads/sdmgr_head.py
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+
+
+class SDMGRHead(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        num_chars=92,
+        visual_dim=16,
+        fusion_dim=1024,
+        node_input=32,
+        node_embed=256,
+        edge_input=5,
+        edge_embed=256,
+        num_gnn=2,
+        num_classes=26,
+        bidirectional=False,
+    ):
+        super().__init__()
+
+        self.fusion = Block([visual_dim, node_embed], node_embed, fusion_dim)
+        self.node_embed = nn.Embedding(num_chars, node_input, 0)
+        hidden = node_embed // 2 if bidirectional else node_embed
+        self.rnn = nn.LSTM(input_size=node_input, hidden_size=hidden, num_layers=1)
+        self.edge_embed = nn.Linear(edge_input, edge_embed)
+        self.gnn_layers = nn.LayerList(
+            [GNNLayer(node_embed, edge_embed) for _ in range(num_gnn)]
+        )
+        self.node_cls = nn.Linear(node_embed, num_classes)
+        self.edge_cls = nn.Linear(edge_embed, 2)
+
+    def forward(self, input, targets):
+        relations, texts, x = input
+        node_nums, char_nums = [], []
+        for text in texts:
+            node_nums.append(text.shape[0])
+            char_nums.append(paddle.sum((text > -1).astype(int), axis=-1))
+
+        max_num = max([char_num.max() for char_num in char_nums])
+        all_nodes = paddle.concat(
+            [
+                paddle.concat(
+                    [text, paddle.zeros((text.shape[0], max_num - text.shape[1]))], -1
+                )
+                for text in texts
+            ]
+        )
+        temp = paddle.clip(all_nodes, min=0).astype(int)
+        embed_nodes = self.node_embed(temp)
+        rnn_nodes, _ = self.rnn(embed_nodes)
+
+        b, h, w = rnn_nodes.shape
+        nodes = paddle.zeros([b, w])
+        all_nums = paddle.concat(char_nums)
+        valid = paddle.nonzero((all_nums > 0).astype(int))
+        temp_all_nums = (paddle.gather(all_nums, valid) - 1).unsqueeze(-1).unsqueeze(-1)
+        temp_all_nums = paddle.expand(
+            temp_all_nums,
+            [temp_all_nums.shape[0], temp_all_nums.shape[1], rnn_nodes.shape[-1]],
+        )
+        temp_all_nodes = paddle.gather(rnn_nodes, valid)
+        N, C, A = temp_all_nodes.shape
+        one_hot = F.one_hot(temp_all_nums[:, 0, :], num_classes=C).transpose([0, 2, 1])
+        one_hot = paddle.multiply(temp_all_nodes, one_hot.astype("float32")).sum(
+            axis=1, keepdim=True
+        )
+        t = one_hot.expand([N, 1, A]).squeeze(1)
+        nodes = paddle.scatter(nodes, valid.squeeze(1), t)
+
+        if x is not None:
+            nodes = self.fusion([x, nodes])
+
+        all_edges = paddle.concat(
+            [rel.reshape([-1, rel.shape[-1]]) for rel in relations]
+        )
+        embed_edges = self.edge_embed(all_edges.astype("float32"))
+        embed_edges = F.normalize(embed_edges)
+
+        for gnn_layer in self.gnn_layers:
+            nodes, cat_nodes = gnn_layer(nodes, embed_edges, node_nums)
+
+        node_cls, edge_cls = self.node_cls(nodes), self.edge_cls(cat_nodes)
+        return node_cls, edge_cls
+
+
+class GNNLayer(nn.Layer):
+    def __init__(self, node_dim=256, edge_dim=256):
+        super().__init__()
+        self.in_fc = nn.Linear(node_dim * 2 + edge_dim, node_dim)
+        self.coef_fc = nn.Linear(node_dim, 1)
+        self.out_fc = nn.Linear(node_dim, node_dim)
+        self.relu = nn.ReLU()
+
+    def forward(self, nodes, edges, nums):
+        start, cat_nodes = 0, []
+        for num in nums:
+            sample_nodes = nodes[start : start + num]
+            cat_nodes.append(
+                paddle.concat(
+                    [
+                        paddle.expand(sample_nodes.unsqueeze(1), [-1, num, -1]),
+                        paddle.expand(sample_nodes.unsqueeze(0), [num, -1, -1]),
+                    ],
+                    -1,
+                ).reshape([num**2, -1])
+            )
+            start += num
+        cat_nodes = paddle.concat([paddle.concat(cat_nodes), edges], -1)
+        cat_nodes = self.relu(self.in_fc(cat_nodes))
+        coefs = self.coef_fc(cat_nodes)
+
+        start, residuals = 0, []
+        for num in nums:
+            residual = F.softmax(
+                -paddle.eye(num).unsqueeze(-1) * 1e9
+                + coefs[start : start + num**2].reshape([num, num, -1]),
+                1,
+            )
+            residuals.append(
+                (
+                    residual * cat_nodes[start : start + num**2].reshape([num, num, -1])
+                ).sum(1)
+            )
+            start += num**2
+
+        nodes += self.relu(self.out_fc(paddle.concat(residuals)))
+        return [nodes, cat_nodes]
+
+
+class Block(nn.Layer):
+    def __init__(
+        self,
+        input_dims,
+        output_dim,
+        mm_dim=1600,
+        chunks=20,
+        rank=15,
+        shared=False,
+        dropout_input=0.0,
+        dropout_pre_lin=0.0,
+        dropout_output=0.0,
+        pos_norm="before_cat",
+    ):
+        super().__init__()
+        self.rank = rank
+        self.dropout_input = dropout_input
+        self.dropout_pre_lin = dropout_pre_lin
+        self.dropout_output = dropout_output
+        assert pos_norm in ["before_cat", "after_cat"]
+        self.pos_norm = pos_norm
+        # Modules
+        self.linear0 = nn.Linear(input_dims[0], mm_dim)
+        self.linear1 = self.linear0 if shared else nn.Linear(input_dims[1], mm_dim)
+        self.merge_linears0 = nn.LayerList()
+        self.merge_linears1 = nn.LayerList()
+        self.chunks = self.chunk_sizes(mm_dim, chunks)
+        for size in self.chunks:
+            ml0 = nn.Linear(size, size * rank)
+            self.merge_linears0.append(ml0)
+            ml1 = ml0 if shared else nn.Linear(size, size * rank)
+            self.merge_linears1.append(ml1)
+        self.linear_out = nn.Linear(mm_dim, output_dim)
+
+    def forward(self, x):
+        x0 = self.linear0(x[0])
+        x1 = self.linear1(x[1])
+        bs = x1.shape[0]
+        if self.dropout_input > 0:
+            x0 = F.dropout(x0, p=self.dropout_input, training=self.training)
+            x1 = F.dropout(x1, p=self.dropout_input, training=self.training)
+        x0_chunks = paddle.split(x0, self.chunks, -1)
+        x1_chunks = paddle.split(x1, self.chunks, -1)
+        zs = []
+        for x0_c, x1_c, m0, m1 in zip(
+            x0_chunks, x1_chunks, self.merge_linears0, self.merge_linears1
+        ):
+            m = m0(x0_c) * m1(x1_c)  # bs x split_size*rank
+            m = m.reshape([bs, self.rank, -1])
+            z = paddle.sum(m, 1)
+            if self.pos_norm == "before_cat":
+                z = paddle.sqrt(F.relu(z)) - paddle.sqrt(F.relu(-z))
+                z = F.normalize(z)
+            zs.append(z)
+        z = paddle.concat(zs, 1)
+        if self.pos_norm == "after_cat":
+            z = paddle.sqrt(F.relu(z)) - paddle.sqrt(F.relu(-z))
+            z = F.normalize(z)
+
+        if self.dropout_pre_lin > 0:
+            z = F.dropout(z, p=self.dropout_pre_lin, training=self.training)
+        z = self.linear_out(z)
+        if self.dropout_output > 0:
+            z = F.dropout(z, p=self.dropout_output, training=self.training)
+        return z
+
+    def chunk_sizes(self, dim, chunks):
+        split_size = (dim + chunks - 1) // chunks
+        sizes_list = [split_size] * chunks
+        sizes_list[-1] = sizes_list[-1] - (sum(sizes_list) - dim)
+        return sizes_list
diff --git a/ppocr/modeling/heads/local_graph.py b/ppocr/modeling/heads/local_graph.py
new file mode 100644
index 0000000..f65c9aa
--- /dev/null
+++ b/ppocr/modeling/heads/local_graph.py
@@ -0,0 +1,425 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textdet/modules/local_graph.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+import paddle
+import paddle.nn as nn
+from ppocr.ext_op import RoIAlignRotated
+
+
+def normalize_adjacent_matrix(A):
+    assert A.ndim == 2
+    assert A.shape[0] == A.shape[1]
+
+    A = A + np.eye(A.shape[0])
+    d = np.sum(A, axis=0)
+    d = np.clip(d, 0, None)
+    d_inv = np.power(d, -0.5).flatten()
+    d_inv[np.isinf(d_inv)] = 0.0
+    d_inv = np.diag(d_inv)
+    G = A.dot(d_inv).transpose().dot(d_inv)
+    return G
+
+
+def euclidean_distance_matrix(A, B):
+    """Calculate the Euclidean distance matrix.
+
+    Args:
+        A (ndarray): The point sequence.
+        B (ndarray): The point sequence with the same dimensions as A.
+
+    returns:
+        D (ndarray): The Euclidean distance matrix.
+    """
+    assert A.ndim == 2
+    assert B.ndim == 2
+    assert A.shape[1] == B.shape[1]
+
+    m = A.shape[0]
+    n = B.shape[0]
+
+    A_dots = (A * A).sum(axis=1).reshape((m, 1)) * np.ones(shape=(1, n))
+    B_dots = (B * B).sum(axis=1) * np.ones(shape=(m, 1))
+    D_squared = A_dots + B_dots - 2 * A.dot(B.T)
+
+    zero_mask = np.less(D_squared, 0.0)
+    D_squared[zero_mask] = 0.0
+    D = np.sqrt(D_squared)
+    return D
+
+
+def feature_embedding(input_feats, out_feat_len):
+    """Embed features. This code was partially adapted from
+    https://github.com/GXYM/DRRG licensed under the MIT license.
+
+    Args:
+        input_feats (ndarray): The input features of shape (N, d), where N is
+            the number of nodes in graph, d is the input feature vector length.
+        out_feat_len (int): The length of output feature vector.
+
+    Returns:
+        embedded_feats (ndarray): The embedded features.
+    """
+    assert input_feats.ndim == 2
+    assert isinstance(out_feat_len, int)
+    assert out_feat_len >= input_feats.shape[1]
+
+    num_nodes = input_feats.shape[0]
+    feat_dim = input_feats.shape[1]
+    feat_repeat_times = out_feat_len // feat_dim
+    residue_dim = out_feat_len % feat_dim
+
+    if residue_dim > 0:
+        embed_wave = np.array(
+            [
+                np.power(1000, 2.0 * (j // 2) / feat_repeat_times + 1)
+                for j in range(feat_repeat_times + 1)
+            ]
+        ).reshape((feat_repeat_times + 1, 1, 1))
+        repeat_feats = np.repeat(
+            np.expand_dims(input_feats, axis=0), feat_repeat_times, axis=0
+        )
+        residue_feats = np.hstack(
+            [
+                input_feats[:, 0:residue_dim],
+                np.zeros((num_nodes, feat_dim - residue_dim)),
+            ]
+        )
+        residue_feats = np.expand_dims(residue_feats, axis=0)
+        repeat_feats = np.concatenate([repeat_feats, residue_feats], axis=0)
+        embedded_feats = repeat_feats / embed_wave
+        embedded_feats[:, 0::2] = np.sin(embedded_feats[:, 0::2])
+        embedded_feats[:, 1::2] = np.cos(embedded_feats[:, 1::2])
+        embedded_feats = np.transpose(embedded_feats, (1, 0, 2)).reshape(
+            (num_nodes, -1)
+        )[:, 0:out_feat_len]
+    else:
+        embed_wave = np.array(
+            [
+                np.power(1000, 2.0 * (j // 2) / feat_repeat_times)
+                for j in range(feat_repeat_times)
+            ]
+        ).reshape((feat_repeat_times, 1, 1))
+        repeat_feats = np.repeat(
+            np.expand_dims(input_feats, axis=0), feat_repeat_times, axis=0
+        )
+        embedded_feats = repeat_feats / embed_wave
+        embedded_feats[:, 0::2] = np.sin(embedded_feats[:, 0::2])
+        embedded_feats[:, 1::2] = np.cos(embedded_feats[:, 1::2])
+        embedded_feats = (
+            np.transpose(embedded_feats, (1, 0, 2))
+            .reshape((num_nodes, -1))
+            .astype(np.float32)
+        )
+
+    return embedded_feats
+
+
+class LocalGraphs:
+    def __init__(
+        self,
+        k_at_hops,
+        num_adjacent_linkages,
+        node_geo_feat_len,
+        pooling_scale,
+        pooling_output_size,
+        local_graph_thr,
+    ):
+        assert len(k_at_hops) == 2
+        assert all(isinstance(n, int) for n in k_at_hops)
+        assert isinstance(num_adjacent_linkages, int)
+        assert isinstance(node_geo_feat_len, int)
+        assert isinstance(pooling_scale, float)
+        assert all(isinstance(n, int) for n in pooling_output_size)
+        assert isinstance(local_graph_thr, float)
+
+        self.k_at_hops = k_at_hops
+        self.num_adjacent_linkages = num_adjacent_linkages
+        self.node_geo_feat_dim = node_geo_feat_len
+        self.pooling = RoIAlignRotated(pooling_output_size, pooling_scale)
+        self.local_graph_thr = local_graph_thr
+
+    def generate_local_graphs(self, sorted_dist_inds, gt_comp_labels):
+        """Generate local graphs for GCN to predict which instance a text
+        component belongs to.
+
+        Args:
+            sorted_dist_inds (ndarray): The complete graph node indices, which
+                is sorted according to the Euclidean distance.
+            gt_comp_labels(ndarray): The ground truth labels define the
+                instance to which the text components (nodes in graphs) belong.
+
+        Returns:
+            pivot_local_graphs(list[list[int]]): The list of local graph
+                neighbor indices of pivots.
+            pivot_knns(list[list[int]]): The list of k-nearest neighbor indices
+                of pivots.
+        """
+
+        assert sorted_dist_inds.ndim == 2
+        assert (
+            sorted_dist_inds.shape[0]
+            == sorted_dist_inds.shape[1]
+            == gt_comp_labels.shape[0]
+        )
+
+        knn_graph = sorted_dist_inds[:, 1 : self.k_at_hops[0] + 1]
+        pivot_local_graphs = []
+        pivot_knns = []
+        for pivot_ind, knn in enumerate(knn_graph):
+            local_graph_neighbors = set(knn)
+
+            for neighbor_ind in knn:
+                local_graph_neighbors.update(
+                    set(sorted_dist_inds[neighbor_ind, 1 : self.k_at_hops[1] + 1])
+                )
+
+            local_graph_neighbors.discard(pivot_ind)
+            pivot_local_graph = list(local_graph_neighbors)
+            pivot_local_graph.insert(0, pivot_ind)
+            pivot_knn = [pivot_ind] + list(knn)
+
+            if pivot_ind < 1:
+                pivot_local_graphs.append(pivot_local_graph)
+                pivot_knns.append(pivot_knn)
+            else:
+                add_flag = True
+                for graph_ind, added_knn in enumerate(pivot_knns):
+                    added_pivot_ind = added_knn[0]
+                    added_local_graph = pivot_local_graphs[graph_ind]
+
+                    union = len(
+                        set(pivot_local_graph[1:]).union(set(added_local_graph[1:]))
+                    )
+                    intersect = len(
+                        set(pivot_local_graph[1:]).intersection(
+                            set(added_local_graph[1:])
+                        )
+                    )
+                    local_graph_iou = intersect / (union + 1e-8)
+
+                    if (
+                        local_graph_iou > self.local_graph_thr
+                        and pivot_ind in added_knn
+                        and gt_comp_labels[added_pivot_ind] == gt_comp_labels[pivot_ind]
+                        and gt_comp_labels[pivot_ind] != 0
+                    ):
+                        add_flag = False
+                        break
+                if add_flag:
+                    pivot_local_graphs.append(pivot_local_graph)
+                    pivot_knns.append(pivot_knn)
+
+        return pivot_local_graphs, pivot_knns
+
+    def generate_gcn_input(
+        self,
+        node_feat_batch,
+        node_label_batch,
+        local_graph_batch,
+        knn_batch,
+        sorted_dist_ind_batch,
+    ):
+        """Generate graph convolution network input data.
+
+        Args:
+            node_feat_batch (List[Tensor]): The batched graph node features.
+            node_label_batch (List[ndarray]): The batched text component
+                labels.
+            local_graph_batch (List[List[list[int]]]): The local graph node
+                indices of image batch.
+            knn_batch (List[List[list[int]]]): The knn graph node indices of
+                image batch.
+            sorted_dist_ind_batch (list[ndarray]): The node indices sorted
+                according to the Euclidean distance.
+
+        Returns:
+            local_graphs_node_feat (Tensor): The node features of graph.
+            adjacent_matrices (Tensor): The adjacent matrices of local graphs.
+            pivots_knn_inds (Tensor): The k-nearest neighbor indices in
+                local graph.
+            gt_linkage (Tensor): The surpervision signal of GCN for linkage
+                prediction.
+        """
+        assert isinstance(node_feat_batch, list)
+        assert isinstance(node_label_batch, list)
+        assert isinstance(local_graph_batch, list)
+        assert isinstance(knn_batch, list)
+        assert isinstance(sorted_dist_ind_batch, list)
+
+        num_max_nodes = max(
+            [
+                len(pivot_local_graph)
+                for pivot_local_graphs in local_graph_batch
+                for pivot_local_graph in pivot_local_graphs
+            ]
+        )
+
+        local_graphs_node_feat = []
+        adjacent_matrices = []
+        pivots_knn_inds = []
+        pivots_gt_linkage = []
+
+        for batch_ind, sorted_dist_inds in enumerate(sorted_dist_ind_batch):
+            node_feats = node_feat_batch[batch_ind]
+            pivot_local_graphs = local_graph_batch[batch_ind]
+            pivot_knns = knn_batch[batch_ind]
+            node_labels = node_label_batch[batch_ind]
+
+            for graph_ind, pivot_knn in enumerate(pivot_knns):
+                pivot_local_graph = pivot_local_graphs[graph_ind]
+                num_nodes = len(pivot_local_graph)
+                pivot_ind = pivot_local_graph[0]
+                node2ind_map = {j: i for i, j in enumerate(pivot_local_graph)}
+
+                knn_inds = paddle.to_tensor([node2ind_map[i] for i in pivot_knn[1:]])
+                pivot_feats = node_feats[pivot_ind]
+                normalized_feats = (
+                    node_feats[paddle.to_tensor(pivot_local_graph)] - pivot_feats
+                )
+
+                adjacent_matrix = np.zeros((num_nodes, num_nodes), dtype=np.float32)
+                for node in pivot_local_graph:
+                    neighbors = sorted_dist_inds[
+                        node, 1 : self.num_adjacent_linkages + 1
+                    ]
+                    for neighbor in neighbors:
+                        if neighbor in pivot_local_graph:
+                            adjacent_matrix[
+                                node2ind_map[node], node2ind_map[neighbor]
+                            ] = 1
+                            adjacent_matrix[
+                                node2ind_map[neighbor], node2ind_map[node]
+                            ] = 1
+
+                adjacent_matrix = normalize_adjacent_matrix(adjacent_matrix)
+                pad_adjacent_matrix = paddle.zeros((num_max_nodes, num_max_nodes))
+                pad_adjacent_matrix[:num_nodes, :num_nodes] = paddle.cast(
+                    paddle.to_tensor(adjacent_matrix), "float32"
+                )
+
+                pad_normalized_feats = paddle.concat(
+                    [
+                        normalized_feats,
+                        paddle.zeros(
+                            (num_max_nodes - num_nodes, normalized_feats.shape[1])
+                        ),
+                    ],
+                    axis=0,
+                )
+                local_graph_labels = node_labels[pivot_local_graph]
+                knn_labels = local_graph_labels[knn_inds.numpy()]
+                link_labels = (
+                    (node_labels[pivot_ind] == knn_labels)
+                    & (node_labels[pivot_ind] > 0)
+                ).astype(np.int64)
+                link_labels = paddle.to_tensor(link_labels)
+
+                local_graphs_node_feat.append(pad_normalized_feats)
+                adjacent_matrices.append(pad_adjacent_matrix)
+                pivots_knn_inds.append(knn_inds)
+                pivots_gt_linkage.append(link_labels)
+
+        local_graphs_node_feat = paddle.stack(local_graphs_node_feat, 0)
+        adjacent_matrices = paddle.stack(adjacent_matrices, 0)
+        pivots_knn_inds = paddle.stack(pivots_knn_inds, 0)
+        pivots_gt_linkage = paddle.stack(pivots_gt_linkage, 0)
+
+        return (
+            local_graphs_node_feat,
+            adjacent_matrices,
+            pivots_knn_inds,
+            pivots_gt_linkage,
+        )
+
+    def __call__(self, feat_maps, comp_attribs):
+        """Generate local graphs as GCN input.
+
+        Args:
+            feat_maps (Tensor): The feature maps to extract the content
+                features of text components.
+            comp_attribs (ndarray): The text component attributes.
+
+        Returns:
+            local_graphs_node_feat (Tensor): The node features of graph.
+            adjacent_matrices (Tensor): The adjacent matrices of local graphs.
+            pivots_knn_inds (Tensor): The k-nearest neighbor indices in local
+                graph.
+            gt_linkage (Tensor): The surpervision signal of GCN for linkage
+                prediction.
+        """
+
+        assert isinstance(feat_maps, paddle.Tensor)
+        assert comp_attribs.ndim == 3
+        assert comp_attribs.shape[2] == 8
+
+        sorted_dist_inds_batch = []
+        local_graph_batch = []
+        knn_batch = []
+        node_feat_batch = []
+        node_label_batch = []
+
+        for batch_ind in range(comp_attribs.shape[0]):
+            num_comps = int(comp_attribs[batch_ind, 0, 0])
+            comp_geo_attribs = comp_attribs[batch_ind, :num_comps, 1:7]
+            node_labels = comp_attribs[batch_ind, :num_comps, 7].astype(np.int32)
+
+            comp_centers = comp_geo_attribs[:, 0:2]
+            distance_matrix = euclidean_distance_matrix(comp_centers, comp_centers)
+
+            batch_id = (
+                np.zeros((comp_geo_attribs.shape[0], 1), dtype=np.float32) * batch_ind
+            )
+            comp_geo_attribs[:, -2] = np.clip(comp_geo_attribs[:, -2], -1, 1)
+            angle = np.arccos(comp_geo_attribs[:, -2]) * np.sign(
+                comp_geo_attribs[:, -1]
+            )
+            angle = angle.reshape((-1, 1))
+            rotated_rois = np.hstack([batch_id, comp_geo_attribs[:, :-2], angle])
+            rois = paddle.to_tensor(rotated_rois)
+            content_feats = self.pooling(feat_maps[batch_ind].unsqueeze(0), rois)
+
+            content_feats = content_feats.reshape([content_feats.shape[0], -1])
+            geo_feats = feature_embedding(comp_geo_attribs, self.node_geo_feat_dim)
+            geo_feats = paddle.to_tensor(geo_feats)
+            node_feats = paddle.concat([content_feats, geo_feats], axis=-1)
+
+            sorted_dist_inds = np.argsort(distance_matrix, axis=1)
+            pivot_local_graphs, pivot_knns = self.generate_local_graphs(
+                sorted_dist_inds, node_labels
+            )
+
+            node_feat_batch.append(node_feats)
+            node_label_batch.append(node_labels)
+            local_graph_batch.append(pivot_local_graphs)
+            knn_batch.append(pivot_knns)
+            sorted_dist_inds_batch.append(sorted_dist_inds)
+
+        (node_feats, adjacent_matrices, knn_inds, gt_linkage) = self.generate_gcn_input(
+            node_feat_batch,
+            node_label_batch,
+            local_graph_batch,
+            knn_batch,
+            sorted_dist_inds_batch,
+        )
+
+        return node_feats, adjacent_matrices, knn_inds, gt_linkage
diff --git a/ppocr/modeling/heads/proposal_local_graph.py b/ppocr/modeling/heads/proposal_local_graph.py
new file mode 100644
index 0000000..7bf0765
--- /dev/null
+++ b/ppocr/modeling/heads/proposal_local_graph.py
@@ -0,0 +1,476 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textdet/modules/proposal_local_graph.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import cv2
+import numpy as np
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from lanms import merge_quadrangle_n9 as la_nms
+
+from ppocr.ext_op import RoIAlignRotated
+from .local_graph import (
+    euclidean_distance_matrix,
+    feature_embedding,
+    normalize_adjacent_matrix,
+)
+
+
+def fill_hole(input_mask):
+    h, w = input_mask.shape
+    canvas = np.zeros((h + 2, w + 2), np.uint8)
+    canvas[1 : h + 1, 1 : w + 1] = input_mask.copy()
+
+    mask = np.zeros((h + 4, w + 4), np.uint8)
+
+    cv2.floodFill(canvas, mask, (0, 0), 1)
+    canvas = canvas[1 : h + 1, 1 : w + 1].astype(np.bool_)
+
+    return ~canvas | input_mask
+
+
+class ProposalLocalGraphs:
+    def __init__(
+        self,
+        k_at_hops,
+        num_adjacent_linkages,
+        node_geo_feat_len,
+        pooling_scale,
+        pooling_output_size,
+        nms_thr,
+        min_width,
+        max_width,
+        comp_shrink_ratio,
+        comp_w_h_ratio,
+        comp_score_thr,
+        text_region_thr,
+        center_region_thr,
+        center_region_area_thr,
+    ):
+        assert len(k_at_hops) == 2
+        assert isinstance(k_at_hops, tuple)
+        assert isinstance(num_adjacent_linkages, int)
+        assert isinstance(node_geo_feat_len, int)
+        assert isinstance(pooling_scale, float)
+        assert isinstance(pooling_output_size, tuple)
+        assert isinstance(nms_thr, float)
+        assert isinstance(min_width, float)
+        assert isinstance(max_width, float)
+        assert isinstance(comp_shrink_ratio, float)
+        assert isinstance(comp_w_h_ratio, float)
+        assert isinstance(comp_score_thr, float)
+        assert isinstance(text_region_thr, float)
+        assert isinstance(center_region_thr, float)
+        assert isinstance(center_region_area_thr, int)
+
+        self.k_at_hops = k_at_hops
+        self.active_connection = num_adjacent_linkages
+        self.local_graph_depth = len(self.k_at_hops)
+        self.node_geo_feat_dim = node_geo_feat_len
+        self.pooling = RoIAlignRotated(pooling_output_size, pooling_scale)
+        self.nms_thr = nms_thr
+        self.min_width = min_width
+        self.max_width = max_width
+        self.comp_shrink_ratio = comp_shrink_ratio
+        self.comp_w_h_ratio = comp_w_h_ratio
+        self.comp_score_thr = comp_score_thr
+        self.text_region_thr = text_region_thr
+        self.center_region_thr = center_region_thr
+        self.center_region_area_thr = center_region_area_thr
+
+    def propose_comps(
+        self,
+        score_map,
+        top_height_map,
+        bot_height_map,
+        sin_map,
+        cos_map,
+        comp_score_thr,
+        min_width,
+        max_width,
+        comp_shrink_ratio,
+        comp_w_h_ratio,
+    ):
+        """Propose text components.
+
+        Args:
+            score_map (ndarray): The score map for NMS.
+            top_height_map (ndarray): The predicted text height map from each
+                pixel in text center region to top sideline.
+            bot_height_map (ndarray): The predicted text height map from each
+                pixel in text center region to bottom sideline.
+            sin_map (ndarray): The predicted sin(theta) map.
+            cos_map (ndarray): The predicted cos(theta) map.
+            comp_score_thr (float): The score threshold of text component.
+            min_width (float): The minimum width of text components.
+            max_width (float): The maximum width of text components.
+            comp_shrink_ratio (float): The shrink ratio of text components.
+            comp_w_h_ratio (float): The width to height ratio of text
+                components.
+
+        Returns:
+            text_comps (ndarray): The text components.
+        """
+
+        comp_centers = np.argwhere(score_map > comp_score_thr)
+        comp_centers = comp_centers[np.argsort(comp_centers[:, 0])]
+        y = comp_centers[:, 0]
+        x = comp_centers[:, 1]
+
+        top_height = top_height_map[y, x].reshape((-1, 1)) * comp_shrink_ratio
+        bot_height = bot_height_map[y, x].reshape((-1, 1)) * comp_shrink_ratio
+        sin = sin_map[y, x].reshape((-1, 1))
+        cos = cos_map[y, x].reshape((-1, 1))
+
+        top_mid_pts = comp_centers + np.hstack([top_height * sin, top_height * cos])
+        bot_mid_pts = comp_centers - np.hstack([bot_height * sin, bot_height * cos])
+
+        width = (top_height + bot_height) * comp_w_h_ratio
+        width = np.clip(width, min_width, max_width)
+        r = width / 2
+
+        tl = top_mid_pts[:, ::-1] - np.hstack([-r * sin, r * cos])
+        tr = top_mid_pts[:, ::-1] + np.hstack([-r * sin, r * cos])
+        br = bot_mid_pts[:, ::-1] + np.hstack([-r * sin, r * cos])
+        bl = bot_mid_pts[:, ::-1] - np.hstack([-r * sin, r * cos])
+        text_comps = np.hstack([tl, tr, br, bl]).astype(np.float32)
+
+        score = score_map[y, x].reshape((-1, 1))
+        text_comps = np.hstack([text_comps, score])
+
+        return text_comps
+
+    def propose_comps_and_attribs(
+        self,
+        text_region_map,
+        center_region_map,
+        top_height_map,
+        bot_height_map,
+        sin_map,
+        cos_map,
+    ):
+        """Generate text components and attributes.
+
+        Args:
+            text_region_map (ndarray): The predicted text region probability
+                map.
+            center_region_map (ndarray): The predicted text center region
+                probability map.
+            top_height_map (ndarray): The predicted text height map from each
+                pixel in text center region to top sideline.
+            bot_height_map (ndarray): The predicted text height map from each
+                pixel in text center region to bottom sideline.
+            sin_map (ndarray): The predicted sin(theta) map.
+            cos_map (ndarray): The predicted cos(theta) map.
+
+        Returns:
+            comp_attribs (ndarray): The text component attributes.
+            text_comps (ndarray): The text components.
+        """
+
+        assert (
+            text_region_map.shape
+            == center_region_map.shape
+            == top_height_map.shape
+            == bot_height_map.shape
+            == sin_map.shape
+            == cos_map.shape
+        )
+        text_mask = text_region_map > self.text_region_thr
+        center_region_mask = (center_region_map > self.center_region_thr) * text_mask
+
+        scale = np.sqrt(1.0 / (sin_map**2 + cos_map**2 + 1e-8))
+        sin_map, cos_map = sin_map * scale, cos_map * scale
+
+        center_region_mask = fill_hole(center_region_mask)
+        center_region_contours, _ = cv2.findContours(
+            center_region_mask.astype(np.uint8), cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE
+        )
+
+        mask_sz = center_region_map.shape
+        comp_list = []
+        for contour in center_region_contours:
+            current_center_mask = np.zeros(mask_sz)
+            cv2.drawContours(current_center_mask, [contour], -1, 1, -1)
+            if current_center_mask.sum() <= self.center_region_area_thr:
+                continue
+            score_map = text_region_map * current_center_mask
+
+            text_comps = self.propose_comps(
+                score_map,
+                top_height_map,
+                bot_height_map,
+                sin_map,
+                cos_map,
+                self.comp_score_thr,
+                self.min_width,
+                self.max_width,
+                self.comp_shrink_ratio,
+                self.comp_w_h_ratio,
+            )
+
+            text_comps = la_nms(text_comps, self.nms_thr)
+            text_comp_mask = np.zeros(mask_sz)
+            text_comp_boxes = text_comps[:, :8].reshape((-1, 4, 2)).astype(np.int32)
+
+            cv2.drawContours(text_comp_mask, text_comp_boxes, -1, 1, -1)
+            if (text_comp_mask * text_mask).sum() < text_comp_mask.sum() * 0.5:
+                continue
+            if text_comps.shape[-1] > 0:
+                comp_list.append(text_comps)
+
+        if len(comp_list) <= 0:
+            return None, None
+
+        text_comps = np.vstack(comp_list)
+        text_comp_boxes = text_comps[:, :8].reshape((-1, 4, 2))
+        centers = np.mean(text_comp_boxes, axis=1).astype(np.int32)
+        x = centers[:, 0]
+        y = centers[:, 1]
+
+        scores = []
+        for text_comp_box in text_comp_boxes:
+            text_comp_box[:, 0] = np.clip(text_comp_box[:, 0], 0, mask_sz[1] - 1)
+            text_comp_box[:, 1] = np.clip(text_comp_box[:, 1], 0, mask_sz[0] - 1)
+            min_coord = np.min(text_comp_box, axis=0).astype(np.int32)
+            max_coord = np.max(text_comp_box, axis=0).astype(np.int32)
+            text_comp_box = text_comp_box - min_coord
+            box_sz = max_coord - min_coord + 1
+            temp_comp_mask = np.zeros((box_sz[1], box_sz[0]), dtype=np.uint8)
+            cv2.fillPoly(temp_comp_mask, [text_comp_box.astype(np.int32)], 1)
+            temp_region_patch = text_region_map[
+                min_coord[1] : (max_coord[1] + 1), min_coord[0] : (max_coord[0] + 1)
+            ]
+            score = cv2.mean(temp_region_patch, temp_comp_mask)[0]
+            scores.append(score)
+        scores = np.array(scores).reshape((-1, 1))
+        text_comps = np.hstack([text_comps[:, :-1], scores])
+
+        h = top_height_map[y, x].reshape((-1, 1)) + bot_height_map[y, x].reshape(
+            (-1, 1)
+        )
+        w = np.clip(h * self.comp_w_h_ratio, self.min_width, self.max_width)
+        sin = sin_map[y, x].reshape((-1, 1))
+        cos = cos_map[y, x].reshape((-1, 1))
+
+        x = x.reshape((-1, 1))
+        y = y.reshape((-1, 1))
+        comp_attribs = np.hstack([x, y, h, w, cos, sin])
+
+        return comp_attribs, text_comps
+
+    def generate_local_graphs(self, sorted_dist_inds, node_feats):
+        """Generate local graphs and graph convolution network input data.
+
+        Args:
+            sorted_dist_inds (ndarray): The node indices sorted according to
+                the Euclidean distance.
+            node_feats (tensor): The features of nodes in graph.
+
+        Returns:
+            local_graphs_node_feats (tensor): The features of nodes in local
+                graphs.
+            adjacent_matrices (tensor): The adjacent matrices.
+            pivots_knn_inds (tensor): The k-nearest neighbor indices in
+                local graphs.
+            pivots_local_graphs (tensor): The indices of nodes in local
+                graphs.
+        """
+
+        assert sorted_dist_inds.ndim == 2
+        assert (
+            sorted_dist_inds.shape[0]
+            == sorted_dist_inds.shape[1]
+            == node_feats.shape[0]
+        )
+
+        knn_graph = sorted_dist_inds[:, 1 : self.k_at_hops[0] + 1]
+        pivot_local_graphs = []
+        pivot_knns = []
+
+        for pivot_ind, knn in enumerate(knn_graph):
+            local_graph_neighbors = set(knn)
+
+            for neighbor_ind in knn:
+                local_graph_neighbors.update(
+                    set(sorted_dist_inds[neighbor_ind, 1 : self.k_at_hops[1] + 1])
+                )
+
+            local_graph_neighbors.discard(pivot_ind)
+            pivot_local_graph = list(local_graph_neighbors)
+            pivot_local_graph.insert(0, pivot_ind)
+            pivot_knn = [pivot_ind] + list(knn)
+
+            pivot_local_graphs.append(pivot_local_graph)
+            pivot_knns.append(pivot_knn)
+
+        num_max_nodes = max(
+            [len(pivot_local_graph) for pivot_local_graph in pivot_local_graphs]
+        )
+
+        local_graphs_node_feat = []
+        adjacent_matrices = []
+        pivots_knn_inds = []
+        pivots_local_graphs = []
+
+        for graph_ind, pivot_knn in enumerate(pivot_knns):
+            pivot_local_graph = pivot_local_graphs[graph_ind]
+            num_nodes = len(pivot_local_graph)
+            pivot_ind = pivot_local_graph[0]
+            node2ind_map = {j: i for i, j in enumerate(pivot_local_graph)}
+
+            knn_inds = paddle.cast(
+                paddle.to_tensor([node2ind_map[i] for i in pivot_knn[1:]]), "int64"
+            )
+            pivot_feats = node_feats[pivot_ind]
+            normalized_feats = (
+                node_feats[paddle.to_tensor(pivot_local_graph)] - pivot_feats
+            )
+
+            adjacent_matrix = np.zeros((num_nodes, num_nodes), dtype=np.float32)
+            for node in pivot_local_graph:
+                neighbors = sorted_dist_inds[node, 1 : self.active_connection + 1]
+                for neighbor in neighbors:
+                    if neighbor in pivot_local_graph:
+                        adjacent_matrix[node2ind_map[node], node2ind_map[neighbor]] = 1
+                        adjacent_matrix[node2ind_map[neighbor], node2ind_map[node]] = 1
+
+            adjacent_matrix = normalize_adjacent_matrix(adjacent_matrix)
+            pad_adjacent_matrix = paddle.zeros(
+                (num_max_nodes, num_max_nodes),
+            )
+            pad_adjacent_matrix[:num_nodes, :num_nodes] = paddle.cast(
+                paddle.to_tensor(adjacent_matrix), "float32"
+            )
+
+            pad_normalized_feats = paddle.concat(
+                [
+                    normalized_feats,
+                    paddle.zeros(
+                        (num_max_nodes - num_nodes, normalized_feats.shape[1]),
+                    ),
+                ],
+                axis=0,
+            )
+
+            local_graph_nodes = paddle.to_tensor(pivot_local_graph)
+            local_graph_nodes = paddle.concat(
+                [
+                    local_graph_nodes,
+                    paddle.zeros([num_max_nodes - num_nodes], dtype="int64"),
+                ],
+                axis=-1,
+            )
+
+            local_graphs_node_feat.append(pad_normalized_feats)
+            adjacent_matrices.append(pad_adjacent_matrix)
+            pivots_knn_inds.append(knn_inds)
+            pivots_local_graphs.append(local_graph_nodes)
+
+        local_graphs_node_feat = paddle.stack(local_graphs_node_feat, 0)
+        adjacent_matrices = paddle.stack(adjacent_matrices, 0)
+        pivots_knn_inds = paddle.stack(pivots_knn_inds, 0)
+        pivots_local_graphs = paddle.stack(pivots_local_graphs, 0)
+
+        return (
+            local_graphs_node_feat,
+            adjacent_matrices,
+            pivots_knn_inds,
+            pivots_local_graphs,
+        )
+
+    def __call__(self, preds, feat_maps):
+        """Generate local graphs and graph convolutional network input data.
+
+        Args:
+            preds (tensor): The predicted maps.
+            feat_maps (tensor): The feature maps to extract content feature of
+                text components.
+
+        Returns:
+            none_flag (bool): The flag showing whether the number of proposed
+                text components is 0.
+            local_graphs_node_feats (tensor): The features of nodes in local
+                graphs.
+            adjacent_matrices (tensor): The adjacent matrices.
+            pivots_knn_inds (tensor): The k-nearest neighbor indices in
+                local graphs.
+            pivots_local_graphs (tensor): The indices of nodes in local
+                graphs.
+            text_comps (ndarray): The predicted text components.
+        """
+        if preds.ndim == 4:
+            assert preds.shape[0] == 1
+            preds = paddle.squeeze(preds)
+        pred_text_region = F.sigmoid(preds[0]).numpy()
+        pred_center_region = F.sigmoid(preds[1]).numpy()
+        pred_sin_map = preds[2].numpy()
+        pred_cos_map = preds[3].numpy()
+        pred_top_height_map = preds[4].numpy()
+        pred_bot_height_map = preds[5].numpy()
+
+        comp_attribs, text_comps = self.propose_comps_and_attribs(
+            pred_text_region,
+            pred_center_region,
+            pred_top_height_map,
+            pred_bot_height_map,
+            pred_sin_map,
+            pred_cos_map,
+        )
+
+        if comp_attribs is None or len(comp_attribs) < 2:
+            none_flag = True
+            return none_flag, (0, 0, 0, 0, 0)
+
+        comp_centers = comp_attribs[:, 0:2]
+        distance_matrix = euclidean_distance_matrix(comp_centers, comp_centers)
+
+        geo_feats = feature_embedding(comp_attribs, self.node_geo_feat_dim)
+        geo_feats = paddle.to_tensor(geo_feats)
+
+        batch_id = np.zeros((comp_attribs.shape[0], 1), dtype=np.float32)
+        comp_attribs = comp_attribs.astype(np.float32)
+        angle = np.arccos(comp_attribs[:, -2]) * np.sign(comp_attribs[:, -1])
+        angle = angle.reshape((-1, 1))
+        rotated_rois = np.hstack([batch_id, comp_attribs[:, :-2], angle])
+        rois = paddle.to_tensor(rotated_rois)
+
+        content_feats = self.pooling(feat_maps, rois)
+        content_feats = content_feats.reshape([content_feats.shape[0], -1])
+        node_feats = paddle.concat([content_feats, geo_feats], axis=-1)
+
+        sorted_dist_inds = np.argsort(distance_matrix, axis=1)
+        (
+            local_graphs_node_feat,
+            adjacent_matrices,
+            pivots_knn_inds,
+            pivots_local_graphs,
+        ) = self.generate_local_graphs(sorted_dist_inds, node_feats)
+
+        none_flag = False
+        return none_flag, (
+            local_graphs_node_feat,
+            adjacent_matrices,
+            pivots_knn_inds,
+            pivots_local_graphs,
+            text_comps,
+        )
diff --git a/ppocr/modeling/heads/rec_abinet_head.py b/ppocr/modeling/heads/rec_abinet_head.py
new file mode 100644
index 0000000..15628c9
--- /dev/null
+++ b/ppocr/modeling/heads/rec_abinet_head.py
@@ -0,0 +1,299 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/FangShancheng/ABINet/tree/main/modules
+"""
+
+import math
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle.nn import LayerList
+from ppocr.modeling.heads.rec_nrtr_head import TransformerBlock, PositionalEncoding
+
+
+class BCNLanguage(nn.Layer):
+    def __init__(
+        self,
+        d_model=512,
+        nhead=8,
+        num_layers=4,
+        dim_feedforward=2048,
+        dropout=0.0,
+        max_length=25,
+        detach=True,
+        num_classes=37,
+    ):
+        super().__init__()
+
+        self.d_model = d_model
+        self.detach = detach
+        self.max_length = max_length + 1  # additional stop token
+        self.proj = nn.Linear(num_classes, d_model, bias_attr=False)
+        self.token_encoder = PositionalEncoding(
+            dropout=0.1, dim=d_model, max_len=self.max_length
+        )
+        self.pos_encoder = PositionalEncoding(
+            dropout=0, dim=d_model, max_len=self.max_length
+        )
+
+        self.decoder = nn.LayerList(
+            [
+                TransformerBlock(
+                    d_model=d_model,
+                    nhead=nhead,
+                    dim_feedforward=dim_feedforward,
+                    attention_dropout_rate=dropout,
+                    residual_dropout_rate=dropout,
+                    with_self_attn=False,
+                    with_cross_attn=True,
+                )
+                for i in range(num_layers)
+            ]
+        )
+
+        self.cls = nn.Linear(d_model, num_classes)
+
+    def forward(self, tokens, lengths):
+        """
+        Args:
+            tokens: (B, N, C) where N is length, B is batch size and C is classes number
+            lengths: (B,)
+        """
+        if self.detach:
+            tokens = tokens.detach()
+        embed = self.proj(tokens)  # (B, N, C)
+        embed = self.token_encoder(embed)  # (B, N, C)
+        padding_mask = _get_mask(lengths, self.max_length)
+        zeros = paddle.zeros_like(embed)  # (B, N, C)
+        query = self.pos_encoder(zeros)
+        for decoder_layer in self.decoder:
+            query = decoder_layer(query, embed, cross_mask=padding_mask)
+        output = query  # (B, N, C)
+
+        logits = self.cls(output)  # (B, N, C)
+
+        return output, logits
+
+
+def encoder_layer(in_c, out_c, k=3, s=2, p=1):
+    return nn.Sequential(
+        nn.Conv2D(in_c, out_c, k, s, p), nn.BatchNorm2D(out_c), nn.ReLU()
+    )
+
+
+def decoder_layer(
+    in_c, out_c, k=3, s=1, p=1, mode="nearest", scale_factor=None, size=None
+):
+    align_corners = False if mode == "nearest" else True
+    return nn.Sequential(
+        nn.Upsample(
+            size=size, scale_factor=scale_factor, mode=mode, align_corners=align_corners
+        ),
+        nn.Conv2D(in_c, out_c, k, s, p),
+        nn.BatchNorm2D(out_c),
+        nn.ReLU(),
+    )
+
+
+class PositionAttention(nn.Layer):
+    def __init__(
+        self,
+        max_length,
+        in_channels=512,
+        num_channels=64,
+        h=8,
+        w=32,
+        mode="nearest",
+        **kwargs,
+    ):
+        super().__init__()
+        self.max_length = max_length
+        self.k_encoder = nn.Sequential(
+            encoder_layer(in_channels, num_channels, s=(1, 2)),
+            encoder_layer(num_channels, num_channels, s=(2, 2)),
+            encoder_layer(num_channels, num_channels, s=(2, 2)),
+            encoder_layer(num_channels, num_channels, s=(2, 2)),
+        )
+        self.k_decoder = nn.Sequential(
+            decoder_layer(num_channels, num_channels, scale_factor=2, mode=mode),
+            decoder_layer(num_channels, num_channels, scale_factor=2, mode=mode),
+            decoder_layer(num_channels, num_channels, scale_factor=2, mode=mode),
+            decoder_layer(num_channels, in_channels, size=(h, w), mode=mode),
+        )
+
+        self.pos_encoder = PositionalEncoding(
+            dropout=0, dim=in_channels, max_len=max_length
+        )
+        self.project = nn.Linear(in_channels, in_channels)
+
+    def forward(self, x):
+        B, C, H, W = x.shape
+        k, v = x, x
+
+        # calculate key vector
+        features = []
+        for i in range(0, len(self.k_encoder)):
+            k = self.k_encoder[i](k)
+            features.append(k)
+        for i in range(0, len(self.k_decoder) - 1):
+            k = self.k_decoder[i](k)
+            # print(k.shape, features[len(self.k_decoder) - 2 - i].shape)
+            k = k + features[len(self.k_decoder) - 2 - i]
+        k = self.k_decoder[-1](k)
+
+        # calculate query vector
+        # TODO q=f(q,k)
+        zeros = paddle.zeros((B, self.max_length, C), dtype=x.dtype)  # (B, N, C)
+        q = self.pos_encoder(zeros)  # (B, N, C)
+        q = self.project(q)  # (B, N, C)
+
+        # calculate attention
+        attn_scores = q @ k.flatten(2)  # (B, N, (H*W))
+        attn_scores = attn_scores / (C**0.5)
+        attn_scores = F.softmax(attn_scores, axis=-1)
+
+        v = v.flatten(2).transpose([0, 2, 1])  # (B, (H*W), C)
+        attn_vecs = attn_scores @ v  # (B, N, C)
+
+        return attn_vecs, attn_scores.reshape([0, self.max_length, H, W])
+
+
+class ABINetHead(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        d_model=512,
+        nhead=8,
+        num_layers=3,
+        dim_feedforward=2048,
+        dropout=0.1,
+        max_length=25,
+        use_lang=False,
+        iter_size=1,
+        image_size=(32, 128),
+    ):
+        super().__init__()
+        self.max_length = max_length + 1
+        h, w = image_size[0] // 4, image_size[1] // 4
+        self.pos_encoder = PositionalEncoding(dropout=0.1, dim=d_model, max_len=h * w)
+        self.encoder = nn.LayerList(
+            [
+                TransformerBlock(
+                    d_model=d_model,
+                    nhead=nhead,
+                    dim_feedforward=dim_feedforward,
+                    attention_dropout_rate=dropout,
+                    residual_dropout_rate=dropout,
+                    with_self_attn=True,
+                    with_cross_attn=False,
+                )
+                for i in range(num_layers)
+            ]
+        )
+        self.decoder = PositionAttention(
+            max_length=max_length + 1, mode="nearest", h=h, w=w  # additional stop token
+        )
+        self.out_channels = out_channels
+        self.cls = nn.Linear(d_model, self.out_channels)
+        self.use_lang = use_lang
+        if use_lang:
+            self.iter_size = iter_size
+            self.language = BCNLanguage(
+                d_model=d_model,
+                nhead=nhead,
+                num_layers=4,
+                dim_feedforward=dim_feedforward,
+                dropout=dropout,
+                max_length=max_length,
+                num_classes=self.out_channels,
+            )
+            # alignment
+            self.w_att_align = nn.Linear(2 * d_model, d_model)
+            self.cls_align = nn.Linear(d_model, self.out_channels)
+
+    def forward(self, x, targets=None):
+        x = x.transpose([0, 2, 3, 1])
+        _, H, W, C = x.shape
+        feature = x.flatten(1, 2)
+        feature = self.pos_encoder(feature)
+        for encoder_layer in self.encoder:
+            feature = encoder_layer(feature)
+        feature = feature.reshape([0, H, W, C]).transpose([0, 3, 1, 2])
+        v_feature, attn_scores = self.decoder(feature)  # (B, N, C), (B, C, H, W)
+        vis_logits = self.cls(v_feature)  # (B, N, C)
+        logits = vis_logits
+        vis_lengths = _get_length(vis_logits)
+        if self.use_lang:
+            align_logits = vis_logits
+            align_lengths = vis_lengths
+            all_l_res, all_a_res = [], []
+            for i in range(self.iter_size):
+                tokens = F.softmax(align_logits, axis=-1)
+                lengths = align_lengths
+                lengths = paddle.clip(
+                    lengths, 2, self.max_length
+                )  # TODO:move to language model
+                l_feature, l_logits = self.language(tokens, lengths)
+
+                # alignment
+                all_l_res.append(l_logits)
+                fuse = paddle.concat((l_feature, v_feature), -1)
+                f_att = F.sigmoid(self.w_att_align(fuse))
+                output = f_att * v_feature + (1 - f_att) * l_feature
+                align_logits = self.cls_align(output)  # (B, N, C)
+
+                align_lengths = _get_length(align_logits)
+                all_a_res.append(align_logits)
+            if self.training:
+                return {"align": all_a_res, "lang": all_l_res, "vision": vis_logits}
+            else:
+                logits = align_logits
+        if self.training:
+            return logits
+        else:
+            return F.softmax(logits, -1)
+
+
+def _get_length(logit):
+    """Greed decoder to obtain length from logit"""
+    out = logit.argmax(-1) == 0
+    abn = out.any(-1)
+    out_int = out.cast("int32")
+    out = (out_int.cumsum(-1) == 1) & out
+    out = out.cast("int32")
+    out = out.argmax(-1)
+    out = out + 1
+    len_seq = paddle.zeros_like(out) + logit.shape[1]
+    out = paddle.where(abn, out, len_seq)
+    return out
+
+
+def _get_mask(length, max_length):
+    """Generate a square mask for the sequence. The masked positions are filled with float('-inf').
+    Unmasked positions are filled with float(0.0).
+    """
+    length = length.unsqueeze(-1)
+    B = length.shape[0]
+    grid = paddle.arange(0, max_length).unsqueeze(0).tile([B, 1])
+    zero_mask = paddle.zeros([B, max_length], dtype="float32")
+    inf_mask = paddle.full([B, max_length], "-inf", dtype="float32")
+    diag_mask = paddle.diag(
+        paddle.full([max_length], "-inf", dtype=paddle.float32), offset=0, name=None
+    )
+    mask = paddle.where(grid >= length, inf_mask, zero_mask)
+    mask = mask.unsqueeze(1) + diag_mask
+    return mask.unsqueeze(1)
diff --git a/ppocr/modeling/heads/rec_aster_head.py b/ppocr/modeling/heads/rec_aster_head.py
new file mode 100644
index 0000000..60eb995
--- /dev/null
+++ b/ppocr/modeling/heads/rec_aster_head.py
@@ -0,0 +1,404 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/ayumiymk/aster.pytorch/blob/master/lib/models/attention_recognition_head.py
+"""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import sys
+
+import paddle
+from paddle import nn
+from paddle.nn import functional as F
+
+
+class AsterHead(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        sDim,
+        attDim,
+        max_len_labels,
+        time_step=25,
+        beam_width=5,
+        **kwargs,
+    ):
+        super(AsterHead, self).__init__()
+        self.num_classes = out_channels
+        self.in_planes = in_channels
+        self.sDim = sDim
+        self.attDim = attDim
+        self.max_len_labels = max_len_labels
+        self.decoder = AttentionRecognitionHead(
+            in_channels, out_channels, sDim, attDim, max_len_labels
+        )
+        self.time_step = time_step
+        self.embedder = Embedding(self.time_step, in_channels)
+        self.beam_width = beam_width
+        self.eos = self.num_classes - 3
+
+    def forward(self, x, targets=None, embed=None):
+        return_dict = {}
+        embedding_vectors = self.embedder(x)
+
+        if self.training:
+            rec_targets, rec_lengths, _ = targets
+            rec_pred = self.decoder([x, rec_targets, rec_lengths], embedding_vectors)
+            return_dict["rec_pred"] = rec_pred
+            return_dict["embedding_vectors"] = embedding_vectors
+        else:
+            rec_pred, rec_pred_scores = self.decoder.beam_search(
+                x, self.beam_width, self.eos, embedding_vectors
+            )
+            return_dict["rec_pred"] = rec_pred
+            return_dict["rec_pred_scores"] = rec_pred_scores
+            return_dict["embedding_vectors"] = embedding_vectors
+
+        return return_dict
+
+
+class Embedding(nn.Layer):
+    def __init__(self, in_timestep, in_planes, mid_dim=4096, embed_dim=300):
+        super(Embedding, self).__init__()
+        self.in_timestep = in_timestep
+        self.in_planes = in_planes
+        self.embed_dim = embed_dim
+        self.mid_dim = mid_dim
+        self.eEmbed = nn.Linear(
+            in_timestep * in_planes, self.embed_dim
+        )  # Embed encoder output to a word-embedding like
+
+    def forward(self, x):
+        x = paddle.reshape(x, [x.shape[0], -1])
+        x = self.eEmbed(x)
+        return x
+
+
+class AttentionRecognitionHead(nn.Layer):
+    """
+    input: [b x 16 x 64 x in_planes]
+    output: probability sequence: [b x T x num_classes]
+    """
+
+    def __init__(self, in_channels, out_channels, sDim, attDim, max_len_labels):
+        super(AttentionRecognitionHead, self).__init__()
+        self.num_classes = (
+            out_channels  # this is the output classes. So it includes the <EOS>.
+        )
+        self.in_planes = in_channels
+        self.sDim = sDim
+        self.attDim = attDim
+        self.max_len_labels = max_len_labels
+
+        self.decoder = DecoderUnit(
+            sDim=sDim, xDim=in_channels, yDim=self.num_classes, attDim=attDim
+        )
+
+    def forward(self, x, embed):
+        x, targets, lengths = x
+        batch_size = x.shape[0]
+        # Decoder
+        state = self.decoder.get_initial_state(embed)
+        outputs = []
+        for i in range(max(lengths)):
+            if i == 0:
+                y_prev = paddle.full(shape=[batch_size], fill_value=self.num_classes)
+            else:
+                y_prev = targets[:, i - 1]
+            output, state = self.decoder(x, state, y_prev)
+            outputs.append(output)
+        outputs = paddle.concat([_.unsqueeze(1) for _ in outputs], 1)
+        return outputs
+
+    # inference stage.
+    def sample(self, x):
+        x, _, _ = x
+        batch_size = x.size(0)
+        # Decoder
+        state = paddle.zeros([1, batch_size, self.sDim])
+
+        predicted_ids, predicted_scores, predicted = [], [], None
+        for i in range(self.max_len_labels):
+            if i == 0:
+                y_prev = paddle.full(shape=[batch_size], fill_value=self.num_classes)
+            else:
+                y_prev = predicted
+
+            output, state = self.decoder(x, state, y_prev)
+            output = F.softmax(output, axis=1)
+            score, predicted = output.max(1)
+            predicted_ids.append(predicted.unsqueeze(1))
+            predicted_scores.append(score.unsqueeze(1))
+        predicted_ids = paddle.concat([predicted_ids, 1])
+        predicted_scores = paddle.concat([predicted_scores, 1])
+        # return predicted_ids.squeeze(), predicted_scores.squeeze()
+        return predicted_ids, predicted_scores
+
+    def beam_search(self, x, beam_width, eos, embed):
+        def _inflate(tensor, times, dim):
+            repeat_dims = [1] * tensor.dim()
+            repeat_dims[dim] = times
+            output = paddle.tile(tensor, repeat_dims)
+            return output
+
+        # https://github.com/IBM/pytorch-seq2seq/blob/fede87655ddce6c94b38886089e05321dc9802af/seq2seq/models/TopKDecoder.py
+        batch_size, l, d = x.shape
+        x = paddle.tile(
+            paddle.transpose(x.unsqueeze(1), perm=[1, 0, 2, 3]), [beam_width, 1, 1, 1]
+        )
+        inflated_encoder_feats = paddle.reshape(
+            paddle.transpose(x, perm=[1, 0, 2, 3]), [-1, l, d]
+        )
+
+        # Initialize the decoder
+        state = self.decoder.get_initial_state(embed, tile_times=beam_width)
+
+        pos_index = paddle.reshape(
+            paddle.arange(batch_size) * beam_width, shape=[-1, 1]
+        )
+
+        # Initialize the scores
+        sequence_scores = paddle.full(
+            shape=[batch_size * beam_width, 1], fill_value=-float("Inf")
+        )
+        index = [i * beam_width for i in range(0, batch_size)]
+        sequence_scores[index] = 0.0
+
+        # Initialize the input vector
+        y_prev = paddle.full(
+            shape=[batch_size * beam_width], fill_value=self.num_classes
+        )
+
+        # Store decisions for backtracking
+        stored_scores = list()
+        stored_predecessors = list()
+        stored_emitted_symbols = list()
+
+        for i in range(self.max_len_labels):
+            output, state = self.decoder(inflated_encoder_feats, state, y_prev)
+            state = paddle.unsqueeze(state, axis=0)
+            log_softmax_output = paddle.nn.functional.log_softmax(output, axis=1)
+
+            sequence_scores = _inflate(sequence_scores, self.num_classes, 1)
+            sequence_scores += log_softmax_output
+            scores, candidates = paddle.topk(
+                paddle.reshape(sequence_scores, [batch_size, -1]), beam_width, axis=1
+            )
+
+            # Reshape input = (bk, 1) and sequence_scores = (bk, 1)
+            y_prev = paddle.reshape(
+                candidates % self.num_classes, shape=[batch_size * beam_width]
+            )
+            sequence_scores = paddle.reshape(scores, shape=[batch_size * beam_width, 1])
+
+            # Update fields for next timestep
+            pos_index = paddle.expand_as(pos_index, candidates)
+            predecessors = paddle.cast(
+                candidates / self.num_classes + pos_index, dtype="int64"
+            )
+            predecessors = paddle.reshape(
+                predecessors, shape=[batch_size * beam_width, 1]
+            )
+            state = paddle.index_select(state, index=predecessors.squeeze(), axis=1)
+
+            # Update sequence scores and erase scores for <eos> symbol so that they aren't expanded
+            stored_scores.append(sequence_scores.clone())
+            y_prev = paddle.reshape(y_prev, shape=[-1, 1])
+            eos_prev = paddle.full_like(y_prev, fill_value=eos)
+            mask = eos_prev == y_prev
+            mask = paddle.nonzero(mask)
+            if mask.dim() > 0:
+                sequence_scores = sequence_scores.numpy()
+                mask = mask.numpy()
+                sequence_scores[mask] = -float("inf")
+                sequence_scores = paddle.to_tensor(sequence_scores)
+
+            # Cache results for backtracking
+            stored_predecessors.append(predecessors)
+            y_prev = paddle.squeeze(y_prev)
+            stored_emitted_symbols.append(y_prev)
+
+        # Do backtracking to return the optimal values
+        # ====== backtrak ======#
+        # Initialize return variables given different types
+        p = list()
+        l = [
+            [self.max_len_labels] * beam_width for _ in range(batch_size)
+        ]  # Placeholder for lengths of top-k sequences
+
+        # the last step output of the beams are not sorted
+        # thus they are sorted here
+        sorted_score, sorted_idx = paddle.topk(
+            paddle.reshape(stored_scores[-1], shape=[batch_size, beam_width]),
+            beam_width,
+        )
+
+        # initialize the sequence scores with the sorted last step beam scores
+        s = sorted_score.clone()
+
+        batch_eos_found = [0] * batch_size  # the number of EOS found
+        # in the backward loop below for each batch
+        t = self.max_len_labels - 1
+        # initialize the back pointer with the sorted order of the last step beams.
+        # add pos_index for indexing variable with b*k as the first dimension.
+        t_predecessors = paddle.reshape(
+            sorted_idx + pos_index.expand_as(sorted_idx),
+            shape=[batch_size * beam_width],
+        )
+        while t >= 0:
+            # Re-order the variables with the back pointer
+            current_symbol = paddle.index_select(
+                stored_emitted_symbols[t], index=t_predecessors, axis=0
+            )
+            t_predecessors = paddle.index_select(
+                stored_predecessors[t].squeeze(), index=t_predecessors, axis=0
+            )
+            eos_indices = stored_emitted_symbols[t] == eos
+            eos_indices = paddle.nonzero(eos_indices)
+
+            if eos_indices.dim() > 0:
+                for i in range(eos_indices.shape[0] - 1, -1, -1):
+                    # Indices of the EOS symbol for both variables
+                    # with b*k as the first dimension, and b, k for
+                    # the first two dimensions
+                    idx = eos_indices[i]
+                    b_idx = int(idx[0] / beam_width)
+                    # The indices of the replacing position
+                    # according to the replacement strategy noted above
+                    res_k_idx = beam_width - (batch_eos_found[b_idx] % beam_width) - 1
+                    batch_eos_found[b_idx] += 1
+                    res_idx = b_idx * beam_width + res_k_idx
+
+                    # Replace the old information in return variables
+                    # with the new ended sequence information
+                    t_predecessors[res_idx] = stored_predecessors[t][idx[0]]
+                    current_symbol[res_idx] = stored_emitted_symbols[t][idx[0]]
+                    s[b_idx, res_k_idx] = stored_scores[t][idx[0], 0]
+                    l[b_idx][res_k_idx] = t + 1
+
+            # record the back tracked results
+            p.append(current_symbol)
+            t -= 1
+
+        # Sort and re-order again as the added ended sequences may change
+        # the order (very unlikely)
+        s, re_sorted_idx = s.topk(beam_width)
+        for b_idx in range(batch_size):
+            l[b_idx] = [l[b_idx][k_idx.item()] for k_idx in re_sorted_idx[b_idx, :]]
+
+        re_sorted_idx = paddle.reshape(
+            re_sorted_idx + pos_index.expand_as(re_sorted_idx),
+            [batch_size * beam_width],
+        )
+
+        # Reverse the sequences and re-order at the same time
+        # It is reversed because the backtracking happens in reverse time order
+        p = [
+            paddle.reshape(
+                paddle.index_select(step, re_sorted_idx, 0),
+                shape=[batch_size, beam_width, -1],
+            )
+            for step in reversed(p)
+        ]
+        p = paddle.concat(p, -1)[:, 0, :]
+        return p, paddle.ones_like(p)
+
+
+class AttentionUnit(nn.Layer):
+    def __init__(self, sDim, xDim, attDim):
+        super(AttentionUnit, self).__init__()
+
+        self.sDim = sDim
+        self.xDim = xDim
+        self.attDim = attDim
+
+        self.sEmbed = nn.Linear(sDim, attDim)
+        self.xEmbed = nn.Linear(xDim, attDim)
+        self.wEmbed = nn.Linear(attDim, 1)
+
+    def forward(self, x, sPrev):
+        batch_size, T, _ = x.shape  # [b x T x xDim]
+        x = paddle.reshape(x, [-1, self.xDim])  # [(b x T) x xDim]
+        xProj = self.xEmbed(x)  # [(b x T) x attDim]
+        xProj = paddle.reshape(xProj, [batch_size, T, -1])  # [b x T x attDim]
+
+        sPrev = sPrev.squeeze(0)
+        sProj = self.sEmbed(sPrev)  # [b x attDim]
+        sProj = paddle.unsqueeze(sProj, 1)  # [b x 1 x attDim]
+        sProj = paddle.expand(sProj, [batch_size, T, self.attDim])  # [b x T x attDim]
+
+        sumTanh = paddle.tanh(sProj + xProj)
+        sumTanh = paddle.reshape(sumTanh, [-1, self.attDim])
+
+        vProj = self.wEmbed(sumTanh)  # [(b x T) x 1]
+        vProj = paddle.reshape(vProj, [batch_size, T])
+        alpha = F.softmax(
+            vProj, axis=1
+        )  # attention weights for each sample in the minibatch
+        return alpha
+
+
+class DecoderUnit(nn.Layer):
+    def __init__(self, sDim, xDim, yDim, attDim):
+        super(DecoderUnit, self).__init__()
+        self.sDim = sDim
+        self.xDim = xDim
+        self.yDim = yDim
+        self.attDim = attDim
+        self.emdDim = attDim
+
+        self.attention_unit = AttentionUnit(sDim, xDim, attDim)
+        self.tgt_embedding = nn.Embedding(
+            yDim + 1, self.emdDim, weight_attr=nn.initializer.Normal(std=0.01)
+        )  # the last is used for <BOS>
+        self.gru = nn.GRUCell(input_size=xDim + self.emdDim, hidden_size=sDim)
+        self.fc = nn.Linear(
+            sDim,
+            yDim,
+            weight_attr=nn.initializer.Normal(std=0.01),
+            bias_attr=nn.initializer.Constant(value=0),
+        )
+        self.embed_fc = nn.Linear(300, self.sDim)
+
+    def get_initial_state(self, embed, tile_times=1):
+        assert embed.shape[1] == 300
+        state = self.embed_fc(embed)  # N * sDim
+        if tile_times != 1:
+            state = state.unsqueeze(1)
+            trans_state = paddle.transpose(state, perm=[1, 0, 2])
+            state = paddle.tile(trans_state, repeat_times=[tile_times, 1, 1])
+            trans_state = paddle.transpose(state, perm=[1, 0, 2])
+            state = paddle.reshape(trans_state, shape=[-1, self.sDim])
+        state = state.unsqueeze(0)  # 1 * N * sDim
+        return state
+
+    def forward(self, x, sPrev, yPrev):
+        # x: feature sequence from the image decoder.
+        batch_size, T, _ = x.shape
+        alpha = self.attention_unit(x, sPrev)
+        context = paddle.squeeze(paddle.matmul(alpha.unsqueeze(1), x), axis=1)
+        yPrev = paddle.cast(yPrev, dtype="int64")
+        yProj = self.tgt_embedding(yPrev)
+
+        concat_context = paddle.concat([yProj, context], 1)
+        concat_context = paddle.squeeze(concat_context, 1)
+        sPrev = paddle.squeeze(sPrev, 0)
+        output, state = self.gru(concat_context, sPrev)
+        output = paddle.squeeze(output, axis=1)
+        output = self.fc(output)
+        return output, state
diff --git a/ppocr/modeling/heads/rec_att_head.py b/ppocr/modeling/heads/rec_att_head.py
new file mode 100644
index 0000000..2c952ce
--- /dev/null
+++ b/ppocr/modeling/heads/rec_att_head.py
@@ -0,0 +1,215 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+import numpy as np
+
+
+class AttentionHead(nn.Layer):
+    def __init__(self, in_channels, out_channels, hidden_size, **kwargs):
+        super(AttentionHead, self).__init__()
+        self.input_size = in_channels
+        self.hidden_size = hidden_size
+        self.num_classes = out_channels
+
+        self.attention_cell = AttentionGRUCell(
+            in_channels, hidden_size, out_channels, use_gru=False
+        )
+        self.generator = nn.Linear(hidden_size, out_channels)
+
+    def _char_to_onehot(self, input_char, onehot_dim):
+        input_ont_hot = F.one_hot(input_char, onehot_dim)
+        return input_ont_hot
+
+    def forward(self, inputs, targets=None, batch_max_length=25):
+        batch_size = inputs.shape[0]
+        num_steps = batch_max_length
+
+        hidden = paddle.zeros((batch_size, self.hidden_size))
+        output_hiddens = []
+
+        if targets is not None:
+            for i in range(num_steps):
+                char_onehots = self._char_to_onehot(
+                    targets[:, i], onehot_dim=self.num_classes
+                )
+                (outputs, hidden), alpha = self.attention_cell(
+                    hidden, inputs, char_onehots
+                )
+                output_hiddens.append(paddle.unsqueeze(outputs, axis=1))
+            output = paddle.concat(output_hiddens, axis=1)
+            probs = self.generator(output)
+        else:
+            targets = paddle.zeros(shape=[batch_size], dtype="int32")
+            probs = None
+            char_onehots = None
+            outputs = None
+            alpha = None
+
+            for i in range(num_steps):
+                char_onehots = self._char_to_onehot(
+                    targets, onehot_dim=self.num_classes
+                )
+                (outputs, hidden), alpha = self.attention_cell(
+                    hidden, inputs, char_onehots
+                )
+                probs_step = self.generator(outputs)
+                if probs is None:
+                    probs = paddle.unsqueeze(probs_step, axis=1)
+                else:
+                    probs = paddle.concat(
+                        [probs, paddle.unsqueeze(probs_step, axis=1)], axis=1
+                    )
+                next_input = probs_step.argmax(axis=1)
+                targets = next_input
+        if not self.training:
+            probs = paddle.nn.functional.softmax(probs, axis=2)
+        return probs
+
+
+class AttentionGRUCell(nn.Layer):
+    def __init__(self, input_size, hidden_size, num_embeddings, use_gru=False):
+        super(AttentionGRUCell, self).__init__()
+        self.i2h = nn.Linear(input_size, hidden_size, bias_attr=False)
+        self.h2h = nn.Linear(hidden_size, hidden_size)
+        self.score = nn.Linear(hidden_size, 1, bias_attr=False)
+
+        self.rnn = nn.GRUCell(
+            input_size=input_size + num_embeddings, hidden_size=hidden_size
+        )
+
+        self.hidden_size = hidden_size
+
+    def forward(self, prev_hidden, batch_H, char_onehots):
+        batch_H_proj = self.i2h(batch_H)
+        prev_hidden_proj = paddle.unsqueeze(self.h2h(prev_hidden), axis=1)
+
+        res = paddle.add(batch_H_proj, prev_hidden_proj)
+        res = paddle.tanh(res)
+        e = self.score(res)
+
+        alpha = F.softmax(e, axis=1)
+        alpha = paddle.transpose(alpha, [0, 2, 1])
+        context = paddle.squeeze(paddle.mm(alpha, batch_H), axis=1)
+        concat_context = paddle.concat([context, char_onehots], 1)
+
+        cur_hidden = self.rnn(concat_context, prev_hidden)
+
+        return cur_hidden, alpha
+
+
+class AttentionLSTM(nn.Layer):
+    def __init__(self, in_channels, out_channels, hidden_size, **kwargs):
+        super(AttentionLSTM, self).__init__()
+        self.input_size = in_channels
+        self.hidden_size = hidden_size
+        self.num_classes = out_channels
+
+        self.attention_cell = AttentionLSTMCell(
+            in_channels, hidden_size, out_channels, use_gru=False
+        )
+        self.generator = nn.Linear(hidden_size, out_channels)
+
+    def _char_to_onehot(self, input_char, onehot_dim):
+        input_ont_hot = F.one_hot(input_char, onehot_dim)
+        return input_ont_hot
+
+    def forward(self, inputs, targets=None, batch_max_length=25):
+        batch_size = inputs.shape[0]
+        num_steps = batch_max_length
+
+        hidden = (
+            paddle.zeros((batch_size, self.hidden_size)),
+            paddle.zeros((batch_size, self.hidden_size)),
+        )
+        output_hiddens = []
+
+        if targets is not None:
+            for i in range(num_steps):
+                # one-hot vectors for a i-th char
+                char_onehots = self._char_to_onehot(
+                    targets[:, i], onehot_dim=self.num_classes
+                )
+                hidden, alpha = self.attention_cell(hidden, inputs, char_onehots)
+
+                hidden = (hidden[1][0], hidden[1][1])
+                output_hiddens.append(paddle.unsqueeze(hidden[0], axis=1))
+            output = paddle.concat(output_hiddens, axis=1)
+            probs = self.generator(output)
+
+        else:
+            targets = paddle.zeros(shape=[batch_size], dtype="int32")
+            probs = None
+            char_onehots = None
+            alpha = None
+
+            for i in range(num_steps):
+                char_onehots = self._char_to_onehot(
+                    targets, onehot_dim=self.num_classes
+                )
+                hidden, alpha = self.attention_cell(hidden, inputs, char_onehots)
+                probs_step = self.generator(hidden[0])
+                hidden = (hidden[1][0], hidden[1][1])
+                if probs is None:
+                    probs = paddle.unsqueeze(probs_step, axis=1)
+                else:
+                    probs = paddle.concat(
+                        [probs, paddle.unsqueeze(probs_step, axis=1)], axis=1
+                    )
+
+                next_input = probs_step.argmax(axis=1)
+
+                targets = next_input
+        if not self.training:
+            probs = paddle.nn.functional.softmax(probs, axis=2)
+        return probs
+
+
+class AttentionLSTMCell(nn.Layer):
+    def __init__(self, input_size, hidden_size, num_embeddings, use_gru=False):
+        super(AttentionLSTMCell, self).__init__()
+        self.i2h = nn.Linear(input_size, hidden_size, bias_attr=False)
+        self.h2h = nn.Linear(hidden_size, hidden_size)
+        self.score = nn.Linear(hidden_size, 1, bias_attr=False)
+        if not use_gru:
+            self.rnn = nn.LSTMCell(
+                input_size=input_size + num_embeddings, hidden_size=hidden_size
+            )
+        else:
+            self.rnn = nn.GRUCell(
+                input_size=input_size + num_embeddings, hidden_size=hidden_size
+            )
+
+        self.hidden_size = hidden_size
+
+    def forward(self, prev_hidden, batch_H, char_onehots):
+        batch_H_proj = self.i2h(batch_H)
+        prev_hidden_proj = paddle.unsqueeze(self.h2h(prev_hidden[0]), axis=1)
+        res = paddle.add(batch_H_proj, prev_hidden_proj)
+        res = paddle.tanh(res)
+        e = self.score(res)
+
+        alpha = F.softmax(e, axis=1)
+        alpha = paddle.transpose(alpha, [0, 2, 1])
+        context = paddle.squeeze(paddle.mm(alpha, batch_H), axis=1)
+        concat_context = paddle.concat([context, char_onehots], 1)
+        cur_hidden = self.rnn(concat_context, prev_hidden)
+
+        return cur_hidden, alpha
diff --git a/ppocr/modeling/heads/rec_can_head.py b/ppocr/modeling/heads/rec_can_head.py
new file mode 100644
index 0000000..e80951c
--- /dev/null
+++ b/ppocr/modeling/heads/rec_can_head.py
@@ -0,0 +1,338 @@
+# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/LBH1024/CAN/models/can.py
+https://github.com/LBH1024/CAN/models/counting.py
+https://github.com/LBH1024/CAN/models/decoder.py
+https://github.com/LBH1024/CAN/models/attention.py
+
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle.nn as nn
+import paddle
+import math
+
+"""
+Counting Module
+"""
+
+
+class ChannelAtt(nn.Layer):
+    def __init__(self, channel, reduction):
+        super(ChannelAtt, self).__init__()
+        self.avg_pool = nn.AdaptiveAvgPool2D(1)
+
+        self.fc = nn.Sequential(
+            nn.Linear(channel, channel // reduction),
+            nn.ReLU(),
+            nn.Linear(channel // reduction, channel),
+            nn.Sigmoid(),
+        )
+
+    def forward(self, x):
+        b, c, _, _ = x.shape
+        y = paddle.reshape(self.avg_pool(x), [b, c])
+        y = paddle.reshape(self.fc(y), [b, c, 1, 1])
+        return x * y
+
+
+class CountingDecoder(nn.Layer):
+    def __init__(self, in_channel, out_channel, kernel_size):
+        super(CountingDecoder, self).__init__()
+        self.in_channel = in_channel
+        self.out_channel = out_channel
+
+        self.trans_layer = nn.Sequential(
+            nn.Conv2D(
+                self.in_channel,
+                512,
+                kernel_size=kernel_size,
+                padding=kernel_size // 2,
+                bias_attr=False,
+            ),
+            nn.BatchNorm2D(512),
+        )
+
+        self.channel_att = ChannelAtt(512, 16)
+
+        self.pred_layer = nn.Sequential(
+            nn.Conv2D(512, self.out_channel, kernel_size=1, bias_attr=False),
+            nn.Sigmoid(),
+        )
+
+    def forward(self, x, mask):
+        b, _, h, w = x.shape
+        x = self.trans_layer(x)
+        x = self.channel_att(x)
+        x = self.pred_layer(x)
+
+        if mask is not None:
+            x = x * mask
+        x = paddle.reshape(x, [b, self.out_channel, -1])
+        x1 = paddle.sum(x, axis=-1)
+
+        return x1, paddle.reshape(x, [b, self.out_channel, h, w])
+
+
+"""
+Attention Decoder
+"""
+
+
+class PositionEmbeddingSine(nn.Layer):
+    def __init__(
+        self, num_pos_feats=64, temperature=10000, normalize=False, scale=None
+    ):
+        super().__init__()
+        self.num_pos_feats = num_pos_feats
+        self.temperature = temperature
+        self.normalize = normalize
+        if scale is not None and normalize is False:
+            raise ValueError("normalize should be True if scale is passed")
+        if scale is None:
+            scale = 2 * math.pi
+        self.scale = scale
+
+    def forward(self, x, mask):
+        y_embed = paddle.cumsum(mask, 1, dtype="float32")
+        x_embed = paddle.cumsum(mask, 2, dtype="float32")
+
+        if self.normalize:
+            eps = 1e-6
+            y_embed = y_embed / (y_embed[:, -1:, :] + eps) * self.scale
+            x_embed = x_embed / (x_embed[:, :, -1:] + eps) * self.scale
+        dim_t = paddle.arange(self.num_pos_feats, dtype="float32")
+        dim_d = paddle.expand(paddle.to_tensor(2), dim_t.shape)
+        dim_t = self.temperature ** (
+            2 * (dim_t / dim_d).astype("int64") / self.num_pos_feats
+        )
+
+        pos_x = paddle.unsqueeze(x_embed, [3]) / dim_t
+        pos_y = paddle.unsqueeze(y_embed, [3]) / dim_t
+
+        pos_x = paddle.flatten(
+            paddle.stack(
+                [paddle.sin(pos_x[:, :, :, 0::2]), paddle.cos(pos_x[:, :, :, 1::2])],
+                axis=4,
+            ),
+            3,
+        )
+        pos_y = paddle.flatten(
+            paddle.stack(
+                [paddle.sin(pos_y[:, :, :, 0::2]), paddle.cos(pos_y[:, :, :, 1::2])],
+                axis=4,
+            ),
+            3,
+        )
+
+        pos = paddle.transpose(paddle.concat([pos_y, pos_x], axis=3), [0, 3, 1, 2])
+
+        return pos
+
+
+class AttDecoder(nn.Layer):
+    def __init__(
+        self,
+        ratio,
+        is_train,
+        input_size,
+        hidden_size,
+        encoder_out_channel,
+        dropout,
+        dropout_ratio,
+        word_num,
+        counting_decoder_out_channel,
+        attention,
+    ):
+        super(AttDecoder, self).__init__()
+        self.input_size = input_size
+        self.hidden_size = hidden_size
+        self.out_channel = encoder_out_channel
+        self.attention_dim = attention["attention_dim"]
+        self.dropout_prob = dropout
+        self.ratio = ratio
+        self.word_num = word_num
+
+        self.counting_num = counting_decoder_out_channel
+        self.is_train = is_train
+
+        self.init_weight = nn.Linear(self.out_channel, self.hidden_size)
+        self.embedding = nn.Embedding(self.word_num, self.input_size)
+        self.word_input_gru = nn.GRUCell(self.input_size, self.hidden_size)
+        self.word_attention = Attention(hidden_size, attention["attention_dim"])
+
+        self.encoder_feature_conv = nn.Conv2D(
+            self.out_channel,
+            self.attention_dim,
+            kernel_size=attention["word_conv_kernel"],
+            padding=attention["word_conv_kernel"] // 2,
+        )
+
+        self.word_state_weight = nn.Linear(self.hidden_size, self.hidden_size)
+        self.word_embedding_weight = nn.Linear(self.input_size, self.hidden_size)
+        self.word_context_weight = nn.Linear(self.out_channel, self.hidden_size)
+        self.counting_context_weight = nn.Linear(self.counting_num, self.hidden_size)
+        self.word_convert = nn.Linear(self.hidden_size, self.word_num)
+
+        if dropout:
+            self.dropout = nn.Dropout(dropout_ratio)
+
+    def forward(self, cnn_features, labels, counting_preds, images_mask):
+        if self.is_train:
+            _, num_steps = labels.shape
+        else:
+            num_steps = 36
+
+        batch_size, _, height, width = cnn_features.shape
+        images_mask = images_mask[:, :, :: self.ratio, :: self.ratio]
+
+        word_probs = paddle.zeros((batch_size, num_steps, self.word_num))
+        word_alpha_sum = paddle.zeros((batch_size, 1, height, width))
+
+        hidden = self.init_hidden(cnn_features, images_mask)
+        counting_context_weighted = self.counting_context_weight(counting_preds)
+        cnn_features_trans = self.encoder_feature_conv(cnn_features)
+
+        position_embedding = PositionEmbeddingSine(256, normalize=True)
+        pos = position_embedding(cnn_features_trans, images_mask[:, 0, :, :])
+
+        cnn_features_trans = cnn_features_trans + pos
+
+        word = paddle.ones([batch_size, 1], dtype="int64")  # init word as sos
+        word = word.squeeze(axis=1)
+        for i in range(num_steps):
+            word_embedding = self.embedding(word)
+            _, hidden = self.word_input_gru(word_embedding, hidden)
+            word_context_vec, _, word_alpha_sum = self.word_attention(
+                cnn_features, cnn_features_trans, hidden, word_alpha_sum, images_mask
+            )
+
+            current_state = self.word_state_weight(hidden)
+            word_weighted_embedding = self.word_embedding_weight(word_embedding)
+            word_context_weighted = self.word_context_weight(word_context_vec)
+
+            if self.dropout_prob:
+                word_out_state = self.dropout(
+                    current_state
+                    + word_weighted_embedding
+                    + word_context_weighted
+                    + counting_context_weighted
+                )
+            else:
+                word_out_state = (
+                    current_state
+                    + word_weighted_embedding
+                    + word_context_weighted
+                    + counting_context_weighted
+                )
+
+            word_prob = self.word_convert(word_out_state)
+            word_probs[:, i] = word_prob
+
+            if self.is_train:
+                word = labels[:, i]
+            else:
+                word = word_prob.argmax(1)
+                word = paddle.multiply(
+                    word, labels[:, i]
+                )  # labels are oneslike tensor in infer/predict mode
+
+        return word_probs
+
+    def init_hidden(self, features, feature_mask):
+        average = paddle.sum(
+            paddle.sum(features * feature_mask, axis=-1), axis=-1
+        ) / paddle.sum((paddle.sum(feature_mask, axis=-1)), axis=-1)
+        average = self.init_weight(average)
+        return paddle.tanh(average)
+
+
+"""
+Attention Module
+"""
+
+
+class Attention(nn.Layer):
+    def __init__(self, hidden_size, attention_dim):
+        super(Attention, self).__init__()
+        self.hidden = hidden_size
+        self.attention_dim = attention_dim
+        self.hidden_weight = nn.Linear(self.hidden, self.attention_dim)
+        self.attention_conv = nn.Conv2D(
+            1, 512, kernel_size=11, padding=5, bias_attr=False
+        )
+        self.attention_weight = nn.Linear(512, self.attention_dim, bias_attr=False)
+        self.alpha_convert = nn.Linear(self.attention_dim, 1)
+
+    def forward(
+        self, cnn_features, cnn_features_trans, hidden, alpha_sum, image_mask=None
+    ):
+        query = self.hidden_weight(hidden)
+        alpha_sum_trans = self.attention_conv(alpha_sum)
+        coverage_alpha = self.attention_weight(
+            paddle.transpose(alpha_sum_trans, [0, 2, 3, 1])
+        )
+        alpha_score = paddle.tanh(
+            paddle.unsqueeze(query, [1, 2])
+            + coverage_alpha
+            + paddle.transpose(cnn_features_trans, [0, 2, 3, 1])
+        )
+        energy = self.alpha_convert(alpha_score)
+        energy = energy - energy.max()
+        energy_exp = paddle.exp(paddle.squeeze(energy, -1))
+
+        if image_mask is not None:
+            energy_exp = energy_exp * paddle.squeeze(image_mask, 1)
+        alpha = energy_exp / (
+            paddle.unsqueeze(paddle.sum(paddle.sum(energy_exp, -1), -1), [1, 2]) + 1e-10
+        )
+        alpha_sum = paddle.unsqueeze(alpha, 1) + alpha_sum
+        context_vector = paddle.sum(
+            paddle.sum((paddle.unsqueeze(alpha, 1) * cnn_features), -1), -1
+        )
+
+        return context_vector, alpha, alpha_sum
+
+
+class CANHead(nn.Layer):
+    def __init__(self, in_channel, out_channel, ratio, attdecoder, **kwargs):
+        super(CANHead, self).__init__()
+
+        self.in_channel = in_channel
+        self.out_channel = out_channel
+
+        self.counting_decoder1 = CountingDecoder(
+            self.in_channel, self.out_channel, 3
+        )  # mscm
+        self.counting_decoder2 = CountingDecoder(self.in_channel, self.out_channel, 5)
+
+        self.decoder = AttDecoder(ratio, **attdecoder)
+
+        self.ratio = ratio
+
+    def forward(self, inputs, targets=None):
+        cnn_features, images_mask, labels = inputs
+
+        counting_mask = images_mask[:, :, :: self.ratio, :: self.ratio]
+        counting_preds1, _ = self.counting_decoder1(cnn_features, counting_mask)
+        counting_preds2, _ = self.counting_decoder2(cnn_features, counting_mask)
+        counting_preds = (counting_preds1 + counting_preds2) / 2
+
+        word_probs = self.decoder(cnn_features, labels, counting_preds, images_mask)
+        return word_probs, counting_preds, counting_preds1, counting_preds2
diff --git a/ppocr/modeling/heads/rec_cppd_head.py b/ppocr/modeling/heads/rec_cppd_head.py
new file mode 100644
index 0000000..c29946b
--- /dev/null
+++ b/ppocr/modeling/heads/rec_cppd_head.py
@@ -0,0 +1,387 @@
+# copyright (c) 2023 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+try:
+    from collections import Callable
+except:
+    from collections.abc import Callable
+
+import numpy as np
+import paddle
+from paddle import nn
+from paddle.nn import functional as F
+from ppocr.modeling.heads.rec_nrtr_head import Embeddings
+from ppocr.modeling.backbones.rec_svtrnet import (
+    DropPath,
+    Identity,
+    trunc_normal_,
+    zeros_,
+    ones_,
+    Mlp,
+)
+
+
+class Attention(nn.Layer):
+    def __init__(
+        self,
+        dim,
+        num_heads=8,
+        qkv_bias=False,
+        qk_scale=None,
+        attn_drop=0.0,
+        proj_drop=0.0,
+    ):
+        super().__init__()
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        self.scale = qk_scale or head_dim**-0.5
+
+        self.q = nn.Linear(dim, dim, bias_attr=qkv_bias)
+        self.kv = nn.Linear(dim, dim * 2, bias_attr=qkv_bias)
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.proj = nn.Linear(dim, dim)
+        self.proj_drop = nn.Dropout(proj_drop)
+
+    def forward(self, q, kv):
+        N, C = kv.shape[1:]
+        QN = q.shape[1]
+        q = (
+            self.q(q)
+            .reshape([-1, QN, self.num_heads, C // self.num_heads])
+            .transpose([0, 2, 1, 3])
+        )
+        k, v = (
+            self.kv(kv)
+            .reshape([-1, N, 2, self.num_heads, C // self.num_heads])
+            .transpose((2, 0, 3, 1, 4))
+        )
+        attn = q.matmul(k.transpose((0, 1, 3, 2))) * self.scale
+        attn = F.softmax(attn, axis=-1)
+        attn = self.attn_drop(attn)
+        x = (attn.matmul(v)).transpose((0, 2, 1, 3)).reshape((-1, QN, C))
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+
+
+class EdgeDecoderLayer(nn.Layer):
+    def __init__(
+        self,
+        dim,
+        num_heads,
+        mlp_ratio=4.0,
+        qkv_bias=False,
+        qk_scale=None,
+        drop=0.0,
+        attn_drop=0.0,
+        drop_path=[0.0, 0.0],
+        act_layer=nn.GELU,
+        norm_layer="nn.LayerNorm",
+        epsilon=1e-6,
+    ):
+        super().__init__()
+
+        self.head_dim = dim // num_heads
+        self.scale = qk_scale or self.head_dim**-0.5
+
+        # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
+        self.drop_path1 = DropPath(drop_path[0]) if drop_path[0] > 0.0 else Identity()
+        self.norm1 = eval(norm_layer)(dim, epsilon=epsilon)
+        self.norm2 = eval(norm_layer)(dim, epsilon=epsilon)
+
+        self.p = nn.Linear(dim, dim)
+        self.cv = nn.Linear(dim, dim)
+        self.pv = nn.Linear(dim, dim)
+
+        self.dim = dim
+        self.num_heads = num_heads
+        self.p_proj = nn.Linear(dim, dim)
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp_ratio = mlp_ratio
+        self.mlp = Mlp(
+            in_features=dim,
+            hidden_features=mlp_hidden_dim,
+            act_layer=act_layer,
+            drop=drop,
+        )
+
+    def forward(self, p, cv, pv):
+        pN = p.shape[1]
+        vN = cv.shape[1]
+        p_shortcut = p
+
+        p1 = (
+            self.p(p)
+            .reshape([-1, pN, self.num_heads, self.dim // self.num_heads])
+            .transpose([0, 2, 1, 3])
+        )
+        cv1 = (
+            self.cv(cv)
+            .reshape([-1, vN, self.num_heads, self.dim // self.num_heads])
+            .transpose([0, 2, 1, 3])
+        )
+        pv1 = (
+            self.pv(pv)
+            .reshape([-1, vN, self.num_heads, self.dim // self.num_heads])
+            .transpose([0, 2, 1, 3])
+        )
+
+        edge = F.softmax(p1.matmul(pv1.transpose((0, 1, 3, 2))), -1)  # B h N N
+        p_c = (edge @ cv1).transpose((0, 2, 1, 3)).reshape((-1, pN, self.dim))
+
+        x1 = self.norm1(p_shortcut + self.drop_path1(self.p_proj(p_c)))
+
+        x = self.norm2(x1 + self.drop_path1(self.mlp(x1)))
+        return x
+
+
+class DecoderLayer(nn.Layer):
+    def __init__(
+        self,
+        dim,
+        num_heads,
+        mlp_ratio=4.0,
+        qkv_bias=False,
+        qk_scale=None,
+        drop=0.0,
+        attn_drop=0.0,
+        drop_path=0.0,
+        act_layer=nn.GELU,
+        norm_layer="nn.LayerNorm",
+        epsilon=1e-6,
+    ):
+        super().__init__()
+        if isinstance(norm_layer, str):
+            self.norm1 = eval(norm_layer)(dim, epsilon=epsilon)
+            self.normkv = eval(norm_layer)(dim, epsilon=epsilon)
+        elif isinstance(norm_layer, Callable):
+            self.norm1 = norm_layer(dim)
+            self.normkv = norm_layer(dim)
+        else:
+            raise TypeError("The norm_layer must be str or paddle.nn.LayerNorm class")
+        self.mixer = Attention(
+            dim,
+            num_heads=num_heads,
+            qkv_bias=qkv_bias,
+            qk_scale=qk_scale,
+            attn_drop=attn_drop,
+            proj_drop=drop,
+        )
+
+        # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
+        self.drop_path = DropPath(drop_path) if drop_path > 0.0 else Identity()
+        if isinstance(norm_layer, str):
+            self.norm2 = eval(norm_layer)(dim, epsilon=epsilon)
+        elif isinstance(norm_layer, Callable):
+            self.norm2 = norm_layer(dim)
+        else:
+            raise TypeError("The norm_layer must be str or paddle.nn.layer.Layer class")
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp_ratio = mlp_ratio
+        self.mlp = Mlp(
+            in_features=dim,
+            hidden_features=mlp_hidden_dim,
+            act_layer=act_layer,
+            drop=drop,
+        )
+
+    def forward(self, q, kv):
+        x1 = self.norm1(q + self.drop_path(self.mixer(q, kv)))
+        x = self.norm2(x1 + self.drop_path(self.mlp(x1)))
+        return x
+
+
+class CPPDHead(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        dim,
+        out_channels,
+        num_layer=2,
+        drop_path_rate=0.1,
+        max_len=25,
+        vis_seq=50,
+        ch=False,
+        **kwargs,
+    ):
+        super(CPPDHead, self).__init__()
+
+        self.out_channels = out_channels  # none + 26 + 10
+        self.dim = dim
+        self.ch = ch
+        self.max_len = max_len + 1  # max_len + eos
+        self.char_node_embed = Embeddings(
+            d_model=dim, vocab=self.out_channels, scale_embedding=True
+        )
+        self.pos_node_embed = Embeddings(
+            d_model=dim, vocab=self.max_len, scale_embedding=True
+        )
+        dpr = np.linspace(0, drop_path_rate, num_layer + 1)
+
+        self.char_node_decoder = nn.LayerList(
+            [
+                DecoderLayer(
+                    dim=dim,
+                    num_heads=dim // 32,
+                    mlp_ratio=4.0,
+                    qkv_bias=True,
+                    drop_path=dpr[i],
+                )
+                for i in range(num_layer)
+            ]
+        )
+        self.pos_node_decoder = nn.LayerList(
+            [
+                DecoderLayer(
+                    dim=dim,
+                    num_heads=dim // 32,
+                    mlp_ratio=4.0,
+                    qkv_bias=True,
+                    drop_path=dpr[i],
+                )
+                for i in range(num_layer)
+            ]
+        )
+
+        self.edge_decoder = EdgeDecoderLayer(
+            dim=dim,
+            num_heads=dim // 32,
+            mlp_ratio=4.0,
+            qkv_bias=True,
+            drop_path=dpr[num_layer : num_layer + 1],
+        )
+
+        self.char_pos_embed = self.create_parameter(
+            shape=[1, self.max_len, dim], default_initializer=zeros_
+        )
+        self.add_parameter("char_pos_embed", self.char_pos_embed)
+        self.vis_pos_embed = self.create_parameter(
+            shape=[1, vis_seq, dim], default_initializer=zeros_
+        )
+        self.add_parameter("vis_pos_embed", self.vis_pos_embed)
+
+        self.char_node_fc1 = nn.Linear(dim, max_len)
+        self.pos_node_fc1 = nn.Linear(dim, self.max_len)
+
+        self.edge_fc = nn.Linear(dim, self.out_channels)
+        trunc_normal_(self.char_pos_embed)
+        trunc_normal_(self.vis_pos_embed)
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight)
+            if isinstance(m, nn.Linear) and m.bias is not None:
+                zeros_(m.bias)
+        elif isinstance(m, nn.LayerNorm):
+            zeros_(m.bias)
+            ones_(m.weight)
+
+    def forward(self, x, targets=None, epoch=0):
+        if self.training:
+            return self.forward_train(x, targets, epoch)
+        else:
+            return self.forward_test(x)
+
+    def forward_test(self, x):
+        visual_feats = x + self.vis_pos_embed
+        bs = visual_feats.shape[0]
+        pos_node_embed = (
+            self.pos_node_embed(paddle.arange(self.max_len)).unsqueeze(0)
+            + self.char_pos_embed
+        )
+        pos_node_embed = paddle.tile(pos_node_embed, [bs, 1, 1])
+        char_vis_node_query = visual_feats
+        pos_vis_node_query = paddle.concat([pos_node_embed, visual_feats], 1)
+
+        for char_decoder_layer, pos_decoder_layer in zip(
+            self.char_node_decoder, self.pos_node_decoder
+        ):
+            char_vis_node_query = char_decoder_layer(
+                char_vis_node_query, char_vis_node_query
+            )
+            pos_vis_node_query = pos_decoder_layer(
+                pos_vis_node_query, pos_vis_node_query[:, self.max_len :, :]
+            )
+        pos_node_query = pos_vis_node_query[:, : self.max_len, :]
+        char_vis_feats = char_vis_node_query
+
+        pos_node_feats = self.edge_decoder(
+            pos_node_query, char_vis_feats, char_vis_feats
+        )  # B, 26, dim
+        edge_feats = self.edge_fc(pos_node_feats)  # B, 26, 37
+        edge_logits = F.softmax(edge_feats, -1)
+
+        return edge_logits
+
+    def forward_train(self, x, targets=None, epoch=0):
+        visual_feats = x + self.vis_pos_embed
+        bs = visual_feats.shape[0]
+
+        if self.ch:
+            char_node_embed = self.char_node_embed(targets[-2])
+        else:
+            char_node_embed = self.char_node_embed(
+                paddle.arange(self.out_channels)
+            ).unsqueeze(0)
+            char_node_embed = paddle.tile(char_node_embed, [bs, 1, 1])
+        counting_char_num = char_node_embed.shape[1]
+        pos_node_embed = (
+            self.pos_node_embed(paddle.arange(self.max_len)).unsqueeze(0)
+            + self.char_pos_embed
+        )
+        pos_node_embed = paddle.tile(pos_node_embed, [bs, 1, 1])
+
+        node_feats = []
+
+        char_vis_node_query = paddle.concat([char_node_embed, visual_feats], 1)
+        pos_vis_node_query = paddle.concat([pos_node_embed, visual_feats], 1)
+
+        for char_decoder_layer, pos_decoder_layer in zip(
+            self.char_node_decoder, self.pos_node_decoder
+        ):
+            char_vis_node_query = char_decoder_layer(
+                char_vis_node_query, char_vis_node_query[:, counting_char_num:, :]
+            )
+            pos_vis_node_query = pos_decoder_layer(
+                pos_vis_node_query, pos_vis_node_query[:, self.max_len :, :]
+            )
+
+        char_node_query = char_vis_node_query[:, :counting_char_num, :]
+        pos_node_query = pos_vis_node_query[:, : self.max_len, :]
+
+        char_vis_feats = char_vis_node_query[:, counting_char_num:, :]
+        char_node_feats1 = self.char_node_fc1(char_node_query)
+
+        pos_node_feats1 = self.pos_node_fc1(pos_node_query)
+        diag_mask = (
+            paddle.eye(pos_node_feats1.shape[1])
+            .unsqueeze(0)
+            .tile([pos_node_feats1.shape[0], 1, 1])
+        )
+        pos_node_feats1 = (pos_node_feats1 * diag_mask).sum(-1)
+
+        node_feats.append(char_node_feats1)
+        node_feats.append(pos_node_feats1)
+
+        pos_node_feats = self.edge_decoder(
+            pos_node_query, char_vis_feats, char_vis_feats
+        )  # B, 26, dim
+        edge_feats = self.edge_fc(pos_node_feats)  # B, 26, 37
+
+        return node_feats, edge_feats
diff --git a/ppocr/modeling/heads/rec_ctc_head.py b/ppocr/modeling/heads/rec_ctc_head.py
new file mode 100755
index 0000000..565f167
--- /dev/null
+++ b/ppocr/modeling/heads/rec_ctc_head.py
@@ -0,0 +1,120 @@
+# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+
+import paddle
+from paddle import ParamAttr, nn
+from paddle.nn import functional as F
+
+
+def get_para_bias_attr(l2_decay, k):
+    regularizer = paddle.regularizer.L2Decay(l2_decay)
+    stdv = 1.0 / math.sqrt(k * 1.0)
+    initializer = nn.initializer.Uniform(-stdv, stdv)
+    weight_attr = ParamAttr(regularizer=regularizer, initializer=initializer)
+    bias_attr = ParamAttr(regularizer=regularizer, initializer=initializer)
+    return [weight_attr, bias_attr]
+
+
+class CTCHead(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        fc_decay=0.0004,
+        mid_channels=None,
+        return_feats=False,
+        use_guide=False,
+        **kwargs,
+    ):
+        super(CTCHead, self).__init__()
+        self.use_guide = use_guide
+        if use_guide:
+            self.guide_layer = nn.Sequential(
+                nn.Conv1D(
+                    in_channels,
+                    in_channels,
+                    kernel_size=5,
+                    padding=2,
+                    groups=in_channels,
+                    bias_attr=False,
+                ),
+                nn.BatchNorm1D(in_channels),
+                nn.Hardswish(),
+                nn.Conv1D(
+                    in_channels,
+                    in_channels,
+                    kernel_size=1,
+                    bias_attr=False,
+                ),
+                nn.BatchNorm1D(in_channels),
+                nn.Hardswish(),
+            )
+        if mid_channels is None:
+            weight_attr, bias_attr = get_para_bias_attr(
+                l2_decay=fc_decay, k=in_channels
+            )
+            self.fc = nn.Linear(
+                in_channels, out_channels, weight_attr=weight_attr, bias_attr=bias_attr
+            )
+        else:
+            weight_attr1, bias_attr1 = get_para_bias_attr(
+                l2_decay=fc_decay, k=in_channels
+            )
+            self.fc1 = nn.Linear(
+                in_channels,
+                mid_channels,
+                weight_attr=weight_attr1,
+                bias_attr=bias_attr1,
+            )
+
+            weight_attr2, bias_attr2 = get_para_bias_attr(
+                l2_decay=fc_decay, k=mid_channels
+            )
+            self.fc2 = nn.Linear(
+                mid_channels,
+                out_channels,
+                weight_attr=weight_attr2,
+                bias_attr=bias_attr2,
+            )
+        self.out_channels = out_channels
+        self.mid_channels = mid_channels
+        self.return_feats = return_feats
+
+    def forward(self, x, targets=None):
+        if self.use_guide:
+            x = x.transpose([0, 2, 1])
+            x = self.guide_layer(x)
+            x = x.transpose([0, 2, 1])
+
+        if self.mid_channels is None:
+            predicts = self.fc(x)
+        else:
+            x = self.fc1(x)
+            predicts = self.fc2(x)
+
+        if self.return_feats:
+            result = (x, predicts)
+        else:
+            result = predicts
+        if not self.training:
+            predicts = F.softmax(predicts, axis=2)
+            result = predicts
+
+        return result
diff --git a/ppocr/modeling/heads/rec_latexocr_head.py b/ppocr/modeling/heads/rec_latexocr_head.py
new file mode 100644
index 0000000..1b01d9d
--- /dev/null
+++ b/ppocr/modeling/heads/rec_latexocr_head.py
@@ -0,0 +1,1030 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+This code is refer from:
+https://github.com/lukas-blecher/LaTeX-OCR/blob/main/pix2tex/models/transformer.py
+"""
+
+import math
+import paddle
+from paddle import nn, einsum
+import paddle.nn.functional as F
+from functools import partial
+from inspect import isfunction
+from collections import namedtuple
+
+from paddle.nn.initializer import (
+    TruncatedNormal,
+    Constant,
+    Normal,
+    KaimingUniform,
+    XavierUniform,
+)
+
+zeros_ = Constant(value=0.0)
+ones_ = Constant(value=1.0)
+normal_ = Normal(std=0.02)
+DEFAULT_DIM_HEAD = 64
+
+Intermediates = namedtuple("Intermediates", ["pre_softmax_attn", "post_softmax_attn"])
+
+LayerIntermediates = namedtuple("Intermediates", ["hiddens", "attn_intermediates"])
+
+# helpers
+
+
+def exists(val):
+    return val is not None
+
+
+def default(val, d):
+    if exists(val):
+        return val
+    return d() if isfunction(d) else d
+
+
+class always:
+    def __init__(self, val):
+        self.val = val
+
+    def __call__(self, *args, **kwargs):
+        return self.val
+
+
+class not_equals:
+    def __init__(self, val):
+        self.val = val
+
+    def __call__(self, x, *args, **kwargs):
+        return x != self.val
+
+
+class equals:
+    def __init__(self, val):
+        self.val = val
+
+    def __call__(self, x, *args, **kwargs):
+        return x == self.val
+
+
+def max_neg_value(tensor):
+    return -paddle.finfo(tensor.dtype).max
+
+
+def pick_and_pop(keys, d):
+    values = list(map(lambda key: d.pop(key), keys))
+    return dict(zip(keys, values))
+
+
+def group_dict_by_key(cond, d):
+    return_val = [dict(), dict()]
+    for key in d.keys():
+        match = bool(cond(key))
+        ind = int(not match)
+        return_val[ind][key] = d[key]
+    return (*return_val,)
+
+
+def string_begins_with(prefix, str):
+    return str.startswith(prefix)
+
+
+def group_by_key_prefix(prefix, d):
+    return group_dict_by_key(partial(string_begins_with, prefix), d)
+
+
+def groupby_prefix_and_trim(prefix, d):
+    kwargs_with_prefix, kwargs = group_dict_by_key(
+        partial(string_begins_with, prefix), d
+    )
+    kwargs_without_prefix = dict(
+        map(lambda x: (x[0][len(prefix) :], x[1]), tuple(kwargs_with_prefix.items()))
+    )
+    return kwargs_without_prefix, kwargs
+
+
+# positional embeddings
+
+
+class DepthWiseConv1d(nn.Layer):
+    def __init__(
+        self, dim_in, dim_out, kernel_size, padding=0, stride=1, bias=True, groups=False
+    ):
+        super().__init__()
+        groups = default(groups, dim_in)
+        self.net = nn.Sequential(
+            nn.Conv1D(
+                dim_in,
+                dim_in,
+                kernel_size=kernel_size,
+                padding=padding,
+                groups=dim_in,
+                stride=stride,
+                bias_attr=bias,
+            ),
+            nn.Conv1D(dim_in, dim_out, 1),
+        )
+
+    def forward(self, x):
+        return self.net(x)
+
+
+class AbsolutePositionalEmbedding(nn.Layer):
+    def __init__(self, dim, max_seq_len):
+        super().__init__()
+        self.emb = nn.Embedding(max_seq_len, dim)
+        self.init_()
+
+    def init_(self):
+
+        normal_(self.emb.weight)
+
+    def forward(self, x):
+        n = paddle.arange(x.shape[1])
+        return self.emb(n)[None, :, :]
+
+
+class FixedPositionalEmbedding(nn.Layer):
+    def __init__(self, dim):
+        super().__init__()
+        inv_freq = 1.0 / (10000 ** (paddle.arange(0, dim, 2).float() / dim))
+        self.register_buffer("inv_freq", inv_freq)
+
+    def forward(self, x, seq_dim=1, offset=0):
+        t = (
+            paddle.arange(
+                x.shape[seq_dim],
+            ).type_as(self.inv_freq)
+            + offset
+        )
+        sinusoid_inp = paddle.einsum("i , j -> i j", t, self.inv_freq)
+        emb = paddle.concat((sinusoid_inp.sin(), sinusoid_inp.cos()), axis=-1)
+        return emb[None, :, :]
+
+
+class Scale(nn.Layer):
+    def __init__(self, value, fn):
+        super().__init__()
+        self.value = value
+        self.fn = fn
+
+    def forward(self, x, **kwargs):
+        x, *rest = self.fn(x, **kwargs)
+        return (x * self.value, *rest)
+
+
+class Rezero(nn.Layer):
+    def __init__(self, fn):
+        super().__init__()
+        self.fn = fn
+        self.g = paddle.create_parameter([1], dtype="float32")
+        zeros_(self.g)
+
+    def forward(self, x, **kwargs):
+        x, *rest = self.fn(x, **kwargs)
+        return (x * self.g, *rest)
+
+
+class ScaleNorm(nn.Layer):
+    def __init__(self, dim, eps=1e-5):
+        super().__init__()
+        self.scale = dim**-0.5
+        self.eps = eps
+        self.g = paddle.create_parameter([1], dtype="float32")
+        ones_(self.g)
+
+    def forward(self, x):
+        norm = paddle.norm(x, axis=-1, keepdim=True) * self.scale
+        return x / norm.clamp(min=self.eps) * self.g
+
+
+class RMSNorm(nn.Layer):
+    def __init__(self, dim, eps=1e-8):
+        super().__init__()
+        self.scale = dim**-0.5
+        self.eps = eps
+        self.g = paddle.create_parameter([dim])
+        ones_(self.g)
+
+    def forward(self, x):
+        norm = paddle.norm(x, axis=-1, keepdim=True) * self.scale
+        return x / norm.clamp(min=self.eps) * self.g
+
+
+class Residual(nn.Layer):
+    def forward(self, x, residual):
+        return x + residual
+
+
+class GEGLU(nn.Layer):
+    def __init__(self, dim_in, dim_out):
+        super().__init__()
+        self.proj = nn.Linear(dim_in, dim_out * 2)
+
+    def forward(self, x):
+        x, gate = self.proj(x).chunk(2, axis=-1)
+        return x * F.gelu(gate)
+
+
+class FeedForward(nn.Layer):
+    def __init__(self, dim, dim_out=None, mult=4, glu=False, dropout=0.0):
+        super().__init__()
+        inner_dim = int(dim * mult)
+        dim_out = default(dim_out, dim)
+        project_in = (
+            nn.Sequential(nn.Linear(dim, inner_dim), nn.GELU())
+            if not glu
+            else GEGLU(dim, inner_dim)
+        )
+
+        self.net = nn.Sequential(
+            project_in, nn.Dropout(dropout), nn.Linear(inner_dim, dim_out)
+        )
+
+    def forward(self, x):
+        return self.net(x)
+
+
+class Attention(nn.Layer):
+    def __init__(
+        self,
+        dim,
+        dim_head=DEFAULT_DIM_HEAD,
+        heads=8,
+        causal=False,
+        mask=None,
+        talking_heads=False,
+        collab_heads=False,
+        collab_compression=0.3,
+        sparse_topk=None,
+        use_entmax15=False,
+        num_mem_kv=0,
+        dropout=0.0,
+        on_attn=False,
+        gate_values=False,
+        is_export=False,
+    ):
+        super().__init__()
+        self.scale = dim_head**-0.5
+        self.heads = heads
+        self.causal = causal
+        self.mask = mask
+        self.is_export = is_export
+
+        qk_dim = v_dim = dim_head * heads
+
+        # collaborative heads
+        self.collab_heads = collab_heads
+        if self.collab_heads:
+            qk_dim = int(collab_compression * qk_dim)
+            self.collab_mixing = nn.Parameter(paddle.randn(heads, qk_dim))
+
+        self.to_q = nn.Linear(dim, qk_dim, bias_attr=False)
+        self.to_k = nn.Linear(dim, qk_dim, bias_attr=False)
+        self.to_v = nn.Linear(dim, v_dim, bias_attr=False)
+
+        self.dropout = nn.Dropout(dropout)
+
+        # add GLU gating for aggregated values, from alphafold2
+        self.to_v_gate = None
+        if gate_values:
+            self.to_v_gate = nn.Linear(dim, v_dim)
+            zeros_(self.to_v_gate.weight)
+            ones_(self.to_v_gate.bias)
+
+        # talking heads
+        self.talking_heads = talking_heads
+        if talking_heads:
+            self.pre_softmax_proj = nn.Parameter(paddle.randn(heads, heads))
+            self.post_softmax_proj = nn.Parameter(paddle.randn(heads, heads))
+
+        # explicit topk sparse attention
+        self.sparse_topk = sparse_topk
+
+        self.attn_fn = F.softmax
+
+        # add memory key / values
+        self.num_mem_kv = num_mem_kv
+        if num_mem_kv > 0:
+            self.mem_k = nn.Parameter(paddle.randn(heads, num_mem_kv, dim_head))
+            self.mem_v = nn.Parameter(paddle.randn(heads, num_mem_kv, dim_head))
+
+        # attention on attention
+        self.attn_on_attn = on_attn
+        self.to_out = (
+            nn.Sequential(nn.Linear(v_dim, dim * 2), nn.GLU())
+            if on_attn
+            else nn.Linear(v_dim, dim)
+        )
+
+    def forward(
+        self,
+        x,
+        context=None,
+        mask=None,
+        context_mask=None,
+        rel_pos=None,
+        sinusoidal_emb=None,
+        rotary_pos_emb=None,
+        prev_attn=None,
+        mem=None,
+        seq_len=0,
+    ):
+        if not self.training:
+            self.is_export = True
+        b, n, _, h, talking_heads, collab_heads, has_context = (
+            *x.shape,
+            self.heads,
+            self.talking_heads,
+            self.collab_heads,
+            exists(context),
+        )
+        kv_input = default(context, x)
+
+        q_input = x
+        k_input = kv_input
+        v_input = kv_input
+
+        if exists(mem):
+            k_input = paddle.concat((mem, k_input), axis=-2)
+            v_input = paddle.concat((mem, v_input), axis=-2)
+
+        if exists(sinusoidal_emb):
+            # in shortformer, the query would start at a position offset depending on the past cached memory
+            offset = k_input.shape[-2] - q_input.shape[-2]
+            q_input = q_input + sinusoidal_emb(q_input, offset=offset)
+            k_input = k_input + sinusoidal_emb(k_input)
+        q = self.to_q(q_input)
+        k = self.to_k(k_input)
+        v = self.to_v(v_input)
+
+        def rearrange_q_k_v(x, h, is_export):
+            if is_export:
+                b, n, h_d = paddle.shape(x)
+            else:
+                b, n, h_d = x.shape
+            d = h_d // h
+            return x.reshape([b, n, h, d]).transpose([0, 2, 1, 3])
+
+        q, k, v = map(
+            lambda t: rearrange_q_k_v(t, h, is_export=self.is_export), (q, k, v)
+        )
+
+        input_mask = None
+        if any(map(exists, (mask, context_mask))):
+            q_mask = default(
+                mask,
+                lambda: paddle.ones(
+                    (b, n),
+                ).cast(paddle.bool),
+            )
+            k_mask = q_mask if not exists(context) else context_mask
+            k_mask = default(
+                k_mask, lambda: paddle.ones((b, k.shape[-2])).cast(paddle.bool)
+            )
+
+            q_mask = q_mask.reshape([q_mask.shape[0], 1, q_mask.shape[1], 1])
+            k_mask = k_mask.reshape([k_mask.shape[0], 1, 1, k_mask.shape[1]])
+            input_mask = q_mask * k_mask
+
+        if collab_heads:
+            k = k.expand(-1, h, -1, -1)
+        dots = einsum("b h i d, b h j d -> b h i j", q, k) * self.scale
+
+        mask_value = max_neg_value(dots)
+
+        if exists(prev_attn):
+            dots = dots + prev_attn
+
+        pre_softmax_attn = dots.clone()
+
+        if talking_heads:
+            dots = einsum(
+                "b h i j, h k -> b k i j", dots, self.pre_softmax_proj
+            ).contiguous()
+
+        if exists(rel_pos):
+            dots = rel_pos(dots)
+
+        input_mask = input_mask.cast(paddle.bool)
+        if exists(input_mask):
+
+            dots.masked_fill_(~input_mask, mask_value)
+            del input_mask
+
+        if self.causal:
+            i, j = dots.shape[-2:]
+            r = paddle.arange(i)
+            r_shape = r.shape[0]
+            mask = r.reshape([1, 1, r_shape, 1]) < r.reshape([1, 1, 1, r_shape])
+
+            if self.is_export:
+                pad_list = [
+                    paddle.to_tensor(0, dtype="int32"),
+                    paddle.to_tensor(0, dtype="int32"),
+                    paddle.to_tensor(j - i, dtype="int32"),
+                    paddle.to_tensor(0, dtype="int32"),
+                ]
+                mask = F.pad(
+                    mask.cast(paddle.int32),
+                    paddle.to_tensor(pad_list).cast(paddle.int32),
+                    value=False,
+                ).cast(paddle.bool)
+                dots = dots.masked_fill_(mask, mask_value)
+            else:
+                mask = F.pad(mask.cast(paddle.int32), (0, 0, j - i, 0), value=False)
+                dots.masked_fill_(mask, mask_value)
+            del mask
+        if exists(self.sparse_topk) and self.sparse_topk < dots.shape[-1]:
+            top, _ = dots.topk(self.sparse_topk, dim=-1)
+            vk = top[..., -1].unsqueeze(-1).expand_as(dots)
+            mask = dots < vk
+            dots.masked_fill_(mask, mask_value)
+            del mask
+
+        attn = self.attn_fn(dots, axis=-1)
+        post_softmax_attn = attn.clone()
+
+        attn = self.dropout(attn)
+
+        if talking_heads:
+            attn = einsum(
+                "b h i j, h k -> b k i j", attn, self.post_softmax_proj
+            ).contiguous()
+        out = einsum("b h i j, b h j d -> b h i d", attn, v)
+
+        b, h, n, d = out.shape
+        out = out.transpose([0, 2, 1, 3]).reshape([b, n, h * d])
+
+        if exists(self.to_v_gate):
+            gates = self.gate_v(x)
+            out = out * gates.sigmoid()
+
+        intermediates = Intermediates(
+            pre_softmax_attn=pre_softmax_attn, post_softmax_attn=post_softmax_attn
+        )
+
+        return self.to_out(out), intermediates
+
+
+class AttentionLayers(nn.Layer):
+    def __init__(
+        self,
+        dim,
+        depth,
+        heads=8,
+        causal=False,
+        cross_attend=False,
+        only_cross=False,
+        use_scalenorm=False,
+        use_rmsnorm=False,
+        use_rezero=False,
+        rel_pos_bias=False,
+        rel_pos_num_buckets=32,
+        rel_pos_max_distance=128,
+        position_infused_attn=False,
+        rotary_pos_emb=False,
+        rotary_emb_dim=None,
+        custom_layers=None,
+        sandwich_coef=None,
+        par_ratio=None,
+        residual_attn=False,
+        cross_residual_attn=False,
+        macaron=False,
+        pre_norm=True,
+        gate_residual=False,
+        is_export=False,
+        **kwargs,
+    ):
+        super().__init__()
+        ff_kwargs, kwargs = groupby_prefix_and_trim("ff_", kwargs)
+        attn_kwargs, _ = groupby_prefix_and_trim("attn_", kwargs)
+
+        dim_head = attn_kwargs.get("dim_head", DEFAULT_DIM_HEAD)
+
+        self.dim = dim
+        self.depth = depth
+        self.layers = nn.LayerList([])
+
+        self.has_pos_emb = position_infused_attn or rel_pos_bias or rotary_pos_emb
+        self.pia_pos_emb = (
+            FixedPositionalEmbedding(dim) if position_infused_attn else None
+        )
+
+        assert (
+            rel_pos_num_buckets <= rel_pos_max_distance
+        ), "number of relative position buckets must be less than the relative position max distance"
+
+        self.pre_norm = pre_norm
+
+        self.residual_attn = residual_attn
+        self.cross_residual_attn = cross_residual_attn
+        self.cross_attend = cross_attend
+        self.rel_pos = None
+
+        norm_class = ScaleNorm if use_scalenorm else nn.LayerNorm
+        norm_class = RMSNorm if use_rmsnorm else norm_class
+        norm_fn = partial(norm_class, dim)
+
+        norm_fn = nn.Identity if use_rezero else norm_fn
+        branch_fn = Rezero if use_rezero else None
+
+        if cross_attend and not only_cross:
+            default_block = ("a", "c", "f")
+        elif cross_attend and only_cross:
+            default_block = ("c", "f")
+        else:
+            default_block = ("a", "f")
+        if macaron:
+            default_block = ("f",) + default_block
+
+        if exists(custom_layers):
+            layer_types = custom_layers
+        elif exists(par_ratio):
+            par_depth = depth * len(default_block)
+            assert 1 < par_ratio <= par_depth, "par ratio out of range"
+            default_block = tuple(filter(not_equals("f"), default_block))
+            par_attn = par_depth // par_ratio
+            depth_cut = (
+                par_depth * 2 // 3
+            )  # 2 / 3 attention layer cutoff suggested by PAR paper
+            par_width = (depth_cut + depth_cut // par_attn) // par_attn
+            assert (
+                len(default_block) <= par_width
+            ), "default block is too large for par_ratio"
+            par_block = default_block + ("f",) * (par_width - len(default_block))
+            par_head = par_block * par_attn
+            layer_types = par_head + ("f",) * (par_depth - len(par_head))
+        elif exists(sandwich_coef):
+            assert (
+                sandwich_coef > 0 and sandwich_coef <= depth
+            ), "sandwich coefficient should be less than the depth"
+            layer_types = (
+                ("a",) * sandwich_coef
+                + default_block * (depth - sandwich_coef)
+                + ("f",) * sandwich_coef
+            )
+        else:
+            layer_types = default_block * depth
+
+        self.layer_types = layer_types
+        self.num_attn_layers = len(list(filter(equals("a"), layer_types)))
+        for layer_type in self.layer_types:
+            if layer_type == "a":
+                layer = Attention(
+                    dim, heads=heads, causal=causal, is_export=is_export, **attn_kwargs
+                )
+            elif layer_type == "c":
+                layer = Attention(dim, heads=heads, is_export=is_export, **attn_kwargs)
+            elif layer_type == "f":
+                layer = FeedForward(dim, **ff_kwargs)
+                layer = layer if not macaron else Scale(0.5, layer)
+            else:
+                raise Exception(f"invalid layer type {layer_type}")
+            if isinstance(layer, Attention) and exists(branch_fn):
+                layer = branch_fn(layer)
+            residual_fn = Residual()
+            self.layers.append(nn.LayerList([norm_fn(), layer, residual_fn]))
+
+    def forward(
+        self,
+        x,
+        context=None,
+        mask=None,
+        context_mask=None,
+        mems=None,
+        seq_len=0,
+        return_hiddens=False,
+    ):
+        assert not (
+            self.cross_attend ^ exists(context)
+        ), "context must be passed in if cross_attend is set to True"
+
+        hiddens = []
+        intermediates = []
+        prev_attn = None
+        prev_cross_attn = None
+        rotary_pos_emb = None
+
+        mems = mems.copy() if exists(mems) else [None] * self.num_attn_layers
+
+        for ind, (layer_type, (norm, block, residual_fn)) in enumerate(
+            zip(self.layer_types, self.layers)
+        ):
+            is_last = ind == (len(self.layers) - 1)
+
+            if layer_type == "a":
+                hiddens.append(x)
+                layer_mem = mems.pop(0)
+
+            residual = x
+
+            if self.pre_norm:
+                x = norm(x)
+
+            if layer_type == "a":
+                out, inter = block(
+                    x,
+                    mask=mask,
+                    sinusoidal_emb=self.pia_pos_emb,
+                    rel_pos=self.rel_pos,
+                    rotary_pos_emb=rotary_pos_emb,
+                    prev_attn=prev_attn,
+                    mem=layer_mem,
+                )
+            elif layer_type == "c":
+                out, inter = block(
+                    x,
+                    context=context,
+                    mask=mask,
+                    context_mask=context_mask,
+                    prev_attn=prev_cross_attn,
+                )
+            elif layer_type == "f":
+                out = block(x)
+
+            x = residual_fn(out, residual)
+
+            if layer_type in ("a", "c"):
+                intermediates.append(inter)
+
+            if layer_type == "a" and self.residual_attn:
+                prev_attn = inter.pre_softmax_attn
+            elif layer_type == "c" and self.cross_residual_attn:
+                prev_cross_attn = inter.pre_softmax_attn
+
+            if not self.pre_norm and not is_last:
+                x = norm(x)
+
+        if return_hiddens:
+            intermediates = LayerIntermediates(
+                hiddens=hiddens, attn_intermediates=intermediates
+            )
+
+            return x, intermediates
+
+        return x
+
+
+class Encoder(AttentionLayers):
+    def __init__(self, **kwargs):
+        assert "causal" not in kwargs, "cannot set causality on encoder"
+        super().__init__(causal=False, **kwargs)
+
+
+class Decoder(AttentionLayers):
+    def __init__(self, **kwargs):
+        assert "causal" not in kwargs, "cannot set causality on decoder"
+        super().__init__(causal=True, **kwargs)
+
+
+class CrossAttender(AttentionLayers):
+    def __init__(self, **kwargs):
+        super().__init__(cross_attend=True, only_cross=True, **kwargs)
+
+
+def create_latex_parameter(shape):
+    return paddle.create_parameter(
+        shape=shape,
+        dtype="float32",
+        default_initializer=paddle.nn.initializer.Assign(paddle.randn(shape)),
+    )
+
+
+class TransformerDecoder(nn.Layer):
+    def __init__(
+        self,
+        *,
+        num_tokens,
+        max_seq_len,
+        attn_layers,
+        emb_dim=None,
+        max_mem_len=0.0,
+        emb_dropout=0.0,
+        num_memory_tokens=None,
+        tie_embedding=False,
+        use_pos_emb=True,
+        is_export=False,
+    ):
+        super().__init__()
+        assert isinstance(
+            attn_layers, AttentionLayers
+        ), "attention layers must be one of Encoder or Decoder"
+
+        dim = attn_layers.dim
+        emb_dim = default(emb_dim, dim)
+
+        self.max_seq_len = max_seq_len
+        self.max_mem_len = max_mem_len
+
+        self.token_emb = nn.Embedding(num_tokens, emb_dim)
+        self.pos_emb = (
+            AbsolutePositionalEmbedding(emb_dim, max_seq_len)
+            if (use_pos_emb and not attn_layers.has_pos_emb)
+            else always(0)
+        )
+        self.emb_dropout = nn.Dropout(emb_dropout)
+
+        self.project_emb = nn.Linear(emb_dim, dim) if emb_dim != dim else nn.Identity()
+        self.attn_layers = attn_layers
+        self.norm = nn.LayerNorm(dim)
+        self.is_export = is_export
+
+        self.init_()
+
+        self.to_logits = (
+            nn.Linear(dim, num_tokens)
+            if not tie_embedding
+            else lambda t: t @ self.token_emb.weight.t()
+        )
+
+        # memory tokens (like [cls]) from Memory Transformers paper
+        num_memory_tokens = default(num_memory_tokens, 0)
+        self.num_memory_tokens = num_memory_tokens
+        if num_memory_tokens > 0:
+            self.memory_tokens = create_latex_parameter([num_memory_tokens, dim])
+
+            # let funnel encoder know number of memory tokens, if specified
+            # TODO: think of a cleaner solution
+            if hasattr(attn_layers, "num_memory_tokens"):
+                attn_layers.num_memory_tokens = num_memory_tokens
+
+    def init_(self):
+        normal_(self.token_emb.weight)
+
+    def forward(
+        self,
+        x,
+        return_embeddings=False,
+        mask=None,
+        return_mems=False,
+        return_attn=False,
+        seq_len=0,
+        mems=None,
+        **kwargs,
+    ):
+        b, n, num_mem = *x.shape, self.num_memory_tokens
+        x = self.token_emb(x)
+        x = x + self.pos_emb(x)
+
+        x = self.emb_dropout(x)
+        x = self.project_emb(x)
+
+        x, intermediates = self.attn_layers(
+            x, mask=mask, mems=mems, return_hiddens=True, seq_len=seq_len, **kwargs
+        )
+        x = self.norm(x)
+        if paddle.device.get_device().startswith("npu"):
+            x = x[:, num_mem:]
+        else:
+            mem, x = x[:, :num_mem], x[:, num_mem:]
+        out = self.to_logits(x) if not return_embeddings else x
+        if return_mems:
+            hiddens = intermediates.hiddens
+            new_mems = (
+                list(map(lambda pair: paddle.concat(pair, axis=-2), zip(mems, hiddens)))
+                if exists(mems)
+                else hiddens
+            )
+            new_mems = list(
+                map(lambda t: t[..., -self.max_mem_len :, :].detach(), new_mems)
+            )
+            return out, new_mems
+
+        if return_attn:
+            attn_maps = list(
+                map(lambda t: t.post_softmax_attn, intermediates.attn_intermediates)
+            )
+            return out, attn_maps
+
+        return out
+
+
+def top_p(logits, thres=0.9):
+    sorted_logits, sorted_indices = paddle.sort(logits, descending=True)
+    cum_probs = paddle.cumsum(F.softmax(sorted_logits, axis=-1), axis=-1)
+
+    sorted_indices_to_remove = cum_probs > (1 - thres)
+    sorted_indices_to_remove[:, 1:] = sorted_indices_to_remove[:, :-1].clone()
+    sorted_indices_to_remove[:, 0] = 0
+
+    sorted_logits[sorted_indices_to_remove] = float("-inf")
+    return sorted_logits.scatter(1, sorted_indices, sorted_logits)
+
+
+# topk
+
+
+def top_k(logits, thres=0.9):
+    k = int((1 - thres) * logits.shape[-1])
+    val, ind = paddle.topk(logits, k)
+    probs = paddle.full_like(logits, float("-inf"))
+    probs = paddle.put_along_axis(probs, ind, val, 1)
+    return probs
+
+
+class LaTeXOCRHead(nn.Layer):
+    """Implementation of LaTeX OCR decoder.
+
+    Args:
+      encoded_feat: The encoded features with shape[N, 1, H//16, W//16]
+      tgt_seq: LaTeX-OCR labels with shape [N, L] , L is the max sequence length
+      xi: The first N-1 LaTeX-OCR sequences in tgt_seq with shape [N, L-1]
+      mask: The first N-1 LaTeX-OCR attention mask with shape [N, L-1]  , L is the max sequence length
+
+    Returns:
+      The predicted LaTeX sequences with shape [N, L-1, C], C is the number of LaTeX classes
+    """
+
+    def __init__(
+        self,
+        net=None,
+        in_channels=256,
+        out_channels=256,
+        pad_value=0,
+        decoder_args=None,
+        is_export=False,
+    ):
+        super().__init__()
+        decoder = Decoder(
+            dim=256, depth=4, heads=8, is_export=is_export, **decoder_args
+        )
+        transformer_decoder = TransformerDecoder(
+            num_tokens=8000,
+            max_seq_len=512,
+            attn_layers=decoder,
+            is_export=is_export,
+        )
+        self.temperature = 0.333
+        self.bos_token = 1
+        self.eos_token = 2
+        self.max_length = 512
+        self.pad_value = pad_value
+
+        self.net = transformer_decoder
+        self.max_seq_len = self.net.max_seq_len
+        self.is_export = is_export
+
+    @paddle.no_grad()
+    def generate(
+        self,
+        start_tokens,
+        seq_len,
+        eos_token=None,
+        temperature=1.0,
+        filter_logits_fn=top_k,
+        filter_thres=0.9,
+        **kwargs,
+    ):
+        was_training = self.net.training
+        num_dims = len(start_tokens.shape)
+
+        if num_dims == 1:
+            start_tokens = start_tokens[None, :]
+
+        b, t = start_tokens.shape
+
+        self.net.eval()
+        out = start_tokens
+        mask = kwargs.pop("mask", None)
+
+        if mask is None:
+            mask = paddle.full_like(out, True, dtype=paddle.bool)
+
+        for _ in range(seq_len):
+            x = out[:, -self.max_seq_len :]
+            mask = mask[:, -self.max_seq_len :]
+            logits = self.net(x, mask=mask, **kwargs)[:, -1, :]
+            if filter_logits_fn in {top_k, top_p}:
+                filtered_logits = filter_logits_fn(logits, thres=filter_thres)
+
+                probs = F.softmax(filtered_logits / temperature, axis=-1)
+            else:
+                raise NotImplementedError("The filter_logits_fn is not supported ")
+
+            sample = paddle.multinomial(probs, 1)
+            out = paddle.concat((out, sample), axis=-1)
+            pad_mask = paddle.full(shape=[mask.shape[0], 1], fill_value=1, dtype="bool")
+            mask = paddle.concat((mask, pad_mask), axis=1)
+            if (
+                eos_token is not None
+                and (
+                    paddle.cumsum((out == eos_token).cast(paddle.int64), 1)[:, -1] >= 1
+                ).all()
+            ):
+                break
+        out = out[:, t:]
+        if num_dims == 1:
+            out = out.squeeze(0)
+        return out
+
+    @paddle.no_grad()
+    def generate_export(
+        self,
+        start_tokens,
+        seq_len,
+        eos_token=None,
+        context=None,
+        temperature=1.0,
+        filter_logits_fn=None,
+        filter_thres=0.9,
+        **kwargs,
+    ):
+        was_training = self.net.training
+        num_dims = len(start_tokens.shape)
+
+        if num_dims == 1:
+            start_tokens = start_tokens[None, :]
+
+        b, t = start_tokens.shape
+
+        self.net.eval()
+        out = start_tokens
+        mask = kwargs.pop("mask", None)
+
+        if mask is None:
+            mask = paddle.full_like(out, True, dtype=paddle.bool)
+
+        i_idx = paddle.full([], 0)
+        while i_idx < paddle.to_tensor(seq_len):
+            x = out[:, -self.max_seq_len :]
+            paddle.jit.api.set_dynamic_shape(x, [-1, -1])
+            mask = mask[:, -self.max_seq_len :]
+            paddle.jit.api.set_dynamic_shape(mask, [-1, -1])
+            logits = self.net(x, mask=mask, context=context, seq_len=i_idx, **kwargs)[
+                :, -1, :
+            ]
+            if filter_logits_fn in {top_k, top_p}:
+                filtered_logits = filter_logits_fn(logits, thres=filter_thres)
+
+                probs = F.softmax(filtered_logits / temperature, axis=-1)
+
+            sample = paddle.multinomial(probs, 1)
+            out = paddle.concat((out, sample), axis=-1)
+
+            pad_mask = paddle.full(shape=[mask.shape[0], 1], fill_value=1, dtype="bool")
+            mask = paddle.concat((mask, pad_mask), axis=1)
+            if (
+                eos_token is not None
+                and (
+                    paddle.cumsum((out == eos_token).cast(paddle.int64), 1)[:, -1] >= 1
+                ).all()
+            ):
+                break
+            i_idx += 1
+        out = out[:, t:]
+        if num_dims == 1:
+            out = out.squeeze(0)
+        return out
+
+    # forward for export
+    def forward(self, inputs, targets=None):
+        if not self.training:
+            self.is_export = True
+            encoded_feat = inputs
+            batch_num = encoded_feat.shape[0]
+            bos_tensor = paddle.full([batch_num, 1], self.bos_token, dtype=paddle.int64)
+            if self.is_export:
+                word_pred = self.generate_export(
+                    bos_tensor,
+                    self.max_seq_len,
+                    eos_token=self.eos_token,
+                    context=encoded_feat,
+                    temperature=self.temperature,
+                    filter_logits_fn=top_k,
+                )
+            else:
+                word_pred = self.generate(
+                    bos_tensor,
+                    self.max_seq_len,
+                    eos_token=self.eos_token,
+                    context=encoded_feat,
+                    temperature=self.temperature,
+                    filter_logits_fn=top_k,
+                )
+            return word_pred
+
+        encoded_feat, tgt_seq, mask = inputs
+        kwargs = {"context": encoded_feat, "mask": mask.cast(paddle.bool)}
+        x = tgt_seq
+        xi = x[:, :-1]
+
+        mask = kwargs.get("mask", None)
+        if mask is not None and mask.shape[1] == x.shape[1]:
+            mask = mask[:, :-1]
+            kwargs["mask"] = mask
+        out = self.net(xi, **kwargs)
+
+        return out
diff --git a/ppocr/modeling/heads/rec_multi_head.py b/ppocr/modeling/heads/rec_multi_head.py
new file mode 100644
index 0000000..be44615
--- /dev/null
+++ b/ppocr/modeling/heads/rec_multi_head.py
@@ -0,0 +1,153 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+from paddle import ParamAttr
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+from ppocr.modeling.necks.rnn import (
+    Im2Seq,
+    EncoderWithRNN,
+    EncoderWithFC,
+    SequenceEncoder,
+    EncoderWithSVTR,
+    trunc_normal_,
+    zeros_,
+)
+from .rec_ctc_head import CTCHead
+from .rec_sar_head import SARHead
+from .rec_nrtr_head import Transformer
+
+
+class FCTranspose(nn.Layer):
+    def __init__(self, in_channels, out_channels, only_transpose=False):
+        super().__init__()
+        self.only_transpose = only_transpose
+        if not self.only_transpose:
+            self.fc = nn.Linear(in_channels, out_channels, bias_attr=False)
+
+    def forward(self, x):
+        if self.only_transpose:
+            return x.transpose([0, 2, 1])
+        else:
+            return self.fc(x.transpose([0, 2, 1]))
+
+
+class AddPos(nn.Layer):
+    def __init__(self, dim, w):
+        super().__init__()
+        self.dec_pos_embed = self.create_parameter(
+            shape=[1, w, dim], default_initializer=zeros_
+        )
+        self.add_parameter("dec_pos_embed", self.dec_pos_embed)
+        trunc_normal_(self.dec_pos_embed)
+
+    def forward(self, x):
+        x = x + self.dec_pos_embed[:, : x.shape[1], :]
+        return x
+
+
+class MultiHead(nn.Layer):
+    def __init__(self, in_channels, out_channels_list, **kwargs):
+        super().__init__()
+        self.head_list = kwargs.pop("head_list")
+        self.use_pool = kwargs.get("use_pool", False)
+        self.use_pos = kwargs.get("use_pos", False)
+        self.in_channels = in_channels
+        if self.use_pool:
+            self.pool = nn.AvgPool2D(kernel_size=[3, 2], stride=[3, 2], padding=0)
+        self.gtc_head = "sar"
+        assert len(self.head_list) >= 2
+        for idx, head_name in enumerate(self.head_list):
+            name = list(head_name)[0]
+            if name == "SARHead":
+                # sar head
+                sar_args = self.head_list[idx][name]
+                self.sar_head = eval(name)(
+                    in_channels=in_channels,
+                    out_channels=out_channels_list["SARLabelDecode"],
+                    **sar_args,
+                )
+            elif name == "NRTRHead":
+                gtc_args = self.head_list[idx][name]
+                max_text_length = gtc_args.get("max_text_length", 25)
+                nrtr_dim = gtc_args.get("nrtr_dim", 256)
+                num_decoder_layers = gtc_args.get("num_decoder_layers", 4)
+                if self.use_pos:
+                    self.before_gtc = nn.Sequential(
+                        nn.Flatten(2),
+                        FCTranspose(in_channels, nrtr_dim),
+                        AddPos(nrtr_dim, 80),
+                    )
+                else:
+                    self.before_gtc = nn.Sequential(
+                        nn.Flatten(2), FCTranspose(in_channels, nrtr_dim)
+                    )
+
+                self.gtc_head = Transformer(
+                    d_model=nrtr_dim,
+                    nhead=nrtr_dim // 32,
+                    num_encoder_layers=-1,
+                    beam_size=-1,
+                    num_decoder_layers=num_decoder_layers,
+                    max_len=max_text_length,
+                    dim_feedforward=nrtr_dim * 4,
+                    out_channels=out_channels_list["NRTRLabelDecode"],
+                )
+            elif name == "CTCHead":
+                # ctc neck
+                self.encoder_reshape = Im2Seq(in_channels)
+                neck_args = self.head_list[idx][name]["Neck"]
+                encoder_type = neck_args.pop("name")
+                self.ctc_encoder = SequenceEncoder(
+                    in_channels=in_channels, encoder_type=encoder_type, **neck_args
+                )
+                # ctc head
+                head_args = self.head_list[idx][name]["Head"]
+                self.ctc_head = eval(name)(
+                    in_channels=self.ctc_encoder.out_channels,
+                    out_channels=out_channels_list["CTCLabelDecode"],
+                    **head_args,
+                )
+            else:
+                raise NotImplementedError(
+                    "{} is not supported in MultiHead yet".format(name)
+                )
+
+    def forward(self, x, targets=None):
+        if self.use_pool:
+            x = self.pool(
+                x.reshape([0, 3, -1, self.in_channels]).transpose([0, 3, 1, 2])
+            )
+        ctc_encoder = self.ctc_encoder(x)
+        ctc_out = self.ctc_head(ctc_encoder, targets)
+        head_out = dict()
+        head_out["ctc"] = ctc_out
+        head_out["ctc_neck"] = ctc_encoder
+        # eval mode
+        if not self.training:
+            return ctc_out
+        if self.gtc_head == "sar":
+            sar_out = self.sar_head(x, targets[1:])
+            head_out["sar"] = sar_out
+        else:
+            gtc_out = self.gtc_head(self.before_gtc(x), targets[1:])
+            head_out["gtc"] = gtc_out
+        return head_out
diff --git a/ppocr/modeling/heads/rec_nrtr_head.py b/ppocr/modeling/heads/rec_nrtr_head.py
new file mode 100644
index 0000000..34f911c
--- /dev/null
+++ b/ppocr/modeling/heads/rec_nrtr_head.py
@@ -0,0 +1,705 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import math
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle.nn import Dropout, LayerNorm
+import numpy as np
+from ppocr.modeling.backbones.rec_svtrnet import Mlp, zeros_
+from paddle.nn.initializer import XavierNormal as xavier_normal_
+
+
+class Transformer(nn.Layer):
+    """A transformer model. User is able to modify the attributes as needed. The architecture
+    is based on the paper "Attention Is All You Need". Ashish Vaswani, Noam Shazeer,
+    Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Lukasz Kaiser, and
+    Illia Polosukhin. 2017. Attention is all you need. In Advances in Neural Information
+    Processing Systems, pages 6000-6010.
+
+    Args:
+        d_model: the number of expected features in the encoder/decoder inputs (default=512).
+        nhead: the number of heads in the multiheadattention models (default=8).
+        num_encoder_layers: the number of sub-encoder-layers in the encoder (default=6).
+        num_decoder_layers: the number of sub-decoder-layers in the decoder (default=6).
+        dim_feedforward: the dimension of the feedforward network model (default=2048).
+        dropout: the dropout value (default=0.1).
+        custom_encoder: custom encoder (default=None).
+        custom_decoder: custom decoder (default=None).
+    """
+
+    def __init__(
+        self,
+        d_model=512,
+        nhead=8,
+        num_encoder_layers=6,
+        beam_size=0,
+        num_decoder_layers=6,
+        max_len=25,
+        dim_feedforward=1024,
+        attention_dropout_rate=0.0,
+        residual_dropout_rate=0.1,
+        in_channels=0,
+        out_channels=0,
+        scale_embedding=True,
+    ):
+        super(Transformer, self).__init__()
+        self.out_channels = out_channels + 1
+        self.max_len = max_len
+        self.embedding = Embeddings(
+            d_model=d_model,
+            vocab=self.out_channels,
+            padding_idx=0,
+            scale_embedding=scale_embedding,
+        )
+        self.positional_encoding = PositionalEncoding(
+            dropout=residual_dropout_rate, dim=d_model
+        )
+
+        if num_encoder_layers > 0:
+            self.encoder = nn.LayerList(
+                [
+                    TransformerBlock(
+                        d_model,
+                        nhead,
+                        dim_feedforward,
+                        attention_dropout_rate,
+                        residual_dropout_rate,
+                        with_self_attn=True,
+                        with_cross_attn=False,
+                    )
+                    for i in range(num_encoder_layers)
+                ]
+            )
+        else:
+            self.encoder = None
+
+        self.decoder = nn.LayerList(
+            [
+                TransformerBlock(
+                    d_model,
+                    nhead,
+                    dim_feedforward,
+                    attention_dropout_rate,
+                    residual_dropout_rate,
+                    with_self_attn=True,
+                    with_cross_attn=True,
+                )
+                for i in range(num_decoder_layers)
+            ]
+        )
+
+        self.beam_size = beam_size
+        self.d_model = d_model
+        self.nhead = nhead
+        self.tgt_word_prj = nn.Linear(d_model, self.out_channels, bias_attr=False)
+        w0 = np.random.normal(
+            0.0, d_model**-0.5, (d_model, self.out_channels)
+        ).astype(np.float32)
+        self.tgt_word_prj.weight.set_value(w0)
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            xavier_normal_(m.weight)
+            if m.bias is not None:
+                zeros_(m.bias)
+
+    def forward_train(self, src, tgt):
+        tgt = tgt[:, :-1]
+
+        tgt = self.embedding(tgt)
+        tgt = self.positional_encoding(tgt)
+        tgt_mask = self.generate_square_subsequent_mask(tgt.shape[1])
+
+        if self.encoder is not None:
+            src = self.positional_encoding(src)
+            for encoder_layer in self.encoder:
+                src = encoder_layer(src)
+            memory = src  # B N C
+        else:
+            memory = src  # B N C
+        for decoder_layer in self.decoder:
+            tgt = decoder_layer(tgt, memory, self_mask=tgt_mask)
+        output = tgt
+        logit = self.tgt_word_prj(output)
+        return logit
+
+    def forward(self, src, targets=None):
+        """Take in and process masked source/target sequences.
+        Args:
+            src: the sequence to the encoder (required).
+            tgt: the sequence to the decoder (required).
+        Shape:
+            - src: :math:`(B, sN, C)`.
+            - tgt: :math:`(B, tN, C)`.
+        Examples:
+            >>> output = transformer_model(src, tgt)
+        """
+
+        if self.training:
+            max_len = targets[1].max()
+            tgt = targets[0][:, : 2 + max_len]
+            return self.forward_train(src, tgt)
+        else:
+            if self.beam_size > 0:
+                return self.forward_beam(src)
+            else:
+                return self.forward_test(src)
+
+    def forward_test(self, src):
+        bs = src.shape[0]
+        if self.encoder is not None:
+            src = self.positional_encoding(src)
+            for encoder_layer in self.encoder:
+                src = encoder_layer(src)
+            memory = src  # B N C
+        else:
+            memory = src
+        dec_seq = paddle.full((bs, 1), 2, dtype=paddle.int64)
+        dec_prob = paddle.full((bs, 1), 1.0, dtype=paddle.float32)
+        for len_dec_seq in range(1, paddle.to_tensor(self.max_len)):
+            dec_seq_embed = self.embedding(dec_seq)
+            dec_seq_embed = self.positional_encoding(dec_seq_embed)
+            tgt_mask = self.generate_square_subsequent_mask(dec_seq_embed.shape[1])
+            tgt = dec_seq_embed
+            for decoder_layer in self.decoder:
+                tgt = decoder_layer(tgt, memory, self_mask=tgt_mask)
+            dec_output = tgt
+            dec_output = dec_output[:, -1, :]
+            word_prob = F.softmax(self.tgt_word_prj(dec_output), axis=-1)
+            preds_idx = paddle.argmax(word_prob, axis=-1)
+            if paddle.equal_all(
+                preds_idx, paddle.full(preds_idx.shape, 3, dtype="int64")
+            ):
+                break
+            preds_prob = paddle.max(word_prob, axis=-1)
+            dec_seq = paddle.concat(
+                [dec_seq, paddle.reshape(preds_idx, [-1, 1])], axis=1
+            )
+            dec_prob = paddle.concat(
+                [dec_prob, paddle.reshape(preds_prob, [-1, 1])], axis=1
+            )
+        return [dec_seq, dec_prob]
+
+    def forward_beam(self, images):
+        """Translation work in one batch"""
+
+        def get_inst_idx_to_tensor_position_map(inst_idx_list):
+            """Indicate the position of an instance in a tensor."""
+            return {
+                inst_idx: tensor_position
+                for tensor_position, inst_idx in enumerate(inst_idx_list)
+            }
+
+        def collect_active_part(
+            beamed_tensor, curr_active_inst_idx, n_prev_active_inst, n_bm
+        ):
+            """Collect tensor parts associated to active instances."""
+
+            beamed_tensor_shape = beamed_tensor.shape
+            n_curr_active_inst = len(curr_active_inst_idx)
+            new_shape = (
+                n_curr_active_inst * n_bm,
+                beamed_tensor_shape[1],
+                beamed_tensor_shape[2],
+            )
+
+            beamed_tensor = beamed_tensor.reshape([n_prev_active_inst, -1])
+            beamed_tensor = beamed_tensor.index_select(curr_active_inst_idx, axis=0)
+            beamed_tensor = beamed_tensor.reshape(new_shape)
+
+            return beamed_tensor
+
+        def collate_active_info(
+            src_enc, inst_idx_to_position_map, active_inst_idx_list
+        ):
+            # Sentences which are still active are collected,
+            # so the decoder will not run on completed sentences.
+
+            n_prev_active_inst = len(inst_idx_to_position_map)
+            active_inst_idx = [
+                inst_idx_to_position_map[k] for k in active_inst_idx_list
+            ]
+            active_inst_idx = paddle.to_tensor(active_inst_idx, dtype="int64")
+            active_src_enc = collect_active_part(
+                src_enc.transpose([1, 0, 2]), active_inst_idx, n_prev_active_inst, n_bm
+            ).transpose([1, 0, 2])
+            active_inst_idx_to_position_map = get_inst_idx_to_tensor_position_map(
+                active_inst_idx_list
+            )
+            return active_src_enc, active_inst_idx_to_position_map
+
+        def beam_decode_step(
+            inst_dec_beams, len_dec_seq, enc_output, inst_idx_to_position_map, n_bm
+        ):
+            """Decode and update beam status, and then return active beam idx"""
+
+            def prepare_beam_dec_seq(inst_dec_beams, len_dec_seq):
+                dec_partial_seq = [
+                    b.get_current_state() for b in inst_dec_beams if not b.done
+                ]
+                dec_partial_seq = paddle.stack(dec_partial_seq)
+                dec_partial_seq = dec_partial_seq.reshape([-1, len_dec_seq])
+                return dec_partial_seq
+
+            def predict_word(dec_seq, enc_output, n_active_inst, n_bm):
+                dec_seq = self.embedding(dec_seq)
+                dec_seq = self.positional_encoding(dec_seq)
+                tgt_mask = self.generate_square_subsequent_mask(dec_seq.shape[1])
+                tgt = dec_seq
+                for decoder_layer in self.decoder:
+                    tgt = decoder_layer(tgt, enc_output, self_mask=tgt_mask)
+                dec_output = tgt
+                dec_output = dec_output[:, -1, :]  # Pick the last step: (bh * bm) * d_h
+                word_prob = F.softmax(self.tgt_word_prj(dec_output), axis=1)
+                word_prob = paddle.reshape(word_prob, [n_active_inst, n_bm, -1])
+                return word_prob
+
+            def collect_active_inst_idx_list(
+                inst_beams, word_prob, inst_idx_to_position_map
+            ):
+                active_inst_idx_list = []
+                for inst_idx, inst_position in inst_idx_to_position_map.items():
+                    is_inst_complete = inst_beams[inst_idx].advance(
+                        word_prob[inst_position]
+                    )
+                    if not is_inst_complete:
+                        active_inst_idx_list += [inst_idx]
+
+                return active_inst_idx_list
+
+            n_active_inst = len(inst_idx_to_position_map)
+            dec_seq = prepare_beam_dec_seq(inst_dec_beams, len_dec_seq)
+            word_prob = predict_word(dec_seq, enc_output, n_active_inst, n_bm)
+            # Update the beam with predicted word prob information and collect incomplete instances
+            active_inst_idx_list = collect_active_inst_idx_list(
+                inst_dec_beams, word_prob, inst_idx_to_position_map
+            )
+            return active_inst_idx_list
+
+        def collect_hypothesis_and_scores(inst_dec_beams, n_best):
+            all_hyp, all_scores = [], []
+            for inst_idx in range(len(inst_dec_beams)):
+                scores, tail_idxs = inst_dec_beams[inst_idx].sort_scores()
+                all_scores += [scores[:n_best]]
+                hyps = [
+                    inst_dec_beams[inst_idx].get_hypothesis(i)
+                    for i in tail_idxs[:n_best]
+                ]
+                all_hyp += [hyps]
+            return all_hyp, all_scores
+
+        with paddle.no_grad():
+            # -- Encode
+            if self.encoder is not None:
+                src = self.positional_encoding(images)
+                src_enc = self.encoder(src)
+            else:
+                src_enc = images
+
+            n_bm = self.beam_size
+            src_shape = src_enc.shape
+            inst_dec_beams = [Beam(n_bm) for _ in range(1)]
+            active_inst_idx_list = list(range(1))
+            # Repeat data for beam search
+            src_enc = paddle.tile(src_enc, [1, n_bm, 1])
+            inst_idx_to_position_map = get_inst_idx_to_tensor_position_map(
+                active_inst_idx_list
+            )
+            # Decode
+            for len_dec_seq in range(1, paddle.to_tensor(self.max_len)):
+                src_enc_copy = src_enc.clone()
+                active_inst_idx_list = beam_decode_step(
+                    inst_dec_beams,
+                    len_dec_seq,
+                    src_enc_copy,
+                    inst_idx_to_position_map,
+                    n_bm,
+                )
+                if not active_inst_idx_list:
+                    break  # all instances have finished their path to <EOS>
+                src_enc, inst_idx_to_position_map = collate_active_info(
+                    src_enc_copy, inst_idx_to_position_map, active_inst_idx_list
+                )
+        batch_hyp, batch_scores = collect_hypothesis_and_scores(inst_dec_beams, 1)
+        result_hyp = []
+        hyp_scores = []
+        for bs_hyp, score in zip(batch_hyp, batch_scores):
+            l = len(bs_hyp[0])
+            bs_hyp_pad = bs_hyp[0] + [3] * (25 - l)
+            result_hyp.append(bs_hyp_pad)
+            score = float(score) / l
+            hyp_score = [score for _ in range(25)]
+            hyp_scores.append(hyp_score)
+        return [
+            paddle.to_tensor(np.array(result_hyp), dtype=paddle.int64),
+            paddle.to_tensor(hyp_scores),
+        ]
+
+    def generate_square_subsequent_mask(self, sz):
+        """Generate a square mask for the sequence. The masked positions are filled with float('-inf').
+        Unmasked positions are filled with float(0.0).
+        """
+        mask = paddle.zeros([sz, sz], dtype="float32")
+        mask_inf = paddle.triu(
+            paddle.full(shape=[sz, sz], dtype="float32", fill_value=float("-inf")),
+            diagonal=1,
+        )
+        mask = mask + mask_inf
+        return mask.unsqueeze([0, 1])
+
+
+class MultiheadAttention(nn.Layer):
+    """Allows the model to jointly attend to information
+    from different representation subspaces.
+    See reference: Attention Is All You Need
+
+    .. math::
+        \text{MultiHead}(Q, K, V) = \text{Concat}(head_1,\dots,head_h)W^O
+        \text{where} head_i = \text{Attention}(QW_i^Q, KW_i^K, VW_i^V)
+
+    Args:
+        embed_dim: total dimension of the model
+        num_heads: parallel attention layers, or heads
+
+    """
+
+    def __init__(self, embed_dim, num_heads, dropout=0.0, self_attn=False):
+        super(MultiheadAttention, self).__init__()
+        self.embed_dim = embed_dim
+        self.num_heads = num_heads
+        # self.dropout = dropout
+        self.head_dim = embed_dim // num_heads
+        assert (
+            self.head_dim * num_heads == self.embed_dim
+        ), "embed_dim must be divisible by num_heads"
+        self.scale = self.head_dim**-0.5
+        self.self_attn = self_attn
+        if self_attn:
+            self.qkv = nn.Linear(embed_dim, embed_dim * 3)
+        else:
+            self.q = nn.Linear(embed_dim, embed_dim)
+            self.kv = nn.Linear(embed_dim, embed_dim * 2)
+        self.attn_drop = nn.Dropout(dropout)
+        self.out_proj = nn.Linear(embed_dim, embed_dim)
+
+    def forward(self, query, key=None, attn_mask=None):
+        qN = query.shape[1]
+
+        if self.self_attn:
+            qkv = (
+                self.qkv(query)
+                .reshape((0, qN, 3, self.num_heads, self.head_dim))
+                .transpose((2, 0, 3, 1, 4))
+            )
+            q, k, v = qkv[0], qkv[1], qkv[2]
+        else:
+            kN = key.shape[1]
+            q = (
+                self.q(query)
+                .reshape([0, qN, self.num_heads, self.head_dim])
+                .transpose([0, 2, 1, 3])
+            )
+            kv = (
+                self.kv(key)
+                .reshape((0, kN, 2, self.num_heads, self.head_dim))
+                .transpose((2, 0, 3, 1, 4))
+            )
+            k, v = kv[0], kv[1]
+
+        attn = (q.matmul(k.transpose((0, 1, 3, 2)))) * self.scale
+
+        if attn_mask is not None:
+            attn += attn_mask
+
+        attn = F.softmax(attn, axis=-1)
+        attn = self.attn_drop(attn)
+
+        x = (attn.matmul(v)).transpose((0, 2, 1, 3)).reshape((0, qN, self.embed_dim))
+        x = self.out_proj(x)
+
+        return x
+
+
+class TransformerBlock(nn.Layer):
+    def __init__(
+        self,
+        d_model,
+        nhead,
+        dim_feedforward=2048,
+        attention_dropout_rate=0.0,
+        residual_dropout_rate=0.1,
+        with_self_attn=True,
+        with_cross_attn=False,
+        epsilon=1e-5,
+    ):
+        super(TransformerBlock, self).__init__()
+        self.with_self_attn = with_self_attn
+        if with_self_attn:
+            self.self_attn = MultiheadAttention(
+                d_model, nhead, dropout=attention_dropout_rate, self_attn=with_self_attn
+            )
+            self.norm1 = LayerNorm(d_model, epsilon=epsilon)
+            self.dropout1 = Dropout(residual_dropout_rate)
+        self.with_cross_attn = with_cross_attn
+        if with_cross_attn:
+            self.cross_attn = (
+                MultiheadAttention(  # for self_attn of encoder or cross_attn of decoder
+                    d_model, nhead, dropout=attention_dropout_rate
+                )
+            )
+            self.norm2 = LayerNorm(d_model, epsilon=epsilon)
+            self.dropout2 = Dropout(residual_dropout_rate)
+
+        self.mlp = Mlp(
+            in_features=d_model,
+            hidden_features=dim_feedforward,
+            act_layer=nn.ReLU,
+            drop=residual_dropout_rate,
+        )
+
+        self.norm3 = LayerNorm(d_model, epsilon=epsilon)
+
+        self.dropout3 = Dropout(residual_dropout_rate)
+
+    def forward(self, tgt, memory=None, self_mask=None, cross_mask=None):
+        if self.with_self_attn:
+            tgt1 = self.self_attn(tgt, attn_mask=self_mask)
+            tgt = self.norm1(tgt + self.dropout1(tgt1))
+
+        if self.with_cross_attn:
+            tgt2 = self.cross_attn(tgt, key=memory, attn_mask=cross_mask)
+            tgt = self.norm2(tgt + self.dropout2(tgt2))
+        tgt = self.norm3(tgt + self.dropout3(self.mlp(tgt)))
+        return tgt
+
+
+class PositionalEncoding(nn.Layer):
+    """Inject some information about the relative or absolute position of the tokens
+        in the sequence. The positional encodings have the same dimension as
+        the embeddings, so that the two can be summed. Here, we use sine and cosine
+        functions of different frequencies.
+    .. math::
+        \text{PosEncoder}(pos, 2i) = sin(pos/10000^(2i/d_model))
+        \text{PosEncoder}(pos, 2i+1) = cos(pos/10000^(2i/d_model))
+        \text{where pos is the word position and i is the embed idx)
+    Args:
+        d_model: the embed dim (required).
+        dropout: the dropout value (default=0.1).
+        max_len: the max. length of the incoming sequence (default=5000).
+    Examples:
+        >>> pos_encoder = PositionalEncoding(d_model)
+    """
+
+    def __init__(self, dropout, dim, max_len=5000):
+        super(PositionalEncoding, self).__init__()
+        self.dropout = nn.Dropout(p=dropout)
+
+        pe = paddle.zeros([max_len, dim])
+        position = paddle.arange(0, max_len, dtype=paddle.float32).unsqueeze(1)
+        div_term = paddle.exp(
+            paddle.arange(0, dim, 2).astype("float32") * (-math.log(10000.0) / dim)
+        )
+        pe[:, 0::2] = paddle.sin(position * div_term)
+        pe[:, 1::2] = paddle.cos(position * div_term)
+        pe = paddle.unsqueeze(pe, 0)
+        pe = paddle.transpose(pe, [1, 0, 2])
+        self.register_buffer("pe", pe)
+
+    def forward(self, x):
+        """Inputs of forward function
+        Args:
+            x: the sequence fed to the positional encoder model (required).
+        Shape:
+            x: [sequence length, batch size, embed dim]
+            output: [sequence length, batch size, embed dim]
+        Examples:
+            >>> output = pos_encoder(x)
+        """
+        x = x.transpose([1, 0, 2])
+        x = x + self.pe[: x.shape[0], :]
+        return self.dropout(x).transpose([1, 0, 2])
+
+
+class PositionalEncoding_2d(nn.Layer):
+    """Inject some information about the relative or absolute position of the tokens
+        in the sequence. The positional encodings have the same dimension as
+        the embeddings, so that the two can be summed. Here, we use sine and cosine
+        functions of different frequencies.
+    .. math::
+        \text{PosEncoder}(pos, 2i) = sin(pos/10000^(2i/d_model))
+        \text{PosEncoder}(pos, 2i+1) = cos(pos/10000^(2i/d_model))
+        \text{where pos is the word position and i is the embed idx)
+    Args:
+        d_model: the embed dim (required).
+        dropout: the dropout value (default=0.1).
+        max_len: the max. length of the incoming sequence (default=5000).
+    Examples:
+        >>> pos_encoder = PositionalEncoding(d_model)
+    """
+
+    def __init__(self, dropout, dim, max_len=5000):
+        super(PositionalEncoding_2d, self).__init__()
+        self.dropout = nn.Dropout(p=dropout)
+
+        pe = paddle.zeros([max_len, dim])
+        position = paddle.arange(0, max_len, dtype=paddle.float32).unsqueeze(1)
+        div_term = paddle.exp(
+            paddle.arange(0, dim, 2).astype("float32") * (-math.log(10000.0) / dim)
+        )
+        pe[:, 0::2] = paddle.sin(position * div_term)
+        pe[:, 1::2] = paddle.cos(position * div_term)
+        pe = paddle.transpose(paddle.unsqueeze(pe, 0), [1, 0, 2])
+        self.register_buffer("pe", pe)
+
+        self.avg_pool_1 = nn.AdaptiveAvgPool2D((1, 1))
+        self.linear1 = nn.Linear(dim, dim)
+        self.linear1.weight.data.fill_(1.0)
+        self.avg_pool_2 = nn.AdaptiveAvgPool2D((1, 1))
+        self.linear2 = nn.Linear(dim, dim)
+        self.linear2.weight.data.fill_(1.0)
+
+    def forward(self, x):
+        """Inputs of forward function
+        Args:
+            x: the sequence fed to the positional encoder model (required).
+        Shape:
+            x: [sequence length, batch size, embed dim]
+            output: [sequence length, batch size, embed dim]
+        Examples:
+            >>> output = pos_encoder(x)
+        """
+        w_pe = self.pe[: x.shape[-1], :]
+        w1 = self.linear1(self.avg_pool_1(x).squeeze()).unsqueeze(0)
+        w_pe = w_pe * w1
+        w_pe = paddle.transpose(w_pe, [1, 2, 0])
+        w_pe = paddle.unsqueeze(w_pe, 2)
+
+        h_pe = self.pe[: x.shape.shape[-2], :]
+        w2 = self.linear2(self.avg_pool_2(x).squeeze()).unsqueeze(0)
+        h_pe = h_pe * w2
+        h_pe = paddle.transpose(h_pe, [1, 2, 0])
+        h_pe = paddle.unsqueeze(h_pe, 3)
+
+        x = x + w_pe + h_pe
+        x = paddle.transpose(
+            paddle.reshape(x, [x.shape[0], x.shape[1], x.shape[2] * x.shape[3]]),
+            [2, 0, 1],
+        )
+
+        return self.dropout(x)
+
+
+class Embeddings(nn.Layer):
+    def __init__(self, d_model, vocab, padding_idx=None, scale_embedding=True):
+        super(Embeddings, self).__init__()
+        self.embedding = nn.Embedding(vocab, d_model, padding_idx=padding_idx)
+        w0 = np.random.normal(0.0, d_model**-0.5, (vocab, d_model)).astype(np.float32)
+        self.embedding.weight.set_value(w0)
+        self.d_model = d_model
+        self.scale_embedding = scale_embedding
+
+    def forward(self, x):
+        if self.scale_embedding:
+            x = self.embedding(x)
+            return x * math.sqrt(self.d_model)
+        return self.embedding(x)
+
+
+class Beam:
+    """Beam search"""
+
+    def __init__(self, size, device=False):
+        self.size = size
+        self._done = False
+        # The score for each translation on the beam.
+        self.scores = paddle.zeros((size,), dtype=paddle.float32)
+        self.all_scores = []
+        # The backpointers at each time-step.
+        self.prev_ks = []
+        # The outputs at each time-step.
+        self.next_ys = [paddle.full((size,), 0, dtype=paddle.int64)]
+        self.next_ys[0][0] = 2
+
+    def get_current_state(self):
+        "Get the outputs for the current timestep."
+        return self.get_tentative_hypothesis()
+
+    def get_current_origin(self):
+        "Get the backpointers for the current timestep."
+        return self.prev_ks[-1]
+
+    @property
+    def done(self):
+        return self._done
+
+    def advance(self, word_prob):
+        "Update beam status and check if finished or not."
+        num_words = word_prob.shape[1]
+
+        # Sum the previous scores.
+        if len(self.prev_ks) > 0:
+            beam_lk = word_prob + self.scores.unsqueeze(1).expand_as(word_prob)
+        else:
+            beam_lk = word_prob[0]
+
+        flat_beam_lk = beam_lk.reshape([-1])
+        best_scores, best_scores_id = flat_beam_lk.topk(
+            self.size, 0, True, True
+        )  # 1st sort
+        self.all_scores.append(self.scores)
+        self.scores = best_scores
+        # bestScoresId is flattened as a (beam x word) array,
+        # so we need to calculate which word and beam each score came from
+        prev_k = best_scores_id // num_words
+        self.prev_ks.append(prev_k)
+        self.next_ys.append(best_scores_id - prev_k * num_words)
+        # End condition is when top-of-beam is EOS.
+        if self.next_ys[-1][0] == 3:
+            self._done = True
+            self.all_scores.append(self.scores)
+
+        return self._done
+
+    def sort_scores(self):
+        "Sort the scores."
+        return self.scores, paddle.to_tensor(
+            [i for i in range(int(self.scores.shape[0]))], dtype="int32"
+        )
+
+    def get_the_best_score_and_idx(self):
+        "Get the score of the best in the beam."
+        scores, ids = self.sort_scores()
+        return scores[1], ids[1]
+
+    def get_tentative_hypothesis(self):
+        "Get the decoded sequence for the current timestep."
+        if len(self.next_ys) == 1:
+            dec_seq = self.next_ys[0].unsqueeze(1)
+        else:
+            _, keys = self.sort_scores()
+            hyps = [self.get_hypothesis(k) for k in keys]
+            hyps = [[2] + h for h in hyps]
+            dec_seq = paddle.to_tensor(hyps, dtype="int64")
+        return dec_seq
+
+    def get_hypothesis(self, k):
+        """Walk back to construct the full hypothesis."""
+        hyp = []
+        for j in range(len(self.prev_ks) - 1, -1, -1):
+            hyp.append(self.next_ys[j + 1][k])
+            k = self.prev_ks[j][k]
+        return list(map(lambda x: x.item(), hyp[::-1]))
diff --git a/ppocr/modeling/heads/rec_parseq_head.py b/ppocr/modeling/heads/rec_parseq_head.py
new file mode 100644
index 0000000..cf79c7e
--- /dev/null
+++ b/ppocr/modeling/heads/rec_parseq_head.py
@@ -0,0 +1,504 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# Code was based on https://github.com/baudm/parseq/blob/main/strhub/models/parseq/system.py
+# reference: https://arxiv.org/abs/2207.06966
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+from paddle import nn, ParamAttr
+from paddle.nn import functional as F
+import numpy as np
+from .self_attention import WrapEncoderForFeature
+from .self_attention import WrapEncoder
+from collections import OrderedDict
+from typing import Optional
+import copy
+from itertools import permutations
+
+
+class DecoderLayer(paddle.nn.Layer):
+    """A Transformer decoder layer supporting two-stream attention (XLNet)
+    This implements a pre-LN decoder, as opposed to the post-LN default in PyTorch."""
+
+    def __init__(
+        self,
+        d_model,
+        nhead,
+        dim_feedforward=2048,
+        dropout=0.1,
+        activation="gelu",
+        layer_norm_eps=1e-05,
+    ):
+        super().__init__()
+        self.self_attn = paddle.nn.MultiHeadAttention(
+            d_model, nhead, dropout=dropout, need_weights=True
+        )  # paddle.nn.MultiHeadAttention默认为batch_first模式
+        self.cross_attn = paddle.nn.MultiHeadAttention(
+            d_model, nhead, dropout=dropout, need_weights=True
+        )
+        self.linear1 = paddle.nn.Linear(
+            in_features=d_model, out_features=dim_feedforward
+        )
+        self.dropout = paddle.nn.Dropout(p=dropout)
+        self.linear2 = paddle.nn.Linear(
+            in_features=dim_feedforward, out_features=d_model
+        )
+        self.norm1 = paddle.nn.LayerNorm(
+            normalized_shape=d_model, epsilon=layer_norm_eps
+        )
+        self.norm2 = paddle.nn.LayerNorm(
+            normalized_shape=d_model, epsilon=layer_norm_eps
+        )
+        self.norm_q = paddle.nn.LayerNorm(
+            normalized_shape=d_model, epsilon=layer_norm_eps
+        )
+        self.norm_c = paddle.nn.LayerNorm(
+            normalized_shape=d_model, epsilon=layer_norm_eps
+        )
+        self.dropout1 = paddle.nn.Dropout(p=dropout)
+        self.dropout2 = paddle.nn.Dropout(p=dropout)
+        self.dropout3 = paddle.nn.Dropout(p=dropout)
+        if activation == "gelu":
+            self.activation = paddle.nn.GELU()
+
+    def __setstate__(self, state):
+        if "activation" not in state:
+            state["activation"] = paddle.nn.functional.gelu
+        super().__setstate__(state)
+
+    def forward_stream(
+        self, tgt, tgt_norm, tgt_kv, memory, tgt_mask, tgt_key_padding_mask
+    ):
+        """Forward pass for a single stream (i.e. content or query)
+        tgt_norm is just a LayerNorm'd tgt. Added as a separate parameter for efficiency.
+        Both tgt_kv and memory are expected to be LayerNorm'd too.
+        memory is LayerNorm'd by ViT.
+        """
+        if tgt_key_padding_mask is not None:
+            tgt_mask1 = (tgt_mask != float("-inf"))[None, None, :, :] & (
+                tgt_key_padding_mask[:, None, None, :] == False
+            )
+            tgt2, sa_weights = self.self_attn(
+                tgt_norm, tgt_kv, tgt_kv, attn_mask=tgt_mask1
+            )
+        else:
+            tgt2, sa_weights = self.self_attn(
+                tgt_norm, tgt_kv, tgt_kv, attn_mask=tgt_mask
+            )
+
+        tgt = tgt + self.dropout1(tgt2)
+        tgt2, ca_weights = self.cross_attn(self.norm1(tgt), memory, memory)
+        tgt = tgt + self.dropout2(tgt2)
+        tgt2 = self.linear2(
+            self.dropout(self.activation(self.linear1(self.norm2(tgt))))
+        )
+        tgt = tgt + self.dropout3(tgt2)
+        return tgt, sa_weights, ca_weights
+
+    def forward(
+        self,
+        query,
+        content,
+        memory,
+        query_mask=None,
+        content_mask=None,
+        content_key_padding_mask=None,
+        update_content=True,
+    ):
+        query_norm = self.norm_q(query)
+        content_norm = self.norm_c(content)
+        query = self.forward_stream(
+            query,
+            query_norm,
+            content_norm,
+            memory,
+            query_mask,
+            content_key_padding_mask,
+        )[0]
+        if update_content:
+            content = self.forward_stream(
+                content,
+                content_norm,
+                content_norm,
+                memory,
+                content_mask,
+                content_key_padding_mask,
+            )[0]
+        return query, content
+
+
+def get_clones(module, N):
+    return paddle.nn.LayerList([copy.deepcopy(module) for i in range(N)])
+
+
+class Decoder(paddle.nn.Layer):
+    __constants__ = ["norm"]
+
+    def __init__(self, decoder_layer, num_layers, norm):
+        super().__init__()
+        self.layers = get_clones(decoder_layer, num_layers)
+        self.num_layers = num_layers
+        self.norm = norm
+
+    def forward(
+        self,
+        query,
+        content,
+        memory,
+        query_mask: Optional[paddle.Tensor] = None,
+        content_mask: Optional[paddle.Tensor] = None,
+        content_key_padding_mask: Optional[paddle.Tensor] = None,
+    ):
+        for i, mod in enumerate(self.layers):
+            last = i == len(self.layers) - 1
+            query, content = mod(
+                query,
+                content,
+                memory,
+                query_mask,
+                content_mask,
+                content_key_padding_mask,
+                update_content=not last,
+            )
+        query = self.norm(query)
+        return query
+
+
+class TokenEmbedding(paddle.nn.Layer):
+    def __init__(self, charset_size: int, embed_dim: int):
+        super().__init__()
+        self.embedding = paddle.nn.Embedding(
+            num_embeddings=charset_size, embedding_dim=embed_dim
+        )
+        self.embed_dim = embed_dim
+
+    def forward(self, tokens: paddle.Tensor):
+        return math.sqrt(self.embed_dim) * self.embedding(tokens.astype(paddle.int64))
+
+
+def trunc_normal_init(param, **kwargs):
+    initializer = nn.initializer.TruncatedNormal(**kwargs)
+    initializer(param, param.block)
+
+
+def constant_init(param, **kwargs):
+    initializer = nn.initializer.Constant(**kwargs)
+    initializer(param, param.block)
+
+
+def kaiming_normal_init(param, **kwargs):
+    initializer = nn.initializer.KaimingNormal(**kwargs)
+    initializer(param, param.block)
+
+
+class ParseQHead(nn.Layer):
+    def __init__(
+        self,
+        out_channels,
+        max_text_length,
+        embed_dim,
+        dec_num_heads,
+        dec_mlp_ratio,
+        dec_depth,
+        perm_num,
+        perm_forward,
+        perm_mirrored,
+        decode_ar,
+        refine_iters,
+        dropout,
+        **kwargs,
+    ):
+        super().__init__()
+
+        self.bos_id = out_channels - 2
+        self.eos_id = 0
+        self.pad_id = out_channels - 1
+
+        self.max_label_length = max_text_length
+        self.decode_ar = decode_ar
+        self.refine_iters = refine_iters
+        decoder_layer = DecoderLayer(
+            embed_dim, dec_num_heads, embed_dim * dec_mlp_ratio, dropout
+        )
+        self.decoder = Decoder(
+            decoder_layer,
+            num_layers=dec_depth,
+            norm=paddle.nn.LayerNorm(normalized_shape=embed_dim),
+        )
+        self.rng = np.random.default_rng()
+        self.max_gen_perms = perm_num // 2 if perm_mirrored else perm_num
+        self.perm_forward = perm_forward
+        self.perm_mirrored = perm_mirrored
+        self.head = paddle.nn.Linear(
+            in_features=embed_dim, out_features=out_channels - 2
+        )
+        self.text_embed = TokenEmbedding(out_channels, embed_dim)
+        self.pos_queries = paddle.create_parameter(
+            shape=paddle.empty(shape=[1, max_text_length + 1, embed_dim]).shape,
+            dtype=paddle.empty(shape=[1, max_text_length + 1, embed_dim]).numpy().dtype,
+            default_initializer=paddle.nn.initializer.Assign(
+                paddle.empty(shape=[1, max_text_length + 1, embed_dim])
+            ),
+        )
+        self.pos_queries.stop_gradient = not True
+        self.dropout = paddle.nn.Dropout(p=dropout)
+        self._device = self.parameters()[0].place
+        trunc_normal_init(self.pos_queries, std=0.02)
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, paddle.nn.Linear):
+            trunc_normal_init(m.weight, std=0.02)
+            if m.bias is not None:
+                constant_init(m.bias, value=0.0)
+        elif isinstance(m, paddle.nn.Embedding):
+            trunc_normal_init(m.weight, std=0.02)
+            if m._padding_idx is not None:
+                m.weight.data[m._padding_idx].zero_()
+        elif isinstance(m, paddle.nn.Conv2D):
+            kaiming_normal_init(m.weight, fan_in=None, nonlinearity="relu")
+            if m.bias is not None:
+                constant_init(m.bias, value=0.0)
+        elif isinstance(
+            m, (paddle.nn.LayerNorm, paddle.nn.BatchNorm2D, paddle.nn.GroupNorm)
+        ):
+            constant_init(m.weight, value=1.0)
+            constant_init(m.bias, value=0.0)
+
+    def no_weight_decay(self):
+        param_names = {"text_embed.embedding.weight", "pos_queries"}
+        enc_param_names = {("encoder." + n) for n in self.encoder.no_weight_decay()}
+        return param_names.union(enc_param_names)
+
+    def encode(self, img):
+        return self.encoder(img)
+
+    def decode(
+        self,
+        tgt,
+        memory,
+        tgt_mask=None,
+        tgt_padding_mask=None,
+        tgt_query=None,
+        tgt_query_mask=None,
+    ):
+        N, L = tgt.shape
+        null_ctx = self.text_embed(tgt[:, :1])
+        if L != 1:
+            tgt_emb = self.pos_queries[:, : L - 1] + self.text_embed(tgt[:, 1:])
+            tgt_emb = self.dropout(paddle.concat(x=[null_ctx, tgt_emb], axis=1))
+        else:
+            tgt_emb = self.dropout(null_ctx)
+        if tgt_query is None:
+            tgt_query = self.pos_queries[:, :L].expand(shape=[N, -1, -1])
+        tgt_query = self.dropout(tgt_query)
+        return self.decoder(
+            tgt_query, tgt_emb, memory, tgt_query_mask, tgt_mask, tgt_padding_mask
+        )
+
+    def forward_test(self, memory, max_length=None):
+        testing = max_length is None
+        max_length = (
+            self.max_label_length
+            if max_length is None
+            else min(max_length, self.max_label_length)
+        )
+        bs = memory.shape[0]
+        num_steps = max_length + 1
+
+        pos_queries = self.pos_queries[:, :num_steps].expand(shape=[bs, -1, -1])
+        tgt_mask = query_mask = paddle.triu(
+            x=paddle.full(shape=(num_steps, num_steps), fill_value=float("-inf")),
+            diagonal=1,
+        )
+        if self.decode_ar:
+            tgt_in = paddle.full(shape=(bs, num_steps), fill_value=self.pad_id).astype(
+                "int64"
+            )
+            tgt_in[:, (0)] = self.bos_id
+
+            logits = []
+            for i in range(paddle.to_tensor(num_steps)):
+                j = i + 1
+                tgt_out = self.decode(
+                    tgt_in[:, :j],
+                    memory,
+                    tgt_mask[:j, :j],
+                    tgt_query=pos_queries[:, i:j],
+                    tgt_query_mask=query_mask[i:j, :j],
+                )
+                p_i = self.head(tgt_out)
+                logits.append(p_i)
+                if j < num_steps:
+                    tgt_in[:, (j)] = p_i.squeeze().argmax(axis=-1)
+                    if (
+                        testing
+                        and (tgt_in == self.eos_id)
+                        .astype("bool")
+                        .any(axis=-1)
+                        .astype("bool")
+                        .all()
+                    ):
+                        break
+            logits = paddle.concat(x=logits, axis=1)
+        else:
+            tgt_in = paddle.full(shape=(bs, 1), fill_value=self.bos_id).astype("int64")
+            tgt_out = self.decode(tgt_in, memory, tgt_query=pos_queries)
+            logits = self.head(tgt_out)
+        if self.refine_iters:
+            temp = paddle.triu(
+                x=paddle.ones(shape=[num_steps, num_steps], dtype="bool"), diagonal=2
+            )
+            posi = paddle.where(temp == True)
+            query_mask[posi] = 0
+            bos = paddle.full(shape=(bs, 1), fill_value=self.bos_id).astype("int64")
+            for i in range(self.refine_iters):
+                tgt_in = paddle.concat(x=[bos, logits[:, :-1].argmax(axis=-1)], axis=1)
+                tgt_padding_mask = (tgt_in == self.eos_id).astype(dtype="int32")
+                tgt_padding_mask = tgt_padding_mask.cpu()
+                tgt_padding_mask = tgt_padding_mask.cumsum(axis=-1) > 0
+                tgt_padding_mask = (
+                    tgt_padding_mask.cuda().astype(dtype="float32") == 1.0
+                )
+                tgt_out = self.decode(
+                    tgt_in,
+                    memory,
+                    tgt_mask,
+                    tgt_padding_mask,
+                    tgt_query=pos_queries,
+                    tgt_query_mask=query_mask[:, : tgt_in.shape[1]],
+                )
+                logits = self.head(tgt_out)
+
+        # transfer to probability
+        logits = F.softmax(logits, axis=-1)
+
+        final_output = {"predict": logits}
+
+        return final_output
+
+    def gen_tgt_perms(self, tgt):
+        """Generate shared permutations for the whole batch.
+        This works because the same attention mask can be used for the shorter sequences
+        because of the padding mask.
+        """
+        max_num_chars = tgt.shape[1] - 2
+        if max_num_chars == 1:
+            return paddle.arange(end=3).unsqueeze(axis=0)
+        perms = [paddle.arange(end=max_num_chars)] if self.perm_forward else []
+        max_perms = math.factorial(max_num_chars)
+        if self.perm_mirrored:
+            max_perms //= 2
+        num_gen_perms = min(self.max_gen_perms, max_perms)
+        if max_num_chars < 5:
+            if max_num_chars == 4 and self.perm_mirrored:
+                selector = [0, 3, 4, 6, 9, 10, 12, 16, 17, 18, 19, 21]
+            else:
+                selector = list(range(max_perms))
+            perm_pool = paddle.to_tensor(
+                data=list(permutations(range(max_num_chars), max_num_chars)),
+                place=self._device,
+            )[selector]
+            if self.perm_forward:
+                perm_pool = perm_pool[1:]
+            perms = paddle.stack(x=perms)
+            if len(perm_pool):
+                i = self.rng.choice(
+                    len(perm_pool), size=num_gen_perms - len(perms), replace=False
+                )
+                perms = paddle.concat(x=[perms, perm_pool[i]])
+        else:
+            perms.extend(
+                [
+                    paddle.randperm(n=max_num_chars)
+                    for _ in range(num_gen_perms - len(perms))
+                ]
+            )
+            perms = paddle.stack(x=perms)
+        if self.perm_mirrored:
+            comp = perms.flip(axis=-1)
+            x = paddle.stack(x=[perms, comp])
+            perm_2 = list(range(x.ndim))
+            perm_2[0] = 1
+            perm_2[1] = 0
+            perms = x.transpose(perm=perm_2).reshape((-1, max_num_chars))
+        bos_idx = paddle.zeros(shape=(len(perms), 1), dtype=perms.dtype)
+        eos_idx = paddle.full(
+            shape=(len(perms), 1), fill_value=max_num_chars + 1, dtype=perms.dtype
+        )
+        perms = paddle.concat(x=[bos_idx, perms + 1, eos_idx], axis=1)
+        if len(perms) > 1:
+            perms[(1), 1:] = max_num_chars + 1 - paddle.arange(end=max_num_chars + 1)
+        return perms
+
+    def generate_attn_masks(self, perm):
+        """Generate attention masks given a sequence permutation (includes pos. for bos and eos tokens)
+        :param perm: the permutation sequence. i = 0 is always the BOS
+        :return: lookahead attention masks
+        """
+        sz = perm.shape[0]
+        mask = paddle.zeros(shape=(sz, sz))
+        for i in range(sz):
+            query_idx = perm[i].cpu().numpy().tolist()
+            masked_keys = perm[i + 1 :].cpu().numpy().tolist()
+            if len(masked_keys) == 0:
+                break
+            mask[query_idx, masked_keys] = float("-inf")
+        content_mask = mask[:-1, :-1].clone()
+        mask[paddle.eye(num_rows=sz).astype("bool")] = float("-inf")
+        query_mask = mask[1:, :-1]
+        return content_mask, query_mask
+
+    def forward_train(self, memory, tgt):
+        tgt_perms = self.gen_tgt_perms(tgt)
+        tgt_in = tgt[:, :-1]
+        tgt_padding_mask = (tgt_in == self.pad_id) | (tgt_in == self.eos_id)
+        logits_list = []
+        final_out = {}
+        for i, perm in enumerate(tgt_perms):
+            tgt_mask, query_mask = self.generate_attn_masks(perm)
+            out = self.decode(
+                tgt_in, memory, tgt_mask, tgt_padding_mask, tgt_query_mask=query_mask
+            )
+            logits = self.head(out)
+            if i == 0:
+                final_out["predict"] = logits
+            logits = logits.flatten(stop_axis=1)
+            logits_list.append(logits)
+
+        final_out["logits_list"] = logits_list
+        final_out["pad_id"] = self.pad_id
+        final_out["eos_id"] = self.eos_id
+
+        return final_out
+
+    def forward(self, feat, targets=None):
+        # feat : B, N, C
+        # targets : labels, labels_len
+
+        if self.training:
+            label = targets[0]  # label
+            label_len = targets[1]
+            max_step = paddle.max(label_len).cpu().numpy()[0] + 2
+            crop_label = label[:, :max_step]
+            final_out = self.forward_train(feat, crop_label)
+        else:
+            final_out = self.forward_test(feat)
+
+        return final_out
diff --git a/ppocr/modeling/heads/rec_ppformulanet_head.py b/ppocr/modeling/heads/rec_ppformulanet_head.py
new file mode 100644
index 0000000..644dc2a
--- /dev/null
+++ b/ppocr/modeling/heads/rec_ppformulanet_head.py
@@ -0,0 +1,1391 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import math
+import re
+import numpy as np
+import inspect
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle.nn import CrossEntropyLoss
+from paddle import Tensor
+from collections import OrderedDict
+from typing import Optional, Tuple, Union, List, Dict, Any
+from dataclasses import dataclass, fields, is_dataclass
+from ppocr.modeling.backbones.rec_donut_swin import DonutSwinModelOutput
+from ppocr.modeling.heads.rec_unimernet_head import (
+    MBartForCausalLM,
+    MBartDecoder,
+    MBartConfig,
+    ModelOutput,
+    BaseModelOutputWithPastAndCrossAttentions,
+    Seq2SeqLMOutput,
+    zeros_,
+    ones_,
+    kaiming_normal_,
+    trunc_normal_,
+    xavier_uniform_,
+    CausalLMOutputWithCrossAttentions,
+    LogitsProcessorList,
+    ForcedEOSTokenLogitsProcessor,
+    UniMERNetHead,
+)
+
+
+@dataclass
+class AttentionMaskConverter:
+    """
+    A class to convert attention masks based on specific configurations.
+
+    This class is designed to handle the conversion of attention masks with options for causal masking
+    and sliding window attention, which are commonly used in transformer models.
+
+    Attributes:
+        is_causal (bool): Flag indicating whether the attention mask should enforce causal masking,
+                          which ensures each position can only attend to previous positions.
+        sliding_window (int, optional): Size of the sliding window for local attention. If set,
+                                        attention is restricted to a local window of this size.
+
+    """
+
+    is_causal: bool
+    sliding_window: int
+
+    def __init__(self, is_causal: bool, sliding_window=None):
+        self.is_causal = is_causal
+        self.sliding_window = sliding_window
+
+        if self.sliding_window is not None and self.sliding_window <= 0:
+            raise ValueError(
+                f"Make sure that when passing `sliding_window` that its value is a strictly positive integer, not `{self.sliding_window}`"
+            )
+
+    @staticmethod
+    def _make_causal_mask(
+        input_ids_shape,
+        dtype,
+        past_key_values_length=0,
+        sliding_window=None,
+        is_export=False,
+    ):
+        """
+        Make causal mask used for bi-directional self-attention.
+        """
+        bsz, tgt_len = input_ids_shape
+        if is_export:
+            mask = paddle.full(
+                (tgt_len, tgt_len), paddle.finfo(dtype).min, dtype="float64"
+            )
+            mask_cond = paddle.arange(mask.shape[-1])
+            mask.masked_fill_(
+                mask_cond < (mask_cond + 1).reshape([mask.shape[-1], 1]), 0
+            )
+        else:
+            mask = paddle.full((tgt_len, tgt_len), paddle.finfo(dtype).min)
+            mask_cond = paddle.arange(mask.shape[-1])
+            mask.masked_fill_(
+                mask_cond < (mask_cond + 1).reshape([mask.shape[-1], 1]), 0
+            )
+            mask = mask.cast(dtype)
+
+        if past_key_values_length > 0:
+            mask = paddle.concat(
+                [paddle.zeros(tgt_len, past_key_values_length, dtype=dtype), mask],
+                axis=-1,
+            )
+
+        # add lower triangular sliding window mask if necessary
+        if sliding_window is not None:
+            diagonal = past_key_values_length - sliding_window - 1
+
+            context_mask = paddle.tril(
+                paddle.ones_like(mask, dtype=paddle.bool), diagonal=diagonal
+            )
+            mask.masked_fill_(context_mask, paddle.finfo(dtype).min)
+
+        return mask[None, None, :, :].expand(
+            [bsz, 1, tgt_len, tgt_len + past_key_values_length]
+        )
+
+    @staticmethod
+    def _make_causal_mask_parallel(
+        input_ids_shape,
+        dtype,
+        past_key_values_length=0,
+        sliding_window=None,
+        parallel_step=1,
+        is_export=False,
+    ):
+        """
+        Make causal mask used for bi-directional self-attention.
+        """
+        bsz, tgt_len = input_ids_shape
+        mask = paddle.full((tgt_len, tgt_len), paddle.finfo(dtype).min)
+        mask_cond = paddle.arange(mask.shape[-1])
+        mask_cond_parallel = paddle.arange(mask.shape[-1])
+
+        mask_parallel = paddle.arange(0, tgt_len, step=parallel_step).reshape([1, -1])
+        mask_parallel = paddle.repeat_interleave(mask_parallel, parallel_step, 1)[
+            :, :tgt_len
+        ]
+        mask.masked_fill_(
+            mask_cond < (mask_parallel + parallel_step).reshape([mask.shape[-1], 1]), 0
+        )
+        mask = mask.cast(dtype)
+
+        if past_key_values_length > 0:
+            mask = paddle.concat(
+                [paddle.zeros([tgt_len, past_key_values_length], dtype=dtype), mask],
+                axis=-1,
+            )
+
+        # add lower triangular sliding window mask if necessary
+        if sliding_window is not None:
+            diagonal = past_key_values_length - sliding_window - 1
+
+            context_mask = paddle.tril(
+                paddle.ones_like(mask, dtype=paddle.bool), diagonal=diagonal
+            )
+            mask.masked_fill_(context_mask, paddle.finfo(dtype).min)
+
+        return mask[None, None, :, :].expand(
+            [bsz, 1, tgt_len, tgt_len + past_key_values_length]
+        )
+
+    def to_4d(
+        self,
+        attention_mask_2d,
+        query_length,
+        dtype,
+        key_value_length,
+        use_parallel=False,
+        parallel_step=3,
+        is_export=False,
+    ):
+        """
+        Converts 2D attention mask to 4D attention mask by expanding mask to (bsz, head_dim=1, query_length,
+        key_value_length) shape and by adding a large negative bias to not-attended positions. If attention_mask is
+        causal, a causal mask will be added.
+        """
+        input_shape = (attention_mask_2d.shape[0], query_length)
+
+        causal_4d_mask = None
+        if use_parallel:
+            step = parallel_step
+        else:
+            step = 1
+        if (
+            input_shape[-1] > step or self.sliding_window is not None
+        ) and self.is_causal:
+
+            if key_value_length is None:
+                raise ValueError(
+                    "This attention mask converter is causal. Make sure to pass `key_value_length` to correctly create a causal mask."
+                )
+
+            past_key_values_length = key_value_length - query_length
+
+            if use_parallel:
+                causal_4d_mask = self._make_causal_mask_parallel(
+                    input_shape,
+                    dtype,
+                    past_key_values_length=past_key_values_length,
+                    sliding_window=self.sliding_window,
+                    parallel_step=parallel_step,
+                    is_export=is_export,
+                )
+            else:
+                causal_4d_mask = self._make_causal_mask(
+                    input_shape,
+                    dtype,
+                    past_key_values_length=past_key_values_length,
+                    sliding_window=self.sliding_window,
+                    is_export=is_export,
+                )
+
+        elif self.sliding_window is not None:
+            raise NotImplementedError(
+                "Sliding window is currently only implemented for causal masking"
+            )
+
+        expanded_attn_mask = self._expand_mask(
+            attention_mask_2d, dtype, tgt_len=input_shape[-1]
+        )
+
+        if causal_4d_mask is not None:
+            expanded_attn_mask = causal_4d_mask.masked_fill_(
+                expanded_attn_mask.cast(paddle.bool), paddle.finfo(dtype).min
+            )
+
+        expanded_4d_mask = expanded_attn_mask
+        return expanded_4d_mask
+
+    def to_4d_export(
+        self,
+        attention_mask_2d,
+        query_length,
+        dtype,
+        key_value_length,
+        use_parallel=False,
+        parallel_step=3,
+        is_export=False,
+    ):
+        input_shape = (attention_mask_2d.shape[0], query_length)
+
+        expanded_attn_mask = self._expand_mask_export(
+            attention_mask_2d, dtype, tgt_len=input_shape[-1]
+        )
+        expanded_4d_mask = expanded_attn_mask
+
+        return expanded_4d_mask
+
+    def _expand_mask(self, mask, dtype, tgt_len=None):
+        """
+        Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
+        """
+        bsz, src_len = mask.shape
+        tgt_len = tgt_len if tgt_len is not None else src_len
+        expanded_mask = (
+            mask[:, None, None, :].expand([bsz, 1, tgt_len, src_len]).cast(dtype)
+        )
+
+        inverted_mask = 1.0 - expanded_mask
+
+        return inverted_mask.masked_fill_(
+            inverted_mask.cast(paddle.bool), paddle.finfo(dtype).min
+        )
+
+    def _expand_mask_export(self, mask, dtype, tgt_len=None):
+        """
+        Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
+        """
+        bsz, src_len = paddle.shape(mask)
+        expanded_mask = (
+            mask[:, None, None, :].expand([bsz, 1, tgt_len, src_len]).cast(dtype)
+        )
+        paddle.jit.api.set_dynamic_shape(expanded_mask, [-1, -1, -1, -1])
+        inverted_mask = 1.0 - expanded_mask
+        return inverted_mask.masked_fill_(
+            inverted_mask.cast(paddle.bool), paddle.finfo(dtype).min
+        )
+
+
+def _prepare_4d_attention_mask(mask, dtype, tgt_len=None):
+    return AttentionMaskConverter._expand_mask(mask=mask, dtype=dtype, tgt_len=tgt_len)
+
+
+def _prepare_4d_causal_attention_mask(
+    attention_mask,
+    input_shape,
+    inputs_embeds,
+    past_key_values_length,
+    sliding_window=None,
+    use_parallel=False,
+    parallel_step=3,
+    is_export=False,
+):
+    """
+    Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape
+    `(batch_size, key_value_length)`
+
+    Args:
+        attention_mask (`paddle.Tensor` or `None`):
+            A 2D attention mask of shape `(batch_size, key_value_length)`
+        input_shape (`tuple(int)` or `list(int)` or `paddle.Size`):
+            The input shape should be a tuple that defines `(batch_size, query_length)`.
+        inputs_embeds (`paddle.Tensor`):
+            The embedded inputs as a paddle Tensor.
+        past_key_values_length (`int`):
+            The length of the key value cache.
+        sliding_window (`int`, *optional*):
+            If the model uses windowed attention, a sliding window should be passed.
+    """
+    attn_mask_converter = AttentionMaskConverter(
+        is_causal=True, sliding_window=sliding_window
+    )
+
+    key_value_length = input_shape[-1] + past_key_values_length
+
+    # 4d mask is passed through the layers
+    if attention_mask is not None and len(attention_mask.shape) == 2:
+        attention_mask = attn_mask_converter.to_4d(
+            attention_mask,
+            input_shape[-1],
+            key_value_length=key_value_length,
+            dtype=inputs_embeds.dtype,
+            use_parallel=use_parallel,
+            parallel_step=parallel_step,
+            is_export=is_export,
+        )
+    elif attention_mask is not None and len(attention_mask.shape) == 4:
+        expected_shape = (input_shape[0], 1, input_shape[1], key_value_length)
+        if tuple(attention_mask.shape) != expected_shape:
+            raise ValueError(
+                f"Incorrect 4D attention_mask shape: {tuple(attention_mask.shape)}; expected: {expected_shape}."
+            )
+        else:
+            # if the 4D mask has correct shape - invert it and fill with negative infinity
+            inverted_mask = 1.0 - attention_mask
+            attention_mask = inverted_mask.masked_fill_(
+                inverted_mask.to(paddle.bool), paddle.finfo(inputs_embeds.dtype).min
+            )
+    else:
+        attention_mask = attn_mask_converter.to_causal_4d(
+            input_shape[0],
+            input_shape[-1],
+            key_value_length,
+            dtype=inputs_embeds.dtype,
+        )
+
+    return attention_mask
+
+
+def _prepare_4d_causal_attention_mask_export(
+    attention_mask,
+    input_shape,
+    inputs_embeds,
+    past_key_values_length,
+    sliding_window=None,
+    use_parallel=False,
+    parallel_step=3,
+    is_export=False,
+):
+    """
+    Prepare a 4D causal attention mask for export.
+
+    This function prepares a 4-dimensional causal attention mask, which is used to ensure that each position in the
+    sequence can only attend to previous positions. It is specifically designed to handle scenarios where the model
+    is being exported, potentially with additional options like sliding window or parallel processing.
+
+    Args:
+        attention_mask: The initial attention mask, typically used to avoid attending to padding tokens.
+        input_shape: Shape of the input tensor, usually in the form (batch_size, sequence_length).
+        inputs_embeds: Embeddings of the input sequence, used to derive certain dimensions if needed.
+        past_key_values_length: Length of past key values, used in contexts like transformer decoders with caching.
+        sliding_window: Optional parameter. If provided, specifies the size of a sliding window for local attention.
+        use_parallel: Flag indicating whether to use parallel processing for attention computation.
+        parallel_step: Number of steps to use in parallel processing, relevant if `use_parallel` is True.
+        is_export: Flag indicating whether the attention mask is being prepared for model export.
+
+    Returns:
+        A 4D causal attention mask suitable for use in transformer models, ensuring correct causal masking.
+    """
+    attn_mask_converter = AttentionMaskConverter(
+        is_causal=True, sliding_window=sliding_window
+    )
+    key_value_length = input_shape[-1] + past_key_values_length
+
+    shape = attention_mask.shape
+    len_shape = len(shape)
+
+    attention_mask = attn_mask_converter.to_4d_export(
+        attention_mask,
+        input_shape[-1],
+        key_value_length=key_value_length,
+        dtype=inputs_embeds.dtype,
+        use_parallel=use_parallel,
+        parallel_step=parallel_step,
+        is_export=is_export,
+    )
+    return attention_mask
+
+
+class CustomMBartDecoder(MBartDecoder):
+    def __init__(self, config):
+        super().__init__(config)
+        hidden_size = config.d_model
+        self.is_export = config.is_export
+        self.config_decoder = config
+
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        head_mask=None,
+        cross_attn_head_mask=None,
+        past_key_values=None,
+        inputs_embeds=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        self.is_export = False if self.training else True
+
+        output_attentions = (
+            output_attentions
+            if output_attentions is not None
+            else self.config.output_attentions
+        )
+        output_hidden_states = (
+            output_hidden_states
+            if output_hidden_states is not None
+            else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = (
+            return_dict if return_dict is not None else self.config.use_return_dict
+        )
+
+        # retrieve input_ids and inputs_embeds
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError(
+                "You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time"
+            )
+        elif input_ids is not None:
+            input = input_ids
+            input_shape = input.shape
+            input_ids = input_ids.reshape([-1, input_shape[-1]])
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.shape[:-1]
+            input = inputs_embeds[:, :, -1]
+        else:
+            raise ValueError(
+                "You have to specify either decoder_input_ids or decoder_inputs_embeds"
+            )
+
+        # past_key_values_length
+        past_key_values_length = (
+            past_key_values[0][0].shape[2] if past_key_values is not None else 0
+        )
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids) * self.embed_scale
+
+        if self._use_flash_attention_2:
+            # 2d mask is passed through the layers
+            attention_mask = (
+                attention_mask
+                if (attention_mask is not None and 0 in attention_mask)
+                else None
+            )
+        else:
+            # 4d mask is passed through the layers
+            if self.is_export:
+                attention_mask = _prepare_4d_causal_attention_mask_export(
+                    attention_mask,
+                    input_shape,
+                    inputs_embeds,
+                    past_key_values_length,
+                    use_parallel=self.config_decoder.use_parallel,
+                    parallel_step=self.config_decoder.parallel_step,
+                    is_export=self.is_export,
+                )
+            else:
+                attention_mask = _prepare_4d_causal_attention_mask(
+                    attention_mask,
+                    input_shape,
+                    inputs_embeds,
+                    past_key_values_length,
+                    use_parallel=self.config_decoder.use_parallel,
+                    parallel_step=self.config_decoder.parallel_step,
+                    is_export=self.is_export,
+                )
+
+        # expand encoder attention mask
+        if encoder_hidden_states is not None and encoder_attention_mask is not None:
+            if self._use_flash_attention_2:
+                encoder_attention_mask = (
+                    encoder_attention_mask if 0 in encoder_attention_mask else None
+                )
+            else:
+                # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+                encoder_attention_mask = _prepare_4d_attention_mask(
+                    encoder_attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]
+                )
+
+        # embed positions
+        positions = self.embed_positions(input, past_key_values_length)
+
+        hidden_states = inputs_embeds + positions
+
+        hidden_states = self.layernorm_embedding(hidden_states)
+        hidden_states = nn.functional.dropout(
+            hidden_states, p=self.dropout, training=self.training
+        )
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                print(
+                    "`use_cache=True` is incompatible with gradient checkpointing`. Setting `use_cache=False`..."
+                )
+                use_cache = False
+
+        # decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        all_cross_attentions = (
+            () if (output_attentions and encoder_hidden_states is not None) else None
+        )
+        next_decoder_cache = () if use_cache else None
+
+        # check if head_mask/cross_attn_head_mask has a correct number of layers specified if desired
+        for attn_mask, mask_name in zip(
+            [head_mask, cross_attn_head_mask], ["head_mask", "cross_attn_head_mask"]
+        ):
+            if attn_mask is not None:
+                if attn_mask.size()[0] != len(self.layers):
+                    raise ValueError(
+                        f"The `{mask_name}` should be specified for {len(self.layers)} layers, but it is for"
+                        f" {attn_mask.size()[0]}."
+                    )
+        for idx, decoder_layer in enumerate(self.layers):
+            # add LayerDrop (see https://arxiv.org/abs/1909.11556 for description)
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+            if self.training:
+                dropout_probability = paddle.rand([])
+                if dropout_probability < self.layerdrop:
+                    continue
+
+            past_key_value = (
+                past_key_values[idx] if past_key_values is not None else None
+            )
+            if self.gradient_checkpointing and self.training:
+                layer_outputs = self._gradient_checkpointing_func(
+                    decoder_layer.__call__,
+                    hidden_states,
+                    attention_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                    head_mask[idx] if head_mask is not None else None,
+                    (
+                        cross_attn_head_mask[idx]
+                        if cross_attn_head_mask is not None
+                        else None
+                    ),
+                    None,
+                    output_attentions,
+                    use_cache,
+                )
+            else:
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    attention_mask=attention_mask,
+                    encoder_hidden_states=encoder_hidden_states,
+                    encoder_attention_mask=encoder_attention_mask,
+                    layer_head_mask=(head_mask[idx] if head_mask is not None else None),
+                    cross_attn_layer_head_mask=(
+                        cross_attn_head_mask[idx]
+                        if cross_attn_head_mask is not None
+                        else None
+                    ),
+                    past_key_value=past_key_value,
+                    output_attentions=output_attentions,
+                    use_cache=use_cache,
+                )
+            hidden_states = layer_outputs[0]
+
+            if self.is_export:
+                next_decoder_cache += (layer_outputs[3 if output_attentions else 1],)
+            else:
+                if use_cache:
+                    next_decoder_cache += (
+                        layer_outputs[3 if output_attentions else 1],
+                    )
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+                if encoder_hidden_states is not None:
+                    all_cross_attentions += (layer_outputs[2],)
+
+        hidden_states = self.layer_norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        if self.is_export:
+            next_cache = next_decoder_cache
+        else:
+            next_cache = next_decoder_cache if use_cache else None
+        if not return_dict:
+            return tuple(
+                v
+                for v in [
+                    hidden_states,
+                    next_cache,
+                    all_hidden_states,
+                    all_self_attns,
+                    all_cross_attentions,
+                ]
+                if v is not None
+            )
+
+        return BaseModelOutputWithPastAndCrossAttentions(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+            cross_attentions=all_cross_attentions,
+        )
+
+
+class CustomMBartForCausalLM(MBartForCausalLM):
+    def __init__(self, config):
+        super().__init__(config)
+        # Modify the decoder within MBartDecoderWrapper
+        self.model.decoder = CustomMBartDecoder(config)
+
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        head_mask=None,
+        cross_attn_head_mask=None,
+        past_key_values=None,
+        inputs_embeds=None,
+        labels=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        output_attentions = (
+            output_attentions
+            if output_attentions is not None
+            else self.config.output_attentions
+        )
+        output_hidden_states = (
+            output_hidden_states
+            if output_hidden_states is not None
+            else self.config.output_hidden_states
+        )
+        return_dict = (
+            return_dict if return_dict is not None else self.config.use_return_dict
+        )
+
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs = self.model.decoder(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+            head_mask=head_mask,
+            cross_attn_head_mask=cross_attn_head_mask,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        logits = self.lm_head(outputs[0])
+
+        return CausalLMOutputWithCrossAttentions(
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+            cross_attentions=outputs.cross_attentions,
+        )
+
+
+class PPFormulaNet_Head(UniMERNetHead):
+    """
+    PPFormulaNet_Head
+    Args:
+        max_new_tokens (int): Maximum number of new tokens to generate. Default is 1536.
+        decoder_start_token_id (int): Start token ID for the decoder. Default is 0.
+        temperature (float): Temperature parameter for controlling randomness in sampling. Default is 0.2.
+        do_sample (bool): Flag to determine whether to use sampling for generation. Default is False.
+        top_p (float): Top-p (nucleus) sampling parameter for controlling diversity. Default is 0.95.
+        in_channels (int): Number of input channels for the model. Default is 1024.
+        decoder_layers (int): Number of layers in the decoder. Default is 8.
+        encoder_hidden_size (int): Size of the hidden layer in the encoder. Default is 1024.
+        decoder_ffn_dim (int): Dimension of the feed-forward network in the decoder. Default is 4096.
+        decoder_hidden_size (int): Size of the hidden layer in the decoder. Default is 1024.
+        is_export (bool): Flag indicating whether the model is to be exported. Default is False.
+        length_aware (bool): Flag to determine if the model should be aware of input sequence length. Default is True.
+        use_parallel (bool): Flag to enable or disable parallel processing. Default is False.
+        parallel_step (int): Number of steps to use in parallel processing. Default is 3.
+    """
+
+    def __init__(
+        self,
+        max_new_tokens=1536,
+        decoder_start_token_id=0,
+        temperature=0.2,
+        do_sample=False,
+        top_p=0.95,
+        in_channels=1024,
+        decoder_layers=8,
+        encoder_hidden_size=1024,
+        decoder_ffn_dim=4096,
+        decoder_hidden_size=1024,
+        is_export=False,
+        length_aware=True,
+        use_parallel=False,
+        parallel_step=3,
+    ):
+
+        super().__init__()
+
+        mbart_config_dict = {
+            "activation_dropout": 0.0,
+            "activation_function": "gelu",
+            "add_cross_attention": True,
+            "add_final_layer_norm": True,
+            "attention_dropout": 0.0,
+            "bos_token_id": 0,
+            "classifier_dropout": 0.0,
+            "d_model": decoder_hidden_size,
+            "decoder_attention_heads": 16,
+            "decoder_ffn_dim": decoder_ffn_dim,
+            "decoder_layerdrop": 0.0,
+            "decoder_layers": decoder_layers,
+            "dropout": 0.1,
+            "encoder_attention_heads": 16,
+            "encoder_ffn_dim": 4096,
+            "encoder_layerdrop": 0.0,
+            "encoder_layers": 12,
+            "eos_token_id": 2,
+            "forced_eos_token_id": 2,
+            "init_std": 0.02,
+            "is_decoder": True,
+            "is_encoder_decoder": False,
+            "output_hidden_states": False,
+            "max_position_embeddings": (
+                max_new_tokens + parallel_step if use_parallel else max_new_tokens
+            ),
+            "model_type": "mbart",
+            "num_hidden_layers": 12,
+            "pad_token_id": 1,
+            "scale_embedding": True,
+            "tie_word_embeddings": False,
+            "transformers_version": "4.40.0",
+            "use_cache": True,
+            "use_return_dict": True,
+            "vocab_size": 50000,
+            "_attn_implementation": "eager",
+            "hidden_size": decoder_hidden_size,
+            "use_parallel": use_parallel,
+            "parallel_step": int(parallel_step),
+            "is_export": is_export,
+        }
+        self.decoder_start_token_id = decoder_start_token_id
+        self.temperature = temperature
+        self.do_sample = do_sample
+        self.top_p = top_p
+        self.is_export = is_export
+        self.max_seq_len = max_new_tokens
+        self.config_decoder = MBartConfig(**mbart_config_dict)
+        self.encoder_hidden_size = encoder_hidden_size
+        self.decoder = CustomMBartForCausalLM(self.config_decoder)
+        if self.config_decoder.hidden_size != self.encoder_hidden_size:
+            self.enc_to_dec_proj = nn.Linear(
+                self.encoder_hidden_size, self.config_decoder.hidden_size
+            )
+        generation_config = {
+            "max_length": 1537,
+            "forced_eos_token_id": 2,
+        }
+        self.eos_token_id = generation_config["forced_eos_token_id"]
+        self.pad_token_id = self.config_decoder.pad_token_id
+        self.logits_processor = LogitsProcessorList()
+        self.logits_processor.append(
+            ForcedEOSTokenLogitsProcessor(
+                generation_config["max_length"],
+                generation_config["forced_eos_token_id"],
+            )
+        )
+
+    def prepare_inputs_for_generation(
+        self,
+        input_ids,
+        past_key_values=None,
+        attention_mask=None,
+        use_cache=None,
+        encoder_outputs=None,
+        **kwargs,
+    ):
+        decoder_inputs = self.prepare_inputs_for_generation_mbart(
+            input_ids, past_key_values=past_key_values
+        )
+        decoder_attention_mask = (
+            decoder_inputs["attention_mask"]
+            if "attention_mask" in decoder_inputs
+            else None
+        )
+        input_dict = {
+            "attention_mask": attention_mask,
+            "decoder_attention_mask": decoder_attention_mask,
+            "decoder_input_ids": decoder_inputs["input_ids"],
+            "past_key_values": decoder_inputs["past_key_values"],
+            "use_cache": use_cache,
+        }
+        return input_dict
+
+    def _extract_past_from_model_output(
+        self, outputs: ModelOutput, standardize_cache_format: bool = False
+    ):
+        past_key_values = None
+        if "past_key_values" in outputs:
+            past_key_values = outputs.past_key_values
+        elif "mems" in outputs:
+            past_key_values = outputs.mems
+        elif "past_buckets_states" in outputs:
+            past_key_values = outputs.past_buckets_states
+        return past_key_values
+
+    def _update_model_kwargs_for_generation(
+        self,
+        outputs: ModelOutput,
+        model_kwargs: Dict[str, Any],
+        is_encoder_decoder: bool = False,
+        standardize_cache_format: bool = False,
+    ) -> Dict[str, Any]:
+        # update past_key_values
+        model_kwargs["past_key_values"] = self._extract_past_from_model_output(
+            outputs, standardize_cache_format=standardize_cache_format
+        )
+        if getattr(outputs, "state", None) is not None:
+            model_kwargs["state"] = outputs.state
+
+        # update token_type_ids with last value
+        if "token_type_ids" in model_kwargs:
+            token_type_ids = model_kwargs["token_type_ids"]
+            model_kwargs["token_type_ids"] = paddle.concat(
+                [token_type_ids, token_type_ids[:, -1].unsqueeze(-1)], axis=-1
+            )
+
+        if not is_encoder_decoder:
+            # update attention mask
+            if "attention_mask" in model_kwargs:
+                attention_mask = model_kwargs["attention_mask"]
+                model_kwargs["attention_mask"] = paddle.concat(
+                    [
+                        attention_mask,
+                        attention_mask.new_ones((attention_mask.shape[0], 1)),
+                    ],
+                    axis=-1,
+                )
+        else:
+            # update decoder attention mask
+            if "decoder_attention_mask" in model_kwargs:
+                decoder_attention_mask = model_kwargs["decoder_attention_mask"]
+                model_kwargs["decoder_attention_mask"] = paddle.concat(
+                    [
+                        decoder_attention_mask,
+                        decoder_attention_mask.new_ones(
+                            (decoder_attention_mask.shape[0], 1)
+                        ),
+                    ],
+                    dim=-1,
+                )
+
+        if (
+            "cache_position" in model_kwargs
+            and model_kwargs["cache_position"] is not None
+        ):
+            model_kwargs["cache_position"] = model_kwargs["cache_position"][-1:] + 1
+        return model_kwargs
+
+    def stopping_criteria(self, input_ids):
+        if self.is_export:
+            return input_ids[:, -1] == paddle.to_tensor([self.eos_token_id])
+        is_done = paddle.isin(input_ids[:, -1], paddle.to_tensor([self.eos_token_id]))
+        return is_done
+
+    def stopping_criteria_parallel(self, input_ids):
+        parallel_step = self.config_decoder.parallel_step
+
+        if self.is_export:
+            is_done_list = []
+            for i in range(parallel_step, 0, -1):
+                cur_is_done = input_ids[:, -i] == paddle.to_tensor([self.eos_token_id])
+                is_done_list.append(cur_is_done)
+            is_done_list = paddle.to_tensor(is_done_list).transpose([1, 0])
+            return is_done_list
+        else:
+            is_done = paddle.isin(
+                input_ids[:, -parallel_step:],
+                paddle.to_tensor([self.eos_token_id]).reshape([1, 1]),
+            )
+            return paddle.to_tensor(is_done)
+
+    def generate_single_iter(
+        self,
+        decoder_input_ids=None,
+        decoder_attention_mask=None,
+        encoder_outputs=None,
+        past_key_values=None,
+        decoder_inputs_embeds=None,
+        labels=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+        **kwargs,
+    ):
+
+        encoder_hidden_states = encoder_outputs[0]
+        if self.config_decoder.hidden_size != self.encoder_hidden_size:
+            encoder_hidden_states = self.enc_to_dec_proj(encoder_hidden_states)
+        kwargs_decoder = {}
+        decoder_outputs = self.decoder(
+            input_ids=decoder_input_ids,
+            attention_mask=decoder_attention_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=None,
+            inputs_embeds=None,
+            output_attentions=False,
+            output_hidden_states=output_hidden_states,
+            use_cache=use_cache,
+            past_key_values=past_key_values,
+            return_dict=return_dict,
+            **kwargs_decoder,
+        )
+
+        return Seq2SeqLMOutput(
+            loss=None,
+            logits=decoder_outputs.logits,
+            past_key_values=decoder_outputs.past_key_values,
+            decoder_hidden_states=decoder_outputs.hidden_states,
+            decoder_attentions=decoder_outputs.attentions,
+            cross_attentions=decoder_outputs.cross_attentions,
+            encoder_last_hidden_state=encoder_outputs.last_hidden_state,
+            encoder_hidden_states=encoder_outputs.hidden_states,
+            encoder_attentions=encoder_outputs.attentions,
+        )
+
+    def _prepare_decoder_input_ids_for_generation(
+        self,
+        batch_size,
+        model_kwargs,
+        decoder_start_token_id=None,
+        bos_token_id=None,
+    ):
+
+        # 1. Check whether the user has defined `decoder_input_ids` manually. To facilitate in terms of input naming,
+        # we also allow the user to pass it under `input_ids`, if the encoder does not use it as the main input.
+        if model_kwargs is not None and "decoder_input_ids" in model_kwargs:
+            decoder_input_ids = model_kwargs.pop("decoder_input_ids")
+        elif "input_ids" in model_kwargs:
+            decoder_input_ids = model_kwargs.pop("input_ids")
+        else:
+            decoder_input_ids = None
+
+        # 2. Encoder-decoder models expect the `decoder_input_ids` to start with a special token. Let's ensure that.
+        decoder_start_token_id = self._get_decoder_start_token_id(
+            decoder_start_token_id, bos_token_id
+        )
+
+        if isinstance(decoder_start_token_id, list):
+            if len(decoder_start_token_id) != batch_size:
+                raise ValueError(
+                    f"`decoder_start_token_id` expected to have length {batch_size} but got {len(decoder_start_token_id)}"
+                )
+            decoder_input_ids_start = paddle.to_tensor(
+                decoder_start_token_id,
+                dtype=paddle.int64,
+            )
+            decoder_input_ids_start = decoder_input_ids_start.view(-1, 1)
+        else:
+            use_parallel = self.config_decoder.use_parallel
+            parallel_step = self.config_decoder.parallel_step
+
+            if use_parallel:
+                decoder_input_ids_start = (
+                    paddle.ones(
+                        (batch_size, parallel_step),
+                        dtype=paddle.int64,
+                    )
+                    * decoder_start_token_id
+                )
+            else:
+                decoder_input_ids_start = (
+                    paddle.ones(
+                        (batch_size, 1),
+                        dtype=paddle.int64,
+                    )
+                    * decoder_start_token_id
+                )
+        # no user input -> use decoder_start_token_id as decoder_input_ids
+        if decoder_input_ids is None:
+            decoder_input_ids = decoder_input_ids_start
+        # exception: Donut checkpoints have task-specific decoder starts and don't expect a BOS token
+        elif (
+            self.config.model_type == "vision-encoder-decoder"
+            and "donut" in self.name_or_path.lower()
+        ):
+            pass
+        elif self.config.model_type in ["whisper"]:
+            pass
+        # user input but doesn't start with decoder_start_token_id -> prepend decoder_start_token_id (and adjust
+        # decoder_attention_mask if provided)
+        elif (
+            isinstance(decoder_start_token_id, int)
+            and (decoder_input_ids[:, 0] != decoder_start_token_id).all().item()
+        ) or (
+            isinstance(decoder_start_token_id, paddle.Tensor)
+            and (decoder_input_ids[:, 0] != decoder_start_token_id[:, 0]).all().item()
+        ):
+            decoder_input_ids = paddle.concat(
+                [decoder_input_ids_start, decoder_input_ids], axis=-1
+            )
+            if "decoder_attention_mask" in model_kwargs:
+                decoder_attention_mask = model_kwargs["decoder_attention_mask"]
+                decoder_attention_mask = paddle.cat(
+                    (
+                        paddle.ones_like(decoder_attention_mask)[:, :1],
+                        decoder_attention_mask,
+                    ),
+                    dim=-1,
+                )
+                model_kwargs["decoder_attention_mask"] = decoder_attention_mask
+
+        return decoder_input_ids, model_kwargs
+
+    @paddle.no_grad()
+    def generate_export(
+        self,
+        encoder_outputs,
+        model_kwargs,
+    ):
+        use_parallel = self.config_decoder.use_parallel
+        parallel_step = self.config_decoder.parallel_step
+        batch_size = encoder_outputs["last_hidden_state"].shape[0]
+        generation_config = {
+            "decoder_start_token_id": 0,
+            "bos_token_id": 0,
+        }
+        input_ids, model_kwargs = self._prepare_decoder_input_ids_for_generation(
+            batch_size=batch_size,
+            model_kwargs=model_kwargs,
+            decoder_start_token_id=generation_config["decoder_start_token_id"],
+            bos_token_id=generation_config["bos_token_id"],
+        )
+        if not use_parallel:
+            input_ids = input_ids.reshape([-1, 1])
+        decoder_input_ids = input_ids
+        model_kwargs["key use_cache"] = True
+        batch_size, cur_len = input_ids.shape
+
+        if "inputs_embeds" in model_kwargs:
+            cur_len = model_kwargs["inputs_embeds"].shape[1]
+
+        cache_position = paddle.arange(cur_len)
+        pad_token_id = self.pad_token_id
+        eos_token_id = [self.eos_token_id]
+        eos_token = self.eos_token_id
+        if use_parallel:
+            unfinished_sequences = paddle.ones(
+                [batch_size, parallel_step], dtype=paddle.int64
+            )
+            parallel_length = math.ceil(self.max_seq_len // parallel_step)
+        else:
+            unfinished_sequences = paddle.ones(batch_size, dtype=paddle.int64)
+            parallel_length = self.max_seq_len
+
+        i_idx = paddle.full([], 0)
+        past_key_values = []
+        decoder_attention_heads = self.config_decoder.decoder_attention_heads
+        decoder_attention_heads_dim = int(
+            self.config_decoder.d_model / decoder_attention_heads
+        )
+        for i in range(self.config_decoder.decoder_layers):
+            init_arr = paddle.zeros(
+                [batch_size, decoder_attention_heads, 0, decoder_attention_heads_dim]
+            )
+            paddle.jit.api.set_dynamic_shape(init_arr, [-1, -1, -1, -1])
+            cache = (init_arr, init_arr, init_arr, init_arr)
+            past_key_values.append(cache)
+
+        while i_idx < paddle.to_tensor(parallel_length):
+
+            model_inputs = self.prepare_inputs_for_generation_export(
+                past_key_values=past_key_values, **model_kwargs
+            )
+            decoder_attention_mask = paddle.ones(paddle.shape(input_ids))
+            paddle.jit.api.set_dynamic_shape(decoder_input_ids, [-1, -1])
+            paddle.jit.api.set_dynamic_shape(decoder_attention_mask, [-1, -1])
+
+            outputs = self.generate_single_iter(
+                decoder_input_ids=decoder_input_ids,
+                decoder_attention_mask=decoder_attention_mask,
+                encoder_outputs=encoder_outputs,
+                past_key_values=past_key_values,
+                return_dict=True,
+                output_attentions=False,
+                output_hidden_states=False,
+            )
+
+            if use_parallel:
+                next_token_logits = outputs.logits[:, -parallel_step:, :]
+            else:
+                next_token_logits = outputs.logits[:, -1, :]
+            next_tokens_scores = self.logits_processor(input_ids, next_token_logits)
+            next_tokens = paddle.argmax(next_tokens_scores, axis=-1)
+
+            if eos_token_id is not None:
+                # False
+                if pad_token_id is None:
+                    raise ValueError(
+                        "If `eos_token_id` is defined, make sure that `pad_token_id` is defined."
+                    )
+                next_tokens = next_tokens * unfinished_sequences + pad_token_id * (
+                    1 - unfinished_sequences
+                )
+            if use_parallel:
+                input_ids = paddle.concat([input_ids, next_tokens], axis=-1)
+                decoder_input_ids = next_tokens
+            else:
+                input_ids = paddle.concat(
+                    [input_ids, next_tokens.unsqueeze(1)], axis=-1
+                )
+                decoder_input_ids = next_tokens.unsqueeze(1)
+
+            past_length = past_key_values[0][0].shape[2]
+
+            past_key_values = outputs.past_key_values
+            cache_position = cache_position[-1:] + 1
+            if use_parallel:
+                unfinished_sequences = (
+                    unfinished_sequences
+                    & ~self.stopping_criteria_parallel(input_ids).cast(paddle.int64)
+                )
+            else:
+                unfinished_sequences = unfinished_sequences & ~self.stopping_criteria(
+                    input_ids
+                ).cast(paddle.int64)
+
+            if (
+                eos_token is not None
+                and (
+                    paddle.cumsum((input_ids == eos_token).cast(paddle.int64), 1)[:, -1]
+                    >= 1
+                ).all()
+            ):
+                break
+            i_idx += 1
+            # break
+
+        return input_ids
+
+    @paddle.no_grad()
+    def generate(
+        self,
+        encoder_outputs,
+        model_kwargs,
+    ):
+        """
+        Generate sequences from the model without computing gradients.
+
+        This method is used to generate sequences from the model based on the given encoder outputs.
+        It does not compute gradients, making it suitable for inference.
+
+        Args:
+            encoder_outputs: The outputs from the encoder, typically including hidden states necessary for generation.
+            model_kwargs: Additional keyword arguments that may include parameters such as maximum length,
+                        temperature, top-k/top-p sampling parameters, and other generation-specific settings.
+
+        Returns:
+            Generated sequences based on the encoder outputs and specified generation parameters.
+        """
+        use_parallel = self.config_decoder.use_parallel
+        parallel_step = self.config_decoder.parallel_step
+        batch_size = encoder_outputs["last_hidden_state"].shape[0]
+        generation_config = {
+            "decoder_start_token_id": 0,
+            "bos_token_id": 0,
+        }
+
+        input_ids, model_kwargs = self._prepare_decoder_input_ids_for_generation(
+            batch_size=batch_size,
+            model_kwargs=model_kwargs,
+            decoder_start_token_id=generation_config["decoder_start_token_id"],
+            bos_token_id=generation_config["bos_token_id"],
+        )
+
+        decoder_input_ids = input_ids
+        model_kwargs["key use_cache"] = True
+        batch_size, cur_len = input_ids.shape
+
+        if "inputs_embeds" in model_kwargs:
+            cur_len = model_kwargs["inputs_embeds"].shape[1]
+        model_kwargs["cache_position"] = paddle.arange(cur_len)
+        pad_token_id = self.pad_token_id
+        eos_token_id = [self.eos_token_id]
+        eos_token = self.eos_token_id
+        if use_parallel:
+            unfinished_sequences = paddle.ones(
+                [batch_size, parallel_step], dtype=paddle.int64
+            )
+            parallel_length = math.ceil(self.max_seq_len // parallel_step)
+        else:
+            unfinished_sequences = paddle.ones(batch_size, dtype=paddle.int64)
+            parallel_length = self.max_seq_len
+        past_key_values = []
+
+        for idx in range(parallel_length):
+
+            model_inputs = self.prepare_inputs_for_generation(input_ids, **model_kwargs)
+            outputs = self.generate_single_iter(
+                **model_inputs,
+                encoder_outputs=encoder_outputs,
+                return_dict=True,
+                output_attentions=False,
+                output_hidden_states=False,
+            )
+
+            if use_parallel:
+                next_token_logits = outputs.logits[:, :, :]
+            else:
+                next_token_logits = outputs.logits[:, -1, :]
+
+            next_tokens_scores = self.logits_processor(input_ids, next_token_logits)
+            next_tokens = paddle.argmax(next_tokens_scores, axis=-1)
+            if eos_token_id is not None:
+                # False
+                if pad_token_id is None:
+                    raise ValueError(
+                        "If `eos_token_id` is defined, make sure that `pad_token_id` is defined."
+                    )
+                next_tokens = next_tokens * unfinished_sequences + pad_token_id * (
+                    1 - unfinished_sequences
+                )
+            if use_parallel:
+                input_ids = paddle.concat([input_ids, next_tokens], axis=-1)
+            else:
+                input_ids = paddle.concat([input_ids, next_tokens[:, None]], axis=-1)
+
+            model_kwargs = self._update_model_kwargs_for_generation(
+                outputs,
+                model_kwargs,
+                is_encoder_decoder=self.config_decoder.is_encoder_decoder,
+            )
+            if use_parallel:
+                unfinished_sequences = (
+                    unfinished_sequences
+                    & ~self.stopping_criteria_parallel(input_ids).cast(paddle.int64)
+                )
+            else:
+                unfinished_sequences = unfinished_sequences & ~self.stopping_criteria(
+                    input_ids
+                ).cast(paddle.int64)
+
+            if (
+                eos_token is not None
+                and (
+                    paddle.cumsum((input_ids == eos_token).cast(paddle.int64), 1)[:, -1]
+                    >= 1
+                ).all()
+            ):
+                break
+        return input_ids
+
+    def forwad_train(
+        self,
+        encoder_outputs,
+        decoder_input_ids,
+        decoder_attention_mask,
+        past_key_values=None,
+        decoder_inputs_embeds=None,
+        labels=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+        **kwargs,
+    ):
+        """
+        Forward pass for training the model.
+
+        Args:
+            encoder_outputs: The outputs from the encoder, typically including hidden states.
+            decoder_input_ids: Input IDs for the decoder.
+            decoder_attention_mask: Attention mask for the decoder inputs to avoid attending to padding tokens.
+            past_key_values: Previously computed key and value states for the decoder, used for fast generation.
+            decoder_inputs_embeds: Optional embeddings for decoder inputs, used instead of decoder_input_ids if provided.
+            labels: Labels for computing the training loss.
+            use_cache: Whether to use a cache of past key values for faster generation.
+            output_attentions: Whether to output attention weights.
+            output_hidden_states: Whether to output hidden states of all layers.
+            return_dict: Whether to return the output as a dictionary.
+            **kwargs: Additional keyword arguments.
+
+        Returns:
+            Depending on the `return_dict` flag, returns either a dictionary of model outputs or a tuple.
+        """
+        if self.config_decoder.use_parallel:
+            batch = decoder_input_ids.shape[0]
+            add_sos_token = self.config_decoder.parallel_step - 1
+            start_token = paddle.zeros([batch, add_sos_token]).cast(paddle.int64)
+            start_mask = paddle.ones([batch, add_sos_token]).cast(paddle.int64)
+            decoder_input_ids = paddle.concat([start_token, decoder_input_ids], axis=1)
+            decoder_attention_mask = paddle.concat(
+                [start_mask, decoder_attention_mask], axis=1
+            )
+
+        labels = decoder_input_ids * 1
+        labels = labels.masked_fill_(labels == self.pad_token_id, -100)
+        if self.config_decoder.use_parallel:
+            input_decoder_input_ids = decoder_input_ids[
+                :, : -self.config_decoder.parallel_step
+            ]
+            input_decoder_attention_mask = decoder_attention_mask[
+                :, : -self.config_decoder.parallel_step
+            ]
+        else:
+            input_decoder_input_ids = decoder_input_ids[:, :-1]
+            input_decoder_attention_mask = decoder_attention_mask[:, :-1]
+
+        encoder_hidden_states = encoder_outputs[0]
+        kwargs_decoder = {}
+        if self.config_decoder.hidden_size != self.encoder_hidden_size:
+            encoder_hidden_states = self.enc_to_dec_proj(encoder_hidden_states)
+
+        decoder_outputs = self.decoder(
+            input_ids=input_decoder_input_ids,
+            attention_mask=input_decoder_attention_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=None,
+            inputs_embeds=None,
+            output_attentions=False,
+            output_hidden_states=output_hidden_states,
+            use_cache=use_cache,
+            past_key_values=past_key_values,
+            return_dict=return_dict,
+            **kwargs_decoder,
+        )
+
+        logits = decoder_outputs.logits
+        return logits, labels
+
+    # forward for export
+    def forward(self, inputs, targets=None):
+        self.is_export = False if self.training else True
+        if not self.training:
+            encoder_outputs = inputs
+            model_kwargs = {
+                "output_attentions": False,
+                "output_hidden_states": False,
+                "use_cache": True,
+            }
+            if self.is_export:
+                word_pred = self.generate_export(encoder_outputs, model_kwargs)
+            else:
+                word_pred = self.generate(encoder_outputs, model_kwargs)
+
+            return word_pred
+        encoder_outputs, tgt_seq, mask = inputs
+        logits, masked_labels = self.forwad_train(encoder_outputs, tgt_seq, mask)
+
+        return logits, masked_labels
diff --git a/ppocr/modeling/heads/rec_pren_head.py b/ppocr/modeling/heads/rec_pren_head.py
new file mode 100644
index 0000000..c9e4b3e
--- /dev/null
+++ b/ppocr/modeling/heads/rec_pren_head.py
@@ -0,0 +1,34 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+from paddle import nn
+from paddle.nn import functional as F
+
+
+class PRENHead(nn.Layer):
+    def __init__(self, in_channels, out_channels, **kwargs):
+        super(PRENHead, self).__init__()
+        self.linear = nn.Linear(in_channels, out_channels)
+
+    def forward(self, x, targets=None):
+        predicts = self.linear(x)
+
+        if not self.training:
+            predicts = F.softmax(predicts, axis=2)
+
+        return predicts
diff --git a/ppocr/modeling/heads/rec_rfl_head.py b/ppocr/modeling/heads/rec_rfl_head.py
new file mode 100644
index 0000000..bd7efd4
--- /dev/null
+++ b/ppocr/modeling/heads/rec_rfl_head.py
@@ -0,0 +1,106 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/hikopensource/DAVAR-Lab-OCR/blob/main/davarocr/davar_rcg/models/sequence_heads/counting_head.py
+"""
+import paddle
+import paddle.nn as nn
+from paddle.nn.initializer import TruncatedNormal, Constant, Normal, KaimingNormal
+
+from .rec_att_head import AttentionLSTM
+
+kaiming_init_ = KaimingNormal()
+zeros_ = Constant(value=0.0)
+ones_ = Constant(value=1.0)
+
+
+class CNTHead(nn.Layer):
+    def __init__(self, embed_size=512, encode_length=26, out_channels=38, **kwargs):
+        super(CNTHead, self).__init__()
+
+        self.out_channels = out_channels
+
+        self.Wv_fusion = nn.Linear(embed_size, embed_size, bias_attr=False)
+        self.Prediction_visual = nn.Linear(
+            encode_length * embed_size, self.out_channels
+        )
+
+    def forward(self, visual_feature):
+        b, c, h, w = visual_feature.shape
+        visual_feature = visual_feature.reshape([b, c, h * w]).transpose([0, 2, 1])
+        visual_feature_num = self.Wv_fusion(visual_feature)  # batch * 26 * 512
+        b, n, c = visual_feature_num.shape
+        # using visual feature directly calculate the text length
+        visual_feature_num = visual_feature_num.reshape([b, n * c])
+        prediction_visual = self.Prediction_visual(visual_feature_num)
+
+        return prediction_visual
+
+
+class RFLHead(nn.Layer):
+    def __init__(
+        self,
+        in_channels=512,
+        hidden_size=256,
+        batch_max_legnth=25,
+        out_channels=38,
+        use_cnt=True,
+        use_seq=True,
+        **kwargs,
+    ):
+        super(RFLHead, self).__init__()
+        assert use_cnt or use_seq
+        self.use_cnt = use_cnt
+        self.use_seq = use_seq
+        if self.use_cnt:
+            self.cnt_head = CNTHead(
+                embed_size=in_channels,
+                encode_length=batch_max_legnth + 1,
+                out_channels=out_channels,
+                **kwargs,
+            )
+        if self.use_seq:
+            self.seq_head = AttentionLSTM(
+                in_channels=in_channels,
+                out_channels=out_channels,
+                hidden_size=hidden_size,
+                **kwargs,
+            )
+        self.batch_max_legnth = batch_max_legnth
+        self.num_class = out_channels
+        self.apply(self.init_weights)
+
+    def init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            kaiming_init_(m.weight)
+            if isinstance(m, nn.Linear) and m.bias is not None:
+                zeros_(m.bias)
+
+    def forward(self, x, targets=None):
+        cnt_inputs, seq_inputs = x
+        if self.use_cnt:
+            cnt_outputs = self.cnt_head(cnt_inputs)
+        else:
+            cnt_outputs = None
+        if self.use_seq:
+            if self.training:
+                seq_outputs = self.seq_head(
+                    seq_inputs, targets[0], self.batch_max_legnth
+                )
+            else:
+                seq_outputs = self.seq_head(seq_inputs, None, self.batch_max_legnth)
+            return cnt_outputs, seq_outputs
+        else:
+            return cnt_outputs
diff --git a/ppocr/modeling/heads/rec_robustscanner_head.py b/ppocr/modeling/heads/rec_robustscanner_head.py
new file mode 100644
index 0000000..5510ee8
--- /dev/null
+++ b/ppocr/modeling/heads/rec_robustscanner_head.py
@@ -0,0 +1,748 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textrecog/encoders/channel_reduction_encoder.py
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textrecog/decoders/robust_scanner_decoder.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+from paddle import ParamAttr
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+
+class BaseDecoder(nn.Layer):
+    def __init__(self, **kwargs):
+        super().__init__()
+
+    def forward_train(self, feat, out_enc, targets, img_metas):
+        raise NotImplementedError
+
+    def forward_test(self, feat, out_enc, img_metas):
+        raise NotImplementedError
+
+    def forward(
+        self,
+        feat,
+        out_enc,
+        label=None,
+        valid_ratios=None,
+        word_positions=None,
+        train_mode=True,
+    ):
+        self.train_mode = train_mode
+
+        if train_mode:
+            return self.forward_train(
+                feat, out_enc, label, valid_ratios, word_positions
+            )
+        return self.forward_test(feat, out_enc, valid_ratios, word_positions)
+
+
+class ChannelReductionEncoder(nn.Layer):
+    """Change the channel number with a one by one convoluational layer.
+
+    Args:
+        in_channels (int): Number of input channels.
+        out_channels (int): Number of output channels.
+    """
+
+    def __init__(self, in_channels, out_channels, **kwargs):
+        super(ChannelReductionEncoder, self).__init__()
+
+        self.layer = nn.Conv2D(
+            in_channels,
+            out_channels,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            weight_attr=nn.initializer.XavierNormal(),
+        )
+
+    def forward(self, feat):
+        """
+        Args:
+            feat (Tensor): Image features with the shape of
+                :math:`(N, C_{in}, H, W)`.
+
+        Returns:
+            Tensor: A tensor of shape :math:`(N, C_{out}, H, W)`.
+        """
+        return self.layer(feat)
+
+
+def masked_fill(x, mask, value):
+    y = paddle.full(x.shape, value, x.dtype)
+    return paddle.where(mask, y, x)
+
+
+class DotProductAttentionLayer(nn.Layer):
+    def __init__(self, dim_model=None):
+        super().__init__()
+
+        self.scale = dim_model**-0.5 if dim_model is not None else 1.0
+
+    def forward(self, query, key, value, h, w, valid_ratios=None):
+        query = paddle.transpose(query, (0, 2, 1))
+        logits = paddle.matmul(query, key) * self.scale
+        n, c, t = logits.shape
+        # reshape to (n, c, h, w)
+        logits = paddle.reshape(logits, [n, c, h, w])
+        if valid_ratios is not None:
+            # cal mask of attention weight
+            with paddle.base.framework._stride_in_no_check_dy2st_diff():
+                for i, valid_ratio in enumerate(valid_ratios):
+                    valid_width = min(w, int(w * valid_ratio + 0.5))
+                    if valid_width < w:
+                        logits[i, :, :, valid_width:] = float("-inf")
+
+        # reshape to (n, c, h, w)
+        logits = paddle.reshape(logits, [n, c, t])
+        weights = F.softmax(logits, axis=2)
+        value = paddle.transpose(value, (0, 2, 1))
+        glimpse = paddle.matmul(weights, value)
+        glimpse = paddle.transpose(glimpse, (0, 2, 1))
+        return glimpse
+
+
+class SequenceAttentionDecoder(BaseDecoder):
+    """Sequence attention decoder for RobustScanner.
+
+    RobustScanner: `RobustScanner: Dynamically Enhancing Positional Clues for
+    Robust Text Recognition <https://arxiv.org/abs/2007.07542>`_
+
+    Args:
+        num_classes (int): Number of output classes :math:`C`.
+        rnn_layers (int): Number of RNN layers.
+        dim_input (int): Dimension :math:`D_i` of input vector ``feat``.
+        dim_model (int): Dimension :math:`D_m` of the model. Should also be the
+            same as encoder output vector ``out_enc``.
+        max_seq_len (int): Maximum output sequence length :math:`T`.
+        start_idx (int): The index of `<SOS>`.
+        mask (bool): Whether to mask input features according to
+            ``img_meta['valid_ratio']``.
+        padding_idx (int): The index of `<PAD>`.
+        dropout (float): Dropout rate.
+        return_feature (bool): Return feature or logits as the result.
+        encode_value (bool): Whether to use the output of encoder ``out_enc``
+            as `value` of attention layer. If False, the original feature
+            ``feat`` will be used.
+
+    Warning:
+        This decoder will not predict the final class which is assumed to be
+        `<PAD>`. Therefore, its output size is always :math:`C - 1`. `<PAD>`
+        is also ignored by loss as specified in
+        :obj:`mmocr.models.textrecog.recognizer.EncodeDecodeRecognizer`.
+    """
+
+    def __init__(
+        self,
+        num_classes=None,
+        rnn_layers=2,
+        dim_input=512,
+        dim_model=128,
+        max_seq_len=40,
+        start_idx=0,
+        mask=True,
+        padding_idx=None,
+        dropout=0,
+        return_feature=False,
+        encode_value=False,
+    ):
+        super().__init__()
+
+        self.num_classes = num_classes
+        self.dim_input = dim_input
+        self.dim_model = dim_model
+        self.return_feature = return_feature
+        self.encode_value = encode_value
+        self.max_seq_len = max_seq_len
+        self.start_idx = start_idx
+        self.mask = mask
+
+        self.embedding = nn.Embedding(
+            self.num_classes, self.dim_model, padding_idx=padding_idx
+        )
+
+        self.sequence_layer = nn.LSTM(
+            input_size=dim_model,
+            hidden_size=dim_model,
+            num_layers=rnn_layers,
+            time_major=False,
+            dropout=dropout,
+        )
+
+        self.attention_layer = DotProductAttentionLayer()
+
+        self.prediction = None
+        if not self.return_feature:
+            pred_num_classes = num_classes - 1
+            self.prediction = nn.Linear(
+                dim_model if encode_value else dim_input, pred_num_classes
+            )
+
+    def forward_train(self, feat, out_enc, targets, valid_ratios):
+        """
+        Args:
+            feat (Tensor): Tensor of shape :math:`(N, D_i, H, W)`.
+            out_enc (Tensor): Encoder output of shape
+                :math:`(N, D_m, H, W)`.
+            targets (Tensor): a tensor of shape :math:`(N, T)`. Each element is the index of a
+                character.
+            valid_ratios (Tensor): valid length ratio of img.
+        Returns:
+            Tensor: A raw logit tensor of shape :math:`(N, T, C-1)` if
+            ``return_feature=False``. Otherwise it would be the hidden feature
+            before the prediction projection layer, whose shape is
+            :math:`(N, T, D_m)`.
+        """
+
+        tgt_embedding = self.embedding(targets)
+
+        n, c_enc, h, w = out_enc.shape
+        assert c_enc == self.dim_model
+        _, c_feat, _, _ = feat.shape
+        assert c_feat == self.dim_input
+        _, len_q, c_q = tgt_embedding.shape
+        assert c_q == self.dim_model
+        assert len_q <= self.max_seq_len
+
+        query, _ = self.sequence_layer(tgt_embedding)
+        query = paddle.transpose(query, (0, 2, 1))
+        key = paddle.reshape(out_enc, [n, c_enc, h * w])
+        if self.encode_value:
+            value = key
+        else:
+            value = paddle.reshape(feat, [n, c_feat, h * w])
+
+        attn_out = self.attention_layer(query, key, value, h, w, valid_ratios)
+        attn_out = paddle.transpose(attn_out, (0, 2, 1))
+
+        if self.return_feature:
+            return attn_out
+
+        out = self.prediction(attn_out)
+
+        return out
+
+    def forward_test(self, feat, out_enc, valid_ratios):
+        """
+        Args:
+            feat (Tensor): Tensor of shape :math:`(N, D_i, H, W)`.
+            out_enc (Tensor): Encoder output of shape
+                :math:`(N, D_m, H, W)`.
+            valid_ratios (Tensor): valid length ratio of img.
+
+        Returns:
+            Tensor: The output logit sequence tensor of shape
+            :math:`(N, T, C-1)`.
+        """
+        seq_len = self.max_seq_len
+        batch_size = feat.shape[0]
+
+        decode_sequence = (
+            paddle.ones((batch_size, seq_len), dtype="int64") * self.start_idx
+        )
+
+        outputs = []
+        for i in range(seq_len):
+            step_out = self.forward_test_step(
+                feat, out_enc, decode_sequence, i, valid_ratios
+            )
+            outputs.append(step_out)
+            max_idx = paddle.argmax(step_out, axis=1, keepdim=False)
+            if i < seq_len - 1:
+                decode_sequence[:, i + 1] = max_idx
+
+        outputs = paddle.stack(outputs, 1)
+
+        return outputs
+
+    def forward_test_step(
+        self, feat, out_enc, decode_sequence, current_step, valid_ratios
+    ):
+        """
+        Args:
+            feat (Tensor): Tensor of shape :math:`(N, D_i, H, W)`.
+            out_enc (Tensor): Encoder output of shape
+                :math:`(N, D_m, H, W)`.
+            decode_sequence (Tensor): Shape :math:`(N, T)`. The tensor that
+                stores history decoding result.
+            current_step (int): Current decoding step.
+            valid_ratios (Tensor): valid length ratio of img
+
+        Returns:
+            Tensor: Shape :math:`(N, C-1)`. The logit tensor of predicted
+            tokens at current time step.
+        """
+
+        embed = self.embedding(decode_sequence)
+
+        n, c_enc, h, w = out_enc.shape
+        assert c_enc == self.dim_model
+        _, c_feat, _, _ = feat.shape
+        assert c_feat == self.dim_input
+        _, _, c_q = embed.shape
+        assert c_q == self.dim_model
+
+        query, _ = self.sequence_layer(embed)
+        query = paddle.transpose(query, (0, 2, 1))
+        key = paddle.reshape(out_enc, [n, c_enc, h * w])
+        if self.encode_value:
+            value = key
+        else:
+            value = paddle.reshape(feat, [n, c_feat, h * w])
+
+        # [n, c, l]
+        attn_out = self.attention_layer(query, key, value, h, w, valid_ratios)
+        out = attn_out[:, :, current_step]
+
+        if self.return_feature:
+            return out
+
+        out = self.prediction(out)
+        out = F.softmax(out, dim=-1)
+
+        return out
+
+
+class PositionAwareLayer(nn.Layer):
+    def __init__(self, dim_model, rnn_layers=2):
+        super().__init__()
+
+        self.dim_model = dim_model
+
+        self.rnn = nn.LSTM(
+            input_size=dim_model,
+            hidden_size=dim_model,
+            num_layers=rnn_layers,
+            time_major=False,
+        )
+
+        self.mixer = nn.Sequential(
+            nn.Conv2D(dim_model, dim_model, kernel_size=3, stride=1, padding=1),
+            nn.ReLU(),
+            nn.Conv2D(dim_model, dim_model, kernel_size=3, stride=1, padding=1),
+        )
+
+    def forward(self, img_feature):
+        n, c, h, w = img_feature.shape
+        rnn_input = paddle.transpose(img_feature, (0, 2, 3, 1))
+        rnn_input = paddle.reshape(rnn_input, (n * h, w, c))
+        rnn_output, _ = self.rnn(rnn_input)
+        rnn_output = paddle.reshape(rnn_output, (n, h, w, c))
+        rnn_output = paddle.transpose(rnn_output, (0, 3, 1, 2))
+        out = self.mixer(rnn_output)
+        return out
+
+
+class PositionAttentionDecoder(BaseDecoder):
+    """Position attention decoder for RobustScanner.
+
+    RobustScanner: `RobustScanner: Dynamically Enhancing Positional Clues for
+    Robust Text Recognition <https://arxiv.org/abs/2007.07542>`_
+
+    Args:
+        num_classes (int): Number of output classes :math:`C`.
+        rnn_layers (int): Number of RNN layers.
+        dim_input (int): Dimension :math:`D_i` of input vector ``feat``.
+        dim_model (int): Dimension :math:`D_m` of the model. Should also be the
+            same as encoder output vector ``out_enc``.
+        max_seq_len (int): Maximum output sequence length :math:`T`.
+        mask (bool): Whether to mask input features according to
+            ``img_meta['valid_ratio']``.
+        return_feature (bool): Return feature or logits as the result.
+        encode_value (bool): Whether to use the output of encoder ``out_enc``
+            as `value` of attention layer. If False, the original feature
+            ``feat`` will be used.
+
+    Warning:
+        This decoder will not predict the final class which is assumed to be
+        `<PAD>`. Therefore, its output size is always :math:`C - 1`. `<PAD>`
+        is also ignored by loss
+
+    """
+
+    def __init__(
+        self,
+        num_classes=None,
+        rnn_layers=2,
+        dim_input=512,
+        dim_model=128,
+        max_seq_len=40,
+        mask=True,
+        return_feature=False,
+        encode_value=False,
+    ):
+        super().__init__()
+
+        self.num_classes = num_classes
+        self.dim_input = dim_input
+        self.dim_model = dim_model
+        self.max_seq_len = max_seq_len
+        self.return_feature = return_feature
+        self.encode_value = encode_value
+        self.mask = mask
+
+        self.embedding = nn.Embedding(self.max_seq_len + 1, self.dim_model)
+
+        self.position_aware_module = PositionAwareLayer(self.dim_model, rnn_layers)
+
+        self.attention_layer = DotProductAttentionLayer()
+
+        self.prediction = None
+        if not self.return_feature:
+            pred_num_classes = num_classes - 1
+            self.prediction = nn.Linear(
+                dim_model if encode_value else dim_input, pred_num_classes
+            )
+
+    def _get_position_index(self, length, batch_size):
+        position_index_list = []
+        for i in range(batch_size):
+            position_index = paddle.arange(0, end=length, step=1, dtype="int64")
+            position_index_list.append(position_index)
+        batch_position_index = paddle.stack(position_index_list, axis=0)
+        return batch_position_index
+
+    def forward_train(self, feat, out_enc, targets, valid_ratios, position_index):
+        """
+        Args:
+            feat (Tensor): Tensor of shape :math:`(N, D_i, H, W)`.
+            out_enc (Tensor): Encoder output of shape
+                :math:`(N, D_m, H, W)`.
+            targets (dict): A dict with the key ``padded_targets``, a
+                tensor of shape :math:`(N, T)`. Each element is the index of a
+                character.
+            valid_ratios (Tensor): valid length ratio of img.
+            position_index (Tensor): The position of each word.
+
+        Returns:
+            Tensor: A raw logit tensor of shape :math:`(N, T, C-1)` if
+            ``return_feature=False``. Otherwise it will be the hidden feature
+            before the prediction projection layer, whose shape is
+            :math:`(N, T, D_m)`.
+        """
+        n, c_enc, h, w = out_enc.shape
+        assert c_enc == self.dim_model
+        _, c_feat, _, _ = feat.shape
+        assert c_feat == self.dim_input
+        _, len_q = targets.shape
+        assert len_q <= self.max_seq_len
+
+        position_out_enc = self.position_aware_module(out_enc)
+
+        query = self.embedding(position_index)
+        query = paddle.transpose(query, (0, 2, 1))
+        key = paddle.reshape(position_out_enc, (n, c_enc, h * w))
+        if self.encode_value:
+            value = paddle.reshape(out_enc, (n, c_enc, h * w))
+        else:
+            value = paddle.reshape(feat, (n, c_feat, h * w))
+
+        attn_out = self.attention_layer(query, key, value, h, w, valid_ratios)
+        attn_out = paddle.transpose(attn_out, (0, 2, 1))  # [n, len_q, dim_v]
+
+        if self.return_feature:
+            return attn_out
+
+        return self.prediction(attn_out)
+
+    def forward_test(self, feat, out_enc, valid_ratios, position_index):
+        """
+        Args:
+            feat (Tensor): Tensor of shape :math:`(N, D_i, H, W)`.
+            out_enc (Tensor): Encoder output of shape
+                :math:`(N, D_m, H, W)`.
+            valid_ratios (Tensor): valid length ratio of img
+            position_index (Tensor): The position of each word.
+
+        Returns:
+            Tensor: A raw logit tensor of shape :math:`(N, T, C-1)` if
+            ``return_feature=False``. Otherwise it would be the hidden feature
+            before the prediction projection layer, whose shape is
+            :math:`(N, T, D_m)`.
+        """
+        n, c_enc, h, w = out_enc.shape
+        assert c_enc == self.dim_model
+        _, c_feat, _, _ = feat.shape
+        assert c_feat == self.dim_input
+
+        position_out_enc = self.position_aware_module(out_enc)
+
+        query = self.embedding(position_index)
+        query = paddle.transpose(query, (0, 2, 1))
+        key = paddle.reshape(position_out_enc, (n, c_enc, h * w))
+        if self.encode_value:
+            value = paddle.reshape(out_enc, (n, c_enc, h * w))
+        else:
+            value = paddle.reshape(feat, (n, c_feat, h * w))
+
+        attn_out = self.attention_layer(query, key, value, h, w, valid_ratios)
+        attn_out = paddle.transpose(attn_out, (0, 2, 1))  # [n, len_q, dim_v]
+
+        if self.return_feature:
+            return attn_out
+
+        return self.prediction(attn_out)
+
+
+class RobustScannerFusionLayer(nn.Layer):
+    def __init__(self, dim_model, dim=-1):
+        super(RobustScannerFusionLayer, self).__init__()
+
+        self.dim_model = dim_model
+        self.dim = dim
+        self.linear_layer = nn.Linear(dim_model * 2, dim_model * 2)
+
+    def forward(self, x0, x1):
+        assert x0.shape == x1.shape
+        fusion_input = paddle.concat([x0, x1], self.dim)
+        output = self.linear_layer(fusion_input)
+        output = F.glu(output, self.dim)
+        return output
+
+
+class RobustScannerDecoder(BaseDecoder):
+    """Decoder for RobustScanner.
+
+    RobustScanner: `RobustScanner: Dynamically Enhancing Positional Clues for
+    Robust Text Recognition <https://arxiv.org/abs/2007.07542>`_
+
+    Args:
+        num_classes (int): Number of output classes :math:`C`.
+        dim_input (int): Dimension :math:`D_i` of input vector ``feat``.
+        dim_model (int): Dimension :math:`D_m` of the model. Should also be the
+            same as encoder output vector ``out_enc``.
+        max_seq_len (int): Maximum output sequence length :math:`T`.
+        start_idx (int): The index of `<SOS>`.
+        mask (bool): Whether to mask input features according to
+            ``img_meta['valid_ratio']``.
+        padding_idx (int): The index of `<PAD>`.
+        encode_value (bool): Whether to use the output of encoder ``out_enc``
+            as `value` of attention layer. If False, the original feature
+            ``feat`` will be used.
+
+    Warning:
+        This decoder will not predict the final class which is assumed to be
+        `<PAD>`. Therefore, its output size is always :math:`C - 1`. `<PAD>`
+        is also ignored by loss as specified in
+        :obj:`mmocr.models.textrecog.recognizer.EncodeDecodeRecognizer`.
+    """
+
+    def __init__(
+        self,
+        num_classes=None,
+        dim_input=512,
+        dim_model=128,
+        hybrid_decoder_rnn_layers=2,
+        hybrid_decoder_dropout=0,
+        position_decoder_rnn_layers=2,
+        max_seq_len=40,
+        start_idx=0,
+        mask=True,
+        padding_idx=None,
+        encode_value=False,
+    ):
+        super().__init__()
+        self.num_classes = num_classes
+        self.dim_input = dim_input
+        self.dim_model = dim_model
+        self.max_seq_len = max_seq_len
+        self.encode_value = encode_value
+        self.start_idx = start_idx
+        self.padding_idx = padding_idx
+        self.mask = mask
+
+        # init hybrid decoder
+        self.hybrid_decoder = SequenceAttentionDecoder(
+            num_classes=num_classes,
+            rnn_layers=hybrid_decoder_rnn_layers,
+            dim_input=dim_input,
+            dim_model=dim_model,
+            max_seq_len=max_seq_len,
+            start_idx=start_idx,
+            mask=mask,
+            padding_idx=padding_idx,
+            dropout=hybrid_decoder_dropout,
+            encode_value=encode_value,
+            return_feature=True,
+        )
+
+        # init position decoder
+        self.position_decoder = PositionAttentionDecoder(
+            num_classes=num_classes,
+            rnn_layers=position_decoder_rnn_layers,
+            dim_input=dim_input,
+            dim_model=dim_model,
+            max_seq_len=max_seq_len,
+            mask=mask,
+            encode_value=encode_value,
+            return_feature=True,
+        )
+
+        self.fusion_module = RobustScannerFusionLayer(
+            self.dim_model if encode_value else dim_input
+        )
+
+        pred_num_classes = num_classes - 1
+        self.prediction = nn.Linear(
+            dim_model if encode_value else dim_input, pred_num_classes
+        )
+
+    def forward_train(self, feat, out_enc, target, valid_ratios, word_positions):
+        """
+        Args:
+            feat (Tensor): Tensor of shape :math:`(N, D_i, H, W)`.
+            out_enc (Tensor): Encoder output of shape
+                :math:`(N, D_m, H, W)`.
+            target (dict): A dict with the key ``padded_targets``, a
+                tensor of shape :math:`(N, T)`. Each element is the index of a
+                character.
+            valid_ratios (Tensor):
+            word_positions (Tensor): The position of each word.
+
+        Returns:
+            Tensor: A raw logit tensor of shape :math:`(N, T, C-1)`.
+        """
+        hybrid_glimpse = self.hybrid_decoder.forward_train(
+            feat, out_enc, target, valid_ratios
+        )
+        position_glimpse = self.position_decoder.forward_train(
+            feat, out_enc, target, valid_ratios, word_positions
+        )
+
+        fusion_out = self.fusion_module(hybrid_glimpse, position_glimpse)
+
+        out = self.prediction(fusion_out)
+
+        return out
+
+    def forward_test(self, feat, out_enc, valid_ratios, word_positions):
+        """
+        Args:
+            feat (Tensor): Tensor of shape :math:`(N, D_i, H, W)`.
+            out_enc (Tensor): Encoder output of shape
+                :math:`(N, D_m, H, W)`.
+            valid_ratios (Tensor):
+            word_positions (Tensor): The position of each word.
+        Returns:
+            Tensor: The output logit sequence tensor of shape
+            :math:`(N, T, C-1)`.
+        """
+        seq_len = self.max_seq_len
+        batch_size = feat.shape[0]
+
+        decode_sequence = (
+            paddle.ones((batch_size, seq_len), dtype="int64") * self.start_idx
+        )
+
+        position_glimpse = self.position_decoder.forward_test(
+            feat, out_enc, valid_ratios, word_positions
+        )
+
+        outputs = []
+        for i in range(seq_len):
+            hybrid_glimpse_step = self.hybrid_decoder.forward_test_step(
+                feat, out_enc, decode_sequence, i, valid_ratios
+            )
+
+            fusion_out = self.fusion_module(
+                hybrid_glimpse_step, position_glimpse[:, i, :]
+            )
+
+            char_out = self.prediction(fusion_out)
+            char_out = F.softmax(char_out, -1)
+            outputs.append(char_out)
+            max_idx = paddle.argmax(char_out, axis=1, keepdim=False)
+            if i < seq_len - 1:
+                decode_sequence[:, i + 1] = max_idx
+
+        outputs = paddle.stack(outputs, 1)
+
+        return outputs
+
+
+class RobustScannerHead(nn.Layer):
+    def __init__(
+        self,
+        out_channels,  # 90 + unknown + start + padding
+        in_channels,
+        enc_outchannles=128,
+        hybrid_dec_rnn_layers=2,
+        hybrid_dec_dropout=0,
+        position_dec_rnn_layers=2,
+        start_idx=0,
+        max_text_length=40,
+        mask=True,
+        padding_idx=None,
+        encode_value=False,
+        **kwargs,
+    ):
+        super(RobustScannerHead, self).__init__()
+
+        # encoder module
+        self.encoder = ChannelReductionEncoder(
+            in_channels=in_channels, out_channels=enc_outchannles
+        )
+
+        # decoder module
+        self.decoder = RobustScannerDecoder(
+            num_classes=out_channels,
+            dim_input=in_channels,
+            dim_model=enc_outchannles,
+            hybrid_decoder_rnn_layers=hybrid_dec_rnn_layers,
+            hybrid_decoder_dropout=hybrid_dec_dropout,
+            position_decoder_rnn_layers=position_dec_rnn_layers,
+            max_seq_len=max_text_length,
+            start_idx=start_idx,
+            mask=mask,
+            padding_idx=padding_idx,
+            encode_value=encode_value,
+        )
+
+    def forward(self, inputs, targets=None):
+        """
+        targets: [label, valid_ratio, word_positions]
+        """
+        out_enc = self.encoder(inputs)
+        valid_ratios = None
+        word_positions = targets[-1]
+
+        if len(targets) > 1:
+            valid_ratios = targets[-2]
+
+        if self.training:
+            label = targets[0]  # label
+            label = paddle.to_tensor(label, dtype="int64")
+            final_out = self.decoder(
+                inputs, out_enc, label, valid_ratios, word_positions
+            )
+        if not self.training:
+            final_out = self.decoder(
+                inputs,
+                out_enc,
+                label=None,
+                valid_ratios=valid_ratios,
+                word_positions=word_positions,
+                train_mode=False,
+            )
+        return final_out
diff --git a/ppocr/modeling/heads/rec_sar_head.py b/ppocr/modeling/heads/rec_sar_head.py
new file mode 100644
index 0000000..1e01999
--- /dev/null
+++ b/ppocr/modeling/heads/rec_sar_head.py
@@ -0,0 +1,407 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textrecog/encoders/sar_encoder.py
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textrecog/decoders/sar_decoder.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+from paddle import ParamAttr
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+
+class SAREncoder(nn.Layer):
+    """
+    Args:
+        enc_bi_rnn (bool): If True, use bidirectional RNN in encoder.
+        enc_drop_rnn (float): Dropout probability of RNN layer in encoder.
+        enc_gru (bool): If True, use GRU, else LSTM in encoder.
+        d_model (int): Dim of channels from backbone.
+        d_enc (int): Dim of encoder RNN layer.
+        mask (bool): If True, mask padding in RNN sequence.
+    """
+
+    def __init__(
+        self,
+        enc_bi_rnn=False,
+        enc_drop_rnn=0.1,
+        enc_gru=False,
+        d_model=512,
+        d_enc=512,
+        mask=True,
+        **kwargs,
+    ):
+        super().__init__()
+        assert isinstance(enc_bi_rnn, bool)
+        assert isinstance(enc_drop_rnn, (int, float))
+        assert 0 <= enc_drop_rnn < 1.0
+        assert isinstance(enc_gru, bool)
+        assert isinstance(d_model, int)
+        assert isinstance(d_enc, int)
+        assert isinstance(mask, bool)
+
+        self.enc_bi_rnn = enc_bi_rnn
+        self.enc_drop_rnn = enc_drop_rnn
+        self.mask = mask
+
+        # LSTM Encoder
+        if enc_bi_rnn:
+            direction = "bidirectional"
+        else:
+            direction = "forward"
+        kwargs = dict(
+            input_size=d_model,
+            hidden_size=d_enc,
+            num_layers=2,
+            time_major=False,
+            dropout=enc_drop_rnn,
+            direction=direction,
+        )
+        if enc_gru:
+            self.rnn_encoder = nn.GRU(**kwargs)
+        else:
+            self.rnn_encoder = nn.LSTM(**kwargs)
+
+        # global feature transformation
+        encoder_rnn_out_size = d_enc * (int(enc_bi_rnn) + 1)
+        self.linear = nn.Linear(encoder_rnn_out_size, encoder_rnn_out_size)
+
+    def forward(self, feat, img_metas=None):
+        if img_metas is not None:
+            assert len(img_metas[0]) == feat.shape[0]
+
+        valid_ratios = None
+        if img_metas is not None and self.mask:
+            valid_ratios = img_metas[-1]
+
+        h_feat = feat.shape[2]  # bsz c h w
+        feat_v = F.max_pool2d(feat, kernel_size=(h_feat, 1), stride=1, padding=0)
+        feat_v = feat_v.squeeze(2)  # bsz * C * W
+        feat_v = paddle.transpose(feat_v, perm=[0, 2, 1])  # bsz * W * C
+        holistic_feat = self.rnn_encoder(feat_v)[0]  # bsz * T * C
+
+        if valid_ratios is not None:
+            valid_hf = []
+            T = paddle.shape(holistic_feat)[1]
+            for i in range(valid_ratios.shape[0]):
+                valid_step = (
+                    paddle.minimum(T, paddle.ceil(valid_ratios[i] * T).astype(T.dtype))
+                    - 1
+                )
+                valid_hf.append(holistic_feat[i, valid_step, :])
+            valid_hf = paddle.stack(valid_hf, axis=0)
+        else:
+            valid_hf = holistic_feat[:, -1, :]  # bsz * C
+        holistic_feat = self.linear(valid_hf)  # bsz * C
+
+        return holistic_feat
+
+
+class BaseDecoder(nn.Layer):
+    def __init__(self, **kwargs):
+        super().__init__()
+
+    def forward_train(self, feat, out_enc, targets, img_metas):
+        raise NotImplementedError
+
+    def forward_test(self, feat, out_enc, img_metas):
+        raise NotImplementedError
+
+    def forward(self, feat, out_enc, label=None, img_metas=None, train_mode=True):
+        self.train_mode = train_mode
+
+        if train_mode:
+            return self.forward_train(feat, out_enc, label, img_metas)
+        return self.forward_test(feat, out_enc, img_metas)
+
+
+class ParallelSARDecoder(BaseDecoder):
+    """
+    Args:
+        out_channels (int): Output class number.
+        enc_bi_rnn (bool): If True, use bidirectional RNN in encoder.
+        dec_bi_rnn (bool): If True, use bidirectional RNN in decoder.
+        dec_drop_rnn (float): Dropout of RNN layer in decoder.
+        dec_gru (bool): If True, use GRU, else LSTM in decoder.
+        d_model (int): Dim of channels from backbone.
+        d_enc (int): Dim of encoder RNN layer.
+        d_k (int): Dim of channels of attention module.
+        pred_dropout (float): Dropout probability of prediction layer.
+        max_seq_len (int): Maximum sequence length for decoding.
+        mask (bool): If True, mask padding in feature map.
+        start_idx (int): Index of start token.
+        padding_idx (int): Index of padding token.
+        pred_concat (bool): If True, concat glimpse feature from
+            attention with holistic feature and hidden state.
+    """
+
+    def __init__(
+        self,
+        out_channels,  # 90 + unknown + start + padding
+        enc_bi_rnn=False,
+        dec_bi_rnn=False,
+        dec_drop_rnn=0.0,
+        dec_gru=False,
+        d_model=512,
+        d_enc=512,
+        d_k=64,
+        pred_dropout=0.1,
+        max_text_length=30,
+        mask=True,
+        pred_concat=True,
+        **kwargs,
+    ):
+        super().__init__()
+
+        self.num_classes = out_channels
+        self.enc_bi_rnn = enc_bi_rnn
+        self.d_k = d_k
+        self.start_idx = out_channels - 2
+        self.padding_idx = out_channels - 1
+        self.max_seq_len = max_text_length
+        self.mask = mask
+        self.pred_concat = pred_concat
+
+        encoder_rnn_out_size = d_enc * (int(enc_bi_rnn) + 1)
+        decoder_rnn_out_size = encoder_rnn_out_size * (int(dec_bi_rnn) + 1)
+
+        # 2D attention layer
+        self.conv1x1_1 = nn.Linear(decoder_rnn_out_size, d_k)
+        self.conv3x3_1 = nn.Conv2D(d_model, d_k, kernel_size=3, stride=1, padding=1)
+        self.conv1x1_2 = nn.Linear(d_k, 1)
+
+        # Decoder RNN layer
+        if dec_bi_rnn:
+            direction = "bidirectional"
+        else:
+            direction = "forward"
+
+        kwargs = dict(
+            input_size=encoder_rnn_out_size,
+            hidden_size=encoder_rnn_out_size,
+            num_layers=2,
+            time_major=False,
+            dropout=dec_drop_rnn,
+            direction=direction,
+        )
+        if dec_gru:
+            self.rnn_decoder = nn.GRU(**kwargs)
+        else:
+            self.rnn_decoder = nn.LSTM(**kwargs)
+
+        # Decoder input embedding
+        self.embedding = nn.Embedding(
+            self.num_classes, encoder_rnn_out_size, padding_idx=self.padding_idx
+        )
+
+        # Prediction layer
+        self.pred_dropout = nn.Dropout(pred_dropout)
+        pred_num_classes = self.num_classes - 1
+        if pred_concat:
+            fc_in_channel = decoder_rnn_out_size + d_model + encoder_rnn_out_size
+        else:
+            fc_in_channel = d_model
+        self.prediction = nn.Linear(fc_in_channel, pred_num_classes)
+
+    def _2d_attention(self, decoder_input, feat, holistic_feat, valid_ratios=None):
+        y = self.rnn_decoder(decoder_input)[0]
+        # y: bsz * (seq_len + 1) * hidden_size
+
+        attn_query = self.conv1x1_1(y)  # bsz * (seq_len + 1) * attn_size
+        bsz, seq_len, attn_size = attn_query.shape
+        attn_query = paddle.unsqueeze(attn_query, axis=[3, 4])
+        # (bsz, seq_len + 1, attn_size, 1, 1)
+
+        attn_key = self.conv3x3_1(feat)
+        # bsz * attn_size * h * w
+        attn_key = attn_key.unsqueeze(1)
+        # bsz * 1 * attn_size * h * w
+
+        attn_weight = paddle.tanh(paddle.add(attn_key, attn_query))
+
+        # bsz * (seq_len + 1) * attn_size * h * w
+        attn_weight = paddle.transpose(attn_weight, perm=[0, 1, 3, 4, 2])
+        # bsz * (seq_len + 1) * h * w * attn_size
+        attn_weight = self.conv1x1_2(attn_weight)
+        # bsz * (seq_len + 1) * h * w * 1
+        bsz, T, h, w, c = paddle.shape(attn_weight)
+        assert c == 1
+
+        if valid_ratios is not None:
+            # cal mask of attention weight
+            for i in range(valid_ratios.shape[0]):
+                valid_width = paddle.minimum(
+                    w.astype("int64"), paddle.ceil(valid_ratios[i] * w).astype("int64")
+                )
+                if valid_width < w:
+                    attn_weight[i, :, :, valid_width:, :] = float("-inf")
+
+        attn_weight = paddle.reshape(attn_weight, [bsz, T, -1])
+        attn_weight = F.softmax(attn_weight, axis=-1)
+
+        attn_weight = paddle.reshape(attn_weight, [bsz, T, h, w, c])
+        attn_weight = paddle.transpose(attn_weight, perm=[0, 1, 4, 2, 3])
+        # attn_weight: bsz * T * c * h * w
+        # feat: bsz * c * h * w
+        attn_feat = paddle.sum(
+            paddle.multiply(feat.unsqueeze(1), attn_weight), (3, 4), keepdim=False
+        )
+        # bsz * (seq_len + 1) * C
+
+        # Linear transformation
+        if self.pred_concat:
+            hf_c = holistic_feat.shape[-1]
+            holistic_feat = paddle.expand(holistic_feat, shape=[bsz, seq_len, hf_c])
+            y = self.prediction(
+                paddle.concat(
+                    (y, attn_feat.astype(y.dtype), holistic_feat.astype(y.dtype)), 2
+                )
+            )
+        else:
+            y = self.prediction(attn_feat)
+        # bsz * (seq_len + 1) * num_classes
+        if self.train_mode:
+            y = self.pred_dropout(y)
+
+        return y
+
+    def forward_train(self, feat, out_enc, label, img_metas):
+        """
+        img_metas: [label, valid_ratio]
+        """
+        if img_metas is not None:
+            assert img_metas[0].shape[0] == feat.shape[0]
+
+        valid_ratios = None
+        if img_metas is not None and self.mask:
+            valid_ratios = img_metas[-1]
+
+        lab_embedding = self.embedding(label)
+        # bsz * seq_len * emb_dim
+        out_enc = out_enc.unsqueeze(1).astype(lab_embedding.dtype)
+        # bsz * 1 * emb_dim
+        in_dec = paddle.concat((out_enc, lab_embedding), axis=1)
+        # bsz * (seq_len + 1) * C
+        out_dec = self._2d_attention(in_dec, feat, out_enc, valid_ratios=valid_ratios)
+
+        return out_dec[:, 1:, :]  # bsz * seq_len * num_classes
+
+    def forward_test(self, feat, out_enc, img_metas):
+        if img_metas is not None:
+            assert len(img_metas[0]) == feat.shape[0]
+
+        valid_ratios = None
+        if img_metas is not None and self.mask:
+            valid_ratios = img_metas[-1]
+
+        seq_len = self.max_seq_len
+        bsz = feat.shape[0]
+        start_token = paddle.full((bsz,), fill_value=self.start_idx, dtype="int64")
+        # bsz
+        start_token = self.embedding(start_token)
+        # bsz * emb_dim
+        emb_dim = start_token.shape[1]
+        start_token = start_token.unsqueeze(1)
+        start_token = paddle.expand(start_token, shape=[bsz, seq_len, emb_dim])
+        # bsz * seq_len * emb_dim
+        out_enc = out_enc.unsqueeze(1)
+        # bsz * 1 * emb_dim
+        decoder_input = paddle.concat((out_enc, start_token), axis=1)
+        # bsz * (seq_len + 1) * emb_dim
+
+        outputs = []
+        for i in range(1, seq_len + 1):
+            decoder_output = self._2d_attention(
+                decoder_input, feat, out_enc, valid_ratios=valid_ratios
+            )
+            char_output = decoder_output[:, i, :]  # bsz * num_classes
+            char_output = F.softmax(char_output, -1)
+            outputs.append(char_output)
+            max_idx = paddle.argmax(char_output, axis=1, keepdim=False)
+            char_embedding = self.embedding(max_idx)  # bsz * emb_dim
+            if i < seq_len:
+                decoder_input[:, i + 1, :] = char_embedding
+
+        outputs = paddle.stack(outputs, 1)  # bsz * seq_len * num_classes
+
+        return outputs
+
+
+class SARHead(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        enc_dim=512,
+        max_text_length=30,
+        enc_bi_rnn=False,
+        enc_drop_rnn=0.1,
+        enc_gru=False,
+        dec_bi_rnn=False,
+        dec_drop_rnn=0.0,
+        dec_gru=False,
+        d_k=512,
+        pred_dropout=0.1,
+        pred_concat=True,
+        **kwargs,
+    ):
+        super(SARHead, self).__init__()
+
+        # encoder module
+        self.encoder = SAREncoder(
+            enc_bi_rnn=enc_bi_rnn,
+            enc_drop_rnn=enc_drop_rnn,
+            enc_gru=enc_gru,
+            d_model=in_channels,
+            d_enc=enc_dim,
+        )
+
+        # decoder module
+        self.decoder = ParallelSARDecoder(
+            out_channels=out_channels,
+            enc_bi_rnn=enc_bi_rnn,
+            dec_bi_rnn=dec_bi_rnn,
+            dec_drop_rnn=dec_drop_rnn,
+            dec_gru=dec_gru,
+            d_model=in_channels,
+            d_enc=enc_dim,
+            d_k=d_k,
+            pred_dropout=pred_dropout,
+            max_text_length=max_text_length,
+            pred_concat=pred_concat,
+        )
+
+    def forward(self, feat, targets=None):
+        """
+        img_metas: [label, valid_ratio]
+        """
+        holistic_feat = self.encoder(feat, targets)  # bsz c
+
+        if self.training:
+            label = targets[0]  # label
+            final_out = self.decoder(feat, holistic_feat, label, img_metas=targets)
+        else:
+            final_out = self.decoder(
+                feat, holistic_feat, label=None, img_metas=targets, train_mode=False
+            )
+            # (bsz, seq_len, num_classes)
+
+        return final_out
diff --git a/ppocr/modeling/heads/rec_satrn_head.py b/ppocr/modeling/heads/rec_satrn_head.py
new file mode 100644
index 0000000..8f20474
--- /dev/null
+++ b/ppocr/modeling/heads/rec_satrn_head.py
@@ -0,0 +1,592 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/1.x/mmocr/models/textrecog/encoders/satrn_encoder.py
+https://github.com/open-mmlab/mmocr/blob/1.x/mmocr/models/textrecog/decoders/nrtr_decoder.py
+"""
+
+import math
+import numpy as np
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle import ParamAttr, reshape, transpose
+from paddle.nn import Conv2D, BatchNorm, Linear, Dropout
+from paddle.nn import AdaptiveAvgPool2D, MaxPool2D, AvgPool2D
+from paddle.nn.initializer import KaimingNormal, Uniform, Constant
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self, num_channels, filter_size, num_filters, stride, padding, num_groups=1
+    ):
+        super(ConvBNLayer, self).__init__()
+
+        self.conv = nn.Conv2D(
+            in_channels=num_channels,
+            out_channels=num_filters,
+            kernel_size=filter_size,
+            stride=stride,
+            padding=padding,
+            groups=num_groups,
+            bias_attr=False,
+        )
+
+        self.bn = nn.BatchNorm2D(
+            num_filters,
+            weight_attr=ParamAttr(initializer=Constant(1)),
+            bias_attr=ParamAttr(initializer=Constant(0)),
+        )
+        self.relu = nn.ReLU()
+
+    def forward(self, inputs):
+        y = self.conv(inputs)
+        y = self.bn(y)
+        y = self.relu(y)
+        return y
+
+
+class SATRNEncoderLayer(nn.Layer):
+    def __init__(
+        self,
+        d_model=512,
+        d_inner=512,
+        n_head=8,
+        d_k=64,
+        d_v=64,
+        dropout=0.1,
+        qkv_bias=False,
+    ):
+        super().__init__()
+        self.norm1 = nn.LayerNorm(d_model)
+        self.attn = MultiHeadAttention(
+            n_head, d_model, d_k, d_v, qkv_bias=qkv_bias, dropout=dropout
+        )
+        self.norm2 = nn.LayerNorm(d_model)
+        self.feed_forward = LocalityAwareFeedforward(d_model, d_inner, dropout=dropout)
+
+    def forward(self, x, h, w, mask=None):
+        n, hw, c = x.shape
+        residual = x
+        x = self.norm1(x)
+        x = residual + self.attn(x, x, x, mask)
+        residual = x
+        x = self.norm2(x)
+        x = x.transpose([0, 2, 1]).reshape([n, c, h, w])
+        x = self.feed_forward(x)
+        x = x.reshape([n, c, hw]).transpose([0, 2, 1])
+        x = residual + x
+        return x
+
+
+class LocalityAwareFeedforward(nn.Layer):
+    def __init__(
+        self,
+        d_in,
+        d_hid,
+        dropout=0.1,
+    ):
+        super().__init__()
+        self.conv1 = ConvBNLayer(d_in, 1, d_hid, stride=1, padding=0)
+
+        self.depthwise_conv = ConvBNLayer(
+            d_hid, 3, d_hid, stride=1, padding=1, num_groups=d_hid
+        )
+
+        self.conv2 = ConvBNLayer(d_hid, 1, d_in, stride=1, padding=0)
+
+    def forward(self, x):
+        x = self.conv1(x)
+        x = self.depthwise_conv(x)
+        x = self.conv2(x)
+
+        return x
+
+
+class Adaptive2DPositionalEncoding(nn.Layer):
+    def __init__(self, d_hid=512, n_height=100, n_width=100, dropout=0.1):
+        super().__init__()
+
+        h_position_encoder = self._get_sinusoid_encoding_table(n_height, d_hid)
+        h_position_encoder = h_position_encoder.transpose([1, 0])
+        h_position_encoder = h_position_encoder.reshape([1, d_hid, n_height, 1])
+
+        w_position_encoder = self._get_sinusoid_encoding_table(n_width, d_hid)
+        w_position_encoder = w_position_encoder.transpose([1, 0])
+        w_position_encoder = w_position_encoder.reshape([1, d_hid, 1, n_width])
+
+        self.register_buffer("h_position_encoder", h_position_encoder)
+        self.register_buffer("w_position_encoder", w_position_encoder)
+
+        self.h_scale = self.scale_factor_generate(d_hid)
+        self.w_scale = self.scale_factor_generate(d_hid)
+        self.pool = nn.AdaptiveAvgPool2D(1)
+        self.dropout = nn.Dropout(p=dropout)
+
+    def _get_sinusoid_encoding_table(self, n_position, d_hid):
+        """Sinusoid position encoding table."""
+        denominator = paddle.to_tensor(
+            [1.0 / np.power(10000, 2 * (hid_j // 2) / d_hid) for hid_j in range(d_hid)]
+        )
+        denominator = denominator.reshape([1, -1])
+        pos_tensor = paddle.cast(paddle.arange(n_position).unsqueeze(-1), "float32")
+        sinusoid_table = pos_tensor * denominator
+        sinusoid_table[:, 0::2] = paddle.sin(sinusoid_table[:, 0::2])
+        sinusoid_table[:, 1::2] = paddle.cos(sinusoid_table[:, 1::2])
+
+        return sinusoid_table
+
+    def scale_factor_generate(self, d_hid):
+        scale_factor = nn.Sequential(
+            nn.Conv2D(d_hid, d_hid, 1),
+            nn.ReLU(),
+            nn.Conv2D(d_hid, d_hid, 1),
+            nn.Sigmoid(),
+        )
+
+        return scale_factor
+
+    def forward(self, x):
+        b, c, h, w = x.shape
+
+        avg_pool = self.pool(x)
+
+        h_pos_encoding = self.h_scale(avg_pool) * self.h_position_encoder[:, :, :h, :]
+        w_pos_encoding = self.w_scale(avg_pool) * self.w_position_encoder[:, :, :, :w]
+
+        out = x + h_pos_encoding + w_pos_encoding
+
+        out = self.dropout(out)
+
+        return out
+
+
+class ScaledDotProductAttention(nn.Layer):
+    def __init__(self, temperature, attn_dropout=0.1):
+        super().__init__()
+        self.temperature = temperature
+        self.dropout = nn.Dropout(attn_dropout)
+
+    def forward(self, q, k, v, mask=None):
+        def masked_fill(x, mask, value):
+            y = paddle.full(x.shape, value, x.dtype)
+            return paddle.where(mask, y, x)
+
+        attn = paddle.matmul(q / self.temperature, k.transpose([0, 1, 3, 2]))
+        if mask is not None:
+            attn = masked_fill(attn, mask == 0, -1e9)
+            # attn = attn.masked_fill(mask == 0, float('-inf'))
+            # attn += mask
+
+        attn = self.dropout(F.softmax(attn, axis=-1))
+        output = paddle.matmul(attn, v)
+
+        return output, attn
+
+
+class MultiHeadAttention(nn.Layer):
+    def __init__(
+        self, n_head=8, d_model=512, d_k=64, d_v=64, dropout=0.1, qkv_bias=False
+    ):
+        super().__init__()
+        self.n_head = n_head
+        self.d_k = d_k
+        self.d_v = d_v
+
+        self.dim_k = n_head * d_k
+        self.dim_v = n_head * d_v
+
+        self.linear_q = nn.Linear(self.dim_k, self.dim_k, bias_attr=qkv_bias)
+        self.linear_k = nn.Linear(self.dim_k, self.dim_k, bias_attr=qkv_bias)
+        self.linear_v = nn.Linear(self.dim_v, self.dim_v, bias_attr=qkv_bias)
+
+        self.attention = ScaledDotProductAttention(d_k**0.5, dropout)
+
+        self.fc = nn.Linear(self.dim_v, d_model, bias_attr=qkv_bias)
+        self.proj_drop = nn.Dropout(dropout)
+
+    def forward(self, q, k, v, mask=None):
+        batch_size, len_q, _ = q.shape
+        _, len_k, _ = k.shape
+
+        q = self.linear_q(q).reshape([batch_size, len_q, self.n_head, self.d_k])
+        k = self.linear_k(k).reshape([batch_size, len_k, self.n_head, self.d_k])
+        v = self.linear_v(v).reshape([batch_size, len_k, self.n_head, self.d_v])
+
+        q, k, v = (
+            q.transpose([0, 2, 1, 3]),
+            k.transpose([0, 2, 1, 3]),
+            v.transpose([0, 2, 1, 3]),
+        )
+
+        if mask is not None:
+            if mask.dim() == 3:
+                mask = mask.unsqueeze(1)
+            elif mask.dim() == 2:
+                mask = mask.unsqueeze(1).unsqueeze(1)
+
+        attn_out, _ = self.attention(q, k, v, mask=mask)
+
+        attn_out = attn_out.transpose([0, 2, 1, 3]).reshape(
+            [batch_size, len_q, self.dim_v]
+        )
+
+        attn_out = self.fc(attn_out)
+        attn_out = self.proj_drop(attn_out)
+
+        return attn_out
+
+
+class SATRNEncoder(nn.Layer):
+    def __init__(
+        self,
+        n_layers=12,
+        n_head=8,
+        d_k=64,
+        d_v=64,
+        d_model=512,
+        n_position=100,
+        d_inner=256,
+        dropout=0.1,
+    ):
+        super().__init__()
+        self.d_model = d_model
+        self.position_enc = Adaptive2DPositionalEncoding(
+            d_hid=d_model, n_height=n_position, n_width=n_position, dropout=dropout
+        )
+        self.layer_stack = nn.LayerList(
+            [
+                SATRNEncoderLayer(d_model, d_inner, n_head, d_k, d_v, dropout=dropout)
+                for _ in range(n_layers)
+            ]
+        )
+        self.layer_norm = nn.LayerNorm(d_model)
+
+    def forward(self, feat, valid_ratios=None):
+        """
+        Args:
+            feat (Tensor): Feature tensor of shape :math:`(N, D_m, H, W)`.
+            img_metas (dict): A dict that contains meta information of input
+                images. Preferably with the key ``valid_ratio``.
+
+        Returns:
+            Tensor: A tensor of shape :math:`(N, T, D_m)`.
+        """
+        if valid_ratios is None:
+            bs = feat.shape[0]
+            valid_ratios = paddle.full((bs, 1), 1.0, dtype=paddle.float32)
+
+        feat = self.position_enc(feat)
+        n, c, h, w = feat.shape
+
+        mask = paddle.zeros((n, h, w))
+        for i, valid_ratio in enumerate(valid_ratios):
+            valid_width = int(min(w, paddle.ceil(w * valid_ratio)))
+            mask[i, :, :valid_width] = 1
+
+        mask = mask.reshape([n, h * w])
+        feat = feat.reshape([n, c, h * w])
+
+        output = feat.transpose([0, 2, 1])
+        for enc_layer in self.layer_stack:
+            output = enc_layer(output, h, w, mask)
+        output = self.layer_norm(output)
+
+        return output
+
+
+class PositionwiseFeedForward(nn.Layer):
+    def __init__(self, d_in, d_hid, dropout=0.1):
+        super().__init__()
+        self.w_1 = nn.Linear(d_in, d_hid)
+        self.w_2 = nn.Linear(d_hid, d_in)
+        self.act = nn.GELU()
+        self.dropout = nn.Dropout(dropout)
+
+    def forward(self, x):
+        x = self.w_1(x)
+        x = self.act(x)
+        x = self.w_2(x)
+        x = self.dropout(x)
+
+        return x
+
+
+class PositionalEncoding(nn.Layer):
+    def __init__(self, d_hid=512, n_position=200, dropout=0):
+        super().__init__()
+        self.dropout = nn.Dropout(p=dropout)
+
+        # Not a parameter
+        # Position table of shape (1, n_position, d_hid)
+        self.register_buffer(
+            "position_table", self._get_sinusoid_encoding_table(n_position, d_hid)
+        )
+
+    def _get_sinusoid_encoding_table(self, n_position, d_hid):
+        """Sinusoid position encoding table."""
+        denominator = paddle.to_tensor(
+            [1.0 / np.power(10000, 2 * (hid_j // 2) / d_hid) for hid_j in range(d_hid)]
+        )
+        denominator = denominator.reshape([1, -1])
+        pos_tensor = paddle.cast(paddle.arange(n_position).unsqueeze(-1), "float32")
+        sinusoid_table = pos_tensor * denominator
+        sinusoid_table[:, 0::2] = paddle.sin(sinusoid_table[:, 0::2])
+        sinusoid_table[:, 1::2] = paddle.cos(sinusoid_table[:, 1::2])
+
+        return sinusoid_table.unsqueeze(0)
+
+    def forward(self, x):
+        x = x + self.position_table[:, : x.shape[1]].clone().detach()
+        return self.dropout(x)
+
+
+class TFDecoderLayer(nn.Layer):
+    def __init__(
+        self,
+        d_model=512,
+        d_inner=256,
+        n_head=8,
+        d_k=64,
+        d_v=64,
+        dropout=0.1,
+        qkv_bias=False,
+        operation_order=None,
+    ):
+        super().__init__()
+
+        self.norm1 = nn.LayerNorm(d_model)
+        self.norm2 = nn.LayerNorm(d_model)
+        self.norm3 = nn.LayerNorm(d_model)
+
+        self.self_attn = MultiHeadAttention(
+            n_head, d_model, d_k, d_v, dropout=dropout, qkv_bias=qkv_bias
+        )
+
+        self.enc_attn = MultiHeadAttention(
+            n_head, d_model, d_k, d_v, dropout=dropout, qkv_bias=qkv_bias
+        )
+
+        self.mlp = PositionwiseFeedForward(d_model, d_inner, dropout=dropout)
+
+        self.operation_order = operation_order
+        if self.operation_order is None:
+            self.operation_order = (
+                "norm",
+                "self_attn",
+                "norm",
+                "enc_dec_attn",
+                "norm",
+                "ffn",
+            )
+        assert self.operation_order in [
+            ("norm", "self_attn", "norm", "enc_dec_attn", "norm", "ffn"),
+            ("self_attn", "norm", "enc_dec_attn", "norm", "ffn", "norm"),
+        ]
+
+    def forward(
+        self, dec_input, enc_output, self_attn_mask=None, dec_enc_attn_mask=None
+    ):
+        if self.operation_order == (
+            "self_attn",
+            "norm",
+            "enc_dec_attn",
+            "norm",
+            "ffn",
+            "norm",
+        ):
+            dec_attn_out = self.self_attn(
+                dec_input, dec_input, dec_input, self_attn_mask
+            )
+            dec_attn_out += dec_input
+            dec_attn_out = self.norm1(dec_attn_out)
+
+            enc_dec_attn_out = self.enc_attn(
+                dec_attn_out, enc_output, enc_output, dec_enc_attn_mask
+            )
+            enc_dec_attn_out += dec_attn_out
+            enc_dec_attn_out = self.norm2(enc_dec_attn_out)
+
+            mlp_out = self.mlp(enc_dec_attn_out)
+            mlp_out += enc_dec_attn_out
+            mlp_out = self.norm3(mlp_out)
+        elif self.operation_order == (
+            "norm",
+            "self_attn",
+            "norm",
+            "enc_dec_attn",
+            "norm",
+            "ffn",
+        ):
+            dec_input_norm = self.norm1(dec_input)
+            dec_attn_out = self.self_attn(
+                dec_input_norm, dec_input_norm, dec_input_norm, self_attn_mask
+            )
+            dec_attn_out += dec_input
+
+            enc_dec_attn_in = self.norm2(dec_attn_out)
+            enc_dec_attn_out = self.enc_attn(
+                enc_dec_attn_in, enc_output, enc_output, dec_enc_attn_mask
+            )
+            enc_dec_attn_out += dec_attn_out
+
+            mlp_out = self.mlp(self.norm3(enc_dec_attn_out))
+            mlp_out += enc_dec_attn_out
+
+        return mlp_out
+
+
+class SATRNDecoder(nn.Layer):
+    def __init__(
+        self,
+        n_layers=6,
+        d_embedding=512,
+        n_head=8,
+        d_k=64,
+        d_v=64,
+        d_model=512,
+        d_inner=256,
+        n_position=200,
+        dropout=0.1,
+        num_classes=93,
+        max_seq_len=40,
+        start_idx=1,
+        padding_idx=92,
+    ):
+        super().__init__()
+
+        self.padding_idx = padding_idx
+        self.start_idx = start_idx
+        self.max_seq_len = max_seq_len
+
+        self.trg_word_emb = nn.Embedding(
+            num_classes, d_embedding, padding_idx=padding_idx
+        )
+
+        self.position_enc = PositionalEncoding(d_embedding, n_position=n_position)
+        self.dropout = nn.Dropout(p=dropout)
+
+        self.layer_stack = nn.LayerList(
+            [
+                TFDecoderLayer(d_model, d_inner, n_head, d_k, d_v, dropout=dropout)
+                for _ in range(n_layers)
+            ]
+        )
+        self.layer_norm = nn.LayerNorm(d_model, epsilon=1e-6)
+
+        pred_num_class = num_classes - 1  # ignore padding_idx
+        self.classifier = nn.Linear(d_model, pred_num_class)
+
+    @staticmethod
+    def get_pad_mask(seq, pad_idx):
+        return (seq != pad_idx).unsqueeze(-2)
+
+    @staticmethod
+    def get_subsequent_mask(seq):
+        """For masking out the subsequent info."""
+        len_s = seq.shape[1]
+        subsequent_mask = 1 - paddle.triu(paddle.ones((len_s, len_s)), diagonal=1)
+        subsequent_mask = paddle.cast(subsequent_mask.unsqueeze(0), "bool")
+
+        return subsequent_mask
+
+    def _attention(self, trg_seq, src, src_mask=None):
+        trg_embedding = self.trg_word_emb(trg_seq)
+        trg_pos_encoded = self.position_enc(trg_embedding)
+        tgt = self.dropout(trg_pos_encoded)
+
+        trg_mask = self.get_pad_mask(
+            trg_seq, pad_idx=self.padding_idx
+        ) & self.get_subsequent_mask(trg_seq)
+        output = tgt
+        for dec_layer in self.layer_stack:
+            output = dec_layer(
+                output, src, self_attn_mask=trg_mask, dec_enc_attn_mask=src_mask
+            )
+        output = self.layer_norm(output)
+
+        return output
+
+    def _get_mask(self, logit, valid_ratios):
+        N, T, _ = logit.shape
+        mask = None
+        if valid_ratios is not None:
+            mask = paddle.zeros((N, T))
+            for i, valid_ratio in enumerate(valid_ratios):
+                valid_width = min(T, math.ceil(T * valid_ratio))
+                mask[i, :valid_width] = 1
+
+        return mask
+
+    def forward_train(self, feat, out_enc, targets, valid_ratio):
+        src_mask = self._get_mask(out_enc, valid_ratio)
+        attn_output = self._attention(targets, out_enc, src_mask=src_mask)
+        outputs = self.classifier(attn_output)
+
+        return outputs
+
+    def forward_test(self, feat, out_enc, valid_ratio):
+        src_mask = self._get_mask(out_enc, valid_ratio)
+        N = out_enc.shape[0]
+        init_target_seq = paddle.full(
+            (N, self.max_seq_len + 1), self.padding_idx, dtype="int64"
+        )
+        # bsz * seq_len
+        init_target_seq[:, 0] = self.start_idx
+
+        outputs = []
+        for step in range(0, paddle.to_tensor(self.max_seq_len)):
+            decoder_output = self._attention(
+                init_target_seq, out_enc, src_mask=src_mask
+            )
+            # bsz * seq_len * C
+            step_result = F.softmax(
+                self.classifier(decoder_output[:, step, :]), axis=-1
+            )
+            # bsz * num_classes
+            outputs.append(step_result)
+            step_max_index = paddle.argmax(step_result, axis=-1)
+            init_target_seq[:, step + 1] = step_max_index
+
+        outputs = paddle.stack(outputs, axis=1)
+
+        return outputs
+
+    def forward(self, feat, out_enc, targets=None, valid_ratio=None):
+        if self.training:
+            return self.forward_train(feat, out_enc, targets, valid_ratio)
+        else:
+            return self.forward_test(feat, out_enc, valid_ratio)
+
+
+class SATRNHead(nn.Layer):
+    def __init__(self, enc_cfg, dec_cfg, **kwargs):
+        super(SATRNHead, self).__init__()
+
+        # encoder module
+        self.encoder = SATRNEncoder(**enc_cfg)
+
+        # decoder module
+        self.decoder = SATRNDecoder(**dec_cfg)
+
+    def forward(self, feat, targets=None):
+        if targets is not None:
+            targets, valid_ratio = targets
+        else:
+            targets, valid_ratio = None, None
+        holistic_feat = self.encoder(feat, valid_ratio)  # bsz c
+        final_out = self.decoder(feat, holistic_feat, targets, valid_ratio)
+
+        return final_out
diff --git a/ppocr/modeling/heads/rec_spin_att_head.py b/ppocr/modeling/heads/rec_spin_att_head.py
new file mode 100644
index 0000000..930d970
--- /dev/null
+++ b/ppocr/modeling/heads/rec_spin_att_head.py
@@ -0,0 +1,124 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+This code is refer from:
+https://github.com/hikopensource/DAVAR-Lab-OCR/davarocr/davar_rcg/models/sequence_heads/att_head.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+
+class SPINAttentionHead(nn.Layer):
+    def __init__(self, in_channels, out_channels, hidden_size, **kwargs):
+        super(SPINAttentionHead, self).__init__()
+        self.input_size = in_channels
+        self.hidden_size = hidden_size
+        self.num_classes = out_channels
+
+        self.attention_cell = AttentionLSTMCell(
+            in_channels, hidden_size, out_channels, use_gru=False
+        )
+        self.generator = nn.Linear(hidden_size, out_channels)
+
+    def _char_to_onehot(self, input_char, onehot_dim):
+        input_ont_hot = F.one_hot(input_char, onehot_dim)
+        return input_ont_hot
+
+    def forward(self, inputs, targets=None, batch_max_length=25):
+        batch_size = inputs.shape[0]
+        num_steps = batch_max_length + 1  # +1 for [sos] at end of sentence
+
+        hidden = (
+            paddle.zeros((batch_size, self.hidden_size)),
+            paddle.zeros((batch_size, self.hidden_size)),
+        )
+        output_hiddens = []
+        if self.training:  # for train
+            targets = targets[0]
+            for i in range(num_steps):
+                char_onehots = self._char_to_onehot(
+                    targets[:, i], onehot_dim=self.num_classes
+                )
+                (outputs, hidden), alpha = self.attention_cell(
+                    hidden, inputs, char_onehots
+                )
+                output_hiddens.append(paddle.unsqueeze(outputs, axis=1))
+            output = paddle.concat(output_hiddens, axis=1)
+            probs = self.generator(output)
+        else:
+            targets = paddle.zeros(shape=[batch_size], dtype="int32")
+            probs = None
+            char_onehots = None
+            outputs = None
+            alpha = None
+
+            for i in range(num_steps):
+                char_onehots = self._char_to_onehot(
+                    targets, onehot_dim=self.num_classes
+                )
+                (outputs, hidden), alpha = self.attention_cell(
+                    hidden, inputs, char_onehots
+                )
+                probs_step = self.generator(outputs)
+                if probs is None:
+                    probs = paddle.unsqueeze(probs_step, axis=1)
+                else:
+                    probs = paddle.concat(
+                        [probs, paddle.unsqueeze(probs_step, axis=1)], axis=1
+                    )
+                next_input = probs_step.argmax(axis=1)
+                targets = next_input
+        if not self.training:
+            probs = paddle.nn.functional.softmax(probs, axis=2)
+        return probs
+
+
+class AttentionLSTMCell(nn.Layer):
+    def __init__(self, input_size, hidden_size, num_embeddings, use_gru=False):
+        super(AttentionLSTMCell, self).__init__()
+        self.i2h = nn.Linear(input_size, hidden_size, bias_attr=False)
+        self.h2h = nn.Linear(hidden_size, hidden_size)
+        self.score = nn.Linear(hidden_size, 1, bias_attr=False)
+        if not use_gru:
+            self.rnn = nn.LSTMCell(
+                input_size=input_size + num_embeddings, hidden_size=hidden_size
+            )
+        else:
+            self.rnn = nn.GRUCell(
+                input_size=input_size + num_embeddings, hidden_size=hidden_size
+            )
+
+        self.hidden_size = hidden_size
+
+    def forward(self, prev_hidden, batch_H, char_onehots):
+        batch_H_proj = self.i2h(batch_H)
+        prev_hidden_proj = paddle.unsqueeze(self.h2h(prev_hidden[0]), axis=1)
+        res = paddle.add(batch_H_proj, prev_hidden_proj)
+        res = paddle.tanh(res)
+        e = self.score(res)
+
+        alpha = F.softmax(e, axis=1)
+        alpha = paddle.transpose(alpha, [0, 2, 1])
+        context = paddle.squeeze(paddle.mm(alpha, batch_H), axis=1)
+        concat_context = paddle.concat([context, char_onehots], 1)
+        cur_hidden = self.rnn(concat_context, prev_hidden)
+
+        return cur_hidden, alpha
diff --git a/ppocr/modeling/heads/rec_srn_head.py b/ppocr/modeling/heads/rec_srn_head.py
new file mode 100644
index 0000000..8f6f139
--- /dev/null
+++ b/ppocr/modeling/heads/rec_srn_head.py
@@ -0,0 +1,315 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+from paddle import nn, ParamAttr
+from paddle.nn import functional as F
+import numpy as np
+from .self_attention import WrapEncoderForFeature
+from .self_attention import WrapEncoder
+from paddle.static import Program
+from ppocr.modeling.backbones.rec_resnet_fpn import ResNetFPN
+
+from collections import OrderedDict
+
+gradient_clip = 10
+
+
+class PVAM(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        char_num,
+        max_text_length,
+        num_heads,
+        num_encoder_tus,
+        hidden_dims,
+    ):
+        super(PVAM, self).__init__()
+        self.char_num = char_num
+        self.max_length = max_text_length
+        self.num_heads = num_heads
+        self.num_encoder_TUs = num_encoder_tus
+        self.hidden_dims = hidden_dims
+        # Transformer encoder
+        t = 256
+        c = 512
+        self.wrap_encoder_for_feature = WrapEncoderForFeature(
+            src_vocab_size=1,
+            max_length=t,
+            n_layer=self.num_encoder_TUs,
+            n_head=self.num_heads,
+            d_key=int(self.hidden_dims / self.num_heads),
+            d_value=int(self.hidden_dims / self.num_heads),
+            d_model=self.hidden_dims,
+            d_inner_hid=self.hidden_dims,
+            prepostprocess_dropout=0.1,
+            attention_dropout=0.1,
+            relu_dropout=0.1,
+            preprocess_cmd="n",
+            postprocess_cmd="da",
+            weight_sharing=True,
+        )
+
+        # PVAM
+        self.flatten0 = paddle.nn.Flatten(start_axis=0, stop_axis=1)
+        self.fc0 = paddle.nn.Linear(
+            in_features=in_channels,
+            out_features=in_channels,
+        )
+        self.emb = paddle.nn.Embedding(
+            num_embeddings=self.max_length, embedding_dim=in_channels
+        )
+        self.flatten1 = paddle.nn.Flatten(start_axis=0, stop_axis=2)
+        self.fc1 = paddle.nn.Linear(
+            in_features=in_channels, out_features=1, bias_attr=False
+        )
+
+    def forward(self, inputs, encoder_word_pos, gsrm_word_pos):
+        b, c, h, w = inputs.shape
+        conv_features = paddle.reshape(inputs, shape=[-1, c, h * w])
+        conv_features = paddle.transpose(conv_features, perm=[0, 2, 1])
+        # transformer encoder
+        b, t, c = conv_features.shape
+
+        enc_inputs = [conv_features, encoder_word_pos, None]
+        word_features = self.wrap_encoder_for_feature(enc_inputs)
+
+        # pvam
+        b, t, c = word_features.shape
+        word_features = self.fc0(word_features)
+        word_features_ = paddle.reshape(word_features, [-1, 1, t, c])
+        word_features_ = paddle.tile(word_features_, [1, self.max_length, 1, 1])
+        word_pos_feature = self.emb(gsrm_word_pos)
+        word_pos_feature_ = paddle.reshape(
+            word_pos_feature, [-1, self.max_length, 1, c]
+        )
+        word_pos_feature_ = paddle.tile(word_pos_feature_, [1, 1, t, 1])
+        y = word_pos_feature_ + word_features_
+        y = F.tanh(y)
+        attention_weight = self.fc1(y)
+        attention_weight = paddle.reshape(
+            attention_weight, shape=[-1, self.max_length, t]
+        )
+        attention_weight = F.softmax(attention_weight, axis=-1)
+        pvam_features = paddle.matmul(
+            attention_weight, word_features
+        )  # [b, max_length, c]
+        return pvam_features
+
+
+class GSRM(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        char_num,
+        max_text_length,
+        num_heads,
+        num_encoder_tus,
+        num_decoder_tus,
+        hidden_dims,
+    ):
+        super(GSRM, self).__init__()
+        self.char_num = char_num
+        self.max_length = max_text_length
+        self.num_heads = num_heads
+        self.num_encoder_TUs = num_encoder_tus
+        self.num_decoder_TUs = num_decoder_tus
+        self.hidden_dims = hidden_dims
+
+        self.fc0 = paddle.nn.Linear(in_features=in_channels, out_features=self.char_num)
+        self.wrap_encoder0 = WrapEncoder(
+            src_vocab_size=self.char_num + 1,
+            max_length=self.max_length,
+            n_layer=self.num_decoder_TUs,
+            n_head=self.num_heads,
+            d_key=int(self.hidden_dims / self.num_heads),
+            d_value=int(self.hidden_dims / self.num_heads),
+            d_model=self.hidden_dims,
+            d_inner_hid=self.hidden_dims,
+            prepostprocess_dropout=0.1,
+            attention_dropout=0.1,
+            relu_dropout=0.1,
+            preprocess_cmd="n",
+            postprocess_cmd="da",
+            weight_sharing=True,
+        )
+
+        self.wrap_encoder1 = WrapEncoder(
+            src_vocab_size=self.char_num + 1,
+            max_length=self.max_length,
+            n_layer=self.num_decoder_TUs,
+            n_head=self.num_heads,
+            d_key=int(self.hidden_dims / self.num_heads),
+            d_value=int(self.hidden_dims / self.num_heads),
+            d_model=self.hidden_dims,
+            d_inner_hid=self.hidden_dims,
+            prepostprocess_dropout=0.1,
+            attention_dropout=0.1,
+            relu_dropout=0.1,
+            preprocess_cmd="n",
+            postprocess_cmd="da",
+            weight_sharing=True,
+        )
+
+        self.mul = lambda x: paddle.matmul(
+            x=x, y=self.wrap_encoder0.prepare_decoder.emb0.weight, transpose_y=True
+        )
+
+    def forward(self, inputs, gsrm_word_pos, gsrm_slf_attn_bias1, gsrm_slf_attn_bias2):
+        # ===== GSRM Visual-to-semantic embedding block =====
+        b, t, c = inputs.shape
+        pvam_features = paddle.reshape(inputs, [-1, c])
+        word_out = self.fc0(pvam_features)
+        word_ids = paddle.argmax(F.softmax(word_out), axis=1)
+        word_ids = paddle.reshape(x=word_ids, shape=[-1, t, 1])
+
+        # ===== GSRM Semantic reasoning block =====
+        """
+        This module is achieved through bi-transformers,
+        ngram_feature1 is the forward one, ngram_fetaure2 is the backward one
+        """
+        pad_idx = self.char_num
+
+        word1 = paddle.cast(word_ids, "float32")
+        word1 = F.pad(word1, [1, 0], value=1.0 * pad_idx, data_format="NLC")
+        word1 = paddle.cast(word1, "int64")
+        word1 = word1[:, :-1, :]
+        word2 = word_ids
+
+        enc_inputs_1 = [word1, gsrm_word_pos, gsrm_slf_attn_bias1]
+        enc_inputs_2 = [word2, gsrm_word_pos, gsrm_slf_attn_bias2]
+
+        gsrm_feature1 = self.wrap_encoder0(enc_inputs_1)
+        gsrm_feature2 = self.wrap_encoder1(enc_inputs_2)
+
+        gsrm_feature2 = F.pad(gsrm_feature2, [0, 1], value=0.0, data_format="NLC")
+        gsrm_feature2 = gsrm_feature2[
+            :,
+            1:,
+        ]
+        gsrm_features = gsrm_feature1 + gsrm_feature2
+
+        gsrm_out = self.mul(gsrm_features)
+
+        b, t, c = gsrm_out.shape
+        gsrm_out = paddle.reshape(gsrm_out, [-1, c])
+
+        return gsrm_features, word_out, gsrm_out
+
+
+class VSFD(nn.Layer):
+    def __init__(self, in_channels=512, pvam_ch=512, char_num=38):
+        super(VSFD, self).__init__()
+        self.char_num = char_num
+        self.fc0 = paddle.nn.Linear(in_features=in_channels * 2, out_features=pvam_ch)
+        self.fc1 = paddle.nn.Linear(in_features=pvam_ch, out_features=self.char_num)
+
+    def forward(self, pvam_feature, gsrm_feature):
+        b, t, c1 = pvam_feature.shape
+        b, t, c2 = gsrm_feature.shape
+        combine_feature_ = paddle.concat([pvam_feature, gsrm_feature], axis=2)
+        img_comb_feature_ = paddle.reshape(combine_feature_, shape=[-1, c1 + c2])
+        img_comb_feature_map = self.fc0(img_comb_feature_)
+        img_comb_feature_map = F.sigmoid(img_comb_feature_map)
+        img_comb_feature_map = paddle.reshape(img_comb_feature_map, shape=[-1, t, c1])
+        combine_feature = (
+            img_comb_feature_map * pvam_feature
+            + (1.0 - img_comb_feature_map) * gsrm_feature
+        )
+        img_comb_feature = paddle.reshape(combine_feature, shape=[-1, c1])
+
+        out = self.fc1(img_comb_feature)
+        return out
+
+
+class SRNHead(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        max_text_length,
+        num_heads,
+        num_encoder_TUs,
+        num_decoder_TUs,
+        hidden_dims,
+        **kwargs,
+    ):
+        super(SRNHead, self).__init__()
+        self.char_num = out_channels
+        self.max_length = max_text_length
+        self.num_heads = num_heads
+        self.num_encoder_TUs = num_encoder_TUs
+        self.num_decoder_TUs = num_decoder_TUs
+        self.hidden_dims = hidden_dims
+
+        self.pvam = PVAM(
+            in_channels=in_channels,
+            char_num=self.char_num,
+            max_text_length=self.max_length,
+            num_heads=self.num_heads,
+            num_encoder_tus=self.num_encoder_TUs,
+            hidden_dims=self.hidden_dims,
+        )
+
+        self.gsrm = GSRM(
+            in_channels=in_channels,
+            char_num=self.char_num,
+            max_text_length=self.max_length,
+            num_heads=self.num_heads,
+            num_encoder_tus=self.num_encoder_TUs,
+            num_decoder_tus=self.num_decoder_TUs,
+            hidden_dims=self.hidden_dims,
+        )
+        self.vsfd = VSFD(in_channels=in_channels, char_num=self.char_num)
+
+        self.gsrm.wrap_encoder1.prepare_decoder.emb0 = (
+            self.gsrm.wrap_encoder0.prepare_decoder.emb0
+        )
+
+    def forward(self, inputs, targets=None):
+        others = targets[-4:]
+        encoder_word_pos = others[0]
+        gsrm_word_pos = others[1]
+        gsrm_slf_attn_bias1 = others[2]
+        gsrm_slf_attn_bias2 = others[3]
+
+        pvam_feature = self.pvam(inputs, encoder_word_pos, gsrm_word_pos)
+
+        gsrm_feature, word_out, gsrm_out = self.gsrm(
+            pvam_feature, gsrm_word_pos, gsrm_slf_attn_bias1, gsrm_slf_attn_bias2
+        )
+
+        final_out = self.vsfd(pvam_feature, gsrm_feature)
+        if not self.training:
+            final_out = F.softmax(final_out, axis=1)
+
+        _, decoded_out = paddle.topk(final_out, k=1)
+
+        predicts = OrderedDict(
+            [
+                ("predict", final_out),
+                ("pvam_feature", pvam_feature),
+                ("decoded_out", decoded_out),
+                ("word_out", word_out),
+                ("gsrm_out", gsrm_out),
+            ]
+        )
+
+        return predicts
diff --git a/ppocr/modeling/heads/rec_unimernet_head.py b/ppocr/modeling/heads/rec_unimernet_head.py
new file mode 100644
index 0000000..fb9f4ac
--- /dev/null
+++ b/ppocr/modeling/heads/rec_unimernet_head.py
@@ -0,0 +1,2674 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/opendatalab/UniMERNet/blob/main/unimernet/models/unimernet/configuration_unimernet_decoder.py
+"""
+
+import copy
+import math
+import re
+import numpy as np
+import inspect
+import warnings
+from collections import OrderedDict
+from typing import Optional, Tuple, Union, List, Dict, Any
+from dataclasses import dataclass, fields, is_dataclass
+
+import paddle
+import paddle.nn as nn
+from paddle import Tensor
+import paddle.nn.functional as F
+from paddle.nn import CrossEntropyLoss
+from paddle.nn.initializer import (
+    TruncatedNormal,
+    Constant,
+    Normal,
+    KaimingUniform,
+    XavierUniform,
+    XavierNormal,
+)
+
+zeros_ = Constant(value=0.0)
+ones_ = Constant(value=1.0)
+kaiming_normal_ = KaimingUniform(nonlinearity="relu")
+trunc_normal_ = TruncatedNormal(std=0.02)
+xavier_uniform_ = XavierUniform()
+xavier_normal_ = XavierNormal()
+
+
+class ModelOutput(OrderedDict):
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+    def __post_init__(self):
+        class_fields = fields(self)
+
+        if not len(class_fields):
+            raise ValueError(f"{self.__class__.__name__} has no fields.")
+        if not all(field.default is None for field in class_fields[1:]):
+            raise ValueError(
+                f"{self.__class__.__name__} should not have more than one required field."
+            )
+
+        first_field = getattr(self, class_fields[0].name)
+        other_fields_are_none = all(
+            getattr(self, field.name) is None for field in class_fields[1:]
+        )
+        if other_fields_are_none:
+            if isinstance(first_field, dict):
+                iterator = first_field.items()
+                first_field_iterator = True
+            else:
+                try:
+                    iterator = iter(first_field)
+                    first_field_iterator = True
+                except TypeError:
+                    first_field_iterator = False
+
+            if first_field_iterator:
+                for idx, element in enumerate(iterator):
+                    if (
+                        not isinstance(element, (list, tuple))
+                        or not len(element) == 2
+                        or not isinstance(element[0], str)
+                    ):
+                        if idx == 0:
+                            self[class_fields[0].name] = first_field
+                        else:
+                            raise ValueError(
+                                f"Cannot set key/value for {element}. It needs to be a tuple (key, value)."
+                            )
+                        break
+                    setattr(self, element[0], element[1])
+                    if element[1] is not None:
+                        self[element[0]] = element[1]
+            elif first_field is not None:
+                self[class_fields[0].name] = first_field
+        else:
+            for field in class_fields:
+                v = getattr(self, field.name)
+                if v is not None:
+                    self[field.name] = v
+
+    def __delitem__(self, *args, **kwargs):
+        raise Exception(
+            f"You cannot use ``__delitem__`` on a {self.__class__.__name__} instance."
+        )
+
+    def setdefault(self, *args, **kwargs):
+        raise Exception(
+            f"You cannot use ``setdefault`` on a {self.__class__.__name__} instance."
+        )
+
+    def pop(self, *args, **kwargs):
+        raise Exception(
+            f"You cannot use ``pop`` on a {self.__class__.__name__} instance."
+        )
+
+    def update(self, *args, **kwargs):
+        raise Exception(
+            f"You cannot use ``update`` on a {self.__class__.__name__} instance."
+        )
+
+    def __getitem__(self, k):
+        if isinstance(k, str):
+            inner_dict = dict(self.items())
+            return inner_dict[k]
+        else:
+            return self.to_tuple()[k]
+
+    def __setattr__(self, name, value):
+        if name in self.keys() and value is not None:
+            super().__setitem__(name, value)
+        super().__setattr__(name, value)
+
+    def __setitem__(self, key, value):
+        super().__setitem__(key, value)
+        super().__setattr__(key, value)
+
+    def __reduce__(self):
+        if not is_dataclass(self):
+            return super().__reduce__()
+        callable, _args, *remaining = super().__reduce__()
+        args = tuple(getattr(self, field.name) for field in fields(self))
+        return callable, args, *remaining
+
+    def to_tuple(self):
+        return tuple(self[k] for k in self.keys())
+
+
+@dataclass
+class BaseModelOutputWithPastAndCrossAttentions(ModelOutput):
+    last_hidden_state = None
+    past_key_values = None
+    hidden_states = None
+    attentions = None
+    cross_attentions = None
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+
+@dataclass
+class Seq2SeqLMOutput(ModelOutput):
+    loss = None
+    logits = None
+    past_key_values = None
+    decoder_hidden_states = None
+    decoder_attentions = None
+    cross_attentions = None
+    encoder_last_hidden_state = None
+    encoder_hidden_states = None
+    encoder_attentions = None
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+
+class MBartConfig(object):
+
+    model_type = "mbart"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    attribute_map = {
+        "num_attention_heads": "encoder_attention_heads",
+        "hidden_size": "d_model",
+    }
+
+    def __init__(
+        self,
+        vocab_size=50265,
+        max_position_embeddings=1024,
+        encoder_layers=12,
+        encoder_ffn_dim=4096,
+        encoder_attention_heads=16,
+        decoder_layers=12,
+        decoder_ffn_dim=4096,
+        decoder_attention_heads=16,
+        encoder_layerdrop=0.0,
+        decoder_layerdrop=0.0,
+        use_cache=True,
+        is_encoder_decoder=True,
+        activation_function="gelu",
+        d_model=1024,
+        dropout=0.1,
+        output_hidden_states=False,
+        use_return_dict=True,
+        attention_dropout=0.0,
+        activation_dropout=0.0,
+        init_std=0.02,
+        classifier_dropout=0.0,
+        scale_embedding=False,
+        pad_token_id=1,
+        bos_token_id=0,
+        eos_token_id=2,
+        forced_eos_token_id=2,
+        _attn_implementation="eager",
+        hidden_size=1024,
+        use_parallel=False,
+        parallel_step=2,
+        is_export=False,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.max_position_embeddings = max_position_embeddings
+        self.d_model = d_model
+        self.encoder_ffn_dim = encoder_ffn_dim
+        self.encoder_layers = encoder_layers
+        self.encoder_attention_heads = encoder_attention_heads
+        self.decoder_ffn_dim = decoder_ffn_dim
+        self.decoder_layers = decoder_layers
+        self.decoder_attention_heads = decoder_attention_heads
+        self.dropout = dropout
+        self.output_hidden_states = output_hidden_states
+        self.use_return_dict = use_return_dict
+        self.attention_dropout = attention_dropout
+        self.activation_dropout = activation_dropout
+        self.activation_function = activation_function
+        self.init_std = init_std
+        self.encoder_layerdrop = encoder_layerdrop
+        self.decoder_layerdrop = decoder_layerdrop
+        self.classifier_dropout = classifier_dropout
+        self.use_cache = use_cache
+        self.num_hidden_layers = encoder_layers
+        self.scale_embedding = (
+            scale_embedding  # scale factor will be sqrt(d_model) if True
+        )
+        self.pad_token_id = pad_token_id
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+        self.is_encoder_decoder = is_encoder_decoder
+        self.forced_eos_token_id = forced_eos_token_id
+        self._attn_implementation = _attn_implementation
+        self.use_parallel = use_parallel
+        self.parallel_step = parallel_step
+        self.is_export = is_export
+        super().__init__()
+
+
+@dataclass
+class AttentionMaskConverter:
+    """
+    A utility class for converting attention masks used in transformer models.
+
+    This class handles the conversion of attention masks based on whether the
+    attention mechanism is causal (i.e., preventing information flow from future
+    tokens to past tokens) and whether a sliding window approach is used.
+
+    Attributes:
+        is_causal (bool): Indicates if the attention mechanism is causal.
+        sliding_window (Optional[int]): Specifies the size of the sliding window
+                                        for local attention, if applicable.
+
+    Args:
+        is_causal (bool): Determines if the attention mask should enforce causality.
+        sliding_window (Optional[int], optional): The size of the sliding window
+                                                  for local attention. Default is None.
+    """
+
+    is_causal: bool
+    sliding_window: int
+
+    def __init__(self, is_causal: bool, sliding_window=None):
+        self.is_causal = is_causal
+        self.sliding_window = sliding_window
+
+        if self.sliding_window is not None and self.sliding_window <= 0:
+            raise ValueError(
+                f"Make sure that when passing `sliding_window` that its value is a strictly positive integer, not `{self.sliding_window}`"
+            )
+
+    @staticmethod
+    def _make_causal_mask(
+        input_ids_shape,
+        dtype,
+        past_key_values_length=0,
+        sliding_window=None,
+        is_export=False,
+    ):
+        bsz, tgt_len = input_ids_shape
+        if is_export:
+            mask = paddle.full(
+                (tgt_len, tgt_len), paddle.finfo(dtype).min, dtype="float64"
+            )
+        else:
+            mask = paddle.full((tgt_len, tgt_len), paddle.finfo(dtype).min)
+        mask_cond = paddle.arange(mask.shape[-1])
+        mask = mask.masked_fill_(
+            mask_cond < (mask_cond + 1).reshape([mask.shape[-1], 1]), 0
+        )
+        return mask[None, None, :, :].expand(
+            [bsz, 1, tgt_len, tgt_len + past_key_values_length]
+        )
+
+    def to_4d_export(
+        self,
+        attention_mask_2d,
+        query_length,
+        dtype,
+        key_value_length,
+        is_export=False,
+    ):
+        input_shape = (attention_mask_2d.shape[0], query_length)
+        expanded_attn_mask = self._expand_mask(
+            attention_mask_2d, dtype, tgt_len=input_shape[-1]
+        )
+        expanded_4d_mask = expanded_attn_mask
+
+        return expanded_4d_mask
+
+    def to_4d(
+        self,
+        attention_mask_2d,
+        query_length,
+        dtype,
+        key_value_length,
+        is_export=False,
+    ):
+
+        input_shape = (attention_mask_2d.shape[0], query_length)
+        causal_4d_mask = None
+        if (input_shape[-1] > 1 or self.sliding_window is not None) and self.is_causal:
+            if key_value_length is None:
+                raise ValueError(
+                    "This attention mask converter is causal. Make sure to pass `key_value_length` to correctly create a causal mask."
+                )
+
+            past_key_values_length = key_value_length - query_length
+
+            causal_4d_mask = self._make_causal_mask(
+                input_shape,
+                dtype,
+                past_key_values_length=past_key_values_length,
+                sliding_window=self.sliding_window,
+                is_export=is_export,
+            )
+        elif self.sliding_window is not None:
+            raise NotImplementedError(
+                "Sliding window is currently only implemented for causal masking"
+            )
+
+        expanded_attn_mask = self._expand_mask(
+            attention_mask_2d, dtype, tgt_len=input_shape[-1]
+        )
+
+        if causal_4d_mask is not None:
+            if is_export:
+                expanded_attn_mask = causal_4d_mask
+                return expanded_attn_mask
+            else:
+                expanded_attn_mask = causal_4d_mask.masked_fill_(
+                    expanded_attn_mask.cast(paddle.bool), paddle.finfo(dtype).min
+                )
+
+        expanded_4d_mask = expanded_attn_mask
+
+        return expanded_4d_mask
+
+    def _expand_mask(self, mask, dtype, tgt_len=None):
+        bsz, src_len = mask.shape
+        tgt_len = tgt_len if tgt_len is not None else src_len
+        expanded_mask = (
+            mask[:, None, None, :].expand([bsz, 1, tgt_len, src_len]).cast(dtype)
+        )
+        inverted_mask = 1.0 - expanded_mask
+        return inverted_mask.masked_fill_(
+            inverted_mask.cast(paddle.bool), paddle.finfo(dtype).min
+        )
+
+
+def _prepare_4d_attention_mask(mask, dtype, tgt_len=None):
+    return AttentionMaskConverter._expand_mask(mask=mask, dtype=dtype, tgt_len=tgt_len)
+
+
+def _prepare_4d_causal_attention_mask_export(
+    attention_mask,
+    input_shape,
+    inputs_embeds,
+    past_key_values_length,
+    sliding_window=None,
+    is_export=False,
+):
+
+    attn_mask_converter = AttentionMaskConverter(
+        is_causal=True, sliding_window=sliding_window
+    )
+    key_value_length = input_shape[-1] + past_key_values_length
+
+    shape = attention_mask.shape
+    len_shape = len(shape)
+
+    attention_mask = attn_mask_converter.to_4d_export(
+        attention_mask,
+        input_shape[-1],
+        key_value_length=key_value_length,
+        dtype=inputs_embeds.dtype,
+        is_export=is_export,
+    )
+    return attention_mask
+
+
+def _prepare_4d_causal_attention_mask(
+    attention_mask,
+    input_shape,
+    inputs_embeds,
+    past_key_values_length,
+    sliding_window=None,
+    is_export=False,
+):
+
+    attn_mask_converter = AttentionMaskConverter(
+        is_causal=True, sliding_window=sliding_window
+    )
+    key_value_length = input_shape[-1] + past_key_values_length
+
+    shape = attention_mask.shape
+    len_shape = len(shape)
+    if (attention_mask is not None) and (len_shape == 2):
+        attention_mask = attn_mask_converter.to_4d(
+            attention_mask,
+            input_shape[-1],
+            key_value_length=key_value_length,
+            dtype=inputs_embeds.dtype,
+            is_export=is_export,
+        )
+
+        return attention_mask
+    elif attention_mask is not None and len(attention_mask.shape) == 4:
+        expected_shape = (input_shape[0], 1, input_shape[1], key_value_length)
+        if tuple(attention_mask.shape) != expected_shape:
+            raise ValueError(
+                f"Incorrect 4D attention_mask shape: {tuple(attention_mask.shape)}; expected: {expected_shape}."
+            )
+        else:
+            inverted_mask = 1.0 - attention_mask
+            attention_mask = inverted_mask.masked_fill_(
+                inverted_mask.to(paddle.bool), paddle.finfo(inputs_embeds.dtype).min
+            )
+    else:
+        attention_mask = attn_mask_converter.to_causal_4d(
+            input_shape[0],
+            input_shape[-1],
+            key_value_length,
+            dtype=inputs_embeds.dtype,
+        )
+
+    return attention_mask
+
+
+class MBartLearnedPositionalEmbedding(nn.Embedding):
+    """
+    This module learns positional embeddings up to a fixed maximum size.
+    """
+
+    def __init__(self, num_embeddings, embedding_dim):
+        self.offset = 2
+        super().__init__(num_embeddings + self.offset, embedding_dim)
+
+    def forward(self, input_ids, past_key_values_length=0):
+        """`input_ids' shape is expected to be [bsz x seqlen]."""
+        bsz, seq_len = input_ids.shape[:2]
+        positions = paddle.arange(
+            past_key_values_length, past_key_values_length + seq_len, dtype=paddle.int64
+        ).expand([bsz, -1])
+        return nn.Embedding.forward(self, positions + self.offset)
+
+
+class MBartPreTrainedModel(nn.Layer):
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["MBartDecoderLayer", "MBartAttention"]
+    _supports_flash_attn_2 = True
+
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+
+    def _initialize_weights(self, module):
+        """
+        Initialize the weights if they are not already initialized.
+        """
+        if getattr(module, "_is_hf_initialized", False):
+            return
+        self._init_weights(module)
+
+    def post_init(self):
+        self.apply(self._initialize_weights)
+
+    def _init_weights(self, module):
+        std = self.config.init_std
+        normal_ = Normal(mean=0.0, std=std)
+        if isinstance(module, nn.Linear):
+            normal_(module.weight)
+            if module.bias is not None:
+                zeros_(module.bias)
+        elif isinstance(module, nn.Embedding):
+            normal_(module.weight)
+            if module._padding_idx is not None:
+                zeros_(module.weight[module._padding_idx])
+
+    @property
+    def dummy_inputs(self):
+        pad_token = self.config.pad_token_id
+        input_ids = paddle.tensor([[0, 6, 10, 4, 2], [0, 8, 12, 2, pad_token]])
+        dummy_inputs = {
+            "attention_mask": input_ids.ne(pad_token),
+            "input_ids": input_ids,
+        }
+        return dummy_inputs
+
+
+class MBartAttention(nn.Layer):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(
+        self,
+        embed_dim,
+        num_heads,
+        dropout: float = 0.0,
+        is_decoder: bool = False,
+        bias: bool = True,
+        is_causal: bool = False,
+        config=None,
+    ):
+        super().__init__()
+        self.embed_dim = embed_dim
+        self.num_heads = num_heads
+        self.dropout = dropout
+        self.head_dim = embed_dim // num_heads
+        self.config = config
+
+        if (self.head_dim * num_heads) != self.embed_dim:
+            raise ValueError(
+                f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim}"
+                f" and `num_heads`: {num_heads})."
+            )
+        self.scaling = self.head_dim**-0.5
+        self.is_decoder = is_decoder
+        self.is_causal = is_causal
+
+        self.k_proj = nn.Linear(embed_dim, embed_dim, bias_attr=bias)
+        self.v_proj = nn.Linear(embed_dim, embed_dim, bias_attr=bias)
+        self.q_proj = nn.Linear(embed_dim, embed_dim, bias_attr=bias)
+        self.out_proj = nn.Linear(embed_dim, embed_dim, bias_attr=bias)
+
+    def _shape(self, tensor, seq_len, bsz):
+        return tensor.reshape([bsz, seq_len, self.num_heads, self.head_dim]).transpose(
+            [0, 2, 1, 3]
+        )
+
+    def forward(
+        self,
+        hidden_states,
+        key_value_states=None,
+        past_key_value=None,
+        attention_mask=None,
+        layer_head_mask=None,
+        output_attentions=False,
+    ):
+
+        is_cross_attention = key_value_states is not None
+
+        bsz, tgt_len, _ = paddle.shape(hidden_states)
+        query_states = self.q_proj(hidden_states) * self.scaling
+        if (
+            is_cross_attention
+            and past_key_value is not None
+            and past_key_value[0].shape[2] == key_value_states.shape[1]
+        ):
+            key_states = past_key_value[0]
+            value_states = past_key_value[1]
+        elif is_cross_attention:
+            key_states = self._shape(self.k_proj(key_value_states), -1, bsz)
+            value_states = self._shape(self.v_proj(key_value_states), -1, bsz)
+        elif past_key_value is not None:
+            key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
+            value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
+            key_states = paddle.concat([past_key_value[0], key_states], axis=2)
+            value_states = paddle.concat([past_key_value[1], value_states], axis=2)
+        else:
+            key_states = self._shape(self.k_proj(hidden_states), -1, bsz)
+            value_states = self._shape(self.v_proj(hidden_states), -1, bsz)
+
+        if self.is_decoder:
+            past_key_value = (key_states, value_states)
+
+        proj_shape = (bsz * self.num_heads, -1, self.head_dim)
+        query_states = self._shape(query_states, tgt_len, bsz).reshape(proj_shape)
+        key_states = key_states.reshape(proj_shape)
+        value_states = value_states.reshape(proj_shape)
+
+        src_len = key_states.shape[1]
+        attn_weights = paddle.bmm(query_states, key_states.transpose([0, 2, 1]))
+
+        if attention_mask is not None:
+            attn_weights = (
+                attn_weights.reshape([bsz, self.num_heads, tgt_len, src_len])
+                + attention_mask
+            )
+            attn_weights = attn_weights.reshape(
+                [bsz * self.num_heads, tgt_len, src_len]
+            )
+
+        attn_weights = nn.functional.softmax(attn_weights, axis=-1)
+        if layer_head_mask is not None:
+            if tuple(layer_head_mask.shape) != (self.num_heads,):
+                raise ValueError(
+                    f"Head mask for a single layer should be of shape {(self.num_heads,)}, but is"
+                    f" {layer_head_mask.shape}"
+                )
+            attn_weights = layer_head_mask.reshape(
+                [1, -1, 1, 1]
+            ) * attn_weights.reshape([bsz, self.num_heads, tgt_len, src_len])
+            attn_weights = attn_weights.reshape(
+                [bsz * self.num_heads, tgt_len, src_len]
+            )
+
+        if output_attentions:
+            attn_weights_reshaped = attn_weights.reshape(
+                [bsz, self.num_heads, tgt_len, src_len]
+            )
+            attn_weights = attn_weights_reshaped.reshape(
+                [bsz * self.num_heads, tgt_len, src_len]
+            )
+        else:
+            attn_weights_reshaped = None
+        attn_probs = nn.functional.dropout(
+            attn_weights, p=self.dropout, training=self.training
+        )
+        attn_output = paddle.bmm(attn_probs, value_states)
+
+        attn_output = attn_output.reshape([bsz, self.num_heads, tgt_len, self.head_dim])
+        attn_output = attn_output.transpose([0, 2, 1, 3])
+
+        attn_output = attn_output.reshape([bsz, tgt_len, self.embed_dim])
+        attn_output = self.out_proj(attn_output)
+        return attn_output, attn_weights_reshaped, past_key_value
+
+
+MBART_ATTENTION_CLASSES = {
+    "eager": MBartAttention,
+}
+
+
+class MBartDecoderLayer(nn.Layer):
+    def __init__(self, config):
+        super().__init__()
+        self.embed_dim = config.d_model
+        self.self_attn = MBART_ATTENTION_CLASSES[config._attn_implementation](
+            embed_dim=self.embed_dim,
+            num_heads=config.decoder_attention_heads,
+            dropout=config.attention_dropout,
+            is_decoder=True,
+            is_causal=True,
+            config=config,
+        )
+        self.is_export = config.is_export
+        self.dropout = config.dropout
+        self.activation_fn = F.gelu
+        self.activation_dropout = config.activation_dropout
+
+        self.self_attn_layer_norm = nn.LayerNorm(self.embed_dim)
+        self.encoder_attn = MBART_ATTENTION_CLASSES[config._attn_implementation](
+            self.embed_dim,
+            config.decoder_attention_heads,
+            dropout=config.attention_dropout,
+            is_decoder=True,
+            config=config,
+        )
+        self.encoder_attn_layer_norm = nn.LayerNorm(self.embed_dim)
+        self.fc1 = nn.Linear(self.embed_dim, config.decoder_ffn_dim)
+        self.fc2 = nn.Linear(config.decoder_ffn_dim, self.embed_dim)
+        self.final_layer_norm = nn.LayerNorm(self.embed_dim)
+
+    def forward(
+        self,
+        hidden_states,
+        attention_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        layer_head_mask=None,
+        cross_attn_layer_head_mask=None,
+        past_key_value: Optional[Tuple[paddle.Tensor]] = None,
+        output_attentions: Optional[bool] = False,
+        use_cache: Optional[bool] = True,
+    ) -> paddle.Tensor:
+
+        residual = hidden_states
+        hidden_states = self.self_attn_layer_norm(hidden_states)
+        self_attn_past_key_value = (
+            past_key_value[:2] if past_key_value is not None else None
+        )
+
+        hidden_states, self_attn_weights, present_key_value = self.self_attn(
+            hidden_states=hidden_states,
+            past_key_value=self_attn_past_key_value,
+            attention_mask=attention_mask,
+            layer_head_mask=layer_head_mask,
+            output_attentions=output_attentions,
+        )
+        hidden_states = nn.functional.dropout(
+            hidden_states, p=self.dropout, training=self.training
+        )
+        hidden_states = residual + hidden_states
+
+        cross_attn_present_key_value = None
+        cross_attn_weights = None
+        if encoder_hidden_states is not None:
+            residual = hidden_states
+            hidden_states = self.encoder_attn_layer_norm(hidden_states)
+            cross_attn_past_key_value = (
+                past_key_value[-2:] if past_key_value is not None else None
+            )
+            (
+                hidden_states,
+                cross_attn_weights,
+                cross_attn_present_key_value,
+            ) = self.encoder_attn(
+                hidden_states=hidden_states,
+                key_value_states=encoder_hidden_states,
+                attention_mask=encoder_attention_mask,
+                layer_head_mask=cross_attn_layer_head_mask,
+                past_key_value=cross_attn_past_key_value,
+                output_attentions=output_attentions,
+            )
+            hidden_states = nn.functional.dropout(
+                hidden_states, p=self.dropout, training=self.training
+            )
+            hidden_states = residual + hidden_states
+
+            present_key_value = present_key_value + cross_attn_present_key_value
+
+        residual = hidden_states
+        hidden_states = self.final_layer_norm(hidden_states)
+        hidden_states = self.activation_fn(self.fc1(hidden_states))
+        hidden_states = nn.functional.dropout(
+            hidden_states, p=self.activation_dropout, training=self.training
+        )
+        hidden_states = self.fc2(hidden_states)
+        hidden_states = nn.functional.dropout(
+            hidden_states, p=self.dropout, training=self.training
+        )
+        hidden_states = residual + hidden_states
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (self_attn_weights, cross_attn_weights)
+
+        if self.is_export:
+            outputs += (present_key_value,)
+        else:
+            if use_cache:
+                outputs += (present_key_value,)
+        return outputs
+
+
+class MBartForCausalLM(MBartPreTrainedModel):
+    _tied_weights_keys = ["lm_head.weight"]
+
+    def __init__(self, config):
+        config = copy.deepcopy(config)
+        config.is_decoder = True
+        config.is_encoder_decoder = False
+        super().__init__(config)
+        self.model = MBartDecoderWrapper(config)
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias_attr=False)
+
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.decoder.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.decoder.embed_tokens = value
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def set_decoder(self, decoder):
+        self.model.decoder = decoder
+
+    def get_decoder(self):
+        return self.model.decoder
+
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        head_mask=None,
+        cross_attn_head_mask=None,
+        past_key_values=None,
+        inputs_embeds=None,
+        labels=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+
+        output_attentions = (
+            output_attentions
+            if output_attentions is not None
+            else self.config.output_attentions
+        )
+        output_hidden_states = (
+            output_hidden_states
+            if output_hidden_states is not None
+            else self.config.output_hidden_states
+        )
+        return_dict = (
+            return_dict if return_dict is not None else self.config.use_return_dict
+        )
+
+        outputs = self.model.decoder(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+            head_mask=head_mask,
+            cross_attn_head_mask=cross_attn_head_mask,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        logits = self.lm_head(outputs[0])
+
+        loss = None
+        if labels is not None:
+            labels = labels
+            loss_fct = CrossEntropyLoss()
+            loss = loss_fct(
+                logits.reshape([-1, self.config.vocab_size]), labels.reshape([-1])
+            )
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return CausalLMOutputWithCrossAttentions(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+            cross_attentions=outputs.cross_attentions,
+        )
+
+    def prepare_inputs_for_generation(
+        self,
+        input_ids,
+        past_key_values=None,
+        attention_mask=None,
+        use_cache=None,
+        **kwargs,
+    ):
+        if attention_mask is None:
+            attention_mask = input_ids.new_ones(input_ids.shape)
+
+        if past_key_values:
+            past_length = past_key_values[0][0].shape[2]
+
+            if input_ids.shape[1] > past_length:
+                remove_prefix_length = past_length
+            else:
+                remove_prefix_length = input_ids.shape[1] - 1
+
+            input_ids = input_ids[:, remove_prefix_length:]
+        return {
+            "input_ids": input_ids,
+            "attention_mask": attention_mask,
+            "past_key_values": past_key_values,
+            "use_cache": use_cache,
+        }
+
+    @staticmethod
+    def _reorder_cache(past_key_values, beam_idx):
+        reordered_past = ()
+        for layer_past in past_key_values:
+            reordered_past += (
+                tuple(
+                    past_state.index_select(0, beam_idx) for past_state in layer_past
+                ),
+            )
+        return reordered_past
+
+
+class myLayerNorm(nn.LayerNorm):
+    """
+    Custom implementation of Layer Normalization, with additional options.
+
+    This class extends the standard LayerNorm to include optional features,
+    such as drop block regularization, which might be used for improving
+    model generalization.
+
+    Args:
+        num_channels (int): The number of features or channels in the input.
+        eps (float, optional): A small value added to the denominator for numerical stability. Default is 1e-5.
+        affine (bool, optional): If True, this module has learnable affine parameters (gamma and beta). Default is True.
+        drop_block (optional): Additional regularization technique that might be applied. Default is None.
+
+    """
+
+    def __init__(
+        self,
+        num_channels,
+        eps=1e-5,
+        affine=True,
+        drop_block=None,
+    ):
+        super(nn.LayerNorm, self).__init__()
+        self._epsilon = eps
+        self.num_channels = num_channels
+        if affine:
+            self.weight = paddle.create_parameter([num_channels], dtype="float32")
+            self.bias = paddle.create_parameter([num_channels], dtype="float32")
+            ones_(self.weight)
+            zeros_(self.bias)
+
+    def forward(self, x):
+        x = F.layer_norm(
+            x,
+            self.num_channels,
+            weight=self.weight,
+            bias=self.bias,
+            epsilon=self._epsilon,
+        )
+        return x
+
+
+class MBartDecoder(MBartPreTrainedModel):
+    """
+    Transformer decoder consisting of *config.decoder_layers* layers. Each layer is a [`MBartDecoderLayer`]
+
+    Args:
+        config
+        embed_tokens (nn.Embedding): output embedding
+    """
+
+    def __init__(self, config, embed_tokens=None):
+        super().__init__(config)
+        self.dropout = config.dropout
+        self.layerdrop = config.decoder_layerdrop
+        self.padding_idx = config.pad_token_id
+        self.max_target_positions = config.max_position_embeddings
+        self.embed_scale = math.sqrt(config.d_model) if config.scale_embedding else 1.0
+
+        self.embed_tokens = nn.Embedding(
+            config.vocab_size, config.d_model, self.padding_idx
+        )
+
+        if embed_tokens is not None:
+            self.embed_tokens.weight = embed_tokens.weight
+
+        self.embed_positions = MBartLearnedPositionalEmbedding(
+            config.max_position_embeddings,
+            config.d_model,
+        )
+        self.layers = nn.LayerList(
+            [MBartDecoderLayer(config) for _ in range(config.decoder_layers)]
+        )
+        self._use_flash_attention_2 = config._attn_implementation == "flash_attention_2"
+        self.layernorm_embedding = myLayerNorm(config.d_model, affine=True)
+        self.layer_norm = nn.LayerNorm(config.d_model)
+
+        self.gradient_checkpointing = False
+        # Initialize weights and apply final processing
+        self.post_init()
+        self.is_export = config.is_export
+
+    def get_input_embeddings(self):
+        return self.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.embed_tokens = value
+
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        head_mask=None,
+        cross_attn_head_mask=None,
+        past_key_values=None,
+        inputs_embeds=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+
+        output_attentions = (
+            output_attentions
+            if output_attentions is not None
+            else self.config.output_attentions
+        )
+        output_hidden_states = (
+            output_hidden_states
+            if output_hidden_states is not None
+            else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = (
+            return_dict if return_dict is not None else self.config.use_return_dict
+        )
+
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError(
+                "You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time"
+            )
+        elif input_ids is not None:
+            input = input_ids
+            input_shape = input.shape
+            input_ids = input_ids.reshape([-1, input_shape[-1]])
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.shape[:-1]
+            input = inputs_embeds[:, :, -1]
+        else:
+            raise ValueError(
+                "You have to specify either decoder_input_ids or decoder_inputs_embeds"
+            )
+
+        past_key_values_length = (
+            past_key_values[0][0].shape[2] if past_key_values is not None else 0
+        )
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids) * self.embed_scale
+
+        if self._use_flash_attention_2:
+            attention_mask = (
+                attention_mask
+                if (attention_mask is not None and 0 in attention_mask)
+                else None
+            )
+        else:
+            attention_mask = _prepare_4d_causal_attention_mask(
+                attention_mask,
+                input_shape,
+                inputs_embeds,
+                past_key_values_length,
+                is_export=self.is_export,
+            )
+
+        if encoder_hidden_states is not None and encoder_attention_mask is not None:
+            if self._use_flash_attention_2:
+                encoder_attention_mask = (
+                    encoder_attention_mask if 0 in encoder_attention_mask else None
+                )
+            else:
+                encoder_attention_mask = _prepare_4d_attention_mask(
+                    encoder_attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]
+                )
+
+        # embed positions
+        positions = self.embed_positions(input, past_key_values_length)
+
+        hidden_states = inputs_embeds + positions
+        hidden_states = self.layernorm_embedding(hidden_states)
+
+        hidden_states = nn.functional.dropout(
+            hidden_states, p=self.dropout, training=self.training
+        )
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                print(
+                    "`use_cache=True` is incompatible with gradient checkpointing`. Setting `use_cache=False`..."
+                )
+                use_cache = False
+
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        all_cross_attentions = (
+            () if (output_attentions and encoder_hidden_states is not None) else None
+        )
+        next_decoder_cache = () if use_cache else None
+
+        for attn_mask, mask_name in zip(
+            [head_mask, cross_attn_head_mask], ["head_mask", "cross_attn_head_mask"]
+        ):
+            if attn_mask is not None:
+                if attn_mask.shape[0] != len(self.layers):
+                    raise ValueError(
+                        f"The `{mask_name}` should be specified for {len(self.layers)} layers, but it is for"
+                        f" {attn_mask.shape[0]}."
+                    )
+
+        for idx, decoder_layer in enumerate(self.layers):
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+            if self.training:
+                dropout_probability = paddle.rand([])
+                if dropout_probability < self.layerdrop:
+                    continue
+
+            past_key_value = (
+                past_key_values[idx] if past_key_values is not None else None
+            )
+
+            if self.gradient_checkpointing and self.training:
+                layer_outputs = self._gradient_checkpointing_func(
+                    decoder_layer.__call__,
+                    hidden_states,
+                    attention_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                    head_mask[idx] if head_mask is not None else None,
+                    (
+                        cross_attn_head_mask[idx]
+                        if cross_attn_head_mask is not None
+                        else None
+                    ),
+                    None,
+                    output_attentions,
+                    use_cache,
+                )
+            else:
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    attention_mask=attention_mask,
+                    encoder_hidden_states=encoder_hidden_states,
+                    encoder_attention_mask=encoder_attention_mask,
+                    layer_head_mask=(head_mask[idx] if head_mask is not None else None),
+                    cross_attn_layer_head_mask=(
+                        cross_attn_head_mask[idx]
+                        if cross_attn_head_mask is not None
+                        else None
+                    ),
+                    past_key_value=past_key_value,
+                    output_attentions=output_attentions,
+                    use_cache=use_cache,
+                )
+            hidden_states = layer_outputs[0]
+
+            if use_cache:
+                next_decoder_cache += (layer_outputs[3 if output_attentions else 1],)
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+                if encoder_hidden_states is not None:
+                    all_cross_attentions += (layer_outputs[2],)
+
+        hidden_states = self.layer_norm(hidden_states)
+
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        next_cache = next_decoder_cache if use_cache else None
+        if not return_dict:
+            return tuple(
+                v
+                for v in [
+                    hidden_states,
+                    next_cache,
+                    all_hidden_states,
+                    all_self_attns,
+                    all_cross_attentions,
+                ]
+                if v is not None
+            )
+        return BaseModelOutputWithPastAndCrossAttentions(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+            cross_attentions=all_cross_attentions,
+        )
+
+
+class MBartDecoderWrapper(MBartPreTrainedModel):
+    """
+    This wrapper class is a helper class to correctly load pretrained checkpoints when the causal language model is
+    used in combination with the [`EncoderDecoderModel`] framework.
+    """
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.decoder = MBartDecoder(config)
+
+    def forward(self, *args, **kwargs):
+        return self.decoder(*args, **kwargs)
+
+
+def _in_projection(
+    q: paddle.Tensor,
+    k: paddle.Tensor,
+    v: paddle.Tensor,
+    w_q: paddle.Tensor,
+    w_k: paddle.Tensor,
+    w_v: paddle.Tensor,
+    b_q: Optional[paddle.Tensor] = None,
+    b_k: Optional[paddle.Tensor] = None,
+    b_v: Optional[paddle.Tensor] = None,
+) -> Tuple[paddle.Tensor, paddle.Tensor, paddle.Tensor]:
+
+    Eq, Ek, Ev = q.shape[-1], k.shape[-1], v.shape[-1]
+    assert w_q.shape == (
+        Eq,
+        Eq,
+    ), f"expecting query weights shape of {(Eq, Eq)}, but got {w_q.shape}"
+    assert w_k.shape == (
+        Eq,
+        Ek,
+    ), f"expecting key weights shape of {(Eq, Ek)}, but got {w_k.shape}"
+    assert w_v.shape == (
+        Eq,
+        Ev,
+    ), f"expecting value weights shape of {(Eq, Ev)}, but got {w_v.shape}"
+    assert b_q is None or b_q.shape == (
+        Eq,
+    ), f"expecting query bias shape of {(Eq,)}, but got {b_q.shape}"
+    assert b_k is None or b_k.shape == (
+        Eq,
+    ), f"expecting key bias shape of {(Eq,)}, but got {b_k.shape}"
+    assert b_v is None or b_v.shape == (
+        Eq,
+    ), f"expecting value bias shape of {(Eq,)}, but got {b_v.shape}"
+    return linear(q, w_q.T, b_q), linear(k, w_k.T, b_k), linear(v, w_v.T, b_v)
+
+
+def _scaled_dot_product_attention(
+    q: paddle.Tensor,
+    k: paddle.Tensor,
+    v: paddle.Tensor,
+    attn_mask: Optional[paddle.Tensor] = None,
+    dropout_p: float = 0.0,
+) -> Tuple[paddle.Tensor, paddle.Tensor]:
+
+    B, Nt, E = q.shape
+    q = q / math.sqrt(E)
+    attn = paddle.bmm(q, k.transpose([0, 2, 1]))
+    if attn_mask is not None:
+        attn += attn_mask
+    attn = F.softmax(attn, axis=-1)
+    if dropout_p > 0.0:
+        attn = F.dropout(attn, p=dropout_p)
+    output = paddle.bmm(attn, v)
+    return output, attn
+
+
+def linear(x, w, b, is_transpose):
+
+    if b is not None:
+        return paddle.matmul(x, w, transpose_y=is_transpose) + b
+    else:
+        return paddle.matmul(x, w, transpose_y=is_transpose)
+
+
+def _in_projection_packed(
+    q: Tensor,
+    k: Tensor,
+    v: Tensor,
+    w: Tensor,
+    b: Optional[Tensor] = None,
+    is_export=False,
+) -> List[Tensor]:
+
+    E = paddle.shape(q)[-1]
+    if k is v:
+        if q is k:
+            proj = linear(q, w, b, is_transpose=True)
+            if is_export:
+                B, D, L = paddle.shape(proj)
+                proj = proj.reshape([B, D, 3, E])
+                proj = (
+                    proj.unsqueeze(0)
+                    .transpose([3, 1, 2, 0, 4])
+                    .squeeze(-2)
+                    .contiguous()
+                )
+            else:
+                proj = (
+                    proj.unflatten(-1, (3, E))
+                    .unsqueeze(0)
+                    .transpose([3, 1, 2, 0, 4])
+                    .squeeze(-2)
+                    .contiguous()
+                )
+            return proj[0], proj[1], proj[2]
+    else:
+        w_q, w_k, w_v = w.chunk(3)
+        if b is None:
+            b_q = b_k = b_v = None
+        else:
+            b_q, b_k, b_v = b.chunk(3)
+        return linear(q, w_q, b_q), linear(k, w_k, b_k), linear(v, w_v, b_v)
+
+
+def multi_head_attention_forward(
+    query: paddle.Tensor,
+    key: paddle.Tensor,
+    value: paddle.Tensor,
+    embed_dim_to_check: int,
+    num_heads: int,
+    in_proj_weight: paddle.Tensor,
+    in_proj_bias: Optional[paddle.Tensor],
+    bias_k: Optional[paddle.Tensor],
+    bias_v: Optional[paddle.Tensor],
+    add_zero_attn: bool,
+    dropout_p: float,
+    out_proj_weight: paddle.Tensor,
+    out_proj_bias: Optional[paddle.Tensor],
+    training: bool = True,
+    key_padding_mask: Optional[paddle.Tensor] = None,
+    need_weights: bool = True,
+    attn_mask: Optional[paddle.Tensor] = None,
+    use_separate_proj_weight: bool = False,
+    q_proj_weight: Optional[paddle.Tensor] = None,
+    k_proj_weight: Optional[paddle.Tensor] = None,
+    v_proj_weight: Optional[paddle.Tensor] = None,
+    static_k: Optional[paddle.Tensor] = None,
+    static_v: Optional[paddle.Tensor] = None,
+    is_export=False,
+):
+
+    tgt_len, bsz, embed_dim = query.shape
+    src_len, _, _ = key.shape
+
+    if isinstance(embed_dim, paddle.Tensor):
+        head_dim = embed_dim.div(num_heads, rounding_mode="trunc")
+    else:
+        head_dim = embed_dim // num_heads
+    q, k, v = _in_projection_packed(
+        query, key, value, in_proj_weight, in_proj_bias, is_export
+    )
+
+    if key_padding_mask is not None and key_padding_mask.dtype == paddle.uint8:
+        warnings.warn(
+            "Byte tensor for key_padding_mask in nn.MultiheadAttention is deprecated. Use bool tensor instead."
+        )
+        key_padding_mask = key_padding_mask.to(paddle.bool)
+
+    if bias_k is not None and bias_v is not None:  # False
+        assert static_k is None, "bias cannot be added to static key."
+        assert static_v is None, "bias cannot be added to static value."
+        k = paddle.concat([k, bias_k.repeat(1, bsz, 1)])
+        v = paddle.concat([v, bias_v.repeat(1, bsz, 1)])
+    else:
+        assert bias_k is None
+        assert bias_v is None
+
+    q = q.reshape([tgt_len, bsz * num_heads, head_dim]).transpose([1, 0, 2])
+    if static_k is None:  # True
+        k = k.reshape([k.shape[0], bsz * num_heads, head_dim]).transpose([1, 0, 2])
+    else:
+        assert (
+            static_k.shape[0] == bsz * num_heads
+        ), f"expecting static_k.size(0) of {bsz * num_heads}, but got {static_k.shape[0]}"
+        assert (
+            static_k.shape[2] == head_dim
+        ), f"expecting static_k.size(2) of {head_dim}, but got {static_k.shape[2]}"
+        k = static_k
+    if static_v is None:  # True
+        v = v.reshape([v.shape[0], bsz * num_heads, head_dim]).transpose([1, 0, 2])
+    else:
+        assert (
+            static_v.shape[0] == bsz * num_heads
+        ), f"expecting static_v.size(0) of {bsz * num_heads}, but got {static_v.shape[0]}"
+        assert (
+            static_v.shape[2] == head_dim
+        ), f"expecting static_v.size(2) of {head_dim}, but got {static_v.shape[2]}"
+        v = static_v
+
+    src_len = k.shape[1]
+
+    if not training:
+        dropout_p = 0.0
+
+    attn_output, attn_output_weights = _scaled_dot_product_attention(
+        q, k, v, attn_mask, dropout_p
+    )
+
+    attn_output = attn_output.transpose([1, 0, 2]).reshape([tgt_len, bsz, embed_dim])
+    attn_output = linear(
+        attn_output, out_proj_weight, out_proj_bias, is_transpose=False
+    )
+
+    if need_weights:
+        attn_output_weights = attn_output_weights.reshape(
+            [bsz, num_heads, tgt_len, src_len]
+        )
+        return attn_output, attn_output_weights.sum(axis=1) / num_heads
+    else:
+        return attn_output, None
+
+
+class MyMultiheadAttention(nn.Layer):
+    """
+    Custom implementation of a multi-head attention layer.
+
+    Attributes:
+        __constants__ (list): List of constant attributes.
+        bias_k (Optional[paddle.Tensor]): Optional tensor for key bias.
+        bias_v (Optional[paddle.Tensor]): Optional tensor for value bias.
+
+    Args:
+        embed_dim (int): Total dimension of the model. This is the size of the input feature vectors.
+        num_heads (int): Number of parallel attention heads. The input dimension must be divisible by the number of heads.
+        dropout (float, optional): Dropout probability on the attention weights. Default is 0.0.
+        bias (bool, optional): If True, adds a learnable bias to the output. Default is True.
+        add_bias_kv (bool, optional): If True, adds bias to the key and value sequences. Default is False.
+        add_zero_attn (bool, optional): If True, adds a zero attention head. Default is False.
+        kdim (int, optional): Total number of features for keys. If None, defaults to embed_dim.
+        vdim (int, optional): Total number of features for values. If None, defaults to embed_dim.
+        batch_first (bool, optional): If True, the input and output tensors are provided as (batch, seq, feature). Default is False.
+        device (optional): The device on which the layer's parameters should be initialized. Default is None.
+        dtype (optional): The data type for the parameters. Default is None.
+        is_export (bool, optional): If True, the layer is set up for export, potentially changing behavior for compatibility. Default is False.
+    """
+
+    __constants__ = ["batch_first"]
+    bias_k: Optional[paddle.Tensor]
+    bias_v: Optional[paddle.Tensor]
+
+    def __init__(
+        self,
+        embed_dim,
+        num_heads,
+        dropout=0.0,
+        bias=True,
+        add_bias_kv=False,
+        add_zero_attn=False,
+        kdim=None,
+        vdim=None,
+        batch_first=False,
+        device=None,
+        dtype=None,
+        is_export=False,
+    ) -> None:
+        super(MyMultiheadAttention, self).__init__()
+        self.embed_dim = embed_dim
+        self.kdim = kdim if kdim is not None else embed_dim
+        self.vdim = vdim if vdim is not None else embed_dim
+        self._qkv_same_embed_dim = self.kdim == embed_dim and self.vdim == embed_dim
+
+        self.num_heads = num_heads
+        self.dropout = dropout
+        self.batch_first = batch_first
+        self.head_dim = embed_dim // num_heads
+        self.is_export = is_export
+        assert (
+            self.head_dim * num_heads == self.embed_dim
+        ), "embed_dim must be divisible by num_heads"
+
+        if self._qkv_same_embed_dim is False:
+            pass
+        else:
+            if dtype is None:
+                dtype = paddle.float32
+            self.in_proj_weight = paddle.create_parameter(
+                (3 * embed_dim, embed_dim), dtype
+            )
+            self.q_proj_weight = None
+            self.k_proj_weight = None
+            self.v_proj_weight = None
+
+        if bias:
+            self.in_proj_bias = paddle.create_parameter((3 * embed_dim,), dtype)
+            zeros_(self.in_proj_bias)
+        else:
+            self.in_proj_bias = None
+        self.out_proj = nn.Linear(embed_dim, embed_dim, bias_attr=bias)
+
+        if add_bias_kv:
+            pass
+        else:
+            self.bias_k = self.bias_v = None
+
+        self.add_zero_attn = add_zero_attn
+
+        self._reset_parameters()
+
+    def _reset_parameters(self):
+
+        if self._qkv_same_embed_dim:
+            xavier_uniform_(self.in_proj_weight)
+        else:
+            xavier_uniform_(self.q_proj_weight)
+            xavier_uniform_(self.k_proj_weight)
+            xavier_uniform_(self.v_proj_weight)
+
+        if self.in_proj_bias is not None:
+            zeros_(self.in_proj_bias)
+            zeros_(self.out_proj.bias)
+        if self.bias_k is not None:
+            xavier_normal_(self.bias_k)
+        if self.bias_v is not None:
+            xavier_normal_(self.bias_v)
+
+    def forward(
+        self,
+        query: paddle.Tensor,
+        key: paddle.Tensor,
+        value: paddle.Tensor,
+        key_padding_mask: Optional[paddle.Tensor] = None,
+        need_weights: bool = True,
+        attn_mask: Optional[paddle.Tensor] = None,
+    ) -> Tuple[paddle.Tensor, Optional[paddle.Tensor]]:
+
+        attn_output, attn_output_weights = multi_head_attention_forward(
+            query,
+            key,
+            value,
+            self.embed_dim,
+            self.num_heads,
+            self.in_proj_weight,
+            self.in_proj_bias,
+            self.bias_k,
+            self.bias_v,
+            self.add_zero_attn,
+            self.dropout,
+            self.out_proj.weight,
+            self.out_proj.bias,
+            training=self.training,
+            key_padding_mask=key_padding_mask,
+            need_weights=need_weights,
+            attn_mask=attn_mask,
+            is_export=self.is_export,
+        )
+
+        return attn_output, attn_output_weights
+
+
+class LogitsProcessorList(list):
+    """
+    A list of logits processors that can be applied sequentially.
+
+    Methods:
+        __call__(input_ids, scores, **kwargs): Apply all processors to the given inputs.
+    """
+
+    def __call__(self, input_ids, scores, **kwargs):
+        for processor in self:
+            function_args = inspect.signature(processor.__call__).parameters
+            if len(function_args) > 2:
+                if not all(arg in kwargs for arg in list(function_args.keys())[2:]):
+                    raise ValueError(
+                        f"Make sure that all the required parameters: {list(function_args.keys())} for "
+                        f"{processor.__class__} are passed to the logits processor."
+                    )
+                scores = processor(input_ids, scores, **kwargs)
+            else:
+                scores = processor(input_ids, scores)
+        return scores
+
+
+class ForcedEOSTokenLogitsProcessor(object):
+    """
+    A processor that forces the generation of an end-of-sequence (EOS) token
+    at a specified position in the sequence.
+
+    This is typically used in language generation tasks to ensure that the
+    generated sequence ends properly when it reaches a certain length.
+
+    Args:
+        max_length (int): The maximum length of the sequence. Forces EOS when this length is reached.
+        eos_token_id (Union[int, List[int]]): The ID(s) of the EOS token(s) to be forced in the sequence.
+    """
+
+    def __init__(self, max_length: int, eos_token_id: Union[int, List[int]]):
+        self.max_length = max_length
+        if isinstance(eos_token_id, int):
+            eos_token_id = [eos_token_id]
+        self.eos_token_id = eos_token_id
+
+    def __call__(self, input_ids, scores):
+        cur_len = input_ids.shape[-1]
+        scores_processed = scores
+        if cur_len == self.max_length - 1:
+            scores_processed = paddle.full_like(scores, -math.inf)
+            scores_processed[:, self.eos_token_id] = 0
+        return scores_processed
+
+
+@dataclass
+class CausalLMOutputWithCrossAttentions(ModelOutput):
+
+    loss = None
+    logits = None
+    past_key_values = None
+    hidden_states = None
+    attentions = None
+    cross_attentions = None
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+
+@dataclass
+class CausalLMOutputWithCrossAttentionsAndCounting(ModelOutput):
+    """
+    Base class for causal language model (or autoregressive) outputs.
+    """
+
+    logits = None
+    counting = None
+    past_key_values = None
+    hidden_states = None
+    attentions = None
+    cross_attentions = None
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+
+class CustomMBartDecoder(MBartDecoder):
+    """
+    A custom MBartDecoder that includes additional processing layers.
+
+    This class extends the MBartDecoder by adding a customizable neural network
+    component called `counting_context_weight`, which applies a series of linear
+    transformations followed by ReLU activations. This can be used to modify or
+    enhance the decoder's behavior for specific tasks.
+
+    Args:
+        config: The configuration object containing model parameters.
+    """
+
+    def __init__(self, config):
+        super().__init__(config)
+        hidden_size = config.d_model
+        self.is_export = config.is_export
+        self.counting_context_weight = nn.Sequential(
+            nn.Linear(config.vocab_size, hidden_size),
+            nn.ReLU(),
+            nn.Linear(hidden_size, hidden_size),
+            nn.ReLU(),
+            nn.Linear(hidden_size, config.d_model),
+        )
+
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        count_pred=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        head_mask=None,
+        cross_attn_head_mask=None,
+        past_key_values=None,
+        inputs_embeds=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+    ):
+        self.is_export = False if self.training else True
+        output_attentions = (
+            output_attentions
+            if output_attentions is not None
+            else self.config.output_attentions
+        )
+        output_hidden_states = (
+            output_hidden_states
+            if output_hidden_states is not None
+            else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = (
+            return_dict if return_dict is not None else self.config.use_return_dict
+        )
+
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError(
+                "You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time"
+            )
+        elif input_ids is not None:
+            input = input_ids
+            input_shape = input.shape
+            input_ids = input_ids.reshape([-1, input_shape[-1]])
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.shape[:-1]
+            input = inputs_embeds[:, :, -1]
+        else:
+            raise ValueError(
+                "You have to specify either decoder_input_ids or decoder_inputs_embeds"
+            )
+
+        past_key_values_length = (
+            past_key_values[0][0].shape[2] if past_key_values is not None else 0
+        )
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids) * self.embed_scale
+
+        if self._use_flash_attention_2:
+            attention_mask = (
+                attention_mask
+                if (attention_mask is not None and 0 in attention_mask)
+                else None
+            )
+        else:
+            if self.is_export:
+                attention_mask = _prepare_4d_causal_attention_mask_export(
+                    attention_mask,
+                    input_shape,
+                    inputs_embeds,
+                    past_key_values_length,
+                    is_export=self.is_export,
+                ).cast(paddle.float32)
+            else:
+                attention_mask = _prepare_4d_causal_attention_mask(
+                    attention_mask,
+                    input_shape,
+                    inputs_embeds,
+                    past_key_values_length,
+                    is_export=self.is_export,
+                )
+
+        if encoder_hidden_states is not None and encoder_attention_mask is not None:
+            if self._use_flash_attention_2:
+                encoder_attention_mask = (
+                    encoder_attention_mask if 0 in encoder_attention_mask else None
+                )
+            else:
+                encoder_attention_mask = _prepare_4d_attention_mask(
+                    encoder_attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]
+                )
+
+        # embed positions
+        positions = self.embed_positions(input, past_key_values_length)
+
+        hidden_states = inputs_embeds + positions
+
+        # TODO: add counting context weight to hidden_states
+        if count_pred is not None:
+            count_context_weight = self.counting_context_weight(count_pred)
+            hidden_states = hidden_states + 0.5 * count_context_weight.unsqueeze(1)
+
+        hidden_states = self.layernorm_embedding(hidden_states)
+        hidden_states = nn.functional.dropout(
+            hidden_states, p=self.dropout, training=self.training
+        )
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                print(
+                    "`use_cache=True` is incompatible with gradient checkpointing`. Setting `use_cache=False`..."
+                )
+                use_cache = False
+
+        # decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        all_cross_attentions = (
+            () if (output_attentions and encoder_hidden_states is not None) else None
+        )
+        next_decoder_cache = () if use_cache else None
+
+        # check if head_mask/cross_attn_head_mask has a correct number of layers specified if desired
+        for attn_mask, mask_name in zip(
+            [head_mask, cross_attn_head_mask], ["head_mask", "cross_attn_head_mask"]
+        ):
+            if attn_mask is not None:
+                if attn_mask.size()[0] != len(self.layers):
+                    raise ValueError(
+                        f"The `{mask_name}` should be specified for {len(self.layers)} layers, but it is for"
+                        f" {attn_mask.size()[0]}."
+                    )
+
+        for idx, decoder_layer in enumerate(self.layers):
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+            if self.training:
+                dropout_probability = paddle.rand([])
+                if dropout_probability < self.layerdrop:
+                    continue
+
+            past_key_value = (
+                past_key_values[idx] if past_key_values is not None else None
+            )
+
+            if self.gradient_checkpointing and self.training:
+                layer_outputs = self._gradient_checkpointing_func(
+                    decoder_layer.__call__,
+                    hidden_states,
+                    attention_mask,
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                    head_mask[idx] if head_mask is not None else None,
+                    (
+                        cross_attn_head_mask[idx]
+                        if cross_attn_head_mask is not None
+                        else None
+                    ),
+                    None,
+                    output_attentions,
+                    use_cache,
+                )
+            else:
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    attention_mask=attention_mask,
+                    encoder_hidden_states=encoder_hidden_states,
+                    encoder_attention_mask=encoder_attention_mask,
+                    layer_head_mask=(head_mask[idx] if head_mask is not None else None),
+                    cross_attn_layer_head_mask=(
+                        cross_attn_head_mask[idx]
+                        if cross_attn_head_mask is not None
+                        else None
+                    ),
+                    past_key_value=past_key_value,
+                    output_attentions=output_attentions,
+                    use_cache=use_cache,
+                )
+            hidden_states = layer_outputs[0]
+            if self.is_export:
+                next_decoder_cache += (layer_outputs[3 if output_attentions else 1],)
+            else:
+                if use_cache:
+                    next_decoder_cache += (
+                        layer_outputs[3 if output_attentions else 1],
+                    )
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+                if encoder_hidden_states is not None:
+                    all_cross_attentions += (layer_outputs[2],)
+
+        hidden_states = self.layer_norm(hidden_states)
+
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+        if self.is_export:
+            next_cache = next_decoder_cache
+        else:
+            next_cache = next_decoder_cache if use_cache else None
+        if not self.is_export:
+            if not return_dict:
+                return tuple(
+                    v
+                    for v in [
+                        hidden_states,
+                        next_cache,
+                        all_hidden_states,
+                        all_self_attns,
+                        all_cross_attentions,
+                    ]
+                    if v is not None
+                )
+        return BaseModelOutputWithPastAndCrossAttentions(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+            cross_attentions=all_cross_attentions,
+        )
+
+
+class SelfAttentionBlock(nn.Layer):
+    """
+    A self-attention block that implements multi-head self-attention
+    followed by a feed-forward network, typically used in transformer architectures.
+
+    Args:
+        embed_size (int): The size of the embedding vector.
+        num_heads (int): The number of attention heads.
+        is_export (bool): Flag indicating whether to configure the layer for export.
+    """
+
+    def __init__(self, embed_size, num_heads, is_export):
+        super(SelfAttentionBlock, self).__init__()
+        self.self_attention = MyMultiheadAttention(
+            embed_dim=embed_size, num_heads=num_heads, is_export=is_export
+        )
+        self.norm = nn.LayerNorm(embed_size)
+
+    def forward(self, x):
+        attn_output, _ = self.self_attention(x, x, x)
+        x = self.norm(attn_output + x)
+        return x
+
+
+class SeqCountingDecoder(nn.Layer):
+    """
+    A custom sequence counting decoder that incorporates multi-head attention layers
+    and feed-forward networks to process sequences, potentially for latex code counting .
+
+    Args:
+        in_features (int): The number of input features.
+        out_features (int): The number of output features.
+        num_heads (int): The number of attention heads. Defaults to 8.
+        num_layers (int): The number of attention layers. Defaults to 4.
+        is_export (bool): Flag indicating whether to configure the layer for export.
+    """
+
+    def __init__(
+        self, in_features, out_features, num_heads=8, num_layers=4, is_export=False
+    ):
+        super(SeqCountingDecoder, self).__init__()
+
+        self.attention_blocks = nn.LayerList(
+            [
+                SelfAttentionBlock(
+                    embed_size=in_features, num_heads=num_heads, is_export=is_export
+                )
+                for i in range(num_layers)
+            ]
+        )
+        self.fc1 = nn.Linear(in_features, in_features // 2)
+        self.relu = nn.ReLU()
+        self.global_avg_pool = nn.AdaptiveAvgPool1D(1)
+        self.fc2 = nn.Linear(in_features // 2, out_features)
+
+    def forward(self, x):
+        for block in self.attention_blocks:
+            x = block(x)
+        x = self.fc1(x)
+        x = self.relu(x)
+        x = x.transpose([0, 2, 1])
+        x = self.global_avg_pool(x)
+        x = x.squeeze(-1)
+        x = self.fc2(x)
+        return x
+
+
+class CustomMBartForCausalLM(MBartForCausalLM):
+    """
+    Custom MBart model for causal language modeling with a custom decoder.
+
+    This class extends the MBartForCausalLM by replacing its decoder with a
+    custom decoder, allowing for additional flexibility and features in the
+    decoding process.
+
+    Args:
+        config: The configuration object containing model parameters.
+        length_aware (bool): A flag to enable or configure length-aware mechanisms.
+    """
+
+    def __init__(self, config, length_aware=True):
+        super().__init__(config)
+        self.model.decoder = CustomMBartDecoder(config)
+        self.counting_decoder = SeqCountingDecoder(
+            config.d_model, config.vocab_size, is_export=config.is_export
+        )
+        self.length_aware = length_aware
+
+    def forward(
+        self,
+        input_ids=None,
+        attention_mask=None,
+        encoder_hidden_states=None,
+        encoder_attention_mask=None,
+        head_mask=None,
+        cross_attn_head_mask=None,
+        past_key_values=None,
+        inputs_embeds=None,
+        labels=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+        count_gt=None,
+    ):
+        output_attentions = (
+            output_attentions
+            if output_attentions is not None
+            else self.config.output_attentions
+        )
+        output_hidden_states = (
+            output_hidden_states
+            if output_hidden_states is not None
+            else self.config.output_hidden_states
+        )
+        return_dict = (
+            return_dict if return_dict is not None else self.config.use_return_dict
+        )
+
+        if self.length_aware:
+            count_pred = self.counting_decoder(encoder_hidden_states)
+        else:
+            count_pred = None
+
+        outputs = self.model.decoder(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            count_pred=count_pred,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+            head_mask=head_mask,
+            cross_attn_head_mask=cross_attn_head_mask,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        logits = self.lm_head(outputs[0])
+
+        return CausalLMOutputWithCrossAttentionsAndCounting(
+            logits=logits,
+            counting=count_pred,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+            cross_attentions=outputs.cross_attentions,
+        )
+
+
+class UniMERNetHead(nn.Layer):
+    """Implementation of UniMERNetHead decoder.
+
+    Args:
+         max_new_tokens (int): Maximum number of new tokens to generate.
+         decoder_start_token_id (int): ID of the token that starts the decoding.
+         temperature (float): Sampling temperature for generation.
+         do_sample (bool): Whether to use sampling; if False, uses greedy decoding.
+         top_p (float): Top-p (nucleus) sampling parameter.
+         in_channels (int): Number of input channels/features.
+         encoder_hidden_size (int): Hidden size of the encoder.
+         decoder_hidden_size (int): Hidden size of the decoder.
+         decoder_ffn_dim (int): Dimension of the decoder's feed-forward network.
+         decoder_layers (int): Number of layers in the decoder.
+         is_export (bool): Flag indicating if the model is being prepared for export.
+         length_aware (bool): Flag to enable length-aware mechanisms.
+    """
+
+    def __init__(
+        self,
+        max_new_tokens=1536,
+        decoder_start_token_id=0,
+        temperature=0.2,
+        do_sample=False,
+        top_p=0.95,
+        in_channels=1024,
+        encoder_hidden_size=1024,
+        decoder_hidden_size=1024,
+        decoder_ffn_dim=4096,
+        decoder_layers=8,
+        is_export=False,
+        length_aware=True,
+    ):
+        super().__init__()
+        mbart_config_dict = {
+            "activation_dropout": 0.0,
+            "activation_function": "gelu",
+            "add_cross_attention": True,
+            "add_final_layer_norm": True,
+            "attention_dropout": 0.0,
+            "bos_token_id": 0,
+            "classifier_dropout": 0.0,
+            "d_model": decoder_hidden_size,
+            "decoder_attention_heads": 16,
+            "decoder_ffn_dim": decoder_ffn_dim,
+            "decoder_layerdrop": 0.0,
+            "decoder_layers": decoder_layers,
+            "dropout": 0.1,
+            "encoder_attention_heads": 16,
+            "encoder_ffn_dim": 4096,
+            "encoder_layerdrop": 0.0,
+            "encoder_layers": 12,
+            "eos_token_id": 2,
+            "forced_eos_token_id": 2,
+            "init_std": 0.02,
+            "is_decoder": True,
+            "is_encoder_decoder": False,
+            "output_hidden_states": False,
+            "max_position_embeddings": max_new_tokens,
+            "model_type": "mbart",
+            "num_hidden_layers": 12,
+            "pad_token_id": 1,
+            "scale_embedding": True,
+            "tie_word_embeddings": False,
+            "transformers_version": "4.40.0",
+            "use_cache": True,
+            "use_return_dict": True,
+            "vocab_size": 50000,
+            "_attn_implementation": "eager",
+            "hidden_size": decoder_hidden_size,
+            "is_export": is_export,
+        }
+
+        self.max_new_tokens = max_new_tokens
+        self.decoder_start_token_id = decoder_start_token_id
+        self.temperature = temperature
+        self.do_sample = do_sample
+        self.top_p = top_p
+        self.max_seq_len = max_new_tokens
+        self.config_decoder = MBartConfig(**mbart_config_dict)
+        self.encoder_hidden_size = encoder_hidden_size
+        self.is_export = self.config_decoder.is_export
+        self.decoder = CustomMBartForCausalLM(
+            self.config_decoder, length_aware=length_aware
+        )
+        if self.config_decoder.hidden_size != self.encoder_hidden_size:
+            self.enc_to_dec_proj = nn.Linear(
+                self.encoder_hidden_size, self.config_decoder.hidden_size
+            )
+        generation_config = {
+            "max_length": 1537,
+            "forced_eos_token_id": 2,
+        }
+        self.eos_token_id = generation_config["forced_eos_token_id"]
+        self.pad_token_id = self.config_decoder.pad_token_id
+        self.logits_processor = LogitsProcessorList()
+        self.logits_processor.append(
+            ForcedEOSTokenLogitsProcessor(
+                generation_config["max_length"],
+                generation_config["forced_eos_token_id"],
+            )
+        )
+
+    def _get_decoder_start_token_id(
+        self, decoder_start_token_id=None, bos_token_id=None
+    ) -> int:
+        decoder_start_token_id = (
+            decoder_start_token_id
+            if decoder_start_token_id is not None
+            else self.generation_config.decoder_start_token_id
+        )
+        bos_token_id = (
+            bos_token_id
+            if bos_token_id is not None
+            else self.generation_config.bos_token_id
+        )
+        if decoder_start_token_id is not None:
+            return decoder_start_token_id
+        elif bos_token_id is not None:
+            return bos_token_id
+        raise ValueError(
+            "`decoder_start_token_id` or `bos_token_id` has to be defined for encoder-decoder generation."
+        )
+
+    def _prepare_decoder_input_ids_for_generation(
+        self,
+        batch_size,
+        model_kwargs,
+        decoder_start_token_id=None,
+        bos_token_id=None,
+    ):
+        if model_kwargs is not None and "decoder_input_ids" in model_kwargs:
+            decoder_input_ids = model_kwargs.pop("decoder_input_ids")
+        elif "input_ids" in model_kwargs:
+            decoder_input_ids = model_kwargs.pop("input_ids")
+        else:
+            decoder_input_ids = None
+
+        decoder_start_token_id = self._get_decoder_start_token_id(
+            decoder_start_token_id, bos_token_id
+        )
+
+        if isinstance(decoder_start_token_id, list):
+            if len(decoder_start_token_id) != batch_size:
+                raise ValueError(
+                    f"`decoder_start_token_id` expected to have length {batch_size} but got {len(decoder_start_token_id)}"
+                )
+            decoder_input_ids_start = paddle.to_tensor(
+                decoder_start_token_id,
+                dtype=paddle.int64,
+            )
+            decoder_input_ids_start = decoder_input_ids_start.view(-1, 1)
+        else:
+            decoder_input_ids_start = (
+                paddle.ones(
+                    (batch_size, 1),
+                    dtype=paddle.int64,
+                )
+                * decoder_start_token_id
+            )
+
+        if decoder_input_ids is None:
+            decoder_input_ids = decoder_input_ids_start
+        elif (
+            self.config.model_type == "vision-encoder-decoder"
+            and "donut" in self.name_or_path.lower()
+        ):
+            pass
+        elif self.config.model_type in ["whisper"]:
+            pass
+        elif (
+            isinstance(decoder_start_token_id, int)
+            and (decoder_input_ids[:, 0] != decoder_start_token_id).all().item()
+        ) or (
+            isinstance(decoder_start_token_id, paddle.Tensor)
+            and (decoder_input_ids[:, 0] != decoder_start_token_id[:, 0]).all().item()
+        ):
+            decoder_input_ids = paddle.concat(
+                [decoder_input_ids_start, decoder_input_ids], axis=-1
+            )
+            if "decoder_attention_mask" in model_kwargs:
+                decoder_attention_mask = model_kwargs["decoder_attention_mask"]
+                decoder_attention_mask = paddle.cat(
+                    (
+                        paddle.ones_like(decoder_attention_mask)[:, :1],
+                        decoder_attention_mask,
+                    ),
+                    dim=-1,
+                )
+                model_kwargs["decoder_attention_mask"] = decoder_attention_mask
+
+        return decoder_input_ids, model_kwargs
+
+    def prepare_inputs_for_generation_mbart(
+        self,
+        input_ids,
+        past_key_values=None,
+        attention_mask=None,
+        use_cache=None,
+        **kwargs,
+    ):
+
+        if attention_mask is None:
+            attention_mask = paddle.ones(input_ids.shape)
+
+        if past_key_values:
+            past_length = past_key_values[0][0].shape[2]
+
+            if input_ids.shape[1] > past_length:
+                remove_prefix_length = past_length
+            else:
+                remove_prefix_length = input_ids.shape[1] - 1
+
+            input_ids = input_ids[:, remove_prefix_length:]
+        return {
+            "input_ids": input_ids,
+            "attention_mask": attention_mask,
+            "past_key_values": past_key_values,
+            "use_cache": use_cache,
+        }
+
+    def prepare_inputs_for_generation(
+        self,
+        input_ids,
+        past_key_values=None,
+        attention_mask=None,
+        use_cache=None,
+        encoder_outputs=None,
+        **kwargs,
+    ):
+        decoder_inputs = self.prepare_inputs_for_generation_mbart(
+            input_ids, past_key_values=past_key_values
+        )
+        decoder_attention_mask = (
+            decoder_inputs["attention_mask"]
+            if "attention_mask" in decoder_inputs
+            else None
+        )
+        input_dict = {
+            "attention_mask": attention_mask,
+            "decoder_attention_mask": decoder_attention_mask,
+            "decoder_input_ids": decoder_inputs["input_ids"],
+            "encoder_outputs": encoder_outputs,
+            "past_key_values": decoder_inputs["past_key_values"],
+            "use_cache": use_cache,
+        }
+        return input_dict
+
+    def prepare_inputs_for_generation_export(
+        self,
+        past_key_values=None,
+        attention_mask=None,
+        use_cache=None,
+        encoder_outputs=None,
+        **kwargs,
+    ):
+
+        input_dict = {
+            "decoder_attention_mask": None,
+            "use_cache": use_cache,
+        }
+        return input_dict
+
+    def _extract_past_from_model_output(
+        self, outputs: ModelOutput, standardize_cache_format: bool = False
+    ):
+        past_key_values = None
+        if "past_key_values" in outputs:
+            past_key_values = outputs.past_key_values
+        elif "mems" in outputs:
+            past_key_values = outputs.mems
+        elif "past_buckets_states" in outputs:
+            past_key_values = outputs.past_buckets_states
+
+        return past_key_values
+
+    def _update_model_kwargs_for_generation(
+        self,
+        outputs: ModelOutput,
+        model_kwargs: Dict[str, Any],
+        is_encoder_decoder: bool = False,
+        standardize_cache_format: bool = False,
+    ) -> Dict[str, Any]:
+        model_kwargs["past_key_values"] = self._extract_past_from_model_output(
+            outputs, standardize_cache_format=standardize_cache_format
+        )
+        if getattr(outputs, "state", None) is not None:
+            model_kwargs["state"] = outputs.state
+
+        if "token_type_ids" in model_kwargs:
+            token_type_ids = model_kwargs["token_type_ids"]
+            model_kwargs["token_type_ids"] = paddle.concat(
+                [token_type_ids, token_type_ids[:, -1].unsqueeze(-1)], axis=-1
+            )
+
+        if not is_encoder_decoder:
+            if "attention_mask" in model_kwargs:
+                attention_mask = model_kwargs["attention_mask"]
+                model_kwargs["attention_mask"] = paddle.concat(
+                    [
+                        attention_mask,
+                        attention_mask.new_ones((attention_mask.shape[0], 1)),
+                    ],
+                    axis=-1,
+                )
+        else:
+            if "decoder_attention_mask" in model_kwargs:
+                decoder_attention_mask = model_kwargs["decoder_attention_mask"]
+                model_kwargs["decoder_attention_mask"] = paddle.concat(
+                    [
+                        decoder_attention_mask,
+                        decoder_attention_mask.new_ones(
+                            (decoder_attention_mask.shape[0], 1)
+                        ),
+                    ],
+                    axis=-1,
+                )
+
+        if (
+            "cache_position" in model_kwargs
+            and model_kwargs["cache_position"] is not None
+        ):
+            model_kwargs["cache_position"] = model_kwargs["cache_position"][-1:] + 1
+
+        return model_kwargs
+
+    def stopping_criteria(self, input_ids):
+        if self.is_export:
+            return input_ids[:, -1] == paddle.to_tensor([self.eos_token_id])
+        is_done = paddle.isin(input_ids[:, -1], paddle.to_tensor([self.eos_token_id]))
+        return is_done
+
+    def generate_single_iter(
+        self,
+        decoder_input_ids=None,
+        decoder_attention_mask=None,
+        encoder_outputs=None,
+        past_key_values=None,
+        decoder_inputs_embeds=None,
+        labels=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+        **kwargs,
+    ):
+        encoder_hidden_states = encoder_outputs[0]
+        if self.config_decoder.hidden_size != self.encoder_hidden_size:
+            encoder_hidden_states = self.enc_to_dec_proj(encoder_hidden_states)
+        kwargs_decoder = {}
+
+        decoder_outputs = self.decoder(
+            input_ids=decoder_input_ids,
+            attention_mask=decoder_attention_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=None,
+            inputs_embeds=None,
+            output_attentions=False,
+            output_hidden_states=output_hidden_states,
+            use_cache=use_cache,
+            past_key_values=past_key_values,
+            return_dict=return_dict,
+            **kwargs_decoder,
+        )
+
+        return Seq2SeqLMOutput(
+            loss=None,
+            logits=decoder_outputs.logits,
+            past_key_values=decoder_outputs.past_key_values,
+            decoder_hidden_states=decoder_outputs.hidden_states,
+            decoder_attentions=decoder_outputs.attentions,
+            cross_attentions=decoder_outputs.cross_attentions,
+            encoder_last_hidden_state=encoder_outputs.last_hidden_state,
+            encoder_hidden_states=encoder_outputs.hidden_states,
+            encoder_attentions=encoder_outputs.attentions,
+        )
+
+    @paddle.no_grad()
+    def generate(
+        self,
+        model_kwargs,
+    ):
+        """
+        Generate sequences using the UniMERNetHead for inference tasks.
+
+        Args:
+            model_kwargs (dict): A dictionary of model configurations and inputs, which typically include:
+                - encoder_outputs: Outputs from the encoder.
+                - use_cache: Boolean flag to indicate if caching should be used.
+                - output_attentions: Boolean flag for outputting attention scores.
+                - output_hidden_states: Boolean flag for outputting hidden states.
+
+        Returns:
+            A tensor containing the generated sequences.
+        """
+        batch_size = model_kwargs["encoder_outputs"]["last_hidden_state"].shape[0]
+        generation_config = {
+            "decoder_start_token_id": 0,
+            "bos_token_id": 0,
+        }
+        input_ids, model_kwargs = self._prepare_decoder_input_ids_for_generation(
+            batch_size=batch_size,
+            model_kwargs=model_kwargs,
+            decoder_start_token_id=generation_config["decoder_start_token_id"],
+            bos_token_id=generation_config["bos_token_id"],
+        )
+        model_kwargs["key use_cache"] = True
+        batch_size, cur_len = input_ids.shape
+
+        if "inputs_embeds" in model_kwargs:
+            cur_len = model_kwargs["inputs_embeds"].shape[1]
+        model_kwargs["cache_position"] = paddle.arange(cur_len)
+        pad_token_id = self.pad_token_id
+        eos_token_id = [self.eos_token_id]
+        eos_token = self.eos_token_id
+        unfinished_sequences = paddle.ones(batch_size, dtype=paddle.int64)
+        for idx in range(self.max_seq_len):
+            model_inputs = self.prepare_inputs_for_generation(input_ids, **model_kwargs)
+            outputs = self.generate_single_iter(
+                **model_inputs,
+                return_dict=True,
+                output_attentions=False,
+                output_hidden_states=False,
+            )
+            next_token_logits = outputs.logits[:, -1, :]
+
+            next_tokens_scores = self.logits_processor(input_ids, next_token_logits)
+            next_tokens = paddle.argmax(next_tokens_scores, axis=-1)
+            if eos_token_id is not None:
+                if pad_token_id is None:
+                    raise ValueError(
+                        "If `eos_token_id` is defined, make sure that `pad_token_id` is defined."
+                    )
+                next_tokens = next_tokens * unfinished_sequences + pad_token_id * (
+                    1 - unfinished_sequences
+                )
+            input_ids = paddle.concat([input_ids, next_tokens[:, None]], axis=-1)
+            model_kwargs = self._update_model_kwargs_for_generation(
+                outputs,
+                model_kwargs,
+                is_encoder_decoder=self.config_decoder.is_encoder_decoder,
+            )
+            unfinished_sequences = unfinished_sequences & ~self.stopping_criteria(
+                input_ids
+            ).cast(paddle.int64)
+
+            if (
+                eos_token is not None
+                and (
+                    paddle.cumsum((input_ids == eos_token).cast(paddle.int64), 1)[:, -1]
+                    >= 1
+                ).all()
+            ):
+                break
+
+        return input_ids
+
+    @paddle.no_grad()
+    def generate_export(
+        self,
+        encoder_outputs,
+        model_kwargs,
+    ):
+        batch_size = encoder_outputs["last_hidden_state"].shape[0]
+        generation_config = {
+            "decoder_start_token_id": 0,
+            "bos_token_id": 0,
+        }
+        input_ids, model_kwargs = self._prepare_decoder_input_ids_for_generation(
+            batch_size=batch_size,
+            model_kwargs=model_kwargs,
+            decoder_start_token_id=generation_config["decoder_start_token_id"],
+            bos_token_id=generation_config["bos_token_id"],
+        )
+        input_ids = input_ids.reshape([-1, 1])
+        decoder_input_ids = input_ids
+        model_kwargs["key use_cache"] = True
+        batch_size, cur_len = input_ids.shape
+
+        if "inputs_embeds" in model_kwargs:
+            cur_len = model_kwargs["inputs_embeds"].shape[1]
+        cache_position = paddle.arange(cur_len)
+        pad_token_id = self.pad_token_id
+        eos_token_id = [self.eos_token_id]
+        eos_token = self.eos_token_id
+        unfinished_sequences = paddle.ones([batch_size], dtype=paddle.int64)
+        i_idx = paddle.full([], 0)
+        past_key_values = []
+        for i in range(8):
+            init_arr = paddle.zeros([batch_size, 16, 0, 64])
+            paddle.jit.api.set_dynamic_shape(init_arr, [-1, -1, -1, -1])
+            cache = (init_arr, init_arr, init_arr, init_arr)
+            past_key_values.append(cache)
+        idx = 0
+        while i_idx < paddle.to_tensor(self.max_seq_len):
+
+            model_inputs = self.prepare_inputs_for_generation_export(
+                past_key_values=past_key_values, **model_kwargs
+            )
+            decoder_attention_mask = model_inputs["decoder_attention_mask"]
+            decoder_attention_mask = paddle.ones(input_ids.shape)
+            paddle.jit.api.set_dynamic_shape(decoder_input_ids, [-1, -1])
+            paddle.jit.api.set_dynamic_shape(decoder_attention_mask, [-1, -1])
+
+            outputs = self.generate_single_iter(
+                decoder_input_ids=decoder_input_ids,
+                decoder_attention_mask=decoder_attention_mask,
+                encoder_outputs=encoder_outputs,
+                past_key_values=past_key_values,
+                return_dict=True,
+                output_attentions=False,
+                output_hidden_states=False,
+            )
+
+            next_token_logits = outputs.logits[:, -1, :]
+
+            next_tokens_scores = self.logits_processor(input_ids, next_token_logits)
+            next_tokens = paddle.argmax(next_tokens_scores, axis=-1)
+            if eos_token_id is not None:
+                next_tokens = next_tokens * unfinished_sequences + pad_token_id * (
+                    1 - unfinished_sequences
+                )
+            input_ids = paddle.concat([input_ids, next_tokens.unsqueeze(1)], axis=-1)
+            past_length = past_key_values[0][0].shape[2]
+            decoder_input_ids = next_tokens.unsqueeze(1)
+            past_key_values = outputs.past_key_values
+            cache_position = cache_position[-1:] + 1
+            unfinished_sequences = unfinished_sequences & ~self.stopping_criteria(
+                input_ids
+            ).cast(paddle.int64)
+            if (
+                eos_token is not None
+                and (
+                    paddle.cumsum((input_ids == eos_token).cast(paddle.int64), 1)[:, -1]
+                    >= 1
+                ).all()
+            ):
+                break
+
+            i_idx += 1
+        return input_ids
+
+    def forwad_train(
+        self,
+        encoder_outputs,
+        decoder_input_ids,
+        decoder_attention_mask,
+        past_key_values=None,
+        decoder_inputs_embeds=None,
+        labels=None,
+        use_cache=None,
+        output_attentions=None,
+        output_hidden_states=None,
+        return_dict=None,
+        **kwargs,
+    ):
+        """
+        Training for the UniMERNetHead.
+
+        Args:
+            encoder_outputs: Outputs from the encoder, used as input to the decoder.
+            decoder_input_ids: Input IDs for the decoder.
+            decoder_attention_mask: Attention mask for the decoder inputs.
+            past_key_values: Cached key/values for faster decoding.
+            decoder_inputs_embeds: Optional embeddings for the decoder inputs.
+            labels: Target labels for calculating loss.
+            use_cache: Whether to use cache during decoding.
+            output_attentions: Whether to return attention scores.
+            output_hidden_states: Whether to return hidden states.
+            return_dict: Whether to return a dictionary of outputs.
+            **kwargs: Additional keyword arguments.
+
+        Returns:
+            logits: The raw, unnormalized predictions from the model.
+            count_pred: Optional prediction related to sequence length or other counts.
+            masked_labels: The labels used during training, possibly masked.
+        """
+        labels = decoder_input_ids * 1
+        labels = labels.masked_fill_(labels == self.pad_token_id, -100)
+        input_decoder_input_ids = decoder_input_ids[:, :-1]
+        input_decoder_attention_mask = decoder_attention_mask[:, :-1]
+        encoder_hidden_states = encoder_outputs[0]
+        if self.config_decoder.hidden_size != self.encoder_hidden_size:
+            encoder_hidden_states = self.enc_to_dec_proj(encoder_hidden_states)
+        kwargs_decoder = {}
+        decoder_outputs = self.decoder(
+            input_ids=input_decoder_input_ids,
+            attention_mask=input_decoder_attention_mask,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=None,
+            inputs_embeds=None,
+            output_attentions=False,
+            output_hidden_states=output_hidden_states,
+            use_cache=use_cache,
+            past_key_values=past_key_values,
+            return_dict=return_dict,
+            **kwargs_decoder,
+        )
+
+        logits = decoder_outputs.logits
+        count_pred = decoder_outputs.counting
+        return logits, count_pred, labels
+
+    def forward(self, inputs, targets=None):
+        """
+        Forward pass for the UniMERNetHead, handling both training and inference.
+
+        Args:
+            inputs: The input data, which can vary based on training or inference.
+            targets: The target labels, used only during training.
+
+        Returns:
+            During inference: Returns predicted latex code.
+            During training: Returns logits, predicted counts, and masked labels.
+        """
+        self.is_export = False if self.training else True
+        if not self.training:
+            encoder_outputs = inputs
+            if self.is_export:
+                model_kwargs = {
+                    "output_attentions": False,
+                    "output_hidden_states": False,
+                    "use_cache": True,
+                }
+                word_pred = self.generate_export(encoder_outputs, model_kwargs)
+            else:
+                model_kwargs = {
+                    "output_attentions": False,
+                    "output_hidden_states": False,
+                    "use_cache": True,
+                    "encoder_outputs": encoder_outputs,
+                }
+                word_pred = self.generate(model_kwargs)
+
+            return word_pred
+
+        encoder_outputs, tgt_seq, mask = inputs
+        logits, count_pred, masked_labels = self.forwad_train(
+            encoder_outputs, tgt_seq, mask
+        )
+        return logits, count_pred, masked_labels
diff --git a/ppocr/modeling/heads/rec_visionlan_head.py b/ppocr/modeling/heads/rec_visionlan_head.py
new file mode 100644
index 0000000..bb34dc3
--- /dev/null
+++ b/ppocr/modeling/heads/rec_visionlan_head.py
@@ -0,0 +1,474 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/wangyuxin87/VisionLAN
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import ParamAttr
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle.nn.initializer import Normal, XavierNormal
+import numpy as np
+
+
+class PositionalEncoding(nn.Layer):
+    def __init__(self, d_hid, n_position=200):
+        super(PositionalEncoding, self).__init__()
+        self.register_buffer(
+            "pos_table", self._get_sinusoid_encoding_table(n_position, d_hid)
+        )
+
+    def _get_sinusoid_encoding_table(self, n_position, d_hid):
+        """Sinusoid position encoding table"""
+
+        def get_position_angle_vec(position):
+            return [
+                position / np.power(10000, 2 * (hid_j // 2) / d_hid)
+                for hid_j in range(d_hid)
+            ]
+
+        sinusoid_table = np.array(
+            [get_position_angle_vec(pos_i) for pos_i in range(n_position)]
+        )
+        sinusoid_table[:, 0::2] = np.sin(sinusoid_table[:, 0::2])  # dim 2i
+        sinusoid_table[:, 1::2] = np.cos(sinusoid_table[:, 1::2])  # dim 2i+1
+        sinusoid_table = paddle.to_tensor(sinusoid_table, dtype="float32")
+        sinusoid_table = paddle.unsqueeze(sinusoid_table, axis=0)
+        return sinusoid_table
+
+    def forward(self, x):
+        return x + self.pos_table[:, : x.shape[1]].clone().detach()
+
+
+class ScaledDotProductAttention(nn.Layer):
+    "Scaled Dot-Product Attention"
+
+    def __init__(self, temperature, attn_dropout=0.1):
+        super(ScaledDotProductAttention, self).__init__()
+        self.temperature = temperature
+        self.dropout = nn.Dropout(attn_dropout)
+        self.softmax = nn.Softmax(axis=2)
+
+    def forward(self, q, k, v, mask=None):
+        k = paddle.transpose(k, perm=[0, 2, 1])
+        attn = paddle.bmm(q, k)
+        attn = attn / self.temperature
+        if mask is not None:
+            attn = attn.masked_fill(mask, -1e9)
+            if mask.dim() == 3:
+                mask = paddle.unsqueeze(mask, axis=1)
+            elif mask.dim() == 2:
+                mask = paddle.unsqueeze(mask, axis=1)
+                mask = paddle.unsqueeze(mask, axis=1)
+            repeat_times = [
+                attn.shape[1] // mask.shape[1],
+                attn.shape[2] // mask.shape[2],
+            ]
+            mask = paddle.tile(mask, [1, repeat_times[0], repeat_times[1], 1])
+            attn[mask == 0] = -1e9
+        attn = self.softmax(attn)
+        attn = self.dropout(attn)
+        output = paddle.bmm(attn, v)
+        return output
+
+
+class MultiHeadAttention(nn.Layer):
+    "Multi-Head Attention module"
+
+    def __init__(self, n_head, d_model, d_k, d_v, dropout=0.1):
+        super(MultiHeadAttention, self).__init__()
+        self.n_head = n_head
+        self.d_k = d_k
+        self.d_v = d_v
+        self.w_qs = nn.Linear(
+            d_model,
+            n_head * d_k,
+            weight_attr=ParamAttr(
+                initializer=Normal(mean=0, std=np.sqrt(2.0 / (d_model + d_k)))
+            ),
+        )
+        self.w_ks = nn.Linear(
+            d_model,
+            n_head * d_k,
+            weight_attr=ParamAttr(
+                initializer=Normal(mean=0, std=np.sqrt(2.0 / (d_model + d_k)))
+            ),
+        )
+        self.w_vs = nn.Linear(
+            d_model,
+            n_head * d_v,
+            weight_attr=ParamAttr(
+                initializer=Normal(mean=0, std=np.sqrt(2.0 / (d_model + d_v)))
+            ),
+        )
+
+        self.attention = ScaledDotProductAttention(temperature=np.power(d_k, 0.5))
+        self.layer_norm = nn.LayerNorm(d_model)
+        self.fc = nn.Linear(
+            n_head * d_v, d_model, weight_attr=ParamAttr(initializer=XavierNormal())
+        )
+        self.dropout = nn.Dropout(dropout)
+
+    def forward(self, q, k, v, mask=None):
+        d_k, d_v, n_head = self.d_k, self.d_v, self.n_head
+        sz_b, len_q, _ = q.shape
+        sz_b, len_k, _ = k.shape
+        sz_b, len_v, _ = v.shape
+        residual = q
+
+        q = self.w_qs(q)
+        q = paddle.reshape(q, shape=[-1, len_q, n_head, d_k])  # 4*21*512 ---- 4*21*8*64
+        k = self.w_ks(k)
+        k = paddle.reshape(k, shape=[-1, len_k, n_head, d_k])
+        v = self.w_vs(v)
+        v = paddle.reshape(v, shape=[-1, len_v, n_head, d_v])
+
+        q = paddle.transpose(q, perm=[2, 0, 1, 3])
+        q = paddle.reshape(q, shape=[-1, len_q, d_k])  # (n*b) x lq x dk
+        k = paddle.transpose(k, perm=[2, 0, 1, 3])
+        k = paddle.reshape(k, shape=[-1, len_k, d_k])  # (n*b) x lk x dk
+        v = paddle.transpose(v, perm=[2, 0, 1, 3])
+        v = paddle.reshape(v, shape=[-1, len_v, d_v])  # (n*b) x lv x dv
+
+        mask = (
+            paddle.tile(mask, [n_head, 1, 1]) if mask is not None else None
+        )  # (n*b) x .. x ..
+        output = self.attention(q, k, v, mask=mask)
+        output = paddle.reshape(output, shape=[n_head, -1, len_q, d_v])
+        output = paddle.transpose(output, perm=[1, 2, 0, 3])
+        output = paddle.reshape(
+            output, shape=[-1, len_q, n_head * d_v]
+        )  # b x lq x (n*dv)
+        output = self.dropout(self.fc(output))
+        output = self.layer_norm(output + residual)
+        return output
+
+
+class PositionwiseFeedForward(nn.Layer):
+    def __init__(self, d_in, d_hid, dropout=0.1):
+        super(PositionwiseFeedForward, self).__init__()
+        self.w_1 = nn.Conv1D(d_in, d_hid, 1)  # position-wise
+        self.w_2 = nn.Conv1D(d_hid, d_in, 1)  # position-wise
+        self.layer_norm = nn.LayerNorm(d_in)
+        self.dropout = nn.Dropout(dropout)
+
+    def forward(self, x):
+        residual = x
+        x = paddle.transpose(x, perm=[0, 2, 1])
+        x = self.w_2(F.relu(self.w_1(x)))
+        x = paddle.transpose(x, perm=[0, 2, 1])
+        x = self.dropout(x)
+        x = self.layer_norm(x + residual)
+        return x
+
+
+class EncoderLayer(nn.Layer):
+    """Compose with two layers"""
+
+    def __init__(self, d_model, d_inner, n_head, d_k, d_v, dropout=0.1):
+        super(EncoderLayer, self).__init__()
+        self.slf_attn = MultiHeadAttention(n_head, d_model, d_k, d_v, dropout=dropout)
+        self.pos_ffn = PositionwiseFeedForward(d_model, d_inner, dropout=dropout)
+
+    def forward(self, enc_input, slf_attn_mask=None):
+        enc_output = self.slf_attn(enc_input, enc_input, enc_input, mask=slf_attn_mask)
+        enc_output = self.pos_ffn(enc_output)
+        return enc_output
+
+
+class Transformer_Encoder(nn.Layer):
+    def __init__(
+        self,
+        n_layers=2,
+        n_head=8,
+        d_word_vec=512,
+        d_k=64,
+        d_v=64,
+        d_model=512,
+        d_inner=2048,
+        dropout=0.1,
+        n_position=256,
+    ):
+        super(Transformer_Encoder, self).__init__()
+        self.position_enc = PositionalEncoding(d_word_vec, n_position=n_position)
+        self.dropout = nn.Dropout(p=dropout)
+        self.layer_stack = nn.LayerList(
+            [
+                EncoderLayer(d_model, d_inner, n_head, d_k, d_v, dropout=dropout)
+                for _ in range(n_layers)
+            ]
+        )
+        self.layer_norm = nn.LayerNorm(d_model, epsilon=1e-6)
+
+    def forward(self, enc_output, src_mask, return_attns=False):
+        enc_output = self.dropout(self.position_enc(enc_output))  # position embedding
+        for enc_layer in self.layer_stack:
+            enc_output = enc_layer(enc_output, slf_attn_mask=src_mask)
+        enc_output = self.layer_norm(enc_output)
+        return enc_output
+
+
+class PP_layer(nn.Layer):
+    def __init__(self, n_dim=512, N_max_character=25, n_position=256):
+        super(PP_layer, self).__init__()
+        self.character_len = N_max_character
+        self.f0_embedding = nn.Embedding(N_max_character, n_dim)
+        self.w0 = nn.Linear(N_max_character, n_position)
+        self.wv = nn.Linear(n_dim, n_dim)
+        self.we = nn.Linear(n_dim, N_max_character)
+        self.active = nn.Tanh()
+        self.softmax = nn.Softmax(axis=2)
+
+    def forward(self, enc_output):
+        # enc_output: b,256,512
+        reading_order = paddle.arange(self.character_len, dtype="int64")
+        reading_order = reading_order.unsqueeze(0).expand(
+            [enc_output.shape[0], self.character_len]
+        )  # (S,) -> (B, S)
+        reading_order = self.f0_embedding(reading_order)  # b,25,512
+
+        # calculate attention
+        reading_order = paddle.transpose(reading_order, perm=[0, 2, 1])
+        t = self.w0(reading_order)  # b,512,256
+        t = self.active(
+            paddle.transpose(t, perm=[0, 2, 1]) + self.wv(enc_output)
+        )  # b,256,512
+        t = self.we(t)  # b,256,25
+        t = self.softmax(paddle.transpose(t, perm=[0, 2, 1]))  # b,25,256
+        g_output = paddle.bmm(t, enc_output)  # b,25,512
+        return g_output
+
+
+class Prediction(nn.Layer):
+    def __init__(self, n_dim=512, n_position=256, N_max_character=25, n_class=37):
+        super(Prediction, self).__init__()
+        self.pp = PP_layer(
+            n_dim=n_dim, N_max_character=N_max_character, n_position=n_position
+        )
+        self.pp_share = PP_layer(
+            n_dim=n_dim, N_max_character=N_max_character, n_position=n_position
+        )
+        self.w_vrm = nn.Linear(n_dim, n_class)  # output layer
+        self.w_share = nn.Linear(n_dim, n_class)  # output layer
+        self.nclass = n_class
+
+    def forward(self, cnn_feature, f_res, f_sub, train_mode=False, use_mlm=True):
+        if train_mode:
+            if not use_mlm:
+                g_output = self.pp(cnn_feature)  # b,25,512
+                g_output = self.w_vrm(g_output)
+                f_res = 0
+                f_sub = 0
+                return g_output, f_res, f_sub
+            g_output = self.pp(cnn_feature)  # b,25,512
+            f_res = self.pp_share(f_res)
+            f_sub = self.pp_share(f_sub)
+            g_output = self.w_vrm(g_output)
+            f_res = self.w_share(f_res)
+            f_sub = self.w_share(f_sub)
+            return g_output, f_res, f_sub
+        else:
+            g_output = self.pp(cnn_feature)  # b,25,512
+            g_output = self.w_vrm(g_output)
+            return g_output
+
+
+class MLM(nn.Layer):
+    "Architecture of MLM"
+
+    def __init__(self, n_dim=512, n_position=256, max_text_length=25):
+        super(MLM, self).__init__()
+        self.MLM_SequenceModeling_mask = Transformer_Encoder(
+            n_layers=2, n_position=n_position
+        )
+        self.MLM_SequenceModeling_WCL = Transformer_Encoder(
+            n_layers=1, n_position=n_position
+        )
+        self.pos_embedding = nn.Embedding(max_text_length, n_dim)
+        self.w0_linear = nn.Linear(1, n_position)
+        self.wv = nn.Linear(n_dim, n_dim)
+        self.active = nn.Tanh()
+        self.we = nn.Linear(n_dim, 1)
+        self.sigmoid = nn.Sigmoid()
+
+    def forward(self, x, label_pos):
+        # transformer unit for generating mask_c
+        feature_v_seq = self.MLM_SequenceModeling_mask(x, src_mask=None)
+        # position embedding layer
+        label_pos = paddle.to_tensor(label_pos, dtype="int64")
+        pos_emb = self.pos_embedding(label_pos)
+        pos_emb = self.w0_linear(paddle.unsqueeze(pos_emb, axis=2))
+        pos_emb = paddle.transpose(pos_emb, perm=[0, 2, 1])
+        # fusion position embedding with features V & generate mask_c
+        att_map_sub = self.active(pos_emb + self.wv(feature_v_seq))
+        att_map_sub = self.we(att_map_sub)  # b,256,1
+        att_map_sub = paddle.transpose(att_map_sub, perm=[0, 2, 1])
+        att_map_sub = self.sigmoid(att_map_sub)  # b,1,256
+        # WCL
+        ## generate inputs for WCL
+        att_map_sub = paddle.transpose(att_map_sub, perm=[0, 2, 1])
+        f_res = x * (1 - att_map_sub)  # second path with remaining string
+        f_sub = x * att_map_sub  # first path with occluded character
+        ## transformer units in WCL
+        f_res = self.MLM_SequenceModeling_WCL(f_res, src_mask=None)
+        f_sub = self.MLM_SequenceModeling_WCL(f_sub, src_mask=None)
+        return f_res, f_sub, att_map_sub
+
+
+def trans_1d_2d(x):
+    b, w_h, c = x.shape  # b, 256, 512
+    x = paddle.transpose(x, perm=[0, 2, 1])
+    x = paddle.reshape(x, [-1, c, 32, 8])
+    x = paddle.transpose(x, perm=[0, 1, 3, 2])  # [b, c, 8, 32]
+    return x
+
+
+class MLM_VRM(nn.Layer):
+    """
+    MLM+VRM, MLM is only used in training.
+    ratio controls the occluded number in a batch.
+    The pipeline of VisionLAN in testing is very concise with only a backbone + sequence modeling(transformer unit) + prediction layer(pp layer).
+    x: input image
+    label_pos: character index
+    training_step: LF or LA process
+    output
+    text_pre: prediction of VRM
+    test_rem: prediction of remaining string in MLM
+    text_mas: prediction of occluded character in MLM
+    mask_c_show: visualization of Mask_c
+    """
+
+    def __init__(
+        self, n_layers=3, n_position=256, n_dim=512, max_text_length=25, nclass=37
+    ):
+        super(MLM_VRM, self).__init__()
+        self.MLM = MLM(
+            n_dim=n_dim, n_position=n_position, max_text_length=max_text_length
+        )
+        self.SequenceModeling = Transformer_Encoder(
+            n_layers=n_layers, n_position=n_position
+        )
+        self.Prediction = Prediction(
+            n_dim=n_dim,
+            n_position=n_position,
+            N_max_character=max_text_length
+            + 1,  # N_max_character = 1 eos + 25 characters
+            n_class=nclass,
+        )
+        self.nclass = nclass
+        self.max_text_length = max_text_length
+
+    def forward(self, x, label_pos, training_step, train_mode=False):
+        b, c, h, w = x.shape
+        nT = self.max_text_length
+        x = paddle.transpose(x, perm=[0, 1, 3, 2])
+        x = paddle.reshape(x, [-1, c, h * w])
+        x = paddle.transpose(x, perm=[0, 2, 1])
+        if train_mode:
+            if training_step == "LF_1":
+                f_res = 0
+                f_sub = 0
+                x = self.SequenceModeling(x, src_mask=None)
+                text_pre, test_rem, text_mas = self.Prediction(
+                    x, f_res, f_sub, train_mode=True, use_mlm=False
+                )
+                return text_pre, text_pre, text_pre, text_pre
+            elif training_step == "LF_2":
+                # MLM
+                f_res, f_sub, mask_c = self.MLM(x, label_pos)
+                x = self.SequenceModeling(x, src_mask=None)
+                text_pre, test_rem, text_mas = self.Prediction(
+                    x, f_res, f_sub, train_mode=True
+                )
+                mask_c_show = trans_1d_2d(mask_c)
+                return text_pre, test_rem, text_mas, mask_c_show
+            elif training_step == "LA":
+                # MLM
+                f_res, f_sub, mask_c = self.MLM(x, label_pos)
+                ## use the mask_c (1 for occluded character and 0 for remaining characters) to occlude input
+                ## ratio controls the occluded number in a batch
+                character_mask = paddle.zeros_like(mask_c)
+
+                ratio = b // 2
+                if ratio >= 1:
+                    with paddle.no_grad():
+                        character_mask[0:ratio, :, :] = mask_c[0:ratio, :, :]
+                else:
+                    character_mask = mask_c
+                x = x * (1 - character_mask)
+                # VRM
+                ## transformer unit for VRM
+                x = self.SequenceModeling(x, src_mask=None)
+                ## prediction layer for MLM and VSR
+                text_pre, test_rem, text_mas = self.Prediction(
+                    x, f_res, f_sub, train_mode=True
+                )
+                mask_c_show = trans_1d_2d(mask_c)
+                return text_pre, test_rem, text_mas, mask_c_show
+            else:
+                raise NotImplementedError
+        else:  # VRM is only used in the testing stage
+            f_res = 0
+            f_sub = 0
+            contextual_feature = self.SequenceModeling(x, src_mask=None)
+            text_pre = self.Prediction(
+                contextual_feature, f_res, f_sub, train_mode=False, use_mlm=False
+            )
+            text_pre = paddle.transpose(text_pre, perm=[1, 0, 2])  # (26, b, 37))
+            return text_pre, x
+
+
+class VLHead(nn.Layer):
+    """
+    Architecture of VisionLAN
+    """
+
+    def __init__(
+        self,
+        in_channels,
+        out_channels=36,
+        n_layers=3,
+        n_position=256,
+        n_dim=512,
+        max_text_length=25,
+        training_step="LA",
+    ):
+        super(VLHead, self).__init__()
+        self.MLM_VRM = MLM_VRM(
+            n_layers=n_layers,
+            n_position=n_position,
+            n_dim=n_dim,
+            max_text_length=max_text_length,
+            nclass=out_channels + 1,
+        )
+        self.training_step = training_step
+
+    def forward(self, feat, targets=None):
+        if self.training:
+            label_pos = targets[-2]
+            text_pre, test_rem, text_mas, mask_map = self.MLM_VRM(
+                feat, label_pos, self.training_step, train_mode=True
+            )
+            return text_pre, test_rem, text_mas, mask_map
+        else:
+            text_pre, x = self.MLM_VRM(
+                feat, targets, self.training_step, train_mode=False
+            )
+            return text_pre, x
diff --git a/ppocr/modeling/heads/self_attention.py b/ppocr/modeling/heads/self_attention.py
new file mode 100644
index 0000000..c02b9cd
--- /dev/null
+++ b/ppocr/modeling/heads/self_attention.py
@@ -0,0 +1,460 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+
+import paddle
+from paddle import ParamAttr, nn
+from paddle import nn, ParamAttr
+from paddle.nn import functional as F
+import numpy as np
+
+gradient_clip = 10
+
+
+class WrapEncoderForFeature(nn.Layer):
+    def __init__(
+        self,
+        src_vocab_size,
+        max_length,
+        n_layer,
+        n_head,
+        d_key,
+        d_value,
+        d_model,
+        d_inner_hid,
+        prepostprocess_dropout,
+        attention_dropout,
+        relu_dropout,
+        preprocess_cmd,
+        postprocess_cmd,
+        weight_sharing,
+        bos_idx=0,
+    ):
+        super(WrapEncoderForFeature, self).__init__()
+
+        self.prepare_encoder = PrepareEncoder(
+            src_vocab_size,
+            d_model,
+            max_length,
+            prepostprocess_dropout,
+            bos_idx=bos_idx,
+            word_emb_param_name="src_word_emb_table",
+        )
+        self.encoder = Encoder(
+            n_layer,
+            n_head,
+            d_key,
+            d_value,
+            d_model,
+            d_inner_hid,
+            prepostprocess_dropout,
+            attention_dropout,
+            relu_dropout,
+            preprocess_cmd,
+            postprocess_cmd,
+        )
+
+    def forward(self, enc_inputs):
+        conv_features, src_pos, src_slf_attn_bias = enc_inputs
+        enc_input = self.prepare_encoder(conv_features, src_pos)
+        enc_output = self.encoder(enc_input, src_slf_attn_bias)
+        return enc_output
+
+
+class WrapEncoder(nn.Layer):
+    """
+    embedder + encoder
+    """
+
+    def __init__(
+        self,
+        src_vocab_size,
+        max_length,
+        n_layer,
+        n_head,
+        d_key,
+        d_value,
+        d_model,
+        d_inner_hid,
+        prepostprocess_dropout,
+        attention_dropout,
+        relu_dropout,
+        preprocess_cmd,
+        postprocess_cmd,
+        weight_sharing,
+        bos_idx=0,
+    ):
+        super(WrapEncoder, self).__init__()
+
+        self.prepare_decoder = PrepareDecoder(
+            src_vocab_size, d_model, max_length, prepostprocess_dropout, bos_idx=bos_idx
+        )
+        self.encoder = Encoder(
+            n_layer,
+            n_head,
+            d_key,
+            d_value,
+            d_model,
+            d_inner_hid,
+            prepostprocess_dropout,
+            attention_dropout,
+            relu_dropout,
+            preprocess_cmd,
+            postprocess_cmd,
+        )
+
+    def forward(self, enc_inputs):
+        src_word, src_pos, src_slf_attn_bias = enc_inputs
+        enc_input = self.prepare_decoder(src_word, src_pos)
+        enc_output = self.encoder(enc_input, src_slf_attn_bias)
+        return enc_output
+
+
+class Encoder(nn.Layer):
+    """
+    encoder
+    """
+
+    def __init__(
+        self,
+        n_layer,
+        n_head,
+        d_key,
+        d_value,
+        d_model,
+        d_inner_hid,
+        prepostprocess_dropout,
+        attention_dropout,
+        relu_dropout,
+        preprocess_cmd="n",
+        postprocess_cmd="da",
+    ):
+        super(Encoder, self).__init__()
+
+        self.encoder_layers = list()
+        for i in range(n_layer):
+            self.encoder_layers.append(
+                self.add_sublayer(
+                    "layer_%d" % i,
+                    EncoderLayer(
+                        n_head,
+                        d_key,
+                        d_value,
+                        d_model,
+                        d_inner_hid,
+                        prepostprocess_dropout,
+                        attention_dropout,
+                        relu_dropout,
+                        preprocess_cmd,
+                        postprocess_cmd,
+                    ),
+                )
+            )
+        self.processor = PrePostProcessLayer(
+            preprocess_cmd, d_model, prepostprocess_dropout
+        )
+
+    def forward(self, enc_input, attn_bias):
+        for encoder_layer in self.encoder_layers:
+            enc_output = encoder_layer(enc_input, attn_bias)
+            enc_input = enc_output
+        enc_output = self.processor(enc_output)
+        return enc_output
+
+
+class EncoderLayer(nn.Layer):
+    """
+    EncoderLayer
+    """
+
+    def __init__(
+        self,
+        n_head,
+        d_key,
+        d_value,
+        d_model,
+        d_inner_hid,
+        prepostprocess_dropout,
+        attention_dropout,
+        relu_dropout,
+        preprocess_cmd="n",
+        postprocess_cmd="da",
+    ):
+        super(EncoderLayer, self).__init__()
+        self.preprocesser1 = PrePostProcessLayer(
+            preprocess_cmd, d_model, prepostprocess_dropout
+        )
+        self.self_attn = MultiHeadAttention(
+            d_key, d_value, d_model, n_head, attention_dropout
+        )
+        self.postprocesser1 = PrePostProcessLayer(
+            postprocess_cmd, d_model, prepostprocess_dropout
+        )
+
+        self.preprocesser2 = PrePostProcessLayer(
+            preprocess_cmd, d_model, prepostprocess_dropout
+        )
+        self.ffn = FFN(d_inner_hid, d_model, relu_dropout)
+        self.postprocesser2 = PrePostProcessLayer(
+            postprocess_cmd, d_model, prepostprocess_dropout
+        )
+
+    def forward(self, enc_input, attn_bias):
+        attn_output = self.self_attn(
+            self.preprocesser1(enc_input), None, None, attn_bias
+        )
+        attn_output = self.postprocesser1(attn_output, enc_input)
+        ffn_output = self.ffn(self.preprocesser2(attn_output))
+        ffn_output = self.postprocesser2(ffn_output, attn_output)
+        return ffn_output
+
+
+class MultiHeadAttention(nn.Layer):
+    """
+    Multi-Head Attention
+    """
+
+    def __init__(self, d_key, d_value, d_model, n_head=1, dropout_rate=0.0):
+        super(MultiHeadAttention, self).__init__()
+        self.n_head = n_head
+        self.d_key = d_key
+        self.d_value = d_value
+        self.d_model = d_model
+        self.dropout_rate = dropout_rate
+        self.q_fc = paddle.nn.Linear(
+            in_features=d_model, out_features=d_key * n_head, bias_attr=False
+        )
+        self.k_fc = paddle.nn.Linear(
+            in_features=d_model, out_features=d_key * n_head, bias_attr=False
+        )
+        self.v_fc = paddle.nn.Linear(
+            in_features=d_model, out_features=d_value * n_head, bias_attr=False
+        )
+        self.proj_fc = paddle.nn.Linear(
+            in_features=d_value * n_head, out_features=d_model, bias_attr=False
+        )
+
+    def _prepare_qkv(self, queries, keys, values, cache=None):
+        if keys is None:  # self-attention
+            keys, values = queries, queries
+            static_kv = False
+        else:  # cross-attention
+            static_kv = True
+
+        q = self.q_fc(queries)
+        q = paddle.reshape(x=q, shape=[0, 0, self.n_head, self.d_key])
+        q = paddle.transpose(x=q, perm=[0, 2, 1, 3])
+
+        if cache is not None and static_kv and "static_k" in cache:
+            # for encoder-decoder attention in inference and has cached
+            k = cache["static_k"]
+            v = cache["static_v"]
+        else:
+            k = self.k_fc(keys)
+            v = self.v_fc(values)
+            k = paddle.reshape(x=k, shape=[0, 0, self.n_head, self.d_key])
+            k = paddle.transpose(x=k, perm=[0, 2, 1, 3])
+            v = paddle.reshape(x=v, shape=[0, 0, self.n_head, self.d_value])
+            v = paddle.transpose(x=v, perm=[0, 2, 1, 3])
+
+        if cache is not None:
+            if static_kv and not "static_k" in cache:
+                # for encoder-decoder attention in inference and has not cached
+                cache["static_k"], cache["static_v"] = k, v
+            elif not static_kv:
+                # for decoder self-attention in inference
+                cache_k, cache_v = cache["k"], cache["v"]
+                k = paddle.concat([cache_k, k], axis=2)
+                v = paddle.concat([cache_v, v], axis=2)
+                cache["k"], cache["v"] = k, v
+
+        return q, k, v
+
+    def forward(self, queries, keys, values, attn_bias, cache=None):
+        # compute q ,k ,v
+        keys = queries if keys is None else keys
+        values = keys if values is None else values
+        q, k, v = self._prepare_qkv(queries, keys, values, cache)
+
+        # scale dot product attention
+        product = paddle.matmul(x=q, y=k, transpose_y=True)
+        product = product * self.d_model**-0.5
+        if attn_bias is not None:
+            product += attn_bias.astype(product.dtype)
+        weights = F.softmax(product)
+        if self.dropout_rate:
+            weights = F.dropout(weights, p=self.dropout_rate, mode="downscale_in_infer")
+        out = paddle.matmul(weights, v)
+
+        # combine heads
+        out = paddle.transpose(out, perm=[0, 2, 1, 3])
+        out = paddle.reshape(x=out, shape=[0, 0, out.shape[2] * out.shape[3]])
+
+        # project to output
+        out = self.proj_fc(out)
+
+        return out
+
+
+class PrePostProcessLayer(nn.Layer):
+    """
+    PrePostProcessLayer
+    """
+
+    def __init__(self, process_cmd, d_model, dropout_rate):
+        super(PrePostProcessLayer, self).__init__()
+        self.process_cmd = process_cmd
+        self.functors = []
+        for cmd in self.process_cmd:
+            if cmd == "a":  # add residual connection
+                self.functors.append(lambda x, y: x + y if y is not None else x)
+            elif cmd == "n":  # add layer normalization
+                self.functors.append(
+                    self.add_sublayer(
+                        "layer_norm_%d" % len(self.sublayers()),
+                        paddle.nn.LayerNorm(
+                            normalized_shape=d_model,
+                            weight_attr=paddle.ParamAttr(
+                                initializer=paddle.nn.initializer.Constant(1.0)
+                            ),
+                            bias_attr=paddle.ParamAttr(
+                                initializer=paddle.nn.initializer.Constant(0.0)
+                            ),
+                        ),
+                    )
+                )
+            elif cmd == "d":  # add dropout
+                self.functors.append(
+                    lambda x: (
+                        F.dropout(x, p=dropout_rate, mode="downscale_in_infer")
+                        if dropout_rate
+                        else x
+                    )
+                )
+
+    def forward(self, x, residual=None):
+        for i, cmd in enumerate(self.process_cmd):
+            if cmd == "a":
+                x = self.functors[i](x, residual)
+            else:
+                x = self.functors[i](x)
+        return x
+
+
+class PrepareEncoder(nn.Layer):
+    def __init__(
+        self,
+        src_vocab_size,
+        src_emb_dim,
+        src_max_len,
+        dropout_rate=0,
+        bos_idx=0,
+        word_emb_param_name=None,
+        pos_enc_param_name=None,
+    ):
+        super(PrepareEncoder, self).__init__()
+        self.src_emb_dim = src_emb_dim
+        self.src_max_len = src_max_len
+        self.emb = paddle.nn.Embedding(
+            num_embeddings=self.src_max_len, embedding_dim=self.src_emb_dim
+        )
+        self.dropout_rate = dropout_rate
+
+    def forward(self, src_word, src_pos):
+        src_word_emb = src_word
+        src_word_emb = paddle.cast(src_word_emb, "float32")
+        src_word_emb = paddle.scale(x=src_word_emb, scale=self.src_emb_dim**0.5)
+        src_pos = paddle.squeeze(src_pos, axis=-1)
+        src_pos_enc = self.emb(src_pos)
+        src_pos_enc.stop_gradient = True
+        enc_input = src_word_emb + src_pos_enc
+        if self.dropout_rate:
+            out = F.dropout(x=enc_input, p=self.dropout_rate, mode="downscale_in_infer")
+        else:
+            out = enc_input
+        return out
+
+
+class PrepareDecoder(nn.Layer):
+    def __init__(
+        self,
+        src_vocab_size,
+        src_emb_dim,
+        src_max_len,
+        dropout_rate=0,
+        bos_idx=0,
+        word_emb_param_name=None,
+        pos_enc_param_name=None,
+    ):
+        super(PrepareDecoder, self).__init__()
+        self.src_emb_dim = src_emb_dim
+        """
+        self.emb0 = Embedding(num_embeddings=src_vocab_size,
+                              embedding_dim=src_emb_dim)
+        """
+        self.emb0 = paddle.nn.Embedding(
+            num_embeddings=src_vocab_size,
+            embedding_dim=self.src_emb_dim,
+            padding_idx=bos_idx,
+            weight_attr=paddle.ParamAttr(
+                name=word_emb_param_name,
+                initializer=nn.initializer.Normal(0.0, src_emb_dim**-0.5),
+            ),
+        )
+        self.emb1 = paddle.nn.Embedding(
+            num_embeddings=src_max_len,
+            embedding_dim=self.src_emb_dim,
+            weight_attr=paddle.ParamAttr(name=pos_enc_param_name),
+        )
+        self.dropout_rate = dropout_rate
+
+    def forward(self, src_word, src_pos):
+        src_word = paddle.cast(src_word, "int64")
+        src_word = paddle.squeeze(src_word, axis=-1)
+        src_word_emb = self.emb0(src_word)
+        src_word_emb = paddle.scale(x=src_word_emb, scale=self.src_emb_dim**0.5)
+        src_pos = paddle.squeeze(src_pos, axis=-1)
+        src_pos_enc = self.emb1(src_pos)
+        src_pos_enc.stop_gradient = True
+        enc_input = src_word_emb + src_pos_enc
+        if self.dropout_rate:
+            out = F.dropout(x=enc_input, p=self.dropout_rate, mode="downscale_in_infer")
+        else:
+            out = enc_input
+        return out
+
+
+class FFN(nn.Layer):
+    """
+    Feed-Forward Network
+    """
+
+    def __init__(self, d_inner_hid, d_model, dropout_rate):
+        super(FFN, self).__init__()
+        self.dropout_rate = dropout_rate
+        self.fc1 = paddle.nn.Linear(in_features=d_model, out_features=d_inner_hid)
+        self.fc2 = paddle.nn.Linear(in_features=d_inner_hid, out_features=d_model)
+
+    def forward(self, x):
+        hidden = self.fc1(x)
+        hidden = F.relu(hidden)
+        if self.dropout_rate:
+            hidden = F.dropout(hidden, p=self.dropout_rate, mode="downscale_in_infer")
+        out = self.fc2(hidden)
+        return out
diff --git a/ppocr/modeling/heads/sr_rensnet_transformer.py b/ppocr/modeling/heads/sr_rensnet_transformer.py
new file mode 100644
index 0000000..dcb8bfb
--- /dev/null
+++ b/ppocr/modeling/heads/sr_rensnet_transformer.py
@@ -0,0 +1,427 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/FudanVI/FudanOCR/blob/main/text-gestalt/loss/transformer_english_decomposition.py
+"""
+import copy
+import math
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+
+def subsequent_mask(size):
+    """Generate a square mask for the sequence. The masked positions are filled with float('-inf').
+    Unmasked positions are filled with float(0.0).
+    """
+    mask = paddle.ones([1, size, size], dtype="float32")
+    mask_inf = paddle.triu(
+        paddle.full(shape=[1, size, size], dtype="float32", fill_value="-inf"),
+        diagonal=1,
+    )
+    mask = mask + mask_inf
+    padding_mask = paddle.equal(mask, paddle.to_tensor(1, dtype=mask.dtype))
+    return padding_mask
+
+
+def clones(module, N):
+    return nn.LayerList([copy.deepcopy(module) for _ in range(N)])
+
+
+def masked_fill(x, mask, value):
+    y = paddle.full(x.shape, value, x.dtype)
+    return paddle.where(mask, y, x)
+
+
+def attention(query, key, value, mask=None, dropout=None, attention_map=None):
+    d_k = query.shape[-1]
+    scores = paddle.matmul(query, paddle.transpose(key, [0, 1, 3, 2])) / math.sqrt(d_k)
+
+    if mask is not None:
+        scores = masked_fill(scores, mask == 0, float("-inf"))
+    else:
+        pass
+
+    p_attn = F.softmax(scores, axis=-1)
+
+    if dropout is not None:
+        p_attn = dropout(p_attn)
+    return paddle.matmul(p_attn, value), p_attn
+
+
+class MultiHeadedAttention(nn.Layer):
+    def __init__(self, h, d_model, dropout=0.1, compress_attention=False):
+        super(MultiHeadedAttention, self).__init__()
+        assert d_model % h == 0
+        self.d_k = d_model // h
+        self.h = h
+        self.linears = clones(nn.Linear(d_model, d_model), 4)
+        self.attn = None
+        self.dropout = nn.Dropout(p=dropout, mode="downscale_in_infer")
+        self.compress_attention = compress_attention
+        self.compress_attention_linear = nn.Linear(h, 1)
+
+    def forward(self, query, key, value, mask=None, attention_map=None):
+        if mask is not None:
+            mask = mask.unsqueeze(1)
+        nbatches = query.shape[0]
+
+        query, key, value = [
+            paddle.transpose(
+                l(x).reshape([nbatches, -1, self.h, self.d_k]), [0, 2, 1, 3]
+            )
+            for l, x in zip(self.linears, (query, key, value))
+        ]
+
+        x, attention_map = attention(
+            query,
+            key,
+            value,
+            mask=mask,
+            dropout=self.dropout,
+            attention_map=attention_map,
+        )
+
+        x = paddle.reshape(
+            paddle.transpose(x, [0, 2, 1, 3]), [nbatches, -1, self.h * self.d_k]
+        )
+
+        return self.linears[-1](x), attention_map
+
+
+class ResNet(nn.Layer):
+    def __init__(self, num_in, block, layers):
+        super(ResNet, self).__init__()
+
+        self.conv1 = nn.Conv2D(num_in, 64, kernel_size=3, stride=1, padding=1)
+        self.bn1 = nn.BatchNorm2D(64, use_global_stats=True)
+        self.relu1 = nn.ReLU()
+        self.pool = nn.MaxPool2D((2, 2), (2, 2))
+
+        self.conv2 = nn.Conv2D(64, 128, kernel_size=3, stride=1, padding=1)
+        self.bn2 = nn.BatchNorm2D(128, use_global_stats=True)
+        self.relu2 = nn.ReLU()
+
+        self.layer1_pool = nn.MaxPool2D((2, 2), (2, 2))
+        self.layer1 = self._make_layer(block, 128, 256, layers[0])
+        self.layer1_conv = nn.Conv2D(256, 256, 3, 1, 1)
+        self.layer1_bn = nn.BatchNorm2D(256, use_global_stats=True)
+        self.layer1_relu = nn.ReLU()
+
+        self.layer2_pool = nn.MaxPool2D((2, 2), (2, 2))
+        self.layer2 = self._make_layer(block, 256, 256, layers[1])
+        self.layer2_conv = nn.Conv2D(256, 256, 3, 1, 1)
+        self.layer2_bn = nn.BatchNorm2D(256, use_global_stats=True)
+        self.layer2_relu = nn.ReLU()
+
+        self.layer3_pool = nn.MaxPool2D((2, 2), (2, 2))
+        self.layer3 = self._make_layer(block, 256, 512, layers[2])
+        self.layer3_conv = nn.Conv2D(512, 512, 3, 1, 1)
+        self.layer3_bn = nn.BatchNorm2D(512, use_global_stats=True)
+        self.layer3_relu = nn.ReLU()
+
+        self.layer4_pool = nn.MaxPool2D((2, 2), (2, 2))
+        self.layer4 = self._make_layer(block, 512, 512, layers[3])
+        self.layer4_conv2 = nn.Conv2D(512, 1024, 3, 1, 1)
+        self.layer4_conv2_bn = nn.BatchNorm2D(1024, use_global_stats=True)
+        self.layer4_conv2_relu = nn.ReLU()
+
+    def _make_layer(self, block, inplanes, planes, blocks):
+        if inplanes != planes:
+            downsample = nn.Sequential(
+                nn.Conv2D(inplanes, planes, 3, 1, 1),
+                nn.BatchNorm2D(planes, use_global_stats=True),
+            )
+        else:
+            downsample = None
+        layers = []
+        layers.append(block(inplanes, planes, downsample))
+        for i in range(1, blocks):
+            layers.append(block(planes, planes, downsample=None))
+
+        return nn.Sequential(*layers)
+
+    def forward(self, x):
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = self.relu1(x)
+        x = self.pool(x)
+
+        x = self.conv2(x)
+        x = self.bn2(x)
+        x = self.relu2(x)
+
+        x = self.layer1_pool(x)
+        x = self.layer1(x)
+        x = self.layer1_conv(x)
+        x = self.layer1_bn(x)
+        x = self.layer1_relu(x)
+
+        x = self.layer2(x)
+        x = self.layer2_conv(x)
+        x = self.layer2_bn(x)
+        x = self.layer2_relu(x)
+
+        x = self.layer3(x)
+        x = self.layer3_conv(x)
+        x = self.layer3_bn(x)
+        x = self.layer3_relu(x)
+
+        x = self.layer4(x)
+        x = self.layer4_conv2(x)
+        x = self.layer4_conv2_bn(x)
+        x = self.layer4_conv2_relu(x)
+
+        return x
+
+
+class Bottleneck(nn.Layer):
+    def __init__(self, input_dim):
+        super(Bottleneck, self).__init__()
+        self.conv1 = nn.Conv2D(input_dim, input_dim, 1)
+        self.bn1 = nn.BatchNorm2D(input_dim, use_global_stats=True)
+        self.relu = nn.ReLU()
+
+        self.conv2 = nn.Conv2D(input_dim, input_dim, 3, 1, 1)
+        self.bn2 = nn.BatchNorm2D(input_dim, use_global_stats=True)
+
+    def forward(self, x):
+        residual = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+
+        out += residual
+        out = self.relu(out)
+
+        return out
+
+
+class PositionalEncoding(nn.Layer):
+    "Implement the PE function."
+
+    def __init__(self, dropout, dim, max_len=5000):
+        super(PositionalEncoding, self).__init__()
+        self.dropout = nn.Dropout(p=dropout, mode="downscale_in_infer")
+
+        pe = paddle.zeros([max_len, dim])
+        position = paddle.arange(0, max_len, dtype=paddle.float32).unsqueeze(1)
+        div_term = paddle.exp(
+            paddle.arange(0, dim, 2).astype("float32") * (-math.log(10000.0) / dim)
+        )
+        pe[:, 0::2] = paddle.sin(position * div_term)
+        pe[:, 1::2] = paddle.cos(position * div_term)
+        pe = paddle.unsqueeze(pe, 0)
+        self.register_buffer("pe", pe)
+
+    def forward(self, x):
+        x = x + self.pe[:, : x.shape[1]]
+        return self.dropout(x)
+
+
+class PositionwiseFeedForward(nn.Layer):
+    "Implements FFN equation."
+
+    def __init__(self, d_model, d_ff, dropout=0.1):
+        super(PositionwiseFeedForward, self).__init__()
+        self.w_1 = nn.Linear(d_model, d_ff)
+        self.w_2 = nn.Linear(d_ff, d_model)
+        self.dropout = nn.Dropout(dropout, mode="downscale_in_infer")
+
+    def forward(self, x):
+        return self.w_2(self.dropout(F.relu(self.w_1(x))))
+
+
+class Generator(nn.Layer):
+    "Define standard linear + softmax generation step."
+
+    def __init__(self, d_model, vocab):
+        super(Generator, self).__init__()
+        self.proj = nn.Linear(d_model, vocab)
+        self.relu = nn.ReLU()
+
+    def forward(self, x):
+        out = self.proj(x)
+        return out
+
+
+class Embeddings(nn.Layer):
+    def __init__(self, d_model, vocab):
+        super(Embeddings, self).__init__()
+        self.lut = nn.Embedding(vocab, d_model)
+        self.d_model = d_model
+
+    def forward(self, x):
+        embed = self.lut(x) * math.sqrt(self.d_model)
+        return embed
+
+
+class LayerNorm(nn.Layer):
+    "Construct a layernorm module (See citation for details)."
+
+    def __init__(self, features, eps=1e-6):
+        super(LayerNorm, self).__init__()
+        self.a_2 = self.create_parameter(
+            shape=[features], default_initializer=paddle.nn.initializer.Constant(1.0)
+        )
+        self.b_2 = self.create_parameter(
+            shape=[features], default_initializer=paddle.nn.initializer.Constant(0.0)
+        )
+        self.eps = eps
+
+    def forward(self, x):
+        mean = x.mean(-1, keepdim=True)
+        std = x.std(-1, keepdim=True)
+        return self.a_2 * (x - mean) / (std + self.eps) + self.b_2
+
+
+class Decoder(nn.Layer):
+    def __init__(self):
+        super(Decoder, self).__init__()
+
+        self.mask_multihead = MultiHeadedAttention(h=16, d_model=1024, dropout=0.1)
+        self.mul_layernorm1 = LayerNorm(1024)
+
+        self.multihead = MultiHeadedAttention(h=16, d_model=1024, dropout=0.1)
+        self.mul_layernorm2 = LayerNorm(1024)
+
+        self.pff = PositionwiseFeedForward(1024, 2048)
+        self.mul_layernorm3 = LayerNorm(1024)
+
+    def forward(self, text, conv_feature, attention_map=None):
+        text_max_length = text.shape[1]
+        mask = subsequent_mask(text_max_length)
+        result = text
+        result = self.mul_layernorm1(
+            result + self.mask_multihead(text, text, text, mask=mask)[0]
+        )
+        b, c, h, w = conv_feature.shape
+        conv_feature = paddle.transpose(conv_feature.reshape([b, c, h * w]), [0, 2, 1])
+        word_image_align, attention_map = self.multihead(
+            result, conv_feature, conv_feature, mask=None, attention_map=attention_map
+        )
+        result = self.mul_layernorm2(result + word_image_align)
+        result = self.mul_layernorm3(result + self.pff(result))
+
+        return result, attention_map
+
+
+class BasicBlock(nn.Layer):
+    def __init__(self, inplanes, planes, downsample):
+        super(BasicBlock, self).__init__()
+        self.conv1 = nn.Conv2D(inplanes, planes, kernel_size=3, stride=1, padding=1)
+        self.bn1 = nn.BatchNorm2D(planes, use_global_stats=True)
+        self.relu = nn.ReLU()
+        self.conv2 = nn.Conv2D(planes, planes, kernel_size=3, stride=1, padding=1)
+        self.bn2 = nn.BatchNorm2D(planes, use_global_stats=True)
+        self.downsample = downsample
+
+    def forward(self, x):
+        residual = x
+
+        out = self.conv1(x)
+        out = self.bn1(out)
+        out = self.relu(out)
+
+        out = self.conv2(out)
+        out = self.bn2(out)
+
+        if self.downsample != None:
+            residual = self.downsample(residual)
+
+        out += residual
+        out = self.relu(out)
+
+        return out
+
+
+class Encoder(nn.Layer):
+    def __init__(self):
+        super(Encoder, self).__init__()
+        self.cnn = ResNet(num_in=1, block=BasicBlock, layers=[1, 2, 5, 3])
+
+    def forward(self, input):
+        conv_result = self.cnn(input)
+        return conv_result
+
+
+class Transformer(nn.Layer):
+    def __init__(self, in_channels=1, alphabet="0123456789"):
+        super(Transformer, self).__init__()
+        self.alphabet = alphabet
+        word_n_class = self.get_alphabet_len()
+        self.embedding_word_with_upperword = Embeddings(512, word_n_class)
+        self.pe = PositionalEncoding(dim=512, dropout=0.1, max_len=5000)
+
+        self.encoder = Encoder()
+        self.decoder = Decoder()
+        self.generator_word_with_upperword = Generator(1024, word_n_class)
+
+        for p in self.parameters():
+            if p.dim() > 1:
+                nn.initializer.XavierNormal(p)
+
+    def get_alphabet_len(self):
+        return len(self.alphabet)
+
+    def forward(self, image, text_length, text_input, attention_map=None):
+        if image.shape[1] == 3:
+            R = image[:, 0:1, :, :]
+            G = image[:, 1:2, :, :]
+            B = image[:, 2:3, :, :]
+            image = 0.299 * R + 0.587 * G + 0.114 * B
+
+        conv_feature = self.encoder(image)  # batch, 1024, 8, 32
+        max_length = max(text_length)
+        text_input = text_input[:, :max_length]
+
+        text_embedding = self.embedding_word_with_upperword(
+            text_input
+        )  # batch, text_max_length, 512
+        postion_embedding = self.pe(
+            paddle.zeros(text_embedding.shape)
+        )  # batch, text_max_length, 512
+        text_input_with_pe = paddle.concat(
+            [text_embedding, postion_embedding], 2
+        )  # batch, text_max_length, 1024
+        batch, seq_len, _ = text_input_with_pe.shape
+
+        text_input_with_pe, word_attention_map = self.decoder(
+            text_input_with_pe, conv_feature
+        )
+
+        word_decoder_result = self.generator_word_with_upperword(text_input_with_pe)
+
+        if self.training:
+            total_length = paddle.sum(text_length)
+            probs_res = paddle.zeros([total_length, self.get_alphabet_len()])
+            start = 0
+
+            for index, length in enumerate(text_length):
+                length = int(length.numpy())
+                probs_res[start : start + length, :] = word_decoder_result[
+                    index, 0 : 0 + length, :
+                ]
+
+                start = start + length
+
+            return probs_res, word_attention_map, None
+        else:
+            return word_decoder_result
diff --git a/ppocr/modeling/heads/table_att_head.py b/ppocr/modeling/heads/table_att_head.py
new file mode 100644
index 0000000..1f43af1
--- /dev/null
+++ b/ppocr/modeling/heads/table_att_head.py
@@ -0,0 +1,421 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+import paddle.nn as nn
+from paddle import ParamAttr
+import paddle.nn.functional as F
+import numpy as np
+
+from .rec_att_head import AttentionGRUCell
+from ppocr.modeling.backbones.rec_svtrnet import DropPath, Identity, Mlp
+
+
+def get_para_bias_attr(l2_decay, k):
+    if l2_decay > 0:
+        regularizer = paddle.regularizer.L2Decay(l2_decay)
+        stdv = 1.0 / math.sqrt(k * 1.0)
+        initializer = nn.initializer.Uniform(-stdv, stdv)
+    else:
+        regularizer = None
+        initializer = None
+    weight_attr = ParamAttr(regularizer=regularizer, initializer=initializer)
+    bias_attr = ParamAttr(regularizer=regularizer, initializer=initializer)
+    return [weight_attr, bias_attr]
+
+
+class TableAttentionHead(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        hidden_size,
+        in_max_len=488,
+        max_text_length=800,
+        out_channels=30,
+        loc_reg_num=4,
+        **kwargs,
+    ):
+        super(TableAttentionHead, self).__init__()
+        self.input_size = in_channels[-1]
+        self.hidden_size = hidden_size
+        self.out_channels = out_channels
+        self.max_text_length = max_text_length
+
+        self.structure_attention_cell = AttentionGRUCell(
+            self.input_size, hidden_size, self.out_channels, use_gru=False
+        )
+        self.structure_generator = nn.Linear(hidden_size, self.out_channels)
+        self.in_max_len = in_max_len
+
+        if self.in_max_len == 640:
+            self.loc_fea_trans = nn.Linear(400, self.max_text_length + 1)
+        elif self.in_max_len == 800:
+            self.loc_fea_trans = nn.Linear(625, self.max_text_length + 1)
+        else:
+            self.loc_fea_trans = nn.Linear(256, self.max_text_length + 1)
+        self.loc_generator = nn.Linear(self.input_size + hidden_size, loc_reg_num)
+
+    def _char_to_onehot(self, input_char, onehot_dim):
+        input_ont_hot = F.one_hot(input_char, onehot_dim)
+        return input_ont_hot
+
+    def forward(self, inputs, targets=None):
+        # if and else branch are both needed when you want to assign a variable
+        # if you modify the var in just one branch, then the modification will not work.
+        fea = inputs[-1]
+        last_shape = int(np.prod(fea.shape[2:]))  # gry added
+        fea = paddle.reshape(fea, [fea.shape[0], fea.shape[1], last_shape])
+        fea = fea.transpose([0, 2, 1])  # (NTC)(batch, width, channels)
+        batch_size = fea.shape[0]
+
+        hidden = paddle.zeros((batch_size, self.hidden_size))
+        output_hiddens = paddle.zeros(
+            (batch_size, self.max_text_length + 1, self.hidden_size)
+        )
+        if self.training and targets is not None:
+            structure = targets[0]
+            for i in range(self.max_text_length + 1):
+                elem_onehots = self._char_to_onehot(
+                    structure[:, i], onehot_dim=self.out_channels
+                )
+                (outputs, hidden), alpha = self.structure_attention_cell(
+                    hidden, fea, elem_onehots
+                )
+                output_hiddens[:, i, :] = outputs
+            structure_probs = self.structure_generator(output_hiddens)
+            loc_fea = fea.transpose([0, 2, 1])
+            loc_fea = self.loc_fea_trans(loc_fea)
+            loc_fea = loc_fea.transpose([0, 2, 1])
+            loc_concat = paddle.concat([output_hiddens, loc_fea], axis=2)
+            loc_preds = self.loc_generator(loc_concat)
+            loc_preds = F.sigmoid(loc_preds)
+        else:
+            temp_elem = paddle.zeros(shape=[batch_size], dtype="int32")
+            structure_probs = None
+            loc_preds = None
+            elem_onehots = None
+            outputs = None
+            alpha = None
+            max_text_length = paddle.to_tensor(self.max_text_length)
+            for i in range(max_text_length + 1):
+                elem_onehots = self._char_to_onehot(
+                    temp_elem, onehot_dim=self.out_channels
+                )
+                (outputs, hidden), alpha = self.structure_attention_cell(
+                    hidden, fea, elem_onehots
+                )
+                output_hiddens[:, i, :] = outputs
+                structure_probs_step = self.structure_generator(outputs)
+                temp_elem = structure_probs_step.argmax(axis=1, dtype="int32")
+
+            structure_probs = self.structure_generator(output_hiddens)
+            structure_probs = F.softmax(structure_probs)
+            loc_fea = fea.transpose([0, 2, 1])
+            loc_fea = self.loc_fea_trans(loc_fea)
+            loc_fea = loc_fea.transpose([0, 2, 1])
+            loc_concat = paddle.concat([output_hiddens, loc_fea], axis=2)
+            loc_preds = self.loc_generator(loc_concat)
+            loc_preds = F.sigmoid(loc_preds)
+        return {"structure_probs": structure_probs, "loc_preds": loc_preds}
+
+
+class HWAttention(nn.Layer):
+    def __init__(
+        self,
+        head_dim=32,
+        qk_scale=None,
+        attn_drop=0.0,
+    ):
+        super().__init__()
+        self.head_dim = head_dim
+        self.scale = qk_scale or self.head_dim**-0.5
+        self.attn_drop = nn.Dropout(attn_drop)
+
+    def forward(self, x):
+        B, N, C = x.shape
+        C = C // 3
+        qkv = x.reshape([B, N, 3, C // self.head_dim, self.head_dim]).transpose(
+            [2, 0, 3, 1, 4]
+        )
+        q, k, v = qkv.unbind(0)
+        attn = q @ k.transpose([0, 1, 3, 2]) * self.scale
+        attn = F.softmax(attn, -1)
+        attn = self.attn_drop(attn)
+        x = attn @ v
+        x = x.transpose([0, 2, 1]).reshape([B, N, C])
+        return x
+
+
+def img2windows(img, H_sp, W_sp):
+    """
+    img: B C H W
+    """
+    B, H, W, C = img.shape
+    img_reshape = img.reshape([B, H // H_sp, H_sp, W // W_sp, W_sp, C])
+    img_perm = img_reshape.transpose([0, 1, 3, 2, 4, 5]).reshape([-1, H_sp * W_sp, C])
+    return img_perm
+
+
+def windows2img(img_splits_hw, H_sp, W_sp, H, W):
+    """
+    img_splits_hw: B' H W C
+    """
+    B = int(img_splits_hw.shape[0] / (H * W / H_sp / W_sp))
+
+    img = img_splits_hw.reshape([B, H // H_sp, W // W_sp, H_sp, W_sp, -1])
+    img = img.transpose([0, 1, 3, 2, 4, 5]).flatten(1, 4)
+    return img
+
+
+class Block(nn.Layer):
+    def __init__(
+        self,
+        dim,
+        num_heads,
+        split_h=4,
+        split_w=4,
+        h_num_heads=None,
+        w_num_heads=None,
+        mlp_ratio=4.0,
+        qkv_bias=False,
+        qk_scale=None,
+        drop=0.0,
+        attn_drop=0.0,
+        drop_path=0.0,
+        act_layer=nn.GELU,
+        norm_layer=nn.LayerNorm,
+        eps=1e-6,
+    ):
+        super().__init__()
+        self.qkv = nn.Linear(dim, dim * 3, bias_attr=qkv_bias)
+        self.proj = nn.Linear(dim, dim)
+        self.split_h = split_h
+        self.split_w = split_w
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.norm1 = norm_layer(dim, epsilon=eps)
+        self.h_num_heads = h_num_heads if h_num_heads is not None else num_heads // 2
+        self.w_num_heads = w_num_heads if w_num_heads is not None else num_heads // 2
+        self.head_dim = dim // num_heads
+        self.mixer = HWAttention(
+            head_dim=dim // num_heads,
+            qk_scale=qk_scale,
+            attn_drop=attn_drop,
+        )
+        self.drop_path = DropPath(drop_path) if drop_path > 0.0 else Identity()
+        self.norm2 = norm_layer(dim, epsilon=eps)
+        self.mlp = Mlp(
+            in_features=dim,
+            hidden_features=mlp_hidden_dim,
+            act_layer=act_layer,
+            drop=drop,
+        )
+
+    def forward(self, x):
+        B, C, H, W = x.shape
+        x = x.flatten(2).transpose([0, 2, 1])
+
+        qkv = self.qkv(x).reshape([B, H, W, 3 * C])
+
+        x1 = qkv[:, :, :, : 3 * self.h_num_heads * self.head_dim]  # b, h, w, 3ch
+        x2 = qkv[:, :, :, 3 * self.h_num_heads * self.head_dim :]  # b, h, w, 3cw
+
+        x1 = self.mixer(img2windows(x1, self.split_h, W))  # b*splith, W, 3ch
+        x2 = self.mixer(img2windows(x2, H, self.split_w))  # b*splitw, h, 3ch
+        x1 = windows2img(x1, self.split_h, W, H, W)
+        x2 = windows2img(x2, H, self.split_w, H, W)
+
+        attened_x = paddle.concat([x1, x2], 2)
+        attened_x = self.proj(attened_x)
+
+        x = self.norm1(x + self.drop_path(attened_x))
+        x = self.norm2(x + self.drop_path(self.mlp(x)))
+        x = x.transpose([0, 2, 1]).reshape([-1, C, H, W])
+        return x
+
+
+class SLAHead(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        hidden_size,
+        out_channels=30,
+        max_text_length=500,
+        loc_reg_num=4,
+        fc_decay=0.0,
+        use_attn=False,
+        **kwargs,
+    ):
+        """
+        @param in_channels: input shape
+        @param hidden_size: hidden_size for RNN and Embedding
+        @param out_channels: num_classes to rec
+        @param max_text_length: max text pred
+        """
+        super().__init__()
+
+        if isinstance(in_channels, int):
+            self.is_next = True
+            in_channels = 512
+        else:
+            self.is_next = False
+            in_channels = in_channels[-1]
+        self.hidden_size = hidden_size
+        self.max_text_length = max_text_length
+        self.emb = self._char_to_onehot
+        self.num_embeddings = out_channels
+        self.loc_reg_num = loc_reg_num
+        self.eos = self.num_embeddings - 1
+
+        # structure
+        self.structure_attention_cell = AttentionGRUCell(
+            in_channels, hidden_size, self.num_embeddings
+        )
+        weight_attr, bias_attr = get_para_bias_attr(l2_decay=fc_decay, k=hidden_size)
+        weight_attr1_1, bias_attr1_1 = get_para_bias_attr(
+            l2_decay=fc_decay, k=hidden_size
+        )
+        weight_attr1_2, bias_attr1_2 = get_para_bias_attr(
+            l2_decay=fc_decay, k=hidden_size
+        )
+        self.structure_generator = nn.Sequential(
+            nn.Linear(
+                self.hidden_size,
+                self.hidden_size,
+                weight_attr=weight_attr1_2,
+                bias_attr=bias_attr1_2,
+            ),
+            nn.Linear(
+                hidden_size, out_channels, weight_attr=weight_attr, bias_attr=bias_attr
+            ),
+        )
+        dpr = np.linspace(0, 0.1, 2)
+
+        self.use_attn = use_attn
+        if use_attn:
+            layer_list = [
+                Block(
+                    in_channels,
+                    num_heads=2,
+                    mlp_ratio=4.0,
+                    qkv_bias=True,
+                    drop_path=dpr[i],
+                )
+                for i in range(2)
+            ]
+            self.cross_atten = nn.Sequential(*layer_list)
+        # loc
+        weight_attr1, bias_attr1 = get_para_bias_attr(
+            l2_decay=fc_decay, k=self.hidden_size
+        )
+        weight_attr2, bias_attr2 = get_para_bias_attr(
+            l2_decay=fc_decay, k=self.hidden_size
+        )
+        self.loc_generator = nn.Sequential(
+            nn.Linear(
+                self.hidden_size,
+                self.hidden_size,
+                weight_attr=weight_attr1,
+                bias_attr=bias_attr1,
+            ),
+            nn.Linear(
+                self.hidden_size,
+                loc_reg_num,
+                weight_attr=weight_attr2,
+                bias_attr=bias_attr2,
+            ),
+            nn.Sigmoid(),
+        )
+
+    def forward(self, inputs, targets=None):
+        if self.is_next == True:
+            fea = inputs
+            batch_size = fea.shape[0]
+        else:
+            fea = inputs[-1]
+            batch_size = fea.shape[0]
+            if self.use_attn:
+                fea = fea + self.cross_atten(fea)
+            # reshape
+            fea = paddle.reshape(fea, [fea.shape[0], fea.shape[1], -1])
+            fea = fea.transpose([0, 2, 1])  # (NTC)(batch, width, channels)
+
+        hidden = paddle.zeros((batch_size, self.hidden_size))
+        structure_preds = paddle.zeros(
+            (batch_size, self.max_text_length + 1, self.num_embeddings)
+        )
+        loc_preds = paddle.zeros(
+            (batch_size, self.max_text_length + 1, self.loc_reg_num)
+        )
+        structure_preds.stop_gradient = True
+        loc_preds.stop_gradient = True
+
+        if self.training and targets is not None:
+            structure = targets[0]
+            max_len = targets[-2].max().astype("int32")
+            for i in range(max_len + 1):
+                hidden, structure_step, loc_step = self._decode(
+                    structure[:, i], fea, hidden
+                )
+                structure_preds[:, i, :] = structure_step
+                loc_preds[:, i, :] = loc_step
+            structure_preds = structure_preds[:, : max_len + 1]
+            loc_preds = loc_preds[:, : max_len + 1]
+        else:
+            structure_ids = paddle.zeros(
+                (batch_size, self.max_text_length + 1), dtype="int32"
+            )
+            pre_chars = paddle.zeros(shape=[batch_size], dtype="int32")
+            max_text_length = paddle.to_tensor(self.max_text_length)
+            for i in range(max_text_length + 1):
+                hidden, structure_step, loc_step = self._decode(pre_chars, fea, hidden)
+                pre_chars = structure_step.argmax(axis=1, dtype="int32")
+                structure_preds[:, i, :] = structure_step
+                loc_preds[:, i, :] = loc_step
+
+                structure_ids[:, i] = pre_chars
+                if (structure_ids == self.eos).any(-1).all():
+                    break
+        if not self.training:
+            structure_preds = F.softmax(structure_preds[:, : i + 1])
+            loc_preds = loc_preds[:, : i + 1]
+        return {"structure_probs": structure_preds, "loc_preds": loc_preds}
+
+    def _decode(self, pre_chars, features, hidden):
+        """
+        Predict table label and coordinates for each step
+        @param pre_chars: Table label in previous step
+        @param features:
+        @param hidden: hidden status in previous step
+        @return:
+        """
+        emb_feature = self.emb(pre_chars)
+        # output shape is b * self.hidden_size
+        (output, hidden), alpha = self.structure_attention_cell(
+            hidden, features, emb_feature
+        )
+
+        # structure
+        structure_step = self.structure_generator(output)
+        # loc
+        loc_step = self.loc_generator(output)
+        return hidden, structure_step, loc_step
+
+    def _char_to_onehot(self, input_char):
+        input_ont_hot = F.one_hot(input_char, self.num_embeddings)
+        return input_ont_hot
diff --git a/ppocr/modeling/heads/table_master_head.py b/ppocr/modeling/heads/table_master_head.py
new file mode 100644
index 0000000..66ad7a4
--- /dev/null
+++ b/ppocr/modeling/heads/table_master_head.py
@@ -0,0 +1,285 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/JiaquanYe/TableMASTER-mmocr/blob/master/mmocr/models/textrecog/decoders/master_decoder.py
+"""
+
+import copy
+import math
+import paddle
+from paddle import nn
+from paddle.nn import functional as F
+
+
+class TableMasterHead(nn.Layer):
+    """
+    Split to two transformer header at the last layer.
+    Cls_layer is used to structure token classification.
+    Bbox_layer is used to regress bbox coord.
+    """
+
+    def __init__(
+        self,
+        in_channels,
+        out_channels=30,
+        headers=8,
+        d_ff=2048,
+        dropout=0,
+        max_text_length=500,
+        loc_reg_num=4,
+        **kwargs,
+    ):
+        super(TableMasterHead, self).__init__()
+        hidden_size = in_channels[-1]
+        self.layers = clones(DecoderLayer(headers, hidden_size, dropout, d_ff), 2)
+        self.cls_layer = clones(DecoderLayer(headers, hidden_size, dropout, d_ff), 1)
+        self.bbox_layer = clones(DecoderLayer(headers, hidden_size, dropout, d_ff), 1)
+        self.cls_fc = nn.Linear(hidden_size, out_channels)
+        self.bbox_fc = nn.Sequential(
+            # nn.Linear(hidden_size, hidden_size),
+            nn.Linear(hidden_size, loc_reg_num),
+            nn.Sigmoid(),
+        )
+        self.norm = nn.LayerNorm(hidden_size)
+        self.embedding = Embeddings(d_model=hidden_size, vocab=out_channels)
+        self.positional_encoding = PositionalEncoding(d_model=hidden_size)
+
+        self.SOS = out_channels - 3
+        self.PAD = out_channels - 1
+        self.out_channels = out_channels
+        self.loc_reg_num = loc_reg_num
+        self.max_text_length = max_text_length
+
+    def make_mask(self, tgt):
+        """
+        Make mask for self attention.
+        :param src: [b, c, h, l_src]
+        :param tgt: [b, l_tgt]
+        :return:
+        """
+        trg_pad_mask = (tgt != self.PAD).unsqueeze(1).unsqueeze(3)
+
+        tgt_len = tgt.shape[1]
+        trg_sub_mask = paddle.tril(
+            paddle.ones(([tgt_len, tgt_len]), dtype=paddle.float32)
+        )
+
+        tgt_mask = paddle.logical_and(trg_pad_mask.astype(paddle.float32), trg_sub_mask)
+        return tgt_mask.astype(paddle.float32)
+
+    def decode(self, input, feature, src_mask, tgt_mask):
+        # main process of transformer decoder.
+        x = self.embedding(input)  # x: 1*x*512, feature: 1*3600,512
+        x = self.positional_encoding(x)
+
+        # origin transformer layers
+        for i, layer in enumerate(self.layers):
+            x = layer(x, feature, src_mask, tgt_mask)
+
+        # cls head
+        cls_x = x
+        for layer in self.cls_layer:
+            cls_x = layer(x, feature, src_mask, tgt_mask)
+        cls_x = self.norm(cls_x)
+
+        # bbox head
+        bbox_x = x
+        for layer in self.bbox_layer:
+            bbox_x = layer(x, feature, src_mask, tgt_mask)
+        bbox_x = self.norm(bbox_x)
+        return self.cls_fc(cls_x), self.bbox_fc(bbox_x)
+
+    def greedy_forward(self, SOS, feature):
+        input = SOS
+        output = paddle.zeros(
+            [input.shape[0], self.max_text_length + 1, self.out_channels]
+        )
+        bbox_output = paddle.zeros(
+            [input.shape[0], self.max_text_length + 1, self.loc_reg_num]
+        )
+        max_text_length = paddle.to_tensor(self.max_text_length)
+        for i in range(max_text_length + 1):
+            target_mask = self.make_mask(input)
+            out_step, bbox_output_step = self.decode(input, feature, None, target_mask)
+            prob = F.softmax(out_step, axis=-1)
+            next_word = prob.argmax(axis=2, dtype="int64")
+            input = paddle.concat([input, next_word[:, -1].unsqueeze(-1)], axis=1)
+            if i == self.max_text_length:
+                output = out_step
+                bbox_output = bbox_output_step
+        return output, bbox_output
+
+    def forward_train(self, out_enc, targets):
+        # x is token of label
+        # feat is feature after backbone before pe.
+        # out_enc is feature after pe.
+        padded_targets = targets[0]
+        src_mask = None
+        tgt_mask = self.make_mask(padded_targets[:, :-1])
+        output, bbox_output = self.decode(
+            padded_targets[:, :-1], out_enc, src_mask, tgt_mask
+        )
+        return {"structure_probs": output, "loc_preds": bbox_output}
+
+    def forward_test(self, out_enc):
+        batch_size = out_enc.shape[0]
+        SOS = paddle.zeros([batch_size, 1], dtype="int64") + self.SOS
+        output, bbox_output = self.greedy_forward(SOS, out_enc)
+        output = F.softmax(output)
+        return {"structure_probs": output, "loc_preds": bbox_output}
+
+    def forward(self, feat, targets=None):
+        feat = feat[-1]
+        b, c, h, w = feat.shape
+        feat = feat.reshape([b, c, h * w])  # flatten 2D feature map
+        feat = feat.transpose((0, 2, 1))
+        out_enc = self.positional_encoding(feat)
+        if self.training:
+            return self.forward_train(out_enc, targets)
+
+        return self.forward_test(out_enc)
+
+
+class DecoderLayer(nn.Layer):
+    """
+    Decoder is made of self attention, source attention and feed forward.
+    """
+
+    def __init__(self, headers, d_model, dropout, d_ff):
+        super(DecoderLayer, self).__init__()
+        self.self_attn = MultiHeadAttention(headers, d_model, dropout)
+        self.src_attn = MultiHeadAttention(headers, d_model, dropout)
+        self.feed_forward = FeedForward(d_model, d_ff, dropout)
+        self.sublayer = clones(SubLayerConnection(d_model, dropout), 3)
+
+    def forward(self, x, feature, src_mask, tgt_mask):
+        x = self.sublayer[0](x, lambda x: self.self_attn(x, x, x, tgt_mask))
+        x = self.sublayer[1](x, lambda x: self.src_attn(x, feature, feature, src_mask))
+        return self.sublayer[2](x, self.feed_forward)
+
+
+class MultiHeadAttention(nn.Layer):
+    def __init__(self, headers, d_model, dropout):
+        super(MultiHeadAttention, self).__init__()
+
+        assert d_model % headers == 0
+        self.d_k = int(d_model / headers)
+        self.headers = headers
+        self.linears = clones(nn.Linear(d_model, d_model), 4)
+        self.attn = None
+        self.dropout = nn.Dropout(dropout)
+
+    def forward(self, query, key, value, mask=None):
+        B = query.shape[0]
+
+        # 1) Do all the linear projections in batch from d_model => h x d_k
+        query, key, value = [
+            l(x).reshape([B, 0, self.headers, self.d_k]).transpose([0, 2, 1, 3])
+            for l, x in zip(self.linears, (query, key, value))
+        ]
+        # 2) Apply attention on all the projected vectors in batch
+        x, self.attn = self_attention(
+            query, key, value, mask=mask, dropout=self.dropout
+        )
+        x = x.transpose([0, 2, 1, 3]).reshape([B, 0, self.headers * self.d_k])
+        return self.linears[-1](x)
+
+
+class FeedForward(nn.Layer):
+    def __init__(self, d_model, d_ff, dropout):
+        super(FeedForward, self).__init__()
+        self.w_1 = nn.Linear(d_model, d_ff)
+        self.w_2 = nn.Linear(d_ff, d_model)
+        self.dropout = nn.Dropout(dropout)
+
+    def forward(self, x):
+        return self.w_2(self.dropout(F.relu(self.w_1(x))))
+
+
+class SubLayerConnection(nn.Layer):
+    """
+    A residual connection followed by a layer norm.
+    Note for code simplicity the norm is first as opposed to last.
+    """
+
+    def __init__(self, size, dropout):
+        super(SubLayerConnection, self).__init__()
+        self.norm = nn.LayerNorm(size)
+        self.dropout = nn.Dropout(dropout)
+
+    def forward(self, x, sublayer):
+        return x + self.dropout(sublayer(self.norm(x)))
+
+
+def masked_fill(x, mask, value):
+    mask = mask.astype(x.dtype)
+    return x * paddle.logical_not(mask).astype(x.dtype) + mask * value
+
+
+def self_attention(query, key, value, mask=None, dropout=None):
+    """
+    Compute 'Scale Dot Product Attention'
+    """
+    d_k = value.shape[-1]
+
+    score = paddle.matmul(query, key.transpose([0, 1, 3, 2]) / math.sqrt(d_k))
+    if mask is not None:
+        # score = score.masked_fill(mask == 0, -1e9) # b, h, L, L
+        score = masked_fill(score, mask == 0, -6.55e4)  # for fp16
+
+    p_attn = F.softmax(score, axis=-1)
+
+    if dropout is not None:
+        p_attn = dropout(p_attn)
+    return paddle.matmul(p_attn, value), p_attn
+
+
+def clones(module, N):
+    """Produce N identical layers"""
+    return nn.LayerList([copy.deepcopy(module) for _ in range(N)])
+
+
+class Embeddings(nn.Layer):
+    def __init__(self, d_model, vocab):
+        super(Embeddings, self).__init__()
+        self.lut = nn.Embedding(vocab, d_model)
+        self.d_model = d_model
+
+    def forward(self, *input):
+        x = input[0]
+        return self.lut(x) * math.sqrt(self.d_model)
+
+
+class PositionalEncoding(nn.Layer):
+    """Implement the PE function."""
+
+    def __init__(self, d_model, dropout=0.0, max_len=5000):
+        super(PositionalEncoding, self).__init__()
+        self.dropout = nn.Dropout(p=dropout)
+
+        # Compute the positional encodings once in log space.
+        pe = paddle.zeros([max_len, d_model])
+        position = paddle.arange(0, max_len).unsqueeze(1).astype("float32")
+        div_term = paddle.exp(
+            paddle.arange(0, d_model, 2) * -math.log(10000.0) / d_model
+        )
+        pe[:, 0::2] = paddle.sin(position * div_term)
+        pe[:, 1::2] = paddle.cos(position * div_term)
+        pe = pe.unsqueeze(0)
+        self.register_buffer("pe", pe)
+
+    def forward(self, feat, **kwargs):
+        feat = feat + self.pe[:, : feat.shape[1]]  # pe 1*5000*512
+        return self.dropout(feat)
diff --git a/ppocr/modeling/necks/__init__.py b/ppocr/modeling/necks/__init__.py
new file mode 100644
index 0000000..bfc6bed
--- /dev/null
+++ b/ppocr/modeling/necks/__init__.py
@@ -0,0 +1,59 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+__all__ = ["build_neck"]
+
+
+def build_neck(config):
+    from .db_fpn import DBFPN, RSEFPN, RepLKFPN, LKPAN, RepLKPAN
+    from .east_fpn import EASTFPN
+    from .sast_fpn import SASTFPN
+    from .rnn import SequenceEncoder
+    from .pg_fpn import PGFPN
+    from .table_fpn import TableFPN
+    from .fpn import FPN
+    from .fce_fpn import FCEFPN
+    from .pren_fpn import PRENFPN
+    from .csp_pan import CSPPAN
+    from .ct_fpn import CTFPN
+    from .fpn_unet import FPN_UNet
+    from .rf_adaptor import RFAdaptor
+
+    support_dict = [
+        "FPN",
+        "FCEFPN",
+        "LKPAN",
+        "RepLKPAN",
+        "DBFPN",
+        "RSEFPN",
+        "RepLKFPN",
+        "EASTFPN",
+        "SASTFPN",
+        "SequenceEncoder",
+        "PGFPN",
+        "TableFPN",
+        "PRENFPN",
+        "CSPPAN",
+        "CTFPN",
+        "RFAdaptor",
+        "FPN_UNet",
+    ]
+
+    module_name = config.pop("name")
+    assert module_name in support_dict, Exception(
+        "neck only support {}".format(support_dict)
+    )
+
+    module_class = eval(module_name)(**config)
+    return module_class
diff --git a/ppocr/modeling/necks/csp_pan.py b/ppocr/modeling/necks/csp_pan.py
new file mode 100755
index 0000000..5e8464d
--- /dev/null
+++ b/ppocr/modeling/necks/csp_pan.py
@@ -0,0 +1,337 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# The code is based on:
+# https://github.com/PaddlePaddle/PaddleDetection/blob/release%2F2.3/ppdet/modeling/necks/csp_pan.py
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+
+__all__ = ["CSPPAN"]
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channel=96,
+        out_channel=96,
+        kernel_size=3,
+        stride=1,
+        groups=1,
+        act="leaky_relu",
+    ):
+        super(ConvBNLayer, self).__init__()
+        initializer = nn.initializer.KaimingUniform()
+        self.act = act
+        assert self.act in ["leaky_relu", "hard_swish"]
+        self.conv = nn.Conv2D(
+            in_channels=in_channel,
+            out_channels=out_channel,
+            kernel_size=kernel_size,
+            groups=groups,
+            padding=(kernel_size - 1) // 2,
+            stride=stride,
+            weight_attr=ParamAttr(initializer=initializer),
+            bias_attr=False,
+        )
+        self.bn = nn.BatchNorm2D(out_channel)
+
+    def forward(self, x):
+        x = self.bn(self.conv(x))
+        if self.act == "leaky_relu":
+            x = F.leaky_relu(x)
+        elif self.act == "hard_swish":
+            x = F.hardswish(x)
+        return x
+
+
+class DPModule(nn.Layer):
+    """
+    Depth-wise and point-wise module.
+     Args:
+        in_channel (int): The input channels of this Module.
+        out_channel (int): The output channels of this Module.
+        kernel_size (int): The conv2d kernel size of this Module.
+        stride (int): The conv2d's stride of this Module.
+        act (str): The activation function of this Module,
+                   Now support `leaky_relu` and `hard_swish`.
+    """
+
+    def __init__(
+        self, in_channel=96, out_channel=96, kernel_size=3, stride=1, act="leaky_relu"
+    ):
+        super(DPModule, self).__init__()
+        initializer = nn.initializer.KaimingUniform()
+        self.act = act
+        self.dwconv = nn.Conv2D(
+            in_channels=in_channel,
+            out_channels=out_channel,
+            kernel_size=kernel_size,
+            groups=out_channel,
+            padding=(kernel_size - 1) // 2,
+            stride=stride,
+            weight_attr=ParamAttr(initializer=initializer),
+            bias_attr=False,
+        )
+        self.bn1 = nn.BatchNorm2D(out_channel)
+        self.pwconv = nn.Conv2D(
+            in_channels=out_channel,
+            out_channels=out_channel,
+            kernel_size=1,
+            groups=1,
+            padding=0,
+            weight_attr=ParamAttr(initializer=initializer),
+            bias_attr=False,
+        )
+        self.bn2 = nn.BatchNorm2D(out_channel)
+
+    def act_func(self, x):
+        if self.act == "leaky_relu":
+            x = F.leaky_relu(x)
+        elif self.act == "hard_swish":
+            x = F.hardswish(x)
+        return x
+
+    def forward(self, x):
+        x = self.act_func(self.bn1(self.dwconv(x)))
+        x = self.act_func(self.bn2(self.pwconv(x)))
+        return x
+
+
+class DarknetBottleneck(nn.Layer):
+    """The basic bottleneck block used in Darknet.
+    Each Block consists of two ConvModules and the input is added to the
+    final output. Each ConvModule is composed of Conv, BN, and act.
+    The first convLayer has filter size of 1x1 and the second one has the
+    filter size of 3x3.
+    Args:
+        in_channels (int): The input channels of this Module.
+        out_channels (int): The output channels of this Module.
+        expansion (int): The kernel size of the convolution. Default: 0.5
+        add_identity (bool): Whether to add identity to the out.
+            Default: True
+        use_depthwise (bool): Whether to use depthwise separable convolution.
+            Default: False
+    """
+
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size=3,
+        expansion=0.5,
+        add_identity=True,
+        use_depthwise=False,
+        act="leaky_relu",
+    ):
+        super(DarknetBottleneck, self).__init__()
+        hidden_channels = int(out_channels * expansion)
+        conv_func = DPModule if use_depthwise else ConvBNLayer
+        self.conv1 = ConvBNLayer(
+            in_channel=in_channels, out_channel=hidden_channels, kernel_size=1, act=act
+        )
+        self.conv2 = conv_func(
+            in_channel=hidden_channels,
+            out_channel=out_channels,
+            kernel_size=kernel_size,
+            stride=1,
+            act=act,
+        )
+        self.add_identity = add_identity and in_channels == out_channels
+
+    def forward(self, x):
+        identity = x
+        out = self.conv1(x)
+        out = self.conv2(out)
+
+        if self.add_identity:
+            return out + identity
+        else:
+            return out
+
+
+class CSPLayer(nn.Layer):
+    """Cross Stage Partial Layer.
+    Args:
+        in_channels (int): The input channels of the CSP layer.
+        out_channels (int): The output channels of the CSP layer.
+        expand_ratio (float): Ratio to adjust the number of channels of the
+            hidden layer. Default: 0.5
+        num_blocks (int): Number of blocks. Default: 1
+        add_identity (bool): Whether to add identity in blocks.
+            Default: True
+        use_depthwise (bool): Whether to depthwise separable convolution in
+            blocks. Default: False
+    """
+
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size=3,
+        expand_ratio=0.5,
+        num_blocks=1,
+        add_identity=True,
+        use_depthwise=False,
+        act="leaky_relu",
+    ):
+        super().__init__()
+        mid_channels = int(out_channels * expand_ratio)
+        self.main_conv = ConvBNLayer(in_channels, mid_channels, 1, act=act)
+        self.short_conv = ConvBNLayer(in_channels, mid_channels, 1, act=act)
+        self.final_conv = ConvBNLayer(2 * mid_channels, out_channels, 1, act=act)
+
+        self.blocks = nn.Sequential(
+            *[
+                DarknetBottleneck(
+                    mid_channels,
+                    mid_channels,
+                    kernel_size,
+                    1.0,
+                    add_identity,
+                    use_depthwise,
+                    act=act,
+                )
+                for _ in range(num_blocks)
+            ]
+        )
+
+    def forward(self, x):
+        x_short = self.short_conv(x)
+
+        x_main = self.main_conv(x)
+        x_main = self.blocks(x_main)
+
+        x_final = paddle.concat((x_main, x_short), axis=1)
+        return self.final_conv(x_final)
+
+
+class Channel_T(nn.Layer):
+    def __init__(self, in_channels=[116, 232, 464], out_channels=96, act="leaky_relu"):
+        super(Channel_T, self).__init__()
+        self.convs = nn.LayerList()
+        for i in range(len(in_channels)):
+            self.convs.append(ConvBNLayer(in_channels[i], out_channels, 1, act=act))
+
+    def forward(self, x):
+        outs = [self.convs[i](x[i]) for i in range(len(x))]
+        return outs
+
+
+class CSPPAN(nn.Layer):
+    """Path Aggregation Network with CSP module.
+    Args:
+        in_channels (List[int]): Number of input channels per scale.
+        out_channels (int): Number of output channels (used at each scale)
+        kernel_size (int): The conv2d kernel size of this Module.
+        num_csp_blocks (int): Number of bottlenecks in CSPLayer. Default: 1
+        use_depthwise (bool): Whether to depthwise separable convolution in
+            blocks. Default: True
+    """
+
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size=5,
+        num_csp_blocks=1,
+        use_depthwise=True,
+        act="hard_swish",
+    ):
+        super(CSPPAN, self).__init__()
+        self.in_channels = in_channels
+        self.out_channels = [out_channels] * len(in_channels)
+        conv_func = DPModule if use_depthwise else ConvBNLayer
+
+        self.conv_t = Channel_T(in_channels, out_channels, act=act)
+
+        # build top-down blocks
+        self.upsample = nn.Upsample(scale_factor=2, mode="nearest")
+        self.top_down_blocks = nn.LayerList()
+        for idx in range(len(in_channels) - 1, 0, -1):
+            self.top_down_blocks.append(
+                CSPLayer(
+                    out_channels * 2,
+                    out_channels,
+                    kernel_size=kernel_size,
+                    num_blocks=num_csp_blocks,
+                    add_identity=False,
+                    use_depthwise=use_depthwise,
+                    act=act,
+                )
+            )
+
+        # build bottom-up blocks
+        self.downsamples = nn.LayerList()
+        self.bottom_up_blocks = nn.LayerList()
+        for idx in range(len(in_channels) - 1):
+            self.downsamples.append(
+                conv_func(
+                    out_channels,
+                    out_channels,
+                    kernel_size=kernel_size,
+                    stride=2,
+                    act=act,
+                )
+            )
+            self.bottom_up_blocks.append(
+                CSPLayer(
+                    out_channels * 2,
+                    out_channels,
+                    kernel_size=kernel_size,
+                    num_blocks=num_csp_blocks,
+                    add_identity=False,
+                    use_depthwise=use_depthwise,
+                    act=act,
+                )
+            )
+
+    def forward(self, inputs):
+        """
+        Args:
+            inputs (tuple[Tensor]): input features.
+        Returns:
+            tuple[Tensor]: CSPPAN features.
+        """
+        assert len(inputs) == len(self.in_channels)
+        inputs = self.conv_t(inputs)
+
+        # top-down path
+        inner_outs = [inputs[-1]]
+        for idx in range(len(self.in_channels) - 1, 0, -1):
+            feat_heigh = inner_outs[0]
+            feat_low = inputs[idx - 1]
+            upsample_feat = F.upsample(
+                feat_heigh, size=feat_low.shape[2:4], mode="nearest"
+            )
+
+            inner_out = self.top_down_blocks[len(self.in_channels) - 1 - idx](
+                paddle.concat([upsample_feat, feat_low], 1)
+            )
+            inner_outs.insert(0, inner_out)
+
+        # bottom-up path
+        outs = [inner_outs[0]]
+        for idx in range(len(self.in_channels) - 1):
+            feat_low = outs[-1]
+            feat_height = inner_outs[idx + 1]
+            downsample_feat = self.downsamples[idx](feat_low)
+            out = self.bottom_up_blocks[idx](
+                paddle.concat([downsample_feat, feat_height], 1)
+            )
+            outs.append(out)
+
+        return tuple(outs)
diff --git a/ppocr/modeling/necks/ct_fpn.py b/ppocr/modeling/necks/ct_fpn.py
new file mode 100644
index 0000000..c7d5773
--- /dev/null
+++ b/ppocr/modeling/necks/ct_fpn.py
@@ -0,0 +1,188 @@
+# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+import os
+import sys
+
+import math
+from paddle.nn.initializer import TruncatedNormal, Constant, Normal
+
+ones_ = Constant(value=1.0)
+zeros_ = Constant(value=0.0)
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "../../..")))
+
+
+class Conv_BN_ReLU(nn.Layer):
+    def __init__(self, in_planes, out_planes, kernel_size=1, stride=1, padding=0):
+        super(Conv_BN_ReLU, self).__init__()
+        self.conv = nn.Conv2D(
+            in_planes,
+            out_planes,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            bias_attr=False,
+        )
+        self.bn = nn.BatchNorm2D(out_planes)
+        self.relu = nn.ReLU()
+
+        for m in self.sublayers():
+            if isinstance(m, nn.Conv2D):
+                n = m._kernel_size[0] * m._kernel_size[1] * m._out_channels
+                normal_ = Normal(mean=0.0, std=math.sqrt(2.0 / n))
+                normal_(m.weight)
+            elif isinstance(m, nn.BatchNorm2D):
+                zeros_(m.bias)
+                ones_(m.weight)
+
+    def forward(self, x):
+        return self.relu(self.bn(self.conv(x)))
+
+
+class FPEM(nn.Layer):
+    def __init__(self, in_channels, out_channels):
+        super(FPEM, self).__init__()
+        planes = out_channels
+        self.dwconv3_1 = nn.Conv2D(
+            planes,
+            planes,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            groups=planes,
+            bias_attr=False,
+        )
+        self.smooth_layer3_1 = Conv_BN_ReLU(planes, planes)
+
+        self.dwconv2_1 = nn.Conv2D(
+            planes,
+            planes,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            groups=planes,
+            bias_attr=False,
+        )
+        self.smooth_layer2_1 = Conv_BN_ReLU(planes, planes)
+
+        self.dwconv1_1 = nn.Conv2D(
+            planes,
+            planes,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            groups=planes,
+            bias_attr=False,
+        )
+        self.smooth_layer1_1 = Conv_BN_ReLU(planes, planes)
+
+        self.dwconv2_2 = nn.Conv2D(
+            planes,
+            planes,
+            kernel_size=3,
+            stride=2,
+            padding=1,
+            groups=planes,
+            bias_attr=False,
+        )
+        self.smooth_layer2_2 = Conv_BN_ReLU(planes, planes)
+
+        self.dwconv3_2 = nn.Conv2D(
+            planes,
+            planes,
+            kernel_size=3,
+            stride=2,
+            padding=1,
+            groups=planes,
+            bias_attr=False,
+        )
+        self.smooth_layer3_2 = Conv_BN_ReLU(planes, planes)
+
+        self.dwconv4_2 = nn.Conv2D(
+            planes,
+            planes,
+            kernel_size=3,
+            stride=2,
+            padding=1,
+            groups=planes,
+            bias_attr=False,
+        )
+        self.smooth_layer4_2 = Conv_BN_ReLU(planes, planes)
+
+    def _upsample_add(self, x, y):
+        return F.upsample(x, scale_factor=2, mode="bilinear") + y
+
+    def forward(self, f1, f2, f3, f4):
+        # up-down
+        f3 = self.smooth_layer3_1(self.dwconv3_1(self._upsample_add(f4, f3)))
+        f2 = self.smooth_layer2_1(self.dwconv2_1(self._upsample_add(f3, f2)))
+        f1 = self.smooth_layer1_1(self.dwconv1_1(self._upsample_add(f2, f1)))
+
+        # down-up
+        f2 = self.smooth_layer2_2(self.dwconv2_2(self._upsample_add(f2, f1)))
+        f3 = self.smooth_layer3_2(self.dwconv3_2(self._upsample_add(f3, f2)))
+        f4 = self.smooth_layer4_2(self.dwconv4_2(self._upsample_add(f4, f3)))
+
+        return f1, f2, f3, f4
+
+
+class CTFPN(nn.Layer):
+    def __init__(self, in_channels, out_channel=128):
+        super(CTFPN, self).__init__()
+        self.out_channels = out_channel * 4
+
+        self.reduce_layer1 = Conv_BN_ReLU(in_channels[0], 128)
+        self.reduce_layer2 = Conv_BN_ReLU(in_channels[1], 128)
+        self.reduce_layer3 = Conv_BN_ReLU(in_channels[2], 128)
+        self.reduce_layer4 = Conv_BN_ReLU(in_channels[3], 128)
+
+        self.fpem1 = FPEM(in_channels=(64, 128, 256, 512), out_channels=128)
+        self.fpem2 = FPEM(in_channels=(64, 128, 256, 512), out_channels=128)
+
+    def _upsample(self, x, scale=1):
+        return F.upsample(x, scale_factor=scale, mode="bilinear")
+
+    def forward(self, f):
+        # # reduce channel
+        f1 = self.reduce_layer1(f[0])  # N,64,160,160    --> N, 128, 160, 160
+        f2 = self.reduce_layer2(f[1])  # N, 128, 80, 80  --> N, 128, 80, 80
+        f3 = self.reduce_layer3(f[2])  # N, 256, 40, 40  --> N, 128, 40, 40
+        f4 = self.reduce_layer4(f[3])  # N, 512, 20, 20  --> N, 128, 20, 20
+
+        # FPEM
+        f1_1, f2_1, f3_1, f4_1 = self.fpem1(f1, f2, f3, f4)
+        f1_2, f2_2, f3_2, f4_2 = self.fpem2(f1_1, f2_1, f3_1, f4_1)
+
+        # FFM
+        f1 = f1_1 + f1_2
+        f2 = f2_1 + f2_2
+        f3 = f3_1 + f3_2
+        f4 = f4_1 + f4_2
+
+        f2 = self._upsample(f2, scale=2)
+        f3 = self._upsample(f3, scale=4)
+        f4 = self._upsample(f4, scale=8)
+        ff = paddle.concat((f1, f2, f3, f4), 1)  # N,512, 160,160
+        return ff
diff --git a/ppocr/modeling/necks/db_fpn.py b/ppocr/modeling/necks/db_fpn.py
new file mode 100644
index 0000000..f8fe496
--- /dev/null
+++ b/ppocr/modeling/necks/db_fpn.py
@@ -0,0 +1,1018 @@
+# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+import os
+import sys
+from ppocr.modeling.necks.intracl import IntraCLBlock
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "../../..")))
+
+from ppocr.modeling.backbones.det_mobilenet_v3 import SEModule
+
+
+class DSConv(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        padding,
+        stride=1,
+        groups=None,
+        if_act=True,
+        act="relu",
+        **kwargs,
+    ):
+        super(DSConv, self).__init__()
+        if groups == None:
+            groups = in_channels
+        self.if_act = if_act
+        self.act = act
+        self.conv1 = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=in_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            groups=groups,
+            bias_attr=False,
+        )
+
+        self.bn1 = nn.BatchNorm(num_channels=in_channels, act=None)
+
+        self.conv2 = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=int(in_channels * 4),
+            kernel_size=1,
+            stride=1,
+            bias_attr=False,
+        )
+
+        self.bn2 = nn.BatchNorm(num_channels=int(in_channels * 4), act=None)
+
+        self.conv3 = nn.Conv2D(
+            in_channels=int(in_channels * 4),
+            out_channels=out_channels,
+            kernel_size=1,
+            stride=1,
+            bias_attr=False,
+        )
+        self._c = [in_channels, out_channels]
+        if in_channels != out_channels:
+            self.conv_end = nn.Conv2D(
+                in_channels=in_channels,
+                out_channels=out_channels,
+                kernel_size=1,
+                stride=1,
+                bias_attr=False,
+            )
+
+    def forward(self, inputs):
+        x = self.conv1(inputs)
+        x = self.bn1(x)
+
+        x = self.conv2(x)
+        x = self.bn2(x)
+        if self.if_act:
+            if self.act == "relu":
+                x = F.relu(x)
+            elif self.act == "hardswish":
+                x = F.hardswish(x)
+            else:
+                print(
+                    "The activation function({}) is selected incorrectly.".format(
+                        self.act
+                    )
+                )
+                exit()
+
+        x = self.conv3(x)
+        if self._c[0] != self._c[1]:
+            x = x + self.conv_end(inputs)
+        return x
+
+
+class DBFPN(nn.Layer):
+    def __init__(self, in_channels, out_channels, use_asf=False, **kwargs):
+        super(DBFPN, self).__init__()
+        self.out_channels = out_channels
+        self.use_asf = use_asf
+        weight_attr = paddle.nn.initializer.KaimingUniform()
+
+        self.in2_conv = nn.Conv2D(
+            in_channels=in_channels[0],
+            out_channels=self.out_channels,
+            kernel_size=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+        self.in3_conv = nn.Conv2D(
+            in_channels=in_channels[1],
+            out_channels=self.out_channels,
+            kernel_size=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+        self.in4_conv = nn.Conv2D(
+            in_channels=in_channels[2],
+            out_channels=self.out_channels,
+            kernel_size=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+        self.in5_conv = nn.Conv2D(
+            in_channels=in_channels[3],
+            out_channels=self.out_channels,
+            kernel_size=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+        self.p5_conv = nn.Conv2D(
+            in_channels=self.out_channels,
+            out_channels=self.out_channels // 4,
+            kernel_size=3,
+            padding=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+        self.p4_conv = nn.Conv2D(
+            in_channels=self.out_channels,
+            out_channels=self.out_channels // 4,
+            kernel_size=3,
+            padding=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+        self.p3_conv = nn.Conv2D(
+            in_channels=self.out_channels,
+            out_channels=self.out_channels // 4,
+            kernel_size=3,
+            padding=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+        self.p2_conv = nn.Conv2D(
+            in_channels=self.out_channels,
+            out_channels=self.out_channels // 4,
+            kernel_size=3,
+            padding=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+
+        if self.use_asf is True:
+            self.asf = ASFBlock(self.out_channels, self.out_channels // 4)
+
+    def forward(self, x):
+        c2, c3, c4, c5 = x
+
+        in5 = self.in5_conv(c5)
+        in4 = self.in4_conv(c4)
+        in3 = self.in3_conv(c3)
+        in2 = self.in2_conv(c2)
+
+        out4 = in4 + F.upsample(
+            in5, scale_factor=2, mode="nearest", align_mode=1
+        )  # 1/16
+        out3 = in3 + F.upsample(
+            out4, scale_factor=2, mode="nearest", align_mode=1
+        )  # 1/8
+        out2 = in2 + F.upsample(
+            out3, scale_factor=2, mode="nearest", align_mode=1
+        )  # 1/4
+
+        p5 = self.p5_conv(in5)
+        p4 = self.p4_conv(out4)
+        p3 = self.p3_conv(out3)
+        p2 = self.p2_conv(out2)
+        p5 = F.upsample(p5, scale_factor=8, mode="nearest", align_mode=1)
+        p4 = F.upsample(p4, scale_factor=4, mode="nearest", align_mode=1)
+        p3 = F.upsample(p3, scale_factor=2, mode="nearest", align_mode=1)
+
+        fuse = paddle.concat([p5, p4, p3, p2], axis=1)
+
+        if self.use_asf is True:
+            fuse = self.asf(fuse, [p5, p4, p3, p2])
+
+        return fuse
+
+
+class RSELayer(nn.Layer):
+    def __init__(self, in_channels, out_channels, kernel_size, shortcut=True):
+        super(RSELayer, self).__init__()
+        weight_attr = paddle.nn.initializer.KaimingUniform()
+        self.out_channels = out_channels
+        self.in_conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=self.out_channels,
+            kernel_size=kernel_size,
+            padding=int(kernel_size // 2),
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+        self.se_block = SEModule(self.out_channels)
+        self.shortcut = shortcut
+
+    def forward(self, ins):
+        x = self.in_conv(ins)
+        if self.shortcut:
+            out = x + self.se_block(x)
+        else:
+            out = self.se_block(x)
+        return out
+
+
+class RSEFPN(nn.Layer):
+    def __init__(self, in_channels, out_channels, shortcut=True, **kwargs):
+        super(RSEFPN, self).__init__()
+        self.out_channels = out_channels
+        self.ins_conv = nn.LayerList()
+        self.inp_conv = nn.LayerList()
+        self.intracl = False
+        if "intracl" in kwargs.keys() and kwargs["intracl"] is True:
+            self.intracl = kwargs["intracl"]
+            self.incl1 = IntraCLBlock(self.out_channels // 4, reduce_factor=2)
+            self.incl2 = IntraCLBlock(self.out_channels // 4, reduce_factor=2)
+            self.incl3 = IntraCLBlock(self.out_channels // 4, reduce_factor=2)
+            self.incl4 = IntraCLBlock(self.out_channels // 4, reduce_factor=2)
+
+        for i in range(len(in_channels)):
+            self.ins_conv.append(
+                RSELayer(in_channels[i], out_channels, kernel_size=1, shortcut=shortcut)
+            )
+            self.inp_conv.append(
+                RSELayer(
+                    out_channels, out_channels // 4, kernel_size=3, shortcut=shortcut
+                )
+            )
+
+    def forward(self, x):
+        c2, c3, c4, c5 = x
+
+        in5 = self.ins_conv[3](c5)
+        in4 = self.ins_conv[2](c4)
+        in3 = self.ins_conv[1](c3)
+        in2 = self.ins_conv[0](c2)
+
+        out4 = in4 + F.upsample(
+            in5, scale_factor=2, mode="nearest", align_mode=1
+        )  # 1/16
+        out3 = in3 + F.upsample(
+            out4, scale_factor=2, mode="nearest", align_mode=1
+        )  # 1/8
+        out2 = in2 + F.upsample(
+            out3, scale_factor=2, mode="nearest", align_mode=1
+        )  # 1/4
+
+        p5 = self.inp_conv[3](in5)
+        p4 = self.inp_conv[2](out4)
+        p3 = self.inp_conv[1](out3)
+        p2 = self.inp_conv[0](out2)
+
+        if self.intracl is True:
+            p5 = self.incl4(p5)
+            p4 = self.incl3(p4)
+            p3 = self.incl2(p3)
+            p2 = self.incl1(p2)
+
+        p5 = F.upsample(p5, scale_factor=8, mode="nearest", align_mode=1)
+        p4 = F.upsample(p4, scale_factor=4, mode="nearest", align_mode=1)
+        p3 = F.upsample(p3, scale_factor=2, mode="nearest", align_mode=1)
+
+        fuse = paddle.concat([p5, p4, p3, p2], axis=1)
+        return fuse
+
+
+class RepLKFPN(nn.Layer):
+    """Optimized RSEFPN: replaces 3x3 standard Conv in inp_conv with
+    DilatedReparamBlock (DW, 5x5) + PWConv 1x1 + SE.
+
+    Uses the existing DilatedReparamBlock from UniRepLKNet to provide
+    multi-branch dilated training with single-conv inference.
+
+    Changes vs RSEFPN:
+      - inp_conv: RSELayer(3x3 std Conv + SE)
+                → DilatedReparamBlock(5x5 DW) + PWConv(1x1) + SE
+      - ins_conv: unchanged (1x1 Conv, no benefit from DW decomposition)
+
+    Parameter comparison (out_channels=96, 4 levels):
+      RSEFPN  inp_conv: 4 × (96×24×9 + SE) = 4 × 21,054 = 84,216
+      RepLKFPN inp_conv: 4 × (DilReparam96 + 96×24 + SE)
+             = 4 × (96×25 + 96×2 + 2×(96×9+96×2) + 96×24 + SE)
+             = 4 × (2,400 + 192 + 2×(864+192) + 2,304 + 318)
+             = 4 × 7,326 = 29,304
+      inp_conv reduction: 65.2%
+      Receptive field: 3×3 → 5×5 (with multi-dilation 3,5 patterns in training)
+
+    Inference: DilatedReparamBlock merges to single 5x5 DWConv, zero extra cost.
+    """
+
+    def __init__(
+        self, in_channels, out_channels, shortcut=True, dilated_kernel_size=7, **kwargs
+    ):
+        super(RepLKFPN, self).__init__()
+        self.out_channels = out_channels
+        self.is_repped = False
+        self.ins_conv = nn.LayerList()
+        self.inp_conv_dw = nn.LayerList()
+        self.inp_conv_pw = nn.LayerList()
+        self.inp_conv_se = nn.LayerList()
+        self.shortcut = shortcut
+
+        self.intracl = False
+        if "intracl" in kwargs.keys() and kwargs["intracl"] is True:
+            self.intracl = kwargs["intracl"]
+            self.incl1 = IntraCLBlock(self.out_channels // 4, reduce_factor=2)
+            self.incl2 = IntraCLBlock(self.out_channels // 4, reduce_factor=2)
+            self.incl3 = IntraCLBlock(self.out_channels // 4, reduce_factor=2)
+            self.incl4 = IntraCLBlock(self.out_channels // 4, reduce_factor=2)
+
+        weight_attr = paddle.nn.initializer.KaimingUniform()
+
+        for i in range(len(in_channels)):
+            self.ins_conv.append(
+                RSELayer(in_channels[i], out_channels, kernel_size=1, shortcut=shortcut)
+            )
+
+            self.inp_conv_dw.append(
+                DilatedReparamBlock(
+                    channels=out_channels, kernel_size=dilated_kernel_size
+                )
+            )
+
+            self.inp_conv_pw.append(
+                nn.Conv2D(
+                    in_channels=out_channels,
+                    out_channels=out_channels // 4,
+                    kernel_size=1,
+                    weight_attr=ParamAttr(initializer=weight_attr),
+                    bias_attr=False,
+                )
+            )
+
+            self.inp_conv_se.append(SEModule(out_channels // 4))
+
+    def _inp_forward(self, x, idx):
+        x = self.inp_conv_dw[idx](x)
+        x = self.inp_conv_pw[idx](x)
+        if self.shortcut:
+            x = x + self.inp_conv_se[idx](x)
+        else:
+            x = self.inp_conv_se[idx](x)
+        return x
+
+    def forward(self, x):
+        c2, c3, c4, c5 = x
+
+        in5 = self.ins_conv[3](c5)
+        in4 = self.ins_conv[2](c4)
+        in3 = self.ins_conv[1](c3)
+        in2 = self.ins_conv[0](c2)
+
+        out4 = in4 + F.upsample(
+            in5, scale_factor=2, mode="nearest", align_mode=1
+        )  # 1/16
+        out3 = in3 + F.upsample(
+            out4, scale_factor=2, mode="nearest", align_mode=1
+        )  # 1/8
+        out2 = in2 + F.upsample(
+            out3, scale_factor=2, mode="nearest", align_mode=1
+        )  # 1/4
+
+        p5 = self._inp_forward(in5, 3)
+        p4 = self._inp_forward(out4, 2)
+        p3 = self._inp_forward(out3, 1)
+        p2 = self._inp_forward(out2, 0)
+
+        if self.intracl is True:
+            p5 = self.incl4(p5)
+            p4 = self.incl3(p4)
+            p3 = self.incl2(p3)
+            p2 = self.incl1(p2)
+
+        p5 = F.upsample(p5, scale_factor=8, mode="nearest", align_mode=1)
+        p4 = F.upsample(p4, scale_factor=4, mode="nearest", align_mode=1)
+        p3 = F.upsample(p3, scale_factor=2, mode="nearest", align_mode=1)
+
+        fuse = paddle.concat([p5, p4, p3, p2], axis=1)
+        if self.training:
+            return {"fuse": fuse, "aux_p4": out4, "aux_p3": out3, "aux_p2": out2}
+        return fuse
+
+    def rep(self):
+        """Merge DilatedReparamBlock branches for inference deployment."""
+        if self.is_repped:
+            return
+        for i in range(len(self.inp_conv_dw)):
+            self.inp_conv_dw[i].rep()
+        self.is_repped = True
+
+
+class LKPAN(nn.Layer):
+    def __init__(self, in_channels, out_channels, mode="large", **kwargs):
+        super(LKPAN, self).__init__()
+        self.out_channels = out_channels
+        weight_attr = paddle.nn.initializer.KaimingUniform()
+
+        self.ins_conv = nn.LayerList()
+        self.inp_conv = nn.LayerList()
+        # pan head
+        self.pan_head_conv = nn.LayerList()
+        self.pan_lat_conv = nn.LayerList()
+
+        if mode.lower() == "lite":
+            p_layer = DSConv
+        elif mode.lower() == "large":
+            p_layer = nn.Conv2D
+        else:
+            raise ValueError(
+                "mode can only be one of ['lite', 'large'], but received {}".format(
+                    mode
+                )
+            )
+
+        for i in range(len(in_channels)):
+            self.ins_conv.append(
+                nn.Conv2D(
+                    in_channels=in_channels[i],
+                    out_channels=self.out_channels,
+                    kernel_size=1,
+                    weight_attr=ParamAttr(initializer=weight_attr),
+                    bias_attr=False,
+                )
+            )
+
+            self.inp_conv.append(
+                p_layer(
+                    in_channels=self.out_channels,
+                    out_channels=self.out_channels // 4,
+                    kernel_size=9,
+                    padding=4,
+                    weight_attr=ParamAttr(initializer=weight_attr),
+                    bias_attr=False,
+                )
+            )
+
+            if i > 0:
+                self.pan_head_conv.append(
+                    nn.Conv2D(
+                        in_channels=self.out_channels // 4,
+                        out_channels=self.out_channels // 4,
+                        kernel_size=3,
+                        padding=1,
+                        stride=2,
+                        weight_attr=ParamAttr(initializer=weight_attr),
+                        bias_attr=False,
+                    )
+                )
+            self.pan_lat_conv.append(
+                p_layer(
+                    in_channels=self.out_channels // 4,
+                    out_channels=self.out_channels // 4,
+                    kernel_size=9,
+                    padding=4,
+                    weight_attr=ParamAttr(initializer=weight_attr),
+                    bias_attr=False,
+                )
+            )
+
+        self.intracl = False
+        if "intracl" in kwargs.keys() and kwargs["intracl"] is True:
+            self.intracl = kwargs["intracl"]
+            self.incl1 = IntraCLBlock(self.out_channels // 4, reduce_factor=2)
+            self.incl2 = IntraCLBlock(self.out_channels // 4, reduce_factor=2)
+            self.incl3 = IntraCLBlock(self.out_channels // 4, reduce_factor=2)
+            self.incl4 = IntraCLBlock(self.out_channels // 4, reduce_factor=2)
+
+    def forward(self, x):
+        c2, c3, c4, c5 = x
+
+        in5 = self.ins_conv[3](c5)
+        in4 = self.ins_conv[2](c4)
+        in3 = self.ins_conv[1](c3)
+        in2 = self.ins_conv[0](c2)
+
+        out4 = in4 + F.upsample(
+            in5, scale_factor=2, mode="nearest", align_mode=1
+        )  # 1/16
+        out3 = in3 + F.upsample(
+            out4, scale_factor=2, mode="nearest", align_mode=1
+        )  # 1/8
+        out2 = in2 + F.upsample(
+            out3, scale_factor=2, mode="nearest", align_mode=1
+        )  # 1/4
+
+        f5 = self.inp_conv[3](in5)
+        f4 = self.inp_conv[2](out4)
+        f3 = self.inp_conv[1](out3)
+        f2 = self.inp_conv[0](out2)
+
+        pan3 = f3 + self.pan_head_conv[0](f2)
+        pan4 = f4 + self.pan_head_conv[1](pan3)
+        pan5 = f5 + self.pan_head_conv[2](pan4)
+
+        p2 = self.pan_lat_conv[0](f2)
+        p3 = self.pan_lat_conv[1](pan3)
+        p4 = self.pan_lat_conv[2](pan4)
+        p5 = self.pan_lat_conv[3](pan5)
+
+        if self.intracl is True:
+            p5 = self.incl4(p5)
+            p4 = self.incl3(p4)
+            p3 = self.incl2(p3)
+            p2 = self.incl1(p2)
+
+        p5 = F.upsample(p5, scale_factor=8, mode="nearest", align_mode=1)
+        p4 = F.upsample(p4, scale_factor=4, mode="nearest", align_mode=1)
+        p3 = F.upsample(p3, scale_factor=2, mode="nearest", align_mode=1)
+
+        fuse = paddle.concat([p5, p4, p3, p2], axis=1)
+        return fuse
+
+
+class DilatedReparamBlock(nn.Layer):
+    """
+    Dilated Reparam Block from UniRepLKNet.
+    Reference: https://github.com/AILab-CVC/UniRepLKNet
+
+    Training: uses multiple parallel dilated depthwise convolutions + BN
+    Inference: all branches merge into a single large-kernel depthwise conv
+
+    For kernel_size=9, the branches are:
+      - origin: 9x9 DW Conv (dil=1)
+      - branch1: 5x5 DW Conv (dil=1, equiv RF=5)
+      - branch2: 5x5 DW Conv (dil=2, equiv RF=9)
+      - branch3: 3x3 DW Conv (dil=3, equiv RF=7)
+      - branch4: 3x3 DW Conv (dil=4, equiv RF=9)
+    """
+
+    def __init__(self, channels, kernel_size=9, deploy=False):
+        super(DilatedReparamBlock, self).__init__()
+        self.channels = channels
+        self.kernel_size = kernel_size
+        self.is_repped = deploy
+
+        if kernel_size == 9:
+            self.kernel_sizes = [5, 5, 3, 3]
+            self.dilates = [1, 2, 3, 4]
+        elif kernel_size == 7:
+            self.kernel_sizes = [5, 3, 3]
+            self.dilates = [1, 2, 3]
+        elif kernel_size == 5:
+            self.kernel_sizes = [3, 3]
+            self.dilates = [1, 2]
+        elif kernel_size == 11:
+            self.kernel_sizes = [5, 5, 3, 3, 3]
+            self.dilates = [1, 2, 3, 4, 5]
+        elif kernel_size == 13:
+            self.kernel_sizes = [5, 7, 3, 3, 3]
+            self.dilates = [1, 2, 3, 4, 5]
+        else:
+            raise ValueError(
+                "DilatedReparamBlock requires kernel_size in [5,7,9,11,13], "
+                "but got {}".format(kernel_size)
+            )
+
+        if not self.is_repped:
+            self.lk_origin = nn.Conv2D(
+                in_channels=channels,
+                out_channels=channels,
+                kernel_size=kernel_size,
+                stride=1,
+                padding=kernel_size // 2,
+                groups=channels,
+                bias_attr=False,
+            )
+            self.origin_bn = nn.BatchNorm2D(channels)
+
+            for k, r in zip(self.kernel_sizes, self.dilates):
+                equiv_ks = r * (k - 1) + 1
+                p = equiv_ks // 2
+                conv = nn.Conv2D(
+                    in_channels=channels,
+                    out_channels=channels,
+                    kernel_size=k,
+                    stride=1,
+                    padding=p,
+                    dilation=r,
+                    groups=channels,
+                    bias_attr=False,
+                )
+                bn = nn.BatchNorm2D(channels)
+                setattr(self, "dil_conv_k{}_{}".format(k, r), conv)
+                setattr(self, "dil_bn_k{}_{}".format(k, r), bn)
+        else:
+            self.lk_origin = nn.Conv2D(
+                in_channels=channels,
+                out_channels=channels,
+                kernel_size=kernel_size,
+                stride=1,
+                padding=kernel_size // 2,
+                groups=channels,
+                bias_attr=True,
+            )
+
+    def forward(self, x):
+        if self.is_repped:
+            return self.lk_origin(x)
+        out = self.origin_bn(self.lk_origin(x))
+        for k, r in zip(self.kernel_sizes, self.dilates):
+            conv = getattr(self, "dil_conv_k{}_{}".format(k, r))
+            bn = getattr(self, "dil_bn_k{}_{}".format(k, r))
+            out = out + bn(conv(x))
+        return out
+
+    @staticmethod
+    def _fuse_bn(conv, bn):
+        """Fuse Conv2D + BatchNorm2D into a single Conv2D (weight, bias)."""
+        kernel = conv.weight
+        gamma = bn.weight
+        beta = bn.bias
+        running_mean = bn._mean
+        running_var = bn._variance
+        eps = bn._epsilon
+        std = paddle.sqrt(running_var + eps)
+        fused_weight = kernel * (gamma / std).reshape([-1, 1, 1, 1])
+        fused_bias = beta - running_mean * gamma / std
+        return fused_weight, fused_bias
+
+    @staticmethod
+    def _convert_dilated_to_nondilated(kernel, dilate_rate):
+        """Convert dilated conv kernel to equivalent non-dilated (sparse) kernel
+        by inserting zeros between kernel elements using transposed convolution."""
+        if dilate_rate == 1:
+            return kernel
+        identity = paddle.ones(shape=[1, 1, 1, 1], dtype=kernel.dtype)
+        # F.conv2d_transpose with stride=dilate_rate inserts zeros
+        # Process each channel independently
+        C = kernel.shape[0]
+        result_list = []
+        for i in range(C):
+            k_i = kernel[i : i + 1]  # (1, 1, kH, kW)
+            dilated = F.conv2d_transpose(k_i, identity, stride=dilate_rate)
+            result_list.append(dilated)
+        return paddle.concat(result_list, axis=0)
+
+    @staticmethod
+    def _merge_dilated_into_large_kernel(large_kernel, dilated_kernel, dilated_r):
+        """Pad dilated equivalent kernel to large kernel size and add."""
+        large_k = large_kernel.shape[2]
+        dilated_k = dilated_kernel.shape[2]
+        equiv_ks = dilated_r * (dilated_k - 1) + 1
+        equiv_kernel = DilatedReparamBlock._convert_dilated_to_nondilated(
+            dilated_kernel, dilated_r
+        )
+        rows_to_pad = large_k // 2 - equiv_ks // 2
+        if rows_to_pad > 0:
+            merged = large_kernel + F.pad(
+                equiv_kernel, [rows_to_pad, rows_to_pad, rows_to_pad, rows_to_pad]
+            )
+        else:
+            merged = large_kernel + equiv_kernel
+        return merged
+
+    @paddle.no_grad()
+    def rep(self):
+        """Merge all parallel branches into a single large-kernel DW conv.
+        Call this before switching to deploy/inference mode."""
+        if self.is_repped:
+            return
+        origin_k, origin_b = self._fuse_bn(self.lk_origin, self.origin_bn)
+        for k, r in zip(self.kernel_sizes, self.dilates):
+            conv = getattr(self, "dil_conv_k{}_{}".format(k, r))
+            bn = getattr(self, "dil_bn_k{}_{}".format(k, r))
+            branch_k, branch_b = self._fuse_bn(conv, bn)
+            origin_k = self._merge_dilated_into_large_kernel(origin_k, branch_k, r)
+            origin_b = origin_b + branch_b
+
+        merged_conv = nn.Conv2D(
+            in_channels=self.channels,
+            out_channels=self.channels,
+            kernel_size=self.kernel_size,
+            stride=1,
+            padding=self.kernel_size // 2,
+            groups=self.channels,
+            bias_attr=True,
+        )
+        merged_conv.weight.set_value(origin_k)
+        merged_conv.bias.set_value(origin_b)
+        self.lk_origin = merged_conv
+        self.is_repped = True
+
+        delattr(self, "origin_bn")
+        for k, r in zip(self.kernel_sizes, self.dilates):
+            delattr(self, "dil_conv_k{}_{}".format(k, r))
+            delattr(self, "dil_bn_k{}_{}".format(k, r))
+
+
+class DilatedReparamConv(nn.Layer):
+    """
+    A drop-in replacement for standard Conv2D (in_ch → out_ch, large kernel)
+    using DilatedReparamBlock (depthwise) + 1x1 pointwise convolution.
+
+    Architecture:
+      input(in_ch) → DilatedReparamBlock(in_ch, DW, kernel_size) → 1x1 Conv(in_ch→out_ch) → BN
+
+    This decomposition replaces a single large standard conv with DW + PW,
+    drastically reducing parameters while maintaining the large receptive field.
+    """
+
+    def __init__(
+        self, in_channels, out_channels, kernel_size=9, deploy=False, **kwargs
+    ):
+        super(DilatedReparamConv, self).__init__()
+        self.is_repped = False
+        weight_attr = paddle.nn.initializer.KaimingUniform()
+        self.dw = DilatedReparamBlock(
+            channels=in_channels, kernel_size=kernel_size, deploy=deploy
+        )
+        self.pw = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+        self.bn = nn.BatchNorm2D(out_channels)
+
+    def forward(self, x):
+        x = self.dw(x)
+        x = self.pw(x)
+        if not self.is_repped:
+            x = self.bn(x)
+        return x
+
+    @paddle.no_grad()
+    def rep(self):
+        """Fuse DW branches + PW Conv + BN for deployment."""
+        if self.is_repped:
+            return
+        self.dw.rep()
+        # Fuse pw(Conv2D, no bias) + bn(BatchNorm2D) into single Conv2D with bias
+        conv, bn = self.pw, self.bn
+        gamma = bn.weight
+        std = paddle.sqrt(bn._variance + bn._epsilon)
+        scale = gamma / std
+        w = conv.weight * scale[:, None, None, None]
+        b = bn.bias - bn._mean * scale
+        fused = nn.Conv2D(
+            conv._in_channels,
+            conv._out_channels,
+            conv._kernel_size,
+            stride=conv._stride,
+            padding=conv._padding,
+            dilation=conv._dilation,
+            groups=conv._groups,
+        )
+        fused.weight.set_value(w)
+        fused.bias.set_value(b)
+        self.pw = fused
+        del self.bn
+        self.is_repped = True
+
+
+class RepLKPAN(nn.Layer):
+    """
+    Optimized LKPAN using UniRepLKNet's DilatedReparamBlock.
+
+    Replaces the 8 standard 9x9 Conv2D in LKPAN (4 inp_conv + 4 pan_lat_conv)
+    with DilatedReparamConv (DW large-kernel reparam + 1x1 pointwise).
+
+    Parameter comparison (out_channels=256, i.e. inner_ch=64):
+    ┌──────────────────────────────────────────────────────────────────────────┐
+    │ Component        │ Original (Std 9x9 Conv)     │ UniRepLK (DW+PW)       │
+    ├──────────────────────────────────────────────────────────────────────────┤
+    │ inp_conv[i]:     │ 256×64×9×9 = 1,327,104      │ DW: 256×1×9×9 = 20,736 │
+    │ (256→64, ×4)     │                              │  +4 dilated DW branches │
+    │                  │                              │  ≈ 256×(25+25+9+9)     │
+    │                  │                              │  = 17,408 (training)    │
+    │                  │                              │ PW: 256×64×1×1 = 16,384 │
+    │                  │                              │ BN: 64×2 = 128          │
+    │                  │                              │ Subtotal/layer ≈ 54,656 │
+    │                  │                              │ (vs 1,327,104 original) │
+    ├──────────────────────────────────────────────────────────────────────────┤
+    │ pan_lat_conv[i]: │ 64×64×9×9 = 331,776         │ DW: 64×1×9×9 = 5,184   │
+    │ (64→64, ×4)      │                              │  +4 dilated DW branches │
+    │                  │                              │  ≈ 64×(25+25+9+9)      │
+    │                  │                              │  = 4,352 (training)     │
+    │                  │                              │ PW: 64×64×1×1 = 4,096   │
+    │                  │                              │ BN: 64×2 = 128          │
+    │                  │                              │ Subtotal/layer ≈ 13,760 │
+    │                  │                              │ (vs 331,776 original)   │
+    ├──────────────────────────────────────────────────────────────────────────┤
+    │ Total 9x9 params │ 4×1,327,104 + 4×331,776     │ 4×54,656 + 4×13,760    │
+    │ (8 layers)       │ = 6,635,520                  │ = 273,664               │
+    │                  │                              │ 95.9% reduction         │
+    ├──────────────────────────────────────────────────────────────────────────┤
+    │ Inference(reparam)│ Same as above                │ DW 9x9 merged (no extra │
+    │                  │                              │ branch overhead)+PW 1x1 │
+    │                  │                              │ FLOPs also greatly reduced│
+    └──────────────────────────────────────────────────────────────────────────┘
+
+    Note: BN params in dilated branches are only present during training and
+    merged into the DW conv weights at inference. The training param count
+    includes these BN params; inference param count is slightly lower.
+    """
+
+    def __init__(self, in_channels, out_channels, mode="large", **kwargs):
+        super(RepLKPAN, self).__init__()
+        self.out_channels = out_channels
+        self.is_repped = False
+        weight_attr = paddle.nn.initializer.KaimingUniform()
+
+        self.ins_conv = nn.LayerList()
+        self.inp_conv = nn.LayerList()
+        # pan head
+        self.pan_head_conv = nn.LayerList()
+        self.pan_lat_conv = nn.LayerList()
+
+        for i in range(len(in_channels)):
+            self.ins_conv.append(
+                nn.Conv2D(
+                    in_channels=in_channels[i],
+                    out_channels=self.out_channels,
+                    kernel_size=1,
+                    weight_attr=ParamAttr(initializer=weight_attr),
+                    bias_attr=False,
+                )
+            )
+
+            self.inp_conv.append(
+                DilatedReparamConv(
+                    in_channels=self.out_channels,
+                    out_channels=self.out_channels // 4,
+                    kernel_size=9,
+                )
+            )
+
+            if i > 0:
+                self.pan_head_conv.append(
+                    nn.Conv2D(
+                        in_channels=self.out_channels // 4,
+                        out_channels=self.out_channels // 4,
+                        kernel_size=3,
+                        padding=1,
+                        stride=2,
+                        weight_attr=ParamAttr(initializer=weight_attr),
+                        bias_attr=False,
+                    )
+                )
+            self.pan_lat_conv.append(
+                DilatedReparamConv(
+                    in_channels=self.out_channels // 4,
+                    out_channels=self.out_channels // 4,
+                    kernel_size=9,
+                )
+            )
+
+        self.intracl = False
+        if "intracl" in kwargs.keys() and kwargs["intracl"] is True:
+            self.intracl = kwargs["intracl"]
+            self.incl1 = IntraCLBlock(self.out_channels // 4, reduce_factor=2)
+            self.incl2 = IntraCLBlock(self.out_channels // 4, reduce_factor=2)
+            self.incl3 = IntraCLBlock(self.out_channels // 4, reduce_factor=2)
+            self.incl4 = IntraCLBlock(self.out_channels // 4, reduce_factor=2)
+
+    def forward(self, x):
+        c2, c3, c4, c5 = x
+
+        in5 = self.ins_conv[3](c5)
+        in4 = self.ins_conv[2](c4)
+        in3 = self.ins_conv[1](c3)
+        in2 = self.ins_conv[0](c2)
+
+        out4 = in4 + F.upsample(
+            in5, scale_factor=2, mode="nearest", align_mode=1
+        )  # 1/16
+        out3 = in3 + F.upsample(
+            out4, scale_factor=2, mode="nearest", align_mode=1
+        )  # 1/8
+        out2 = in2 + F.upsample(
+            out3, scale_factor=2, mode="nearest", align_mode=1
+        )  # 1/4
+
+        f5 = self.inp_conv[3](in5)
+        f4 = self.inp_conv[2](out4)
+        f3 = self.inp_conv[1](out3)
+        f2 = self.inp_conv[0](out2)
+
+        pan3 = f3 + self.pan_head_conv[0](f2)
+        pan4 = f4 + self.pan_head_conv[1](pan3)
+        pan5 = f5 + self.pan_head_conv[2](pan4)
+
+        p2 = self.pan_lat_conv[0](f2)
+        p3 = self.pan_lat_conv[1](pan3)
+        p4 = self.pan_lat_conv[2](pan4)
+        p5 = self.pan_lat_conv[3](pan5)
+
+        if self.intracl is True:
+            p5 = self.incl4(p5)
+            p4 = self.incl3(p4)
+            p3 = self.incl2(p3)
+            p2 = self.incl1(p2)
+
+        p5 = F.upsample(p5, scale_factor=8, mode="nearest", align_mode=1)
+        p4 = F.upsample(p4, scale_factor=4, mode="nearest", align_mode=1)
+        p3 = F.upsample(p3, scale_factor=2, mode="nearest", align_mode=1)
+
+        fuse = paddle.concat([p5, p4, p3, p2], axis=1)
+        if self.training:
+            return {"fuse": fuse, "aux_p4": out4, "aux_p3": out3, "aux_p2": out2}
+        return fuse
+
+    def rep(self):
+        """Merge all DilatedReparamBlock branches and fuse PW+BN for deployment."""
+        if self.is_repped:
+            return
+        for i in range(len(self.inp_conv)):
+            self.inp_conv[i].rep()
+        for i in range(len(self.pan_lat_conv)):
+            self.pan_lat_conv[i].rep()
+        self.is_repped = True
+
+
+class ASFBlock(nn.Layer):
+    """
+    This code is referred from:
+        https://github.com/MhLiao/DB/blob/master/decoders/feature_attention.py
+    """
+
+    def __init__(self, in_channels, inter_channels, out_features_num=4):
+        """
+        Adaptive Scale Fusion (ASF) block of DBNet++
+        Args:
+            in_channels: the number of channels in the input data
+            inter_channels: the number of middle channels
+            out_features_num: the number of fused stages
+        """
+        super(ASFBlock, self).__init__()
+        weight_attr = paddle.nn.initializer.KaimingUniform()
+        self.in_channels = in_channels
+        self.inter_channels = inter_channels
+        self.out_features_num = out_features_num
+        self.conv = nn.Conv2D(in_channels, inter_channels, 3, padding=1)
+
+        self.spatial_scale = nn.Sequential(
+            # Nx1xHxW
+            nn.Conv2D(
+                in_channels=1,
+                out_channels=1,
+                kernel_size=3,
+                bias_attr=False,
+                padding=1,
+                weight_attr=ParamAttr(initializer=weight_attr),
+            ),
+            nn.ReLU(),
+            nn.Conv2D(
+                in_channels=1,
+                out_channels=1,
+                kernel_size=1,
+                bias_attr=False,
+                weight_attr=ParamAttr(initializer=weight_attr),
+            ),
+            nn.Sigmoid(),
+        )
+
+        self.channel_scale = nn.Sequential(
+            nn.Conv2D(
+                in_channels=inter_channels,
+                out_channels=out_features_num,
+                kernel_size=1,
+                bias_attr=False,
+                weight_attr=ParamAttr(initializer=weight_attr),
+            ),
+            nn.Sigmoid(),
+        )
+
+    def forward(self, fuse_features, features_list):
+        fuse_features = self.conv(fuse_features)
+        spatial_x = paddle.mean(fuse_features, axis=1, keepdim=True)
+        attention_scores = self.spatial_scale(spatial_x) + fuse_features
+        attention_scores = self.channel_scale(attention_scores)
+        assert len(features_list) == self.out_features_num
+
+        out_list = []
+        for i in range(self.out_features_num):
+            out_list.append(attention_scores[:, i : i + 1] * features_list[i])
+        return paddle.concat(out_list, axis=1)
diff --git a/ppocr/modeling/necks/east_fpn.py b/ppocr/modeling/necks/east_fpn.py
new file mode 100644
index 0000000..1b7f50d
--- /dev/null
+++ b/ppocr/modeling/necks/east_fpn.py
@@ -0,0 +1,203 @@
+# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride,
+        padding,
+        groups=1,
+        if_act=True,
+        act=None,
+        name=None,
+    ):
+        super(ConvBNLayer, self).__init__()
+        self.if_act = if_act
+        self.act = act
+        self.conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            groups=groups,
+            weight_attr=ParamAttr(name=name + "_weights"),
+            bias_attr=False,
+        )
+
+        self.bn = nn.BatchNorm(
+            num_channels=out_channels,
+            act=act,
+            param_attr=ParamAttr(name="bn_" + name + "_scale"),
+            bias_attr=ParamAttr(name="bn_" + name + "_offset"),
+            moving_mean_name="bn_" + name + "_mean",
+            moving_variance_name="bn_" + name + "_variance",
+        )
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        return x
+
+
+class DeConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride,
+        padding,
+        groups=1,
+        if_act=True,
+        act=None,
+        name=None,
+    ):
+        super(DeConvBNLayer, self).__init__()
+        self.if_act = if_act
+        self.act = act
+        self.deconv = nn.Conv2DTranspose(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            groups=groups,
+            weight_attr=ParamAttr(name=name + "_weights"),
+            bias_attr=False,
+        )
+        self.bn = nn.BatchNorm(
+            num_channels=out_channels,
+            act=act,
+            param_attr=ParamAttr(name="bn_" + name + "_scale"),
+            bias_attr=ParamAttr(name="bn_" + name + "_offset"),
+            moving_mean_name="bn_" + name + "_mean",
+            moving_variance_name="bn_" + name + "_variance",
+        )
+
+    def forward(self, x):
+        x = self.deconv(x)
+        x = self.bn(x)
+        return x
+
+
+class EASTFPN(nn.Layer):
+    def __init__(self, in_channels, model_name, **kwargs):
+        super(EASTFPN, self).__init__()
+        self.model_name = model_name
+        if self.model_name == "large":
+            self.out_channels = 128
+        else:
+            self.out_channels = 64
+        self.in_channels = in_channels[::-1]
+        self.h1_conv = ConvBNLayer(
+            in_channels=self.out_channels + self.in_channels[1],
+            out_channels=self.out_channels,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            if_act=True,
+            act="relu",
+            name="unet_h_1",
+        )
+        self.h2_conv = ConvBNLayer(
+            in_channels=self.out_channels + self.in_channels[2],
+            out_channels=self.out_channels,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            if_act=True,
+            act="relu",
+            name="unet_h_2",
+        )
+        self.h3_conv = ConvBNLayer(
+            in_channels=self.out_channels + self.in_channels[3],
+            out_channels=self.out_channels,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            if_act=True,
+            act="relu",
+            name="unet_h_3",
+        )
+        self.g0_deconv = DeConvBNLayer(
+            in_channels=self.in_channels[0],
+            out_channels=self.out_channels,
+            kernel_size=4,
+            stride=2,
+            padding=1,
+            if_act=True,
+            act="relu",
+            name="unet_g_0",
+        )
+        self.g1_deconv = DeConvBNLayer(
+            in_channels=self.out_channels,
+            out_channels=self.out_channels,
+            kernel_size=4,
+            stride=2,
+            padding=1,
+            if_act=True,
+            act="relu",
+            name="unet_g_1",
+        )
+        self.g2_deconv = DeConvBNLayer(
+            in_channels=self.out_channels,
+            out_channels=self.out_channels,
+            kernel_size=4,
+            stride=2,
+            padding=1,
+            if_act=True,
+            act="relu",
+            name="unet_g_2",
+        )
+        self.g3_conv = ConvBNLayer(
+            in_channels=self.out_channels,
+            out_channels=self.out_channels,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            if_act=True,
+            act="relu",
+            name="unet_g_3",
+        )
+
+    def forward(self, x):
+        f = x[::-1]
+
+        h = f[0]
+        g = self.g0_deconv(h)
+        h = paddle.concat([g, f[1]], axis=1)
+        h = self.h1_conv(h)
+        g = self.g1_deconv(h)
+        h = paddle.concat([g, f[2]], axis=1)
+        h = self.h2_conv(h)
+        g = self.g2_deconv(h)
+        h = paddle.concat([g, f[3]], axis=1)
+        h = self.h3_conv(h)
+        g = self.g3_conv(h)
+
+        return g
diff --git a/ppocr/modeling/necks/fce_fpn.py b/ppocr/modeling/necks/fce_fpn.py
new file mode 100644
index 0000000..a456fd1
--- /dev/null
+++ b/ppocr/modeling/necks/fce_fpn.py
@@ -0,0 +1,304 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/PaddlePaddle/PaddleDetection/blob/release/2.3/ppdet/modeling/necks/fpn.py
+"""
+
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+from paddle.nn.initializer import XavierUniform
+from paddle.nn.initializer import Normal
+from paddle.regularizer import L2Decay
+
+__all__ = ["FCEFPN"]
+
+
+class ConvNormLayer(nn.Layer):
+    def __init__(
+        self,
+        ch_in,
+        ch_out,
+        filter_size,
+        stride,
+        groups=1,
+        norm_type="bn",
+        norm_decay=0.0,
+        norm_groups=32,
+        lr_scale=1.0,
+        freeze_norm=False,
+        initializer=Normal(mean=0.0, std=0.01),
+    ):
+        super(ConvNormLayer, self).__init__()
+        assert norm_type in ["bn", "sync_bn", "gn"]
+
+        bias_attr = False
+
+        self.conv = nn.Conv2D(
+            in_channels=ch_in,
+            out_channels=ch_out,
+            kernel_size=filter_size,
+            stride=stride,
+            padding=(filter_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(initializer=initializer, learning_rate=1.0),
+            bias_attr=bias_attr,
+        )
+
+        norm_lr = 0.0 if freeze_norm else 1.0
+        param_attr = ParamAttr(
+            learning_rate=norm_lr,
+            regularizer=L2Decay(norm_decay) if norm_decay is not None else None,
+        )
+        bias_attr = ParamAttr(
+            learning_rate=norm_lr,
+            regularizer=L2Decay(norm_decay) if norm_decay is not None else None,
+        )
+        if norm_type == "bn":
+            self.norm = nn.BatchNorm2D(
+                ch_out, weight_attr=param_attr, bias_attr=bias_attr
+            )
+        elif norm_type == "sync_bn":
+            self.norm = nn.SyncBatchNorm(
+                ch_out, weight_attr=param_attr, bias_attr=bias_attr
+            )
+        elif norm_type == "gn":
+            self.norm = nn.GroupNorm(
+                num_groups=norm_groups,
+                num_channels=ch_out,
+                weight_attr=param_attr,
+                bias_attr=bias_attr,
+            )
+
+    def forward(self, inputs):
+        out = self.conv(inputs)
+        out = self.norm(out)
+        return out
+
+
+class FCEFPN(nn.Layer):
+    """
+    Feature Pyramid Network, see https://arxiv.org/abs/1612.03144
+    Args:
+        in_channels (list[int]): input channels of each level which can be
+            derived from the output shape of backbone by from_config
+        out_channels (list[int]): output channel of each level
+        spatial_scales (list[float]): the spatial scales between input feature
+            maps and original input image which can be derived from the output
+            shape of backbone by from_config
+        has_extra_convs (bool): whether to add extra conv to the last level.
+            default False
+        extra_stage (int): the number of extra stages added to the last level.
+            default 1
+        use_c5 (bool): Whether to use c5 as the input of extra stage,
+            otherwise p5 is used. default True
+        norm_type (string|None): The normalization type in FPN module. If
+            norm_type is None, norm will not be used after conv and if
+            norm_type is string, bn, gn, sync_bn are available. default None
+        norm_decay (float): weight decay for normalization layer weights.
+            default 0.
+        freeze_norm (bool): whether to freeze normalization layer.
+            default False
+        relu_before_extra_convs (bool): whether to add relu before extra convs.
+            default False
+
+    """
+
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        spatial_scales=[0.25, 0.125, 0.0625, 0.03125],
+        has_extra_convs=False,
+        extra_stage=1,
+        use_c5=True,
+        norm_type=None,
+        norm_decay=0.0,
+        freeze_norm=False,
+        relu_before_extra_convs=True,
+    ):
+        super(FCEFPN, self).__init__()
+        self.out_channels = out_channels
+        for s in range(extra_stage):
+            spatial_scales = spatial_scales + [spatial_scales[-1] / 2.0]
+        self.spatial_scales = spatial_scales
+        self.has_extra_convs = has_extra_convs
+        self.extra_stage = extra_stage
+        self.use_c5 = use_c5
+        self.relu_before_extra_convs = relu_before_extra_convs
+        self.norm_type = norm_type
+        self.norm_decay = norm_decay
+        self.freeze_norm = freeze_norm
+
+        self.lateral_convs = []
+        self.fpn_convs = []
+        fan = out_channels * 3 * 3
+
+        # stage index 0,1,2,3 stands for res2,res3,res4,res5 on ResNet Backbone
+        # 0 <= st_stage < ed_stage <= 3
+        st_stage = 4 - len(in_channels)
+        ed_stage = st_stage + len(in_channels) - 1
+        for i in range(st_stage, ed_stage + 1):
+            if i == 3:
+                lateral_name = "fpn_inner_res5_sum"
+            else:
+                lateral_name = "fpn_inner_res{}_sum_lateral".format(i + 2)
+            in_c = in_channels[i - st_stage]
+            if self.norm_type is not None:
+                lateral = self.add_sublayer(
+                    lateral_name,
+                    ConvNormLayer(
+                        ch_in=in_c,
+                        ch_out=out_channels,
+                        filter_size=1,
+                        stride=1,
+                        norm_type=self.norm_type,
+                        norm_decay=self.norm_decay,
+                        freeze_norm=self.freeze_norm,
+                        initializer=XavierUniform(fan_out=in_c),
+                    ),
+                )
+            else:
+                lateral = self.add_sublayer(
+                    lateral_name,
+                    nn.Conv2D(
+                        in_channels=in_c,
+                        out_channels=out_channels,
+                        kernel_size=1,
+                        weight_attr=ParamAttr(initializer=XavierUniform(fan_out=in_c)),
+                    ),
+                )
+            self.lateral_convs.append(lateral)
+
+        for i in range(st_stage, ed_stage + 1):
+            fpn_name = "fpn_res{}_sum".format(i + 2)
+            if self.norm_type is not None:
+                fpn_conv = self.add_sublayer(
+                    fpn_name,
+                    ConvNormLayer(
+                        ch_in=out_channels,
+                        ch_out=out_channels,
+                        filter_size=3,
+                        stride=1,
+                        norm_type=self.norm_type,
+                        norm_decay=self.norm_decay,
+                        freeze_norm=self.freeze_norm,
+                        initializer=XavierUniform(fan_out=fan),
+                    ),
+                )
+            else:
+                fpn_conv = self.add_sublayer(
+                    fpn_name,
+                    nn.Conv2D(
+                        in_channels=out_channels,
+                        out_channels=out_channels,
+                        kernel_size=3,
+                        padding=1,
+                        weight_attr=ParamAttr(initializer=XavierUniform(fan_out=fan)),
+                    ),
+                )
+            self.fpn_convs.append(fpn_conv)
+
+        # add extra conv levels for RetinaNet(use_c5)/FCOS(use_p5)
+        if self.has_extra_convs:
+            for i in range(self.extra_stage):
+                lvl = ed_stage + 1 + i
+                if i == 0 and self.use_c5:
+                    in_c = in_channels[-1]
+                else:
+                    in_c = out_channels
+                extra_fpn_name = "fpn_{}".format(lvl + 2)
+                if self.norm_type is not None:
+                    extra_fpn_conv = self.add_sublayer(
+                        extra_fpn_name,
+                        ConvNormLayer(
+                            ch_in=in_c,
+                            ch_out=out_channels,
+                            filter_size=3,
+                            stride=2,
+                            norm_type=self.norm_type,
+                            norm_decay=self.norm_decay,
+                            freeze_norm=self.freeze_norm,
+                            initializer=XavierUniform(fan_out=fan),
+                        ),
+                    )
+                else:
+                    extra_fpn_conv = self.add_sublayer(
+                        extra_fpn_name,
+                        nn.Conv2D(
+                            in_channels=in_c,
+                            out_channels=out_channels,
+                            kernel_size=3,
+                            stride=2,
+                            padding=1,
+                            weight_attr=ParamAttr(
+                                initializer=XavierUniform(fan_out=fan)
+                            ),
+                        ),
+                    )
+                self.fpn_convs.append(extra_fpn_conv)
+
+    @classmethod
+    def from_config(cls, cfg, input_shape):
+        return {
+            "in_channels": [i.channels for i in input_shape],
+            "spatial_scales": [1.0 / i.stride for i in input_shape],
+        }
+
+    def forward(self, body_feats):
+        laterals = []
+        num_levels = len(body_feats)
+
+        for i in range(num_levels):
+            laterals.append(self.lateral_convs[i](body_feats[i]))
+
+        for i in range(1, num_levels):
+            lvl = num_levels - i
+            upsample = F.interpolate(
+                laterals[lvl],
+                scale_factor=2.0,
+                mode="nearest",
+            )
+            laterals[lvl - 1] += upsample
+
+        fpn_output = []
+        for lvl in range(num_levels):
+            fpn_output.append(self.fpn_convs[lvl](laterals[lvl]))
+
+        if self.extra_stage > 0:
+            # use max pool to get more levels on top of outputs (Faster R-CNN, Mask R-CNN)
+            if not self.has_extra_convs:
+                assert (
+                    self.extra_stage == 1
+                ), "extra_stage should be 1 if FPN has not extra convs"
+                fpn_output.append(F.max_pool2d(fpn_output[-1], 1, stride=2))
+            # add extra conv levels for RetinaNet(use_c5)/FCOS(use_p5)
+            else:
+                if self.use_c5:
+                    extra_source = body_feats[-1]
+                else:
+                    extra_source = fpn_output[-1]
+                fpn_output.append(self.fpn_convs[num_levels](extra_source))
+
+                for i in range(1, self.extra_stage):
+                    if self.relu_before_extra_convs:
+                        fpn_output.append(
+                            self.fpn_convs[num_levels + i](F.relu(fpn_output[-1]))
+                        )
+                    else:
+                        fpn_output.append(
+                            self.fpn_convs[num_levels + i](fpn_output[-1])
+                        )
+        return fpn_output
diff --git a/ppocr/modeling/necks/fpn.py b/ppocr/modeling/necks/fpn.py
new file mode 100644
index 0000000..ea5253c
--- /dev/null
+++ b/ppocr/modeling/necks/fpn.py
@@ -0,0 +1,149 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/whai362/PSENet/blob/python3/models/neck/fpn.py
+"""
+
+import paddle.nn as nn
+import paddle
+import math
+import paddle.nn.functional as F
+
+
+class Conv_BN_ReLU(nn.Layer):
+    def __init__(self, in_planes, out_planes, kernel_size=1, stride=1, padding=0):
+        super(Conv_BN_ReLU, self).__init__()
+        self.conv = nn.Conv2D(
+            in_planes,
+            out_planes,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            bias_attr=False,
+        )
+        self.bn = nn.BatchNorm2D(out_planes, momentum=0.1)
+        self.relu = nn.ReLU()
+
+        for m in self.sublayers():
+            if isinstance(m, nn.Conv2D):
+                n = m._kernel_size[0] * m._kernel_size[1] * m._out_channels
+                m.weight = paddle.create_parameter(
+                    shape=m.weight.shape,
+                    dtype="float32",
+                    default_initializer=paddle.nn.initializer.Normal(
+                        0, math.sqrt(2.0 / n)
+                    ),
+                )
+            elif isinstance(m, nn.BatchNorm2D):
+                m.weight = paddle.create_parameter(
+                    shape=m.weight.shape,
+                    dtype="float32",
+                    default_initializer=paddle.nn.initializer.Constant(1.0),
+                )
+                m.bias = paddle.create_parameter(
+                    shape=m.bias.shape,
+                    dtype="float32",
+                    default_initializer=paddle.nn.initializer.Constant(0.0),
+                )
+
+    def forward(self, x):
+        return self.relu(self.bn(self.conv(x)))
+
+
+class FPN(nn.Layer):
+    def __init__(self, in_channels, out_channels):
+        super(FPN, self).__init__()
+
+        # Top layer
+        self.toplayer_ = Conv_BN_ReLU(
+            in_channels[3], out_channels, kernel_size=1, stride=1, padding=0
+        )
+        # Lateral layers
+        self.latlayer1_ = Conv_BN_ReLU(
+            in_channels[2], out_channels, kernel_size=1, stride=1, padding=0
+        )
+
+        self.latlayer2_ = Conv_BN_ReLU(
+            in_channels[1], out_channels, kernel_size=1, stride=1, padding=0
+        )
+
+        self.latlayer3_ = Conv_BN_ReLU(
+            in_channels[0], out_channels, kernel_size=1, stride=1, padding=0
+        )
+
+        # Smooth layers
+        self.smooth1_ = Conv_BN_ReLU(
+            out_channels, out_channels, kernel_size=3, stride=1, padding=1
+        )
+
+        self.smooth2_ = Conv_BN_ReLU(
+            out_channels, out_channels, kernel_size=3, stride=1, padding=1
+        )
+
+        self.smooth3_ = Conv_BN_ReLU(
+            out_channels, out_channels, kernel_size=3, stride=1, padding=1
+        )
+
+        self.out_channels = out_channels * 4
+        for m in self.sublayers():
+            if isinstance(m, nn.Conv2D):
+                n = m._kernel_size[0] * m._kernel_size[1] * m._out_channels
+                m.weight = paddle.create_parameter(
+                    shape=m.weight.shape,
+                    dtype="float32",
+                    default_initializer=paddle.nn.initializer.Normal(
+                        0, math.sqrt(2.0 / n)
+                    ),
+                )
+            elif isinstance(m, nn.BatchNorm2D):
+                m.weight = paddle.create_parameter(
+                    shape=m.weight.shape,
+                    dtype="float32",
+                    default_initializer=paddle.nn.initializer.Constant(1.0),
+                )
+                m.bias = paddle.create_parameter(
+                    shape=m.bias.shape,
+                    dtype="float32",
+                    default_initializer=paddle.nn.initializer.Constant(0.0),
+                )
+
+    def _upsample(self, x, scale=1):
+        return F.upsample(x, scale_factor=scale, mode="bilinear")
+
+    def _upsample_add(self, x, y, scale=1):
+        return F.upsample(x, scale_factor=scale, mode="bilinear") + y
+
+    def forward(self, x):
+        f2, f3, f4, f5 = x
+        p5 = self.toplayer_(f5)
+
+        f4 = self.latlayer1_(f4)
+        p4 = self._upsample_add(p5, f4, 2)
+        p4 = self.smooth1_(p4)
+
+        f3 = self.latlayer2_(f3)
+        p3 = self._upsample_add(p4, f3, 2)
+        p3 = self.smooth2_(p3)
+
+        f2 = self.latlayer3_(f2)
+        p2 = self._upsample_add(p3, f2, 2)
+        p2 = self.smooth3_(p2)
+
+        p3 = self._upsample(p3, 2)
+        p4 = self._upsample(p4, 4)
+        p5 = self._upsample(p5, 8)
+
+        fuse = paddle.concat([p2, p3, p4, p5], axis=1)
+        return fuse
diff --git a/ppocr/modeling/necks/fpn_unet.py b/ppocr/modeling/necks/fpn_unet.py
new file mode 100644
index 0000000..e560cdb
--- /dev/null
+++ b/ppocr/modeling/necks/fpn_unet.py
@@ -0,0 +1,103 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textdet/necks/fpn_unet.py
+"""
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+
+
+class UpBlock(nn.Layer):
+    def __init__(self, in_channels, out_channels):
+        super().__init__()
+
+        assert isinstance(in_channels, int)
+        assert isinstance(out_channels, int)
+
+        self.conv1x1 = nn.Conv2D(
+            in_channels, in_channels, kernel_size=1, stride=1, padding=0
+        )
+        self.conv3x3 = nn.Conv2D(
+            in_channels, out_channels, kernel_size=3, stride=1, padding=1
+        )
+        self.deconv = nn.Conv2DTranspose(
+            out_channels, out_channels, kernel_size=4, stride=2, padding=1
+        )
+
+    def forward(self, x):
+        x = F.relu(self.conv1x1(x))
+        x = F.relu(self.conv3x3(x))
+        x = self.deconv(x)
+        return x
+
+
+class FPN_UNet(nn.Layer):
+    def __init__(self, in_channels, out_channels):
+        super().__init__()
+
+        assert len(in_channels) == 4
+        assert isinstance(out_channels, int)
+        self.out_channels = out_channels
+
+        blocks_out_channels = [out_channels] + [
+            min(out_channels * 2**i, 256) for i in range(4)
+        ]
+        blocks_in_channels = (
+            [blocks_out_channels[1]]
+            + [in_channels[i] + blocks_out_channels[i + 2] for i in range(3)]
+            + [in_channels[3]]
+        )
+
+        self.up4 = nn.Conv2DTranspose(
+            blocks_in_channels[4],
+            blocks_out_channels[4],
+            kernel_size=4,
+            stride=2,
+            padding=1,
+        )
+        self.up_block3 = UpBlock(blocks_in_channels[3], blocks_out_channels[3])
+        self.up_block2 = UpBlock(blocks_in_channels[2], blocks_out_channels[2])
+        self.up_block1 = UpBlock(blocks_in_channels[1], blocks_out_channels[1])
+        self.up_block0 = UpBlock(blocks_in_channels[0], blocks_out_channels[0])
+
+    def forward(self, x):
+        """
+        Args:
+            x (list[Tensor] | tuple[Tensor]): A list of four tensors of shape
+                :math:`(N, C_i, H_i, W_i)`, representing C2, C3, C4, C5
+                features respectively. :math:`C_i` should matches the number in
+                ``in_channels``.
+
+        Returns:
+            Tensor: Shape :math:`(N, C, H, W)` where :math:`H=4H_0` and
+            :math:`W=4W_0`.
+        """
+        c2, c3, c4, c5 = x
+
+        x = F.relu(self.up4(c5))
+
+        x = paddle.concat([x, c4], axis=1)
+        x = F.relu(self.up_block3(x))
+
+        x = paddle.concat([x, c3], axis=1)
+        x = F.relu(self.up_block2(x))
+
+        x = paddle.concat([x, c2], axis=1)
+        x = F.relu(self.up_block1(x))
+
+        x = self.up_block0(x)
+        return x
diff --git a/ppocr/modeling/necks/intracl.py b/ppocr/modeling/necks/intracl.py
new file mode 100644
index 0000000..2c4809c
--- /dev/null
+++ b/ppocr/modeling/necks/intracl.py
@@ -0,0 +1,121 @@
+import paddle
+from paddle import nn
+
+# refer from: https://github.com/ViTAE-Transformer/I3CL/blob/736c80237f66d352d488e83b05f3e33c55201317/mmdet/models/detectors/intra_cl_module.py
+
+
+class IntraCLBlock(nn.Layer):
+    def __init__(self, in_channels=96, reduce_factor=4):
+        super(IntraCLBlock, self).__init__()
+        self.channels = in_channels
+        self.rf = reduce_factor
+        weight_attr = paddle.nn.initializer.KaimingUniform()
+        self.conv1x1_reduce_channel = nn.Conv2D(
+            self.channels, self.channels // self.rf, kernel_size=1, stride=1, padding=0
+        )
+        self.conv1x1_return_channel = nn.Conv2D(
+            self.channels // self.rf, self.channels, kernel_size=1, stride=1, padding=0
+        )
+
+        self.v_layer_7x1 = nn.Conv2D(
+            self.channels // self.rf,
+            self.channels // self.rf,
+            kernel_size=(7, 1),
+            stride=(1, 1),
+            padding=(3, 0),
+        )
+        self.v_layer_5x1 = nn.Conv2D(
+            self.channels // self.rf,
+            self.channels // self.rf,
+            kernel_size=(5, 1),
+            stride=(1, 1),
+            padding=(2, 0),
+        )
+        self.v_layer_3x1 = nn.Conv2D(
+            self.channels // self.rf,
+            self.channels // self.rf,
+            kernel_size=(3, 1),
+            stride=(1, 1),
+            padding=(1, 0),
+        )
+
+        self.q_layer_1x7 = nn.Conv2D(
+            self.channels // self.rf,
+            self.channels // self.rf,
+            kernel_size=(1, 7),
+            stride=(1, 1),
+            padding=(0, 3),
+        )
+        self.q_layer_1x5 = nn.Conv2D(
+            self.channels // self.rf,
+            self.channels // self.rf,
+            kernel_size=(1, 5),
+            stride=(1, 1),
+            padding=(0, 2),
+        )
+        self.q_layer_1x3 = nn.Conv2D(
+            self.channels // self.rf,
+            self.channels // self.rf,
+            kernel_size=(1, 3),
+            stride=(1, 1),
+            padding=(0, 1),
+        )
+
+        # base
+        self.c_layer_7x7 = nn.Conv2D(
+            self.channels // self.rf,
+            self.channels // self.rf,
+            kernel_size=(7, 7),
+            stride=(1, 1),
+            padding=(3, 3),
+        )
+        self.c_layer_5x5 = nn.Conv2D(
+            self.channels // self.rf,
+            self.channels // self.rf,
+            kernel_size=(5, 5),
+            stride=(1, 1),
+            padding=(2, 2),
+        )
+        self.c_layer_3x3 = nn.Conv2D(
+            self.channels // self.rf,
+            self.channels // self.rf,
+            kernel_size=(3, 3),
+            stride=(1, 1),
+            padding=(1, 1),
+        )
+
+        self.bn = nn.BatchNorm2D(self.channels)
+        self.relu = nn.ReLU()
+
+    def forward(self, x):
+        x_new = self.conv1x1_reduce_channel(x)
+
+        x_7_c = self.c_layer_7x7(x_new)
+        x_7_v = self.v_layer_7x1(x_new)
+        x_7_q = self.q_layer_1x7(x_new)
+        x_7 = x_7_c + x_7_v + x_7_q
+
+        x_5_c = self.c_layer_5x5(x_7)
+        x_5_v = self.v_layer_5x1(x_7)
+        x_5_q = self.q_layer_1x5(x_7)
+        x_5 = x_5_c + x_5_v + x_5_q
+
+        x_3_c = self.c_layer_3x3(x_5)
+        x_3_v = self.v_layer_3x1(x_5)
+        x_3_q = self.q_layer_1x3(x_5)
+        x_3 = x_3_c + x_3_v + x_3_q
+
+        x_relation = self.conv1x1_return_channel(x_3)
+
+        x_relation = self.bn(x_relation)
+        x_relation = self.relu(x_relation)
+
+        return x + x_relation
+
+
+def build_intraclblock_list(num_block):
+    IntraCLBlock_list = nn.LayerList()
+    for i in range(num_block):
+        IntraCLBlock_list.append(IntraCLBlock())
+
+    return IntraCLBlock_list
diff --git a/ppocr/modeling/necks/pg_fpn.py b/ppocr/modeling/necks/pg_fpn.py
new file mode 100644
index 0000000..cae4b22
--- /dev/null
+++ b/ppocr/modeling/necks/pg_fpn.py
@@ -0,0 +1,345 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride=1,
+        groups=1,
+        is_vd_mode=False,
+        act=None,
+        name=None,
+    ):
+        super(ConvBNLayer, self).__init__()
+
+        self.is_vd_mode = is_vd_mode
+        self._pool2d_avg = nn.AvgPool2D(
+            kernel_size=2, stride=2, padding=0, ceil_mode=True
+        )
+        self._conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=(kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(name=name + "_weights"),
+            bias_attr=False,
+        )
+        if name == "conv1":
+            bn_name = "bn_" + name
+        else:
+            bn_name = "bn" + name[3:]
+        self._batch_norm = nn.BatchNorm(
+            out_channels,
+            act=act,
+            param_attr=ParamAttr(name=bn_name + "_scale"),
+            bias_attr=ParamAttr(bn_name + "_offset"),
+            moving_mean_name=bn_name + "_mean",
+            moving_variance_name=bn_name + "_variance",
+            use_global_stats=False,
+        )
+
+    def forward(self, inputs):
+        y = self._conv(inputs)
+        y = self._batch_norm(y)
+        return y
+
+
+class DeConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size=4,
+        stride=2,
+        padding=1,
+        groups=1,
+        if_act=True,
+        act=None,
+        name=None,
+    ):
+        super(DeConvBNLayer, self).__init__()
+
+        self.if_act = if_act
+        self.act = act
+        self.deconv = nn.Conv2DTranspose(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=padding,
+            groups=groups,
+            weight_attr=ParamAttr(name=name + "_weights"),
+            bias_attr=False,
+        )
+        self.bn = nn.BatchNorm(
+            num_channels=out_channels,
+            act=act,
+            param_attr=ParamAttr(name="bn_" + name + "_scale"),
+            bias_attr=ParamAttr(name="bn_" + name + "_offset"),
+            moving_mean_name="bn_" + name + "_mean",
+            moving_variance_name="bn_" + name + "_variance",
+            use_global_stats=False,
+        )
+
+    def forward(self, x):
+        x = self.deconv(x)
+        x = self.bn(x)
+        return x
+
+
+class PGFPN(nn.Layer):
+    def __init__(self, in_channels, **kwargs):
+        super(PGFPN, self).__init__()
+        num_inputs = [2048, 2048, 1024, 512, 256]
+        num_outputs = [256, 256, 192, 192, 128]
+        self.out_channels = 128
+        self.conv_bn_layer_1 = ConvBNLayer(
+            in_channels=3,
+            out_channels=32,
+            kernel_size=3,
+            stride=1,
+            act=None,
+            name="FPN_d1",
+        )
+        self.conv_bn_layer_2 = ConvBNLayer(
+            in_channels=64,
+            out_channels=64,
+            kernel_size=3,
+            stride=1,
+            act=None,
+            name="FPN_d2",
+        )
+        self.conv_bn_layer_3 = ConvBNLayer(
+            in_channels=256,
+            out_channels=128,
+            kernel_size=3,
+            stride=1,
+            act=None,
+            name="FPN_d3",
+        )
+        self.conv_bn_layer_4 = ConvBNLayer(
+            in_channels=32,
+            out_channels=64,
+            kernel_size=3,
+            stride=2,
+            act=None,
+            name="FPN_d4",
+        )
+        self.conv_bn_layer_5 = ConvBNLayer(
+            in_channels=64,
+            out_channels=64,
+            kernel_size=3,
+            stride=1,
+            act="relu",
+            name="FPN_d5",
+        )
+        self.conv_bn_layer_6 = ConvBNLayer(
+            in_channels=64,
+            out_channels=128,
+            kernel_size=3,
+            stride=2,
+            act=None,
+            name="FPN_d6",
+        )
+        self.conv_bn_layer_7 = ConvBNLayer(
+            in_channels=128,
+            out_channels=128,
+            kernel_size=3,
+            stride=1,
+            act="relu",
+            name="FPN_d7",
+        )
+        self.conv_bn_layer_8 = ConvBNLayer(
+            in_channels=128,
+            out_channels=128,
+            kernel_size=1,
+            stride=1,
+            act=None,
+            name="FPN_d8",
+        )
+
+        self.conv_h0 = ConvBNLayer(
+            in_channels=num_inputs[0],
+            out_channels=num_outputs[0],
+            kernel_size=1,
+            stride=1,
+            act=None,
+            name="conv_h{}".format(0),
+        )
+        self.conv_h1 = ConvBNLayer(
+            in_channels=num_inputs[1],
+            out_channels=num_outputs[1],
+            kernel_size=1,
+            stride=1,
+            act=None,
+            name="conv_h{}".format(1),
+        )
+        self.conv_h2 = ConvBNLayer(
+            in_channels=num_inputs[2],
+            out_channels=num_outputs[2],
+            kernel_size=1,
+            stride=1,
+            act=None,
+            name="conv_h{}".format(2),
+        )
+        self.conv_h3 = ConvBNLayer(
+            in_channels=num_inputs[3],
+            out_channels=num_outputs[3],
+            kernel_size=1,
+            stride=1,
+            act=None,
+            name="conv_h{}".format(3),
+        )
+        self.conv_h4 = ConvBNLayer(
+            in_channels=num_inputs[4],
+            out_channels=num_outputs[4],
+            kernel_size=1,
+            stride=1,
+            act=None,
+            name="conv_h{}".format(4),
+        )
+
+        self.dconv0 = DeConvBNLayer(
+            in_channels=num_outputs[0],
+            out_channels=num_outputs[0 + 1],
+            name="dconv_{}".format(0),
+        )
+        self.dconv1 = DeConvBNLayer(
+            in_channels=num_outputs[1],
+            out_channels=num_outputs[1 + 1],
+            act=None,
+            name="dconv_{}".format(1),
+        )
+        self.dconv2 = DeConvBNLayer(
+            in_channels=num_outputs[2],
+            out_channels=num_outputs[2 + 1],
+            act=None,
+            name="dconv_{}".format(2),
+        )
+        self.dconv3 = DeConvBNLayer(
+            in_channels=num_outputs[3],
+            out_channels=num_outputs[3 + 1],
+            act=None,
+            name="dconv_{}".format(3),
+        )
+        self.conv_g1 = ConvBNLayer(
+            in_channels=num_outputs[1],
+            out_channels=num_outputs[1],
+            kernel_size=3,
+            stride=1,
+            act="relu",
+            name="conv_g{}".format(1),
+        )
+        self.conv_g2 = ConvBNLayer(
+            in_channels=num_outputs[2],
+            out_channels=num_outputs[2],
+            kernel_size=3,
+            stride=1,
+            act="relu",
+            name="conv_g{}".format(2),
+        )
+        self.conv_g3 = ConvBNLayer(
+            in_channels=num_outputs[3],
+            out_channels=num_outputs[3],
+            kernel_size=3,
+            stride=1,
+            act="relu",
+            name="conv_g{}".format(3),
+        )
+        self.conv_g4 = ConvBNLayer(
+            in_channels=num_outputs[4],
+            out_channels=num_outputs[4],
+            kernel_size=3,
+            stride=1,
+            act="relu",
+            name="conv_g{}".format(4),
+        )
+        self.convf = ConvBNLayer(
+            in_channels=num_outputs[4],
+            out_channels=num_outputs[4],
+            kernel_size=1,
+            stride=1,
+            act=None,
+            name="conv_f{}".format(4),
+        )
+
+    def forward(self, x):
+        c0, c1, c2, c3, c4, c5, c6 = x
+        # FPN_Down_Fusion
+        f = [c0, c1, c2]
+        g = [None, None, None]
+        h = [None, None, None]
+        h[0] = self.conv_bn_layer_1(f[0])
+        h[1] = self.conv_bn_layer_2(f[1])
+        h[2] = self.conv_bn_layer_3(f[2])
+
+        g[0] = self.conv_bn_layer_4(h[0])
+        g[1] = paddle.add(g[0], h[1])
+        g[1] = F.relu(g[1])
+        g[1] = self.conv_bn_layer_5(g[1])
+        g[1] = self.conv_bn_layer_6(g[1])
+
+        g[2] = paddle.add(g[1], h[2])
+        g[2] = F.relu(g[2])
+        g[2] = self.conv_bn_layer_7(g[2])
+        f_down = self.conv_bn_layer_8(g[2])
+
+        # FPN UP Fusion
+        f1 = [c6, c5, c4, c3, c2]
+        g = [None, None, None, None, None]
+        h = [None, None, None, None, None]
+        h[0] = self.conv_h0(f1[0])
+        h[1] = self.conv_h1(f1[1])
+        h[2] = self.conv_h2(f1[2])
+        h[3] = self.conv_h3(f1[3])
+        h[4] = self.conv_h4(f1[4])
+
+        g[0] = self.dconv0(h[0])
+        g[1] = paddle.add(g[0], h[1])
+        g[1] = F.relu(g[1])
+        g[1] = self.conv_g1(g[1])
+        g[1] = self.dconv1(g[1])
+
+        g[2] = paddle.add(g[1], h[2])
+        g[2] = F.relu(g[2])
+        g[2] = self.conv_g2(g[2])
+        g[2] = self.dconv2(g[2])
+
+        g[3] = paddle.add(g[2], h[3])
+        g[3] = F.relu(g[3])
+        g[3] = self.conv_g3(g[3])
+        g[3] = self.dconv3(g[3])
+
+        g[4] = paddle.add(x=g[3], y=h[4])
+        g[4] = F.relu(g[4])
+        g[4] = self.conv_g4(g[4])
+        f_up = self.convf(g[4])
+        f_common = paddle.add(f_down, f_up)
+        f_common = F.relu(f_common)
+        return f_common
diff --git a/ppocr/modeling/necks/pren_fpn.py b/ppocr/modeling/necks/pren_fpn.py
new file mode 100644
index 0000000..29c98e9
--- /dev/null
+++ b/ppocr/modeling/necks/pren_fpn.py
@@ -0,0 +1,177 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+Code is refer from:
+https://github.com/RuijieJ/pren/blob/main/Nets/Aggregation.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+
+
+class PoolAggregate(nn.Layer):
+    def __init__(self, n_r, d_in, d_middle=None, d_out=None):
+        super(PoolAggregate, self).__init__()
+        if not d_middle:
+            d_middle = d_in
+        if not d_out:
+            d_out = d_in
+
+        self.d_in = d_in
+        self.d_middle = d_middle
+        self.d_out = d_out
+        self.act = nn.Swish()
+
+        self.n_r = n_r
+        self.aggs = self._build_aggs()
+
+    def _build_aggs(self):
+        aggs = []
+        for i in range(self.n_r):
+            aggs.append(
+                self.add_sublayer(
+                    "{}".format(i),
+                    nn.Sequential(
+                        (
+                            "conv1",
+                            nn.Conv2D(
+                                self.d_in, self.d_middle, 3, 2, 1, bias_attr=False
+                            ),
+                        ),
+                        ("bn1", nn.BatchNorm(self.d_middle)),
+                        ("act", self.act),
+                        (
+                            "conv2",
+                            nn.Conv2D(
+                                self.d_middle, self.d_out, 3, 2, 1, bias_attr=False
+                            ),
+                        ),
+                        ("bn2", nn.BatchNorm(self.d_out)),
+                    ),
+                )
+            )
+        return aggs
+
+    def forward(self, x):
+        b = x.shape[0]
+        outs = []
+        for agg in self.aggs:
+            y = agg(x)
+            p = F.adaptive_avg_pool2d(y, 1)
+            outs.append(p.reshape((b, 1, self.d_out)))
+        out = paddle.concat(outs, 1)
+        return out
+
+
+class WeightAggregate(nn.Layer):
+    def __init__(self, n_r, d_in, d_middle=None, d_out=None):
+        super(WeightAggregate, self).__init__()
+        if not d_middle:
+            d_middle = d_in
+        if not d_out:
+            d_out = d_in
+
+        self.n_r = n_r
+        self.d_out = d_out
+        self.act = nn.Swish()
+
+        self.conv_n = nn.Sequential(
+            ("conv1", nn.Conv2D(d_in, d_in, 3, 1, 1, bias_attr=False)),
+            ("bn1", nn.BatchNorm(d_in)),
+            ("act1", self.act),
+            ("conv2", nn.Conv2D(d_in, n_r, 1, bias_attr=False)),
+            ("bn2", nn.BatchNorm(n_r)),
+            ("act2", nn.Sigmoid()),
+        )
+        self.conv_d = nn.Sequential(
+            ("conv1", nn.Conv2D(d_in, d_middle, 3, 1, 1, bias_attr=False)),
+            ("bn1", nn.BatchNorm(d_middle)),
+            ("act1", self.act),
+            ("conv2", nn.Conv2D(d_middle, d_out, 1, bias_attr=False)),
+            ("bn2", nn.BatchNorm(d_out)),
+        )
+
+    def forward(self, x):
+        b, _, h, w = x.shape
+
+        hmaps = self.conv_n(x)
+        fmaps = self.conv_d(x)
+        r = paddle.bmm(
+            hmaps.reshape((b, self.n_r, h * w)),
+            fmaps.reshape((b, self.d_out, h * w)).transpose((0, 2, 1)),
+        )
+        return r
+
+
+class GCN(nn.Layer):
+    def __init__(self, d_in, n_in, d_out=None, n_out=None, dropout=0.1):
+        super(GCN, self).__init__()
+        if not d_out:
+            d_out = d_in
+        if not n_out:
+            n_out = d_in
+
+        self.conv_n = nn.Conv1D(n_in, n_out, 1)
+        self.linear = nn.Linear(d_in, d_out)
+        self.dropout = nn.Dropout(dropout)
+        self.act = nn.Swish()
+
+    def forward(self, x):
+        x = self.conv_n(x)
+        x = self.dropout(self.linear(x))
+        return self.act(x)
+
+
+class PRENFPN(nn.Layer):
+    def __init__(self, in_channels, n_r, d_model, max_len, dropout):
+        super(PRENFPN, self).__init__()
+        assert len(in_channels) == 3, "in_channels' length must be 3."
+        c1, c2, c3 = in_channels  # the depths are from big to small
+        # build fpn
+        assert d_model % 3 == 0, "{} can't be divided by 3.".format(d_model)
+        self.agg_p1 = PoolAggregate(n_r, c1, d_out=d_model // 3)
+        self.agg_p2 = PoolAggregate(n_r, c2, d_out=d_model // 3)
+        self.agg_p3 = PoolAggregate(n_r, c3, d_out=d_model // 3)
+
+        self.agg_w1 = WeightAggregate(n_r, c1, 4 * c1, d_model // 3)
+        self.agg_w2 = WeightAggregate(n_r, c2, 4 * c2, d_model // 3)
+        self.agg_w3 = WeightAggregate(n_r, c3, 4 * c3, d_model // 3)
+
+        self.gcn_pool = GCN(d_model, n_r, d_model, max_len, dropout)
+        self.gcn_weight = GCN(d_model, n_r, d_model, max_len, dropout)
+
+        self.out_channels = d_model
+
+    def forward(self, inputs):
+        f3, f5, f7 = inputs
+
+        rp1 = self.agg_p1(f3)
+        rp2 = self.agg_p2(f5)
+        rp3 = self.agg_p3(f7)
+        rp = paddle.concat([rp1, rp2, rp3], 2)  # [b,nr,d]
+
+        rw1 = self.agg_w1(f3)
+        rw2 = self.agg_w2(f5)
+        rw3 = self.agg_w3(f7)
+        rw = paddle.concat([rw1, rw2, rw3], 2)  # [b,nr,d]
+
+        y1 = self.gcn_pool(rp)
+        y2 = self.gcn_weight(rw)
+        y = 0.5 * (y1 + y2)
+        return y  # [b,max_len,d]
diff --git a/ppocr/modeling/necks/rf_adaptor.py b/ppocr/modeling/necks/rf_adaptor.py
new file mode 100644
index 0000000..e27e894
--- /dev/null
+++ b/ppocr/modeling/necks/rf_adaptor.py
@@ -0,0 +1,146 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/hikopensource/DAVAR-Lab-OCR/blob/main/davarocr/davar_rcg/models/connects/single_block/RFAdaptor.py
+"""
+
+import paddle
+import paddle.nn as nn
+from paddle.nn.initializer import TruncatedNormal, Constant, Normal, KaimingNormal
+
+kaiming_init_ = KaimingNormal()
+zeros_ = Constant(value=0.0)
+ones_ = Constant(value=1.0)
+
+
+class S2VAdaptor(nn.Layer):
+    """Semantic to Visual adaptation module"""
+
+    def __init__(self, in_channels=512):
+        super(S2VAdaptor, self).__init__()
+
+        self.in_channels = in_channels  # 512
+
+        # feature strengthen module, channel attention
+        self.channel_inter = nn.Linear(
+            self.in_channels, self.in_channels, bias_attr=False
+        )
+        self.channel_bn = nn.BatchNorm1D(self.in_channels)
+        self.channel_act = nn.ReLU()
+        self.apply(self.init_weights)
+
+    def init_weights(self, m):
+        if isinstance(m, nn.Conv2D):
+            kaiming_init_(m.weight)
+            if isinstance(m, nn.Conv2D) and m.bias is not None:
+                zeros_(m.bias)
+        elif isinstance(m, (nn.BatchNorm, nn.BatchNorm2D, nn.BatchNorm1D)):
+            zeros_(m.bias)
+            ones_(m.weight)
+
+    def forward(self, semantic):
+        semantic_source = semantic  # batch, channel, height, width
+
+        # feature transformation
+        semantic = semantic.squeeze(2).transpose([0, 2, 1])  # batch, width, channel
+        channel_att = self.channel_inter(semantic)  # batch, width, channel
+        channel_att = channel_att.transpose([0, 2, 1])  # batch, channel, width
+        channel_bn = self.channel_bn(channel_att)  # batch, channel, width
+        channel_att = self.channel_act(channel_bn)  # batch, channel, width
+
+        # Feature enhancement
+        channel_output = semantic_source * channel_att.unsqueeze(
+            -2
+        )  # batch, channel, 1, width
+
+        return channel_output
+
+
+class V2SAdaptor(nn.Layer):
+    """Visual to Semantic adaptation module"""
+
+    def __init__(self, in_channels=512, return_mask=False):
+        super(V2SAdaptor, self).__init__()
+
+        # parameter initialization
+        self.in_channels = in_channels
+        self.return_mask = return_mask
+
+        # output transformation
+        self.channel_inter = nn.Linear(
+            self.in_channels, self.in_channels, bias_attr=False
+        )
+        self.channel_bn = nn.BatchNorm1D(self.in_channels)
+        self.channel_act = nn.ReLU()
+
+    def forward(self, visual):
+        # Feature enhancement
+        visual = visual.squeeze(2).transpose([0, 2, 1])  # batch, width, channel
+        channel_att = self.channel_inter(visual)  # batch, width, channel
+        channel_att = channel_att.transpose([0, 2, 1])  # batch, channel, width
+        channel_bn = self.channel_bn(channel_att)  # batch, channel, width
+        channel_att = self.channel_act(channel_bn)  # batch, channel, width
+
+        # size alignment
+        channel_output = channel_att.unsqueeze(-2)  # batch, width, channel
+
+        if self.return_mask:
+            return channel_output, channel_att
+        return channel_output
+
+
+class RFAdaptor(nn.Layer):
+    def __init__(self, in_channels=512, use_v2s=True, use_s2v=True, **kwargs):
+        super(RFAdaptor, self).__init__()
+        if use_v2s is True:
+            self.neck_v2s = V2SAdaptor(in_channels=in_channels, **kwargs)
+        else:
+            self.neck_v2s = None
+        if use_s2v is True:
+            self.neck_s2v = S2VAdaptor(in_channels=in_channels, **kwargs)
+        else:
+            self.neck_s2v = None
+        self.out_channels = in_channels
+
+    def forward(self, x):
+        visual_feature, rcg_feature = x
+        if visual_feature is not None:
+            (
+                batch,
+                source_channels,
+                v_source_height,
+                v_source_width,
+            ) = visual_feature.shape
+            visual_feature = visual_feature.reshape(
+                [batch, source_channels, 1, v_source_height * v_source_width]
+            )
+
+        if self.neck_v2s is not None:
+            v_rcg_feature = rcg_feature * self.neck_v2s(visual_feature)
+        else:
+            v_rcg_feature = rcg_feature
+
+        if self.neck_s2v is not None:
+            v_visual_feature = visual_feature + self.neck_s2v(rcg_feature)
+        else:
+            v_visual_feature = visual_feature
+        if v_rcg_feature is not None:
+            batch, source_channels, source_height, source_width = v_rcg_feature.shape
+            v_rcg_feature = v_rcg_feature.reshape(
+                [batch, source_channels, 1, source_height * source_width]
+            )
+
+            v_rcg_feature = v_rcg_feature.squeeze(2).transpose([0, 2, 1])
+        return v_visual_feature, v_rcg_feature
diff --git a/ppocr/modeling/necks/rnn.py b/ppocr/modeling/necks/rnn.py
new file mode 100644
index 0000000..1d935be
--- /dev/null
+++ b/ppocr/modeling/necks/rnn.py
@@ -0,0 +1,392 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+from paddle.nn.initializer import KaimingNormal
+
+from ppocr.modeling.heads.rec_ctc_head import get_para_bias_attr
+from ppocr.modeling.backbones.rec_svtrnet import (
+    Block,
+    ConvBNLayer,
+    trunc_normal_,
+    zeros_,
+    ones_,
+)
+
+
+class Im2Seq(nn.Layer):
+    def __init__(self, in_channels, **kwargs):
+        super().__init__()
+        self.out_channels = in_channels
+
+    def forward(self, x):
+        B, C, H, W = x.shape
+        assert H == 1
+        x = x.squeeze(axis=2)
+        x = x.transpose([0, 2, 1])  # (NTC)(batch, width, channels)
+        return x
+
+
+class EncoderWithRNN(nn.Layer):
+    def __init__(self, in_channels, hidden_size):
+        super(EncoderWithRNN, self).__init__()
+        self.out_channels = hidden_size * 2
+        self.lstm = nn.LSTM(
+            in_channels, hidden_size, direction="bidirectional", num_layers=2
+        )
+
+    def forward(self, x):
+        x, _ = self.lstm(x)
+        return x
+
+
+class BidirectionalLSTM(nn.Layer):
+    def __init__(
+        self,
+        input_size,
+        hidden_size,
+        output_size=None,
+        num_layers=1,
+        dropout=0,
+        direction=False,
+        time_major=False,
+        with_linear=False,
+    ):
+        super(BidirectionalLSTM, self).__init__()
+        self.with_linear = with_linear
+        self.rnn = nn.LSTM(
+            input_size,
+            hidden_size,
+            num_layers=num_layers,
+            dropout=dropout,
+            direction=direction,
+            time_major=time_major,
+        )
+
+        # text recognition the specified structure LSTM with linear
+        if self.with_linear:
+            self.linear = nn.Linear(hidden_size * 2, output_size)
+
+    def forward(self, input_feature):
+        recurrent, _ = self.rnn(
+            input_feature
+        )  # batch_size x T x input_size -> batch_size x T x (2*hidden_size)
+        if self.with_linear:
+            output = self.linear(recurrent)  # batch_size x T x output_size
+            return output
+        return recurrent
+
+
+class EncoderWithCascadeRNN(nn.Layer):
+    def __init__(
+        self, in_channels, hidden_size, out_channels, num_layers=2, with_linear=False
+    ):
+        super(EncoderWithCascadeRNN, self).__init__()
+        self.out_channels = out_channels[-1]
+        self.encoder = nn.LayerList(
+            [
+                BidirectionalLSTM(
+                    in_channels if i == 0 else out_channels[i - 1],
+                    hidden_size,
+                    output_size=out_channels[i],
+                    num_layers=1,
+                    direction="bidirectional",
+                    with_linear=with_linear,
+                )
+                for i in range(num_layers)
+            ]
+        )
+
+    def forward(self, x):
+        for i, l in enumerate(self.encoder):
+            x = l(x)
+        return x
+
+
+class EncoderWithFC(nn.Layer):
+    def __init__(self, in_channels, hidden_size):
+        super(EncoderWithFC, self).__init__()
+        self.out_channels = hidden_size
+        weight_attr, bias_attr = get_para_bias_attr(l2_decay=0.00001, k=in_channels)
+        self.fc = nn.Linear(
+            in_channels,
+            hidden_size,
+            weight_attr=weight_attr,
+            bias_attr=bias_attr,
+            name="reduce_encoder_fea",
+        )
+
+    def forward(self, x):
+        x = self.fc(x)
+        return x
+
+
+class EncoderWithSVTR(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        dims=64,  # XS
+        depth=2,
+        hidden_dims=120,
+        use_guide=False,
+        num_heads=8,
+        qkv_bias=True,
+        mlp_ratio=2.0,
+        drop_rate=0.1,
+        attn_drop_rate=0.1,
+        drop_path=0.0,
+        kernel_size=[3, 3],
+        qk_scale=None,
+    ):
+        super(EncoderWithSVTR, self).__init__()
+        self.depth = depth
+        self.use_guide = use_guide
+        self.conv1 = ConvBNLayer(
+            in_channels,
+            in_channels // 8,
+            kernel_size=kernel_size,
+            padding=[kernel_size[0] // 2, kernel_size[1] // 2],
+            act=nn.Swish,
+        )
+        self.conv2 = ConvBNLayer(
+            in_channels // 8, hidden_dims, kernel_size=1, act=nn.Swish
+        )
+
+        self.svtr_block = nn.LayerList(
+            [
+                Block(
+                    dim=hidden_dims,
+                    num_heads=num_heads,
+                    mixer="Global",
+                    HW=None,
+                    mlp_ratio=mlp_ratio,
+                    qkv_bias=qkv_bias,
+                    qk_scale=qk_scale,
+                    drop=drop_rate,
+                    act_layer=nn.Swish,
+                    attn_drop=attn_drop_rate,
+                    drop_path=drop_path,
+                    norm_layer="nn.LayerNorm",
+                    epsilon=1e-05,
+                    prenorm=False,
+                )
+                for i in range(depth)
+            ]
+        )
+        self.norm = nn.LayerNorm(hidden_dims, epsilon=1e-6)
+        self.conv3 = ConvBNLayer(hidden_dims, in_channels, kernel_size=1, act=nn.Swish)
+        # last conv-nxn, the input is concat of input tensor and conv3 output tensor
+        self.conv4 = ConvBNLayer(
+            2 * in_channels,
+            in_channels // 8,
+            kernel_size=kernel_size,
+            padding=[kernel_size[0] // 2, kernel_size[1] // 2],
+            act=nn.Swish,
+        )
+
+        self.conv1x1 = ConvBNLayer(in_channels // 8, dims, kernel_size=1, act=nn.Swish)
+        self.out_channels = dims
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight)
+            if isinstance(m, nn.Linear) and m.bias is not None:
+                zeros_(m.bias)
+        elif isinstance(m, nn.LayerNorm):
+            zeros_(m.bias)
+            ones_(m.weight)
+
+    def forward(self, x):
+        # for use guide
+        if self.use_guide:
+            z = x.clone()
+            z.stop_gradient = True
+        else:
+            z = x
+        # for short cut
+        h = z
+        # reduce dim
+        z = self.conv1(z)
+        z = self.conv2(z)
+        # SVTR global block
+        B, C, H, W = z.shape
+        z = z.flatten(2).transpose([0, 2, 1])
+        for blk in self.svtr_block:
+            z = blk(z)
+        z = self.norm(z)
+        # last stage
+        z = z.reshape([0, H, W, C]).transpose([0, 3, 1, 2])
+        z = self.conv3(z)
+        z = paddle.concat((h, z), axis=1)
+        z = self.conv1x1(self.conv4(z))
+        return z
+
+
+class EncoderWithLightSVTR(nn.Layer):
+    """Light SVTR neck: attention with lightweight skip connection.
+
+    Replaces heavy guide fusion with a lightweight 1x1 skip conv + add.
+    Adds a DWConv before attention for local context priming.
+
+    When use_guide=True, applies stop_gradient to input so that neck
+    gradients do not propagate back to the backbone (implicit regularization).
+    """
+
+    def __init__(
+        self,
+        in_channels,
+        dims=64,
+        depth=1,
+        num_heads=8,
+        qkv_bias=True,
+        mlp_ratio=4.0,
+        drop_rate=0.1,
+        attn_drop_rate=0.1,
+        drop_path=0.0,
+        qk_scale=None,
+        local_kernel=7,
+        use_guide=False,
+    ):
+        super().__init__()
+        self.use_guide = use_guide
+        self.conv_reduce = ConvBNLayer(
+            in_channels,
+            dims,
+            kernel_size=1,
+            act=nn.Swish,
+        )
+        self.local_conv = nn.Sequential(
+            nn.Conv2D(
+                dims,
+                dims,
+                [1, local_kernel],
+                padding=[0, local_kernel // 2],
+                groups=dims,
+                bias_attr=False,
+                weight_attr=paddle.ParamAttr(initializer=KaimingNormal()),
+            ),
+            nn.BatchNorm2D(dims),
+            nn.Swish(),
+        )
+
+        self.svtr_block = nn.LayerList(
+            [
+                Block(
+                    dim=dims,
+                    num_heads=num_heads,
+                    mixer="Global",
+                    HW=None,
+                    mlp_ratio=mlp_ratio,
+                    qkv_bias=qkv_bias,
+                    qk_scale=qk_scale,
+                    drop=drop_rate,
+                    act_layer=nn.Swish,
+                    attn_drop=attn_drop_rate,
+                    drop_path=drop_path,
+                    norm_layer="nn.LayerNorm",
+                    epsilon=1e-05,
+                    prenorm=False,
+                )
+                for _ in range(depth)
+            ]
+        )
+        self.norm = nn.LayerNorm(dims, epsilon=1e-6)
+
+        self.skip_conv = ConvBNLayer(
+            in_channels,
+            dims,
+            kernel_size=1,
+            act=nn.Swish,
+        )
+
+        self.out_channels = dims
+        self.apply(self._init_weights)
+
+    def _init_weights(self, m):
+        if isinstance(m, nn.Linear):
+            trunc_normal_(m.weight)
+            if m.bias is not None:
+                zeros_(m.bias)
+        elif isinstance(m, nn.LayerNorm):
+            zeros_(m.bias)
+            ones_(m.weight)
+
+    def forward(self, x):
+        if self.use_guide:
+            x = x.clone()
+            x.stop_gradient = True
+        skip = self.skip_conv(x)
+        z = self.conv_reduce(x)
+        z = z + self.local_conv(z)
+        B, C, H, W = z.shape
+        z = z.flatten(2).transpose([0, 2, 1])
+        for blk in self.svtr_block:
+            z = blk(z)
+        z = self.norm(z)
+        z = z.reshape([0, H, W, C]).transpose([0, 3, 1, 2])
+        z = z + skip
+        return z
+
+
+class SequenceEncoder(nn.Layer):
+    def __init__(self, in_channels, encoder_type, hidden_size=48, **kwargs):
+        super(SequenceEncoder, self).__init__()
+        self.encoder_reshape = Im2Seq(in_channels)
+        self.out_channels = self.encoder_reshape.out_channels
+        self.encoder_type = encoder_type
+        if encoder_type == "reshape":
+            self.only_reshape = True
+        else:
+            support_encoder_dict = {
+                "reshape": Im2Seq,
+                "fc": EncoderWithFC,
+                "rnn": EncoderWithRNN,
+                "svtr": EncoderWithSVTR,
+                "lightsvtr": EncoderWithLightSVTR,
+                "cascadernn": EncoderWithCascadeRNN,
+            }
+            assert encoder_type in support_encoder_dict, "{} must in {}".format(
+                encoder_type, support_encoder_dict.keys()
+            )
+            if encoder_type in ("svtr", "lightsvtr"):
+                self.encoder = support_encoder_dict[encoder_type](
+                    self.encoder_reshape.out_channels, **kwargs
+                )
+            elif encoder_type == "cascadernn":
+                self.encoder = support_encoder_dict[encoder_type](
+                    self.encoder_reshape.out_channels, hidden_size, **kwargs
+                )
+            else:
+                self.encoder = support_encoder_dict[encoder_type](
+                    self.encoder_reshape.out_channels, hidden_size
+                )
+            self.out_channels = self.encoder.out_channels
+            self.only_reshape = False
+
+    def forward(self, x):
+        if self.encoder_type not in ("svtr", "lightsvtr"):
+            x = self.encoder_reshape(x)
+            if not self.only_reshape:
+                x = self.encoder(x)
+            return x
+        else:
+            x = self.encoder(x)
+            x = self.encoder_reshape(x)
+            return x
diff --git a/ppocr/modeling/necks/sast_fpn.py b/ppocr/modeling/necks/sast_fpn.py
new file mode 100644
index 0000000..4804806
--- /dev/null
+++ b/ppocr/modeling/necks/sast_fpn.py
@@ -0,0 +1,368 @@
+# copyright (c) 2019 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride,
+        groups=1,
+        if_act=True,
+        act=None,
+        name=None,
+    ):
+        super(ConvBNLayer, self).__init__()
+        self.if_act = if_act
+        self.act = act
+        self.conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=(kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(name=name + "_weights"),
+            bias_attr=False,
+        )
+
+        self.bn = nn.BatchNorm(
+            num_channels=out_channels,
+            act=act,
+            param_attr=ParamAttr(name="bn_" + name + "_scale"),
+            bias_attr=ParamAttr(name="bn_" + name + "_offset"),
+            moving_mean_name="bn_" + name + "_mean",
+            moving_variance_name="bn_" + name + "_variance",
+        )
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        return x
+
+
+class DeConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride,
+        groups=1,
+        if_act=True,
+        act=None,
+        name=None,
+    ):
+        super(DeConvBNLayer, self).__init__()
+        self.if_act = if_act
+        self.act = act
+        self.deconv = nn.Conv2DTranspose(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=(kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(name=name + "_weights"),
+            bias_attr=False,
+        )
+        self.bn = nn.BatchNorm(
+            num_channels=out_channels,
+            act=act,
+            param_attr=ParamAttr(name="bn_" + name + "_scale"),
+            bias_attr=ParamAttr(name="bn_" + name + "_offset"),
+            moving_mean_name="bn_" + name + "_mean",
+            moving_variance_name="bn_" + name + "_variance",
+        )
+
+    def forward(self, x):
+        x = self.deconv(x)
+        x = self.bn(x)
+        return x
+
+
+class FPN_Up_Fusion(nn.Layer):
+    def __init__(self, in_channels):
+        super(FPN_Up_Fusion, self).__init__()
+        in_channels = in_channels[::-1]
+        out_channels = [256, 256, 192, 192, 128]
+
+        self.h0_conv = ConvBNLayer(
+            in_channels[0], out_channels[0], 1, 1, act=None, name="fpn_up_h0"
+        )
+        self.h1_conv = ConvBNLayer(
+            in_channels[1], out_channels[1], 1, 1, act=None, name="fpn_up_h1"
+        )
+        self.h2_conv = ConvBNLayer(
+            in_channels[2], out_channels[2], 1, 1, act=None, name="fpn_up_h2"
+        )
+        self.h3_conv = ConvBNLayer(
+            in_channels[3], out_channels[3], 1, 1, act=None, name="fpn_up_h3"
+        )
+        self.h4_conv = ConvBNLayer(
+            in_channels[4], out_channels[4], 1, 1, act=None, name="fpn_up_h4"
+        )
+
+        self.g0_conv = DeConvBNLayer(
+            out_channels[0], out_channels[1], 4, 2, act=None, name="fpn_up_g0"
+        )
+
+        self.g1_conv = nn.Sequential(
+            ConvBNLayer(
+                out_channels[1], out_channels[1], 3, 1, act="relu", name="fpn_up_g1_1"
+            ),
+            DeConvBNLayer(
+                out_channels[1], out_channels[2], 4, 2, act=None, name="fpn_up_g1_2"
+            ),
+        )
+        self.g2_conv = nn.Sequential(
+            ConvBNLayer(
+                out_channels[2], out_channels[2], 3, 1, act="relu", name="fpn_up_g2_1"
+            ),
+            DeConvBNLayer(
+                out_channels[2], out_channels[3], 4, 2, act=None, name="fpn_up_g2_2"
+            ),
+        )
+        self.g3_conv = nn.Sequential(
+            ConvBNLayer(
+                out_channels[3], out_channels[3], 3, 1, act="relu", name="fpn_up_g3_1"
+            ),
+            DeConvBNLayer(
+                out_channels[3], out_channels[4], 4, 2, act=None, name="fpn_up_g3_2"
+            ),
+        )
+
+        self.g4_conv = nn.Sequential(
+            ConvBNLayer(
+                out_channels[4],
+                out_channels[4],
+                3,
+                1,
+                act="relu",
+                name="fpn_up_fusion_1",
+            ),
+            ConvBNLayer(
+                out_channels[4], out_channels[4], 1, 1, act=None, name="fpn_up_fusion_2"
+            ),
+        )
+
+    def _add_relu(self, x1, x2):
+        x = paddle.add(x=x1, y=x2)
+        x = F.relu(x)
+        return x
+
+    def forward(self, x):
+        f = x[2:][::-1]
+        h0 = self.h0_conv(f[0])
+        h1 = self.h1_conv(f[1])
+        h2 = self.h2_conv(f[2])
+        h3 = self.h3_conv(f[3])
+        h4 = self.h4_conv(f[4])
+
+        g0 = self.g0_conv(h0)
+        g1 = self._add_relu(g0, h1)
+        g1 = self.g1_conv(g1)
+        g2 = self.g2_conv(self._add_relu(g1, h2))
+        g3 = self.g3_conv(self._add_relu(g2, h3))
+        g4 = self.g4_conv(self._add_relu(g3, h4))
+
+        return g4
+
+
+class FPN_Down_Fusion(nn.Layer):
+    def __init__(self, in_channels):
+        super(FPN_Down_Fusion, self).__init__()
+        out_channels = [32, 64, 128]
+
+        self.h0_conv = ConvBNLayer(
+            in_channels[0], out_channels[0], 3, 1, act=None, name="fpn_down_h0"
+        )
+        self.h1_conv = ConvBNLayer(
+            in_channels[1], out_channels[1], 3, 1, act=None, name="fpn_down_h1"
+        )
+        self.h2_conv = ConvBNLayer(
+            in_channels[2], out_channels[2], 3, 1, act=None, name="fpn_down_h2"
+        )
+
+        self.g0_conv = ConvBNLayer(
+            out_channels[0], out_channels[1], 3, 2, act=None, name="fpn_down_g0"
+        )
+
+        self.g1_conv = nn.Sequential(
+            ConvBNLayer(
+                out_channels[1], out_channels[1], 3, 1, act="relu", name="fpn_down_g1_1"
+            ),
+            ConvBNLayer(
+                out_channels[1], out_channels[2], 3, 2, act=None, name="fpn_down_g1_2"
+            ),
+        )
+
+        self.g2_conv = nn.Sequential(
+            ConvBNLayer(
+                out_channels[2],
+                out_channels[2],
+                3,
+                1,
+                act="relu",
+                name="fpn_down_fusion_1",
+            ),
+            ConvBNLayer(
+                out_channels[2],
+                out_channels[2],
+                1,
+                1,
+                act=None,
+                name="fpn_down_fusion_2",
+            ),
+        )
+
+    def forward(self, x):
+        f = x[:3]
+        h0 = self.h0_conv(f[0])
+        h1 = self.h1_conv(f[1])
+        h2 = self.h2_conv(f[2])
+        g0 = self.g0_conv(h0)
+        g1 = paddle.add(x=g0, y=h1)
+        g1 = F.relu(g1)
+        g1 = self.g1_conv(g1)
+        g2 = paddle.add(x=g1, y=h2)
+        g2 = F.relu(g2)
+        g2 = self.g2_conv(g2)
+        return g2
+
+
+class Cross_Attention(nn.Layer):
+    def __init__(self, in_channels):
+        super(Cross_Attention, self).__init__()
+        self.theta_conv = ConvBNLayer(
+            in_channels, in_channels, 1, 1, act="relu", name="f_theta"
+        )
+        self.phi_conv = ConvBNLayer(
+            in_channels, in_channels, 1, 1, act="relu", name="f_phi"
+        )
+        self.g_conv = ConvBNLayer(
+            in_channels, in_channels, 1, 1, act="relu", name="f_g"
+        )
+
+        self.fh_weight_conv = ConvBNLayer(
+            in_channels, in_channels, 1, 1, act=None, name="fh_weight"
+        )
+        self.fh_sc_conv = ConvBNLayer(
+            in_channels, in_channels, 1, 1, act=None, name="fh_sc"
+        )
+
+        self.fv_weight_conv = ConvBNLayer(
+            in_channels, in_channels, 1, 1, act=None, name="fv_weight"
+        )
+        self.fv_sc_conv = ConvBNLayer(
+            in_channels, in_channels, 1, 1, act=None, name="fv_sc"
+        )
+
+        self.f_attn_conv = ConvBNLayer(
+            in_channels * 2, in_channels, 1, 1, act="relu", name="f_attn"
+        )
+
+    def _cal_fweight(self, f, shape):
+        f_theta, f_phi, f_g = f
+        # flatten
+        f_theta = paddle.transpose(f_theta, [0, 2, 3, 1])
+        f_theta = paddle.reshape(f_theta, [shape[0] * shape[1], shape[2], 128])
+        f_phi = paddle.transpose(f_phi, [0, 2, 3, 1])
+        f_phi = paddle.reshape(f_phi, [shape[0] * shape[1], shape[2], 128])
+        f_g = paddle.transpose(f_g, [0, 2, 3, 1])
+        f_g = paddle.reshape(f_g, [shape[0] * shape[1], shape[2], 128])
+        # correlation
+        f_attn = paddle.matmul(f_theta, paddle.transpose(f_phi, [0, 2, 1]))
+        # scale
+        f_attn = f_attn / (128**0.5)
+        f_attn = F.softmax(f_attn)
+        # weighted sum
+        f_weight = paddle.matmul(f_attn, f_g)
+        f_weight = paddle.reshape(f_weight, [shape[0], shape[1], shape[2], 128])
+        return f_weight
+
+    def forward(self, f_common):
+        f_shape = f_common.shape
+        # print('f_shape: ', f_shape)
+
+        f_theta = self.theta_conv(f_common)
+        f_phi = self.phi_conv(f_common)
+        f_g = self.g_conv(f_common)
+
+        ######## horizon ########
+        fh_weight = self._cal_fweight(
+            [f_theta, f_phi, f_g], [f_shape[0], f_shape[2], f_shape[3]]
+        )
+        fh_weight = paddle.transpose(fh_weight, [0, 3, 1, 2])
+        fh_weight = self.fh_weight_conv(fh_weight)
+        # short cut
+        fh_sc = self.fh_sc_conv(f_common)
+        f_h = F.relu(fh_weight + fh_sc)
+
+        ######## vertical ########
+        fv_theta = paddle.transpose(f_theta, [0, 1, 3, 2])
+        fv_phi = paddle.transpose(f_phi, [0, 1, 3, 2])
+        fv_g = paddle.transpose(f_g, [0, 1, 3, 2])
+        fv_weight = self._cal_fweight(
+            [fv_theta, fv_phi, fv_g], [f_shape[0], f_shape[3], f_shape[2]]
+        )
+        fv_weight = paddle.transpose(fv_weight, [0, 3, 2, 1])
+        fv_weight = self.fv_weight_conv(fv_weight)
+        # short cut
+        fv_sc = self.fv_sc_conv(f_common)
+        f_v = F.relu(fv_weight + fv_sc)
+
+        ######## merge ########
+        f_attn = paddle.concat([f_h, f_v], axis=1)
+        f_attn = self.f_attn_conv(f_attn)
+        return f_attn
+
+
+class SASTFPN(nn.Layer):
+    def __init__(self, in_channels, with_cab=False, **kwargs):
+        super(SASTFPN, self).__init__()
+        self.in_channels = in_channels
+        self.with_cab = with_cab
+        self.FPN_Down_Fusion = FPN_Down_Fusion(self.in_channels)
+        self.FPN_Up_Fusion = FPN_Up_Fusion(self.in_channels)
+        self.out_channels = 128
+        self.cross_attention = Cross_Attention(self.out_channels)
+
+    def forward(self, x):
+        # down fpn
+        f_down = self.FPN_Down_Fusion(x)
+
+        # up fpn
+        f_up = self.FPN_Up_Fusion(x)
+
+        # fusion
+        f_common = paddle.add(x=f_down, y=f_up)
+        f_common = F.relu(f_common)
+
+        if self.with_cab:
+            # print('enhence f_common with CAB.')
+            f_common = self.cross_attention(f_common)
+
+        return f_common
diff --git a/ppocr/modeling/necks/table_fpn.py b/ppocr/modeling/necks/table_fpn.py
new file mode 100644
index 0000000..d273965
--- /dev/null
+++ b/ppocr/modeling/necks/table_fpn.py
@@ -0,0 +1,123 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import paddle
+from paddle import nn
+import paddle.nn.functional as F
+from paddle import ParamAttr
+
+
+class TableFPN(nn.Layer):
+    def __init__(self, in_channels, out_channels, **kwargs):
+        super(TableFPN, self).__init__()
+        self.out_channels = 512
+        weight_attr = paddle.nn.initializer.KaimingUniform()
+        self.in2_conv = nn.Conv2D(
+            in_channels=in_channels[0],
+            out_channels=self.out_channels,
+            kernel_size=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+        self.in3_conv = nn.Conv2D(
+            in_channels=in_channels[1],
+            out_channels=self.out_channels,
+            kernel_size=1,
+            stride=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+        self.in4_conv = nn.Conv2D(
+            in_channels=in_channels[2],
+            out_channels=self.out_channels,
+            kernel_size=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+        self.in5_conv = nn.Conv2D(
+            in_channels=in_channels[3],
+            out_channels=self.out_channels,
+            kernel_size=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+        self.p5_conv = nn.Conv2D(
+            in_channels=self.out_channels,
+            out_channels=self.out_channels // 4,
+            kernel_size=3,
+            padding=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+        self.p4_conv = nn.Conv2D(
+            in_channels=self.out_channels,
+            out_channels=self.out_channels // 4,
+            kernel_size=3,
+            padding=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+        self.p3_conv = nn.Conv2D(
+            in_channels=self.out_channels,
+            out_channels=self.out_channels // 4,
+            kernel_size=3,
+            padding=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+        self.p2_conv = nn.Conv2D(
+            in_channels=self.out_channels,
+            out_channels=self.out_channels // 4,
+            kernel_size=3,
+            padding=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+        self.fuse_conv = nn.Conv2D(
+            in_channels=self.out_channels * 4,
+            out_channels=512,
+            kernel_size=3,
+            padding=1,
+            weight_attr=ParamAttr(initializer=weight_attr),
+            bias_attr=False,
+        )
+
+    def forward(self, x):
+        c2, c3, c4, c5 = x
+
+        in5 = self.in5_conv(c5)
+        in4 = self.in4_conv(c4)
+        in3 = self.in3_conv(c3)
+        in2 = self.in2_conv(c2)
+
+        out4 = in4 + F.upsample(
+            in5, size=in4.shape[2:4], mode="nearest", align_mode=1
+        )  # 1/16
+        out3 = in3 + F.upsample(
+            out4, size=in3.shape[2:4], mode="nearest", align_mode=1
+        )  # 1/8
+        out2 = in2 + F.upsample(
+            out3, size=in2.shape[2:4], mode="nearest", align_mode=1
+        )  # 1/4
+
+        p4 = F.upsample(out4, size=in5.shape[2:4], mode="nearest", align_mode=1)
+        p3 = F.upsample(out3, size=in5.shape[2:4], mode="nearest", align_mode=1)
+        p2 = F.upsample(out2, size=in5.shape[2:4], mode="nearest", align_mode=1)
+        fuse = paddle.concat([in5, p4, p3, p2], axis=1)
+        fuse_conv = self.fuse_conv(fuse) * 0.005
+        return [c5 + fuse_conv]
diff --git a/ppocr/modeling/transforms/__init__.py b/ppocr/modeling/transforms/__init__.py
new file mode 100755
index 0000000..24aeca3
--- /dev/null
+++ b/ppocr/modeling/transforms/__init__.py
@@ -0,0 +1,32 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+__all__ = ["build_transform"]
+
+
+def build_transform(config):
+    from .tps import TPS
+    from .stn import STN_ON
+    from .tsrn import TSRN
+    from .tbsrn import TBSRN
+    from .gaspin_transformer import GA_SPIN_Transformer as GA_SPIN
+
+    support_dict = ["TPS", "STN_ON", "GA_SPIN", "TSRN", "TBSRN"]
+
+    module_name = config.pop("name")
+    assert module_name in support_dict, Exception(
+        "transform only support {}".format(support_dict)
+    )
+    module_class = eval(module_name)(**config)
+    return module_class
diff --git a/ppocr/modeling/transforms/gaspin_transformer.py b/ppocr/modeling/transforms/gaspin_transformer.py
new file mode 100644
index 0000000..22e9242
--- /dev/null
+++ b/ppocr/modeling/transforms/gaspin_transformer.py
@@ -0,0 +1,319 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+from paddle import nn, ParamAttr
+from paddle.nn import functional as F
+import numpy as np
+import functools
+from .tps import GridGenerator
+
+"""This code is refer from:
+https://github.com/hikopensource/DAVAR-Lab-OCR/davarocr/davar_rcg/models/transformations/gaspin_transformation.py
+"""
+
+
+class SP_TransformerNetwork(nn.Layer):
+    """
+    Sturture-Preserving Transformation (SPT) as Equa. (2) in Ref. [1]
+    Ref: [1] SPIN: Structure-Preserving Inner Offset Network for Scene Text Recognition. AAAI-2021.
+    """
+
+    def __init__(self, nc=1, default_type=5):
+        """Based on SPIN
+        Args:
+            nc (int): number of input channels (usually in 1 or 3)
+            default_type (int): the complexity of transformation intensities (by default set to 6 as the paper)
+        """
+        super(SP_TransformerNetwork, self).__init__()
+        self.power_list = self.cal_K(default_type)
+        self.sigmoid = nn.Sigmoid()
+        self.bn = nn.InstanceNorm2D(nc)
+
+    def cal_K(self, k=5):
+        """
+
+        Args:
+            k (int): the complexity of transformation intensities (by default set to 6 as the paper)
+
+        Returns:
+            List: the normalized intensity of each pixel in [0,1], denoted as \beta [1x(2K+1)]
+
+        """
+        from math import log
+
+        x = []
+        if k != 0:
+            for i in range(1, k + 1):
+                lower = round(
+                    log(1 - (0.5 / (k + 1)) * i) / log((0.5 / (k + 1)) * i), 2
+                )
+                upper = round(1 / lower, 2)
+                x.append(lower)
+                x.append(upper)
+        x.append(1.00)
+        return x
+
+    def forward(self, batch_I, weights, offsets, lambda_color=None):
+        """
+
+        Args:
+            batch_I (Tensor): batch of input images [batch_size x nc x I_height x I_width]
+            weights:
+            offsets: the predicted offset by AIN, a scalar
+            lambda_color: the learnable update gate \alpha in Equa. (5) as
+                          g(x) = (1 - \alpha) \odot x + \alpha \odot x_{offsets}
+
+        Returns:
+            Tensor: transformed images by SPN as Equa. (4) in Ref. [1]
+                        [batch_size x I_channel_num x I_r_height x I_r_width]
+
+        """
+        batch_I = (batch_I + 1) * 0.5
+        if offsets is not None:
+            batch_I = batch_I * (1 - lambda_color) + offsets * lambda_color
+        batch_weight_params = paddle.unsqueeze(paddle.unsqueeze(weights, -1), -1)
+        batch_I_power = paddle.stack([batch_I.pow(p) for p in self.power_list], axis=1)
+
+        batch_weight_sum = paddle.sum(batch_I_power * batch_weight_params, axis=1)
+        batch_weight_sum = self.bn(batch_weight_sum)
+        batch_weight_sum = self.sigmoid(batch_weight_sum)
+        batch_weight_sum = batch_weight_sum * 2 - 1
+        return batch_weight_sum
+
+
+class GA_SPIN_Transformer(nn.Layer):
+    """
+    Geometric-Absorbed SPIN Transformation (GA-SPIN) proposed in Ref. [1]
+
+
+    Ref: [1] SPIN: Structure-Preserving Inner Offset Network for Scene Text Recognition. AAAI-2021.
+    """
+
+    def __init__(
+        self,
+        in_channels=1,
+        I_r_size=(32, 100),
+        offsets=False,
+        norm_type="BN",
+        default_type=6,
+        loc_lr=1,
+        stn=True,
+    ):
+        """
+        Args:
+            in_channels (int): channel of input features,
+                                set it to 1 if the grayscale images and 3 if RGB input
+            I_r_size (tuple): size of rectified images (used in STN transformations)
+            offsets (bool): set it to False if use SPN w.o. AIN,
+                            and set it to True if use SPIN (both with SPN and AIN)
+            norm_type (str): the normalization type of the module,
+                            set it to 'BN' by default, 'IN' optionally
+            default_type (int): the K chromatic space,
+                                set it to 3/5/6 depend on the complexity of transformation intensities
+            loc_lr (float): learning rate of location network
+            stn (bool): whether to use stn.
+
+        """
+        super(GA_SPIN_Transformer, self).__init__()
+        self.nc = in_channels
+        self.spt = True
+        self.offsets = offsets
+        self.stn = stn  # set to True in GA-SPIN, while set it to False in SPIN
+        self.I_r_size = I_r_size
+        self.out_channels = in_channels
+        if norm_type == "BN":
+            norm_layer = functools.partial(nn.BatchNorm2D, use_global_stats=True)
+        elif norm_type == "IN":
+            norm_layer = functools.partial(
+                nn.InstanceNorm2D, weight_attr=False, use_global_stats=False
+            )
+        else:
+            raise NotImplementedError(
+                "normalization layer [%s] is not found" % norm_type
+            )
+
+        if self.spt:
+            self.sp_net = SP_TransformerNetwork(in_channels, default_type)
+            self.spt_convnet = nn.Sequential(
+                # 32*100
+                nn.Conv2D(in_channels, 32, 3, 1, 1, bias_attr=False),
+                norm_layer(32),
+                nn.ReLU(),
+                nn.MaxPool2D(kernel_size=2, stride=2),
+                # 16*50
+                nn.Conv2D(32, 64, 3, 1, 1, bias_attr=False),
+                norm_layer(64),
+                nn.ReLU(),
+                nn.MaxPool2D(kernel_size=2, stride=2),
+                # 8*25
+                nn.Conv2D(64, 128, 3, 1, 1, bias_attr=False),
+                norm_layer(128),
+                nn.ReLU(),
+                nn.MaxPool2D(kernel_size=2, stride=2),
+                # 4*12
+            )
+            self.stucture_fc1 = nn.Sequential(
+                nn.Conv2D(128, 256, 3, 1, 1, bias_attr=False),
+                norm_layer(256),
+                nn.ReLU(),
+                nn.MaxPool2D(kernel_size=2, stride=2),
+                nn.Conv2D(256, 256, 3, 1, 1, bias_attr=False),
+                norm_layer(256),
+                nn.ReLU(),  # 2*6
+                nn.MaxPool2D(kernel_size=2, stride=2),
+                nn.Conv2D(256, 512, 3, 1, 1, bias_attr=False),
+                norm_layer(512),
+                nn.ReLU(),  # 1*3
+                nn.AdaptiveAvgPool2D(1),
+                nn.Flatten(1, -1),  # batch_size x 512
+                nn.Linear(512, 256, weight_attr=nn.initializer.Normal(0.001)),
+                nn.BatchNorm1D(256),
+                nn.ReLU(),
+            )
+            self.out_weight = 2 * default_type + 1
+            self.spt_length = 2 * default_type + 1
+            if offsets:
+                self.out_weight += 1
+            if self.stn:
+                self.F = 20
+                self.out_weight += self.F * 2
+                self.GridGenerator = GridGenerator(self.F * 2, self.F)
+
+            # self.out_weight*=nc
+            # Init structure_fc2 in LocalizationNetwork
+            initial_bias = self.init_spin(default_type * 2)
+            initial_bias = initial_bias.reshape(-1)
+            param_attr = ParamAttr(
+                learning_rate=loc_lr,
+                initializer=nn.initializer.Assign(np.zeros([256, self.out_weight])),
+            )
+            bias_attr = ParamAttr(
+                learning_rate=loc_lr, initializer=nn.initializer.Assign(initial_bias)
+            )
+            self.stucture_fc2 = nn.Linear(
+                256, self.out_weight, weight_attr=param_attr, bias_attr=bias_attr
+            )
+            self.sigmoid = nn.Sigmoid()
+
+            if offsets:
+                self.offset_fc1 = nn.Sequential(
+                    nn.Conv2D(128, 16, 3, 1, 1, bias_attr=False),
+                    norm_layer(16),
+                    nn.ReLU(),
+                )
+                self.offset_fc2 = nn.Conv2D(16, in_channels, 3, 1, 1)
+                self.pool = nn.MaxPool2D(2, 2)
+
+    def init_spin(self, nz):
+        """
+        Args:
+            nz (int): number of paired \betas exponents, which means the value of K x 2
+
+        """
+        init_id = [0.00] * nz + [5.00]
+        if self.offsets:
+            init_id += [-5.00]
+            # init_id *=3
+        init = np.array(init_id)
+
+        if self.stn:
+            F = self.F
+            ctrl_pts_x = np.linspace(-1.0, 1.0, int(F / 2))
+            ctrl_pts_y_top = np.linspace(0.0, -1.0, num=int(F / 2))
+            ctrl_pts_y_bottom = np.linspace(1.0, 0.0, num=int(F / 2))
+            ctrl_pts_top = np.stack([ctrl_pts_x, ctrl_pts_y_top], axis=1)
+            ctrl_pts_bottom = np.stack([ctrl_pts_x, ctrl_pts_y_bottom], axis=1)
+            initial_bias = np.concatenate([ctrl_pts_top, ctrl_pts_bottom], axis=0)
+            initial_bias = initial_bias.reshape(-1)
+            init = np.concatenate([init, initial_bias], axis=0)
+        return init
+
+    def forward(self, x, return_weight=False):
+        """
+        Args:
+            x (Tensor): input image batch
+            return_weight (bool): set to False by default,
+                                  if set to True return the predicted offsets of AIN, denoted as x_{offsets}
+
+        Returns:
+            Tensor: rectified image [batch_size x I_channel_num x I_height x I_width], the same as the input size
+        """
+
+        if self.spt:
+            feat = self.spt_convnet(x)
+            fc1 = self.stucture_fc1(feat)
+            sp_weight_fusion = self.stucture_fc2(fc1)
+            sp_weight_fusion = sp_weight_fusion.reshape(
+                [x.shape[0], self.out_weight, 1]
+            )
+            if self.offsets:  # SPIN w. AIN
+                lambda_color = sp_weight_fusion[:, self.spt_length, 0]
+                lambda_color = (
+                    self.sigmoid(lambda_color).unsqueeze(-1).unsqueeze(-1).unsqueeze(-1)
+                )
+                sp_weight = sp_weight_fusion[:, : self.spt_length, :]
+                offsets = self.pool(self.offset_fc2(self.offset_fc1(feat)))
+
+                assert offsets.shape[2] == 2  # 2
+                assert offsets.shape[3] == 6  # 16
+                offsets = self.sigmoid(offsets)  # v12
+
+                if return_weight:
+                    return offsets
+                offsets = nn.functional.upsample(
+                    offsets, size=(x.shape[2], x.shape[3]), mode="bilinear"
+                )
+
+                if self.stn:
+                    batch_C_prime = sp_weight_fusion[
+                        :, (self.spt_length + 1) :, :
+                    ].reshape([x.shape[0], self.F, 2])
+                    build_P_prime = self.GridGenerator(batch_C_prime, self.I_r_size)
+                    build_P_prime_reshape = build_P_prime.reshape(
+                        [build_P_prime.shape[0], self.I_r_size[0], self.I_r_size[1], 2]
+                    )
+
+            else:  # SPIN w.o. AIN
+                sp_weight = sp_weight_fusion[:, : self.spt_length, :]
+                lambda_color, offsets = None, None
+
+                if self.stn:
+                    batch_C_prime = sp_weight_fusion[:, self.spt_length :, :].reshape(
+                        [x.shape[0], self.F, 2]
+                    )
+                    build_P_prime = self.GridGenerator(batch_C_prime, self.I_r_size)
+                    build_P_prime_reshape = build_P_prime.reshape(
+                        [build_P_prime.shape[0], self.I_r_size[0], self.I_r_size[1], 2]
+                    )
+
+            x = self.sp_net(x, sp_weight, offsets, lambda_color)
+            if self.stn:
+                is_fp16 = False
+                if build_P_prime_reshape.dtype != paddle.float32:
+                    data_type = build_P_prime_reshape.dtype
+                    x = x.cast(paddle.float32)
+                    build_P_prime_reshape = build_P_prime_reshape.cast(paddle.float32)
+                    is_fp16 = True
+                x = F.grid_sample(
+                    x=x, grid=build_P_prime_reshape, padding_mode="border"
+                )
+                if is_fp16:
+                    x = x.cast(data_type)
+        return x
diff --git a/ppocr/modeling/transforms/stn.py b/ppocr/modeling/transforms/stn.py
new file mode 100644
index 0000000..a721184
--- /dev/null
+++ b/ppocr/modeling/transforms/stn.py
@@ -0,0 +1,147 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/ayumiymk/aster.pytorch/blob/master/lib/models/stn_head.py
+"""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+from paddle import nn, ParamAttr
+from paddle.nn import functional as F
+import numpy as np
+
+from .tps_spatial_transformer import TPSSpatialTransformer
+
+
+def conv3x3_block(in_channels, out_channels, stride=1):
+    n = 3 * 3 * out_channels
+    w = math.sqrt(2.0 / n)
+    conv_layer = nn.Conv2D(
+        in_channels,
+        out_channels,
+        kernel_size=3,
+        stride=stride,
+        padding=1,
+        weight_attr=nn.initializer.Normal(mean=0.0, std=w),
+        bias_attr=nn.initializer.Constant(0),
+    )
+    block = nn.Sequential(conv_layer, nn.BatchNorm2D(out_channels), nn.ReLU())
+    return block
+
+
+class STN(nn.Layer):
+    def __init__(self, in_channels, num_ctrlpoints, activation="none"):
+        super(STN, self).__init__()
+        self.in_channels = in_channels
+        self.num_ctrlpoints = num_ctrlpoints
+        self.activation = activation
+        self.stn_convnet = nn.Sequential(
+            conv3x3_block(in_channels, 32),  # 32x64
+            nn.MaxPool2D(kernel_size=2, stride=2),
+            conv3x3_block(32, 64),  # 16x32
+            nn.MaxPool2D(kernel_size=2, stride=2),
+            conv3x3_block(64, 128),  # 8*16
+            nn.MaxPool2D(kernel_size=2, stride=2),
+            conv3x3_block(128, 256),  # 4*8
+            nn.MaxPool2D(kernel_size=2, stride=2),
+            conv3x3_block(256, 256),  # 2*4,
+            nn.MaxPool2D(kernel_size=2, stride=2),
+            conv3x3_block(256, 256),
+        )  # 1*2
+        self.stn_fc1 = nn.Sequential(
+            nn.Linear(
+                2 * 256,
+                512,
+                weight_attr=nn.initializer.Normal(0, 0.001),
+                bias_attr=nn.initializer.Constant(0),
+            ),
+            nn.BatchNorm1D(512),
+            nn.ReLU(),
+        )
+        fc2_bias = self.init_stn()
+        self.stn_fc2 = nn.Linear(
+            512,
+            num_ctrlpoints * 2,
+            weight_attr=nn.initializer.Constant(0.0),
+            bias_attr=nn.initializer.Assign(fc2_bias),
+        )
+
+    def init_stn(self):
+        margin = 0.01
+        sampling_num_per_side = int(self.num_ctrlpoints / 2)
+        ctrl_pts_x = np.linspace(margin, 1.0 - margin, sampling_num_per_side)
+        ctrl_pts_y_top = np.ones(sampling_num_per_side) * margin
+        ctrl_pts_y_bottom = np.ones(sampling_num_per_side) * (1 - margin)
+        ctrl_pts_top = np.stack([ctrl_pts_x, ctrl_pts_y_top], axis=1)
+        ctrl_pts_bottom = np.stack([ctrl_pts_x, ctrl_pts_y_bottom], axis=1)
+        ctrl_points = np.concatenate([ctrl_pts_top, ctrl_pts_bottom], axis=0).astype(
+            np.float32
+        )
+        if self.activation == "none":
+            pass
+        elif self.activation == "sigmoid":
+            ctrl_points = -np.log(1.0 / ctrl_points - 1.0)
+        ctrl_points = paddle.to_tensor(ctrl_points)
+        fc2_bias = paddle.reshape(
+            ctrl_points, shape=[ctrl_points.shape[0] * ctrl_points.shape[1]]
+        )
+        return fc2_bias
+
+    def forward(self, x):
+        x = self.stn_convnet(x)
+        batch_size, _, h, w = x.shape
+        x = paddle.reshape(x, shape=(batch_size, -1))
+        img_feat = self.stn_fc1(x)
+        x = self.stn_fc2(0.1 * img_feat)
+        if self.activation == "sigmoid":
+            x = F.sigmoid(x)
+        x = paddle.reshape(x, shape=[-1, self.num_ctrlpoints, 2])
+        return img_feat, x
+
+
+class STN_ON(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        tps_inputsize,
+        tps_outputsize,
+        num_control_points,
+        tps_margins,
+        stn_activation,
+    ):
+        super(STN_ON, self).__init__()
+        self.tps = TPSSpatialTransformer(
+            output_image_size=tuple(tps_outputsize),
+            num_control_points=num_control_points,
+            margins=tuple(tps_margins),
+        )
+        self.stn_head = STN(
+            in_channels=in_channels,
+            num_ctrlpoints=num_control_points,
+            activation=stn_activation,
+        )
+        self.tps_inputsize = tps_inputsize
+        self.out_channels = in_channels
+
+    def forward(self, image):
+        stn_input = paddle.nn.functional.interpolate(
+            image, self.tps_inputsize, mode="bilinear", align_corners=True
+        )
+        stn_img_feat, ctrl_points = self.stn_head(stn_input)
+        x, _ = self.tps(image, ctrl_points)
+        return x
diff --git a/ppocr/modeling/transforms/tbsrn.py b/ppocr/modeling/transforms/tbsrn.py
new file mode 100644
index 0000000..a1fa167
--- /dev/null
+++ b/ppocr/modeling/transforms/tbsrn.py
@@ -0,0 +1,298 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/FudanVI/FudanOCR/blob/main/scene-text-telescope/model/tbsrn.py
+"""
+
+import math
+import warnings
+import numpy as np
+import paddle
+from paddle import nn
+import string
+
+warnings.filterwarnings("ignore")
+
+from .tps_spatial_transformer import TPSSpatialTransformer
+from .stn import STN as STNHead
+from .tsrn import GruBlock, mish, UpsampleBLock
+from ppocr.modeling.heads.sr_rensnet_transformer import (
+    Transformer,
+    LayerNorm,
+    PositionwiseFeedForward,
+    MultiHeadedAttention,
+)
+
+
+def positionalencoding2d(d_model, height, width):
+    """
+    :param d_model: dimension of the model
+    :param height: height of the positions
+    :param width: width of the positions
+    :return: d_model*height*width position matrix
+    """
+    if d_model % 4 != 0:
+        raise ValueError(
+            "Cannot use sin/cos positional encoding with "
+            "odd dimension (got dim={:d})".format(d_model)
+        )
+    pe = paddle.zeros([d_model, height, width])
+    # Each dimension use half of d_model
+    d_model = int(d_model / 2)
+    div_term = paddle.exp(
+        paddle.arange(0.0, d_model, 2, dtype="int64") * -(math.log(10000.0) / d_model)
+    )
+    pos_w = paddle.arange(0.0, width, dtype="float32").unsqueeze(1)
+    pos_h = paddle.arange(0.0, height, dtype="float32").unsqueeze(1)
+
+    pe[0:d_model:2, :, :] = (
+        paddle.sin(pos_w * div_term).transpose([1, 0]).unsqueeze(1).tile([1, height, 1])
+    )
+    pe[1:d_model:2, :, :] = (
+        paddle.cos(pos_w * div_term).transpose([1, 0]).unsqueeze(1).tile([1, height, 1])
+    )
+    pe[d_model::2, :, :] = (
+        paddle.sin(pos_h * div_term).transpose([1, 0]).unsqueeze(2).tile([1, 1, width])
+    )
+    pe[d_model + 1 :: 2, :, :] = (
+        paddle.cos(pos_h * div_term).transpose([1, 0]).unsqueeze(2).tile([1, 1, width])
+    )
+
+    return pe
+
+
+class FeatureEnhancer(nn.Layer):
+    def __init__(self):
+        super(FeatureEnhancer, self).__init__()
+
+        self.multihead = MultiHeadedAttention(h=4, d_model=128, dropout=0.1)
+        self.mul_layernorm1 = LayerNorm(features=128)
+
+        self.pff = PositionwiseFeedForward(128, 128)
+        self.mul_layernorm3 = LayerNorm(features=128)
+
+        self.linear = nn.Linear(128, 64)
+
+    def forward(self, conv_feature):
+        """
+        text : (batch, seq_len, embedding_size)
+        global_info: (batch, embedding_size, 1, 1)
+        conv_feature: (batch, channel, H, W)
+        """
+        batch = conv_feature.shape[0]
+        position2d = (
+            positionalencoding2d(64, 16, 64)
+            .cast("float32")
+            .unsqueeze(0)
+            .reshape([1, 64, 1024])
+        )
+        position2d = position2d.tile([batch, 1, 1])
+        conv_feature = paddle.concat(
+            [conv_feature, position2d], 1
+        )  # batch, 128(64+64), 32, 128
+        result = conv_feature.transpose([0, 2, 1])
+        origin_result = result
+        result = self.mul_layernorm1(
+            origin_result + self.multihead(result, result, result, mask=None)[0]
+        )
+        origin_result = result
+        result = self.mul_layernorm3(origin_result + self.pff(result))
+        result = self.linear(result)
+        return result.transpose([0, 2, 1])
+
+
+def str_filt(str_, voc_type):
+    alpha_dict = {
+        "digit": string.digits,
+        "lower": string.digits + string.ascii_lowercase,
+        "upper": string.digits + string.ascii_letters,
+        "all": string.digits + string.ascii_letters + string.punctuation,
+    }
+    if voc_type == "lower":
+        str_ = str_.lower()
+    for char in str_:
+        if char not in alpha_dict[voc_type]:
+            str_ = str_.replace(char, "")
+    str_ = str_.lower()
+    return str_
+
+
+class TBSRN(nn.Layer):
+    def __init__(
+        self,
+        in_channels=3,
+        scale_factor=2,
+        width=128,
+        height=32,
+        STN=True,
+        srb_nums=5,
+        mask=False,
+        hidden_units=32,
+        infer_mode=False,
+    ):
+        super(TBSRN, self).__init__()
+        in_planes = 3
+        if mask:
+            in_planes = 4
+        assert math.log(scale_factor, 2) % 1 == 0
+        upsample_block_num = int(math.log(scale_factor, 2))
+        self.block1 = nn.Sequential(
+            nn.Conv2D(in_planes, 2 * hidden_units, kernel_size=9, padding=4),
+            nn.PReLU(),
+            # nn.ReLU()
+        )
+        self.srb_nums = srb_nums
+        for i in range(srb_nums):
+            setattr(self, "block%d" % (i + 2), RecurrentResidualBlock(2 * hidden_units))
+
+        setattr(
+            self,
+            "block%d" % (srb_nums + 2),
+            nn.Sequential(
+                nn.Conv2D(2 * hidden_units, 2 * hidden_units, kernel_size=3, padding=1),
+                nn.BatchNorm2D(2 * hidden_units),
+            ),
+        )
+
+        # self.non_local = NonLocalBlock2D(64, 64)
+        block_ = [UpsampleBLock(2 * hidden_units, 2) for _ in range(upsample_block_num)]
+        block_.append(nn.Conv2D(2 * hidden_units, in_planes, kernel_size=9, padding=4))
+        setattr(self, "block%d" % (srb_nums + 3), nn.Sequential(*block_))
+        self.tps_inputsize = [height // scale_factor, width // scale_factor]
+        tps_outputsize = [height // scale_factor, width // scale_factor]
+        num_control_points = 20
+        tps_margins = [0.05, 0.05]
+        self.stn = STN
+        self.out_channels = in_channels
+        if self.stn:
+            self.tps = TPSSpatialTransformer(
+                output_image_size=tuple(tps_outputsize),
+                num_control_points=num_control_points,
+                margins=tuple(tps_margins),
+            )
+
+            self.stn_head = STNHead(
+                in_channels=in_planes,
+                num_ctrlpoints=num_control_points,
+                activation="none",
+            )
+        self.infer_mode = infer_mode
+
+        self.english_alphabet = (
+            "-0123456789abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ"
+        )
+        self.english_dict = {}
+        for index in range(len(self.english_alphabet)):
+            self.english_dict[self.english_alphabet[index]] = index
+        transformer = Transformer(alphabet="-0123456789abcdefghijklmnopqrstuvwxyz")
+        self.transformer = transformer
+        for param in self.transformer.parameters():
+            param.trainable = False
+
+    def label_encoder(self, label):
+        batch = len(label)
+
+        length = [len(i) for i in label]
+        length_tensor = paddle.to_tensor(length, dtype="int64")
+
+        max_length = max(length)
+        input_tensor = np.zeros((batch, max_length))
+        for i in range(batch):
+            for j in range(length[i] - 1):
+                input_tensor[i][j + 1] = self.english_dict[label[i][j]]
+
+        text_gt = []
+        for i in label:
+            for j in i:
+                text_gt.append(self.english_dict[j])
+        text_gt = paddle.to_tensor(text_gt, dtype="int64")
+
+        input_tensor = paddle.to_tensor(input_tensor, dtype="int64")
+        return length_tensor, input_tensor, text_gt
+
+    def forward(self, x):
+        output = {}
+        if self.infer_mode:
+            output["lr_img"] = x
+            y = x
+        else:
+            output["lr_img"] = x[0]
+            output["hr_img"] = x[1]
+            y = x[0]
+        if self.stn and self.training:
+            _, ctrl_points_x = self.stn_head(y)
+            y, _ = self.tps(y, ctrl_points_x)
+        block = {"1": self.block1(y)}
+        for i in range(self.srb_nums + 1):
+            block[str(i + 2)] = getattr(self, "block%d" % (i + 2))(block[str(i + 1)])
+
+        block[str(self.srb_nums + 3)] = getattr(self, "block%d" % (self.srb_nums + 3))(
+            (block["1"] + block[str(self.srb_nums + 2)])
+        )
+
+        sr_img = paddle.tanh(block[str(self.srb_nums + 3)])
+        output["sr_img"] = sr_img
+
+        if self.training:
+            hr_img = x[1]
+
+            # add transformer
+            label = [str_filt(i, "lower") + "-" for i in x[2]]
+            length_tensor, input_tensor, text_gt = self.label_encoder(label)
+            hr_pred, word_attention_map_gt, hr_correct_list = self.transformer(
+                hr_img, length_tensor, input_tensor
+            )
+            sr_pred, word_attention_map_pred, sr_correct_list = self.transformer(
+                sr_img, length_tensor, input_tensor
+            )
+            output["hr_img"] = hr_img
+            output["hr_pred"] = hr_pred
+            output["text_gt"] = text_gt
+            output["word_attention_map_gt"] = word_attention_map_gt
+            output["sr_pred"] = sr_pred
+            output["word_attention_map_pred"] = word_attention_map_pred
+
+        return output
+
+
+class RecurrentResidualBlock(nn.Layer):
+    def __init__(self, channels):
+        super(RecurrentResidualBlock, self).__init__()
+        self.conv1 = nn.Conv2D(channels, channels, kernel_size=3, padding=1)
+        self.bn1 = nn.BatchNorm2D(channels)
+        self.gru1 = GruBlock(channels, channels)
+        # self.prelu = nn.ReLU()
+        self.prelu = mish()
+        self.conv2 = nn.Conv2D(channels, channels, kernel_size=3, padding=1)
+        self.bn2 = nn.BatchNorm2D(channels)
+        self.gru2 = GruBlock(channels, channels)
+        self.feature_enhancer = FeatureEnhancer()
+
+        for p in self.parameters():
+            if p.dim() > 1:
+                paddle.nn.initializer.XavierUniform(p)
+
+    def forward(self, x):
+        residual = self.conv1(x)
+        residual = self.bn1(residual)
+        residual = self.prelu(residual)
+        residual = self.conv2(residual)
+        residual = self.bn2(residual)
+
+        size = residual.shape
+        residual = residual.reshape([size[0], size[1], -1])
+        residual = self.feature_enhancer(residual)
+        residual = residual.reshape([size[0], size[1], size[2], size[3]])
+        return x + residual
diff --git a/ppocr/modeling/transforms/tps.py b/ppocr/modeling/transforms/tps.py
new file mode 100644
index 0000000..d5681db
--- /dev/null
+++ b/ppocr/modeling/transforms/tps.py
@@ -0,0 +1,321 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/clovaai/deep-text-recognition-benchmark/blob/master/modules/transformation.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+from paddle import nn, ParamAttr
+from paddle.nn import functional as F
+import numpy as np
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        kernel_size,
+        stride=1,
+        groups=1,
+        act=None,
+        name=None,
+    ):
+        super(ConvBNLayer, self).__init__()
+        self.conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=(kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(name=name + "_weights"),
+            bias_attr=False,
+        )
+        bn_name = "bn_" + name
+        self.bn = nn.BatchNorm(
+            out_channels,
+            act=act,
+            param_attr=ParamAttr(name=bn_name + "_scale"),
+            bias_attr=ParamAttr(bn_name + "_offset"),
+            moving_mean_name=bn_name + "_mean",
+            moving_variance_name=bn_name + "_variance",
+        )
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        return x
+
+
+class LocalizationNetwork(nn.Layer):
+    def __init__(self, in_channels, num_fiducial, loc_lr, model_name):
+        super(LocalizationNetwork, self).__init__()
+        self.F = num_fiducial
+        F = num_fiducial
+        if model_name == "large":
+            num_filters_list = [64, 128, 256, 512]
+            fc_dim = 256
+        else:
+            num_filters_list = [16, 32, 64, 128]
+            fc_dim = 64
+
+        self.block_list = []
+        for fno in range(0, len(num_filters_list)):
+            num_filters = num_filters_list[fno]
+            name = "loc_conv%d" % fno
+            conv = self.add_sublayer(
+                name,
+                ConvBNLayer(
+                    in_channels=in_channels,
+                    out_channels=num_filters,
+                    kernel_size=3,
+                    act="relu",
+                    name=name,
+                ),
+            )
+            self.block_list.append(conv)
+            if fno == len(num_filters_list) - 1:
+                pool = nn.AdaptiveAvgPool2D(1)
+            else:
+                pool = nn.MaxPool2D(kernel_size=2, stride=2, padding=0)
+            in_channels = num_filters
+            self.block_list.append(pool)
+        name = "loc_fc1"
+        stdv = 1.0 / math.sqrt(num_filters_list[-1] * 1.0)
+        self.fc1 = nn.Linear(
+            in_channels,
+            fc_dim,
+            weight_attr=ParamAttr(
+                learning_rate=loc_lr,
+                name=name + "_w",
+                initializer=nn.initializer.Uniform(-stdv, stdv),
+            ),
+            bias_attr=ParamAttr(name=name + ".b_0"),
+            name=name,
+        )
+
+        # Init fc2 in LocalizationNetwork
+        initial_bias = self.get_initial_fiducials()
+        initial_bias = initial_bias.reshape(-1)
+        name = "loc_fc2"
+        param_attr = ParamAttr(
+            learning_rate=loc_lr,
+            initializer=nn.initializer.Assign(np.zeros([fc_dim, F * 2])),
+            name=name + "_w",
+        )
+        bias_attr = ParamAttr(
+            learning_rate=loc_lr,
+            initializer=nn.initializer.Assign(initial_bias),
+            name=name + "_b",
+        )
+        self.fc2 = nn.Linear(
+            fc_dim, F * 2, weight_attr=param_attr, bias_attr=bias_attr, name=name
+        )
+        self.out_channels = F * 2
+
+    def forward(self, x):
+        """
+        Estimating parameters of geometric transformation
+        Args:
+            image: input
+        Return:
+            batch_C_prime: the matrix of the geometric transformation
+        """
+        B = x.shape[0]
+        i = 0
+        for block in self.block_list:
+            x = block(x)
+        x = x.squeeze(axis=2).squeeze(axis=2)
+        x = self.fc1(x)
+
+        x = F.relu(x)
+        x = self.fc2(x)
+        x = x.reshape(shape=[-1, self.F, 2])
+        return x
+
+    def get_initial_fiducials(self):
+        """see RARE paper Fig. 6 (a)"""
+        F = self.F
+        ctrl_pts_x = np.linspace(-1.0, 1.0, int(F / 2))
+        ctrl_pts_y_top = np.linspace(0.0, -1.0, num=int(F / 2))
+        ctrl_pts_y_bottom = np.linspace(1.0, 0.0, num=int(F / 2))
+        ctrl_pts_top = np.stack([ctrl_pts_x, ctrl_pts_y_top], axis=1)
+        ctrl_pts_bottom = np.stack([ctrl_pts_x, ctrl_pts_y_bottom], axis=1)
+        initial_bias = np.concatenate([ctrl_pts_top, ctrl_pts_bottom], axis=0)
+        return initial_bias
+
+
+class GridGenerator(nn.Layer):
+    def __init__(self, in_channels, num_fiducial):
+        super(GridGenerator, self).__init__()
+        self.eps = 1e-6
+        self.F = num_fiducial
+
+        name = "ex_fc"
+        initializer = nn.initializer.Constant(value=0.0)
+        param_attr = ParamAttr(
+            learning_rate=0.0, initializer=initializer, name=name + "_w"
+        )
+        bias_attr = ParamAttr(
+            learning_rate=0.0, initializer=initializer, name=name + "_b"
+        )
+        self.fc = nn.Linear(
+            in_channels, 6, weight_attr=param_attr, bias_attr=bias_attr, name=name
+        )
+
+    def forward(self, batch_C_prime, I_r_size):
+        """
+        Generate the grid for the grid_sampler.
+        Args:
+            batch_C_prime: the matrix of the geometric transformation
+            I_r_size: the shape of the input image
+        Return:
+            batch_P_prime: the grid for the grid_sampler
+        """
+        C = self.build_C_paddle()
+        P = self.build_P_paddle(I_r_size)
+
+        inv_delta_C_tensor = self.build_inv_delta_C_paddle(C).astype("float32")
+        P_hat_tensor = self.build_P_hat_paddle(C, paddle.to_tensor(P)).astype("float32")
+
+        inv_delta_C_tensor.stop_gradient = True
+        P_hat_tensor.stop_gradient = True
+
+        batch_C_ex_part_tensor = self.get_expand_tensor(batch_C_prime)
+
+        batch_C_ex_part_tensor.stop_gradient = True
+
+        batch_C_prime_with_zeros = paddle.concat(
+            [batch_C_prime, batch_C_ex_part_tensor], axis=1
+        )
+        batch_T = paddle.matmul(inv_delta_C_tensor, batch_C_prime_with_zeros)
+        batch_P_prime = paddle.matmul(P_hat_tensor, batch_T)
+        return batch_P_prime
+
+    def build_C_paddle(self):
+        """Return coordinates of fiducial points in I_r; C"""
+        F = self.F
+        ctrl_pts_x = paddle.linspace(-1.0, 1.0, int(F / 2), dtype="float64")
+        ctrl_pts_y_top = -1 * paddle.ones([int(F / 2)], dtype="float64")
+        ctrl_pts_y_bottom = paddle.ones([int(F / 2)], dtype="float64")
+        ctrl_pts_top = paddle.stack([ctrl_pts_x, ctrl_pts_y_top], axis=1)
+        ctrl_pts_bottom = paddle.stack([ctrl_pts_x, ctrl_pts_y_bottom], axis=1)
+        C = paddle.concat([ctrl_pts_top, ctrl_pts_bottom], axis=0)
+        return C  # F x 2
+
+    def build_P_paddle(self, I_r_size):
+        I_r_height, I_r_width = I_r_size
+        I_r_grid_x = (
+            paddle.arange(-I_r_width, I_r_width, 2, dtype="float64") + 1.0
+        ) / paddle.to_tensor(np.array([I_r_width])).astype("float64")
+
+        I_r_grid_y = (
+            paddle.arange(-I_r_height, I_r_height, 2, dtype="float64") + 1.0
+        ) / paddle.to_tensor(np.array([I_r_height])).astype("float64")
+
+        # P: self.I_r_width x self.I_r_height x 2
+        P = paddle.stack(paddle.meshgrid(I_r_grid_x, I_r_grid_y), axis=2)
+        P = paddle.transpose(P, perm=[1, 0, 2])
+        # n (= self.I_r_width x self.I_r_height) x 2
+        return P.reshape([-1, 2])
+
+    def build_inv_delta_C_paddle(self, C):
+        """Return inv_delta_C which is needed to calculate T"""
+        F = self.F
+        hat_eye = paddle.eye(F, dtype="float64")  # F x F
+        hat_C = (
+            paddle.norm(C.reshape([1, F, 2]) - C.reshape([F, 1, 2]), axis=2) + hat_eye
+        )
+        hat_C = (hat_C**2) * paddle.log(hat_C)
+        delta_C = paddle.concat(  # F+3 x F+3
+            [
+                paddle.concat(
+                    [paddle.ones((F, 1), dtype="float64"), C, hat_C], axis=1
+                ),  # F x F+3
+                paddle.concat(
+                    [
+                        paddle.zeros((2, 3), dtype="float64"),
+                        paddle.transpose(C, perm=[1, 0]),
+                    ],
+                    axis=1,
+                ),  # 2 x F+3
+                paddle.concat(
+                    [
+                        paddle.zeros((1, 3), dtype="float64"),
+                        paddle.ones((1, F), dtype="float64"),
+                    ],
+                    axis=1,
+                ),  # 1 x F+3
+            ],
+            axis=0,
+        )
+        inv_delta_C = paddle.inverse(delta_C)
+        return inv_delta_C  # F+3 x F+3
+
+    def build_P_hat_paddle(self, C, P):
+        F = self.F
+        eps = self.eps
+        n = P.shape[0]  # n (= self.I_r_width x self.I_r_height)
+        # P_tile: n x 2 -> n x 1 x 2 -> n x F x 2
+        P_tile = paddle.tile(paddle.unsqueeze(P, axis=1), (1, F, 1))
+        C_tile = paddle.unsqueeze(C, axis=0)  # 1 x F x 2
+        P_diff = P_tile - C_tile  # n x F x 2
+        # rbf_norm: n x F
+        rbf_norm = paddle.norm(P_diff, p=2, axis=2, keepdim=False)
+
+        # rbf: n x F
+        rbf = paddle.multiply(paddle.square(rbf_norm), paddle.log(rbf_norm + eps))
+        P_hat = paddle.concat([paddle.ones((n, 1), dtype="float64"), P, rbf], axis=1)
+        return P_hat  # n x F+3
+
+    def get_expand_tensor(self, batch_C_prime):
+        B, H, C = batch_C_prime.shape
+        batch_C_prime = batch_C_prime.reshape([B, H * C])
+        batch_C_ex_part_tensor = self.fc(batch_C_prime)
+        batch_C_ex_part_tensor = batch_C_ex_part_tensor.reshape([-1, 3, 2])
+        return batch_C_ex_part_tensor
+
+
+class TPS(nn.Layer):
+    def __init__(self, in_channels, num_fiducial, loc_lr, model_name):
+        super(TPS, self).__init__()
+        self.loc_net = LocalizationNetwork(
+            in_channels, num_fiducial, loc_lr, model_name
+        )
+        self.grid_generator = GridGenerator(self.loc_net.out_channels, num_fiducial)
+        self.out_channels = in_channels
+
+    def forward(self, image):
+        image.stop_gradient = False
+        batch_C_prime = self.loc_net(image)
+        batch_P_prime = self.grid_generator(batch_C_prime, image.shape[2:])
+        batch_P_prime = batch_P_prime.reshape([-1, image.shape[2], image.shape[3], 2])
+        is_fp16 = False
+        if batch_P_prime.dtype != paddle.float32:
+            data_type = batch_P_prime.dtype
+            image = image.cast(paddle.float32)
+            batch_P_prime = batch_P_prime.cast(paddle.float32)
+            is_fp16 = True
+        batch_I_r = F.grid_sample(x=image, grid=batch_P_prime)
+        if is_fp16:
+            batch_I_r = batch_I_r.cast(data_type)
+
+        return batch_I_r
diff --git a/ppocr/modeling/transforms/tps_spatial_transformer.py b/ppocr/modeling/transforms/tps_spatial_transformer.py
new file mode 100644
index 0000000..fc230b0
--- /dev/null
+++ b/ppocr/modeling/transforms/tps_spatial_transformer.py
@@ -0,0 +1,170 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/ayumiymk/aster.pytorch/blob/master/lib/models/tps_spatial_transformer.py
+"""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import math
+import paddle
+from paddle import nn, ParamAttr
+from paddle.nn import functional as F
+import numpy as np
+import itertools
+
+
+def grid_sample(input, grid, canvas=None):
+    input.stop_gradient = False
+
+    is_fp16 = False
+    if grid.dtype != paddle.float32:
+        data_type = grid.dtype
+        input = input.cast(paddle.float32)
+        grid = grid.cast(paddle.float32)
+        is_fp16 = True
+    output = F.grid_sample(input, grid)
+    if is_fp16:
+        output = output.cast(data_type)
+        grid = grid.cast(data_type)
+
+    if canvas is None:
+        return output
+    else:
+        input_mask = paddle.ones(shape=input.shape)
+        if is_fp16:
+            input_mask = input_mask.cast(paddle.float32)
+            grid = grid.cast(paddle.float32)
+        output_mask = F.grid_sample(input_mask, grid)
+        if is_fp16:
+            output_mask = output_mask.cast(data_type)
+        padded_output = output * output_mask + canvas * (1 - output_mask)
+        return padded_output
+
+
+# phi(x1, x2) = r^2 * log(r), where r = ||x1 - x2||_2
+def compute_partial_repr(input_points, control_points):
+    N = input_points.shape[0]
+    M = control_points.shape[0]
+    pairwise_diff = paddle.reshape(input_points, shape=[N, 1, 2]) - paddle.reshape(
+        control_points, shape=[1, M, 2]
+    )
+    # original implementation, very slow
+    # pairwise_dist = torch.sum(pairwise_diff ** 2, dim = 2) # square of distance
+    pairwise_diff_square = pairwise_diff * pairwise_diff
+    pairwise_dist = pairwise_diff_square[:, :, 0] + pairwise_diff_square[:, :, 1]
+    repr_matrix = 0.5 * pairwise_dist * paddle.log(pairwise_dist)
+    # fix numerical error for 0 * log(0), substitute all nan with 0
+    mask = np.array(repr_matrix != repr_matrix)
+    repr_matrix[mask] = 0
+    return repr_matrix
+
+
+# output_ctrl_pts are specified, according to our task.
+def build_output_control_points(num_control_points, margins):
+    margin_x, margin_y = margins
+    num_ctrl_pts_per_side = num_control_points // 2
+    ctrl_pts_x = np.linspace(margin_x, 1.0 - margin_x, num_ctrl_pts_per_side)
+    ctrl_pts_y_top = np.ones(num_ctrl_pts_per_side) * margin_y
+    ctrl_pts_y_bottom = np.ones(num_ctrl_pts_per_side) * (1.0 - margin_y)
+    ctrl_pts_top = np.stack([ctrl_pts_x, ctrl_pts_y_top], axis=1)
+    ctrl_pts_bottom = np.stack([ctrl_pts_x, ctrl_pts_y_bottom], axis=1)
+    output_ctrl_pts_arr = np.concatenate([ctrl_pts_top, ctrl_pts_bottom], axis=0)
+    output_ctrl_pts = paddle.to_tensor(output_ctrl_pts_arr)
+    return output_ctrl_pts
+
+
+class TPSSpatialTransformer(nn.Layer):
+    def __init__(self, output_image_size=None, num_control_points=None, margins=None):
+        super(TPSSpatialTransformer, self).__init__()
+        self.output_image_size = output_image_size
+        self.num_control_points = num_control_points
+        self.margins = margins
+
+        self.target_height, self.target_width = output_image_size
+        target_control_points = build_output_control_points(num_control_points, margins)
+        N = num_control_points
+
+        # create padded kernel matrix
+        forward_kernel = paddle.zeros(shape=[N + 3, N + 3])
+        target_control_partial_repr = compute_partial_repr(
+            target_control_points, target_control_points
+        )
+        target_control_partial_repr = paddle.cast(
+            target_control_partial_repr, forward_kernel.dtype
+        )
+        forward_kernel[:N, :N] = target_control_partial_repr
+        forward_kernel[:N, -3] = 1
+        forward_kernel[-3, :N] = 1
+        target_control_points = paddle.cast(target_control_points, forward_kernel.dtype)
+        forward_kernel[:N, -2:] = target_control_points
+        forward_kernel[-2:, :N] = paddle.transpose(target_control_points, perm=[1, 0])
+        # compute inverse matrix
+        inverse_kernel = paddle.inverse(forward_kernel)
+
+        # create target coordinate matrix
+        HW = self.target_height * self.target_width
+        target_coordinate = list(
+            itertools.product(range(self.target_height), range(self.target_width))
+        )
+        target_coordinate = paddle.to_tensor(target_coordinate)  # HW x 2
+        Y, X = paddle.split(target_coordinate, target_coordinate.shape[1], axis=1)
+        Y = Y / (self.target_height - 1)
+        X = X / (self.target_width - 1)
+        target_coordinate = paddle.concat(
+            [X, Y], axis=1
+        )  # convert from (y, x) to (x, y)
+        target_coordinate_partial_repr = compute_partial_repr(
+            target_coordinate, target_control_points
+        )
+        target_coordinate_repr = paddle.concat(
+            [
+                target_coordinate_partial_repr,
+                paddle.ones(shape=[HW, 1]),
+                target_coordinate,
+            ],
+            axis=1,
+        )
+
+        # register precomputed matrices
+        self.inverse_kernel = inverse_kernel
+        self.padding_matrix = paddle.zeros(shape=[3, 2])
+        self.target_coordinate_repr = target_coordinate_repr
+        self.target_control_points = target_control_points
+
+    def forward(self, input, source_control_points):
+        assert source_control_points.ndimension() == 3
+        assert source_control_points.shape[1] == self.num_control_points
+        assert source_control_points.shape[2] == 2
+        batch_size = source_control_points.shape[0]
+
+        padding_matrix = paddle.expand(self.padding_matrix, shape=[batch_size, 3, 2])
+        Y = paddle.concat(
+            [source_control_points.astype(padding_matrix.dtype), padding_matrix], 1
+        )
+        mapping_matrix = paddle.matmul(self.inverse_kernel, Y)
+        source_coordinate = paddle.matmul(self.target_coordinate_repr, mapping_matrix)
+
+        grid = paddle.reshape(
+            source_coordinate, shape=[-1, self.target_height, self.target_width, 2]
+        )
+        grid = paddle.clip(
+            grid, 0, 1
+        )  # the source_control_points may be out of [0, 1].
+        # the input to grid_sample is normalized [-1, 1], but what we get is [0, 1]
+        grid = 2.0 * grid - 1.0
+        output_maps = grid_sample(input, grid, canvas=None)
+        return output_maps, source_coordinate
diff --git a/ppocr/modeling/transforms/tsrn.py b/ppocr/modeling/transforms/tsrn.py
new file mode 100644
index 0000000..93238fa
--- /dev/null
+++ b/ppocr/modeling/transforms/tsrn.py
@@ -0,0 +1,215 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/FudanVI/FudanOCR/blob/main/text-gestalt/model/tsrn.py
+"""
+
+import math
+import paddle
+import paddle.nn.functional as F
+from paddle import nn
+from collections import OrderedDict
+import sys
+import numpy as np
+import warnings
+import math, copy
+import cv2
+
+warnings.filterwarnings("ignore")
+
+from .tps_spatial_transformer import TPSSpatialTransformer
+from .stn import STN as STN_model
+from ppocr.modeling.heads.sr_rensnet_transformer import Transformer
+
+
+class TSRN(nn.Layer):
+    def __init__(
+        self,
+        in_channels,
+        scale_factor=2,
+        width=128,
+        height=32,
+        STN=False,
+        srb_nums=5,
+        mask=False,
+        hidden_units=32,
+        infer_mode=False,
+        **kwargs,
+    ):
+        super(TSRN, self).__init__()
+        in_planes = 3
+        if mask:
+            in_planes = 4
+        assert math.log(scale_factor, 2) % 1 == 0
+        upsample_block_num = int(math.log(scale_factor, 2))
+        self.block1 = nn.Sequential(
+            nn.Conv2D(in_planes, 2 * hidden_units, kernel_size=9, padding=4), nn.PReLU()
+        )
+        self.srb_nums = srb_nums
+        for i in range(srb_nums):
+            setattr(self, "block%d" % (i + 2), RecurrentResidualBlock(2 * hidden_units))
+
+        setattr(
+            self,
+            "block%d" % (srb_nums + 2),
+            nn.Sequential(
+                nn.Conv2D(2 * hidden_units, 2 * hidden_units, kernel_size=3, padding=1),
+                nn.BatchNorm2D(2 * hidden_units),
+            ),
+        )
+
+        block_ = [UpsampleBLock(2 * hidden_units, 2) for _ in range(upsample_block_num)]
+        block_.append(nn.Conv2D(2 * hidden_units, in_planes, kernel_size=9, padding=4))
+        setattr(self, "block%d" % (srb_nums + 3), nn.Sequential(*block_))
+        self.tps_inputsize = [height // scale_factor, width // scale_factor]
+        tps_outputsize = [height // scale_factor, width // scale_factor]
+        num_control_points = 20
+        tps_margins = [0.05, 0.05]
+        self.stn = STN
+        if self.stn:
+            self.tps = TPSSpatialTransformer(
+                output_image_size=tuple(tps_outputsize),
+                num_control_points=num_control_points,
+                margins=tuple(tps_margins),
+            )
+
+            self.stn_head = STN_model(
+                in_channels=in_planes,
+                num_ctrlpoints=num_control_points,
+                activation="none",
+            )
+        self.out_channels = in_channels
+
+        self.r34_transformer = Transformer()
+        for param in self.r34_transformer.parameters():
+            param.trainable = False
+        self.infer_mode = infer_mode
+
+    def forward(self, x):
+        output = {}
+        if self.infer_mode:
+            output["lr_img"] = x
+            y = x
+        else:
+            output["lr_img"] = x[0]
+            output["hr_img"] = x[1]
+            y = x[0]
+        if self.stn and self.training:
+            _, ctrl_points_x = self.stn_head(y)
+            y, _ = self.tps(y, ctrl_points_x)
+        block = {"1": self.block1(y)}
+        for i in range(self.srb_nums + 1):
+            block[str(i + 2)] = getattr(self, "block%d" % (i + 2))(block[str(i + 1)])
+
+        block[str(self.srb_nums + 3)] = getattr(self, "block%d" % (self.srb_nums + 3))(
+            (block["1"] + block[str(self.srb_nums + 2)])
+        )
+
+        sr_img = paddle.tanh(block[str(self.srb_nums + 3)])
+
+        output["sr_img"] = sr_img
+
+        if self.training:
+            hr_img = x[1]
+            length = x[2]
+            input_tensor = x[3]
+
+            # add transformer
+            sr_pred, word_attention_map_pred, _ = self.r34_transformer(
+                sr_img, length, input_tensor
+            )
+
+            hr_pred, word_attention_map_gt, _ = self.r34_transformer(
+                hr_img, length, input_tensor
+            )
+
+            output["hr_img"] = hr_img
+            output["hr_pred"] = hr_pred
+            output["word_attention_map_gt"] = word_attention_map_gt
+            output["sr_pred"] = sr_pred
+            output["word_attention_map_pred"] = word_attention_map_pred
+
+        return output
+
+
+class RecurrentResidualBlock(nn.Layer):
+    def __init__(self, channels):
+        super(RecurrentResidualBlock, self).__init__()
+        self.conv1 = nn.Conv2D(channels, channels, kernel_size=3, padding=1)
+        self.bn1 = nn.BatchNorm2D(channels)
+        self.gru1 = GruBlock(channels, channels)
+        self.prelu = mish()
+        self.conv2 = nn.Conv2D(channels, channels, kernel_size=3, padding=1)
+        self.bn2 = nn.BatchNorm2D(channels)
+        self.gru2 = GruBlock(channels, channels)
+
+    def forward(self, x):
+        residual = self.conv1(x)
+        residual = self.bn1(residual)
+        residual = self.prelu(residual)
+        residual = self.conv2(residual)
+        residual = self.bn2(residual)
+        residual = self.gru1(residual.transpose([0, 1, 3, 2])).transpose([0, 1, 3, 2])
+
+        return self.gru2(x + residual)
+
+
+class UpsampleBLock(nn.Layer):
+    def __init__(self, in_channels, up_scale):
+        super(UpsampleBLock, self).__init__()
+        self.conv = nn.Conv2D(
+            in_channels, in_channels * up_scale**2, kernel_size=3, padding=1
+        )
+
+        self.pixel_shuffle = nn.PixelShuffle(up_scale)
+        self.prelu = mish()
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.pixel_shuffle(x)
+        x = self.prelu(x)
+        return x
+
+
+class mish(nn.Layer):
+    def __init__(
+        self,
+    ):
+        super(mish, self).__init__()
+        self.activated = True
+
+    def forward(self, x):
+        if self.activated:
+            x = x * (paddle.tanh(F.softplus(x)))
+        return x
+
+
+class GruBlock(nn.Layer):
+    def __init__(self, in_channels, out_channels):
+        super(GruBlock, self).__init__()
+        assert out_channels % 2 == 0
+        self.conv1 = nn.Conv2D(in_channels, out_channels, kernel_size=1, padding=0)
+        self.gru = nn.GRU(out_channels, out_channels // 2, direction="bidirectional")
+
+    def forward(self, x):
+        # x: b, c, w, h
+        x = self.conv1(x)
+        x = x.transpose([0, 2, 3, 1])  # b, w, h, c
+        batch_size, w, h, c = x.shape
+        x = x.reshape([-1, h, c])  # b*w, h, c
+        x, _ = self.gru(x)
+        x = x.reshape([-1, w, h, c])
+        x = x.transpose([0, 3, 1, 2])
+        return x
diff --git a/ppocr/optimizer/__init__.py b/ppocr/optimizer/__init__.py
new file mode 100644
index 0000000..09b1d98
--- /dev/null
+++ b/ppocr/optimizer/__init__.py
@@ -0,0 +1,123 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+import copy
+import math
+import paddle
+
+__all__ = ["build_optimizer"]
+
+
+class CosineWeightDecayScheduler(object):
+    """Cosine-anneal the optimizer's weight decay each step.
+
+    wd(t) = end + 0.5 * (start - end) * (1 + cos(pi * t / T))
+
+    During warmup the coefficient is held at `start_factor`.
+    """
+
+    def __init__(
+        self, optimizer, start_factor, end_factor, total_steps, warmup_steps=0
+    ):
+        self.optimizer = optimizer
+        self.start_factor = start_factor
+        self.end_factor = end_factor
+        self.total_steps = total_steps
+        self.warmup_steps = warmup_steps
+        self._step = 0
+
+    def step(self):
+        self._step += 1
+        if self._step <= self.warmup_steps:
+            wd = self.start_factor
+        else:
+            progress = (self._step - self.warmup_steps) / max(
+                1, self.total_steps - self.warmup_steps
+            )
+            progress = min(progress, 1.0)
+            wd = self.end_factor + 0.5 * (self.start_factor - self.end_factor) * (
+                1 + math.cos(math.pi * progress)
+            )
+        self.optimizer.regularization._coeff = wd
+
+    def get_wd(self):
+        return self.optimizer.regularization._coeff
+
+
+def build_lr_scheduler(lr_config, epochs, step_each_epoch):
+    from . import learning_rate
+
+    lr_config.update({"epochs": epochs, "step_each_epoch": step_each_epoch})
+    lr_name = lr_config.pop("name", "Const")
+    lr = getattr(learning_rate, lr_name)(**lr_config)()
+    return lr
+
+
+def build_optimizer(config, epochs, step_each_epoch, model):
+    from . import regularizer, optimizer
+
+    config = copy.deepcopy(config)
+    # step1 build lr
+    lr = build_lr_scheduler(config.pop("lr"), epochs, step_each_epoch)
+
+    # step2 build regularization
+    wd_scheduler = None
+    if "regularizer" in config and config["regularizer"] is not None:
+        reg_config = config.pop("regularizer")
+        reg_name = reg_config.pop("name")
+        if not hasattr(regularizer, reg_name):
+            reg_name += "Decay"
+        reg_obj = getattr(regularizer, reg_name)(**reg_config)
+        reg = reg_obj()
+
+        # Build weight decay scheduler for CosineL2Decay
+        if isinstance(reg_obj, regularizer.CosineL2Decay):
+            warmup_epoch = reg_obj.warmup_epoch
+            warmup_steps = round(warmup_epoch * step_each_epoch)
+            total_steps = step_each_epoch * epochs
+            wd_scheduler = {
+                "start_factor": reg_obj.start_factor,
+                "end_factor": reg_obj.end_factor,
+                "total_steps": total_steps,
+                "warmup_steps": warmup_steps,
+            }
+    elif "weight_decay" in config:
+        reg = config.pop("weight_decay")
+    else:
+        reg = None
+
+    # step3 build optimizer
+    optim_name = config.pop("name")
+    if "clip_norm" in config:
+        clip_norm = config.pop("clip_norm")
+        grad_clip = paddle.nn.ClipGradByNorm(clip_norm=clip_norm)
+    elif "clip_norm_global" in config:
+        clip_norm = config.pop("clip_norm_global")
+        grad_clip = paddle.nn.ClipGradByGlobalNorm(clip_norm=clip_norm)
+    else:
+        grad_clip = None
+    optim = getattr(optimizer, optim_name)(
+        learning_rate=lr, weight_decay=reg, grad_clip=grad_clip, **config
+    )
+    built_optim = optim(model)
+
+    # Instantiate the scheduler now that we have the real optimizer
+    if wd_scheduler is not None:
+        wd_scheduler = CosineWeightDecayScheduler(built_optim, **wd_scheduler)
+
+    return built_optim, lr, wd_scheduler
diff --git a/ppocr/optimizer/learning_rate.py b/ppocr/optimizer/learning_rate.py
new file mode 100644
index 0000000..257dd3d
--- /dev/null
+++ b/ppocr/optimizer/learning_rate.py
@@ -0,0 +1,503 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+from paddle.optimizer import lr
+from .lr_scheduler import CyclicalCosineDecay, OneCycleDecay, TwoStepCosineDecay
+
+
+class Linear(object):
+    """
+    Linear learning rate decay
+    Args:
+        lr (float): The initial learning rate. It is a python float number.
+        epochs(int): The decay step size. It determines the decay cycle.
+        end_lr(float, optional): The minimum final learning rate. Default: 0.0001.
+        power(float, optional): Power of polynomial. Default: 1.0.
+        last_epoch (int, optional):  The index of last epoch. Can be set to restart training. Default: -1, means initial learning rate.
+    """
+
+    def __init__(
+        self,
+        learning_rate,
+        epochs,
+        step_each_epoch,
+        end_lr=0.0,
+        power=1.0,
+        warmup_epoch=0,
+        last_epoch=-1,
+        **kwargs,
+    ):
+        super(Linear, self).__init__()
+        self.learning_rate = learning_rate
+        self.epochs = epochs * step_each_epoch
+        self.end_lr = end_lr
+        self.power = power
+        self.last_epoch = last_epoch
+        self.warmup_epoch = round(warmup_epoch * step_each_epoch)
+
+    def __call__(self):
+        learning_rate = lr.PolynomialDecay(
+            learning_rate=self.learning_rate,
+            decay_steps=self.epochs,
+            end_lr=self.end_lr,
+            power=self.power,
+            last_epoch=self.last_epoch,
+        )
+        if self.warmup_epoch > 0:
+            learning_rate = lr.LinearWarmup(
+                learning_rate=learning_rate,
+                warmup_steps=self.warmup_epoch,
+                start_lr=0.0,
+                end_lr=self.learning_rate,
+                last_epoch=self.last_epoch,
+            )
+        return learning_rate
+
+
+class Cosine(object):
+    """
+    Cosine learning rate decay
+    lr = 0.05 * (math.cos(epoch * (math.pi / epochs)) + 1)
+    Args:
+        lr(float): initial learning rate
+        step_each_epoch(int): steps each epoch
+        epochs(int): total training epochs
+        last_epoch (int, optional):  The index of last epoch. Can be set to restart training. Default: -1, means initial learning rate.
+    """
+
+    def __init__(
+        self,
+        learning_rate,
+        step_each_epoch,
+        epochs,
+        warmup_epoch=0,
+        last_epoch=-1,
+        **kwargs,
+    ):
+        super(Cosine, self).__init__()
+        self.learning_rate = learning_rate
+        self.T_max = step_each_epoch * epochs
+        self.last_epoch = last_epoch
+        self.warmup_epoch = round(warmup_epoch * step_each_epoch)
+
+    def __call__(self):
+        learning_rate = lr.CosineAnnealingDecay(
+            learning_rate=self.learning_rate,
+            T_max=self.T_max,
+            last_epoch=self.last_epoch,
+        )
+        if self.warmup_epoch > 0:
+            learning_rate = lr.LinearWarmup(
+                learning_rate=learning_rate,
+                warmup_steps=self.warmup_epoch,
+                start_lr=0.0,
+                end_lr=self.learning_rate,
+                last_epoch=self.last_epoch,
+            )
+        return learning_rate
+
+
+class LinearWarmupCosine(object):
+    """
+    LinearWarmupCosine learning rate decay
+    Args:
+        learning_rate(float): initial learning rate
+        step_each_epoch(int): steps each epoch
+        epochs(int): total training epochs
+        start_lr (float): Initial learning rate of warm up.
+        min_lr (float): Minimum learning rate in CosineAnnealingDecay.
+        last_epoch (int, optional):  The index of last epoch. Can be set to restart training. Default: -1, means initial learning rate.
+    """
+
+    def __init__(
+        self,
+        learning_rate,
+        step_each_epoch,
+        epochs,
+        warmup_steps=5000,
+        start_lr=1e-5,
+        min_lr=1e-8,
+        last_epoch=-1,
+        **kwargs,
+    ):
+        super(LinearWarmupCosine, self).__init__()
+        self.learning_rate = float(learning_rate)
+        self.T_max = step_each_epoch * epochs
+        self.last_epoch = last_epoch
+        self.warmup_steps = warmup_steps
+        self.start_lr = float(start_lr)
+        self.min_lr = float(min_lr)
+
+    def __call__(self):
+        learning_rate = lr.CosineAnnealingDecay(
+            learning_rate=self.learning_rate,
+            T_max=self.T_max,
+            eta_min=self.min_lr,
+            last_epoch=self.last_epoch,
+        )
+        if self.warmup_steps > 0:
+            learning_rate = lr.LinearWarmup(
+                learning_rate=learning_rate,
+                warmup_steps=self.warmup_steps,
+                start_lr=self.start_lr,
+                end_lr=self.learning_rate,
+                last_epoch=self.last_epoch,
+            )
+        return learning_rate
+
+
+class Step(object):
+    """
+    Piecewise learning rate decay
+    Args:
+        step_each_epoch(int): steps each epoch
+        learning_rate (float): The initial learning rate. It is a python float number.
+        step_size (int): the interval to update.
+        gamma (float, optional): The Ratio that the learning rate will be reduced. ``new_lr = origin_lr * gamma`` .
+            It should be less than 1.0. Default: 0.1.
+        last_epoch (int, optional):  The index of last epoch. Can be set to restart training. Default: -1, means initial learning rate.
+    """
+
+    def __init__(
+        self,
+        learning_rate,
+        step_size,
+        step_each_epoch,
+        gamma,
+        warmup_epoch=0,
+        last_epoch=-1,
+        **kwargs,
+    ):
+        super(Step, self).__init__()
+        self.step_size = step_each_epoch * step_size
+        self.learning_rate = learning_rate
+        self.gamma = gamma
+        self.last_epoch = last_epoch
+        self.warmup_epoch = round(warmup_epoch * step_each_epoch)
+
+    def __call__(self):
+        learning_rate = lr.StepDecay(
+            learning_rate=self.learning_rate,
+            step_size=self.step_size,
+            gamma=self.gamma,
+            last_epoch=self.last_epoch,
+        )
+        if self.warmup_epoch > 0:
+            learning_rate = lr.LinearWarmup(
+                learning_rate=learning_rate,
+                warmup_steps=self.warmup_epoch,
+                start_lr=0.0,
+                end_lr=self.learning_rate,
+                last_epoch=self.last_epoch,
+            )
+        return learning_rate
+
+
+class Piecewise(object):
+    """
+    Piecewise learning rate decay
+    Args:
+        boundaries(list): A list of steps numbers. The type of element in the list is python int.
+        values(list): A list of learning rate values that will be picked during different epoch boundaries.
+            The type of element in the list is python float.
+        last_epoch (int, optional):  The index of last epoch. Can be set to restart training. Default: -1, means initial learning rate.
+    """
+
+    def __init__(
+        self,
+        step_each_epoch,
+        decay_epochs,
+        values,
+        warmup_epoch=0,
+        last_epoch=-1,
+        **kwargs,
+    ):
+        super(Piecewise, self).__init__()
+        self.boundaries = [step_each_epoch * e for e in decay_epochs]
+        self.values = values
+        self.last_epoch = last_epoch
+        self.warmup_epoch = round(warmup_epoch * step_each_epoch)
+
+    def __call__(self):
+        learning_rate = lr.PiecewiseDecay(
+            boundaries=self.boundaries, values=self.values, last_epoch=self.last_epoch
+        )
+        if self.warmup_epoch > 0:
+            learning_rate = lr.LinearWarmup(
+                learning_rate=learning_rate,
+                warmup_steps=self.warmup_epoch,
+                start_lr=0.0,
+                end_lr=self.values[0],
+                last_epoch=self.last_epoch,
+            )
+        return learning_rate
+
+
+class CyclicalCosine(object):
+    """
+    Cyclical cosine learning rate decay
+    Args:
+        learning_rate(float): initial learning rate
+        step_each_epoch(int): steps each epoch
+        epochs(int): total training epochs
+        cycle(int): period of the cosine learning rate
+        last_epoch (int, optional):  The index of last epoch. Can be set to restart training. Default: -1, means initial learning rate.
+    """
+
+    def __init__(
+        self,
+        learning_rate,
+        step_each_epoch,
+        epochs,
+        cycle,
+        warmup_epoch=0,
+        last_epoch=-1,
+        **kwargs,
+    ):
+        super(CyclicalCosine, self).__init__()
+        self.learning_rate = learning_rate
+        self.T_max = step_each_epoch * epochs
+        self.last_epoch = last_epoch
+        self.warmup_epoch = round(warmup_epoch * step_each_epoch)
+        self.cycle = round(cycle * step_each_epoch)
+
+    def __call__(self):
+        learning_rate = CyclicalCosineDecay(
+            learning_rate=self.learning_rate,
+            T_max=self.T_max,
+            cycle=self.cycle,
+            last_epoch=self.last_epoch,
+        )
+        if self.warmup_epoch > 0:
+            learning_rate = lr.LinearWarmup(
+                learning_rate=learning_rate,
+                warmup_steps=self.warmup_epoch,
+                start_lr=0.0,
+                end_lr=self.learning_rate,
+                last_epoch=self.last_epoch,
+            )
+        return learning_rate
+
+
+class OneCycle(object):
+    """
+    One Cycle learning rate decay
+    Args:
+        max_lr(float): Upper learning rate boundaries
+        epochs(int): total training epochs
+        step_each_epoch(int): steps each epoch
+        anneal_strategy(str): {‘cos’, ‘linear’} Specifies the annealing strategy: “cos” for cosine annealing, “linear” for linear annealing.
+            Default: ‘cos’
+        three_phase(bool): If True, use a third phase of the schedule to annihilate the learning rate according to ‘final_div_factor’
+            instead of modifying the second phase (the first two phases will be symmetrical about the step indicated by ‘pct_start’).
+        last_epoch (int, optional):  The index of last epoch. Can be set to restart training. Default: -1, means initial learning rate.
+    """
+
+    def __init__(
+        self,
+        max_lr,
+        epochs,
+        step_each_epoch,
+        anneal_strategy="cos",
+        three_phase=False,
+        warmup_epoch=0,
+        last_epoch=-1,
+        **kwargs,
+    ):
+        super(OneCycle, self).__init__()
+        self.max_lr = max_lr
+        self.epochs = epochs
+        self.steps_per_epoch = step_each_epoch
+        self.anneal_strategy = anneal_strategy
+        self.three_phase = three_phase
+        self.last_epoch = last_epoch
+        self.warmup_epoch = round(warmup_epoch * step_each_epoch)
+
+    def __call__(self):
+        learning_rate = OneCycleDecay(
+            max_lr=self.max_lr,
+            epochs=self.epochs,
+            steps_per_epoch=self.steps_per_epoch,
+            anneal_strategy=self.anneal_strategy,
+            three_phase=self.three_phase,
+            last_epoch=self.last_epoch,
+        )
+        if self.warmup_epoch > 0:
+            learning_rate = lr.LinearWarmup(
+                learning_rate=learning_rate,
+                warmup_steps=self.warmup_epoch,
+                start_lr=0.0,
+                end_lr=self.max_lr,
+                last_epoch=self.last_epoch,
+            )
+        return learning_rate
+
+
+class Const(object):
+    """
+    Const learning rate decay
+    Args:
+        learning_rate(float): initial learning rate
+        step_each_epoch(int): steps each epoch
+        last_epoch (int, optional):  The index of last epoch. Can be set to restart training. Default: -1, means initial learning rate.
+    """
+
+    def __init__(
+        self, learning_rate, step_each_epoch, warmup_epoch=0, last_epoch=-1, **kwargs
+    ):
+        super(Const, self).__init__()
+        self.learning_rate = learning_rate
+        self.last_epoch = last_epoch
+        self.warmup_epoch = round(warmup_epoch * step_each_epoch)
+
+    def __call__(self):
+        learning_rate = self.learning_rate
+        if self.warmup_epoch > 0:
+            learning_rate = lr.LinearWarmup(
+                learning_rate=learning_rate,
+                warmup_steps=self.warmup_epoch,
+                start_lr=0.0,
+                end_lr=self.learning_rate,
+                last_epoch=self.last_epoch,
+            )
+        return learning_rate
+
+
+class DecayLearningRate(object):
+    """
+    DecayLearningRate learning rate decay
+    new_lr = (lr - end_lr) * (1 - epoch/decay_steps)**power + end_lr
+    Args:
+        learning_rate(float): initial learning rate
+        step_each_epoch(int): steps each epoch
+        epochs(int): total training epochs
+        factor(float): Power of polynomial, should greater than 0.0 to get learning rate decay. Default: 0.9
+        end_lr(float): The minimum final learning rate. Default: 0.0.
+    """
+
+    def __init__(
+        self, learning_rate, step_each_epoch, epochs, factor=0.9, end_lr=0, **kwargs
+    ):
+        super(DecayLearningRate, self).__init__()
+        self.learning_rate = learning_rate
+        self.epochs = epochs + 1
+        self.factor = factor
+        self.end_lr = 0
+        self.decay_steps = step_each_epoch * epochs
+
+    def __call__(self):
+        learning_rate = lr.PolynomialDecay(
+            learning_rate=self.learning_rate,
+            decay_steps=self.decay_steps,
+            power=self.factor,
+            end_lr=self.end_lr,
+        )
+        return learning_rate
+
+
+class MultiStepDecay(object):
+    """
+    Piecewise learning rate decay
+    Args:
+        step_each_epoch(int): steps each epoch
+        learning_rate (float): The initial learning rate. It is a python float number.
+        step_size (int): the interval to update.
+        gamma (float, optional): The Ratio that the learning rate will be reduced. ``new_lr = origin_lr * gamma`` .
+            It should be less than 1.0. Default: 0.1.
+        last_epoch (int, optional):  The index of last epoch. Can be set to restart training. Default: -1, means initial learning rate.
+    """
+
+    def __init__(
+        self,
+        learning_rate,
+        milestones,
+        step_each_epoch,
+        gamma,
+        warmup_epoch=0,
+        last_epoch=-1,
+        **kwargs,
+    ):
+        super(MultiStepDecay, self).__init__()
+        self.milestones = [step_each_epoch * e for e in milestones]
+        self.learning_rate = learning_rate
+        self.gamma = gamma
+        self.last_epoch = last_epoch
+        self.warmup_epoch = round(warmup_epoch * step_each_epoch)
+
+    def __call__(self):
+        learning_rate = lr.MultiStepDecay(
+            learning_rate=self.learning_rate,
+            milestones=self.milestones,
+            gamma=self.gamma,
+            last_epoch=self.last_epoch,
+        )
+        if self.warmup_epoch > 0:
+            learning_rate = lr.LinearWarmup(
+                learning_rate=learning_rate,
+                warmup_steps=self.warmup_epoch,
+                start_lr=0.0,
+                end_lr=self.learning_rate,
+                last_epoch=self.last_epoch,
+            )
+        return learning_rate
+
+
+class TwoStepCosine(object):
+    """
+    Cosine learning rate decay
+    lr = 0.05 * (math.cos(epoch * (math.pi / epochs)) + 1)
+    Args:
+        lr(float): initial learning rate
+        step_each_epoch(int): steps each epoch
+        epochs(int): total training epochs
+        last_epoch (int, optional):  The index of last epoch. Can be set to restart training. Default: -1, means initial learning rate.
+    """
+
+    def __init__(
+        self,
+        learning_rate,
+        step_each_epoch,
+        epochs,
+        warmup_epoch=0,
+        last_epoch=-1,
+        **kwargs,
+    ):
+        super(TwoStepCosine, self).__init__()
+        self.learning_rate = learning_rate
+        self.T_max1 = step_each_epoch * 200
+        self.T_max2 = step_each_epoch * epochs
+        self.last_epoch = last_epoch
+        self.warmup_epoch = round(warmup_epoch * step_each_epoch)
+
+    def __call__(self):
+        learning_rate = TwoStepCosineDecay(
+            learning_rate=self.learning_rate,
+            T_max1=self.T_max1,
+            T_max2=self.T_max2,
+            last_epoch=self.last_epoch,
+        )
+        if self.warmup_epoch > 0:
+            learning_rate = lr.LinearWarmup(
+                learning_rate=learning_rate,
+                warmup_steps=self.warmup_epoch,
+                start_lr=0.0,
+                end_lr=self.learning_rate,
+                last_epoch=self.last_epoch,
+            )
+        return learning_rate
diff --git a/ppocr/optimizer/lr_scheduler.py b/ppocr/optimizer/lr_scheduler.py
new file mode 100644
index 0000000..1393e40
--- /dev/null
+++ b/ppocr/optimizer/lr_scheduler.py
@@ -0,0 +1,240 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import math
+from paddle.optimizer.lr import LRScheduler
+
+
+class CyclicalCosineDecay(LRScheduler):
+    def __init__(
+        self, learning_rate, T_max, cycle=1, last_epoch=-1, eta_min=0.0, verbose=False
+    ):
+        """
+        Cyclical cosine learning rate decay
+        A learning rate which can be referred in https://arxiv.org/pdf/2012.12645.pdf
+        Args:
+            learning rate(float): learning rate
+            T_max(int): maximum epoch num
+            cycle(int): period of the cosine decay
+            last_epoch (int, optional):  The index of last epoch. Can be set to restart training. Default: -1, means initial learning rate.
+            eta_min(float): minimum learning rate during training
+            verbose(bool): whether to print learning rate for each epoch
+        """
+        super(CyclicalCosineDecay, self).__init__(learning_rate, last_epoch, verbose)
+        self.cycle = cycle
+        self.eta_min = eta_min
+
+    def get_lr(self):
+        if self.last_epoch == 0:
+            return self.base_lr
+        reletive_epoch = self.last_epoch % self.cycle
+        lr = self.eta_min + 0.5 * (self.base_lr - self.eta_min) * (
+            1 + math.cos(math.pi * reletive_epoch / self.cycle)
+        )
+        return lr
+
+
+class OneCycleDecay(LRScheduler):
+    """
+    One Cycle learning rate decay
+    A learning rate which can be referred in https://arxiv.org/abs/1708.07120
+    Code referred in https://pytorch.org/docs/stable/_modules/torch/optim/lr_scheduler.html#OneCycleLR
+    """
+
+    def __init__(
+        self,
+        max_lr,
+        epochs=None,
+        steps_per_epoch=None,
+        pct_start=0.3,
+        anneal_strategy="cos",
+        div_factor=25.0,
+        final_div_factor=1e4,
+        three_phase=False,
+        last_epoch=-1,
+        verbose=False,
+    ):
+        # Validate total_steps
+        if epochs <= 0 or not isinstance(epochs, int):
+            raise ValueError(
+                "Expected positive integer epochs, but got {}".format(epochs)
+            )
+        if steps_per_epoch <= 0 or not isinstance(steps_per_epoch, int):
+            raise ValueError(
+                "Expected positive integer steps_per_epoch, but got {}".format(
+                    steps_per_epoch
+                )
+            )
+        self.total_steps = epochs * steps_per_epoch
+
+        self.max_lr = max_lr
+        self.initial_lr = self.max_lr / div_factor
+        self.min_lr = self.initial_lr / final_div_factor
+
+        if three_phase:
+            self._schedule_phases = [
+                {
+                    "end_step": float(pct_start * self.total_steps) - 1,
+                    "start_lr": self.initial_lr,
+                    "end_lr": self.max_lr,
+                },
+                {
+                    "end_step": float(2 * pct_start * self.total_steps) - 2,
+                    "start_lr": self.max_lr,
+                    "end_lr": self.initial_lr,
+                },
+                {
+                    "end_step": self.total_steps - 1,
+                    "start_lr": self.initial_lr,
+                    "end_lr": self.min_lr,
+                },
+            ]
+        else:
+            self._schedule_phases = [
+                {
+                    "end_step": float(pct_start * self.total_steps) - 1,
+                    "start_lr": self.initial_lr,
+                    "end_lr": self.max_lr,
+                },
+                {
+                    "end_step": self.total_steps - 1,
+                    "start_lr": self.max_lr,
+                    "end_lr": self.min_lr,
+                },
+            ]
+
+        # Validate pct_start
+        if pct_start < 0 or pct_start > 1 or not isinstance(pct_start, float):
+            raise ValueError(
+                "Expected float between 0 and 1 pct_start, but got {}".format(pct_start)
+            )
+
+        # Validate anneal_strategy
+        if anneal_strategy not in ["cos", "linear"]:
+            raise ValueError(
+                "anneal_strategy must by one of 'cos' or 'linear', instead got {}".format(
+                    anneal_strategy
+                )
+            )
+        elif anneal_strategy == "cos":
+            self.anneal_func = self._annealing_cos
+        elif anneal_strategy == "linear":
+            self.anneal_func = self._annealing_linear
+
+        super(OneCycleDecay, self).__init__(max_lr, last_epoch, verbose)
+
+    def _annealing_cos(self, start, end, pct):
+        "Cosine anneal from `start` to `end` as pct goes from 0.0 to 1.0."
+        cos_out = math.cos(math.pi * pct) + 1
+        return end + (start - end) / 2.0 * cos_out
+
+    def _annealing_linear(self, start, end, pct):
+        "Linearly anneal from `start` to `end` as pct goes from 0.0 to 1.0."
+        return (end - start) * pct + start
+
+    def get_lr(self):
+        computed_lr = 0.0
+        step_num = self.last_epoch
+
+        if step_num > self.total_steps:
+            raise ValueError(
+                "Tried to step {} times. The specified number of total steps is {}".format(
+                    step_num + 1, self.total_steps
+                )
+            )
+        start_step = 0
+        for i, phase in enumerate(self._schedule_phases):
+            end_step = phase["end_step"]
+            if step_num <= end_step or i == len(self._schedule_phases) - 1:
+                pct = (step_num - start_step) / (end_step - start_step)
+                computed_lr = self.anneal_func(phase["start_lr"], phase["end_lr"], pct)
+                break
+            start_step = phase["end_step"]
+
+        return computed_lr
+
+
+class TwoStepCosineDecay(LRScheduler):
+    def __init__(
+        self, learning_rate, T_max1, T_max2, eta_min=0, last_epoch=-1, verbose=False
+    ):
+        if not isinstance(T_max1, int):
+            raise TypeError(
+                "The type of 'T_max1' in 'CosineAnnealingDecay' must be 'int', but received %s."
+                % type(T_max1)
+            )
+        if not isinstance(T_max2, int):
+            raise TypeError(
+                "The type of 'T_max2' in 'CosineAnnealingDecay' must be 'int', but received %s."
+                % type(T_max2)
+            )
+        if not isinstance(eta_min, (float, int)):
+            raise TypeError(
+                "The type of 'eta_min' in 'CosineAnnealingDecay' must be 'float, int', but received %s."
+                % type(eta_min)
+            )
+        assert T_max1 > 0 and isinstance(
+            T_max1, int
+        ), " 'T_max1' must be a positive integer."
+        assert T_max2 > 0 and isinstance(
+            T_max2, int
+        ), " 'T_max1' must be a positive integer."
+        self.T_max1 = T_max1
+        self.T_max2 = T_max2
+        self.eta_min = float(eta_min)
+        super(TwoStepCosineDecay, self).__init__(learning_rate, last_epoch, verbose)
+
+    def get_lr(self):
+        if self.last_epoch <= self.T_max1:
+            if self.last_epoch == 0:
+                return self.base_lr
+            elif (self.last_epoch - 1 - self.T_max1) % (2 * self.T_max1) == 0:
+                return (
+                    self.last_lr
+                    + (self.base_lr - self.eta_min)
+                    * (1 - math.cos(math.pi / self.T_max1))
+                    / 2
+                )
+
+            return (1 + math.cos(math.pi * self.last_epoch / self.T_max1)) / (
+                1 + math.cos(math.pi * (self.last_epoch - 1) / self.T_max1)
+            ) * (self.last_lr - self.eta_min) + self.eta_min
+        else:
+            if (self.last_epoch - 1 - self.T_max2) % (2 * self.T_max2) == 0:
+                return (
+                    self.last_lr
+                    + (self.base_lr - self.eta_min)
+                    * (1 - math.cos(math.pi / self.T_max2))
+                    / 2
+                )
+
+            return (1 + math.cos(math.pi * self.last_epoch / self.T_max2)) / (
+                1 + math.cos(math.pi * (self.last_epoch - 1) / self.T_max2)
+            ) * (self.last_lr - self.eta_min) + self.eta_min
+
+    def _get_closed_form_lr(self):
+        if self.last_epoch <= self.T_max1:
+            return (
+                self.eta_min
+                + (self.base_lr - self.eta_min)
+                * (1 + math.cos(math.pi * self.last_epoch / self.T_max1))
+                / 2
+            )
+        else:
+            return (
+                self.eta_min
+                + (self.base_lr - self.eta_min)
+                * (1 + math.cos(math.pi * self.last_epoch / self.T_max2))
+                / 2
+            )
diff --git a/ppocr/optimizer/optimizer.py b/ppocr/optimizer/optimizer.py
new file mode 100644
index 0000000..d7f78a5
--- /dev/null
+++ b/ppocr/optimizer/optimizer.py
@@ -0,0 +1,292 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+from paddle import optimizer as optim
+
+
+class Momentum(object):
+    """
+    Simple Momentum optimizer with velocity state.
+    Args:
+        learning_rate (float|Variable) - The learning rate used to update parameters.
+            Can be a float value or a Variable with one float value as data element.
+        momentum (float) - Momentum factor.
+        regularization (WeightDecayRegularizer, optional) - The strategy of regularization.
+    """
+
+    def __init__(
+        self, learning_rate, momentum, weight_decay=None, grad_clip=None, **args
+    ):
+        super(Momentum, self).__init__()
+        self.learning_rate = learning_rate
+        self.momentum = momentum
+        self.weight_decay = weight_decay
+        self.grad_clip = grad_clip
+
+    def __call__(self, model):
+        train_params = [
+            param for param in model.parameters() if param.trainable is True
+        ]
+        opt = optim.Momentum(
+            learning_rate=self.learning_rate,
+            momentum=self.momentum,
+            weight_decay=self.weight_decay,
+            grad_clip=self.grad_clip,
+            parameters=train_params,
+        )
+        return opt
+
+
+class Adam(object):
+    def __init__(
+        self,
+        learning_rate=0.001,
+        beta1=0.9,
+        beta2=0.999,
+        epsilon=1e-08,
+        parameter_list=None,
+        weight_decay=None,
+        grad_clip=None,
+        name=None,
+        lazy_mode=False,
+        **kwargs,
+    ):
+        self.learning_rate = learning_rate
+        self.beta1 = beta1
+        self.beta2 = beta2
+        self.epsilon = epsilon
+        self.parameter_list = parameter_list
+        self.learning_rate = learning_rate
+        self.weight_decay = weight_decay
+        self.grad_clip = grad_clip
+        self.name = name
+        self.lazy_mode = lazy_mode
+        self.group_lr = kwargs.get("group_lr", False)
+        self.training_step = kwargs.get("training_step", None)
+
+    def __call__(self, model):
+        if self.group_lr:
+            if self.training_step == "LF_2":
+                import paddle
+
+                if isinstance(model, paddle.DataParallel):  # multi gpu
+                    mlm = model._layers.head.MLM_VRM.MLM.parameters()
+                    pre_mlm_pp = (
+                        model._layers.head.MLM_VRM.Prediction.pp_share.parameters()
+                    )
+                    pre_mlm_w = (
+                        model._layers.head.MLM_VRM.Prediction.w_share.parameters()
+                    )
+                else:  # single gpu
+                    mlm = model.head.MLM_VRM.MLM.parameters()
+                    pre_mlm_pp = model.head.MLM_VRM.Prediction.pp_share.parameters()
+                    pre_mlm_w = model.head.MLM_VRM.Prediction.w_share.parameters()
+
+                total = []
+                for param in mlm:
+                    total.append(id(param))
+                for param in pre_mlm_pp:
+                    total.append(id(param))
+                for param in pre_mlm_w:
+                    total.append(id(param))
+
+                group_base_params = [
+                    param for param in model.parameters() if id(param) in total
+                ]
+                group_small_params = [
+                    param for param in model.parameters() if id(param) not in total
+                ]
+                train_params = [
+                    {"params": group_base_params},
+                    {
+                        "params": group_small_params,
+                        "learning_rate": self.learning_rate.values[0] * 0.1,
+                    },
+                ]
+
+            else:
+                print("group lr currently only support VisionLAN in LF_2 training step")
+                train_params = [
+                    param for param in model.parameters() if param.trainable is True
+                ]
+        else:
+            train_params = [
+                param for param in model.parameters() if param.trainable is True
+            ]
+
+        opt = optim.Adam(
+            learning_rate=self.learning_rate,
+            beta1=self.beta1,
+            beta2=self.beta2,
+            epsilon=self.epsilon,
+            weight_decay=self.weight_decay,
+            grad_clip=self.grad_clip,
+            name=self.name,
+            lazy_mode=self.lazy_mode,
+            parameters=train_params,
+        )
+        return opt
+
+
+class RMSProp(object):
+    """
+    Root Mean Squared Propagation (RMSProp) is an unpublished, adaptive learning rate method.
+    Args:
+        learning_rate (float|Variable) - The learning rate used to update parameters.
+            Can be a float value or a Variable with one float value as data element.
+        momentum (float) - Momentum factor.
+        rho (float) - rho value in equation.
+        epsilon (float) - avoid division by zero, default is 1e-6.
+        regularization (WeightDecayRegularizer, optional) - The strategy of regularization.
+    """
+
+    def __init__(
+        self,
+        learning_rate,
+        momentum=0.0,
+        rho=0.95,
+        epsilon=1e-6,
+        weight_decay=None,
+        grad_clip=None,
+        **args,
+    ):
+        super(RMSProp, self).__init__()
+        self.learning_rate = learning_rate
+        self.momentum = momentum
+        self.rho = rho
+        self.epsilon = epsilon
+        self.weight_decay = weight_decay
+        self.grad_clip = grad_clip
+
+    def __call__(self, model):
+        train_params = [
+            param for param in model.parameters() if param.trainable is True
+        ]
+        opt = optim.RMSProp(
+            learning_rate=self.learning_rate,
+            momentum=self.momentum,
+            rho=self.rho,
+            epsilon=self.epsilon,
+            weight_decay=self.weight_decay,
+            grad_clip=self.grad_clip,
+            parameters=train_params,
+        )
+        return opt
+
+
+class Adadelta(object):
+    def __init__(
+        self,
+        learning_rate=0.001,
+        epsilon=1e-08,
+        rho=0.95,
+        parameter_list=None,
+        weight_decay=None,
+        grad_clip=None,
+        name=None,
+        **kwargs,
+    ):
+        self.learning_rate = learning_rate
+        self.epsilon = epsilon
+        self.rho = rho
+        self.parameter_list = parameter_list
+        self.learning_rate = learning_rate
+        self.weight_decay = weight_decay
+        self.grad_clip = grad_clip
+        self.name = name
+
+    def __call__(self, model):
+        train_params = [
+            param for param in model.parameters() if param.trainable is True
+        ]
+        opt = optim.Adadelta(
+            learning_rate=self.learning_rate,
+            epsilon=self.epsilon,
+            rho=self.rho,
+            weight_decay=self.weight_decay,
+            grad_clip=self.grad_clip,
+            name=self.name,
+            parameters=train_params,
+        )
+        return opt
+
+
+class AdamW(object):
+    def __init__(
+        self,
+        learning_rate=0.001,
+        beta1=0.9,
+        beta2=0.999,
+        epsilon=1e-8,
+        weight_decay=0.01,
+        multi_precision=False,
+        grad_clip=None,
+        no_weight_decay_name=None,
+        one_dim_param_no_weight_decay=False,
+        name=None,
+        lazy_mode=False,
+        **args,
+    ):
+        super().__init__()
+        self.learning_rate = learning_rate
+        self.beta1 = beta1
+        self.beta2 = beta2
+        self.epsilon = epsilon
+        self.grad_clip = grad_clip
+        self.weight_decay = 0.01 if weight_decay is None else weight_decay
+        self.grad_clip = grad_clip
+        self.name = name
+        self.lazy_mode = lazy_mode
+        self.multi_precision = multi_precision
+        self.no_weight_decay_name_list = (
+            no_weight_decay_name.split() if no_weight_decay_name else []
+        )
+        self.one_dim_param_no_weight_decay = one_dim_param_no_weight_decay
+
+    def __call__(self, model):
+        parameters = [param for param in model.parameters() if param.trainable is True]
+
+        self.no_weight_decay_param_name_list = [
+            p.name
+            for n, p in model.named_parameters()
+            if any(nd in n for nd in self.no_weight_decay_name_list)
+        ]
+
+        if self.one_dim_param_no_weight_decay:
+            self.no_weight_decay_param_name_list += [
+                p.name for n, p in model.named_parameters() if len(p.shape) == 1
+            ]
+
+        opt = optim.AdamW(
+            learning_rate=self.learning_rate,
+            beta1=self.beta1,
+            beta2=self.beta2,
+            epsilon=self.epsilon,
+            parameters=parameters,
+            weight_decay=self.weight_decay,
+            multi_precision=self.multi_precision,
+            grad_clip=self.grad_clip,
+            name=self.name,
+            lazy_mode=self.lazy_mode,
+            apply_decay_param_fun=self._apply_decay_param_fun,
+        )
+        return opt
+
+    def _apply_decay_param_fun(self, name):
+        return name not in self.no_weight_decay_param_name_list
diff --git a/ppocr/optimizer/regularizer.py b/ppocr/optimizer/regularizer.py
new file mode 100644
index 0000000..611ed99
--- /dev/null
+++ b/ppocr/optimizer/regularizer.py
@@ -0,0 +1,78 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+import paddle
+
+
+class L1Decay(object):
+    """
+    L1 Weight Decay Regularization, which encourages the weights to be sparse.
+    Args:
+        factor(float): regularization coeff. Default:0.0.
+    """
+
+    def __init__(self, factor=0.0):
+        super(L1Decay, self).__init__()
+        self.coeff = factor
+
+    def __call__(self):
+        reg = paddle.regularizer.L1Decay(self.coeff)
+        return reg
+
+
+class L2Decay(object):
+    """
+    L2 Weight Decay Regularization, which helps to prevent the model over-fitting.
+    Args:
+        factor(float): regularization coeff. Default:0.0.
+    """
+
+    def __init__(self, factor=0.0):
+        super(L2Decay, self).__init__()
+        self.coeff = float(factor)
+
+    def __call__(self):
+        return self.coeff
+
+
+class CosineL2Decay(object):
+    """
+    L2 Weight Decay with cosine annealing schedule.
+
+    Anneals the weight decay coefficient from `factor` to `end_factor`
+    following a cosine curve over total training steps, with optional
+    linear warmup. Avoids over-regularizing small-capacity models.
+
+    Reference: EfficientNetV2 (Tan & Le, 2021) - "annealing the loss
+    incurred by weight decay regularization over the course of training".
+
+    Args:
+        factor(float): initial weight decay coefficient.
+        end_factor(float): final weight decay coefficient. Default: 0.0.
+        warmup_epoch(int|float): warmup epochs (same as lr warmup). Default: 0.
+    """
+
+    def __init__(self, factor=5e-5, end_factor=0.0, warmup_epoch=0):
+        super(CosineL2Decay, self).__init__()
+        self.start_factor = float(factor)
+        self.end_factor = float(end_factor)
+        self.warmup_epoch = warmup_epoch
+
+    def __call__(self):
+        return self.start_factor
diff --git a/ppocr/postprocess/__init__.py b/ppocr/postprocess/__init__.py
new file mode 100644
index 0000000..e427ee9
--- /dev/null
+++ b/ppocr/postprocess/__init__.py
@@ -0,0 +1,121 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+from __future__ import unicode_literals
+
+import os
+import copy
+
+__all__ = ["build_post_process"]
+
+from .db_postprocess import DBPostProcess, DistillationDBPostProcess
+from .east_postprocess import EASTPostProcess
+from .sast_postprocess import SASTPostProcess
+from .fce_postprocess import FCEPostProcess
+from .rec_postprocess import (
+    CTCLabelDecode,
+    AttnLabelDecode,
+    SRNLabelDecode,
+    DistillationCTCLabelDecode,
+    NRTRLabelDecode,
+    SARLabelDecode,
+    SEEDLabelDecode,
+    PRENLabelDecode,
+    ViTSTRLabelDecode,
+    ABINetLabelDecode,
+    SPINLabelDecode,
+    VLLabelDecode,
+    RFLLabelDecode,
+    SATRNLabelDecode,
+    ParseQLabelDecode,
+    CPPDLabelDecode,
+    LaTeXOCRDecode,
+    UniMERNetDecode,
+)
+from .cls_postprocess import ClsPostProcess
+from .pg_postprocess import PGPostProcess
+from .vqa_token_ser_layoutlm_postprocess import (
+    VQASerTokenLayoutLMPostProcess,
+    DistillationSerPostProcess,
+)
+from .vqa_token_re_layoutlm_postprocess import (
+    VQAReTokenLayoutLMPostProcess,
+    DistillationRePostProcess,
+)
+from .table_postprocess import TableMasterLabelDecode, TableLabelDecode
+from .picodet_postprocess import PicoDetPostProcess
+from .ct_postprocess import CTPostProcess
+from .drrg_postprocess import DRRGPostprocess
+from .rec_postprocess import CANLabelDecode
+
+
+def build_post_process(config, global_config=None):
+    support_dict = [
+        "DBPostProcess",
+        "EASTPostProcess",
+        "SASTPostProcess",
+        "FCEPostProcess",
+        "CTCLabelDecode",
+        "AttnLabelDecode",
+        "ClsPostProcess",
+        "SRNLabelDecode",
+        "PGPostProcess",
+        "DistillationCTCLabelDecode",
+        "TableLabelDecode",
+        "DistillationDBPostProcess",
+        "NRTRLabelDecode",
+        "SARLabelDecode",
+        "SEEDLabelDecode",
+        "VQASerTokenLayoutLMPostProcess",
+        "VQAReTokenLayoutLMPostProcess",
+        "PRENLabelDecode",
+        "DistillationSARLabelDecode",
+        "ViTSTRLabelDecode",
+        "ABINetLabelDecode",
+        "TableMasterLabelDecode",
+        "SPINLabelDecode",
+        "DistillationSerPostProcess",
+        "DistillationRePostProcess",
+        "VLLabelDecode",
+        "PicoDetPostProcess",
+        "CTPostProcess",
+        "RFLLabelDecode",
+        "DRRGPostprocess",
+        "CANLabelDecode",
+        "SATRNLabelDecode",
+        "ParseQLabelDecode",
+        "CPPDLabelDecode",
+        "LaTeXOCRDecode",
+        "UniMERNetDecode",
+    ]
+
+    if config["name"] == "PSEPostProcess":
+        from .pse_postprocess import PSEPostProcess
+
+        support_dict.append("PSEPostProcess")
+
+    config = copy.deepcopy(config)
+    module_name = config.pop("name")
+    if module_name == "None":
+        return
+    if global_config is not None:
+        config.update(global_config)
+    assert module_name in support_dict, Exception(
+        "post process only support {}".format(support_dict)
+    )
+    module_class = eval(module_name)(**config)
+    return module_class
diff --git a/ppocr/postprocess/cls_postprocess.py b/ppocr/postprocess/cls_postprocess.py
new file mode 100644
index 0000000..06c7693
--- /dev/null
+++ b/ppocr/postprocess/cls_postprocess.py
@@ -0,0 +1,43 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import paddle
+
+
+class ClsPostProcess(object):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, label_list=None, key=None, **kwargs):
+        super(ClsPostProcess, self).__init__()
+        self.label_list = label_list
+        self.key = key
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if self.key is not None:
+            preds = preds[self.key]
+
+        label_list = self.label_list
+        if label_list is None:
+            label_list = {idx: idx for idx in range(preds.shape[-1])}
+
+        if isinstance(preds, paddle.Tensor):
+            preds = preds.numpy()
+
+        pred_idxs = preds.argmax(axis=1)
+        decode_out = [
+            (label_list[idx], preds[i, idx]) for i, idx in enumerate(pred_idxs)
+        ]
+        if label is None:
+            return decode_out
+        label = [(label_list[idx], 1.0) for idx in label]
+        return decode_out, label
diff --git a/ppocr/postprocess/ct_postprocess.py b/ppocr/postprocess/ct_postprocess.py
new file mode 100755
index 0000000..3a9ce5c
--- /dev/null
+++ b/ppocr/postprocess/ct_postprocess.py
@@ -0,0 +1,158 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is referred from:
+https://github.com/shengtao96/CentripetalText/blob/main/test.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import os.path as osp
+import numpy as np
+import cv2
+import paddle
+import pyclipper
+
+
+class CTPostProcess(object):
+    """
+    The post process for Centripetal Text (CT).
+    """
+
+    def __init__(self, min_score=0.88, min_area=16, box_type="poly", **kwargs):
+        self.min_score = min_score
+        self.min_area = min_area
+        self.box_type = box_type
+
+        self.coord = np.zeros((2, 300, 300), dtype=np.int32)
+        for i in range(300):
+            for j in range(300):
+                self.coord[0, i, j] = j
+                self.coord[1, i, j] = i
+
+    def __call__(self, preds, batch):
+        outs = preds["maps"]
+        out_scores = preds["score"]
+
+        if isinstance(outs, paddle.Tensor):
+            outs = outs.numpy()
+        if isinstance(out_scores, paddle.Tensor):
+            out_scores = out_scores.numpy()
+
+        batch_size = outs.shape[0]
+        boxes_batch = []
+        for idx in range(batch_size):
+            bboxes = []
+            scores = []
+
+            img_shape = batch[idx]
+
+            org_img_size = img_shape[:3]
+            img_shape = img_shape[3:]
+            img_size = img_shape[:2]
+
+            out = np.expand_dims(outs[idx], axis=0)
+            outputs = dict()
+
+            score = np.expand_dims(out_scores[idx], axis=0)
+
+            kernel = out[:, 0, :, :] > 0.2
+            loc = out[:, 1:, :, :].astype("float32")
+
+            score = score[0].astype(np.float32)
+            kernel = kernel[0].astype(np.uint8)
+            loc = loc[0].astype(np.float32)
+
+            label_num, label_kernel = cv2.connectedComponents(kernel, connectivity=4)
+
+            for i in range(1, label_num):
+                ind = label_kernel == i
+                if ind.sum() < 10:  # pixel number less than 10, treated as background
+                    label_kernel[ind] = 0
+
+            label = np.zeros_like(label_kernel)
+            h, w = label_kernel.shape
+            pixels = self.coord[:, :h, :w].reshape(2, -1)
+            points = pixels.transpose([1, 0]).astype(np.float32)
+
+            off_points = (points + 10.0 / 4.0 * loc[:, pixels[1], pixels[0]].T).astype(
+                np.int32
+            )
+            off_points[:, 0] = np.clip(off_points[:, 0], 0, label.shape[1] - 1)
+            off_points[:, 1] = np.clip(off_points[:, 1], 0, label.shape[0] - 1)
+
+            label[pixels[1], pixels[0]] = label_kernel[
+                off_points[:, 1], off_points[:, 0]
+            ]
+            label[label_kernel > 0] = label_kernel[label_kernel > 0]
+
+            score_pocket = [0.0]
+            for i in range(1, label_num):
+                ind = label_kernel == i
+                if ind.sum() == 0:
+                    score_pocket.append(0.0)
+                    continue
+                score_i = np.mean(score[ind])
+                score_pocket.append(score_i)
+
+            label_num = np.max(label) + 1
+            label = cv2.resize(
+                label, (img_size[1], img_size[0]), interpolation=cv2.INTER_NEAREST
+            )
+
+            scale = (
+                float(org_img_size[1]) / float(img_size[1]),
+                float(org_img_size[0]) / float(img_size[0]),
+            )
+
+            for i in range(1, label_num):
+                ind = label == i
+                points = np.array(np.where(ind)).transpose((1, 0))
+
+                if points.shape[0] < self.min_area:
+                    continue
+
+                score_i = score_pocket[i]
+                if score_i < self.min_score:
+                    continue
+
+                if self.box_type == "rect":
+                    rect = cv2.minAreaRect(points[:, ::-1])
+                    bbox = cv2.boxPoints(rect) * scale
+                    z = bbox.mean(0)
+                    bbox = z + (bbox - z) * 0.85
+                elif self.box_type == "poly":
+                    binary = np.zeros(label.shape, dtype="uint8")
+                    binary[ind] = 1
+                    try:
+                        _, contours, _ = cv2.findContours(
+                            binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE
+                        )
+                    except BaseException:
+                        contours, _ = cv2.findContours(
+                            binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE
+                        )
+
+                    bbox = contours[0] * scale
+
+                bbox = bbox.astype("int32")
+                bboxes.append(bbox.reshape(-1, 2))
+                scores.append(score_i)
+
+            boxes_batch.append({"points": bboxes})
+
+        return boxes_batch
diff --git a/ppocr/postprocess/db_postprocess.py b/ppocr/postprocess/db_postprocess.py
new file mode 100755
index 0000000..f7e84d3
--- /dev/null
+++ b/ppocr/postprocess/db_postprocess.py
@@ -0,0 +1,289 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is referred from:
+https://github.com/WenmuZhou/DBNet.pytorch/blob/master/post_processing/seg_detector_representer.py
+"""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+import cv2
+import paddle
+from shapely.geometry import Polygon
+import pyclipper
+
+
+class DBPostProcess(object):
+    """
+    The post process for Differentiable Binarization (DB).
+    """
+
+    def __init__(
+        self,
+        thresh=0.3,
+        box_thresh=0.7,
+        max_candidates=1000,
+        unclip_ratio=2.0,
+        use_dilation=False,
+        score_mode="fast",
+        box_type="quad",
+        **kwargs,
+    ):
+        self.thresh = thresh
+        self.box_thresh = box_thresh
+        self.max_candidates = max_candidates
+        self.unclip_ratio = unclip_ratio
+        self.min_size = 3
+        self.score_mode = score_mode
+        self.box_type = box_type
+        assert score_mode in [
+            "slow",
+            "fast",
+        ], "Score mode must be in [slow, fast] but got: {}".format(score_mode)
+
+        self.dilation_kernel = None if not use_dilation else np.array([[1, 1], [1, 1]])
+
+    def polygons_from_bitmap(self, pred, _bitmap, dest_width, dest_height):
+        """
+        _bitmap: single map with shape (1, H, W),
+            whose values are binarized as {0, 1}
+        """
+
+        bitmap = _bitmap
+        height, width = bitmap.shape
+
+        boxes = []
+        scores = []
+
+        contours, _ = cv2.findContours(
+            (bitmap * 255).astype(np.uint8), cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE
+        )
+
+        for contour in contours[: self.max_candidates]:
+            epsilon = 0.002 * cv2.arcLength(contour, True)
+            approx = cv2.approxPolyDP(contour, epsilon, True)
+            points = approx.reshape((-1, 2))
+            if points.shape[0] < 4:
+                continue
+
+            score = self.box_score_fast(pred, points.reshape(-1, 2))
+            if self.box_thresh > score:
+                continue
+
+            if points.shape[0] > 2:
+                box = self.unclip(points, self.unclip_ratio)
+                if len(box) > 1:
+                    continue
+            else:
+                continue
+            box = np.array(box).reshape(-1, 2)
+            if len(box) == 0:
+                continue
+
+            _, sside = self.get_mini_boxes(box.reshape((-1, 1, 2)))
+            if sside < self.min_size + 2:
+                continue
+
+            box = np.array(box)
+            box[:, 0] = np.clip(np.round(box[:, 0] / width * dest_width), 0, dest_width)
+            box[:, 1] = np.clip(
+                np.round(box[:, 1] / height * dest_height), 0, dest_height
+            )
+            boxes.append(box.tolist())
+            scores.append(score)
+        return boxes, scores
+
+    def boxes_from_bitmap(self, pred, _bitmap, dest_width, dest_height):
+        """
+        _bitmap: single map with shape (1, H, W),
+                whose values are binarized as {0, 1}
+        """
+
+        bitmap = _bitmap
+        height, width = bitmap.shape
+
+        outs = cv2.findContours(
+            (bitmap * 255).astype(np.uint8), cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE
+        )
+        if len(outs) == 3:
+            img, contours, _ = outs[0], outs[1], outs[2]
+        elif len(outs) == 2:
+            contours, _ = outs[0], outs[1]
+
+        num_contours = min(len(contours), self.max_candidates)
+
+        boxes = []
+        scores = []
+        for index in range(num_contours):
+            contour = contours[index]
+            points, sside = self.get_mini_boxes(contour)
+            if sside < self.min_size:
+                continue
+            points = np.array(points)
+            if self.score_mode == "fast":
+                score = self.box_score_fast(pred, points.reshape(-1, 2))
+            else:
+                score = self.box_score_slow(pred, contour)
+            if self.box_thresh > score:
+                continue
+
+            box = self.unclip(points, self.unclip_ratio)
+            if len(box) > 1:
+                continue
+            box = np.array(box).reshape(-1, 1, 2)
+            box, sside = self.get_mini_boxes(box)
+            if sside < self.min_size + 2:
+                continue
+            box = np.array(box)
+
+            box[:, 0] = np.clip(np.round(box[:, 0] / width * dest_width), 0, dest_width)
+            box[:, 1] = np.clip(
+                np.round(box[:, 1] / height * dest_height), 0, dest_height
+            )
+            boxes.append(box.astype("int32"))
+            scores.append(score)
+        return np.array(boxes, dtype="int32"), scores
+
+    def unclip(self, box, unclip_ratio):
+        poly = Polygon(box)
+        distance = poly.area * unclip_ratio / poly.length
+        offset = pyclipper.PyclipperOffset()
+        offset.AddPath(box, pyclipper.JT_ROUND, pyclipper.ET_CLOSEDPOLYGON)
+        expanded = offset.Execute(distance)
+        return expanded
+
+    def get_mini_boxes(self, contour):
+        bounding_box = cv2.minAreaRect(contour)
+        points = sorted(list(cv2.boxPoints(bounding_box)), key=lambda x: x[0])
+
+        index_1, index_2, index_3, index_4 = 0, 1, 2, 3
+        if points[1][1] > points[0][1]:
+            index_1 = 0
+            index_4 = 1
+        else:
+            index_1 = 1
+            index_4 = 0
+        if points[3][1] > points[2][1]:
+            index_2 = 2
+            index_3 = 3
+        else:
+            index_2 = 3
+            index_3 = 2
+
+        box = [points[index_1], points[index_2], points[index_3], points[index_4]]
+        return box, min(bounding_box[1])
+
+    def box_score_fast(self, bitmap, _box):
+        """
+        box_score_fast: use bbox mean score as the mean score
+        """
+        h, w = bitmap.shape[:2]
+        box = _box.copy()
+        xmin = np.clip(np.floor(box[:, 0].min()).astype("int32"), 0, w - 1)
+        xmax = np.clip(np.ceil(box[:, 0].max()).astype("int32"), 0, w - 1)
+        ymin = np.clip(np.floor(box[:, 1].min()).astype("int32"), 0, h - 1)
+        ymax = np.clip(np.ceil(box[:, 1].max()).astype("int32"), 0, h - 1)
+
+        mask = np.zeros((ymax - ymin + 1, xmax - xmin + 1), dtype=np.uint8)
+        box[:, 0] = box[:, 0] - xmin
+        box[:, 1] = box[:, 1] - ymin
+        cv2.fillPoly(mask, box.reshape(1, -1, 2).astype("int32"), 1)
+        return cv2.mean(bitmap[ymin : ymax + 1, xmin : xmax + 1], mask)[0]
+
+    def box_score_slow(self, bitmap, contour):
+        """
+        box_score_slow: use polyon mean score as the mean score
+        """
+        h, w = bitmap.shape[:2]
+        contour = contour.copy()
+        contour = np.reshape(contour, (-1, 2))
+
+        xmin = np.clip(np.min(contour[:, 0]), 0, w - 1)
+        xmax = np.clip(np.max(contour[:, 0]), 0, w - 1)
+        ymin = np.clip(np.min(contour[:, 1]), 0, h - 1)
+        ymax = np.clip(np.max(contour[:, 1]), 0, h - 1)
+
+        mask = np.zeros((ymax - ymin + 1, xmax - xmin + 1), dtype=np.uint8)
+
+        contour[:, 0] = contour[:, 0] - xmin
+        contour[:, 1] = contour[:, 1] - ymin
+
+        cv2.fillPoly(mask, contour.reshape(1, -1, 2).astype("int32"), 1)
+        return cv2.mean(bitmap[ymin : ymax + 1, xmin : xmax + 1], mask)[0]
+
+    def __call__(self, outs_dict, shape_list):
+        pred = outs_dict["maps"]
+        if isinstance(pred, paddle.Tensor):
+            pred = pred.numpy()
+        pred = pred[:, 0, :, :]
+        segmentation = pred > self.thresh
+
+        boxes_batch = []
+        for batch_index in range(pred.shape[0]):
+            src_h, src_w, ratio_h, ratio_w = shape_list[batch_index]
+            if self.dilation_kernel is not None:
+                mask = cv2.dilate(
+                    np.array(segmentation[batch_index]).astype(np.uint8),
+                    self.dilation_kernel,
+                )
+            else:
+                mask = segmentation[batch_index]
+            if self.box_type == "poly":
+                boxes, scores = self.polygons_from_bitmap(
+                    pred[batch_index], mask, src_w, src_h
+                )
+            elif self.box_type == "quad":
+                boxes, scores = self.boxes_from_bitmap(
+                    pred[batch_index], mask, src_w, src_h
+                )
+            else:
+                raise ValueError("box_type can only be one of ['quad', 'poly']")
+
+            boxes_batch.append({"points": boxes})
+        return boxes_batch
+
+
+class DistillationDBPostProcess(object):
+    def __init__(
+        self,
+        model_name=["student"],
+        key=None,
+        thresh=0.3,
+        box_thresh=0.6,
+        max_candidates=1000,
+        unclip_ratio=1.5,
+        use_dilation=False,
+        score_mode="fast",
+        box_type="quad",
+        **kwargs,
+    ):
+        self.model_name = model_name
+        self.key = key
+        self.post_process = DBPostProcess(
+            thresh=thresh,
+            box_thresh=box_thresh,
+            max_candidates=max_candidates,
+            unclip_ratio=unclip_ratio,
+            use_dilation=use_dilation,
+            score_mode=score_mode,
+            box_type=box_type,
+        )
+
+    def __call__(self, predicts, shape_list):
+        results = {}
+        for k in self.model_name:
+            results[k] = self.post_process(predicts[k], shape_list=shape_list)
+        return results
diff --git a/ppocr/postprocess/drrg_postprocess.py b/ppocr/postprocess/drrg_postprocess.py
new file mode 100644
index 0000000..f1241c4
--- /dev/null
+++ b/ppocr/postprocess/drrg_postprocess.py
@@ -0,0 +1,338 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/main/mmocr/models/textdet/postprocess/drrg_postprocessor.py
+"""
+
+import functools
+import operator
+
+import numpy as np
+import paddle
+from numpy.linalg import norm
+import cv2
+
+
+class Node:
+    def __init__(self, ind):
+        self.__ind = ind
+        self.__links = set()
+
+    @property
+    def ind(self):
+        return self.__ind
+
+    @property
+    def links(self):
+        return set(self.__links)
+
+    def add_link(self, link_node):
+        self.__links.add(link_node)
+        link_node.__links.add(self)
+
+
+def graph_propagation(edges, scores, text_comps, edge_len_thr=50.0):
+    assert edges.ndim == 2
+    assert edges.shape[1] == 2
+    assert edges.shape[0] == scores.shape[0]
+    assert text_comps.ndim == 2
+    assert isinstance(edge_len_thr, float)
+
+    edges = np.sort(edges, axis=1)
+    score_dict = {}
+    for i, edge in enumerate(edges):
+        if text_comps is not None:
+            box1 = text_comps[edge[0], :8].reshape(4, 2)
+            box2 = text_comps[edge[1], :8].reshape(4, 2)
+            center1 = np.mean(box1, axis=0)
+            center2 = np.mean(box2, axis=0)
+            distance = norm(center1 - center2)
+            if distance > edge_len_thr:
+                scores[i] = 0
+        if (edge[0], edge[1]) in score_dict:
+            score_dict[edge[0], edge[1]] = 0.5 * (
+                score_dict[edge[0], edge[1]] + scores[i]
+            )
+        else:
+            score_dict[edge[0], edge[1]] = scores[i]
+
+    nodes = np.sort(np.unique(edges.flatten()))
+    mapping = -1 * np.ones((np.max(nodes) + 1), dtype=np.int32)
+    mapping[nodes] = np.arange(nodes.shape[0])
+    order_inds = mapping[edges]
+    vertices = [Node(node) for node in nodes]
+    for ind in order_inds:
+        vertices[ind[0]].add_link(vertices[ind[1]])
+
+    return vertices, score_dict
+
+
+def connected_components(nodes, score_dict, link_thr):
+    assert isinstance(nodes, list)
+    assert all([isinstance(node, Node) for node in nodes])
+    assert isinstance(score_dict, dict)
+    assert isinstance(link_thr, float)
+
+    clusters = []
+    nodes = set(nodes)
+    while nodes:
+        node = nodes.pop()
+        cluster = {node}
+        node_queue = [node]
+        while node_queue:
+            node = node_queue.pop(0)
+            neighbors = set(
+                [
+                    neighbor
+                    for neighbor in node.links
+                    if score_dict[tuple(sorted([node.ind, neighbor.ind]))] >= link_thr
+                ]
+            )
+            neighbors.difference_update(cluster)
+            nodes.difference_update(neighbors)
+            cluster.update(neighbors)
+            node_queue.extend(neighbors)
+        clusters.append(list(cluster))
+    return clusters
+
+
+def clusters2labels(clusters, num_nodes):
+    assert isinstance(clusters, list)
+    assert all([isinstance(cluster, list) for cluster in clusters])
+    assert all([isinstance(node, Node) for cluster in clusters for node in cluster])
+    assert isinstance(num_nodes, int)
+
+    node_labels = np.zeros(num_nodes)
+    for cluster_ind, cluster in enumerate(clusters):
+        for node in cluster:
+            node_labels[node.ind] = cluster_ind
+    return node_labels
+
+
+def remove_single(text_comps, comp_pred_labels):
+    assert text_comps.ndim == 2
+    assert text_comps.shape[0] == comp_pred_labels.shape[0]
+
+    single_flags = np.zeros_like(comp_pred_labels)
+    pred_labels = np.unique(comp_pred_labels)
+    for label in pred_labels:
+        current_label_flag = comp_pred_labels == label
+        if np.sum(current_label_flag) == 1:
+            single_flags[np.where(current_label_flag)[0][0]] = 1
+    keep_ind = [i for i in range(len(comp_pred_labels)) if not single_flags[i]]
+    filtered_text_comps = text_comps[keep_ind, :]
+    filtered_labels = comp_pred_labels[keep_ind]
+
+    return filtered_text_comps, filtered_labels
+
+
+def norm2(point1, point2):
+    return ((point1[0] - point2[0]) ** 2 + (point1[1] - point2[1]) ** 2) ** 0.5
+
+
+def min_connect_path(points):
+    assert isinstance(points, list)
+    assert all([isinstance(point, list) for point in points])
+    assert all([isinstance(coord, int) for point in points for coord in point])
+
+    points_queue = points.copy()
+    shortest_path = []
+    current_edge = [[], []]
+
+    edge_dict0 = {}
+    edge_dict1 = {}
+    current_edge[0] = points_queue[0]
+    current_edge[1] = points_queue[0]
+    points_queue.remove(points_queue[0])
+    while points_queue:
+        for point in points_queue:
+            length0 = norm2(point, current_edge[0])
+            edge_dict0[length0] = [point, current_edge[0]]
+            length1 = norm2(current_edge[1], point)
+            edge_dict1[length1] = [current_edge[1], point]
+        key0 = min(edge_dict0.keys())
+        key1 = min(edge_dict1.keys())
+
+        if key0 <= key1:
+            start = edge_dict0[key0][0]
+            end = edge_dict0[key0][1]
+            shortest_path.insert(0, [points.index(start), points.index(end)])
+            points_queue.remove(start)
+            current_edge[0] = start
+        else:
+            start = edge_dict1[key1][0]
+            end = edge_dict1[key1][1]
+            shortest_path.append([points.index(start), points.index(end)])
+            points_queue.remove(end)
+            current_edge[1] = end
+
+        edge_dict0 = {}
+        edge_dict1 = {}
+
+    shortest_path = functools.reduce(operator.concat, shortest_path)
+    shortest_path = sorted(set(shortest_path), key=shortest_path.index)
+
+    return shortest_path
+
+
+def in_contour(cont, point):
+    x, y = point
+    is_inner = cv2.pointPolygonTest(cont, (int(x), int(y)), False) > 0.5
+    return is_inner
+
+
+def fix_corner(top_line, bot_line, start_box, end_box):
+    assert isinstance(top_line, list)
+    assert all(isinstance(point, list) for point in top_line)
+    assert isinstance(bot_line, list)
+    assert all(isinstance(point, list) for point in bot_line)
+    assert start_box.shape == end_box.shape == (4, 2)
+
+    contour = np.array(top_line + bot_line[::-1])
+    start_left_mid = (start_box[0] + start_box[3]) / 2
+    start_right_mid = (start_box[1] + start_box[2]) / 2
+    end_left_mid = (end_box[0] + end_box[3]) / 2
+    end_right_mid = (end_box[1] + end_box[2]) / 2
+    if not in_contour(contour, start_left_mid):
+        top_line.insert(0, start_box[0].tolist())
+        bot_line.insert(0, start_box[3].tolist())
+    elif not in_contour(contour, start_right_mid):
+        top_line.insert(0, start_box[1].tolist())
+        bot_line.insert(0, start_box[2].tolist())
+    if not in_contour(contour, end_left_mid):
+        top_line.append(end_box[0].tolist())
+        bot_line.append(end_box[3].tolist())
+    elif not in_contour(contour, end_right_mid):
+        top_line.append(end_box[1].tolist())
+        bot_line.append(end_box[2].tolist())
+    return top_line, bot_line
+
+
+def comps2boundaries(text_comps, comp_pred_labels):
+    assert text_comps.ndim == 2
+    assert len(text_comps) == len(comp_pred_labels)
+    boundaries = []
+    if len(text_comps) < 1:
+        return boundaries
+    for cluster_ind in range(0, int(np.max(comp_pred_labels)) + 1):
+        cluster_comp_inds = np.where(comp_pred_labels == cluster_ind)
+        text_comp_boxes = (
+            text_comps[cluster_comp_inds, :8].reshape((-1, 4, 2)).astype(np.int32)
+        )
+        score = np.mean(text_comps[cluster_comp_inds, -1])
+
+        if text_comp_boxes.shape[0] < 1:
+            continue
+
+        elif text_comp_boxes.shape[0] > 1:
+            centers = np.mean(text_comp_boxes, axis=1).astype(np.int32).tolist()
+            shortest_path = min_connect_path(centers)
+            text_comp_boxes = text_comp_boxes[shortest_path]
+            top_line = (
+                np.mean(text_comp_boxes[:, 0:2, :], axis=1).astype(np.int32).tolist()
+            )
+            bot_line = (
+                np.mean(text_comp_boxes[:, 2:4, :], axis=1).astype(np.int32).tolist()
+            )
+            top_line, bot_line = fix_corner(
+                top_line, bot_line, text_comp_boxes[0], text_comp_boxes[-1]
+            )
+            boundary_points = top_line + bot_line[::-1]
+
+        else:
+            top_line = text_comp_boxes[0, 0:2, :].astype(np.int32).tolist()
+            bot_line = text_comp_boxes[0, 2:4:-1, :].astype(np.int32).tolist()
+            boundary_points = top_line + bot_line
+
+        boundary = [p for coord in boundary_points for p in coord] + [score]
+        boundaries.append(boundary)
+
+    return boundaries
+
+
+class DRRGPostprocess(object):
+    """Merge text components and construct boundaries of text instances.
+
+    Args:
+        link_thr (float): The edge score threshold.
+    """
+
+    def __init__(self, link_thr, **kwargs):
+        assert isinstance(link_thr, float)
+        self.link_thr = link_thr
+
+    def __call__(self, preds, shape_list):
+        """
+        Args:
+            edges (ndarray): The edge array of shape N * 2, each row is a node
+                index pair that makes up an edge in graph.
+            scores (ndarray): The edge score array of shape (N,).
+            text_comps (ndarray): The text components.
+
+        Returns:
+            List[list[float]]: The predicted boundaries of text instances.
+        """
+        edges, scores, text_comps = preds
+        if edges is not None:
+            if isinstance(edges, paddle.Tensor):
+                edges = edges.numpy()
+            if isinstance(scores, paddle.Tensor):
+                scores = scores.numpy()
+            if isinstance(text_comps, paddle.Tensor):
+                text_comps = text_comps.numpy()
+            assert len(edges) == len(scores)
+            assert text_comps.ndim == 2
+            assert text_comps.shape[1] == 9
+
+            vertices, score_dict = graph_propagation(edges, scores, text_comps)
+            clusters = connected_components(vertices, score_dict, self.link_thr)
+            pred_labels = clusters2labels(clusters, text_comps.shape[0])
+            text_comps, pred_labels = remove_single(text_comps, pred_labels)
+            boundaries = comps2boundaries(text_comps, pred_labels)
+        else:
+            boundaries = []
+
+        boundaries, scores = self.resize_boundary(
+            boundaries, (1 / shape_list[0, 2:]).tolist()[::-1]
+        )
+        boxes_batch = [dict(points=boundaries, scores=scores)]
+        return boxes_batch
+
+    def resize_boundary(self, boundaries, scale_factor):
+        """Rescale boundaries via scale_factor.
+
+        Args:
+            boundaries (list[list[float]]): The boundary list. Each boundary
+            with size 2k+1 with k>=4.
+            scale_factor(ndarray): The scale factor of size (4,).
+
+        Returns:
+            boundaries (list[list[float]]): The scaled boundaries.
+        """
+        boxes = []
+        scores = []
+        for b in boundaries:
+            sz = len(b)
+            scores.append(b[-1])
+            b = (
+                (
+                    np.array(b[: sz - 1])
+                    * (np.tile(scale_factor[:2], int((sz - 1) / 2)).reshape(1, sz - 1))
+                )
+                .flatten()
+                .tolist()
+            )
+            boxes.append(np.array(b).reshape([-1, 2]))
+        return boxes, scores
diff --git a/ppocr/postprocess/east_postprocess.py b/ppocr/postprocess/east_postprocess.py
new file mode 100755
index 0000000..3db2fe8
--- /dev/null
+++ b/ppocr/postprocess/east_postprocess.py
@@ -0,0 +1,141 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+from .locality_aware_nms import nms_locality
+import cv2
+import paddle
+
+import os
+from ppocr.utils.utility import check_install
+import sys
+
+
+class EASTPostProcess(object):
+    """
+    The post process for EAST.
+    """
+
+    def __init__(self, score_thresh=0.8, cover_thresh=0.1, nms_thresh=0.2, **kwargs):
+        self.score_thresh = score_thresh
+        self.cover_thresh = cover_thresh
+        self.nms_thresh = nms_thresh
+
+    def restore_rectangle_quad(self, origin, geometry):
+        """
+        Restore rectangle from quadrangle.
+        """
+        # quad
+        origin_concat = np.concatenate(
+            (origin, origin, origin, origin), axis=1
+        )  # (n, 8)
+        pred_quads = origin_concat - geometry
+        pred_quads = pred_quads.reshape((-1, 4, 2))  # (n, 4, 2)
+        return pred_quads
+
+    def detect(
+        self, score_map, geo_map, score_thresh=0.8, cover_thresh=0.1, nms_thresh=0.2
+    ):
+        """
+        restore text boxes from score map and geo map
+        """
+
+        score_map = score_map[0]
+        geo_map = np.swapaxes(geo_map, 1, 0)
+        geo_map = np.swapaxes(geo_map, 1, 2)
+        # filter the score map
+        xy_text = np.argwhere(score_map > score_thresh)
+        if len(xy_text) == 0:
+            return []
+        # sort the text boxes via the y axis
+        xy_text = xy_text[np.argsort(xy_text[:, 0])]
+        # restore quad proposals
+        text_box_restored = self.restore_rectangle_quad(
+            xy_text[:, ::-1] * 4, geo_map[xy_text[:, 0], xy_text[:, 1], :]
+        )
+        boxes = np.zeros((text_box_restored.shape[0], 9), dtype=np.float32)
+        boxes[:, :8] = text_box_restored.reshape((-1, 8))
+        boxes[:, 8] = score_map[xy_text[:, 0], xy_text[:, 1]]
+
+        try:
+            check_install("lanms", "lanms-nova")
+            import lanms
+
+            boxes = lanms.merge_quadrangle_n9(boxes, nms_thresh)
+        except:
+            print(
+                "You should install lanms by pip3 install lanms-nova to speed up nms_locality"
+            )
+            boxes = nms_locality(boxes.astype(np.float64), nms_thresh)
+        if boxes.shape[0] == 0:
+            return []
+        # Here we filter some low score boxes by the average score map,
+        #   this is different from the original paper.
+        for i, box in enumerate(boxes):
+            mask = np.zeros_like(score_map, dtype=np.uint8)
+            cv2.fillPoly(mask, box[:8].reshape((-1, 4, 2)).astype(np.int32) // 4, 1)
+            boxes[i, 8] = cv2.mean(score_map, mask)[0]
+        boxes = boxes[boxes[:, 8] > cover_thresh]
+        return boxes
+
+    def sort_poly(self, p):
+        """
+        Sort polygons.
+        """
+        min_axis = np.argmin(np.sum(p, axis=1))
+        p = p[[min_axis, (min_axis + 1) % 4, (min_axis + 2) % 4, (min_axis + 3) % 4]]
+        if abs(p[0, 0] - p[1, 0]) > abs(p[0, 1] - p[1, 1]):
+            return p
+        else:
+            return p[[0, 3, 2, 1]]
+
+    def __call__(self, outs_dict, shape_list):
+        score_list = outs_dict["f_score"]
+        geo_list = outs_dict["f_geo"]
+        if isinstance(score_list, paddle.Tensor):
+            score_list = score_list.numpy()
+            geo_list = geo_list.numpy()
+        img_num = len(shape_list)
+        dt_boxes_list = []
+        for ino in range(img_num):
+            score = score_list[ino]
+            geo = geo_list[ino]
+            boxes = self.detect(
+                score_map=score,
+                geo_map=geo,
+                score_thresh=self.score_thresh,
+                cover_thresh=self.cover_thresh,
+                nms_thresh=self.nms_thresh,
+            )
+            boxes_norm = []
+            if len(boxes) > 0:
+                h, w = score.shape[1:]
+                src_h, src_w, ratio_h, ratio_w = shape_list[ino]
+                boxes = boxes[:, :8].reshape((-1, 4, 2))
+                boxes[:, :, 0] /= ratio_w
+                boxes[:, :, 1] /= ratio_h
+                for i_box, box in enumerate(boxes):
+                    box = self.sort_poly(box.astype(np.int32))
+                    if (
+                        np.linalg.norm(box[0] - box[1]) < 5
+                        or np.linalg.norm(box[3] - box[0]) < 5
+                    ):
+                        continue
+                    boxes_norm.append(box)
+            dt_boxes_list.append({"points": np.array(boxes_norm)})
+        return dt_boxes_list
diff --git a/ppocr/postprocess/fce_postprocess.py b/ppocr/postprocess/fce_postprocess.py
new file mode 100755
index 0000000..bff9316
--- /dev/null
+++ b/ppocr/postprocess/fce_postprocess.py
@@ -0,0 +1,250 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/open-mmlab/mmocr/blob/v0.3.0/mmocr/models/textdet/postprocess/wrapper.py
+"""
+
+import cv2
+import paddle
+import numpy as np
+from numpy.fft import ifft
+from ppocr.utils.poly_nms import poly_nms, valid_boundary
+
+
+def fill_hole(input_mask):
+    h, w = input_mask.shape
+    canvas = np.zeros((h + 2, w + 2), np.uint8)
+    canvas[1 : h + 1, 1 : w + 1] = input_mask.copy()
+
+    mask = np.zeros((h + 4, w + 4), np.uint8)
+
+    cv2.floodFill(canvas, mask, (0, 0), 1)
+    canvas = canvas[1 : h + 1, 1 : w + 1].astype(np.bool_)
+
+    return ~canvas | input_mask
+
+
+def fourier2poly(fourier_coeff, num_reconstr_points=50):
+    """Inverse Fourier transform
+    Args:
+        fourier_coeff (ndarray): Fourier coefficients shaped (n, 2k+1),
+            with n and k being candidates number and Fourier degree
+            respectively.
+        num_reconstr_points (int): Number of reconstructed polygon points.
+    Returns:
+        Polygons (ndarray): The reconstructed polygons shaped (n, n')
+    """
+
+    a = np.zeros((len(fourier_coeff), num_reconstr_points), dtype="complex")
+    k = (len(fourier_coeff[0]) - 1) // 2
+
+    a[:, 0 : k + 1] = fourier_coeff[:, k:]
+    a[:, -k:] = fourier_coeff[:, :k]
+
+    poly_complex = ifft(a) * num_reconstr_points
+    polygon = np.zeros((len(fourier_coeff), num_reconstr_points, 2))
+    polygon[:, :, 0] = poly_complex.real
+    polygon[:, :, 1] = poly_complex.imag
+    return polygon.astype("int32").reshape((len(fourier_coeff), -1))
+
+
+class FCEPostProcess(object):
+    """
+    The post process for FCENet.
+    """
+
+    def __init__(
+        self,
+        scales,
+        fourier_degree=5,
+        num_reconstr_points=50,
+        decoding_type="fcenet",
+        score_thr=0.3,
+        nms_thr=0.1,
+        alpha=1.0,
+        beta=1.0,
+        box_type="poly",
+        **kwargs,
+    ):
+        self.scales = scales
+        self.fourier_degree = fourier_degree
+        self.num_reconstr_points = num_reconstr_points
+        self.decoding_type = decoding_type
+        self.score_thr = score_thr
+        self.nms_thr = nms_thr
+        self.alpha = alpha
+        self.beta = beta
+        self.box_type = box_type
+
+    def __call__(self, preds, shape_list):
+        score_maps = []
+        for key, value in preds.items():
+            if isinstance(value, paddle.Tensor):
+                value = value.numpy()
+            cls_res = value[:, :4, :, :]
+            reg_res = value[:, 4:, :, :]
+            score_maps.append([cls_res, reg_res])
+
+        return self.get_boundary(score_maps, shape_list)
+
+    def resize_boundary(self, boundaries, scale_factor):
+        """Rescale boundaries via scale_factor.
+
+        Args:
+            boundaries (list[list[float]]): The boundary list. Each boundary
+            with size 2k+1 with k>=4.
+            scale_factor(ndarray): The scale factor of size (4,).
+
+        Returns:
+            boundaries (list[list[float]]): The scaled boundaries.
+        """
+        boxes = []
+        scores = []
+        for b in boundaries:
+            sz = len(b)
+            valid_boundary(b, True)
+            scores.append(b[-1])
+            b = (
+                (
+                    np.array(b[: sz - 1])
+                    * (np.tile(scale_factor[:2], int((sz - 1) / 2)).reshape(1, sz - 1))
+                )
+                .flatten()
+                .tolist()
+            )
+            boxes.append(np.array(b).reshape([-1, 2]))
+
+        return np.array(boxes, dtype=np.float32), scores
+
+    def get_boundary(self, score_maps, shape_list):
+        assert len(score_maps) == len(self.scales)
+        boundaries = []
+        for idx, score_map in enumerate(score_maps):
+            scale = self.scales[idx]
+            boundaries = boundaries + self._get_boundary_single(score_map, scale)
+
+        # nms
+        boundaries = poly_nms(boundaries, self.nms_thr)
+        boundaries, scores = self.resize_boundary(
+            boundaries, (1 / shape_list[0, 2:]).tolist()[::-1]
+        )
+
+        boxes_batch = [dict(points=boundaries, scores=scores)]
+        return boxes_batch
+
+    def _get_boundary_single(self, score_map, scale):
+        assert len(score_map) == 2
+        assert score_map[1].shape[1] == 4 * self.fourier_degree + 2
+
+        return self.fcenet_decode(
+            preds=score_map,
+            fourier_degree=self.fourier_degree,
+            num_reconstr_points=self.num_reconstr_points,
+            scale=scale,
+            alpha=self.alpha,
+            beta=self.beta,
+            box_type=self.box_type,
+            score_thr=self.score_thr,
+            nms_thr=self.nms_thr,
+        )
+
+    def fcenet_decode(
+        self,
+        preds,
+        fourier_degree,
+        num_reconstr_points,
+        scale,
+        alpha=1.0,
+        beta=2.0,
+        box_type="poly",
+        score_thr=0.3,
+        nms_thr=0.1,
+    ):
+        """Decoding predictions of FCENet to instances.
+
+        Args:
+            preds (list(Tensor)): The head output tensors.
+            fourier_degree (int): The maximum Fourier transform degree k.
+            num_reconstr_points (int): The points number of the polygon
+                reconstructed from predicted Fourier coefficients.
+            scale (int): The down-sample scale of the prediction.
+            alpha (float) : The parameter to calculate final scores. Score_{final}
+                    = (Score_{text region} ^ alpha)
+                    * (Score_{text center region}^ beta)
+            beta (float) : The parameter to calculate final score.
+            box_type (str):  Boundary encoding type 'poly' or 'quad'.
+            score_thr (float) : The threshold used to filter out the final
+                candidates.
+            nms_thr (float) :  The threshold of nms.
+
+        Returns:
+            boundaries (list[list[float]]): The instance boundary and confidence
+                list.
+        """
+        assert isinstance(preds, list)
+        assert len(preds) == 2
+        assert box_type in ["poly", "quad"]
+
+        cls_pred = preds[0][0]
+        tr_pred = cls_pred[0:2]
+        tcl_pred = cls_pred[2:]
+
+        reg_pred = preds[1][0].transpose([1, 2, 0])
+        x_pred = reg_pred[:, :, : 2 * fourier_degree + 1]
+        y_pred = reg_pred[:, :, 2 * fourier_degree + 1 :]
+
+        score_pred = (tr_pred[1] ** alpha) * (tcl_pred[1] ** beta)
+        tr_pred_mask = (score_pred) > score_thr
+        tr_mask = fill_hole(tr_pred_mask)
+
+        tr_contours, _ = cv2.findContours(
+            tr_mask.astype(np.uint8), cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE
+        )  # opencv4
+
+        mask = np.zeros_like(tr_mask)
+        boundaries = []
+        for cont in tr_contours:
+            deal_map = mask.copy().astype(np.int8)
+            cv2.drawContours(deal_map, [cont], -1, 1, -1)
+
+            score_map = score_pred * deal_map
+            score_mask = score_map > 0
+            xy_text = np.argwhere(score_mask)
+            dxy = xy_text[:, 1] + xy_text[:, 0] * 1j
+
+            x, y = x_pred[score_mask], y_pred[score_mask]
+            c = x + y * 1j
+            c[:, fourier_degree] = c[:, fourier_degree] + dxy
+            c *= scale
+
+            polygons = fourier2poly(c, num_reconstr_points)
+            score = score_map[score_mask].reshape(-1, 1)
+            polygons = poly_nms(np.hstack((polygons, score)).tolist(), nms_thr)
+
+            boundaries = boundaries + polygons
+
+        boundaries = poly_nms(boundaries, nms_thr)
+
+        if box_type == "quad":
+            new_boundaries = []
+            for boundary in boundaries:
+                poly = np.array(boundary[:-1]).reshape(-1, 2).astype(np.float32)
+                score = boundary[-1]
+                points = cv2.boxPoints(cv2.minAreaRect(poly))
+                points = np.int64(points)
+                new_boundaries.append(points.reshape(-1).tolist() + [score])
+                boundaries = new_boundaries
+
+        return boundaries
diff --git a/ppocr/postprocess/locality_aware_nms.py b/ppocr/postprocess/locality_aware_nms.py
new file mode 100644
index 0000000..296ea6f
--- /dev/null
+++ b/ppocr/postprocess/locality_aware_nms.py
@@ -0,0 +1,198 @@
+"""
+Locality aware nms.
+This code is referred from: https://github.com/songdejia/EAST/blob/master/locality_aware_nms.py
+"""
+
+import numpy as np
+from shapely.geometry import Polygon
+
+
+def intersection(g, p):
+    """
+    Intersection.
+    """
+    g = Polygon(g[:8].reshape((4, 2)))
+    p = Polygon(p[:8].reshape((4, 2)))
+    g = g.buffer(0)
+    p = p.buffer(0)
+    if not g.is_valid or not p.is_valid:
+        return 0
+    inter = Polygon(g).intersection(Polygon(p)).area
+    union = g.area + p.area - inter
+    if union == 0:
+        return 0
+    else:
+        return inter / union
+
+
+def intersection_iog(g, p):
+    """
+    Intersection_iog.
+    """
+    g = Polygon(g[:8].reshape((4, 2)))
+    p = Polygon(p[:8].reshape((4, 2)))
+    if not g.is_valid or not p.is_valid:
+        return 0
+    inter = Polygon(g).intersection(Polygon(p)).area
+    # union = g.area + p.area - inter
+    union = p.area
+    if union == 0:
+        print("p_area is very small")
+        return 0
+    else:
+        return inter / union
+
+
+def weighted_merge(g, p):
+    """
+    Weighted merge.
+    """
+    g[:8] = (g[8] * g[:8] + p[8] * p[:8]) / (g[8] + p[8])
+    g[8] = g[8] + p[8]
+    return g
+
+
+def standard_nms(S, thres):
+    """
+    Standard nms.
+    """
+    order = np.argsort(S[:, 8])[::-1]
+    keep = []
+    while order.size > 0:
+        i = order[0]
+        keep.append(i)
+        ovr = np.array([intersection(S[i], S[t]) for t in order[1:]])
+
+        inds = np.where(ovr <= thres)[0]
+        order = order[inds + 1]
+
+    return S[keep]
+
+
+def standard_nms_inds(S, thres):
+    """
+    Standard nms, return inds.
+    """
+    order = np.argsort(S[:, 8])[::-1]
+    keep = []
+    while order.size > 0:
+        i = order[0]
+        keep.append(i)
+        ovr = np.array([intersection(S[i], S[t]) for t in order[1:]])
+
+        inds = np.where(ovr <= thres)[0]
+        order = order[inds + 1]
+
+    return keep
+
+
+def nms(S, thres):
+    """
+    nms.
+    """
+    order = np.argsort(S[:, 8])[::-1]
+    keep = []
+    while order.size > 0:
+        i = order[0]
+        keep.append(i)
+        ovr = np.array([intersection(S[i], S[t]) for t in order[1:]])
+
+        inds = np.where(ovr <= thres)[0]
+        order = order[inds + 1]
+
+    return keep
+
+
+def soft_nms(boxes_in, Nt_thres=0.3, threshold=0.8, sigma=0.5, method=2):
+    """
+    soft_nms
+    :para boxes_in, N x 9 (coords + score)
+    :para threshould, eliminate cases min score(0.001)
+    :para Nt_thres, iou_threshi
+    :para sigma, gaussian weght
+    :method, linear or gaussian
+    """
+    boxes = boxes_in.copy()
+    N = boxes.shape[0]
+    if N is None or N < 1:
+        return np.array([])
+    pos, maxpos = 0, 0
+    weight = 0.0
+    inds = np.arange(N)
+    tbox, sbox = boxes[0].copy(), boxes[0].copy()
+    for i in range(N):
+        maxscore = boxes[i, 8]
+        maxpos = i
+        tbox = boxes[i].copy()
+        ti = inds[i]
+        pos = i + 1
+        # get max box
+        while pos < N:
+            if maxscore < boxes[pos, 8]:
+                maxscore = boxes[pos, 8]
+                maxpos = pos
+            pos = pos + 1
+        # add max box as a detection
+        boxes[i, :] = boxes[maxpos, :]
+        inds[i] = inds[maxpos]
+        # swap
+        boxes[maxpos, :] = tbox
+        inds[maxpos] = ti
+        tbox = boxes[i].copy()
+        pos = i + 1
+        # NMS iteration
+        while pos < N:
+            sbox = boxes[pos].copy()
+            ts_iou_val = intersection(tbox, sbox)
+            if ts_iou_val > 0:
+                if method == 1:
+                    if ts_iou_val > Nt_thres:
+                        weight = 1 - ts_iou_val
+                    else:
+                        weight = 1
+                elif method == 2:
+                    weight = np.exp(-1.0 * ts_iou_val**2 / sigma)
+                else:
+                    if ts_iou_val > Nt_thres:
+                        weight = 0
+                    else:
+                        weight = 1
+                boxes[pos, 8] = weight * boxes[pos, 8]
+                # if box score falls below threshold, discard the box by
+                # swapping last box update N
+                if boxes[pos, 8] < threshold:
+                    boxes[pos, :] = boxes[N - 1, :]
+                    inds[pos] = inds[N - 1]
+                    N = N - 1
+                    pos = pos - 1
+            pos = pos + 1
+
+    return boxes[:N]
+
+
+def nms_locality(polys, thres=0.3):
+    """
+    locality aware nms of EAST
+    :param polys: a N*9 numpy array. first 8 coordinates, then prob
+    :return: boxes after nms
+    """
+    S = []
+    p = None
+    for g in polys:
+        if p is not None and intersection(g, p) > thres:
+            p = weighted_merge(g, p)
+        else:
+            if p is not None:
+                S.append(p)
+            p = g
+    if p is not None:
+        S.append(p)
+
+    if len(S) == 0:
+        return np.array([])
+    return standard_nms(np.array(S), thres)
+
+
+if __name__ == "__main__":
+    # 343,350,448,135,474,143,369,359
+    print(Polygon(np.array([[343, 350], [448, 135], [474, 143], [369, 359]])).area)
diff --git a/ppocr/postprocess/pg_postprocess.py b/ppocr/postprocess/pg_postprocess.py
new file mode 100644
index 0000000..70d1bdf
--- /dev/null
+++ b/ppocr/postprocess/pg_postprocess.py
@@ -0,0 +1,61 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+__dir__ = os.path.dirname(__file__)
+sys.path.append(__dir__)
+sys.path.append(os.path.join(__dir__, ".."))
+from ppocr.utils.e2e_utils.pgnet_pp_utils import PGNet_PostProcess
+
+
+class PGPostProcess(object):
+    """
+    The post process for PGNet.
+    """
+
+    def __init__(
+        self,
+        character_dict_path,
+        valid_set,
+        score_thresh,
+        mode,
+        point_gather_mode=None,
+        **kwargs,
+    ):
+        self.character_dict_path = character_dict_path
+        self.valid_set = valid_set
+        self.score_thresh = score_thresh
+        self.mode = mode
+        self.point_gather_mode = point_gather_mode
+
+    def __call__(self, outs_dict, shape_list):
+        post = PGNet_PostProcess(
+            self.character_dict_path,
+            self.valid_set,
+            self.score_thresh,
+            outs_dict,
+            shape_list,
+            point_gather_mode=self.point_gather_mode,
+        )
+        if self.mode == "fast":
+            data = post.pg_postprocess_fast()
+        else:
+            data = post.pg_postprocess_slow()
+        return data
diff --git a/ppocr/postprocess/picodet_postprocess.py b/ppocr/postprocess/picodet_postprocess.py
new file mode 100644
index 0000000..9189c8e
--- /dev/null
+++ b/ppocr/postprocess/picodet_postprocess.py
@@ -0,0 +1,297 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import numpy as np
+from scipy.special import softmax
+
+
+def hard_nms(box_scores, iou_threshold, top_k=-1, candidate_size=200):
+    """
+    Args:
+        box_scores (N, 5): boxes in corner-form and probabilities.
+        iou_threshold: intersection over union threshold.
+        top_k: keep top_k results. If k <= 0, keep all the results.
+        candidate_size: only consider the candidates with the highest scores.
+    Returns:
+         picked: a list of indexes of the kept boxes
+    """
+    scores = box_scores[:, -1]
+    boxes = box_scores[:, :-1]
+    picked = []
+    indexes = np.argsort(scores)
+    indexes = indexes[-candidate_size:]
+    while len(indexes) > 0:
+        current = indexes[-1]
+        picked.append(current)
+        if 0 < top_k == len(picked) or len(indexes) == 1:
+            break
+        current_box = boxes[current, :]
+        indexes = indexes[:-1]
+        rest_boxes = boxes[indexes, :]
+        iou = iou_of(
+            rest_boxes,
+            np.expand_dims(current_box, axis=0),
+        )
+        indexes = indexes[iou <= iou_threshold]
+
+    return box_scores[picked, :]
+
+
+def iou_of(boxes0, boxes1, eps=1e-5):
+    """Return intersection-over-union (Jaccard index) of boxes.
+    Args:
+        boxes0 (N, 4): ground truth boxes.
+        boxes1 (N or 1, 4): predicted boxes.
+        eps: a small number to avoid 0 as denominator.
+    Returns:
+        iou (N): IoU values.
+    """
+    overlap_left_top = np.maximum(boxes0[..., :2], boxes1[..., :2])
+    overlap_right_bottom = np.minimum(boxes0[..., 2:], boxes1[..., 2:])
+
+    overlap_area = area_of(overlap_left_top, overlap_right_bottom)
+    area0 = area_of(boxes0[..., :2], boxes0[..., 2:])
+    area1 = area_of(boxes1[..., :2], boxes1[..., 2:])
+    return overlap_area / (area0 + area1 - overlap_area + eps)
+
+
+def area_of(left_top, right_bottom):
+    """Compute the areas of rectangles given two corners.
+    Args:
+        left_top (N, 2): left top corner.
+        right_bottom (N, 2): right bottom corner.
+    Returns:
+        area (N): return the area.
+    """
+    hw = np.clip(right_bottom - left_top, 0.0, None)
+    return hw[..., 0] * hw[..., 1]
+
+
+def calculate_containment(boxes0, boxes1):
+    """
+    Calculate the containment of the boxes.
+    Args:
+        boxes0 (N, 4): ground truth boxes.
+        boxes1 (N or 1, 4): predicted boxes.
+    Returns:
+        containment (N): containment values.
+    """
+    overlap_left_top = np.maximum(boxes0[..., :2], boxes1[..., :2])
+    overlap_right_bottom = np.minimum(boxes0[..., 2:], boxes1[..., 2:])
+
+    overlap_area = area_of(overlap_left_top, overlap_right_bottom)
+    area0 = area_of(boxes0[..., :2], boxes0[..., 2:])
+    area1 = area_of(boxes1[..., :2], boxes1[..., 2:])
+    return overlap_area / np.minimum(area0, np.expand_dims(area1, axis=0))
+
+
+class PicoDetPostProcess(object):
+    """
+    Args:
+        input_shape (int): network input image size
+        ori_shape (int): ori image shape of before padding
+        scale_factor (float): scale factor of ori image
+        enable_mkldnn (bool): whether to open MKLDNN
+    """
+
+    def __init__(
+        self,
+        layout_dict_path,
+        strides=[8, 16, 32, 64],
+        score_threshold=0.4,
+        nms_threshold=0.5,
+        nms_top_k=1000,
+        keep_top_k=100,
+    ):
+        self.labels = self.load_layout_dict(layout_dict_path)
+        self.strides = strides
+        self.score_threshold = score_threshold
+        self.nms_threshold = nms_threshold
+        self.nms_top_k = nms_top_k
+        self.keep_top_k = keep_top_k
+
+    def load_layout_dict(self, layout_dict_path):
+        with open(layout_dict_path, "r", encoding="utf-8") as fp:
+            labels = fp.readlines()
+        return [label.strip("\n") for label in labels]
+
+    def warp_boxes(self, boxes, ori_shape):
+        """Apply transform to boxes"""
+        width, height = ori_shape[1], ori_shape[0]
+        n = len(boxes)
+        if n:
+            # warp points
+            xy = np.ones((n * 4, 3))
+            xy[:, :2] = boxes[:, [0, 1, 2, 3, 0, 3, 2, 1]].reshape(
+                n * 4, 2
+            )  # x1y1, x2y2, x1y2, x2y1
+            # xy = xy @ M.T  # transform
+            xy = (xy[:, :2] / xy[:, 2:3]).reshape(n, 8)  # rescale
+            # create new boxes
+            x = xy[:, [0, 2, 4, 6]]
+            y = xy[:, [1, 3, 5, 7]]
+            xy = (
+                np.concatenate((x.min(1), y.min(1), x.max(1), y.max(1))).reshape(4, n).T
+            )
+            # clip boxes
+            xy[:, [0, 2]] = xy[:, [0, 2]].clip(0, width)
+            xy[:, [1, 3]] = xy[:, [1, 3]].clip(0, height)
+            return xy.astype(np.float32)
+        else:
+            return boxes
+
+    def img_info(self, ori_img, img):
+        origin_shape = ori_img.shape
+        resize_shape = img.shape
+        im_scale_y = resize_shape[2] / float(origin_shape[0])
+        im_scale_x = resize_shape[3] / float(origin_shape[1])
+        scale_factor = np.array([im_scale_y, im_scale_x], dtype=np.float32)
+        img_shape = np.array(img.shape[2:], dtype=np.float32)
+
+        input_shape = np.array(img).astype("float32").shape[2:]
+        ori_shape = np.array((img_shape,)).astype("float32")
+        scale_factor = np.array((scale_factor,)).astype("float32")
+        return ori_shape, input_shape, scale_factor
+
+    def __call__(self, ori_img, img, preds):
+        scores, raw_boxes = preds["boxes"], preds["boxes_num"]
+        batch_size = raw_boxes[0].shape[0]
+        reg_max = int(raw_boxes[0].shape[-1] / 4 - 1)
+        out_boxes_num = []
+        out_boxes_list = []
+        results = []
+        ori_shape, input_shape, scale_factor = self.img_info(ori_img, img)
+
+        for batch_id in range(batch_size):
+            # generate centers
+            decode_boxes = []
+            select_scores = []
+            for stride, box_distribute, score in zip(self.strides, raw_boxes, scores):
+                box_distribute = box_distribute[batch_id]
+                score = score[batch_id]
+                # centers
+                fm_h = input_shape[0] / stride
+                fm_w = input_shape[1] / stride
+                h_range = np.arange(fm_h)
+                w_range = np.arange(fm_w)
+                ww, hh = np.meshgrid(w_range, h_range)
+                ct_row = (hh.flatten() + 0.5) * stride
+                ct_col = (ww.flatten() + 0.5) * stride
+                center = np.stack((ct_col, ct_row, ct_col, ct_row), axis=1)
+
+                # box distribution to distance
+                reg_range = np.arange(reg_max + 1)
+                box_distance = box_distribute.reshape((-1, reg_max + 1))
+                box_distance = softmax(box_distance, axis=1)
+                box_distance = box_distance * np.expand_dims(reg_range, axis=0)
+                box_distance = np.sum(box_distance, axis=1).reshape((-1, 4))
+                box_distance = box_distance * stride
+
+                # top K candidate
+                topk_idx = np.argsort(score.max(axis=1))[::-1]
+                topk_idx = topk_idx[: self.nms_top_k]
+                center = center[topk_idx]
+                score = score[topk_idx]
+                box_distance = box_distance[topk_idx]
+
+                # decode box
+                decode_box = center + [-1, -1, 1, 1] * box_distance
+
+                select_scores.append(score)
+                decode_boxes.append(decode_box)
+
+            # nms
+            bboxes = np.concatenate(decode_boxes, axis=0)
+            confidences = np.concatenate(select_scores, axis=0)
+            picked_box_probs = []
+            picked_labels = []
+            for class_index in range(0, confidences.shape[1]):
+                probs = confidences[:, class_index]
+                mask = probs > self.score_threshold
+                probs = probs[mask]
+                if probs.shape[0] == 0:
+                    continue
+                subset_boxes = bboxes[mask, :]
+                box_probs = np.concatenate([subset_boxes, probs.reshape(-1, 1)], axis=1)
+                box_probs = hard_nms(
+                    box_probs,
+                    iou_threshold=self.nms_threshold,
+                    top_k=self.keep_top_k,
+                )
+                picked_box_probs.append(box_probs)
+                picked_labels.extend([class_index] * box_probs.shape[0])
+
+            if len(picked_box_probs) == 0:
+                out_boxes_list.append(np.empty((0, 4)))
+                out_boxes_num.append(0)
+
+            else:
+                picked_box_probs = np.concatenate(picked_box_probs)
+
+                # resize output boxes
+                picked_box_probs[:, :4] = self.warp_boxes(
+                    picked_box_probs[:, :4], ori_shape[batch_id]
+                )
+                im_scale = np.concatenate(
+                    [scale_factor[batch_id][::-1], scale_factor[batch_id][::-1]]
+                )
+                picked_box_probs[:, :4] /= im_scale
+                # clas score box
+                out_boxes_list.append(
+                    np.concatenate(
+                        [
+                            np.expand_dims(np.array(picked_labels), axis=-1),
+                            np.expand_dims(picked_box_probs[:, 4], axis=-1),
+                            picked_box_probs[:, :4],
+                        ],
+                        axis=1,
+                    )
+                )
+                out_boxes_num.append(len(picked_labels))
+
+        out_boxes_list = np.concatenate(out_boxes_list, axis=0)
+        out_boxes_num = np.asarray(out_boxes_num).astype(np.int32)
+
+        for dt in out_boxes_list:
+            clsid, bbox, score = int(dt[0]), dt[2:], dt[1]
+            label = self.labels[clsid]
+            result = {"bbox": bbox, "label": label, "score": score}
+            results.append(result)
+
+        # Handle conflict where a box is simultaneously recognized as multiple labels.
+        # Use IoU to find similar boxes. Prioritize labels as table, text, and others when deduplicate similar boxes.
+        bboxes = np.array([x["bbox"] for x in results])
+        duplicate_idx = list()
+        for i in range(len(results)):
+            if i in duplicate_idx:
+                continue
+            containments = calculate_containment(bboxes, bboxes[i, ...])
+            overlaps = np.where(containments > 0.5)[0]
+            if len(overlaps) > 1:
+                table_box = [x for x in overlaps if results[x]["label"] == "table"]
+                if len(table_box) > 0:
+                    keep = sorted(
+                        [(x, results[x]) for x in table_box],
+                        key=lambda x: x[1]["score"],
+                        reverse=True,
+                    )[0][0]
+                else:
+                    keep = sorted(
+                        [(x, results[x]) for x in overlaps],
+                        key=lambda x: x[1]["score"],
+                        reverse=True,
+                    )[0][0]
+                duplicate_idx.extend([x for x in overlaps if x != keep])
+        results = [x for i, x in enumerate(results) if i not in duplicate_idx]
+        return results
diff --git a/ppocr/postprocess/pse_postprocess/__init__.py b/ppocr/postprocess/pse_postprocess/__init__.py
new file mode 100644
index 0000000..eed166a
--- /dev/null
+++ b/ppocr/postprocess/pse_postprocess/__init__.py
@@ -0,0 +1,15 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .pse_postprocess import PSEPostProcess
diff --git a/ppocr/postprocess/pse_postprocess/pse/README.md b/ppocr/postprocess/pse_postprocess/pse/README.md
new file mode 100644
index 0000000..6a19d5d
--- /dev/null
+++ b/ppocr/postprocess/pse_postprocess/pse/README.md
@@ -0,0 +1,6 @@
+## 编译
+This code is refer from:
+https://github.com/whai362/PSENet/blob/python3/models/post_processing/pse
+```python
+python3 setup.py build_ext --inplace
+```
diff --git a/ppocr/postprocess/pse_postprocess/pse/__init__.py b/ppocr/postprocess/pse_postprocess/pse/__init__.py
new file mode 100644
index 0000000..6028802
--- /dev/null
+++ b/ppocr/postprocess/pse_postprocess/pse/__init__.py
@@ -0,0 +1,33 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import sys
+import os
+import subprocess
+
+python_path = sys.executable
+
+ori_path = os.getcwd()
+os.chdir("ppocr/postprocess/pse_postprocess/pse")
+if (
+    subprocess.call("{} setup.py build_ext --inplace".format(python_path), shell=True)
+    != 0
+):
+    raise RuntimeError(
+        "Cannot compile pse: {}, if your system is windows, you need to install all the default components of `desktop development using C++` in visual studio 2019+".format(
+            os.path.dirname(os.path.realpath(__file__))
+        )
+    )
+os.chdir(ori_path)
+
+from .pse import pse
diff --git a/ppocr/postprocess/pse_postprocess/pse/pse.pyx b/ppocr/postprocess/pse_postprocess/pse/pse.pyx
new file mode 100644
index 0000000..dfa90f3
--- /dev/null
+++ b/ppocr/postprocess/pse_postprocess/pse/pse.pyx
@@ -0,0 +1,70 @@
+
+import numpy as np
+import cv2
+cimport numpy as np
+cimport cython
+cimport libcpp
+cimport libcpp.pair
+cimport libcpp.queue
+from libcpp.pair cimport *
+from libcpp.queue  cimport *
+
+@cython.boundscheck(False)
+@cython.wraparound(False)
+cdef np.ndarray[np.int32_t, ndim=2] _pse(np.ndarray[np.uint8_t, ndim=3] kernels,
+                                         np.ndarray[np.int32_t, ndim=2] label,
+                                         int kernel_num,
+                                         int label_num,
+                                         float min_area=0):
+    cdef np.ndarray[np.int32_t, ndim=2] pred
+    pred = np.zeros((label.shape[0], label.shape[1]), dtype=np.int32)
+
+    for label_idx in range(1, label_num):
+        if np.sum(label == label_idx) < min_area:
+            label[label == label_idx] = 0
+
+    cdef libcpp.queue.queue[libcpp.pair.pair[np.int16_t,np.int16_t]] que = \
+        queue[libcpp.pair.pair[np.int16_t,np.int16_t]]()
+    cdef libcpp.queue.queue[libcpp.pair.pair[np.int16_t,np.int16_t]] nxt_que = \
+        queue[libcpp.pair.pair[np.int16_t,np.int16_t]]()
+    cdef np.int16_t* dx = [-1, 1, 0, 0]
+    cdef np.int16_t* dy = [0, 0, -1, 1]
+    cdef np.int16_t tmpx, tmpy
+
+    points = np.array(np.where(label > 0)).transpose((1, 0))
+    for point_idx in range(points.shape[0]):
+        tmpx, tmpy = points[point_idx, 0], points[point_idx, 1]
+        que.push(pair[np.int16_t,np.int16_t](tmpx, tmpy))
+        pred[tmpx, tmpy] = label[tmpx, tmpy]
+
+    cdef libcpp.pair.pair[np.int16_t,np.int16_t] cur
+    cdef int cur_label
+    for kernel_idx in range(kernel_num - 1, -1, -1):
+        while not que.empty():
+            cur = que.front()
+            que.pop()
+            cur_label = pred[cur.first, cur.second]
+
+            is_edge = True
+            for j in range(4):
+                tmpx = cur.first + dx[j]
+                tmpy = cur.second + dy[j]
+                if tmpx < 0 or tmpx >= label.shape[0] or tmpy < 0 or tmpy >= label.shape[1]:
+                    continue
+                if kernels[kernel_idx, tmpx, tmpy] == 0 or pred[tmpx, tmpy] > 0:
+                    continue
+
+                que.push(pair[np.int16_t,np.int16_t](tmpx, tmpy))
+                pred[tmpx, tmpy] = cur_label
+                is_edge = False
+            if is_edge:
+                nxt_que.push(cur)
+
+        que, nxt_que = nxt_que, que
+
+    return pred
+
+def pse(kernels, min_area):
+    kernel_num = kernels.shape[0]
+    label_num, label = cv2.connectedComponents(kernels[-1], connectivity=4)
+    return _pse(kernels[:-1], label, kernel_num, label_num, min_area)
diff --git a/ppocr/postprocess/pse_postprocess/pse/setup.py b/ppocr/postprocess/pse_postprocess/pse/setup.py
new file mode 100644
index 0000000..2ae2d53
--- /dev/null
+++ b/ppocr/postprocess/pse_postprocess/pse/setup.py
@@ -0,0 +1,18 @@
+from setuptools import setup, Extension
+from Cython.Build import cythonize
+import numpy
+
+setup(
+    ext_modules=cythonize(
+        Extension(
+            "pse",
+            sources=["pse.pyx"],
+            language="c++",
+            include_dirs=[numpy.get_include()],
+            library_dirs=[],
+            libraries=[],
+            extra_compile_args=["-O3"],
+            extra_link_args=[],
+        )
+    )
+)
diff --git a/ppocr/postprocess/pse_postprocess/pse_postprocess.py b/ppocr/postprocess/pse_postprocess/pse_postprocess.py
new file mode 100755
index 0000000..4050dce
--- /dev/null
+++ b/ppocr/postprocess/pse_postprocess/pse_postprocess.py
@@ -0,0 +1,122 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/whai362/PSENet/blob/python3/models/head/psenet_head.py
+"""
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+import cv2
+import paddle
+from paddle.nn import functional as F
+
+from ppocr.postprocess.pse_postprocess.pse import pse
+
+
+class PSEPostProcess(object):
+    """
+    The post process for PSE.
+    """
+
+    def __init__(
+        self,
+        thresh=0.5,
+        box_thresh=0.85,
+        min_area=16,
+        box_type="quad",
+        scale=4,
+        **kwargs,
+    ):
+        assert box_type in ["quad", "poly"], "Only quad and poly is supported"
+        self.thresh = thresh
+        self.box_thresh = box_thresh
+        self.min_area = min_area
+        self.box_type = box_type
+        self.scale = scale
+
+    def __call__(self, outs_dict, shape_list):
+        pred = outs_dict["maps"]
+        if not isinstance(pred, paddle.Tensor):
+            pred = paddle.to_tensor(pred)
+        pred = F.interpolate(pred, scale_factor=4 // self.scale, mode="bilinear")
+
+        score = F.sigmoid(pred[:, 0, :, :])
+
+        kernels = (pred > self.thresh).astype("float32")
+        text_mask = kernels[:, 0, :, :]
+        text_mask = paddle.unsqueeze(text_mask, axis=1)
+
+        kernels[:, 0:, :, :] = kernels[:, 0:, :, :] * text_mask
+
+        score = score.numpy()
+        kernels = kernels.numpy().astype(np.uint8)
+
+        boxes_batch = []
+        for batch_index in range(pred.shape[0]):
+            boxes, scores = self.boxes_from_bitmap(
+                score[batch_index], kernels[batch_index], shape_list[batch_index]
+            )
+
+            boxes_batch.append({"points": boxes, "scores": scores})
+        return boxes_batch
+
+    def boxes_from_bitmap(self, score, kernels, shape):
+        label = pse(kernels, self.min_area)
+        return self.generate_box(score, label, shape)
+
+    def generate_box(self, score, label, shape):
+        src_h, src_w, ratio_h, ratio_w = shape
+        label_num = np.max(label) + 1
+
+        boxes = []
+        scores = []
+        for i in range(1, label_num):
+            ind = label == i
+            points = np.array(np.where(ind)).transpose((1, 0))[:, ::-1]
+
+            if points.shape[0] < self.min_area:
+                label[ind] = 0
+                continue
+
+            score_i = np.mean(score[ind])
+            if score_i < self.box_thresh:
+                label[ind] = 0
+                continue
+
+            if self.box_type == "quad":
+                rect = cv2.minAreaRect(points)
+                bbox = cv2.boxPoints(rect)
+            elif self.box_type == "poly":
+                box_height = np.max(points[:, 1]) + 10
+                box_width = np.max(points[:, 0]) + 10
+
+                mask = np.zeros((box_height, box_width), np.uint8)
+                mask[points[:, 1], points[:, 0]] = 255
+
+                contours, _ = cv2.findContours(
+                    mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE
+                )
+                bbox = np.squeeze(contours[0], 1)
+            else:
+                raise NotImplementedError
+
+            bbox[:, 0] = np.clip(np.round(bbox[:, 0] / ratio_w), 0, src_w)
+            bbox[:, 1] = np.clip(np.round(bbox[:, 1] / ratio_h), 0, src_h)
+            boxes.append(bbox)
+            scores.append(score_i)
+        return boxes, scores
diff --git a/ppocr/postprocess/rec_postprocess.py b/ppocr/postprocess/rec_postprocess.py
new file mode 100644
index 0000000..4700bbf
--- /dev/null
+++ b/ppocr/postprocess/rec_postprocess.py
@@ -0,0 +1,1565 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import numpy as np
+import paddle
+from paddle.nn import functional as F
+import re
+import json
+
+
+class BaseRecLabelDecode(object):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, character_dict_path=None, use_space_char=False):
+        self.beg_str = "sos"
+        self.end_str = "eos"
+        self.reverse = False
+        self.character_str = []
+
+        if character_dict_path is None:
+            self.character_str = "0123456789abcdefghijklmnopqrstuvwxyz"
+            dict_character = list(self.character_str)
+        else:
+            with open(character_dict_path, "rb") as fin:
+                lines = fin.readlines()
+                for line in lines:
+                    line = line.decode("utf-8").strip("\n").strip("\r\n")
+                    self.character_str.append(line)
+            if use_space_char:
+                self.character_str.append(" ")
+            dict_character = list(self.character_str)
+            if "arabic" in character_dict_path:
+                self.reverse = True
+
+        dict_character = self.add_special_char(dict_character)
+        self.dict = {}
+        for i, char in enumerate(dict_character):
+            self.dict[char] = i
+        self.character = dict_character
+
+    def pred_reverse(self, pred):
+        pred_re = []
+        c_current = ""
+        for c in pred:
+            if not bool(re.search("[a-zA-Z0-9 :*./%+-]", c)):
+                if c_current != "":
+                    pred_re.append(c_current)
+                pred_re.append(c)
+                c_current = ""
+            else:
+                c_current += c
+        if c_current != "":
+            pred_re.append(c_current)
+
+        return "".join(pred_re[::-1])
+
+    def add_special_char(self, dict_character):
+        return dict_character
+
+    def get_word_info(self, text, selection):
+        """
+        Group the decoded characters and record the corresponding decoded positions.
+
+        Args:
+            text: the decoded text
+            selection: the bool array that identifies which columns of features are decoded as non-separated characters
+        Returns:
+            word_list: list of the grouped words
+            word_col_list: list of decoding positions corresponding to each character in the grouped word
+            state_list: list of marker to identify the type of grouping words, including two types of grouping words:
+                        - 'cn': continuous chinese characters (e.g., 你好啊)
+                        - 'en&num': continuous english characters (e.g., hello), number (e.g., 123, 1.123), or mixed of them connected by '-' (e.g., VGG-16)
+                        The remaining characters in text are treated as separators between groups (e.g., space, '(', ')', etc.).
+        """
+        state = None
+        word_content = []
+        word_col_content = []
+        word_list = []
+        word_col_list = []
+        state_list = []
+        valid_col = np.where(selection == True)[0]
+
+        for c_i, char in enumerate(text):
+            if "\u4e00" <= char <= "\u9fff":
+                c_state = "cn"
+            # Use \w with UNICODE flag to match letters (including accented chars like ä, ö, ü, é, etc.) and digits
+            # Exclude underscore since \w includes it but we want to treat it as splitter
+            elif bool(re.search(r"[\w]", char, re.UNICODE)) and char != "_":
+                c_state = "en&num"
+            else:
+                c_state = "splitter"
+
+            # Handle apostrophes in French words like "n'êtes"
+            if char == "'" and state == "en&num":
+                c_state = "en&num"
+
+            if (
+                char == "."
+                and state == "en&num"
+                and c_i + 1 < len(text)
+                and bool(re.search("[0-9]", text[c_i + 1]))
+            ):  # grouping floating number
+                c_state = "en&num"
+            if (
+                char == "-" and state == "en&num"
+            ):  # grouping word with '-', such as 'state-of-the-art'
+                c_state = "en&num"
+
+            if state == None:
+                state = c_state
+
+            if state != c_state:
+                if len(word_content) != 0:
+                    word_list.append(word_content)
+                    word_col_list.append(word_col_content)
+                    state_list.append(state)
+                    word_content = []
+                    word_col_content = []
+                state = c_state
+
+            if state != "splitter":
+                word_content.append(char)
+                word_col_content.append(valid_col[c_i])
+
+        if len(word_content) != 0:
+            word_list.append(word_content)
+            word_col_list.append(word_col_content)
+            state_list.append(state)
+
+        return word_list, word_col_list, state_list
+
+    def decode(
+        self,
+        text_index,
+        text_prob=None,
+        is_remove_duplicate=False,
+        return_word_box=False,
+    ):
+        """convert text-index into text-label."""
+        result_list = []
+        ignored_tokens = self.get_ignored_tokens()
+        batch_size = len(text_index)
+        for batch_idx in range(batch_size):
+            selection = np.ones(len(text_index[batch_idx]), dtype=bool)
+            if is_remove_duplicate:
+                selection[1:] = text_index[batch_idx][1:] != text_index[batch_idx][:-1]
+            for ignored_token in ignored_tokens:
+                selection &= text_index[batch_idx] != ignored_token
+
+            char_list = [
+                self.character[text_id] for text_id in text_index[batch_idx][selection]
+            ]
+            if text_prob is not None:
+                conf_list = text_prob[batch_idx][selection]
+            else:
+                conf_list = [1] * len(selection)
+            if len(conf_list) == 0:
+                conf_list = [0]
+
+            text = "".join(char_list)
+
+            if self.reverse:  # for arabic rec
+                text = self.pred_reverse(text)
+
+            if return_word_box:
+                word_list, word_col_list, state_list = self.get_word_info(
+                    text, selection
+                )
+                result_list.append(
+                    (
+                        text,
+                        np.mean(conf_list).tolist(),
+                        [
+                            len(text_index[batch_idx]),
+                            word_list,
+                            word_col_list,
+                            state_list,
+                        ],
+                    )
+                )
+            else:
+                result_list.append((text, np.mean(conf_list).tolist()))
+        return result_list
+
+    def get_ignored_tokens(self):
+        return [0]  # for ctc blank
+
+
+class CTCLabelDecode(BaseRecLabelDecode):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, character_dict_path=None, use_space_char=False, **kwargs):
+        super(CTCLabelDecode, self).__init__(character_dict_path, use_space_char)
+
+    def __call__(self, preds, label=None, return_word_box=False, *args, **kwargs):
+        if isinstance(preds, tuple) or isinstance(preds, list):
+            preds = preds[-1]
+        if isinstance(preds, paddle.Tensor):
+            preds = preds.numpy()
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+        text = self.decode(
+            preds_idx,
+            preds_prob,
+            is_remove_duplicate=True,
+            return_word_box=return_word_box,
+        )
+        if return_word_box:
+            for rec_idx, rec in enumerate(text):
+                wh_ratio = kwargs["wh_ratio_list"][rec_idx]
+                max_wh_ratio = kwargs["max_wh_ratio"]
+                rec[2][0] = rec[2][0] * (wh_ratio / max_wh_ratio)
+        if label is None:
+            return text
+        label = self.decode(label)
+        return text, label
+
+    def add_special_char(self, dict_character):
+        dict_character = ["blank"] + dict_character
+        return dict_character
+
+
+class DistillationCTCLabelDecode(CTCLabelDecode):
+    """
+    Convert
+    Convert between text-label and text-index
+    """
+
+    def __init__(
+        self,
+        character_dict_path=None,
+        use_space_char=False,
+        model_name=["student"],
+        key=None,
+        multi_head=False,
+        **kwargs,
+    ):
+        super(DistillationCTCLabelDecode, self).__init__(
+            character_dict_path, use_space_char
+        )
+        if not isinstance(model_name, list):
+            model_name = [model_name]
+        self.model_name = model_name
+
+        self.key = key
+        self.multi_head = multi_head
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        output = dict()
+        for name in self.model_name:
+            pred = preds[name]
+            if self.key is not None:
+                pred = pred[self.key]
+            if self.multi_head and isinstance(pred, dict):
+                pred = pred["ctc"]
+            output[name] = super().__call__(pred, label=label, *args, **kwargs)
+        return output
+
+
+class AttnLabelDecode(BaseRecLabelDecode):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, character_dict_path=None, use_space_char=False, **kwargs):
+        super(AttnLabelDecode, self).__init__(character_dict_path, use_space_char)
+
+    def add_special_char(self, dict_character):
+        self.beg_str = "sos"
+        self.end_str = "eos"
+        dict_character = dict_character
+        dict_character = [self.beg_str] + dict_character + [self.end_str]
+        return dict_character
+
+    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
+        """convert text-index into text-label."""
+        result_list = []
+        ignored_tokens = self.get_ignored_tokens()
+        [beg_idx, end_idx] = self.get_ignored_tokens()
+        batch_size = len(text_index)
+        for batch_idx in range(batch_size):
+            char_list = []
+            conf_list = []
+            for idx in range(len(text_index[batch_idx])):
+                if text_index[batch_idx][idx] in ignored_tokens:
+                    continue
+                if int(text_index[batch_idx][idx]) == int(end_idx):
+                    break
+                if is_remove_duplicate:
+                    # only for predict
+                    if (
+                        idx > 0
+                        and text_index[batch_idx][idx - 1] == text_index[batch_idx][idx]
+                    ):
+                        continue
+                char_list.append(self.character[int(text_index[batch_idx][idx])])
+                if text_prob is not None:
+                    conf_list.append(text_prob[batch_idx][idx])
+                else:
+                    conf_list.append(1)
+            text = "".join(char_list)
+            result_list.append((text, np.mean(conf_list).tolist()))
+        return result_list
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        """
+        text = self.decode(text)
+        if label is None:
+            return text
+        else:
+            label = self.decode(label, is_remove_duplicate=False)
+            return text, label
+        """
+        if isinstance(preds, paddle.Tensor):
+            preds = preds.numpy()
+
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+        if label is None:
+            return text
+        label = self.decode(label, is_remove_duplicate=False)
+        return text, label
+
+    def get_ignored_tokens(self):
+        beg_idx = self.get_beg_end_flag_idx("beg")
+        end_idx = self.get_beg_end_flag_idx("end")
+        return [beg_idx, end_idx]
+
+    def get_beg_end_flag_idx(self, beg_or_end):
+        if beg_or_end == "beg":
+            idx = np.array(self.dict[self.beg_str])
+        elif beg_or_end == "end":
+            idx = np.array(self.dict[self.end_str])
+        else:
+            assert False, "unsupported type %s in get_beg_end_flag_idx" % beg_or_end
+        return idx
+
+
+class RFLLabelDecode(BaseRecLabelDecode):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, character_dict_path=None, use_space_char=False, **kwargs):
+        super(RFLLabelDecode, self).__init__(character_dict_path, use_space_char)
+
+    def add_special_char(self, dict_character):
+        self.beg_str = "sos"
+        self.end_str = "eos"
+        dict_character = dict_character
+        dict_character = [self.beg_str] + dict_character + [self.end_str]
+        return dict_character
+
+    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
+        """convert text-index into text-label."""
+        result_list = []
+        ignored_tokens = self.get_ignored_tokens()
+        [beg_idx, end_idx] = self.get_ignored_tokens()
+        batch_size = len(text_index)
+        for batch_idx in range(batch_size):
+            char_list = []
+            conf_list = []
+            for idx in range(len(text_index[batch_idx])):
+                if text_index[batch_idx][idx] in ignored_tokens:
+                    continue
+                if int(text_index[batch_idx][idx]) == int(end_idx):
+                    break
+                if is_remove_duplicate:
+                    # only for predict
+                    if (
+                        idx > 0
+                        and text_index[batch_idx][idx - 1] == text_index[batch_idx][idx]
+                    ):
+                        continue
+                char_list.append(self.character[int(text_index[batch_idx][idx])])
+                if text_prob is not None:
+                    conf_list.append(text_prob[batch_idx][idx])
+                else:
+                    conf_list.append(1)
+            text = "".join(char_list)
+            result_list.append((text, np.mean(conf_list).tolist()))
+        return result_list
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        # if seq_outputs is not None:
+        if isinstance(preds, tuple) or isinstance(preds, list):
+            cnt_outputs, seq_outputs = preds
+            if isinstance(seq_outputs, paddle.Tensor):
+                seq_outputs = seq_outputs.numpy()
+            preds_idx = seq_outputs.argmax(axis=2)
+            preds_prob = seq_outputs.max(axis=2)
+            text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+
+            if label is None:
+                return text
+            label = self.decode(label, is_remove_duplicate=False)
+            return text, label
+
+        else:
+            cnt_outputs = preds
+            if isinstance(cnt_outputs, paddle.Tensor):
+                cnt_outputs = cnt_outputs.numpy()
+            cnt_length = []
+            for lens in cnt_outputs:
+                length = round(np.sum(lens))
+                cnt_length.append(length)
+            if label is None:
+                return cnt_length
+            label = self.decode(label, is_remove_duplicate=False)
+            length = [len(res[0]) for res in label]
+            return cnt_length, length
+
+    def get_ignored_tokens(self):
+        beg_idx = self.get_beg_end_flag_idx("beg")
+        end_idx = self.get_beg_end_flag_idx("end")
+        return [beg_idx, end_idx]
+
+    def get_beg_end_flag_idx(self, beg_or_end):
+        if beg_or_end == "beg":
+            idx = np.array(self.dict[self.beg_str])
+        elif beg_or_end == "end":
+            idx = np.array(self.dict[self.end_str])
+        else:
+            assert False, "unsupported type %s in get_beg_end_flag_idx" % beg_or_end
+        return idx
+
+
+class SEEDLabelDecode(BaseRecLabelDecode):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, character_dict_path=None, use_space_char=False, **kwargs):
+        super(SEEDLabelDecode, self).__init__(character_dict_path, use_space_char)
+
+    def add_special_char(self, dict_character):
+        self.padding_str = "padding"
+        self.end_str = "eos"
+        self.unknown = "unknown"
+        dict_character = dict_character + [self.end_str, self.padding_str, self.unknown]
+        return dict_character
+
+    def get_ignored_tokens(self):
+        end_idx = self.get_beg_end_flag_idx("eos")
+        return [end_idx]
+
+    def get_beg_end_flag_idx(self, beg_or_end):
+        if beg_or_end == "sos":
+            idx = np.array(self.dict[self.beg_str])
+        elif beg_or_end == "eos":
+            idx = np.array(self.dict[self.end_str])
+        else:
+            assert False, "unsupported type %s in get_beg_end_flag_idx" % beg_or_end
+        return idx
+
+    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
+        """convert text-index into text-label."""
+        result_list = []
+        [end_idx] = self.get_ignored_tokens()
+        batch_size = len(text_index)
+        for batch_idx in range(batch_size):
+            char_list = []
+            conf_list = []
+            for idx in range(len(text_index[batch_idx])):
+                if int(text_index[batch_idx][idx]) == int(end_idx):
+                    break
+                if is_remove_duplicate:
+                    # only for predict
+                    if (
+                        idx > 0
+                        and text_index[batch_idx][idx - 1] == text_index[batch_idx][idx]
+                    ):
+                        continue
+                char_list.append(self.character[int(text_index[batch_idx][idx])])
+                if text_prob is not None:
+                    conf_list.append(text_prob[batch_idx][idx])
+                else:
+                    conf_list.append(1)
+            text = "".join(char_list)
+            result_list.append((text, np.mean(conf_list).tolist()))
+        return result_list
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        """
+        text = self.decode(text)
+        if label is None:
+            return text
+        else:
+            label = self.decode(label, is_remove_duplicate=False)
+            return text, label
+        """
+        preds_idx = preds["rec_pred"]
+        if isinstance(preds_idx, paddle.Tensor):
+            preds_idx = preds_idx.numpy()
+        if "rec_pred_scores" in preds:
+            preds_idx = preds["rec_pred"]
+            preds_prob = preds["rec_pred_scores"]
+        else:
+            preds_idx = preds["rec_pred"].argmax(axis=2)
+            preds_prob = preds["rec_pred"].max(axis=2)
+        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+        if label is None:
+            return text
+        label = self.decode(label, is_remove_duplicate=False)
+        return text, label
+
+
+class SRNLabelDecode(BaseRecLabelDecode):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, character_dict_path=None, use_space_char=False, **kwargs):
+        super(SRNLabelDecode, self).__init__(character_dict_path, use_space_char)
+        self.max_text_length = kwargs.get("max_text_length", 25)
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        pred = preds["predict"]
+        char_num = len(self.character_str) + 2
+        if isinstance(pred, paddle.Tensor):
+            pred = pred.numpy()
+        pred = np.reshape(pred, [-1, char_num])
+
+        preds_idx = np.argmax(pred, axis=1)
+        preds_prob = np.max(pred, axis=1)
+
+        preds_idx = np.reshape(preds_idx, [-1, self.max_text_length])
+
+        preds_prob = np.reshape(preds_prob, [-1, self.max_text_length])
+
+        text = self.decode(preds_idx, preds_prob)
+
+        if label is None:
+            text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+            return text
+        label = self.decode(label)
+        return text, label
+
+    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
+        """convert text-index into text-label."""
+        result_list = []
+        ignored_tokens = self.get_ignored_tokens()
+        batch_size = len(text_index)
+
+        for batch_idx in range(batch_size):
+            char_list = []
+            conf_list = []
+            for idx in range(len(text_index[batch_idx])):
+                if text_index[batch_idx][idx] in ignored_tokens:
+                    continue
+                if is_remove_duplicate:
+                    # only for predict
+                    if (
+                        idx > 0
+                        and text_index[batch_idx][idx - 1] == text_index[batch_idx][idx]
+                    ):
+                        continue
+                char_list.append(self.character[int(text_index[batch_idx][idx])])
+                if text_prob is not None:
+                    conf_list.append(text_prob[batch_idx][idx])
+                else:
+                    conf_list.append(1)
+
+            text = "".join(char_list)
+            result_list.append((text, np.mean(conf_list).tolist()))
+        return result_list
+
+    def add_special_char(self, dict_character):
+        dict_character = dict_character + [self.beg_str, self.end_str]
+        return dict_character
+
+    def get_ignored_tokens(self):
+        beg_idx = self.get_beg_end_flag_idx("beg")
+        end_idx = self.get_beg_end_flag_idx("end")
+        return [beg_idx, end_idx]
+
+    def get_beg_end_flag_idx(self, beg_or_end):
+        if beg_or_end == "beg":
+            idx = np.array(self.dict[self.beg_str])
+        elif beg_or_end == "end":
+            idx = np.array(self.dict[self.end_str])
+        else:
+            assert False, "unsupported type %s in get_beg_end_flag_idx" % beg_or_end
+        return idx
+
+
+class ParseQLabelDecode(BaseRecLabelDecode):
+    """Convert between text-label and text-index"""
+
+    BOS = "[B]"
+    EOS = "[E]"
+    PAD = "[P]"
+
+    def __init__(self, character_dict_path=None, use_space_char=False, **kwargs):
+        super(ParseQLabelDecode, self).__init__(character_dict_path, use_space_char)
+        self.max_text_length = kwargs.get("max_text_length", 25)
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if isinstance(preds, dict):
+            pred = preds["predict"]
+        else:
+            pred = preds
+
+        char_num = (
+            len(self.character_str) + 1
+        )  # We don't predict <bos> nor <pad>, with only addition <eos>
+        if isinstance(pred, paddle.Tensor):
+            pred = pred.numpy()
+        B, L = pred.shape[:2]
+        pred = np.reshape(pred, [-1, char_num])
+
+        preds_idx = np.argmax(pred, axis=1)
+        preds_prob = np.max(pred, axis=1)
+
+        preds_idx = np.reshape(preds_idx, [B, L])
+        preds_prob = np.reshape(preds_prob, [B, L])
+
+        if label is None:
+            text = self.decode(preds_idx, preds_prob, raw=False)
+            return text
+
+        text = self.decode(preds_idx, preds_prob, raw=False)
+        label = self.decode(label, None, False)
+
+        return text, label
+
+    def decode(self, text_index, text_prob=None, raw=False):
+        """convert text-index into text-label."""
+        result_list = []
+        ignored_tokens = self.get_ignored_tokens()
+        batch_size = len(text_index)
+
+        for batch_idx in range(batch_size):
+            char_list = []
+            conf_list = []
+
+            index = text_index[batch_idx, :]
+            prob = None
+            if text_prob is not None:
+                prob = text_prob[batch_idx, :]
+
+            if not raw:
+                index, prob = self._filter(index, prob)
+
+            for idx in range(len(index)):
+                if index[idx] in ignored_tokens:
+                    continue
+                char_list.append(self.character[int(index[idx])])
+                if text_prob is not None:
+                    conf_list.append(prob[idx])
+                else:
+                    conf_list.append(1)
+
+            text = "".join(char_list)
+            result_list.append((text, np.mean(conf_list).tolist()))
+
+        return result_list
+
+    def add_special_char(self, dict_character):
+        dict_character = [self.EOS] + dict_character + [self.BOS, self.PAD]
+        return dict_character
+
+    def _filter(self, ids, probs=None):
+        ids = ids.tolist()
+        try:
+            eos_idx = ids.index(self.dict[self.EOS])
+        except ValueError:
+            eos_idx = len(ids)  # Nothing to truncate.
+        # Truncate after EOS
+        ids = ids[:eos_idx]
+        if probs is not None:
+            probs = probs[: eos_idx + 1]  # but include prob. for EOS (if it exists)
+        return ids, probs
+
+    def get_ignored_tokens(self):
+        return [self.dict[self.BOS], self.dict[self.EOS], self.dict[self.PAD]]
+
+
+class SARLabelDecode(BaseRecLabelDecode):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, character_dict_path=None, use_space_char=False, **kwargs):
+        super(SARLabelDecode, self).__init__(character_dict_path, use_space_char)
+
+        self.rm_symbol = kwargs.get("rm_symbol", False)
+
+    def add_special_char(self, dict_character):
+        beg_end_str = "<BOS/EOS>"
+        unknown_str = "<UKN>"
+        padding_str = "<PAD>"
+        dict_character = dict_character + [unknown_str]
+        self.unknown_idx = len(dict_character) - 1
+        dict_character = dict_character + [beg_end_str]
+        self.start_idx = len(dict_character) - 1
+        self.end_idx = len(dict_character) - 1
+        dict_character = dict_character + [padding_str]
+        self.padding_idx = len(dict_character) - 1
+        return dict_character
+
+    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
+        """convert text-index into text-label."""
+        result_list = []
+        ignored_tokens = self.get_ignored_tokens()
+
+        batch_size = len(text_index)
+        for batch_idx in range(batch_size):
+            char_list = []
+            conf_list = []
+            for idx in range(len(text_index[batch_idx])):
+                if text_index[batch_idx][idx] in ignored_tokens:
+                    continue
+                if int(text_index[batch_idx][idx]) == int(self.end_idx):
+                    if text_prob is None and idx == 0:
+                        continue
+                    else:
+                        break
+                if is_remove_duplicate:
+                    # only for predict
+                    if (
+                        idx > 0
+                        and text_index[batch_idx][idx - 1] == text_index[batch_idx][idx]
+                    ):
+                        continue
+                char_list.append(self.character[int(text_index[batch_idx][idx])])
+                if text_prob is not None:
+                    conf_list.append(text_prob[batch_idx][idx])
+                else:
+                    conf_list.append(1)
+            text = "".join(char_list)
+            if self.rm_symbol:
+                comp = re.compile("[^A-Z^a-z^0-9^\u4e00-\u9fa5]")
+                text = text.lower()
+                text = comp.sub("", text)
+            result_list.append((text, np.mean(conf_list).tolist()))
+        return result_list
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if isinstance(preds, paddle.Tensor):
+            preds = preds.numpy()
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+
+        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+
+        if label is None:
+            return text
+        label = self.decode(label, is_remove_duplicate=False)
+        return text, label
+
+    def get_ignored_tokens(self):
+        return [self.padding_idx]
+
+
+class SATRNLabelDecode(BaseRecLabelDecode):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, character_dict_path=None, use_space_char=False, **kwargs):
+        super(SATRNLabelDecode, self).__init__(character_dict_path, use_space_char)
+
+        self.rm_symbol = kwargs.get("rm_symbol", False)
+
+    def add_special_char(self, dict_character):
+        beg_end_str = "<BOS/EOS>"
+        unknown_str = "<UKN>"
+        padding_str = "<PAD>"
+        dict_character = dict_character + [unknown_str]
+        self.unknown_idx = len(dict_character) - 1
+        dict_character = dict_character + [beg_end_str]
+        self.start_idx = len(dict_character) - 1
+        self.end_idx = len(dict_character) - 1
+        dict_character = dict_character + [padding_str]
+        self.padding_idx = len(dict_character) - 1
+        return dict_character
+
+    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
+        """convert text-index into text-label."""
+        result_list = []
+        ignored_tokens = self.get_ignored_tokens()
+
+        batch_size = len(text_index)
+        for batch_idx in range(batch_size):
+            char_list = []
+            conf_list = []
+            for idx in range(len(text_index[batch_idx])):
+                if text_index[batch_idx][idx] in ignored_tokens:
+                    continue
+                if int(text_index[batch_idx][idx]) == int(self.end_idx):
+                    if text_prob is None and idx == 0:
+                        continue
+                    else:
+                        break
+                if is_remove_duplicate:
+                    # only for predict
+                    if (
+                        idx > 0
+                        and text_index[batch_idx][idx - 1] == text_index[batch_idx][idx]
+                    ):
+                        continue
+                char_list.append(self.character[int(text_index[batch_idx][idx])])
+                if text_prob is not None:
+                    conf_list.append(text_prob[batch_idx][idx])
+                else:
+                    conf_list.append(1)
+            text = "".join(char_list)
+            if self.rm_symbol:
+                comp = re.compile("[^A-Z^a-z^0-9^\u4e00-\u9fa5]")
+                text = text.lower()
+                text = comp.sub("", text)
+            result_list.append((text, np.mean(conf_list).tolist()))
+        return result_list
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if isinstance(preds, paddle.Tensor):
+            preds = preds.numpy()
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+
+        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+
+        if label is None:
+            return text
+        label = self.decode(label, is_remove_duplicate=False)
+        return text, label
+
+    def get_ignored_tokens(self):
+        return [self.padding_idx]
+
+
+class DistillationSARLabelDecode(SARLabelDecode):
+    """
+    Convert
+    Convert between text-label and text-index
+    """
+
+    def __init__(
+        self,
+        character_dict_path=None,
+        use_space_char=False,
+        model_name=["student"],
+        key=None,
+        multi_head=False,
+        **kwargs,
+    ):
+        super(DistillationSARLabelDecode, self).__init__(
+            character_dict_path, use_space_char
+        )
+        if not isinstance(model_name, list):
+            model_name = [model_name]
+        self.model_name = model_name
+
+        self.key = key
+        self.multi_head = multi_head
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        output = dict()
+        for name in self.model_name:
+            pred = preds[name]
+            if self.key is not None:
+                pred = pred[self.key]
+            if self.multi_head and isinstance(pred, dict):
+                pred = pred["sar"]
+            output[name] = super().__call__(pred, label=label, *args, **kwargs)
+        return output
+
+
+class PRENLabelDecode(BaseRecLabelDecode):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, character_dict_path=None, use_space_char=False, **kwargs):
+        super(PRENLabelDecode, self).__init__(character_dict_path, use_space_char)
+
+    def add_special_char(self, dict_character):
+        padding_str = "<PAD>"  # 0
+        end_str = "<EOS>"  # 1
+        unknown_str = "<UNK>"  # 2
+
+        dict_character = [padding_str, end_str, unknown_str] + dict_character
+        self.padding_idx = 0
+        self.end_idx = 1
+        self.unknown_idx = 2
+
+        return dict_character
+
+    def decode(self, text_index, text_prob=None):
+        """convert text-index into text-label."""
+        result_list = []
+        batch_size = len(text_index)
+
+        for batch_idx in range(batch_size):
+            char_list = []
+            conf_list = []
+            for idx in range(len(text_index[batch_idx])):
+                if text_index[batch_idx][idx] == self.end_idx:
+                    break
+                if text_index[batch_idx][idx] in [self.padding_idx, self.unknown_idx]:
+                    continue
+                char_list.append(self.character[int(text_index[batch_idx][idx])])
+                if text_prob is not None:
+                    conf_list.append(text_prob[batch_idx][idx])
+                else:
+                    conf_list.append(1)
+
+            text = "".join(char_list)
+            if len(text) > 0:
+                result_list.append((text, np.mean(conf_list).tolist()))
+            else:
+                # here confidence of empty recog result is 1
+                result_list.append(("", 1))
+        return result_list
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if isinstance(preds, paddle.Tensor):
+            preds = preds.numpy()
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+        text = self.decode(preds_idx, preds_prob)
+        if label is None:
+            return text
+        label = self.decode(label)
+        return text, label
+
+
+class NRTRLabelDecode(BaseRecLabelDecode):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, character_dict_path=None, use_space_char=True, **kwargs):
+        super(NRTRLabelDecode, self).__init__(character_dict_path, use_space_char)
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if len(preds) == 2:
+            preds_id = preds[0]
+            preds_prob = preds[1]
+            if isinstance(preds_id, paddle.Tensor):
+                preds_id = preds_id.numpy()
+            if isinstance(preds_prob, paddle.Tensor):
+                preds_prob = preds_prob.numpy()
+            if preds_id[0][0] == 2:
+                preds_idx = preds_id[:, 1:]
+                preds_prob = preds_prob[:, 1:]
+            else:
+                preds_idx = preds_id
+            text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+            if label is None:
+                return text
+            label = self.decode(label[:, 1:])
+        else:
+            if isinstance(preds, paddle.Tensor):
+                preds = preds.numpy()
+            preds_idx = preds.argmax(axis=2)
+            preds_prob = preds.max(axis=2)
+            text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+            if label is None:
+                return text
+            label = self.decode(label[:, 1:])
+        return text, label
+
+    def add_special_char(self, dict_character):
+        dict_character = ["blank", "<unk>", "<s>", "</s>"] + dict_character
+        return dict_character
+
+    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
+        """convert text-index into text-label."""
+        result_list = []
+        batch_size = len(text_index)
+        for batch_idx in range(batch_size):
+            char_list = []
+            conf_list = []
+            for idx in range(len(text_index[batch_idx])):
+                try:
+                    char_idx = self.character[int(text_index[batch_idx][idx])]
+                except:
+                    continue
+                if char_idx == "</s>":  # end
+                    break
+                char_list.append(char_idx)
+                if text_prob is not None:
+                    conf_list.append(text_prob[batch_idx][idx])
+                else:
+                    conf_list.append(1)
+            text = "".join(char_list)
+            result_list.append((text, np.mean(conf_list).tolist()))
+        return result_list
+
+
+class ViTSTRLabelDecode(NRTRLabelDecode):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, character_dict_path=None, use_space_char=False, **kwargs):
+        super(ViTSTRLabelDecode, self).__init__(character_dict_path, use_space_char)
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if isinstance(preds, paddle.Tensor):
+            preds = preds[:, 1:].numpy()
+        else:
+            preds = preds[:, 1:]
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+        if label is None:
+            return text
+        label = self.decode(label[:, 1:])
+        return text, label
+
+    def add_special_char(self, dict_character):
+        dict_character = ["<s>", "</s>"] + dict_character
+        return dict_character
+
+
+class ABINetLabelDecode(NRTRLabelDecode):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, character_dict_path=None, use_space_char=False, **kwargs):
+        super(ABINetLabelDecode, self).__init__(character_dict_path, use_space_char)
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if isinstance(preds, dict):
+            preds = preds["align"][-1].numpy()
+        elif isinstance(preds, paddle.Tensor):
+            preds = preds.numpy()
+        else:
+            preds = preds
+
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+        if label is None:
+            return text
+        label = self.decode(label)
+        return text, label
+
+    def add_special_char(self, dict_character):
+        dict_character = ["</s>"] + dict_character
+        return dict_character
+
+
+class SPINLabelDecode(AttnLabelDecode):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, character_dict_path=None, use_space_char=False, **kwargs):
+        super(SPINLabelDecode, self).__init__(character_dict_path, use_space_char)
+
+    def add_special_char(self, dict_character):
+        self.beg_str = "sos"
+        self.end_str = "eos"
+        dict_character = dict_character
+        dict_character = [self.beg_str] + [self.end_str] + dict_character
+        return dict_character
+
+
+class VLLabelDecode(BaseRecLabelDecode):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, character_dict_path=None, use_space_char=False, **kwargs):
+        super(VLLabelDecode, self).__init__(character_dict_path, use_space_char)
+        self.max_text_length = kwargs.get("max_text_length", 25)
+        self.nclass = len(self.character) + 1
+
+    def decode(self, text_index, text_prob=None, is_remove_duplicate=False):
+        """convert text-index into text-label."""
+        result_list = []
+        ignored_tokens = self.get_ignored_tokens()
+        batch_size = len(text_index)
+        for batch_idx in range(batch_size):
+            selection = np.ones(len(text_index[batch_idx]), dtype=bool)
+            if is_remove_duplicate:
+                selection[1:] = text_index[batch_idx][1:] != text_index[batch_idx][:-1]
+            for ignored_token in ignored_tokens:
+                selection &= text_index[batch_idx] != ignored_token
+
+            char_list = [
+                self.character[text_id - 1]
+                for text_id in text_index[batch_idx][selection]
+            ]
+            if text_prob is not None:
+                conf_list = text_prob[batch_idx][selection]
+            else:
+                conf_list = [1] * len(selection)
+            if len(conf_list) == 0:
+                conf_list = [0]
+
+            text = "".join(char_list)
+            result_list.append((text, np.mean(conf_list).tolist()))
+        return result_list
+
+    def __call__(self, preds, label=None, length=None, *args, **kwargs):
+        if len(preds) == 2:  # eval mode
+            text_pre, x = preds
+            b = text_pre.shape[1]
+            lenText = self.max_text_length
+            nsteps = self.max_text_length
+
+            if not isinstance(text_pre, paddle.Tensor):
+                text_pre = paddle.to_tensor(text_pre, dtype="float32")
+
+            out_res = paddle.zeros(shape=[lenText, b, self.nclass], dtype=x.dtype)
+            out_length = paddle.zeros(shape=[b], dtype=x.dtype)
+            now_step = 0
+            for _ in range(nsteps):
+                if 0 in out_length and now_step < nsteps:
+                    tmp_result = text_pre[now_step, :, :]
+                    out_res[now_step] = tmp_result
+                    tmp_result = tmp_result.topk(1)[1].squeeze(axis=1)
+                    for j in range(b):
+                        if out_length[j] == 0 and tmp_result[j] == 0:
+                            out_length[j] = now_step + 1
+                    now_step += 1
+            for j in range(0, b):
+                if int(out_length[j]) == 0:
+                    out_length[j] = nsteps
+            start = 0
+            output = paddle.zeros(
+                shape=[int(out_length.sum()), self.nclass], dtype=x.dtype
+            )
+            for i in range(0, b):
+                cur_length = int(out_length[i])
+                output[start : start + cur_length] = out_res[0:cur_length, i, :]
+                start += cur_length
+            net_out = output
+            length = out_length
+
+        else:  # train mode
+            net_out = preds[0]
+            length = length
+            net_out = paddle.concat([t[:l] for t, l in zip(net_out, length)])
+        text = []
+        if not isinstance(net_out, paddle.Tensor):
+            net_out = paddle.to_tensor(net_out, dtype="float32")
+        net_out = F.softmax(net_out, axis=1)
+        for i in range(0, length.shape[0]):
+            if i == 0:
+                start_idx = 0
+                end_idx = int(length[i])
+            else:
+                start_idx = int(length[:i].sum())
+                end_idx = int(length[:i].sum() + length[i])
+            preds_idx = net_out[start_idx:end_idx].topk(1)[1][:, 0].tolist()
+            preds_text = "".join(
+                [
+                    (
+                        self.character[idx - 1]
+                        if idx > 0 and idx <= len(self.character)
+                        else ""
+                    )
+                    for idx in preds_idx
+                ]
+            )
+            preds_prob = net_out[start_idx:end_idx].topk(1)[0][:, 0]
+            preds_prob = paddle.exp(
+                paddle.log(preds_prob).sum() / (preds_prob.shape[0] + 1e-6)
+            )
+            text.append((preds_text, float(preds_prob)))
+        if label is None:
+            return text
+        label = self.decode(label)
+        return text, label
+
+
+class CANLabelDecode(BaseRecLabelDecode):
+    """Convert between latex-symbol and symbol-index"""
+
+    def __init__(self, character_dict_path=None, use_space_char=False, **kwargs):
+        super(CANLabelDecode, self).__init__(character_dict_path, use_space_char)
+
+    def decode(self, text_index, preds_prob=None):
+        result_list = []
+        batch_size = len(text_index)
+        for batch_idx in range(batch_size):
+            seq_end = text_index[batch_idx].argmin(0)
+            idx_list = text_index[batch_idx][:seq_end].tolist()
+            symbol_list = [self.character[idx] for idx in idx_list]
+            probs = []
+            if preds_prob is not None:
+                probs = preds_prob[batch_idx][: len(symbol_list)].tolist()
+
+            result_list.append([" ".join(symbol_list), probs])
+        return result_list
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        pred_prob, _, _, _ = preds
+        preds_idx = pred_prob.argmax(axis=2)
+
+        text = self.decode(preds_idx)
+        if label is None:
+            return text
+        label = self.decode(label)
+        return text, label
+
+
+class CPPDLabelDecode(NRTRLabelDecode):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, character_dict_path=None, use_space_char=False, **kwargs):
+        super(CPPDLabelDecode, self).__init__(character_dict_path, use_space_char)
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        if isinstance(preds, tuple):
+            if isinstance(preds[-1], dict):
+                preds = preds[-1]["align"][-1].numpy()
+            else:
+                preds = preds[-1].numpy()
+        if isinstance(preds, paddle.Tensor):
+            preds = preds.numpy()
+        else:
+            preds = preds
+        preds_idx = preds.argmax(axis=2)
+        preds_prob = preds.max(axis=2)
+        text = self.decode(preds_idx, preds_prob, is_remove_duplicate=False)
+        if label is None:
+            return text
+        label = self.decode(label)
+        return text, label
+
+    def add_special_char(self, dict_character):
+        dict_character = ["</s>"] + dict_character
+        return dict_character
+
+
+class LaTeXOCRDecode(object):
+    """Convert between latex-symbol and symbol-index"""
+
+    def __init__(self, rec_char_dict_path, **kwargs):
+        # Set the TOKENIZERS_PARALLELISM environment variable to 'false' to suppress
+        # the warning: "The current process just got forked, Disabling parallelism to avoid deadlocks..
+        #  To disable this warning, please explicitly set TOKENIZERS_PARALLELISM=(true | false)" from tokenizers
+        os.environ["TOKENIZERS_PARALLELISM"] = "false"
+        from tokenizers import Tokenizer as TokenizerFast
+
+        super(LaTeXOCRDecode, self).__init__()
+        self.tokenizer = TokenizerFast.from_file(rec_char_dict_path)
+
+    def post_process(self, s):
+        text_reg = r"(\\(operatorname|mathrm|text|mathbf)\s?\*? {.*?})"
+        letter = "[a-zA-Z]"
+        noletter = "[\W_^\d]"
+        names = [x[0].replace(" ", "") for x in re.findall(text_reg, s)]
+        s = re.sub(text_reg, lambda match: str(names.pop(0)), s)
+        news = s
+        while True:
+            s = news
+            news = re.sub(r"(?!\\ )(%s)\s+?(%s)" % (noletter, noletter), r"\1\2", s)
+            news = re.sub(r"(?!\\ )(%s)\s+?(%s)" % (noletter, letter), r"\1\2", news)
+            news = re.sub(r"(%s)\s+?(%s)" % (letter, noletter), r"\1\2", news)
+            if news == s:
+                break
+        return s
+
+    def decode(self, tokens):
+        if len(tokens.shape) == 1:
+            tokens = tokens[None, :]
+        dec = [self.tokenizer.decode(tok) for tok in tokens]
+        dec_str_list = [
+            "".join(detok.split(" "))
+            .replace("Ġ", " ")
+            .replace("[EOS]", "")
+            .replace("[BOS]", "")
+            .replace("[PAD]", "")
+            .strip()
+            for detok in dec
+        ]
+        return [self.post_process(dec_str) for dec_str in dec_str_list]
+
+    def __call__(self, preds, label=None, mode="eval", *args, **kwargs):
+        if mode == "train":
+            preds_idx = np.array(preds.argmax(axis=2))
+            text = self.decode(preds_idx)
+        else:
+            text = self.decode(np.array(preds))
+        if label is None:
+            return text
+        label = self.decode(np.array(label))
+        return text, label
+
+
+class UniMERNetDecode(object):
+
+    SPECIAL_TOKENS_ATTRIBUTES = [
+        "bos_token",
+        "eos_token",
+        "unk_token",
+        "sep_token",
+        "pad_token",
+        "cls_token",
+        "mask_token",
+        "additional_special_tokens",
+    ]
+
+    def __init__(
+        self,
+        rec_char_dict_path,
+        is_infer=False,
+        **kwargs,
+    ):
+        # Set the TOKENIZERS_PARALLELISM environment variable to 'false' to suppress
+        # the warning: "The current process just got forked, Disabling parallelism to avoid deadlocks..
+        #  To disable this warning, please explicitly set TOKENIZERS_PARALLELISM=(true | false)" from tokenizers
+        os.environ["TOKENIZERS_PARALLELISM"] = "false"
+        from tokenizers import Tokenizer as TokenizerFast
+        from tokenizers import AddedToken
+
+        self.is_infer = is_infer
+        self._unk_token = "<unk>"
+        self._bos_token = "<s>"
+        self._eos_token = "</s>"
+        self._pad_token = "<pad>"
+        self._sep_token = None
+        self._cls_token = None
+        self._mask_token = None
+        self._additional_special_tokens = []
+        self.model_input_names = ["input_ids", "token_type_ids", "attention_mask"]
+        self.max_seq_len = 2048
+        self.pad_token_id = 1
+        self.bos_token_id = 0
+        self.eos_token_id = 2
+        self.padding_side = "right"
+        self.pad_token_id = 1
+        self.pad_token = "<pad>"
+        self.pad_token_type_id = 0
+        self.pad_to_multiple_of = None
+        fast_tokenizer_file = os.path.join(rec_char_dict_path, "tokenizer.json")
+        tokenizer_config_file = os.path.join(
+            rec_char_dict_path, "tokenizer_config.json"
+        )
+        self.tokenizer = TokenizerFast.from_file(fast_tokenizer_file)
+        added_tokens_decoder = {}
+        added_tokens_map = {}
+        if tokenizer_config_file is not None:
+            with open(
+                tokenizer_config_file, encoding="utf-8"
+            ) as tokenizer_config_handle:
+                init_kwargs = json.load(tokenizer_config_handle)
+                if "added_tokens_decoder" in init_kwargs:
+                    for idx, token in init_kwargs["added_tokens_decoder"].items():
+                        if isinstance(token, dict):
+                            token = AddedToken(**token)
+                        if isinstance(token, AddedToken):
+                            added_tokens_decoder[int(idx)] = token
+                            added_tokens_map[str(token)] = token
+                        else:
+                            raise ValueError(
+                                f"Found a {token.__class__} in the saved `added_tokens_decoder`, should be a dictionary or an AddedToken instance"
+                            )
+                init_kwargs["added_tokens_decoder"] = added_tokens_decoder
+                added_tokens_decoder = init_kwargs.pop("added_tokens_decoder", {})
+                tokens_to_add = [
+                    token
+                    for index, token in sorted(
+                        added_tokens_decoder.items(), key=lambda x: x[0]
+                    )
+                    if token not in added_tokens_decoder
+                ]
+                added_tokens_encoder = self.added_tokens_encoder(added_tokens_decoder)
+                encoder = list(added_tokens_encoder.keys()) + [
+                    str(token) for token in tokens_to_add
+                ]
+                tokens_to_add += [
+                    token
+                    for token in self.all_special_tokens_extended
+                    if token not in encoder and token not in tokens_to_add
+                ]
+                if len(tokens_to_add) > 0:
+                    is_last_special = None
+                    tokens = []
+                    special_tokens = self.all_special_tokens
+                    for token in tokens_to_add:
+                        is_special = (
+                            (token.special or str(token) in special_tokens)
+                            if isinstance(token, AddedToken)
+                            else str(token) in special_tokens
+                        )
+                        if is_last_special is None or is_last_special == is_special:
+                            tokens.append(token)
+                        else:
+                            self._add_tokens(tokens, special_tokens=is_last_special)
+                            tokens = [token]
+                        is_last_special = is_special
+                    if tokens:
+                        self._add_tokens(tokens, special_tokens=is_last_special)
+
+    def _add_tokens(self, new_tokens, special_tokens=False) -> int:
+        if special_tokens:
+            return self.tokenizer.add_special_tokens(new_tokens)
+
+        return self.tokenizer.add_tokens(new_tokens)
+
+    def added_tokens_encoder(self, added_tokens_decoder):
+        return {
+            k.content: v
+            for v, k in sorted(added_tokens_decoder.items(), key=lambda item: item[0])
+        }
+
+    @property
+    def all_special_tokens(self):
+
+        all_toks = [str(s) for s in self.all_special_tokens_extended]
+        return all_toks
+
+    @property
+    def all_special_tokens_extended(self):
+        all_tokens = []
+        seen = set()
+        for value in self.special_tokens_map_extended.values():
+            if isinstance(value, (list, tuple)):
+                tokens_to_add = [token for token in value if str(token) not in seen]
+            else:
+                tokens_to_add = [value] if str(value) not in seen else []
+            seen.update(map(str, tokens_to_add))
+            all_tokens.extend(tokens_to_add)
+        return all_tokens
+
+    @property
+    def special_tokens_map_extended(self):
+        set_attr = {}
+        for attr in self.SPECIAL_TOKENS_ATTRIBUTES:
+            attr_value = getattr(self, "_" + attr)
+            if attr_value:
+                set_attr[attr] = attr_value
+        return set_attr
+
+    def convert_ids_to_tokens(self, ids, skip_special_tokens: bool = False):
+        if isinstance(ids, int):
+            return self.tokenizer.id_to_token(ids)
+        tokens = []
+        for index in ids:
+            index = int(index)
+            if skip_special_tokens and index in self.all_special_ids:
+                continue
+            tokens.append(self.tokenizer.id_to_token(index))
+        return tokens
+
+    def detokenize(self, tokens):
+        self.tokenizer.bos_token = "<s>"
+        self.tokenizer.eos_token = "</s>"
+        self.tokenizer.pad_token = "<pad>"
+        toks = [self.convert_ids_to_tokens(tok) for tok in tokens]
+        for b in range(len(toks)):
+            for i in reversed(range(len(toks[b]))):
+                if toks[b][i] is None:
+                    toks[b][i] = ""
+                toks[b][i] = toks[b][i].replace("Ġ", " ").strip()
+                if toks[b][i] in (
+                    [
+                        self.tokenizer.bos_token,
+                        self.tokenizer.eos_token,
+                        self.tokenizer.pad_token,
+                    ]
+                ):
+                    del toks[b][i]
+        return toks
+
+    def token2str(self, token_ids) -> list:
+        generated_text = []
+        for tok_id in token_ids:
+            end_idx = np.argwhere(tok_id == 2)
+            if len(end_idx) > 0:
+                end_idx = int(end_idx[0][0])
+                tok_id = tok_id[: end_idx + 1]
+            generated_text.append(
+                self.tokenizer.decode(tok_id, skip_special_tokens=True)
+            )
+        generated_text = [self.post_process(text) for text in generated_text]
+        return generated_text
+
+    def normalize_infer(self, s: str) -> str:
+        """Normalizes a string by removing unnecessary spaces.
+
+        Args:
+            s (str): String to normalize.
+
+        Returns:
+            str: Normalized string.
+        """
+        text_reg = r"(\\(operatorname|mathrm|text|mathbf)\s?\*? {.*?})"
+        letter = "[a-zA-Z]"
+        noletter = "[\W_^\d]"
+        names = []
+        for x in re.findall(text_reg, s):
+            pattern = r"\\[a-zA-Z]+"
+            pattern = r"(\\[a-zA-Z]+)\s(?=\w)|\\[a-zA-Z]+\s(?=})"
+            matches = re.findall(pattern, x[0])
+            for m in matches:
+                if (
+                    m
+                    not in [
+                        "\\operatorname",
+                        "\\mathrm",
+                        "\\text",
+                        "\\mathbf",
+                    ]
+                    and m.strip() != ""
+                ):
+                    s = s.replace(m, m + "XXXXXXX")
+                    s = s.replace(" ", "")
+                    names.append(s)
+        if len(names) > 0:
+            s = re.sub(text_reg, lambda match: str(names.pop(0)), s)
+        news = s
+        while True:
+            s = news
+            news = re.sub(r"(?!\\ )(%s)\s+?(%s)" % (noletter, noletter), r"\1\2", s)
+            news = re.sub(r"(?!\\ )(%s)\s+?(%s)" % (noletter, letter), r"\1\2", news)
+            news = re.sub(r"(%s)\s+?(%s)" % (letter, noletter), r"\1\2", news)
+            if news == s:
+                break
+        return s.replace("XXXXXXX", " ")
+
+    def remove_chinese_text_wrapping(self, formula):
+        pattern = re.compile(r"\\text\s*{\s*([^}]*?[\u4e00-\u9fff]+[^}]*?)\s*}")
+
+        def replacer(match):
+            return match.group(1)
+
+        replaced_formula = pattern.sub(replacer, formula)
+        return replaced_formula.replace('"', "")
+
+    def normalize(self, s):
+        text_reg = r"(\\(operatorname|mathrm|text|mathbf)\s?\*? {.*?})"
+        letter = "[a-zA-Z]"
+        noletter = "[\W_^\d]"
+        names = [x[0].replace(" ", "") for x in re.findall(text_reg, s)]
+        s = re.sub(text_reg, lambda match: str(names.pop(0)), s)
+        news = s
+        while True:
+            s = news
+            news = re.sub(r"(?!\\ )(%s)\s+?(%s)" % (noletter, noletter), r"\1\2", s)
+            news = re.sub(r"(?!\\ )(%s)\s+?(%s)" % (noletter, letter), r"\1\2", news)
+            news = re.sub(r"(%s)\s+?(%s)" % (letter, noletter), r"\1\2", news)
+            if news == s:
+                break
+        return s
+
+    def post_process(self, text: str) -> str:
+        """Post-processes a string by fixing text and normalizing it.
+
+        Args:
+            text (str): String to post-process.
+
+        Returns:
+            str: Post-processed string.
+        """
+        from ftfy import fix_text
+
+        if self.is_infer:
+            text = self.remove_chinese_text_wrapping(text)
+            text = fix_text(text)
+            text = self.normalize_infer(text)
+        else:
+            text = fix_text(text)
+            text = self.normalize(text)
+        return text
+
+    def __call__(self, preds, label=None, mode="eval", *args, **kwargs):
+        if mode == "train":
+            preds_idx = np.array(preds.argmax(axis=2))
+            text = self.token2str(preds_idx)
+        else:
+            text = self.token2str(np.array(preds))
+        if label is None:
+            return text
+        label = self.token2str(np.array(label))
+        return text, label
diff --git a/ppocr/postprocess/sast_postprocess.py b/ppocr/postprocess/sast_postprocess.py
new file mode 100755
index 0000000..6c8b8e3
--- /dev/null
+++ b/ppocr/postprocess/sast_postprocess.py
@@ -0,0 +1,358 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+__dir__ = os.path.dirname(__file__)
+sys.path.append(__dir__)
+sys.path.append(os.path.join(__dir__, ".."))
+
+import numpy as np
+from .locality_aware_nms import nms_locality
+import paddle
+import cv2
+import time
+
+
+class SASTPostProcess(object):
+    """
+    The post process for SAST.
+    """
+
+    def __init__(
+        self,
+        score_thresh=0.5,
+        nms_thresh=0.2,
+        sample_pts_num=2,
+        shrink_ratio_of_width=0.3,
+        expand_scale=1.0,
+        tcl_map_thresh=0.5,
+        **kwargs,
+    ):
+        self.score_thresh = score_thresh
+        self.nms_thresh = nms_thresh
+        self.sample_pts_num = sample_pts_num
+        self.shrink_ratio_of_width = shrink_ratio_of_width
+        self.expand_scale = expand_scale
+        self.tcl_map_thresh = tcl_map_thresh
+
+    def point_pair2poly(self, point_pair_list):
+        """
+        Transfer vertical point_pairs into poly point in clockwise.
+        """
+        # construct poly
+        point_num = len(point_pair_list) * 2
+        point_list = [0] * point_num
+        for idx, point_pair in enumerate(point_pair_list):
+            point_list[idx] = point_pair[0]
+            point_list[point_num - 1 - idx] = point_pair[1]
+        return np.array(point_list).reshape(-1, 2)
+
+    def shrink_quad_along_width(self, quad, begin_width_ratio=0.0, end_width_ratio=1.0):
+        """
+        Generate shrink_quad_along_width.
+        """
+        ratio_pair = np.array(
+            [[begin_width_ratio], [end_width_ratio]], dtype=np.float32
+        )
+        p0_1 = quad[0] + (quad[1] - quad[0]) * ratio_pair
+        p3_2 = quad[3] + (quad[2] - quad[3]) * ratio_pair
+        return np.array([p0_1[0], p0_1[1], p3_2[1], p3_2[0]])
+
+    def expand_poly_along_width(self, poly, shrink_ratio_of_width=0.3):
+        """
+        expand poly along width.
+        """
+        point_num = poly.shape[0]
+        left_quad = np.array([poly[0], poly[1], poly[-2], poly[-1]], dtype=np.float32)
+        left_ratio = (
+            -shrink_ratio_of_width
+            * np.linalg.norm(left_quad[0] - left_quad[3])
+            / (np.linalg.norm(left_quad[0] - left_quad[1]) + 1e-6)
+        )
+        left_quad_expand = self.shrink_quad_along_width(left_quad, left_ratio, 1.0)
+        right_quad = np.array(
+            [
+                poly[point_num // 2 - 2],
+                poly[point_num // 2 - 1],
+                poly[point_num // 2],
+                poly[point_num // 2 + 1],
+            ],
+            dtype=np.float32,
+        )
+        right_ratio = 1.0 + shrink_ratio_of_width * np.linalg.norm(
+            right_quad[0] - right_quad[3]
+        ) / (np.linalg.norm(right_quad[0] - right_quad[1]) + 1e-6)
+        right_quad_expand = self.shrink_quad_along_width(right_quad, 0.0, right_ratio)
+        poly[0] = left_quad_expand[0]
+        poly[-1] = left_quad_expand[-1]
+        poly[point_num // 2 - 1] = right_quad_expand[1]
+        poly[point_num // 2] = right_quad_expand[2]
+        return poly
+
+    def restore_quad(self, tcl_map, tcl_map_thresh, tvo_map):
+        """Restore quad."""
+        xy_text = np.argwhere(tcl_map[:, :, 0] > tcl_map_thresh)
+        xy_text = xy_text[:, ::-1]  # (n, 2)
+
+        # Sort the text boxes via the y axis
+        xy_text = xy_text[np.argsort(xy_text[:, 1])]
+
+        scores = tcl_map[xy_text[:, 1], xy_text[:, 0], 0]
+        scores = scores[:, np.newaxis]
+
+        # Restore
+        point_num = int(tvo_map.shape[-1] / 2)
+        assert point_num == 4
+        tvo_map = tvo_map[xy_text[:, 1], xy_text[:, 0], :]
+        xy_text_tile = np.tile(xy_text, (1, point_num))  # (n, point_num * 2)
+        quads = xy_text_tile - tvo_map
+
+        return scores, quads, xy_text
+
+    def quad_area(self, quad):
+        """
+        compute area of a quad.
+        """
+        edge = [
+            (quad[1][0] - quad[0][0]) * (quad[1][1] + quad[0][1]),
+            (quad[2][0] - quad[1][0]) * (quad[2][1] + quad[1][1]),
+            (quad[3][0] - quad[2][0]) * (quad[3][1] + quad[2][1]),
+            (quad[0][0] - quad[3][0]) * (quad[0][1] + quad[3][1]),
+        ]
+        return np.sum(edge) / 2.0
+
+    def nms(self, dets):
+        dets = nms_locality(dets, self.nms_thresh)
+        return dets
+
+    def cluster_by_quads_tco(self, tcl_map, tcl_map_thresh, quads, tco_map):
+        """
+        Cluster pixels in tcl_map based on quads.
+        """
+        instance_count = quads.shape[0] + 1  # contain background
+        instance_label_map = np.zeros(tcl_map.shape[:2], dtype=np.int32)
+        if instance_count == 1:
+            return instance_count, instance_label_map
+
+        # predict text center
+        xy_text = np.argwhere(tcl_map[:, :, 0] > tcl_map_thresh)
+        n = xy_text.shape[0]
+        xy_text = xy_text[:, ::-1]  # (n, 2)
+        tco = tco_map[xy_text[:, 1], xy_text[:, 0], :]  # (n, 2)
+        pred_tc = xy_text - tco
+
+        # get gt text center
+        m = quads.shape[0]
+        gt_tc = np.mean(quads, axis=1)  # (m, 2)
+
+        pred_tc_tile = np.tile(pred_tc[:, np.newaxis, :], (1, m, 1))  # (n, m, 2)
+        gt_tc_tile = np.tile(gt_tc[np.newaxis, :, :], (n, 1, 1))  # (n, m, 2)
+        dist_mat = np.linalg.norm(pred_tc_tile - gt_tc_tile, axis=2)  # (n, m)
+        xy_text_assign = np.argmin(dist_mat, axis=1) + 1  # (n,)
+
+        instance_label_map[xy_text[:, 1], xy_text[:, 0]] = xy_text_assign
+        return instance_count, instance_label_map
+
+    def estimate_sample_pts_num(self, quad, xy_text):
+        """
+        Estimate sample points number.
+        """
+        eh = (
+            np.linalg.norm(quad[0] - quad[3]) + np.linalg.norm(quad[1] - quad[2])
+        ) / 2.0
+        ew = (
+            np.linalg.norm(quad[0] - quad[1]) + np.linalg.norm(quad[2] - quad[3])
+        ) / 2.0
+
+        dense_sample_pts_num = max(2, int(ew))
+        dense_xy_center_line = xy_text[
+            np.linspace(
+                0,
+                xy_text.shape[0] - 1,
+                dense_sample_pts_num,
+                endpoint=True,
+                dtype=np.float32,
+            ).astype(np.int32)
+        ]
+
+        dense_xy_center_line_diff = dense_xy_center_line[1:] - dense_xy_center_line[:-1]
+        estimate_arc_len = np.sum(np.linalg.norm(dense_xy_center_line_diff, axis=1))
+
+        sample_pts_num = max(2, int(estimate_arc_len / eh))
+        return sample_pts_num
+
+    def detect_sast(
+        self,
+        tcl_map,
+        tvo_map,
+        tbo_map,
+        tco_map,
+        ratio_w,
+        ratio_h,
+        src_w,
+        src_h,
+        shrink_ratio_of_width=0.3,
+        tcl_map_thresh=0.5,
+        offset_expand=1.0,
+        out_strid=4.0,
+    ):
+        """
+        first resize the tcl_map, tvo_map and tbo_map to the input_size, then restore the polys
+        """
+        # restore quad
+        scores, quads, xy_text = self.restore_quad(tcl_map, tcl_map_thresh, tvo_map)
+        dets = np.hstack((quads, scores)).astype(np.float32, copy=False)
+        dets = self.nms(dets)
+        if dets.shape[0] == 0:
+            return []
+        quads = dets[:, :-1].reshape(-1, 4, 2)
+
+        # Compute quad area
+        quad_areas = []
+        for quad in quads:
+            quad_areas.append(-self.quad_area(quad))
+
+        # instance segmentation
+        # instance_count, instance_label_map = cv2.connectedComponents(tcl_map.astype(np.uint8), connectivity=8)
+        instance_count, instance_label_map = self.cluster_by_quads_tco(
+            tcl_map, tcl_map_thresh, quads, tco_map
+        )
+
+        # restore single poly with tcl instance.
+        poly_list = []
+        for instance_idx in range(1, instance_count):
+            xy_text = np.argwhere(instance_label_map == instance_idx)[:, ::-1]
+            quad = quads[instance_idx - 1]
+            q_area = quad_areas[instance_idx - 1]
+            if q_area < 5:
+                continue
+
+            #
+            len1 = float(np.linalg.norm(quad[0] - quad[1]))
+            len2 = float(np.linalg.norm(quad[1] - quad[2]))
+            min_len = min(len1, len2)
+            if min_len < 3:
+                continue
+
+            # filter small CC
+            if xy_text.shape[0] <= 0:
+                continue
+
+            # filter low confidence instance
+            xy_text_scores = tcl_map[xy_text[:, 1], xy_text[:, 0], 0]
+            if np.sum(xy_text_scores) / quad_areas[instance_idx - 1] < 0.1:
+                # if np.sum(xy_text_scores) / quad_areas[instance_idx - 1] < 0.05:
+                continue
+
+            # sort xy_text
+            left_center_pt = np.array(
+                [[(quad[0, 0] + quad[-1, 0]) / 2.0, (quad[0, 1] + quad[-1, 1]) / 2.0]]
+            )  # (1, 2)
+            right_center_pt = np.array(
+                [[(quad[1, 0] + quad[2, 0]) / 2.0, (quad[1, 1] + quad[2, 1]) / 2.0]]
+            )  # (1, 2)
+            proj_unit_vec = (right_center_pt - left_center_pt) / (
+                np.linalg.norm(right_center_pt - left_center_pt) + 1e-6
+            )
+            proj_value = np.sum(xy_text * proj_unit_vec, axis=1)
+            xy_text = xy_text[np.argsort(proj_value)]
+
+            # Sample pts in tcl map
+            if self.sample_pts_num == 0:
+                sample_pts_num = self.estimate_sample_pts_num(quad, xy_text)
+            else:
+                sample_pts_num = self.sample_pts_num
+            xy_center_line = xy_text[
+                np.linspace(
+                    0,
+                    xy_text.shape[0] - 1,
+                    sample_pts_num,
+                    endpoint=True,
+                    dtype=np.float32,
+                ).astype(np.int32)
+            ]
+
+            point_pair_list = []
+            for x, y in xy_center_line:
+                # get corresponding offset
+                offset = tbo_map[y, x, :].reshape(2, 2)
+                if offset_expand != 1.0:
+                    offset_length = np.linalg.norm(offset, axis=1, keepdims=True)
+                    expand_length = np.clip(
+                        offset_length * (offset_expand - 1), a_min=0.5, a_max=3.0
+                    )
+                    offset_detal = offset / offset_length * expand_length
+                    offset = offset + offset_detal
+                    # original point
+                ori_yx = np.array([y, x], dtype=np.float32)
+                point_pair = (
+                    (ori_yx + offset)[:, ::-1]
+                    * out_strid
+                    / np.array([ratio_w, ratio_h]).reshape(-1, 2)
+                )
+                point_pair_list.append(point_pair)
+
+            # ndarry: (x, 2), expand poly along width
+            detected_poly = self.point_pair2poly(point_pair_list)
+            detected_poly = self.expand_poly_along_width(
+                detected_poly, shrink_ratio_of_width
+            )
+            detected_poly[:, 0] = np.clip(detected_poly[:, 0], a_min=0, a_max=src_w)
+            detected_poly[:, 1] = np.clip(detected_poly[:, 1], a_min=0, a_max=src_h)
+            poly_list.append(detected_poly)
+
+        return poly_list
+
+    def __call__(self, outs_dict, shape_list):
+        score_list = outs_dict["f_score"]
+        border_list = outs_dict["f_border"]
+        tvo_list = outs_dict["f_tvo"]
+        tco_list = outs_dict["f_tco"]
+        if isinstance(score_list, paddle.Tensor):
+            score_list = score_list.numpy()
+            border_list = border_list.numpy()
+            tvo_list = tvo_list.numpy()
+            tco_list = tco_list.numpy()
+
+        img_num = len(shape_list)
+        poly_lists = []
+        for ino in range(img_num):
+            p_score = score_list[ino].transpose((1, 2, 0))
+            p_border = border_list[ino].transpose((1, 2, 0))
+            p_tvo = tvo_list[ino].transpose((1, 2, 0))
+            p_tco = tco_list[ino].transpose((1, 2, 0))
+            src_h, src_w, ratio_h, ratio_w = shape_list[ino]
+
+            poly_list = self.detect_sast(
+                p_score,
+                p_tvo,
+                p_border,
+                p_tco,
+                ratio_w,
+                ratio_h,
+                src_w,
+                src_h,
+                shrink_ratio_of_width=self.shrink_ratio_of_width,
+                tcl_map_thresh=self.tcl_map_thresh,
+                offset_expand=self.expand_scale,
+            )
+            poly_lists.append({"points": np.array(poly_list)})
+
+        return poly_lists
diff --git a/ppocr/postprocess/table_postprocess.py b/ppocr/postprocess/table_postprocess.py
new file mode 100644
index 0000000..11a877a
--- /dev/null
+++ b/ppocr/postprocess/table_postprocess.py
@@ -0,0 +1,191 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import numpy as np
+import paddle
+
+from .rec_postprocess import AttnLabelDecode
+
+
+class TableLabelDecode(AttnLabelDecode):
+    """ """
+
+    def __init__(self, character_dict_path, merge_no_span_structure=False, **kwargs):
+        dict_character = []
+        with open(character_dict_path, "rb") as fin:
+            lines = fin.readlines()
+            for line in lines:
+                line = line.decode("utf-8").strip("\n").strip("\r\n")
+                dict_character.append(line)
+
+        if merge_no_span_structure:
+            if "<td></td>" not in dict_character:
+                dict_character.append("<td></td>")
+            if "<td>" in dict_character:
+                dict_character.remove("<td>")
+
+        dict_character = self.add_special_char(dict_character)
+        self.dict = {}
+        for i, char in enumerate(dict_character):
+            self.dict[char] = i
+        self.character = dict_character
+        self.td_token = ["<td>", "<td", "<td></td>"]
+
+    def __call__(self, preds, batch=None):
+        structure_probs = preds["structure_probs"]
+        bbox_preds = preds["loc_preds"]
+        if isinstance(structure_probs, paddle.Tensor):
+            structure_probs = structure_probs.numpy()
+        if isinstance(bbox_preds, paddle.Tensor):
+            bbox_preds = bbox_preds.numpy()
+        shape_list = batch[-1]
+        result = self.decode(structure_probs, bbox_preds, shape_list)
+        if len(batch) == 1:  # only contains shape
+            return result
+
+        label_decode_result = self.decode_label(batch)
+        return result, label_decode_result
+
+    def decode(self, structure_probs, bbox_preds, shape_list):
+        """convert text-label into text-index."""
+        ignored_tokens = self.get_ignored_tokens()
+        end_idx = self.dict[self.end_str]
+
+        structure_idx = structure_probs.argmax(axis=2)
+        structure_probs = structure_probs.max(axis=2)
+
+        structure_batch_list = []
+        bbox_batch_list = []
+        batch_size = len(structure_idx)
+        for batch_idx in range(batch_size):
+            structure_list = []
+            bbox_list = []
+            score_list = []
+            for idx in range(len(structure_idx[batch_idx])):
+                char_idx = int(structure_idx[batch_idx][idx])
+                if idx > 0 and char_idx == end_idx:
+                    break
+                if char_idx in ignored_tokens:
+                    continue
+                text = self.character[char_idx]
+                if text in self.td_token:
+                    bbox = bbox_preds[batch_idx, idx]
+                    bbox = self._bbox_decode(bbox, shape_list[batch_idx])
+                    bbox_list.append(bbox)
+                structure_list.append(text)
+                score_list.append(structure_probs[batch_idx, idx])
+            structure_batch_list.append([structure_list, np.mean(score_list)])
+            bbox_batch_list.append(np.array(bbox_list))
+        result = {
+            "bbox_batch_list": bbox_batch_list,
+            "structure_batch_list": structure_batch_list,
+        }
+        return result
+
+    def decode_label(self, batch):
+        """convert text-label into text-index."""
+        structure_idx = batch[1]
+        gt_bbox_list = batch[2]
+        shape_list = batch[-1]
+        ignored_tokens = self.get_ignored_tokens()
+        end_idx = self.dict[self.end_str]
+
+        structure_batch_list = []
+        bbox_batch_list = []
+        batch_size = len(structure_idx)
+        for batch_idx in range(batch_size):
+            structure_list = []
+            bbox_list = []
+            for idx in range(len(structure_idx[batch_idx])):
+                char_idx = int(structure_idx[batch_idx][idx])
+                if idx > 0 and char_idx == end_idx:
+                    break
+                if char_idx in ignored_tokens:
+                    continue
+                structure_list.append(self.character[char_idx])
+
+                bbox = gt_bbox_list[batch_idx][idx]
+                if bbox.sum() != 0:
+                    bbox = self._bbox_decode(bbox, shape_list[batch_idx])
+                    bbox_list.append(bbox)
+            structure_batch_list.append(structure_list)
+            bbox_batch_list.append(bbox_list)
+        result = {
+            "bbox_batch_list": bbox_batch_list,
+            "structure_batch_list": structure_batch_list,
+        }
+        return result
+
+    def _bbox_decode(self, bbox, shape):
+        h, w, ratio_h, ratio_w, pad_h, pad_w = shape
+        h, w = pad_h, pad_w
+        bbox[0::2] *= w
+        bbox[1::2] *= h
+        bbox[0::2] /= ratio_w
+        bbox[1::2] /= ratio_h
+        return bbox
+
+
+class TableMasterLabelDecode(TableLabelDecode):
+    """ """
+
+    def __init__(
+        self,
+        character_dict_path,
+        box_shape="ori",
+        merge_no_span_structure=True,
+        **kwargs,
+    ):
+        super(TableMasterLabelDecode, self).__init__(
+            character_dict_path, merge_no_span_structure
+        )
+        self.box_shape = box_shape
+        assert box_shape in [
+            "ori",
+            "pad",
+        ], "The shape used for box normalization must be ori or pad"
+
+    def add_special_char(self, dict_character):
+        self.beg_str = "<SOS>"
+        self.end_str = "<EOS>"
+        self.unknown_str = "<UKN>"
+        self.pad_str = "<PAD>"
+        dict_character = dict_character
+        dict_character = dict_character + [
+            self.unknown_str,
+            self.beg_str,
+            self.end_str,
+            self.pad_str,
+        ]
+        return dict_character
+
+    def get_ignored_tokens(self):
+        pad_idx = self.dict[self.pad_str]
+        start_idx = self.dict[self.beg_str]
+        end_idx = self.dict[self.end_str]
+        unknown_idx = self.dict[self.unknown_str]
+        return [start_idx, end_idx, pad_idx, unknown_idx]
+
+    def _bbox_decode(self, bbox, shape):
+        h, w, ratio_h, ratio_w, pad_h, pad_w = shape
+        if self.box_shape == "pad":
+            h, w = pad_h, pad_w
+        bbox[0::2] *= w
+        bbox[1::2] *= h
+        bbox[0::2] /= ratio_w
+        bbox[1::2] /= ratio_h
+        x, y, w, h = bbox
+        x1, y1, x2, y2 = x - w // 2, y - h // 2, x + w // 2, y + h // 2
+        bbox = np.array([x1, y1, x2, y2])
+        return bbox
diff --git a/ppocr/postprocess/vqa_token_re_layoutlm_postprocess.py b/ppocr/postprocess/vqa_token_re_layoutlm_postprocess.py
new file mode 100644
index 0000000..efdfffe
--- /dev/null
+++ b/ppocr/postprocess/vqa_token_re_layoutlm_postprocess.py
@@ -0,0 +1,96 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import paddle
+
+
+class VQAReTokenLayoutLMPostProcess(object):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, **kwargs):
+        super(VQAReTokenLayoutLMPostProcess, self).__init__()
+
+    def __call__(self, preds, label=None, *args, **kwargs):
+        pred_relations = preds["pred_relations"]
+        if isinstance(preds["pred_relations"], paddle.Tensor):
+            pred_relations = pred_relations.numpy()
+        pred_relations = self.decode_pred(pred_relations)
+
+        if label is not None:
+            return self._metric(pred_relations, label)
+        else:
+            return self._infer(pred_relations, *args, **kwargs)
+
+    def _metric(self, pred_relations, label):
+        return pred_relations, label[-1], label[-2]
+
+    def _infer(self, pred_relations, *args, **kwargs):
+        ser_results = kwargs["ser_results"]
+        entity_idx_dict_batch = kwargs["entity_idx_dict_batch"]
+
+        # merge relations and ocr info
+        results = []
+        for pred_relation, ser_result, entity_idx_dict in zip(
+            pred_relations, ser_results, entity_idx_dict_batch
+        ):
+            result = []
+            used_tail_id = []
+            for relation in pred_relation:
+                if relation["tail_id"] in used_tail_id:
+                    continue
+                used_tail_id.append(relation["tail_id"])
+                ocr_info_head = ser_result[entity_idx_dict[relation["head_id"]]]
+                ocr_info_tail = ser_result[entity_idx_dict[relation["tail_id"]]]
+                result.append((ocr_info_head, ocr_info_tail))
+            results.append(result)
+        return results
+
+    def decode_pred(self, pred_relations):
+        pred_relations_new = []
+        for pred_relation in pred_relations:
+            pred_relation_new = []
+            pred_relation = pred_relation[1 : pred_relation[0, 0, 0] + 1]
+            for relation in pred_relation:
+                relation_new = dict()
+                relation_new["head_id"] = relation[0, 0]
+                relation_new["head"] = tuple(relation[1])
+                relation_new["head_type"] = relation[2, 0]
+                relation_new["tail_id"] = relation[3, 0]
+                relation_new["tail"] = tuple(relation[4])
+                relation_new["tail_type"] = relation[5, 0]
+                relation_new["type"] = relation[6, 0]
+                pred_relation_new.append(relation_new)
+            pred_relations_new.append(pred_relation_new)
+        return pred_relations_new
+
+
+class DistillationRePostProcess(VQAReTokenLayoutLMPostProcess):
+    """
+    DistillationRePostProcess
+    """
+
+    def __init__(self, model_name=["Student"], key=None, **kwargs):
+        super().__init__(**kwargs)
+        if not isinstance(model_name, list):
+            model_name = [model_name]
+        self.model_name = model_name
+        self.key = key
+
+    def __call__(self, preds, *args, **kwargs):
+        output = dict()
+        for name in self.model_name:
+            pred = preds[name]
+            if self.key is not None:
+                pred = pred[self.key]
+            output[name] = super().__call__(pred, *args, **kwargs)
+        return output
diff --git a/ppocr/postprocess/vqa_token_ser_layoutlm_postprocess.py b/ppocr/postprocess/vqa_token_ser_layoutlm_postprocess.py
new file mode 100644
index 0000000..a10f070
--- /dev/null
+++ b/ppocr/postprocess/vqa_token_ser_layoutlm_postprocess.py
@@ -0,0 +1,116 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import numpy as np
+import paddle
+from ppocr.utils.utility import load_vqa_bio_label_maps
+
+
+class VQASerTokenLayoutLMPostProcess(object):
+    """Convert between text-label and text-index"""
+
+    def __init__(self, class_path, **kwargs):
+        super(VQASerTokenLayoutLMPostProcess, self).__init__()
+        label2id_map, self.id2label_map = load_vqa_bio_label_maps(class_path)
+
+        self.label2id_map_for_draw = dict()
+        for key in label2id_map:
+            if key.startswith("I-"):
+                self.label2id_map_for_draw[key] = label2id_map["B" + key[1:]]
+            else:
+                self.label2id_map_for_draw[key] = label2id_map[key]
+
+        self.id2label_map_for_show = dict()
+        for key in self.label2id_map_for_draw:
+            val = self.label2id_map_for_draw[key]
+            if key == "O":
+                self.id2label_map_for_show[val] = key
+            if key.startswith("B-") or key.startswith("I-"):
+                self.id2label_map_for_show[val] = key[2:]
+            else:
+                self.id2label_map_for_show[val] = key
+
+    def __call__(self, preds, batch=None, *args, **kwargs):
+        if isinstance(preds, tuple):
+            preds = preds[0]
+        if isinstance(preds, paddle.Tensor):
+            preds = preds.numpy()
+
+        if batch is not None:
+            return self._metric(preds, batch[5])
+        else:
+            return self._infer(preds, **kwargs)
+
+    def _metric(self, preds, label):
+        pred_idxs = preds.argmax(axis=2)
+        decode_out_list = [[] for _ in range(pred_idxs.shape[0])]
+        label_decode_out_list = [[] for _ in range(pred_idxs.shape[0])]
+
+        for i in range(pred_idxs.shape[0]):
+            for j in range(pred_idxs.shape[1]):
+                if label[i, j] != -100:
+                    label_decode_out_list[i].append(self.id2label_map[label[i, j]])
+                    decode_out_list[i].append(self.id2label_map[pred_idxs[i, j]])
+        return decode_out_list, label_decode_out_list
+
+    def _infer(self, preds, segment_offset_ids, ocr_infos):
+        results = []
+
+        for pred, segment_offset_id, ocr_info in zip(
+            preds, segment_offset_ids, ocr_infos
+        ):
+            pred = np.argmax(pred, axis=1)
+            pred = [self.id2label_map[idx] for idx in pred]
+
+            for idx in range(len(segment_offset_id)):
+                if idx == 0:
+                    start_id = 0
+                else:
+                    start_id = segment_offset_id[idx - 1]
+
+                end_id = segment_offset_id[idx]
+
+                curr_pred = pred[start_id:end_id]
+                curr_pred = [self.label2id_map_for_draw[p] for p in curr_pred]
+
+                if len(curr_pred) <= 0:
+                    pred_id = 0
+                else:
+                    counts = np.bincount(curr_pred)
+                    pred_id = np.argmax(counts)
+                ocr_info[idx]["pred_id"] = int(pred_id)
+                ocr_info[idx]["pred"] = self.id2label_map_for_show[int(pred_id)]
+            results.append(ocr_info)
+        return results
+
+
+class DistillationSerPostProcess(VQASerTokenLayoutLMPostProcess):
+    """
+    DistillationSerPostProcess
+    """
+
+    def __init__(self, class_path, model_name=["Student"], key=None, **kwargs):
+        super().__init__(class_path, **kwargs)
+        if not isinstance(model_name, list):
+            model_name = [model_name]
+        self.model_name = model_name
+        self.key = key
+
+    def __call__(self, preds, batch=None, *args, **kwargs):
+        output = dict()
+        for name in self.model_name:
+            pred = preds[name]
+            if self.key is not None:
+                pred = pred[self.key]
+            output[name] = super().__call__(pred, batch=batch, *args, **kwargs)
+        return output
diff --git a/ppocr/utils/EN_symbol_dict.txt b/ppocr/utils/EN_symbol_dict.txt
new file mode 100644
index 0000000..87c6d67
--- /dev/null
+++ b/ppocr/utils/EN_symbol_dict.txt
@@ -0,0 +1,94 @@
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+:
+;
+<
+=
+>
+?
+@
+[
+\
+]
+^
+_
+`
+{
+|
+}
+~
diff --git a/ppocr/utils/__init__.py b/ppocr/utils/__init__.py
new file mode 100755
index 0000000..abf198b
--- /dev/null
+++ b/ppocr/utils/__init__.py
@@ -0,0 +1,13 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/ppocr/utils/dict/README.md b/ppocr/utils/dict/README.md
new file mode 100644
index 0000000..7b68200
--- /dev/null
+++ b/ppocr/utils/dict/README.md
@@ -0,0 +1,5 @@
+## Dictionary and Corpus
+
+Dictionary files (usually character level vocabulary) are included here for easier configuration. Corpus contributed by OSS contributors are listed here, please respect copyrights when using them at your own risk.
+
+- Burmese corpus:  https://github.com/1chimaruGin/BurmeseCorpus
diff --git a/ppocr/utils/dict/ar_dict.txt b/ppocr/utils/dict/ar_dict.txt
new file mode 100644
index 0000000..fc63802
--- /dev/null
+++ b/ppocr/utils/dict/ar_dict.txt
@@ -0,0 +1,117 @@
+a
+r
+b
+i
+c
+_
+m
+g
+/
+1
+0
+I
+L
+S
+V
+R
+C
+2
+v
+l
+6
+3
+9
+.
+j
+p	
+ا
+ل
+م
+ر
+ج
+و
+ح
+ي
+ة
+5
+8
+7
+أ
+ب
+ض
+4
+ك
+س
+ه
+ث
+ن
+ط
+ع
+ت
+غ
+خ
+ف
+ئ
+ز
+إ
+د
+ص
+ظ
+ذ
+ش
+ى
+ق
+ؤ
+آ
+ء
+s
+e
+n
+w
+t
+u
+z
+d
+A
+N
+G
+h
+o
+E
+T
+H
+O
+B
+y
+F
+U
+J
+X
+W
+P
+Z
+M
+k
+q
+Y
+Q
+D
+f
+K
+x
+'
+%
+-
+#
+@
+!
+&
+$
+,
+:
+é
+?
++
+É
+(
+ 
diff --git a/ppocr/utils/dict/arabic_dict.txt b/ppocr/utils/dict/arabic_dict.txt
new file mode 100644
index 0000000..916d421
--- /dev/null
+++ b/ppocr/utils/dict/arabic_dict.txt
@@ -0,0 +1,161 @@
+!
+#
+$
+%
+&
+'
+(
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+?
+@
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+_
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+É
+é
+ء
+آ
+أ
+ؤ
+إ
+ئ
+ا
+ب
+ة
+ت
+ث
+ج
+ح
+خ
+د
+ذ
+ر
+ز
+س
+ش
+ص
+ض
+ط
+ظ
+ع
+غ
+ف
+ق
+ك
+ل
+م
+ن
+ه
+و
+ى
+ي
+ً
+ٌ
+ٍ
+َ
+ُ
+ِ
+ّ
+ْ
+ٓ
+ٔ
+ٰ
+ٱ
+ٹ
+پ
+چ
+ڈ
+ڑ
+ژ
+ک
+ڭ
+گ
+ں
+ھ
+ۀ
+ہ
+ۂ
+ۃ
+ۆ
+ۇ
+ۈ
+ۋ
+ی
+ې
+ے
+ۓ
+ە
+١
+٢
+٣
+٤
+٥
+٦
+٧
+٨
+٩
diff --git a/ppocr/utils/dict/be_dict.txt b/ppocr/utils/dict/be_dict.txt
new file mode 100644
index 0000000..f8458ba
--- /dev/null
+++ b/ppocr/utils/dict/be_dict.txt
@@ -0,0 +1,145 @@
+b
+e
+_
+i
+m
+g
+/
+2
+0
+I
+L
+S
+V
+R
+C
+1
+v
+a
+l
+6
+9
+4
+3
+.
+j
+p	
+п
+а
+з
+б
+у
+г
+н
+ц
+ь
+8
+м
+л
+і
+о
+ў
+ы
+7
+5
+М
+х
+с
+р
+ф
+я
+е
+д
+ж
+ю
+ч
+й
+к
+Д
+в
+Б
+т
+І
+ш
+ё
+э
+К
+Л
+Н
+А
+Ж
+Г
+В
+П
+З
+Е
+О
+Р
+С
+У
+Ё
+Й
+Т
+Ч
+Э
+Ц
+Ю
+Ш
+Ф
+Х
+Я
+Ь
+Ы
+Ў
+s
+c
+n
+w
+M
+o
+t
+T
+E
+A
+B
+u
+h
+y
+k
+r
+H
+d
+Y
+O
+U
+F
+f
+x
+D
+G
+N
+K
+P
+z
+J
+X
+W
+Z
+Q
+%
+-
+q
+@
+'
+!
+#
+&
+,
+:
+$
+(
+?
+é
++
+É
+ 
diff --git a/ppocr/utils/dict/bengali_dict.txt b/ppocr/utils/dict/bengali_dict.txt
new file mode 100644
index 0000000..89ffadf
--- /dev/null
+++ b/ppocr/utils/dict/bengali_dict.txt
@@ -0,0 +1,74 @@
+হ
+থ
+শ
+৫
+ক
+ও
+য
+০
+গ
+দ
+ড়
+খ
+য়
+ঋ
+ন
+অ
+৪
+এ
+ব
+ঠ
+ঢ
+৭
+৯
+ধ
+ঙ
+ট
+ঝ
+ৎ
+ণ
+ত
+র
+২
+চ
+ঌ
+ড
+৬
+ঔ
+প
+ভ
+ম
+ঢ়
+ঈ
+৮
+ঘ
+১
+ষ
+৩
+ফ
+ছ
+ল
+জ
+আ
+।
+ঊ
+ই
+স
+ঐ
+উ
+ঞ
+া
+্
+ু
+ী
+ে
+ং
+ি
+়
+ঁ
+ৃ
+ো
+ূ
+ৈ
+ৌ
+ঃ
diff --git a/ppocr/utils/dict/bg_dict.txt b/ppocr/utils/dict/bg_dict.txt
new file mode 100644
index 0000000..84713c3
--- /dev/null
+++ b/ppocr/utils/dict/bg_dict.txt
@@ -0,0 +1,140 @@
+!
+#
+$
+%
+&
+'
+(
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+?
+@
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+_
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+É
+é
+А
+Б
+В
+Г
+Д
+Е
+Ж
+З
+И
+Й
+К
+Л
+М
+Н
+О
+П
+Р
+С
+Т
+У
+Ф
+Х
+Ц
+Ч
+Ш
+Щ
+Ъ
+Ю
+Я
+а
+б
+в
+г
+д
+е
+ж
+з
+и
+й
+к
+л
+м
+н
+о
+п
+р
+с
+т
+у
+ф
+х
+ц
+ч
+ш
+щ
+ъ
+ь
+ю
+я
+ 
diff --git a/ppocr/utils/dict/bm_dict.txt b/ppocr/utils/dict/bm_dict.txt
new file mode 100644
index 0000000..bd68de9
--- /dev/null
+++ b/ppocr/utils/dict/bm_dict.txt
@@ -0,0 +1,160 @@
+က
+ခ
+ဂ
+ဃ
+င
+စ
+ဆ
+ဇ
+ဈ
+ဉ
+ည
+ဋ
+ဌ
+ဍ
+ဎ
+ဏ
+တ
+ထ
+ဒ
+ဓ
+န
+ပ
+ဖ
+ဗ
+ဘ
+မ
+ယ
+ရ
+လ
+ဝ
+သ
+ဟ
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
diff --git a/ppocr/utils/dict/bm_dict_add.txt b/ppocr/utils/dict/bm_dict_add.txt
new file mode 100644
index 0000000..29b6155
--- /dev/null
+++ b/ppocr/utils/dict/bm_dict_add.txt
@@ -0,0 +1,3219 @@
+က
+ခ
+ဂ
+ဃ
+င
+စ
+ဆ
+ဇ
+ဈ
+ဉ
+ည
+ဋ
+ဌ
+ဍ
+ဎ
+ဏ
+တ
+ထ
+ဒ
+ဓ
+န
+ပ
+ဖ
+ဗ
+ဘ
+မ
+ယ
+ရ
+လ
+ဝ
+သ
+ဟ
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+Ⴀ
+Ⴁ
+Ⴂ
+Ⴃ
+Ⴄ
+Ⴅ
+Ⴆ
+Ⴇ
+Ⴈ
+Ⴉ
+Ⴊ
+Ⴋ
+ᄀ
+ᄁ
+ᄂ
+ᄃ
+ᄄ
+ᄅ
+ᄆ
+ᄇ
+ᄈ
+ᄉ
+ᄊ
+ᄋ
+ᄌ
+ᄍ
+ᄎ
+ᄏ
+ᄐ
+ᄑ
+ᄒ
+
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+က
+ခ
+ဂ
+ဃ
+င
+စ
+ဆ
+ဇ
+ဈ
+ဉ
+ည
+ဋ
+ဌ
+ဍ
+ဎ
+ဏ
+တ
+ထ
+ဒ
+ဓ
+န
+ပ
+ဖ
+ဗ
+ဘ
+မ
+ယ
+ရ
+လ
+ဝ
+သ
+ဟ
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+Ⴀ
+Ⴁ
+Ⴂ
+Ⴃ
+Ⴄ
+Ⴅ
+Ⴆ
+Ⴇ
+Ⴈ
+Ⴉ
+Ⴊ
+Ⴋ
+ᄀ
+ᄁ
+ᄂ
+ᄃ
+ᄄ
+ᄅ
+ᄆ
+ᄇ
+ᄈ
+ᄉ
+ᄊ
+ᄋ
+ᄌ
+ᄍ
+ᄎ
+ᄏ
+ᄐ
+ᄑ
+ᄒ
+
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+က
+ခ
+ဂ
+ဃ
+င
+စ
+ဆ
+ဇ
+ဈ
+ဉ
+ည
+ဋ
+ဌ
+ဍ
+ဎ
+ဏ
+တ
+ထ
+ဒ
+ဓ
+န
+ပ
+ဖ
+ဗ
+ဘ
+မ
+ယ
+ရ
+လ
+ဝ
+သ
+ဟ
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+Ⴀ
+Ⴁ
+Ⴂ
+Ⴃ
+Ⴄ
+Ⴅ
+Ⴆ
+Ⴇ
+Ⴈ
+Ⴉ
+Ⴊ
+Ⴋ
+ᄀ
+ᄁ
+ᄂ
+ᄃ
+ᄄ
+ᄅ
+ᄆ
+ᄇ
+ᄈ
+ᄉ
+ᄊ
+ᄋ
+ᄌ
+ᄍ
+ᄎ
+ᄏ
+ᄐ
+ᄑ
+ᄒ
+
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+က
+ခ
+ဂ
+ဃ
+င
+စ
+ဆ
+ဇ
+ဈ
+ဉ
+ည
+ဋ
+ဌ
+ဍ
+ဎ
+ဏ
+တ
+ထ
+ဒ
+ဓ
+န
+ပ
+ဖ
+ဗ
+ဘ
+မ
+ယ
+ရ
+လ
+ဝ
+သ
+ဟ
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+Ⴀ
+Ⴁ
+Ⴂ
+Ⴃ
+Ⴄ
+Ⴅ
+Ⴆ
+Ⴇ
+Ⴈ
+Ⴉ
+Ⴊ
+Ⴋ
+ᄀ
+ᄁ
+ᄂ
+ᄃ
+ᄄ
+ᄅ
+ᄆ
+ᄇ
+ᄈ
+ᄉ
+ᄊ
+ᄋ
+ᄌ
+ᄍ
+ᄎ
+ᄏ
+ᄐ
+ᄑ
+ᄒ
+
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+က
+ခ
+ဂ
+ဃ
+င
+စ
+ဆ
+ဇ
+ဈ
+ဉ
+ည
+ဋ
+ဌ
+ဍ
+ဎ
+ဏ
+တ
+ထ
+ဒ
+ဓ
+န
+ပ
+ဖ
+ဗ
+ဘ
+မ
+ယ
+ရ
+လ
+ဝ
+သ
+ဟ
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+Ⴀ
+Ⴁ
+Ⴂ
+Ⴃ
+Ⴄ
+Ⴅ
+Ⴆ
+Ⴇ
+Ⴈ
+Ⴉ
+Ⴊ
+Ⴋ
+ᄀ
+ᄁ
+ᄂ
+ᄃ
+ᄄ
+ᄅ
+ᄆ
+ᄇ
+ᄈ
+ᄉ
+ᄊ
+ᄋ
+ᄌ
+ᄍ
+ᄎ
+ᄏ
+ᄐ
+ᄑ
+ᄒ
+
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+က
+ခ
+ဂ
+ဃ
+င
+စ
+ဆ
+ဇ
+ဈ
+ဉ
+ည
+ဋ
+ဌ
+ဍ
+ဎ
+ဏ
+တ
+ထ
+ဒ
+ဓ
+န
+ပ
+ဖ
+ဗ
+ဘ
+မ
+ယ
+ရ
+လ
+ဝ
+သ
+ဟ
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+Ⴀ
+Ⴁ
+Ⴂ
+Ⴃ
+Ⴄ
+Ⴅ
+Ⴆ
+Ⴇ
+Ⴈ
+Ⴉ
+Ⴊ
+Ⴋ
+ᄀ
+ᄁ
+ᄂ
+ᄃ
+ᄄ
+ᄅ
+ᄆ
+ᄇ
+ᄈ
+ᄉ
+ᄊ
+ᄋ
+ᄌ
+ᄍ
+ᄎ
+ᄏ
+ᄐ
+ᄑ
+ᄒ
+
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+က
+ခ
+ဂ
+ဃ
+င
+စ
+ဆ
+ဇ
+ဈ
+ဉ
+ည
+ဋ
+ဌ
+ဍ
+ဎ
+ဏ
+တ
+ထ
+ဒ
+ဓ
+န
+ပ
+ဖ
+ဗ
+ဘ
+မ
+ယ
+ရ
+လ
+ဝ
+သ
+ဟ
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+Ⴀ
+Ⴁ
+Ⴂ
+Ⴃ
+Ⴄ
+Ⴅ
+Ⴆ
+Ⴇ
+Ⴈ
+Ⴉ
+Ⴊ
+Ⴋ
+ᄀ
+ᄁ
+ᄂ
+ᄃ
+ᄄ
+ᄅ
+ᄆ
+ᄇ
+ᄈ
+ᄉ
+ᄊ
+ᄋ
+ᄌ
+ᄍ
+ᄎ
+ᄏ
+ᄐ
+ᄑ
+ᄒ
+
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+က
+ခ
+ဂ
+ဃ
+င
+စ
+ဆ
+ဇ
+ဈ
+ဉ
+ည
+ဋ
+ဌ
+ဍ
+ဎ
+ဏ
+တ
+ထ
+ဒ
+ဓ
+န
+ပ
+ဖ
+ဗ
+ဘ
+မ
+ယ
+ရ
+လ
+ဝ
+သ
+ဟ
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+Ⴀ
+Ⴁ
+Ⴂ
+Ⴃ
+Ⴄ
+Ⴅ
+Ⴆ
+Ⴇ
+Ⴈ
+Ⴉ
+Ⴊ
+Ⴋ
+ᄀ
+ᄁ
+ᄂ
+ᄃ
+ᄄ
+ᄅ
+ᄆ
+ᄇ
+ᄈ
+ᄉ
+ᄊ
+ᄋ
+ᄌ
+ᄍ
+ᄎ
+ᄏ
+ᄐ
+ᄑ
+ᄒ
+
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+က
+ခ
+ဂ
+ဃ
+င
+စ
+ဆ
+ဇ
+ဈ
+ဉ
+ည
+ဋ
+ဌ
+ဍ
+ဎ
+ဏ
+တ
+ထ
+ဒ
+ဓ
+န
+ပ
+ဖ
+ဗ
+ဘ
+မ
+ယ
+ရ
+လ
+ဝ
+သ
+ဟ
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+Ⴀ
+Ⴁ
+Ⴂ
+Ⴃ
+Ⴄ
+Ⴅ
+Ⴆ
+Ⴇ
+Ⴈ
+Ⴉ
+Ⴊ
+Ⴋ
+ᄀ
+ᄁ
+ᄂ
+ᄃ
+ᄄ
+ᄅ
+ᄆ
+ᄇ
+ᄈ
+ᄉ
+ᄊ
+ᄋ
+ᄌ
+ᄍ
+ᄎ
+ᄏ
+ᄐ
+ᄑ
+ᄒ
+
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+က
+ခ
+ဂ
+ဃ
+င
+စ
+ဆ
+ဇ
+ဈ
+ဉ
+ည
+ဋ
+ဌ
+ဍ
+ဎ
+ဏ
+တ
+ထ
+ဒ
+ဓ
+န
+ပ
+ဖ
+ဗ
+ဘ
+မ
+ယ
+ရ
+လ
+ဝ
+သ
+ဟ
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
+
+Ⴀ
+Ⴁ
+Ⴂ
+Ⴃ
+Ⴄ
+Ⴅ
+Ⴆ
+Ⴇ
+Ⴈ
+Ⴉ
+Ⴊ
+Ⴋ
+ᄀ
+ᄁ
+ᄂ
+ᄃ
+ᄄ
+ᄅ
+ᄆ
+ᄇ
+ᄈ
+ᄉ
+ᄊ
+ᄋ
+ᄌ
+ᄍ
+ᄎ
+ᄏ
+ᄐ
+ᄑ
+ᄒ
+
+ဠ
+အ
+ဢ
+ဣ
+ဤ
+ဥ
+ဦ
+ဧ
+ဨ
+ဩ
+ဪ
+ါ
+ာ
+ိ
+ီ
+ု
+ူ
+ေ
+ဲ
+ဳ
+ဴ
+ဵ
+ံ
+့
+း
+္
+်
+ျ
+ြ
+ွ
+ှ
+ဿ
+၀
+၁
+၂
+၃
+၄
+၅
+၆
+၇
+၈
+၉
+၊
+။
+၌
+၍
+၎
+၏
+ၐ
+ၑ
+ၒ
+ၓ
+ၔ
+ၕ
+ၖ
+ၗ
+ၘ
+ၙ
+ၚ
+ၛ
+ၜ
+ၝ
+ၞ
+ၟ
+ၠ
+ၡ
+ၢ
+ၣ
+ၤ
+ၥ
+ၦ
+ၧ
+ၨ
+ၩ
+ၪ
+ၫ
+ၬ
+ၭ
+ၮ
+ၯ
+ၰ
+ၱ
+ၲ
+ၳ
+ၴ
+ၵ
+ၶ
+ၷ
+ၸ
+ၹ
+ၺ
+ၻ
+ၼ
+ၽ
+ၾ
+ၿ
+ႀ
+ႁ
+ႂ
+ႃ
+ႄ
+ႅ
+ႆ
+ႇ
+ႈ
+ႉ
+ႊ
+ႋ
+ႌ
+ႍ
+ႎ
+ႏ
+႐
+႑
+႒
+႓
+႔
+႕
+႖
+႗
+႘
+႙
+ႚ
+ႛ
+ႜ
+ႝ
+႞
+႟
diff --git a/ppocr/utils/dict/bn_dict.txt b/ppocr/utils/dict/bn_dict.txt
new file mode 100644
index 0000000..655c585
--- /dev/null
+++ b/ppocr/utils/dict/bn_dict.txt
@@ -0,0 +1,477 @@
+ 
+!
+#
+$
+%
+&
+'
+(
++
+
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+?
+@
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+_
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+É
+é
+'
+"
+ঃ
+,
+;
+!
+?
+।
+—
+ ঃ-
+-
+(
+)
+[
+]
+{
+}
+√
+...
+.
+/
+\
+=
+<
+>
+৳
+°
+%
+অ
+আ
+ই
+ঈ
+উ
+ঊ
+ঋ
+এ
+ঐ
+ও
+ঔ
+্‌
+া
+ি
+ী
+ু
+ূ
+ে
+ৈ
+ো
+ৌ
+ৃ
+‍্য
+‍্র
+‍্ব
+ক
+খ
+গ
+ঘ
+ঙ
+চ
+ছ
+জ
+ঝ
+ঞ
+ট
+ঠ
+ড
+ঢ
+ণ
+ত
+থ
+দ
+ধ
+ন
+প
+ফ
+ব
+ভ
+ম
+য
+র
+ল
+শ
+ষ
+স
+হ
+ড়
+ঢ়
+য়
+ং
+ৎ
+ঁ
+ক্ক
+ক্ট
+ক্ত
+ত্র
+ক্ব
+ক্ম
+ক্য
+ক্র
+ক্ল
+ক্ষ
+ক্ষ্ণ
+ক্ষ্ব
+ক্ষ্ম
+ক্ষ্য
+ক্স
+র্ক
+র্ক্য
+খ্য
+খ্র
+র্খ
+গ্ণ
+গ্ধ
+গ্ধ্য
+গ্ধ্র
+গ্ন
+গ্ন্য
+গ্ব
+গ্ম
+গ্য
+গ্র
+গ্র্য
+গ্ল
+র্গ
+র্গ্য
+র্গ্র
+ঘ্ন
+ঘ্য
+ঘ্র
+র্ঘ্য
+র্ঘ
+ঙ্ক
+ঙ্ক্য
+ঙ্ক্ষ
+ঙ্খ
+ঙ্খ্য
+ঙ্গ
+ঙ্গ্য
+ঙ্ঘ
+ঙ্ঘ্য
+ঙ্ঘ্র
+ঙ্ম
+র্ঙ্গ
+চ্চ
+চ্ছ
+চ্ছ্ব
+চ্ছ্র
+চ্ঞ
+চ্ব
+চ্য
+র্চ্য
+র্চ
+র্ছ
+জ্জ
+জ্জ্ব
+জ্ঝ
+জ্ঞ
+জ্ব
+জ্য
+জ্র
+র্জ্য
+র্জ্জ
+র্জ্ঞ
+র্জ
+র্ঝ
+ঞ্চ
+ঞ্ছ
+ঞ্জ
+ঞ্ঝ
+ট্ট
+ট্ব
+ট্ম
+ট্য
+ট্র
+র্ট
+ড্ড
+ড্ব
+ড্ম
+ড্য
+ড্র
+র্ড
+ঢ্য
+ঢ্র
+র্ঢ্য
+ণ্ট
+ণ্ঠ
+ণ্ঠ্য
+ণ্ড
+ণ্ড্য
+ণ্ড্র
+ণ্ঢ
+ণ্ণ
+ণ্ব
+ণ্ম
+ণ্য
+র্ণ্য
+র্ণ
+ত্ত
+ত্ত্ব
+ত্ত্য
+ত্থ
+ত্ন
+ত্ব
+ত্ম
+ত্ম্য
+ত্য
+ত্র
+ত্র্য
+র্ত্য
+র্ত
+র্ত্ম
+র্ত্র
+থ্ব
+থ্য
+থ্র
+র্থ্য
+র্থ
+দ্গ
+দ্ঘ
+দ্দ
+দ্দ্ব
+দ্ধ
+দ্ব
+দ্ভ
+দ্ভ্র
+দ্ম
+দ্য
+দ্র
+দ্র্য
+র্দ
+র্দ্ব
+র্দ্র
+ধ্ন
+ধ্ব
+ধ্ম
+ধ্য
+ধ্র
+র্ধ
+র্ধ্ব
+ন্ট
+ন্ট্র
+ন্ঠ
+ন্ড
+ন্ড্র
+ন্ত
+ন্ত্ব
+ন্ত্য
+ন্ত্র
+ন্ত্র্য
+ন্থ
+ন্থ্র
+ন্দ
+ন্দ্য
+ন্দ্ব
+ন্দ্র
+ন্ধ
+ন্ধ্য
+ন্ধ্র
+ন্ন
+ন্ব
+ন্ম
+ন্য
+র্ন
+প্ট
+প্ত
+প্ন
+প্প
+প্য
+প্র
+প্র্য
+প্ল
+প্স
+র্প
+ফ্র
+ফ্ল
+র্ফ
+ব্জ
+ব্দ
+ব্ধ
+ব্ব
+ব্য
+ব্র
+ব্ল
+র্ব্য
+র্ব
+ভ্ব
+ভ্য
+ভ্র
+ভ্ল
+র্ভ
+ম্ন
+ম্প
+ম্প্র
+ম্ফ
+ম্ব
+ম্ব্র
+ম্ভ
+ম্ভ্র
+ম্ম
+ম্য
+ম্র
+ম্ল
+র্ম্য
+র্ম
+য্য
+র্য
+ল্ক
+ল্ক্য
+ল্গ
+ল্ট
+ল্ড
+ল্প
+ল্ফ
+ল্ব
+ল্ভ
+ল্ম
+ল্য
+ল্ল
+র্ল
+শ্চ
+শ্ছ
+শ্ন
+শ্ব
+শ্ম
+শ্য
+শ্র
+শ্ল
+র্শ্য
+র্শ
+র্শ্ব
+ষ্ক
+ষ্ক্ব
+ষ্ক্র
+ষ্ট
+ষ্ট্য
+ষ্ট্র
+ষ্ঠ
+ষ্ঠ্য
+ষ্ণ
+ষ্ণ্ব
+ষ্প
+ষ্প্র
+ষ্ফ
+ষ্ব
+ষ্ম
+ষ্য
+র্ষ্য
+র্ষ
+র্ষ্ট
+র্ষ্ণ
+র্ষ্ণ্য
+স্ক
+স্ক্র
+স্খ
+স্ট
+স্ট্র
+স্ত
+স্ত্ব
+স্ত্য
+স্ত্র
+স্থ
+স্থ্য
+স্ন
+স্ন্য
+স্প
+স্প্র
+স্প্ল
+স্ফ
+স্ব
+স্ম
+স্য
+স্র
+স্ল
+স্ক্ল
+র্স
+হ্ণ
+হ্ন
+হ্ব
+হ্ম
+হ্য
+হ্র
+হ্ল
+র্হ্য
+র্হ
+ড়্গ
+র্ৎ
+০
+১
+২
+৩
+৪
+৫
+৬
+৭
+৮
+৯
diff --git a/ppocr/utils/dict/chinese_cht_dict.txt b/ppocr/utils/dict/chinese_cht_dict.txt
new file mode 100644
index 0000000..cc1aa47
--- /dev/null
+++ b/ppocr/utils/dict/chinese_cht_dict.txt
@@ -0,0 +1,8421 @@
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+<
+=
+>
+?
+@
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+[
+\
+]
+^
+_
+`
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+{
+|
+}
+~
+¥
+®
+°
+±
+²
+´
+·
+»
+É
+Ë
+Ó
+×
+Ü
+à
+á
+ä
+è
+é
+ì
+í
+ò
+ó
+÷
+ú
+ü
+ā
+ē
+ī
+ō
+ū
+ǐ
+ǒ
+ɔ
+ɡ
+ʌ
+ˋ
+Λ
+Ο
+Φ
+Ω
+α
+β
+ε
+θ
+μ
+π
+З
+И
+Й
+П
+Я
+г
+—
+‖
+‘
+’
+“
+”
+•
+…
+‧
+′
+″
+※
+℃
+№
+™
+Ⅱ
+Ⅲ
+Ⅳ
+←
+↑
+→
+↓
+⇋
+∈
+∑
+√
+∞
+∣
+∧
+∩
+∫
+∶
+≈
+≠
+≤
+≥
+⊙
+⊥
+①
+②
+③
+④
+⑧
+⑴
+⑵
+⑶
+─
+│
+┅
+┌
+├
+█
+▎
+▏
+▕
+■
+□
+▪
+▲
+△
+▼
+◆
+◇
+○
+◎
+●
+◥
+★
+☆
+❋
+❤
+　
+、
+。
+〇
+〉
+《
+》
+「
+」
+『
+』
+【
+】
+〔
+〕
+〖
+〗
+の
+サ
+シ
+ジ
+マ
+ㄱ
+ㆍ
+㎏
+㎡
+㐂
+㐱
+㙟
+㴪
+㸃
+䖝
+䝉
+䰾
+䲁
+一
+丁
+七
+丄
+丈
+三
+上
+下
+丌
+不
+与
+丏
+丐
+丑
+且
+丕
+世
+丘
+丙
+丞
+丟
+両
+並
+丨
+丫
+中
+丰
+串
+丶
+丸
+丹
+主
+丼
+丿
+乂
+乃
+久
+么
+之
+乍
+乎
+乏
+乒
+乓
+乖
+乗
+乘
+乙
+乚
+乜
+九
+乞
+也
+乩
+乭
+乳
+乸
+乹
+乾
+亀
+亂
+亅
+了
+予
+亊
+事
+二
+亍
+云
+互
+亓
+五
+井
+亘
+些
+亜
+亞
+亟
+亠
+亡
+亢
+交
+亥
+亦
+亨
+享
+京
+亭
+亮
+亰
+亳
+亶
+亹
+人
+亻
+什
+仁
+仂
+仃
+仄
+仇
+仉
+今
+介
+仍
+仏
+仔
+仕
+他
+仗
+付
+仙
+仛
+仝
+仞
+仟
+仡
+代
+令
+以
+仨
+仫
+仮
+仰
+仲
+仳
+仵
+件
+仺
+任
+仼
+份
+仿
+企
+伃
+伈
+伉
+伊
+伋
+伍
+伎
+伏
+伐
+休
+伕
+伙
+伝
+伢
+伯
+估
+伱
+伴
+伶
+伷
+伸
+伺
+似
+伽
+伾
+佀
+佁
+佃
+但
+佇
+佈
+佉
+佋
+位
+低
+住
+佐
+佑
+体
+佔
+何
+佗
+佘
+余
+佚
+佛
+作
+佝
+佞
+佟
+你
+佣
+佤
+佧
+佩
+佬
+佯
+佰
+佳
+併
+佶
+佹
+佺
+佼
+佾
+使
+侁
+侃
+侄
+侅
+來
+侈
+侊
+例
+侍
+侏
+侑
+侖
+侗
+侘
+侚
+供
+依
+侞
+価
+侮
+侯
+侵
+侶
+侷
+侹
+便
+俁
+係
+促
+俄
+俅
+俊
+俋
+俌
+俍
+俎
+俏
+俐
+俑
+俗
+俘
+俚
+俛
+保
+俞
+俟
+俠
+信
+俬
+修
+俯
+俱
+俳
+俴
+俵
+俶
+俸
+俺
+俽
+俾
+倆
+倈
+倉
+個
+倌
+倍
+們
+倒
+倓
+倔
+倖
+倗
+倘
+候
+倚
+倜
+倞
+借
+倡
+倢
+倣
+値
+倦
+倧
+倩
+倪
+倫
+倬
+倭
+倮
+倻
+值
+偁
+偃
+假
+偈
+偉
+偊
+偌
+偍
+偎
+偏
+偓
+偕
+做
+停
+健
+偪
+偲
+側
+偵
+偶
+偷
+偸
+偽
+傀
+傃
+傅
+傈
+傉
+傍
+傑
+傒
+傕
+傖
+傘
+備
+傜
+傢
+傣
+催
+傭
+傲
+傳
+債
+傷
+傻
+傾
+僅
+僉
+僊
+働
+像
+僑
+僔
+僕
+僖
+僙
+僚
+僜
+僡
+僧
+僩
+僭
+僮
+僰
+僱
+僳
+僴
+僵
+價
+僻
+儀
+儁
+儂
+億
+儆
+儇
+儈
+儉
+儋
+儐
+儒
+儔
+儕
+儘
+儚
+儞
+償
+儡
+儥
+儦
+優
+儫
+儱
+儲
+儷
+儺
+儻
+儼
+兀
+允
+元
+兄
+充
+兆
+先
+光
+克
+兌
+免
+児
+兒
+兔
+兕
+兗
+兜
+入
+內
+全
+兩
+兪
+八
+公
+六
+兮
+共
+兵
+其
+具
+典
+兼
+兿
+冀
+冂
+円
+冇
+冉
+冊
+再
+冏
+冑
+冒
+冕
+冖
+冗
+冚
+冠
+冢
+冤
+冥
+冧
+冨
+冪
+冫
+冬
+冮
+冰
+冴
+冶
+冷
+冼
+冽
+凃
+凄
+准
+凈
+凋
+凌
+凍
+凖
+凜
+凝
+凞
+几
+凡
+処
+凪
+凬
+凰
+凱
+凳
+凵
+凶
+凸
+凹
+出
+函
+刀
+刁
+刂
+刃
+刄
+分
+切
+刈
+刊
+刎
+刑
+划
+列
+初
+判
+別
+刦
+刧
+刨
+利
+刪
+刮
+到
+制
+刷
+券
+刺
+刻
+刼
+剁
+剃
+則
+削
+剋
+剌
+前
+剎
+剏
+剔
+剖
+剛
+剝
+剡
+剣
+剩
+剪
+剮
+副
+割
+創
+剿
+劃
+劄
+劇
+劈
+劉
+劊
+劌
+劍
+劑
+劔
+力
+功
+加
+劣
+助
+努
+劫
+劬
+劭
+劵
+効
+劼
+劾
+勁
+勃
+勅
+勇
+勉
+勐
+勑
+勒
+勔
+動
+勖
+勗
+勘
+務
+勛
+勝
+勞
+募
+勢
+勣
+勤
+勦
+勰
+勱
+勲
+勳
+勵
+勷
+勸
+勺
+勻
+勾
+勿
+匂
+匄
+包
+匆
+匈
+匋
+匍
+匏
+匐
+匕
+化
+北
+匙
+匚
+匝
+匠
+匡
+匣
+匪
+匯
+匱
+匸
+匹
+匾
+匿
+區
+十
+千
+卅
+升
+午
+卉
+半
+卋
+卍
+卐
+卑
+卒
+卓
+協
+南
+博
+卜
+卞
+卟
+占
+卡
+卣
+卦
+卧
+卩
+卬
+卮
+卯
+印
+危
+卲
+即
+卵
+卷
+卸
+卹
+卺
+卻
+卽
+卿
+厄
+厓
+厔
+厙
+厚
+厝
+原
+厥
+厭
+厰
+厲
+厴
+厶
+去
+參
+叄
+又
+叉
+及
+友
+反
+収
+叔
+叕
+取
+受
+叛
+叟
+叡
+叢
+口
+古
+句
+另
+叨
+叩
+只
+叫
+召
+叭
+叮
+可
+台
+叱
+史
+右
+叵
+司
+叻
+叼
+吁
+吃
+各
+吆
+合
+吉
+吊
+吋
+同
+名
+后
+吏
+吐
+向
+吒
+吔
+吖
+君
+吝
+吞
+吟
+吠
+吡
+吥
+否
+吧
+吩
+含
+吮
+吱
+吲
+吳
+吵
+吶
+吸
+吹
+吻
+吼
+吾
+呀
+呂
+呃
+呈
+呉
+告
+呋
+呎
+呢
+呤
+呦
+周
+呱
+味
+呵
+呷
+呸
+呼
+命
+呾
+咀
+咁
+咂
+咄
+咅
+咆
+咋
+和
+咎
+咑
+咒
+咔
+咕
+咖
+咗
+咘
+咚
+咟
+咤
+咥
+咧
+咨
+咩
+咪
+咫
+咬
+咭
+咯
+咱
+咲
+咳
+咸
+咻
+咼
+咽
+咾
+咿
+哀
+品
+哂
+哄
+哆
+哇
+哈
+哉
+哌
+哎
+哏
+哐
+哖
+哚
+哞
+員
+哥
+哦
+哨
+哩
+哪
+哭
+哮
+哱
+哲
+哺
+哼
+唃
+唄
+唆
+唇
+唉
+唏
+唐
+唑
+唔
+唘
+唧
+唫
+唬
+唭
+售
+唯
+唱
+唳
+唵
+唷
+唸
+唻
+唾
+啁
+啃
+啄
+商
+啉
+啊
+啍
+問
+啓
+啖
+啚
+啜
+啞
+啟
+啡
+啣
+啤
+啥
+啦
+啪
+啫
+啯
+啰
+啱
+啲
+啵
+啶
+啷
+啻
+啼
+啾
+喀
+喂
+喃
+善
+喆
+喇
+喈
+喉
+喊
+喋
+喏
+喔
+喘
+喙
+喚
+喜
+喝
+喢
+喦
+喧
+喪
+喫
+喬
+單
+喰
+喱
+喲
+喳
+喵
+喹
+喻
+喼
+嗄
+嗅
+嗆
+嗇
+嗊
+嗎
+嗑
+嗒
+嗓
+嗔
+嗖
+嗚
+嗜
+嗝
+嗞
+嗡
+嗢
+嗣
+嗦
+嗨
+嗩
+嗪
+嗮
+嗯
+嗲
+嗶
+嗹
+嗽
+嘀
+嘅
+嘆
+嘉
+嘌
+嘍
+嘎
+嘏
+嘔
+嘗
+嘚
+嘛
+嘜
+嘞
+嘟
+嘢
+嘣
+嘥
+嘧
+嘩
+嘬
+嘮
+嘯
+嘰
+嘲
+嘴
+嘶
+嘸
+嘹
+嘻
+嘿
+噁
+噌
+噍
+噏
+噓
+噗
+噝
+噠
+噢
+噤
+噥
+噦
+器
+噩
+噪
+噬
+噯
+噰
+噲
+噴
+噶
+噸
+噹
+噻
+嚇
+嚈
+嚎
+嚏
+嚐
+嚒
+嚓
+嚕
+嚗
+嚙
+嚞
+嚟
+嚤
+嚦
+嚧
+嚨
+嚩
+嚮
+嚳
+嚴
+嚶
+嚷
+嚼
+嚿
+囀
+囂
+囃
+囉
+囊
+囍
+囑
+囒
+囓
+囗
+囚
+四
+囝
+回
+因
+囡
+団
+囤
+囧
+囪
+囮
+囯
+困
+囲
+図
+囶
+囷
+囹
+固
+囿
+圂
+圃
+圄
+圈
+圉
+國
+圍
+圏
+園
+圓
+圖
+圗
+團
+圜
+土
+圧
+在
+圩
+圪
+圭
+圯
+地
+圳
+圻
+圾
+址
+均
+坊
+坋
+坌
+坍
+坎
+坐
+坑
+坖
+坡
+坣
+坤
+坦
+坨
+坩
+坪
+坫
+坬
+坭
+坮
+坯
+坳
+坵
+坶
+坷
+坻
+垂
+垃
+垈
+型
+垍
+垓
+垕
+垚
+垛
+垞
+垟
+垠
+垢
+垣
+垮
+垯
+垰
+垵
+垸
+垻
+垿
+埃
+埅
+埇
+埈
+埋
+埌
+城
+埏
+埒
+埔
+埕
+埗
+埜
+域
+埠
+埡
+埤
+埧
+埨
+埪
+埭
+埮
+埴
+埵
+執
+培
+基
+埻
+埼
+堀
+堂
+堃
+堅
+堆
+堇
+堈
+堉
+堊
+堍
+堖
+堝
+堡
+堤
+堦
+堪
+堮
+堯
+堰
+報
+場
+堵
+堷
+堺
+塀
+塅
+塆
+塊
+塋
+塌
+塍
+塏
+塑
+塔
+塗
+塘
+塙
+塜
+塞
+塡
+塢
+塤
+塨
+塩
+填
+塬
+塭
+塰
+塱
+塲
+塵
+塹
+塽
+塾
+墀
+境
+墅
+墉
+墊
+墎
+墓
+増
+墘
+墜
+增
+墟
+墡
+墣
+墨
+墩
+墫
+墬
+墮
+墱
+墳
+墺
+墼
+墾
+壁
+壄
+壆
+壇
+壋
+壌
+壎
+壐
+壑
+壓
+壔
+壕
+壘
+壙
+壞
+壟
+壠
+壢
+壤
+壩
+士
+壬
+壯
+壱
+壴
+壹
+壺
+壽
+夀
+夆
+変
+夊
+夋
+夌
+夏
+夔
+夕
+外
+夙
+多
+夜
+夠
+夢
+夤
+夥
+大
+天
+太
+夫
+夬
+夭
+央
+夯
+失
+夷
+夾
+奀
+奄
+奇
+奈
+奉
+奎
+奏
+奐
+契
+奓
+奔
+奕
+套
+奘
+奚
+奠
+奢
+奣
+奧
+奩
+奪
+奫
+奭
+奮
+女
+奴
+奶
+她
+好
+妀
+妁
+如
+妃
+妄
+妊
+妍
+妏
+妑
+妒
+妓
+妖
+妙
+妝
+妞
+妠
+妤
+妥
+妧
+妨
+妭
+妮
+妯
+妲
+妳
+妸
+妹
+妺
+妻
+妾
+姀
+姁
+姃
+姆
+姈
+姉
+姊
+始
+姌
+姍
+姐
+姑
+姒
+姓
+委
+姚
+姜
+姝
+姣
+姥
+姦
+姨
+姪
+姫
+姬
+姮
+姵
+姶
+姸
+姻
+姿
+威
+娃
+娉
+娋
+娌
+娍
+娎
+娑
+娖
+娘
+娛
+娜
+娟
+娠
+娣
+娥
+娩
+娫
+娳
+娶
+娸
+娼
+娽
+婀
+婁
+婆
+婉
+婊
+婑
+婕
+婚
+婢
+婦
+婧
+婪
+婭
+婯
+婷
+婺
+婻
+婼
+婿
+媃
+媄
+媊
+媐
+媒
+媓
+媖
+媗
+媚
+媛
+媜
+媞
+媧
+媭
+媯
+媲
+媳
+媺
+媼
+媽
+媾
+媿
+嫁
+嫂
+嫄
+嫈
+嫉
+嫌
+嫖
+嫘
+嫚
+嫡
+嫣
+嫦
+嫩
+嫪
+嫲
+嫳
+嫵
+嫺
+嫻
+嬅
+嬈
+嬉
+嬋
+嬌
+嬗
+嬛
+嬝
+嬡
+嬤
+嬨
+嬪
+嬬
+嬭
+嬰
+嬴
+嬸
+嬾
+嬿
+孀
+孃
+孆
+孋
+孌
+子
+孑
+孔
+孕
+孖
+字
+存
+孚
+孛
+孜
+孝
+孟
+孢
+季
+孤
+孩
+孫
+孬
+孮
+孰
+孳
+孵
+學
+孺
+孻
+孽
+孿
+宀
+它
+宅
+宇
+守
+安
+宋
+完
+宍
+宏
+宓
+宕
+宗
+官
+宙
+定
+宛
+宜
+実
+客
+宣
+室
+宥
+宦
+宧
+宮
+宰
+害
+宴
+宵
+家
+宸
+容
+宿
+寀
+寁
+寂
+寄
+寅
+密
+寇
+寈
+寊
+富
+寐
+寒
+寓
+寔
+寕
+寖
+寗
+寘
+寛
+寜
+寞
+察
+寡
+寢
+寤
+寥
+實
+寧
+寨
+審
+寫
+寬
+寮
+寯
+寰
+寳
+寵
+寶
+寸
+寺
+対
+封
+専
+尃
+射
+將
+專
+尉
+尊
+尋
+對
+導
+小
+尐
+少
+尓
+尕
+尖
+尗
+尙
+尚
+尢
+尤
+尨
+尪
+尬
+就
+尷
+尹
+尺
+尻
+尼
+尾
+尿
+局
+屁
+屄
+居
+屆
+屇
+屈
+屋
+屌
+屍
+屎
+屏
+屐
+屑
+屓
+展
+屚
+屜
+屠
+屢
+層
+履
+屬
+屭
+屯
+山
+屹
+屺
+屻
+岀
+岈
+岌
+岐
+岑
+岔
+岡
+岢
+岣
+岧
+岩
+岪
+岫
+岬
+岰
+岱
+岳
+岵
+岷
+岸
+岻
+峁
+峅
+峇
+峋
+峍
+峒
+峘
+峙
+峚
+峠
+峨
+峩
+峪
+峭
+峯
+峰
+峴
+島
+峻
+峼
+峽
+崁
+崆
+崇
+崈
+崋
+崍
+崎
+崐
+崑
+崒
+崔
+崖
+崗
+崘
+崙
+崚
+崛
+崞
+崟
+崠
+崢
+崤
+崧
+崩
+崬
+崮
+崱
+崴
+崵
+崶
+崽
+嵇
+嵊
+嵋
+嵌
+嵎
+嵐
+嵒
+嵕
+嵖
+嵗
+嵙
+嵛
+嵜
+嵨
+嵩
+嵬
+嵮
+嵯
+嵰
+嵴
+嵻
+嵿
+嶁
+嶂
+嶃
+嶄
+嶇
+嶋
+嶌
+嶍
+嶒
+嶔
+嶗
+嶝
+嶠
+嶢
+嶦
+嶧
+嶪
+嶬
+嶰
+嶲
+嶴
+嶷
+嶸
+嶺
+嶼
+嶽
+巂
+巄
+巆
+巋
+巌
+巍
+巎
+巑
+巒
+巔
+巖
+巘
+巛
+川
+州
+巡
+巢
+工
+左
+巧
+巨
+巫
+差
+巰
+己
+已
+巳
+巴
+巶
+巷
+巻
+巽
+巾
+巿
+市
+布
+帆
+希
+帑
+帔
+帕
+帖
+帘
+帙
+帚
+帛
+帝
+帡
+帢
+帥
+師
+席
+帯
+帰
+帳
+帶
+帷
+常
+帽
+幀
+幃
+幄
+幅
+幌
+幔
+幕
+幗
+幚
+幛
+幟
+幡
+幢
+幣
+幪
+幫
+干
+平
+年
+幵
+幷
+幸
+幹
+幺
+幻
+幼
+幽
+幾
+庀
+庁
+広
+庇
+床
+序
+底
+庖
+店
+庚
+府
+庠
+庢
+庥
+度
+座
+庫
+庭
+庲
+庵
+庶
+康
+庸
+庹
+庼
+庾
+廁
+廂
+廄
+廆
+廈
+廉
+廊
+廋
+廌
+廍
+廑
+廓
+廔
+廕
+廖
+廙
+廚
+廝
+廞
+廟
+廠
+廡
+廢
+廣
+廧
+廨
+廩
+廬
+廰
+廱
+廳
+延
+廷
+廸
+建
+廻
+廼
+廿
+弁
+弄
+弅
+弇
+弈
+弉
+弊
+弋
+弍
+式
+弐
+弒
+弓
+弔
+引
+弖
+弗
+弘
+弛
+弟
+弢
+弦
+弧
+弨
+弩
+弭
+弱
+張
+強
+弸
+弼
+弾
+彀
+彄
+彅
+彆
+彈
+彊
+彌
+彎
+彐
+彔
+彖
+彗
+彘
+彙
+彜
+彞
+彠
+彡
+形
+彣
+彤
+彥
+彧
+彩
+彪
+彫
+彬
+彭
+彰
+影
+彳
+彷
+役
+彼
+彿
+往
+征
+徂
+待
+徇
+很
+徉
+徊
+律
+後
+徐
+徑
+徒
+得
+徘
+徙
+徜
+從
+徠
+御
+徧
+徨
+復
+循
+徫
+徬
+徭
+微
+徳
+徴
+徵
+德
+徸
+徹
+徽
+心
+忄
+必
+忉
+忌
+忍
+忐
+忑
+忒
+志
+忘
+忙
+応
+忝
+忞
+忠
+快
+忬
+忯
+忱
+忳
+念
+忻
+忽
+忿
+怍
+怎
+怒
+怕
+怖
+怙
+怛
+思
+怠
+怡
+急
+怦
+性
+怨
+怪
+怯
+怵
+恁
+恂
+恃
+恆
+恊
+恍
+恐
+恕
+恙
+恢
+恣
+恤
+恥
+恨
+恩
+恪
+恬
+恭
+息
+恰
+恵
+恿
+悄
+悅
+悆
+悉
+悌
+悍
+悔
+悖
+悚
+悛
+悝
+悞
+悟
+悠
+患
+悧
+您
+悪
+悰
+悲
+悳
+悵
+悶
+悸
+悼
+情
+惆
+惇
+惑
+惔
+惕
+惘
+惚
+惜
+惟
+惠
+惡
+惣
+惦
+惰
+惱
+惲
+想
+惶
+惹
+惺
+愁
+愃
+愆
+愈
+愉
+愍
+意
+愐
+愒
+愔
+愕
+愚
+愛
+愜
+感
+愣
+愧
+愨
+愫
+愭
+愴
+愷
+愼
+愾
+愿
+慄
+慈
+態
+慌
+慎
+慕
+慘
+慚
+慜
+慟
+慢
+慣
+慥
+慧
+慨
+慮
+慰
+慳
+慵
+慶
+慷
+慾
+憂
+憊
+憋
+憍
+憎
+憐
+憑
+憓
+憕
+憙
+憚
+憤
+憧
+憨
+憩
+憫
+憬
+憲
+憶
+憺
+憻
+憾
+懂
+懃
+懇
+懈
+應
+懋
+懌
+懍
+懐
+懣
+懦
+懮
+懲
+懵
+懶
+懷
+懸
+懺
+懼
+懽
+懾
+懿
+戀
+戇
+戈
+戊
+戌
+戍
+戎
+成
+我
+戒
+戔
+戕
+或
+戙
+戚
+戛
+戟
+戡
+戢
+戥
+戦
+戩
+截
+戮
+戰
+戱
+戲
+戳
+戴
+戶
+戸
+戻
+戽
+戾
+房
+所
+扁
+扆
+扇
+扈
+扉
+手
+扌
+才
+扎
+扒
+打
+扔
+托
+扙
+扛
+扞
+扣
+扥
+扦
+扭
+扮
+扯
+扳
+扶
+批
+扼
+找
+承
+技
+抃
+抄
+抇
+抉
+把
+抑
+抒
+抓
+投
+抖
+抗
+折
+抦
+披
+抬
+抱
+抵
+抹
+抻
+押
+抽
+抿
+拂
+拆
+拇
+拈
+拉
+拋
+拌
+拍
+拎
+拏
+拐
+拒
+拓
+拔
+拖
+拗
+拘
+拙
+拚
+招
+拜
+拝
+拡
+括
+拭
+拮
+拯
+拱
+拳
+拴
+拷
+拺
+拼
+拽
+拾
+拿
+持
+指
+按
+挎
+挑
+挖
+挙
+挨
+挪
+挫
+振
+挲
+挵
+挹
+挺
+挻
+挾
+捂
+捆
+捉
+捌
+捍
+捎
+捏
+捐
+捒
+捕
+捜
+捦
+捧
+捨
+捩
+捫
+捭
+捱
+捲
+捶
+捷
+捺
+捻
+掀
+掂
+掃
+掄
+掇
+授
+掉
+掌
+掏
+掐
+排
+掖
+掘
+掙
+掛
+掞
+掟
+掠
+採
+探
+掣
+接
+控
+推
+掩
+措
+掬
+掰
+掾
+揀
+揄
+揆
+揉
+揍
+描
+提
+插
+揔
+揖
+揚
+換
+握
+揪
+揭
+揮
+援
+揸
+揺
+損
+搏
+搐
+搓
+搔
+搖
+搗
+搜
+搞
+搠
+搢
+搪
+搬
+搭
+搳
+搴
+搵
+搶
+搽
+搾
+摂
+摒
+摔
+摘
+摜
+摞
+摟
+摠
+摧
+摩
+摭
+摯
+摳
+摴
+摵
+摶
+摸
+摹
+摺
+摻
+摽
+撃
+撇
+撈
+撐
+撒
+撓
+撕
+撖
+撙
+撚
+撞
+撣
+撤
+撥
+撩
+撫
+撬
+播
+撮
+撰
+撲
+撳
+撻
+撼
+撾
+撿
+擀
+擁
+擂
+擅
+擇
+擊
+擋
+操
+擎
+擒
+擔
+擘
+據
+擠
+擢
+擥
+擦
+擬
+擯
+擰
+擱
+擲
+擴
+擷
+擺
+擼
+擾
+攀
+攏
+攔
+攖
+攘
+攜
+攝
+攞
+攢
+攣
+攤
+攪
+攫
+攬
+支
+攴
+攵
+收
+攷
+攸
+改
+攻
+攽
+放
+政
+故
+效
+敍
+敎
+敏
+救
+敔
+敕
+敖
+敗
+敘
+教
+敝
+敞
+敟
+敢
+散
+敦
+敫
+敬
+敭
+敲
+整
+敵
+敷
+數
+敻
+敾
+斂
+斃
+文
+斌
+斎
+斐
+斑
+斕
+斖
+斗
+料
+斛
+斜
+斝
+斟
+斡
+斤
+斥
+斧
+斬
+斯
+新
+斷
+方
+於
+施
+斿
+旁
+旂
+旃
+旄
+旅
+旉
+旋
+旌
+旎
+族
+旖
+旗
+旙
+旛
+旡
+既
+日
+旦
+旨
+早
+旬
+旭
+旱
+旲
+旳
+旺
+旻
+旼
+旽
+旾
+旿
+昀
+昂
+昃
+昆
+昇
+昉
+昊
+昌
+昍
+明
+昏
+昐
+易
+昔
+昕
+昚
+昛
+昜
+昝
+昞
+星
+映
+昡
+昣
+昤
+春
+昧
+昨
+昪
+昫
+昭
+是
+昰
+昱
+昴
+昵
+昶
+昺
+晁
+時
+晃
+晈
+晉
+晊
+晏
+晗
+晙
+晚
+晛
+晝
+晞
+晟
+晤
+晦
+晧
+晨
+晩
+晪
+晫
+晭
+普
+景
+晰
+晳
+晴
+晶
+晷
+晸
+智
+晾
+暃
+暄
+暅
+暇
+暈
+暉
+暊
+暌
+暎
+暏
+暐
+暑
+暕
+暖
+暗
+暘
+暝
+暟
+暠
+暢
+暦
+暨
+暫
+暮
+暱
+暲
+暴
+暸
+暹
+暻
+暾
+曄
+曅
+曆
+曇
+曉
+曌
+曔
+曖
+曙
+曜
+曝
+曠
+曦
+曧
+曨
+曩
+曬
+曮
+曰
+曲
+曳
+更
+曶
+曷
+書
+曹
+曺
+曼
+曽
+曾
+替
+最
+會
+月
+有
+朊
+朋
+服
+朏
+朐
+朓
+朔
+朕
+朖
+朗
+望
+朝
+期
+朦
+朧
+木
+未
+末
+本
+札
+朱
+朴
+朵
+朶
+朽
+朿
+杁
+杉
+杋
+杌
+李
+杏
+材
+村
+杓
+杖
+杙
+杜
+杞
+束
+杠
+杣
+杤
+杧
+杬
+杭
+杯
+東
+杲
+杳
+杴
+杵
+杷
+杻
+杼
+松
+板
+极
+枇
+枉
+枋
+枏
+析
+枕
+枖
+林
+枚
+枛
+果
+枝
+枠
+枡
+枯
+枰
+枱
+枲
+枳
+架
+枷
+枸
+枹
+枼
+柁
+柃
+柄
+柉
+柊
+柎
+柏
+某
+柑
+柒
+染
+柔
+柘
+柚
+柜
+柝
+柞
+柟
+查
+柩
+柬
+柯
+柰
+柱
+柳
+柴
+柵
+柶
+柷
+査
+柾
+柿
+栃
+栄
+栐
+栒
+栓
+栜
+栝
+栞
+校
+栢
+栨
+栩
+株
+栲
+栴
+核
+根
+栻
+格
+栽
+桀
+桁
+桂
+桃
+桄
+桅
+框
+案
+桉
+桌
+桎
+桐
+桑
+桓
+桔
+桕
+桖
+桙
+桜
+桝
+桫
+桱
+桲
+桴
+桶
+桷
+桼
+桿
+梀
+梁
+梂
+梃
+梅
+梆
+梉
+梏
+梓
+梔
+梗
+梘
+條
+梟
+梠
+梢
+梣
+梧
+梨
+梫
+梭
+梯
+械
+梱
+梳
+梵
+梶
+梽
+棄
+棆
+棉
+棋
+棍
+棐
+棒
+棓
+棕
+棖
+棗
+棘
+棚
+棛
+棟
+棠
+棡
+棣
+棧
+棨
+棩
+棪
+棫
+森
+棱
+棲
+棵
+棶
+棹
+棺
+棻
+棼
+棽
+椅
+椆
+椇
+椋
+植
+椎
+椏
+椒
+椙
+椥
+椪
+椰
+椲
+椴
+椵
+椹
+椽
+椿
+楂
+楊
+楓
+楔
+楗
+楙
+楚
+楝
+楞
+楠
+楡
+楢
+楣
+楤
+楦
+楧
+楨
+楫
+業
+楮
+楯
+楳
+極
+楷
+楸
+楹
+楽
+楿
+概
+榆
+榊
+榍
+榎
+榑
+榔
+榕
+榖
+榗
+榘
+榛
+榜
+榞
+榢
+榣
+榤
+榦
+榧
+榨
+榫
+榭
+榮
+榲
+榴
+榷
+榻
+榿
+槀
+槁
+槃
+槊
+構
+槌
+槍
+槎
+槐
+槓
+槔
+槗
+様
+槙
+槤
+槩
+槭
+槰
+槱
+槲
+槳
+槺
+槻
+槼
+槽
+槿
+樀
+樁
+樂
+樅
+樆
+樊
+樋
+樑
+樓
+樗
+樘
+標
+樞
+樟
+模
+樣
+樨
+権
+樫
+樵
+樸
+樹
+樺
+樻
+樽
+樾
+橄
+橇
+橈
+橋
+橐
+橒
+橓
+橘
+橙
+橚
+機
+橡
+橢
+橪
+橫
+橿
+檀
+檄
+檇
+檉
+檊
+檎
+檐
+檔
+檗
+檜
+檞
+檠
+檡
+檢
+檣
+檦
+檨
+檫
+檬
+檯
+檳
+檵
+檸
+檻
+檽
+櫂
+櫃
+櫆
+櫈
+櫓
+櫚
+櫛
+櫞
+櫟
+櫥
+櫨
+櫪
+櫱
+櫸
+櫻
+櫾
+櫿
+欄
+欉
+權
+欏
+欒
+欖
+欞
+欠
+次
+欣
+欥
+欲
+欸
+欹
+欺
+欽
+款
+歆
+歇
+歉
+歊
+歌
+歎
+歐
+歓
+歙
+歛
+歡
+止
+正
+此
+步
+武
+歧
+歩
+歪
+歲
+歳
+歴
+歷
+歸
+歹
+死
+歿
+殂
+殃
+殄
+殆
+殉
+殊
+殑
+殖
+殘
+殛
+殞
+殟
+殤
+殭
+殮
+殯
+殲
+殳
+段
+殷
+殺
+殻
+殼
+殿
+毀
+毅
+毆
+毉
+毋
+毌
+母
+毎
+每
+毐
+毒
+毓
+比
+毖
+毗
+毘
+毛
+毫
+毬
+毯
+毴
+毸
+毽
+毿
+氂
+氈
+氍
+氏
+氐
+民
+氓
+氖
+気
+氘
+氙
+氚
+氛
+氟
+氣
+氦
+氧
+氨
+氪
+氫
+氬
+氮
+氯
+氰
+水
+氵
+氷
+永
+氹
+氻
+氽
+氾
+汀
+汁
+求
+汊
+汎
+汐
+汕
+汗
+汛
+汜
+汝
+汞
+江
+池
+污
+汧
+汨
+汩
+汪
+汭
+汰
+汲
+汴
+汶
+決
+汽
+汾
+沁
+沂
+沃
+沄
+沅
+沆
+沇
+沈
+沉
+沌
+沍
+沏
+沐
+沒
+沓
+沔
+沖
+沘
+沙
+沚
+沛
+沜
+沢
+沨
+沫
+沭
+沮
+沯
+沱
+河
+沸
+油
+沺
+治
+沼
+沽
+沾
+沿
+況
+泂
+泄
+泆
+泇
+泉
+泊
+泌
+泐
+泓
+泔
+法
+泖
+泗
+泚
+泛
+泠
+泡
+波
+泣
+泥
+泩
+泫
+泮
+泯
+泰
+泱
+泳
+泵
+洄
+洋
+洌
+洎
+洗
+洙
+洛
+洞
+洢
+洣
+洤
+津
+洨
+洩
+洪
+洮
+洱
+洲
+洳
+洵
+洸
+洹
+洺
+活
+洽
+派
+流
+浄
+浙
+浚
+浛
+浜
+浞
+浟
+浠
+浡
+浣
+浤
+浥
+浦
+浩
+浪
+浮
+浯
+浴
+浵
+海
+浸
+浹
+涅
+涇
+消
+涉
+涌
+涎
+涑
+涓
+涔
+涕
+涙
+涪
+涫
+涮
+涯
+液
+涵
+涸
+涼
+涿
+淄
+淅
+淆
+淇
+淋
+淌
+淍
+淎
+淏
+淑
+淓
+淖
+淘
+淙
+淚
+淛
+淝
+淞
+淠
+淡
+淤
+淥
+淦
+淨
+淩
+淪
+淫
+淬
+淮
+淯
+淰
+深
+淳
+淵
+淶
+混
+淸
+淹
+淺
+添
+淼
+淽
+渃
+清
+済
+渉
+渋
+渕
+渙
+渚
+減
+渝
+渟
+渠
+渡
+渣
+渤
+渥
+渦
+渫
+測
+渭
+港
+渲
+渴
+游
+渺
+渼
+渽
+渾
+湃
+湄
+湉
+湊
+湍
+湓
+湔
+湖
+湘
+湛
+湜
+湞
+湟
+湣
+湥
+湧
+湫
+湮
+湯
+湳
+湴
+湼
+満
+溁
+溇
+溈
+溉
+溋
+溎
+溏
+源
+準
+溙
+溜
+溝
+溟
+溢
+溥
+溦
+溧
+溪
+溫
+溯
+溱
+溲
+溴
+溵
+溶
+溺
+溼
+滀
+滁
+滂
+滄
+滅
+滇
+滈
+滉
+滋
+滌
+滎
+滏
+滑
+滓
+滔
+滕
+滘
+滙
+滝
+滬
+滯
+滲
+滴
+滷
+滸
+滹
+滻
+滽
+滾
+滿
+漁
+漂
+漆
+漇
+漈
+漎
+漏
+漓
+演
+漕
+漚
+漠
+漢
+漣
+漩
+漪
+漫
+漬
+漯
+漱
+漲
+漳
+漴
+漵
+漷
+漸
+漼
+漾
+漿
+潁
+潑
+潔
+潘
+潛
+潞
+潟
+潢
+潤
+潭
+潮
+潯
+潰
+潲
+潺
+潼
+潽
+潾
+潿
+澀
+澁
+澂
+澄
+澆
+澇
+澈
+澉
+澋
+澌
+澍
+澎
+澔
+澗
+澠
+澡
+澣
+澤
+澥
+澧
+澪
+澮
+澯
+澱
+澳
+澶
+澹
+澻
+激
+濁
+濂
+濃
+濉
+濊
+濋
+濕
+濘
+濙
+濛
+濞
+濟
+濠
+濡
+濤
+濫
+濬
+濮
+濯
+濰
+濱
+濲
+濶
+濺
+濼
+濾
+瀁
+瀅
+瀆
+瀉
+瀍
+瀏
+瀑
+瀔
+瀕
+瀘
+瀚
+瀛
+瀝
+瀞
+瀟
+瀠
+瀣
+瀦
+瀧
+瀨
+瀬
+瀰
+瀲
+瀴
+瀶
+瀹
+瀾
+灃
+灊
+灌
+灑
+灘
+灝
+灞
+灡
+灣
+灤
+灧
+火
+灰
+灴
+灸
+灼
+災
+炁
+炅
+炆
+炊
+炎
+炒
+炔
+炕
+炘
+炙
+炟
+炣
+炤
+炫
+炬
+炭
+炮
+炯
+炱
+炲
+炳
+炷
+炸
+為
+炻
+烈
+烉
+烊
+烋
+烏
+烒
+烔
+烘
+烙
+烜
+烝
+烤
+烯
+烱
+烴
+烷
+烹
+烺
+烽
+焃
+焄
+焉
+焊
+焌
+焓
+焗
+焙
+焚
+焜
+焞
+無
+焦
+焯
+焰
+焱
+焴
+然
+焻
+焼
+焿
+煇
+煉
+煊
+煌
+煎
+煐
+煒
+煔
+煕
+煖
+煙
+煚
+煜
+煞
+煠
+煤
+煥
+煦
+照
+煨
+煩
+煬
+煮
+煲
+煳
+煵
+煶
+煸
+煽
+熄
+熅
+熇
+熈
+熊
+熏
+熒
+熔
+熖
+熗
+熘
+熙
+熜
+熟
+熠
+熤
+熥
+熨
+熬
+熯
+熱
+熲
+熳
+熵
+熹
+熺
+熼
+熾
+熿
+燁
+燃
+燄
+燈
+燉
+燊
+燎
+燏
+燐
+燒
+燔
+燕
+燘
+燙
+燚
+燜
+燝
+營
+燥
+燦
+燧
+燫
+燬
+燭
+燮
+燴
+燹
+燻
+燼
+燾
+燿
+爀
+爆
+爌
+爍
+爐
+爔
+爚
+爛
+爝
+爨
+爪
+爬
+爭
+爯
+爰
+爲
+爵
+父
+爸
+爹
+爺
+爻
+爽
+爾
+爿
+牁
+牂
+牆
+片
+版
+牌
+牒
+牕
+牖
+牘
+牙
+牛
+牝
+牟
+牠
+牡
+牢
+牧
+物
+牯
+牲
+特
+牻
+牼
+牽
+犀
+犁
+犂
+犇
+犍
+犎
+犖
+犛
+犢
+犧
+犨
+犬
+犯
+犰
+犴
+犽
+狀
+狂
+狄
+狍
+狎
+狐
+狒
+狓
+狗
+狙
+狛
+狟
+狠
+狡
+狦
+狨
+狩
+狳
+狶
+狷
+狸
+狹
+狻
+狼
+猁
+猄
+猇
+猊
+猗
+猙
+猛
+猜
+猝
+猞
+猢
+猥
+猨
+猩
+猳
+猴
+猶
+猷
+猺
+猻
+猾
+猿
+獁
+獃
+獄
+獅
+獇
+獎
+獏
+獐
+獒
+獠
+獢
+獣
+獨
+獬
+獮
+獯
+獰
+獲
+獴
+獵
+獷
+獸
+獺
+獻
+獼
+獾
+玀
+玄
+玆
+率
+玉
+王
+玎
+玏
+玓
+玕
+玖
+玗
+玘
+玙
+玟
+玠
+玡
+玢
+玥
+玧
+玨
+玩
+玫
+玭
+玲
+玳
+玶
+玷
+玹
+玻
+玾
+珀
+珂
+珅
+珈
+珉
+珊
+珌
+珍
+珎
+珏
+珖
+珙
+珝
+珞
+珠
+珡
+珣
+珤
+珥
+珦
+珧
+珩
+珪
+班
+珮
+珵
+珹
+珺
+珽
+現
+琁
+球
+琄
+琅
+理
+琇
+琉
+琊
+琍
+琎
+琚
+琛
+琡
+琢
+琤
+琥
+琦
+琨
+琪
+琬
+琮
+琯
+琰
+琱
+琳
+琴
+琵
+琶
+琹
+琺
+琿
+瑀
+瑁
+瑂
+瑄
+瑅
+瑆
+瑈
+瑊
+瑋
+瑑
+瑒
+瑕
+瑗
+瑙
+瑚
+瑛
+瑜
+瑝
+瑞
+瑟
+瑠
+瑢
+瑣
+瑤
+瑥
+瑧
+瑨
+瑩
+瑪
+瑭
+瑯
+瑰
+瑱
+瑳
+瑴
+瑺
+瑾
+璀
+璁
+璃
+璄
+璆
+璇
+璈
+璉
+璋
+璌
+璐
+璕
+璘
+璙
+璚
+璜
+璞
+璟
+璠
+璡
+璣
+璥
+璦
+璧
+璨
+璩
+璪
+璫
+璬
+璮
+環
+璱
+璵
+璸
+璹
+璽
+璿
+瓈
+瓊
+瓌
+瓏
+瓑
+瓔
+瓖
+瓘
+瓚
+瓛
+瓜
+瓞
+瓠
+瓢
+瓣
+瓤
+瓦
+瓮
+瓴
+瓶
+瓷
+瓿
+甂
+甄
+甌
+甍
+甑
+甕
+甘
+甙
+甚
+甜
+生
+甡
+產
+産
+甥
+甦
+用
+甩
+甪
+甫
+甬
+甯
+田
+由
+甲
+申
+男
+甸
+甹
+町
+甾
+畀
+畇
+畈
+畊
+畋
+界
+畎
+畏
+畐
+畑
+畔
+留
+畜
+畝
+畠
+畢
+略
+畦
+畧
+番
+畫
+畬
+畯
+異
+畲
+畳
+畵
+當
+畷
+畸
+畹
+畿
+疃
+疆
+疇
+疊
+疋
+疌
+疍
+疏
+疑
+疒
+疕
+疙
+疚
+疝
+疣
+疤
+疥
+疫
+疲
+疳
+疵
+疸
+疹
+疼
+疽
+疾
+痂
+病
+症
+痊
+痍
+痔
+痕
+痘
+痙
+痛
+痞
+痟
+痠
+痢
+痣
+痤
+痧
+痩
+痰
+痱
+痲
+痴
+痹
+痺
+痿
+瘀
+瘁
+瘊
+瘋
+瘍
+瘓
+瘙
+瘜
+瘞
+瘟
+瘠
+瘡
+瘢
+瘤
+瘦
+瘧
+瘩
+瘰
+瘴
+瘺
+癀
+療
+癆
+癇
+癌
+癒
+癖
+癘
+癜
+癟
+癡
+癢
+癤
+癥
+癩
+癬
+癭
+癮
+癯
+癰
+癱
+癲
+癸
+発
+登
+發
+白
+百
+皂
+的
+皆
+皇
+皈
+皋
+皎
+皐
+皓
+皖
+皙
+皚
+皛
+皝
+皞
+皮
+皰
+皴
+皷
+皸
+皺
+皿
+盂
+盃
+盅
+盆
+盈
+益
+盋
+盌
+盎
+盒
+盔
+盛
+盜
+盞
+盟
+盡
+監
+盤
+盥
+盦
+盧
+盨
+盩
+盪
+盫
+目
+盯
+盱
+盲
+直
+盷
+相
+盹
+盺
+盼
+盾
+眀
+省
+眉
+看
+県
+眙
+眛
+眜
+眞
+真
+眠
+眥
+眨
+眩
+眭
+眯
+眵
+眶
+眷
+眸
+眺
+眼
+眾
+着
+睇
+睛
+睜
+睞
+睡
+睢
+督
+睥
+睦
+睨
+睪
+睫
+睭
+睹
+睺
+睽
+睾
+睿
+瞄
+瞅
+瞋
+瞌
+瞎
+瞑
+瞓
+瞞
+瞢
+瞥
+瞧
+瞪
+瞫
+瞬
+瞭
+瞰
+瞳
+瞻
+瞼
+瞽
+瞿
+矇
+矍
+矗
+矚
+矛
+矜
+矞
+矢
+矣
+知
+矧
+矩
+短
+矮
+矯
+石
+矸
+矽
+砂
+砋
+砌
+砍
+砒
+研
+砝
+砢
+砥
+砦
+砧
+砩
+砫
+砭
+砮
+砯
+砰
+砲
+砳
+破
+砵
+砷
+砸
+砼
+硂
+硃
+硅
+硇
+硏
+硐
+硒
+硓
+硚
+硜
+硝
+硤
+硨
+硫
+硬
+硭
+硯
+硼
+碁
+碇
+碉
+碌
+碎
+碑
+碓
+碕
+碗
+碘
+碚
+碟
+碡
+碣
+碧
+碩
+碪
+碭
+碰
+碲
+碳
+碴
+碶
+碸
+確
+碻
+碼
+碽
+碾
+磁
+磅
+磊
+磋
+磐
+磔
+磕
+磘
+磙
+磚
+磜
+磡
+磨
+磪
+磬
+磯
+磱
+磲
+磵
+磷
+磺
+磻
+磾
+礁
+礄
+礎
+礐
+礑
+礒
+礙
+礠
+礦
+礪
+礫
+礬
+礮
+礱
+礴
+示
+礻
+礽
+社
+祀
+祁
+祂
+祆
+祇
+祈
+祉
+祋
+祏
+祐
+祓
+祕
+祖
+祗
+祙
+祚
+祛
+祜
+祝
+神
+祟
+祠
+祥
+祧
+票
+祭
+祹
+祺
+祼
+祿
+禁
+禃
+禇
+禍
+禎
+福
+禑
+禓
+禔
+禕
+禘
+禛
+禟
+禠
+禤
+禦
+禧
+禨
+禩
+禪
+禮
+禰
+禱
+禵
+禹
+禺
+禼
+禽
+禾
+禿
+秀
+私
+秈
+秉
+秋
+科
+秒
+秕
+秘
+租
+秠
+秣
+秤
+秦
+秧
+秩
+秭
+秳
+秸
+移
+稀
+稅
+稈
+稉
+程
+稍
+稑
+稔
+稗
+稘
+稙
+稚
+稜
+稞
+稟
+稠
+種
+稱
+稲
+稷
+稹
+稺
+稻
+稼
+稽
+稾
+稿
+穀
+穂
+穆
+穈
+穉
+穌
+積
+穎
+穗
+穟
+穠
+穡
+穢
+穣
+穩
+穫
+穰
+穴
+穵
+究
+穹
+空
+穿
+突
+窄
+窅
+窈
+窋
+窒
+窕
+窖
+窗
+窘
+窟
+窠
+窣
+窨
+窩
+窪
+窮
+窯
+窰
+窶
+窺
+窿
+竄
+竅
+竇
+竈
+竊
+立
+竑
+站
+竜
+竟
+章
+竣
+童
+竦
+竩
+竭
+端
+競
+竹
+竺
+竻
+竿
+笄
+笆
+笈
+笏
+笑
+笘
+笙
+笛
+笞
+笠
+笥
+符
+笨
+笩
+笪
+第
+笭
+笮
+笯
+笱
+笳
+笹
+筅
+筆
+等
+筊
+筋
+筌
+筍
+筏
+筐
+筒
+答
+策
+筘
+筠
+筥
+筦
+筧
+筬
+筭
+筱
+筲
+筳
+筵
+筶
+筷
+筻
+箆
+箇
+箋
+箍
+箏
+箐
+箑
+箒
+箔
+箕
+算
+箜
+管
+箬
+箭
+箱
+箴
+箸
+節
+篁
+範
+篆
+篇
+築
+篊
+篋
+篌
+篔
+篙
+篝
+篠
+篡
+篤
+篥
+篦
+篩
+篪
+篭
+篯
+篳
+篷
+簀
+簃
+簇
+簉
+簋
+簍
+簑
+簕
+簗
+簞
+簠
+簡
+簧
+簪
+簫
+簷
+簸
+簹
+簺
+簽
+簾
+簿
+籀
+籃
+籌
+籍
+籐
+籙
+籛
+籜
+籝
+籟
+籠
+籣
+籤
+籥
+籪
+籬
+籮
+籲
+米
+籽
+籾
+粄
+粉
+粍
+粑
+粒
+粕
+粗
+粘
+粟
+粢
+粥
+粦
+粧
+粩
+粱
+粲
+粳
+粵
+粹
+粼
+粽
+精
+粿
+糀
+糅
+糊
+糌
+糍
+糎
+糕
+糖
+糙
+糜
+糝
+糞
+糟
+糠
+糢
+糧
+糬
+糯
+糰
+糴
+糶
+糸
+糹
+糺
+系
+糾
+紀
+紂
+約
+紅
+紆
+紇
+紈
+紉
+紊
+紋
+納
+紐
+紑
+紓
+純
+紕
+紗
+紘
+紙
+級
+紛
+紜
+紝
+紞
+素
+紡
+索
+紫
+紮
+累
+細
+紱
+紲
+紳
+紵
+紹
+紺
+紿
+終
+絃
+組
+絆
+経
+絎
+結
+絕
+絛
+絜
+絞
+絡
+絢
+給
+絨
+絪
+絮
+統
+絲
+絳
+絵
+絶
+絹
+絺
+綁
+綃
+綈
+綉
+綎
+綏
+經
+綖
+継
+続
+綜
+綝
+綞
+綠
+綢
+綣
+綦
+綧
+綫
+綬
+維
+綮
+綰
+綱
+網
+綳
+綴
+綸
+綺
+綻
+綽
+綾
+綿
+緁
+緃
+緄
+緈
+緊
+緋
+総
+緑
+緒
+緖
+緘
+線
+緜
+緝
+緞
+締
+緡
+緣
+緤
+編
+緩
+緬
+緯
+緱
+緲
+練
+緹
+緻
+縂
+縄
+縈
+縉
+縊
+縕
+縛
+縝
+縞
+縠
+縡
+縣
+縤
+縫
+縮
+縯
+縱
+縴
+縵
+縷
+縹
+縻
+總
+績
+繁
+繃
+繆
+繇
+繒
+織
+繕
+繖
+繙
+繚
+繞
+繡
+繩
+繪
+繫
+繭
+繰
+繳
+繹
+繻
+繼
+繽
+繾
+纁
+纂
+纈
+續
+纍
+纏
+纓
+纔
+纕
+纖
+纘
+纛
+纜
+缐
+缶
+缸
+缺
+缽
+罃
+罄
+罅
+罈
+罉
+罌
+罍
+罐
+罔
+罕
+罘
+罟
+罡
+罨
+罩
+罪
+置
+罰
+罱
+署
+罳
+罵
+罶
+罷
+罹
+罽
+羂
+羅
+羆
+羈
+羊
+羋
+羌
+美
+羔
+羕
+羗
+羙
+羚
+羞
+羡
+羣
+群
+羥
+羧
+羨
+義
+羯
+羰
+羱
+羲
+羸
+羹
+羽
+羿
+翀
+翁
+翂
+翃
+翅
+翊
+翌
+翎
+翏
+習
+翔
+翕
+翙
+翜
+翟
+翠
+翡
+翥
+翦
+翩
+翬
+翮
+翰
+翱
+翳
+翹
+翻
+翼
+耀
+老
+考
+耄
+者
+耆
+而
+耍
+耎
+耐
+耑
+耒
+耔
+耕
+耗
+耘
+耙
+耜
+耦
+耨
+耬
+耳
+耵
+耶
+耷
+耽
+耿
+聃
+聆
+聊
+聒
+聖
+聘
+聚
+聞
+聟
+聨
+聯
+聰
+聱
+聲
+聳
+聴
+聶
+職
+聽
+聾
+聿
+肄
+肅
+肆
+肇
+肉
+肋
+肌
+肏
+肖
+肘
+肚
+肛
+肜
+肝
+肟
+股
+肢
+肥
+肩
+肪
+肫
+肯
+肱
+育
+肸
+肹
+肺
+肼
+肽
+胂
+胃
+胄
+胅
+胇
+胊
+背
+胍
+胎
+胖
+胗
+胙
+胚
+胛
+胝
+胞
+胡
+胤
+胥
+胬
+胭
+胰
+胱
+胳
+胴
+胸
+胺
+胼
+能
+脂
+脅
+脆
+脇
+脈
+脊
+脒
+脖
+脘
+脛
+脣
+脩
+脫
+脬
+脭
+脯
+脲
+脳
+脷
+脹
+脾
+腆
+腈
+腊
+腋
+腌
+腎
+腐
+腑
+腓
+腔
+腕
+腥
+腦
+腧
+腩
+腫
+腮
+腰
+腱
+腳
+腴
+腸
+腹
+腺
+腿
+膀
+膂
+膈
+膊
+膏
+膚
+膛
+膜
+膝
+膠
+膣
+膥
+膦
+膨
+膩
+膮
+膳
+膺
+膽
+膾
+膿
+臀
+臂
+臃
+臆
+臉
+臊
+臍
+臏
+臘
+臚
+臞
+臟
+臠
+臣
+臧
+臨
+自
+臭
+臯
+至
+致
+臺
+臻
+臼
+臾
+舂
+舅
+與
+興
+舉
+舊
+舌
+舍
+舎
+舒
+舔
+舖
+舘
+舛
+舜
+舞
+舟
+舢
+舥
+舨
+舩
+航
+舫
+般
+舲
+舵
+舶
+舷
+舸
+船
+舺
+艅
+艇
+艉
+艋
+艎
+艏
+艔
+艘
+艙
+艚
+艦
+艮
+良
+艱
+色
+艶
+艷
+艸
+艽
+艾
+艿
+芃
+芊
+芋
+芍
+芎
+芑
+芒
+芘
+芙
+芛
+芝
+芡
+芥
+芨
+芩
+芪
+芫
+芬
+芭
+芮
+芯
+花
+芳
+芴
+芷
+芸
+芹
+芻
+芽
+芾
+苄
+苅
+苑
+苒
+苓
+苔
+苕
+苗
+苛
+苜
+苝
+苞
+苟
+苡
+苣
+苤
+若
+苦
+苧
+苪
+苫
+苯
+英
+苳
+苴
+苷
+苺
+苻
+苼
+苾
+茀
+茁
+茂
+范
+茄
+茅
+茆
+茇
+茈
+茉
+茌
+茗
+茘
+茚
+茛
+茜
+茝
+茨
+茫
+茬
+茭
+茮
+茯
+茱
+茲
+茴
+茵
+茶
+茷
+茸
+茹
+茺
+茼
+荀
+荃
+荅
+荇
+草
+荊
+荎
+荏
+荒
+荔
+荖
+荘
+荳
+荷
+荸
+荻
+荼
+荽
+莆
+莉
+莊
+莎
+莒
+莓
+莕
+莖
+莘
+莙
+莛
+莜
+莞
+莠
+莢
+莧
+莨
+莩
+莪
+莫
+莽
+莿
+菀
+菁
+菅
+菇
+菈
+菉
+菊
+菌
+菍
+菏
+菑
+菓
+菔
+菖
+菘
+菜
+菝
+菟
+菠
+菡
+菥
+菩
+菪
+菫
+華
+菰
+菱
+菲
+菴
+菶
+菸
+菹
+菺
+菼
+菽
+菾
+萁
+萃
+萄
+萇
+萊
+萌
+萍
+萎
+萐
+萘
+萜
+萠
+萡
+萣
+萩
+萬
+萭
+萱
+萵
+萸
+萹
+萼
+落
+葃
+葆
+葉
+葊
+葎
+葑
+葒
+著
+葙
+葚
+葛
+葜
+葝
+葡
+董
+葦
+葩
+葫
+葬
+葭
+葯
+葰
+葳
+葵
+葶
+葷
+葺
+蒂
+蒄
+蒍
+蒎
+蒐
+蒓
+蒔
+蒗
+蒙
+蒜
+蒞
+蒟
+蒡
+蒢
+蒤
+蒧
+蒨
+蒭
+蒯
+蒲
+蒴
+蒸
+蒹
+蒺
+蒻
+蒼
+蒽
+蒾
+蒿
+蓀
+蓁
+蓂
+蓄
+蓆
+蓉
+蓋
+蓍
+蓑
+蓓
+蓖
+蓘
+蓚
+蓧
+蓨
+蓪
+蓬
+蓭
+蓮
+蓯
+蓳
+蓼
+蓽
+蓿
+蔆
+蔎
+蔑
+蔓
+蔔
+蔕
+蔗
+蔘
+蔚
+蔝
+蔞
+蔡
+蔣
+蔥
+蔦
+蔬
+蔭
+蔴
+蔵
+蔻
+蔽
+蕁
+蕃
+蕅
+蕈
+蕉
+蕊
+蕎
+蕑
+蕒
+蕖
+蕘
+蕙
+蕚
+蕟
+蕡
+蕢
+蕤
+蕨
+蕩
+蕪
+蕭
+蕷
+蕹
+蕺
+蕻
+蕾
+薀
+薄
+薆
+薇
+薈
+薊
+薌
+薏
+薐
+薑
+薔
+薗
+薘
+薙
+薛
+薜
+薞
+薟
+薡
+薦
+薨
+薩
+薪
+薫
+薬
+薯
+薰
+薲
+薷
+薸
+薹
+薺
+薾
+薿
+藁
+藉
+藍
+藎
+藏
+藐
+藔
+藕
+藜
+藝
+藟
+藤
+藥
+藦
+藨
+藩
+藪
+藶
+藸
+藹
+藺
+藻
+藿
+蘂
+蘄
+蘅
+蘆
+蘇
+蘊
+蘋
+蘐
+蘑
+蘓
+蘗
+蘘
+蘚
+蘞
+蘢
+蘧
+蘩
+蘭
+蘵
+蘶
+蘸
+蘼
+蘿
+虉
+虎
+虐
+虓
+虔
+處
+虖
+虛
+虜
+虞
+號
+虢
+虧
+虨
+虯
+虱
+虵
+虹
+虺
+虻
+蚆
+蚊
+蚋
+蚌
+蚍
+蚓
+蚖
+蚜
+蚝
+蚡
+蚢
+蚣
+蚤
+蚧
+蚨
+蚩
+蚪
+蚯
+蚱
+蚴
+蚵
+蚶
+蚺
+蚼
+蛀
+蛄
+蛇
+蛉
+蛋
+蛍
+蛐
+蛑
+蛔
+蛙
+蛛
+蛞
+蛟
+蛤
+蛭
+蛯
+蛸
+蛹
+蛺
+蛻
+蛾
+蜀
+蜂
+蜃
+蜆
+蜇
+蜈
+蜉
+蜊
+蜍
+蜑
+蜒
+蜓
+蜘
+蜚
+蜛
+蜜
+蜞
+蜢
+蜣
+蜥
+蜨
+蜮
+蜯
+蜱
+蜴
+蜷
+蜻
+蜾
+蜿
+蝀
+蝌
+蝍
+蝎
+蝓
+蝕
+蝗
+蝘
+蝙
+蝚
+蝟
+蝠
+蝣
+蝤
+蝦
+蝨
+蝮
+蝯
+蝰
+蝲
+蝴
+蝶
+蝸
+蝽
+螂
+螃
+螄
+螅
+螈
+螋
+融
+螐
+螔
+螞
+螟
+螠
+螢
+螣
+螥
+螫
+螭
+螯
+螳
+螶
+螺
+螻
+螽
+螾
+蟀
+蟄
+蟅
+蟆
+蟊
+蟋
+蟌
+蟎
+蟑
+蟒
+蟜
+蟠
+蟥
+蟪
+蟫
+蟬
+蟯
+蟲
+蟳
+蟴
+蟶
+蟹
+蟻
+蟾
+蠂
+蠃
+蠄
+蠅
+蠆
+蠊
+蠋
+蠍
+蠐
+蠑
+蠓
+蠔
+蠕
+蠖
+蠘
+蠙
+蠟
+蠡
+蠢
+蠣
+蠱
+蠲
+蠵
+蠶
+蠷
+蠹
+蠻
+血
+衂
+衆
+行
+衍
+衎
+術
+衕
+衖
+街
+衙
+衚
+衛
+衜
+衝
+衞
+衡
+衢
+衣
+表
+衩
+衫
+衰
+衲
+衷
+衽
+衾
+衿
+袁
+袂
+袈
+袋
+袍
+袓
+袖
+袛
+袞
+袤
+袪
+被
+袱
+袴
+袾
+裁
+裂
+裊
+裎
+裒
+裔
+裕
+裖
+裘
+裙
+補
+裝
+裟
+裡
+裨
+裬
+裱
+裳
+裴
+裵
+裸
+裹
+製
+裾
+裿
+褀
+褂
+複
+褌
+褍
+褎
+褐
+褒
+褓
+褔
+褘
+褙
+褚
+褞
+褥
+褧
+褪
+褫
+褭
+褲
+褶
+褸
+褻
+襄
+襌
+襖
+襞
+襟
+襠
+襤
+襦
+襪
+襯
+襲
+襴
+襶
+襻
+襾
+西
+要
+覃
+覆
+覇
+覈
+見
+覌
+規
+覓
+視
+覚
+覡
+覦
+覧
+親
+覬
+覲
+観
+覺
+覽
+覿
+觀
+角
+觔
+觙
+觚
+觜
+解
+觭
+觱
+觴
+觶
+觸
+觿
+言
+訁
+訂
+訃
+訇
+計
+訊
+訌
+討
+訏
+訐
+訒
+訓
+訔
+訕
+訖
+託
+記
+訛
+訝
+訟
+訣
+訥
+訪
+設
+許
+訴
+訶
+診
+註
+証
+訾
+詁
+詆
+詈
+詐
+詒
+詔
+評
+詛
+詞
+詠
+詡
+詢
+詣
+詥
+試
+詧
+詩
+詫
+詭
+詮
+詰
+話
+該
+詳
+詵
+詹
+詼
+誄
+誅
+誇
+誌
+認
+誒
+誓
+誕
+誘
+語
+誠
+誡
+誣
+誤
+誥
+誦
+誨
+說
+説
+読
+誰
+課
+誴
+誹
+誼
+誾
+調
+談
+請
+諍
+諏
+諒
+論
+諗
+諜
+諟
+諠
+諡
+諤
+諦
+諧
+諪
+諫
+諭
+諮
+諱
+諲
+諳
+諴
+諶
+諷
+諸
+諺
+諼
+諾
+謀
+謁
+謂
+謄
+謇
+謊
+謌
+謎
+謏
+謐
+謔
+謖
+謗
+謙
+謚
+講
+謜
+謝
+謠
+謢
+謤
+謨
+謩
+謫
+謬
+謳
+謹
+謾
+證
+譏
+譓
+譔
+識
+譙
+譚
+譜
+譞
+警
+譫
+譬
+譭
+譯
+議
+譲
+譳
+譴
+護
+譽
+譿
+讀
+讃
+變
+讌
+讎
+讓
+讖
+讙
+讚
+讜
+讞
+谷
+谿
+豁
+豆
+豇
+豈
+豉
+豊
+豌
+豎
+豐
+豔
+豕
+豚
+象
+豢
+豨
+豪
+豫
+豬
+豳
+豸
+豹
+豺
+豿
+貂
+貅
+貉
+貊
+貌
+貐
+貒
+貓
+貔
+貘
+貝
+貞
+負
+財
+貢
+貤
+貧
+貨
+販
+貪
+貫
+責
+貭
+貮
+貯
+貲
+貳
+貴
+貶
+買
+貸
+貺
+費
+貼
+貽
+貿
+賀
+賁
+賂
+賃
+賄
+資
+賈
+賊
+賑
+賒
+賓
+賔
+賕
+賚
+賜
+賞
+賠
+賡
+賢
+賣
+賤
+賦
+賨
+質
+賬
+賭
+賴
+賹
+賺
+賻
+購
+賽
+賾
+贄
+贅
+贇
+贈
+贊
+贌
+贍
+贏
+贓
+贔
+贖
+贛
+赤
+赦
+赧
+赫
+赬
+赭
+走
+赳
+赴
+起
+趁
+超
+越
+趐
+趕
+趖
+趙
+趟
+趣
+趨
+足
+趴
+趵
+趺
+趼
+趾
+跅
+跆
+跋
+跌
+跏
+跑
+跖
+跗
+跛
+距
+跟
+跡
+跣
+跤
+跨
+跩
+跪
+路
+跳
+踎
+踏
+踐
+踝
+踞
+踢
+踩
+踰
+踴
+踹
+踺
+蹂
+蹄
+蹇
+蹈
+蹉
+蹊
+蹋
+蹕
+蹙
+蹟
+蹠
+蹤
+蹦
+蹬
+蹭
+蹯
+蹲
+蹴
+蹶
+蹺
+蹻
+蹼
+躁
+躂
+躄
+躉
+躋
+躍
+躑
+躒
+躔
+躝
+躪
+身
+躬
+躰
+躲
+躺
+軀
+車
+軋
+軌
+軍
+軎
+軒
+軔
+軛
+軟
+転
+軫
+軲
+軸
+軹
+軺
+軻
+軼
+軽
+軾
+較
+輄
+輅
+載
+輋
+輒
+輓
+輔
+輕
+輛
+輝
+輞
+輟
+輥
+輦
+輩
+輪
+輬
+輭
+輯
+輶
+輸
+輻
+輾
+輿
+轀
+轂
+轄
+轅
+轆
+轉
+轍
+轎
+轘
+轝
+轟
+轤
+辛
+辜
+辟
+辣
+辦
+辧
+辨
+辭
+辮
+辯
+辰
+辱
+農
+辵
+辺
+辻
+込
+迂
+迄
+迅
+迎
+近
+返
+迢
+迤
+迥
+迦
+迪
+迫
+迭
+迮
+述
+迴
+迵
+迷
+迸
+迺
+追
+退
+送
+逃
+逄
+逅
+逆
+逈
+逋
+逌
+逍
+逎
+透
+逐
+逑
+途
+逕
+逖
+逗
+這
+通
+逛
+逝
+逞
+速
+造
+逢
+連
+逤
+逨
+逮
+逯
+進
+逴
+逵
+逸
+逹
+逺
+逼
+逾
+遁
+遂
+遄
+遇
+遊
+運
+遍
+過
+遏
+遐
+遒
+道
+達
+違
+遘
+遙
+遛
+遜
+遞
+遠
+遢
+遣
+遨
+適
+遭
+遮
+遯
+遲
+遴
+遵
+遶
+遷
+選
+遹
+遺
+遼
+避
+邀
+邁
+邂
+邃
+還
+邇
+邈
+邉
+邊
+邋
+邏
+邑
+邕
+邗
+邙
+邛
+邠
+邡
+邢
+那
+邦
+邨
+邪
+邯
+邰
+邱
+邲
+邳
+邴
+邵
+邸
+邽
+邾
+郁
+郃
+郄
+郅
+郇
+郊
+郋
+郎
+郗
+郛
+郜
+郝
+郞
+郟
+郡
+郢
+郤
+部
+郪
+郫
+郭
+郯
+郳
+郴
+郵
+郷
+都
+郾
+郿
+鄂
+鄃
+鄄
+鄆
+鄉
+鄋
+鄑
+鄒
+鄔
+鄖
+鄗
+鄘
+鄙
+鄚
+鄜
+鄞
+鄠
+鄢
+鄣
+鄤
+鄧
+鄩
+鄫
+鄭
+鄯
+鄰
+鄱
+鄲
+鄳
+鄴
+鄺
+酃
+酆
+酈
+酉
+酊
+酋
+酌
+配
+酎
+酏
+酐
+酒
+酔
+酗
+酚
+酞
+酡
+酢
+酣
+酥
+酩
+酪
+酬
+酮
+酯
+酰
+酴
+酵
+酶
+酷
+酸
+酺
+酼
+醁
+醂
+醃
+醅
+醇
+醉
+醋
+醌
+醍
+醐
+醒
+醚
+醛
+醜
+醞
+醢
+醣
+醪
+醫
+醬
+醮
+醯
+醴
+醺
+醾
+醿
+釀
+釁
+釆
+采
+釉
+釋
+里
+重
+野
+量
+釐
+金
+釒
+釓
+釔
+釕
+釗
+釘
+釙
+釚
+釜
+針
+釣
+釤
+釦
+釧
+釩
+釪
+釭
+釴
+釵
+釷
+釹
+釺
+鈀
+鈁
+鈄
+鈇
+鈈
+鈉
+鈊
+鈍
+鈏
+鈐
+鈑
+鈔
+鈕
+鈖
+鈞
+鈢
+鈣
+鈥
+鈦
+鈫
+鈮
+鈰
+鈳
+鈴
+鈷
+鈸
+鈹
+鈺
+鈾
+鈿
+鉀
+鉄
+鉅
+鉆
+鉈
+鉉
+鉋
+鉌
+鉍
+鉏
+鉑
+鉓
+鉗
+鉚
+鉛
+鉞
+鉟
+鉤
+鉦
+鉬
+鉭
+鉲
+鉶
+鉷
+鉸
+鉻
+鉾
+鉿
+銀
+銂
+銃
+銅
+銋
+銍
+銑
+銓
+銕
+銖
+銘
+銚
+銜
+銠
+銣
+銥
+銦
+銨
+銩
+銪
+銫
+銬
+銭
+銱
+銲
+銳
+銶
+銷
+銹
+銻
+銼
+銾
+鋁
+鋅
+鋆
+鋇
+鋌
+鋏
+鋐
+鋒
+鋕
+鋗
+鋙
+鋡
+鋤
+鋥
+鋦
+鋨
+鋪
+鋮
+鋯
+鋰
+鋱
+鋳
+鋶
+鋸
+鋹
+鋼
+錀
+錄
+錏
+錐
+錒
+錕
+錘
+錚
+錞
+錟
+錠
+錡
+錢
+錦
+錨
+錫
+錬
+錮
+錯
+錳
+錶
+錸
+錻
+鍀
+鍇
+鍈
+鍉
+鍊
+鍋
+鍍
+鍏
+鍔
+鍘
+鍛
+鍝
+鍟
+鍠
+鍥
+鍩
+鍬
+鍱
+鍳
+鍵
+鍶
+鍷
+鍺
+鍼
+鍾
+鎂
+鎅
+鎊
+鎌
+鎏
+鎓
+鎔
+鎖
+鎗
+鎘
+鎚
+鎛
+鎢
+鎣
+鎦
+鎧
+鎪
+鎬
+鎭
+鎮
+鎰
+鎳
+鎵
+鎻
+鏃
+鏇
+鏈
+鏊
+鏌
+鏐
+鏑
+鏓
+鏖
+鏗
+鏘
+鏜
+鏝
+鏞
+鏟
+鏡
+鏢
+鏤
+鏦
+鏳
+鏴
+鏵
+鏷
+鏻
+鏽
+鐃
+鐇
+鐈
+鐓
+鐔
+鐘
+鐙
+鐠
+鐡
+鐤
+鐦
+鐧
+鐫
+鐬
+鐭
+鐮
+鐲
+鐳
+鐵
+鐸
+鐺
+鐽
+鐿
+鑀
+鑁
+鑂
+鑄
+鑅
+鑊
+鑌
+鑑
+鑒
+鑛
+鑠
+鑣
+鑨
+鑪
+鑫
+鑭
+鑰
+鑲
+鑴
+鑷
+鑼
+鑽
+鑾
+鑿
+長
+門
+閂
+閃
+閆
+閉
+開
+閎
+閏
+閑
+閒
+間
+閔
+閘
+閜
+閞
+閟
+関
+閣
+閥
+閦
+閨
+閩
+閬
+閭
+閰
+閱
+閶
+閹
+閻
+閼
+閾
+閿
+闆
+闇
+闈
+闊
+闋
+闌
+闍
+闐
+闓
+闔
+闕
+闖
+闘
+關
+闞
+闡
+闢
+闥
+阜
+阝
+阡
+阪
+阭
+阮
+阯
+阱
+防
+阻
+阿
+陀
+陁
+陂
+附
+陋
+陌
+降
+限
+陔
+陘
+陛
+陜
+陝
+陞
+陟
+陡
+院
+陣
+除
+陪
+陬
+陰
+陲
+陳
+陵
+陶
+陷
+陸
+険
+陽
+隄
+隅
+隆
+隈
+隊
+隋
+隍
+階
+隔
+隕
+隗
+隘
+隙
+際
+障
+隣
+隧
+隨
+險
+隰
+隱
+隲
+隳
+隴
+隷
+隸
+隹
+隻
+隼
+雀
+雁
+雄
+雅
+集
+雇
+雉
+雋
+雌
+雍
+雎
+雑
+雒
+雕
+雖
+雙
+雛
+雜
+雝
+雞
+離
+難
+雨
+雩
+雪
+雫
+雯
+雱
+雲
+零
+雷
+雹
+電
+需
+霄
+霅
+霆
+震
+霈
+霉
+霊
+霍
+霎
+霏
+霑
+霓
+霖
+霙
+霜
+霞
+霤
+霧
+霨
+霰
+露
+霶
+霸
+霹
+霽
+霾
+靁
+靂
+靄
+靈
+靉
+靑
+青
+靖
+靚
+靛
+靜
+非
+靠
+靡
+面
+革
+靫
+靬
+靭
+靳
+靴
+靶
+靺
+靼
+鞅
+鞆
+鞋
+鞍
+鞏
+鞘
+鞞
+鞠
+鞣
+鞥
+鞦
+鞨
+鞭
+鞮
+鞴
+韁
+韃
+韆
+韋
+韌
+韑
+韓
+韙
+韜
+韞
+韠
+韡
+韭
+韮
+音
+韶
+韺
+韻
+韾
+響
+頁
+頂
+頃
+項
+順
+須
+頊
+頌
+頍
+頎
+頏
+預
+頑
+頒
+頓
+頔
+頗
+領
+頜
+頠
+頡
+頤
+頦
+頫
+頭
+頰
+頴
+頵
+頷
+頸
+頹
+頻
+頼
+顆
+題
+額
+顎
+顏
+顒
+顓
+顔
+顕
+顗
+願
+顙
+顛
+類
+顥
+顧
+顫
+顯
+顰
+顱
+顳
+顴
+風
+颮
+颯
+颱
+颶
+颺
+颼
+飄
+飆
+飈
+飛
+食
+飠
+飡
+飢
+飥
+飩
+飪
+飫
+飬
+飭
+飮
+飯
+飲
+飴
+飼
+飽
+飾
+餃
+餄
+餅
+餉
+養
+餌
+餎
+餐
+餒
+餓
+餗
+餘
+餚
+餛
+餞
+餠
+餡
+館
+餮
+餵
+餺
+餾
+餿
+饃
+饅
+饋
+饌
+饑
+饒
+饕
+饗
+饞
+饟
+饢
+首
+馗
+馘
+香
+馛
+馥
+馦
+馨
+馬
+馭
+馮
+馯
+馱
+馳
+馴
+馼
+駁
+駄
+駅
+駆
+駐
+駑
+駒
+駔
+駕
+駘
+駙
+駛
+駝
+駟
+駢
+駭
+駰
+駱
+駿
+騁
+騂
+騄
+騅
+騋
+騎
+騏
+験
+騖
+騙
+騤
+騨
+騫
+騭
+騮
+騰
+騶
+騷
+騾
+驁
+驃
+驄
+驅
+驊
+驌
+驍
+驎
+驒
+驕
+驗
+驚
+驛
+驟
+驢
+驤
+驥
+驩
+驪
+骨
+骯
+骰
+骶
+骷
+骸
+骼
+髀
+髂
+髎
+髏
+髑
+髒
+髓
+體
+高
+髙
+髡
+髦
+髪
+髭
+髮
+髯
+髲
+髷
+髹
+髻
+鬃
+鬄
+鬅
+鬆
+鬍
+鬚
+鬟
+鬢
+鬣
+鬥
+鬧
+鬨
+鬩
+鬪
+鬬
+鬮
+鬯
+鬱
+鬲
+鬹
+鬻
+鬼
+魁
+魂
+魃
+魄
+魅
+魈
+魋
+魍
+魎
+魏
+魔
+魕
+魘
+魚
+魛
+魞
+魟
+魣
+魨
+魩
+魮
+魯
+魴
+魷
+鮀
+鮁
+鮃
+鮄
+鮊
+鮋
+鮍
+鮐
+鮑
+鮒
+鮓
+鮗
+鮜
+鮟
+鮠
+鮡
+鮣
+鮨
+鮪
+鮫
+鮭
+鮮
+鮰
+鮸
+鮹
+鮻
+鯀
+鯁
+鯃
+鯇
+鯉
+鯊
+鯏
+鯒
+鯓
+鯔
+鯕
+鯖
+鯗
+鯙
+鯛
+鯡
+鯢
+鯤
+鯧
+鯨
+鯪
+鯭
+鯮
+鯰
+鯶
+鯷
+鯻
+鯽
+鯿
+鰂
+鰃
+鰆
+鰈
+鰉
+鰍
+鰏
+鰒
+鰓
+鰕
+鰗
+鰛
+鰜
+鰟
+鰣
+鰤
+鰧
+鰨
+鰩
+鰭
+鰮
+鰱
+鰲
+鰳
+鰶
+鰷
+鰹
+鰺
+鰻
+鰼
+鰾
+鱀
+鱂
+鱅
+鱇
+鱈
+鱉
+鱊
+鱒
+鱓
+鱔
+鱖
+鱗
+鱘
+鱚
+鱝
+鱟
+鱠
+鱣
+鱥
+鱧
+鱨
+鱬
+鱮
+鱰
+鱲
+鱵
+鱷
+鱸
+鱺
+鱻
+鳥
+鳧
+鳩
+鳯
+鳰
+鳳
+鳴
+鳶
+鳽
+鴆
+鴇
+鴉
+鴒
+鴓
+鴕
+鴗
+鴛
+鴝
+鴞
+鴟
+鴡
+鴣
+鴦
+鴨
+鴫
+鴯
+鴰
+鴴
+鴻
+鴿
+鵂
+鵄
+鵎
+鵐
+鵑
+鵒
+鵓
+鵙
+鵜
+鵝
+鵞
+鵟
+鵠
+鵡
+鵪
+鵬
+鵯
+鵰
+鵲
+鵵
+鵼
+鵾
+鶆
+鶇
+鶉
+鶏
+鶒
+鶓
+鶘
+鶚
+鶡
+鶥
+鶩
+鶬
+鶯
+鶲
+鶴
+鶹
+鶺
+鶻
+鶼
+鶿
+鷂
+鷄
+鷉
+鷎
+鷓
+鷗
+鷙
+鷚
+鷟
+鷥
+鷦
+鷫
+鷯
+鷲
+鷳
+鷸
+鷹
+鷺
+鸊
+鸌
+鸐
+鸑
+鸕
+鸘
+鸚
+鸛
+鸜
+鸝
+鸞
+鹮
+鹵
+鹹
+鹼
+鹽
+鹿
+麂
+麅
+麇
+麈
+麊
+麋
+麐
+麒
+麓
+麗
+麝
+麞
+麟
+麥
+麩
+麪
+麯
+麴
+麵
+麹
+麺
+麻
+麼
+麽
+麾
+麿
+黁
+黃
+黇
+黌
+黍
+黎
+黏
+黐
+黑
+黒
+黔
+默
+黙
+黛
+黜
+黝
+點
+黟
+黥
+黧
+黨
+黯
+黴
+黶
+黻
+黼
+黽
+黿
+鼂
+鼇
+鼈
+鼉
+鼎
+鼐
+鼒
+鼓
+鼕
+鼙
+鼠
+鼢
+鼩
+鼬
+鼯
+鼱
+鼴
+鼷
+鼻
+鼽
+鼾
+齊
+齋
+齒
+齕
+齡
+齣
+齦
+齧
+齲
+齶
+龍
+龎
+龐
+龑
+龔
+龕
+龜
+龝
+龠
+龢
+郎
+凉
+﹑
+﹗
+﹝
+﹞
+﹢
+！
+＂
+＃
+＄
+％
+＆
+＇
+（
+）
+＊
+＋
+，
+－
+．
+／
+０
+１
+２
+３
+４
+５
+６
+７
+８
+９
+：
+；
+＜
+＝
+＞
+？
+Ａ
+Ｂ
+Ｃ
+Ｄ
+Ｅ
+Ｆ
+Ｇ
+Ｈ
+Ｉ
+Ｋ
+Ｌ
+Ｍ
+Ｎ
+Ｏ
+Ｐ
+Ｒ
+Ｓ
+Ｔ
+Ｕ
+Ｖ
+Ｗ
+Ｙ
+Ｚ
+［
+］
+｀
+ａ
+ｂ
+ｃ
+ｄ
+ｅ
+ｆ
+ｇ
+ｈ
+ｉ
+ｊ
+ｋ
+ｌ
+ｍ
+ｎ
+ｏ
+ｐ
+ｒ
+ｓ
+ｔ
+ｕ
+ｚ
+｛
+｜
+｝
+～
+￥
+𣇉
+ 
diff --git a/ppocr/utils/dict/confuse.pkl b/ppocr/utils/dict/confuse.pkl
new file mode 100644
index 0000000..e5d4853
Binary files /dev/null and b/ppocr/utils/dict/confuse.pkl differ
diff --git a/ppocr/utils/dict/cyrillic_dict.txt b/ppocr/utils/dict/cyrillic_dict.txt
new file mode 100644
index 0000000..2b6f664
--- /dev/null
+++ b/ppocr/utils/dict/cyrillic_dict.txt
@@ -0,0 +1,163 @@
+ 
+!
+#
+$
+%
+&
+'
+(
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+?
+@
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+_
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+É
+é
+Ё
+Є
+І
+Ј
+Љ
+Ў
+А
+Б
+В
+Г
+Д
+Е
+Ж
+З
+И
+Й
+К
+Л
+М
+Н
+О
+П
+Р
+С
+Т
+У
+Ф
+Х
+Ц
+Ч
+Ш
+Щ
+Ъ
+Ы
+Ь
+Э
+Ю
+Я
+а
+б
+в
+г
+д
+е
+ж
+з
+и
+й
+к
+л
+м
+н
+о
+п
+р
+с
+т
+у
+ф
+х
+ц
+ч
+ш
+щ
+ъ
+ы
+ь
+э
+ю
+я
+ё
+ђ
+є
+і
+ј
+љ
+њ
+ћ
+ў
+џ
+Ґ
+ґ
diff --git a/ppocr/utils/dict/devanagari_dict.txt b/ppocr/utils/dict/devanagari_dict.txt
new file mode 100644
index 0000000..f559230
--- /dev/null
+++ b/ppocr/utils/dict/devanagari_dict.txt
@@ -0,0 +1,167 @@
+ 
+!
+#
+$
+%
+&
+'
+(
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+?
+@
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+_
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+É
+é
+ँ
+ं
+ः
+अ
+आ
+इ
+ई
+उ
+ऊ
+ऋ
+ए
+ऐ
+ऑ
+ओ
+औ
+क
+ख
+ग
+घ
+ङ
+च
+छ
+ज
+झ
+ञ
+ट
+ठ
+ड
+ढ
+ण
+त
+थ
+द
+ध
+न
+ऩ
+प
+फ
+ब
+भ
+म
+य
+र
+ऱ
+ल
+ळ
+व
+श
+ष
+स
+ह
+़
+ा
+ि
+ी
+ु
+ू
+ृ
+ॅ
+े
+ै
+ॉ
+ो
+ौ
+्
+॒
+क़
+ख़
+ग़
+ज़
+ड़
+ढ़
+फ़
+ॠ
+।
+०
+१
+२
+३
+४
+५
+६
+७
+८
+९
+॰
diff --git a/ppocr/utils/dict/en_dict.txt b/ppocr/utils/dict/en_dict.txt
new file mode 100644
index 0000000..6fbd99f
--- /dev/null
+++ b/ppocr/utils/dict/en_dict.txt
@@ -0,0 +1,63 @@
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+ 
diff --git a/ppocr/utils/dict/fa_dict.txt b/ppocr/utils/dict/fa_dict.txt
new file mode 100644
index 0000000..2328fbd
--- /dev/null
+++ b/ppocr/utils/dict/fa_dict.txt
@@ -0,0 +1,136 @@
+f
+a
+_
+i
+m
+g
+/
+1
+3
+I
+L
+S
+V
+R
+C
+2
+0
+v
+l
+6
+8
+5
+.
+j
+p	
+و
+د
+ر
+ك
+ن
+ش
+ه
+ا
+4
+9
+ی
+ج
+ِ
+7
+غ
+ل
+س
+ز
+ّ
+ت
+ک
+گ
+ي
+م
+ب
+ف
+چ
+خ
+ق
+ژ
+آ
+ص
+پ
+َ
+ع
+ئ
+ح
+ٔ
+ض
+ُ
+ذ
+أ
+ى
+ط
+ظ
+ث
+ة
+ً
+ء
+ؤ
+ْ
+ۀ
+إ
+ٍ
+ٌ
+ٰ
+ٓ
+ٱ
+s
+c
+e
+n
+w
+N
+E
+W
+Y
+D
+O
+H
+A
+d
+z
+r
+T
+G
+o
+t
+x
+h
+b
+B
+M
+Z
+u
+P
+F
+y
+q
+U
+K
+k
+J
+Q
+'
+X
+#
+?
+%
+$
+,
+:
+&
+!
+-
+(
+É
+@
+é
++
+ 
diff --git a/ppocr/utils/dict/french_dict.txt b/ppocr/utils/dict/french_dict.txt
new file mode 100644
index 0000000..e8f657d
--- /dev/null
+++ b/ppocr/utils/dict/french_dict.txt
@@ -0,0 +1,136 @@
+f
+e
+n
+c
+h
+_
+i
+m
+g
+/
+r
+v
+a
+l
+t
+w
+o
+d
+6
+1
+.
+p
+B
+u
+2
+à
+3
+R
+y
+4
+U
+E
+A
+5
+P
+O
+S
+T
+D
+7
+Z
+8
+I
+N
+L
+G
+M
+H
+0
+J
+K
+-
+9
+F
+C
+V
+é
+X
+'
+s
+Q
+:
+è
+x
+b
+Y
+Œ
+É
+z
+W
+Ç
+È
+k
+Ô
+ô
+€
+À
+Ê
+q
+ù
+°
+ê
+î
+*
+Â
+j
+"
+,
+â
+%
+û
+ç
+ü
+?
+!
+;
+ö
+(
+)
+ï
+º
+ó
+ø
+å
++
+™
+á
+Ë
+<
+²
+Á
+Î
+&
+@
+œ
+ε
+Ü
+ë
+[
+]
+í
+ò
+Ö
+ä
+ß
+«
+»
+ú
+ñ
+æ
+µ
+³
+Å
+$
+#
+ 
diff --git a/ppocr/utils/dict/german_dict.txt b/ppocr/utils/dict/german_dict.txt
new file mode 100644
index 0000000..5e121af
--- /dev/null
+++ b/ppocr/utils/dict/german_dict.txt
@@ -0,0 +1,143 @@
+ 
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+=
+>
+?
+@
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+[
+]
+_
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p	
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+£
+§
+­
+°
+´
+µ
+·
+º
+¿
+Á
+Ä
+Å
+É
+Ï
+Ô
+Ö
+Ü
+ß
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+í
+ï
+ñ
+ò
+ó
+ô
+ö
+ø
+ù
+ú
+û
+ü
+ō
+Š
+Ÿ
+ʒ
+β
+δ
+з
+Ṡ
+‘
+€
+©
+ª
+«
+¬
diff --git a/ppocr/utils/dict/gujarati_dict.txt b/ppocr/utils/dict/gujarati_dict.txt
new file mode 100644
index 0000000..36f5589
--- /dev/null
+++ b/ppocr/utils/dict/gujarati_dict.txt
@@ -0,0 +1,48 @@
+અ
+આ
+ઇ
+ઈ
+ઉ
+ઊ
+ઋ
+ઌ
+એ
+ઐ
+ઓ
+ઔ
+અં
+અઃ
+ક
+ખ
+ગ
+ઘ
+ઙ
+ચ
+છ
+જ
+ઝ
+ઞ
+ટ
+ઠ
+ડ
+ઢ
+ણ
+ત
+થ
+દ
+ધ
+ન
+પ
+ફ
+બ
+ભ
+મ
+ય
+ર
+લ
+ળ
+વ
+શ
+ષ
+સ
+હ
diff --git a/ppocr/utils/dict/hebrew_dict.txt b/ppocr/utils/dict/hebrew_dict.txt
new file mode 100644
index 0000000..ed30137
--- /dev/null
+++ b/ppocr/utils/dict/hebrew_dict.txt
@@ -0,0 +1,214 @@
+!
+#
+$
+%
+&
+'
+(
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+?
+@
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+_
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+É
+é
+֑
+֒
+֓
+֔
+֕
+֖
+֗
+֘
+֙
+֚
+֛
+֜
+֝
+֞
+֟
+֠
+֡
+֢
+֣
+֤
+֥
+֦
+֧
+֨
+֩
+֪
+֫
+֬
+֭
+֮
+֯
+ְ
+ֱ
+ֲ
+ֳ
+ִ
+ֵ
+ֶ
+ַ
+ָ
+ֹ
+ֺ
+ֻ
+ּ
+ֽ
+־
+ֿ
+׀
+ׁ
+ׂ
+׃
+ׄ
+ׅ
+׆
+ׇ
+א
+ב
+ג
+ד
+ה
+ו
+ז
+ח
+ט
+י
+ך
+כ
+ל
+ם
+מ
+ן
+נ
+ס
+ע
+ף
+פ
+ץ
+צ
+ק
+ר
+ש
+ת
+ׯ
+װ
+ױ
+ײ
+׳
+״
+יִ
+ﬞ
+ײַ
+ﬠ
+ﬡ
+ﬢ
+ﬣ
+ﬤ
+ﬥ
+ﬦ
+ﬧ
+ﬨ
+﬩
+שׁ
+שׂ
+שּׁ
+שּׂ
+אַ
+אָ
+אּ
+בּ
+גּ
+דּ
+הּ
+וּ
+זּ
+טּ
+יּ
+ךּ
+כּ
+לּ
+מּ
+נּ
+סּ
+ףּ
+פּ
+צּ
+קּ
+רּ
+שּ
+תּ
+וֹ
+בֿ
+כֿ
+פֿ
+ﭏ
diff --git a/ppocr/utils/dict/hi_dict.txt b/ppocr/utils/dict/hi_dict.txt
new file mode 100644
index 0000000..8dfedb5
--- /dev/null
+++ b/ppocr/utils/dict/hi_dict.txt
@@ -0,0 +1,162 @@
+ 
+!
+#
+$
+%
+&
+'
+(
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+?
+@
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+_
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+É
+é
+ँ
+ं
+ः
+अ
+आ
+इ
+ई
+उ
+ऊ
+ऋ
+ए
+ऐ
+ऑ
+ओ
+औ
+क
+ख
+ग
+घ
+ङ
+च
+छ
+ज
+झ
+ञ
+ट
+ठ
+ड
+ढ
+ण
+त
+थ
+द
+ध
+न
+प
+फ
+ब
+भ
+म
+य
+र
+ल
+ळ
+व
+श
+ष
+स
+ह
+़
+ा
+ि
+ी
+ु
+ू
+ृ
+ॅ
+े
+ै
+ॉ
+ो
+ौ
+्
+क़
+ख़
+ग़
+ज़
+ड़
+ढ़
+फ़
+०
+१
+२
+३
+४
+५
+६
+७
+८
+९
+॰
diff --git a/ppocr/utils/dict/it_dict.txt b/ppocr/utils/dict/it_dict.txt
new file mode 100644
index 0000000..e692c6d
--- /dev/null
+++ b/ppocr/utils/dict/it_dict.txt
@@ -0,0 +1,118 @@
+i
+t
+_
+m
+g
+/
+5
+I
+L
+S
+V
+R
+C
+2
+0
+1
+v
+a
+l
+7
+8
+9
+6
+.
+j
+p
+	
+e
+r
+o
+d
+s
+n
+3
+4
+P
+u
+c
+A
+-
+,
+"
+z
+h
+f
+b
+q
+ì
+'
+à
+O
+è
+G
+ù
+é
+ò
+;
+F
+E
+B
+N
+H
+k
+:
+U
+T
+X
+D
+K
+?
+[
+M
+­
+x
+y
+(
+)
+W
+ö
+º
+w
+]
+Q
+J
++
+ü
+!
+È
+á
+%
+=
+»
+ñ
+Ö
+Y
+ä
+í
+Z
+«
+@
+ó
+ø
+ï
+ú
+ê
+ç
+Á
+É
+Å
+ß
+{
+}
+&
+`
+û
+î
+#
+$
diff --git a/ppocr/utils/dict/japan_dict.txt b/ppocr/utils/dict/japan_dict.txt
new file mode 100644
index 0000000..339d4b8
--- /dev/null
+++ b/ppocr/utils/dict/japan_dict.txt
@@ -0,0 +1,4399 @@
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+<
+=
+>
+?
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+[
+]
+_
+`
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+©
+°
+²
+´
+½
+Á
+Ä
+Å
+Ç
+È
+É
+Í
+Ó
+Ö
+×
+Ü
+ß
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+í
+ð
+ñ
+ò
+ó
+ô
+õ
+ö
+ø
+ú
+û
+ü
+ý
+ā
+ă
+ą
+ć
+Č
+č
+đ
+ē
+ė
+ę
+ğ
+ī
+ı
+Ł
+ł
+ń
+ň
+ō
+ř
+Ş
+ş
+Š
+š
+ţ
+ū
+ż
+Ž
+ž
+Ș
+ș
+ț
+Δ
+α
+λ
+μ
+φ
+Г
+О
+а
+в
+л
+о
+р
+с
+т
+я
+ồ
+​
+—
+―
+’
+“
+”
+…
+℃
+→
+∇
+−
+■
+☆
+　
+、
+。
+々
+〆
+〈
+〉
+「
+」
+『
+』
+〔
+〕
+〜
+ぁ
+あ
+ぃ
+い
+う
+ぇ
+え
+ぉ
+お
+か
+が
+き
+ぎ
+く
+ぐ
+け
+げ
+こ
+ご
+さ
+ざ
+し
+じ
+す
+ず
+せ
+ぜ
+そ
+ぞ
+た
+だ
+ち
+ぢ
+っ
+つ
+づ
+て
+で
+と
+ど
+な
+に
+ぬ
+ね
+の
+は
+ば
+ぱ
+ひ
+び
+ぴ
+ふ
+ぶ
+ぷ
+へ
+べ
+ぺ
+ほ
+ぼ
+ぽ
+ま
+み
+む
+め
+も
+ゃ
+や
+ゅ
+ゆ
+ょ
+よ
+ら
+り
+る
+れ
+ろ
+わ
+ゑ
+を
+ん
+ゝ
+ゞ
+ァ
+ア
+ィ
+イ
+ゥ
+ウ
+ェ
+エ
+ォ
+オ
+カ
+ガ
+キ
+ギ
+ク
+グ
+ケ
+ゲ
+コ
+ゴ
+サ
+ザ
+シ
+ジ
+ス
+ズ
+セ
+ゼ
+ソ
+ゾ
+タ
+ダ
+チ
+ヂ
+ッ
+ツ
+ヅ
+テ
+デ
+ト
+ド
+ナ
+ニ
+ヌ
+ネ
+ノ
+ハ
+バ
+パ
+ヒ
+ビ
+ピ
+フ
+ブ
+プ
+ヘ
+ベ
+ペ
+ホ
+ボ
+ポ
+マ
+ミ
+ム
+メ
+モ
+ャ
+ヤ
+ュ
+ユ
+ョ
+ヨ
+ラ
+リ
+ル
+レ
+ロ
+ワ
+ヰ
+ン
+ヴ
+ヵ
+ヶ
+・
+ー
+㈱
+一
+丁
+七
+万
+丈
+三
+上
+下
+不
+与
+丑
+且
+世
+丘
+丙
+丞
+両
+並
+中
+串
+丸
+丹
+主
+丼
+丿
+乃
+久
+之
+乎
+乏
+乗
+乘
+乙
+九
+乞
+也
+乱
+乳
+乾
+亀
+了
+予
+争
+事
+二
+于
+互
+五
+井
+亘
+亙
+些
+亜
+亟
+亡
+交
+亥
+亦
+亨
+享
+京
+亭
+亮
+人
+什
+仁
+仇
+今
+介
+仍
+仏
+仔
+仕
+他
+仗
+付
+仙
+代
+令
+以
+仮
+仰
+仲
+件
+任
+企
+伊
+伍
+伎
+伏
+伐
+休
+会
+伝
+伯
+估
+伴
+伶
+伸
+伺
+似
+伽
+佃
+但
+位
+低
+住
+佐
+佑
+体
+何
+余
+佚
+佛
+作
+佩
+佳
+併
+佶
+使
+侈
+例
+侍
+侏
+侑
+侘
+供
+依
+侠
+価
+侮
+侯
+侵
+侶
+便
+係
+促
+俄
+俊
+俔
+俗
+俘
+保
+信
+俣
+俤
+修
+俯
+俳
+俵
+俸
+俺
+倉
+個
+倍
+倒
+候
+借
+倣
+値
+倫
+倭
+倶
+倹
+偃
+假
+偈
+偉
+偏
+偐
+偕
+停
+健
+側
+偵
+偶
+偽
+傀
+傅
+傍
+傑
+傘
+備
+催
+傭
+傲
+傳
+債
+傷
+傾
+僊
+働
+像
+僑
+僕
+僚
+僧
+僭
+僮
+儀
+億
+儇
+儒
+儛
+償
+儡
+優
+儲
+儺
+儼
+兀
+允
+元
+兄
+充
+兆
+先
+光
+克
+兌
+免
+兎
+児
+党
+兜
+入
+全
+八
+公
+六
+共
+兵
+其
+具
+典
+兼
+内
+円
+冊
+再
+冑
+冒
+冗
+写
+冠
+冤
+冥
+冨
+冬
+冲
+决
+冶
+冷
+准
+凉
+凋
+凌
+凍
+凛
+凝
+凞
+几
+凡
+処
+凪
+凰
+凱
+凶
+凸
+凹
+出
+函
+刀
+刃
+分
+切
+刈
+刊
+刎
+刑
+列
+初
+判
+別
+利
+刪
+到
+制
+刷
+券
+刹
+刺
+刻
+剃
+則
+削
+剋
+前
+剖
+剛
+剣
+剤
+剥
+剪
+副
+剰
+割
+創
+剽
+劇
+劉
+劔
+力
+功
+加
+劣
+助
+努
+劫
+劭
+励
+労
+効
+劾
+勃
+勅
+勇
+勉
+勒
+動
+勘
+務
+勝
+募
+勢
+勤
+勧
+勲
+勺
+勾
+勿
+匁
+匂
+包
+匏
+化
+北
+匙
+匝
+匠
+匡
+匣
+匯
+匲
+匹
+区
+医
+匿
+十
+千
+升
+午
+卉
+半
+卍
+卑
+卒
+卓
+協
+南
+単
+博
+卜
+占
+卦
+卯
+印
+危
+即
+却
+卵
+卸
+卿
+厄
+厚
+原
+厠
+厨
+厩
+厭
+厳
+去
+参
+又
+叉
+及
+友
+双
+反
+収
+叔
+取
+受
+叙
+叛
+叟
+叡
+叢
+口
+古
+句
+叩
+只
+叫
+召
+可
+台
+叱
+史
+右
+叶
+号
+司
+吃
+各
+合
+吉
+吊
+同
+名
+后
+吏
+吐
+向
+君
+吝
+吟
+吠
+否
+含
+吸
+吹
+吻
+吽
+吾
+呂
+呆
+呈
+呉
+告
+呑
+周
+呪
+呰
+味
+呼
+命
+咀
+咄
+咋
+和
+咒
+咫
+咲
+咳
+咸
+哀
+品
+哇
+哉
+員
+哨
+哩
+哭
+哲
+哺
+唄
+唆
+唇
+唐
+唖
+唯
+唱
+唳
+唸
+唾
+啄
+商
+問
+啓
+啼
+善
+喋
+喚
+喜
+喝
+喧
+喩
+喪
+喫
+喬
+單
+喰
+営
+嗅
+嗇
+嗔
+嗚
+嗜
+嗣
+嘆
+嘉
+嘗
+嘘
+嘩
+嘯
+嘱
+嘲
+嘴
+噂
+噌
+噛
+器
+噴
+噺
+嚆
+嚢
+囀
+囃
+囉
+囚
+四
+回
+因
+団
+困
+囲
+図
+固
+国
+圀
+圃
+國
+圏
+園
+圓
+團
+圜
+土
+圧
+在
+圭
+地
+址
+坂
+均
+坊
+坐
+坑
+坡
+坤
+坦
+坪
+垂
+型
+垢
+垣
+埃
+埋
+城
+埒
+埔
+域
+埠
+埴
+埵
+執
+培
+基
+埼
+堀
+堂
+堅
+堆
+堕
+堤
+堪
+堯
+堰
+報
+場
+堵
+堺
+塀
+塁
+塊
+塑
+塔
+塗
+塘
+塙
+塚
+塞
+塩
+填
+塵
+塾
+境
+墉
+墓
+増
+墜
+墟
+墨
+墳
+墺
+墻
+墾
+壁
+壇
+壊
+壌
+壕
+士
+壬
+壮
+声
+壱
+売
+壷
+壹
+壺
+壽
+変
+夏
+夕
+外
+夙
+多
+夜
+夢
+夥
+大
+天
+太
+夫
+夬
+夭
+央
+失
+夷
+夾
+奄
+奇
+奈
+奉
+奎
+奏
+契
+奔
+奕
+套
+奘
+奠
+奢
+奥
+奨
+奪
+奮
+女
+奴
+奸
+好
+如
+妃
+妄
+妊
+妍
+妓
+妖
+妙
+妥
+妨
+妬
+妲
+妹
+妻
+妾
+姉
+始
+姐
+姓
+委
+姚
+姜
+姞
+姥
+姦
+姨
+姪
+姫
+姶
+姻
+姿
+威
+娑
+娘
+娟
+娠
+娩
+娯
+娼
+婆
+婉
+婚
+婢
+婦
+婬
+婿
+媄
+媒
+媓
+媚
+媛
+媞
+媽
+嫁
+嫄
+嫉
+嫌
+嫐
+嫗
+嫡
+嬉
+嬌
+嬢
+嬪
+嬬
+嬾
+孁
+子
+孔
+字
+存
+孚
+孝
+孟
+季
+孤
+学
+孫
+孵
+學
+宅
+宇
+守
+安
+宋
+完
+宍
+宏
+宕
+宗
+官
+宙
+定
+宛
+宜
+宝
+実
+客
+宣
+室
+宥
+宮
+宰
+害
+宴
+宵
+家
+宸
+容
+宿
+寂
+寄
+寅
+密
+寇
+富
+寒
+寓
+寔
+寛
+寝
+察
+寡
+實
+寧
+審
+寮
+寵
+寶
+寸
+寺
+対
+寿
+封
+専
+射
+将
+尉
+尊
+尋
+對
+導
+小
+少
+尖
+尚
+尤
+尪
+尭
+就
+尹
+尺
+尻
+尼
+尽
+尾
+尿
+局
+居
+屈
+届
+屋
+屍
+屎
+屏
+屑
+屓
+展
+属
+屠
+層
+履
+屯
+山
+岐
+岑
+岡
+岩
+岫
+岬
+岳
+岷
+岸
+峠
+峡
+峨
+峯
+峰
+島
+峻
+崇
+崋
+崎
+崑
+崖
+崗
+崛
+崩
+嵌
+嵐
+嵩
+嵯
+嶂
+嶋
+嶠
+嶺
+嶼
+嶽
+巀
+巌
+巒
+巖
+川
+州
+巡
+巣
+工
+左
+巧
+巨
+巫
+差
+己
+巳
+巴
+巷
+巻
+巽
+巾
+市
+布
+帆
+希
+帖
+帚
+帛
+帝
+帥
+師
+席
+帯
+帰
+帳
+帷
+常
+帽
+幄
+幅
+幇
+幌
+幔
+幕
+幟
+幡
+幢
+幣
+干
+平
+年
+并
+幸
+幹
+幻
+幼
+幽
+幾
+庁
+広
+庄
+庇
+床
+序
+底
+庖
+店
+庚
+府
+度
+座
+庫
+庭
+庵
+庶
+康
+庸
+廂
+廃
+廉
+廊
+廓
+廟
+廠
+廣
+廬
+延
+廷
+建
+廻
+廼
+廿
+弁
+弄
+弉
+弊
+弌
+式
+弐
+弓
+弔
+引
+弖
+弗
+弘
+弛
+弟
+弥
+弦
+弧
+弱
+張
+強
+弼
+弾
+彈
+彊
+彌
+彎
+当
+彗
+彙
+彝
+形
+彦
+彩
+彫
+彬
+彭
+彰
+影
+彷
+役
+彼
+往
+征
+徂
+径
+待
+律
+後
+徐
+徑
+徒
+従
+得
+徠
+御
+徧
+徨
+復
+循
+徭
+微
+徳
+徴
+德
+徹
+徽
+心
+必
+忉
+忌
+忍
+志
+忘
+忙
+応
+忠
+快
+忯
+念
+忻
+忽
+忿
+怒
+怖
+思
+怠
+怡
+急
+性
+怨
+怪
+怯
+恂
+恋
+恐
+恒
+恕
+恣
+恤
+恥
+恨
+恩
+恬
+恭
+息
+恵
+悉
+悌
+悍
+悔
+悟
+悠
+患
+悦
+悩
+悪
+悲
+悼
+情
+惇
+惑
+惚
+惜
+惟
+惠
+惣
+惧
+惨
+惰
+想
+惹
+惺
+愈
+愉
+愍
+意
+愔
+愚
+愛
+感
+愷
+愿
+慈
+態
+慌
+慎
+慕
+慢
+慣
+慧
+慨
+慮
+慰
+慶
+憂
+憎
+憐
+憑
+憙
+憤
+憧
+憩
+憬
+憲
+憶
+憾
+懇
+應
+懌
+懐
+懲
+懸
+懺
+懽
+懿
+戈
+戊
+戌
+戎
+成
+我
+戒
+戔
+或
+戚
+戟
+戦
+截
+戮
+戯
+戴
+戸
+戻
+房
+所
+扁
+扇
+扈
+扉
+手
+才
+打
+払
+托
+扮
+扱
+扶
+批
+承
+技
+抄
+把
+抑
+抓
+投
+抗
+折
+抜
+択
+披
+抱
+抵
+抹
+押
+抽
+担
+拇
+拈
+拉
+拍
+拏
+拐
+拒
+拓
+拘
+拙
+招
+拝
+拠
+拡
+括
+拭
+拳
+拵
+拶
+拾
+拿
+持
+挂
+指
+按
+挑
+挙
+挟
+挨
+振
+挺
+挽
+挿
+捉
+捕
+捗
+捜
+捧
+捨
+据
+捺
+捻
+掃
+掄
+授
+掌
+排
+掖
+掘
+掛
+掟
+採
+探
+掣
+接
+控
+推
+掩
+措
+掬
+掲
+掴
+掻
+掾
+揃
+揄
+揆
+揉
+描
+提
+揖
+揚
+換
+握
+揮
+援
+揶
+揺
+損
+搦
+搬
+搭
+携
+搾
+摂
+摘
+摩
+摸
+摺
+撃
+撒
+撞
+撤
+撥
+撫
+播
+撮
+撰
+撲
+撹
+擁
+操
+擔
+擦
+擬
+擾
+攘
+攝
+攣
+支
+收
+改
+攻
+放
+政
+故
+敏
+救
+敗
+教
+敢
+散
+敦
+敬
+数
+整
+敵
+敷
+斂
+文
+斉
+斎
+斐
+斑
+斗
+料
+斜
+斟
+斤
+斥
+斧
+斬
+断
+斯
+新
+方
+於
+施
+旁
+旅
+旋
+旌
+族
+旗
+旛
+无
+旡
+既
+日
+旦
+旧
+旨
+早
+旬
+旭
+旺
+旻
+昂
+昆
+昇
+昉
+昌
+明
+昏
+易
+昔
+星
+映
+春
+昧
+昨
+昪
+昭
+是
+昵
+昼
+晁
+時
+晃
+晋
+晏
+晒
+晟
+晦
+晧
+晩
+普
+景
+晴
+晶
+智
+暁
+暇
+暈
+暉
+暑
+暖
+暗
+暘
+暢
+暦
+暫
+暮
+暲
+暴
+暹
+暾
+曄
+曇
+曉
+曖
+曙
+曜
+曝
+曠
+曰
+曲
+曳
+更
+書
+曹
+曼
+曽
+曾
+替
+最
+會
+月
+有
+朋
+服
+朏
+朔
+朕
+朗
+望
+朝
+期
+朧
+木
+未
+末
+本
+札
+朱
+朴
+机
+朽
+杁
+杉
+李
+杏
+材
+村
+杓
+杖
+杜
+杞
+束
+条
+杢
+杣
+来
+杭
+杮
+杯
+東
+杲
+杵
+杷
+杼
+松
+板
+枅
+枇
+析
+枓
+枕
+林
+枚
+果
+枝
+枠
+枡
+枢
+枯
+枳
+架
+柄
+柊
+柏
+某
+柑
+染
+柔
+柘
+柚
+柯
+柱
+柳
+柴
+柵
+査
+柾
+柿
+栂
+栃
+栄
+栖
+栗
+校
+株
+栲
+栴
+核
+根
+栻
+格
+栽
+桁
+桂
+桃
+框
+案
+桐
+桑
+桓
+桔
+桜
+桝
+桟
+桧
+桴
+桶
+桾
+梁
+梅
+梆
+梓
+梔
+梗
+梛
+條
+梟
+梢
+梧
+梨
+械
+梱
+梲
+梵
+梶
+棄
+棋
+棒
+棗
+棘
+棚
+棟
+棠
+森
+棲
+棹
+棺
+椀
+椅
+椋
+植
+椎
+椏
+椒
+椙
+検
+椥
+椹
+椿
+楊
+楓
+楕
+楚
+楞
+楠
+楡
+楢
+楨
+楪
+楫
+業
+楮
+楯
+楳
+極
+楷
+楼
+楽
+概
+榊
+榎
+榕
+榛
+榜
+榮
+榱
+榴
+槃
+槇
+槊
+構
+槌
+槍
+槐
+様
+槙
+槻
+槽
+槿
+樂
+樋
+樓
+樗
+標
+樟
+模
+権
+横
+樫
+樵
+樹
+樺
+樽
+橇
+橋
+橘
+機
+橿
+檀
+檄
+檎
+檐
+檗
+檜
+檣
+檥
+檬
+檮
+檸
+檻
+櫃
+櫓
+櫛
+櫟
+櫨
+櫻
+欄
+欅
+欠
+次
+欣
+欧
+欲
+欺
+欽
+款
+歌
+歎
+歓
+止
+正
+此
+武
+歩
+歪
+歯
+歳
+歴
+死
+殆
+殉
+殊
+残
+殖
+殯
+殴
+段
+殷
+殺
+殻
+殿
+毀
+毅
+母
+毎
+毒
+比
+毘
+毛
+毫
+毬
+氈
+氏
+民
+気
+水
+氷
+永
+氾
+汀
+汁
+求
+汎
+汐
+汗
+汚
+汝
+江
+池
+汪
+汰
+汲
+決
+汽
+沂
+沃
+沅
+沆
+沈
+沌
+沐
+沓
+沖
+沙
+没
+沢
+沱
+河
+沸
+油
+治
+沼
+沽
+沿
+況
+泉
+泊
+泌
+法
+泗
+泡
+波
+泣
+泥
+注
+泯
+泰
+泳
+洋
+洒
+洗
+洛
+洞
+津
+洩
+洪
+洲
+洸
+洹
+活
+洽
+派
+流
+浄
+浅
+浙
+浚
+浜
+浣
+浦
+浩
+浪
+浮
+浴
+海
+浸
+涅
+消
+涌
+涙
+涛
+涯
+液
+涵
+涼
+淀
+淄
+淆
+淇
+淋
+淑
+淘
+淡
+淤
+淨
+淫
+深
+淳
+淵
+混
+淹
+添
+清
+済
+渉
+渋
+渓
+渕
+渚
+減
+渟
+渠
+渡
+渤
+渥
+渦
+温
+渫
+測
+港
+游
+渾
+湊
+湖
+湘
+湛
+湧
+湫
+湯
+湾
+湿
+満
+源
+準
+溜
+溝
+溢
+溥
+溪
+溶
+溺
+滄
+滅
+滋
+滌
+滑
+滕
+滝
+滞
+滴
+滸
+滹
+滿
+漁
+漂
+漆
+漉
+漏
+漑
+演
+漕
+漠
+漢
+漣
+漫
+漬
+漱
+漸
+漿
+潅
+潔
+潙
+潜
+潟
+潤
+潭
+潮
+潰
+潴
+澁
+澂
+澄
+澎
+澗
+澤
+澪
+澱
+澳
+激
+濁
+濃
+濟
+濠
+濡
+濤
+濫
+濯
+濱
+濾
+瀉
+瀋
+瀑
+瀕
+瀞
+瀟
+瀧
+瀬
+瀾
+灌
+灑
+灘
+火
+灯
+灰
+灸
+災
+炉
+炊
+炎
+炒
+炭
+炮
+炷
+点
+為
+烈
+烏
+烙
+烝
+烹
+焔
+焙
+焚
+無
+焦
+然
+焼
+煇
+煉
+煌
+煎
+煕
+煙
+煤
+煥
+照
+煩
+煬
+煮
+煽
+熈
+熊
+熙
+熟
+熨
+熱
+熹
+熾
+燃
+燈
+燎
+燔
+燕
+燗
+燥
+燭
+燻
+爆
+爐
+爪
+爬
+爲
+爵
+父
+爺
+爼
+爽
+爾
+片
+版
+牌
+牒
+牘
+牙
+牛
+牝
+牟
+牡
+牢
+牧
+物
+牲
+特
+牽
+犂
+犠
+犬
+犯
+状
+狂
+狄
+狐
+狗
+狙
+狛
+狡
+狩
+独
+狭
+狷
+狸
+狼
+猊
+猛
+猟
+猥
+猨
+猩
+猪
+猫
+献
+猴
+猶
+猷
+猾
+猿
+獄
+獅
+獏
+獣
+獲
+玄
+玅
+率
+玉
+王
+玖
+玩
+玲
+珀
+珂
+珈
+珉
+珊
+珍
+珎
+珞
+珠
+珣
+珥
+珪
+班
+現
+球
+理
+琉
+琢
+琥
+琦
+琮
+琲
+琳
+琴
+琵
+琶
+瑁
+瑋
+瑙
+瑚
+瑛
+瑜
+瑞
+瑠
+瑤
+瑩
+瑪
+瑳
+瑾
+璃
+璋
+璜
+璞
+璧
+璨
+環
+璵
+璽
+璿
+瓊
+瓔
+瓜
+瓢
+瓦
+瓶
+甍
+甑
+甕
+甘
+甚
+甞
+生
+産
+甥
+用
+甫
+田
+由
+甲
+申
+男
+町
+画
+界
+畏
+畑
+畔
+留
+畜
+畝
+畠
+畢
+略
+番
+異
+畳
+當
+畷
+畸
+畺
+畿
+疆
+疇
+疋
+疎
+疏
+疑
+疫
+疱
+疲
+疹
+疼
+疾
+病
+症
+痒
+痔
+痕
+痘
+痙
+痛
+痢
+痩
+痴
+痺
+瘍
+瘡
+瘧
+療
+癇
+癌
+癒
+癖
+癡
+癪
+発
+登
+白
+百
+的
+皆
+皇
+皋
+皐
+皓
+皮
+皺
+皿
+盂
+盃
+盆
+盈
+益
+盒
+盗
+盛
+盞
+盟
+盡
+監
+盤
+盥
+盧
+目
+盲
+直
+相
+盾
+省
+眉
+看
+県
+眞
+真
+眠
+眷
+眺
+眼
+着
+睡
+督
+睦
+睨
+睿
+瞋
+瞑
+瞞
+瞬
+瞭
+瞰
+瞳
+瞻
+瞼
+瞿
+矍
+矛
+矜
+矢
+知
+矧
+矩
+短
+矮
+矯
+石
+砂
+砌
+研
+砕
+砥
+砦
+砧
+砲
+破
+砺
+硝
+硫
+硬
+硯
+碁
+碇
+碌
+碑
+碓
+碕
+碗
+碣
+碧
+碩
+確
+碾
+磁
+磐
+磔
+磧
+磨
+磬
+磯
+礁
+礎
+礒
+礙
+礫
+礬
+示
+礼
+社
+祀
+祁
+祇
+祈
+祉
+祐
+祓
+祕
+祖
+祗
+祚
+祝
+神
+祟
+祠
+祢
+祥
+票
+祭
+祷
+祺
+禁
+禄
+禅
+禊
+禍
+禎
+福
+禔
+禖
+禛
+禦
+禧
+禮
+禰
+禹
+禽
+禿
+秀
+私
+秋
+科
+秒
+秘
+租
+秤
+秦
+秩
+称
+移
+稀
+程
+税
+稔
+稗
+稙
+稚
+稜
+稠
+種
+稱
+稲
+稷
+稻
+稼
+稽
+稿
+穀
+穂
+穆
+積
+穎
+穏
+穗
+穜
+穢
+穣
+穫
+穴
+究
+空
+突
+窃
+窄
+窒
+窓
+窟
+窠
+窩
+窪
+窮
+窯
+竃
+竄
+竈
+立
+站
+竜
+竝
+竟
+章
+童
+竪
+竭
+端
+竴
+競
+竹
+竺
+竽
+竿
+笄
+笈
+笏
+笑
+笙
+笛
+笞
+笠
+笥
+符
+第
+笹
+筅
+筆
+筇
+筈
+等
+筋
+筌
+筍
+筏
+筐
+筑
+筒
+答
+策
+筝
+筥
+筧
+筬
+筮
+筯
+筰
+筵
+箆
+箇
+箋
+箏
+箒
+箔
+箕
+算
+箙
+箜
+管
+箪
+箭
+箱
+箸
+節
+篁
+範
+篆
+篇
+築
+篋
+篌
+篝
+篠
+篤
+篥
+篦
+篩
+篭
+篳
+篷
+簀
+簒
+簡
+簧
+簪
+簫
+簺
+簾
+簿
+籀
+籃
+籌
+籍
+籐
+籟
+籠
+籤
+籬
+米
+籾
+粂
+粉
+粋
+粒
+粕
+粗
+粘
+粛
+粟
+粥
+粧
+粮
+粳
+精
+糊
+糖
+糜
+糞
+糟
+糠
+糧
+糯
+糸
+糺
+系
+糾
+紀
+約
+紅
+紋
+納
+紐
+純
+紗
+紘
+紙
+級
+紛
+素
+紡
+索
+紫
+紬
+累
+細
+紳
+紵
+紹
+紺
+絁
+終
+絃
+組
+絅
+経
+結
+絖
+絞
+絡
+絣
+給
+統
+絲
+絵
+絶
+絹
+絽
+綏
+經
+継
+続
+綜
+綟
+綬
+維
+綱
+網
+綴
+綸
+綺
+綽
+綾
+綿
+緊
+緋
+総
+緑
+緒
+線
+締
+緥
+編
+緩
+緬
+緯
+練
+緻
+縁
+縄
+縅
+縒
+縛
+縞
+縢
+縣
+縦
+縫
+縮
+縹
+總
+績
+繁
+繊
+繋
+繍
+織
+繕
+繝
+繦
+繧
+繰
+繹
+繼
+纂
+纈
+纏
+纐
+纒
+纛
+缶
+罔
+罠
+罧
+罪
+置
+罰
+署
+罵
+罷
+罹
+羂
+羅
+羆
+羇
+羈
+羊
+羌
+美
+群
+羨
+義
+羯
+羲
+羹
+羽
+翁
+翅
+翌
+習
+翔
+翛
+翠
+翡
+翫
+翰
+翺
+翻
+翼
+耀
+老
+考
+者
+耆
+而
+耐
+耕
+耗
+耨
+耳
+耶
+耽
+聊
+聖
+聘
+聚
+聞
+聟
+聡
+聨
+聯
+聰
+聲
+聴
+職
+聾
+肄
+肆
+肇
+肉
+肋
+肌
+肖
+肘
+肛
+肝
+股
+肢
+肥
+肩
+肪
+肯
+肱
+育
+肴
+肺
+胃
+胆
+背
+胎
+胖
+胚
+胝
+胞
+胡
+胤
+胱
+胴
+胸
+能
+脂
+脅
+脆
+脇
+脈
+脊
+脚
+脛
+脩
+脱
+脳
+腋
+腎
+腐
+腑
+腔
+腕
+腫
+腰
+腱
+腸
+腹
+腺
+腿
+膀
+膏
+膚
+膜
+膝
+膠
+膣
+膨
+膩
+膳
+膵
+膾
+膿
+臂
+臆
+臈
+臍
+臓
+臘
+臚
+臣
+臥
+臨
+自
+臭
+至
+致
+臺
+臼
+舂
+舅
+與
+興
+舌
+舍
+舎
+舒
+舖
+舗
+舘
+舜
+舞
+舟
+舩
+航
+般
+舳
+舶
+船
+艇
+艘
+艦
+艮
+良
+色
+艶
+芋
+芒
+芙
+芝
+芥
+芦
+芬
+芭
+芯
+花
+芳
+芸
+芹
+芻
+芽
+芿
+苅
+苑
+苔
+苗
+苛
+苞
+苡
+若
+苦
+苧
+苫
+英
+苴
+苻
+茂
+范
+茄
+茅
+茎
+茗
+茘
+茜
+茨
+茲
+茵
+茶
+茸
+茹
+草
+荊
+荏
+荒
+荘
+荷
+荻
+荼
+莞
+莪
+莫
+莬
+莱
+莵
+莽
+菅
+菊
+菌
+菓
+菖
+菘
+菜
+菟
+菩
+菫
+華
+菱
+菴
+萄
+萊
+萌
+萍
+萎
+萠
+萩
+萬
+萱
+落
+葉
+著
+葛
+葡
+董
+葦
+葩
+葬
+葭
+葱
+葵
+葺
+蒋
+蒐
+蒔
+蒙
+蒟
+蒡
+蒲
+蒸
+蒻
+蒼
+蒿
+蓄
+蓆
+蓉
+蓋
+蓑
+蓬
+蓮
+蓼
+蔀
+蔑
+蔓
+蔚
+蔡
+蔦
+蔬
+蔭
+蔵
+蔽
+蕃
+蕉
+蕊
+蕎
+蕨
+蕩
+蕪
+蕭
+蕾
+薄
+薇
+薊
+薔
+薗
+薙
+薛
+薦
+薨
+薩
+薪
+薫
+薬
+薭
+薮
+藁
+藉
+藍
+藏
+藐
+藝
+藤
+藩
+藪
+藷
+藹
+藺
+藻
+蘂
+蘆
+蘇
+蘊
+蘭
+虎
+虐
+虔
+虚
+虜
+虞
+號
+虫
+虹
+虻
+蚊
+蚕
+蛇
+蛉
+蛍
+蛎
+蛙
+蛛
+蛟
+蛤
+蛭
+蛮
+蛸
+蛹
+蛾
+蜀
+蜂
+蜃
+蜆
+蜊
+蜘
+蜜
+蜷
+蜻
+蝉
+蝋
+蝕
+蝙
+蝠
+蝦
+蝶
+蝿
+螂
+融
+螣
+螺
+蟄
+蟇
+蟠
+蟷
+蟹
+蟻
+蠢
+蠣
+血
+衆
+行
+衍
+衒
+術
+街
+衙
+衛
+衝
+衞
+衡
+衢
+衣
+表
+衫
+衰
+衵
+衷
+衽
+衾
+衿
+袁
+袈
+袋
+袍
+袒
+袖
+袙
+袞
+袢
+被
+袰
+袱
+袴
+袷
+袿
+裁
+裂
+裃
+装
+裏
+裔
+裕
+裘
+裙
+補
+裟
+裡
+裲
+裳
+裴
+裸
+裹
+製
+裾
+褂
+褄
+複
+褌
+褐
+褒
+褥
+褪
+褶
+褻
+襄
+襖
+襞
+襟
+襠
+襦
+襪
+襲
+襴
+襷
+西
+要
+覆
+覇
+覈
+見
+規
+視
+覗
+覚
+覧
+親
+覲
+観
+覺
+觀
+角
+解
+触
+言
+訂
+計
+討
+訓
+託
+記
+訛
+訟
+訢
+訥
+訪
+設
+許
+訳
+訴
+訶
+診
+註
+証
+詐
+詔
+評
+詛
+詞
+詠
+詢
+詣
+試
+詩
+詫
+詮
+詰
+話
+該
+詳
+誄
+誅
+誇
+誉
+誌
+認
+誓
+誕
+誘
+語
+誠
+誡
+誣
+誤
+誥
+誦
+説
+読
+誰
+課
+誼
+誾
+調
+談
+請
+諌
+諍
+諏
+諒
+論
+諚
+諜
+諟
+諡
+諦
+諧
+諫
+諭
+諮
+諱
+諶
+諷
+諸
+諺
+諾
+謀
+謄
+謌
+謎
+謗
+謙
+謚
+講
+謝
+謡
+謫
+謬
+謹
+證
+識
+譚
+譛
+譜
+警
+譬
+譯
+議
+譲
+譴
+護
+讀
+讃
+讐
+讒
+谷
+谿
+豅
+豆
+豊
+豎
+豐
+豚
+象
+豪
+豫
+豹
+貌
+貝
+貞
+負
+財
+貢
+貧
+貨
+販
+貪
+貫
+責
+貯
+貰
+貴
+買
+貸
+費
+貼
+貿
+賀
+賁
+賂
+賃
+賄
+資
+賈
+賊
+賎
+賑
+賓
+賛
+賜
+賞
+賠
+賢
+賣
+賤
+賦
+質
+賭
+購
+賽
+贄
+贅
+贈
+贋
+贔
+贖
+赤
+赦
+走
+赴
+起
+超
+越
+趙
+趣
+足
+趺
+趾
+跋
+跏
+距
+跡
+跨
+跪
+路
+跳
+践
+踊
+踏
+踐
+踞
+踪
+踵
+蹄
+蹉
+蹊
+蹟
+蹲
+蹴
+躅
+躇
+躊
+躍
+躑
+躙
+躪
+身
+躬
+躯
+躰
+車
+軋
+軌
+軍
+軒
+軟
+転
+軸
+軻
+軽
+軾
+較
+載
+輌
+輔
+輜
+輝
+輦
+輩
+輪
+輯
+輸
+輿
+轄
+轍
+轟
+轢
+辛
+辞
+辟
+辥
+辦
+辨
+辰
+辱
+農
+辺
+辻
+込
+迂
+迅
+迎
+近
+返
+迢
+迦
+迪
+迫
+迭
+述
+迷
+迹
+追
+退
+送
+逃
+逅
+逆
+逍
+透
+逐
+逓
+途
+逕
+逗
+這
+通
+逝
+逞
+速
+造
+逢
+連
+逮
+週
+進
+逸
+逼
+遁
+遂
+遅
+遇
+遊
+運
+遍
+過
+遐
+道
+達
+違
+遙
+遜
+遠
+遡
+遣
+遥
+適
+遭
+遮
+遯
+遵
+遷
+選
+遺
+遼
+避
+邀
+邁
+邂
+邃
+還
+邇
+邉
+邊
+邑
+那
+邦
+邨
+邪
+邯
+邵
+邸
+郁
+郊
+郎
+郡
+郢
+部
+郭
+郴
+郵
+郷
+都
+鄂
+鄙
+鄭
+鄰
+鄲
+酉
+酋
+酌
+配
+酎
+酒
+酔
+酢
+酥
+酪
+酬
+酵
+酷
+酸
+醍
+醐
+醒
+醗
+醜
+醤
+醪
+醵
+醸
+采
+釈
+釉
+釋
+里
+重
+野
+量
+釐
+金
+釘
+釜
+針
+釣
+釧
+釿
+鈍
+鈎
+鈐
+鈔
+鈞
+鈦
+鈴
+鈷
+鈸
+鈿
+鉄
+鉇
+鉉
+鉋
+鉛
+鉢
+鉤
+鉦
+鉱
+鉾
+銀
+銃
+銅
+銈
+銑
+銕
+銘
+銚
+銜
+銭
+鋏
+鋒
+鋤
+鋭
+鋲
+鋳
+鋸
+鋺
+鋼
+錆
+錍
+錐
+錘
+錠
+錣
+錦
+錫
+錬
+錯
+録
+錵
+鍋
+鍍
+鍑
+鍔
+鍛
+鍬
+鍮
+鍵
+鍼
+鍾
+鎌
+鎖
+鎗
+鎚
+鎧
+鎬
+鎮
+鎰
+鎹
+鏃
+鏑
+鏡
+鐃
+鐇
+鐐
+鐔
+鐘
+鐙
+鐚
+鐡
+鐵
+鐸
+鑁
+鑊
+鑑
+鑒
+鑚
+鑠
+鑢
+鑰
+鑵
+鑷
+鑼
+鑽
+鑿
+長
+門
+閃
+閇
+閉
+開
+閏
+閑
+間
+閔
+閘
+関
+閣
+閤
+閥
+閦
+閨
+閬
+閲
+閻
+閼
+閾
+闇
+闍
+闔
+闕
+闘
+關
+闡
+闢
+闥
+阜
+阪
+阮
+阯
+防
+阻
+阿
+陀
+陂
+附
+陌
+降
+限
+陛
+陞
+院
+陣
+除
+陥
+陪
+陬
+陰
+陳
+陵
+陶
+陸
+険
+陽
+隅
+隆
+隈
+隊
+隋
+階
+随
+隔
+際
+障
+隠
+隣
+隧
+隷
+隻
+隼
+雀
+雁
+雄
+雅
+集
+雇
+雉
+雊
+雋
+雌
+雍
+雑
+雖
+雙
+雛
+離
+難
+雨
+雪
+雫
+雰
+雲
+零
+雷
+雹
+電
+需
+震
+霊
+霍
+霖
+霜
+霞
+霧
+霰
+露
+靈
+青
+靖
+静
+靜
+非
+面
+革
+靫
+靭
+靱
+靴
+靺
+鞁
+鞄
+鞆
+鞋
+鞍
+鞏
+鞘
+鞠
+鞨
+鞭
+韋
+韓
+韜
+韮
+音
+韶
+韻
+響
+頁
+頂
+頃
+項
+順
+須
+頌
+預
+頑
+頒
+頓
+領
+頚
+頬
+頭
+頴
+頸
+頻
+頼
+顆
+題
+額
+顎
+顔
+顕
+顗
+願
+顛
+類
+顧
+顯
+風
+飛
+食
+飢
+飩
+飫
+飯
+飲
+飴
+飼
+飽
+飾
+餃
+餅
+餉
+養
+餌
+餐
+餓
+餘
+餝
+餡
+館
+饂
+饅
+饉
+饋
+饌
+饒
+饗
+首
+馗
+香
+馨
+馬
+馳
+馴
+駄
+駅
+駆
+駈
+駐
+駒
+駕
+駝
+駿
+騁
+騎
+騏
+騒
+験
+騙
+騨
+騰
+驕
+驚
+驛
+驢
+骨
+骸
+髄
+體
+高
+髙
+髢
+髪
+髭
+髮
+髷
+髻
+鬘
+鬚
+鬢
+鬨
+鬯
+鬱
+鬼
+魁
+魂
+魄
+魅
+魏
+魔
+魚
+魯
+鮎
+鮑
+鮒
+鮪
+鮫
+鮭
+鮮
+鯉
+鯔
+鯖
+鯛
+鯨
+鯰
+鯱
+鰐
+鰒
+鰭
+鰯
+鰰
+鰹
+鰻
+鱈
+鱒
+鱗
+鱧
+鳥
+鳩
+鳰
+鳳
+鳴
+鳶
+鴈
+鴉
+鴎
+鴛
+鴟
+鴦
+鴨
+鴫
+鴻
+鵄
+鵜
+鵞
+鵡
+鵬
+鵲
+鵺
+鶉
+鶏
+鶯
+鶴
+鷄
+鷙
+鷲
+鷹
+鷺
+鸚
+鸞
+鹸
+鹽
+鹿
+麁
+麒
+麓
+麗
+麝
+麞
+麟
+麦
+麩
+麹
+麺
+麻
+麾
+麿
+黄
+黌
+黍
+黒
+黙
+黛
+黠
+鼈
+鼉
+鼎
+鼓
+鼠
+鼻
+齊
+齋
+齟
+齢
+齬
+龍
+龕
+龗
+！
+＃
+％
+＆
+（
+）
+＋
+，
+－
+．
+／
+０
+１
+２
+３
+４
+５
+６
+７
+８
+９
+：
+；
+＝
+？
+＠
+Ａ
+Ｂ
+Ｃ
+Ｄ
+Ｅ
+Ｆ
+Ｇ
+Ｈ
+Ｉ
+Ｊ
+Ｋ
+Ｌ
+Ｍ
+Ｎ
+Ｏ
+Ｐ
+Ｒ
+Ｓ
+Ｔ
+Ｕ
+Ｖ
+Ｗ
+Ｘ
+Ｚ
+ａ
+ｃ
+ｄ
+ｅ
+ｆ
+ｈ
+ｉ
+ｊ
+ｋ
+ｌ
+ｍ
+ｎ
+ｏ
+ｐ
+ｒ
+ｓ
+ｔ
+ｕ
+ｙ
+ｚ
+～
+･
+ 
diff --git a/ppocr/utils/dict/ka_dict.txt b/ppocr/utils/dict/ka_dict.txt
new file mode 100644
index 0000000..d506b69
--- /dev/null
+++ b/ppocr/utils/dict/ka_dict.txt
@@ -0,0 +1,153 @@
+k
+a
+_
+i
+m
+g
+/
+1
+2
+I
+L
+S
+V
+R
+C
+0
+v
+l
+6
+4
+8
+.
+j
+p
+ಗ
+ು
+ಣ
+ಪ
+ಡ
+ಿ
+ಸ
+ಲ
+ಾ
+ದ
+್
+7
+5
+3
+ವ
+ಷ
+ಬ
+ಹ
+ೆ
+9
+ಅ
+ಳ
+ನ
+ರ
+ಉ
+ಕ
+ಎ
+ೇ
+ಂ
+ೈ
+ೊ
+ೀ
+ಯ
+ೋ
+ತ
+ಶ
+ಭ
+ಧ
+ಚ
+ಜ
+ೂ
+ಮ
+ಒ
+ೃ
+ಥ
+ಇ
+ಟ
+ಖ
+ಆ
+ಞ
+ಫ
+-
+ಢ
+ಊ
+ಓ
+ಐ
+ಃ
+ಘ
+ಝ
+ೌ
+ಠ
+ಛ
+ಔ
+ಏ
+ಈ
+ಋ
+೨
+೦
+೧
+೮
+೯
+೪
+,
+೫
+೭
+೩
+೬
+ಙ
+s
+c
+e
+n
+w
+o
+u
+t
+d
+E
+A
+T
+B
+Z
+N
+G
+O
+q
+z
+r
+x
+P
+K
+M
+J
+U
+D
+f
+F
+h
+b
+W
+Y
+y
+H
+X
+Q
+'
+#
+&
+!
+@
+$
+:
+%
+é
+É
+(
+?
++
+ 
diff --git a/ppocr/utils/dict/kazakh_dict.txt b/ppocr/utils/dict/kazakh_dict.txt
new file mode 100644
index 0000000..007d4f5
--- /dev/null
+++ b/ppocr/utils/dict/kazakh_dict.txt
@@ -0,0 +1,42 @@
+А
+Ә
+Б
+В
+Г
+Ғ
+Д
+Е
+Ё
+Ж
+З
+И
+Й
+К
+Қ
+Л
+М
+Н
+Ң
+О
+Ө
+П
+Р
+С
+Т
+У
+Ұ
+Ү
+Ф
+Х
+Һ
+Ц
+Ч
+Ш
+Щ
+Ъ
+Ы
+І
+Ь
+Э
+Ю
+Я
diff --git a/ppocr/utils/dict/kie_dict/xfund_class_list.txt b/ppocr/utils/dict/kie_dict/xfund_class_list.txt
new file mode 100644
index 0000000..faded9f
--- /dev/null
+++ b/ppocr/utils/dict/kie_dict/xfund_class_list.txt
@@ -0,0 +1,4 @@
+OTHER
+QUESTION
+ANSWER
+HEADER
diff --git a/ppocr/utils/dict/korean_dict.txt b/ppocr/utils/dict/korean_dict.txt
new file mode 100644
index 0000000..a13899f
--- /dev/null
+++ b/ppocr/utils/dict/korean_dict.txt
@@ -0,0 +1,3688 @@
+!
+"
+#
+$
+%
+&
+'
+*
++
+-
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+<
+=
+>
+?
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+[
+\
+]
+^
+_
+`
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+{
+|
+}
+~
+©
+°
+²
+½
+Á
+Ä
+Å
+Ç
+É
+Í
+Î
+Ó
+Ö
+×
+Ü
+ß
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+ì
+í
+î
+ï
+ð
+ñ
+ò
+ó
+ô
+õ
+ö
+ø
+ú
+û
+ü
+ý
+ā
+ă
+ą
+ć
+Č
+č
+đ
+ē
+ė
+ę
+ě
+ğ
+ī
+İ
+ı
+Ł
+ł
+ń
+ň
+ō
+ř
+Ş
+ş
+Š
+š
+ţ
+ū
+ź
+ż
+Ž
+ž
+Ș
+ș
+Α
+Δ
+α
+λ
+φ
+Г
+О
+а
+в
+л
+о
+р
+с
+т
+я
+​
+’
+“
+”
+→
+∇
+∼
+「
+」
+ア
+カ
+グ
+ニ
+ラ
+ン
+ㄱ
+ㄴ
+ㄷ
+ㄸ
+ㄹ
+ㅂ
+ㅅ
+ㅆ
+ㅇ
+ㅈ
+ㅊ
+ㅋ
+ㅌ
+ㅎ
+ㅓ
+ㅜ
+ㅣ
+一
+丁
+七
+三
+上
+下
+不
+丑
+世
+丘
+丞
+中
+丸
+丹
+主
+乃
+久
+之
+乎
+乘
+九
+也
+乳
+乾
+事
+二
+云
+互
+五
+井
+亞
+亡
+交
+亥
+亨
+享
+京
+亭
+人
+仁
+今
+他
+仙
+代
+令
+以
+仰
+仲
+件
+任
+企
+伊
+伍
+伎
+伏
+伐
+休
+伯
+伴
+伸
+佃
+佈
+位
+低
+住
+佐
+何
+佛
+作
+使
+來
+供
+依
+侯
+侵
+侶
+便
+俗
+保
+俠
+信
+修
+俱
+俳
+倉
+個
+倍
+倒
+候
+借
+値
+倫
+倭
+假
+偈
+偉
+偏
+停
+偶
+傅
+傑
+傳
+傷
+傾
+像
+僞
+僥
+僧
+價
+儀
+儉
+儒
+優
+儼
+兀
+允
+元
+兆
+先
+光
+克
+兒
+入
+內
+全
+八
+公
+六
+共
+兵
+其
+具
+典
+兼
+再
+冠
+冥
+冶
+准
+凞
+凡
+凱
+出
+函
+刀
+分
+刊
+刑
+列
+初
+判
+別
+利
+到
+制
+券
+刺
+刻
+則
+前
+剛
+副
+創
+劃
+劑
+力
+功
+加
+劣
+助
+劫
+勇
+動
+務
+勝
+勢
+勳
+勸
+匈
+化
+北
+匠
+區
+十
+千
+午
+半
+卍
+卑
+卒
+卓
+南
+博
+卜
+占
+卦
+印
+危
+卵
+卷
+卽
+卿
+厄
+原
+厦
+去
+參
+又
+叉
+友
+反
+叔
+受
+口
+古
+句
+可
+台
+史
+右
+司
+各
+合
+吉
+同
+名
+后
+吏
+吐
+君
+吠
+吳
+呂
+告
+周
+味
+呵
+命
+和
+咳
+咸
+咽
+哀
+品
+哨
+哮
+哲
+唐
+唯
+唱
+商
+問
+啼
+善
+喆
+喉
+喜
+喩
+喪
+嘗
+器
+嚴
+囊
+四
+回
+因
+困
+固
+圈
+國
+圍
+園
+圓
+圖
+團
+土
+在
+地
+均
+坊
+坐
+坑
+坵
+型
+垢
+城
+域
+埴
+執
+培
+基
+堂
+堅
+堆
+堤
+堯
+報
+場
+塔
+塚
+塞
+塵
+境
+墜
+墟
+墨
+墳
+墾
+壁
+壇
+壓
+壤
+士
+壬
+壯
+壺
+壽
+夏
+夕
+外
+多
+夜
+夢
+大
+天
+太
+夫
+央
+失
+夷
+奄
+奇
+奉
+奎
+奏
+契
+奔
+奮
+女
+奴
+好
+如
+妄
+妊
+妖
+妙
+始
+姑
+姓
+姚
+姜
+威
+婆
+婚
+婦
+媒
+媚
+子
+孔
+字
+存
+孝
+孟
+季
+孤
+孫
+學
+孺
+宇
+守
+安
+宋
+宗
+官
+宙
+定
+客
+宣
+室
+宮
+害
+家
+容
+寂
+寃
+寄
+寅
+密
+寇
+富
+寒
+寓
+實
+審
+寫
+寬
+寶
+寸
+寺
+封
+將
+專
+尊
+對
+小
+少
+尙
+尹
+尼
+尿
+局
+居
+屈
+屋
+屍
+屎
+屛
+層
+屬
+山
+岐
+岡
+岩
+岳
+岸
+峙
+峰
+島
+峻
+峽
+崇
+崔
+崖
+崩
+嶋
+巖
+川
+州
+巢
+工
+左
+巧
+巨
+巫
+差
+己
+巷
+市
+布
+帝
+師
+帶
+常
+帽
+幕
+干
+平
+年
+幹
+幻
+幼
+幽
+庇
+序
+店
+府
+度
+座
+庫
+庭
+康
+廟
+廣
+廳
+延
+廷
+建
+廻
+弁
+式
+弑
+弓
+引
+弘
+弟
+弱
+張
+强
+弼
+彌
+彛
+形
+彬
+影
+役
+彼
+彿
+往
+征
+待
+律
+後
+徐
+徑
+得
+從
+循
+微
+德
+徹
+心
+必
+忌
+忍
+志
+忠
+思
+怡
+急
+性
+恐
+恒
+恨
+恩
+悅
+悖
+患
+悲
+情
+惑
+惟
+惠
+惡
+想
+惺
+愁
+意
+愚
+愛
+感
+愼
+慈
+態
+慕
+慣
+慧
+慾
+憂
+憤
+憺
+應
+懸
+戎
+成
+我
+戟
+戮
+戰
+戴
+戶
+房
+所
+手
+才
+打
+批
+承
+技
+抄
+把
+抗
+抱
+抽
+拇
+拓
+拘
+拙
+拜
+拾
+持
+指
+捌
+捨
+捿
+授
+掌
+排
+接
+推
+提
+揚
+揭
+援
+損
+搗
+摩
+播
+操
+擒
+擔
+擘
+據
+擧
+攘
+攝
+攬
+支
+改
+攻
+放
+政
+故
+敍
+敎
+救
+敗
+散
+敬
+整
+數
+文
+斗
+料
+斛
+斜
+斧
+斯
+新
+斷
+方
+於
+施
+旋
+族
+旗
+日
+旨
+早
+旱
+昌
+明
+易
+昔
+星
+春
+昧
+昭
+是
+時
+晉
+晋
+晩
+普
+景
+晴
+晶
+智
+暈
+暑
+暗
+暘
+曉
+曜
+曠
+曦
+曰
+曲
+書
+曹
+曼
+曾
+最
+會
+月
+有
+朋
+服
+望
+朝
+期
+木
+未
+末
+本
+朱
+朴
+李
+材
+村
+杖
+杜
+杞
+杭
+杯
+東
+松
+板
+林
+果
+枝
+枯
+枰
+枾
+柏
+柑
+柱
+栗
+校
+栢
+核
+根
+格
+桀
+桂
+案
+桎
+桑
+桓
+桔
+梁
+梏
+梓
+梗
+條
+梨
+梵
+棗
+棟
+森
+植
+椒
+楊
+楓
+楚
+業
+楮
+極
+榮
+槃
+槍
+樂
+樓
+樗
+樣
+樸
+樹
+樺
+樽
+橄
+橋
+橘
+機
+橡
+檀
+檎
+權
+欌
+欖
+次
+欲
+歌
+歐
+止
+正
+此
+步
+武
+歲
+歸
+死
+殖
+段
+殷
+殺
+殿
+毅
+母
+毒
+比
+毛
+氏
+民
+氣
+水
+永
+求
+汎
+汗
+江
+池
+沅
+沒
+沖
+沙
+沛
+河
+油
+治
+沼
+沿
+泉
+泊
+法
+泗
+泡
+波
+注
+泰
+洋
+洙
+洛
+洞
+津
+洲
+活
+派
+流
+浅
+浦
+浮
+浴
+海
+涅
+涇
+消
+涌
+液
+淑
+淡
+淨
+淫
+深
+淳
+淵
+淸
+渠
+渡
+游
+渾
+湖
+湯
+源
+溪
+溫
+溶
+滄
+滅
+滋
+滯
+滿
+漁
+漆
+漢
+漫
+漸
+潑
+潤
+潭
+澄
+澎
+澤
+澳
+澹
+濁
+濕
+濟
+濤
+濯
+瀋
+瀝
+灣
+火
+灰
+灸
+災
+炎
+炭
+点
+烈
+烏
+烙
+焚
+無
+焦
+然
+煌
+煎
+照
+煬
+煮
+熟
+熱
+燁
+燈
+燔
+燕
+燥
+燧
+燮
+爲
+爵
+父
+片
+版
+牌
+牛
+牝
+牟
+牡
+物
+特
+犧
+犬
+狀
+狗
+猥
+猩
+猪
+獨
+獵
+獸
+獻
+玄
+玉
+王
+玲
+珍
+珠
+珪
+班
+現
+球
+理
+琴
+瑞
+瑟
+瑪
+璃
+璋
+璽
+瓜
+瓦
+甑
+甘
+生
+産
+用
+甫
+田
+由
+甲
+申
+男
+界
+畏
+留
+畜
+畢
+略
+番
+異
+畵
+當
+畸
+疏
+疑
+疫
+疹
+疼
+病
+症
+痔
+痛
+痺
+瘀
+瘍
+瘡
+療
+癌
+癖
+登
+發
+白
+百
+的
+皆
+皇
+皮
+盂
+盆
+益
+盛
+盜
+盟
+盡
+盤
+盧
+目
+直
+相
+省
+看
+眞
+眼
+睡
+督
+瞋
+矢
+矣
+知
+短
+石
+破
+碍
+碑
+磁
+磨
+磬
+示
+社
+祇
+祖
+祝
+神
+祥
+祭
+祺
+禁
+禅
+禍
+福
+禦
+禪
+禮
+禹
+禽
+禾
+秀
+私
+秉
+秋
+科
+秘
+秤
+秦
+秩
+移
+稀
+稗
+種
+稱
+稷
+稼
+稽
+穀
+穆
+積
+空
+窮
+竅
+立
+章
+童
+竭
+端
+竹
+笑
+符
+第
+筆
+等
+筍
+答
+策
+箋
+箕
+管
+箱
+節
+篇
+簡
+米
+粉
+粘
+粥
+精
+糖
+糞
+系
+紀
+紂
+約
+紅
+紋
+純
+紙
+級
+素
+索
+紫
+紬
+累
+細
+紳
+終
+組
+結
+絡
+統
+絲
+絶
+絹
+經
+綠
+維
+綱
+網
+綸
+綽
+緖
+線
+緣
+緯
+縣
+縱
+總
+織
+繡
+繩
+繪
+繭
+纂
+續
+罕
+置
+罰
+羅
+羊
+美
+群
+義
+羽
+翁
+習
+翟
+老
+考
+者
+而
+耐
+耕
+耳
+聃
+聖
+聞
+聰
+聲
+職
+肇
+肉
+肖
+肝
+股
+肥
+育
+肺
+胃
+胎
+胚
+胞
+胡
+胥
+能
+脂
+脈
+脚
+脛
+脣
+脩
+脫
+脯
+脾
+腋
+腎
+腫
+腸
+腹
+膜
+膠
+膨
+膽
+臆
+臟
+臣
+臥
+臨
+自
+至
+致
+臺
+臼
+臾
+與
+興
+舊
+舌
+舍
+舒
+舜
+舟
+般
+船
+艦
+良
+色
+芋
+花
+芳
+芽
+苑
+苔
+苕
+苛
+苞
+若
+苦
+英
+茂
+茵
+茶
+茹
+荀
+荇
+草
+荒
+荷
+莊
+莫
+菊
+菌
+菜
+菩
+菫
+華
+菴
+菽
+萊
+萍
+萬
+落
+葉
+著
+葛
+董
+葬
+蒙
+蒜
+蒲
+蒸
+蒿
+蓮
+蔓
+蔘
+蔡
+蔬
+蕃
+蕉
+蕓
+薄
+薑
+薛
+薩
+薪
+薺
+藏
+藝
+藤
+藥
+藩
+藻
+蘆
+蘇
+蘊
+蘚
+蘭
+虎
+處
+虛
+虞
+虹
+蜀
+蜂
+蜜
+蝕
+蝶
+融
+蟬
+蟲
+蠶
+蠻
+血
+衆
+行
+術
+衛
+衡
+衣
+表
+袁
+裔
+裕
+裙
+補
+製
+複
+襄
+西
+要
+見
+視
+親
+覺
+觀
+角
+解
+言
+訂
+訊
+訓
+託
+記
+訣
+設
+診
+註
+評
+詩
+話
+詵
+誅
+誌
+認
+誕
+語
+誠
+誤
+誥
+誦
+說
+調
+談
+諍
+論
+諡
+諫
+諭
+諸
+謙
+講
+謝
+謠
+證
+識
+譚
+譜
+譯
+議
+護
+讀
+變
+谷
+豆
+豊
+豚
+象
+豪
+豫
+貝
+貞
+財
+貧
+貨
+貪
+貫
+貴
+貸
+費
+資
+賊
+賓
+賞
+賢
+賣
+賦
+質
+贍
+赤
+赫
+走
+起
+超
+越
+趙
+趣
+趨
+足
+趾
+跋
+跡
+路
+踏
+蹟
+身
+躬
+車
+軍
+軒
+軟
+載
+輓
+輕
+輪
+輯
+輸
+輻
+輿
+轅
+轉
+辨
+辭
+辯
+辰
+農
+近
+迦
+述
+追
+逆
+透
+逐
+通
+逝
+造
+逢
+連
+進
+逵
+遂
+遊
+運
+遍
+過
+道
+達
+遠
+遡
+適
+遷
+選
+遺
+遽
+還
+邊
+邑
+那
+邪
+郞
+郡
+部
+都
+鄒
+鄕
+鄭
+鄲
+配
+酒
+酸
+醉
+醫
+醯
+釋
+里
+重
+野
+量
+釐
+金
+針
+鈍
+鈴
+鉞
+銀
+銅
+銘
+鋼
+錄
+錢
+錦
+鎭
+鏡
+鐘
+鐵
+鑑
+鑛
+長
+門
+閃
+開
+間
+閔
+閣
+閥
+閭
+閻
+闕
+關
+阪
+防
+阿
+陀
+降
+限
+陝
+院
+陰
+陳
+陵
+陶
+陸
+陽
+隆
+隊
+隋
+階
+際
+障
+隣
+隨
+隱
+隷
+雀
+雄
+雅
+集
+雇
+雌
+雖
+雙
+雜
+離
+難
+雨
+雪
+雲
+電
+霜
+露
+靈
+靑
+靖
+靜
+非
+面
+革
+靴
+鞏
+韓
+音
+韶
+韻
+順
+須
+頊
+頌
+領
+頭
+顔
+願
+顚
+類
+顯
+風
+飛
+食
+飢
+飮
+飯
+飾
+養
+餓
+餘
+首
+香
+馨
+馬
+駒
+騫
+騷
+驕
+骨
+骸
+髓
+體
+高
+髥
+髮
+鬪
+鬱
+鬼
+魏
+魔
+魚
+魯
+鮮
+鰍
+鰐
+鳥
+鳧
+鳳
+鴨
+鵲
+鶴
+鷄
+鷹
+鹽
+鹿
+麗
+麥
+麻
+黃
+黑
+默
+點
+黨
+鼎
+齊
+齋
+齒
+龍
+龜
+가
+각
+간
+갇
+갈
+갉
+감
+갑
+값
+갓
+갔
+강
+갖
+갗
+같
+갚
+갛
+개
+객
+갠
+갤
+갬
+갭
+갯
+갰
+갱
+갸
+걀
+걔
+걘
+거
+걱
+건
+걷
+걸
+검
+겁
+것
+겄
+겅
+겆
+겉
+겊
+겋
+게
+겐
+겔
+겟
+겠
+겡
+겨
+격
+겪
+견
+결
+겸
+겹
+겻
+겼
+경
+곁
+계
+곕
+곗
+고
+곡
+곤
+곧
+골
+곪
+곬
+곯
+곰
+곱
+곳
+공
+곶
+과
+곽
+관
+괄
+괌
+광
+괘
+괜
+괭
+괴
+괸
+굉
+교
+구
+국
+군
+굳
+굴
+굵
+굶
+굼
+굽
+굿
+궁
+궂
+궈
+권
+궐
+궜
+궝
+궤
+귀
+귄
+귈
+귓
+규
+균
+귤
+그
+극
+근
+글
+긁
+금
+급
+긋
+긍
+기
+긴
+길
+김
+깁
+깃
+깅
+깊
+까
+깍
+깎
+깐
+깔
+깜
+깝
+깟
+깡
+깥
+깨
+깬
+깰
+깻
+깼
+깽
+꺄
+꺼
+꺽
+꺾
+껀
+껄
+껌
+껍
+껏
+껐
+껑
+께
+껴
+꼈
+꼍
+꼐
+꼬
+꼭
+꼴
+꼼
+꼽
+꼿
+꽁
+꽂
+꽃
+꽉
+꽝
+꽤
+꽥
+꾀
+꾜
+꾸
+꾹
+꾼
+꿀
+꿇
+꿈
+꿉
+꿋
+꿍
+꿎
+꿔
+꿨
+꿩
+꿰
+꿴
+뀄
+뀌
+뀐
+뀔
+뀜
+뀝
+끄
+끈
+끊
+끌
+끓
+끔
+끕
+끗
+끙
+끝
+끼
+끽
+낀
+낄
+낌
+낍
+낏
+낑
+나
+낙
+낚
+난
+낟
+날
+낡
+남
+납
+낫
+났
+낭
+낮
+낯
+낱
+낳
+내
+낵
+낸
+낼
+냄
+냅
+냇
+냈
+냉
+냐
+냔
+냘
+냥
+너
+넉
+넋
+넌
+널
+넓
+넘
+넙
+넛
+넜
+넝
+넣
+네
+넥
+넨
+넬
+넴
+넵
+넷
+넸
+넹
+녀
+녁
+년
+념
+녔
+녕
+녘
+녜
+노
+녹
+논
+놀
+놈
+놋
+농
+높
+놓
+놔
+놨
+뇌
+뇨
+뇩
+뇽
+누
+눅
+눈
+눌
+눔
+눕
+눗
+눠
+눴
+뉘
+뉜
+뉩
+뉴
+늄
+늅
+늉
+느
+늑
+는
+늘
+늙
+늠
+늡
+능
+늦
+늪
+늬
+니
+닉
+닌
+닐
+님
+닙
+닛
+닝
+닢
+다
+닥
+닦
+단
+닫
+달
+닭
+닮
+닯
+닳
+담
+답
+닷
+당
+닻
+닿
+대
+댁
+댄
+댈
+댐
+댑
+댓
+댔
+댕
+댜
+더
+덕
+덖
+던
+덜
+덟
+덤
+덥
+덧
+덩
+덫
+덮
+데
+덱
+덴
+델
+뎀
+뎃
+뎅
+뎌
+뎠
+뎨
+도
+독
+돈
+돋
+돌
+돔
+돕
+돗
+동
+돛
+돝
+돼
+됐
+되
+된
+될
+됨
+됩
+됴
+두
+둑
+둔
+둘
+둠
+둡
+둣
+둥
+둬
+뒀
+뒤
+뒬
+뒷
+뒹
+듀
+듈
+듐
+드
+득
+든
+듣
+들
+듦
+듬
+듭
+듯
+등
+듸
+디
+딕
+딘
+딛
+딜
+딤
+딥
+딧
+딨
+딩
+딪
+따
+딱
+딴
+딸
+땀
+땄
+땅
+때
+땐
+땔
+땜
+땝
+땠
+땡
+떠
+떡
+떤
+떨
+떫
+떰
+떱
+떳
+떴
+떵
+떻
+떼
+떽
+뗀
+뗄
+뗍
+뗏
+뗐
+뗑
+또
+똑
+똘
+똥
+뙤
+뚜
+뚝
+뚤
+뚫
+뚱
+뛰
+뛴
+뛸
+뜀
+뜁
+뜨
+뜩
+뜬
+뜯
+뜰
+뜸
+뜻
+띄
+띈
+띌
+띔
+띕
+띠
+띤
+띨
+띱
+띵
+라
+락
+란
+랄
+람
+랍
+랏
+랐
+랑
+랒
+랗
+래
+랙
+랜
+랠
+램
+랩
+랫
+랬
+랭
+랴
+략
+량
+러
+럭
+런
+럴
+럼
+럽
+럿
+렀
+렁
+렇
+레
+렉
+렌
+렐
+렘
+렙
+렛
+렝
+려
+력
+련
+렬
+렴
+렵
+렷
+렸
+령
+례
+로
+록
+론
+롤
+롬
+롭
+롯
+롱
+롸
+롹
+뢰
+뢴
+뢸
+룃
+료
+룐
+룡
+루
+룩
+룬
+룰
+룸
+룹
+룻
+룽
+뤄
+뤘
+뤼
+류
+륙
+륜
+률
+륨
+륭
+르
+륵
+른
+를
+름
+릅
+릇
+릉
+릎
+리
+릭
+린
+릴
+림
+립
+릿
+링
+마
+막
+만
+많
+맏
+말
+맑
+맘
+맙
+맛
+망
+맞
+맡
+맣
+매
+맥
+맨
+맬
+맴
+맵
+맷
+맸
+맹
+맺
+먀
+먁
+머
+먹
+먼
+멀
+멈
+멋
+멍
+멎
+메
+멕
+멘
+멜
+멤
+멥
+멧
+멩
+며
+멱
+면
+멸
+몄
+명
+몇
+모
+목
+몫
+몬
+몰
+몸
+몹
+못
+몽
+뫼
+묘
+무
+묵
+묶
+문
+묻
+물
+묽
+뭄
+뭅
+뭇
+뭉
+뭍
+뭏
+뭐
+뭔
+뭘
+뭡
+뭣
+뮈
+뮌
+뮐
+뮤
+뮬
+므
+믈
+믐
+미
+믹
+민
+믿
+밀
+밈
+밉
+밋
+밌
+밍
+및
+밑
+바
+박
+밖
+반
+받
+발
+밝
+밟
+밤
+밥
+밧
+방
+밭
+배
+백
+밴
+밸
+뱀
+뱁
+뱃
+뱄
+뱅
+뱉
+뱍
+뱐
+버
+벅
+번
+벌
+범
+법
+벗
+벙
+벚
+베
+벡
+벤
+벨
+벰
+벱
+벳
+벵
+벼
+벽
+변
+별
+볍
+볏
+볐
+병
+볕
+보
+복
+볶
+본
+볼
+봄
+봅
+봇
+봉
+봐
+봤
+뵈
+뵐
+뵙
+부
+북
+분
+붇
+불
+붉
+붐
+붓
+붕
+붙
+뷔
+뷰
+뷴
+뷸
+브
+븐
+블
+비
+빅
+빈
+빌
+빔
+빕
+빗
+빙
+빚
+빛
+빠
+빡
+빤
+빨
+빳
+빴
+빵
+빻
+빼
+빽
+뺀
+뺄
+뺌
+뺏
+뺐
+뺑
+뺨
+뻐
+뻑
+뻔
+뻗
+뻘
+뻣
+뻤
+뻥
+뻬
+뼈
+뼉
+뼘
+뽀
+뽈
+뽐
+뽑
+뽕
+뾰
+뿌
+뿍
+뿐
+뿔
+뿜
+쁘
+쁜
+쁠
+쁨
+삐
+삔
+삘
+사
+삭
+삯
+산
+살
+삵
+삶
+삼
+삽
+삿
+샀
+상
+샅
+새
+색
+샌
+샐
+샘
+샙
+샛
+샜
+생
+샤
+샨
+샬
+샴
+샵
+샷
+샹
+서
+석
+섞
+선
+섣
+설
+섬
+섭
+섯
+섰
+성
+섶
+세
+섹
+센
+셀
+셈
+셉
+셋
+셌
+셍
+셔
+션
+셜
+셨
+셰
+셴
+셸
+소
+속
+손
+솔
+솜
+솝
+솟
+송
+솥
+쇄
+쇠
+쇤
+쇳
+쇼
+숀
+숄
+숍
+수
+숙
+순
+숟
+술
+숨
+숩
+숫
+숭
+숯
+숱
+숲
+숴
+쉐
+쉘
+쉬
+쉭
+쉰
+쉴
+쉼
+쉽
+슈
+슐
+슘
+슛
+슝
+스
+슥
+슨
+슬
+슭
+슴
+습
+슷
+승
+시
+식
+신
+싣
+실
+싫
+심
+십
+싯
+싱
+싶
+싸
+싹
+싼
+쌀
+쌈
+쌉
+쌌
+쌍
+쌓
+쌔
+쌘
+쌩
+써
+썩
+썬
+썰
+썸
+썹
+썼
+썽
+쎄
+쎈
+쏘
+쏙
+쏜
+쏟
+쏠
+쏭
+쏴
+쐈
+쐐
+쐬
+쑤
+쑥
+쑨
+쒀
+쒔
+쓰
+쓱
+쓴
+쓸
+씀
+씁
+씌
+씨
+씩
+씬
+씰
+씸
+씹
+씻
+씽
+아
+악
+안
+앉
+않
+알
+앎
+앓
+암
+압
+앗
+았
+앙
+앞
+애
+액
+앤
+앨
+앰
+앱
+앳
+앴
+앵
+야
+약
+얀
+얄
+얇
+얌
+얍
+얏
+양
+얕
+얗
+얘
+얜
+어
+억
+언
+얹
+얻
+얼
+얽
+엄
+업
+없
+엇
+었
+엉
+엊
+엌
+엎
+에
+엑
+엔
+엘
+엠
+엡
+엣
+엥
+여
+역
+엮
+연
+열
+엷
+염
+엽
+엾
+엿
+였
+영
+옅
+옆
+옇
+예
+옌
+옐
+옙
+옛
+오
+옥
+온
+올
+옭
+옮
+옳
+옴
+옵
+옷
+옹
+옻
+와
+왁
+완
+왈
+왑
+왓
+왔
+왕
+왜
+왠
+왱
+외
+왼
+요
+욕
+욘
+욜
+욤
+용
+우
+욱
+운
+울
+움
+웁
+웃
+웅
+워
+웍
+원
+월
+웜
+웠
+웡
+웨
+웬
+웰
+웸
+웹
+위
+윅
+윈
+윌
+윔
+윗
+윙
+유
+육
+윤
+율
+윱
+윳
+융
+으
+윽
+은
+을
+읊
+음
+읍
+응
+의
+읜
+읠
+이
+익
+인
+일
+읽
+잃
+임
+입
+잇
+있
+잉
+잊
+잎
+자
+작
+잔
+잖
+잘
+잠
+잡
+잣
+잤
+장
+잦
+재
+잭
+잰
+잴
+잽
+잿
+쟀
+쟁
+쟈
+쟉
+쟤
+저
+적
+전
+절
+젊
+점
+접
+젓
+정
+젖
+제
+젝
+젠
+젤
+젬
+젭
+젯
+져
+젼
+졀
+졌
+졍
+조
+족
+존
+졸
+좀
+좁
+종
+좇
+좋
+좌
+좍
+좽
+죄
+죠
+죤
+주
+죽
+준
+줄
+줌
+줍
+줏
+중
+줘
+줬
+쥐
+쥔
+쥘
+쥬
+쥴
+즈
+즉
+즌
+즐
+즘
+즙
+증
+지
+직
+진
+짇
+질
+짊
+짐
+집
+짓
+징
+짖
+짙
+짚
+짜
+짝
+짠
+짢
+짤
+짧
+짬
+짭
+짰
+짱
+째
+짹
+짼
+쨀
+쨉
+쨋
+쨌
+쨍
+쩄
+쩌
+쩍
+쩐
+쩔
+쩜
+쩝
+쩡
+쩨
+쪄
+쪘
+쪼
+쪽
+쪾
+쫀
+쫄
+쫑
+쫓
+쫙
+쬐
+쭈
+쭉
+쭐
+쭙
+쯔
+쯤
+쯧
+찌
+찍
+찐
+찔
+찜
+찝
+찡
+찢
+찧
+차
+착
+찬
+찮
+찰
+참
+찹
+찻
+찼
+창
+찾
+채
+책
+챈
+챌
+챔
+챕
+챗
+챘
+챙
+챠
+챤
+처
+척
+천
+철
+첨
+첩
+첫
+청
+체
+첵
+첸
+첼
+쳄
+쳇
+쳉
+쳐
+쳔
+쳤
+초
+촉
+촌
+촘
+촛
+총
+촨
+촬
+최
+쵸
+추
+축
+춘
+출
+춤
+춥
+춧
+충
+춰
+췄
+췌
+취
+췬
+츄
+츠
+측
+츨
+츰
+층
+치
+칙
+친
+칠
+칡
+침
+칩
+칫
+칭
+카
+칵
+칸
+칼
+캄
+캅
+캇
+캉
+캐
+캔
+캘
+캠
+캡
+캣
+캤
+캥
+캬
+커
+컥
+컨
+컫
+컬
+컴
+컵
+컷
+컸
+컹
+케
+켄
+켈
+켐
+켓
+켕
+켜
+켠
+켤
+켭
+켯
+켰
+코
+콕
+콘
+콜
+콤
+콥
+콧
+콩
+콰
+콱
+콴
+콸
+쾅
+쾌
+쾡
+쾨
+쾰
+쿄
+쿠
+쿡
+쿤
+쿨
+쿰
+쿵
+쿼
+퀀
+퀄
+퀘
+퀭
+퀴
+퀵
+퀸
+퀼
+큐
+큘
+크
+큰
+클
+큼
+큽
+키
+킥
+킨
+킬
+킴
+킵
+킷
+킹
+타
+탁
+탄
+탈
+탉
+탐
+탑
+탓
+탔
+탕
+태
+택
+탠
+탤
+탬
+탭
+탯
+탰
+탱
+터
+턱
+턴
+털
+텀
+텁
+텃
+텄
+텅
+테
+텍
+텐
+텔
+템
+텝
+텡
+텨
+톈
+토
+톡
+톤
+톨
+톰
+톱
+톳
+통
+퇴
+툇
+투
+툭
+툰
+툴
+툼
+퉁
+퉈
+퉜
+튀
+튄
+튈
+튕
+튜
+튠
+튤
+튬
+트
+특
+튼
+튿
+틀
+틈
+틉
+틋
+틔
+티
+틱
+틴
+틸
+팀
+팁
+팅
+파
+팍
+팎
+판
+팔
+팜
+팝
+팟
+팠
+팡
+팥
+패
+팩
+팬
+팰
+팸
+팻
+팼
+팽
+퍼
+퍽
+펀
+펄
+펌
+펍
+펐
+펑
+페
+펙
+펜
+펠
+펨
+펩
+펫
+펭
+펴
+편
+펼
+폄
+폈
+평
+폐
+포
+폭
+폰
+폴
+폼
+폿
+퐁
+표
+푭
+푸
+푹
+푼
+풀
+품
+풋
+풍
+퓨
+퓬
+퓰
+퓸
+프
+픈
+플
+픔
+픕
+피
+픽
+핀
+필
+핌
+핍
+핏
+핑
+하
+학
+한
+할
+핥
+함
+합
+핫
+항
+해
+핵
+핸
+핼
+햄
+햅
+햇
+했
+행
+햐
+향
+헀
+허
+헉
+헌
+헐
+험
+헙
+헛
+헝
+헤
+헥
+헨
+헬
+헴
+헵
+헷
+헹
+혀
+혁
+현
+혈
+혐
+협
+혓
+혔
+형
+혜
+호
+혹
+혼
+홀
+홈
+홉
+홋
+홍
+홑
+화
+확
+환
+활
+홧
+황
+홰
+홱
+횃
+회
+획
+횝
+횟
+횡
+효
+후
+훅
+훈
+훌
+훑
+훔
+훗
+훤
+훨
+훼
+휄
+휑
+휘
+휙
+휜
+휠
+휩
+휭
+휴
+휼
+흄
+흉
+흐
+흑
+흔
+흘
+흙
+흠
+흡
+흣
+흥
+흩
+희
+흰
+흽
+히
+힉
+힌
+힐
+힘
+힙
+힝
+車
+滑
+金
+奈
+羅
+洛
+卵
+欄
+蘭
+郎
+來
+盧
+老
+魯
+綠
+鹿
+論
+雷
+樓
+縷
+凌
+樂
+不
+參
+葉
+沈
+若
+兩
+凉
+梁
+呂
+女
+廬
+麗
+黎
+曆
+歷
+戀
+蓮
+連
+列
+烈
+裂
+念
+獵
+靈
+領
+例
+禮
+醴
+惡
+尿
+料
+遼
+龍
+暈
+柳
+流
+類
+六
+陸
+倫
+律
+栗
+利
+李
+梨
+理
+離
+燐
+林
+臨
+立
+茶
+切
+宅
+ 
diff --git a/ppocr/utils/dict/latex_ocr_tokenizer.json b/ppocr/utils/dict/latex_ocr_tokenizer.json
new file mode 100644
index 0000000..e8fd4f6
--- /dev/null
+++ b/ppocr/utils/dict/latex_ocr_tokenizer.json
@@ -0,0 +1 @@
+{"version":"1.0","truncation":null,"padding":null,"added_tokens":[{"id":0,"special":true,"content":"[PAD]","single_word":false,"lstrip":false,"rstrip":false,"normalized":false},{"id":1,"special":true,"content":"[BOS]","single_word":false,"lstrip":false,"rstrip":false,"normalized":false},{"id":2,"special":true,"content":"[EOS]","single_word":false,"lstrip":false,"rstrip":false,"normalized":false}],"normalizer":null,"pre_tokenizer":{"type":"ByteLevel","add_prefix_space":false,"trim_offsets":true},"post_processor":null,"decoder":null,"model":{"dropout":null,"unk_token":null,"continuing_subword_prefix":null,"end_of_word_suffix":null,"fuse_unk":false,"vocab":{"[PAD]":0,"[BOS]":1,"[EOS]":2,"!":3,"\"":4,"#":5,"$":6,"&":7,"'":8,"(":9,")":10,"*":11,"+":12,",":13,"-":14,".":15,"/":16,"0":17,"1":18,"2":19,"3":20,"4":21,"5":22,"6":23,"7":24,"8":25,"9":26,":":27,";":28,"<":29,"=":30,">":31,"?":32,"@":33,"A":34,"B":35,"C":36,"D":37,"E":38,"F":39,"G":40,"H":41,"I":42,"J":43,"K":44,"L":45,"M":46,"N":47,"O":48,"P":49,"Q":50,"R":51,"S":52,"T":53,"U":54,"V":55,"W":56,"X":57,"Y":58,"Z":59,"[":60,"\\":61,"]":62,"^":63,"_":64,"`":65,"a":66,"b":67,"c":68,"d":69,"e":70,"f":71,"g":72,"h":73,"i":74,"j":75,"k":76,"l":77,"m":78,"n":79,"o":80,"p":81,"q":82,"r":83,"s":84,"t":85,"u":86,"v":87,"w":88,"x":89,"y":90,"z":91,"{":92,"|":93,"}":94,"~":95,"½":96,"¿":97,"ï":98,"Ċ":99,"č":100,"Ġ":101,"Ġ}":102,"Ġ{":103,"Ġ\\":104,"Ġ_":105,"Ġ^":106,"Ġ2":107,"Ġ)":108,"Ġ(":109,"Ġ1":110,"ra":111,"Ġ=":112,"Ġ-":113,"čĊ":114,"Ġ,":115,"fra":116,"frac":117,"Ġ+":118,"ma":119,"ta":120,"ig":121,"Ġ0":122,"ar":123,"al":124,"le":125,"Ġi":126,"th":127,"Ġx":128,"ft":129,"igh":130,"me":131,"righ":132,"math":133,"Ġn":134,"Ġ.":135,"Ġ\\,":136,"in":137,"ph":138,"Ġd":139,"left":140,"Ġa":141,"right":142,"am":143,"eta":144,"ti":145,"Ġm":146,"mu":147,"Ġ3":148,"Ġk":149,"Ġt":150,"Ġe":151,"Ġr":152,"Ġ&":153,"Ġc":154,"Ġp":155,"si":156,"rm":157,"de":158,"mathrm":159,"Ġ4":160,"Ġs":161,"pr":162,"Ġ~":163,"pha":164,"Ġl":165,"alpha":166,"da":167,"ĠA":168,"Ġ\\;":169,"ot":170,"pi":171,"par":172,"tial":173,"partial":174,"ime":175,"prime":176,"psi":177,"dot":178,"Ġj":179,"Ġb":180,"Ġf":181,"lta":182,"Ġ|":183,"amma":184,"bda":185,"ambda":186,"phi":187,"Ġq":188,"bf":189,"Ġg":190,"nu":191,"Ġz":192,"ray":193,"array":194,"ĠM":195,"ĠT":196,"Ġy":197,"cal":198,"bar":199,"ĠN":200,"igma":201,"ĠR":202,"rt":203,"lambda":204,"int":205,"ĠB":206,"ve":207,"ng":208,"qu":209,"ĠL":210,"Ġ/":211,"lo":212,"beta":213,"ngle":214,"Ġu":215,"delta":216,"sq":217,"sqrt":218,"theta":219,"Ġ\\\\":220,"gamma":221,"Ġ]":222,"sigma":223,"ga":224,"mega":225,"ĠD":226,"ĠF":227,"Ġ[":228,"ĠS":229,"mathbf":230,"su":231,"ĠP":232,"lon":233,"Ġv":234,"sum":235,"psilon":236,"ĠV":237,"ĠC":238,"cdot":239,"epsilon":240,"at":241,"hat":242,"ad":243,"quad":244,"Ġh":245,"ho":246,"rho":247,"hi":248,"to":249,"ĠE":250,"la":251,"ĠH":252,"lde":253,"tilde":254,"ĠQ":255,"Ġ5":256,"var":257,"ĠX":258,"ĠG":259,"be":260,"nd":261,"omega":262,"end":263,"gin":264,"begin":265,"tau":266,"Ġ6":267,"na":268,"vec":269,"ĠI":270,"Ġo":271,"rangle":272,"Ġ*":273,"De":274,"Delta":275,"Gamma":276,"pe":277,"fty":278,"infty":279,"ĠK":280,"xi":281,"Ġ8":282,"ow":283,"ĠJ":284,"ĠU":285,"row":286,"tar":287,"ge":288,"Phi":289,"ap":290,"ĠW":291,"co":292,"mes":293,"times":294,"sin":295,"ĠZ":296,"langle":297,"ope":298,"rna":299,"rato":300,"operato":301,"rname":302,"operatorname":303,"tarrow":304,"lin":305,"line":306,"varphi":307,"pm":308,"rline":309,"Lambda":310,"Ġ\\!":311,"Ġ;":312,"dots":313,"cos":314,"Ġw":315,"rightarrow":316,"big":317,"chi":318,"ove":319,"Ġ\\}":320,"overline":321,"Ġ7":322,"ex":323,"pa":324,"st":325,"pro":326,"qquad":327,"iv":328,"equ":329,"equiv":330,"ĠO":331,"ln":332,"Omega":333,"ll":334,"Ġ9":335,"kap":336,"kappa":337,"Big":338,"Ġ\\{":339,"dag":340,"ĠY":341,"\\{":342,"varepsilon":343,"cdots":344,"Ġ:":345,"mathcal":346,"Psi":347,"Ġ>":348,"bo":349,"bol":350,"Ġ<":351,"ger":352,"dagger":353,"ldots":354,"ell":355,"bla":356,"nabla":357,"exp":358,"yle":359,"style":360,"zeta":361,"Sigma":362,"wi":363,"wide":364,"sim":365,"leq":366,"Ġ!":367,"bigg":368,"mathb":369,"mathbb":370,"Ġ\\:":371,"hbar":372,"otimes":373,"bold":374,"\\}":375,"mi":376,"prox":377,"approx":378,"Pi":379,"log":380,"mid":381,"sp":382,"vert":383,"di":384,"prod":385,"per":386,"perp":387,"ystyle":388,"laystyle":389,"splaystyle":390,"displaystyle":391,"meq":392,"simeq":393,"ed":394,"wed":395,"wedge":396,"widetilde":397,"sy":398,"sym":399,"symbol":400,"boldsymbol":401,"ck":402,"tex":403,"text":404,"ri":405,"Th":406,"Theta":407,"geq":408,"se":409,"eq":410,"nde":411,"unde":412,"tan":413,"sc":414,"ast":415,"rc":416,"set":417,"pt":418,"widehat":419,"ci":420,"circ":421,"re":422,"ript":423,"script":424,"underline":425,"Ġ\\|":426,"rel":427,"neq":428,"sta":429,"stack":430,"stackrel":431,"sinh":432,"op":433,"us":434,"cosh":435,"Bigg":436,"ce":437,"textstyle":438,"star":439,"not":440,"frak":441,"mathfrak":442,"mp":443,"biggr":444,"lus":445,"oplus":446,"vartheta":447,"biggl":448,"Bigr":449,"bra":450,"Bigl":451,"fo":452,"sf":453,"sub":454,"subset":455,"ngrightarrow":456,"ec":457,"boldmath":458,"rall":459,"forall":460,"scriptstyle":461,"ect":462,"parrow":463,"uparrow":464,"bj":465,"bject":466,"pto":467,"propto":468,"Ġ'":469,"longrightarrow":470,"bigl":471,"bigr":472,"oint":473,"ps":474,"maps":475,"mapsto":476,"om":477,"lle":478,"\\|":479,"ddot":480,"cu":481,"bin":482,"binom":483,"vdots":484,"angle":485,"leftrightarrow":486,"over":487,"or":488,"mathsf":489,"cup":490,"brace":491,"no":492,"arc":493,"flo":494,"floor":495,"tri":496,"triangle":497,"Xi":498,"cot":499,"bot":500,"cong":501,"it":502,"mbe":503,"numbe":504,"nonumbe":505,"nonumber":506,"cap":507,"Righ":508,"Rightarrow":509,"ze":510,"size":511,"textrm":512,"ne":513,"arctan":514,"ralle":515,"paralle":516,"parallel":517,"cfrac":518,"Ġ--":519,"object":520,"ĠObject":521,"brack":522,"sh":523,"arrow":524,"own":525,"varrho":526,"subseteq":527,"rbrace":528,"textbf":529,"imath":530,"od":531,"down":532,"he":533,"land":534,"scriptscriptstyle":535,"scriptsize":536,"che":537,"check":538,"sla":539,"overrightarrow":540,"downarrow":541,"Biggl":542,"gg":543,"nto":544,"phanto":545,"phantom":546,"exi":547,"hline":548,"sts":549,"exists":550,"Biggr":551,"bu":552,"rfloor":553,"ddots":554,"io":555,"iota":556,"llet":557,"bullet":558,"colon":559,"inus":560,"Upsilon":561,"lfloor":562,"lbrack":563,"underbrace":564,"neg":565,"Im":566,"mathit":567,"tin":568,"tiny":569,"jmath":570,"lef":571,"slash":572,"vee":573,"minus":574,"setminus":575,"Re":576,"iint":577,"leftarrow":578,"Ve":579,"Vert":580,"atop":581,"sup":582,"bigcup":583,"wp":584,"dim":585,"sec":586,"supset":587,"Lo":588,"lor":589,"pmod":590,"mod":591,"bigoplus":592,"il":593,"bmod":594,"coth":595,"Le":596,"ftrightarrow":597,"Leftrightarrow":598,"ngleftrightarrow":599,"sma":600,"upsilon":601,"\\,":602,"csc":603,"eph":604,"aleph":605,"bigwedge":606,"arcsin":607,"small":608,"odot":609,"overset":610,"rbrack":611,"mit":612,"lbrace":613,"li":614,"arp":615,"arge":616,"Ġ\\#":617,"bre":618,"textsf":619,"Longrightarrow":620,"breve":621,"em":622,"yset":623,"varpi":624,"ptyset":625,"emptyset":626,"ff":627,"iff":628,"nt":629,"er":630,"lap":631,"lnot":632,"dash":633,"under":634,"slant":635,"arg":636,"underset":637,"Bo":638,"Box":639,"Ġ\"":640,"spa":641,"space":642,"deg":643,"iiint":644,"oo":645,"otnot":646,"footnot":647,"arpoo":648,"footnote":649,"rlap":650,"es":651,"imp":652,"sb":653,"te":654,"bigtriangle":655,"lies":656,"implies":657,"\\;":658,"ker":659,"footnotesize":660,"tharpoo":661,"up":662,"acu":663,"acute":664,"longleftrightarrow":665,"eil":666,"lce":667,"rceil":668,"lceil":669,"vphantom":670,"en":671,"thin":672,"ack":673,"back":674,"tt":675,"backslash":676,"xrightarrow":677,"vdash":678,"top":679,"rightharpoo":680,"varsigma":681,"Longleftrightarrow":682,"mathop":683,"large":684,"bigcap":685,"leqslant":686,"Ġ`":687,"overbrace":688,"nup":689,"rightharpoonup":690,"bigotimes":691,"triangleq":692,"Large":693,"ru":694,"null":695,"bigtriangleup":696,"varno":697,"thing":698,"varnothing":699,"doteq":700,"Ġ\\_":701,"overleftarrow":702,"hf":703,"bigstar":704,"enspace":705,"\\!":706,"stru":707,"strut":708,"ominus":709,"div":710,"ond":711,"amond":712,"ddagger":713,"Ġcm":714,"ni":715,"sk":716,"diamond":717,"rVert":718,"prot":719,"protect":720,"ip":721,"varDelta":722,"notin":723,"skip":724,"lVert":725,"Ġ\\/":726,"dotsc":727,"ill":728,"ule":729,"\\:":730,"hfill":731,"krightarrow":732,"okrightarrow":733,"hookrightarrow":734,"sharp":735,"Vdash":736,"bigvee":737,"subsetneq":738,"supseteq":739,"Ġ?":740,"ngmapsto":741,"longmapsto":742,"cdotp":743,"geqslant":744,"bigtriangledown":745,"dotsb":746,"lim":747,"fl":748,"triangleleft":749,"flat":750,"sl":751,"box":752,"Ġ---":753,"sqcup":754,"jlim":755,"ls":756,"mo":757,"dels":758,"ref":759,"models":760,"tag":761,"Pr":762,"mal":763,"ou":764,"llap":765,"thinspace":766,"enskip":767,"Vec":768,"ebox":769,"kebox":770,"nor":771,"rd":772,"squ":773,"vline":774,"¿½":775,"�":776,"Ġ�":777,"makebox":778,"surd":779,"normal":780,"are":781,"square":782,"pou":783,"mathrel":784,"varOmega":785,"nds":786,"smallsetminus":787,"pounds":788,"ns":789,"ss":790,"smi":791,"mathor":792,"rightlef":793,"textup":794,"tharpoons":795,"smile":796,"mathord":797,"rightleftharpoons":798,"cc":799,"Ġ\\-":800,"succ":801,"ftarrow":802,"rtimes":803,"det":804,"prec":805,"texttt":806,"oslash":807,"Ġ\\&":808,"arrowvert":809,"lg":810,"Ġmm":811,"inter":812,"ngleftarrow":813,"hfil":814,"intercal":815,"frow":816,"Ġ\\*":817,"frown":818,"mpe":819,"Ġpt":820,"varpro":821,"searrow":822,"bumpe":823,"varprojlim":824,"bumpeq":825,"Down":826,"SS":827,"cd":828,"ere":829,"gcd":830,"ohe":831,"tw":832,"leme":833,"there":834,"injlim":835,"tit":836,"adrightarrow":837,"varinjlim":838,"comp":839,"textit":840,"fore":841,"overleftrightarrow":842,"Downarrow":843,"oheadrightarrow":844,"twoheadrightarrow":845,"lement":846,"therefore":847,"complement":848,"ca":849,"thi":850,"longleftarrow":851,"bigm":852,"triangleright":853,"nearrow":854,"\\#":855,"nce":856,"ral":857,"cance":858,"thick":859,"cancel":860,"Uparrow":861,"nat":862,"ural":863,"mathstrut":864,"suit":865,"bigcirc":866,"smallskip":867,"diamondsuit":868,"normalsize":869,"natural":870,"gt":871,"less":872,"mathtt":873,"bigsqcup":874,"thicksim":875,"lesssim":876,"bow":877,"llde":878,"tie":879,"nullde":880,"miter":881,"limiter":882,"kern":883,"bowtie":884,"nulldelimiter":885,"nulldelimiterspace":886,"Da":887,"hphantom":888,"ro":889,"vDa":890,"barwedge":891,"beth":892,"eqno":893,"vDash":894,"AR":895,"Di":896,"GE":897,"LAR":898,"dskip":899,"ts":900,"Ġ@":901,"medskip":902,"ndown":903,"gets":904,"coprod":905,"dotsm":906,"smash":907,"rightharpoondown":908,"Diamond":909,"LARGE":910,"nrightarrow":911,"nleftrightarrow":912,"rsim":913,"rne":914,"warrow":915,"mathc":916,"corne":917,"textnormal":918,"preceq":919,"gtrsim":920,"roup":921,"corner":922,"Ġ\\[":923,"Ġ\\]":924,"mathope":925,"lefteq":926,"lose":927,"varkappa":928,"Bigm":929,"Biggm":930,"mathclose":931,"mathopen":932,"lefteqn":933,"Bar":934,"Ti":935,"lr":936,"swarrow":937,"uge":938,"vru":939,"xleftarrow":940,"mathnormal":941,"rightrightarrow":942,"rightleftarrow":943,"sqsubseteq":944,"succeq":945,"Tilde":946,"lrcorner":947,"vrule":948,"rightrightarrows":949,"rightleftarrows":950,"AA":951,"Hat":952,"ak":953,"ble":954,"dou":955,"hss":956,"min":957,"nright":958,"nleftarrow":959,"uph":960,"wbre":961,"allo":962,"side":963,"sqcap":964,"hom":965,"bigodot":966,"arpoonright":967,"blebarwedge":968,"doublebarwedge":969,"upharpoonright":970,"wbreak":971,"allowbreak":972,"sideset":973,"--":974,"Huge":975,"amal":976,"do":977,"fbox":978,"group":979,"hskip":980,"lse":981,"pprox":982,"rk":983,"rgroup":984,"rapprox":985,"Ġin":986,"arrayco":987,"sure":988,"varlim":989,"pmb":990,"cite":991,"substack":992,"leftrightarrows":993,"supsetneq":994,"Longleftarrow":995,"updownarrow":996,"ensure":997,"lgroup":998,"gtrapprox":999,"amalg":1000,"lsep":1001,"arraycolsep":1002,"ensuremath":1003,"asym":1004,"ch":1005,"dig":1006,"ddag":1007,"ew":1008,"gra":1009,"gime":1010,"jo":1011,"ltimes":1012,"nleq":1013,"tch":1014,"frame":1015,"max":1016,"thde":1017,"inrel":1018,"ver":1019,"withde":1020,"ointop":1021,"notag":1022,"smallint":1023,"skew":1024,"lims":1025,"asymp":1026,"digamma":1027,"grave":1028,"gimel":1029,"joinrel":1030,"framebox":1031,"withdelims":1032,"Ar":1033,"Rrightarrow":1034,"ae":1035,"ag":1036,"fill":1037,"hspace":1038,"huge":1039,"lq":1040,"nwarrow":1041,"wline":1042,"Ġ14":1043,"mark":1044,"led":1045,"inf":1046,"inde":1047,"Ġex":1048,"pitch":1049,"dotsi":1050,"intop":1051,"rowvert":1052,"llcorner":1053,"black":1054,"leqq":1055,"biggm":1056,"approxeq":1057,"diag":1058,"textsc":1059,"textsl":1060,"circled":1061,"fork":1062,"cur":1063,"newline":1064,"negthick":1065,"atopwithdelims":1066,"Leftarrow":1067,"footnotemark":1068,"uplus":1069,"subsetneqq":1070,"---":1071,"varlimsup":1072,"varliminf":1073,"verb":1074,"Arrowvert":1075,"pitchfork":1076,"blacksquare":1077,"diagup":1078,"negthickspace":1079,"23":1080,"25":1081,"\\-":1082,"\\/":1083,"ape":1084,"ckap":1085,"dddot":1086,"erline":1087,"ever":1088,"ij":1089,"ice":1090,"ly":1091,"md":1092,"nda":1093,"nnu":1094,"nmid":1095,"nRightarrow":1096,"nVdash":1097,"of":1098,"off":1099,"sho":1100,"spe":1101,"wr":1102,"ymath":1103,"Ġ#":1104,"Ġ\\'":1105,"Ġ\\^":1106,"Ġ10":1107,"Ġ15":1108,"mannu":1109,"igarrow":1110,"fter":1111,"meral":1112,"leftrightharpoo":1113,"rightsqu":1114,"def":1115,"arrayst":1116,"rtmid":1117,"interline":1118,"vearrow":1119,"ngeq":1120,"hoice":1121,"lax":1122,"varGamma":1123,"varpropto":1124,"vartriangle":1125,"varUpsilon":1126,"biguplus":1127,"expa":1128,"Ġ<$":1129,"mathbin":1130,"perca":1131,"textcircled":1132,"textmd":1133,"scsh":1134,"cial":1135,"retch":1136,"relax":1137,"overwithdelims":1138,"noinde":1139,"owns":1140,"veebar":1141,"underbar":1142,"underrightarrow":1143,"upperca":1144,"backsimeq":1145,"trianglelefteq":1146,"boxtimes":1147,"boxed":1148,"preccur":1149,"thickap":1150,"root":1151,"romannu":1152,"mathchoice":1153,"index":1154,"circledcirc":1155,"curvearrow":1156,"everymath":1157,"lyeq":1158,"ndafter":1159,"offinterline":1160,"shortmid":1161,"special":1162,"leftrightharpoons":1163,"rightsquigarrow":1164,"arraystretch":1165,"expandafter":1166,"scshape":1167,"noindent":1168,"uppercase":1169,"preccurlyeq":1170,"thickapprox":1171,"romannumeral":1172,"curvearrowright":1173,"offinterlineskip":1174},"merges":["Ġ }","Ġ {","Ġ \\","Ġ _","Ġ ^","Ġ 2","Ġ )","Ġ (","Ġ 1","r a","Ġ =","Ġ -","č Ċ","Ġ ,","f ra","fra c","Ġ +","m a","t a","i g","Ġ 0","a r","a l","l e","Ġ i","t h","Ġ x","f t","ig h","m e","r igh","ma th","Ġ n","Ġ .","Ġ\\ ,","i n","p h","Ġ d","le ft","Ġ a","righ t","a m","e ta","t i","Ġ m","m u","Ġ 3","Ġ k","Ġ t","Ġ e","Ġ r","Ġ &","Ġ c","Ġ p","s i","r m","d e","math rm","Ġ 4","Ġ s","p r","Ġ ~","ph a","Ġ l","al pha","d a","Ġ A","Ġ\\ ;","o t","p i","p ar","ti al","par tial","i me","pr ime","p si","d ot","Ġ j","Ġ b","Ġ f","l ta","Ġ |","am ma","b da","am bda","ph i","Ġ q","b f","Ġ g","n u","Ġ z","ra y","ar ray","Ġ M","Ġ T","Ġ y","c al","b ar","Ġ N","ig ma","Ġ R","r t","l ambda","in t","Ġ B","v e","n g","q u","Ġ L","Ġ /","l o","b eta","ng le","Ġ u","de lta","s q","sq rt","th eta","Ġ\\ \\","g amma","Ġ ]","s igma","g a","me ga","Ġ D","Ġ F","Ġ [","Ġ S","math bf","s u","Ġ P","lo n","Ġ v","su m","psi lon","Ġ V","Ġ C","c dot","e psilon","a t","h at","a d","qu ad","Ġ h","h o","r ho","h i","t o","Ġ E","l a","Ġ H","l de","ti lde","Ġ Q","Ġ 5","v ar","Ġ X","Ġ G","b e","n d","o mega","e nd","g in","be gin","ta u","Ġ 6","n a","ve c","Ġ I","Ġ o","ra ngle","Ġ *","D e","De lta","G amma","p e","ft y","in fty","Ġ K","x i","Ġ 8","o w","Ġ J","Ġ U","r ow","ta r","g e","P hi","a p","Ġ W","c o","me s","ti mes","s in","Ġ Z","la ngle","o pe","r na","ra to","ope rato","rna me","operato rname","tar row","l in","lin e","var phi","p m","r line","L ambda","Ġ\\ !","Ġ ;","dot s","co s","Ġ w","righ tarrow","b ig","c hi","o ve","Ġ\\ }","ove rline","Ġ 7","e x","p a","s t","pr o","q quad","i v","e qu","equ iv","Ġ O","l n","O mega","l l","Ġ 9","k ap","kap pa","B ig","Ġ\\ {","da g","Ġ Y","\\ {","var epsilon","cdot s","Ġ :","math cal","P si","Ġ >","b o","bo l","Ġ <","ge r","dag ger","l dots","e ll","b la","na bla","ex p","y le","st yle","z eta","S igma","w i","wi de","si m","le q","Ġ !","big g","math b","mathb b","Ġ\\ :","h bar","o times","bol d","\\ }","m i","pro x","ap prox","P i","lo g","mi d","s p","ve rt","d i","pro d","pe r","per p","y style","la ystyle","sp laystyle","di splaystyle","me q","si meq","e d","w ed","wed ge","wide tilde","s y","sy m","sym bol","bold symbol","c k","t ex","tex t","r i","T h","Th eta","ge q","s e","e q","n de","u nde","ta n","s c","a st","r c","se t","p t","wide hat","c i","ci rc","r e","ri pt","sc ript","unde rline","Ġ\\ |","re l","n eq","s ta","sta ck","stack rel","sin h","o p","u s","cos h","Big g","c e","text style","s tar","n ot","fra k","math frak","m p","bigg r","l us","op lus","var theta","bigg l","Big r","b ra","Big l","f o","s f","su b","sub set","ng rightarrow","e c","bold math","ra ll","fo rall","script style","ec t","par row","u parrow","b j","bj ect","p to","pro pto","Ġ '","lo ngrightarrow","big l","big r","o int","p s","ma ps","maps to","o m","l le","\\ |","d dot","c u","b in","bin om","v dots","a ngle","left rightarrow","ove r","o r","math sf","cu p","bra ce","n o","ar c","f lo","flo or","t ri","tri angle","X i","c ot","b ot","co ng","i t","m be","nu mbe","no numbe","nonumbe r","c ap","R igh","Righ tarrow","z e","si ze","text rm","n e","arc tan","ra lle","pa ralle","paralle l","c frac","Ġ- -","o bject","ĠO bject","bra ck","s h","ar row","ow n","var rho","subset eq","r brace","text bf","i math","o d","d own","h e","la nd","script scriptstyle","script size","c he","che ck","s la","over rightarrow","down arrow","Bigg l","g g","n to","pha nto","phanto m","e xi","h line","st s","exi sts","Bigg r","b u","r floor","d dots","i o","io ta","lle t","bu llet","co lon","in us","U psilon","l floor","l brack","unde rbrace","ne g","I m","math it","t in","tin y","j math","le f","sla sh","ve e","m inus","set minus","R e","i int","lef tarrow","V e","Ve rt","at op","su p","big cup","w p","di m","se c","sup set","L o","lo r","pm od","m od","big oplus","i l","b mod","co th","L e","ft rightarrow","Le ftrightarrow","ng leftrightarrow","s ma","u psilon","\\ ,","c sc","e ph","al eph","big wedge","arc sin","sma ll","o dot","over set","r brack","mi t","l brace","l i","ar p","ar ge","Ġ\\ #","b re","text sf","Lo ngrightarrow","bre ve","e m","y set","var pi","pt yset","em ptyset","f f","i ff","n t","e r","la p","ln ot","da sh","unde r","sla nt","ar g","under set","B o","Bo x","Ġ \"","s pa","spa ce","de g","i iint","o o","ot not","fo otnot","arp oo","footnot e","r lap","e s","i mp","s b","t e","big triangle","li es","imp lies","\\ ;","k er","footnote size","th arpoo","u p","a cu","acu te","lo ngleftrightarrow","e il","l ce","rc eil","lce il","v phantom","e n","th in","a ck","b ack","t t","back slash","x rightarrow","v dash","to p","righ tharpoo","var sigma","Lo ngleftrightarrow","math op","l arge","big cap","leq slant","Ġ `","over brace","nu p","rightharpoo nup","big otimes","triangle q","L arge","r u","nu ll","bigtriangle up","var no","thin g","varno thing","dot eq","Ġ\\ _","over leftarrow","h f","big star","en space","\\ !","st ru","stru t","om inus","d iv","o nd","am ond","d dagger","Ġc m","n i","s k","di amond","r Vert","pr ot","prot ect","i p","var Delta","not in","sk ip","l Vert","Ġ\\ /","dots c","i ll","u le","\\ :","hf ill","k rightarrow","o krightarrow","ho okrightarrow","sh arp","V dash","big vee","subset neq","supset eq","Ġ ?","ng mapsto","lo ngmapsto","cdot p","geq slant","bigtriangle down","dots b","li m","f l","triangle left","fl at","s l","bo x","Ġ-- -","sq cup","j lim","l s","m o","de ls","re f","mo dels","ta g","P r","ma l","o u","l lap","thin space","en skip","V ec","e box","k ebox","n or","r d","s qu","v line","¿ ½","ï ¿½","Ġ �","ma kebox","su rd","nor mal","ar e","squ are","p ou","math rel","var Omega","nd s","small setminus","pou nds","n s","s s","s mi","math or","right lef","text up","tharpoo ns","smi le","mathor d","rightlef tharpoons","c c","Ġ\\ -","su cc","f tarrow","r times","de t","pr ec","text tt","o slash","Ġ\\ &","arrow vert","l g","Ġm m","int er","ngle ftarrow","hf il","inter cal","f row","Ġ\\ *","frow n","m pe","Ġp t","var pro","se arrow","bu mpe","varpro jlim","bumpe q","D own","S S","c d","e re","g cd","o he","t w","le me","th ere","in jlim","ti t","ad rightarrow","var injlim","co mp","tex tit","fo re","over leftrightarrow","Down arrow","ohe adrightarrow","tw oheadrightarrow","leme nt","there fore","comp lement","c a","th i","lo ngleftarrow","big m","triangle right","ne arrow","\\ #","n ce","ra l","ca nce","thi ck","cance l","U parrow","n at","u ral","math strut","su it","big circ","small skip","diamond suit","normal size","nat ural","g t","le ss","math tt","big sqcup","thick sim","less sim","b ow","l lde","ti e","nu llde","mit er","li miter","ker n","bow tie","nullde limiter","nulldelimiter space","D a","h phantom","r o","v Da","bar wedge","be th","eq no","vDa sh","A R","D i","G E","L AR","d skip","t s","Ġ @","me dskip","nd own","ge ts","co prod","dots m","sma sh","rightharpoo ndown","Di amond","LAR GE","n rightarrow","n leftrightarrow","r sim","r ne","w arrow","math c","co rne","text normal","prec eq","gt rsim","ro up","corne r","Ġ\\ [","Ġ\\ ]","math ope","left eq","lo se","var kappa","Big m","Bigg m","mathc lose","mathope n","lefteq n","B ar","T i","l r","s warrow","u ge","v ru","x leftarrow","math normal","right rightarrow","right leftarrow","sq subseteq","succ eq","Ti lde","lr corner","vru le","rightrightarrow s","rightleftarrow s","A A","H at","a k","b le","d ou","h ss","m in","n right","n leftarrow","u ph","w bre","al lo","si de","sq cap","ho m","big odot","arpoo nright","ble barwedge","dou blebarwedge","uph arpoonright","wbre ak","allo wbreak","side set","- -","H uge","a mal","d o","f box","g roup","h skip","l se","p prox","r k","r group","ra pprox","Ġi n","array co","su re","var lim","pm b","ci te","sub stack","leftrightarrow s","supset neq","Lo ngleftarrow","up downarrow","en sure","lg roup","gt rapprox","amal g","lse p","arrayco lsep","ensure math","a sym","c h","d ig","d dag","e w","g ra","g ime","j o","l times","n leq","t ch","fra me","ma x","th de","in rel","ve r","wi thde","oint op","no tag","small int","sk ew","lim s","asym p","dig amma","gra ve","gime l","jo inrel","frame box","withde lims","A r","R rightarrow","a e","a g","f ill","h space","h uge","l q","n warrow","w line","Ġ1 4","ma rk","le d","in f","in de","Ġe x","pi tch","dot si","int op","row vert","ll corner","bla ck","leq q","bigg m","approx eq","di ag","text sc","text sl","circ led","fo rk","cu r","ne wline","neg thick","atop withdelims","Le ftarrow","footnote mark","up lus","subsetneq q","-- -","varlim sup","varlim inf","ver b","Ar rowvert","pitch fork","black square","diag up","negthick space","2 3","2 5","\\ -","\\ /","a pe","c kap","d ddot","e rline","e ver","i j","i ce","l y","m d","n da","n nu","n mid","n Rightarrow","n Vdash","o f","o ff","s ho","s pe","w r","y math","Ġ #","Ġ\\ '","Ġ\\ ^","Ġ1 0","Ġ1 5","ma nnu","ig arrow","ft er","me ral","left rightharpoo","right squ","de f","array st","rt mid","int erline","ve arrow","ng eq","ho ice","la x","var Gamma","var propto","var triangle","var Upsilon","big uplus","ex pa","Ġ< $","mathb in","per ca","text circled","text md","sc sh","ci al","re tch","re lax","over withdelims","no inde","own s","vee bar","under bar","under rightarrow","up perca","back simeq","triangleleft eq","box times","box ed","prec cur","thi ckap","ro ot","ro mannu","mathc hoice","inde x","circled circ","cur vearrow","ever ymath","ly eq","nda fter","off interline","sho rtmid","spe cial","leftrightharpoo ns","rightsqu igarrow","arrayst retch","expa ndafter","scsh ape","noinde nt","upperca se","preccur lyeq","thickap prox","romannu meral","curvearrow right","offinterline skip"]}}
diff --git a/ppocr/utils/dict/latex_symbol_dict.txt b/ppocr/utils/dict/latex_symbol_dict.txt
new file mode 100644
index 0000000..69b1b84
--- /dev/null
+++ b/ppocr/utils/dict/latex_symbol_dict.txt
@@ -0,0 +1,111 @@
+eos
+sos
+!
+'
+(
+)
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+<
+=
+>
+A
+B
+C
+E
+F
+G
+H
+I
+L
+M
+N
+P
+R
+S
+T
+V
+X
+Y
+[
+\Delta
+\alpha
+\beta
+\cdot
+\cdots
+\cos
+\div
+\exists
+\forall
+\frac
+\gamma
+\geq
+\in
+\infty
+\int
+\lambda
+\ldots
+\leq
+\lim
+\log
+\mu
+\neq
+\phi
+\pi
+\pm
+\prime
+\rightarrow
+\sigma
+\sin
+\sqrt
+\sum
+\tan
+\theta
+\times
+]
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+\{
+|
+\}
+{
+}
+^
+_
diff --git a/ppocr/utils/dict/latin_dict.txt b/ppocr/utils/dict/latin_dict.txt
new file mode 100644
index 0000000..e166bf3
--- /dev/null
+++ b/ppocr/utils/dict/latin_dict.txt
@@ -0,0 +1,185 @@
+ 
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+<
+=
+>
+?
+@
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+[
+]
+_
+`
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+{
+}
+¡
+£
+§
+ª
+«
+­
+°
+²
+³
+´
+µ
+·
+º
+»
+¿
+À
+Á
+Â
+Ä
+Å
+Ç
+È
+É
+Ê
+Ë
+Ì
+Í
+Î
+Ï
+Ò
+Ó
+Ô
+Õ
+Ö
+Ú
+Ü
+Ý
+ß
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+ì
+í
+î
+ï
+ñ
+ò
+ó
+ô
+õ
+ö
+ø
+ù
+ú
+û
+ü
+ý
+ą
+Ć
+ć
+Č
+č
+Đ
+đ
+ę
+ı
+Ł
+ł
+ō
+Œ
+œ
+Š
+š
+Ÿ
+Ž
+ž
+ʒ
+β
+δ
+ε
+з
+Ṡ
+‘
+€
+™
diff --git a/ppocr/utils/dict/layout_dict/layout_cdla_dict.txt b/ppocr/utils/dict/layout_dict/layout_cdla_dict.txt
new file mode 100644
index 0000000..5380334
--- /dev/null
+++ b/ppocr/utils/dict/layout_dict/layout_cdla_dict.txt
@@ -0,0 +1,10 @@
+text
+title
+figure
+figure_caption
+table
+table_caption
+header
+footer
+reference
+equation
diff --git a/ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt b/ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt
new file mode 100644
index 0000000..d9c92da
--- /dev/null
+++ b/ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt
@@ -0,0 +1,5 @@
+text
+title
+list
+table
+figure
diff --git a/ppocr/utils/dict/layout_dict/layout_table_dict.txt b/ppocr/utils/dict/layout_dict/layout_table_dict.txt
new file mode 100644
index 0000000..ecbe2e6
--- /dev/null
+++ b/ppocr/utils/dict/layout_dict/layout_table_dict.txt
@@ -0,0 +1 @@
+table
diff --git a/ppocr/utils/dict/mr_dict.txt b/ppocr/utils/dict/mr_dict.txt
new file mode 100644
index 0000000..283b150
--- /dev/null
+++ b/ppocr/utils/dict/mr_dict.txt
@@ -0,0 +1,153 @@
+ 
+!
+#
+$
+%
+&
+'
+(
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+?
+@
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+_
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+É
+é
+ँ
+ं
+ः
+अ
+आ
+इ
+ई
+उ
+ऊ
+ए
+ऐ
+ऑ
+ओ
+औ
+क
+ख
+ग
+घ
+च
+छ
+ज
+झ
+ञ
+ट
+ठ
+ड
+ढ
+ण
+त
+थ
+द
+ध
+न
+प
+फ
+ब
+भ
+म
+य
+र
+ऱ
+ल
+ळ
+व
+श
+ष
+स
+ह
+़
+ा
+ि
+ी
+ु
+ू
+ृ
+ॅ
+े
+ै
+ॉ
+ो
+ौ
+्
+०
+१
+२
+३
+४
+५
+६
+७
+८
+९
diff --git a/ppocr/utils/dict/ne_dict.txt b/ppocr/utils/dict/ne_dict.txt
new file mode 100644
index 0000000..5a7df95
--- /dev/null
+++ b/ppocr/utils/dict/ne_dict.txt
@@ -0,0 +1,153 @@
+ 
+!
+#
+$
+%
+&
+'
+(
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+?
+@
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+_
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+É
+é
+ः
+अ
+आ
+इ
+ई
+उ
+ऊ
+ऋ
+ए
+ऐ
+ओ
+औ
+क
+ख
+ग
+घ
+ङ
+च
+छ
+ज
+झ
+ञ
+ट
+ठ
+ड
+ढ
+ण
+त
+थ
+द
+ध
+न
+ऩ
+प
+फ
+ब
+भ
+म
+य
+र
+ऱ
+ल
+व
+श
+ष
+स
+ह
+़
+ा
+ि
+ी
+ु
+ू
+ृ
+े
+ै
+ो
+ौ
+्
+॒
+ॠ
+।
+०
+१
+२
+३
+४
+५
+६
+७
+८
+९
diff --git a/ppocr/utils/dict/oc_dict.txt b/ppocr/utils/dict/oc_dict.txt
new file mode 100644
index 0000000..e88af8b
--- /dev/null
+++ b/ppocr/utils/dict/oc_dict.txt
@@ -0,0 +1,96 @@
+o
+c
+_
+i
+m
+g
+/
+2
+0
+I
+L
+S
+V
+R
+C
+1
+v
+a
+l
+4
+3
+.
+j
+p	
+r
+e
+è
+t
+9
+7
+5
+8
+n
+'
+b
+s
+6
+q
+u
+á
+d
+ò
+à
+h
+z
+f
+ï
+í
+A
+ç
+x
+ó
+é
+P
+O
+Ò
+ü
+k
+À
+F
+-
+ú
+­
+æ
+Á
+D
+E
+w
+K
+T
+N
+y
+U
+Z
+G
+B
+J
+H
+M
+W
+Y
+X
+Q
+%
+$
+,
+@
+&
+!
+:
+(
+#
+?
++
+É
+ 
diff --git a/ppocr/utils/dict/parseq_dict.txt b/ppocr/utils/dict/parseq_dict.txt
new file mode 100644
index 0000000..87c6d67
--- /dev/null
+++ b/ppocr/utils/dict/parseq_dict.txt
@@ -0,0 +1,94 @@
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+:
+;
+<
+=
+>
+?
+@
+[
+\
+]
+^
+_
+`
+{
+|
+}
+~
diff --git a/ppocr/utils/dict/ppocrv4_doc_dict.txt b/ppocr/utils/dict/ppocrv4_doc_dict.txt
new file mode 100644
index 0000000..09e275b
--- /dev/null
+++ b/ppocr/utils/dict/ppocrv4_doc_dict.txt
@@ -0,0 +1,15629 @@
+'
+疗
+绚
+诚
+娇
+溜
+题
+贿
+者
+廖
+更
+纳
+加
+奉
+公
+一
+就
+汴
+计
+与
+路
+房
+原
+妇
+2
+0
+8
+-
+7
+其
+>
+:
+]
+,
+，
+骑
+刈
+全
+消
+昏
+傈
+安
+久
+钟
+嗅
+不
+影
+处
+驽
+蜿
+资
+关
+椤
+地
+瘸
+专
+问
+忖
+票
+嫉
+炎
+韵
+要
+月
+田
+节
+陂
+鄙
+捌
+备
+拳
+伺
+眼
+网
+盎
+大
+傍
+心
+东
+愉
+汇
+蹿
+科
+每
+业
+里
+航
+晏
+字
+平
+录
+先
+1
+3
+彤
+鲶
+产
+稍
+督
+腴
+有
+象
+岳
+注
+绍
+在
+泺
+文
+定
+核
+名
+水
+过
+理
+让
+偷
+率
+等
+这
+发
+”
+为
+含
+肥
+酉
+相
+鄱
+七
+编
+猥
+锛
+日
+镀
+蒂
+掰
+倒
+辆
+栾
+栗
+综
+涩
+州
+雌
+滑
+馀
+了
+机
+块
+司
+宰
+甙
+兴
+矽
+抚
+保
+用
+沧
+秩
+如
+收
+息
+滥
+页
+疑
+埠
+!
+！
+姥
+异
+橹
+钇
+向
+下
+跄
+的
+椴
+沫
+国
+绥
+獠
+报
+开
+民
+蜇
+何
+分
+凇
+长
+讥
+藏
+掏
+施
+羽
+中
+讲
+派
+嘟
+人
+提
+浼
+间
+世
+而
+古
+多
+倪
+唇
+饯
+控
+庚
+首
+赛
+蜓
+味
+断
+制
+觉
+技
+替
+艰
+溢
+潮
+夕
+钺
+外
+摘
+枋
+动
+双
+单
+啮
+户
+枇
+确
+锦
+曜
+杜
+或
+能
+效
+霜
+盒
+然
+侗
+电
+晁
+放
+步
+鹃
+新
+杖
+蜂
+吒
+濂
+瞬
+评
+总
+隍
+对
+独
+合
+也
+是
+府
+青
+天
+诲
+墙
+组
+滴
+级
+邀
+帘
+示
+已
+时
+骸
+仄
+泅
+和
+遨
+店
+雇
+疫
+持
+巍
+踮
+境
+只
+亨
+目
+鉴
+崤
+闲
+体
+泄
+杂
+作
+般
+轰
+化
+解
+迂
+诿
+蛭
+璀
+腾
+告
+版
+服
+省
+师
+小
+规
+程
+线
+海
+办
+引
+二
+桧
+牌
+砺
+洄
+裴
+修
+图
+痫
+胡
+许
+犊
+事
+郛
+基
+柴
+呼
+食
+研
+奶
+律
+蛋
+因
+葆
+察
+戏
+褒
+戒
+再
+李
+骁
+工
+貂
+油
+鹅
+章
+啄
+休
+场
+给
+睡
+纷
+豆
+器
+捎
+说
+敏
+学
+会
+浒
+设
+诊
+格
+廓
+查
+来
+霓
+室
+溆
+￠
+诡
+寥
+焕
+舜
+柒
+狐
+回
+戟
+砾
+厄
+实
+翩
+尿
+五
+入
+径
+惭
+喹
+股
+宇
+篝
+|
+;
+美
+期
+云
+九
+祺
+扮
+靠
+锝
+槌
+系
+企
+酰
+阊
+暂
+蚕
+忻
+豁
+本
+羹
+执
+条
+钦
+H
+獒
+限
+进
+季
+楦
+于
+芘
+玖
+铋
+茯
+未
+答
+粘
+括
+样
+精
+欠
+矢
+甥
+帷
+嵩
+扣
+令
+仔
+风
+皈
+行
+支
+部
+蓉
+刮
+站
+蜡
+救
+钊
+汗
+松
+嫌
+成
+可
+.
+鹤
+院
+从
+交
+政
+怕
+活
+调
+球
+局
+验
+髌
+第
+韫
+谗
+串
+到
+圆
+年
+米
+/
+*
+友
+忿
+检
+区
+看
+自
+敢
+刃
+个
+兹
+弄
+流
+留
+同
+没
+齿
+星
+聆
+轼
+湖
+什
+三
+建
+蛔
+儿
+椋
+汕
+震
+颧
+鲤
+跟
+力
+情
+璺
+铨
+陪
+务
+指
+族
+训
+滦
+鄣
+濮
+扒
+商
+箱
+十
+召
+慷
+辗
+所
+莞
+管
+护
+臭
+横
+硒
+嗓
+接
+侦
+六
+露
+党
+馋
+驾
+剖
+高
+侬
+妪
+幂
+猗
+绺
+骐
+央
+酐
+孝
+筝
+课
+徇
+缰
+门
+男
+西
+项
+句
+谙
+瞒
+秃
+篇
+教
+碲
+罚
+声
+呐
+景
+前
+富
+嘴
+鳌
+稀
+免
+朋
+啬
+睐
+去
+赈
+鱼
+住
+肩
+愕
+速
+旁
+波
+厅
+健
+茼
+厥
+鲟
+谅
+投
+攸
+炔
+数
+方
+击
+呋
+谈
+绩
+别
+愫
+僚
+躬
+鹧
+胪
+炳
+招
+喇
+膨
+泵
+蹦
+毛
+结
+5
+4
+谱
+识
+陕
+粽
+婚
+拟
+构
+且
+搜
+任
+潘
+比
+郢
+妨
+醪
+陀
+桔
+碘
+扎
+选
+哈
+骷
+楷
+亿
+明
+缆
+脯
+监
+睫
+逻
+婵
+共
+赴
+淝
+凡
+惦
+及
+达
+揖
+谩
+澹
+减
+焰
+蛹
+番
+祁
+柏
+员
+禄
+怡
+峤
+龙
+白
+叽
+生
+闯
+起
+细
+装
+谕
+竟
+聚
+钙
+上
+导
+渊
+按
+艾
+辘
+挡
+耒
+盹
+饪
+臀
+记
+邮
+蕙
+受
+各
+医
+搂
+普
+滇
+朗
+茸
+带
+翻
+酚
+(
+光
+堤
+墟
+蔷
+万
+幻
+〓
+瑙
+辈
+昧
+盏
+亘
+蛀
+吉
+铰
+请
+子
+假
+闻
+税
+井
+诩
+哨
+嫂
+好
+面
+琐
+校
+馊
+鬣
+缂
+营
+访
+炖
+占
+农
+缀
+否
+经
+钚
+棵
+趟
+张
+亟
+吏
+茶
+谨
+捻
+论
+迸
+堂
+玉
+信
+吧
+瞠
+乡
+姬
+寺
+咬
+溏
+苄
+皿
+意
+赉
+宝
+尔
+钰
+艺
+特
+唳
+踉
+都
+荣
+倚
+登
+荐
+丧
+奇
+涵
+批
+炭
+近
+符
+傩
+感
+道
+着
+菊
+虹
+仲
+众
+懈
+濯
+颞
+眺
+南
+释
+北
+缝
+标
+既
+茗
+整
+撼
+迤
+贲
+挎
+耱
+拒
+某
+妍
+卫
+哇
+英
+矶
+藩
+治
+他
+元
+领
+膜
+遮
+穗
+蛾
+飞
+荒
+棺
+劫
+么
+市
+火
+温
+拈
+棚
+洼
+转
+果
+奕
+卸
+迪
+伸
+泳
+斗
+邡
+侄
+涨
+屯
+萋
+胭
+氡
+崮
+枞
+惧
+冒
+彩
+斜
+手
+豚
+随
+旭
+淑
+妞
+形
+菌
+吲
+沱
+争
+驯
+歹
+挟
+兆
+柱
+传
+至
+包
+内
+响
+临
+红
+功
+弩
+衡
+寂
+禁
+老
+棍
+耆
+渍
+织
+害
+氵
+渑
+布
+载
+靥
+嗬
+虽
+苹
+咨
+娄
+库
+雉
+榜
+帜
+嘲
+套
+瑚
+亲
+簸
+欧
+边
+6
+腿
+旮
+抛
+吹
+瞳
+得
+镓
+梗
+厨
+继
+漾
+愣
+憨
+士
+策
+窑
+抑
+躯
+襟
+脏
+参
+贸
+言
+干
+绸
+鳄
+穷
+藜
+音
+折
+详
+)
+举
+悍
+甸
+癌
+黎
+谴
+死
+罩
+迁
+寒
+驷
+袖
+媒
+蒋
+掘
+模
+纠
+恣
+观
+祖
+蛆
+碍
+位
+稿
+主
+澧
+跌
+筏
+京
+锏
+帝
+贴
+证
+糠
+才
+黄
+鲸
+略
+炯
+饱
+四
+出
+园
+犀
+牧
+容
+汉
+杆
+浈
+汰
+瑷
+造
+虫
+瘩
+怪
+驴
+济
+应
+花
+沣
+谔
+夙
+旅
+价
+矿
+以
+考
+s
+u
+呦
+晒
+巡
+茅
+准
+肟
+瓴
+詹
+仟
+褂
+译
+桌
+混
+宁
+怦
+郑
+抿
+些
+余
+鄂
+饴
+攒
+珑
+群
+阖
+岔
+琨
+藓
+预
+环
+洮
+岌
+宀
+杲
+瀵
+最
+常
+囡
+周
+踊
+女
+鼓
+袭
+喉
+简
+范
+薯
+遐
+疏
+粱
+黜
+禧
+法
+箔
+斤
+遥
+汝
+奥
+直
+贞
+撑
+置
+绱
+集
+她
+馅
+逗
+钧
+橱
+魉
+[
+恙
+躁
+唤
+9
+旺
+膘
+待
+脾
+惫
+购
+吗
+依
+盲
+度
+瘿
+蠖
+俾
+之
+镗
+拇
+鲵
+厝
+簧
+续
+款
+展
+啃
+表
+剔
+品
+钻
+腭
+损
+清
+锶
+统
+涌
+寸
+滨
+贪
+链
+吠
+冈
+伎
+迥
+咏
+吁
+览
+防
+迅
+失
+汾
+阔
+逵
+绀
+蔑
+列
+川
+凭
+努
+熨
+揪
+利
+俱
+绉
+抢
+鸨
+我
+即
+责
+膦
+易
+毓
+鹊
+刹
+玷
+岿
+空
+嘞
+绊
+排
+术
+估
+锷
+违
+们
+苟
+铜
+播
+肘
+件
+烫
+审
+鲂
+广
+像
+铌
+惰
+铟
+巳
+胍
+鲍
+康
+憧
+色
+恢
+想
+拷
+尤
+疳
+知
+S
+Y
+F
+D
+A
+峄
+裕
+帮
+握
+搔
+氐
+氘
+难
+墒
+沮
+雨
+叁
+缥
+悴
+藐
+湫
+娟
+苑
+稠
+颛
+簇
+后
+阕
+闭
+蕤
+缚
+怎
+佞
+码
+嘤
+蔡
+痊
+舱
+螯
+帕
+赫
+昵
+升
+烬
+岫
+、
+疵
+蜻
+髁
+蕨
+隶
+烛
+械
+丑
+盂
+梁
+强
+鲛
+由
+拘
+揉
+劭
+龟
+撤
+钩
+呕
+孛
+费
+妻
+漂
+求
+阑
+崖
+秤
+甘
+通
+深
+补
+赃
+坎
+床
+啪
+承
+吼
+量
+暇
+钼
+烨
+阂
+擎
+脱
+逮
+称
+P
+神
+属
+矗
+华
+届
+狍
+葑
+汹
+育
+患
+窒
+蛰
+佼
+静
+槎
+运
+鳗
+庆
+逝
+曼
+疱
+克
+代
+官
+此
+麸
+耧
+蚌
+晟
+例
+础
+榛
+副
+测
+唰
+缢
+迹
+灬
+霁
+身
+岁
+赭
+扛
+又
+菡
+乜
+雾
+板
+读
+陷
+徉
+贯
+郁
+虑
+变
+钓
+菜
+圾
+现
+琢
+式
+乐
+维
+渔
+浜
+左
+吾
+脑
+钡
+警
+T
+啵
+拴
+偌
+漱
+湿
+硕
+止
+骼
+魄
+积
+燥
+联
+踢
+玛
+则
+窿
+见
+振
+畿
+送
+班
+钽
+您
+赵
+刨
+印
+讨
+踝
+籍
+谡
+舌
+崧
+汽
+蔽
+沪
+酥
+绒
+怖
+财
+帖
+肱
+私
+莎
+勋
+羔
+霸
+励
+哼
+帐
+将
+帅
+渠
+纪
+婴
+娩
+岭
+厘
+滕
+吻
+伤
+坝
+冠
+戊
+隆
+瘁
+介
+涧
+物
+黍
+并
+姗
+奢
+蹑
+掣
+垸
+锴
+命
+箍
+捉
+病
+辖
+琰
+眭
+迩
+艘
+绌
+繁
+寅
+若
+毋
+思
+诉
+类
+诈
+燮
+轲
+酮
+狂
+重
+反
+职
+筱
+县
+委
+磕
+绣
+奖
+晋
+濉
+志
+徽
+肠
+呈
+獐
+坻
+口
+片
+碰
+几
+村
+柿
+劳
+料
+获
+亩
+惕
+晕
+厌
+号
+罢
+池
+正
+鏖
+煨
+家
+棕
+复
+尝
+懋
+蜥
+锅
+岛
+扰
+队
+坠
+瘾
+钬
+@
+卧
+疣
+镇
+譬
+冰
+彷
+频
+黯
+据
+垄
+采
+八
+缪
+瘫
+型
+熹
+砰
+楠
+襁
+箐
+但
+嘶
+绳
+啤
+拍
+盥
+穆
+傲
+洗
+盯
+塘
+怔
+筛
+丿
+台
+恒
+喂
+葛
+永
+￥
+烟
+酒
+桦
+书
+砂
+蚝
+缉
+态
+瀚
+袄
+圳
+轻
+蛛
+超
+榧
+遛
+姒
+奘
+铮
+右
+荽
+望
+偻
+卡
+丶
+氰
+附
+做
+革
+索
+戚
+坨
+桷
+唁
+垅
+榻
+岐
+偎
+坛
+莨
+山
+殊
+微
+骇
+陈
+爨
+推
+嗝
+驹
+澡
+藁
+呤
+卤
+嘻
+糅
+逛
+侵
+郓
+酌
+德
+摇
+※
+鬃
+被
+慨
+殡
+羸
+昌
+泡
+戛
+鞋
+河
+宪
+沿
+玲
+鲨
+翅
+哽
+源
+铅
+语
+照
+邯
+址
+荃
+佬
+顺
+鸳
+町
+霭
+睾
+瓢
+夸
+椁
+晓
+酿
+痈
+咔
+侏
+券
+噎
+湍
+签
+嚷
+离
+午
+尚
+社
+锤
+背
+孟
+使
+浪
+缦
+潍
+鞅
+军
+姹
+驶
+笑
+鳟
+鲁
+》
+孽
+钜
+绿
+洱
+礴
+焯
+椰
+颖
+囔
+乌
+孔
+巴
+互
+性
+椽
+哞
+聘
+昨
+早
+暮
+胶
+炀
+隧
+低
+彗
+昝
+铁
+呓
+氽
+藉
+喔
+癖
+瑗
+姨
+权
+胱
+韦
+堑
+蜜
+酋
+楝
+砝
+毁
+靓
+歙
+锲
+究
+屋
+喳
+骨
+辨
+碑
+武
+鸠
+宫
+辜
+烊
+适
+坡
+殃
+培
+佩
+供
+走
+蜈
+迟
+翼
+况
+姣
+凛
+浔
+吃
+飘
+债
+犟
+金
+促
+苛
+崇
+坂
+莳
+畔
+绂
+兵
+蠕
+斋
+根
+砍
+亢
+欢
+恬
+崔
+剁
+餐
+榫
+快
+扶
+‖
+濒
+缠
+鳜
+当
+彭
+驭
+浦
+篮
+昀
+锆
+秸
+钳
+弋
+娣
+瞑
+夷
+龛
+苫
+拱
+致
+%
+嵊
+障
+隐
+弑
+初
+娓
+抉
+汩
+累
+蓖
+"
+唬
+助
+苓
+昙
+押
+毙
+破
+城
+郧
+逢
+嚏
+獭
+瞻
+溱
+婿
+赊
+跨
+恼
+璧
+萃
+姻
+貉
+灵
+炉
+密
+氛
+陶
+砸
+谬
+衔
+点
+琛
+沛
+枳
+层
+岱
+诺
+脍
+榈
+埂
+征
+冷
+裁
+打
+蹴
+素
+瘘
+逞
+蛐
+聊
+激
+腱
+萘
+踵
+飒
+蓟
+吆
+取
+咙
+簋
+涓
+矩
+曝
+挺
+揣
+座
+你
+史
+舵
+焱
+尘
+苏
+笈
+脚
+溉
+榨
+诵
+樊
+邓
+焊
+义
+庶
+儋
+蟋
+蒲
+赦
+呷
+杞
+诠
+豪
+还
+试
+颓
+茉
+太
+除
+紫
+逃
+痴
+草
+充
+鳕
+珉
+祗
+墨
+渭
+烩
+蘸
+慕
+璇
+镶
+穴
+嵘
+恶
+骂
+险
+绋
+幕
+碉
+肺
+戳
+刘
+潞
+秣
+纾
+潜
+銮
+洛
+须
+罘
+销
+瘪
+汞
+兮
+屉
+r
+林
+厕
+质
+探
+划
+狸
+殚
+善
+煊
+烹
+〒
+锈
+逯
+宸
+辍
+泱
+柚
+袍
+远
+蹋
+嶙
+绝
+峥
+娥
+缍
+雀
+徵
+认
+镱
+谷
+=
+贩
+勉
+撩
+鄯
+斐
+洋
+非
+祚
+泾
+诒
+饿
+撬
+威
+晷
+搭
+芍
+锥
+笺
+蓦
+候
+琊
+档
+礁
+沼
+卵
+荠
+忑
+朝
+凹
+瑞
+头
+仪
+弧
+孵
+畏
+铆
+突
+衲
+车
+浩
+气
+茂
+悖
+厢
+枕
+酝
+戴
+湾
+邹
+飚
+攘
+锂
+写
+宵
+翁
+岷
+无
+喜
+丈
+挑
+嗟
+绛
+殉
+议
+槽
+具
+醇
+淞
+笃
+郴
+阅
+饼
+底
+壕
+砚
+弈
+询
+缕
+庹
+翟
+零
+筷
+暨
+舟
+闺
+甯
+撞
+麂
+茌
+蔼
+很
+珲
+捕
+棠
+角
+阉
+媛
+娲
+诽
+剿
+尉
+爵
+睬
+韩
+诰
+匣
+危
+糍
+镯
+立
+浏
+阳
+少
+盆
+舔
+擘
+匪
+申
+尬
+铣
+旯
+抖
+赘
+瓯
+居
+ˇ
+哮
+游
+锭
+茏
+歌
+坏
+甚
+秒
+舞
+沙
+仗
+劲
+潺
+阿
+燧
+郭
+嗖
+霏
+忠
+材
+奂
+耐
+跺
+砀
+输
+岖
+媳
+氟
+极
+摆
+灿
+今
+扔
+腻
+枝
+奎
+药
+熄
+吨
+话
+q
+额
+慑
+嘌
+协
+喀
+壳
+埭
+视
+著
+於
+愧
+陲
+翌
+峁
+颅
+佛
+腹
+聋
+侯
+咎
+叟
+秀
+颇
+存
+较
+罪
+哄
+岗
+扫
+栏
+钾
+羌
+己
+璨
+枭
+霉
+煌
+涸
+衿
+键
+镝
+益
+岢
+奏
+连
+夯
+睿
+冥
+均
+糖
+狞
+蹊
+稻
+爸
+刿
+胥
+煜
+丽
+肿
+璃
+掸
+跚
+灾
+垂
+樾
+濑
+乎
+莲
+窄
+犹
+撮
+战
+馄
+软
+络
+显
+鸢
+胸
+宾
+妲
+恕
+埔
+蝌
+份
+遇
+巧
+瞟
+粒
+恰
+剥
+桡
+博
+讯
+凯
+堇
+阶
+滤
+卖
+斌
+骚
+彬
+兑
+磺
+樱
+舷
+两
+娱
+福
+仃
+差
+找
+桁
+÷
+净
+把
+阴
+污
+戬
+雷
+碓
+蕲
+楚
+罡
+焖
+抽
+妫
+咒
+仑
+闱
+尽
+邑
+菁
+爱
+贷
+沥
+鞑
+牡
+嗉
+崴
+骤
+塌
+嗦
+订
+拮
+滓
+捡
+锻
+次
+坪
+杩
+臃
+箬
+融
+珂
+鹗
+宗
+枚
+降
+鸬
+妯
+阄
+堰
+盐
+毅
+必
+杨
+崃
+俺
+甬
+状
+莘
+货
+耸
+菱
+腼
+铸
+唏
+痤
+孚
+澳
+懒
+溅
+翘
+疙
+杷
+淼
+缙
+骰
+喊
+悉
+砻
+坷
+艇
+赁
+界
+谤
+纣
+宴
+晃
+茹
+归
+饭
+梢
+铡
+街
+抄
+肼
+鬟
+苯
+颂
+撷
+戈
+炒
+咆
+茭
+瘙
+负
+仰
+客
+琉
+铢
+封
+卑
+珥
+椿
+镧
+窨
+鬲
+寿
+御
+袤
+铃
+萎
+砖
+餮
+脒
+裳
+肪
+孕
+嫣
+馗
+嵇
+恳
+氯
+江
+石
+褶
+冢
+祸
+阻
+狈
+羞
+银
+靳
+透
+咳
+叼
+敷
+芷
+啥
+它
+瓤
+兰
+痘
+懊
+逑
+肌
+往
+捺
+坊
+甩
+呻
+〃
+沦
+忘
+膻
+祟
+菅
+剧
+崆
+智
+坯
+臧
+霍
+墅
+攻
+眯
+倘
+拢
+骠
+铐
+庭
+岙
+瓠
+′
+缺
+泥
+迢
+捶
+?
+？
+郏
+喙
+掷
+沌
+纯
+秘
+种
+听
+绘
+固
+螨
+团
+香
+盗
+妒
+埚
+蓝
+拖
+旱
+荞
+铀
+血
+遏
+汲
+辰
+叩
+拽
+幅
+硬
+惶
+桀
+漠
+措
+泼
+唑
+齐
+肾
+念
+酱
+虚
+屁
+耶
+旗
+砦
+闵
+婉
+馆
+拭
+绅
+韧
+忏
+窝
+醋
+葺
+顾
+辞
+倜
+堆
+辋
+逆
+玟
+贱
+疾
+董
+惘
+倌
+锕
+淘
+嘀
+莽
+俭
+笏
+绑
+鲷
+杈
+择
+蟀
+粥
+嗯
+驰
+逾
+案
+谪
+褓
+胫
+哩
+昕
+颚
+鲢
+绠
+躺
+鹄
+崂
+儒
+俨
+丝
+尕
+泌
+啊
+萸
+彰
+幺
+吟
+骄
+苣
+弦
+脊
+瑰
+〈
+诛
+镁
+析
+闪
+剪
+侧
+哟
+框
+螃
+守
+嬗
+燕
+狭
+铈
+缮
+概
+迳
+痧
+鲲
+俯
+售
+笼
+痣
+扉
+挖
+满
+咋
+援
+邱
+扇
+歪
+便
+玑
+绦
+峡
+蛇
+叨
+〖
+泽
+胃
+斓
+喋
+怂
+坟
+猪
+该
+蚬
+炕
+弥
+赞
+棣
+晔
+娠
+挲
+狡
+创
+疖
+铕
+镭
+稷
+挫
+弭
+啾
+翔
+粉
+履
+苘
+哦
+楼
+秕
+铂
+土
+锣
+瘟
+挣
+栉
+习
+享
+桢
+袅
+磨
+桂
+谦
+延
+坚
+蔚
+噗
+署
+谟
+猬
+钎
+恐
+嬉
+雒
+倦
+衅
+亏
+璩
+睹
+刻
+殿
+王
+算
+雕
+麻
+丘
+柯
+骆
+丸
+塍
+谚
+添
+鲈
+垓
+桎
+蚯
+芥
+予
+飕
+镦
+谌
+窗
+醚
+菀
+亮
+搪
+莺
+蒿
+羁
+足
+J
+真
+轶
+悬
+衷
+靛
+翊
+掩
+哒
+炅
+掐
+冼
+妮
+l
+谐
+稚
+荆
+擒
+犯
+陵
+虏
+浓
+崽
+刍
+陌
+傻
+孜
+千
+靖
+演
+矜
+钕
+煽
+杰
+酗
+渗
+伞
+栋
+俗
+泫
+戍
+罕
+沾
+疽
+灏
+煦
+芬
+磴
+叱
+阱
+榉
+湃
+蜀
+叉
+醒
+彪
+租
+郡
+篷
+屎
+良
+垢
+隗
+弱
+陨
+峪
+砷
+掴
+颁
+胎
+雯
+绵
+贬
+沐
+撵
+隘
+篙
+暖
+曹
+陡
+栓
+填
+臼
+彦
+瓶
+琪
+潼
+哪
+鸡
+摩
+啦
+俟
+锋
+域
+耻
+蔫
+疯
+纹
+撇
+毒
+绶
+痛
+酯
+忍
+爪
+赳
+歆
+嘹
+辕
+烈
+册
+朴
+钱
+吮
+毯
+癜
+娃
+谀
+邵
+厮
+炽
+璞
+邃
+丐
+追
+词
+瓒
+忆
+轧
+芫
+谯
+喷
+弟
+半
+冕
+裙
+掖
+墉
+绮
+寝
+苔
+势
+顷
+褥
+切
+衮
+君
+佳
+嫒
+蚩
+霞
+佚
+洙
+逊
+镖
+暹
+唛
+&
+殒
+顶
+碗
+獗
+轭
+铺
+蛊
+废
+恹
+汨
+崩
+珍
+那
+杵
+曲
+纺
+夏
+薰
+傀
+闳
+淬
+姘
+舀
+拧
+卷
+楂
+恍
+讪
+厩
+寮
+篪
+赓
+乘
+灭
+盅
+鞣
+沟
+慎
+挂
+饺
+鼾
+杳
+树
+缨
+丛
+絮
+娌
+臻
+嗳
+篡
+侩
+述
+衰
+矛
+圈
+蚜
+匕
+筹
+匿
+濞
+晨
+叶
+骋
+郝
+挚
+蚴
+滞
+增
+侍
+描
+瓣
+吖
+嫦
+蟒
+匾
+圣
+赌
+毡
+癞
+恺
+百
+曳
+需
+篓
+肮
+庖
+帏
+卿
+驿
+遗
+蹬
+鬓
+骡
+歉
+芎
+胳
+屐
+禽
+烦
+晌
+寄
+媾
+狄
+翡
+苒
+船
+廉
+终
+痞
+殇
+々
+畦
+饶
+改
+拆
+悻
+萄
+￡
+瓿
+乃
+訾
+桅
+匮
+溧
+拥
+纱
+铍
+骗
+蕃
+龋
+缬
+父
+佐
+疚
+栎
+醍
+掳
+蓄
+x
+惆
+颜
+鲆
+榆
+〔
+猎
+敌
+暴
+谥
+鲫
+贾
+罗
+玻
+缄
+扦
+芪
+癣
+落
+徒
+臾
+恿
+猩
+托
+邴
+肄
+牵
+春
+陛
+耀
+刊
+拓
+蓓
+邳
+堕
+寇
+枉
+淌
+啡
+湄
+兽
+酷
+萼
+碚
+濠
+萤
+夹
+旬
+戮
+梭
+琥
+椭
+昔
+勺
+蜊
+绐
+晚
+孺
+僵
+宣
+摄
+冽
+旨
+萌
+忙
+蚤
+眉
+噼
+蟑
+付
+契
+瓜
+悼
+颡
+壁
+曾
+窕
+颢
+澎
+仿
+俑
+浑
+嵌
+浣
+乍
+碌
+褪
+乱
+蔟
+隙
+玩
+剐
+葫
+箫
+纲
+围
+伐
+决
+伙
+漩
+瑟
+刑
+肓
+镳
+缓
+蹭
+氨
+皓
+典
+畲
+坍
+铑
+檐
+塑
+洞
+倬
+储
+胴
+淳
+戾
+吐
+灼
+惺
+妙
+毕
+珐
+缈
+虱
+盖
+羰
+鸿
+磅
+谓
+髅
+娴
+苴
+唷
+蚣
+霹
+抨
+贤
+唠
+犬
+誓
+逍
+庠
+逼
+麓
+籼
+釉
+呜
+碧
+秧
+氩
+摔
+霄
+穸
+纨
+辟
+妈
+映
+完
+牛
+缴
+嗷
+炊
+恩
+荔
+茆
+掉
+紊
+慌
+莓
+羟
+阙
+萁
+磐
+另
+蕹
+辱
+鳐
+湮
+吡
+吩
+唐
+睦
+垠
+舒
+圜
+冗
+瞿
+溺
+芾
+囱
+匠
+僳
+汐
+菩
+饬
+漓
+黑
+霰
+浸
+濡
+窥
+毂
+蒡
+兢
+驻
+鹉
+芮
+诙
+迫
+雳
+厂
+忐
+臆
+猴
+鸣
+蚪
+栈
+箕
+羡
+渐
+莆
+捍
+眈
+哓
+趴
+蹼
+埕
+嚣
+骛
+宏
+淄
+斑
+噜
+严
+瑛
+垃
+椎
+诱
+压
+庾
+绞
+焘
+廿
+抡
+迄
+棘
+夫
+纬
+锹
+眨
+瞌
+侠
+脐
+竞
+瀑
+孳
+骧
+遁
+姜
+颦
+荪
+滚
+萦
+伪
+逸
+粳
+爬
+锁
+矣
+役
+趣
+洒
+颔
+诏
+逐
+奸
+甭
+惠
+攀
+蹄
+泛
+尼
+拼
+阮
+鹰
+亚
+颈
+惑
+勒
+〉
+际
+肛
+爷
+刚
+钨
+丰
+养
+冶
+鲽
+辉
+蔻
+画
+覆
+皴
+妊
+麦
+返
+醉
+皂
+擀
+〗
+酶
+凑
+粹
+悟
+诀
+硖
+港
+卜
+z
+杀
+涕
+±
+舍
+铠
+抵
+弛
+段
+敝
+镐
+奠
+拂
+轴
+跛
+袱
+e
+t
+沉
+菇
+俎
+薪
+峦
+秭
+蟹
+历
+盟
+菠
+寡
+液
+肢
+喻
+染
+裱
+悱
+抱
+氙
+赤
+捅
+猛
+跑
+氮
+谣
+仁
+尺
+辊
+窍
+烙
+衍
+架
+擦
+倏
+璐
+瑁
+币
+楞
+胖
+夔
+趸
+邛
+惴
+饕
+虔
+蝎
+§
+哉
+贝
+宽
+辫
+炮
+扩
+饲
+籽
+魏
+菟
+锰
+伍
+猝
+末
+琳
+哚
+蛎
+邂
+呀
+姿
+鄞
+却
+歧
+仙
+恸
+椐
+森
+牒
+寤
+袒
+婆
+虢
+雅
+钉
+朵
+贼
+欲
+苞
+寰
+故
+龚
+坭
+嘘
+咫
+礼
+硷
+兀
+睢
+汶
+’
+铲
+烧
+绕
+诃
+浃
+钿
+哺
+柜
+讼
+颊
+璁
+腔
+洽
+咐
+脲
+簌
+筠
+镣
+玮
+鞠
+谁
+兼
+姆
+挥
+梯
+蝴
+谘
+漕
+刷
+躏
+宦
+弼
+b
+垌
+劈
+麟
+莉
+揭
+笙
+渎
+仕
+嗤
+仓
+配
+怏
+抬
+错
+泯
+镊
+孰
+猿
+邪
+仍
+秋
+鼬
+壹
+歇
+吵
+炼
+<
+尧
+射
+柬
+廷
+胧
+霾
+凳
+隋
+肚
+浮
+梦
+祥
+株
+堵
+退
+L
+鹫
+跎
+凶
+毽
+荟
+炫
+栩
+玳
+甜
+沂
+鹿
+顽
+伯
+爹
+赔
+蛴
+徐
+匡
+欣
+狰
+缸
+雹
+蟆
+疤
+默
+沤
+啜
+痂
+衣
+禅
+w
+i
+h
+辽
+葳
+黝
+钗
+停
+沽
+棒
+馨
+颌
+肉
+吴
+硫
+悯
+劾
+娈
+马
+啧
+吊
+悌
+镑
+峭
+帆
+瀣
+涉
+咸
+疸
+滋
+泣
+翦
+拙
+癸
+钥
+蜒
++
+尾
+庄
+凝
+泉
+婢
+渴
+谊
+乞
+陆
+锉
+糊
+鸦
+淮
+I
+B
+N
+晦
+弗
+乔
+庥
+葡
+尻
+席
+橡
+傣
+渣
+拿
+惩
+麋
+斛
+缃
+矮
+蛏
+岘
+鸽
+姐
+膏
+催
+奔
+镒
+喱
+蠡
+摧
+钯
+胤
+柠
+拐
+璋
+鸥
+卢
+荡
+倾
+^
+_
+珀
+逄
+萧
+塾
+掇
+贮
+笆
+聂
+圃
+冲
+嵬
+M
+滔
+笕
+值
+炙
+偶
+蜱
+搐
+梆
+汪
+蔬
+腑
+鸯
+蹇
+敞
+绯
+仨
+祯
+谆
+梧
+糗
+鑫
+啸
+豺
+囹
+猾
+巢
+柄
+瀛
+筑
+踌
+沭
+暗
+苁
+鱿
+蹉
+脂
+蘖
+牢
+热
+木
+吸
+溃
+宠
+序
+泞
+偿
+拜
+檩
+厚
+朐
+毗
+螳
+吞
+媚
+朽
+担
+蝗
+橘
+畴
+祈
+糟
+盱
+隼
+郜
+惜
+珠
+裨
+铵
+焙
+琚
+唯
+咚
+噪
+骊
+丫
+滢
+勤
+棉
+呸
+咣
+淀
+隔
+蕾
+窈
+饨
+挨
+煅
+短
+匙
+粕
+镜
+赣
+撕
+墩
+酬
+馁
+豌
+颐
+抗
+酣
+氓
+佑
+搁
+哭
+递
+耷
+涡
+桃
+贻
+碣
+截
+瘦
+昭
+镌
+蔓
+氚
+甲
+猕
+蕴
+蓬
+散
+拾
+纛
+狼
+猷
+铎
+埋
+旖
+矾
+讳
+囊
+糜
+迈
+粟
+蚂
+紧
+鲳
+瘢
+栽
+稼
+羊
+锄
+斟
+睁
+桥
+瓮
+蹙
+祉
+醺
+鼻
+昱
+剃
+跳
+篱
+跷
+蒜
+翎
+宅
+晖
+嗑
+壑
+峻
+癫
+屏
+狠
+陋
+袜
+途
+憎
+祀
+莹
+滟
+佶
+溥
+臣
+约
+盛
+峰
+磁
+慵
+婪
+拦
+莅
+朕
+鹦
+粲
+裤
+哎
+疡
+嫖
+琵
+窟
+堪
+谛
+嘉
+儡
+鳝
+斩
+郾
+驸
+酊
+妄
+胜
+贺
+徙
+傅
+噌
+钢
+栅
+庇
+恋
+匝
+巯
+邈
+尸
+锚
+粗
+佟
+蛟
+薹
+纵
+蚊
+郅
+绢
+锐
+苗
+俞
+篆
+淆
+膀
+鲜
+煎
+诶
+秽
+寻
+涮
+刺
+怀
+噶
+巨
+褰
+魅
+灶
+灌
+桉
+藕
+谜
+舸
+薄
+搀
+恽
+借
+牯
+痉
+渥
+愿
+亓
+耘
+杠
+柩
+锔
+蚶
+钣
+珈
+喘
+蹒
+幽
+赐
+稗
+晤
+莱
+泔
+扯
+肯
+菪
+裆
+腩
+豉
+疆
+骜
+腐
+倭
+珏
+唔
+粮
+亡
+润
+慰
+伽
+橄
+玄
+誉
+醐
+胆
+龊
+粼
+塬
+陇
+彼
+削
+嗣
+绾
+芽
+妗
+垭
+瘴
+爽
+薏
+寨
+龈
+泠
+弹
+赢
+漪
+猫
+嘧
+涂
+恤
+圭
+茧
+烽
+屑
+痕
+巾
+赖
+荸
+凰
+腮
+畈
+亵
+蹲
+偃
+苇
+澜
+艮
+换
+骺
+烘
+苕
+梓
+颉
+肇
+哗
+悄
+氤
+涠
+葬
+屠
+鹭
+植
+竺
+佯
+诣
+鲇
+瘀
+鲅
+邦
+移
+滁
+冯
+耕
+癔
+戌
+茬
+沁
+巩
+悠
+湘
+洪
+痹
+锟
+循
+谋
+腕
+鳃
+钠
+捞
+焉
+迎
+碱
+伫
+急
+榷
+奈
+邝
+卯
+辄
+皲
+卟
+醛
+畹
+忧
+稳
+雄
+昼
+缩
+阈
+睑
+扌
+耗
+曦
+涅
+捏
+瞧
+邕
+淖
+漉
+铝
+耦
+禹
+湛
+喽
+莼
+琅
+诸
+苎
+纂
+硅
+始
+嗨
+傥
+燃
+臂
+赅
+嘈
+呆
+贵
+屹
+壮
+肋
+亍
+蚀
+卅
+豹
+腆
+邬
+迭
+浊
+}
+童
+螂
+捐
+圩
+勐
+触
+寞
+汊
+壤
+荫
+膺
+渌
+芳
+懿
+遴
+螈
+泰
+蓼
+蛤
+茜
+舅
+枫
+朔
+膝
+眙
+避
+梅
+判
+鹜
+璜
+牍
+缅
+垫
+藻
+黔
+侥
+惚
+懂
+踩
+腰
+腈
+札
+丞
+唾
+慈
+顿
+摹
+荻
+琬
+~
+斧
+沈
+滂
+胁
+胀
+幄
+莜
+Z
+匀
+鄄
+掌
+绰
+茎
+焚
+赋
+萱
+谑
+汁
+铒
+瞎
+夺
+蜗
+野
+娆
+冀
+弯
+篁
+懵
+灞
+隽
+芡
+脘
+俐
+辩
+芯
+掺
+喏
+膈
+蝈
+觐
+悚
+踹
+蔗
+熠
+鼠
+呵
+抓
+橼
+峨
+畜
+缔
+禾
+崭
+弃
+熊
+摒
+凸
+拗
+穹
+蒙
+抒
+祛
+劝
+闫
+扳
+阵
+醌
+踪
+喵
+侣
+搬
+仅
+荧
+赎
+蝾
+琦
+买
+婧
+瞄
+寓
+皎
+冻
+赝
+箩
+莫
+瞰
+郊
+笫
+姝
+筒
+枪
+遣
+煸
+袋
+舆
+痱
+涛
+母
+〇
+启
+践
+耙
+绲
+盘
+遂
+昊
+搞
+槿
+诬
+纰
+泓
+惨
+檬
+亻
+越
+C
+o
+憩
+熵
+祷
+钒
+暧
+塔
+阗
+胰
+咄
+娶
+魔
+琶
+钞
+邻
+扬
+杉
+殴
+咽
+弓
+〆
+髻
+】
+吭
+揽
+霆
+拄
+殖
+脆
+彻
+岩
+芝
+勃
+辣
+剌
+钝
+嘎
+甄
+佘
+皖
+伦
+授
+徕
+憔
+挪
+皇
+庞
+稔
+芜
+踏
+溴
+兖
+卒
+擢
+饥
+鳞
+煲
+‰
+账
+颗
+叻
+斯
+捧
+鳍
+琮
+讹
+蛙
+纽
+谭
+酸
+兔
+莒
+睇
+伟
+觑
+羲
+嗜
+宜
+褐
+旎
+辛
+卦
+诘
+筋
+鎏
+溪
+挛
+熔
+阜
+晰
+鳅
+丢
+奚
+灸
+呱
+献
+陉
+黛
+鸪
+甾
+萨
+疮
+拯
+洲
+疹
+辑
+叙
+恻
+谒
+允
+柔
+烂
+氏
+逅
+漆
+拎
+惋
+扈
+湟
+纭
+啕
+掬
+擞
+哥
+忽
+涤
+鸵
+靡
+郗
+瓷
+扁
+廊
+怨
+雏
+钮
+敦
+E
+懦
+憋
+汀
+拚
+啉
+腌
+岸
+f
+痼
+瞅
+尊
+咀
+眩
+飙
+忌
+仝
+迦
+熬
+毫
+胯
+篑
+茄
+腺
+凄
+舛
+碴
+锵
+诧
+羯
+後
+漏
+汤
+宓
+仞
+蚁
+壶
+谰
+皑
+铄
+棰
+罔
+辅
+晶
+苦
+牟
+闽
+\
+烃
+饮
+聿
+丙
+蛳
+朱
+煤
+涔
+鳖
+犁
+罐
+荼
+砒
+淦
+妤
+黏
+戎
+孑
+婕
+瑾
+戢
+钵
+枣
+捋
+砥
+衩
+狙
+桠
+稣
+阎
+肃
+梏
+诫
+孪
+昶
+婊
+衫
+嗔
+侃
+塞
+蜃
+樵
+峒
+貌
+屿
+欺
+缫
+阐
+栖
+诟
+珞
+荭
+吝
+萍
+嗽
+恂
+啻
+蜴
+磬
+峋
+俸
+豫
+谎
+徊
+镍
+韬
+魇
+晴
+U
+囟
+猜
+蛮
+坐
+囿
+伴
+亭
+肝
+佗
+蝠
+妃
+胞
+滩
+榴
+氖
+垩
+苋
+砣
+扪
+馏
+姓
+轩
+厉
+夥
+侈
+禀
+垒
+岑
+赏
+钛
+辐
+痔
+披
+纸
+碳
+“
+坞
+蠓
+挤
+荥
+沅
+悔
+铧
+帼
+蒌
+蝇
+a
+p
+y
+n
+g
+哀
+浆
+瑶
+凿
+桶
+馈
+皮
+奴
+苜
+佤
+伶
+晗
+铱
+炬
+优
+弊
+氢
+恃
+甫
+攥
+端
+锌
+灰
+稹
+炝
+曙
+邋
+亥
+眶
+碾
+拉
+萝
+绔
+捷
+浍
+腋
+姑
+菖
+凌
+涞
+麽
+锢
+桨
+潢
+绎
+镰
+殆
+锑
+渝
+铬
+困
+绽
+觎
+匈
+糙
+暑
+裹
+鸟
+盔
+肽
+迷
+綦
+『
+亳
+佝
+俘
+钴
+觇
+骥
+仆
+疝
+跪
+婶
+郯
+瀹
+唉
+脖
+踞
+针
+晾
+忒
+扼
+瞩
+叛
+椒
+疟
+嗡
+邗
+肆
+跆
+玫
+忡
+捣
+咧
+唆
+艄
+蘑
+潦
+笛
+阚
+沸
+泻
+掊
+菽
+贫
+斥
+髂
+孢
+镂
+赂
+麝
+鸾
+屡
+衬
+苷
+恪
+叠
+希
+粤
+爻
+喝
+茫
+惬
+郸
+绻
+庸
+撅
+碟
+宄
+妹
+膛
+叮
+饵
+崛
+嗲
+椅
+冤
+搅
+咕
+敛
+尹
+垦
+闷
+蝉
+霎
+勰
+败
+蓑
+泸
+肤
+鹌
+幌
+焦
+浠
+鞍
+刁
+舰
+乙
+竿
+裔
+。
+茵
+函
+伊
+兄
+丨
+娜
+匍
+謇
+莪
+宥
+似
+蝽
+翳
+酪
+翠
+粑
+薇
+祢
+骏
+赠
+叫
+Q
+噤
+噻
+竖
+芗
+莠
+潭
+俊
+羿
+耜
+O
+郫
+趁
+嗪
+囚
+蹶
+芒
+洁
+笋
+鹑
+敲
+硝
+啶
+堡
+渲
+揩
+』
+携
+宿
+遒
+颍
+扭
+棱
+割
+萜
+蔸
+葵
+琴
+捂
+饰
+衙
+耿
+掠
+募
+岂
+窖
+涟
+蔺
+瘤
+柞
+瞪
+怜
+匹
+距
+楔
+炜
+哆
+秦
+缎
+幼
+茁
+绪
+痨
+恨
+楸
+娅
+瓦
+桩
+雪
+嬴
+伏
+榔
+妥
+铿
+拌
+眠
+雍
+缇
+‘
+卓
+搓
+哌
+觞
+噩
+屈
+哧
+髓
+咦
+巅
+娑
+侑
+淫
+膳
+祝
+勾
+姊
+莴
+胄
+疃
+薛
+蜷
+胛
+巷
+芙
+芋
+熙
+闰
+勿
+窃
+狱
+剩
+钏
+幢
+陟
+铛
+慧
+靴
+耍
+k
+浙
+浇
+飨
+惟
+绗
+祜
+澈
+啼
+咪
+磷
+摞
+诅
+郦
+抹
+跃
+壬
+吕
+肖
+琏
+颤
+尴
+剡
+抠
+凋
+赚
+泊
+津
+宕
+殷
+倔
+氲
+漫
+邺
+涎
+怠
+$
+垮
+荬
+遵
+俏
+叹
+噢
+饽
+蜘
+孙
+筵
+疼
+鞭
+羧
+牦
+箭
+潴
+c
+眸
+祭
+髯
+啖
+坳
+愁
+芩
+驮
+倡
+巽
+穰
+沃
+胚
+怒
+凤
+槛
+剂
+趵
+嫁
+v
+邢
+灯
+鄢
+桐
+睽
+檗
+锯
+槟
+婷
+嵋
+圻
+诗
+蕈
+颠
+遭
+痢
+芸
+怯
+馥
+竭
+锗
+徜
+恭
+遍
+籁
+剑
+嘱
+苡
+龄
+僧
+桑
+潸
+弘
+澶
+楹
+悲
+讫
+愤
+腥
+悸
+谍
+椹
+呢
+桓
+葭
+攫
+阀
+翰
+躲
+敖
+柑
+郎
+笨
+橇
+呃
+魁
+燎
+脓
+葩
+磋
+垛
+玺
+狮
+沓
+砜
+蕊
+锺
+罹
+蕉
+翱
+虐
+闾
+巫
+旦
+茱
+嬷
+枯
+鹏
+贡
+芹
+汛
+矫
+绁
+拣
+禺
+佃
+讣
+舫
+惯
+乳
+趋
+疲
+挽
+岚
+虾
+衾
+蠹
+蹂
+飓
+氦
+铖
+孩
+稞
+瑜
+壅
+掀
+勘
+妓
+畅
+髋
+W
+庐
+牲
+蓿
+榕
+练
+垣
+唱
+邸
+菲
+昆
+婺
+穿
+绡
+麒
+蚱
+掂
+愚
+泷
+涪
+漳
+妩
+娉
+榄
+讷
+觅
+旧
+藤
+煮
+呛
+柳
+腓
+叭
+庵
+烷
+阡
+罂
+蜕
+擂
+猖
+咿
+媲
+脉
+【
+沏
+貅
+黠
+熏
+哲
+烁
+坦
+酵
+兜
+×
+潇
+撒
+剽
+珩
+圹
+乾
+摸
+樟
+帽
+嗒
+襄
+魂
+轿
+憬
+锡
+〕
+喃
+皆
+咖
+隅
+脸
+残
+泮
+袂
+鹂
+珊
+囤
+捆
+咤
+误
+徨
+闹
+淙
+芊
+淋
+怆
+囗
+拨
+梳
+渤
+R
+G
+绨
+蚓
+婀
+幡
+狩
+麾
+谢
+唢
+裸
+旌
+伉
+纶
+裂
+驳
+砼
+咛
+澄
+樨
+蹈
+宙
+澍
+倍
+貔
+操
+勇
+蟠
+摈
+砧
+虬
+够
+缁
+悦
+藿
+撸
+艹
+摁
+淹
+豇
+虎
+榭
+ˉ
+吱
+d
+°
+喧
+荀
+踱
+侮
+奋
+偕
+饷
+犍
+惮
+坑
+璎
+徘
+宛
+妆
+袈
+倩
+窦
+昂
+荏
+乖
+K
+怅
+撰
+鳙
+牙
+袁
+酞
+X
+痿
+琼
+闸
+雁
+趾
+荚
+虻
+涝
+《
+杏
+韭
+偈
+烤
+绫
+鞘
+卉
+症
+遢
+蓥
+诋
+杭
+荨
+匆
+竣
+簪
+辙
+敕
+虞
+丹
+缭
+咩
+黟
+m
+淤
+瑕
+咂
+铉
+硼
+茨
+嶂
+痒
+畸
+敬
+涿
+粪
+窘
+熟
+叔
+嫔
+盾
+忱
+裘
+憾
+梵
+赡
+珙
+咯
+娘
+庙
+溯
+胺
+葱
+痪
+摊
+荷
+卞
+乒
+髦
+寐
+铭
+坩
+胗
+枷
+爆
+溟
+嚼
+羚
+砬
+轨
+惊
+挠
+罄
+竽
+菏
+氧
+浅
+楣
+盼
+枢
+炸
+阆
+杯
+谏
+噬
+淇
+渺
+俪
+秆
+墓
+泪
+跻
+砌
+痰
+垡
+渡
+耽
+釜
+讶
+鳎
+煞
+呗
+韶
+舶
+绷
+鹳
+缜
+旷
+铊
+皱
+龌
+檀
+霖
+奄
+槐
+艳
+蝶
+旋
+哝
+赶
+骞
+蚧
+腊
+盈
+丁
+`
+蜚
+矸
+蝙
+睨
+嚓
+僻
+鬼
+醴
+夜
+彝
+磊
+笔
+拔
+栀
+糕
+厦
+邰
+纫
+逭
+纤
+眦
+膊
+馍
+躇
+烯
+蘼
+冬
+诤
+暄
+骶
+哑
+瘠
+」
+臊
+丕
+愈
+咱
+螺
+擅
+跋
+搏
+硪
+谄
+笠
+淡
+嘿
+骅
+谧
+鼎
+皋
+姚
+歼
+蠢
+驼
+耳
+胬
+挝
+涯
+狗
+蒽
+孓
+犷
+凉
+芦
+箴
+铤
+孤
+嘛
+坤
+V
+茴
+朦
+挞
+尖
+橙
+诞
+搴
+碇
+洵
+浚
+帚
+蜍
+漯
+柘
+嚎
+讽
+芭
+荤
+咻
+祠
+秉
+跖
+埃
+吓
+糯
+眷
+馒
+惹
+娼
+鲑
+嫩
+讴
+轮
+瞥
+靶
+褚
+乏
+缤
+宋
+帧
+删
+驱
+碎
+扑
+俩
+俄
+偏
+涣
+竹
+噱
+皙
+佰
+渚
+唧
+斡
+#
+镉
+刀
+崎
+筐
+佣
+夭
+贰
+肴
+峙
+哔
+艿
+匐
+牺
+镛
+缘
+仡
+嫡
+劣
+枸
+堀
+梨
+簿
+鸭
+蒸
+亦
+稽
+浴
+{
+衢
+束
+槲
+j
+阁
+揍
+疥
+棋
+潋
+聪
+窜
+乓
+睛
+插
+冉
+阪
+苍
+搽
+「
+蟾
+螟
+幸
+仇
+樽
+撂
+慢
+跤
+幔
+俚
+淅
+覃
+觊
+溶
+妖
+帛
+侨
+曰
+妾
+泗
+·
+：
+瀘
+風
+Ë
+（
+）
+∶
+紅
+紗
+瑭
+雲
+頭
+鶏
+財
+許
+•
+¥
+樂
+焗
+麗
+—
+；
+滙
+東
+榮
+繪
+興
+…
+門
+業
+π
+楊
+國
+顧
+é
+盤
+寳
+Λ
+龍
+鳳
+島
+誌
+緣
+結
+銭
+萬
+勝
+祎
+璟
+優
+歡
+臨
+時
+購
+＝
+★
+藍
+昇
+鐵
+觀
+勅
+農
+聲
+畫
+兿
+術
+發
+劉
+記
+專
+耑
+園
+書
+壴
+種
+Ο
+●
+褀
+號
+銀
+匯
+敟
+锘
+葉
+橪
+廣
+進
+蒄
+鑽
+阝
+祙
+貢
+鍋
+豊
+夬
+喆
+團
+閣
+開
+燁
+賓
+館
+酡
+沔
+順
+＋
+硚
+劵
+饸
+陽
+車
+湓
+復
+萊
+氣
+軒
+華
+堃
+迮
+纟
+戶
+馬
+學
+裡
+電
+嶽
+獨
+マ
+シ
+サ
+ジ
+燘
+袪
+環
+❤
+臺
+灣
+専
+賣
+孖
+聖
+攝
+線
+▪
+α
+傢
+俬
+夢
+達
+莊
+喬
+貝
+薩
+劍
+羅
+壓
+棛
+饦
+尃
+璈
+囍
+醫
+Ｇ
+Ｉ
+Ａ
+＃
+Ｎ
+鷄
+髙
+嬰
+啓
+約
+隹
+潔
+賴
+藝
+～
+寶
+籣
+麺
+　
+嶺
+√
+義
+網
+峩
+長
+∧
+魚
+機
+構
+②
+鳯
+偉
+Ｌ
+Ｂ
+㙟
+畵
+鴿
+＇
+詩
+溝
+嚞
+屌
+藔
+佧
+玥
+蘭
+織
+１
+３
+９
+０
+７
+點
+砭
+鴨
+鋪
+銘
+廳
+弍
+‧
+創
+湯
+坶
+℃
+卩
+骝
+＆
+烜
+荘
+當
+潤
+扞
+係
+懷
+碶
+钅
+蚨
+讠
+☆
+叢
+爲
+埗
+涫
+塗
+→
+楽
+現
+鯨
+愛
+瑪
+鈺
+忄
+悶
+藥
+飾
+樓
+視
+孬
+ㆍ
+燚
+苪
+師
+①
+丼
+锽
+│
+韓
+標
+è
+兒
+閏
+匋
+張
+漢
+Ü
+髪
+會
+閑
+檔
+習
+裝
+の
+峯
+菘
+輝
+И
+雞
+釣
+億
+浐
+Ｋ
+Ｏ
+Ｒ
+８
+Ｈ
+Ｅ
+Ｐ
+Ｔ
+Ｗ
+Ｄ
+Ｓ
+Ｃ
+Ｍ
+Ｆ
+姌
+饹
+»
+晞
+廰
+ä
+嵯
+鷹
+負
+飲
+絲
+冚
+楗
+澤
+綫
+區
+❋
+←
+質
+靑
+揚
+③
+滬
+統
+産
+協
+﹑
+乸
+畐
+經
+運
+際
+洺
+岽
+為
+粵
+諾
+崋
+豐
+碁
+ɔ
+Ｖ
+２
+６
+齋
+誠
+訂
+´
+勑
+雙
+陳
+無
+í
+泩
+媄
+夌
+刂
+ｉ
+ｃ
+ｔ
+ｏ
+ｒ
+ａ
+嘢
+耄
+燴
+暃
+壽
+媽
+靈
+抻
+體
+唻
+É
+冮
+甹
+鎮
+錦
+ʌ
+蜛
+蠄
+尓
+駕
+戀
+飬
+逹
+倫
+貴
+極
+Я
+Й
+寬
+磚
+嶪
+郎
+職
+｜
+間
+ｎ
+ｄ
+剎
+伈
+課
+飛
+橋
+瘊
+№
+譜
+骓
+圗
+滘
+縣
+粿
+咅
+養
+濤
+彳
+®
+％
+Ⅱ
+啰
+㴪
+見
+矞
+薬
+糁
+邨
+鲮
+顔
+罱
+З
+選
+話
+贏
+氪
+俵
+競
+瑩
+繡
+枱
+β
+綉
+á
+獅
+爾
+™
+麵
+戋
+淩
+徳
+個
+劇
+場
+務
+簡
+寵
+ｈ
+實
+膠
+轱
+圖
+築
+嘣
+樹
+㸃
+營
+耵
+孫
+饃
+鄺
+飯
+麯
+遠
+輸
+坫
+孃
+乚
+閃
+鏢
+㎡
+題
+廠
+關
+↑
+爺
+將
+軍
+連
+篦
+覌
+參
+箸
+－
+窠
+棽
+寕
+夀
+爰
+歐
+呙
+閥
+頡
+熱
+雎
+垟
+裟
+凬
+勁
+帑
+馕
+夆
+疌
+枼
+馮
+貨
+蒤
+樸
+彧
+旸
+靜
+龢
+暢
+㐱
+鳥
+珺
+鏡
+灡
+爭
+堷
+廚
+Ó
+騰
+診
+┅
+蘇
+褔
+凱
+頂
+豕
+亞
+帥
+嘬
+⊥
+仺
+桖
+複
+饣
+絡
+穂
+顏
+棟
+納
+▏
+濟
+親
+設
+計
+攵
+埌
+烺
+ò
+頤
+燦
+蓮
+撻
+節
+講
+濱
+濃
+娽
+洳
+朿
+燈
+鈴
+護
+膚
+铔
+過
+補
+Ｚ
+Ｕ
+５
+４
+坋
+闿
+䖝
+餘
+缐
+铞
+貿
+铪
+桼
+趙
+鍊
+［
+㐂
+垚
+菓
+揸
+捲
+鐘
+滏
+𣇉
+爍
+輪
+燜
+鴻
+鮮
+動
+鹞
+鷗
+丄
+慶
+鉌
+翥
+飮
+腸
+⇋
+漁
+覺
+來
+熘
+昴
+翏
+鲱
+圧
+鄉
+萭
+頔
+爐
+嫚
+г
+貭
+類
+聯
+幛
+輕
+訓
+鑒
+夋
+锨
+芃
+珣
+䝉
+扙
+嵐
+銷
+處
+ㄱ
+語
+誘
+苝
+歸
+儀
+燒
+楿
+內
+粢
+葒
+奧
+麥
+礻
+滿
+蠔
+穵
+瞭
+態
+鱬
+榞
+硂
+鄭
+黃
+煙
+祐
+奓
+逺
+＊
+瑄
+獲
+聞
+薦
+讀
+這
+樣
+決
+問
+啟
+們
+執
+説
+轉
+單
+隨
+唘
+帶
+倉
+庫
+還
+贈
+尙
+皺
+■
+餅
+產
+○
+∈
+報
+狀
+楓
+賠
+琯
+嗮
+禮
+｀
+傳
+＞
+≤
+嗞
+Φ
+≥
+換
+咭
+∣
+↓
+曬
+ε
+応
+寫
+″
+終
+様
+純
+費
+療
+聨
+凍
+壐
+郵
+ü
+黒
+∫
+製
+塊
+調
+軽
+確
+撃
+級
+馴
+Ⅲ
+涇
+繹
+數
+碼
+證
+狒
+処
+劑
+＜
+晧
+賀
+衆
+］
+櫥
+兩
+陰
+絶
+對
+鯉
+憶
+◎
+ｐ
+ｅ
+Ｙ
+蕒
+煖
+頓
+測
+試
+鼽
+僑
+碩
+妝
+帯
+≈
+鐡
+舖
+權
+喫
+倆
+ˋ
+該
+悅
+ā
+俫
+．
+ｆ
+ｓ
+ｂ
+ｍ
+ｋ
+ｇ
+ｕ
+ｊ
+貼
+淨
+濕
+針
+適
+備
+ｌ
+／
+給
+謢
+強
+觸
+衛
+與
+⊙
+＄
+緯
+變
+⑴
+⑵
+⑶
+㎏
+殺
+∩
+幚
+─
+價
+▲
+離
+ú
+ó
+飄
+烏
+関
+閟
+﹝
+﹞
+邏
+輯
+鍵
+驗
+訣
+導
+歷
+屆
+層
+▼
+儱
+錄
+熳
+ē
+艦
+吋
+錶
+辧
+飼
+顯
+④
+禦
+販
+気
+対
+枰
+閩
+紀
+幹
+瞓
+貊
+淚
+△
+眞
+墊
+Ω
+獻
+褲
+縫
+緑
+亜
+鉅
+餠
+｛
+｝
+◆
+蘆
+薈
+█
+◇
+溫
+彈
+晳
+粧
+犸
+穩
+訊
+崬
+凖
+熥
+П
+舊
+條
+紋
+圍
+Ⅳ
+筆
+尷
+難
+雜
+錯
+綁
+識
+頰
+鎖
+艶
+□
+殁
+殼
+⑧
+├
+▕
+鵬
+ǐ
+ō
+ǒ
+糝
+綱
+▎
+μ
+盜
+饅
+醬
+籤
+蓋
+釀
+鹽
+據
+à
+ɡ
+辦
+◥
+彐
+┌
+婦
+獸
+鲩
+伱
+ī
+蒟
+蒻
+齊
+袆
+腦
+寧
+凈
+妳
+煥
+詢
+偽
+謹
+啫
+鯽
+騷
+鱸
+損
+傷
+鎻
+髮
+買
+冏
+儥
+両
+﹢
+∞
+載
+喰
+ｚ
+羙
+悵
+燙
+曉
+員
+組
+徹
+艷
+痠
+鋼
+鼙
+縮
+細
+嚒
+爯
+≠
+維
+＂
+鱻
+壇
+厍
+帰
+浥
+犇
+薡
+軎
+²
+應
+醜
+刪
+緻
+鶴
+賜
+噁
+軌
+尨
+镔
+鷺
+槗
+彌
+葚
+濛
+請
+溇
+緹
+賢
+訪
+獴
+瑅
+資
+縤
+陣
+蕟
+栢
+韻
+祼
+恁
+伢
+謝
+劃
+涑
+總
+衖
+踺
+砋
+凉
+籃
+駿
+苼
+瘋
+昽
+紡
+驊
+腎
+﹗
+響
+杋
+剛
+嚴
+禪
+歓
+槍
+傘
+檸
+檫
+炣
+勢
+鏜
+鎢
+銑
+尐
+減
+奪
+惡
+θ
+僮
+婭
+臘
+ū
+ì
+殻
+鉄
+∑
+蛲
+焼
+緖
+續
+紹
+懮!
+䰾
+䲁
+丌
+丏
+丟
+並
+乂
+乗
+乩
+乭
+乹
+亀
+亂
+亅
+亊
+亠
+亰
+亶
+亹
+仂
+仉
+仏
+仛
+仫
+仮
+仳
+仵
+仼
+伃
+伋
+伕
+伝
+伷
+伾
+佀
+佁
+佇
+佈
+佉
+佋
+佔
+併
+佹
+佺
+佾
+侁
+侅
+侊
+侖
+侘
+侚
+侞
+価
+侶
+侷
+侹
+俁
+俅
+俋
+俌
+俍
+俛
+俠
+俳
+俴
+俶
+俽
+倈
+倓
+倖
+倗
+倞
+倢
+倣
+値
+倧
+倮
+倻
+偁
+偊
+偍
+偓
+偪
+偲
+側
+偵
+偸
+傃
+傉
+傑
+傒
+傕
+傖
+傜
+傭
+債
+傾
+僅
+僉
+僊
+働
+僔
+僕
+僖
+僙
+僜
+僡
+僩
+僭
+僰
+僱
+僴
+儁
+儂
+儆
+儇
+儈
+儉
+儐
+儔
+儕
+儘
+儚
+儞
+償
+儦
+儫
+儲
+儷
+儺
+儻
+儼
+兌
+児
+兕
+兗
+兪
+冂
+円
+冇
+冊
+冑
+冖
+冧
+冨
+冪
+冫
+冴
+凃
+凜
+凞
+凪
+凵
+刄
+刎
+別
+刦
+刧
+刼
+則
+剋
+剏
+剝
+剣
+剮
+劄
+劊
+劌
+劔
+劬
+効
+劼
+勔
+勖
+勗
+勛
+勞
+勣
+勦
+勱
+勲
+勳
+勵
+勷
+勸
+勻
+匂
+匄
+匏
+匚
+匱
+匸
+卋
+卍
+卐
+卣
+卬
+卮
+卲
+卹
+卺
+卻
+卽
+厓
+厔
+厙
+厭
+厰
+厲
+厴
+厶
+叄
+収
+叕
+叡
+叵
+吔
+吥
+吳
+吶
+呂
+呉
+呎
+呾
+咁
+咑
+咗
+咘
+咟
+咥
+咲
+咼
+咾
+哂
+哏
+哐
+哖
+哱
+唃
+唄
+唫
+唭
+唵
+唸
+啁
+啍
+啚
+啞
+啣
+啯
+啱
+啲
+啷
+喈
+喚
+喢
+喦
+喪
+喲
+喼
+嗄
+嗆
+嗇
+嗊
+嗎
+嗚
+嗢
+嗩
+嗶
+嗹
+嘅
+嘆
+嘍
+嘏
+嘔
+嘗
+嘚
+嘜
+嘥
+嘩
+嘮
+嘯
+嘰
+嘸
+噍
+噏
+噓
+噝
+噠
+噥
+噦
+噯
+噰
+噲
+噴
+噸
+噹
+嚇
+嚈
+嚐
+嚕
+嚗
+嚙
+嚟
+嚤
+嚦
+嚧
+嚨
+嚩
+嚮
+嚳
+嚶
+嚿
+囀
+囂
+囃
+囉
+囑
+囒
+囓
+囝
+団
+囧
+囪
+囮
+囯
+囲
+図
+囶
+囷
+圂
+圄
+圉
+圏
+圓
+圪
+圯
+坌
+坖
+坣
+坬
+坮
+坵
+垈
+垍
+垕
+垞
+垯
+垰
+垵
+垻
+垿
+埅
+埇
+埈
+埏
+埒
+埜
+埡
+埤
+埧
+埨
+埪
+埮
+埴
+埵
+埻
+埼
+堅
+堈
+堉
+堊
+堍
+堖
+堝
+堦
+堮
+堯
+堺
+塀
+塅
+塆
+塋
+塏
+塙
+塜
+塡
+塢
+塤
+塨
+塩
+塭
+塰
+塱
+塲
+塵
+塹
+塽
+墀
+墎
+増
+墘
+墜
+墡
+墣
+墫
+墬
+墮
+墱
+墳
+墺
+墼
+墾
+壄
+壆
+壋
+壌
+壎
+壔
+壘
+壙
+壞
+壟
+壠
+壢
+壩
+壯
+壱
+壺
+変
+夊
+夠
+夤
+夾
+奀
+奐
+奣
+奩
+奫
+奭
+奮
+妀
+妁
+妏
+妑
+妠
+妧
+妭
+妸
+妺
+姀
+姁
+姃
+姈
+姉
+姍
+姦
+姪
+姫
+姮
+姵
+姶
+姸
+娋
+娍
+娎
+娖
+娛
+娫
+娳
+娸
+婁
+婑
+婯
+婻
+婼
+媃
+媊
+媐
+媓
+媖
+媗
+媜
+媞
+媧
+媭
+媯
+媺
+媼
+媿
+嫄
+嫈
+嫘
+嫪
+嫲
+嫳
+嫵
+嫺
+嫻
+嬅
+嬈
+嬋
+嬌
+嬛
+嬝
+嬡
+嬤
+嬨
+嬪
+嬬
+嬭
+嬸
+嬾
+嬿
+孀
+孆
+孋
+孌
+孮
+孻
+孿
+宍
+実
+宧
+宮
+寀
+寁
+寈
+寊
+寔
+寖
+寗
+寘
+寛
+寜
+寢
+審
+寯
+尋
+尗
+尢
+尪
+屄
+屇
+屍
+屓
+屚
+屜
+屢
+屬
+屭
+屺
+屻
+岀
+岈
+岡
+岣
+岧
+岪
+岬
+岰
+岵
+岻
+峅
+峇
+峍
+峘
+峚
+峠
+峴
+峼
+峽
+崁
+崈
+崍
+崐
+崑
+崒
+崗
+崘
+崙
+崚
+崞
+崟
+崠
+崢
+崱
+崵
+崶
+嵎
+嵒
+嵕
+嵖
+嵗
+嵙
+嵛
+嵜
+嵨
+嵮
+嵰
+嵴
+嵻
+嵿
+嶁
+嶃
+嶄
+嶇
+嶋
+嶌
+嶍
+嶒
+嶔
+嶗
+嶝
+嶠
+嶢
+嶦
+嶧
+嶬
+嶰
+嶲
+嶴
+嶷
+嶸
+嶼
+巂
+巄
+巆
+巋
+巌
+巎
+巑
+巒
+巔
+巖
+巘
+巛
+巰
+巶
+巻
+巿
+帔
+帙
+帡
+帢
+帳
+幀
+幃
+幗
+幟
+幣
+幪
+幫
+幵
+幷
+幾
+庀
+庁
+広
+庢
+庲
+庼
+廁
+廂
+廄
+廆
+廈
+廋
+廌
+廍
+廑
+廔
+廕
+廙
+廝
+廞
+廟
+廡
+廢
+廧
+廨
+廩
+廬
+廱
+廸
+廻
+廼
+弁
+弅
+弇
+弉
+弐
+弒
+弔
+弖
+弢
+弨
+弸
+弾
+彀
+彄
+彅
+彆
+彊
+彎
+彔
+彖
+彘
+彙
+彜
+彞
+彠
+彡
+彣
+彥
+彫
+彿
+徂
+徑
+從
+徠
+徧
+徫
+徬
+徭
+徴
+徸
+忉
+忝
+忞
+忬
+忯
+忳
+怍
+怙
+怛
+怵
+恆
+恊
+恥
+恵
+悆
+悛
+悝
+悞
+悧
+悪
+悰
+悳
+惇
+惔
+惣
+惱
+惲
+愃
+愆
+愍
+愐
+愒
+愔
+愜
+愨
+愭
+愴
+愷
+愼
+愾
+慄
+慘
+慚
+慜
+慟
+慣
+慥
+慮
+慳
+慾
+憂
+憊
+憍
+憐
+憑
+憓
+憕
+憙
+憚
+憤
+憫
+憲
+憺
+憻
+懃
+懇
+懌
+懍
+懐
+懣
+懮
+懲
+懶
+懸
+懺
+懼
+懽
+懾
+戇
+戔
+戕
+戙
+戡
+戥
+戦
+戩
+戰
+戱
+戲
+戸
+戻
+戽
+扆
+扥
+抃
+抇
+抦
+拋
+拏
+拝
+拡
+拺
+挙
+挵
+挹
+挻
+挾
+捒
+捜
+捦
+捨
+捩
+捫
+捭
+捱
+掃
+掄
+掙
+掛
+掞
+掟
+採
+掾
+揀
+揄
+揆
+揔
+揮
+揺
+搖
+搗
+搠
+搢
+搳
+搵
+搶
+搾
+摂
+摜
+摟
+摠
+摭
+摯
+摳
+摴
+摵
+摶
+摺
+摻
+摽
+撈
+撐
+撓
+撖
+撙
+撚
+撣
+撥
+撫
+撲
+撳
+撾
+撿
+擁
+擇
+擊
+擋
+擔
+擠
+擥
+擬
+擯
+擰
+擱
+擲
+擴
+擷
+擺
+擼
+擾
+攏
+攔
+攖
+攜
+攞
+攢
+攣
+攤
+攪
+攬
+攴
+攷
+攽
+敍
+敎
+敔
+敗
+敘
+敫
+敭
+敵
+敻
+敾
+斂
+斃
+斎
+斕
+斖
+斝
+斬
+斷
+斿
+旂
+旃
+旄
+旉
+旙
+旛
+旡
+旲
+旳
+旻
+旼
+旽
+旾
+旿
+昃
+昉
+昍
+昐
+昚
+昛
+昜
+昞
+昡
+昣
+昤
+昪
+昫
+昰
+昺
+晈
+晉
+晊
+晙
+晛
+晝
+晩
+晪
+晫
+晭
+晸
+暅
+暈
+暉
+暊
+暌
+暎
+暏
+暐
+暕
+暘
+暝
+暟
+暠
+暦
+暫
+暱
+暲
+暸
+暻
+暾
+曄
+曅
+曆
+曇
+曌
+曔
+曖
+曠
+曧
+曨
+曩
+曮
+曶
+曷
+曺
+曽
+朊
+朏
+朓
+朖
+朧
+朶
+杁
+杌
+杓
+杙
+杣
+杤
+杧
+杬
+杴
+杻
+杼
+枏
+枖
+枛
+枠
+枡
+枲
+枹
+柁
+柃
+柉
+柊
+柎
+柝
+柟
+柰
+柵
+柶
+柷
+査
+柾
+栃
+栄
+栐
+栒
+栜
+栝
+栞
+栨
+栲
+栴
+栻
+桄
+桕
+桙
+桜
+桝
+桫
+桱
+桲
+桴
+桿
+梀
+梂
+梃
+梉
+梔
+梘
+梟
+梠
+梣
+梫
+梱
+梶
+梽
+棄
+棆
+棐
+棓
+棖
+棗
+棡
+棧
+棨
+棩
+棪
+棫
+棲
+棶
+棹
+棻
+棼
+椆
+椇
+椏
+椙
+椥
+椪
+椲
+椵
+楙
+楡
+楢
+楤
+楧
+楨
+楫
+楮
+楯
+楳
+榊
+榍
+榎
+榑
+榖
+榗
+榘
+榢
+榣
+榤
+榦
+榲
+榿
+槀
+槁
+槃
+槊
+槓
+槔
+槙
+槤
+槩
+槭
+槰
+槱
+槳
+槺
+槻
+槼
+樀
+樁
+樅
+樆
+樋
+樑
+樗
+樘
+樞
+権
+樫
+樺
+樻
+橈
+橐
+橒
+橓
+橚
+橢
+橫
+橿
+檄
+檇
+檉
+檊
+檎
+檜
+檞
+檠
+檡
+檢
+檣
+檦
+檨
+檯
+檳
+檵
+檻
+檽
+櫂
+櫃
+櫆
+櫈
+櫓
+櫚
+櫛
+櫞
+櫟
+櫨
+櫪
+櫱
+櫸
+櫻
+櫾
+櫿
+欄
+欉
+欏
+欒
+欖
+欞
+欥
+欸
+欹
+欽
+歊
+歎
+歛
+歩
+歲
+歳
+歴
+歿
+殂
+殄
+殑
+殘
+殛
+殞
+殟
+殤
+殭
+殮
+殯
+殲
+殳
+毀
+毆
+毉
+毌
+毎
+毐
+毖
+毘
+毬
+毴
+毸
+毿
+氂
+氈
+氍
+氫
+氬
+氷
+氹
+氻
+氾
+汎
+汜
+汧
+汭
+沄
+沆
+沇
+沍
+沒
+沖
+沘
+沚
+沜
+沢
+沨
+沯
+沺
+況
+泂
+泆
+泇
+泐
+泖
+泚
+洌
+洎
+洢
+洣
+洤
+洨
+洩
+洸
+洹
+浄
+浛
+浞
+浟
+浡
+浤
+浯
+浵
+浹
+涙
+涼
+淍
+淎
+淏
+淓
+淛
+淠
+淥
+淪
+淯
+淰
+淵
+淶
+淸
+淺
+淽
+渃
+済
+渉
+渋
+渕
+渙
+渟
+渦
+渫
+渼
+渽
+渾
+湉
+湊
+湔
+湜
+湞
+湣
+湥
+湧
+湳
+湴
+湼
+満
+溁
+溈
+溋
+溎
+準
+溙
+溦
+溲
+溵
+溼
+滀
+滄
+滅
+滈
+滉
+滌
+滎
+滝
+滯
+滲
+滷
+滸
+滹
+滻
+滽
+滾
+漇
+漈
+漎
+漚
+漣
+漬
+漲
+漴
+漵
+漷
+漸
+漼
+漿
+潁
+潑
+潛
+潟
+潯
+潰
+潲
+潽
+潾
+潿
+澀
+澁
+澂
+澆
+澇
+澉
+澋
+澌
+澔
+澗
+澠
+澣
+澥
+澪
+澮
+澯
+澱
+澻
+濁
+濊
+濋
+濘
+濙
+濫
+濬
+濰
+濲
+濶
+濺
+濼
+濾
+瀁
+瀅
+瀆
+瀉
+瀍
+瀏
+瀔
+瀕
+瀝
+瀞
+瀟
+瀠
+瀦
+瀧
+瀨
+瀬
+瀰
+瀲
+瀴
+瀶
+瀾
+灃
+灊
+灑
+灘
+灝
+灤
+灧
+灴
+災
+炁
+炆
+炘
+炟
+炤
+炱
+炲
+炷
+炻
+烉
+烋
+烒
+烔
+烝
+烱
+烴
+焃
+焄
+焌
+焓
+焜
+焞
+焴
+焻
+焿
+煇
+煉
+煐
+煒
+煔
+煕
+煚
+煠
+煩
+煬
+煳
+煵
+煶
+熅
+熇
+熈
+熒
+熖
+熗
+熜
+熤
+熯
+熲
+熺
+熼
+熾
+熿
+燄
+燉
+燊
+燏
+燐
+燔
+燝
+燫
+燬
+燭
+燹
+燻
+燼
+燾
+燿
+爀
+爌
+爔
+爚
+爛
+爝
+爿
+牁
+牂
+牆
+牕
+牖
+牘
+牝
+牠
+牻
+牼
+牽
+犂
+犎
+犖
+犛
+犢
+犧
+犨
+犰
+犴
+犽
+狎
+狓
+狛
+狟
+狦
+狨
+狳
+狶
+狷
+狹
+狻
+猁
+猄
+猇
+猊
+猙
+猞
+猢
+猨
+猳
+猶
+猺
+猻
+獁
+獃
+獄
+獇
+獎
+獏
+獢
+獣
+獬
+獮
+獯
+獰
+獵
+獷
+獺
+獼
+獾
+玀
+玆
+玎
+玏
+玓
+玕
+玗
+玘
+玙
+玠
+玡
+玢
+玧
+玨
+玭
+玶
+玹
+玾
+珅
+珌
+珎
+珖
+珝
+珡
+珤
+珦
+珧
+珪
+珮
+珵
+珹
+珽
+琁
+琄
+琇
+琍
+琎
+琡
+琤
+琱
+琹
+琺
+琿
+瑀
+瑂
+瑆
+瑈
+瑊
+瑋
+瑑
+瑒
+瑝
+瑠
+瑢
+瑣
+瑤
+瑥
+瑧
+瑨
+瑯
+瑱
+瑳
+瑴
+瑺
+璄
+璆
+璉
+璌
+璕
+璘
+璙
+璚
+璠
+璡
+璣
+璥
+璦
+璪
+璫
+璬
+璮
+璱
+璵
+璸
+璹
+璽
+璿
+瓈
+瓊
+瓌
+瓏
+瓑
+瓔
+瓖
+瓘
+瓚
+瓛
+瓞
+甂
+甌
+甍
+甑
+甕
+甡
+甦
+甪
+畀
+畇
+畊
+畋
+畎
+畑
+畝
+畠
+畢
+畧
+畬
+畯
+異
+畳
+畷
+疇
+疊
+疋
+疍
+疒
+疕
+痍
+痙
+痟
+痩
+痲
+痺
+瘍
+瘓
+瘜
+瘞
+瘡
+瘧
+瘰
+瘺
+癀
+癆
+癇
+癒
+癘
+癟
+癡
+癢
+癤
+癥
+癩
+癬
+癭
+癮
+癯
+癰
+癱
+癲
+発
+皐
+皚
+皛
+皝
+皞
+皰
+皷
+皸
+盃
+盋
+盌
+盞
+盡
+監
+盦
+盧
+盨
+盩
+盪
+盫
+盷
+盺
+眀
+県
+眛
+眜
+眥
+眵
+眾
+睜
+睞
+睥
+睪
+睭
+睺
+瞋
+瞞
+瞢
+瞫
+瞼
+瞽
+矇
+矍
+矚
+矧
+矯
+砢
+砩
+砫
+砮
+砯
+砲
+砳
+砵
+硃
+硇
+硏
+硐
+硓
+硜
+硤
+硨
+硭
+硯
+碕
+碡
+碪
+碭
+碸
+碻
+碽
+磔
+磘
+磙
+磜
+磡
+磪
+磯
+磱
+磲
+磵
+磻
+磾
+礄
+礎
+礐
+礑
+礒
+礙
+礠
+礦
+礪
+礫
+礬
+礮
+礱
+礽
+祂
+祆
+祇
+祋
+祏
+祓
+祕
+祧
+祹
+祿
+禃
+禇
+禍
+禎
+禑
+禓
+禔
+禕
+禘
+禛
+禟
+禠
+禤
+禨
+禩
+禰
+禱
+禵
+禼
+禿
+秈
+秠
+秳
+稅
+稈
+稉
+稑
+稘
+稙
+稜
+稟
+稱
+稲
+稺
+稾
+穀
+穈
+穉
+穌
+積
+穎
+穟
+穠
+穡
+穢
+穣
+穫
+窅
+窋
+窣
+窩
+窪
+窮
+窯
+窰
+窶
+窺
+竄
+竅
+竇
+竈
+竊
+竑
+竜
+竦
+竩
+竻
+笄
+笘
+笞
+笥
+笩
+笪
+笭
+笮
+笯
+笱
+笳
+笹
+筅
+筊
+筌
+筍
+筘
+筥
+筦
+筧
+筬
+筭
+筲
+筳
+筶
+筻
+箆
+箇
+箋
+箏
+箑
+箒
+箜
+範
+篊
+篋
+篌
+篔
+篠
+篤
+篥
+篩
+篭
+篯
+篳
+簀
+簃
+簉
+簍
+簑
+簕
+簗
+簞
+簠
+簫
+簷
+簹
+簺
+簽
+簾
+籀
+籌
+籐
+籙
+籛
+籜
+籝
+籟
+籠
+籥
+籪
+籬
+籮
+籲
+籾
+粄
+粍
+粦
+粩
+糀
+糌
+糎
+糞
+糢
+糧
+糬
+糰
+糴
+糶
+糸
+糹
+糺
+糾
+紂
+紆
+紇
+紈
+紉
+紐
+紑
+紓
+紕
+紘
+紙
+紛
+紜
+紝
+紞
+紮
+紱
+紲
+紳
+紵
+紺
+紿
+絃
+絆
+経
+絎
+絕
+絛
+絜
+絞
+絢
+絨
+絪
+絳
+絵
+絹
+絺
+綃
+綈
+綎
+綏
+綖
+継
+続
+綜
+綝
+綞
+綠
+綢
+綣
+綧
+綬
+綮
+綰
+綳
+綴
+綸
+綺
+綻
+綽
+綾
+綿
+緁
+緃
+緄
+緈
+緊
+緋
+総
+緒
+緘
+緜
+緝
+緞
+締
+緡
+緤
+編
+緩
+緬
+緱
+緲
+練
+縂
+縄
+縈
+縉
+縊
+縕
+縛
+縝
+縞
+縠
+縡
+縯
+縱
+縴
+縵
+縷
+縹
+縻
+績
+繃
+繆
+繇
+繒
+繕
+繖
+繙
+繚
+繞
+繩
+繫
+繭
+繰
+繳
+繻
+繼
+繽
+繾
+纁
+纈
+纍
+纏
+纓
+纔
+纕
+纖
+纘
+纜
+缶
+缽
+罃
+罅
+罈
+罉
+罌
+罍
+罟
+罨
+罰
+罳
+罵
+罶
+罷
+罽
+羂
+羆
+羈
+羋
+羕
+羗
+羣
+羥
+羨
+羱
+翀
+翂
+翃
+翕
+翙
+翜
+翬
+翮
+翹
+耎
+耔
+耨
+耬
+聃
+聒
+聟
+聰
+聱
+聳
+聴
+聶
+聽
+聾
+肅
+肏
+肜
+肫
+肸
+肹
+胂
+胅
+胇
+胊
+胙
+胝
+胼
+脅
+脇
+脈
+脛
+脣
+脩
+脫
+脬
+脭
+脳
+脷
+脹
+腧
+腫
+腳
+膂
+膣
+膥
+膩
+膮
+膽
+膾
+膿
+臉
+臍
+臏
+臚
+臞
+臟
+臠
+臯
+舂
+舉
+舎
+舘
+舢
+舥
+舨
+舩
+舲
+舺
+艅
+艉
+艋
+艎
+艏
+艔
+艙
+艚
+艱
+艸
+艽
+芑
+芛
+芨
+芴
+芻
+苅
+苤
+苧
+苳
+苺
+苻
+苾
+茀
+茇
+茈
+茘
+茚
+茛
+茝
+茮
+茲
+茷
+茺
+荅
+荇
+荊
+荎
+荖
+荳
+莕
+莖
+莙
+莛
+莢
+莧
+莩
+莿
+菈
+菉
+菍
+菑
+菔
+菝
+菥
+菫
+菰
+菴
+菶
+菸
+菹
+菺
+菼
+菾
+萇
+萐
+萠
+萡
+萣
+萩
+萵
+萹
+葃
+葊
+葎
+葙
+葜
+葝
+葦
+葯
+葰
+葶
+葷
+蒍
+蒎
+蒐
+蒓
+蒔
+蒗
+蒞
+蒢
+蒧
+蒨
+蒭
+蒯
+蒴
+蒹
+蒺
+蒼
+蒾
+蓀
+蓁
+蓂
+蓆
+蓍
+蓘
+蓚
+蓧
+蓨
+蓪
+蓭
+蓯
+蓳
+蓽
+蔆
+蔎
+蔔
+蔕
+蔘
+蔝
+蔞
+蔣
+蔥
+蔦
+蔭
+蔴
+蔵
+蕁
+蕅
+蕎
+蕑
+蕖
+蕘
+蕚
+蕡
+蕢
+蕩
+蕪
+蕭
+蕷
+蕺
+蕻
+薀
+薆
+薊
+薌
+薐
+薑
+薔
+薗
+薘
+薙
+薜
+薞
+薟
+薨
+薫
+薲
+薷
+薸
+薺
+薾
+薿
+藎
+藟
+藦
+藨
+藪
+藶
+藸
+藹
+藺
+蘂
+蘄
+蘅
+蘊
+蘋
+蘐
+蘓
+蘗
+蘘
+蘚
+蘞
+蘢
+蘧
+蘩
+蘵
+蘶
+蘿
+虉
+虓
+虖
+虛
+虜
+虧
+虨
+虯
+虵
+虺
+蚆
+蚋
+蚍
+蚖
+蚡
+蚢
+蚵
+蚺
+蚼
+蛄
+蛉
+蛍
+蛑
+蛞
+蛯
+蛸
+蛺
+蛻
+蜆
+蜉
+蜑
+蜞
+蜢
+蜣
+蜨
+蜮
+蜯
+蜾
+蝀
+蝍
+蝓
+蝕
+蝘
+蝚
+蝟
+蝣
+蝤
+蝦
+蝨
+蝮
+蝯
+蝰
+蝲
+蝸
+螄
+螅
+螋
+螐
+螔
+螞
+螠
+螢
+螣
+螥
+螫
+螭
+螶
+螻
+螽
+螾
+蟄
+蟅
+蟊
+蟌
+蟎
+蟜
+蟥
+蟪
+蟫
+蟬
+蟯
+蟲
+蟳
+蟴
+蟶
+蟻
+蠂
+蠃
+蠅
+蠆
+蠊
+蠋
+蠍
+蠐
+蠑
+蠘
+蠙
+蠟
+蠣
+蠱
+蠲
+蠵
+蠶
+蠷
+蠻
+衂
+衎
+衕
+衚
+衜
+衝
+衞
+衽
+袓
+袛
+袞
+袴
+袾
+裊
+裎
+裒
+裖
+裬
+裵
+裾
+裿
+褌
+褍
+褎
+褘
+褙
+褞
+褧
+褫
+褭
+褸
+褻
+襌
+襖
+襞
+襠
+襤
+襦
+襪
+襯
+襲
+襴
+襶
+襻
+襾
+覇
+覈
+規
+覓
+覚
+覡
+覦
+覧
+覬
+覲
+観
+覽
+覿
+觔
+觙
+觚
+觜
+觭
+觱
+觴
+觶
+觿
+訁
+訃
+訇
+訌
+討
+訏
+訐
+訒
+訔
+訕
+訖
+託
+訛
+訝
+訟
+訥
+訴
+訶
+註
+証
+詁
+詆
+詈
+詐
+詒
+詔
+評
+詛
+詞
+詠
+詡
+詣
+詥
+詧
+詫
+詭
+詮
+詰
+詳
+詵
+詼
+誄
+誅
+誇
+認
+誒
+誕
+誡
+誣
+誤
+誥
+誦
+誨
+說
+読
+誰
+誴
+誹
+誼
+誾
+談
+諍
+諏
+諒
+論
+諗
+諜
+諟
+諠
+諡
+諤
+諦
+諧
+諪
+諫
+諭
+諮
+諱
+諲
+諳
+諴
+諶
+諷
+諸
+諺
+諼
+謀
+謁
+謂
+謄
+謊
+謌
+謎
+謏
+謐
+謔
+謖
+謗
+謙
+謚
+謜
+謠
+謤
+謨
+謩
+謫
+謬
+謳
+謾
+譏
+譓
+譔
+譙
+譚
+譞
+譫
+譭
+譯
+議
+譲
+譳
+譴
+譽
+譿
+讃
+讌
+讎
+讓
+讖
+讙
+讚
+讜
+讞
+谿
+豈
+豎
+豔
+豢
+豨
+豬
+豳
+豸
+豿
+貐
+貒
+貓
+貘
+貞
+貤
+貧
+貪
+貫
+責
+貮
+貯
+貲
+貳
+貶
+貸
+貺
+貽
+賁
+賂
+賃
+賄
+賈
+賊
+賑
+賒
+賔
+賕
+賚
+賞
+賡
+賤
+賦
+賨
+賬
+賭
+賹
+賺
+賻
+賽
+賾
+贄
+贅
+贇
+贊
+贌
+贍
+贓
+贔
+贖
+贛
+赧
+赬
+趐
+趕
+趖
+趨
+趺
+趼
+跅
+跏
+跗
+跡
+跣
+跩
+踎
+踐
+踰
+踴
+蹕
+蹟
+蹠
+蹤
+蹯
+蹺
+蹻
+躂
+躄
+躉
+躋
+躍
+躑
+躒
+躔
+躝
+躪
+躰
+軀
+軋
+軔
+軛
+軟
+転
+軫
+軲
+軸
+軹
+軺
+軻
+軼
+軾
+較
+輄
+輅
+輋
+輒
+輓
+輔
+輛
+輞
+輟
+輥
+輦
+輩
+輬
+輭
+輶
+輻
+輾
+輿
+轀
+轂
+轄
+轅
+轆
+轍
+轎
+轘
+轝
+轟
+轤
+辭
+辮
+辯
+辵
+辺
+辻
+込
+迴
+迵
+迺
+逈
+逋
+逌
+逎
+逕
+逖
+逤
+逨
+逴
+遄
+遊
+違
+遘
+遙
+遜
+遞
+遯
+遲
+遶
+遷
+遹
+遺
+遼
+邁
+邇
+邉
+邊
+邙
+邠
+邲
+邽
+邾
+郃
+郄
+郇
+郋
+郞
+郟
+郤
+郪
+郳
+郷
+郿
+鄃
+鄆
+鄋
+鄑
+鄒
+鄔
+鄖
+鄗
+鄘
+鄚
+鄜
+鄠
+鄤
+鄧
+鄩
+鄫
+鄰
+鄲
+鄳
+鄴
+酃
+酆
+酈
+酎
+酏
+酔
+酢
+酩
+酴
+酺
+酼
+醁
+醂
+醃
+醅
+醞
+醢
+醣
+醮
+醯
+醾
+醿
+釁
+釆
+釋
+釐
+釒
+釓
+釔
+釕
+釗
+釘
+釙
+釚
+釤
+釦
+釧
+釩
+釪
+釭
+釴
+釵
+釷
+釹
+釺
+鈀
+鈁
+鈄
+鈇
+鈈
+鈉
+鈊
+鈍
+鈏
+鈐
+鈑
+鈔
+鈕
+鈖
+鈞
+鈢
+鈣
+鈥
+鈦
+鈫
+鈮
+鈰
+鈳
+鈷
+鈸
+鈹
+鈾
+鈿
+鉀
+鉆
+鉈
+鉉
+鉋
+鉍
+鉏
+鉑
+鉓
+鉗
+鉚
+鉛
+鉞
+鉟
+鉤
+鉦
+鉬
+鉭
+鉲
+鉶
+鉷
+鉸
+鉻
+鉾
+鉿
+銂
+銃
+銅
+銋
+銍
+銓
+銕
+銖
+銚
+銜
+銠
+銣
+銥
+銦
+銨
+銩
+銪
+銫
+銬
+銱
+銲
+銳
+銶
+銹
+銻
+銼
+銾
+鋁
+鋅
+鋆
+鋇
+鋌
+鋏
+鋐
+鋒
+鋕
+鋗
+鋙
+鋡
+鋤
+鋥
+鋦
+鋨
+鋮
+鋯
+鋰
+鋱
+鋳
+鋶
+鋸
+鋹
+錀
+錏
+錐
+錒
+錕
+錘
+錚
+錞
+錟
+錠
+錡
+錢
+錨
+錫
+錬
+錮
+錳
+錸
+錻
+鍀
+鍇
+鍈
+鍉
+鍍
+鍏
+鍔
+鍘
+鍛
+鍝
+鍟
+鍠
+鍥
+鍩
+鍬
+鍱
+鍳
+鍶
+鍷
+鍺
+鍼
+鍾
+鎂
+鎅
+鎊
+鎌
+鎓
+鎔
+鎗
+鎘
+鎚
+鎛
+鎣
+鎦
+鎧
+鎪
+鎬
+鎭
+鎰
+鎳
+鎵
+鏃
+鏇
+鏈
+鏊
+鏌
+鏐
+鏑
+鏓
+鏗
+鏘
+鏝
+鏞
+鏟
+鏤
+鏦
+鏳
+鏴
+鏵
+鏷
+鏻
+鏽
+鐃
+鐇
+鐈
+鐓
+鐔
+鐙
+鐠
+鐤
+鐦
+鐧
+鐫
+鐬
+鐭
+鐮
+鐲
+鐳
+鐸
+鐺
+鐽
+鐿
+鑀
+鑁
+鑂
+鑄
+鑅
+鑊
+鑌
+鑑
+鑛
+鑠
+鑣
+鑨
+鑪
+鑭
+鑰
+鑲
+鑴
+鑷
+鑼
+鑾
+鑿
+閂
+閆
+閉
+閎
+閒
+閔
+閘
+閜
+閞
+閦
+閨
+閬
+閭
+閰
+閱
+閶
+閹
+閻
+閼
+閾
+閿
+闆
+闇
+闈
+闊
+闋
+闌
+闍
+闐
+闓
+闔
+闕
+闖
+闘
+闞
+闡
+闢
+闥
+阭
+阯
+陁
+陔
+陘
+陜
+陝
+陞
+陬
+陸
+険
+隄
+隈
+隊
+階
+隕
+隣
+險
+隰
+隱
+隲
+隳
+隴
+隷
+隸
+隻
+雋
+雑
+雖
+雛
+雝
+雩
+雫
+雱
+霅
+霈
+霊
+霑
+霙
+霤
+霧
+霨
+霶
+霽
+靁
+靂
+靄
+靉
+靚
+靫
+靬
+靭
+靺
+靼
+鞆
+鞏
+鞞
+鞥
+鞦
+鞨
+鞮
+鞴
+韁
+韃
+韆
+韋
+韌
+韑
+韙
+韜
+韞
+韠
+韡
+韮
+韺
+韾
+頁
+頃
+項
+須
+頊
+頌
+頍
+頎
+頏
+預
+頑
+頒
+頗
+領
+頜
+頠
+頦
+頫
+頴
+頵
+頷
+頸
+頹
+頻
+頼
+顆
+額
+顎
+顒
+顓
+顕
+顗
+願
+顙
+顛
+顥
+顫
+顰
+顱
+顳
+顴
+颮
+颯
+颱
+颶
+颺
+颼
+飆
+飈
+飠
+飡
+飢
+飥
+飩
+飪
+飫
+飭
+飴
+飽
+餃
+餄
+餉
+餌
+餎
+餒
+餓
+餗
+餚
+餛
+餞
+餡
+餵
+餺
+餾
+餿
+饋
+饌
+饑
+饒
+饗
+饞
+饟
+饢
+馘
+馛
+馦
+馭
+馯
+馱
+馳
+馼
+駁
+駄
+駅
+駆
+駐
+駑
+駒
+駔
+駘
+駙
+駛
+駝
+駟
+駢
+駭
+駰
+駱
+騁
+騂
+騄
+騅
+騋
+騎
+騏
+験
+騖
+騙
+騤
+騨
+騫
+騭
+騮
+騶
+騾
+驁
+驃
+驄
+驅
+驌
+驍
+驎
+驒
+驕
+驚
+驛
+驟
+驢
+驤
+驥
+驩
+驪
+骯
+髀
+髎
+髏
+髑
+髒
+髡
+髭
+髲
+髷
+髹
+鬄
+鬅
+鬆
+鬍
+鬚
+鬢
+鬥
+鬧
+鬨
+鬩
+鬪
+鬬
+鬮
+鬯
+鬱
+鬹
+鬻
+魃
+魈
+魋
+魍
+魎
+魕
+魘
+魛
+魞
+魟
+魣
+魨
+魩
+魮
+魯
+魴
+魷
+鮀
+鮁
+鮃
+鮄
+鮊
+鮋
+鮍
+鮐
+鮑
+鮒
+鮓
+鮗
+鮜
+鮟
+鮠
+鮡
+鮣
+鮨
+鮪
+鮫
+鮭
+鮰
+鮸
+鮹
+鮻
+鯀
+鯁
+鯃
+鯇
+鯊
+鯏
+鯒
+鯓
+鯔
+鯕
+鯖
+鯗
+鯙
+鯛
+鯡
+鯢
+鯤
+鯧
+鯪
+鯭
+鯮
+鯰
+鯶
+鯷
+鯻
+鯿
+鰂
+鰃
+鰆
+鰈
+鰉
+鰍
+鰏
+鰒
+鰓
+鰕
+鰗
+鰛
+鰜
+鰟
+鰣
+鰤
+鰧
+鰨
+鰩
+鰭
+鰮
+鰱
+鰲
+鰳
+鰶
+鰷
+鰹
+鰺
+鰻
+鰼
+鰾
+鱀
+鱂
+鱅
+鱇
+鱈
+鱉
+鱊
+鱒
+鱓
+鱔
+鱖
+鱗
+鱘
+鱚
+鱝
+鱟
+鱠
+鱣
+鱥
+鱧
+鱨
+鱮
+鱰
+鱲
+鱵
+鱷
+鱺
+鳧
+鳩
+鳰
+鳴
+鳶
+鳽
+鴆
+鴇
+鴉
+鴒
+鴓
+鴕
+鴗
+鴛
+鴝
+鴞
+鴟
+鴡
+鴣
+鴦
+鴫
+鴯
+鴰
+鴴
+鵂
+鵄
+鵎
+鵐
+鵑
+鵒
+鵓
+鵙
+鵜
+鵝
+鵞
+鵟
+鵠
+鵡
+鵪
+鵯
+鵰
+鵲
+鵵
+鵼
+鵾
+鶆
+鶇
+鶉
+鶒
+鶓
+鶘
+鶚
+鶡
+鶥
+鶩
+鶬
+鶯
+鶲
+鶹
+鶺
+鶻
+鶼
+鶿
+鷂
+鷉
+鷎
+鷓
+鷙
+鷚
+鷟
+鷥
+鷦
+鷫
+鷯
+鷲
+鷳
+鷸
+鸊
+鸌
+鸐
+鸑
+鸕
+鸘
+鸚
+鸛
+鸜
+鸝
+鸞
+鹮
+鹵
+鹹
+鹼
+麅
+麇
+麈
+麊
+麐
+麞
+麩
+麪
+麴
+麹
+麼
+麿
+黁
+黇
+黌
+黐
+黙
+黥
+黧
+黨
+黴
+黶
+黻
+黼
+黽
+黿
+鼂
+鼇
+鼈
+鼉
+鼐
+鼒
+鼕
+鼢
+鼩
+鼯
+鼱
+鼴
+鼷
+齒
+齕
+齡
+齣
+齦
+齧
+齲
+齶
+龎
+龐
+龑
+龔
+龕
+龜
+龝
+龠 
+ず
+梌
+叀
+晢
+媸
+錾
+鐖
+䰡
+櫬
+锱
+υ
+鼗
+媪
+澴
+苈
+眴
+𝜏
+缱
+𝜶
+조
+晡
+≡
+ࠀ
+н
+廇
+嗛
+篚
+ώ
+莰
+윤
+纚
+𢢞
+闼
+熌
+饎
+蓊
+倅
+년
+聭
+耩
+≅
+
+≺
+诌
+
+
+
+耰
+菗
+僦
+⇣
+甊
+冓
+缷
+枊
+沕
+𝐴
+❹
+형
+秾
+
+щ
+厹
+
+˗
+疔
+䩦
+髴
+⨂
+莏
+≧
+垆
+銌
+桤
+隤
+ギ
+벽
+⑸
+✘
+̣
+辶
+铼
+게
+へ
+獶
+藳
+祍
+黉
+跱
+⽬
+埙
+だ
+蓣
+亯
+구
+
+鹎
+
+⾃
+楩
+⌘
+汏
+虒
+谖
+
+﹜
+劖
+じ
+瑇
+㮑
+揕
+⇔
+𤔲
+薉
+𝑾
+硗
+〈
+は
+盍
+狽
+ж
+я
+挆
+槨
+γ
+阏
+襕
+𝜉
+❖
+└
+총
+시
+
+ν
+刲
+ด
+嬲
+绤
+𝐰
+飦
+扱
+帻
+辀
+廴
+к
+蔖
+–
+같
+熭
+巣
+
+裛
+𝑶
+蓺
+蔊
+그
+匳
+玚
+Ц
+璲
+련
+𨒅
+변
+㤵
+饫
+𨚵
+Ｘ
+筇
+镡
+ⅳ
+𝛿
+轸
+𝑭
+鋈
+鵩
+縁
+˙
+ɿ
+𝒴
+㝮
+𝜂
+栠
+橦
+緇
+肰
+
+跼
+䭜
+蜅
+訸
+㻶
+𝑉
+เ
+
+嚢
+鼔
+𝒆
+閫
+阃
+𥞹
+杪
+誊
+
+鲋
+骍
+τ
+莾
+凊
+﹡
+箚
+蛱
+樯
+喾
+幞
+欕
+搡
+戉
+瘖
+᙭
+砟
+ས
+∤
+ี
+メ
+𝝁
+穑
+渶
+𦬁
+서
+⊗
+穇
+⌊
+を
+鐻
+蘤
+≫
+◐
+汙
+蒒
+⑷
+蹨
+ｘ
+裥
+嶤
+ァ
+従
+침
+稂
+𪧶
+で
+𝑹
+⑫
+闩
+槫
+舮
+𝑿
+戁
+간
+戯
+
+ོ
+æ
+わ
+チ
+砉
+Ψ
+劂
+･
+В
+鬭
+钔
+盭
+黓
+⎯
+𝐏
+함
+钪
+𝑸
+澰
+래
+藒
+龃
+瞀
+伧
+♂
+¹
+ƞ
+澼
+餍
+倶
+ð
+
+嚱
+跬
+貙
+磿
+娬
+氿
+鹘
+𝐁
+摅
+ヱ
+傰
+พ
+湝
+ˆ
+Л
+翾
+≃
+에
+滫
+С
+嫕
+あ
+㈣
+ⅇ
+垧
+⺮
+∠
+躐
+硌
+眢
+乧
+𝑐
+泃
+轫
+↔
+㎝
+≜
+⽇
+撟
+⟹
+脿
+
+㸁
+靯
+う
+⁠
+懬
+搷
+瀓
+ˁ
+ⅲ
+훈
+
+お
+𝛄
+瓅
+葻
+猋
+ら
+⾳
+喣
+⽿
+č
+鈎
+⑤
+å
+阸
+름
+て
+圮
+⚫
+⻄
+胨
+琠
+戄
+箄
+𝒳
+鼍
+й
+⼲
+廪
+睃
+囫
+͞
+죄
+
+호
+み
+饩
+
+⊆
+х
+欚
+瘚
+≯
+瞗
+ž
+嗵
+근
+ま
+
+⾔
+罥
+ʹ
+鼃
+д
+✳
+ゃ
+悊
+𝐅
+영
+＠
+ɣ
+𝛷
+𝜁
+ǜ
+犄
+⽂
+ཆ
+胒
+﹦
+谫
+є
+・
+𝐻
+狺
+백
+舳
+𝑁
+ษ
+
+𝜓
+𝒦
+盕
+유
+𪯐
+茑
+礤
+거
+コ
+肂
+鸻
+ã
+⑬
+铚
+걸
+磳
+綷
+𝒚
+舭
+腚
+㈩
+榱
+𝐌
+畾
+馐
+罾
+∕
+𝔛
+𝑬
+ç
+楬
+櫽
+顼
+阋
+
+꺼
+諛
+̌
+้
+㮀
+乵
+沬
+⼀
+ư
+鲠
+䜩
+樉
+鹈
+搧
+轾
+䟒
+등
+𝝉
+잠
+짤
+า
+蘨
+愪
+ྟ
+慪
+鮝
+𝛑
+び
+𥞪
+𝐾
+レ
+교
+ྲ
+달
+𝐩
+
+殹
+踇
+狥
+ベ
+미
+매
+⑭
+钁
+Θ
+못
+𝜇
+侂
+ę
+ฟ
+邶
+諣
+颃
+𡢕
+昑
+𝒖
+讱
+﹤
+緵
+骢
+朢
+骘
+ℜ
+
+ゞ
+愬
+鹬
+
+ッ
+ར
+급
+‚
+鸶
+蒫
+餽
+蓃
+ข
+辠
+ğ
+氺
+暆
+笿
+迚
+甝
+ή
+徼
+旣
+ϖ
+ヲ
+倕
+匽
+蓱
+리
+剷
+ู
+逪
+
+나
+堋
+焠
+Δ
+炑
+爫
+蒖
+𝒓
+悫
+𝛱
+
+𝐮
+騧
+ⅴ
+饾
+贠
+𝚲
+崀
+磀
+柤
+肈
+⻮
+鶄
+狲
+跫
+지
+鳇
+痖
+跂
+秫
+ʒ
+합
+ไ
+迨
+𝜐
+
+屦
+𝐶
+;
+辎
+∵
+鴁
+撏
+ς
+⟶
+薮
+㟪
+犮
+ب
+ビ
+藡
+甏
+
+眡
+訿
+鉥
+媵
+
+柫
+𝒞
+ь
+萏
+ค
+트
+訮
+汚
+眚
+〞
+き
+ほ
+刖
+髄
+蘀
+や
+ة
+诹
+т
+ན
+𝒃
+掼
+䓁
+僥
+팰
+枵
+✔
+³
+ེ
+鼖
+屖
+鍮
+砇
+カ
+舐
+牴
+𝜎
+㡿
+攉
+⽤
+晅
+労
+蛕
+𝐽
+Ʃ
+く
+穽
+孥
+𝒏
+
+
+ɬ
+玦
+檮
+ョ
+∥
+중
+萯
+呲
+䰈
+새
+
+釶
+
+ɢ
+⊂
+臮
+
+梼
+デ
+骖
+ス
+蹩
+羼
+▽
+Π
+≪
+匛
+𝐼
+稊
+่
+茠
+䢉
+秝
+茐
+齎
+そ
+
+芕
+噚
+癉
+蹱
+蓜
+𝐬
+ϑ
+е
+瀋
+ϕ
+χ
+镟
+霂
+隒
+▱
+ヶ
+撄
+둔
+¢
+こ
+跲
+莻
+𝑠
+輮
+็
+堠
+푟
+赕
+◦
+ا
+런
+帒
+汘
+̱
+尥
+蘠
+𦟜
+옥
+腠
+夨
+⩾
+𝑝
+歯
+刱
+여
+け
+溘
+釰
+肍
+擗
+矱
+鍌
+芧
+술
+발
+鼫
+舾
+⼯
+𝝓
+ƒ
+怸
+པ
+𣐼
+疎
+铷
+Η
+⑺
+蒏
+림
+⃛
+゜
+褴
+𨒪
+れ
+揢
+さ
+櫫
+櫑
+䋎
+灋
+櫜
+诓
+❶
+𝐃
+Ｑ
+袳
+ℒ
+菂
+
+荙
+ℛ
+⁄
+堙
+贋
+̅
+鳏
+̂
+､
+茍
+泜
+𝑈
+즉
+噔
+
+
+迓
+Ⅸ
+❷
+이
+＿
+⾊
+Ö
+铥
+耹
+䶮
+
+무
+
+セ
+饳
+อ
+篾
+통
+‒
+ย
+덕
+말
+艨
+Ω
+𝐨
+螓
+澐
+巠
+⋅
+钶
+도
+鸱
+齍
+恑
+褛
+剟
+준
+勶
+𠟠
+ß
+箅
+𝑆
+悃
+蘥
+
+Ξ
+𝑘
+妣
+𝑖
+𝐑
+纡
+釿
+⺌
+ヴ
+𝕀
+涻
+箙
+塚
+
+⼠
+墈
+∷
+疴
+ク
+ㄕ
+𝒂
+蒪
+蓡
+
+鷇
+瘏
+𣹳
+橰
+嵚
+帀
+주
+ド
+盓
+爇
+φ
+觋
+𝜑
+钍
+화
+표
+Ɛ
+篰
+명
+週
+с
+蓛
+裢
+穜
+㱃
+玊
+鲕
+蒕
+箪
+⑯
+苽
+矦
+偰
+盝
+佊
+僨
+駉
+𝑳
+머
+ª
+絅
+
+锒
+
+苆
+ั
+𝛻
+碹
+咺
+竝
+и
+づ
+강
+辁
+́
+铽
+纩
+齑
+𝝎
+어
+ユ
+躡
+𝒄
+ซ
+畛
+鸰
+ླ
+
+骉
+❸
+揲
+廃
+湋
+𝑲
+
+
+旤
+
+蹷
+钌
+국
+豙
+鬳
+
+ɛ
+轳
+俜
+眄
+萮
+𝐡
+颵
+箓
+魑
+𝑅
+漍
+ℤ
+
+逡
+학
+浖
+ょ
+¬
+怴
+𝛤
+怿
+祌
+纥
+𝒑
+⃑
+棅
+笵
+낭
+栦
+⑰
+บ
+𝔽
+𝑇
+埝
+⽓
+孱
+埶
+匜
+鸼
+
+
+벌
+ル
+锸
+斫
+妟
+뽀
+昬
+댁
+ʂ
+暯
+夳
+ノ
+堞
+懘
+榼
+鞫
+오
+𝑡
+偑
+戗
+∴
+伥
+끝
+𬌗
+稯
+岜
+Ε
+犲
+𩓞
+연
+鹚
+
+ག
+诜
+嗍
+倥
+鳣
+庑
+屾
+雚
+
+椄
+颏
+酤
+𝒋
+欛
+
+း
+려
+缋
+¾
+ゴ
+籑
+笤
+鞛
+鏺
+蓒
+설
+緍
+⑩
+迀
+鼋
+ɮ
+위
+锪
+∨
+滆
+€
+躅
+鋓
+柀
+䐶
+啎
+𝛵
+骃
+ć
+갈
+卨
+い
+𝑺
+鸲
+壻
+偯
+𝑞
+譖
+곤
+溍
+
+噫
+순
+
+𝑽
+ы
+赑
+蓸
+鸮
+稃
+っ
+詗
+으
+⨀
+屮
+俦
+伛
+畱
+늬
+𝑂
+朼
+沰
+겨
+з
+骀
+鸩
+𝜈
+º
+苊
+诎
+皤
+
+하
+̀
+砑
+凷
+翄
+𝑛
+赪
+≮
+浗
+𝐍
+û
+オ
+ƹ
+𝜅
+묘
+曛
+鳊
+𝛩
+癹
+磒
+ば
+⑨
+礆
+
+乼
+
+∽
+褱
+藴
+縶
+觥
+に
+식
+凫
+
+佥
+槷
+阍
+䰍
+졸
+전
+葢
+㝸
+も
+⻔
+遽
+
+蹰
+𝛺
+裏
+།
+를
+
+ろ
+짭
+
+ぐ
+싶
+渰
+⊤
+浳
+൯
+∃
+옛
+蟞
+과
+芠
+飖
+⼆
+敶
+粝
+𥃩
+坿
+䩉
+𝑯
+｢
+矰
+
+사
+𝛶
+𝑎
+挐
+푎
+동
+ℝ
+Γ
+︃
+珒
+鹍
+κ
+鑓
+傁
+惓
+臿
+丣
+悒
+侔
+ñ
+訳
+櫭
+賛
+觏
+辂
+覅
+濓
+堿
+擪
+฀
+𝑵
+扨
+嫫
+珰
+
+寃
+𝒔
+曱
+髣
+인
+≌
+莵
+踳
+ⅱ
+Ø
+⌋
+¯
+挢
+̇
+﹪
+哕
+𦫳
+
+襛
+昳
+铙
+铫
+軱
+汔
+ネ
+
+躩
+옷
+ถ
+엄
+皊
+臑
+𧄝
+𝑃
+
+䢅
+𝐝
+𝒍
+ℱ
+𝐓
+蓾
+𝑻
+䋁
+裼
+개
+ത
+𝒊
+僪
+瞂
+𦞠
+요
+￣
+荍
+𝜔
+ф
+峣
+庋
+檏
+袢
+绬
+Σ
+향
+钫
+え
+枅
+≝
+荦
+들
+勍
+ö
+𝒕
+툰
+遬
+𝐵
+擧
+咢
+钘
+
+𝒢
+Ⅷ
+➢
+讧
+ω
+簟
+廐
+刳
+阘
+б
+⊘
+髟
+臓
+루
+⎧
+诳
+у
+诮
+蠪
+梹
+耤
+パ
+ن
+∆
+
+𝑫
+น
+べ
+坼
+
+𝑤
+褽
+憼
+심
+∇
+迖
+휆
+叚
+없
+⼿
+钖
+斠
+䪵
+胠
+𝜋
+殽
+剜
+⾝
+−
+慸
+𝛽
+椔
+⟩
+皦
+筚
+奰
+Å
+물
+𝒐
+嫱
+钆
+ï
+∪
+⇢
+ş
+
+㖞
+璗
+葸
+殢
+𝜺
+夲
+骒
+ち
+회
+선
+睒
+轡
+ξ
+
+鲧
+镞
+碜
+놈
+Å
+紴
+
+⇤
+ྷ
+⑪
+喟
+𦼰
+
+蔩
+埦
+𝜆
+耋
+˜
+한
+舣
+馓
+⑻
+
+ɐ
+椘
+し
+莐
+辔
+憰
+碛
+⁃
+飏
+颀
+跽
+⇥
+赀
+撺
+襜
+ɒ
+袧
+л
+정
+꾸
+콩
+
+박
+缑
+柈
+
+樲
+𝑮
+詘
+µ
+𝑷
+鹪
+𝛼
+차
+讬
+掯
+硎
+𝑨
+舄
+‹
+누
+バ
+ก
+萀
+兇
+숙
+貍
+
+踈
+친
+𝜽
+摰
+甿
+坜
+遑
+삼
+배
+Μ
+을
+譊
+沩
+빈
+飑
+钹
+镨
+鐉
+宭
+桗
+ə
+歺
+А
+⇒
+锞
+𝒪
+棊
+愓
+莶
+琲
+འ
+プ
+་
+𝐿
+艟
+欬
+ิ
+в
+ų
+纻
+㎎
+婄
+Ρ
+歟
+椢
+粜
+종
+خ
+ね
+剞
+베
+斄
+幠
+ト
+疛
+よ
+╳
+醳
+군
+諂
+芰
+穋
+禆
+길
+秊
+噙
+ｙ
+锓
+⁵
+
+拠
+Ĥ
+𝑒
+窬
+抈
+︰
+퐶
+铳
+각
+ش
+錉
+ù
+臝
+闶
+𝒟
+芐
+韎
+권
+萚
+
+ど
+羮
+镕
+欔
+瘐
+받
+𝑚
+𢦟
+髤
+腙
+⽽
+상
+铘
+장
+𤇈
+ニ
+凂
+ȷ
+육
+а
+살
+雠
+荑
+태
+穤
+ɯ
+
+圬
+楑
+단
+ง
+⾯
+λ
+⁰
+성
+萿
+缌
+
+毣
+矅
+
+푚
+˘
+貣
+∂
+은
+ė
+䌛
+경
+せ
+
+拫
+⅞
+餕
+鐨
+翭
+ื
+ɵ
+⺍
+Փ
+▬
+ว
+희
+짐
+屙
+洫
+ေ
+∏
+臜
+
+剸
+芓
+운
+∓
+계
+祔
+鶵
+𝝅
+柂
+訢
+禊
+扽
+恫
+𝜙
+狢
+勠
+ི
+𝜒
+จ
+铯
+ྱ
+𝑙
+蟇
+울
+莤
+牱
+𝒗
+詇
+靃
+殓
+栍
+踟
+ي
+
+鲄
+㓷
+贳
+ナ
+鲓
+𝒙
+薁
+Χ
+侪
+恌
+㰤
+목
+̄
+丱
+―
+𝛔
+𝑔
+
+鸷
+﹣
+籢
+脢
+δ
+窭
+‐
+阒
+석
+아
+ォ
+두
+𝐦
+浬
+搰
+褃
+
+ལ
+乇
+腘
+眊
+偬
+Ⅻ
+ℳ
+畤
+芟
+曈
+飧
+堌
+═
+谶
+櫝
+嬑
+冋
+嗌
+抜
+
+腜
+공
+𝜕
+ん
+鲭
+郐
+酓
+𝑍
+⾏
+⼹
+㐬
+고
+𝟑
+缯
+碤
+濩
+ʰ
+佻
+Υ
+∗
+賅
+집
+跹
+
+ɾ
+蔧
+다
+栫
+庰
+欤
+洿
+捾
+𝜍
+𝑄
+
+攆
+夂
+檿
+荜
+ц
+柖
+唅
+ท
+ɦ
+讦
+습
+锿
+
+纆
+檑
+殰
+歠
+鼑
+Ä
+و
+☑
+緦
+悁
+偞
+ส
+絭
+저
+踯
+騀
+쉰
+蒷
+揗
+儵
+ρ
+薅
+ャ
+‗
+犒
+旟
+
+승
+ང
+소
+𝛴
+瀜
+锜
+𣱼
+谳
+
+軑
+ポ
+楁
+𝑜
+袚
+ྐ
+Á
+𝑑
+鲀
+牾
+鬌
+푥
+¤
+呴
+‑
+✓
+민
+⼦
+ⅰ
+⽉
+擿
+ч
+➝
+가
+≳
+漥
+踖
+枧
+莝
+⻘
+傧
+𝑢
+ю
+赍
+ｑ
+絫
+о
+ア
+ℐ
+髫
+齢
+湎
+甓
+揿
+
+ℋ
+怹
+자
+⑦
+져
+椟
+鶟
+浕
+ー
+𝛂
+偾
+⃗
+喑
+鹡
+≦
+磽
+ⅆ
+
+葂
+鶱
+ン
+貇
+褡
+▴
+것
+喿
+つ
+闚
+
+盳
+𝟒
+雔
+洭
+殫
+楎
+£
+＾
+葲
+𝟖
+眗
+棸
+潏
+熕
+𝟐
+품
+သ
+樳
+⁴
+イ
+㈢
+렴
+脰
+돈
+⑮
+钲
+𝒘
+訽
+爟
+幨
+枻
+亷
+猃
+σ
+黩
+嘑
+榹
+⁡
+鍧
+𝑋
+枘
+𝑥
+원
+睚
+饔
+酲
+
+顸
+람
+𝐫
+脁
+½
+긴
+ʔ
+Ⅰ
+旆
+죽
+
+궐
+
+奡
+㭃
+杝
+忾
+ม
+掮
+饍
+摛
+쓰
+慊
+踣
+푅
+悽
+礅
+毄
+俓
+데
+冣
+만
+驖
+𤉣
+̃
+廾
+匵
+阇
+𤸫
+戣
+刌
+剕
+杅
+ο
+蒥
+ː
+癃
+蒬
+â
+À
+嗥
+우
+ケ
+｣
+聩
+ë
+吽
+檌
+苰
+⑹
+Ÿ
+
+⑥
+노
+
+˄
+鸫
+廛
+㱿
+鹛
+赟
+℅
+菿
+辳
+阼
+𝒇
+哋
+♀
+氕
+砤
+†
+舡
+偝
+飜
+넓
+鈜
+ầ
+닝
+禚
+匲
+〉
+Ф
+锊
+ϵ
+∙
+惛
+箧
+𝑦
+宬
+𝑀
+臙
+𩡶
+
+¡
+潀
+수
+敃
+か
+أ
+
+苌
+饘
+咝
+𝑼
+∘
+涷
+樍
+厣
+蝼
+墻
+Ñ
+秅
+︒
+∅
+↵
+葹
+ỽ
+𝑗
+た
+일
+蒊
+치
+竢
+¨
+佢
+潵
+櫼
+軵
+𧕿
+倨
+歱
+瘅
+𝐭
+黾
+脼
+ê
+땋
+鶷
+ё
+鹯
+掲
+＼
+𨳡
+
+Г
+ι
+탁
+溞
+殪
+菭
+𝛥
+擛
+録
+㥥
+∀
+锇
+锃
+편
+餬
+瘻
+ཟ
+豤
+로
+ɸ
+ℎ
+랑
+ʃ
+鼹
+臬
+ŋ
+巵
+譁
+ｗ
+窳
+蓔
+䉜
+浧
+酂
+⒀
+མ
+椠
+槖
+
+衄
+𨐨
+̿
+ご
+⺗
+顇
+𝒫
+搕
+ミ
+𪪋
+«
+䣛
+鹩
+鴈
+п
+는
+䋰
+𝛹
+犕
+呌
+𝒮
+𝑪
+鋎
+嚻
+杚
+䕊
+蠜
+ザ
+𝐂
+☐
+𥘔
+
+赜
+Ν
+廦
+瓾
+↦
+龉
+⽅
+棂
+𝜌
+큰
+踔
+ラ
+｡
+剤
+황
+⅜
+僈
+骈
+ɕ
+О
+м
+椑
+𝑟
+纇
+𝑓
+𝐖
+Ш
+⎦
+旹
+삶
+ึ
+囵
+
+す
+ⅈ
+ت
+踽
+陴
+餱
+ป
+막
+紟
+방
+剀
+簖
+闬
+キ
+鋉
+납
+タ
+谵
+詑
+족
+垔
+荋
+旰
+𥘸
+窾
+
+신
+𝐎
+𝛃
+
+
+﹒
+縰
+猲
+郘
+파
+⊕
+镘
+𠊃
+
+呔
+𝜗
+ʊ
+𝛬
+辏
+陭
+𝑕
+庴
+ʐ
+瀌
+倄
+蕞
+ل
+絷
+러
+든
+བ
+柅
+›
+傚
+睠
+Ⅺ
+饐
+蔮
+ɟ
+莈
+𤨨
+⋯
+犭
+𝜃
+𥹳
+초
+⎞
+遌
+眇
+蓗
+綅
+토
+裰
+
+⼼
+虘
+𝑌
+觯
+漶
+钤
+讒
+げ
+螬
+鲹
+咷
+蓞
+僂
+𝒉
+が
+桮
+포
+쟈
+柽
+ウ
+綟
+缟
+䁾
+钭
+烕
+厠
+
+孭
+礉
+­
+谲
+
+⼤
+𝒒
+旒
+㫄
+
+
+鳓
+挈
+재
+ད
+𧊒
+蝝
+𝐺
+懱
+芢
+
+ⅼ
+Ú
+𝑱
+翯
+芶
+厽
+遉
+鲒
+η
+𝛾
+趮
+虆
+汸
+嬖
+糈
+窸
+
+추
+棬
+懔
+硁
+ぶ
+抟
+胕
+𝑧
+⌦
+碫
+Ⅵ
+속
+𝐚
+
+Ç
+행
+Ɵ
+⑱
+贽
+箤
+р
+敒
+衤
+풍
+⊛
+慉
+ψ
+©
+광
+ℕ
+屣
+臌
+旵
+臁
+‡
+癎
+閡
+𡵂
+襐
+畟
+
+萪
+娒
+瘼
+庳
+천
+觌
+Α
+と
+奁
+煃
+؟
+◯
+의
+攎
+
+𝐞
+Ｊ
+𢦒
+❑
+벤
+𝐒
+リ
+蒉
+𝐱
+朹
+㈤
+„
+䗬
+Ι
+ཀ
+𡜵
+俣
+疬
+
+墥
+吣
+У
+榀
+絟
+
+旓
+𝐛
+𝜷
+瑮
+≔
+笾
+ζ
+김
+暵
+𝜹
+逶
+萙
+欇
+俧
+籴
+絰
+揶
+ǔ
+宂
+伩
+Ө
+菞
+梕
+エ
+蹚
+제
+Š
+沝
+
+𦳢
+𝒱
+揠
+ℏ
+𝐹
+箝
+규
+氒
+⼊
+鰌
+筮
+⼩
+대
+𝔾
+䄃
+𝐸
+﹕
+부
+
+刓
+ȵ
+缛
+기
+缊
+𝟎
+𨟻
+め
+捃
+⽚
+鍪
+灆
+迻
+⾦
+荗
+ｖ
+𡒊
+汍
+斲
+姕
+
+儴
+偒
+辤
+芀
+蝥
+ń
+臥
+椀
+㪚
+š
+담
+ø
+䈰
+睏
+テ
+﹐
+
+참
+楘
+𝒌
+劓
+ɪ
+醑
+绹
+諓
+𝛉
+ズ
+怼
+埘
+臽
+잡
+镢
+𝜖
+진
+踬
+谠
+﹥
+髺
+腞
+현
+嘭
+ʑ
+蓌
+〜
+锠
+蓶
+る
+
+∼
+枎
+緗
+薠
+芈
+耪
+𝒎
+謼
+
+瘳
+诨
+忤
+œ
+⇡
+
+鲣
+ⅵ
+Τ
+㯮
+
+㶲
+ⅹ
+䙴
+坴
+馑
+缹
+椦
+ô
+⼈
+フ
+誏
+э
+哙
+愎
+
+埽
+祲
+마
+殍
+菋
+懑
+
+辇
+鍤
+𝜀
+ɜ
+り
+𝐷
+㕞
+瑵
+
+蔨
+Ⅶ
+镴
+ภ
+𝝀
+𢶡
+⃝
+ơ
+柢
+𧴗
+ʁ
+攙
+
+な
+𝑏
+挴
+餧
+絇
+怄
+曏
+洟
+軷
+∉
+咍
+⎪
+樛
+𝑣
+웃
+椊
+黢
+𝑩
+誩
+伓
+戠
+橥
+⟨
+豰
+懥
+涖
+綘
+詬
+ွ
+˚
+刽
+ɑ
+격
+稖
+𝟏
+禝
+墦
+𝑊
+
+択
+檙
+∝
+颟
+诂
+𝐧
+踲
+𝜛
+𝑰
+
+鲬
+⁸
+ǎ
+문
+
+紬
+楲
+䊭
+枨
+膑
+õ
+던
+Ⅴ
+秏
+馔
+拊
+缗
+隠
+牀
+淲
+鬰
+綵
+鶑
+詎
+慙
+劒
+閲
+鎡
+淒
+屨
+鉢
+扃
+鳬
+閤
+馿
+翛
+駸
+蛩
+驂
+嵲
+覩
+牋
+湲
+蓴
+賸
+遡
+翫
+嫋
+惻
+妬
+罇
+龏
+鵷
+閙
+鎸
+朅
+巉
+僞
+洶
+磧
+筯
+慇
+鷁
+斾
+斸
+酹
+幘
+羶
+閽
+薤
+泝
+覯
+怱
+缾
+氳
+躊
+檝
+擣
+虀
+誚
+甃
+艤
+櫳
+醖
+壚
+涴
+崦
+秪
+潄
+濆
+駡
+坰
+闉
+縑
+躕
+颻
+燠
+輙
+鏁
+嶮
+薖
+輈
+綆
+覷
+蹔
+攄
+鐩
+鑱
+羃
+轓
+麤
+驀
+欵
+亙
+朮
+邐
+箠
+怳
+鋩
+鷃
+篘
+蔌
+諄
+旐
+慍
+欷
+頽
+蜺
+覊
+禋
+秔
+蜩
+嚬
+櫺
+軿
+痾
+笻
+猱
+毳
+泬
+竚
+齪
+搘
+欻
+釂
+嚥
+誑
+籩
+韉
+幙
+舠
+飣
+颭
+颸
+歔
+屧
+巇
+艫
+壖
+牓
+薝
+銛
+皪
+枿
+剗
+歘
+鸂
+邅
+衒
+荄
+鴂
+嫗
+顦
+瀼
+繄
+搆
+狖
+貰
+醆
+秖
+蹀
+頳
+纒
+憇
+溽
+澦
+讐
+灩
+箎
+螿
+鰥
+飀
+澒
+矻
+枌
+擡
+鷖
+齬
+纊
+挼
+齟
+錙
+屩
+蠧
+氅
+漭
+躚
+翺
+瘵
+螘
+鵶
+㶁
+斵
+饜
+岏
+䍦
+哢
+彴
+豗
+靨
+鋋
+禳
+覘
+鞚
+擻
+涘
+溷
+沴
+嶓
+褊
+罏
+齏
+醲
+繅
+舴
+釃
+厖
+闤
+閴
+藂
+譍
+糲
+籞
+躞
+餳
+遰
+倐
+嘖
+鷀
+暍
+韝
+蘺
+齁
+醽
+醨
+憀
+䕷
+跕
+拶
+垤
+鸎
+漙
+躭
+傴
+蕕
+嘒
+晻
+糵
+閈
+嫠
+斁
+鶗
+詶
+囘
+羇
+橛
+鞬
+磈
+粃
+阨
+塿
+敧
+氊
+芼
+襆
+迍
+鬛
+憒
+釅
+蓐
+奬
+頲
+髠
+抔
+葅
+槧
+跧
+揜
+渇
+餔
+罝
+裯
+蹁
+椶
+幰
+逰
+饁
+棃
+噀
+轔
+囁
+惸
+騑
+呪
+鬐
+綌
+醤
+䆉
+艣
+鐶
+夐
+摐
+鸇
+睎
+羝
+紼
+鞿
+噉
+磑
+闗
+筩
+駮
+蹌
+縢
+闠
+鬙
+谹
+榾
+觳
+皁
+晼
+啗
+簣
+騕
+蹣
+麰
+觧
+怊
+朞
+鱍
+蟣
+畚
+蠁
+舁
+瞇
+劚
+鰐
+籯
+鬖
+柮
+飱
+帟
+酇
+崿
+霪
+緌
+踆
+欃
+縟
+搦
+琖
+搥
+倀
+觫
+遝
+嚅
+聵
+藋
+筴
+喁
+窻
+穏
+牥
+鎩
+礲
+膴
+鞵
+醵
+斚
+縲
+裀
+齷
+騃
+袠
+谺
+靦
+帬
+鯈
+曀
+灔
+崷
+趂
+徯
+闃
+洧
+獪
+稏
+煢
+歈
+呶
+壈
+襃
+旴
+檟
+簦
+谽
+箵
+謡
+慝
+餖
+稌
+朣
+禖
+嚀
+嵂
+黷
+甖
+洑
+釡
+蕝
+甆
+翣
+篸
+隮
+滃
+裌
+蔀
+籖
+秬
+鷴
+啅
+慼
+捄
+咮
+睟
+譎
+嘷
+駃
+檥
+蹐
+窊
+駞
+雘
+趯
+篲
+讋
+睍
+毰
+憗
+鳷
+嚲
+圞
+歃
+緼
+賫
+籋
+繐
+麏
+灕
+礧
+歜
+飇
+鵁
+疢
+賖
+窆
+螮
+毹
+硉
+耡
+甔
+篛
+娭
+髩
+燋
+輜
+籧
+撝
+嬙
+徃
+驦
+𡏖
+麕
+馹
+覔
+鶠
+褷
+綍
+螗
+嗈
+彯
+篨
+炰
+鄮
+噞
+尅
+鷰
+鋭
+饉
+霢
+䔩
+坱
+裓
+帨
+忺
+豅
+栱
+謦
+傯
+誵
+骭
+潩
+鬒
+嵫
+悮
+扊
+扅
+轢
+惝
+臈
+舃
+鞾
+譟
+袵
+眎
+簏
+埸
+堧
+憸
+雰
+腷
+嵓
+隩
+趄
+墐
+褦
+艑
+狴
+玿
+竪
+恧
+姱
+抆
+恓
+霣
+躓
+鞲
+晬
+簴
+唼
+曵
+褕
+罣
+縐
+衘
+巃
+攲
+輀
+貎
+哳
+醭
+鋣
+僛
+迕
+蠭
+膓
+欝
+洊
+敺
+纎
+栟
+鞓
+蛬
+灺
+痏
+恡
+踸
+霔
+濵
+昻
+鉘
+楖
+竛
+竮
+窱
+幬
+慤
+儗
+黮
+嘐
+睆
+頇
+麑
+壼
+㦸
+顋
+瘥
+苖
+韈
+盻
+袷
+矼
+塼
+鐍
+傞
+苶
+吷
+噇
+鶖
+僣
+髧
+䅉
+鯫
+襏
+縳
+蠨
+痡
+髽
+剉
+蝱
+鄽
+匼
+嚚
+襫
+缿
+鵊
+燖
+忸
+摝
+攅
+牷
+氎
+騣
+颿
+虡
+腯
+漘
+矓
+祫
+顢
+綀
+弮
+柙
+蔾
+胾
+筤
+馽
+砆
+冩
+賙
+唶
+麛
+簜
+蹏
+屼
+鞶
+煑
+踠
+愀
+蠒
+頬
+韲
+戞
+畆
+笊
+搨
+捽
+絙
+覉
+澨
+趫
+矹
+穮
+愠
+劘
+轣
+卭
+鼪
+杕
+轗
+擐
+蚿
+恚
+檛
+𩕳
+靆
+轕
+餼
+頮
+槹
+蔉
+皜
+扄
+鮆
+轑
+蹡
+嵽
+甎
+蟈
+橅
+笴
+膰
+蕣
+澘
+髿
+樕
+褵
+蜋
+窼
+櫧
+雊
+胷
+嘵
+麄
+裋
+繢
+啐
+臛
+簁
+巓
+羜
+攧
+簮
+壊
+齩
+晹
+臲
+鬵
+齠
+媮
+幮
+壍
+蠛
+槜
+羖
+窓
+隃
+嚘
+輳
+籹
+凴
+崕
+獍
+嗸
+趦
+囅
+戺
+涬
+諉
+箯
+輊
+桹
+嵷
+㲲
+愊
+蒱
+洚
+赩
+輴
+幈
+齔
+嗁
+阽
+圠
+荈
+碔
+揎
+巀
+洏
+卼
+𨁝
+痁
+黳
+嗾
+䆗
+戃
+蕆
+頋
+悤
+掎
+㯝
+吚
+猘
+鮎
+鬴
+墁
+飋
+呿
+窀
+沲
+枒
+窌
+爼
+頞
+譡
+鶋
+湩
+㦬
+僾
+斒
+醼
+鶂
+磥
+揫
+犗
+齗
+鄶
+囏
+崪
+爞
+籓
+斮
+觝
+嵸
+驔
+䨴
+頺
+萑
+珓
+牸
+闒
+凘
+悢
+蟭
+濈
+嬄
+翽
+旍
+鶢
+罋
+輠
+怩
+頖
+趍
+壝
+嫮
+蕋
+踦
+轇
+眘
+巗
+嶭
+糓
+甽
+籺
+矟
+佖
+絏
+憮
+懡
+駈
+擕
+淟
+皡
+膋
+潨
+鳲
+趠
+麨
+頩
+漻
+輗
+墄
+賮
+㴩
+莟
+縦
+岝
+醻
+曚
+䙰
+噭
+醥
+筰
+躧
+踘
+鑕
+咈
+僶
+鶊
+鬂
+聼
+騐
+壒
+磎
+歗
+淈
+隟
+狃
+縋
+媻
+趲
+惙
+呫
+聮
+羾
+尫
+佽
+髼
+繋
+鬘
+旜
+疐
+阬
+䰐
+塈
+徤
+祊
+灂
+祅
+樷
+颾
+凟
+頀
+蠏
+塒
+衹
+婬
+裩
+粞
+憯
+匭
+筈
+盬
+霮
+黕
+靮
+伻
+緺
+瘝
+羑
+醸
+樝
+僎
+絓
+噆
+愞
+痗
+愽
+岊
+黤
+湑
+搉
+綯
+焮
+疉
+楛
+玼
+喤
+粔
+飂
+贐
+緉
+覰
+胔
+鞳
+摑
+墢
+斅
+誶
+僝
+鹺
+諌
+齅
+嵼
+讟
+冦
+脝
+婣
+緐
+茰
+飶
+欎
+慁
+抝
+瘉
+𡎺
+鈯
+瘃
+麫
+匊
+窞
+羓
+氄
+嚌
+姤
+橑
+駬
+冺
+騠
+㕙
+楶
+靸
+圎
+尀
+䙀
+鄏
+軃
+竁
+㹠
+刜
+剨
+罛
+鏹
+鬉
+簨
+藭
+藷
+僇
+瀫
+袨
+忮
+冡
+齯
+楪
+囋
+蟉
+醱
+尠
+牣
+攟
+袿
+齾
+甞
+啑
+潚
+樐
+絖
+酖
+觖
+骹
+嶅
+玃
+嫜
+廹
+儤
+矂
+艓
+挱
+骳
+嵳
+洴
+礓
+厪
+﨑
+禜
+籊
+瓻
+彛
+狁
+腪
+骾
+娯
+罻
+璅
+簳
+姢
+猰
+眹
+䴥
+堘
+搯
+怤
+緫
+聫
+涊
+熛
+輤
+䡾
+譌
+髇
+攛
+稭
+媕
+鬷
+跰
+縚
+鉧
+踧
+嚭
+襮
+藞
+滮
+颷
+荂
+蓰
+怫
+閧
+臕
+稛
+怗
+闑
+抶
+薶
+嶕
+瀺
+𥫗
+墝
+埆
+皥
+惷
+鞔
+鞺
+蟛
+瀡
+鎁
+酧
+恝
+齓
+嚄
+簔
+蟺
+㔶
+胹
+憖
+惄
+鸒
+貛
+軏
+縗
+蓻
+娵
+抺
+鼛
+虩
+歕
+矑
+繂
+襚
+倂
+廥
+諝
+虗
+弜
+兟
+繿
+偘
+翶
+肻
+棙
+斆
+碨
+醎
+蟢
+銙
+躠
+櫩
+椮
+絀
+鷾
+溳
+詖
+葓
+谼
+𦩘
+韔
+翿
+呑
+馡
+騊
+堁
+蓏
+䟃
+頟
+渢
+趑
+堄
+滛
+擫
+豭
+騩
+騘
+䍧
+彍
+忭
+餙
+馺
+忩
+芣
+矴
+噂
+滍
+慫
+𨍏
+怲
+扵
+搊
+昩
+嶻
+禬
+憃
+忼
+榰
+箾
+撁
+鈆
+袗
+脤
+騞
+哤
+螀
+靧
+梲
+囦
+魖
+褠
+䭔
+煆
+挃
+宷
+熉
+朘
+憭
+䒠
+謭
+鷤
+拕
+骫
+穾
+襭
+喓
+冞
+勩
+媢
+麚
+椓
+俙
+幐
+磝
+蜎
+灙
+漦
+㛹
+䭀
+㜷
+粻
+懟
+箳
+滣
+糉
+冐
+韤
+湱
+糭
+栳
+勌
+慱
+㸌
+罫
+筞
+霿
+躶
+玞
+磉
+罦
+祴
+媟
+猒
+擭
+恠
+嵁
+屴
+慆
+庬
+蟁
+㹀
+薧
+鷕
+渻
+朂
+愯
+齚
+蝻
+胏
+饙
+鳦
+鸃
+叅
+肧
+篂
+脗
+雺
+飰
+筀
+頥
+毶
+弌
+逓
+瞍
+絁
+鏚
+嚵
+攂
+醄
+奼
+獫
+絣
+靷
+畮
+褉
+棁
+揑
+楥
+橤
+襥
+蹮
+窔
+忪
+潠
+杇
+黲
+擄
+蚻
+蘙
+虙
+袐
+陿
+帊
+醟
+髖
+㞦
+鱭
+譸
+鮞
+栧
+扺
+脽
+擉
+岨
+黈
+餻
+佪
+遻
+鋟
+瞶
+廽
+懨
+墖
+玁
+籉
+宼
+鰋
+瑽
+垖
+酕
+漰
+戹
+蝛
+瑲
+阤
+褆
+儛
+䍽
+觕
+箘
+碯
+灨
+燀
+膇
+韀
+䳏
+詿
+禂
+韣
+踡
+碏
+尵
+莭
+庻
+篿
+狤
+㘞
+艭
+殱
+鵔
+槮
+猧
+劙
+獝
+㭊
+㾕
+蠚
+帤
+蹢
+蛚
+輼
+麀
+檃
+䰒
+䪫
+悾
+濳
+杗
+揾
+駏
+撦
+耈
+蟟
+狌
+鸖
+䨥
+餫
+鍰
+耉
+毚
+袽
+䱥
+慓
+䔿
+艖
+舋
+弰
+蠺
+嫓
+蚳
+髾
+喨
+鴐
+䍥
+韍
+柹
+掁
+薋
+攕
+飺
+凢
+麌
+嫰
+鑚
+黦
+葠
+吿
+栰
+踶
+芿
+穭
+啝
+筓
+褁
+稇
+顚
+䎘
+恇
+珷
+緪
+墠
+蛣
+蛜
+讕
+疻
+惎
+袝
+霡
+罸
+鬽
+苢
+喭
+飊
+唎
+澾
+襋
+皭
+廏
+蔿
+穊
+䝟
+駊
+獹
+夣
+褾
+慴
+軥
+讁
+軰
+瞷
+𡋯
+晜
+潗
+衋
+揵
+覼
+鱐
+醡
+䏰
+侐
+亁
+桞
+驘
+鬋
+鷽
+懞
+㵳
+儳
+豝
+傺
+搒
+縧
+硾
+䏶
+覻
+薍
+憝
+榠
+湆
+皵
+鎞
+菆
+糇
+矉
+搤
+紃
+峿
+磹
+甒
+琭
+𩥇
+菢
+禡
+渹
+刅
+迒
+敂
+蹜
+磓
+傪
+縿
+㕮
+涏
+䰀
+㡛
+韛
+犠
+餦
+圝
+焫
+㝢
+潬
+馵
+澟
+鱏
+譾
+㪍
+煼
+鍜
+窽
+紾
+堨
+䕸
+穅
+戅
+穄
+駴
+偫
+煗
+媠
+酘
+矬
+貆
+茞
+骩
+扠
+岞
+潓
+炧
+陊
+栭
+釱
+㡚
+篴
+耞
+鞉
+䋏
+𤫩
+椸
+儜
+痀
+謷
+潙
+寠
+牐
+嫭
+慅
+獧
+鈒
+欿
+薳
+蟂
+郲
+軨
+斨
+訦
+𠴲
+剺
+駪
+贙
+禫
+噣
+茢
+茙
+鄼
+揷
+魌
+䫻
+嗋
+噐
+侲
+諵
+𠺕
+挍
+䑳
+㨷
+槸
+靘
+㩧
+虣
+瑿
+衱
+襹
+餭
+㗶
+枑
+悋
+纑
+嶫
+儓
+髵
+甗
+榝
+㗭
+贗
+熸
+嬃
+礌
+偭
+樠
+栮
+鷼
+鵀
+澬
+眂
+牿
+骴
+呞
+爕
+牎
+巹
+帉
+砠
+梴
+䛏
+攃
+餁
+哿
+蹝
+崺
+閌
+醝
+臡
+麖
+駼
+賵
+夘
+骻
+愡
+俔
+諐
+觩
+莂
+饈
+殣
+溠
+冱
+埓
+厫
+虥
+芄
+慽
+竃
+埿
+仭
+褼
+倛
+韸
+牗
+幖
+禈
+穧
+蜧
+諞
+脞
+蝃
+飃
+煁
+涒
+谾
+覢
+赮
+鼘
+艗
+䶉
+鴥
+轒
+睅
+傔
+惵
+唈
+懆
+磣
+膢
+堶
+囈
+瘕
+誷
+瑘
+絝
+鬈
+嘽
+鷅
+梜
+喎
+鼟
+㟧
+劻
+眑
+剴
+痎
+餟
+庌
+菷
+梐
+吺
+躘
+慞
+罼
+穨
+摏
+釄
+莋
+呺
+砅
+鴽
+㘭
+㟅
+艴
+犉
+籕
+跐
+惏
+陗
+刋
+襘
+醹
+紽
+痌
+㗀
+撋
+陼
+駷
+艼
+踼
+癏
+慠
+趒
+邍
+姞
+䂬
+堲
+苙
+椌
+嗃
+挶
+岯
+禗
+嵔
+觡
+豜
+睩
+㒿
+塠
+燂
+扤
+恟
+鬝
+鬇
+鬡
+揳
+霠
+㗫
+苐
+蒀
+圌
+戭
+䖃
+𥈭
+勮
+耝
+轞
+胮
+墯
+枮
+罿
+浺
+綪
+爓
+蘃
+襍
+轜
+閠
+畽
+鄊
+嶆
+籭
+蠯
+陑
+瘽
+迆
+賷
+䍡
+韂
+躃
+禴
+簄
+瓟
+碐
+躨
+侜
+岍
+䃸
+趚
+髐
+榅
+粣
+屝
+鴃
+圁
+蝜
+黫
+僽
+丗
+靣
+湏
+抏
+㟽
+跙
+餤
+朙
+㹞
+瞖
+繣
+㨫
+罙
+糒
+惉
+葽
+鼮
+蕳
+豏
+𥱼
+鵮
+獦
+悕
+𠴨
+闟
+惽
+慿
+隉
+椷
+𩅰
+艛
+眽
+凓
+儃
+奨
+埀
+瑫
+駚
+濇
+緶
+峉
+礨
+髢
+瞯
+壥
+姡
+㟯
+髬
+啀
+㶿
+歅
+殀
+縩
+疈
+鳸
+霳
+稬
+圊
+彚
+裠
+埳
+褋
+㔩
+矲
+剶
+硋
+聦
+峞
+浰
+窵
+嘂
+睘
+簵
+腒
+韘
+躣
+甈
+忲
+舽
+襂
+硠
+脃
+鐏
+奯
+脧
+矕
+䠞
+駹
+豶
+訑
+柸
+鰅
+瘨
+趿
+糦
+蟏
+饛
+尰
+諑
+汃
+毺
+鋃
+絚
+馧
+艬
+枍
+爊
+峗
+泙
+碖
+鵕
+尩
+閗
+𤧚
+幩
+塉
+箊
+覂
+玒
+橧
+謟
+庨
+籔
+欑
+厎
+尭
+氉
+蠈
+䓞
+矙
+梡
+瀩
+溔
+煴
+蔲
+僬
+嵢
+梩
+弝
+𣙙
+鞟
+敉
+鮚
+湠
+鐐
+爣
+裻
+䶎
+𦨴
+謿
+垾
+蝂
+睂
+癙
+韽
+㟳
+桒
+鳿
+樏
+峛
+瑉
+僄
+顣
+衺
+殗
+肦
+圑
+朒
+喌
+犦
+㰅
+疁
+氃
+吰
+陻
+盰
+娀
+魶
+㖃
+曒
+娿
+獱
+孏
+酅
+蝡
+齰
+莬
+鄀
+逥
+挿
+觵
+縆
+㟝
+繍
+碙
+㑂
+䎳
+兾
+壸
+賝
+桯
+跁
+跒
+蔍
+舼
+忀
+懭
+媌
+罭
+菵
+狔
+靿
+拪
+㲉
+䔲
+嬀
+鵽
+涳
+朾
+𡸣
+𢫫
+虈
+㜮
+顑
+櫋
+蔪
+旝
+湡
+蹛
+稆
+唽
+㟏
+熂
+龡
+煟
+韅
+韐
+慂
+剳
+掫
+兠
+摋
+羫
+璊
+鵻
+駓
+佌
+蜹
+晲
+矒
+玅
+剰
+斶
+紖
+懴
+駜
+羢
+麳
+㳷
+馞
+爥
+鍚
+鑢
+螵
+嗺
+鏨
+𠙶
+疪
+鷔
+鮧
+轊
+栘
+鼜
+睗
+蟘
+枓
+䖟
+剠
+瞤
+圛
+椳
+籸
+䪌
+鯹
+湌
+丳
+賧
+縭
+檾
+𦨻
+撆
+䩫
+磢
+惥
+譀
+罤
+鞸
+鉎
+㶏
+膁
+甋
+瓀
+懹
+槢
+硊
+弆
+琫
+嵠
+駻
+湢
+杮
+䌨
+訹
+藇
+穯
+蠉
+曭
+蹎
+詄
+毷
+𩃎
+熁
+灜
+蜫
+蜳
+昈
+帩
+鈋
+䐹
+顖
+鄹
+匶
+毾
+礜
+堭
+婞
+鷿
+㙞
+詀
+瘮
+䫜
+㾪
+捘
+屫
+誧
+䲔
+閍
+蒳
+㬋
+遟
+嶀
+葐
+蜼
+㻱
+曡
+䃜
+濴
+䦱
+霫
+譆
+霋
+蕰
+襓
+氋
+鴷
+魦
+㩻
+㡠
+灉
+贑
+燑
+峝
+輷
+烻
+耼
+螉
+跜
+豩
+㑃
+藙
+鋂
+胐
+𣔻
+紒
+瓓
+塯
+辴
+趷
+堛
+㒟
+㗲
+㬊
+䄡
+卄
+姧
+猓
+躗
+覤
+醊
+兎
+罯
+痯
+覸
+詉
+癿
+岋
+歝
+茟
+㘆
+㮰
+淜
+𥉌
+㫰
+鈌
+毵
+狉
+贜
+峬
+汻
+誖
+烓
+睋
+潎
+䲺
+㠓
+歖
+𠜱
+槵
+熚
+萷
+磤
+絸
+鷞
+聻
+屷
+㝵
+諕
+瘂
+㺷
+蚰
+柦
+䍐
+泿
+礰
+摎
+㜕
+㻞
+洓
+喍
+囌
+囐
+䙱
+腨
+妉
+鄛
+鄥
+㵝
+輧
+鱄
+騟
+鈚
+廜
+𨗨
+㶼
+膞
+崯
+硞
+萆
+眒
+譩
+揬
+藑
+匌
+㠾
+㥏
+㢮
+䕢
+帣
+酭
+枦
+孅
+鞙
+丷
+鍭
+䤴
+餂
+愗
+冘
+埛
+㒇
+郕
+蔯
+簰
+刔
+蠩
+耏
+鞹
+𧑅
+觹
+䐑
+磶
+蹵
+鵃
+耛
+蓤
+臄
+轙
+庤
+㒩
+翐
+榥
+晀
+輣
+蟚
+拲
+皠
+穱
+䃔
+䃧
+窡
+絍
+礿
+鑞
+栯
+㾓
+掿
+厞
+淂
+撶
+伹
+鹻
+軓
+岹
+蚷
+榸
+刾
+艂
+㤝
+塕
+蚔
+藾
+攓
+鏬
+珫
+黪
+蟧
+猭
+漑
+粺
+驆
+撘
+亾
+㼌
+蝑
+澓
+揞
+欱
+愶
+泲
+醷
+螴
+芚
+絻
+轃
+漮
+唪
+岉
+鬀
+䱹
+齖
+䂓
+趢
+荓
+覶
+鯾
+諿
+槥
+嚆
+爢
+瓬
+笐
+篢
+舝
+襵
+鎒
+𤝞
+肭
+瘇
+笓
+餑
+豋
+湗
+緎
+肐
+胲
+掤
+潫
+䖴
+𠎝
+𨺗
+諢
+毈
+寱
+唲
+䃭
+峮
+狘
+韊
+䬝
+呰
+㹱
+碞
+畞
+㠌
+黭
+蚘
+豵
+穥
+尯
+㳇
+隵
+灇
+壜
+楰
+彲
+甤
+綹
+旞
+𡏟
+曁
+喩
+𥲤
+郈
+塺
+訧
+絿
+掔
+蠮
+𡱰
+䃺
+宻
+灎
+羵
+𨠵
+糚
+摉
+壷
+勴
+瑃
+鎝
+𥜥
+婥
+鬺
+扢
+肣
+溰
+磩
+耇
+宎
+㔇
+霱
+敚
+汳
+鏄
+儹
+隥
+㿉
+膆
+崏
+𦭵
+郔
+扂
+垗
+㳂
+礛
+缻
+垜
+晱
+訩
+蘪
+珇
+怮
+垝
+㔢
+憛
+痝
+蟨
+鞁
+鶤
+肎
+傝
+䢆
+䰄
+𥊚
+㖀
+㠭
+壵
+墋
+㠔
+橜
+怓
+蚹
+塛
+憪
+鋝
+腶
+嶾
+翍
+溓
+齼
+蔂
+䃂
+鉺
+攑
+瓐
+泎
+眤
+邘
+崝
+稡
+愸
+髥
+輹
+詨
+髆
+麃
+虤
+洐
+婐
+挏
+峑
+嶣
+篬
+葄
+瑎
+瓉
+㳅
+葼
+姙
+䪜
+𩇕
+焭
+剚
+濪
+霵
+僒
+
+羭
diff --git a/ppocr/utils/dict/ppocrv5_arabic_dict.txt b/ppocr/utils/dict/ppocrv5_arabic_dict.txt
new file mode 100644
index 0000000..9c5955b
--- /dev/null
+++ b/ppocr/utils/dict/ppocrv5_arabic_dict.txt
@@ -0,0 +1,747 @@
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+<
+=
+>
+?
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+[
+]
+_
+`
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+©
+‥
+{
+}
+\
+|
+@
+^
+~
+÷
+∕
+∙
+⋅
+·
+±
+∓
+∩
+∪
+□
+←
+↔
+⇒
+⇐
+⇔
+∀
+∃
+∄
+∴
+∵
+∝
+∞
+⊥
+∟
+∠
+∡
+∢
+′
+″
+∥
+⊾
+⊿
+∂
+∫
+∬
+∭
+∮
+∯
+∰
+∑
+∏
+√
+∛
+∜
+∱
+∲
+∳
+∶
+∷
+∼
+®
+℉
+Ω
+℧
+Å
+⌀
+ℏ
+⅀
+⍺
+⍵
+¢
+€
+£
+¥
+₿
+Ⅰ
+Ⅱ
+Ⅲ
+Ⅳ
+Ⅴ
+Ⅵ
+Ⅶ
+Ⅷ
+Ⅸ
+Ⅹ
+Ⅺ
+Ⅻ
+ⅰ
+ⅱ
+ⅲ
+ⅳ
+ⅴ
+ⅵ
+ⅶ
+ⅷ
+ⅸ
+ⅹ
+ⅺ
+ⅻ
+➀
+➁
+➂
+➃
+➄
+➅
+➆
+➇
+➈
+➉
+➊
+➋
+➌
+➍
+➎
+➏
+➐
+➑
+➒
+➓
+❶
+❷
+❸
+❹
+❺
+❻
+❼
+❽
+❾
+❿
+①
+②
+③
+④
+⑤
+⑥
+⑦
+⑧
+⑨
+⑩
+●
+▶
+𝑢
+︽
+–
+﹥
+𝜓
+•
+∋
+ƒ
+०
+⬆
+Ạ
+◀
+
+▫
+︾
+À
+Á
+Â
+Ã
+Ä
+Å
+Æ
+Ç
+È
+É
+Ê
+Ë
+Ì
+Í
+Î
+Ï
+Ð
+Ñ
+Ò
+Ó
+Ô
+Õ
+Ö
+Ø
+Ù
+Ú
+Û
+Ü
+Ý
+Þ
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+ì
+í
+î
+ï
+ð
+ñ
+ò
+ó
+ô
+õ
+ö
+ø
+ù
+ú
+û
+ü
+ý
+þ
+ÿ
+¡
+¤
+¦
+§
+¨
+ª
+«
+¬
+¯
+°
+²
+³
+´
+µ
+¶
+¸
+¹
+º
+»
+¼
+½
+¾
+¿
+×
+‐
+‑
+‒
+—
+―
+‖
+‗
+‘
+’
+‚
+‛
+“
+”
+„
+‟
+†
+‡
+‣
+․
+…
+‧
+‰
+‴
+‵
+‶
+‷
+‸
+‹
+›
+※
+‼
+‽
+‾
+₤
+₡
+₹
+−
+∖
+∗
+≈
+≠
+≡
+≤
+≥
+⊂
+⊃
+↑
+→
+↓
+↕
+™
+Ω
+℮
+∆
+✓
+✗
+✘
+▪
+◼
+✔
+✕
+☑
+☒
+№
+₽
+₴
+Α
+α
+Β
+β
+Γ
+γ
+Δ
+δ
+Ε
+ε
+Ζ
+ζ
+Η
+η
+Θ
+θ
+Ι
+ι
+Κ
+κ
+Λ
+λ
+Μ
+μ
+Ν
+ν
+Ξ
+ξ
+Ο
+ο
+Π
+π
+Ρ
+ρ
+Σ
+σ
+ς
+Τ
+τ
+Υ
+υ
+Φ
+φ
+Χ
+χ
+Ψ
+ψ
+ω
+ـ
+،
+؍
+؎
+؏
+ؐ
+ؑ
+ؒ
+ؓ
+ؔ
+ؕ
+ؖ
+ؗ
+ؘ
+ؙ
+ؚ
+؛
+؞
+؟
+ء
+آ
+أ
+ؤ
+إ
+ئ
+ا
+ب
+ة
+ت
+ث
+ج
+ح
+خ
+د
+ذ
+ر
+ز
+س
+ش
+ص
+ض
+ط
+ظ
+ع
+غ
+ف
+ق
+ك
+ل
+م
+ن
+ه
+و
+ى
+ي
+ً
+ٌ
+ٍ
+َ
+ُ
+ِ
+ّ
+ْ
+ٓ
+ٔ
+ٕ
+ٖ
+ٗ
+ٚ
+ٛ
+ٜ
+ٝ
+ٞ
+ٟ
+ٰ
+ٱ
+ٲ
+ٳ
+ٴ
+ٵ
+ٶ
+ٷ
+ٸ
+ٹ
+ٺ
+ٻ
+ټ
+ٽ
+پ
+ٿ
+ڀ
+ځ
+ڂ
+ڃ
+ڄ
+څ
+چ
+ڇ
+ڈ
+ډ
+ڊ
+ڋ
+ڌ
+ڍ
+ڎ
+ڏ
+ڐ
+ڑ
+ڒ
+ړ
+ڔ
+ڕ
+ږ
+ڗ
+ژ
+ڙ
+ښ
+ڛ
+ڜ
+ڝ
+ڞ
+ڟ
+ڠ
+ڡ
+ڢ
+ڣ
+ڤ
+ڥ
+ڦ
+ڧ
+ڨ
+ک
+ڪ
+ګ
+ڬ
+ڭ
+ڮ
+گ
+ڰ
+ڱ
+ڲ
+ڳ
+ڴ
+ڵ
+ڶ
+ڷ
+ڸ
+ڹ
+ں
+ڻ
+ڼ
+ڽ
+ھ
+ڿ
+ۀ
+ہ
+ۂ
+ۃ
+ۄ
+ۅ
+ۆ
+ۇ
+ۈ
+ۉ
+ۊ
+ۋ
+ی
+ۍ
+ێ
+ۏ
+ې
+ۑ
+ے
+ۓ
+۔
+ە
+ۖ
+ۗ
+ۘ
+ۙ
+ۚ
+ۛ
+ۜ
+۞
+ۡ
+ۢ
+ۣ
+ۤ
+ۥ
+ۦ
+ۧ
+ۨ
+۩
+ۭ
+ۮ
+ۯ
+ݐ
+ݑ
+ݒ
+ݓ
+ݔ
+ݕ
+ݖ
+ݗ
+ݘ
+ݙ
+ݚ
+ݛ
+ݜ
+ݝ
+ݞ
+ݟ
+ݠ
+ݡ
+ݢ
+ݣ
+ݤ
+ݥ
+ݦ
+ݧ
+ݨ
+ݩ
+ݪ
+ݫ
+ݬ
+ݭ
+ݮ
+ݯ
+ݰ
+ݱ
+ݲ
+ݳ
+ݴ
+ݵ
+ݶ
+ݷ
+ݸ
+ݹ
+ݺ
+ݻ
+ݼ
+ݽ
+ݾ
+ݿ
+٠
+١
+٢
+٣
+٤
+٥
+٦
+٧
+٨
+٩
+٪
+٫
+٬
+۰
+۱
+۲
+۳
+۴
+۵
+۶
+۷
+۸
+۹
+ﭙ
+ﭚ
+ﭪ
+ﭭ
+ﭺ
+ﭽ
+ﷲ
+﷼
+٠
+١
+٢
+٣
+٤
+٥
+٦
+٧
+٨
+٩
+٪
+٫
+٬
+۰
+۱
+۲
+۳
+۴
+۵
+۶
+۷
+۸
+۹
+ﭙ
+ﭚ
+ﭪ
+ﭭ
+ﭺ
+ﭽ
+ﷲ
+﷼
diff --git a/ppocr/utils/dict/ppocrv5_cyrillic_dict.txt b/ppocr/utils/dict/ppocrv5_cyrillic_dict.txt
new file mode 100644
index 0000000..7625a74
--- /dev/null
+++ b/ppocr/utils/dict/ppocrv5_cyrillic_dict.txt
@@ -0,0 +1,850 @@
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+<
+=
+>
+?
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+[
+]
+_
+`
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+©
+‥
+{
+}
+\
+|
+@
+^
+~
+÷
+∕
+∙
+⋅
+·
+±
+∓
+∩
+∪
+□
+←
+↔
+⇒
+⇐
+⇔
+∀
+∃
+∄
+∴
+∵
+∝
+∞
+⊥
+∟
+∠
+∡
+∢
+′
+″
+∥
+⊾
+⊿
+∂
+∫
+∬
+∭
+∮
+∯
+∰
+∑
+∏
+√
+∛
+∜
+∱
+∲
+∳
+∶
+∷
+∼
+®
+℉
+Ω
+℧
+Å
+⌀
+ℏ
+⅀
+⍺
+⍵
+¢
+€
+£
+¥
+₿
+Ⅰ
+Ⅱ
+Ⅲ
+Ⅳ
+Ⅴ
+Ⅵ
+Ⅶ
+Ⅷ
+Ⅸ
+Ⅹ
+Ⅺ
+Ⅻ
+ⅰ
+ⅱ
+ⅲ
+ⅳ
+ⅴ
+ⅵ
+ⅶ
+ⅷ
+ⅸ
+ⅹ
+ⅺ
+ⅻ
+➀
+➁
+➂
+➃
+➄
+➅
+➆
+➇
+➈
+➉
+➊
+➋
+➌
+➍
+➎
+➏
+➐
+➑
+➒
+➓
+❶
+❷
+❸
+❹
+❺
+❻
+❼
+❽
+❾
+❿
+①
+②
+③
+④
+⑤
+⑥
+⑦
+⑧
+⑨
+⑩
+●
+▶
+𝑢
+︽
+–
+﹥
+𝜓
+•
+∋
+ƒ
+०
+⬆
+Ạ
+◀
+
+▫
+︾
+À
+Á
+Â
+Ã
+Ä
+Å
+Æ
+Ç
+È
+É
+Ê
+Ë
+Ì
+Í
+Î
+Ï
+Ð
+Ñ
+Ò
+Ó
+Ô
+Õ
+Ö
+Ø
+Ù
+Ú
+Û
+Ü
+Ý
+Þ
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+ì
+í
+î
+ï
+ð
+ñ
+ò
+ó
+ô
+õ
+ö
+ø
+ù
+ú
+û
+ü
+ý
+þ
+ÿ
+¡
+¤
+¦
+§
+¨
+ª
+«
+¬
+¯
+°
+²
+³
+´
+µ
+¶
+¸
+¹
+º
+»
+¼
+½
+¾
+¿
+×
+‐
+‑
+‒
+—
+―
+‖
+‗
+‘
+’
+‚
+‛
+“
+”
+„
+‟
+†
+‡
+‣
+․
+…
+‧
+‰
+‴
+‵
+‶
+‷
+‸
+‹
+›
+※
+‼
+‽
+‾
+₤
+₡
+₹
+−
+∖
+∗
+≈
+≠
+≡
+≤
+≥
+⊂
+⊃
+↑
+→
+↓
+↕
+™
+Ω
+℮
+∆
+✓
+✗
+✘
+▪
+◼
+✔
+✕
+☑
+☒
+№
+₽
+₴
+Α
+α
+Β
+β
+Γ
+γ
+Δ
+δ
+Ε
+ε
+Ζ
+ζ
+Η
+η
+Θ
+θ
+Ι
+ι
+Κ
+κ
+Λ
+λ
+Μ
+μ
+Ν
+ν
+Ξ
+ξ
+Ο
+ο
+Π
+π
+Ρ
+ρ
+Σ
+σ
+ς
+Τ
+τ
+Υ
+υ
+Φ
+φ
+Χ
+χ
+Ψ
+ψ
+ω
+Ѐ
+Ё
+Ђ
+Ѓ
+Є
+Ѕ
+І
+Ї
+Ј
+Љ
+Њ
+Ћ
+Ќ
+Ѝ
+Ў
+Џ
+А
+Б
+В
+Г
+Д
+Е
+Ж
+З
+И
+Й
+К
+Л
+М
+Н
+О
+П
+Р
+С
+Т
+У
+Ф
+Х
+Ц
+Ч
+Ш
+Щ
+Ъ
+Ы
+Ь
+Э
+Ю
+Я
+а
+б
+в
+г
+д
+е
+ж
+з
+и
+й
+к
+л
+м
+н
+о
+п
+р
+с
+т
+у
+ф
+х
+ц
+ч
+ш
+щ
+ъ
+ы
+ь
+э
+ю
+я
+ѐ
+ё
+ђ
+ѓ
+є
+ѕ
+і
+ї
+ј
+љ
+њ
+ћ
+ќ
+ѝ
+ў
+џ
+Ѡ
+ѡ
+Ѣ
+ѣ
+Ѥ
+ѥ
+Ѧ
+ѧ
+Ѩ
+ѩ
+Ѫ
+ѫ
+Ѭ
+ѭ
+Ѯ
+ѯ
+Ѱ
+ѱ
+Ѳ
+ѳ
+Ѵ
+ѵ
+Ѷ
+ѷ
+Ѹ
+ѹ
+Ѻ
+ѻ
+Ѽ
+ѽ
+Ѿ
+ѿ
+Ҁ
+ҁ
+Ҋ
+ҋ
+Ҍ
+ҍ
+Ҏ
+ҏ
+Ґ
+ґ
+Ғ
+ғ
+Ҕ
+ҕ
+Җ
+җ
+Ҙ
+ҙ
+Қ
+қ
+Ҝ
+ҝ
+Ҟ
+ҟ
+Ҡ
+ҡ
+Ң
+ң
+Ҧ
+ҧ
+Ҩ
+ҩ
+Ҫ
+ҫ
+Ҭ
+ҭ
+Ү
+ү
+Ұ
+ұ
+Ҳ
+ҳ
+Ҷ
+ҷ
+Ҹ
+ҹ
+Һ
+һ
+Ҽ
+ҽ
+Ҿ
+ҿ
+Ӏ
+Ӂ
+ӂ
+Ӄ
+ӄ
+Ӆ
+ӆ
+Ӈ
+ӈ
+Ӊ
+ӊ
+Ӌ
+ӌ
+Ӎ
+ӎ
+ӏ
+Ӑ
+ӑ
+Ӓ
+ӓ
+Ӗ
+ӗ
+Ә
+ә
+Ӛ
+ӛ
+Ӝ
+ӝ
+Ӟ
+ӟ
+Ӡ
+ӡ
+Ӣ
+ӣ
+Ӥ
+ӥ
+Ӧ
+ӧ
+Ө
+ө
+Ӫ
+ӫ
+Ӭ
+ӭ
+Ӯ
+ӯ
+Ӱ
+ӱ
+Ӳ
+ӳ
+Ӵ
+ӵ
+Ӷ
+ӷ
+Ӹ
+ӹ
+Ӻ
+ӻ
+Ӽ
+ӽ
+Ӿ
+ӿ
+Ԁ
+ԁ
+Ԃ
+ԃ
+Ԅ
+ԅ
+Ԇ
+ԇ
+Ԉ
+ԉ
+Ԋ
+ԋ
+Ԍ
+ԍ
+Ԏ
+ԏ
+Ԑ
+ԑ
+Ԓ
+ԓ
+Ԕ
+ԕ
+Ԗ
+ԗ
+Ԙ
+ԙ
+Ԛ
+ԛ
+Ԝ
+ԝ
+Ԟ
+ԟ
+Ԡ
+ԡ
+Ԣ
+ԣ
+Ԥ
+ԥ
+Ԧ
+ԧ
+Ԩ
+ԩ
+Ԫ
+ԫ
+Ԭ
+ԭ
+Ԯ
+ԯ
+ⷠ
+ⷡ
+ⷢ
+ⷣ
+ⷤ
+ⷥ
+ⷦ
+ⷧ
+ⷨ
+ⷩ
+ⷪ
+ⷫ
+ⷬ
+ⷭ
+ⷮ
+ⷯ
+ⷰ
+ⷱ
+ⷲ
+ⷳ
+ⷴ
+ⷵ
+ⷶ
+ⷷ
+ⷸ
+ⷹ
+ⷺ
+ⷻ
+ⷼ
+ⷽ
+ⷾ
+ⷿ
+Ꙁ
+ꙁ
+Ꙃ
+ꙃ
+Ꙅ
+ꙅ
+Ꙇ
+ꙇ
+Ꙉ
+ꙉ
+Ꙋ
+ꙋ
+Ꙍ
+ꙍ
+Ꙏ
+ꙏ
+Ꙑ
+ꙑ
+Ꙓ
+ꙓ
+Ꙕ
+ꙕ
+Ꙗ
+ꙗ
+Ꙙ
+ꙙ
+Ꙛ
+ꙛ
+Ꙝ
+ꙝ
+Ꙟ
+ꙟ
+Ꙡ
+ꙡ
+Ꙣ
+ꙣ
+Ꙥ
+ꙥ
+Ꙧ
+ꙧ
+Ꙩ
+ꙩ
+Ꙫ
+ꙫ
+Ꙭ
+ꙭ
+ꙮ
+ꙴ
+ꙵ
+ꙶ
+ꙷ
+ꙸ
+ꙹ
+ꙺ
+ꙻ
+Ꚁ
+ꚁ
+Ꚃ
+ꚃ
+Ꚅ
+ꚅ
+Ꚇ
+ꚇ
+Ꚉ
+ꚉ
+Ꚋ
+ꚋ
+Ꚍ
+ꚍ
+Ꚏ
+ꚏ
+Ꚑ
+ꚑ
+Ꚓ
+ꚓ
+Ꚕ
+ꚕ
+Ꚗ
+ꚗ
+Ꚙ
+ꚙ
+Ꚛ
+ꚛ
+ꚜ
+ꚝ
+ꚞ
+ꚟ
diff --git a/ppocr/utils/dict/ppocrv5_devanagari_dict.txt b/ppocr/utils/dict/ppocrv5_devanagari_dict.txt
new file mode 100644
index 0000000..498ec2a
--- /dev/null
+++ b/ppocr/utils/dict/ppocrv5_devanagari_dict.txt
@@ -0,0 +1,568 @@
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+<
+=
+>
+?
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+[
+]
+_
+`
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+©
+‥
+{
+}
+\
+|
+@
+^
+~
+÷
+∕
+∙
+⋅
+·
+±
+∓
+∩
+∪
+□
+←
+↔
+⇒
+⇐
+⇔
+∀
+∃
+∄
+∴
+∵
+∝
+∞
+⊥
+∟
+∠
+∡
+∢
+′
+″
+∥
+⊾
+⊿
+∂
+∫
+∬
+∭
+∮
+∯
+∰
+∑
+∏
+√
+∛
+∜
+∱
+∲
+∳
+∶
+∷
+∼
+®
+℉
+Ω
+℧
+Å
+⌀
+ℏ
+⅀
+⍺
+⍵
+¢
+€
+£
+¥
+₿
+Ⅰ
+Ⅱ
+Ⅲ
+Ⅳ
+Ⅴ
+Ⅵ
+Ⅶ
+Ⅷ
+Ⅸ
+Ⅹ
+Ⅺ
+Ⅻ
+ⅰ
+ⅱ
+ⅲ
+ⅳ
+ⅴ
+ⅵ
+ⅶ
+ⅷ
+ⅸ
+ⅹ
+ⅺ
+ⅻ
+➀
+➁
+➂
+➃
+➄
+➅
+➆
+➇
+➈
+➉
+➊
+➋
+➌
+➍
+➎
+➏
+➐
+➑
+➒
+➓
+❶
+❷
+❸
+❹
+❺
+❻
+❼
+❽
+❾
+❿
+①
+②
+③
+④
+⑤
+⑥
+⑦
+⑧
+⑨
+⑩
+●
+▶
+𝑢
+︽
+–
+﹥
+𝜓
+•
+∋
+ƒ
+⬆
+Ạ
+◀
+
+▫
+︾
+À
+Á
+Â
+Ã
+Ä
+Å
+Æ
+Ç
+È
+É
+Ê
+Ë
+Ì
+Í
+Î
+Ï
+Ð
+Ñ
+Ò
+Ó
+Ô
+Õ
+Ö
+Ø
+Ù
+Ú
+Û
+Ü
+Ý
+Þ
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+ì
+í
+î
+ï
+ð
+ñ
+ò
+ó
+ô
+õ
+ö
+ø
+ù
+ú
+û
+ü
+ý
+þ
+ÿ
+¡
+¤
+¦
+§
+¨
+ª
+«
+¬
+¯
+°
+²
+³
+´
+µ
+¶
+¸
+¹
+º
+»
+¼
+½
+¾
+¿
+×
+‐
+‑
+‒
+—
+―
+‖
+‗
+‘
+’
+‚
+‛
+“
+”
+„
+‟
+†
+‡
+‣
+․
+…
+‧
+‰
+‴
+‵
+‶
+‷
+‸
+‹
+›
+※
+‼
+‽
+‾
+₤
+₡
+₹
+−
+∖
+∗
+≈
+≠
+≡
+≤
+≥
+⊂
+⊃
+↑
+→
+↓
+↕
+™
+Ω
+℮
+∆
+✓
+✗
+✘
+▪
+◼
+✔
+✕
+☑
+☒
+№
+₽
+₴
+Α
+α
+Β
+β
+Γ
+γ
+Δ
+δ
+Ε
+ε
+Ζ
+ζ
+Η
+η
+Θ
+θ
+Ι
+ι
+Κ
+κ
+Λ
+λ
+Μ
+μ
+Ν
+ν
+Ξ
+ξ
+Ο
+ο
+Π
+π
+Ρ
+ρ
+Σ
+σ
+ς
+Τ
+τ
+Υ
+υ
+Φ
+φ
+Χ
+χ
+Ψ
+ψ
+ω
+ऀ
+ँ
+ं
+ः
+ऄ
+अ
+आ
+इ
+ई
+उ
+ऊ
+ऋ
+ऌ
+ऍ
+ऎ
+ए
+ऐ
+ऑ
+ऒ
+ओ
+औ
+क
+ख
+ग
+घ
+ङ
+च
+छ
+ज
+झ
+ञ
+ट
+ठ
+ड
+ढ
+ण
+त
+थ
+द
+ध
+न
+ऩ
+प
+फ
+ब
+भ
+म
+य
+र
+ऱ
+ल
+ळ
+ऴ
+व
+श
+ष
+स
+ह
+ऺ
+ऻ
+़
+ऽ
+ा
+ि
+ी
+ु
+ू
+ृ
+ॄ
+ॅ
+ॆ
+े
+ै
+ॉ
+ॊ
+ो
+ौ
+्
+ॎ
+ॏ
+ॐ
+॑
+॒
+॓
+॔
+ॕ
+ॖ
+ॗ
+क़
+ख़
+ग़
+ज़
+ड़
+ढ़
+फ़
+य़
+ॠ
+ॡ
+ॢ
+ॣ
+।
+॥
+०
+१
+२
+३
+४
+५
+६
+७
+८
+९
+॰
+ॱ
+ॲ
+ॳ
+ॴ
+ॵ
+ॶ
+ॷ
+ॸ
+ॹ
+ॺ
+ॻ
+ॼ
+ॽ
+ॾ
+ॿ
diff --git a/ppocr/utils/dict/ppocrv5_dict.txt b/ppocr/utils/dict/ppocrv5_dict.txt
new file mode 100644
index 0000000..567898b
--- /dev/null
+++ b/ppocr/utils/dict/ppocrv5_dict.txt
@@ -0,0 +1,18383 @@
+　
+一
+乙
+二
+十
+丁
+厂
+七
+卜
+八
+人
+入
+儿
+匕
+几
+九
+刁
+了
+刀
+力
+乃
+又
+三
+干
+于
+亏
+工
+土
+士
+才
+下
+寸
+大
+丈
+与
+万
+上
+小
+口
+山
+巾
+千
+乞
+川
+亿
+个
+夕
+久
+么
+勺
+凡
+丸
+及
+广
+亡
+门
+丫
+义
+之
+尸
+己
+已
+巳
+弓
+子
+卫
+也
+女
+刃
+飞
+习
+叉
+马
+乡
+丰
+王
+开
+井
+天
+夫
+元
+无
+云
+专
+丐
+扎
+艺
+木
+五
+支
+厅
+不
+犬
+太
+区
+历
+歹
+友
+尤
+匹
+车
+巨
+牙
+屯
+戈
+比
+互
+切
+瓦
+止
+少
+曰
+日
+中
+贝
+冈
+内
+水
+见
+午
+牛
+手
+气
+毛
+壬
+升
+夭
+长
+仁
+什
+片
+仆
+化
+仇
+币
+仍
+仅
+斤
+爪
+反
+介
+父
+从
+仑
+今
+凶
+分
+乏
+公
+仓
+月
+氏
+勿
+欠
+风
+丹
+匀
+乌
+勾
+凤
+六
+文
+亢
+方
+火
+为
+斗
+忆
+计
+订
+户
+认
+冗
+讥
+心
+尺
+引
+丑
+巴
+孔
+队
+办
+以
+允
+予
+邓
+劝
+双
+书
+幻
+玉
+刊
+未
+末
+示
+击
+打
+巧
+正
+扑
+卉
+扒
+功
+扔
+去
+甘
+世
+艾
+古
+节
+本
+术
+可
+丙
+左
+厉
+石
+右
+布
+夯
+戊
+龙
+平
+灭
+轧
+东
+卡
+北
+占
+凸
+卢
+业
+旧
+帅
+归
+旦
+目
+且
+叶
+甲
+申
+叮
+电
+号
+田
+由
+只
+叭
+史
+央
+兄
+叽
+叼
+叫
+叩
+叨
+另
+叹
+冉
+皿
+凹
+囚
+四
+生
+矢
+失
+乍
+禾
+丘
+付
+仗
+代
+仙
+们
+仪
+白
+仔
+他
+斥
+瓜
+乎
+丛
+令
+用
+甩
+印
+尔
+乐
+句
+匆
+册
+卯
+犯
+外
+处
+冬
+鸟
+务
+包
+饥
+主
+市
+立
+冯
+玄
+闪
+兰
+半
+汁
+汇
+头
+汉
+宁
+穴
+它
+讨
+写
+让
+礼
+训
+议
+必
+讯
+记
+永
+司
+尼
+民
+弗
+弘
+出
+辽
+奶
+奴
+召
+加
+皮
+边
+孕
+发
+圣
+对
+台
+矛
+纠
+母
+幼
+丝
+邦
+式
+迂
+刑
+戎
+动
+扛
+寺
+吉
+扣
+考
+托
+老
+巩
+圾
+执
+扩
+扫
+地
+场
+扬
+耳
+芋
+共
+芒
+亚
+芝
+朽
+朴
+机
+权
+过
+臣
+吏
+再
+协
+西
+压
+厌
+戌
+在
+百
+有
+存
+而
+页
+匠
+夸
+夺
+灰
+达
+列
+死
+成
+夹
+夷
+轨
+邪
+尧
+划
+迈
+毕
+至
+此
+贞
+师
+尘
+尖
+劣
+光
+当
+早
+吁
+吐
+吓
+虫
+曲
+团
+吕
+同
+吊
+吃
+因
+吸
+吗
+吆
+屿
+屹
+岁
+帆
+回
+岂
+则
+刚
+网
+肉
+年
+朱
+先
+丢
+廷
+舌
+竹
+迁
+乔
+迄
+伟
+传
+乒
+乓
+休
+伍
+伏
+优
+臼
+伐
+延
+仲
+件
+任
+伤
+价
+伦
+份
+华
+仰
+仿
+伙
+伪
+自
+伊
+血
+向
+似
+后
+行
+舟
+全
+会
+杀
+合
+兆
+企
+众
+爷
+伞
+创
+肌
+肋
+朵
+杂
+危
+旬
+旨
+旭
+负
+匈
+名
+各
+多
+争
+色
+壮
+冲
+妆
+冰
+庄
+庆
+亦
+刘
+齐
+交
+衣
+次
+产
+决
+亥
+充
+妄
+闭
+问
+闯
+羊
+并
+关
+米
+灯
+州
+汗
+污
+江
+汛
+池
+汝
+汤
+忙
+兴
+宇
+守
+宅
+字
+安
+讲
+讳
+军
+讶
+许
+讹
+论
+讼
+农
+讽
+设
+访
+诀
+寻
+那
+迅
+尽
+导
+异
+弛
+孙
+阵
+阳
+收
+阶
+阴
+防
+奸
+如
+妇
+妃
+好
+她
+妈
+戏
+羽
+观
+欢
+买
+红
+驮
+纤
+驯
+约
+级
+纪
+驰
+纫
+巡
+寿
+弄
+麦
+玖
+玛
+形
+进
+戒
+吞
+远
+违
+韧
+运
+扶
+抚
+坛
+技
+坏
+抠
+扰
+扼
+拒
+找
+批
+址
+扯
+走
+抄
+贡
+汞
+坝
+攻
+赤
+折
+抓
+扳
+抡
+扮
+抢
+孝
+坎
+均
+抑
+抛
+投
+坟
+坑
+抗
+坊
+抖
+护
+壳
+志
+块
+扭
+声
+把
+报
+拟
+却
+抒
+劫
+芙
+芜
+苇
+芽
+花
+芹
+芥
+芬
+苍
+芳
+严
+芦
+芯
+劳
+克
+芭
+苏
+杆
+杠
+杜
+材
+村
+杖
+杏
+杉
+巫
+极
+李
+杨
+求
+甫
+匣
+更
+束
+吾
+豆
+两
+酉
+丽
+医
+辰
+励
+否
+还
+尬
+歼
+来
+连
+轩
+步
+卤
+坚
+肖
+旱
+盯
+呈
+时
+吴
+助
+县
+里
+呆
+吱
+吠
+呕
+园
+旷
+围
+呀
+吨
+足
+邮
+男
+困
+吵
+串
+员
+呐
+听
+吟
+吩
+呛
+吻
+吹
+呜
+吭
+吧
+邑
+吼
+囤
+别
+吮
+岖
+岗
+帐
+财
+针
+钉
+牡
+告
+我
+乱
+利
+秃
+秀
+私
+每
+兵
+估
+体
+何
+佐
+佑
+但
+伸
+佃
+作
+伯
+伶
+佣
+低
+你
+住
+位
+伴
+身
+皂
+伺
+佛
+囱
+近
+彻
+役
+返
+余
+希
+坐
+谷
+妥
+含
+邻
+岔
+肝
+肛
+肚
+肘
+肠
+龟
+甸
+免
+狂
+犹
+狈
+角
+删
+条
+彤
+卵
+灸
+岛
+刨
+迎
+饭
+饮
+系
+言
+冻
+状
+亩
+况
+床
+库
+庇
+疗
+吝
+应
+这
+冷
+庐
+序
+辛
+弃
+冶
+忘
+闰
+闲
+间
+闷
+判
+兑
+灶
+灿
+灼
+弟
+汪
+沐
+沛
+汰
+沥
+沙
+汽
+沃
+沦
+汹
+泛
+沧
+没
+沟
+沪
+沈
+沉
+沁
+怀
+忧
+忱
+快
+完
+宋
+宏
+牢
+究
+穷
+灾
+良
+证
+启
+评
+补
+初
+社
+祀
+识
+诈
+诉
+罕
+诊
+词
+译
+君
+灵
+即
+层
+屁
+尿
+尾
+迟
+局
+改
+张
+忌
+际
+陆
+阿
+陈
+阻
+附
+坠
+妓
+妙
+妖
+姊
+妨
+妒
+努
+忍
+劲
+矣
+鸡
+纬
+驱
+纯
+纱
+纲
+纳
+驳
+纵
+纷
+纸
+纹
+纺
+驴
+纽
+奉
+玩
+环
+武
+青
+责
+现
+玫
+表
+规
+抹
+卦
+坷
+坯
+拓
+拢
+拔
+坪
+拣
+坦
+担
+坤
+押
+抽
+拐
+拖
+者
+拍
+顶
+拆
+拎
+拥
+抵
+拘
+势
+抱
+拄
+垃
+拉
+拦
+幸
+拌
+拧
+拂
+拙
+招
+坡
+披
+拨
+择
+抬
+拇
+拗
+其
+取
+茉
+苦
+昔
+苛
+若
+茂
+苹
+苗
+英
+苟
+苑
+苞
+范
+直
+茁
+茄
+茎
+苔
+茅
+枉
+林
+枝
+杯
+枢
+柜
+枚
+析
+板
+松
+枪
+枫
+构
+杭
+杰
+述
+枕
+丧
+或
+画
+卧
+事
+刺
+枣
+雨
+卖
+郁
+矾
+矿
+码
+厕
+奈
+奔
+奇
+奋
+态
+欧
+殴
+垄
+妻
+轰
+顷
+转
+斩
+轮
+软
+到
+非
+叔
+歧
+肯
+齿
+些
+卓
+虎
+虏
+肾
+贤
+尚
+旺
+具
+味
+果
+昆
+国
+哎
+咕
+昌
+呵
+畅
+明
+易
+咙
+昂
+迪
+典
+固
+忠
+呻
+咒
+咋
+咐
+呼
+鸣
+咏
+呢
+咄
+咖
+岸
+岩
+帖
+罗
+帜
+帕
+岭
+凯
+败
+账
+贩
+贬
+购
+贮
+图
+钓
+制
+知
+迭
+氛
+垂
+牧
+物
+乖
+刮
+秆
+和
+季
+委
+秉
+佳
+侍
+岳
+供
+使
+例
+侠
+侥
+版
+侄
+侦
+侣
+侧
+凭
+侨
+佩
+货
+侈
+依
+卑
+的
+迫
+质
+欣
+征
+往
+爬
+彼
+径
+所
+舍
+金
+刹
+命
+肴
+斧
+爸
+采
+觅
+受
+乳
+贪
+念
+贫
+忿
+肤
+肺
+肢
+肿
+胀
+朋
+股
+肮
+肪
+肥
+服
+胁
+周
+昏
+鱼
+兔
+狐
+忽
+狗
+狞
+备
+饰
+饱
+饲
+变
+京
+享
+庞
+店
+夜
+庙
+府
+底
+疟
+疙
+疚
+剂
+卒
+郊
+庚
+废
+净
+盲
+放
+刻
+育
+氓
+闸
+闹
+郑
+券
+卷
+单
+炬
+炒
+炊
+炕
+炎
+炉
+沫
+浅
+法
+泄
+沽
+河
+沾
+泪
+沮
+油
+泊
+沿
+泡
+注
+泣
+泞
+泻
+泌
+泳
+泥
+沸
+沼
+波
+泼
+泽
+治
+怔
+怯
+怖
+性
+怕
+怜
+怪
+怡
+学
+宝
+宗
+定
+宠
+宜
+审
+宙
+官
+空
+帘
+宛
+实
+试
+郎
+诗
+肩
+房
+诚
+衬
+衫
+视
+祈
+话
+诞
+诡
+询
+该
+详
+建
+肃
+录
+隶
+帚
+屉
+居
+届
+刷
+屈
+弧
+弥
+弦
+承
+孟
+陋
+陌
+孤
+陕
+降
+函
+限
+妹
+姑
+姐
+姓
+妮
+始
+姆
+迢
+驾
+叁
+参
+艰
+线
+练
+组
+绅
+细
+驶
+织
+驹
+终
+驻
+绊
+驼
+绍
+绎
+经
+贯
+契
+贰
+奏
+春
+帮
+玷
+珍
+玲
+玻
+毒
+型
+拭
+挂
+封
+持
+拷
+拱
+项
+垮
+挎
+城
+挟
+挠
+政
+赴
+赵
+挡
+拽
+哉
+挺
+括
+垢
+拴
+拾
+挑
+垛
+指
+垫
+挣
+挤
+拼
+挖
+按
+挥
+挪
+拯
+某
+甚
+荆
+茸
+革
+茬
+荐
+巷
+带
+草
+茧
+茵
+茶
+荒
+茫
+荡
+荣
+荤
+荧
+故
+胡
+荫
+荔
+南
+药
+标
+栈
+柑
+枯
+柄
+栋
+相
+查
+柏
+栅
+柳
+柱
+柿
+栏
+柠
+树
+勃
+要
+柬
+咸
+威
+歪
+研
+砖
+厘
+厚
+砌
+砂
+泵
+砚
+砍
+面
+耐
+耍
+牵
+鸥
+残
+殃
+轴
+轻
+鸦
+皆
+韭
+背
+战
+点
+虐
+临
+览
+竖
+省
+削
+尝
+昧
+盹
+是
+盼
+眨
+哇
+哄
+哑
+显
+冒
+映
+星
+昨
+咧
+昭
+畏
+趴
+胃
+贵
+界
+虹
+虾
+蚁
+思
+蚂
+虽
+品
+咽
+骂
+勋
+哗
+咱
+响
+哈
+哆
+咬
+咳
+咪
+哪
+哟
+炭
+峡
+罚
+贱
+贴
+贻
+骨
+幽
+钙
+钝
+钞
+钟
+钢
+钠
+钥
+钦
+钧
+钩
+钮
+卸
+缸
+拜
+看
+矩
+毡
+氢
+怎
+牲
+选
+适
+秒
+香
+种
+秋
+科
+重
+复
+竿
+段
+便
+俩
+贷
+顺
+修
+俏
+保
+促
+俄
+俐
+侮
+俭
+俗
+俘
+信
+皇
+泉
+鬼
+侵
+禹
+侯
+追
+俊
+盾
+待
+徊
+衍
+律
+很
+须
+叙
+剑
+逃
+食
+盆
+胚
+胧
+胆
+胜
+胞
+胖
+脉
+胎
+勉
+狭
+狮
+独
+狰
+狡
+狱
+狠
+贸
+怨
+急
+饵
+饶
+蚀
+饺
+饼
+峦
+弯
+将
+奖
+哀
+亭
+亮
+度
+迹
+庭
+疮
+疯
+疫
+疤
+咨
+姿
+亲
+音
+帝
+施
+闺
+闻
+闽
+阀
+阁
+差
+养
+美
+姜
+叛
+送
+类
+迷
+籽
+娄
+前
+首
+逆
+兹
+总
+炼
+炸
+烁
+炮
+炫
+烂
+剃
+洼
+洁
+洪
+洒
+柒
+浇
+浊
+洞
+测
+洗
+活
+派
+洽
+染
+洛
+浏
+济
+洋
+洲
+浑
+浓
+津
+恃
+恒
+恢
+恍
+恬
+恤
+恰
+恼
+恨
+举
+觉
+宣
+宦
+室
+宫
+宪
+突
+穿
+窃
+客
+诫
+冠
+诬
+语
+扁
+袄
+祖
+神
+祝
+祠
+误
+诱
+诲
+说
+诵
+垦
+退
+既
+屋
+昼
+屏
+屎
+费
+陡
+逊
+眉
+孩
+陨
+除
+险
+院
+娃
+姥
+姨
+姻
+娇
+姚
+娜
+怒
+架
+贺
+盈
+勇
+怠
+癸
+蚤
+柔
+垒
+绑
+绒
+结
+绕
+骄
+绘
+给
+绚
+骆
+络
+绝
+绞
+骇
+统
+耕
+耘
+耗
+耙
+艳
+泰
+秦
+珠
+班
+素
+匿
+蚕
+顽
+盏
+匪
+捞
+栽
+捕
+埂
+捂
+振
+载
+赶
+起
+盐
+捎
+捍
+捏
+埋
+捉
+捆
+捐
+损
+袁
+捌
+都
+哲
+逝
+捡
+挫
+换
+挽
+挚
+热
+恐
+捣
+壶
+捅
+埃
+挨
+耻
+耿
+耽
+聂
+恭
+莽
+莱
+莲
+莫
+莉
+荷
+获
+晋
+恶
+莹
+莺
+真
+框
+梆
+桂
+桔
+栖
+档
+桐
+株
+桥
+桦
+栓
+桃
+格
+桩
+校
+核
+样
+根
+索
+哥
+速
+逗
+栗
+贾
+酌
+配
+翅
+辱
+唇
+夏
+砸
+砰
+砾
+础
+破
+原
+套
+逐
+烈
+殊
+殉
+顾
+轿
+较
+顿
+毙
+致
+柴
+桌
+虑
+监
+紧
+党
+逞
+晒
+眠
+晓
+哮
+唠
+鸭
+晃
+哺
+晌
+剔
+晕
+蚌
+畔
+蚣
+蚊
+蚪
+蚓
+哨
+哩
+圃
+哭
+哦
+恩
+鸯
+唤
+唁
+哼
+唧
+啊
+唉
+唆
+罢
+峭
+峨
+峰
+圆
+峻
+贼
+贿
+赂
+赃
+钱
+钳
+钻
+钾
+铁
+铃
+铅
+缺
+氧
+氨
+特
+牺
+造
+乘
+敌
+秤
+租
+积
+秧
+秩
+称
+秘
+透
+笔
+笑
+笋
+债
+借
+值
+倚
+俺
+倾
+倒
+倘
+俱
+倡
+候
+赁
+俯
+倍
+倦
+健
+臭
+射
+躬
+息
+倔
+徒
+徐
+殷
+舰
+舱
+般
+航
+途
+拿
+耸
+爹
+舀
+爱
+豺
+豹
+颁
+颂
+翁
+胰
+脆
+脂
+胸
+胳
+脏
+脐
+胶
+脑
+脓
+逛
+狸
+狼
+卿
+逢
+鸵
+留
+鸳
+皱
+饿
+馁
+凌
+凄
+恋
+桨
+浆
+衰
+衷
+高
+郭
+席
+准
+座
+症
+病
+疾
+斋
+疹
+疼
+疲
+脊
+效
+离
+紊
+唐
+瓷
+资
+凉
+站
+剖
+竞
+部
+旁
+旅
+畜
+阅
+羞
+羔
+瓶
+拳
+粉
+料
+益
+兼
+烤
+烘
+烦
+烧
+烛
+烟
+烙
+递
+涛
+浙
+涝
+浦
+酒
+涉
+消
+涡
+浩
+海
+涂
+浴
+浮
+涣
+涤
+流
+润
+涧
+涕
+浪
+浸
+涨
+烫
+涩
+涌
+悖
+悟
+悄
+悍
+悔
+悯
+悦
+害
+宽
+家
+宵
+宴
+宾
+窍
+窄
+容
+宰
+案
+请
+朗
+诸
+诺
+读
+扇
+诽
+袜
+袖
+袍
+被
+祥
+课
+冥
+谁
+调
+冤
+谅
+谆
+谈
+谊
+剥
+恳
+展
+剧
+屑
+弱
+陵
+祟
+陶
+陷
+陪
+娱
+娟
+恕
+娥
+娘
+通
+能
+难
+预
+桑
+绢
+绣
+验
+继
+骏
+球
+琐
+理
+琉
+琅
+捧
+堵
+措
+描
+域
+捺
+掩
+捷
+排
+焉
+掉
+捶
+赦
+堆
+推
+埠
+掀
+授
+捻
+教
+掏
+掐
+掠
+掂
+培
+接
+掷
+控
+探
+据
+掘
+掺
+职
+基
+聆
+勘
+聊
+娶
+著
+菱
+勒
+黄
+菲
+萌
+萝
+菌
+萎
+菜
+萄
+菊
+菩
+萍
+菠
+萤
+营
+乾
+萧
+萨
+菇
+械
+彬
+梦
+婪
+梗
+梧
+梢
+梅
+检
+梳
+梯
+桶
+梭
+救
+曹
+副
+票
+酝
+酗
+厢
+戚
+硅
+硕
+奢
+盔
+爽
+聋
+袭
+盛
+匾
+雪
+辅
+辆
+颅
+虚
+彪
+雀
+堂
+常
+眶
+匙
+晨
+睁
+眯
+眼
+悬
+野
+啪
+啦
+曼
+晦
+晚
+啄
+啡
+距
+趾
+啃
+跃
+略
+蚯
+蛀
+蛇
+唬
+累
+鄂
+唱
+患
+啰
+唾
+唯
+啤
+啥
+啸
+崖
+崎
+崭
+逻
+崔
+帷
+崩
+崇
+崛
+婴
+圈
+铐
+铛
+铝
+铜
+铭
+铲
+银
+矫
+甜
+秸
+梨
+犁
+秽
+移
+笨
+笼
+笛
+笙
+符
+第
+敏
+做
+袋
+悠
+偿
+偶
+偎
+偷
+您
+售
+停
+偏
+躯
+兜
+假
+衅
+徘
+徙
+得
+衔
+盘
+舶
+船
+舵
+斜
+盒
+鸽
+敛
+悉
+欲
+彩
+领
+脚
+脖
+脯
+豚
+脸
+脱
+象
+够
+逸
+猜
+猪
+猎
+猫
+凰
+猖
+猛
+祭
+馅
+馆
+凑
+减
+毫
+烹
+庶
+麻
+庵
+痊
+痒
+痕
+廊
+康
+庸
+鹿
+盗
+章
+竟
+商
+族
+旋
+望
+率
+阎
+阐
+着
+羚
+盖
+眷
+粘
+粗
+粒
+断
+剪
+兽
+焊
+焕
+清
+添
+鸿
+淋
+涯
+淹
+渠
+渐
+淑
+淌
+混
+淮
+淆
+渊
+淫
+渔
+淘
+淳
+液
+淤
+淡
+淀
+深
+涮
+涵
+婆
+梁
+渗
+情
+惜
+惭
+悼
+惧
+惕
+惟
+惊
+惦
+悴
+惋
+惨
+惯
+寇
+寅
+寄
+寂
+宿
+窒
+窑
+密
+谋
+谍
+谎
+谐
+袱
+祷
+祸
+谓
+谚
+谜
+逮
+敢
+尉
+屠
+弹
+隋
+堕
+随
+蛋
+隅
+隆
+隐
+婚
+婶
+婉
+颇
+颈
+绩
+绪
+续
+骑
+绰
+绳
+维
+绵
+绷
+绸
+综
+绽
+绿
+缀
+巢
+琴
+琳
+琢
+琼
+斑
+替
+揍
+款
+堪
+塔
+搭
+堰
+揩
+越
+趁
+趋
+超
+揽
+堤
+提
+博
+揭
+喜
+彭
+揣
+插
+揪
+搜
+煮
+援
+搀
+裁
+搁
+搓
+搂
+搅
+壹
+握
+搔
+揉
+斯
+期
+欺
+联
+葫
+散
+惹
+葬
+募
+葛
+董
+葡
+敬
+葱
+蒋
+蒂
+落
+韩
+朝
+辜
+葵
+棒
+棱
+棋
+椰
+植
+森
+焚
+椅
+椒
+棵
+棍
+椎
+棉
+棚
+棕
+棺
+榔
+椭
+惠
+惑
+逼
+粟
+棘
+酣
+酥
+厨
+厦
+硬
+硝
+确
+硫
+雁
+殖
+裂
+雄
+颊
+雳
+暂
+雅
+翘
+辈
+悲
+紫
+凿
+辉
+敞
+棠
+赏
+掌
+晴
+睐
+暑
+最
+晰
+量
+鼎
+喷
+喳
+晶
+喇
+遇
+喊
+遏
+晾
+景
+畴
+践
+跋
+跌
+跑
+跛
+遗
+蛙
+蛛
+蜓
+蜒
+蛤
+喝
+鹃
+喂
+喘
+喉
+喻
+啼
+喧
+嵌
+幅
+帽
+赋
+赌
+赎
+赐
+赔
+黑
+铸
+铺
+链
+销
+锁
+锄
+锅
+锈
+锋
+锌
+锐
+甥
+掰
+短
+智
+氮
+毯
+氯
+鹅
+剩
+稍
+程
+稀
+税
+筐
+等
+筑
+策
+筛
+筒
+筏
+答
+筋
+筝
+傲
+傅
+牌
+堡
+集
+焦
+傍
+储
+皓
+皖
+粤
+奥
+街
+惩
+御
+循
+艇
+舒
+逾
+番
+释
+禽
+腊
+脾
+腋
+腔
+腕
+鲁
+猩
+猬
+猾
+猴
+惫
+然
+馈
+馋
+装
+蛮
+就
+敦
+斌
+痘
+痢
+痪
+痛
+童
+竣
+阔
+善
+翔
+羡
+普
+粪
+尊
+奠
+道
+遂
+曾
+焰
+港
+滞
+湖
+湘
+渣
+渤
+渺
+湿
+温
+渴
+溃
+溅
+滑
+湃
+渝
+湾
+渡
+游
+滋
+渲
+溉
+愤
+慌
+惰
+愕
+愣
+惶
+愧
+愉
+慨
+割
+寒
+富
+寓
+窜
+窝
+窖
+窗
+窘
+遍
+雇
+裕
+裤
+裙
+禅
+禄
+谢
+谣
+谤
+谦
+犀
+属
+屡
+强
+粥
+疏
+隔
+隙
+隘
+媒
+絮
+嫂
+媚
+婿
+登
+缅
+缆
+缉
+缎
+缓
+缔
+缕
+骗
+编
+骚
+缘
+瑟
+鹉
+瑞
+瑰
+瑙
+魂
+肆
+摄
+摸
+填
+搏
+塌
+鼓
+摆
+携
+搬
+摇
+搞
+塘
+摊
+聘
+斟
+蒜
+勤
+靴
+靶
+鹊
+蓝
+墓
+幕
+蓬
+蓄
+蒲
+蓉
+蒙
+蒸
+献
+椿
+禁
+楚
+楷
+榄
+想
+槐
+榆
+楼
+概
+赖
+酪
+酬
+感
+碍
+碘
+碑
+碎
+碰
+碗
+碌
+尴
+雷
+零
+雾
+雹
+辐
+辑
+输
+督
+频
+龄
+鉴
+睛
+睹
+睦
+瞄
+睫
+睡
+睬
+嗜
+鄙
+嗦
+愚
+暖
+盟
+歇
+暗
+暇
+照
+畸
+跨
+跷
+跳
+跺
+跪
+路
+跤
+跟
+遣
+蜈
+蜗
+蛾
+蜂
+蜕
+嗅
+嗡
+嗓
+署
+置
+罪
+罩
+蜀
+幌
+错
+锚
+锡
+锣
+锤
+锥
+锦
+键
+锯
+锰
+矮
+辞
+稚
+稠
+颓
+愁
+筹
+签
+简
+筷
+毁
+舅
+鼠
+催
+傻
+像
+躲
+魁
+衙
+微
+愈
+遥
+腻
+腰
+腥
+腮
+腹
+腺
+鹏
+腾
+腿
+鲍
+猿
+颖
+触
+解
+煞
+雏
+馍
+馏
+酱
+禀
+痹
+廓
+痴
+痰
+廉
+靖
+新
+韵
+意
+誊
+粮
+数
+煎
+塑
+慈
+煤
+煌
+满
+漠
+滇
+源
+滤
+滥
+滔
+溪
+溜
+漓
+滚
+溢
+溯
+滨
+溶
+溺
+粱
+滩
+慎
+誉
+塞
+寞
+窥
+窟
+寝
+谨
+褂
+裸
+福
+谬
+群
+殿
+辟
+障
+媳
+嫉
+嫌
+嫁
+叠
+缚
+缝
+缠
+缤
+剿
+静
+碧
+璃
+赘
+熬
+墙
+墟
+嘉
+摧
+赫
+截
+誓
+境
+摘
+摔
+撇
+聚
+慕
+暮
+摹
+蔓
+蔑
+蔡
+蔗
+蔽
+蔼
+熙
+蔚
+兢
+模
+槛
+榴
+榜
+榨
+榕
+歌
+遭
+酵
+酷
+酿
+酸
+碟
+碱
+碳
+磁
+愿
+需
+辖
+辗
+雌
+裳
+颗
+瞅
+墅
+嗽
+踊
+蜻
+蜡
+蝇
+蜘
+蝉
+嘛
+嘀
+赚
+锹
+锻
+镀
+舞
+舔
+稳
+熏
+箕
+算
+箩
+管
+箫
+舆
+僚
+僧
+鼻
+魄
+魅
+貌
+膜
+膊
+膀
+鲜
+疑
+孵
+馒
+裹
+敲
+豪
+膏
+遮
+腐
+瘩
+瘟
+瘦
+辣
+彰
+竭
+端
+旗
+精
+粹
+歉
+弊
+熄
+熔
+煽
+潇
+漆
+漱
+漂
+漫
+滴
+漾
+演
+漏
+慢
+慷
+寨
+赛
+寡
+察
+蜜
+寥
+谭
+肇
+褐
+褪
+谱
+隧
+嫩
+翠
+熊
+凳
+骡
+缩
+慧
+撵
+撕
+撒
+撩
+趣
+趟
+撑
+撮
+撬
+播
+擒
+墩
+撞
+撤
+增
+撰
+聪
+鞋
+鞍
+蕉
+蕊
+蔬
+蕴
+横
+槽
+樱
+橡
+樟
+橄
+敷
+豌
+飘
+醋
+醇
+醉
+磕
+磊
+磅
+碾
+震
+霄
+霉
+瞒
+题
+暴
+瞎
+嘻
+嘶
+嘲
+嘹
+影
+踢
+踏
+踩
+踪
+蝶
+蝴
+蝠
+蝎
+蝌
+蝗
+蝙
+嘿
+嘱
+幢
+墨
+镇
+镐
+镑
+靠
+稽
+稻
+黎
+稿
+稼
+箱
+篓
+箭
+篇
+僵
+躺
+僻
+德
+艘
+膝
+膛
+鲤
+鲫
+熟
+摩
+褒
+瘪
+瘤
+瘫
+凛
+颜
+毅
+糊
+遵
+憋
+潜
+澎
+潮
+潭
+鲨
+澳
+潘
+澈
+澜
+澄
+懂
+憔
+懊
+憎
+额
+翩
+褥
+谴
+鹤
+憨
+慰
+劈
+履
+豫
+缭
+撼
+擂
+操
+擅
+燕
+蕾
+薯
+薛
+薇
+擎
+薪
+薄
+颠
+翰
+噩
+橱
+橙
+橘
+整
+融
+瓢
+醒
+霍
+霎
+辙
+冀
+餐
+嘴
+踱
+蹄
+蹂
+蟆
+螃
+器
+噪
+鹦
+赠
+默
+黔
+镜
+赞
+穆
+篮
+篡
+篷
+篱
+儒
+邀
+衡
+膨
+雕
+鲸
+磨
+瘾
+瘸
+凝
+辨
+辩
+糙
+糖
+糕
+燃
+濒
+澡
+激
+懒
+憾
+懈
+窿
+壁
+避
+缰
+缴
+戴
+擦
+藉
+鞠
+藏
+藐
+檬
+檐
+檀
+礁
+磷
+霜
+霞
+瞭
+瞧
+瞬
+瞳
+瞩
+瞪
+曙
+蹋
+蹈
+螺
+蟋
+蟀
+嚎
+赡
+穗
+魏
+簧
+簇
+繁
+徽
+爵
+朦
+臊
+鳄
+癌
+辫
+赢
+糟
+糠
+燥
+懦
+豁
+臀
+臂
+翼
+骤
+藕
+鞭
+藤
+覆
+瞻
+蹦
+嚣
+镰
+翻
+鳍
+鹰
+瀑
+襟
+璧
+戳
+孽
+警
+蘑
+藻
+攀
+曝
+蹲
+蹭
+蹬
+巅
+簸
+簿
+蟹
+颤
+靡
+癣
+瓣
+羹
+鳖
+爆
+疆
+鬓
+壤
+馨
+耀
+躁
+蠕
+嚼
+嚷
+巍
+籍
+鳞
+魔
+糯
+灌
+譬
+蠢
+霸
+露
+霹
+躏
+黯
+髓
+赣
+囊
+镶
+瓤
+罐
+矗
+乂
+乜
+兀
+弋
+孑
+孓
+幺
+亓
+韦
+廿
+丏
+卅
+仄
+厄
+仃
+仉
+仂
+兮
+刈
+爻
+卞
+闩
+讣
+尹
+夬
+爿
+毋
+邗
+邛
+艽
+艿
+札
+叵
+匝
+丕
+匜
+劢
+卟
+叱
+叻
+仨
+仕
+仟
+仡
+仫
+仞
+卮
+氐
+犰
+刍
+邝
+邙
+汀
+讦
+讧
+讪
+讫
+尻
+阡
+尕
+弁
+驭
+匡
+耒
+玎
+玑
+邢
+圩
+圬
+圭
+扦
+圪
+圳
+圹
+扪
+圮
+圯
+芊
+芍
+芄
+芨
+芑
+芎
+芗
+亘
+厍
+夼
+戍
+尥
+乩
+旯
+曳
+岌
+屺
+凼
+囡
+钇
+缶
+氘
+氖
+牝
+伎
+伛
+伢
+佤
+仵
+伥
+伧
+伉
+伫
+囟
+汆
+刖
+夙
+旮
+刎
+犷
+犸
+舛
+凫
+邬
+饧
+汕
+汔
+汐
+汲
+汜
+汊
+忖
+忏
+讴
+讵
+祁
+讷
+聿
+艮
+厾
+阱
+阮
+阪
+丞
+妁
+牟
+纡
+纣
+纥
+纨
+玕
+玙
+抟
+抔
+圻
+坂
+坍
+坞
+抃
+抉
+㧐
+芫
+邯
+芸
+芾
+苈
+苣
+芷
+芮
+苋
+芼
+苌
+苁
+芩
+芪
+芡
+芟
+苄
+苎
+苡
+杌
+杓
+杞
+杈
+忑
+孛
+邴
+邳
+矶
+奁
+豕
+忒
+欤
+轫
+迓
+邶
+忐
+卣
+邺
+旰
+呋
+呒
+呓
+呔
+呖
+呃
+旸
+吡
+町
+虬
+呗
+吽
+吣
+吲
+帏
+岐
+岈
+岘
+岑
+岚
+兕
+囵
+囫
+钊
+钋
+钌
+迕
+氙
+氚
+牤
+佞
+邱
+攸
+佚
+佝
+佟
+佗
+伽
+彷
+佘
+佥
+孚
+豸
+坌
+肟
+邸
+奂
+劬
+狄
+狁
+鸠
+邹
+饨
+饩
+饪
+饫
+饬
+亨
+庑
+庋
+疔
+疖
+肓
+闱
+闳
+闵
+羌
+炀
+沣
+沅
+沔
+沤
+沌
+沏
+沚
+汩
+汨
+沂
+汾
+沨
+汴
+汶
+沆
+沩
+泐
+怃
+怄
+忡
+忤
+忾
+怅
+忻
+忪
+怆
+忭
+忸
+诂
+诃
+诅
+诋
+诌
+诏
+诒
+孜
+陇
+陀
+陂
+陉
+妍
+妩
+妪
+妣
+妊
+妗
+妫
+妞
+姒
+妤
+邵
+劭
+刭
+甬
+邰
+纭
+纰
+纴
+纶
+纾
+玮
+玡
+玭
+玠
+玢
+玥
+玦
+盂
+忝
+匦
+坩
+抨
+拤
+坫
+拈
+垆
+抻
+劼
+拃
+拊
+坼
+坻
+㧟
+坨
+坭
+抿
+坳
+耶
+苷
+苯
+苤
+茏
+苫
+苜
+苴
+苒
+苘
+茌
+苻
+苓
+茚
+茆
+茑
+茓
+茔
+茕
+茀
+苕
+枥
+枇
+杪
+杳
+枧
+杵
+枨
+枞
+枋
+杻
+杷
+杼
+矸
+砀
+刳
+奄
+瓯
+殁
+郏
+轭
+郅
+鸢
+盱
+昊
+昙
+杲
+昃
+咂
+呸
+昕
+昀
+旻
+昉
+炅
+咔
+畀
+虮
+咀
+呷
+黾
+呱
+呤
+咚
+咆
+咛
+呶
+呣
+呦
+咝
+岢
+岿
+岬
+岫
+帙
+岣
+峁
+刿
+迥
+岷
+剀
+帔
+峄
+沓
+囹
+罔
+钍
+钎
+钏
+钒
+钕
+钗
+邾
+迮
+牦
+竺
+迤
+佶
+佬
+佰
+侑
+侉
+臾
+岱
+侗
+侃
+侏
+侩
+佻
+佾
+侪
+佼
+佯
+侬
+帛
+阜
+侔
+徂
+刽
+郄
+怂
+籴
+瓮
+戗
+肼
+䏝
+肽
+肱
+肫
+剁
+迩
+郇
+狙
+狎
+狍
+狒
+咎
+炙
+枭
+饯
+饴
+冽
+冼
+庖
+疠
+疝
+疡
+兖
+妾
+劾
+炜
+𬉼
+炖
+炘
+炝
+炔
+泔
+沭
+泷
+泸
+泱
+泅
+泗
+泠
+泺
+泖
+泫
+泮
+沱
+泯
+泓
+泾
+怙
+怵
+怦
+怛
+怏
+怍
+㤘
+怩
+怫
+怿
+宕
+穹
+宓
+诓
+诔
+诖
+诘
+戾
+诙
+戽
+郓
+衩
+祆
+祎
+祉
+祇
+诛
+诜
+诟
+诠
+诣
+诤
+诧
+诨
+诩
+戕
+孢
+亟
+陔
+妲
+妯
+姗
+帑
+弩
+孥
+驽
+虱
+迦
+迨
+绀
+绁
+绂
+驷
+驸
+绉
+绌
+驿
+骀
+甾
+珏
+珐
+珂
+珑
+玳
+珀
+顸
+珉
+珈
+拮
+垭
+挝
+垣
+挞
+垤
+赳
+贲
+垱
+垌
+郝
+垧
+垓
+挦
+垠
+茜
+荚
+荑
+贳
+荜
+莒
+茼
+茴
+茱
+莛
+荞
+茯
+荏
+荇
+荃
+荟
+荀
+茗
+荠
+茭
+茨
+垩
+荥
+荦
+荨
+荩
+剋
+荪
+茹
+荬
+荮
+柰
+栉
+柯
+柘
+栊
+柩
+枰
+栌
+柙
+枵
+柚
+枳
+柞
+柝
+栀
+柢
+栎
+枸
+柈
+柁
+枷
+柽
+剌
+酊
+郦
+甭
+砗
+砘
+砒
+斫
+砭
+砜
+奎
+耷
+虺
+殂
+殇
+殄
+殆
+轱
+轲
+轳
+轶
+轸
+虿
+毖
+觇
+尜
+哐
+眄
+眍
+𠳐
+郢
+眇
+眊
+眈
+禺
+哂
+咴
+曷
+昴
+昱
+昵
+咦
+哓
+哔
+畎
+毗
+呲
+胄
+畋
+畈
+虼
+虻
+盅
+咣
+哕
+剐
+郧
+咻
+囿
+咿
+哌
+哙
+哚
+咯
+咩
+咤
+哝
+哏
+哞
+峙
+峣
+罘
+帧
+峒
+峤
+峋
+峥
+贶
+钚
+钛
+钡
+钣
+钤
+钨
+钫
+钯
+氡
+氟
+牯
+郜
+秕
+秭
+竽
+笈
+笃
+俦
+俨
+俅
+俪
+叟
+垡
+牮
+俣
+俚
+皈
+俑
+俟
+逅
+徇
+徉
+舢
+俞
+郗
+俎
+郤
+爰
+郛
+瓴
+胨
+胪
+胛
+胂
+胙
+胍
+胗
+胝
+朐
+胫
+鸨
+匍
+狨
+狯
+飑
+狩
+狲
+訇
+逄
+昝
+饷
+饸
+饹
+胤
+孪
+娈
+弈
+奕
+庥
+疬
+疣
+疥
+疭
+庠
+竑
+彦
+飒
+闼
+闾
+闿
+阂
+羑
+迸
+籼
+酋
+炳
+炻
+炽
+炯
+烀
+炷
+烃
+洱
+洹
+洧
+洌
+浃
+洇
+洄
+洙
+涎
+洎
+洫
+浍
+洮
+洵
+浒
+浔
+浕
+洳
+恸
+恓
+恹
+恫
+恺
+恻
+恂
+恪
+恽
+宥
+扃
+衲
+衽
+衿
+袂
+祛
+祜
+祓
+祚
+诮
+祗
+祢
+诰
+诳
+鸩
+昶
+郡
+咫
+弭
+牁
+胥
+陛
+陟
+娅
+姮
+娆
+姝
+姣
+姘
+姹
+怼
+羿
+炱
+矜
+绔
+骁
+骅
+绗
+绛
+骈
+耖
+挈
+珥
+珙
+顼
+珰
+珩
+珧
+珣
+珞
+琤
+珲
+敖
+恚
+埔
+埕
+埘
+埙
+埚
+挹
+耆
+耄
+埒
+捋
+贽
+垸
+捃
+盍
+荸
+莆
+莳
+莴
+莪
+莠
+莓
+莜
+莅
+荼
+莩
+荽
+莸
+荻
+莘
+莎
+莞
+莨
+渇
+鸪
+莼
+栲
+栳
+郴
+桓
+桡
+桎
+桢
+桤
+梃
+栝
+桕
+桁
+桧
+桅
+栟
+桉
+栩
+逑
+逋
+彧
+鬲
+豇
+酐
+逦
+厝
+孬
+砝
+砹
+砺
+砧
+砷
+砟
+砼
+砥
+砣
+剞
+砻
+轼
+轾
+辂
+鸫
+趸
+龀
+鸬
+虔
+逍
+眬
+唛
+晟
+眩
+眙
+哧
+哽
+唔
+晁
+晏
+鸮
+趵
+趿
+畛
+蚨
+蚜
+蚍
+蚋
+蚬
+蚝
+蚧
+唢
+圄
+唣
+唏
+盎
+唑
+崂
+崃
+罡
+罟
+峪
+觊
+赅
+钰
+钲
+钴
+钵
+钹
+钺
+钽
+钼
+钿
+铀
+铂
+铄
+铆
+铈
+铉
+铊
+铋
+铌
+铍
+䥽
+铎
+氩
+氤
+氦
+毪
+舐
+秣
+秫
+盉
+笄
+笕
+笊
+笏
+笆
+俸
+倩
+俵
+偌
+俳
+俶
+倬
+倏
+恁
+倭
+倪
+俾
+倜
+隼
+隽
+倌
+倥
+臬
+皋
+郫
+倨
+衄
+颀
+徕
+舫
+釜
+奚
+衾
+胯
+胱
+胴
+胭
+脍
+胼
+朕
+脒
+胺
+鸱
+玺
+鸲
+狷
+猁
+狳
+猃
+狺
+逖
+桀
+袅
+饽
+凇
+栾
+挛
+亳
+疳
+疴
+疸
+疽
+痈
+疱
+痂
+痉
+衮
+凋
+颃
+恣
+旆
+旄
+旃
+阃
+阄
+訚
+阆
+恙
+粑
+朔
+郸
+烜
+烨
+烩
+烊
+剡
+郯
+烬
+涑
+浯
+涞
+涟
+娑
+涅
+涠
+浞
+涓
+浥
+涔
+浜
+浠
+浣
+浚
+悚
+悭
+悝
+悒
+悌
+悛
+宸
+窈
+剜
+诹
+冢
+诼
+袒
+袢
+祯
+诿
+谀
+谂
+谄
+谇
+屐
+屙
+陬
+勐
+奘
+牂
+蚩
+陲
+姬
+娠
+娌
+娉
+娲
+娩
+娴
+娣
+娓
+婀
+畚
+逡
+绠
+骊
+绡
+骋
+绥
+绦
+绨
+骎
+邕
+鸶
+彗
+耜
+焘
+舂
+琏
+琇
+麸
+揶
+埴
+埯
+捯
+掳
+掴
+埸
+埵
+赧
+埤
+捭
+逵
+埝
+堋
+堍
+掬
+鸷
+掖
+捽
+掊
+堉
+掸
+捩
+掮
+悫
+埭
+埽
+掇
+掼
+聃
+菁
+萁
+菘
+堇
+萘
+萋
+菽
+菖
+萜
+萸
+萑
+棻
+菔
+菟
+萏
+萃
+菏
+菹
+菪
+菅
+菀
+萦
+菰
+菡
+梵
+梿
+梏
+觋
+桴
+桷
+梓
+棁
+桫
+棂
+啬
+郾
+匮
+敕
+豉
+鄄
+酞
+酚
+戛
+硎
+硭
+硒
+硖
+硗
+硐
+硇
+硌
+鸸
+瓠
+匏
+厩
+龚
+殒
+殓
+殍
+赉
+雩
+辄
+堑
+眭
+眦
+啧
+晡
+晤
+眺
+眵
+眸
+圊
+喏
+喵
+啉
+勖
+晞
+唵
+晗
+冕
+啭
+畦
+趺
+啮
+跄
+蚶
+蛄
+蛎
+蛆
+蚰
+蛊
+圉
+蚱
+蛉
+蛏
+蚴
+啁
+啕
+唿
+啐
+唼
+唷
+啖
+啵
+啶
+啷
+唳
+唰
+啜
+帻
+崚
+崦
+帼
+崮
+崤
+崆
+赇
+赈
+赊
+铑
+铒
+铗
+铙
+铟
+铠
+铡
+铢
+铣
+铤
+铧
+铨
+铩
+铪
+铫
+铬
+铮
+铯
+铰
+铱
+铳
+铵
+铷
+氪
+牾
+鸹
+秾
+逶
+笺
+筇
+笸
+笪
+笮
+笠
+笥
+笤
+笳
+笾
+笞
+偾
+偃
+偕
+偈
+傀
+偬
+偻
+皑
+皎
+鸻
+徜
+舸
+舻
+舴
+舷
+龛
+翎
+脬
+脘
+脲
+匐
+猗
+猡
+猞
+猝
+斛
+猕
+馗
+馃
+馄
+鸾
+孰
+庹
+庾
+痔
+痍
+疵
+翊
+旌
+旎
+袤
+阇
+阈
+阉
+阊
+阋
+阍
+阏
+羟
+粝
+粕
+敝
+焐
+烯
+焓
+烽
+焖
+烷
+焗
+渍
+渚
+淇
+淅
+淞
+渎
+涿
+淖
+挲
+淠
+涸
+渑
+淦
+淝
+淬
+涪
+淙
+涫
+渌
+淄
+惬
+悻
+悱
+惝
+惘
+悸
+惆
+惚
+惇
+惮
+窕
+谌
+谏
+扈
+皲
+谑
+裆
+袷
+裉
+谒
+谔
+谕
+谖
+谗
+谙
+谛
+谝
+逯
+郿
+隈
+粜
+隍
+隗
+婧
+婊
+婕
+娼
+婢
+婵
+胬
+袈
+翌
+恿
+欸
+绫
+骐
+绮
+绯
+绱
+骒
+绲
+骓
+绶
+绺
+绻
+绾
+骖
+缁
+耠
+琫
+琵
+琶
+琪
+瑛
+琦
+琥
+琨
+靓
+琰
+琮
+琯
+琬
+琛
+琚
+辇
+鼋
+揳
+堞
+搽
+揸
+揠
+堙
+趄
+揖
+颉
+塄
+揿
+耋
+揄
+蛩
+蛰
+塆
+摒
+揆
+掾
+聒
+葑
+葚
+靰
+靸
+葳
+葺
+葸
+萼
+葆
+葩
+葶
+蒌
+萱
+戟
+葭
+楮
+棼
+椟
+棹
+椤
+棰
+赍
+椋
+椁
+椪
+棣
+椐
+鹁
+覃
+酤
+酢
+酡
+鹂
+厥
+殚
+殛
+雯
+雱
+辊
+辋
+椠
+辍
+辎
+斐
+睄
+睑
+睇
+睃
+戢
+喋
+嗒
+喃
+喱
+喹
+晷
+喈
+跖
+跗
+跞
+跚
+跎
+跏
+跆
+蛱
+蛲
+蛭
+蛳
+蛐
+蛔
+蛞
+蛴
+蛟
+蛘
+喁
+喟
+啾
+嗖
+喑
+嗟
+喽
+嗞
+喀
+喔
+喙
+嵘
+嵖
+崴
+遄
+詈
+嵎
+崽
+嵬
+嵛
+嵯
+嵝
+嵫
+幄
+嵋
+赕
+铻
+铼
+铿
+锃
+锂
+锆
+锇
+锉
+锏
+锑
+锒
+锔
+锕
+掣
+矬
+氰
+毳
+毽
+犊
+犄
+犋
+鹄
+犍
+嵇
+黍
+稃
+稂
+筚
+筵
+筌
+傣
+傈
+舄
+牍
+傥
+傧
+遑
+傩
+遁
+徨
+媭
+畲
+弑
+颌
+翕
+釉
+鹆
+舜
+貂
+腈
+腌
+腓
+腆
+腴
+腑
+腚
+腱
+鱿
+鲀
+鲂
+颍
+猢
+猹
+猥
+飓
+觞
+觚
+猱
+颎
+飧
+馇
+馊
+亵
+脔
+裒
+痣
+痨
+痦
+痞
+痤
+痫
+痧
+赓
+竦
+瓿
+啻
+颏
+鹇
+阑
+阒
+阕
+粞
+遒
+孳
+焯
+焜
+焙
+焱
+鹈
+湛
+渫
+湮
+湎
+湜
+渭
+湍
+湫
+溲
+湟
+溆
+湲
+湔
+湉
+渥
+湄
+滁
+愠
+惺
+愦
+惴
+愀
+愎
+愔
+喾
+寐
+谟
+扉
+裢
+裎
+裥
+祾
+祺
+谠
+幂
+谡
+谥
+谧
+遐
+孱
+弼
+巽
+骘
+媪
+媛
+婷
+巯
+翚
+皴
+婺
+骛
+缂
+缃
+缄
+彘
+缇
+缈
+缌
+缑
+缒
+缗
+飨
+耢
+瑚
+瑁
+瑜
+瑗
+瑄
+瑕
+遨
+骜
+韫
+髡
+塬
+鄢
+趔
+趑
+摅
+摁
+蜇
+搋
+搪
+搐
+搛
+搠
+摈
+彀
+毂
+搦
+搡
+蓁
+戡
+蓍
+鄞
+靳
+蓐
+蓦
+鹋
+蒽
+蓓
+蓖
+蓊
+蒯
+蓟
+蓑
+蒿
+蒺
+蓠
+蒟
+蒡
+蒹
+蒴
+蒗
+蓥
+颐
+楔
+楠
+楂
+楝
+楫
+楸
+椴
+槌
+楯
+皙
+榈
+槎
+榉
+楦
+楣
+楹
+椽
+裘
+剽
+甄
+酮
+酰
+酯
+酩
+蜃
+碛
+碓
+硼
+碉
+碚
+碇
+碜
+鹌
+辏
+龃
+龅
+訾
+粲
+虞
+睚
+嗪
+韪
+嗷
+嗉
+睨
+睢
+雎
+睥
+嘟
+嗑
+嗫
+嗬
+嗔
+嗝
+戥
+嗄
+煦
+暄
+遢
+暌
+跬
+跶
+跸
+跐
+跣
+跹
+跻
+蛸
+蜊
+蜍
+蜉
+蜣
+畹
+蛹
+嗣
+嗯
+嗥
+嗲
+嗳
+嗌
+嗍
+嗨
+嗐
+嗤
+嗵
+罨
+嵊
+嵩
+嵴
+骰
+锗
+锛
+锜
+锝
+锞
+锟
+锢
+锨
+锩
+锭
+锱
+雉
+氲
+犏
+歃
+稞
+稗
+稔
+筠
+筢
+筮
+筲
+筱
+牒
+煲
+敫
+徭
+愆
+艄
+觎
+毹
+貊
+貅
+貉
+颔
+腠
+腩
+腼
+腭
+腧
+塍
+媵
+詹
+鲅
+鲆
+鲇
+鲈
+稣
+鲋
+鲐
+肄
+鹐
+飕
+觥
+遛
+馐
+鹑
+亶
+瘃
+痱
+痼
+痿
+瘐
+瘁
+瘆
+麂
+裔
+歆
+旒
+雍
+阖
+阗
+阙
+羧
+豢
+粳
+猷
+煳
+煜
+煨
+煅
+煊
+煸
+煺
+滟
+溱
+溘
+漭
+滢
+溥
+溧
+溽
+裟
+溻
+溷
+滗
+滫
+溴
+滏
+滃
+滦
+溏
+滂
+滓
+溟
+滪
+愫
+慑
+慊
+鲎
+骞
+窦
+窠
+窣
+裱
+褚
+裨
+裾
+裰
+禊
+谩
+谪
+媾
+嫫
+媲
+嫒
+嫔
+媸
+缙
+缜
+缛
+辔
+骝
+缟
+缡
+缢
+缣
+骟
+耥
+璈
+瑶
+瑭
+獒
+觏
+慝
+嫠
+韬
+叆
+髦
+摽
+墁
+撂
+摞
+撄
+翥
+踅
+摭
+墉
+墒
+榖
+綦
+蔫
+蔷
+靺
+靼
+鞅
+靿
+甍
+蔸
+蔟
+蔺
+戬
+蕖
+蔻
+蓿
+斡
+鹕
+蓼
+榛
+榧
+榻
+榫
+榭
+槔
+榱
+槁
+槟
+槠
+榷
+僰
+酽
+酶
+酹
+厮
+碡
+碴
+碣
+碲
+磋
+臧
+豨
+殡
+霆
+霁
+辕
+蜚
+裴
+翡
+龇
+龈
+睿
+䁖
+睽
+嘞
+嘈
+嘌
+嘁
+嘎
+暧
+暝
+踌
+踉
+蜞
+蜥
+蜮
+蝈
+蜴
+蜱
+蜩
+蜷
+蜿
+螂
+蜢
+嘘
+嘡
+鹗
+嘣
+嘤
+嘚
+嗾
+嘧
+罴
+罱
+幔
+嶂
+幛
+赙
+罂
+骷
+骶
+鹘
+锲
+锴
+锶
+锷
+锸
+锵
+镁
+镂
+犒
+箐
+箦
+箧
+箍
+箸
+箬
+箅
+箪
+箔
+箜
+箢
+箓
+毓
+僖
+儆
+僳
+僭
+劁
+僮
+魃
+魆
+睾
+艋
+鄱
+膈
+膑
+鲑
+鲔
+鲚
+鲛
+鲟
+獐
+觫
+雒
+夤
+馑
+銮
+塾
+麽
+瘌
+瘊
+瘘
+瘙
+廖
+韶
+旖
+膂
+阚
+鄯
+鲞
+粿
+粼
+粽
+糁
+槊
+鹚
+熘
+熥
+潢
+漕
+滹
+漯
+漶
+潋
+潴
+漪
+漉
+漳
+漩
+澉
+潍
+慵
+搴
+窨
+寤
+綮
+谮
+褡
+褙
+褓
+褛
+褊
+谯
+谰
+谲
+暨
+屣
+鹛
+嫣
+嫱
+嫖
+嫦
+嫚
+嫘
+嫡
+鼐
+翟
+瞀
+鹜
+骠
+缥
+缦
+缧
+缨
+骢
+缪
+缫
+耦
+耧
+瑾
+璜
+璀
+璎
+璁
+璋
+璇
+奭
+髯
+髫
+撷
+撅
+赭
+撸
+鋆
+撙
+撺
+墀
+聩
+觐
+鞑
+蕙
+鞒
+蕈
+蕨
+蕤
+蕞
+蕺
+瞢
+蕃
+蕲
+赜
+槿
+樯
+槭
+樗
+樘
+樊
+槲
+醌
+醅
+靥
+魇
+餍
+磔
+磙
+霈
+辘
+龉
+龊
+觑
+瞌
+瞋
+瞑
+嘭
+噎
+噶
+颙
+暹
+噘
+踔
+踝
+踟
+踒
+踬
+踮
+踯
+踺
+踞
+蝽
+蝾
+蝻
+蝰
+蝮
+螋
+蝓
+蝣
+蝼
+噗
+嘬
+颚
+噍
+噢
+噙
+噜
+噌
+噔
+颛
+幞
+幡
+嶙
+嶝
+骺
+骼
+骸
+镊
+镉
+镌
+镍
+镏
+镒
+镓
+镔
+稷
+箴
+篑
+篁
+篌
+篆
+牖
+儋
+徵
+磐
+虢
+鹞
+膘
+滕
+鲠
+鲡
+鲢
+鲣
+鲥
+鲧
+鲩
+獗
+獠
+觯
+馓
+馔
+麾
+廛
+瘛
+瘼
+瘢
+瘠
+齑
+羯
+羰
+𥻗
+遴
+糌
+糍
+糅
+熜
+熵
+熠
+澍
+澌
+潸
+潦
+潲
+鋈
+潟
+潼
+潺
+憬
+憧
+寮
+窳
+谳
+褴
+褟
+褫
+谵
+熨
+屦
+嬉
+勰
+戮
+蝥
+缬
+缮
+缯
+骣
+畿
+耩
+耨
+耪
+璞
+璟
+靛
+璠
+璘
+聱
+螯
+髻
+髭
+髹
+擀
+熹
+甏
+擞
+縠
+磬
+颞
+蕻
+鞘
+颟
+薤
+薨
+檠
+薏
+薮
+薜
+薅
+樾
+橛
+橇
+樵
+檎
+橹
+樽
+樨
+橼
+墼
+橐
+翮
+醛
+醐
+醍
+醚
+磲
+赝
+飙
+殪
+霖
+霏
+霓
+錾
+辚
+臻
+遽
+氅
+瞟
+瞠
+瞰
+嚄
+嚆
+噤
+暾
+蹀
+踹
+踵
+踽
+蹉
+蹁
+螨
+蟒
+螈
+螅
+螭
+螠
+螟
+噱
+噬
+噫
+噻
+噼
+罹
+圜
+䦃
+镖
+镗
+镘
+镚
+镛
+镝
+镞
+镠
+氇
+氆
+憩
+穑
+篝
+篥
+篦
+篪
+篙
+盥
+劓
+翱
+魉
+魈
+徼
+歙
+膳
+膦
+膙
+鲮
+鲱
+鲲
+鲳
+鲴
+鲵
+鲷
+鲻
+獴
+獭
+獬
+邂
+鹧
+廨
+赟
+瘰
+廪
+瘿
+瘵
+瘴
+癃
+瘳
+斓
+麇
+麈
+嬴
+壅
+羲
+糗
+瞥
+甑
+燎
+燠
+燔
+燧
+濑
+濉
+潞
+澧
+澹
+澥
+澶
+濂
+褰
+寰
+窸
+褶
+禧
+嬖
+犟
+隰
+嬗
+颡
+缱
+缲
+缳
+璨
+璩
+璐
+璪
+螫
+擤
+壕
+觳
+罄
+擢
+薹
+鞡
+鞬
+薷
+薰
+藓
+藁
+檄
+檩
+懋
+醢
+翳
+礅
+磴
+鹩
+龋
+龌
+豳
+壑
+黻
+嚏
+嚅
+蹑
+蹒
+蹊
+蟥
+螬
+螵
+疃
+螳
+蟑
+嚓
+羁
+罽
+罾
+嶷
+黜
+黝
+髁
+髀
+镡
+镢
+镣
+镦
+镧
+镩
+镪
+镫
+罅
+黏
+簌
+篾
+篼
+簖
+簋
+鼢
+黛
+儡
+鹪
+鼾
+皤
+魍
+龠
+繇
+貘
+邈
+貔
+臌
+膻
+臆
+臃
+鲼
+鲽
+鳀
+鳃
+鳅
+鳇
+鳊
+螽
+燮
+鹫
+襄
+糜
+縻
+膺
+癍
+麋
+懑
+濡
+濮
+濞
+濠
+濯
+蹇
+謇
+邃
+襁
+檗
+擘
+孺
+隳
+嬷
+蟊
+鹬
+鍪
+鏊
+鳌
+鬈
+鬃
+瞽
+鞯
+鞨
+鞫
+鞧
+鞣
+藜
+藠
+藩
+醪
+蹙
+礓
+燹
+餮
+瞿
+曛
+颢
+曜
+躇
+蹚
+鹭
+蟛
+蟪
+蟠
+蟮
+鹮
+黠
+黟
+髅
+髂
+镬
+镭
+镯
+馥
+簟
+簪
+鼬
+雠
+艟
+鳎
+鳏
+鳐
+癞
+癔
+癜
+癖
+糨
+蹩
+鎏
+懵
+彝
+邋
+鬏
+攉
+攒
+鞲
+鞴
+藿
+蘧
+蘅
+麓
+醮
+醯
+酃
+霪
+霭
+霨
+黼
+嚯
+蹰
+蹶
+蹽
+蹼
+蹴
+蹾
+蹿
+蠖
+蠓
+蟾
+蠊
+黢
+髋
+髌
+镲
+籀
+籁
+齁
+魑
+艨
+鳓
+鳔
+鳕
+鳗
+鳙
+麒
+鏖
+羸
+㸆
+瀚
+瀣
+瀛
+襦
+谶
+襞
+骥
+缵
+瓒
+攘
+蘩
+蘖
+醴
+霰
+酆
+矍
+曦
+躅
+鼍
+巉
+黩
+黥
+黪
+镳
+镴
+黧
+纂
+璺
+鼯
+臜
+鳜
+鳝
+鳟
+獾
+孀
+骧
+瓘
+鼙
+醺
+礴
+颦
+曩
+鳢
+癫
+麝
+夔
+爝
+灏
+禳
+鐾
+羼
+蠡
+耱
+懿
+蘸
+鹳
+霾
+氍
+饕
+躐
+髑
+镵
+穰
+饔
+鬻
+鬟
+趱
+攫
+攥
+颧
+躜
+鼹
+癯
+麟
+蠲
+蠹
+躞
+衢
+鑫
+灞
+襻
+纛
+鬣
+攮
+囔
+馕
+戆
+爨
+齉
+亍
+尢
+彳
+卬
+殳
+𠙶
+毌
+邘
+戋
+圢
+氕
+伋
+仝
+冮
+氿
+汈
+氾
+忉
+宄
+讱
+扞
+圲
+圫
+芏
+芃
+朳
+朸
+𨙸
+邨
+吒
+吖
+屼
+屾
+辿
+钆
+仳
+伣
+伈
+癿
+甪
+邠
+犴
+冱
+邡
+闫
+汋
+䜣
+讻
+孖
+纩
+玒
+玓
+玘
+玚
+刬
+坜
+坉
+扽
+坋
+扺
+㧑
+毐
+芰
+芣
+苊
+苉
+芘
+芴
+芠
+芤
+杕
+杙
+杄
+杧
+杩
+尪
+尨
+轪
+坒
+芈
+旴
+旵
+呙
+㕮
+岍
+岠
+岜
+呇
+冏
+觃
+岙
+伾
+㑇
+伭
+佖
+伲
+佁
+飏
+狃
+闶
+汧
+汫
+𣲘
+𣲗
+沄
+沘
+汭
+㳇
+沇
+忮
+忳
+忺
+祃
+诇
+邲
+诎
+诐
+屃
+岊
+阽
+䢺
+阼
+妧
+妘
+𨚕
+纮
+驲
+纻
+纼
+玤
+玞
+玱
+玟
+邽
+邿
+坥
+坰
+坬
+坽
+弆
+耵
+䢼
+𦭜
+茋
+苧
+苾
+苠
+枅
+㭎
+枘
+枍
+矼
+矻
+匼
+旿
+昇
+昄
+昒
+昈
+咉
+咇
+咍
+岵
+岽
+岨
+岞
+峂
+㟃
+囷
+钐
+钔
+钖
+牥
+佴
+垈
+侁
+侹
+佸
+佺
+隹
+㑊
+侂
+佽
+侘
+郈
+舠
+郐
+郃
+攽
+肭
+肸
+肷
+狉
+狝
+饳
+忞
+於
+炌
+炆
+泙
+沺
+泂
+泜
+泃
+泇
+怊
+峃
+穸
+祋
+祊
+鸤
+弢
+弨
+陑
+陎
+卺
+乸
+妭
+姈
+迳
+叕
+驵
+䌹
+驺
+绋
+绐
+砉
+耔
+㛃
+玶
+珇
+珅
+珋
+玹
+珌
+玿
+韨
+垚
+垯
+垙
+垲
+埏
+垍
+耇
+垎
+垴
+垟
+垞
+挓
+垵
+垏
+拶
+荖
+荁
+荙
+荛
+茈
+茽
+荄
+茺
+荓
+茳
+𦰡
+茛
+荭
+㭕
+柷
+柃
+柊
+枹
+栐
+柖
+郚
+剅
+䴓
+迺
+厖
+砆
+砑
+砄
+耏
+奓
+䶮
+轵
+轷
+轹
+轺
+昺
+昽
+盷
+咡
+咺
+昳
+昣
+哒
+昤
+昫
+昡
+咥
+昪
+虷
+虸
+哃
+峘
+耑
+峛
+峗
+峧
+帡
+钘
+钜
+钪
+钬
+钭
+矧
+秬
+俫
+舁
+俜
+俙
+俍
+垕
+衎
+舣
+弇
+侴
+鸧
+䏡
+胠
+𦙶
+胈
+胩
+胣
+朏
+飐
+訄
+饻
+庤
+疢
+炣
+炟
+㶲
+洭
+洘
+洓
+洿
+㳚
+泚
+浈
+浉
+洸
+洑
+洢
+洈
+洚
+洺
+洨
+浐
+㳘
+洴
+洣
+恔
+宬
+窀
+扂
+袆
+祏
+祐
+祕
+叚
+陧
+陞
+娀
+姞
+姱
+姤
+姶
+姽
+枲
+绖
+骃
+彖
+骉
+恝
+珪
+珛
+珹
+琊
+玼
+珖
+珽
+珦
+珫
+珒
+珢
+珕
+珝
+埗
+垾
+垺
+埆
+垿
+埌
+埇
+莰
+茝
+鄀
+莶
+莝
+䓖
+莙
+栻
+桠
+桄
+梠
+栴
+梴
+栒
+酎
+酏
+砵
+砠
+砫
+砬
+硁
+恧
+翃
+郪
+𨐈
+辀
+辁
+剕
+赀
+哢
+晅
+晊
+唝
+哳
+哱
+冔
+晔
+晐
+晖
+畖
+蚄
+蚆
+帱
+崁
+峿
+崄
+帨
+崀
+赆
+钷
+眚
+甡
+笫
+倻
+倴
+脩
+倮
+倕
+倞
+倓
+倧
+衃
+虒
+舭
+舯
+舥
+瓞
+鬯
+鸰
+脎
+朓
+胲
+虓
+鱽
+狴
+峱
+狻
+眢
+勍
+痄
+疰
+痃
+竘
+羖
+羓
+桊
+敉
+烠
+烔
+烶
+烻
+涍
+浡
+浭
+浬
+涄
+涢
+涐
+浰
+浟
+浛
+浼
+浲
+涘
+悈
+悃
+悢
+宧
+窅
+窊
+窎
+扅
+扆
+袪
+袗
+袯
+祧
+隺
+堲
+疍
+𨺙
+陴
+烝
+砮
+㛚
+哿
+翀
+翂
+剟
+绤
+骍
+䂮
+琎
+珸
+珵
+琄
+琈
+琀
+珺
+掭
+堎
+堐
+埼
+掎
+埫
+堌
+晢
+掞
+埪
+壸
+㙍
+聍
+菝
+萚
+菥
+莿
+䓫
+勚
+䓬
+萆
+菂
+菍
+菼
+萣
+䓨
+菉
+䓛
+梼
+梽
+桲
+梾
+桯
+梣
+梌
+桹
+敔
+厣
+硔
+硙
+硚
+硊
+硍
+勔
+䴕
+龁
+逴
+唪
+啫
+翈
+㫰
+晙
+畤
+趼
+跂
+蛃
+蚲
+蚺
+啴
+䎃
+崧
+崟
+崞
+崒
+崌
+崡
+铏
+铕
+铖
+铘
+铚
+铞
+铥
+铴
+牻
+牿
+稆
+笱
+笯
+偰
+偡
+鸺
+偭
+偲
+偁
+㿠
+鄅
+偓
+徛
+衒
+舳
+舲
+鸼
+悆
+鄃
+瓻
+䝙
+脶
+脞
+脟
+䏲
+鱾
+猇
+猊
+猄
+觖
+𠅤
+庱
+庼
+庳
+痓
+䴔
+竫
+堃
+阌
+羝
+羕
+焆
+烺
+焌
+淏
+淟
+淜
+淴
+淯
+湴
+涴
+㥄
+惛
+惔
+悰
+惙
+寁
+逭
+袼
+裈
+祲
+谞
+艴
+弸
+弶
+隃
+婞
+娵
+婼
+媖
+婳
+婍
+婌
+婫
+婤
+婘
+婠
+绹
+骕
+絜
+珷
+琲
+琡
+琟
+琔
+琭
+堾
+堼
+揕
+㙘
+堧
+喆
+堨
+塅
+堠
+絷
+𡎚
+葜
+惎
+萳
+葙
+靬
+葴
+蒇
+蒈
+鄚
+蒉
+蓇
+萩
+蒐
+葰
+葎
+鄑
+蒎
+葖
+蒄
+萹
+棤
+棽
+棫
+椓
+椑
+鹀
+椆
+棓
+棬
+棪
+椀
+楗
+甦
+酦
+觌
+奡
+皕
+硪
+欹
+詟
+辌
+棐
+龂
+黹
+牚
+睎
+晫
+晪
+晱
+𧿹
+蛑
+畯
+斝
+喤
+崶
+嵁
+崾
+嵅
+崿
+嵚
+翙
+圌
+圐
+赑
+淼
+赒
+铹
+铽
+𨱇
+锊
+锍
+锎
+锓
+犇
+颋
+稌
+筀
+筘
+筜
+筥
+筅
+傃
+傉
+翛
+傒
+傕
+舾
+畬
+脿
+腘
+䐃
+腙
+腒
+鲃
+猰
+猯
+㺄
+馉
+鄗
+廋
+廆
+鄌
+粢
+遆
+旐
+焞
+欻
+𣸣
+溚
+溁
+湝
+渰
+湓
+㴔
+渟
+溠
+渼
+溇
+湣
+湑
+溞
+愐
+愃
+敩
+甯
+棨
+扊
+裣
+祼
+婻
+媆
+媞
+㛹
+媓
+媂
+媄
+毵
+矞
+缊
+缐
+骙
+瑃
+瑓
+瑅
+瑆
+䴖
+瑖
+瑝
+瑔
+瑀
+𤧛
+瑳
+瑂
+嶅
+瑑
+遘
+髢
+塥
+堽
+赪
+摛
+塝
+搒
+搌
+蒱
+蒨
+蓏
+蔀
+蓢
+蓂
+蒻
+蓣
+椹
+楪
+榃
+榅
+楒
+楞
+楩
+榇
+椸
+楙
+歅
+碃
+碏
+碈
+䃅
+硿
+鄠
+辒
+龆
+觜
+䣘
+暕
+鹍
+㬊
+暅
+跱
+蜐
+蜎
+嵲
+赗
+骱
+锖
+锘
+锳
+锧
+锪
+锫
+锬
+稑
+稙
+䅟
+筻
+筼
+筶
+筦
+筤
+傺
+鹎
+僇
+艅
+艉
+谼
+貆
+腽
+腨
+腯
+鲉
+鲊
+鲌
+䲟
+鲏
+雊
+猺
+飔
+觟
+𦝼
+馌
+裛
+廒
+瘀
+瘅
+鄘
+鹒
+鄜
+麀
+鄣
+阘
+煁
+煃
+煴
+煋
+煟
+煓
+滠
+溍
+溹
+滆
+滉
+溦
+溵
+漷
+滧
+滘
+滍
+愭
+慥
+慆
+塱
+裼
+禋
+禔
+禘
+禒
+谫
+鹔
+愍
+嫄
+媱
+戤
+戣
+缞
+耤
+瑧
+瑨
+瑱
+瑷
+瑢
+斠
+摏
+墕
+墈
+墐
+墘
+摴
+銎
+𡐓
+墚
+撖
+靽
+鞁
+蔌
+蔈
+蓰
+蔹
+蔊
+嘏
+榰
+榑
+槚
+𣗋
+槜
+榍
+疐
+酺
+酾
+酲
+酴
+碶
+䃎
+碨
+𥔲
+碹
+碥
+劂
+䴗
+夥
+瞍
+鹖
+㬎
+跽
+蜾
+幖
+嶍
+圙
+𨱏
+锺
+锼
+锽
+锾
+锿
+镃
+镄
+镅
+馝
+鹙
+箨
+箖
+劄
+僬
+僦
+僔
+僎
+槃
+㙦
+鲒
+鲕
+鲖
+鲗
+鲘
+鲙
+𩽾
+夐
+獍
+飗
+凘
+廑
+廙
+瘗
+瘥
+瘕
+鲝
+鄫
+熇
+漹
+漖
+潆
+漤
+潩
+漼
+漴
+㽏
+漈
+漋
+漻
+慬
+窬
+窭
+㮾
+褕
+禛
+禚
+隩
+嫕
+嫭
+嫜
+嫪
+㻬
+麹
+璆
+漦
+叇
+墣
+墦
+墡
+劐
+薁
+蕰
+蔃
+鼒
+槱
+鹝
+磏
+磉
+殣
+慭
+霅
+暵
+暲
+暶
+踦
+踣
+䗖
+蝘
+蝲
+蝤
+噇
+噂
+噀
+罶
+嶲
+嶓
+㠇
+嶟
+嶒
+镆
+镈
+镋
+镎
+镕
+稹
+儇
+皞
+皛
+䴘
+艎
+艏
+鹟
+𩾃
+鲦
+鲪
+鲬
+橥
+觭
+鹠
+鹡
+糇
+糈
+翦
+鹢
+鹣
+熛
+潖
+潵
+㵐
+澂
+澛
+瑬
+潽
+潾
+潏
+憭
+憕
+戭
+褯
+禤
+嫽
+遹
+璥
+璲
+璒
+憙
+擐
+鄹
+薳
+鞔
+黇
+蕗
+薢
+蕹
+橞
+橑
+橦
+醑
+觱
+磡
+𥕢
+磜
+豮
+鹾
+虤
+暿
+曌
+曈
+㬚
+蹅
+踶
+䗛
+螗
+疁
+㠓
+幪
+嶦
+𨱑
+馞
+穄
+篚
+篯
+簉
+鼽
+衠
+盦
+螣
+縢
+鲭
+鲯
+鲰
+鲺
+鲹
+亸
+癀
+瘭
+羱
+糒
+燋
+熻
+燊
+燚
+燏
+濩
+濋
+澪
+澽
+澴
+澭
+澼
+憷
+憺
+懔
+黉
+嬛
+鹨
+翯
+璱
+𤩽
+璬
+璮
+髽
+擿
+薿
+薸
+檑
+櫆
+檞
+醨
+繄
+磹
+磻
+瞫
+瞵
+蹐
+蟏
+㘎
+镤
+镥
+镨
+𨱔
+矰
+穙
+穜
+穟
+簕
+簃
+簏
+儦
+魋
+斶
+艚
+谿
+䲠
+鲾
+鲿
+鳁
+鳂
+鳈
+鳉
+獯
+䗪
+馘
+襕
+襚
+螱
+甓
+嬬
+嬥
+𦈡
+瓀
+釐
+鬶
+爇
+鞳
+鞮
+藟
+藦
+藨
+鹲
+檫
+黡
+礞
+礌
+𥖨
+蹢
+蹜
+蟫
+䗴
+嚚
+髃
+镮
+镱
+酂
+馧
+簠
+簝
+簰
+鼫
+鼩
+皦
+臑
+䲢
+鳑
+鳒
+鹱
+鹯
+癗
+𦒍
+旞
+翷
+冁
+䎖
+瀔
+瀍
+瀌
+襜
+䴙
+嚭
+㰀
+鬷
+醭
+蹯
+蠋
+翾
+鳘
+儳
+儴
+鼗
+𩾌
+鳚
+鳛
+麑
+麖
+蠃
+彟
+嬿
+鬒
+蘘
+欂
+醵
+颥
+甗
+𨟠
+巇
+酅
+髎
+犨
+𨭉
+㸌
+爔
+瀱
+瀹
+瀼
+瀵
+襫
+孅
+骦
+耰
+𤫉
+瓖
+鬘
+趯
+罍
+鼱
+鳠
+鳡
+鳣
+爟
+爚
+灈
+韂
+糵
+蘼
+礵
+鹴
+躔
+皭
+龢
+鳤
+亹
+籥
+鼷
+玃
+醾
+齇
+觿
+蠼
+𬣙
+𬇕
+𬣞
+𬘓
+𫭟
+𫭢
+𫇭
+𫐄
+𫵷
+𬇙
+𬣡
+𫸩
+𫘜
+𬘘
+𫘝
+𬨂
+𬀩
+𬀪
+𬬩
+𫍣
+𬣳
+𬩽
+𬮿
+𬯀
+𫰛
+𬳵
+𬳶
+𫠊
+𬍛
+鿍
+𬜬
+𪾢
+𪨰
+𫓧
+𬬮
+𬬱
+𬬭
+𬘡
+𬳽
+𬘩
+𫄧
+𪟝
+𬍤
+𫭼
+𬜯
+𬂩
+𫠆
+𬌗
+𫑡
+𪨶
+𬬸
+𬬻
+𬬹
+𬬿
+𬭁
+𫢸
+𫗧
+𬊈
+𬒈
+𬳿
+𫄨
+𬘫
+𫮃
+鿎
+𬱖
+𬟽
+𫓯
+𫟹
+𫟼
+𬇹
+𬍡
+𬤇
+𫍯
+𬤊
+𫍲
+𬯎
+𬘬
+𬘭
+𬴂
+𫘦
+𫟅
+𬘯
+𫘧
+𪣻
+𬃊
+𬷕
+𫐐
+𬹼
+𫶇
+𫖮
+鿏
+𬭊
+𫓶
+𬭎
+𫖯
+𬱟
+𫛭
+𫷷
+𬮱
+𬊤
+𬴃
+𫘨
+𬪩
+𬒔
+𬨎
+𫐓
+𫫇
+𫓹
+𬭚
+𬭛
+𬕂
+𬶋
+𬶍
+𫔶
+𫌀
+𫖳
+𫘪
+𫘬
+𫞩
+𪤗
+𬸘
+𬒗
+𫚖
+𬭤
+𫚕
+𬶐
+𬶏
+𬸚
+𬤝
+𬙂
+𬭩
+𬸣
+𫍽
+𬴊
+𬞟
+𫟦
+𬺈
+𫠜
+𪩘
+𬭬
+𬭯
+𫗴
+𬸦
+𫄷
+𬭳
+𬭶
+𫔍
+𬭸
+𬭼
+𫔎
+𬸪
+𬶟
+𬶠
+𬶨
+𫄸
+𬟁
+𬙊
+𬶭
+𬶮
+𬙋
+𬺓
+𫚭
+廠
+蔔
+兒
+幾
+幹
+虧
+纔
+與
+萬
+韆
+億
+個
+廣
+門
+義
+衛
+飛
+習
+馬
+鄉
+豐
+開
+無
+雲
+專
+藝
+廳
+區
+歷
+曆
+車
+貝
+岡
+見
+氣
+長
+僕
+幣
+僅
+從
+侖
+倉
+風
+烏
+鳳
+爲
+鬥
+憶
+計
+訂
+認
+譏
+醜
+隊
+辦
+鄧
+勸
+雙
+書
+擊
+撲
+節
+術
+厲
+龍
+滅
+軋
+東
+盧
+業
+舊
+帥
+歸
+葉
+電
+號
+衹
+隻
+嘰
+嘆
+們
+儀
+叢
+爾
+樂
+處
+鼕
+鳥
+務
+飢
+饑
+馮
+閃
+蘭
+匯
+彙
+頭
+漢
+寧
+討
+寫
+讓
+禮
+訓
+議
+訊
+記
+齣
+遼
+邊
+發
+髮
+聖
+對
+臺
+颱
+檯
+糾
+絲
+動
+鞏
+執
+擴
+掃
+場
+揚
+亞
+樸
+機
+權
+過
+協
+壓
+厭
+頁
+誇
+奪
+達
+夾
+軌
+堯
+劃
+邁
+畢
+貞
+師
+塵
+當
+噹
+籲
+嚇
+蟲
+麯
+團
+糰
+嗎
+嶼
+歲
+迴
+豈
+則
+剛
+網
+硃
+遷
+喬
+偉
+傳
+優
+傷
+價
+倫
+華
+僞
+嚮
+後
+會
+殺
+閤
+衆
+爺
+傘
+創
+雜
+負
+壯
+衝
+妝
+莊
+慶
+劉
+齊
+産
+閉
+問
+闖
+關
+燈
+湯
+興
+講
+諱
+軍
+訝
+許
+訛
+論
+訟
+農
+諷
+設
+訪
+訣
+尋
+盡
+儘
+導
+孫
+陣
+陽
+階
+陰
+婦
+媽
+戲
+觀
+歡
+買
+紅
+馱
+纖
+縴
+馴
+約
+級
+紀
+馳
+紉
+壽
+麥
+瑪
+進
+遠
+違
+韌
+運
+撫
+壇
+罎
+壞
+摳
+擾
+貢
+垻
+壩
+摺
+掄
+搶
+墳
+護
+殻
+塊
+聲
+報
+擬
+蕪
+葦
+蒼
+嚴
+蘆
+勞
+蘇
+囌
+極
+楊
+兩
+麗
+醫
+勵
+還
+殲
+來
+連
+軒
+鹵
+滷
+堅
+時
+縣
+裏
+嘔
+園
+曠
+圍
+噸
+郵
+睏
+員
+聽
+嗆
+嗚
+彆
+嶇
+崗
+帳
+財
+針
+釘
+亂
+體
+傭
+徹
+餘
+穀
+鄰
+腸
+龜
+猶
+狽
+條
+島
+飯
+飲
+係
+繫
+凍
+狀
+畝
+庫
+療
+應
+這
+廬
+閏
+閑
+間
+悶
+竈
+燦
+瀝
+淪
+滄
+溝
+滬
+瀋
+懷
+憂
+窮
+證
+啓
+評
+補
+識
+詐
+訴
+診
+詞
+譯
+靈
+層
+遲
+張
+際
+陸
+陳
+墜
+勁
+鷄
+緯
+驅
+純
+紗
+綱
+納
+駁
+縱
+紛
+紙
+紋
+紡
+驢
+紐
+環
+責
+現
+錶
+規
+攏
+揀
+擔
+頂
+擁
+勢
+攔
+擰
+撥
+擇
+蘋
+範
+莖
+樞
+櫃
+闆
+鬆
+槍
+楓
+構
+喪
+畫
+棗
+賣
+鬱
+礬
+礦
+碼
+厠
+奮
+態
+歐
+毆
+壟
+轟
+頃
+轉
+斬
+輪
+軟
+齒
+虜
+腎
+賢
+國
+暢
+嚨
+鳴
+羅
+幟
+嶺
+凱
+敗
+賬
+販
+貶
+購
+貯
+圖
+釣
+製
+颳
+俠
+僥
+偵
+側
+憑
+僑
+貨
+質
+徑
+捨
+覓
+貪
+貧
+膚
+腫
+脹
+骯
+脅
+魚
+獰
+備
+飾
+飽
+飼
+變
+龐
+廟
+瘧
+劑
+廢
+閘
+鬧
+鄭
+捲
+單
+爐
+淺
+濘
+瀉
+潑
+澤
+憐
+學
+寶
+寵
+審
+簾
+實
+試
+詩
+誠
+襯
+視
+話
+誕
+詭
+詢
+該
+詳
+肅
+録
+隸
+彌
+瀰
+陝
+駕
+參
+艱
+綫
+練
+組
+紳
+細
+駛
+織
+駒
+終
+駐
+絆
+駝
+紹
+繹
+經
+貫
+貳
+幫
+項
+挾
+撓
+趙
+擋
+墊
+擠
+揮
+薦
+帶
+繭
+蕩
+榮
+葷
+熒
+鬍
+蔭
+藥
+標
+棧
+棟
+欄
+檸
+樹
+鹹
+磚
+硯
+麵
+牽
+鷗
+殘
+軸
+輕
+鴉
+戰
+點
+臨
+覽
+竪
+嘗
+啞
+顯
+貴
+蝦
+蟻
+螞
+雖
+駡
+勛
+嘩
+響
+喲
+峽
+罰
+賤
+貼
+貽
+鈣
+鈍
+鈔
+鍾
+鐘
+鋼
+鈉
+鑰
+欽
+鈞
+鈎
+鈕
+氈
+氫
+選
+適
+種
+鞦
+復
+複
+倆
+貸
+順
+儉
+須
+鬚
+劍
+朧
+膽
+勝
+狹
+獅
+獨
+獄
+貿
+餌
+饒
+蝕
+餃
+餅
+巒
+彎
+將
+奬
+瘡
+瘋
+親
+閨
+聞
+閩
+閥
+閣
+養
+薑
+類
+婁
+總
+煉
+爍
+爛
+窪
+潔
+灑
+澆
+濁
+測
+瀏
+濟
+渾
+濃
+惱
+舉
+覺
+憲
+竊
+誡
+誣
+語
+襖
+誤
+誘
+誨
+説
+誦
+墾
+晝
+費
+遜
+隕
+險
+嬌
+賀
+壘
+綁
+絨
+結
+繞
+驕
+繪
+給
+絢
+駱
+絡
+絶
+絞
+駭
+統
+艷
+蠶
+頑
+盞
+撈
+載
+趕
+鹽
+損
+撿
+摯
+剝
+熱
+搗
+壺
+聶
+萊
+蓮
+獲
+穫
+惡
+噁
+瑩
+鶯
+檔
+橋
+樺
+樁
+樣
+賈
+礫
+礎
+顧
+轎
+較
+頓
+斃
+緻
+慮
+監
+緊
+黨
+曬
+曉
+嘮
+鴨
+暈
+鴦
+罷
+圓
+賊
+賄
+賂
+贜
+錢
+鉗
+鑽
+鉀
+鐵
+鈴
+鉛
+犧
+敵
+積
+稱
+筆
+債
+傾
+賃
+艦
+艙
+聳
+愛
+頒
+頌
+臟
+髒
+臍
+膠
+腦
+膿
+鴕
+鴛
+皺
+餓
+餒
+戀
+槳
+漿
+準
+癥
+齋
+離
+資
+競
+閲
+煩
+燒
+燭
+遞
+濤
+澇
+渦
+塗
+滌
+潤
+澗
+漲
+燙
+澀
+憫
+寬
+傢
+賓
+竅
+請
+諸
+諾
+讀
+誹
+襪
+課
+誰
+調
+諒
+諄
+談
+誼
+懇
+劇
+難
+預
+絹
+綉
+驗
+繼
+駿
+瑣
+擲
+據
+摻
+職
+蘿
+螢
+營
+蕭
+薩
+夢
+檢
+醖
+碩
+聾
+襲
+輔
+輛
+顱
+懸
+躍
+纍
+囉
+嘯
+嶄
+邏
+嬰
+銬
+鐺
+鋁
+銅
+銘
+鏟
+銀
+矯
+穢
+籠
+償
+軀
+釁
+銜
+盤
+鴿
+斂
+領
+臉
+獵
+餡
+館
+癢
+鏇
+閻
+闡
+蓋
+斷
+獸
+鴻
+漸
+淵
+漁
+澱
+滲
+慚
+懼
+驚
+慘
+慣
+謀
+諜
+謊
+諧
+禱
+禍
+謂
+諺
+謎
+彈
+墮
+隨
+隱
+嬸
+頗
+頸
+績
+緒
+續
+騎
+綽
+繩
+維
+綿
+綳
+綢
+綜
+綻
+緑
+綴
+瓊
+趨
+攬
+攙
+擱
+摟
+攪
+聯
+蔣
+韓
+橢
+確
+頰
+靂
+暫
+翹
+輩
+鑿
+輝
+賞
+睞
+噴
+疇
+踐
+遺
+鵑
+賦
+賭
+贖
+賜
+賠
+鑄
+鋪
+鏈
+銷
+鎖
+鋤
+鍋
+銹
+鋒
+鋅
+鋭
+鵝
+築
+篩
+儲
+懲
+禦
+釋
+臘
+魯
+憊
+饋
+饞
+裝
+蠻
+闊
+糞
+滯
+濕
+潰
+濺
+灣
+憤
+竄
+窩
+褲
+禪
+謝
+謡
+謗
+謙
+屬
+屢
+緬
+纜
+緝
+緞
+緩
+締
+縷
+騙
+編
+騷
+緣
+鵡
+攝
+擺
+襬
+攤
+鵲
+藍
+濛
+懞
+矇
+獻
+欖
+樓
+賴
+礙
+尷
+霧
+輻
+輯
+輸
+頻
+齡
+鑒
+蹺
+蝸
+錯
+錨
+錫
+鑼
+錘
+錐
+錦
+鍵
+鋸
+錳
+辭
+頽
+籌
+簽
+籤
+簡
+膩
+鵬
+騰
+鮑
+穎
+觸
+雛
+饃
+餾
+醬
+謄
+糧
+數
+滿
+濾
+濫
+灕
+濱
+灘
+譽
+窺
+寢
+謹
+謬
+闢
+縛
+縫
+纏
+繽
+贅
+墻
+衊
+藹
+檻
+釀
+願
+轄
+輾
+顆
+踴
+蠟
+蠅
+蟬
+賺
+鍬
+鍛
+鍍
+穩
+籮
+簫
+輿
+鮮
+饅
+瀟
+賽
+譚
+譜
+騾
+縮
+攆
+聰
+藴
+櫻
+飄
+黴
+瞞
+題
+囑
+鎮
+鎬
+鎊
+簍
+鯉
+鯽
+癟
+癱
+顔
+鯊
+瀾
+額
+譴
+鶴
+繚
+顛
+轍
+鸚
+贈
+鏡
+贊
+籃
+籬
+鯨
+癮
+辯
+瀕
+懶
+繮
+繳
+矚
+贍
+鰐
+辮
+贏
+驟
+囂
+鐮
+鰭
+鷹
+巔
+顫
+癬
+鱉
+鬢
+鱗
+躪
+贛
+鑲
+韋
+閂
+訃
+勱
+芻
+鄺
+訐
+訌
+訕
+訖
+馭
+璣
+壙
+捫
+薌
+厙
+釔
+傴
+倀
+傖
+獷
+獁
+鳬
+鄔
+餳
+懺
+謳
+詎
+訥
+紆
+紂
+紇
+紈
+璵
+摶
+塢
+㩳
+蕓
+藶
+莧
+萇
+蓯
+磯
+奩
+歟
+軔
+鄴
+嘸
+囈
+嚦
+暘
+唄
+幃
+峴
+嵐
+圇
+釗
+釙
+釕
+僉
+鳩
+鄒
+飩
+餼
+飪
+飫
+飭
+廡
+癤
+闈
+閎
+閔
+煬
+灃
+漚
+渢
+潙
+憮
+慪
+愾
+悵
+愴
+詁
+訶
+詛
+詆
+謅
+詔
+詒
+隴
+陘
+嫵
+嫗
+嬀
+剄
+紜
+紕
+紝
+綸
+紓
+瑋
+匭
+壚
+擓
+蘢
+蔦
+塋
+煢
+櫪
+梘
+棖
+樅
+碭
+甌
+郟
+軛
+鳶
+曇
+蟣
+黽
+嚀
+噝
+巋
+劌
+剴
+嶧
+釷
+釺
+釧
+釩
+釹
+釵
+儈
+儕
+儂
+劊
+慫
+糴
+戧
+膞
+邇
+梟
+餞
+飴
+癘
+瘍
+煒
+熰
+熗
+瀧
+瀘
+濼
+涇
+㥮
+懌
+誆
+誄
+詿
+詰
+詼
+鄆
+禕
+誅
+詵
+詬
+詮
+詣
+諍
+詫
+諢
+詡
+駑
+紺
+紲
+紱
+駟
+駙
+縐
+絀
+驛
+駘
+瓏
+頇
+埡
+撾
+撻
+賁
+壋
+撏
+莢
+貰
+蓽
+蕎
+薈
+薺
+堊
+滎
+犖
+蕁
+藎
+蓀
+蕒
+葤
+櫛
+櫳
+櫨
+櫟
+檉
+酈
+硨
+碸
+殤
+軲
+軻
+轤
+軼
+軫
+蠆
+覘
+瞘
+嘵
+嗶
+噦
+剮
+鄖
+噲
+噥
+嶢
+幀
+嶠
+貺
+鈈
+鈦
+鋇
+鈑
+鈐
+鎢
+鈁
+鈀
+篤
+儔
+儼
+儷
+腖
+臚
+脛
+鴇
+獪
+颮
+猻
+餉
+餄
+餎
+孿
+孌
+癧
+瘲
+颯
+闥
+閭
+闓
+閡
+熾
+烴
+浹
+澮
+滸
+潯
+濜
+慟
+懨
+愷
+惻
+惲
+誚
+禰
+誥
+誑
+鴆
+婭
+嬈
+懟
+絝
+驍
+驊
+絎
+絳
+駢
+頊
+璫
+琿
+塒
+塤
+堝
+贄
+蒔
+萵
+蕕
+鴣
+蒓
+橈
+楨
+榿
+檜
+邐
+礪
+礱
+軾
+輊
+輅
+鶇
+躉
+齔
+鸕
+矓
+嘜
+鴞
+蜆
+嗩
+嶗
+崍
+覬
+賅
+鈺
+鉦
+鈷
+鉢
+鈸
+鉞
+鉭
+鉬
+鈿
+鈾
+鉑
+鑠
+鉚
+鈰
+鉉
+鉈
+鉍
+鈮
+鈹
+鏺
+鐸
+氬
+筧
+頎
+徠
+膾
+鴟
+璽
+鴝
+獫
+裊
+餑
+欒
+攣
+癰
+痙
+頏
+閫
+鬮
+誾
+閬
+鄲
+燁
+燴
+燼
+淶
+漣
+潿
+慳
+諏
+諑
+禎
+諉
+諛
+諗
+諂
+誶
+媧
+嫻
+綆
+驪
+綃
+騁
+綏
+縧
+綈
+駸
+鷥
+燾
+璉
+麩
+擄
+摑
+鷙
+撣
+慤
+摜
+縈
+槤
+覡
+欞
+嗇
+匱
+硤
+磽
+鴯
+龔
+殞
+殮
+賚
+輒
+塹
+嘖
+囀
+嚙
+蹌
+蠣
+蠱
+蟶
+幘
+幗
+賕
+賑
+賒
+銠
+鉺
+鋏
+鐃
+銦
+鎧
+鍘
+銖
+銑
+鋌
+鏵
+銓
+鎩
+鉿
+銚
+鉻
+錚
+銫
+鉸
+銥
+銃
+銨
+銣
+鴰
+穠
+箋
+籩
+僨
+僂
+皚
+鴴
+艫
+龕
+玀
+獼
+餜
+餛
+鸞
+闍
+閾
+閹
+閶
+鬩
+閽
+閼
+羥
+糲
+燜
+漬
+瀆
+澠
+愜
+憚
+諶
+諫
+皸
+謔
+襠
+謁
+諤
+諭
+諼
+讒
+諳
+諦
+諞
+糶
+嬋
+綾
+騏
+綺
+緋
+緔
+騍
+緄
+騅
+綬
+綹
+綣
+綰
+驂
+緇
+靚
+輦
+黿
+頡
+撳
+蟄
+壪
+蔞
+櫝
+欏
+賫
+鵓
+鸝
+殫
+輥
+輞
+槧
+輟
+輜
+瞼
+躒
+蛺
+蟯
+螄
+蠐
+嘍
+嶸
+嶁
+賧
+鋙
+錸
+鏗
+鋥
+鋰
+鋯
+鋨
+銼
+鐧
+銻
+鋃
+鋦
+錒
+犢
+鵠
+篳
+牘
+儻
+儐
+儺
+嬃
+頜
+鵒
+魷
+魨
+魴
+潁
+颶
+觴
+熲
+餷
+餿
+褻
+臠
+癆
+癇
+賡
+頦
+鷳
+闌
+闃
+闋
+鵜
+憒
+嚳
+謨
+褳
+襇
+讜
+謖
+謚
+謐
+騭
+巰
+翬
+騖
+緙
+緗
+緘
+緹
+緲
+緦
+緱
+縋
+緡
+饗
+耮
+驁
+韞
+攄
+擯
+轂
+驀
+鶓
+薊
+蘺
+鎣
+頤
+櫚
+櫸
+磧
+磣
+鵪
+輳
+齟
+齙
+韙
+囁
+躂
+蹕
+躚
+躋
+噯
+鍺
+錛
+錡
+鍀
+錁
+錕
+錮
+鍁
+錈
+錠
+錙
+覦
+頷
+鮁
+鮃
+鮎
+鱸
+穌
+鮒
+鮐
+鵮
+颼
+饈
+鶉
+瘮
+闔
+闐
+闕
+灧
+瀅
+潷
+灤
+澦
+懾
+鱟
+騫
+竇
+謾
+謫
+嬡
+嬪
+縉
+縝
+縟
+轡
+騮
+縞
+縭
+縊
+縑
+騸
+覯
+韜
+靉
+攖
+薔
+藺
+鶘
+檳
+櫧
+釅
+殯
+霽
+轅
+齜
+齦
+瞜
+曖
+躊
+蟈
+鶚
+嚶
+羆
+賻
+罌
+鶻
+鍥
+鍇
+鍶
+鍔
+鍤
+鏘
+鎂
+鏤
+簀
+篋
+簞
+籙
+臏
+鮭
+鮪
+鱭
+鮫
+鱘
+饉
+鑾
+瘻
+闞
+鮝
+糝
+鷀
+瀲
+濰
+譖
+褸
+譙
+讕
+譎
+鶥
+嬙
+鶩
+驃
+縹
+縵
+縲
+纓
+驄
+繆
+繅
+耬
+瓔
+擷
+擼
+攛
+聵
+覲
+韃
+鞽
+蘄
+賾
+檣
+靨
+魘
+饜
+轆
+齬
+齪
+覷
+顒
+躓
+躑
+蠑
+螻
+顎
+嚕
+顓
+鑷
+鎘
+鎸
+鎳
+鎦
+鎰
+鎵
+鑌
+簣
+鷂
+鯁
+鱺
+鰱
+鰹
+鰣
+鯀
+鯇
+觶
+饊
+饌
+齏
+讞
+襤
+譫
+屨
+纈
+繕
+繒
+驏
+擻
+顳
+顢
+藪
+櫓
+櫞
+贋
+飆
+鏨
+轔
+蟎
+鐯
+鏢
+鏜
+鏝
+鏰
+鏞
+鏑
+鏃
+鏐
+氌
+穡
+魎
+鯪
+鯡
+鯤
+鯧
+鯝
+鯢
+鯛
+鯔
+獺
+鷓
+贇
+癭
+斕
+瀨
+顙
+繾
+繰
+繯
+蘚
+鷯
+齲
+齷
+躡
+蹣
+羈
+鐔
+鐝
+鐐
+鐓
+鑭
+鑹
+鏹
+鐙
+籪
+鷦
+鱝
+鰈
+鯷
+鰓
+鰍
+鰉
+鯿
+鷲
+懣
+鷸
+鰲
+韉
+顥
+鷺
+䴉
+髏
+鑊
+鐳
+鐲
+讎
+鰨
+鰥
+鰩
+癩
+攢
+靄
+躥
+髖
+髕
+鑔
+籟
+鰳
+鰾
+鱈
+鰻
+鱅
+讖
+驥
+纘
+瓚
+鼉
+黷
+黲
+鑣
+鑞
+臢
+鱖
+鱔
+鱒
+驤
+顰
+鱧
+癲
+灝
+鸛
+鑱
+趲
+顴
+躦
+饢
+戇
+戔
+訏
+訒
+釓
+俔
+閆
+澫
+訢
+訩
+詝
+紃
+纊
+瑒
+剗
+塸
+壢
+埨
+撝
+蔿
+榪
+軑
+軏
+咼
+㠣
+覎
+㑳
+颺
+閌
+潕
+湋
+澐
+浿
+諓
+禡
+詗
+詘
+詖
+屓
+彄
+紘
+馹
+馼
+紵
+紞
+駃
+紖
+瑲
+薴
+棡
+軝
+暐
+晛
+崬
+釴
+釤
+鍆
+鍚
+鄶
+獮
+飿
+嶨
+詷
+詪
+鄩
+鳲
+隑
+隮
+娙
+逕
+駓
+駔
+駉
+絅
+騶
+䮄
+紼
+紿
+瓅
+韍
+墶
+塏
+薘
+蕘
+蔄
+葒
+鳾
+龑
+軹
+軤
+轢
+軺
+睍
+曨
+噠
+鈃
+鈇
+鉅
+鋹
+釿
+錀
+鈧
+鈥
+鈄
+倈
+艤
+鶬
+颭
+餏
+湞
+溮
+滻
+褘
+絰
+駰
+絪
+駪
+綎
+綖
+驫
+勣
+璕
+𡑍
+䓣
+薟
+藭
+椏
+梜
+頍
+硜
+輄
+輈
+輇
+貲
+嗊
+曄
+暉
+鄳
+幬
+輋
+嶮
+贐
+鉥
+鉕
+鑪
+鉮
+鉊
+鉧
+僤
+鴒
+魛
+餗
+燖
+溳
+礐
+窵
+襏
+駼
+絺
+綌
+騂
+綄
+璡
+墠
+壼
+聹
+蘀
+勩
+罃
+檮
+棶
+厴
+䃮
+磑
+礄
+鴷
+齕
+頔
+廼
+凢
+亾
+枒
+屍
+匃
+匄
+紥
+紮
+疋
+殀
+讐
+觔
+兇
+宂
+㕥
+㠯
+栞
+佈
+佔
+呌
+敂
+冄
+坵
+僊
+怱
+悤
+冊
+夘
+戼
+牠
+妳
+嬭
+摃
+釦
+攷
+託
+衺
+衕
+弔
+喫
+囙
+㠶
+颿
+秊
+倣
+髣
+佀
+朶
+氷
+決
+併
+並
+竝
+汙
+汚
+異
+姦
+廵
+挵
+衖
+搤
+阯
+撦
+埳
+阬
+誌
+㕁
+卻
+刦
+刧
+刼
+芲
+蘤
+桿
+槓
+荳
+獃
+唫
+脗
+皁
+彿
+髴
+疘
+刪
+鉋
+鑤
+況
+牀
+恡
+棄
+洶
+汎
+災
+烖
+菑
+禩
+侷
+跼
+坿
+玅
+姉
+妬
+翫
+搨
+柺
+拕
+牴
+觝
+倖
+抝
+盃
+桮
+傑
+逩
+肎
+菓
+崐
+崑
+呪
+虖
+嘑
+謼
+詠
+㟁
+嵒
+巗
+巖
+雰
+稈
+咊
+嶽
+妷
+姪
+廹
+徃
+餚
+採
+寀
+唸
+週
+昬
+兎
+兔
+亯
+亱
+䘚
+淨
+劵
+匟
+㳒
+灋
+洩
+霑
+淚
+註
+恠
+箒
+屆
+絃
+圅
+旾
+珎
+掛
+垜
+艸
+茘
+査
+栢
+柵
+栁
+桺
+柹
+韮
+揹
+昰
+閧
+鬨
+冐
+暎
+嚥
+倃
+𠴰
+偺
+喒
+齩
+欬
+榘
+㑺
+儁
+敍
+敘
+肧
+脈
+䘑
+衇
+跡
+蹟
+砲
+礮
+薙
+鬀
+恆
+怳
+卹
+䘏
+賉
+婣
+畊
+揑
+綑
+輓
+恥
+躭
+晉
+棲
+覈
+慄
+翄
+脣
+槕
+㨪
+螡
+蟁
+㤙
+陗
+峩
+峯
+乗
+椉
+咲
+筍
+俛
+頫
+勌
+䠶
+躳
+慇
+拏
+㧱
+挐
+脃
+胷
+肐
+貍
+㽞
+畱
+淒
+悽
+蓆
+効
+傚
+涼
+缾
+菸
+煙
+淛
+湧
+誖
+猂
+醼
+讌
+㝠
+寃
+孃
+桒
+毬
+瑠
+璢
+瑯
+㨗
+搥
+搯
+蔆
+惏
+楳
+槑
+捄
+廂
+慽
+慼
+瞇
+埜
+畧
+虵
+稭
+棃
+犂
+迻
+媮
+兠
+舩
+慾
+綵
+腳
+𩓐
+夠
+豬
+貓
+湊
+減
+庻
+蔴
+菴
+朢
+睠
+觕
+麤
+釬
+銲
+痳
+殽
+婬
+滛
+湻
+㴱
+樑
+顇
+㝛
+窰
+窯
+琹
+欵
+墖
+趂
+隄
+愽
+揷
+揫
+煑
+朞
+㪚
+塟
+蔥
+蔕
+稜
+棊
+碁
+椶
+偪
+㕑
+廚
+廈
+鴈
+冣
+㝡
+晳
+鼃
+餧
+餵
+嗁
+諠
+㡌
+賸
+筴
+筞
+筩
+栰
+暠
+皜
+踰
+蝟
+㪟
+燄
+遊
+媿
+嘅
+庽
+窓
+牎
+牕
+窻
+徧
+僱
+帬
+裠
+強
+彊
+疎
+壻
+瓌
+䰟
+皷
+擕
+㩗
+㩦
+攜
+懃
+鞾
+幙
+㮣
+酧
+詶
+醻
+掽
+踫
+㼝
+盌
+磟
+覩
+倸
+㬉
+煗
+煖
+晻
+闇
+炤
+跥
+䗬
+蠭
+寘
+辠
+稺
+穉
+燬
+譭
+瘉
+癒
+顋
+骽
+猨
+蝯
+稟
+痺
+癡
+亷
+㢘
+韻
+泝
+遡
+昚
+躶
+臝
+羣
+㬪
+曡
+疊
+勦
+琍
+瓈
+𤋮
+熈
+牓
+搾
+謌
+堿
+鹻
+鹼
+矁
+燻
+髈
+𤺥
+辢
+旂
+𡚁
+潄
+砦
+詧
+嫰
+櫈
+撐
+墪
+譔
+鞵
+鞌
+蕋
+橤
+蘂
+醕
+譆
+跴
+蹤
+蜨
+蠍
+稾
+殭
+惪
+厀
+襃
+癅
+䊀
+餬
+潛
+癄
+顦
+鷰
+藷
+櫥
+螎
+蹏
+蟇
+譟
+簒
+彫
+琱
+鵰
+餹
+餻
+簷
+粦
+燐
+緐
+幑
+蹧
+粇
+穅
+臋
+籐
+繙
+飜
+孼
+蠏
+燿
+蝡
+稬
+穤
+惷
+覇
+鑵
+戹
+阨
+剳
+帀
+巵
+亙
+佇
+竚
+穽
+岅
+虯
+𦍑
+羗
+啎
+姙
+㘭
+袟
+袠
+逈
+㒺
+犛
+氂
+偘
+甕
+罋
+冺
+姍
+蝨
+琺
+瑇
+尅
+梔
+斮
+斲
+斵
+暱
+毘
+蝱
+吚
+哶
+峝
+粃
+竢
+狥
+秈
+烱
+㳄
+袵
+盇
+涖
+蒞
+碪
+蠔
+唕
+倐
+儵
+雋
+皐
+臯
+衂
+䶊
+臙
+獧
+痾
+皰
+湼
+澣
+濬
+塚
+襢
+娿
+勅
+勑
+戞
+廐
+廄
+眥
+覜
+勗
+啗
+噉
+傯
+挱
+㥫
+惥
+慂
+陻
+蕚
+萲
+蕿
+蘐
+藼
+櫂
+箠
+槨
+啑
+蹠
+蚘
+痐
+蛕
+蜖
+瘖
+遯
+醃
+飱
+冪
+簑
+枏
+柟
+檝
+楥
+矴
+椗
+嘷
+獋
+粺
+䈰
+諐
+齶
+堘
+疿
+雝
+秔
+稉
+槀
+搉
+廝
+叡
+嘠
+蜋
+筯
+篛
+麞
+糉
+緥
+璿
+髥
+臕
+餈
+剹
+橜
+罇
+蜺
+矙
+憇
+翺
+饍
+瞖
+羴
+羶
+爕
+繦
+騌
+鬉
+騣
+蔾
+䠀
+簮
+躕
+蹵
+䝔
+貛
+鼴
+麐
+塡
+あ
+い
+う
+え
+お
+か
+き
+く
+け
+こ
+さ
+し
+す
+せ
+そ
+た
+ち
+つ
+て
+と
+な
+に
+ぬ
+ね
+の
+は
+ひ
+ふ
+へ
+ほ
+ま
+み
+む
+め
+も
+や
+ゆ
+よ
+ら
+り
+る
+れ
+ろ
+わ
+を
+ん
+が
+ぎ
+ぐ
+げ
+ご
+ざ
+じ
+ず
+ぜ
+ぞ
+だ
+ぢ
+づ
+で
+ど
+ば
+び
+ぶ
+べ
+ぼ
+ぱ
+ぴ
+ぷ
+ぺ
+ぽ
+ぁ
+ぃ
+ぅ
+ぇ
+ぉ
+っ
+ゃ
+ゅ
+ょ
+ゎ
+ゕ
+ゖ
+ア
+イ
+ウ
+エ
+オ
+カ
+キ
+ク
+ケ
+コ
+サ
+シ
+ス
+セ
+ソ
+タ
+チ
+ツ
+テ
+ト
+ナ
+ニ
+ヌ
+ネ
+ノ
+ハ
+ヒ
+フ
+ヘ
+ホ
+マ
+ミ
+ム
+メ
+モ
+ヤ
+ユ
+ヨ
+ラ
+リ
+ル
+レ
+ロ
+ワ
+ヲ
+ン
+ガ
+ギ
+グ
+ゲ
+ゴ
+ザ
+ジ
+ズ
+ゼ
+ゾ
+ダ
+ヂ
+ヅ
+デ
+ド
+バ
+ビ
+ブ
+ベ
+ボ
+パ
+ピ
+プ
+ペ
+ポ
+ァ
+ィ
+ゥ
+ェ
+ォ
+ッ
+ャ
+ュ
+ョ
+ヮ
+ヵ
+ヶ
+ヷ
+ヸ
+ヹ
+ヺ
+・
+ー
+ヽ
+ヾ
+ヿ
+ｱ
+ｲ
+ｳ
+ｴ
+ｵ
+ｶ
+ｷ
+ｸ
+ｹ
+ｺ
+ｻ
+ｼ
+ｽ
+ｾ
+ｿ
+ﾀ
+ﾁ
+ﾂ
+ﾃ
+ﾄ
+ﾅ
+ﾆ
+ﾇ
+ﾈ
+ﾉ
+ﾊ
+ﾋ
+ﾌ
+ﾍ
+ﾎ
+ﾏ
+ﾐ
+ﾑ
+ﾒ
+ﾓ
+ﾔ
+ﾕ
+ﾖ
+ﾗ
+ﾘ
+ﾙ
+ﾚ
+ﾛ
+ﾜ
+ｦ
+ﾝ
+ﾞ
+ﾟ
+ｧ
+ｨ
+ｩ
+ｪ
+ｫ
+ｯ
+ｬ
+ｭ
+ｮ
+円
+気
+糸
+絵
+楽
+帰
+戸
+広
+黒
+図
+線
+読
+売
+歩
+毎
+亜
+悪
+圧
+扱
+囲
+為
+壱
+隠
+栄
+営
+駅
+塩
+縁
+艶
+応
+桜
+穏
+仮
+価
+箇
+ゑ
+ゝ
+ゞ
+ヰ
+ヴ
+㈱
+両
+丼
+丿
+亀
+仏
+伝
+侶
+俤
+値
+倶
+倹
+偐
+偽
+働
+儛
+兌
+児
+冑
+冨
+凞
+処
+凪
+別
+剣
+剤
+剰
+劔
+労
+勧
+勲
+匁
+匂
+匲
+卍
+単
+厳
+収
+呂
+呉
+呑
+呰
+唖
+喚
+喩
+喰
+噛
+噺
+嚢
+囃
+団
+圀
+圏
+堀
+堺
+塀
+塁
+塙
+増
+墺
+壊
+壌
+壷
+変
+奨
+姫
+娯
+嫐
+嬢
+嬾
+孁
+宍
+実
+宮
+寔
+寛
+対
+専
+尭
+峠
+崋
+嶋
+巀
+巌
+巣
+巻
+帯
+幇
+庁
+廃
+廻
+弉
+弌
+弐
+弖
+弾
+従
+徳
+徴
+忯
+恵
+悩
+惣
+懐
+懽
+戦
+戯
+戻
+払
+抜
+択
+拝
+拠
+拡
+拵
+挙
+挿
+捗
+捜
+掟
+掲
+掻
+揃
+換
+揺
+摂
+撃
+撹
+斉
+斎
+旛
+旡
+晧
+晩
+暁
+暦
+曽
+杁
+杢
+杣
+杮
+枓
+枠
+枡
+柾
+栂
+栃
+桝
+桟
+桾
+梛
+梱
+梲
+梶
+椙
+検
+椥
+楕
+楡
+楢
+榊
+榎
+槇
+様
+槙
+槻
+樋
+権
+樫
+橿
+檥
+欅
+歎
+歓
+歯
+歳
+歴
+毀
+沖
+沢
+浄
+涙
+済
+渉
+渋
+渓
+渕
+満
+滝
+漑
+潅
+澁
+瀞
+瀬
+焔
+焼
+煇
+煕
+煥
+燗
+爼
+犠
+狛
+猟
+獏
+獣
+珊
+瑤
+甞
+畑
+畠
+畳
+畷
+畺
+痩
+癪
+発
+県
+眞
+砕
+碕
+礒
+禖
+禿
+稲
+穂
+穣
+竃
+竜
+竴
+笹
+筈
+筬
+筰
+箆
+箏
+箙
+篠
+篭
+簺
+籾
+粂
+粋
+粛
+粧
+糺
+紬
+絁
+経
+絖
+絣
+絽
+継
+続
+綟
+総
+縄
+縅
+縒
+縦
+繊
+繋
+繍
+繝
+繧
+纐
+纒
+罠
+罧
+罵
+羂
+羇
+羨
+聟
+聡
+聨
+聴
+脇
+脳
+膣
+膵
+臈
+臓
+臥
+舎
+舖
+舗
+舘
+芿
+苅
+茲
+荊
+荘
+莬
+莵
+菫
+萠
+蔵
+薗
+薫
+薬
+薭
+蘊
+蛍
+蝋
+蝿
+蟷
+衞
+衵
+袙
+袞
+袰
+袴
+袿
+裃
+裡
+裲
+褄
+褌
+襴
+襷
+覗
+覚
+覧
+観
+訳
+証
+諌
+諚
+諟
+諡
+諮
+譛
+譲
+讃
+豅
+豊
+豎
+賎
+賛
+贔
+躙
+躰
+転
+軽
+輌
+辥
+辺
+辻
+込
+逓
+遅
+遙
+邉
+郷
+酔
+醗
+醤
+醸
+釈
+鉄
+鉇
+鉤
+鉱
+鉾
+銈
+銕
+銭
+鋲
+鋳
+鋺
+錆
+錍
+錣
+錬
+錵
+鍑
+鍮
+鍼
+鎌
+鎗
+鎚
+鎹
+鐇
+鐚
+鐡
+鑁
+鑑
+鑚
+鑢
+閇
+関
+閦
+闘
+陥
+険
+隣
+隷
+雑
+雫
+霊
+靜
+靫
+靭
+靱
+鞄
+鞆
+頚
+頬
+頴
+頼
+顕
+顗
+餝
+饂
+駄
+駆
+駈
+騒
+験
+騨
+髄
+髙
+髪
+髷
+鯖
+鯰
+鯱
+鰒
+鰯
+鰰
+鳰
+鴎
+鴫
+鵄
+鵞
+鵺
+鶏
+鹸
+麁
+麺
+麿
+黌
+黙
+鼈
+齢
+龗
+縯
+蟅
+坖
+祂
+鼂
+鱚
+蛻
+屌
+呾
+煔
+吶
+扥
+蚖
+銂
+尃
+夋
+鵼
+徬
+寳
+彡
+舨
+湳
+麼
+鍈
+崈
+鱣
+盺
+拺
+瑥
+茷
+焻
+奀
+驎
+鱰
+砢
+痟
+廱
+僜
+瘺
+鱊
+擥
+嶰
+淓
+跅
+浵
+媗
+璦
+煠
+檊
+媃
+峅
+躄
+鉟
+塽
+蟴
+鯮
+弍
+烒
+鵵
+妑
+孋
+蚡
+恊
+輭
+廞
+產
+曅
+盜
+騤
+囪
+鱀
+茇
+葊
+逹
+狓
+崢
+趖
+凃
+羙
+鮸
+昞
+楿
+渽
+圗
+麪
+屇
+鍉
+葝
+沯
+爭
+幵
+筭
+寊
+銋
+貮
+鎭
+熺
+昜
+鍱
+墬
+愒
+磺
+嚈
+稘
+珮
+釆
+殑
+鍩
+䲁
+蕷
+鐿
+僡
+佹
+輶
+冴
+襶
+賔
+猙
+辧
+絛
+磾
+韁
+螔
+譳
+礑
+鋱
+魩
+嚗
+棆
+牆
+敟
+柶
+瓛
+魣
+巎
+轘
+襌
+枼
+鸌
+逺
+錏
+縡
+帢
+騄
+媼
+埅
+鄤
+萐
+祙
+旼
+詥
+鶲
+燉
+卲
+銱
+庲
+伱
+氽
+嵿
+挻
+煵
+窋
+鐤
+鮊
+鱬
+鰧
+嬤
+譞
+諲
+脭
+悳
+崘
+阭
+內
+袾
+冚
+壐
+咗
+礠
+孮
+痲
+埈
+肹
+鰮
+鮓
+濊
+塜
+凜
+蒢
+噰
+桼
+峍
+焴
+鶒
+鋮
+綠
+鶹
+熿
+毴
+咟
+嘥
+睺
+繡
+郎
+瘞
+鉶
+蔎
+秠
+緤
+蝀
+躝
+蟜
+繃
+囮
+墫
+乭
+胊
+濙
+瘓
+榣
+鑛
+鐫
+嶴
+甹
+坮
+銾
+蒭
+睜
+俋
+餠
+榢
+蓳
+盋
+堷
+鍏
+苝
+巛
+蚵
+暏
+熤
+嬨
+墎
+鏽
+戶
+菺
+膮
+熖
+睪
+栜
+捱
+榗
+鍷
+曧
+犽
+韑
+袓
+䖝
+焄
+喦
+髲
+疌
+㴪
+侊
+貐
+蕅
+禠
+蕑
+囯
+暊
+儞
+佋
+柎
+㐱
+鰤
+苳
+鱥
+謤
+遶
+眀
+鑀
+羋
+顏
+陜
+銩
+黶
+苼
+蒤
+棛
+儫
+咁
+抦
+衚
+棩
+焿
+脫
+麅
+玏
+埧
+淸
+黁
+淽
+彠
+鮨
+沜
+糀
+厓
+楧
+嶌
+簹
+檵
+鱇
+嶬
+廸
+卽
+樀
+贌
+酼
+籛
+沒
+晸
+諪
+蕡
+妏
+鄋
+蒍
+奧
+抇
+蓨
+薆
+鱷
+巘
+䝉
+亰
+寈
+槩
+誒
+麴
+蕟
+溎
+蘗
+榦
+斿
+暟
+炲
+拚
+娖
+繖
+橚
+寜
+爀
+饟
+悅
+鯏
+彜
+眾
+葯
+嬝
+埮
+獇
+馛
+溙
+瀦
+熼
+硓
+鈢
+樆
+輬
+鰜
+蔘
+渙
+澔
+嗮
+旉
+籜
+媊
+燘
+儚
+頹
+缽
+俽
+逨
+鱓
+郞
+歊
+杴
+珡
+杋
+醁
+鰏
+鵾
+鐽
+鮋
+巶
+荅
+薾
+囓
+蹻
+獎
+禑
+鎓
+榲
+僴
+綞
+尓
+敭
+曔
+褔
+鬅
+亊
+鏦
+蓘
+裬
+鱲
+薡
+鰗
+箑
+鬪
+縂
+璸
+甙
+茮
+辵
+岻
+覿
+滈
+鯶
+鑂
+囶
+舺
+溋
+拋
+菾
+敾
+虨
+綝
+蝍
+醂
+禨
+賹
+廧
+絕
+槗
+徫
+鎔
+曮
+蠂
+捒
+堈
+莕
+蓪
+敎
+禃
+櫱
+綧
+瀶
+逌
+浤
+碻
+刄
+逤
+剏
+氹
+菈
+娫
+蜛
+嵗
+糎
+螶
+譓
+鏳
+嵙
+瑊
+隲
+檨
+緈
+畵
+砯
+簗
+彅
+鰺
+騋
+窶
+嚒
+嵻
+尙
+頵
+槰
+虉
+醞
+巂
+彔
+偊
+畇
+鱨
+妸
+塲
+畐
+鈫
+錟
+磪
+摠
+彥
+璙
+囝
+寗
+耎
+鮡
+蘓
+弅
+焃
+飥
+戙
+塰
+儱
+槺
+噏
+魟
+禵
+佧
+咘
+盪
+瑈
+鉲
+睭
+鏌
+鼇
+郋
+魮
+朖
+滽
+渃
+滙
+熯
+醿
+鎅
+褀
+鬬
+巄
+螥
+眜
+釚
+柉
+壎
+峇
+姸
+唭
+鮜
+鈖
+嫈
+壄
+洤
+黃
+伕
+堦
+嶔
+鮰
+鞞
+漎
+鉓
+鮗
+壴
+阝
+妀
+矽
+獢
+倗
+銪
+鴓
+橒
+凈
+哖
+屚
+偍
+瑺
+媯
+淍
+驌
+椇
+赬
+薐
+糹
+碽
+濲
+釭
+晭
+纕
+寖
+閞
+歿
+呎
+鶆
+屄
+櫿
+犎
+旲
+㙟
+龎
+翜
+螾
+說
+衜
+泆
+軎
+鵂
+荎
+嚧
+硂
+桖
+褭
+筊
+鰷
+秳
+戩
+轀
+鬹
+飬
+卋
+暸
+狦
+搢
+娋
+鏴
+溫
+毉
+淰
+謩
+餺
+鵙
+鳽
+鮀
+狶
+氻
+轝
+妺
+袛
+蓭
+梂
+娛
+牼
+稅
+兿
+玾
+煚
+僩
+鶿
+鬄
+崠
+鉆
+鯓
+蚢
+庀
+鵟
+坣
+殼
+悞
+熅
+敻
+鍠
+曶
+愼
+搳
+姃
+砳
+槼
+臞
+韾
+靑
+鸊
+薲
+虛
+蠄
+啟
+鶺
+苺
+滾
+褞
+仺
+胇
+憻
+郳
+烉
+驩
+冇
+枖
+夌
+搵
+匸
+盨
+櫾
+霤
+麊
+貒
+噓
+嗢
+笩
+晈
+冂
+銳
+毿
+慜
+囧
+閜
+娸
+庢
+壆
+馯
+桱
+兗
+葃
+侅
+煐
+鐦
+藸
+鷎
+嵰
+逎
+弒
+匋
+鐭
+廔
+砩
+孆
+灴
+伷
+兪
+鴗
+澯
+幚
+旙
+勻
+礽
+婑
+鱮
+娍
+銶
+吳
+鍟
+仼
+鳧
+彞
+娽
+昛
+鰼
+剎
+佉
+鉏
+偸
+鰆
+讙
+橪
+啱
+岀
+孻
+釪
+乹
+鈳
+漇
+檦
+埻
+祿
+爌
+禇
+鱵
+㸃
+梉
+燝
+霙
+炁
+飮
+蠙
+勷
+鵎
+儥
+鐠
+唻
+廰
+嚿
+嵕
+墱
+紑
+搖
+瘜
+皝
+鸑
+瀁
+粵
+撚
+巑
+梀
+啯
+眛
+諴
+夊
+僙
+鍝
+裖
+鮣
+凬
+飡
+灊
+橓
+嫳
+筳
+咑
+粍
+瓑
+璌
+伃
+閰
+傜
+黐
+謢
+驒
+橫
+蛯
+寕
+蠵
+瞓
+旳
+翏
+硏
+寯
+韡
+楤
+鰃
+朿
+侞
+鵯
+愨
+祹
+厔
+丌
+盩
+謏
+魕
+啣
+閱
+曺
+枛
+罉
+卐
+樻
+鷉
+鯒
+鋡
+磱
+枱
+攴
+蠷
+穈
+嚟
+檽
+趐
+奐
+鋐
+檇
+薀
+峼
+咭
+訔
+韠
+鑴
+鸐
+唃
+捦
+鸜
+誴
+罳
+璄
+暃
+夀
+賨
+鞥
+鈊
+灡
+鮍
+懮
+籣
+昐
+陁
+襾
+鮠
+鈏
+囍
+婯
+艔
+貭
+䰾
+姁
+禼
+堖
+鋶
+仛
+鏷
+謜
+鑅
+忬
+蘶
+謠
+觙
+奫
+狟
+泩
+桙
+飈
+垰
+啍
+嚞
+鯕
+蒧
+榞
+徸
+璹
+揔
+欉
+魞
+菶
+玧
+鳯
+廍
+侚
+岰
+岧
+鋕
+凵
+彣
+崱
+媜
+倢
+鵐
+砋
+鷚
+鱠
+鮻
+繻
+摵
+贓
+磵
+錻
+痠
+粩
+胅
+奣
+塨
+瀠
+鸘
+啚
+娳
+霶
+壔
+峚
+甂
+廁
+覌
+鰂
+猳
+鱻
+盫
+裿
+杬
+歛
+澋
+蘞
+嵜
+尐
+旽
+鉌
+鎛
+豿
+凖
+榤
+禓
+龝
+悧
+鷟
+鮟
+吋
+喢
+岪
+吥
+漵
+頠
+豔
+巿
+鑨
+醣
+熳
+懍
+湥
+檡
+韺
+戱
+緖
+鐈
+凉
+緃
+鮹
+媐
+爯
+巆
+褍
+鐬
+昍
+扙
+鍳
+芛
+蟳
+嬅
+糬
+吔
+塭
+譿
+冧
+鏓
+嶪
+嗹
+椵
+姀
+閿
+褧
+錞
+玆
+笘
+篔
+萡
+鶡
+螐
+鮄
+鰟
+脷
+啲
+杤
+蓚
+尗
+娎
+殟
+淥
+蝚
+蓧
+彐
+嚤
+銍
+囒
+坶
+淩
+鶼
+鱂
+喼
+燫
+肏
+姵
+廌
+禟
+籝
+迵
+嵨
+堮
+蟌
+憍
+廕
+蜑
+緁
+唘
+竩
+崙
+璚
+粄
+栨
+罈
+梫
+貤
+藔
+蜯
+訁
+斖
+煶
+馦
+妠
+閟
+疕
+夆
+鎪
+膥
+澻
+嘢
+嚐
+靁
+鎻
+鰛
+穵
+烋
+縕
+褎
+疒
+壠
+溼
+圂
+咅
+鯭
+鯙
+磘
+玨
+珤
+朊
+蚼
+濶
+薞
+嚩
+丟
+嫺
+鯻
+椲
+鰕
+刂
+蠘
+踎
+瀴
+琁
+鰶
+瑴
+肜
+㐂
+欥
+媺
+竻
+讚
+𣇉
+裵
+緜
+廩
+齧
+叄
+俌
+厰
+滀
+錄
+鷫
+鯗
+攞
+姌
+蔝
+幷
+縤
+屻
+鯃
+雞
+纁
+嫲
+嵮
+屭
+嶃
+跩
+鋗
+蕢
+篊
+俬
+淎
+暻
+鏻
+憓
+玗
+溈
+笭
+糢
+勳
+閒
+沍
+咾
+鉷
+蘵
+俁
+崵
+毸
+苪
+掙
+鴡
+萭
+俴
+屜
+蒾
+艹
+剷
+慍
+朮
+枴
+氳
+猓
+甽
+箝
+譁
+贗
+迆
+鈽
+鍊
+鍰
+鏍
+靦
+餽
+丮
+丱
+仜
+仩
+伬
+伔
+仱
+伀
+伻
+佢
+佒
+侀
+侇
+佷
+佌
+佪
+侐
+侜
+俓
+侲
+俉
+侻
+侳
+俇
+倅
+倇
+倰
+倛
+倳
+倷
+俷
+倠
+偯
+偞
+偠
+偋
+偝
+偛
+偢
+偅
+偟
+偩
+偫
+傛
+傔
+傞
+傋
+傌
+傎
+傝
+偨
+傂
+傽
+傿
+僆
+傮
+僄
+僈
+傰
+僁
+傱
+僋
+僗
+僛
+僪
+僝
+僓
+僿
+儃
+儰
+僸
+僶
+僾
+儌
+僽
+儜
+儓
+儗
+儑
+儢
+儤
+儠
+儸
+儹
+儽
+冓
+冘
+冞
+凊
+凅
+凔
+刌
+刉
+刓
+刜
+刞
+刵
+刲
+剆
+刱
+剉
+剚
+剒
+剫
+剭
+剬
+剺
+剸
+剻
+剼
+劀
+劋
+劖
+劘
+劗
+劙
+劦
+勴
+匊
+匢
+匰
+匴
+匷
+匽
+卌
+卼
+厎
+厒
+厗
+厞
+厜
+厤
+厬
+厹
+吰
+吷
+吪
+呿
+咈
+呫
+呺
+呥
+呬
+呴
+茍
+咷
+咮
+咶
+哅
+咠
+咢
+唦
+唗
+唒
+哤
+唚
+唈
+哫
+唅
+唴
+啢
+唶
+啒
+啅
+唌
+唲
+喨
+喥
+喭
+噅
+喓
+喣
+啽
+喌
+嗃
+嗛
+嗋
+嗀
+喿
+喍
+嗏
+嗕
+嗈
+嘕
+嘒
+嗼
+嘐
+嘓
+嘂
+嗺
+嘝
+嘄
+嗿
+噈
+噊
+噆
+噚
+嘳
+嘽
+嘾
+噮
+噳
+噣
+噭
+噞
+嚌
+嚍
+嚃
+嚘
+嚜
+嚫
+嚪
+嚬
+嚲
+嚵
+嚽
+嚾
+囆
+囅
+囋
+囗
+圁
+圞
+圠
+坁
+坅
+坲
+坱
+垀
+坴
+垗
+垝
+垔
+垘
+垽
+垼
+埢
+埶
+堩
+堣
+塈
+堥
+塓
+塉
+塯
+塕
+塼
+墆
+塿
+塴
+墋
+塺
+墝
+墯
+壈
+墽
+壖
+壝
+壛
+壾
+壿
+夃
+夎
+夒
+夗
+奅
+奊
+奰
+奲
+奼
+妦
+妎
+妢
+妐
+妵
+姏
+姎
+㚷
+姡
+姺
+姼
+娭
+婐
+婟
+婥
+婓
+婗
+媔
+媟
+媢
+婸
+媦
+媥
+媬
+媕
+娷
+嫇
+嫋
+媰
+媻
+嫮
+嫥
+嫢
+嫛
+嫿
+嫴
+嫷
+嫶
+嬎
+嬓
+嬐
+嬲
+嬽
+孈
+屘
+孲
+孷
+宎
+宨
+寪
+寍
+寋
+寑
+寙
+寠
+寱
+尌
+尒
+尟
+尰
+尳
+屖
+屔
+屝
+屧
+屩
+屮
+屴
+岏
+岋
+岉
+岒
+岮
+岤
+岯
+岟
+岝
+峐
+峌
+峞
+峉
+峊
+峬
+峮
+峷
+崝
+崨
+崥
+崏
+崰
+崣
+崷
+嵃
+嵑
+崳
+崺
+嵂
+嵱
+嵣
+嵥
+嵞
+嶀
+嵽
+嶆
+嵺
+嵷
+嶊
+嶉
+嶈
+嵾
+嶕
+嶜
+嶡
+嶚
+嶞
+嶱
+嶩
+嶵
+嶭
+巃
+巏
+巕
+巟
+巹
+帊
+帗
+帟
+帣
+帠
+帤
+帩
+帾
+帴
+幏
+幎
+幓
+幩
+幝
+幠
+幧
+幨
+幦
+幭
+幰
+庂
+庉
+庌
+庈
+庰
+庛
+庣
+庨
+庮
+庪
+庬
+庴
+廅
+廇
+廘
+廗
+廎
+廜
+緳
+廦
+廥
+廮
+廯
+蠯
+廾
+弚
+弝
+弣
+弤
+弮
+弳
+彃
+彉
+彋
+彏
+彯
+彴
+彸
+彾
+徦
+徥
+徯
+徲
+徾
+徿
+忀
+忁
+忔
+忕
+忨
+忣
+忷
+忥
+怭
+怲
+怋
+怴
+怗
+怚
+怞
+怬
+怢
+怐
+怮
+怓
+怷
+怹
+恲
+恞
+恅
+恇
+恉
+恛
+恌
+恀
+恟
+悀
+悁
+悕
+悗
+悇
+悊
+悐
+悾
+悺
+惓
+惤
+惈
+悷
+惉
+悹
+惌
+惢
+惄
+愊
+愖
+愅
+惵
+愓
+惸
+惼
+惾
+慉
+慅
+愶
+愲
+愮
+愯
+愬
+慁
+慞
+慱
+慒
+慓
+慲
+憀
+慴
+慔
+慺
+慛
+憃
+慹
+憱
+憰
+憢
+憉
+憛
+憯
+憟
+憪
+憡
+憝
+憖
+懅
+憴
+懆
+懁
+憿
+憸
+憵
+憼
+懧
+懠
+懥
+懤
+懘
+懭
+懱
+懪
+懰
+懫
+懻
+戁
+戃
+戄
+戉
+戠
+酨
+戺
+扐
+扜
+扤
+扡
+扢
+抆
+抌
+抎
+抏
+扻
+抭
+抴
+拑
+抾
+抪
+抶
+抮
+挍
+挋
+挃
+拫
+拹
+挏
+挌
+拸
+挀
+拲
+捖
+挬
+挶
+揤
+捊
+挼
+挩
+捁
+挴
+捘
+捔
+捥
+掝
+掗
+掫
+掯
+捵
+掜
+捼
+掤
+掔
+掱
+揎
+揥
+揨
+揯
+揊
+揲
+揵
+摡
+揟
+揝
+揜
+揘
+揅
+揱
+搆
+搟
+搕
+搘
+搹
+搷
+搣
+搰
+搊
+搚
+摀
+搧
+搫
+摍
+摝
+摲
+摦
+摎
+摋
+摓
+摐
+摿
+摮
+摰
+撢
+撠
+撗
+撜
+撋
+撊
+撌
+撟
+擗
+擖
+擏
+擉
+撽
+擩
+擣
+擫
+擭
+擨
+擽
+擸
+攇
+攐
+攍
+攌
+攗
+攕
+攓
+攡
+攠
+攦
+攩
+攭
+攲
+攳
+敁
+敊
+敆
+敓
+敧
+敪
+敤
+敜
+敯
+敳
+敶
+敺
+敹
+敿
+斁
+斀
+斄
+斒
+斔
+斞
+斨
+斪
+斻
+旍
+旓
+旚
+旝
+旟
+昲
+昦
+昢
+晇
+晥
+晜
+晼
+晬
+暀
+暆
+暍
+暋
+暡
+暰
+暩
+曀
+曊
+曋
+曏
+曒
+曚
+曣
+曭
+朁
+朅
+朄
+朒
+朘
+朣
+朾
+朹
+朻
+朼
+杅
+杇
+杝
+杗
+枎
+杶
+枆
+枌
+柲
+枺
+枻
+柸
+柀
+柅
+柫
+柤
+柍
+柮
+柣
+柂
+柧
+栚
+桋
+桏
+栱
+栵
+栫
+栭
+栯
+栘
+栔
+梡
+梇
+梐
+桭
+梮
+楖
+梬
+梩
+桵
+梒
+椌
+椄
+棜
+棷
+棳
+棌
+椈
+楰
+棯
+椔
+棸
+楟
+楎
+楱
+楅
+楺
+楈
+楛
+楉
+楬
+椳
+楀
+楄
+楶
+楘
+榶
+槉
+榠
+榬
+榼
+榙
+榩
+榾
+榯
+槄
+榽
+榹
+槥
+槸
+樕
+樠
+槬
+槢
+樛
+樝
+槾
+樧
+槮
+樔
+槷
+橀
+樴
+橉
+橧
+樲
+橨
+橝
+橭
+橶
+樿
+橁
+檍
+檖
+檁
+檟
+橾
+檛
+檓
+檕
+檃
+櫅
+檹
+櫡
+櫠
+櫌
+櫑
+櫙
+櫋
+櫜
+櫐
+櫫
+櫬
+櫰
+櫹
+櫺
+櫼
+欃
+欋
+欈
+欐
+欑
+欘
+欨
+欴
+欯
+欭
+欱
+欶
+欳
+欷
+欿
+歂
+歈
+歍
+歋
+歕
+歔
+歜
+歠
+歭
+歾
+肂
+殈
+殏
+殔
+殗
+殙
+殠
+殥
+殢
+殦
+殧
+殰
+殶
+毃
+毄
+毈
+毇
+毊
+毚
+毞
+毦
+毤
+毨
+毣
+毰
+毲
+毻
+毼
+毾
+氁
+氀
+氄
+氠
+氶
+汃
+汒
+汏
+汍
+汸
+沋
+汱
+汯
+沕
+汦
+汳
+泬
+沶
+沬
+泧
+沷
+泭
+泲
+泒
+沴
+洟
+洊
+洀
+浺
+浶
+洍
+涒
+浘
+浢
+涊
+涆
+浧
+涗
+涳
+涬
+淢
+涷
+淔
+渀
+淈
+涾
+淊
+涽
+淭
+湆
+湇
+湅
+湢
+渿
+湁
+渜
+渳
+湀
+渻
+渮
+湨
+湡
+渱
+渨
+湠
+湱
+湩
+渹
+溛
+滖
+溓
+溔
+滒
+溰
+溾
+滜
+滵
+滱
+漃
+漥
+漮
+潎
+漙
+漧
+漘
+漒
+滭
+漊
+潳
+滮
+潀
+漰
+潃
+漅
+濆
+澒
+澅
+潚
+潠
+澖
+潶
+潬
+潒
+潐
+潗
+澓
+潝
+濇
+濎
+濈
+濄
+澞
+澨
+瀄
+濌
+澩
+濴
+濔
+濣
+濭
+濧
+濦
+瀇
+瀎
+濿
+瀀
+濻
+瀙
+瀖
+瀫
+瀡
+瀢
+瀩
+瀯
+瀷
+灂
+瀸
+瀿
+瀺
+灄
+灉
+灖
+灗
+灛
+灟
+灨
+灩
+灪
+炾
+炰
+烓
+烑
+缹
+焍
+烰
+焠
+焮
+焣
+煆
+煣
+煝
+熐
+熉
+熀
+熂
+熚
+燅
+燂
+熸
+燀
+燡
+爁
+爊
+爂
+爓
+爞
+爢
+爣
+牄
+牉
+牋
+牏
+牣
+牬
+牰
+牸
+牷
+犈
+犉
+犆
+犅
+犌
+犑
+犐
+犗
+犕
+犓
+犘
+犚
+犝
+犞
+犥
+犦
+犤
+犣
+犩
+犪
+犮
+犵
+犿
+狆
+狖
+狋
+狘
+狜
+狔
+狚
+狌
+狑
+狊
+狤
+狫
+狪
+狣
+猀
+狾
+猑
+猘
+猈
+狿
+猏
+猋
+猒
+猧
+猲
+猭
+猦
+猣
+猵
+猼
+獂
+獀
+獊
+獑
+獌
+獘
+獞
+獟
+獝
+獛
+獡
+獩
+獦
+獥
+獳
+獶
+獽
+獿
+玂
+玁
+玈
+玊
+玔
+珓
+珶
+琖
+瑵
+璊
+瑽
+璅
+瑿
+璗
+瓁
+瓋
+瓝
+瓟
+瓡
+瓥
+瓨
+瓬
+瓵
+瓾
+瓽
+甀
+甃
+甈
+甋
+甐
+甒
+甔
+甖
+甝
+甮
+甿
+畟
+畣
+畽
+疀
+疧
+痁
+疻
+痀
+痎
+痏
+痋
+痌
+痑
+痚
+痡
+痝
+痗
+痯
+瘏
+痷
+痸
+痻
+瘈
+瘑
+瘝
+瘣
+瘯
+瘱
+瘽
+癈
+癉
+癙
+癐
+癓
+癠
+癵
+癹
+皊
+皏
+皫
+皯
+皵
+皻
+皽
+皾
+盄
+盓
+盝
+盬
+盭
+盳
+眃
+眅
+盻
+眝
+眐
+眓
+眒
+眣
+眑
+眕
+眹
+眱
+眲
+眴
+眳
+眽
+睆
+睅
+睊
+睋
+睌
+睕
+睟
+睒
+睖
+睩
+睧
+睔
+瞁
+睼
+瞂
+睮
+睯
+瞏
+瞉
+瞚
+瞝
+瞡
+瞛
+瞲
+瞷
+瞶
+瞴
+矂
+矉
+矊
+矌
+矎
+矏
+矐
+矔
+矕
+矘
+矠
+矱
+矲
+矹
+矺
+砅
+砐
+砏
+砎
+砨
+硈
+硉
+硠
+硥
+硱
+硰
+硩
+碔
+碄
+碅
+碆
+硾
+碫
+碞
+磍
+磌
+磎
+磈
+磃
+磝
+磩
+磥
+磞
+磛
+磳
+磼
+磿
+礔
+礉
+礝
+礛
+礜
+礥
+礣
+礧
+礨
+礭
+礿
+祌
+祅
+祔
+祒
+祑
+祤
+祩
+祪
+祣
+祫
+祡
+祴
+祳
+禂
+禗
+禜
+禫
+禭
+禬
+禴
+禷
+禸
+歶
+秅
+秏
+秖
+秎
+秮
+秪
+秺
+秶
+稊
+稒
+稫
+穊
+稰
+稯
+穋
+穛
+穖
+穧
+穨
+穮
+穬
+穭
+穱
+穾
+窆
+窉
+窌
+窏
+窔
+窐
+窙
+窢
+窞
+窫
+窲
+窴
+窱
+窾
+竀
+竁
+竷
+笐
+笓
+笅
+笵
+笻
+笴
+笰
+笢
+笝
+笲
+筄
+筡
+箈
+箊
+箌
+箛
+箎
+箘
+箄
+箷
+箾
+篎
+箯
+箹
+篞
+篣
+篧
+篕
+篨
+篹
+簅
+篲
+篿
+篻
+簎
+篴
+簂
+簁
+篸
+篽
+簜
+簩
+簙
+簭
+簦
+簨
+簢
+簥
+簳
+簼
+簬
+簻
+籉
+籈
+籊
+籔
+籗
+籧
+籦
+籯
+籺
+籸
+籹
+粊
+粔
+粻
+糔
+糪
+糱
+糷
+紎
+紟
+紒
+紽
+紸
+紶
+紩
+絇
+紾
+絘
+絯
+絓
+絧
+絏
+絭
+絫
+綀
+綍
+絿
+綅
+絻
+絼
+綔
+綷
+緂
+綪
+緀
+緅
+緎
+緆
+緌
+綯
+綼
+緷
+緛
+緪
+緧
+縃
+緺
+緶
+緰
+縗
+縌
+縓
+縎
+縜
+縚
+縏
+縼
+繂
+縳
+顈
+繈
+縸
+縪
+繉
+繀
+縩
+緵
+縰
+縿
+縶
+繜
+繐
+繣
+繘
+繢
+繟
+繑
+繠
+繶
+繵
+繸
+繷
+繺
+繲
+繴
+纀
+纇
+纋
+纆
+纑
+纗
+纚
+缿
+罊
+罏
+罜
+罞
+罝
+罛
+罣
+罥
+罦
+罭
+罫
+罬
+罻
+罼
+罺
+罿
+羃
+羉
+羍
+羒
+羜
+羛
+羢
+羠
+羦
+羬
+羭
+羵
+羳
+羷
+羺
+羾
+翋
+翍
+翐
+翑
+翇
+翢
+翣
+翭
+翪
+翨
+翴
+翲
+翽
+翿
+耟
+耞
+耡
+耴
+耾
+耹
+聇
+聈
+聑
+聏
+聝
+肕
+肙
+肒
+肣
+肵
+胘
+胑
+胐
+胕
+胉
+胏
+胹
+胵
+脁
+胻
+脀
+胾
+胔
+脰
+脥
+脤
+脙
+脡
+脕
+脧
+腃
+腏
+腄
+腇
+脽
+腍
+腤
+腷
+腜
+腛
+腢
+腲
+朡
+腞
+腶
+膉
+膆
+膃
+膇
+膍
+膌
+膋
+膟
+膕
+膢
+膱
+膹
+膫
+膰
+膬
+膴
+膲
+臇
+膷
+臄
+臅
+臒
+臐
+臗
+臛
+臡
+臦
+臩
+臮
+臲
+臷
+臸
+臿
+舋
+舑
+舕
+舝
+舡
+舼
+舽
+艀
+艂
+艓
+艒
+艐
+艑
+艕
+艛
+艵
+艼
+芀
+芐
+芅
+芓
+芔
+苀
+芚
+芵
+芧
+芞
+芺
+苙
+苨
+苖
+苬
+苲
+苵
+苶
+茙
+茥
+茿
+茦
+茢
+荂
+茪
+荍
+茖
+茤
+茠
+茩
+茻
+莐
+莣
+莍
+荺
+莤
+荴
+莏
+莁
+荵
+莔
+莃
+莌
+莋
+荾
+莥
+菨
+萒
+菧
+菤
+菆
+菣
+菿
+菋
+菎
+菵
+萉
+菞
+菳
+菕
+蓱
+萿
+葹
+葥
+葀
+葧
+萰
+葍
+葽
+蔇
+葞
+萷
+萺
+萴
+葅
+菙
+葋
+萯
+葂
+葟
+葌
+蓎
+蒬
+蒮
+蒫
+蒪
+蒚
+蒝
+蓌
+蒛
+蒩
+蒘
+蒶
+蒠
+蔤
+蔏
+蔩
+蔉
+蔍
+蔧
+蔜
+蓻
+蓺
+蓴
+蔪
+蓲
+蓷
+蓫
+蔒
+蓩
+蔖
+蓾
+蔨
+蔮
+蔂
+蓶
+蔱
+蓹
+蔠
+蔰
+蕫
+蕍
+蕀
+蕆
+蕄
+蕇
+蕣
+蕛
+蕱
+蕵
+蕮
+蕧
+蕠
+蕦
+蕝
+薃
+薧
+薕
+薠
+薋
+薣
+薚
+蕼
+薉
+蕸
+薎
+薖
+薍
+薝
+薂
+藆
+藀
+藃
+藂
+薵
+薽
+藇
+藄
+藋
+藈
+藅
+薱
+薶
+藒
+藫
+藱
+藙
+藡
+藚
+藗
+藲
+藬
+藘
+藣
+藑
+藰
+蘁
+藾
+蘛
+蘉
+蘌
+蘪
+蘦
+蘟
+蘣
+蘜
+蘙
+蘮
+蘡
+蘠
+蘥
+蘴
+蘳
+蘬
+虀
+蘹
+蘱
+蘻
+蘾
+虃
+虆
+虇
+虈
+虌
+虋
+虙
+虡
+虣
+虩
+虪
+虰
+虭
+虴
+蚑
+蚞
+蚇
+蚗
+蚚
+蚅
+蚥
+蚙
+蚿
+蚷
+蛂
+蛁
+蛅
+蛈
+蚹
+蚳
+蚸
+蛌
+蚻
+蛢
+蛦
+蛓
+蛣
+蛚
+蛪
+蛝
+蛫
+蛜
+蛬
+蛗
+蜄
+蛷
+蜌
+蛖
+蛵
+蜁
+蛶
+蜳
+蝫
+蜙
+蝃
+蜬
+蝁
+蝆
+蜠
+蜲
+蜪
+蜭
+蜼
+蜵
+蝂
+蜦
+蜧
+蜸
+蜤
+蜰
+蝖
+蝷
+蟡
+蝳
+蝔
+蝛
+蝒
+蝑
+蝞
+蝭
+蝪
+蝐
+蝝
+蝬
+蝺
+蝜
+螛
+螏
+螓
+螒
+螁
+螖
+螘
+蝹
+螇
+螑
+螝
+螜
+螚
+螪
+螰
+螹
+螼
+螮
+蟉
+蟃
+蟂
+螷
+螴
+螿
+螸
+蟞
+蟧
+蟦
+蟢
+蟟
+蟤
+蟔
+蟓
+蟭
+蟘
+螤
+蟗
+蟙
+蠁
+蟨
+蠀
+蟺
+蠉
+蠌
+蟼
+蠈
+蟿
+蠗
+蠩
+蠝
+蠛
+蠠
+蠤
+蠜
+蠫
+蠬
+蠨
+蠦
+蠪
+蠥
+蠰
+蠮
+蠳
+蠸
+蠾
+蠽
+蠿
+衁
+衈
+衋
+衧
+衪
+衭
+衶
+袀
+衱
+衯
+袃
+袉
+袕
+袨
+袚
+袑
+袡
+袘
+袧
+袬
+袌
+袺
+裗
+袹
+袸
+裀
+袶
+袽
+袲
+裋
+裍
+裞
+裚
+裷
+裧
+裺
+裮
+裶
+裯
+裻
+褁
+褅
+褋
+褗
+褆
+褖
+褑
+褦
+褮
+褱
+褢
+褩
+褵
+褼
+褾
+襒
+褷
+襂
+褽
+襓
+襋
+襆
+襐
+襛
+襗
+襡
+襘
+襝
+襣
+襭
+襩
+襮
+襳
+襹
+襺
+覂
+覅
+覕
+覛
+覝
+覢
+覤
+覣
+覭
+覮
+覶
+觓
+觤
+觡
+觠
+觢
+觩
+觰
+觬
+觲
+觷
+觺
+觻
+觼
+觾
+訑
+訰
+訧
+訬
+訞
+詍
+訹
+詙
+詀
+詄
+詅
+訿
+誂
+詻
+誃
+誫
+誙
+誋
+諆
+誸
+諔
+諕
+誻
+諀
+諅
+諵
+諝
+諰
+諈
+謞
+謘
+謑
+謋
+謒
+謕
+謍
+謈
+謪
+謧
+謣
+謰
+謵
+譇
+謯
+謱
+謥
+謷
+謦
+譐
+譈
+譊
+譀
+譋
+譕
+譑
+譠
+譪
+譝
+譨
+譣
+譥
+譹
+譸
+譅
+譺
+譻
+譾
+讄
+讂
+讆
+讋
+讔
+讘
+讟
+谹
+谻
+谽
+谾
+豃
+豋
+豍
+豏
+豗
+豜
+豝
+豟
+豥
+豤
+豦
+豭
+豰
+豲
+豱
+豯
+豵
+豷
+豶
+豻
+豽
+貁
+貀
+貄
+貏
+貑
+貕
+貙
+貗
+貜
+貣
+貾
+賌
+賥
+賟
+賙
+賵
+賮
+贆
+贕
+贙
+赨
+赩
+赮
+赸
+趀
+趌
+趎
+趏
+趍
+趓
+趠
+趜
+趡
+趥
+趧
+趬
+趪
+趭
+趫
+趮
+趷
+趹
+跘
+跓
+跍
+跇
+跜
+跕
+跙
+跈
+跰
+跠
+跮
+跦
+跢
+跧
+跲
+跫
+踂
+跿
+踍
+踃
+踇
+踆
+跾
+踠
+踥
+踤
+踡
+踕
+踛
+踖
+踑
+踙
+踧
+踘
+踓
+踳
+踾
+踸
+踼
+蹎
+蹍
+蹓
+蹗
+蹖
+蹞
+蹥
+蹛
+蹡
+蹝
+蹔
+蹸
+蹳
+蹪
+躆
+躈
+躖
+躗
+躟
+躠
+躤
+躣
+躩
+躨
+躽
+軓
+軘
+軞
+軯
+軷
+軦
+軮
+軥
+軵
+軧
+軨
+軶
+軱
+軬
+輆
+軿
+輁
+輀
+輂
+輐
+輑
+輤
+輘
+輚
+輠
+輣
+輖
+輗
+輮
+輵
+輲
+輹
+輷
+輴
+轃
+轇
+轈
+轒
+轑
+轏
+轐
+轓
+轙
+轖
+轗
+轕
+轚
+轞
+轛
+轠
+辴
+迉
+迒
+迋
+迍
+迖
+迣
+迡
+迾
+迿
+逜
+逿
+遝
+遳
+遰
+遻
+邆
+邅
+遾
+邍
+邔
+邟
+邥
+邞
+邧
+郱
+郕
+郖
+郠
+郙
+郣
+郥
+郘
+郰
+郲
+郔
+鄬
+郼
+鄈
+郹
+郻
+鄁
+鄇
+郺
+鄐
+鄍
+鄏
+鄎
+鄟
+鄝
+鄡
+鄛
+鄨
+鄪
+鄦
+鄮
+鄵
+鄸
+鄻
+鄾
+酀
+酁
+酄
+酇
+酖
+酘
+酓
+酟
+酳
+醆
+醊
+醓
+醙
+醟
+醥
+醧
+醰
+醱
+醷
+醲
+醳
+醹
+醽
+釂
+釃
+釢
+釱
+釳
+釸
+鈚
+鈌
+鈒
+釽
+鈆
+鉒
+鉠
+鉯
+鈶
+鉼
+銤
+銛
+銔
+鉹
+銗
+鋄
+鋀
+鋟
+鋘
+鋩
+鋝
+鋂
+鋊
+錧
+錼
+錭
+錎
+鋋
+鎡
+鎃
+鎯
+鍖
+鍜
+鍐
+鍭
+鍌
+鎒
+鎷
+鎝
+鎉
+鎎
+鎞
+鏏
+鏂
+鏚
+鏬
+鏙
+鐋
+鐏
+鏾
+鐕
+鐨
+鐍
+鐀
+鐎
+鐖
+鐻
+鐶
+鑐
+鑋
+鑕
+鑮
+鑯
+钂
+钀
+钁
+钃
+镺
+镻
+镼
+镽
+閈
+閍
+閺
+閵
+闀
+闉
+闅
+閷
+闒
+闑
+闚
+闛
+闠
+闟
+闤
+阞
+阢
+阤
+阠
+阰
+阹
+阸
+阺
+陏
+陓
+陊
+陼
+陭
+陫
+隇
+陾
+隉
+隒
+隓
+隞
+隤
+隿
+雂
+雈
+雓
+雔
+雗
+雚
+雟
+雘
+雺
+雽
+雿
+霂
+霋
+霒
+霐
+霠
+霣
+霢
+霩
+霫
+霬
+霮
+霵
+霿
+靆
+靃
+靪
+靮
+靷
+靲
+靾
+鞃
+鞀
+鞂
+靻
+鞊
+鞎
+鞈
+鞙
+鞗
+鞚
+鞜
+鞤
+鞪
+鞷
+鞶
+鞹
+鞻
+鞿
+韄
+韅
+韇
+韎
+韐
+韏
+韕
+韔
+韗
+韝
+韟
+韣
+韥
+韰
+韱
+韹
+韽
+頄
+頖
+頞
+頝
+頩
+頨
+頯
+頲
+顁
+顄
+顊
+顉
+顅
+顐
+顑
+顜
+顝
+顠
+顣
+顟
+顤
+顪
+顩
+顲
+颬
+颲
+颸
+颽
+颻
+颾
+飁
+飂
+飉
+飋
+飌
+飣
+飶
+餂
+餀
+飺
+餔
+餖
+餕
+餤
+餟
+餥
+餫
+餪
+餲
+餯
+餭
+餱
+餰
+饁
+饇
+饐
+饎
+饙
+饘
+饛
+饡
+馣
+馲
+馰
+馵
+馻
+馺
+駂
+馽
+駜
+駍
+駏
+駎
+駖
+駮
+駬
+駥
+駤
+駣
+駩
+駺
+駴
+駷
+駹
+駶
+駻
+駽
+駾
+騃
+騉
+騑
+騊
+騇
+騚
+騕
+騥
+騝
+騛
+騢
+騠
+騧
+騞
+騜
+騵
+騲
+騴
+騱
+騬
+騪
+騩
+騹
+騽
+驆
+騺
+驓
+驔
+驈
+驉
+驖
+驞
+驠
+驦
+驨
+骭
+骫
+骹
+骿
+骴
+骾
+髇
+髊
+髆
+髍
+髐
+髟
+髧
+髬
+髳
+髶
+髺
+髾
+鬁
+髼
+鬋
+鬊
+鬎
+鬌
+鬐
+鬕
+鬗
+鬖
+鬙
+鬞
+鬠
+鬤
+鬫
+鬳
+鬵
+鬺
+鬾
+鬿
+魊
+魌
+魖
+魠
+魡
+魧
+魱
+魦
+魶
+魵
+鮅
+鮇
+魼
+魾
+魻
+鮂
+鮚
+鮞
+鮛
+鮦
+鮥
+鮤
+鮆
+鯆
+鮿
+鮵
+鯈
+鯫
+鯠
+鯞
+鯦
+鯬
+鰌
+鰋
+鰅
+鯸
+鰫
+鰝
+鰬
+鱆
+鰿
+鱄
+鱁
+鰴
+鱐
+鱍
+鱋
+鱕
+鱦
+鱢
+鱞
+鱴
+鱳
+鱹
+鳦
+鳪
+鳭
+鳱
+鳵
+鳼
+鳺
+鳿
+鳷
+鴀
+鳹
+鳻
+鴅
+鴃
+鴥
+鴠
+鴔
+鴩
+鴘
+鴢
+鴐
+鴳
+鵁
+鵧
+鴶
+鴮
+鴱
+鴸
+鵅
+鵃
+鴾
+鵀
+鴽
+鵏
+鵊
+鵛
+鵋
+鵖
+鵌
+鵗
+鵔
+鵷
+鶁
+鶊
+鶄
+鶈
+鵱
+鶀
+鵸
+鶋
+鶌
+鵽
+鵫
+鵴
+鵩
+鶅
+鵳
+鵻
+鶂
+鵹
+鶟
+鶙
+鶤
+鶝
+鶐
+鶛
+鶠
+鶔
+鶜
+鶪
+鶗
+鶢
+鶨
+鶞
+鶣
+鶖
+鶷
+鶶
+鷁
+鷇
+鷊
+鷏
+鶾
+鷅
+鷃
+鶵
+鷈
+鶱
+鶭
+鷛
+鷒
+鷞
+鷋
+鷐
+鷜
+鷑
+鷩
+鷘
+鷖
+鷵
+鷕
+鷻
+鷷
+鷣
+鷤
+鷶
+鷡
+鷮
+鷢
+鸂
+鷾
+鸇
+鸃
+鸆
+鸅
+鸀
+鸁
+鸉
+鷿
+鷽
+鸄
+鸋
+鸍
+鸏
+鸒
+鸔
+鸓
+鸗
+鸙
+鹺
+麃
+麆
+麉
+麎
+麌
+麔
+麙
+麛
+麚
+麜
+麠
+麡
+麧
+麮
+麰
+麶
+麷
+黀
+黂
+黈
+黓
+黕
+黖
+黚
+黤
+黫
+黮
+黭
+黰
+黳
+黵
+黺
+鼁
+鼀
+鼆
+鼊
+鼏
+鼖
+鼛
+鼘
+鼜
+鼤
+鼣
+鼥
+鼪
+鼨
+鼭
+鼰
+鼮
+鼵
+鼳
+鼲
+鼸
+鼶
+齀
+齂
+齃
+齌
+齍
+齎
+齖
+齗
+齘
+齛
+齠
+齞
+齝
+齥
+齤
+齫
+齱
+齰
+齮
+齯
+齴
+齵
+齸
+齻
+齺
+齹
+齾
+龒
+龤
+堔
+礂
+蒏
+蒆
+兙
+兛
+兞
+兝
+兡
+兣
+嗧
+瓩
+忼
+擡
+氊
+穇
+擧
+譌
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+<
+=
+>
+?
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+[
+]
+_
+`
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+©
+°
+²
+´
+½
+Á
+Ä
+Å
+Ç
+È
+É
+Í
+Ó
+Ö
+×
+Ü
+ß
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+í
+ð
+ñ
+ò
+ó
+ô
+õ
+ö
+ø
+ú
+û
+ü
+ý
+ā
+ă
+ą
+ć
+Č
+č
+đ
+ē
+ė
+ę
+ğ
+ī
+ı
+Ł
+ł
+ń
+ň
+ō
+ř
+Ş
+ş
+Š
+š
+ţ
+ū
+ż
+Ž
+ž
+Ș
+ș
+ț
+Δ
+α
+λ
+μ
+φ
+Г
+О
+а
+в
+л
+о
+р
+с
+т
+я
+ồ
+—
+―
+’
+“
+”
+…
+℃
+→
+∇
+−
+■
+☆
+、
+。
+々
+〆
+〈
+〉
+「
+」
+『
+』
+〔
+〕
+〜
+！
+＃
+％
+＆
+（
+）
+＋
+，
+－
+．
+／
+０
+１
+２
+３
+４
+５
+６
+７
+８
+９
+：
+；
+＝
+？
+＠
+Ａ
+Ｂ
+Ｃ
+Ｄ
+Ｅ
+Ｆ
+Ｇ
+Ｈ
+Ｉ
+Ｊ
+Ｋ
+Ｌ
+Ｍ
+Ｎ
+Ｏ
+Ｐ
+Ｒ
+Ｓ
+Ｔ
+Ｕ
+Ｖ
+Ｗ
+Ｘ
+Ｚ
+ａ
+ｂ
+ｃ
+ｄ
+ｅ
+ｆ
+ｇ
+ｈ
+ｉ
+ｊ
+ｋ
+ｌ
+ｍ
+ｎ
+ｏ
+ｐ
+ｑ
+ｒ
+ｓ
+ｔ
+ｕ
+ｖ
+ｗ
+ｘ
+ｙ
+ｚ
+～
+･
+ǎ
+ǒ
+ě
+ǐ
+ì
+ǔ
+ù
+ǖ
+ǘ
+ǚ
+ǜ
+【
+】
+《
+》
+‥
+{
+}
+\
+|
+@
+^
+~
+÷
+∕
+∙
+⋅
+·
+⊕
+⊖
+⊗
+⊘
+⊙
+±
+∓
+∩
+∪
+□
+⊎
+⊓
+⊔
+≠
+≈
+≡
+≤
+≥
+≪
+≫
+≲
+≳
+≶
+≷
+≺
+≻
+≼
+≽
+∈
+∉
+⊂
+⊃
+⊆
+⊇
+⊄
+⊅
+∅
+∖
+∁
+∆
+∧
+∨
+¬
+⊻
+⊼
+⊽
+←
+↔
+⇒
+⇐
+⇔
+∀
+∃
+∄
+∴
+∵
+∝
+∞
+⊥
+∟
+∠
+∡
+∢
+′
+″
+∥
+⊾
+⊿
+∂
+∫
+∬
+∭
+∮
+∯
+∰
+∑
+∏
+√
+∛
+∜
+∱
+∲
+∳
+∶
+∷
+∼
+®
+≄
+≅
+≃
+≦
+≧
+⊈
+⊉
+⊢
+⊤
+⊨
+⊧
+℉
+Ω
+℧
+Å
+⌀
+ℏ
+⅀
+⍺
+⍵
+¢
+€
+£
+¥
+￥
+₿
+↑
+↓
+↕
+↖
+↗
+↘
+↙
+↺
+↻
+↼
+↽
+↾
+↿
+⇀
+⇁
+⇂
+⇃
+⇋
+⇌
+ª
+º
+⁰
+¹
+³
+⁴
+⁵
+⁶
+⁷
+⁸
+⁹
+⁺
+⁻
+⁼
+⁽
+⁾
+ⁿ
+₀
+₁
+₂
+₃
+₄
+₅
+₆
+₇
+₈
+₉
+₊
+₋
+₌
+₍
+₎
+Ⅰ
+Ⅱ
+Ⅲ
+Ⅳ
+Ⅴ
+Ⅵ
+Ⅶ
+Ⅷ
+Ⅸ
+Ⅹ
+Ⅺ
+Ⅻ
+ⅰ
+ⅱ
+ⅲ
+ⅳ
+ⅴ
+ⅵ
+ⅶ
+ⅷ
+ⅸ
+ⅹ
+ⅺ
+ⅻ
+☰
+☱
+☲
+☳
+☴
+☵
+☶
+☷
+♀
+♂
+♳
+♴
+♵
+♶
+♷
+♸
+♹
+♺
+♩
+♪
+♫
+♬
+⚪
+⚫
+⚬
+✶
+✷
+✸
+➀
+➁
+➂
+➃
+➄
+➅
+➆
+➇
+➈
+➉
+➊
+➋
+➌
+➍
+➎
+➏
+➐
+➑
+➒
+➓
+⏀
+⏁
+⏂
+⏃
+⏄
+⏅
+⏆
+⏇
+⏈
+⏉
+⏊
+⏋
+⏌
+⏚
+⏴
+⏵
+⏶
+⏷
+⏸
+⏹
+⏺
+⏻
+⏼
+Α
+Β
+Γ
+Ε
+Ζ
+Η
+Θ
+Ι
+Κ
+Λ
+Μ
+Ν
+Ξ
+Ο
+Π
+Ρ
+Σ
+Τ
+Υ
+Φ
+Χ
+Ψ
+β
+γ
+δ
+ε
+ζ
+η
+θ
+ι
+κ
+ν
+ξ
+ο
+π
+ρ
+σ
+τ
+υ
+χ
+ψ
+ω
+ϐ
+ϑ
+ϒ
+ϕ
+█
+ϖ
+ϰ
+ϱ
+ϴ
+ϵ
+ϝ
+Ϟ
+ϟ
+Ϡ
+ϡ
+Ϣ
+ϣ
+Ϥ
+ϥ
+Ϧ
+ϧ
+Ϩ
+ϩ
+Ϫ
+ϫ
+Ϭ
+ϭ
+Ϯ
+ϯ
+∸
+∹
+∺
+∻
+∽
+∾
+∿
+≀
+≁
+≂
+≆
+≇
+≉
+≊
+≋
+≌
+≍
+≎
+≏
+≐
+≑
+≒
+≓
+≔
+≕
+≖
+≗
+≘
+≙
+≚
+≛
+≜
+≝
+≞
+≟
+≢
+≣
+≨
+≩
+≬
+≭
+≮
+≯
+≰
+≱
+≴
+≵
+≸
+≹
+≾
+≿
+⊀
+⊁
+⊊
+⊋
+⊌
+⊍
+⊏
+⊐
+⊑
+⊒
+⊚
+⊛
+⊜
+⊝
+⊞
+⊟
+⊠
+⊡
+⊣
+⊦
+⊩
+⊪
+⊫
+⊬
+⊭
+⊮
+⊯
+⊰
+⊱
+⊲
+⊳
+⊴
+⊵
+⊶
+⊷
+⊸
+⊹
+⊺
+ℎ
+℘
+ℜ
+ℑ
+ℵ
+ℶ
+ℷ
+ℸ
+⌬
+⌭
+⌮
+⌯
+⎔
+¤
+₠
+₡
+₢
+₣
+₤
+₥
+₦
+₧
+₨
+₩
+₪
+₫
+₭
+₮
+₯
+₰
+₱
+₲
+₳
+₴
+₵
+₶
+₷
+₸
+₹
+₺
+₻
+₼
+₽
+₾
+↚
+↛
+↜
+↝
+↞
+↟
+↠
+↡
+↢
+↣
+↤
+↥
+↦
+↧
+↨
+↩
+↪
+↫
+↬
+↭
+↮
+↯
+↰
+↱
+↲
+↳
+↴
+↵
+↶
+↷
+↸
+↹
+⇄
+⇅
+⇆
+⇇
+⇈
+⇉
+⇊
+⇍
+⇎
+⇏
+⇑
+⇓
+⇕
+⇖
+⇗
+⇘
+⇙
+⇚
+⇛
+⇜
+⇝
+⇞
+⇟
+⇠
+⇡
+⇢
+⇣
+⇤
+⇥
+⇦
+⇧
+⇨
+⇩
+⇪
+⇫
+⇬
+⇭
+⇮
+⇯
+⇰
+⇱
+⇲
+⇳
+⇴
+⇵
+⇶
+⇷
+⇸
+⇹
+⇺
+⇻
+⇼
+⇽
+⇾
+⇿
+ↀ
+ↁ
+ↂ
+☀
+☁
+☂
+☃
+☄
+★
+☇
+☈
+☉
+☊
+☋
+☌
+☍
+☎
+☏
+☐
+☑
+☒
+☓
+☔
+☕
+☖
+☗
+☘
+☙
+☚
+☛
+☜
+☝
+☞
+☟
+☠
+☡
+☢
+☣
+☤
+☥
+☦
+☧
+☨
+☩
+☪
+☫
+☬
+☭
+☮
+☯
+☸
+☹
+☺
+☻
+☼
+☽
+☾
+☿
+♁
+♃
+♄
+♅
+♆
+♇
+♔
+♕
+♖
+♗
+♘
+♙
+♚
+♛
+♜
+♝
+♞
+♟
+♠
+♡
+♢
+♣
+♤
+♥
+♦
+♧
+♨
+♭
+♮
+♯
+♰
+♱
+♲
+♻
+♼
+♽
+♾
+⚀
+⚁
+⚂
+⚃
+⚄
+⚅
+⚆
+⚇
+⚈
+⚉
+⚊
+⚋
+⚌
+⚍
+⚎
+⚏
+⚐
+⚑
+⚒
+⚓
+⚔
+⚕
+⚖
+⚗
+⚘
+⚙
+⚚
+⚛
+⚜
+⚝
+⚞
+⚟
+⚠
+⚡
+⚢
+⚣
+⚤
+⚥
+⚦
+⚧
+⚨
+⚩
+⚭
+⚮
+⚯
+⚰
+⚱
+⚲
+⚳
+⚴
+⚵
+⚶
+⚷
+⚸
+⚹
+⚺
+⚻
+⚼
+⚿
+⛀
+⛁
+⛂
+⛃
+⛆
+⛇
+⛈
+⛉
+⛊
+⛋
+⛌
+⛍
+⛏
+⛐
+⛑
+⛒
+⛓
+⛕
+⛖
+⛗
+⛘
+⛙
+⛚
+⛛
+⛜
+⛝
+⛞
+⛠
+⛡
+⛢
+⛣
+⛤
+⛥
+⛦
+⛧
+⛨
+⛩
+⛪
+⛫
+⛬
+⛭
+⛮
+⛯
+⛶
+⛾
+⛿
+✆
+✇
+✈
+✉
+✌
+✍
+✎
+✏
+✐
+✑
+✒
+✓
+✔
+✕
+✙
+✚
+✛
+✜
+✝
+✞
+✟
+✠
+✡
+✢
+✣
+✤
+✥
+✦
+✧
+✩
+✪
+✫
+✬
+✭
+✮
+✯
+✰
+✱
+✲
+✳
+✴
+✵
+✹
+✺
+✻
+✼
+✽
+✾
+✿
+❀
+❁
+❂
+❃
+❄
+❅
+❆
+❇
+❈
+❉
+❊
+❋
+❍
+❏
+❐
+❑
+❒
+❖
+❘
+❙
+❚
+❛
+❜
+❝
+❞
+❡
+❢
+❣
+❤
+❥
+❦
+❧
+❨
+❩
+❪
+❫
+❬
+❭
+❮
+❯
+❰
+❱
+❲
+❳
+❴
+❵
+❶
+❷
+❸
+❹
+❺
+❻
+❼
+❽
+❾
+❿
+①
+②
+③
+④
+⑤
+⑥
+⑦
+⑧
+⑨
+⑩
+➔
+➕
+➖
+➗
+➘
+➙
+➚
+➛
+➜
+➝
+➞
+➟
+➠
+➡
+➢
+➣
+➤
+➥
+➦
+➧
+➨
+➩
+➪
+➫
+➬
+➭
+➮
+➯
+➰
+➱
+➲
+➳
+➴
+➵
+➶
+➷
+➸
+➹
+➺
+➻
+➼
+➽
+➾
+➿
+⌘
+⌥
+⌃
+⎋
+⌫
+⌦
+⏏
+⌤
+⌧
+⌨
+⎆
+⎇
+⎈
+⎉
+⎊
+⎌
+⎍
+⎎
+⎏
+⎐
+⎑
+⎒
+⎓
+⎕
+⎖
+⎗
+⎘
+⎙
+⎚
+⎛
+⎜
+⎝
+⎞
+⎟
+⎠
+⎡
+⎢
+⎣
+⎤
+⎥
+⎦
+⎧
+⎨
+⎩
+⎪
+⎫
+⎬
+⎭
+⎮
+⎯
+⎰
+⎱
+⎲
+⎳
+⎴
+⎵
+⎶
+⎷
+⎸
+⎹
+⎺
+⎻
+⎼
+⎽
+⎾
+⎿
+⏍
+⏎
+⏐
+⏑
+⏒
+⏓
+⏔
+⏕
+⏖
+⏗
+⏘
+⏙
+⏛
+⏜
+⏝
+⏞
+⏟
+⏠
+⏡
+⏢
+⏣
+⏤
+⏥
+⏦
+⏧
+⏨
+⏭
+⏮
+⏯
+⏱
+⏲
+▲
+▽
+◐
+⏽
+⏾
+⏿
+ɐ
+ɑ
+ɒ
+ɓ
+ɔ
+ɕ
+ɖ
+ɗ
+ɘ
+ə
+ɚ
+ɛ
+ɜ
+ɝ
+ɞ
+ɟ
+ɠ
+ɡ
+ɢ
+ɣ
+ɤ
+ɥ
+ɦ
+ɧ
+ɨ
+ɩ
+ɪ
+ɫ
+ɬ
+ɭ
+ɮ
+ɯ
+ɰ
+ɱ
+ɲ
+ɳ
+ɴ
+ɵ
+ɶ
+ɷ
+ɸ
+ɹ
+ɺ
+ɻ
+ɼ
+ɽ
+ɾ
+ɿ
+ʀ
+ʁ
+ʂ
+ʃ
+ʄ
+ʅ
+ʆ
+ʇ
+ʈ
+ʉ
+ʊ
+ʋ
+ʌ
+ʍ
+ʎ
+ʏ
+ʐ
+ʑ
+ʒ
+ʓ
+ʔ
+ʕ
+ʖ
+ʗ
+ʘ
+ʙ
+ʚ
+ʛ
+ʜ
+ʝ
+ʞ
+ʟ
+ʠ
+ʡ
+ʢ
+ʣ
+ʤ
+ʥ
+ʦ
+ʧ
+ʨ
+ʩ
+ʪ
+ʫ
+ʬ
+ʭ
+ʮ
+ʯ
+━
+Ǝ
+Ã
+●
+▶
+｜
+𝑢
+〖
+〗
+︽
+–
+﹥
+𝜓
+•
+∋
+ƒ
+०
+✘
+Е
+◉
+〒
+𝒱
+𝜆
+⟹
+﹪
+◊
+╆
+오
+˂
+〉
+𝝎
+▪
+△
+▁
+◼
+〇
+▷
+▬
+𝒮
+†
+ₒ
+⼁
+〵
+⭐
+╳
+⟶
+으
+⬆
+Ạ
+◀
+
+▫
+丄
+︾
+◥
+‖
+𝜌
+ⅼ
+▼
+⁎
+﹏
+😁
+😂
+😃
+😄
+😅
+😆
+😉
+😊
+😋
+😌
+😍
+😏
+😒
+😓
+😔
+😖
+😘
+😚
+😜
+😝
+😞
+😠
+😡
+😢
+😣
+😤
+😥
+😨
+😩
+😪
+😫
+😭
+😰
+😱
+😲
+😳
+😵
+😷
+😸
+😹
+😺
+😻
+😼
+😽
+😾
+😿
+🙀
+🙅
+🙆
+🙇
+🙈
+🙉
+🙊
+🙋
+🙌
+🙍
+🙎
+🙏
+✂
+✅
+✊
+✋
+✖
+✨
+❌
+❎
+❓
+❔
+❕
+❗
+🚀
+🚃
+🚄
+🚅
+🚇
+🚉
+🚌
+🚏
+🚑
+🚒
+🚓
+🚕
+🚗
+🚙
+🚚
+🚢
+🚤
+🚥
+🚧
+🚨
+🚩
+🚪
+🚫
+🚬
+🚭
+🚲
+🚶
+🚹
+🚺
+🚻
+🚼
+🚽
+🚾
+🛀
+Ⓜ
+🅰
+🅱
+🅾
+🅿
+🆎
+🆑
+🆒
+🆓
+🆔
+🆕
+🆖
+🆗
+🆘
+🆙
+🆚
+🇩🇪
+🇬🇧
+🇨🇳
+🇯🇵
+🇫🇷
+🇰🇷
+🇪🇸
+🇮🇹
+🇷🇺
+🇺🇸
+🈁
+ℹ
+⌚
+⌛
+⏩
+⏪
+⏫
+⏬
+⏰
+⏳
+◻
+◽
+◾
+♈
+♉
+♊
+♋
+♌
+♍
+♎
+♏
+♐
+♑
+♒
+♓
+♿
+⚽
+⚾
+⛄
+⛅
+⛎
+⛔
+⛲
+⛳
+⛵
+⛺
+⛽
+⤴
+⤵
+⬅
+⬇
+⬛
+⬜
+⭕
+〰
+〽
+㊗
+㊙
+🀄
+🃏
+🌀
+🌁
+🌂
+🌃
+🌄
+🌅
+🌆
+🌇
+🌈
+🌉
+🌊
+🌋
+🌌
+🌏
+🌑
+🌓
+🌔
+🌕
+🌙
+🌛
+🌟
+🌠
+🌰
+🌱
+🌴
+🌵
+🌷
+🌸
+🌹
+🌺
+🌻
+🌼
+🌽
+🌾
+🌿
+🍀
+🍁
+🍂
+🍃
+🍄
+🍅
+🍆
+🍇
+🍈
+🍉
+🍊
+🍌
+🍍
+🍎
+🍏
+🍑
+🍒
+🍓
+🍔
+🍕
+🍖
+🍗
+🍘
+🍙
+🍚
+🍛
+🍜
+🍝
+🍞
+🍟
+🍠
+🍡
+🍢
+🍣
+🍤
+🍥
+🍦
+🍧
+🍨
+🍩
+🍪
+🍫
+🍬
+🍭
+🍮
+🍯
+🍰
+🍱
+🍲
+🍳
+🍴
+🍵
+🍶
+🍷
+🍸
+🍹
+🍺
+🍻
+🎀
+🎁
+🎂
+🎃
+🎄
+🎅
+🎆
+🎇
+🎈
+🎉
+🎊
+🎋
+🎌
+🎍
+🎎
+🎏
+🎐
+🎑
+🎒
+🎓
+🎠
+🎡
+🎢
+🎣
+🎤
+🎥
+🎦
+🎧
+🎨
+🎩
+🎪
+🎫
+🎬
+🎭
+🎮
+🎯
+🎰
+🎱
+🎲
+🎳
+🎴
+🎵
+🎶
+🎷
+🎸
+🎹
+🎺
+🎻
+🎼
+🎽
+🎾
+🎿
+🏀
+🏁
+🏂
+🏃
+🏄
+🏆
+🏈
+🏊
+🏠
+🏡
+🏢
+🏣
+🏥
+🏦
+🏧
+🏨
+🏩
+🏪
+🏫
+🏬
+🏭
+🏮
+🏯
+🏰
+🐌
+🐍
+🐎
+🐑
+🐒
+🐔
+🐗
+🐘
+🐙
+🐚
+🐛
+🐜
+🐝
+🐞
+🐟
+🐠
+🐡
+🐢
+🐣
+🐤
+🐥
+🐦
+🐧
+🐨
+🐩
+🐫
+🐬
+🐭
+🐮
+🐯
+🐰
+🐱
+🐲
+🐳
+🐴
+🐵
+🐶
+🐷
+🐸
+🐹
+🐺
+🐻
+🐼
+🐽
+🐾
+👀
+👂
+👃
+👄
+👅
+👆
+👇
+👈
+👉
+👊
+👋
+👌
+👍
+👎
+👏
+👐
+👑
+👒
+👓
+👔
+👕
+👖
+👗
+👘
+👙
+👚
+👛
+👜
+👝
+👞
+👟
+👠
+👡
+👢
+👣
+👤
+👦
+👧
+👨
+👩
+👪
+👫
+👮
+👯
+👰
+👱
+👲
+👳
+👴
+👵
+👶
+👷
+👸
+👹
+👺
+👻
+👼
+👽
+👾
+👿
+💀
+💁
+💂
+💃
+💄
+💅
+💆
+💇
+💈
+💉
+💊
+💋
+💌
+💍
+💎
+💏
+💐
+💑
+💒
+💓
+💔
+💕
+💖
+💗
+💘
+💙
+💚
+💛
+💜
+💝
+💞
+💟
+💠
+💡
+💢
+💣
+💤
+💥
+💦
+💧
+💨
+💩
+💪
+💫
+💬
+💮
+💯
+💰
+💲
+💳
+💴
+💵
+💸
+💹
+💺
+💻
+💼
+💽
+💾
+💿
+📀
+📁
+📂
+📃
+📄
+📅
+📆
+📇
+📈
+📉
+📊
+📋
+📌
+📍
+📎
+📏
+📐
+📑
+📒
+📓
+📔
+📕
+📖
+📗
+📘
+📙
+📚
+📛
+📜
+📝
+📞
+📟
+📠
+📡
+📢
+📣
+📤
+📥
+📦
+📧
+📨
+📩
+📪
+📫
+📮
+📰
+📱
+📲
+📳
+📴
+📶
+📷
+📹
+📺
+📻
+📼
+🔃
+🔊
+🔋
+🔌
+🔍
+🔎
+🔏
+🔐
+🔑
+🔒
+🔓
+🔔
+🔖
+🔗
+🔘
+🔙
+🔚
+🔛
+🔜
+🔝
+🔞
+🔟
+🔠
+🔡
+🔢
+🔣
+🔤
+🔥
+🔦
+🔧
+🔨
+🔩
+🔪
+🔫
+🔮
+🔯
+🔰
+🔱
+🔲
+🔳
+🔴
+🔵
+🔶
+🔷
+🔸
+🔹
+🔺
+🔻
+🔼
+🔽
+🕐
+🕑
+🕒
+🕓
+🕔
+🕕
+🕖
+🕗
+🕘
+🕙
+🕚
+🕛
+🗻
+🗼
+🗽
+🗾
+🗿
+😀
+😇
+😈
+😎
+😐
+😑
+😕
+😗
+😙
+😛
+😟
+😦
+😧
+😬
+😮
+😯
+😴
+😶
+🚁
+🚂
+🚆
+🚈
+🚊
+🚍
+🚎
+🚐
+🚔
+🚖
+🚘
+🚛
+🚜
+🚝
+🚞
+🚟
+🚠
+🚡
+🚣
+🚦
+🚮
+🚯
+🚰
+🚱
+🚳
+🚴
+🚵
+🚷
+🚸
+🚿
+🛁
+🛂
+🛃
+🛄
+🛅
+🌍
+🌎
+🌐
+🌒
+🌖
+🌗
+🌘
+🌚
+🌜
+🌝
+🌞
+🌲
+🌳
+🍋
+🍐
+🍼
+🏇
+🏉
+🏤
+🐀
+🐁
+🐂
+🐃
+🐄
+🐅
+🐆
+🐇
+🐈
+🐉
+🐊
+🐋
+🐏
+🐐
+🐓
+🐕
+🐖
+🐪
+👥
+👬
+👭
+💭
+💶
+💷
+📬
+📭
+📯
+📵
+🔀
+🔁
+🔂
+🔄
+🔅
+🔆
+🔇
+🔉
+🔕
+🔬
+🔭
+🕜
+🕝
+🕞
+🕟
+🕠
+🕡
+🕢
+🕣
+🕤
+🕥
+🕦
+🕧
diff --git a/ppocr/utils/dict/ppocrv5_el_dict.txt b/ppocr/utils/dict/ppocrv5_el_dict.txt
new file mode 100644
index 0000000..a2de2e6
--- /dev/null
+++ b/ppocr/utils/dict/ppocrv5_el_dict.txt
@@ -0,0 +1,354 @@
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+<
+=
+>
+?
+@
+[
+]
+_
+`
+^
+~
+©
+®
+™
+°
+²
+³
+´
+µ
+¶
+·
+¸
+¹
+º
+»
+¼
+½
+¾
+¿
+×
+÷
+‖
+‗
+‘
+’
+‚
+‛
+“
+”
+„
+†
+‡
+…
+‰
+′
+″
+‴
+‵
+‶
+‷
+‹
+›
+‼
+‽
+‾
+€
+£
+¥
+¢
+₤
+₡
+₽
+₴
+₹
+₿
+±
+∓
+≈
+≠
+≡
+≤
+≥
+∑
+∏
+∫
+∬
+∭
+∮
+∯
+∰
+√
+∛
+∜
+∝
+∞
+∂
+∇
+∴
+∵
+∷
+∠
+∡
+∢
+∟
+∦
+∩
+∪
+∗
+∖
+∥
+∧
+∨
+⊂
+⊃
+⊥
+∈
+∋
+∉
+∅
+↑
+→
+↓
+↔
+↕
+←
+⇒
+⇐
+⇔
+∀
+∃
+∄
+Ⅰ
+Ⅱ
+Ⅲ
+Ⅳ
+Ⅴ
+Ⅵ
+Ⅶ
+Ⅷ
+Ⅸ
+Ⅹ
+Ⅺ
+Ⅻ
+ⅰ
+ⅱ
+ⅲ
+ⅳ
+ⅴ
+ⅵ
+ⅶ
+ⅷ
+ⅸ
+ⅹ
+ⅺ
+ⅻ
+①
+②
+③
+④
+⑤
+⑥
+⑦
+⑧
+⑨
+⑩
+❶
+❷
+❸
+❹
+❺
+❻
+❼
+❽
+❾
+❿
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+Α
+Β
+Γ
+Δ
+Ε
+Ζ
+Η
+Θ
+Ι
+Κ
+Λ
+Μ
+Ν
+Ξ
+Ο
+Π
+Ρ
+Σ
+Τ
+Υ
+Φ
+Χ
+Ψ
+Ω
+ά
+έ
+ή
+ί
+ό
+ύ
+ώ
+Ϊ
+Ϋ
+Ά
+Έ
+Ή
+Ί
+Ό
+Ύ
+Ώ
+ϊ
+ΐ
+ϋ
+ΰ
+α
+β
+γ
+δ
+ε
+ζ
+η
+θ
+ι
+κ
+λ
+μ
+ν
+ξ
+ο
+π
+ρ
+ς
+σ
+τ
+υ
+φ
+χ
+ψ
+ω
+ϐ
+ϑ
+ϒ
+ϓ
+ϔ
+ϕ
+ϖ
+Ϛ
+ϛ
+Ϝ
+ϝ
+Ϟ
+ϟ
+Ϡ
+ϡ
+Ϣ
+ϣ
+Ϥ
+ϥ
+Ϧ
+ϧ
+Ϩ
+ϩ
+Ϫ
+ϫ
+Ϭ
+ϭ
+Ϯ
+ϯ
+ϰ
+ϱ
+ϲ
+ϳ
+ϴ
+ϵ
+϶
+Ϸ
+ϸ
+Ϲ
+Ϻ
+ϻ
diff --git a/ppocr/utils/dict/ppocrv5_en_dict.txt b/ppocr/utils/dict/ppocrv5_en_dict.txt
new file mode 100644
index 0000000..7bf3f5f
--- /dev/null
+++ b/ppocr/utils/dict/ppocrv5_en_dict.txt
@@ -0,0 +1,436 @@
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+:
+;
+<
+=
+>
+?
+@
+[
+\
+]
+_
+`
+{
+|
+}
+^
+~
+©
+®
+℉
+№
+Ω
+℮
+™
+∆
+✓
+✔
+✗
+✘
+✕
+☑
+☒
+●
+▪
+▫
+◼
+▶
+◀
+⬆
+¤
+¦
+§
+¨
+ª
+«
+¬
+¯
+°
+²
+³
+´
+µ
+¶
+¸
+¹
+º
+»
+¼
+½
+¾
+¿
+×
+‐
+‑
+‒
+—
+―
+‖
+‗
+‘
+’
+‚
+‛
+“
+”
+„
+‟
+†
+‡
+‣
+․
+…
+‧
+‰
+‴
+‵
+‶
+‷
+‸
+‹
+›
+※
+‼
+‽
+‾
+−
+₤
+₡
+₹
+₽
+₴
+₿
+¢
+€
+£
+¥
+Ⅰ
+Ⅱ
+Ⅲ
+Ⅳ
+Ⅴ
+Ⅵ
+Ⅶ
+Ⅷ
+Ⅸ
+Ⅹ
+Ⅺ
+Ⅻ
+ⅰ
+ⅱ
+ⅲ
+ⅳ
+ⅴ
+ⅵ
+ⅶ
+ⅷ
+ⅸ
+ⅹ
+ⅺ
+ⅻ
+➀
+➁
+➂
+➃
+➄
+➅
+➆
+➇
+➈
+➉
+➊
+➋
+➌
+➍
+➎
+➏
+➐
+➑
+➒
+➓
+❶
+❷
+❸
+❹
+❺
+❻
+❼
+❽
+❾
+❿
+①
+②
+③
+④
+⑤
+⑥
+⑦
+⑧
+⑨
+⑩
+↑
+→
+↓
+↕
+←
+↔
+⇒
+⇐
+⇔
+∀
+∃
+∄
+∴
+∵
+∝
+∞
+∩
+∪
+∂
+∫
+∬
+∭
+∮
+∯
+∰
+∑
+∏
+√
+∛
+∜
+∱
+∲
+∳
+∶
+∷
+∼
+∖
+∗
+≈
+≠
+≡
+≤
+≥
+⊂
+⊃
+⊥
+⊾
+⊿
+□
+∥
+∋
+ƒ
+′
+″
+À
+Á
+Â
+Ã
+Ä
+Å
+Æ
+Ç
+È
+É
+Ê
+Ë
+Ì
+Í
+Î
+Ï
+Ð
+Ñ
+Ò
+Ó
+Ô
+Õ
+Ö
+Ø
+Ù
+Ú
+Û
+Ü
+Ý
+Þ
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+ì
+í
+î
+ï
+ð
+ñ
+ò
+ó
+ô
+õ
+ö
+ø
+ù
+ú
+û
+ü
+ý
+þ
+ÿ
+Α
+Β
+Γ
+Δ
+Ε
+Ζ
+Η
+Θ
+Ι
+Κ
+Λ
+Μ
+Ν
+Ξ
+Ο
+Π
+Ρ
+Σ
+Τ
+Υ
+Φ
+Χ
+Ψ
+Ω
+α
+β
+γ
+δ
+ε
+ζ
+η
+θ
+ι
+κ
+λ
+μ
+ν
+ξ
+ο
+π
+ρ
+σ
+ς
+τ
+υ
+φ
+χ
+ψ
+ω
+Å
+ℏ
+⌀
+⍺
+⍵
+𝑢
+𝜓
+०
+‥
+︽
+﹥
+•
+÷
+∕
+∙
+⋅
+·
+±
+∓
+∟
+∠
+∡
+∢
+℧
+☺
diff --git a/ppocr/utils/dict/ppocrv5_eslav_dict.txt b/ppocr/utils/dict/ppocrv5_eslav_dict.txt
new file mode 100644
index 0000000..d50d6e5
--- /dev/null
+++ b/ppocr/utils/dict/ppocrv5_eslav_dict.txt
@@ -0,0 +1,517 @@
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+<
+=
+>
+?
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+[
+]
+_
+`
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+©
+‥
+{
+}
+\
+|
+@
+^
+~
+÷
+∕
+∙
+⋅
+·
+±
+∓
+∩
+∪
+□
+←
+↔
+⇒
+⇐
+⇔
+∀
+∃
+∄
+∴
+∵
+∝
+∞
+⊥
+∟
+∠
+∡
+∢
+′
+″
+∥
+⊾
+⊿
+∂
+∫
+∬
+∭
+∮
+∯
+∰
+∑
+∏
+√
+∛
+∜
+∱
+∲
+∳
+∶
+∷
+∼
+®
+℉
+Ω
+℧
+Å
+⌀
+ℏ
+⅀
+⍺
+⍵
+¢
+€
+£
+¥
+₿
+Ⅰ
+Ⅱ
+Ⅲ
+Ⅳ
+Ⅴ
+Ⅵ
+Ⅶ
+Ⅷ
+Ⅸ
+Ⅹ
+Ⅺ
+Ⅻ
+ⅰ
+ⅱ
+ⅲ
+ⅳ
+ⅴ
+ⅵ
+ⅶ
+ⅷ
+ⅸ
+ⅹ
+ⅺ
+ⅻ
+➀
+➁
+➂
+➃
+➄
+➅
+➆
+➇
+➈
+➉
+➊
+➋
+➌
+➍
+➎
+➏
+➐
+➑
+➒
+➓
+❶
+❷
+❸
+❹
+❺
+❻
+❼
+❽
+❾
+❿
+①
+②
+③
+④
+⑤
+⑥
+⑦
+⑧
+⑨
+⑩
+●
+▶
+𝑢
+︽
+–
+﹥
+𝜓
+•
+∋
+ƒ
+०
+⬆
+Ạ
+◀
+
+▫
+︾
+À
+Á
+Â
+Ã
+Ä
+Å
+Æ
+Ç
+È
+É
+Ê
+Ë
+Ì
+Í
+Î
+Ï
+Ð
+Ñ
+Ò
+Ó
+Ô
+Õ
+Ö
+Ø
+Ù
+Ú
+Û
+Ü
+Ý
+Þ
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+ì
+í
+î
+ï
+ð
+ñ
+ò
+ó
+ô
+õ
+ö
+ø
+ù
+ú
+û
+ü
+ý
+þ
+ÿ
+¡
+¤
+¦
+§
+¨
+ª
+«
+¬
+¯
+°
+²
+³
+´
+µ
+¶
+¸
+¹
+º
+»
+¼
+½
+¾
+¿
+×
+‐
+‑
+‒
+—
+―
+‖
+‗
+‘
+’
+‚
+‛
+“
+”
+„
+‟
+†
+‡
+‣
+․
+…
+‧
+‰
+‴
+‵
+‶
+‷
+‸
+‹
+›
+※
+‼
+‽
+‾
+₤
+₡
+₹
+−
+∖
+∗
+≈
+≠
+≡
+≤
+≥
+⊂
+⊃
+↑
+→
+↓
+↕
+™
+Ω
+℮
+∆
+✓
+✗
+✘
+▪
+◼
+✔
+✕
+☑
+☒
+№
+₽
+₴
+Α
+α
+Β
+β
+Γ
+γ
+Δ
+δ
+Ε
+ε
+Ζ
+ζ
+Η
+η
+Θ
+θ
+Ι
+ι
+Κ
+κ
+Λ
+λ
+Μ
+μ
+Ν
+ν
+Ξ
+ξ
+Ο
+ο
+Π
+π
+Ρ
+ρ
+Σ
+σ
+ς
+Τ
+τ
+Υ
+υ
+Φ
+φ
+Χ
+χ
+Ψ
+ψ
+ω
+А
+Б
+В
+Г
+Ґ
+Д
+Е
+Ё
+Є
+Ж
+З
+И
+І
+Ї
+Й
+К
+Л
+М
+Н
+О
+П
+Р
+С
+Т
+У
+Ў
+Ф
+Х
+Ц
+Ч
+Ш
+Щ
+Ъ
+Ы
+Ь
+Э
+Ю
+Я
+а
+б
+в
+г
+ґ
+д
+е
+ё
+є
+ж
+з
+и
+і
+ї
+й
+к
+л
+м
+н
+о
+п
+р
+с
+т
+у
+ў
+ф
+х
+ц
+ч
+ш
+щ
+ъ
+ы
+ь
+э
+ю
+я
diff --git a/ppocr/utils/dict/ppocrv5_korean_dict.txt b/ppocr/utils/dict/ppocrv5_korean_dict.txt
new file mode 100644
index 0000000..e220263
--- /dev/null
+++ b/ppocr/utils/dict/ppocrv5_korean_dict.txt
@@ -0,0 +1,11945 @@
+ᄀ
+ᄁ
+ᄂ
+ᄃ
+ᄄ
+ᄅ
+ᄆ
+ᄇ
+ᄈ
+ᄉ
+ᄊ
+ᄋ
+ᄌ
+ᄍ
+ᄎ
+ᄏ
+ᄐ
+ᄑ
+ᄒ
+ᄓ
+ᄔ
+ᄕ
+ᄖ
+ᄗ
+ᄘ
+ᄙ
+ᄚ
+ᄛ
+ᄜ
+ᄝ
+ᄞ
+ᄟ
+ᄠ
+ᄡ
+ᄢ
+ᄣ
+ᄤ
+ᄥ
+ᄦ
+ᄧ
+ᄨ
+ᄩ
+ᄪ
+ᄫ
+ᄬ
+ᄭ
+ᄮ
+ᄯ
+ᄰ
+ᄱ
+ᄲ
+ᄳ
+ᄴ
+ᄵ
+ᄶ
+ᄷ
+ᄸ
+ᄹ
+ᄺ
+ᄻ
+ᄼ
+ᄽ
+ᄾ
+ᄿ
+ᅀ
+ᅁ
+ᅂ
+ᅃ
+ᅄ
+ᅅ
+ᅆ
+ᅇ
+ᅈ
+ᅉ
+ᅊ
+ᅋ
+ᅌ
+ᅍ
+ᅎ
+ᅏ
+ᅐ
+ᅑ
+ᅒ
+ᅓ
+ᅔ
+ᅕ
+ᅖ
+ᅗ
+ᅘ
+ᅙ
+ᅡ
+ᅢ
+ᅣ
+ᅤ
+ᅥ
+ᅦ
+ᅧ
+ᅨ
+ᅩ
+ᅪ
+ᅫ
+ᅬ
+ᅭ
+ᅮ
+ᅯ
+ᅰ
+ᅱ
+ᅲ
+ᅳ
+ᅴ
+ᅵ
+ᅶ
+ᅷ
+ᅸ
+ᅹ
+ᅺ
+ᅻ
+ᅼ
+ᅽ
+ᅾ
+ᅿ
+ᆀ
+ᆁ
+ᆂ
+ᆃ
+ᆄ
+ᆅ
+ᆆ
+ᆇ
+ᆈ
+ᆉ
+ᆊ
+ᆋ
+ᆌ
+ᆍ
+ᆎ
+ᆏ
+ᆐ
+ᆑ
+ᆒ
+ᆓ
+ᆔ
+ᆕ
+ᆖ
+ᆗ
+ᆘ
+ᆙ
+ᆚ
+ᆛ
+ᆜ
+ᆝ
+ᆞ
+ᆟ
+ᆠ
+ᆡ
+ᆢ
+ᆨ
+ᆩ
+ᆪ
+ᆫ
+ᆬ
+ᆭ
+ᆮ
+ᆯ
+ᆰ
+ᆱ
+ᆲ
+ᆳ
+ᆴ
+ᆵ
+ᆶ
+ᆷ
+ᆸ
+ᆹ
+ᆺ
+ᆻ
+ᆼ
+ᆽ
+ᆾ
+ᆿ
+ᇀ
+ᇁ
+ᇂ
+ᇃ
+ᇄ
+ᇅ
+ᇆ
+ᇇ
+ᇈ
+ᇉ
+ᇊ
+ᇋ
+ᇌ
+ᇍ
+ᇎ
+ᇏ
+ᇐ
+ᇑ
+ᇒ
+ᇓ
+ᇔ
+ᇕ
+ᇖ
+ᇗ
+ᇘ
+ᇙ
+ᇚ
+ᇛ
+ᇜ
+ᇝ
+ᇞ
+ᇟ
+ᇠ
+ᇡ
+ᇢ
+ᇣ
+ᇤ
+ᇥ
+ᇦ
+ᇧ
+ᇨ
+ᇩ
+ᇪ
+ᇫ
+ᇬ
+ᇭ
+ᇮ
+ᇯ
+ᇰ
+ᇱ
+ᇲ
+ᇳ
+ᇴ
+ᇵ
+ᇶ
+ᇷ
+ᇸ
+ᇹ
+ㄱ
+ㄲ
+ㄳ
+ㄴ
+ㄵ
+ㄶ
+ㄷ
+ㄸ
+ㄹ
+ㄺ
+ㄻ
+ㄼ
+ㄽ
+ㄾ
+ㄿ
+ㅀ
+ㅁ
+ㅂ
+ㅃ
+ㅄ
+ㅅ
+ㅆ
+ㅇ
+ㅈ
+ㅉ
+ㅊ
+ㅋ
+ㅌ
+ㅍ
+ㅎ
+ㅏ
+ㅐ
+ㅑ
+ㅒ
+ㅓ
+ㅔ
+ㅕ
+ㅖ
+ㅗ
+ㅘ
+ㅙ
+ㅚ
+ㅛ
+ㅜ
+ㅝ
+ㅞ
+ㅟ
+ㅠ
+ㅡ
+ㅢ
+ㅣ
+ㅤ
+ㅥ
+ㅦ
+ㅧ
+ㅨ
+ㅩ
+ㅪ
+ㅫ
+ㅬ
+ㅭ
+ㅮ
+ㅯ
+ㅰ
+ㅱ
+ㅲ
+ㅳ
+ㅴ
+ㅵ
+ㅶ
+ㅷ
+ㅸ
+ㅹ
+ㅺ
+ㅻ
+ㅼ
+ㅽ
+ㅾ
+ㅿ
+ㆀ
+ㆁ
+ㆂ
+ㆃ
+ㆄ
+ㆅ
+ㆆ
+ㆇ
+ㆈ
+ㆉ
+ㆊ
+ㆋ
+ㆌ
+ㆍ
+ㆎ
+가
+각
+갂
+갃
+간
+갅
+갆
+갇
+갈
+갉
+갊
+갋
+갌
+갍
+갎
+갏
+감
+갑
+값
+갓
+갔
+강
+갖
+갗
+갘
+같
+갚
+갛
+개
+객
+갞
+갟
+갠
+갡
+갢
+갣
+갤
+갥
+갦
+갧
+갨
+갩
+갪
+갫
+갬
+갭
+갮
+갯
+갰
+갱
+갲
+갳
+갴
+갵
+갶
+갷
+갸
+갹
+갺
+갻
+갼
+갽
+갾
+갿
+걀
+걁
+걂
+걃
+걄
+걅
+걆
+걇
+걈
+걉
+걊
+걋
+걌
+걍
+걎
+걏
+걐
+걑
+걒
+걓
+걔
+걕
+걖
+걗
+걘
+걙
+걚
+걛
+걜
+걝
+걞
+걟
+걠
+걡
+걢
+걣
+걤
+걥
+걦
+걧
+걨
+걩
+걪
+걫
+걬
+걭
+걮
+걯
+거
+걱
+걲
+걳
+건
+걵
+걶
+걷
+걸
+걹
+걺
+걻
+걼
+걽
+걾
+걿
+검
+겁
+겂
+것
+겄
+겅
+겆
+겇
+겈
+겉
+겊
+겋
+게
+겍
+겎
+겏
+겐
+겑
+겒
+겓
+겔
+겕
+겖
+겗
+겘
+겙
+겚
+겛
+겜
+겝
+겞
+겟
+겠
+겡
+겢
+겣
+겤
+겥
+겦
+겧
+겨
+격
+겪
+겫
+견
+겭
+겮
+겯
+결
+겱
+겲
+겳
+겴
+겵
+겶
+겷
+겸
+겹
+겺
+겻
+겼
+경
+겾
+겿
+곀
+곁
+곂
+곃
+계
+곅
+곆
+곇
+곈
+곉
+곊
+곋
+곌
+곍
+곎
+곏
+곐
+곑
+곒
+곓
+곔
+곕
+곖
+곗
+곘
+곙
+곚
+곛
+곜
+곝
+곞
+곟
+고
+곡
+곢
+곣
+곤
+곥
+곦
+곧
+골
+곩
+곪
+곫
+곬
+곭
+곮
+곯
+곰
+곱
+곲
+곳
+곴
+공
+곶
+곷
+곸
+곹
+곺
+곻
+과
+곽
+곾
+곿
+관
+괁
+괂
+괃
+괄
+괅
+괆
+괇
+괈
+괉
+괊
+괋
+괌
+괍
+괎
+괏
+괐
+광
+괒
+괓
+괔
+괕
+괖
+괗
+괘
+괙
+괚
+괛
+괜
+괝
+괞
+괟
+괠
+괡
+괢
+괣
+괤
+괥
+괦
+괧
+괨
+괩
+괪
+괫
+괬
+괭
+괮
+괯
+괰
+괱
+괲
+괳
+괴
+괵
+괶
+괷
+괸
+괹
+괺
+괻
+괼
+괽
+괾
+괿
+굀
+굁
+굂
+굃
+굄
+굅
+굆
+굇
+굈
+굉
+굊
+굋
+굌
+굍
+굎
+굏
+교
+굑
+굒
+굓
+굔
+굕
+굖
+굗
+굘
+굙
+굚
+굛
+굜
+굝
+굞
+굟
+굠
+굡
+굢
+굣
+굤
+굥
+굦
+굧
+굨
+굩
+굪
+굫
+구
+국
+굮
+굯
+군
+굱
+굲
+굳
+굴
+굵
+굶
+굷
+굸
+굹
+굺
+굻
+굼
+굽
+굾
+굿
+궀
+궁
+궂
+궃
+궄
+궅
+궆
+궇
+궈
+궉
+궊
+궋
+권
+궍
+궎
+궏
+궐
+궑
+궒
+궓
+궔
+궕
+궖
+궗
+궘
+궙
+궚
+궛
+궜
+궝
+궞
+궟
+궠
+궡
+궢
+궣
+궤
+궥
+궦
+궧
+궨
+궩
+궪
+궫
+궬
+궭
+궮
+궯
+궰
+궱
+궲
+궳
+궴
+궵
+궶
+궷
+궸
+궹
+궺
+궻
+궼
+궽
+궾
+궿
+귀
+귁
+귂
+귃
+귄
+귅
+귆
+귇
+귈
+귉
+귊
+귋
+귌
+귍
+귎
+귏
+귐
+귑
+귒
+귓
+귔
+귕
+귖
+귗
+귘
+귙
+귚
+귛
+규
+귝
+귞
+귟
+균
+귡
+귢
+귣
+귤
+귥
+귦
+귧
+귨
+귩
+귪
+귫
+귬
+귭
+귮
+귯
+귰
+귱
+귲
+귳
+귴
+귵
+귶
+귷
+그
+극
+귺
+귻
+근
+귽
+귾
+귿
+글
+긁
+긂
+긃
+긄
+긅
+긆
+긇
+금
+급
+긊
+긋
+긌
+긍
+긎
+긏
+긐
+긑
+긒
+긓
+긔
+긕
+긖
+긗
+긘
+긙
+긚
+긛
+긜
+긝
+긞
+긟
+긠
+긡
+긢
+긣
+긤
+긥
+긦
+긧
+긨
+긩
+긪
+긫
+긬
+긭
+긮
+긯
+기
+긱
+긲
+긳
+긴
+긵
+긶
+긷
+길
+긹
+긺
+긻
+긼
+긽
+긾
+긿
+김
+깁
+깂
+깃
+깄
+깅
+깆
+깇
+깈
+깉
+깊
+깋
+까
+깍
+깎
+깏
+깐
+깑
+깒
+깓
+깔
+깕
+깖
+깗
+깘
+깙
+깚
+깛
+깜
+깝
+깞
+깟
+깠
+깡
+깢
+깣
+깤
+깥
+깦
+깧
+깨
+깩
+깪
+깫
+깬
+깭
+깮
+깯
+깰
+깱
+깲
+깳
+깴
+깵
+깶
+깷
+깸
+깹
+깺
+깻
+깼
+깽
+깾
+깿
+꺀
+꺁
+꺂
+꺃
+꺄
+꺅
+꺆
+꺇
+꺈
+꺉
+꺊
+꺋
+꺌
+꺍
+꺎
+꺏
+꺐
+꺑
+꺒
+꺓
+꺔
+꺕
+꺖
+꺗
+꺘
+꺙
+꺚
+꺛
+꺜
+꺝
+꺞
+꺟
+꺠
+꺡
+꺢
+꺣
+꺤
+꺥
+꺦
+꺧
+꺨
+꺩
+꺪
+꺫
+꺬
+꺭
+꺮
+꺯
+꺰
+꺱
+꺲
+꺳
+꺴
+꺵
+꺶
+꺷
+꺸
+꺹
+꺺
+꺻
+꺼
+꺽
+꺾
+꺿
+껀
+껁
+껂
+껃
+껄
+껅
+껆
+껇
+껈
+껉
+껊
+껋
+껌
+껍
+껎
+껏
+껐
+껑
+껒
+껓
+껔
+껕
+껖
+껗
+께
+껙
+껚
+껛
+껜
+껝
+껞
+껟
+껠
+껡
+껢
+껣
+껤
+껥
+껦
+껧
+껨
+껩
+껪
+껫
+껬
+껭
+껮
+껯
+껰
+껱
+껲
+껳
+껴
+껵
+껶
+껷
+껸
+껹
+껺
+껻
+껼
+껽
+껾
+껿
+꼀
+꼁
+꼂
+꼃
+꼄
+꼅
+꼆
+꼇
+꼈
+꼉
+꼊
+꼋
+꼌
+꼍
+꼎
+꼏
+꼐
+꼑
+꼒
+꼓
+꼔
+꼕
+꼖
+꼗
+꼘
+꼙
+꼚
+꼛
+꼜
+꼝
+꼞
+꼟
+꼠
+꼡
+꼢
+꼣
+꼤
+꼥
+꼦
+꼧
+꼨
+꼩
+꼪
+꼫
+꼬
+꼭
+꼮
+꼯
+꼰
+꼱
+꼲
+꼳
+꼴
+꼵
+꼶
+꼷
+꼸
+꼹
+꼺
+꼻
+꼼
+꼽
+꼾
+꼿
+꽀
+꽁
+꽂
+꽃
+꽄
+꽅
+꽆
+꽇
+꽈
+꽉
+꽊
+꽋
+꽌
+꽍
+꽎
+꽏
+꽐
+꽑
+꽒
+꽓
+꽔
+꽕
+꽖
+꽗
+꽘
+꽙
+꽚
+꽛
+꽜
+꽝
+꽞
+꽟
+꽠
+꽡
+꽢
+꽣
+꽤
+꽥
+꽦
+꽧
+꽨
+꽩
+꽪
+꽫
+꽬
+꽭
+꽮
+꽯
+꽰
+꽱
+꽲
+꽳
+꽴
+꽵
+꽶
+꽷
+꽸
+꽹
+꽺
+꽻
+꽼
+꽽
+꽾
+꽿
+꾀
+꾁
+꾂
+꾃
+꾄
+꾅
+꾆
+꾇
+꾈
+꾉
+꾊
+꾋
+꾌
+꾍
+꾎
+꾏
+꾐
+꾑
+꾒
+꾓
+꾔
+꾕
+꾖
+꾗
+꾘
+꾙
+꾚
+꾛
+꾜
+꾝
+꾞
+꾟
+꾠
+꾡
+꾢
+꾣
+꾤
+꾥
+꾦
+꾧
+꾨
+꾩
+꾪
+꾫
+꾬
+꾭
+꾮
+꾯
+꾰
+꾱
+꾲
+꾳
+꾴
+꾵
+꾶
+꾷
+꾸
+꾹
+꾺
+꾻
+꾼
+꾽
+꾾
+꾿
+꿀
+꿁
+꿂
+꿃
+꿄
+꿅
+꿆
+꿇
+꿈
+꿉
+꿊
+꿋
+꿌
+꿍
+꿎
+꿏
+꿐
+꿑
+꿒
+꿓
+꿔
+꿕
+꿖
+꿗
+꿘
+꿙
+꿚
+꿛
+꿜
+꿝
+꿞
+꿟
+꿠
+꿡
+꿢
+꿣
+꿤
+꿥
+꿦
+꿧
+꿨
+꿩
+꿪
+꿫
+꿬
+꿭
+꿮
+꿯
+꿰
+꿱
+꿲
+꿳
+꿴
+꿵
+꿶
+꿷
+꿸
+꿹
+꿺
+꿻
+꿼
+꿽
+꿾
+꿿
+뀀
+뀁
+뀂
+뀃
+뀄
+뀅
+뀆
+뀇
+뀈
+뀉
+뀊
+뀋
+뀌
+뀍
+뀎
+뀏
+뀐
+뀑
+뀒
+뀓
+뀔
+뀕
+뀖
+뀗
+뀘
+뀙
+뀚
+뀛
+뀜
+뀝
+뀞
+뀟
+뀠
+뀡
+뀢
+뀣
+뀤
+뀥
+뀦
+뀧
+뀨
+뀩
+뀪
+뀫
+뀬
+뀭
+뀮
+뀯
+뀰
+뀱
+뀲
+뀳
+뀴
+뀵
+뀶
+뀷
+뀸
+뀹
+뀺
+뀻
+뀼
+뀽
+뀾
+뀿
+끀
+끁
+끂
+끃
+끄
+끅
+끆
+끇
+끈
+끉
+끊
+끋
+끌
+끍
+끎
+끏
+끐
+끑
+끒
+끓
+끔
+끕
+끖
+끗
+끘
+끙
+끚
+끛
+끜
+끝
+끞
+끟
+끠
+끡
+끢
+끣
+끤
+끥
+끦
+끧
+끨
+끩
+끪
+끫
+끬
+끭
+끮
+끯
+끰
+끱
+끲
+끳
+끴
+끵
+끶
+끷
+끸
+끹
+끺
+끻
+끼
+끽
+끾
+끿
+낀
+낁
+낂
+낃
+낄
+낅
+낆
+낇
+낈
+낉
+낊
+낋
+낌
+낍
+낎
+낏
+낐
+낑
+낒
+낓
+낔
+낕
+낖
+낗
+나
+낙
+낚
+낛
+난
+낝
+낞
+낟
+날
+낡
+낢
+낣
+낤
+낥
+낦
+낧
+남
+납
+낪
+낫
+났
+낭
+낮
+낯
+낰
+낱
+낲
+낳
+내
+낵
+낶
+낷
+낸
+낹
+낺
+낻
+낼
+낽
+낾
+낿
+냀
+냁
+냂
+냃
+냄
+냅
+냆
+냇
+냈
+냉
+냊
+냋
+냌
+냍
+냎
+냏
+냐
+냑
+냒
+냓
+냔
+냕
+냖
+냗
+냘
+냙
+냚
+냛
+냜
+냝
+냞
+냟
+냠
+냡
+냢
+냣
+냤
+냥
+냦
+냧
+냨
+냩
+냪
+냫
+냬
+냭
+냮
+냯
+냰
+냱
+냲
+냳
+냴
+냵
+냶
+냷
+냸
+냹
+냺
+냻
+냼
+냽
+냾
+냿
+넀
+넁
+넂
+넃
+넄
+넅
+넆
+넇
+너
+넉
+넊
+넋
+넌
+넍
+넎
+넏
+널
+넑
+넒
+넓
+넔
+넕
+넖
+넗
+넘
+넙
+넚
+넛
+넜
+넝
+넞
+넟
+넠
+넡
+넢
+넣
+네
+넥
+넦
+넧
+넨
+넩
+넪
+넫
+넬
+넭
+넮
+넯
+넰
+넱
+넲
+넳
+넴
+넵
+넶
+넷
+넸
+넹
+넺
+넻
+넼
+넽
+넾
+넿
+녀
+녁
+녂
+녃
+년
+녅
+녆
+녇
+녈
+녉
+녊
+녋
+녌
+녍
+녎
+녏
+념
+녑
+녒
+녓
+녔
+녕
+녖
+녗
+녘
+녙
+녚
+녛
+녜
+녝
+녞
+녟
+녠
+녡
+녢
+녣
+녤
+녥
+녦
+녧
+녨
+녩
+녪
+녫
+녬
+녭
+녮
+녯
+녰
+녱
+녲
+녳
+녴
+녵
+녶
+녷
+노
+녹
+녺
+녻
+논
+녽
+녾
+녿
+놀
+놁
+놂
+놃
+놄
+놅
+놆
+놇
+놈
+놉
+놊
+놋
+놌
+농
+놎
+놏
+놐
+놑
+높
+놓
+놔
+놕
+놖
+놗
+놘
+놙
+놚
+놛
+놜
+놝
+놞
+놟
+놠
+놡
+놢
+놣
+놤
+놥
+놦
+놧
+놨
+놩
+놪
+놫
+놬
+놭
+놮
+놯
+놰
+놱
+놲
+놳
+놴
+놵
+놶
+놷
+놸
+놹
+놺
+놻
+놼
+놽
+놾
+놿
+뇀
+뇁
+뇂
+뇃
+뇄
+뇅
+뇆
+뇇
+뇈
+뇉
+뇊
+뇋
+뇌
+뇍
+뇎
+뇏
+뇐
+뇑
+뇒
+뇓
+뇔
+뇕
+뇖
+뇗
+뇘
+뇙
+뇚
+뇛
+뇜
+뇝
+뇞
+뇟
+뇠
+뇡
+뇢
+뇣
+뇤
+뇥
+뇦
+뇧
+뇨
+뇩
+뇪
+뇫
+뇬
+뇭
+뇮
+뇯
+뇰
+뇱
+뇲
+뇳
+뇴
+뇵
+뇶
+뇷
+뇸
+뇹
+뇺
+뇻
+뇼
+뇽
+뇾
+뇿
+눀
+눁
+눂
+눃
+누
+눅
+눆
+눇
+눈
+눉
+눊
+눋
+눌
+눍
+눎
+눏
+눐
+눑
+눒
+눓
+눔
+눕
+눖
+눗
+눘
+눙
+눚
+눛
+눜
+눝
+눞
+눟
+눠
+눡
+눢
+눣
+눤
+눥
+눦
+눧
+눨
+눩
+눪
+눫
+눬
+눭
+눮
+눯
+눰
+눱
+눲
+눳
+눴
+눵
+눶
+눷
+눸
+눹
+눺
+눻
+눼
+눽
+눾
+눿
+뉀
+뉁
+뉂
+뉃
+뉄
+뉅
+뉆
+뉇
+뉈
+뉉
+뉊
+뉋
+뉌
+뉍
+뉎
+뉏
+뉐
+뉑
+뉒
+뉓
+뉔
+뉕
+뉖
+뉗
+뉘
+뉙
+뉚
+뉛
+뉜
+뉝
+뉞
+뉟
+뉠
+뉡
+뉢
+뉣
+뉤
+뉥
+뉦
+뉧
+뉨
+뉩
+뉪
+뉫
+뉬
+뉭
+뉮
+뉯
+뉰
+뉱
+뉲
+뉳
+뉴
+뉵
+뉶
+뉷
+뉸
+뉹
+뉺
+뉻
+뉼
+뉽
+뉾
+뉿
+늀
+늁
+늂
+늃
+늄
+늅
+늆
+늇
+늈
+늉
+늊
+늋
+늌
+늍
+늎
+늏
+느
+늑
+늒
+늓
+는
+늕
+늖
+늗
+늘
+늙
+늚
+늛
+늜
+늝
+늞
+늟
+늠
+늡
+늢
+늣
+늤
+능
+늦
+늧
+늨
+늩
+늪
+늫
+늬
+늭
+늮
+늯
+늰
+늱
+늲
+늳
+늴
+늵
+늶
+늷
+늸
+늹
+늺
+늻
+늼
+늽
+늾
+늿
+닀
+닁
+닂
+닃
+닄
+닅
+닆
+닇
+니
+닉
+닊
+닋
+닌
+닍
+닎
+닏
+닐
+닑
+닒
+닓
+닔
+닕
+닖
+닗
+님
+닙
+닚
+닛
+닜
+닝
+닞
+닟
+닠
+닡
+닢
+닣
+다
+닥
+닦
+닧
+단
+닩
+닪
+닫
+달
+닭
+닮
+닯
+닰
+닱
+닲
+닳
+담
+답
+닶
+닷
+닸
+당
+닺
+닻
+닼
+닽
+닾
+닿
+대
+댁
+댂
+댃
+댄
+댅
+댆
+댇
+댈
+댉
+댊
+댋
+댌
+댍
+댎
+댏
+댐
+댑
+댒
+댓
+댔
+댕
+댖
+댗
+댘
+댙
+댚
+댛
+댜
+댝
+댞
+댟
+댠
+댡
+댢
+댣
+댤
+댥
+댦
+댧
+댨
+댩
+댪
+댫
+댬
+댭
+댮
+댯
+댰
+댱
+댲
+댳
+댴
+댵
+댶
+댷
+댸
+댹
+댺
+댻
+댼
+댽
+댾
+댿
+덀
+덁
+덂
+덃
+덄
+덅
+덆
+덇
+덈
+덉
+덊
+덋
+덌
+덍
+덎
+덏
+덐
+덑
+덒
+덓
+더
+덕
+덖
+덗
+던
+덙
+덚
+덛
+덜
+덝
+덞
+덟
+덠
+덡
+덢
+덣
+덤
+덥
+덦
+덧
+덨
+덩
+덪
+덫
+덬
+덭
+덮
+덯
+데
+덱
+덲
+덳
+덴
+덵
+덶
+덷
+델
+덹
+덺
+덻
+덼
+덽
+덾
+덿
+뎀
+뎁
+뎂
+뎃
+뎄
+뎅
+뎆
+뎇
+뎈
+뎉
+뎊
+뎋
+뎌
+뎍
+뎎
+뎏
+뎐
+뎑
+뎒
+뎓
+뎔
+뎕
+뎖
+뎗
+뎘
+뎙
+뎚
+뎛
+뎜
+뎝
+뎞
+뎟
+뎠
+뎡
+뎢
+뎣
+뎤
+뎥
+뎦
+뎧
+뎨
+뎩
+뎪
+뎫
+뎬
+뎭
+뎮
+뎯
+뎰
+뎱
+뎲
+뎳
+뎴
+뎵
+뎶
+뎷
+뎸
+뎹
+뎺
+뎻
+뎼
+뎽
+뎾
+뎿
+돀
+돁
+돂
+돃
+도
+독
+돆
+돇
+돈
+돉
+돊
+돋
+돌
+돍
+돎
+돏
+돐
+돑
+돒
+돓
+돔
+돕
+돖
+돗
+돘
+동
+돚
+돛
+돜
+돝
+돞
+돟
+돠
+돡
+돢
+돣
+돤
+돥
+돦
+돧
+돨
+돩
+돪
+돫
+돬
+돭
+돮
+돯
+돰
+돱
+돲
+돳
+돴
+돵
+돶
+돷
+돸
+돹
+돺
+돻
+돼
+돽
+돾
+돿
+됀
+됁
+됂
+됃
+됄
+됅
+됆
+됇
+됈
+됉
+됊
+됋
+됌
+됍
+됎
+됏
+됐
+됑
+됒
+됓
+됔
+됕
+됖
+됗
+되
+됙
+됚
+됛
+된
+됝
+됞
+됟
+될
+됡
+됢
+됣
+됤
+됥
+됦
+됧
+됨
+됩
+됪
+됫
+됬
+됭
+됮
+됯
+됰
+됱
+됲
+됳
+됴
+됵
+됶
+됷
+됸
+됹
+됺
+됻
+됼
+됽
+됾
+됿
+둀
+둁
+둂
+둃
+둄
+둅
+둆
+둇
+둈
+둉
+둊
+둋
+둌
+둍
+둎
+둏
+두
+둑
+둒
+둓
+둔
+둕
+둖
+둗
+둘
+둙
+둚
+둛
+둜
+둝
+둞
+둟
+둠
+둡
+둢
+둣
+둤
+둥
+둦
+둧
+둨
+둩
+둪
+둫
+둬
+둭
+둮
+둯
+둰
+둱
+둲
+둳
+둴
+둵
+둶
+둷
+둸
+둹
+둺
+둻
+둼
+둽
+둾
+둿
+뒀
+뒁
+뒂
+뒃
+뒄
+뒅
+뒆
+뒇
+뒈
+뒉
+뒊
+뒋
+뒌
+뒍
+뒎
+뒏
+뒐
+뒑
+뒒
+뒓
+뒔
+뒕
+뒖
+뒗
+뒘
+뒙
+뒚
+뒛
+뒜
+뒝
+뒞
+뒟
+뒠
+뒡
+뒢
+뒣
+뒤
+뒥
+뒦
+뒧
+뒨
+뒩
+뒪
+뒫
+뒬
+뒭
+뒮
+뒯
+뒰
+뒱
+뒲
+뒳
+뒴
+뒵
+뒶
+뒷
+뒸
+뒹
+뒺
+뒻
+뒼
+뒽
+뒾
+뒿
+듀
+듁
+듂
+듃
+듄
+듅
+듆
+듇
+듈
+듉
+듊
+듋
+듌
+듍
+듎
+듏
+듐
+듑
+듒
+듓
+듔
+듕
+듖
+듗
+듘
+듙
+듚
+듛
+드
+득
+듞
+듟
+든
+듡
+듢
+듣
+들
+듥
+듦
+듧
+듨
+듩
+듪
+듫
+듬
+듭
+듮
+듯
+듰
+등
+듲
+듳
+듴
+듵
+듶
+듷
+듸
+듹
+듺
+듻
+듼
+듽
+듾
+듿
+딀
+딁
+딂
+딃
+딄
+딅
+딆
+딇
+딈
+딉
+딊
+딋
+딌
+딍
+딎
+딏
+딐
+딑
+딒
+딓
+디
+딕
+딖
+딗
+딘
+딙
+딚
+딛
+딜
+딝
+딞
+딟
+딠
+딡
+딢
+딣
+딤
+딥
+딦
+딧
+딨
+딩
+딪
+딫
+딬
+딭
+딮
+딯
+따
+딱
+딲
+딳
+딴
+딵
+딶
+딷
+딸
+딹
+딺
+딻
+딼
+딽
+딾
+딿
+땀
+땁
+땂
+땃
+땄
+땅
+땆
+땇
+땈
+땉
+땊
+땋
+때
+땍
+땎
+땏
+땐
+땑
+땒
+땓
+땔
+땕
+땖
+땗
+땘
+땙
+땚
+땛
+땜
+땝
+땞
+땟
+땠
+땡
+땢
+땣
+땤
+땥
+땦
+땧
+땨
+땩
+땪
+땫
+땬
+땭
+땮
+땯
+땰
+땱
+땲
+땳
+땴
+땵
+땶
+땷
+땸
+땹
+땺
+땻
+땼
+땽
+땾
+땿
+떀
+떁
+떂
+떃
+떄
+떅
+떆
+떇
+떈
+떉
+떊
+떋
+떌
+떍
+떎
+떏
+떐
+떑
+떒
+떓
+떔
+떕
+떖
+떗
+떘
+떙
+떚
+떛
+떜
+떝
+떞
+떟
+떠
+떡
+떢
+떣
+떤
+떥
+떦
+떧
+떨
+떩
+떪
+떫
+떬
+떭
+떮
+떯
+떰
+떱
+떲
+떳
+떴
+떵
+떶
+떷
+떸
+떹
+떺
+떻
+떼
+떽
+떾
+떿
+뗀
+뗁
+뗂
+뗃
+뗄
+뗅
+뗆
+뗇
+뗈
+뗉
+뗊
+뗋
+뗌
+뗍
+뗎
+뗏
+뗐
+뗑
+뗒
+뗓
+뗔
+뗕
+뗖
+뗗
+뗘
+뗙
+뗚
+뗛
+뗜
+뗝
+뗞
+뗟
+뗠
+뗡
+뗢
+뗣
+뗤
+뗥
+뗦
+뗧
+뗨
+뗩
+뗪
+뗫
+뗬
+뗭
+뗮
+뗯
+뗰
+뗱
+뗲
+뗳
+뗴
+뗵
+뗶
+뗷
+뗸
+뗹
+뗺
+뗻
+뗼
+뗽
+뗾
+뗿
+똀
+똁
+똂
+똃
+똄
+똅
+똆
+똇
+똈
+똉
+똊
+똋
+똌
+똍
+똎
+똏
+또
+똑
+똒
+똓
+똔
+똕
+똖
+똗
+똘
+똙
+똚
+똛
+똜
+똝
+똞
+똟
+똠
+똡
+똢
+똣
+똤
+똥
+똦
+똧
+똨
+똩
+똪
+똫
+똬
+똭
+똮
+똯
+똰
+똱
+똲
+똳
+똴
+똵
+똶
+똷
+똸
+똹
+똺
+똻
+똼
+똽
+똾
+똿
+뙀
+뙁
+뙂
+뙃
+뙄
+뙅
+뙆
+뙇
+뙈
+뙉
+뙊
+뙋
+뙌
+뙍
+뙎
+뙏
+뙐
+뙑
+뙒
+뙓
+뙔
+뙕
+뙖
+뙗
+뙘
+뙙
+뙚
+뙛
+뙜
+뙝
+뙞
+뙟
+뙠
+뙡
+뙢
+뙣
+뙤
+뙥
+뙦
+뙧
+뙨
+뙩
+뙪
+뙫
+뙬
+뙭
+뙮
+뙯
+뙰
+뙱
+뙲
+뙳
+뙴
+뙵
+뙶
+뙷
+뙸
+뙹
+뙺
+뙻
+뙼
+뙽
+뙾
+뙿
+뚀
+뚁
+뚂
+뚃
+뚄
+뚅
+뚆
+뚇
+뚈
+뚉
+뚊
+뚋
+뚌
+뚍
+뚎
+뚏
+뚐
+뚑
+뚒
+뚓
+뚔
+뚕
+뚖
+뚗
+뚘
+뚙
+뚚
+뚛
+뚜
+뚝
+뚞
+뚟
+뚠
+뚡
+뚢
+뚣
+뚤
+뚥
+뚦
+뚧
+뚨
+뚩
+뚪
+뚫
+뚬
+뚭
+뚮
+뚯
+뚰
+뚱
+뚲
+뚳
+뚴
+뚵
+뚶
+뚷
+뚸
+뚹
+뚺
+뚻
+뚼
+뚽
+뚾
+뚿
+뛀
+뛁
+뛂
+뛃
+뛄
+뛅
+뛆
+뛇
+뛈
+뛉
+뛊
+뛋
+뛌
+뛍
+뛎
+뛏
+뛐
+뛑
+뛒
+뛓
+뛔
+뛕
+뛖
+뛗
+뛘
+뛙
+뛚
+뛛
+뛜
+뛝
+뛞
+뛟
+뛠
+뛡
+뛢
+뛣
+뛤
+뛥
+뛦
+뛧
+뛨
+뛩
+뛪
+뛫
+뛬
+뛭
+뛮
+뛯
+뛰
+뛱
+뛲
+뛳
+뛴
+뛵
+뛶
+뛷
+뛸
+뛹
+뛺
+뛻
+뛼
+뛽
+뛾
+뛿
+뜀
+뜁
+뜂
+뜃
+뜄
+뜅
+뜆
+뜇
+뜈
+뜉
+뜊
+뜋
+뜌
+뜍
+뜎
+뜏
+뜐
+뜑
+뜒
+뜓
+뜔
+뜕
+뜖
+뜗
+뜘
+뜙
+뜚
+뜛
+뜜
+뜝
+뜞
+뜟
+뜠
+뜡
+뜢
+뜣
+뜤
+뜥
+뜦
+뜧
+뜨
+뜩
+뜪
+뜫
+뜬
+뜭
+뜮
+뜯
+뜰
+뜱
+뜲
+뜳
+뜴
+뜵
+뜶
+뜷
+뜸
+뜹
+뜺
+뜻
+뜼
+뜽
+뜾
+뜿
+띀
+띁
+띂
+띃
+띄
+띅
+띆
+띇
+띈
+띉
+띊
+띋
+띌
+띍
+띎
+띏
+띐
+띑
+띒
+띓
+띔
+띕
+띖
+띗
+띘
+띙
+띚
+띛
+띜
+띝
+띞
+띟
+띠
+띡
+띢
+띣
+띤
+띥
+띦
+띧
+띨
+띩
+띪
+띫
+띬
+띭
+띮
+띯
+띰
+띱
+띲
+띳
+띴
+띵
+띶
+띷
+띸
+띹
+띺
+띻
+라
+락
+띾
+띿
+란
+랁
+랂
+랃
+랄
+랅
+랆
+랇
+랈
+랉
+랊
+랋
+람
+랍
+랎
+랏
+랐
+랑
+랒
+랓
+랔
+랕
+랖
+랗
+래
+랙
+랚
+랛
+랜
+랝
+랞
+랟
+랠
+랡
+랢
+랣
+랤
+랥
+랦
+랧
+램
+랩
+랪
+랫
+랬
+랭
+랮
+랯
+랰
+랱
+랲
+랳
+랴
+략
+랶
+랷
+랸
+랹
+랺
+랻
+랼
+랽
+랾
+랿
+럀
+럁
+럂
+럃
+럄
+럅
+럆
+럇
+럈
+량
+럊
+럋
+럌
+럍
+럎
+럏
+럐
+럑
+럒
+럓
+럔
+럕
+럖
+럗
+럘
+럙
+럚
+럛
+럜
+럝
+럞
+럟
+럠
+럡
+럢
+럣
+럤
+럥
+럦
+럧
+럨
+럩
+럪
+럫
+러
+럭
+럮
+럯
+런
+럱
+럲
+럳
+럴
+럵
+럶
+럷
+럸
+럹
+럺
+럻
+럼
+럽
+럾
+럿
+렀
+렁
+렂
+렃
+렄
+렅
+렆
+렇
+레
+렉
+렊
+렋
+렌
+렍
+렎
+렏
+렐
+렑
+렒
+렓
+렔
+렕
+렖
+렗
+렘
+렙
+렚
+렛
+렜
+렝
+렞
+렟
+렠
+렡
+렢
+렣
+려
+력
+렦
+렧
+련
+렩
+렪
+렫
+렬
+렭
+렮
+렯
+렰
+렱
+렲
+렳
+렴
+렵
+렶
+렷
+렸
+령
+렺
+렻
+렼
+렽
+렾
+렿
+례
+롁
+롂
+롃
+롄
+롅
+롆
+롇
+롈
+롉
+롊
+롋
+롌
+롍
+롎
+롏
+롐
+롑
+롒
+롓
+롔
+롕
+롖
+롗
+롘
+롙
+롚
+롛
+로
+록
+롞
+롟
+론
+롡
+롢
+롣
+롤
+롥
+롦
+롧
+롨
+롩
+롪
+롫
+롬
+롭
+롮
+롯
+롰
+롱
+롲
+롳
+롴
+롵
+롶
+롷
+롸
+롹
+롺
+롻
+롼
+롽
+롾
+롿
+뢀
+뢁
+뢂
+뢃
+뢄
+뢅
+뢆
+뢇
+뢈
+뢉
+뢊
+뢋
+뢌
+뢍
+뢎
+뢏
+뢐
+뢑
+뢒
+뢓
+뢔
+뢕
+뢖
+뢗
+뢘
+뢙
+뢚
+뢛
+뢜
+뢝
+뢞
+뢟
+뢠
+뢡
+뢢
+뢣
+뢤
+뢥
+뢦
+뢧
+뢨
+뢩
+뢪
+뢫
+뢬
+뢭
+뢮
+뢯
+뢰
+뢱
+뢲
+뢳
+뢴
+뢵
+뢶
+뢷
+뢸
+뢹
+뢺
+뢻
+뢼
+뢽
+뢾
+뢿
+룀
+룁
+룂
+룃
+룄
+룅
+룆
+룇
+룈
+룉
+룊
+룋
+료
+룍
+룎
+룏
+룐
+룑
+룒
+룓
+룔
+룕
+룖
+룗
+룘
+룙
+룚
+룛
+룜
+룝
+룞
+룟
+룠
+룡
+룢
+룣
+룤
+룥
+룦
+룧
+루
+룩
+룪
+룫
+룬
+룭
+룮
+룯
+룰
+룱
+룲
+룳
+룴
+룵
+룶
+룷
+룸
+룹
+룺
+룻
+룼
+룽
+룾
+룿
+뤀
+뤁
+뤂
+뤃
+뤄
+뤅
+뤆
+뤇
+뤈
+뤉
+뤊
+뤋
+뤌
+뤍
+뤎
+뤏
+뤐
+뤑
+뤒
+뤓
+뤔
+뤕
+뤖
+뤗
+뤘
+뤙
+뤚
+뤛
+뤜
+뤝
+뤞
+뤟
+뤠
+뤡
+뤢
+뤣
+뤤
+뤥
+뤦
+뤧
+뤨
+뤩
+뤪
+뤫
+뤬
+뤭
+뤮
+뤯
+뤰
+뤱
+뤲
+뤳
+뤴
+뤵
+뤶
+뤷
+뤸
+뤹
+뤺
+뤻
+뤼
+뤽
+뤾
+뤿
+륀
+륁
+륂
+륃
+륄
+륅
+륆
+륇
+륈
+륉
+륊
+륋
+륌
+륍
+륎
+륏
+륐
+륑
+륒
+륓
+륔
+륕
+륖
+륗
+류
+륙
+륚
+륛
+륜
+륝
+륞
+륟
+률
+륡
+륢
+륣
+륤
+륥
+륦
+륧
+륨
+륩
+륪
+륫
+륬
+륭
+륮
+륯
+륰
+륱
+륲
+륳
+르
+륵
+륶
+륷
+른
+륹
+륺
+륻
+를
+륽
+륾
+륿
+릀
+릁
+릂
+릃
+름
+릅
+릆
+릇
+릈
+릉
+릊
+릋
+릌
+릍
+릎
+릏
+릐
+릑
+릒
+릓
+릔
+릕
+릖
+릗
+릘
+릙
+릚
+릛
+릜
+릝
+릞
+릟
+릠
+릡
+릢
+릣
+릤
+릥
+릦
+릧
+릨
+릩
+릪
+릫
+리
+릭
+릮
+릯
+린
+릱
+릲
+릳
+릴
+릵
+릶
+릷
+릸
+릹
+릺
+릻
+림
+립
+릾
+릿
+맀
+링
+맂
+맃
+맄
+맅
+맆
+맇
+마
+막
+맊
+맋
+만
+맍
+많
+맏
+말
+맑
+맒
+맓
+맔
+맕
+맖
+맗
+맘
+맙
+맚
+맛
+맜
+망
+맞
+맟
+맠
+맡
+맢
+맣
+매
+맥
+맦
+맧
+맨
+맩
+맪
+맫
+맬
+맭
+맮
+맯
+맰
+맱
+맲
+맳
+맴
+맵
+맶
+맷
+맸
+맹
+맺
+맻
+맼
+맽
+맾
+맿
+먀
+먁
+먂
+먃
+먄
+먅
+먆
+먇
+먈
+먉
+먊
+먋
+먌
+먍
+먎
+먏
+먐
+먑
+먒
+먓
+먔
+먕
+먖
+먗
+먘
+먙
+먚
+먛
+먜
+먝
+먞
+먟
+먠
+먡
+먢
+먣
+먤
+먥
+먦
+먧
+먨
+먩
+먪
+먫
+먬
+먭
+먮
+먯
+먰
+먱
+먲
+먳
+먴
+먵
+먶
+먷
+머
+먹
+먺
+먻
+먼
+먽
+먾
+먿
+멀
+멁
+멂
+멃
+멄
+멅
+멆
+멇
+멈
+멉
+멊
+멋
+멌
+멍
+멎
+멏
+멐
+멑
+멒
+멓
+메
+멕
+멖
+멗
+멘
+멙
+멚
+멛
+멜
+멝
+멞
+멟
+멠
+멡
+멢
+멣
+멤
+멥
+멦
+멧
+멨
+멩
+멪
+멫
+멬
+멭
+멮
+멯
+며
+멱
+멲
+멳
+면
+멵
+멶
+멷
+멸
+멹
+멺
+멻
+멼
+멽
+멾
+멿
+몀
+몁
+몂
+몃
+몄
+명
+몆
+몇
+몈
+몉
+몊
+몋
+몌
+몍
+몎
+몏
+몐
+몑
+몒
+몓
+몔
+몕
+몖
+몗
+몘
+몙
+몚
+몛
+몜
+몝
+몞
+몟
+몠
+몡
+몢
+몣
+몤
+몥
+몦
+몧
+모
+목
+몪
+몫
+몬
+몭
+몮
+몯
+몰
+몱
+몲
+몳
+몴
+몵
+몶
+몷
+몸
+몹
+몺
+못
+몼
+몽
+몾
+몿
+뫀
+뫁
+뫂
+뫃
+뫄
+뫅
+뫆
+뫇
+뫈
+뫉
+뫊
+뫋
+뫌
+뫍
+뫎
+뫏
+뫐
+뫑
+뫒
+뫓
+뫔
+뫕
+뫖
+뫗
+뫘
+뫙
+뫚
+뫛
+뫜
+뫝
+뫞
+뫟
+뫠
+뫡
+뫢
+뫣
+뫤
+뫥
+뫦
+뫧
+뫨
+뫩
+뫪
+뫫
+뫬
+뫭
+뫮
+뫯
+뫰
+뫱
+뫲
+뫳
+뫴
+뫵
+뫶
+뫷
+뫸
+뫹
+뫺
+뫻
+뫼
+뫽
+뫾
+뫿
+묀
+묁
+묂
+묃
+묄
+묅
+묆
+묇
+묈
+묉
+묊
+묋
+묌
+묍
+묎
+묏
+묐
+묑
+묒
+묓
+묔
+묕
+묖
+묗
+묘
+묙
+묚
+묛
+묜
+묝
+묞
+묟
+묠
+묡
+묢
+묣
+묤
+묥
+묦
+묧
+묨
+묩
+묪
+묫
+묬
+묭
+묮
+묯
+묰
+묱
+묲
+묳
+무
+묵
+묶
+묷
+문
+묹
+묺
+묻
+물
+묽
+묾
+묿
+뭀
+뭁
+뭂
+뭃
+뭄
+뭅
+뭆
+뭇
+뭈
+뭉
+뭊
+뭋
+뭌
+뭍
+뭎
+뭏
+뭐
+뭑
+뭒
+뭓
+뭔
+뭕
+뭖
+뭗
+뭘
+뭙
+뭚
+뭛
+뭜
+뭝
+뭞
+뭟
+뭠
+뭡
+뭢
+뭣
+뭤
+뭥
+뭦
+뭧
+뭨
+뭩
+뭪
+뭫
+뭬
+뭭
+뭮
+뭯
+뭰
+뭱
+뭲
+뭳
+뭴
+뭵
+뭶
+뭷
+뭸
+뭹
+뭺
+뭻
+뭼
+뭽
+뭾
+뭿
+뮀
+뮁
+뮂
+뮃
+뮄
+뮅
+뮆
+뮇
+뮈
+뮉
+뮊
+뮋
+뮌
+뮍
+뮎
+뮏
+뮐
+뮑
+뮒
+뮓
+뮔
+뮕
+뮖
+뮗
+뮘
+뮙
+뮚
+뮛
+뮜
+뮝
+뮞
+뮟
+뮠
+뮡
+뮢
+뮣
+뮤
+뮥
+뮦
+뮧
+뮨
+뮩
+뮪
+뮫
+뮬
+뮭
+뮮
+뮯
+뮰
+뮱
+뮲
+뮳
+뮴
+뮵
+뮶
+뮷
+뮸
+뮹
+뮺
+뮻
+뮼
+뮽
+뮾
+뮿
+므
+믁
+믂
+믃
+믄
+믅
+믆
+믇
+믈
+믉
+믊
+믋
+믌
+믍
+믎
+믏
+믐
+믑
+믒
+믓
+믔
+믕
+믖
+믗
+믘
+믙
+믚
+믛
+믜
+믝
+믞
+믟
+믠
+믡
+믢
+믣
+믤
+믥
+믦
+믧
+믨
+믩
+믪
+믫
+믬
+믭
+믮
+믯
+믰
+믱
+믲
+믳
+믴
+믵
+믶
+믷
+미
+믹
+믺
+믻
+민
+믽
+믾
+믿
+밀
+밁
+밂
+밃
+밄
+밅
+밆
+밇
+밈
+밉
+밊
+밋
+밌
+밍
+밎
+및
+밐
+밑
+밒
+밓
+바
+박
+밖
+밗
+반
+밙
+밚
+받
+발
+밝
+밞
+밟
+밠
+밡
+밢
+밣
+밤
+밥
+밦
+밧
+밨
+방
+밪
+밫
+밬
+밭
+밮
+밯
+배
+백
+밲
+밳
+밴
+밵
+밶
+밷
+밸
+밹
+밺
+밻
+밼
+밽
+밾
+밿
+뱀
+뱁
+뱂
+뱃
+뱄
+뱅
+뱆
+뱇
+뱈
+뱉
+뱊
+뱋
+뱌
+뱍
+뱎
+뱏
+뱐
+뱑
+뱒
+뱓
+뱔
+뱕
+뱖
+뱗
+뱘
+뱙
+뱚
+뱛
+뱜
+뱝
+뱞
+뱟
+뱠
+뱡
+뱢
+뱣
+뱤
+뱥
+뱦
+뱧
+뱨
+뱩
+뱪
+뱫
+뱬
+뱭
+뱮
+뱯
+뱰
+뱱
+뱲
+뱳
+뱴
+뱵
+뱶
+뱷
+뱸
+뱹
+뱺
+뱻
+뱼
+뱽
+뱾
+뱿
+벀
+벁
+벂
+벃
+버
+벅
+벆
+벇
+번
+벉
+벊
+벋
+벌
+벍
+벎
+벏
+벐
+벑
+벒
+벓
+범
+법
+벖
+벗
+벘
+벙
+벚
+벛
+벜
+벝
+벞
+벟
+베
+벡
+벢
+벣
+벤
+벥
+벦
+벧
+벨
+벩
+벪
+벫
+벬
+벭
+벮
+벯
+벰
+벱
+벲
+벳
+벴
+벵
+벶
+벷
+벸
+벹
+벺
+벻
+벼
+벽
+벾
+벿
+변
+볁
+볂
+볃
+별
+볅
+볆
+볇
+볈
+볉
+볊
+볋
+볌
+볍
+볎
+볏
+볐
+병
+볒
+볓
+볔
+볕
+볖
+볗
+볘
+볙
+볚
+볛
+볜
+볝
+볞
+볟
+볠
+볡
+볢
+볣
+볤
+볥
+볦
+볧
+볨
+볩
+볪
+볫
+볬
+볭
+볮
+볯
+볰
+볱
+볲
+볳
+보
+복
+볶
+볷
+본
+볹
+볺
+볻
+볼
+볽
+볾
+볿
+봀
+봁
+봂
+봃
+봄
+봅
+봆
+봇
+봈
+봉
+봊
+봋
+봌
+봍
+봎
+봏
+봐
+봑
+봒
+봓
+봔
+봕
+봖
+봗
+봘
+봙
+봚
+봛
+봜
+봝
+봞
+봟
+봠
+봡
+봢
+봣
+봤
+봥
+봦
+봧
+봨
+봩
+봪
+봫
+봬
+봭
+봮
+봯
+봰
+봱
+봲
+봳
+봴
+봵
+봶
+봷
+봸
+봹
+봺
+봻
+봼
+봽
+봾
+봿
+뵀
+뵁
+뵂
+뵃
+뵄
+뵅
+뵆
+뵇
+뵈
+뵉
+뵊
+뵋
+뵌
+뵍
+뵎
+뵏
+뵐
+뵑
+뵒
+뵓
+뵔
+뵕
+뵖
+뵗
+뵘
+뵙
+뵚
+뵛
+뵜
+뵝
+뵞
+뵟
+뵠
+뵡
+뵢
+뵣
+뵤
+뵥
+뵦
+뵧
+뵨
+뵩
+뵪
+뵫
+뵬
+뵭
+뵮
+뵯
+뵰
+뵱
+뵲
+뵳
+뵴
+뵵
+뵶
+뵷
+뵸
+뵹
+뵺
+뵻
+뵼
+뵽
+뵾
+뵿
+부
+북
+붂
+붃
+분
+붅
+붆
+붇
+불
+붉
+붊
+붋
+붌
+붍
+붎
+붏
+붐
+붑
+붒
+붓
+붔
+붕
+붖
+붗
+붘
+붙
+붚
+붛
+붜
+붝
+붞
+붟
+붠
+붡
+붢
+붣
+붤
+붥
+붦
+붧
+붨
+붩
+붪
+붫
+붬
+붭
+붮
+붯
+붰
+붱
+붲
+붳
+붴
+붵
+붶
+붷
+붸
+붹
+붺
+붻
+붼
+붽
+붾
+붿
+뷀
+뷁
+뷂
+뷃
+뷄
+뷅
+뷆
+뷇
+뷈
+뷉
+뷊
+뷋
+뷌
+뷍
+뷎
+뷏
+뷐
+뷑
+뷒
+뷓
+뷔
+뷕
+뷖
+뷗
+뷘
+뷙
+뷚
+뷛
+뷜
+뷝
+뷞
+뷟
+뷠
+뷡
+뷢
+뷣
+뷤
+뷥
+뷦
+뷧
+뷨
+뷩
+뷪
+뷫
+뷬
+뷭
+뷮
+뷯
+뷰
+뷱
+뷲
+뷳
+뷴
+뷵
+뷶
+뷷
+뷸
+뷹
+뷺
+뷻
+뷼
+뷽
+뷾
+뷿
+븀
+븁
+븂
+븃
+븄
+븅
+븆
+븇
+븈
+븉
+븊
+븋
+브
+븍
+븎
+븏
+븐
+븑
+븒
+븓
+블
+븕
+븖
+븗
+븘
+븙
+븚
+븛
+븜
+븝
+븞
+븟
+븠
+븡
+븢
+븣
+븤
+븥
+븦
+븧
+븨
+븩
+븪
+븫
+븬
+븭
+븮
+븯
+븰
+븱
+븲
+븳
+븴
+븵
+븶
+븷
+븸
+븹
+븺
+븻
+븼
+븽
+븾
+븿
+빀
+빁
+빂
+빃
+비
+빅
+빆
+빇
+빈
+빉
+빊
+빋
+빌
+빍
+빎
+빏
+빐
+빑
+빒
+빓
+빔
+빕
+빖
+빗
+빘
+빙
+빚
+빛
+빜
+빝
+빞
+빟
+빠
+빡
+빢
+빣
+빤
+빥
+빦
+빧
+빨
+빩
+빪
+빫
+빬
+빭
+빮
+빯
+빰
+빱
+빲
+빳
+빴
+빵
+빶
+빷
+빸
+빹
+빺
+빻
+빼
+빽
+빾
+빿
+뺀
+뺁
+뺂
+뺃
+뺄
+뺅
+뺆
+뺇
+뺈
+뺉
+뺊
+뺋
+뺌
+뺍
+뺎
+뺏
+뺐
+뺑
+뺒
+뺓
+뺔
+뺕
+뺖
+뺗
+뺘
+뺙
+뺚
+뺛
+뺜
+뺝
+뺞
+뺟
+뺠
+뺡
+뺢
+뺣
+뺤
+뺥
+뺦
+뺧
+뺨
+뺩
+뺪
+뺫
+뺬
+뺭
+뺮
+뺯
+뺰
+뺱
+뺲
+뺳
+뺴
+뺵
+뺶
+뺷
+뺸
+뺹
+뺺
+뺻
+뺼
+뺽
+뺾
+뺿
+뻀
+뻁
+뻂
+뻃
+뻄
+뻅
+뻆
+뻇
+뻈
+뻉
+뻊
+뻋
+뻌
+뻍
+뻎
+뻏
+뻐
+뻑
+뻒
+뻓
+뻔
+뻕
+뻖
+뻗
+뻘
+뻙
+뻚
+뻛
+뻜
+뻝
+뻞
+뻟
+뻠
+뻡
+뻢
+뻣
+뻤
+뻥
+뻦
+뻧
+뻨
+뻩
+뻪
+뻫
+뻬
+뻭
+뻮
+뻯
+뻰
+뻱
+뻲
+뻳
+뻴
+뻵
+뻶
+뻷
+뻸
+뻹
+뻺
+뻻
+뻼
+뻽
+뻾
+뻿
+뼀
+뼁
+뼂
+뼃
+뼄
+뼅
+뼆
+뼇
+뼈
+뼉
+뼊
+뼋
+뼌
+뼍
+뼎
+뼏
+뼐
+뼑
+뼒
+뼓
+뼔
+뼕
+뼖
+뼗
+뼘
+뼙
+뼚
+뼛
+뼜
+뼝
+뼞
+뼟
+뼠
+뼡
+뼢
+뼣
+뼤
+뼥
+뼦
+뼧
+뼨
+뼩
+뼪
+뼫
+뼬
+뼭
+뼮
+뼯
+뼰
+뼱
+뼲
+뼳
+뼴
+뼵
+뼶
+뼷
+뼸
+뼹
+뼺
+뼻
+뼼
+뼽
+뼾
+뼿
+뽀
+뽁
+뽂
+뽃
+뽄
+뽅
+뽆
+뽇
+뽈
+뽉
+뽊
+뽋
+뽌
+뽍
+뽎
+뽏
+뽐
+뽑
+뽒
+뽓
+뽔
+뽕
+뽖
+뽗
+뽘
+뽙
+뽚
+뽛
+뽜
+뽝
+뽞
+뽟
+뽠
+뽡
+뽢
+뽣
+뽤
+뽥
+뽦
+뽧
+뽨
+뽩
+뽪
+뽫
+뽬
+뽭
+뽮
+뽯
+뽰
+뽱
+뽲
+뽳
+뽴
+뽵
+뽶
+뽷
+뽸
+뽹
+뽺
+뽻
+뽼
+뽽
+뽾
+뽿
+뾀
+뾁
+뾂
+뾃
+뾄
+뾅
+뾆
+뾇
+뾈
+뾉
+뾊
+뾋
+뾌
+뾍
+뾎
+뾏
+뾐
+뾑
+뾒
+뾓
+뾔
+뾕
+뾖
+뾗
+뾘
+뾙
+뾚
+뾛
+뾜
+뾝
+뾞
+뾟
+뾠
+뾡
+뾢
+뾣
+뾤
+뾥
+뾦
+뾧
+뾨
+뾩
+뾪
+뾫
+뾬
+뾭
+뾮
+뾯
+뾰
+뾱
+뾲
+뾳
+뾴
+뾵
+뾶
+뾷
+뾸
+뾹
+뾺
+뾻
+뾼
+뾽
+뾾
+뾿
+뿀
+뿁
+뿂
+뿃
+뿄
+뿅
+뿆
+뿇
+뿈
+뿉
+뿊
+뿋
+뿌
+뿍
+뿎
+뿏
+뿐
+뿑
+뿒
+뿓
+뿔
+뿕
+뿖
+뿗
+뿘
+뿙
+뿚
+뿛
+뿜
+뿝
+뿞
+뿟
+뿠
+뿡
+뿢
+뿣
+뿤
+뿥
+뿦
+뿧
+뿨
+뿩
+뿪
+뿫
+뿬
+뿭
+뿮
+뿯
+뿰
+뿱
+뿲
+뿳
+뿴
+뿵
+뿶
+뿷
+뿸
+뿹
+뿺
+뿻
+뿼
+뿽
+뿾
+뿿
+쀀
+쀁
+쀂
+쀃
+쀄
+쀅
+쀆
+쀇
+쀈
+쀉
+쀊
+쀋
+쀌
+쀍
+쀎
+쀏
+쀐
+쀑
+쀒
+쀓
+쀔
+쀕
+쀖
+쀗
+쀘
+쀙
+쀚
+쀛
+쀜
+쀝
+쀞
+쀟
+쀠
+쀡
+쀢
+쀣
+쀤
+쀥
+쀦
+쀧
+쀨
+쀩
+쀪
+쀫
+쀬
+쀭
+쀮
+쀯
+쀰
+쀱
+쀲
+쀳
+쀴
+쀵
+쀶
+쀷
+쀸
+쀹
+쀺
+쀻
+쀼
+쀽
+쀾
+쀿
+쁀
+쁁
+쁂
+쁃
+쁄
+쁅
+쁆
+쁇
+쁈
+쁉
+쁊
+쁋
+쁌
+쁍
+쁎
+쁏
+쁐
+쁑
+쁒
+쁓
+쁔
+쁕
+쁖
+쁗
+쁘
+쁙
+쁚
+쁛
+쁜
+쁝
+쁞
+쁟
+쁠
+쁡
+쁢
+쁣
+쁤
+쁥
+쁦
+쁧
+쁨
+쁩
+쁪
+쁫
+쁬
+쁭
+쁮
+쁯
+쁰
+쁱
+쁲
+쁳
+쁴
+쁵
+쁶
+쁷
+쁸
+쁹
+쁺
+쁻
+쁼
+쁽
+쁾
+쁿
+삀
+삁
+삂
+삃
+삄
+삅
+삆
+삇
+삈
+삉
+삊
+삋
+삌
+삍
+삎
+삏
+삐
+삑
+삒
+삓
+삔
+삕
+삖
+삗
+삘
+삙
+삚
+삛
+삜
+삝
+삞
+삟
+삠
+삡
+삢
+삣
+삤
+삥
+삦
+삧
+삨
+삩
+삪
+삫
+사
+삭
+삮
+삯
+산
+삱
+삲
+삳
+살
+삵
+삶
+삷
+삸
+삹
+삺
+삻
+삼
+삽
+삾
+삿
+샀
+상
+샂
+샃
+샄
+샅
+샆
+샇
+새
+색
+샊
+샋
+샌
+샍
+샎
+샏
+샐
+샑
+샒
+샓
+샔
+샕
+샖
+샗
+샘
+샙
+샚
+샛
+샜
+생
+샞
+샟
+샠
+샡
+샢
+샣
+샤
+샥
+샦
+샧
+샨
+샩
+샪
+샫
+샬
+샭
+샮
+샯
+샰
+샱
+샲
+샳
+샴
+샵
+샶
+샷
+샸
+샹
+샺
+샻
+샼
+샽
+샾
+샿
+섀
+섁
+섂
+섃
+섄
+섅
+섆
+섇
+섈
+섉
+섊
+섋
+섌
+섍
+섎
+섏
+섐
+섑
+섒
+섓
+섔
+섕
+섖
+섗
+섘
+섙
+섚
+섛
+서
+석
+섞
+섟
+선
+섡
+섢
+섣
+설
+섥
+섦
+섧
+섨
+섩
+섪
+섫
+섬
+섭
+섮
+섯
+섰
+성
+섲
+섳
+섴
+섵
+섶
+섷
+세
+섹
+섺
+섻
+센
+섽
+섾
+섿
+셀
+셁
+셂
+셃
+셄
+셅
+셆
+셇
+셈
+셉
+셊
+셋
+셌
+셍
+셎
+셏
+셐
+셑
+셒
+셓
+셔
+셕
+셖
+셗
+션
+셙
+셚
+셛
+셜
+셝
+셞
+셟
+셠
+셡
+셢
+셣
+셤
+셥
+셦
+셧
+셨
+셩
+셪
+셫
+셬
+셭
+셮
+셯
+셰
+셱
+셲
+셳
+셴
+셵
+셶
+셷
+셸
+셹
+셺
+셻
+셼
+셽
+셾
+셿
+솀
+솁
+솂
+솃
+솄
+솅
+솆
+솇
+솈
+솉
+솊
+솋
+소
+속
+솎
+솏
+손
+솑
+솒
+솓
+솔
+솕
+솖
+솗
+솘
+솙
+솚
+솛
+솜
+솝
+솞
+솟
+솠
+송
+솢
+솣
+솤
+솥
+솦
+솧
+솨
+솩
+솪
+솫
+솬
+솭
+솮
+솯
+솰
+솱
+솲
+솳
+솴
+솵
+솶
+솷
+솸
+솹
+솺
+솻
+솼
+솽
+솾
+솿
+쇀
+쇁
+쇂
+쇃
+쇄
+쇅
+쇆
+쇇
+쇈
+쇉
+쇊
+쇋
+쇌
+쇍
+쇎
+쇏
+쇐
+쇑
+쇒
+쇓
+쇔
+쇕
+쇖
+쇗
+쇘
+쇙
+쇚
+쇛
+쇜
+쇝
+쇞
+쇟
+쇠
+쇡
+쇢
+쇣
+쇤
+쇥
+쇦
+쇧
+쇨
+쇩
+쇪
+쇫
+쇬
+쇭
+쇮
+쇯
+쇰
+쇱
+쇲
+쇳
+쇴
+쇵
+쇶
+쇷
+쇸
+쇹
+쇺
+쇻
+쇼
+쇽
+쇾
+쇿
+숀
+숁
+숂
+숃
+숄
+숅
+숆
+숇
+숈
+숉
+숊
+숋
+숌
+숍
+숎
+숏
+숐
+숑
+숒
+숓
+숔
+숕
+숖
+숗
+수
+숙
+숚
+숛
+순
+숝
+숞
+숟
+술
+숡
+숢
+숣
+숤
+숥
+숦
+숧
+숨
+숩
+숪
+숫
+숬
+숭
+숮
+숯
+숰
+숱
+숲
+숳
+숴
+숵
+숶
+숷
+숸
+숹
+숺
+숻
+숼
+숽
+숾
+숿
+쉀
+쉁
+쉂
+쉃
+쉄
+쉅
+쉆
+쉇
+쉈
+쉉
+쉊
+쉋
+쉌
+쉍
+쉎
+쉏
+쉐
+쉑
+쉒
+쉓
+쉔
+쉕
+쉖
+쉗
+쉘
+쉙
+쉚
+쉛
+쉜
+쉝
+쉞
+쉟
+쉠
+쉡
+쉢
+쉣
+쉤
+쉥
+쉦
+쉧
+쉨
+쉩
+쉪
+쉫
+쉬
+쉭
+쉮
+쉯
+쉰
+쉱
+쉲
+쉳
+쉴
+쉵
+쉶
+쉷
+쉸
+쉹
+쉺
+쉻
+쉼
+쉽
+쉾
+쉿
+슀
+슁
+슂
+슃
+슄
+슅
+슆
+슇
+슈
+슉
+슊
+슋
+슌
+슍
+슎
+슏
+슐
+슑
+슒
+슓
+슔
+슕
+슖
+슗
+슘
+슙
+슚
+슛
+슜
+슝
+슞
+슟
+슠
+슡
+슢
+슣
+스
+슥
+슦
+슧
+슨
+슩
+슪
+슫
+슬
+슭
+슮
+슯
+슰
+슱
+슲
+슳
+슴
+습
+슶
+슷
+슸
+승
+슺
+슻
+슼
+슽
+슾
+슿
+싀
+싁
+싂
+싃
+싄
+싅
+싆
+싇
+싈
+싉
+싊
+싋
+싌
+싍
+싎
+싏
+싐
+싑
+싒
+싓
+싔
+싕
+싖
+싗
+싘
+싙
+싚
+싛
+시
+식
+싞
+싟
+신
+싡
+싢
+싣
+실
+싥
+싦
+싧
+싨
+싩
+싪
+싫
+심
+십
+싮
+싯
+싰
+싱
+싲
+싳
+싴
+싵
+싶
+싷
+싸
+싹
+싺
+싻
+싼
+싽
+싾
+싿
+쌀
+쌁
+쌂
+쌃
+쌄
+쌅
+쌆
+쌇
+쌈
+쌉
+쌊
+쌋
+쌌
+쌍
+쌎
+쌏
+쌐
+쌑
+쌒
+쌓
+쌔
+쌕
+쌖
+쌗
+쌘
+쌙
+쌚
+쌛
+쌜
+쌝
+쌞
+쌟
+쌠
+쌡
+쌢
+쌣
+쌤
+쌥
+쌦
+쌧
+쌨
+쌩
+쌪
+쌫
+쌬
+쌭
+쌮
+쌯
+쌰
+쌱
+쌲
+쌳
+쌴
+쌵
+쌶
+쌷
+쌸
+쌹
+쌺
+쌻
+쌼
+쌽
+쌾
+쌿
+썀
+썁
+썂
+썃
+썄
+썅
+썆
+썇
+썈
+썉
+썊
+썋
+썌
+썍
+썎
+썏
+썐
+썑
+썒
+썓
+썔
+썕
+썖
+썗
+썘
+썙
+썚
+썛
+썜
+썝
+썞
+썟
+썠
+썡
+썢
+썣
+썤
+썥
+썦
+썧
+써
+썩
+썪
+썫
+썬
+썭
+썮
+썯
+썰
+썱
+썲
+썳
+썴
+썵
+썶
+썷
+썸
+썹
+썺
+썻
+썼
+썽
+썾
+썿
+쎀
+쎁
+쎂
+쎃
+쎄
+쎅
+쎆
+쎇
+쎈
+쎉
+쎊
+쎋
+쎌
+쎍
+쎎
+쎏
+쎐
+쎑
+쎒
+쎓
+쎔
+쎕
+쎖
+쎗
+쎘
+쎙
+쎚
+쎛
+쎜
+쎝
+쎞
+쎟
+쎠
+쎡
+쎢
+쎣
+쎤
+쎥
+쎦
+쎧
+쎨
+쎩
+쎪
+쎫
+쎬
+쎭
+쎮
+쎯
+쎰
+쎱
+쎲
+쎳
+쎴
+쎵
+쎶
+쎷
+쎸
+쎹
+쎺
+쎻
+쎼
+쎽
+쎾
+쎿
+쏀
+쏁
+쏂
+쏃
+쏄
+쏅
+쏆
+쏇
+쏈
+쏉
+쏊
+쏋
+쏌
+쏍
+쏎
+쏏
+쏐
+쏑
+쏒
+쏓
+쏔
+쏕
+쏖
+쏗
+쏘
+쏙
+쏚
+쏛
+쏜
+쏝
+쏞
+쏟
+쏠
+쏡
+쏢
+쏣
+쏤
+쏥
+쏦
+쏧
+쏨
+쏩
+쏪
+쏫
+쏬
+쏭
+쏮
+쏯
+쏰
+쏱
+쏲
+쏳
+쏴
+쏵
+쏶
+쏷
+쏸
+쏹
+쏺
+쏻
+쏼
+쏽
+쏾
+쏿
+쐀
+쐁
+쐂
+쐃
+쐄
+쐅
+쐆
+쐇
+쐈
+쐉
+쐊
+쐋
+쐌
+쐍
+쐎
+쐏
+쐐
+쐑
+쐒
+쐓
+쐔
+쐕
+쐖
+쐗
+쐘
+쐙
+쐚
+쐛
+쐜
+쐝
+쐞
+쐟
+쐠
+쐡
+쐢
+쐣
+쐤
+쐥
+쐦
+쐧
+쐨
+쐩
+쐪
+쐫
+쐬
+쐭
+쐮
+쐯
+쐰
+쐱
+쐲
+쐳
+쐴
+쐵
+쐶
+쐷
+쐸
+쐹
+쐺
+쐻
+쐼
+쐽
+쐾
+쐿
+쑀
+쑁
+쑂
+쑃
+쑄
+쑅
+쑆
+쑇
+쑈
+쑉
+쑊
+쑋
+쑌
+쑍
+쑎
+쑏
+쑐
+쑑
+쑒
+쑓
+쑔
+쑕
+쑖
+쑗
+쑘
+쑙
+쑚
+쑛
+쑜
+쑝
+쑞
+쑟
+쑠
+쑡
+쑢
+쑣
+쑤
+쑥
+쑦
+쑧
+쑨
+쑩
+쑪
+쑫
+쑬
+쑭
+쑮
+쑯
+쑰
+쑱
+쑲
+쑳
+쑴
+쑵
+쑶
+쑷
+쑸
+쑹
+쑺
+쑻
+쑼
+쑽
+쑾
+쑿
+쒀
+쒁
+쒂
+쒃
+쒄
+쒅
+쒆
+쒇
+쒈
+쒉
+쒊
+쒋
+쒌
+쒍
+쒎
+쒏
+쒐
+쒑
+쒒
+쒓
+쒔
+쒕
+쒖
+쒗
+쒘
+쒙
+쒚
+쒛
+쒜
+쒝
+쒞
+쒟
+쒠
+쒡
+쒢
+쒣
+쒤
+쒥
+쒦
+쒧
+쒨
+쒩
+쒪
+쒫
+쒬
+쒭
+쒮
+쒯
+쒰
+쒱
+쒲
+쒳
+쒴
+쒵
+쒶
+쒷
+쒸
+쒹
+쒺
+쒻
+쒼
+쒽
+쒾
+쒿
+쓀
+쓁
+쓂
+쓃
+쓄
+쓅
+쓆
+쓇
+쓈
+쓉
+쓊
+쓋
+쓌
+쓍
+쓎
+쓏
+쓐
+쓑
+쓒
+쓓
+쓔
+쓕
+쓖
+쓗
+쓘
+쓙
+쓚
+쓛
+쓜
+쓝
+쓞
+쓟
+쓠
+쓡
+쓢
+쓣
+쓤
+쓥
+쓦
+쓧
+쓨
+쓩
+쓪
+쓫
+쓬
+쓭
+쓮
+쓯
+쓰
+쓱
+쓲
+쓳
+쓴
+쓵
+쓶
+쓷
+쓸
+쓹
+쓺
+쓻
+쓼
+쓽
+쓾
+쓿
+씀
+씁
+씂
+씃
+씄
+씅
+씆
+씇
+씈
+씉
+씊
+씋
+씌
+씍
+씎
+씏
+씐
+씑
+씒
+씓
+씔
+씕
+씖
+씗
+씘
+씙
+씚
+씛
+씜
+씝
+씞
+씟
+씠
+씡
+씢
+씣
+씤
+씥
+씦
+씧
+씨
+씩
+씪
+씫
+씬
+씭
+씮
+씯
+씰
+씱
+씲
+씳
+씴
+씵
+씶
+씷
+씸
+씹
+씺
+씻
+씼
+씽
+씾
+씿
+앀
+앁
+앂
+앃
+아
+악
+앆
+앇
+안
+앉
+않
+앋
+알
+앍
+앎
+앏
+앐
+앑
+앒
+앓
+암
+압
+앖
+앗
+았
+앙
+앚
+앛
+앜
+앝
+앞
+앟
+애
+액
+앢
+앣
+앤
+앥
+앦
+앧
+앨
+앩
+앪
+앫
+앬
+앭
+앮
+앯
+앰
+앱
+앲
+앳
+앴
+앵
+앶
+앷
+앸
+앹
+앺
+앻
+야
+약
+앾
+앿
+얀
+얁
+얂
+얃
+얄
+얅
+얆
+얇
+얈
+얉
+얊
+얋
+얌
+얍
+얎
+얏
+얐
+양
+얒
+얓
+얔
+얕
+얖
+얗
+얘
+얙
+얚
+얛
+얜
+얝
+얞
+얟
+얠
+얡
+얢
+얣
+얤
+얥
+얦
+얧
+얨
+얩
+얪
+얫
+얬
+얭
+얮
+얯
+얰
+얱
+얲
+얳
+어
+억
+얶
+얷
+언
+얹
+얺
+얻
+얼
+얽
+얾
+얿
+엀
+엁
+엂
+엃
+엄
+업
+없
+엇
+었
+엉
+엊
+엋
+엌
+엍
+엎
+엏
+에
+엑
+엒
+엓
+엔
+엕
+엖
+엗
+엘
+엙
+엚
+엛
+엜
+엝
+엞
+엟
+엠
+엡
+엢
+엣
+엤
+엥
+엦
+엧
+엨
+엩
+엪
+엫
+여
+역
+엮
+엯
+연
+엱
+엲
+엳
+열
+엵
+엶
+엷
+엸
+엹
+엺
+엻
+염
+엽
+엾
+엿
+였
+영
+옂
+옃
+옄
+옅
+옆
+옇
+예
+옉
+옊
+옋
+옌
+옍
+옎
+옏
+옐
+옑
+옒
+옓
+옔
+옕
+옖
+옗
+옘
+옙
+옚
+옛
+옜
+옝
+옞
+옟
+옠
+옡
+옢
+옣
+오
+옥
+옦
+옧
+온
+옩
+옪
+옫
+올
+옭
+옮
+옯
+옰
+옱
+옲
+옳
+옴
+옵
+옶
+옷
+옸
+옹
+옺
+옻
+옼
+옽
+옾
+옿
+와
+왁
+왂
+왃
+완
+왅
+왆
+왇
+왈
+왉
+왊
+왋
+왌
+왍
+왎
+왏
+왐
+왑
+왒
+왓
+왔
+왕
+왖
+왗
+왘
+왙
+왚
+왛
+왜
+왝
+왞
+왟
+왠
+왡
+왢
+왣
+왤
+왥
+왦
+왧
+왨
+왩
+왪
+왫
+왬
+왭
+왮
+왯
+왰
+왱
+왲
+왳
+왴
+왵
+왶
+왷
+외
+왹
+왺
+왻
+왼
+왽
+왾
+왿
+욀
+욁
+욂
+욃
+욄
+욅
+욆
+욇
+욈
+욉
+욊
+욋
+욌
+욍
+욎
+욏
+욐
+욑
+욒
+욓
+요
+욕
+욖
+욗
+욘
+욙
+욚
+욛
+욜
+욝
+욞
+욟
+욠
+욡
+욢
+욣
+욤
+욥
+욦
+욧
+욨
+용
+욪
+욫
+욬
+욭
+욮
+욯
+우
+욱
+욲
+욳
+운
+욵
+욶
+욷
+울
+욹
+욺
+욻
+욼
+욽
+욾
+욿
+움
+웁
+웂
+웃
+웄
+웅
+웆
+웇
+웈
+웉
+웊
+웋
+워
+웍
+웎
+웏
+원
+웑
+웒
+웓
+월
+웕
+웖
+웗
+웘
+웙
+웚
+웛
+웜
+웝
+웞
+웟
+웠
+웡
+웢
+웣
+웤
+웥
+웦
+웧
+웨
+웩
+웪
+웫
+웬
+웭
+웮
+웯
+웰
+웱
+웲
+웳
+웴
+웵
+웶
+웷
+웸
+웹
+웺
+웻
+웼
+웽
+웾
+웿
+윀
+윁
+윂
+윃
+위
+윅
+윆
+윇
+윈
+윉
+윊
+윋
+윌
+윍
+윎
+윏
+윐
+윑
+윒
+윓
+윔
+윕
+윖
+윗
+윘
+윙
+윚
+윛
+윜
+윝
+윞
+윟
+유
+육
+윢
+윣
+윤
+윥
+윦
+윧
+율
+윩
+윪
+윫
+윬
+윭
+윮
+윯
+윰
+윱
+윲
+윳
+윴
+융
+윶
+윷
+윸
+윹
+윺
+윻
+으
+윽
+윾
+윿
+은
+읁
+읂
+읃
+을
+읅
+읆
+읇
+읈
+읉
+읊
+읋
+음
+읍
+읎
+읏
+읐
+응
+읒
+읓
+읔
+읕
+읖
+읗
+의
+읙
+읚
+읛
+읜
+읝
+읞
+읟
+읠
+읡
+읢
+읣
+읤
+읥
+읦
+읧
+읨
+읩
+읪
+읫
+읬
+읭
+읮
+읯
+읰
+읱
+읲
+읳
+이
+익
+읶
+읷
+인
+읹
+읺
+읻
+일
+읽
+읾
+읿
+잀
+잁
+잂
+잃
+임
+입
+잆
+잇
+있
+잉
+잊
+잋
+잌
+잍
+잎
+잏
+자
+작
+잒
+잓
+잔
+잕
+잖
+잗
+잘
+잙
+잚
+잛
+잜
+잝
+잞
+잟
+잠
+잡
+잢
+잣
+잤
+장
+잦
+잧
+잨
+잩
+잪
+잫
+재
+잭
+잮
+잯
+잰
+잱
+잲
+잳
+잴
+잵
+잶
+잷
+잸
+잹
+잺
+잻
+잼
+잽
+잾
+잿
+쟀
+쟁
+쟂
+쟃
+쟄
+쟅
+쟆
+쟇
+쟈
+쟉
+쟊
+쟋
+쟌
+쟍
+쟎
+쟏
+쟐
+쟑
+쟒
+쟓
+쟔
+쟕
+쟖
+쟗
+쟘
+쟙
+쟚
+쟛
+쟜
+쟝
+쟞
+쟟
+쟠
+쟡
+쟢
+쟣
+쟤
+쟥
+쟦
+쟧
+쟨
+쟩
+쟪
+쟫
+쟬
+쟭
+쟮
+쟯
+쟰
+쟱
+쟲
+쟳
+쟴
+쟵
+쟶
+쟷
+쟸
+쟹
+쟺
+쟻
+쟼
+쟽
+쟾
+쟿
+저
+적
+젂
+젃
+전
+젅
+젆
+젇
+절
+젉
+젊
+젋
+젌
+젍
+젎
+젏
+점
+접
+젒
+젓
+젔
+정
+젖
+젗
+젘
+젙
+젚
+젛
+제
+젝
+젞
+젟
+젠
+젡
+젢
+젣
+젤
+젥
+젦
+젧
+젨
+젩
+젪
+젫
+젬
+젭
+젮
+젯
+젰
+젱
+젲
+젳
+젴
+젵
+젶
+젷
+져
+젹
+젺
+젻
+젼
+젽
+젾
+젿
+졀
+졁
+졂
+졃
+졄
+졅
+졆
+졇
+졈
+졉
+졊
+졋
+졌
+졍
+졎
+졏
+졐
+졑
+졒
+졓
+졔
+졕
+졖
+졗
+졘
+졙
+졚
+졛
+졜
+졝
+졞
+졟
+졠
+졡
+졢
+졣
+졤
+졥
+졦
+졧
+졨
+졩
+졪
+졫
+졬
+졭
+졮
+졯
+조
+족
+졲
+졳
+존
+졵
+졶
+졷
+졸
+졹
+졺
+졻
+졼
+졽
+졾
+졿
+좀
+좁
+좂
+좃
+좄
+종
+좆
+좇
+좈
+좉
+좊
+좋
+좌
+좍
+좎
+좏
+좐
+좑
+좒
+좓
+좔
+좕
+좖
+좗
+좘
+좙
+좚
+좛
+좜
+좝
+좞
+좟
+좠
+좡
+좢
+좣
+좤
+좥
+좦
+좧
+좨
+좩
+좪
+좫
+좬
+좭
+좮
+좯
+좰
+좱
+좲
+좳
+좴
+좵
+좶
+좷
+좸
+좹
+좺
+좻
+좼
+좽
+좾
+좿
+죀
+죁
+죂
+죃
+죄
+죅
+죆
+죇
+죈
+죉
+죊
+죋
+죌
+죍
+죎
+죏
+죐
+죑
+죒
+죓
+죔
+죕
+죖
+죗
+죘
+죙
+죚
+죛
+죜
+죝
+죞
+죟
+죠
+죡
+죢
+죣
+죤
+죥
+죦
+죧
+죨
+죩
+죪
+죫
+죬
+죭
+죮
+죯
+죰
+죱
+죲
+죳
+죴
+죵
+죶
+죷
+죸
+죹
+죺
+죻
+주
+죽
+죾
+죿
+준
+줁
+줂
+줃
+줄
+줅
+줆
+줇
+줈
+줉
+줊
+줋
+줌
+줍
+줎
+줏
+줐
+중
+줒
+줓
+줔
+줕
+줖
+줗
+줘
+줙
+줚
+줛
+줜
+줝
+줞
+줟
+줠
+줡
+줢
+줣
+줤
+줥
+줦
+줧
+줨
+줩
+줪
+줫
+줬
+줭
+줮
+줯
+줰
+줱
+줲
+줳
+줴
+줵
+줶
+줷
+줸
+줹
+줺
+줻
+줼
+줽
+줾
+줿
+쥀
+쥁
+쥂
+쥃
+쥄
+쥅
+쥆
+쥇
+쥈
+쥉
+쥊
+쥋
+쥌
+쥍
+쥎
+쥏
+쥐
+쥑
+쥒
+쥓
+쥔
+쥕
+쥖
+쥗
+쥘
+쥙
+쥚
+쥛
+쥜
+쥝
+쥞
+쥟
+쥠
+쥡
+쥢
+쥣
+쥤
+쥥
+쥦
+쥧
+쥨
+쥩
+쥪
+쥫
+쥬
+쥭
+쥮
+쥯
+쥰
+쥱
+쥲
+쥳
+쥴
+쥵
+쥶
+쥷
+쥸
+쥹
+쥺
+쥻
+쥼
+쥽
+쥾
+쥿
+즀
+즁
+즂
+즃
+즄
+즅
+즆
+즇
+즈
+즉
+즊
+즋
+즌
+즍
+즎
+즏
+즐
+즑
+즒
+즓
+즔
+즕
+즖
+즗
+즘
+즙
+즚
+즛
+즜
+증
+즞
+즟
+즠
+즡
+즢
+즣
+즤
+즥
+즦
+즧
+즨
+즩
+즪
+즫
+즬
+즭
+즮
+즯
+즰
+즱
+즲
+즳
+즴
+즵
+즶
+즷
+즸
+즹
+즺
+즻
+즼
+즽
+즾
+즿
+지
+직
+짂
+짃
+진
+짅
+짆
+짇
+질
+짉
+짊
+짋
+짌
+짍
+짎
+짏
+짐
+집
+짒
+짓
+짔
+징
+짖
+짗
+짘
+짙
+짚
+짛
+짜
+짝
+짞
+짟
+짠
+짡
+짢
+짣
+짤
+짥
+짦
+짧
+짨
+짩
+짪
+짫
+짬
+짭
+짮
+짯
+짰
+짱
+짲
+짳
+짴
+짵
+짶
+짷
+째
+짹
+짺
+짻
+짼
+짽
+짾
+짿
+쨀
+쨁
+쨂
+쨃
+쨄
+쨅
+쨆
+쨇
+쨈
+쨉
+쨊
+쨋
+쨌
+쨍
+쨎
+쨏
+쨐
+쨑
+쨒
+쨓
+쨔
+쨕
+쨖
+쨗
+쨘
+쨙
+쨚
+쨛
+쨜
+쨝
+쨞
+쨟
+쨠
+쨡
+쨢
+쨣
+쨤
+쨥
+쨦
+쨧
+쨨
+쨩
+쨪
+쨫
+쨬
+쨭
+쨮
+쨯
+쨰
+쨱
+쨲
+쨳
+쨴
+쨵
+쨶
+쨷
+쨸
+쨹
+쨺
+쨻
+쨼
+쨽
+쨾
+쨿
+쩀
+쩁
+쩂
+쩃
+쩄
+쩅
+쩆
+쩇
+쩈
+쩉
+쩊
+쩋
+쩌
+쩍
+쩎
+쩏
+쩐
+쩑
+쩒
+쩓
+쩔
+쩕
+쩖
+쩗
+쩘
+쩙
+쩚
+쩛
+쩜
+쩝
+쩞
+쩟
+쩠
+쩡
+쩢
+쩣
+쩤
+쩥
+쩦
+쩧
+쩨
+쩩
+쩪
+쩫
+쩬
+쩭
+쩮
+쩯
+쩰
+쩱
+쩲
+쩳
+쩴
+쩵
+쩶
+쩷
+쩸
+쩹
+쩺
+쩻
+쩼
+쩽
+쩾
+쩿
+쪀
+쪁
+쪂
+쪃
+쪄
+쪅
+쪆
+쪇
+쪈
+쪉
+쪊
+쪋
+쪌
+쪍
+쪎
+쪏
+쪐
+쪑
+쪒
+쪓
+쪔
+쪕
+쪖
+쪗
+쪘
+쪙
+쪚
+쪛
+쪜
+쪝
+쪞
+쪟
+쪠
+쪡
+쪢
+쪣
+쪤
+쪥
+쪦
+쪧
+쪨
+쪩
+쪪
+쪫
+쪬
+쪭
+쪮
+쪯
+쪰
+쪱
+쪲
+쪳
+쪴
+쪵
+쪶
+쪷
+쪸
+쪹
+쪺
+쪻
+쪼
+쪽
+쪾
+쪿
+쫀
+쫁
+쫂
+쫃
+쫄
+쫅
+쫆
+쫇
+쫈
+쫉
+쫊
+쫋
+쫌
+쫍
+쫎
+쫏
+쫐
+쫑
+쫒
+쫓
+쫔
+쫕
+쫖
+쫗
+쫘
+쫙
+쫚
+쫛
+쫜
+쫝
+쫞
+쫟
+쫠
+쫡
+쫢
+쫣
+쫤
+쫥
+쫦
+쫧
+쫨
+쫩
+쫪
+쫫
+쫬
+쫭
+쫮
+쫯
+쫰
+쫱
+쫲
+쫳
+쫴
+쫵
+쫶
+쫷
+쫸
+쫹
+쫺
+쫻
+쫼
+쫽
+쫾
+쫿
+쬀
+쬁
+쬂
+쬃
+쬄
+쬅
+쬆
+쬇
+쬈
+쬉
+쬊
+쬋
+쬌
+쬍
+쬎
+쬏
+쬐
+쬑
+쬒
+쬓
+쬔
+쬕
+쬖
+쬗
+쬘
+쬙
+쬚
+쬛
+쬜
+쬝
+쬞
+쬟
+쬠
+쬡
+쬢
+쬣
+쬤
+쬥
+쬦
+쬧
+쬨
+쬩
+쬪
+쬫
+쬬
+쬭
+쬮
+쬯
+쬰
+쬱
+쬲
+쬳
+쬴
+쬵
+쬶
+쬷
+쬸
+쬹
+쬺
+쬻
+쬼
+쬽
+쬾
+쬿
+쭀
+쭁
+쭂
+쭃
+쭄
+쭅
+쭆
+쭇
+쭈
+쭉
+쭊
+쭋
+쭌
+쭍
+쭎
+쭏
+쭐
+쭑
+쭒
+쭓
+쭔
+쭕
+쭖
+쭗
+쭘
+쭙
+쭚
+쭛
+쭜
+쭝
+쭞
+쭟
+쭠
+쭡
+쭢
+쭣
+쭤
+쭥
+쭦
+쭧
+쭨
+쭩
+쭪
+쭫
+쭬
+쭭
+쭮
+쭯
+쭰
+쭱
+쭲
+쭳
+쭴
+쭵
+쭶
+쭷
+쭸
+쭹
+쭺
+쭻
+쭼
+쭽
+쭾
+쭿
+쮀
+쮁
+쮂
+쮃
+쮄
+쮅
+쮆
+쮇
+쮈
+쮉
+쮊
+쮋
+쮌
+쮍
+쮎
+쮏
+쮐
+쮑
+쮒
+쮓
+쮔
+쮕
+쮖
+쮗
+쮘
+쮙
+쮚
+쮛
+쮜
+쮝
+쮞
+쮟
+쮠
+쮡
+쮢
+쮣
+쮤
+쮥
+쮦
+쮧
+쮨
+쮩
+쮪
+쮫
+쮬
+쮭
+쮮
+쮯
+쮰
+쮱
+쮲
+쮳
+쮴
+쮵
+쮶
+쮷
+쮸
+쮹
+쮺
+쮻
+쮼
+쮽
+쮾
+쮿
+쯀
+쯁
+쯂
+쯃
+쯄
+쯅
+쯆
+쯇
+쯈
+쯉
+쯊
+쯋
+쯌
+쯍
+쯎
+쯏
+쯐
+쯑
+쯒
+쯓
+쯔
+쯕
+쯖
+쯗
+쯘
+쯙
+쯚
+쯛
+쯜
+쯝
+쯞
+쯟
+쯠
+쯡
+쯢
+쯣
+쯤
+쯥
+쯦
+쯧
+쯨
+쯩
+쯪
+쯫
+쯬
+쯭
+쯮
+쯯
+쯰
+쯱
+쯲
+쯳
+쯴
+쯵
+쯶
+쯷
+쯸
+쯹
+쯺
+쯻
+쯼
+쯽
+쯾
+쯿
+찀
+찁
+찂
+찃
+찄
+찅
+찆
+찇
+찈
+찉
+찊
+찋
+찌
+찍
+찎
+찏
+찐
+찑
+찒
+찓
+찔
+찕
+찖
+찗
+찘
+찙
+찚
+찛
+찜
+찝
+찞
+찟
+찠
+찡
+찢
+찣
+찤
+찥
+찦
+찧
+차
+착
+찪
+찫
+찬
+찭
+찮
+찯
+찰
+찱
+찲
+찳
+찴
+찵
+찶
+찷
+참
+찹
+찺
+찻
+찼
+창
+찾
+찿
+챀
+챁
+챂
+챃
+채
+책
+챆
+챇
+챈
+챉
+챊
+챋
+챌
+챍
+챎
+챏
+챐
+챑
+챒
+챓
+챔
+챕
+챖
+챗
+챘
+챙
+챚
+챛
+챜
+챝
+챞
+챟
+챠
+챡
+챢
+챣
+챤
+챥
+챦
+챧
+챨
+챩
+챪
+챫
+챬
+챭
+챮
+챯
+챰
+챱
+챲
+챳
+챴
+챵
+챶
+챷
+챸
+챹
+챺
+챻
+챼
+챽
+챾
+챿
+첀
+첁
+첂
+첃
+첄
+첅
+첆
+첇
+첈
+첉
+첊
+첋
+첌
+첍
+첎
+첏
+첐
+첑
+첒
+첓
+첔
+첕
+첖
+첗
+처
+척
+첚
+첛
+천
+첝
+첞
+첟
+철
+첡
+첢
+첣
+첤
+첥
+첦
+첧
+첨
+첩
+첪
+첫
+첬
+청
+첮
+첯
+첰
+첱
+첲
+첳
+체
+첵
+첶
+첷
+첸
+첹
+첺
+첻
+첼
+첽
+첾
+첿
+쳀
+쳁
+쳂
+쳃
+쳄
+쳅
+쳆
+쳇
+쳈
+쳉
+쳊
+쳋
+쳌
+쳍
+쳎
+쳏
+쳐
+쳑
+쳒
+쳓
+쳔
+쳕
+쳖
+쳗
+쳘
+쳙
+쳚
+쳛
+쳜
+쳝
+쳞
+쳟
+쳠
+쳡
+쳢
+쳣
+쳤
+쳥
+쳦
+쳧
+쳨
+쳩
+쳪
+쳫
+쳬
+쳭
+쳮
+쳯
+쳰
+쳱
+쳲
+쳳
+쳴
+쳵
+쳶
+쳷
+쳸
+쳹
+쳺
+쳻
+쳼
+쳽
+쳾
+쳿
+촀
+촁
+촂
+촃
+촄
+촅
+촆
+촇
+초
+촉
+촊
+촋
+촌
+촍
+촎
+촏
+촐
+촑
+촒
+촓
+촔
+촕
+촖
+촗
+촘
+촙
+촚
+촛
+촜
+총
+촞
+촟
+촠
+촡
+촢
+촣
+촤
+촥
+촦
+촧
+촨
+촩
+촪
+촫
+촬
+촭
+촮
+촯
+촰
+촱
+촲
+촳
+촴
+촵
+촶
+촷
+촸
+촹
+촺
+촻
+촼
+촽
+촾
+촿
+쵀
+쵁
+쵂
+쵃
+쵄
+쵅
+쵆
+쵇
+쵈
+쵉
+쵊
+쵋
+쵌
+쵍
+쵎
+쵏
+쵐
+쵑
+쵒
+쵓
+쵔
+쵕
+쵖
+쵗
+쵘
+쵙
+쵚
+쵛
+최
+쵝
+쵞
+쵟
+쵠
+쵡
+쵢
+쵣
+쵤
+쵥
+쵦
+쵧
+쵨
+쵩
+쵪
+쵫
+쵬
+쵭
+쵮
+쵯
+쵰
+쵱
+쵲
+쵳
+쵴
+쵵
+쵶
+쵷
+쵸
+쵹
+쵺
+쵻
+쵼
+쵽
+쵾
+쵿
+춀
+춁
+춂
+춃
+춄
+춅
+춆
+춇
+춈
+춉
+춊
+춋
+춌
+춍
+춎
+춏
+춐
+춑
+춒
+춓
+추
+축
+춖
+춗
+춘
+춙
+춚
+춛
+출
+춝
+춞
+춟
+춠
+춡
+춢
+춣
+춤
+춥
+춦
+춧
+춨
+충
+춪
+춫
+춬
+춭
+춮
+춯
+춰
+춱
+춲
+춳
+춴
+춵
+춶
+춷
+춸
+춹
+춺
+춻
+춼
+춽
+춾
+춿
+췀
+췁
+췂
+췃
+췄
+췅
+췆
+췇
+췈
+췉
+췊
+췋
+췌
+췍
+췎
+췏
+췐
+췑
+췒
+췓
+췔
+췕
+췖
+췗
+췘
+췙
+췚
+췛
+췜
+췝
+췞
+췟
+췠
+췡
+췢
+췣
+췤
+췥
+췦
+췧
+취
+췩
+췪
+췫
+췬
+췭
+췮
+췯
+췰
+췱
+췲
+췳
+췴
+췵
+췶
+췷
+췸
+췹
+췺
+췻
+췼
+췽
+췾
+췿
+츀
+츁
+츂
+츃
+츄
+츅
+츆
+츇
+츈
+츉
+츊
+츋
+츌
+츍
+츎
+츏
+츐
+츑
+츒
+츓
+츔
+츕
+츖
+츗
+츘
+츙
+츚
+츛
+츜
+츝
+츞
+츟
+츠
+측
+츢
+츣
+츤
+츥
+츦
+츧
+츨
+츩
+츪
+츫
+츬
+츭
+츮
+츯
+츰
+츱
+츲
+츳
+츴
+층
+츶
+츷
+츸
+츹
+츺
+츻
+츼
+츽
+츾
+츿
+칀
+칁
+칂
+칃
+칄
+칅
+칆
+칇
+칈
+칉
+칊
+칋
+칌
+칍
+칎
+칏
+칐
+칑
+칒
+칓
+칔
+칕
+칖
+칗
+치
+칙
+칚
+칛
+친
+칝
+칞
+칟
+칠
+칡
+칢
+칣
+칤
+칥
+칦
+칧
+침
+칩
+칪
+칫
+칬
+칭
+칮
+칯
+칰
+칱
+칲
+칳
+카
+칵
+칶
+칷
+칸
+칹
+칺
+칻
+칼
+칽
+칾
+칿
+캀
+캁
+캂
+캃
+캄
+캅
+캆
+캇
+캈
+캉
+캊
+캋
+캌
+캍
+캎
+캏
+캐
+캑
+캒
+캓
+캔
+캕
+캖
+캗
+캘
+캙
+캚
+캛
+캜
+캝
+캞
+캟
+캠
+캡
+캢
+캣
+캤
+캥
+캦
+캧
+캨
+캩
+캪
+캫
+캬
+캭
+캮
+캯
+캰
+캱
+캲
+캳
+캴
+캵
+캶
+캷
+캸
+캹
+캺
+캻
+캼
+캽
+캾
+캿
+컀
+컁
+컂
+컃
+컄
+컅
+컆
+컇
+컈
+컉
+컊
+컋
+컌
+컍
+컎
+컏
+컐
+컑
+컒
+컓
+컔
+컕
+컖
+컗
+컘
+컙
+컚
+컛
+컜
+컝
+컞
+컟
+컠
+컡
+컢
+컣
+커
+컥
+컦
+컧
+컨
+컩
+컪
+컫
+컬
+컭
+컮
+컯
+컰
+컱
+컲
+컳
+컴
+컵
+컶
+컷
+컸
+컹
+컺
+컻
+컼
+컽
+컾
+컿
+케
+켁
+켂
+켃
+켄
+켅
+켆
+켇
+켈
+켉
+켊
+켋
+켌
+켍
+켎
+켏
+켐
+켑
+켒
+켓
+켔
+켕
+켖
+켗
+켘
+켙
+켚
+켛
+켜
+켝
+켞
+켟
+켠
+켡
+켢
+켣
+켤
+켥
+켦
+켧
+켨
+켩
+켪
+켫
+켬
+켭
+켮
+켯
+켰
+켱
+켲
+켳
+켴
+켵
+켶
+켷
+켸
+켹
+켺
+켻
+켼
+켽
+켾
+켿
+콀
+콁
+콂
+콃
+콄
+콅
+콆
+콇
+콈
+콉
+콊
+콋
+콌
+콍
+콎
+콏
+콐
+콑
+콒
+콓
+코
+콕
+콖
+콗
+콘
+콙
+콚
+콛
+콜
+콝
+콞
+콟
+콠
+콡
+콢
+콣
+콤
+콥
+콦
+콧
+콨
+콩
+콪
+콫
+콬
+콭
+콮
+콯
+콰
+콱
+콲
+콳
+콴
+콵
+콶
+콷
+콸
+콹
+콺
+콻
+콼
+콽
+콾
+콿
+쾀
+쾁
+쾂
+쾃
+쾄
+쾅
+쾆
+쾇
+쾈
+쾉
+쾊
+쾋
+쾌
+쾍
+쾎
+쾏
+쾐
+쾑
+쾒
+쾓
+쾔
+쾕
+쾖
+쾗
+쾘
+쾙
+쾚
+쾛
+쾜
+쾝
+쾞
+쾟
+쾠
+쾡
+쾢
+쾣
+쾤
+쾥
+쾦
+쾧
+쾨
+쾩
+쾪
+쾫
+쾬
+쾭
+쾮
+쾯
+쾰
+쾱
+쾲
+쾳
+쾴
+쾵
+쾶
+쾷
+쾸
+쾹
+쾺
+쾻
+쾼
+쾽
+쾾
+쾿
+쿀
+쿁
+쿂
+쿃
+쿄
+쿅
+쿆
+쿇
+쿈
+쿉
+쿊
+쿋
+쿌
+쿍
+쿎
+쿏
+쿐
+쿑
+쿒
+쿓
+쿔
+쿕
+쿖
+쿗
+쿘
+쿙
+쿚
+쿛
+쿜
+쿝
+쿞
+쿟
+쿠
+쿡
+쿢
+쿣
+쿤
+쿥
+쿦
+쿧
+쿨
+쿩
+쿪
+쿫
+쿬
+쿭
+쿮
+쿯
+쿰
+쿱
+쿲
+쿳
+쿴
+쿵
+쿶
+쿷
+쿸
+쿹
+쿺
+쿻
+쿼
+쿽
+쿾
+쿿
+퀀
+퀁
+퀂
+퀃
+퀄
+퀅
+퀆
+퀇
+퀈
+퀉
+퀊
+퀋
+퀌
+퀍
+퀎
+퀏
+퀐
+퀑
+퀒
+퀓
+퀔
+퀕
+퀖
+퀗
+퀘
+퀙
+퀚
+퀛
+퀜
+퀝
+퀞
+퀟
+퀠
+퀡
+퀢
+퀣
+퀤
+퀥
+퀦
+퀧
+퀨
+퀩
+퀪
+퀫
+퀬
+퀭
+퀮
+퀯
+퀰
+퀱
+퀲
+퀳
+퀴
+퀵
+퀶
+퀷
+퀸
+퀹
+퀺
+퀻
+퀼
+퀽
+퀾
+퀿
+큀
+큁
+큂
+큃
+큄
+큅
+큆
+큇
+큈
+큉
+큊
+큋
+큌
+큍
+큎
+큏
+큐
+큑
+큒
+큓
+큔
+큕
+큖
+큗
+큘
+큙
+큚
+큛
+큜
+큝
+큞
+큟
+큠
+큡
+큢
+큣
+큤
+큥
+큦
+큧
+큨
+큩
+큪
+큫
+크
+큭
+큮
+큯
+큰
+큱
+큲
+큳
+클
+큵
+큶
+큷
+큸
+큹
+큺
+큻
+큼
+큽
+큾
+큿
+킀
+킁
+킂
+킃
+킄
+킅
+킆
+킇
+킈
+킉
+킊
+킋
+킌
+킍
+킎
+킏
+킐
+킑
+킒
+킓
+킔
+킕
+킖
+킗
+킘
+킙
+킚
+킛
+킜
+킝
+킞
+킟
+킠
+킡
+킢
+킣
+키
+킥
+킦
+킧
+킨
+킩
+킪
+킫
+킬
+킭
+킮
+킯
+킰
+킱
+킲
+킳
+킴
+킵
+킶
+킷
+킸
+킹
+킺
+킻
+킼
+킽
+킾
+킿
+타
+탁
+탂
+탃
+탄
+탅
+탆
+탇
+탈
+탉
+탊
+탋
+탌
+탍
+탎
+탏
+탐
+탑
+탒
+탓
+탔
+탕
+탖
+탗
+탘
+탙
+탚
+탛
+태
+택
+탞
+탟
+탠
+탡
+탢
+탣
+탤
+탥
+탦
+탧
+탨
+탩
+탪
+탫
+탬
+탭
+탮
+탯
+탰
+탱
+탲
+탳
+탴
+탵
+탶
+탷
+탸
+탹
+탺
+탻
+탼
+탽
+탾
+탿
+턀
+턁
+턂
+턃
+턄
+턅
+턆
+턇
+턈
+턉
+턊
+턋
+턌
+턍
+턎
+턏
+턐
+턑
+턒
+턓
+턔
+턕
+턖
+턗
+턘
+턙
+턚
+턛
+턜
+턝
+턞
+턟
+턠
+턡
+턢
+턣
+턤
+턥
+턦
+턧
+턨
+턩
+턪
+턫
+턬
+턭
+턮
+턯
+터
+턱
+턲
+턳
+턴
+턵
+턶
+턷
+털
+턹
+턺
+턻
+턼
+턽
+턾
+턿
+텀
+텁
+텂
+텃
+텄
+텅
+텆
+텇
+텈
+텉
+텊
+텋
+테
+텍
+텎
+텏
+텐
+텑
+텒
+텓
+텔
+텕
+텖
+텗
+텘
+텙
+텚
+텛
+템
+텝
+텞
+텟
+텠
+텡
+텢
+텣
+텤
+텥
+텦
+텧
+텨
+텩
+텪
+텫
+텬
+텭
+텮
+텯
+텰
+텱
+텲
+텳
+텴
+텵
+텶
+텷
+텸
+텹
+텺
+텻
+텼
+텽
+텾
+텿
+톀
+톁
+톂
+톃
+톄
+톅
+톆
+톇
+톈
+톉
+톊
+톋
+톌
+톍
+톎
+톏
+톐
+톑
+톒
+톓
+톔
+톕
+톖
+톗
+톘
+톙
+톚
+톛
+톜
+톝
+톞
+톟
+토
+톡
+톢
+톣
+톤
+톥
+톦
+톧
+톨
+톩
+톪
+톫
+톬
+톭
+톮
+톯
+톰
+톱
+톲
+톳
+톴
+통
+톶
+톷
+톸
+톹
+톺
+톻
+톼
+톽
+톾
+톿
+퇀
+퇁
+퇂
+퇃
+퇄
+퇅
+퇆
+퇇
+퇈
+퇉
+퇊
+퇋
+퇌
+퇍
+퇎
+퇏
+퇐
+퇑
+퇒
+퇓
+퇔
+퇕
+퇖
+퇗
+퇘
+퇙
+퇚
+퇛
+퇜
+퇝
+퇞
+퇟
+퇠
+퇡
+퇢
+퇣
+퇤
+퇥
+퇦
+퇧
+퇨
+퇩
+퇪
+퇫
+퇬
+퇭
+퇮
+퇯
+퇰
+퇱
+퇲
+퇳
+퇴
+퇵
+퇶
+퇷
+퇸
+퇹
+퇺
+퇻
+퇼
+퇽
+퇾
+퇿
+툀
+툁
+툂
+툃
+툄
+툅
+툆
+툇
+툈
+툉
+툊
+툋
+툌
+툍
+툎
+툏
+툐
+툑
+툒
+툓
+툔
+툕
+툖
+툗
+툘
+툙
+툚
+툛
+툜
+툝
+툞
+툟
+툠
+툡
+툢
+툣
+툤
+툥
+툦
+툧
+툨
+툩
+툪
+툫
+투
+툭
+툮
+툯
+툰
+툱
+툲
+툳
+툴
+툵
+툶
+툷
+툸
+툹
+툺
+툻
+툼
+툽
+툾
+툿
+퉀
+퉁
+퉂
+퉃
+퉄
+퉅
+퉆
+퉇
+퉈
+퉉
+퉊
+퉋
+퉌
+퉍
+퉎
+퉏
+퉐
+퉑
+퉒
+퉓
+퉔
+퉕
+퉖
+퉗
+퉘
+퉙
+퉚
+퉛
+퉜
+퉝
+퉞
+퉟
+퉠
+퉡
+퉢
+퉣
+퉤
+퉥
+퉦
+퉧
+퉨
+퉩
+퉪
+퉫
+퉬
+퉭
+퉮
+퉯
+퉰
+퉱
+퉲
+퉳
+퉴
+퉵
+퉶
+퉷
+퉸
+퉹
+퉺
+퉻
+퉼
+퉽
+퉾
+퉿
+튀
+튁
+튂
+튃
+튄
+튅
+튆
+튇
+튈
+튉
+튊
+튋
+튌
+튍
+튎
+튏
+튐
+튑
+튒
+튓
+튔
+튕
+튖
+튗
+튘
+튙
+튚
+튛
+튜
+튝
+튞
+튟
+튠
+튡
+튢
+튣
+튤
+튥
+튦
+튧
+튨
+튩
+튪
+튫
+튬
+튭
+튮
+튯
+튰
+튱
+튲
+튳
+튴
+튵
+튶
+튷
+트
+특
+튺
+튻
+튼
+튽
+튾
+튿
+틀
+틁
+틂
+틃
+틄
+틅
+틆
+틇
+틈
+틉
+틊
+틋
+틌
+틍
+틎
+틏
+틐
+틑
+틒
+틓
+틔
+틕
+틖
+틗
+틘
+틙
+틚
+틛
+틜
+틝
+틞
+틟
+틠
+틡
+틢
+틣
+틤
+틥
+틦
+틧
+틨
+틩
+틪
+틫
+틬
+틭
+틮
+틯
+티
+틱
+틲
+틳
+틴
+틵
+틶
+틷
+틸
+틹
+틺
+틻
+틼
+틽
+틾
+틿
+팀
+팁
+팂
+팃
+팄
+팅
+팆
+팇
+팈
+팉
+팊
+팋
+파
+팍
+팎
+팏
+판
+팑
+팒
+팓
+팔
+팕
+팖
+팗
+팘
+팙
+팚
+팛
+팜
+팝
+팞
+팟
+팠
+팡
+팢
+팣
+팤
+팥
+팦
+팧
+패
+팩
+팪
+팫
+팬
+팭
+팮
+팯
+팰
+팱
+팲
+팳
+팴
+팵
+팶
+팷
+팸
+팹
+팺
+팻
+팼
+팽
+팾
+팿
+퍀
+퍁
+퍂
+퍃
+퍄
+퍅
+퍆
+퍇
+퍈
+퍉
+퍊
+퍋
+퍌
+퍍
+퍎
+퍏
+퍐
+퍑
+퍒
+퍓
+퍔
+퍕
+퍖
+퍗
+퍘
+퍙
+퍚
+퍛
+퍜
+퍝
+퍞
+퍟
+퍠
+퍡
+퍢
+퍣
+퍤
+퍥
+퍦
+퍧
+퍨
+퍩
+퍪
+퍫
+퍬
+퍭
+퍮
+퍯
+퍰
+퍱
+퍲
+퍳
+퍴
+퍵
+퍶
+퍷
+퍸
+퍹
+퍺
+퍻
+퍼
+퍽
+퍾
+퍿
+펀
+펁
+펂
+펃
+펄
+펅
+펆
+펇
+펈
+펉
+펊
+펋
+펌
+펍
+펎
+펏
+펐
+펑
+펒
+펓
+펔
+펕
+펖
+펗
+페
+펙
+펚
+펛
+펜
+펝
+펞
+펟
+펠
+펡
+펢
+펣
+펤
+펥
+펦
+펧
+펨
+펩
+펪
+펫
+펬
+펭
+펮
+펯
+펰
+펱
+펲
+펳
+펴
+펵
+펶
+펷
+편
+펹
+펺
+펻
+펼
+펽
+펾
+펿
+폀
+폁
+폂
+폃
+폄
+폅
+폆
+폇
+폈
+평
+폊
+폋
+폌
+폍
+폎
+폏
+폐
+폑
+폒
+폓
+폔
+폕
+폖
+폗
+폘
+폙
+폚
+폛
+폜
+폝
+폞
+폟
+폠
+폡
+폢
+폣
+폤
+폥
+폦
+폧
+폨
+폩
+폪
+폫
+포
+폭
+폮
+폯
+폰
+폱
+폲
+폳
+폴
+폵
+폶
+폷
+폸
+폹
+폺
+폻
+폼
+폽
+폾
+폿
+퐀
+퐁
+퐂
+퐃
+퐄
+퐅
+퐆
+퐇
+퐈
+퐉
+퐊
+퐋
+퐌
+퐍
+퐎
+퐏
+퐐
+퐑
+퐒
+퐓
+퐔
+퐕
+퐖
+퐗
+퐘
+퐙
+퐚
+퐛
+퐜
+퐝
+퐞
+퐟
+퐠
+퐡
+퐢
+퐣
+퐤
+퐥
+퐦
+퐧
+퐨
+퐩
+퐪
+퐫
+퐬
+퐭
+퐮
+퐯
+퐰
+퐱
+퐲
+퐳
+퐴
+퐵
+퐶
+퐷
+퐸
+퐹
+퐺
+퐻
+퐼
+퐽
+퐾
+퐿
+푀
+푁
+푂
+푃
+푄
+푅
+푆
+푇
+푈
+푉
+푊
+푋
+푌
+푍
+푎
+푏
+푐
+푑
+푒
+푓
+푔
+푕
+푖
+푗
+푘
+푙
+푚
+푛
+표
+푝
+푞
+푟
+푠
+푡
+푢
+푣
+푤
+푥
+푦
+푧
+푨
+푩
+푪
+푫
+푬
+푭
+푮
+푯
+푰
+푱
+푲
+푳
+푴
+푵
+푶
+푷
+푸
+푹
+푺
+푻
+푼
+푽
+푾
+푿
+풀
+풁
+풂
+풃
+풄
+풅
+풆
+풇
+품
+풉
+풊
+풋
+풌
+풍
+풎
+풏
+풐
+풑
+풒
+풓
+풔
+풕
+풖
+풗
+풘
+풙
+풚
+풛
+풜
+풝
+풞
+풟
+풠
+풡
+풢
+풣
+풤
+풥
+풦
+풧
+풨
+풩
+풪
+풫
+풬
+풭
+풮
+풯
+풰
+풱
+풲
+풳
+풴
+풵
+풶
+풷
+풸
+풹
+풺
+풻
+풼
+풽
+풾
+풿
+퓀
+퓁
+퓂
+퓃
+퓄
+퓅
+퓆
+퓇
+퓈
+퓉
+퓊
+퓋
+퓌
+퓍
+퓎
+퓏
+퓐
+퓑
+퓒
+퓓
+퓔
+퓕
+퓖
+퓗
+퓘
+퓙
+퓚
+퓛
+퓜
+퓝
+퓞
+퓟
+퓠
+퓡
+퓢
+퓣
+퓤
+퓥
+퓦
+퓧
+퓨
+퓩
+퓪
+퓫
+퓬
+퓭
+퓮
+퓯
+퓰
+퓱
+퓲
+퓳
+퓴
+퓵
+퓶
+퓷
+퓸
+퓹
+퓺
+퓻
+퓼
+퓽
+퓾
+퓿
+픀
+픁
+픂
+픃
+프
+픅
+픆
+픇
+픈
+픉
+픊
+픋
+플
+픍
+픎
+픏
+픐
+픑
+픒
+픓
+픔
+픕
+픖
+픗
+픘
+픙
+픚
+픛
+픜
+픝
+픞
+픟
+픠
+픡
+픢
+픣
+픤
+픥
+픦
+픧
+픨
+픩
+픪
+픫
+픬
+픭
+픮
+픯
+픰
+픱
+픲
+픳
+픴
+픵
+픶
+픷
+픸
+픹
+픺
+픻
+피
+픽
+픾
+픿
+핀
+핁
+핂
+핃
+필
+핅
+핆
+핇
+핈
+핉
+핊
+핋
+핌
+핍
+핎
+핏
+핐
+핑
+핒
+핓
+핔
+핕
+핖
+핗
+하
+학
+핚
+핛
+한
+핝
+핞
+핟
+할
+핡
+핢
+핣
+핤
+핥
+핦
+핧
+함
+합
+핪
+핫
+핬
+항
+핮
+핯
+핰
+핱
+핲
+핳
+해
+핵
+핶
+핷
+핸
+핹
+핺
+핻
+핼
+핽
+핾
+핿
+햀
+햁
+햂
+햃
+햄
+햅
+햆
+햇
+했
+행
+햊
+햋
+햌
+햍
+햎
+햏
+햐
+햑
+햒
+햓
+햔
+햕
+햖
+햗
+햘
+햙
+햚
+햛
+햜
+햝
+햞
+햟
+햠
+햡
+햢
+햣
+햤
+향
+햦
+햧
+햨
+햩
+햪
+햫
+햬
+햭
+햮
+햯
+햰
+햱
+햲
+햳
+햴
+햵
+햶
+햷
+햸
+햹
+햺
+햻
+햼
+햽
+햾
+햿
+헀
+헁
+헂
+헃
+헄
+헅
+헆
+헇
+허
+헉
+헊
+헋
+헌
+헍
+헎
+헏
+헐
+헑
+헒
+헓
+헔
+헕
+헖
+헗
+험
+헙
+헚
+헛
+헜
+헝
+헞
+헟
+헠
+헡
+헢
+헣
+헤
+헥
+헦
+헧
+헨
+헩
+헪
+헫
+헬
+헭
+헮
+헯
+헰
+헱
+헲
+헳
+헴
+헵
+헶
+헷
+헸
+헹
+헺
+헻
+헼
+헽
+헾
+헿
+혀
+혁
+혂
+혃
+현
+혅
+혆
+혇
+혈
+혉
+혊
+혋
+혌
+혍
+혎
+혏
+혐
+협
+혒
+혓
+혔
+형
+혖
+혗
+혘
+혙
+혚
+혛
+혜
+혝
+혞
+혟
+혠
+혡
+혢
+혣
+혤
+혥
+혦
+혧
+혨
+혩
+혪
+혫
+혬
+혭
+혮
+혯
+혰
+혱
+혲
+혳
+혴
+혵
+혶
+혷
+호
+혹
+혺
+혻
+혼
+혽
+혾
+혿
+홀
+홁
+홂
+홃
+홄
+홅
+홆
+홇
+홈
+홉
+홊
+홋
+홌
+홍
+홎
+홏
+홐
+홑
+홒
+홓
+화
+확
+홖
+홗
+환
+홙
+홚
+홛
+활
+홝
+홞
+홟
+홠
+홡
+홢
+홣
+홤
+홥
+홦
+홧
+홨
+황
+홪
+홫
+홬
+홭
+홮
+홯
+홰
+홱
+홲
+홳
+홴
+홵
+홶
+홷
+홸
+홹
+홺
+홻
+홼
+홽
+홾
+홿
+횀
+횁
+횂
+횃
+횄
+횅
+횆
+횇
+횈
+횉
+횊
+횋
+회
+획
+횎
+횏
+횐
+횑
+횒
+횓
+횔
+횕
+횖
+횗
+횘
+횙
+횚
+횛
+횜
+횝
+횞
+횟
+횠
+횡
+횢
+횣
+횤
+횥
+횦
+횧
+효
+횩
+횪
+횫
+횬
+횭
+횮
+횯
+횰
+횱
+횲
+횳
+횴
+횵
+횶
+횷
+횸
+횹
+횺
+횻
+횼
+횽
+횾
+횿
+훀
+훁
+훂
+훃
+후
+훅
+훆
+훇
+훈
+훉
+훊
+훋
+훌
+훍
+훎
+훏
+훐
+훑
+훒
+훓
+훔
+훕
+훖
+훗
+훘
+훙
+훚
+훛
+훜
+훝
+훞
+훟
+훠
+훡
+훢
+훣
+훤
+훥
+훦
+훧
+훨
+훩
+훪
+훫
+훬
+훭
+훮
+훯
+훰
+훱
+훲
+훳
+훴
+훵
+훶
+훷
+훸
+훹
+훺
+훻
+훼
+훽
+훾
+훿
+휀
+휁
+휂
+휃
+휄
+휅
+휆
+휇
+휈
+휉
+휊
+휋
+휌
+휍
+휎
+휏
+휐
+휑
+휒
+휓
+휔
+휕
+휖
+휗
+휘
+휙
+휚
+휛
+휜
+휝
+휞
+휟
+휠
+휡
+휢
+휣
+휤
+휥
+휦
+휧
+휨
+휩
+휪
+휫
+휬
+휭
+휮
+휯
+휰
+휱
+휲
+휳
+휴
+휵
+휶
+휷
+휸
+휹
+휺
+휻
+휼
+휽
+휾
+휿
+흀
+흁
+흂
+흃
+흄
+흅
+흆
+흇
+흈
+흉
+흊
+흋
+흌
+흍
+흎
+흏
+흐
+흑
+흒
+흓
+흔
+흕
+흖
+흗
+흘
+흙
+흚
+흛
+흜
+흝
+흞
+흟
+흠
+흡
+흢
+흣
+흤
+흥
+흦
+흧
+흨
+흩
+흪
+흫
+희
+흭
+흮
+흯
+흰
+흱
+흲
+흳
+흴
+흵
+흶
+흷
+흸
+흹
+흺
+흻
+흼
+흽
+흾
+흿
+힀
+힁
+힂
+힃
+힄
+힅
+힆
+힇
+히
+힉
+힊
+힋
+힌
+힍
+힎
+힏
+힐
+힑
+힒
+힓
+힔
+힕
+힖
+힗
+힘
+힙
+힚
+힛
+힜
+힝
+힞
+힟
+힠
+힡
+힢
+힣
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+<
+=
+>
+?
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+[
+]
+_
+`
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+©
+‥
+{
+}
+\
+|
+@
+^
+~
+÷
+∕
+∙
+⋅
+·
+±
+∓
+∩
+∪
+□
+←
+↔
+⇒
+⇐
+⇔
+∀
+∃
+∄
+∴
+∵
+∝
+∞
+⊥
+∟
+∠
+∡
+∢
+′
+″
+∥
+⊾
+⊿
+∂
+∫
+∬
+∭
+∮
+∯
+∰
+∑
+∏
+√
+∛
+∜
+∱
+∲
+∳
+∶
+∷
+∼
+®
+℉
+Ω
+℧
+Å
+⌀
+ℏ
+⅀
+⍺
+⍵
+¢
+€
+£
+¥
+₿
+Ⅰ
+Ⅱ
+Ⅲ
+Ⅳ
+Ⅴ
+Ⅵ
+Ⅶ
+Ⅷ
+Ⅸ
+Ⅹ
+Ⅺ
+Ⅻ
+ⅰ
+ⅱ
+ⅲ
+ⅳ
+ⅴ
+ⅵ
+ⅶ
+ⅷ
+ⅸ
+ⅹ
+ⅺ
+ⅻ
+➀
+➁
+➂
+➃
+➄
+➅
+➆
+➇
+➈
+➉
+➊
+➋
+➌
+➍
+➎
+➏
+➐
+➑
+➒
+➓
+❶
+❷
+❸
+❹
+❺
+❻
+❼
+❽
+❾
+❿
+①
+②
+③
+④
+⑤
+⑥
+⑦
+⑧
+⑨
+⑩
+●
+▶
+𝑢
+︽
+–
+﹥
+𝜓
+•
+∋
+ƒ
+०
+⬆
+Ạ
+◀
+
+▫
+︾
+À
+Á
+Â
+Ã
+Ä
+Å
+Æ
+Ç
+È
+É
+Ê
+Ë
+Ì
+Í
+Î
+Ï
+Ð
+Ñ
+Ò
+Ó
+Ô
+Õ
+Ö
+Ø
+Ù
+Ú
+Û
+Ü
+Ý
+Þ
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+ì
+í
+î
+ï
+ð
+ñ
+ò
+ó
+ô
+õ
+ö
+ø
+ù
+ú
+û
+ü
+ý
+þ
+ÿ
+¡
+¤
+¦
+§
+¨
+ª
+«
+¬
+¯
+°
+²
+³
+´
+µ
+¶
+¸
+¹
+º
+»
+¼
+½
+¾
+¿
+×
+‐
+‑
+‒
+—
+―
+‖
+‗
+‘
+’
+‚
+‛
+“
+”
+„
+‟
+†
+‡
+‣
+․
+…
+‧
+‰
+‴
+‵
+‶
+‷
+‸
+‹
+›
+※
+‼
+‽
+‾
+₤
+₡
+₹
+−
+∖
+∗
+≈
+≠
+≡
+≤
+≥
+⊂
+⊃
+↑
+→
+↓
+↕
+™
+Ω
+℮
+∆
+✓
+✗
+✘
+▪
+◼
+✔
+✕
+☑
+☒
+№
+₽
+₴
+Α
+α
+Β
+β
+Γ
+γ
+Δ
+δ
+Ε
+ε
+Ζ
+ζ
+Η
+η
+Θ
+θ
+Ι
+ι
+Κ
+κ
+Λ
+λ
+Μ
+μ
+Ν
+ν
+Ξ
+ξ
+Ο
+ο
+Π
+π
+Ρ
+ρ
+Σ
+σ
+ς
+Τ
+τ
+Υ
+υ
+Φ
+φ
+Χ
+χ
+Ψ
+ψ
+ω
diff --git a/ppocr/utils/dict/ppocrv5_latin_dict.txt b/ppocr/utils/dict/ppocrv5_latin_dict.txt
new file mode 100644
index 0000000..c618257
--- /dev/null
+++ b/ppocr/utils/dict/ppocrv5_latin_dict.txt
@@ -0,0 +1,836 @@
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+À
+Á
+Â
+Ã
+Ä
+Å
+Æ
+Ç
+È
+É
+Ê
+Ë
+Ì
+Í
+Î
+Ï
+Ð
+Ñ
+Ò
+Ó
+Ô
+Õ
+Ö
+×
+Ø
+Ù
+Ú
+Û
+Ü
+Ý
+Þ
+ß
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+ì
+í
+î
+ï
+ð
+ñ
+ò
+ó
+ô
+õ
+ö
+÷
+ø
+ù
+ú
+û
+ü
+ý
+þ
+ÿ
+Ā
+ā
+Ă
+ă
+Ą
+ą
+Ć
+ć
+Ĉ
+ĉ
+Ċ
+ċ
+Č
+č
+Ď
+ď
+Đ
+đ
+Ē
+ē
+Ĕ
+ĕ
+Ė
+ė
+Ę
+ę
+Ě
+ě
+Ĝ
+ĝ
+Ğ
+ğ
+Ġ
+ġ
+Ģ
+ģ
+Ĥ
+ĥ
+Ħ
+ħ
+Ĩ
+ĩ
+Ī
+ī
+Ĭ
+ĭ
+Į
+į
+İ
+ı
+IJ
+ij
+Ĵ
+ĵ
+Ķ
+ķ
+ĸ
+Ĺ
+ĺ
+Ļ
+ļ
+Ľ
+ľ
+Ŀ
+ŀ
+Ł
+ł
+Ń
+ń
+Ņ
+ņ
+Ň
+ň
+ʼn
+Ŋ
+ŋ
+Ō
+ō
+Ŏ
+ŏ
+Ő
+ő
+Œ
+œ
+Ŕ
+ŕ
+Ŗ
+ŗ
+Ř
+ř
+Ś
+ś
+Ŝ
+ŝ
+Ş
+ş
+Š
+š
+Ţ
+ţ
+Ť
+ť
+Ŧ
+ŧ
+Ũ
+ũ
+Ū
+ū
+Ŭ
+ŭ
+Ů
+ů
+Ű
+ű
+Ų
+ų
+Ŵ
+ŵ
+Ŷ
+ŷ
+Ÿ
+Ź
+ź
+Ż
+ż
+Ž
+ž
+ſ
+ƀ
+Ɓ
+Ƃ
+ƃ
+Ƅ
+ƅ
+Ɔ
+Ƈ
+ƈ
+Ɖ
+Ɗ
+Ƌ
+ƌ
+ƍ
+Ǝ
+Ə
+Ɛ
+Ƒ
+ƒ
+Ɠ
+Ɣ
+ƕ
+Ɩ
+Ɨ
+Ƙ
+ƙ
+ƚ
+ƛ
+Ɯ
+Ɲ
+ƞ
+Ɵ
+Ơ
+ơ
+Ƣ
+ƣ
+Ƥ
+ƥ
+Ʀ
+Ƨ
+ƨ
+Ʃ
+ƪ
+ƫ
+Ƭ
+ƭ
+Ʈ
+Ư
+ư
+Ʊ
+Ʋ
+Ƴ
+ƴ
+Ƶ
+ƶ
+Ʒ
+Ƹ
+ƹ
+ƺ
+ƻ
+Ƽ
+ƽ
+ƾ
+ƿ
+ǀ
+ǁ
+ǂ
+ǃ
+DŽ
+Dž
+dž
+LJ
+Lj
+lj
+NJ
+Nj
+nj
+Ǎ
+ǎ
+Ǐ
+ǐ
+Ǒ
+ǒ
+Ǔ
+ǔ
+Ǖ
+ǖ
+Ǘ
+ǘ
+Ǚ
+ǚ
+Ǜ
+ǜ
+ǝ
+Ǟ
+ǟ
+Ǡ
+ǡ
+Ǣ
+ǣ
+Ǥ
+ǥ
+Ǧ
+ǧ
+Ǩ
+ǩ
+Ǫ
+ǫ
+Ǭ
+ǭ
+Ǯ
+ǯ
+ǰ
+DZ
+Dz
+dz
+Ǵ
+ǵ
+Ƕ
+Ƿ
+Ǹ
+ǹ
+Ǻ
+ǻ
+Ǽ
+ǽ
+Ǿ
+ǿ
+Ȁ
+ȁ
+Ȃ
+ȃ
+Ȅ
+ȅ
+Ȇ
+ȇ
+Ȉ
+ȉ
+Ȋ
+ȋ
+Ȍ
+ȍ
+Ȏ
+ȏ
+Ȑ
+ȑ
+Ȓ
+ȓ
+Ȕ
+ȕ
+Ȗ
+ȗ
+Ș
+ș
+Ț
+ț
+Ȝ
+ȝ
+Ȟ
+ȟ
+Ƞ
+ȡ
+Ȣ
+ȣ
+Ȥ
+ȥ
+Ȧ
+ȧ
+Ȩ
+ȩ
+Ȫ
+ȫ
+Ȭ
+ȭ
+Ȯ
+ȯ
+Ȱ
+ȱ
+Ȳ
+ȳ
+ȴ
+ȵ
+ȶ
+ȷ
+ȸ
+ȹ
+Ⱥ
+Ȼ
+ȼ
+Ƚ
+Ⱦ
+ȿ
+ɀ
+Ɂ
+ɂ
+Ƀ
+Ʉ
+Ʌ
+Ɇ
+ɇ
+Ɉ
+ɉ
+Ɋ
+ɋ
+Ɍ
+ɍ
+Ɏ
+ɏ
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+:
+;
+<
+=
+>
+?
+@
+[
+\
+]
+_
+`
+{
+|
+}
+^
+~
+©
+®
+℉
+№
+Ω
+℮
+™
+∆
+✓
+✔
+✗
+✘
+✕
+☑
+☒
+●
+▪
+▫
+◼
+▶
+◀
+⬆
+¤
+¦
+§
+¨
+ª
+«
+¬
+¯
+°
+²
+³
+´
+µ
+¶
+¸
+¹
+º
+»
+¼
+½
+¾
+¿
+×
+‐
+‑
+‒
+—
+―
+‖
+‗
+‘
+’
+‚
+‛
+“
+”
+„
+‟
+†
+‡
+‣
+․
+…
+‧
+‰
+‴
+‵
+‶
+‷
+‸
+‹
+›
+※
+‼
+‽
+‾
+−
+₤
+₡
+₹
+₽
+₴
+₿
+¢
+€
+£
+¥
+Ⅰ
+Ⅱ
+Ⅲ
+Ⅳ
+Ⅴ
+Ⅵ
+Ⅶ
+Ⅷ
+Ⅸ
+Ⅹ
+Ⅺ
+Ⅻ
+ⅰ
+ⅱ
+ⅲ
+ⅳ
+ⅴ
+ⅵ
+ⅶ
+ⅷ
+ⅸ
+ⅹ
+ⅺ
+ⅻ
+➀
+➁
+➂
+➃
+➄
+➅
+➆
+➇
+➈
+➉
+➊
+➋
+➌
+➍
+➎
+➏
+➐
+➑
+➒
+➓
+❶
+❷
+❸
+❹
+❺
+❻
+❼
+❽
+❾
+❿
+①
+②
+③
+④
+⑤
+⑥
+⑦
+⑧
+⑨
+⑩
+↑
+→
+↓
+↕
+←
+↔
+⇒
+⇐
+⇔
+∀
+∃
+∄
+∴
+∵
+∝
+∞
+∩
+∪
+∂
+∫
+∬
+∭
+∮
+∯
+∰
+∑
+∏
+√
+∛
+∜
+∱
+∲
+∳
+∶
+∷
+∼
+∖
+∗
+≈
+≠
+≡
+≤
+≥
+⊂
+⊃
+⊥
+⊾
+⊿
+□
+∥
+∋
+ƒ
+′
+″
+À
+Á
+Â
+Ã
+Ä
+Å
+Æ
+Ç
+È
+É
+Ê
+Ë
+Ì
+Í
+Î
+Ï
+Ð
+Ñ
+Ò
+Ó
+Ô
+Õ
+Ö
+Ø
+Ù
+Ú
+Û
+Ü
+Ý
+Þ
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+ì
+í
+î
+ï
+ð
+ñ
+ò
+ó
+ô
+õ
+ö
+ø
+ù
+ú
+û
+ü
+ý
+þ
+ÿ
+Α
+Β
+Γ
+Δ
+Ε
+Ζ
+Η
+Θ
+Ι
+Κ
+Λ
+Μ
+Ν
+Ξ
+Ο
+Π
+Ρ
+Σ
+Τ
+Υ
+Φ
+Χ
+Ψ
+Ω
+α
+β
+γ
+δ
+ε
+ζ
+η
+θ
+ι
+κ
+λ
+μ
+ν
+ξ
+ο
+π
+ρ
+σ
+ς
+τ
+υ
+φ
+χ
+ψ
+ω
+Å
+ℏ
+⌀
+⍺
+⍵
+𝑢
+𝜓
+०
+‥
+︽
+﹥
+•
+÷
+∕
+∙
+⋅
+·
+±
+∓
+∟
+∠
+∡
+∢
+℧
+☺
diff --git a/ppocr/utils/dict/ppocrv5_ta_dict.txt b/ppocr/utils/dict/ppocrv5_ta_dict.txt
new file mode 100644
index 0000000..1290f65
--- /dev/null
+++ b/ppocr/utils/dict/ppocrv5_ta_dict.txt
@@ -0,0 +1,513 @@
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+<
+=
+>
+?
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+[
+]
+_
+`
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+©
+‥
+{
+}
+\
+|
+@
+^
+~
+÷
+∕
+∙
+⋅
+·
+±
+∓
+∩
+∪
+□
+←
+↔
+⇒
+⇐
+⇔
+∀
+∃
+∄
+∴
+∵
+∝
+∞
+⊥
+∟
+∠
+∡
+∢
+′
+″
+∥
+⊾
+⊿
+∂
+∫
+∬
+∭
+∮
+∯
+∰
+∑
+∏
+√
+∛
+∜
+∱
+∲
+∳
+∶
+∷
+∼
+®
+℉
+Ω
+℧
+Å
+⌀
+ℏ
+⅀
+⍺
+⍵
+¢
+€
+£
+¥
+₿
+Ⅰ
+Ⅱ
+Ⅲ
+Ⅳ
+Ⅴ
+Ⅵ
+Ⅶ
+Ⅷ
+Ⅸ
+Ⅹ
+Ⅺ
+Ⅻ
+ⅰ
+ⅱ
+ⅲ
+ⅳ
+ⅴ
+ⅵ
+ⅶ
+ⅷ
+ⅸ
+ⅹ
+ⅺ
+ⅻ
+➀
+➁
+➂
+➃
+➄
+➅
+➆
+➇
+➈
+➉
+➊
+➋
+➌
+➍
+➎
+➏
+➐
+➑
+➒
+➓
+❶
+❷
+❸
+❹
+❺
+❻
+❼
+❽
+❾
+❿
+①
+②
+③
+④
+⑤
+⑥
+⑦
+⑧
+⑨
+⑩
+●
+▶
+𝑢
+︽
+–
+﹥
+𝜓
+•
+∋
+ƒ
+०
+⬆
+Ạ
+◀
+
+▫
+︾
+À
+Á
+Â
+Ã
+Ä
+Å
+Æ
+Ç
+È
+É
+Ê
+Ë
+Ì
+Í
+Î
+Ï
+Ð
+Ñ
+Ò
+Ó
+Ô
+Õ
+Ö
+Ø
+Ù
+Ú
+Û
+Ü
+Ý
+Þ
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+ì
+í
+î
+ï
+ð
+ñ
+ò
+ó
+ô
+õ
+ö
+ø
+ù
+ú
+û
+ü
+ý
+þ
+ÿ
+¡
+¤
+¦
+§
+¨
+ª
+«
+¬
+¯
+°
+²
+³
+´
+µ
+¶
+¸
+¹
+º
+»
+¼
+½
+¾
+¿
+×
+‐
+‑
+‒
+—
+―
+‖
+‗
+‘
+’
+‚
+‛
+“
+”
+„
+‟
+†
+‡
+‣
+․
+…
+‧
+‰
+‴
+‵
+‶
+‷
+‸
+‹
+›
+※
+‼
+‽
+‾
+₤
+₡
+₹
+−
+∖
+∗
+≈
+≠
+≡
+≤
+≥
+⊂
+⊃
+↑
+→
+↓
+↕
+™
+Ω
+℮
+∆
+✓
+✗
+✘
+▪
+◼
+✔
+✕
+☑
+☒
+№
+₽
+₴
+Α
+α
+Β
+β
+Γ
+γ
+Δ
+δ
+Ε
+ε
+Ζ
+ζ
+Η
+η
+Θ
+θ
+Ι
+ι
+Κ
+κ
+Λ
+λ
+Μ
+μ
+Ν
+ν
+Ξ
+ξ
+Ο
+ο
+Π
+π
+Ρ
+ρ
+Σ
+σ
+ς
+Τ
+τ
+Υ
+υ
+Φ
+φ
+Χ
+χ
+Ψ
+ψ
+ω
+ஂ
+ஃ
+அ
+ஆ
+இ
+ஈ
+உ
+ஊ
+எ
+ஏ
+ஐ
+ஒ
+ஓ
+ஔ
+க
+ங
+ச
+ஜ
+ஞ
+ட
+ண
+த
+ந
+ன
+ப
+ம
+ய
+ர
+ற
+ல
+ள
+ழ
+வ
+ஶ
+ஷ
+ஸ
+ஹ
+ா
+ி
+ீ
+ு
+ூ
+ெ
+ே
+ை
+ொ
+ோ
+ௌ
+்
+ௐ
+ௗ
+௦
+௧
+௨
+௩
+௪
+௫
+௬
+௭
+௮
+௯
+௰
+௱
+௲
+௳
+௴
+௵
+௶
+௷
+௸
+௹
+௺
diff --git a/ppocr/utils/dict/ppocrv5_te_dict.txt b/ppocr/utils/dict/ppocrv5_te_dict.txt
new file mode 100644
index 0000000..878f639
--- /dev/null
+++ b/ppocr/utils/dict/ppocrv5_te_dict.txt
@@ -0,0 +1,540 @@
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+<
+=
+>
+?
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+[
+]
+_
+`
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+©
+‥
+{
+}
+\
+|
+@
+^
+~
+÷
+∕
+∙
+⋅
+·
+±
+∓
+∩
+∪
+□
+←
+↔
+⇒
+⇐
+⇔
+∀
+∃
+∄
+∴
+∵
+∝
+∞
+⊥
+∟
+∠
+∡
+∢
+′
+″
+∥
+⊾
+⊿
+∂
+∫
+∬
+∭
+∮
+∯
+∰
+∑
+∏
+√
+∛
+∜
+∱
+∲
+∳
+∶
+∷
+∼
+®
+℉
+Ω
+℧
+Å
+⌀
+ℏ
+⅀
+⍺
+⍵
+¢
+€
+£
+¥
+₿
+Ⅰ
+Ⅱ
+Ⅲ
+Ⅳ
+Ⅴ
+Ⅵ
+Ⅶ
+Ⅷ
+Ⅸ
+Ⅹ
+Ⅺ
+Ⅻ
+ⅰ
+ⅱ
+ⅲ
+ⅳ
+ⅴ
+ⅵ
+ⅶ
+ⅷ
+ⅸ
+ⅹ
+ⅺ
+ⅻ
+➀
+➁
+➂
+➃
+➄
+➅
+➆
+➇
+➈
+➉
+➊
+➋
+➌
+➍
+➎
+➏
+➐
+➑
+➒
+➓
+❶
+❷
+❸
+❹
+❺
+❻
+❼
+❽
+❾
+❿
+①
+②
+③
+④
+⑤
+⑥
+⑦
+⑧
+⑨
+⑩
+●
+▶
+𝑢
+︽
+–
+﹥
+𝜓
+•
+∋
+ƒ
+⬆
+Ạ
+◀
+
+▫
+︾
+À
+Á
+Â
+Ã
+Ä
+Å
+Æ
+Ç
+È
+É
+Ê
+Ë
+Ì
+Í
+Î
+Ï
+Ð
+Ñ
+Ò
+Ó
+Ô
+Õ
+Ö
+Ø
+Ù
+Ú
+Û
+Ü
+Ý
+Þ
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+ì
+í
+î
+ï
+ð
+ñ
+ò
+ó
+ô
+õ
+ö
+ø
+ù
+ú
+û
+ü
+ý
+þ
+ÿ
+¡
+¤
+¦
+§
+¨
+ª
+«
+¬
+¯
+°
+²
+³
+´
+µ
+¶
+¸
+¹
+º
+»
+¼
+½
+¾
+¿
+×
+‐
+‑
+‒
+—
+―
+‖
+‗
+‘
+’
+‚
+‛
+“
+”
+„
+‟
+†
+‡
+‣
+․
+…
+‧
+‰
+‴
+‵
+‶
+‷
+‸
+‹
+›
+※
+‼
+‽
+‾
+₤
+₡
+₹
+−
+∖
+∗
+≈
+≠
+≡
+≤
+≥
+⊂
+⊃
+↑
+→
+↓
+↕
+™
+Ω
+℮
+∆
+✓
+✗
+✘
+▪
+◼
+✔
+✕
+☑
+☒
+№
+₽
+₴
+Α
+α
+Β
+β
+Γ
+γ
+Δ
+δ
+Ε
+ε
+Ζ
+ζ
+Η
+η
+Θ
+θ
+Ι
+ι
+Κ
+κ
+Λ
+λ
+Μ
+μ
+Ν
+ν
+Ξ
+ξ
+Ο
+ο
+Π
+π
+Ρ
+ρ
+Σ
+σ
+ς
+Τ
+τ
+Υ
+υ
+Φ
+φ
+Χ
+χ
+Ψ
+ψ
+ω
+ఀ
+ఁ
+ం
+ః
+ఄ
+అ
+ఆ
+ఇ
+ఈ
+ఉ
+ఊ
+ఋ
+ఌ
+ఎ
+ఏ
+ఐ
+ఒ
+ఓ
+ఔ
+క
+ఖ
+గ
+ఘ
+ఙ
+చ
+ఛ
+జ
+ఝ
+ఞ
+ట
+ఠ
+డ
+ఢ
+ణ
+త
+థ
+ద
+ధ
+న
+ప
+ఫ
+బ
+భ
+మ
+య
+ర
+ఱ
+ల
+ళ
+ఴ
+వ
+శ
+ష
+స
+హ
+఼
+ఽ
+ా
+ి
+ీ
+ు
+ూ
+ృ
+ౄ
+ె
+ే
+ై
+ొ
+ో
+ౌ
+్
+ౕ
+ౖ
+ౘ
+ౙ
+ౚ
+ౝ
+ౠ
+ౡ
+ౢ
+ౣ
+౦
+౧
+౨
+౩
+౪
+౫
+౬
+౭
+౮
+౯
+౷
+౸
+౹
+౺
+౻
+౼
+౽
+౾
+౿
diff --git a/ppocr/utils/dict/ppocrv5_th_dict.txt b/ppocr/utils/dict/ppocrv5_th_dict.txt
new file mode 100644
index 0000000..afab9d5
--- /dev/null
+++ b/ppocr/utils/dict/ppocrv5_th_dict.txt
@@ -0,0 +1,524 @@
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+<
+=
+>
+?
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+[
+]
+_
+`
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+©
+‥
+{
+}
+\
+|
+@
+^
+~
+÷
+∕
+∙
+⋅
+·
+±
+∓
+∩
+∪
+□
+←
+↔
+⇒
+⇐
+⇔
+∀
+∃
+∄
+∴
+∵
+∝
+∞
+⊥
+∟
+∠
+∡
+∢
+′
+″
+∥
+⊾
+⊿
+∂
+∫
+∬
+∭
+∮
+∯
+∰
+∑
+∏
+√
+∛
+∜
+∱
+∲
+∳
+∶
+∷
+∼
+®
+℉
+Ω
+℧
+Å
+⌀
+ℏ
+⅀
+⍺
+⍵
+¢
+€
+£
+¥
+₿
+Ⅰ
+Ⅱ
+Ⅲ
+Ⅳ
+Ⅴ
+Ⅵ
+Ⅶ
+Ⅷ
+Ⅸ
+Ⅹ
+Ⅺ
+Ⅻ
+ⅰ
+ⅱ
+ⅲ
+ⅳ
+ⅴ
+ⅵ
+ⅶ
+ⅷ
+ⅸ
+ⅹ
+ⅺ
+ⅻ
+➀
+➁
+➂
+➃
+➄
+➅
+➆
+➇
+➈
+➉
+➊
+➋
+➌
+➍
+➎
+➏
+➐
+➑
+➒
+➓
+❶
+❷
+❸
+❹
+❺
+❻
+❼
+❽
+❾
+❿
+①
+②
+③
+④
+⑤
+⑥
+⑦
+⑧
+⑨
+⑩
+●
+▶
+𝑢
+︽
+–
+﹥
+𝜓
+•
+∋
+ƒ
+०
+⬆
+Ạ
+◀
+
+▫
+︾
+À
+Á
+Â
+Ã
+Ä
+Å
+Æ
+Ç
+È
+É
+Ê
+Ë
+Ì
+Í
+Î
+Ï
+Ð
+Ñ
+Ò
+Ó
+Ô
+Õ
+Ö
+Ø
+Ù
+Ú
+Û
+Ü
+Ý
+Þ
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+ì
+í
+î
+ï
+ð
+ñ
+ò
+ó
+ô
+õ
+ö
+ø
+ù
+ú
+û
+ü
+ý
+þ
+ÿ
+¡
+¤
+¦
+§
+¨
+ª
+«
+¬
+¯
+°
+²
+³
+´
+µ
+¶
+¸
+¹
+º
+»
+¼
+½
+¾
+¿
+×
+‐
+‑
+‒
+—
+―
+‖
+‗
+‘
+’
+‚
+‛
+“
+”
+„
+‟
+†
+‡
+‣
+․
+…
+‧
+‰
+‴
+‵
+‶
+‷
+‸
+‹
+›
+※
+‼
+‽
+‾
+₤
+₡
+₹
+−
+∖
+∗
+≈
+≠
+≡
+≤
+≥
+⊂
+⊃
+↑
+→
+↓
+↕
+™
+Ω
+℮
+∆
+✓
+✗
+✘
+▪
+◼
+✔
+✕
+☑
+☒
+№
+₽
+₴
+Α
+α
+Β
+β
+Γ
+γ
+Δ
+δ
+Ε
+ε
+Ζ
+ζ
+Η
+η
+Θ
+θ
+Ι
+ι
+Κ
+κ
+Λ
+λ
+Μ
+μ
+Ν
+ν
+Ξ
+ξ
+Ο
+ο
+Π
+π
+Ρ
+ρ
+Σ
+σ
+ς
+Τ
+τ
+Υ
+υ
+Φ
+φ
+Χ
+χ
+Ψ
+ψ
+ω
+ก
+ข
+ค
+ฅ
+ฆ
+ง
+จ
+ฉ
+ช
+ซ
+ฌ
+ญ
+ฎ
+ฏ
+ฐ
+ฑ
+ฒ
+ณ
+ด
+ต
+ถ
+ท
+ธ
+น
+บ
+ป
+ผ
+ฝ
+พ
+ฟ
+ภ
+ม
+ย
+ร
+ฤ
+ล
+ฦ
+ว
+ศ
+ษ
+ส
+ห
+ฬ
+อ
+ฮ
+ฯ
+ะ
+ั
+า
+ำ
+ิ
+ี
+ึ
+ื
+ุ
+ู
+ฺ
+฿
+เ
+แ
+โ
+ใ
+ไ
+ๅ
+ๆ
+็
+่
+้
+๊
+๋
+์
+ํ
+๐
+๑
+๒
+๓
+๔
+๕
+๖
+๗
+๘
+๙
+๚
diff --git a/ppocr/utils/dict/ppocrv6_dict.txt b/ppocr/utils/dict/ppocrv6_dict.txt
new file mode 100644
index 0000000..270082a
--- /dev/null
+++ b/ppocr/utils/dict/ppocrv6_dict.txt
@@ -0,0 +1,18708 @@
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+:
+;
+<
+=
+>
+?
+@
+[
+\
+]
+^
+_
+`
+{
+|
+}
+~
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+¢
+£
+¤
+¥
+¦
+§
+¨
+©
+ª
+«
+¬
+®
+¯
+°
+±
+²
+³
+´
+µ
+¶
+·
+¸
+¹
+º
+»
+¼
+½
+¾
+¿
+À
+Á
+Â
+Ã
+Ä
+Å
+Æ
+Ç
+È
+É
+Ê
+Ë
+Ì
+Í
+Î
+Ï
+Ð
+Ñ
+Ò
+Ó
+Ô
+Õ
+Ö
+×
+Ø
+Ù
+Ú
+Û
+Ü
+Ý
+Þ
+ß
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+ì
+í
+î
+ï
+ð
+ñ
+ò
+ó
+ô
+õ
+ö
+÷
+ø
+ù
+ú
+û
+ü
+ý
+þ
+ÿ
+Ā
+ā
+Ă
+ă
+Ą
+ą
+Ć
+ć
+Ĉ
+ĉ
+Ċ
+ċ
+Č
+č
+Ď
+ď
+Đ
+đ
+Ē
+ē
+Ĕ
+ĕ
+Ė
+ė
+Ę
+ę
+Ě
+ě
+Ĝ
+ĝ
+Ğ
+ğ
+Ġ
+ġ
+Ģ
+ģ
+Ĥ
+ĥ
+Ħ
+ħ
+Ĩ
+ĩ
+Ī
+ī
+Ĭ
+ĭ
+Į
+į
+İ
+ı
+IJ
+ij
+Ĵ
+ĵ
+Ķ
+ķ
+ĸ
+Ĺ
+ĺ
+Ļ
+ļ
+Ľ
+ľ
+Ŀ
+ŀ
+Ł
+ł
+Ń
+ń
+Ņ
+ņ
+Ň
+ň
+ʼn
+Ŋ
+ŋ
+Ō
+ō
+Ŏ
+ŏ
+Ő
+ő
+Œ
+œ
+Ŕ
+ŕ
+Ŗ
+ŗ
+Ř
+ř
+Ś
+ś
+Ŝ
+ŝ
+Ş
+ş
+Š
+š
+Ţ
+ţ
+Ť
+ť
+Ŧ
+ŧ
+Ũ
+ũ
+Ū
+ū
+Ŭ
+ŭ
+Ů
+ů
+Ű
+ű
+Ų
+ų
+Ŵ
+ŵ
+Ŷ
+ŷ
+Ÿ
+Ź
+ź
+Ż
+ż
+Ž
+ž
+ſ
+ƀ
+Ɓ
+Ƃ
+ƃ
+Ƅ
+ƅ
+Ɔ
+Ƈ
+ƈ
+Ɖ
+Ɗ
+Ƌ
+ƌ
+ƍ
+Ǝ
+Ə
+Ɛ
+Ƒ
+ƒ
+Ɠ
+Ɣ
+ƕ
+Ɩ
+Ɨ
+Ƙ
+ƙ
+ƚ
+ƛ
+Ɯ
+Ɲ
+ƞ
+Ɵ
+Ơ
+ơ
+Ƣ
+ƣ
+Ƥ
+ƥ
+Ʀ
+Ƨ
+ƨ
+Ʃ
+ƪ
+ƫ
+Ƭ
+ƭ
+Ʈ
+Ư
+ư
+Ʊ
+Ʋ
+Ƴ
+ƴ
+Ƶ
+ƶ
+Ʒ
+Ƹ
+ƹ
+ƺ
+ƻ
+Ƽ
+ƽ
+ƾ
+ƿ
+ǀ
+ǁ
+ǂ
+ǃ
+DŽ
+Dž
+dž
+LJ
+Lj
+lj
+NJ
+Nj
+nj
+Ǎ
+ǎ
+Ǐ
+ǐ
+Ǒ
+ǒ
+Ǔ
+ǔ
+Ǖ
+ǖ
+Ǘ
+ǘ
+Ǚ
+ǚ
+Ǜ
+ǜ
+ǝ
+Ǟ
+ǟ
+Ǡ
+ǡ
+Ǣ
+ǣ
+Ǥ
+ǥ
+Ǧ
+ǧ
+Ǩ
+ǩ
+Ǫ
+ǫ
+Ǭ
+ǭ
+Ǯ
+ǯ
+ǰ
+DZ
+Dz
+dz
+Ǵ
+ǵ
+Ƕ
+Ƿ
+Ǹ
+ǹ
+Ǻ
+ǻ
+Ǽ
+ǽ
+Ǿ
+ǿ
+Ȁ
+ȁ
+Ȃ
+ȃ
+Ȅ
+ȅ
+Ȇ
+ȇ
+Ȉ
+ȉ
+Ȋ
+ȋ
+Ȍ
+ȍ
+Ȏ
+ȏ
+Ȑ
+ȑ
+Ȓ
+ȓ
+Ȕ
+ȕ
+Ȗ
+ȗ
+Ș
+ș
+Ț
+ț
+Ȝ
+ȝ
+Ȟ
+ȟ
+Ƞ
+ȡ
+Ȣ
+ȣ
+Ȥ
+ȥ
+Ȧ
+ȧ
+Ȩ
+ȩ
+Ȫ
+ȫ
+Ȭ
+ȭ
+Ȯ
+ȯ
+Ȱ
+ȱ
+Ȳ
+ȳ
+ȴ
+ȵ
+ȶ
+ȷ
+ȸ
+ȹ
+Ⱥ
+Ȼ
+ȼ
+Ƚ
+Ⱦ
+ȿ
+ɀ
+Ɂ
+ɂ
+Ƀ
+Ʉ
+Ʌ
+Ɇ
+ɇ
+Ɉ
+ɉ
+Ɋ
+ɋ
+Ɍ
+ɍ
+Ɏ
+ɏ
+ɐ
+ɑ
+ɒ
+ɓ
+ɔ
+ɕ
+ɖ
+ɗ
+ɘ
+ə
+ɚ
+ɛ
+ɜ
+ɝ
+ɞ
+ɟ
+ɠ
+ɡ
+ɢ
+ɣ
+ɤ
+ɥ
+ɦ
+ɧ
+ɨ
+ɩ
+ɪ
+ɫ
+ɬ
+ɭ
+ɮ
+ɯ
+ɰ
+ɱ
+ɲ
+ɳ
+ɴ
+ɵ
+ɶ
+ɷ
+ɸ
+ɹ
+ɺ
+ɻ
+ɼ
+ɽ
+ɾ
+ɿ
+ʀ
+ʁ
+ʂ
+ʃ
+ʄ
+ʅ
+ʆ
+ʇ
+ʈ
+ʉ
+ʊ
+ʋ
+ʌ
+ʍ
+ʎ
+ʏ
+ʐ
+ʑ
+ʒ
+ʓ
+ʔ
+ʕ
+ʖ
+ʗ
+ʘ
+ʙ
+ʚ
+ʛ
+ʜ
+ʝ
+ʞ
+ʟ
+ʠ
+ʡ
+ʢ
+ʣ
+ʤ
+ʥ
+ʦ
+ʧ
+ʨ
+ʩ
+ʪ
+ʫ
+ʬ
+ʭ
+ʮ
+ʯ
+Α
+Β
+Γ
+Δ
+Ε
+Ζ
+Η
+Θ
+Ι
+Κ
+Λ
+Μ
+Ν
+Ξ
+Ο
+Π
+Ρ
+Σ
+Τ
+Υ
+Φ
+Χ
+Ψ
+Ω
+α
+β
+γ
+δ
+ε
+ζ
+η
+θ
+ι
+κ
+λ
+μ
+ν
+ξ
+ο
+π
+ρ
+σ
+τ
+υ
+φ
+χ
+ψ
+ω
+ϐ
+ϑ
+ϒ
+ϕ
+ϖ
+ϝ
+Ϟ
+ϟ
+Ϡ
+ϡ
+Ϣ
+ϣ
+Ϥ
+ϥ
+Ϧ
+ϧ
+Ϩ
+ϩ
+Ϫ
+ϫ
+Ϭ
+ϭ
+Ϯ
+ϯ
+ϰ
+ϱ
+ϴ
+ϵ
+Ạ
+ồ
+–
+—
+―
+‖
+‘
+’
+“
+”
+†
+‡
+•
+‥
+…
+‰
+′
+″
+※
+⁎
+⁰
+⁴
+⁵
+⁶
+⁷
+⁸
+⁹
+⁺
+⁻
+⁼
+⁽
+⁾
+ⁿ
+₀
+₁
+₂
+₃
+₄
+₅
+₆
+₇
+₈
+₉
+₊
+₋
+₌
+₍
+₎
+ₒ
+₠
+₡
+₢
+₣
+₤
+₥
+₦
+₧
+₨
+₩
+₪
+₫
+€
+₭
+₮
+₯
+₰
+₱
+₲
+₳
+₴
+₵
+₶
+₷
+₸
+₹
+₺
+₻
+₼
+₽
+₾
+₿
+℃
+℉
+ℎ
+ℏ
+ℑ
+℘
+ℜ
+™
+℧
+Å
+ℵ
+ℶ
+ℷ
+ℸ
+ℹ
+⅀
+Ⅰ
+Ⅱ
+Ⅲ
+Ⅳ
+Ⅴ
+Ⅵ
+Ⅶ
+Ⅷ
+Ⅸ
+Ⅹ
+Ⅺ
+Ⅻ
+ⅰ
+ⅱ
+ⅲ
+ⅳ
+ⅴ
+ⅵ
+ⅶ
+ⅷ
+ⅸ
+ⅹ
+ⅺ
+ⅻ
+ⅼ
+ↀ
+ↁ
+ↂ
+←
+↑
+→
+↓
+↔
+↕
+↖
+↗
+↘
+↙
+↚
+↛
+↜
+↝
+↞
+↟
+↠
+↡
+↢
+↣
+↤
+↥
+↦
+↧
+↨
+↩
+↪
+↫
+↬
+↭
+↮
+↯
+↰
+↱
+↲
+↳
+↴
+↵
+↶
+↷
+↸
+↹
+↺
+↻
+↼
+↽
+↾
+↿
+⇀
+⇁
+⇂
+⇃
+⇄
+⇅
+⇆
+⇇
+⇈
+⇉
+⇊
+⇋
+⇌
+⇍
+⇎
+⇏
+⇐
+⇑
+⇒
+⇓
+⇔
+⇕
+⇖
+⇗
+⇘
+⇙
+⇚
+⇛
+⇜
+⇝
+⇞
+⇟
+⇠
+⇡
+⇢
+⇣
+⇤
+⇥
+⇦
+⇧
+⇨
+⇩
+⇪
+⇫
+⇬
+⇭
+⇮
+⇯
+⇰
+⇱
+⇲
+⇳
+⇴
+⇵
+⇶
+⇷
+⇸
+⇹
+⇺
+⇻
+⇼
+⇽
+⇾
+⇿
+∀
+∁
+∂
+∃
+∄
+∅
+∆
+∇
+∈
+∉
+∋
+∏
+∑
+−
+∓
+∕
+∖
+∙
+√
+∛
+∜
+∝
+∞
+∟
+∠
+∡
+∢
+∥
+∧
+∨
+∩
+∪
+∫
+∬
+∭
+∮
+∯
+∰
+∱
+∲
+∳
+∴
+∵
+∶
+∷
+∸
+∹
+∺
+∻
+∼
+∽
+∾
+∿
+≀
+≁
+≂
+≃
+≄
+≅
+≆
+≇
+≈
+≉
+≊
+≋
+≌
+≍
+≎
+≏
+≐
+≑
+≒
+≓
+≔
+≕
+≖
+≗
+≘
+≙
+≚
+≛
+≜
+≝
+≞
+≟
+≠
+≡
+≢
+≣
+≤
+≥
+≦
+≧
+≨
+≩
+≪
+≫
+≬
+≭
+≮
+≯
+≰
+≱
+≲
+≳
+≴
+≵
+≶
+≷
+≸
+≹
+≺
+≻
+≼
+≽
+≾
+≿
+⊀
+⊁
+⊂
+⊃
+⊄
+⊅
+⊆
+⊇
+⊈
+⊉
+⊊
+⊋
+⊌
+⊍
+⊎
+⊏
+⊐
+⊑
+⊒
+⊓
+⊔
+⊕
+⊖
+⊗
+⊘
+⊙
+⊚
+⊛
+⊜
+⊝
+⊞
+⊟
+⊠
+⊡
+⊢
+⊣
+⊤
+⊥
+⊦
+⊧
+⊨
+⊩
+⊪
+⊫
+⊬
+⊭
+⊮
+⊯
+⊰
+⊱
+⊲
+⊳
+⊴
+⊵
+⊶
+⊷
+⊸
+⊹
+⊺
+⊻
+⊼
+⊽
+⊾
+⊿
+⋅
+⌀
+⌃
+⌘
+⌚
+⌛
+⌤
+⌥
+⌦
+⌧
+⌨
+〉
+⌫
+⌬
+⌭
+⌮
+⌯
+⍵
+⍺
+⎆
+⎇
+⎈
+⎉
+⎊
+⎋
+⎌
+⎍
+⎎
+⎏
+⎐
+⎑
+⎒
+⎓
+⎔
+⎕
+⎖
+⎗
+⎘
+⎙
+⎚
+⎛
+⎜
+⎝
+⎞
+⎟
+⎠
+⎡
+⎢
+⎣
+⎤
+⎥
+⎦
+⎧
+⎨
+⎩
+⎪
+⎫
+⎬
+⎭
+⎮
+⎯
+⎰
+⎱
+⎲
+⎳
+⎴
+⎵
+⎶
+⎷
+⎸
+⎹
+⎺
+⎻
+⎼
+⎽
+⎾
+⎿
+⏀
+⏁
+⏂
+⏃
+⏄
+⏅
+⏆
+⏇
+⏈
+⏉
+⏊
+⏋
+⏌
+⏍
+⏎
+⏏
+⏐
+⏑
+⏒
+⏓
+⏔
+⏕
+⏖
+⏗
+⏘
+⏙
+⏚
+⏛
+⏜
+⏝
+⏞
+⏟
+⏠
+⏡
+⏢
+⏣
+⏤
+⏥
+⏦
+⏧
+⏨
+⏩
+⏪
+⏫
+⏬
+⏭
+⏮
+⏯
+⏰
+⏱
+⏲
+⏳
+⏴
+⏵
+⏶
+⏷
+⏸
+⏹
+⏺
+⏻
+⏼
+⏽
+⏾
+⏿
+①
+②
+③
+④
+⑤
+⑥
+⑦
+⑧
+⑨
+⑩
+⑪
+⑫
+⑬
+⑭
+⑮
+⑯
+⑰
+⑱
+⑲
+⑳
+Ⓜ
+━
+╆
+╳
+▁
+█
+■
+□
+▪
+▫
+▬
+▲
+△
+▶
+▷
+▼
+▽
+◀
+◆
+◇
+◉
+◊
+○
+◎
+●
+◐
+◥
+◻
+◼
+◽
+◾
+☀
+☁
+☂
+☃
+☄
+★
+☆
+☇
+☈
+☉
+☊
+☋
+☌
+☍
+☎
+☏
+☐
+☑
+☒
+☓
+☔
+☕
+☖
+☗
+☘
+☙
+☚
+☛
+☜
+☝
+☞
+☟
+☠
+☡
+☢
+☣
+☤
+☥
+☦
+☧
+☨
+☩
+☪
+☫
+☬
+☭
+☮
+☯
+☰
+☱
+☲
+☳
+☴
+☵
+☶
+☷
+☸
+☹
+☺
+☻
+☼
+☽
+☾
+☿
+♀
+♁
+♂
+♃
+♄
+♅
+♆
+♇
+♈
+♉
+♊
+♋
+♌
+♍
+♎
+♏
+♐
+♑
+♒
+♓
+♔
+♕
+♖
+♗
+♘
+♙
+♚
+♛
+♜
+♝
+♞
+♟
+♠
+♡
+♢
+♣
+♤
+♥
+♦
+♧
+♨
+♩
+♪
+♫
+♬
+♭
+♮
+♯
+♰
+♱
+♲
+♳
+♴
+♵
+♶
+♷
+♸
+♹
+♺
+♻
+♼
+♽
+♾
+♿
+⚀
+⚁
+⚂
+⚃
+⚄
+⚅
+⚆
+⚇
+⚈
+⚉
+⚊
+⚋
+⚌
+⚍
+⚎
+⚏
+⚐
+⚑
+⚒
+⚓
+⚔
+⚕
+⚖
+⚗
+⚘
+⚙
+⚚
+⚛
+⚜
+⚝
+⚞
+⚟
+⚠
+⚡
+⚢
+⚣
+⚤
+⚥
+⚦
+⚧
+⚨
+⚩
+⚪
+⚫
+⚬
+⚭
+⚮
+⚯
+⚰
+⚱
+⚲
+⚳
+⚴
+⚵
+⚶
+⚷
+⚸
+⚹
+⚺
+⚻
+⚼
+⚽
+⚾
+⚿
+⛀
+⛁
+⛂
+⛃
+⛄
+⛅
+⛆
+⛇
+⛈
+⛉
+⛊
+⛋
+⛌
+⛍
+⛎
+⛏
+⛐
+⛑
+⛒
+⛓
+⛔
+⛕
+⛖
+⛗
+⛘
+⛙
+⛚
+⛛
+⛜
+⛝
+⛞
+⛠
+⛡
+⛢
+⛣
+⛤
+⛥
+⛦
+⛧
+⛨
+⛩
+⛪
+⛫
+⛬
+⛭
+⛮
+⛯
+⛲
+⛳
+⛵
+⛶
+⛺
+⛽
+⛾
+⛿
+✂
+✅
+✆
+✇
+✈
+✉
+✊
+✋
+✌
+✍
+✎
+✏
+✐
+✑
+✒
+✓
+✔
+✕
+✖
+✘
+✙
+✚
+✛
+✜
+✝
+✞
+✟
+✠
+✡
+✢
+✣
+✤
+✥
+✦
+✧
+✨
+✩
+✪
+✫
+✬
+✭
+✮
+✯
+✰
+✱
+✲
+✳
+✴
+✵
+✶
+✷
+✸
+✹
+✺
+✻
+✼
+✽
+✾
+✿
+❀
+❁
+❂
+❃
+❄
+❅
+❆
+❇
+❈
+❉
+❊
+❋
+❌
+❍
+❎
+❏
+❐
+❑
+❒
+❓
+❔
+❕
+❖
+❗
+❘
+❙
+❚
+❛
+❜
+❝
+❞
+❡
+❢
+❣
+❤
+❥
+❦
+❧
+❨
+❩
+❪
+❫
+❬
+❭
+❮
+❯
+❰
+❱
+❲
+❳
+❴
+❵
+❶
+❷
+❸
+❹
+❺
+❻
+❼
+❽
+❾
+❿
+➀
+➁
+➂
+➃
+➄
+➅
+➆
+➇
+➈
+➉
+➊
+➋
+➌
+➍
+➎
+➏
+➐
+➑
+➒
+➓
+➔
+➕
+➖
+➗
+➘
+➙
+➚
+➛
+➜
+➝
+➞
+➟
+➠
+➡
+➢
+➣
+➤
+➥
+➦
+➧
+➨
+➩
+➪
+➫
+➬
+➭
+➮
+➯
+➰
+➱
+➲
+➳
+➴
+➵
+➶
+➷
+➸
+➹
+➺
+➻
+➼
+➽
+➾
+➿
+⟶
+⟹
+⤴
+⤵
+⬅
+⬆
+⬇
+⬛
+⬜
+⭐
+⭕
+　
+、
+。
+々
+〆
+〇
+〈
+〉
+《
+》
+「
+」
+『
+』
+【
+】
+〒
+〔
+〕
+〖
+〗
+〜
+〰
+〵
+〽
+ぁ
+あ
+ぃ
+い
+ぅ
+う
+ぇ
+え
+ぉ
+お
+か
+が
+き
+ぎ
+く
+ぐ
+け
+げ
+こ
+ご
+さ
+ざ
+し
+じ
+す
+ず
+せ
+ぜ
+そ
+ぞ
+た
+だ
+ち
+ぢ
+っ
+つ
+づ
+て
+で
+と
+ど
+な
+に
+ぬ
+ね
+の
+は
+ば
+ぱ
+ひ
+び
+ぴ
+ふ
+ぶ
+ぷ
+へ
+べ
+ぺ
+ほ
+ぼ
+ぽ
+ま
+み
+む
+め
+も
+ゃ
+や
+ゅ
+ゆ
+ょ
+よ
+ら
+り
+る
+れ
+ろ
+ゎ
+わ
+ゑ
+を
+ん
+ゕ
+ゖ
+ゝ
+ゞ
+ァ
+ア
+ィ
+イ
+ゥ
+ウ
+ェ
+エ
+ォ
+オ
+カ
+ガ
+キ
+ギ
+ク
+グ
+ケ
+ゲ
+コ
+ゴ
+サ
+ザ
+シ
+ジ
+ス
+ズ
+セ
+ゼ
+ソ
+ゾ
+タ
+ダ
+チ
+ヂ
+ッ
+ツ
+ヅ
+テ
+デ
+ト
+ド
+ナ
+ニ
+ヌ
+ネ
+ノ
+ハ
+バ
+パ
+ヒ
+ビ
+ピ
+フ
+ブ
+プ
+ヘ
+ベ
+ペ
+ホ
+ボ
+ポ
+マ
+ミ
+ム
+メ
+モ
+ャ
+ヤ
+ュ
+ユ
+ョ
+ヨ
+ラ
+リ
+ル
+レ
+ロ
+ヮ
+ワ
+ヰ
+ヲ
+ン
+ヴ
+ヵ
+ヶ
+ヷ
+ヸ
+ヹ
+ヺ
+・
+ー
+ヽ
+ヾ
+ヿ
+㈱
+㊗
+㊙
+㐂
+㐱
+㑇
+㑊
+㑳
+㑺
+㒺
+㕁
+㕑
+㕥
+㕮
+㘎
+㘭
+㙍
+㙘
+㙟
+㙦
+㚷
+㛃
+㛚
+㛹
+㝛
+㝠
+㝡
+㟁
+㟃
+㠇
+㠓
+㠣
+㠯
+㠶
+㡌
+㢘
+㤘
+㤙
+㥄
+㥫
+㥮
+㧐
+㧑
+㧟
+㧱
+㨗
+㨪
+㩗
+㩦
+㩳
+㪚
+㪟
+㫰
+㬉
+㬊
+㬎
+㬚
+㬪
+㭎
+㭕
+㮣
+㮾
+㰀
+㳄
+㳇
+㳒
+㳘
+㳚
+㴔
+㴪
+㴱
+㵐
+㶲
+㸃
+㸆
+㸌
+㺄
+㻬
+㼝
+㽏
+㽞
+㿠
+䁖
+䂮
+䃅
+䃎
+䃮
+䅟
+䈰
+䊀
+䌹
+䎃
+䎖
+䏝
+䏡
+䏲
+䐃
+䓖
+䓛
+䓣
+䓨
+䓫
+䓬
+䖝
+䗖
+䗛
+䗪
+䗬
+䗴
+䘏
+䘑
+䘚
+䜣
+䝉
+䝔
+䝙
+䠀
+䠶
+䢺
+䢼
+䣘
+䥽
+䦃
+䮄
+䰟
+䰾
+䲁
+䲟
+䲠
+䲢
+䴉
+䴓
+䴔
+䴕
+䴖
+䴗
+䴘
+䴙
+䶊
+䶮
+一
+丁
+七
+丄
+万
+丈
+三
+上
+下
+丌
+不
+与
+丏
+丐
+丑
+专
+且
+丕
+世
+丘
+丙
+业
+丛
+东
+丝
+丞
+丟
+両
+丢
+两
+严
+並
+丧
+个
+丫
+中
+丮
+丰
+丱
+串
+临
+丸
+丹
+为
+主
+丼
+丽
+举
+丿
+乂
+乃
+久
+么
+义
+之
+乌
+乍
+乎
+乏
+乐
+乒
+乓
+乔
+乖
+乗
+乘
+乙
+乜
+九
+乞
+也
+习
+乡
+书
+乩
+乭
+买
+乱
+乳
+乸
+乹
+乾
+亀
+亂
+了
+予
+争
+亊
+事
+二
+亍
+于
+亏
+云
+互
+亓
+五
+井
+亘
+亙
+亚
+些
+亜
+亞
+亟
+亡
+亢
+交
+亥
+亦
+产
+亨
+亩
+享
+京
+亭
+亮
+亯
+亰
+亱
+亲
+亳
+亵
+亶
+亷
+亸
+亹
+人
+亾
+亿
+什
+仁
+仂
+仃
+仄
+仅
+仆
+仇
+仉
+今
+介
+仍
+从
+仏
+仑
+仓
+仔
+仕
+他
+仗
+付
+仙
+仛
+仜
+仝
+仞
+仟
+仡
+代
+令
+以
+仨
+仩
+仪
+仫
+们
+仮
+仰
+仱
+仲
+仳
+仵
+件
+价
+仺
+任
+仼
+份
+仿
+伀
+企
+伃
+伈
+伉
+伊
+伋
+伍
+伎
+伏
+伐
+休
+伔
+伕
+众
+优
+伙
+会
+伛
+伝
+伞
+伟
+传
+伢
+伣
+伤
+伥
+伦
+伧
+伪
+伫
+伬
+伭
+伯
+估
+伱
+伲
+伴
+伶
+伷
+伸
+伺
+伻
+似
+伽
+伾
+佀
+佁
+佃
+但
+佇
+佈
+佉
+佋
+佌
+位
+低
+住
+佐
+佑
+佒
+体
+佔
+何
+佖
+佗
+佘
+余
+佚
+佛
+作
+佝
+佞
+佟
+你
+佢
+佣
+佤
+佥
+佧
+佩
+佪
+佬
+佯
+佰
+佳
+佴
+併
+佶
+佷
+佸
+佹
+佺
+佻
+佼
+佽
+佾
+使
+侀
+侁
+侂
+侃
+侄
+侅
+來
+侇
+侈
+侉
+侊
+例
+侍
+侏
+侐
+侑
+侔
+侖
+侗
+侘
+侚
+供
+侜
+依
+侞
+侠
+価
+侣
+侥
+侦
+侧
+侨
+侩
+侪
+侬
+侮
+侯
+侲
+侳
+侴
+侵
+侶
+侷
+侹
+侻
+便
+俁
+係
+促
+俄
+俅
+俇
+俉
+俊
+俋
+俌
+俍
+俎
+俏
+俐
+俑
+俓
+俔
+俗
+俘
+俙
+俚
+俛
+俜
+保
+俞
+俟
+俠
+信
+俣
+俤
+俦
+俨
+俩
+俪
+俫
+俬
+俭
+修
+俯
+俱
+俳
+俴
+俵
+俶
+俷
+俸
+俺
+俽
+俾
+倀
+倃
+倅
+倆
+倇
+倈
+倉
+個
+倌
+倍
+倏
+倐
+們
+倒
+倓
+倔
+倕
+倖
+倗
+倘
+候
+倚
+倛
+倜
+倞
+借
+倠
+倡
+倢
+倣
+値
+倥
+倦
+倧
+倨
+倩
+倪
+倫
+倬
+倭
+倮
+倰
+倳
+倴
+倶
+倷
+倸
+倹
+债
+倻
+值
+倾
+偁
+偃
+偅
+假
+偈
+偉
+偊
+偋
+偌
+偍
+偎
+偏
+偐
+偓
+偕
+偘
+做
+偛
+停
+偝
+偞
+偟
+偠
+偡
+偢
+健
+偨
+偩
+偪
+偫
+偬
+偭
+偯
+偰
+偲
+側
+偵
+偶
+偷
+偸
+偺
+偻
+偽
+偾
+偿
+傀
+傂
+傃
+傅
+傈
+傉
+傋
+傌
+傍
+傎
+傑
+傒
+傔
+傕
+傖
+傘
+備
+傚
+傛
+傜
+傝
+傞
+傢
+傣
+傥
+傧
+储
+傩
+催
+傭
+傮
+傯
+傰
+傱
+傲
+傳
+傴
+債
+傷
+傺
+傻
+傽
+傾
+傿
+僁
+僂
+僄
+僅
+僆
+僇
+僈
+僉
+僊
+僋
+働
+僎
+像
+僑
+僓
+僔
+僕
+僖
+僗
+僙
+僚
+僛
+僜
+僝
+僞
+僡
+僤
+僥
+僦
+僧
+僨
+僩
+僪
+僬
+僭
+僮
+僰
+僱
+僳
+僴
+僵
+僶
+僸
+價
+僻
+僽
+僾
+僿
+儀
+儁
+儂
+儃
+億
+儆
+儇
+儈
+儉
+儋
+儌
+儐
+儑
+儒
+儓
+儔
+儕
+儗
+儘
+儚
+儛
+儜
+儞
+償
+儠
+儡
+儢
+儤
+儥
+儦
+優
+儫
+儰
+儱
+儲
+儳
+儴
+儵
+儷
+儸
+儹
+儺
+儻
+儼
+儽
+儿
+兀
+允
+元
+兄
+充
+兆
+兇
+先
+光
+克
+兌
+免
+兎
+児
+兑
+兒
+兔
+兕
+兖
+兗
+兙
+党
+兛
+兜
+兝
+兞
+兠
+兡
+兢
+兣
+入
+內
+全
+兩
+兪
+八
+公
+六
+兮
+兰
+共
+关
+兴
+兵
+其
+具
+典
+兹
+养
+兼
+兽
+兿
+冀
+冁
+冂
+冄
+内
+円
+冇
+冈
+冉
+冊
+册
+再
+冏
+冐
+冑
+冒
+冓
+冔
+冕
+冗
+冘
+写
+冚
+军
+农
+冞
+冠
+冢
+冣
+冤
+冥
+冧
+冨
+冪
+冬
+冮
+冯
+冰
+冱
+冲
+决
+冴
+况
+冶
+冷
+冺
+冻
+冼
+冽
+净
+凃
+凄
+凅
+准
+凇
+凈
+凉
+凊
+凋
+凌
+凍
+减
+凑
+凔
+凖
+凘
+凛
+凜
+凝
+凞
+几
+凡
+凢
+凤
+処
+凪
+凫
+凬
+凭
+凯
+凰
+凱
+凳
+凵
+凶
+凸
+凹
+出
+击
+凼
+函
+凿
+刀
+刁
+刂
+刃
+刄
+分
+切
+刈
+刉
+刊
+刌
+刍
+刎
+刑
+划
+刓
+刖
+列
+刘
+则
+刚
+创
+刜
+初
+刞
+删
+判
+別
+刦
+刧
+刨
+利
+刪
+别
+刬
+刭
+刮
+到
+刱
+刲
+刳
+刵
+制
+刷
+券
+刹
+刺
+刻
+刼
+刽
+刿
+剀
+剁
+剂
+剃
+剄
+剅
+剆
+則
+剉
+削
+剋
+剌
+前
+剎
+剏
+剐
+剑
+剒
+剔
+剕
+剖
+剗
+剚
+剛
+剜
+剝
+剞
+剟
+剡
+剣
+剤
+剥
+剧
+剩
+剪
+剫
+剬
+剭
+剮
+副
+剰
+割
+剳
+剴
+創
+剷
+剸
+剹
+剺
+剻
+剼
+剽
+剿
+劀
+劁
+劂
+劃
+劄
+劇
+劈
+劉
+劊
+劋
+劌
+劍
+劐
+劑
+劓
+劔
+劖
+劗
+劘
+劙
+力
+劝
+办
+功
+加
+务
+劢
+劣
+劦
+动
+助
+努
+劫
+劬
+劭
+励
+劲
+劳
+労
+劵
+効
+劼
+劾
+势
+勁
+勃
+勅
+勇
+勉
+勋
+勌
+勍
+勐
+勑
+勒
+勔
+動
+勖
+勗
+勘
+務
+勚
+勛
+勝
+勞
+募
+勢
+勣
+勤
+勦
+勧
+勩
+勰
+勱
+勲
+勳
+勴
+勵
+勷
+勸
+勺
+勻
+勾
+勿
+匀
+匁
+匂
+匃
+匄
+包
+匆
+匈
+匊
+匋
+匍
+匏
+匐
+匕
+化
+北
+匙
+匜
+匝
+匟
+匠
+匡
+匢
+匣
+匦
+匪
+匭
+匮
+匯
+匰
+匱
+匲
+匴
+匷
+匸
+匹
+区
+医
+匼
+匽
+匾
+匿
+區
+十
+千
+卅
+升
+午
+卉
+半
+卋
+卌
+卍
+华
+协
+卐
+卑
+卒
+卓
+協
+单
+卖
+南
+単
+博
+卜
+卞
+卟
+占
+卡
+卢
+卣
+卤
+卦
+卧
+卫
+卬
+卮
+卯
+印
+危
+卲
+即
+却
+卵
+卷
+卸
+卹
+卺
+卻
+卼
+卽
+卿
+厀
+厂
+厄
+厅
+历
+厉
+压
+厌
+厍
+厎
+厒
+厓
+厔
+厕
+厖
+厗
+厘
+厙
+厚
+厜
+厝
+厞
+原
+厠
+厢
+厣
+厤
+厥
+厦
+厨
+厩
+厬
+厭
+厮
+厰
+厲
+厳
+厴
+厹
+去
+厾
+县
+叁
+参
+參
+叄
+叆
+叇
+又
+叉
+及
+友
+双
+反
+収
+发
+叔
+叕
+取
+受
+变
+叙
+叚
+叛
+叟
+叠
+叡
+叢
+口
+古
+句
+另
+叨
+叩
+只
+叫
+召
+叭
+叮
+可
+台
+叱
+史
+右
+叵
+叶
+号
+司
+叹
+叻
+叼
+叽
+吁
+吃
+各
+吆
+合
+吉
+吊
+吋
+同
+名
+后
+吏
+吐
+向
+吒
+吓
+吔
+吕
+吖
+吗
+吚
+君
+吝
+吞
+吟
+吠
+吡
+吣
+吥
+否
+吧
+吨
+吩
+吪
+含
+听
+吭
+吮
+启
+吰
+吱
+吲
+吳
+吴
+吵
+吶
+吷
+吸
+吹
+吻
+吼
+吽
+吾
+呀
+呂
+呃
+呆
+呇
+呈
+呉
+告
+呋
+呌
+呎
+呐
+呑
+呒
+呓
+呔
+呕
+呖
+呗
+员
+呙
+呛
+呜
+呢
+呣
+呤
+呥
+呦
+周
+呪
+呫
+呬
+呰
+呱
+呲
+味
+呴
+呵
+呶
+呷
+呸
+呺
+呻
+呼
+命
+呾
+呿
+咀
+咁
+咂
+咄
+咅
+咆
+咇
+咈
+咉
+咊
+咋
+和
+咍
+咎
+咏
+咐
+咑
+咒
+咔
+咕
+咖
+咗
+咘
+咙
+咚
+咛
+咝
+咟
+咠
+咡
+咢
+咣
+咤
+咥
+咦
+咧
+咨
+咩
+咪
+咫
+咬
+咭
+咮
+咯
+咱
+咲
+咳
+咴
+咶
+咷
+咸
+咺
+咻
+咼
+咽
+咾
+咿
+哀
+品
+哂
+哃
+哄
+哅
+哆
+哇
+哈
+哉
+哌
+响
+哎
+哏
+哐
+哑
+哒
+哓
+哔
+哕
+哖
+哗
+哙
+哚
+哝
+哞
+哟
+員
+哢
+哤
+哥
+哦
+哧
+哨
+哩
+哪
+哫
+哭
+哮
+哱
+哲
+哳
+哶
+哺
+哼
+哽
+哿
+唁
+唃
+唄
+唅
+唆
+唇
+唈
+唉
+唌
+唏
+唐
+唑
+唒
+唔
+唕
+唖
+唗
+唘
+唚
+唛
+唝
+唠
+唢
+唣
+唤
+唦
+唧
+唪
+唫
+唬
+唭
+售
+唯
+唰
+唱
+唲
+唳
+唴
+唵
+唶
+唷
+唸
+唻
+唼
+唾
+唿
+啁
+啃
+啄
+啅
+商
+啉
+啊
+啍
+啎
+問
+啐
+啑
+啒
+啓
+啕
+啖
+啗
+啚
+啜
+啞
+啟
+啡
+啢
+啣
+啤
+啥
+啦
+啧
+啪
+啫
+啬
+啭
+啮
+啯
+啰
+啱
+啲
+啴
+啵
+啶
+啷
+啸
+啻
+啼
+啽
+啾
+喀
+喁
+喂
+喃
+善
+喆
+喇
+喈
+喉
+喊
+喋
+喌
+喍
+喏
+喑
+喒
+喓
+喔
+喘
+喙
+喚
+喜
+喝
+喟
+喢
+喣
+喤
+喥
+喦
+喧
+喨
+喩
+喪
+喫
+喬
+喭
+單
+喰
+喱
+喲
+喳
+喵
+営
+喷
+喹
+喻
+喼
+喽
+喾
+喿
+嗀
+嗁
+嗃
+嗄
+嗅
+嗆
+嗇
+嗈
+嗉
+嗊
+嗋
+嗌
+嗍
+嗎
+嗏
+嗐
+嗑
+嗒
+嗓
+嗔
+嗕
+嗖
+嗚
+嗛
+嗜
+嗝
+嗞
+嗟
+嗡
+嗢
+嗣
+嗤
+嗥
+嗦
+嗧
+嗨
+嗩
+嗪
+嗫
+嗬
+嗮
+嗯
+嗲
+嗳
+嗵
+嗶
+嗷
+嗹
+嗺
+嗼
+嗽
+嗾
+嗿
+嘀
+嘁
+嘂
+嘄
+嘅
+嘆
+嘈
+嘉
+嘌
+嘍
+嘎
+嘏
+嘐
+嘑
+嘒
+嘓
+嘔
+嘕
+嘖
+嘗
+嘘
+嘚
+嘛
+嘜
+嘝
+嘞
+嘟
+嘠
+嘡
+嘢
+嘣
+嘤
+嘥
+嘧
+嘩
+嘬
+嘭
+嘮
+嘯
+嘰
+嘱
+嘲
+嘳
+嘴
+嘵
+嘶
+嘷
+嘸
+嘹
+嘻
+嘽
+嘾
+嘿
+噀
+噁
+噂
+噅
+噆
+噇
+噈
+噉
+噊
+噌
+噍
+噎
+噏
+噓
+噔
+噗
+噘
+噙
+噚
+噛
+噜
+噝
+噞
+噠
+噢
+噣
+噤
+噥
+噦
+器
+噩
+噪
+噫
+噬
+噭
+噮
+噯
+噰
+噱
+噲
+噳
+噴
+噶
+噸
+噹
+噺
+噻
+噼
+嚀
+嚃
+嚄
+嚅
+嚆
+嚇
+嚈
+嚌
+嚍
+嚎
+嚏
+嚐
+嚒
+嚓
+嚕
+嚗
+嚘
+嚙
+嚚
+嚜
+嚞
+嚟
+嚢
+嚣
+嚤
+嚥
+嚦
+嚧
+嚨
+嚩
+嚪
+嚫
+嚬
+嚭
+嚮
+嚯
+嚲
+嚳
+嚴
+嚵
+嚶
+嚷
+嚼
+嚽
+嚾
+嚿
+囀
+囁
+囂
+囃
+囅
+囆
+囈
+囉
+囊
+囋
+囌
+囍
+囑
+囒
+囓
+囔
+囗
+囙
+囚
+四
+囝
+回
+囟
+因
+囡
+团
+団
+囤
+囧
+囪
+囫
+园
+囮
+囯
+困
+囱
+囲
+図
+围
+囵
+囶
+囷
+囹
+固
+国
+图
+囿
+圀
+圁
+圂
+圃
+圄
+圅
+圆
+圇
+圈
+圉
+圊
+國
+圌
+圍
+圏
+圐
+園
+圓
+圖
+圗
+團
+圙
+圜
+圞
+土
+圠
+圢
+圣
+圧
+在
+圩
+圪
+圫
+圬
+圭
+圮
+圯
+地
+圲
+圳
+圹
+场
+圻
+圾
+址
+坁
+坂
+坅
+均
+坉
+坊
+坋
+坌
+坍
+坎
+坏
+坐
+坑
+坒
+坖
+块
+坚
+坛
+坜
+坝
+坞
+坟
+坠
+坡
+坣
+坤
+坥
+坦
+坨
+坩
+坪
+坫
+坬
+坭
+坮
+坯
+坰
+坱
+坲
+坳
+坴
+坵
+坶
+坷
+坻
+坼
+坽
+坿
+垀
+垂
+垃
+垄
+垆
+垈
+型
+垌
+垍
+垎
+垏
+垒
+垓
+垔
+垕
+垗
+垘
+垙
+垚
+垛
+垜
+垝
+垞
+垟
+垠
+垡
+垢
+垣
+垤
+垦
+垧
+垩
+垫
+垭
+垮
+垯
+垰
+垱
+垲
+垴
+垵
+垸
+垺
+垻
+垼
+垽
+垾
+垿
+埂
+埃
+埅
+埆
+埇
+埈
+埋
+埌
+城
+埏
+埒
+埔
+埕
+埗
+埘
+埙
+埚
+埜
+埝
+域
+埠
+埡
+埢
+埤
+埧
+埨
+埪
+埫
+埭
+埮
+埯
+埳
+埴
+埵
+埶
+執
+埸
+培
+基
+埻
+埼
+埽
+堀
+堂
+堃
+堅
+堆
+堇
+堈
+堉
+堊
+堋
+堌
+堍
+堎
+堐
+堑
+堔
+堕
+堖
+堘
+堙
+堝
+堞
+堠
+堡
+堣
+堤
+堥
+堦
+堧
+堨
+堩
+堪
+堮
+堯
+堰
+報
+堲
+場
+堵
+堷
+堺
+堼
+堽
+堾
+堿
+塀
+塁
+塄
+塅
+塆
+塈
+塉
+塊
+塋
+塌
+塍
+塏
+塑
+塒
+塓
+塔
+塕
+塗
+塘
+塙
+塚
+塜
+塝
+塞
+塟
+塡
+塢
+塤
+塥
+塨
+塩
+填
+塬
+塭
+塯
+塰
+塱
+塲
+塴
+塵
+塸
+塹
+塺
+塼
+塽
+塾
+塿
+墀
+墁
+境
+墅
+墆
+墈
+墉
+墊
+墋
+墎
+墐
+墒
+墓
+墕
+墖
+増
+墘
+墙
+墚
+墜
+墝
+增
+墟
+墠
+墡
+墣
+墦
+墨
+墩
+墪
+墫
+墬
+墮
+墯
+墱
+墳
+墶
+墺
+墻
+墼
+墽
+墾
+壁
+壄
+壅
+壆
+壇
+壈
+壊
+壋
+壌
+壎
+壐
+壑
+壓
+壔
+壕
+壖
+壘
+壙
+壚
+壛
+壝
+壞
+壟
+壠
+壢
+壤
+壩
+壪
+士
+壬
+壮
+壯
+声
+壱
+売
+壳
+壴
+壶
+壷
+壸
+壹
+壺
+壻
+壼
+壽
+壾
+壿
+夀
+夃
+处
+夆
+备
+変
+夊
+夋
+夌
+复
+夎
+夏
+夐
+夒
+夔
+夕
+外
+夗
+夘
+夙
+多
+夜
+够
+夠
+夢
+夤
+夥
+大
+天
+太
+夫
+夬
+夭
+央
+夯
+失
+头
+夷
+夸
+夹
+夺
+夼
+夾
+奀
+奁
+奂
+奄
+奅
+奇
+奈
+奉
+奊
+奋
+奎
+奏
+奐
+契
+奓
+奔
+奕
+奖
+套
+奘
+奚
+奠
+奡
+奢
+奣
+奥
+奧
+奨
+奩
+奪
+奫
+奬
+奭
+奮
+奰
+奲
+女
+奴
+奶
+奸
+她
+奼
+好
+妀
+妁
+如
+妃
+妄
+妆
+妇
+妈
+妊
+妍
+妎
+妏
+妐
+妑
+妒
+妓
+妖
+妗
+妘
+妙
+妝
+妞
+妠
+妢
+妣
+妤
+妥
+妦
+妧
+妨
+妩
+妪
+妫
+妬
+妭
+妮
+妯
+妲
+妳
+妵
+妷
+妸
+妹
+妺
+妻
+妾
+姀
+姁
+姃
+姆
+姈
+姉
+姊
+始
+姌
+姍
+姎
+姏
+姐
+姑
+姒
+姓
+委
+姗
+姘
+姙
+姚
+姜
+姝
+姞
+姡
+姣
+姤
+姥
+姦
+姨
+姪
+姫
+姬
+姮
+姱
+姵
+姶
+姸
+姹
+姺
+姻
+姼
+姽
+姿
+娀
+威
+娃
+娄
+娅
+娆
+娇
+娈
+娉
+娋
+娌
+娍
+娎
+娑
+娓
+娖
+娘
+娙
+娛
+娜
+娟
+娠
+娣
+娥
+娩
+娫
+娭
+娯
+娱
+娲
+娳
+娴
+娵
+娶
+娷
+娸
+娼
+娽
+娿
+婀
+婁
+婆
+婉
+婊
+婌
+婍
+婐
+婑
+婓
+婕
+婗
+婘
+婚
+婞
+婟
+婠
+婢
+婣
+婤
+婥
+婦
+婧
+婪
+婫
+婬
+婭
+婯
+婳
+婴
+婵
+婶
+婷
+婸
+婺
+婻
+婼
+婿
+媂
+媃
+媄
+媆
+媊
+媐
+媒
+媓
+媔
+媕
+媖
+媗
+媚
+媛
+媜
+媞
+媟
+媢
+媥
+媦
+媧
+媪
+媬
+媭
+媮
+媯
+媰
+媱
+媲
+媳
+媵
+媸
+媺
+媻
+媼
+媽
+媾
+媿
+嫁
+嫂
+嫄
+嫇
+嫈
+嫉
+嫋
+嫌
+嫐
+嫒
+嫔
+嫕
+嫖
+嫗
+嫘
+嫚
+嫛
+嫜
+嫠
+嫡
+嫢
+嫣
+嫥
+嫦
+嫩
+嫪
+嫫
+嫭
+嫮
+嫰
+嫱
+嫲
+嫳
+嫴
+嫵
+嫶
+嫷
+嫺
+嫻
+嫽
+嫿
+嬀
+嬃
+嬅
+嬈
+嬉
+嬋
+嬌
+嬎
+嬐
+嬓
+嬖
+嬗
+嬙
+嬛
+嬝
+嬡
+嬢
+嬤
+嬥
+嬨
+嬪
+嬬
+嬭
+嬰
+嬲
+嬴
+嬷
+嬸
+嬽
+嬾
+嬿
+孀
+孁
+孃
+孅
+孆
+孈
+孋
+孌
+子
+孑
+孓
+孔
+孕
+孖
+字
+存
+孙
+孚
+孛
+孜
+孝
+孟
+孢
+季
+孤
+孥
+学
+孩
+孪
+孫
+孬
+孮
+孰
+孱
+孲
+孳
+孵
+孷
+學
+孺
+孻
+孼
+孽
+孿
+宁
+宂
+它
+宄
+宅
+宇
+守
+安
+宋
+完
+宍
+宎
+宏
+宓
+宕
+宗
+官
+宙
+定
+宛
+宜
+宝
+实
+実
+宠
+审
+客
+宣
+室
+宥
+宦
+宧
+宨
+宪
+宫
+宬
+宮
+宰
+害
+宴
+宵
+家
+宸
+容
+宽
+宾
+宿
+寀
+寁
+寂
+寃
+寄
+寅
+密
+寇
+寈
+寊
+寋
+富
+寍
+寐
+寑
+寒
+寓
+寔
+寕
+寖
+寗
+寘
+寙
+寛
+寜
+寝
+寞
+察
+寠
+寡
+寢
+寤
+寥
+實
+寧
+寨
+審
+寪
+寫
+寬
+寮
+寯
+寰
+寱
+寳
+寵
+寶
+寸
+对
+寺
+寻
+导
+対
+寿
+封
+専
+尃
+射
+尅
+将
+將
+專
+尉
+尊
+尋
+尌
+對
+導
+小
+尐
+少
+尒
+尓
+尔
+尕
+尖
+尗
+尘
+尙
+尚
+尜
+尝
+尟
+尢
+尤
+尥
+尧
+尨
+尪
+尬
+尭
+尰
+就
+尳
+尴
+尷
+尸
+尹
+尺
+尻
+尼
+尽
+尾
+尿
+局
+屁
+层
+屃
+屄
+居
+屆
+屇
+屈
+屉
+届
+屋
+屌
+屍
+屎
+屏
+屐
+屑
+屓
+屔
+展
+屖
+屘
+屙
+屚
+屜
+屝
+属
+屠
+屡
+屢
+屣
+層
+履
+屦
+屧
+屨
+屩
+屬
+屭
+屮
+屯
+山
+屴
+屹
+屺
+屻
+屼
+屾
+屿
+岀
+岁
+岂
+岅
+岈
+岉
+岊
+岋
+岌
+岍
+岏
+岐
+岑
+岒
+岔
+岖
+岗
+岘
+岙
+岚
+岛
+岜
+岝
+岞
+岟
+岠
+岡
+岢
+岣
+岤
+岧
+岨
+岩
+岪
+岫
+岬
+岭
+岮
+岯
+岰
+岱
+岳
+岵
+岷
+岸
+岻
+岽
+岿
+峁
+峂
+峃
+峄
+峅
+峇
+峉
+峊
+峋
+峌
+峍
+峐
+峒
+峗
+峘
+峙
+峚
+峛
+峝
+峞
+峠
+峡
+峣
+峤
+峥
+峦
+峧
+峨
+峩
+峪
+峬
+峭
+峮
+峯
+峰
+峱
+峴
+島
+峷
+峻
+峼
+峽
+峿
+崀
+崁
+崂
+崃
+崄
+崆
+崇
+崈
+崋
+崌
+崍
+崎
+崏
+崐
+崑
+崒
+崔
+崖
+崗
+崘
+崙
+崚
+崛
+崝
+崞
+崟
+崠
+崡
+崢
+崣
+崤
+崥
+崦
+崧
+崨
+崩
+崬
+崭
+崮
+崰
+崱
+崳
+崴
+崵
+崶
+崷
+崺
+崽
+崾
+崿
+嵁
+嵂
+嵃
+嵅
+嵇
+嵊
+嵋
+嵌
+嵎
+嵐
+嵑
+嵒
+嵕
+嵖
+嵗
+嵘
+嵙
+嵚
+嵛
+嵜
+嵝
+嵞
+嵣
+嵥
+嵨
+嵩
+嵫
+嵬
+嵮
+嵯
+嵰
+嵱
+嵲
+嵴
+嵷
+嵺
+嵻
+嵽
+嵾
+嵿
+嶀
+嶁
+嶂
+嶃
+嶄
+嶅
+嶆
+嶇
+嶈
+嶉
+嶊
+嶋
+嶌
+嶍
+嶒
+嶓
+嶔
+嶕
+嶗
+嶙
+嶚
+嶜
+嶝
+嶞
+嶟
+嶠
+嶡
+嶢
+嶦
+嶧
+嶨
+嶩
+嶪
+嶬
+嶭
+嶮
+嶰
+嶱
+嶲
+嶴
+嶵
+嶷
+嶸
+嶺
+嶼
+嶽
+巀
+巂
+巃
+巄
+巅
+巆
+巇
+巉
+巋
+巌
+巍
+巎
+巏
+巑
+巒
+巔
+巕
+巖
+巗
+巘
+巛
+川
+州
+巟
+巡
+巢
+巣
+工
+左
+巧
+巨
+巩
+巫
+差
+巯
+巰
+己
+已
+巳
+巴
+巵
+巶
+巷
+巹
+巻
+巽
+巾
+巿
+帀
+币
+市
+布
+帅
+帆
+师
+帊
+希
+帏
+帐
+帑
+帔
+帕
+帖
+帗
+帘
+帙
+帚
+帛
+帜
+帝
+帟
+帠
+帡
+帢
+帣
+帤
+帥
+带
+帧
+帨
+帩
+師
+帬
+席
+帮
+帯
+帰
+帱
+帳
+帴
+帶
+帷
+常
+帻
+帼
+帽
+帾
+幀
+幂
+幃
+幄
+幅
+幇
+幌
+幎
+幏
+幑
+幓
+幔
+幕
+幖
+幗
+幘
+幙
+幚
+幛
+幝
+幞
+幟
+幠
+幡
+幢
+幣
+幦
+幧
+幨
+幩
+幪
+幫
+幬
+幭
+幰
+干
+平
+年
+幵
+并
+幷
+幸
+幹
+幺
+幻
+幼
+幽
+幾
+广
+庀
+庁
+庂
+広
+庄
+庆
+庇
+庈
+庉
+床
+庋
+庌
+序
+庐
+庑
+库
+应
+底
+庖
+店
+庙
+庚
+庛
+府
+庞
+废
+庠
+庢
+庣
+庤
+庥
+度
+座
+庨
+庪
+庫
+庬
+庭
+庮
+庰
+庱
+庲
+庳
+庴
+庵
+庶
+康
+庸
+庹
+庻
+庼
+庽
+庾
+廁
+廂
+廃
+廄
+廅
+廆
+廇
+廈
+廉
+廊
+廋
+廌
+廍
+廎
+廐
+廑
+廒
+廓
+廔
+廕
+廖
+廗
+廘
+廙
+廚
+廛
+廜
+廝
+廞
+廟
+廠
+廡
+廢
+廣
+廥
+廦
+廧
+廨
+廩
+廪
+廬
+廮
+廯
+廰
+廱
+廳
+廵
+延
+廷
+廸
+廹
+建
+廻
+廼
+廾
+廿
+开
+弁
+异
+弃
+弄
+弅
+弆
+弇
+弈
+弉
+弊
+弋
+弌
+弍
+式
+弐
+弑
+弒
+弓
+弔
+引
+弖
+弗
+弘
+弚
+弛
+弝
+弟
+张
+弢
+弣
+弤
+弥
+弦
+弧
+弨
+弩
+弭
+弮
+弯
+弱
+弳
+張
+弶
+強
+弸
+弹
+强
+弼
+弾
+彀
+彃
+彄
+彅
+彆
+彈
+彉
+彊
+彋
+彌
+彎
+彏
+彐
+归
+当
+彔
+录
+彖
+彗
+彘
+彙
+彜
+彝
+彞
+彟
+彠
+彡
+形
+彣
+彤
+彥
+彦
+彧
+彩
+彪
+彫
+彬
+彭
+彯
+彰
+影
+彳
+彴
+彷
+彸
+役
+彻
+彼
+彾
+彿
+往
+征
+徂
+徃
+径
+待
+徇
+很
+徉
+徊
+律
+後
+徐
+徑
+徒
+従
+徕
+得
+徘
+徙
+徛
+徜
+從
+徠
+御
+徥
+徦
+徧
+徨
+復
+循
+徫
+徬
+徭
+微
+徯
+徲
+徳
+徴
+徵
+德
+徸
+徹
+徼
+徽
+徾
+徿
+忀
+忁
+心
+必
+忆
+忉
+忌
+忍
+忏
+忐
+忑
+忒
+忔
+忕
+忖
+志
+忘
+忙
+応
+忝
+忞
+忠
+忡
+忣
+忤
+忥
+忧
+忨
+忪
+快
+忬
+忭
+忮
+忯
+忱
+忳
+念
+忷
+忸
+忺
+忻
+忼
+忽
+忾
+忿
+怀
+态
+怂
+怃
+怄
+怅
+怆
+怊
+怋
+怍
+怎
+怏
+怐
+怒
+怓
+怔
+怕
+怖
+怗
+怙
+怚
+怛
+怜
+思
+怞
+怠
+怡
+怢
+急
+怦
+性
+怨
+怩
+怪
+怫
+怬
+怭
+怮
+怯
+怱
+怲
+怳
+怴
+怵
+怷
+怹
+总
+怼
+怿
+恀
+恁
+恂
+恃
+恅
+恆
+恇
+恉
+恊
+恋
+恌
+恍
+恐
+恒
+恓
+恔
+恕
+恙
+恚
+恛
+恝
+恞
+恟
+恠
+恡
+恢
+恣
+恤
+恥
+恧
+恨
+恩
+恪
+恫
+恬
+恭
+息
+恰
+恲
+恳
+恵
+恶
+恸
+恹
+恺
+恻
+恼
+恽
+恿
+悀
+悁
+悃
+悄
+悅
+悆
+悇
+悈
+悉
+悊
+悌
+悍
+悐
+悒
+悔
+悕
+悖
+悗
+悚
+悛
+悝
+悞
+悟
+悠
+悢
+患
+悤
+悦
+悧
+您
+悩
+悪
+悫
+悬
+悭
+悯
+悰
+悱
+悲
+悳
+悴
+悵
+悶
+悷
+悸
+悹
+悺
+悻
+悼
+悽
+悾
+惄
+情
+惆
+惇
+惈
+惉
+惊
+惋
+惌
+惎
+惏
+惑
+惓
+惔
+惕
+惘
+惙
+惚
+惛
+惜
+惝
+惟
+惠
+惡
+惢
+惣
+惤
+惥
+惦
+惧
+惨
+惩
+惪
+惫
+惬
+惭
+惮
+惯
+惰
+惱
+惲
+想
+惴
+惵
+惶
+惷
+惸
+惹
+惺
+惻
+惼
+惾
+愀
+愁
+愃
+愅
+愆
+愈
+愉
+愊
+愍
+愎
+意
+愐
+愒
+愓
+愔
+愕
+愖
+愚
+愛
+愜
+感
+愠
+愣
+愤
+愦
+愧
+愨
+愫
+愬
+愭
+愮
+愯
+愲
+愴
+愶
+愷
+愼
+愽
+愾
+愿
+慁
+慂
+慄
+慅
+慆
+慇
+慈
+慉
+慊
+態
+慌
+慍
+慎
+慑
+慒
+慓
+慔
+慕
+慘
+慚
+慛
+慜
+慝
+慞
+慟
+慢
+慣
+慤
+慥
+慧
+慨
+慪
+慫
+慬
+慭
+慮
+慰
+慱
+慲
+慳
+慴
+慵
+慶
+慷
+慹
+慺
+慼
+慽
+慾
+憀
+憂
+憃
+憇
+憉
+憊
+憋
+憍
+憎
+憐
+憑
+憒
+憓
+憔
+憕
+憖
+憙
+憚
+憛
+憝
+憟
+憡
+憢
+憤
+憧
+憨
+憩
+憪
+憫
+憬
+憭
+憮
+憯
+憰
+憱
+憲
+憴
+憵
+憶
+憷
+憸
+憺
+憻
+憼
+憾
+憿
+懁
+懂
+懃
+懅
+懆
+懇
+懈
+應
+懊
+懋
+懌
+懍
+懐
+懑
+懒
+懔
+懘
+懞
+懟
+懠
+懣
+懤
+懥
+懦
+懧
+懨
+懪
+懫
+懭
+懮
+懰
+懱
+懲
+懵
+懶
+懷
+懸
+懺
+懻
+懼
+懽
+懾
+懿
+戀
+戁
+戃
+戄
+戆
+戇
+戈
+戉
+戊
+戋
+戌
+戍
+戎
+戏
+成
+我
+戒
+戔
+戕
+或
+戗
+战
+戙
+戚
+戛
+戞
+戟
+戠
+戡
+戢
+戣
+戤
+戥
+戦
+戧
+戩
+截
+戬
+戭
+戮
+戯
+戰
+戱
+戲
+戳
+戴
+戶
+户
+戸
+戹
+戺
+戻
+戼
+戽
+戾
+房
+所
+扁
+扂
+扃
+扅
+扆
+扇
+扈
+扉
+扊
+手
+才
+扎
+扐
+扑
+扒
+打
+扔
+払
+托
+扙
+扛
+扜
+扞
+扡
+扢
+扣
+扤
+扥
+扦
+执
+扩
+扪
+扫
+扬
+扭
+扮
+扯
+扰
+扱
+扳
+扶
+批
+扺
+扻
+扼
+扽
+找
+承
+技
+抃
+抄
+抆
+抇
+抉
+把
+抌
+抎
+抏
+抑
+抒
+抓
+抔
+投
+抖
+抗
+折
+抚
+抛
+抜
+抝
+択
+抟
+抠
+抡
+抢
+护
+报
+抦
+抨
+抪
+披
+抬
+抭
+抮
+抱
+抴
+抵
+抶
+抹
+抻
+押
+抽
+抾
+抿
+拂
+拃
+拄
+担
+拆
+拇
+拈
+拉
+拊
+拋
+拌
+拍
+拎
+拏
+拐
+拑
+拒
+拓
+拔
+拕
+拖
+拗
+拘
+拙
+拚
+招
+拜
+拝
+拟
+拠
+拡
+拢
+拣
+拤
+拥
+拦
+拧
+拨
+择
+拫
+括
+拭
+拮
+拯
+拱
+拲
+拳
+拴
+拵
+拶
+拷
+拸
+拹
+拺
+拼
+拽
+拾
+拿
+挀
+持
+挂
+挃
+指
+挈
+按
+挋
+挌
+挍
+挎
+挏
+挐
+挑
+挓
+挖
+挙
+挚
+挛
+挝
+挞
+挟
+挠
+挡
+挣
+挤
+挥
+挦
+挨
+挩
+挪
+挫
+挬
+振
+挱
+挲
+挴
+挵
+挶
+挹
+挺
+挻
+挼
+挽
+挾
+挿
+捁
+捂
+捃
+捄
+捅
+捆
+捉
+捊
+捋
+捌
+捍
+捎
+捏
+捐
+捒
+捔
+捕
+捖
+捗
+捘
+捜
+捞
+损
+捡
+换
+捣
+捥
+捦
+捧
+捨
+捩
+捫
+捭
+据
+捯
+捱
+捲
+捵
+捶
+捷
+捺
+捻
+捼
+捽
+掀
+掂
+掃
+掄
+掇
+授
+掉
+掊
+掌
+掎
+掏
+掐
+排
+掔
+掖
+掗
+掘
+掙
+掛
+掜
+掝
+掞
+掟
+掠
+採
+探
+掣
+掤
+接
+控
+推
+掩
+措
+掫
+掬
+掭
+掮
+掯
+掰
+掱
+掲
+掳
+掴
+掷
+掸
+掺
+掻
+掼
+掽
+掾
+揀
+揃
+揄
+揅
+揆
+揉
+揊
+揍
+揎
+描
+提
+揑
+插
+揔
+揕
+揖
+揘
+揚
+換
+揜
+揝
+揟
+揠
+握
+揣
+揤
+揥
+揨
+揩
+揪
+揫
+揭
+揮
+揯
+揱
+揲
+揳
+援
+揵
+揶
+揷
+揸
+揹
+揺
+揽
+揿
+搀
+搁
+搂
+搅
+搆
+搉
+搊
+搋
+搌
+損
+搏
+搐
+搒
+搓
+搔
+搕
+搖
+搗
+搘
+搚
+搛
+搜
+搞
+搟
+搠
+搡
+搢
+搣
+搤
+搥
+搦
+搧
+搨
+搪
+搫
+搬
+搭
+搯
+搰
+搳
+搴
+搵
+搶
+搷
+搹
+携
+搽
+搾
+摀
+摁
+摂
+摃
+摄
+摅
+摆
+摇
+摈
+摊
+摋
+摍
+摎
+摏
+摐
+摑
+摒
+摓
+摔
+摘
+摛
+摜
+摝
+摞
+摟
+摠
+摡
+摦
+摧
+摩
+摭
+摮
+摯
+摰
+摲
+摳
+摴
+摵
+摶
+摸
+摹
+摺
+摻
+摽
+摿
+撂
+撃
+撄
+撅
+撇
+撈
+撊
+撋
+撌
+撏
+撐
+撑
+撒
+撓
+撕
+撖
+撗
+撙
+撚
+撜
+撝
+撞
+撟
+撠
+撢
+撣
+撤
+撥
+撦
+撩
+撫
+撬
+播
+撮
+撰
+撲
+撳
+撵
+撷
+撸
+撹
+撺
+撻
+撼
+撽
+撾
+撿
+擀
+擁
+擂
+擄
+擅
+擇
+擉
+擊
+擋
+操
+擎
+擏
+擐
+擒
+擓
+擔
+擕
+擖
+擗
+擘
+據
+擞
+擠
+擡
+擢
+擣
+擤
+擥
+擦
+擧
+擨
+擩
+擫
+擬
+擭
+擯
+擰
+擱
+擲
+擴
+擷
+擸
+擺
+擻
+擼
+擽
+擾
+擿
+攀
+攄
+攆
+攇
+攉
+攌
+攍
+攏
+攐
+攒
+攓
+攔
+攕
+攖
+攗
+攘
+攙
+攛
+攜
+攝
+攞
+攠
+攡
+攢
+攣
+攤
+攥
+攦
+攩
+攪
+攫
+攬
+攭
+攮
+支
+攲
+攳
+攴
+收
+攷
+攸
+改
+攻
+攽
+放
+政
+敁
+敂
+故
+敆
+效
+敉
+敊
+敌
+敍
+敎
+敏
+救
+敓
+敔
+敕
+敖
+敗
+敘
+教
+敛
+敜
+敝
+敞
+敟
+敢
+散
+敤
+敦
+敧
+敩
+敪
+敫
+敬
+敭
+敯
+数
+敲
+敳
+整
+敵
+敶
+敷
+數
+敹
+敺
+敻
+敾
+敿
+斀
+斁
+斂
+斃
+斄
+文
+斉
+斋
+斌
+斎
+斐
+斑
+斒
+斓
+斔
+斕
+斖
+斗
+料
+斛
+斜
+斝
+斞
+斟
+斠
+斡
+斤
+斥
+斧
+斨
+斩
+斪
+斫
+斬
+断
+斮
+斯
+新
+斲
+斵
+斶
+斷
+方
+斻
+於
+施
+斿
+旁
+旂
+旃
+旄
+旅
+旆
+旉
+旋
+旌
+旍
+旎
+族
+旐
+旒
+旓
+旖
+旗
+旙
+旚
+旛
+旝
+旞
+旟
+无
+旡
+既
+日
+旦
+旧
+旨
+早
+旬
+旭
+旮
+旯
+旰
+旱
+旲
+旳
+旴
+旵
+时
+旷
+旸
+旺
+旻
+旼
+旽
+旾
+旿
+昀
+昂
+昃
+昄
+昆
+昇
+昈
+昉
+昊
+昌
+昍
+明
+昏
+昐
+昒
+易
+昔
+昕
+昙
+昚
+昛
+昜
+昝
+昞
+星
+映
+昡
+昢
+昣
+昤
+春
+昦
+昧
+昨
+昪
+昫
+昬
+昭
+是
+昰
+昱
+昲
+昳
+昴
+昵
+昶
+昺
+昼
+昽
+显
+晁
+時
+晃
+晅
+晇
+晈
+晉
+晊
+晋
+晌
+晏
+晐
+晒
+晓
+晔
+晕
+晖
+晗
+晙
+晚
+晛
+晜
+晝
+晞
+晟
+晡
+晢
+晤
+晥
+晦
+晧
+晨
+晩
+晪
+晫
+晬
+晭
+普
+景
+晰
+晱
+晳
+晴
+晶
+晷
+晸
+智
+晻
+晼
+晾
+暀
+暁
+暂
+暃
+暄
+暅
+暆
+暇
+暈
+暉
+暊
+暋
+暌
+暍
+暎
+暏
+暐
+暑
+暕
+暖
+暗
+暘
+暝
+暟
+暠
+暡
+暢
+暦
+暧
+暨
+暩
+暫
+暮
+暰
+暱
+暲
+暴
+暵
+暶
+暸
+暹
+暻
+暾
+暿
+曀
+曄
+曅
+曆
+曇
+曈
+曉
+曊
+曋
+曌
+曏
+曒
+曔
+曖
+曙
+曚
+曛
+曜
+曝
+曠
+曡
+曣
+曦
+曧
+曨
+曩
+曬
+曭
+曮
+曰
+曲
+曳
+更
+曶
+曷
+書
+曹
+曺
+曼
+曽
+曾
+替
+最
+朁
+會
+朄
+朅
+月
+有
+朊
+朋
+服
+朏
+朐
+朒
+朓
+朔
+朕
+朖
+朗
+朘
+望
+朝
+朞
+期
+朡
+朢
+朣
+朦
+朧
+木
+未
+末
+本
+札
+朮
+术
+朱
+朳
+朴
+朵
+朶
+朸
+朹
+机
+朻
+朼
+朽
+朾
+朿
+杀
+杁
+杂
+权
+杄
+杅
+杆
+杇
+杈
+杉
+杋
+杌
+李
+杏
+材
+村
+杓
+杕
+杖
+杗
+杙
+杜
+杝
+杞
+束
+杠
+条
+杢
+杣
+杤
+来
+杧
+杨
+杩
+杪
+杬
+杭
+杮
+杯
+杰
+東
+杲
+杳
+杴
+杵
+杶
+杷
+杻
+杼
+松
+板
+极
+构
+枅
+枆
+枇
+枉
+枋
+枌
+枍
+枎
+枏
+析
+枒
+枓
+枕
+枖
+林
+枘
+枚
+枛
+果
+枝
+枞
+枠
+枡
+枢
+枣
+枥
+枧
+枨
+枪
+枫
+枭
+枯
+枰
+枱
+枲
+枳
+枴
+枵
+架
+枷
+枸
+枹
+枺
+枻
+枼
+柀
+柁
+柂
+柃
+柄
+柅
+柈
+柉
+柊
+柍
+柎
+柏
+某
+柑
+柒
+染
+柔
+柖
+柘
+柙
+柚
+柜
+柝
+柞
+柟
+柠
+柢
+柣
+柤
+查
+柧
+柩
+柫
+柬
+柮
+柯
+柰
+柱
+柲
+柳
+柴
+柵
+柶
+柷
+柸
+柹
+柺
+査
+柽
+柾
+柿
+栀
+栁
+栂
+栃
+栄
+栅
+标
+栈
+栉
+栊
+栋
+栌
+栎
+栏
+栐
+树
+栒
+栓
+栔
+栖
+栗
+栘
+栚
+栜
+栝
+栞
+栟
+校
+栢
+栨
+栩
+株
+栫
+栭
+栯
+栰
+栱
+栲
+栳
+栴
+栵
+样
+核
+根
+栻
+格
+栽
+栾
+桀
+桁
+桂
+桃
+桄
+桅
+框
+案
+桉
+桊
+桋
+桌
+桎
+桏
+桐
+桑
+桒
+桓
+桔
+桕
+桖
+桙
+桜
+桝
+桟
+桠
+桡
+桢
+档
+桤
+桥
+桦
+桧
+桨
+桩
+桫
+桭
+桮
+桯
+桱
+桲
+桴
+桵
+桶
+桷
+桹
+桺
+桼
+桾
+桿
+梀
+梁
+梂
+梃
+梅
+梆
+梇
+梉
+梌
+梏
+梐
+梒
+梓
+梔
+梗
+梘
+梛
+梜
+條
+梟
+梠
+梡
+梢
+梣
+梦
+梧
+梨
+梩
+梫
+梬
+梭
+梮
+梯
+械
+梱
+梲
+梳
+梴
+梵
+梶
+梼
+梽
+梾
+梿
+检
+棁
+棂
+棃
+棄
+棆
+棉
+棊
+棋
+棌
+棍
+棐
+棒
+棓
+棕
+棖
+棗
+棘
+棚
+棛
+棜
+棟
+棠
+棡
+棣
+棤
+棧
+棨
+棩
+棪
+棫
+棬
+森
+棯
+棰
+棱
+棲
+棳
+棵
+棶
+棷
+棸
+棹
+棺
+棻
+棼
+棽
+椀
+椁
+椄
+椅
+椆
+椇
+椈
+椉
+椋
+椌
+植
+椎
+椏
+椐
+椑
+椒
+椓
+椔
+椗
+椙
+検
+椟
+椠
+椤
+椥
+椪
+椭
+椰
+椲
+椳
+椴
+椵
+椶
+椸
+椹
+椽
+椿
+楀
+楂
+楄
+楅
+楈
+楉
+楊
+楎
+楒
+楓
+楔
+楕
+楖
+楗
+楘
+楙
+楚
+楛
+楝
+楞
+楟
+楠
+楡
+楢
+楣
+楤
+楥
+楦
+楧
+楨
+楩
+楪
+楫
+楬
+業
+楮
+楯
+楰
+楱
+楳
+極
+楶
+楷
+楸
+楹
+楺
+楼
+楽
+楿
+概
+榃
+榄
+榅
+榆
+榇
+榈
+榉
+榊
+榍
+榎
+榑
+榔
+榕
+榖
+榗
+榘
+榙
+榛
+榜
+榞
+榠
+榢
+榣
+榤
+榦
+榧
+榨
+榩
+榪
+榫
+榬
+榭
+榮
+榯
+榰
+榱
+榲
+榴
+榶
+榷
+榹
+榻
+榼
+榽
+榾
+榿
+槀
+槁
+槃
+槄
+槇
+槉
+槊
+構
+槌
+槍
+槎
+槐
+槑
+槓
+槔
+槕
+槗
+様
+槙
+槚
+槛
+槜
+槟
+槠
+槢
+槤
+槥
+槧
+槨
+槩
+槬
+槭
+槮
+槰
+槱
+槲
+槳
+槷
+槸
+槺
+槻
+槼
+槽
+槾
+槿
+樀
+樁
+樂
+樅
+樆
+樊
+樋
+樑
+樓
+樔
+樕
+樗
+樘
+標
+樛
+樝
+樞
+樟
+樠
+模
+樣
+樧
+樨
+権
+横
+樫
+樯
+樱
+樲
+樴
+樵
+樸
+樹
+樺
+樻
+樽
+樾
+樿
+橀
+橁
+橄
+橇
+橈
+橉
+橋
+橐
+橑
+橒
+橓
+橘
+橙
+橚
+橛
+橜
+橝
+橞
+機
+橡
+橢
+橤
+橥
+橦
+橧
+橨
+橪
+橫
+橭
+橱
+橶
+橹
+橼
+橾
+橿
+檀
+檁
+檃
+檄
+檇
+檉
+檊
+檍
+檎
+檐
+檑
+檓
+檔
+檕
+檖
+檗
+檛
+檜
+檝
+檞
+檟
+檠
+檡
+檢
+檣
+檥
+檦
+檨
+檩
+檫
+檬
+檮
+檯
+檳
+檵
+檸
+檹
+檻
+檽
+櫂
+櫃
+櫅
+櫆
+櫈
+櫋
+櫌
+櫐
+櫑
+櫓
+櫙
+櫚
+櫛
+櫜
+櫝
+櫞
+櫟
+櫠
+櫡
+櫥
+櫧
+櫨
+櫪
+櫫
+櫬
+櫰
+櫱
+櫳
+櫸
+櫹
+櫺
+櫻
+櫼
+櫾
+櫿
+欂
+欃
+欄
+欅
+欈
+欉
+權
+欋
+欏
+欐
+欑
+欒
+欖
+欘
+欞
+欠
+次
+欢
+欣
+欤
+欥
+欧
+欨
+欬
+欭
+欯
+欱
+欲
+欳
+欴
+欵
+欶
+欷
+欸
+欹
+欺
+欻
+欽
+款
+欿
+歂
+歃
+歅
+歆
+歇
+歈
+歉
+歊
+歋
+歌
+歍
+歎
+歐
+歓
+歔
+歕
+歙
+歛
+歜
+歟
+歠
+歡
+止
+正
+此
+步
+武
+歧
+歩
+歪
+歭
+歯
+歲
+歳
+歴
+歶
+歷
+歸
+歹
+死
+歼
+歾
+歿
+殀
+殁
+殂
+殃
+殄
+殆
+殇
+殈
+殉
+殊
+残
+殍
+殏
+殑
+殒
+殓
+殔
+殖
+殗
+殘
+殙
+殚
+殛
+殞
+殟
+殠
+殡
+殢
+殣
+殤
+殥
+殦
+殧
+殪
+殫
+殭
+殮
+殯
+殰
+殲
+殳
+殴
+段
+殶
+殷
+殺
+殻
+殼
+殽
+殿
+毀
+毁
+毂
+毃
+毄
+毅
+毆
+毇
+毈
+毉
+毊
+毋
+毌
+母
+毎
+每
+毐
+毒
+毓
+比
+毕
+毖
+毗
+毘
+毙
+毚
+毛
+毞
+毡
+毣
+毤
+毦
+毨
+毪
+毫
+毬
+毯
+毰
+毲
+毳
+毴
+毵
+毸
+毹
+毻
+毼
+毽
+毾
+毿
+氀
+氁
+氂
+氄
+氅
+氆
+氇
+氈
+氊
+氌
+氍
+氏
+氐
+民
+氓
+气
+氕
+氖
+気
+氘
+氙
+氚
+氛
+氟
+氠
+氡
+氢
+氣
+氤
+氦
+氧
+氨
+氩
+氪
+氫
+氬
+氮
+氯
+氰
+氲
+氳
+水
+氶
+氷
+永
+氹
+氻
+氽
+氾
+氿
+汀
+汁
+求
+汃
+汆
+汇
+汈
+汉
+汊
+汋
+汍
+汎
+汏
+汐
+汒
+汔
+汕
+汗
+汙
+汚
+汛
+汜
+汝
+汞
+江
+池
+污
+汤
+汦
+汧
+汨
+汩
+汪
+汫
+汭
+汯
+汰
+汱
+汲
+汳
+汴
+汶
+汸
+汹
+決
+汽
+汾
+沁
+沂
+沃
+沄
+沅
+沆
+沇
+沈
+沉
+沋
+沌
+沍
+沏
+沐
+沒
+沓
+沔
+沕
+沖
+沘
+沙
+沚
+沛
+沜
+沟
+没
+沢
+沣
+沤
+沥
+沦
+沧
+沨
+沩
+沪
+沫
+沬
+沭
+沮
+沯
+沱
+河
+沴
+沶
+沷
+沸
+油
+沺
+治
+沼
+沽
+沾
+沿
+況
+泂
+泃
+泄
+泅
+泆
+泇
+泉
+泊
+泌
+泐
+泒
+泓
+泔
+法
+泖
+泗
+泙
+泚
+泛
+泜
+泝
+泞
+泠
+泡
+波
+泣
+泥
+泧
+注
+泩
+泪
+泫
+泬
+泭
+泮
+泯
+泰
+泱
+泲
+泳
+泵
+泷
+泸
+泺
+泻
+泼
+泽
+泾
+洀
+洁
+洄
+洇
+洈
+洊
+洋
+洌
+洍
+洎
+洑
+洒
+洓
+洗
+洘
+洙
+洚
+洛
+洞
+洟
+洢
+洣
+洤
+津
+洧
+洨
+洩
+洪
+洫
+洭
+洮
+洱
+洲
+洳
+洴
+洵
+洶
+洸
+洹
+洺
+活
+洼
+洽
+派
+洿
+流
+浃
+浄
+浅
+浆
+浇
+浈
+浉
+浊
+测
+浍
+济
+浏
+浐
+浑
+浒
+浓
+浔
+浕
+浘
+浙
+浚
+浛
+浜
+浞
+浟
+浠
+浡
+浢
+浣
+浤
+浥
+浦
+浧
+浩
+浪
+浬
+浭
+浮
+浯
+浰
+浲
+浴
+浵
+浶
+海
+浸
+浹
+浺
+浼
+浿
+涂
+涄
+涅
+涆
+涇
+消
+涉
+涊
+涌
+涍
+涎
+涐
+涑
+涒
+涓
+涔
+涕
+涖
+涗
+涘
+涙
+涛
+涝
+涞
+涟
+涠
+涡
+涢
+涣
+涤
+润
+涧
+涨
+涩
+涪
+涫
+涬
+涮
+涯
+液
+涳
+涴
+涵
+涷
+涸
+涼
+涽
+涾
+涿
+淀
+淄
+淅
+淆
+淇
+淈
+淊
+淋
+淌
+淍
+淎
+淏
+淑
+淒
+淓
+淔
+淖
+淘
+淙
+淚
+淛
+淜
+淝
+淞
+淟
+淠
+淡
+淢
+淤
+淥
+淦
+淨
+淩
+淪
+淫
+淬
+淭
+淮
+淯
+淰
+深
+淳
+淴
+淵
+淶
+混
+淸
+淹
+淺
+添
+淼
+淽
+渀
+渃
+清
+渇
+済
+渉
+渊
+渋
+渌
+渍
+渎
+渐
+渑
+渓
+渔
+渕
+渗
+渙
+渚
+減
+渜
+渝
+渟
+渠
+渡
+渢
+渣
+渤
+渥
+渦
+渨
+温
+渫
+測
+渭
+渮
+港
+渰
+渱
+渲
+渳
+渴
+游
+渹
+渺
+渻
+渼
+渽
+渾
+渿
+湀
+湁
+湃
+湄
+湅
+湆
+湇
+湉
+湊
+湋
+湍
+湎
+湑
+湓
+湔
+湖
+湘
+湛
+湜
+湝
+湞
+湟
+湠
+湡
+湢
+湣
+湥
+湧
+湨
+湩
+湫
+湮
+湯
+湱
+湲
+湳
+湴
+湻
+湼
+湾
+湿
+満
+溁
+溃
+溅
+溆
+溇
+溈
+溉
+溋
+溍
+溎
+溏
+源
+溓
+溔
+準
+溘
+溙
+溚
+溛
+溜
+溝
+溞
+溟
+溠
+溢
+溥
+溦
+溧
+溪
+溫
+溮
+溯
+溰
+溱
+溲
+溳
+溴
+溵
+溶
+溷
+溹
+溺
+溻
+溼
+溽
+溾
+滀
+滁
+滂
+滃
+滄
+滅
+滆
+滇
+滈
+滉
+滋
+滌
+滍
+滎
+滏
+滑
+滒
+滓
+滔
+滕
+滖
+滗
+滘
+滙
+滚
+滛
+滜
+滝
+滞
+滟
+滠
+满
+滢
+滤
+滥
+滦
+滧
+滨
+滩
+滪
+滫
+滬
+滭
+滮
+滯
+滱
+滲
+滴
+滵
+滷
+滸
+滹
+滻
+滽
+滾
+滿
+漁
+漂
+漃
+漅
+漆
+漇
+漈
+漉
+漊
+漋
+漎
+漏
+漑
+漒
+漓
+演
+漕
+漖
+漘
+漙
+漚
+漠
+漢
+漣
+漤
+漥
+漦
+漧
+漩
+漪
+漫
+漬
+漭
+漮
+漯
+漰
+漱
+漲
+漳
+漴
+漵
+漶
+漷
+漸
+漹
+漻
+漼
+漾
+漿
+潀
+潁
+潃
+潄
+潅
+潆
+潇
+潋
+潍
+潎
+潏
+潐
+潑
+潒
+潔
+潕
+潖
+潗
+潘
+潙
+潚
+潛
+潜
+潝
+潞
+潟
+潠
+潢
+潤
+潦
+潩
+潬
+潭
+潮
+潯
+潰
+潲
+潳
+潴
+潵
+潶
+潷
+潸
+潺
+潼
+潽
+潾
+潿
+澀
+澁
+澂
+澄
+澅
+澆
+澇
+澈
+澉
+澋
+澌
+澍
+澎
+澐
+澒
+澓
+澔
+澖
+澗
+澛
+澜
+澞
+澠
+澡
+澣
+澤
+澥
+澦
+澧
+澨
+澩
+澪
+澫
+澭
+澮
+澯
+澱
+澳
+澴
+澶
+澹
+澻
+澼
+澽
+激
+濁
+濂
+濃
+濄
+濆
+濇
+濈
+濉
+濊
+濋
+濌
+濎
+濑
+濒
+濔
+濕
+濘
+濙
+濛
+濜
+濞
+濟
+濠
+濡
+濣
+濤
+濦
+濧
+濩
+濫
+濬
+濭
+濮
+濯
+濰
+濱
+濲
+濴
+濶
+濺
+濻
+濼
+濾
+濿
+瀀
+瀁
+瀄
+瀅
+瀆
+瀇
+瀉
+瀋
+瀌
+瀍
+瀎
+瀏
+瀑
+瀔
+瀕
+瀖
+瀘
+瀙
+瀚
+瀛
+瀝
+瀞
+瀟
+瀠
+瀡
+瀢
+瀣
+瀦
+瀧
+瀨
+瀩
+瀫
+瀬
+瀯
+瀰
+瀱
+瀲
+瀴
+瀵
+瀶
+瀷
+瀸
+瀹
+瀺
+瀼
+瀾
+瀿
+灂
+灃
+灄
+灈
+灉
+灊
+灋
+灌
+灏
+灑
+灕
+灖
+灗
+灘
+灛
+灝
+灞
+灟
+灡
+灣
+灤
+灧
+灨
+灩
+灪
+火
+灭
+灯
+灰
+灴
+灵
+灶
+灸
+灼
+災
+灾
+灿
+炀
+炁
+炅
+炆
+炉
+炊
+炌
+炎
+炒
+炔
+炕
+炖
+炘
+炙
+炜
+炝
+炟
+炣
+炤
+炫
+炬
+炭
+炮
+炯
+炰
+炱
+炲
+炳
+炷
+炸
+点
+為
+炻
+炼
+炽
+炾
+烀
+烁
+烂
+烃
+烈
+烉
+烊
+烋
+烏
+烑
+烒
+烓
+烔
+烖
+烘
+烙
+烛
+烜
+烝
+烟
+烠
+烤
+烦
+烧
+烨
+烩
+烫
+烬
+热
+烯
+烰
+烱
+烴
+烶
+烷
+烹
+烺
+烻
+烽
+焃
+焄
+焆
+焉
+焊
+焌
+焍
+焐
+焓
+焔
+焕
+焖
+焗
+焘
+焙
+焚
+焜
+焞
+焠
+無
+焣
+焦
+焮
+焯
+焰
+焱
+焴
+然
+焻
+焼
+焿
+煁
+煃
+煅
+煆
+煇
+煉
+煊
+煋
+煌
+煎
+煐
+煑
+煒
+煓
+煔
+煕
+煖
+煗
+煙
+煚
+煜
+煝
+煞
+煟
+煠
+煢
+煣
+煤
+煥
+煦
+照
+煨
+煩
+煬
+煮
+煲
+煳
+煴
+煵
+煶
+煸
+煺
+煽
+熀
+熂
+熄
+熅
+熇
+熈
+熉
+熊
+熏
+熐
+熒
+熔
+熖
+熗
+熘
+熙
+熚
+熛
+熜
+熟
+熠
+熤
+熥
+熨
+熬
+熯
+熰
+熱
+熲
+熳
+熵
+熸
+熹
+熺
+熻
+熼
+熾
+熿
+燀
+燁
+燂
+燃
+燄
+燅
+燈
+燉
+燊
+燋
+燎
+燏
+燐
+燒
+燔
+燕
+燖
+燗
+燘
+燙
+燚
+燜
+燝
+營
+燠
+燡
+燥
+燦
+燧
+燫
+燬
+燭
+燮
+燴
+燹
+燻
+燼
+燾
+燿
+爀
+爁
+爂
+爆
+爇
+爊
+爌
+爍
+爐
+爓
+爔
+爕
+爚
+爛
+爝
+爞
+爟
+爢
+爣
+爨
+爪
+爬
+爭
+爯
+爰
+爱
+爲
+爵
+父
+爷
+爸
+爹
+爺
+爻
+爼
+爽
+爾
+爿
+牀
+牁
+牂
+牄
+牆
+片
+版
+牉
+牋
+牌
+牍
+牎
+牏
+牒
+牓
+牕
+牖
+牘
+牙
+牚
+牛
+牝
+牟
+牠
+牡
+牢
+牣
+牤
+牥
+牦
+牧
+物
+牬
+牮
+牯
+牰
+牲
+牴
+牵
+牷
+牸
+特
+牺
+牻
+牼
+牽
+牾
+牿
+犀
+犁
+犂
+犄
+犅
+犆
+犇
+犈
+犉
+犊
+犋
+犌
+犍
+犎
+犏
+犐
+犑
+犒
+犓
+犕
+犖
+犗
+犘
+犚
+犛
+犝
+犞
+犟
+犠
+犢
+犣
+犤
+犥
+犦
+犧
+犨
+犩
+犪
+犬
+犮
+犯
+犰
+犴
+犵
+状
+犷
+犸
+犹
+犽
+犿
+狀
+狁
+狂
+狃
+狄
+狆
+狈
+狉
+狊
+狋
+狌
+狍
+狎
+狐
+狑
+狒
+狓
+狔
+狖
+狗
+狘
+狙
+狚
+狛
+狜
+狝
+狞
+狟
+狠
+狡
+狣
+狤
+狥
+狦
+狨
+狩
+狪
+狫
+独
+狭
+狮
+狯
+狰
+狱
+狲
+狳
+狴
+狶
+狷
+狸
+狹
+狺
+狻
+狼
+狽
+狾
+狿
+猀
+猁
+猂
+猃
+猄
+猇
+猈
+猊
+猋
+猎
+猏
+猑
+猒
+猓
+猕
+猖
+猗
+猘
+猙
+猛
+猜
+猝
+猞
+猟
+猡
+猢
+猣
+猥
+猦
+猧
+猨
+猩
+猪
+猫
+猬
+猭
+献
+猯
+猰
+猱
+猲
+猳
+猴
+猵
+猶
+猷
+猹
+猺
+猻
+猼
+猾
+猿
+獀
+獁
+獂
+獃
+獄
+獅
+獇
+獊
+獋
+獌
+獍
+獎
+獏
+獐
+獑
+獒
+獗
+獘
+獛
+獝
+獞
+獟
+獠
+獡
+獢
+獣
+獥
+獦
+獧
+獨
+獩
+獪
+獫
+獬
+獭
+獮
+獯
+獰
+獲
+獳
+獴
+獵
+獶
+獷
+獸
+獺
+獻
+獼
+獽
+獾
+獿
+玀
+玁
+玂
+玃
+玄
+玅
+玆
+率
+玈
+玉
+玊
+王
+玎
+玏
+玑
+玒
+玓
+玔
+玕
+玖
+玗
+玘
+玙
+玚
+玛
+玞
+玟
+玠
+玡
+玢
+玤
+玥
+玦
+玧
+玨
+玩
+玫
+玭
+玮
+环
+现
+玱
+玲
+玳
+玶
+玷
+玹
+玺
+玻
+玼
+玾
+玿
+珀
+珂
+珅
+珇
+珈
+珉
+珊
+珋
+珌
+珍
+珎
+珏
+珐
+珑
+珒
+珓
+珕
+珖
+珙
+珛
+珝
+珞
+珠
+珡
+珢
+珣
+珤
+珥
+珦
+珧
+珩
+珪
+珫
+班
+珮
+珰
+珲
+珵
+珶
+珷
+珸
+珹
+珺
+珽
+現
+琀
+琁
+球
+琄
+琅
+理
+琇
+琈
+琉
+琊
+琍
+琎
+琏
+琐
+琔
+琖
+琚
+琛
+琟
+琡
+琢
+琤
+琥
+琦
+琨
+琪
+琫
+琬
+琭
+琮
+琯
+琰
+琱
+琲
+琳
+琴
+琵
+琶
+琹
+琺
+琼
+琿
+瑀
+瑁
+瑂
+瑃
+瑄
+瑅
+瑆
+瑇
+瑈
+瑊
+瑋
+瑑
+瑒
+瑓
+瑔
+瑕
+瑖
+瑗
+瑙
+瑚
+瑛
+瑜
+瑝
+瑞
+瑟
+瑠
+瑢
+瑣
+瑤
+瑥
+瑧
+瑨
+瑩
+瑪
+瑬
+瑭
+瑯
+瑰
+瑱
+瑲
+瑳
+瑴
+瑵
+瑶
+瑷
+瑺
+瑽
+瑾
+瑿
+璀
+璁
+璃
+璄
+璅
+璆
+璇
+璈
+璉
+璊
+璋
+璌
+璎
+璐
+璒
+璕
+璗
+璘
+璙
+璚
+璜
+璞
+璟
+璠
+璡
+璢
+璣
+璥
+璦
+璧
+璨
+璩
+璪
+璫
+璬
+璮
+環
+璱
+璲
+璵
+璸
+璹
+璺
+璽
+璿
+瓀
+瓁
+瓅
+瓈
+瓊
+瓋
+瓌
+瓏
+瓑
+瓒
+瓔
+瓖
+瓘
+瓚
+瓛
+瓜
+瓝
+瓞
+瓟
+瓠
+瓡
+瓢
+瓣
+瓤
+瓥
+瓦
+瓨
+瓩
+瓬
+瓮
+瓯
+瓴
+瓵
+瓶
+瓷
+瓻
+瓽
+瓾
+瓿
+甀
+甂
+甃
+甄
+甈
+甋
+甌
+甍
+甏
+甐
+甑
+甒
+甓
+甔
+甕
+甖
+甗
+甘
+甙
+甚
+甜
+甝
+甞
+生
+甡
+產
+産
+甥
+甦
+用
+甩
+甪
+甫
+甬
+甭
+甮
+甯
+田
+由
+甲
+申
+电
+男
+甸
+甹
+町
+画
+甽
+甾
+甿
+畀
+畅
+畇
+畈
+畊
+畋
+界
+畎
+畏
+畐
+畑
+畔
+畖
+留
+畚
+畛
+畜
+畝
+畟
+畠
+畢
+畣
+畤
+略
+畦
+畧
+番
+畫
+畬
+畯
+異
+畱
+畲
+畳
+畴
+畵
+當
+畷
+畸
+畹
+畺
+畽
+畿
+疀
+疁
+疃
+疆
+疇
+疊
+疋
+疌
+疍
+疎
+疏
+疐
+疑
+疒
+疔
+疕
+疖
+疗
+疘
+疙
+疚
+疝
+疟
+疠
+疡
+疢
+疣
+疤
+疥
+疧
+疫
+疬
+疭
+疮
+疯
+疰
+疱
+疲
+疳
+疴
+疵
+疸
+疹
+疻
+疼
+疽
+疾
+疿
+痀
+痁
+痂
+痃
+痄
+病
+症
+痈
+痉
+痊
+痋
+痌
+痍
+痎
+痏
+痐
+痑
+痒
+痓
+痔
+痕
+痗
+痘
+痙
+痚
+痛
+痝
+痞
+痟
+痠
+痡
+痢
+痣
+痤
+痦
+痧
+痨
+痩
+痪
+痫
+痯
+痰
+痱
+痲
+痳
+痴
+痷
+痸
+痹
+痺
+痻
+痼
+痾
+痿
+瘀
+瘁
+瘃
+瘅
+瘆
+瘈
+瘉
+瘊
+瘋
+瘌
+瘍
+瘏
+瘐
+瘑
+瘓
+瘕
+瘖
+瘗
+瘘
+瘙
+瘛
+瘜
+瘝
+瘞
+瘟
+瘠
+瘡
+瘢
+瘣
+瘤
+瘥
+瘦
+瘧
+瘩
+瘪
+瘫
+瘭
+瘮
+瘯
+瘰
+瘱
+瘲
+瘳
+瘴
+瘵
+瘸
+瘺
+瘻
+瘼
+瘽
+瘾
+瘿
+癀
+療
+癃
+癄
+癅
+癆
+癇
+癈
+癉
+癌
+癍
+癐
+癒
+癓
+癔
+癖
+癗
+癘
+癙
+癜
+癞
+癟
+癠
+癡
+癢
+癣
+癤
+癥
+癧
+癩
+癪
+癫
+癬
+癭
+癮
+癯
+癰
+癱
+癲
+癵
+癸
+癹
+発
+登
+發
+白
+百
+癿
+皁
+皂
+的
+皆
+皇
+皈
+皊
+皋
+皎
+皏
+皐
+皑
+皓
+皕
+皖
+皙
+皚
+皛
+皜
+皝
+皞
+皤
+皦
+皫
+皭
+皮
+皯
+皰
+皱
+皲
+皴
+皵
+皷
+皸
+皺
+皻
+皽
+皾
+皿
+盂
+盃
+盄
+盅
+盆
+盇
+盈
+盉
+益
+盋
+盌
+盍
+盎
+盏
+盐
+监
+盒
+盓
+盔
+盖
+盗
+盘
+盛
+盜
+盝
+盞
+盟
+盡
+監
+盤
+盥
+盦
+盧
+盨
+盩
+盪
+盫
+盬
+盭
+目
+盯
+盱
+盲
+盳
+直
+盷
+相
+盹
+盺
+盻
+盼
+盾
+眀
+省
+眃
+眄
+眅
+眇
+眈
+眉
+眊
+看
+県
+眍
+眐
+眑
+眒
+眓
+眕
+眙
+眚
+眛
+眜
+眝
+眞
+真
+眠
+眢
+眣
+眥
+眦
+眨
+眩
+眬
+眭
+眯
+眱
+眲
+眳
+眴
+眵
+眶
+眷
+眸
+眹
+眺
+眼
+眽
+眾
+着
+睁
+睃
+睄
+睅
+睆
+睇
+睊
+睋
+睌
+睍
+睎
+睏
+睐
+睑
+睒
+睔
+睕
+睖
+睚
+睛
+睜
+睞
+睟
+睠
+睡
+睢
+督
+睥
+睦
+睧
+睨
+睩
+睪
+睫
+睬
+睭
+睮
+睯
+睹
+睺
+睼
+睽
+睾
+睿
+瞀
+瞁
+瞂
+瞄
+瞅
+瞇
+瞉
+瞋
+瞌
+瞍
+瞎
+瞏
+瞑
+瞒
+瞓
+瞖
+瞘
+瞚
+瞛
+瞜
+瞝
+瞞
+瞟
+瞠
+瞡
+瞢
+瞥
+瞧
+瞩
+瞪
+瞫
+瞬
+瞭
+瞰
+瞲
+瞳
+瞴
+瞵
+瞶
+瞷
+瞻
+瞼
+瞽
+瞿
+矁
+矂
+矇
+矉
+矊
+矌
+矍
+矎
+矏
+矐
+矓
+矔
+矕
+矗
+矘
+矙
+矚
+矛
+矜
+矞
+矠
+矢
+矣
+知
+矧
+矩
+矫
+矬
+短
+矮
+矯
+矰
+矱
+矲
+石
+矴
+矶
+矸
+矹
+矺
+矻
+矼
+矽
+矾
+矿
+砀
+码
+砂
+砄
+砅
+砆
+砉
+砋
+砌
+砍
+砎
+砏
+砐
+砑
+砒
+研
+砕
+砖
+砗
+砘
+砚
+砜
+砝
+砟
+砠
+砢
+砣
+砥
+砦
+砧
+砨
+砩
+砫
+砬
+砭
+砮
+砯
+砰
+砲
+砳
+破
+砵
+砷
+砸
+砹
+砺
+砻
+砼
+砾
+础
+硁
+硂
+硃
+硅
+硇
+硈
+硉
+硊
+硌
+硍
+硎
+硏
+硐
+硒
+硓
+硔
+硕
+硖
+硗
+硙
+硚
+硜
+硝
+硠
+硤
+硥
+硨
+硩
+硪
+硫
+硬
+硭
+确
+硯
+硰
+硱
+硼
+硾
+硿
+碁
+碃
+碄
+碅
+碆
+碇
+碈
+碉
+碌
+碍
+碎
+碏
+碑
+碓
+碔
+碕
+碗
+碘
+碚
+碛
+碜
+碞
+碟
+碡
+碣
+碥
+碧
+碨
+碩
+碪
+碫
+碭
+碰
+碱
+碲
+碳
+碴
+碶
+碸
+碹
+確
+碻
+碼
+碽
+碾
+磁
+磃
+磅
+磈
+磉
+磊
+磋
+磌
+磍
+磎
+磏
+磐
+磑
+磔
+磕
+磘
+磙
+磚
+磛
+磜
+磝
+磞
+磟
+磡
+磣
+磥
+磧
+磨
+磩
+磪
+磬
+磯
+磱
+磲
+磳
+磴
+磵
+磷
+磹
+磺
+磻
+磼
+磽
+磾
+磿
+礁
+礂
+礄
+礅
+礉
+礌
+礎
+礐
+礑
+礒
+礓
+礔
+礙
+礛
+礜
+礝
+礞
+礠
+礣
+礥
+礦
+礧
+礨
+礪
+礫
+礬
+礭
+礮
+礱
+礴
+礵
+示
+礼
+礽
+社
+礿
+祀
+祁
+祂
+祃
+祅
+祆
+祇
+祈
+祉
+祊
+祋
+祌
+祎
+祏
+祐
+祑
+祒
+祓
+祔
+祕
+祖
+祗
+祙
+祚
+祛
+祜
+祝
+神
+祟
+祠
+祡
+祢
+祣
+祤
+祥
+祧
+票
+祩
+祪
+祫
+祭
+祯
+祲
+祳
+祴
+祷
+祸
+祹
+祺
+祼
+祾
+祿
+禀
+禁
+禂
+禃
+禄
+禅
+禇
+禊
+禋
+禍
+禎
+福
+禑
+禒
+禓
+禔
+禕
+禖
+禗
+禘
+禚
+禛
+禜
+禟
+禠
+禡
+禤
+禦
+禧
+禨
+禩
+禪
+禫
+禬
+禭
+禮
+禰
+禱
+禳
+禴
+禵
+禷
+禸
+禹
+禺
+离
+禼
+禽
+禾
+禿
+秀
+私
+秃
+秅
+秆
+秈
+秉
+秊
+秋
+种
+秎
+秏
+科
+秒
+秔
+秕
+秖
+秘
+租
+秠
+秣
+秤
+秦
+秧
+秩
+秪
+秫
+秬
+秭
+秮
+积
+称
+秳
+秶
+秸
+秺
+移
+秽
+秾
+稀
+稂
+稃
+稅
+稆
+稈
+稉
+稊
+程
+稌
+稍
+税
+稑
+稒
+稔
+稗
+稘
+稙
+稚
+稜
+稞
+稟
+稠
+稣
+稫
+稬
+稭
+種
+稯
+稰
+稱
+稲
+稳
+稷
+稹
+稺
+稻
+稼
+稽
+稾
+稿
+穀
+穂
+穄
+穅
+穆
+穇
+穈
+穉
+穊
+穋
+穌
+積
+穎
+穏
+穑
+穖
+穗
+穙
+穛
+穜
+穟
+穠
+穡
+穢
+穣
+穤
+穧
+穨
+穩
+穫
+穬
+穭
+穮
+穰
+穱
+穴
+穵
+究
+穷
+穸
+穹
+空
+穽
+穾
+穿
+窀
+突
+窃
+窄
+窅
+窆
+窈
+窉
+窊
+窋
+窌
+窍
+窎
+窏
+窐
+窑
+窒
+窓
+窔
+窕
+窖
+窗
+窘
+窙
+窜
+窝
+窞
+窟
+窠
+窢
+窣
+窥
+窦
+窨
+窩
+窪
+窫
+窬
+窭
+窮
+窯
+窰
+窱
+窲
+窳
+窴
+窵
+窶
+窸
+窺
+窻
+窾
+窿
+竀
+竁
+竃
+竄
+竅
+竇
+竈
+竊
+立
+竑
+竖
+竘
+站
+竚
+竜
+竝
+竞
+竟
+章
+竢
+竣
+童
+竦
+竩
+竪
+竫
+竭
+端
+竴
+競
+竷
+竹
+竺
+竻
+竽
+竿
+笃
+笄
+笅
+笆
+笈
+笊
+笋
+笏
+笐
+笑
+笓
+笔
+笕
+笘
+笙
+笛
+笝
+笞
+笠
+笢
+笤
+笥
+符
+笨
+笩
+笪
+笫
+第
+笭
+笮
+笯
+笰
+笱
+笲
+笳
+笴
+笵
+笸
+笹
+笺
+笻
+笼
+笾
+筀
+筄
+筅
+筆
+筇
+筈
+等
+筊
+筋
+筌
+筍
+筏
+筐
+筑
+筒
+答
+策
+筘
+筚
+筛
+筜
+筝
+筞
+筠
+筡
+筢
+筤
+筥
+筦
+筧
+筩
+筬
+筭
+筮
+筯
+筰
+筱
+筲
+筳
+筴
+筵
+筶
+筷
+筹
+筻
+筼
+签
+简
+箄
+箅
+箆
+箇
+箈
+箊
+箋
+箌
+箍
+箎
+箏
+箐
+箑
+箒
+箓
+箔
+箕
+箖
+算
+箘
+箙
+箛
+箜
+箝
+箠
+管
+箢
+箦
+箧
+箨
+箩
+箪
+箫
+箬
+箭
+箯
+箱
+箴
+箷
+箸
+箹
+箾
+節
+篁
+範
+篆
+篇
+築
+篊
+篋
+篌
+篎
+篑
+篓
+篔
+篕
+篙
+篚
+篛
+篝
+篞
+篠
+篡
+篣
+篤
+篥
+篦
+篧
+篨
+篩
+篪
+篭
+篮
+篯
+篱
+篲
+篳
+篴
+篷
+篸
+篹
+篻
+篼
+篽
+篾
+篿
+簀
+簁
+簂
+簃
+簅
+簇
+簉
+簋
+簌
+簍
+簎
+簏
+簑
+簒
+簕
+簖
+簗
+簙
+簜
+簝
+簞
+簟
+簠
+簡
+簢
+簣
+簥
+簦
+簧
+簨
+簩
+簪
+簫
+簬
+簭
+簮
+簰
+簳
+簷
+簸
+簹
+簺
+簻
+簼
+簽
+簾
+簿
+籀
+籁
+籃
+籈
+籉
+籊
+籌
+籍
+籐
+籔
+籗
+籙
+籛
+籜
+籝
+籟
+籠
+籣
+籤
+籥
+籦
+籧
+籩
+籪
+籬
+籮
+籯
+籲
+米
+籴
+籸
+籹
+籺
+类
+籼
+籽
+籾
+粂
+粃
+粄
+粇
+粉
+粊
+粋
+粍
+粑
+粒
+粔
+粕
+粗
+粘
+粛
+粜
+粝
+粞
+粟
+粢
+粤
+粥
+粦
+粧
+粩
+粪
+粮
+粱
+粲
+粳
+粵
+粹
+粺
+粻
+粼
+粽
+精
+粿
+糀
+糁
+糅
+糇
+糈
+糉
+糊
+糌
+糍
+糎
+糒
+糔
+糕
+糖
+糗
+糙
+糜
+糝
+糞
+糟
+糠
+糢
+糧
+糨
+糪
+糬
+糯
+糰
+糱
+糲
+糴
+糵
+糶
+糷
+糸
+糹
+糺
+系
+糾
+紀
+紂
+紃
+約
+紅
+紆
+紇
+紈
+紉
+紊
+紋
+納
+紎
+紐
+紑
+紒
+紓
+純
+紕
+紖
+紗
+紘
+紙
+級
+紛
+紜
+紝
+紞
+紟
+素
+紡
+索
+紥
+紧
+紩
+紫
+紬
+紮
+累
+細
+紱
+紲
+紳
+紵
+紶
+紸
+紹
+紺
+紼
+紽
+紾
+紿
+絀
+絁
+終
+絃
+組
+絅
+絆
+絇
+経
+絎
+絏
+結
+絓
+絕
+絖
+絘
+絛
+絜
+絝
+絞
+絡
+絢
+絣
+給
+絧
+絨
+絪
+絫
+絭
+絮
+絯
+絰
+統
+絲
+絳
+絵
+絶
+絷
+絹
+絺
+絻
+絼
+絽
+絿
+綀
+綁
+綃
+綄
+綅
+綆
+綈
+綉
+綌
+綍
+綎
+綏
+綑
+經
+綔
+綖
+継
+続
+綜
+綝
+綞
+綟
+綠
+綢
+綣
+綦
+綧
+綪
+綫
+綬
+維
+綮
+綯
+綰
+綱
+網
+綳
+綴
+綵
+綷
+綸
+綹
+綺
+綻
+綼
+綽
+綾
+綿
+緀
+緁
+緂
+緃
+緄
+緅
+緆
+緇
+緈
+緊
+緋
+緌
+緎
+総
+緐
+緑
+緒
+緔
+緖
+緗
+緘
+緙
+線
+緛
+緜
+緝
+緞
+締
+緡
+緣
+緤
+緥
+緦
+緧
+編
+緩
+緪
+緬
+緯
+緰
+緱
+緲
+緳
+練
+緵
+緶
+緷
+緹
+緺
+緻
+縁
+縂
+縃
+縄
+縅
+縈
+縉
+縊
+縋
+縌
+縎
+縏
+縐
+縑
+縒
+縓
+縕
+縗
+縚
+縛
+縜
+縝
+縞
+縟
+縠
+縡
+縢
+縣
+縤
+縦
+縧
+縩
+縪
+縫
+縭
+縮
+縯
+縰
+縱
+縲
+縳
+縴
+縵
+縶
+縷
+縸
+縹
+縻
+縼
+總
+績
+縿
+繀
+繁
+繂
+繃
+繄
+繅
+繆
+繇
+繈
+繉
+繊
+繋
+繍
+繐
+繑
+繒
+織
+繕
+繖
+繘
+繙
+繚
+繜
+繝
+繞
+繟
+繠
+繡
+繢
+繣
+繦
+繧
+繩
+繪
+繫
+繭
+繮
+繯
+繰
+繲
+繳
+繴
+繵
+繶
+繷
+繸
+繹
+繺
+繻
+繼
+繽
+繾
+纀
+纁
+纂
+纆
+纇
+纈
+纊
+纋
+續
+纍
+纏
+纐
+纑
+纒
+纓
+纔
+纕
+纖
+纗
+纘
+纚
+纛
+纜
+纠
+纡
+红
+纣
+纤
+纥
+约
+级
+纨
+纩
+纪
+纫
+纬
+纭
+纮
+纯
+纰
+纱
+纲
+纳
+纴
+纵
+纶
+纷
+纸
+纹
+纺
+纻
+纼
+纽
+纾
+线
+绀
+绁
+绂
+练
+组
+绅
+细
+织
+终
+绉
+绊
+绋
+绌
+绍
+绎
+经
+绐
+绑
+绒
+结
+绔
+绕
+绖
+绗
+绘
+给
+绚
+绛
+络
+绝
+绞
+统
+绠
+绡
+绢
+绣
+绤
+绥
+绦
+继
+绨
+绩
+绪
+绫
+续
+绮
+绯
+绰
+绱
+绲
+绳
+维
+绵
+绶
+绷
+绸
+绹
+绺
+绻
+综
+绽
+绾
+绿
+缀
+缁
+缂
+缃
+缄
+缅
+缆
+缇
+缈
+缉
+缊
+缌
+缎
+缐
+缑
+缒
+缓
+缔
+缕
+编
+缗
+缘
+缙
+缚
+缛
+缜
+缝
+缞
+缟
+缠
+缡
+缢
+缣
+缤
+缥
+缦
+缧
+缨
+缩
+缪
+缫
+缬
+缭
+缮
+缯
+缰
+缱
+缲
+缳
+缴
+缵
+缶
+缸
+缹
+缺
+缽
+缾
+缿
+罂
+罃
+罄
+罅
+罇
+罈
+罉
+罊
+罋
+罌
+罍
+罎
+罏
+罐
+网
+罔
+罕
+罗
+罘
+罚
+罛
+罜
+罝
+罞
+罟
+罠
+罡
+罢
+罣
+罥
+罦
+罧
+罨
+罩
+罪
+罫
+罬
+罭
+置
+罰
+罱
+署
+罳
+罴
+罵
+罶
+罷
+罹
+罺
+罻
+罼
+罽
+罾
+罿
+羁
+羂
+羃
+羅
+羆
+羇
+羈
+羉
+羊
+羋
+羌
+羍
+美
+羑
+羒
+羓
+羔
+羕
+羖
+羗
+羙
+羚
+羛
+羜
+羝
+羞
+羟
+羠
+羡
+羢
+羣
+群
+羥
+羦
+羧
+羨
+義
+羬
+羭
+羯
+羰
+羱
+羲
+羳
+羴
+羵
+羶
+羷
+羸
+羹
+羺
+羼
+羽
+羾
+羿
+翀
+翁
+翂
+翃
+翄
+翅
+翇
+翈
+翊
+翋
+翌
+翍
+翎
+翏
+翐
+翑
+習
+翔
+翕
+翘
+翙
+翚
+翛
+翜
+翟
+翠
+翡
+翢
+翣
+翥
+翦
+翨
+翩
+翪
+翫
+翬
+翭
+翮
+翯
+翰
+翱
+翲
+翳
+翴
+翷
+翹
+翺
+翻
+翼
+翽
+翾
+翿
+耀
+老
+考
+耄
+者
+耆
+耇
+耋
+而
+耍
+耎
+耏
+耐
+耑
+耒
+耔
+耕
+耖
+耗
+耘
+耙
+耜
+耞
+耟
+耠
+耡
+耢
+耤
+耥
+耦
+耧
+耨
+耩
+耪
+耬
+耮
+耰
+耱
+耳
+耴
+耵
+耶
+耷
+耸
+耹
+耻
+耽
+耾
+耿
+聂
+聃
+聆
+聇
+聈
+聊
+聋
+职
+聍
+聏
+聑
+聒
+联
+聖
+聘
+聚
+聝
+聞
+聟
+聡
+聨
+聩
+聪
+聯
+聰
+聱
+聲
+聳
+聴
+聵
+聶
+職
+聹
+聽
+聾
+聿
+肂
+肃
+肄
+肅
+肆
+肇
+肉
+肋
+肌
+肎
+肏
+肐
+肒
+肓
+肕
+肖
+肘
+肙
+肚
+肛
+肜
+肝
+肟
+肠
+股
+肢
+肣
+肤
+肥
+肧
+肩
+肪
+肫
+肭
+肮
+肯
+肱
+育
+肴
+肵
+肷
+肸
+肹
+肺
+肼
+肽
+肾
+肿
+胀
+胁
+胂
+胃
+胄
+胅
+胆
+胇
+胈
+胉
+胊
+背
+胍
+胎
+胏
+胐
+胑
+胔
+胕
+胖
+胗
+胘
+胙
+胚
+胛
+胜
+胝
+胞
+胠
+胡
+胣
+胤
+胥
+胧
+胨
+胩
+胪
+胫
+胬
+胭
+胯
+胰
+胱
+胲
+胳
+胴
+胵
+胶
+胷
+胸
+胹
+胺
+胻
+胼
+能
+胾
+脀
+脁
+脂
+脃
+脅
+脆
+脇
+脈
+脉
+脊
+脍
+脎
+脏
+脐
+脑
+脒
+脓
+脔
+脕
+脖
+脗
+脘
+脙
+脚
+脛
+脞
+脟
+脡
+脣
+脤
+脥
+脧
+脩
+脫
+脬
+脭
+脯
+脰
+脱
+脲
+脳
+脶
+脷
+脸
+脹
+脽
+脾
+脿
+腃
+腄
+腆
+腇
+腈
+腊
+腋
+腌
+腍
+腎
+腏
+腐
+腑
+腒
+腓
+腔
+腕
+腖
+腘
+腙
+腚
+腛
+腜
+腞
+腠
+腢
+腤
+腥
+腦
+腧
+腨
+腩
+腫
+腭
+腮
+腯
+腰
+腱
+腲
+腳
+腴
+腶
+腷
+腸
+腹
+腺
+腻
+腼
+腽
+腾
+腿
+膀
+膂
+膃
+膆
+膇
+膈
+膉
+膊
+膋
+膌
+膍
+膏
+膑
+膕
+膘
+膙
+膚
+膛
+膜
+膝
+膞
+膟
+膠
+膢
+膣
+膥
+膦
+膨
+膩
+膫
+膬
+膮
+膰
+膱
+膲
+膳
+膴
+膵
+膷
+膹
+膺
+膻
+膽
+膾
+膿
+臀
+臂
+臃
+臄
+臅
+臆
+臇
+臈
+臉
+臊
+臋
+臌
+臍
+臏
+臐
+臑
+臒
+臓
+臕
+臗
+臘
+臙
+臚
+臛
+臜
+臝
+臞
+臟
+臠
+臡
+臢
+臣
+臥
+臦
+臧
+臨
+臩
+自
+臬
+臭
+臮
+臯
+臲
+至
+致
+臷
+臸
+臺
+臻
+臼
+臾
+臿
+舀
+舁
+舂
+舄
+舅
+舆
+與
+興
+舉
+舊
+舋
+舌
+舍
+舎
+舐
+舑
+舒
+舔
+舕
+舖
+舗
+舘
+舛
+舜
+舝
+舞
+舟
+舠
+舡
+舢
+舣
+舥
+舨
+舩
+航
+舫
+般
+舭
+舯
+舰
+舱
+舲
+舳
+舴
+舵
+舶
+舷
+舸
+船
+舺
+舻
+舼
+舽
+舾
+艀
+艂
+艄
+艅
+艇
+艉
+艋
+艎
+艏
+艐
+艑
+艒
+艓
+艔
+艕
+艘
+艙
+艚
+艛
+艟
+艤
+艦
+艨
+艫
+艮
+良
+艰
+艱
+色
+艳
+艴
+艵
+艶
+艷
+艸
+艹
+艺
+艼
+艽
+艾
+艿
+芀
+节
+芃
+芄
+芅
+芈
+芊
+芋
+芍
+芎
+芏
+芐
+芑
+芒
+芓
+芔
+芗
+芘
+芙
+芚
+芛
+芜
+芝
+芞
+芟
+芠
+芡
+芣
+芤
+芥
+芦
+芧
+芨
+芩
+芪
+芫
+芬
+芭
+芮
+芯
+芰
+花
+芲
+芳
+芴
+芵
+芷
+芸
+芹
+芺
+芻
+芼
+芽
+芾
+芿
+苀
+苁
+苄
+苅
+苇
+苈
+苉
+苊
+苋
+苌
+苍
+苎
+苏
+苑
+苒
+苓
+苔
+苕
+苖
+苗
+苘
+苙
+苛
+苜
+苝
+苞
+苟
+苠
+苡
+苣
+苤
+若
+苦
+苧
+苨
+苪
+苫
+苬
+苯
+英
+苲
+苳
+苴
+苵
+苶
+苷
+苹
+苺
+苻
+苼
+苾
+茀
+茁
+茂
+范
+茄
+茅
+茆
+茇
+茈
+茉
+茋
+茌
+茍
+茎
+茏
+茑
+茓
+茔
+茕
+茖
+茗
+茘
+茙
+茚
+茛
+茜
+茝
+茠
+茢
+茤
+茥
+茦
+茧
+茨
+茩
+茪
+茫
+茬
+茭
+茮
+茯
+茱
+茲
+茳
+茴
+茵
+茶
+茷
+茸
+茹
+茺
+茻
+茼
+茽
+茿
+荀
+荁
+荂
+荃
+荄
+荅
+荆
+荇
+草
+荊
+荍
+荎
+荏
+荐
+荑
+荒
+荓
+荔
+荖
+荘
+荙
+荚
+荛
+荜
+荞
+荟
+荠
+荡
+荣
+荤
+荥
+荦
+荧
+荨
+荩
+荪
+荫
+荬
+荭
+荮
+药
+荳
+荴
+荵
+荷
+荸
+荺
+荻
+荼
+荽
+荾
+莁
+莃
+莅
+莆
+莉
+莊
+莋
+莌
+莍
+莎
+莏
+莐
+莒
+莓
+莔
+莕
+莖
+莘
+莙
+莛
+莜
+莝
+莞
+莠
+莢
+莣
+莤
+莥
+莧
+莨
+莩
+莪
+莫
+莬
+莰
+莱
+莲
+莳
+莴
+莵
+莶
+获
+莸
+莹
+莺
+莼
+莽
+莿
+菀
+菁
+菂
+菅
+菆
+菇
+菈
+菉
+菊
+菋
+菌
+菍
+菎
+菏
+菑
+菓
+菔
+菕
+菖
+菘
+菙
+菜
+菝
+菞
+菟
+菠
+菡
+菣
+菤
+菥
+菧
+菨
+菩
+菪
+菫
+華
+菰
+菱
+菲
+菳
+菴
+菵
+菶
+菸
+菹
+菺
+菼
+菽
+菾
+菿
+萁
+萃
+萄
+萆
+萇
+萉
+萊
+萋
+萌
+萍
+萎
+萏
+萐
+萑
+萒
+萘
+萚
+萜
+萝
+萠
+萡
+萣
+萤
+营
+萦
+萧
+萨
+萩
+萬
+萭
+萯
+萰
+萱
+萲
+萳
+萴
+萵
+萷
+萸
+萹
+萺
+萼
+落
+萿
+葀
+葂
+葃
+葅
+葆
+葉
+葊
+葋
+葌
+葍
+葎
+葑
+葒
+葖
+著
+葙
+葚
+葛
+葜
+葝
+葞
+葟
+葡
+董
+葤
+葥
+葦
+葧
+葩
+葫
+葬
+葭
+葯
+葰
+葱
+葳
+葴
+葵
+葶
+葷
+葸
+葹
+葺
+葽
+蒂
+蒄
+蒆
+蒇
+蒈
+蒉
+蒋
+蒌
+蒍
+蒎
+蒏
+蒐
+蒓
+蒔
+蒗
+蒘
+蒙
+蒚
+蒛
+蒜
+蒝
+蒞
+蒟
+蒠
+蒡
+蒢
+蒤
+蒧
+蒨
+蒩
+蒪
+蒫
+蒬
+蒭
+蒮
+蒯
+蒱
+蒲
+蒴
+蒶
+蒸
+蒹
+蒺
+蒻
+蒼
+蒽
+蒾
+蒿
+蓀
+蓁
+蓂
+蓄
+蓆
+蓇
+蓉
+蓊
+蓋
+蓌
+蓍
+蓎
+蓏
+蓐
+蓑
+蓓
+蓖
+蓘
+蓚
+蓝
+蓟
+蓠
+蓢
+蓣
+蓥
+蓦
+蓧
+蓨
+蓩
+蓪
+蓫
+蓬
+蓭
+蓮
+蓯
+蓰
+蓱
+蓲
+蓳
+蓴
+蓶
+蓷
+蓹
+蓺
+蓻
+蓼
+蓽
+蓾
+蓿
+蔀
+蔂
+蔃
+蔄
+蔆
+蔇
+蔈
+蔉
+蔊
+蔌
+蔍
+蔎
+蔏
+蔑
+蔒
+蔓
+蔔
+蔕
+蔖
+蔗
+蔘
+蔚
+蔜
+蔝
+蔞
+蔟
+蔠
+蔡
+蔣
+蔤
+蔥
+蔦
+蔧
+蔨
+蔩
+蔪
+蔫
+蔬
+蔭
+蔮
+蔰
+蔱
+蔴
+蔵
+蔷
+蔸
+蔹
+蔺
+蔻
+蔼
+蔽
+蔾
+蔿
+蕀
+蕁
+蕃
+蕄
+蕅
+蕆
+蕇
+蕈
+蕉
+蕊
+蕋
+蕍
+蕎
+蕑
+蕒
+蕓
+蕕
+蕖
+蕗
+蕘
+蕙
+蕚
+蕛
+蕝
+蕞
+蕟
+蕠
+蕡
+蕢
+蕣
+蕤
+蕦
+蕧
+蕨
+蕩
+蕪
+蕫
+蕭
+蕮
+蕰
+蕱
+蕲
+蕴
+蕵
+蕷
+蕸
+蕹
+蕺
+蕻
+蕼
+蕾
+蕿
+薀
+薁
+薂
+薃
+薄
+薅
+薆
+薇
+薈
+薉
+薊
+薋
+薌
+薍
+薎
+薏
+薐
+薑
+薔
+薕
+薖
+薗
+薘
+薙
+薚
+薛
+薜
+薝
+薞
+薟
+薠
+薡
+薢
+薣
+薤
+薦
+薧
+薨
+薩
+薪
+薫
+薬
+薭
+薮
+薯
+薰
+薱
+薲
+薳
+薴
+薵
+薶
+薷
+薸
+薹
+薺
+薽
+薾
+薿
+藀
+藁
+藂
+藃
+藄
+藅
+藆
+藇
+藈
+藉
+藋
+藍
+藎
+藏
+藐
+藑
+藒
+藓
+藔
+藕
+藗
+藘
+藙
+藚
+藜
+藝
+藟
+藠
+藡
+藣
+藤
+藥
+藦
+藨
+藩
+藪
+藫
+藬
+藭
+藰
+藱
+藲
+藴
+藶
+藷
+藸
+藹
+藺
+藻
+藼
+藾
+藿
+蘀
+蘁
+蘂
+蘄
+蘅
+蘆
+蘇
+蘉
+蘊
+蘋
+蘌
+蘐
+蘑
+蘓
+蘖
+蘗
+蘘
+蘙
+蘚
+蘛
+蘜
+蘞
+蘟
+蘠
+蘡
+蘢
+蘣
+蘤
+蘥
+蘦
+蘧
+蘩
+蘪
+蘬
+蘭
+蘮
+蘱
+蘳
+蘴
+蘵
+蘶
+蘸
+蘹
+蘺
+蘻
+蘼
+蘾
+蘿
+虀
+虃
+虆
+虇
+虈
+虉
+虋
+虌
+虎
+虏
+虐
+虑
+虒
+虓
+虔
+處
+虖
+虙
+虚
+虛
+虜
+虞
+號
+虡
+虢
+虣
+虤
+虧
+虨
+虩
+虪
+虫
+虬
+虭
+虮
+虯
+虰
+虱
+虴
+虵
+虷
+虸
+虹
+虺
+虻
+虼
+虽
+虾
+虿
+蚀
+蚁
+蚂
+蚄
+蚅
+蚆
+蚇
+蚊
+蚋
+蚌
+蚍
+蚑
+蚓
+蚕
+蚖
+蚗
+蚘
+蚙
+蚚
+蚜
+蚝
+蚞
+蚡
+蚢
+蚣
+蚤
+蚥
+蚧
+蚨
+蚩
+蚪
+蚬
+蚯
+蚰
+蚱
+蚲
+蚳
+蚴
+蚵
+蚶
+蚷
+蚸
+蚹
+蚺
+蚻
+蚼
+蚿
+蛀
+蛁
+蛂
+蛃
+蛄
+蛅
+蛆
+蛇
+蛈
+蛉
+蛊
+蛋
+蛌
+蛍
+蛎
+蛏
+蛐
+蛑
+蛓
+蛔
+蛕
+蛖
+蛗
+蛘
+蛙
+蛚
+蛛
+蛜
+蛝
+蛞
+蛟
+蛢
+蛣
+蛤
+蛦
+蛩
+蛪
+蛫
+蛬
+蛭
+蛮
+蛯
+蛰
+蛱
+蛲
+蛳
+蛴
+蛵
+蛶
+蛷
+蛸
+蛹
+蛺
+蛻
+蛾
+蜀
+蜁
+蜂
+蜃
+蜄
+蜆
+蜇
+蜈
+蜉
+蜊
+蜋
+蜌
+蜍
+蜎
+蜐
+蜑
+蜒
+蜓
+蜕
+蜖
+蜗
+蜘
+蜙
+蜚
+蜛
+蜜
+蜞
+蜠
+蜡
+蜢
+蜣
+蜤
+蜥
+蜦
+蜧
+蜨
+蜩
+蜪
+蜬
+蜭
+蜮
+蜯
+蜰
+蜱
+蜲
+蜳
+蜴
+蜵
+蜷
+蜸
+蜺
+蜻
+蜼
+蜾
+蜿
+蝀
+蝁
+蝂
+蝃
+蝆
+蝇
+蝈
+蝉
+蝋
+蝌
+蝍
+蝎
+蝐
+蝑
+蝒
+蝓
+蝔
+蝕
+蝖
+蝗
+蝘
+蝙
+蝚
+蝛
+蝜
+蝝
+蝞
+蝟
+蝠
+蝡
+蝣
+蝤
+蝥
+蝦
+蝨
+蝪
+蝫
+蝬
+蝭
+蝮
+蝯
+蝰
+蝱
+蝲
+蝳
+蝴
+蝶
+蝷
+蝸
+蝹
+蝺
+蝻
+蝼
+蝽
+蝾
+蝿
+螁
+螂
+螃
+螄
+螅
+螇
+螈
+螋
+融
+螎
+螏
+螐
+螑
+螒
+螓
+螔
+螖
+螗
+螘
+螚
+螛
+螜
+螝
+螞
+螟
+螠
+螡
+螢
+螣
+螤
+螥
+螨
+螪
+螫
+螬
+螭
+螮
+螯
+螰
+螱
+螳
+螴
+螵
+螶
+螷
+螸
+螹
+螺
+螻
+螼
+螽
+螾
+螿
+蟀
+蟁
+蟂
+蟃
+蟄
+蟅
+蟆
+蟇
+蟈
+蟉
+蟊
+蟋
+蟌
+蟎
+蟏
+蟑
+蟒
+蟓
+蟔
+蟗
+蟘
+蟙
+蟛
+蟜
+蟞
+蟟
+蟠
+蟡
+蟢
+蟣
+蟤
+蟥
+蟦
+蟧
+蟨
+蟪
+蟫
+蟬
+蟭
+蟮
+蟯
+蟲
+蟳
+蟴
+蟶
+蟷
+蟹
+蟺
+蟻
+蟼
+蟾
+蟿
+蠀
+蠁
+蠂
+蠃
+蠄
+蠅
+蠆
+蠈
+蠉
+蠊
+蠋
+蠌
+蠍
+蠏
+蠐
+蠑
+蠓
+蠔
+蠕
+蠖
+蠗
+蠘
+蠙
+蠛
+蠜
+蠝
+蠟
+蠠
+蠡
+蠢
+蠣
+蠤
+蠥
+蠦
+蠨
+蠩
+蠪
+蠫
+蠬
+蠭
+蠮
+蠯
+蠰
+蠱
+蠲
+蠳
+蠵
+蠶
+蠷
+蠸
+蠹
+蠻
+蠼
+蠽
+蠾
+蠿
+血
+衁
+衂
+衃
+衄
+衅
+衆
+衇
+衈
+衊
+衋
+行
+衍
+衎
+衒
+術
+衔
+衕
+衖
+街
+衙
+衚
+衛
+衜
+衝
+衞
+衠
+衡
+衢
+衣
+补
+衧
+表
+衩
+衪
+衫
+衬
+衭
+衮
+衯
+衰
+衱
+衲
+衵
+衶
+衷
+衹
+衺
+衽
+衾
+衿
+袀
+袁
+袂
+袃
+袄
+袅
+袆
+袈
+袉
+袋
+袌
+袍
+袑
+袒
+袓
+袕
+袖
+袗
+袘
+袙
+袚
+袛
+袜
+袞
+袟
+袠
+袡
+袢
+袤
+袧
+袨
+袪
+被
+袬
+袭
+袯
+袰
+袱
+袲
+袴
+袵
+袶
+袷
+袸
+袹
+袺
+袼
+袽
+袾
+袿
+裀
+裁
+裂
+裃
+装
+裆
+裈
+裉
+裊
+裋
+裍
+裎
+裏
+裒
+裔
+裕
+裖
+裗
+裘
+裙
+裚
+裛
+補
+裝
+裞
+裟
+裠
+裡
+裢
+裣
+裤
+裥
+裧
+裨
+裬
+裮
+裯
+裰
+裱
+裲
+裳
+裴
+裵
+裶
+裷
+裸
+裹
+裺
+裻
+裼
+製
+裾
+裿
+褀
+褁
+褂
+褄
+褅
+褆
+複
+褊
+褋
+褌
+褍
+褎
+褐
+褑
+褒
+褓
+褔
+褕
+褖
+褗
+褘
+褙
+褚
+褛
+褞
+褟
+褡
+褢
+褥
+褦
+褧
+褩
+褪
+褫
+褭
+褮
+褯
+褰
+褱
+褲
+褳
+褴
+褵
+褶
+褷
+褸
+褻
+褼
+褽
+褾
+襁
+襂
+襃
+襄
+襆
+襇
+襋
+襌
+襏
+襐
+襒
+襓
+襕
+襖
+襗
+襘
+襚
+襛
+襜
+襝
+襞
+襟
+襠
+襡
+襢
+襣
+襤
+襦
+襩
+襪
+襫
+襬
+襭
+襮
+襯
+襲
+襳
+襴
+襶
+襷
+襹
+襺
+襻
+襾
+西
+要
+覂
+覃
+覅
+覆
+覇
+覈
+見
+覌
+覎
+規
+覓
+覕
+視
+覗
+覘
+覚
+覛
+覜
+覝
+覡
+覢
+覣
+覤
+覦
+覧
+覩
+親
+覬
+覭
+覮
+覯
+覲
+観
+覶
+覷
+覺
+覽
+覿
+觀
+见
+观
+觃
+规
+觅
+视
+觇
+览
+觉
+觊
+觋
+觌
+觎
+觏
+觐
+觑
+角
+觓
+觔
+觕
+觖
+觙
+觚
+觜
+觝
+觞
+觟
+觠
+觡
+觢
+解
+觤
+觥
+触
+觩
+觫
+觬
+觭
+觯
+觰
+觱
+觲
+觳
+觴
+觶
+觷
+觸
+觺
+觻
+觼
+觾
+觿
+言
+訁
+訂
+訃
+訄
+訇
+計
+訊
+訌
+討
+訏
+訐
+訑
+訒
+訓
+訔
+訕
+訖
+託
+記
+訚
+訛
+訝
+訞
+訟
+訢
+訣
+訥
+訧
+訩
+訪
+訬
+設
+訰
+許
+訳
+訴
+訶
+訹
+診
+註
+証
+訾
+訿
+詀
+詁
+詄
+詅
+詆
+詈
+詍
+詎
+詐
+詒
+詔
+評
+詖
+詗
+詘
+詙
+詛
+詝
+詞
+詟
+詠
+詡
+詢
+詣
+詥
+試
+詧
+詩
+詪
+詫
+詬
+詭
+詮
+詰
+話
+該
+詳
+詵
+詶
+詷
+詹
+詻
+詼
+詿
+誂
+誃
+誄
+誅
+誆
+誇
+誉
+誊
+誋
+誌
+認
+誑
+誒
+誓
+誕
+誖
+誘
+誙
+誚
+語
+誠
+誡
+誣
+誤
+誥
+誦
+誨
+說
+誫
+説
+読
+誰
+課
+誴
+誶
+誸
+誹
+誻
+誼
+誾
+調
+諀
+諂
+諄
+諅
+諆
+談
+諈
+諉
+請
+諌
+諍
+諏
+諐
+諑
+諒
+諓
+諔
+諕
+論
+諗
+諚
+諛
+諜
+諝
+諞
+諟
+諠
+諡
+諢
+諤
+諦
+諧
+諪
+諫
+諭
+諮
+諰
+諱
+諲
+諳
+諴
+諵
+諶
+諷
+諸
+諺
+諼
+諾
+謀
+謁
+謂
+謄
+謅
+謇
+謈
+謊
+謋
+謌
+謍
+謎
+謏
+謐
+謑
+謒
+謔
+謕
+謖
+謗
+謘
+謙
+謚
+講
+謜
+謝
+謞
+謠
+謡
+謢
+謣
+謤
+謥
+謦
+謧
+謨
+謩
+謪
+謫
+謬
+謯
+謰
+謱
+謳
+謵
+謷
+謹
+謼
+謾
+譀
+譁
+譅
+譆
+譇
+譈
+證
+譊
+譋
+譌
+譎
+譏
+譐
+譑
+譓
+譔
+譕
+譖
+識
+譙
+譚
+譛
+譜
+譝
+譞
+譟
+譠
+譣
+譥
+警
+譨
+譪
+譫
+譬
+譭
+譯
+議
+譲
+譳
+譴
+護
+譸
+譹
+譺
+譻
+譽
+譾
+譿
+讀
+讂
+讃
+讄
+讆
+變
+讋
+讌
+讎
+讐
+讒
+讓
+讔
+讕
+讖
+讘
+讙
+讚
+讜
+讞
+讟
+计
+订
+讣
+认
+讥
+讦
+讧
+讨
+让
+讪
+讫
+训
+议
+讯
+记
+讱
+讲
+讳
+讴
+讵
+讶
+讷
+许
+讹
+论
+讻
+讼
+讽
+设
+访
+诀
+证
+诂
+诃
+评
+诅
+识
+诇
+诈
+诉
+诊
+诋
+诌
+词
+诎
+诏
+诐
+译
+诒
+诓
+诔
+试
+诖
+诗
+诘
+诙
+诚
+诛
+诜
+话
+诞
+诟
+诠
+诡
+询
+诣
+诤
+该
+详
+诧
+诨
+诩
+诫
+诬
+语
+诮
+误
+诰
+诱
+诲
+诳
+说
+诵
+请
+诸
+诹
+诺
+读
+诼
+诽
+课
+诿
+谀
+谁
+谂
+调
+谄
+谅
+谆
+谇
+谈
+谊
+谋
+谌
+谍
+谎
+谏
+谐
+谑
+谒
+谓
+谔
+谕
+谖
+谗
+谙
+谚
+谛
+谜
+谝
+谞
+谟
+谠
+谡
+谢
+谣
+谤
+谥
+谦
+谧
+谨
+谩
+谪
+谫
+谬
+谭
+谮
+谯
+谰
+谱
+谲
+谳
+谴
+谵
+谶
+谷
+谹
+谻
+谼
+谽
+谾
+谿
+豁
+豃
+豅
+豆
+豇
+豈
+豉
+豊
+豋
+豌
+豍
+豎
+豏
+豐
+豔
+豕
+豗
+豚
+豜
+豝
+豟
+象
+豢
+豤
+豥
+豦
+豨
+豪
+豫
+豬
+豭
+豮
+豯
+豰
+豱
+豲
+豳
+豵
+豶
+豷
+豸
+豹
+豺
+豻
+豽
+豿
+貀
+貁
+貂
+貄
+貅
+貆
+貉
+貊
+貌
+貍
+貏
+貐
+貑
+貒
+貓
+貔
+貕
+貗
+貘
+貙
+貛
+貜
+貝
+貞
+負
+財
+貢
+貣
+貤
+貧
+貨
+販
+貪
+貫
+責
+貭
+貮
+貯
+貰
+貲
+貳
+貴
+貶
+買
+貸
+貺
+費
+貼
+貽
+貾
+貿
+賀
+賁
+賂
+賃
+賄
+賅
+資
+賈
+賉
+賊
+賌
+賎
+賑
+賒
+賓
+賔
+賕
+賙
+賚
+賛
+賜
+賞
+賟
+賠
+賡
+賢
+賣
+賤
+賥
+賦
+賧
+賨
+質
+賫
+賬
+賭
+賮
+賴
+賵
+賸
+賹
+賺
+賻
+購
+賽
+賾
+贄
+贅
+贆
+贇
+贈
+贊
+贋
+贌
+贍
+贏
+贐
+贓
+贔
+贕
+贖
+贗
+贙
+贛
+贜
+贝
+贞
+负
+贡
+财
+责
+贤
+败
+账
+货
+质
+贩
+贪
+贫
+贬
+购
+贮
+贯
+贰
+贱
+贲
+贳
+贴
+贵
+贶
+贷
+贸
+费
+贺
+贻
+贼
+贽
+贾
+贿
+赀
+赁
+赂
+赃
+资
+赅
+赆
+赇
+赈
+赉
+赊
+赋
+赌
+赍
+赎
+赏
+赐
+赑
+赒
+赓
+赔
+赕
+赖
+赗
+赘
+赙
+赚
+赛
+赜
+赝
+赞
+赟
+赠
+赡
+赢
+赣
+赤
+赦
+赧
+赨
+赩
+赪
+赫
+赬
+赭
+赮
+走
+赳
+赴
+赵
+赶
+起
+赸
+趀
+趁
+趂
+趄
+超
+越
+趋
+趌
+趍
+趎
+趏
+趐
+趑
+趓
+趔
+趕
+趖
+趙
+趜
+趟
+趠
+趡
+趣
+趥
+趧
+趨
+趪
+趫
+趬
+趭
+趮
+趯
+趱
+趲
+足
+趴
+趵
+趷
+趸
+趹
+趺
+趼
+趾
+趿
+跂
+跃
+跄
+跅
+跆
+跇
+跈
+跋
+跌
+跍
+跎
+跏
+跐
+跑
+跓
+跕
+跖
+跗
+跘
+跙
+跚
+跛
+跜
+距
+跞
+跟
+跠
+跡
+跢
+跣
+跤
+跥
+跦
+跧
+跨
+跩
+跪
+跫
+跬
+跮
+路
+跰
+跱
+跲
+跳
+跴
+践
+跶
+跷
+跸
+跹
+跺
+跻
+跼
+跽
+跾
+跿
+踂
+踃
+踅
+踆
+踇
+踉
+踊
+踌
+踍
+踎
+踏
+踐
+踑
+踒
+踓
+踔
+踕
+踖
+踘
+踙
+踛
+踝
+踞
+踟
+踠
+踡
+踢
+踣
+踤
+踥
+踦
+踧
+踩
+踪
+踫
+踬
+踮
+踯
+踰
+踱
+踳
+踴
+踵
+踶
+踸
+踹
+踺
+踼
+踽
+踾
+蹀
+蹁
+蹂
+蹄
+蹅
+蹇
+蹈
+蹉
+蹊
+蹋
+蹌
+蹍
+蹎
+蹏
+蹐
+蹑
+蹒
+蹓
+蹔
+蹕
+蹖
+蹗
+蹙
+蹚
+蹛
+蹜
+蹝
+蹞
+蹟
+蹠
+蹡
+蹢
+蹣
+蹤
+蹥
+蹦
+蹧
+蹩
+蹪
+蹬
+蹭
+蹯
+蹰
+蹲
+蹳
+蹴
+蹵
+蹶
+蹸
+蹺
+蹻
+蹼
+蹽
+蹾
+蹿
+躁
+躂
+躄
+躅
+躆
+躇
+躈
+躉
+躊
+躋
+躍
+躏
+躐
+躑
+躒
+躓
+躔
+躕
+躖
+躗
+躙
+躚
+躜
+躝
+躞
+躟
+躠
+躡
+躣
+躤
+躥
+躦
+躨
+躩
+躪
+身
+躬
+躭
+躯
+躰
+躲
+躳
+躶
+躺
+躽
+軀
+車
+軋
+軌
+軍
+軎
+軏
+軑
+軒
+軓
+軔
+軘
+軛
+軝
+軞
+軟
+転
+軤
+軥
+軦
+軧
+軨
+軫
+軬
+軮
+軯
+軱
+軲
+軵
+軶
+軷
+軸
+軹
+軺
+軻
+軼
+軽
+軾
+軿
+輀
+輁
+輂
+較
+輄
+輅
+輆
+輇
+輈
+載
+輊
+輋
+輌
+輐
+輑
+輒
+輓
+輔
+輕
+輖
+輗
+輘
+輚
+輛
+輜
+輝
+輞
+輟
+輠
+輣
+輤
+輥
+輦
+輩
+輪
+輬
+輭
+輮
+輯
+輲
+輳
+輴
+輵
+輶
+輷
+輸
+輹
+輻
+輾
+輿
+轀
+轂
+轃
+轄
+轅
+轆
+轇
+轈
+轉
+轍
+轎
+轏
+轐
+轑
+轒
+轓
+轔
+轕
+轖
+轗
+轘
+轙
+轚
+轛
+轝
+轞
+轟
+轠
+轡
+轢
+轤
+车
+轧
+轨
+轩
+轪
+轫
+转
+轭
+轮
+软
+轰
+轱
+轲
+轳
+轴
+轵
+轶
+轷
+轸
+轹
+轺
+轻
+轼
+载
+轾
+轿
+辀
+辁
+辂
+较
+辄
+辅
+辆
+辇
+辈
+辉
+辊
+辋
+辌
+辍
+辎
+辏
+辐
+辑
+辒
+输
+辔
+辕
+辖
+辗
+辘
+辙
+辚
+辛
+辜
+辞
+辟
+辠
+辢
+辣
+辥
+辦
+辧
+辨
+辩
+辫
+辭
+辮
+辯
+辰
+辱
+農
+辴
+辵
+边
+辺
+辻
+込
+辽
+达
+辿
+迁
+迂
+迄
+迅
+迆
+过
+迈
+迉
+迋
+迍
+迎
+运
+近
+迒
+迓
+返
+迕
+迖
+还
+这
+进
+远
+违
+连
+迟
+迡
+迢
+迣
+迤
+迥
+迦
+迨
+迩
+迪
+迫
+迭
+迮
+述
+迳
+迴
+迵
+迷
+迸
+迹
+迺
+迻
+追
+迾
+迿
+退
+送
+适
+逃
+逄
+逅
+逆
+逈
+选
+逊
+逋
+逌
+逍
+逎
+透
+逐
+逑
+递
+逓
+途
+逕
+逖
+逗
+這
+通
+逛
+逜
+逝
+逞
+速
+造
+逡
+逢
+連
+逤
+逦
+逨
+逩
+逭
+逮
+逯
+週
+進
+逴
+逵
+逶
+逸
+逹
+逺
+逻
+逼
+逾
+逿
+遁
+遂
+遄
+遅
+遆
+遇
+遊
+運
+遍
+過
+遏
+遐
+遑
+遒
+道
+達
+違
+遗
+遘
+遙
+遛
+遜
+遝
+遞
+遠
+遡
+遢
+遣
+遥
+遨
+適
+遭
+遮
+遯
+遰
+遲
+遳
+遴
+遵
+遶
+遷
+選
+遹
+遺
+遻
+遼
+遽
+遾
+避
+邀
+邁
+邂
+邃
+還
+邅
+邆
+邇
+邈
+邉
+邊
+邋
+邍
+邏
+邐
+邑
+邓
+邔
+邕
+邗
+邘
+邙
+邛
+邝
+邞
+邟
+邠
+邡
+邢
+那
+邥
+邦
+邧
+邨
+邪
+邬
+邮
+邯
+邰
+邱
+邲
+邳
+邴
+邵
+邶
+邸
+邹
+邺
+邻
+邽
+邾
+邿
+郁
+郃
+郄
+郅
+郇
+郈
+郊
+郋
+郎
+郏
+郐
+郑
+郓
+郔
+郕
+郖
+郗
+郘
+郙
+郚
+郛
+郜
+郝
+郞
+郟
+郠
+郡
+郢
+郣
+郤
+郥
+郦
+郧
+部
+郪
+郫
+郭
+郯
+郰
+郱
+郲
+郳
+郴
+郵
+郷
+郸
+郹
+郺
+郻
+郼
+都
+郾
+郿
+鄀
+鄁
+鄂
+鄃
+鄄
+鄅
+鄆
+鄇
+鄈
+鄉
+鄋
+鄌
+鄍
+鄎
+鄏
+鄐
+鄑
+鄒
+鄔
+鄖
+鄗
+鄘
+鄙
+鄚
+鄛
+鄜
+鄝
+鄞
+鄟
+鄠
+鄡
+鄢
+鄣
+鄤
+鄦
+鄧
+鄨
+鄩
+鄪
+鄫
+鄬
+鄭
+鄮
+鄯
+鄰
+鄱
+鄲
+鄳
+鄴
+鄵
+鄶
+鄸
+鄹
+鄺
+鄻
+鄾
+酀
+酁
+酂
+酃
+酄
+酅
+酆
+酇
+酈
+酉
+酊
+酋
+酌
+配
+酎
+酏
+酐
+酒
+酓
+酔
+酖
+酗
+酘
+酚
+酝
+酞
+酟
+酡
+酢
+酣
+酤
+酥
+酦
+酧
+酨
+酩
+酪
+酬
+酮
+酯
+酰
+酱
+酲
+酳
+酴
+酵
+酶
+酷
+酸
+酹
+酺
+酼
+酽
+酾
+酿
+醁
+醂
+醃
+醅
+醆
+醇
+醉
+醊
+醋
+醌
+醍
+醐
+醑
+醒
+醓
+醕
+醖
+醗
+醙
+醚
+醛
+醜
+醞
+醟
+醢
+醣
+醤
+醥
+醧
+醨
+醪
+醫
+醬
+醭
+醮
+醯
+醰
+醱
+醲
+醳
+醴
+醵
+醷
+醸
+醹
+醺
+醻
+醼
+醽
+醾
+醿
+釀
+釁
+釂
+釃
+釅
+釆
+采
+釈
+釉
+释
+釋
+里
+重
+野
+量
+釐
+金
+釓
+釔
+釕
+釗
+釘
+釙
+釚
+釜
+針
+釢
+釣
+釤
+釦
+釧
+釩
+釪
+釬
+釭
+釱
+釳
+釴
+釵
+釷
+釸
+釹
+釺
+釽
+釿
+鈀
+鈁
+鈃
+鈄
+鈆
+鈇
+鈈
+鈉
+鈊
+鈌
+鈍
+鈎
+鈏
+鈐
+鈑
+鈒
+鈔
+鈕
+鈖
+鈚
+鈞
+鈢
+鈣
+鈥
+鈦
+鈧
+鈫
+鈮
+鈰
+鈳
+鈴
+鈶
+鈷
+鈸
+鈹
+鈺
+鈽
+鈾
+鈿
+鉀
+鉄
+鉅
+鉆
+鉇
+鉈
+鉉
+鉊
+鉋
+鉌
+鉍
+鉏
+鉑
+鉒
+鉓
+鉕
+鉗
+鉚
+鉛
+鉞
+鉟
+鉠
+鉢
+鉤
+鉥
+鉦
+鉧
+鉬
+鉭
+鉮
+鉯
+鉱
+鉲
+鉴
+鉶
+鉷
+鉸
+鉹
+鉺
+鉻
+鉼
+鉾
+鉿
+銀
+銂
+銃
+銅
+銈
+銋
+銍
+銎
+銑
+銓
+銔
+銕
+銖
+銗
+銘
+銚
+銛
+銜
+銠
+銣
+銤
+銥
+銦
+銨
+銩
+銪
+銫
+銬
+銭
+銮
+銱
+銲
+銳
+銶
+銷
+銹
+銻
+銼
+銾
+鋀
+鋁
+鋂
+鋃
+鋄
+鋅
+鋆
+鋇
+鋈
+鋊
+鋋
+鋌
+鋏
+鋐
+鋒
+鋕
+鋗
+鋘
+鋙
+鋝
+鋟
+鋡
+鋤
+鋥
+鋦
+鋨
+鋩
+鋪
+鋭
+鋮
+鋯
+鋰
+鋱
+鋲
+鋳
+鋶
+鋸
+鋹
+鋺
+鋼
+錀
+錁
+錄
+錆
+錈
+錍
+錎
+錏
+錐
+錒
+錕
+錘
+錙
+錚
+錛
+錞
+錟
+錠
+錡
+錢
+錣
+錦
+錧
+錨
+錫
+錬
+錭
+錮
+錯
+録
+錳
+錵
+錶
+錸
+錻
+錼
+錾
+鍀
+鍁
+鍆
+鍇
+鍈
+鍉
+鍊
+鍋
+鍌
+鍍
+鍏
+鍐
+鍑
+鍔
+鍖
+鍘
+鍚
+鍛
+鍜
+鍝
+鍟
+鍠
+鍤
+鍥
+鍩
+鍪
+鍬
+鍭
+鍮
+鍰
+鍱
+鍳
+鍵
+鍶
+鍷
+鍺
+鍼
+鍾
+鎂
+鎃
+鎅
+鎉
+鎊
+鎌
+鎎
+鎏
+鎒
+鎓
+鎔
+鎖
+鎗
+鎘
+鎚
+鎛
+鎝
+鎞
+鎡
+鎢
+鎣
+鎦
+鎧
+鎩
+鎪
+鎬
+鎭
+鎮
+鎯
+鎰
+鎳
+鎵
+鎷
+鎸
+鎹
+鎻
+鏂
+鏃
+鏇
+鏈
+鏊
+鏌
+鏍
+鏏
+鏐
+鏑
+鏓
+鏖
+鏗
+鏘
+鏙
+鏚
+鏜
+鏝
+鏞
+鏟
+鏡
+鏢
+鏤
+鏦
+鏨
+鏬
+鏰
+鏳
+鏴
+鏵
+鏷
+鏹
+鏺
+鏻
+鏽
+鏾
+鐀
+鐃
+鐇
+鐈
+鐋
+鐍
+鐎
+鐏
+鐐
+鐓
+鐔
+鐕
+鐖
+鐘
+鐙
+鐚
+鐝
+鐠
+鐡
+鐤
+鐦
+鐧
+鐨
+鐫
+鐬
+鐭
+鐮
+鐯
+鐲
+鐳
+鐵
+鐶
+鐸
+鐺
+鐻
+鐽
+鐾
+鐿
+鑀
+鑁
+鑂
+鑄
+鑅
+鑊
+鑋
+鑌
+鑐
+鑑
+鑒
+鑔
+鑕
+鑚
+鑛
+鑞
+鑠
+鑢
+鑣
+鑤
+鑨
+鑪
+鑫
+鑭
+鑮
+鑯
+鑰
+鑱
+鑲
+鑴
+鑵
+鑷
+鑹
+鑼
+鑽
+鑾
+鑿
+钀
+钁
+钂
+钃
+钆
+钇
+针
+钉
+钊
+钋
+钌
+钍
+钎
+钏
+钐
+钒
+钓
+钔
+钕
+钖
+钗
+钘
+钙
+钚
+钛
+钜
+钝
+钞
+钟
+钠
+钡
+钢
+钣
+钤
+钥
+钦
+钧
+钨
+钩
+钪
+钫
+钬
+钭
+钮
+钯
+钰
+钱
+钲
+钳
+钴
+钵
+钷
+钹
+钺
+钻
+钼
+钽
+钾
+钿
+铀
+铁
+铂
+铃
+铄
+铅
+铆
+铈
+铉
+铊
+铋
+铌
+铍
+铎
+铏
+铐
+铑
+铒
+铕
+铖
+铗
+铘
+铙
+铚
+铛
+铜
+铝
+铞
+铟
+铠
+铡
+铢
+铣
+铤
+铥
+铧
+铨
+铩
+铪
+铫
+铬
+铭
+铮
+铯
+铰
+铱
+铲
+铳
+铴
+铵
+银
+铷
+铸
+铹
+铺
+铻
+铼
+铽
+链
+铿
+销
+锁
+锂
+锃
+锄
+锅
+锆
+锇
+锈
+锉
+锊
+锋
+锌
+锍
+锎
+锏
+锐
+锑
+锒
+锓
+锔
+锕
+锖
+锗
+锘
+错
+锚
+锛
+锜
+锝
+锞
+锟
+锡
+锢
+锣
+锤
+锥
+锦
+锧
+锨
+锩
+锪
+锫
+锬
+锭
+键
+锯
+锰
+锱
+锲
+锳
+锴
+锵
+锶
+锷
+锸
+锹
+锺
+锻
+锼
+锽
+锾
+锿
+镀
+镁
+镂
+镃
+镄
+镅
+镆
+镇
+镈
+镉
+镊
+镋
+镌
+镍
+镎
+镏
+镐
+镑
+镒
+镓
+镔
+镕
+镖
+镗
+镘
+镚
+镛
+镜
+镝
+镞
+镠
+镡
+镢
+镣
+镤
+镥
+镦
+镧
+镨
+镩
+镪
+镫
+镬
+镭
+镮
+镯
+镰
+镱
+镲
+镳
+镴
+镵
+镶
+長
+镺
+镻
+镼
+镽
+长
+門
+閂
+閃
+閆
+閇
+閈
+閉
+開
+閌
+閍
+閎
+閏
+閑
+閒
+間
+閔
+閘
+閜
+閞
+閟
+閡
+関
+閣
+閤
+閥
+閦
+閧
+閨
+閩
+閫
+閬
+閭
+閰
+閱
+閲
+閵
+閶
+閷
+閹
+閺
+閻
+閼
+閽
+閾
+閿
+闀
+闃
+闅
+闆
+闇
+闈
+闉
+闊
+闋
+闌
+闍
+闐
+闑
+闒
+闓
+闔
+闕
+闖
+闘
+闚
+闛
+關
+闞
+闟
+闠
+闡
+闢
+闤
+闥
+门
+闩
+闪
+闫
+闭
+问
+闯
+闰
+闱
+闲
+闳
+间
+闵
+闶
+闷
+闸
+闹
+闺
+闻
+闼
+闽
+闾
+闿
+阀
+阁
+阂
+阃
+阄
+阅
+阆
+阇
+阈
+阉
+阊
+阋
+阌
+阍
+阎
+阏
+阐
+阑
+阒
+阔
+阕
+阖
+阗
+阘
+阙
+阚
+阜
+阝
+阞
+队
+阠
+阡
+阢
+阤
+阨
+阪
+阬
+阭
+阮
+阯
+阰
+阱
+防
+阳
+阴
+阵
+阶
+阸
+阹
+阺
+阻
+阼
+阽
+阿
+陀
+陁
+陂
+附
+际
+陆
+陇
+陈
+陉
+陊
+陋
+陌
+降
+陎
+陏
+限
+陑
+陓
+陔
+陕
+陗
+陘
+陛
+陜
+陝
+陞
+陟
+陡
+院
+陣
+除
+陥
+陧
+陨
+险
+陪
+陫
+陬
+陭
+陰
+陲
+陳
+陴
+陵
+陶
+陷
+陸
+険
+陻
+陼
+陽
+陾
+隃
+隄
+隅
+隆
+隇
+隈
+隉
+隊
+隋
+隍
+階
+随
+隐
+隑
+隒
+隓
+隔
+隕
+隗
+隘
+隙
+際
+障
+隞
+隠
+隣
+隤
+隧
+隨
+隩
+險
+隮
+隰
+隱
+隲
+隳
+隴
+隶
+隷
+隸
+隹
+隺
+隻
+隼
+隽
+难
+隿
+雀
+雁
+雂
+雄
+雅
+集
+雇
+雈
+雉
+雊
+雋
+雌
+雍
+雎
+雏
+雑
+雒
+雓
+雔
+雕
+雖
+雗
+雘
+雙
+雚
+雛
+雜
+雝
+雞
+雟
+雠
+離
+難
+雨
+雩
+雪
+雫
+雯
+雰
+雱
+雲
+雳
+零
+雷
+雹
+雺
+電
+雽
+雾
+雿
+需
+霁
+霂
+霄
+霅
+霆
+震
+霈
+霉
+霊
+霋
+霍
+霎
+霏
+霐
+霑
+霒
+霓
+霖
+霙
+霜
+霞
+霠
+霢
+霣
+霤
+霧
+霨
+霩
+霪
+霫
+霬
+霭
+霮
+霰
+露
+霵
+霶
+霸
+霹
+霽
+霾
+霿
+靁
+靂
+靃
+靄
+靆
+靈
+靉
+靑
+青
+靓
+靖
+静
+靚
+靛
+靜
+非
+靠
+靡
+面
+靥
+靦
+靨
+革
+靪
+靫
+靬
+靭
+靮
+靰
+靱
+靲
+靳
+靴
+靶
+靷
+靸
+靺
+靻
+靼
+靽
+靾
+靿
+鞀
+鞁
+鞂
+鞃
+鞄
+鞅
+鞆
+鞈
+鞊
+鞋
+鞌
+鞍
+鞎
+鞏
+鞑
+鞒
+鞔
+鞗
+鞘
+鞙
+鞚
+鞜
+鞞
+鞠
+鞡
+鞣
+鞤
+鞥
+鞦
+鞧
+鞨
+鞪
+鞫
+鞬
+鞭
+鞮
+鞯
+鞲
+鞳
+鞴
+鞵
+鞶
+鞷
+鞹
+鞻
+鞽
+鞾
+鞿
+韁
+韂
+韃
+韄
+韅
+韆
+韇
+韉
+韋
+韌
+韍
+韎
+韏
+韐
+韑
+韓
+韔
+韕
+韗
+韙
+韜
+韝
+韞
+韟
+韠
+韡
+韣
+韥
+韦
+韧
+韨
+韩
+韪
+韫
+韬
+韭
+韮
+韰
+韱
+音
+韵
+韶
+韹
+韺
+韻
+韽
+韾
+響
+頁
+頂
+頃
+頄
+項
+順
+頇
+須
+頊
+頌
+頍
+頎
+頏
+預
+頑
+頒
+頓
+頔
+頖
+頗
+領
+頚
+頜
+頝
+頞
+頠
+頡
+頤
+頦
+頨
+頩
+頫
+頬
+頭
+頯
+頰
+頲
+頴
+頵
+頷
+頸
+頹
+頻
+頼
+頽
+顁
+顄
+顅
+顆
+顇
+顈
+顉
+顊
+顋
+題
+額
+顎
+顏
+顐
+顑
+顒
+顓
+顔
+顕
+顗
+願
+顙
+顛
+顜
+顝
+類
+顟
+顠
+顢
+顣
+顤
+顥
+顦
+顧
+顩
+顪
+顫
+顯
+顰
+顱
+顲
+顳
+顴
+页
+顶
+顷
+顸
+项
+顺
+须
+顼
+顽
+顾
+顿
+颀
+颁
+颂
+颃
+预
+颅
+领
+颇
+颈
+颉
+颊
+颋
+颌
+颍
+颎
+颏
+颐
+频
+颓
+颔
+颖
+颗
+题
+颙
+颚
+颛
+颜
+额
+颞
+颟
+颠
+颡
+颢
+颤
+颥
+颦
+颧
+風
+颬
+颭
+颮
+颯
+颱
+颲
+颳
+颶
+颸
+颺
+颻
+颼
+颽
+颾
+颿
+飁
+飂
+飄
+飆
+飈
+飉
+飋
+飌
+风
+飏
+飐
+飑
+飒
+飓
+飔
+飕
+飗
+飘
+飙
+飛
+飜
+飞
+食
+飡
+飢
+飣
+飥
+飧
+飨
+飩
+飪
+飫
+飬
+飭
+飮
+飯
+飱
+飲
+飴
+飶
+飺
+飼
+飽
+飾
+飿
+餀
+餂
+餃
+餄
+餅
+餈
+餉
+養
+餌
+餍
+餎
+餏
+餐
+餑
+餒
+餓
+餔
+餕
+餖
+餗
+餘
+餚
+餛
+餜
+餝
+餞
+餟
+餠
+餡
+餤
+餥
+餧
+館
+餪
+餫
+餬
+餭
+餮
+餯
+餰
+餱
+餲
+餳
+餵
+餷
+餹
+餺
+餻
+餼
+餽
+餾
+餿
+饁
+饂
+饃
+饅
+饇
+饈
+饉
+饊
+饋
+饌
+饍
+饎
+饐
+饑
+饒
+饔
+饕
+饗
+饘
+饙
+饛
+饜
+饞
+饟
+饡
+饢
+饥
+饧
+饨
+饩
+饪
+饫
+饬
+饭
+饮
+饯
+饰
+饱
+饲
+饳
+饴
+饵
+饶
+饷
+饸
+饹
+饺
+饻
+饼
+饽
+饿
+馁
+馃
+馄
+馅
+馆
+馇
+馈
+馉
+馊
+馋
+馌
+馍
+馏
+馐
+馑
+馒
+馓
+馔
+馕
+首
+馗
+馘
+香
+馛
+馝
+馞
+馣
+馥
+馦
+馧
+馨
+馬
+馭
+馮
+馯
+馰
+馱
+馲
+馳
+馴
+馵
+馹
+馺
+馻
+馼
+馽
+駁
+駂
+駃
+駄
+駅
+駆
+駈
+駉
+駍
+駎
+駏
+駐
+駑
+駒
+駓
+駔
+駕
+駖
+駘
+駙
+駛
+駜
+駝
+駟
+駡
+駢
+駣
+駤
+駥
+駩
+駪
+駬
+駭
+駮
+駰
+駱
+駴
+駶
+駷
+駸
+駹
+駺
+駻
+駼
+駽
+駾
+駿
+騁
+騂
+騃
+騄
+騅
+騇
+騉
+騊
+騋
+騌
+騍
+騎
+騏
+騑
+騒
+験
+騕
+騖
+騙
+騚
+騛
+騜
+騝
+騞
+騠
+騢
+騣
+騤
+騥
+騧
+騨
+騩
+騪
+騫
+騬
+騭
+騮
+騰
+騱
+騲
+騴
+騵
+騶
+騷
+騸
+騹
+騺
+騽
+騾
+驀
+驁
+驂
+驃
+驄
+驅
+驆
+驈
+驉
+驊
+驌
+驍
+驎
+驏
+驒
+驓
+驔
+驕
+驖
+驗
+驚
+驛
+驞
+驟
+驠
+驢
+驤
+驥
+驦
+驨
+驩
+驪
+驫
+马
+驭
+驮
+驯
+驰
+驱
+驲
+驳
+驴
+驵
+驶
+驷
+驸
+驹
+驺
+驻
+驼
+驽
+驾
+驿
+骀
+骁
+骂
+骃
+骄
+骅
+骆
+骇
+骈
+骉
+骊
+骋
+验
+骍
+骎
+骏
+骐
+骑
+骒
+骓
+骕
+骖
+骗
+骘
+骙
+骚
+骛
+骜
+骝
+骞
+骟
+骠
+骡
+骢
+骣
+骤
+骥
+骦
+骧
+骨
+骫
+骭
+骯
+骰
+骱
+骴
+骶
+骷
+骸
+骹
+骺
+骼
+骽
+骾
+骿
+髀
+髁
+髂
+髃
+髄
+髅
+髆
+髇
+髈
+髊
+髋
+髌
+髍
+髎
+髏
+髐
+髑
+髒
+髓
+體
+髕
+髖
+高
+髙
+髟
+髡
+髢
+髣
+髥
+髦
+髧
+髪
+髫
+髬
+髭
+髮
+髯
+髲
+髳
+髴
+髶
+髷
+髹
+髺
+髻
+髼
+髽
+髾
+鬀
+鬁
+鬃
+鬄
+鬅
+鬆
+鬈
+鬉
+鬊
+鬋
+鬌
+鬍
+鬎
+鬏
+鬐
+鬒
+鬓
+鬕
+鬖
+鬗
+鬘
+鬙
+鬚
+鬞
+鬟
+鬠
+鬢
+鬣
+鬤
+鬥
+鬧
+鬨
+鬩
+鬪
+鬫
+鬬
+鬮
+鬯
+鬱
+鬲
+鬳
+鬵
+鬶
+鬷
+鬹
+鬺
+鬻
+鬼
+鬾
+鬿
+魁
+魂
+魃
+魄
+魅
+魆
+魇
+魈
+魉
+魊
+魋
+魌
+魍
+魎
+魏
+魑
+魔
+魕
+魖
+魘
+魚
+魛
+魞
+魟
+魠
+魡
+魣
+魦
+魧
+魨
+魩
+魮
+魯
+魱
+魴
+魵
+魶
+魷
+魻
+魼
+魾
+鮀
+鮁
+鮂
+鮃
+鮄
+鮅
+鮆
+鮇
+鮊
+鮋
+鮍
+鮎
+鮐
+鮑
+鮒
+鮓
+鮗
+鮚
+鮛
+鮜
+鮝
+鮞
+鮟
+鮠
+鮡
+鮣
+鮤
+鮥
+鮦
+鮨
+鮪
+鮫
+鮭
+鮮
+鮰
+鮵
+鮸
+鮹
+鮻
+鮿
+鯀
+鯁
+鯃
+鯆
+鯇
+鯈
+鯉
+鯊
+鯏
+鯒
+鯓
+鯔
+鯕
+鯖
+鯗
+鯙
+鯛
+鯝
+鯞
+鯠
+鯡
+鯢
+鯤
+鯦
+鯧
+鯨
+鯪
+鯫
+鯬
+鯭
+鯮
+鯰
+鯱
+鯶
+鯷
+鯸
+鯻
+鯽
+鯿
+鰂
+鰃
+鰅
+鰆
+鰈
+鰉
+鰋
+鰌
+鰍
+鰏
+鰐
+鰒
+鰓
+鰕
+鰗
+鰛
+鰜
+鰝
+鰟
+鰣
+鰤
+鰥
+鰧
+鰨
+鰩
+鰫
+鰬
+鰭
+鰮
+鰯
+鰰
+鰱
+鰲
+鰳
+鰴
+鰶
+鰷
+鰹
+鰺
+鰻
+鰼
+鰾
+鰿
+鱀
+鱁
+鱂
+鱄
+鱅
+鱆
+鱇
+鱈
+鱉
+鱊
+鱋
+鱍
+鱐
+鱒
+鱓
+鱔
+鱕
+鱖
+鱗
+鱘
+鱚
+鱝
+鱞
+鱟
+鱠
+鱢
+鱣
+鱥
+鱦
+鱧
+鱨
+鱬
+鱭
+鱮
+鱰
+鱲
+鱳
+鱴
+鱵
+鱷
+鱸
+鱹
+鱺
+鱻
+鱼
+鱽
+鱾
+鱿
+鲀
+鲁
+鲂
+鲃
+鲅
+鲆
+鲇
+鲈
+鲉
+鲊
+鲋
+鲌
+鲍
+鲎
+鲏
+鲐
+鲑
+鲒
+鲔
+鲕
+鲖
+鲗
+鲘
+鲙
+鲚
+鲛
+鲜
+鲝
+鲞
+鲟
+鲠
+鲡
+鲢
+鲣
+鲤
+鲥
+鲦
+鲧
+鲨
+鲩
+鲪
+鲫
+鲬
+鲭
+鲮
+鲯
+鲰
+鲱
+鲲
+鲳
+鲴
+鲵
+鲷
+鲸
+鲹
+鲺
+鲻
+鲼
+鲽
+鲾
+鲿
+鳀
+鳁
+鳂
+鳃
+鳄
+鳅
+鳇
+鳈
+鳉
+鳊
+鳌
+鳍
+鳎
+鳏
+鳐
+鳑
+鳒
+鳓
+鳔
+鳕
+鳖
+鳗
+鳘
+鳙
+鳚
+鳛
+鳜
+鳝
+鳞
+鳟
+鳠
+鳡
+鳢
+鳣
+鳤
+鳥
+鳦
+鳧
+鳩
+鳪
+鳬
+鳭
+鳯
+鳰
+鳱
+鳲
+鳳
+鳴
+鳵
+鳶
+鳷
+鳹
+鳺
+鳻
+鳼
+鳽
+鳾
+鳿
+鴀
+鴃
+鴅
+鴆
+鴇
+鴈
+鴉
+鴎
+鴐
+鴒
+鴓
+鴔
+鴕
+鴗
+鴘
+鴛
+鴝
+鴞
+鴟
+鴠
+鴡
+鴢
+鴣
+鴥
+鴦
+鴨
+鴩
+鴫
+鴮
+鴯
+鴰
+鴱
+鴳
+鴴
+鴶
+鴷
+鴸
+鴻
+鴽
+鴾
+鴿
+鵀
+鵁
+鵂
+鵃
+鵄
+鵅
+鵊
+鵋
+鵌
+鵎
+鵏
+鵐
+鵑
+鵒
+鵓
+鵔
+鵖
+鵗
+鵙
+鵛
+鵜
+鵝
+鵞
+鵟
+鵠
+鵡
+鵧
+鵩
+鵪
+鵫
+鵬
+鵮
+鵯
+鵰
+鵱
+鵲
+鵳
+鵴
+鵵
+鵷
+鵸
+鵹
+鵺
+鵻
+鵼
+鵽
+鵾
+鶀
+鶁
+鶂
+鶄
+鶅
+鶆
+鶇
+鶈
+鶉
+鶊
+鶋
+鶌
+鶏
+鶐
+鶒
+鶓
+鶔
+鶖
+鶗
+鶘
+鶙
+鶚
+鶛
+鶜
+鶝
+鶞
+鶟
+鶠
+鶡
+鶢
+鶣
+鶤
+鶥
+鶨
+鶩
+鶪
+鶬
+鶭
+鶯
+鶱
+鶲
+鶴
+鶵
+鶶
+鶷
+鶹
+鶺
+鶻
+鶼
+鶾
+鶿
+鷀
+鷁
+鷂
+鷃
+鷄
+鷅
+鷇
+鷈
+鷉
+鷊
+鷋
+鷎
+鷏
+鷐
+鷑
+鷒
+鷓
+鷕
+鷖
+鷗
+鷘
+鷙
+鷚
+鷛
+鷜
+鷞
+鷟
+鷡
+鷢
+鷣
+鷤
+鷥
+鷦
+鷩
+鷫
+鷮
+鷯
+鷰
+鷲
+鷳
+鷵
+鷶
+鷷
+鷸
+鷹
+鷺
+鷻
+鷽
+鷾
+鷿
+鸀
+鸁
+鸂
+鸃
+鸄
+鸅
+鸆
+鸇
+鸉
+鸊
+鸋
+鸌
+鸍
+鸏
+鸐
+鸑
+鸒
+鸓
+鸔
+鸕
+鸗
+鸘
+鸙
+鸚
+鸛
+鸜
+鸝
+鸞
+鸟
+鸠
+鸡
+鸢
+鸣
+鸤
+鸥
+鸦
+鸧
+鸨
+鸩
+鸪
+鸫
+鸬
+鸭
+鸮
+鸯
+鸰
+鸱
+鸲
+鸳
+鸵
+鸶
+鸷
+鸸
+鸹
+鸺
+鸻
+鸼
+鸽
+鸾
+鸿
+鹀
+鹁
+鹂
+鹃
+鹄
+鹅
+鹆
+鹇
+鹈
+鹉
+鹊
+鹋
+鹌
+鹍
+鹎
+鹏
+鹐
+鹑
+鹒
+鹔
+鹕
+鹖
+鹗
+鹘
+鹙
+鹚
+鹛
+鹜
+鹝
+鹞
+鹟
+鹠
+鹡
+鹢
+鹣
+鹤
+鹦
+鹧
+鹨
+鹩
+鹪
+鹫
+鹬
+鹭
+鹮
+鹯
+鹰
+鹱
+鹲
+鹳
+鹴
+鹵
+鹸
+鹹
+鹺
+鹻
+鹼
+鹽
+鹾
+鹿
+麀
+麁
+麂
+麃
+麅
+麆
+麇
+麈
+麉
+麊
+麋
+麌
+麎
+麐
+麑
+麒
+麓
+麔
+麖
+麗
+麙
+麚
+麛
+麜
+麝
+麞
+麟
+麠
+麡
+麤
+麥
+麦
+麧
+麩
+麪
+麮
+麯
+麰
+麴
+麵
+麶
+麷
+麸
+麹
+麺
+麻
+麼
+麽
+麾
+麿
+黀
+黁
+黂
+黃
+黄
+黇
+黈
+黉
+黌
+黍
+黎
+黏
+黐
+黑
+黒
+黓
+黔
+黕
+黖
+默
+黙
+黚
+黛
+黜
+黝
+點
+黟
+黠
+黡
+黢
+黤
+黥
+黧
+黨
+黩
+黪
+黫
+黭
+黮
+黯
+黰
+黲
+黳
+黴
+黵
+黶
+黷
+黹
+黺
+黻
+黼
+黽
+黾
+黿
+鼀
+鼁
+鼂
+鼃
+鼆
+鼇
+鼈
+鼉
+鼊
+鼋
+鼍
+鼎
+鼏
+鼐
+鼒
+鼓
+鼕
+鼖
+鼗
+鼘
+鼙
+鼛
+鼜
+鼠
+鼢
+鼣
+鼤
+鼥
+鼨
+鼩
+鼪
+鼫
+鼬
+鼭
+鼮
+鼯
+鼰
+鼱
+鼲
+鼳
+鼴
+鼵
+鼶
+鼷
+鼸
+鼹
+鼻
+鼽
+鼾
+齀
+齁
+齂
+齃
+齇
+齉
+齊
+齋
+齌
+齍
+齎
+齏
+齐
+齑
+齒
+齔
+齕
+齖
+齗
+齘
+齙
+齛
+齜
+齝
+齞
+齟
+齠
+齡
+齢
+齣
+齤
+齥
+齦
+齧
+齩
+齪
+齫
+齬
+齮
+齯
+齰
+齱
+齲
+齴
+齵
+齶
+齷
+齸
+齹
+齺
+齻
+齾
+齿
+龀
+龁
+龂
+龃
+龄
+龅
+龆
+龇
+龈
+龉
+龊
+龋
+龌
+龍
+龎
+龐
+龑
+龒
+龔
+龕
+龗
+龙
+龚
+龛
+龜
+龝
+龟
+龠
+龢
+龤
+鿍
+鿎
+鿏
+郎
+凉
+︽
+︾
+﹏
+﹥
+﹪
+！
+＃
+％
+＆
+（
+）
+＋
+，
+－
+．
+／
+０
+１
+２
+３
+４
+５
+６
+７
+８
+９
+：
+；
+＝
+？
+＠
+Ａ
+Ｂ
+Ｃ
+Ｄ
+Ｅ
+Ｆ
+Ｇ
+Ｈ
+Ｉ
+Ｊ
+Ｋ
+Ｌ
+Ｍ
+Ｎ
+Ｏ
+Ｐ
+Ｑ
+Ｒ
+Ｓ
+Ｔ
+Ｕ
+Ｖ
+Ｗ
+Ｘ
+Ｙ
+Ｚ
+ａ
+ｂ
+ｃ
+ｄ
+ｅ
+ｆ
+ｇ
+ｈ
+ｉ
+ｊ
+ｋ
+ｌ
+ｍ
+ｎ
+ｏ
+ｐ
+ｑ
+ｒ
+ｓ
+ｔ
+ｕ
+ｖ
+ｗ
+ｘ
+ｙ
+ｚ
+｜
+～
+･
+ｦ
+ｧ
+ｨ
+ｩ
+ｪ
+ｫ
+ｬ
+ｭ
+ｮ
+ｯ
+ｱ
+ｲ
+ｳ
+ｴ
+ｵ
+ｶ
+ｷ
+ｸ
+ｹ
+ｺ
+ｻ
+ｼ
+ｽ
+ｾ
+ｿ
+ﾀ
+ﾁ
+ﾂ
+ﾃ
+ﾄ
+ﾅ
+ﾆ
+ﾇ
+ﾈ
+ﾉ
+ﾊ
+ﾋ
+ﾌ
+ﾍ
+ﾎ
+ﾏ
+ﾐ
+ﾑ
+ﾒ
+ﾓ
+ﾔ
+ﾕ
+ﾖ
+ﾗ
+ﾘ
+ﾙ
+ﾚ
+ﾛ
+ﾜ
+ﾝ
+ﾞ
+ﾟ
+￥
+𠅤
+𠙶
+𠳐
+𠴰
+𡎚
+𡐓
+𡑍
+𡚁
+𣇉
+𣗋
+𣲗
+𣲘
+𣸣
+𤋮
+𤧛
+𤩽
+𤫉
+𤺥
+𥔲
+𥕢
+𥖨
+𥻗
+𦈡
+𦍑
+𦒍
+𦙶
+𦝼
+𦭜
+𦰡
+𧿹
+𨐈
+𨙸
+𨚕
+𨟠
+𨭉
+𨱇
+𨱏
+𨱑
+𨱔
+𨺙
+𩓐
+𩽾
+𩾃
+𩾌
+𪟝
+𪣻
+𪤗
+𪨰
+𪨶
+𪩘
+𪾢
+𫄧
+𫄨
+𫄷
+𫄸
+𫇭
+𫌀
+𫍣
+𫍯
+𫍲
+𫍽
+𫐄
+𫐐
+𫐓
+𫑡
+𫓧
+𫓯
+𫓶
+𫓹
+𫔍
+𫔎
+𫔶
+𫖮
+𫖯
+𫖳
+𫗧
+𫗴
+𫘜
+𫘝
+𫘦
+𫘧
+𫘨
+𫘪
+𫘬
+𫚕
+𫚖
+𫚭
+𫛭
+𫞩
+𫟅
+𫟦
+𫟹
+𫟼
+𫠆
+𫠊
+𫠜
+𫢸
+𫫇
+𫭟
+𫭢
+𫭼
+𫮃
+𫰛
+𫵷
+𫶇
+𫷷
+𫸩
+𬀩
+𬀪
+𬂩
+𬃊
+𬇕
+𬇙
+𬇹
+𬉼
+𬊈
+𬊤
+𬌗
+𬍛
+𬍡
+𬍤
+𬒈
+𬒔
+𬒗
+𬕂
+𬘓
+𬘘
+𬘡
+𬘩
+𬘫
+𬘬
+𬘭
+𬘯
+𬙂
+𬙊
+𬙋
+𬜬
+𬜯
+𬞟
+𬟁
+𬟽
+𬣙
+𬣞
+𬣡
+𬣳
+𬤇
+𬤊
+𬤝
+𬨂
+𬨎
+𬩽
+𬪩
+𬬩
+𬬭
+𬬮
+𬬱
+𬬸
+𬬹
+𬬻
+𬬿
+𬭁
+𬭊
+𬭎
+𬭚
+𬭛
+𬭤
+𬭩
+𬭬
+𬭯
+𬭳
+𬭶
+𬭸
+𬭼
+𬮱
+𬮿
+𬯀
+𬯎
+𬱖
+𬱟
+𬳵
+𬳶
+𬳽
+𬳿
+𬴂
+𬴃
+𬴊
+𬶋
+𬶍
+𬶏
+𬶐
+𬶟
+𬶠
+𬶨
+𬶭
+𬶮
+𬷕
+𬸘
+𬸚
+𬸣
+𬸦
+𬸪
+𬹼
+𬺈
+𬺓
+🀄
+🃏
+🅰
+🅱
+🅾
+🅿
+🆎
+🆑
+🆒
+🆓
+🆔
+🆕
+🆖
+🆗
+🆘
+🆙
+🆚
+🇨
+🇩
+🇪
+🇫
+🇬
+🇮
+🇯
+🇰
+🇷
+🇺
+🈁
+🌀
+🌁
+🌂
+🌃
+🌄
+🌅
+🌆
+🌇
+🌈
+🌉
+🌊
+🌋
+🌌
+🌍
+🌎
+🌏
+🌐
+🌑
+🌒
+🌓
+🌔
+🌕
+🌖
+🌗
+🌘
+🌙
+🌚
+🌛
+🌜
+🌝
+🌞
+🌟
+🌠
+🌰
+🌱
+🌲
+🌳
+🌴
+🌵
+🌷
+🌸
+🌹
+🌺
+🌻
+🌼
+🌽
+🌾
+🌿
+🍀
+🍁
+🍂
+🍃
+🍄
+🍅
+🍆
+🍇
+🍈
+🍉
+🍊
+🍋
+🍌
+🍍
+🍎
+🍏
+🍐
+🍑
+🍒
+🍓
+🍔
+🍕
+🍖
+🍗
+🍘
+🍙
+🍚
+🍛
+🍜
+🍝
+🍞
+🍟
+🍠
+🍡
+🍢
+🍣
+🍤
+🍥
+🍦
+🍧
+🍨
+🍩
+🍪
+🍫
+🍬
+🍭
+🍮
+🍯
+🍰
+🍱
+🍲
+🍳
+🍴
+🍵
+🍶
+🍷
+🍸
+🍹
+🍺
+🍻
+🍼
+🎀
+🎁
+🎂
+🎃
+🎄
+🎅
+🎆
+🎇
+🎈
+🎉
+🎊
+🎋
+🎌
+🎍
+🎎
+🎏
+🎐
+🎑
+🎒
+🎓
+🎠
+🎡
+🎢
+🎣
+🎤
+🎥
+🎦
+🎧
+🎨
+🎩
+🎪
+🎫
+🎬
+🎭
+🎮
+🎯
+🎰
+🎱
+🎲
+🎳
+🎴
+🎵
+🎶
+🎷
+🎸
+🎹
+🎺
+🎻
+🎼
+🎽
+🎾
+🎿
+🏀
+🏁
+🏂
+🏃
+🏄
+🏆
+🏇
+🏈
+🏉
+🏊
+🏠
+🏡
+🏢
+🏣
+🏤
+🏥
+🏦
+🏧
+🏨
+🏩
+🏪
+🏫
+🏬
+🏭
+🏮
+🏯
+🏰
+🐀
+🐁
+🐂
+🐃
+🐄
+🐅
+🐆
+🐇
+🐈
+🐉
+🐊
+🐋
+🐌
+🐍
+🐎
+🐏
+🐐
+🐑
+🐒
+🐓
+🐔
+🐕
+🐖
+🐗
+🐘
+🐙
+🐚
+🐛
+🐜
+🐝
+🐞
+🐟
+🐠
+🐡
+🐢
+🐣
+🐤
+🐥
+🐦
+🐧
+🐨
+🐩
+🐪
+🐫
+🐬
+🐭
+🐮
+🐯
+🐰
+🐱
+🐲
+🐳
+🐴
+🐵
+🐶
+🐷
+🐸
+🐹
+🐺
+🐻
+🐼
+🐽
+🐾
+👀
+👂
+👃
+👄
+👅
+👆
+👇
+👈
+👉
+👊
+👋
+👌
+👍
+👎
+👏
+👐
+👑
+👒
+👓
+👔
+👕
+👖
+👗
+👘
+👙
+👚
+👛
+👜
+👝
+👞
+👟
+👠
+👡
+👢
+👣
+👤
+👥
+👦
+👧
+👨
+👩
+👪
+👫
+👬
+👭
+👮
+👯
+👰
+👱
+👲
+👳
+👴
+👵
+👶
+👷
+👸
+👹
+👺
+👻
+👼
+👽
+👾
+👿
+💀
+💁
+💂
+💃
+💄
+💅
+💆
+💇
+💈
+💉
+💊
+💋
+💌
+💍
+💎
+💏
+💐
+💑
+💒
+💓
+💔
+💕
+💖
+💗
+💘
+💙
+💚
+💛
+💜
+💝
+💞
+💟
+💠
+💡
+💢
+💣
+💤
+💥
+💦
+💧
+💨
+💩
+💪
+💫
+💬
+💭
+💮
+💯
+💰
+💲
+💳
+💴
+💵
+💶
+💷
+💸
+💹
+💺
+💻
+💼
+💽
+💾
+💿
+📀
+📁
+📂
+📃
+📄
+📅
+📆
+📇
+📈
+📉
+📊
+📋
+📌
+📍
+📎
+📏
+📐
+📑
+📒
+📓
+📔
+📕
+📖
+📗
+📘
+📙
+📚
+📛
+📜
+📝
+📞
+📟
+📠
+📡
+📢
+📣
+📤
+📥
+📦
+📧
+📨
+📩
+📪
+📫
+📬
+📭
+📮
+📯
+📰
+📱
+📲
+📳
+📴
+📵
+📶
+📷
+📹
+📺
+📻
+📼
+🔀
+🔁
+🔂
+🔃
+🔄
+🔅
+🔆
+🔇
+🔉
+🔊
+🔋
+🔌
+🔍
+🔎
+🔏
+🔐
+🔑
+🔒
+🔓
+🔔
+🔕
+🔖
+🔗
+🔘
+🔙
+🔚
+🔛
+🔜
+🔝
+🔞
+🔟
+🔠
+🔡
+🔢
+🔣
+🔤
+🔥
+🔦
+🔧
+🔨
+🔩
+🔪
+🔫
+🔬
+🔭
+🔮
+🔯
+🔰
+🔱
+🔲
+🔳
+🔴
+🔵
+🔶
+🔷
+🔸
+🔹
+🔺
+🔻
+🔼
+🔽
+🕐
+🕑
+🕒
+🕓
+🕔
+🕕
+🕖
+🕗
+🕘
+🕙
+🕚
+🕛
+🕜
+🕝
+🕞
+🕟
+🕠
+🕡
+🕢
+🕣
+🕤
+🕥
+🕦
+🕧
+🗻
+🗼
+🗽
+🗾
+🗿
+😀
+😁
+😂
+😃
+😄
+😅
+😆
+😇
+😈
+😉
+😊
+😋
+😌
+😍
+😎
+😏
+😐
+😑
+😒
+😓
+😔
+😕
+😖
+😗
+😘
+😙
+😚
+😛
+😜
+😝
+😞
+😟
+😠
+😡
+😢
+😣
+😤
+😥
+😦
+😧
+😨
+😩
+😪
+😫
+😬
+😭
+😮
+😯
+😰
+😱
+😲
+😳
+😴
+😵
+😶
+😷
+😸
+😹
+😺
+😻
+😼
+😽
+😾
+😿
+🙀
+🙅
+🙆
+🙇
+🙈
+🙉
+🙊
+🙋
+🙌
+🙍
+🙎
+🙏
+🚀
+🚁
+🚂
+🚃
+🚄
+🚅
+🚆
+🚇
+🚈
+🚉
+🚊
+🚌
+🚍
+🚎
+🚏
+🚐
+🚑
+🚒
+🚓
+🚔
+🚕
+🚖
+🚗
+🚘
+🚙
+🚚
+🚛
+🚜
+🚝
+🚞
+🚟
+🚠
+🚡
+🚢
+🚣
+🚤
+🚥
+🚦
+🚧
+🚨
+🚩
+🚪
+🚫
+🚬
+🚭
+🚮
+🚯
+🚰
+🚱
+🚲
+🚳
+🚴
+🚵
+🚶
+🚷
+🚸
+🚹
+🚺
+🚻
+🚼
+🚽
+🚾
+🚿
+🛀
+🛁
+🛂
+🛃
+🛄
+🛅
diff --git a/ppocr/utils/dict/ppocrv6_tiny_dict.txt b/ppocr/utils/dict/ppocrv6_tiny_dict.txt
new file mode 100644
index 0000000..3757d18
--- /dev/null
+++ b/ppocr/utils/dict/ppocrv6_tiny_dict.txt
@@ -0,0 +1,6904 @@
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+:
+;
+<
+=
+>
+?
+@
+[
+\
+]
+^
+_
+`
+{
+|
+}
+~
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+¢
+£
+¤
+¥
+¦
+§
+¨
+©
+ª
+«
+¬
+®
+¯
+°
+±
+²
+³
+´
+µ
+¶
+·
+¸
+¹
+º
+»
+¼
+½
+¾
+¿
+À
+Á
+Â
+Ã
+Ä
+Å
+Æ
+Ç
+È
+É
+Ê
+Ë
+Ì
+Í
+Î
+Ï
+Ð
+Ñ
+Ò
+Ó
+Ô
+Õ
+Ö
+×
+Ø
+Ù
+Ú
+Û
+Ü
+Ý
+Þ
+ß
+à
+á
+â
+ã
+ä
+å
+æ
+ç
+è
+é
+ê
+ë
+ì
+í
+î
+ï
+ð
+ñ
+ò
+ó
+ô
+õ
+ö
+÷
+ø
+ù
+ú
+û
+ü
+ý
+þ
+ÿ
+Ā
+ā
+Ă
+ă
+Ą
+ą
+Ć
+ć
+Ĉ
+ĉ
+Ċ
+ċ
+Č
+č
+Ď
+ď
+Đ
+đ
+Ē
+ē
+Ĕ
+ĕ
+Ė
+ė
+Ę
+ę
+Ě
+ě
+Ĝ
+ĝ
+Ğ
+ğ
+Ġ
+ġ
+Ģ
+ģ
+Ĥ
+ĥ
+Ħ
+ħ
+Ĩ
+ĩ
+Ī
+ī
+Ĭ
+ĭ
+Į
+į
+İ
+ı
+IJ
+ij
+Ĵ
+ĵ
+Ķ
+ķ
+ĸ
+Ĺ
+ĺ
+Ļ
+ļ
+Ľ
+ľ
+Ŀ
+ŀ
+Ł
+ł
+Ń
+ń
+Ņ
+ņ
+Ň
+ň
+ʼn
+Ŋ
+ŋ
+Ō
+ō
+Ŏ
+ŏ
+Ő
+ő
+Œ
+œ
+Ŕ
+ŕ
+Ŗ
+ŗ
+Ř
+ř
+Ś
+ś
+Ŝ
+ŝ
+Ş
+ş
+Š
+š
+Ţ
+ţ
+Ť
+ť
+Ŧ
+ŧ
+Ũ
+ũ
+Ū
+ū
+Ŭ
+ŭ
+Ů
+ů
+Ű
+ű
+Ų
+ų
+Ŵ
+ŵ
+Ŷ
+ŷ
+Ÿ
+Ź
+ź
+Ż
+ż
+Ž
+ž
+ſ
+ƒ
+Α
+Β
+Γ
+Δ
+Ε
+Ζ
+Η
+Θ
+Ι
+Κ
+Λ
+Μ
+Ν
+Ξ
+Ο
+Π
+Ρ
+Σ
+Τ
+Υ
+Φ
+Χ
+Ψ
+Ω
+α
+β
+γ
+δ
+ε
+ζ
+η
+θ
+ι
+κ
+λ
+μ
+ν
+ξ
+ο
+π
+ρ
+σ
+τ
+υ
+φ
+χ
+ψ
+ω
+ϕ
+ϵ
+Ạ
+ồ
+–
+—
+―
+‖
+‘
+’
+“
+”
+†
+‡
+•
+‥
+…
+‰
+′
+″
+※
+⁎
+⁰
+⁴
+⁵
+⁶
+⁷
+⁸
+⁹
+⁺
+⁻
+⁼
+⁽
+⁾
+ⁿ
+₀
+₁
+₂
+₃
+₄
+₅
+₆
+₇
+₈
+₉
+₊
+₋
+₌
+₍
+₎
+ₒ
+₠
+₡
+₤
+₧
+₩
+₪
+₫
+€
+₭
+₮
+₱
+₹
+₺
+₽
+₿
+℃
+℉
+ℎ
+ℏ
+ℑ
+℘
+ℜ
+™
+℧
+Å
+ℵ
+ℶ
+ℷ
+ℸ
+ℹ
+⅀
+Ⅰ
+Ⅱ
+Ⅲ
+Ⅳ
+Ⅴ
+Ⅵ
+Ⅶ
+Ⅷ
+Ⅸ
+Ⅹ
+Ⅺ
+Ⅻ
+ⅰ
+ⅱ
+ⅲ
+ⅳ
+ⅴ
+ⅵ
+ⅶ
+ⅷ
+ⅸ
+ⅹ
+ⅺ
+ⅻ
+←
+↑
+→
+↓
+↔
+↕
+↖
+↗
+↘
+↙
+⇐
+⇑
+⇒
+⇓
+⇔
+⇕
+∀
+∂
+∃
+∅
+∆
+∇
+∈
+∉
+∏
+∑
+−
+∓
+∕
+∙
+√
+∝
+∞
+∠
+∥
+∧
+∨
+∩
+∪
+∫
+∬
+∮
+∴
+∵
+∶
+∷
+∼
+∽
+≈
+≌
+≒
+≠
+≡
+≤
+≥
+≦
+≧
+≪
+≫
+≮
+≯
+⊂
+⊃
+⊆
+⊇
+⊊
+⊋
+⊕
+⊗
+⊙
+⊥
+⋅
+①
+②
+③
+④
+⑤
+⑥
+⑦
+⑧
+⑨
+⑩
+⑪
+⑫
+⑬
+⑭
+⑮
+⑯
+⑰
+⑱
+⑲
+⑳
+■
+□
+▲
+△
+▶
+▼
+▽
+◀
+◆
+◇
+○
+◎
+●
+☀
+☁
+☂
+★
+☆
+☎
+☐
+☑
+☒
+☔
+☕
+☠
+♀
+♂
+♠
+♡
+♣
+♤
+♥
+♦
+♻
+⚓
+⚠
+⚡
+✓
+✔
+✕
+✖
+✘
+✚
+✠
+✨
+✳
+✴
+✶
+✽
+❄
+❌
+❎
+❓
+❤
+　
+、
+。
+々
+〆
+〇
+〈
+〉
+《
+》
+「
+」
+『
+』
+【
+】
+〒
+〔
+〕
+〖
+〗
+〜
+〰
+〵
+〽
+㈱
+㊗
+㊙
+一
+丁
+七
+丄
+万
+丈
+三
+上
+下
+不
+与
+丐
+丑
+专
+且
+丕
+世
+丘
+丙
+业
+丛
+东
+丝
+丞
+両
+丢
+两
+严
+丧
+个
+丫
+中
+丰
+串
+临
+丸
+丹
+为
+主
+丼
+丽
+举
+丿
+乂
+乃
+久
+么
+义
+之
+乌
+乍
+乎
+乏
+乐
+乒
+乓
+乔
+乖
+乘
+乙
+乜
+九
+乞
+也
+习
+乡
+书
+买
+乱
+乳
+乸
+乾
+了
+予
+争
+事
+二
+亍
+于
+亏
+云
+互
+亓
+五
+井
+亘
+亚
+些
+亜
+亞
+亟
+亡
+亢
+交
+亥
+亦
+产
+亨
+亩
+享
+京
+亭
+亮
+亲
+亳
+亵
+人
+亿
+什
+仁
+仃
+仄
+仅
+仆
+仇
+今
+介
+仍
+从
+仑
+仓
+仔
+仕
+他
+仗
+付
+仙
+仝
+仞
+仟
+仡
+代
+令
+以
+仨
+仪
+们
+仰
+仲
+件
+价
+仺
+任
+份
+仿
+企
+伈
+伉
+伊
+伍
+伎
+伏
+伐
+休
+众
+优
+伙
+会
+伞
+伟
+传
+伤
+伦
+伪
+伫
+伯
+估
+伱
+伴
+伶
+伸
+伺
+似
+伽
+佃
+但
+位
+低
+住
+佐
+佑
+体
+何
+佗
+佘
+余
+佚
+佛
+作
+佝
+佞
+佟
+你
+佣
+佤
+佧
+佩
+佬
+佯
+佰
+佳
+佶
+佼
+使
+侃
+侄
+來
+侈
+例
+侍
+侏
+侑
+侗
+供
+依
+侠
+侣
+侥
+侦
+侧
+侨
+侩
+侬
+侮
+侯
+侵
+便
+係
+促
+俄
+俊
+俎
+俏
+俐
+俑
+俗
+俘
+俚
+保
+俞
+俟
+信
+俨
+俩
+俪
+俫
+俬
+俭
+修
+俯
+俱
+俵
+俸
+俺
+俾
+倆
+倉
+個
+倌
+倍
+倏
+們
+倒
+倔
+倘
+候
+倚
+倜
+借
+倡
+倦
+倩
+倪
+倫
+倬
+倭
+债
+值
+倾
+偃
+假
+偈
+偉
+偌
+偎
+偏
+偕
+做
+停
+健
+偶
+偷
+偻
+偽
+偿
+傀
+傅
+傈
+傍
+備
+傢
+傣
+傥
+储
+傩
+催
+傲
+傳
+傷
+傻
+像
+僑
+僚
+僧
+僳
+僵
+價
+僻
+儀
+億
+儋
+儒
+儡
+儥
+優
+儱
+儿
+兀
+允
+元
+兄
+充
+兆
+先
+光
+克
+免
+兑
+兒
+兔
+兖
+党
+兜
+兢
+入
+內
+全
+兩
+八
+公
+六
+兮
+兰
+共
+关
+兴
+兵
+其
+具
+典
+兹
+养
+兼
+兽
+兿
+冀
+内
+冈
+冉
+册
+再
+冏
+冒
+冕
+冗
+写
+冚
+军
+农
+冠
+冢
+冤
+冥
+冬
+冮
+冯
+冰
+冲
+决
+况
+冶
+冷
+冻
+冼
+冽
+净
+凄
+准
+凇
+凈
+凉
+凋
+凌
+凍
+减
+凑
+凖
+凛
+凝
+几
+凡
+凤
+処
+凬
+凭
+凯
+凰
+凱
+凳
+凶
+凸
+凹
+出
+击
+函
+凿
+刀
+刁
+刂
+刃
+分
+切
+刈
+刊
+刍
+刑
+划
+列
+刘
+则
+刚
+创
+初
+删
+判
+刨
+利
+刪
+别
+刮
+到
+制
+刷
+券
+刹
+刺
+刻
+刿
+剁
+剂
+剃
+削
+剌
+前
+剎
+剐
+剑
+剔
+剖
+剡
+剥
+剧
+剩
+剪
+副
+割
+創
+剽
+剿
+劇
+劈
+劉
+劍
+劑
+力
+劝
+办
+功
+加
+务
+劣
+动
+助
+努
+劫
+劭
+励
+劲
+劳
+劵
+劾
+势
+勁
+勃
+勅
+勇
+勉
+勋
+勐
+勑
+勒
+動
+勘
+務
+勝
+募
+勤
+勰
+勺
+勾
+勿
+匀
+包
+匆
+匈
+匋
+匍
+匐
+匕
+化
+北
+匙
+匝
+匠
+匡
+匣
+匪
+匮
+匯
+匹
+区
+医
+匾
+匿
+區
+十
+千
+卅
+升
+午
+卉
+半
+华
+协
+卑
+卒
+卓
+協
+单
+卖
+南
+博
+卜
+卞
+卟
+占
+卡
+卢
+卤
+卦
+卧
+卫
+卯
+印
+危
+即
+却
+卵
+卷
+卸
+卿
+厂
+厄
+厅
+历
+厉
+压
+厌
+厍
+厕
+厘
+厚
+厝
+原
+厢
+厥
+厦
+厨
+厩
+厮
+去
+县
+叁
+参
+參
+又
+叉
+及
+友
+双
+反
+发
+叔
+取
+受
+变
+叙
+叛
+叟
+叠
+叢
+口
+古
+句
+另
+叨
+叩
+只
+叫
+召
+叭
+叮
+可
+台
+叱
+史
+右
+叶
+号
+司
+叹
+叻
+叼
+叽
+吁
+吃
+各
+吆
+合
+吉
+吊
+吋
+同
+名
+后
+吏
+吐
+向
+吒
+吓
+吕
+吖
+吗
+君
+吝
+吞
+吟
+吠
+吡
+否
+吧
+吨
+吩
+含
+听
+吭
+吮
+启
+吱
+吲
+吴
+吵
+吸
+吹
+吻
+吼
+吾
+呀
+呃
+呆
+呈
+告
+呋
+呐
+呓
+呕
+呗
+员
+呙
+呛
+呜
+呢
+呤
+呦
+周
+呱
+味
+呵
+呷
+呸
+呻
+呼
+命
+咀
+咂
+咄
+咅
+咆
+咋
+和
+咎
+咏
+咐
+咒
+咔
+咕
+咖
+咙
+咚
+咛
+咣
+咤
+咦
+咧
+咨
+咩
+咪
+咫
+咬
+咭
+咯
+咱
+咳
+咸
+咻
+咽
+咿
+哀
+品
+哄
+哆
+哇
+哈
+哉
+哌
+响
+哎
+哑
+哒
+哓
+哔
+哗
+哚
+哝
+哞
+哟
+員
+哥
+哦
+哧
+哨
+哩
+哪
+哭
+哮
+哲
+哺
+哼
+哽
+唁
+唆
+唇
+唉
+唏
+唐
+唑
+唔
+唘
+唛
+唠
+唢
+唤
+唧
+唬
+售
+唯
+唰
+唱
+唳
+唷
+唻
+唾
+啃
+啄
+商
+啉
+啊
+問
+啓
+啕
+啖
+啜
+啟
+啡
+啤
+啥
+啦
+啧
+啪
+啫
+啬
+啮
+啰
+啵
+啶
+啸
+啻
+啼
+啾
+喀
+喂
+喃
+善
+喆
+喇
+喉
+喊
+喋
+喏
+喔
+喘
+喙
+喜
+喝
+喧
+喫
+喬
+單
+喰
+喱
+喳
+喵
+喷
+喹
+喻
+喽
+嗅
+嗉
+嗑
+嗒
+嗓
+嗔
+嗖
+嗜
+嗝
+嗞
+嗟
+嗡
+嗣
+嗤
+嗦
+嗨
+嗪
+嗬
+嗮
+嗯
+嗲
+嗳
+嗷
+嗽
+嘀
+嘈
+嘉
+嘌
+嘎
+嘘
+嘛
+嘞
+嘟
+嘢
+嘣
+嘤
+嘧
+嘬
+嘱
+嘲
+嘴
+嘶
+嘹
+嘻
+嘿
+噌
+噎
+噗
+噜
+噢
+噤
+器
+噩
+噪
+噬
+噱
+噶
+噻
+噼
+嚎
+嚏
+嚒
+嚓
+嚞
+嚣
+嚷
+嚼
+囊
+囍
+囔
+囗
+囚
+四
+回
+囟
+因
+囡
+团
+囤
+园
+困
+囱
+围
+囹
+固
+国
+图
+囿
+圃
+圆
+圈
+國
+圍
+園
+圖
+圗
+團
+圜
+土
+圣
+圧
+在
+圩
+圭
+地
+圳
+圹
+场
+圻
+圾
+址
+坂
+均
+坊
+坋
+坍
+坎
+坏
+坐
+坑
+块
+坚
+坛
+坝
+坞
+坟
+坠
+坡
+坤
+坦
+坨
+坩
+坪
+坫
+坭
+坯
+坳
+坶
+坷
+坻
+垂
+垃
+垄
+型
+垌
+垒
+垓
+垚
+垛
+垟
+垠
+垡
+垢
+垣
+垦
+垩
+垫
+垭
+垮
+垸
+埂
+埃
+埋
+埌
+城
+埔
+埕
+埗
+埚
+域
+埠
+埭
+執
+培
+基
+堀
+堂
+堃
+堆
+堇
+堑
+堕
+堡
+堤
+堪
+堰
+報
+場
+堵
+堷
+塊
+塌
+塍
+塑
+塔
+塗
+塘
+塞
+填
+塬
+塾
+境
+墅
+墉
+墊
+墒
+墓
+墙
+增
+墟
+墨
+墩
+壁
+壅
+壇
+壐
+壑
+壓
+壕
+壤
+士
+壬
+壮
+声
+壳
+壴
+壶
+壹
+壽
+夀
+处
+夆
+备
+夋
+夌
+复
+夏
+夔
+夕
+外
+夙
+多
+夜
+够
+夢
+夥
+大
+天
+太
+夫
+夬
+夭
+央
+夯
+失
+头
+夷
+夸
+夹
+夺
+奂
+奄
+奇
+奈
+奉
+奋
+奎
+奏
+契
+奓
+奔
+奕
+奖
+套
+奘
+奚
+奠
+奢
+奥
+奧
+女
+奴
+奶
+奸
+她
+好
+如
+妃
+妄
+妆
+妇
+妈
+妊
+妍
+妒
+妓
+妖
+妗
+妙
+妝
+妞
+妤
+妥
+妨
+妩
+妪
+妫
+妮
+妯
+妲
+妳
+妹
+妻
+妾
+姆
+姊
+始
+姌
+姐
+姑
+姒
+姓
+委
+姗
+姘
+姚
+姜
+姝
+姣
+姥
+姨
+姬
+姹
+姻
+姿
+威
+娃
+娄
+娅
+娆
+娇
+娈
+娉
+娌
+娑
+娓
+娘
+娜
+娟
+娠
+娣
+娥
+娩
+娱
+娲
+娴
+娶
+娼
+娽
+婀
+婆
+婉
+婊
+婕
+婚
+婢
+婦
+婧
+婪
+婴
+婵
+婶
+婷
+婺
+婿
+媄
+媒
+媚
+媛
+媲
+媳
+媽
+媾
+嫁
+嫂
+嫉
+嫌
+嫒
+嫔
+嫖
+嫚
+嫡
+嫣
+嫦
+嫩
+嬉
+嬗
+嬰
+嬴
+嬷
+孃
+子
+孑
+孓
+孔
+孕
+孖
+字
+存
+孙
+孚
+孛
+孜
+孝
+孟
+孢
+季
+孤
+学
+孩
+孪
+孫
+孬
+孰
+孳
+孵
+學
+孺
+孽
+宁
+它
+宄
+宅
+宇
+守
+安
+宋
+完
+宏
+宓
+宕
+宗
+官
+宙
+定
+宛
+宜
+宝
+实
+宠
+审
+客
+宣
+室
+宥
+宦
+宪
+宫
+宰
+害
+宴
+宵
+家
+宸
+容
+宽
+宾
+宿
+寂
+寄
+寅
+密
+寇
+富
+寐
+寒
+寓
+寕
+寝
+寞
+察
+寡
+寤
+寥
+實
+寧
+寨
+寫
+寬
+寮
+寰
+寳
+寵
+寶
+寸
+对
+寺
+寻
+导
+対
+寿
+封
+専
+尃
+射
+将
+將
+專
+尉
+尊
+對
+導
+小
+少
+尓
+尔
+尕
+尖
+尘
+尙
+尚
+尝
+尤
+尧
+尬
+就
+尴
+尷
+尸
+尹
+尺
+尻
+尼
+尽
+尾
+尿
+局
+屁
+层
+居
+屆
+屈
+屉
+届
+屋
+屌
+屎
+屏
+屐
+屑
+展
+属
+屠
+屡
+層
+履
+屯
+山
+屹
+屿
+岁
+岂
+岌
+岐
+岑
+岔
+岖
+岗
+岘
+岙
+岚
+岛
+岢
+岩
+岫
+岭
+岱
+岳
+岷
+岸
+岽
+岿
+峁
+峄
+峋
+峒
+峙
+峡
+峤
+峥
+峦
+峨
+峩
+峪
+峭
+峯
+峰
+島
+峻
+崂
+崃
+崆
+崇
+崋
+崎
+崔
+崖
+崛
+崤
+崧
+崩
+崬
+崭
+崮
+崴
+崽
+嵇
+嵊
+嵋
+嵌
+嵐
+嵘
+嵩
+嵬
+嵯
+嶂
+嶙
+嶪
+嶺
+嶽
+巅
+巍
+川
+州
+巡
+巢
+工
+左
+巧
+巨
+巩
+巫
+差
+巯
+己
+已
+巳
+巴
+巷
+巽
+巾
+币
+市
+布
+帅
+帆
+师
+希
+帏
+帐
+帑
+帕
+帖
+帘
+帚
+帛
+帜
+帝
+帥
+带
+帧
+師
+席
+帮
+帯
+帰
+帶
+帷
+常
+帼
+帽
+幂
+幄
+幅
+幌
+幔
+幕
+幚
+幛
+幡
+幢
+干
+平
+年
+并
+幸
+幹
+幺
+幻
+幼
+幽
+广
+庄
+庆
+庇
+床
+序
+庐
+库
+应
+底
+庖
+店
+庙
+庚
+府
+庞
+废
+庠
+庥
+度
+座
+庫
+庭
+庵
+庶
+康
+庸
+庹
+庾
+廉
+廊
+廓
+廖
+廚
+廠
+廣
+廰
+廳
+延
+廷
+建
+廿
+开
+异
+弃
+弄
+弈
+弊
+弋
+弍
+式
+弑
+弓
+引
+弗
+弘
+弛
+弟
+张
+弥
+弦
+弧
+弩
+弭
+弯
+弱
+張
+強
+弹
+强
+弼
+彈
+彐
+归
+当
+录
+彗
+彝
+形
+彤
+彦
+彧
+彩
+彪
+彬
+彭
+彰
+影
+彳
+彷
+役
+彻
+彼
+往
+征
+径
+待
+徇
+很
+徉
+徊
+律
+後
+徐
+徒
+徕
+得
+徘
+徙
+徜
+御
+徨
+復
+循
+微
+徳
+徵
+德
+徹
+徽
+心
+必
+忆
+忌
+忍
+忏
+忐
+忑
+忒
+忖
+志
+忘
+忙
+応
+忠
+忡
+忧
+快
+忱
+念
+忻
+忽
+忿
+怀
+态
+怂
+怅
+怆
+怎
+怏
+怒
+怔
+怕
+怖
+怜
+思
+怠
+怡
+急
+怦
+性
+怨
+怪
+怯
+总
+恂
+恃
+恋
+恍
+恐
+恒
+恕
+恙
+恢
+恣
+恤
+恨
+恩
+恪
+恬
+恭
+息
+恰
+恳
+恶
+恸
+恹
+恺
+恻
+恼
+恽
+恿
+悄
+悅
+悉
+悌
+悍
+悔
+悖
+悚
+悟
+悠
+患
+悦
+您
+悬
+悯
+悱
+悲
+悴
+悵
+悶
+悸
+悻
+悼
+情
+惆
+惊
+惋
+惑
+惕
+惘
+惚
+惜
+惟
+惠
+惦
+惧
+惨
+惩
+惫
+惬
+惭
+惮
+惯
+惰
+想
+惴
+惶
+惹
+惺
+愁
+愈
+愉
+意
+愕
+愚
+愛
+感
+愣
+愤
+愧
+愫
+愿
+慈
+態
+慌
+慎
+慑
+慕
+慢
+慧
+慨
+慰
+慵
+慶
+慷
+憋
+憎
+憔
+憧
+憨
+憩
+憬
+憶
+憾
+懂
+懈
+應
+懊
+懋
+懒
+懦
+懵
+懷
+懿
+戀
+戈
+戊
+戋
+戌
+戍
+戎
+戏
+成
+我
+戒
+或
+战
+戚
+戛
+戟
+戢
+截
+戬
+戮
+戳
+戴
+戶
+户
+戾
+房
+所
+扁
+扇
+扈
+扉
+手
+才
+扎
+扑
+扒
+打
+扔
+托
+扙
+扛
+扞
+扣
+扦
+执
+扩
+扪
+扫
+扬
+扭
+扮
+扯
+扰
+扳
+扶
+批
+扼
+找
+承
+技
+抄
+抉
+把
+抑
+抒
+抓
+投
+抖
+抗
+折
+抚
+抛
+抠
+抡
+抢
+护
+报
+抨
+披
+抬
+抱
+抵
+抹
+抻
+押
+抽
+抿
+拂
+拄
+担
+拆
+拇
+拈
+拉
+拌
+拍
+拎
+拐
+拒
+拓
+拔
+拖
+拗
+拘
+拙
+拚
+招
+拜
+拟
+拢
+拣
+拥
+拦
+拧
+拨
+择
+括
+拭
+拮
+拯
+拱
+拳
+拴
+拷
+拼
+拽
+拾
+拿
+持
+挂
+指
+按
+挎
+挑
+挖
+挚
+挛
+挝
+挞
+挟
+挠
+挡
+挣
+挤
+挥
+挨
+挪
+挫
+振
+挲
+挺
+挽
+捂
+捅
+捆
+捉
+捋
+捌
+捍
+捎
+捏
+捐
+捕
+捞
+损
+捡
+换
+捣
+捧
+据
+捲
+捶
+捷
+捺
+捻
+掀
+掂
+掇
+授
+掉
+掊
+掌
+掏
+掐
+排
+掖
+掘
+掠
+探
+掣
+接
+控
+推
+掩
+措
+掬
+掰
+掳
+掴
+掷
+掸
+掺
+揉
+揍
+描
+提
+插
+揖
+揚
+換
+握
+揣
+揩
+揪
+揭
+援
+揸
+揽
+搀
+搁
+搂
+搅
+損
+搏
+搐
+搓
+搔
+搜
+搞
+搪
+搬
+搭
+搴
+携
+搽
+摁
+摄
+摆
+摇
+摈
+摊
+摒
+摔
+摘
+摞
+摧
+摩
+摸
+摹
+撂
+撃
+撅
+撇
+撑
+撒
+撕
+撞
+撤
+撩
+撬
+播
+撮
+撰
+撵
+撷
+撸
+撻
+撼
+擀
+擂
+擅
+操
+擎
+擒
+擘
+據
+擞
+擢
+擦
+攀
+攒
+攘
+攝
+攥
+攫
+支
+收
+攸
+改
+攻
+放
+政
+故
+效
+敌
+敏
+救
+敕
+敖
+教
+敛
+敝
+敞
+敟
+敢
+散
+敦
+敬
+数
+敲
+整
+敷
+數
+文
+斋
+斌
+斐
+斑
+斓
+斗
+料
+斛
+斜
+斟
+斡
+斤
+斥
+斧
+斩
+断
+斯
+新
+方
+於
+施
+旁
+旅
+旋
+旌
+旎
+族
+旖
+旗
+无
+既
+日
+旦
+旧
+旨
+早
+旬
+旭
+旮
+旯
+旱
+时
+旷
+旸
+旺
+昀
+昂
+昆
+昇
+昊
+昌
+明
+昏
+易
+昔
+昕
+昙
+昝
+星
+映
+春
+昧
+昨
+昭
+是
+昱
+昴
+昵
+昶
+昼
+显
+晁
+時
+晃
+晋
+晌
+晏
+晒
+晓
+晔
+晕
+晖
+晗
+晚
+晞
+晟
+晤
+晦
+晧
+晨
+普
+景
+晰
+晳
+晴
+晶
+晷
+智
+晾
+暂
+暃
+暄
+暇
+暑
+暖
+暗
+暢
+暧
+暨
+暮
+暴
+暹
+曉
+曙
+曜
+曝
+曦
+曬
+曰
+曲
+曳
+更
+書
+曹
+曼
+曾
+替
+最
+會
+月
+有
+朋
+服
+朐
+朔
+朕
+朗
+望
+朝
+期
+朦
+木
+未
+末
+本
+札
+术
+朱
+朴
+朵
+机
+朽
+朿
+杀
+杂
+权
+杆
+杈
+杉
+李
+杏
+材
+村
+杖
+杜
+杞
+束
+杠
+条
+来
+杨
+杩
+杭
+杯
+杰
+東
+杲
+杳
+杵
+杷
+松
+板
+极
+构
+枇
+枉
+枋
+析
+枕
+林
+枚
+果
+枝
+枞
+枢
+枣
+枪
+枫
+枭
+枯
+枰
+枱
+枳
+架
+枷
+枸
+枼
+柄
+柏
+某
+柑
+柒
+染
+柔
+柘
+柚
+柜
+柞
+柠
+查
+柩
+柬
+柯
+柱
+柳
+柴
+柿
+栀
+栅
+标
+栈
+栉
+栋
+栎
+栏
+树
+栓
+栖
+栗
+校
+栩
+株
+样
+核
+根
+格
+栽
+栾
+桀
+桁
+桂
+桃
+桅
+框
+案
+桉
+桌
+桎
+桐
+桑
+桓
+桔
+桖
+桠
+桡
+桢
+档
+桥
+桦
+桧
+桨
+桩
+桶
+桷
+桼
+梁
+梅
+梆
+梏
+梓
+梗
+條
+梢
+梦
+梧
+梨
+梭
+梯
+械
+梳
+梵
+检
+棉
+棋
+棍
+棒
+棕
+棘
+棚
+棛
+棟
+棠
+棣
+森
+棰
+棱
+棵
+棺
+棽
+椁
+椅
+椋
+植
+椎
+椐
+椒
+椤
+椭
+椰
+椴
+椹
+椽
+椿
+楂
+楊
+楓
+楔
+楗
+楚
+楝
+楞
+楠
+楣
+楦
+業
+極
+楷
+楸
+楹
+楼
+楽
+楿
+概
+榄
+榆
+榈
+榉
+榔
+榕
+榛
+榜
+榞
+榧
+榨
+榫
+榭
+榮
+榴
+榷
+榻
+構
+槌
+槎
+槐
+様
+槛
+槟
+槲
+槽
+槿
+樂
+樊
+樓
+標
+樟
+模
+樣
+樨
+横
+樱
+樵
+樸
+樹
+樽
+樾
+橄
+橇
+橋
+橘
+橙
+機
+橡
+橪
+橱
+橹
+橼
+檀
+檐
+檔
+檗
+檩
+檬
+櫥
+權
+欠
+次
+欢
+欣
+欧
+欲
+欺
+款
+歆
+歇
+歉
+歌
+歐
+歙
+歡
+止
+正
+此
+步
+武
+歧
+歪
+歷
+歸
+歹
+死
+歼
+殁
+殃
+殆
+殇
+殉
+殊
+残
+殒
+殖
+殚
+殡
+殴
+段
+殷
+殺
+殼
+殿
+毁
+毂
+毅
+毋
+母
+每
+毒
+毓
+比
+毕
+毗
+毙
+毛
+毡
+毫
+毯
+毽
+氏
+氐
+民
+氓
+气
+氖
+気
+氘
+氙
+氚
+氛
+氟
+氡
+氢
+氣
+氤
+氦
+氧
+氨
+氩
+氪
+氮
+氯
+氰
+氲
+水
+永
+氽
+汀
+汁
+求
+汇
+汉
+汊
+汐
+汕
+汗
+汛
+汝
+汞
+江
+池
+污
+汤
+汨
+汩
+汪
+汰
+汲
+汴
+汶
+汹
+決
+汽
+汾
+沁
+沂
+沃
+沅
+沈
+沉
+沌
+沏
+沐
+沓
+沔
+沙
+沛
+沟
+没
+沣
+沤
+沥
+沦
+沧
+沪
+沫
+沭
+沮
+沱
+河
+沸
+油
+治
+沼
+沽
+沾
+沿
+泄
+泅
+泉
+泊
+泌
+泓
+泔
+法
+泗
+泛
+泞
+泠
+泡
+波
+泣
+泥
+注
+泩
+泪
+泫
+泮
+泯
+泰
+泱
+泳
+泵
+泷
+泸
+泺
+泻
+泼
+泽
+泾
+洁
+洄
+洋
+洒
+洗
+洙
+洛
+洞
+津
+洪
+洮
+洱
+洲
+洳
+洵
+洺
+活
+洼
+洽
+派
+流
+浃
+浅
+浆
+浇
+浈
+浊
+测
+浍
+济
+浏
+浐
+浑
+浒
+浓
+浔
+浙
+浚
+浜
+浠
+浣
+浥
+浦
+浩
+浪
+浮
+浴
+海
+浸
+浼
+涂
+涅
+涇
+消
+涉
+涌
+涎
+涓
+涔
+涕
+涛
+涝
+涞
+涟
+涠
+涡
+涣
+涤
+润
+涧
+涨
+涩
+涪
+涫
+涮
+涯
+液
+涵
+涸
+涿
+淀
+淄
+淅
+淆
+淇
+淋
+淌
+淑
+淖
+淘
+淙
+淚
+淝
+淞
+淡
+淤
+淦
+淨
+淩
+淫
+淬
+淮
+深
+淳
+混
+淹
+添
+淼
+清
+渊
+渌
+渍
+渎
+渐
+渑
+渔
+渗
+渚
+渝
+渠
+渡
+渣
+渤
+渥
+温
+測
+渭
+港
+渲
+渴
+游
+渺
+湃
+湄
+湍
+湓
+湖
+湘
+湛
+湟
+湫
+湮
+湯
+湾
+湿
+溃
+溅
+溆
+溉
+溏
+源
+溜
+溝
+溟
+溢
+溥
+溧
+溪
+溫
+溯
+溱
+溴
+溶
+溺
+滁
+滂
+滇
+滋
+滏
+滑
+滓
+滔
+滕
+滘
+滙
+滚
+滞
+滟
+满
+滢
+滤
+滥
+滦
+滨
+滩
+滬
+滴
+滿
+漁
+漂
+漆
+漉
+漏
+漓
+演
+漕
+漠
+漢
+漩
+漪
+漫
+漯
+漱
+漳
+漾
+潇
+潋
+潍
+潔
+潘
+潜
+潞
+潢
+潤
+潦
+潭
+潮
+潴
+潸
+潺
+潼
+澄
+澈
+澍
+澎
+澜
+澡
+澤
+澧
+澳
+澶
+澹
+激
+濂
+濃
+濉
+濑
+濒
+濕
+濞
+濟
+濠
+濡
+濤
+濮
+濯
+濱
+瀑
+瀘
+瀚
+瀛
+瀣
+瀵
+瀹
+灌
+灏
+灞
+灡
+灣
+火
+灭
+灯
+灰
+灵
+灶
+灸
+灼
+灾
+灿
+炀
+炅
+炉
+炊
+炎
+炒
+炔
+炕
+炖
+炙
+炜
+炝
+炫
+炬
+炭
+炮
+炯
+炳
+炸
+点
+為
+炼
+炽
+烁
+烂
+烃
+烈
+烊
+烏
+烘
+烙
+烛
+烜
+烟
+烤
+烦
+烧
+烨
+烩
+烫
+烬
+热
+烯
+烷
+烹
+烺
+烽
+焉
+焊
+焕
+焖
+焗
+焘
+焙
+焚
+無
+焦
+焯
+焰
+焱
+然
+煅
+煊
+煌
+煎
+煖
+煙
+煜
+煞
+煤
+煥
+煦
+照
+煨
+煮
+煲
+煸
+煽
+熄
+熊
+熏
+熔
+熘
+熙
+熟
+熠
+熥
+熨
+熬
+熱
+熳
+熵
+熹
+燁
+燃
+燈
+燎
+燒
+燕
+燘
+燙
+燚
+燜
+營
+燥
+燦
+燧
+燮
+燴
+爆
+爍
+爐
+爨
+爪
+爬
+爭
+爯
+爰
+爱
+爲
+爵
+父
+爷
+爸
+爹
+爺
+爻
+爽
+爾
+片
+版
+牌
+牍
+牒
+牙
+牛
+牟
+牡
+牢
+牦
+牧
+物
+牯
+牲
+牵
+特
+牺
+犀
+犁
+犇
+犊
+犍
+犟
+犬
+犯
+状
+犷
+犸
+犹
+狀
+狂
+狄
+狈
+狍
+狐
+狒
+狗
+狙
+狞
+狠
+狡
+狩
+独
+狭
+狮
+狰
+狱
+狸
+狼
+猎
+猕
+猖
+猗
+猛
+猜
+猝
+猥
+猩
+猪
+猫
+猬
+献
+猴
+猷
+猾
+猿
+獅
+獐
+獒
+獗
+獠
+獨
+獭
+獲
+獸
+獻
+玄
+率
+玉
+王
+玑
+玖
+玛
+玟
+玥
+玩
+玫
+玮
+环
+现
+玲
+玳
+玷
+玺
+玻
+珀
+珂
+珈
+珉
+珊
+珍
+珏
+珐
+珑
+珙
+珞
+珠
+珣
+珥
+珩
+班
+珲
+珺
+現
+球
+琅
+理
+琉
+琊
+琏
+琐
+琚
+琛
+琢
+琥
+琦
+琨
+琪
+琬
+琮
+琯
+琰
+琳
+琴
+琵
+琶
+琼
+瑁
+瑄
+瑕
+瑗
+瑙
+瑚
+瑛
+瑜
+瑞
+瑟
+瑩
+瑪
+瑭
+瑰
+瑶
+瑷
+瑾
+璀
+璁
+璃
+璇
+璈
+璋
+璎
+璐
+璜
+璞
+璟
+璧
+璨
+璩
+環
+璺
+瓒
+瓜
+瓠
+瓢
+瓣
+瓤
+瓦
+瓮
+瓯
+瓴
+瓶
+瓷
+瓿
+甄
+甘
+甙
+甚
+甜
+生
+產
+産
+甥
+用
+甩
+甫
+甬
+甭
+甯
+田
+由
+甲
+申
+电
+男
+甸
+甹
+町
+画
+甾
+畅
+畈
+界
+畏
+畐
+畔
+留
+畜
+略
+畦
+番
+畫
+畲
+畴
+畵
+當
+畸
+畹
+畿
+疃
+疆
+疌
+疏
+疑
+疖
+疗
+疙
+疚
+疝
+疟
+疡
+疣
+疤
+疥
+疫
+疮
+疯
+疱
+疲
+疳
+疵
+疸
+疹
+疼
+疽
+疾
+痂
+病
+症
+痈
+痉
+痊
+痒
+痔
+痕
+痘
+痛
+痞
+痠
+痢
+痣
+痤
+痧
+痨
+痪
+痫
+痰
+痱
+痴
+痹
+痼
+痿
+瘀
+瘁
+瘊
+瘘
+瘙
+瘟
+瘠
+瘢
+瘤
+瘦
+瘩
+瘪
+瘫
+瘴
+瘸
+瘾
+瘿
+療
+癌
+癔
+癖
+癜
+癞
+癣
+癫
+癸
+登
+發
+白
+百
+皂
+的
+皆
+皇
+皈
+皋
+皎
+皑
+皓
+皖
+皙
+皮
+皱
+皲
+皴
+皺
+皿
+盂
+盅
+盆
+盈
+益
+盎
+盏
+盐
+监
+盒
+盔
+盖
+盗
+盘
+盛
+盜
+盟
+盤
+盥
+目
+盯
+盱
+盲
+直
+相
+盹
+盼
+盾
+省
+眈
+眉
+看
+眙
+眞
+真
+眠
+眦
+眨
+眩
+眭
+眯
+眶
+眷
+眸
+眺
+眼
+着
+睁
+睇
+睐
+睑
+睛
+睡
+睢
+督
+睦
+睨
+睫
+睬
+睹
+睽
+睾
+睿
+瞄
+瞅
+瞌
+瞎
+瞑
+瞒
+瞓
+瞟
+瞠
+瞥
+瞧
+瞩
+瞪
+瞬
+瞭
+瞰
+瞳
+瞻
+瞿
+矗
+矛
+矜
+矞
+矢
+矣
+知
+矩
+矫
+短
+矮
+石
+矶
+矸
+矽
+矾
+矿
+砀
+码
+砂
+砌
+砍
+砒
+研
+砖
+砚
+砜
+砝
+砣
+砥
+砦
+砧
+砬
+砭
+砰
+破
+砷
+砸
+砺
+砻
+砼
+砾
+础
+硂
+硅
+硒
+硕
+硖
+硚
+硝
+硪
+硫
+硬
+确
+硼
+碁
+碇
+碉
+碌
+碍
+碎
+碑
+碓
+碗
+碘
+碚
+碟
+碣
+碧
+碩
+碰
+碱
+碲
+碳
+碴
+碶
+確
+碼
+碾
+磁
+磅
+磊
+磋
+磐
+磕
+磚
+磨
+磬
+磴
+磷
+磺
+礁
+礴
+示
+礼
+社
+祀
+祁
+祈
+祉
+祎
+祐
+祖
+祗
+祙
+祚
+祛
+祜
+祝
+神
+祟
+祠
+祢
+祥
+票
+祭
+祯
+祷
+祸
+祺
+禀
+禁
+禄
+禅
+福
+禦
+禧
+禮
+禹
+禺
+离
+禽
+禾
+秀
+私
+秃
+秆
+秉
+秋
+种
+科
+秒
+秕
+秘
+租
+秣
+秤
+秦
+秧
+秩
+秭
+积
+称
+秸
+移
+秽
+稀
+程
+稍
+税
+稔
+稗
+稚
+稞
+稠
+稣
+種
+稳
+稷
+稹
+稻
+稼
+稽
+稿
+穂
+穆
+穗
+穩
+穰
+穴
+穵
+究
+穷
+穸
+穹
+空
+穿
+突
+窃
+窄
+窈
+窍
+窑
+窒
+窕
+窖
+窗
+窘
+窜
+窝
+窟
+窠
+窥
+窦
+窨
+窿
+立
+竖
+站
+竞
+竟
+章
+竣
+童
+竭
+端
+競
+竹
+竺
+竽
+竿
+笃
+笆
+笈
+笋
+笏
+笑
+笔
+笕
+笙
+笛
+笠
+符
+笨
+笫
+第
+笺
+笼
+筆
+等
+筋
+筏
+筐
+筑
+筒
+答
+策
+筛
+筝
+筠
+筱
+筵
+筷
+筹
+签
+简
+箍
+箐
+箔
+箕
+算
+管
+箩
+箫
+箬
+箭
+箱
+箴
+箸
+節
+篁
+篆
+篇
+築
+篑
+篓
+篙
+篝
+篡
+篦
+篪
+篮
+篱
+篷
+簇
+簋
+簌
+簡
+簧
+簪
+簸
+簿
+籁
+籍
+籣
+籤
+米
+类
+籼
+籽
+粉
+粑
+粒
+粕
+粗
+粘
+粟
+粢
+粤
+粥
+粧
+粪
+粮
+粱
+粲
+粳
+粵
+粹
+粼
+粽
+精
+粿
+糁
+糅
+糊
+糍
+糕
+糖
+糗
+糙
+糜
+糝
+糟
+糠
+糯
+系
+紀
+約
+紅
+紊
+紋
+納
+純
+紗
+級
+素
+索
+紧
+紫
+累
+細
+終
+組
+結
+絡
+給
+絮
+統
+絲
+絶
+綁
+綉
+經
+綦
+綫
+維
+綱
+網
+緑
+線
+緣
+緯
+緻
+縣
+縫
+縮
+繁
+織
+繡
+繪
+繹
+纂
+纛
+纠
+红
+纣
+纤
+约
+级
+纨
+纪
+纫
+纬
+纭
+纯
+纰
+纱
+纲
+纳
+纵
+纶
+纷
+纸
+纹
+纺
+纽
+纾
+线
+绀
+绁
+绂
+练
+组
+绅
+细
+织
+终
+绉
+绊
+绋
+绌
+绍
+绎
+经
+绐
+绑
+绒
+结
+绔
+绕
+绗
+绘
+给
+绚
+绛
+络
+绝
+绞
+统
+绠
+绡
+绢
+绣
+绥
+绦
+继
+绨
+绩
+绪
+绫
+续
+绮
+绯
+绰
+绱
+绲
+绳
+维
+绵
+绶
+绷
+绸
+绺
+绻
+综
+绽
+绾
+绿
+缀
+缁
+缂
+缃
+缄
+缅
+缆
+缇
+缈
+缉
+缎
+缐
+缓
+缔
+缕
+编
+缘
+缙
+缚
+缜
+缝
+缠
+缢
+缤
+缥
+缦
+缨
+缩
+缪
+缫
+缬
+缭
+缮
+缰
+缴
+缸
+缺
+罂
+罄
+罐
+网
+罔
+罕
+罗
+罘
+罚
+罡
+罢
+罩
+罪
+置
+罱
+署
+罹
+羁
+羅
+羊
+羌
+美
+羔
+羙
+羚
+羞
+羟
+羡
+群
+羧
+義
+羯
+羰
+羲
+羸
+羹
+羽
+羿
+翁
+翅
+翊
+翌
+翎
+翏
+習
+翔
+翘
+翟
+翠
+翡
+翥
+翦
+翩
+翰
+翱
+翳
+翻
+翼
+耀
+老
+考
+耄
+者
+耆
+而
+耍
+耐
+耑
+耒
+耕
+耗
+耘
+耙
+耜
+耦
+耧
+耱
+耳
+耵
+耶
+耷
+耸
+耻
+耽
+耿
+聂
+聆
+聊
+聋
+职
+联
+聖
+聘
+聚
+聞
+聨
+聪
+聯
+聲
+職
+聿
+肃
+肄
+肆
+肇
+肉
+肋
+肌
+肓
+肖
+肘
+肚
+肛
+肝
+肟
+肠
+股
+肢
+肤
+肥
+肩
+肪
+肮
+肯
+肱
+育
+肴
+肺
+肼
+肽
+肾
+肿
+胀
+胁
+胃
+胄
+胆
+背
+胍
+胎
+胖
+胗
+胚
+胛
+胜
+胞
+胡
+胤
+胥
+胧
+胪
+胫
+胬
+胭
+胯
+胰
+胱
+胳
+胴
+胶
+胸
+胺
+能
+脂
+脆
+脉
+脊
+脍
+脏
+脐
+脑
+脒
+脓
+脖
+脘
+脚
+脯
+脱
+脲
+脸
+脾
+腆
+腈
+腊
+腋
+腌
+腐
+腑
+腓
+腔
+腕
+腥
+腦
+腩
+腭
+腮
+腰
+腱
+腴
+腸
+腹
+腺
+腻
+腼
+腾
+腿
+膀
+膈
+膊
+膏
+膘
+膚
+膛
+膜
+膝
+膠
+膦
+膨
+膳
+膺
+膻
+臀
+臂
+臃
+臆
+臊
+臣
+臧
+臨
+自
+臭
+至
+致
+臺
+臻
+臼
+臾
+舀
+舅
+舆
+與
+興
+舊
+舌
+舍
+舒
+舔
+舖
+舛
+舜
+舞
+舟
+航
+舫
+般
+舰
+舱
+舵
+舶
+舷
+舸
+船
+艄
+艇
+艘
+艦
+艮
+良
+艰
+色
+艳
+艶
+艷
+艹
+艺
+艾
+艿
+节
+芃
+芊
+芋
+芍
+芎
+芒
+芗
+芘
+芙
+芜
+芝
+芡
+芥
+芦
+芩
+芪
+芫
+芬
+芭
+芮
+芯
+花
+芳
+芷
+芸
+芹
+芽
+芾
+苁
+苄
+苇
+苋
+苍
+苎
+苏
+苑
+苒
+苓
+苔
+苕
+苗
+苘
+苛
+苜
+苝
+苞
+苟
+苡
+苣
+若
+苦
+苪
+苫
+苯
+英
+苴
+苷
+苹
+茁
+茂
+范
+茄
+茅
+茆
+茉
+茌
+茎
+茏
+茗
+茜
+茧
+茨
+茫
+茬
+茭
+茯
+茱
+茴
+茵
+茶
+茸
+茹
+茼
+荀
+荃
+荆
+草
+荏
+荐
+荒
+荔
+荘
+荚
+荞
+荟
+荠
+荡
+荣
+荤
+荥
+荧
+荨
+荪
+荫
+荬
+荭
+药
+荷
+荸
+荻
+荼
+荽
+莅
+莆
+莉
+莊
+莎
+莒
+莓
+莘
+莜
+莞
+莠
+莨
+莪
+莫
+莱
+莲
+莳
+莴
+获
+莹
+莺
+莼
+莽
+菀
+菁
+菅
+菇
+菊
+菌
+菏
+菓
+菖
+菘
+菜
+菟
+菠
+菡
+菩
+菪
+華
+菱
+菲
+菽
+萁
+萃
+萄
+萊
+萋
+萌
+萍
+萎
+萘
+萜
+萝
+萤
+营
+萦
+萧
+萨
+萬
+萭
+萱
+萸
+萼
+落
+葆
+葉
+葑
+葒
+著
+葛
+葡
+董
+葩
+葫
+葬
+葭
+葱
+葳
+葵
+葺
+蒂
+蒄
+蒋
+蒌
+蒙
+蒜
+蒟
+蒡
+蒤
+蒲
+蒸
+蒻
+蒽
+蒿
+蓄
+蓉
+蓋
+蓑
+蓓
+蓖
+蓝
+蓟
+蓥
+蓦
+蓬
+蓮
+蓼
+蓿
+蔑
+蔓
+蔗
+蔚
+蔟
+蔡
+蔫
+蔬
+蔷
+蔸
+蔺
+蔻
+蔼
+蔽
+蕃
+蕈
+蕉
+蕊
+蕒
+蕙
+蕤
+蕨
+蕲
+蕴
+蕹
+蕾
+薄
+薇
+薈
+薏
+薛
+薡
+薦
+薩
+薪
+薬
+薯
+薰
+薹
+藁
+藉
+藍
+藏
+藐
+藓
+藔
+藕
+藜
+藝
+藤
+藥
+藩
+藻
+藿
+蘆
+蘇
+蘑
+蘖
+蘭
+蘸
+蘼
+虎
+虏
+虐
+虑
+虔
+處
+虚
+虞
+號
+虢
+虫
+虬
+虱
+虹
+虻
+虽
+虾
+蚀
+蚁
+蚂
+蚊
+蚌
+蚓
+蚕
+蚜
+蚝
+蚣
+蚤
+蚧
+蚨
+蚩
+蚪
+蚬
+蚯
+蚱
+蚴
+蚶
+蛀
+蛆
+蛇
+蛊
+蛋
+蛎
+蛏
+蛐
+蛔
+蛙
+蛛
+蛟
+蛤
+蛭
+蛮
+蛰
+蛳
+蛴
+蛹
+蛾
+蜀
+蜂
+蜃
+蜇
+蜈
+蜊
+蜍
+蜒
+蜓
+蜕
+蜗
+蜘
+蜚
+蜛
+蜜
+蜡
+蜥
+蜱
+蜴
+蜷
+蜻
+蜿
+蝇
+蝈
+蝉
+蝌
+蝎
+蝗
+蝙
+蝠
+蝴
+蝶
+蝽
+蝾
+螂
+螃
+螈
+融
+螟
+螨
+螯
+螳
+螺
+蟀
+蟆
+蟋
+蟑
+蟒
+蟠
+蟹
+蟾
+蠄
+蠓
+蠔
+蠕
+蠖
+蠡
+蠢
+蠹
+血
+衅
+衆
+行
+衍
+術
+衔
+街
+衙
+衛
+衡
+衢
+衣
+补
+表
+衩
+衫
+衬
+衮
+衰
+衲
+衷
+衾
+衿
+袁
+袂
+袄
+袅
+袆
+袈
+袋
+袍
+袒
+袖
+袜
+袤
+袪
+被
+袭
+袱
+裁
+裂
+装
+裆
+裔
+裕
+裘
+裙
+補
+裝
+裟
+裡
+裤
+裨
+裱
+裳
+裴
+裸
+裹
+製
+褀
+褂
+複
+褐
+褒
+褓
+褔
+褚
+褥
+褪
+褰
+褲
+褶
+襁
+襄
+襟
+西
+要
+覃
+覆
+見
+覌
+視
+親
+覺
+觀
+见
+观
+规
+觅
+视
+觇
+览
+觉
+觊
+觎
+觐
+觑
+角
+觞
+解
+触
+觸
+言
+訂
+計
+訊
+訓
+記
+訣
+設
+許
+診
+訾
+詢
+試
+詩
+話
+該
+詹
+誉
+誊
+誌
+誓
+誘
+語
+誠
+説
+課
+調
+諾
+謇
+講
+謢
+謹
+證
+識
+譜
+警
+譬
+護
+讀
+變
+计
+订
+讣
+认
+讥
+讨
+让
+讪
+讫
+训
+议
+讯
+记
+讲
+讳
+讴
+讶
+讷
+许
+讹
+论
+讼
+讽
+设
+访
+诀
+证
+诃
+评
+诅
+识
+诈
+诉
+诊
+诋
+词
+诏
+译
+诒
+试
+诗
+诘
+诙
+诚
+诛
+话
+诞
+诟
+诠
+诡
+询
+诣
+诤
+该
+详
+诧
+诩
+诫
+诬
+语
+误
+诰
+诱
+诲
+说
+诵
+请
+诸
+诺
+读
+诽
+课
+诿
+谀
+谁
+调
+谄
+谅
+谆
+谈
+谊
+谋
+谌
+谍
+谎
+谏
+谐
+谑
+谒
+谓
+谔
+谕
+谗
+谙
+谚
+谛
+谜
+谟
+谡
+谢
+谣
+谤
+谥
+谦
+谧
+谨
+谩
+谪
+谬
+谭
+谯
+谰
+谱
+谴
+谷
+豁
+豆
+豇
+豉
+豊
+豌
+豐
+豕
+豚
+象
+豪
+豫
+豹
+豺
+貂
+貅
+貉
+貊
+貌
+貔
+貝
+負
+財
+貢
+貨
+販
+貭
+貴
+買
+費
+貼
+貿
+賀
+賓
+賠
+賣
+質
+賴
+購
+贈
+贏
+贝
+贞
+负
+贡
+财
+责
+贤
+败
+账
+货
+质
+贩
+贪
+贫
+贬
+购
+贮
+贯
+贰
+贱
+贲
+贴
+贵
+贷
+贸
+费
+贺
+贻
+贼
+贾
+贿
+赁
+赂
+赃
+资
+赅
+赈
+赉
+赊
+赋
+赌
+赎
+赏
+赐
+赓
+赔
+赖
+赘
+赚
+赛
+赝
+赞
+赠
+赡
+赢
+赣
+赤
+赦
+赫
+赭
+走
+赳
+赴
+赵
+赶
+起
+趁
+超
+越
+趋
+趙
+趟
+趣
+足
+趴
+趵
+趸
+趾
+跃
+跄
+跆
+跋
+跌
+跎
+跑
+跖
+跚
+跛
+距
+跟
+跤
+跨
+跪
+路
+跳
+践
+跷
+跺
+跻
+踉
+踊
+踌
+踏
+踝
+踞
+踢
+踩
+踪
+踮
+踱
+踵
+踹
+蹂
+蹄
+蹇
+蹈
+蹉
+蹊
+蹋
+蹑
+蹒
+蹙
+蹦
+蹬
+蹭
+蹲
+蹴
+蹶
+蹼
+蹿
+躁
+躇
+躏
+身
+躬
+躯
+躲
+躺
+車
+軍
+軎
+軒
+軽
+載
+輕
+輝
+輪
+輯
+輸
+轉
+车
+轧
+轨
+轩
+转
+轭
+轮
+软
+轰
+轱
+轲
+轴
+轶
+轻
+轼
+载
+轿
+较
+辄
+辅
+辆
+辈
+辉
+辊
+辋
+辍
+辐
+辑
+输
+辕
+辖
+辗
+辘
+辙
+辛
+辜
+辞
+辟
+辣
+辦
+辧
+辨
+辩
+辫
+辰
+辱
+農
+边
+辽
+达
+迁
+迂
+迄
+迅
+过
+迈
+迎
+运
+近
+返
+还
+这
+进
+远
+违
+连
+迟
+迢
+迤
+迥
+迦
+迩
+迪
+迫
+迭
+迮
+述
+迳
+迷
+迸
+迹
+追
+退
+送
+适
+逃
+逄
+逅
+逆
+选
+逊
+逍
+透
+逐
+逑
+递
+途
+逗
+這
+通
+逛
+逝
+逞
+速
+造
+逢
+連
+逭
+逮
+逯
+進
+逵
+逸
+逹
+逺
+逻
+逼
+逾
+遁
+遂
+遇
+運
+遍
+過
+遏
+遐
+遒
+道
+達
+遗
+遛
+遠
+遢
+遣
+遥
+遨
+適
+遭
+遮
+遴
+遵
+選
+避
+邀
+邂
+邃
+還
+邈
+邋
+邏
+邑
+邓
+邕
+邗
+邛
+邝
+邡
+邢
+那
+邦
+邨
+邪
+邬
+邮
+邯
+邰
+邱
+邳
+邴
+邵
+邸
+邹
+邺
+邻
+郁
+郅
+郊
+郎
+郏
+郑
+郓
+郗
+郛
+郜
+郝
+郡
+郢
+郦
+郧
+部
+郫
+郭
+郯
+郴
+郵
+郸
+都
+郾
+鄂
+鄄
+鄉
+鄙
+鄞
+鄢
+鄣
+鄭
+鄯
+鄱
+鄺
+酉
+酊
+酋
+酌
+配
+酐
+酒
+酗
+酚
+酝
+酞
+酡
+酣
+酥
+酪
+酬
+酮
+酯
+酰
+酱
+酵
+酶
+酷
+酸
+酿
+醇
+醉
+醋
+醌
+醍
+醐
+醒
+醚
+醛
+醜
+醪
+醫
+醬
+醴
+醺
+釀
+采
+釉
+释
+里
+重
+野
+量
+金
+釜
+針
+釣
+鈴
+鈺
+鉅
+鉌
+鉴
+銀
+銘
+銭
+銮
+銷
+鋪
+鋼
+錄
+錦
+錯
+錶
+鍊
+鍋
+鍵
+鎏
+鎖
+鎮
+鎻
+鏖
+鏡
+鏢
+鐘
+鐡
+鐵
+鑒
+鑫
+鑽
+钇
+针
+钉
+钊
+钎
+钏
+钒
+钓
+钕
+钗
+钙
+钚
+钛
+钜
+钝
+钞
+钟
+钠
+钡
+钢
+钣
+钥
+钦
+钧
+钨
+钩
+钬
+钮
+钯
+钰
+钱
+钳
+钴
+钵
+钺
+钻
+钼
+钽
+钾
+钿
+铀
+铁
+铂
+铃
+铄
+铅
+铆
+铈
+铉
+铊
+铋
+铌
+铍
+铎
+铐
+铑
+铒
+铕
+铖
+铛
+铜
+铝
+铞
+铟
+铠
+铡
+铢
+铣
+铤
+铧
+铨
+铪
+铬
+铭
+铮
+铰
+铱
+铲
+铵
+银
+铸
+铺
+链
+铿
+销
+锁
+锂
+锄
+锅
+锆
+锈
+锉
+锋
+锌
+锏
+锐
+锑
+锔
+锕
+锗
+锘
+错
+锚
+锛
+锝
+锟
+锡
+锢
+锣
+锤
+锥
+锦
+锨
+锭
+键
+锯
+锰
+锲
+锴
+锵
+锶
+锷
+锹
+锺
+锻
+锽
+镀
+镁
+镂
+镇
+镉
+镊
+镌
+镍
+镐
+镑
+镒
+镓
+镖
+镗
+镛
+镜
+镝
+镣
+镦
+镧
+镭
+镯
+镰
+镱
+镳
+镶
+長
+长
+門
+閃
+開
+閏
+閑
+間
+閟
+関
+閣
+閥
+閩
+關
+门
+闪
+闫
+闭
+问
+闯
+闰
+闱
+闲
+闳
+间
+闵
+闷
+闸
+闹
+闺
+闻
+闽
+闾
+闿
+阀
+阁
+阂
+阄
+阅
+阆
+阈
+阉
+阊
+阎
+阐
+阑
+阔
+阕
+阖
+阗
+阙
+阚
+阜
+阝
+队
+阡
+阪
+阮
+阱
+防
+阳
+阴
+阵
+阶
+阻
+阿
+陀
+陂
+附
+际
+陆
+陇
+陈
+陉
+陋
+陌
+降
+限
+陕
+陛
+陟
+陡
+院
+除
+陨
+险
+陪
+陰
+陲
+陳
+陵
+陶
+陷
+陽
+隅
+隆
+隋
+隍
+随
+隐
+隔
+隗
+隘
+隙
+際
+障
+隧
+隨
+隶
+隹
+隼
+隽
+难
+雀
+雁
+雄
+雅
+集
+雇
+雉
+雌
+雍
+雎
+雏
+雒
+雕
+雙
+雜
+雞
+離
+難
+雨
+雪
+雯
+雲
+雳
+零
+雷
+雹
+電
+雾
+需
+霁
+霄
+霆
+震
+霉
+霍
+霎
+霏
+霓
+霖
+霜
+霞
+霭
+霰
+露
+霸
+霹
+霾
+靈
+靑
+青
+靓
+靖
+静
+靛
+靜
+非
+靠
+靡
+面
+靥
+革
+靳
+靴
+靶
+鞅
+鞋
+鞍
+鞑
+鞘
+鞠
+鞣
+鞭
+韓
+韦
+韧
+韩
+韫
+韬
+韭
+音
+韵
+韶
+頂
+順
+頓
+頔
+頡
+頤
+頭
+頰
+題
+顏
+顔
+類
+顧
+顯
+页
+顶
+顷
+项
+顺
+须
+顽
+顾
+顿
+颁
+颂
+预
+颅
+领
+颇
+颈
+颉
+颊
+颌
+颍
+颐
+频
+颓
+颔
+颖
+颗
+题
+颚
+颛
+颜
+额
+颞
+颠
+颡
+颢
+颤
+颦
+颧
+風
+飄
+风
+飒
+飓
+飕
+飘
+飙
+飛
+飞
+食
+飨
+飬
+飮
+飯
+飲
+飼
+飾
+餅
+養
+餐
+餘
+餠
+館
+餮
+饃
+饅
+饕
+饥
+饨
+饪
+饬
+饭
+饮
+饯
+饰
+饱
+饲
+饴
+饵
+饶
+饷
+饸
+饹
+饺
+饼
+饽
+饿
+馁
+馄
+馅
+馆
+馈
+馊
+馋
+馍
+馏
+馒
+馕
+首
+馗
+香
+馥
+馨
+馬
+馮
+馴
+駕
+騰
+騷
+驗
+马
+驭
+驮
+驯
+驰
+驱
+驳
+驴
+驶
+驷
+驸
+驹
+驻
+驼
+驽
+驾
+驿
+骁
+骂
+骄
+骅
+骆
+骇
+骊
+骋
+验
+骏
+骐
+骑
+骓
+骗
+骚
+骛
+骜
+骝
+骞
+骠
+骡
+骤
+骥
+骧
+骨
+骰
+骶
+骷
+骸
+骺
+骼
+髁
+髂
+髅
+髋
+髌
+髓
+體
+高
+髙
+髦
+髪
+髮
+髯
+髻
+鬃
+鬓
+鬟
+鬣
+鬲
+鬼
+魁
+魂
+魄
+魅
+魇
+魉
+魏
+魔
+魚
+鮮
+鯉
+鯨
+鯽
+鱬
+鱸
+鱻
+鱼
+鱿
+鲁
+鲂
+鲅
+鲆
+鲇
+鲈
+鲍
+鲑
+鲛
+鲜
+鲟
+鲢
+鲤
+鲨
+鲩
+鲫
+鲮
+鲱
+鲲
+鲳
+鲵
+鲷
+鲸
+鲽
+鳃
+鳄
+鳅
+鳌
+鳍
+鳎
+鳐
+鳕
+鳖
+鳗
+鳙
+鳜
+鳝
+鳞
+鳟
+鳥
+鳯
+鳳
+鴨
+鴻
+鴿
+鵬
+鶏
+鷄
+鷗
+鷹
+鸟
+鸠
+鸡
+鸢
+鸣
+鸥
+鸦
+鸨
+鸪
+鸬
+鸭
+鸯
+鸳
+鸵
+鸽
+鸾
+鸿
+鹂
+鹃
+鹄
+鹅
+鹉
+鹊
+鹌
+鹏
+鹑
+鹗
+鹜
+鹞
+鹤
+鹦
+鹧
+鹫
+鹭
+鹰
+鹳
+鹽
+鹿
+麂
+麋
+麒
+麓
+麗
+麝
+麟
+麥
+麦
+麯
+麵
+麸
+麺
+麻
+麽
+麾
+黃
+黄
+黍
+黎
+黏
+黑
+黒
+黔
+默
+黛
+黜
+黝
+點
+黟
+黠
+黯
+鼎
+鼓
+鼙
+鼠
+鼬
+鼻
+鼽
+鼾
+齊
+齋
+齐
+齿
+龄
+龈
+龊
+龋
+龌
+龍
+龙
+龚
+龛
+龟
+龢
+︽
+︾
+﹏
+﹥
+﹪
+！
+＃
+％
+＆
+（
+）
+＋
+，
+－
+．
+／
+０
+１
+２
+３
+４
+５
+６
+７
+８
+９
+：
+；
+＝
+？
+＠
+Ａ
+Ｂ
+Ｃ
+Ｄ
+Ｅ
+Ｆ
+Ｇ
+Ｈ
+Ｉ
+Ｊ
+Ｋ
+Ｌ
+Ｍ
+Ｎ
+Ｏ
+Ｐ
+Ｑ
+Ｒ
+Ｓ
+Ｔ
+Ｕ
+Ｖ
+Ｗ
+Ｘ
+Ｙ
+Ｚ
+ａ
+ｂ
+ｃ
+ｄ
+ｅ
+ｆ
+ｇ
+ｈ
+ｉ
+ｊ
+ｋ
+ｌ
+ｍ
+ｎ
+ｏ
+ｐ
+ｑ
+ｒ
+ｓ
+ｔ
+ｕ
+ｖ
+ｗ
+ｘ
+ｙ
+ｚ
+｜
+～
+￥
diff --git a/ppocr/utils/dict/pu_dict.txt b/ppocr/utils/dict/pu_dict.txt
new file mode 100644
index 0000000..9500fae
--- /dev/null
+++ b/ppocr/utils/dict/pu_dict.txt
@@ -0,0 +1,130 @@
+p
+u
+_
+i
+m
+g
+/
+8
+I
+L
+S
+V
+R
+C
+2
+0
+1
+v
+a
+l
+6
+7
+4
+5
+.
+j
+	
+q
+e
+s
+t
+ã
+o
+x
+9
+c
+n
+r
+z
+ç
+õ
+3
+A
+U
+d
+º
+ô
+­
+,
+E
+;
+ó
+á
+b
+D
+?
+ú
+ê
+-
+h
+P
+f
+à
+N
+í
+O
+M
+G
+É
+é
+â
+F
+:
+T
+Á
+"
+Q
+)
+W
+J
+B
+H
+(
+ö
+%
+Ö
+«
+w
+K
+y
+!
+k
+]
+'
+Z
++
+Ç
+Õ
+Y
+À
+X
+µ
+»
+ª
+Í
+ü
+ä
+´
+è
+ñ
+ß
+ï
+Ú
+ë
+Ô
+Ï
+Ó
+[
+Ì
+<
+Â
+ò
+§
+³
+ø
+å
+#
+$
+&
+@
diff --git a/ppocr/utils/dict/rs_dict.txt b/ppocr/utils/dict/rs_dict.txt
new file mode 100644
index 0000000..d1ce46d
--- /dev/null
+++ b/ppocr/utils/dict/rs_dict.txt
@@ -0,0 +1,91 @@
+r
+s
+_
+i
+m
+g
+/
+1
+I
+L
+S
+V
+R
+C
+2
+0
+v
+a
+l
+7
+5
+8
+6
+.
+j
+p
+	
+t
+d
+9
+3
+e
+š
+4
+k
+u
+ć
+c
+n
+đ
+o
+z
+č
+b
+ž
+f
+Z
+T
+h
+M
+F
+O
+Š
+B
+H
+A
+E
+Đ
+Ž
+D
+P
+G
+Č
+K
+U
+N
+J
+Ć
+w
+y
+W
+x
+Y
+X
+q
+Q
+#
+&
+$
+,
+-
+%
+'
+@
+!
+:
+?
+(
+É
+é
++
diff --git a/ppocr/utils/dict/rsc_dict.txt b/ppocr/utils/dict/rsc_dict.txt
new file mode 100644
index 0000000..95dd463
--- /dev/null
+++ b/ppocr/utils/dict/rsc_dict.txt
@@ -0,0 +1,134 @@
+r
+s
+c
+_
+i
+m
+g
+/
+5
+I
+L
+S
+V
+R
+C
+2
+0
+1
+v
+a
+l
+9
+7
+8
+.
+j
+p	
+м
+а
+с
+и
+р
+ћ
+е
+ш
+3
+4
+о
+г
+н
+з
+в
+л
+6
+т
+ж
+у
+к
+п
+њ
+д
+ч
+С
+ј
+ф
+ц
+љ
+х
+О
+И
+А
+б
+Ш
+К
+ђ
+џ
+М
+В
+З
+Д
+Р
+У
+Н
+Т
+Б
+?
+П
+Х
+Ј
+Ц
+Г
+Љ
+Л
+Ф
+e
+n
+w
+E
+F
+A
+N
+f
+o
+b
+M
+G
+t
+y
+W
+k
+P
+u
+H
+B
+T
+z
+h
+O
+Y
+d
+U
+K
+D
+x
+X
+J
+Z
+Q
+q
+'
+-
+@
+é
+#
+!
+,
+%
+$
+:
+&
++
+(
+É
+ 
diff --git a/ppocr/utils/dict/ru_dict.txt b/ppocr/utils/dict/ru_dict.txt
new file mode 100644
index 0000000..3b0cf3a
--- /dev/null
+++ b/ppocr/utils/dict/ru_dict.txt
@@ -0,0 +1,125 @@
+к
+в
+а
+з
+и
+у
+р
+о
+н
+я
+х
+п
+л
+ы 
+г
+е
+т
+м
+д
+ж
+ш
+ь
+с
+ё
+б
+й
+ч
+ю
+ц
+щ
+М
+э
+ф
+А
+ъ
+С
+Ф
+Ю
+В
+К
+Т
+Н
+О
+Э
+У
+И
+Г
+Л
+Р
+Д
+Б
+Ш
+П
+З
+Х
+Е
+Ж
+Я
+Ц
+Ч
+Й
+Щ
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+ 
diff --git a/ppocr/utils/dict/samaritan_dict.txt b/ppocr/utils/dict/samaritan_dict.txt
new file mode 100644
index 0000000..1fe9187
--- /dev/null
+++ b/ppocr/utils/dict/samaritan_dict.txt
@@ -0,0 +1,222 @@
+!
+#
+$
+%
+&
+'
+(
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+?
+@
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+_
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+É
+é
+ء
+آ
+أ
+ؤ
+إ
+ئ
+ا
+ب
+ة
+ت
+ث
+ج
+ح
+خ
+د
+ذ
+ر
+ز
+س
+ش
+ص
+ض
+ط
+ظ
+ع
+غ
+ف
+ق
+ك
+ل
+م
+ن
+ه
+و
+ى
+ي
+ً
+ٌ
+ٍ
+َ
+ُ
+ِ
+ّ
+ْ
+ٓ
+ٔ
+ٰ
+ٱ
+ٹ
+پ
+چ
+ڈ
+ڑ
+ژ
+ک
+ڭ
+گ
+ں
+ھ
+ۀ
+ہ
+ۂ
+ۃ
+ۆ
+ۇ
+ۈ
+ۋ
+ی
+ې
+ے
+ۓ
+ە
+١
+٢
+٣
+٤
+٥
+٦
+٧
+٨
+٩
+ࠀ
+ࠁ
+ࠂ
+ࠃ
+ࠄ
+ࠅ
+ࠆ
+ࠇ
+ࠈ
+ࠉ
+ࠊ
+ࠋ
+ࠌ
+ࠍ
+ࠎ
+ࠏ
+ࠐ
+ࠑ
+ࠒ
+ࠓ
+ࠔ
+ࠕ
+ࠖ
+ࠗ
+࠘
+࠙
+ࠚ
+ࠛ
+ࠜ
+ࠝ
+ࠞ
+ࠟ
+ࠠ
+ࠡ
+ࠢ
+ࠣ
+ࠤ
+ࠥ
+ࠦ
+ࠧ
+ࠨ
+ࠩ
+ࠪ
+ࠫ
+ࠬ
+࠭
+࠰
+࠱
+࠲
+࠳
+࠴
+࠵
+࠶
+࠷
+࠸
+࠹
+࠺
+࠻
+࠼
+࠽
+࠾
diff --git a/ppocr/utils/dict/spin_dict.txt b/ppocr/utils/dict/spin_dict.txt
new file mode 100644
index 0000000..b6c16c8
--- /dev/null
+++ b/ppocr/utils/dict/spin_dict.txt
@@ -0,0 +1,68 @@
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+:
+(
+'
+-
+,
+%
+>
+.
+[
+?
+)
+"
+=
+_
+*
+]
+;
+&
++
+$
+@
+/
+|
+!
+<
+#
+`
+{
+~
+\
+}
+^
diff --git a/ppocr/utils/dict/syriac_dict.txt b/ppocr/utils/dict/syriac_dict.txt
new file mode 100644
index 0000000..4417f4a
--- /dev/null
+++ b/ppocr/utils/dict/syriac_dict.txt
@@ -0,0 +1,157 @@
+!
+#
+$
+%
+&
+'
+(
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+?
+@
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+_
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+É
+é
+܀
+܁
+܂
+܃
+܄
+܅
+܆
+܇
+܈
+܉
+܊
+܋
+܌
+܍
+܏
+ܐ
+ܑ
+ܒ
+ܓ
+ܔ
+ܕ
+ܖ
+ܗ
+ܘ
+ܙ
+ܚ
+ܛ
+ܜ
+ܝ
+ܞ
+ܟ
+ܠ
+ܡ
+ܢ
+ܣ
+ܤ
+ܥ
+ܦ
+ܧ
+ܨ
+ܩ
+ܪ
+ܫ
+ܬ
+ܭ
+ܮ
+ܯ
+ܰ
+ܱ
+ܲ
+ܳ
+ܴ
+ܵ
+ܶ
+ܷ
+ܸ
+ܹ
+ܺ
+ܻ
+ܼ
+ܽ
+ܾ
+ܿ
+݀
+݁
+݂
+݃
+݄
+݅
+݆
+݇
+݈
+݉
+݊
+ݍ
+ݎ
+ݏ
diff --git a/ppocr/utils/dict/ta_dict.txt b/ppocr/utils/dict/ta_dict.txt
new file mode 100644
index 0000000..19d8189
--- /dev/null
+++ b/ppocr/utils/dict/ta_dict.txt
@@ -0,0 +1,128 @@
+t
+a
+_
+i
+m
+g
+/
+3
+I
+L
+S
+V
+R
+C
+2
+0
+1
+v
+l
+9
+7
+8
+.
+j
+p
+ப
+ூ
+த
+ம
+ி
+வ
+ர
+்
+ந
+ோ
+ன
+6
+ஆ
+ற
+ல
+5
+ள
+ா
+ொ
+ழ
+ு
+4
+ெ
+ண
+க
+ட
+ை
+ே
+ச
+ய
+ஒ
+இ
+அ
+ங
+உ
+ீ
+ஞ
+எ
+ஓ
+ஃ
+ஜ
+ஷ
+ஸ
+ஏ
+ஊ
+ஹ
+ஈ
+ஐ
+ௌ
+ஔ
+s
+c
+e
+n
+w
+F
+T
+O
+P
+K
+A
+N
+G
+Y
+E
+M
+H
+U
+B
+o
+b
+D
+d
+r
+W
+u
+y
+f
+X
+k
+q
+h
+J
+z
+Z
+Q
+x
+-
+'
+$
+,
+%
+@
+é
+!
+#
++
+É
+&
+:
+(
+?
+ 
diff --git a/ppocr/utils/dict/table_dict.txt b/ppocr/utils/dict/table_dict.txt
new file mode 100644
index 0000000..2ef028c
--- /dev/null
+++ b/ppocr/utils/dict/table_dict.txt
@@ -0,0 +1,277 @@
+←
+</overline>
+☆
+─
+α
+
+ 
+⋅
+$
+ω
+ψ
+χ
+(
+υ
+≥
+σ
+,
+ρ
+ε
+0
+■
+4
+8
+✗
+b
+<
+✓
+Ψ
+Ω
+€
+D
+3
+Π
+H
+║
+</strike>
+L
+Φ
+Χ
+θ
+P
+κ
+λ
+μ
+T
+ξ
+X
+β
+γ
+δ
+\
+ζ
+η
+`
+d
+<strike>
+h
+f
+l
+Θ
+p
+√
+t
+</sub>
+x
+Β
+Γ
+Δ
+|
+ǂ
+ɛ
+j
+̧
+➢
+⁡
+̌
+′
+«
+△
+▲
+#
+</b>
+'
+Ι
++
+¶
+/
+▼
+⇑
+□
+·
+7
+▪
+;
+?
+➔
+∩
+C
+÷
+G
+⇒
+K
+<sup>
+O
+S
+С
+W
+Α
+[
+○
+_
+●
+‡
+c
+z
+g
+<i>
+o
+<sub>
+〈
+〉
+s
+⩽
+w
+φ
+ʹ
+{
+»
+∣
+̆
+e
+ˆ
+∈
+τ
+◆
+ι
+∅
+∆
+∙
+∘
+Ø
+ß
+✔
+∞
+∑
+−
+×
+◊
+∗
+∖
+˃
+˂
+∫
+"
+i
+&
+π
+↔
+*
+∥
+æ
+∧
+.
+⁄
+ø
+Q
+∼
+6
+⁎
+:
+★
+>
+a
+B
+≈
+F
+J
+̄
+N
+♯
+R
+V
+<overline>
+―
+Z
+♣
+^
+¤
+¥
+§
+<underline>
+¢
+£
+≦
+­
+≤
+‖
+Λ
+©
+n
+↓
+→
+↑
+r
+°
+±
+v
+<b>
+♂
+k
+♀
+~
+ᅟ
+̇
+@
+”
+♦
+ł
+®
+⊕
+„
+!
+</sup>
+%
+⇓
+)
+-
+1
+5
+9
+=
+А
+A
+‰
+⋆
+Σ
+E
+◦
+I
+※
+M
+m
+̨
+⩾
+†
+</i>
+•
+U
+Y
+

+]
+̸
+2
+‐
+–
+‒
+̂
+—
+̀
+́
+’
+‘
+⋮
+⋯
+̊
+“
+̈
+≧
+q
+u
+ı
+y
+</underline>
+​
+̃
+}
+ν
diff --git a/ppocr/utils/dict/table_master_structure_dict.txt b/ppocr/utils/dict/table_master_structure_dict.txt
new file mode 100644
index 0000000..95ab253
--- /dev/null
+++ b/ppocr/utils/dict/table_master_structure_dict.txt
@@ -0,0 +1,39 @@
+<thead>
+<tr>
+<td></td>
+</tr>
+</thead>
+<tbody>
+<eb></eb>
+</tbody>
+<td
+ colspan="5"
+>
+</td>
+ colspan="2"
+ colspan="3"
+<eb2></eb2>
+<eb1></eb1>
+ rowspan="2"
+ colspan="4"
+ colspan="6"
+ rowspan="3"
+ colspan="9"
+ colspan="10"
+ colspan="7"
+ rowspan="4"
+ rowspan="5"
+ rowspan="9"
+ colspan="8"
+ rowspan="8"
+ rowspan="6"
+ rowspan="7"
+ rowspan="10"
+<eb3></eb3>
+<eb4></eb4>
+<eb5></eb5>
+<eb6></eb6>
+<eb7></eb7>
+<eb8></eb8>
+<eb9></eb9>
+<eb10></eb10>
diff --git a/ppocr/utils/dict/table_structure_dict.txt b/ppocr/utils/dict/table_structure_dict.txt
new file mode 100644
index 0000000..fec6f7d
--- /dev/null
+++ b/ppocr/utils/dict/table_structure_dict.txt
@@ -0,0 +1,28 @@
+<thead>
+<tr>
+<td>
+</td>
+</tr>
+</thead>
+<tbody>
+</tbody>
+<td
+ colspan="5"
+>
+ colspan="2"
+ colspan="3"
+ rowspan="2"
+ colspan="4"
+ colspan="6"
+ rowspan="3"
+ colspan="9"
+ colspan="10"
+ colspan="7"
+ rowspan="4"
+ rowspan="5"
+ rowspan="9"
+ colspan="8"
+ rowspan="8"
+ rowspan="6"
+ rowspan="7"
+ rowspan="10"
diff --git a/ppocr/utils/dict/table_structure_dict_ch.txt b/ppocr/utils/dict/table_structure_dict_ch.txt
new file mode 100644
index 0000000..0c59c0e
--- /dev/null
+++ b/ppocr/utils/dict/table_structure_dict_ch.txt
@@ -0,0 +1,48 @@
+<thead>
+</thead>
+<tbody>
+</tbody>
+<tr>
+</tr>
+<td>
+<td
+>
+</td>
+ colspan="2"
+ colspan="3"
+ colspan="4"
+ colspan="5"
+ colspan="6"
+ colspan="7"
+ colspan="8"
+ colspan="9"
+ colspan="10"
+ colspan="11"
+ colspan="12"
+ colspan="13"
+ colspan="14"
+ colspan="15"
+ colspan="16"
+ colspan="17"
+ colspan="18"
+ colspan="19"
+ colspan="20"
+ rowspan="2"
+ rowspan="3"
+ rowspan="4"
+ rowspan="5"
+ rowspan="6"
+ rowspan="7"
+ rowspan="8"
+ rowspan="9"
+ rowspan="10"
+ rowspan="11"
+ rowspan="12"
+ rowspan="13"
+ rowspan="14"
+ rowspan="15"
+ rowspan="16"
+ rowspan="17"
+ rowspan="18"
+ rowspan="19"
+ rowspan="20"
diff --git a/ppocr/utils/dict/te_dict.txt b/ppocr/utils/dict/te_dict.txt
new file mode 100644
index 0000000..83d74cc
--- /dev/null
+++ b/ppocr/utils/dict/te_dict.txt
@@ -0,0 +1,151 @@
+t
+e
+_
+i
+m
+g
+/
+5
+I
+L
+S
+V
+R
+C
+2
+0
+1
+v
+a
+l
+3
+4
+8
+9
+.
+j
+p
+త
+ె
+ర
+క
+్
+ి
+ం
+చ
+ే
+ద
+ు
+7
+6
+ఉ
+ా
+మ
+ట
+ో
+వ
+ప
+ల
+శ
+ఆ
+య
+ై
+భ
+'
+ీ
+గ
+ూ
+డ
+ధ
+హ
+న
+జ
+స
+[
+‌
+ష
+అ
+ణ
+ఫ
+బ
+ఎ
+;
+ళ
+థ
+ొ
+ఠ
+ృ
+ఒ
+ఇ
+ః
+ఊ
+ఖ
+-
+ఐ
+ఘ
+ౌ
+ఏ
+ఈ
+ఛ
+,
+ఓ
+ఞ
+|
+?
+:
+ఢ
+"
+(
+”
+!
++
+)
+*
+=
+&
+“
+€
+]
+£
+$
+s
+c
+n
+w
+k
+J
+G
+u
+d
+r
+E
+o
+h
+y
+b
+f
+B
+M
+O
+T
+N
+D
+P
+A
+F
+x
+W
+Y
+U
+H
+K
+X
+z
+Z
+Q
+q
+É
+%
+#
+@
+é
diff --git a/ppocr/utils/dict/th_dict.txt b/ppocr/utils/dict/th_dict.txt
new file mode 100644
index 0000000..83a5b54
--- /dev/null
+++ b/ppocr/utils/dict/th_dict.txt
@@ -0,0 +1,81 @@
+ก
+ข
+ฃ
+ค
+ฅ
+ฆ
+ง
+จ
+ฉ
+ช
+ซ
+ฌ
+ญ
+ฎ
+ฏ
+ฐ
+ฑ
+ฒ
+ณ
+ด
+ต
+ถ
+ท
+ธ
+น
+บ
+ป
+ผ
+ฝ
+พ
+ฟ
+ภ
+ม
+ย
+ร
+ล
+ว
+ศ
+ษ
+ส
+ห
+ฬ
+อ
+ฮ
+ะ
+ั
+า
+ำ
+ิ
+ี
+ึ
+ื
+ุ
+ู
+เ
+แ
+โ
+ใ
+ไ
+็
+่
+้
+๊
+๋
+์
+๐
+๑
+๒
+๓
+๔
+๕
+๖
+๗
+๘
+๙
+ฯ
+ๆ
+ฤ
+ฤา
+ฦ
+ฦา
diff --git a/ppocr/utils/dict/ug_dict.txt b/ppocr/utils/dict/ug_dict.txt
new file mode 100644
index 0000000..8fc0db7
--- /dev/null
+++ b/ppocr/utils/dict/ug_dict.txt
@@ -0,0 +1,131 @@
+`
+~
+=
+|
+ 
+_
+-
+,
+;
+:
+!
+?
+/
+.
+'
+"
+(
+)
+[
+]
+{
+}
+@
+$
+*
+\
+&
+#
+%
++
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+a
+A
+b
+B
+c
+C
+d
+D
+e
+E
+f
+F
+g
+G
+h
+H
+i
+I
+j
+J
+k
+K
+l
+L
+m
+M
+n
+N
+o
+O
+p
+P
+q
+Q
+r
+R
+s
+S
+t
+T
+u
+U
+v
+V
+w
+W
+x
+X
+y
+Y
+z
+Z
+Ö
+ö
+Ü
+ü
+Ë
+ë
+ا
+ە
+ب
+پ
+ت
+ج
+چ
+خ
+د
+ر
+ز
+ژ
+س
+ش
+غ
+ف
+ق
+ك
+گ
+ڭ
+ل
+م
+ن
+ھ
+و
+ۇ
+ۆ
+ۈ
+ۋ
+ې
+ى
+ي
+ئ
diff --git a/ppocr/utils/dict/uk_dict.txt b/ppocr/utils/dict/uk_dict.txt
new file mode 100644
index 0000000..c5ffc0a
--- /dev/null
+++ b/ppocr/utils/dict/uk_dict.txt
@@ -0,0 +1,142 @@
+u
+k
+_
+i
+m
+g
+/
+1
+6
+I
+L
+S
+V
+R
+C
+2
+0
+v
+a
+l
+7
+9
+.
+j
+p	
+в
+і
+д
+п
+о
+н
+с
+т
+ю
+4
+5
+3
+а
+и
+м
+е
+р
+ч
+у
+Б
+з
+л
+к
+8
+А
+В
+г
+є
+б
+ь
+х
+ґ
+ш
+ц
+ф
+я
+щ
+ж
+Г
+Х
+У
+Т
+Е
+І
+Н
+П
+З
+Л
+Ю
+С
+Д
+М
+К
+Р
+Ф
+О
+Ц
+И
+Я
+Ч
+Ш
+Ж
+Є
+Ґ
+Ь
+s
+c
+e
+n
+w
+A
+P
+r
+E
+t
+o
+h
+d
+y
+M
+G
+N
+F
+B
+T
+D
+U
+O
+W
+Z
+f
+H
+Y
+b
+K
+z
+x
+Q
+X
+q
+J
+$
+-
+'
+#
+&
+%
+?
+:
+!
+,
++
+@
+(
+é
+É
+ 
diff --git a/ppocr/utils/dict/unimernet_tokenizer/tokenizer.json b/ppocr/utils/dict/unimernet_tokenizer/tokenizer.json
new file mode 100644
index 0000000..55a7760
--- /dev/null
+++ b/ppocr/utils/dict/unimernet_tokenizer/tokenizer.json
@@ -0,0 +1,100067 @@
+{
+  "version": "1.0",
+  "truncation": {
+    "direction": "Right",
+    "max_length": 4096,
+    "strategy": "LongestFirst",
+    "stride": 0
+  },
+  "padding": {
+    "strategy": {
+      "Fixed": 4096
+    },
+    "direction": "Right",
+    "pad_to_multiple_of": null,
+    "pad_id": 1,
+    "pad_type_id": 0,
+    "pad_token": "<pad>"
+  },
+  "added_tokens": [
+    {
+      "id": 0,
+      "content": "<s>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 1,
+      "content": "<pad>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 2,
+      "content": "</s>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 3,
+      "content": "<unk>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 4,
+      "content": "[START_REF]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 5,
+      "content": "[END_REF]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 6,
+      "content": "[IMAGE]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 7,
+      "content": "<fragments>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 8,
+      "content": "</fragments>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 9,
+      "content": "<work>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 10,
+      "content": "</work>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 11,
+      "content": "[START_SUP]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 12,
+      "content": "[END_SUP]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 13,
+      "content": "[START_SUB]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 14,
+      "content": "[END_SUB]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 15,
+      "content": "[START_DNA]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 16,
+      "content": "[END_DNA]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 17,
+      "content": "[START_AMINO]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 18,
+      "content": "[END_AMINO]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 19,
+      "content": "[START_SMILES]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 20,
+      "content": "[END_SMILES]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 21,
+      "content": "[START_I_SMILES]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    },
+    {
+      "id": 22,
+      "content": "[END_I_SMILES]",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+    }
+  ],
+  "normalizer": {
+    "type": "NFKC"
+  },
+  "pre_tokenizer": {
+    "type": "Sequence",
+    "pretokenizers": [
+      {
+        "type": "Split",
+        "pattern": {
+          "String": "SPL1T-TH1S-Pl3A5E"
+        },
+        "behavior": "Removed",
+        "invert": false
+      },
+      {
+        "type": "Digits",
+        "individual_digits": true
+      },
+      {
+        "type": "Split",
+        "pattern": {
+          "Regex": "[\\(\\)\\[\\]\\{\\}]|([!\"\\#\\$%\\&'\\*\\+,\\-\\./:;<=>\\?\\\\\\^_`\\|\\~])\\1*"
+        },
+        "behavior": "Isolated",
+        "invert": false
+      },
+      {
+        "type": "Split",
+        "pattern": {
+          "String": "\n"
+        },
+        "behavior": "Isolated",
+        "invert": false
+      },
+      {
+        "type": "ByteLevel",
+        "add_prefix_space": false,
+        "trim_offsets": true,
+        "use_regex": true
+      }
+    ]
+  },
+  "post_processor": {
+    "type": "TemplateProcessing",
+    "single": [
+      {
+        "SpecialToken": {
+          "id": "<s>",
+          "type_id": 0
+        }
+      },
+      {
+        "Sequence": {
+          "id": "A",
+          "type_id": 0
+        }
+      },
+      {
+        "SpecialToken": {
+          "id": "</s>",
+          "type_id": 0
+        }
+      }
+    ],
+    "pair": [
+      {
+        "Sequence": {
+          "id": "A",
+          "type_id": 0
+        }
+      },
+      {
+        "Sequence": {
+          "id": "B",
+          "type_id": 1
+        }
+      }
+    ],
+    "special_tokens": {
+      "</s>": {
+        "id": "</s>",
+        "ids": [
+          2
+        ],
+        "tokens": [
+          "</s>"
+        ]
+      },
+      "<s>": {
+        "id": "<s>",
+        "ids": [
+          0
+        ],
+        "tokens": [
+          "<s>"
+        ]
+      }
+    }
+  },
+  "decoder": {
+    "type": "ByteLevel",
+    "add_prefix_space": true,
+    "trim_offsets": true,
+    "use_regex": true
+  },
+  "model": {
+    "type": "BPE",
+    "dropout": null,
+    "unk_token": null,
+    "continuing_subword_prefix": null,
+    "end_of_word_suffix": null,
+    "fuse_unk": false,
+    "byte_fallback": false,
+    "vocab": {
+      "<s>": 0,
+      "<pad>": 1,
+      "</s>": 2,
+      "<unk>": 3,
+      "[START_REF]": 4,
+      "[END_REF]": 5,
+      "[IMAGE]": 6,
+      "<fragments>": 7,
+      "</fragments>": 8,
+      "<work>": 9,
+      "</work>": 10,
+      "[START_SUP]": 11,
+      "[END_SUP]": 12,
+      "[START_SUB]": 13,
+      "[END_SUB]": 14,
+      "[START_DNA]": 15,
+      "[END_DNA]": 16,
+      "[START_AMINO]": 17,
+      "[END_AMINO]": 18,
+      "[START_SMILES]": 19,
+      "[END_SMILES]": 20,
+      "[START_I_SMILES]": 21,
+      "[END_I_SMILES]": 22,
+      "!": 23,
+      "\"": 24,
+      "#": 25,
+      "$": 26,
+      "%": 27,
+      "&": 28,
+      "'": 29,
+      "(": 30,
+      ")": 31,
+      "*": 32,
+      "+": 33,
+      ",": 34,
+      "-": 35,
+      ".": 36,
+      "/": 37,
+      "0": 38,
+      "1": 39,
+      "2": 40,
+      "3": 41,
+      "4": 42,
+      "5": 43,
+      "6": 44,
+      "7": 45,
+      "8": 46,
+      "9": 47,
+      ":": 48,
+      ";": 49,
+      "<": 50,
+      "=": 51,
+      ">": 52,
+      "?": 53,
+      "@": 54,
+      "A": 55,
+      "B": 56,
+      "C": 57,
+      "D": 58,
+      "E": 59,
+      "F": 60,
+      "G": 61,
+      "H": 62,
+      "I": 63,
+      "J": 64,
+      "K": 65,
+      "L": 66,
+      "M": 67,
+      "N": 68,
+      "O": 69,
+      "P": 70,
+      "Q": 71,
+      "R": 72,
+      "S": 73,
+      "T": 74,
+      "U": 75,
+      "V": 76,
+      "W": 77,
+      "X": 78,
+      "Y": 79,
+      "Z": 80,
+      "[": 81,
+      "\\": 82,
+      "]": 83,
+      "^": 84,
+      "_": 85,
+      "`": 86,
+      "a": 87,
+      "b": 88,
+      "c": 89,
+      "d": 90,
+      "e": 91,
+      "f": 92,
+      "g": 93,
+      "h": 94,
+      "i": 95,
+      "j": 96,
+      "k": 97,
+      "l": 98,
+      "m": 99,
+      "n": 100,
+      "o": 101,
+      "p": 102,
+      "q": 103,
+      "r": 104,
+      "s": 105,
+      "t": 106,
+      "u": 107,
+      "v": 108,
+      "w": 109,
+      "x": 110,
+      "y": 111,
+      "z": 112,
+      "{": 113,
+      "|": 114,
+      "}": 115,
+      "~": 116,
+      "¡": 117,
+      "¢": 118,
+      "£": 119,
+      "¤": 120,
+      "¥": 121,
+      "¦": 122,
+      "§": 123,
+      "¨": 124,
+      "©": 125,
+      "ª": 126,
+      "«": 127,
+      "¬": 128,
+      "®": 129,
+      "¯": 130,
+      "°": 131,
+      "±": 132,
+      "²": 133,
+      "³": 134,
+      "´": 135,
+      "µ": 136,
+      "¶": 137,
+      "·": 138,
+      "¸": 139,
+      "¹": 140,
+      "º": 141,
+      "»": 142,
+      "¼": 143,
+      "½": 144,
+      "¾": 145,
+      "¿": 146,
+      "À": 147,
+      "Á": 148,
+      "Â": 149,
+      "Ã": 150,
+      "Ä": 151,
+      "Å": 152,
+      "Æ": 153,
+      "Ç": 154,
+      "È": 155,
+      "É": 156,
+      "Ê": 157,
+      "Ë": 158,
+      "Ì": 159,
+      "Í": 160,
+      "Î": 161,
+      "Ï": 162,
+      "Ð": 163,
+      "Ñ": 164,
+      "Ò": 165,
+      "Ó": 166,
+      "Ô": 167,
+      "Õ": 168,
+      "Ö": 169,
+      "×": 170,
+      "Ø": 171,
+      "Ù": 172,
+      "Ú": 173,
+      "Û": 174,
+      "Ü": 175,
+      "Ý": 176,
+      "Þ": 177,
+      "ß": 178,
+      "à": 179,
+      "á": 180,
+      "â": 181,
+      "ã": 182,
+      "ä": 183,
+      "å": 184,
+      "æ": 185,
+      "ç": 186,
+      "è": 187,
+      "é": 188,
+      "ê": 189,
+      "ë": 190,
+      "ì": 191,
+      "í": 192,
+      "î": 193,
+      "ï": 194,
+      "ð": 195,
+      "ñ": 196,
+      "ò": 197,
+      "ó": 198,
+      "ô": 199,
+      "õ": 200,
+      "ö": 201,
+      "÷": 202,
+      "ø": 203,
+      "ù": 204,
+      "ú": 205,
+      "û": 206,
+      "ü": 207,
+      "ý": 208,
+      "þ": 209,
+      "ÿ": 210,
+      "Ā": 211,
+      "ā": 212,
+      "Ă": 213,
+      "ă": 214,
+      "Ą": 215,
+      "ą": 216,
+      "Ć": 217,
+      "ć": 218,
+      "Ĉ": 219,
+      "ĉ": 220,
+      "Ċ": 221,
+      "ċ": 222,
+      "Č": 223,
+      "č": 224,
+      "Ď": 225,
+      "ď": 226,
+      "Đ": 227,
+      "đ": 228,
+      "Ē": 229,
+      "ē": 230,
+      "Ĕ": 231,
+      "ĕ": 232,
+      "Ė": 233,
+      "ė": 234,
+      "Ę": 235,
+      "ę": 236,
+      "Ě": 237,
+      "ě": 238,
+      "Ĝ": 239,
+      "ĝ": 240,
+      "Ğ": 241,
+      "ğ": 242,
+      "Ġ": 243,
+      "ġ": 244,
+      "Ģ": 245,
+      "ģ": 246,
+      "Ĥ": 247,
+      "ĥ": 248,
+      "Ħ": 249,
+      "ħ": 250,
+      "Ĩ": 251,
+      "ĩ": 252,
+      "Ī": 253,
+      "ī": 254,
+      "Ĭ": 255,
+      "ĭ": 256,
+      "Į": 257,
+      "į": 258,
+      "İ": 259,
+      "ı": 260,
+      "IJ": 261,
+      "ij": 262,
+      "Ĵ": 263,
+      "ĵ": 264,
+      "Ķ": 265,
+      "ķ": 266,
+      "ĸ": 267,
+      "Ĺ": 268,
+      "ĺ": 269,
+      "Ļ": 270,
+      "ļ": 271,
+      "Ľ": 272,
+      "ľ": 273,
+      "Ŀ": 274,
+      "ŀ": 275,
+      "Ł": 276,
+      "ł": 277,
+      "Ń": 278,
+      "Ġt": 279,
+      "in": 280,
+      "Ġa": 281,
+      "he": 282,
+      "on": 283,
+      "re": 284,
+      "at": 285,
+      "Ġthe": 286,
+      "er": 287,
+      "Ġs": 288,
+      "Ġo": 289,
+      "en": 290,
+      "al": 291,
+      "Ġc": 292,
+      "ti": 293,
+      "or": 294,
+      "ed": 295,
+      "es": 296,
+      "is": 297,
+      "Ġp": 298,
+      "Ġof": 299,
+      "nd": 300,
+      "Ġin": 301,
+      "Ġf": 302,
+      "Ġw": 303,
+      "ĠĠ": 304,
+      "it": 305,
+      "an": 306,
+      "ro": 307,
+      "ar": 308,
+      "Ġd": 309,
+      "Ġm": 310,
+      "Ġb": 311,
+      "Ġand": 312,
+      "ic": 313,
+      "le": 314,
+      "ing": 315,
+      "ion": 316,
+      "as": 317,
+      "Ġe": 318,
+      "Ġre": 319,
+      "ation": 320,
+      "Ġto": 321,
+      "el": 322,
+      "ent": 323,
+      "ac": 324,
+      "et": 325,
+      "ec": 326,
+      "tion": 327,
+      "om": 328,
+      "st": 329,
+      "ĠT": 330,
+      "Ġn": 331,
+      "Ġth": 332,
+      "ol": 333,
+      "ul": 334,
+      "im": 335,
+      "RE": 336,
+      "ig": 337,
+      "us": 338,
+      "REF": 339,
+      "Ġl": 340,
+      "Ġh": 341,
+      "ur": 342,
+      "Ġis": 343,
+      "ĠĠĠĠ": 344,
+      "Ġfor": 345,
+      "id": 346,
+      "am": 347,
+      "ĠS": 348,
+      "ve": 349,
+      "il": 350,
+      "ĠA": 351,
+      "ĠC": 352,
+      "Ġg": 353,
+      "ot": 354,
+      "ith": 355,
+      "ly": 356,
+      "ce": 357,
+      "Ġcon": 358,
+      "ow": 359,
+      "Ġst": 360,
+      "ut": 361,
+      "os": 362,
+      "Ġwith": 363,
+      "od": 364,
+      "ra": 365,
+      "Ġv": 366,
+      "Ġpro": 367,
+      "um": 368,
+      "ĠI": 369,
+      "if": 370,
+      "uc": 371,
+      "ter": 372,
+      "un": 373,
+      "AR": 374,
+      "ST": 375,
+      "res": 376,
+      "Ġon": 377,
+      "EN": 378,
+      "ere": 379,
+      "ĠP": 380,
+      "ĠThe": 381,
+      "ĠM": 382,
+      "Ġas": 383,
+      "ART": 384,
+      "Ġan": 385,
+      "END": 386,
+      "START": 387,
+      "Ġthat": 388,
+      "qu": 389,
+      "em": 390,
+      "Ġbe": 391,
+      "Ġex": 392,
+      "ri": 393,
+      "ab": 394,
+      "ity": 395,
+      "tic": 396,
+      "ver": 397,
+      "Ġal": 398,
+      "pl": 399,
+      "ts": 400,
+      "ĠF": 401,
+      "Ġâ": 402,
+      "ure": 403,
+      "Ġby": 404,
+      "ate": 405,
+      "ag": 406,
+      "ir": 407,
+      "oc": 408,
+      "per": 409,
+      "ĠB": 410,
+      "ay": 411,
+      "ĠD": 412,
+      "Ġcom": 413,
+      "ĠH": 414,
+      "ated": 415,
+      "ĠR": 416,
+      "Ġare": 417,
+      "rom": 418,
+      "ĠE": 419,
+      "op": 420,
+      "ad": 421,
+      "se": 422,
+      "ĠL": 423,
+      "igh": 424,
+      "ĠN": 425,
+      "ment": 426,
+      "her": 427,
+      "og": 428,
+      "ain": 429,
+      "ect": 430,
+      "ud": 431,
+      "Ġde": 432,
+      "Ġr": 433,
+      "Ġat": 434,
+      "Ġwas": 435,
+      "Ġus": 436,
+      "Ġres": 437,
+      "ell": 438,
+      "iz": 439,
+      "ine": 440,
+      "ph": 441,
+      "Ġac": 442,
+      "ess": 443,
+      "ore": 444,
+      "ical": 445,
+      "th": 446,
+      "und": 447,
+      "rac": 448,
+      "Ġwe": 449,
+      "ath": 450,
+      "ĠG": 451,
+      "Ġfrom": 452,
+      "ati": 453,
+      "up": 454,
+      "ist": 455,
+      "ant": 456,
+      "Ġor": 457,
+      "ff": 458,
+      "Ġcomp": 459,
+      "Ġwh": 460,
+      "ĠW": 461,
+      "ch": 462,
+      "ers": 463,
+      "Ġsp": 464,
+      "orm": 465,
+      "Ġch": 466,
+      "ations": 467,
+      "ran": 468,
+      "ub": 469,
+      "te": 470,
+      "di": 471,
+      "Ġsh": 472,
+      "ge": 473,
+      "ase": 474,
+      "Ġwere": 475,
+      "ĠĠĠĠĠĠĠĠ": 476,
+      "ĠÎ": 477,
+      "ap": 478,
+      "ĠIn": 479,
+      "and": 480,
+      "Ġse": 481,
+      "vel": 482,
+      "Ġim": 483,
+      "ĠâĪ": 484,
+      "ens": 485,
+      "ies": 486,
+      "ich": 487,
+      "ight": 488,
+      "duc": 489,
+      "ĠO": 490,
+      "Ġit": 491,
+      "tions": 492,
+      "end": 493,
+      "Ġco": 494,
+      "Ġthis": 495,
+      "Ġcan": 496,
+      "Ġk": 497,
+      "âĢ": 498,
+      "lec": 499,
+      "ted": 500,
+      "Ġmod": 501,
+      "math": 502,
+      "Ġcont": 503,
+      "Ġne": 504,
+      "Ġpar": 505,
+      "ib": 506,
+      "ĠĠĠ": 507,
+      "Ġle": 508,
+      "iv": 509,
+      "ug": 510,
+      "ence": 511,
+      "ign": 512,
+      "ous": 513,
+      "ents": 514,
+      "ys": 515,
+      "ave": 516,
+      "red": 517,
+      "ress": 518,
+      "able": 519,
+      "por": 520,
+      "all": 521,
+      "iff": 522,
+      "est": 523,
+      "Ġap": 524,
+      "Ġinc": 525,
+      "nt": 526,
+      "ary": 527,
+      "iti": 528,
+      "Ġwhich": 529,
+      "Ġnot": 530,
+      "form": 531,
+      "Ġsy": 532,
+      "Ġad": 533,
+      "low": 534,
+      "ak": 535,
+      "Ġper": 536,
+      "Ġhe": 537,
+      "pro": 538,
+      "ance": 539,
+      "ial": 540,
+      "ue": 541,
+      "Ġen": 542,
+      "Ġcl": 543,
+      "ass": 544,
+      "ip": 545,
+      "rans": 546,
+      "Ġob": 547,
+      "Ġgen": 548,
+      "tim": 549,
+      "Ġdis": 550,
+      "unc": 551,
+      "Ġint": 552,
+      "ep": 553,
+      "etw": 554,
+      "Ġdiff": 555,
+      "ach": 556,
+      "ther": 557,
+      "ime": 558,
+      "age": 559,
+      "ple": 560,
+      "ill": 561,
+      "yp": 562,
+      "ĠK": 563,
+      "act": 564,
+      "ari": 565,
+      "Ġmet": 566,
+      "ors": 567,
+      "Ġhave": 568,
+      "Ġstud": 569,
+      "ong": 570,
+      "ĠU": 571,
+      "Ġpl": 572,
+      "ide": 573,
+      "ma": 574,
+      "hen": 575,
+      "ific": 576,
+      "ome": 577,
+      "Ġi": 578,
+      "ular": 579,
+      "ĠV": 580,
+      "ally": 581,
+      "Ġshow": 582,
+      "rib": 583,
+      "ia": 584,
+      "enti": 585,
+      "Ġass": 586,
+      "ond": 587,
+      "ft": 588,
+      "Ġab": 589,
+      "Ġinter": 590,
+      "ĠTh": 591,
+      "The": 592,
+      "str": 593,
+      "Ġcell": 594,
+      "cal": 595,
+      "Ġmodel": 596,
+      "ata": 597,
+      "ast": 598,
+      "Ġeff": 599,
+      "Ġtrans": 600,
+      "ates": 601,
+      "ased": 602,
+      "ost": 603,
+      "vi": 604,
+      "ang": 605,
+      "our": 606,
+      "Ġme": 607,
+      "ard": 608,
+      "Ġdiffere": 609,
+      "Ġpre": 610,
+      "Ġdi": 611,
+      "ĠâĪĴ": 612,
+      "olog": 613,
+      "ution": 614,
+      "ound": 615,
+      "ace": 616,
+      "Ġresul": 617,
+      "erm": 618,
+      "pos": 619,
+      "here": 620,
+      "tive": 621,
+      "ord": 622,
+      "so": 623,
+      "stem": 624,
+      "yl": 625,
+      "Ġph": 626,
+      "Ġy": 627,
+      "ame": 628,
+      "ork": 629,
+      "ative": 630,
+      "Ġqu": 631,
+      "ric": 632,
+      "SU": 633,
+      "wo": 634,
+      "Ġun": 635,
+      "Ġev": 636,
+      "are": 637,
+      "##": 638,
+      "de": 639,
+      "een": 640,
+      "tiv": 641,
+      "Ġgro": 642,
+      "ory": 643,
+      "Ġcons": 644,
+      "Ġsub": 645,
+      "ta": 646,
+      "--": 647,
+      "Ġstr": 648,
+      "ber": 649,
+      "erv": 650,
+      "etween": 651,
+      "enc": 652,
+      "Ġanal": 653,
+      "int": 654,
+      "Ġhas": 655,
+      "uch": 656,
+      "Ġreg": 657,
+      "Ġbetween": 658,
+      "Ġdet": 659,
+      "Ġall": 660,
+      "cess": 661,
+      "Ġexp": 662,
+      "ection": 663,
+      "ĠâĢ": 664,
+      "ind": 665,
+      "ater": 666,
+      "Ġsign": 667,
+      "pt": 668,
+      "ugh": 669,
+      "ite": 670,
+      "ility": 671,
+      "Ġusing": 672,
+      "Ġval": 673,
+      "Ġro": 674,
+      "ree": 675,
+      "Ġrel": 676,
+      "out": 677,
+      "Ġfunc": 678,
+      "ition": 679,
+      "Ġcor": 680,
+      "Ġalso": 681,
+      "Ġtwo": 682,
+      "ne": 683,
+      "ĠJ": 684,
+      "Ġsystem": 685,
+      "cl": 686,
+      "uct": 687,
+      "Ġsim": 688,
+      "tain": 689,
+      "ust": 690,
+      "ied": 691,
+      "port": 692,
+      "Ġrec": 693,
+      "Ġresp": 694,
+      "Ġdata": 695,
+      "rm": 696,
+      "resent": 697,
+      "uld": 698,
+      "xt": 699,
+      "Ġj": 700,
+      "ry": 701,
+      "ack": 702,
+      "Ġra": 703,
+      "par": 704,
+      "Ġform": 705,
+      "Ġsc": 706,
+      "frac": 707,
+      "ĠWe": 708,
+      "ating": 709,
+      "ech": 710,
+      "hod": 711,
+      "Ġfol": 712,
+      "ined": 713,
+      "ĠSt": 714,
+      "ual": 715,
+      "Ġused": 716,
+      "Ġone": 717,
+      "Ġdes": 718,
+      "ĠÏ": 719,
+      "Ġvari": 720,
+      "Ġdist": 721,
+      "Ġnum": 722,
+      "ym": 723,
+      "ew": 724,
+      "rec": 725,
+      "ob": 726,
+      "Ġinf": 727,
+      "Ġar": 728,
+      "lect": 729,
+      "ll": 730,
+      "ons": 731,
+      "ĠThis": 732,
+      "ose": 733,
+      "ile": 734,
+      "play": 735,
+      "ear": 736,
+      "ox": 737,
+      "ures": 738,
+      "one": 739,
+      "Ġstudy": 740,
+      "ysis": 741,
+      "Ġfollow": 742,
+      "yle": 743,
+      "ract": 744,
+      "dis": 745,
+      "Ġpos": 746,
+      "right": 747,
+      "Ġthan": 748,
+      "ros": 749,
+      "av": 750,
+      "Fig": 751,
+      "Ġtime": 752,
+      "ization": 753,
+      "ulation": 754,
+      "ized": 755,
+      "Ġsur": 756,
+      "oth": 757,
+      "Ġout": 758,
+      "Ġcol": 759,
+      "ature": 760,
+      "ive": 761,
+      "Ġsol": 762,
+      "Ġx": 763,
+      "eld": 764,
+      "Ġother": 765,
+      "plic": 766,
+      "Ġdef": 767,
+      "erg": 768,
+      "Ġgener": 769,
+      "ely": 770,
+      "Ġbeen": 771,
+      "Ġincre": 772,
+      "Ġthese": 773,
+      "Ġno": 774,
+      "ax": 775,
+      "style": 776,
+      "arg": 777,
+      "ian": 778,
+      "Ġind": 779,
+      "Ġsuch": 780,
+      "Ġfunction": 781,
+      "ting": 782,
+      "Ġequ": 783,
+      "aus": 784,
+      "Ġund": 785,
+      "mathb": 786,
+      "tical": 787,
+      "Ġhigh": 788,
+      "rain": 789,
+      "Ġam": 790,
+      "ield": 791,
+      "oun": 792,
+      "ression": 793,
+      "Ġspec": 794,
+      "Ġop": 795,
+      "Ġdec": 796,
+      "Ġover": 797,
+      "Ġmethod": 798,
+      "Ġset": 799,
+      "âĪ": 800,
+      "Ġif": 801,
+      "dition": 802,
+      "ues": 803,
+      "ects": 804,
+      "display": 805,
+      "hem": 806,
+      "Ġpati": 807,
+      "Ġresults": 808,
+      "old": 809,
+      "anc": 810,
+      "displaystyle": 811,
+      "Ġeach": 812,
+      "Ġmore": 813,
+      "les": 814,
+      "pr": 815,
+      "acter": 816,
+      "Ġtheir": 817,
+      "Ġacc": 818,
+      "Ġappro": 819,
+      "iss": 820,
+      "ize": 821,
+      "Ġinv": 822,
+      "ases": 823,
+      "Ġcells": 824,
+      "irst": 825,
+      "lu": 826,
+      "ail": 827,
+      "Ġmeas": 828,
+      "Ġlow": 829,
+      "ov": 830,
+      "the": 831,
+      "ik": 832,
+      "**": 833,
+      "ef": 834,
+      "Ġbut": 835,
+      "hes": 836,
+      "fter": 837,
+      "Ġdifferent": 838,
+      "vely": 839,
+      "Ġext": 840,
+      "Ġthere": 841,
+      "oci": 842,
+      "Ġprob": 843,
+      "Ġits": 844,
+      "ron": 845,
+      "ments": 846,
+      "Ġag": 847,
+      "NA": 848,
+      "Ġpo": 849,
+      "ice": 850,
+      "ype": 851,
+      "Ġgroup": 852,
+      "âĢĵ": 853,
+      "ever": 854,
+      "ult": 855,
+      "ism": 856,
+      "tern": 857,
+      "ability": 858,
+      "ions": 859,
+      "ark": 860,
+      "Ġnon": 861,
+      "to": 862,
+      "ĠĠĠĠĠĠĠ": 863,
+      "Ġobs": 864,
+      "Ġtre": 865,
+      "als": 866,
+      "left": 867,
+      "ĠPro": 868,
+      "Ġonly": 869,
+      "Ġman": 870,
+      "der": 871,
+      "Ġpol": 872,
+      "uring": 873,
+      "amet": 874,
+      "rol": 875,
+      "In": 876,
+      "yn": 877,
+      "Ġunder": 878,
+      "ĠCh": 879,
+      "Ġwhere": 880,
+      "ood": 881,
+      "ĠX": 882,
+      "nce": 883,
+      "Ġpartic": 884,
+      "ected": 885,
+      "ĠFig": 886,
+      "Ġem": 887,
+      "Ġfact": 888,
+      "ĠAn": 889,
+      "Ġperform": 890,
+      "Ġso": 891,
+      "Ġanalysis": 892,
+      "stract": 893,
+      "hed": 894,
+      "Ġmay": 895,
+      "atic": 896,
+      "Ġrep": 897,
+      "tein": 898,
+      "duced": 899,
+      "Ġup": 900,
+      "Ġinto": 901,
+      "Ġnumber": 902,
+      "Ġour": 903,
+      "Ġet": 904,
+      "eg": 905,
+      "itle": 906,
+      "over": 907,
+      "ix": 908,
+      "ator": 909,
+      "ulti": 910,
+      "Ġincl": 911,
+      "ould": 912,
+      "ici": 913,
+      "bstract": 914,
+      "Ġcomple": 915,
+      "Ġpatients": 916,
+      "Ġdo": 917,
+      "Ġexper": 918,
+      "vid": 919,
+      "ange": 920,
+      "Ġlevel": 921,
+      "Ġprocess": 922,
+      "mathcal": 923,
+      "ps": 924,
+      "Ġsignific": 925,
+      "Ġsam": 926,
+      "Title": 927,
+      "Ġbl": 928,
+      "Ġstruct": 929,
+      "eta": 930,
+      "Ġobserv": 931,
+      "raph": 932,
+      "gr": 933,
+      "Ġactiv": 934,
+      "Ġfirst": 935,
+      "velop": 936,
+      "gen": 937,
+      "ible": 938,
+      "Ġsm": 939,
+      "Ġwill": 940,
+      "ĠQ": 941,
+      "Ġmeasure": 942,
+      "put": 943,
+      "Ġloc": 944,
+      "Ġmo": 945,
+      "vers": 946,
+      "of": 947,
+      "tal": 948,
+      "ered": 949,
+      "own": 950,
+      "Ġmat": 951,
+      "ities": 952,
+      "til": 953,
+      "inal": 954,
+      "Ġcar": 955,
+      "pha": 956,
+      "Ġboth": 957,
+      "Ġcur": 958,
+      "SUB": 959,
+      "its": 960,
+      "rel": 961,
+      "Ġwhen": 962,
+      "Ġz": 963,
+      "Ġchar": 964,
+      "Ġbi": 965,
+      "cent": 966,
+      "Ġthen": 967,
+      "ise": 968,
+      "owever": 969,
+      "Ġmin": 970,
+      "ĠFor": 971,
+      "ĠY": 972,
+      "ption": 973,
+      "Ġes": 974,
+      "mun": 975,
+      "Ġinclud": 976,
+      "istic": 977,
+      "con": 978,
+      "Ġobtain": 979,
+      "ared": 980,
+      "duction": 981,
+      "Ġsignificant": 982,
+      "ĠZ": 983,
+      "Ġpresent": 984,
+      "ann": 985,
+      "Ġid": 986,
+      "ency": 987,
+      "Ġver": 988,
+      "val": 989,
+      "yd": 990,
+      "rough": 991,
+      "SUP": 992,
+      "fore": 993,
+      "Ġsome": 994,
+      "ĠAs": 995,
+      "Ġsup": 996,
+      "Ġafter": 997,
+      "ological": 998,
+      "entif": 999,
+      "Ġcase": 1000,
+      "Ġsec": 1001,
+      "elf": 1002,
+      "Ġdep": 1003,
+      "ks": 1004,
+      "Ġcal": 1005,
+      "ved": 1006,
+      "Ġtem": 1007,
+      "Ġuse": 1008,
+      "ĠCom": 1009,
+      "lam": 1010,
+      "ines": 1011,
+      "ays": 1012,
+      "Ġgiv": 1013,
+      "Ġconsid": 1014,
+      "Ġelect": 1015,
+      "ational": 1016,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 1017,
+      "iqu": 1018,
+      "ties": 1019,
+      "Ġline": 1020,
+      "Ġsu": 1021,
+      "Abstract": 1022,
+      "ount": 1023,
+      "Ġdevelop": 1024,
+      "ĠCon": 1025,
+      "ology": 1026,
+      "alpha": 1027,
+      "ans": 1028,
+      "prime": 1029,
+      "cc": 1030,
+      "ogen": 1031,
+      "Ġwork": 1032,
+      "ven": 1033,
+      "ium": 1034,
+      "ective": 1035,
+      "Ġpa": 1036,
+      "ten": 1037,
+      "ĠAl": 1038,
+      "Ġï": 1039,
+      "Ġfe": 1040,
+      "âĢĻ": 1041,
+      "ential": 1042,
+      "line": 1043,
+      "Ġparamet": 1044,
+      "Ġprotein": 1045,
+      "Ġdisc": 1046,
+      "face": 1047,
+      "ces": 1048,
+      "Ġwell": 1049,
+      "ural": 1050,
+      "eng": 1051,
+      "Ġduring": 1052,
+      "row": 1053,
+      "ants": 1054,
+      "Ġrem": 1055,
+      "formation": 1056,
+      "Ġexam": 1057,
+      "Ġmic": 1058,
+      "âĪĴ": 1059,
+      "lem": 1060,
+      "ergy": 1061,
+      "Ġassoci": 1062,
+      "ĠÃ": 1063,
+      "rop": 1064,
+      "Ġfield": 1065,
+      "ty": 1066,
+      "Ġclass": 1067,
+      "Ġu": 1068,
+      "ie": 1069,
+      "Ġbec": 1070,
+      "Ġexperim": 1071,
+      "sp": 1072,
+      "Ġpr": 1073,
+      "ilar": 1074,
+      "tial": 1075,
+      "Ġconst": 1076,
+      "ĠIt": 1077,
+      "Ġcontrol": 1078,
+      "da": 1079,
+      "Ġmulti": 1080,
+      "itive": 1081,
+      "ics": 1082,
+      "urn": 1083,
+      "Ġindic": 1084,
+      "Ġfound": 1085,
+      "text": 1086,
+      "Ġnew": 1087,
+      "Ġref": 1088,
+      "gor": 1089,
+      "rap": 1090,
+      "Ġdesc": 1091,
+      "Ġsame": 1092,
+      "Ġfollowing": 1093,
+      "Ġdistrib": 1094,
+      "Figure": 1095,
+      "ild": 1096,
+      "Ġanti": 1097,
+      "etwork": 1098,
+      "ove": 1099,
+      "Ġthrough": 1100,
+      "Ġmost": 1101,
+      "cer": 1102,
+      "Ġdeterm": 1103,
+      "ha": 1104,
+      "elta": 1105,
+      "arge": 1106,
+      "Ġshown": 1107,
+      "ince": 1108,
+      "Ġany": 1109,
+      "ren": 1110,
+      "dot": 1111,
+      "ral": 1112,
+      "ration": 1113,
+      "amma": 1114,
+      "oid": 1115,
+      "Ġmed": 1116,
+      "ension": 1117,
+      "art": 1118,
+      "Ġpred": 1119,
+      "met": 1120,
+      "mathbb": 1121,
+      "ake": 1122,
+      "Ġcalc": 1123,
+      "Ġhig": 1124,
+      "Ġthree": 1125,
+      "Ġbased": 1126,
+      "mon": 1127,
+      "arch": 1128,
+      "----": 1129,
+      "ples": 1130,
+      "ages": 1131,
+      "ause": 1132,
+      "ish": 1133,
+      "tively": 1134,
+      "qui": 1135,
+      "resp": 1136,
+      "Ġcharacter": 1137,
+      "ock": 1138,
+      "Ġtreat": 1139,
+      "Ġproper": 1140,
+      "ex": 1141,
+      "Ġsmall": 1142,
+      "Ġterm": 1143,
+      "bda": 1144,
+      "Ġkn": 1145,
+      "ode": 1146,
+      "ings": 1147,
+      "Ġexpression": 1148,
+      "Ġmon": 1149,
+      "emb": 1150,
+      "ute": 1151,
+      "echn": 1152,
+      "hib": 1153,
+      "Ġdirec": 1154,
+      "ination": 1155,
+      "ithm": 1156,
+      "ulated": 1157,
+      "Ġcy": 1158,
+      "Ġpot": 1159,
+      "Ġorder": 1160,
+      "ote": 1161,
+      "ically": 1162,
+      "Ġvalues": 1163,
+      "ort": 1164,
+      "urther": 1165,
+      "cept": 1166,
+      "ynam": 1167,
+      "ough": 1168,
+      "echan": 1169,
+      "Ġâī": 1170,
+      "ok": 1171,
+      "ement": 1172,
+      "Ġμ": 1173,
+      "Ġestim": 1174,
+      "Ġeffect": 1175,
+      "Ġpath": 1176,
+      "Ġconf": 1177,
+      "Ġapp": 1178,
+      "Ġgiven": 1179,
+      "Ġend": 1180,
+      "set": 1181,
+      "Ġgl": 1182,
+      "Ġthey": 1183,
+      "ning": 1184,
+      "Ġtest": 1185,
+      "Ġtemper": 1186,
+      "ves": 1187,
+      "Ġvalue": 1188,
+      "ited": 1189,
+      "ality": 1190,
+      "Ġlim": 1191,
+      "Ġspect": 1192,
+      "ently": 1193,
+      "tit": 1194,
+      "Ġsequ": 1195,
+      "Ġidentif": 1196,
+      "//": 1197,
+      "igma": 1198,
+      "Ġenergy": 1199,
+      "inc": 1200,
+      "ness": 1201,
+      "ensity": 1202,
+      "Ġproblem": 1203,
+      "ydro": 1204,
+      "agn": 1205,
+      "ane": 1206,
+      "rent": 1207,
+      "com": 1208,
+      "ject": 1209,
+      "Ġimport": 1210,
+      "ĉĉ": 1211,
+      "Ġoper": 1212,
+      "olution": 1213,
+      "Ġaut": 1214,
+      "ectively": 1215,
+      "ĠHowever": 1216,
+      "ho": 1217,
+      "ental": 1218,
+      "Ġsing": 1219,
+      "ey": 1220,
+      "mu": 1221,
+      "ross": 1222,
+      "action": 1223,
+      "epend": 1224,
+      "ĠEx": 1225,
+      "vious": 1226,
+      "Ġstudies": 1227,
+      "sc": 1228,
+      "ormal": 1229,
+      "Ġhad": 1230,
+      "Ġmain": 1231,
+      "alth": 1232,
+      "gorithm": 1233,
+      "Ġfl": 1234,
+      "omet": 1235,
+      "ĠÂ": 1236,
+      "..": 1237,
+      "err": 1238,
+      "Ġposs": 1239,
+      "Ġdifferen": 1240,
+      "Ġobserved": 1241,
+      "ray": 1242,
+      "Ġpredic": 1243,
+      "Ġgene": 1244,
+      "Ġstate": 1245,
+      "We": 1246,
+      "Ġstructure": 1247,
+      "Ġret": 1248,
+      "respond": 1249,
+      "requ": 1250,
+      "ily": 1251,
+      "ĠâĪĪ": 1252,
+      "Ġser": 1253,
+      "Ġbound": 1254,
+      "Ġrepresent": 1255,
+      "phi": 1256,
+      "Ġtreatment": 1257,
+      "hat": 1258,
+      "Ġrequi": 1259,
+      "app": 1260,
+      "uman": 1261,
+      "Ġhigher": 1262,
+      "Ġlarge": 1263,
+      "Ġtra": 1264,
+      "ward": 1265,
+      "Ġobtained": 1266,
+      "Ġcould": 1267,
+      "tig": 1268,
+      "ĠUn": 1269,
+      "Ġdescrib": 1270,
+      "Ġsimilar": 1271,
+      "ported": 1272,
+      "ins": 1273,
+      "Ġaddition": 1274,
+      "osis": 1275,
+      "Ġnetwork": 1276,
+      "Ġele": 1277,
+      "pi": 1278,
+      "rix": 1279,
+      "Ġrate": 1280,
+      "gan": 1281,
+      "ugg": 1282,
+      "uss": 1283,
+      "Ġmechan": 1284,
+      "Ġdise": 1285,
+      "Ġeffects": 1286,
+      "Ġmodels": 1287,
+      "orph": 1288,
+      "ike": 1289,
+      "Ġsecond": 1290,
+      "mathbf": 1291,
+      "Ġdue": 1292,
+      "Ġq": 1293,
+      "Ġpres": 1294,
+      "Ġtechn": 1295,
+      "els": 1296,
+      "Ġcorrespond": 1297,
+      "Ġassociated": 1298,
+      "posed": 1299,
+      "Ġmass": 1300,
+      "round": 1301,
+      "view": 1302,
+      "Ġins": 1303,
+      "ĠâĢ¢": 1304,
+      "ditions": 1305,
+      "Ġwhile": 1306,
+      "ole": 1307,
+      "Ġlong": 1308,
+      "alu": 1309,
+      "Ġcap": 1310,
+      "Ġsurface": 1311,
+      "Ġcomplex": 1312,
+      "Ġcent": 1313,
+      "Ġcompared": 1314,
+      "Ġfind": 1315,
+      "arget": 1316,
+      "atory": 1317,
+      "fer": 1318,
+      "Ġsize": 1319,
+      "Ġcontain": 1320,
+      "usion": 1321,
+      "utions": 1322,
+      "Ġdem": 1323,
+      "ES": 1324,
+      "Ġdepend": 1325,
+      "atis": 1326,
+      "sum": 1327,
+      "ffici": 1328,
+      "Ġbas": 1329,
+      "lambda": 1330,
+      "ier": 1331,
+      "AT": 1332,
+      "Ġmax": 1333,
+      "Ġimp": 1334,
+      "Ġevalu": 1335,
+      "Ġtemperature": 1336,
+      "ink": 1337,
+      "ector": 1338,
+      "Ġscal": 1339,
+      "Ġgrow": 1340,
+      "ower": 1341,
+      "Ġrespectively": 1342,
+      "lear": 1343,
+      "sh": 1344,
+      "ick": 1345,
+      "Ġfil": 1346,
+      "irc": 1347,
+      "ilon": 1348,
+      "ram": 1349,
+      "Ġα": 1350,
+      "ification": 1351,
+      "Ġocc": 1352,
+      "Ġyear": 1353,
+      "Ġsugg": 1354,
+      "Ġradi": 1355,
+      "ified": 1356,
+      "havi": 1357,
+      "Ġwithin": 1358,
+      "Ġsens": 1359,
+      "Ġinte": 1360,
+      "Ġwould": 1361,
+      "Ġconcent": 1362,
+      "Ġmicro": 1363,
+      "Ġsingle": 1364,
+      "ĠSp": 1365,
+      "ou": 1366,
+      "Ġatt": 1367,
+      "Ġself": 1368,
+      "Ġabout": 1369,
+      "ength": 1370,
+      "Ġel": 1371,
+      "ĠRe": 1372,
+      "xim": 1373,
+      "Ġconditions": 1374,
+      "ude": 1375,
+      "ĠAt": 1376,
+      "where": 1377,
+      "med": 1378,
+      "Ġneed": 1379,
+      "iron": 1380,
+      "Ġpop": 1381,
+      "Ġresult": 1382,
+      "Ġpoint": 1383,
+      "Ġlo": 1384,
+      "Ġalgorithm": 1385,
+      "Ġactivity": 1386,
+      "leq": 1387,
+      "plement": 1388,
+      "ĠRes": 1389,
+      "Ġsym": 1390,
+      "onstr": 1391,
+      "atures": 1392,
+      "Ġimpro": 1393,
+      "for": 1394,
+      "Ġgeneral": 1395,
+      "iter": 1396,
+      "Ġexpl": 1397,
+      "###": 1398,
+      "Ġdom": 1399,
+      "Ġtri": 1400,
+      "min": 1401,
+      "Ġdistribution": 1402,
+      "Ġtr": 1403,
+      "ĠThere": 1404,
+      "oss": 1405,
+      "uce": 1406,
+      "mathrm": 1407,
+      "ull": 1408,
+      "ER": 1409,
+      "reg": 1410,
+      "Ġpe": 1411,
+      "Ġtotal": 1412,
+      "Ġlead": 1413,
+      "==": 1414,
+      "iod": 1415,
+      "Ġassum": 1416,
+      "Ġchang": 1417,
+      "Ġgra": 1418,
+      "MI": 1419,
+      "Ġcomput": 1420,
+      "Ġcomb": 1421,
+      "Ġinformation": 1422,
+      "Ġdesign": 1423,
+      "Ġiniti": 1424,
+      "Ġfrequ": 1425,
+      "imension": 1426,
+      "cop": 1427,
+      "Ġproperties": 1428,
+      "Ġconsider": 1429,
+      "Ġlevels": 1430,
+      "ene": 1431,
+      "Ġtype": 1432,
+      "ived": 1433,
+      "ĠHe": 1434,
+      "ependent": 1435,
+      "Ġapplic": 1436,
+      "Ġinves": 1437,
+      "Ġprevious": 1438,
+      "aw": 1439,
+      "Ġspace": 1440,
+      "Ġprovid": 1441,
+      "hyl": 1442,
+      "Ġinvestig": 1443,
+      "Ġapproach": 1444,
+      "aterial": 1445,
+      "onse": 1446,
+      "lecular": 1447,
+      "Ġparameters": 1448,
+      "Ġphase": 1449,
+      "ulations": 1450,
+      "ubl": 1451,
+      "beta": 1452,
+      "Ġav": 1453,
+      "Ġflu": 1454,
+      "Ġpotential": 1455,
+      "ĠThese": 1456,
+      "sigma": 1457,
+      "lo": 1458,
+      "times": 1459,
+      "Ġoptim": 1460,
+      "ision": 1461,
+      "Ġaff": 1462,
+      "Ġmean": 1463,
+      "Ġbehavi": 1464,
+      "Ġvol": 1465,
+      "orem": 1466,
+      "agne": 1467,
+      "Ġdecre": 1468,
+      "tional": 1469,
+      "Ġsolution": 1470,
+      "Ġhuman": 1471,
+      "ger": 1472,
+      "Ġpaper": 1473,
+      "Ġcompar": 1474,
+      "Ġlower": 1475,
+      "andard": 1476,
+      "Ġcorrel": 1477,
+      "cri": 1478,
+      "Ġcurrent": 1479,
+      "Ġder": 1480,
+      "ission": 1481,
+      "ĠFigure": 1482,
+      "Ġproduc": 1483,
+      "Ġwater": 1484,
+      "ĠTo": 1485,
+      "Ġthose": 1486,
+      "Ġacid": 1487,
+      "Ġcancer": 1488,
+      "Ġlocal": 1489,
+      "ton": 1490,
+      "Ġflow": 1491,
+      "Ġregion": 1492,
+      "Ġhealth": 1493,
+      "Ġimportant": 1494,
+      "ograph": 1495,
+      "abl": 1496,
+      "Ġselec": 1497,
+      "Ġgre": 1498,
+      "Ġindi": 1499,
+      "ade": 1500,
+      "rid": 1501,
+      "Ġshould": 1502,
+      "based": 1503,
+      "Ġabove": 1504,
+      "ld": 1505,
+      "Ġsystems": 1506,
+      "ication": 1507,
+      "Ġed": 1508,
+      "Ġtyp": 1509,
+      "Ġphys": 1510,
+      "oper": 1511,
+      "Ġcompon": 1512,
+      "ON": 1513,
+      "Ġsuper": 1514,
+      "ga": 1515,
+      "hemical": 1516,
+      "isk": 1517,
+      "oph": 1518,
+      "Ġhy": 1519,
+      "Ġanaly": 1520,
+      "inu": 1521,
+      "Ġtarget": 1522,
+      "ĠAd": 1523,
+      "Ġpat": 1524,
+      "gamma": 1525,
+      "Ġsamples": 1526,
+      "Ġsl": 1527,
+      "Ġpart": 1528,
+      "olds": 1529,
+      "Ġbel": 1530,
+      "imum": 1531,
+      "ĠIm": 1532,
+      "Ġdisease": 1533,
+      "II": 1534,
+      "ists": 1535,
+      "iver": 1536,
+      "Ġperformance": 1537,
+      "ĠĠĠĠĠĠĠĠĠĠĠ": 1538,
+      "gle": 1539,
+      "Ġox": 1540,
+      "ndom": 1541,
+      "ĠĠĠĠĠ": 1542,
+      "Ġbecause": 1543,
+      "ayer": 1544,
+      "Ġrange": 1545,
+      "Ġcoun": 1546,
+      "Ġincreased": 1547,
+      "och": 1548,
+      "onal": 1549,
+      "Ġvery": 1550,
+      "Ġdynam": 1551,
+      "anti": 1552,
+      "Ġadd": 1553,
+      "Ġinhib": 1554,
+      "Ġmethods": 1555,
+      "idence": 1556,
+      "inical": 1557,
+      "erence": 1558,
+      "ival": 1559,
+      "ule": 1560,
+      "Ġfactor": 1561,
+      "Ġfin": 1562,
+      "ints": 1563,
+      "viron": 1564,
+      "Ġsour": 1565,
+      "verage": 1566,
+      "equ": 1567,
+      "Ġear": 1568,
+      "Ġshowed": 1569,
+      "ites": 1570,
+      "Ġperformed": 1571,
+      "Ġrese": 1572,
+      "ĠEn": 1573,
+      "Ġspecies": 1574,
+      "AC": 1575,
+      "ĠCl": 1576,
+      "hip": 1577,
+      "tilde": 1578,
+      "io": 1579,
+      "ately": 1580,
+      "Th": 1581,
+      "ody": 1582,
+      "Ġincrease": 1583,
+      "ĠPh": 1584,
+      "âĢĿ": 1585,
+      "Ġshows": 1586,
+      "ĠAc": 1587,
+      "Ġpost": 1588,
+      "ording": 1589,
+      "ences": 1590,
+      "oy": 1591,
+      "ner": 1592,
+      "Ġresponse": 1593,
+      "Ġoccur": 1594,
+      "rho": 1595,
+      "Ġperiod": 1596,
+      "ars": 1597,
+      "Ġred": 1598,
+      "ĠOn": 1599,
+      "Ġdensity": 1600,
+      "Ġexample": 1601,
+      "get": 1602,
+      "Ġreal": 1603,
+      "ĠCount": 1604,
+      "acy": 1605,
+      "Ġpower": 1606,
+      "Ġabs": 1607,
+      "ital": 1608,
+      "Ġprim": 1609,
+      "âĢIJ": 1610,
+      "Ġdefined": 1611,
+      "Ġnormal": 1612,
+      "aj": 1613,
+      "Ġinst": 1614,
+      "Ġallow": 1615,
+      "Ġpossible": 1616,
+      "Ġvis": 1617,
+      "Ġreported": 1618,
+      "Ġsignal": 1619,
+      "theta": 1620,
+      "Ġden": 1621,
+      "ables": 1622,
+      "Ġdeg": 1623,
+      "Ġindivid": 1624,
+      "agnetic": 1625,
+      "Ġgroups": 1626,
+      "ae": 1627,
+      "arrow": 1628,
+      "Ġstat": 1629,
+      "Ġmechanism": 1630,
+      "osp": 1631,
+      "mer": 1632,
+      "other": 1633,
+      "Ġprot": 1634,
+      "Ġcases": 1635,
+      "Ġcr": 1636,
+      "Ġte": 1637,
+      "Ġintegr": 1638,
+      "ets": 1639,
+      "Ġdevelopment": 1640,
+      "Ġrandom": 1641,
+      "Ġinvol": 1642,
+      "Ġincluding": 1643,
+      "Ġerr": 1644,
+      "gram": 1645,
+      "Ġparticular": 1646,
+      "eps": 1647,
+      "Ġstandard": 1648,
+      "position": 1649,
+      "Ġcontrib": 1650,
+      "sequ": 1651,
+      "Ġmany": 1652,
+      "Ġfurther": 1653,
+      "Ġsignificantly": 1654,
+      "ators": 1655,
+      "urb": 1656,
+      "Ġagain": 1657,
+      "bar": 1658,
+      "Ġwithout": 1659,
+      "Ġsever": 1660,
+      "Ġtop": 1661,
+      "ret": 1662,
+      "led": 1663,
+      "Ġmatrix": 1664,
+      "Ġspecific": 1665,
+      "ateg": 1666,
+      "ĨĴ": 1667,
+      "Ġdirect": 1668,
+      "Ġsample": 1669,
+      "Ġthem": 1670,
+      "SA": 1671,
+      "oint": 1672,
+      "Ġrole": 1673,
+      "Ġchanges": 1674,
+      "raction": 1675,
+      "Ġsum": 1676,
+      "Ġindividual": 1677,
+      "IN": 1678,
+      "Ġimmun": 1679,
+      "ced": 1680,
+      "oh": 1681,
+      "Ġstrong": 1682,
+      "Ġep": 1683,
+      "Ġlinear": 1684,
+      "ually": 1685,
+      "delta": 1686,
+      "way": 1687,
+      "asing": 1688,
+      "Ġtim": 1689,
+      "Ġvi": 1690,
+      "ison": 1691,
+      "Ġfunctions": 1692,
+      "Ġamong": 1693,
+      "Ġsee": 1694,
+      "erest": 1695,
+      "Ġgrowth": 1696,
+      "Ġrati": 1697,
+      "ĠSc": 1698,
+      "ixed": 1699,
+      "RNA": 1700,
+      "eed": 1701,
+      "tau": 1702,
+      "Ġent": 1703,
+      "Ġdr": 1704,
+      "ores": 1705,
+      "Ġapproxim": 1706,
+      "ful": 1707,
+      "Ġrele": 1708,
+      "Ġfactors": 1709,
+      "Ġdiscuss": 1710,
+      "Ġphot": 1711,
+      "Ġproposed": 1712,
+      "ero": 1713,
+      "omega": 1714,
+      "Ġfour": 1715,
+      "astic": 1716,
+      "Ġyears": 1717,
+      "hesis": 1718,
+      "ique": 1719,
+      "Ġmaterial": 1720,
+      "Ġbre": 1721,
+      "Ġprof": 1722,
+      "ĠAp": 1723,
+      "Ġneg": 1724,
+      "Ġbu": 1725,
+      "Ġassess": 1726,
+      "ĠâĢľ": 1727,
+      "Ġvir": 1728,
+      "atter": 1729,
+      "Ġdescribed": 1730,
+      "istics": 1731,
+      "Ġcompos": 1732,
+      "az": 1733,
+      "struc": 1734,
+      "Ġtum": 1735,
+      "partial": 1736,
+      "af": 1737,
+      "Ġwho": 1738,
+      "atal": 1739,
+      "Ġdemonstr": 1740,
+      "ances": 1741,
+      "yt": 1742,
+      "Ġremain": 1743,
+      "Ġless": 1744,
+      "Ġpositive": 1745,
+      "omic": 1746,
+      "Ġsince": 1747,
+      "ogn": 1748,
+      "Ġcondition": 1749,
+      "::": 1750,
+      "Ġdoes": 1751,
+      "tice": 1752,
+      "osph": 1753,
+      "Ġprov": 1754,
+      "ĠCO": 1755,
+      "Ġrat": 1756,
+      "Ġterms": 1757,
+      "box": 1758,
+      "Ġtak": 1759,
+      "Ġpattern": 1760,
+      "ale": 1761,
+      "Ġnan": 1762,
+      "ules": 1763,
+      "Ġmut": 1764,
+      "ished": 1765,
+      "Ġrelated": 1766,
+      "Ġtheory": 1767,
+      "bol": 1768,
+      "cdot": 1769,
+      "vironment": 1770,
+      "air": 1771,
+      "ivers": 1772,
+      "ĠAr": 1773,
+      "Ġï£": 1774,
+      "ressed": 1775,
+      "Ġâī¤": 1776,
+      "ĠMet": 1777,
+      "ID": 1778,
+      "ults": 1779,
+      "Ġβ": 1780,
+      "Ġdat": 1781,
+      "pose": 1782,
+      "Ġorig": 1783,
+      "Ġreturn": 1784,
+      "Ġchange": 1785,
+      "Ġlarg": 1786,
+      "au": 1787,
+      "aces": 1788,
+      "Ġarea": 1789,
+      "Ġgenes": 1790,
+      "AS": 1791,
+      "Ġhydro": 1792,
+      "Ġconsist": 1793,
+      "man": 1794,
+      "Ġresearch": 1795,
+      "ĠDe": 1796,
+      "Ġorgan": 1797,
+      "ask": 1798,
+      "Ġback": 1799,
+      "Ġfollows": 1800,
+      "ung": 1801,
+      "roll": 1802,
+      "Ġequation": 1803,
+      "plied": 1804,
+      "tr": 1805,
+      "Ġcorresponding": 1806,
+      "odes": 1807,
+      "ested": 1808,
+      "Ġrelations": 1809,
+      "nal": 1810,
+      "Ġfr": 1811,
+      "Ġlimit": 1812,
+      "mit": 1813,
+      "Ġoff": 1814,
+      "uted": 1815,
+      "Ġrisk": 1816,
+      "read": 1817,
+      "Ġknown": 1818,
+      "plit": 1819,
+      "tivity": 1820,
+      "Ġsequence": 1821,
+      "Ġconsidered": 1822,
+      "xi": 1823,
+      "ĠMod": 1824,
+      "vity": 1825,
+      "Ġnuc": 1826,
+      "cle": 1827,
+      "ices": 1828,
+      "Ġlength": 1829,
+      "Ġseveral": 1830,
+      "sing": 1831,
+      "oot": 1832,
+      "not": 1833,
+      "Ġstress": 1834,
+      "ĠIf": 1835,
+      "CT": 1836,
+      "roph": 1837,
+      "Ġcommun": 1838,
+      "Ġclust": 1839,
+      "ĠLe": 1840,
+      "me": 1841,
+      "antum": 1842,
+      "Ġmemb": 1843,
+      "Ġlab": 1844,
+      "Ġeven": 1845,
+      "Ġinflu": 1846,
+      "ck": 1847,
+      "ĠÃĹ": 1848,
+      "Ġlog": 1849,
+      "ving": 1850,
+      "ests": 1851,
+      "Ġhis": 1852,
+      "ank": 1853,
+      "ĠInd": 1854,
+      "actions": 1855,
+      "fty": 1856,
+      "mod": 1857,
+      "Ġreview": 1858,
+      "though": 1859,
+      "Ġeffici": 1860,
+      "Ġmap": 1861,
+      "infty": 1862,
+      "Ġbeing": 1863,
+      "land": 1864,
+      "Ġclinical": 1865,
+      "Ġmeasured": 1866,
+      "ering": 1867,
+      "ĠTable": 1868,
+      "Ġshe": 1869,
+      "see": 1870,
+      "Ġsection": 1871,
+      "Ġavail": 1872,
+      "omen": 1873,
+      "Ġvers": 1874,
+      "Ġdel": 1875,
+      "ither": 1876,
+      "eration": 1877,
+      "Ġhand": 1878,
+      "Ġcontinu": 1879,
+      "Ġconn": 1880,
+      "hors": 1881,
+      "rad": 1882,
+      "Ġfam": 1883,
+      "Ġlear": 1884,
+      "Ġinitial": 1885,
+      "ystem": 1886,
+      "Ġge": 1887,
+      "ĠâĢ²": 1888,
+      "Ġcirc": 1889,
+      "Ġpubl": 1890,
+      "ĠIs": 1891,
+      "Ġvia": 1892,
+      "Ġcommon": 1893,
+      "ife": 1894,
+      "Ġmark": 1895,
+      "Ġever": 1896,
+      "arc": 1897,
+      "big": 1898,
+      "ertain": 1899,
+      "\\\\": 1900,
+      "var": 1901,
+      "As": 1902,
+      "roscop": 1903,
+      "Ġage": 1904,
+      "Ġhow": 1905,
+      "ĠLet": 1906,
+      "struct": 1907,
+      "Ġaverage": 1908,
+      "vant": 1909,
+      "ĠSh": 1910,
+      "imensional": 1911,
+      "SC": 1912,
+      "ape": 1913,
+      "nu": 1914,
+      "Ġloss": 1915,
+      "ason": 1916,
+      "ides": 1917,
+      "Ġpopulation": 1918,
+      "Ġdomain": 1919,
+      "inding": 1920,
+      "we": 1921,
+      "AL": 1922,
+      "Ġaccur": 1923,
+      "ety": 1924,
+      "Ġcaus": 1925,
+      "Delta": 1926,
+      "rapy": 1927,
+      "Ġprom": 1928,
+      "time": 1929,
+      "Ġintro": 1930,
+      "Ġmultiple": 1931,
+      "Ġconstant": 1932,
+      "pling": 1933,
+      "ino": 1934,
+      "ajor": 1935,
+      "ior": 1936,
+      "abol": 1937,
+      "def": 1938,
+      "Ġpoints": 1939,
+      "verse": 1940,
+      "name": 1941,
+      "ĠSe": 1942,
+      "itor": 1943,
+      "Pro": 1944,
+      "arm": 1945,
+      "Ġtiss": 1946,
+      "Ġfib": 1947,
+      "Ġgraph": 1948,
+      "Ġcall": 1949,
+      "atisf": 1950,
+      "Ġconduc": 1951,
+      "dex": 1952,
+      "ĠNe": 1953,
+      "Ġpers": 1954,
+      "ern": 1955,
+      "CR": 1956,
+      "angle": 1957,
+      "Ġfrequency": 1958,
+      "AP": 1959,
+      "Ġpresented": 1960,
+      "amp": 1961,
+      "Ġbefore": 1962,
+      "ords": 1963,
+      "Ġinput": 1964,
+      "ĠâĨĴ": 1965,
+      "Ġparticip": 1966,
+      "OR": 1967,
+      "Ġchild": 1968,
+      "Ġcre": 1969,
+      "fficient": 1970,
+      "Ġsepar": 1971,
+      "uration": 1972,
+      "α": 1973,
+      "Ġexist": 1974,
+      "ised": 1975,
+      "Ġlight": 1976,
+      "imal": 1977,
+      "****": 1978,
+      "ĠDNA": 1979,
+      "hel": 1980,
+      "Ġinterest": 1981,
+      "bf": 1982,
+      "ke": 1983,
+      "Ġcollec": 1984,
+      "Ġtrain": 1985,
+      "ai": 1986,
+      "ĠPl": 1987,
+      "Ġλ": 1988,
+      "ĠCo": 1989,
+      "Ġimage": 1990,
+      "Ġhyp": 1991,
+      "oma": 1992,
+      "Ġweight": 1993,
+      "Ġcross": 1994,
+      "rt": 1995,
+      "Ġdifference": 1996,
+      "Ġfeatures": 1997,
+      "medi": 1998,
+      "type": 1999,
+      "Ġpress": 2000,
+      "IC": 2001,
+      "Ġtherm": 2002,
+      "Ġstates": 2003,
+      "ustr": 2004,
+      "till": 2005,
+      "Ġhist": 2006,
+      "Ġratio": 2007,
+      "aging": 2008,
+      "ĠAll": 2009,
+      "Ġhel": 2010,
+      "bon": 2011,
+      "Ġbehavior": 2012,
+      "Ġpri": 2013,
+      "Ġsynt": 2014,
+      "ended": 2015,
+      "ĠInt": 2016,
+      "tt": 2017,
+      "Ġvarious": 2018,
+      "rect": 2019,
+      "Ġprec": 2020,
+      "Ġtimes": 2021,
+      "MS": 2022,
+      "Ġanalyz": 2023,
+      "Ġcare": 2024,
+      "mat": 2025,
+      "Ġalong": 2026,
+      "Ġpur": 2027,
+      "atively": 2028,
+      "Ġstar": 2029,
+      "jects": 2030,
+      "ii": 2031,
+      "istance": 2032,
+      "ĠThen": 2033,
+      "AN": 2034,
+      "Ġparameter": 2035,
+      "ulate": 2036,
+      "Ġevery": 2037,
+      "Ġsatisf": 2038,
+      "Ġdetermined": 2039,
+      "ina": 2040,
+      "rane": 2041,
+      "Ġpair": 2042,
+      "ool": 2043,
+      "Table": 2044,
+      "Ġthus": 2045,
+      "ogene": 2046,
+      "ĠÏĨ": 2047,
+      "Ġprogram": 2048,
+      "asc": 2049,
+      "Ġenvironment": 2050,
+      "MP": 2051,
+      "Ġread": 2052,
+      "Ġach": 2053,
+      "Ġpresence": 2054,
+      "Ġmice": 2055,
+      "For": 2056,
+      "Ġproduction": 2057,
+      "Ġdifferences": 2058,
+      "Ġprovide": 2059,
+      "ste": 2060,
+      "ames": 2061,
+      "ĉĠ": 2062,
+      "Ġ±": 2063,
+      "roup": 2064,
+      "Ġelectron": 2065,
+      "Ġhyper": 2066,
+      "bit": 2067,
+      "ĠRec": 2068,
+      "Ġvector": 2069,
+      "uble": 2070,
+      "rangle": 2071,
+      "Ġwr": 2072,
+      "wide": 2073,
+      "ĠâĬ": 2074,
+      "rack": 2075,
+      "ryst": 2076,
+      "Ġinj": 2077,
+      "ega": 2078,
+      "Ġwhe": 2079,
+      "psilon": 2080,
+      "Ġagainst": 2081,
+      "Ġdiagn": 2082,
+      "Ġhom": 2083,
+      "Ġachie": 2084,
+      "ns": 2085,
+      "Ġrece": 2086,
+      "--------": 2087,
+      "Ġavailable": 2088,
+      "inf": 2089,
+      "Ġsuc": 2090,
+      "Ġgu": 2091,
+      "Ġmajor": 2092,
+      "ĠThus": 2093,
+      "ware": 2094,
+      "Ġsupport": 2095,
+      "lor": 2096,
+      "Ġexperimental": 2097,
+      "ĠMo": 2098,
+      "Ġconcentration": 2099,
+      "tics": 2100,
+      "Ġnec": 2101,
+      "Ġphen": 2102,
+      "sq": 2103,
+      "Ġclos": 2104,
+      "sub": 2105,
+      "Ġknow": 2106,
+      "Ġformation": 2107,
+      "Ġdid": 2108,
+      "ouse": 2109,
+      "inary": 2110,
+      "ict": 2111,
+      "ĠCD": 2112,
+      "This": 2113,
+      "less": 2114,
+      "Ġnear": 2115,
+      "Ġimprove": 2116,
+      "abil": 2117,
+      "Ġreve": 2118,
+      "Ġexperiments": 2119,
+      "ience": 2120,
+      "ula": 2121,
+      "ored": 2122,
+      "Ġunc": 2123,
+      "__": 2124,
+      "Ġapplied": 2125,
+      "Ġreduced": 2126,
+      "Ġdetail": 2127,
+      "stand": 2128,
+      "Ġcho": 2129,
+      "omy": 2130,
+      "Ġcalculated": 2131,
+      "Ġenh": 2132,
+      "LES": 2133,
+      "itro": 2134,
+      "Ġrespons": 2135,
+      "Ġest": 2136,
+      "Ġmi": 2137,
+      "Ġcoe": 2138,
+      "ĠTherefore": 2139,
+      "ĠMore": 2140,
+      "bl": 2141,
+      "anced": 2142,
+      "ume": 2143,
+      "Ġband": 2144,
+      "Ġact": 2145,
+      "Ġeither": 2146,
+      "omes": 2147,
+      "ĠGen": 2148,
+      "vare": 2149,
+      "ET": 2150,
+      "reen": 2151,
+      "ĠPar": 2152,
+      "ĠSim": 2153,
+      "Ġidentified": 2154,
+      "Ġinteraction": 2155,
+      "Ġmade": 2156,
+      "Ġsource": 2157,
+      "tis": 2158,
+      "ots": 2159,
+      "mega": 2160,
+      "Ġserv": 2161,
+      "ms": 2162,
+      "alysis": 2163,
+      "vent": 2164,
+      "ense": 2165,
+      "gl": 2166,
+      "Ġlines": 2167,
+      "Ġappear": 2168,
+      "tif": 2169,
+      "Ġfree": 2170,
+      "oms": 2171,
+      "ining": 2172,
+      "eren": 2173,
+      "Ġchann": 2174,
+      "varepsilon": 2175,
+      "sim": 2176,
+      "Ġcou": 2177,
+      "°": 2178,
+      "Ġerror": 2179,
+      "Ġquanti": 2180,
+      "ĠEq": 2181,
+      "by": 2182,
+      "ĠII": 2183,
+      "tex": 2184,
+      "ĠSch": 2185,
+      "sqrt": 2186,
+      "ocus": 2187,
+      "Ġdev": 2188,
+      "quad": 2189,
+      "ters": 2190,
+      "Ġrelationship": 2191,
+      "oll": 2192,
+      "Ġgo": 2193,
+      "Ġwave": 2194,
+      "Ġleft": 2195,
+      "ways": 2196,
+      "hi": 2197,
+      "Ġright": 2198,
+      "obal": 2199,
+      "Ġdown": 2200,
+      "uk": 2201,
+      "Ġcoll": 2202,
+      "Ġmagnetic": 2203,
+      "Ġprog": 2204,
+      "dots": 2205,
+      "Ġstrateg": 2206,
+      "bs": 2207,
+      "unction": 2208,
+      "Ġenc": 2209,
+      "Ġclear": 2210,
+      "Ġcost": 2211,
+      "geb": 2212,
+      "etter": 2213,
+      "MILES": 2214,
+      "lamm": 2215,
+      "Ġmust": 2216,
+      "Ġeffective": 2217,
+      "Ġexc": 2218,
+      "Ġplas": 2219,
+      "Ġsuggest": 2220,
+      "itions": 2221,
+      "Ġleast": 2222,
+      "ying": 2223,
+      "lying": 2224,
+      "Ġlik": 2225,
+      "Omega": 2226,
+      "aking": 2227,
+      "Ġmaximum": 2228,
+      "Ġrelative": 2229,
+      "é": 2230,
+      "Ġaccording": 2231,
+      "ient": 2232,
+      "Ġway": 2233,
+      "Ġsem": 2234,
+      "atural": 2235,
+      "like": 2236,
+      "resh": 2237,
+      "ĠMe": 2238,
+      "Ps": 2239,
+      "ĠTrans": 2240,
+      "isc": 2241,
+      "Ġprac": 2242,
+      "Ġrun": 2243,
+      "Ġconver": 2244,
+      "Ġsk": 2245,
+      "Ġyield": 2246,
+      "geq": 2247,
+      "ably": 2248,
+      "Ġantib": 2249,
+      "izing": 2250,
+      "β": 2251,
+      "mission": 2252,
+      "Ġnow": 2253,
+      "Ġdetection": 2254,
+      "eloc": 2255,
+      "Ġget": 2256,
+      "ert": 2257,
+      "Ġvariables": 2258,
+      "Ġopen": 2259,
+      "Ġpressure": 2260,
+      "Ġstrain": 2261,
+      "ument": 2262,
+      "ĠFurther": 2263,
+      "Ġquantum": 2264,
+      "Ġimplement": 2265,
+      "Ġearly": 2266,
+      "Ġframe": 2267,
+      "Ġshort": 2268,
+      "Ġdrug": 2269,
+      "Ġrequired": 2270,
+      "PS": 2271,
+      "Ġmy": 2272,
+      "Ġmuch": 2273,
+      "Ġmem": 2274,
+      "CC": 2275,
+      "Ġquality": 2276,
+      "Ġproteins": 2277,
+      "Ġlayer": 2278,
+      "Ġques": 2279,
+      "Ġrecept": 2280,
+      "Ġhere": 2281,
+      "Ġproced": 2282,
+      "ured": 2283,
+      "Ġdeveloped": 2284,
+      "Ġposition": 2285,
+      "rum": 2286,
+      "Ġlat": 2287,
+      "Ġincreasing": 2288,
+      "EM": 2289,
+      "Ġmeasurements": 2290,
+      "Ġben": 2291,
+      "Ġisol": 2292,
+      "wh": 2293,
+      "To": 2294,
+      "Ġvalid": 2295,
+      "Ġfunctional": 2296,
+      "emma": 2297,
+      "...": 2298,
+      "orld": 2299,
+      "ries": 2300,
+      "Ġprobability": 2301,
+      "ĠNew": 2302,
+      "Ġmm": 2303,
+      "OS": 2304,
+      "AD": 2305,
+      "Ġδ": 2306,
+      "Ġscale": 2307,
+      "ĠFe": 2308,
+      "ĠTheorem": 2309,
+      "ĠQu": 2310,
+      "Ġcomponents": 2311,
+      "Ġblood": 2312,
+      "ĠÏĥ": 2313,
+      "acc": 2314,
+      "Ġbetter": 2315,
+      "Ġstep": 2316,
+      "Ġγ": 2317,
+      "Ġfac": 2318,
+      "aneous": 2319,
+      "Ġload": 2320,
+      "Ġmetabol": 2321,
+      "Ġevolution": 2322,
+      "son": 2323,
+      "ream": 2324,
+      "Ġeas": 2325,
+      "ird": 2326,
+      "dimensional": 2327,
+      "bor": 2328,
+      "Ġmus": 2329,
+      "Ġequations": 2330,
+      "psi": 2331,
+      "order": 2332,
+      "olar": 2333,
+      "Ġnumer": 2334,
+      "Ġkey": 2335,
+      "orth": 2336,
+      "Ġsimple": 2337,
+      "ift": 2338,
+      "cale": 2339,
+      "Ġindex": 2340,
+      "ĠâĢĵ": 2341,
+      "Ġconcentr": 2342,
+      "ges": 2343,
+      "Ġnegative": 2344,
+      "Ġveloc": 2345,
+      "Ġax": 2346,
+      "ĠEff": 2347,
+      "Ġfinite": 2348,
+      "Ġill": 2349,
+      "ching": 2350,
+      "Ġpatient": 2351,
+      "epsilon": 2352,
+      "Ġmen": 2353,
+      "Ġcri": 2354,
+      "IS": 2355,
+      "Cl": 2356,
+      "Ġconcl": 2357,
+      "Ġθ": 2358,
+      "ibility": 2359,
+      "Ġsymmet": 2360,
+      "enter": 2361,
+      "Ġdistance": 2362,
+      "Ġpolym": 2363,
+      "ights": 2364,
+      "Ġcult": 2365,
+      "Ġpeak": 2366,
+      "Ġacross": 2367,
+      "inition": 2368,
+      "Ġlet": 2369,
+      "Ġconstruc": 2370,
+      "Ġincluded": 2371,
+      "Ġhowever": 2372,
+      "Ġregions": 2373,
+      "Ġlearning": 2374,
+      "Ġevidence": 2375,
+      "inally": 2376,
+      "Ġneut": 2377,
+      "itation": 2378,
+      "Ġwhether": 2379,
+      "Ġoutput": 2380,
+      "ĠSection": 2381,
+      "Ġgood": 2382,
+      "IT": 2383,
+      "uation": 2384,
+      "Ġtypes": 2385,
+      "bm": 2386,
+      "cos": 2387,
+      "with": 2388,
+      "lim": 2389,
+      "otic": 2390,
+      "Ġstill": 2391,
+      "Ġdays": 2392,
+      "Ġstudied": 2393,
+      "Ġimages": 2394,
+      "ble": 2395,
+      "Ġarg": 2396,
+      "linear": 2397,
+      "Ġprocesses": 2398,
+      "Ġwid": 2399,
+      "Ġtraining": 2400,
+      "Ġindependent": 2401,
+      "plac": 2402,
+      "Ġresid": 2403,
+      "Ġsuccess": 2404,
+      "Ġnucle": 2405,
+      "GF": 2406,
+      "let": 2407,
+      "ploy": 2408,
+      "Ġtumor": 2409,
+      "Gamma": 2410,
+      "Ġtherefore": 2411,
+      "rast": 2412,
+      "Ġfocus": 2413,
+      "ash": 2414,
+      "Ġbelow": 2415,
+      "ially": 2416,
+      "Ġcomparison": 2417,
+      "Ġadj": 2418,
+      "Ġlike": 2419,
+      "Ġmolecular": 2420,
+      "ried": 2421,
+      "Ġfit": 2422,
+      "ĠDi": 2423,
+      "log": 2424,
+      "Ġplay": 2425,
+      "work": 2426,
+      "ections": 2427,
+      "Ġelectro": 2428,
+      "uit": 2429,
+      "more": 2430,
+      "Ġmight": 2431,
+      "Ġanalys": 2432,
+      "Ġmeans": 2433,
+      "Ġcorrelation": 2434,
+      "kn": 2435,
+      "Ġcontroll": 2436,
+      "IV": 2437,
+      "Ch": 2438,
+      "pec": 2439,
+      "rag": 2440,
+      "Ġmagn": 2441,
+      "Ġphysical": 2442,
+      "ION": 2443,
+      "Ġreveal": 2444,
+      "Ġphosph": 2445,
+      "Ġrates": 2446,
+      "Ġlarger": 2447,
+      "Ġstim": 2448,
+      "Ġsoft": 2449,
+      "Ġcompound": 2450,
+      "be": 2451,
+      "chi": 2452,
+      "ĠNo": 2453,
+      "Ġimpact": 2454,
+      "tor": 2455,
+      "Ġprimary": 2456,
+      "ocial": 2457,
+      "Ġapplication": 2458,
+      "Ġsolutions": 2459,
+      "duce": 2460,
+      "Ġcharacteristics": 2461,
+      "Ġelements": 2462,
+      "Ġview": 2463,
+      "Ġlater": 2464,
+      "uture": 2465,
+      "Ġfamily": 2466,
+      "rial": 2467,
+      "Ġtranscri": 2468,
+      "orption": 2469,
+      "Ġsw": 2470,
+      "CD": 2471,
+      "ED": 2472,
+      "Ġemb": 2473,
+      "Ġzero": 2474,
+      "ols": 2475,
+      "Ġlife": 2476,
+      "cep": 2477,
+      "ĠLi": 2478,
+      "ths": 2479,
+      "Ġseries": 2480,
+      "Ġaround": 2481,
+      "Ġtransition": 2482,
+      "ĠCor": 2483,
+      "ĠâĪĤ": 2484,
+      "Ġdatas": 2485,
+      "Ġher": 2486,
+      "ĠBy": 2487,
+      "AM": 2488,
+      "spec": 2489,
+      "oles": 2490,
+      "ography": 2491,
+      "tle": 2492,
+      "ĠCar": 2493,
+      "alle": 2494,
+      "Ġestabl": 2495,
+      "agement": 2496,
+      "Ġschem": 2497,
+      "ground": 2498,
+      "Ġfail": 2499,
+      "Ġexpected": 2500,
+      "Ġrequire": 2501,
+      "array": 2502,
+      "Ġexperiment": 2503,
+      "Ġelement": 2504,
+      "Ġneu": 2505,
+      "Ġgenerated": 2506,
+      "Ġsite": 2507,
+      "ĠCont": 2508,
+      "ĠRNA": 2509,
+      "eral": 2510,
+      "Ġcontent": 2511,
+      "Ġbacter": 2512,
+      "ler": 2513,
+      "Ġtransfer": 2514,
+      "ulf": 2515,
+      "rightarrow": 2516,
+      "any": 2517,
+      "ĠSince": 2518,
+      "induced": 2519,
+      "Ġreaction": 2520,
+      "heck": 2521,
+      "Ġstructures": 2522,
+      "Ġcount": 2523,
+      "Ġdetermine": 2524,
+      "zym": 2525,
+      "ĠBl": 2526,
+      "Ġunderstand": 2527,
+      "ocal": 2528,
+      "Ġsyn": 2529,
+      "Ġpoly": 2530,
+      "ury": 2531,
+      "Ġbest": 2532,
+      "Ġfixed": 2533,
+      "reng": 2534,
+      "Ġchemical": 2535,
+      "Ġtissue": 2536,
+      "Ġpul": 2537,
+      "Ġboundary": 2538,
+      "ising": 2539,
+      "Ġbro": 2540,
+      "atistical": 2541,
+      "icity": 2542,
+      "sk": 2543,
+      "ring": 2544,
+      "Ġlast": 2545,
+      "Ġchildren": 2546,
+      "rim": 2547,
+      "Ġreduction": 2548,
+      "Ġspin": 2549,
+      "Ġbody": 2550,
+      "operator": 2551,
+      "vari": 2552,
+      "Ġdiv": 2553,
+      "ymbol": 2554,
+      "Ġmal": 2555,
+      "Ġspati": 2556,
+      "ah": 2557,
+      "ĠBi": 2558,
+      "back": 2559,
+      "sy": 2560,
+      "Ġseen": 2561,
+      "ĠWith": 2562,
+      "ids": 2563,
+      "plications": 2564,
+      "Ġnecess": 2565,
+      "Ġside": 2566,
+      "Ġbrain": 2567,
+      "Ġfew": 2568,
+      "Ġapplications": 2569,
+      "utes": 2570,
+      "aches": 2571,
+      "Ġactive": 2572,
+      "varphi": 2573,
+      "term": 2574,
+      "Ġmom": 2575,
+      "iversity": 2576,
+      "Ġfinal": 2577,
+      "ledge": 2578,
+      "Ġdynamics": 2579,
+      "aving": 2580,
+      "erc": 2581,
+      "orphism": 2582,
+      "ones": 2583,
+      "off": 2584,
+      "pm": 2585,
+      "Ġaction": 2586,
+      "Ġnatural": 2587,
+      "ĠGe": 2588,
+      "Ġyou": 2589,
+      "lex": 2590,
+      "ĠĠĠĠĠĠ": 2591,
+      "stit": 2592,
+      "Ġgas": 2593,
+      "Ġmake": 2594,
+      "Ġinduced": 2595,
+      "ĠAfter": 2596,
+      "ĠWh": 2597,
+      "Ġcomponent": 2598,
+      "Ġinfection": 2599,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 2600,
+      "Ġconfir": 2601,
+      "igen": 2602,
+      "ĠSystem": 2603,
+      "ticle": 2604,
+      "Ġprovided": 2605,
+      "ternal": 2606,
+      "bers": 2607,
+      "OD": 2608,
+      "ĠInter": 2609,
+      "ott": 2610,
+      "aves": 2611,
+      "ĠStud": 2612,
+      "py": 2613,
+      "Ġresistance": 2614,
+      "ĠSur": 2615,
+      "atch": 2616,
+      "Ġdim": 2617,
+      "Ġinterp": 2618,
+      "Ġcycl": 2619,
+      "ont": 2620,
+      "iting": 2621,
+      "AG": 2622,
+      "Ġequival": 2623,
+      "otype": 2624,
+      "Ġpreviously": 2625,
+      "Ġadditional": 2626,
+      "outh": 2627,
+      "Ġimpl": 2628,
+      "Ġion": 2629,
+      "Ġir": 2630,
+      "Ġcop": 2631,
+      "Ġhal": 2632,
+      "Ġactivation": 2633,
+      "langle": 2634,
+      "Ġfull": 2635,
+      "SS": 2636,
+      "ĠOp": 2637,
+      "idd": 2638,
+      "Ġproof": 2639,
+      "Ġproblems": 2640,
+      "Ġtransform": 2641,
+      "Ġinteractions": 2642,
+      "Ġsupp": 2643,
+      "des": 2644,
+      "ĠReg": 2645,
+      "operatorname": 2646,
+      "egin": 2647,
+      "Ġcryst": 2648,
+      "Ġincreases": 2649,
+      "ronic": 2650,
+      "Ġadap": 2651,
+      "inant": 2652,
+      "Ġvelocity": 2653,
+      "ĠAss": 2654,
+      "iques": 2655,
+      "Ġcontinuous": 2656,
+      "ĠComp": 2657,
+      "ĠProper": 2658,
+      "Ġprior": 2659,
+      "orb": 2660,
+      "Ġnovel": 2661,
+      "Ġblock": 2662,
+      "Ġvolume": 2663,
+      "Ġregard": 2664,
+      "ometry": 2665,
+      "EC": 2666,
+      "Ġresulting": 2667,
+      "ĠOr": 2668,
+      "Ġcarbon": 2669,
+      "arent": 2670,
+      "Ġbinding": 2671,
+      "ij": 2672,
+      "Ġaccess": 2673,
+      "Ġweak": 2674,
+      "Ġunit": 2675,
+      "Ġide": 2676,
+      "\"\"": 2677,
+      "Ġcm": 2678,
+      "Ġcritical": 2679,
+      "Ġrespect": 2680,
+      "trans": 2681,
+      "Ġâī¥": 2682,
+      "Ġsal": 2683,
+      "ead": 2684,
+      "Ġsimulation": 2685,
+      "Ġcapac": 2686,
+      "itivity": 2687,
+      "Ġrecord": 2688,
+      "rak": 2689,
+      "Ġneur": 2690,
+      "onic": 2691,
+      "ople": 2692,
+      "Ġmg": 2693,
+      "Ġstreng": 2694,
+      "erve": 2695,
+      "Ġreduc": 2696,
+      "Ġpass": 2697,
+      "ordin": 2698,
+      "exp": 2699,
+      "jective": 2700,
+      "ensor": 2701,
+      "Ġparticles": 2702,
+      "Ġair": 2703,
+      "Ġlink": 2704,
+      "ĠÏĦ": 2705,
+      "Ġlist": 2706,
+      "cin": 2707,
+      "ĠOur": 2708,
+      "pri": 2709,
+      "vere": 2710,
+      "ibr": 2711,
+      "iform": 2712,
+      "Ġexplain": 2713,
+      "Ġfem": 2714,
+      "Ġutil": 2715,
+      "St": 2716,
+      "overline": 2717,
+      "Ġoften": 2718,
+      "ery": 2719,
+      "ope": 2720,
+      "ĠUsing": 2721,
+      "begin": 2722,
+      "Ġdifferenti": 2723,
+      "pers": 2724,
+      "self": 2725,
+      "izes": 2726,
+      "Ġconcentrations": 2727,
+      "IR": 2728,
+      "ĠSup": 2729,
+      "Ġbasis": 2730,
+      "Ġinclude": 2731,
+      "ĠBond": 2732,
+      "Ġextrac": 2733,
+      "ĠMethod": 2734,
+      "ĠData": 2735,
+      "ĠDef": 2736,
+      "wn": 2737,
+      "Ġnetworks": 2738,
+      "igned": 2739,
+      "âĢ¢": 2740,
+      "Ġexpressed": 2741,
+      "Ġcontrast": 2742,
+      "esis": 2743,
+      "col": 2744,
+      "inter": 2745,
+      "pid": 2746,
+      "Ġdri": 2747,
+      "Ġdefine": 2748,
+      "Ġinfluence": 2749,
+      "Ġselected": 2750,
+      "EL": 2751,
+      "Ġcontaining": 2752,
+      "Ġsil": 2753,
+      "gebra": 2754,
+      "reat": 2755,
+      "bolds": 2756,
+      "Ġinvestigated": 2757,
+      "ĠCol": 2758,
+      "ymmet": 2759,
+      "ytes": 2760,
+      "Ġmolec": 2761,
+      "Ġinvolved": 2762,
+      "Ġday": 2763,
+      "Ġchain": 2764,
+      "ĠMoreover": 2765,
+      "Ġdiag": 2766,
+      "Ġang": 2767,
+      "Ġlikely": 2768,
+      "Ġspectrum": 2769,
+      "Ġderiv": 2770,
+      "boldsymbol": 2771,
+      "Ġhelp": 2772,
+      "ĠAm": 2773,
+      "Ġtreated": 2774,
+      "Ġvariable": 2775,
+      "ellular": 2776,
+      "ĠDes": 2777,
+      "aps": 2778,
+      "Ġnm": 2779,
+      "ĠÏģ": 2780,
+      "ĠWhen": 2781,
+      "Ġhighly": 2782,
+      "amin": 2783,
+      "Ġwhat": 2784,
+      "related": 2785,
+      "Ġchrom": 2786,
+      "Ġsurv": 2787,
+      "ĠAnalysis": 2788,
+      "Ġsit": 2789,
+      "fact": 2790,
+      "oding": 2791,
+      "Ġproduct": 2792,
+      "Ġevents": 2793,
+      "ras": 2794,
+      "ĠPer": 2795,
+      "max": 2796,
+      "ĠAg": 2797,
+      "cont": 2798,
+      "icro": 2799,
+      "Ġadv": 2800,
+      "Ġcalled": 2801,
+      "Ġdegree": 2802,
+      "AB": 2803,
+      "TR": 2804,
+      "Ġseg": 2805,
+      "ĠCan": 2806,
+      "Ġdemonstrated": 2807,
+      "wise": 2808,
+      "Ġve": 2809,
+      "ĠCa": 2810,
+      "Ġdetected": 2811,
+      "co": 2812,
+      "Ġderived": 2813,
+      "Ġexhib": 2814,
+      "Ġglobal": 2815,
+      "alax": 2816,
+      "ulating": 2817,
+      "Al": 2818,
+      "angu": 2819,
+      "bo": 2820,
+      "Ġrecom": 2821,
+      "Ġfeature": 2822,
+      "dependent": 2823,
+      "Ġrot": 2824,
+      "vention": 2825,
+      "Ġremov": 2826,
+      "Ġwind": 2827,
+      "Ġaccuracy": 2828,
+      "size": 2829,
+      "Ġsumm": 2830,
+      "Ġmeasurement": 2831,
+      "Ġfields": 2832,
+      "wards": 2833,
+      "Ġliter": 2834,
+      "ataly": 2835,
+      "ĠStr": 2836,
+      "Ġreport": 2837,
+      "Ġcentral": 2838,
+      "Ġsqu": 2839,
+      "Ġtherapy": 2840,
+      "hest": 2841,
+      "Ġfeed": 2842,
+      "SMILES": 2843,
+      "ĠAN": 2844,
+      "Ġsites": 2845,
+      "âĢ²": 2846,
+      "ours": 2847,
+      "omal": 2848,
+      "Ġlip": 2849,
+      "Ġanalyzed": 2850,
+      "Ġ°": 2851,
+      "Ġwee": 2852,
+      "tem": 2853,
+      "Ġanother": 2854,
+      "iles": 2855,
+      "Ġcomplete": 2856,
+      "Ġnext": 2857,
+      "ĠOne": 2858,
+      "bi": 2859,
+      "rip": 2860,
+      "state": 2861,
+      "ĠModel": 2862,
+      "Ġfindings": 2863,
+      "ĠPre": 2864,
+      "Ġrecent": 2865,
+      "ascular": 2866,
+      "Ġestimate": 2867,
+      "Ġmechanisms": 2868,
+      "ĠResults": 2869,
+      "Ġparticipants": 2870,
+      "Ġeng": 2871,
+      "most": 2872,
+      "ometric": 2873,
+      "Ġequal": 2874,
+      "Ġrob": 2875,
+      "Ġpolar": 2876,
+      "Ġgenetic": 2877,
+      "Ġbo": 2878,
+      "Ġrest": 2879,
+      "ĠÏĢ": 2880,
+      "Ġrelation": 2881,
+      "Ġquestion": 2882,
+      "epti": 2883,
+      "Ġdiffic": 2884,
+      "ems": 2885,
+      "Ġfuture": 2886,
+      "ify": 2887,
+      "Ġmode": 2888,
+      "Ġmembrane": 2889,
+      "Ġheat": 2890,
+      "Aut": 2891,
+      "ding": 2892,
+      "Ġoxid": 2893,
+      "Ġconfig": 2894,
+      "plication": 2895,
+      "ĠMon": 2896,
+      "allel": 2897,
+      "ided": 2898,
+      "Ġdirection": 2899,
+      "pled": 2900,
+      "Ġprovides": 2901,
+      "Ġindicate": 2902,
+      "Ġsets": 2903,
+      "Ġtechnique": 2904,
+      "Ġmac": 2905,
+      "Ġhypot": 2906,
+      "Ġatten": 2907,
+      "Ġevent": 2908,
+      "Ġstage": 2909,
+      "Ġnode": 2910,
+      "Ġreference": 2911,
+      "Ġupper": 2912,
+      "Ġtechniques": 2913,
+      "Ġgreater": 2914,
+      "Ġdirectly": 2915,
+      "Ġareas": 2916,
+      "Ġdiss": 2917,
+      "hor": 2918,
+      "ĠPol": 2919,
+      "Ġevaluation": 2920,
+      "Ġpatterns": 2921,
+      "ĠAbstract": 2922,
+      "Ġvirus": 2923,
+      "vey": 2924,
+      "PC": 2925,
+      "Ġwomen": 2926,
+      "rient": 2927,
+      "Ġplasma": 2928,
+      "Ġproduced": 2929,
+      "Ġε": 2930,
+      "Ġanalyses": 2931,
+      "ĠSub": 2932,
+      "Ġsetting": 2933,
+      "Ġmoment": 2934,
+      "Ġthermal": 2935,
+      "Ġoptimal": 2936,
+      "Ġtaken": 2937,
+      "Ġrecogn": 2938,
+      "Ġvariation": 2939,
+      "ĠLemma": 2940,
+      "Ġsus": 2941,
+      "frak": 2942,
+      "ĠIL": 2943,
+      "Ġprocedure": 2944,
+      "hood": 2945,
+      "Ġaim": 2946,
+      "aries": 2947,
+      "mathfrak": 2948,
+      "Ġplant": 2949,
+      "brid": 2950,
+      "elect": 2951,
+      "Ġvisual": 2952,
+      "urs": 2953,
+      "cence": 2954,
+      "Ġfive": 2955,
+      "Ġspatial": 2956,
+      "Ġreceptor": 2957,
+      "Ġindicated": 2958,
+      "Ġess": 2959,
+      "Ġconsistent": 2960,
+      "Ġturn": 2961,
+      "tices": 2962,
+      "Ġexists": 2963,
+      "ectors": 2964,
+      "Ġenzym": 2965,
+      "meric": 2966,
+      "Ġnoise": 2967,
+      "Ġground": 2968,
+      "Ġestimated": 2969,
+      "eline": 2970,
+      "Ġchannel": 2971,
+      "tition": 2972,
+      "Ġdiscussed": 2973,
+      "omer": 2974,
+      "otes": 2975,
+      "Ġexact": 2976,
+      "ĠSec": 2977,
+      "Ġtake": 2978,
+      "Ġknowledge": 2979,
+      "Ġprop": 2980,
+      "Ġinflamm": 2981,
+      "Ġdouble": 2982,
+      "It": 2983,
+      "Ġcontext": 2984,
+      "ĠMed": 2985,
+      "MA": 2986,
+      "Ġfat": 2987,
+      "ams": 2988,
+      "data": 2989,
+      "ands": 2990,
+      "Ġcardi": 2991,
+      "ĠFurthermore": 2992,
+      "ocy": 2993,
+      "Ġobservations": 2994,
+      "apping": 2995,
+      "ĠInf": 2996,
+      "omial": 2997,
+      "Ġpublic": 2998,
+      "Ġemploy": 2999,
+      "Ġreason": 3000,
+      "ygen": 3001,
+      "Ġfollowed": 3002,
+      "Ġamount": 3003,
+      "Ġcertain": 3004,
+      "which": 3005,
+      "otyp": 3006,
+      "ĠCell": 3007,
+      "Ġchall": 3008,
+      "Ġparticle": 3009,
+      "ambda": 3010,
+      "Ġens": 3011,
+      "Ġpeople": 3012,
+      "ault": 3013,
+      "ĠUnd": 3014,
+      "ĠBe": 3015,
+      "umin": 3016,
+      "roscopy": 3017,
+      "MR": 3018,
+      "lation": 3019,
+      "Ġrepe": 3020,
+      "Ġable": 3021,
+      "ĠSo": 3022,
+      "ĠâĪŀ": 3023,
+      "Ġenti": 3024,
+      "Ġmove": 3025,
+      "Ġtrac": 3026,
+      "CO": 3027,
+      "Ġheter": 3028,
+      "Ġspeed": 3029,
+      "Ġefficiency": 3030,
+      "Ġoptical": 3031,
+      "Ġcombination": 3032,
+      "eness": 3033,
+      "Ġchem": 3034,
+      "LE": 3035,
+      "appa": 3036,
+      "Ġdecrease": 3037,
+      "μ": 3038,
+      "ped": 3039,
+      "note": 3040,
+      "ĠMulti": 3041,
+      "Ġaltern": 3042,
+      "Ġassume": 3043,
+      "ĠForm": 3044,
+      "stric": 3045,
+      "que": 3046,
+      "Ġiss": 3047,
+      "urrent": 3048,
+      "Ġprinc": 3049,
+      "Ġtask": 3050,
+      "ops": 3051,
+      "Ġwhereas": 3052,
+      "CH": 3053,
+      "Ġrevealed": 3054,
+      "Ġcannot": 3055,
+      "active": 3056,
+      "enz": 3057,
+      "Ġfore": 3058,
+      "Ġoperator": 3059,
+      "Ġcolum": 3060,
+      "atin": 3061,
+      "Ġoriginal": 3062,
+      "Ġsmaller": 3063,
+      "Ġmaterials": 3064,
+      "hydro": 3065,
+      "Ġcurve": 3066,
+      "Ġselection": 3067,
+      "akes": 3068,
+      "Ġexpos": 3069,
+      "ats": 3070,
+      "ĠÏī": 3071,
+      "Ġpack": 3072,
+      "Ġstability": 3073,
+      "Ġoverall": 3074,
+      "Ġmorph": 3075,
+      "Ġmetric": 3076,
+      "Ġol": 3077,
+      "Ġbar": 3078,
+      "ĠIN": 3079,
+      "IM": 3080,
+      "cy": 3081,
+      "ethyl": 3082,
+      "SP": 3083,
+      "Ġresponses": 3084,
+      "ancy": 3085,
+      "Ġlay": 3086,
+      "specific": 3087,
+      "Ġvs": 3088,
+      "aged": 3089,
+      "Ġsocial": 3090,
+      "Ġcut": 3091,
+      "IP": 3092,
+      "Ġlimited": 3093,
+      "encies": 3094,
+      "Ġprotoc": 3095,
+      "Ġcomposition": 3096,
+      "ĠThey": 3097,
+      "Ġnumbers": 3098,
+      "mbox": 3099,
+      "Ġdecreased": 3100,
+      "vec": 3101,
+      "RO": 3102,
+      "Authors": 3103,
+      "Ġthick": 3104,
+      "Ġcoordin": 3105,
+      "Ġmes": 3106,
+      "Ġaffect": 3107,
+      "Ġclose": 3108,
+      "Ġtransport": 3109,
+      "CA": 3110,
+      "rete": 3111,
+      "come": 3112,
+      "Ġcollected": 3113,
+      "ĠFrom": 3114,
+      "Ġcontains": 3115,
+      "chit": 3116,
+      "ĠDet": 3117,
+      "Ġflux": 3118,
+      "overy": 3119,
+      "eu": 3120,
+      "aff": 3121,
+      "Ġconducted": 3122,
+      "Ġcriter": 3123,
+      "Ġliterature": 3124,
+      "Ġmemory": 3125,
+      "Ġsequences": 3126,
+      "Ġpan": 3127,
+      "plicit": 3128,
+      "Ġtrue": 3129,
+      "Ġmedium": 3130,
+      "Ġdam": 3131,
+      "ire": 3132,
+      "cell": 3133,
+      "Let": 3134,
+      "eful": 3135,
+      "ĠAmeric": 3136,
+      "Ġnodes": 3137,
+      "gether": 3138,
+      "Ġtogether": 3139,
+      "TP": 3140,
+      "Ġrather": 3141,
+      "Ġauthors": 3142,
+      "Ġsch": 3143,
+      "Ġprocessing": 3144,
+      "Ġspectra": 3145,
+      "Ġevaluated": 3146,
+      "alk": 3147,
+      "Ġreduce": 3148,
+      "ĠHigh": 3149,
+      "ĠCons": 3150,
+      "Ġcycle": 3151,
+      "orn": 3152,
+      "iers": 3153,
+      "Ġpropor": 3154,
+      "ories": 3155,
+      "rate": 3156,
+      "Ġhost": 3157,
+      "ooth": 3158,
+      "ynt": 3159,
+      "Ġsources": 3160,
+      "Ġindividuals": 3161,
+      "Ġaccount": 3162,
+      "ĠAlthough": 3163,
+      "Ġcorrec": 3164,
+      "Ġplan": 3165,
+      "entially": 3166,
+      "Ġdistinc": 3167,
+      "Ġsoil": 3168,
+      "Ġsearch": 3169,
+      "Ġmanagement": 3170,
+      "Ġversion": 3171,
+      "âĢĶ": 3172,
+      "Ġfig": 3173,
+      "ĠNote": 3174,
+      "Ġhead": 3175,
+      "ditional": 3176,
+      "Ġbuild": 3177,
+      "ĠGl": 3178,
+      "asis": 3179,
+      "group": 3180,
+      "Ġdisplay": 3181,
+      "ĠUniversity": 3182,
+      "ootnote": 3183,
+      "ameter": 3184,
+      "minist": 3185,
+      "opl": 3186,
+      "ymph": 3187,
+      "Lambda": 3188,
+      "Ġidentify": 3189,
+      "ĠStere": 3190,
+      "ĠïĢ": 3191,
+      "Ġprol": 3192,
+      "ource": 3193,
+      "icial": 3194,
+      "Ġsimulations": 3195,
+      "Ġthresh": 3196,
+      "point": 3197,
+      "earch": 3198,
+      "elling": 3199,
+      "ĠAcc": 3200,
+      "Ġframework": 3201,
+      "Ġstrength": 3202,
+      "ĠAb": 3203,
+      "ticles": 3204,
+      "Ġcos": 3205,
+      "Footnote": 3206,
+      "ru": 3207,
+      "ospital": 3208,
+      "Ġstable": 3209,
+      "Ġmotion": 3210,
+      "Ġtested": 3211,
+      "Ġtests": 3212,
+      "aster": 3213,
+      "ldots": 3214,
+      "CL": 3215,
+      "inite": 3216,
+      "Ġspecial": 3217,
+      "====": 3218,
+      "Ġapproaches": 3219,
+      "ping": 3220,
+      "Ġconsum": 3221,
+      "SD": 3222,
+      "Ġjust": 3223,
+      "kappa": 3224,
+      "Ġthough": 3225,
+      "faces": 3226,
+      "Ġrapid": 3227,
+      "ensive": 3228,
+      "Ġnecessary": 3229,
+      "Ġtub": 3230,
+      "Ġforce": 3231,
+      "Ġblack": 3232,
+      "volution": 3233,
+      "ĠAtom": 3234,
+      "ĠHere": 3235,
+      "itude": 3236,
+      "ensions": 3237,
+      "ffer": 3238,
+      "rich": 3239,
+      "Ġgives": 3240,
+      "Ġshape": 3241,
+      "Ġhard": 3242,
+      "omp": 3243,
+      "Ġrepresentation": 3244,
+      "ling": 3245,
+      "ĠDec": 3246,
+      "Ġnumerical": 3247,
+      "Ġplace": 3248,
+      "Ġleading": 3249,
+      "Ġbenef": 3250,
+      "Ġregular": 3251,
+      "Ġcluster": 3252,
+      "Ġrelatively": 3253,
+      "Ġpercent": 3254,
+      "Ġautom": 3255,
+      "Ġsympt": 3256,
+      "ibri": 3257,
+      "ches": 3258,
+      "henyl": 3259,
+      "car": 3260,
+      "Ġillustr": 3261,
+      "ports": 3262,
+      "emic": 3263,
+      "Ġgive": 3264,
+      "Ġconven": 3265,
+      "lection": 3266,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠ": 3267,
+      "ĠAnd": 3268,
+      "Ġfood": 3269,
+      "mic": 3270,
+      "ographic": 3271,
+      "Ġcheck": 3272,
+      "Ġability": 3273,
+      "iquid": 3274,
+      "Ġsubstr": 3275,
+      "ĠâĪĨ": 3276,
+      "Ġedge": 3277,
+      "ĠPD": 3278,
+      "Ġclassification": 3279,
+      "Ġsurvival": 3280,
+      "ĠCal": 3281,
+      "erate": 3282,
+      "Ġuseful": 3283,
+      "Ġcarried": 3284,
+      "Ġintensity": 3285,
+      "HE": 3286,
+      "ocenter": 3287,
+      "Ġpathway": 3288,
+      "Ġdefinition": 3289,
+      "Ġscheme": 3290,
+      "Ġsubsequ": 3291,
+      "ĠFirst": 3292,
+      "Ġconsequ": 3293,
+      "ĠDiff": 3294,
+      "Ġinhibit": 3295,
+      "Ġamplit": 3296,
+      "aser": 3297,
+      "ĠNetwork": 3298,
+      "normal": 3299,
+      "ĠST": 3300,
+      "Ġsolid": 3301,
+      "perim": 3302,
+      "comes": 3303,
+      "Ġcyt": 3304,
+      "odies": 3305,
+      "IF": 3306,
+      "radi": 3307,
+      "Ġmor": 3308,
+      "Ġcore": 3309,
+      "BS": 3310,
+      "********": 3311,
+      "Ġsoftware": 3312,
+      "ĠGu": 3313,
+      "ired": 3314,
+      "ident": 3315,
+      "Ġdifficult": 3316,
+      "use": 3317,
+      "Ġadded": 3318,
+      "ley": 3319,
+      "Ġcaused": 3320,
+      "gence": 3321,
+      "Ġbase": 3322,
+      "####": 3323,
+      "ogenic": 3324,
+      "from": 3325,
+      "Ġstatus": 3326,
+      "Ġassociation": 3327,
+      "ĠStereocenter": 3328,
+      "Ġgalax": 3329,
+      "NO": 3330,
+      "anguage": 3331,
+      "Ġdimension": 3332,
+      "ogenesis": 3333,
+      "Ġemission": 3334,
+      "Ġdeath": 3335,
+      "ulin": 3336,
+      "Ġagre": 3337,
+      "turb": 3338,
+      "nabl": 3339,
+      "poral": 3340,
+      "Ġpor": 3341,
+      "Ġcombined": 3342,
+      "Ġalgorithms": 3343,
+      "Cs": 3344,
+      "Ġsensitivity": 3345,
+      "Ġallows": 3346,
+      "Ġcapacity": 3347,
+      "version": 3348,
+      "Ġrestric": 3349,
+      "rome": 3350,
+      "Ġexposure": 3351,
+      "hy": 3352,
+      "anning": 3353,
+      "Ġobject": 3354,
+      "Ġcode": 3355,
+      "fl": 3356,
+      "roduction": 3357,
+      "resents": 3358,
+      "rup": 3359,
+      "Ġtext": 3360,
+      "ĠMat": 3361,
+      "Ġleads": 3362,
+      "Ġreson": 3363,
+      "Ġproducts": 3364,
+      "Ġwhole": 3365,
+      "Ġmatter": 3366,
+      "Phi": 3367,
+      "opt": 3368,
+      "encing": 3369,
+      "fficients": 3370,
+      "na": 3371,
+      "pecially": 3372,
+      "Ġhaving": 3373,
+      "ropy": 3374,
+      "Ġuncertain": 3375,
+      "enari": 3376,
+      "rical": 3377,
+      "Ġminim": 3378,
+      "Ġorigin": 3379,
+      "uper": 3380,
+      "ĠNon": 3381,
+      "Ġevaluate": 3382,
+      "Proof": 3383,
+      "cap": 3384,
+      "Ġsignaling": 3385,
+      "Ġpolymer": 3386,
+      "tically": 3387,
+      "itten": 3388,
+      "antit": 3389,
+      "Ġuser": 3390,
+      "level": 3391,
+      "Ġmeasures": 3392,
+      "Ġdynamic": 3393,
+      "Ġmonths": 3394,
+      "oti": 3395,
+      "rand": 3396,
+      "Ġuntil": 3397,
+      "Ġdenote": 3398,
+      "Ġnote": 3399,
+      "Ġmaintain": 3400,
+      "Ġkin": 3401,
+      "scill": 3402,
+      "Ġimaging": 3403,
+      "Ġpain": 3404,
+      "avy": 3405,
+      "Ġmit": 3406,
+      "othe": 3407,
+      "Ġregul": 3408,
+      "known": 3409,
+      "Ġplot": 3410,
+      "nabla": 3411,
+      "Ġfraction": 3412,
+      "wer": 3413,
+      "Ġstrategy": 3414,
+      "Ġgreat": 3415,
+      "Ġdataset": 3416,
+      "Ġunique": 3417,
+      "CM": 3418,
+      "Ġtw": 3419,
+      "han": 3420,
+      "ĠEu": 3421,
+      "andid": 3422,
+      "Ġbackground": 3423,
+      "Ġbroad": 3424,
+      "ilt": 3425,
+      "Ġimproved": 3426,
+      "Ġdiagnosis": 3427,
+      "ious": 3428,
+      "Ġdig": 3429,
+      "rem": 3430,
+      "era": 3431,
+      "Ġexcl": 3432,
+      "Ġmetal": 3433,
+      "Ġsix": 3434,
+      "Ġminimum": 3435,
+      "usions": 3436,
+      "ee": 3437,
+      "Ġcompounds": 3438,
+      "Ġasp": 3439,
+      "Ġeth": 3440,
+      "Ġdetect": 3441,
+      "ference": 3442,
+      "Ġη": 3443,
+      "Ġstatistical": 3444,
+      "atives": 3445,
+      "ris": 3446,
+      "Ġtheorem": 3447,
+      "ĠOF": 3448,
+      "ww": 3449,
+      "arily": 3450,
+      "ception": 3451,
+      "iving": 3452,
+      "Ġtesting": 3453,
+      "Ġdiagnos": 3454,
+      "Ġrepresents": 3455,
+      "Sigma": 3456,
+      "onical": 3457,
+      "Ġequivalent": 3458,
+      "Ġbiom": 3459,
+      "Ġsubst": 3460,
+      "raints": 3461,
+      "ĠRef": 3462,
+      "Ġscore": 3463,
+      "Ġdoc": 3464,
+      "Ġimplies": 3465,
+      "eter": 3466,
+      "Ġsynthesis": 3467,
+      "ilibri": 3468,
+      "attering": 3469,
+      "CS": 3470,
+      "alse": 3471,
+      "Ġneuro": 3472,
+      "Ġalthough": 3473,
+      "irus": 3474,
+      "methyl": 3475,
+      "Ġtranscription": 3476,
+      "ÏĢ": 3477,
+      "ĠMolecular": 3478,
+      "Ġcause": 3479,
+      "mut": 3480,
+      "ĠId": 3481,
+      "λ": 3482,
+      "add": 3483,
+      "Ġplac": 3484,
+      "Ġagg": 3485,
+      "ture": 3486,
+      "Ġlack": 3487,
+      "Ġprediction": 3488,
+      "raw": 3489,
+      "An": 3490,
+      "Ġult": 3491,
+      "ynomial": 3492,
+      "Ġimmune": 3493,
+      "ili": 3494,
+      "Ġprep": 3495,
+      "γ": 3496,
+      "class": 3497,
+      "Ġmach": 3498,
+      "ample": 3499,
+      "Ġresolution": 3500,
+      "Ġcoupling": 3501,
+      "seud": 3502,
+      "Ġindicates": 3503,
+      "Ġgeneration": 3504,
+      "Ġhar": 3505,
+      "Ġfund": 3506,
+      "scale": 3507,
+      "Ġeigen": 3508,
+      "ĠRel": 3509,
+      "abor": 3510,
+      "ĠCH": 3511,
+      "ext": 3512,
+      "amm": 3513,
+      "Ġcorrect": 3514,
+      "Ġscreen": 3515,
+      "Ġstructural": 3516,
+      "ĠpH": 3517,
+      "Ġrelevant": 3518,
+      "Ġangle": 3519,
+      "IG": 3520,
+      "Ġalgebra": 3521,
+      "helial": 3522,
+      "Ġworld": 3523,
+      "Ġcurves": 3524,
+      "ĠIntroduction": 3525,
+      "Ġthird": 3526,
+      "Ġintroduced": 3527,
+      "Big": 3528,
+      "no": 3529,
+      "auss": 3530,
+      "subset": 3531,
+      "Ġtransmission": 3532,
+      "Ġprofile": 3533,
+      "Ġν": 3534,
+      "Ġespecially": 3535,
+      "Ġattrib": 3536,
+      "uction": 3537,
+      "Ġcoefficients": 3538,
+      "Ġremains": 3539,
+      "Ġneigh": 3540,
+      "osen": 3541,
+      "Ġreli": 3542,
+      "Ġhighest": 3543,
+      "Ġuniform": 3544,
+      "Ġfar": 3545,
+      "chitect": 3546,
+      "||": 3547,
+      "Ġappropri": 3548,
+      "plex": 3549,
+      "ĠMass": 3550,
+      "ogeneous": 3551,
+      "ales": 3552,
+      "Ġrefer": 3553,
+      "Ġneeded": 3554,
+      "Ġdifferential": 3555,
+      "ceed": 3556,
+      "$$": 3557,
+      "ynamic": 3558,
+      "Ġsex": 3559,
+      "Ġspectral": 3560,
+      "char": 3561,
+      "PE": 3562,
+      "TS": 3563,
+      "Ġapproximately": 3564,
+      "value": 3565,
+      "Ġhalf": 3566,
+      "ending": 3567,
+      "Ġgradi": 3568,
+      "Ġcoefficient": 3569,
+      "ĠPhys": 3570,
+      "Ġconcer": 3571,
+      "Ġlabel": 3572,
+      "iral": 3573,
+      "Ġcharge": 3574,
+      "Ġoxygen": 3575,
+      "Ġdevi": 3576,
+      "Ġinternal": 3577,
+      "Ġexpans": 3578,
+      "load": 3579,
+      "ĠSm": 3580,
+      "rang": 3581,
+      "Con": 3582,
+      "ĠNa": 3583,
+      "Ġke": 3584,
+      "Ġdiab": 3585,
+      "ached": 3586,
+      "Ġlocation": 3587,
+      "Ġvolt": 3588,
+      "ĠDisc": 3589,
+      "---": 3590,
+      "ocytes": 3591,
+      "oretical": 3592,
+      "Ġgain": 3593,
+      "Ġmedi": 3594,
+      "ympt": 3595,
+      "oted": 3596,
+      "ĠVal": 3597,
+      "Ġcommunity": 3598,
+      "plementary": 3599,
+      "Ġtree": 3600,
+      "ĠTwo": 3601,
+      "Ġwhose": 3602,
+      "Ġdone": 3603,
+      "amine": 3604,
+      "Ġbiological": 3605,
+      "inks": 3606,
+      "Ġalmost": 3607,
+      "Ġslight": 3608,
+      "Ġrepro": 3609,
+      "ģĦ": 3610,
+      "Ġtherap": 3611,
+      "ocation": 3612,
+      "Ġgly": 3613,
+      "ĠEqu": 3614,
+      "Ġcolor": 3615,
+      "Ġnam": 3616,
+      "section": 3617,
+      "ĠEm": 3618,
+      "ready": 3619,
+      "Hz": 3620,
+      "PD": 3621,
+      "function": 3622,
+      "change": 3623,
+      "Ġprincip": 3624,
+      "Ġbecome": 3625,
+      "ĠâĢĺ": 3626,
+      "Ġcour": 3627,
+      "Ġlocated": 3628,
+      "Ġrang": 3629,
+      "inity": 3630,
+      "Ġinterval": 3631,
+      "gin": 3632,
+      "Ġinvestigate": 3633,
+      "free": 3634,
+      "Ġvitro": 3635,
+      "Ġsubset": 3636,
+      "Ġmov": 3637,
+      "Ġprove": 3638,
+      "Ġliver": 3639,
+      "ategor": 3640,
+      "etes": 3641,
+      "Ġlymph": 3642,
+      "dom": 3643,
+      "ĠElect": 3644,
+      "Ġserum": 3645,
+      "Ġscenari": 3646,
+      "ends": 3647,
+      "ĠFinally": 3648,
+      "Ġfilter": 3649,
+      "IL": 3650,
+      "Ġabund": 3651,
+      "mentation": 3652,
+      "imals": 3653,
+      "num": 3654,
+      "enced": 3655,
+      "Ġproperty": 3656,
+      "matrix": 3657,
+      "ĠCompar": 3658,
+      "Ġland": 3659,
+      "ĠChar": 3660,
+      "ressive": 3661,
+      "ulus": 3662,
+      "Ġbone": 3663,
+      "Ex": 3664,
+      "Ġradiation": 3665,
+      "Ġsuggested": 3666,
+      "ĠComput": 3667,
+      "Ġthreshold": 3668,
+      "ĠAD": 3669,
+      "Ġhor": 3670,
+      "Ġinduc": 3671,
+      "Ġapproximation": 3672,
+      "Ġadminist": 3673,
+      "Ġord": 3674,
+      "Ġlung": 3675,
+      "Ġreceived": 3676,
+      "Ġnorm": 3677,
+      "Ġestimates": 3678,
+      "Ġlaw": 3679,
+      "Ġoutcomes": 3680,
+      "ĠPr": 3681,
+      "Ġdepth": 3682,
+      "Ġelse": 3683,
+      "Ġcontribution": 3684,
+      "hetic": 3685,
+      "Ġconserv": 3686,
+      "Ġupon": 3687,
+      "Ġdeep": 3688,
+      "MD": 3689,
+      "Ġmel": 3690,
+      "Ġfilm": 3691,
+      "ilibrium": 3692,
+      "Ġoscill": 3693,
+      "olved": 3694,
+      "Ġbreast": 3695,
+      "CP": 3696,
+      "ĠDist": 3697,
+      "rices": 3698,
+      "inated": 3699,
+      "Ġoptimization": 3700,
+      "Ġpredicted": 3701,
+      "sf": 3702,
+      "dim": 3703,
+      "ĠSN": 3704,
+      "Ġavoid": 3705,
+      "Ġneural": 3706,
+      "Ġwa": 3707,
+      "rope": 3708,
+      "Ġdistributions": 3709,
+      "oxid": 3710,
+      "Ġsmooth": 3711,
+      "path": 3712,
+      "Ġfluid": 3713,
+      "Ġsaf": 3714,
+      "Ġchoice": 3715,
+      "AA": 3716,
+      "Ġmolecules": 3717,
+      "US": 3718,
+      "Ġalways": 3719,
+      "ivo": 3720,
+      "Ġregression": 3721,
+      "Ġsuccessful": 3722,
+      "Ġwall": 3723,
+      "oung": 3724,
+      "Ġactivities": 3725,
+      "Ġdependence": 3726,
+      "Ġrequires": 3727,
+      "Ġplane": 3728,
+      "Ġdesigned": 3729,
+      "PI": 3730,
+      "down": 3731,
+      "Ġpopulations": 3732,
+      "cor": 3733,
+      "mediate": 3734,
+      "Ġdose": 3735,
+      "Ġbond": 3736,
+      "Co": 3737,
+      "ĠMan": 3738,
+      "Ġdiagram": 3739,
+      "gs": 3740,
+      "Ġtool": 3741,
+      "Ġisolated": 3742,
+      "Ġversus": 3743,
+      "ney": 3744,
+      "Ġemerg": 3745,
+      "ĠAut": 3746,
+      "aim": 3747,
+      "field": 3748,
+      "Ġexamined": 3749,
+      "Ġsat": 3750,
+      "SM": 3751,
+      "ĠSpec": 3752,
+      "Ġparallel": 3753,
+      "isation": 3754,
+      "Ġdistinct": 3755,
+      "Ġpredict": 3756,
+      "Ġfer": 3757,
+      "Ġunderstanding": 3758,
+      "ĠSimilar": 3759,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 3760,
+      "udes": 3761,
+      "Ġorient": 3762,
+      "hic": 3763,
+      "uz": 3764,
+      "Ġmodified": 3765,
+      "ĠâĪ¼": 3766,
+      "FF": 3767,
+      "There": 3768,
+      "Ġtrial": 3769,
+      "xy": 3770,
+      "gery": 3771,
+      "Ġalready": 3772,
+      "define": 3773,
+      "ming": 3774,
+      "ĠSD": 3775,
+      "Ġmonitor": 3776,
+      "Ġpsy": 3777,
+      "Ġbecomes": 3778,
+      "istry": 3779,
+      "ĠÎĵ": 3780,
+      "Ġhum": 3781,
+      "rier": 3782,
+      "ession": 3783,
+      "Ġhistory": 3784,
+      "ö": 3785,
+      "Ġξ": 3786,
+      "Ġestablished": 3787,
+      "Ġachieved": 3788,
+      "estern": 3789,
+      "ÏĨ": 3790,
+      "ĠHence": 3791,
+      "Ġassessment": 3792,
+      "otor": 3793,
+      "Ġdescribe": 3794,
+      "ochond": 3795,
+      "ylation": 3796,
+      "sts": 3797,
+      "space": 3798,
+      "Ġdiseases": 3799,
+      "jection": 3800,
+      "Ġslow": 3801,
+      "Ġnonlinear": 3802,
+      "ply": 3803,
+      "ml": 3804,
+      "Ġembed": 3805,
+      "comp": 3806,
+      "Ġefficient": 3807,
+      "Ġoperation": 3808,
+      "Ġcontact": 3809,
+      "oz": 3810,
+      "Ġinvari": 3811,
+      "Ġcenter": 3812,
+      "Ġconc": 3813,
+      "widetilde": 3814,
+      "Ġbeam": 3815,
+      "Ġclosed": 3816,
+      "ĠMethods": 3817,
+      "Ġchronic": 3818,
+      "aling": 3819,
+      "Ġsevere": 3820,
+      "Ġforms": 3821,
+      "ilit": 3822,
+      "side": 3823,
+      "pen": 3824,
+      "Ġbran": 3825,
+      "oud": 3826,
+      "tality": 3827,
+      "Ġmaps": 3828,
+      "acts": 3829,
+      "OL": 3830,
+      "PR": 3831,
+      "ĠÍ": 3832,
+      "sl": 3833,
+      "Ġinstance": 3834,
+      "ully": 3835,
+      "Ġestimation": 3836,
+      "Ġplate": 3837,
+      "Ġdevice": 3838,
+      "ĠIII": 3839,
+      "sin": 3840,
+      "Ġplants": 3841,
+      "ittle": 3842,
+      "Ġproduce": 3843,
+      "Ġhence": 3844,
+      "Ġnature": 3845,
+      "Ġrelease": 3846,
+      "ĠMin": 3847,
+      "rict": 3848,
+      "Ġconnected": 3849,
+      "ottom": 3850,
+      "ellar": 3851,
+      "Ġformed": 3852,
+      "Ġmob": 3853,
+      "Ġcomputed": 3854,
+      "ĠRE": 3855,
+      "Ġpolynomial": 3856,
+      "Ġliquid": 3857,
+      "gn": 3858,
+      "Ġassay": 3859,
+      "Ġmanif": 3860,
+      "ĠSi": 3861,
+      "rence": 3862,
+      "Ġaxis": 3863,
+      "VID": 3864,
+      "Ġsignals": 3865,
+      "θ": 3866,
+      "tok": 3867,
+      "ds": 3868,
+      "Ġrats": 3869,
+      "Ġtor": 3870,
+      "olecular": 3871,
+      "ched": 3872,
+      "Ġdescri": 3873,
+      "Ġexpon": 3874,
+      "Ġperturb": 3875,
+      "Ġgluc": 3876,
+      "Ġcolumn": 3877,
+      "UL": 3878,
+      "Ġmainly": 3879,
+      "Ġmul": 3880,
+      "ider": 3881,
+      "ĠCR": 3882,
+      "Ġcataly": 3883,
+      "Ġlaser": 3884,
+      "tioned": 3885,
+      "den": 3886,
+      "Ġsuggests": 3887,
+      "fig": 3888,
+      "Ġpropag": 3889,
+      "org": 3890,
+      "rep": 3891,
+      "Ġcharacterized": 3892,
+      "ologies": 3893,
+      "Ġaccum": 3894,
+      "Ġvary": 3895,
+      "Ġcontrolled": 3896,
+      "Ġupd": 3897,
+      "ĠBr": 3898,
+      "Ġentire": 3899,
+      "Ġ@": 3900,
+      "âģĦ": 3901,
+      "ĠÌ": 3902,
+      "Ġdatab": 3903,
+      "ano": 3904,
+      "amil": 3905,
+      "Ġadjust": 3906,
+      "ye": 3907,
+      "pression": 3908,
+      "erences": 3909,
+      "Ġessential": 3910,
+      "ĠHydro": 3911,
+      "ĠTr": 3912,
+      "Ġappropriate": 3913,
+      "Ġformula": 3914,
+      "Ġlattice": 3915,
+      "Ġacute": 3916,
+      "Ġusually": 3917,
+      "itable": 3918,
+      "Ġmar": 3919,
+      "Ġμm": 3920,
+      "ĠUSA": 3921,
+      "Ġincub": 3922,
+      "ocks": 3923,
+      "Ġpepti": 3924,
+      "iddle": 3925,
+      "Ġdecom": 3926,
+      "Ġdamage": 3927,
+      "Ġgenome": 3928,
+      "Ġmouse": 3929,
+      "circ": 3930,
+      "Ġlayers": 3931,
+      "Ġtrack": 3932,
+      "Ġtox": 3933,
+      "Ġreplac": 3934,
+      "Ġadvant": 3935,
+      "izon": 3936,
+      "Ġrecorded": 3937,
+      "Ġstart": 3938,
+      "Ġrank": 3939,
+      "ser": 3940,
+      "ĠGene": 3941,
+      "aussian": 3942,
+      "ingu": 3943,
+      "Ġconstraints": 3944,
+      "flow": 3945,
+      "Ġmig": 3946,
+      "PL": 3947,
+      "Ġincor": 3948,
+      "appro": 3949,
+      "Ġfast": 3950,
+      "Ġmuscle": 3951,
+      "Ġhome": 3952,
+      "eq": 3953,
+      "ĠÏĪ": 3954,
+      "Ġstrongly": 3955,
+      "ĠEurope": 3956,
+      "Ġsubjects": 3957,
+      "Ġobjects": 3958,
+      "test": 3959,
+      "tered": 3960,
+      "ĠWhile": 3961,
+      "Ġsymmetry": 3962,
+      "Ġquantif": 3963,
+      "``": 3964,
+      "Ġbreak": 3965,
+      "ĠExperim": 3966,
+      "Ġmixt": 3967,
+      "<<": 3968,
+      "ĠChina": 3969,
+      "ĠIdentif": 3970,
+      "Ġaffected": 3971,
+      "Ġsecondary": 3972,
+      "Ġinequ": 3973,
+      "incl": 3974,
+      "EG": 3975,
+      "FT": 3976,
+      "Ġfailure": 3977,
+      "ectiv": 3978,
+      "Ġkm": 3979,
+      "Ġsampling": 3980,
+      "Ġexpansion": 3981,
+      "Ġpractice": 3982,
+      "uations": 3983,
+      "ognitive": 3984,
+      "Ġdiet": 3985,
+      "Ġtemperatures": 3986,
+      "Ġcontrols": 3987,
+      "Ġchosen": 3988,
+      "Ġgenerally": 3989,
+      "ancer": 3990,
+      "Ġdegrad": 3991,
+      "uli": 3992,
+      "sm": 3993,
+      "otherapy": 3994,
+      "Ġtowards": 3995,
+      "ĠProperties": 3996,
+      "Ġclusters": 3997,
+      "Ġdelay": 3998,
+      "Ġhep": 3999,
+      "PA": 4000,
+      "ĠStudy": 4001,
+      "antitative": 4002,
+      "Ġclassical": 4003,
+      "ĠZh": 4004,
+      "ĠΩ": 4005,
+      "ĠBo": 4006,
+      "Ġseed": 4007,
+      "ĠStruct": 4008,
+      "Ġtrend": 4009,
+      "iological": 4010,
+      "Ġconfirmed": 4011,
+      "Ġdistributed": 4012,
+      "bial": 4013,
+      "Ġname": 4014,
+      "CN": 4015,
+      "valence": 4016,
+      "erior": 4017,
+      "iven": 4018,
+      "ned": 4019,
+      "Ġbehaviour": 4020,
+      "asks": 4021,
+      "gra": 4022,
+      "mark": 4023,
+      "Ġerrors": 4024,
+      "ĠRep": 4025,
+      "light": 4026,
+      "cript": 4027,
+      "If": 4028,
+      "Ġcandid": 4029,
+      "Ġdepends": 4030,
+      "ĠNational": 4031,
+      "Ġholds": 4032,
+      "Ġprotocol": 4033,
+      "ĠUnited": 4034,
+      "Ġinterface": 4035,
+      "Ġexpect": 4036,
+      "Ġïģ": 4037,
+      "ĠHIV": 4038,
+      "Ġroot": 4039,
+      "Ġscattering": 4040,
+      "words": 4041,
+      "Ġobservation": 4042,
+      "otop": 4043,
+      "Ġoccurs": 4044,
+      "ources": 4045,
+      "pite": 4046,
+      "ĠSte": 4047,
+      "Ġorth": 4048,
+      "Ġstain": 4049,
+      "Ġsteps": 4050,
+      "Ġcompare": 4051,
+      "Ġbasic": 4052,
+      "Ġinhibition": 4053,
+      "Ġsymptoms": 4054,
+      "ĠHealth": 4055,
+      "Ġpublished": 4056,
+      "fold": 4057,
+      "Ġtun": 4058,
+      "Ġvivo": 4059,
+      "Ġreconstr": 4060,
+      "ĠmRNA": 4061,
+      "icy": 4062,
+      "Ġhybrid": 4063,
+      "yr": 4064,
+      "Ġmixed": 4065,
+      "vis": 4066,
+      "ChI": 4067,
+      "Ġmedical": 4068,
+      "Ġfrag": 4069,
+      "Ġanimals": 4070,
+      "Ġimportance": 4071,
+      "Ġengine": 4072,
+      "ĠCT": 4073,
+      "Ġpairs": 4074,
+      "Ġbal": 4075,
+      "ĠEar": 4076,
+      "hers": 4077,
+      "Ġsynd": 4078,
+      "Ġarchitect": 4079,
+      "Ġidentification": 4080,
+      "Ġstrategies": 4081,
+      "Ġregulation": 4082,
+      "ĠLa": 4083,
+      "ror": 4084,
+      "Ġfluores": 4085,
+      "urity": 4086,
+      "Ġconcept": 4087,
+      "Ġattention": 4088,
+      "Ġtransformation": 4089,
+      "ucle": 4090,
+      "ĠResearch": 4091,
+      "Ġsimpl": 4092,
+      "Ġculture": 4093,
+      "aring": 4094,
+      "ifically": 4095,
+      "pir": 4096,
+      "ze": 4097,
+      "PT": 4098,
+      "mosp": 4099,
+      "Ġswit": 4100,
+      "Ġnor": 4101,
+      "Ġenhance": 4102,
+      "Ġenvironmental": 4103,
+      "rary": 4104,
+      "ĠMicro": 4105,
+      "Ġwide": 4106,
+      "opath": 4107,
+      "auge": 4108,
+      "zeta": 4109,
+      "Ġste": 4110,
+      "ĠEl": 4111,
+      "Ġwords": 4112,
+      "Ġnuclear": 4113,
+      "Ġlanguage": 4114,
+      "Ġdetails": 4115,
+      "opar": 4116,
+      "ĠRed": 4117,
+      "water": 4118,
+      "Ġcategor": 4119,
+      "Ġfile": 4120,
+      "Ġcover": 4121,
+      "Ġachieve": 4122,
+      "á": 4123,
+      "umm": 4124,
+      "Ġlig": 4125,
+      "Ġsurvey": 4126,
+      "Ġextended": 4127,
+      "lab": 4128,
+      "ĠInc": 4129,
+      "Ġdispers": 4130,
+      "Ġrecomm": 4131,
+      "ĠBased": 4132,
+      "Ġabsence": 4133,
+      "Ġconstruction": 4134,
+      "Ġpoor": 4135,
+      "Ġvoltage": 4136,
+      "Ġcellular": 4137,
+      "Ġmortality": 4138,
+      "Ġshowing": 4139,
+      "Ġprolif": 4140,
+      "mp": 4141,
+      "Ġneurons": 4142,
+      "Ġsupported": 4143,
+      "Ġprevent": 4144,
+      "eli": 4145,
+      "oxy": 4146,
+      "ica": 4147,
+      "Ġfully": 4148,
+      "Ġenough": 4149,
+      "otimes": 4150,
+      "ĠMR": 4151,
+      "Ġbul": 4152,
+      "Ġphenomen": 4153,
+      "FA": 4154,
+      "Ġdecision": 4155,
+      "Ġdual": 4156,
+      "Ġdecay": 4157,
+      "Ġown": 4158,
+      "Ġuses": 4159,
+      "Ġchalleng": 4160,
+      "Ġaddress": 4161,
+      "OC": 4162,
+      "tivation": 4163,
+      "Ġmill": 4164,
+      "Ġmodes": 4165,
+      "atus": 4166,
+      "iction": 4167,
+      "Ġabsorption": 4168,
+      "Ġepit": 4169,
+      "Ġconstra": 4170,
+      "Ġagreement": 4171,
+      "ĠAf": 4172,
+      "Ġbias": 4173,
+      "uded": 4174,
+      "Ġparts": 4175,
+      "Ġvan": 4176,
+      "Ġcolon": 4177,
+      "Ġexternal": 4178,
+      "Ġtheoretical": 4179,
+      "asi": 4180,
+      "Ġles": 4181,
+      "abilities": 4182,
+      "LA": 4183,
+      "ttps": 4184,
+      "Ġinstead": 4185,
+      "Ġmembers": 4186,
+      "++": 4187,
+      "Ġrecently": 4188,
+      "Ġprepared": 4189,
+      "Ġarticle": 4190,
+      "day": 4191,
+      "Ġextract": 4192,
+      "Ġâİ": 4193,
+      "Ġpathways": 4194,
+      "ÏĦ": 4195,
+      "mid": 4196,
+      "orage": 4197,
+      "Ġcommunication": 4198,
+      "Ġaccel": 4199,
+      "Ġunits": 4200,
+      "itis": 4201,
+      "ynthesis": 4202,
+      "Ġamplitude": 4203,
+      "rie": 4204,
+      "ultaneous": 4205,
+      "ĠLear": 4206,
+      "ecause": 4207,
+      "do": 4208,
+      "eff": 4209,
+      "Ġexplicit": 4210,
+      "Ġcriteria": 4211,
+      "bre": 4212,
+      "Ġexec": 4213,
+      "Ġmechanical": 4214,
+      "eros": 4215,
+      "ĠConcl": 4216,
+      "ĠExt": 4217,
+      "Ġclasses": 4218,
+      "Ġlonger": 4219,
+      "Ġcalculations": 4220,
+      "eutic": 4221,
+      "ociated": 4222,
+      "ardi": 4223,
+      "Ġcourse": 4224,
+      "Ġpartial": 4225,
+      "Ġsensor": 4226,
+      "Ïĥ": 4227,
+      "Ġoperators": 4228,
+      "ĠAmerican": 4229,
+      "ĠmM": 4230,
+      "Ġvacc": 4231,
+      "occ": 4232,
+      "icon": 4233,
+      "Ġoutcome": 4234,
+      "Ġanalog": 4235,
+      "Ġthickness": 4236,
+      "Ġreach": 4237,
+      "Ġassumed": 4238,
+      "ender": 4239,
+      "Ġmale": 4240,
+      "SE": 4241,
+      "Ġintra": 4242,
+      "Ġimplementation": 4243,
+      "emia": 4244,
+      "Ġenhanced": 4245,
+      "bility": 4246,
+      "Ġeasily": 4247,
+      "ump": 4248,
+      "Ġcarcin": 4249,
+      "osa": 4250,
+      "Ġcorresponds": 4251,
+      "neg": 4252,
+      "Ġmagnitude": 4253,
+      "const": 4254,
+      "Ġlatter": 4255,
+      "Ġrepresented": 4256,
+      "Ġsed": 4257,
+      "Ġparticularly": 4258,
+      "Ġwritten": 4259,
+      "part": 4260,
+      "Ġoil": 4261,
+      "berg": 4262,
+      "ĠBar": 4263,
+      "Ġdys": 4264,
+      "ĠSome": 4265,
+      "ĠMar": 4266,
+      "Ġalternative": 4267,
+      "ĠGerm": 4268,
+      "Ġgenerate": 4269,
+      "Ġconstruct": 4270,
+      "ians": 4271,
+      "stream": 4272,
+      "Ġec": 4273,
+      "ochemical": 4274,
+      "ibration": 4275,
+      "operative": 4276,
+      "ister": 4277,
+      "Ġrobust": 4278,
+      "tre": 4279,
+      "Ġmodeling": 4280,
+      "oring": 4281,
+      "ese": 4282,
+      "ded": 4283,
+      "ideo": 4284,
+      "Ġhydrogen": 4285,
+      "uments": 4286,
+      "Ġdemonstrate": 4287,
+      "Ġcorrelated": 4288,
+      "Ġsystematic": 4289,
+      "Ġsurgery": 4290,
+      "Ġindicating": 4291,
+      "Ġhypothesis": 4292,
+      "year": 4293,
+      "mitted": 4294,
+      "Ġstars": 4295,
+      "Ġprofiles": 4296,
+      "Ġconsists": 4297,
+      "tri": 4298,
+      "Ġdependent": 4299,
+      "ishing": 4300,
+      "top": 4301,
+      "Ġheart": 4302,
+      "atically": 4303,
+      "Ġinjury": 4304,
+      "Ġquad": 4305,
+      "Ġweeks": 4306,
+      "uting": 4307,
+      "ĠTe": 4308,
+      "Ġidenti": 4309,
+      "Ġgradient": 4310,
+      "Ġcalculation": 4311,
+      "Ġur": 4312,
+      "RT": 4313,
+      "zation": 4314,
+      "Ġeduc": 4315,
+      "ening": 4316,
+      "PP": 4317,
+      "zed": 4318,
+      "ush": 4319,
+      "Ġcharacteristic": 4320,
+      "Ġstrains": 4321,
+      "eth": 4322,
+      "Ġdivers": 4323,
+      "âĪĪ": 4324,
+      "oids": 4325,
+      "olic": 4326,
+      "Ġinterpret": 4327,
+      "Key": 4328,
+      "Ġattack": 4329,
+      "pective": 4330,
+      "Ġlabor": 4331,
+      "Ġmetast": 4332,
+      "NF": 4333,
+      "Ġtissues": 4334,
+      "Ġradius": 4335,
+      "ĠEach": 4336,
+      "Ġcat": 4337,
+      "Ġdon": 4338,
+      "Ġelev": 4339,
+      "Ġassemb": 4340,
+      "rons": 4341,
+      "Ġarbit": 4342,
+      "Ġpanel": 4343,
+      "Ġgrid": 4344,
+      "Ġtable": 4345,
+      "roscopic": 4346,
+      "Ġcle": 4347,
+      "ĠIntern": 4348,
+      "obacter": 4349,
+      "Ġassumption": 4350,
+      "ĠCOVID": 4351,
+      "Ġbounded": 4352,
+      "Ġothers": 4353,
+      "Ġschool": 4354,
+      "Ġhospital": 4355,
+      "lected": 4356,
+      "ĠCu": 4357,
+      "ÃĹ": 4358,
+      "Ġcomplet": 4359,
+      "Ġwidth": 4360,
+      "Ġlinks": 4361,
+      "po": 4362,
+      "ollow": 4363,
+      "Ġnut": 4364,
+      "Ġappears": 4365,
+      "rown": 4366,
+      "aro": 4367,
+      "Ġusers": 4368,
+      "Ġclim": 4369,
+      "Ġslightly": 4370,
+      "Ġblue": 4371,
+      "rab": 4372,
+      "ĠSer": 4373,
+      "Ġfigure": 4374,
+      "ĠRad": 4375,
+      "Ġelectric": 4376,
+      "mm": 4377,
+      "ochastic": 4378,
+      "rief": 4379,
+      "Ġcollection": 4380,
+      "Ġstem": 4381,
+      "Ġgover": 4382,
+      "Ġbur": 4383,
+      "Ġtypical": 4384,
+      "sup": 4385,
+      "Ġaggreg": 4386,
+      "raz": 4387,
+      "ĉĉĉ": 4388,
+      "Ġstation": 4389,
+      "Ġarter": 4390,
+      "ively": 4391,
+      "itrogen": 4392,
+      "Ġconstit": 4393,
+      "empt": 4394,
+      "ĠEffect": 4395,
+      "Ġdescription": 4396,
+      "Ġscores": 4397,
+      "Ġmethyl": 4398,
+      "ĠOb": 4399,
+      "ĠStates": 4400,
+      "Ġsplit": 4401,
+      "ĠVari": 4402,
+      "ĠWang": 4403,
+      "Ġcere": 4404,
+      "ĠFran": 4405,
+      "Ġneeds": 4406,
+      "ĠFour": 4407,
+      "Ġproject": 4408,
+      "Ġdevices": 4409,
+      "Ġintegral": 4410,
+      "ĠEs": 4411,
+      "ymmetric": 4412,
+      "Ġmess": 4413,
+      "Ġplays": 4414,
+      "ĠLearning": 4415,
+      "Ġoverl": 4416,
+      "Here": 4417,
+      "ignment": 4418,
+      "Ġdeliver": 4419,
+      "apan": 4420,
+      "CE": 4421,
+      "Ġgauge": 4422,
+      "ĠJoh": 4423,
+      "----------------": 4424,
+      "Ġunderlying": 4425,
+      "Ġthin": 4426,
+      "Ġassessed": 4427,
+      "Ġdiffusion": 4428,
+      "Ġheight": 4429,
+      "ĠSw": 4430,
+      "Ġdark": 4431,
+      "print": 4432,
+      "range": 4433,
+      "ĠCI": 4434,
+      "ises": 4435,
+      "lier": 4436,
+      "rant": 4437,
+      "omorphism": 4438,
+      "Ġcompact": 4439,
+      "ips": 4440,
+      "ĠName": 4441,
+      "Ġtechnology": 4442,
+      "agen": 4443,
+      "Ġconfiguration": 4444,
+      "Ġduration": 4445,
+      "ĠClass": 4446,
+      "Ġput": 4447,
+      "Ġmaking": 4448,
+      "Ġasympt": 4449,
+      "aid": 4450,
+      "Ġcoh": 4451,
+      "Ġcomplexity": 4452,
+      "Ġsections": 4453,
+      "ĠMD": 4454,
+      "ĠĠĠĠĠĠĠĠĠ": 4455,
+      "Ġrad": 4456,
+      "Ġsubstrate": 4457,
+      "dd": 4458,
+      "Ġann": 4459,
+      "Ġorganic": 4460,
+      "Ġtaking": 4461,
+      "Ġincludes": 4462,
+      "Ġkine": 4463,
+      "ares": 4464,
+      "Ġrow": 4465,
+      "ategory": 4466,
+      "Ġmitochond": 4467,
+      "UT": 4468,
+      "Ġsyndrome": 4469,
+      "ĠProb": 4470,
+      "retion": 4471,
+      "Ġfluct": 4472,
+      "ĠDis": 4473,
+      "Ġtransl": 4474,
+      "plas": 4475,
+      "Ġpsych": 4476,
+      "Ġsurfaces": 4477,
+      "Ġdetailed": 4478,
+      "amilton": 4479,
+      "Ġhold": 4480,
+      "ĠâĬĹ": 4481,
+      "ĠCN": 4482,
+      "Ġdil": 4483,
+      "ĠOver": 4484,
+      "atform": 4485,
+      "Ġvertical": 4486,
+      "Ġcomputation": 4487,
+      "Ġpure": 4488,
+      "Ġmakes": 4489,
+      "Ġexisting": 4490,
+      "Ġexamples": 4491,
+      "SO": 4492,
+      "orders": 4493,
+      "Ġmix": 4494,
+      "Ġincorpor": 4495,
+      "Ġrequ": 4496,
+      "antic": 4497,
+      "DNA": 4498,
+      "δ": 4499,
+      "Ġcloud": 4500,
+      "ĠTechn": 4501,
+      "Ġïĥ": 4502,
+      "ements": 4503,
+      "Ġbaseline": 4504,
+      "stein": 4505,
+      "Ġbelong": 4506,
+      "Ġtrials": 4507,
+      "Ġhorizon": 4508,
+      "Ġphosphor": 4509,
+      "Ġans": 4510,
+      "dix": 4511,
+      "roid": 4512,
+      "Ġapply": 4513,
+      "ued": 4514,
+      "ernel": 4515,
+      "Ġfemale": 4516,
+      "icacy": 4517,
+      "Ġvectors": 4518,
+      "Ġmatrices": 4519,
+      "atric": 4520,
+      "ĠMc": 4521,
+      "Ġpy": 4522,
+      "Ġchlor": 4523,
+      "len": 4524,
+      "Ġclearly": 4525,
+      "static": 4526,
+      "ref": 4527,
+      "ĠSouth": 4528,
+      "Ġmedia": 4529,
+      "ĠShe": 4530,
+      "ĠBay": 4531,
+      "Ġagents": 4532,
+      "By": 4533,
+      "Ġdifferentiation": 4534,
+      "istant": 4535,
+      "orphic": 4536,
+      "Ġvariety": 4537,
+      "Ġservice": 4538,
+      "Ġmapping": 4539,
+      "velength": 4540,
+      "Ġchannels": 4541,
+      "Ġcompute": 4542,
+      "Ġstream": 4543,
+      "uls": 4544,
+      "amide": 4545,
+      "oking": 4546,
+      "vit": 4547,
+      "Ġyields": 4548,
+      "omb": 4549,
+      "ĠGaussian": 4550,
+      "Ġpen": 4551,
+      "une": 4552,
+      "Ġexperience": 4553,
+      "band": 4554,
+      "ĠDo": 4555,
+      "mathsf": 4556,
+      "Ġallowed": 4557,
+      "Ar": 4558,
+      "RA": 4559,
+      "Ġbacterial": 4560,
+      "Ġmiss": 4561,
+      "Ġbacteria": 4562,
+      "Ġmomentum": 4563,
+      "Ġhours": 4564,
+      "uck": 4565,
+      "ĠProposition": 4566,
+      "bert": 4567,
+      "otrop": 4568,
+      "Ġvariance": 4569,
+      "Ġtrig": 4570,
+      "Ġshift": 4571,
+      "Ġequilibrium": 4572,
+      "bu": 4573,
+      "ING": 4574,
+      "Ġwhite": 4575,
+      "Ġkind": 4576,
+      "Ġjoint": 4577,
+      "Ġtemporal": 4578,
+      "ĠIV": 4579,
+      "ĠAfric": 4580,
+      "Ġsubject": 4581,
+      "ĠPo": 4582,
+      "head": 4583,
+      "idel": 4584,
+      "Ġantibody": 4585,
+      "ĠEffects": 4586,
+      "Ġspe": 4587,
+      "Ġsufficient": 4588,
+      "jected": 4589,
+      "rees": 4590,
+      "ĠTop": 4591,
+      "Ġmutations": 4592,
+      "isions": 4593,
+      "BC": 4594,
+      "Ġinduction": 4595,
+      "Ġinteresting": 4596,
+      "ella": 4597,
+      "can": 4598,
+      "Ġsusp": 4599,
+      "ĠGroup": 4600,
+      "Ġextracted": 4601,
+      "istically": 4602,
+      "coh": 4603,
+      "map": 4604,
+      "Ġaccurate": 4605,
+      "Ġtoo": 4606,
+      "Ġdimensions": 4607,
+      "tegr": 4608,
+      "Ġgreen": 4609,
+      "ĠRo": 4610,
+      "Ġwild": 4611,
+      "Ġloop": 4612,
+      "Ġmeta": 4613,
+      "Ġsubstit": 4614,
+      "osome": 4615,
+      "Ġsuggesting": 4616,
+      "Ġspecim": 4617,
+      "amental": 4618,
+      "iment": 4619,
+      "Ġij": 4620,
+      "Ġclaim": 4621,
+      "Ġauthor": 4622,
+      "Ġfilms": 4623,
+      "Ġcounter": 4624,
+      "Ġconventional": 4625,
+      "rin": 4626,
+      "otypes": 4627,
+      "Ġpast": 4628,
+      "Since": 4629,
+      "mediated": 4630,
+      "reatment": 4631,
+      "Ġextension": 4632,
+      "Ġbio": 4633,
+      "Ġsent": 4634,
+      "hal": 4635,
+      "Ġobjective": 4636,
+      "Ġarray": 4637,
+      "Ġsuitable": 4638,
+      "ĠBut": 4639,
+      "ĠHuman": 4640,
+      "organ": 4641,
+      "but": 4642,
+      "model": 4643,
+      "SI": 4644,
+      "Ġhealthy": 4645,
+      "Ġvac": 4646,
+      "Ġlate": 4647,
+      "Ġring": 4648,
+      "Ġlittle": 4649,
+      "MT": 4650,
+      "Ġsquare": 4651,
+      "Ġgeometry": 4652,
+      "ĠTHE": 4653,
+      "ĠSing": 4654,
+      "jug": 4655,
+      "Ġstudents": 4656,
+      ",,": 4657,
+      "Ġadult": 4658,
+      "Ġcharacterization": 4659,
+      "Ġatmosp": 4660,
+      "Ġmonitoring": 4661,
+      "ani": 4662,
+      "net": 4663,
+      "ĠPa": 4664,
+      "optosis": 4665,
+      "Ġcontin": 4666,
+      "ĠSol": 4667,
+      "Ġdatabase": 4668,
+      "import": 4669,
+      "mann": 4670,
+      "ĠProcess": 4671,
+      "ĠChen": 4672,
+      "Ġgap": 4673,
+      "Ġenzyme": 4674,
+      "OT": 4675,
+      "Ġsimultaneous": 4676,
+      "Ġexistence": 4677,
+      "BP": 4678,
+      "ĠJapan": 4679,
+      "ounts": 4680,
+      "Ġturb": 4681,
+      "Ġspaces": 4682,
+      "ĠWeight": 4683,
+      "ophil": 4684,
+      "Ġast": 4685,
+      "Ġwrite": 4686,
+      "Ġdiabetes": 4687,
+      "ĠCA": 4688,
+      "Ġneutral": 4689,
+      "Ġvariations": 4690,
+      "axon": 4691,
+      "Ġbegin": 4692,
+      "under": 4693,
+      "Ġextraction": 4694,
+      "ĠPati": 4695,
+      "Ġfron": 4696,
+      "efined": 4697,
+      "Ġacids": 4698,
+      "Ġservices": 4699,
+      "Ġsense": 4700,
+      "Ġagent": 4701,
+      "hens": 4702,
+      "electric": 4703,
+      "values": 4704,
+      "Ġimprovement": 4705,
+      "herent": 4706,
+      "actic": 4707,
+      "Ġacet": 4708,
+      "cdots": 4709,
+      "Ġamino": 4710,
+      "Ġroom": 4711,
+      "Ġexpress": 4712,
+      "Ġexcept": 4713,
+      "Ġold": 4714,
+      "plant": 4715,
+      "cepti": 4716,
+      "ĠPCR": 4717,
+      "ĠER": 4718,
+      "ĠBoth": 4719,
+      "vex": 4720,
+      "Ġadults": 4721,
+      "Ġpseud": 4722,
+      "Ġalle": 4723,
+      "Ġworks": 4724,
+      "Ġconsumption": 4725,
+      "ipher": 4726,
+      "cm": 4727,
+      "cast": 4728,
+      "Ġnanopar": 4729,
+      "Ïī": 4730,
+      "Ġecon": 4731,
+      "ynamics": 4732,
+      "Ġalter": 4733,
+      "Ġskin": 4734,
+      "Ġdiameter": 4735,
+      "GC": 4736,
+      "ĠSign": 4737,
+      "vial": 4738,
+      "Ġglucose": 4739,
+      "ĠNorth": 4740,
+      "otox": 4741,
+      "Ġprote": 4742,
+      "dx": 4743,
+      "ĠCr": 4744,
+      "Ġfract": 4745,
+      "Ġinside": 4746,
+      "Ġstatic": 4747,
+      "wid": 4748,
+      "Ġstorage": 4749,
+      "ĠAL": 4750,
+      "ĠMark": 4751,
+      "ĠAT": 4752,
+      "Ġsensitive": 4753,
+      "Ġads": 4754,
+      "Ġedges": 4755,
+      "ana": 4756,
+      "Re": 4757,
+      "Ġsummar": 4758,
+      "ĠAND": 4759,
+      "Ġremaining": 4760,
+      "ditionally": 4761,
+      "Ġmid": 4762,
+      "ĠTheory": 4763,
+      "MC": 4764,
+      "Ġflex": 4765,
+      "oly": 4766,
+      "Ġdegradation": 4767,
+      "Ġintr": 4768,
+      "ota": 4769,
+      "isms": 4770,
+      "Ġampl": 4771,
+      "ĠAre": 4772,
+      "Ġworking": 4773,
+      "Ġdiversity": 4774,
+      "Ġtensor": 4775,
+      "Ġbinary": 4776,
+      "\"\"\"": 4777,
+      "vals": 4778,
+      "Ġhem": 4779,
+      "ML": 4780,
+      "Ġμg": 4781,
+      "neq": 4782,
+      "ensities": 4783,
+      "Ġtakes": 4784,
+      "Ġcharg": 4785,
+      "Ġintervention": 4786,
+      "Ġalb": 4787,
+      "Ġqual": 4788,
+      "Ġmentioned": 4789,
+      "Ġones": 4790,
+      "ĠAccording": 4791,
+      "illed": 4792,
+      "OH": 4793,
+      "Sup": 4794,
+      "Ġgalaxies": 4795,
+      "aily": 4796,
+      "Ġrule": 4797,
+      "Ġcognitive": 4798,
+      "hern": 4799,
+      "Ġrecognition": 4800,
+      "Ġbuffer": 4801,
+      "Ġmarg": 4802,
+      "ĠNi": 4803,
+      "ĠâĪļ": 4804,
+      "Ġclin": 4805,
+      "Ġintegration": 4806,
+      "Ġsin": 4807,
+      "ĠAlso": 4808,
+      "Ġmachine": 4809,
+      "wr": 4810,
+      "idity": 4811,
+      "Ġsubsequent": 4812,
+      "Fe": 4813,
+      "Ġnames": 4814,
+      "ather": 4815,
+      "ĠCy": 4816,
+      "Ġmetabolism": 4817,
+      "Ġreactions": 4818,
+      "Ġiter": 4819,
+      "Ġnoted": 4820,
+      "Ġcauses": 4821,
+      "ĠHamilton": 4822,
+      "go": 4823,
+      "Ġrare": 4824,
+      "VA": 4825,
+      "ĠMy": 4826,
+      "vol": 4827,
+      "asure": 4828,
+      "Ġsignificance": 4829,
+      "ĠNone": 4830,
+      "Ġvehic": 4831,
+      "SR": 4832,
+      "Ġvariability": 4833,
+      "ĠDevelop": 4834,
+      "aren": 4835,
+      "Ġpromot": 4836,
+      "ards": 4837,
+      "Ġcomputational": 4838,
+      "Ġshall": 4839,
+      "izations": 4840,
+      "ĠHydrogen": 4841,
+      "Ġproliferation": 4842,
+      "Ġcoupled": 4843,
+      "chron": 4844,
+      "Ġconvergence": 4845,
+      "Ġgast": 4846,
+      "Ġcalculate": 4847,
+      "raft": 4848,
+      "paration": 4849,
+      "heric": 4850,
+      "ĠPC": 4851,
+      "plate": 4852,
+      "ptions": 4853,
+      "ĠAlgorithm": 4854,
+      "Ġresulted": 4855,
+      "DE": 4856,
+      "Ġinvestigation": 4857,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 4858,
+      "olation": 4859,
+      "Ġtasks": 4860,
+      "Ġleg": 4861,
+      "iness": 4862,
+      "Ġemployed": 4863,
+      "On": 4864,
+      "Ġexperi": 4865,
+      "Ġtraject": 4866,
+      "GA": 4867,
+      "Ġpurpose": 4868,
+      "ĠNum": 4869,
+      "Ġcompletely": 4870,
+      "that": 4871,
+      "ĠOptim": 4872,
+      "Ġformal": 4873,
+      "eck": 4874,
+      "ĠProtein": 4875,
+      "Ġgoal": 4876,
+      "Ġthroughout": 4877,
+      "Ġconsidering": 4878,
+      "Ġreflect": 4879,
+      "treated": 4880,
+      "oration": 4881,
+      "ribution": 4882,
+      "Ġtherapeutic": 4883,
+      "Ġfinding": 4884,
+      "UN": 4885,
+      "Then": 4886,
+      "ilities": 4887,
+      "Ġunknown": 4888,
+      "overed": 4889,
+      "Ġvertex": 4890,
+      "Ġexchange": 4891,
+      "Ġdrugs": 4892,
+      "ĠCP": 4893,
+      "Ġinstr": 4894,
+      "Ġsymmetric": 4895,
+      "ĠDep": 4896,
+      "Ġconstructed": 4897,
+      "Ġprevalence": 4898,
+      "Ġdecreases": 4899,
+      "ĠmiR": 4900,
+      "Ġyet": 4901,
+      "Ġbox": 4902,
+      "graph": 4903,
+      "widehat": 4904,
+      "alian": 4905,
+      "ufact": 4906,
+      "LR": 4907,
+      "cription": 4908,
+      "Ġnp": 4909,
+      "ĠCharacter": 4910,
+      "Ġepid": 4911,
+      "ν": 4912,
+      "Ġstages": 4913,
+      "Ġsay": 4914,
+      "ĠDuring": 4915,
+      "atur": 4916,
+      "ientif": 4917,
+      "abric": 4918,
+      "ü": 4919,
+      "ament": 4920,
+      "inations": 4921,
+      "Ġsolar": 4922,
+      "Ġdiscrete": 4923,
+      "ĠEr": 4924,
+      "ĠGeneral": 4925,
+      "bal": 4926,
+      "ĠCent": 4927,
+      "uel": 4928,
+      "Ġmixture": 4929,
+      "Ġwidely": 4930,
+      "ĠSecond": 4931,
+      "Ġresources": 4932,
+      "ĠAppro": 4933,
+      "ĠIR": 4934,
+      "Ġstring": 4935,
+      "opro": 4936,
+      "Ġinner": 4937,
+      "ĠComplex": 4938,
+      "OP": 4939,
+      "Ġatoms": 4940,
+      "Ġphases": 4941,
+      "Ġdomains": 4942,
+      "ada": 4943,
+      "Ġcountries": 4944,
+      "acet": 4945,
+      "ociation": 4946,
+      "izer": 4947,
+      "Ġitself": 4948,
+      "Ġminimal": 4949,
+      "ĠControl": 4950,
+      "ttp": 4951,
+      "Ġbottom": 4952,
+      "ball": 4953,
+      "ĠMay": 4954,
+      "dev": 4955,
+      "now": 4956,
+      "ember": 4957,
+      "Ġpercentage": 4958,
+      "ĠOther": 4959,
+      "omas": 4960,
+      "Ġled": 4961,
+      "Res": 4962,
+      "ĠEng": 4963,
+      "kg": 4964,
+      "Ġfrequencies": 4965,
+      "kin": 4966,
+      "Ġincidence": 4967,
+      "Ġanimal": 4968,
+      "Ġadop": 4969,
+      "Ġidentity": 4970,
+      "ĠRT": 4971,
+      "Ġyoung": 4972,
+      "istent": 4973,
+      "weight": 4974,
+      "gu": 4975,
+      "Ġseason": 4976,
+      "Ġexplained": 4977,
+      "ĠUnder": 4978,
+      "iotic": 4979,
+      "well": 4980,
+      "Ġmetabolic": 4981,
+      "gical": 4982,
+      "±": 4983,
+      "Theorem": 4984,
+      "ades": 4985,
+      "plicated": 4986,
+      "Ġcontained": 4987,
+      "Ġsulf": 4988,
+      "Ġcool": 4989,
+      "Ġperson": 4990,
+      "Ïģ": 4991,
+      "Ġpix": 4992,
+      "ĠSal": 4993,
+      "link": 4994,
+      "ini": 4995,
+      "tual": 4996,
+      "SH": 4997,
+      "ged": 4998,
+      "ky": 4999,
+      "asts": 5000,
+      "ercise": 5001,
+      "ĠHar": 5002,
+      "Ġrelax": 5003,
+      "equiv": 5004,
+      "Ġyour": 5005,
+      "Ġunderg": 5006,
+      "Ġrecovery": 5007,
+      "Ġcomm": 5008,
+      "Ġdenotes": 5009,
+      "formed": 5010,
+      "aria": 5011,
+      "etic": 5012,
+      "Ġtumors": 5013,
+      "ĠHy": 5014,
+      "Ġmarkers": 5015,
+      "Ġplaced": 5016,
+      "olute": 5017,
+      "Ġwaves": 5018,
+      "Ġuncertainty": 5019,
+      "Ġcontribute": 5020,
+      "ĠHist": 5021,
+      "Ġaver": 5022,
+      "Ġfav": 5023,
+      "Ġpow": 5024,
+      "ĠSee": 5025,
+      "Ġteam": 5026,
+      "Ġscales": 5027,
+      "ientific": 5028,
+      "ierarch": 5029,
+      "Ġearlier": 5030,
+      "Ġsatisfies": 5031,
+      "Ġcrystal": 5032,
+      "Ġpregn": 5033,
+      "Ġobserve": 5034,
+      "Ġonline": 5035,
+      "Ġcontributions": 5036,
+      "ogram": 5037,
+      "ĠMa": 5038,
+      "Ġfrac": 5039,
+      "Ġspread": 5040,
+      "Ġonce": 5041,
+      "det": 5042,
+      "Ġrespond": 5043,
+      "Ġplatform": 5044,
+      "Ġinflammatory": 5045,
+      "utive": 5046,
+      "ĠSumm": 5047,
+      "place": 5048,
+      "Ġions": 5049,
+      "Ġwindow": 5050,
+      "axis": 5051,
+      "estinal": 5052,
+      "Ġdepending": 5053,
+      "Ġseparation": 5054,
+      "Ġforward": 5055,
+      "ĠTi": 5056,
+      "Ġglass": 5057,
+      "Ġaccept": 5058,
+      "Ġfeedback": 5059,
+      "Ġonto": 5060,
+      "ME": 5061,
+      "merc": 5062,
+      "unctional": 5063,
+      "Ġapoptosis": 5064,
+      "ĠProperty": 5065,
+      "Ġintegrated": 5066,
+      "Ġorb": 5067,
+      "Ġdeviation": 5068,
+      "Ġantibodies": 5069,
+      "Ġremoved": 5070,
+      "Ġlipid": 5071,
+      "armac": 5072,
+      "Ġarbitrary": 5073,
+      "agger": 5074,
+      "Ġembry": 5075,
+      "Ġgrain": 5076,
+      "Ġdrop": 5077,
+      "Ġstarting": 5078,
+      "Ġrelationships": 5079,
+      "ĠÏĩ": 5080,
+      "SF": 5081,
+      "Ġsimply": 5082,
+      "Ġfacilit": 5083,
+      "Ġzone": 5084,
+      "ils": 5085,
+      "Psi": 5086,
+      "Ġinequality": 5087,
+      "Keywords": 5088,
+      "Ġtoler": 5089,
+      "edge": 5090,
+      "Ġeasy": 5091,
+      "Ġalpha": 5092,
+      "Ġperf": 5093,
+      "width": 5094,
+      "init": 5095,
+      "Ġimplemented": 5096,
+      "CF": 5097,
+      "osity": 5098,
+      "ocyte": 5099,
+      "Ġproportion": 5100,
+      "rest": 5101,
+      "ĠSuper": 5102,
+      "Ġpref": 5103,
+      "Ġword": 5104,
+      "ev": 5105,
+      "Ġextent": 5106,
+      "Ġinjection": 5107,
+      "alled": 5108,
+      "ĠAnti": 5109,
+      "Ġbeta": 5110,
+      "ĠJan": 5111,
+      "ĠGa": 5112,
+      "ĠZhang": 5113,
+      "Ġiron": 5114,
+      "Ġquantitative": 5115,
+      "roc": 5116,
+      "Ġfall": 5117,
+      "Ġregarding": 5118,
+      "Ġfix": 5119,
+      "Ġdatasets": 5120,
+      "Ġtend": 5121,
+      "Ġscalar": 5122,
+      "Ġresidual": 5123,
+      "Ġratios": 5124,
+      "ĠΦ": 5125,
+      "king": 5126,
+      "Ġinflammation": 5127,
+      "Ġsingular": 5128,
+      "ĠPark": 5129,
+      "omatic": 5130,
+      "unctions": 5131,
+      "Ġwar": 5132,
+      "ÍĴ": 5133,
+      "hemat": 5134,
+      "Ġface": 5135,
+      "ĠHu": 5136,
+      "Ġfundamental": 5137,
+      "Ġwavelength": 5138,
+      "eling": 5139,
+      "ĠSuch": 5140,
+      "RNAs": 5141,
+      "ct": 5142,
+      "Ġiden": 5143,
+      "cean": 5144,
+      "new": 5145,
+      "Type": 5146,
+      "ĠFormula": 5147,
+      "Ġmedic": 5148,
+      "ussion": 5149,
+      "Ġdistingu": 5150,
+      "Ġresonance": 5151,
+      "ATION": 5152,
+      "inear": 5153,
+      "Ġhyd": 5154,
+      "ln": 5155,
+      "âĨĴ": 5156,
+      "ĠUp": 5157,
+      "Ġactual": 5158,
+      "Ġadapt": 5159,
+      "hene": 5160,
+      "Ġmotor": 5161,
+      "list": 5162,
+      "abit": 5163,
+      "Ind": 5164,
+      "otal": 5165,
+      "Ġneighbor": 5166,
+      "ĠPT": 5167,
+      "gener": 5168,
+      "Ġpossibility": 5169,
+      "ergies": 5170,
+      "Ġseems": 5171,
+      "ĠUS": 5172,
+      "Ġimm": 5173,
+      "Ġtypically": 5174,
+      "Ġsimulated": 5175,
+      "ĠSystems": 5176,
+      "ectiveness": 5177,
+      "rying": 5178,
+      "Ġkinase": 5179,
+      "Ġdecomposition": 5180,
+      "ateral": 5181,
+      "Ġrotation": 5182,
+      "pendix": 5183,
+      "enn": 5184,
+      "att": 5185,
+      "vate": 5186,
+      "Ġtargets": 5187,
+      "Ġsituation": 5188,
+      "Ġinvolve": 5189,
+      "Ġcreated": 5190,
+      "hesized": 5191,
+      "Ġalone": 5192,
+      "ci": 5193,
+      "ĠmL": 5194,
+      "Ġdivided": 5195,
+      "Ġbulk": 5196,
+      "oin": 5197,
+      "HC": 5198,
+      "Ġarm": 5199,
+      "LO": 5200,
+      "ills": 5201,
+      "Ġmedian": 5202,
+      "ham": 5203,
+      "imer": 5204,
+      "flu": 5205,
+      "Ġfiber": 5206,
+      "ĠSU": 5207,
+      "file": 5208,
+      "tivated": 5209,
+      "Ġradio": 5210,
+      "ĠNames": 5211,
+      "pe": 5212,
+      "Ġoste": 5213,
+      "Ġelim": 5214,
+      "Ġsuscepti": 5215,
+      "rehens": 5216,
+      "Ġdiscussion": 5217,
+      "ĠSep": 5218,
+      "Ġarchitecture": 5219,
+      "Ġdest": 5220,
+      "typ": 5221,
+      "rame": 5222,
+      "Ġpartition": 5223,
+      "Ġoccurred": 5224,
+      "Ġsizes": 5225,
+      "cles": 5226,
+      "Ġsched": 5227,
+      "Molecular": 5228,
+      "Ġκ": 5229,
+      "Ġinvas": 5230,
+      "cup": 5231,
+      "PCR": 5232,
+      "ĠSMILES": 5233,
+      "tially": 5234,
+      "oxide": 5235,
+      "ĠEd": 5236,
+      "Ġmanufact": 5237,
+      "ĠMaterial": 5238,
+      "Ġflat": 5239,
+      "Ġmutation": 5240,
+      "Ġintroduce": 5241,
+      "bound": 5242,
+      "Ġdisorders": 5243,
+      "regulated": 5244,
+      "ĠMor": 5245,
+      "Ġfalse": 5246,
+      "inger": 5247,
+      "ĠTR": 5248,
+      "Ġextrem": 5249,
+      "war": 5250,
+      "Ġsymbol": 5251,
+      "Ġanomal": 5252,
+      "ĠAR": 5253,
+      "Ġissues": 5254,
+      "Ġcoordinates": 5255,
+      "Ġreceptors": 5256,
+      "Ġprogression": 5257,
+      "ĠFl": 5258,
+      "ublic": 5259,
+      "Ġelectronic": 5260,
+      "Ġaspects": 5261,
+      "Ġdocument": 5262,
+      "flo": 5263,
+      "ĠPred": 5264,
+      "Ġgraphs": 5265,
+      "Ġtraditional": 5266,
+      "DM": 5267,
+      "Ġsafety": 5268,
+      "ĠDr": 5269,
+      "ĠSequ": 5270,
+      "Ġcomposite": 5271,
+      "ĠÎĽ": 5272,
+      "Ġresponsible": 5273,
+      "Ġgran": 5274,
+      "Ġintermediate": 5275,
+      "odium": 5276,
+      "posite": 5277,
+      "phase": 5278,
+      "dt": 5279,
+      "Ġweek": 5280,
+      "Ġdos": 5281,
+      "Ġstabil": 5282,
+      "LC": 5283,
+      "ĠKey": 5284,
+      "Ġvertices": 5285,
+      "Ġcomputer": 5286,
+      "ĠCanonical": 5287,
+      "Ġinvariant": 5288,
+      "emark": 5289,
+      "benz": 5290,
+      "Ġice": 5291,
+      "tile": 5292,
+      "zy": 5293,
+      "ĠOut": 5294,
+      "Ġmovement": 5295,
+      "Ġshif": 5296,
+      "leep": 5297,
+      "Ġdaily": 5298,
+      "Ġpositions": 5299,
+      "Ġhim": 5300,
+      "Ġcreate": 5301,
+      "Our": 5302,
+      "Ġresearc": 5303,
+      "Ġprogn": 5304,
+      "duct": 5305,
+      "Ġscreening": 5306,
+      "Ġchoose": 5307,
+      "process": 5308,
+      "mal": 5309,
+      "Ġlaboratory": 5310,
+      "Ġoperations": 5311,
+      "Ġtools": 5312,
+      "ologic": 5313,
+      "qquad": 5314,
+      "Ġcommonly": 5315,
+      "Ġvoid": 5316,
+      "Ġoccup": 5317,
+      "associated": 5318,
+      "Ġcorrelations": 5319,
+      "Ġcarcinoma": 5320,
+      "lin": 5321,
+      "Ġvideo": 5322,
+      "Ġheavy": 5323,
+      "Ġlargest": 5324,
+      "Ġmiddle": 5325,
+      "ĉĉĉĉ": 5326,
+      "ĠBas": 5327,
+      "asons": 5328,
+      "iding": 5329,
+      "Ġetc": 5330,
+      "ache": 5331,
+      "ĠEval": 5332,
+      "ira": 5333,
+      "romagnetic": 5334,
+      "Ġcovari": 5335,
+      "LI": 5336,
+      "Ġdele": 5337,
+      "Ġstra": 5338,
+      "amples": 5339,
+      "oder": 5340,
+      "Ġcategory": 5341,
+      "ĠInstit": 5342,
+      "Ġpolicy": 5343,
+      "Based": 5344,
+      "ibly": 5345,
+      "Ġdetermination": 5346,
+      "Ġrespir": 5347,
+      "otropic": 5348,
+      "Ġolder": 5349,
+      "ĠMal": 5350,
+      "Ġcytok": 5351,
+      "Ġdegrees": 5352,
+      "aut": 5353,
+      "illing": 5354,
+      "eting": 5355,
+      "Ġreduces": 5356,
+      "Ġideal": 5357,
+      "binding": 5358,
+      "ĠSpect": 5359,
+      "unit": 5360,
+      "Ġdiver": 5361,
+      "ĠWorld": 5362,
+      "Ġmarked": 5363,
+      "aly": 5364,
+      "Ġcomplexes": 5365,
+      "ĠSummary": 5366,
+      "Ġpropose": 5367,
+      "ĠAustr": 5368,
+      "Ġmaxim": 5369,
+      "Ġround": 5370,
+      "Ġinhibitor": 5371,
+      "Ġefficacy": 5372,
+      "actor": 5373,
+      "bur": 5374,
+      "Ġtransf": 5375,
+      "ĠGal": 5376,
+      "Ġproved": 5377,
+      "ĠDefined": 5378,
+      "At": 5379,
+      "Ġselect": 5380,
+      "Ġnanoparticles": 5381,
+      "Wh": 5382,
+      "ken": 5383,
+      "ĠSP": 5384,
+      "enge": 5385,
+      "Ġdelivery": 5386,
+      "Ġdisorder": 5387,
+      "ĠInChI": 5388,
+      "ĠComparison": 5389,
+      "ifying": 5390,
+      "ĠMechan": 5391,
+      "Ġconclude": 5392,
+      "Ġrepeated": 5393,
+      "ellow": 5394,
+      "ĠÃĢ": 5395,
+      "CI": 5396,
+      "ĠHz": 5397,
+      "analysis": 5398,
+      "Tr": 5399,
+      "ÃŃ": 5400,
+      "elihood": 5401,
+      "Ġexpand": 5402,
+      "ĠDevelopment": 5403,
+      "ĠState": 5404,
+      "Ġtet": 5405,
+      "ffic": 5406,
+      "Ġparent": 5407,
+      "Ġscenario": 5408,
+      "rs": 5409,
+      "ĠWhat": 5410,
+      "âī": 5411,
+      "Ġstimulation": 5412,
+      "ĠObs": 5413,
+      "zero": 5414,
+      "Ġmanner": 5415,
+      "ashed": 5416,
+      "ĠLog": 5417,
+      "Ġoxide": 5418,
+      "phosph": 5419,
+      "Ġmigration": 5420,
+      "Ġsubgroup": 5421,
+      "rosis": 5422,
+      "ipp": 5423,
+      "DR": 5424,
+      "dec": 5425,
+      "osomal": 5426,
+      "Ġsegment": 5427,
+      "ogenous": 5428,
+      "FP": 5429,
+      "hand": 5430,
+      "ĠSurface": 5431,
+      "itz": 5432,
+      "Ġcrystall": 5433,
+      "this": 5434,
+      "Ġbuilding": 5435,
+      "tag": 5436,
+      "Ġreducing": 5437,
+      "Ġuns": 5438,
+      "Ġrecomb": 5439,
+      "Ġcam": 5440,
+      "Ġlimits": 5441,
+      "ocardi": 5442,
+      "&&": 5443,
+      "Ġseparate": 5444,
+      "Ġsupplement": 5445,
+      "kele": 5446,
+      "Ġgrad": 5447,
+      "Ġissue": 5448,
+      "ĠQuantum": 5449,
+      "Ġcurrently": 5450,
+      "Ġquite": 5451,
+      "EP": 5452,
+      "Ġrules": 5453,
+      "Ġweights": 5454,
+      "uary": 5455,
+      "illi": 5456,
+      "Ġbecame": 5457,
+      "ó": 5458,
+      "Ġnormalized": 5459,
+      "ĠNetworks": 5460,
+      "erved": 5461,
+      "Ġstatistics": 5462,
+      "ĠTime": 5463,
+      "ĠUV": 5464,
+      "Ġcav": 5465,
+      "used": 5466,
+      "Ġfish": 5467,
+      "Ġmajority": 5468,
+      "ĠPe": 5469,
+      "Ġcohort": 5470,
+      "Ġsemi": 5471,
+      "Ġgame": 5472,
+      "monary": 5473,
+      "MM": 5474,
+      "oded": 5475,
+      "Ġvent": 5476,
+      "Ġauto": 5477,
+      "Ġabundance": 5478,
+      "nov": 5479,
+      "Ġasymptotic": 5480,
+      "Ġtreatments": 5481,
+      "uly": 5482,
+      "Ġconstraint": 5483,
+      "Ġbey": 5484,
+      "ĠSO": 5485,
+      "Ġstd": 5486,
+      "Ġdeveloping": 5487,
+      "ĠNot": 5488,
+      "Lemma": 5489,
+      "Ġapparent": 5490,
+      "Ġcircuit": 5491,
+      "From": 5492,
+      "ĠEuropean": 5493,
+      "Ġsolve": 5494,
+      "ĠÍij": 5495,
+      "ux": 5496,
+      "Ġbeyond": 5497,
+      "ept": 5498,
+      "Ġappe": 5499,
+      "requency": 5500,
+      "Ġvacu": 5501,
+      "ĠIndeed": 5502,
+      "ĠChemical": 5503,
+      "ĠUndefined": 5504,
+      "Note": 5505,
+      "Ġnull": 5506,
+      "Ġinverse": 5507,
+      "Ġnamely": 5508,
+      "Ġshear": 5509,
+      "mL": 5510,
+      "All": 5511,
+      "Rec": 5512,
+      "Ġgeneralized": 5513,
+      "ranes": 5514,
+      "ĠTest": 5515,
+      "iling": 5516,
+      "Ġfluorescence": 5517,
+      "ĠΣ": 5518,
+      "Ġindepend": 5519,
+      "diff": 5520,
+      "Ġproviding": 5521,
+      "phenyl": 5522,
+      "hing": 5523,
+      "Ġviral": 5524,
+      "ĠBecause": 5525,
+      "Ġintrac": 5526,
+      "ĠHig": 5527,
+      "Ġwant": 5528,
+      "Ġprinciple": 5529,
+      "anol": 5530,
+      "Ġha": 5531,
+      "ovascular": 5532,
+      "Ġformer": 5533,
+      "Ġestablish": 5534,
+      "Ġadvantage": 5535,
+      "III": 5536,
+      "Ġsequencing": 5537,
+      "Ġprocedures": 5538,
+      "tra": 5539,
+      "index": 5540,
+      "fe": 5541,
+      "Ġpi": 5542,
+      "Ġobvious": 5543,
+      "Ġregime": 5544,
+      "sur": 5545,
+      "Ġpresents": 5546,
+      "Ġdisplac": 5547,
+      "Ġdecl": 5548,
+      "ĠAppendix": 5549,
+      "Ġinteract": 5550,
+      "lands": 5551,
+      "inate": 5552,
+      "omorphic": 5553,
+      "Ġlowest": 5554,
+      "Ġartif": 5555,
+      "Ġinvolving": 5556,
+      "Ġcommerc": 5557,
+      "Ġdop": 5558,
+      "Ġconform": 5559,
+      "ĠIg": 5560,
+      "rolog": 5561,
+      "vised": 5562,
+      "Ġflo": 5563,
+      "Ġcardiac": 5564,
+      "pts": 5565,
+      "rig": 5566,
+      "Ġensure": 5567,
+      "Ġaccumulation": 5568,
+      "Ġentropy": 5569,
+      "Ġidea": 5570,
+      "perature": 5571,
+      "Ġquestions": 5572,
+      "ĠPR": 5573,
+      "Ġstatistically": 5574,
+      "dagger": 5575,
+      "Ġnitrogen": 5576,
+      "scr": 5577,
+      "ĠDiscussion": 5578,
+      "Ġreports": 5579,
+      "Ġpulse": 5580,
+      "Ġrequirements": 5581,
+      "Ġcomparing": 5582,
+      "quired": 5583,
+      "layer": 5584,
+      "Ġspectroscopy": 5585,
+      "vironments": 5586,
+      "Ġscaling": 5587,
+      "Ġexposed": 5588,
+      "MB": 5589,
+      "ξ": 5590,
+      "Ġhole": 5591,
+      "Ġá": 5592,
+      "Ġsimilarity": 5593,
+      "Ġvariants": 5594,
+      "body": 5595,
+      "Ġkeep": 5596,
+      "ĠCancer": 5597,
+      "edi": 5598,
+      "osomes": 5599,
+      "Ç«": 5600,
+      "Ad": 5601,
+      "âĪŀ": 5602,
+      "monic": 5603,
+      "ging": 5604,
+      "split": 5605,
+      "know": 5606,
+      "Ġrough": 5607,
+      "hematical": 5608,
+      "vision": 5609,
+      "Ġded": 5610,
+      "Ġcycles": 5611,
+      "Ġfamil": 5612,
+      "Ġadministration": 5613,
+      "etal": 5614,
+      "Ġcoron": 5615,
+      "Ġinfections": 5616,
+      "Ġmacroph": 5617,
+      "atics": 5618,
+      "Ġpredictions": 5619,
+      "isher": 5620,
+      "erent": 5621,
+      "reted": 5622,
+      "include": 5623,
+      "Ġclimate": 5624,
+      "sec": 5625,
+      "========": 5626,
+      "ĠMS": 5627,
+      "Ġcompe": 5628,
+      "ratic": 5629,
+      "lig": 5630,
+      "poses": 5631,
+      "Ġpolarization": 5632,
+      "llip": 5633,
+      "derived": 5634,
+      "Ġreleased": 5635,
+      "Ġconnection": 5636,
+      "lic": 5637,
+      "Ġcoli": 5638,
+      "Ġoutside": 5639,
+      "Ġabsolute": 5640,
+      "esian": 5641,
+      "ĠEnd": 5642,
+      "ĠOf": 5643,
+      "Ġidentical": 5644,
+      "Ġmodule": 5645,
+      "Ġmitochondrial": 5646,
+      "Ġadvanced": 5647,
+      "ingly": 5648,
+      "formance": 5649,
+      "Ġtoward": 5650,
+      "uding": 5651,
+      "ek": 5652,
+      "Ġmeaning": 5653,
+      "crib": 5654,
+      "ulator": 5655,
+      "FN": 5656,
+      "key": 5657,
+      "cons": 5658,
+      "Ġapplying": 5659,
+      "ishes": 5660,
+      "Ġmamm": 5661,
+      "Ġderivatives": 5662,
+      "Ġorientation": 5663,
+      "Ġstochastic": 5664,
+      "ĠAug": 5665,
+      "Ġrenal": 5666,
+      "ĠGreen": 5667,
+      "Ġcomplement": 5668,
+      "obl": 5669,
+      "pirical": 5670,
+      "orts": 5671,
+      "BM": 5672,
+      "Ġexcess": 5673,
+      "Ġmorphology": 5674,
+      "Ġsound": 5675,
+      "ifier": 5676,
+      "Ġimplications": 5677,
+      "ĠDesign": 5678,
+      "approx": 5679,
+      "prop": 5680,
+      "Ġcandidate": 5681,
+      "Ġdepos": 5682,
+      "Ġequip": 5683,
+      "ustain": 5684,
+      "inese": 5685,
+      "etry": 5686,
+      "Ġpotentially": 5687,
+      "Ġstraight": 5688,
+      "Ġcruc": 5689,
+      "iology": 5690,
+      "Ġkernel": 5691,
+      "Ġalcoh": 5692,
+      "idden": 5693,
+      "return": 5694,
+      "Ġcorrection": 5695,
+      "rot": 5696,
+      "Ġmicroscopy": 5697,
+      "Ġfoot": 5698,
+      "GL": 5699,
+      "ĠCells": 5700,
+      "irth": 5701,
+      "yg": 5702,
+      "ĠPath": 5703,
+      "outhern": 5704,
+      "ĠLong": 5705,
+      "Ġrevers": 5706,
+      "ε": 5707,
+      "arse": 5708,
+      "Ġcereb": 5709,
+      "isted": 5710,
+      "Ġpuls": 5711,
+      "Ġdisk": 5712,
+      "itud": 5713,
+      "Ġdu": 5714,
+      "Ġangular": 5715,
+      "chem": 5716,
+      "length": 5717,
+      "Ġexactly": 5718,
+      "roke": 5719,
+      "uth": 5720,
+      "Ġcond": 5721,
+      "insic": 5722,
+      "Ġrise": 5723,
+      "take": 5724,
+      "Ġtopological": 5725,
+      "Ġremark": 5726,
+      "ollary": 5727,
+      "Ġcer": 5728,
+      "TE": 5729,
+      "nment": 5730,
+      "Ġbuilt": 5731,
+      "Ġfre": 5732,
+      "Ġenergies": 5733,
+      "ecting": 5734,
+      "ĠTem": 5735,
+      "rared": 5736,
+      "ĠNow": 5737,
+      "charge": 5738,
+      "Ġlocations": 5739,
+      "Ġbalance": 5740,
+      "Ġla": 5741,
+      "Ġreached": 5742,
+      "lammatory": 5743,
+      "Ġfabric": 5744,
+      "ifications": 5745,
+      "Ġdiagnostic": 5746,
+      "Ġmutant": 5747,
+      "ĠNO": 5748,
+      "HD": 5749,
+      "ĠAB": 5750,
+      "Ġdiscrim": 5751,
+      "Ġprecip": 5752,
+      "ĠThree": 5753,
+      "Ġinser": 5754,
+      "Ġinfected": 5755,
+      "Ġconstants": 5756,
+      "Ω": 5757,
+      "negative": 5758,
+      "Ġconfidence": 5759,
+      "ĠPatients": 5760,
+      "ollowing": 5761,
+      "ads": 5762,
+      "Ġhypert": 5763,
+      "ĠInternational": 5764,
+      "Def": 5765,
+      "ariate": 5766,
+      "Ġintervals": 5767,
+      "Ġexercise": 5768,
+      "Ġeducation": 5769,
+      "Ġremoval": 5770,
+      "thern": 5771,
+      "ster": 5772,
+      "Ġinteger": 5773,
+      "ĠPA": 5774,
+      "Ġkid": 5775,
+      "Ġcategories": 5776,
+      "ĠGiven": 5777,
+      "Ġvascular": 5778,
+      "herence": 5779,
+      "mathscr": 5780,
+      "ĠRet": 5781,
+      "Ġinsulin": 5782,
+      "ticip": 5783,
+      "ĠCF": 5784,
+      "Ġlook": 5785,
+      "ymmetry": 5786,
+      "Ġforces": 5787,
+      "ĠPhysical": 5788,
+      "LS": 5789,
+      "care": 5790,
+      "Ġhouse": 5791,
+      "Ġinduce": 5792,
+      "Ġbelie": 5793,
+      "ria": 5794,
+      "ĠAssum": 5795,
+      "Ġcomputing": 5796,
+      "Ġbus": 5797,
+      "âĪİ": 5798,
+      "Ġpractical": 5799,
+      "train": 5800,
+      "TT": 5801,
+      "Ġplastic": 5802,
+      "ĠNor": 5803,
+      "Ġfeas": 5804,
+      "ĠHamiltonian": 5805,
+      "Ġtail": 5806,
+      "ĠZn": 5807,
+      "Ġinterpretation": 5808,
+      "ducing": 5809,
+      "Is": 5810,
+      "Ġexamine": 5811,
+      "ulates": 5812,
+      "Ġmatch": 5813,
+      "ĠÄ": 5814,
+      "ives": 5815,
+      "ameters": 5816,
+      "ĠμM": 5817,
+      "Ġexhibit": 5818,
+      "Ġnit": 5819,
+      "oto": 5820,
+      "ĠClinical": 5821,
+      "ervation": 5822,
+      "ĠAdditionally": 5823,
+      "arant": 5824,
+      "Ġelastic": 5825,
+      "DA": 5826,
+      "otopic": 5827,
+      "Ġactivated": 5828,
+      "Ġter": 5829,
+      "Ġconsequence": 5830,
+      "Ġendot": 5831,
+      "ophag": 5832,
+      "Ġcomparable": 5833,
+      "Ġdominant": 5834,
+      "η": 5835,
+      "Ġvalidation": 5836,
+      "Im": 5837,
+      "ĠÅ": 5838,
+      "Ġleaf": 5839,
+      "Ġfung": 5840,
+      "taining": 5841,
+      "Ġunivers": 5842,
+      "Ġphyl": 5843,
+      "Ġlibr": 5844,
+      "Ġextra": 5845,
+      "Ġprint": 5846,
+      "mediately": 5847,
+      "Ġmaximal": 5848,
+      "idae": 5849,
+      "Ġoral": 5850,
+      "bin": 5851,
+      "Ġpeptide": 5852,
+      "ĠMax": 5853,
+      "arp": 5854,
+      "Ġconclusion": 5855,
+      "Ġsatisfy": 5856,
+      "Ġanalyze": 5857,
+      "ois": 5858,
+      "Ġinfer": 5859,
+      "Ġdraw": 5860,
+      "Ġdepression": 5861,
+      "Ġmetall": 5862,
+      "Ġposterior": 5863,
+      "Ġpeaks": 5864,
+      "sol": 5865,
+      "Ġhorizontal": 5866,
+      "Ġlateral": 5867,
+      "ĠOR": 5868,
+      "NN": 5869,
+      "Ġemo": 5870,
+      "PV": 5871,
+      "TA": 5872,
+      "Ġincubated": 5873,
+      "Ġretrie": 5874,
+      "Ġhumans": 5875,
+      "Ġri": 5876,
+      "Ġsoci": 5877,
+      "onia": 5878,
+      "Ġinterven": 5879,
+      "Ġvarying": 5880,
+      "Ġsti": 5881,
+      "ĠImmun": 5882,
+      "Ġonset": 5883,
+      "Ġleaves": 5884,
+      "Ġotherwise": 5885,
+      "Ġblocks": 5886,
+      "Ġassigned": 5887,
+      "SCs": 5888,
+      "Ġbios": 5889,
+      "Ġmixing": 5890,
+      "ara": 5891,
+      "li": 5892,
+      "Ġdeformation": 5893,
+      "Ġcosts": 5894,
+      "Ġperipher": 5895,
+      "ĠTra": 5896,
+      "Ġatomic": 5897,
+      "Ġrandomly": 5898,
+      "Ġargument": 5899,
+      "Ġitems": 5900,
+      "Ġsuff": 5901,
+      "Ġprobably": 5902,
+      "ners": 5903,
+      "Ġinhibitors": 5904,
+      "Ġbeh": 5905,
+      "ĠDeep": 5906,
+      "Ġpig": 5907,
+      "ĠType": 5908,
+      "ĠMost": 5909,
+      "ura": 5910,
+      "itudinal": 5911,
+      "Ġderivative": 5912,
+      "Ġexplore": 5913,
+      "ĠInformation": 5914,
+      "Ġgrap": 5915,
+      "ĠÎĶ": 5916,
+      "Ġprogress": 5917,
+      "****************": 5918,
+      "Ġul": 5919,
+      "ARS": 5920,
+      "oral": 5921,
+      "ostic": 5922,
+      "Com": 5923,
+      "ĠExternal": 5924,
+      "ĠStatistical": 5925,
+      "ĠRam": 5926,
+      "ĠLo": 5927,
+      "Ġelectrical": 5928,
+      "long": 5929,
+      "Net": 5930,
+      "ENT": 5931,
+      "va": 5932,
+      "ä": 5933,
+      "urations": 5934,
+      "Ġdesired": 5935,
+      "iring": 5936,
+      "Ġphysics": 5937,
+      "Ġmasses": 5938,
+      "ki": 5939,
+      "Ġbands": 5940,
+      "Ġalk": 5941,
+      "ĠSimilarly": 5942,
+      "Ġsurround": 5943,
+      "Ġconvex": 5944,
+      "oster": 5945,
+      "Ġlinked": 5946,
+      "Ġfocused": 5947,
+      "Ġhot": 5948,
+      "Ġmatching": 5949,
+      "Ġoxidation": 5950,
+      "Ġanten": 5951,
+      "miss": 5952,
+      "Ġmental": 5953,
+      "ille": 5954,
+      "iciency": 5955,
+      "ĠLiu": 5956,
+      "Ġprobe": 5957,
+      "ĠEstim": 5958,
+      "Ġindices": 5959,
+      "che": 5960,
+      "ĠRob": 5961,
+      "Ġconv": 5962,
+      "ĠVer": 5963,
+      "apse": 5964,
+      "Si": 5965,
+      "phal": 5966,
+      "Ġlesions": 5967,
+      "Ġmolecule": 5968,
+      "Ġadi": 5969,
+      "Ġdate": 5970,
+      "Ġcomposed": 5971,
+      "Ġaud": 5972,
+      "structure": 5973,
+      "oton": 5974,
+      "infor": 5975,
+      "Ġclustering": 5976,
+      "acent": 5977,
+      "star": 5978,
+      "PO": 5979,
+      "ĠChinese": 5980,
+      "Ġspecifically": 5981,
+      "erential": 5982,
+      "Ġcapture": 5983,
+      "ĠLow": 5984,
+      "Ġfine": 5985,
+      "Ġfemales": 5986,
+      "ĠHow": 5987,
+      "Ġaer": 5988,
+      "vector": 5989,
+      "portun": 5990,
+      "forms": 5991,
+      "zo": 5992,
+      "Ġprecision": 5993,
+      "ypt": 5994,
+      "Ġminutes": 5995,
+      "κ": 5996,
+      "Ġoxidative": 5997,
+      "conn": 5998,
+      "ensus": 5999,
+      "Ġtrace": 6000,
+      "Ġconjug": 6001,
+      "Ġhighlight": 6002,
+      "ss": 6003,
+      "ĠExperimental": 6004,
+      "ĠThat": 6005,
+      "artment": 6006,
+      "MO": 6007,
+      "''": 6008,
+      "ometer": 6009,
+      "Ġstop": 6010,
+      "Ġrib": 6011,
+      "Ġouter": 6012,
+      "rh": 6013,
+      "ript": 6014,
+      "Ġfluctuations": 6015,
+      "obs": 6016,
+      "non": 6017,
+      "Ġquark": 6018,
+      "Ġð": 6019,
+      "ĠMac": 6020,
+      "Ġperiods": 6021,
+      "rolled": 6022,
+      "AV": 6023,
+      "ĠOc": 6024,
+      "ĠImage": 6025,
+      "ĠBel": 6026,
+      "Ġpropagation": 6027,
+      "ĠDon": 6028,
+      "www": 6029,
+      "glish": 6030,
+      "Ġexhibited": 6031,
+      "ogeneity": 6032,
+      "ĠBack": 6033,
+      "Ġactions": 6034,
+      "ski": 6035,
+      "ĠAmong": 6036,
+      "Ġbrief": 6037,
+      "riers": 6038,
+      "ĠNF": 6039,
+      "positive": 6040,
+      "sequently": 6041,
+      "ulence": 6042,
+      "Ġenvironments": 6043,
+      "Ġcurv": 6044,
+      "omics": 6045,
+      "Ġbit": 6046,
+      "Ġgel": 6047,
+      "Ġrepresentations": 6048,
+      "Ġaway": 6049,
+      "ĠField": 6050,
+      "obic": 6051,
+      "CG": 6052,
+      "Ġcomprehens": 6053,
+      "Ġhierarch": 6054,
+      "Ġinduces": 6055,
+      "BD": 6056,
+      "Ġhapp": 6057,
+      "Ġeight": 6058,
+      "Ġgravity": 6059,
+      "Ġadaptive": 6060,
+      "BL": 6061,
+      "genic": 6062,
+      "Ġinstruc": 6063,
+      "Ġanalytical": 6064,
+      "ĠOx": 6065,
+      "ĠCON": 6066,
+      "Ġsurgical": 6067,
+      "Ġdip": 6068,
+      "ato": 6069,
+      "Ġrandomized": 6070,
+      "Ġroles": 6071,
+      "dep": 6072,
+      "ĠâĪĩ": 6073,
+      "chang": 6074,
+      "Ġdispersion": 6075,
+      "Ġseparated": 6076,
+      "ĠOrgan": 6077,
+      "ĠVi": 6078,
+      "ĠJohn": 6079,
+      "Ġannot": 6080,
+      "Ġresource": 6081,
+      "energy": 6082,
+      "relation": 6083,
+      "mean": 6084,
+      "ĠBen": 6085,
+      "Ġconfirm": 6086,
+      "With": 6087,
+      "Ġinfinite": 6088,
+      "ĠScience": 6089,
+      "Ġsuccessfully": 6090,
+      "Ġlocalization": 6091,
+      "mode": 6092,
+      "https": 6093,
+      "gebras": 6094,
+      "idelines": 6095,
+      "Ġeffectiveness": 6096,
+      "hyd": 6097,
+      "Ġsaid": 6098,
+      "ico": 6099,
+      "Ġtransitions": 6100,
+      "eding": 6101,
+      "Ġprograms": 6102,
+      "Ġmobile": 6103,
+      "Ġimmediately": 6104,
+      "ectivity": 6105,
+      "ĠTherm": 6106,
+      "ogenetic": 6107,
+      "Ġseven": 6108,
+      "Ġemph": 6109,
+      "GE": 6110,
+      "neum": 6111,
+      "Ġfusion": 6112,
+      "limits": 6113,
+      "Ġcalcium": 6114,
+      "raf": 6115,
+      "minus": 6116,
+      "Ġtrap": 6117,
+      "Ġspecimens": 6118,
+      "ancing": 6119,
+      "ĠMarch": 6120,
+      "Ġten": 6121,
+      "Ġfamilies": 6122,
+      "ĠHD": 6123,
+      "isons": 6124,
+      "Ġpreparation": 6125,
+      "hold": 6126,
+      "ether": 6127,
+      "ĠVol": 6128,
+      "ĠDise": 6129,
+      "Ġrunning": 6130,
+      "Ġqualit": 6131,
+      "Ġeffectively": 6132,
+      "fficiently": 6133,
+      "BI": 6134,
+      "Ġdenoted": 6135,
+      "ĠEquation": 6136,
+      "Ġdemand": 6137,
+      "itory": 6138,
+      "aching": 6139,
+      "Ġsodium": 6140,
+      "Ġreproduc": 6141,
+      "cho": 6142,
+      "Ġbil": 6143,
+      "Pi": 6144,
+      "umb": 6145,
+      "Ġreconstruction": 6146,
+      "forward": 6147,
+      "One": 6148,
+      "Ġconversion": 6149,
+      "Ġformulation": 6150,
+      "Ġnearly": 6151,
+      "ĠLag": 6152,
+      "Str": 6153,
+      "terior": 6154,
+      "Ġoperating": 6155,
+      "andom": 6156,
+      "Ġmoving": 6157,
+      "ĠReview": 6158,
+      "////": 6159,
+      "nai": 6160,
+      "pp": 6161,
+      "otide": 6162,
+      "label": 6163,
+      "ococc": 6164,
+      "Ġnever": 6165,
+      "aker": 6166,
+      "Ġdigital": 6167,
+      "Bl": 6168,
+      "Un": 6169,
+      "Ġmember": 6170,
+      "sel": 6171,
+      "Ġpotenti": 6172,
+      "Ġcopy": 6173,
+      "Ġelectrons": 6174,
+      "chlor": 6175,
+      "annel": 6176,
+      "ylind": 6177,
+      "Ġmis": 6178,
+      "ĠSet": 6179,
+      "Ġnutri": 6180,
+      "Ġdescribes": 6181,
+      "Ġassumptions": 6182,
+      "Ġvirtual": 6183,
+      "Ġcoordinate": 6184,
+      "Ġvor": 6185,
+      "ĠArab": 6186,
+      "ĠImp": 6187,
+      "Ġdeposition": 6188,
+      "Ġinstit": 6189,
+      "Ġrepresentative": 6190,
+      "everal": 6191,
+      "Ġmillion": 6192,
+      "ĠMA": 6193,
+      "Ġmales": 6194,
+      "Ġcrucial": 6195,
+      "Ġcold": 6196,
+      "Ġloading": 6197,
+      "Ġtranslation": 6198,
+      "Ġstead": 6199,
+      "rays": 6200,
+      "Ġchallenge": 6201,
+      "activity": 6202,
+      "idal": 6203,
+      "uff": 6204,
+      "Ġseem": 6205,
+      "Ġnational": 6206,
+      "Ġfa": 6207,
+      "Ġminor": 6208,
+      "Ġundergo": 6209,
+      "cr": 6210,
+      "Ġcapt": 6211,
+      "ele": 6212,
+      "uple": 6213,
+      "ĠMg": 6214,
+      "lege": 6215,
+      "GR": 6216,
+      "Ġrig": 6217,
+      "Ġarri": 6218,
+      "Ġdetector": 6219,
+      "Ġstrict": 6220,
+      "Ġadhes": 6221,
+      "Ġsea": 6222,
+      "theless": 6223,
+      "Ġsleep": 6224,
+      "ĠCommun": 6225,
+      "Ġantioxid": 6226,
+      "Ġmarker": 6227,
+      "Ġflows": 6228,
+      "ancre": 6229,
+      "ĠJanuary": 6230,
+      "input": 6231,
+      "UP": 6232,
+      "Ġstored": 6233,
+      "ading": 6234,
+      "itively": 6235,
+      "Ġslope": 6236,
+      "Ġshell": 6237,
+      "Ġelevated": 6238,
+      "ilk": 6239,
+      "Ġfrequently": 6240,
+      "Ġball": 6241,
+      "urban": 6242,
+      "Ġml": 6243,
+      "usive": 6244,
+      "ĠAnt": 6245,
+      "amino": 6246,
+      "Sim": 6247,
+      "Ġphysiological": 6248,
+      "regulation": 6249,
+      "esity": 6250,
+      "Ġexplan": 6251,
+      "Ġaden": 6252,
+      "reme": 6253,
+      "Ġdiffer": 6254,
+      "Ġmodification": 6255,
+      "Ġirradi": 6256,
+      "He": 6257,
+      "acial": 6258,
+      "Ġsuppress": 6259,
+      "quis": 6260,
+      "Ġdry": 6261,
+      "erated": 6262,
+      "Ġprojection": 6263,
+      "Ġpool": 6264,
+      "plete": 6265,
+      "Ġdirections": 6266,
+      "Ġchanged": 6267,
+      "ĠIts": 6268,
+      "Ġster": 6269,
+      "Ġradial": 6270,
+      "Ġgr": 6271,
+      "Ġperiodic": 6272,
+      "Ġbin": 6273,
+      "Ġpip": 6274,
+      "men": 6275,
+      "then": 6276,
+      "pc": 6277,
+      "amily": 6278,
+      "ĠDM": 6279,
+      "Ġsediment": 6280,
+      "mi": 6281,
+      "Ġclosely": 6282,
+      "Ġrepair": 6283,
+      "Ġrespiratory": 6284,
+      "Ġhorm": 6285,
+      "Ans": 6286,
+      "dr": 6287,
+      "ls": 6288,
+      "Ġhomogeneous": 6289,
+      "etric": 6290,
+      "DS": 6291,
+      "Ġresidues": 6292,
+      "ĠValue": 6293,
+      "Fs": 6294,
+      "Ġwhy": 6295,
+      "Sp": 6296,
+      "Ġca": 6297,
+      "Ġnarrow": 6298,
+      "gent": 6299,
+      "Ġbr": 6300,
+      "Ġquasi": 6301,
+      "Ġpict": 6302,
+      "mo": 6303,
+      "Ġatom": 6304,
+      "Ġhabit": 6305,
+      "Ġlimitations": 6306,
+      "conduc": 6307,
+      "Ġshock": 6308,
+      "ceptor": 6309,
+      "ĠDetection": 6310,
+      "Sh": 6311,
+      "ube": 6312,
+      "Ġellip": 6313,
+      "UR": 6314,
+      "Ġstaining": 6315,
+      "Ġrapidly": 6316,
+      "ĠBur": 6317,
+      "ĠBro": 6318,
+      "Ġuptake": 6319,
+      "Ġchallenges": 6320,
+      "SN": 6321,
+      "Ġanis": 6322,
+      "Ġbounds": 6323,
+      "step": 6324,
+      "omeric": 6325,
+      "tention": 6326,
+      "ĠEvaluation": 6327,
+      "Ġrecommend": 6328,
+      "Me": 6329,
+      "Ġmoderate": 6330,
+      "elled": 6331,
+      "Ġtit": 6332,
+      "ĠYang": 6333,
+      "Ġpharmac": 6334,
+      "inflammatory": 6335,
+      "ĠJune": 6336,
+      "Ġsensors": 6337,
+      "aired": 6338,
+      "Ġapproximate": 6339,
+      "SV": 6340,
+      "Ġbund": 6341,
+      "rc": 6342,
+      "oman": 6343,
+      "Ġvisible": 6344,
+      "Ġmeasuring": 6345,
+      "ogonal": 6346,
+      "ĠFourier": 6347,
+      "Ġtheories": 6348,
+      "Ġprofession": 6349,
+      "tained": 6350,
+      "atas": 6351,
+      "ĠInterest": 6352,
+      "param": 6353,
+      "ĠStructure": 6354,
+      "Ġliving": 6355,
+      "Data": 6356,
+      "ĠSM": 6357,
+      "Ġnet": 6358,
+      "Ġsimultaneously": 6359,
+      "continu": 6360,
+      "Ġshor": 6361,
+      "########": 6362,
+      "Ġdecreasing": 6363,
+      "Ġreferred": 6364,
+      "gg": 6365,
+      "Thus": 6366,
+      "Ġdro": 6367,
+      "pril": 6368,
+      "ĠPers": 6369,
+      "Ġencoding": 6370,
+      "Ġarc": 6371,
+      "Ġregulatory": 6372,
+      "Ġtrained": 6373,
+      "cepts": 6374,
+      "Ġrout": 6375,
+      "lys": 6376,
+      "Par": 6377,
+      "ĠUl": 6378,
+      "ĠGraph": 6379,
+      "âĪĤ": 6380,
+      "Ġirre": 6381,
+      "oidal": 6382,
+      "Ġexceed": 6383,
+      "Ġmostly": 6384,
+      "ĠPat": 6385,
+      "aternal": 6386,
+      "Ġer": 6387,
+      "Ġcoverage": 6388,
+      "FS": 6389,
+      "ĠRot": 6390,
+      "Ġclassified": 6391,
+      "Ġexcitation": 6392,
+      "Ġconductivity": 6393,
+      "Ġcommercial": 6394,
+      "ĠDel": 6395,
+      "ĠPolar": 6396,
+      "HR": 6397,
+      "Ġtraffic": 6398,
+      "zing": 6399,
+      "Ġsettings": 6400,
+      "Ġinclusion": 6401,
+      "Answer": 6402,
+      "Ġvit": 6403,
+      "vitational": 6404,
+      "Ġbind": 6405,
+      "Ġoc": 6406,
+      "ĠWestern": 6407,
+      "Ġprosp": 6408,
+      "Ġnorth": 6409,
+      "itch": 6410,
+      "ĠRiver": 6411,
+      "Ġvehicle": 6412,
+      "Ġlikelihood": 6413,
+      "LD": 6414,
+      "Ġinsp": 6415,
+      "âĪĨ": 6416,
+      "Ġleuk": 6417,
+      "ĠBre": 6418,
+      "Ġsynthetic": 6419,
+      "ĠGermany": 6420,
+      "ĠTheir": 6421,
+      "target": 6422,
+      "ĠEnglish": 6423,
+      "Ġnotation": 6424,
+      "ĠATP": 6425,
+      "ĠModels": 6426,
+      "Ġabnormal": 6427,
+      "ĠConclusions": 6428,
+      "Ġoccurrence": 6429,
+      "Ġmicrobi": 6430,
+      "ĠWar": 6431,
+      "tember": 6432,
+      "Ġlocally": 6433,
+      "born": 6434,
+      "Ġbarrier": 6435,
+      "Ġexpressions": 6436,
+      "oval": 6437,
+      "Ġflav": 6438,
+      "emble": 6439,
+      "Ġdynamical": 6440,
+      "Ġphoton": 6441,
+      "apped": 6442,
+      "Ġglut": 6443,
+      "Ġkinetic": 6444,
+      "Ġalcohol": 6445,
+      "Ġtransplant": 6446,
+      "LP": 6447,
+      "Ġdefault": 6448,
+      "Ġopportun": 6449,
+      "args": 6450,
+      "ĠDav": 6451,
+      "Ġfront": 6452,
+      "hom": 6453,
+      "Ġways": 6454,
+      "ĠAssociation": 6455,
+      "Ġkidney": 6456,
+      "Ġproportional": 6457,
+      "When": 6458,
+      "Ġepithelial": 6459,
+      "Ġfresh": 6460,
+      "Ġrecall": 6461,
+      "Ġenzymes": 6462,
+      "br": 6463,
+      "Ġmalign": 6464,
+      "textrm": 6465,
+      "ĠUse": 6466,
+      "Now": 6467,
+      "ĠLie": 6468,
+      "Ġimpair": 6469,
+      "Ġguarant": 6470,
+      "Ġinver": 6471,
+      "Ġtranscript": 6472,
+      "Ġsustain": 6473,
+      "Ġactually": 6474,
+      "alities": 6475,
+      "ĠMic": 6476,
+      "ĠIC": 6477,
+      "ĠMeasure": 6478,
+      "Ġ": 6479,
+      "Ġdensities": 6480,
+      "Ġgalaxy": 6481,
+      "Ġsufficiently": 6482,
+      "Ġorbit": 6483,
+      "ford": 6484,
+      "Ġpartially": 6485,
+      "ĠPy": 6486,
+      "Ġreverse": 6487,
+      "Ġsurve": 6488,
+      "ĠWork": 6489,
+      "Ġask": 6490,
+      "However": 6491,
+      "Ġsitu": 6492,
+      "Ġvacuum": 6493,
+      "tober": 6494,
+      "Ġspac": 6495,
+      "anth": 6496,
+      "Or": 6497,
+      "ags": 6498,
+      "Ġbig": 6499,
+      "herical": 6500,
+      "erge": 6501,
+      "ellite": 6502,
+      "Ġinvolves": 6503,
+      "ĠVis": 6504,
+      "Ġsummary": 6505,
+      "ĠSupplementary": 6506,
+      "ĠColl": 6507,
+      "Ġadjacent": 6508,
+      "ontaneous": 6509,
+      "abs": 6510,
+      "Ġresearchers": 6511,
+      "ka": 6512,
+      "Ġintern": 6513,
+      "Ġmonth": 6514,
+      "ĠNeural": 6515,
+      "apor": 6516,
+      "ĠNan": 6517,
+      "Ġstri": 6518,
+      "EE": 6519,
+      "Ġconsisting": 6520,
+      "Ġupdate": 6521,
+      "Ġphoto": 6522,
+      "Val": 6523,
+      "sens": 6524,
+      "Ġveget": 6525,
+      "BR": 6526,
+      "Ġcoinc": 6527,
+      "ĠJuly": 6528,
+      "tility": 6529,
+      "ĠExpression": 6530,
+      "Ġtopology": 6531,
+      "Ġgrowing": 6532,
+      "aptic": 6533,
+      "uced": 6534,
+      "Ġperipheral": 6535,
+      "enes": 6536,
+      "Ġplots": 6537,
+      "Ġexplo": 6538,
+      "Ġwor": 6539,
+      "ba": 6540,
+      "atitis": 6541,
+      "ief": 6542,
+      "wave": 6543,
+      "Ġprotection": 6544,
+      "Ġdefects": 6545,
+      "Ġadsorption": 6546,
+      "Ġshared": 6547,
+      "Ġstellar": 6548,
+      "ĠBa": 6549,
+      "ĠEnergy": 6550,
+      "queous": 6551,
+      "ĠAugust": 6552,
+      "Ġlys": 6553,
+      "Ġplus": 6554,
+      "irel": 6555,
+      "ĠGP": 6556,
+      "ĠNeu": 6557,
+      "dist": 6558,
+      "gers": 6559,
+      "ifer": 6560,
+      "isp": 6561,
+      "Ġstrat": 6562,
+      "ione": 6563,
+      "ĠMaterials": 6564,
+      "Ġln": 6565,
+      "Ġpulmonary": 6566,
+      "ened": 6567,
+      "plan": 6568,
+      "Mod": 6569,
+      "Ġorganization": 6570,
+      "Ġrelaxation": 6571,
+      "Ġcortex": 6572,
+      "Ġmodulation": 6573,
+      "ogl": 6574,
+      "shift": 6575,
+      "Ġsecurity": 6576,
+      "Ġfatty": 6577,
+      "Ġms": 6578,
+      "local": 6579,
+      "ergic": 6580,
+      "Ġinterference": 6581,
+      "inson": 6582,
+      "cf": 6583,
+      "Ġreasons": 6584,
+      "pred": 6585,
+      "Ġinterventions": 6586,
+      "Ġjo": 6587,
+      "ĠID": 6588,
+      "ĠArea": 6589,
+      "ĠHa": 6590,
+      "uits": 6591,
+      "output": 6592,
+      "Le": 6593,
+      "ycl": 6594,
+      "inted": 6595,
+      "Ġnano": 6596,
+      "NC": 6597,
+      "ĠCap": 6598,
+      "Ġchanging": 6599,
+      "Ġcust": 6600,
+      "Ġappeared": 6601,
+      "Ġgrown": 6602,
+      "ĠUK": 6603,
+      "Ġradical": 6604,
+      "ĠPot": 6605,
+      "ĠProgram": 6606,
+      "ĠSR": 6607,
+      "Ġshap": 6608,
+      "oscop": 6609,
+      "ĠChang": 6610,
+      "Ġquantity": 6611,
+      "ĠTaxon": 6612,
+      "idation": 6613,
+      "Ġadding": 6614,
+      "ĠLee": 6615,
+      "Ġamounts": 6616,
+      "Ġdespite": 6617,
+      "Ġremained": 6618,
+      "Ġscenarios": 6619,
+      "lets": 6620,
+      "oming": 6621,
+      "Ġcurvature": 6622,
+      "Ġdimensional": 6623,
+      "Ġpromising": 6624,
+      "ĠFil": 6625,
+      "string": 6626,
+      "Ġattributed": 6627,
+      "ymer": 6628,
+      "Ġneighb": 6629,
+      "Ġinputs": 6630,
+      "Ġmagnet": 6631,
+      "Ġtrees": 6632,
+      "Ġenter": 6633,
+      "ruit": 6634,
+      "stable": 6635,
+      "toplas": 6636,
+      "Ġmessage": 6637,
+      "rophic": 6638,
+      "Ġisolates": 6639,
+      "tz": 6640,
+      "Ġdisplayed": 6641,
+      "HA": 6642,
+      "ocl": 6643,
+      "Ġderive": 6644,
+      "Ġsynchron": 6645,
+      "QU": 6646,
+      "Ãŀ": 6647,
+      "Ġexamination": 6648,
+      "Ġdeb": 6649,
+      "Ġdefin": 6650,
+      "Ġfault": 6651,
+      "Ġsteady": 6652,
+      "Ġphenotype": 6653,
+      "Ġperspective": 6654,
+      "Ġstatement": 6655,
+      "df": 6656,
+      "void": 6657,
+      "Ġpromote": 6658,
+      "illary": 6659,
+      "ĠEth": 6660,
+      "Ġwalk": 6661,
+      "Ġrepresenting": 6662,
+      "Ġgenomic": 6663,
+      "ĠGr": 6664,
+      "shape": 6665,
+      "ĠPet": 6666,
+      "ĠLocal": 6667,
+      "plicity": 6668,
+      "ĠProblem": 6669,
+      "GS": 6670,
+      "Ġcompleted": 6671,
+      "inking": 6672,
+      "Ġreads": 6673,
+      "Ġinde": 6674,
+      "ceived": 6675,
+      "ĠPL": 6676,
+      "ĠMean": 6677,
+      "ĠSchool": 6678,
+      "Ġbiomark": 6679,
+      "ireless": 6680,
+      "cut": 6681,
+      "osing": 6682,
+      "nel": 6683,
+      "ĠApril": 6684,
+      "ĠBal": 6685,
+      "Ġadopted": 6686,
+      "Ġcomplications": 6687,
+      "Ġassembly": 6688,
+      "fort": 6689,
+      "har": 6690,
+      "Ġadoles": 6691,
+      "Ġanswer": 6692,
+      "Ġcommunities": 6693,
+      "ĠInstitute": 6694,
+      "Ġvariant": 6695,
+      "Finally": 6696,
+      "mitte": 6697,
+      "Ġrestricted": 6698,
+      "Ġmanip": 6699,
+      "aters": 6700,
+      "EX": 6701,
+      "Ġdust": 6702,
+      "Ġsupply": 6703,
+      "Ġperme": 6704,
+      "Ġreliable": 6705,
+      "ĠResp": 6706,
+      "Ġsubt": 6707,
+      "oks": 6708,
+      "Ġpoll": 6709,
+      "Ġcanc": 6710,
+      "ĠUnit": 6711,
+      "Ġendothelial": 6712,
+      "dy": 6713,
+      "ĠBlack": 6714,
+      "Ġempirical": 6715,
+      "Ġport": 6716,
+      "opy": 6717,
+      "Ġinitially": 6718,
+      "Ġcondens": 6719,
+      "Ġeye": 6720,
+      "Ġlisted": 6721,
+      "urrence": 6722,
+      "Ġreplaced": 6723,
+      "Ġselective": 6724,
+      "Ġdistances": 6725,
+      "Ġparas": 6726,
+      "ĠPost": 6727,
+      "ĠSeptember": 6728,
+      "Ġmissing": 6729,
+      "verex": 6730,
+      "Er": 6731,
+      "Ġthought": 6732,
+      "thal": 6733,
+      "Ġchromat": 6734,
+      "Ġbenefit": 6735,
+      "rames": 6736,
+      "ĠSuppose": 6737,
+      "Ġsubs": 6738,
+      "Ġangi": 6739,
+      "ori": 6740,
+      "Ġreplic": 6741,
+      "Ġschemes": 6742,
+      "pre": 6743,
+      "plane": 6744,
+      "Ġsouth": 6745,
+      "ager": 6746,
+      "Ġbeginning": 6747,
+      "vents": 6748,
+      "onent": 6749,
+      "iples": 6750,
+      "ĠHer": 6751,
+      "Ġspectrom": 6752,
+      "Ġdense": 6753,
+      "Ġtook": 6754,
+      "iverse": 6755,
+      "Ġdisturb": 6756,
+      "pass": 6757,
+      "Ġillustrated": 6758,
+      "Ġreveals": 6759,
+      "ama": 6760,
+      "Ġreflec": 6761,
+      "Ġallowing": 6762,
+      "Ġexponential": 6763,
+      "oustic": 6764,
+      "subseteq": 6765,
+      "Ġsn": 6766,
+      "Ġurban": 6767,
+      "Ġextend": 6768,
+      "Ġassays": 6769,
+      "rice": 6770,
+      "CoV": 6771,
+      "quisition": 6772,
+      "rine": 6773,
+      "ĠIntegr": 6774,
+      "fil": 6775,
+      "VD": 6776,
+      "Ġfibro": 6777,
+      "Ġcompens": 6778,
+      "ĠImpro": 6779,
+      "ĠĠĠĠĠĠĠĠĠĠ": 6780,
+      "ĠGR": 6781,
+      "ÏĪ": 6782,
+      "Ġbasal": 6783,
+      "Ġolig": 6784,
+      "HT": 6785,
+      "Ġvess": 6786,
+      "uzzy": 6787,
+      "Ġpossibly": 6788,
+      "Ġtolerance": 6789,
+      "Theta": 6790,
+      "Ġviol": 6791,
+      "uclear": 6792,
+      "ĠLim": 6793,
+      "gel": 6794,
+      "Ġmetrics": 6795,
+      "ĠMus": 6796,
+      "amination": 6797,
+      "Ġelectrode": 6798,
+      "Ġpersonal": 6799,
+      "Ġcooling": 6800,
+      "Ġacquired": 6801,
+      "ĠFunction": 6802,
+      "ows": 6803,
+      "olester": 6804,
+      "DP": 6805,
+      "Ġreliability": 6806,
+      "Ġmuc": 6807,
+      "ĠOctober": 6808,
+      "Ġgold": 6809,
+      "ca": 6810,
+      "Ġcul": 6811,
+      "fit": 6812,
+      "Ġlem": 6813,
+      "Ġexcit": 6814,
+      "Ġnucleus": 6815,
+      "iation": 6816,
+      "Ġpregnancy": 6817,
+      "Ġsynthesized": 6818,
+      "hemistry": 6819,
+      "Ġmembranes": 6820,
+      "vert": 6821,
+      "ĠKim": 6822,
+      "tenance": 6823,
+      "Ġquantities": 6824,
+      "Ġeconomic": 6825,
+      "Ġbenefits": 6826,
+      "Ġcylind": 6827,
+      "pler": 6828,
+      "ĠLarge": 6829,
+      "Ġengineering": 6830,
+      "ĠEp": 6831,
+      "Ġcoating": 6832,
+      "ativ": 6833,
+      "Ġconduct": 6834,
+      "Ġabsorb": 6835,
+      "ĠDecember": 6836,
+      "Ġopposite": 6837,
+      "ĠGlobal": 6838,
+      "Ġlif": 6839,
+      "ĠDue": 6840,
+      "Ġintake": 6841,
+      "odynamic": 6842,
+      "TM": 6843,
+      "Ġfed": 6844,
+      "Ġspecified": 6845,
+      "Ġgeometric": 6846,
+      "Ġrespective": 6847,
+      "Ġbirth": 6848,
+      "ĠCompound": 6849,
+      "Ġstarted": 6850,
+      "Ġmother": 6851,
+      "arr": 6852,
+      "Ġprimarily": 6853,
+      "Ġparen": 6854,
+      "Ġtube": 6855,
+      "Ġinters": 6856,
+      "Ġgraphene": 6857,
+      "itial": 6858,
+      "ously": 6859,
+      "Ġcardiovascular": 6860,
+      "ĠeV": 6861,
+      "Ġheating": 6862,
+      "Ġmathematical": 6863,
+      "Ġindependently": 6864,
+      "BA": 6865,
+      "Ġaffects": 6866,
+      "umor": 6867,
+      "ĠMP": 6868,
+      "ĠDem": 6869,
+      "ĠWest": 6870,
+      "ĠDom": 6871,
+      "itter": 6872,
+      "Ġdisrup": 6873,
+      "oped": 6874,
+      "Ġphenomenon": 6875,
+      "Ġlumin": 6876,
+      "Ac": 6877,
+      "Ġprefer": 6878,
+      "omers": 6879,
+      "Ġgender": 6880,
+      "ĠGL": 6881,
+      "FC": 6882,
+      "Ġindeed": 6883,
+      "Ġrational": 6884,
+      "ĠSC": 6885,
+      "Ġprincipal": 6886,
+      "Ġperfect": 6887,
+      "Ġintroduction": 6888,
+      "tes": 6889,
+      "Ġpiec": 6890,
+      "Ġcity": 6891,
+      "Ġpopular": 6892,
+      "Ġcoding": 6893,
+      "cler": 6894,
+      "ague": 6895,
+      "ĠHR": 6896,
+      "Ġtracking": 6897,
+      "ker": 6898,
+      "Ġphosphorylation": 6899,
+      "Ġpaths": 6900,
+      "Ġsolving": 6901,
+      "Ġdy": 6902,
+      "Ġplayed": 6903,
+      "Ġprecise": 6904,
+      "ĠSl": 6905,
+      "ĠSem": 6906,
+      "Ġgenerating": 6907,
+      "ĠSun": 6908,
+      "Ġcriterion": 6909,
+      "Ġbranch": 6910,
+      "Ġζ": 6911,
+      "tish": 6912,
+      "Se": 6913,
+      "Ġantigen": 6914,
+      "Ġcalibration": 6915,
+      "Es": 6916,
+      "ĠItal": 6917,
+      "Ġmassive": 6918,
+      "En": 6919,
+      "No": 6920,
+      "YP": 6921,
+      "ya": 6922,
+      "Ġsatisfying": 6923,
+      "Ġquick": 6924,
+      "HO": 6925,
+      "Ġbehaviors": 6926,
+      "icrobial": 6927,
+      "Ġamb": 6928,
+      "Ġproton": 6929,
+      "SL": 6930,
+      "Ġusual": 6931,
+      "rows": 6932,
+      "ench": 6933,
+      "UC": 6934,
+      "Ġweighted": 6935,
+      "Ġrecords": 6936,
+      "ĠAC": 6937,
+      "GT": 6938,
+      "inn": 6939,
+      "Ġeq": 6940,
+      "ĠWil": 6941,
+      "yroid": 6942,
+      "Ġsetup": 6943,
+      "IA": 6944,
+      "press": 6945,
+      "isely": 6946,
+      "Ġentry": 6947,
+      "%%": 6948,
+      "ĠSil": 6949,
+      "east": 6950,
+      "ĠEvolution": 6951,
+      "ĠRandom": 6952,
+      "Ġcavity": 6953,
+      "Ġnamed": 6954,
+      "knowled": 6955,
+      "mber": 6956,
+      "uestion": 6957,
+      "ĠâĪ©": 6958,
+      "gi": 6959,
+      "Ġdetermining": 6960,
+      "tin": 6961,
+      "Ġgenus": 6962,
+      "Ġtoxicity": 6963,
+      "ocyt": 6964,
+      "Ġperturbation": 6965,
+      "rought": 6966,
+      "ĠBri": 6967,
+      "Ġcarb": 6968,
+      "ĠGra": 6969,
+      "ĠFlu": 6970,
+      "uns": 6971,
+      "Ġdriven": 6972,
+      "Ġbatch": 6973,
+      "rif": 6974,
+      "Pl": 6975,
+      "Ġdisplacement": 6976,
+      "ĠCL": 6977,
+      "Ġdepic": 6978,
+      "Ġpredictive": 6979,
+      "Int": 6980,
+      "hydroxy": 6981,
+      "tid": 6982,
+      "dri": 6983,
+      "Ġpancre": 6984,
+      "Ġdiagonal": 6985,
+      "Ġseverity": 6986,
+      "Ġlongitudinal": 6987,
+      "ĠED": 6988,
+      "atible": 6989,
+      "dir": 6990,
+      "ĠAnother": 6991,
+      "ĠHel": 6992,
+      "van": 6993,
+      "Ġpneum": 6994,
+      "Ġspecificity": 6995,
+      "squ": 6996,
+      "Ġign": 6997,
+      "Ġbed": 6998,
+      "ĠWT": 6999,
+      "awa": 7000,
+      "ester": 7001,
+      "Ġkg": 7002,
+      "Ġcompression": 7003,
+      "evertheless": 7004,
+      "Ġmask": 7005,
+      "-----------": 7006,
+      "Ġtens": 7007,
+      "rowth": 7008,
+      "ĠGo": 7009,
+      "Ġfaster": 7010,
+      "Ġcanonical": 7011,
+      "Ġdetermin": 7012,
+      "ustrial": 7013,
+      "ĠEarth": 7014,
+      "while": 7015,
+      "ournal": 7016,
+      "Ġcountry": 7017,
+      "Ġferm": 7018,
+      "rist": 7019,
+      "Ġproxim": 7020,
+      "Ġmicrobial": 7021,
+      "Ġextensive": 7022,
+      "Ġcham": 7023,
+      "Ġ§": 7024,
+      "such": 7025,
+      "went": 7026,
+      "Ġlar": 7027,
+      "Using": 7028,
+      "ĠPM": 7029,
+      "Ġoffset": 7030,
+      "ĠPI": 7031,
+      "ĠBayesian": 7032,
+      "HS": 7033,
+      "ĠAfrica": 7034,
+      "Ġsusceptibility": 7035,
+      "ĠâĬĤ": 7036,
+      "ococcus": 7037,
+      "ĠDir": 7038,
+      "Ġbos": 7039,
+      "Ġdysfunction": 7040,
+      "ovember": 7041,
+      "Ġunderst": 7042,
+      "Ġlargely": 7043,
+      "ĠCM": 7044,
+      "Ġmaintained": 7045,
+      "Ġpossess": 7046,
+      "Ġexcluded": 7047,
+      "ensis": 7048,
+      "ĠDC": 7049,
+      "opsis": 7050,
+      "Ġtorch": 7051,
+      "idine": 7052,
+      "Ġforest": 7053,
+      "ĠExact": 7054,
+      "ĠStudies": 7055,
+      "iffiff": 7056,
+      "ĠCam": 7057,
+      "angular": 7058,
+      "Ġremove": 7059,
+      "oir": 7060,
+      "ava": 7061,
+      "ida": 7062,
+      "Ġmant": 7063,
+      "Log": 7064,
+      "Ġranging": 7065,
+      "rog": 7066,
+      "Ġchains": 7067,
+      "ĠÇ«": 7068,
+      "ĠCase": 7069,
+      "ĠAP": 7070,
+      "points": 7071,
+      "Ġtargeting": 7072,
+      "Ġscience": 7073,
+      "Ġepis": 7074,
+      "ĠSoci": 7075,
+      "Ġphysic": 7076,
+      "Ġpromoter": 7077,
+      "ĠEarly": 7078,
+      "estic": 7079,
+      "tives": 7080,
+      "Ġassuming": 7081,
+      "ĠMi": 7082,
+      "Ġlemma": 7083,
+      "Ġconfigurations": 7084,
+      "alia": 7085,
+      "Ġpay": 7086,
+      "rino": 7087,
+      "eb": 7088,
+      "Ġvaried": 7089,
+      "ounted": 7090,
+      "Ġinterview": 7091,
+      "ĠGeV": 7092,
+      "OM": 7093,
+      "ognition": 7094,
+      "Ġenhancement": 7095,
+      "ĠMach": 7096,
+      "plies": 7097,
+      "Ob": 7098,
+      "setminus": 7099,
+      "Ġintrinsic": 7100,
+      "Ġcomparisons": 7101,
+      "bold": 7102,
+      "xiety": 7103,
+      "Ġstroke": 7104,
+      "GB": 7105,
+      "ancial": 7106,
+      "stead": 7107,
+      "Ġrock": 7108,
+      "thon": 7109,
+      "ĠCurrent": 7110,
+      "cat": 7111,
+      "Ġguidelines": 7112,
+      "cycl": 7113,
+      "Ġintracellular": 7114,
+      "oney": 7115,
+      "ko": 7116,
+      "Ġdirected": 7117,
+      "ripts": 7118,
+      "Ġtravel": 7119,
+      "Ġlens": 7120,
+      "idi": 7121,
+      "ĠAssess": 7122,
+      "Ġdx": 7123,
+      "ĠPos": 7124,
+      "Ġmethodology": 7125,
+      "Ġpredom": 7126,
+      "defined": 7127,
+      "ĠPop": 7128,
+      "Ġgovernment": 7129,
+      "ellig": 7130,
+      "phyl": 7131,
+      "oli": 7132,
+      "ropical": 7133,
+      "Ġembedded": 7134,
+      "edom": 7135,
+      "cribed": 7136,
+      "ĠDisease": 7137,
+      "Ġmediated": 7138,
+      "Ġcircular": 7139,
+      "ĠTopological": 7140,
+      "Ġearth": 7141,
+      "ritis": 7142,
+      "gal": 7143,
+      "mass": 7144,
+      "Ġcomprehensive": 7145,
+      "ĠAir": 7146,
+      "Ġnerve": 7147,
+      "Ġimplant": 7148,
+      "Ġextremely": 7149,
+      "ĠSE": 7150,
+      "Ġmarket": 7151,
+      "Ġconserved": 7152,
+      "embrane": 7153,
+      "Ġschedul": 7154,
+      "Ġruns": 7155,
+      "Ph": 7156,
+      "Ġtechnical": 7157,
+      "TL": 7158,
+      "Ġregional": 7159,
+      "Ġgerm": 7160,
+      "ĠProt": 7161,
+      "Ġbright": 7162,
+      "Ġartery": 7163,
+      "Ġmacrophages": 7164,
+      "mittee": 7165,
+      "ĠSingle": 7166,
+      "Ġcome": 7167,
+      "wa": 7168,
+      "acchar": 7169,
+      "plet": 7170,
+      "Ġsensing": 7171,
+      "rosp": 7172,
+      "atom": 7173,
+      "Ġcompr": 7174,
+      "ĠLu": 7175,
+      "Ġavailability": 7176,
+      "prot": 7177,
+      "Ġfitting": 7178,
+      "selves": 7179,
+      "ĠPrim": 7180,
+      "rew": 7181,
+      "Ġwaste": 7182,
+      "ĠKing": 7183,
+      "pot": 7184,
+      "Ġinstrument": 7185,
+      "ĠYork": 7186,
+      "AF": 7187,
+      "antial": 7188,
+      "standing": 7189,
+      "Ġplanning": 7190,
+      "uster": 7191,
+      "ĠâĨ": 7192,
+      "NT": 7193,
+      "icular": 7194,
+      "Ġmelan": 7195,
+      "Ġexcell": 7196,
+      "iller": 7197,
+      "ĠLD": 7198,
+      "info": 7199,
+      "Ġshare": 7200,
+      "vas": 7201,
+      "Ġlum": 7202,
+      "Ġaqueous": 7203,
+      "Ġquery": 7204,
+      "Ġmag": 7205,
+      "ulture": 7206,
+      "ĠBer": 7207,
+      "Ġoffer": 7208,
+      "ĠNMR": 7209,
+      "aceae": 7210,
+      "Ġmodern": 7211,
+      "Ġcircum": 7212,
+      "Ġcultures": 7213,
+      "Ġdog": 7214,
+      "Ġcir": 7215,
+      "Ġpoli": 7216,
+      "Ġchemotherapy": 7217,
+      "Ġplates": 7218,
+      "Ġrestriction": 7219,
+      "stack": 7220,
+      "ĠFlow": 7221,
+      "ĠBu": 7222,
+      "ĠCenter": 7223,
+      "Ġproceed": 7224,
+      "timicrobial": 7225,
+      "she": 7226,
+      "Ġthereby": 7227,
+      "Ġknock": 7228,
+      "Ġdiverse": 7229,
+      "ustry": 7230,
+      "Ġstated": 7231,
+      "ĠHol": 7232,
+      "More": 7233,
+      "Ġconservation": 7234,
+      "Ġprevention": 7235,
+      "norm": 7236,
+      "Ġpal": 7237,
+      "ĠCalc": 7238,
+      "Ġclean": 7239,
+      "ĠPlas": 7240,
+      "```": 7241,
+      "perp": 7242,
+      "prod": 7243,
+      "Ġâī¡": 7244,
+      "porter": 7245,
+      "Ġtransient": 7246,
+      "asp": 7247,
+      "Ġtargeted": 7248,
+      "ĠPri": 7249,
+      "Supplementary": 7250,
+      "ĠTreatment": 7251,
+      "zen": 7252,
+      "ĠMart": 7253,
+      "ĠFerm": 7254,
+      "uscript": 7255,
+      "ĠSynthesis": 7256,
+      "Ġcombinations": 7257,
+      "ULL": 7258,
+      "Ġweb": 7259,
+      "Ġthrom": 7260,
+      "Ġexplicitly": 7261,
+      "anks": 7262,
+      "Ġadaptation": 7263,
+      "ĠSequence": 7264,
+      "Ġacts": 7265,
+      "Ġranges": 7266,
+      "fs": 7267,
+      "bru": 7268,
+      "Ġsystemic": 7269,
+      "Ġsteel": 7270,
+      "Ġprivate": 7271,
+      "Ġobesity": 7272,
+      "ĠPart": 7273,
+      "mented": 7274,
+      "break": 7275,
+      "ERT": 7276,
+      "Ġfibers": 7277,
+      "Ġiso": 7278,
+      "Ġtransverse": 7279,
+      "CTION": 7280,
+      "ĠRi": 7281,
+      "itin": 7282,
+      "ĠRepresent": 7283,
+      "ophys": 7284,
+      "Ġcoast": 7285,
+      "Ġalignment": 7286,
+      "ACT": 7287,
+      "esides": 7288,
+      "open": 7289,
+      "gly": 7290,
+      "Ġsalt": 7291,
+      "unced": 7292,
+      "iaz": 7293,
+      "Ġcosm": 7294,
+      "Ġangles": 7295,
+      "ĠâĢł": 7296,
+      "ĠIdentification": 7297,
+      "hex": 7298,
+      "ĠHall": 7299,
+      "Ġhepat": 7300,
+      "Ġsegments": 7301,
+      "ĠPhase": 7302,
+      "ĠLand": 7303,
+      "forming": 7304,
+      "hbox": 7305,
+      "ications": 7306,
+      "Ġsubsequently": 7307,
+      "ĠCur": 7308,
+      "Ġlabels": 7309,
+      "vidence": 7310,
+      "uality": 7311,
+      "Ġheld": 7312,
+      "emann": 7313,
+      "Ġcamera": 7314,
+      "cing": 7315,
+      "ubic": 7316,
+      "ĠSARS": 7317,
+      "ulatory": 7318,
+      "keletal": 7319,
+      "ĠInflu": 7320,
+      "ĠIndia": 7321,
+      "amic": 7322,
+      "Ġsand": 7323,
+      "Ġcomes": 7324,
+      "Ġassociations": 7325,
+      "Ġcharged": 7326,
+      "Ġsper": 7327,
+      "oprotein": 7328,
+      "iii": 7329,
+      "odal": 7330,
+      "Ġboundaries": 7331,
+      "tization": 7332,
+      "ĠHeavy": 7333,
+      "ĠReal": 7334,
+      "ĠAF": 7335,
+      "Ġcontroller": 7336,
+      "Ġantioxidant": 7337,
+      "Ġbars": 7338,
+      "Ġwet": 7339,
+      "ener": 7340,
+      "ĠComplexity": 7341,
+      "Ġstack": 7342,
+      "Therefore": 7343,
+      "Ġreplication": 7344,
+      "Ġappearance": 7345,
+      "Ġtrajectory": 7346,
+      "Ġunderstood": 7347,
+      "Ġdot": 7348,
+      "Ġimag": 7349,
+      "Ġscanning": 7350,
+      "Ti": 7351,
+      "ruct": 7352,
+      "ĠLy": 7353,
+      "Ġspontaneous": 7354,
+      "lat": 7355,
+      "omon": 7356,
+      "Ġroots": 7357,
+      "Ġlive": 7358,
+      "Ġfinally": 7359,
+      "¿½": 7360,
+      "Ġapproved": 7361,
+      "ĠApplications": 7362,
+      "ĠPan": 7363,
+      "Ġlost": 7364,
+      "Ġsatisfied": 7365,
+      "Ġgamma": 7366,
+      "ional": 7367,
+      "Ġimproving": 7368,
+      "Ġmanifold": 7369,
+      "Ġcodes": 7370,
+      "bb": 7371,
+      "ĠNovember": 7372,
+      "Ġrich": 7373,
+      "NP": 7374,
+      "ĠEle": 7375,
+      "SB": 7376,
+      "Ġdeal": 7377,
+      "Ġoptions": 7378,
+      "Ġcultured": 7379,
+      "Ġvul": 7380,
+      ">>": 7381,
+      "arithm": 7382,
+      "oys": 7383,
+      "These": 7384,
+      "ĠDeterm": 7385,
+      "Ġquadratic": 7386,
+      "ĠComb": 7387,
+      "isson": 7388,
+      "ĠPerformance": 7389,
+      "Ġexception": 7390,
+      "Ġnuclei": 7391,
+      "Ġadverse": 7392,
+      "ket": 7393,
+      "ĠPal": 7394,
+      "ĠMany": 7395,
+      "Ġdiffraction": 7396,
+      "Ġtransmit": 7397,
+      "Ġphosphate": 7398,
+      "olesterol": 7399,
+      "Ġquestionnai": 7400,
+      "ĠSea": 7401,
+      "bruary": 7402,
+      "Ġmodelling": 7403,
+      "ĠDR": 7404,
+      "olin": 7405,
+      "chmark": 7406,
+      "Ġprecisely": 7407,
+      "gans": 7408,
+      "vin": 7409,
+      "ridge": 7410,
+      "ĠIncre": 7411,
+      "Ġneuronal": 7412,
+      "ĠâīĪ": 7413,
+      "Ġexcellent": 7414,
+      "etary": 7415,
+      "Ġoverlap": 7416,
+      "Ġstronger": 7417,
+      "Ġfracture": 7418,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 7419,
+      "Ġclinic": 7420,
+      "ĠList": 7421,
+      "Ġhistor": 7422,
+      "generation": 7423,
+      "riched": 7424,
+      "illus": 7425,
+      "ĠÃħ": 7426,
+      "ĠRole": 7427,
+      "Ġlabeled": 7428,
+      "Ġorthogonal": 7429,
+      "Ġischem": 7430,
+      "Ġinstability": 7431,
+      "loop": 7432,
+      "Ġplotted": 7433,
+      "ĠProcessing": 7434,
+      "ĠTa": 7435,
+      "ĠConclusion": 7436,
+      "Ġmagne": 7437,
+      "Ġuniversal": 7438,
+      "Ġjet": 7439,
+      "Ġregim": 7440,
+      "float": 7441,
+      "Ġcod": 7442,
+      "adj": 7443,
+      "boldmath": 7444,
+      "Ġarrang": 7445,
+      "Ġtrends": 7446,
+      "Ġprecipitation": 7447,
+      "frequency": 7448,
+      "Ġcontrad": 7449,
+      "Ġtransferred": 7450,
+      "Ġmaintenance": 7451,
+      "ÎĶ": 7452,
+      "np": 7453,
+      "istence": 7454,
+      "heres": 7455,
+      "lective": 7456,
+      "ĠSurvey": 7457,
+      "ĠÐ": 7458,
+      "Ġstand": 7459,
+      "Ġdiscovery": 7460,
+      "ains": 7461,
+      "versely": 7462,
+      "Ġnumerous": 7463,
+      "ylated": 7464,
+      "Ġembedding": 7465,
+      "Ġcollabor": 7466,
+      "ename": 7467,
+      "immun": 7468,
+      "Ġadjusted": 7469,
+      "ires": 7470,
+      "cur": 7471,
+      "Ġvaccine": 7472,
+      "Ġtraits": 7473,
+      "Ġmorphological": 7474,
+      "Ġprecurs": 7475,
+      "roscope": 7476,
+      "adi": 7477,
+      "ecutive": 7478,
+      "uan": 7479,
+      "Ġtract": 7480,
+      "ĠPres": 7481,
+      "Ġmyel": 7482,
+      "Ġadequ": 7483,
+      "Ġethanol": 7484,
+      "ih": 7485,
+      "Ġmeth": 7486,
+      "Ġcounts": 7487,
+      "Ġqualitative": 7488,
+      "Ġmusic": 7489,
+      "Ġreinfor": 7490,
+      "After": 7491,
+      "Ġacquisition": 7492,
+      "Ġhttps": 7493,
+      "alling": 7494,
+      "ita": 7495,
+      "icate": 7496,
+      "script": 7497,
+      "Ġoptimized": 7498,
+      "ĠHo": 7499,
+      "Ġmild": 7500,
+      "oplas": 7501,
+      "Ġoverex": 7502,
+      "ĠâĪ§": 7503,
+      "Ġcollect": 7504,
+      "ĠMain": 7505,
+      "Ġextracellular": 7506,
+      "Ġanc": 7507,
+      "rawn": 7508,
+      "Ġexplored": 7509,
+      "Ġreserv": 7510,
+      "ĠApplication": 7511,
+      "case": 7512,
+      "Ġmarine": 7513,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠ": 7514,
+      "iled": 7515,
+      "Ġmesh": 7516,
+      "ĠMonte": 7517,
+      "clos": 7518,
+      "Ġperforming": 7519,
+      "Ag": 7520,
+      "regular": 7521,
+      "Ġcatal": 7522,
+      "Ġpotentials": 7523,
+      "antly": 7524,
+      "URE": 7525,
+      "Ġaccomp": 7526,
+      "Ġreasonable": 7527,
+      "Ġpresentation": 7528,
+      "abolic": 7529,
+      "ĠOnly": 7530,
+      "anned": 7531,
+      "Ġsubstantial": 7532,
+      "Ġdietary": 7533,
+      "Ġsubstrates": 7534,
+      "apter": 7535,
+      "Ġintestinal": 7536,
+      "Ġproduces": 7537,
+      "Proposition": 7538,
+      "rogen": 7539,
+      "ĠStat": 7540,
+      "burg": 7541,
+      "rench": 7542,
+      "textbf": 7543,
+      "ystems": 7544,
+      "atable": 7545,
+      "ĠVir": 7546,
+      "Ġsolved": 7547,
+      "icense": 7548,
+      "Ġsong": 7549,
+      "Ġextreme": 7550,
+      "pty": 7551,
+      "ĠCity": 7552,
+      "vered": 7553,
+      "ĠMRI": 7554,
+      "Ġtwice": 7555,
+      "ĠMn": 7556,
+      "Ġmerg": 7557,
+      "activation": 7558,
+      "Ġng": 7559,
+      "Ġodd": 7560,
+      "Ġattrac": 7561,
+      "Ġattempt": 7562,
+      "Ġseparately": 7563,
+      "Ġrobot": 7564,
+      "ĠMultiple": 7565,
+      "Ġscientific": 7566,
+      "ĠPP": 7567,
+      "Ġmineral": 7568,
+      "Ġprotocols": 7569,
+      "Ġsuperior": 7570,
+      "ocamp": 7571,
+      "boxyl": 7572,
+      "Ġuniformly": 7573,
+      "ĠSeveral": 7574,
+      "Ġmol": 7575,
+      "Cor": 7576,
+      "underline": 7577,
+      "Ġinfluenced": 7578,
+      "Ġcurren": 7579,
+      "using": 7580,
+      "race": 7581,
+      "ĠNevertheless": 7582,
+      "Ġaccom": 7583,
+      "Ġgravitational": 7584,
+      "Ġindirect": 7585,
+      "Ġcapable": 7586,
+      "Ġanalysed": 7587,
+      "Ġdischarge": 7588,
+      "Ġves": 7589,
+      "Ġligand": 7590,
+      "lik": 7591,
+      "Ġsi": 7592,
+      "Ġaged": 7593,
+      "Ġcrystals": 7594,
+      "Ġspeech": 7595,
+      "Ġcopper": 7596,
+      "ĠSan": 7597,
+      "ĠArm": 7598,
+      "Ġmanuscript": 7599,
+      "Ġsecretion": 7600,
+      "wedge": 7601,
+      "·": 7602,
+      "Ġraw": 7603,
+      "Ġaimed": 7604,
+      "Ġevolutionary": 7605,
+      "Ġconsequences": 7606,
+      "Ġitem": 7607,
+      "Ġwestern": 7608,
+      "Ġsolvent": 7609,
+      "Ġstimuli": 7610,
+      "Ġrequirement": 7611,
+      "http": 7612,
+      "efore": 7613,
+      "ĠAtl": 7614,
+      "Ġatmospheric": 7615,
+      "Ġpackage": 7616,
+      "Ġmyocardi": 7617,
+      "Ġdashed": 7618,
+      "Ġverify": 7619,
+      "ativistic": 7620,
+      "Ġtom": 7621,
+      "avirus": 7622,
+      "aken": 7623,
+      "ĠNumer": 7624,
+      "Ġadvantages": 7625,
+      "FR": 7626,
+      "ĠSelf": 7627,
+      "rected": 7628,
+      "config": 7629,
+      "Ġiteration": 7630,
+      "Ġeigenvalues": 7631,
+      "Ġprobabilities": 7632,
+      "FIG": 7633,
+      "ĠWater": 7634,
+      "ĠAu": 7635,
+      "Ġgave": 7636,
+      "Ġvar": 7637,
+      "ricular": 7638,
+      "opathy": 7639,
+      "Ġrh": 7640,
+      "ordance": 7641,
+      "Ġwin": 7642,
+      "ĠScale": 7643,
+      "Ġannual": 7644,
+      "ataset": 7645,
+      "Ġpel": 7646,
+      "ĠâĪª": 7647,
+      "ĠCC": 7648,
+      "itors": 7649,
+      "Ġlith": 7650,
+      "Ġchromosome": 7651,
+      "Ġfuel": 7652,
+      "Ġmultiv": 7653,
+      "Ġmanufacture": 7654,
+      "la": 7655,
+      "ĠSa": 7656,
+      "umes": 7657,
+      "igm": 7658,
+      "Ġnanoc": 7659,
+      "EGF": 7660,
+      "Ġsignature": 7661,
+      "NS": 7662,
+      "Ġmeet": 7663,
+      "Ġfair": 7664,
+      "meth": 7665,
+      "Ġlocalized": 7666,
+      "ĠCentral": 7667,
+      "deg": 7668,
+      "Ġsurrounding": 7669,
+      "Ġnone": 7670,
+      "ĠMO": 7671,
+      "ĠInterestingly": 7672,
+      "Ġmultic": 7673,
+      "ĠKe": 7674,
+      "Ġinhibited": 7675,
+      "ĠCare": 7676,
+      "ĠOpen": 7677,
+      "Ġglob": 7678,
+      "EA": 7679,
+      "ĠFound": 7680,
+      "Ġpixel": 7681,
+      "oke": 7682,
+      "RD": 7683,
+      "loc": 7684,
+      "tious": 7685,
+      "Ġdistinguish": 7686,
+      "Ġanterior": 7687,
+      "urch": 7688,
+      "Ġjud": 7689,
+      "ĠPower": 7690,
+      "Ġswitch": 7691,
+      "ĠSyn": 7692,
+      "Ġinvolvement": 7693,
+      "ucl": 7694,
+      "Ġlibrary": 7695,
+      "ĠConst": 7696,
+      "Ġspherical": 7697,
+      "ĠTNF": 7698,
+      "Ġaltered": 7699,
+      "vance": 7700,
+      "transfer": 7701,
+      "Ms": 7702,
+      "ĠOper": 7703,
+      "inement": 7704,
+      "seq": 7705,
+      "Cons": 7706,
+      "hole": 7707,
+      "ĠPhot": 7708,
+      "Ġgut": 7709,
+      "acterial": 7710,
+      "ĠIP": 7711,
+      "unt": 7712,
+      "Ġnom": 7713,
+      "has": 7714,
+      "ĠFebruary": 7715,
+      "Ġprostate": 7716,
+      "ĠML": 7717,
+      "high": 7718,
+      "ĠBackground": 7719,
+      "ulent": 7720,
+      "Ġocean": 7721,
+      "after": 7722,
+      "ĠOff": 7723,
+      "loss": 7724,
+      "Ġfavor": 7725,
+      "Ġworkers": 7726,
+      "Ġhidden": 7727,
+      "Ġextracts": 7728,
+      "razil": 7729,
+      "sign": 7730,
+      "None": 7731,
+      "Ġcolumns": 7732,
+      "Ġfractions": 7733,
+      "Ġcovered": 7734,
+      "ĠServ": 7735,
+      "Ġinform": 7736,
+      "bed": 7737,
+      "Ġattem": 7738,
+      "raining": 7739,
+      "Ġneutron": 7740,
+      "Ġrice": 7741,
+      "Ġmotif": 7742,
+      "Ġartificial": 7743,
+      "Ġinhibitory": 7744,
+      "Ġdt": 7745,
+      "AGE": 7746,
+      "Ġsampled": 7747,
+      "Ġbatter": 7748,
+      "Ġsubjected": 7749,
+      "Ġgeneric": 7750,
+      "ĠNH": 7751,
+      "Ġcontinue": 7752,
+      "utional": 7753,
+      "Ġaug": 7754,
+      "ius": 7755,
+      "Ġexecution": 7756,
+      "ĠWilli": 7757,
+      "ĠDespite": 7758,
+      "AMI": 7759,
+      "Ġcontents": 7760,
+      "ĠSens": 7761,
+      "ogens": 7762,
+      "Col": 7763,
+      "Ġfo": 7764,
+      "Ġaddi": 7765,
+      "uated": 7766,
+      "Ġrecommended": 7767,
+      "ĠSW": 7768,
+      "Ġarch": 7769,
+      "ĠYes": 7770,
+      "Ġhol": 7771,
+      "aturally": 7772,
+      "titive": 7773,
+      "Ġche": 7774,
+      "Ġsector": 7775,
+      "ĠDefinition": 7776,
+      "Ġconcepts": 7777,
+      "orous": 7778,
+      "small": 7779,
+      "erson": 7780,
+      "inator": 7781,
+      "ĠMT": 7782,
+      "Ġhypertension": 7783,
+      "cks": 7784,
+      "Ġnative": 7785,
+      "Ġtax": 7786,
+      "ryl": 7787,
+      "Ġreactive": 7788,
+      "rb": 7789,
+      "ducible": 7790,
+      "omm": 7791,
+      "Ġdiagnosed": 7792,
+      "Ġdriving": 7793,
+      "Ġbiomass": 7794,
+      "uate": 7795,
+      "Ġpil": 7796,
+      "called": 7797,
+      "Ġserve": 7798,
+      "Ġinterfer": 7799,
+      "ippocamp": 7800,
+      "Ġalgebraic": 7801,
+      "Ġbegan": 7802,
+      "Ġpicture": 7803,
+      "independent": 7804,
+      "Ġutilized": 7805,
+      "going": 7806,
+      "ora": 7807,
+      "nm": 7808,
+      "Ġdownstream": 7809,
+      "Ġorbital": 7810,
+      "ountain": 7811,
+      "ĠHis": 7812,
+      "Ġresol": 7813,
+      "Ġcorrections": 7814,
+      "onym": 7815,
+      "scripts": 7816,
+      "Ġsilicon": 7817,
+      "Ġcum": 7818,
+      "ĠTri": 7819,
+      "Ġpeptides": 7820,
+      "Ġreceiving": 7821,
+      "Ġstationary": 7822,
+      "ĠμL": 7823,
+      "clerosis": 7824,
+      "Ġmodules": 7825,
+      "ema": 7826,
+      "ĠAfrican": 7827,
+      "struction": 7828,
+      "Ġfarm": 7829,
+      "Ġlearn": 7830,
+      "node": 7831,
+      "®": 7832,
+      "Ġsuperconduc": 7833,
+      "ĠLinear": 7834,
+      "Ġtechnologies": 7835,
+      "Ġnecessarily": 7836,
+      "Ġcoronary": 7837,
+      "ĠEast": 7838,
+      "Ġframes": 7839,
+      "Ġsegmentation": 7840,
+      "Vs": 7841,
+      "Ġbehavioral": 7842,
+      "Îĵ": 7843,
+      "Ġlogic": 7844,
+      "Ġaccompan": 7845,
+      "tified": 7846,
+      "hanol": 7847,
+      "ĠInhib": 7848,
+      "ilation": 7849,
+      "ander": 7850,
+      "Ġeffort": 7851,
+      "ĠDen": 7852,
+      "DI": 7853,
+      "optim": 7854,
+      "terminal": 7855,
+      "Ġmobility": 7856,
+      "Ġconsideration": 7857,
+      "OVA": 7858,
+      "Ġparad": 7859,
+      "oxo": 7860,
+      "Ġdeficiency": 7861,
+      "ultural": 7862,
+      "Ġvalidity": 7863,
+      "Ġorders": 7864,
+      "Ġlocus": 7865,
+      "Ġarth": 7866,
+      "emat": 7867,
+      "Ġfeeding": 7868,
+      "Ġprogramming": 7869,
+      "Ġtemplate": 7870,
+      "elian": 7871,
+      "Ġoption": 7872,
+      "ĠFollowing": 7873,
+      "Ġenable": 7874,
+      "Ġassign": 7875,
+      "Ġformul": 7876,
+      "pu": 7877,
+      "Ġatmosphere": 7878,
+      "slant": 7879,
+      "ĠRuss": 7880,
+      "ĠEvidence": 7881,
+      "Ġsimilarly": 7882,
+      "Ġcamp": 7883,
+      "Ġwound": 7884,
+      "ĠCharacterization": 7885,
+      "ĠPBS": 7886,
+      "ees": 7887,
+      "ĠDirect": 7888,
+      "ĠSL": 7889,
+      "Ġfruit": 7890,
+      "Ġgate": 7891,
+      "ito": 7892,
+      "Chem": 7893,
+      "Ġcollision": 7894,
+      "ortic": 7895,
+      "Ġpolymorphism": 7896,
+      "enza": 7897,
+      "what": 7898,
+      "Ġexperimentally": 7899,
+      "Ġultra": 7900,
+      "ez": 7901,
+      "Ġnerv": 7902,
+      "Ġessentially": 7903,
+      "ĠAustralia": 7904,
+      "ĠStandard": 7905,
+      "Ġmedicine": 7906,
+      "adian": 7907,
+      "ĠHiggs": 7908,
+      "uge": 7909,
+      "Ġsupports": 7910,
+      "uma": 7911,
+      "Ġcomplicated": 7912,
+      "date": 7913,
+      "ophagy": 7914,
+      "ĠMarkov": 7915,
+      "Ġoccurring": 7916,
+      "oplus": 7917,
+      "Pub": 7918,
+      "prob": 7919,
+      "urable": 7920,
+      "Ġkept": 7921,
+      "Ġisolation": 7922,
+      "Ġevol": 7923,
+      "iliary": 7924,
+      "Ġregist": 7925,
+      "Ġholes": 7926,
+      "Ġclar": 7927,
+      "ipar": 7928,
+      "Ġenrich": 7929,
+      "Ġroute": 7930,
+      "ayers": 7931,
+      "ediatric": 7932,
+      "Ġpolynomials": 7933,
+      "Ġtrivial": 7934,
+      "ĠSam": 7935,
+      "variant": 7936,
+      "Ġfreedom": 7937,
+      "poss": 7938,
+      "Ġinference": 7939,
+      "ola": 7940,
+      "Ġinterpreted": 7941,
+      "Ca": 7942,
+      "emory": 7943,
+      "Ġcentury": 7944,
+      "ĠRem": 7945,
+      "ĠWu": 7946,
+      "Ġsuppression": 7947,
+      "Ġgenerator": 7948,
+      "ĠHom": 7949,
+      "Ġviscos": 7950,
+      "Ġpseudo": 7951,
+      "ĠChild": 7952,
+      "ĠSA": 7953,
+      "iber": 7954,
+      "Ġequivalence": 7955,
+      "ifies": 7956,
+      "ĠConsider": 7957,
+      "oline": 7958,
+      "âī¤": 7959,
+      "Ġdeple": 7960,
+      "Ġaveraged": 7961,
+      "Ġsouthern": 7962,
+      "Ġordered": 7963,
+      "ĠBrown": 7964,
+      "Ġmethylation": 7965,
+      "ĠAdap": 7966,
+      "Ġmaternal": 7967,
+      "onded": 7968,
+      "ĠBehavi": 7969,
+      "Ġidentifiers": 7970,
+      "Ġprocessed": 7971,
+      "GG": 7972,
+      "VI": 7973,
+      "Ġcha": 7974,
+      "unk": 7975,
+      "ĠFunctional": 7976,
+      "Ġhydroph": 7977,
+      "Ġfinancial": 7978,
+      "econd": 7979,
+      "ĠΨ": 7980,
+      "Ġemphas": 7981,
+      "Ġdefect": 7982,
+      "mar": 7983,
+      "Ġnorthern": 7984,
+      "core": 7985,
+      "Ġadhesion": 7986,
+      "Ġtele": 7987,
+      "Ġwarm": 7988,
+      "rifug": 7989,
+      "rangian": 7990,
+      "resolution": 7991,
+      "Ġhex": 7992,
+      "hbar": 7993,
+      "Ġharmonic": 7994,
+      "Ġcontrac": 7995,
+      "Ġreading": 7996,
+      "Ġefforts": 7997,
+      "ĠOl": 7998,
+      "Ġanxiety": 7999,
+      "bul": 8000,
+      "TC": 8001,
+      "ipid": 8002,
+      "Remark": 8003,
+      "Ġforming": 8004,
+      "ilbert": 8005,
+      "amond": 8006,
+      "Ġanalytic": 8007,
+      "orec": 8008,
+      "cha": 8009,
+      "ĠConsequently": 8010,
+      "ĠSu": 8011,
+      "forall": 8012,
+      "ĠÃŀ": 8013,
+      "Ġaspect": 8014,
+      "Ġinsights": 8015,
+      "ativity": 8016,
+      "iotics": 8017,
+      "heimer": 8018,
+      "ĠLabor": 8019,
+      "Ġaware": 8020,
+      "ĠBritish": 8021,
+      "chemical": 8022,
+      "Ġâĭ": 8023,
+      "clusion": 8024,
+      "ĠMich": 8025,
+      "Ġgrade": 8026,
+      "ĠSEM": 8027,
+      "ĠCirc": 8028,
+      "heses": 8029,
+      "WL": 8030,
+      "Ġenabl": 8031,
+      "Ġdend": 8032,
+      "Ġindustry": 8033,
+      "Ġimproves": 8034,
+      "tet": 8035,
+      "Ġtel": 8036,
+      "Ġwashed": 8037,
+      "Ġshorter": 8038,
+      "Ġincident": 8039,
+      "ĠActivity": 8040,
+      "Ġdoses": 8041,
+      "ĠBrazil": 8042,
+      "Ġtransformations": 8043,
+      "Ġformat": 8044,
+      "ĠProof": 8045,
+      "Ġlen": 8046,
+      "ulative": 8047,
+      "Ġcyclic": 8048,
+      "Ġrecruit": 8049,
+      "ptr": 8050,
+      "TH": 8051,
+      "Ġreceive": 8052,
+      "ĠNext": 8053,
+      "ĠExp": 8054,
+      "iant": 8055,
+      "instein": 8056,
+      "Set": 8057,
+      "rene": 8058,
+      "Ġgeomet": 8059,
+      "Ġconsiderable": 8060,
+      "So": 8061,
+      "ught": 8062,
+      "Ġpapers": 8063,
+      "ĠCS": 8064,
+      "za": 8065,
+      "Ġisomorphism": 8066,
+      "hou": 8067,
+      "Ġmutants": 8068,
+      "Ġportion": 8069,
+      "Ġþ": 8070,
+      "Ġcontinuum": 8071,
+      "Cu": 8072,
+      "ĠComputed": 8073,
+      "Ġcombining": 8074,
+      "ova": 8075,
+      "ĠNP": 8076,
+      "Ġcrack": 8077,
+      "Ġsometimes": 8078,
+      "Ġcontinued": 8079,
+      "Definition": 8080,
+      "arcin": 8081,
+      "ĠCd": 8082,
+      "ĠMedical": 8083,
+      "iences": 8084,
+      "ĠCross": 8085,
+      "Ġtranscriptional": 8086,
+      "ĠZe": 8087,
+      "std": 8088,
+      "iforn": 8089,
+      "Ġfailed": 8090,
+      "Ġidentifying": 8091,
+      "Ġmir": 8092,
+      "Ġmetastasis": 8093,
+      "OF": 8094,
+      "nn": 8095,
+      "ĠCID": 8096,
+      "Ġoscillations": 8097,
+      "ancies": 8098,
+      "write": 8099,
+      "Ġbandwidth": 8100,
+      "Ġtrade": 8101,
+      "Ġaging": 8102,
+      "ĠModeling": 8103,
+      "Ġassert": 8104,
+      "Ġcurrents": 8105,
+      "Ġfire": 8106,
+      "ubiqu": 8107,
+      "Ġalbum": 8108,
+      "Ġfrequent": 8109,
+      "Name": 8110,
+      "Ġpurch": 8111,
+      "Ġplayer": 8112,
+      "ĠEsc": 8113,
+      "Ġnotion": 8114,
+      "Ġinternational": 8115,
+      "ulum": 8116,
+      "oic": 8117,
+      "Ġincubation": 8118,
+      "Ġphenomena": 8119,
+      "Ġserver": 8120,
+      "uter": 8121,
+      "Ġven": 8122,
+      "quin": 8123,
+      "Ġhypox": 8124,
+      "ĠRF": 8125,
+      "iton": 8126,
+      "Error": 8127,
+      "Ġhemat": 8128,
+      "Ġthemselves": 8129,
+      "Ġperp": 8130,
+      "idual": 8131,
+      "Ġpurposes": 8132,
+      "mes": 8133,
+      "wing": 8134,
+      "rov": 8135,
+      "Ġemiss": 8136,
+      "Ġexperienced": 8137,
+      "ques": 8138,
+      "ĠLC": 8139,
+      "ĠRecent": 8140,
+      "book": 8141,
+      "Ġalkal": 8142,
+      "idx": 8143,
+      "hyth": 8144,
+      "Ġconcrete": 8145,
+      "Ġswitching": 8146,
+      "Ġexplanation": 8147,
+      "irds": 8148,
+      "Ġsigns": 8149,
+      "Ġobj": 8150,
+      "Ġcytokines": 8151,
+      "ubble": 8152,
+      "adder": 8153,
+      "Ġuncertainties": 8154,
+      "Ġpromotes": 8155,
+      "Ġcompl": 8156,
+      "Ġscan": 8157,
+      "Ġprime": 8158,
+      "PH": 8159,
+      "Ġheterogeneous": 8160,
+      "ĠYou": 8161,
+      "Although": 8162,
+      "Ġserious": 8163,
+      "Ġdrive": 8164,
+      "Ġheterogeneity": 8165,
+      "rystall": 8166,
+      "Ġod": 8167,
+      "Ġconvolution": 8168,
+      "ĠâĬĨ": 8169,
+      "ĠSpace": 8170,
+      "Ġgastric": 8171,
+      "ĠStre": 8172,
+      "ĠPV": 8173,
+      "base": 8174,
+      "Met": 8175,
+      "Ġlosses": 8176,
+      "Ġcytotox": 8177,
+      "Ġcontrolling": 8178,
+      "lease": 8179,
+      "Ġregulated": 8180,
+      "ĠEngine": 8181,
+      "ĠHospital": 8182,
+      "Br": 8183,
+      "onom": 8184,
+      "hyde": 8185,
+      "stage": 8186,
+      "Ġgiving": 8187,
+      "ĠPen": 8188,
+      "ĠSociety": 8189,
+      "driven": 8190,
+      "iang": 8191,
+      "Ġmodifications": 8192,
+      "BV": 8193,
+      "Ġacceleration": 8194,
+      "Ġmilk": 8195,
+      "onomic": 8196,
+      "Ġthink": 8197,
+      "oglob": 8198,
+      "Ġfeasible": 8199,
+      "nam": 8200,
+      "Ġreflection": 8201,
+      "ĠPoly": 8202,
+      "Ġsummarized": 8203,
+      "FL": 8204,
+      "Ġrect": 8205,
+      "Ġpredominant": 8206,
+      "Ġblot": 8207,
+      "dehyde": 8208,
+      "Ġtransformed": 8209,
+      "Ġfacilitate": 8210,
+      "ĠCarlo": 8211,
+      "Ġgreatly": 8212,
+      "ĠSocial": 8213,
+      "Ġparents": 8214,
+      "bigg": 8215,
+      "rospective": 8216,
+      "Ġprognosis": 8217,
+      "Ġcharacterize": 8218,
+      "Ġconnectivity": 8219,
+      "Ġtrajectories": 8220,
+      "ĠSH": 8221,
+      "Ġlies": 8222,
+      "Ġcandidates": 8223,
+      "romy": 8224,
+      "Ġsor": 8225,
+      "ĠIns": 8226,
+      "Ġthor": 8227,
+      "Ġmetals": 8228,
+      "ĠSV": 8229,
+      "Ġtiming": 8230,
+      "Ġutility": 8231,
+      "Ġnewly": 8232,
+      "ĠIFN": 8233,
+      "Ġaffecting": 8234,
+      "cement": 8235,
+      "ĠMel": 8236,
+      "ĠÌģ": 8237,
+      "types": 8238,
+      "lysis": 8239,
+      "ercul": 8240,
+      "Ġdistor": 8241,
+      "actors": 8242,
+      "psy": 8243,
+      "Ġbook": 8244,
+      "ĠEven": 8245,
+      "temperature": 8246,
+      "Ġinvasion": 8247,
+      "Ġrecognized": 8248,
+      "factor": 8249,
+      "Ne": 8250,
+      "Ġintersection": 8251,
+      "Ġcortical": 8252,
+      "ng": 8253,
+      "Ġdeploy": 8254,
+      "Ġamplitudes": 8255,
+      "Ġda": 8256,
+      "ĠGC": 8257,
+      "Ġchallenging": 8258,
+      "Ġprelim": 8259,
+      "GM": 8260,
+      "Acc": 8261,
+      "Ġfourth": 8262,
+      "alc": 8263,
+      "ĠPS": 8264,
+      "ĠGenetic": 8265,
+      "lock": 8266,
+      "error": 8267,
+      "skip": 8268,
+      "sime": 8269,
+      "Ġana": 8270,
+      "simeq": 8271,
+      "Ġcerebral": 8272,
+      "ĠEX": 8273,
+      "aved": 8274,
+      "rophy": 8275,
+      "idopsis": 8276,
+      "Ġbehind": 8277,
+      "Ġenables": 8278,
+      "Ġindustrial": 8279,
+      "ĠPac": 8280,
+      "Ġdefinitions": 8281,
+      "Ġcatalytic": 8282,
+      "Ġdissip": 8283,
+      "ervical": 8284,
+      "Ġcommut": 8285,
+      "Ġrepeat": 8286,
+      "Ġchiral": 8287,
+      "Ġpron": 8288,
+      "pol": 8289,
+      "Ġgoing": 8290,
+      "Ġmicroscope": 8291,
+      "Ġhealthcare": 8292,
+      "ĠClassification": 8293,
+      "titude": 8294,
+      "ĠFermi": 8295,
+      "Ġhttp": 8296,
+      "arest": 8297,
+      "Ġsupporting": 8298,
+      "Ġwood": 8299,
+      "night": 8300,
+      "Ġkinetics": 8301,
+      "Ġsubsets": 8302,
+      "Ġsubunit": 8303,
+      "ĠCanada": 8304,
+      "aton": 8305,
+      "Ġaccurately": 8306,
+      "Ġresistant": 8307,
+      "ĠïĢ½": 8308,
+      "riction": 8309,
+      "Ġchamber": 8310,
+      "igue": 8311,
+      "ĠPhil": 8312,
+      "Ġrecover": 8313,
+      "cs": 8314,
+      "Ġsphere": 8315,
+      "ĠSpecifically": 8316,
+      "Ġanne": 8317,
+      "Ġinitiation": 8318,
+      "ĠTH": 8319,
+      "Ġbud": 8320,
+      "ordered": 8321,
+      "Ġdielectric": 8322,
+      "ĠCollege": 8323,
+      "Ġproducing": 8324,
+      "Ġantenna": 8325,
+      "Bs": 8326,
+      "ĠFrench": 8327,
+      "OX": 8328,
+      "ĠAmerica": 8329,
+      "ĠâĢĶ": 8330,
+      "ounting": 8331,
+      "fully": 8332,
+      "Ġserved": 8333,
+      "Ġresidue": 8334,
+      "Ġarguments": 8335,
+      "Ġpand": 8336,
+      "Ġcompany": 8337,
+      "Ġconditional": 8338,
+      "mia": 8339,
+      "ĠQCD": 8340,
+      "Ġviscosity": 8341,
+      "Ġprospective": 8342,
+      "asonal": 8343,
+      "Ġdominated": 8344,
+      "Ġpenet": 8345,
+      "opo": 8346,
+      "Ġnine": 8347,
+      "ĠIll": 8348,
+      "ĠVisual": 8349,
+      "Ġfiles": 8350,
+      "Ġyeast": 8351,
+      "Ġthank": 8352,
+      "GN": 8353,
+      "real": 8354,
+      "Ġverified": 8355,
+      "ĠIndian": 8356,
+      "Ġstiff": 8357,
+      "rological": 8358,
+      "Ġdram": 8359,
+      "Ġtight": 8360,
+      "ĠGerman": 8361,
+      "ĠTechnology": 8362,
+      "ĠApproach": 8363,
+      "romatic": 8364,
+      "Ġacoustic": 8365,
+      "tian": 8366,
+      "osin": 8367,
+      "ĠDepartment": 8368,
+      "otropy": 8369,
+      "Ġempty": 8370,
+      "trivial": 8371,
+      "ofil": 8372,
+      "Ġalgebras": 8373,
+      "texts": 8374,
+      "Ġwebs": 8375,
+      "Ġpore": 8376,
+      "Ġpacket": 8377,
+      "Time": 8378,
+      "img": 8379,
+      "ony": 8380,
+      "ritic": 8381,
+      "Ġvelocities": 8382,
+      "ĠDynamics": 8383,
+      "Ġcancers": 8384,
+      "Ġtrunc": 8385,
+      "ĠFormation": 8386,
+      "ĠDonor": 8387,
+      "ĠMit": 8388,
+      "IST": 8389,
+      "Ġconcluded": 8390,
+      "Ġantag": 8391,
+      "ĠSoft": 8392,
+      "append": 8393,
+      "Ġfragments": 8394,
+      "ĠProf": 8395,
+      "Ġfluor": 8396,
+      "ĠJac": 8397,
+      "ĠSn": 8398,
+      "Ġlept": 8399,
+      "Ġsplitting": 8400,
+      "Ġsexual": 8401,
+      "ĠFore": 8402,
+      "ĠGener": 8403,
+      "Ġneighborhood": 8404,
+      "Ġbenchmark": 8405,
+      "ĠRA": 8406,
+      "Ġdivision": 8407,
+      "ifornia": 8408,
+      "True": 8409,
+      "Ġfuzzy": 8410,
+      "Ġtro": 8411,
+      "cents": 8412,
+      "Ġconstitu": 8413,
+      "atial": 8414,
+      "astern": 8415,
+      "ĠTim": 8416,
+      "Ġperception": 8417,
+      "Ġsubstanti": 8418,
+      "Ġmacro": 8419,
+      "Ġoutl": 8420,
+      "ĠObserv": 8421,
+      "prising": 8422,
+      "oked": 8423,
+      "orectal": 8424,
+      "ĠCho": 8425,
+      "ĠDifferent": 8426,
+      "Ġinvestigations": 8427,
+      "Ġconsistency": 8428,
+      "ients": 8429,
+      "ĠFOR": 8430,
+      "ASS": 8431,
+      "ĠVan": 8432,
+      "Ġsituations": 8433,
+      "ĠBR": 8434,
+      "Ġinfrared": 8435,
+      "ymal": 8436,
+      "Ġpixels": 8437,
+      "Ġcarrier": 8438,
+      "sen": 8439,
+      "INT": 8440,
+      "Ġefficiently": 8441,
+      "DT": 8442,
+      "ĠExpl": 8443,
+      "ionic": 8444,
+      "Ġnaturally": 8445,
+      "Ġpropos": 8446,
+      "Ġguide": 8447,
+      "Ġconclusions": 8448,
+      "oon": 8449,
+      "Ġgrant": 8450,
+      "Ġinstances": 8451,
+      "Ġreviewed": 8452,
+      "Ġelectromagnetic": 8453,
+      "Ġthreat": 8454,
+      "edia": 8455,
+      "ĠOptimization": 8456,
+      "ĠBio": 8457,
+      "Ġtrigger": 8458,
+      "icient": 8459,
+      "otypic": 8460,
+      "Ġstret": 8461,
+      "Ġantic": 8462,
+      "Ġtoxic": 8463,
+      "Ġspinal": 8464,
+      "UPAC": 8465,
+      "Ġoverview": 8466,
+      "otion": 8467,
+      "Ġstraightforward": 8468,
+      "Ġpositively": 8469,
+      "aste": 8470,
+      "Ġreferences": 8471,
+      "ulose": 8472,
+      "ĠGre": 8473,
+      "Ġantagon": 8474,
+      "Ġshifts": 8475,
+      "Ġdrawn": 8476,
+      "ĠWhite": 8477,
+      "Ġfractional": 8478,
+      "Ġbundle": 8479,
+      "Ġexhibits": 8480,
+      "Ġreservoir": 8481,
+      "ĠAlex": 8482,
+      "Ġaggregation": 8483,
+      "Ġcircle": 8484,
+      "Ġpractices": 8485,
+      "ĠCoval": 8486,
+      "ĠDistribution": 8487,
+      "Ġtang": 8488,
+      "ĠMut": 8489,
+      "Ġregulate": 8490,
+      "osphere": 8491,
+      "iro": 8492,
+      "AMINO": 8493,
+      "vest": 8494,
+      "Ġphotos": 8495,
+      "Ġevident": 8496,
+      "Ġbusiness": 8497,
+      "control": 8498,
+      "Ġworth": 8499,
+      "ĠPoisson": 8500,
+      "ĠArabidopsis": 8501,
+      "ĠTarget": 8502,
+      "Ġregulates": 8503,
+      "ĠIr": 8504,
+      "ĠAdv": 8505,
+      "Ġensemble": 8506,
+      "pring": 8507,
+      "Ġprice": 8508,
+      "ĠFL": 8509,
+      "ĠImpact": 8510,
+      "Ġeventually": 8511,
+      "inating": 8512,
+      "Ġcentrifug": 8513,
+      "frame": 8514,
+      "Ġdiagrams": 8515,
+      "Ġtag": 8516,
+      "Ġtry": 8517,
+      "surface": 8518,
+      "ĠIdentifiers": 8519,
+      "rained": 8520,
+      "Ġsides": 8521,
+      "Ġinn": 8522,
+      "Ġflexible": 8523,
+      "Ġsatellite": 8524,
+      "Ġaffinity": 8525,
+      "Ġsummer": 8526,
+      "GP": 8527,
+      "amb": 8528,
+      "Ġaqu": 8529,
+      "String": 8530,
+      "treatment": 8531,
+      "ĠDynamic": 8532,
+      "mathop": 8533,
+      "Ġnotice": 8534,
+      "nes": 8535,
+      "rowave": 8536,
+      "vestig": 8537,
+      "Ġoutputs": 8538,
+      "Ġcoherent": 8539,
+      "Ġillustrate": 8540,
+      "Ġvalidated": 8541,
+      "ĠSchem": 8542,
+      "Ġasked": 8543,
+      "batch": 8544,
+      "Ġpurified": 8545,
+      "Ġminimize": 8546,
+      "ĠDE": 8547,
+      "UM": 8548,
+      "check": 8549,
+      "varian": 8550,
+      "ĠGold": 8551,
+      "ylene": 8552,
+      "IO": 8553,
+      "Ġcholesterol": 8554,
+      "PubChem": 8555,
+      "ĠKore": 8556,
+      "ĠCounty": 8557,
+      "Ġii": 8558,
+      "ĠMAP": 8559,
+      "ectomy": 8560,
+      "Ġsemantic": 8561,
+      "Ġcollagen": 8562,
+      "Ġperceived": 8563,
+      "ichia": 8564,
+      "Ġadministered": 8565,
+      "containing": 8566,
+      "rank": 8567,
+      "InChI": 8568,
+      "Ġirradiation": 8569,
+      "Ġlogarithm": 8570,
+      "Ġgames": 8571,
+      "Ġinjected": 8572,
+      "ĠMHz": 8573,
+      "Ġdors": 8574,
+      "Ġevaluating": 8575,
+      "ĠHyper": 8576,
+      "Ġchromatography": 8577,
+      "phen": 8578,
+      "ĠKar": 8579,
+      "Ġantimicrobial": 8580,
+      "riend": 8581,
+      "Ġdescribing": 8582,
+      "Ġwt": 8583,
+      "Ġhormone": 8584,
+      "AK": 8585,
+      "ĠIUPAC": 8586,
+      "Ga": 8587,
+      "Ġvitamin": 8588,
+      "Ġconnections": 8589,
+      "uous": 8590,
+      "ĠLine": 8591,
+      "Ġbeneficial": 8592,
+      "cases": 8593,
+      "icated": 8594,
+      "isks": 8595,
+      "parent": 8596,
+      "Id": 8597,
+      "eries": 8598,
+      "run": 8599,
+      "Ġmind": 8600,
+      "itt": 8601,
+      "sulf": 8602,
+      "zheimer": 8603,
+      "Ġinterf": 8604,
+      "Vert": 8605,
+      "Ġanth": 8606,
+      "ologous": 8607,
+      "ĠLife": 8608,
+      "Ġmur": 8609,
+      "Ġpermut": 8610,
+      "oting": 8611,
+      "Ġneutrino": 8612,
+      "Ġborn": 8613,
+      "pmatrix": 8614,
+      "ĠCalifornia": 8615,
+      "agent": 8616,
+      "Ġcollisions": 8617,
+      "ĠNS": 8618,
+      "Ġhippocamp": 8619,
+      "Ġpowder": 8620,
+      "Ġvaries": 8621,
+      "Ġepidem": 8622,
+      "ĠWeb": 8623,
+      "uler": 8624,
+      "Ġinterested": 8625,
+      "Ġdevelopmental": 8626,
+      "Ġlengths": 8627,
+      "Ġcolour": 8628,
+      "Ġquas": 8629,
+      "ĠRich": 8630,
+      "Eq": 8631,
+      "Ġinfants": 8632,
+      "ĠPH": 8633,
+      "ophila": 8634,
+      "Ġcausing": 8635,
+      "Ge": 8636,
+      "module": 8637,
+      "IB": 8638,
+      "Ġcontributed": 8639,
+      "rose": 8640,
+      "Ġcytoplas": 8641,
+      "--------------------------------": 8642,
+      "Ġroad": 8643,
+      "symmetric": 8644,
+      "Us": 8645,
+      "Ġweakly": 8646,
+      "tite": 8647,
+      "Ġdefines": 8648,
+      "ĠPE": 8649,
+      "Ġmetabolites": 8650,
+      "Ġlob": 8651,
+      "Ġterminal": 8652,
+      "Ġdemonstrates": 8653,
+      "ĠAcceptor": 8654,
+      "ĠClo": 8655,
+      "Ġinferred": 8656,
+      "Ġvill": 8657,
+      "First": 8658,
+      "Ġneglig": 8659,
+      "Ġwireless": 8660,
+      "Ab": 8661,
+      "particle": 8662,
+      "oisotopic": 8663,
+      "Ġexcited": 8664,
+      "PM": 8665,
+      "Ġconsecutive": 8666,
+      "ĠIsotype": 8667,
+      "Ġstimulus": 8668,
+      "ĠMC": 8669,
+      "timate": 8670,
+      "ĠCovalently": 8671,
+      "Bonded": 8672,
+      "Ġyellow": 8673,
+      "Ġalloy": 8674,
+      "density": 8675,
+      "Ġfilters": 8676,
+      "Ġamplification": 8677,
+      "Ġwon": 8678,
+      "ht": 8679,
+      "Ġimpacts": 8680,
+      "Ġstaff": 8681,
+      "ĠâĪĢ": 8682,
+      "ĠIsomeric": 8683,
+      "Ġsmoking": 8684,
+      "Qu": 8685,
+      "Ġcaptured": 8686,
+      "haps": 8687,
+      "ĠNULL": 8688,
+      "Ġriver": 8689,
+      "count": 8690,
+      "Ġmanifest": 8691,
+      "Ġdiabetic": 8692,
+      "Ġalterations": 8693,
+      "ĠRotatable": 8694,
+      "ĠPRO": 8695,
+      "ĠMonoisotopic": 8696,
+      "ĠïĤ": 8697,
+      "spect": 8698,
+      "Ġcatalyst": 8699,
+      "Ġmodeled": 8700,
+      "Ġpage": 8701,
+      "ĠROS": 8702,
+      "ĠCanonicalized": 8703,
+      "ĠTw": 8704,
+      "Ġaux": 8705,
+      "avage": 8706,
+      "ĠRaman": 8707,
+      "sto": 8708,
+      "perf": 8709,
+      "Ġreplacement": 8710,
+      "ĠEnvironment": 8711,
+      "Ġacting": 8712,
+      "pati": 8713,
+      "ificant": 8714,
+      "through": 8715,
+      "Ġsaturation": 8716,
+      "Ġtip": 8717,
+      "Ġrecurrence": 8718,
+      "ĠHistory": 8719,
+      "Ġprotective": 8720,
+      "Ġburden": 8721,
+      "ado": 8722,
+      "yes": 8723,
+      "inst": 8724,
+      "Ap": 8725,
+      "ĠSy": 8726,
+      "Ġphon": 8727,
+      "ĠâĪij": 8728,
+      "Ġgenotype": 8729,
+      "Ġcovariance": 8730,
+      "Ġquickly": 8731,
+      "ĠDu": 8732,
+      "Ġsug": 8733,
+      "Ġdecline": 8734,
+      "ĠTB": 8735,
+      "Ġstrictly": 8736,
+      "Ġmoist": 8737,
+      "undred": 8738,
+      "ĠCB": 8739,
+      "atile": 8740,
+      "ĠHF": 8741,
+      "Ġarticles": 8742,
+      "Ġps": 8743,
+      "ĠEnh": 8744,
+      "isting": 8745,
+      "Ġbiology": 8746,
+      "Ġbodies": 8747,
+      "ĠAk": 8748,
+      "ĠNumerical": 8749,
+      "ĠLagrangian": 8750,
+      "Ġdiscovered": 8751,
+      "Ġvic": 8752,
+      "opes": 8753,
+      "Ġfragment": 8754,
+      "Ġty": 8755,
+      "ismic": 8756,
+      "Ġhepatic": 8757,
+      "Ġenriched": 8758,
+      "pan": 8759,
+      "Ġinfluences": 8760,
+      "ĠLake": 8761,
+      "color": 8762,
+      "Ġenrichment": 8763,
+      "ochemistry": 8764,
+      "Ġunstable": 8765,
+      "ĠIgG": 8766,
+      "derly": 8767,
+      "Ġecos": 8768,
+      "Ġconcerning": 8769,
+      "ĠRisk": 8770,
+      "Ġmargin": 8771,
+      "Ġpathogenesis": 8772,
+      "Ġpump": 8773,
+      "Ġpreliminary": 8774,
+      "Ġtumour": 8775,
+      "Further": 8776,
+      "azole": 8777,
+      "Ġelectrodes": 8778,
+      "Ġdial": 8779,
+      "ubes": 8780,
+      "ĠNatural": 8781,
+      "ĠMul": 8782,
+      "ĠïĢŃ": 8783,
+      "Ġnic": 8784,
+      "Ġimped": 8785,
+      "only": 8786,
+      "Ġcomparative": 8787,
+      "rection": 8788,
+      "aki": 8789,
+      "Ġrend": 8790,
+      "Ġsparse": 8791,
+      "Ġindicator": 8792,
+      "location": 8793,
+      "tism": 8794,
+      "activated": 8795,
+      "ĠPb": 8796,
+      "eptide": 8797,
+      "Ġendogenous": 8798,
+      "Ġcenters": 8799,
+      "ao": 8800,
+      "sw": 8801,
+      "Ġconsensus": 8802,
+      "Ġattributes": 8803,
+      "Ġsafe": 8804,
+      "Ġbelieve": 8805,
+      "ovirus": 8806,
+      "Ġimmunity": 8807,
+      "Ġfitted": 8808,
+      "Ġcontributes": 8809,
+      "iable": 8810,
+      "Ġviruses": 8811,
+      "Ġinsight": 8812,
+      "ĠNovel": 8813,
+      "ĠAlzheimer": 8814,
+      "cepted": 8815,
+      "ĠPt": 8816,
+      "Ġcentre": 8817,
+      "nat": 8818,
+      "Ġbiosynthesis": 8819,
+      "mits": 8820,
+      "Ġchemistry": 8821,
+      "Ġjus": 8822,
+      "anish": 8823,
+      "Ġrefrac": 8824,
+      "ĠTor": 8825,
+      "Ġpanels": 8826,
+      "Ġimply": 8827,
+      "Ġmatched": 8828,
+      "usc": 8829,
+      "word": 8830,
+      "vae": 8831,
+      "ĠStar": 8832,
+      "syn": 8833,
+      "Mat": 8834,
+      "Ġapplicable": 8835,
+      "ĠPseud": 8836,
+      "ampions": 8837,
+      "ĠRen": 8838,
+      "Ġusage": 8839,
+      "ĠLight": 8840,
+      "prec": 8841,
+      "Ġfibrosis": 8842,
+      "Ġreconstruc": 8843,
+      "ĠON": 8844,
+      "ĠGHz": 8845,
+      "GD": 8846,
+      "algebra": 8847,
+      "iger": 8848,
+      "Ġdecisions": 8849,
+      "infected": 8850,
+      "knowledg": 8851,
+      "Ġexpressing": 8852,
+      "Ġmyocardial": 8853,
+      "ordination": 8854,
+      "Ġprognostic": 8855,
+      "Ġfibrobl": 8856,
+      "Ġacceler": 8857,
+      "ĠAssessment": 8858,
+      "Ġconstrained": 8859,
+      "Ġallele": 8860,
+      "ride": 8861,
+      "Ġrequest": 8862,
+      "abilistic": 8863,
+      "teb": 8864,
+      "Ġga": 8865,
+      "Ġrecovered": 8866,
+      "Ġpromin": 8867,
+      "urses": 8868,
+      "ĠHC": 8869,
+      "ĠMur": 8870,
+      "ĠEqs": 8871,
+      "Ġdefining": 8872,
+      "Ġmer": 8873,
+      "image": 8874,
+      "Ġorganisms": 8875,
+      "grad": 8876,
+      "Ġreflected": 8877,
+      "elastic": 8878,
+      "eties": 8879,
+      "dimethyl": 8880,
+      "ELO": 8881,
+      "random": 8882,
+      "ĠDiagn": 8883,
+      "erculosis": 8884,
+      "rob": 8885,
+      "Ġmoments": 8886,
+      "ĠEC": 8887,
+      "Ġexperiences": 8888,
+      "erving": 8889,
+      "ĠNC": 8890,
+      "Ġvortex": 8891,
+      "gre": 8892,
+      "structures": 8893,
+      "elt": 8894,
+      "Ġcarry": 8895,
+      "ĠThrough": 8896,
+      "Ġpreced": 8897,
+      "rastruct": 8898,
+      "itus": 8899,
+      "Ġpsychological": 8900,
+      "Ġlimiting": 8901,
+      "two": 8902,
+      "ĠBound": 8903,
+      "ĠCre": 8904,
+      "ĠSmith": 8905,
+      "Ġcast": 8906,
+      "Ġcompetition": 8907,
+      "sch": 8908,
+      "Ġcapability": 8909,
+      "achment": 8910,
+      "Ġinhibits": 8911,
+      "ð": 8912,
+      "ĠDifferential": 8913,
+      "Ġautomatically": 8914,
+      "Ġgest": 8915,
+      "Ġwaters": 8916,
+      "Ġuniqu": 8917,
+      "zer": 8918,
+      "Equ": 8919,
+      "Ġstudying": 8920,
+      "Ġdied": 8921,
+      "Ġos": 8922,
+      "Ġrecombination": 8923,
+      "uncil": 8924,
+      "Ġpathogen": 8925,
+      "GFR": 8926,
+      "UV": 8927,
+      "eneration": 8928,
+      "ĠSta": 8929,
+      "Ġinstant": 8930,
+      "Ġproven": 8931,
+      "Ġds": 8932,
+      "Ġdamp": 8933,
+      "Next": 8934,
+      "ĠYoung": 8935,
+      "Ġpowerful": 8936,
+      "Ġwriting": 8937,
+      "kl": 8938,
+      "Ġcareer": 8939,
+      "ĠCorollary": 8940,
+      "Ns": 8941,
+      "Ġ�": 8942,
+      "ĠMil": 8943,
+      "Ġburn": 8944,
+      "ticular": 8945,
+      "ondon": 8946,
+      "Pr": 8947,
+      "ĠLin": 8948,
+      "ĠJapanese": 8949,
+      "ĠLab": 8950,
+      "Ġstrip": 8951,
+      "protein": 8952,
+      "Ġhour": 8953,
+      "anglement": 8954,
+      "anguages": 8955,
+      "rd": 8956,
+      "parse": 8957,
+      "Ġemissions": 8958,
+      "Hence": 8959,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 8960,
+      "Ġjob": 8961,
+      "ĠAS": 8962,
+      "Ġaxial": 8963,
+      "ĠTur": 8964,
+      "carbon": 8965,
+      "MF": 8966,
+      "ĠNE": 8967,
+      "Ġarise": 8968,
+      "Ġlinearly": 8969,
+      "Ġprolong": 8970,
+      "Ġleak": 8971,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 8972,
+      "Ġmoved": 8973,
+      "orbidity": 8974,
+      "Ġprofessional": 8975,
+      "code": 8976,
+      "osine": 8977,
+      "Ġpolic": 8978,
+      "Ġbonds": 8979,
+      "mask": 8980,
+      "Ġconverted": 8981,
+      "ville": 8982,
+      "ectious": 8983,
+      "parallel": 8984,
+      "ĠHal": 8985,
+      "ĠTGF": 8986,
+      "mental": 8987,
+      "Ġreader": 8988,
+      "Ġstandards": 8989,
+      "ago": 8990,
+      "ĠEN": 8991,
+      "Ġstations": 8992,
+      "Ġnormalization": 8993,
+      "ĠÎĺ": 8994,
+      "chain": 8995,
+      "What": 8996,
+      "Ġtomography": 8997,
+      "Ġentries": 8998,
+      "blue": 8999,
+      "ĠPrevious": 9000,
+      "ias": 9001,
+      "Ġquestionnaire": 9002,
+      "Ġhaz": 9003,
+      "Ġhomology": 9004,
+      "very": 9005,
+      "Ġnucleotide": 9006,
+      "ĠGenome": 9007,
+      "Ġμl": 9008,
+      "Ġutilization": 9009,
+      "Ġpolymers": 9010,
+      "rote": 9011,
+      "Ġsmallest": 9012,
+      "calc": 9013,
+      "Ġspl": 9014,
+      "Ġtension": 9015,
+      "Ġdiscontinu": 9016,
+      "ala": 9017,
+      "hol": 9018,
+      "Ġdetermines": 9019,
+      "Ġproj": 9020,
+      "ĠOverall": 9021,
+      "Ġble": 9022,
+      "fo": 9023,
+      "Ġprinciples": 9024,
+      "Ġinteracting": 9025,
+      "Ġhardware": 9026,
+      "life": 9027,
+      "ails": 9028,
+      "Ġdifficulty": 9029,
+      "Ġchoices": 9030,
+      "Ġcard": 9031,
+      "Ġlact": 9032,
+      "Ġroll": 9033,
+      "Ġquantified": 9034,
+      "ĠScientific": 9035,
+      "Ġlandsc": 9036,
+      "aligned": 9037,
+      "Ġcomposites": 9038,
+      "herichia": 9039,
+      "Ġenvelop": 9040,
+      "itig": 9041,
+      "Ste": 9042,
+      "Ġcompet": 9043,
+      "Ġimpairment": 9044,
+      "Ġclosure": 9045,
+      "Ġreturned": 9046,
+      "Ġreceiver": 9047,
+      "Ġpeer": 9048,
+      "Ġconsent": 9049,
+      "Ġultras": 9050,
+      "Ġphotons": 9051,
+      "Ġsuppose": 9052,
+      "Ġpredicting": 9053,
+      "ĠâĬķ": 9054,
+      "Ġcompan": 9055,
+      "Ġnegligible": 9056,
+      "current": 9057,
+      "umber": 9058,
+      "Ġcompatible": 9059,
+      "iop": 9060,
+      "ĠStructural": 9061,
+      "Ref": 9062,
+      "Ġson": 9063,
+      "Ġequality": 9064,
+      "Ġconsisted": 9065,
+      "Ġvibr": 9066,
+      "oupling": 9067,
+      "vation": 9068,
+      "Ġovercome": 9069,
+      "super": 9070,
+      "lict": 9071,
+      "Ġpancreatic": 9072,
+      "Gs": 9073,
+      "aped": 9074,
+      "asal": 9075,
+      "wan": 9076,
+      "Ġlatent": 9077,
+      "Ġcovering": 9078,
+      "Ġlesion": 9079,
+      "iance": 9080,
+      "ĠFT": 9081,
+      "wood": 9082,
+      "jecture": 9083,
+      "ĠBC": 9084,
+      "linked": 9085,
+      "ĠLaw": 9086,
+      "Ġemit": 9087,
+      "Ġunclear": 9088,
+      "Ġprem": 9089,
+      "acted": 9090,
+      "polar": 9091,
+      "cre": 9092,
+      "Ġmodulus": 9093,
+      "ropath": 9094,
+      "Sub": 9095,
+      "ami": 9096,
+      "Ġpick": 9097,
+      "ERR": 9098,
+      "Ġmovements": 9099,
+      "Ni": 9100,
+      "Ġmechanics": 9101,
+      "odic": 9102,
+      "Ġgal": 9103,
+      "ĠManagement": 9104,
+      "host": 9105,
+      "ewise": 9106,
+      "ĠTotal": 9107,
+      "ĠInfluence": 9108,
+      "Ġubiqu": 9109,
+      "rophys": 9110,
+      "Ġcaps": 9111,
+      "Ġparticipant": 9112,
+      "Ġpolyp": 9113,
+      "td": 9114,
+      "Ġiterations": 9115,
+      "dominal": 9116,
+      "BB": 9117,
+      "Ġcharacters": 9118,
+      "Ġdeviations": 9119,
+      "resistant": 9120,
+      "Ġmalaria": 9121,
+      "Ġremote": 9122,
+      "hskip": 9123,
+      "Ġunderwent": 9124,
+      "util": 9125,
+      "block": 9126,
+      "uclide": 9127,
+      "Φ": 9128,
+      "electron": 9129,
+      "Ġsensory": 9130,
+      "ĠSimulation": 9131,
+      "Ġreward": 9132,
+      "Ġpandemic": 9133,
+      "Ġbor": 9134,
+      "ynthetic": 9135,
+      "Ġinvasive": 9136,
+      "RF": 9137,
+      "ĠSmall": 9138,
+      "ĠFisher": 9139,
+      "valent": 9140,
+      "ĠMI": 9141,
+      "rocytes": 9142,
+      "ĠTE": 9143,
+      "Ġstre": 9144,
+      "Ġperturbations": 9145,
+      "Ġsimplicity": 9146,
+      "ĠGrowth": 9147,
+      "ĠÎł": 9148,
+      "Ġinoc": 9149,
+      "arding": 9150,
+      "atum": 9151,
+      "multi": 9152,
+      "ĠDiv": 9153,
+      "anes": 9154,
+      "acillus": 9155,
+      "Ġlifetime": 9156,
+      "ĠHep": 9157,
+      "Ġaz": 9158,
+      "usp": 9159,
+      "ĠAssume": 9160,
+      "Ġbreaking": 9161,
+      "ĠAtt": 9162,
+      "ticipants": 9163,
+      "Ġluminosity": 9164,
+      "Ġdonor": 9165,
+      "params": 9166,
+      "ohyd": 9167,
+      "Ġprogen": 9168,
+      "ĠPO": 9169,
+      "GO": 9170,
+      "ĠLeg": 9171,
+      "Ġbiomarkers": 9172,
+      "Ġrural": 9173,
+      "Ġneon": 9174,
+      "gluc": 9175,
+      "ĠPB": 9176,
+      "Ġguid": 9177,
+      "Ġcervical": 9178,
+      "pace": 9179,
+      "Ġcord": 9180,
+      "umn": 9181,
+      "Ġsubspace": 9182,
+      "Ġattached": 9183,
+      "Ġdeposited": 9184,
+      "Ġindicators": 9185,
+      "ĠSF": 9186,
+      "quire": 9187,
+      "Ġdissolved": 9188,
+      "rite": 9189,
+      "ĠNA": 9190,
+      "Ġju": 9191,
+      "Ġaddressed": 9192,
+      "Ġsuppressed": 9193,
+      "Ġpneumonia": 9194,
+      "Ġsession": 9195,
+      "ĠChe": 9196,
+      "ĠFer": 9197,
+      "Ġaccordance": 9198,
+      "Des": 9199,
+      "Ġquar": 9200,
+      "Ġfitness": 9201,
+      "Ġviability": 9202,
+      "osh": 9203,
+      "Ġphylogenetic": 9204,
+      "ectin": 9205,
+      "pat": 9206,
+      "ĠFrance": 9207,
+      "Ġmessages": 9208,
+      "Ġloci": 9209,
+      "Ġconflict": 9210,
+      "Ġrelevance": 9211,
+      "Ġinstructions": 9212,
+      "Ġsomewhat": 9213,
+      "changed": 9214,
+      "Ġcorrectly": 9215,
+      "ozyg": 9216,
+      "avig": 9217,
+      "ĠLat": 9218,
+      "Ġovarian": 9219,
+      "ĠRemark": 9220,
+      "joint": 9221,
+      "aint": 9222,
+      "west": 9223,
+      "sample": 9224,
+      "Ġdivergence": 9225,
+      "Ġhair": 9226,
+      "agonal": 9227,
+      "Ġmim": 9228,
+      "Ġimmediate": 9229,
+      "ĠPort": 9230,
+      "Ġoffers": 9231,
+      "Ġdepicted": 9232,
+      "Ġhydrox": 9233,
+      "ĠTow": 9234,
+      "Ġemerging": 9235,
+      "oupled": 9236,
+      "Ġhundred": 9237,
+      "Ġadapted": 9238,
+      "eller": 9239,
+      "ĠRelations": 9240,
+      "ette": 9241,
+      "Ġgastro": 9242,
+      "Ġmorphism": 9243,
+      "Ġequipment": 9244,
+      "pop": 9245,
+      "unately": 9246,
+      "Ġtransplantation": 9247,
+      "ifiers": 9248,
+      "Ġelderly": 9249,
+      "onucle": 9250,
+      "Ġrefers": 9251,
+      "arial": 9252,
+      "ĠCommittee": 9253,
+      "Ġmalignant": 9254,
+      "omonas": 9255,
+      "Ġallocation": 9256,
+      "ogether": 9257,
+      "Ġnanot": 9258,
+      "plot": 9259,
+      "ĠMes": 9260,
+      "Ġplanar": 9261,
+      "ells": 9262,
+      "source": 9263,
+      "owski": 9264,
+      "Ġna": 9265,
+      "Ġclock": 9266,
+      "Ġambient": 9267,
+      "ocene": 9268,
+      "Ġfluorescent": 9269,
+      "Ġvalu": 9270,
+      "ĠMagnetic": 9271,
+      "Ġdepart": 9272,
+      "phosphate": 9273,
+      "Ġroughly": 9274,
+      "Ġneither": 9275,
+      "ĠAltern": 9276,
+      "Ġstay": 9277,
+      "Ġspot": 9278,
+      "ĠEnt": 9279,
+      "Ġseconds": 9280,
+      "hard": 9281,
+      "Ġrecurrent": 9282,
+      "Ġpatch": 9283,
+      "Ġlimitation": 9284,
+      "ĠDer": 9285,
+      "Ġsharp": 9286,
+      "Ġexpectation": 9287,
+      "ĠLore": 9288,
+      "dict": 9289,
+      "Reg": 9290,
+      "Ġneutroph": 9291,
+      "Ġnur": 9292,
+      "Ġstarts": 9293,
+      "ostasis": 9294,
+      "Ġorganized": 9295,
+      "ĠcDNA": 9296,
+      "orient": 9297,
+      "ĠExample": 9298,
+      "ĠFund": 9299,
+      "aylor": 9300,
+      "idering": 9301,
+      "Ġtriple": 9302,
+      "nic": 9303,
+      "Ġattacks": 9304,
+      "ĠDros": 9305,
+      "è": 9306,
+      "ĠEM": 9307,
+      "Ġoptimum": 9308,
+      "Ġpull": 9309,
+      "Ġce": 9310,
+      "eryth": 9311,
+      "Ġrating": 9312,
+      "Ġreproductive": 9313,
+      "Ġdecades": 9314,
+      "Ġreplace": 9315,
+      "List": 9316,
+      "ĠFast": 9317,
+      "Ġredshift": 9318,
+      "opsy": 9319,
+      "illa": 9320,
+      "double": 9321,
+      "tera": 9322,
+      "Ġgoals": 9323,
+      "ĠSk": 9324,
+      "INE": 9325,
+      "Ġbiochemical": 9326,
+      "uint": 9327,
+      "Ġfetal": 9328,
+      "ĠRiemann": 9329,
+      "uries": 9330,
+      "Ġpp": 9331,
+      "Ġsymbols": 9332,
+      "ĠKa": 9333,
+      "Di": 9334,
+      "ĠGalax": 9335,
+      "ĠCompared": 9336,
+      "Ġcasc": 9337,
+      "Ġbits": 9338,
+      "Ġscaff": 9339,
+      "Ġestimator": 9340,
+      "ĠAdditional": 9341,
+      "Ġimprovements": 9342,
+      "ectives": 9343,
+      "Ġhous": 9344,
+      "ĠMagn": 9345,
+      "Ġmultivariate": 9346,
+      "Ġagric": 9347,
+      "vo": 9348,
+      "utter": 9349,
+      "ĠAcknowledg": 9350,
+      "su": 9351,
+      "Ġammon": 9352,
+      "Ġaims": 9353,
+      "Ġzinc": 9354,
+      "Ġelong": 9355,
+      "ĠGO": 9356,
+      "Question": 9357,
+      "including": 9358,
+      "LogP": 9359,
+      "Ġintellig": 9360,
+      "Ġcone": 9361,
+      "ĠFoundation": 9362,
+      "Ġimpaired": 9363,
+      "Ġillness": 9364,
+      "ĠEscherichia": 9365,
+      "Ġabundant": 9366,
+      "scal": 9367,
+      "ensively": 9368,
+      "Ġnegatively": 9369,
+      "parameter": 9370,
+      "Ġpermeability": 9371,
+      "domain": 9372,
+      "rated": 9373,
+      "Ġepoch": 9374,
+      "Ġadolescents": 9375,
+      "Ġdefic": 9376,
+      "ĠEstimation": 9377,
+      "Ġroutine": 9378,
+      "Per": 9379,
+      "tol": 9380,
+      "Ġelliptic": 9381,
+      "ĠHE": 9382,
+      "oblast": 9383,
+      "Ġreaches": 9384,
+      "Ġfluxes": 9385,
+      "Ġsun": 9386,
+      "ĠAnaly": 9387,
+      "âĢľ": 9388,
+      "ĠXLogP": 9389,
+      "Ġfiltering": 9390,
+      "rian": 9391,
+      "ĠScal": 9392,
+      "Ġpin": 9393,
+      "ĠTiO": 9394,
+      "iments": 9395,
+      "Ġmarginal": 9396,
+      "Ġrecombinant": 9397,
+      "Ġencour": 9398,
+      "Ġalumin": 9399,
+      "Ġtf": 9400,
+      "atalytic": 9401,
+      "Ġobservational": 9402,
+      "Ġgeneralization": 9403,
+      "Ġ": 9404,
+      "Ġantibiotic": 9405,
+      "Ġgenerates": 9406,
+      "ĠdB": 9407,
+      "Spec": 9408,
+      "rically": 9409,
+      "Ġvaluable": 9410,
+      "Ġtopic": 9411,
+      "Ġtermin": 9412,
+      "Ġsemicon": 9413,
+      "Ġquantification": 9414,
+      "ubb": 9415,
+      "Ġkinem": 9416,
+      "erring": 9417,
+      "Ġaeros": 9418,
+      "pack": 9419,
+      "Ġfewer": 9420,
+      "Ġfatigue": 9421,
+      "Ġgoes": 9422,
+      "Ġnight": 9423,
+      "ĠUs": 9424,
+      "âĢ¬": 9425,
+      "ĠPrinc": 9426,
+      "Ġspring": 9427,
+      "Ġconcerns": 9428,
+      "Ġsmart": 9429,
+      "Ġsecret": 9430,
+      "Ġmmol": 9431,
+      "Ġbelief": 9432,
+      "DC": 9433,
+      "Ġsubstantially": 9434,
+      "âĪĩ": 9435,
+      "Ġsubstitution": 9436,
+      "mapsto": 9437,
+      "sky": 9438,
+      "illance": 9439,
+      "Ġstudent": 9440,
+      "okine": 9441,
+      "Ġinterior": 9442,
+      "Ġeigenvalue": 9443,
+      "my": 9444,
+      "Ġcloser": 9445,
+      "erenti": 9446,
+      "Ġecological": 9447,
+      "ĠFigures": 9448,
+      "olytic": 9449,
+      "Ġarrays": 9450,
+      "ĠCas": 9451,
+      "Ġloops": 9452,
+      "Ġcorrected": 9453,
+      "Ġrhe": 9454,
+      "Ġinversion": 9455,
+      "Ġpreferred": 9456,
+      "umab": 9457,
+      "ĠDI": 9458,
+      "Ġadequate": 9459,
+      "irm": 9460,
+      "Ġimplicit": 9461,
+      "ship": 9462,
+      "Ġplayers": 9463,
+      "Ġdelayed": 9464,
+      "Ġwinter": 9465,
+      "Ġvulner": 9466,
+      "Ġshapes": 9467,
+      "Ġstained": 9468,
+      "ĠMajor": 9469,
+      "Ġhierarchical": 9470,
+      "ĠDig": 9471,
+      "ersion": 9472,
+      "ĠEfficient": 9473,
+      "Ġwalls": 9474,
+      "dfrac": 9475,
+      "Ġclassifier": 9476,
+      "Ġmonol": 9477,
+      "Ġupdated": 9478,
+      "Ġmature": 9479,
+      "ĠLI": 9480,
+      "earing": 9481,
+      "Ġfinger": 9482,
+      "ounter": 9483,
+      "ankton": 9484,
+      "While": 9485,
+      "Ġrealistic": 9486,
+      "ĠCamp": 9487,
+      "Ġfilled": 9488,
+      "Ġdead": 9489,
+      "ĠPacific": 9490,
+      "Ïĩ": 9491,
+      "ĠDavid": 9492,
+      "Ġadditive": 9493,
+      "enchymal": 9494,
+      "Ġobser": 9495,
+      "Ġstere": 9496,
+      "Ġultrasound": 9497,
+      "ĠPredic": 9498,
+      "Ġends": 9499,
+      "sectional": 9500,
+      "mas": 9501,
+      "omat": 9502,
+      "ivity": 9503,
+      "Ġhandle": 9504,
+      "Ġmetastatic": 9505,
+      "olet": 9506,
+      "ryp": 9507,
+      "ACE": 9508,
+      "Ġporous": 9509,
+      "Ġconcern": 9510,
+      "itored": 9511,
+      "Ġcircles": 9512,
+      "Ġemotional": 9513,
+      "gered": 9514,
+      "Ġfriction": 9515,
+      "first": 9516,
+      "ophy": 9517,
+      "escop": 9518,
+      "aded": 9519,
+      "Ġresolved": 9520,
+      "ERS": 9521,
+      "Ġpathogens": 9522,
+      "Ġgradually": 9523,
+      "ĠBrain": 9524,
+      "xf": 9525,
+      "anium": 9526,
+      "ael": 9527,
+      "New": 9528,
+      "Ġcytokine": 9529,
+      "ĠBP": 9530,
+      "Ġspecimen": 9531,
+      "olean": 9532,
+      "Ġtaxon": 9533,
+      "Ġsequential": 9534,
+      "κB": 9535,
+      "ademic": 9536,
+      "plings": 9537,
+      "~~": 9538,
+      "ermal": 9539,
+      "tree": 9540,
+      "Ġcausal": 9541,
+      "arian": 9542,
+      "Ġcrop": 9543,
+      "opol": 9544,
+      "channel": 9545,
+      "ĠMex": 9546,
+      "Ġclon": 9547,
+      "ĠRecently": 9548,
+      "ĠInvestig": 9549,
+      "Ġrecommendations": 9550,
+      "format": 9551,
+      "ĠMET": 9552,
+      "Ġsentence": 9553,
+      "Ġbp": 9554,
+      "ĠGW": 9555,
+      "Ġrecording": 9556,
+      "Ġple": 9557,
+      "totic": 9558,
+      "Ġ": 9559,
+      "Ġranged": 9560,
+      "ention": 9561,
+      "obacteria": 9562,
+      "ceptions": 9563,
+      "ĠImport": 9564,
+      "dynamic": 9565,
+      "porary": 9566,
+      "Given": 9567,
+      "Ġturbulence": 9568,
+      "Ġgram": 9569,
+      "Ġequally": 9570,
+      "cd": 9571,
+      "ĠOs": 9572,
+      "Ġturns": 9573,
+      "Ġdetecting": 9574,
+      "atio": 9575,
+      "generate": 9576,
+      "grade": 9577,
+      "Ġcirculation": 9578,
+      "Ġmanufacturer": 9579,
+      "La": 9580,
+      "ĠHilbert": 9581,
+      "Ts": 9582,
+      "integr": 9583,
+      "Ġbelongs": 9584,
+      "ĠInternet": 9585,
+      "angl": 9586,
+      "ĠâĬ¥": 9587,
+      "ĠDrosophila": 9588,
+      "uclidean": 9589,
+      "tan": 9590,
+      "Ġextends": 9591,
+      "Ġexpanded": 9592,
+      "illin": 9593,
+      "square": 9594,
+      "ysacchar": 9595,
+      "Ġquantify": 9596,
+      "Ġpulses": 9597,
+      "Ġvesic": 9598,
+      "ĠNK": 9599,
+      "orescence": 9600,
+      "ĠPhosph": 9601,
+      "Ġvision": 9602,
+      "ĠHuang": 9603,
+      "ĠResponse": 9604,
+      "house": 9605,
+      "ears": 9606,
+      "Ġeg": 9607,
+      "Ġaccepted": 9608,
+      "ĠTM": 9609,
+      "ametric": 9610,
+      "Ġpathological": 9611,
+      "Ġrecruitment": 9612,
+      "ATA": 9613,
+      "Ġfigures": 9614,
+      "ĠPress": 9615,
+      "Ġaligned": 9616,
+      "Ġpostoperative": 9617,
+      "ĠMeV": 9618,
+      "Ġconsiderably": 9619,
+      "Ġconformal": 9620,
+      "ĠIsland": 9621,
+      "number": 9622,
+      "Ġautomatic": 9623,
+      "Ġsplic": 9624,
+      "Ġcytos": 9625,
+      "Ġdescrip": 9626,
+      "ĠSant": 9627,
+      "lies": 9628,
+      "uity": 9629,
+      "itone": 9630,
+      "ECT": 9631,
+      "ĠBon": 9632,
+      "Ġdisapp": 9633,
+      "board": 9634,
+      "orrh": 9635,
+      "Ġcalculating": 9636,
+      "nee": 9637,
+      "ĠMeas": 9638,
+      "Ġgenomes": 9639,
+      "Ġphotoc": 9640,
+      "Ġreadily": 9641,
+      "ovine": 9642,
+      "ĠDev": 9643,
+      "Ġsatur": 9644,
+      "Ġkinds": 9645,
+      "ĠPK": 9646,
+      "Ġrod": 9647,
+      "Ġjunction": 9648,
+      "ĠHA": 9649,
+      "Ġdesigns": 9650,
+      "hn": 9651,
+      "Ġordering": 9652,
+      "Ġcosmological": 9653,
+      "Ġpilot": 9654,
+      "Ġcolorectal": 9655,
+      "ĠLondon": 9656,
+      "ĠDirac": 9657,
+      "Cont": 9658,
+      "ĠWind": 9659,
+      "ĠTre": 9660,
+      "idin": 9661,
+      "ĠïĢ«": 9662,
+      "iltration": 9663,
+      "Moreover": 9664,
+      "Ġretention": 9665,
+      "timately": 9666,
+      "hydrogen": 9667,
+      "del": 9668,
+      "bolic": 9669,
+      "ĠQuanti": 9670,
+      "period": 9671,
+      "Ġretrieval": 9672,
+      "atase": 9673,
+      "endicular": 9674,
+      "ulties": 9675,
+      "RS": 9676,
+      "NH": 9677,
+      "Ġinformed": 9678,
+      "Ġfiltered": 9679,
+      "membrane": 9680,
+      "Ġstiffness": 9681,
+      "ĠOcean": 9682,
+      "ĠSY": 9683,
+      "Ġlot": 9684,
+      "ĠFigs": 9685,
+      "Ġansw": 9686,
+      "ĠEngland": 9687,
+      "ĠAtlantic": 9688,
+      "processing": 9689,
+      "Ġdogs": 9690,
+      "Ġlie": 9691,
+      "Ġunion": 9692,
+      "ĠTan": 9693,
+      "Ġhalo": 9694,
+      "Ġcontinuously": 9695,
+      "Bu": 9696,
+      "AMP": 9697,
+      "ĠApp": 9698,
+      "Ġmoisture": 9699,
+      "Ġthyroid": 9700,
+      "Ġaccompanied": 9701,
+      "Ġfold": 9702,
+      "Ġoriginally": 9703,
+      "Ġspan": 9704,
+      "ĠFA": 9705,
+      "connected": 9706,
+      "Ġrecurs": 9707,
+      "vian": 9708,
+      "ĠEquations": 9709,
+      "ena": 9710,
+      "arcinoma": 9711,
+      "....": 9712,
+      "Ġdiscrep": 9713,
+      "UH": 9714,
+      "о": 9715,
+      "anger": 9716,
+      "Ġmonitored": 9717,
+      "Ġinfluenza": 9718,
+      "Ġsure": 9719,
+      "black": 9720,
+      "oe": 9721,
+      "Ġalloc": 9722,
+      "Ġhabitat": 9723,
+      "ophenyl": 9724,
+      "Ġventricular": 9725,
+      "Ġpolicies": 9726,
+      "amate": 9727,
+      "Ġreporting": 9728,
+      "Ġsoluble": 9729,
+      "================": 9730,
+      "Ġdipole": 9731,
+      "Ġirreducible": 9732,
+      "ĠPrec": 9733,
+      "acetyl": 9734,
+      "Ġthread": 9735,
+      "ĠApproxim": 9736,
+      "Ġmapped": 9737,
+      "ipro": 9738,
+      "Ġtropical": 9739,
+      "Sch": 9740,
+      "ĠANOVA": 9741,
+      "Ġlanguages": 9742,
+      "icine": 9743,
+      "ĠFamily": 9744,
+      "functions": 9745,
+      "EF": 9746,
+      "Ġnutrient": 9747,
+      "Ġanalyzing": 9748,
+      "inescence": 9749,
+      "Ġthromb": 9750,
+      "Ġkit": 9751,
+      "Ġmammalian": 9752,
+      "optotic": 9753,
+      "Ġequipped": 9754,
+      "ona": 9755,
+      "Ġque": 9756,
+      "Ġcame": 9757,
+      "Ġsimplified": 9758,
+      "Ġdecays": 9759,
+      "Ġpassive": 9760,
+      "Ġdeletion": 9761,
+      "Ġobtaining": 9762,
+      "Ġmixtures": 9763,
+      "Ġprimers": 9764,
+      "ĠPsy": 9765,
+      "osc": 9766,
+      "oment": 9767,
+      "Ġchloride": 9768,
+      "ĠPaul": 9769,
+      "start": 9770,
+      "intestinal": 9771,
+      "helium": 9772,
+      "arth": 9773,
+      "odot": 9774,
+      "Ġfits": 9775,
+      "Ġsquares": 9776,
+      "ĠCardi": 9777,
+      "aka": 9778,
+      "ributed": 9779,
+      "Ġinequalities": 9780,
+      "omething": 9781,
+      "hedral": 9782,
+      "ĠFuture": 9783,
+      "Ġgli": 9784,
+      "Ġmetallic": 9785,
+      "Ġfacilities": 9786,
+      "Ġobst": 9787,
+      "possible": 9788,
+      "Ġzones": 9789,
+      "ucid": 9790,
+      "Ġdrift": 9791,
+      "depend": 9792,
+      "valued": 9793,
+      "Ġnons": 9794,
+      "Ġworldwide": 9795,
+      "Ġtrust": 9796,
+      "Ġsole": 9797,
+      "ĠLevel": 9798,
+      "ĠSha": 9799,
+      "Ġregardless": 9800,
+      "Ġspectrometry": 9801,
+      "ductor": 9802,
+      "leuk": 9803,
+      "Ġskills": 9804,
+      "Ġincorporated": 9805,
+      "Ġlearned": 9806,
+      "Ġure": 9807,
+      "Ġextinc": 9808,
+      "ODU": 9809,
+      "Ġgrains": 9810,
+      "atern": 9811,
+      "ĠIndex": 9812,
+      "comput": 9813,
+      "ua": 9814,
+      "Ġcontamination": 9815,
+      "ĠAff": 9816,
+      "uning": 9817,
+      "Ġasymmetric": 9818,
+      "Ġopening": 9819,
+      "Ġbat": 9820,
+      "Ġagree": 9821,
+      "ITY": 9822,
+      "ĠChanges": 9823,
+      "organic": 9824,
+      "ĠRay": 9825,
+      "ĠHand": 9826,
+      "ni": 9827,
+      "inic": 9828,
+      "Ġrisks": 9829,
+      "Ġstock": 9830,
+      "Ġneck": 9831,
+      "Ġvolumes": 9832,
+      "ĠPrac": 9833,
+      "Ġincreasingly": 9834,
+      "Sc": 9835,
+      "oses": 9836,
+      "GFP": 9837,
+      "Ġassignment": 9838,
+      "ĠFed": 9839,
+      "ospit": 9840,
+      "Ġoverexpression": 9841,
+      "Ġmaster": 9842,
+      "Ġopt": 9843,
+      "iler": 9844,
+      "invariant": 9845,
+      "Ġconverges": 9846,
+      "Similar": 9847,
+      "ny": 9848,
+      "Ġstore": 9849,
+      "Ġelevation": 9850,
+      "Ġcoal": 9851,
+      "het": 9852,
+      "item": 9853,
+      "PLC": 9854,
+      "ohist": 9855,
+      "Gen": 9856,
+      "ĠChem": 9857,
+      "ĠCost": 9858,
+      "pair": 9859,
+      "Ġnumerically": 9860,
+      "Ġpreference": 9861,
+      "ĠNucle": 9862,
+      "ĠBD": 9863,
+      "TI": 9864,
+      "ĠHyp": 9865,
+      "roy": 9866,
+      "Te": 9867,
+      "ĠFin": 9868,
+      "Ġclaims": 9869,
+      "ibilities": 9870,
+      "Ġlarvae": 9871,
+      "ima": 9872,
+      "embly": 9873,
+      "Ġcit": 9874,
+      "LL": 9875,
+      "Ġsilica": 9876,
+      "ĠVI": 9877,
+      "Ġreaching": 9878,
+      "Of": 9879,
+      "ĠAustralian": 9880,
+      "tub": 9881,
+      "world": 9882,
+      "oni": 9883,
+      "ĠFP": 9884,
+      "Ġbriefly": 9885,
+      "ĠDescription": 9886,
+      "ζ": 9887,
+      "charg": 9888,
+      "Ġcis": 9889,
+      "ĠCat": 9890,
+      "Ġrecip": 9891,
+      "Ġemergency": 9892,
+      "Ġstrand": 9893,
+      "Ġrealized": 9894,
+      "posing": 9895,
+      "otope": 9896,
+      "Ġmaintaining": 9897,
+      "ĠChrist": 9898,
+      "Ġcreating": 9899,
+      "Ġembryos": 9900,
+      "Ġskeletal": 9901,
+      "Ġages": 9902,
+      "represent": 9903,
+      "Cr": 9904,
+      "Ġestimating": 9905,
+      "Ġrear": 9906,
+      "ĠYu": 9907,
+      "ĠPi": 9908,
+      "mg": 9909,
+      "Ġfloat": 9910,
+      "ĠRoy": 9911,
+      "pus": 9912,
+      "Ġchick": 9913,
+      "Ġmicrobiota": 9914,
+      "vasive": 9915,
+      "ĠBern": 9916,
+      "ĠPattern": 9917,
+      "lines": 9918,
+      "Ġflood": 9919,
+      "ĠLou": 9920,
+      "ilitary": 9921,
+      "rosion": 9922,
+      "Ġsurveys": 9923,
+      "FI": 9924,
+      "iae": 9925,
+      "Ġsearc": 9926,
+      "mol": 9927,
+      "Ġtitle": 9928,
+      "ĠMachine": 9929,
+      "Ġcircuits": 9930,
+      "ĠNumber": 9931,
+      "zi": 9932,
+      "ĠBMI": 9933,
+      "Ġautomated": 9934,
+      "plicate": 9935,
+      "ĠLPS": 9936,
+      "Ġelectrochemical": 9937,
+      "Ġwebsite": 9938,
+      "Ġanisotropy": 9939,
+      "Ġrings": 9940,
+      "Ġinnov": 9941,
+      "bits": 9942,
+      "win": 9943,
+      "ĠNAD": 9944,
+      "According": 9945,
+      "ĠConn": 9946,
+      "ureus": 9947,
+      "ĠFeature": 9948,
+      "ĠInstead": 9949,
+      "Comp": 9950,
+      "itudes": 9951,
+      "Mo": 9952,
+      "Ġscope": 9953,
+      "tification": 9954,
+      "ĠIS": 9955,
+      "ĠNeut": 9956,
+      "Ġregulating": 9957,
+      "coding": 9958,
+      "Ġrows": 9959,
+      "hl": 9960,
+      "ĠKn": 9961,
+      "istor": 9962,
+      "ampionship": 9963,
+      "Ġprominent": 9964,
+      "Ġrs": 9965,
+      "umatic": 9966,
+      "Am": 9967,
+      "Ġdifferentially": 9968,
+      "ugin": 9969,
+      "Ġadvance": 9970,
+      "phys": 9971,
+      "Ġsharing": 9972,
+      "Ġart": 9973,
+      "vacy": 9974,
+      "titions": 9975,
+      "Ġstyle": 9976,
+      "Figures": 9977,
+      "Ġglu": 9978,
+      "Ġvaccination": 9979,
+      "ĠOptical": 9980,
+      "fluid": 9981,
+      "ĠFre": 9982,
+      "Ġgradients": 9983,
+      "ophyl": 9984,
+      "ĠPubl": 9985,
+      "Ġaccretion": 9986,
+      "ĠâĢ²âĢ²": 9987,
+      "ressing": 9988,
+      "Ġtransmitted": 9989,
+      "Ġnervous": 9990,
+      "umar": 9991,
+      "Ġreviews": 9992,
+      "Ġgenotypes": 9993,
+      "lower": 9994,
+      "ĠEV": 9995,
+      "Ġcontract": 9996,
+      "atibility": 9997,
+      "Ġchildhood": 9998,
+      "Ġonc": 9999,
+      "Ġbiofil": 10000,
+      "Ġautophagy": 10001,
+      "Ġadsorb": 10002,
+      "ĠSupport": 10003,
+      "Ġligands": 10004,
+      "power": 10005,
+      "rectional": 10006,
+      "ĠRap": 10007,
+      "similar": 10008,
+      "Ġinfarc": 10009,
+      "Ġelectroly": 10010,
+      "Ġincome": 10011,
+      "arity": 10012,
+      "ĠAv": 10013,
+      "eric": 10014,
+      "Ġclinically": 10015,
+      "unch": 10016,
+      "Ġattribute": 10017,
+      "Ġcommand": 10018,
+      "ributions": 10019,
+      "Ġglyc": 10020,
+      "Ġtranscripts": 10021,
+      "ograms": 10022,
+      "Ġassessing": 10023,
+      "FO": 10024,
+      "scriptstyle": 10025,
+      "ji": 10026,
+      "rick": 10027,
+      "environment": 10028,
+      "Ġlaws": 10029,
+      "Ġnormally": 10030,
+      "Ġdepletion": 10031,
+      "ĠRO": 10032,
+      "Ġencoded": 10033,
+      "hma": 10034,
+      "Ġbranches": 10035,
+      "Ġargs": 10036,
+      "ounger": 10037,
+      "orge": 10038,
+      "umps": 10039,
+      "Ġviewed": 10040,
+      "Ġultr": 10041,
+      "RR": 10042,
+      "ulsion": 10043,
+      "ĠHor": 10044,
+      "Ġfro": 10045,
+      "ĠMeasurement": 10046,
+      "xx": 10047,
+      "erman": 10048,
+      "ĠOnce": 10049,
+      "Ġoriented": 10050,
+      "ĠPoint": 10051,
+      "Ġtown": 10052,
+      "Ġformulas": 10053,
+      "SY": 10054,
+      "ĠAM": 10055,
+      "Ġconsiderations": 10056,
+      "ĠTC": 10057,
+      "ĠKit": 10058,
+      "Ġactin": 10059,
+      "Ġplasmid": 10060,
+      "Ġhistorical": 10061,
+      "Ġdye": 10062,
+      "Ġheur": 10063,
+      "ĠLeague": 10064,
+      "ĠMad": 10065,
+      "Ġgraft": 10066,
+      "Ġsilver": 10067,
+      "Over": 10068,
+      "ĠCos": 10069,
+      "ographical": 10070,
+      "Ġprecursor": 10071,
+      "rus": 10072,
+      "Ġregarded": 10073,
+      "ĠHam": 10074,
+      "functional": 10075,
+      "iveness": 10076,
+      "fficiency": 10077,
+      "igene": 10078,
+      "ocol": 10079,
+      "Ġcumulative": 10080,
+      "Ġseasonal": 10081,
+      "Ġmu": 10082,
+      "ĠBan": 10083,
+      "omycin": 10084,
+      "Ġbool": 10085,
+      "ĠMag": 10086,
+      "ĠAnal": 10087,
+      "entia": 10088,
+      "aign": 10089,
+      "Ġfootball": 10090,
+      "acting": 10091,
+      "Ġreturns": 10092,
+      "ĠTom": 10093,
+      "shaped": 10094,
+      "itance": 10095,
+      "ĠExperiment": 10096,
+      "ĠOS": 10097,
+      "Ġabsent": 10098,
+      "ranial": 10099,
+      "Ġtherapies": 10100,
+      "Op": 10101,
+      "ounced": 10102,
+      "ATE": 10103,
+      "Value": 10104,
+      "green": 10105,
+      "Ġvegetation": 10106,
+      "Ds": 10107,
+      "Ġincom": 10108,
+      "ç": 10109,
+      "Ġmarrow": 10110,
+      "ĠCouncil": 10111,
+      "Ġinvest": 10112,
+      "Ġclub": 10113,
+      "Trans": 10114,
+      "device": 10115,
+      "Ġvibration": 10116,
+      "ĠXu": 10117,
+      "////////": 10118,
+      "ĠHen": 10119,
+      "vier": 10120,
+      "Ġanalogous": 10121,
+      "Ġdelta": 10122,
+      "Ġsaline": 10123,
+      "Ġrequiring": 10124,
+      "Ġneuron": 10125,
+      "oo": 10126,
+      "ĠQuality": 10127,
+      "Ġteac": 10128,
+      "ĠEc": 10129,
+      "Li": 10130,
+      "Ġpublication": 10131,
+      "ĠPhysics": 10132,
+      "Ġppm": 10133,
+      "thase": 10134,
+      "Ġcreation": 10135,
+      "ĠAge": 10136,
+      "Ġbelonging": 10137,
+      "Ġionic": 10138,
+      "ĠSI": 10139,
+      "uating": 10140,
+      "endif": 10141,
+      "ĠCour": 10142,
+      "а": 10143,
+      "Ġdots": 10144,
+      "Ġeast": 10145,
+      "arcom": 10146,
+      "Ġâĩ": 10147,
+      "Ġrights": 10148,
+      "essions": 10149,
+      "Ġversions": 10150,
+      "ĠFree": 10151,
+      "ĠStress": 10152,
+      "Ġsediments": 10153,
+      "Ġmitig": 10154,
+      "Ġbow": 10155,
+      "ĠAct": 10156,
+      "ĠCarbon": 10157,
+      "there": 10158,
+      "teen": 10159,
+      "Ġphenotypes": 10160,
+      "Ġnearest": 10161,
+      "ĠPotential": 10162,
+      "Ġdeform": 10163,
+      "Ġreflects": 10164,
+      "Ġpartners": 10165,
+      "Ġanest": 10166,
+      "Ġadvers": 10167,
+      "ĠFactor": 10168,
+      "Ġconvenient": 10169,
+      "ulos": 10170,
+      "ĠPur": 10171,
+      "ĠMer": 10172,
+      "Ġflag": 10173,
+      "Ġtriang": 10174,
+      "Ġseeds": 10175,
+      "Ġfif": 10176,
+      "obil": 10177,
+      "ĠCK": 10178,
+      "mentioned": 10179,
+      "Ġvapor": 10180,
+      "ogue": 10181,
+      "Ġpredictor": 10182,
+      "Out": 10183,
+      "Ġcompletion": 10184,
+      "ĠSeg": 10185,
+      "Ġdiffuse": 10186,
+      "Ġraised": 10187,
+      "Ġcoordination": 10188,
+      "Ġsynaptic": 10189,
+      "ĠBor": 10190,
+      "ĠBol": 10191,
+      "Ġpolymerase": 10192,
+      "Ġwheat": 10193,
+      "Ġinsertion": 10194,
+      "Ġesc": 10195,
+      "ĠWal": 10196,
+      "Ġdistal": 10197,
+      "transferase": 10198,
+      "Ġinterfaces": 10199,
+      "Ġinsu": 10200,
+      "Ġpoorly": 10201,
+      "Ġaureus": 10202,
+      "Ġbenz": 10203,
+      "Ġuniverse": 10204,
+      "ĠInteraction": 10205,
+      "ĠFrame": 10206,
+      "ĠImaging": 10207,
+      "Ġexploration": 10208,
+      "ĠEngineering": 10209,
+      "ĠBesides": 10210,
+      "tia": 10211,
+      "Ġenum": 10212,
+      "anine": 10213,
+      "Ġtot": 10214,
+      "ĠEduc": 10215,
+      "Ġderivation": 10216,
+      "Array": 10217,
+      "yloid": 10218,
+      "ĠArch": 10219,
+      "isen": 10220,
+      "acity": 10221,
+      "akers": 10222,
+      "Ġsheet": 10223,
+      "ĠEst": 10224,
+      "Ġwear": 10225,
+      "Ġeryth": 10226,
+      "ECK": 10227,
+      "hematics": 10228,
+      "Ġarterial": 10229,
+      "criptstyle": 10230,
+      "scriptscriptstyle": 10231,
+      "echanical": 10232,
+      "Ġparticipation": 10233,
+      "cher": 10234,
+      "urance": 10235,
+      "ĠFR": 10236,
+      "ĠCV": 10237,
+      "Ġcomplementary": 10238,
+      "aine": 10239,
+      "empty": 10240,
+      "Ġdiges": 10241,
+      "Ġexponent": 10242,
+      "Ġsimulate": 10243,
+      "UE": 10244,
+      "Ġantibiotics": 10245,
+      "ĠUnivers": 10246,
+      "Ġpathology": 10247,
+      "thermal": 10248,
+      "pa": 10249,
+      "Ġstresses": 10250,
+      "ĠLaboratory": 10251,
+      "Node": 10252,
+      "Ġleave": 10253,
+      "ashing": 10254,
+      "Ġdiscre": 10255,
+      "Ġsuspension": 10256,
+      "reek": 10257,
+      "Ġscheduling": 10258,
+      "ĠDA": 10259,
+      "aryn": 10260,
+      "ĠNaCl": 10261,
+      "strain": 10262,
+      "STR": 10263,
+      "ĠCong": 10264,
+      "olf": 10265,
+      "Ġcalibr": 10266,
+      "ĠOptimal": 10267,
+      "Ġó": 10268,
+      "Gl": 10269,
+      "ĠRh": 10270,
+      "Ġdifficulties": 10271,
+      "Ġvessels": 10272,
+      "Ġasymmetry": 10273,
+      "Ġcoherence": 10274,
+      "ĠTaxonomy": 10275,
+      "Ġped": 10276,
+      "ĠHouse": 10277,
+      "titudes": 10278,
+      "ĠFar": 10279,
+      "OY": 10280,
+      "Ġconcentrated": 10281,
+      "Ġsignalling": 10282,
+      "Ġfungal": 10283,
+      "Ġconsistently": 10284,
+      "Ġenhances": 10285,
+      "Ġforecast": 10286,
+      "Ġcubic": 10287,
+      "ĠEP": 10288,
+      "Ġparticipate": 10289,
+      "ĠPlant": 10290,
+      "risk": 10291,
+      "And": 10292,
+      "adic": 10293,
+      "oflu": 10294,
+      "Ġsperm": 10295,
+      "ĠChris": 10296,
+      "ND": 10297,
+      "colon": 10298,
+      "Ġfaces": 10299,
+      "Ġtuberculosis": 10300,
+      "rystal": 10301,
+      "floor": 10302,
+      "ups": 10303,
+      "Ġgray": 10304,
+      "ĠPublic": 10305,
+      "tensor": 10306,
+      "Ġrigid": 10307,
+      "Ġeastern": 10308,
+      "ĠItaly": 10309,
+      "Ġsignatures": 10310,
+      "Ġshallow": 10311,
+      "ón": 10312,
+      "ĠCe": 10313,
+      "Ġprojects": 10314,
+      "Ġrouting": 10315,
+      "Ġpredicts": 10316,
+      "ĠFeatures": 10317,
+      "ĠDistrict": 10318,
+      "Ġcarrying": 10319,
+      "ĉĠĠĠĠ": 10320,
+      "ĠTO": 10321,
+      "HM": 10322,
+      "dings": 10323,
+      "Ġrenormal": 10324,
+      "Ġbring": 10325,
+      "pin": 10326,
+      "aled": 10327,
+      "Ġclouds": 10328,
+      "names": 10329,
+      "oxin": 10330,
+      "Ġperpendicular": 10331,
+      "WT": 10332,
+      "ership": 10333,
+      "Ġrecon": 10334,
+      "Ġworked": 10335,
+      "ĠâĢ«": 10336,
+      "rastructure": 10337,
+      "Ġpointed": 10338,
+      "EV": 10339,
+      "ĠTaylor": 10340,
+      "Ġhepatitis": 10341,
+      "Ġorbits": 10342,
+      "ĠFactors": 10343,
+      "cellular": 10344,
+      "Ġfocal": 10345,
+      "Ġboost": 10346,
+      "Ġmicrowave": 10347,
+      "ĠProject": 10348,
+      "BF": 10349,
+      "Ġpolitical": 10350,
+      "Ġsupplemented": 10351,
+      "Ġillustrates": 10352,
+      "Ġideas": 10353,
+      "ĠDrug": 10354,
+      "obile": 10355,
+      "ĠHO": 10356,
+      "Ġrobustness": 10357,
+      "rosine": 10358,
+      "ĠNormal": 10359,
+      "Ġstimulated": 10360,
+      "Ġimpedance": 10361,
+      "fortunately": 10362,
+      "zyme": 10363,
+      "Ġbarriers": 10364,
+      "actory": 10365,
+      "learly": 10366,
+      "Ġpreprint": 10367,
+      "sensitive": 10368,
+      "Ġturbulent": 10369,
+      "thing": 10370,
+      "Ġboard": 10371,
+      "Ġpit": 10372,
+      "Ġintegrity": 10373,
+      "Ġrotating": 10374,
+      "uda": 10375,
+      "Ġventi": 10376,
+      "ĠSNPs": 10377,
+      "Ġcorrespondence": 10378,
+      "Ġvisualization": 10379,
+      "avail": 10380,
+      "Ġbeams": 10381,
+      "ĠContinu": 10382,
+      "Ġpersistent": 10383,
+      "Ġbath": 10384,
+      "ĠmiRNAs": 10385,
+      "Ġcustom": 10386,
+      "Ġordinary": 10387,
+      "Ġgenerators": 10388,
+      "Ġbridge": 10389,
+      "Ġdomin": 10390,
+      "amy": 10391,
+      "Ġlooking": 10392,
+      "table": 10393,
+      "False": 10394,
+      "Ġsoils": 10395,
+      "Ġmatches": 10396,
+      "Ġprogressive": 10397,
+      "states": 10398,
+      "ĠShort": 10399,
+      "Ġcores": 10400,
+      "Ġintroducing": 10401,
+      "Ġarrest": 10402,
+      "Ġtexture": 10403,
+      "Ġdorsal": 10404,
+      "Ġdrain": 10405,
+      "izoph": 10406,
+      "ĠQue": 10407,
+      "ñ": 10408,
+      "disc": 10409,
+      "Index": 10410,
+      "Ġextensively": 10411,
+      "Ġplasticity": 10412,
+      "Ġreally": 10413,
+      "ĠError": 10414,
+      "Ġsugges": 10415,
+      "Ġconsequently": 10416,
+      "Ġperforms": 10417,
+      "likely": 10418,
+      "ivered": 10419,
+      "Ġthermodynamic": 10420,
+      "Ġker": 10421,
+      "Ġacetate": 10422,
+      "Ġgets": 10423,
+      "leqslant": 10424,
+      "Ġpredictors": 10425,
+      "ĠSwed": 10426,
+      "nan": 10427,
+      "heter": 10428,
+      "Ġanomaly": 10429,
+      "Ġoperational": 10430,
+      "Ġretrospective": 10431,
+      "Ġtends": 10432,
+      "aden": 10433,
+      "Ġborder": 10434,
+      "Ġmethanol": 10435,
+      "ĠEnter": 10436,
+      "Ġcollapse": 10437,
+      "Ġpurchased": 10438,
+      "Da": 10439,
+      "ĠHT": 10440,
+      "Ġfulf": 10441,
+      "Ġcrust": 10442,
+      "stone": 10443,
+      "Ġpenal": 10444,
+      "Ġtunn": 10445,
+      "ĠTemperature": 10446,
+      "Ġpotent": 10447,
+      "lecule": 10448,
+      "Ġcovers": 10449,
+      "Ġbattery": 10450,
+      "Ġbeg": 10451,
+      "Ġorgans": 10452,
+      "ĠThomas": 10453,
+      "Ġsolub": 10454,
+      "ocrine": 10455,
+      "ĠSpin": 10456,
+      "Ġinterests": 10457,
+      "doc": 10458,
+      "Ġundergoing": 10459,
+      "ui": 10460,
+      "Ġinherent": 10461,
+      "Ġintegrals": 10462,
+      "irable": 10463,
+      "ashi": 10464,
+      "Ġregeneration": 10465,
+      "Ġinflation": 10466,
+      "manif": 10467,
+      "ĠRecognition": 10468,
+      "Ġdisplays": 10469,
+      "Another": 10470,
+      "Ġcontamin": 10471,
+      "junction": 10472,
+      "Ġcopies": 10473,
+      "MRI": 10474,
+      "Ġvehicles": 10475,
+      "Get": 10476,
+      "Ġperhaps": 10477,
+      "Ġwest": 10478,
+      "Ġintensive": 10479,
+      "Ġsomething": 10480,
+      "Ġhypoxia": 10481,
+      "Ġcouplings": 10482,
+      "Ġfeasibility": 10483,
+      "azine": 10484,
+      "unic": 10485,
+      "iner": 10486,
+      "ĠIT": 10487,
+      "Ġdistrict": 10488,
+      "ĠJames": 10489,
+      "eval": 10490,
+      "Ġplacebo": 10491,
+      "aque": 10492,
+      "Ġelucid": 10493,
+      "ĠJacob": 10494,
+      "Ġcounting": 10495,
+      "Ġflexibility": 10496,
+      "Ġperman": 10497,
+      "Ġadvances": 10498,
+      "ulph": 10499,
+      "Ġentanglement": 10500,
+      "Ġintegers": 10501,
+      "Ġfocusing": 10502,
+      "kov": 10503,
+      "Ġhospit": 10504,
+      "Ġapplies": 10505,
+      "Ġcot": 10506,
+      "Sm": 10507,
+      "assium": 10508,
+      "Ġdocumented": 10509,
+      "Ġloaded": 10510,
+      "Ġrely": 10511,
+      "Ġinfectious": 10512,
+      "Ġprobes": 10513,
+      "Ġhighlighted": 10514,
+      "Ġpediatric": 10515,
+      "Ġweather": 10516,
+      "Ġmanual": 10517,
+      "Ġcation": 10518,
+      "Ġinterpolation": 10519,
+      "ĠStep": 10520,
+      "ĠKal": 10521,
+      "DH": 10522,
+      "db": 10523,
+      "izophren": 10524,
+      "ader": 10525,
+      "carb": 10526,
+      "Ġagon": 10527,
+      "orphous": 10528,
+      "tors": 10529,
+      "atz": 10530,
+      "Ġbif": 10531,
+      "Ġcharges": 10532,
+      "ĠAgain": 10533,
+      "Ġbron": 10534,
+      "ĠGover": 10535,
+      "Ġmining": 10536,
+      "aver": 10537,
+      "Ġearthqu": 10538,
+      "Ġviews": 10539,
+      "Ġscene": 10540,
+      "parameters": 10541,
+      "Ġbroken": 10542,
+      "Test": 10543,
+      "ĠSum": 10544,
+      "ĠProm": 10545,
+      "ÎĽ": 10546,
+      "Ġcutoff": 10547,
+      "Ġbirds": 10548,
+      "Ġarising": 10549,
+      "ĠAI": 10550,
+      "ĠCE": 10551,
+      "Ġpronounced": 10552,
+      "aspase": 10553,
+      "Ġintended": 10554,
+      "Ġaffine": 10555,
+      "Ġurine": 10556,
+      "Ġbelieved": 10557,
+      "ĠPrimary": 10558,
+      "ĠConf": 10559,
+      "Ġabdominal": 10560,
+      "spin": 10561,
+      "uniform": 10562,
+      "ĠStochastic": 10563,
+      "ĠProv": 10564,
+      "ĠmiRNA": 10565,
+      "ĠBell": 10566,
+      "BO": 10567,
+      "ĠSoftware": 10568,
+      "ĠTs": 10569,
+      "utri": 10570,
+      "icking": 10571,
+      "ien": 10572,
+      "Ġmicros": 10573,
+      "ĠNR": 10574,
+      "Ġleukemia": 10575,
+      "Ġsupernat": 10576,
+      "family": 10577,
+      "Ġalloys": 10578,
+      "ĠPET": 10579,
+      "ĠAbs": 10580,
+      "ĠGA": 10581,
+      "ĠQuantitative": 10582,
+      "Lo": 10583,
+      "Ġisland": 10584,
+      "second": 10585,
+      "pectives": 10586,
+      "Ġlatency": 10587,
+      "angi": 10588,
+      "Ġflight": 10589,
+      "ĠEuclidean": 10590,
+      "emy": 10591,
+      "ĠBlood": 10592,
+      "leukin": 10593,
+      "LT": 10594,
+      "enh": 10595,
+      "Ġswe": 10596,
+      "Ġunitary": 10597,
+      "ĠRepublic": 10598,
+      "Ġstructured": 10599,
+      "ĠSen": 10600,
+      "Mn": 10601,
+      "centric": 10602,
+      "Ġtransgenic": 10603,
+      "Ġhelpful": 10604,
+      "pyx": 10605,
+      "Ġhomeostasis": 10606,
+      "Na": 10607,
+      "Ġpassed": 10608,
+      "Ġeyes": 10609,
+      "Ġabstract": 10610,
+      "ulse": 10611,
+      "Ġmirror": 10612,
+      "Ġregulator": 10613,
+      "Ġmurine": 10614,
+      "loaded": 10615,
+      "Ġmodular": 10616,
+      "Ġlandscape": 10617,
+      "icks": 10618,
+      "Ġsnow": 10619,
+      "Ġbovine": 10620,
+      "elli": 10621,
+      "Ġdatabases": 10622,
+      "Ġoutbreak": 10623,
+      "larg": 10624,
+      "ĠRun": 10625,
+      "BE": 10626,
+      "Ġsurprising": 10627,
+      "Ġacceptable": 10628,
+      "Ġrotational": 10629,
+      "pg": 10630,
+      "FE": 10631,
+      "wik": 10632,
+      "Ġyounger": 10633,
+      "ashion": 10634,
+      "Ġmicroscopic": 10635,
+      "regation": 10636,
+      "Ġfibr": 10637,
+      "ĠPlan": 10638,
+      "Ġhapl": 10639,
+      "Ġmanifolds": 10640,
+      "Ġoutper": 10641,
+      "Ġchoosing": 10642,
+      "eper": 10643,
+      "ĠkeV": 10644,
+      "ĠTyp": 10645,
+      "pread": 10646,
+      "ntz": 10647,
+      "ĠReport": 10648,
+      "ĠMatrix": 10649,
+      "Ġintu": 10650,
+      "Ġproperly": 10651,
+      "ogly": 10652,
+      "oscopic": 10653,
+      "ĠAMP": 10654,
+      "ĠBM": 10655,
+      "Ġelementary": 10656,
+      "keleton": 10657,
+      "Ġsynthase": 10658,
+      "Ġionization": 10659,
+      "bes": 10660,
+      "ophage": 10661,
+      "duces": 10662,
+      "acco": 10663,
+      "Ġprotect": 10664,
+      "ĠCoul": 10665,
+      "Ġspent": 10666,
+      "Ġmand": 10667,
+      "Ġhind": 10668,
+      "fluor": 10669,
+      "ĠGood": 10670,
+      "Ġdoing": 10671,
+      "Object": 10672,
+      "ducts": 10673,
+      "oyl": 10674,
+      "chiatric": 10675,
+      "Ġov": 10676,
+      "cel": 10677,
+      "Ġbases": 10678,
+      "Ġmitochondria": 10679,
+      "pted": 10680,
+      "artz": 10681,
+      "Ġbrown": 10682,
+      "Ġequals": 10683,
+      "tible": 10684,
+      "Ġopportunity": 10685,
+      "azol": 10686,
+      "Ġofficial": 10687,
+      "ailed": 10688,
+      "Ġurinary": 10689,
+      "ĠHan": 10690,
+      "Be": 10691,
+      "result": 10692,
+      "units": 10693,
+      "Ġbad": 10694,
+      "ĠString": 10695,
+      "izable": 10696,
+      "condition": 10697,
+      "ĠElectron": 10698,
+      "immune": 10699,
+      "ĠME": 10700,
+      "hao": 10701,
+      "Σ": 10702,
+      "ĠMAT": 10703,
+      "Ġadopt": 10704,
+      "Ġelic": 10705,
+      "Ġshr": 10706,
+      "Ġproximal": 10707,
+      "FD": 10708,
+      "ĠSS": 10709,
+      "Ġentirely": 10710,
+      "esium": 10711,
+      "ĠEEG": 10712,
+      "Ġpaired": 10713,
+      "ĠTP": 10714,
+      "ĠDO": 10715,
+      "NAL": 10716,
+      "idespread": 10717,
+      "Ġmoves": 10718,
+      "site": 10719,
+      "Ġrain": 10720,
+      "Ġlap": 10721,
+      "ĠFu": 10722,
+      "ĠMeta": 10723,
+      "ircraft": 10724,
+      "Ġmagnetization": 10725,
+      "operation": 10726,
+      "Ġprost": 10727,
+      "Step": 10728,
+      "Ġsubgroups": 10729,
+      "ĠSouthern": 10730,
+      "Ġathe": 10731,
+      "luor": 10732,
+      "ĠTaxonomic": 10733,
+      "ĠEinstein": 10734,
+      "Ġrace": 10735,
+      "ĠKen": 10736,
+      "Ġattempts": 10737,
+      "Ġcosmic": 10738,
+      "ĠDop": 10739,
+      "Ġfixation": 10740,
+      "Ġremoving": 10741,
+      "BT": 10742,
+      "Ġlimb": 10743,
+      "Ġalign": 10744,
+      "Ġdried": 10745,
+      "du": 10746,
+      "Ġputative": 10747,
+      "uccess": 10748,
+      "pert": 10749,
+      "Ġslowly": 10750,
+      "also": 10751,
+      "olip": 10752,
+      "Ġclient": 10753,
+      "Ġbasin": 10754,
+      "Ġsusceptible": 10755,
+      "Ġcoming": 10756,
+      "nson": 10757,
+      "ĠNGC": 10758,
+      "assert": 10759,
+      "Ġtensile": 10760,
+      "Ġarises": 10761,
+      "cutaneous": 10762,
+      "Ġcaro": 10763,
+      "Bi": 10764,
+      "Ġdiscussions": 10765,
+      "Ġabnormalities": 10766,
+      "Ġpollution": 10767,
+      "ĠAx": 10768,
+      "Ġloads": 10769,
+      "Do": 10770,
+      "iao": 10771,
+      "Ġmedication": 10772,
+      "Ġintact": 10773,
+      "ĠCX": 10774,
+      "Ġbreeding": 10775,
+      "ĠUnion": 10776,
+      "ĠBat": 10777,
+      "ĠParticipants": 10778,
+      "ĠRegulation": 10779,
+      "Ġcontradiction": 10780,
+      "Ġintensities": 10781,
+      "encephal": 10782,
+      "rile": 10783,
+      "ĠTLR": 10784,
+      "Ġredund": 10785,
+      "Ġpersons": 10786,
+      "ĠArc": 10787,
+      "solid": 10788,
+      "law": 10789,
+      "Results": 10790,
+      "ilic": 10791,
+      "zone": 10792,
+      "ocytosis": 10793,
+      "Ġtriangle": 10794,
+      "STM": 10795,
+      "ĠVirus": 10796,
+      "Ġaid": 10797,
+      "soft": 10798,
+      "Ġsoon": 10799,
+      "expected": 10800,
+      "Ġanch": 10801,
+      "ĠMu": 10802,
+      "ĠSr": 10803,
+      "ĠLO": 10804,
+      "Ġcry": 10805,
+      "Ġupstream": 10806,
+      "oxic": 10807,
+      "mathit": 10808,
+      "ĠKle": 10809,
+      "Ġisotropic": 10810,
+      "Ġspatially": 10811,
+      "ĠHard": 10812,
+      "Ġextr": 10813,
+      "bas": 10814,
+      "eor": 10815,
+      "ivil": 10816,
+      "yan": 10817,
+      "Ġshifted": 10818,
+      "Ġbiopsy": 10819,
+      "Ġfeel": 10820,
+      "glut": 10821,
+      "Size": 10822,
+      "Ġerg": 10823,
+      "ĠTer": 10824,
+      "Ġdeaths": 10825,
+      "borne": 10826,
+      "Ġrelativistic": 10827,
+      "ĠVEGF": 10828,
+      "atab": 10829,
+      "spring": 10830,
+      "restim": 10831,
+      "ĠSearch": 10832,
+      "yphenyl": 10833,
+      "ecal": 10834,
+      "urc": 10835,
+      "Ġlamin": 10836,
+      "Ġserial": 10837,
+      "las": 10838,
+      "ĠProduction": 10839,
+      "Ġsocio": 10840,
+      "Ġmodify": 10841,
+      "ĠService": 10842,
+      "Ġbary": 10843,
+      "Ġradiative": 10844,
+      "bigl": 10845,
+      "Ġparadigm": 10846,
+      "patient": 10847,
+      "Ġspp": 10848,
+      "phone": 10849,
+      "Ġî": 10850,
+      "Ġrocks": 10851,
+      "ĠMartin": 10852,
+      "mn": 10853,
+      "Ġfluids": 10854,
+      "ĠINTR": 10855,
+      "ods": 10856,
+      "Ġdivis": 10857,
+      "Consider": 10858,
+      "component": 10859,
+      "Ġanomalies": 10860,
+      "Ġknee": 10861,
+      "ĠRelationship": 10862,
+      "aud": 10863,
+      "Ġovernight": 10864,
+      "Ġrainf": 10865,
+      "Ġannealing": 10866,
+      "Ġtreating": 10867,
+      "Ġcoarse": 10868,
+      "Model": 10869,
+      "Ġpose": 10870,
+      "Ġoccas": 10871,
+      "ĠWilliam": 10872,
+      "oor": 10873,
+      "Ġadjustment": 10874,
+      "ĠFunctions": 10875,
+      "imeter": 10876,
+      "Ġdetectors": 10877,
+      "Ġinstitutional": 10878,
+      "Ġthroughput": 10879,
+      "ividual": 10880,
+      "Ġentities": 10881,
+      "Ġprolonged": 10882,
+      "Ġship": 10883,
+      "Ġpreserved": 10884,
+      "ODUCTION": 10885,
+      "Ġlogistic": 10886,
+      "ĠPrediction": 10887,
+      "tized": 10888,
+      "ĠOrig": 10889,
+      "ĠHem": 10890,
+      "onomous": 10891,
+      "################": 10892,
+      "ĠGeneration": 10893,
+      "bottom": 10894,
+      "ĠKnow": 10895,
+      "clinical": 10896,
+      "Ġtrauma": 10897,
+      "Ġiterative": 10898,
+      "Ġfacility": 10899,
+      "ront": 10900,
+      "ĠBus": 10901,
+      "Ġretinal": 10902,
+      "Ġconduction": 10903,
+      "Ġchecked": 10904,
+      "Ġcalls": 10905,
+      "ologists": 10906,
+      "CON": 10907,
+      "ĠSciences": 10908,
+      "Ġnonzero": 10909,
+      "Ġbrack": 10910,
+      "Ġmelting": 10911,
+      "Ġasc": 10912,
+      "Ġmention": 10913,
+      "ĠBL": 10914,
+      "Ġverification": 10915,
+      "ukary": 10916,
+      "ĠSpatial": 10917,
+      "ĠGram": 10918,
+      "Ġplaces": 10919,
+      "Ġnecrosis": 10920,
+      "ĠChildren": 10921,
+      "Ġdelivered": 10922,
+      "Ġresection": 10923,
+      "Ġdeterministic": 10924,
+      "Section": 10925,
+      "Ġmultim": 10926,
+      "DF": 10927,
+      "Ġhypotheses": 10928,
+      "Ġraise": 10929,
+      "Ġseismic": 10930,
+      "Ġlam": 10931,
+      "ĠHCC": 10932,
+      "bigr": 10933,
+      "Ġhealing": 10934,
+      "isy": 10935,
+      "Ġoptimize": 10936,
+      "obacterium": 10937,
+      "edy": 10938,
+      "Ġtruth": 10939,
+      "Ġspacetime": 10940,
+      "Ġchromatin": 10941,
+      "Ġdomestic": 10942,
+      "Ġrecru": 10943,
+      "ĠJose": 10944,
+      "ĠThermal": 10945,
+      "Ġenvelope": 10946,
+      "vable": 10947,
+      "Ġincons": 10948,
+      "Ġnod": 10949,
+      "и": 10950,
+      "Ġcontributing": 10951,
+      "Ġguarantee": 10952,
+      "ĠPhen": 10953,
+      "Ġrab": 10954,
+      "Man": 10955,
+      "Ġsurveillance": 10956,
+      "Ġthings": 10957,
+      "Ġprev": 10958,
+      "ĠNonlinear": 10959,
+      "Ġgaps": 10960,
+      "aya": 10961,
+      "ĠCri": 10962,
+      "Ġcrystalline": 10963,
+      "strict": 10964,
+      "Ġcomputations": 10965,
+      "Ġunable": 10966,
+      "habil": 10967,
+      "umina": 10968,
+      "Ġpromoting": 10969,
+      "egrad": 10970,
+      "Ġregister": 10971,
+      "Ġcrossing": 10972,
+      "ulators": 10973,
+      "ĠLanguage": 10974,
+      "ĠAA": 10975,
+      "Ġiner": 10976,
+      "ĠLV": 10977,
+      "osan": 10978,
+      "Ġcoastal": 10979,
+      "Ġbiod": 10980,
+      "ĠMOD": 10981,
+      "Ġneighbour": 10982,
+      "Ġpredominantly": 10983,
+      "ĠNewton": 10984,
+      "ĠStrateg": 10985,
+      "being": 10986,
+      "Ġì": 10987,
+      "Ġcapabilities": 10988,
+      "Ġunless": 10989,
+      "formal": 10990,
+      "Ġvessel": 10991,
+      "bmatrix": 10992,
+      "ESS": 10993,
+      "Ġrainfall": 10994,
+      "ã": 10995,
+      "Ġprepar": 10996,
+      "axial": 10997,
+      "Ġdental": 10998,
+      "ĠProte": 10999,
+      "Ġworse": 11000,
+      "doped": 11001,
+      "hentic": 11002,
+      "Ġvalidate": 11003,
+      "Zn": 11004,
+      "Ġspecification": 11005,
+      "si": 11006,
+      "ĠAng": 11007,
+      "Ġtubes": 11008,
+      "ulic": 11009,
+      "ĠAny": 11010,
+      "ĠMap": 11011,
+      "Ġfabricated": 11012,
+      "Ġforced": 11013,
+      "ĠWilson": 11014,
+      "olysis": 11015,
+      "ĠWave": 11016,
+      "ĠCast": 11017,
+      "Ġasthma": 11018,
+      "Ġperi": 11019,
+      "ĠCyt": 11020,
+      "asty": 11021,
+      "Ġsky": 11022,
+      "rupt": 11023,
+      "Dec": 11024,
+      "Ġmelanoma": 11025,
+      "PER": 11026,
+      "Ġcontinuity": 11027,
+      "Box": 11028,
+      "system": 11029,
+      "Ġnavig": 11030,
+      "Ġcirculating": 11031,
+      "Ġcolony": 11032,
+      "lesssim": 11033,
+      "adium": 11034,
+      "Ġtetra": 11035,
+      "Ġaccounts": 11036,
+      "Ġpresenting": 11037,
+      "ĠLik": 11038,
+      "Ġresis": 11039,
+      "Ġdamping": 11040,
+      "ĠGly": 11041,
+      "ĠNeuro": 11042,
+      "user": 11043,
+      "Ġcapital": 11044,
+      "urate": 11045,
+      "ĠMW": 11046,
+      "Ġcorrelates": 11047,
+      "ĠGib": 11048,
+      "Ġhappens": 11049,
+      "Ġgall": 11050,
+      "ĠWithin": 11051,
+      "Ġcombine": 11052,
+      "Ġsinus": 11053,
+      "ĠKin": 11054,
+      "********************************": 11055,
+      "Map": 11056,
+      "Ġmaturation": 11057,
+      "Ġblocking": 11058,
+      "ĠCloud": 11059,
+      "Ġcontacts": 11060,
+      "Ġsac": 11061,
+      "ALL": 11062,
+      "ĠRab": 11063,
+      "zz": 11064,
+      "utch": 11065,
+      "Ġcarriers": 11066,
+      "ĠSNR": 11067,
+      "erb": 11068,
+      "Ġprotected": 11069,
+      "racking": 11070,
+      "radient": 11071,
+      "Ġattractive": 11072,
+      "Ġlag": 11073,
+      "Ġopin": 11074,
+      "ĠGi": 11075,
+      "Ġdefense": 11076,
+      "Ġtuning": 11077,
+      "Ġelectroph": 11078,
+      "Ġgreatest": 11079,
+      "Ġreconstructed": 11080,
+      "ĠPopulation": 11081,
+      "MAP": 11082,
+      "Ġwrote": 11083,
+      "AND": 11084,
+      "economic": 11085,
+      "ĠMichael": 11086,
+      "ĠBlock": 11087,
+      "Ġvo": 11088,
+      "oprop": 11089,
+      "Ġprofiling": 11090,
+      "ootst": 11091,
+      "ĠAsian": 11092,
+      "Ġoscillation": 11093,
+      "ĠâĨIJ": 11094,
+      "UD": 11095,
+      "Ġsigned": 11096,
+      "ĠEuler": 11097,
+      "ĠComparative": 11098,
+      "ĠWhere": 11099,
+      "ĠJack": 11100,
+      "Ġpassing": 11101,
+      "Ġvillage": 11102,
+      "Ġau": 11103,
+      "ĠNorthern": 11104,
+      "essage": 11105,
+      "matic": 11106,
+      "Ġaffili": 11107,
+      "ĠFac": 11108,
+      "Ġoverlapping": 11109,
+      "shell": 11110,
+      "Ġobstac": 11111,
+      "Ġbecoming": 11112,
+      "entive": 11113,
+      "Ġeasier": 11114,
+      "initely": 11115,
+      "Ġcentered": 11116,
+      "Ġacademic": 11117,
+      "annels": 11118,
+      "Ġirregular": 11119,
+      "Ġprojections": 11120,
+      "Ġproposition": 11121,
+      "Ġdiscrimination": 11122,
+      "Ġremod": 11123,
+      "Ġshoot": 11124,
+      "month": 11125,
+      "essor": 11126,
+      "Ġdiffers": 11127,
+      "ĠTV": 11128,
+      "ĠZhou": 11129,
+      "Ġinher": 11130,
+      "Ġmachines": 11131,
+      "Ġmell": 11132,
+      "Ġconjugate": 11133,
+      "Ġcoc": 11134,
+      "una": 11135,
+      "anyl": 11136,
+      "Ġoffic": 11137,
+      "Ġopportunities": 11138,
+      "Ġvein": 11139,
+      "ĠCharacteristics": 11140,
+      "Ġpathogenic": 11141,
+      "OYSA": 11142,
+      "ĠParkinson": 11143,
+      "ĠGalactic": 11144,
+      "FFFA": 11145,
+      "yses": 11146,
+      "UHFFFA": 11147,
+      "UHFFFAOYSA": 11148,
+      "actin": 11149,
+      "Ġunus": 11150,
+      "hesia": 11151,
+      "aceu": 11152,
+      "adow": 11153,
+      "oside": 11154,
+      "Ġglycos": 11155,
+      "Ġdiluted": 11156,
+      "ĠSource": 11157,
+      "olated": 11158,
+      "armaceu": 11159,
+      "antom": 11160,
+      "Ġmusc": 11161,
+      "Ġaveraging": 11162,
+      "Ġvisit": 11163,
+      "Ġcatch": 11164,
+      "Ġsatisfaction": 11165,
+      "Ġvon": 11166,
+      "valid": 11167,
+      "Ġyielded": 11168,
+      "Ġpackets": 11169,
+      "Ġresonant": 11170,
+      "pret": 11171,
+      "ĠGFP": 11172,
+      "Ġcutting": 11173,
+      "Ġreplacing": 11174,
+      "aze": 11175,
+      "Pa": 11176,
+      "Ġtoday": 11177,
+      "Ġdecided": 11178,
+      "ilateral": 11179,
+      "imate": 11180,
+      "lings": 11181,
+      "ĠRobust": 11182,
+      "ĠAst": 11183,
+      "odynamics": 11184,
+      "Ġlacking": 11185,
+      "izophrenia": 11186,
+      "Ġcontraction": 11187,
+      "umann": 11188,
+      "ĠSample": 11189,
+      "Ġdiamond": 11190,
+      "method": 11191,
+      "TOR": 11192,
+      "Ġcomments": 11193,
+      "sey": 11194,
+      "Ġmanufacturing": 11195,
+      "ĠDa": 11196,
+      "NR": 11197,
+      "Ġoperated": 11198,
+      "rates": 11199,
+      "Ġextinction": 11200,
+      "uvant": 11201,
+      "ĠFinite": 11202,
+      "Ġlymphocytes": 11203,
+      "bro": 11204,
+      "omology": 11205,
+      "Ġinstruments": 11206,
+      "bec": 11207,
+      "ogle": 11208,
+      "Ġquoti": 11209,
+      "Ġhyperbolic": 11210,
+      "Ġtrim": 11211,
+      "Ġpap": 11212,
+      "aturated": 11213,
+      "haus": 11214,
+      "Ġsessions": 11215,
+      "Ġcampaign": 11216,
+      "Ġvarieties": 11217,
+      "Ġprojected": 11218,
+      "Ġrid": 11219,
+      "bone": 11220,
+      "Ġancest": 11221,
+      "ĠET": 11222,
+      "mail": 11223,
+      "ĠTransport": 11224,
+      "///": 11225,
+      "ĠAnn": 11226,
+      "Ġcompositions": 11227,
+      "ĠINTRODUCTION": 11228,
+      "ĠâĪĴâĨĴ": 11229,
+      "Ġwhenever": 11230,
+      "ĠLip": 11231,
+      "parts": 11232,
+      "Ġisomorphic": 11233,
+      "Ġsulfate": 11234,
+      "Ġhop": 11235,
+      "Ġgon": 11236,
+      "ĠObject": 11237,
+      "Ġpipeline": 11238,
+      "Ġma": 11239,
+      "ĠGas": 11240,
+      "Ġtendency": 11241,
+      "object": 11242,
+      "Ġparametric": 11243,
+      "ĠReturn": 11244,
+      "Ġdwar": 11245,
+      "Ġpressures": 11246,
+      "ĠBios": 11247,
+      "Ġmultiplication": 11248,
+      "Ġdimin": 11249,
+      "Ġcolors": 11250,
+      "ĠTrue": 11251,
+      "Max": 11252,
+      "ĠDepend": 11253,
+      "Ġpairwise": 11254,
+      "Ġlake": 11255,
+      "Ġhierarchy": 11256,
+      "Ġthresholds": 11257,
+      "ĠAdaptive": 11258,
+      "making": 11259,
+      "Ġcatalysts": 11260,
+      "ipal": 11261,
+      "Ġeggs": 11262,
+      "Ġwire": 11263,
+      "ophyll": 11264,
+      "ictor": 11265,
+      "labeled": 11266,
+      "Ġmuscles": 11267,
+      "ĠUnderstanding": 11268,
+      "Ġfibre": 11269,
+      "controlled": 11270,
+      "Ġinvariance": 11271,
+      "Ġcache": 11272,
+      "Ġboson": 11273,
+      "Ġnearby": 11274,
+      "ĠWomen": 11275,
+      "ĠInitial": 11276,
+      "Ġprobabilistic": 11277,
+      "Ġembryonic": 11278,
+      "ĠBetween": 11279,
+      "Ġconjecture": 11280,
+      "ienti": 11281,
+      "tx": 11282,
+      "gens": 11283,
+      "anck": 11284,
+      "Ġgir": 11285,
+      "ĠLower": 11286,
+      "Ġhospitals": 11287,
+      "bridge": 11288,
+      "Method": 11289,
+      "Ġtheta": 11290,
+      "ja": 11291,
+      "Ġconceptual": 11292,
+      "Ġcolle": 11293,
+      "ĠSaf": 11294,
+      "dic": 11295,
+      "Ġpet": 11296,
+      "Ġprimer": 11297,
+      "ĠOh": 11298,
+      "Ġuntreated": 11299,
+      "longrightarrow": 11300,
+      "Ġlicense": 11301,
+      "Ġhelps": 11302,
+      "Ġcleavage": 11303,
+      "Ġamplified": 11304,
+      "е": 11305,
+      "Ġaccessible": 11306,
+      "ĠSelection": 11307,
+      "ĠLorentz": 11308,
+      "Py": 11309,
+      "Ġpolarized": 11310,
+      "ĠSTAT": 11311,
+      "mitt": 11312,
+      "Up": 11313,
+      "Ġongoing": 11314,
+      "Ġneph": 11315,
+      "efficient": 11316,
+      "activ": 11317,
+      "ĠRR": 11318,
+      "Ġfunctioning": 11319,
+      "otin": 11320,
+      "Ġlists": 11321,
+      "Ġformalism": 11322,
+      "Ġoscillator": 11323,
+      "Ġgastrointestinal": 11324,
+      "ootstrap": 11325,
+      "ĠAsia": 11326,
+      "ĠDay": 11327,
+      "Ġcompeting": 11328,
+      "ivalent": 11329,
+      "Ġbladder": 11330,
+      "Ġhit": 11331,
+      "Ġapproximations": 11332,
+      "ĠEg": 11333,
+      "ĠClust": 11334,
+      "Ġrelies": 11335,
+      "NE": 11336,
+      "copro": 11337,
+      "Ġbank": 11338,
+      "Ġintegrating": 11339,
+      "ĠHear": 11340,
+      "Ġinitiated": 11341,
+      "acryl": 11342,
+      "ĠBH": 11343,
+      "racted": 11344,
+      "yc": 11345,
+      "ĠRa": 11346,
+      "Ġremarkable": 11347,
+      "ĠË": 11348,
+      "teness": 11349,
+      "Ġemploying": 11350,
+      "steine": 11351,
+      "Ġï£Ń": 11352,
+      "Ġtransfected": 11353,
+      "Ġinjuries": 11354,
+      "ĠBrief": 11355,
+      "Ġwidespread": 11356,
+      "ĠAK": 11357,
+      "IVE": 11358,
+      "Ġharm": 11359,
+      "Ġpole": 11360,
+      "Ġanisotropic": 11361,
+      "aten": 11362,
+      "gene": 11363,
+      "ivariate": 11364,
+      "Inter": 11365,
+      "ductors": 11366,
+      "Ġaccompl": 11367,
+      "oglobin": 11368,
+      "cong": 11369,
+      "Ġqueries": 11370,
+      "escope": 11371,
+      "ĠHop": 11372,
+      "Ġentity": 11373,
+      "Ġoffered": 11374,
+      "State": 11375,
+      "ĠExperiments": 11376,
+      "anner": 11377,
+      "ĠWood": 11378,
+      "arded": 11379,
+      "agon": 11380,
+      "Ġfibroblasts": 11381,
+      "Ġnanos": 11382,
+      "Ġperoxid": 11383,
+      "Ġevid": 11384,
+      "Ġ": 11385,
+      "Ġretained": 11386,
+      "osqu": 11387,
+      "Ġleaving": 11388,
+      "Ġfashion": 11389,
+      "ĠnM": 11390,
+      "Ġmutual": 11391,
+      "approxim": 11392,
+      "Ġwalking": 11393,
+      "Ġimpossible": 11394,
+      "Ġdemonstrating": 11395,
+      "Ġdegener": 11396,
+      "ĠAV": 11397,
+      "Ġcontrary": 11398,
+      "ustion": 11399,
+      "oclonal": 11400,
+      "Anal": 11401,
+      "Ġperformances": 11402,
+      "Ġcomprom": 11403,
+      "orms": 11404,
+      "Ġbudget": 11405,
+      "ĠHaw": 11406,
+      "Ġarthritis": 11407,
+      "obj": 11408,
+      "noise": 11409,
+      "TiO": 11410,
+      "ochrome": 11411,
+      "Ġgeodes": 11412,
+      "bean": 11413,
+      "Ġselectivity": 11414,
+      "ĠFood": 11415,
+      "ughter": 11416,
+      "Ġpermutation": 11417,
+      "ĠRP": 11418,
+      "osal": 11419,
+      "Ġadip": 11420,
+      "armaceutical": 11421,
+      "when": 11422,
+      "ĠText": 11423,
+      "week": 11424,
+      "Ġbonding": 11425,
+      "arb": 11426,
+      "ocor": 11427,
+      "Ġvoc": 11428,
+      "Ġupregulated": 11429,
+      "Ġneighbors": 11430,
+      "Ġtrait": 11431,
+      "Ġtheore": 11432,
+      "Ġcf": 11433,
+      "ĠBerg": 11434,
+      "ĠLA": 11435,
+      "Ġlas": 11436,
+      "unte": 11437,
+      "ceptual": 11438,
+      "ASE": 11439,
+      "Ġischemic": 11440,
+      "Ġbending": 11441,
+      "dataset": 11442,
+      "Ġkeeping": 11443,
+      "Ġarrows": 11444,
+      "Ġsubstances": 11445,
+      "Ġns": 11446,
+      "Ġextending": 11447,
+      "ĠRu": 11448,
+      "Ġsupplementation": 11449,
+      "critical": 11450,
+      "ĠTraining": 11451,
+      "bullet": 11452,
+      "Ġpara": 11453,
+      "tail": 11454,
+      "ĠReference": 11455,
+      "Ġ": 11456,
+      "Ġdissipation": 11457,
+      "Ġauxiliary": 11458,
+      "ĠCycl": 11459,
+      "stim": 11460,
+      "Ġdilution": 11461,
+      "buf": 11462,
+      "ĠMiss": 11463,
+      "Ġultimately": 11464,
+      "Ġpowers": 11465,
+      "Ġstands": 11466,
+      "usted": 11467,
+      "ĠOH": 11468,
+      "habilitation": 11469,
+      "analy": 11470,
+      "ĠBra": 11471,
+      "adding": 11472,
+      "Corollary": 11473,
+      "Ġdrought": 11474,
+      "quality": 11475,
+      "Ġstandardized": 11476,
+      "ĠJe": 11477,
+      "ĠAcid": 11478,
+      "Ġmism": 11479,
+      "ĠChrom": 11480,
+      "draw": 11481,
+      "ĠBiom": 11482,
+      "ĠStability": 11483,
+      "Furthermore": 11484,
+      "last": 11485,
+      "vic": 11486,
+      "Ġabst": 11487,
+      "Ġbis": 11488,
+      "Ġemergence": 11489,
+      "Ġgiant": 11490,
+      "De": 11491,
+      "ĠSamples": 11492,
+      "ABA": 11493,
+      "nas": 11494,
+      "Ġont": 11495,
+      "Ġevap": 11496,
+      "levant": 11497,
+      "main": 11498,
+      "ĠRod": 11499,
+      "Ġcros": 11500,
+      "itary": 11501,
+      "Ġdoub": 11502,
+      "rö": 11503,
+      "igenetic": 11504,
+      "Ġincomplete": 11505,
+      "depth": 11506,
+      "ïģ": 11507,
+      "Ġsaturated": 11508,
+      "Ġaerosol": 11509,
+      "Assum": 11510,
+      "Ġimmunos": 11511,
+      "Ġlipids": 11512,
+      "itoneal": 11513,
+      "Ġbearing": 11514,
+      "ĠImplications": 11515,
+      "Ġsustained": 11516,
+      "Ġcompetitive": 11517,
+      "Ġmotivation": 11518,
+      "Ġdisturbance": 11519,
+      "rystalline": 11520,
+      "Ġtaxa": 11521,
+      "Ġdementia": 11522,
+      "Ġconcerned": 11523,
+      "PIO": 11524,
+      "homogeneous": 11525,
+      "ĠEv": 11526,
+      "ĠGeorge": 11527,
+      "ĠAlgorithms": 11528,
+      "ickel": 11529,
+      "usively": 11530,
+      "Ġcorner": 11531,
+      "ĠRest": 11532,
+      "Ġinfinity": 11533,
+      "ĠTransform": 11534,
+      "heng": 11535,
+      "Ġneurode": 11536,
+      "olim": 11537,
+      "Íij": 11538,
+      "Ġskew": 11539,
+      "ĠBS": 11540,
+      "score": 11541,
+      "YPE": 11542,
+      "eman": 11543,
+      "elle": 11544,
+      "ĠCorrelation": 11545,
+      "Ġcultural": 11546,
+      "ophosph": 11547,
+      "Ġattenuation": 11548,
+      "Ġaggregate": 11549,
+      "Ġambig": 11550,
+      "Ġanomalous": 11551,
+      "Ġtors": 11552,
+      "Ġplanet": 11553,
+      "ĠNPs": 11554,
+      "hr": 11555,
+      "ĠDivision": 11556,
+      "ĠEducation": 11557,
+      "lectic": 11558,
+      "Ġbrought": 11559,
+      "ĠMorph": 11560,
+      "Ġplanes": 11561,
+      "Ġsugar": 11562,
+      "Ġdendritic": 11563,
+      "Ġcontour": 11564,
+      "Ġcylinder": 11565,
+      "post": 11566,
+      "Ġwent": 11567,
+      "RL": 11568,
+      "Ġadmission": 11569,
+      "MSE": 11570,
+      "IX": 11571,
+      "Ġdisjoint": 11572,
+      "Ġannotation": 11573,
+      "Ġisotope": 11574,
+      "Ġμν": 11575,
+      "Ġeliminate": 11576,
+      "Ġreactor": 11577,
+      "onents": 11578,
+      "Ġreasoning": 11579,
+      "Ġmorbidity": 11580,
+      "Ġcorrosion": 11581,
+      "othermal": 11582,
+      "arctic": 11583,
+      "ĠMB": 11584,
+      "ĠZhao": 11585,
+      "Ġhistological": 11586,
+      "Ġsuperconducting": 11587,
+      "attered": 11588,
+      "Ġhousehold": 11589,
+      "ĠProp": 11590,
+      "Ġasser": 11591,
+      "hered": 11592,
+      "Ġteams": 11593,
+      "Ġvanishes": 11594,
+      "Pre": 11595,
+      "aments": 11596,
+      "Ġamorphous": 11597,
+      "ĠDetermination": 11598,
+      "missions": 11599,
+      "Ġoverhead": 11600,
+      "determ": 11601,
+      "Ġutilizing": 11602,
+      "fa": 11603,
+      "ipolar": 11604,
+      "Ġformulated": 11605,
+      "Ġextrap": 11606,
+      "grid": 11607,
+      "Ġhumidity": 11608,
+      "uber": 11609,
+      "tumor": 11610,
+      "rous": 11611,
+      "Ġdistortion": 11612,
+      "dynamics": 11613,
+      "ĠLoss": 11614,
+      "Ġscaled": 11615,
+      "Ġischemia": 11616,
+      "Ġaxes": 11617,
+      "Ġquantit": 11618,
+      "nit": 11619,
+      "ĠRegion": 11620,
+      "ained": 11621,
+      "Ġfill": 11622,
+      "Ġbranching": 11623,
+      "ĠTiss": 11624,
+      "cross": 11625,
+      "Ġplatelet": 11626,
+      "iffiffiffiff": 11627,
+      "rops": 11628,
+      "lux": 11629,
+      "join": 11630,
+      "uracy": 11631,
+      "icide": 11632,
+      "ĠLouis": 11633,
+      "Ġ": 11634,
+      "Ġstrings": 11635,
+      "yset": 11636,
+      "Ġfacial": 11637,
+      "ĠMMP": 11638,
+      "RES": 11639,
+      "Ġhydrolysis": 11640,
+      "ĠCanadian": 11641,
+      "Ġprojective": 11642,
+      "Ġscatter": 11643,
+      "uron": 11644,
+      "ĠPsych": 11645,
+      "complex": 11646,
+      "ĠNam": 11647,
+      "Ġconcurrent": 11648,
+      "IONS": 11649,
+      "Ġthous": 11650,
+      "Ġchance": 11651,
+      "Ġplacement": 11652,
+      "Ġawareness": 11653,
+      "Ġtrib": 11654,
+      "ĠTex": 11655,
+      "ĠThird": 11656,
+      "Ġlabeling": 11657,
+      "cerol": 11658,
+      "Ġsaw": 11659,
+      "ĠBand": 11660,
+      "ĠPear": 11661,
+      "Ġpregnant": 11662,
+      "ĠDown": 11663,
+      "platin": 11664,
+      "Seq": 11665,
+      "xe": 11666,
+      "ethylene": 11667,
+      "ĠHigher": 11668,
+      "Ġreality": 11669,
+      "uris": 11670,
+      "ĠPAR": 11671,
+      "lb": 11672,
+      "dose": 11673,
+      "shif": 11674,
+      "iliar": 11675,
+      "total": 11676,
+      "SW": 11677,
+      "Ġvalve": 11678,
+      "nder": 11679,
+      "н": 11680,
+      "amous": 11681,
+      "Ġendomet": 11682,
+      "LISA": 11683,
+      "Ġfractures": 11684,
+      "Ġfilt": 11685,
+      "role": 11686,
+      "Ġmicrostructure": 11687,
+      "ĠSNP": 11688,
+      "TER": 11689,
+      "ĠZnO": 11690,
+      "oving": 11691,
+      "ali": 11692,
+      "ĠGM": 11693,
+      "unct": 11694,
+      "Ġextensions": 11695,
+      "expression": 11696,
+      "Ġescape": 11697,
+      "ĠMas": 11698,
+      "ĠSpanish": 11699,
+      "Ġfloor": 11700,
+      "ĠCommon": 11701,
+      "otopy": 11702,
+      "plementation": 11703,
+      "Ġrhyth": 11704,
+      "Ġserves": 11705,
+      "yto": 11706,
+      "Ġwavelengths": 11707,
+      "emptyset": 11708,
+      "ĠHill": 11709,
+      "nor": 11710,
+      "ĠElectro": 11711,
+      "Ġdehydrogen": 11712,
+      "Ġwhom": 11713,
+      "imetric": 11714,
+      "ĠRoman": 11715,
+      "ĠVe": 11716,
+      "âī¥": 11717,
+      "ĠKu": 11718,
+      "ĠTransfer": 11719,
+      "Äĩ": 11720,
+      "ĠTF": 11721,
+      "brain": 11722,
+      "coprotein": 11723,
+      "ĠGreat": 11724,
+      "aven": 11725,
+      "ĠIndividual": 11726,
+      "uri": 11727,
+      "Ġfungi": 11728,
+      "Ġparam": 11729,
+      "pton": 11730,
+      "symmetry": 11731,
+      "Ġlock": 11732,
+      "meas": 11733,
+      "Ġhaem": 11734,
+      "Ġhip": 11735,
+      "Ass": 11736,
+      "enger": 11737,
+      "Ġpotassium": 11738,
+      "anal": 11739,
+      "ibrary": 11740,
+      "Ġschools": 11741,
+      "natal": 11742,
+      "Ġalleles": 11743,
+      "ĠHLA": 11744,
+      "oxygen": 11745,
+      "ĠCup": 11746,
+      "Ġpurely": 11747,
+      "DO": 11748,
+      "Ġchip": 11749,
+      "ôı": 11750,
+      "Car": 11751,
+      "sil": 11752,
+      "Ġunlikely": 11753,
+      "correspond": 11754,
+      "ĠDP": 11755,
+      "Ġintense": 11756,
+      "Ġforcing": 11757,
+      "ĠJournal": 11758,
+      "Ġarrow": 11759,
+      "ocyan": 11760,
+      "Ġcultiv": 11761,
+      "Ġblind": 11762,
+      "Ġselecting": 11763,
+      "ocarcinoma": 11764,
+      "rance": 11765,
+      "Ġhydrophobic": 11766,
+      "closed": 11767,
+      "Ġensures": 11768,
+      "Ġpromoted": 11769,
+      "Ġdetectable": 11770,
+      "ranean": 11771,
+      "Ġschedule": 11772,
+      "Ġpartly": 11773,
+      "Ġgland": 11774,
+      "Ġcouple": 11775,
+      "ĠEmerg": 11776,
+      "Ġtraces": 11777,
+      "poly": 11778,
+      "Ġprotease": 11779,
+      "ystic": 11780,
+      "Ġdocuments": 11781,
+      "positions": 11782,
+      "Ġdriver": 11783,
+      "tium": 11784,
+      "ĠCYP": 11785,
+      "close": 11786,
+      "ĠRecep": 11787,
+      "Ġpermit": 11788,
+      "Ġblocked": 11789,
+      "Ġinvestigating": 11790,
+      "ĠTumor": 11791,
+      "ĠBig": 11792,
+      "Ġwavegu": 11793,
+      "Ġsubstance": 11794,
+      "Ġweaker": 11795,
+      "ĠMont": 11796,
+      "rovers": 11797,
+      "ĠMexico": 11798,
+      "pres": 11799,
+      "ĠAcute": 11800,
+      "Ġmicrogl": 11801,
+      "ĠES": 11802,
+      "itoring": 11803,
+      "ĠSeries": 11804,
+      "lights": 11805,
+      "Ġhypothesized": 11806,
+      "Ġconstructs": 11807,
+      "Ġfiltration": 11808,
+      "Black": 11809,
+      "Ġunchanged": 11810,
+      "Ġobservable": 11811,
+      "Ġray": 11812,
+      "between": 11813,
+      "Ġ": 11814,
+      "ĠPosition": 11815,
+      "Ġthi": 11816,
+      "ĠSystematic": 11817,
+      "Class": 11818,
+      "km": 11819,
+      "ĠTak": 11820,
+      "Ġrespondents": 11821,
+      "Ġinnate": 11822,
+      "Ġant": 11823,
+      "Ġconnecting": 11824,
+      "Rel": 11825,
+      "Ġmanipulation": 11826,
+      "ĠNeg": 11827,
+      "NPs": 11828,
+      "ĠDiab": 11829,
+      "ĠActive": 11830,
+      "ĠGall": 11831,
+      "ĠCoulomb": 11832,
+      "Ġspacing": 11833,
+      "ĠFlor": 11834,
+      "Ġconductance": 11835,
+      "Ġtracks": 11836,
+      "ĠZhu": 11837,
+      "weighted": 11838,
+      "rocy": 11839,
+      "Ġfather": 11840,
+      "idium": 11841,
+      "structured": 11842,
+      "ĠTel": 11843,
+      "Ġstrom": 11844,
+      "ithub": 11845,
+      "certain": 11846,
+      "But": 11847,
+      "ĠAccess": 11848,
+      "Ġpreventing": 11849,
+      "restrial": 11850,
+      "ĠConsidering": 11851,
+      "true": 11852,
+      "Ġhosts": 11853,
+      "Ġworst": 11854,
+      "ĠPd": 11855,
+      "gredi": 11856,
+      "Ġglycol": 11857,
+      "Ġstory": 11858,
+      "osquito": 11859,
+      "paratus": 11860,
+      "Ġmeeting": 11861,
+      "Ġepisode": 11862,
+      "nc": 11863,
+      "ĠSand": 11864,
+      "Ġuint": 11865,
+      "ynamical": 11866,
+      "urt": 11867,
+      "Ġeducational": 11868,
+      "Ġfocuses": 11869,
+      "gt": 11870,
+      "ĠHS": 11871,
+      "Ġdeterminant": 11872,
+      "Ġlithium": 11873,
+      "ĠDigital": 11874,
+      "Ġguidance": 11875,
+      "Ġpriority": 11876,
+      "Ġparty": 11877,
+      "orial": 11878,
+      "Two": 11879,
+      "ĠProblems": 11880,
+      "Ġseman": 11881,
+      "ĠCNN": 11882,
+      "ĠEpid": 11883,
+      "Ġplaying": 11884,
+      "Ġelimination": 11885,
+      "ĠSat": 11886,
+      "Ġobjectives": 11887,
+      "plectic": 11888,
+      "Ġcircumst": 11889,
+      "ĠGS": 11890,
+      "ocellular": 11891,
+      "otrans": 11892,
+      "Ġfinds": 11893,
+      "Ġaromatic": 11894,
+      "izers": 11895,
+      "Ġfavorable": 11896,
+      "standard": 11897,
+      "ichlor": 11898,
+      "models": 11899,
+      "otyping": 11900,
+      "Ġstabilization": 11901,
+      "Ġhandling": 11902,
+      "Ġcoated": 11903,
+      "even": 11904,
+      "Ġletter": 11905,
+      "ZE": 11906,
+      "Ġultrason": 11907,
+      "Ġfriend": 11908,
+      "Ġsensiti": 11909,
+      "Ġattachment": 11910,
+      "Ġapart": 11911,
+      "Ġgrey": 11912,
+      "Ġaircraft": 11913,
+      "ĠrRNA": 11914,
+      "Ġenabled": 11915,
+      "Ġbuff": 11916,
+      "Ġredox": 11917,
+      "assisted": 11918,
+      "Ġgenerality": 11919,
+      "PSS": 11920,
+      "Ġelection": 11921,
+      "response": 11922,
+      "Ġdedicated": 11923,
+      "Ġdemographic": 11924,
+      "Ġimposed": 11925,
+      "ĠKir": 11926,
+      "ĠRadio": 11927,
+      "ĠELISA": 11928,
+      "gae": 11929,
+      "Ġresc": 11930,
+      "ĠRic": 11931,
+      "raphic": 11932,
+      "Ġrail": 11933,
+      "Ġjournal": 11934,
+      "oler": 11935,
+      "WS": 11936,
+      "Ġincorporation": 11937,
+      "wind": 11938,
+      "Ġauditory": 11939,
+      "AE": 11940,
+      "task": 11941,
+      "Ġpc": 11942,
+      "wall": 11943,
+      "Ġapprec": 11944,
+      "aterials": 11945,
+      "Ġpartner": 11946,
+      "Ġcollective": 11947,
+      "Ġscoring": 11948,
+      "ĠFrank": 11949,
+      "Ġpermanent": 11950,
+      "ĠIran": 11951,
+      "umination": 11952,
+      "Med": 11953,
+      "ĠHybrid": 11954,
+      "Ġphenotypic": 11955,
+      "Ġdisruption": 11956,
+      "violet": 11957,
+      "ospheric": 11958,
+      "Ġregimes": 11959,
+      "ĠColor": 11960,
+      "ĠPatient": 11961,
+      "Ġfever": 11962,
+      "Ġnn": 11963,
+      "Ġvariational": 11964,
+      "keys": 11965,
+      "Ġdistill": 11966,
+      "Ġspectroscopic": 11967,
+      "ĠArchitect": 11968,
+      "acing": 11969,
+      "Ġproves": 11970,
+      "Ġverteb": 11971,
+      "ĠComputer": 11972,
+      "Ġexpensive": 11973,
+      "Ġfrozen": 11974,
+      "arcoma": 11975,
+      "NK": 11976,
+      "Ġhistone": 11977,
+      "Ġpolymerization": 11978,
+      "Ġtob": 11979,
+      "Ġturned": 11980,
+      "effective": 11981,
+      "ĠAuthor": 11982,
+      "API": 11983,
+      "Ġdecade": 11984,
+      "ĠRobert": 11985,
+      "Example": 11986,
+      "overset": 11987,
+      "ABLE": 11988,
+      "ĠBehavior": 11989,
+      "feed": 11990,
+      "ĠTai": 11991,
+      "Ġ": 11992,
+      "Ġegg": 11993,
+      "Ġcath": 11994,
+      "aux": 11995,
+      "ĠJohnson": 11996,
+      "Ġtorque": 11997,
+      "Ġpurification": 11998,
+      "White": 11999,
+      "cious": 12000,
+      "ĠSong": 12001,
+      "Ġprecipit": 12002,
+      "reshold": 12003,
+      "Ġmilitary": 12004,
+      "Ġconvection": 12005,
+      "ĠMiddle": 12006,
+      "ĠWhe": 12007,
+      "Ġôı": 12008,
+      "aland": 12009,
+      "aration": 12010,
+      "figure": 12011,
+      "Ġdeduce": 12012,
+      "chloro": 12013,
+      "cost": 12014,
+      "ithmetic": 12015,
+      "ĠItalian": 12016,
+      "missible": 12017,
+      "ĠCommunity": 12018,
+      "ĠNature": 12019,
+      "Ġdioxide": 12020,
+      "Ġbalanced": 12021,
+      "ett": 12022,
+      "STAT": 12023,
+      "ilding": 12024,
+      "Ġevolved": 12025,
+      "Ġmonot": 12026,
+      "pur": 12027,
+      "Ġpreferences": 12028,
+      "dinger": 12029,
+      "Ġargue": 12030,
+      "Ġmotions": 12031,
+      "Ġinfant": 12032,
+      "Ġaccelerated": 12033,
+      "Ġobserver": 12034,
+      "Ġfabrication": 12035,
+      "ĠMechanisms": 12036,
+      "Ġfunctor": 12037,
+      "Ġharves": 12038,
+      "rase": 12039,
+      "ĠSpecial": 12040,
+      "Ġdeposits": 12041,
+      "Ġrub": 12042,
+      "à¸": 12043,
+      "ĠCPU": 12044,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 12045,
+      "atomical": 12046,
+      "Ġfinit": 12047,
+      "Ġsecure": 12048,
+      "Ġnutritional": 12049,
+      "renal": 12050,
+      "ĠFalse": 12051,
+      "Ġshel": 12052,
+      "Ġrecruited": 12053,
+      "ambig": 12054,
+      "ĠSignaling": 12055,
+      "KO": 12056,
+      "organisms": 12057,
+      "ĠLT": 12058,
+      "elen": 12059,
+      "ĠMarc": 12060,
+      "abatic": 12061,
+      "Ġtables": 12062,
+      "Ġconfined": 12063,
+      "ĠAz": 12064,
+      "Ġproductivity": 12065,
+      "Ġadherence": 12066,
+      "Ġreplicates": 12067,
+      "Ġvirt": 12068,
+      "fin": 12069,
+      "Ġagricultural": 12070,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 12071,
+      "ĠChampionship": 12072,
+      "anda": 12073,
+      "ĠChurch": 12074,
+      "During": 12075,
+      "Ġinserted": 12076,
+      "ighter": 12077,
+      "Ġxen": 12078,
+      "Ġsave": 12079,
+      "Ġtangent": 12080,
+      "venous": 12081,
+      "Ġconverge": 12082,
+      "Ġdistinguished": 12083,
+      "Ġexplos": 12084,
+      "Ġaortic": 12085,
+      "Ġjump": 12086,
+      "Ġneonatal": 12087,
+      "udden": 12088,
+      "Ġslower": 12089,
+      "Ġinfarction": 12090,
+      "Ġprevents": 12091,
+      "uer": 12092,
+      "Ġeros": 12093,
+      "RP": 12094,
+      "Ġcontinues": 12095,
+      "ORT": 12096,
+      "Ġconsiders": 12097,
+      "ĠNuclear": 12098,
+      "lymp": 12099,
+      "Ġaccounted": 12100,
+      "oresis": 12101,
+      "Ġneighboring": 12102,
+      "ĠRichard": 12103,
+      "Ġenfor": 12104,
+      "ĠChronic": 12105,
+      "Ġdiscover": 12106,
+      "ĠHong": 12107,
+      "cells": 12108,
+      "ĠChall": 12109,
+      "Ġhomogen": 12110,
+      "Ġatheros": 12111,
+      "Ġisolate": 12112,
+      "ĠPlasma": 12113,
+      "ĠDL": 12114,
+      "parametric": 12115,
+      "ĠUpper": 12116,
+      "HP": 12117,
+      "Ġintroduces": 12118,
+      "Ġmothers": 12119,
+      "Ġattract": 12120,
+      "Ġexclusion": 12121,
+      "gravity": 12122,
+      "ĠKr": 12123,
+      "Ġspike": 12124,
+      "ĠHeat": 12125,
+      "vival": 12126,
+      "ĠRNAs": 12127,
+      "bach": 12128,
+      "atorial": 12129,
+      "ĠLtd": 12130,
+      "onomy": 12131,
+      "invasive": 12132,
+      "lass": 12133,
+      "Ġwells": 12134,
+      "Ġimaginary": 12135,
+      "Ġcarbohyd": 12136,
+      "oda": 12137,
+      "Ġactivate": 12138,
+      "µĦ": 12139,
+      "Ġenzymatic": 12140,
+      "pes": 12141,
+      "Ġstatements": 12142,
+      "Ġapproximated": 12143,
+      "ĠSalmon": 12144,
+      "ophageal": 12145,
+      "ĠHPV": 12146,
+      "conf": 12147,
+      "umat": 12148,
+      "Ġsulfur": 12149,
+      "ĠRecall": 12150,
+      "Ġchond": 12151,
+      "Ġviable": 12152,
+      "poration": 12153,
+      "Ġcarefully": 12154,
+      "tetra": 12155,
+      "Ġlymphoma": 12156,
+      "stat": 12157,
+      "Ġconservative": 12158,
+      "atabase": 12159,
+      "mand": 12160,
+      "Ġscored": 12161,
+      "Ġvas": 12162,
+      "Ġprivacy": 12163,
+      "onymous": 12164,
+      "Ġlogarithmic": 12165,
+      "ĠEcon": 12166,
+      "Ġachieves": 12167,
+      "Ġabundances": 12168,
+      "cam": 12169,
+      "Ġcyan": 12170,
+      "ĠEL": 12171,
+      "idelity": 12172,
+      "jo": 12173,
+      "Ġanticip": 12174,
+      "reported": 12175,
+      "Ġarrangement": 12176,
+      "iterranean": 12177,
+      "psis": 12178,
+      "ichi": 12179,
+      "Ġta": 12180,
+      "umping": 12181,
+      "ĠActivation": 12182,
+      "Ġmelt": 12183,
+      "Ġanno": 12184,
+      "oge": 12185,
+      "ĠDam": 12186,
+      "optimal": 12187,
+      "Ġneurological": 12188,
+      "sa": 12189,
+      "ĠParameters": 12190,
+      "offset": 12191,
+      "Ġcement": 12192,
+      "Ġinhibiting": 12193,
+      "Ġchose": 12194,
+      "itzer": 12195,
+      "attr": 12196,
+      "Ġmoder": 12197,
+      "atories": 12198,
+      "Ġteaching": 12199,
+      "ĠCore": 12200,
+      "phthal": 12201,
+      "ĠLuc": 12202,
+      "Ġingredi": 12203,
+      "Ġclearance": 12204,
+      "Ġachieving": 12205,
+      "tage": 12206,
+      "Ġburst": 12207,
+      "vie": 12208,
+      "ĠSpain": 12209,
+      "pto": 12210,
+      "Ġtransmembrane": 12211,
+      "Ġsupplementary": 12212,
+      "Ġtoken": 12213,
+      "Ġobviously": 12214,
+      "ĠVector": 12215,
+      "Ġdestr": 12216,
+      "HOD": 12217,
+      "Ġassumes": 12218,
+      "Ġpenetration": 12219,
+      "Ġsubjective": 12220,
+      "holds": 12221,
+      "ão": 12222,
+      "Ġmotiv": 12223,
+      "Ġproviders": 12224,
+      "vascular": 12225,
+      "Ġdepartment": 12226,
+      "ocket": 12227,
+      "File": 12228,
+      "Ġbreath": 12229,
+      "ĠBest": 12230,
+      "grable": 12231,
+      "Ġliqu": 12232,
+      "ĠArg": 12233,
+      "ĠBob": 12234,
+      "Ġfragmentation": 12235,
+      "ectic": 12236,
+      "Ġvital": 12237,
+      "since": 12238,
+      "alloc": 12239,
+      "oxyphenyl": 12240,
+      "Ġradiotherapy": 12241,
+      "ĠSDS": 12242,
+      "Ġcytometry": 12243,
+      "nucle": 12244,
+      "ĠIM": 12245,
+      "ĠTeV": 12246,
+      "rafish": 12247,
+      "ĠKorea": 12248,
+      "Ġstrengthen": 12249,
+      "Ġbare": 12250,
+      "Ġwoman": 12251,
+      "Ġradar": 12252,
+      "Ġplatforms": 12253,
+      "ozygous": 12254,
+      "ĠAh": 12255,
+      "Ġsubtypes": 12256,
+      "pyrid": 12257,
+      "ĠTranscription": 12258,
+      "Ġáº": 12259,
+      "ĠMeasurements": 12260,
+      "Ġsurviv": 12261,
+      "ĠNear": 12262,
+      "Ġcascade": 12263,
+      "outhe": 12264,
+      "BU": 12265,
+      "Ġexponentially": 12266,
+      "Ġhazard": 12267,
+      "ĠsiRNA": 12268,
+      "Ġcellulose": 12269,
+      "Figs": 12270,
+      "Ġdifferentiated": 12271,
+      "Ġimplicated": 12272,
+      "metric": 12273,
+      "Ġcorrelate": 12274,
+      "Ġmission": 12275,
+      "Ġmantle": 12276,
+      "ĠPhyl": 12277,
+      "ĠHart": 12278,
+      "Ġgases": 12279,
+      "Ġunity": 12280,
+      "Ġexpert": 12281,
+      "Ġchart": 12282,
+      "Ġdict": 12283,
+      "Ġepile": 12284,
+      "Ġoffspring": 12285,
+      "Ġemerged": 12286,
+      "Ġdemands": 12287,
+      "Ġpresum": 12288,
+      "orbid": 12289,
+      "ĠMedicine": 12290,
+      "Ġstreams": 12291,
+      "ticed": 12292,
+      "ĠNic": 12293,
+      "Ġfilling": 12294,
+      "ĠCro": 12295,
+      "Ġrestrictions": 12296,
+      "See": 12297,
+      "ĠMill": 12298,
+      "Ġparental": 12299,
+      "Ġdeterminants": 12300,
+      "Ġecosystem": 12301,
+      "ĠWall": 12302,
+      "ĠMemory": 12303,
+      "plets": 12304,
+      "Ġaggregates": 12305,
+      "perturb": 12306,
+      "Ġresidents": 12307,
+      "ACK": 12308,
+      "vectors": 12309,
+      "Ġmanually": 12310,
+      "Ġïĺ": 12311,
+      "ĠFramework": 12312,
+      "Ġvag": 12313,
+      "ebrafish": 12314,
+      "lib": 12315,
+      "ĠHeart": 12316,
+      "ĠAnimal": 12317,
+      "Ġwider": 12318,
+      "Gene": 12319,
+      "ĠRos": 12320,
+      "Ġoperate": 12321,
+      "Ġpossibilities": 12322,
+      "ĠStrong": 12323,
+      "Ġpyro": 12324,
+      "respectively": 12325,
+      "Ġhybridization": 12326,
+      "ipedia": 12327,
+      "xin": 12328,
+      "Ġstom": 12329,
+      "fish": 12330,
+      "ĠForce": 12331,
+      "Ġdimer": 12332,
+      "SUL": 12333,
+      "else": 12334,
+      "Ġunde": 12335,
+      "gar": 12336,
+      "conv": 12337,
+      "Ġarrival": 12338,
+      "Ġmonoclonal": 12339,
+      "IAL": 12340,
+      "Ġly": 12341,
+      "Ġsymmetries": 12342,
+      "Ġnursing": 12343,
+      "rach": 12344,
+      "ĠóµĦ": 12345,
+      "Ġbiased": 12346,
+      "Ġcues": 12347,
+      "Ġbiomarker": 12348,
+      "ders": 12349,
+      "Ġcrow": 12350,
+      "ernels": 12351,
+      "Ġbilateral": 12352,
+      "Ġphysically": 12353,
+      "Ġpatches": 12354,
+      "Ġuncon": 12355,
+      "ĠBefore": 12356,
+      "default": 12357,
+      "estyle": 12358,
+      "tfrac": 12359,
+      "ĠCox": 12360,
+      "Ġinfiltration": 12361,
+      "Ġconvert": 12362,
+      "Ġstrengths": 12363,
+      "ĠSar": 12364,
+      "igible": 12365,
+      "ocomp": 12366,
+      "Ġstir": 12367,
+      "Ġschizophrenia": 12368,
+      "was": 12369,
+      "Ġow": 12370,
+      "eterm": 12371,
+      "ĠOrder": 12372,
+      "Ġfoss": 12373,
+      "Ġlineage": 12374,
+      "Ġrabbit": 12375,
+      "Ġregularization": 12376,
+      "ranch": 12377,
+      "oplastic": 12378,
+      "TO": 12379,
+      "Ġmeasurable": 12380,
+      "Ġmang": 12381,
+      "initial": 12382,
+      "Ġbuildings": 12383,
+      "Ġsystematically": 12384,
+      "Ġfermions": 12385,
+      "Ġlibraries": 12386,
+      "Ġablation": 12387,
+      "ideos": 12388,
+      "ĠWi": 12389,
+      "photon": 12390,
+      "ĠTesting": 12391,
+      "ĠComputing": 12392,
+      "tier": 12393,
+      "inet": 12394,
+      "Ġprimitive": 12395,
+      "Ġcapillary": 12396,
+      "Ġslip": 12397,
+      "vergence": 12398,
+      "rapeutic": 12399,
+      "ĠBlue": 12400,
+      "ĠAcad": 12401,
+      "hai": 12402,
+      "ĠLew": 12403,
+      "Ġtriangular": 12404,
+      "MSO": 12405,
+      "Ġsalinity": 12406,
+      "Ġnanocom": 12407,
+      "oa": 12408,
+      "Ġhomomorphism": 12409,
+      "ĠMM": 12410,
+      "Ġresin": 12411,
+      "DB": 12412,
+      "uminescence": 12413,
+      "dashed": 12414,
+      "ĠKh": 12415,
+      "quark": 12416,
+      "embles": 12417,
+      "Ġidentifies": 12418,
+      "Ġfollic": 12419,
+      "Ġmetam": 12420,
+      "ĠHerm": 12421,
+      "Ġtobacco": 12422,
+      "Ġrealization": 12423,
+      "hydrox": 12424,
+      "ĠBet": 12425,
+      "Because": 12426,
+      "Ġpieces": 12427,
+      "Ġtalk": 12428,
+      "Ġopened": 12429,
+      "asome": 12430,
+      "Ġsurge": 12431,
+      "Ġfluctuation": 12432,
+      "github": 12433,
+      "ĠBacter": 12434,
+      "Ġbinds": 12435,
+      "ĠRapid": 12436,
+      "auer": 12437,
+      "pH": 12438,
+      "embed": 12439,
+      "ĠDoc": 12440,
+      "uchi": 12441,
+      "ĠCandid": 12442,
+      "Ġrarely": 12443,
+      "Ġmountain": 12444,
+      "ĠFat": 12445,
+      "Ġsend": 12446,
+      "ovsk": 12447,
+      "ĠOrganization": 12448,
+      "ĠFranc": 12449,
+      "ĠOP": 12450,
+      "âĪ¼": 12451,
+      "okes": 12452,
+      "ece": 12453,
+      "deficient": 12454,
+      "Ġlinkage": 12455,
+      "odon": 12456,
+      "Ġfly": 12457,
+      "Ġtidal": 12458,
+      "ĠExamples": 12459,
+      "ĠRout": 12460,
+      "Ġaccommod": 12461,
+      "Suppose": 12462,
+      "adap": 12463,
+      "Ġdie": 12464,
+      "root": 12465,
+      "Ġhon": 12466,
+      "Ġminimizing": 12467,
+      "Ġroughness": 12468,
+      "Ġgrass": 12469,
+      "enta": 12470,
+      "ĠLang": 12471,
+      "edu": 12472,
+      "ĠSimple": 12473,
+      "enic": 12474,
+      "Ġinducing": 12475,
+      "tf": 12476,
+      "Ġcontexts": 12477,
+      "ĠGeneralized": 12478,
+      "ĠWnt": 12479,
+      "Pb": 12480,
+      "atomic": 12481,
+      "dem": 12482,
+      "ĠPreparation": 12483,
+      "Ġinsufficient": 12484,
+      "sam": 12485,
+      "ĠSpecies": 12486,
+      "ĠSolar": 12487,
+      "Ġunsigned": 12488,
+      "ĠHER": 12489,
+      "âĬ": 12490,
+      "Ġparity": 12491,
+      "Ġnitrate": 12492,
+      "ĠCer": 12493,
+      "ptic": 12494,
+      "identif": 12495,
+      "geal": 12496,
+      "Ġemotion": 12497,
+      "ĠLP": 12498,
+      "Ġenhancing": 12499,
+      "Ġmeaningful": 12500,
+      "station": 12501,
+      "Ġrelig": 12502,
+      "yo": 12503,
+      "Ġperspectives": 12504,
+      "Ġscans": 12505,
+      "uginosa": 12506,
+      "Ġsummarize": 12507,
+      "relations": 12508,
+      "Ġdistant": 12509,
+      "Ġfunctionality": 12510,
+      "Ġdeeper": 12511,
+      "olate": 12512,
+      "ĠPor": 12513,
+      "graphs": 12514,
+      "ĠWa": 12515,
+      "ophilic": 12516,
+      "CLUS": 12517,
+      "ropathy": 12518,
+      "Ġcred": 12519,
+      "Ġuniversity": 12520,
+      "seg": 12521,
+      "vee": 12522,
+      "OG": 12523,
+      "ĠMen": 12524,
+      "ĠCritical": 12525,
+      "ãģ": 12526,
+      "Ġexit": 12527,
+      "vartheta": 12528,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 12529,
+      "Ġunf": 12530,
+      "Ġproposal": 12531,
+      "Ġtyrosine": 12532,
+      "otides": 12533,
+      "Ġproximity": 12534,
+      "Ġboxes": 12535,
+      "caten": 12536,
+      "ĠEnvironmental": 12537,
+      "bounded": 12538,
+      "downarrow": 12539,
+      "Ġfalls": 12540,
+      "Ġfertil": 12541,
+      "Ġcomprised": 12542,
+      "Ġmellitus": 12543,
+      "Ġleakage": 12544,
+      "uty": 12545,
+      "Ġchromosomes": 12546,
+      "ĠStatistics": 12547,
+      "%%%%": 12548,
+      "Ġcombinator": 12549,
+      "Ġket": 12550,
+      "advant": 12551,
+      "Ther": 12552,
+      "Ġtopics": 12553,
+      "flat": 12554,
+      "nia": 12555,
+      "ĠSpectral": 12556,
+      "Ġsynchronization": 12557,
+      "varrho": 12558,
+      "Ġcolonies": 12559,
+      "ĠFive": 12560,
+      "agues": 12561,
+      "ĠFC": 12562,
+      "IDS": 12563,
+      "Ġaward": 12564,
+      "Ġyielding": 12565,
+      "Ġarchitectures": 12566,
+      "ashington": 12567,
+      "chitz": 12568,
+      "perty": 12569,
+      "Ġmoduli": 12570,
+      "moment": 12571,
+      "speed": 12572,
+      "Ġmesenchymal": 12573,
+      "optera": 12574,
+      "Ġincomp": 12575,
+      "Cell": 12576,
+      "ĠMice": 12577,
+      "Ġgot": 12578,
+      "teger": 12579,
+      "Ġtau": 12580,
+      "ĠAdS": 12581,
+      "Ġbill": 12582,
+      "Ġdrinking": 12583,
+      "ulsive": 12584,
+      "Ġknockdown": 12585,
+      "Ġarms": 12586,
+      "ĠAutom": 12587,
+      "ĠIncreased": 12588,
+      "HF": 12589,
+      "Ġglobally": 12590,
+      "Ġdoping": 12591,
+      "Ġath": 12592,
+      "ĠCop": 12593,
+      "Ġsuccessive": 12594,
+      "ULT": 12595,
+      "eless": 12596,
+      "Ġbleeding": 12597,
+      "Ġfoods": 12598,
+      "Ġimmunohist": 12599,
+      "Ġdefinite": 12600,
+      "ĠJones": 12601,
+      "ĠTS": 12602,
+      "Ġjoined": 12603,
+      "ĠTowards": 12604,
+      "ĠCs": 12605,
+      "Ġunlike": 12606,
+      "Ġvalence": 12607,
+      "dor": 12608,
+      "oS": 12609,
+      "Ġpush": 12610,
+      "Ġoffice": 12611,
+      "Ġaluminum": 12612,
+      "idyl": 12613,
+      "idirectional": 12614,
+      "written": 12615,
+      "Ġbubble": 12616,
+      "HI": 12617,
+      "Ġmarkedly": 12618,
+      "ĠTok": 12619,
+      "Ġvesicles": 12620,
+      "Ġquotient": 12621,
+      "Ġreproduce": 12622,
+      "Ġelsewhere": 12623,
+      "ĠMyc": 12624,
+      "Ġinfrastructure": 12625,
+      "Ġgained": 12626,
+      "abel": 12627,
+      "ĠSex": 12628,
+      "ĠTables": 12629,
+      "etin": 12630,
+      "Ġhomolog": 12631,
+      "Ġlegal": 12632,
+      "hea": 12633,
+      "Ġsociety": 12634,
+      "Ġmanaged": 12635,
+      "idase": 12636,
+      "ĠInhibition": 12637,
+      "Ġparasite": 12638,
+      "Ġvolunte": 12639,
+      "ATP": 12640,
+      "ios": 12641,
+      "Ġsepsis": 12642,
+      "Ġribosomal": 12643,
+      "Ġconfound": 12644,
+      "ĠStaphyl": 12645,
+      "aryngeal": 12646,
+      "ïĢ": 12647,
+      "comb": 12648,
+      "ĠObjective": 12649,
+      "SULTS": 12650,
+      "Ġthorough": 12651,
+      "mt": 12652,
+      "Ġchest": 12653,
+      "Vector": 12654,
+      "element": 12655,
+      "Ġvirulence": 12656,
+      "Ġhemisp": 12657,
+      "Ġsought": 12658,
+      "ĠKo": 12659,
+      "Ġnutrition": 12660,
+      "uling": 12661,
+      "iana": 12662,
+      "Ġprototype": 12663,
+      "ĠOnt": 12664,
+      "cine": 12665,
+      "Ġdotted": 12666,
+      "Ġobese": 12667,
+      "ountered": 12668,
+      "Ġphysicians": 12669,
+      "Ġmini": 12670,
+      "Ľľ": 12671,
+      "spaces": 12672,
+      "Ġexclusively": 12673,
+      "ĠConvolution": 12674,
+      "Ġcaspase": 12675,
+      "ĠLink": 12676,
+      "div": 12677,
+      "ĠRoyal": 12678,
+      "hist": 12679,
+      "itness": 12680,
+      "Ġester": 12681,
+      "Ġconducting": 12682,
+      "Ġparticipated": 12683,
+      "Ġairway": 12684,
+      "Ġaeruginosa": 12685,
+      "Ext": 12686,
+      "argument": 12687,
+      "ocking": 12688,
+      "Ġintegrate": 12689,
+      "Ġcontrovers": 12690,
+      "apes": 12691,
+      "training": 12692,
+      "ĠPrevalence": 12693,
+      "temp": 12694,
+      "both": 12695,
+      "Ġreactivity": 12696,
+      "Ġranking": 12697,
+      "Ġtunneling": 12698,
+      "ODE": 12699,
+      "ĠMediterranean": 12700,
+      "Ġresonances": 12701,
+      "Mg": 12702,
+      "Ġlib": 12703,
+      "ĠHeter": 12704,
+      "Ġnothing": 12705,
+      "Ġindication": 12706,
+      "ĠHM": 12707,
+      "ocytic": 12708,
+      "strand": 12709,
+      "Ġcollaboration": 12710,
+      "Ġelectrostatic": 12711,
+      "Ġindependence": 12712,
+      "hab": 12713,
+      "Ġconflic": 12714,
+      "Ġiod": 12715,
+      "inus": 12716,
+      "Ġdependency": 12717,
+      "ĠLam": 12718,
+      "Ġexamining": 12719,
+      "Ġoccupied": 12720,
+      "Ġqueue": 12721,
+      "ĠBul": 12722,
+      "Ġregistered": 12723,
+      "Ġindividually": 12724,
+      "Rx": 12725,
+      "ausal": 12726,
+      "VE": 12727,
+      "Ġbrightness": 12728,
+      "respons": 12729,
+      "balance": 12730,
+      "Ġcytotoxic": 12731,
+      "fall": 12732,
+      "commut": 12733,
+      "ICAL": 12734,
+      "uran": 12735,
+      "aining": 12736,
+      "raulic": 12737,
+      "results": 12738,
+      "Ġepisodes": 12739,
+      "YS": 12740,
+      "ĠGar": 12741,
+      "Ġsurfact": 12742,
+      "drug": 12743,
+      "Ġcities": 12744,
+      "ĠChange": 12745,
+      "osition": 12746,
+      "Ġtriggered": 12747,
+      "Ġcytoplasmic": 12748,
+      "erves": 12749,
+      "Ġlex": 12750,
+      "Ġasymptotically": 12751,
+      "phy": 12752,
+      "Ġfrontal": 12753,
+      "ĠDensity": 12754,
+      "Ġsynerg": 12755,
+      "cycle": 12756,
+      "ĠImproved": 12757,
+      "ø": 12758,
+      "Ġmono": 12759,
+      "Ġaccumulated": 12760,
+      "oriented": 12761,
+      "bour": 12762,
+      "Ġtunnel": 12763,
+      "coming": 12764,
+      "Ġapparatus": 12765,
+      "Ġencountered": 12766,
+      "Cre": 12767,
+      "Ġletters": 12768,
+      "etch": 12769,
+      "Ġexcessive": 12770,
+      "Ġbiofilm": 12771,
+      "Ġrearrang": 12772,
+      "Ġpolymorphisms": 12773,
+      "erobic": 12774,
+      "Ġconnect": 12775,
+      "resolved": 12776,
+      "ĠNN": 12777,
+      "Ġretro": 12778,
+      "ĠIniti": 12779,
+      "ĠQuantif": 12780,
+      "Ġpup": 12781,
+      "Tensor": 12782,
+      "Ġsentences": 12783,
+      "lay": 12784,
+      "rants": 12785,
+      "ploid": 12786,
+      "ĠAnderson": 12787,
+      "Ġdesirable": 12788,
+      "stud": 12789,
+      "iability": 12790,
+      "Ġdrying": 12791,
+      "ecess": 12792,
+      "Ġdens": 12793,
+      "Ġdescript": 12794,
+      "ĠËĨ": 12795,
+      "Ġclones": 12796,
+      "Ġjuven": 12797,
+      "bp": 12798,
+      "Ġkil": 12799,
+      "HL": 12800,
+      "Ġhemorrh": 12801,
+      "ĠKi": 12802,
+      "How": 12803,
+      "Ġenerge": 12804,
+      "Ġsubsection": 12805,
+      "ĠSac": 12806,
+      "dial": 12807,
+      "Ġcardiomy": 12808,
+      "Ġtouch": 12809,
+      "dm": 12810,
+      "Ġscienti": 12811,
+      "oides": 12812,
+      "ĠÃĤ": 12813,
+      "ysaccharide": 12814,
+      "Ġsclerosis": 12815,
+      "ĠZealand": 12816,
+      "inine": 12817,
+      "Ġunusual": 12818,
+      "ĠBA": 12819,
+      "ipschitz": 12820,
+      "gap": 12821,
+      "ĠDifferences": 12822,
+      "Ġduality": 12823,
+      "edical": 12824,
+      "Ġlign": 12825,
+      "Ġfails": 12826,
+      "Ġlect": 12827,
+      "Ġrelate": 12828,
+      "Ġincorrect": 12829,
+      "Ġspecify": 12830,
+      "Ġcylindrical": 12831,
+      "ĠPF": 12832,
+      "ĠLind": 12833,
+      "Ġdeterior": 12834,
+      "Ġherb": 12835,
+      "dz": 12836,
+      "Ġweld": 12837,
+      "Ġnominal": 12838,
+      "copy": 12839,
+      "Ġacetyl": 12840,
+      "html": 12841,
+      "Ġrecognize": 12842,
+      "***": 12843,
+      "itian": 12844,
+      "WA": 12845,
+      "ĠMN": 12846,
+      "ĠFind": 12847,
+      "Ġauthentic": 12848,
+      "perture": 12849,
+      "Ġcytotoxicity": 12850,
+      "ofl": 12851,
+      "ĠGet": 12852,
+      "Ġcohomology": 12853,
+      "Ġremainder": 12854,
+      "Ġexpanding": 12855,
+      "Ġheav": 12856,
+      "osterone": 12857,
+      "Right": 12858,
+      "Ġcopol": 12859,
+      "Ġshed": 12860,
+      "Ġcompliance": 12861,
+      "Ġacidic": 12862,
+      "oric": 12863,
+      "Ġamyloid": 12864,
+      "Ġevaporation": 12865,
+      "dl": 12866,
+      "Ġdelays": 12867,
+      "Po": 12868,
+      "ĠCHECK": 12869,
+      "tains": 12870,
+      "Ġreversed": 12871,
+      "ĠMPa": 12872,
+      "Ġprocessor": 12873,
+      "Ġhall": 12874,
+      "ĠLast": 12875,
+      "Ġplasm": 12876,
+      "ĠAssociated": 12877,
+      "ĠBasic": 12878,
+      "inos": 12879,
+      "Ġsymptom": 12880,
+      "ãĢ": 12881,
+      "Ġanthrop": 12882,
+      "Ġjudg": 12883,
+      "Ġeti": 12884,
+      "kle": 12885,
+      "Ġwrong": 12886,
+      "room": 12887,
+      "Ġdevelopments": 12888,
+      "ĠMaximum": 12889,
+      "Ġcoatings": 12890,
+      "Ġheuristic": 12891,
+      "rontal": 12892,
+      "Some": 12893,
+      "Ġutilize": 12894,
+      "ĠâĪħ": 12895,
+      "coll": 12896,
+      "ĠRelated": 12897,
+      "Ġdegeneration": 12898,
+      "template": 12899,
+      "Ġmodulated": 12900,
+      "Ġparametri": 12901,
+      "Ġsaliv": 12902,
+      "ĠPseudomonas": 12903,
+      "Ġantigens": 12904,
+      "Ġharmon": 12905,
+      "ĠLHC": 12906,
+      "doi": 12907,
+      "ensitive": 12908,
+      "ĠNotice": 12909,
+      "ĠMoh": 12910,
+      "tilage": 12911,
+      "ACS": 12912,
+      "Ġdiscrepancy": 12913,
+      "Ġspik": 12914,
+      "Ġrestrict": 12915,
+      "itrile": 12916,
+      "leg": 12917,
+      "ĠBase": 12918,
+      "Ġconvolutional": 12919,
+      "ĠResistance": 12920,
+      "Ġappearing": 12921,
+      "ĠImages": 12922,
+      "ĠMann": 12923,
+      "Ġreact": 12924,
+      "Ġmacrophage": 12925,
+      "Ġwavelet": 12926,
+      "ochrom": 12927,
+      "Ġfairly": 12928,
+      "Ġpreceding": 12929,
+      "Ġspir": 12930,
+      "network": 12931,
+      "ĠNak": 12932,
+      "IFT": 12933,
+      "Ġago": 12934,
+      "Ġencryp": 12935,
+      "ald": 12936,
+      "ensin": 12937,
+      "Ġsulph": 12938,
+      "ĠPolymer": 12939,
+      "ĠArt": 12940,
+      "Ġsubunits": 12941,
+      "shot": 12942,
+      "Ġbegins": 12943,
+      "Ġexer": 12944,
+      "propto": 12945,
+      "Ġnurses": 12946,
+      "Ġsuffices": 12947,
+      "Ġgraded": 12948,
+      "ĠRock": 12949,
+      "Ġuniquely": 12950,
+      "itol": 12951,
+      "Ġspiral": 12952,
+      "Ġthanks": 12953,
+      "character": 12954,
+      "ĠDistributed": 12955,
+      "ĠCart": 12956,
+      "Form": 12957,
+      "Ġformulations": 12958,
+      "ictionary": 12959,
+      "Ġspreading": 12960,
+      "Ġsingularity": 12961,
+      "Ġpigs": 12962,
+      "itu": 12963,
+      "otrophic": 12964,
+      "ÑĢ": 12965,
+      "Ġsemiconductor": 12966,
+      "Ġdrag": 12967,
+      "next": 12968,
+      "maxim": 12969,
+      "unn": 12970,
+      "Ġargued": 12971,
+      "plastic": 12972,
+      "Ġdehydrogenase": 12973,
+      "Ġreinforcement": 12974,
+      "entral": 12975,
+      "ĠDS": 12976,
+      "Ġcompanies": 12977,
+      "Ġquantization": 12978,
+      "ĠDri": 12979,
+      "Ġsimpler": 12980,
+      "Ġradii": 12981,
+      "ĠEthics": 12982,
+      "ĠElectronic": 12983,
+      "taken": 12984,
+      "Ġpharmacological": 12985,
+      "pson": 12986,
+      "Ġpairing": 12987,
+      "Ġnest": 12988,
+      "ĠRS": 12989,
+      "Ġlic": 12990,
+      "ocon": 12991,
+      "Ġobserving": 12992,
+      "ĠFM": 12993,
+      "IES": 12994,
+      "Ġsubmitted": 12995,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 12996,
+      "Ġnoisy": 12997,
+      "Ġvanishing": 12998,
+      "ĠTechnologies": 12999,
+      "ilst": 13000,
+      "agic": 13001,
+      "Ġembeddings": 13002,
+      "Ġplans": 13003,
+      "reak": 13004,
+      "oct": 13005,
+      "Ġepithelium": 13006,
+      "Ġreversible": 13007,
+      "Ġrequests": 13008,
+      "Vi": 13009,
+      "ĠProg": 13010,
+      "methoxy": 13011,
+      "uria": 13012,
+      "Ġslice": 13013,
+      "Ġmetastases": 13014,
+      "ĠMary": 13015,
+      "Ġpriori": 13016,
+      "Ġexplains": 13017,
+      "ĠSigma": 13018,
+      "ĠArmy": 13019,
+      "Ġprey": 13020,
+      "KL": 13021,
+      "ĠPass": 13022,
+      "Ġreproduction": 13023,
+      "Ġfermentation": 13024,
+      "ulo": 13025,
+      "Ġproofs": 13026,
+      "ĠAccordingly": 13027,
+      "tist": 13028,
+      "ĠïĢ©": 13029,
+      "Ġmeat": 13030,
+      "Ġplanned": 13031,
+      "Ġangiogenesis": 13032,
+      "WR": 13033,
+      "ĠAust": 13034,
+      "Similarly": 13035,
+      "ĠWashington": 13036,
+      "Ġrefinement": 13037,
+      "Ġembryo": 13038,
+      "Ġdissociation": 13039,
+      "án": 13040,
+      "plasia": 13041,
+      "ĠGro": 13042,
+      "Ġsimilarities": 13043,
+      "Ġsolubility": 13044,
+      "Ġimmobil": 13045,
+      "ĠScot": 13046,
+      "ĠSubsequently": 13047,
+      "divid": 13048,
+      "Ġclosest": 13049,
+      "ĠWat": 13050,
+      "ĠâĮ": 13051,
+      "ĠAGN": 13052,
+      "Ġprescribed": 13053,
+      "Ġmosquito": 13054,
+      "Ġfirm": 13055,
+      "Ġdegenerate": 13056,
+      "Ġethyl": 13057,
+      "Ġharvest": 13058,
+      "ĠSpecific": 13059,
+      "Ġcompartment": 13060,
+      "public": 13061,
+      "ĠBiological": 13062,
+      "Ġpiece": 13063,
+      "Ġattitudes": 13064,
+      "Ġspray": 13065,
+      "ĠSix": 13066,
+      "Ġprofessionals": 13067,
+      "Ġslot": 13068,
+      "Ġretrieved": 13069,
+      "vement": 13070,
+      "Ġexecuted": 13071,
+      "seed": 13072,
+      "Ġoutflow": 13073,
+      "distance": 13074,
+      "ĠTerm": 13075,
+      "ady": 13076,
+      "ĠProvince": 13077,
+      "ĠCentre": 13078,
+      "ĠDFT": 13079,
+      "Ġsudden": 13080,
+      "Ġseiz": 13081,
+      "rat": 13082,
+      "romo": 13083,
+      "otechn": 13084,
+      "Ġhighlights": 13085,
+      "Ġelectrolyte": 13086,
+      "ĠAdvanced": 13087,
+      "allow": 13088,
+      "px": 13089,
+      "osed": 13090,
+      "subarray": 13091,
+      "racks": 13092,
+      "PRO": 13093,
+      "ogeny": 13094,
+      "Ġpooled": 13095,
+      "Ġdtype": 13096,
+      "Ġopposed": 13097,
+      "ĠGrand": 13098,
+      "Ġdesigning": 13099,
+      "bel": 13100,
+      "itability": 13101,
+      "Ġminimization": 13102,
+      "Ġdramatically": 13103,
+      "Ġsoy": 13104,
+      "agents": 13105,
+      "ĠMetal": 13106,
+      "ĠMV": 13107,
+      "ribute": 13108,
+      "DD": 13109,
+      "itan": 13110,
+      "Ġspeeds": 13111,
+      "Ġmarried": 13112,
+      "Ġevaluations": 13113,
+      "ĠKingdom": 13114,
+      "Ġclay": 13115,
+      "ĠTissue": 13116,
+      "leftarrow": 13117,
+      "Ġcompensation": 13118,
+      "child": 13119,
+      "pool": 13120,
+      "uparrow": 13121,
+      "ĠDomain": 13122,
+      "species": 13123,
+      "Ġmethane": 13124,
+      "ĠEGFR": 13125,
+      "Ġparser": 13126,
+      "have": 13127,
+      "Ġneglected": 13128,
+      "func": 13129,
+      "apsed": 13130,
+      "Ġsays": 13131,
+      "adata": 13132,
+      "binom": 13133,
+      "Case": 13134,
+      "Ġreporter": 13135,
+      "Sn": 13136,
+      "Ġmaximize": 13137,
+      "Ġbifurc": 13138,
+      "ĠCNS": 13139,
+      "ĠOlymp": 13140,
+      "Ġdeclare": 13141,
+      "Ġencoder": 13142,
+      "Ġabelian": 13143,
+      "Ġsingularities": 13144,
+      "Ġech": 13145,
+      "Ψ": 13146,
+      "Ġproto": 13147,
+      "Ġphag": 13148,
+      "Ġpolyg": 13149,
+      "Ġbott": 13150,
+      "Ġadipose": 13151,
+      "uing": 13152,
+      "jk": 13153,
+      "uchy": 13154,
+      "ĠStudent": 13155,
+      "Ġnanow": 13156,
+      "Ġthym": 13157,
+      "Ed": 13158,
+      "End": 13159,
+      "Ġtransforms": 13160,
+      "ĠPCA": 13161,
+      "kern": 13162,
+      "regn": 13163,
+      "Ġcomment": 13164,
+      "ĠLL": 13165,
+      "elles": 13166,
+      "Ġengagement": 13167,
+      "ĠPeter": 13168,
+      "ISPR": 13169,
+      "ĠChannel": 13170,
+      "iny": 13171,
+      "Ġbundles": 13172,
+      "Ald": 13173,
+      "Ġpublications": 13174,
+      "TG": 13175,
+      "stra": 13176,
+      "Ġfear": 13177,
+      "Ġretic": 13178,
+      "plements": 13179,
+      "Ġcorpus": 13180,
+      "ĠCluster": 13181,
+      "ĠRate": 13182,
+      "Ġsimplest": 13183,
+      "acic": 13184,
+      "rbrack": 13185,
+      "Ġblow": 13186,
+      "Ġcompress": 13187,
+      "ĠDark": 13188,
+      "Ġpsychiatric": 13189,
+      "ĠConversely": 13190,
+      "Ġowing": 13191,
+      "Ġabsor": 13192,
+      "ĠHP": 13193,
+      "Ġcrude": 13194,
+      "equal": 13195,
+      "ĠArray": 13196,
+      "ĠRelative": 13197,
+      "Ġcombustion": 13198,
+      "Red": 13199,
+      "kt": 13200,
+      "ĠmA": 13201,
+      "Ġtex": 13202,
+      "porters": 13203,
+      "Ġdiffered": 13204,
+      "Ġaudio": 13205,
+      "zon": 13206,
+      "odi": 13207,
+      "Ġmacroscopic": 13208,
+      "acin": 13209,
+      "Ġzeros": 13210,
+      "Ġforeign": 13211,
+      "Ġduct": 13212,
+      "bow": 13213,
+      "worth": 13214,
+      "ĠRoad": 13215,
+      "rey": 13216,
+      "aceous": 13217,
+      "Ġblast": 13218,
+      "Ġgranul": 13219,
+      "Ġwing": 13220,
+      "Ġannotated": 13221,
+      "ĠFull": 13222,
+      "Ġinfluencing": 13223,
+      "vy": 13224,
+      "iazol": 13225,
+      "Ġpitch": 13226,
+      "Ġrehabilitation": 13227,
+      "ĠPrior": 13228,
+      "comit": 13229,
+      "mathtt": 13230,
+      "dia": 13231,
+      "ĠIon": 13232,
+      "Ġabuse": 13233,
+      "Ġharvested": 13234,
+      "Ġepidemic": 13235,
+      "Ġfilament": 13236,
+      "Ġnucleation": 13237,
+      "ĠKnowledge": 13238,
+      "rinos": 13239,
+      "Ġbent": 13240,
+      "Ġsquared": 13241,
+      "Ġhippocampal": 13242,
+      "ĠTG": 13243,
+      "ANT": 13244,
+      "modified": 13245,
+      "ario": 13246,
+      "ĠFace": 13247,
+      "Ġgrows": 13248,
+      "Ġfaults": 13249,
+      "virus": 13250,
+      "Ġpartitioning": 13251,
+      "airs": 13252,
+      "Ġhearing": 13253,
+      "Ġcongen": 13254,
+      "Ġrip": 13255,
+      "ĠCollabor": 13256,
+      "Ġinterviews": 13257,
+      "Ġhuge": 13258,
+      "Ġbreakdown": 13259,
+      "Ġmonthly": 13260,
+      "ĠCONCLUS": 13261,
+      "Each": 13262,
+      "Diff": 13263,
+      "Ġrelay": 13264,
+      "ĠMuse": 13265,
+      "oscopy": 13266,
+      "Ġrenew": 13267,
+      "gb": 13268,
+      "Ġbrid": 13269,
+      "Ġoutlined": 13270,
+      "orig": 13271,
+      "eat": 13272,
+      "ĠWithout": 13273,
+      "Ġspor": 13274,
+      "ĠTN": 13275,
+      "ĠJo": 13276,
+      "ĠAU": 13277,
+      "Not": 13278,
+      "Ġretin": 13279,
+      "ĠAngel": 13280,
+      "Ġtried": 13281,
+      "eyond": 13282,
+      "je": 13283,
+      "ĠRussian": 13284,
+      "ĠUnfortunately": 13285,
+      "ĠMeanwhile": 13286,
+      "ographs": 13287,
+      "Ġaccounting": 13288,
+      "ĠAβ": 13289,
+      "mb": 13290,
+      "Ġdopamine": 13291,
+      "ĠBriefly": 13292,
+      "ĠFrequency": 13293,
+      "Matrix": 13294,
+      "ĠJoseph": 13295,
+      "Ġexperts": 13296,
+      "Ġdrops": 13297,
+      "ĠRESULTS": 13298,
+      "Ġrectangular": 13299,
+      "athione": 13300,
+      "center": 13301,
+      "ĠLeft": 13302,
+      "inform": 13303,
+      "kins": 13304,
+      "Ġmil": 13305,
+      "ĠMah": 13306,
+      "Ġmedial": 13307,
+      "ĠCompany": 13308,
+      "Ġpassage": 13309,
+      "Ġleader": 13310,
+      "Ġscreened": 13311,
+      "eri": 13312,
+      "posites": 13313,
+      "rarily": 13314,
+      "Ġphone": 13315,
+      "ietic": 13316,
+      "Ġexpectations": 13317,
+      "ĠParticle": 13318,
+      "ĠMountain": 13319,
+      "Ġinterleukin": 13320,
+      "Ġfifth": 13321,
+      "Ġvast": 13322,
+      "Ġlogical": 13323,
+      "Ġterr": 13324,
+      "Ġcreates": 13325,
+      "Ġfinitely": 13326,
+      "Ġswim": 13327,
+      "Ġsupernatant": 13328,
+      "opathological": 13329,
+      "ĠUltra": 13330,
+      "ĠTy": 13331,
+      "Ġgrand": 13332,
+      "Ġconstitute": 13333,
+      "ologist": 13334,
+      "ĠBroad": 13335,
+      "aware": 13336,
+      "Ġvicinity": 13337,
+      "agulation": 13338,
+      "unsigned": 13339,
+      "ĠSize": 13340,
+      "ĠCognitive": 13341,
+      "Ġsuspected": 13342,
+      "Ġupl": 13343,
+      "Ġautoimmune": 13344,
+      "ĠSK": 13345,
+      "CB": 13346,
+      "Ġslices": 13347,
+      "ĠChi": 13348,
+      "Ġobservables": 13349,
+      "Ġhippocampus": 13350,
+      "sover": 13351,
+      "Ġfunding": 13352,
+      "Ġconformation": 13353,
+      "ĠQuestion": 13354,
+      "ĠSqu": 13355,
+      "ĠWill": 13356,
+      "Ġscattered": 13357,
+      "irty": 13358,
+      "Ġplaus": 13359,
+      "correlation": 13360,
+      "Ġventilation": 13361,
+      "ĠGenes": 13362,
+      "Ġbenign": 13363,
+      "Ġhetero": 13364,
+      "Status": 13365,
+      "angled": 13366,
+      "Ġbootstrap": 13367,
+      "Ġvaccines": 13368,
+      "Ġmicroorganisms": 13369,
+      "Ġvisits": 13370,
+      "Ġtheorems": 13371,
+      "drop": 13372,
+      "ĠTA": 13373,
+      "Ġcycling": 13374,
+      "Ġspectrometer": 13375,
+      "Ġgroundwater": 13376,
+      "Ġnanotubes": 13377,
+      "Ġjoints": 13378,
+      "ĠEll": 13379,
+      "Ġconsult": 13380,
+      "Ġwindows": 13381,
+      "Ġdisability": 13382,
+      "Ġgains": 13383,
+      "Ġdischarg": 13384,
+      "Ġheated": 13385,
+      "Ġafore": 13386,
+      "arying": 13387,
+      "incre": 13388,
+      "Ġaggressive": 13389,
+      "Ġhemod": 13390,
+      "arium": 13391,
+      "ĠInst": 13392,
+      "vm": 13393,
+      "Ġdroplet": 13394,
+      "ptive": 13395,
+      "viously": 13396,
+      "Ġstarch": 13397,
+      "Ġdf": 13398,
+      "osyl": 13399,
+      "Ġdonors": 13400,
+      "ĠUnlike": 13401,
+      "Ġalkaline": 13402,
+      "Ġintelligence": 13403,
+      "aa": 13404,
+      "Ġacceptance": 13405,
+      "Ġsliding": 13406,
+      "apses": 13407,
+      "ĠDiss": 13408,
+      "istan": 13409,
+      "auc": 13410,
+      "Ġbins": 13411,
+      "Ġmodulate": 13412,
+      "Ġmanage": 13413,
+      "outs": 13414,
+      "Ġsenes": 13415,
+      "Ġdifferentiate": 13416,
+      "Ġcounted": 13417,
+      "ASK": 13418,
+      "Ġantibacterial": 13419,
+      "Ġentered": 13420,
+      "Ġdisadvant": 13421,
+      "ĠSalmonella": 13422,
+      "Ġisotopic": 13423,
+      "Ġannounced": 13424,
+      "ĠBoard": 13425,
+      "Ġrestoration": 13426,
+      "Ġallevi": 13427,
+      "Ġprogramme": 13428,
+      "Ġalbumin": 13429,
+      "Ġcatalog": 13430,
+      "estine": 13431,
+      "Ġdifferently": 13432,
+      "Ġmolar": 13433,
+      "rödinger": 13434,
+      "ĠEvent": 13435,
+      "ministration": 13436,
+      "ĠSerum": 13437,
+      "ROM": 13438,
+      "kw": 13439,
+      "bot": 13440,
+      "Ġjets": 13441,
+      "ĠDouble": 13442,
+      "eler": 13443,
+      "Ġinfusion": 13444,
+      "Ġconsumed": 13445,
+      "ĠIron": 13446,
+      "ĠProcesses": 13447,
+      "Ġadmits": 13448,
+      "Ġjuris": 13449,
+      "ĠPeriod": 13450,
+      "Ġremodeling": 13451,
+      "alley": 13452,
+      "Ġenabling": 13453,
+      "Ġbackward": 13454,
+      "ĠMid": 13455,
+      "brevi": 13456,
+      "Ġclassify": 13457,
+      "Ġcrypt": 13458,
+      "Ġhelix": 13459,
+      "ĠJiang": 13460,
+      "Ġhoney": 13461,
+      "gestion": 13462,
+      "xc": 13463,
+      "Ġcoincides": 13464,
+      "ĠDN": 13465,
+      "Ġapoptotic": 13466,
+      "Ġinstall": 13467,
+      "ĠRever": 13468,
+      "ĠDoppler": 13469,
+      "icago": 13470,
+      "erals": 13471,
+      "Ġpie": 13472,
+      "ĠMars": 13473,
+      "ĠStaphylococcus": 13474,
+      "Ġnoting": 13475,
+      "Ġgenera": 13476,
+      "ĠIo": 13477,
+      "Ġhope": 13478,
+      "Ġpreserve": 13479,
+      "MAX": 13480,
+      "ynchron": 13481,
+      "Ġrup": 13482,
+      "Ġcomprising": 13483,
+      "ĠWay": 13484,
+      "Ġviolation": 13485,
+      "QR": 13486,
+      "Ġreflecting": 13487,
+      "Ġregularity": 13488,
+      "ĠSiO": 13489,
+      "ĠJun": 13490,
+      "Ġcommunications": 13491,
+      "rating": 13492,
+      "Ġfamiliar": 13493,
+      "Ġinstantaneous": 13494,
+      "Ġcortic": 13495,
+      "Ġapparently": 13496,
+      "XX": 13497,
+      "Ġexcitations": 13498,
+      "ĠAward": 13499,
+      "Num": 13500,
+      "ĠUN": 13501,
+      "Ġqubit": 13502,
+      "ĠAction": 13503,
+      "ĠFried": 13504,
+      "Ġeliminated": 13505,
+      "Ġaspir": 13506,
+      "hler": 13507,
+      "Ġdecoding": 13508,
+      "unov": 13509,
+      "Ġanalogue": 13510,
+      "ulmonary": 13511,
+      "Ġgeographic": 13512,
+      "Ġsort": 13513,
+      "ĠCRC": 13514,
+      "Aldrich": 13515,
+      "ĠkDa": 13516,
+      "ĠND": 13517,
+      "Ġsettle": 13518,
+      "exists": 13519,
+      "Ġstatistic": 13520,
+      "ĠBow": 13521,
+      "ĠCG": 13522,
+      "Ġorganizations": 13523,
+      "ĠMobile": 13524,
+      "Ġinvent": 13525,
+      "Ġincorporate": 13526,
+      "ĠFib": 13527,
+      "ordan": 13528,
+      "Ġcolleagues": 13529,
+      "ĠStation": 13530,
+      "Ġsen": 13531,
+      "Ġencaps": 13532,
+      "ĠRH": 13533,
+      "relim": 13534,
+      "Ġcarbonate": 13535,
+      "ĠNether": 13536,
+      "mem": 13537,
+      "EEE": 13538,
+      "Ġaforementioned": 13539,
+      "Ġpent": 13540,
+      "ĠSignal": 13541,
+      "Ġsuspended": 13542,
+      "Color": 13543,
+      "Ġspins": 13544,
+      "Ġproportions": 13545,
+      "ulty": 13546,
+      "Ġenrolled": 13547,
+      "ĠTEM": 13548,
+      "ĠReceptor": 13549,
+      "Ġprevalent": 13550,
+      "large": 13551,
+      "vs": 13552,
+      "Ġtruncated": 13553,
+      "Ġâĭħ": 13554,
+      "lm": 13555,
+      "anil": 13556,
+      "Ġannih": 13557,
+      "ĠGalaxy": 13558,
+      "eras": 13559,
+      "Ġepigenetic": 13560,
+      "Ġtooth": 13561,
+      "Ġcondensation": 13562,
+      "ĠTensor": 13563,
+      "Ġinorganic": 13564,
+      "ymers": 13565,
+      "uf": 13566,
+      "anese": 13567,
+      "aret": 13568,
+      "Ġarithmetic": 13569,
+      "âĨ": 13570,
+      "Ġtrying": 13571,
+      "Ġimplementing": 13572,
+      "xd": 13573,
+      "Ġillumination": 13574,
+      "ela": 13575,
+      "Ġdeficits": 13576,
+      "Ġspots": 13577,
+      "Ġdoesn": 13578,
+      "Ġresting": 13579,
+      "trained": 13580,
+      "Ġerosion": 13581,
+      "Ġgranular": 13582,
+      "Ġscar": 13583,
+      "Ġpollen": 13584,
+      "lie": 13585,
+      "Ġconvers": 13586,
+      "Ġdisturbances": 13587,
+      "ĠGod": 13588,
+      "Ġenlarg": 13589,
+      "ĠLate": 13590,
+      "ylase": 13591,
+      "Ġfacts": 13592,
+      "enty": 13593,
+      "ĠStreet": 13594,
+      "sequence": 13595,
+      "Ġvenous": 13596,
+      "ĠCheck": 13597,
+      "agg": 13598,
+      "Ġabsorbed": 13599,
+      "Ġcommit": 13600,
+      "sets": 13601,
+      "Ġdestroy": 13602,
+      "Ġbowel": 13603,
+      "Ġfinished": 13604,
+      "ĠFeed": 13605,
+      "Ġdoped": 13606,
+      "ĠAlb": 13607,
+      "ĠMitochond": 13608,
+      "Ġtheoretically": 13609,
+      "RI": 13610,
+      "Ġmeteor": 13611,
+      "ĠMG": 13612,
+      "Ġnation": 13613,
+      "ĠBasin": 13614,
+      "nik": 13615,
+      "Ġdepths": 13616,
+      "ĠMechanism": 13617,
+      "Ġmotifs": 13618,
+      "ĠHay": 13619,
+      "Ġmotivated": 13620,
+      "ĠCopy": 13621,
+      "ĠEastern": 13622,
+      "Ġpersistence": 13623,
+      "Ġrays": 13624,
+      "FB": 13625,
+      "andem": 13626,
+      "layers": 13627,
+      "eyer": 13628,
+      "ĠStrept": 13629,
+      "Ġregistration": 13630,
+      "ĠAntarctic": 13631,
+      "CV": 13632,
+      "ĠPap": 13633,
+      "ĠSpe": 13634,
+      "Ġsplicing": 13635,
+      "performance": 13636,
+      "Ġsemantics": 13637,
+      "Ġlocom": 13638,
+      "oblastoma": 13639,
+      "Ġmoney": 13640,
+      "Ġtransparent": 13641,
+      "Ġhr": 13642,
+      "ĠInteractions": 13643,
+      "Ġsap": 13644,
+      "Ġbiases": 13645,
+      "Ġteeth": 13646,
+      "ynolds": 13647,
+      "omethyl": 13648,
+      "ĠmV": 13649,
+      "Ġsolely": 13650,
+      "Ġorange": 13651,
+      "blast": 13652,
+      "ATIONS": 13653,
+      "call": 13654,
+      "opoietic": 13655,
+      "sided": 13656,
+      "ĠFox": 13657,
+      "ĠVideo": 13658,
+      "Ġinspection": 13659,
+      "Ġbuck": 13660,
+      "hesize": 13661,
+      "present": 13662,
+      "ĠAntib": 13663,
+      "Ġham": 13664,
+      "alam": 13665,
+      "ĠPG": 13666,
+      "ĠAE": 13667,
+      "Ġjoin": 13668,
+      "Ġmonocytes": 13669,
+      "estiv": 13670,
+      "Ġrandomised": 13671,
+      "Ġtranslocation": 13672,
+      "Ġincorporating": 13673,
+      "Ġprolifer": 13674,
+      "Ġodds": 13675,
+      "ITH": 13676,
+      "Ġran": 13677,
+      "Ġinstruction": 13678,
+      "Ġresolve": 13679,
+      "Ġft": 13680,
+      "ĠHead": 13681,
+      "Ġreagent": 13682,
+      "Ġadmitted": 13683,
+      "human": 13684,
+      "posure": 13685,
+      "ĠCha": 13686,
+      "ĠFr": 13687,
+      "Ġbroadcast": 13688,
+      "Ġnutrients": 13689,
+      "nob": 13690,
+      "Ġnotable": 13691,
+      "ĠIGF": 13692,
+      "ĠClearly": 13693,
+      "Ġquarks": 13694,
+      "Ġeukary": 13695,
+      "ĠAdd": 13696,
+      "itosan": 13697,
+      "Ġinteractive": 13698,
+      "itting": 13699,
+      "ĠComputational": 13700,
+      "Ġdissolution": 13701,
+      "istribution": 13702,
+      "product": 13703,
+      "ĠABC": 13704,
+      "olimits": 13705,
+      "biased": 13706,
+      "Ġtrapped": 13707,
+      "PK": 13708,
+      "ĠHPLC": 13709,
+      "rophot": 13710,
+      "zes": 13711,
+      "ourse": 13712,
+      "ĠHot": 13713,
+      "Ġrecipro": 13714,
+      "nolimits": 13715,
+      "ello": 13716,
+      "Ġassessments": 13717,
+      "ENTS": 13718,
+      "Ġalteration": 13719,
+      "tw": 13720,
+      "Ġchaotic": 13721,
+      "ĠLoc": 13722,
+      "Ġcattle": 13723,
+      "Ray": 13724,
+      "Ġformally": 13725,
+      "leave": 13726,
+      "textstyle": 13727,
+      "Ġventral": 13728,
+      "ĠWilliams": 13729,
+      "ĠPeople": 13730,
+      "ixing": 13731,
+      "ĠTherapy": 13732,
+      "Ġiii": 13733,
+      "ĠDT": 13734,
+      "Ġbic": 13735,
+      "Ġspheres": 13736,
+      "Ġvisc": 13737,
+      "Ġestablishment": 13738,
+      "Ġdescriptions": 13739,
+      "ĠAverage": 13740,
+      "Ġtour": 13741,
+      "ĠInfection": 13742,
+      "ĠLicense": 13743,
+      "Ġprepare": 13744,
+      "Hs": 13745,
+      "finite": 13746,
+      "rium": 13747,
+      "oreg": 13748,
+      "entry": 13749,
+      "Ġdisks": 13750,
+      "Ġelongation": 13751,
+      "cpu": 13752,
+      "ĠCharles": 13753,
+      "FIGURE": 13754,
+      "ston": 13755,
+      "ĠObservations": 13756,
+      "Add": 13757,
+      "ĠTask": 13758,
+      "atomy": 13759,
+      "igration": 13760,
+      "ĠDatabase": 13761,
+      "ĠTexas": 13762,
+      "Ġphyt": 13763,
+      "ller": 13764,
+      "conjug": 13765,
+      "onald": 13766,
+      "Ġheavily": 13767,
+      "Ġsple": 13768,
+      "Ġassist": 13769,
+      "ĠCp": 13770,
+      "Ġhappen": 13771,
+      "uv": 13772,
+      "ĠUniverse": 13773,
+      "ĠGPS": 13774,
+      "WE": 13775,
+      "Xi": 13776,
+      "Ġadministr": 13777,
+      "strong": 13778,
+      "Ġmagnitudes": 13779,
+      "Ġsimplify": 13780,
+      "Ġelegans": 13781,
+      "esh": 13782,
+      "ĠBody": 13783,
+      "ĠNetherlands": 13784,
+      "ï": 13785,
+      "ometers": 13786,
+      "Bo": 13787,
+      "FM": 13788,
+      "ĠNiger": 13789,
+      "plus": 13790,
+      "instance": 13791,
+      "Ġdistress": 13792,
+      "Organ": 13793,
+      "Cas": 13794,
+      "Ġsymplectic": 13795,
+      "Ġbreaks": 13796,
+      "ÑĤ": 13797,
+      "Ġfermion": 13798,
+      "emporal": 13799,
+      "Ġsomatic": 13800,
+      "event": 13801,
+      "neut": 13802,
+      "lammation": 13803,
+      "ĠLibrary": 13804,
+      "Ġmultiplic": 13805,
+      "ĠInstr": 13806,
+      "ethel": 13807,
+      "urys": 13808,
+      "Ġhelped": 13809,
+      "Ġcollege": 13810,
+      "Ġcartilage": 13811,
+      "Ġrpm": 13812,
+      "western": 13813,
+      "resis": 13814,
+      "Ġlobe": 13815,
+      "QL": 13816,
+      "Input": 13817,
+      "Ġemphasis": 13818,
+      "best": 13819,
+      "Ġtotally": 13820,
+      "ĠMETHOD": 13821,
+      "ĠFa": 13822,
+      "ĠReduction": 13823,
+      "icious": 13824,
+      "Ġimplantation": 13825,
+      "potential": 13826,
+      "problem": 13827,
+      "Ġobtains": 13828,
+      "urons": 13829,
+      "Ġconstructing": 13830,
+      "ĠMusic": 13831,
+      "Ġcancell": 13832,
+      "Ġnews": 13833,
+      "ĠChapter": 13834,
+      "Ġlabelled": 13835,
+      "Ġzebrafish": 13836,
+      "ĠSolid": 13837,
+      "Ġglutamate": 13838,
+      "ĉĉĉĉĉ": 13839,
+      "Ġchapter": 13840,
+      "ĠPresident": 13841,
+      "Min": 13842,
+      "Ġatrial": 13843,
+      "cp": 13844,
+      "fi": 13845,
+      "final": 13846,
+      "Ġtok": 13847,
+      "Ġeffector": 13848,
+      "Ġspine": 13849,
+      "Ġidentities": 13850,
+      "isco": 13851,
+      "olis": 13852,
+      "ĠCle": 13853,
+      "Ġinvariants": 13854,
+      "Path": 13855,
+      "ĠGon": 13856,
+      "factory": 13857,
+      "Ġexogenous": 13858,
+      "ĠMAPK": 13859,
+      "Ġanswers": 13860,
+      "Ġgetting": 13861,
+      "Rs": 13862,
+      "IH": 13863,
+      "ĠDefine": 13864,
+      "ĠConvolutional": 13865,
+      "Ġgeometrical": 13866,
+      "ĠInput": 13867,
+      "Ġà": 13868,
+      "Ġattenuated": 13869,
+      "Ġradicals": 13870,
+      "ĠAcademy": 13871,
+      "ãĥ": 13872,
+      "ichlet": 13873,
+      "Ġtorus": 13874,
+      "ĠTheoretical": 13875,
+      "ĠTD": 13876,
+      "Ġantiv": 13877,
+      "onge": 13878,
+      "Ġintravenous": 13879,
+      "Ġhypoth": 13880,
+      "Ġwastewater": 13881,
+      "ĠFlo": 13882,
+      "Ġporosity": 13883,
+      "Ġpall": 13884,
+      "aci": 13885,
+      "Ġrecordings": 13886,
+      "Ġeating": 13887,
+      "ĠDW": 13888,
+      "unting": 13889,
+      "ĠDim": 13890,
+      "Ġemitted": 13891,
+      "ĠJoint": 13892,
+      "ofib": 13893,
+      "Ġearthquake": 13894,
+      "Ġmunic": 13895,
+      "Ġreductions": 13896,
+      "Ġconjunction": 13897,
+      "ĠLocation": 13898,
+      "Ġestablishing": 13899,
+      "ĠMathematical": 13900,
+      "ĠSolution": 13901,
+      "buffer": 13902,
+      "arin": 13903,
+      "iley": 13904,
+      "ĠCommission": 13905,
+      "ĠGABA": 13906,
+      "ĠMuseum": 13907,
+      "Ġverb": 13908,
+      "lecules": 13909,
+      "infection": 13910,
+      "Ġinsect": 13911,
+      "iser": 13912,
+      "Ġprovision": 13913,
+      "Ġagreed": 13914,
+      "Ġafford": 13915,
+      "theory": 13916,
+      "knowledge": 13917,
+      "Protein": 13918,
+      "Ġkernels": 13919,
+      "Ġderm": 13920,
+      "Ġwish": 13921,
+      "Ġvox": 13922,
+      "Scale": 13923,
+      "hu": 13924,
+      "Ġcounterparts": 13925,
+      "ĠRoss": 13926,
+      "Ġunp": 13927,
+      "ĠOnline": 13928,
+      "Ġtransporter": 13929,
+      "Graph": 13930,
+      "Ġuter": 13931,
+      "Ġminute": 13932,
+      "Ġautomorphism": 13933,
+      "iltr": 13934,
+      "ĠRespons": 13935,
+      "ĠSym": 13936,
+      "Ġfactorization": 13937,
+      "sem": 13938,
+      "Ġmediates": 13939,
+      "Ġunexpected": 13940,
+      "Ġorganism": 13941,
+      "Ġattempted": 13942,
+      "aran": 13943,
+      "venue": 13944,
+      "etheless": 13945,
+      "Ġnoticed": 13946,
+      "ĠInvestigation": 13947,
+      "Ġcareg": 13948,
+      "Ġgrouped": 13949,
+      "orbit": 13950,
+      "Ġshortest": 13951,
+      "Ġbroader": 13952,
+      "ĠMIM": 13953,
+      "rises": 13954,
+      "veloper": 13955,
+      "ĠHi": 13956,
+      "ĠkHz": 13957,
+      "Ġbeads": 13958,
+      "Ġphyto": 13959,
+      "ĠDoes": 13960,
+      "Ġmammals": 13961,
+      "Ġrefined": 13962,
+      "volume": 13963,
+      "Ser": 13964,
+      "Ġresistivity": 13965,
+      "Ġterrestrial": 13966,
+      "Ġaxi": 13967,
+      "ifluor": 13968,
+      "Ġ£": 13969,
+      "Ġvice": 13970,
+      "ĠKel": 13971,
+      "VM": 13972,
+      "ĠTown": 13973,
+      "adm": 13974,
+      "plates": 13975,
+      "Ġholomorphic": 13976,
+      "ĠRib": 13977,
+      "ĠSB": 13978,
+      "ĠTemporal": 13979,
+      "src": 13980,
+      "Ġupdates": 13981,
+      "Ġseek": 13982,
+      "endix": 13983,
+      "oretic": 13984,
+      "warz": 13985,
+      "Ġroutes": 13986,
+      "Ġstanding": 13987,
+      "ĠÃģ": 13988,
+      "Ġclassic": 13989,
+      "Ġpale": 13990,
+      "lections": 13991,
+      "Ġclassifiers": 13992,
+      "Ġpathophys": 13993,
+      "Ġmounted": 13994,
+      "Ġdesignated": 13995,
+      "Ġvideos": 13996,
+      "Ġincoming": 13997,
+      "Ġguarantees": 13998,
+      "Ġparasites": 13999,
+      "ĠBacillus": 14000,
+      "four": 14001,
+      "ĠâĪ¨": 14002,
+      "Ġcommutative": 14003,
+      "stackrel": 14004,
+      "ĠBanach": 14005,
+      "Ġdealing": 14006,
+      "emporary": 14007,
+      "Multi": 14008,
+      "otomy": 14009,
+      "reting": 14010,
+      "Ġnond": 14011,
+      "ĠConference": 14012,
+      "tzmann": 14013,
+      "Ġphosphorus": 14014,
+      "Ġchemicals": 14015,
+      "Ġdispar": 14016,
+      "degree": 14017,
+      "Ġarbitrarily": 14018,
+      "rocyte": 14019,
+      "Ġparabolic": 14020,
+      "Ġdimensionless": 14021,
+      "Ġosm": 14022,
+      "Ġphonon": 14023,
+      "tiary": 14024,
+      "ĠSect": 14025,
+      "ophysical": 14026,
+      "ĠMapping": 14027,
+      "bis": 14028,
+      "ĠCommunication": 14029,
+      "Ġmimic": 14030,
+      "Ġregulators": 14031,
+      "Ġneutrophils": 14032,
+      "fn": 14033,
+      "ĠImportantly": 14034,
+      "Ġmere": 14035,
+      "Ġconfirms": 14036,
+      "agram": 14037,
+      "Ġattend": 14038,
+      "ungal": 14039,
+      "ĠGroups": 14040,
+      "Ġzo": 14041,
+      "Ġmouth": 14042,
+      "Ġsteep": 14043,
+      "Ġprevented": 14044,
+      "Ġdepressive": 14045,
+      "acies": 14046,
+      "ĠLS": 14047,
+      "Ġnitric": 14048,
+      "Ġvisualized": 14049,
+      "Ġtranscriptome": 14050,
+      "Ġgait": 14051,
+      "ercury": 14052,
+      "Ġshot": 14053,
+      "ĠVen": 14054,
+      "Ġexchang": 14055,
+      "Ġintention": 14056,
+      "ĠTang": 14057,
+      "Ġfavour": 14058,
+      "veolar": 14059,
+      "Ġpermission": 14060,
+      "Ġhabitats": 14061,
+      "Ġmaize": 14062,
+      "inct": 14063,
+      "Ġtelevision": 14064,
+      "rystals": 14065,
+      "ĠRadi": 14066,
+      "Ġflavon": 14067,
+      "Ġcann": 14068,
+      "iota": 14069,
+      "ĠOT": 14070,
+      "pic": 14071,
+      "Rad": 14072,
+      "titial": 14073,
+      "ĠOrth": 14074,
+      "stellar": 14075,
+      "ĠKine": 14076,
+      "Ġnavigation": 14077,
+      "fast": 14078,
+      "ĠCRISPR": 14079,
+      "Ġkinematic": 14080,
+      "Ġsearching": 14081,
+      "Ġmicrom": 14082,
+      "Ġinstalled": 14083,
+      "ĠTaiwan": 14084,
+      "ila": 14085,
+      "rf": 14086,
+      "riage": 14087,
+      "plinary": 14088,
+      "Ġecho": 14089,
+      "rav": 14090,
+      "ĠLes": 14091,
+      "create": 14092,
+      "Ġubiquit": 14093,
+      "Ġprecursors": 14094,
+      "KE": 14095,
+      "Ġdivide": 14096,
+      "Ġlnc": 14097,
+      "ĠConstruction": 14098,
+      "anic": 14099,
+      "estim": 14100,
+      "isters": 14101,
+      "Ġfeet": 14102,
+      "ariant": 14103,
+      "ĠSchw": 14104,
+      "Ġexclude": 14105,
+      "Ġvolcan": 14106,
+      "ĠOverview": 14107,
+      "Ġyr": 14108,
+      "olk": 14109,
+      "Ġ©": 14110,
+      "ĠFE": 14111,
+      "Ġspermat": 14112,
+      "Ġcapacitance": 14113,
+      "ĠSchrödinger": 14114,
+      "ĠGE": 14115,
+      "Ġcalibrated": 14116,
+      "SEM": 14117,
+      "Ġlattices": 14118,
+      "plier": 14119,
+      "Arg": 14120,
+      "ĠNT": 14121,
+      "ĠEnhanced": 14122,
+      "Ġbrom": 14123,
+      "Ġmultip": 14124,
+      "Ġcertified": 14125,
+      "Ġislands": 14126,
+      "Ġcyst": 14127,
+      "Ġaltitude": 14128,
+      "edef": 14129,
+      "Ġconstrain": 14130,
+      "Ġsatisfactory": 14131,
+      "Ġspecialized": 14132,
+      "Ġjunctions": 14133,
+      "Ġcoronavirus": 14134,
+      "udge": 14135,
+      "exc": 14136,
+      "Ġalt": 14137,
+      "ĠBacterial": 14138,
+      "Ġseasons": 14139,
+      "ĠLM": 14140,
+      "Ġhistogram": 14141,
+      "Ġsolvents": 14142,
+      "average": 14143,
+      "Ġcardinal": 14144,
+      "chrom": 14145,
+      "python": 14146,
+      "dered": 14147,
+      "enia": 14148,
+      "ĠGH": 14149,
+      "ĠEss": 14150,
+      "____": 14151,
+      "ĠPak": 14152,
+      "sized": 14153,
+      "ĠHg": 14154,
+      "Ġelif": 14155,
+      "ĠSchematic": 14156,
+      "Ġcytoplasm": 14157,
+      "ĠFort": 14158,
+      "ania": 14159,
+      "Ġcareful": 14160,
+      "ĠDual": 14161,
+      "Ġtranslated": 14162,
+      "Ġnasal": 14163,
+      "Inv": 14164,
+      "Ġdaughter": 14165,
+      "Ġemphasize": 14166,
+      "modules": 14167,
+      "Ġlives": 14168,
+      "Ġhomotopy": 14169,
+      "Ġbot": 14170,
+      "Ġdisordered": 14171,
+      "mato": 14172,
+      "Second": 14173,
+      "Ġclaimed": 14174,
+      "addle": 14175,
+      "Ġinterfacial": 14176,
+      "Ġviscous": 14177,
+      "Ġdestination": 14178,
+      "ĠPlanck": 14179,
+      "Ġabsorbance": 14180,
+      "Ġvolatile": 14181,
+      "Ġstorm": 14182,
+      "Ġcarboxyl": 14183,
+      "ĠBank": 14184,
+      "ĠPack": 14185,
+      "Ġscaffold": 14186,
+      "tebr": 14187,
+      "ipot": 14188,
+      "Ġtumours": 14189,
+      "ĠGol": 14190,
+      "Ġelectrophoresis": 14191,
+      "Ġrealize": 14192,
+      "Ġconstituents": 14193,
+      "Sol": 14194,
+      "ĠEvery": 14195,
+      "Ġmediate": 14196,
+      "Ġcoincide": 14197,
+      "Ġexploit": 14198,
+      "Ġmonoton": 14199,
+      "measure": 14200,
+      "Ġsupplied": 14201,
+      "racellular": 14202,
+      "Ġferro": 14203,
+      "Ġpurs": 14204,
+      "erentially": 14205,
+      "trast": 14206,
+      "ĠRB": 14207,
+      "Ġdissem": 14208,
+      "asy": 14209,
+      "Ġrelating": 14210,
+      "null": 14211,
+      "uates": 14212,
+      "constant": 14213,
+      "ĠContinuous": 14214,
+      "Ġgeometries": 14215,
+      "rust": 14216,
+      "ĠSTR": 14217,
+      "cluster": 14218,
+      "Ġprogenitor": 14219,
+      "ĠCSF": 14220,
+      "ĠYam": 14221,
+      "ĠReynolds": 14222,
+      "ĠMY": 14223,
+      "ĠKO": 14224,
+      "ĠWalk": 14225,
+      "ariable": 14226,
+      "inder": 14227,
+      "ĠRight": 14228,
+      "ĠAlgebra": 14229,
+      "ĠWik": 14230,
+      "Ġinactivation": 14231,
+      "tmp": 14232,
+      "access": 14233,
+      "ĠLater": 14234,
+      "Ġmicrobiome": 14235,
+      "Ġgeodesic": 14236,
+      "Ġrejection": 14237,
+      "uses": 14238,
+      "Ġhardness": 14239,
+      "Ġhydrodynamic": 14240,
+      "Ġvanish": 14241,
+      "Ġpollut": 14242,
+      "amycin": 14243,
+      "ĠÏŃ": 14244,
+      "ipitation": 14245,
+      "Ġaugmented": 14246,
+      "ĠTT": 14247,
+      "aval": 14248,
+      "Ġencode": 14249,
+      "Ġtoxin": 14250,
+      "eto": 14251,
+      "ighbor": 14252,
+      "addr": 14253,
+      "Ġdamaged": 14254,
+      "oi": 14255,
+      "Ġtransduction": 14256,
+      "Ġinteracts": 14257,
+      "ÃŃa": 14258,
+      "ĠCall": 14259,
+      "riends": 14260,
+      "ĠMonitoring": 14261,
+      "ĠVariation": 14262,
+      "Ġôı¼": 14263,
+      "Ġdich": 14264,
+      "Ġspars": 14265,
+      "align": 14266,
+      "Ġanatomical": 14267,
+      "Ġcentrifuged": 14268,
+      "urally": 14269,
+      "ĠZr": 14270,
+      "ĠCarl": 14271,
+      "Recall": 14272,
+      "Ġopinion": 14273,
+      "Ġera": 14274,
+      "Ġdrainage": 14275,
+      "Ġmicroarray": 14276,
+      "status": 14277,
+      "umental": 14278,
+      "Ġcomprises": 14279,
+      "pressure": 14280,
+      "Ġpractition": 14281,
+      "mac": 14282,
+      "Ġcongr": 14283,
+      "urnal": 14284,
+      "ĠAPI": 14285,
+      "ĠLR": 14286,
+      "Ġtransfection": 14287,
+      "Ġslopes": 14288,
+      "ĠCode": 14289,
+      "Ġphil": 14290,
+      "bool": 14291,
+      "Ws": 14292,
+      "ĠâĻ": 14293,
+      "Ġassociate": 14294,
+      "otoxicity": 14295,
+      "rade": 14296,
+      "ĠMiller": 14297,
+      "ĠϪ": 14298,
+      "Ġshorten": 14299,
+      "Ġadditionally": 14300,
+      "ĠEffective": 14301,
+      "Ġsupervised": 14302,
+      "Ġelabor": 14303,
+      "ĠCellular": 14304,
+      "Ġtell": 14305,
+      "ĠRC": 14306,
+      "save": 14307,
+      "imid": 14308,
+      "Ġratings": 14309,
+      "ĠTaking": 14310,
+      "Ġapproval": 14311,
+      "Ġpenalty": 14312,
+      "KK": 14313,
+      "context": 14314,
+      "aks": 14315,
+      "pecific": 14316,
+      "Ġtempor": 14317,
+      "Ġupregulation": 14318,
+      "VAL": 14319,
+      "Ġencodes": 14320,
+      "inin": 14321,
+      "Ġnotes": 14322,
+      "ĠForest": 14323,
+      "Ġcombinatorial": 14324,
+      "ymptotic": 14325,
+      "Ġsquamous": 14326,
+      "ĠAsh": 14327,
+      "ourn": 14328,
+      "Ġmyeloid": 14329,
+      "elines": 14330,
+      "Bio": 14331,
+      "Ġbreed": 14332,
+      "ĠRub": 14333,
+      "uzz": 14334,
+      "Ġsinglet": 14335,
+      "enna": 14336,
+      "Ġcritically": 14337,
+      "dig": 14338,
+      "disci": 14339,
+      "Ġdropped": 14340,
+      "Ġlipoprotein": 14341,
+      "ĠEt": 14342,
+      "Ġnov": 14343,
+      "ophen": 14344,
+      "Ġancient": 14345,
+      "Base": 14346,
+      "Ġsmoothing": 14347,
+      "itives": 14348,
+      "pine": 14349,
+      "Ġsolver": 14350,
+      "perm": 14351,
+      "ĠHome": 14352,
+      "Ġazim": 14353,
+      "lVert": 14354,
+      "Ġtransportation": 14355,
+      "Ġdex": 14356,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 14357,
+      "opathic": 14358,
+      "experim": 14359,
+      "âĢ¢âĢ¢": 14360,
+      "perfusion": 14361,
+      "Ġdoi": 14362,
+      "ĠLact": 14363,
+      "Ġhepatocellular": 14364,
+      "Ġmismatch": 14365,
+      "Ġadenocarcinoma": 14366,
+      "ĠPain": 14367,
+      "Ġspr": 14368,
+      "Ġconfinement": 14369,
+      "Ġexceeds": 14370,
+      "Ġhash": 14371,
+      "ĠComparing": 14372,
+      "ĠSensor": 14373,
+      "Ġfiring": 14374,
+      "kes": 14375,
+      "vir": 14376,
+      "inea": 14377,
+      "affected": 14378,
+      "Ġmodelled": 14379,
+      "Ġether": 14380,
+      "Ġsuffer": 14381,
+      "âĢ²âĢ²": 14382,
+      "оÐ": 14383,
+      "ĠBir": 14384,
+      "Äģ": 14385,
+      "Ġsecreted": 14386,
+      "Ġcatheter": 14387,
+      "Ġyouth": 14388,
+      "expl": 14389,
+      "ĠDar": 14390,
+      "ĠWHO": 14391,
+      "Ġfoundation": 14392,
+      "Ġhydraulic": 14393,
+      "ĠCarol": 14394,
+      "SSION": 14395,
+      "Ġá¹": 14396,
+      "feld": 14397,
+      "avor": 14398,
+      "Ġpasses": 14399,
+      "visiae": 14400,
+      "Ġapplicability": 14401,
+      "Ġnested": 14402,
+      "Fl": 14403,
+      "ĠCatal": 14404,
+      "Ġmicroenvironment": 14405,
+      "labels": 14406,
+      "Ġcrystallization": 14407,
+      "Info": 14408,
+      "Ġpositioning": 14409,
+      "Ġtriangles": 14410,
+      "Ġtryp": 14411,
+      "ĠTransition": 14412,
+      "Ġsett": 14413,
+      "Ġneurot": 14414,
+      "Mon": 14415,
+      "Ġdroplets": 14416,
+      "ĠART": 14417,
+      "Ġcorne": 14418,
+      "Ġmultiplicity": 14419,
+      "Ġeccentric": 14420,
+      "Ġiv": 14421,
+      "ĠMatter": 14422,
+      "learning": 14423,
+      "electro": 14424,
+      "ĠWeyl": 14425,
+      "Ġdecide": 14426,
+      "ĠWr": 14427,
+      "ĠHierarch": 14428,
+      "Ġapical": 14429,
+      "Ġfailures": 14430,
+      "Ġdigestion": 14431,
+      "MIC": 14432,
+      "Ġgeographical": 14433,
+      "ĠElement": 14434,
+      "ĠThough": 14435,
+      "Ġchron": 14436,
+      "limited": 14437,
+      "ĠDISC": 14438,
+      "ĠArchitecture": 14439,
+      "Ġvibrational": 14440,
+      "ĠVarious": 14441,
+      "Ġdynamically": 14442,
+      "aked": 14443,
+      "Ġconvenience": 14444,
+      "ĠIsra": 14445,
+      "ĠMDA": 14446,
+      "itic": 14447,
+      "Au": 14448,
+      "Ġassistance": 14449,
+      "ventional": 14450,
+      "midt": 14451,
+      "ospor": 14452,
+      "Following": 14453,
+      "Ġinferior": 14454,
+      "Ġnickel": 14455,
+      "raine": 14456,
+      "paren": 14457,
+      "Ġtitanium": 14458,
+      "Field": 14459,
+      "Ġhoc": 14460,
+      "ĠCauchy": 14461,
+      "ĠMcC": 14462,
+      "ĠScreen": 14463,
+      "Ġneglect": 14464,
+      "classes": 14465,
+      "ĠIF": 14466,
+      "Ġstratified": 14467,
+      "enses": 14468,
+      "ĠPlate": 14469,
+      "ozoic": 14470,
+      "Ġinstitutions": 14471,
+      "ĠThose": 14472,
+      "Ġgenerations": 14473,
+      "transform": 14474,
+      "Ġpartitions": 14475,
+      "Rxiv": 14476,
+      "enth": 14477,
+      "Ġstic": 14478,
+      "olith": 14479,
+      "ĠFem": 14480,
+      "Ġagar": 14481,
+      "beam": 14482,
+      "Ġprotons": 14483,
+      "LU": 14484,
+      "Ġworkload": 14485,
+      "Ġminerals": 14486,
+      "Ġmt": 14487,
+      "lla": 14488,
+      "ĠPharmac": 14489,
+      "Ġconverter": 14490,
+      "ĠMechanical": 14491,
+      "Ġflavor": 14492,
+      "Ġphosphatase": 14493,
+      "Ġsums": 14494,
+      "PCs": 14495,
+      "Ġisoforms": 14496,
+      "igroup": 14497,
+      "pyr": 14498,
+      "features": 14499,
+      "Ġperc": 14500,
+      "Ġcompleteness": 14501,
+      "Ġforests": 14502,
+      "Ġdividing": 14503,
+      "ĠLipschitz": 14504,
+      "periodic": 14505,
+      "Ġrecycl": 14506,
+      "ĠNag": 14507,
+      "Ġtwin": 14508,
+      "eptides": 14509,
+      "Ġcohor": 14510,
+      "Ġsearches": 14511,
+      "eated": 14512,
+      "Hg": 14513,
+      "ĠPU": 14514,
+      "ĠTree": 14515,
+      "allic": 14516,
+      "PF": 14517,
+      "Ġappendix": 14518,
+      "ĠCov": 14519,
+      "Ġchecking": 14520,
+      "Ġbackbone": 14521,
+      "Thermo": 14522,
+      "Ġactivating": 14523,
+      "ĠVictor": 14524,
+      "Ġcritic": 14525,
+      "ĠLem": 14526,
+      "groups": 14527,
+      "REG": 14528,
+      "ĠOcc": 14529,
+      "SCC": 14530,
+      "ĠXRD": 14531,
+      "ĠValues": 14532,
+      "Ġsubtype": 14533,
+      "Ġstretching": 14534,
+      "ORM": 14535,
+      "some": 14536,
+      "Ġflip": 14537,
+      "Ġphenolic": 14538,
+      "Ġkilled": 14539,
+      "Ġsequenced": 14540,
+      "uscular": 14541,
+      "abin": 14542,
+      "Ġquadr": 14543,
+      "Ġtranslational": 14544,
+      "Ġsolids": 14545,
+      "direct": 14546,
+      "Ġpromotion": 14547,
+      "Ġcohorts": 14548,
+      "ĠClimate": 14549,
+      "ĠOld": 14550,
+      "ĠSir": 14551,
+      "gue": 14552,
+      "strate": 14553,
+      "ĠPoss": 14554,
+      "Ġreceives": 14555,
+      "ĠValidation": 14556,
+      "uctive": 14557,
+      "Ġcerevisiae": 14558,
+      "Gu": 14559,
+      "isis": 14560,
+      "ceil": 14561,
+      "ĠPearson": 14562,
+      "ĠPrelim": 14563,
+      "ĠGran": 14564,
+      "CSF": 14565,
+      "Ġsterile": 14566,
+      "ofluorescence": 14567,
+      "bad": 14568,
+      "Ġcolored": 14569,
+      "compass": 14570,
+      "equation": 14571,
+      "jan": 14572,
+      "Ġconditioning": 14573,
+      "Ġvoice": 14574,
+      "Ġmening": 14575,
+      "Ġgranted": 14576,
+      "Ġrenormalization": 14577,
+      "ĠLimit": 14578,
+      "thi": 14579,
+      "Ġaperture": 14580,
+      "Ġdosage": 14581,
+      "directed": 14582,
+      "ĠBreast": 14583,
+      "ocular": 14584,
+      "bearing": 14585,
+      "sal": 14586,
+      "ascul": 14587,
+      "upervised": 14588,
+      "Ġmonolayer": 14589,
+      "Ġmembership": 14590,
+      "ĠWireless": 14591,
+      "show": 14592,
+      "ĠMedia": 14593,
+      "ĠVL": 14594,
+      "essel": 14595,
+      "Ġdecoder": 14596,
+      "ĠMF": 14597,
+      "ĠComposition": 14598,
+      "ĠClark": 14599,
+      "Point": 14600,
+      "ĠNano": 14601,
+      "ĠDeg": 14602,
+      "NL": 14603,
+      "ĠBox": 14604,
+      "Ġexploring": 14605,
+      "molecular": 14606,
+      "Other": 14607,
+      "ĠDiabetes": 14608,
+      "height": 14609,
+      "Ġkinases": 14610,
+      "Ġadjusting": 14611,
+      "Ġsports": 14612,
+      "offs": 14613,
+      "ĠIEEE": 14614,
+      "Ġtil": 14615,
+      "ĠIntra": 14616,
+      "Ġplanets": 14617,
+      "ĠEpidem": 14618,
+      "Ġtomato": 14619,
+      "Ġscaffolds": 14620,
+      "ĠMetabol": 14621,
+      "ĠGeometry": 14622,
+      "imetry": 14623,
+      "ĠTen": 14624,
+      "thread": 14625,
+      "ohex": 14626,
+      "Ġproposes": 14627,
+      "prim": 14628,
+      "ĠParty": 14629,
+      "Ġquarter": 14630,
+      "ĠShi": 14631,
+      "Ġaberr": 14632,
+      "ĠIntr": 14633,
+      "Ġdirector": 14634,
+      "affe": 14635,
+      "ĠSus": 14636,
+      "ensors": 14637,
+      "Ele": 14638,
+      "Ġpoles": 14639,
+      "Additional": 14640,
+      "Ġbypass": 14641,
+      "catenin": 14642,
+      "Ġundertaken": 14643,
+      "imation": 14644,
+      "opor": 14645,
+      "Ġpreserving": 14646,
+      "Ġmultiplex": 14647,
+      "ĠRepresentative": 14648,
+      "sis": 14649,
+      "ĠAG": 14650,
+      "achy": 14651,
+      "Ġfruits": 14652,
+      "Ġreconstruct": 14653,
+      "ensen": 14654,
+      "Ġstrongest": 14655,
+      "Ġscav": 14656,
+      "ĠCheng": 14657,
+      "ĠCoron": 14658,
+      "ĠObservation": 14659,
+      "ĠAch": 14660,
+      "ĠGeorg": 14661,
+      "ĠSVM": 14662,
+      "ĠChern": 14663,
+      "Ġreversal": 14664,
+      "via": 14665,
+      "imp": 14666,
+      "Ġdeployment": 14667,
+      "ĠHad": 14668,
+      "Ġcircumstances": 14669,
+      "obi": 14670,
+      "Ġcurved": 14671,
+      "Induced": 14672,
+      "ĠPositive": 14673,
+      "imb": 14674,
+      "ĠParis": 14675,
+      "ĠStein": 14676,
+      "icz": 14677,
+      "ĠCath": 14678,
+      "Ġdrawing": 14679,
+      "tory": 14680,
+      "Ġcontinental": 14681,
+      "Ġquantitatively": 14682,
+      "acerb": 14683,
+      "Ġnorms": 14684,
+      "ĠBE": 14685,
+      "Several": 14686,
+      "door": 14687,
+      "Ġplateau": 14688,
+      "Gal": 14689,
+      "Ġcivil": 14690,
+      "ĠFix": 14691,
+      "LAB": 14692,
+      "occal": 14693,
+      "Ġsorted": 14694,
+      "ĠâĢĿ": 14695,
+      "Ġediting": 14696,
+      "ĠChristian": 14697,
+      "Ġclarify": 14698,
+      "Ġwaveguide": 14699,
+      "bell": 14700,
+      "Ġdeduced": 14701,
+      "odec": 14702,
+      "utrition": 14703,
+      "Ġcompressive": 14704,
+      "ĠEU": 14705,
+      "ĠRegression": 14706,
+      "Ġranked": 14707,
+      "Ġestimators": 14708,
+      "Ġabilities": 14709,
+      "Ġbeliefs": 14710,
+      "three": 14711,
+      "ĠâĩĴ": 14712,
+      "rology": 14713,
+      "Ġautonomous": 14714,
+      "ĠSz": 14715,
+      "schem": 14716,
+      "ĠALT": 14717,
+      "ĠPatterns": 14718,
+      "Ġexon": 14719,
+      "Ġlifestyle": 14720,
+      "fill": 14721,
+      "ĠCAR": 14722,
+      "ĠDomains": 14723,
+      "Ġpaid": 14724,
+      "Ġtab": 14725,
+      "ĠCohen": 14726,
+      "airy": 14727,
+      "Ġsheep": 14728,
+      "Ġseaw": 14729,
+      "ĠKong": 14730,
+      "gas": 14731,
+      "Ġreserved": 14732,
+      "Ġresil": 14733,
+      "Ġobl": 14734,
+      "carbox": 14735,
+      "ĠGovernment": 14736,
+      "upper": 14737,
+      "racting": 14738,
+      "Ġgangl": 14739,
+      "ĠRV": 14740,
+      "Ġbronch": 14741,
+      "Methods": 14742,
+      "ĠLiver": 14743,
+      "Ġguess": 14744,
+      "charomy": 14745,
+      "ICE": 14746,
+      "Ġcongenital": 14747,
+      "Ġka": 14748,
+      "Ġspanning": 14749,
+      "ĠRecomm": 14750,
+      "ea": 14751,
+      "Ġconvention": 14752,
+      "Ġsheets": 14753,
+      "Ġthermo": 14754,
+      "Ġqualitatively": 14755,
+      "Ġoxides": 14756,
+      "Ġcongru": 14757,
+      "ĠJer": 14758,
+      "Ġpreservation": 14759,
+      "ĠBT": 14760,
+      "ĠDMSO": 14761,
+      "Ġcomplication": 14762,
+      "Ġsurvivors": 14763,
+      "Ġreduct": 14764,
+      "Ġdescent": 14765,
+      "Ġsucrose": 14766,
+      "ĠCourt": 14767,
+      "Ġmetabolite": 14768,
+      "ĠMath": 14769,
+      "ĠSecurity": 14770,
+      "ĠNotably": 14771,
+      "ĠStem": 14772,
+      "Ġdwarf": 14773,
+      "bc": 14774,
+      "Ġrevis": 14775,
+      "ĠKl": 14776,
+      "ĠGh": 14777,
+      "Ġmanager": 14778,
+      "Ġinvestment": 14779,
+      "Ġmotility": 14780,
+      "Em": 14781,
+      "ĠMr": 14782,
+      "asic": 14783,
+      "ĠBos": 14784,
+      "Ġinspired": 14785,
+      "placian": 14786,
+      "Ġease": 14787,
+      "Ġtorsion": 14788,
+      "ĠDirichlet": 14789,
+      "Ġspleen": 14790,
+      "agation": 14791,
+      "onate": 14792,
+      "ĠTrial": 14793,
+      "Ġturnover": 14794,
+      "Ġselectively": 14795,
+      "ĠÍĴ": 14796,
+      "iano": 14797,
+      "Ġnontrivial": 14798,
+      "iasis": 14799,
+      "Ñģ": 14800,
+      "ĠGuo": 14801,
+      "Ġaddresses": 14802,
+      "Ġuniqueness": 14803,
+      "Ġwithdraw": 14804,
+      "riz": 14805,
+      "Ġcomputationally": 14806,
+      "Ġpersonality": 14807,
+      "AX": 14808,
+      "wenty": 14809,
+      "Ġgovern": 14810,
+      "berts": 14811,
+      "Ġrobots": 14812,
+      "Ġready": 14813,
+      "Ġdiets": 14814,
+      "lit": 14815,
+      "My": 14816,
+      "ĠReve": 14817,
+      "ĠLos": 14818,
+      "infrared": 14819,
+      "Ġintram": 14820,
+      "lated": 14821,
+      "plankton": 14822,
+      "ĠGrant": 14823,
+      "piper": 14824,
+      "Ġantennas": 14825,
+      "Ġbol": 14826,
+      "fp": 14827,
+      "ĠVit": 14828,
+      "Compar": 14829,
+      "oken": 14830,
+      "Ġkeys": 14831,
+      "ĠClub": 14832,
+      "inery": 14833,
+      "ĠFoot": 14834,
+      "Ġwarming": 14835,
+      "mond": 14836,
+      "Ġmiles": 14837,
+      "Ġspeaking": 14838,
+      "ĠIv": 14839,
+      "Ġconformational": 14840,
+      "ĠOk": 14841,
+      "Ġunified": 14842,
+      "Ġassembled": 14843,
+      "Ġinverted": 14844,
+      "Ġfelt": 14845,
+      "corresponding": 14846,
+      "ĠECM": 14847,
+      "ĠNSC": 14848,
+      "Ġindoor": 14849,
+      "gov": 14850,
+      "Ġantagonist": 14851,
+      "unched": 14852,
+      "ĠJava": 14853,
+      "ĠCombined": 14854,
+      "tivities": 14855,
+      "Ġalternating": 14856,
+      "ãĤ": 14857,
+      "ĠDiagnosis": 14858,
+      "Ġdistinction": 14859,
+      "leigh": 14860,
+      "ĠTogether": 14861,
+      "Ġparticipating": 14862,
+      "Ġglomer": 14863,
+      "oche": 14864,
+      "Ġcopyright": 14865,
+      "ĠGTP": 14866,
+      "ĠVar": 14867,
+      "Ġammonium": 14868,
+      "Ġfacilitates": 14869,
+      "Ġperfusion": 14870,
+      "ĠLB": 14871,
+      "full": 14872,
+      "Ġreti": 14873,
+      "iferase": 14874,
+      "Ġimmunosup": 14875,
+      "ĠImplementation": 14876,
+      "Ġpores": 14877,
+      "ĠBB": 14878,
+      "ĠBud": 14879,
+      "ĠVO": 14880,
+      "ĠVo": 14881,
+      "Ġphysician": 14882,
+      "ĠAUC": 14883,
+      "Ġcertainly": 14884,
+      "μm": 14885,
+      "ĠKol": 14886,
+      "Ġwrap": 14887,
+      "middle": 14888,
+      "Ġsilencing": 14889,
+      "Ġfreshwater": 14890,
+      "igan": 14891,
+      "area": 14892,
+      "AI": 14893,
+      "Ġmicrotub": 14894,
+      "Ġarranged": 14895,
+      "structive": 14896,
+      "ĠRegular": 14897,
+      "ĠFile": 14898,
+      "alks": 14899,
+      "Ġplain": 14900,
+      "Ġintegrable": 14901,
+      "ĠMembrane": 14902,
+      "istors": 14903,
+      "Ġaquatic": 14904,
+      "Ġworkflow": 14905,
+      "ĠGer": 14906,
+      "ulant": 14907,
+      "Ġactivates": 14908,
+      "Term": 14909,
+      "ĠUpon": 14910,
+      "ĠPut": 14911,
+      "Var": 14912,
+      "ĠOD": 14913,
+      "half": 14914,
+      "Ġulcer": 14915,
+      "ĠBO": 14916,
+      "ĠGy": 14917,
+      "rences": 14918,
+      "Ġpurity": 14919,
+      "Ġarrive": 14920,
+      "ĠSignificant": 14921,
+      "ĠMAC": 14922,
+      "ĠOtherwise": 14923,
+      "oured": 14924,
+      "Ġtan": 14925,
+      "ĠRL": 14926,
+      "ĠQTL": 14927,
+      "Ġammonia": 14928,
+      "vmode": 14929,
+      "Ġmagnesium": 14930,
+      "Ġacknowled": 14931,
+      "Ġalternatives": 14932,
+      "idents": 14933,
+      "rVert": 14934,
+      "ĠComplete": 14935,
+      "ĠBone": 14936,
+      "yer": 14937,
+      "ĠBab": 14938,
+      "Ġeut": 14939,
+      "Ġnovo": 14940,
+      "disciplinary": 14941,
+      "Ġseverely": 14942,
+      "uki": 14943,
+      "ĠPN": 14944,
+      "leavevmode": 14945,
+      "clip": 14946,
+      "ĠSynd": 14947,
+      "ĠMIMO": 14948,
+      "adequ": 14949,
+      "ĠArctic": 14950,
+      "lycer": 14951,
+      "RET": 14952,
+      "ensed": 14953,
+      "coated": 14954,
+      "VP": 14955,
+      "Ġlakes": 14956,
+      "Ġchurch": 14957,
+      "Ġhomologous": 14958,
+      "Ġoxidase": 14959,
+      "ĠAud": 14960,
+      "Ġincrement": 14961,
+      "Ġneutrinos": 14962,
+      "arbon": 14963,
+      "TYPE": 14964,
+      "izumab": 14965,
+      "utable": 14966,
+      "Ġimplying": 14967,
+      "ĠMotion": 14968,
+      "Ġâīĥ": 14969,
+      "Ġpages": 14970,
+      "Ġplausible": 14971,
+      "ĠNL": 14972,
+      "Ġisotop": 14973,
+      "ĠHyd": 14974,
+      "Att": 14975,
+      "lattice": 14976,
+      "shore": 14977,
+      "Ġsucceed": 14978,
+      "Ġsupposed": 14979,
+      "ĠTransmission": 14980,
+      "Dimensional": 14981,
+      "inguistic": 14982,
+      "Ġcontours": 14983,
+      "Ġconcomit": 14984,
+      "Ġagrees": 14985,
+      "ĠDani": 14986,
+      "quar": 14987,
+      "Ġshield": 14988,
+      "Ġozone": 14989,
+      "ĠTet": 14990,
+      "lbrack": 14991,
+      "Ġwat": 14992,
+      "Ġcytochrome": 14993,
+      "tailed": 14994,
+      "pix": 14995,
+      "Ġcoex": 14996,
+      "ĠView": 14997,
+      "odef": 14998,
+      "ĠWild": 14999,
+      "ĠLE": 15000,
+      "hop": 15001,
+      "Ġpointing": 15002,
+      "uncture": 15003,
+      "Ġecology": 15004,
+      "Ġbab": 15005,
+      "rea": 15006,
+      "ego": 15007,
+      "Ġviolence": 15008,
+      "ĠtRNA": 15009,
+      "ĠRN": 15010,
+      "pent": 15011,
+      "orel": 15012,
+      "ĠParallel": 15013,
+      "Ġdrives": 15014,
+      "nobreak": 15015,
+      "Ġholog": 15016,
+      "Ġprobable": 15017,
+      "Ġentering": 15018,
+      "Ġsink": 15019,
+      "Ġswelling": 15020,
+      "producing": 15021,
+      "âĨĴâĪŀ": 15022,
+      "ĠSafety": 15023,
+      "Ġanalyse": 15024,
+      "series": 15025,
+      "Ġdrivers": 15026,
+      "KS": 15027,
+      "ĠRMS": 15028,
+      "Ġgenetics": 15029,
+      "ĠFred": 15030,
+      "Ġsubm": 15031,
+      "Ġscientists": 15032,
+      "ĠFD": 15033,
+      "ĠSolutions": 15034,
+      "ĠFab": 15035,
+      "Ġencompass": 15036,
+      "commutative": 15037,
+      "Ġadiabatic": 15038,
+      "butyl": 15039,
+      "PEG": 15040,
+      "Ġαβ": 15041,
+      "ĠStan": 15042,
+      "Ġclustered": 15043,
+      "Ġholding": 15044,
+      "ĠBeck": 15045,
+      "ĠYan": 15046,
+      "Ġaster": 15047,
+      "Ġeconom": 15048,
+      "Ġignored": 15049,
+      "uro": 15050,
+      "yles": 15051,
+      "ubbles": 15052,
+      "Ġfate": 15053,
+      "Ġperceptions": 15054,
+      "Ġlin": 15055,
+      "én": 15056,
+      "Ġactu": 15057,
+      "Ġarsen": 15058,
+      "Ġba": 15059,
+      "epoch": 15060,
+      "ĠStim": 15061,
+      "Ġmedications": 15062,
+      "ECs": 15063,
+      "ĠMinistry": 15064,
+      "ĠPublisher": 15065,
+      "Ġdepri": 15066,
+      "Ġobstruction": 15067,
+      "ĠmRNAs": 15068,
+      "Ġbrother": 15069,
+      "Ġcrossover": 15070,
+      "ĠTurb": 15071,
+      "tation": 15072,
+      "Ġtank": 15073,
+      "ĠMem": 15074,
+      "Ġintestine": 15075,
+      "Ġmicroglia": 15076,
+      "ĠMaxwell": 15077,
+      "Ġjurisdic": 15078,
+      "Ġphenyl": 15079,
+      "hyper": 15080,
+      "ums": 15081,
+      "ĠHIF": 15082,
+      "ĠShen": 15083,
+      "Ġcheckpoint": 15084,
+      "ĠBrownian": 15085,
+      "ĠâĭĨ": 15086,
+      "ĠStrain": 15087,
+      "ĠExtraction": 15088,
+      "Ġbatteries": 15089,
+      "ĠPle": 15090,
+      "ĠConditions": 15091,
+      "Ġinconsistent": 15092,
+      "ĠHost": 15093,
+      "ypical": 15094,
+      "Ġcrops": 15095,
+      "alg": 15096,
+      "ĠFI": 15097,
+      "anta": 15098,
+      "Ġfounded": 15099,
+      "Ġmarks": 15100,
+      "distribution": 15101,
+      "Ġι": 15102,
+      "Ġhors": 15103,
+      "Ġsnap": 15104,
+      "WM": 15105,
+      "Ġmanifestations": 15106,
+      "empl": 15107,
+      "Ġproving": 15108,
+      "leading": 15109,
+      "ĠACE": 15110,
+      "ĠLED": 15111,
+      "channels": 15112,
+      "Ġlift": 15113,
+      "Function": 15114,
+      "inase": 15115,
+      "supervised": 15116,
+      "ĠUser": 15117,
+      "Ġphysiology": 15118,
+      "Ġlinking": 15119,
+      "pressed": 15120,
+      "Ġiff": 15121,
+      "ĠJim": 15122,
+      "Ġglutathione": 15123,
+      "ĠTI": 15124,
+      "Ġane": 15125,
+      "enosis": 15126,
+      "Ġcollections": 15127,
+      "Ġgenetically": 15128,
+      "ĠFilter": 15129,
+      "ĠChicago": 15130,
+      "ĠServices": 15131,
+      "Ġsupersymmetric": 15132,
+      "Ġstriking": 15133,
+      "Ġirrig": 15134,
+      "ococcal": 15135,
+      "Ġfibres": 15136,
+      "Ġecosystems": 15137,
+      "uming": 15138,
+      "fly": 15139,
+      "Ġlungs": 15140,
+      "Ġcovariates": 15141,
+      "Ġlayout": 15142,
+      "ĠRaj": 15143,
+      "Ġsummation": 15144,
+      "abled": 15145,
+      "Ġfreely": 15146,
+      "Ġrevised": 15147,
+      "Ġcuts": 15148,
+      "ĠIntegrated": 15149,
+      "Ġpharmaceutical": 15150,
+      "Ġrespiration": 15151,
+      "ĠBill": 15152,
+      "Ġestrogen": 15153,
+      "raint": 15154,
+      "Ġpercentages": 15155,
+      "ĠPf": 15156,
+      "ĠGF": 15157,
+      "methylene": 15158,
+      "Ġorigins": 15159,
+      "trim": 15160,
+      "match": 15161,
+      "itney": 15162,
+      "ĠYe": 15163,
+      "Ġallocated": 15164,
+      "manifold": 15165,
+      "ĠTris": 15166,
+      "ĠLys": 15167,
+      "Ġcompressed": 15168,
+      "orer": 15169,
+      "Ġhimself": 15170,
+      "Ġquin": 15171,
+      "ĠAssembly": 15172,
+      "single": 15173,
+      "temporal": 15174,
+      "Ġsoph": 15175,
+      "Ġepidemiological": 15176,
+      "Ġknockout": 15177,
+      "Ġcompares": 15178,
+      "ĠSensitivity": 15179,
+      "Ġgirls": 15180,
+      "ĠValley": 15181,
+      "alid": 15182,
+      "ĠScheme": 15183,
+      "ĠCOMP": 15184,
+      "Ġrefractive": 15185,
+      "ĠOffice": 15186,
+      "Ġlatest": 15187,
+      "Ġprices": 15188,
+      "carboxyl": 15189,
+      "Ġeconomy": 15190,
+      "Ġbooks": 15191,
+      "ĠDD": 15192,
+      "Ġneoplas": 15193,
+      "appings": 15194,
+      "Ġfolding": 15195,
+      "momentum": 15196,
+      "potent": 15197,
+      "Ġprefix": 15198,
+      "ĠRiemannian": 15199,
+      "ĠERK": 15200,
+      "ĠPathway": 15201,
+      "Ġlarval": 15202,
+      "olor": 15203,
+      "Ġattitude": 15204,
+      "geqslant": 15205,
+      "Ġgates": 15206,
+      "Ġagonist": 15207,
+      "ĠïĢ¨": 15208,
+      "ĠMCF": 15209,
+      "ostatic": 15210,
+      "micro": 15211,
+      "Ġdoubl": 15212,
+      "ĠParameter": 15213,
+      "Ġequivalently": 15214,
+      "Ġsrc": 15215,
+      "Most": 15216,
+      "ĉĠĠĠ": 15217,
+      "Ġrheumat": 15218,
+      "ĠHum": 15219,
+      "region": 15220,
+      "Ġwinds": 15221,
+      "Ġquadrup": 15222,
+      "cales": 15223,
+      "ulfide": 15224,
+      "balanced": 15225,
+      "Under": 15226,
+      "generated": 15227,
+      "oplasmic": 15228,
+      "Ġweighting": 15229,
+      "ĠNov": 15230,
+      "veloc": 15231,
+      "utils": 15232,
+      "ĠACT": 15233,
+      "Ġvulnerable": 15234,
+      "dc": 15235,
+      "Ġstromal": 15236,
+      "Ġexacerb": 15237,
+      "HV": 15238,
+      "Ġperfectly": 15239,
+      "txt": 15240,
+      "direction": 15241,
+      "ogon": 15242,
+      "Ġbim": 15243,
+      "ĠMarg": 15244,
+      "itons": 15245,
+      "Ġtermination": 15246,
+      "eda": 15247,
+      "Ġpretreatment": 15248,
+      "Ġimportantly": 15249,
+      "Ġduc": 15250,
+      "Ġartifacts": 15251,
+      "Stud": 15252,
+      "otensin": 15253,
+      "reland": 15254,
+      "ahn": 15255,
+      "Ġdeployed": 15256,
+      "ĠEF": 15257,
+      "ensing": 15258,
+      "ĠCard": 15259,
+      "ĠJordan": 15260,
+      "apunov": 15261,
+      "Ġanesthesia": 15262,
+      "Ġatherosclerosis": 15263,
+      "inner": 15264,
+      "structural": 15265,
+      "ĠAsp": 15266,
+      "throughput": 15267,
+      "urities": 15268,
+      "Ġinset": 15269,
+      "without": 15270,
+      "Ġacquire": 15271,
+      "Ġcombines": 15272,
+      "ĠShar": 15273,
+      "MASK": 15274,
+      "ĠLiter": 15275,
+      "Ġconscious": 15276,
+      "iscell": 15277,
+      "consistent": 15278,
+      "yst": 15279,
+      "Ġfilaments": 15280,
+      "ĠAlice": 15281,
+      "ĠGround": 15282,
+      "ĠmTOR": 15283,
+      "versal": 15284,
+      "Ġlineages": 15285,
+      "particles": 15286,
+      "aroscopic": 15287,
+      "ĠProced": 15288,
+      "Ġorientations": 15289,
+      "ĠMouse": 15290,
+      "Ġaccordingly": 15291,
+      "Ġsuppressor": 15292,
+      "Ġdestruction": 15293,
+      "OV": 15294,
+      "ĠProteins": 15295,
+      "PECT": 15296,
+      "Ġcup": 15297,
+      "Ġmonomer": 15298,
+      "plemental": 15299,
+      "Ġneutrophil": 15300,
+      "Ġerup": 15301,
+      "Ġtac": 15302,
+      "Ġasymptomatic": 15303,
+      "ĠEmbed": 15304,
+      "ĠRadiation": 15305,
+      "ĠGame": 15306,
+      "Ġneedle": 15307,
+      "Ġreuse": 15308,
+      "ĠDutch": 15309,
+      "Ġjuvenile": 15310,
+      "Ġmomenta": 15311,
+      "ĠBose": 15312,
+      "Ġdeveloper": 15313,
+      "Ġresiduals": 15314,
+      "Å¡": 15315,
+      "Ġcognition": 15316,
+      "ĠRegional": 15317,
+      "You": 15318,
+      "ĠConcent": 15319,
+      "ocin": 15320,
+      "ĠPartial": 15321,
+      "Ġcompletes": 15322,
+      "ĠSingh": 15323,
+      "ĠExc": 15324,
+      "ĠIsolation": 15325,
+      "ĠStructures": 15326,
+      "Ġintermitt": 15327,
+      "Exception": 15328,
+      "Ġanalytically": 15329,
+      "Ġelectricity": 15330,
+      "âĭ": 15331,
+      "Äį": 15332,
+      "Ġproteome": 15333,
+      "Ġic": 15334,
+      "kal": 15335,
+      "inux": 15336,
+      "ĠBeyond": 15337,
+      "Ġimplied": 15338,
+      "ASH": 15339,
+      "Ġclone": 15340,
+      "ĠRussia": 15341,
+      "ĠHod": 15342,
+      "tebrates": 15343,
+      "Ġproxy": 15344,
+      "holder": 15345,
+      "elve": 15346,
+      "Ġvalley": 15347,
+      "utely": 15348,
+      "Ġjobs": 15349,
+      "ruption": 15350,
+      "roids": 15351,
+      "ĠWhy": 15352,
+      "eping": 15353,
+      "ĠYet": 15354,
+      "Ġpyl": 15355,
+      "Ġbra": 15356,
+      "ilization": 15357,
+      "eters": 15358,
+      "Ġadver": 15359,
+      "Ġove": 15360,
+      "kernel": 15361,
+      "samples": 15362,
+      "ordinate": 15363,
+      "ĠAssuming": 15364,
+      "Ġcontaminated": 15365,
+      "Ġbipolar": 15366,
+      "Ġlac": 15367,
+      "Ġluc": 15368,
+      "Ġcentrifugation": 15369,
+      "Both": 15370,
+      "Ġnd": 15371,
+      "Ġtib": 15372,
+      "Before": 15373,
+      "ĠImmune": 15374,
+      "Ġash": 15375,
+      "Ġconditioned": 15376,
+      "ĠRank": 15377,
+      "NOS": 15378,
+      "Ġnanoparticle": 15379,
+      "Ġdependencies": 15380,
+      "Ġhouseholds": 15381,
+      "agers": 15382,
+      "Ġspectrophot": 15383,
+      "Ġbile": 15384,
+      "ĠHans": 15385,
+      "ĠAcknowledgements": 15386,
+      "ratio": 15387,
+      "ĠSecondary": 15388,
+      "Ġdownregulated": 15389,
+      "fixed": 15390,
+      "Obs": 15391,
+      "ĠHL": 15392,
+      "Ġsends": 15393,
+      "tings": 15394,
+      "Ġfi": 15395,
+      "ĠPaper": 15396,
+      "Ġultraviolet": 15397,
+      "ĠBall": 15398,
+      "Ġdrastic": 15399,
+      "ailure": 15400,
+      "oil": 15401,
+      "exchange": 15402,
+      "ĠDan": 15403,
+      "ĠAuto": 15404,
+      "Ġarchae": 15405,
+      "ĠCollection": 15406,
+      "Ġantiviral": 15407,
+      "ĠChemistry": 15408,
+      "Ġferr": 15409,
+      "choice": 15410,
+      "vac": 15411,
+      "olipid": 15412,
+      "Ġdanger": 15413,
+      "ĠLittle": 15414,
+      "Ġdehyd": 15415,
+      "Ġoccasion": 15416,
+      "opropyl": 15417,
+      "abe": 15418,
+      "Ġinterferon": 15419,
+      "Ġexport": 15420,
+      "onitrile": 15421,
+      "pd": 15422,
+      "ĠContext": 15423,
+      "ruz": 15424,
+      "ĠDys": 15425,
+      "Ġassembl": 15426,
+      "Ġoils": 15427,
+      "Image": 15428,
+      "rowing": 15429,
+      "Ġaneurys": 15430,
+      "Ġliquids": 15431,
+      "Ġactively": 15432,
+      "Ġevapor": 15433,
+      "ĠPresent": 15434,
+      "Ġconstitutive": 15435,
+      "ĠSite": 15436,
+      "Ġscript": 15437,
+      "Ġrepeats": 15438,
+      "ĠSIR": 15439,
+      "ĠFilm": 15440,
+      "ĠSanta": 15441,
+      "ĠRepresentation": 15442,
+      "ĠAma": 15443,
+      "ordon": 15444,
+      "ĠMolecule": 15445,
+      "Ġgoverning": 15446,
+      "ĠSoil": 15447,
+      "Ver": 15448,
+      "Ġphotonic": 15449,
+      "tify": 15450,
+      "ĠLewis": 15451,
+      "athered": 15452,
+      "Ġcategorical": 15453,
+      "iscellaneous": 15454,
+      "update": 15455,
+      "Ġdeficit": 15456,
+      "Ġadjuvant": 15457,
+      "ĠHenry": 15458,
+      "Group": 15459,
+      "istency": 15460,
+      "agraph": 15461,
+      "ĠImproving": 15462,
+      "El": 15463,
+      "Ġflame": 15464,
+      "rogate": 15465,
+      "omorph": 15466,
+      "Ġqubits": 15467,
+      "Ġillustration": 15468,
+      "ĠFlorida": 15469,
+      "ĠDG": 15470,
+      "bigcup": 15471,
+      "Ġprovince": 15472,
+      "egradation": 15473,
+      "ĠLandau": 15474,
+      "Ġgrating": 15475,
+      "Ġinsects": 15476,
+      "Ġdraft": 15477,
+      "ĠHb": 15478,
+      "Ġss": 15479,
+      "ĠRas": 15480,
+      "Ġmucosa": 15481,
+      "Ġhydroxyl": 15482,
+      "Ġmodest": 15483,
+      "Ġconfirming": 15484,
+      "ĠGalaxies": 15485,
+      "Gaussian": 15486,
+      "ĠRetrie": 15487,
+      "Ġrestored": 15488,
+      "memory": 15489,
+      "Ġreinforced": 15490,
+      "rific": 15491,
+      "Ġassisted": 15492,
+      "Ġaffiliations": 15493,
+      "RC": 15494,
+      "ducer": 15495,
+      "ĠIntellig": 15496,
+      "ĠASD": 15497,
+      "modium": 15498,
+      "Ġomitted": 15499,
+      "okers": 15500,
+      "Ġguided": 15501,
+      "Ġgraphical": 15502,
+      "ĠQual": 15503,
+      "Due": 15504,
+      "Ġnemat": 15505,
+      "variable": 15506,
+      "Ġsenescence": 15507,
+      "Ġpipe": 15508,
+      "Ġsustainable": 15509,
+      "Ġteacher": 15510,
+      "Ġthing": 15511,
+      "ĠGPU": 15512,
+      "TB": 15513,
+      "Ġreform": 15514,
+      "Ġreflex": 15515,
+      "Ġindicative": 15516,
+      "about": 15517,
+      "Ġopi": 15518,
+      "effect": 15519,
+      "Ġdispersed": 15520,
+      "kh": 15521,
+      "ithelial": 15522,
+      "ĠTreg": 15523,
+      "ipl": 15524,
+      "ĠAutomatic": 15525,
+      "Ġnitro": 15526,
+      "complete": 15527,
+      "Ġbosons": 15528,
+      "Ġpac": 15529,
+      "Ġavoiding": 15530,
+      "isl": 15531,
+      "plasty": 15532,
+      "responsive": 15533,
+      "dest": 15534,
+      "ĠBrad": 15535,
+      "ĠDecision": 15536,
+      "ĠDiscovery": 15537,
+      "Ġchicken": 15538,
+      "mus": 15539,
+      "ĠWITH": 15540,
+      "Ġtric": 15541,
+      "Ġquartz": 15542,
+      "onstruction": 15543,
+      "ĠFields": 15544,
+      "Ġassim": 15545,
+      "oprot": 15546,
+      "Ġguaranteed": 15547,
+      "fat": 15548,
+      "icts": 15549,
+      "Ġchol": 15550,
+      "ido": 15551,
+      "ĠKL": 15552,
+      "Ġchitosan": 15553,
+      "ĠNd": 15554,
+      "ĠOscill": 15555,
+      "Ġevolve": 15556,
+      "cu": 15557,
+      "Ġmast": 15558,
+      "Ġamph": 15559,
+      "torch": 15560,
+      "Vis": 15561,
+      "entity": 15562,
+      "ĠAdam": 15563,
+      "Ġdevoted": 15564,
+      "Ġethical": 15565,
+      "Ġpremature": 15566,
+      "Ġconsumer": 15567,
+      "Ġrecursive": 15568,
+      "Ġgluon": 15569,
+      "Ġmoderately": 15570,
+      "Ġmodalities": 15571,
+      "Ġcanal": 15572,
+      "force": 15573,
+      "ĠChlor": 15574,
+      "slash": 15575,
+      "sten": 15576,
+      "Ġcommercially": 15577,
+      "ongs": 15578,
+      "Ġstimulate": 15579,
+      "atinum": 15580,
+      "ĠRail": 15581,
+      "Ġconvective": 15582,
+      "Ġarteries": 15583,
+      "inv": 15584,
+      "ĠWol": 15585,
+      "ĠLung": 15586,
+      "letes": 15587,
+      "raphy": 15588,
+      "ĠHI": 15589,
+      "Ġgraphite": 15590,
+      "Ġhousing": 15591,
+      "each": 15592,
+      "Ġcalor": 15593,
+      "acetamide": 15594,
+      "rochemical": 15595,
+      "Ġhands": 15596,
+      "Ġelucidate": 15597,
+      "ĠChand": 15598,
+      "road": 15599,
+      "nova": 15600,
+      "ĠLineage": 15601,
+      "Ġram": 15602,
+      "Ġfight": 15603,
+      "Ġrecommendation": 15604,
+      "Ġamongst": 15605,
+      "Ġswitches": 15606,
+      "berry": 15607,
+      "Ġtherein": 15608,
+      "algebras": 15609,
+      "ĠTaken": 15610,
+      "azz": 15611,
+      "Ġfurn": 15612,
+      "Ġamel": 15613,
+      "Ġteachers": 15614,
+      "arn": 15615,
+      "Ġavoided": 15616,
+      "Ġaverages": 15617,
+      "amer": 15618,
+      "ĠCondition": 15619,
+      "Ġdislocation": 15620,
+      "ircon": 15621,
+      "Ġadolescent": 15622,
+      "Ġtur": 15623,
+      "env": 15624,
+      "Ġze": 15625,
+      "DL": 15626,
+      "loading": 15627,
+      "icidal": 15628,
+      "category": 15629,
+      "ĠDB": 15630,
+      "Ġmucosal": 15631,
+      "ĠRG": 15632,
+      "Ġtaxonomic": 15633,
+      "Ġmutagen": 15634,
+      "ĠStage": 15635,
+      "necess": 15636,
+      "ĠPerm": 15637,
+      "Ġocclusion": 15638,
+      "Ġexploited": 15639,
+      "Ġanaerobic": 15640,
+      "uled": 15641,
+      "Ġwanted": 15642,
+      "ĠCombining": 15643,
+      "Ġsubcutaneous": 15644,
+      "Recomm": 15645,
+      "Ġdiscusses": 15646,
+      "Ġcounterpart": 15647,
+      "ĠFB": 15648,
+      "Ġadsorbed": 15649,
+      "don": 15650,
+      "Many": 15651,
+      "ĠSweden": 15652,
+      "ĠAndrew": 15653,
+      "enhanced": 15654,
+      "Ġdoctor": 15655,
+      "ĠKorean": 15656,
+      "ĠSAR": 15657,
+      "Ġmating": 15658,
+      "aturation": 15659,
+      "ĠLatin": 15660,
+      "Ġsorting": 15661,
+      "Ġskip": 15662,
+      "Os": 15663,
+      "Ġwife": 15664,
+      "Ġcommittee": 15665,
+      "lvert": 15666,
+      "ĠACC": 15667,
+      "ĠComm": 15668,
+      "Ġsubtle": 15669,
+      "ĠSurvival": 15670,
+      "because": 15671,
+      "Ġfeat": 15672,
+      "ĠPortug": 15673,
+      "ARY": 15674,
+      "ĠISB": 15675,
+      "itron": 15676,
+      "Ġsectors": 15677,
+      "Ġadjoint": 15678,
+      "ĠAlexander": 15679,
+      "Ġimpurity": 15680,
+      "ĠMarine": 15681,
+      "lact": 15682,
+      "Ġtrapping": 15683,
+      "Ġgeneralize": 15684,
+      "filter": 15685,
+      "Ġpolarity": 15686,
+      "Also": 15687,
+      "Ġstabilized": 15688,
+      "ĠVirgin": 15689,
+      "Ġstores": 15690,
+      "PAGE": 15691,
+      "Ġdrawback": 15692,
+      "Ġâݪ": 15693,
+      "jet": 15694,
+      "Ġsubstituted": 15695,
+      "LINE": 15696,
+      "Ġoutperforms": 15697,
+      "Ġtermed": 15698,
+      "Ġweekly": 15699,
+      "Ġpolyc": 15700,
+      "Ġfused": 15701,
+      "Ġferromagnetic": 15702,
+      "lr": 15703,
+      "ellites": 15704,
+      "ĠTurn": 15705,
+      "ĠCulture": 15706,
+      "prise": 15707,
+      "ÅĤ": 15708,
+      "omposition": 15709,
+      "elfare": 15710,
+      "ĠGoogle": 15711,
+      "oarth": 15712,
+      "Ġë": 15713,
+      "Ġmist": 15714,
+      "ĠMathematics": 15715,
+      "SET": 15716,
+      "Ġepochs": 15717,
+      "Ġcontras": 15718,
+      "ishment": 15719,
+      "ĠFirstly": 15720,
+      "Ġdeclared": 15721,
+      "aur": 15722,
+      "ĠPed": 15723,
+      "Ġreplicate": 15724,
+      "Ġeligible": 15725,
+      "Ġconcaten": 15726,
+      "Ġcig": 15727,
+      "Ġtriplet": 15728,
+      "found": 15729,
+      "ĠCz": 15730,
+      "Ġaccomplished": 15731,
+      "Ġgoverned": 15732,
+      "onuclear": 15733,
+      "ĠNY": 15734,
+      "ĠEthiop": 15735,
+      "Ġinject": 15736,
+      "Ġeosin": 15737,
+      "annon": 15738,
+      "olo": 15739,
+      "ĠMHC": 15740,
+      "Ġpreoperative": 15741,
+      "Ġdates": 15742,
+      "Ġsigma": 15743,
+      "Long": 15744,
+      "ĠReson": 15745,
+      "Ġsymptomatic": 15746,
+      "Ġvolunteers": 15747,
+      "Ġcooperation": 15748,
+      "Ġarr": 15749,
+      "Ġcloned": 15750,
+      "Ġdent": 15751,
+      "ĠSob": 15752,
+      "Ġcathode": 15753,
+      "ctx": 15754,
+      "Ġencephal": 15755,
+      "Ġpiv": 15756,
+      "vive": 15757,
+      "umetric": 15758,
+      "ĠFF": 15759,
+      "Ġunderestim": 15760,
+      "Ġcoded": 15761,
+      "Ġanalges": 15762,
+      "spectral": 15763,
+      "Ġattracted": 15764,
+      "Ġtwenty": 15765,
+      "Ġinactive": 15766,
+      "Ġvictim": 15767,
+      "Ġholder": 15768,
+      "ogenes": 15769,
+      "Ġsuffering": 15770,
+      "rex": 15771,
+      "Ġprophyl": 15772,
+      "ĠUniversal": 15773,
+      "Ġdenom": 15774,
+      "stolic": 15775,
+      "ansion": 15776,
+      "SIZE": 15777,
+      "ĠHCV": 15778,
+      "Ġtechnological": 15779,
+      "CNN": 15780,
+      "enching": 15781,
+      "Ġdebris": 15782,
+      "ĠBoundary": 15783,
+      "linking": 15784,
+      "Ġstopped": 15785,
+      "ĠDie": 15786,
+      "ĠCosm": 15787,
+      "Ġturning": 15788,
+      "Ġglycoprotein": 15789,
+      "ĠKumar": 15790,
+      "Ġpg": 15791,
+      "ĠBY": 15792,
+      "Ġrising": 15793,
+      "ĠROC": 15794,
+      "Despite": 15795,
+      "ĠBoolean": 15796,
+      "ilder": 15797,
+      "Ġexponents": 15798,
+      "inters": 15799,
+      "printf": 15800,
+      "Ġlit": 15801,
+      "track": 15802,
+      "Ġfidelity": 15803,
+      "Ġsmoke": 15804,
+      "otemporal": 15805,
+      "Ġadmissible": 15806,
+      "ĠBoltzmann": 15807,
+      "TF": 15808,
+      "olite": 15809,
+      "liament": 15810,
+      "Ġcalculus": 15811,
+      "itized": 15812,
+      "Ġdivergent": 15813,
+      "Ġcolonization": 15814,
+      "Ġconvergent": 15815,
+      "ĠHas": 15816,
+      "Ġconsumers": 15817,
+      "Ġmyc": 15818,
+      "Ġcontig": 15819,
+      "Ġepidemiology": 15820,
+      "és": 15821,
+      "ĠAssoci": 15822,
+      "given": 15823,
+      "Ġwhilst": 15824,
+      "ĠKur": 15825,
+      "Ġreasonably": 15826,
+      "Ġaerobic": 15827,
+      "separ": 15828,
+      "Ġchecks": 15829,
+      "ĠSemantic": 15830,
+      "Ġserving": 15831,
+      "ĠAtmosp": 15832,
+      "Ġoxidized": 15833,
+      "coupled": 15834,
+      "ĠbioRxiv": 15835,
+      "Ġtuned": 15836,
+      "uspended": 15837,
+      "Ġindirectly": 15838,
+      "ĠCAD": 15839,
+      "ĠCurrently": 15840,
+      "Ġbehaviours": 15841,
+      "ĠPPAR": 15842,
+      "rors": 15843,
+      "ereb": 15844,
+      "Ġwidths": 15845,
+      "diagonal": 15846,
+      "ervice": 15847,
+      "Ġole": 15848,
+      "means": 15849,
+      "IME": 15850,
+      "ĠTracking": 15851,
+      "Ġacknowledge": 15852,
+      "ĠHon": 15853,
+      "ĠTechniques": 15854,
+      "ĠOxid": 15855,
+      "blind": 15856,
+      "Ġdiast": 15857,
+      "named": 15858,
+      "asitic": 15859,
+      "Ġpreparations": 15860,
+      "ĠArth": 15861,
+      "Ġpreserves": 15862,
+      "Ġfasc": 15863,
+      "Ġwaveform": 15864,
+      "ĠCrystal": 15865,
+      "Ġuncom": 15866,
+      "Ġelast": 15867,
+      "Ġfunctionally": 15868,
+      "Hom": 15869,
+      "ĠCoast": 15870,
+      "Ġoptic": 15871,
+      "ĠAlternatively": 15872,
+      "onyl": 15873,
+      "ĠLig": 15874,
+      "aldehyde": 15875,
+      "Ġsimulator": 15876,
+      "Ġdramatic": 15877,
+      "ifera": 15878,
+      "Ġexhibiting": 15879,
+      "Ġbehavioural": 15880,
+      "thick": 15881,
+      "xture": 15882,
+      "Ġexecutive": 15883,
+      "Ġcondensate": 15884,
+      "ĠOutcomes": 15885,
+      "Text": 15886,
+      "ointed": 15887,
+      "ĠCopyright": 15888,
+      "Ġdc": 15889,
+      "odd": 15890,
+      "ĠDiversity": 15891,
+      "chip": 15892,
+      "ĠBuilding": 15893,
+      "Ġpulsed": 15894,
+      "harmonic": 15895,
+      "Ġclinicians": 15896,
+      "dp": 15897,
+      "ĠqPCR": 15898,
+      "marks": 15899,
+      "Ġappreci": 15900,
+      "ĠLaser": 15901,
+      "Ġsizeof": 15902,
+      "yrene": 15903,
+      "Ġcooperative": 15904,
+      "generative": 15905,
+      "ĠLib": 15906,
+      "Ġdispersal": 15907,
+      "Ġevolving": 15908,
+      "ĠStatus": 15909,
+      "Ġsupercon": 15910,
+      "ĠMamm": 15911,
+      "Ġinterstitial": 15912,
+      "isenberg": 15913,
+      "Ġâľ": 15914,
+      "Ġconfocal": 15915,
+      "Ġmodulates": 15916,
+      "hour": 15917,
+      "Ġperoxide": 15918,
+      "dependence": 15919,
+      "Ġperturbed": 15920,
+      "illation": 15921,
+      "Ġplaque": 15922,
+      "ĠNeumann": 15923,
+      "Ġtriggers": 15924,
+      "omain": 15925,
+      "ĠAdministration": 15926,
+      "olia": 15927,
+      "ĠMIC": 15928,
+      "osaic": 15929,
+      "ĠGB": 15930,
+      "textnormal": 15931,
+      "Ġdominance": 15932,
+      "ĠExper": 15933,
+      "CAM": 15934,
+      "ĠAbout": 15935,
+      "ĠGarc": 15936,
+      "Ġsummarizes": 15937,
+      "App": 15938,
+      "charomyces": 15939,
+      "tificial": 15940,
+      "Ġglycerol": 15941,
+      "ĠAssumption": 15942,
+      "Ġtect": 15943,
+      "ĠFW": 15944,
+      "Ġcotton": 15945,
+      "general": 15946,
+      "ĠFern": 15947,
+      "Pt": 15948,
+      "Ġworker": 15949,
+      "Ġanion": 15950,
+      "grams": 15951,
+      "req": 15952,
+      "Ġlooks": 15953,
+      "Ġimplementations": 15954,
+      "ĠColumb": 15955,
+      "agi": 15956,
+      "ĠAttention": 15957,
+      "ĠTeam": 15958,
+      "oning": 15959,
+      "onential": 15960,
+      "tiny": 15961,
+      "ĠHighly": 15962,
+      "textup": 15963,
+      "Ġinvertible": 15964,
+      "ocortic": 15965,
+      "Inf": 15966,
+      "ĠOfficial": 15967,
+      "ĠModelling": 15968,
+      "Ġinclusions": 15969,
+      "Ġblank": 15970,
+      "Ġsight": 15971,
+      "ĠGamma": 15972,
+      "Ġlepton": 15973,
+      "Ġpneumoniae": 15974,
+      "Ġrotor": 15975,
+      "Ġethnic": 15976,
+      "Ġretain": 15977,
+      "varying": 15978,
+      "ĠEB": 15979,
+      "Ġastrocytes": 15980,
+      "ĠNorm": 15981,
+      "Ġnanom": 15982,
+      "classical": 15983,
+      "Ġshadow": 15984,
+      "ĠReferences": 15985,
+      "ĠFS": 15986,
+      "Ġnonnegative": 15987,
+      "bond": 15988,
+      "ĠCoh": 15989,
+      "Ġnumpy": 15990,
+      "Ġoct": 15991,
+      "span": 15992,
+      "racts": 15993,
+      "Ġnotably": 15994,
+      "Ġsophistic": 15995,
+      "PAR": 15996,
+      "Ġhormones": 15997,
+      "Ġtensors": 15998,
+      "ĠÌĦ": 15999,
+      "ĠConstraints": 16000,
+      "ĠâIJ": 16001,
+      "Ġtransit": 16002,
+      "Ġruntime": 16003,
+      "author": 16004,
+      "Ġprompt": 16005,
+      "ĠSG": 16006,
+      "Ġgrate": 16007,
+      "cemia": 16008,
+      "ĠLyapunov": 16009,
+      "convex": 16010,
+      "Ġforecasting": 16011,
+      "push": 16012,
+      "Ġjurisdictional": 16013,
+      "ÃĢ": 16014,
+      "Ġbiomedical": 16015,
+      "Ġepilepsy": 16016,
+      "feature": 16017,
+      "wiki": 16018,
+      "View": 16019,
+      "Ġlesser": 16020,
+      "Ġconjugated": 16021,
+      "Ġwaiting": 16022,
+      "ĠWord": 16023,
+      "IZ": 16024,
+      "Ġhydroxy": 16025,
+      "Ġdisp": 16026,
+      "Ġseeded": 16027,
+      "fitting": 16028,
+      "Ġstratification": 16029,
+      "Ġendpoint": 16030,
+      "Ġmediators": 16031,
+      "ductive": 16032,
+      "Ġinjections": 16033,
+      "ĠMicrobi": 16034,
+      "Ġinsert": 16035,
+      "ĠEmb": 16036,
+      "Ġstopping": 16037,
+      "welling": 16038,
+      "Ġirradiated": 16039,
+      "Ġmetallicity": 16040,
+      "vinyl": 16041,
+      "Ġplasmids": 16042,
+      "Rep": 16043,
+      "ĠDifferenti": 16044,
+      "ĠSmart": 16045,
+      "ĠIdentifier": 16046,
+      "ĠBF": 16047,
+      "ropic": 16048,
+      "Ġkinematics": 16049,
+      "Ġinoculated": 16050,
+      "CK": 16051,
+      "auses": 16052,
+      "ĠReturns": 16053,
+      "reement": 16054,
+      "Ġanticancer": 16055,
+      "Ġspecifications": 16056,
+      "Ġadds": 16057,
+      "Ġstake": 16058,
+      "Ġwheel": 16059,
+      "üller": 16060,
+      "ĠSon": 16061,
+      "Ġrupture": 16062,
+      "Ġsold": 16063,
+      "than": 16064,
+      "Ġintermedi": 16065,
+      "ĠNik": 16066,
+      "Ġtuple": 16067,
+      "establ": 16068,
+      "Ġnorthe": 16069,
+      "Ġsuppresses": 16070,
+      "Ġfet": 16071,
+      "Ġwashing": 16072,
+      "Ġinterplay": 16073,
+      "Ġregularly": 16074,
+      "EXT": 16075,
+      "Ġemployees": 16076,
+      "yz": 16077,
+      "rupted": 16078,
+      "etts": 16079,
+      "ĠUAV": 16080,
+      "Ġdifferentiable": 16081,
+      "inge": 16082,
+      "MDA": 16083,
+      "Ġho": 16084,
+      "Ġtags": 16085,
+      "Ġcompatibility": 16086,
+      "ĠÃĥ": 16087,
+      "bus": 16088,
+      "ĠUC": 16089,
+      "Ġtokens": 16090,
+      "Ġclients": 16091,
+      "Ġprescription": 16092,
+      "ĠÌĪ": 16093,
+      "ĠReaction": 16094,
+      "velocity": 16095,
+      "ĠNLR": 16096,
+      "ĠGast": 16097,
+      "ĠPlasmodium": 16098,
+      "ĠCut": 16099,
+      "Ġnas": 16100,
+      "grained": 16101,
+      "Ġchromosomal": 16102,
+      "Ġpossesses": 16103,
+      "Ġmath": 16104,
+      "Ġelected": 16105,
+      "placement": 16106,
+      "Ġcollecting": 16107,
+      "Ġgels": 16108,
+      "aire": 16109,
+      "Ġdeformations": 16110,
+      "raise": 16111,
+      "Ġflank": 16112,
+      "sulfanyl": 16113,
+      "zens": 16114,
+      "priate": 16115,
+      "Ġchlorophyll": 16116,
+      "abi": 16117,
+      "available": 16118,
+      "ا": 16119,
+      "Ġtack": 16120,
+      "fields": 16121,
+      "Ġrichness": 16122,
+      "Ġimplants": 16123,
+      "obenz": 16124,
+      "idential": 16125,
+      "Ġbillion": 16126,
+      "utor": 16127,
+      "ĠISBN": 16128,
+      "Ġinsurance": 16129,
+      "NET": 16130,
+      "Ġinadequ": 16131,
+      "Ġmerged": 16132,
+      "ĠRange": 16133,
+      "Ġavoidance": 16134,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 16135,
+      "rics": 16136,
+      "Ġexclusive": 16137,
+      "LV": 16138,
+      "Ġ": 16139,
+      "Ġcategorized": 16140,
+      "Ġultrasonic": 16141,
+      "ipe": 16142,
+      "icans": 16143,
+      "ĠAPP": 16144,
+      "Ġtraumatic": 16145,
+      "Ba": 16146,
+      "ĠAssay": 16147,
+      "ĠGrid": 16148,
+      "ĠClassical": 16149,
+      "ĠDES": 16150,
+      "Ġsoybean": 16151,
+      "Ġtopography": 16152,
+      "ĠControll": 16153,
+      "Ġemotions": 16154,
+      "Ġcarbohydrate": 16155,
+      "Ġconsol": 16156,
+      "oxyl": 16157,
+      "Ġbifurcation": 16158,
+      "Ġcoil": 16159,
+      "find": 16160,
+      "Ġwitness": 16161,
+      "ĠLF": 16162,
+      "threshold": 16163,
+      "Ġaddressing": 16164,
+      "Ġscrew": 16165,
+      "Ġactor": 16166,
+      "ĠWell": 16167,
+      "Ġ": 16168,
+      "ïĺ": 16169,
+      "ĠDF": 16170,
+      "ĠCorporation": 16171,
+      "ĠMitochondrial": 16172,
+      "Ġkpc": 16173,
+      "unders": 16174,
+      "Ġfibrin": 16175,
+      "axel": 16176,
+      "Ġpolyt": 16177,
+      "Ġshaped": 16178,
+      "rez": 16179,
+      "steresis": 16180,
+      "ĠComprehens": 16181,
+      "Ġ": 16182,
+      "dh": 16183,
+      "Ġsemic": 16184,
+      "Ġmot": 16185,
+      "ĠDavis": 16186,
+      "ska": 16187,
+      "ĠLH": 16188,
+      "Ġexpansions": 16189,
+      "acks": 16190,
+      "Ġoptimizing": 16191,
+      "eak": 16192,
+      "ĠQi": 16193,
+      "mul": 16194,
+      "ograft": 16195,
+      "Ġsuicide": 16196,
+      "calar": 16197,
+      "ĠScott": 16198,
+      "Ġthinking": 16199,
+      "Ġdirectional": 16200,
+      "Ġsurfactant": 16201,
+      "Ġdegraded": 16202,
+      "Ġregimen": 16203,
+      "itative": 16204,
+      "ĠVersion": 16205,
+      "ĠMaster": 16206,
+      "ĠSimulations": 16207,
+      "NCBI": 16208,
+      "lip": 16209,
+      "Ġreagents": 16210,
+      "Ġposted": 16211,
+      "osus": 16212,
+      "Ġlayered": 16213,
+      "ĠSpectrum": 16214,
+      "ĠGraphs": 16215,
+      "burst": 16216,
+      "Ġlived": 16217,
+      "Ġelemental": 16218,
+      "Ġ": 16219,
+      "ĠDiscrete": 16220,
+      "Ġexcluding": 16221,
+      "Ġoriginating": 16222,
+      "ĠGames": 16223,
+      "continuous": 16224,
+      "ATED": 16225,
+      "Ġpyram": 16226,
+      "luent": 16227,
+      "Ġtwisted": 16228,
+      "ĠNb": 16229,
+      "oxicity": 16230,
+      "Ġscr": 16231,
+      "Ġfun": 16232,
+      "ĠSegmentation": 16233,
+      "Ġphenol": 16234,
+      "Ġmeters": 16235,
+      "ĠEigen": 16236,
+      "ĠWeak": 16237,
+      "Ġschematic": 16238,
+      "rone": 16239,
+      "Ġphilos": 16240,
+      "titis": 16241,
+      "ĠIreland": 16242,
+      "Ġgy": 16243,
+      "ĠPTM": 16244,
+      "Ġpacking": 16245,
+      "ilinear": 16246,
+      "zeros": 16247,
+      "Ġubiquitin": 16248,
+      "ĠPressure": 16249,
+      "Ġinfiltr": 16250,
+      "ENS": 16251,
+      "validation": 16252,
+      "Ġprone": 16253,
+      "Ġoutline": 16254,
+      "hs": 16255,
+      "rength": 16256,
+      "Ġattain": 16257,
+      "Ġtwe": 16258,
+      "Ġtandem": 16259,
+      "Can": 16260,
+      "Ġlatitude": 16261,
+      "uitary": 16262,
+      "Ġvoltages": 16263,
+      "ĠGao": 16264,
+      "Ġpharmacokine": 16265,
+      "Ġcontextual": 16266,
+      "Ġxyl": 16267,
+      "elson": 16268,
+      "ĠMetabolic": 16269,
+      "oden": 16270,
+      "tiles": 16271,
+      "fficking": 16272,
+      "Ġdistilled": 16273,
+      "Ġalph": 16274,
+      "Ġpiezo": 16275,
+      "growth": 16276,
+      "Ġbore": 16277,
+      "Ġredundant": 16278,
+      "Ġdemonstration": 16279,
+      "Ġik": 16280,
+      "Ġrounds": 16281,
+      "ĠSri": 16282,
+      "figuration": 16283,
+      "ĠRayleigh": 16284,
+      "Line": 16285,
+      "ovol": 16286,
+      "Ġobstacle": 16287,
+      "cn": 16288,
+      "Ġbioactive": 16289,
+      "ĠOA": 16290,
+      "physical": 16291,
+      "atidyl": 16292,
+      "ACC": 16293,
+      "how": 16294,
+      "Ġresultant": 16295,
+      "ĠHubble": 16296,
+      "ĠVor": 16297,
+      "Ġensuring": 16298,
+      "Ġannotations": 16299,
+      "acyl": 16300,
+      "stituted": 16301,
+      "ĠAmb": 16302,
+      "feeding": 16303,
+      "Ġpresumably": 16304,
+      "Ġblockade": 16305,
+      "Ġsoc": 16306,
+      "ĠUrb": 16307,
+      "Ġmultiplied": 16308,
+      "Ġdiffe": 16309,
+      "Ġreflectance": 16310,
+      "ĠKeywords": 16311,
+      "ĠBayes": 16312,
+      "odeficiency": 16313,
+      "ĠBinding": 16314,
+      "inely": 16315,
+      "except": 16316,
+      "ĠUltr": 16317,
+      "ĠBrazilian": 16318,
+      "Number": 16319,
+      "Ġmassless": 16320,
+      "ĠConsistent": 16321,
+      "Ġcrisis": 16322,
+      "ogs": 16323,
+      "Ġresidence": 16324,
+      "Ġimper": 16325,
+      "fts": 16326,
+      "Ġcaptures": 16327,
+      "ĠSyndrome": 16328,
+      "Ġdimensionality": 16329,
+      "jun": 16330,
+      "Ġexhaus": 16331,
+      "ĠModern": 16332,
+      "Ġpercenti": 16333,
+      "Level": 16334,
+      "ĠResponses": 16335,
+      "Ġlaunched": 16336,
+      "Ġrepos": 16337,
+      "ĠKam": 16338,
+      "atility": 16339,
+      "Ġcarotid": 16340,
+      "rotic": 16341,
+      "ĠMand": 16342,
+      "UB": 16343,
+      "ĠMixed": 16344,
+      "Ġindexes": 16345,
+      "Ġcisplatin": 16346,
+      "ican": 16347,
+      "ionine": 16348,
+      "Ġhab": 16349,
+      "ĠIce": 16350,
+      "ĠGT": 16351,
+      "ĠAgg": 16352,
+      "ĠLDL": 16353,
+      "Ġvolcanic": 16354,
+      "dB": 16355,
+      "ĠElectric": 16356,
+      "Ġtmp": 16357,
+      "Ġgrids": 16358,
+      "liquid": 16359,
+      "prom": 16360,
+      "ĠGAL": 16361,
+      "Ġpestic": 16362,
+      "Ġhelium": 16363,
+      "Ġ": 16364,
+      "ĠDong": 16365,
+      "Ġmagnification": 16366,
+      "kip": 16367,
+      "ĠGrad": 16368,
+      "ĠWei": 16369,
+      "ĠPDF": 16370,
+      "ĠGluc": 16371,
+      "Pol": 16372,
+      "Ġtumorigen": 16373,
+      "yrin": 16374,
+      "Ġshelf": 16375,
+      "adher": 16376,
+      "entials": 16377,
+      "sn": 16378,
+      "Ġcultivars": 16379,
+      "Ġorbitals": 16380,
+      "ĠPEG": 16381,
+      "ĠAnne": 16382,
+      "eno": 16383,
+      "Ġattended": 16384,
+      "ophore": 16385,
+      "ishop": 16386,
+      "Ġfriends": 16387,
+      "posable": 16388,
+      "Ġimpose": 16389,
+      "Ġendemic": 16390,
+      "Ġsick": 16391,
+      "shifts": 16392,
+      "ĠOutput": 16393,
+      "LM": 16394,
+      "ĠMiscellaneous": 16395,
+      "Ġthousands": 16396,
+      "ĠDataset": 16397,
+      "Ġperturbative": 16398,
+      "oprec": 16399,
+      "Ġbene": 16400,
+      "Ġreef": 16401,
+      "Ġfossil": 16402,
+      "Ġcited": 16403,
+      "plicates": 16404,
+      "Ġrelates": 16405,
+      "ĠVII": 16406,
+      "Ġantifer": 16407,
+      "Ġglasses": 16408,
+      "closure": 16409,
+      "Ġrubber": 16410,
+      "Ġbird": 16411,
+      "Ġsupersymmetry": 16412,
+      "Ġmeson": 16413,
+      "hell": 16414,
+      "Ġparties": 16415,
+      "kar": 16416,
+      "ĠHur": 16417,
+      "ĠEA": 16418,
+      "ĠStars": 16419,
+      "othing": 16420,
+      "hot": 16421,
+      "illar": 16422,
+      "ASP": 16423,
+      "hev": 16424,
+      "ïĥ": 16425,
+      "aques": 16426,
+      "Ġcoordinated": 16427,
+      "ĠIslands": 16428,
+      "enable": 16429,
+      "SiO": 16430,
+      "Ġexceptional": 16431,
+      "Comb": 16432,
+      "ĠLike": 16433,
+      "Ġbroadly": 16434,
+      "ĠBac": 16435,
+      "Ġnil": 16436,
+      "ipartite": 16437,
+      "rations": 16438,
+      "Ġrewrite": 16439,
+      "Ġsalts": 16440,
+      "dimension": 16441,
+      "ĠVehic": 16442,
+      "Ġhundreds": 16443,
+      "ĠUr": 16444,
+      "Ġendpoints": 16445,
+      "ĠMODEL": 16446,
+      "ĠHBV": 16447,
+      "ĠVirtual": 16448,
+      "ĠConfl": 16449,
+      "ĠPractice": 16450,
+      "ĠAFM": 16451,
+      "Ġadversarial": 16452,
+      "Ġdiameters": 16453,
+      "Ġtransported": 16454,
+      "REM": 16455,
+      "ĠBart": 16456,
+      "Ġedition": 16457,
+      "Ġturbine": 16458,
+      "Ġminus": 16459,
+      "otechnology": 16460,
+      "Ig": 16461,
+      "Ġbigger": 16462,
+      "abul": 16463,
+      "Ġperoxidase": 16464,
+      "white": 16465,
+      "ĠSed": 16466,
+      "dihydro": 16467,
+      "Ġsegregation": 16468,
+      "Ġreductase": 16469,
+      "Ġhoriz": 16470,
+      "Ġinfinitely": 16471,
+      "availability": 16472,
+      "Ġactivator": 16473,
+      "Ġcensus": 16474,
+      "pressing": 16475,
+      "Ġspirit": 16476,
+      "conver": 16477,
+      "ĠQuantification": 16478,
+      "omerase": 16479,
+      "Ġrelapse": 16480,
+      "ĠFinal": 16481,
+      "Ġoverweight": 16482,
+      "aper": 16483,
+      "Ġformulae": 16484,
+      "rr": 16485,
+      "Ġfemoral": 16486,
+      "Ġfoam": 16487,
+      "otics": 16488,
+      "Ġprovider": 16489,
+      "Ġinstrumental": 16490,
+      "Ġadvice": 16491,
+      "Ġoccupation": 16492,
+      "assembly": 16493,
+      "bias": 16494,
+      "ĠNOT": 16495,
+      "restric": 16496,
+      "ĠProtocol": 16497,
+      "ĠCandida": 16498,
+      "ĠRhod": 16499,
+      "arden": 16500,
+      "funder": 16501,
+      "osens": 16502,
+      "Ġparams": 16503,
+      "front": 16504,
+      "Ġexerc": 16505,
+      "Ġgalactic": 16506,
+      "rvert": 16507,
+      "Ġimbalance": 16508,
+      "Ġkilling": 16509,
+      "ĠGenomic": 16510,
+      "Ġip": 16511,
+      "Ġcave": 16512,
+      "Ġfalc": 16513,
+      "ĠRM": 16514,
+      "Ġcarries": 16515,
+      "global": 16516,
+      "Ġcube": 16517,
+      "Ġrigorous": 16518,
+      "Ġcomputes": 16519,
+      "QP": 16520,
+      "Ġexposures": 16521,
+      "cover": 16522,
+      "ologically": 16523,
+      "Oper": 16524,
+      "Ġpec": 16525,
+      "Ġinhomogeneous": 16526,
+      "Ġservers": 16527,
+      "aliana": 16528,
+      "nb": 16529,
+      "Ġexplaining": 16530,
+      "Ġshrink": 16531,
+      "Ġcomorbid": 16532,
+      "ethoxy": 16533,
+      "outheast": 16534,
+      "Ġcourses": 16535,
+      "ĠNM": 16536,
+      "ĠShape": 16537,
+      "Ġflies": 16538,
+      "ĠMir": 16539,
+      "Ġpublicly": 16540,
+      "Ġphotometric": 16541,
+      "versible": 16542,
+      "olev": 16543,
+      "Ġvulnerability": 16544,
+      "Ġcations": 16545,
+      "Ġseeking": 16546,
+      "UTR": 16547,
+      "Ġdecomposed": 16548,
+      "Ġhus": 16549,
+      "Ġdisappear": 16550,
+      "Ġencounter": 16551,
+      "Ġtransforming": 16552,
+      "Ġpolymeric": 16553,
+      "Ġdiscretization": 16554,
+      "otoxic": 16555,
+      "ĠIter": 16556,
+      "ĠMari": 16557,
+      "Ġunfold": 16558,
+      "ĠAdult": 16559,
+      "obacillus": 16560,
+      "metal": 16561,
+      "berger": 16562,
+      "raphene": 16563,
+      "respective": 16564,
+      "Ġsurvive": 16565,
+      "ovich": 16566,
+      "Ġprotects": 16567,
+      "ĠRog": 16568,
+      "Ġimmunotherapy": 16569,
+      "ĠDSM": 16570,
+      "Ġanalogy": 16571,
+      "ĠPER": 16572,
+      "ĠPython": 16573,
+      "hum": 16574,
+      "ĠAdj": 16575,
+      "ĠLikewise": 16576,
+      "Ġ": 16577,
+      "Ġstomach": 16578,
+      "Ġinit": 16579,
+      "Ġwires": 16580,
+      "Ġingredients": 16581,
+      "Ġperceptual": 16582,
+      "Hand": 16583,
+      "Back": 16584,
+      "Ġmood": 16585,
+      "Ġdeformed": 16586,
+      "ĠRead": 16587,
+      "Ġrhiz": 16588,
+      "ĠOrganism": 16589,
+      "ĠIndones": 16590,
+      "annot": 16591,
+      "ictory": 16592,
+      "Ġtended": 16593,
+      "ĠSound": 16594,
+      "iax": 16595,
+      "Sr": 16596,
+      "ĠTab": 16597,
+      "ĠLaplacian": 16598,
+      "oluminescence": 16599,
+      "backslash": 16600,
+      "iologic": 16601,
+      "Ġtypename": 16602,
+      "ĠYear": 16603,
+      "Dependent": 16604,
+      "Ġslides": 16605,
+      "Ġsacrific": 16606,
+      "Ġconcomitant": 16607,
+      "opsies": 16608,
+      "Bigg": 16609,
+      "peak": 16610,
+      "ĠApplying": 16611,
+      "Ġcodon": 16612,
+      "ĠSimultaneous": 16613,
+      "tise": 16614,
+      "Ġtertiary": 16615,
+      "ĠPoll": 16616,
+      "Ġrevision": 16617,
+      "RAF": 16618,
+      "xmm": 16619,
+      "Ġsuited": 16620,
+      "ĠRecommend": 16621,
+      "ĠRy": 16622,
+      "Ġsake": 16623,
+      "Ġstretch": 16624,
+      "ĠSampling": 16625,
+      "Ġtubular": 16626,
+      "Ġpark": 16627,
+      "Ġultimate": 16628,
+      "Ġlands": 16629,
+      "ĠCriter": 16630,
+      "assay": 16631,
+      "mor": 16632,
+      "Ġdocking": 16633,
+      "Ġgradual": 16634,
+      "Ġeditor": 16635,
+      "Ġpolice": 16636,
+      "affin": 16637,
+      "ĠDeath": 16638,
+      "Ġpromoters": 16639,
+      "assic": 16640,
+      "Ġwriter": 16641,
+      "ĠVolume": 16642,
+      "iso": 16643,
+      "Ġdisag": 16644,
+      "token": 16645,
+      "Ġsteroid": 16646,
+      "Non": 16647,
+      "ĠMethyl": 16648,
+      "Americ": 16649,
+      "due": 16650,
+      "ĠLess": 16651,
+      "Ġdyst": 16652,
+      "ĠStatement": 16653,
+      "ĠTwenty": 16654,
+      "Ġaccessed": 16655,
+      "Ġblotting": 16656,
+      "ĠCOPD": 16657,
+      "Ġsteam": 16658,
+      "Ġdescriptive": 16659,
+      "ĠVery": 16660,
+      "Ġcapacities": 16661,
+      "ĠPersonal": 16662,
+      "acid": 16663,
+      "ähler": 16664,
+      "estival": 16665,
+      "Context": 16666,
+      "Ġastr": 16667,
+      "Analysis": 16668,
+      "Ġsept": 16669,
+      "Ġprinted": 16670,
+      "dual": 16671,
+      "aman": 16672,
+      "erer": 16673,
+      "Ġweakness": 16674,
+      "ìĿ": 16675,
+      "ĠTranslation": 16676,
+      "Ġpropagating": 16677,
+      "ĠSections": 16678,
+      "aca": 16679,
+      "Ġconfusion": 16680,
+      "IK": 16681,
+      "Ġframeworks": 16682,
+      "Ġsituated": 16683,
+      "Ġstays": 16684,
+      "nodes": 16685,
+      "chen": 16686,
+      "artments": 16687,
+      "Ġfreezing": 16688,
+      "ws": 16689,
+      "nett": 16690,
+      "Ġcontrollers": 16691,
+      "Ġsilic": 16692,
+      "LAST": 16693,
+      "foot": 16694,
+      "ĠDISCU": 16695,
+      "RH": 16696,
+      "ridine": 16697,
+      "ĠRev": 16698,
+      "perg": 16699,
+      "pyrim": 16700,
+      "flags": 16701,
+      "ĠGuide": 16702,
+      "Ġspeaker": 16703,
+      "tisol": 16704,
+      "rell": 16705,
+      "ĠDEG": 16706,
+      "Ġfu": 16707,
+      "ĠGut": 16708,
+      "Ġshar": 16709,
+      "Ġgross": 16710,
+      "Ġcrosses": 16711,
+      "wavelength": 16712,
+      "ĠApplied": 16713,
+      "ïve": 16714,
+      "ĠHB": 16715,
+      "ĠEdge": 16716,
+      "Ġinertial": 16717,
+      "Ġvocal": 16718,
+      "production": 16719,
+      "pathetic": 16720,
+      "Ġplanetary": 16721,
+      "Ġsister": 16722,
+      "Ġminima": 16723,
+      "Ġlongest": 16724,
+      "Ġflash": 16725,
+      "Ġperiodon": 16726,
+      "Ġepidermal": 16727,
+      "Ġfloating": 16728,
+      "GET": 16729,
+      "ĠTake": 16730,
+      "pdf": 16731,
+      "ĠLiquid": 16732,
+      "Ġremarkably": 16733,
+      "Sign": 16734,
+      "Ġshells": 16735,
+      "oglobulin": 16736,
+      "quilibrium": 16737,
+      "ĠMoore": 16738,
+      "ĠAdvers": 16739,
+      "ĠMycobacterium": 16740,
+      "Invitrogen": 16741,
+      "Ġthaliana": 16742,
+      "BY": 16743,
+      "ĠBit": 16744,
+      "Ġts": 16745,
+      "Ġsynchronous": 16746,
+      "yx": 16747,
+      "Ġpropagator": 16748,
+      "ĠIncreasing": 16749,
+      "iparum": 16750,
+      "Ġfreeze": 16751,
+      "ĠSelective": 16752,
+      "afe": 16753,
+      "Ġstrept": 16754,
+      "phantom": 16755,
+      "ĠGenerally": 16756,
+      "Ġalternate": 16757,
+      "ĠConvergence": 16758,
+      "////////////////": 16759,
+      "enging": 16760,
+      "ĠRandomized": 16761,
+      "develop": 16762,
+      "predict": 16763,
+      "ressor": 16764,
+      "Ġmathematics": 16765,
+      "fr": 16766,
+      "ĠComputation": 16767,
+      "ĠMalays": 16768,
+      "Ġbreathing": 16769,
+      "Through": 16770,
+      "ĠSIM": 16771,
+      "Ġanode": 16772,
+      "oad": 16773,
+      "ĠATCC": 16774,
+      "Ġconstituent": 16775,
+      "ĠMeasuring": 16776,
+      "ĠfMRI": 16777,
+      "Ġanemia": 16778,
+      "liest": 16779,
+      "Ġhemisphere": 16780,
+      "Ġmaxima": 16781,
+      "Ġtemporary": 16782,
+      "Ġdz": 16783,
+      "otoxin": 16784,
+      "Count": 16785,
+      "oned": 16786,
+      "ú": 16787,
+      "Ġcollaborative": 16788,
+      "Ġkb": 16789,
+      "Ġversa": 16790,
+      "ĠSwedish": 16791,
+      "ika": 16792,
+      "Ġdialysis": 16793,
+      "Ġperovsk": 16794,
+      "Ġwilling": 16795,
+      "ĠGreek": 16796,
+      "Output": 16797,
+      "Ġsemigroup": 16798,
+      "Ġbottlen": 16799,
+      "ĠGibbs": 16800,
+      "dark": 16801,
+      "Ġrheumatoid": 16802,
+      "urring": 16803,
+      "matched": 16804,
+      "Ġsophisticated": 16805,
+      "Ġcustomer": 16806,
+      "tetrahydro": 16807,
+      "XY": 16808,
+      "bug": 16809,
+      "Ġmorning": 16810,
+      "ĠCVD": 16811,
+      "Ġmappings": 16812,
+      "ĠMSCs": 16813,
+      "ĠDH": 16814,
+      "Ġquatern": 16815,
+      "health": 16816,
+      "ı": 16817,
+      "Ġtemp": 16818,
+      "ĠJew": 16819,
+      "ĠIl": 16820,
+      "Ġvortices": 16821,
+      "Ġserine": 16822,
+      "ĠOxygen": 16823,
+      "weg": 16824,
+      "Ġexplanations": 16825,
+      "PG": 16826,
+      "Ġciti": 16827,
+      "Ġlocality": 16828,
+      "===": 16829,
+      "ĠThom": 16830,
+      "Ġdairy": 16831,
+      "Block": 16832,
+      "ordial": 16833,
+      "akov": 16834,
+      "Ġglioma": 16835,
+      "Ġtransaction": 16836,
+      "Ġincremental": 16837,
+      "anche": 16838,
+      "Ret": 16839,
+      "magnetic": 16840,
+      "pyrrol": 16841,
+      "ĠPic": 16842,
+      "Ġamelior": 16843,
+      "oxidant": 16844,
+      "roviral": 16845,
+      "oratory": 16846,
+      "Ġsav": 16847,
+      "ĠStream": 16848,
+      "Ġsuperf": 16849,
+      "ĠICU": 16850,
+      "Ġevidenced": 16851,
+      "Ġrepeatedly": 16852,
+      "Ġrated": 16853,
+      "ĠPit": 16854,
+      "FAULT": 16855,
+      "Ġhat": 16856,
+      "ĠContent": 16857,
+      "Ġisoform": 16858,
+      "VER": 16859,
+      "Ġnodal": 16860,
+      "Ġscheduled": 16861,
+      "Ġshoulder": 16862,
+      "Ġtap": 16863,
+      "Ġportal": 16864,
+      "Ġtraps": 16865,
+      "aev": 16866,
+      "ĠSOD": 16867,
+      "ematic": 16868,
+      "Ġenj": 16869,
+      "Ġreticulum": 16870,
+      "ĠMinister": 16871,
+      "ĠSel": 16872,
+      "Ġfalling": 16873,
+      "rost": 16874,
+      "NG": 16875,
+      "fd": 16876,
+      "nitro": 16877,
+      "ĠMove": 16878,
+      "relativistic": 16879,
+      "enges": 16880,
+      "ĠSST": 16881,
+      "ĠInv": 16882,
+      "Ġfinish": 16883,
+      "ĠPoland": 16884,
+      "osecond": 16885,
+      "ĠBAL": 16886,
+      "oarthritis": 16887,
+      "Ġoptics": 16888,
+      "ĠSky": 16889,
+      "Ġadvoc": 16890,
+      "Ġhemorrhage": 16891,
+      "Ġmodulating": 16892,
+      "nis": 16893,
+      "Ġmachinery": 16894,
+      "Ġupdating": 16895,
+      "Ġcharacterizing": 16896,
+      "ishman": 16897,
+      "Ġtemplates": 16898,
+      "ĠLaplace": 16899,
+      "ĠEns": 16900,
+      "Recently": 16901,
+      "orus": 16902,
+      "arts": 16903,
+      "diffusion": 16904,
+      "ĠLevels": 16905,
+      "aga": 16906,
+      "ĠInj": 16907,
+      "ĠLayer": 16908,
+      "Ġremn": 16909,
+      "Ġelasticity": 16910,
+      "Ġmerely": 16911,
+      "Ġfission": 16912,
+      "engue": 16913,
+      "make": 16914,
+      "Ġmonop": 16915,
+      "Ġurea": 16916,
+      "ĠSimon": 16917,
+      "miR": 16918,
+      "ĠSecondly": 16919,
+      "uric": 16920,
+      "ĠVariable": 16921,
+      "ilis": 16922,
+      "Ġmultiplicative": 16923,
+      "ĠNoise": 16924,
+      "Ġswitched": 16925,
+      "Ġnicot": 16926,
+      "Ġefficiencies": 16927,
+      "hema": 16928,
+      "Ġappointed": 16929,
+      "guided": 16930,
+      "Ġwinning": 16931,
+      "ĠMechanics": 16932,
+      "Ġneo": 16933,
+      "ĠBRCA": 16934,
+      "udi": 16935,
+      "Ġcontainer": 16936,
+      "shop": 16937,
+      "Ġsuggestions": 16938,
+      "KB": 16939,
+      "Ġsubstitute": 16940,
+      "Ox": 16941,
+      "VC": 16942,
+      "Ġstone": 16943,
+      "anna": 16944,
+      "ĠDepression": 16945,
+      "Ġcontemporary": 16946,
+      "Ġoutliers": 16947,
+      "quet": 16948,
+      "ĠZheng": 16949,
+      "Ġoccl": 16950,
+      "Ġalveolar": 16951,
+      "expressing": 16952,
+      "Ġcomfort": 16953,
+      "Ġignore": 16954,
+      "Among": 16955,
+      "ĠKlein": 16956,
+      "Ġrhythm": 16957,
+      "Ġimmers": 16958,
+      "Ġfaith": 16959,
+      "bling": 16960,
+      "Ġaugmentation": 16961,
+      "ĠPrevention": 16962,
+      "Ġhepar": 16963,
+      "Ġnotations": 16964,
+      "Ġhematopoietic": 16965,
+      "perfect": 16966,
+      "Ġshares": 16967,
+      "notin": 16968,
+      "Ġpictures": 16969,
+      "ĠAcknowledgments": 16970,
+      "Ġtick": 16971,
+      "Ġunrelated": 16972,
+      "ĠTool": 16973,
+      "Ġmas": 16974,
+      "osocial": 16975,
+      "gest": 16976,
+      "ushed": 16977,
+      "Ġphosphorylated": 16978,
+      "Ġceramic": 16979,
+      "cool": 16980,
+      "orylation": 16981,
+      "Ġdeficient": 16982,
+      "Ġrelaxed": 16983,
+      "ĠAnalyses": 16984,
+      "ecraft": 16985,
+      "Ġretina": 16986,
+      "ĠInternal": 16987,
+      "Ġspite": 16988,
+      "Ġrecipients": 16989,
+      "Ġshut": 16990,
+      "Ġethylene": 16991,
+      "ĠGulf": 16992,
+      "Ġunaffected": 16993,
+      "ĠResource": 16994,
+      "ĠNet": 16995,
+      "Ġperpet": 16996,
+      "Ġslab": 16997,
+      "report": 16998,
+      "Ġμmol": 16999,
+      "Ġidx": 17000,
+      "Ġskill": 17001,
+      "ĠInduction": 17002,
+      "Ġmalignancy": 17003,
+      "Ġcv": 17004,
+      "Ġdiffering": 17005,
+      "Ġappropriately": 17006,
+      "ijing": 17007,
+      "Ġwarrant": 17008,
+      "rally": 17009,
+      "Ġalgae": 17010,
+      "weights": 17011,
+      "casts": 17012,
+      "Ġocular": 17013,
+      "racycl": 17014,
+      "Ġdominates": 17015,
+      "Ġleuc": 17016,
+      "Where": 17017,
+      "phon": 17018,
+      "Ġsocioeconomic": 17019,
+      "itzerland": 17020,
+      "Ġresilience": 17021,
+      "Ġneighbourhood": 17022,
+      "Ġtone": 17023,
+      "psych": 17024,
+      "ĠOrganic": 17025,
+      "Ġgather": 17026,
+      "Ġfalciparum": 17027,
+      "Ġengineered": 17028,
+      "ĠAvail": 17029,
+      "intering": 17030,
+      "Ġclimatic": 17031,
+      "ĠEvolutionary": 17032,
+      "NMR": 17033,
+      "Ġrev": 17034,
+      "central": 17035,
+      "ĠSin": 17036,
+      "Ġdeclined": 17037,
+      "opausal": 17038,
+      "Ġalarm": 17039,
+      "Rightarrow": 17040,
+      "sex": 17041,
+      "Ġenergetic": 17042,
+      "ïĤ": 17043,
+      "Ġdiscs": 17044,
+      "Ġolfactory": 17045,
+      "uripot": 17046,
+      "spectrum": 17047,
+      "spot": 17048,
+      "Ġhemoglobin": 17049,
+      "Mark": 17050,
+      "cov": 17051,
+      "arboxyl": 17052,
+      "Ġindications": 17053,
+      "Ġsalmon": 17054,
+      "Ġsearched": 17055,
+      "Ġended": 17056,
+      "rologic": 17057,
+      "rfloor": 17058,
+      "Ġautism": 17059,
+      "Ġselen": 17060,
+      "ĠHung": 17061,
+      "ĠInference": 17062,
+      "Ġmammary": 17063,
+      "lfloor": 17064,
+      "Ġseroton": 17065,
+      "Ġfunded": 17066,
+      "ĠViet": 17067,
+      "Ġrivers": 17068,
+      "ĠReinfor": 17069,
+      "urg": 17070,
+      "Ġalbicans": 17071,
+      "ĠThermo": 17072,
+      "ERROR": 17073,
+      "Ġmutually": 17074,
+      "Ġirr": 17075,
+      "ĠRat": 17076,
+      "Ġimg": 17077,
+      "Ġlymphocyte": 17078,
+      "ĠRefs": 17079,
+      "ĠSparse": 17080,
+      "holders": 17081,
+      "Free": 17082,
+      "RED": 17083,
+      "ĠGauss": 17084,
+      "Ġcircadian": 17085,
+      "ĠJin": 17086,
+      "Ġconstitutes": 17087,
+      "Ġwors": 17088,
+      "Ġfeatured": 17089,
+      "ocent": 17090,
+      "lete": 17091,
+      "Ġontology": 17092,
+      "Ġbilayer": 17093,
+      "ĠCambridge": 17094,
+      "Ġencryption": 17095,
+      "rotron": 17096,
+      "etti": 17097,
+      "ĠAer": 17098,
+      "Ġcouples": 17099,
+      "rail": 17100,
+      "Ġtwist": 17101,
+      "Ġridge": 17102,
+      "GAN": 17103,
+      "iders": 17104,
+      "SHIFT": 17105,
+      "Ġdiffus": 17106,
+      "Ġmeant": 17107,
+      "ĠSchwarz": 17108,
+      "Sb": 17109,
+      "Ġarcs": 17110,
+      "Notice": 17111,
+      "iy": 17112,
+      "Ġemerge": 17113,
+      "kwargs": 17114,
+      "Eff": 17115,
+      "Ent": 17116,
+      "ionization": 17117,
+      "choline": 17118,
+      "ustries": 17119,
+      "acher": 17120,
+      "spl": 17121,
+      "population": 17122,
+      "fol": 17123,
+      "Ġquestionnaires": 17124,
+      "Ġallergic": 17125,
+      "wich": 17126,
+      "ĠVacc": 17127,
+      "Ġattained": 17128,
+      "ĠAnimals": 17129,
+      "amics": 17130,
+      "ĠRegarding": 17131,
+      "ĠSemi": 17132,
+      "Ġglac": 17133,
+      "ĠEfficacy": 17134,
+      "Ġsynergistic": 17135,
+      "ISH": 17136,
+      "Ġmaintains": 17137,
+      "Ġsongs": 17138,
+      "ĠNegative": 17139,
+      "amoto": 17140,
+      "ĠModified": 17141,
+      "Ġseparable": 17142,
+      "Ġbinaries": 17143,
+      "Ġaccessibility": 17144,
+      "Iter": 17145,
+      "din": 17146,
+      "ĠBinary": 17147,
+      "equilibrium": 17148,
+      "Ġcue": 17149,
+      "magn": 17150,
+      "Ġedema": 17151,
+      "�": 17152,
+      "Ġpositioned": 17153,
+      "Ġcharging": 17154,
+      "Ġunivariate": 17155,
+      "hep": 17156,
+      "Ġclade": 17157,
+      "Ġcysteine": 17158,
+      "racle": 17159,
+      "Ġrescue": 17160,
+      "habit": 17161,
+      "ĠDISCUSSION": 17162,
+      "Ġdepicts": 17163,
+      "pole": 17164,
+      "Ġstenosis": 17165,
+      "Ġveter": 17166,
+      "pringer": 17167,
+      "ĠPow": 17168,
+      "Ġcovariant": 17169,
+      "Ġmodifying": 17170,
+      "Algorithm": 17171,
+      "averaged": 17172,
+      "alo": 17173,
+      "reson": 17174,
+      "Ġcharacterised": 17175,
+      "Ġni": 17176,
+      "Ġseemed": 17177,
+      "ĠRom": 17178,
+      "short": 17179,
+      "NV": 17180,
+      "Ġfertility": 17181,
+      "ĠMemb": 17182,
+      "Ġlying": 17183,
+      "Ġinstitution": 17184,
+      "images": 17185,
+      "ĠBorel": 17186,
+      "fsys": 17187,
+      "cataly": 17188,
+      "Ġseparating": 17189,
+      "biotic": 17190,
+      "mel": 17191,
+      "pgfsys": 17192,
+      "ĠJackson": 17193,
+      "Ġbag": 17194,
+      "ograp": 17195,
+      "propyl": 17196,
+      "ĠProgramming": 17197,
+      "ocratic": 17198,
+      "Ġpion": 17199,
+      "ĠGradient": 17200,
+      "Ġsphe": 17201,
+      "Ġinline": 17202,
+      "Ġdominate": 17203,
+      "Ġsuffered": 17204,
+      "ĠDiseases": 17205,
+      "igenous": 17206,
+      "will": 17207,
+      "Ġamin": 17208,
+      "adherin": 17209,
+      "ĠTro": 17210,
+      "adjusted": 17211,
+      "EW": 17212,
+      "Ġdebut": 17213,
+      "nea": 17214,
+      "ĠDun": 17215,
+      "Ġdictionary": 17216,
+      "operatively": 17217,
+      "KA": 17218,
+      "beit": 17219,
+      "Ġpersonnel": 17220,
+      "ĠŽ": 17221,
+      "review": 17222,
+      "into": 17223,
+      "ĠTokyo": 17224,
+      "Ġtrop": 17225,
+      "Ġventric": 17226,
+      "ĠMETHODS": 17227,
+      "Ġimplication": 17228,
+      "akis": 17229,
+      "ĠCMB": 17230,
+      "Ġtransmitter": 17231,
+      "oichi": 17232,
+      "ĠNigeria": 17233,
+      "ĠKon": 17234,
+      "Ġbear": 17235,
+      "ĠKan": 17236,
+      "ĠPlot": 17237,
+      "ĠSPSS": 17238,
+      "ĠBiology": 17239,
+      "Ġbaryon": 17240,
+      "ĠmicroRNA": 17241,
+      "Ġreproducibility": 17242,
+      "Ġlactate": 17243,
+      "Ġpolyphen": 17244,
+      "ĠMt": 17245,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 17246,
+      "endit": 17247,
+      "Ġhydrothermal": 17248,
+      "Ġwealth": 17249,
+      "Ġhadron": 17250,
+      "Ġwhereby": 17251,
+      "ellum": 17252,
+      "ĠDiffusion": 17253,
+      "ĠOrigin": 17254,
+      "Ġnonlinearity": 17255,
+      "Ġinformative": 17256,
+      "Ġvisited": 17257,
+      "Ġvirtually": 17258,
+      "ĠTun": 17259,
+      "Ġreset": 17260,
+      "ĠElectrical": 17261,
+      "ĠGlu": 17262,
+      "ĠSAM": 17263,
+      "ĠIsing": 17264,
+      "ĠStra": 17265,
+      "onder": 17266,
+      "Ġdies": 17267,
+      "Ġreciprocal": 17268,
+      "Check": 17269,
+      "ĠGuidelines": 17270,
+      "hester": 17271,
+      "Ġproblematic": 17272,
+      "ĠAtomic": 17273,
+      "Ġconcentrate": 17274,
+      "steps": 17275,
+      "json": 17276,
+      "Recommended": 17277,
+      "ĠScreening": 17278,
+      "Ġnaive": 17279,
+      "Ġpractitioners": 17280,
+      "Ġfasting": 17281,
+      "Ġmechanistic": 17282,
+      "options": 17283,
+      "Ptr": 17284,
+      "ITE": 17285,
+      "Work": 17286,
+      "âĢĺ": 17287,
+      "rafts": 17288,
+      "Ġunw": 17289,
+      "Ġannihilation": 17290,
+      "objective": 17291,
+      "ĠDynamical": 17292,
+      "adec": 17293,
+      "ĠLith": 17294,
+      "Ġextracting": 17295,
+      "Ġcoral": 17296,
+      "ĠStable": 17297,
+      "Ġbackgrounds": 17298,
+      "omorphisms": 17299,
+      "ĠâĪ«": 17300,
+      "Ġgrew": 17301,
+      "Inst": 17302,
+      "gels": 17303,
+      "Ġinhal": 17304,
+      "dam": 17305,
+      "heim": 17306,
+      "benzyl": 17307,
+      "Ġpelvic": 17308,
+      "Ġdiarr": 17309,
+      "Ġdiode": 17310,
+      "Ġempir": 17311,
+      "ĠAlf": 17312,
+      "ĠUncertain": 17313,
+      "ĠHCl": 17314,
+      "Ġjointly": 17315,
+      "Ġdepar": 17316,
+      "Ġmerging": 17317,
+      "Ġchi": 17318,
+      "apt": 17319,
+      "Ġplt": 17320,
+      "Ġidi": 17321,
+      "Ġperfor": 17322,
+      "stituting": 17323,
+      "page": 17324,
+      "aré": 17325,
+      "indices": 17326,
+      "putation": 17327,
+      "different": 17328,
+      "burn": 17329,
+      "Ġsurrounded": 17330,
+      "ĠTL": 17331,
+      "untary": 17332,
+      "strip": 17333,
+      "lan": 17334,
+      "Ġcow": 17335,
+      "ĠSab": 17336,
+      "ĠGaAs": 17337,
+      "pf": 17338,
+      "Ġesophageal": 17339,
+      "ĠAlt": 17340,
+      "Ġhospitalization": 17341,
+      "ĠApproximation": 17342,
+      "Organism": 17343,
+      "ĠFair": 17344,
+      "Ġtracing": 17345,
+      "Ġpreferentially": 17346,
+      "Ġlowering": 17347,
+      "uliar": 17348,
+      "ĠDeriv": 17349,
+      "Ġphytoplankton": 17350,
+      "omyc": 17351,
+      "That": 17352,
+      "ĠIsrael": 17353,
+      "Ġminimized": 17354,
+      "Ġanything": 17355,
+      "rule": 17356,
+      "pow": 17357,
+      "Ġfamous": 17358,
+      "ĠAccuracy": 17359,
+      "Ġphotocatalytic": 17360,
+      "ĠNonetheless": 17361,
+      "Ġdivisor": 17362,
+      "vb": 17363,
+      "Ġcameras": 17364,
+      "ĠWales": 17365,
+      "ĠContributions": 17366,
+      "Ġdisplacements": 17367,
+      "ĠTam": 17368,
+      "Ġvolumetric": 17369,
+      "essional": 17370,
+      "Ġcompensate": 17371,
+      "Ġace": 17372,
+      "triangle": 17373,
+      "buff": 17374,
+      "Ġnamespace": 17375,
+      "Ġbounding": 17376,
+      "ynchronous": 17377,
+      "md": 17378,
+      "Ġimagery": 17379,
+      "itated": 17380,
+      "Ġoriginated": 17381,
+      "ĠBelg": 17382,
+      "ĠECG": 17383,
+      "existing": 17384,
+      "ĠStokes": 17385,
+      "sensitivity": 17386,
+      "tidine": 17387,
+      "ĠWM": 17388,
+      "Ġmonotone": 17389,
+      "Ġproceeds": 17390,
+      "ĠClustering": 17391,
+      "ĠIoT": 17392,
+      "ernary": 17393,
+      "alamic": 17394,
+      "ĠCollaboration": 17395,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 17396,
+      "OLD": 17397,
+      "Îĺ": 17398,
+      "ĠNanopar": 17399,
+      "ĠMultiv": 17400,
+      "Ġcystic": 17401,
+      "pire": 17402,
+      "Ġoperates": 17403,
+      "Ġmediating": 17404,
+      "Ġbeneath": 17405,
+      "obe": 17406,
+      "gate": 17407,
+      "Ġoocytes": 17408,
+      "Ġmargins": 17409,
+      "ymmetries": 17410,
+      "Ġreligious": 17411,
+      "ĠNit": 17412,
+      "Ġcutaneous": 17413,
+      "ANS": 17414,
+      "Ġdevelops": 17415,
+      "asia": 17416,
+      "ĠRoberts": 17417,
+      "avier": 17418,
+      "Ġsimplic": 17419,
+      "Ġrevealing": 17420,
+      "UND": 17421,
+      "Ġtea": 17422,
+      "Ġlysis": 17423,
+      "Ġaggregated": 17424,
+      "ĠRGB": 17425,
+      "Ġcorro": 17426,
+      "Ġbir": 17427,
+      "inae": 17428,
+      "vd": 17429,
+      "Ġcourt": 17430,
+      "Ġcontroversial": 17431,
+      "Ġtow": 17432,
+      "Ġhysteresis": 17433,
+      "enberg": 17434,
+      "Ġenters": 17435,
+      "png": 17436,
+      "ĠFlex": 17437,
+      "Assume": 17438,
+      "ĠBad": 17439,
+      "ĠSimilarities": 17440,
+      "Experim": 17441,
+      "ATH": 17442,
+      "Ġut": 17443,
+      "terms": 17444,
+      "ĠMol": 17445,
+      "Ġvisually": 17446,
+      "Ġadoption": 17447,
+      "Ġprinting": 17448,
+      "Ġequiv": 17449,
+      "ĠPert": 17450,
+      "Ġpercol": 17451,
+      "Ġsomeone": 17452,
+      "abulary": 17453,
+      "Ġlever": 17454,
+      "ĠHaus": 17455,
+      "icillin": 17456,
+      "itar": 17457,
+      "Ġtourn": 17458,
+      "Altern": 17459,
+      "Exp": 17460,
+      "~~~~": 17461,
+      "ĠFo": 17462,
+      "Ġabol": 17463,
+      "median": 17464,
+      "Ġrolling": 17465,
+      "hm": 17466,
+      "Ġtelescope": 17467,
+      "ĠCav": 17468,
+      "Ġseedlings": 17469,
+      "inhib": 17470,
+      "Ġdin": 17471,
+      "Ġimpurities": 17472,
+      "Ġamplifier": 17473,
+      "ĠKer": 17474,
+      "Ġdiminished": 17475,
+      "PB": 17476,
+      "fib": 17477,
+      "rock": 17478,
+      "ĠBin": 17479,
+      "Ġphotosynthetic": 17480,
+      "ĠCrypt": 17481,
+      "Ġpreterm": 17482,
+      "Ġhits": 17483,
+      "Ġfractal": 17484,
+      "Ġdiscarded": 17485,
+      "Ġendocrine": 17486,
+      "oshi": 17487,
+      "Ġmodulo": 17488,
+      "wt": 17489,
+      "Ġquenching": 17490,
+      "Ġsounds": 17491,
+      "ĠEDTA": 17492,
+      "reactive": 17493,
+      "Ġresist": 17494,
+      "anghai": 17495,
+      "Ġnarr": 17496,
+      "Ġinitiate": 17497,
+      "ĠSaint": 17498,
+      "XR": 17499,
+      "GeV": 17500,
+      "ĠIndependent": 17501,
+      "Ġinjective": 17502,
+      "upus": 17503,
+      "Ġlinguistic": 17504,
+      "Ġanalogues": 17505,
+      "Ġdissection": 17506,
+      "Ġlasers": 17507,
+      "diab": 17508,
+      "ĠTele": 17509,
+      "Ġcracks": 17510,
+      "Ġbrane": 17511,
+      "VO": 17512,
+      "ĠExtended": 17513,
+      "Ġtells": 17514,
+      "Ġremarks": 17515,
+      "ulting": 17516,
+      "ĠBurn": 17517,
+      "dL": 17518,
+      "ressible": 17519,
+      "ĠChap": 17520,
+      "Ġsq": 17521,
+      "Ġreproduced": 17522,
+      "ĠBcl": 17523,
+      "Ġswarm": 17524,
+      "opathology": 17525,
+      "chrotron": 17526,
+      "Ġmine": 17527,
+      "Ġhadronic": 17528,
+      "ĠLocalization": 17529,
+      "ĠMotor": 17530,
+      "Ġvisualize": 17531,
+      "Ġcats": 17532,
+      "Ġbalancing": 17533,
+      "ĠSched": 17534,
+      "CoA": 17535,
+      "Ġthermodynamics": 17536,
+      "ĠDiagnostic": 17537,
+      "Ġrelief": 17538,
+      "Ġpositivity": 17539,
+      "Ġhub": 17540,
+      "ĠInfrared": 17541,
+      "Sur": 17542,
+      "omed": 17543,
+      "Ġoptically": 17544,
+      "Ġvascul": 17545,
+      "isations": 17546,
+      "encoder": 17547,
+      "Ġcopolymer": 17548,
+      "Ġrestore": 17549,
+      "Ġinertia": 17550,
+      "ubicin": 17551,
+      "Ġetiology": 17552,
+      "ĠSecret": 17553,
+      "ĠCW": 17554,
+      "Const": 17555,
+      "ĠBrit": 17556,
+      "ĠConstant": 17557,
+      "ĠDIS": 17558,
+      "Ġdiscipl": 17559,
+      "bra": 17560,
+      "ĠOral": 17561,
+      "ĠUL": 17562,
+      "Ġdeline": 17563,
+      "Ġnucleon": 17564,
+      "Ġemployment": 17565,
+      "ĠRD": 17566,
+      "qq": 17567,
+      "ĠCarolina": 17568,
+      "ĠGab": 17569,
+      "Ġassertion": 17570,
+      "CMC": 17571,
+      "rgb": 17572,
+      "Frame": 17573,
+      "ĠJust": 17574,
+      "Ġinoculation": 17575,
+      "cluding": 17576,
+      "Ġoscillatory": 17577,
+      "Ġcancel": 17578,
+      "ĠPoinc": 17579,
+      "pora": 17580,
+      "ĠJul": 17581,
+      "ruvate": 17582,
+      "Ġpolitic": 17583,
+      "urus": 17584,
+      "ĠAdvances": 17585,
+      "ĠRoot": 17586,
+      "thood": 17587,
+      "oxygenase": 17588,
+      "msg": 17589,
+      "ĠkV": 17590,
+      "Ġadmit": 17591,
+      "Ġrefractory": 17592,
+      "Ġcloning": 17593,
+      "Ġfatal": 17594,
+      "plantation": 17595,
+      "ĠGir": 17596,
+      "Ġtes": 17597,
+      "ĠRho": 17598,
+      "ohn": 17599,
+      "Ġinnovation": 17600,
+      "Ġsending": 17601,
+      "Ġcable": 17602,
+      "Ġniche": 17603,
+      "Ġreserve": 17604,
+      "Ġatrophy": 17605,
+      "athan": 17606,
+      "ĠÃij": 17607,
+      "itization": 17608,
+      "Ġfan": 17609,
+      "Ġbubbles": 17610,
+      "ĠTheorems": 17611,
+      "ĠSwitzerland": 17612,
+      "ĠHeisenberg": 17613,
+      "ĠReduced": 17614,
+      "Ra": 17615,
+      "Zr": 17616,
+      "ĠPossible": 17617,
+      "Upsilon": 17618,
+      "ĠAgric": 17619,
+      "ellect": 17620,
+      "nds": 17621,
+      "mathds": 17622,
+      "atre": 17623,
+      "Ġforaging": 17624,
+      "Ġupward": 17625,
+      "idene": 17626,
+      "Ġglands": 17627,
+      "fed": 17628,
+      "uccessful": 17629,
+      "ĠWolf": 17630,
+      "Ġusefulness": 17631,
+      "oporous": 17632,
+      "Ġpunct": 17633,
+      "ardo": 17634,
+      "Ġsystolic": 17635,
+      "ĠTargeting": 17636,
+      "Ġillumin": 17637,
+      "Ġpigment": 17638,
+      "Ġsimulating": 17639,
+      "Ġportions": 17640,
+      "ĠPrinciples": 17641,
+      "ĠHopf": 17642,
+      "lipid": 17643,
+      "ĠLU": 17644,
+      "ubation": 17645,
+      "ĠArtificial": 17646,
+      "Ġprison": 17647,
+      "aning": 17648,
+      "ĠGN": 17649,
+      "ĠStrategies": 17650,
+      "ĠPas": 17651,
+      "Ta": 17652,
+      "ĠProbability": 17653,
+      "orum": 17654,
+      "Ġskeleton": 17655,
+      "Ġcompartments": 17656,
+      "Read": 17657,
+      "Ġcoach": 17658,
+      "Ġmodality": 17659,
+      "ĠRegister": 17660,
+      "Ġje": 17661,
+      "Ġheights": 17662,
+      "inyl": 17663,
+      "Ġsubspaces": 17664,
+      "tip": 17665,
+      "Ġá¸": 17666,
+      "ĠGI": 17667,
+      "Char": 17668,
+      "rogenic": 17669,
+      "rett": 17670,
+      "eutics": 17671,
+      "Ġadhesive": 17672,
+      "ĠPier": 17673,
+      "Left": 17674,
+      "idental": 17675,
+      "NAc": 17676,
+      "Ġconjugation": 17677,
+      "orov": 17678,
+      "idge": 17679,
+      "imaging": 17680,
+      "ĠTW": 17681,
+      "Ġpresident": 17682,
+      "ĠOste": 17683,
+      "assemb": 17684,
+      "Ġinternet": 17685,
+      "Ġdeals": 17686,
+      "ĠGAP": 17687,
+      "Ġformulate": 17688,
+      "ĠUpdate": 17689,
+      "ĠRNAi": 17690,
+      "clero": 17691,
+      "Ġpermutations": 17692,
+      "Ġisotopes": 17693,
+      "opic": 17694,
+      "ĠQU": 17695,
+      "romes": 17696,
+      "ĠPolicy": 17697,
+      "ĠCreek": 17698,
+      "ĠWindows": 17699,
+      "Ġmerge": 17700,
+      "Ġaccident": 17701,
+      "Ġsuperposition": 17702,
+      "Ġdebate": 17703,
+      "Ġdocumentation": 17704,
+      "Ġeigenvectors": 17705,
+      "sor": 17706,
+      "ĠPhoto": 17707,
+      "Ġdeposit": 17708,
+      "Ġgermination": 17709,
+      "Ġsubgraph": 17710,
+      "ĠRecords": 17711,
+      "Ġchemically": 17712,
+      "ĠPredicting": 17713,
+      "ĠKy": 17714,
+      "selective": 17715,
+      "ynman": 17716,
+      "dispers": 17717,
+      "Ġlumbar": 17718,
+      "Ġmusical": 17719,
+      "inates": 17720,
+      "Ġinherited": 17721,
+      "ju": 17722,
+      "Ġtracer": 17723,
+      "Ġending": 17724,
+      "Ġengaged": 17725,
+      "handed": 17726,
+      "Ġproducer": 17727,
+      "Ġentangled": 17728,
+      "ĠDelta": 17729,
+      "Ġpiecewise": 17730,
+      "NAME": 17731,
+      "stop": 17732,
+      "Ġmutated": 17733,
+      "Ġrecess": 17734,
+      "Ġimmuno": 17735,
+      "cancer": 17736,
+      "ĠAkt": 17737,
+      "iters": 17738,
+      "ĠBMP": 17739,
+      "Ġcompanion": 17740,
+      "Ġcommunicate": 17741,
+      "Ġhollow": 17742,
+      "Ġpad": 17743,
+      "Ġsph": 17744,
+      "omod": 17745,
+      "Ġparton": 17746,
+      "Ġspontaneously": 17747,
+      "eared": 17748,
+      "Ġrotations": 17749,
+      "Ġcosmology": 17750,
+      "Ġmoreover": 17751,
+      "princ": 17752,
+      "Ġeverywhere": 17753,
+      "brane": 17754,
+      "lational": 17755,
+      "eme": 17756,
+      "Ġbehave": 17757,
+      "umen": 17758,
+      "oston": 17759,
+      "oves": 17760,
+      "Ġgar": 17761,
+      "Ġadrenal": 17762,
+      "ĠEstimating": 17763,
+      "Nb": 17764,
+      "Ġechocardi": 17765,
+      "Ġemphasized": 17766,
+      "Ġengines": 17767,
+      "Ġbrackets": 17768,
+      "Ġleaders": 17769,
+      "Ġdistinctive": 17770,
+      "ĠLymph": 17771,
+      "Ġexert": 17772,
+      "Ġinnovative": 17773,
+      "coupling": 17774,
+      "ĠSignific": 17775,
+      "sheet": 17776,
+      "ĠCover": 17777,
+      "ĠCCD": 17778,
+      "ĠFall": 17779,
+      "stimulated": 17780,
+      "Ġsuperoxide": 17781,
+      "Ġpollutants": 17782,
+      "Ġbytes": 17783,
+      "ĠLipid": 17784,
+      "Ġtrafficking": 17785,
+      "Ġleadership": 17786,
+      "informatics": 17787,
+      "Ġbiodiversity": 17788,
+      "ador": 17789,
+      "Ġinterconn": 17790,
+      "Ġharmonics": 17791,
+      "Ġseawater": 17792,
+      "ĠIllumina": 17793,
+      "necessary": 17794,
+      "ĠAnton": 17795,
+      "Ġprocessors": 17796,
+      "typename": 17797,
+      "Det": 17798,
+      "proton": 17799,
+      "Ġsubtraction": 17800,
+      "Ġshifting": 17801,
+      "Ġcustomers": 17802,
+      "Ke": 17803,
+      "ĠOB": 17804,
+      "atonin": 17805,
+      "atellite": 17806,
+      "ĠSUS": 17807,
+      "ĠColon": 17808,
+      "ĠTimes": 17809,
+      "TV": 17810,
+      "ĠMink": 17811,
+      "ĠIntegration": 17812,
+      "Ġprofound": 17813,
+      "ITC": 17814,
+      "Ġgras": 17815,
+      "ĠNASA": 17816,
+      "ĠACK": 17817,
+      "radiol": 17818,
+      "ĠMale": 17819,
+      "ĠWorking": 17820,
+      "ticity": 17821,
+      "ilibria": 17822,
+      "boundary": 17823,
+      "ĠRI": 17824,
+      "ĠAli": 17825,
+      "cardi": 17826,
+      "ĠFGF": 17827,
+      "branes": 17828,
+      "Ġbeet": 17829,
+      "Ġmissed": 17830,
+      "Source": 17831,
+      "ĠBot": 17832,
+      "ieve": 17833,
+      "Ġisother": 17834,
+      "neys": 17835,
+      "nl": 17836,
+      "ortion": 17837,
+      "Ġcooled": 17838,
+      "MV": 17839,
+      "Ġomit": 17840,
+      "Ġverbal": 17841,
+      "arette": 17842,
+      "Ġconference": 17843,
+      "Ġtransformer": 17844,
+      "Ġrejected": 17845,
+      "Ġprogressively": 17846,
+      "ĠTurkey": 17847,
+      "Ġathletes": 17848,
+      "Ġanatomy": 17849,
+      "EQ": 17850,
+      "Ġdeterioration": 17851,
+      "ĠDietary": 17852,
+      "Ġcorn": 17853,
+      "Ġcapsule": 17854,
+      "Ġvibrations": 17855,
+      "Ġoccupational": 17856,
+      "Ġexosomes": 17857,
+      "Ġrewritten": 17858,
+      "Ġlignin": 17859,
+      "Ġbiopsies": 17860,
+      "ĠAdversarial": 17861,
+      "Ġmercury": 17862,
+      "Ġplatinum": 17863,
+      "Ġirrelevant": 17864,
+      "Ġkeratin": 17865,
+      "ĠEmission": 17866,
+      "Ġeukaryotic": 17867,
+      "Ġinteg": 17868,
+      "Ġknot": 17869,
+      "Ġsera": 17870,
+      "Ġcavities": 17871,
+      "ĠMedi": 17872,
+      "Indeed": 17873,
+      "Eu": 17874,
+      "ĠâŁ": 17875,
+      "Ġscenes": 17876,
+      "Ġlaparoscopic": 17877,
+      "Ġsenior": 17878,
+      "ĠDistance": 17879,
+      "predic": 17880,
+      "Ġearliest": 17881,
+      "Ġorg": 17882,
+      "ĠThor": 17883,
+      "bury": 17884,
+      "oblasts": 17885,
+      "Ġpumping": 17886,
+      "targeted": 17887,
+      "Ġrap": 17888,
+      "ĠPil": 17889,
+      "Îł": 17890,
+      "Ġneurom": 17891,
+      "oft": 17892,
+      "ostat": 17893,
+      "Ġpadding": 17894,
+      "Ġconflicts": 17895,
+      "Ġstems": 17896,
+      "ĠSaccharomyces": 17897,
+      "engine": 17898,
+      "Ġalkyl": 17899,
+      "Ġtill": 17900,
+      "ĠQuad": 17901,
+      "good": 17902,
+      "rox": 17903,
+      "ĠFuzzy": 17904,
+      "Ġrobotic": 17905,
+      "ĠDenote": 17906,
+      "ĠNIR": 17907,
+      "ĠYuk": 17908,
+      "parency": 17909,
+      "Ġlegs": 17910,
+      "ylvan": 17911,
+      "Ġtightly": 17912,
+      "Ġdecor": 17913,
+      "ĠVP": 17914,
+      "ĠMun": 17915,
+      "atoms": 17916,
+      "ĠSilver": 17917,
+      "Ġneurodegenerative": 17918,
+      "Ġresponded": 17919,
+      "Ġrecons": 17920,
+      "GEN": 17921,
+      "ĠFine": 17922,
+      "fc": 17923,
+      "Ġparagraph": 17924,
+      "Ġintens": 17925,
+      "Ġalongside": 17926,
+      "Ġbrand": 17927,
+      "monium": 17928,
+      "Ġpm": 17929,
+      "Ġsimplex": 17930,
+      "ĠPreliminary": 17931,
+      "Ġdownregulation": 17932,
+      "Ġxy": 17933,
+      "ĠMak": 17934,
+      "opter": 17935,
+      "ushing": 17936,
+      "ĠBog": 17937,
+      "oxia": 17938,
+      "================================": 17939,
+      "common": 17940,
+      "ĠASS": 17941,
+      "ĠHDL": 17942,
+      "alamus": 17943,
+      "Ġirrigation": 17944,
+      "NM": 17945,
+      "Ġfading": 17946,
+      "Ġpreventive": 17947,
+      "Ġreliably": 17948,
+      "ĠEthiopia": 17949,
+      "othesis": 17950,
+      "izability": 17951,
+      "OB": 17952,
+      "Ġtriglycer": 17953,
+      "Ġgestational": 17954,
+      "Ġbesides": 17955,
+      "ĠIii": 17956,
+      "ĠZone": 17957,
+      "Ġcoping": 17958,
+      "Ġminority": 17959,
+      "Ġdeprivation": 17960,
+      "Ġhexagonal": 17961,
+      "chlorophenyl": 17962,
+      "ĠóµĦ¨": 17963,
+      "Ġgyr": 17964,
+      "Ġviewing": 17965,
+      "Newton": 17966,
+      "ĠHierarchical": 17967,
+      "oL": 17968,
+      "eces": 17969,
+      "Ġconcludes": 17970,
+      "Ġfungus": 17971,
+      "Ġpylori": 17972,
+      "Ġobstacles": 17973,
+      "thiazol": 17974,
+      "conjugated": 17975,
+      "rass": 17976,
+      "Ġlose": 17977,
+      "Ġforth": 17978,
+      "ĠAllen": 17979,
+      "oplast": 17980,
+      "ĠProtection": 17981,
+      "Ġintermittent": 17982,
+      "Ġluciferase": 17983,
+      "ĠMK": 17984,
+      "Ġgaug": 17985,
+      "ĠFan": 17986,
+      "Ġmodal": 17987,
+      "ĠExercise": 17988,
+      "scattering": 17989,
+      "ĠShim": 17990,
+      "Ġexcretion": 17991,
+      "Ġatypical": 17992,
+      "Ġmalignancies": 17993,
+      "anglades": 17994,
+      "ĠSpectroscopy": 17995,
+      "Ġadenosine": 17996,
+      "lif": 17997,
+      "Ġnucleic": 17998,
+      "Ġinclination": 17999,
+      "ĠCass": 18000,
+      "Ġethn": 18001,
+      "Ġexempl": 18002,
+      "ĠDy": 18003,
+      "Ġlambda": 18004,
+      "Ġjac": 18005,
+      "ĠPRE": 18006,
+      "Ġrailway": 18007,
+      "Ġfle": 18008,
+      "Ġreflections": 18009,
+      "Ġnanostructures": 18010,
+      "tists": 18011,
+      "prints": 18012,
+      "ĠCAT": 18013,
+      "Ġsib": 18014,
+      "Ġchloro": 18015,
+      "Ġrecipient": 18016,
+      "optic": 18017,
+      "Ġcounty": 18018,
+      "Ġnucleotides": 18019,
+      "Ġzircon": 18020,
+      "Ġhorses": 18021,
+      "ĠMental": 18022,
+      "inline": 18023,
+      "ĠNorway": 18024,
+      "They": 18025,
+      "Ġmuscular": 18026,
+      "acetic": 18027,
+      "ĠJu": 18028,
+      "Ġcommunic": 18029,
+      "files": 18030,
+      "filled": 18031,
+      "HB": 18032,
+      "Ġregulations": 18033,
+      "Ġaccumulate": 18034,
+      "ĠPanel": 18035,
+      "Cy": 18036,
+      "öl": 18037,
+      "ĠPakistan": 18038,
+      "Ġthoracic": 18039,
+      "ĠMPI": 18040,
+      "portion": 18041,
+      "Ġinductive": 18042,
+      "ĠCongress": 18043,
+      "Ġfibroblast": 18044,
+      "clust": 18045,
+      "Ġcentres": 18046,
+      "adel": 18047,
+      "Ġsubstitutions": 18048,
+      "Ġtruncation": 18049,
+      "rification": 18050,
+      "oka": 18051,
+      "Flow": 18052,
+      "ĠReduc": 18053,
+      "polarized": 18054,
+      "ibular": 18055,
+      "Pe": 18056,
+      "ĠAML": 18057,
+      "ĠAgency": 18058,
+      "Ġtilt": 18059,
+      "ublished": 18060,
+      "Ġdepolar": 18061,
+      "Ġbelt": 18062,
+      "Ġoptimizer": 18063,
+      "ELL": 18064,
+      "ĠHandbook": 18065,
+      "ĠVirginia": 18066,
+      "sense": 18067,
+      "ĠDur": 18068,
+      "Ġpiezoelectric": 18069,
+      "Ġawarded": 18070,
+      "ailing": 18071,
+      "Pos": 18072,
+      "pref": 18073,
+      "ĠSummer": 18074,
+      "edo": 18075,
+      "ĠIde": 18076,
+      "ĠBSA": 18077,
+      "Ġmonomers": 18078,
+      "Ġcoagulation": 18079,
+      "Ġgam": 18080,
+      "Ġhomes": 18081,
+      "Ġheads": 18082,
+      "admium": 18083,
+      "ĠOC": 18084,
+      "Ġoccupancy": 18085,
+      "ĠEmpirical": 18086,
+      "ĠIi": 18087,
+      "Ġchir": 18088,
+      "Ġdegeneracy": 18089,
+      "Ġflowers": 18090,
+      "Ġsuperconductivity": 18091,
+      "Ġinversely": 18092,
+      "optical": 18093,
+      "were": 18094,
+      "ĠAsymptotic": 18095,
+      "Sec": 18096,
+      "title": 18097,
+      "posal": 18098,
+      "ĠProgn": 18099,
+      "Ġposes": 18100,
+      "ĠBorn": 18101,
+      "Ġcontinuation": 18102,
+      "Ġcultivated": 18103,
+      "entiment": 18104,
+      "Ġmanaging": 18105,
+      "Ġthrombosis": 18106,
+      "aug": 18107,
+      "CNT": 18108,
+      "urea": 18109,
+      "Ġspind": 18110,
+      "ĠWhereas": 18111,
+      "ĠPerson": 18112,
+      "Ġbipartite": 18113,
+      "Ġrescal": 18114,
+      "Ġmarkets": 18115,
+      "phan": 18116,
+      "perties": 18117,
+      "Ġfermionic": 18118,
+      "Ġmunicip": 18119,
+      "Ġachievable": 18120,
+      "tab": 18121,
+      "Åį": 18122,
+      "ĠRelation": 18123,
+      "Total": 18124,
+      "xia": 18125,
+      "Ġintelligent": 18126,
+      "ĠUT": 18127,
+      "ĠDal": 18128,
+      "Ġmedicinal": 18129,
+      "Ġinadequate": 18130,
+      "iently": 18131,
+      "ersen": 18132,
+      "Ġprecondition": 18133,
+      "Ġmethodological": 18134,
+      "Ġcanopy": 18135,
+      "Ġbacterium": 18136,
+      "column": 18137,
+      "Cal": 18138,
+      "ĠDiego": 18139,
+      "ĠSak": 18140,
+      "ĠComprehensive": 18141,
+      "Ġantitumor": 18142,
+      "Ġflower": 18143,
+      "ĠKhan": 18144,
+      "Ġmetadata": 18145,
+      "Ġphotore": 18146,
+      "ogenicity": 18147,
+      "Ġleague": 18148,
+      "olating": 18149,
+      "Ġpromise": 18150,
+      "ĠPere": 18151,
+      "Ġpermits": 18152,
+      "Ġthreads": 18153,
+      "ĠDCs": 18154,
+      "ĠCham": 18155,
+      "razol": 18156,
+      "Bank": 18157,
+      "Ġwithdrawal": 18158,
+      "Ġappend": 18159,
+      "othelial": 18160,
+      "ĠMeasures": 18161,
+      "Ġguideline": 18162,
+      "Ġmitigate": 18163,
+      "adjoint": 18164,
+      "Ġbracket": 18165,
+      "Pad": 18166,
+      "Mills": 18167,
+      "Buffer": 18168,
+      "Ġcass": 18169,
+      "hoc": 18170,
+      "manifolds": 18171,
+      "herry": 18172,
+      "Ġfacilitated": 18173,
+      "Event": 18174,
+      "ĠÈ": 18175,
+      "ĠCruz": 18176,
+      "ĠBrand": 18177,
+      "Ġnecessity": 18178,
+      "burgh": 18179,
+      "ĠmeV": 18180,
+      "ĠcAMP": 18181,
+      "Off": 18182,
+      "selected": 18183,
+      "Ġengage": 18184,
+      "Ġredundancy": 18185,
+      "Ġnanocomposites": 18186,
+      "solution": 18187,
+      "onset": 18188,
+      "ĠExposure": 18189,
+      "Ġrepetitive": 18190,
+      "Ãł": 18191,
+      "ĠRAD": 18192,
+      "ĠTurk": 18193,
+      "Ġcorneal": 18194,
+      "Ġexploiting": 18195,
+      "Ġobstructive": 18196,
+      "gramming": 18197,
+      "ĠMED": 18198,
+      "Ġmathem": 18199,
+      "Ġconductive": 18200,
+      "Ġphotosynthesis": 18201,
+      "Einstein": 18202,
+      "ĠPeng": 18203,
+      "MW": 18204,
+      "ĠSchmidt": 18205,
+      "Ġrepetition": 18206,
+      "identified": 18207,
+      "Ġinjured": 18208,
+      "Ġdefective": 18209,
+      "ĠPel": 18210,
+      "Ġcultivation": 18211,
+      "Ġfirstly": 18212,
+      "Ġanalyzer": 18213,
+      "Ġstainless": 18214,
+      "Ġjoining": 18215,
+      "ĠOxidative": 18216,
+      "Ġphage": 18217,
+      "Ġexpendit": 18218,
+      "Ġhomogeneity": 18219,
+      "iple": 18220,
+      "ovic": 18221,
+      "Ġcrossed": 18222,
+      "ĠTrust": 18223,
+      "ĠFract": 18224,
+      "rophysiological": 18225,
+      "Ġbasically": 18226,
+      "Ġcoales": 18227,
+      "Ġgravit": 18228,
+      "fulness": 18229,
+      "cano": 18230,
+      "Ġcolitis": 18231,
+      "Ġchaos": 18232,
+      "carbons": 18233,
+      "Once": 18234,
+      "ĠToward": 18235,
+      "orf": 18236,
+      "topic": 18237,
+      "ĠPlay": 18238,
+      "ĠCorrespond": 18239,
+      "ĠSleep": 18240,
+      "ticularly": 18241,
+      "cumin": 18242,
+      "vdots": 18243,
+      "ĠRhe": 18244,
+      "Ġultraf": 18245,
+      "Ġtimescale": 18246,
+      "ĠDetails": 18247,
+      "angles": 18248,
+      "Ġsurrogate": 18249,
+      "ĠFluid": 18250,
+      "cz": 18251,
+      "Ġinitialization": 18252,
+      "ĠTelescope": 18253,
+      "rases": 18254,
+      "ĠStock": 18255,
+      "ĠCond": 18256,
+      "Ġimmunodeficiency": 18257,
+      "Bel": 18258,
+      "oser": 18259,
+      "shown": 18260,
+      "Ġkcal": 18261,
+      "Equation": 18262,
+      "protective": 18263,
+      "Ġcalling": 18264,
+      "Ġanticipated": 18265,
+      "Ġambiguity": 18266,
+      "ĠNode": 18267,
+      "ĠGD": 18268,
+      "Ġinlet": 18269,
+      "Ġbread": 18270,
+      "Ġexceeded": 18271,
+      "Ġimmunization": 18272,
+      "Ġprohib": 18273,
+      "ytic": 18274,
+      "Ġboys": 18275,
+      "tu": 18276,
+      "Ġtower": 18277,
+      "Like": 18278,
+      "ĠAnomal": 18279,
+      "âĮ": 18280,
+      "ĠShow": 18281,
+      "Ġimaged": 18282,
+      "Ġequil": 18283,
+      "Ġrendering": 18284,
+      "obility": 18285,
+      "Ġgeological": 18286,
+      "friend": 18287,
+      "ör": 18288,
+      "carboxamide": 18289,
+      "ovolta": 18290,
+      "Current": 18291,
+      "ĠSti": 18292,
+      "ĠMU": 18293,
+      "Ġvalued": 18294,
+      "Ġpoison": 18295,
+      "Ġpractically": 18296,
+      "Ġrequested": 18297,
+      "Code": 18298,
+      "Ġbrings": 18299,
+      "Ġdimethyl": 18300,
+      "hyp": 18301,
+      "cemic": 18302,
+      "Vol": 18303,
+      "quanti": 18304,
+      "Ġexha": 18305,
+      "Ġresponsibility": 18306,
+      "ĠControlled": 18307,
+      "Ġfur": 18308,
+      "Ġresemb": 18309,
+      "ĠKaw": 18310,
+      "Ġevoked": 18311,
+      "Ġuterine": 18312,
+      "л": 18313,
+      "Ġanonymous": 18314,
+      "ĠChallenges": 18315,
+      "Ġanchor": 18316,
+      "ĠAbd": 18317,
+      "Der": 18318,
+      "Ġthermally": 18319,
+      "ĠCAP": 18320,
+      "oblot": 18321,
+      "ĠFire": 18322,
+      "Ġdiagnostics": 18323,
+      "Ġexecute": 18324,
+      "alis": 18325,
+      "roni": 18326,
+      "ĠHarris": 18327,
+      "ĠGonz": 18328,
+      "Ġvig": 18329,
+      "ĠProfessor": 18330,
+      "Ġinventory": 18331,
+      "intensity": 18332,
+      "ĠNSCLC": 18333,
+      "Ġinterfere": 18334,
+      "ysaccharides": 18335,
+      "Ġregener": 18336,
+      "ĠAuthors": 18337,
+      "Ġtranslate": 18338,
+      "ĠTests": 18339,
+      "ĠLove": 18340,
+      "ĠInduced": 18341,
+      "ennis": 18342,
+      "ĠGEN": 18343,
+      "Ġoligonucle": 18344,
+      "Ġmeter": 18345,
+      "satisf": 18346,
+      "hesion": 18347,
+      "Ġtransporters": 18348,
+      "BIT": 18349,
+      "ĠConc": 18350,
+      "Ġglauc": 18351,
+      "scores": 18352,
+      "Ġmerger": 18353,
+      "GH": 18354,
+      "Ġstoichi": 18355,
+      "ĠXia": 18356,
+      "effects": 18357,
+      "ĠExploring": 18358,
+      "dorff": 18359,
+      "Ġcardinality": 18360,
+      "ĠKaz": 18361,
+      "false": 18362,
+      "ĠHSP": 18363,
+      "Ġunsupervised": 18364,
+      "inguish": 18365,
+      "ischer": 18366,
+      "Ġrelativity": 18367,
+      "onormal": 18368,
+      "oothed": 18369,
+      "edges": 18370,
+      "ĠIMP": 18371,
+      "Ġimpulse": 18372,
+      "ĠColumbia": 18373,
+      "Ġparticulate": 18374,
+      "ĠSupporting": 18375,
+      "ĠSDSS": 18376,
+      "voltage": 18377,
+      "ĠAmazon": 18378,
+      "Ġepoxy": 18379,
+      "Call": 18380,
+      "Bigl": 18381,
+      "Ġmeets": 18382,
+      "Ġequatorial": 18383,
+      "Ġneuros": 18384,
+      "Ġperitoneal": 18385,
+      "desc": 18386,
+      "inputs": 18387,
+      "Ġexterior": 18388,
+      "aco": 18389,
+      "Ġmeal": 18390,
+      "ĠDaniel": 18391,
+      "Ġintuitive": 18392,
+      "Ġcouns": 18393,
+      "depress": 18394,
+      "inis": 18395,
+      "phot": 18396,
+      "ĠAmin": 18397,
+      "Ġreservoirs": 18398,
+      "ĠWhole": 18399,
+      "Ġcaud": 18400,
+      "Ġbosonic": 18401,
+      "Ġreaders": 18402,
+      "Ġcrim": 18403,
+      "Ġpathophysiology": 18404,
+      "argo": 18405,
+      "these": 18406,
+      "income": 18407,
+      "Ġissued": 18408,
+      "Ġhepatocytes": 18409,
+      "ĠCi": 18410,
+      "deriv": 18411,
+      "upta": 18412,
+      "tuple": 18413,
+      "ĠChan": 18414,
+      "Ġauthentication": 18415,
+      "ygd": 18416,
+      "Ġinfin": 18417,
+      "Ġaccelerate": 18418,
+      "eptive": 18419,
+      "Ġhydrogel": 18420,
+      "aska": 18421,
+      "ONE": 18422,
+      "Ġfederal": 18423,
+      "ographics": 18424,
+      "Ġmuon": 18425,
+      "Ġslide": 18426,
+      "Ġelliptical": 18427,
+      "atite": 18428,
+      "Ġcc": 18429,
+      "ETs": 18430,
+      "Ġclarity": 18431,
+      "ocycl": 18432,
+      "isal": 18433,
+      "rections": 18434,
+      "ayan": 18435,
+      "roweak": 18436,
+      "ĠSOC": 18437,
+      "oderm": 18438,
+      "tun": 18439,
+      "asm": 18440,
+      "ĠHir": 18441,
+      "likelihood": 18442,
+      "Ġadul": 18443,
+      "tl": 18444,
+      "High": 18445,
+      "Ġalters": 18446,
+      "plitude": 18447,
+      "ĠRelease": 18448,
+      "Ġharmful": 18449,
+      "late": 18450,
+      "ounds": 18451,
+      "ĠFederal": 18452,
+      "ĠEconomic": 18453,
+      "Ġrabb": 18454,
+      "Ġaccommodate": 18455,
+      "emission": 18456,
+      "ĠBah": 18457,
+      "cox": 18458,
+      "ĠModulation": 18459,
+      "Ġconstructions": 18460,
+      "igner": 18461,
+      "ĠUrban": 18462,
+      "Ġwake": 18463,
+      "Ġadversary": 18464,
+      "wikipedia": 18465,
+      "Ġsuite": 18466,
+      "wick": 18467,
+      "expressed": 18468,
+      "rod": 18469,
+      "KD": 18470,
+      "Ġcomputers": 18471,
+      "ĠBanglades": 18472,
+      "Ġpersist": 18473,
+      "Ġburning": 18474,
+      "Ġadministrative": 18475,
+      "Ġplug": 18476,
+      "ĠRepresentations": 18477,
+      "ĠScattering": 18478,
+      "Ġendometrial": 18479,
+      "Ġdescriptors": 18480,
+      "Ġcommission": 18481,
+      "Bar": 18482,
+      "ighth": 18483,
+      "ĠMarsh": 18484,
+      "sampling": 18485,
+      "Ġhull": 18486,
+      "icin": 18487,
+      "Prob": 18488,
+      "Ġnurse": 18489,
+      "Ġsham": 18490,
+      "ĠKerr": 18491,
+      "Ġprefrontal": 18492,
+      "Ġfixing": 18493,
+      "OK": 18494,
+      "Ġbold": 18495,
+      "Ġcorollary": 18496,
+      "cfg": 18497,
+      "ĠOxford": 18498,
+      "Ġboron": 18499,
+      "RB": 18500,
+      "ĠCab": 18501,
+      "Bigr": 18502,
+      "ĠPredict": 18503,
+      "Ġpeculiar": 18504,
+      "hidden": 18505,
+      "isa": 18506,
+      "iden": 18507,
+      "appropriate": 18508,
+      "orh": 18509,
+      "ellectual": 18510,
+      "Ġseizures": 18511,
+      "asser": 18512,
+      "tilis": 18513,
+      "handle": 18514,
+      "iaxial": 18515,
+      "sym": 18516,
+      "Ġcarcinomas": 18517,
+      "sea": 18518,
+      "spired": 18519,
+      "Ġabrupt": 18520,
+      "tests": 18521,
+      "Ġwelfare": 18522,
+      "ĠOil": 18523,
+      "ĠLoad": 18524,
+      "FLAG": 18525,
+      "uthal": 18526,
+      "Ġfacing": 18527,
+      "American": 18528,
+      "LAS": 18529,
+      "Ġirrespective": 18530,
+      "Ġroutinely": 18531,
+      "wal": 18532,
+      "Ġsettlement": 18533,
+      "ĠAqu": 18534,
+      "Ġelectronics": 18535,
+      "Ġhandled": 18536,
+      "Ġbiologically": 18537,
+      "smooth": 18538,
+      "ĠBelongs": 18539,
+      "tib": 18540,
+      "Ġtrav": 18541,
+      "pressive": 18542,
+      "ournals": 18543,
+      "к": 18544,
+      "filename": 18545,
+      "Ġhelical": 18546,
+      "Ġbacteri": 18547,
+      "Ġsatellites": 18548,
+      "BH": 18549,
+      "ented": 18550,
+      "ĠFootball": 18551,
+      "Ġ": 18552,
+      "ĠHV": 18553,
+      "Ġtrip": 18554,
+      "ĠCKD": 18555,
+      "rani": 18556,
+      "Ġcleaning": 18557,
+      "limit": 18558,
+      "ĠTCP": 18559,
+      "Ġscin": 18560,
+      "Ġsludge": 18561,
+      "Ġsymbolic": 18562,
+      "ĠSequencing": 18563,
+      "adal": 18564,
+      "ĠPhilipp": 18565,
+      "ICS": 18566,
+      "Ġvaginal": 18567,
+      "Ġcommitment": 18568,
+      "ĠAwards": 18569,
+      "trig": 18570,
+      "Ġguitar": 18571,
+      "acetate": 18572,
+      "Ġbet": 18573,
+      "ClN": 18574,
+      "Ġagriculture": 18575,
+      "Ġchief": 18576,
+      "Ġembol": 18577,
+      "build": 18578,
+      "Ġtexts": 18579,
+      "ĠCooper": 18580,
+      "lived": 18581,
+      "ĠDelay": 18582,
+      "ĠMode": 18583,
+      "yal": 18584,
+      "BN": 18585,
+      "Ġindexed": 18586,
+      "expr": 18587,
+      "ERN": 18588,
+      "vens": 18589,
+      "Ġpointer": 18590,
+      "cv": 18591,
+      "acon": 18592,
+      "tance": 18593,
+      "ĠâĪĿ": 18594,
+      "Ġlowered": 18595,
+      "Ġmitotic": 18596,
+      "rhosis": 18597,
+      "ĠPage": 18598,
+      "ür": 18599,
+      "imm": 18600,
+      "ĠTherapeutic": 18601,
+      "Ġosteopor": 18602,
+      "Ġbilinear": 18603,
+      "ĠCatholic": 18604,
+      "ĠAlternative": 18605,
+      "oxidation": 18606,
+      "Ġinitio": 18607,
+      "benzo": 18608,
+      "ĠAdi": 18609,
+      "person": 18610,
+      "peritoneal": 18611,
+      "ĉĉĠ": 18612,
+      "Ġattraction": 18613,
+      "Ġdiarrhea": 18614,
+      "Ġren": 18615,
+      "ĠISO": 18616,
+      "imir": 18617,
+      "Ġterminology": 18618,
+      "ukey": 18619,
+      "Ġresonator": 18620,
+      "Ġsubstituting": 18621,
+      "Ġharbor": 18622,
+      "provid": 18623,
+      "decay": 18624,
+      "ĠHDAC": 18625,
+      "ĠAnalytical": 18626,
+      "Ġpostnatal": 18627,
+      "Ġundes": 18628,
+      "Specific": 18629,
+      "dichlor": 18630,
+      "ARI": 18631,
+      "tot": 18632,
+      "Ġdigit": 18633,
+      "oping": 18634,
+      "ĠZinc": 18635,
+      "Ġlethal": 18636,
+      "Whitney": 18637,
+      "Fi": 18638,
+      "quantum": 18639,
+      "ĠFailure": 18640,
+      "Ġsolves": 18641,
+      "ĠSpaces": 18642,
+      "earman": 18643,
+      "Ġgoat": 18644,
+      "Ġsynapses": 18645,
+      "Ġresuspended": 18646,
+      "Ġresident": 18647,
+      "Ġcompac": 18648,
+      "Ġcortisol": 18649,
+      "Ġphotometry": 18650,
+      "WP": 18651,
+      "select": 18652,
+      "Ġcele": 18653,
+      "orubicin": 18654,
+      "ĠMultic": 18655,
+      "ĠJean": 18656,
+      "Ġclip": 18657,
+      "Ġsa": 18658,
+      "oco": 18659,
+      "geometric": 18660,
+      "Ġhelic": 18661,
+      "Ġempirically": 18662,
+      "Ġmicrofluid": 18663,
+      "idis": 18664,
+      "Ġautocor": 18665,
+      "WF": 18666,
+      "ĠRespir": 18667,
+      "radiation": 18668,
+      "Ġthemes": 18669,
+      "Ġtaste": 18670,
+      "ricing": 18671,
+      "Ġexaminations": 18672,
+      "ĠSensing": 18673,
+      "same": 18674,
+      "DEFAULT": 18675,
+      "Ġphylogeny": 18676,
+      "hig": 18677,
+      "Ġplatelets": 18678,
+      "ĠHistor": 18679,
+      "aba": 18680,
+      "Ġresidential": 18681,
+      "Ġunbounded": 18682,
+      "anding": 18683,
+      "hedron": 18684,
+      "rys": 18685,
+      "ĠCCR": 18686,
+      "Ġconce": 18687,
+      "Ġparasitic": 18688,
+      "cb": 18689,
+      "ĠFeynman": 18690,
+      "ĠKepler": 18691,
+      "ô": 18692,
+      "ĠGil": 18693,
+      "ĠMATLAB": 18694,
+      "ben": 18695,
+      "scope": 18696,
+      "Ġdiscrimin": 18697,
+      "Ġjustified": 18698,
+      "plasma": 18699,
+      "ĠChoi": 18700,
+      "Ġroof": 18701,
+      "PCA": 18702,
+      "ĠTCR": 18703,
+      "Ġvoxel": 18704,
+      "ĠWard": 18705,
+      "Ġuncor": 18706,
+      "Stok": 18707,
+      "Ġspur": 18708,
+      "TRA": 18709,
+      "Ġdiagnoses": 18710,
+      "rophysical": 18711,
+      "ategories": 18712,
+      "Ġoverestim": 18713,
+      "Ġstreaming": 18714,
+      "ĠRecovery": 18715,
+      "Ġeverything": 18716,
+      "LOW": 18717,
+      "Gener": 18718,
+      "Ġunbiased": 18719,
+      "Ġvariances": 18720,
+      "compact": 18721,
+      "espan": 18722,
+      "inj": 18723,
+      "Ġendoscopic": 18724,
+      "Ġideals": 18725,
+      "ĠRice": 18726,
+      "ĠKaplan": 18727,
+      "Ġfecal": 18728,
+      "ferred": 18729,
+      "ĠCycle": 18730,
+      "Ġimplanted": 18731,
+      "Ġwine": 18732,
+      "PET": 18733,
+      "Ġassignments": 18734,
+      "Ġabsol": 18735,
+      "XT": 18736,
+      "Ġswimming": 18737,
+      "MN": 18738,
+      "ĠGeometric": 18739,
+      "ĠHealthcare": 18740,
+      "Ġpowders": 18741,
+      "ĠGel": 18742,
+      "Ġdownward": 18743,
+      "Ġexceeding": 18744,
+      "ĠHEK": 18745,
+      "lym": 18746,
+      "ĠBV": 18747,
+      "Ġvisco": 18748,
+      "iet": 18749,
+      "ĠCOX": 18750,
+      "ployment": 18751,
+      "inski": 18752,
+      "Ġoutdoor": 18753,
+      "ĠLiterature": 18754,
+      "anted": 18755,
+      "methoxyphenyl": 18756,
+      "ĠMedium": 18757,
+      "Ġdia": 18758,
+      "ailand": 18759,
+      "variance": 18760,
+      "ĠEvaluating": 18761,
+      "oxacin": 18762,
+      "Ġantif": 18763,
+      "Ġpulp": 18764,
+      "Ġcorrobor": 18765,
+      "ĠOt": 18766,
+      "Ġrabbits": 18767,
+      "Ru": 18768,
+      "Ġfunctionals": 18769,
+      "âĩ": 18770,
+      "Ġimmersion": 18771,
+      "Ġcreatin": 18772,
+      "ĠqRT": 18773,
+      "Ġcondensed": 18774,
+      "nr": 18775,
+      "ĠVA": 18776,
+      "had": 18777,
+      "Ġking": 18778,
+      "oble": 18779,
+      "Ġexisted": 18780,
+      "Ġthesis": 18781,
+      "ubbard": 18782,
+      "apoptotic": 18783,
+      "Ġflowering": 18784,
+      "ĠAdaptation": 18785,
+      "ĠKalman": 18786,
+      "trl": 18787,
+      "Ġment": 18788,
+      "utation": 18789,
+      "ĠConv": 18790,
+      "Ġhistories": 18791,
+      "Ġenanti": 18792,
+      "nell": 18793,
+      "onian": 18794,
+      "ĠFabric": 18795,
+      "Ġxx": 18796,
+      "Ġfell": 18797,
+      "Ġcytosolic": 18798,
+      "Ġmud": 18799,
+      "Ġsuspensions": 18800,
+      "ĠMicrobial": 18801,
+      "measured": 18802,
+      "Ġdownload": 18803,
+      "Ġinvalid": 18804,
+      "Ġcapturing": 18805,
+      "ĠHH": 18806,
+      "ĠGray": 18807,
+      "ĠAZ": 18808,
+      "ĠNash": 18809,
+      "viation": 18810,
+      "naire": 18811,
+      "ortium": 18812,
+      "ynch": 18813,
+      "aminergic": 18814,
+      "Ġwait": 18815,
+      "Schem": 18816,
+      "trace": 18817,
+      "ĠVill": 18818,
+      "Ġpools": 18819,
+      "Ġhypoxic": 18820,
+      "xp": 18821,
+      "Ġshaded": 18822,
+      "ORY": 18823,
+      "turn": 18824,
+      "interacting": 18825,
+      "Ġdestroyed": 18826,
+      "akh": 18827,
+      "ĠCpG": 18828,
+      "dotted": 18829,
+      "ĠTranscript": 18830,
+      "planar": 18831,
+      "Ġpreclinical": 18832,
+      "ĠRepro": 18833,
+      "ĠSurgery": 18834,
+      "Stokes": 18835,
+      "ifdef": 18836,
+      "Ġdiscriminate": 18837,
+      "ĠGross": 18838,
+      "Ġflags": 18839,
+      "iety": 18840,
+      "ummy": 18841,
+      "Ġtransfers": 18842,
+      "SG": 18843,
+      "ĠSci": 18844,
+      "Ġheader": 18845,
+      "ĠFunding": 18846,
+      "Ġdetrim": 18847,
+      "Ġinstabilities": 18848,
+      "ĠPhylogenetic": 18849,
+      "ymethyl": 18850,
+      "ĠAssessing": 18851,
+      "ROC": 18852,
+      "elsen": 18853,
+      "Equal": 18854,
+      "Ġcas": 18855,
+      "Ġvertically": 18856,
+      "Ġvisibility": 18857,
+      "ĠFTIR": 18858,
+      "scrib": 18859,
+      "Ġbursts": 18860,
+      "ĠDoug": 18861,
+      "ĠFrancisco": 18862,
+      "ĠMSC": 18863,
+      "Ġpredis": 18864,
+      "established": 18865,
+      "Ġfaced": 18866,
+      "ĠWI": 18867,
+      "Sl": 18868,
+      "Ġcharts": 18869,
+      "orthy": 18870,
+      "izontal": 18871,
+      "ialysis": 18872,
+      "Ġtunable": 18873,
+      "Ġexplosion": 18874,
+      "Sw": 18875,
+      "TNF": 18876,
+      "Ġdiscontinuous": 18877,
+      "ecture": 18878,
+      "ciences": 18879,
+      "mathbbm": 18880,
+      "look": 18881,
+      "Ġtachy": 18882,
+      "Ġbrow": 18883,
+      "observed": 18884,
+      "Ġanaest": 18885,
+      "Sal": 18886,
+      "qPCR": 18887,
+      "Ġsees": 18888,
+      "Ġspacecraft": 18889,
+      "Ġsales": 18890,
+      "ĠTrac": 18891,
+      "Tem": 18892,
+      "ivest": 18893,
+      "ĠFc": 18894,
+      "ĠNews": 18895,
+      "Ġharvesting": 18896,
+      "ĠEG": 18897,
+      "pad": 18898,
+      "Ġnanowires": 18899,
+      "Ġpotato": 18900,
+      "pliers": 18901,
+      "onin": 18902,
+      "Ġworm": 18903,
+      "sue": 18904,
+      "tie": 18905,
+      "Ġmasks": 18906,
+      "Ġthrow": 18907,
+      "!!": 18908,
+      "behavi": 18909,
+      "Ġpine": 18910,
+      "ogy": 18911,
+      "TEST": 18912,
+      "onto": 18913,
+      "Ġcreatinine": 18914,
+      "ĠBoston": 18915,
+      "Ġchair": 18916,
+      "ploys": 18917,
+      "oven": 18918,
+      "Ġentrance": 18919,
+      "Ġcoch": 18920,
+      "Ġdyes": 18921,
+      "Tor": 18922,
+      "ĠPDE": 18923,
+      "underset": 18924,
+      "atasets": 18925,
+      "Ġternary": 18926,
+      "choose": 18927,
+      "five": 18928,
+      "chloride": 18929,
+      "onium": 18930,
+      "Property": 18931,
+      "Ġtu": 18932,
+      "Ġadequately": 18933,
+      "romycin": 18934,
+      "Ġcooper": 18935,
+      "ïĽľ": 18936,
+      "Ġpapill": 18937,
+      "ĠStreptococcus": 18938,
+      "ĠCY": 18939,
+      "Ġgrouping": 18940,
+      "Ġbioc": 18941,
+      "ĠCardiac": 18942,
+      "ĠBook": 18943,
+      "reference": 18944,
+      "Ġconfirmation": 18945,
+      "ivery": 18946,
+      "Ġwarning": 18947,
+      "pretation": 18948,
+      "Ġlove": 18949,
+      "Ġoscillators": 18950,
+      "sed": 18951,
+      "ĠTX": 18952,
+      "ilent": 18953,
+      "ĠVas": 18954,
+      "Ġclamp": 18955,
+      "Ġahead": 18956,
+      "acs": 18957,
+      "Ġdepleted": 18958,
+      "Ġmethodologies": 18959,
+      "may": 18960,
+      "Ġcaffe": 18961,
+      "Ġsequentially": 18962,
+      "osacchar": 18963,
+      "Ġcomprise": 18964,
+      "Ġchel": 18965,
+      "Ġinacc": 18966,
+      "Ġtendon": 18967,
+      "Sequ": 18968,
+      "ought": 18969,
+      "server": 18970,
+      "ĠPerturb": 18971,
+      "Ġterrain": 18972,
+      "curve": 18973,
+      "ĠArgent": 18974,
+      "TABLE": 18975,
+      "Ġimplicitly": 18976,
+      "Ġenjoy": 18977,
+      "ĠSitter": 18978,
+      "Ġmicron": 18979,
+      "ĠEvans": 18980,
+      "nsylvan": 18981,
+      "Ġlooked": 18982,
+      "spe": 18983,
+      "volving": 18984,
+      "ĠLSTM": 18985,
+      "agnetism": 18986,
+      "ĠNotch": 18987,
+      "ĠTal": 18988,
+      "ĠDEGs": 18989,
+      "leman": 18990,
+      "Ġboolean": 18991,
+      "Ġobey": 18992,
+      "organization": 18993,
+      "seen": 18994,
+      "ĠEnc": 18995,
+      "schild": 18996,
+      "ĠOntario": 18997,
+      "Element": 18998,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 18999,
+      "mouse": 19000,
+      "Ġpolyethylene": 19001,
+      "Ġacetic": 19002,
+      "sections": 19003,
+      "uronal": 19004,
+      "ĠDick": 19005,
+      "Ġkill": 19006,
+      "Ġbroadening": 19007,
+      "Ġfluoride": 19008,
+      "Ġsaved": 19009,
+      "Ġdeem": 19010,
+      "Stream": 19011,
+      "aced": 19012,
+      "ĠJeff": 19013,
+      "QA": 19014,
+      "Ġscalable": 19015,
+      "ĠFif": 19016,
+      "ĠMini": 19017,
+      "Ġsupergravity": 19018,
+      "Ġcolloidal": 19019,
+      "LY": 19020,
+      "OA": 19021,
+      "Ġperic": 19022,
+      "Ġshortly": 19023,
+      "Ġvap": 19024,
+      "Ġsplits": 19025,
+      "move": 19026,
+      "Ġstimulating": 19027,
+      "ĠBeijing": 19028,
+      "Ġpyr": 19029,
+      "ÏŃ": 19030,
+      "Ġlexical": 19031,
+      "âĢł": 19032,
+      "ÅĦ": 19033,
+      "itories": 19034,
+      "olerance": 19035,
+      "Ġinsulator": 19036,
+      "ĠLeon": 19037,
+      "Ġpropagate": 19038,
+      "ĠElements": 19039,
+      "yen": 19040,
+      "Module": 19041,
+      "ĠWhether": 19042,
+      "Ġaph": 19043,
+      "ĠLaure": 19044,
+      "ĠMutations": 19045,
+      "Ġhypertrophy": 19046,
+      "Ġoceanic": 19047,
+      "ographically": 19048,
+      "patients": 19049,
+      "ĠAngeles": 19050,
+      "Ġphe": 19051,
+      "Ġsquee": 19052,
+      "Ġcaroten": 19053,
+      "fine": 19054,
+      "Ġsketch": 19055,
+      "Ġansatz": 19056,
+      "titution": 19057,
+      "ĠFus": 19058,
+      "ĠSug": 19059,
+      "obacterial": 19060,
+      "Ħĥ": 19061,
+      "Related": 19062,
+      "Ġartist": 19063,
+      "Ġacryl": 19064,
+      "lined": 19065,
+      "rafted": 19066,
+      "ĠQoS": 19067,
+      "ĠFeng": 19068,
+      "search": 19069,
+      "Ġnanotube": 19070,
+      "ĠVM": 19071,
+      "ahl": 19072,
+      "Ġstride": 19073,
+      "ĠTag": 19074,
+      "ĠLar": 19075,
+      "Ġdesorption": 19076,
+      "dtype": 19077,
+      "Ġbug": 19078,
+      "Ġcaregivers": 19079,
+      "ĠHun": 19080,
+      "ĠPractical": 19081,
+      "Ġoblig": 19082,
+      "rer": 19083,
+      "ĠKang": 19084,
+      "ĠProducts": 19085,
+      "ometh": 19086,
+      "ĠHeLa": 19087,
+      "Ġlaboratories": 19088,
+      "natural": 19089,
+      "Ġful": 19090,
+      "Ġmold": 19091,
+      "abine": 19092,
+      "ĠSpring": 19093,
+      "Ġcobal": 19094,
+      "Ġhighlighting": 19095,
+      "ĠPref": 19096,
+      "cyclic": 19097,
+      "ĠCONCLUSION": 19098,
+      "ĠSources": 19099,
+      "Ġapex": 19100,
+      "parser": 19101,
+      "ĠLogic": 19102,
+      "Ġpond": 19103,
+      "Ġtold": 19104,
+      "ĠShap": 19105,
+      "pergillus": 19106,
+      "Ġsaying": 19107,
+      "Ġmutagenesis": 19108,
+      "ĠmmHg": 19109,
+      "ĠPAN": 19110,
+      "Ġsmokers": 19111,
+      "oday": 19112,
+      "Ġherein": 19113,
+      "CMV": 19114,
+      "ĠPW": 19115,
+      "Ġredshifts": 19116,
+      "ĠMinim": 19117,
+      "yman": 19118,
+      "ulli": 19119,
+      "dense": 19120,
+      "Ġarsenic": 19121,
+      "ĠEMT": 19122,
+      "ogaster": 19123,
+      "carboxylate": 19124,
+      "sys": 19125,
+      "Ro": 19126,
+      "anch": 19127,
+      "ĠAlpha": 19128,
+      "ĠTechnical": 19129,
+      "sv": 19130,
+      "Ġbones": 19131,
+      "Ġacceptor": 19132,
+      "Ġnewborn": 19133,
+      "private": 19134,
+      "Ġnanor": 19135,
+      "ĠSwiss": 19136,
+      "around": 19137,
+      "Ġsyntax": 19138,
+      "ĠKähler": 19139,
+      "Ġaerial": 19140,
+      "ĠPale": 19141,
+      "typedef": 19142,
+      "namespace": 19143,
+      "Ġconfounding": 19144,
+      "viÄĩ": 19145,
+      "Ġretard": 19146,
+      "Ġzeta": 19147,
+      "ĠTum": 19148,
+      "isch": 19149,
+      "Ġsulfide": 19150,
+      "ĠTian": 19151,
+      "uy": 19152,
+      "Ġintuition": 19153,
+      "Ġphospholip": 19154,
+      "ĠSher": 19155,
+      "ricts": 19156,
+      "----------------------------------------------------------------": 19157,
+      "okines": 19158,
+      "glucose": 19159,
+      "toler": 19160,
+      "iferative": 19161,
+      "ĠFluor": 19162,
+      "Ġencourage": 19163,
+      "Ġresponsive": 19164,
+      "perturbative": 19165,
+      "Ġsaddle": 19166,
+      "lers": 19167,
+      "ndez": 19168,
+      "ĠZero": 19169,
+      "ĠDiet": 19170,
+      "Ġdevelopers": 19171,
+      "Syn": 19172,
+      "Ġconfer": 19173,
+      "Ġoriginate": 19174,
+      "ropol": 19175,
+      "haw": 19176,
+      "letion": 19177,
+      "mskip": 19178,
+      "Ġber": 19179,
+      "Ġpeat": 19180,
+      "vially": 19181,
+      "Ġgranules": 19182,
+      "ĠÌĥ": 19183,
+      "Ġpluripot": 19184,
+      "Ġassimilation": 19185,
+      "Ġdenominator": 19186,
+      "abilization": 19187,
+      "ĠEpidemiology": 19188,
+      "MIN": 19189,
+      "eeds": 19190,
+      "ĠVR": 19191,
+      "Eval": 19192,
+      "store": 19193,
+      "ĠBaseline": 19194,
+      "Ġcu": 19195,
+      "ĠSpectra": 19196,
+      "Ġfractionation": 19197,
+      "Ġplacing": 19198,
+      "Ġburied": 19199,
+      "eleration": 19200,
+      "Ġalkali": 19201,
+      "ĠIU": 19202,
+      "Calc": 19203,
+      "weak": 19204,
+      "Ġmorphisms": 19205,
+      "Ġligase": 19206,
+      "Ġfs": 19207,
+      "Ġutilizes": 19208,
+      "Comput": 19209,
+      "â": 19210,
+      "Ġstig": 19211,
+      "relative": 19212,
+      "Ġimmature": 19213,
+      "ĠFrac": 19214,
+      "api": 19215,
+      "Ġoutpatient": 19216,
+      "Ġachievement": 19217,
+      "Ġstacking": 19218,
+      "Ġnodules": 19219,
+      "IND": 19220,
+      "ĠGPa": 19221,
+      "Ġpercolation": 19222,
+      "mspace": 19223,
+      "Ġbrains": 19224,
+      "uffle": 19225,
+      "entropy": 19226,
+      "Lab": 19227,
+      "Ġstabilize": 19228,
+      "ĠRicci": 19229,
+      "ĠAntimicrobial": 19230,
+      "personal": 19231,
+      "Ġfarms": 19232,
+      "ĠPin": 19233,
+      "Ġporcine": 19234,
+      "Ġoccasionally": 19235,
+      "whe": 19236,
+      "Ġundergoes": 19237,
+      "Ġregimens": 19238,
+      "Ġblade": 19239,
+      "Ġlinearized": 19240,
+      "Ġdecon": 19241,
+      "Ġpacked": 19242,
+      "Ġfishes": 19243,
+      "ĠMend": 19244,
+      "Ġapproaching": 19245,
+      "Ġballs": 19246,
+      "Ġproinflammatory": 19247,
+      "imeric": 19248,
+      "ĠDirector": 19249,
+      "Ġsoliton": 19250,
+      "Ġmosaic": 19251,
+      "viet": 19252,
+      "Mean": 19253,
+      "ĠPad": 19254,
+      "Ġtriplicate": 19255,
+      "supported": 19256,
+      "Ġcart": 19257,
+      "<<<<": 19258,
+      "Ġremission": 19259,
+      "aseous": 19260,
+      "asticity": 19261,
+      "ĠMik": 19262,
+      "ĠStrategy": 19263,
+      "ramer": 19264,
+      "ĠPolish": 19265,
+      "Ġenthal": 19266,
+      "Ġheterozygous": 19267,
+      "ĠGravity": 19268,
+      "Ax": 19269,
+      "Ġorganizational": 19270,
+      "Ġmovie": 19271,
+      "Ġexploratory": 19272,
+      "WLED": 19273,
+      "Ġmoiety": 19274,
+      "decre": 19275,
+      "ĠStill": 19276,
+      "Ġ¡": 19277,
+      "Ġgreenhouse": 19278,
+      "Ġsuperconductors": 19279,
+      "enum": 19280,
+      "elin": 19281,
+      "Ġoffering": 19282,
+      "stad": 19283,
+      "ĠTrich": 19284,
+      "Ġrepl": 19285,
+      "Ġrecycling": 19286,
+      "phor": 19287,
+      "Ġinelastic": 19288,
+      "ockey": 19289,
+      "ĠâĢĻ": 19290,
+      "Ġsequel": 19291,
+      "EB": 19292,
+      "ĠChile": 19293,
+      "Ġfibrillation": 19294,
+      "Ġdisulfide": 19295,
+      "obtained": 19296,
+      "ubin": 19297,
+      "Ĥ¬": 19298,
+      "Ġfacilitating": 19299,
+      "Ġhopping": 19300,
+      "Ġmediator": 19301,
+      "Ġhydration": 19302,
+      "Ġsparsity": 19303,
+      "Ġsati": 19304,
+      "Ġisothermal": 19305,
+      "Ġreturning": 19306,
+      "Ġtraveling": 19307,
+      "Ġing": 19308,
+      "Ġstent": 19309,
+      "Ġcapacitor": 19310,
+      "Ġcompromise": 19311,
+      "ĠSud": 19312,
+      "ĠVision": 19313,
+      "Ġtopologies": 19314,
+      "opolysaccharide": 19315,
+      "ĠProfile": 19316,
+      "ĠRing": 19317,
+      "Ġdiscrepancies": 19318,
+      "Dis": 19319,
+      "ARD": 19320,
+      "cccc": 19321,
+      "Ġdirectory": 19322,
+      "ĠCMOS": 19323,
+      "owed": 19324,
+      "illo": 19325,
+      "ĠInsights": 19326,
+      "ĠTib": 19327,
+      "Ġaband": 19328,
+      "arose": 19329,
+      "Order": 19330,
+      "Ġ¬": 19331,
+      "Ġintracranial": 19332,
+      "Ġintermediates": 19333,
+      "Ġhabits": 19334,
+      "Ġcarp": 19335,
+      "property": 19336,
+      "IMAGE": 19337,
+      "ĠUk": 19338,
+      "Ġhydrophilic": 19339,
+      "Wid": 19340,
+      "Ġabiotic": 19341,
+      "Ġobservers": 19342,
+      "Ġchor": 19343,
+      "ĠConservation": 19344,
+      "ĠEnhance": 19345,
+      "ĠAutomated": 19346,
+      "ĠGlut": 19347,
+      "iratory": 19348,
+      "Ġspaw": 19349,
+      "ĠEfficiency": 19350,
+      "vast": 19351,
+      "initi": 19352,
+      "Ġoptional": 19353,
+      "ĠScaling": 19354,
+      "ifold": 19355,
+      "ĠmtDNA": 19356,
+      "ĠReconstruction": 19357,
+      "Ġcountable": 19358,
+      "ĠGrass": 19359,
+      "Den": 19360,
+      "ĠChain": 19361,
+      "enzyme": 19362,
+      "Ġwaveforms": 19363,
+      "Ġpancreas": 19364,
+      "ĠDetailed": 19365,
+      "cmd": 19366,
+      "Ġâİľ": 19367,
+      "Ġmagneto": 19368,
+      "ĠFPGA": 19369,
+      "Ġabsolutely": 19370,
+      "Ġstimulates": 19371,
+      "achus": 19372,
+      "ĠArn": 19373,
+      "message": 19374,
+      "ocompatibility": 19375,
+      "HCl": 19376,
+      "ĠFish": 19377,
+      "Ġphenomenological": 19378,
+      "Ġsalivary": 19379,
+      "ondo": 19380,
+      "Ġnotions": 19381,
+      "fur": 19382,
+      "UCT": 19383,
+      "Ġwww": 19384,
+      "abet": 19385,
+      "ĠSulf": 19386,
+      "Fil": 19387,
+      "dominated": 19388,
+      "arser": 19389,
+      "Ġpackages": 19390,
+      "Ġsplice": 19391,
+      "Flo": 19392,
+      "NOWLED": 19393,
+      "xa": 19394,
+      "ĠYuan": 19395,
+      "Ġacetone": 19396,
+      "ĠVitamin": 19397,
+      "ĠÎŀ": 19398,
+      "Ġobsc": 19399,
+      "Ġchaper": 19400,
+      "Ġmort": 19401,
+      "MAN": 19402,
+      "Ġsubtilis": 19403,
+      "Ġoptimality": 19404,
+      "Ġcontinuing": 19405,
+      "Ġduplication": 19406,
+      "Ġmultiplying": 19407,
+      "Ġimmunological": 19408,
+      "Ġcirrhosis": 19409,
+      "hospital": 19410,
+      "ĠProbabilistic": 19411,
+      "Ġdeletions": 19412,
+      "Ġcaution": 19413,
+      "Ġowner": 19414,
+      "oxorubicin": 19415,
+      "Ġlaunch": 19416,
+      "Ġcure": 19417,
+      "thus": 19418,
+      "ĠHermitian": 19419,
+      "canonical": 19420,
+      "Ġimmunore": 19421,
+      "formin": 19422,
+      "Ġbroadband": 19423,
+      "partum": 19424,
+      "ophe": 19425,
+      "ĠBeta": 19426,
+      "ĠBI": 19427,
+      "Ġïĺº": 19428,
+      "Ġjumps": 19429,
+      "Ġparadox": 19430,
+      "umped": 19431,
+      "Ġdoctors": 19432,
+      "Ġhospitalized": 19433,
+      "Ġwash": 19434,
+      "precision": 19435,
+      "Ġruled": 19436,
+      "Ġduplicate": 19437,
+      "ante": 19438,
+      "Ġneurotrans": 19439,
+      "Ġïĥ§": 19440,
+      "Ġtheme": 19441,
+      "Taking": 19442,
+      "ĠPlants": 19443,
+      "following": 19444,
+      "Ġageing": 19445,
+      "Ġcongestion": 19446,
+      "osarcoma": 19447,
+      "Ġrepository": 19448,
+      "ĠHess": 19449,
+      "ĠCatalytic": 19450,
+      "ĠDV": 19451,
+      "INK": 19452,
+      "priv": 19453,
+      "ĠAna": 19454,
+      "ĠSLE": 19455,
+      "ĠThailand": 19456,
+      "íķ": 19457,
+      "Ġduty": 19458,
+      "locations": 19459,
+      "oter": 19460,
+      "Ġlysine": 19461,
+      "Ġindist": 19462,
+      "Ġagonists": 19463,
+      "Ack": 19464,
+      "Ġminimally": 19465,
+      "Ġetching": 19466,
+      "ugging": 19467,
+      "cuda": 19468,
+      "ndef": 19469,
+      "Ġreferring": 19470,
+      "Ġlysates": 19471,
+      "Ġserotonin": 19472,
+      "cribing": 19473,
+      "ĠInterface": 19474,
+      "dV": 19475,
+      "Ġdurations": 19476,
+      "Ġphotod": 19477,
+      "Ġdating": 19478,
+      "Ġirreversible": 19479,
+      "osidase": 19480,
+      "ĠFROM": 19481,
+      "within": 19482,
+      "SNR": 19483,
+      "Ġarrhyth": 19484,
+      "ĠRatio": 19485,
+      "ĠThin": 19486,
+      "centered": 19487,
+      "Ġshocks": 19488,
+      "ĠVers": 19489,
+      "Ġnoticeable": 19490,
+      "Ġfoci": 19491,
+      "Ġorthonormal": 19492,
+      "ĠâİŁ": 19493,
+      "Ġluminescence": 19494,
+      "ĠSUSY": 19495,
+      "internal": 19496,
+      "ĠTour": 19497,
+      "Ġabbrevi": 19498,
+      "ĠMAL": 19499,
+      "vertex": 19500,
+      "Ġemploys": 19501,
+      "INS": 19502,
+      "Ġimmunohistochemistry": 19503,
+      "Ġheparin": 19504,
+      "Ġidiopathic": 19505,
+      "Ġimmobilized": 19506,
+      "ishe": 19507,
+      "phth": 19508,
+      "thin": 19509,
+      "ĠStorage": 19510,
+      "Ġperovskite": 19511,
+      "Prot": 19512,
+      "ĠDepending": 19513,
+      "Ġblends": 19514,
+      "Ġpredator": 19515,
+      "Ġdisplaying": 19516,
+      "Ġvesicle": 19517,
+      "ĠKra": 19518,
+      "Ġlane": 19519,
+      "Ġmultilayer": 19520,
+      "Ġhomozygous": 19521,
+      "cosh": 19522,
+      "Ġsuperficial": 19523,
+      "Ġil": 19524,
+      "ĠKR": 19525,
+      "ĠBrun": 19526,
+      "ĠEW": 19527,
+      "opa": 19528,
+      "ĠCartesian": 19529,
+      "ĠCytoplas": 19530,
+      "ĠPennsylvan": 19531,
+      "bands": 19532,
+      "Ġangiotensin": 19533,
+      "ĠLattice": 19534,
+      "GI": 19535,
+      "jee": 19536,
+      "Ġenlarged": 19537,
+      "enius": 19538,
+      "ĠIa": 19539,
+      "oux": 19540,
+      "Ġgent": 19541,
+      "Ġcarbonyl": 19542,
+      "chers": 19543,
+      "Ġhypothe": 19544,
+      "Ġmicrosp": 19545,
+      "Ġaffective": 19546,
+      "Ġaxons": 19547,
+      "ei": 19548,
+      "yptoph": 19549,
+      "ĠJon": 19550,
+      "queue": 19551,
+      "ĠGauge": 19552,
+      "menopausal": 19553,
+      "ĠDas": 19554,
+      "ĠEssential": 19555,
+      "ĠFault": 19556,
+      "ĠBil": 19557,
+      "Ġtestosterone": 19558,
+      "Ġchambers": 19559,
+      "dione": 19560,
+      "Ġelicited": 19561,
+      "IGN": 19562,
+      "Ġantioxidants": 19563,
+      "populations": 19564,
+      "Ġovary": 19565,
+      "Ġâĸ": 19566,
+      "Ġabstraction": 19567,
+      "Ġhydrocarbons": 19568,
+      "Ġrectal": 19569,
+      "Ġtriggering": 19570,
+      "Ġthoroughly": 19571,
+      "Run": 19572,
+      "acteria": 19573,
+      "information": 19574,
+      "ĠBed": 19575,
+      "Ġquenc": 19576,
+      "Ġunders": 19577,
+      "ĠScotland": 19578,
+      "Ġrevolution": 19579,
+      "Ġpituitary": 19580,
+      "Ġanthropogenic": 19581,
+      "focus": 19582,
+      "Ġmethan": 19583,
+      "Ġinflow": 19584,
+      "Ġdeflection": 19585,
+      "ĠCape": 19586,
+      "Ġmultidimensional": 19587,
+      "Ġarrived": 19588,
+      "ĠSpar": 19589,
+      "dv": 19590,
+      "Ġcows": 19591,
+      "ĠBh": 19592,
+      "Ġjk": 19593,
+      "tolyl": 19594,
+      "Ġeigenstates": 19595,
+      "Ġpreprocessing": 19596,
+      "ĠRain": 19597,
+      "ä¸": 19598,
+      "inz": 19599,
+      "Ġmn": 19600,
+      "REE": 19601,
+      "atrick": 19602,
+      "Dev": 19603,
+      "Ġfulfilled": 19604,
+      "Ġartic": 19605,
+      "Ġrealizations": 19606,
+      "ĠComponent": 19607,
+      "ĠWS": 19608,
+      "Ġinfo": 19609,
+      "printed": 19610,
+      "atosis": 19611,
+      "cache": 19612,
+      "anov": 19613,
+      "ĠTg": 19614,
+      "content": 19615,
+      "junc": 19616,
+      "ĠCDK": 19617,
+      "Ġbehaves": 19618,
+      "ĠKid": 19619,
+      "difference": 19620,
+      "ĠPs": 19621,
+      "ĠUg": 19622,
+      "Ġstructurally": 19623,
+      "erebral": 19624,
+      "ĠSurve": 19625,
+      "heal": 19626,
+      "onite": 19627,
+      "Ġdeleted": 19628,
+      "itim": 19629,
+      "Star": 19630,
+      "ĠSpeech": 19631,
+      "ĠAstr": 19632,
+      "gradient": 19633,
+      "Ġfellow": 19634,
+      "Ġsyring": 19635,
+      "NB": 19636,
+      "ĠNB": 19637,
+      "Ġcreep": 19638,
+      "Ġlogging": 19639,
+      "Ġinten": 19640,
+      "scalar": 19641,
+      "ĠAtmospheric": 19642,
+      "Ġlupus": 19643,
+      "Ġidentically": 19644,
+      "processed": 19645,
+      "signal": 19646,
+      "ĠClostr": 19647,
+      "ancers": 19648,
+      "Ġdb": 19649,
+      "Ġsubsystem": 19650,
+      "situ": 19651,
+      "Ġferroelectric": 19652,
+      "ĠïĽľ": 19653,
+      "Ġore": 19654,
+      "ĠRb": 19655,
+      "ĠMicrosoft": 19656,
+      "ĠCoch": 19657,
+      "ĠActin": 19658,
+      "Ġnerves": 19659,
+      "Ġexpertise": 19660,
+      "otive": 19661,
+      "ĠPoincaré": 19662,
+      "ĠRig": 19663,
+      "Ġpsychosocial": 19664,
+      "Ġprogenitors": 19665,
+      "ĠMyr": 19666,
+      "ĠHug": 19667,
+      "Ġbiogenesis": 19668,
+      "Ġincorporates": 19669,
+      "Ġnevertheless": 19670,
+      "ĠDecl": 19671,
+      "observ": 19672,
+      "Ġmultiplier": 19673,
+      "Ġresponding": 19674,
+      "hoff": 19675,
+      "Ġimpacted": 19676,
+      "Ġsyndromes": 19677,
+      "kel": 19678,
+      "ĠSynt": 19679,
+      "ĠConcer": 19680,
+      "ĠAmericans": 19681,
+      "Ġspaced": 19682,
+      "umption": 19683,
+      "ĠThompson": 19684,
+      "ĠJacobian": 19685,
+      "Tra": 19686,
+      "evolution": 19687,
+      "Ġdidn": 19688,
+      "Ġpercentile": 19689,
+      "Ġlid": 19690,
+      "equivalent": 19691,
+      "Ġantico": 19692,
+      "Ġmultiply": 19693,
+      "Ġpenicillin": 19694,
+      "Ġresponsiveness": 19695,
+      "Ġrunoff": 19696,
+      "alanine": 19697,
+      "squares": 19698,
+      "ĠInsulin": 19699,
+      "rele": 19700,
+      "ĠLif": 19701,
+      "ĠMinkowski": 19702,
+      "Ġblend": 19703,
+      "ĠPand": 19704,
+      "Ġtwelve": 19705,
+      "Ġhybrids": 19706,
+      "Ġbass": 19707,
+      "interaction": 19708,
+      "ĠBangladesh": 19709,
+      "Ġopens": 19710,
+      "ĠArts": 19711,
+      "Ġconcave": 19712,
+      "Ġpedest": 19713,
+      "Ġfist": 19714,
+      "ĠAdults": 19715,
+      "openia": 19716,
+      "ENCE": 19717,
+      "ĠFusion": 19718,
+      "Ġmicroc": 19719,
+      "ĠSurgical": 19720,
+      "ylate": 19721,
+      "Ġpackaging": 19722,
+      "OCK": 19723,
+      "QC": 19724,
+      "Tri": 19725,
+      "scan": 19726,
+      "Ġregards": 19727,
+      "Ġdiscriminant": 19728,
+      "Ġindustries": 19729,
+      "icus": 19730,
+      "ĠWalker": 19731,
+      "Ġpeers": 19732,
+      "synt": 19733,
+      "Ġhorse": 19734,
+      "Ġflowing": 19735,
+      "urred": 19736,
+      "ĠCRP": 19737,
+      "ĠCareer": 19738,
+      "iffiffiffiffiffiffiffiff": 19739,
+      "ĠMSE": 19740,
+      "hana": 19741,
+      "ĠMortality": 19742,
+      "Ġtumorigenesis": 19743,
+      "ĠIslam": 19744,
+      "Ġazimuthal": 19745,
+      "wen": 19746,
+      "Ġsys": 19747,
+      "azin": 19748,
+      "neighbor": 19749,
+      "Config": 19750,
+      "they": 19751,
+      "Ġsorption": 19752,
+      "Ġspanned": 19753,
+      "Ġviewpoint": 19754,
+      "MOD": 19755,
+      "Ġthrust": 19756,
+      "uplex": 19757,
+      "Ġhistograms": 19758,
+      "Ġprogrammed": 19759,
+      "Ġethics": 19760,
+      "ectable": 19761,
+      "representation": 19762,
+      "umns": 19763,
+      "Ġstreet": 19764,
+      "ĠSobolev": 19765,
+      "Ġexcision": 19766,
+      "ĠRud": 19767,
+      "quires": 19768,
+      "Ġowned": 19769,
+      "Ġthousand": 19770,
+      "Ġantagonists": 19771,
+      "UST": 19772,
+      "Ġdrastically": 19773,
+      "ĠóµĦ©": 19774,
+      "ĠDor": 19775,
+      "ĠMOS": 19776,
+      "pn": 19777,
+      "ĠDecre": 19778,
+      "Dep": 19779,
+      "Ġsintering": 19780,
+      "Ġpurple": 19781,
+      "ethanol": 19782,
+      "Ġhydrocarbon": 19783,
+      "ĠFO": 19784,
+      "leftrightarrow": 19785,
+      "Ġimmunofluorescence": 19786,
+      "ĠOM": 19787,
+      "Ġmaturity": 19788,
+      "Ġearthquakes": 19789,
+      "Ġaxon": 19790,
+      "Ġprobed": 19791,
+      "ORD": 19792,
+      "ĠADP": 19793,
+      "sg": 19794,
+      "omere": 19795,
+      "Ġtranscribed": 19796,
+      "Mar": 19797,
+      "ĠUtil": 19798,
+      "ĠIA": 19799,
+      "Ġcompiled": 19800,
+      "Ġsupervision": 19801,
+      "ĠXen": 19802,
+      "ĠJur": 19803,
+      "compar": 19804,
+      "Ġhypertensive": 19805,
+      "ilized": 19806,
+      "rae": 19807,
+      "Conclusion": 19808,
+      "'''": 19809,
+      "Double": 19810,
+      "ĠFas": 19811,
+      "Ġinsectic": 19812,
+      "ĠPrem": 19813,
+      "Pri": 19814,
+      "ĠCao": 19815,
+      "ĠQuestionnaire": 19816,
+      "Ġgathered": 19817,
+      "GW": 19818,
+      "ĠNV": 19819,
+      "ĠLactobacillus": 19820,
+      "Ġcyclin": 19821,
+      "Ġreject": 19822,
+      "Ġskull": 19823,
+      "Ġaw": 19824,
+      "ĠCold": 19825,
+      "Ġmesons": 19826,
+      "bd": 19827,
+      "Ġdetrimental": 19828,
+      "apore": 19829,
+      "nowled": 19830,
+      "ĠCXCL": 19831,
+      "Ġspikes": 19832,
+      "Ġtent": 19833,
+      "ĠLength": 19834,
+      "Ġdoor": 19835,
+      "Ġflour": 19836,
+      "ustration": 19837,
+      "Health": 19838,
+      "Ġtransparency": 19839,
+      "Ġdisrupted": 19840,
+      "Hy": 19841,
+      "overl": 19842,
+      "ĠReinforcement": 19843,
+      "ceptors": 19844,
+      "ĠKos": 19845,
+      "retroviral": 19846,
+      "ĠINT": 19847,
+      "ĠSor": 19848,
+      "Ġadopting": 19849,
+      "Ġendoplasmic": 19850,
+      "Ġsuit": 19851,
+      "Ġopioid": 19852,
+      "Ġintegrin": 19853,
+      "away": 19854,
+      "Ġtailored": 19855,
+      "ĠSoc": 19856,
+      "Ġquies": 19857,
+      "Ġhusband": 19858,
+      "Ġumb": 19859,
+      "ĠCai": 19860,
+      "ĠAspergillus": 19861,
+      "ĠGaN": 19862,
+      "Ġdistinguishing": 19863,
+      "Ġextrapolation": 19864,
+      "Ġcage": 19865,
+      "Ġscavenging": 19866,
+      "KF": 19867,
+      "Tree": 19868,
+      "ĠConflict": 19869,
+      "UNC": 19870,
+      "Ġmanganese": 19871,
+      "days": 19872,
+      "ÃŁ": 19873,
+      "ĠLive": 19874,
+      "sd": 19875,
+      "ractor": 19876,
+      "Ġlute": 19877,
+      "Ġdissimilar": 19878,
+      "Ġib": 19879,
+      "ĠVeg": 19880,
+      "Ġoccurrences": 19881,
+      "Ġbinomial": 19882,
+      "Scheme": 19883,
+      "Ġtape": 19884,
+      "ĠCant": 19885,
+      "Ġelectrosp": 19886,
+      "Cd": 19887,
+      "made": 19888,
+      "Ġsevent": 19889,
+      "shared": 19890,
+      "Ġaccession": 19891,
+      "orp": 19892,
+      "DATA": 19893,
+      "leted": 19894,
+      "Vari": 19895,
+      "Ġrose": 19896,
+      "tagged": 19897,
+      "ĠAth": 19898,
+      "Ġeddy": 19899,
+      "estone": 19900,
+      "Ġesters": 19901,
+      "Ġtyping": 19902,
+      "ĠStudents": 19903,
+      "yi": 19904,
+      "oresistance": 19905,
+      "inois": 19906,
+      "Ġglucocortic": 19907,
+      "iosis": 19908,
+      "Ġcoronal": 19909,
+      "Ġsheath": 19910,
+      "ĠTrack": 19911,
+      "Ġequilibria": 19912,
+      "amming": 19913,
+      "Ġpione": 19914,
+      "Ġsciences": 19915,
+      "Ġsuppressing": 19916,
+      "Ġdeco": 19917,
+      "ifndef": 19918,
+      "His": 19919,
+      "Ġpellet": 19920,
+      "Linear": 19921,
+      "orbent": 19922,
+      "Ġflatten": 19923,
+      "Ġstraw": 19924,
+      "Ġalbeit": 19925,
+      "ĠPredictive": 19926,
+      "Ġgaze": 19927,
+      "Ġhydroly": 19928,
+      "uther": 19929,
+      "oders": 19930,
+      "Ġflap": 19931,
+      "Ġsimplicial": 19932,
+      "System": 19933,
+      "Ġstressed": 19934,
+      "Ġimmunoglobulin": 19935,
+      "ilia": 19936,
+      "Ġconsuming": 19937,
+      "Ġé": 19938,
+      "galact": 19939,
+      "Ġadulthood": 19940,
+      "Ġvorticity": 19941,
+      "yclic": 19942,
+      "ovoltaic": 19943,
+      "ivestock": 19944,
+      "Ġbeds": 19945,
+      "ĠPlanning": 19946,
+      "Ġparameterized": 19947,
+      "Ġghost": 19948,
+      "maximum": 19949,
+      "Ġsuperim": 19950,
+      "Ġphysicochemical": 19951,
+      "gp": 19952,
+      "ongue": 19953,
+      "Ġprimordial": 19954,
+      "xff": 19955,
+      "insula": 19956,
+      "Mc": 19957,
+      "Ġminimizes": 19958,
+      "ĠGravitational": 19959,
+      "osoma": 19960,
+      "ignificant": 19961,
+      "Ġelucidated": 19962,
+      "Ġsubsurface": 19963,
+      "significant": 19964,
+      "Ġrelatives": 19965,
+      "ferroni": 19966,
+      "transf": 19967,
+      "Ġtails": 19968,
+      "beck": 19969,
+      "omagnetic": 19970,
+      "Ġunnecessary": 19971,
+      "Ġmonomial": 19972,
+      "delay": 19973,
+      "Ġsta": 19974,
+      "ĠSuz": 19975,
+      "Ġaltering": 19976,
+      "LOG": 19977,
+      "ĠLac": 19978,
+      "Ġranks": 19979,
+      "hw": 19980,
+      "ĠNep": 19981,
+      "Ġneuropath": 19982,
+      "ĠCompe": 19983,
+      "Gr": 19984,
+      "Pati": 19985,
+      "reduce": 19986,
+      "ĠMalaysia": 19987,
+      "ceral": 19988,
+      "Ġmicrobes": 19989,
+      "Ġlensing": 19990,
+      "ĠCalcium": 19991,
+      "ĠDetermin": 19992,
+      "ĠCosta": 19993,
+      "Ġkeeps": 19994,
+      "printing": 19995,
+      "ĉĉĉĉĉĉ": 19996,
+      "chin": 19997,
+      "exposed": 19998,
+      "Ġperiodically": 19999,
+      "Ġrender": 20000,
+      "ĠCardiovascular": 20001,
+      "entin": 20002,
+      "Ġbioavailability": 20003,
+      "Ġinterpretations": 20004,
+      "ĠCU": 20005,
+      "Ġnegoti": 20006,
+      "Ġantim": 20007,
+      "Ġdeemed": 20008,
+      "Ġae": 20009,
+      "Ġhalos": 20010,
+      "ĠMichigan": 20011,
+      "Ġosteoarthritis": 20012,
+      "diag": 20013,
+      "ĠBeng": 20014,
+      "Ġmetagen": 20015,
+      "Ġparameterization": 20016,
+      "diagn": 20017,
+      "ĠMatching": 20018,
+      "Ġcatalysis": 20019,
+      "uts": 20020,
+      "Ġdissemination": 20021,
+      "Ġoutlet": 20022,
+      "ĠMoon": 20023,
+      "ĠGST": 20024,
+      "sphere": 20025,
+      "Ġresearcher": 20026,
+      "ambiguation": 20027,
+      "Ġraises": 20028,
+      "Ġflavonoids": 20029,
+      "ĠMultivariate": 20030,
+      "Ġaccl": 20031,
+      "WI": 20032,
+      "Ġnu": 20033,
+      "Ġergodic": 20034,
+      "unique": 20035,
+      "atinib": 20036,
+      "Ġresolutions": 20037,
+      "Ġhouses": 20038,
+      "DEC": 20039,
+      "ighed": 20040,
+      "Ġsixth": 20041,
+      "Ġpolitician": 20042,
+      "apache": 20043,
+      "Ġsolute": 20044,
+      "Ġaugment": 20045,
+      "stress": 20046,
+      "HIV": 20047,
+      "ĠSets": 20048,
+      "Ġtransistors": 20049,
+      "qubit": 20050,
+      "amines": 20051,
+      "Ġfarmers": 20052,
+      "Ġnt": 20053,
+      "ĠLagrange": 20054,
+      "Ġvegetables": 20055,
+      "Ġpret": 20056,
+      "ĠSynthetic": 20057,
+      "Ġcones": 20058,
+      "Ġmedicines": 20059,
+      "Ġgenomics": 20060,
+      "Ġexperiencing": 20061,
+      "agland": 20062,
+      "Ġgenital": 20063,
+      "ĠObservatory": 20064,
+      "ĠSkin": 20065,
+      "ĠRosen": 20066,
+      "ĠBritain": 20067,
+      "genome": 20068,
+      "ĠEntropy": 20069,
+      "Ġrac": 20070,
+      "Go": 20071,
+      "Ġwalks": 20072,
+      "criptor": 20073,
+      "ĠBaker": 20074,
+      "oker": 20075,
+      "Ġpropensity": 20076,
+      "Ġpopularity": 20077,
+      "restricted": 20078,
+      "ĠBert": 20079,
+      "before": 20080,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 20081,
+      "auto": 20082,
+      "Rank": 20083,
+      "ĠRCT": 20084,
+      "Ġpocket": 20085,
+      "obut": 20086,
+      "Ġbenzene": 20087,
+      "ĠCNT": 20088,
+      "yptophan": 20089,
+      "allis": 20090,
+      "ĠResources": 20091,
+      "ĠBerlin": 20092,
+      "Ġscholar": 20093,
+      "glob": 20094,
+      "ĠSpeed": 20095,
+      "ĠXiao": 20096,
+      "biggl": 20097,
+      "ANCE": 20098,
+      "ĠPrime": 20099,
+      "Phys": 20100,
+      "idia": 20101,
+      "Ġmonoc": 20102,
+      "ĠCommunications": 20103,
+      "ĠPrecision": 20104,
+      "ĠPauli": 20105,
+      "Ġinvestigators": 20106,
+      "ĠLiang": 20107,
+      "Ġmeteorological": 20108,
+      "mog": 20109,
+      "reens": 20110,
+      "ubric": 20111,
+      "Ġrearrangement": 20112,
+      "orta": 20113,
+      "Elect": 20114,
+      "ĠTukey": 20115,
+      "ĠMis": 20116,
+      "Ġepiderm": 20117,
+      "ĠACKNOWLED": 20118,
+      "wart": 20119,
+      "Ġexciton": 20120,
+      "Ġassociative": 20121,
+      "styrene": 20122,
+      "Ġlosing": 20123,
+      "ĠOd": 20124,
+      "prep": 20125,
+      "essation": 20126,
+      "Ġattributable": 20127,
+      "ĠNavier": 20128,
+      "anz": 20129,
+      "Ġcorrectness": 20130,
+      "oints": 20131,
+      "ĠRather": 20132,
+      "Ġassemblies": 20133,
+      "Ġbridges": 20134,
+      "OSS": 20135,
+      "MET": 20136,
+      "Ġperm": 20137,
+      "Ġauthorities": 20138,
+      "Ġiodine": 20139,
+      "shire": 20140,
+      "interval": 20141,
+      "eptid": 20142,
+      "Ġpotency": 20143,
+      "Ġrenewable": 20144,
+      "vard": 20145,
+      "Ġsurjective": 20146,
+      "Ġsubsequence": 20147,
+      "ĠEVs": 20148,
+      "itching": 20149,
+      "Ġgenotyping": 20150,
+      "ĠAccurate": 20151,
+      "iophene": 20152,
+      "Gly": 20153,
+      "plified": 20154,
+      "ĠDistinct": 20155,
+      "ACH": 20156,
+      "Ġspeakers": 20157,
+      "holm": 20158,
+      "Ġpros": 20159,
+      "ĠDevice": 20160,
+      "mc": 20161,
+      "ĠDense": 20162,
+      "ĠVa": 20163,
+      "rison": 20164,
+      "Ġacyl": 20165,
+      "ĠPrincipal": 20166,
+      "ĠViral": 20167,
+      "Ġcosine": 20168,
+      "ĠResidual": 20169,
+      "Ġefflux": 20170,
+      "ĠSubjects": 20171,
+      "Ġrectangle": 20172,
+      "workers": 20173,
+      "Ġrotated": 20174,
+      "Ġbomb": 20175,
+      "ĠResolution": 20176,
+      "near": 20177,
+      "Ġ®": 20178,
+      "Ġestablishes": 20179,
+      "amed": 20180,
+      "Ġcompetence": 20181,
+      "Glu": 20182,
+      "ĠDend": 20183,
+      "ĠHsp": 20184,
+      "ensation": 20185,
+      "ĠLead": 20186,
+      "Ġlogger": 20187,
+      "sinh": 20188,
+      "Ġintellectual": 20189,
+      "former": 20190,
+      "Ce": 20191,
+      "Ġmonocyte": 20192,
+      "hores": 20193,
+      "Ġdiastolic": 20194,
+      "Ġlifespan": 20195,
+      "ĠSilva": 20196,
+      "arum": 20197,
+      "Ġtransducer": 20198,
+      "Ġoutgoing": 20199,
+      "entation": 20200,
+      "Ġabsorbing": 20201,
+      "itage": 20202,
+      "Ġsynthesize": 20203,
+      "Ġfeeling": 20204,
+      "asian": 20205,
+      "Ġceramics": 20206,
+      "iph": 20207,
+      "Ġnonlocal": 20208,
+      "Part": 20209,
+      "Ġimmersed": 20210,
+      "stationary": 20211,
+      "lecting": 20212,
+      "Ġwelding": 20213,
+      "Ġresembles": 20214,
+      "ĠKat": 20215,
+      "master": 20216,
+      "Ġintersect": 20217,
+      "ĠOlig": 20218,
+      "ĠTrends": 20219,
+      "agh": 20220,
+      "ĠNav": 20221,
+      "ĠTu": 20222,
+      "Ġepist": 20223,
+      "Ġclinics": 20224,
+      "Ġrepresentatives": 20225,
+      "Ġgrateful": 20226,
+      "GPIO": 20227,
+      "HH": 20228,
+      "Ġunambig": 20229,
+      "tuning": 20230,
+      "Ġnewsp": 20231,
+      "cohol": 20232,
+      "################################": 20233,
+      "%%%%%%%%": 20234,
+      "represented": 20235,
+      "ocic": 20236,
+      "ĠFuk": 20237,
+      "ĠSund": 20238,
+      "hasone": 20239,
+      "Mode": 20240,
+      "olone": 20241,
+      "ĠSb": 20242,
+      "Three": 20243,
+      "Link": 20244,
+      "cephal": 20245,
+      "ĠKap": 20246,
+      "Ġeliminating": 20247,
+      "Ġmelanogaster": 20248,
+      "âŁ": 20249,
+      "ĠBMD": 20250,
+      "ISE": 20251,
+      "ĠBattle": 20252,
+      "Ġshrinkage": 20253,
+      "ĠSeven": 20254,
+      "ĠGlass": 20255,
+      "romagn": 20256,
+      "Ġkl": 20257,
+      "ĠObviously": 20258,
+      "preserving": 20259,
+      "ĠPlatform": 20260,
+      "ĠÌĩ": 20261,
+      "omavirus": 20262,
+      "ĠEight": 20263,
+      "Ġallerg": 20264,
+      "ĠNanoparticles": 20265,
+      "aryl": 20266,
+      "Ġpriors": 20267,
+      "pattern": 20268,
+      "Ġlinearity": 20269,
+      "Ġtruly": 20270,
+      "Process": 20271,
+      "Ġdescending": 20272,
+      "ĠVictoria": 20273,
+      "cond": 20274,
+      "ĠICP": 20275,
+      "orescent": 20276,
+      "Ġauthority": 20277,
+      "Ġmock": 20278,
+      "igmoid": 20279,
+      "Ġcomorbidities": 20280,
+      "simple": 20281,
+      "Ġblo": 20282,
+      "ĠCompute": 20283,
+      "Ġgestation": 20284,
+      "achusetts": 20285,
+      "Ġphantom": 20286,
+      "ĠEdward": 20287,
+      "ĠFBS": 20288,
+      "factors": 20289,
+      "ĠEstimates": 20290,
+      "clear": 20291,
+      "WB": 20292,
+      "products": 20293,
+      "numpy": 20294,
+      "brief": 20295,
+      "Ġshop": 20296,
+      "ĠPoli": 20297,
+      "ĠRespiratory": 20298,
+      "Ġsurprisingly": 20299,
+      "Ġnanocomposite": 20300,
+      "dividual": 20301,
+      "Ġholographic": 20302,
+      "ygdala": 20303,
+      "roplasty": 20304,
+      "otactic": 20305,
+      "ĠPennsylvania": 20306,
+      "ĠScore": 20307,
+      "Obj": 20308,
+      "Ġstories": 20309,
+      "Ġmaximizing": 20310,
+      "Ġgelatin": 20311,
+      "rites": 20312,
+      "ĠTau": 20313,
+      "Ġtrypsin": 20314,
+      "Ġith": 20315,
+      "Ġfaint": 20316,
+      "Ġpriming": 20317,
+      "eworthy": 20318,
+      "ĠInverse": 20319,
+      "Ġknots": 20320,
+      "sharp": 20321,
+      "Ġtrains": 20322,
+      "Ġcredit": 20323,
+      "ĠBelow": 20324,
+      "pixel": 20325,
+      "Ġspindle": 20326,
+      "ĠPast": 20327,
+      "Ġenumerate": 20328,
+      "olateral": 20329,
+      "Ġattending": 20330,
+      "Ġquantized": 20331,
+      "Ġhaplotypes": 20332,
+      "encl": 20333,
+      "Ġwaven": 20334,
+      "Ġfurthermore": 20335,
+      "Ġchallenged": 20336,
+      "Ġmanufactured": 20337,
+      "ipheral": 20338,
+      "Ġinfinites": 20339,
+      "ĠRand": 20340,
+      "Ġstaging": 20341,
+      "agan": 20342,
+      "Ġperox": 20343,
+      "trifluor": 20344,
+      "ĠMcK": 20345,
+      "ĠFOX": 20346,
+      "ĠLank": 20347,
+      "ĠLuo": 20348,
+      "ĠAnth": 20349,
+      "ibrio": 20350,
+      "yel": 20351,
+      "ĠJi": 20352,
+      "ĠIO": 20353,
+      "ĠBridge": 20354,
+      "ĠRow": 20355,
+      "Ġcompensated": 20356,
+      "atsu": 20357,
+      "Ġhypothetical": 20358,
+      "Ġterminals": 20359,
+      "Ġcobalt": 20360,
+      "mers": 20361,
+      "ĠMang": 20362,
+      "NI": 20363,
+      "ĠRac": 20364,
+      "ALS": 20365,
+      "fen": 20366,
+      "ĠUb": 20367,
+      "Ġpredation": 20368,
+      "cadherin": 20369,
+      "ĠShanghai": 20370,
+      "Ġtries": 20371,
+      "Ġsport": 20372,
+      "acrylate": 20373,
+      "ĠAlgebraic": 20374,
+      "aints": 20375,
+      "Expr": 20376,
+      "Ġandrogen": 20377,
+      "Ġwedge": 20378,
+      "disp": 20379,
+      "Ġstirred": 20380,
+      "ĠAle": 20381,
+      "Ġcock": 20382,
+      "Four": 20383,
+      "Ġscanner": 20384,
+      "Ġplasmon": 20385,
+      "ĠGender": 20386,
+      "ĠRecord": 20387,
+      "ĠInjury": 20388,
+      "oblastic": 20389,
+      "ĠFluorescence": 20390,
+      "Ġantidepress": 20391,
+      "Ġdefinitive": 20392,
+      "Ġrepression": 20393,
+      "ordinates": 20394,
+      "Ġangiography": 20395,
+      "ĠHelical": 20396,
+      "Ġcancellation": 20397,
+      "release": 20398,
+      "Ġrelational": 20399,
+      "ĠAndre": 20400,
+      "molecule": 20401,
+      "Ġshaping": 20402,
+      "ĠDenmark": 20403,
+      "ĠALS": 20404,
+      "ĠNW": 20405,
+      "overrightarrow": 20406,
+      "Ġcombat": 20407,
+      "boxes": 20408,
+      "subject": 20409,
+      "Ġnanoscale": 20410,
+      "Ġcanine": 20411,
+      "Ġsaving": 20412,
+      "Ġstrategic": 20413,
+      "Stat": 20414,
+      "ĠDub": 20415,
+      "Ġpermitted": 20416,
+      "ĠTwitter": 20417,
+      "âĶ": 20418,
+      "Ġmemories": 20419,
+      "ĠBusiness": 20420,
+      "adays": 20421,
+      "Ġpooling": 20422,
+      "ĠClusters": 20423,
+      "imide": 20424,
+      "ounters": 20425,
+      "fraction": 20426,
+      "ĠCliff": 20427,
+      "Cam": 20428,
+      "Even": 20429,
+      "KY": 20430,
+      "kit": 20431,
+      "ibrated": 20432,
+      "Ġaccompanying": 20433,
+      "anus": 20434,
+      "Ġbuoy": 20435,
+      "Ġproliferative": 20436,
+      "Ġproc": 20437,
+      "Ġstabilizing": 20438,
+      "ĠNamely": 20439,
+      "posp": 20440,
+      "soon": 20441,
+      "Ġaberrant": 20442,
+      "Ġinterstellar": 20443,
+      "Overall": 20444,
+      "ĠGn": 20445,
+      "ĠFeedback": 20446,
+      "Ġoracle": 20447,
+      "Ġprenatal": 20448,
+      "commun": 20449,
+      "Ġoutbreaks": 20450,
+      "Ġfertilization": 20451,
+      "ĠMAG": 20452,
+      "Ġsinger": 20453,
+      "ĠMicrowave": 20454,
+      "ĠParliament": 20455,
+      "casting": 20456,
+      "General": 20457,
+      "algorithm": 20458,
+      "Ġphrase": 20459,
+      "Ġavian": 20460,
+      "ĠPLA": 20461,
+      "Ġhardly": 20462,
+      "approximately": 20463,
+      "ARCH": 20464,
+      "Ġtransc": 20465,
+      "Ġdecomp": 20466,
+      "contin": 20467,
+      "ĠMilky": 20468,
+      "Ġherpes": 20469,
+      "Range": 20470,
+      "OFF": 20471,
+      "prisingly": 20472,
+      "lx": 20473,
+      "ĠABA": 20474,
+      "Ġshore": 20475,
+      "Ġderiving": 20476,
+      "Ġpellets": 20477,
+      "nowledg": 20478,
+      "Item": 20479,
+      "stranded": 20480,
+      "built": 20481,
+      "Glc": 20482,
+      "quist": 20483,
+      "ĠSubstrate": 20484,
+      "Ġtraditionally": 20485,
+      "ĠMount": 20486,
+      "ivalence": 20487,
+      "axation": 20488,
+      "Ġlocate": 20489,
+      "Ġgun": 20490,
+      "Ġvocabulary": 20491,
+      "ĠPolym": 20492,
+      "Ġect": 20493,
+      "Ġmult": 20494,
+      "Ġsedimentary": 20495,
+      "Ġautocorrelation": 20496,
+      "ĠSympt": 20497,
+      "Ġterritory": 20498,
+      "Ġexcitatory": 20499,
+      "Ġvote": 20500,
+      "Ġhered": 20501,
+      "acea": 20502,
+      "ĠFocus": 20503,
+      "ampling": 20504,
+      "ffee": 20505,
+      "Ġprimes": 20506,
+      "ĠMaking": 20507,
+      "irs": 20508,
+      "MPs": 20509,
+      "Ġlitter": 20510,
+      "amethasone": 20511,
+      "ĠkJ": 20512,
+      "Ġsecretory": 20513,
+      "Ġcostly": 20514,
+      "Ġpartnership": 20515,
+      "ĠBacteria": 20516,
+      "Ġperoxidation": 20517,
+      "stroke": 20518,
+      "ĠSav": 20519,
+      "ĠBW": 20520,
+      "Ġconnects": 20521,
+      "Ġamine": 20522,
+      "ril": 20523,
+      "Ġbattle": 20524,
+      "ĠNotes": 20525,
+      "ĠProvid": 20526,
+      "ĠInstitutional": 20527,
+      "Ġpropri": 20528,
+      "fan": 20529,
+      "Ġpun": 20530,
+      "romb": 20531,
+      "vities": 20532,
+      "ĠCAM": 20533,
+      "ĠIsh": 20534,
+      "ĠHN": 20535,
+      "ĠRecomb": 20536,
+      "sche": 20537,
+      "Ġsynchrotron": 20538,
+      "rik": 20539,
+      "synaptic": 20540,
+      "ĠGeorgia": 20541,
+      "??": 20542,
+      "CY": 20543,
+      "Ġcorresponded": 20544,
+      "kinase": 20545,
+      "ĠITS": 20546,
+      "Ġproposals": 20547,
+      "Ġbioge": 20548,
+      "ĠESR": 20549,
+      "ĠWen": 20550,
+      "ĠJa": 20551,
+      "ĠSevere": 20552,
+      "ĠAden": 20553,
+      "ĠCCL": 20554,
+      "Ġseat": 20555,
+      "ĠKre": 20556,
+      "Ġhelping": 20557,
+      "Ġnets": 20558,
+      "ĠLep": 20559,
+      "hedra": 20560,
+      "opoulos": 20561,
+      "ĠBak": 20562,
+      "ansas": 20563,
+      "Ġrefrig": 20564,
+      "Ġubiquitous": 20565,
+      "Ġmatters": 20566,
+      "Ġsilicate": 20567,
+      "ĠLastly": 20568,
+      "ĠTheories": 20569,
+      "Ġagarose": 20570,
+      "biggr": 20571,
+      "transition": 20572,
+      "ĠDecomposition": 20573,
+      "bromo": 20574,
+      "Ġstakeholders": 20575,
+      "ĠEE": 20576,
+      "Only": 20577,
+      "ĠKenya": 20578,
+      "Ġargon": 20579,
+      "ĠIdentifying": 20580,
+      "Ġtournament": 20581,
+      "clock": 20582,
+      "ĠCFU": 20583,
+      "ĠBehavioral": 20584,
+      "Ġpod": 20585,
+      "Ġtaxonomy": 20586,
+      "ĠProduct": 20587,
+      "ĠAlong": 20588,
+      "Ġfamilial": 20589,
+      "Ġdescriptor": 20590,
+      "vated": 20591,
+      "ĠVariables": 20592,
+      "tp": 20593,
+      "Ġgoods": 20594,
+      "ĠAST": 20595,
+      "ĠAnis": 20596,
+      "Ġspinor": 20597,
+      "attention": 20598,
+      "Ġbasket": 20599,
+      "Struct": 20600,
+      "Ġimmunohistochemical": 20601,
+      "engers": 20602,
+      "CAT": 20603,
+      "Ġtangential": 20604,
+      "Cap": 20605,
+      "ĠPair": 20606,
+      "Ġviscoelastic": 20607,
+      "ĠAds": 20608,
+      "Ġglycosylation": 20609,
+      "Ġdur": 20610,
+      "ĠMinimum": 20611,
+      "Ġrigidity": 20612,
+      "stats": 20613,
+      "tillation": 20614,
+      "ĠDiscrim": 20615,
+      "ĠLegend": 20616,
+      "Previous": 20617,
+      "film": 20618,
+      "Ġaluminium": 20619,
+      "Micro": 20620,
+      "inia": 20621,
+      "egel": 20622,
+      "ĠSubcellular": 20623,
+      "Ġbottleneck": 20624,
+      "Ġsyll": 20625,
+      "icle": 20626,
+      "Ġsheaf": 20627,
+      "chell": 20628,
+      "example": 20629,
+      "ĠSelected": 20630,
+      "Ġpredators": 20631,
+      "Ġreper": 20632,
+      "Ġstrugg": 20633,
+      "ĠMaria": 20634,
+      "lyl": 20635,
+      "LF": 20636,
+      "Ġexercises": 20637,
+      "obium": 20638,
+      "ILITY": 20639,
+      "corrected": 20640,
+      "Ġbenchmarks": 20641,
+      "ĠTol": 20642,
+      "Ġintercept": 20643,
+      "ĠCalculation": 20644,
+      "ĠIndonesia": 20645,
+      "Ġglioblastoma": 20646,
+      "KM": 20647,
+      "ĠSupplemental": 20648,
+      "Ġcitizens": 20649,
+      "adren": 20650,
+      "Ġmultimodal": 20651,
+      "Ġmosquitoes": 20652,
+      "iva": 20653,
+      "ĠFindings": 20654,
+      "ĠPub": 20655,
+      "ĠMacroph": 20656,
+      "Acknowledg": 20657,
+      "Ġbasins": 20658,
+      "exact": 20659,
+      "Ġgrades": 20660,
+      "Ġfir": 20661,
+      "iga": 20662,
+      "ĠPolynomial": 20663,
+      "ĠLongitudinal": 20664,
+      "Ġsemiconductors": 20665,
+      "Top": 20666,
+      "iptera": 20667,
+      "Ġlacks": 20668,
+      "rograph": 20669,
+      "Ġselects": 20670,
+      "Ġsweet": 20671,
+      "Ġbac": 20672,
+      "Ġdownloaded": 20673,
+      "aponic": 20674,
+      "ijk": 20675,
+      "otonic": 20676,
+      "normalized": 20677,
+      "ĠVariability": 20678,
+      "division": 20679,
+      "ĠSupers": 20680,
+      "ilab": 20681,
+      "Human": 20682,
+      "Ġleptin": 20683,
+      "Ġosmotic": 20684,
+      "Ġhur": 20685,
+      "ĠSingapore": 20686,
+      "ĠOPT": 20687,
+      "ĠSoviet": 20688,
+      "litaxel": 20689,
+      "retaceous": 20690,
+      "ĠOnc": 20691,
+      "ĠIX": 20692,
+      "ulas": 20693,
+      "uent": 20694,
+      "Ġlymphoid": 20695,
+      "Tc": 20696,
+      "Ġrationale": 20697,
+      "Layer": 20698,
+      "osities": 20699,
+      "Ġdesire": 20700,
+      "ĠAnnual": 20701,
+      "uba": 20702,
+      "ĠCompounds": 20703,
+      "Ġantifungal": 20704,
+      "Ġcationic": 20705,
+      "items": 20706,
+      "acterium": 20707,
+      "amilies": 20708,
+      "Ġelongated": 20709,
+      "ĠMassachusetts": 20710,
+      "ĠIrish": 20711,
+      "asso": 20712,
+      "azo": 20713,
+      "ĠBurk": 20714,
+      "robenius": 20715,
+      "Ġisinstance": 20716,
+      "bion": 20717,
+      "Ġgreedy": 20718,
+      "Ġnicotine": 20719,
+      "Ġretrieve": 20720,
+      "Ġsympathetic": 20721,
+      "quee": 20722,
+      "Ġfoli": 20723,
+      "Ġsputter": 20724,
+      "Ġgrading": 20725,
+      "determined": 20726,
+      "Ġabnorm": 20727,
+      "Ġmanagers": 20728,
+      "Ġtopical": 20729,
+      "Ġimmig": 20730,
+      "ĠDNN": 20731,
+      "gtr": 20732,
+      "Ġdetections": 20733,
+      "ĠObesity": 20734,
+      "suc": 20735,
+      "ĠSche": 20736,
+      "Ġtrunk": 20737,
+      "Ġtough": 20738,
+      "ĠBN": 20739,
+      "Ġru": 20740,
+      "oxif": 20741,
+      "Ġaiming": 20742,
+      "ĠExtracellular": 20743,
+      "Ġhaplotype": 20744,
+      "Du": 20745,
+      "ĠDing": 20746,
+      "ĠDol": 20747,
+      "Ġhumid": 20748,
+      "brom": 20749,
+      "Ġoffline": 20750,
+      "Combining": 20751,
+      "Ġpulsar": 20752,
+      "Ġpari": 20753,
+      "partate": 20754,
+      "imated": 20755,
+      "Ġwatershed": 20756,
+      "acrylamide": 20757,
+      "exec": 20758,
+      "ĠComposite": 20759,
+      "Ġdispersive": 20760,
+      "Ġtons": 20761,
+      "rometry": 20762,
+      "ĠJud": 20763,
+      "aza": 20764,
+      "Ġchickens": 20765,
+      "register": 20766,
+      "nz": 20767,
+      "Util": 20768,
+      "ĠVes": 20769,
+      "eV": 20770,
+      "ĠRule": 20771,
+      "substituted": 20772,
+      "Conv": 20773,
+      "query": 20774,
+      "Mac": 20775,
+      "ĠTar": 20776,
+      "implies": 20777,
+      "ĠRates": 20778,
+      "Ġrins": 20779,
+      "Ġtimescales": 20780,
+      "ĠCzech": 20781,
+      "Such": 20782,
+      "restimate": 20783,
+      "ĠMb": 20784,
+      "ĠFuj": 20785,
+      "ĠIMD": 20786,
+      "cit": 20787,
+      "Ġraising": 20788,
+      "........": 20789,
+      "home": 20790,
+      "asted": 20791,
+      "Ġocta": 20792,
+      "Ġcadmium": 20793,
+      "Ġpsori": 20794,
+      "roleum": 20795,
+      "ĠStellar": 20796,
+      "ĠKinase": 20797,
+      "ĠGard": 20798,
+      "ieu": 20799,
+      "ĠMoS": 20800,
+      "MG": 20801,
+      "ĠGSH": 20802,
+      "Ġhazards": 20803,
+      "Ġnice": 20804,
+      "heating": 20805,
+      "Ġreproducible": 20806,
+      "genesis": 20807,
+      "ĠIgM": 20808,
+      "Ġbeat": 20809,
+      "onuclease": 20810,
+      "entralized": 20811,
+      "ĠLé": 20812,
+      "Ġdol": 20813,
+      "Ġdeeply": 20814,
+      "ractive": 20815,
+      "Ġglial": 20816,
+      "iella": 20817,
+      "Ġinitialized": 20818,
+      "ĠMethodology": 20819,
+      "Ġbenthic": 20820,
+      "omi": 20821,
+      "ĠAlter": 20822,
+      "Ordered": 20823,
+      "ĠLIN": 20824,
+      "Ġunilateral": 20825,
+      "Ġcorticoster": 20826,
+      "LEN": 20827,
+      "Ġdilute": 20828,
+      "Ġmetalloprotein": 20829,
+      "abeth": 20830,
+      "ampion": 20831,
+      "Ġmoral": 20832,
+      "ĠSiC": 20833,
+      "Ġquadrature": 20834,
+      "Ġsedimentation": 20835,
+      "ete": 20836,
+      "ĠFrag": 20837,
+      "Ġpeaked": 20838,
+      "Ġmitigation": 20839,
+      "Ġsoldi": 20840,
+      "Ġdoubly": 20841,
+      "Ġellipso": 20842,
+      "ĠlncRNAs": 20843,
+      "Ġâİ¢": 20844,
+      "ĠSame": 20845,
+      "ĠSustain": 20846,
+      "ĠCapacity": 20847,
+      "Ġsomat": 20848,
+      "Ġtransistor": 20849,
+      "Ġassayed": 20850,
+      "ĠNur": 20851,
+      "tools": 20852,
+      "Sing": 20853,
+      "Ġligament": 20854,
+      "atever": 20855,
+      "Ġperce": 20856,
+      "hence": 20857,
+      "UX": 20858,
+      "sent": 20859,
+      "EGG": 20860,
+      "third": 20861,
+      "enders": 20862,
+      "theoretic": 20863,
+      "Ġrewards": 20864,
+      "uto": 20865,
+      "Ġinstallation": 20866,
+      "ĠKinetic": 20867,
+      "ĠInnov": 20868,
+      "ĠSolving": 20869,
+      "ĠSymmetry": 20870,
+      "Ġramp": 20871,
+      "Ġneuropathy": 20872,
+      "omerization": 20873,
+      "Ġcatech": 20874,
+      "Pred": 20875,
+      "ĠBoh": 20876,
+      "EMENT": 20877,
+      "Ġarmy": 20878,
+      "ĠYukawa": 20879,
+      "Ġalignments": 20880,
+      "ĠDependence": 20881,
+      "Ġenv": 20882,
+      "ean": 20883,
+      "sr": 20884,
+      "Ġinterpreting": 20885,
+      "elocity": 20886,
+      "Ġpsychology": 20887,
+      "Ġbiofilms": 20888,
+      "Ġeccentricity": 20889,
+      "lot": 20890,
+      "analytic": 20891,
+      "Ġperiodicity": 20892,
+      "nings": 20893,
+      "ĠKent": 20894,
+      "flag": 20895,
+      "Ġmp": 20896,
+      "ĠNich": 20897,
+      "hire": 20898,
+      "Ġflare": 20899,
+      "Ġcitrate": 20900,
+      "Ġpaste": 20901,
+      "Ġdelete": 20902,
+      "zymes": 20903,
+      "orientation": 20904,
+      "ĠHY": 20905,
+      "Ġcommands": 20906,
+      "Ġstrike": 20907,
+      "symbol": 20908,
+      "ĠMind": 20909,
+      "Ġoptimisation": 20910,
+      "Ġosteoporosis": 20911,
+      "ĠInflammation": 20912,
+      "ĠIntelligence": 20913,
+      "eh": 20914,
+      "utum": 20915,
+      "Ġvec": 20916,
+      "ellation": 20917,
+      "ĠBloch": 20918,
+      "ĠMajorana": 20919,
+      "enor": 20920,
+      "ĠNgu": 20921,
+      "Ġdeuter": 20922,
+      "opedia": 20923,
+      "Ġutter": 20924,
+      "Ġribosome": 20925,
+      "Ġactors": 20926,
+      "electronic": 20927,
+      "ée": 20928,
+      "Ġfeaturing": 20929,
+      "agle": 20930,
+      "Ġperin": 20931,
+      "ĠCivil": 20932,
+      "Ġpredefined": 20933,
+      "lag": 20934,
+      "ĠJAK": 20935,
+      "jamin": 20936,
+      "individual": 20937,
+      "onc": 20938,
+      "Ġfishing": 20939,
+      "ditive": 20940,
+      "Norm": 20941,
+      "ĠScanning": 20942,
+      "vanishing": 20943,
+      "Ġcessation": 20944,
+      "ĠHole": 20945,
+      "ributes": 20946,
+      "IE": 20947,
+      "ĠMpc": 20948,
+      "wegian": 20949,
+      "Ma": 20950,
+      "Ġrevisited": 20951,
+      "ĠPlus": 20952,
+      "abilized": 20953,
+      "Ġscanned": 20954,
+      "ĠExchange": 20955,
+      "Ġbromide": 20956,
+      "Life": 20957,
+      "otroph": 20958,
+      "ADS": 20959,
+      "âĭħ": 20960,
+      "Ġoperative": 20961,
+      "ĠBERT": 20962,
+      "Ġplume": 20963,
+      "Ġpoorer": 20964,
+      "Ġtrout": 20965,
+      "Ġmicrotubule": 20966,
+      "Ġphosphatidyl": 20967,
+      "radius": 20968,
+      "ĠMuscle": 20969,
+      "Ġcarcinogenesis": 20970,
+      "Ġseeing": 20971,
+      "uclein": 20972,
+      "follow": 20973,
+      "Ġsupplements": 20974,
+      "olars": 20975,
+      "specially": 20976,
+      "Ġcompleting": 20977,
+      "Ġnaïve": 20978,
+      "ĠÏ©": 20979,
+      "clerotic": 20980,
+      "Disc": 20981,
+      "ĠFestival": 20982,
+      "Ġclick": 20983,
+      "clusive": 20984,
+      "Ġcatalogue": 20985,
+      "Ġapps": 20986,
+      "ĠSED": 20987,
+      "Ġstacked": 20988,
+      "Ġtune": 20989,
+      "ĠDMEM": 20990,
+      "Ġaerosols": 20991,
+      "Ġgear": 20992,
+      "antine": 20993,
+      "ĠStone": 20994,
+      "Ġpositives": 20995,
+      "triang": 20996,
+      "probability": 20997,
+      "Ġdecoupling": 20998,
+      "ĠÍĵ": 20999,
+      "ĠVin": 21000,
+      "Ġsurvived": 21001,
+      "Ġreplicated": 21002,
+      "utrient": 21003,
+      "Ġtemperate": 21004,
+      "Ġensembles": 21005,
+      "Ġmulticenter": 21006,
+      "Ġgaseous": 21007,
+      "idea": 21008,
+      "classification": 21009,
+      "ĠOutcome": 21010,
+      "clonal": 21011,
+      "Ġdiscontinuity": 21012,
+      "Ġadvantageous": 21013,
+      "Ġdistricts": 21014,
+      "ĠIBM": 21015,
+      "inguishable": 21016,
+      "Ġcars": 21017,
+      "cult": 21018,
+      "enriched": 21019,
+      "argin": 21020,
+      "novae": 21021,
+      "steady": 21022,
+      "Ġbuy": 21023,
+      "piration": 21024,
+      "Ġpartitioned": 21025,
+      "Ġinability": 21026,
+      "pq": 21027,
+      "Ġbull": 21028,
+      "odend": 21029,
+      "Ġassistant": 21030,
+      "Ġlumen": 21031,
+      "Ġconverting": 21032,
+      "PY": 21033,
+      "zol": 21034,
+      "utors": 21035,
+      "ĠNLRP": 21036,
+      "apply": 21037,
+      "ĠBonferroni": 21038,
+      "Ls": 21039,
+      "Ġtips": 21040,
+      "ĠLN": 21041,
+      "rolase": 21042,
+      "Ġadvis": 21043,
+      "ĠMetast": 21044,
+      "Ġsaliva": 21045,
+      "Ġinhabit": 21046,
+      "Ġrim": 21047,
+      "debug": 21048,
+      "Any": 21049,
+      "Ġforb": 21050,
+      "Ġversatile": 21051,
+      "ĠCompact": 21052,
+      "voc": 21053,
+      "ĠIso": 21054,
+      "ĠJus": 21055,
+      "bodies": 21056,
+      "ARM": 21057,
+      "ĠGWAS": 21058,
+      "hetized": 21059,
+      "Ġmicrofluidic": 21060,
+      "Ġacetonitrile": 21061,
+      "Ġinhom": 21062,
+      "Ġparench": 21063,
+      "Ġinsensitive": 21064,
+      "Ġagency": 21065,
+      "poor": 21066,
+      "ĠAngi": 21067,
+      "Ġapproached": 21068,
+      "Ġemulsion": 21069,
+      "Ġvoluntary": 21070,
+      "utt": 21071,
+      "ĠRecurrent": 21072,
+      "riculum": 21073,
+      "ê": 21074,
+      "Ġtall": 21075,
+      "ĠDepth": 21076,
+      "Ġff": 21077,
+      "ĠIncidence": 21078,
+      "Ġmanifestation": 21079,
+      "Ġcompromised": 21080,
+      "iaceae": 21081,
+      "ĠMIT": 21082,
+      "otransfer": 21083,
+      "ĠWD": 21084,
+      "mov": 21085,
+      "ĠManual": 21086,
+      "Medi": 21087,
+      "Ġinterfering": 21088,
+      "ĠJacobi": 21089,
+      "KT": 21090,
+      "Ġsarcoma": 21091,
+      "Ġkidneys": 21092,
+      "Ġodor": 21093,
+      "Ġti": 21094,
+      "yday": 21095,
+      "although": 21096,
+      "visible": 21097,
+      "Ġdengue": 21098,
+      "ĠCAL": 21099,
+      "strat": 21100,
+      "ĠVariations": 21101,
+      "inib": 21102,
+      "components": 21103,
+      "ĠTob": 21104,
+      "ĠAntioxidant": 21105,
+      "ÍĶ": 21106,
+      "Ġkiller": 21107,
+      "Ġsubtracted": 21108,
+      "ĠEvents": 21109,
+      "Ġimplements": 21110,
+      "ĠGAN": 21111,
+      "Ġprophylaxis": 21112,
+      "Ġnozz": 21113,
+      "Ġsmoothed": 21114,
+      "Ġdecaying": 21115,
+      "ĠInitially": 21116,
+      "Ġuncommon": 21117,
+      "Ġconductor": 21118,
+      "ĠWOR": 21119,
+      "avity": 21120,
+      "ĠXie": 21121,
+      "ĠAcet": 21122,
+      "Ġine": 21123,
+      "ĠBeam": 21124,
+      "opolymer": 21125,
+      "ĠXML": 21126,
+      "ĠWide": 21127,
+      "Ñĥ": 21128,
+      "Ġejection": 21129,
+      "BMI": 21130,
+      "tc": 21131,
+      "uez": 21132,
+      "Ġcerebellar": 21133,
+      "Ġcatchment": 21134,
+      "coxon": 21135,
+      "ĠShannon": 21136,
+      "Ġcentrality": 21137,
+      "Ġsafely": 21138,
+      "probe": 21139,
+      "ĠLaboratories": 21140,
+      "Ġnc": 21141,
+      "Ġspher": 21142,
+      "Ġprobing": 21143,
+      "ĠLev": 21144,
+      "Ġaf": 21145,
+      "ĠMig": 21146,
+      "ĠVascular": 21147,
+      "Ġprogrammes": 21148,
+      "Ġcontaminants": 21149,
+      "sequent": 21150,
+      "Ġbonded": 21151,
+      "integration": 21152,
+      "bos": 21153,
+      "ĠFew": 21154,
+      "ĠIllinois": 21155,
+      "She": 21156,
+      "WC": 21157,
+      "ĠGPIO": 21158,
+      "oC": 21159,
+      "ĠMaternal": 21160,
+      "ercetin": 21161,
+      "ĠMassive": 21162,
+      "Ġenorm": 21163,
+      "imgur": 21164,
+      "Ġbidirectional": 21165,
+      "ĠGraphene": 21166,
+      "insky": 21167,
+      "ĠObserve": 21168,
+      "Ġstops": 21169,
+      "bio": 21170,
+      "ĠLines": 21171,
+      "ĠGill": 21172,
+      "Ġeigenvector": 21173,
+      "Space": 21174,
+      "ĠMining": 21175,
+      "Ġmelatonin": 21176,
+      "ĠSET": 21177,
+      "onsequ": 21178,
+      "oscale": 21179,
+      "ĠRaw": 21180,
+      "Ġreviewers": 21181,
+      "Ġnanofib": 21182,
+      "taking": 21183,
+      "ammad": 21184,
+      "Ġrecursion": 21185,
+      "usal": 21186,
+      "Ġpositron": 21187,
+      "ĠNIH": 21188,
+      "ĠINTER": 21189,
+      "ĠDocument": 21190,
+      "Ġconstantly": 21191,
+      "Ġundergone": 21192,
+      "Ġelectroweak": 21193,
+      "Ġiteratively": 21194,
+      "folio": 21195,
+      "Ġsubfamily": 21196,
+      "Ġâİ¥": 21197,
+      "Page": 21198,
+      "ferm": 21199,
+      "avir": 21200,
+      "Ġagencies": 21201,
+      "Ġpolys": 21202,
+      "ĠSquare": 21203,
+      "ymm": 21204,
+      "Ġhydrogels": 21205,
+      "almost": 21206,
+      "arter": 21207,
+      "Ġankle": 21208,
+      "Ġrises": 21209,
+      "Ġmedull": 21210,
+      "gated": 21211,
+      "Ġmononuclear": 21212,
+      "Ġdiscussing": 21213,
+      "Ġprofessor": 21214,
+      "transformed": 21215,
+      "Ġcolours": 21216,
+      "ragg": 21217,
+      "emicon": 21218,
+      "Ġsymmetrical": 21219,
+      "Ġplacental": 21220,
+      "Ġli": 21221,
+      "Ġstudio": 21222,
+      "sequences": 21223,
+      "Ġtam": 21224,
+      "ĠLap": 21225,
+      "ĠCriteria": 21226,
+      "Ġhappened": 21227,
+      "Ġantiferromagnetic": 21228,
+      "ĠHausdorff": 21229,
+      "ĠCONCLUSIONS": 21230,
+      "HER": 21231,
+      "VR": 21232,
+      "ĠKor": 21233,
+      "ĠAPO": 21234,
+      "Ġprotecting": 21235,
+      "ĠSOL": 21236,
+      "ĠBuck": 21237,
+      "phia": 21238,
+      "ĠMultim": 21239,
+      "onine": 21240,
+      "ulsions": 21241,
+      "Ġgp": 21242,
+      "benzamide": 21243,
+      "ĠNADPH": 21244,
+      "ĠOhio": 21245,
+      "ĠMEG": 21246,
+      "COVID": 21247,
+      "Ġdisplaced": 21248,
+      "ĠAbb": 21249,
+      "Ġbranched": 21250,
+      "ĠNavy": 21251,
+      "ĠNrf": 21252,
+      "ĠODE": 21253,
+      "achi": 21254,
+      "ĠTransient": 21255,
+      "Ġcircumference": 21256,
+      "Ġbees": 21257,
+      "iration": 21258,
+      "Ġfaculty": 21259,
+      "IGHT": 21260,
+      "ĠMetabolism": 21261,
+      "MK": 21262,
+      "ĠInequ": 21263,
+      "ĠQualitative": 21264,
+      "PBS": 21265,
+      "terminus": 21266,
+      "kary": 21267,
+      "ovian": 21268,
+      "ĠTHz": 21269,
+      "ĠReliability": 21270,
+      "furan": 21271,
+      "Ġcorners": 21272,
+      "Ġattacker": 21273,
+      "Ġmarriage": 21274,
+      "oprecipitation": 21275,
+      "ĠCry": 21276,
+      "ĠâĬĻ": 21277,
+      "Ġevolves": 21278,
+      "Ġban": 21279,
+      "Ġdiurnal": 21280,
+      "ounce": 21281,
+      "Ġoverw": 21282,
+      "ĠHoff": 21283,
+      "Ġextrinsic": 21284,
+      "amps": 21285,
+      "ULAR": 21286,
+      "opher": 21287,
+      "Ġlighting": 21288,
+      "Ġarchitectural": 21289,
+      "hesive": 21290,
+      "Ġsavings": 21291,
+      "Ġglaucoma": 21292,
+      "ozoa": 21293,
+      "ĠOption": 21294,
+      "controll": 21295,
+      "ecker": 21296,
+      "Ġosteocl": 21297,
+      "Ġglycine": 21298,
+      "analyses": 21299,
+      "ĠAld": 21300,
+      "ĠSyd": 21301,
+      "ĠCx": 21302,
+      "Ġscalars": 21303,
+      "Ġknowing": 21304,
+      "Ġremember": 21305,
+      "ĠEmbry": 21306,
+      "TEM": 21307,
+      "ĠBran": 21308,
+      "FORM": 21309,
+      "Ġsurviving": 21310,
+      "Ġglobular": 21311,
+      "Ġinclusive": 21312,
+      "sched": 21313,
+      "UTION": 21314,
+      "Ġquadrupole": 21315,
+      "ĠHubbard": 21316,
+      "Ġaxonal": 21317,
+      "ĠCosmic": 21318,
+      "Ġslots": 21319,
+      "ĠProcedure": 21320,
+      "agin": 21321,
+      "ĠLoop": 21322,
+      "arer": 21323,
+      "Ġbutter": 21324,
+      "Ġhistopathological": 21325,
+      "fusion": 21326,
+      "ANOVA": 21327,
+      "Ġclosing": 21328,
+      "ĠLord": 21329,
+      "ĠBis": 21330,
+      "ĠRAM": 21331,
+      "IDE": 21332,
+      "Ġjournals": 21333,
+      "Ġmonkeys": 21334,
+      "Ġattenuates": 21335,
+      "Ġsegmented": 21336,
+      "TOF": 21337,
+      "otional": 21338,
+      "polymer": 21339,
+      "ĠShah": 21340,
+      "Akt": 21341,
+      "Wr": 21342,
+      "lov": 21343,
+      "Ġpolymorphic": 21344,
+      "Ġarrangements": 21345,
+      "UF": 21346,
+      "lon": 21347,
+      "Ġdepressed": 21348,
+      "NAT": 21349,
+      "ĠOperation": 21350,
+      "ι": 21351,
+      "ĠRan": 21352,
+      "âIJ": 21353,
+      "Ġthereafter": 21354,
+      "Ġmyeloma": 21355,
+      "jor": 21356,
+      "Ã¥": 21357,
+      "ĠWinter": 21358,
+      "ptosis": 21359,
+      "Dir": 21360,
+      "verty": 21361,
+      "ĠFinn": 21362,
+      "Ġortholog": 21363,
+      "Ġmonotonically": 21364,
+      "Ġtectonic": 21365,
+      "ĠGBM": 21366,
+      "ĠAO": 21367,
+      "Ġgenerative": 21368,
+      "Clearly": 21369,
+      "Ġtile": 21370,
+      "ĠRNN": 21371,
+      "Ġgrounds": 21372,
+      "GaAs": 21373,
+      "Ġbee": 21374,
+      "ĠBoy": 21375,
+      "ĠTranscriptional": 21376,
+      "urin": 21377,
+      "otom": 21378,
+      "Ġsinusoidal": 21379,
+      "ĠAy": 21380,
+      "ĠClinic": 21381,
+      "utorial": 21382,
+      "ĠADC": 21383,
+      "ERIAL": 21384,
+      "cation": 21385,
+      "ĠADHD": 21386,
+      "cyclohex": 21387,
+      "ĠHawai": 21388,
+      "astom": 21389,
+      "Ġmorphologies": 21390,
+      "Ġrodents": 21391,
+      "Ġscalability": 21392,
+      "ROS": 21393,
+      "aemia": 21394,
+      "Ġdecompose": 21395,
+      "Ġpivotal": 21396,
+      "Ġdiffusivity": 21397,
+      "Ġcovalent": 21398,
+      "ĠKD": 21399,
+      "atalyst": 21400,
+      "Ġoldest": 21401,
+      "Ġsuitability": 21402,
+      "Ġwants": 21403,
+      "ifts": 21404,
+      "ĠDistributions": 21405,
+      "ĠQueen": 21406,
+      "lich": 21407,
+      "Ġparse": 21408,
+      "ĠMHD": 21409,
+      "Ġrecre": 21410,
+      "Ġhydroxide": 21411,
+      "eum": 21412,
+      "Ġlev": 21413,
+      "Ġreferral": 21414,
+      "planes": 21415,
+      "ĠEgypt": 21416,
+      "Ġlenti": 21417,
+      "Ġtransactions": 21418,
+      "Ġexpense": 21419,
+      "Ġcysts": 21420,
+      "Ġabscess": 21421,
+      "ĠmicroRNAs": 21422,
+      "effectiveness": 21423,
+      "ĠDifferentiation": 21424,
+      "Ġcertif": 21425,
+      "cience": 21426,
+      "ĠREL": 21427,
+      "Ġreadout": 21428,
+      "ĠQuasi": 21429,
+      "Ġrounded": 21430,
+      "otti": 21431,
+      "efficients": 21432,
+      "Ġsynchronized": 21433,
+      "Ġsilico": 21434,
+      "Ġforecasts": 21435,
+      "Ġdμ": 21436,
+      "Ġexotic": 21437,
+      "ĠOCT": 21438,
+      "xb": 21439,
+      "Ġasynchronous": 21440,
+      "nez": 21441,
+      "chiat": 21442,
+      "Ġhaemat": 21443,
+      "Ġfulfill": 21444,
+      "ĠMix": 21445,
+      "ibli": 21446,
+      "fm": 21447,
+      "Ġjava": 21448,
+      "soluble": 21449,
+      "Ġincompressible": 21450,
+      "âĨij": 21451,
+      "CDM": 21452,
+      "Ġdilation": 21453,
+      "LYP": 21454,
+      "ashes": 21455,
+      "ĠSports": 21456,
+      "Ġfundament": 21457,
+      "ĠSaudi": 21458,
+      "Ġenroll": 21459,
+      "ĠNaOH": 21460,
+      "Ġcrustal": 21461,
+      "ĠInstruments": 21462,
+      "Ġïģ¡": 21463,
+      "Result": 21464,
+      "Ġpreferential": 21465,
+      "Ġsugars": 21466,
+      "Ġdimers": 21467,
+      "ĠEmerging": 21468,
+      "ère": 21469,
+      "diabetic": 21470,
+      "Ġstrengthening": 21471,
+      "epi": 21472,
+      "ĠMeg": 21473,
+      "ĠYour": 21474,
+      "ĠSetting": 21475,
+      "lez": 21476,
+      "ĠBou": 21477,
+      "Ġhistology": 21478,
+      "Ġolive": 21479,
+      "ĠDisorders": 21480,
+      "Ġdistorted": 21481,
+      "Ġcompete": 21482,
+      "cens": 21483,
+      "ĠAe": 21484,
+      "ĠGG": 21485,
+      "Ġquantifying": 21486,
+      "Ġaur": 21487,
+      "ĠWright": 21488,
+      "Ġsuperconductor": 21489,
+      "eds": 21490,
+      "stalk": 21491,
+      "concent": 21492,
+      "ĠLimited": 21493,
+      "Ġstyles": 21494,
+      "design": 21495,
+      "ĠEllip": 21496,
+      "PLA": 21497,
+      "mogorov": 21498,
+      "ĠRidge": 21499,
+      "Ġrandomization": 21500,
+      "aft": 21501,
+      "icially": 21502,
+      "ĠBiotechnology": 21503,
+      "Ġseizure": 21504,
+      "KI": 21505,
+      "AVE": 21506,
+      "receptor": 21507,
+      "Ġgrammar": 21508,
+      "Ġcrime": 21509,
+      "nection": 21510,
+      "inces": 21511,
+      "ĠCompton": 21512,
+      "Ġventricle": 21513,
+      "Ġredistribution": 21514,
+      "ynaptic": 21515,
+      "Parameter": 21516,
+      "Normal": 21517,
+      "Pack": 21518,
+      "ermann": 21519,
+      "ulants": 21520,
+      "degenerate": 21521,
+      "ĠNewtonian": 21522,
+      "Ġancestral": 21523,
+      "phrag": 21524,
+      "Ġimpression": 21525,
+      "Ġnormalize": 21526,
+      "Ġambiguous": 21527,
+      "Ġingredient": 21528,
+      "ĠClaim": 21529,
+      "Ġcleaved": 21530,
+      "ĠApproaches": 21531,
+      "ĠSPECT": 21532,
+      "csv": 21533,
+      "ĠReveals": 21534,
+      "ĠWaves": 21535,
+      "Ġdwarfs": 21536,
+      "ĠProgress": 21537,
+      "Ġaorta": 21538,
+      "Ġnig": 21539,
+      "ĠAdams": 21540,
+      "ĠMüller": 21541,
+      "ĠYellow": 21542,
+      "ĠCord": 21543,
+      "ĠPhill": 21544,
+      "ĠFormal": 21545,
+      "besgue": 21546,
+      "termin": 21547,
+      "rn": 21548,
+      "bn": 21549,
+      "kine": 21550,
+      "rit": 21551,
+      "qi": 21552,
+      "ĠRoute": 21553,
+      "enol": 21554,
+      "ĠASC": 21555,
+      "ĠPu": 21556,
+      "mill": 21557,
+      "umer": 21558,
+      "Ġsupernova": 21559,
+      "iative": 21560,
+      "differenti": 21561,
+      "Ġtolu": 21562,
+      "opus": 21563,
+      "RM": 21564,
+      "Ġpoverty": 21565,
+      "ĠXX": 21566,
+      "ĠïĤ¶": 21567,
+      "ultry": 21568,
+      "Optim": 21569,
+      "Ġglacial": 21570,
+      "ĠDispers": 21571,
+      "Ġdifferentiating": 21572,
+      "ández": 21573,
+      "project": 21574,
+      "ĠEliz": 21575,
+      "scaling": 21576,
+      "ĠToll": 21577,
+      "Ġnonempty": 21578,
+      "Ġpredicate": 21579,
+      "Ġgyrus": 21580,
+      "minute": 21581,
+      "âĸ": 21582,
+      "ĠHind": 21583,
+      "ĠLiving": 21584,
+      "VS": 21585,
+      "prior": 21586,
+      "ĠVertical": 21587,
+      "arks": 21588,
+      "ĠSFR": 21589,
+      "ĠVietnam": 21590,
+      "compare": 21591,
+      ">>>": 21592,
+      "Ġbanks": 21593,
+      "Ġseptic": 21594,
+      "ĠBif": 21595,
+      "ĠEPS": 21596,
+      "ĠIntel": 21597,
+      "ĠDisorder": 21598,
+      "PN": 21599,
+      "ĠNord": 21600,
+      "tiveness": 21601,
+      "Ġdrilling": 21602,
+      "ĠSubject": 21603,
+      "enario": 21604,
+      "Ġrms": 21605,
+      "phones": 21606,
+      "hang": 21607,
+      "ĠTechnique": 21608,
+      "Ġclot": 21609,
+      "Ġintersections": 21610,
+      "Ġanions": 21611,
+      "above": 21612,
+      "Ġclause": 21613,
+      "Ġgenu": 21614,
+      "ozo": 21615,
+      "rhiz": 21616,
+      "Ġlobes": 21617,
+      "ĠBian": 21618,
+      "Ġexerted": 21619,
+      "ureth": 21620,
+      "roma": 21621,
+      "ĠCharge": 21622,
+      "ĠSynchron": 21623,
+      "Ġconting": 21624,
+      "otherapeutic": 21625,
+      "gtrsim": 21626,
+      "ĠResonance": 21627,
+      "ĠFal": 21628,
+      "undle": 21629,
+      "Ġdropout": 21630,
+      "ĠPerspective": 21631,
+      "OLOG": 21632,
+      "atches": 21633,
+      "ĠSequences": 21634,
+      "Considering": 21635,
+      "Ġprospects": 21636,
+      "Ġaliqu": 21637,
+      "Ġstrata": 21638,
+      "Ġanalogs": 21639,
+      "Ġencouraged": 21640,
+      "ĠPulmonary": 21641,
+      "Ġchim": 21642,
+      "ĠCFT": 21643,
+      "unar": 21644,
+      "izz": 21645,
+      "endocrine": 21646,
+      "ĠCRE": 21647,
+      "ĠStroke": 21648,
+      "âĩĴ": 21649,
+      "upuncture": 21650,
+      "translational": 21651,
+      "ĠGriff": 21652,
+      "ĠSter": 21653,
+      "erged": 21654,
+      "phrine": 21655,
+      "Ġlivestock": 21656,
+      "ĠHash": 21657,
+      "Ġdosing": 21658,
+      "Ġplasmas": 21659,
+      "ĠComparisons": 21660,
+      "Ġencouraging": 21661,
+      "Ġcomparatively": 21662,
+      "Ġcharacterisation": 21663,
+      "Ġascending": 21664,
+      "ĠFixed": 21665,
+      "Ġvegetable": 21666,
+      "especially": 21667,
+      "ĠLange": 21668,
+      "ĠCoding": 21669,
+      "Ġvertebrate": 21670,
+      "FW": 21671,
+      "ĠORF": 21672,
+      "ĠTub": 21673,
+      "lee": 21674,
+      "Ġtimely": 21675,
+      "Ep": 21676,
+      "ĠâĪĴâĪŀ": 21677,
+      "Ġliposomes": 21678,
+      "Ġextremal": 21679,
+      "ropolitan": 21680,
+      "ĠCay": 21681,
+      "ĠBiod": 21682,
+      "oulli": 21683,
+      "Dri": 21684,
+      "ĠRats": 21685,
+      "Ġcentroid": 21686,
+      "ospin": 21687,
+      "rospinal": 21688,
+      "Ġsolitons": 21689,
+      "portive": 21690,
+      "ĠMcG": 21691,
+      "Bbb": 21692,
+      "Ġparaffin": 21693,
+      "lectively": 21694,
+      "Ġmetastable": 21695,
+      "Ġdissipative": 21696,
+      "Ġassemblages": 21697,
+      "Ġcolonic": 21698,
+      "Ġsized": 21699,
+      "Ġcryp": 21700,
+      "processor": 21701,
+      "ção": 21702,
+      "Ġacknowledged": 21703,
+      "ĠUncertainty": 21704,
+      "ĠIndustrial": 21705,
+      "Ġuncont": 21706,
+      "Ġrefere": 21707,
+      "ĠNitrogen": 21708,
+      "Ġlifting": 21709,
+      "Ġforget": 21710,
+      "Ġfeelings": 21711,
+      "Ġdigits": 21712,
+      "Ġstratig": 21713,
+      "ypes": 21714,
+      "Ġcompositional": 21715,
+      "Ġsupernatants": 21716,
+      "Ġconflicting": 21717,
+      "Ġdisadvantage": 21718,
+      "adelphia": 21719,
+      "Pd": 21720,
+      "ĠCoupling": 21721,
+      "Ġexpenditure": 21722,
+      "iki": 21723,
+      "described": 21724,
+      "ĠRNase": 21725,
+      "ĠConvex": 21726,
+      "ĠBax": 21727,
+      "ungsten": 21728,
+      "Ġboiling": 21729,
+      "Ġbasement": 21730,
+      "ocardial": 21731,
+      "history": 21732,
+      "inton": 21733,
+      "trimethyl": 21734,
+      "Ġgrafting": 21735,
+      "ĠHardy": 21736,
+      "ĠFemale": 21737,
+      "ĠFollow": 21738,
+      "ĠEST": 21739,
+      "tistic": 21740,
+      "Open": 21741,
+      "Ġreflux": 21742,
+      "elements": 21743,
+      "Ġpolysaccharide": 21744,
+      "dims": 21745,
+      "acency": 21746,
+      "Ġbiore": 21747,
+      "capac": 21748,
+      "Ġoverexpressed": 21749,
+      "either": 21750,
+      "Ġlaid": 21751,
+      "Ġincision": 21752,
+      "Ġassets": 21753,
+      "inflammation": 21754,
+      "Ġreconstructions": 21755,
+      "Ġglomerular": 21756,
+      "Ġconvey": 21757,
+      "ĠCXCR": 21758,
+      "oro": 21759,
+      "Ġclassifying": 21760,
+      "Ġcope": 21761,
+      "Ġpd": 21762,
+      "linic": 21763,
+      "Ġchord": 21764,
+      "ĠAging": 21765,
+      "Ġpalm": 21766,
+      "Ġpermittivity": 21767,
+      "ĠReverse": 21768,
+      "Ġoffshore": 21769,
+      "Ġdoubt": 21770,
+      "imoto": 21771,
+      "ĠColomb": 21772,
+      "Ġrodent": 21773,
+      "ĠElectrochemical": 21774,
+      "ĠImprovement": 21775,
+      "inescent": 21776,
+      "ĠTriton": 21777,
+      "Ġtransfusion": 21778,
+      "Ġlocomotion": 21779,
+      "Ġdangerous": 21780,
+      "Ġweighed": 21781,
+      "ĠHSV": 21782,
+      "techn": 21783,
+      "ĠDiagram": 21784,
+      "Ġparietal": 21785,
+      "six": 21786,
+      "Ġtitles": 21787,
+      "ylon": 21788,
+      "Ġheuristics": 21789,
+      "Ġjaponic": 21790,
+      "Ġtranslations": 21791,
+      "Ġtiters": 21792,
+      "Ġworms": 21793,
+      "ĠDPP": 21794,
+      "Ġcytoskeleton": 21795,
+      "Mediated": 21796,
+      "ariance": 21797,
+      "thel": 21798,
+      "Ãħ": 21799,
+      "ĠInflammatory": 21800,
+      "Ġoscillating": 21801,
+      "Ġavoids": 21802,
+      "Define": 21803,
+      "ĠOlympics": 21804,
+      "ogel": 21805,
+      "Ġheme": 21806,
+      "Ġmicrop": 21807,
+      "Ġthreats": 21808,
+      "QCD": 21809,
+      "XRD": 21810,
+      "ĠCoupled": 21811,
+      "Ġlm": 21812,
+      "ĠHelic": 21813,
+      "Ġdischarged": 21814,
+      "Ġrooted": 21815,
+      "Ġalleviate": 21816,
+      "Ġcausality": 21817,
+      "ĠCrow": 21818,
+      "ĠMack": 21819,
+      "ĠAirport": 21820,
+      "Ġchemokine": 21821,
+      "Ġll": 21822,
+      "ĠNar": 21823,
+      "omyces": 21824,
+      "ethoxyphenyl": 21825,
+      "ĠDaily": 21826,
+      "ĠFinland": 21827,
+      "Ġhorn": 21828,
+      "ĠOrient": 21829,
+      "Ġionized": 21830,
+      "ĠYears": 21831,
+      "Ġquasipar": 21832,
+      "Ġpercutaneous": 21833,
+      "Phase": 21834,
+      "Ġforeground": 21835,
+      "ĠANAL": 21836,
+      "Ġincrements": 21837,
+      "stan": 21838,
+      "Ġspeculate": 21839,
+      "TX": 21840,
+      "Ġpile": 21841,
+      "Ġdic": 21842,
+      "ipy": 21843,
+      "window": 21844,
+      "neutral": 21845,
+      "ĠAtlas": 21846,
+      "ĠMTT": 21847,
+      "ĠNy": 21848,
+      "ĠVIII": 21849,
+      "ĠFilms": 21850,
+      "singular": 21851,
+      "remove": 21852,
+      "Length": 21853,
+      "ĠRece": 21854,
+      "wait": 21855,
+      "Ġpurchase": 21856,
+      "ĠWikipedia": 21857,
+      "ĠLars": 21858,
+      "Ġsyntactic": 21859,
+      "Ġactuator": 21860,
+      "ĠAKT": 21861,
+      "ĠBry": 21862,
+      "ĠResult": 21863,
+      "ĠVariational": 21864,
+      "Ġjudgment": 21865,
+      "JECT": 21866,
+      "ximab": 21867,
+      "Ġtraced": 21868,
+      "Ġcardiomyopathy": 21869,
+      "WN": 21870,
+      "ĠRodrig": 21871,
+      "bt": 21872,
+      "Ġbid": 21873,
+      "acle": 21874,
+      "amura": 21875,
+      "Ġepic": 21876,
+      "Ġpuzz": 21877,
+      "ĠSox": 21878,
+      "Ġinflux": 21879,
+      "ÃŃn": 21880,
+      "uloskeletal": 21881,
+      "Dim": 21882,
+      "ĠSCC": 21883,
+      "ĠRAS": 21884,
+      "mr": 21885,
+      "UI": 21886,
+      "Ġjun": 21887,
+      "ĠSpearman": 21888,
+      "Ġfairness": 21889,
+      "etz": 21890,
+      "ĠPPI": 21891,
+      "inance": 21892,
+      "enko": 21893,
+      "Ġgalact": 21894,
+      "öm": 21895,
+      "Ġexceptions": 21896,
+      "ĠCretaceous": 21897,
+      "MY": 21898,
+      "Resp": 21899,
+      "Ġpep": 21900,
+      "ĠOrd": 21901,
+      "STE": 21902,
+      "Ġhelicity": 21903,
+      "Ġofficer": 21904,
+      "Target": 21905,
+      "ĠNorwegian": 21906,
+      "Ġdehydration": 21907,
+      "ĠSIRT": 21908,
+      "ĠRobinson": 21909,
+      "ĠBenchmark": 21910,
+      "viral": 21911,
+      "Real": 21912,
+      "Ġdoxorubicin": 21913,
+      "Ġestimations": 21914,
+      "ĠCauc": 21915,
+      "Ġadditives": 21916,
+      "modes": 21917,
+      "ĠHend": 21918,
+      "Ġaccelerating": 21919,
+      "ĠGordon": 21920,
+      "ĠMagnet": 21921,
+      "Ġgonad": 21922,
+      "Ġpyrolysis": 21923,
+      "coholic": 21924,
+      "ĠPKC": 21925,
+      "SAR": 21926,
+      "Ġwinding": 21927,
+      "terious": 21928,
+      "ĠMountains": 21929,
+      "ĠSymbol": 21930,
+      "ĠMatthe": 21931,
+      "ĠShin": 21932,
+      "Script": 21933,
+      "rug": 21934,
+      "ĠmW": 21935,
+      "ĠISM": 21936,
+      "ĠNg": 21937,
+      "Ġappoint": 21938,
+      "ĠAIDS": 21939,
+      "Ġports": 21940,
+      "differential": 21941,
+      "ĠJes": 21942,
+      "ĠNeed": 21943,
+      "Ġlenses": 21944,
+      "ĠAMPK": 21945,
+      "à¤": 21946,
+      "leaf": 21947,
+      "ĠBron": 21948,
+      "Ġprofit": 21949,
+      "Local": 21950,
+      "ĠExamination": 21951,
+      "ĠChief": 21952,
+      "Ġopinions": 21953,
+      "ĠRound": 21954,
+      "formations": 21955,
+      "Ġcollinear": 21956,
+      "Ġdigested": 21957,
+      "lassical": 21958,
+      "ervative": 21959,
+      "Ġcephal": 21960,
+      "Ġdisadvantages": 21961,
+      "Ġïĥ·": 21962,
+      "Ġsubtracting": 21963,
+      "Ġweigh": 21964,
+      "Bound": 21965,
+      "DG": 21966,
+      "Ġinfluential": 21967,
+      "Ġtoxins": 21968,
+      "ĠBenjamin": 21969,
+      "ĠNumbers": 21970,
+      "crystal": 21971,
+      "Ġstocks": 21972,
+      "ĠBour": 21973,
+      "ĠCompeting": 21974,
+      "Ġacqu": 21975,
+      "tRNA": 21976,
+      "ĠSeparation": 21977,
+      "Ġtagged": 21978,
+      "Ġconject": 21979,
+      "ĠPrince": 21980,
+      "Ġgrazing": 21981,
+      "Ġreleases": 21982,
+      "ĠChallenge": 21983,
+      "ATPase": 21984,
+      "Ġemail": 21985,
+      "insically": 21986,
+      "ĠRegulatory": 21987,
+      "Message": 21988,
+      "Ġslit": 21989,
+      "Ġpolygon": 21990,
+      "Ġdoubling": 21991,
+      "Ġreceivers": 21992,
+      "Ġtracked": 21993,
+      "Ġengineer": 21994,
+      "stained": 21995,
+      "ĠDanish": 21996,
+      "shock": 21997,
+      "ĠMaz": 21998,
+      "Ġcough": 21999,
+      "ĠImmunohist": 22000,
+      "Consequ": 22001,
+      "armacy": 22002,
+      "Ġchemo": 22003,
+      "ĠMH": 22004,
+      "Ġemerges": 22005,
+      "Ġannealed": 22006,
+      "Ġhypothesize": 22007,
+      "ĠTypically": 22008,
+      "ĠBang": 22009,
+      "ĠPuls": 22010,
+      "Ġgirl": 22011,
+      "Ġherbiv": 22012,
+      "ĠANN": 22013,
+      "Ġseism": 22014,
+      "ĠCytok": 22015,
+      "ĠThroughout": 22016,
+      "Ġadaptations": 22017,
+      "lang": 22018,
+      "Ġclonal": 22019,
+      "umulation": 22020,
+      "ĠUniform": 22021,
+      "Ġhi": 22022,
+      "opent": 22023,
+      "Ġbutton": 22024,
+      "tene": 22025,
+      "Ġproteasome": 22026,
+      "bred": 22027,
+      "ĠNelson": 22028,
+      "racycline": 22029,
+      "ĠDY": 22030,
+      "Ġimmunoblot": 22031,
+      "prol": 22032,
+      "Ġpic": 22033,
+      "Ġcompilation": 22034,
+      "ĠDevices": 22035,
+      "etermined": 22036,
+      "ĠFrancis": 22037,
+      "notation": 22038,
+      "writing": 22039,
+      "terase": 22040,
+      "ĠStephen": 22041,
+      "amel": 22042,
+      "ĠChu": 22043,
+      "alone": 22044,
+      "Ġexhaust": 22045,
+      "relevant": 22046,
+      "ĠStrat": 22047,
+      "Ġparametrization": 22048,
+      "ĠBull": 22049,
+      "ĠRemote": 22050,
+      "increasing": 22051,
+      "Ġdd": 22052,
+      "âĢ°": 22053,
+      "yroidism": 22054,
+      "ilin": 22055,
+      "ĠHip": 22056,
+      "ICA": 22057,
+      "ĠApoptosis": 22058,
+      "Ġmachining": 22059,
+      "LDL": 22060,
+      "Ġgem": 22061,
+      "ĠFFT": 22062,
+      "ĠGuang": 22063,
+      "Ġoriginates": 22064,
+      "dat": 22065,
+      "cone": 22066,
+      "ĠAdoles": 22067,
+      "ucci": 22068,
+      "avoid": 22069,
+      "ulpt": 22070,
+      "urium": 22071,
+      "Ġliteracy": 22072,
+      "Recent": 22073,
+      "avg": 22074,
+      "Ġinvited": 22075,
+      "ĠPeak": 22076,
+      "ĠEnterobacter": 22077,
+      "Ġaneurysm": 22078,
+      "ĠMorris": 22079,
+      "tida": 22080,
+      "ĠSER": 22081,
+      "ĠMichel": 22082,
+      "ĠIBD": 22083,
+      "ĠNG": 22084,
+      "Ġscarce": 22085,
+      "web": 22086,
+      "Ġexpresses": 22087,
+      "Ġschema": 22088,
+      "Ġlessons": 22089,
+      "Ġarginine": 22090,
+      "Ġphotographs": 22091,
+      "ĠNeurons": 22092,
+      "ĠATPase": 22093,
+      "Ġfiller": 22094,
+      "rapped": 22095,
+      "Ġrandomness": 22096,
+      "Ġveins": 22097,
+      "Ġwounds": 22098,
+      "ĠApart": 22099,
+      "Ġracial": 22100,
+      "Ġnoteworthy": 22101,
+      "Ġremoves": 22102,
+      "Ġganglion": 22103,
+      "Ġlaminar": 22104,
+      "ĠSSR": 22105,
+      "Ġpolysaccharides": 22106,
+      "Ġbuf": 22107,
+      "Ġendothelium": 22108,
+      "ĠCAS": 22109,
+      "ĠGolgi": 22110,
+      "Ġinheritance": 22111,
+      "isite": 22112,
+      "COMP": 22113,
+      "Ġpt": 22114,
+      "Ġmeshes": 22115,
+      "Ġtherapeutics": 22116,
+      "OST": 22117,
+      "olinergic": 22118,
+      "UG": 22119,
+      "squared": 22120,
+      "Ġdegrade": 22121,
+      "uum": 22122,
+      "Ġretrosp": 22123,
+      "Loc": 22124,
+      "ĠJNK": 22125,
+      "Options": 22126,
+      "Ġinsulating": 22127,
+      "Ġspecifies": 22128,
+      "Ġoven": 22129,
+      "yy": 22130,
+      "ĠConver": 22131,
+      "Ġdisappeared": 22132,
+      "ĠPrognostic": 22133,
+      "ĠNguyen": 22134,
+      "Ġperiphery": 22135,
+      "bank": 22136,
+      "Ġimid": 22137,
+      "Ġassigning": 22138,
+      "ĠMess": 22139,
+      "propan": 22140,
+      "ioma": 22141,
+      "olyb": 22142,
+      "Ġepitope": 22143,
+      "Ġemitting": 22144,
+      "DIR": 22145,
+      "ync": 22146,
+      "Ġimpairments": 22147,
+      "ĠMicroscopy": 22148,
+      "ĠFWHM": 22149,
+      "gray": 22150,
+      "Ġfing": 22151,
+      "ucial": 22152,
+      "plemented": 22153,
+      "eas": 22154,
+      "estamp": 22155,
+      "Ġcrest": 22156,
+      "ĠMos": 22157,
+      "Ġneutrons": 22158,
+      "Ġbroth": 22159,
+      "Ġheadache": 22160,
+      "ongevity": 22161,
+      "Ġreass": 22162,
+      "ĠPSF": 22163,
+      "ĠBuch": 22164,
+      "visor": 22165,
+      "Ġdenoting": 22166,
+      "integer": 22167,
+      "ouin": 22168,
+      "efficacy": 22169,
+      "Ġglutamine": 22170,
+      "Ġpicked": 22171,
+      "ĠCampbell": 22172,
+      "ĠKernel": 22173,
+      "Ġships": 22174,
+      "lt": 22175,
+      "ondyl": 22176,
+      "Ġcredi": 22177,
+      "Ġpeptid": 22178,
+      "ĠEstabl": 22179,
+      "bons": 22180,
+      "Ġaggl": 22181,
+      "USE": 22182,
+      "supp": 22183,
+      "upsilon": 22184,
+      "characterized": 22185,
+      "isheries": 22186,
+      "May": 22187,
+      "ARC": 22188,
+      "Ġroads": 22189,
+      "Ġdeparture": 22190,
+      "ĠMAX": 22191,
+      "ĠTRA": 22192,
+      "imod": 22193,
+      "ĠAlber": 22194,
+      "Ġterminated": 22195,
+      "ölder": 22196,
+      "Scalar": 22197,
+      "hash": 22198,
+      "ĠMSS": 22199,
+      "Ġsmoothness": 22200,
+      "Ġresemble": 22201,
+      "ĠEffectiveness": 22202,
+      "rx": 22203,
+      "ĠEye": 22204,
+      "Ġfaecal": 22205,
+      "þ": 22206,
+      "ĠClostridium": 22207,
+      "achine": 22208,
+      "ĠBDNF": 22209,
+      "Ġcab": 22210,
+      "ĠWong": 22211,
+      "ĠDouglas": 22212,
+      "Ġreperfusion": 22213,
+      "ĠXi": 22214,
+      "Ġconfused": 22215,
+      "ĠPhiladelphia": 22216,
+      "Ġapple": 22217,
+      "Ġile": 22218,
+      "Ġfavored": 22219,
+      "Ġplaques": 22220,
+      "Ġtrivially": 22221,
+      "ĠTypical": 22222,
+      "Ġcentralized": 22223,
+      "ĠFacebook": 22224,
+      "Ġnortheast": 22225,
+      "Ġnormality": 22226,
+      "ĠTb": 22227,
+      "Ġapt": 22228,
+      "Ġfacet": 22229,
+      "ĠRenal": 22230,
+      "clk": 22231,
+      "Ġligation": 22232,
+      "ifferenti": 22233,
+      "Ġputting": 22234,
+      "Ġintrig": 22235,
+      "walled": 22236,
+      "Et": 22237,
+      "ĠCow": 22238,
+      "ĠNations": 22239,
+      "Ġcampus": 22240,
+      "ĠKinetics": 22241,
+      "ĠMexican": 22242,
+      "ERK": 22243,
+      "Ġlatitudes": 22244,
+      "ĠRoll": 22245,
+      "ĠQD": 22246,
+      "adaptive": 22247,
+      "Ġquenched": 22248,
+      "Ġfram": 22249,
+      "Qi": 22250,
+      "Ġtongue": 22251,
+      "edes": 22252,
+      "Ġascorb": 22253,
+      "ĠGlucose": 22254,
+      "ouri": 22255,
+      "Ġdefeated": 22256,
+      "ophilus": 22257,
+      "ralateral": 22258,
+      "xrightarrow": 22259,
+      "ĠJup": 22260,
+      "axes": 22261,
+      "eger": 22262,
+      "MIT": 22263,
+      "ĠMember": 22264,
+      "ĠNu": 22265,
+      "Ġtransloc": 22266,
+      "ĠFlux": 22267,
+      "ĠColorado": 22268,
+      "Ġrelying": 22269,
+      "atrol": 22270,
+      "Ġcontrasts": 22271,
+      "centage": 22272,
+      "Ġleukocyte": 22273,
+      "Ġcoincidence": 22274,
+      "Ġcontractions": 22275,
+      "oga": 22276,
+      "ANN": 22277,
+      "ĠAbsorption": 22278,
+      "Return": 22279,
+      "reprene": 22280,
+      "baum": 22281,
+      "traumatic": 22282,
+      "incial": 22283,
+      "Ġautophag": 22284,
+      "Ġalgorithmic": 22285,
+      "rimp": 22286,
+      "Ġdivides": 22287,
+      "ĠRose": 22288,
+      "ĠEric": 22289,
+      "Ġaddiction": 22290,
+      "plification": 22291,
+      "Ġdiffusive": 22292,
+      "ĠVehicle": 22293,
+      "enerate": 22294,
+      "tising": 22295,
+      "Ġstarvation": 22296,
+      "absorption": 22297,
+      "ĠAra": 22298,
+      "Ġgrav": 22299,
+      "ĠSubunit": 22300,
+      "Ġamide": 22301,
+      "Ġenhancer": 22302,
+      "Ġmerid": 22303,
+      "ermost": 22304,
+      "Ġalgal": 22305,
+      "ĠQueens": 22306,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 22307,
+      "Ġjudge": 22308,
+      "ĠGreenland": 22309,
+      "brace": 22310,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 22311,
+      "Ġhypergly": 22312,
+      "Ġemergent": 22313,
+      "Fisher": 22314,
+      "ĠLas": 22315,
+      "Ġsexes": 22316,
+      "Sep": 22317,
+      "Ġphrases": 22318,
+      "ĠSequential": 22319,
+      "inki": 22320,
+      "Ġaxioms": 22321,
+      "study": 22322,
+      "Ġtiny": 22323,
+      "Ġcd": 22324,
+      "catalyzed": 22325,
+      "asaki": 22326,
+      "ĠWR": 22327,
+      "ĠMinimal": 22328,
+      "Ġsubcellular": 22329,
+      "Ġphospho": 22330,
+      "ESI": 22331,
+      "Ġvow": 22332,
+      "Ġsupplies": 22333,
+      "operand": 22334,
+      "Fix": 22335,
+      "anian": 22336,
+      "writer": 22337,
+      "âĪ¶": 22338,
+      "Ġwinner": 22339,
+      "ĠPID": 22340,
+      "ĠLebesgue": 22341,
+      "Ġsimplification": 22342,
+      "ĠRelationships": 22343,
+      "Ġautomata": 22344,
+      "ĠContribution": 22345,
+      "Ġhereditary": 22346,
+      "errin": 22347,
+      "ĠBLAST": 22348,
+      "aea": 22349,
+      "yleth": 22350,
+      "ĠTc": 22351,
+      "adeh": 22352,
+      "adjuvant": 22353,
+      "Wave": 22354,
+      "counter": 22355,
+      "ĠGupta": 22356,
+      "ĠGhana": 22357,
+      "Cho": 22358,
+      "Ġourselves": 22359,
+      "Ġevenly": 22360,
+      "lymph": 22361,
+      "Ġcerebellum": 22362,
+      "Ġcopolymers": 22363,
+      "modular": 22364,
+      "Ġharder": 22365,
+      "Ġplease": 22366,
+      "ĠPSD": 22367,
+      "Ġlimbs": 22368,
+      "Ġexploitation": 22369,
+      "iry": 22370,
+      "Ġperiodontal": 22371,
+      "ATCH": 22372,
+      "Ġmalicious": 22373,
+      "ĠSlov": 22374,
+      "HY": 22375,
+      "Consequently": 22376,
+      "oren": 22377,
+      "ĠPare": 22378,
+      "agine": 22379,
+      "ĠROI": 22380,
+      "ĠWhich": 22381,
+      "ĠNative": 22382,
+      "amen": 22383,
+      "reshape": 22384,
+      "oplankton": 22385,
+      "Ġartifact": 22386,
+      "Ġrhin": 22387,
+      "gpu": 22388,
+      "Ġundet": 22389,
+      "Ġsporadic": 22390,
+      "Ġorally": 22391,
+      "Ġstepwise": 22392,
+      "ĠCohort": 22393,
+      "Ġrhod": 22394,
+      "cyt": 22395,
+      "Ġierr": 22396,
+      "Ġmotors": 22397,
+      "ĠIgE": 22398,
+      "calculated": 22399,
+      "ĠChampionships": 22400,
+      "pel": 22401,
+      "ĠFerr": 22402,
+      "Ġisometric": 22403,
+      "nutrition": 22404,
+      "Ġunsaturated": 22405,
+      "Ġdoll": 22406,
+      "ĠRMSE": 22407,
+      "Ġsolitary": 22408,
+      "approximation": 22409,
+      "Ġreperto": 22410,
+      "sight": 22411,
+      "Ġcranial": 22412,
+      "ilical": 22413,
+      "ĠOst": 22414,
+      "oul": 22415,
+      "Ġdg": 22416,
+      "ĠProceed": 22417,
+      "Ġmilling": 22418,
+      "sz": 22419,
+      "Ġmineralization": 22420,
+      "Ġcigarette": 22421,
+      "Ġporph": 22422,
+      "Ġspons": 22423,
+      "ĠGreece": 22424,
+      "ipore": 22425,
+      "accept": 22426,
+      "ĠPTSD": 22427,
+      "Å«": 22428,
+      "Ġcipher": 22429,
+      "Ġfunctionalized": 22430,
+      "Poly": 22431,
+      "Ġabd": 22432,
+      "flight": 22433,
+      "ĠSydney": 22434,
+      "Ġdisaster": 22435,
+      "ĠHaving": 22436,
+      "Ġdiesel": 22437,
+      "ĠGreg": 22438,
+      "Ġspans": 22439,
+      "ĠSeasonal": 22440,
+      "STEM": 22441,
+      "ierr": 22442,
+      "ĠIB": 22443,
+      "Ġlemm": 22444,
+      "anum": 22445,
+      "ĠBottom": 22446,
+      "Ġseal": 22447,
+      "boost": 22448,
+      "Ġlegend": 22449,
+      "bing": 22450,
+      "abis": 22451,
+      "Ġchitin": 22452,
+      "Ġmaximally": 22453,
+      "Ġimmunosuppressive": 22454,
+      "âĪĴâĪĴ": 22455,
+      "Ġabolished": 22456,
+      "ige": 22457,
+      "Ġesophag": 22458,
+      "Ġlasted": 22459,
+      "Ġcarbohydrates": 22460,
+      "Ġchips": 22461,
+      "ĠFernand": 22462,
+      "far": 22463,
+      "ĠPoints": 22464,
+      "calation": 22465,
+      "ĠRegions": 22466,
+      "CHK": 22467,
+      "veratrol": 22468,
+      "truth": 22469,
+      "Ġstrange": 22470,
+      "Interest": 22471,
+      "sho": 22472,
+      "ĠInduc": 22473,
+      "Ġmigraine": 22474,
+      "ĠVac": 22475,
+      "ophores": 22476,
+      "Ġerrone": 22477,
+      "scriptsize": 22478,
+      "ĠNeutron": 22479,
+      "Ġindistinguishable": 22480,
+      "istine": 22481,
+      "Ġhelper": 22482,
+      "specified": 22483,
+      "Ġjuice": 22484,
+      "oxal": 22485,
+      "ĠJung": 22486,
+      "Ġmagazine": 22487,
+      "Ġtelephone": 22488,
+      "ĠPhyt": 22489,
+      "Ġum": 22490,
+      "ĠAvailability": 22491,
+      "ĠTropical": 22492,
+      "ĠCases": 22493,
+      "Ġdescend": 22494,
+      "Har": 22495,
+      "âĪĹ": 22496,
+      "ĠâĨĵ": 22497,
+      "Ks": 22498,
+      "Ġê": 22499,
+      "oluble": 22500,
+      "Ġchampionship": 22501,
+      "ĠMovement": 22502,
+      "ĠXY": 22503,
+      "kappaB": 22504,
+      "years": 22505,
+      "memb": 22506,
+      "quine": 22507,
+      "Ġletting": 22508,
+      "Ġbiggest": 22509,
+      "Ġcards": 22510,
+      "Ġbiotin": 22511,
+      "ĠAur": 22512,
+      "modal": 22513,
+      "Ġvaccinated": 22514,
+      "Ġtranslates": 22515,
+      "ĠPAC": 22516,
+      "lli": 22517,
+      "reonine": 22518,
+      "Ġcurcumin": 22519,
+      "ĠConstruct": 22520,
+      "Ġconvinc": 22521,
+      "ĠNat": 22522,
+      "Ġamygdala": 22523,
+      "Ġprotr": 22524,
+      "ĠSingular": 22525,
+      "ĠContact": 22526,
+      "kind": 22527,
+      "ĠDaw": 22528,
+      "ogroup": 22529,
+      "ĠKCl": 22530,
+      "Ġhygi": 22531,
+      "erenced": 22532,
+      "Ġsurveyed": 22533,
+      "ĠMull": 22534,
+      "esthetic": 22535,
+      "Ġmsg": 22536,
+      "ĠRequire": 22537,
+      "Ġdistortions": 22538,
+      "Control": 22539,
+      "BERT": 22540,
+      "Ġautonomic": 22541,
+      "Ġhormonal": 22542,
+      "Ġstrips": 22543,
+      "Ġtrophic": 22544,
+      "ifting": 22545,
+      "opod": 22546,
+      "ĠSpontaneous": 22547,
+      "Ġlogs": 22548,
+      "OPT": 22549,
+      "ĠMot": 22550,
+      "ĠGmb": 22551,
+      "aharan": 22552,
+      "ĠPOL": 22553,
+      "Ġvisceral": 22554,
+      "blocks": 22555,
+      "Ġsitting": 22556,
+      "Ġsine": 22557,
+      "Ġoncogenic": 22558,
+      "ERRQ": 22559,
+      "quinone": 22560,
+      "Ġsmartphone": 22561,
+      "ĠTanz": 22562,
+      "lactam": 22563,
+      "ĠSignificance": 22564,
+      "Ġeu": 22565,
+      "ĠISS": 22566,
+      "ĠTrig": 22567,
+      "ĠMaj": 22568,
+      "tingale": 22569,
+      "Ġdilat": 22570,
+      "ennes": 22571,
+      "ĠBelgium": 22572,
+      "lev": 22573,
+      "ĠContr": 22574,
+      "ĠGalois": 22575,
+      "ĠCombination": 22576,
+      "ĠThi": 22577,
+      "ĠAustria": 22578,
+      "Prom": 22579,
+      "Ġelicit": 22580,
+      "biosis": 22581,
+      "Ġlymphatic": 22582,
+      "ĠMurray": 22583,
+      "ĠXPS": 22584,
+      "Ġcong": 22585,
+      "screen": 22586,
+      "tide": 22587,
+      "amoyl": 22588,
+      "ĠMcD": 22589,
+      "Ġretired": 22590,
+      "mixed": 22591,
+      "ELD": 22592,
+      "ĠMaps": 22593,
+      "ĠVE": 22594,
+      "cession": 22595,
+      "numer": 22596,
+      "idated": 22597,
+      "ĠBishop": 22598,
+      "Ġneonates": 22599,
+      "Ġlandsl": 22600,
+      "ĠFractional": 22601,
+      "Ġspecifying": 22602,
+      "ĠJr": 22603,
+      "Ġnanowire": 22604,
+      "Ġconsultation": 22605,
+      "language": 22606,
+      "Ġpricing": 22607,
+      "ĠLimitations": 22608,
+      "ĠPediatric": 22609,
+      "ĠDimension": 22610,
+      "Ġpreparing": 22611,
+      "Lag": 22612,
+      "segment": 22613,
+      "Ġspend": 22614,
+      "athe": 22615,
+      "Ġweap": 22616,
+      "ĠJos": 22617,
+      "textit": 22618,
+      "outputs": 22619,
+      "ordering": 22620,
+      "Ġplacenta": 22621,
+      "ationally": 22622,
+      "ĠKun": 22623,
+      "Ġoutstanding": 22624,
+      "Ġthicknesses": 22625,
+      "ĠChIP": 22626,
+      "deoxy": 22627,
+      "ĠZo": 22628,
+      "ĠDeveloping": 22629,
+      "Ġstringent": 22630,
+      "iency": 22631,
+      "perse": 22632,
+      "Ġpend": 22633,
+      "ĠDevelopmental": 22634,
+      "Ġextern": 22635,
+      "Ġinverter": 22636,
+      "ĠDAPI": 22637,
+      "lectivity": 22638,
+      "Ġtablets": 22639,
+      "Ġprogester": 22640,
+      "ĠïģŃ": 22641,
+      "Ġanswered": 22642,
+      "entary": 22643,
+      "ORS": 22644,
+      "Ġdir": 22645,
+      "Ġdeleterious": 22646,
+      "Ġdopaminergic": 22647,
+      "Random": 22648,
+      "diss": 22649,
+      "Ġmonolayers": 22650,
+      "Ġintegrand": 22651,
+      "ĠComponents": 22652,
+      "ĠPerc": 22653,
+      "agit": 22654,
+      "ARN": 22655,
+      "esophageal": 22656,
+      "ivan": 22657,
+      "neider": 22658,
+      "ĠStarting": 22659,
+      "PORT": 22660,
+      "yellow": 22661,
+      "Ġregisters": 22662,
+      "pairs": 22663,
+      "Ġethnicity": 22664,
+      "Ġboy": 22665,
+      "auti": 22666,
+      "Ġchromium": 22667,
+      "POS": 22668,
+      "vature": 22669,
+      "ayashi": 22670,
+      "Ġinappropriate": 22671,
+      "ĠSNA": 22672,
+      "Domain": 22673,
+      "ĠPrice": 22674,
+      "Ġmacular": 22675,
+      "Ġoverload": 22676,
+      "ĠUnified": 22677,
+      "Ġattach": 22678,
+      "ĠScottish": 22679,
+      "maps": 22680,
+      "agl": 22681,
+      "emi": 22682,
+      "Ġseam": 22683,
+      "ĠAnalog": 22684,
+      "dated": 22685,
+      "uo": 22686,
+      "Ġplated": 22687,
+      "Ġasset": 22688,
+      "Ġscreens": 22689,
+      "Ġspurious": 22690,
+      "Besides": 22691,
+      "Ġbaselines": 22692,
+      "heads": 22693,
+      "Ġcoat": 22694,
+      "ĠRemoval": 22695,
+      "Ġinfinitesimal": 22696,
+      "ĠTransformation": 22697,
+      "Ġcommens": 22698,
+      "Float": 22699,
+      "AUC": 22700,
+      "ĠLay": 22701,
+      "Ġintron": 22702,
+      "ĠDetecting": 22703,
+      "ĠHerein": 22704,
+      "ĠAssociations": 22705,
+      "Ġprogesterone": 22706,
+      "Bacteria": 22707,
+      "Ġsentiment": 22708,
+      "ĠPhenomen": 22709,
+      "matter": 22710,
+      "Ġcylinders": 22711,
+      "Ġtoluene": 22712,
+      "Ġspatiotemporal": 22713,
+      "Ġlanding": 22714,
+      "ĠCoronavirus": 22715,
+      "ĠBerry": 22716,
+      "ĠBragg": 22717,
+      "Ġregistry": 22718,
+      "Ġenthalpy": 22719,
+      "tica": 22720,
+      "razine": 22721,
+      "Ġcargo": 22722,
+      "otation": 22723,
+      "Ġcontradicts": 22724,
+      "Ġpesticides": 22725,
+      "ĠFischer": 22726,
+      "Ġmechanically": 22727,
+      "ĠInterfer": 22728,
+      "ĠCyp": 22729,
+      "ĠKas": 22730,
+      "Ġmetres": 22731,
+      "Ġantiretroviral": 22732,
+      "Ġtravers": 22733,
+      "selection": 22734,
+      "ĠWA": 22735,
+      "Ġdoublet": 22736,
+      "meta": 22737,
+      "ENTR": 22738,
+      "sonic": 22739,
+      "Ġmarking": 22740,
+      "ĠOverex": 22741,
+      "Ġpyruvate": 22742,
+      "Ġextrusion": 22743,
+      "Ġingestion": 22744,
+      "Ġcocaine": 22745,
+      "ĠFellow": 22746,
+      "CNTs": 22747,
+      "BG": 22748,
+      "ĠMorphological": 22749,
+      "Ġdefence": 22750,
+      "ĠYosh": 22751,
+      "mitter": 22752,
+      "rystallization": 22753,
+      "STRACT": 22754,
+      "Ġinflammasome": 22755,
+      "ĠGd": 22756,
+      "Ġshaft": 22757,
+      "Ġeruption": 22758,
+      "ĠOxide": 22759,
+      "ifolds": 22760,
+      "ĠGam": 22761,
+      "ĠGap": 22762,
+      "command": 22763,
+      "ĠIgA": 22764,
+      "Ġshortening": 22765,
+      "assembled": 22766,
+      "isopropyl": 22767,
+      "Ġalumina": 22768,
+      "ĠATM": 22769,
+      "Ġct": 22770,
+      "Ġspinning": 22771,
+      "ĠPetsc": 22772,
+      "prefix": 22773,
+      "Ġperpetuity": 22774,
+      "PRE": 22775,
+      "Ġfruct": 22776,
+      "GHz": 22777,
+      "elike": 22778,
+      "enyl": 22779,
+      "Ġwherein": 22780,
+      "UK": 22781,
+      "visual": 22782,
+      "lipidemia": 22783,
+      "reduction": 22784,
+      "anin": 22785,
+      "olas": 22786,
+      "Ġamplic": 22787,
+      "ĠSAT": 22788,
+      "Ġmodulator": 22789,
+      "forth": 22790,
+      "rl": 22791,
+      "Ġcrew": 22792,
+      "ĠiP": 22793,
+      "Ġxi": 22794,
+      "ADD": 22795,
+      "ĠAlexand": 22796,
+      "constrained": 22797,
+      "ratory": 22798,
+      "ĠkW": 22799,
+      "ĠMDR": 22800,
+      "ĠlncRNA": 22801,
+      "Mill": 22802,
+      "ĠMgO": 22803,
+      "circuit": 22804,
+      "Ġpersonalized": 22805,
+      "ĠOperator": 22806,
+      "stock": 22807,
+      "ĠPSA": 22808,
+      "ensable": 22809,
+      "Ġlean": 22810,
+      "yield": 22811,
+      "Ġopacity": 22812,
+      "ĠCommons": 22813,
+      "Ġsummed": 22814,
+      "ucker": 22815,
+      "ecke": 22816,
+      "epithelial": 22817,
+      "Ġasking": 22818,
+      "uese": 22819,
+      "ĠFlav": 22820,
+      "Ġlactic": 22821,
+      "Ġlubric": 22822,
+      "Ġisn": 22823,
+      "regions": 22824,
+      "support": 22825,
+      "Below": 22826,
+      "ĠNom": 22827,
+      "Ġhyal": 22828,
+      "ikh": 22829,
+      "ban": 22830,
+      "ĠBG": 22831,
+      "rometer": 22832,
+      "indic": 22833,
+      "opharyngeal": 22834,
+      "ITION": 22835,
+      "ĠPropagation": 22836,
+      "ĠPlace": 22837,
+      "ĠCircuit": 22838,
+      "ĠCOL": 22839,
+      "Green": 22840,
+      "Ir": 22841,
+      "lav": 22842,
+      "ĠdS": 22843,
+      "ĠMoment": 22844,
+      "Ġinducible": 22845,
+      "Ġdischarges": 22846,
+      "habdi": 22847,
+      "ĠExperience": 22848,
+      "Ġsg": 22849,
+      "Ġoutward": 22850,
+      "Ġportable": 22851,
+      "ĠOperators": 22852,
+      "Av": 22853,
+      "ĠDQ": 22854,
+      "ostatin": 22855,
+      "Ġeosinophil": 22856,
+      "Ġstriatum": 22857,
+      "ĠConsensus": 22858,
+      "Ġimperfect": 22859,
+      "NOT": 22860,
+      "ĠDemocratic": 22861,
+      ";;": 22862,
+      "Body": 22863,
+      "dii": 22864,
+      "Ho": 22865,
+      "ĠRailway": 22866,
+      "ĠUganda": 22867,
+      "Ġunpaired": 22868,
+      "friendly": 22869,
+      "Ġreprogramming": 22870,
+      "Alternative": 22871,
+      "RG": 22872,
+      "imet": 22873,
+      "enez": 22874,
+      "ĠHypothesis": 22875,
+      "Ġton": 22876,
+      "ĠCombin": 22877,
+      "ĠDelivery": 22878,
+      "Last": 22879,
+      "Ġowners": 22880,
+      "razole": 22881,
+      "ĠKob": 22882,
+      "Ġformats": 22883,
+      "Ġpolyclonal": 22884,
+      "Ġidentifier": 22885,
+      "ILL": 22886,
+      "Ġsurgeon": 22887,
+      "Ġpostp": 22888,
+      "ĠGenerative": 22889,
+      "ĠMall": 22890,
+      "abc": 22891,
+      "ĠHaz": 22892,
+      "Ġsmoothly": 22893,
+      "Ġcrystallographic": 22894,
+      "ĠFDA": 22895,
+      "Ġcoexistence": 22896,
+      "ionized": 22897,
+      "Ġcompiler": 22898,
+      "ĠArter": 22899,
+      "Ġappearances": 22900,
+      "amiltonian": 22901,
+      "Ġencapsulated": 22902,
+      "atia": 22903,
+      "wi": 22904,
+      "reb": 22905,
+      "Ġwafer": 22906,
+      "ubs": 22907,
+      "ĠUE": 22908,
+      "ĠGSK": 22909,
+      "Ġviv": 22910,
+      "Ġflooding": 22911,
+      "ĠGyr": 22912,
+      "Ġstably": 22913,
+      "Ġdislocations": 22914,
+      "Ġescap": 22915,
+      "ĠPhysiological": 22916,
+      "tidal": 22917,
+      "yme": 22918,
+      "ĠMaxim": 22919,
+      "iterator": 22920,
+      "ordant": 22921,
+      "Ġattentional": 22922,
+      "Ġcatalyzed": 22923,
+      "ĠTryp": 22924,
+      "PIN": 22925,
+      "ĠCorrelations": 22926,
+      "Ġhydrological": 22927,
+      "Ġnose": 22928,
+      "export": 22929,
+      "Ġdext": 22930,
+      "ĠBenef": 22931,
+      "ĠBiosystems": 22932,
+      "ĠPars": 22933,
+      "Ġreadings": 22934,
+      "Ġinstrumentation": 22935,
+      "ĠIQ": 22936,
+      "RIC": 22937,
+      "Ġgrafts": 22938,
+      "overs": 22939,
+      "ĠMedic": 22940,
+      "Ġmonod": 22941,
+      "Ġuniformity": 22942,
+      "ĠATLAS": 22943,
+      "Ġmasked": 22944,
+      "Ri": 22945,
+      "ĠPhysic": 22946,
+      "Ġimposing": 22947,
+      "ĠParad": 22948,
+      "imetic": 22949,
+      "Ġdemanding": 22950,
+      "unks": 22951,
+      "Ġfolds": 22952,
+      "ĠAnc": 22953,
+      "Ġvolatility": 22954,
+      "Ġbringing": 22955,
+      "acil": 22956,
+      "ĠNMDA": 22957,
+      "reduced": 22958,
+      "tii": 22959,
+      "Ġnorthwest": 22960,
+      "ĠBessel": 22961,
+      "ventions": 22962,
+      "Ġconsolidation": 22963,
+      "Meier": 22964,
+      "Ġmicrof": 22965,
+      "Ġqualified": 22966,
+      "Ġinsignificant": 22967,
+      "ĠMorphology": 22968,
+      "Ġpointwise": 22969,
+      "Ġlearns": 22970,
+      "Ġguard": 22971,
+      "CHECK": 22972,
+      "phonon": 22973,
+      "ĠEnhancement": 22974,
+      "Ġzonal": 22975,
+      "ERG": 22976,
+      "Start": 22977,
+      "Ġhistoric": 22978,
+      "ĠPure": 22979,
+      "ĠGmbH": 22980,
+      "glu": 22981,
+      "Ġpatterning": 22982,
+      "Ġstick": 22983,
+      "uminosity": 22984,
+      "Dataset": 22985,
+      "Ġoverride": 22986,
+      "ĠSteel": 22987,
+      "Ġfuels": 22988,
+      "mechanical": 22989,
+      "Ġautologous": 22990,
+      "Ġdepartments": 22991,
+      "ĠBlo": 22992,
+      "Ġimported": 22993,
+      "Ġrestrictive": 22994,
+      "eigen": 22995,
+      "ĠRome": 22996,
+      "ĠÌĬ": 22997,
+      "Ġepitopes": 22998,
+      "Ġlabelling": 22999,
+      "Ġownership": 23000,
+      "ĠEspecially": 23001,
+      "Ġcoffee": 23002,
+      "ĠGRB": 23003,
+      "Head": 23004,
+      "ĠVent": 23005,
+      "esare": 23006,
+      "ĠParticles": 23007,
+      "UNCTION": 23008,
+      "jj": 23009,
+      "uents": 23010,
+      "elic": 23011,
+      "ĠTat": 23012,
+      "ĠFle": 23013,
+      "Ġgating": 23014,
+      "Ġrefuge": 23015,
+      "Additionally": 23016,
+      "Ġrhs": 23017,
+      "Ġmaybe": 23018,
+      "ĠFang": 23019,
+      "Ġadvent": 23020,
+      "otransferase": 23021,
+      "should": 23022,
+      "Ġproteomic": 23023,
+      "Ġlegitim": 23024,
+      "PERIM": 23025,
+      "ĠGiant": 23026,
+      "Ġgraphics": 23027,
+      "onomical": 23028,
+      "scatter": 23029,
+      "Ġsuggestive": 23030,
+      "plots": 23031,
+      "Ġmultidrug": 23032,
+      "Ġabsorber": 23033,
+      "XS": 23034,
+      "consuming": 23035,
+      "Ġsustainability": 23036,
+      "opre": 23037,
+      "fix": 23038,
+      "Ġvolcano": 23039,
+      "ĠTypes": 23040,
+      "ĠCreate": 23041,
+      "Ġchooses": 23042,
+      "Ġstirring": 23043,
+      "Ġsurgeons": 23044,
+      "dS": 23045,
+      "Ġcharacterizes": 23046,
+      "Ġadjustments": 23047,
+      "texttt": 23048,
+      "etra": 23049,
+      "Ġclassifications": 23050,
+      "spots": 23051,
+      "ĠâĻ¯": 23052,
+      "erex": 23053,
+      "dehyd": 23054,
+      "ĠBrig": 23055,
+      "ĠSuperconduc": 23056,
+      "Ġgrants": 23057,
+      "ĠCen": 23058,
+      "ĠYin": 23059,
+      "ĠReactions": 23060,
+      "description": 23061,
+      "transcription": 23062,
+      "important": 23063,
+      "Ġhemodynamic": 23064,
+      "ĠYi": 23065,
+      "ĠGolden": 23066,
+      "kk": 23067,
+      "alb": 23068,
+      "Ġrooms": 23069,
+      "Ġsegreg": 23070,
+      "Ġsumming": 23071,
+      "Ġsuccession": 23072,
+      "Ġfollicular": 23073,
+      "Ġtackle": 23074,
+      "Down": 23075,
+      "Ġevaluates": 23076,
+      "atica": 23077,
+      "annual": 23078,
+      "ĠAlbert": 23079,
+      "Ġtal": 23080,
+      "orbital": 23081,
+      "fted": 23082,
+      "variables": 23083,
+      "Ġwetland": 23084,
+      "outheastern": 23085,
+      "MEM": 23086,
+      "ĠBrill": 23087,
+      "ĠSodium": 23088,
+      "ĠAlexa": 23089,
+      "umed": 23090,
+      "BUG": 23091,
+      "arine": 23092,
+      "Ġrevenue": 23093,
+      "habditis": 23094,
+      "Ġdissol": 23095,
+      "amplitude": 23096,
+      "Ġartists": 23097,
+      "Ġnormalised": 23098,
+      "Ġfluctuating": 23099,
+      "Ġaspar": 23100,
+      "ĠFi": 23101,
+      "olates": 23102,
+      "ispanic": 23103,
+      "Ġacetylation": 23104,
+      "ĠConcentration": 23105,
+      "Ġthro": 23106,
+      "shots": 23107,
+      "Ġnarrative": 23108,
+      "ĠWaals": 23109,
+      "ammonium": 23110,
+      "ureau": 23111,
+      "------------": 23112,
+      "Ġresearches": 23113,
+      "Ġbaby": 23114,
+      "Ġsharply": 23115,
+      "ÙĦ": 23116,
+      "ĠCel": 23117,
+      "CX": 23118,
+      "uminal": 23119,
+      "Ġgermline": 23120,
+      "ĠTransformer": 23121,
+      "pseud": 23122,
+      "HG": 23123,
+      "Ka": 23124,
+      "ĠSMC": 23125,
+      "ĠNutrition": 23126,
+      "Ġbarc": 23127,
+      "ĠWrite": 23128,
+      "Ġproteases": 23129,
+      "Ġsweep": 23130,
+      "ĠKolmogorov": 23131,
+      "morph": 23132,
+      "inducible": 23133,
+      "Ġexciting": 23134,
+      "lein": 23135,
+      "ĠHass": 23136,
+      "Ġproductive": 23137,
+      "mesh": 23138,
+      "ĠCMS": 23139,
+      "Ġheavier": 23140,
+      "Ġmeetings": 23141,
+      "ĠCopper": 23142,
+      "Ġvirtue": 23143,
+      "asant": 23144,
+      "ĠDEN": 23145,
+      "Ġinherently": 23146,
+      "rio": 23147,
+      "Ġhoused": 23148,
+      "Ġintraoperative": 23149,
+      "Ġcrown": 23150,
+      "conditions": 23151,
+      "ANG": 23152,
+      "YSIS": 23153,
+      "iman": 23154,
+      "Ġnmol": 23155,
+      "ĠRetrieval": 23156,
+      "algae": 23157,
+      "Ġkappa": 23158,
+      "deep": 23159,
+      "inence": 23160,
+      "ĠCarcinoma": 23161,
+      "Ġchromatographic": 23162,
+      "Ġascribed": 23163,
+      "Ġleverage": 23164,
+      "ĠKK": 23165,
+      "omyel": 23166,
+      "pet": 23167,
+      "ĠNJ": 23168,
+      "comm": 23169,
+      "Ġannually": 23170,
+      "gran": 23171,
+      "Ġaval": 23172,
+      "ĠNish": 23173,
+      "Ġevac": 23174,
+      "Ġmultif": 23175,
+      "Ġfunds": 23176,
+      "enny": 23177,
+      "ĠMong": 23178,
+      "ĠException": 23179,
+      "paths": 23180,
+      "ymen": 23181,
+      "hpp": 23182,
+      "Ġrestricting": 23183,
+      "saturated": 23184,
+      "âĻ": 23185,
+      "Ġlearners": 23186,
+      "ĠLanka": 23187,
+      "inities": 23188,
+      "ĠGDP": 23189,
+      "Ġspeciation": 23190,
+      "Ġensured": 23191,
+      "Ġneutralizing": 23192,
+      "Ġballoon": 23193,
+      "Comparison": 23194,
+      "ĠCalibration": 23195,
+      "ĠInfluenza": 23196,
+      "Ġvapour": 23197,
+      "XA": 23198,
+      "tracking": 23199,
+      "ĠICD": 23200,
+      "fluoro": 23201,
+      "ĠDamage": 23202,
+      "Ġpra": 23203,
+      "Ġconceived": 23204,
+      "ĠCosmological": 23205,
+      "Ġloose": 23206,
+      "inositol": 23207,
+      "ĠClifford": 23208,
+      "owa": 23209,
+      "Ġoffsets": 23210,
+      "document": 23211,
+      "Ġenormous": 23212,
+      "Ġphotoelectron": 23213,
+      "record": 23214,
+      "esticular": 23215,
+      "Ġvocals": 23216,
+      "Ġconsciousness": 23217,
+      "Ġtrem": 23218,
+      "Ġlandscapes": 23219,
+      "ĠFundamental": 23220,
+      "tebrate": 23221,
+      "Ġvertebral": 23222,
+      "Ġregenerative": 23223,
+      "Ġtroposp": 23224,
+      "Integr": 23225,
+      "Ġassociates": 23226,
+      "oved": 23227,
+      "ussed": 23228,
+      "aws": 23229,
+      "ĠSide": 23230,
+      "Ġinterconnected": 23231,
+      "Ġsuperfamily": 23232,
+      "ĠCook": 23233,
+      "loader": 23234,
+      "Ġpython": 23235,
+      "ĠCounter": 23236,
+      "books": 23237,
+      "Ġïģ²": 23238,
+      "breaking": 23239,
+      "gy": 23240,
+      "Ġcarbox": 23241,
+      "Ġedited": 23242,
+      "otyped": 23243,
+      "Ġduoden": 23244,
+      "anne": 23245,
+      "Ġanastom": 23246,
+      "ginate": 23247,
+      "ĠBiosciences": 23248,
+      "rage": 23249,
+      "ĠChiral": 23250,
+      "Ġsimplifies": 23251,
+      "Ġtestis": 23252,
+      "ström": 23253,
+      "ials": 23254,
+      "Ġmicelles": 23255,
+      "correct": 23256,
+      "ĠGenetics": 23257,
+      "along": 23258,
+      "Rem": 23259,
+      "resistance": 23260,
+      "Ġdrink": 23261,
+      "orbed": 23262,
+      "ĠTreat": 23263,
+      "ĠSho": 23264,
+      "shows": 23265,
+      "ér": 23266,
+      "Ġmimics": 23267,
+      "occup": 23268,
+      "eclam": 23269,
+      "ONG": 23270,
+      "Ġmarketing": 23271,
+      "ĠFinding": 23272,
+      "Ġendometri": 23273,
+      "âĶĢ": 23274,
+      "strained": 23275,
+      "ĠMuch": 23276,
+      "Ġexons": 23277,
+      "ĠHil": 23278,
+      "TD": 23279,
+      "ĠWW": 23280,
+      "ĠVic": 23281,
+      "enda": 23282,
+      "Ġfactory": 23283,
+      "ĠHepG": 23284,
+      "ĠStatic": 23285,
+      "blastoma": 23286,
+      "wd": 23287,
+      "raisal": 23288,
+      "ĠBasis": 23289,
+      "Ins": 23290,
+      "ĠUnsupervised": 23291,
+      "elo": 23292,
+      "oselective": 23293,
+      "Ġaccomplish": 23294,
+      "ĠProspective": 23295,
+      "Ġuncorrelated": 23296,
+      "ĠGate": 23297,
+      "icycl": 23298,
+      "Ġurgent": 23299,
+      "ĠPathways": 23300,
+      "Ġoblique": 23301,
+      "ĠIndividuals": 23302,
+      "Ġinitiative": 23303,
+      "Ġcatast": 23304,
+      "jections": 23305,
+      "Ġautosomal": 23306,
+      "ĠPhilip": 23307,
+      "Ġcomprehension": 23308,
+      "mM": 23309,
+      "pain": 23310,
+      "ĠmicroM": 23311,
+      "Ġencounters": 23312,
+      "goto": 23313,
+      "Ġladder": 23314,
+      "Ġoccupy": 23315,
+      "ĠSurfaces": 23316,
+      "Doc": 23317,
+      "ugby": 23318,
+      "Ġexamines": 23319,
+      "osynthesis": 23320,
+      "ĠKEGG": 23321,
+      "glass": 23322,
+      "slice": 23323,
+      "propagation": 23324,
+      "stry": 23325,
+      "Ġillustrating": 23326,
+      "imi": 23327,
+      "Ġspores": 23328,
+      "Ġastrophysical": 23329,
+      "Ġenclosed": 23330,
+      "Ġinferences": 23331,
+      "Ġbijection": 23332,
+      "Ġeveryday": 23333,
+      "Ġalternatively": 23334,
+      "reaction": 23335,
+      "iants": 23336,
+      "contact": 23337,
+      "Ġging": 23338,
+      "ĠBias": 23339,
+      "Ġautomaton": 23340,
+      "background": 23341,
+      "Ġneighbouring": 23342,
+      "Ġdetects": 23343,
+      "porate": 23344,
+      "ĠSharma": 23345,
+      "Hydro": 23346,
+      "Ġsacc": 23347,
+      "ĠFiber": 23348,
+      "ĠChlam": 23349,
+      "Ġbuffers": 23350,
+      "Applying": 23351,
+      "lceil": 23352,
+      "emph": 23353,
+      "ĠGSE": 23354,
+      "metry": 23355,
+      "Ġimmunost": 23356,
+      "ĠHistorical": 23357,
+      "ĠDrag": 23358,
+      "Ġtransplanted": 23359,
+      "Ġfrail": 23360,
+      "Ġanthocyan": 23361,
+      "inte": 23362,
+      "ĠBhat": 23363,
+      "ĠOg": 23364,
+      "Ġsteering": 23365,
+      "benzene": 23366,
+      "****************************************************************": 23367,
+      "Ġsynthet": 23368,
+      "Act": 23369,
+      "Ġcin": 23370,
+      "Ġherbal": 23371,
+      "Ġdyn": 23372,
+      "Ġhyperplasia": 23373,
+      "header": 23374,
+      "Ġcalculates": 23375,
+      "ĠDifference": 23376,
+      "Ġbats": 23377,
+      "ductivity": 23378,
+      "Ġconformations": 23379,
+      "city": 23380,
+      "Ġseparates": 23381,
+      "ĠCDC": 23382,
+      "ĠPrism": 23383,
+      "ĠBehaviour": 23384,
+      "ĠKelly": 23385,
+      "ĠSey": 23386,
+      "ĠÃł": 23387,
+      "LEX": 23388,
+      "gkin": 23389,
+      "strom": 23390,
+      "Ġvom": 23391,
+      "ĠWin": 23392,
+      "ĠWigner": 23393,
+      "Ġcontralateral": 23394,
+      "ĠMinor": 23395,
+      "Ġstereo": 23396,
+      "ĠApproximately": 23397,
+      "LED": 23398,
+      "say": 23399,
+      "ĠJS": 23400,
+      "Ġalcohols": 23401,
+      "Ġsan": 23402,
+      "Ġhardening": 23403,
+      "IFN": 23404,
+      "Ġretrospectively": 23405,
+      "Ġgeneralised": 23406,
+      "Ġtibial": 23407,
+      "ĠWeek": 23408,
+      "Ġaryl": 23409,
+      "ĠPeninsula": 23410,
+      "Ġdeterminations": 23411,
+      "Ġphotovoltaic": 23412,
+      "Ġsuggestion": 23413,
+      "Jac": 23414,
+      "ĠVitro": 23415,
+      "Ġcyclo": 23416,
+      "Ġfibrous": 23417,
+      "disambiguation": 23418,
+      "program": 23419,
+      "Ġguest": 23420,
+      "ĠDust": 23421,
+      "rceil": 23422,
+      "Ġpowered": 23423,
+      "Ġcardiomyocytes": 23424,
+      "heat": 23425,
+      "ylic": 23426,
+      "Ġpresentations": 23427,
+      "Ġtransmitting": 23428,
+      "WD": 23429,
+      "added": 23430,
+      "Initial": 23431,
+      "Del": 23432,
+      "ĠVelocity": 23433,
+      "Ġmole": 23434,
+      "Ġoval": 23435,
+      "Ġplankton": 23436,
+      "their": 23437,
+      "ĠQED": 23438,
+      "volutions": 23439,
+      "Ġmandatory": 23440,
+      "Ġrepulsive": 23441,
+      "ĉĠĠ": 23442,
+      "Ġpostulated": 23443,
+      "ĠCortex": 23444,
+      "ĠCarb": 23445,
+      "CHKERRQ": 23446,
+      "Ġoverlay": 23447,
+      "ĠFarm": 23448,
+      "enorhabditis": 23449,
+      "Ġposed": 23450,
+      "Ġinstanti": 23451,
+      "ZT": 23452,
+      "ĠVisualization": 23453,
+      "ĠGAPDH": 23454,
+      "lecom": 23455,
+      "ochron": 23456,
+      "ĠBj": 23457,
+      "ĠTrib": 23458,
+      "Ġbyte": 23459,
+      "Ġsuperimposed": 23460,
+      "Ġundi": 23461,
+      "Ġaccelerator": 23462,
+      "criptions": 23463,
+      "ĠSmooth": 23464,
+      "Ġzip": 23465,
+      "nesota": 23466,
+      "ĠEFF": 23467,
+      "ĠCole": 23468,
+      "ĠBru": 23469,
+      "rend": 23470,
+      "utz": 23471,
+      "Ġdiagnose": 23472,
+      "basis": 23473,
+      "diamond": 23474,
+      "ĠInn": 23475,
+      "ĠMedian": 23476,
+      "Ġmarginally": 23477,
+      "Ġlemmas": 23478,
+      "rectomy": 23479,
+      "Ġdialogue": 23480,
+      "ĠBrid": 23481,
+      "Ġå": 23482,
+      "oxane": 23483,
+      "aris": 23484,
+      "Ġmunicipality": 23485,
+      "Ġproducers": 23486,
+      "Regarding": 23487,
+      "ĠFV": 23488,
+      "ideal": 23489,
+      "exponential": 23490,
+      "Label": 23491,
+      "ĠFrobenius": 23492,
+      "Ġell": 23493,
+      "ĠLTE": 23494,
+      "Ġlipase": 23495,
+      "rp": 23496,
+      "Ġdm": 23497,
+      "otri": 23498,
+      "cloud": 23499,
+      "ĠAgent": 23500,
+      "MSCs": 23501,
+      "osom": 23502,
+      "hydropy": 23503,
+      "neurons": 23504,
+      "Ġsolvable": 23505,
+      "ducting": 23506,
+      "Ġrendered": 23507,
+      "Ġattractor": 23508,
+      "Ġbrac": 23509,
+      "Ãģ": 23510,
+      "Ġhosted": 23511,
+      "ĠOct": 23512,
+      "Ġguiding": 23513,
+      "Ġdigestive": 23514,
+      "js": 23515,
+      "Ġintent": 23516,
+      "flux": 23517,
+      "Ġbiosynthetic": 23518,
+      "Ġelections": 23519,
+      "ĠWilcoxon": 23520,
+      "Ġspectrophotometer": 23521,
+      "Ġimpairs": 23522,
+      "Ġabdomen": 23523,
+      "kb": 23524,
+      "ĠWho": 23525,
+      "ASSERT": 23526,
+      "Ġeluted": 23527,
+      "Ġmaximization": 23528,
+      "Ġcollector": 23529,
+      "ĠPreviously": 23530,
+      "aq": 23531,
+      "ambo": 23532,
+      "ĠOz": 23533,
+      "Cur": 23534,
+      "Ġcaffeine": 23535,
+      "Mass": 23536,
+      "pal": 23537,
+      "piece": 23538,
+      "ouville": 23539,
+      "ĠMeyer": 23540,
+      "uta": 23541,
+      "chan": 23542,
+      "ĠKS": 23543,
+      "omotor": 23544,
+      "ĠGPR": 23545,
+      "Ġeval": 23546,
+      "ĠCooperative": 23547,
+      "oglycan": 23548,
+      "Ġnozzle": 23549,
+      "ĠShel": 23550,
+      "Ġinterchange": 23551,
+      "Ġundergrad": 23552,
+      "Ġexplanatory": 23553,
+      "Ġphagocytosis": 23554,
+      "Ġctx": 23555,
+      "hess": 23556,
+      "Ġuniversality": 23557,
+      "ĠKilling": 23558,
+      "onsin": 23559,
+      "Ġlasting": 23560,
+      "ĠImm": 23561,
+      "Ġconcordance": 23562,
+      "yma": 23563,
+      "Ġautumn": 23564,
+      "Ġbarley": 23565,
+      "Ġconsequent": 23566,
+      "isi": 23567,
+      "Ġconjugates": 23568,
+      "Ġtaught": 23569,
+      "Ġcovariate": 23570,
+      "Ġadolescence": 23571,
+      "Ġvillages": 23572,
+      "Ġeigenfunctions": 23573,
+      "Ġtemporally": 23574,
+      "ĠMinnesota": 23575,
+      "yrate": 23576,
+      "iesis": 23577,
+      "definite": 23578,
+      "Ġalphabet": 23579,
+      "ĠYun": 23580,
+      "ĠMAR": 23581,
+      "Ġsealed": 23582,
+      "ronectin": 23583,
+      "ĠSepar": 23584,
+      "nx": 23585,
+      "CAA": 23586,
+      "Ġreception": 23587,
+      "ucky": 23588,
+      "ĠPTEN": 23589,
+      "ĠMorgan": 23590,
+      "Ġdiodes": 23591,
+      "Ġmetformin": 23592,
+      "Ġsynthes": 23593,
+      "ĠParticip": 23594,
+      "ĠJersey": 23595,
+      "Ġamphib": 23596,
+      "chel": 23597,
+      "Ġlamp": 23598,
+      "ĠHels": 23599,
+      "ĠFN": 23600,
+      "Ġexcav": 23601,
+      "isecond": 23602,
+      "intro": 23603,
+      "Ġnoncommutative": 23604,
+      "Ġsubsystems": 23605,
+      "summ": 23606,
+      "Ġcontrasting": 23607,
+      "ĠSilicon": 23608,
+      "ĠPartition": 23609,
+      "GlcNAc": 23610,
+      "Ġdiscern": 23611,
+      "ĠBounds": 23612,
+      "ĠRah": 23613,
+      "Ġapproximating": 23614,
+      "ĠHypert": 23615,
+      "ĠDil": 23616,
+      "Ġcompactness": 23617,
+      "Ġcaught": 23618,
+      "ĠImprove": 23619,
+      "ĠToronto": 23620,
+      "ĠBiomark": 23621,
+      "ĠBag": 23622,
+      "ĠInvent": 23623,
+      "Ġelaborate": 23624,
+      "ĠMott": 23625,
+      "ABC": 23626,
+      "ĠGraham": 23627,
+      "Ġpoultry": 23628,
+      "ĠConjecture": 23629,
+      "ĠAlgebras": 23630,
+      "ĠNLO": 23631,
+      "apsing": 23632,
+      "pathy": 23633,
+      "ĠElizabeth": 23634,
+      "ĠTit": 23635,
+      "ĠSCI": 23636,
+      "anton": 23637,
+      "Ġvoting": 23638,
+      "mathrel": 23639,
+      "ĠFord": 23640,
+      "igibility": 23641,
+      "Ġallergy": 23642,
+      "acoustic": 23643,
+      "ĠDyn": 23644,
+      "ĠDSC": 23645,
+      "ĠGRO": 23646,
+      "ĠThirty": 23647,
+      "Ġanalysing": 23648,
+      "ĠEmpire": 23649,
+      "fire": 23650,
+      "Ġpathologic": 23651,
+      "Ġpatent": 23652,
+      "Ġheard": 23653,
+      "ĠFront": 23654,
+      "isconsin": 23655,
+      "hypert": 23656,
+      "uzumab": 23657,
+      "ĠMutation": 23658,
+      "Ġbiliary": 23659,
+      "Ġsuperfluid": 23660,
+      "ĠWC": 23661,
+      "ustom": 23662,
+      "ĠActivities": 23663,
+      "Ġpolypeptide": 23664,
+      "heets": 23665,
+      "Ġborders": 23666,
+      "early": 23667,
+      "Ġorthogon": 23668,
+      "Ġbulge": 23669,
+      "ï£": 23670,
+      "Ġconical": 23671,
+      "ĠLept": 23672,
+      "Ġelectrolytes": 23673,
+      "Ġ«": 23674,
+      "regulating": 23675,
+      "Ġviolated": 23676,
+      "âĺ": 23677,
+      "ALT": 23678,
+      "ĠWorks": 23679,
+      "ĠHepat": 23680,
+      "urgical": 23681,
+      "obar": 23682,
+      "ĠReactive": 23683,
+      "possibly": 23684,
+      "ĠAdsorption": 23685,
+      "ĠRio": 23686,
+      "anoic": 23687,
+      "ĠâĨij": 23688,
+      "Ġintriguing": 23689,
+      "Ġom": 23690,
+      "hertz": 23691,
+      "ĠApproximate": 23692,
+      "ĠParent": 23693,
+      "Ġcoin": 23694,
+      "expand": 23695,
+      "в": 23696,
+      "Ġnonparametric": 23697,
+      "extern": 23698,
+      "aeus": 23699,
+      "glycerol": 23700,
+      "Ġcp": 23701,
+      "Ġbatches": 23702,
+      "Ġnanomaterials": 23703,
+      "Use": 23704,
+      "ĠVivo": 23705,
+      "Rh": 23706,
+      "Ġtiles": 23707,
+      "Ġdepict": 23708,
+      "Ġsouthwest": 23709,
+      "ĠCasimir": 23710,
+      "layered": 23711,
+      "ĠLeaf": 23712,
+      "fem": 23713,
+      "bered": 23714,
+      "Ġsubalgebra": 23715,
+      "Ġdetachment": 23716,
+      "ĠLeuk": 23717,
+      "olus": 23718,
+      "ĠRick": 23719,
+      "Ġabortion": 23720,
+      "Ġclarified": 23721,
+      "Ġganglia": 23722,
+      "QS": 23723,
+      "oising": 23724,
+      "ĠForward": 23725,
+      "ĠPeripheral": 23726,
+      "shifted": 23727,
+      "bula": 23728,
+      "ramolecular": 23729,
+      "ĠFEM": 23730,
+      "ĠProton": 23731,
+      "AME": 23732,
+      "Ġschedules": 23733,
+      "Ġaa": 23734,
+      "ĠUDP": 23735,
+      "stere": 23736,
+      "Ġmorphine": 23737,
+      "Ġspecialist": 23738,
+      "ĠAndroid": 23739,
+      "Identif": 23740,
+      "Ġunexpl": 23741,
+      "Ġheterozyg": 23742,
+      "Ġfid": 23743,
+      "pyridyl": 23744,
+      "ĠWy": 23745,
+      "phosphor": 23746,
+      "Ġfriendly": 23747,
+      "Ġmicrol": 23748,
+      "ĠSplit": 23749,
+      "agner": 23750,
+      "cribe": 23751,
+      "Ġmoth": 23752,
+      "ĠEuro": 23753,
+      "igs": 23754,
+      "ĠConditional": 23755,
+      "ĠStewart": 23756,
+      "properties": 23757,
+      "ASC": 23758,
+      "ĠTraditional": 23759,
+      "ĠPortugal": 23760,
+      "Ġearned": 23761,
+      "Ġcathe": 23762,
+      "Create": 23763,
+      "iciencies": 23764,
+      "Ġsphing": 23765,
+      "xml": 23766,
+      "Ġimmunomod": 23767,
+      "Ġcommute": 23768,
+      "Ġselenium": 23769,
+      "anges": 23770,
+      "hook": 23771,
+      "denoted": 23772,
+      "Ġjustify": 23773,
+      "ĠPool": 23774,
+      "Ġguinea": 23775,
+      "Ġcontra": 23776,
+      "Ġfolded": 23777,
+      "Ġlisting": 23778,
+      "ĠLG": 23779,
+      "ĠLane": 23780,
+      "Ġsurely": 23781,
+      "vet": 23782,
+      "fluorophenyl": 23783,
+      "Ġcorona": 23784,
+      "ĠAbund": 23785,
+      "ĠObjects": 23786,
+      "Ġtrough": 23787,
+      "cht": 23788,
+      "Ġdish": 23789,
+      "ithi": 23790,
+      "ĠMatlab": 23791,
+      "worm": 23792,
+      "Ġproteomics": 23793,
+      "Ġintermolecular": 23794,
+      "ĠPeters": 23795,
+      "Ġmirrors": 23796,
+      "quinoline": 23797,
+      "artens": 23798,
+      "ĠJewish": 23799,
+      "kB": 23800,
+      "ĠDegradation": 23801,
+      "Ġreleasing": 23802,
+      "VEGF": 23803,
+      "Ġsubpopulations": 23804,
+      "ĠTraffic": 23805,
+      "Ġproline": 23806,
+      "ĠHf": 23807,
+      "Ġadren": 23808,
+      "birth": 23809,
+      "Ġsender": 23810,
+      "Ġatlas": 23811,
+      "Ġworkplace": 23812,
+      "Ġreflectivity": 23813,
+      "ĠExistence": 23814,
+      "cls": 23815,
+      "Ġfiner": 23816,
+      "Ġbreastfeeding": 23817,
+      "onectin": 23818,
+      "Ġcogn": 23819,
+      "ellate": 23820,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 23821,
+      "byte": 23822,
+      "Ġsket": 23823,
+      "NULL": 23824,
+      "systems": 23825,
+      "ĠBranch": 23826,
+      "ĠProposed": 23827,
+      "learn": 23828,
+      "Ġtolerant": 23829,
+      "Ġvertebrates": 23830,
+      "Ġmultilevel": 23831,
+      "ĠPAH": 23832,
+      "Ġaudience": 23833,
+      "ĠWL": 23834,
+      "nitrop": 23835,
+      "ĠCt": 23836,
+      "Ġsativa": 23837,
+      "eight": 23838,
+      "Ġmeg": 23839,
+      "ocell": 23840,
+      "Ġstating": 23841,
+      "dominant": 23842,
+      "bytes": 23843,
+      "Ġpu": 23844,
+      "ĠBatter": 23845,
+      "otaxis": 23846,
+      "ĠEBV": 23847,
+      "Ġnanocrystals": 23848,
+      "Ġmonopole": 23849,
+      "Ġdiaphrag": 23850,
+      "ĠVel": 23851,
+      "Appendix": 23852,
+      "atten": 23853,
+      "impl": 23854,
+      "Ġlandmark": 23855,
+      "enclature": 23856,
+      "ĠSTAR": 23857,
+      "Ġprostagland": 23858,
+      "oprotective": 23859,
+      "Ġloadings": 23860,
+      "ĠPresence": 23861,
+      "ĠNSF": 23862,
+      "resses": 23863,
+      "FU": 23864,
+      "ilers": 23865,
+      "Ġerythrocytes": 23866,
+      "trac": 23867,
+      "islation": 23868,
+      "ĠNight": 23869,
+      "Ġsteroids": 23870,
+      "tiz": 23871,
+      "ĠDMA": 23872,
+      "Ġric": 23873,
+      "Ġsalient": 23874,
+      "ĠFur": 23875,
+      "special": 23876,
+      "Ġbioinformatics": 23877,
+      "ignant": 23878,
+      "ĠEXPERIM": 23879,
+      "avorable": 23880,
+      "disk": 23881,
+      "Ġcurriculum": 23882,
+      "imidazol": 23883,
+      "higher": 23884,
+      "Ġdesigner": 23885,
+      "ĠStrength": 23886,
+      "Ġcytosol": 23887,
+      "ĠChannels": 23888,
+      "Land": 23889,
+      "spar": 23890,
+      "Expression": 23891,
+      "Ġdaytime": 23892,
+      "mercial": 23893,
+      "vbox": 23894,
+      "inar": 23895,
+      "ieving": 23896,
+      "cein": 23897,
+      "ĠNCBI": 23898,
+      "RAN": 23899,
+      "¸Ģ": 23900,
+      "Hig": 23901,
+      "ĠDHA": 23902,
+      "Ġsubscript": 23903,
+      "Ġ¢": 23904,
+      "orange": 23905,
+      "Ġknows": 23906,
+      "ĠNAF": 23907,
+      "produced": 23908,
+      "epid": 23909,
+      "Ġdexamethasone": 23910,
+      "Ġformaldehyde": 23911,
+      "yll": 23912,
+      "Ġectopic": 23913,
+      "ĠVerification": 23914,
+      "activating": 23915,
+      "ĠIG": 23916,
+      "ĠPav": 23917,
+      "Ġtrading": 23918,
+      "Ġgraduate": 23919,
+      "ĠFIR": 23920,
+      "encil": 23921,
+      "every": 23922,
+      "Ġradiological": 23923,
+      "ĠMammalian": 23924,
+      "MES": 23925,
+      "inium": 23926,
+      "ĠSAS": 23927,
+      "ĠWH": 23928,
+      "Override": 23929,
+      "ĠScheduling": 23930,
+      "ĠBes": 23931,
+      "ĠYao": 23932,
+      "Ġglad": 23933,
+      "ĠStandards": 23934,
+      "Ġprovinces": 23935,
+      "eners": 23936,
+      "Ġnr": 23937,
+      "Ġtranspos": 23938,
+      "ĠCarib": 23939,
+      "Ġfauna": 23940,
+      "umi": 23941,
+      "reset": 23942,
+      "Ġsupra": 23943,
+      "Ġdivisions": 23944,
+      "Ġbiodegrad": 23945,
+      "metrics": 23946,
+      "ografts": 23947,
+      "Ġfunctors": 23948,
+      "Ġsupportive": 23949,
+      "Ġcaudal": 23950,
+      "Ġexerts": 23951,
+      "Ġcub": 23952,
+      "odimer": 23953,
+      "Ġairborne": 23954,
+      "Ġdelivering": 23955,
+      "Ġmultivariable": 23956,
+      "Ġfurnace": 23957,
+      "Ġremnant": 23958,
+      "Ġinco": 23959,
+      "ĠElectromagnetic": 23960,
+      "mapping": 23961,
+      "Ġdeclines": 23962,
+      "cold": 23963,
+      "ĠSeed": 23964,
+      "conversion": 23965,
+      "Ġglycogen": 23966,
+      "dT": 23967,
+      "awi": 23968,
+      "APP": 23969,
+      "Hol": 23970,
+      "atalysts": 23971,
+      "ĠSatellite": 23972,
+      "garis": 23973,
+      "card": 23974,
+      "ĠBreak": 23975,
+      "ĠAgainst": 23976,
+      "ddot": 23977,
+      "Ġpruning": 23978,
+      "ĠCaenorhabditis": 23979,
+      "Ġsucceeded": 23980,
+      "ubert": 23981,
+      "ĠÏħ": 23982,
+      "IDs": 23983,
+      "Ġasymptotics": 23984,
+      "Ġautoanti": 23985,
+      "ĠScalar": 23986,
+      "Ġnematode": 23987,
+      "hd": 23988,
+      "Ġgyn": 23989,
+      "istocene": 23990,
+      "Ġunderground": 23991,
+      "ĠEthical": 23992,
+      "Ġsial": 23993,
+      "ĠMigration": 23994,
+      "cope": 23995,
+      "Ġstigma": 23996,
+      "Ġeleven": 23997,
+      "Ġcoloring": 23998,
+      "initions": 23999,
+      "ĠJay": 24000,
+      "oba": 24001,
+      "ĠLDA": 24002,
+      "Ġbuilds": 24003,
+      "gences": 24004,
+      "ĠEcology": 24005,
+      "scheme": 24006,
+      "ĠUltras": 24007,
+      "Ġmediation": 24008,
+      "ĠTaq": 24009,
+      "Ġflying": 24010,
+      "ĠEquilibrium": 24011,
+      "ophosphate": 24012,
+      "ĠArgentina": 24013,
+      "psia": 24014,
+      "ttes": 24015,
+      "Ġdisparity": 24016,
+      "Ġadvertis": 24017,
+      "aggreg": 24018,
+      "ISA": 24019,
+      "odem": 24020,
+      "ĠRational": 24021,
+      "Ġsilent": 24022,
+      "divided": 24023,
+      "Pan": 24024,
+      "JA": 24025,
+      "claim": 24026,
+      "Ġradioactive": 24027,
+      "Ġpink": 24028,
+      "Ġconverse": 24029,
+      "ĠMell": 24030,
+      "enib": 24031,
+      "ruskal": 24032,
+      "slope": 24033,
+      "henol": 24034,
+      "ĠPon": 24035,
+      "partition": 24036,
+      "SMGR": 24037,
+      "titled": 24038,
+      "ĠInterference": 24039,
+      "tosecond": 24040,
+      "Ġseq": 24041,
+      "Ġtransitive": 24042,
+      "ĠWid": 24043,
+      "reviewed": 24044,
+      "×¥": 24045,
+      "ĠVC": 24046,
+      "recall": 24047,
+      "ogeneic": 24048,
+      "ĠOverexpression": 24049,
+      "Ġcommitted": 24050,
+      "Ġsynapse": 24051,
+      "Short": 24052,
+      "ĠNeutral": 24053,
+      "icles": 24054,
+      "ISM": 24055,
+      "Ġintrinsically": 24056,
+      "Ġmicrosatellite": 24057,
+      "RN": 24058,
+      "ĠâĪĥ": 24059,
+      "detection": 24060,
+      "Ġcodimension": 24061,
+      "Ġdrawbacks": 24062,
+      "ĠTurner": 24063,
+      "Ġsputtering": 24064,
+      "Ġdismut": 24065,
+      "Ġhypogly": 24066,
+      "Ġspeak": 24067,
+      "JD": 24068,
+      "Ġsul": 24069,
+      "Ġperinatal": 24070,
+      "Ġink": 24071,
+      "iest": 24072,
+      "Ġofficers": 24073,
+      "tick": 24074,
+      "Ġretaining": 24075,
+      "ĠNET": 24076,
+      "Ġexchanges": 24077,
+      "Ġanyone": 24078,
+      "ĠEndothelial": 24079,
+      "send": 24080,
+      "injection": 24081,
+      "ĠPeru": 24082,
+      "Ġclades": 24083,
+      "uctuations": 24084,
+      "Ġsulphate": 24085,
+      "pio": 24086,
+      "Ġphysi": 24087,
+      "ĠMiy": 24088,
+      "ĠBAS": 24089,
+      "arius": 24090,
+      "Ġlipopolysaccharide": 24091,
+      "Ġneurodegeneration": 24092,
+      "ĠTurkish": 24093,
+      "Ġophthal": 24094,
+      "Ġacted": 24095,
+      "entre": 24096,
+      "Ġshaking": 24097,
+      "Ġchloroplast": 24098,
+      "ĠSid": 24099,
+      "regnancy": 24100,
+      "asion": 24101,
+      "ĠHs": 24102,
+      "Ġinitiating": 24103,
+      "Ġflexural": 24104,
+      "Ϫ": 24105,
+      "Ġparac": 24106,
+      "Ġinterlayer": 24107,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 24108,
+      "cause": 24109,
+      "ractions": 24110,
+      "Ġvaluation": 24111,
+      "SYSMGR": 24112,
+      "ĠGarcia": 24113,
+      "arrays": 24114,
+      "Ġcasting": 24115,
+      "ĠPFN": 24116,
+      "ĠLanc": 24117,
+      "ĠGlob": 24118,
+      "Ġdenti": 24119,
+      "Ġportfolio": 24120,
+      "ĠHolocene": 24121,
+      "ĠMATERIAL": 24122,
+      "Ġsarc": 24123,
+      "Lear": 24124,
+      "Ġtin": 24125,
+      "ĠClear": 24126,
+      "below": 24127,
+      "Ġadvection": 24128,
+      "Ġoverlaps": 24129,
+      "Ġarthroplasty": 24130,
+      "compute": 24131,
+      "Ġglycolysis": 24132,
+      "hept": 24133,
+      "lora": 24134,
+      "frames": 24135,
+      "ĠHern": 24136,
+      "proto": 24137,
+      "Ġswine": 24138,
+      "Ġjejun": 24139,
+      "Ġrepeating": 24140,
+      "ancreatic": 24141,
+      "ĠCollins": 24142,
+      "ĠPrinciple": 24143,
+      "Ġnanof": 24144,
+      "Ġadjacency": 24145,
+      "Ġsynov": 24146,
+      "chet": 24147,
+      "ĠAlmost": 24148,
+      "Ġintrusion": 24149,
+      "Ġechocardiography": 24150,
+      "liferation": 24151,
+      "Ġquiescent": 24152,
+      "ĠMuk": 24153,
+      "Ġlifetimes": 24154,
+      "graded": 24155,
+      "Ġoverwhel": 24156,
+      "zel": 24157,
+      "Ġnitride": 24158,
+      "Ġdisturbed": 24159,
+      "Ġfastest": 24160,
+      "grability": 24161,
+      "Ġtolerated": 24162,
+      "frag": 24163,
+      "ĠExtension": 24164,
+      "anoate": 24165,
+      "iferous": 24166,
+      "Ġhydrodynamics": 24167,
+      "IONAL": 24168,
+      "ĠToday": 24169,
+      "ĠExpansion": 24170,
+      "Ġvenom": 24171,
+      "ĠHepatitis": 24172,
+      "ño": 24173,
+      "onation": 24174,
+      "synuclein": 24175,
+      "Ġbasketball": 24176,
+      "clusions": 24177,
+      "Ġsettled": 24178,
+      "IQR": 24179,
+      "ĠCra": 24180,
+      "Ġautomation": 24181,
+      "ĠHealthy": 24182,
+      "ĠPortuguese": 24183,
+      "ĠAbelian": 24184,
+      "Ġgad": 24185,
+      "ĠHG": 24186,
+      "ĠRoth": 24187,
+      "Ġconsume": 24188,
+      "FG": 24189,
+      "inals": 24190,
+      "ĠMCMC": 24191,
+      "Ġpregnancies": 24192,
+      "DES": 24193,
+      "portional": 24194,
+      "ĠBiochemical": 24195,
+      "Ġmissions": 24196,
+      "ĠAntibody": 24197,
+      "ĠBCG": 24198,
+      "ĠLAS": 24199,
+      "marine": 24200,
+      "DMA": 24201,
+      "Ġlongevity": 24202,
+      "ĠDry": 24203,
+      "ĠRao": 24204,
+      "Ġinterferometer": 24205,
+      "Ġdiscretized": 24206,
+      "osensory": 24207,
+      "sit": 24208,
+      "etta": 24209,
+      "tainer": 24210,
+      "otherwise": 24211,
+      "AKT": 24212,
+      "ĠFaculty": 24213,
+      "Ġascertain": 24214,
+      "ĠSimulated": 24215,
+      "Ġpayload": 24216,
+      "OUT": 24217,
+      "Ġsuffers": 24218,
+      "Ġtungsten": 24219,
+      "ĠAnxiety": 24220,
+      "ĠHeterogeneous": 24221,
+      "lingual": 24222,
+      "Ġpherom": 24223,
+      "bors": 24224,
+      "linux": 24225,
+      "Ġmonkey": 24226,
+      "£": 24227,
+      "url": 24228,
+      "ĠAcross": 24229,
+      "ĠAKI": 24230,
+      "Ġopp": 24231,
+      "ocalization": 24232,
+      "Ġmorphogenesis": 24233,
+      "gic": 24234,
+      "ĠPCM": 24235,
+      "Ġoligomers": 24236,
+      "Ġexhaustive": 24237,
+      "ĠGIS": 24238,
+      "Ġpristine": 24239,
+      "ĠActiv": 24240,
+      "ĠScilab": 24241,
+      "ĠAcoustic": 24242,
+      "ĠPick": 24243,
+      "integral": 24244,
+      "Ġphilosophy": 24245,
+      "ĠDeng": 24246,
+      "ĠHab": 24247,
+      "scape": 24248,
+      "ĠEmergency": 24249,
+      "Ġepi": 24250,
+      "ĠBET": 24251,
+      "ricket": 24252,
+      "Ġannulus": 24253,
+      "Ġlysosomal": 24254,
+      "Ġstrands": 24255,
+      "CAP": 24256,
+      "ĠAmino": 24257,
+      "ĠStri": 24258,
+      "ependence": 24259,
+      "Ġfootprint": 24260,
+      "ĠFatty": 24261,
+      "ĠNaz": 24262,
+      "nest": 24263,
+      "ĠExplicit": 24264,
+      "planetary": 24265,
+      "lead": 24266,
+      "Ġgrip": 24267,
+      "need": 24268,
+      "ATT": 24269,
+      "ERV": 24270,
+      "ĠTargeted": 24271,
+      "CRP": 24272,
+      "Ġparamagnetic": 24273,
+      "ĠTyr": 24274,
+      "ĠMicroRNA": 24275,
+      "hline": 24276,
+      "gh": 24277,
+      "pit": 24278,
+      "ĠIsolated": 24279,
+      "jectory": 24280,
+      "Ġcleaned": 24281,
+      "oste": 24282,
+      "Ġpathologies": 24283,
+      "propylene": 24284,
+      "ĠReason": 24285,
+      "ĠINFO": 24286,
+      "RAY": 24287,
+      "Values": 24288,
+      "Ġalive": 24289,
+      "Ġbiof": 24290,
+      "ewicz": 24291,
+      "Ġcracking": 24292,
+      "google": 24293,
+      "locked": 24294,
+      "crop": 24295,
+      "eca": 24296,
+      "urane": 24297,
+      "SVM": 24298,
+      "utta": 24299,
+      "ĠMetric": 24300,
+      "ĠEncycl": 24301,
+      "ĠModule": 24302,
+      "Ġwarranted": 24303,
+      "Ġmultidisciplinary": 24304,
+      "ĠElastic": 24305,
+      "labelled": 24306,
+      "ĠSchwarzschild": 24307,
+      "ĠPCC": 24308,
+      "major": 24309,
+      "video": 24310,
+      "Ġstoring": 24311,
+      "ĠMake": 24312,
+      "ako": 24313,
+      "ĠJia": 24314,
+      "Ġtoroidal": 24315,
+      "ĠHMM": 24316,
+      "Ġmasking": 24317,
+      "Again": 24318,
+      "Ġnephropathy": 24319,
+      "gf": 24320,
+      "Ġdominating": 24321,
+      "erkin": 24322,
+      "ĠFabrication": 24323,
+      "ĠFel": 24324,
+      "DEF": 24325,
+      "culture": 24326,
+      "ĠIra": 24327,
+      "ĠREG": 24328,
+      "ilingual": 24329,
+      "Ġmuss": 24330,
+      "plain": 24331,
+      "zh": 24332,
+      "iston": 24333,
+      "ĠÎ¥": 24334,
+      "minimal": 24335,
+      "cmp": 24336,
+      "GaN": 24337,
+      "Ġmonotonic": 24338,
+      "Ġinvolution": 24339,
+      "Ġwhatever": 24340,
+      "ĠInstrument": 24341,
+      "imple": 24342,
+      "ĠPCI": 24343,
+      "ĠNeuronal": 24344,
+      "Ġfacets": 24345,
+      "Ġhemodialysis": 24346,
+      "apatite": 24347,
+      "ĠKil": 24348,
+      "ontally": 24349,
+      "Ġinserting": 24350,
+      "ĠRIP": 24351,
+      "Ġconnective": 24352,
+      "ĠFederation": 24353,
+      "nut": 24354,
+      "ĠGun": 24355,
+      "inuous": 24356,
+      "Mor": 24357,
+      "ĠWisconsin": 24358,
+      "Ġmush": 24359,
+      "ITS": 24360,
+      "Ġeject": 24361,
+      "ĠBPS": 24362,
+      "ĠHorn": 24363,
+      "ĠEmbedding": 24364,
+      "Ġraces": 24365,
+      "ĠJam": 24366,
+      "Ġposture": 24367,
+      "ĠInvol": 24368,
+      "ĠIMDb": 24369,
+      "ĠPlease": 24370,
+      "proportion": 24371,
+      "ĠInterleukin": 24372,
+      "Ġarte": 24373,
+      "Ġsubsp": 24374,
+      "oderma": 24375,
+      "Find": 24376,
+      "imit": 24377,
+      "ĠClin": 24378,
+      "Hel": 24379,
+      "FILE": 24380,
+      "original": 24381,
+      "ervoir": 24382,
+      "Ġpleural": 24383,
+      "clipse": 24384,
+      "encer": 24385,
+      "inaries": 24386,
+      "Ġvictory": 24387,
+      "Ġinvestigates": 24388,
+      "ĠImportance": 24389,
+      "ĠMIN": 24390,
+      "Ġphonons": 24391,
+      "integrated": 24392,
+      "Ġexchanged": 24393,
+      "ystis": 24394,
+      "Ġmigrate": 24395,
+      "Rob": 24396,
+      "eland": 24397,
+      "proof": 24398,
+      "ĠIntegral": 24399,
+      "Ġmergers": 24400,
+      "Ġpolyphenols": 24401,
+      "ĠFully": 24402,
+      "Ġuro": 24403,
+      "Ġhomogenous": 24404,
+      "Ġrecognizing": 24405,
+      "ĠSignals": 24406,
+      "vat": 24407,
+      "igms": 24408,
+      "Ġaccuracies": 24409,
+      "Substituting": 24410,
+      "Ġpoisoning": 24411,
+      "Ġshrimp": 24412,
+      "ĠHölder": 24413,
+      "ĠTanzania": 24414,
+      "JS": 24415,
+      "MENT": 24416,
+      "ĠTopology": 24417,
+      "Ġinvers": 24418,
+      "ĠDU": 24419,
+      "Ġuniaxial": 24420,
+      "ĠSEC": 24421,
+      "party": 24422,
+      "Ġcontrollable": 24423,
+      "Ġfum": 24424,
+      "ostics": 24425,
+      "Ġmanifested": 24426,
+      "Ġpropagated": 24427,
+      "Ġsuffix": 24428,
+      "ĠCAN": 24429,
+      "ĠPret": 24430,
+      "keeping": 24431,
+      "Assuming": 24432,
+      "Ġsuture": 24433,
+      "Ġpest": 24434,
+      "Ġgamet": 24435,
+      "ĠAlignment": 24436,
+      "esarean": 24437,
+      "tum": 24438,
+      "Ġrefine": 24439,
+      "Ġpopulated": 24440,
+      "Ġestu": 24441,
+      "ĠDefense": 24442,
+      "ĠPrivacy": 24443,
+      "ĠWein": 24444,
+      "ĠSenate": 24445,
+      "Ġazimuth": 24446,
+      "ĠProfessional": 24447,
+      "Ġlabour": 24448,
+      "Ġseminal": 24449,
+      "ĠIntervention": 24450,
+      "ĠOlder": 24451,
+      "AU": 24452,
+      "Wind": 24453,
+      "dynamical": 24454,
+      "ĠVeter": 24455,
+      "ación": 24456,
+      "Ġcooking": 24457,
+      "Ġâīª": 24458,
+      "Ġbead": 24459,
+      "Ġdensely": 24460,
+      "Ġpalliative": 24461,
+      "mort": 24462,
+      "ĠAAV": 24463,
+      "ĠRyan": 24464,
+      "Prim": 24465,
+      "galax": 24466,
+      "muir": 24467,
+      "sters": 24468,
+      "ĠSalt": 24469,
+      "queeze": 24470,
+      "ĠPlateau": 24471,
+      "Ġí": 24472,
+      "Ġlighter": 24473,
+      "ordinary": 24474,
+      "formaldehyde": 24475,
+      "ĠWer": 24476,
+      "Ġbark": 24477,
+      "Ġhomogenized": 24478,
+      "Ġpyramidal": 24479,
+      "Ġinert": 24480,
+      "ĠAPC": 24481,
+      "ĠMicros": 24482,
+      "ĠProteobacteria": 24483,
+      "ĠPurification": 24484,
+      "Ġparametrized": 24485,
+      "Ġille": 24486,
+      "accuracy": 24487,
+      "embedding": 24488,
+      "Ġtoughness": 24489,
+      "Ġisometry": 24490,
+      "backs": 24491,
+      "ĠFIG": 24492,
+      "ĠRon": 24493,
+      "ĠESP": 24494,
+      "Ġmicroglial": 24495,
+      "interp": 24496,
+      "ĠIntegrating": 24497,
+      "ĠReducing": 24498,
+      "Ġhearts": 24499,
+      "Ġseriously": 24500,
+      "Ġspecially": 24501,
+      "CTRL": 24502,
+      "ĠSurprisingly": 24503,
+      "Ġhyperplane": 24504,
+      "polynomial": 24505,
+      "Ġreconc": 24506,
+      "Ġpharmacokinetic": 24507,
+      "Mart": 24508,
+      "ĠBright": 24509,
+      "mable": 24510,
+      "Ġionizing": 24511,
+      "Ġtrich": 24512,
+      "zymatic": 24513,
+      "Ġleptons": 24514,
+      "etting": 24515,
+      "ĠHex": 24516,
+      "Ġneurop": 24517,
+      "Ġadipocytes": 24518,
+      "Ġrods": 24519,
+      "Ġsupercritical": 24520,
+      "Ġsuccin": 24521,
+      "Ġanter": 24522,
+      "ĠNAC": 24523,
+      "ĠSubsequent": 24524,
+      "IGH": 24525,
+      "Ġsoutheast": 24526,
+      "Ġendowed": 24527,
+      "Ġconverging": 24528,
+      "Ġspatio": 24529,
+      "Ġcelebr": 24530,
+      "helix": 24531,
+      "Ġaccessions": 24532,
+      "Ġimmobilization": 24533,
+      "ĠEQ": 24534,
+      "spatial": 24535,
+      "Ġinformal": 24536,
+      "Ġdere": 24537,
+      "ĠEnzyme": 24538,
+      "ĠBBC": 24539,
+      "ĠEPR": 24540,
+      "Ġelectrically": 24541,
+      "Ġleukocytes": 24542,
+      "Ġalanine": 24543,
+      "Ġmitogen": 24544,
+      "Ġintramolecular": 24545,
+      "ĠNI": 24546,
+      "Ġprokary": 24547,
+      "ISO": 24548,
+      "Ġdodec": 24549,
+      "ĠTrade": 24550,
+      "ĠDai": 24551,
+      "ccc": 24552,
+      "ĠWalter": 24553,
+      "ĠNeither": 24554,
+      "Ġvulgaris": 24555,
+      "Ġlongitude": 24556,
+      "ĠIntro": 24557,
+      "option": 24558,
+      "ĠQC": 24559,
+      "ĠâĿ": 24560,
+      "protection": 24561,
+      "ĠIMF": 24562,
+      "aprote": 24563,
+      "Ġlinker": 24564,
+      "Ġfounder": 24565,
+      "Ġaspiration": 24566,
+      "clusters": 24567,
+      "ĠPay": 24568,
+      "ĠRoles": 24569,
+      "Ġacyclic": 24570,
+      "overing": 24571,
+      "Ġremind": 24572,
+      "ĠTong": 24573,
+      "ĠAtten": 24574,
+      "Ġengineers": 24575,
+      "Ġdysregulation": 24576,
+      "ĠFourth": 24577,
+      "Ġfilename": 24578,
+      "ĠCool": 24579,
+      "protected": 24580,
+      "Ġnilpotent": 24581,
+      "ĠHK": 24582,
+      "clone": 24583,
+      "ĠStadium": 24584,
+      "ais": 24585,
+      "osamine": 24586,
+      "ABILITY": 24587,
+      "rovascular": 24588,
+      "ĠAH": 24589,
+      "ĠConcept": 24590,
+      "Ġcerebrospinal": 24591,
+      "owitz": 24592,
+      "Ġresolving": 24593,
+      "Ġwings": 24594,
+      "ĠEGF": 24595,
+      "ĠCommand": 24596,
+      "iazep": 24597,
+      "Ġbeef": 24598,
+      "Ġspines": 24599,
+      "Ġpriorities": 24600,
+      "Ġattempting": 24601,
+      "Ġtelomere": 24602,
+      "BQU": 24603,
+      "Ġviolations": 24604,
+      "LB": 24605,
+      "omnia": 24606,
+      "osm": 24607,
+      "irq": 24608,
+      "Ġdiversification": 24609,
+      "alt": 24610,
+      "ĠBRAF": 24611,
+      "Ġorganisation": 24612,
+      "die": 24613,
+      "Ġautoreg": 24614,
+      "icked": 24615,
+      "ĠEcological": 24616,
+      "ĠTrain": 24617,
+      "ĠPY": 24618,
+      "Ġmusculoskeletal": 24619,
+      "Ġhorizons": 24620,
+      "Ġomega": 24621,
+      "Ġquasars": 24622,
+      "eption": 24623,
+      "Ġerad": 24624,
+      "Ġluminal": 24625,
+      "Interestingly": 24626,
+      "Ġpayment": 24627,
+      "cnt": 24628,
+      "Ġdipl": 24629,
+      "Ġrecognised": 24630,
+      "Cat": 24631,
+      "ĠChl": 24632,
+      "Ġmillions": 24633,
+      "Ġdisappearance": 24634,
+      "GAP": 24635,
+      "Ġradiographic": 24636,
+      "Ġpostpartum": 24637,
+      "developed": 24638,
+      "xual": 24639,
+      "Ġhed": 24640,
+      "idered": 24641,
+      "ĠCertain": 24642,
+      "Ġdysplasia": 24643,
+      "________": 24644,
+      "ĠHalf": 24645,
+      "Ġasymmetries": 24646,
+      "ĠAlcohol": 24647,
+      "Sum": 24648,
+      "Ġfm": 24649,
+      "Ġchap": 24650,
+      "Ġpretreated": 24651,
+      "ĠGallery": 24652,
+      "Ġoutperform": 24653,
+      "Ġbreeds": 24654,
+      "Ġtied": 24655,
+      "Ġdiffeomorphism": 24656,
+      "Ġcausative": 24657,
+      "Ġcollectively": 24658,
+      "Ġsuboptimal": 24659,
+      "Ġinsulation": 24660,
+      "Ġmanipulate": 24661,
+      "Ġkilomet": 24662,
+      "Ġrepulsion": 24663,
+      "Ġchloroform": 24664,
+      "Ġbean": 24665,
+      "Ġhero": 24666,
+      "rophysics": 24667,
+      "ĠPeptide": 24668,
+      "Ġoutlier": 24669,
+      "Derived": 24670,
+      "isser": 24671,
+      "ĠInfant": 24672,
+      "sulfonyl": 24673,
+      "Ġrecursively": 24674,
+      "Hu": 24675,
+      "ĠKoh": 24676,
+      "pyridine": 24677,
+      "Ġsquad": 24678,
+      "Ġthirty": 24679,
+      "Ġspoken": 24680,
+      "ĠZar": 24681,
+      "othermic": 24682,
+      "Ġcalcification": 24683,
+      "ĠHelsinki": 24684,
+      "Ġbeach": 24685,
+      "ĠFDR": 24686,
+      "Ġprobiotic": 24687,
+      "Ġfinishing": 24688,
+      "ymmetrical": 24689,
+      "Ġvacancy": 24690,
+      "Ġthrombo": 24691,
+      "Compared": 24692,
+      "AST": 24693,
+      "sted": 24694,
+      "otherap": 24695,
+      "Ġiodide": 24696,
+      "Ġtt": 24697,
+      "alignment": 24698,
+      "Ġmicrovascular": 24699,
+      "Ġinitialize": 24700,
+      "ĠANALYSIS": 24701,
+      "Ġtopographic": 24702,
+      "ĠReporting": 24703,
+      "Ġunderestimated": 24704,
+      "puted": 24705,
+      "Ġatherosclerotic": 24706,
+      "Qiagen": 24707,
+      "gut": 24708,
+      "ĠCortical": 24709,
+      "Ġdisrupt": 24710,
+      "este": 24711,
+      "Ġglue": 24712,
+      "Ġnarrower": 24713,
+      "Ġinpatient": 24714,
+      "Ġscholars": 24715,
+      "Ġbc": 24716,
+      "ĠPsychological": 24717,
+      "ĠHamiltonians": 24718,
+      "Ġhonor": 24719,
+      "tibular": 24720,
+      "Ġinsertions": 24721,
+      "oscope": 24722,
+      "Ġpharmacokinetics": 24723,
+      "Ġmathematically": 24724,
+      "Ġfork": 24725,
+      "ipital": 24726,
+      "ĠArgs": 24727,
+      "abolism": 24728,
+      "Ġâİł": 24729,
+      "ĠRobot": 24730,
+      "ĠCasc": 24731,
+      "Ġleaching": 24732,
+      "ĠLack": 24733,
+      "Ġendocytosis": 24734,
+      "Ġtris": 24735,
+      "Ġsensitivities": 24736,
+      "Ġlicensed": 24737,
+      "Ġsponge": 24738,
+      "carbonyl": 24739,
+      "feat": 24740,
+      "Ġprecl": 24741,
+      "Ġwaist": 24742,
+      "tifications": 24743,
+      "Ġoliv": 24744,
+      "binary": 24745,
+      "atri": 24746,
+      "ĠBiot": 24747,
+      "TZ": 24748,
+      "Ġfake": 24749,
+      "ĠMosc": 24750,
+      "ĠHPS": 24751,
+      "ĠVoltage": 24752,
+      "ĠâİĿ": 24753,
+      "ĠAhmed": 24754,
+      "ĠSexual": 24755,
+      "dehydes": 24756,
+      "ĠCot": 24757,
+      "Ġmagma": 24758,
+      "oxylin": 24759,
+      "ÐĪ": 24760,
+      "amethyl": 24761,
+      "ĠLOS": 24762,
+      "diphenyl": 24763,
+      "experimental": 24764,
+      "Ġpluripotent": 24765,
+      "agittal": 24766,
+      "walk": 24767,
+      "Ġplasmonic": 24768,
+      "Ġcontracts": 24769,
+      "Ġexped": 24770,
+      "ĠArabia": 24771,
+      "Ġshoots": 24772,
+      "ĠRAN": 24773,
+      "ustrated": 24774,
+      "Ġconvexity": 24775,
+      "ĠmJ": 24776,
+      "ĠAbsolute": 24777,
+      "ĠSEL": 24778,
+      "MIP": 24779,
+      "ĠActually": 24780,
+      "sole": 24781,
+      "QI": 24782,
+      "ĠTGFβ": 24783,
+      "Ġâİŀ": 24784,
+      "Ġrearrangements": 24785,
+      "Ġcuring": 24786,
+      "expensive": 24787,
+      "ceptibility": 24788,
+      "Ġours": 24789,
+      "ĠKidney": 24790,
+      "Ġassigns": 24791,
+      "Ġvoxels": 24792,
+      "oreal": 24793,
+      "Ġevening": 24794,
+      "hus": 24795,
+      "ĠãĢ": 24796,
+      "oradi": 24797,
+      "ĠCorrection": 24798,
+      "Ġnanofibers": 24799,
+      "Ġcantile": 24800,
+      "bigoplus": 24801,
+      "uminous": 24802,
+      "eclampsia": 24803,
+      "ĠCult": 24804,
+      "ECH": 24805,
+      "atology": 24806,
+      "Ġji": 24807,
+      "cryp": 24808,
+      "ĠAspects": 24809,
+      "eni": 24810,
+      "Ġsemis": 24811,
+      "IRS": 24812,
+      "ĠPho": 24813,
+      "encoding": 24814,
+      "ĠJustice": 24815,
+      "ococci": 24816,
+      "Ġhypothalamic": 24817,
+      "ractable": 24818,
+      "ĠOrb": 24819,
+      "Simons": 24820,
+      "Ġmanipulated": 24821,
+      "attribute": 24822,
+      "onov": 24823,
+      "orously": 24824,
+      "endar": 24825,
+      "uder": 24826,
+      "insert": 24827,
+      "Ġlysed": 24828,
+      "ĠHodge": 24829,
+      "Ġfootballer": 24830,
+      "Device": 24831,
+      "ĠLeast": 24832,
+      "Ġstratum": 24833,
+      "Ġmitral": 24834,
+      "Ġsell": 24835,
+      "ĠMuc": 24836,
+      "glycer": 24837,
+      "oj": 24838,
+      "Ġpathogenicity": 24839,
+      "ĠDeclaration": 24840,
+      "opause": 24841,
+      "ĠArticle": 24842,
+      "Ġrinsed": 24843,
+      "ĠLévy": 24844,
+      "rement": 24845,
+      "Ġants": 24846,
+      "ĠDic": 24847,
+      "ĠkPa": 24848,
+      "urry": 24849,
+      "motion": 24850,
+      "client": 24851,
+      "Ġaccessory": 24852,
+      "Ġdepolarization": 24853,
+      "namely": 24854,
+      "Ġdisparities": 24855,
+      "Ġfavourable": 24856,
+      "ĠTibet": 24857,
+      "Ġoocyte": 24858,
+      "istration": 24859,
+      "Ġunresolved": 24860,
+      "criptive": 24861,
+      "physics": 24862,
+      "Ġbenzo": 24863,
+      "Ġcrystallinity": 24864,
+      "Ġpayoff": 24865,
+      "Ġumbilical": 24866,
+      "osil": 24867,
+      "ĠSystemic": 24868,
+      "ĠSTM": 24869,
+      "Ġstabilizer": 24870,
+      "USA": 24871,
+      "ĠJensen": 24872,
+      "Aug": 24873,
+      "ĠHat": 24874,
+      "AGG": 24875,
+      "underbrace": 24876,
+      "Ġmanipulations": 24877,
+      "ĠManc": 24878,
+      "nedy": 24879,
+      "Ġscratch": 24880,
+      "Cherry": 24881,
+      "osaccharides": 24882,
+      "Ġprecipitate": 24883,
+      "quarters": 24884,
+      "icul": 24885,
+      "Ġoptimally": 24886,
+      "many": 24887,
+      "Ġneoplasms": 24888,
+      "Ġinward": 24889,
+      "aryng": 24890,
+      "Ġmoll": 24891,
+      "ĠWel": 24892,
+      "ĠWiley": 24893,
+      "Ġnewspaper": 24894,
+      "Ġinhabitants": 24895,
+      "ĠSuccess": 24896,
+      "Ġbridging": 24897,
+      "Ġdisconnected": 24898,
+      "Ġhygiene": 24899,
+      "Dist": 24900,
+      "Ġscripts": 24901,
+      "Ġmesoporous": 24902,
+      "Ġrestricts": 24903,
+      "actone": 24904,
+      "Ġaquifer": 24905,
+      "ĠïĤ·": 24906,
+      "Ġplex": 24907,
+      "Ġpresumed": 24908,
+      "Ġips": 24909,
+      "ĠMilitary": 24910,
+      "Ġjudged": 24911,
+      "Ġald": 24912,
+      "Ġsequest": 24913,
+      "compared": 24914,
+      "ULATION": 24915,
+      "adapted": 24916,
+      "Ġinstructed": 24917,
+      "pulse": 24918,
+      "Ġcusp": 24919,
+      "matching": 24920,
+      "carrier": 24921,
+      "Ġenforce": 24922,
+      "ĠInterview": 24923,
+      "ometrics": 24924,
+      "Ġnullptr": 24925,
+      "Ġflavour": 24926,
+      "ĠPareto": 24927,
+      "ĠBER": 24928,
+      "Ġuv": 24929,
+      "Ġcrash": 24930,
+      "ĠCann": 24931,
+      "ĠMineral": 24932,
+      "ĠOlympic": 24933,
+      "Ġpolycrystalline": 24934,
+      "lett": 24935,
+      "Tables": 24936,
+      "requent": 24937,
+      "Ġsedentary": 24938,
+      "unsaturated": 24939,
+      "ĠBernoulli": 24940,
+      "Ġadmissions": 24941,
+      "itorial": 24942,
+      "acute": 24943,
+      "Ġadditions": 24944,
+      "weet": 24945,
+      "ALE": 24946,
+      "ĠManip": 24947,
+      "tokens": 24948,
+      "preced": 24949,
+      "dk": 24950,
+      "consider": 24951,
+      "Ġïĺ¹": 24952,
+      "Ġwrites": 24953,
+      "cardia": 24954,
+      "ctomy": 24955,
+      "omatous": 24956,
+      "Symbol": 24957,
+      "usten": 24958,
+      "Ġproteolytic": 24959,
+      "categories": 24960,
+      "Ġfic": 24961,
+      "Ġswing": 24962,
+      "Ġpassenger": 24963,
+      "Ġoverlapped": 24964,
+      "ifi": 24965,
+      "Ġmutational": 24966,
+      "ĠJosephson": 24967,
+      "Ġregret": 24968,
+      "ĠArk": 24969,
+      "ĠCFD": 24970,
+      "Ġmaneu": 24971,
+      "encoded": 24972,
+      "textsc": 24973,
+      "Ġdecompositions": 24974,
+      "ĠDeb": 24975,
+      "Ġmandibular": 24976,
+      "dU": 24977,
+      "ĠPIC": 24978,
+      "Ġtranscriptomic": 24979,
+      "Ġtelescop": 24980,
+      "ĠSantos": 24981,
+      "oE": 24982,
+      "ĠMCP": 24983,
+      "Ġindigenous": 24984,
+      "Ġmicrospheres": 24985,
+      "Ġcodew": 24986,
+      "zip": 24987,
+      "Ġfingers": 24988,
+      "Ġcampaigns": 24989,
+      "¸Ģł": 24990,
+      "Ġaccidents": 24991,
+      "ĠTools": 24992,
+      "Planck": 24993,
+      "»": 24994,
+      "eder": 24995,
+      "ingham": 24996,
+      "oxidase": 24997,
+      "Ġancestor": 24998,
+      "whose": 24999,
+      "Ġphospholipid": 25000,
+      "Ġconversation": 25001,
+      "ĠHof": 25002,
+      "cortical": 25003,
+      "glycos": 25004,
+      "Ġmanufacturers": 25005,
+      "opulmonary": 25006,
+      "Ġinclined": 25007,
+      "ĠBethe": 25008,
+      "Ġspending": 25009,
+      "ĠFusarium": 25010,
+      "uitively": 25011,
+      "Ġfemur": 25012,
+      "ĠLinks": 25013,
+      "Ġnitrite": 25014,
+      "Main": 25015,
+      "Ġflora": 25016,
+      "ĠPhD": 25017,
+      "ĠWriting": 25018,
+      "ĠHessian": 25019,
+      "Ġμs": 25020,
+      "ools": 25021,
+      "Ġvictims": 25022,
+      "ĠRew": 25023,
+      "ansen": 25024,
+      "Ear": 25025,
+      "Ġorn": 25026,
+      "Ġthermoelectric": 25027,
+      "ENSE": 25028,
+      "ĠWeighted": 25029,
+      "holes": 25030,
+      "Ġcen": 25031,
+      "Ġacuity": 25032,
+      "Ġvacancies": 25033,
+      "ĠDuke": 25034,
+      "Ġpaclitaxel": 25035,
+      "Ġconverts": 25036,
+      "bourne": 25037,
+      "ĠACS": 25038,
+      "osi": 25039,
+      "Ġcriminal": 25040,
+      "ĠIb": 25041,
+      "unes": 25042,
+      "ĠNanoc": 25043,
+      "Post": 25044,
+      "ĠMDS": 25045,
+      "Ġeconomics": 25046,
+      "Ġthoughts": 25047,
+      "Ġneuroprotective": 25048,
+      "Ġintersects": 25049,
+      "cers": 25050,
+      "atid": 25051,
+      "usa": 25052,
+      "ĠAns": 25053,
+      "Ġafterwards": 25054,
+      "ĠOFDM": 25055,
+      "ĠCMV": 25056,
+      "ĠCum": 25057,
+      "ATG": 25058,
+      "ĠImageNet": 25059,
+      "ĠAttack": 25060,
+      "ogeneities": 25061,
+      "Ġcounseling": 25062,
+      "ĠCONTR": 25063,
+      "ález": 25064,
+      "ĠDh": 25065,
+      "ĠGV": 25066,
+      "Ġpositional": 25067,
+      "Ġgang": 25068,
+      "ĠInteractive": 25069,
+      "wig": 25070,
+      "ĠTrace": 25071,
+      "ĠDSS": 25072,
+      "Ġsynthetase": 25073,
+      "ĠGalile": 25074,
+      "usually": 25075,
+      "ĠBass": 25076,
+      "ardless": 25077,
+      "Ġexecuting": 25078,
+      "KP": 25079,
+      "ĠNepal": 25080,
+      "READ": 25081,
+      "ĠLock": 25082,
+      "ohydro": 25083,
+      "rotation": 25084,
+      "dil": 25085,
+      "roscopically": 25086,
+      "reperfusion": 25087,
+      "Ġdishes": 25088,
+      "ĠProceedings": 25089,
+      "ĠNPC": 25090,
+      "Ġmonsoon": 25091,
+      "ĠLemmas": 25092,
+      "ĠChandra": 25093,
+      "Ġreactors": 25094,
+      "Ġtryptophan": 25095,
+      "ĠVT": 25096,
+      "ĠDEM": 25097,
+      "Ġlegislation": 25098,
+      "mk": 25099,
+      "Ġtoric": 25100,
+      "ĠPrograms": 25101,
+      "ĠPubMed": 25102,
+      "ĠrDNA": 25103,
+      "Ġposts": 25104,
+      "ĠâİĽ": 25105,
+      "Ġshedding": 25106,
+      "tolerant": 25107,
+      "Ġvoids": 25108,
+      "ĠCaribbean": 25109,
+      "CODE": 25110,
+      "Tube": 25111,
+      "ALSE": 25112,
+      "Ġchlorine": 25113,
+      "Ġcoerc": 25114,
+      "ĠRhiz": 25115,
+      "ĠKirk": 25116,
+      "ĠÃĸ": 25117,
+      "rout": 25118,
+      "icides": 25119,
+      "agu": 25120,
+      "ĠKw": 25121,
+      "Ġcru": 25122,
+      "Observe": 25123,
+      "ĠRevis": 25124,
+      "Ġanonym": 25125,
+      "Ġprerequ": 25126,
+      "ocortical": 25127,
+      "Ġrestaur": 25128,
+      "ĠPopulations": 25129,
+      "dst": 25130,
+      "Ġfort": 25131,
+      "regs": 25132,
+      "ĠPolarization": 25133,
+      "Ġpancreatitis": 25134,
+      "aph": 25135,
+      "threat": 25136,
+      "ften": 25137,
+      "ĠAlaska": 25138,
+      "ĠFlexible": 25139,
+      "Ġrepertoire": 25140,
+      "kan": 25141,
+      "mathchoice": 25142,
+      "Ġmitosis": 25143,
+      "Ġeat": 25144,
+      "utin": 25145,
+      "Ġrt": 25146,
+      "Ġdummy": 25147,
+      "ĠCys": 25148,
+      "ĠGor": 25149,
+      "earchers": 25150,
+      "HPLC": 25151,
+      "Ġbay": 25152,
+      "ĠNielsen": 25153,
+      "ĠRoc": 25154,
+      "iani": 25155,
+      "icit": 25156,
+      "rague": 25157,
+      "Ġcourts": 25158,
+      "testing": 25159,
+      "Ġamplify": 25160,
+      "Ġtuples": 25161,
+      "proliferative": 25162,
+      "ĠParas": 25163,
+      "Ġmagnets": 25164,
+      "Ġchemokines": 25165,
+      "ĠMitchell": 25166,
+      "ĠPetri": 25167,
+      "holtz": 25168,
+      "ych": 25169,
+      "matrices": 25170,
+      "Ġcorrecting": 25171,
+      "ĠPCa": 25172,
+      "ynamically": 25173,
+      "ĠNAFLD": 25174,
+      "Ġeffluent": 25175,
+      "itum": 25176,
+      "Ġthrows": 25177,
+      "ĠGuid": 25178,
+      "ochromatic": 25179,
+      "ĠFro": 25180,
+      "idad": 25181,
+      "romagnetism": 25182,
+      "Herm": 25183,
+      "ĠSpi": 25184,
+      "ĠQuas": 25185,
+      "domains": 25186,
+      "Ġquadrant": 25187,
+      "ĠSOX": 25188,
+      "ĠGovernor": 25189,
+      "Ġamenable": 25190,
+      "held": 25191,
+      "ĠCul": 25192,
+      "Ġunderwater": 25193,
+      "ĠKron": 25194,
+      "ĠSpati": 25195,
+      "anoyl": 25196,
+      "CU": 25197,
+      "ovir": 25198,
+      "Ġdemographics": 25199,
+      "Within": 25200,
+      "ĠMé": 25201,
+      "textsf": 25202,
+      "ĠLabel": 25203,
+      "Ġgenuine": 25204,
+      "Ġhill": 25205,
+      "ĠLaz": 25206,
+      "Ġtesticular": 25207,
+      "ĠBrow": 25208,
+      "ICATION": 25209,
+      "¡": 25210,
+      "ĠAIC": 25211,
+      "ancomycin": 25212,
+      "strual": 25213,
+      "Ġarrested": 25214,
+      "ĠSom": 25215,
+      "ĠIHC": 25216,
+      "ĠPose": 25217,
+      "ĠMö": 25218,
+      "istar": 25219,
+      "ĠPAM": 25220,
+      "ĠHCT": 25221,
+      "Ġtypedef": 25222,
+      "ĠMorse": 25223,
+      "ĠLeishman": 25224,
+      "limb": 25225,
+      "Ġspheroid": 25226,
+      "osely": 25227,
+      "ĠGuinea": 25228,
+      "renew": 25229,
+      "Ġpsoriasis": 25230,
+      "ista": 25231,
+      "ĠChung": 25232,
+      "orthogonal": 25233,
+      "ĠShear": 25234,
+      "ĠMuslim": 25235,
+      "ĠPict": 25236,
+      "Integer": 25237,
+      "Ġspacer": 25238,
+      "Ly": 25239,
+      "Ġdermal": 25240,
+      "Ġoncology": 25241,
+      "Ġdp": 25242,
+      "Ġphotoluminescence": 25243,
+      "regon": 25244,
+      "aminase": 25245,
+      "Ġáºĭ": 25246,
+      "Instance": 25247,
+      "verb": 25248,
+      "Ġmethylated": 25249,
+      "ĠGem": 25250,
+      "istently": 25251,
+      "ĠMgCl": 25252,
+      "ĠElevated": 25253,
+      "⣩": 25254,
+      "onstruct": 25255,
+      "Ġsnapshot": 25256,
+      "enem": 25257,
+      "ĠDisk": 25258,
+      "Ġhydrostatic": 25259,
+      "Ġïĥª": 25260,
+      "vor": 25261,
+      "ĠIE": 25262,
+      "ĠLY": 25263,
+      "ORF": 25264,
+      "Ġfoil": 25265,
+      "male": 25266,
+      "Ġdepended": 25267,
+      "sparse": 25268,
+      "Ġmetas": 25269,
+      "Ġtextures": 25270,
+      "Ġstacks": 25271,
+      "MHz": 25272,
+      "Ġfn": 25273,
+      "Ġultrac": 25274,
+      "ĠShould": 25275,
+      "Vec": 25276,
+      "nine": 25277,
+      "infinite": 25278,
+      "ĠLawrence": 25279,
+      "ĠInventory": 25280,
+      "ĠProstate": 25281,
+      "Ġgesture": 25282,
+      "ĠSuzuki": 25283,
+      "Abs": 25284,
+      "ricane": 25285,
+      "ĠPeriodic": 25286,
+      "Myc": 25287,
+      "ifiable": 25288,
+      "Ġinefficient": 25289,
+      "Ġcollapsed": 25290,
+      "Ġtopologically": 25291,
+      "Ġpreferable": 25292,
+      "Ġbronchial": 25293,
+      "uston": 25294,
+      "Ġflexion": 25295,
+      "ourney": 25296,
+      "translation": 25297,
+      "Ġepitaxial": 25298,
+      "Ġirradiance": 25299,
+      "Ġneighbours": 25300,
+      "switch": 25301,
+      "Ġactuators": 25302,
+      "SOD": 25303,
+      "mir": 25304,
+      "dies": 25305,
+      "ikawa": 25306,
+      "ĠALL": 25307,
+      "ĠRSV": 25308,
+      "ĠHEP": 25309,
+      "Ġendurance": 25310,
+      "connection": 25311,
+      "Ġgestures": 25312,
+      "odontic": 25313,
+      "ĠUnc": 25314,
+      "Ġdismutase": 25315,
+      "Having": 25316,
+      "mix": 25317,
+      "Ġneurogenesis": 25318,
+      "Ġmyocardium": 25319,
+      "ĠRussell": 25320,
+      "Hist": 25321,
+      "ĠSPI": 25322,
+      "triazol": 25323,
+      "agulant": 25324,
+      "ĠRequired": 25325,
+      "ĠshRNA": 25326,
+      "ĠArthur": 25327,
+      "Ġspawning": 25328,
+      "dried": 25329,
+      "Ġrectif": 25330,
+      "ĠÃī": 25331,
+      "Ġosteogenic": 25332,
+      "replace": 25333,
+      "Ġgaining": 25334,
+      "Ġneutralization": 25335,
+      "ĠHartree": 25336,
+      "Ġfollicles": 25337,
+      "Ġreligion": 25338,
+      "Ġduplex": 25339,
+      "Ġtransients": 25340,
+      "amped": 25341,
+      "Ġmicrotubules": 25342,
+      "interest": 25343,
+      "Ġsteels": 25344,
+      "Batch": 25345,
+      "Ġdenaturation": 25346,
+      "ĠPhillips": 25347,
+      "Ġquiet": 25348,
+      "ĠBureau": 25349,
+      "ĠRare": 25350,
+      "Ġquercetin": 25351,
+      "aults": 25352,
+      "Ġelution": 25353,
+      "uka": 25354,
+      "ĠInterpretation": 25355,
+      "RV": 25356,
+      "ĠESC": 25357,
+      "ĠKom": 25358,
+      "arettes": 25359,
+      "ĠïģĦ": 25360,
+      "Ġtradition": 25361,
+      "Ġdissected": 25362,
+      "Neigh": 25363,
+      "Ġsheaves": 25364,
+      "Ġbelonged": 25365,
+      "ĠHistoric": 25366,
+      "ĠOE": 25367,
+      "Ġjson": 25368,
+      "lemma": 25369,
+      "ĠYAP": 25370,
+      "odext": 25371,
+      "interface": 25372,
+      "Ġextremity": 25373,
+      "crossing": 25374,
+      "precedented": 25375,
+      "according": 25376,
+      "Ġconstructive": 25377,
+      "ĠStimulation": 25378,
+      "ĠHFD": 25379,
+      "Ġwavenumber": 25380,
+      "Ġhrs": 25381,
+      "Ġpapillomavirus": 25382,
+      "Ġvomiting": 25383,
+      "Ġreactivation": 25384,
+      "ometrically": 25385,
+      "ĠDimensions": 25386,
+      "objects": 25387,
+      "orton": 25388,
+      "ĠMathem": 25389,
+      "ĠOlive": 25390,
+      "Ġcrosstalk": 25391,
+      "partite": 25392,
+      "opathies": 25393,
+      "ĠCNTs": 25394,
+      "rousal": 25395,
+      "Ġcrowd": 25396,
+      "ĠLangmuir": 25397,
+      "ĠTox": 25398,
+      "echanics": 25399,
+      "imus": 25400,
+      "ĠShock": 25401,
+      "tanh": 25402,
+      "ĠBrillouin": 25403,
+      "Ġtransferring": 25404,
+      "Ġellipse": 25405,
+      "ĠAddition": 25406,
+      "ĠRural": 25407,
+      "Ġgeodesics": 25408,
+      "GEM": 25409,
+      "ĠPOS": 25410,
+      "ĠMission": 25411,
+      "ocarp": 25412,
+      "ĠJane": 25413,
+      "Lie": 25414,
+      "freq": 25415,
+      "opot": 25416,
+      "ĠVibrio": 25417,
+      "ĠObj": 25418,
+      "erts": 25419,
+      "ĠTrials": 25420,
+      "CFT": 25421,
+      "ĠCodes": 25422,
+      "μg": 25423,
+      "Reference": 25424,
+      "ĠFung": 25425,
+      "ĠSuppression": 25426,
+      "hog": 25427,
+      "Ġresistive": 25428,
+      "Chi": 25429,
+      "intered": 25430,
+      "Ġpostmenopausal": 25431,
+      "Statistical": 25432,
+      "ĠEdwards": 25433,
+      "Ġses": 25434,
+      "Ġfarming": 25435,
+      "quartile": 25436,
+      "cooled": 25437,
+      "Ġnanop": 25438,
+      "ĠProbing": 25439,
+      "ĠBernard": 25440,
+      "uni": 25441,
+      "ieties": 25442,
+      "ĠMarket": 25443,
+      "osum": 25444,
+      "ĠMessage": 25445,
+      "Ġaxiom": 25446,
+      "cg": 25447,
+      "ĠMoving": 25448,
+      "Resolution": 25449,
+      "Ġadsorbent": 25450,
+      "Ġmultin": 25451,
+      "Ġineffective": 25452,
+      "propag": 25453,
+      "hardt": 25454,
+      "Saharan": 25455,
+      "Wil": 25456,
+      "ĠIvan": 25457,
+      "irubin": 25458,
+      "Ġtrabec": 25459,
+      "alli": 25460,
+      "ĠCDCl": 25461,
+      "Ġsew": 25462,
+      "ĠIss": 25463,
+      "Ġaggression": 25464,
+      "ĠJuan": 25465,
+      "Ġdispersions": 25466,
+      "Ġauxin": 25467,
+      "FET": 25468,
+      "lp": 25469,
+      "reach": 25470,
+      "ĠPGE": 25471,
+      "chestr": 25472,
+      "Ġlecture": 25473,
+      "ĠDonald": 25474,
+      "slip": 25475,
+      "ĠHbA": 25476,
+      "ĠSecure": 25477,
+      "ĠBeh": 25478,
+      "Ġdamages": 25479,
+      "WH": 25480,
+      "alkyl": 25481,
+      "Ha": 25482,
+      "ĠThanks": 25483,
+      "Ġsensitization": 25484,
+      "Ġwaterm": 25485,
+      "Ġtwins": 25486,
+      "Ġcultivar": 25487,
+      "Ġzeolite": 25488,
+      "Variable": 25489,
+      "ĠBent": 25490,
+      "Ġantisense": 25491,
+      "ĠHansen": 25492,
+      "repreneur": 25493,
+      "ĠSNe": 25494,
+      "ĠEMG": 25495,
+      "Ġreacted": 25496,
+      "Ġoverflow": 25497,
+      "Ġformalin": 25498,
+      "ĠUsually": 25499,
+      "olybden": 25500,
+      "Ġacad": 25501,
+      "ATURE": 25502,
+      "Ġwaveguides": 25503,
+      "Ġchunk": 25504,
+      "Ġmodifies": 25505,
+      "Ġeryt": 25506,
+      "ĠZhong": 25507,
+      "Ġgranule": 25508,
+      "Ġcs": 25509,
+      "ĠGrade": 25510,
+      "Ġlandmarks": 25511,
+      "uristic": 25512,
+      "Ġamines": 25513,
+      "ĠIntrinsic": 25514,
+      "Ġerroneous": 25515,
+      "Ġlockdown": 25516,
+      "ypti": 25517,
+      "Child": 25518,
+      "Ġuniversities": 25519,
+      "Ġparasit": 25520,
+      "Ġignition": 25521,
+      "Tim": 25522,
+      "araj": 25523,
+      "ravel": 25524,
+      "ĠLands": 25525,
+      "ĠCircular": 25526,
+      "Ġrotate": 25527,
+      "Patients": 25528,
+      "ĠWB": 25529,
+      "Ġmyelin": 25530,
+      "ĠWeiss": 25531,
+      "Ġdipolar": 25532,
+      "Ġfollicle": 25533,
+      "ĠWatson": 25534,
+      "ĠIncor": 25535,
+      "Ġfoundations": 25536,
+      "ĠPip": 25537,
+      "Ġpressing": 25538,
+      "Ġforbidden": 25539,
+      "avan": 25540,
+      "ĠmAb": 25541,
+      "union": 25542,
+      "ĠFresh": 25543,
+      "ĠCorp": 25544,
+      "floxacin": 25545,
+      "coordinate": 25546,
+      "Ġshunt": 25547,
+      "Ġconstituted": 25548,
+      "aniline": 25549,
+      "Ġtweets": 25550,
+      "ĠChow": 25551,
+      "Ġmobilization": 25552,
+      "zyk": 25553,
+      "EST": 25554,
+      "neigh": 25555,
+      "ĠMeng": 25556,
+      "ĠResNet": 25557,
+      "ĠJet": 25558,
+      "Ġluminous": 25559,
+      "Ġstressors": 25560,
+      "does": 25561,
+      "trifluoromethyl": 25562,
+      "Ġconcert": 25563,
+      "ĠChoice": 25564,
+      "phim": 25565,
+      "alcoholic": 25566,
+      "ochem": 25567,
+      "iltered": 25568,
+      "Ġpredictable": 25569,
+      "Ġtran": 25570,
+      "ĠPra": 25571,
+      "Ġvalves": 25572,
+      "Ġautonomy": 25573,
+      "regulate": 25574,
+      "ĠBeach": 25575,
+      "ĠOntology": 25576,
+      "Ġisofl": 25577,
+      "Ġquoted": 25578,
+      "ĠLex": 25579,
+      "thy": 25580,
+      "Ġcomplaints": 25581,
+      "ĠTrees": 25582,
+      "Ġopposing": 25583,
+      "ĠAcceler": 25584,
+      "contrast": 25585,
+      "Ġcompeted": 25586,
+      "OE": 25587,
+      "ĠRoche": 25588,
+      "issance": 25589,
+      "Ġpeace": 25590,
+      "ĠAim": 25591,
+      "Ġinfertility": 25592,
+      "ĠAntarctica": 25593,
+      "thien": 25594,
+      "Summ": 25595,
+      "Ġjudgments": 25596,
+      "amides": 25597,
+      "Ġspill": 25598,
+      "Ġhereafter": 25599,
+      "ĠConstit": 25600,
+      "computer": 25601,
+      "Ġbegun": 25602,
+      "ocentric": 25603,
+      "Ġpumps": 25604,
+      "medium": 25605,
+      "chol": 25606,
+      "metallic": 25607,
+      "Ġflares": 25608,
+      "Ġpetroleum": 25609,
+      "Ġwithd": 25610,
+      "ĠTheatre": 25611,
+      "Ġunlabeled": 25612,
+      "Ġregularized": 25613,
+      "osteric": 25614,
+      "ĠPFS": 25615,
+      "Ġunem": 25616,
+      "Ġpresently": 25617,
+      "Ġbuffered": 25618,
+      "affinity": 25619,
+      "ĠDemographic": 25620,
+      "ĠKondo": 25621,
+      "Ġcenturies": 25622,
+      "Ġmigratory": 25623,
+      "arynx": 25624,
+      "Associated": 25625,
+      "anilino": 25626,
+      "grown": 25627,
+      "ĠExecutive": 25628,
+      "ĠEk": 25629,
+      "ĠHemat": 25630,
+      "ĠPlayer": 25631,
+      "ĠCHD": 25632,
+      "flex": 25633,
+      "ĠSever": 25634,
+      "altham": 25635,
+      "impro": 25636,
+      "anet": 25637,
+      "ocyst": 25638,
+      "ĠAster": 25639,
+      "COL": 25640,
+      "ĠSimilarity": 25641,
+      "ĠHoward": 25642,
+      "Ġmulticast": 25643,
+      "ĠEnsemble": 25644,
+      "ìĹ": 25645,
+      "olys": 25646,
+      "ĠGenomics": 25647,
+      "Ġresonators": 25648,
+      "Ġfistula": 25649,
+      "onen": 25650,
+      "users": 25651,
+      "Ġhypo": 25652,
+      "rogens": 25653,
+      "Ġmedal": 25654,
+      "ĠMIP": 25655,
+      "Ġvoltam": 25656,
+      "Ġappreciated": 25657,
+      "ĠPé": 25658,
+      "ĠGaia": 25659,
+      "Ġbuckling": 25660,
+      "Ġcongruence": 25661,
+      "furyl": 25662,
+      "ĠEpstein": 25663,
+      "Ġcascades": 25664,
+      "gold": 25665,
+      "Ġanhyd": 25666,
+      "Ġgraduated": 25667,
+      "Memory": 25668,
+      "ĠIndustry": 25669,
+      "ĠSchneider": 25670,
+      "Ġemployee": 25671,
+      "ĠCorn": 25672,
+      "MAC": 25673,
+      "rove": 25674,
+      "ropod": 25675,
+      "service": 25676,
+      "ĠOxidation": 25677,
+      "Ġenumeration": 25678,
+      "mad": 25679,
+      "ĠClose": 25680,
+      "ĠModular": 25681,
+      "Ġprogeny": 25682,
+      "Ġgt": 25683,
+      "reading": 25684,
+      "ĠIndic": 25685,
+      "opathologic": 25686,
+      "ĠPFNGL": 25687,
+      "XL": 25688,
+      "cis": 25689,
+      "ĠMike": 25690,
+      "ĠBBB": 25691,
+      "ĠExtreme": 25692,
+      "ĠChoose": 25693,
+      "Ġhorizontally": 25694,
+      "ĠASSERT": 25695,
+      "Ġglucocorticoid": 25696,
+      "Bay": 25697,
+      "Ġpdf": 25698,
+      "Ġcontainers": 25699,
+      "ĠLOC": 25700,
+      "ĠYield": 25701,
+      "oprote": 25702,
+      "Ġfructose": 25703,
+      "ĠICC": 25704,
+      "Ġdecid": 25705,
+      "rimidine": 25706,
+      "Ġfragmented": 25707,
+      "Ġisomorphisms": 25708,
+      "м": 25709,
+      "Ġintegrates": 25710,
+      "Ġfibration": 25711,
+      "ĠâĬ¤": 25712,
+      "Ġxenograft": 25713,
+      "nucleon": 25714,
+      "ĠCSP": 25715,
+      "Ġsut": 25716,
+      "ĠSpir": 25717,
+      "Ġdissoci": 25718,
+      "ĠTBI": 25719,
+      "ĠForces": 25720,
+      "Ġhypersurface": 25721,
+      "Ġmyosin": 25722,
+      "ĠQueensland": 25723,
+      "Neg": 25724,
+      "ĠURL": 25725,
+      "bind": 25726,
+      "Applied": 25727,
+      "ĠDob": 25728,
+      "ĠKE": 25729,
+      "Ġmemor": 25730,
+      "ĠArabic": 25731,
+      "ĠLateral": 25732,
+      "ĠStart": 25733,
+      "nose": 25734,
+      "tibility": 25735,
+      "asters": 25736,
+      "Ġusability": 25737,
+      "Ġincenti": 25738,
+      "ymn": 25739,
+      "ĠAnalytic": 25740,
+      "Pet": 25741,
+      "ĠMask": 25742,
+      "World": 25743,
+      "brand": 25744,
+      "Ġeliminates": 25745,
+      "Ġmerit": 25746,
+      "ĠPhilippines": 25747,
+      "ĠBCL": 25748,
+      "ĠOri": 25749,
+      "Ġparadigms": 25750,
+      "ĠInters": 25751,
+      "rizona": 25752,
+      "Ġconception": 25753,
+      "Ġrelied": 25754,
+      "ĠJoe": 25755,
+      "ĠApple": 25756,
+      "Ġlightweight": 25757,
+      "mortem": 25758,
+      "olig": 25759,
+      "Ġviz": 25760,
+      "Ġstones": 25761,
+      "Ġkeywords": 25762,
+      "ĠSecretary": 25763,
+      "TN": 25764,
+      "older": 25765,
+      "ĠIntestinal": 25766,
+      "Ġpossessed": 25767,
+      "Ġmonotonicity": 25768,
+      "emitting": 25769,
+      "ĠDefining": 25770,
+      "ĠParticularly": 25771,
+      "Ġautomorphisms": 25772,
+      "Ġerythemat": 25773,
+      "ĠWaters": 25774,
+      "ĠCyclic": 25775,
+      "maximal": 25776,
+      "xty": 25777,
+      "ĠSad": 25778,
+      "Ġuranium": 25779,
+      "Ġhypothalamus": 25780,
+      "ĠSUMO": 25781,
+      "Ġdealt": 25782,
+      "Ġkits": 25783,
+      "Ġpainting": 25784,
+      "ĠSier": 25785,
+      "chool": 25786,
+      "ODO": 25787,
+      "surfaces": 25788,
+      "ĠPneum": 25789,
+      "organized": 25790,
+      "ĠCPT": 25791,
+      "Ġinsoluble": 25792,
+      "ĠCoherent": 25793,
+      "Ġrecessive": 25794,
+      "Ġbivariate": 25795,
+      "Ġedit": 25796,
+      "Ġnationwide": 25797,
+      "MODE": 25798,
+      "chest": 25799,
+      "ĠSLC": 25800,
+      "Ġintraperitoneal": 25801,
+      "ĠDisordered": 25802,
+      "Ġinsufficiency": 25803,
+      "iev": 25804,
+      "iazole": 25805,
+      "Write": 25806,
+      "ĠDATA": 25807,
+      "toral": 25808,
+      "Ġqualities": 25809,
+      "Ġpossessing": 25810,
+      "ĠMats": 25811,
+      "Ġretinopathy": 25812,
+      "ĠBK": 25813,
+      "Ġnovelty": 25814,
+      "ceans": 25815,
+      "Ġreserves": 25816,
+      "ĠNADH": 25817,
+      "Ġisotherm": 25818,
+      "Ġsoldiers": 25819,
+      "pb": 25820,
+      "iterpen": 25821,
+      "ĠAgents": 25822,
+      "zu": 25823,
+      "Ġunwanted": 25824,
+      "Ġhyperparameters": 25825,
+      "ecan": 25826,
+      "ĠSES": 25827,
+      "ĠFG": 25828,
+      "ĠNavig": 25829,
+      "Ġtriangulation": 25830,
+      "Ġnetworking": 25831,
+      "Ġpolystyrene": 25832,
+      "Ġinductively": 25833,
+      "breviations": 25834,
+      "Ġneuromuscular": 25835,
+      "ĠLinux": 25836,
+      "studied": 25837,
+      "ĠBeing": 25838,
+      "Ġdeficiencies": 25839,
+      "ĠMatrices": 25840,
+      "Ġwearing": 25841,
+      "Ġhadrons": 25842,
+      "amyl": 25843,
+      "Ġdiscourse": 25844,
+      "ochlor": 25845,
+      "ĠMelan": 25846,
+      "ĠLan": 25847,
+      "VL": 25848,
+      "Ġmunicipal": 25849,
+      "Ġenrollment": 25850,
+      "ĠSymmetric": 25851,
+      "Ġdisciplines": 25852,
+      "ĠBaron": 25853,
+      "Research": 25854,
+      "Ġmagnetite": 25855,
+      "omide": 25856,
+      "polarization": 25857,
+      "leys": 25858,
+      "Ġseemingly": 25859,
+      "hepatic": 25860,
+      "Ġclo": 25861,
+      "ĠQuatern": 25862,
+      "Ġcompetit": 25863,
+      "Requ": 25864,
+      "gauge": 25865,
+      "Ġhydrochloride": 25866,
+      "dropout": 25867,
+      "panel": 25868,
+      "Ġaspirin": 25869,
+      "ĠRUN": 25870,
+      "Ġribbon": 25871,
+      "Ġinaccurate": 25872,
+      "ĠPall": 25873,
+      "ducers": 25874,
+      "Throughout": 25875,
+      "Ġcellul": 25876,
+      "Ġsuspect": 25877,
+      "Ġallelic": 25878,
+      "Ġsnake": 25879,
+      "ordinated": 25880,
+      "ĠAutophagy": 25881,
+      "Ġeig": 25882,
+      "Ġrif": 25883,
+      "ĠKennedy": 25884,
+      "Ġbottle": 25885,
+      "ĠYouth": 25886,
+      "awed": 25887,
+      "linearity": 25888,
+      "uker": 25889,
+      "ĠOX": 25890,
+      "extension": 25891,
+      "Ġward": 25892,
+      "ĠComplexes": 25893,
+      "Ġbiosensor": 25894,
+      "ĠCartan": 25895,
+      "dn": 25896,
+      "Ġsonic": 25897,
+      "Ġindexing": 25898,
+      "Ġdv": 25899,
+      "reliable": 25900,
+      "pk": 25901,
+      "RENT": 25902,
+      "Ġtanks": 25903,
+      "ĠHet": 25904,
+      "ĠWing": 25905,
+      "ĠCuO": 25906,
+      "Ġprintf": 25907,
+      "Ġluminosities": 25908,
+      "course": 25909,
+      "Ġscram": 25910,
+      "Ġsampler": 25911,
+      "Ġmultipliers": 25912,
+      "Default": 25913,
+      "odil": 25914,
+      "intr": 25915,
+      "sequencing": 25916,
+      "Ġtransmissions": 25917,
+      "ĠWhit": 25918,
+      "ĠOpportun": 25919,
+      "Ġinternally": 25920,
+      "Ġacknowledges": 25921,
+      "ĠEdition": 25922,
+      "Ġarteri": 25923,
+      "Ġalbedo": 25924,
+      "ĠNucleotide": 25925,
+      "Ġyes": 25926,
+      "ĠRelativistic": 25927,
+      "Ġvotes": 25928,
+      "ĠFormulation": 25929,
+      "uscitation": 25930,
+      "Ġconcurrently": 25931,
+      "uin": 25932,
+      "Ġnoninvasive": 25933,
+      "Ġprimates": 25934,
+      "μl": 25935,
+      "Ġsubtropical": 25936,
+      "gun": 25937,
+      "ĠSoutheast": 25938,
+      "ön": 25939,
+      "Ġequator": 25940,
+      "Ġworkshop": 25941,
+      "Ġschist": 25942,
+      "undant": 25943,
+      "ĠMODIS": 25944,
+      "tar": 25945,
+      "Ġaeg": 25946,
+      "Ġplotting": 25947,
+      "ĠDET": 25948,
+      "Manager": 25949,
+      "uned": 25950,
+      "oxifen": 25951,
+      "ĠInver": 25952,
+      "Ġxanth": 25953,
+      "ĠServer": 25954,
+      "Ġstretched": 25955,
+      "Global": 25956,
+      "Core": 25957,
+      "ĠWeber": 25958,
+      "yard": 25959,
+      "Ġexplores": 25960,
+      "ĠBiography": 25961,
+      "SNP": 25962,
+      "ĠNeutrino": 25963,
+      "Ġkilometres": 25964,
+      "Ġcommutes": 25965,
+      "Ġacceptability": 25966,
+      "ĠAntibodies": 25967,
+      "icol": 25968,
+      "Ġmuseum": 25969,
+      "Ġdenit": 25970,
+      "Ġextrapolated": 25971,
+      "Ġacetylcholine": 25972,
+      "Token": 25973,
+      "ĠFock": 25974,
+      "onde": 25975,
+      "Ġdiscriminative": 25976,
+      "ĠMant": 25977,
+      "Ġessence": 25978,
+      "celand": 25979,
+      "ĠChair": 25980,
+      "Ġintegrative": 25981,
+      "ĠSPD": 25982,
+      "henium": 25983,
+      "arbonate": 25984,
+      "BASE": 25985,
+      "regulates": 25986,
+      "patch": 25987,
+      "Ġdib": 25988,
+      "Ġantisymmetric": 25989,
+      "Ġwearable": 25990,
+      "Edge": 25991,
+      "rets": 25992,
+      "Ġperceive": 25993,
+      "ĠMagnesium": 25994,
+      "adows": 25995,
+      "Ġdisposal": 25996,
+      "Ġairport": 25997,
+      "ausea": 25998,
+      "fits": 25999,
+      "Ġnecro": 26000,
+      "ĠSIN": 26001,
+      "ĠDuc": 26002,
+      "ĠReading": 26003,
+      "bys": 26004,
+      "Ġreflective": 26005,
+      "his": 26006,
+      "ometries": 26007,
+      "Ġvirial": 26008,
+      "Ġartificially": 26009,
+      "children": 26010,
+      "ĠUltrasound": 26011,
+      "VIEW": 26012,
+      "Ġsculpt": 26013,
+      "Ġsurf": 26014,
+      "Ġsexually": 26015,
+      "Ġgeometrically": 26016,
+      "Ġdivisors": 26017,
+      "Ġinitiatives": 26018,
+      "acci": 26019,
+      "Ġkeratinocytes": 26020,
+      "aR": 26021,
+      "arot": 26022,
+      "Ġïĥ¨": 26023,
+      "computed": 26024,
+      "ĠTCGA": 26025,
+      "psychological": 26026,
+      "ĠMAN": 26027,
+      "ĠMPC": 26028,
+      "ticing": 26029,
+      "limiting": 26030,
+      "amins": 26031,
+      "Ġsurfactants": 26032,
+      "ĠSerb": 26033,
+      "Ġrhythms": 26034,
+      "ĠRouting": 26035,
+      "wang": 26036,
+      "Ġmicrostructures": 26037,
+      "ophytes": 26038,
+      "Ġanalgesic": 26039,
+      "FOR": 26040,
+      "qual": 26041,
+      "Ġpublish": 26042,
+      "ĠTiming": 26043,
+      "porous": 26044,
+      "ranging": 26045,
+      "eron": 26046,
+      "ĠZi": 26047,
+      "ĠMarshall": 26048,
+      "Width": 26049,
+      "Ġisomers": 26050,
+      "Ġ·": 26051,
+      "phenoxy": 26052,
+      "Ġureth": 26053,
+      "robl": 26054,
+      "Ġmentioning": 26055,
+      "ozyme": 26056,
+      "ĠLud": 26057,
+      "Ġopposition": 26058,
+      "Ġabandoned": 26059,
+      "Ġroutines": 26060,
+      "ĠHST": 26061,
+      "mutex": 26062,
+      "coded": 26063,
+      "eating": 26064,
+      "tert": 26065,
+      "emiconductor": 26066,
+      "dw": 26067,
+      "Ġbaryons": 26068,
+      "Ġleucine": 26069,
+      "otron": 26070,
+      "Ġendos": 26071,
+      "Ġreproduces": 26072,
+      "Ġanalgesia": 26073,
+      "Ġimmunoreactivity": 26074,
+      "ĠPrep": 26075,
+      "ĠGarcÃŃa": 26076,
+      "Ġincoherent": 26077,
+      "aned": 26078,
+      "lepton": 26079,
+      "andra": 26080,
+      "ulae": 26081,
+      "ĠHidden": 26082,
+      "FV": 26083,
+      "Ġgeneralizes": 26084,
+      "ĠStevens": 26085,
+      "ĠFoster": 26086,
+      "Ġfreshly": 26087,
+      "Ġhf": 26088,
+      "Denote": 26089,
+      "oes": 26090,
+      "ĠDin": 26091,
+      "Ġdetox": 26092,
+      "Ġdecoupled": 26093,
+      "Ġseparations": 26094,
+      "ucleotide": 26095,
+      "Ġelectrophysiological": 26096,
+      "ĠBALB": 26097,
+      "QTL": 26098,
+      "ĠACh": 26099,
+      "ĠRele": 26100,
+      "quez": 26101,
+      "MnO": 26102,
+      "ectures": 26103,
+      "Ġischa": 26104,
+      "Ġinsulators": 26105,
+      "cellulose": 26106,
+      "ĠFLAG": 26107,
+      "ombic": 26108,
+      "ĠUsed": 26109,
+      "jiang": 26110,
+      "expansion": 26111,
+      "ĠRepeat": 26112,
+      "ĠReserve": 26113,
+      "abelian": 26114,
+      "ĠHunting": 26115,
+      "GRO": 26116,
+      "lyte": 26117,
+      "ĠBark": 26118,
+      "Ġcreative": 26119,
+      "Ġbend": 26120,
+      "elerated": 26121,
+      "dish": 26122,
+      "Ġhighway": 26123,
+      "Ġcrossings": 26124,
+      "just": 26125,
+      "ono": 26126,
+      "ullivan": 26127,
+      "ĠDead": 26128,
+      "Ġtradeoff": 26129,
+      "eon": 26130,
+      "ogical": 26131,
+      "experiment": 26132,
+      "Ġconfers": 26133,
+      "ĠDot": 26134,
+      "Ġcoils": 26135,
+      "Ġaxion": 26136,
+      "ĠIRS": 26137,
+      "ĠÅ©": 26138,
+      "Ġglacier": 26139,
+      "ĠMoscow": 26140,
+      "ĠSpringer": 26141,
+      "Ġinvis": 26142,
+      "ĠArnold": 26143,
+      "University": 26144,
+      "attern": 26145,
+      "peror": 26146,
+      "ĠLimits": 26147,
+      "Ġincompatible": 26148,
+      "rather": 26149,
+      "ĠTes": 26150,
+      "Ġfailing": 26151,
+      "Ġthickening": 26152,
+      "Ġestradiol": 26153,
+      "asse": 26154,
+      "Ġnecessit": 26155,
+      "Ġsacrificed": 26156,
+      "ĠSear": 26157,
+      "ĠNorthe": 26158,
+      "raisebox": 26159,
+      "ĠSlow": 26160,
+      "ĠMunic": 26161,
+      "Ġlearner": 26162,
+      "igenic": 26163,
+      "Ġdermatitis": 26164,
+      "uten": 26165,
+      "Ġdeer": 26166,
+      "Ġhistamine": 26167,
+      "Lat": 26168,
+      "Mal": 26169,
+      "illy": 26170,
+      "Ġgeochemical": 26171,
+      "Ġspermatozoa": 26172,
+      "Ġvinyl": 26173,
+      "emet": 26174,
+      "Ġeffectors": 26175,
+      "ĠEncyclopedia": 26176,
+      "Ġordinal": 26177,
+      "Ġcontroversy": 26178,
+      "ĠPerspectives": 26179,
+      "oviruses": 26180,
+      "marked": 26181,
+      "ĠSPE": 26182,
+      "ĠNutri": 26183,
+      "Ġadhere": 26184,
+      "ĠHighway": 26185,
+      "Ġdistillation": 26186,
+      "MRT": 26187,
+      "pletion": 26188,
+      "Ġannihil": 26189,
+      "Ġwavefunction": 26190,
+      "Ġconfigured": 26191,
+      "Ġmethionine": 26192,
+      "Low": 26193,
+      "sensor": 26194,
+      "ĠSnow": 26195,
+      "Sample": 26196,
+      "Ġdefinitely": 26197,
+      "ĠMeth": 26198,
+      "rypt": 26199,
+      "Ġprompted": 26200,
+      "Ġmonolith": 26201,
+      "ĠEnvironments": 26202,
+      "tm": 26203,
+      "ĠCOD": 26204,
+      "oris": 26205,
+      "equations": 26206,
+      "âĺĨ": 26207,
+      "ĠNeighbor": 26208,
+      "Ġimagine": 26209,
+      "ĠUsers": 26210,
+      "ĠCamera": 26211,
+      "ĠModification": 26212,
+      "ĠAttacks": 26213,
+      "Ġinhalation": 26214,
+      "áº": 26215,
+      "Ġventil": 26216,
+      "ĠNU": 26217,
+      "ĠContrast": 26218,
+      "Ġconfining": 26219,
+      "Service": 26220,
+      "Wallis": 26221,
+      "ĠATR": 26222,
+      "Ġsubduction": 26223,
+      "Ġïģ¢": 26224,
+      "Ġtitration": 26225,
+      "Roche": 26226,
+      "viv": 26227,
+      "Ġbears": 26228,
+      "bola": 26229,
+      "Ġblinded": 26230,
+      "measures": 26231,
+      "ĠStack": 26232,
+      "occurrence": 26233,
+      "Ġpermeation": 26234,
+      "lar": 26235,
+      "eptors": 26236,
+      "ĠDIF": 26237,
+      "corrhiz": 26238,
+      "ĠVisc": 26239,
+      "figurable": 26240,
+      "Ġscheduler": 26241,
+      "Ġoccasions": 26242,
+      "amboo": 26243,
+      "Ġamp": 26244,
+      "gain": 26245,
+      "ĠCit": 26246,
+      "Ġpreceded": 26247,
+      "Ġtactile": 26248,
+      "Ġïĥ¦": 26249,
+      "generic": 26250,
+      "Ġretrograde": 26251,
+      "Ġfans": 26252,
+      "Ġfisher": 26253,
+      "Ġlights": 26254,
+      "eeper": 26255,
+      "Ġundesirable": 26256,
+      "wald": 26257,
+      "embol": 26258,
+      "Ġwrist": 26259,
+      "Ġauthorized": 26260,
+      "Ġchondrocytes": 26261,
+      "ĠEPA": 26262,
+      "neu": 26263,
+      "ĠOperations": 26264,
+      "Ġcheap": 26265,
+      "Ġanionic": 26266,
+      "ĠOregon": 26267,
+      "cot": 26268,
+      "reason": 26269,
+      "existence": 26270,
+      "ĠFinancial": 26271,
+      "olybdenum": 26272,
+      "cus": 26273,
+      "ĠNON": 26274,
+      "Ġlocked": 26275,
+      "Bit": 26276,
+      "Sil": 26277,
+      "mixing": 26278,
+      "ĠSites": 26279,
+      "aproteobacteria": 26280,
+      "ĠInner": 26281,
+      "Ġcarc": 26282,
+      "Ġbiotic": 26283,
+      "ĠFlag": 26284,
+      "Ġmagic": 26285,
+      "kinetic": 26286,
+      "icted": 26287,
+      "Ġbulb": 26288,
+      "supset": 26289,
+      "pez": 26290,
+      "derivative": 26291,
+      "ĠeIF": 26292,
+      "ĠRough": 26293,
+      "directional": 26294,
+      "exit": 26295,
+      "axy": 26296,
+      "xtures": 26297,
+      "phimurium": 26298,
+      "ĠTFs": 26299,
+      "athin": 26300,
+      "Ġorch": 26301,
+      "Ġspectro": 26302,
+      "ductase": 26303,
+      "quinolin": 26304,
+      "Ġgrasp": 26305,
+      "Ġparsing": 26306,
+      "Ġdifficile": 26307,
+      "ĠLDH": 26308,
+      "ĠJupiter": 26309,
+      "ĠFIF": 26310,
+      "ĠPrize": 26311,
+      "Ġintentions": 26312,
+      "session": 26313,
+      "powered": 26314,
+      "ĠBam": 26315,
+      "phasic": 26316,
+      "Ġignoring": 26317,
+      "ĠRichardson": 26318,
+      "principles": 26319,
+      "Ġofficially": 26320,
+      "Ct": 26321,
+      "Ġincon": 26322,
+      "ĠRegulates": 26323,
+      "Ġmisc": 26324,
+      "ĠEZ": 26325,
+      "Ġsynonym": 26326,
+      "Ġunfolding": 26327,
+      "ĠDEC": 26328,
+      "ĠRX": 26329,
+      "PDF": 26330,
+      "Ġbranes": 26331,
+      "typically": 26332,
+      "Ġcages": 26333,
+      "ifolia": 26334,
+      "ugu": 26335,
+      "ollen": 26336,
+      "Ġtablet": 26337,
+      "ĠSah": 26338,
+      "ĠPVD": 26339,
+      "Ġalert": 26340,
+      "Ġformerly": 26341,
+      "ĠKRAS": 26342,
+      "sun": 26343,
+      "Ġdeacetyl": 26344,
+      "Mer": 26345,
+      "Ġskewed": 26346,
+      "ĠPleistocene": 26347,
+      "ĠBetter": 26348,
+      "ĠHud": 26349,
+      "ĠBrook": 26350,
+      "Ġpts": 26351,
+      "ĠHU": 26352,
+      "omo": 26353,
+      "agrass": 26354,
+      "Ġenvironmentally": 26355,
+      "Ġhonest": 26356,
+      "ĠNine": 26357,
+      "Ġpigments": 26358,
+      "links": 26359,
+      "ĠTOP": 26360,
+      "ĠCytoplasm": 26361,
+      "Gib": 26362,
+      "Ġaccessing": 26363,
+      "mias": 26364,
+      "Ġexplosive": 26365,
+      "Ġreside": 26366,
+      "artan": 26367,
+      "Ġtransitional": 26368,
+      "Ġunprecedented": 26369,
+      "Ġrom": 26370,
+      "ĠTNFα": 26371,
+      "Ġprecipitated": 26372,
+      "Ġtie": 26373,
+      "ISS": 26374,
+      "Ġthicker": 26375,
+      "ĠLatent": 26376,
+      "ĠValueError": 26377,
+      "dq": 26378,
+      "dma": 26379,
+      "Ġchromatic": 26380,
+      "ĠSubsection": 26381,
+      "ĠFACS": 26382,
+      "Ġrenormalized": 26383,
+      "Prop": 26384,
+      "mTOR": 26385,
+      "ĠHCO": 26386,
+      "Ġoverlo": 26387,
+      "bsiella": 26388,
+      "ylobacter": 26389,
+      "Ġneuroimaging": 26390,
+      "Ġassemblage": 26391,
+      "Ġexpands": 26392,
+      "ĠîĪ": 26393,
+      "ĠFun": 26394,
+      "Ġcitation": 26395,
+      "IKV": 26396,
+      "Ġtroops": 26397,
+      "inistic": 26398,
+      "Ġcubes": 26399,
+      "Ġfont": 26400,
+      "ĠHos": 26401,
+      "geries": 26402,
+      "Ġsuccessively": 26403,
+      "Ġdecoherence": 26404,
+      "Springer": 26405,
+      "hin": 26406,
+      "atine": 26407,
+      "ĠâĪ¥": 26408,
+      "SAS": 26409,
+      "ét": 26410,
+      "ĠSediment": 26411,
+      "uously": 26412,
+      "ĠWars": 26413,
+      "indicated": 26414,
+      "Ġflask": 26415,
+      "AIDS": 26416,
+      "Ġcra": 26417,
+      "ĠLot": 26418,
+      "Ġprimal": 26419,
+      "Ġjustice": 26420,
+      "zag": 26421,
+      "Ġmaxillary": 26422,
+      "Ġgeneralizations": 26423,
+      "uela": 26424,
+      "Ġtagging": 26425,
+      "Ġpupil": 26426,
+      "Ġinexpensive": 26427,
+      "Ġwatch": 26428,
+      "ĠAMD": 26429,
+      "ĠFir": 26430,
+      "Ġneuroblastoma": 26431,
+      "Ġmaximizes": 26432,
+      "ĠObserved": 26433,
+      "mixture": 26434,
+      "Ġopportunistic": 26435,
+      "trial": 26436,
+      "ahan": 26437,
+      "Ġïģ¬": 26438,
+      "Ġcatar": 26439,
+      "ĠControls": 26440,
+      "ĠNewman": 26441,
+      "Ġmicrostructural": 26442,
+      "borns": 26443,
+      "Ġoxygenation": 26444,
+      "ĠMacro": 26445,
+      "ĠJak": 26446,
+      "plicating": 26447,
+      "Ġoligodend": 26448,
+      "Ġresorption": 26449,
+      "Ġdorm": 26450,
+      "Ġsolvers": 26451,
+      "ĠKruskal": 26452,
+      "ĠRevolution": 26453,
+      "ĠGastro": 26454,
+      "Driven": 26455,
+      "Ġtiter": 26456,
+      "Ġori": 26457,
+      "ĠPCL": 26458,
+      "Ġwetlands": 26459,
+      "Ġarticular": 26460,
+      "CCA": 26461,
+      "enoic": 26462,
+      "Ġtrick": 26463,
+      "operiod": 26464,
+      "ĠCochrane": 26465,
+      "aday": 26466,
+      "ĠCerebral": 26467,
+      "Ġmodulators": 26468,
+      "ĠSSC": 26469,
+      "Ġactivations": 26470,
+      "Ġadapting": 26471,
+      "ĠScalable": 26472,
+      "none": 26473,
+      "pip": 26474,
+      "Ġprivi": 26475,
+      "ĠPseudo": 26476,
+      "Ġdisappears": 26477,
+      "ĠEur": 26478,
+      "Ġunconstrained": 26479,
+      "Ġsubmit": 26480,
+      "Ġreputation": 26481,
+      "atar": 26482,
+      "ĠBai": 26483,
+      "arians": 26484,
+      "ĠIntracellular": 26485,
+      "trees": 26486,
+      "Ġwetting": 26487,
+      "ĠFrances": 26488,
+      "Ġeligibility": 26489,
+      "folder": 26490,
+      "ĠStaff": 26491,
+      "oki": 26492,
+      "Ġstrengthened": 26493,
+      "ĠCob": 26494,
+      "teral": 26495,
+      "ĠYeast": 26496,
+      "bye": 26497,
+      "decoder": 26498,
+      "Ġrainbow": 26499,
+      "perturbed": 26500,
+      "vc": 26501,
+      "Ġsupplemental": 26502,
+      "Ġbirths": 26503,
+      "WO": 26504,
+      "conc": 26505,
+      "stitution": 26506,
+      "hybrid": 26507,
+      "Ġki": 26508,
+      "Ġhypere": 26509,
+      "ĠSMA": 26510,
+      "formula": 26511,
+      "Ġundefined": 26512,
+      "naphth": 26513,
+      "Ġdeclining": 26514,
+      "Ġshielding": 26515,
+      "Yau": 26516,
+      "Ġrever": 26517,
+      "ĠWilk": 26518,
+      "Ġdecimal": 26519,
+      "HCO": 26520,
+      "angered": 26521,
+      "Ġerythrocyte": 26522,
+      "ĉĉĠĠĠ": 26523,
+      "nuclear": 26524,
+      "Ġabnormality": 26525,
+      "Pres": 26526,
+      "Participants": 26527,
+      "ĠWagner": 26528,
+      "Ġfibrils": 26529,
+      "Ġfetus": 26530,
+      "ĠExpress": 26531,
+      "request": 26532,
+      "minimum": 26533,
+      "ĠBooks": 26534,
+      "hetamine": 26535,
+      "ushes": 26536,
+      "ĠBach": 26537,
+      "ĠDOS": 26538,
+      "lectric": 26539,
+      "ĠTween": 26540,
+      "ĠHughes": 26541,
+      "Ġmartens": 26542,
+      "Ġnematic": 26543,
+      "Ġexperimentation": 26544,
+      "ĠParker": 26545,
+      "Ġepisodic": 26546,
+      "Ġtelem": 26547,
+      "ADE": 26548,
+      "columns": 26549,
+      "Ġfundamentally": 26550,
+      "enet": 26551,
+      "ĠVl": 26552,
+      "earth": 26553,
+      "Ġquantile": 26554,
+      "ĠReplication": 26555,
+      "Ġcleared": 26556,
+      "Energy": 26557,
+      "Smith": 26558,
+      "Ġantidepressant": 26559,
+      "mx": 26560,
+      "pmod": 26561,
+      "amid": 26562,
+      "Ġserotype": 26563,
+      "Ġundergraduate": 26564,
+      "ĠArizona": 26565,
+      "Ġpushed": 26566,
+      "ulu": 26567,
+      "ĠNIC": 26568,
+      "Ġrheological": 26569,
+      "omegal": 26570,
+      "ĠQing": 26571,
+      "orch": 26572,
+      "irmed": 26573,
+      "ĠQuery": 26574,
+      "Ġsandwich": 26575,
+      "Ġclinician": 26576,
+      "ĠElliptic": 26577,
+      "ĠMeh": 26578,
+      "DEV": 26579,
+      "ĠDetermining": 26580,
+      "alcogen": 26581,
+      "bench": 26582,
+      "azep": 26583,
+      "ĠMississ": 26584,
+      "tizing": 26585,
+      "ĠRBC": 26586,
+      "Ġofficials": 26587,
+      "Tag": 26588,
+      "kT": 26589,
+      "luence": 26590,
+      "ĠRoom": 26591,
+      "Ġlectin": 26592,
+      "bara": 26593,
+      "kyl": 26594,
+      "OND": 26595,
+      "ĠDose": 26596,
+      "Ġprism": 26597,
+      "Ġreductive": 26598,
+      "ĠSpectroscopic": 26599,
+      "odied": 26600,
+      "colone": 26601,
+      "ĠCONFIG": 26602,
+      "Ġbrittle": 26603,
+      "inverse": 26604,
+      "ĠBuff": 26605,
+      "ytocin": 26606,
+      "Ġformations": 26607,
+      "ĠConventional": 26608,
+      "prev": 26609,
+      "Ġferrite": 26610,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 26611,
+      "Ġadopts": 26612,
+      "ĠMiocene": 26613,
+      "management": 26614,
+      "ĠCRF": 26615,
+      "ĠHelm": 26616,
+      "Ġdoubled": 26617,
+      "ĠEFFECT": 26618,
+      "Ġdance": 26619,
+      "structions": 26620,
+      "rait": 26621,
+      "ifers": 26622,
+      "ellip": 26623,
+      "utting": 26624,
+      "prof": 26625,
+      "ĠQin": 26626,
+      "Ġabsc": 26627,
+      "Ġexploits": 26628,
+      "Ġcyber": 26629,
+      "definition": 26630,
+      "ĠCoronary": 26631,
+      "Ġdeterg": 26632,
+      "ĠPerception": 26633,
+      "ĠCurves": 26634,
+      "Ġnematodes": 26635,
+      "Ġlistening": 26636,
+      "Ġcatalase": 26637,
+      "Coll": 26638,
+      "ré": 26639,
+      "islative": 26640,
+      "Ġarriving": 26641,
+      "Ġviolating": 26642,
+      "д": 26643,
+      "hetics": 26644,
+      "ĠJar": 26645,
+      "concept": 26646,
+      "Ġbrush": 26647,
+      "immunity": 26648,
+      "Ġfingerprint": 26649,
+      "resid": 26650,
+      "Ġelevations": 26651,
+      "ockets": 26652,
+      "Ġcatechol": 26653,
+      "иÑ": 26654,
+      "Ġprecipitates": 26655,
+      "Ġsoccer": 26656,
+      "insulin": 26657,
+      "Ġpursue": 26658,
+      "ĠICA": 26659,
+      "ĠPolice": 26660,
+      "ĠMurphy": 26661,
+      "Task": 26662,
+      "ĠCoc": 26663,
+      "ĠHabit": 26664,
+      "ĠKP": 26665,
+      "Ġfloral": 26666,
+      "Ġhun": 26667,
+      "Ġhydrogenation": 26668,
+      "Ġspong": 26669,
+      "Ġchimeric": 26670,
+      "ĠKoch": 26671,
+      "gon": 26672,
+      "ĠSchur": 26673,
+      "ĠGreater": 26674,
+      "RX": 26675,
+      "Ġcing": 26676,
+      "ĠWaltham": 26677,
+      "angling": 26678,
+      "Ġcounties": 26679,
+      "Ġlamina": 26680,
+      "Ġcouncil": 26681,
+      "sort": 26682,
+      "ĠBarc": 26683,
+      "ĠDow": 26684,
+      "ĠZeng": 26685,
+      "Ġdevised": 26686,
+      "uitable": 26687,
+      "Ġmethylene": 26688,
+      "Ġsuperiority": 26689,
+      "Ġepidermis": 26690,
+      "Ġprag": 26691,
+      "ĠPED": 26692,
+      "threatening": 26693,
+      "ishi": 26694,
+      "Ġepsilon": 26695,
+      "address": 26696,
+      "ENTAL": 26697,
+      "ĠBle": 26698,
+      "ĠAntonio": 26699,
+      "oother": 26700,
+      "ĠAgar": 26701,
+      "Ġneighborhoods": 26702,
+      "Ġshortened": 26703,
+      "STATE": 26704,
+      "ĠSerial": 26705,
+      "MAR": 26706,
+      "OU": 26707,
+      "Ġencapsulation": 26708,
+      "ĠConsortium": 26709,
+      "Dr": 26710,
+      "profile": 26711,
+      "Ġemitter": 26712,
+      "Ġnecrotic": 26713,
+      "ĠAutonomous": 26714,
+      "ĠPhosphorylation": 26715,
+      "minim": 26716,
+      "anthin": 26717,
+      "ĠSph": 26718,
+      "ĠGur": 26719,
+      "dihydroxy": 26720,
+      "distributed": 26721,
+      "ĠRPMI": 26722,
+      "stones": 26723,
+      "Ġhyperfine": 26724,
+      "Ġislet": 26725,
+      "ĠSlo": 26726,
+      "pletely": 26727,
+      "Ġinactivated": 26728,
+      "ĠAgriculture": 26729,
+      "Ġtremend": 26730,
+      "Ġeveryone": 26731,
+      "omponent": 26732,
+      "ZnO": 26733,
+      "MPI": 26734,
+      "ĠDiamond": 26735,
+      "Ġ⣨": 26736,
+      "Cost": 26737,
+      "Ġdisabilities": 26738,
+      "inver": 26739,
+      "ĠCensus": 26740,
+      "echo": 26741,
+      "Ġvegetative": 26742,
+      "Ġwillingness": 26743,
+      "Ġrecap": 26744,
+      "ĠConstraint": 26745,
+      "ĠPatrick": 26746,
+      "Ġovert": 26747,
+      "Ġmoieties": 26748,
+      "orax": 26749,
+      "ippi": 26750,
+      "Direct": 26751,
+      "Ġcaries": 26752,
+      "Ġlocalities": 26753,
+      "lattices": 26754,
+      "ĠExploration": 26755,
+      "ĠAW": 26756,
+      "Ġlocking": 26757,
+      "Ġcoincident": 26758,
+      "Ġmultimedia": 26759,
+      "Ġtemporarily": 26760,
+      "ĠCaus": 26761,
+      "encia": 26762,
+      "Ġweathering": 26763,
+      "ĠHelicobacter": 26764,
+      "ĠThings": 26765,
+      "hips": 26766,
+      "moving": 26767,
+      "Ġsigmoid": 26768,
+      "isin": 26769,
+      "ĠBec": 26770,
+      "Ġmicrograms": 26771,
+      "bounds": 26772,
+      "ĠColumn": 26773,
+      "Ġcommuting": 26774,
+      "ĠJen": 26775,
+      "Ġhourly": 26776,
+      "MSC": 26777,
+      "Ġattendance": 26778,
+      "ĠâIJ£": 26779,
+      "ĠEO": 26780,
+      "prog": 26781,
+      "Ġrapamycin": 26782,
+      "ĠPredictors": 26783,
+      "ĠRetrieved": 26784,
+      "Ġsubspecies": 26785,
+      "Ġderives": 26786,
+      "ĠĤ": 26787,
+      "ĠGenerating": 26788,
+      "anners": 26789,
+      "Ġvolat": 26790,
+      "Ġvisiting": 26791,
+      "ĠCalculations": 26792,
+      "ña": 26793,
+      "Ġdesert": 26794,
+      "Ġexpectancy": 26795,
+      "BMCs": 26796,
+      "ĠExplo": 26797,
+      "Ġtravelling": 26798,
+      "icum": 26799,
+      "Ġsubdivision": 26800,
+      "Ġcrosslinking": 26801,
+      "benzoth": 26802,
+      "ĠTon": 26803,
+      "REN": 26804,
+      "Ġleth": 26805,
+      "rabbit": 26806,
+      "ĠAbove": 26807,
+      "ulted": 26808,
+      "Ġconstric": 26809,
+      "Jones": 26810,
+      "zhou": 26811,
+      "vern": 26812,
+      "ĠLady": 26813,
+      "ĠBuffer": 26814,
+      "ĠControlling": 26815,
+      "Ġmultiscale": 26816,
+      "nikov": 26817,
+      "acycl": 26818,
+      "Ġprosthesis": 26819,
+      "Af": 26820,
+      "ĠCorps": 26821,
+      "structed": 26822,
+      "Grid": 26823,
+      "inning": 26824,
+      "olding": 26825,
+      "Ġthiol": 26826,
+      "ikov": 26827,
+      "âĢ¢âĢ¢âĢ¢": 26828,
+      "Ġgovernments": 26829,
+      "rapping": 26830,
+      "Ġthrombocyt": 26831,
+      "Leg": 26832,
+      "RY": 26833,
+      "ĠIceland": 26834,
+      "ocycle": 26835,
+      "ĠMemorial": 26836,
+      "got": 26837,
+      "Ġidem": 26838,
+      "ĠBuild": 26839,
+      "olipoprotein": 26840,
+      "DV": 26841,
+      "Ġphthal": 26842,
+      "richment": 26843,
+      "ĠHaem": 26844,
+      "Ġanswering": 26845,
+      "ĠIJ": 26846,
+      "Ġtransgene": 26847,
+      "Ġrenamed": 26848,
+      "ĠImageJ": 26849,
+      "Ġcassette": 26850,
+      "Ġcoalescence": 26851,
+      "Ġcompaction": 26852,
+      "Ġwildlife": 26853,
+      "Ġwins": 26854,
+      "Ġsupernovae": 26855,
+      "enteric": 26856,
+      "isphere": 26857,
+      "Ġtracker": 26858,
+      "Ġevidences": 26859,
+      "Ġcomorbidity": 26860,
+      "ĠRules": 26861,
+      "phasing": 26862,
+      "ĠLangevin": 26863,
+      "ĠFit": 26864,
+      "Ġpsychiat": 26865,
+      "Ġbreakthrough": 26866,
+      "Ġcholinergic": 26867,
+      "ĠMetall": 26868,
+      "breeding": 26869,
+      "itinib": 26870,
+      "Ġsolo": 26871,
+      "abling": 26872,
+      "elief": 26873,
+      "oscill": 26874,
+      "rev": 26875,
+      "arya": 26876,
+      "Ġgoodness": 26877,
+      "ĠPBE": 26878,
+      "Ġawards": 26879,
+      "Ġcrani": 26880,
+      "Ġphotograp": 26881,
+      "arents": 26882,
+      "Ġfixes": 26883,
+      "rÃŃ": 26884,
+      "assuming": 26885,
+      "Ġcongruent": 26886,
+      "ĠMother": 26887,
+      "ĠNap": 26888,
+      "ĠProc": 26889,
+      "Ġcategorization": 26890,
+      "inch": 26891,
+      "ĠHorm": 26892,
+      "ĠInterventions": 26893,
+      "Ġnonequilibrium": 26894,
+      "Ġencrypted": 26895,
+      "primary": 26896,
+      "iens": 26897,
+      "lac": 26898,
+      "rams": 26899,
+      "Ġboards": 26900,
+      "ĠHell": 26901,
+      "charged": 26902,
+      "Ġperioperative": 26903,
+      "emp": 26904,
+      "ĠInvolvement": 26905,
+      "Russ": 26906,
+      "univers": 26907,
+      "ĠDJ": 26908,
+      "Ġdisagreement": 26909,
+      "Ġpert": 26910,
+      "Ġstroma": 26911,
+      "Ġcalcite": 26912,
+      "Ġrotary": 26913,
+      "Ġmethyltransferase": 26914,
+      "Ġancestry": 26915,
+      "ĠWitten": 26916,
+      "CRC": 26917,
+      "uretic": 26918,
+      "ophyta": 26919,
+      "provided": 26920,
+      "Ġcorrespondingly": 26921,
+      "bigcap": 26922,
+      "ĠAgilent": 26923,
+      "ë": 26924,
+      "rooms": 26925,
+      "Ġdisent": 26926,
+      "Ġdilutions": 26927,
+      "ĠMyel": 26928,
+      "Ġquasar": 26929,
+      "Ġtilted": 26930,
+      "Ġinternalization": 26931,
+      "ĠPrivate": 26932,
+      "ĠFriedman": 26933,
+      "Ġseventh": 26934,
+      "ĠClosed": 26935,
+      "CTC": 26936,
+      "gren": 26937,
+      "ĠColombia": 26938,
+      "odel": 26939,
+      "Ġpolitics": 26940,
+      "ĠMSSM": 26941,
+      "Ġmate": 26942,
+      "Ġcommod": 26943,
+      "ĠRus": 26944,
+      "Ġanesthetized": 26945,
+      "together": 26946,
+      "ĠBCS": 26947,
+      "ewski": 26948,
+      "romagnet": 26949,
+      "ĠCun": 26950,
+      "Ġcurative": 26951,
+      "Ġimputation": 26952,
+      "Ġcarbide": 26953,
+      "DFT": 26954,
+      "nsic": 26955,
+      "bee": 26956,
+      "Ġsplen": 26957,
+      "ĠMaryland": 26958,
+      "Ġoligonucleotide": 26959,
+      "ĠVeget": 26960,
+      "buffered": 26961,
+      "National": 26962,
+      "letic": 26963,
+      "ĠSyl": 26964,
+      "Ġseab": 26965,
+      "ardial": 26966,
+      "Ġportray": 26967,
+      "Ġaberrations": 26968,
+      "Ġstorms": 26969,
+      "ĠShan": 26970,
+      "ĠGenBank": 26971,
+      "issa": 26972,
+      "Ġcet": 26973,
+      "Ġbench": 26974,
+      "ĠRecommendations": 26975,
+      "Ġtriples": 26976,
+      "Ġïĥ¥": 26977,
+      "ĠNeuros": 26978,
+      "Ġdiscom": 26979,
+      "season": 26980,
+      "ĠExec": 26981,
+      "changing": 26982,
+      "Ġarrives": 26983,
+      "Hash": 26984,
+      "mRNA": 26985,
+      "Ġfric": 26986,
+      "asa": 26987,
+      "obia": 26988,
+      "Ġpostsynaptic": 26989,
+      "optimizer": 26990,
+      "ĠClouds": 26991,
+      "Ġhypersensitivity": 26992,
+      "vacc": 26993,
+      "ĠSig": 26994,
+      "philic": 26995,
+      "Ġgrounded": 26996,
+      "ĠWan": 26997,
+      "ĠCalabi": 26998,
+      "ĠMachines": 26999,
+      "Ġaxisymmetric": 27000,
+      "ĠSteve": 27001,
+      "Ġpulled": 27002,
+      "ĠExcel": 27003,
+      "Ġdiamonds": 27004,
+      "KR": 27005,
+      "West": 27006,
+      "ĠDest": 27007,
+      "Ġannular": 27008,
+      "Ġarchive": 27009,
+      "Ġparenchyma": 27010,
+      "ĠEH": 27011,
+      "ópez": 27012,
+      "Ġunpublished": 27013,
+      "Ġsoutheastern": 27014,
+      "Ġnests": 27015,
+      "dimensions": 27016,
+      "latitude": 27017,
+      "Orig": 27018,
+      "eced": 27019,
+      "ĠDraw": 27020,
+      "redshift": 27021,
+      "Ġamyl": 27022,
+      "omyelitis": 27023,
+      "Why": 27024,
+      "caro": 27025,
+      "iq": 27026,
+      "assess": 27027,
+      "ĠContin": 27028,
+      "Ġchirality": 27029,
+      "matical": 27030,
+      "Ġchaperone": 27031,
+      "Ġendometriosis": 27032,
+      "relu": 27033,
+      "Ġconverged": 27034,
+      "broad": 27035,
+      "ĠIterative": 27036,
+      "Ġvasculature": 27037,
+      "fund": 27038,
+      "ĠFly": 27039,
+      "Ġantigenic": 27040,
+      "Ġmeningitis": 27041,
+      "Ġentails": 27042,
+      "horn": 27043,
+      "Ġlocomotor": 27044,
+      "izard": 27045,
+      "Ġuneven": 27046,
+      "parity": 27047,
+      "packet": 27048,
+      "tubulin": 27049,
+      "Ġsewage": 27050,
+      "Ġdecentralized": 27051,
+      "Ġgrafted": 27052,
+      "Ġsep": 27053,
+      "ĠExtensive": 27054,
+      "Ġspline": 27055,
+      "quer": 27056,
+      "archit": 27057,
+      "Ġprimate": 27058,
+      "Ġïģ±": 27059,
+      "pyrimidin": 27060,
+      "ĠSAP": 27061,
+      "Ġunderlie": 27062,
+      "Ġanalyzes": 27063,
+      "ĠCCA": 27064,
+      "recogn": 27065,
+      "IPT": 27066,
+      "Different": 27067,
+      "ĠTEST": 27068,
+      "Ġunfavorable": 27069,
+      "edic": 27070,
+      "ĠAbnormal": 27071,
+      "pyrimidine": 27072,
+      "urine": 27073,
+      "embedded": 27074,
+      "varies": 27075,
+      "otropin": 27076,
+      "Ġsemen": 27077,
+      "Ġtransmittance": 27078,
+      "Ġabras": 27079,
+      "Ġó¸Ģł": 27080,
+      "Ġtriglyceride": 27081,
+      "bundle": 27082,
+      "ĠYb": 27083,
+      "ĠCarr": 27084,
+      "Ġnaming": 27085,
+      "Weight": 27086,
+      "Ġcondensates": 27087,
+      "Ġnos": 27088,
+      "amard": 27089,
+      "vertices": 27090,
+      "ELS": 27091,
+      "idone": 27092,
+      "Ġcontest": 27093,
+      "Ġheading": 27094,
+      "ĠGalerkin": 27095,
+      "GV": 27096,
+      "ĠGli": 27097,
+      "Ġfermented": 27098,
+      "Ġbilingual": 27099,
+      "Ġticks": 27100,
+      "Ġkary": 27101,
+      "ragal": 27102,
+      "ĠAber": 27103,
+      "ĠYouTube": 27104,
+      "UCTURE": 27105,
+      "branch": 27106,
+      "ر": 27107,
+      "ĠFH": 27108,
+      "onoi": 27109,
+      "imotor": 27110,
+      "Ġverifying": 27111,
+      "ĠConceptual": 27112,
+      "ĠDeterminants": 27113,
+      "urm": 27114,
+      "uronic": 27115,
+      "ĠKau": 27116,
+      "ĠConformal": 27117,
+      "Ġdropping": 27118,
+      "ĠFlows": 27119,
+      "gluon": 27120,
+      "again": 27121,
+      "ĠMRSA": 27122,
+      "warf": 27123,
+      "Ġemphasizes": 27124,
+      "Entry": 27125,
+      "ĠASP": 27126,
+      "resol": 27127,
+      "ventricular": 27128,
+      "ĠâĨĶ": 27129,
+      "Ġoverexpressing": 27130,
+      "omegalovirus": 27131,
+      "inoc": 27132,
+      "SCO": 27133,
+      "ĠPARP": 27134,
+      "ĠSchul": 27135,
+      "ĠCamb": 27136,
+      "ĠPod": 27137,
+      "ĠPun": 27138,
+      "ĠCompetition": 27139,
+      "ĠGATA": 27140,
+      "Ġmoon": 27141,
+      "Ġputs": 27142,
+      "angiogenic": 27143,
+      "ĠRepublican": 27144,
+      "ĠUbiqu": 27145,
+      "eys": 27146,
+      "ĠGong": 27147,
+      "arger": 27148,
+      "ĠIntermediate": 27149,
+      "Ġinterpolated": 27150,
+      "Ġenlargement": 27151,
+      "Ġinstruct": 27152,
+      "Ġrc": 27153,
+      "dioxo": 27154,
+      "eye": 27155,
+      "ĠCarls": 27156,
+      "ĠMeasured": 27157,
+      "ircles": 27158,
+      "ĠRaf": 27159,
+      "Ġarb": 27160,
+      "examples": 27161,
+      "Mi": 27162,
+      "ĠStern": 27163,
+      "ĠFK": 27164,
+      "Ġmillisecond": 27165,
+      "ĠIRF": 27166,
+      "ĠEpithelial": 27167,
+      "edicine": 27168,
+      "eles": 27169,
+      "sig": 27170,
+      "âĪĢ": 27171,
+      "ĠWiener": 27172,
+      "bauer": 27173,
+      "ouses": 27174,
+      "Ġcoloured": 27175,
+      "ĠIncrease": 27176,
+      "Ġtriglycerides": 27177,
+      "Ġaegypti": 27178,
+      "ĠNumerous": 27179,
+      "Ġretardation": 27180,
+      "Ġintercellular": 27181,
+      "ĠKlebsiella": 27182,
+      "ĠDra": 27183,
+      "ĠDIC": 27184,
+      "ĠThreshold": 27185,
+      "rainment": 27186,
+      "Ġreproducing": 27187,
+      "Ġulcers": 27188,
+      "Ġarousal": 27189,
+      "ĠHills": 27190,
+      "Ġcalves": 27191,
+      "ĠReservoir": 27192,
+      "ĠRadar": 27193,
+      "Ġpsychosis": 27194,
+      "ĠFORM": 27195,
+      "duration": 27196,
+      "ĠAcademic": 27197,
+      "catal": 27198,
+      "olla": 27199,
+      "olol": 27200,
+      "ĠCron": 27201,
+      "iko": 27202,
+      "Ġextremes": 27203,
+      "ĠTrypan": 27204,
+      "Ġbip": 27205,
+      "Ġalginate": 27206,
+      "ĠHoch": 27207,
+      "ĠBennett": 27208,
+      "ĠHippocamp": 27209,
+      "ĠGeological": 27210,
+      "Nevertheless": 27211,
+      "ĠHes": 27212,
+      "ĠAdding": 27213,
+      "Ġexternally": 27214,
+      "Ġslag": 27215,
+      "Ġteach": 27216,
+      "ĠStanley": 27217,
+      "controller": 27218,
+      "ĠUnits": 27219,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 27220,
+      "Ġaerodynamic": 27221,
+      "ovalent": 27222,
+      "cube": 27223,
+      "ÅŁ": 27224,
+      "require": 27225,
+      "romolecules": 27226,
+      "irteen": 27227,
+      "Ġclauses": 27228,
+      "Ġdefeat": 27229,
+      "policy": 27230,
+      "Ġfaithful": 27231,
+      "Ġpq": 27232,
+      "ĠTanaka": 27233,
+      "ĠEver": 27234,
+      "Ġunpredict": 27235,
+      "auty": 27236,
+      "ĠGALAX": 27237,
+      "Ġtide": 27238,
+      "ĠFiltering": 27239,
+      "Ġeuthan": 27240,
+      "merce": 27241,
+      "DEX": 27242,
+      "Ġnesting": 27243,
+      "DN": 27244,
+      "IRT": 27245,
+      "ĠThr": 27246,
+      "tissue": 27247,
+      "Ġpalae": 27248,
+      "Ï©": 27249,
+      "Ġdilated": 27250,
+      "Ġpinning": 27251,
+      "Rb": 27252,
+      "ĠSap": 27253,
+      "ragonal": 27254,
+      "ĠSPR": 27255,
+      "ĠDial": 27256,
+      "Ġacupuncture": 27257,
+      "diameter": 27258,
+      "ĠPCB": 27259,
+      "Parameters": 27260,
+      "ĠProfiles": 27261,
+      "transfected": 27262,
+      "liter": 27263,
+      "ĠRights": 27264,
+      "Ġcontributor": 27265,
+      "ĠCorrel": 27266,
+      "Ġregressions": 27267,
+      "Ġsegmental": 27268,
+      "Shape": 27269,
+      "IAN": 27270,
+      "ecom": 27271,
+      "comings": 27272,
+      "Ġhemorrhagic": 27273,
+      "opos": 27274,
+      "Ġrefraction": 27275,
+      "PFC": 27276,
+      "proj": 27277,
+      "ovo": 27278,
+      "ĠDerived": 27279,
+      "Ġundirected": 27280,
+      "Ġlos": 27281,
+      "Ġengaging": 27282,
+      "cans": 27283,
+      "Ġdestructive": 27284,
+      "Pop": 27285,
+      "Ġmakers": 27286,
+      "ĠWor": 27287,
+      "ĠAreas": 27288,
+      "vasion": 27289,
+      "Ġparaformaldehyde": 27290,
+      "abinoid": 27291,
+      "cpy": 27292,
+      "proxim": 27293,
+      "Ġenamel": 27294,
+      "Ġpaediatric": 27295,
+      "ĠChildhood": 27296,
+      "Ġpectin": 27297,
+      "ofilm": 27298,
+      "Ġcarboxylic": 27299,
+      "Ġausten": 27300,
+      "Ġunequal": 27301,
+      "ĠCountry": 27302,
+      "Ġiterated": 27303,
+      "Ġflanking": 27304,
+      "Ġtraction": 27305,
+      "anson": 27306,
+      "iscus": 27307,
+      "ĠDavies": 27308,
+      "raham": 27309,
+      "terozoic": 27310,
+      "ĠBrass": 27311,
+      "Oc": 27312,
+      "Ġunification": 27313,
+      "meter": 27314,
+      "ĠNeon": 27315,
+      "building": 27316,
+      "icting": 27317,
+      "Ġjustification": 27318,
+      "Prior": 27319,
+      "Ġfirms": 27320,
+      "Ġeducated": 27321,
+      "Ġintersecting": 27322,
+      "Ġboosting": 27323,
+      "Pass": 27324,
+      "member": 27325,
+      "contains": 27326,
+      "rano": 27327,
+      "relax": 27328,
+      "ĠCollaborative": 27329,
+      "Ġpx": 27330,
+      "Ġseeding": 27331,
+      "cripts": 27332,
+      "inez": 27333,
+      "omeres": 27334,
+      "Ġsiblings": 27335,
+      "anging": 27336,
+      "fert": 27337,
+      "Ġrecovering": 27338,
+      "pure": 27339,
+      "Ġsd": 27340,
+      "ĠVul": 27341,
+      "pedance": 27342,
+      "Ġfighting": 27343,
+      "Super": 27344,
+      "ĠIto": 27345,
+      "Ġperimeter": 27346,
+      "ĠInhibitors": 27347,
+      "electrode": 27348,
+      "enabled": 27349,
+      "fb": 27350,
+      "ĠPCs": 27351,
+      "Ġnausea": 27352,
+      "ĠConversion": 27353,
+      "Ġsla": 27354,
+      "Ġinvertebrates": 27355,
+      "ĠBrian": 27356,
+      "Ġcontiguous": 27357,
+      "ĠACKNOWLEDGM": 27358,
+      "urface": 27359,
+      "Ġcoars": 27360,
+      "ĠLeh": 27361,
+      "ĠCompression": 27362,
+      "cycles": 27363,
+      "Ġsinh": 27364,
+      "ĠOccup": 27365,
+      "strength": 27366,
+      "Ġconstr": 27367,
+      "Ġpesticide": 27368,
+      "Ġbisp": 27369,
+      "ĠTn": 27370,
+      "Ġparentheses": 27371,
+      "degrad": 27372,
+      "Ġhyperglycemia": 27373,
+      "PW": 27374,
+      "kj": 27375,
+      "ecological": 27376,
+      "Ġthy": 27377,
+      "Ġeleg": 27378,
+      "ĠSynaptic": 27379,
+      "scaled": 27380,
+      "tity": 27381,
+      "Ġequity": 27382,
+      "Ġblockchain": 27383,
+      "ĠLithium": 27384,
+      "Ġspark": 27385,
+      "Ġentitled": 27386,
+      "Ġconventions": 27387,
+      "Argument": 27388,
+      "Ġretail": 27389,
+      "Ġneoplastic": 27390,
+      "Ġdamped": 27391,
+      "ĠSurveillance": 27392,
+      "ĠAnna": 27393,
+      "Ġspacetimes": 27394,
+      "inges": 27395,
+      "ahashi": 27396,
+      "ĠInfections": 27397,
+      "Ġneglecting": 27398,
+      "Ġevaporated": 27399,
+      "vastatin": 27400,
+      "Ġgh": 27401,
+      "ĠNLP": 27402,
+      "Ġphones": 27403,
+      "Ġlifted": 27404,
+      "Ġdivisible": 27405,
+      "Ġdurability": 27406,
+      "osited": 27407,
+      "Ġexcitability": 27408,
+      "Ġbuoyancy": 27409,
+      "Ġuncontrolled": 27410,
+      "bran": 27411,
+      "ĠPhe": 27412,
+      "Ġimmunocomp": 27413,
+      "Ġeventual": 27414,
+      "Ġclassroom": 27415,
+      "Ġmicrographs": 27416,
+      "Ġrecharge": 27417,
+      "ettes": 27418,
+      "ĠDiver": 27419,
+      "ĠDall": 27420,
+      "Ġmetac": 27421,
+      "Ġneuroendocrine": 27422,
+      "topology": 27423,
+      "ĠHawking": 27424,
+      "omson": 27425,
+      "ĠHarry": 27426,
+      "mouth": 27427,
+      "Ġdeciding": 27428,
+      "Ġuncovered": 27429,
+      "Ġgolden": 27430,
+      "ĠCastle": 27431,
+      "Ġfiducial": 27432,
+      "Aware": 27433,
+      "ĠGan": 27434,
+      "erahertz": 27435,
+      "ĠSaturn": 27436,
+      "LN": 27437,
+      "Unit": 27438,
+      "ĥĹ": 27439,
+      "Ġbinder": 27440,
+      "INFO": 27441,
+      "ĠTemper": 27442,
+      "ipel": 27443,
+      "Ġnumerator": 27444,
+      "Ġwebsites": 27445,
+      "Ġthreatened": 27446,
+      "Ġremnants": 27447,
+      "ĠFinnish": 27448,
+      "hof": 27449,
+      "media": 27450,
+      "concentration": 27451,
+      "ĠReed": 27452,
+      "ĠLeishmania": 27453,
+      "Ġmultifunctional": 27454,
+      "racy": 27455,
+      "Ġdistribute": 27456,
+      "ĠDecay": 27457,
+      "Ġgrinding": 27458,
+      "Loss": 27459,
+      "MPL": 27460,
+      "ĠLakes": 27461,
+      "ĠQR": 27462,
+      "ĠStructured": 27463,
+      "ĠMalaria": 27464,
+      "Ġflavonoid": 27465,
+      "Ġtowns": 27466,
+      "opia": 27467,
+      "ĠVec": 27468,
+      "othy": 27469,
+      "Ġsingles": 27470,
+      "Ġpenetrate": 27471,
+      "ĠPig": 27472,
+      "ieved": 27473,
+      "Ġderivations": 27474,
+      "Ġdiscomfort": 27475,
+      "afenib": 27476,
+      "ĠLegendre": 27477,
+      "ĠPax": 27478,
+      "ĠMX": 27479,
+      "ĠExtrem": 27480,
+      "ĠForeign": 27481,
+      "ĠCourse": 27482,
+      "ĠHit": 27483,
+      "vage": 27484,
+      "Ġclique": 27485,
+      "Ġcompensatory": 27486,
+      "User": 27487,
+      "Ġdraws": 27488,
+      "ĠProtective": 27489,
+      "Ġallocate": 27490,
+      "ĠPant": 27491,
+      "Ġdash": 27492,
+      "Ġparal": 27493,
+      "ĠCirculating": 27494,
+      "ĠHistone": 27495,
+      "ĠÅ«": 27496,
+      "Ġprojec": 27497,
+      "ĠAAA": 27498,
+      "ĠPMS": 27499,
+      "glacial": 27500,
+      "ĠMeeting": 27501,
+      "ĠAntibiotic": 27502,
+      "ategorical": 27503,
+      "Ġattenuate": 27504,
+      "Power": 27505,
+      "owicz": 27506,
+      "ĠDefault": 27507,
+      "Ġmarsh": 27508,
+      "plasm": 27509,
+      "ĠPathology": 27510,
+      "ĠEf": 27511,
+      "Lys": 27512,
+      "flies": 27513,
+      "Ġinterviewed": 27514,
+      "ĠQA": 27515,
+      "Ġimpuls": 27516,
+      "Ġpapillary": 27517,
+      "dR": 27518,
+      "uh": 27519,
+      "ĠJing": 27520,
+      "Ġrescaled": 27521,
+      "efficiency": 27522,
+      "Ġef": 27523,
+      "ĠEisen": 27524,
+      "Ġattacked": 27525,
+      "Ġopto": 27526,
+      "Ġspeculated": 27527,
+      "haz": 27528,
+      "Ġideally": 27529,
+      "ymenoptera": 27530,
+      "Ġlr": 27531,
+      "ĠIz": 27532,
+      "resource": 27533,
+      "ĠFacility": 27534,
+      "ĠAcquisition": 27535,
+      "Ġpostural": 27536,
+      "autiful": 27537,
+      "Ġgingival": 27538,
+      "Ġpertaining": 27539,
+      "ĠExtra": 27540,
+      "ĠProgramme": 27541,
+      "hesus": 27542,
+      "fermion": 27543,
+      "Ġsteadily": 27544,
+      "Ġterminus": 27545,
+      "Parser": 27546,
+      "ĠInclusion": 27547,
+      "ĠWuhan": 27548,
+      "Ġrepetitions": 27549,
+      "done": 27550,
+      "ĠCep": 27551,
+      "Ġunstructured": 27552,
+      "ĠCollectively": 27553,
+      "Ġsettling": 27554,
+      "Ġjaw": 27555,
+      "ĠUni": 27556,
+      "Ġrestoring": 27557,
+      "urtles": 27558,
+      "Full": 27559,
+      "Ġdynamo": 27560,
+      "IGO": 27561,
+      "ĠBAT": 27562,
+      "ová": 27563,
+      "venues": 27564,
+      "ĠPerhaps": 27565,
+      "sensing": 27566,
+      "ĠIschem": 27567,
+      "odemographic": 27568,
+      "Ss": 27569,
+      "ĠLund": 27570,
+      "Ġelite": 27571,
+      "protocol": 27572,
+      "ĠChristopher": 27573,
+      "basic": 27574,
+      "Ġpuber": 27575,
+      "Ġmagnetism": 27576,
+      "vars": 27577,
+      "inducing": 27578,
+      "Ġdated": 27579,
+      "Ġenemy": 27580,
+      "ĠStop": 27581,
+      "social": 27582,
+      "ĠdÏĦ": 27583,
+      "ĠBun": 27584,
+      "Small": 27585,
+      "purpose": 27586,
+      "Ġhunting": 27587,
+      "CPU": 27588,
+      "ĠJunior": 27589,
+      "REL": 27590,
+      "Ġcontractile": 27591,
+      "Ġsilicone": 27592,
+      "adrenergic": 27593,
+      "bz": 27594,
+      "Ġfus": 27595,
+      "ifted": 27596,
+      "sep": 27597,
+      "âĪĴâĪŀ": 27598,
+      "Ġdrum": 27599,
+      "----------": 27600,
+      "ĠTregs": 27601,
+      "itarian": 27602,
+      "century": 27603,
+      "âĬ¥": 27604,
+      "Numer": 27605,
+      "ĠBenz": 27606,
+      "Ġcommunicating": 27607,
+      "Ġpaternal": 27608,
+      "ĠFGFR": 27609,
+      "ĠâĤ¬": 27610,
+      "Ġdeviate": 27611,
+      "fre": 27612,
+      "Ġmolten": 27613,
+      "Ġstandardization": 27614,
+      "Ġfunctionalities": 27615,
+      "ĠPaulo": 27616,
+      "Ġbucket": 27617,
+      "ĠConcentrations": 27618,
+      "ĠKum": 27619,
+      "Ġmimicking": 27620,
+      "Drop": 27621,
+      "zoa": 27622,
+      "ĠNuclei": 27623,
+      "brack": 27624,
+      "ecolor": 27625,
+      "Ġcarn": 27626,
+      "Ġveterinary": 27627,
+      "Ġchemotherapeutic": 27628,
+      "Ġferment": 27629,
+      "lasting": 27630,
+      "ĠRogers": 27631,
+      "ieri": 27632,
+      "Ġconverters": 27633,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 27634,
+      "ĠRepair": 27635,
+      "Europe": 27636,
+      "TIME": 27637,
+      "Ġties": 27638,
+      "ĠPIN": 27639,
+      "Ġtribut": 27640,
+      "Ġhomogenization": 27641,
+      "excitation": 27642,
+      "atization": 27643,
+      "ĠRash": 27644,
+      "Ġprecession": 27645,
+      "ás": 27646,
+      "Ġspiking": 27647,
+      "ĠGrassmann": 27648,
+      "minister": 27649,
+      "Ġfactorial": 27650,
+      "ĠDeut": 27651,
+      "sampled": 27652,
+      "Ġeukaryotes": 27653,
+      "overlapping": 27654,
+      "agglut": 27655,
+      "Ġprescribing": 27656,
+      "Ġcro": 27657,
+      "omechanical": 27658,
+      "iza": 27659,
+      "ĠManufact": 27660,
+      "native": 27661,
+      "ursive": 27662,
+      "ĠIssues": 27663,
+      "Ġstreptomycin": 27664,
+      "endi": 27665,
+      "ĠSpr": 27666,
+      "ceq": 27667,
+      "arginine": 27668,
+      "ixon": 27669,
+      "ĠFoundations": 27670,
+      "Single": 27671,
+      "Ġoxal": 27672,
+      "Ġhydrate": 27673,
+      "Iterator": 27674,
+      "kii": 27675,
+      "aminated": 27676,
+      "Ġsprings": 27677,
+      "oln": 27678,
+      "ĠSetup": 27679,
+      "Ġripening": 27680,
+      "Ġtheoretic": 27681,
+      "Ġcfg": 27682,
+      "μL": 27683,
+      "Gordon": 27684,
+      "SK": 27685,
+      "Ġnations": 27686,
+      "Query": 27687,
+      "Ùħ": 27688,
+      "Ġfores": 27689,
+      "requencies": 27690,
+      "ĠPharmaceutical": 27691,
+      "ĠAllocation": 27692,
+      "otypical": 27693,
+      "ĠPilot": 27694,
+      "thora": 27695,
+      "ĠVand": 27696,
+      "Ġsyringe": 27697,
+      "ĠRAP": 27698,
+      "rometric": 27699,
+      "Ġïģ´": 27700,
+      "Ġcitations": 27701,
+      "would": 27702,
+      "Ġnortheastern": 27703,
+      "comparison": 27704,
+      "locus": 27705,
+      "ethe": 27706,
+      "ĠKB": 27707,
+      "Ġhomologs": 27708,
+      "Ġencephalitis": 27709,
+      "Ġzig": 27710,
+      "Ġincentive": 27711,
+      "Ġconfidential": 27712,
+      "Ġvestibular": 27713,
+      "ĠOTUs": 27714,
+      "Ġsynovial": 27715,
+      "ĠRelativity": 27716,
+      "Ġsubdivided": 27717,
+      "chez": 27718,
+      "Ġlikewise": 27719,
+      "ĠPDMS": 27720,
+      "ĠÅł": 27721,
+      "Ġsocieties": 27722,
+      "ocyanate": 27723,
+      "gia": 27724,
+      "Ġlocalize": 27725,
+      "Ġlactation": 27726,
+      "Ġnodule": 27727,
+      "ĠCOR": 27728,
+      "Ġharboring": 27729,
+      "ĠEQU": 27730,
+      "harvest": 27731,
+      "Ġbandgap": 27732,
+      "rk": 27733,
+      "Ġresistor": 27734,
+      "Ġye": 27735,
+      "ĠAsymmetric": 27736,
+      "Ġpropagators": 27737,
+      "Ġdiagnosing": 27738,
+      "ĠAffairs": 27739,
+      "Ġejecta": 27740,
+      "Ġisomer": 27741,
+      "Ġix": 27742,
+      "Ġfoliation": 27743,
+      "Ġcapacitors": 27744,
+      "Ġcad": 27745,
+      "ĠNeutroph": 27746,
+      "pliance": 27747,
+      "Ġcompressible": 27748,
+      "ĠHunter": 27749,
+      "ĠMZ": 27750,
+      "ĠWeib": 27751,
+      "Ġnoncoding": 27752,
+      "Ġmountains": 27753,
+      "Ġadvertising": 27754,
+      "alez": 27755,
+      "bright": 27756,
+      "limsup": 27757,
+      "Ci": 27758,
+      "ĠNev": 27759,
+      "ĠStrains": 27760,
+      "ostomy": 27761,
+      "opal": 27762,
+      "Ġconcatenated": 27763,
+      "ĠPerf": 27764,
+      "CHO": 27765,
+      "Ġturtles": 27766,
+      "ĠFra": 27767,
+      "Ġallogeneic": 27768,
+      "Ġunsuccessful": 27769,
+      "YM": 27770,
+      "erver": 27771,
+      "Ġcuc": 27772,
+      "Ġfires": 27773,
+      "chart": 27774,
+      "Ġinterrupted": 27775,
+      "Ġdecides": 27776,
+      "Ġauction": 27777,
+      "ĠUntil": 27778,
+      "ĠATG": 27779,
+      "Ġdiam": 27780,
+      "magnitude": 27781,
+      "Ġdl": 27782,
+      "Vertex": 27783,
+      "mont": 27784,
+      "Ġfemtosecond": 27785,
+      "Params": 27786,
+      "Ġlysate": 27787,
+      "ishers": 27788,
+      "ĠPAT": 27789,
+      "ĠKev": 27790,
+      "ĠKnock": 27791,
+      "Ġgroove": 27792,
+      "Lu": 27793,
+      "ĠJohann": 27794,
+      "Ġreplica": 27795,
+      "ĠMATERIALS": 27796,
+      "Ġlots": 27797,
+      "Ġgenerically": 27798,
+      "ĠAltered": 27799,
+      "ĠIdentity": 27800,
+      "Ġunfolded": 27801,
+      "CES": 27802,
+      "ingular": 27803,
+      "ĠFraction": 27804,
+      "ĠProliferation": 27805,
+      "ĠVienna": 27806,
+      "acia": 27807,
+      "pless": 27808,
+      "ĠSevent": 27809,
+      "Ġturbines": 27810,
+      "lysine": 27811,
+      "Ġperoxis": 27812,
+      "ARP": 27813,
+      "ĠEpis": 27814,
+      "ĠSYBR": 27815,
+      "Builder": 27816,
+      "Ġspherically": 27817,
+      "Ġdefend": 27818,
+      "Performance": 27819,
+      "Ġmortar": 27820,
+      "ĠConcepts": 27821,
+      "works": 27822,
+      "Ġreinforce": 27823,
+      "á¹": 27824,
+      "Ġcus": 27825,
+      "ĠCIF": 27826,
+      "ĠAgricultural": 27827,
+      "crystalline": 27828,
+      "rish": 27829,
+      "Ġreferenced": 27830,
+      "Ġactress": 27831,
+      "Ġboundedness": 27832,
+      "SiC": 27833,
+      "Ġâ": 27834,
+      "Ġjack": 27835,
+      "Ġterminate": 27836,
+      "ĠJA": 27837,
+      "ĠKrish": 27838,
+      "MMP": 27839,
+      "kx": 27840,
+      "ĠPSR": 27841,
+      "endl": 27842,
+      "WHO": 27843,
+      "ĠSão": 27844,
+      "ĠCultural": 27845,
+      "ĠEh": 27846,
+      "ulis": 27847,
+      "vik": 27848,
+      "prises": 27849,
+      "ixel": 27850,
+      "ĠMetrics": 27851,
+      "Ġdiscontinuities": 27852,
+      "ĠUne": 27853,
+      "SCR": 27854,
+      "Ġprojecting": 27855,
+      "ĠOriginal": 27856,
+      "ĠHumans": 27857,
+      "transcriptional": 27858,
+      "HK": 27859,
+      "ĠJain": 27860,
+      "atisfaction": 27861,
+      "mesenchymal": 27862,
+      "Ġpyramid": 27863,
+      "Ġascorbic": 27864,
+      "game": 27865,
+      "Ġnoun": 27866,
+      "otoxins": 27867,
+      "peptide": 27868,
+      "Ġglassy": 27869,
+      "Ġtalking": 27870,
+      "Dem": 27871,
+      "ĠSchro": 27872,
+      "ĠAssumptions": 27873,
+      "Ġðx": 27874,
+      "Ġaneurysms": 27875,
+      "MASS": 27876,
+      "ĠHou": 27877,
+      "exposure": 27878,
+      "ĠLLC": 27879,
+      "Ġnoises": 27880,
+      "CTG": 27881,
+      "ĠElementary": 27882,
+      "flip": 27883,
+      "Ġdysp": 27884,
+      "Ġmessenger": 27885,
+      "ĠImportant": 27886,
+      "Ġimposes": 27887,
+      "Ġorganelles": 27888,
+      "assertEqual": 27889,
+      "Ġjustif": 27890,
+      "ucine": 27891,
+      "Ġformic": 27892,
+      "ormalization": 27893,
+      "ĠRadial": 27894,
+      "ĠCurve": 27895,
+      "ĠCrohn": 27896,
+      "Ġbrowser": 27897,
+      "Ġeffusion": 27898,
+      "Ġhandles": 27899,
+      "varsigma": 27900,
+      "Ġspecialists": 27901,
+      "Ġpainful": 27902,
+      "Ġerythematosus": 27903,
+      "Ġfen": 27904,
+      "nitrophenyl": 27905,
+      "Ġlegacy": 27906,
+      "ĠQDs": 27907,
+      "rapper": 27908,
+      "Ġmonotherapy": 27909,
+      "ĠBelt": 27910,
+      "ZZ": 27911,
+      "Ġsintered": 27912,
+      "enedi": 27913,
+      "Hb": 27914,
+      "tv": 27915,
+      "ĠNas": 27916,
+      "ovis": 27917,
+      "Ġmucin": 27918,
+      "Ġaccelerates": 27919,
+      "Ġacquiring": 27920,
+      "luc": 27921,
+      "Ġdilaton": 27922,
+      "ĠPitts": 27923,
+      "Ġequivariant": 27924,
+      "ĠLyman": 27925,
+      "ĠYa": 27926,
+      "Ġprogressed": 27927,
+      "ĠAfterwards": 27928,
+      "ĠCHAR": 27929,
+      "Don": 27930,
+      "Ġhistologic": 27931,
+      "Ġcircuitry": 27932,
+      "pene": 27933,
+      "opres": 27934,
+      "ĠStefan": 27935,
+      "Ġsemiclassical": 27936,
+      "mund": 27937,
+      "ĠWaste": 27938,
+      "BQ": 27939,
+      "Ġadiponectin": 27940,
+      "Ġunseen": 27941,
+      "Ġbiomechanical": 27942,
+      "Ġhazardous": 27943,
+      "ructive": 27944,
+      "xyl": 27945,
+      "opf": 27946,
+      "Ġprion": 27947,
+      "ĠInfinite": 27948,
+      "Ġtracers": 27949,
+      "ĠHarrison": 27950,
+      "Ġfibrinogen": 27951,
+      "Ġhydrolys": 27952,
+      "Ġislets": 27953,
+      "Ġparallelism": 27954,
+      "Spect": 27955,
+      "Ġimperative": 27956,
+      "Ġcured": 27957,
+      "ĠDSB": 27958,
+      "idefinite": 27959,
+      "icker": 27960,
+      "Ġdivergences": 27961,
+      "ĠShapiro": 27962,
+      "abd": 27963,
+      "ĠLum": 27964,
+      "ĠVD": 27965,
+      "Ġfisheries": 27966,
+      "geon": 27967,
+      "copenia": 27968,
+      "ĠClay": 27969,
+      "Ġmaximized": 27970,
+      "ĠGrey": 27971,
+      "ĠBatch": 27972,
+      "Ġinfest": 27973,
+      "Ġample": 27974,
+      "Ġestate": 27975,
+      "ĠSupreme": 27976,
+      "AO": 27977,
+      "isia": 27978,
+      "ĠSmad": 27979,
+      "Carlo": 27980,
+      "ĠSubst": 27981,
+      "Ġmonoidal": 27982,
+      "Ġnumeric": 27983,
+      "Plot": 27984,
+      "Ġdystrophy": 27985,
+      "hypertensive": 27986,
+      "Ġstool": 27987,
+      "alsy": 27988,
+      "Ġcheese": 27989,
+      "nih": 27990,
+      "Ġbought": 27991,
+      "ĠSQ": 27992,
+      "Ġclues": 27993,
+      "Ġmeiotic": 27994,
+      "Ġgoats": 27995,
+      "ĠGTPase": 27996,
+      "Ġrescaling": 27997,
+      "NUM": 27998,
+      "icing": 27999,
+      "ĠÄĢ": 28000,
+      "Ġpretty": 28001,
+      "ligand": 28002,
+      "English": 28003,
+      "ĠIntelligent": 28004,
+      "Every": 28005,
+      "ĠPolitical": 28006,
+      "enton": 28007,
+      "Ġpassages": 28008,
+      "ĠRemarks": 28009,
+      "sb": 28010,
+      "Network": 28011,
+      "ĠLRR": 28012,
+      "Ġcurl": 28013,
+      "ursion": 28014,
+      "ĠAver": 28015,
+      "ĠGLP": 28016,
+      "heren": 28017,
+      "atan": 28018,
+      "ICENSE": 28019,
+      "Ġlatex": 28020,
+      "EMI": 28021,
+      "quasi": 28022,
+      "ĠOm": 28023,
+      "Ġreviewing": 28024,
+      "Background": 28025,
+      "Ġsom": 28026,
+      "Ġsnapshots": 28027,
+      "brow": 28028,
+      "who": 28029,
+      "ĠTail": 28030,
+      "ĠMSM": 28031,
+      "ĠGm": 28032,
+      "Ġphi": 28033,
+      "rency": 28034,
+      "separated": 28035,
+      "Ġgig": 28036,
+      "osides": 28037,
+      "Ġpean": 28038,
+      "Ġappealing": 28039,
+      "PU": 28040,
+      "nk": 28041,
+      "Ġquer": 28042,
+      "ĠCharg": 28043,
+      "ĠMolecules": 28044,
+      "localization": 28045,
+      "Idx": 28046,
+      "lap": 28047,
+      "ĠTax": 28048,
+      "ĠExponential": 28049,
+      "ĠInhibitor": 28050,
+      "ĠBiomedical": 28051,
+      "urethane": 28052,
+      "lerene": 28053,
+      "rogenesis": 28054,
+      "ĠLai": 28055,
+      "ĠAggregation": 28056,
+      "ĠCaCl": 28057,
+      "Ġsensible": 28058,
+      "Ġconjunc": 28059,
+      "paper": 28060,
+      "ĠCovid": 28061,
+      "ĠProcedures": 28062,
+      "Ġknew": 28063,
+      "Ġsetae": 28064,
+      "ĠAlle": 28065,
+      "ĠExcept": 28066,
+      "Ġpresynaptic": 28067,
+      "flower": 28068,
+      "Ġultrasonography": 28069,
+      "Ġentertain": 28070,
+      "iors": 28071,
+      "ĠEry": 28072,
+      "ĠInteger": 28073,
+      "Ġrepressor": 28074,
+      "Ġlaterally": 28075,
+      "Ġcomplemented": 28076,
+      "TAG": 28077,
+      "ĠAround": 28078,
+      "ĠLister": 28079,
+      "bitrary": 28080,
+      "backward": 28081,
+      "MeV": 28082,
+      "Ġwhisk": 28083,
+      "AMs": 28084,
+      "ĠBulk": 28085,
+      "Ġquiver": 28086,
+      "Ġdamaging": 28087,
+      "ĠQuantifying": 28088,
+      "Ġsuprem": 28089,
+      "tel": 28090,
+      "Ġtear": 28091,
+      "oters": 28092,
+      "vidin": 28093,
+      "Ġtubules": 28094,
+      "Ġipsilateral": 28095,
+      "isive": 28096,
+      "Ġsuitably": 28097,
+      "riel": 28098,
+      "Ġtuber": 28099,
+      "Ġfavors": 28100,
+      "Ġcentim": 28101,
+      "Ġtransversal": 28102,
+      "ĠCHO": 28103,
+      "Ġtrimester": 28104,
+      "CAC": 28105,
+      "cognitive": 28106,
+      "ĠUTC": 28107,
+      "pute": 28108,
+      "Ġmidline": 28109,
+      "amers": 28110,
+      "evaluation": 28111,
+      "Dav": 28112,
+      "Ġbags": 28113,
+      "timer": 28114,
+      "Ġshortcomings": 28115,
+      "ĠErd": 28116,
+      "Ġdiscriminator": 28117,
+      "Ant": 28118,
+      "sizes": 28119,
+      "Ġbist": 28120,
+      "ingual": 28121,
+      "ĠCategory": 28122,
+      "Ġpulsars": 28123,
+      "ĠSchwartz": 28124,
+      "ĠDrop": 28125,
+      "Sequence": 28126,
+      "Ġtann": 28127,
+      "ĠSymptoms": 28128,
+      "Dict": 28129,
+      "ĠBlu": 28130,
+      "Supplemental": 28131,
+      "Ġdisabled": 28132,
+      "ĠKoz": 28133,
+      "Ġinvoked": 28134,
+      "ĠCQ": 28135,
+      "ĠConnectivity": 28136,
+      "Ġtelescopes": 28137,
+      "oso": 28138,
+      "Ġphytochemical": 28139,
+      "Ġorthogonality": 28140,
+      "Ġinvisible": 28141,
+      "ĠSCF": 28142,
+      "ĠAvoid": 28143,
+      "ĠHus": 28144,
+      "micron": 28145,
+      "aternity": 28146,
+      "Project": 28147,
+      "Ġadvancing": 28148,
+      "ĠLorentzian": 28149,
+      "Sa": 28150,
+      "tÃŀ": 28151,
+      "ĠUP": 28152,
+      "Ġarts": 28153,
+      "Ġzer": 28154,
+      "asket": 28155,
+      "Ġappeal": 28156,
+      "nick": 28157,
+      "ĠCloning": 28158,
+      "Ġswap": 28159,
+      "Ġphospholipids": 28160,
+      "bg": 28161,
+      "othel": 28162,
+      "asco": 28163,
+      "Track": 28164,
+      "Ġsubmanifold": 28165,
+      "Offset": 28166,
+      "ĠBird": 28167,
+      "problems": 28168,
+      "DCs": 28169,
+      "Ġdow": 28170,
+      "Ġdeionized": 28171,
+      "Ġsubclass": 28172,
+      "Ġpublishing": 28173,
+      "ĠCarter": 28174,
+      "Ġsynergy": 28175,
+      "Ġweakened": 28176,
+      "ĠGlas": 28177,
+      "ĠPie": 28178,
+      "henko": 28179,
+      "Ġsetups": 28180,
+      "ĠBernstein": 28181,
+      "Ġÿ": 28182,
+      "ĠShu": 28183,
+      "ĠChanging": 28184,
+      "osov": 28185,
+      "ĠMeteor": 28186,
+      "inth": 28187,
+      "rah": 28188,
+      "paramet": 28189,
+      "rena": 28190,
+      "Ġnewborns": 28191,
+      "ische": 28192,
+      "rotating": 28193,
+      "Ġconfident": 28194,
+      "fac": 28195,
+      "ĠTerr": 28196,
+      "Ġlinewidth": 28197,
+      "ICP": 28198,
+      "thony": 28199,
+      "Ġlanes": 28200,
+      "Ġsmoother": 28201,
+      "mony": 28202,
+      "ĠCNNs": 28203,
+      "Port": 28204,
+      "Ġtransiently": 28205,
+      "Ġsurgeries": 28206,
+      "Ġsubmerged": 28207,
+      "Ġpuncture": 28208,
+      "Ġdichlor": 28209,
+      "Ġsystematics": 28210,
+      "Ġcontigs": 28211,
+      "Ġresiding": 28212,
+      "BW": 28213,
+      "EO": 28214,
+      "Gold": 28215,
+      "ionate": 28216,
+      "vocab": 28217,
+      "dW": 28218,
+      "STAR": 28219,
+      "ĠPLC": 28220,
+      "athi": 28221,
+      "ĠInfectious": 28222,
+      "Light": 28223,
+      "á»": 28224,
+      "ĠRal": 28225,
+      "Ġpropagates": 28226,
+      "ĠLikelihood": 28227,
+      "hill": 28228,
+      "curl": 28229,
+      "checkpoint": 28230,
+      "rax": 28231,
+      "Ġvancomycin": 28232,
+      "ĠUSD": 28233,
+      "opheles": 28234,
+      "Ġfiltr": 28235,
+      "Ġstoichiometry": 28236,
+      "âĶĢâĶĢ": 28237,
+      "ĠNad": 28238,
+      "accessible": 28239,
+      "Ġtoy": 28240,
+      "Ġnude": 28241,
+      "ĠSut": 28242,
+      "essential": 28243,
+      "ĠOL": 28244,
+      "Ġpertin": 28245,
+      "Ġrecur": 28246,
+      "Ġcapill": 28247,
+      "Ġcomputable": 28248,
+      "Ġsuction": 28249,
+      "Ġsoftening": 28250,
+      "ĠESI": 28251,
+      "Ġmonitors": 28252,
+      "Ġpyridine": 28253,
+      "ĠSensors": 28254,
+      "ĠCombinatorial": 28255,
+      "atta": 28256,
+      "ĠAMS": 28257,
+      "ĠDul": 28258,
+      "pleteness": 28259,
+      "Eth": 28260,
+      "Ġû": 28261,
+      "Ġexcised": 28262,
+      "ĠDiabetic": 28263,
+      "ĠIowa": 28264,
+      "Ġimmunostaining": 28265,
+      "Ġillnesses": 28266,
+      "Ġenumer": 28267,
+      "ĠIranian": 28268,
+      "Ġthumb": 28269,
+      "orphisms": 28270,
+      "Ġlegitimate": 28271,
+      "lg": 28272,
+      "ĠSVD": 28273,
+      "Ġdesk": 28274,
+      "Format": 28275,
+      "Bon": 28276,
+      "Ġgarden": 28277,
+      "Ġinterpersonal": 28278,
+      "Ġelbow": 28279,
+      "ĠDemonstr": 28280,
+      "Ġnonspecific": 28281,
+      "Ferm": 28282,
+      "ivalently": 28283,
+      "phthalene": 28284,
+      "ARGET": 28285,
+      "Valid": 28286,
+      "Ġsunlight": 28287,
+      "Ġrescued": 28288,
+      "DAR": 28289,
+      "ĠInvariant": 28290,
+      "Ġidle": 28291,
+      "Ġalkaloids": 28292,
+      "scales": 28293,
+      "ses": 28294,
+      "obicity": 28295,
+      "beat": 28296,
+      "Ġcentrifugal": 28297,
+      "analytical": 28298,
+      "pv": 28299,
+      "Ġtutorial": 28300,
+      "ĠNation": 28301,
+      "generator": 28302,
+      "Ġcollisional": 28303,
+      "ĠCME": 28304,
+      "Ġscrap": 28305,
+      "ĠQSO": 28306,
+      "Ġwax": 28307,
+      "ĠScenario": 28308,
+      "Ġminimizer": 28309,
+      "ĠMDPI": 28310,
+      "Ġprostaglandin": 28311,
+      "olites": 28312,
+      "ocysteine": 28313,
+      "Ġcompactification": 28314,
+      "Ġfrailty": 28315,
+      "opsin": 28316,
+      "Ġjunior": 28317,
+      "loud": 28318,
+      "Ġtitled": 28319,
+      "Ġeconomically": 28320,
+      "thiophene": 28321,
+      "ĠInvestigating": 28322,
+      "ĠEsp": 28323,
+      "Ġelusive": 28324,
+      "Ġmalware": 28325,
+      "ĠTHP": 28326,
+      "imidazole": 28327,
+      "Ġretains": 28328,
+      "ĠMIR": 28329,
+      "ffl": 28330,
+      "jac": 28331,
+      "ĠPART": 28332,
+      "ĠDCM": 28333,
+      "transport": 28334,
+      "MAPK": 28335,
+      "Problem": 28336,
+      "Su": 28337,
+      "Ġdelim": 28338,
+      "Ġpsychometric": 28339,
+      "vitably": 28340,
+      "Ġhypergeometric": 28341,
+      "Ġuterus": 28342,
+      "Ġanaesthesia": 28343,
+      "ĠAvenue": 28344,
+      "Ġmeanings": 28345,
+      "Ġrapidity": 28346,
+      "Ġdendrites": 28347,
+      "grain": 28348,
+      "ĠNile": 28349,
+      "Ġfacies": 28350,
+      "Ġpipelines": 28351,
+      "ĠCampylobacter": 28352,
+      "ĠMembers": 28353,
+      "benzoate": 28354,
+      "Request": 28355,
+      "Ġpk": 28356,
+      "Ġrefused": 28357,
+      "caus": 28358,
+      "ĠSay": 28359,
+      "lane": 28360,
+      "ĠPSO": 28361,
+      "Ġgathering": 28362,
+      "Ġrefriger": 28363,
+      "RCC": 28364,
+      "Ġfibronectin": 28365,
+      "help": 28366,
+      "ĠIntensity": 28367,
+      "CLC": 28368,
+      "Que": 28369,
+      "elly": 28370,
+      "Ġilluminated": 28371,
+      "Ġpedestrian": 28372,
+      "ĠMercury": 28373,
+      "Ġafforded": 28374,
+      "Ġpathophysiological": 28375,
+      "ĠNGS": 28376,
+      "assa": 28377,
+      "Ġendors": 28378,
+      "Ġsensation": 28379,
+      "Ġstreamflow": 28380,
+      "avin": 28381,
+      "ĠGABAergic": 28382,
+      "Ġretirement": 28383,
+      "Cells": 28384,
+      "oca": 28385,
+      "Ġoptimizations": 28386,
+      "Ġdigraph": 28387,
+      "ĠAutism": 28388,
+      "octurnal": 28389,
+      "oscience": 28390,
+      "ĠEllis": 28391,
+      "ĠAj": 28392,
+      "ĠWSN": 28393,
+      "Ġshooting": 28394,
+      "iper": 28395,
+      "îĦĥ": 28396,
+      "ĠWeather": 28397,
+      "Ġreceptive": 28398,
+      "Ġquartic": 28399,
+      "ocyclic": 28400,
+      "PATH": 28401,
+      "sizeof": 28402,
+      "Ġmelts": 28403,
+      "Ġdipoles": 28404,
+      "Ġbimodal": 28405,
+      "summary": 28406,
+      "Ġinsomnia": 28407,
+      "opyran": 28408,
+      "Ġwrapped": 28409,
+      "ĠJosé": 28410,
+      "AH": 28411,
+      "cia": 28412,
+      "Ġobeys": 28413,
+      "ĠKay": 28414,
+      "intervention": 28415,
+      "Ġrouter": 28416,
+      "ĠDrugs": 28417,
+      "owska": 28418,
+      "ĠArr": 28419,
+      "ĠCaptain": 28420,
+      "ĠTMS": 28421,
+      "adv": 28422,
+      "Ġboat": 28423,
+      "Ġtrusted": 28424,
+      "sever": 28425,
+      "illars": 28426,
+      "ĠMissouri": 28427,
+      "Ġequivalents": 28428,
+      "ĠHarvard": 28429,
+      "ĠClarke": 28430,
+      "resonant": 28431,
+      "rady": 28432,
+      "triggered": 28433,
+      "Ġcleft": 28434,
+      "Ġunic": 28435,
+      "Ġbrainstem": 28436,
+      "Ġthrombin": 28437,
+      "ĠFlight": 28438,
+      "Ġsectional": 28439,
+      "Ġconcatenation": 28440,
+      "Ġcantilever": 28441,
+      "eton": 28442,
+      "Ġdecode": 28443,
+      "ofacial": 28444,
+      "Action": 28445,
+      "ĠIllustration": 28446,
+      "vertical": 28447,
+      "chall": 28448,
+      "ĠRegistry": 28449,
+      "MAT": 28450,
+      "Ġconson": 28451,
+      "Ġneoadjuvant": 28452,
+      "ĠWistar": 28453,
+      "ĠImper": 28454,
+      "Ġaltitudes": 28455,
+      "Ġsubpopulation": 28456,
+      "ĠScene": 28457,
+      "tensorflow": 28458,
+      "slow": 28459,
+      "Ġhint": 28460,
+      "Ġbeamforming": 28461,
+      "ein": 28462,
+      "Ġimpregn": 28463,
+      "ĠRFID": 28464,
+      "ĠAnalyzing": 28465,
+      "ĠPent": 28466,
+      "ĠDNS": 28467,
+      "ĠGilbert": 28468,
+      "Ġcrater": 28469,
+      "Comparing": 28470,
+      "Ġbf": 28471,
+      "Ġflights": 28472,
+      "Ġmalnutrition": 28473,
+      "SMC": 28474,
+      "Ġerythrop": 28475,
+      "ĠTumors": 28476,
+      "Tx": 28477,
+      "Ġisospin": 28478,
+      "ĠKub": 28479,
+      "iking": 28480,
+      "Ġcorticosteroids": 28481,
+      "ursor": 28482,
+      "ĠBurg": 28483,
+      "inspired": 28484,
+      "ĠIgn": 28485,
+      "Ġmycel": 28486,
+      "prediction": 28487,
+      "methods": 28488,
+      "Copy": 28489,
+      "ĠRW": 28490,
+      "ĠKnight": 28491,
+      "Ġdemethyl": 28492,
+      "ìĦ": 28493,
+      "Ġcili": 28494,
+      "Ġbes": 28495,
+      "ĠEck": 28496,
+      "Ġdilatation": 28497,
+      "Ġanimation": 28498,
+      "abstract": 28499,
+      "Ġcircumvent": 28500,
+      "Ġinoculum": 28501,
+      "Seg": 28502,
+      "ĠCaps": 28503,
+      "erers": 28504,
+      "PLS": 28505,
+      "ĠPeer": 28506,
+      "Ġverifies": 28507,
+      "ategy": 28508,
+      "ogenetics": 28509,
+      "Ġoligonucleotides": 28510,
+      "ractical": 28511,
+      "Ġdiverges": 28512,
+      "ĠStanford": 28513,
+      "ĠAi": 28514,
+      "Ġweighing": 28515,
+      "Tg": 28516,
+      "reinfor": 28517,
+      "ĠAlam": 28518,
+      "quiry": 28519,
+      "ĠNob": 28520,
+      "Ġlinearization": 28521,
+      "ĠVenez": 28522,
+      "nexin": 28523,
+      "levels": 28524,
+      "Lip": 28525,
+      "ĠPatel": 28526,
+      "ĠMagnitude": 28527,
+      "etitive": 28528,
+      "ĠEagle": 28529,
+      "Ġsputum": 28530,
+      "ĠCOS": 28531,
+      "Ġincubator": 28532,
+      "Ul": 28533,
+      "ĠReceptors": 28534,
+      "ĠSchott": 28535,
+      "GCG": 28536,
+      "ĠZeiss": 28537,
+      "ĠEntanglement": 28538,
+      "ĠVaccine": 28539,
+      "orted": 28540,
+      "Ġnb": 28541,
+      "ĠSj": 28542,
+      "ĠMrs": 28543,
+      "Ġcalf": 28544,
+      "Ġintegrability": 28545,
+      "ĠPhoton": 28546,
+      "Ġgondii": 28547,
+      "ĠMIL": 28548,
+      "Ġaliph": 28549,
+      "ĠDip": 28550,
+      "falls": 28551,
+      "ctrl": 28552,
+      "ku": 28553,
+      "etent": 28554,
+      "plt": 28555,
+      "Ġpersisted": 28556,
+      "ĠManager": 28557,
+      "Ġprerequisite": 28558,
+      "filling": 28559,
+      "ĠMEA": 28560,
+      "Sym": 28561,
+      "ĠGrain": 28562,
+      "Ġductal": 28563,
+      "ĠTODO": 28564,
+      "Ġaffinities": 28565,
+      "Ġdegenerative": 28566,
+      "ĠFitz": 28567,
+      "ovar": 28568,
+      "ĠTriple": 28569,
+      "Ġdendrim": 28570,
+      "ĠFranklin": 28571,
+      "mag": 28572,
+      "otely": 28573,
+      "Ġstabilizes": 28574,
+      "Ġcash": 28575,
+      "ĠSquad": 28576,
+      "Ġchampion": 28577,
+      "PDB": 28578,
+      "Ġurg": 28579,
+      "Ġalcoholic": 28580,
+      "Ġtar": 28581,
+      "yled": 28582,
+      "Version": 28583,
+      "Ġsale": 28584,
+      "ĠMLP": 28585,
+      "outer": 28586,
+      "Ġsimplifying": 28587,
+      "ĠExtract": 28588,
+      "Param": 28589,
+      "ĠRestric": 28590,
+      "Ġtractable": 28591,
+      "ĠArchive": 28592,
+      "Response": 28593,
+      "ADDR": 28594,
+      "Ġcommutation": 28595,
+      "Rich": 28596,
+      "ĠAndrews": 28597,
+      "Ġosteoclast": 28598,
+      "romic": 28599,
+      "ĠShift": 28600,
+      "Ġaccelerometer": 28601,
+      "ĠSent": 28602,
+      "Ġchances": 28603,
+      "osting": 28604,
+      "Ġmethacrylate": 28605,
+      "Ġgluons": 28606,
+      "Ġôı½": 28607,
+      "Ġpolygons": 28608,
+      "ĠRCTs": 28609,
+      "Ġinfancy": 28610,
+      "Ġproceeded": 28611,
+      "ĠHorizontal": 28612,
+      "COR": 28613,
+      "Ġcaching": 28614,
+      "ĠNHS": 28615,
+      "phobic": 28616,
+      "ĠXMM": 28617,
+      "Ġmicrobiological": 28618,
+      "GMP": 28619,
+      "ÙĨ": 28620,
+      "ĠTSS": 28621,
+      "ĠSul": 28622,
+      "ĠFact": 28623,
+      "ĠWE": 28624,
+      "Ġcertainty": 28625,
+      "ensitivity": 28626,
+      "Ġdeconvolution": 28627,
+      "ĠGain": 28628,
+      "Ġblots": 28629,
+      "Ġseeks": 28630,
+      "Ġcosh": 28631,
+      "ennessee": 28632,
+      "Ġslave": 28633,
+      "ĠTran": 28634,
+      "Ġtranspose": 28635,
+      "reated": 28636,
+      "Ġshading": 28637,
+      "ĠBU": 28638,
+      "ĠOV": 28639,
+      "ĠLook": 28640,
+      "Ġcomprehensively": 28641,
+      "ĠFreder": 28642,
+      "Handler": 28643,
+      "fibr": 28644,
+      "Ġmissense": 28645,
+      "targets": 28646,
+      "promoting": 28647,
+      "ĠPep": 28648,
+      "varpi": 28649,
+      "ĠHarmonic": 28650,
+      "ĠAIS": 28651,
+      "Ġmonocyt": 28652,
+      "Ġthinning": 28653,
+      "Ġpheromone": 28654,
+      "Water": 28655,
+      "anase": 28656,
+      "ĠSang": 28657,
+      "Ġsubstructure": 28658,
+      "wp": 28659,
+      "ĠKansas": 28660,
+      "DEBUG": 28661,
+      "ĠProbe": 28662,
+      "Ġpatterned": 28663,
+      "clean": 28664,
+      "Ġbroiler": 28665,
+      "odextrin": 28666,
+      "aided": 28667,
+      "oprol": 28668,
+      "ublin": 28669,
+      "inum": 28670,
+      "Ġanatomic": 28671,
+      "Ġplating": 28672,
+      "arro": 28673,
+      "ucal": 28674,
+      "Ġspeedup": 28675,
+      "Ġhaemorrh": 28676,
+      "eptidase": 28677,
+      "Ġsagittal": 28678,
+      "Ġintim": 28679,
+      "ĠFISH": 28680,
+      "Ġscarc": 28681,
+      "ATCC": 28682,
+      "incor": 28683,
+      "Ġserological": 28684,
+      "ente": 28685,
+      "Ġshale": 28686,
+      "Ġoverfitting": 28687,
+      "ĠExcess": 28688,
+      "ĠALP": 28689,
+      "Pool": 28690,
+      "dry": 28691,
+      "yu": 28692,
+      "ĠPMMA": 28693,
+      "ĠHypoxia": 28694,
+      "nothing": 28695,
+      "chestra": 28696,
+      "coloneqq": 28697,
+      "Ġbibli": 28698,
+      "ĠEXPECT": 28699,
+      "BAL": 28700,
+      "ethan": 28701,
+      "ĠâĪĺ": 28702,
+      "Ġjourney": 28703,
+      "Ġbiocompatibility": 28704,
+      "PAN": 28705,
+      "Ġbon": 28706,
+      "ĠRoh": 28707,
+      "Ġpolarisation": 28708,
+      "Spin": 28709,
+      "idences": 28710,
+      "ĠBCR": 28711,
+      "ĠHIP": 28712,
+      "ĠThick": 28713,
+      "Ġrecognizes": 28714,
+      "Ġsar": 28715,
+      "Ġamend": 28716,
+      "questions": 28717,
+      "Ġcaregiver": 28718,
+      "ĠMarie": 28719,
+      "Ġmetalloproteinase": 28720,
+      "Ġaldehydes": 28721,
+      "Ġinterneurons": 28722,
+      "Ġtetrahedral": 28723,
+      "guez": 28724,
+      "Ġquasiparticle": 28725,
+      "Ġot": 28726,
+      "decreasing": 28727,
+      "stre": 28728,
+      "Ġphotoperiod": 28729,
+      "Ġprioriti": 28730,
+      "Ġapo": 28731,
+      "Ġimmunosuppression": 28732,
+      "ĠPierre": 28733,
+      "LPS": 28734,
+      "Ġclumps": 28735,
+      "ĠPlane": 28736,
+      "Ġturbidity": 28737,
+      "Ġpollutant": 28738,
+      "Ġbioch": 28739,
+      "ĠTRE": 28740,
+      "Ġdesigners": 28741,
+      "Ġrenders": 28742,
+      "Ġreplaces": 28743,
+      "ĠPLS": 28744,
+      "Ġhumoral": 28745,
+      "Bas": 28746,
+      "reira": 28747,
+      "ĠAedes": 28748,
+      "vitamin": 28749,
+      "curves": 28750,
+      "ociceptive": 28751,
+      "Ġindisp": 28752,
+      "Ġoxy": 28753,
+      "Ġedible": 28754,
+      "ĠMesenchymal": 28755,
+      "ĠDegree": 28756,
+      "ž": 28757,
+      "ĠOak": 28758,
+      "ĠBhatt": 28759,
+      "onso": 28760,
+      "ĠSBP": 28761,
+      "ĠAux": 28762,
+      "Ġmartingale": 28763,
+      "ĠMicrobiota": 28764,
+      "glow": 28765,
+      "Ġexud": 28766,
+      "apolis": 28767,
+      "Ġsomehow": 28768,
+      "Ġcentred": 28769,
+      "Channel": 28770,
+      "ĠNormalized": 28771,
+      "ilitation": 28772,
+      "Ġtranscriptase": 28773,
+      "Ġcryo": 28774,
+      "predicted": 28775,
+      "ĠDAG": 28776,
+      "Ġrf": 28777,
+      "endor": 28778,
+      "INTER": 28779,
+      "ĠMesh": 28780,
+      "ĠFundament": 28781,
+      "ycle": 28782,
+      "Ġprimitives": 28783,
+      "radiated": 28784,
+      "Ġrho": 28785,
+      "enesulf": 28786,
+      "ĠFSH": 28787,
+      "ĠEcos": 28788,
+      "localized": 28789,
+      "Ġenterprise": 28790,
+      "cephalus": 28791,
+      "Ġcarcass": 28792,
+      "AY": 28793,
+      "ecurity": 28794,
+      "ĠTMD": 28795,
+      "Ġlb": 28796,
+      "ĠAeros": 28797,
+      "ĠMER": 28798,
+      "Attr": 28799,
+      "ĠACL": 28800,
+      "ĠBarb": 28801,
+      "cout": 28802,
+      "Ġdeoxy": 28803,
+      "atios": 28804,
+      "Ġpersists": 28805,
+      "Ġviolent": 28806,
+      "Abelian": 28807,
+      "Ġellips": 28808,
+      "iong": 28809,
+      "Ġsuccessor": 28810,
+      "ĠGonzález": 28811,
+      "living": 28812,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 28813,
+      "imentin": 28814,
+      "Ġcapsules": 28815,
+      "VIS": 28816,
+      "ĠPOP": 28817,
+      "arithmic": 28818,
+      "OO": 28819,
+      "wl": 28820,
+      "inoic": 28821,
+      "ĠCenters": 28822,
+      "roblasts": 28823,
+      "those": 28824,
+      "ĠMJ": 28825,
+      "Ġfronts": 28826,
+      "Ġunint": 28827,
+      "Ġfacile": 28828,
+      "coherent": 28829,
+      "avour": 28830,
+      "ceptive": 28831,
+      "tah": 28832,
+      "Ġrelatedness": 28833,
+      "dE": 28834,
+      "ungen": 28835,
+      "#####": 28836,
+      "Ġamphi": 28837,
+      "ĠGuy": 28838,
+      "stars": 28839,
+      "ectom": 28840,
+      "Ġlaying": 28841,
+      "Ġspider": 28842,
+      "ACs": 28843,
+      "Ġseedling": 28844,
+      "Ġduplicated": 28845,
+      "iche": 28846,
+      "ĠMST": 28847,
+      "grass": 28848,
+      "Ġprophylactic": 28849,
+      "eks": 28850,
+      "Ġlaryngeal": 28851,
+      "ĠSper": 28852,
+      "ĠWals": 28853,
+      "Ġcholec": 28854,
+      "ĠPlanet": 28855,
+      "ĠHEPES": 28856,
+      "Ġdiploid": 28857,
+      "constraint": 28858,
+      "Pyx": 28859,
+      "ACh": 28860,
+      "ĠCui": 28861,
+      "ĠShared": 28862,
+      "ĠCand": 28863,
+      "ĠGö": 28864,
+      "Ġdetached": 28865,
+      "Ġpassengers": 28866,
+      "Ġaliphatic": 28867,
+      "Ġpour": 28868,
+      "Ġaccesses": 28869,
+      "ĠWald": 28870,
+      "Ġdecorated": 28871,
+      "Ġcarotenoids": 28872,
+      "uestions": 28873,
+      "ĠImpacts": 28874,
+      "SAT": 28875,
+      "aru": 28876,
+      "ĠPir": 28877,
+      "ĠConfiguration": 28878,
+      "ĠCongo": 28879,
+      "ĠLing": 28880,
+      "Ġdesic": 28881,
+      "Ġmacrom": 28882,
+      "Ġlacked": 28883,
+      "Ġencompasses": 28884,
+      "Ġpumped": 28885,
+      "ĠForty": 28886,
+      "rexate": 28887,
+      "ifferentiated": 28888,
+      "Ġnoble": 28889,
+      "Ġradion": 28890,
+      "Ġimmigrants": 28891,
+      "Ġbiodegradable": 28892,
+      "Ġmigrating": 28893,
+      "argv": 28894,
+      "COM": 28895,
+      "ĠObservational": 28896,
+      "Ġcannabis": 28897,
+      "yama": 28898,
+      "Ġconcentric": 28899,
+      "Conn": 28900,
+      "talion": 28901,
+      "Ġresponders": 28902,
+      "utenant": 28903,
+      "ĠTrim": 28904,
+      "Ġcontributors": 28905,
+      "Ġcontracted": 28906,
+      "ĠXenopus": 28907,
+      "Ġloud": 28908,
+      "ĠEnhancing": 28909,
+      "Ġinfarct": 28910,
+      "Ġok": 28911,
+      "Ġasks": 28912,
+      "relin": 28913,
+      "Ġillustrative": 28914,
+      "vdash": 28915,
+      "dg": 28916,
+      "Ġfoc": 28917,
+      "Ġlivers": 28918,
+      "ĠOtt": 28919,
+      "ĠTSP": 28920,
+      "logger": 28921,
+      "depending": 28922,
+      "Ġdisproportion": 28923,
+      "Ġintric": 28924,
+      "Ġimmunized": 28925,
+      "varez": 28926,
+      "Ġsalic": 28927,
+      "ĠInstitutes": 28928,
+      "KEY": 28929,
+      "Ġendoscopy": 28930,
+      "erk": 28931,
+      "eliness": 28932,
+      "ĠSag": 28933,
+      "athyroid": 28934,
+      "Ġacidity": 28935,
+      "arov": 28936,
+      "ĠVoronoi": 28937,
+      "Experimental": 28938,
+      "Ġgently": 28939,
+      "Measure": 28940,
+      "ïĺº": 28941,
+      "Ġwonder": 28942,
+      "ĠPancreatic": 28943,
+      "ĠHispanic": 28944,
+      "ĠEug": 28945,
+      "reducing": 28946,
+      "tainment": 28947,
+      "Ġsurprise": 28948,
+      "Ġæ": 28949,
+      "criter": 28950,
+      "ĠHypertension": 28951,
+      "tique": 28952,
+      "ĠCris": 28953,
+      "compatible": 28954,
+      "enson": 28955,
+      "Ġdistributional": 28956,
+      "ĠNAT": 28957,
+      "widths": 28958,
+      "Ġisotherms": 28959,
+      "ĠPrad": 28960,
+      "Ġbiodies": 28961,
+      "Ġorbifold": 28962,
+      "ĠEOS": 28963,
+      "Ġatax": 28964,
+      "ĠBod": 28965,
+      "ĠNMD": 28966,
+      "Ġmonoxide": 28967,
+      "ĠUkraine": 28968,
+      "foli": 28969,
+      "ĠDro": 28970,
+      "Ġunavailable": 28971,
+      "Ġbrighter": 28972,
+      "âĬĹ": 28973,
+      "omethane": 28974,
+      "Ġdream": 28975,
+      "Ġspo": 28976,
+      "ĠMaur": 28977,
+      "Ġoccasional": 28978,
+      "Ġinconsistency": 28979,
+      "ĠTac": 28980,
+      "opts": 28981,
+      "ĠGAB": 28982,
+      "ĠTao": 28983,
+      "ĠMatthew": 28984,
+      "ý": 28985,
+      "Ġpiano": 28986,
+      "ĠRCC": 28987,
+      "ĠOK": 28988,
+      "ĠKul": 28989,
+      "methan": 28990,
+      "ĠPROC": 28991,
+      "Ġconversations": 28992,
+      "ĠCSI": 28993,
+      "angent": 28994,
+      "ĠXue": 28995,
+      "Ġgraphic": 28996,
+      "dening": 28997,
+      "healthy": 28998,
+      "Ġfp": 28999,
+      "azone": 29000,
+      "Ġdiscipline": 29001,
+      "Ġprogresses": 29002,
+      "Ġbamboo": 29003,
+      "Ġcharm": 29004,
+      "ĠActivated": 29005,
+      "ĠSharp": 29006,
+      "ynes": 29007,
+      "Ġtoolbox": 29008,
+      "Ġheterostructures": 29009,
+      "piperazin": 29010,
+      "Ġarose": 29011,
+      "ĠInterval": 29012,
+      "Ġstripe": 29013,
+      "ĠChak": 29014,
+      "Ġcuff": 29015,
+      "RESS": 29016,
+      "Ġnonuniform": 29017,
+      "Ġbeetle": 29018,
+      "Prec": 29019,
+      "zc": 29020,
+      "Thread": 29021,
+      "bet": 29022,
+      "Ġee": 29023,
+      "ĠOptional": 29024,
+      "Ġtroph": 29025,
+      "ĠPuer": 29026,
+      "ĠFron": 29027,
+      "Ġmultiplet": 29028,
+      "Ġcalorimetry": 29029,
+      "Ġmonocytogenes": 29030,
+      "ĠHimal": 29031,
+      "Ġdrill": 29032,
+      "AGA": 29033,
+      "Ġferritin": 29034,
+      "Ġdpi": 29035,
+      "ĠCarm": 29036,
+      "Ġgone": 29037,
+      "Ġunidirectional": 29038,
+      "Ġreminis": 29039,
+      "Ġadjustable": 29040,
+      "ĠAustin": 29041,
+      "SARS": 29042,
+      "dal": 29043,
+      "Ġcef": 29044,
+      "equivariant": 29045,
+      "baseline": 29046,
+      "Ġspinors": 29047,
+      "ĠPrint": 29048,
+      "Ġmile": 29049,
+      "ĠLinc": 29050,
+      "mutation": 29051,
+      "Ġmucus": 29052,
+      "ĠHSC": 29053,
+      "Ġthermod": 29054,
+      "Ġpaint": 29055,
+      "Ġdistinctly": 29056,
+      "athy": 29057,
+      "Ġpharmacy": 29058,
+      "ĠBulg": 29059,
+      "ĠGang": 29060,
+      "hicle": 29061,
+      "ogan": 29062,
+      "ĠJian": 29063,
+      "ĠIndiana": 29064,
+      "Ġinstanton": 29065,
+      "Ġpalladium": 29066,
+      "fiber": 29067,
+      "npy": 29068,
+      "ĠUA": 29069,
+      "ĠQT": 29070,
+      "ceptible": 29071,
+      "etine": 29072,
+      "ĠHoles": 29073,
+      "Ġdependences": 29074,
+      "Ġthresholding": 29075,
+      "ĠMaintenance": 29076,
+      "Ġparticipates": 29077,
+      "ĠGenomes": 29078,
+      "factorial": 29079,
+      "ĠLiber": 29080,
+      "ĠThermodynamic": 29081,
+      "Ġelective": 29082,
+      "ucher": 29083,
+      "Ġhyperther": 29084,
+      "Ġstomatal": 29085,
+      "ĠBirth": 29086,
+      "cholesterol": 29087,
+      "Ġnotch": 29088,
+      "Ġsymbiotic": 29089,
+      "Ġbusinesses": 29090,
+      "Ġappreciable": 29091,
+      "Ġspecialization": 29092,
+      "ár": 29093,
+      "actyl": 29094,
+      "ĠGraphPad": 29095,
+      "osper": 29096,
+      "Ġorchestr": 29097,
+      "Ġdihydro": 29098,
+      "Ġconcluding": 29099,
+      "CLK": 29100,
+      "Ġeqs": 29101,
+      "ĠProgression": 29102,
+      "Ġclubs": 29103,
+      "aku": 29104,
+      "events": 29105,
+      "Ġsplenic": 29106,
+      "Ġbunch": 29107,
+      "ĠTm": 29108,
+      "ĠMobility": 29109,
+      "Ġtwofold": 29110,
+      "Ġradially": 29111,
+      "LSTM": 29112,
+      "MH": 29113,
+      "ĠCoal": 29114,
+      "Ġfrontier": 29115,
+      "Jan": 29116,
+      "Jun": 29117,
+      "ĠSimpson": 29118,
+      "Ġabstracts": 29119,
+      "Pal": 29120,
+      "Ġunim": 29121,
+      "Ġrobo": 29122,
+      "ĠIIB": 29123,
+      "depleted": 29124,
+      "Ġmorphologically": 29125,
+      "Ġenforcement": 29126,
+      "Ġdwell": 29127,
+      "Ġstagn": 29128,
+      "Ġlimestone": 29129,
+      "Ġmicrov": 29130,
+      "Ġïĥ¸": 29131,
+      "Luc": 29132,
+      "pacs": 29133,
+      "cyano": 29134,
+      "Ġintraocular": 29135,
+      "ĠCalculate": 29136,
+      "Support": 29137,
+      "SYS": 29138,
+      "ĠVS": 29139,
+      "CMs": 29140,
+      "Constant": 29141,
+      "ĠDj": 29142,
+      "Ġunbalanced": 29143,
+      "Ġrepeatability": 29144,
+      "gins": 29145,
+      "irect": 29146,
+      "ĠMOR": 29147,
+      "ĠBailey": 29148,
+      "Ġadvancement": 29149,
+      "Ġpursuit": 29150,
+      "Ġarom": 29151,
+      "proced": 29152,
+      "ĠInitiative": 29153,
+      "Ġincentives": 29154,
+      "Ġsurpass": 29155,
+      "genes": 29156,
+      "ĠIND": 29157,
+      "LH": 29158,
+      "Ġsuicidal": 29159,
+      "Ġbiodiesel": 29160,
+      "xz": 29161,
+      "ÙĬ": 29162,
+      "lea": 29163,
+      "ĠAnthony": 29164,
+      "Learning": 29165,
+      "Ġundo": 29166,
+      "Ġïĥº": 29167,
+      "ĠCommunities": 29168,
+      "hua": 29169,
+      "itime": 29170,
+      "ĠDean": 29171,
+      "Ġplasmin": 29172,
+      "ÃŃnez": 29173,
+      "ohydrate": 29174,
+      "Ġneurodevelop": 29175,
+      "Ġstoichiometric": 29176,
+      "ĠOncology": 29177,
+      "Ġshower": 29178,
+      "ĠDMS": 29179,
+      "WOR": 29180,
+      "ĠPIP": 29181,
+      "Ġsteric": 29182,
+      "mittees": 29183,
+      "istol": 29184,
+      "oxins": 29185,
+      "noon": 29186,
+      "FFT": 29187,
+      "Ġá»": 29188,
+      "opoiesis": 29189,
+      "Ġresembling": 29190,
+      "ĠBord": 29191,
+      "Ġprobiotics": 29192,
+      "ocysts": 29193,
+      "grey": 29194,
+      "ĠCatalog": 29195,
+      "IZATION": 29196,
+      "illes": 29197,
+      "ĠAlan": 29198,
+      "ĠÅ·": 29199,
+      "ĠLeib": 29200,
+      "ĠReasoning": 29201,
+      "biological": 29202,
+      "uterine": 29203,
+      "vacizumab": 29204,
+      "lecommun": 29205,
+      "ĠWarm": 29206,
+      "epage": 29207,
+      "variants": 29208,
+      "BSA": 29209,
+      "Ġïĥ¶": 29210,
+      "Ġhepatocyte": 29211,
+      "ketch": 29212,
+      "Ġstripping": 29213,
+      "ĠAdverse": 29214,
+      "ĠFeas": 29215,
+      "Ġïĥ¯": 29216,
+      "Pac": 29217,
+      "Ġindentation": 29218,
+      "Ġsecular": 29219,
+      "Ġidentifiable": 29220,
+      "running": 29221,
+      "Ġrd": 29222,
+      "Ġzyg": 29223,
+      "ĠDictionary": 29224,
+      "Ġresveratrol": 29225,
+      "inesterase": 29226,
+      "Ġtetracycline": 29227,
+      "ubles": 29228,
+      "Ġthroat": 29229,
+      "ĠLamb": 29230,
+      "aryon": 29231,
+      "ĠSQL": 29232,
+      "ĠÃľ": 29233,
+      "Ġglycemic": 29234,
+      "Ġcompetent": 29235,
+      "ĠAgreement": 29236,
+      "oiced": 29237,
+      "Ġconstitutively": 29238,
+      "Ġelectrocardi": 29239,
+      "oplasma": 29240,
+      "ĠîĦĥ": 29241,
+      "anide": 29242,
+      "Ġreorganization": 29243,
+      "Ġuninfected": 29244,
+      "UTE": 29245,
+      "Ġroyal": 29246,
+      "ĠSit": 29247,
+      "Ġmarital": 29248,
+      "ĠKobayashi": 29249,
+      "Barr": 29250,
+      "ĠTennessee": 29251,
+      "ĠChromat": 29252,
+      "ĠDerm": 29253,
+      "projection": 29254,
+      "ĠJob": 29255,
+      "Ġâīł": 29256,
+      "ĠTrip": 29257,
+      "Ġisop": 29258,
+      "Ġprojector": 29259,
+      "Ġatmospheres": 29260,
+      "Ġperforation": 29261,
+      "storage": 29262,
+      "iths": 29263,
+      "Ġmonomeric": 29264,
+      "ĠUSB": 29265,
+      "ĠEve": 29266,
+      "Ġspore": 29267,
+      "ĠmT": 29268,
+      "oxazole": 29269,
+      "ĠDeformation": 29270,
+      "Ġtextual": 29271,
+      "Ġwarf": 29272,
+      "Ġneuropathic": 29273,
+      "prepared": 29274,
+      "Ġblended": 29275,
+      "ĠHouston": 29276,
+      "************************************************************************": 29277,
+      "esters": 29278,
+      "Equals": 29279,
+      "Ġallergen": 29280,
+      "Ġpertinent": 29281,
+      "facts": 29282,
+      "uctions": 29283,
+      "Ġclocks": 29284,
+      "ĠVia": 29285,
+      "ĠCDF": 29286,
+      "Ġestuary": 29287,
+      "Ġphenomenology": 29288,
+      "arus": 29289,
+      "APH": 29290,
+      "Ġargues": 29291,
+      "Ġinserts": 29292,
+      "gow": 29293,
+      "hart": 29294,
+      "Ġchemotaxis": 29295,
+      "Ġpv": 29296,
+      "Ġrein": 29297,
+      "ĠGrim": 29298,
+      "ĠVF": 29299,
+      "Ġeffic": 29300,
+      "ĠProfiling": 29301,
+      "Ġanodic": 29302,
+      "ĠDENV": 29303,
+      "ĠWit": 29304,
+      "ĠSYSTEM": 29305,
+      "ĠCayley": 29306,
+      "Eng": 29307,
+      "ĠAQP": 29308,
+      "interactions": 29309,
+      "iliarity": 29310,
+      "ĠPromotes": 29311,
+      "Ġdams": 29312,
+      "ington": 29313,
+      "ffff": 29314,
+      "Ġintran": 29315,
+      "ĠTurbulence": 29316,
+      "ĠBianchi": 29317,
+      "CRE": 29318,
+      "ĠNOD": 29319,
+      "apine": 29320,
+      "ĠKane": 29321,
+      "ĠPDGF": 29322,
+      "ĠAxis": 29323,
+      "ĠCausal": 29324,
+      "ĠPoor": 29325,
+      "ĠWords": 29326,
+      "ĠHRV": 29327,
+      "Ġcyanobacteria": 29328,
+      "Ġreminiscent": 29329,
+      "ĠRemarkably": 29330,
+      "heet": 29331,
+      "@@": 29332,
+      "bil": 29333,
+      "Ġdiscriminating": 29334,
+      "ĠBaltic": 29335,
+      "ĠQuebec": 29336,
+      "Ġdefensive": 29337,
+      "âĪ©": 29338,
+      "kr": 29339,
+      "ĠRPE": 29340,
+      "seeking": 29341,
+      "ĠMovie": 29342,
+      "Ġinnovations": 29343,
+      "lept": 29344,
+      "Ġkw": 29345,
+      "Ġtibia": 29346,
+      "Ġneat": 29347,
+      "ytest": 29348,
+      "Ġthinner": 29349,
+      "Ġosteoblasts": 29350,
+      "ĠNorthwest": 29351,
+      "MOS": 29352,
+      "ĠPQ": 29353,
+      "Ġspi": 29354,
+      "Ġresponds": 29355,
+      "Ġhistorically": 29356,
+      "ĠPackage": 29357,
+      "ĠCoastal": 29358,
+      "ĠMississippi": 29359,
+      "ĠPVA": 29360,
+      "pering": 29361,
+      "indole": 29362,
+      "Ġprospectively": 29363,
+      "ĠHemisphere": 29364,
+      "Ġbarely": 29365,
+      "ánchez": 29366,
+      "aggered": 29367,
+      "yptian": 29368,
+      "ĠGest": 29369,
+      "yline": 29370,
+      "Ġphotochemical": 29371,
+      "oscalar": 29372,
+      "porated": 29373,
+      "Ġmetabolomics": 29374,
+      "Ġosteoblast": 29375,
+      "EGFP": 29376,
+      "eriatric": 29377,
+      "DW": 29378,
+      "quest": 29379,
+      "ĠHave": 29380,
+      "Ġspondyl": 29381,
+      "ĠPrimer": 29382,
+      "Ġsinks": 29383,
+      "Ġgaussian": 29384,
+      "ĠKhal": 29385,
+      "Enc": 29386,
+      "ĠAnopheles": 29387,
+      "Thanks": 29388,
+      "Ġconstrued": 29389,
+      "ĠUSS": 29390,
+      "ĠZeeman": 29391,
+      "Ġexported": 29392,
+      "ĠLevi": 29393,
+      "Ġcommander": 29394,
+      "connect": 29395,
+      "Ġnomenclature": 29396,
+      "therefore": 29397,
+      "ulata": 29398,
+      "Ġentrepreneur": 29399,
+      "Ġneuroscience": 29400,
+      "zan": 29401,
+      "Ġextant": 29402,
+      "ATIVE": 29403,
+      "opez": 29404,
+      "Ġenforced": 29405,
+      "ĠInnovation": 29406,
+      "earance": 29407,
+      "Ġimpressive": 29408,
+      "ĠPlac": 29409,
+      "ĠMoz": 29410,
+      "ĠStark": 29411,
+      "Ġrival": 29412,
+      "ĠCapital": 29413,
+      "Ġgranularity": 29414,
+      "Ġdiaphragm": 29415,
+      "utaneous": 29416,
+      "inds": 29417,
+      "Ġphotograph": 29418,
+      "Ġrectangles": 29419,
+      "TGF": 29420,
+      "Ġseaf": 29421,
+      "Ġmaze": 29422,
+      "ĠHW": 29423,
+      "Ġcorrelators": 29424,
+      "Ġdistinguishable": 29425,
+      "Ġconfounders": 29426,
+      "Ġlandslide": 29427,
+      "Ġtoll": 29428,
+      "Ġwastes": 29429,
+      "ĠWF": 29430,
+      "Ġendoc": 29431,
+      "Ġcapsid": 29432,
+      "ecund": 29433,
+      "ĠRBD": 29434,
+      "psin": 29435,
+      "Ġobstetric": 29436,
+      "Ġnanosheets": 29437,
+      "ocols": 29438,
+      "rens": 29439,
+      "ĠSubstituting": 29440,
+      "Ġcustomized": 29441,
+      "Ġresuscitation": 29442,
+      "Ġtubulin": 29443,
+      "ophyte": 29444,
+      "~~~~~~~~": 29445,
+      "plants": 29446,
+      "hicillin": 29447,
+      "halo": 29448,
+      "ruitment": 29449,
+      "ĠConcrete": 29450,
+      "Ġnanorods": 29451,
+      "ĠForms": 29452,
+      "Ġdying": 29453,
+      "discharge": 29454,
+      "Ġwellbeing": 29455,
+      "Ġwarmer": 29456,
+      "ĠSSD": 29457,
+      "ĠAUT": 29458,
+      "ĠConjug": 29459,
+      "Ġjuveniles": 29460,
+      "Ġinevitably": 29461,
+      "ĠMCS": 29462,
+      "approach": 29463,
+      "ĠMason": 29464,
+      "ĠGust": 29465,
+      "ĠThermodynamics": 29466,
+      "Ġpeel": 29467,
+      "ĠTranscriptome": 29468,
+      "Ġindispensable": 29469,
+      "urgery": 29470,
+      "posity": 29471,
+      "Ġpolarizations": 29472,
+      "ĠOthers": 29473,
+      "Ġsandy": 29474,
+      "Ġgliomas": 29475,
+      "Ġpursued": 29476,
+      "VEL": 29477,
+      "Ġrst": 29478,
+      "posium": 29479,
+      "nearest": 29480,
+      "Ġdisseminated": 29481,
+      "ĠMYC": 29482,
+      "Ġaldehyde": 29483,
+      "ĠDiagnostics": 29484,
+      "mans": 29485,
+      "Ġasphal": 29486,
+      "ĠSelect": 29487,
+      "ĠRecon": 29488,
+      "andro": 29489,
+      "DIM": 29490,
+      "Ġfeces": 29491,
+      "illon": 29492,
+      "ĠMALDI": 29493,
+      "nf": 29494,
+      "ĠElim": 29495,
+      "Ġhappy": 29496,
+      "ĠKarl": 29497,
+      "ĠInser": 29498,
+      "Ġinterrog": 29499,
+      "Intern": 29500,
+      "Ġtensorflow": 29501,
+      "Ġhaloes": 29502,
+      "Ġanticipate": 29503,
+      "ĠDPPH": 29504,
+      "rÃŃguez": 29505,
+      "Her": 29506,
+      "anate": 29507,
+      "Ġdressing": 29508,
+      "ĠHoly": 29509,
+      "Ġnewer": 29510,
+      "rides": 29511,
+      "placed": 29512,
+      "inetobacter": 29513,
+      "ĠOccurrence": 29514,
+      "edema": 29515,
+      "ĠIk": 29516,
+      "abad": 29517,
+      "ĠTransitions": 29518,
+      "Ġoutlines": 29519,
+      "Ġcochlear": 29520,
+      "Gy": 29521,
+      "success": 29522,
+      "ĠMEM": 29523,
+      "astype": 29524,
+      "Ġnormalizing": 29525,
+      "Ġterminates": 29526,
+      "Ġsuddenly": 29527,
+      "bbox": 29528,
+      "ĠPul": 29529,
+      "ĠPTP": 29530,
+      "aginal": 29531,
+      "Ġpretrained": 29532,
+      "Ġunreliable": 29533,
+      "ĠGraphical": 29534,
+      "ĠSeyfert": 29535,
+      "Ġcharacterizations": 29536,
+      "Ġtx": 29537,
+      "Ġbicarbonate": 29538,
+      "mathord": 29539,
+      "Ġheritability": 29540,
+      "stackexchange": 29541,
+      "iri": 29542,
+      "âĢĸ": 29543,
+      "ipit": 29544,
+      "attle": 29545,
+      "Ġarena": 29546,
+      "iba": 29547,
+      "ĠAX": 29548,
+      "ĠGPs": 29549,
+      "ophilia": 29550,
+      "SEL": 29551,
+      "osystem": 29552,
+      "ĠâĬ¢": 29553,
+      "ĠNucleus": 29554,
+      "redited": 29555,
+      "ACR": 29556,
+      "ĠAntenna": 29557,
+      "ĠCdc": 29558,
+      "orie": 29559,
+      "Ġequilibration": 29560,
+      "elong": 29561,
+      "stability": 29562,
+      "ĠSchist": 29563,
+      "Ġinjecting": 29564,
+      "hp": 29565,
+      "Ġvitamins": 29566,
+      "Poisson": 29567,
+      "ortal": 29568,
+      "ĠÃĬ": 29569,
+      "ĠÄı": 29570,
+      "Ill": 29571,
+      "Ġutils": 29572,
+      "ов": 29573,
+      "ĠGrom": 29574,
+      "::::": 29575,
+      "ĠGnRH": 29576,
+      "ĠSierra": 29577,
+      "Ġdrafted": 29578,
+      "Ġcapita": 29579,
+      "ships": 29580,
+      "Ġtimestamp": 29581,
+      "Ġsubstituents": 29582,
+      "ĠNotable": 29583,
+      "ĠPurpose": 29584,
+      "inol": 29585,
+      "Ġai": 29586,
+      "Ġfog": 29587,
+      "otone": 29588,
+      "ĠPlaces": 29589,
+      "byshev": 29590,
+      "tiology": 29591,
+      "ription": 29592,
+      "Ġyards": 29593,
+      "ĠXI": 29594,
+      "Ġtechnically": 29595,
+      "GAM": 29596,
+      "ĠABS": 29597,
+      "platform": 29598,
+      "ĠWO": 29599,
+      "PROC": 29600,
+      "Ġreconstit": 29601,
+      "ĠAnomalous": 29602,
+      "ĠBiol": 29603,
+      "Stage": 29604,
+      "ĠReviews": 29605,
+      "Ġrecalling": 29606,
+      "Ġillegal": 29607,
+      "lund": 29608,
+      "¬": 29609,
+      "uthenium": 29610,
+      "ĠPes": 29611,
+      "Ġovaries": 29612,
+      "solutions": 29613,
+      "massive": 29614,
+      "ĠRAW": 29615,
+      "Ġreconnection": 29616,
+      "ĠSusceptibility": 29617,
+      "Ġeconomical": 29618,
+      "cultured": 29619,
+      "ĠSham": 29620,
+      "sqcup": 29621,
+      "Ġpear": 29622,
+      "deposition": 29623,
+      "uchs": 29624,
+      "ĠSaw": 29625,
+      "Ġembolism": 29626,
+      "Bur": 29627,
+      "nar": 29628,
+      "oule": 29629,
+      "Ġtextile": 29630,
+      "seven": 29631,
+      "thio": 29632,
+      "Ġdenoising": 29633,
+      "CEP": 29634,
+      "Ġubiquitination": 29635,
+      "ĠCarlos": 29636,
+      "aP": 29637,
+      "Ġfolder": 29638,
+      "Ġhematological": 29639,
+      "iluminescence": 29640,
+      "ĠFuel": 29641,
+      "icion": 29642,
+      "aculture": 29643,
+      "ARB": 29644,
+      "ĠTravel": 29645,
+      "Func": 29646,
+      "acles": 29647,
+      "ĠInte": 29648,
+      "Ġvacua": 29649,
+      "Ġcocktail": 29650,
+      "ĠInsp": 29651,
+      "Ġcorporate": 29652,
+      "Ġdepicting": 29653,
+      "Ġsprint": 29654,
+      "ĠmTORC": 29655,
+      "Ġcimg": 29656,
+      "ocarbon": 29657,
+      "ĠDave": 29658,
+      "ĠGb": 29659,
+      "iji": 29660,
+      "targeting": 29661,
+      "Ġsequestration": 29662,
+      "Bri": 29663,
+      "IGF": 29664,
+      "Ġanalytics": 29665,
+      "ĠAcinetobacter": 29666,
+      "gets": 29667,
+      "MPS": 29668,
+      "ogluc": 29669,
+      "Cent": 29670,
+      "Ġverbs": 29671,
+      "Ġinductance": 29672,
+      "diagram": 29673,
+      "Ġrecalled": 29674,
+      "Ġcosme": 29675,
+      "Ġautomotive": 29676,
+      "ĠPDEs": 29677,
+      "ĠReid": 29678,
+      "Ġadapter": 29679,
+      "ĠOliver": 29680,
+      "Ġavalanche": 29681,
+      "Vir": 29682,
+      "ĠToxicity": 29683,
+      "ĠLeu": 29684,
+      "Conclusions": 29685,
+      "Ġtetragonal": 29686,
+      "ĠDMF": 29687,
+      "umannii": 29688,
+      "ĠRequirements": 29689,
+      "toc": 29690,
+      "ité": 29691,
+      "Ġcontinent": 29692,
+      "ĠHank": 29693,
+      "ĠDefinitions": 29694,
+      "GPU": 29695,
+      "origin": 29696,
+      "Ġdichro": 29697,
+      "Mus": 29698,
+      "Ġbival": 29699,
+      "Ġimpulsive": 29700,
+      "Ġassemble": 29701,
+      "Ġpipes": 29702,
+      "docs": 29703,
+      "Ġexchanger": 29704,
+      "Ġallograft": 29705,
+      "loyd": 29706,
+      "ĠÌĭ": 29707,
+      "Ġantenatal": 29708,
+      "Ġgrassland": 29709,
+      "Ġhystere": 29710,
+      "ĠAntigen": 29711,
+      "ĠGeneric": 29712,
+      "ĠTuring": 29713,
+      "ĠExcell": 29714,
+      "ĠHein": 29715,
+      "aja": 29716,
+      "uminum": 29717,
+      "citabine": 29718,
+      "facial": 29719,
+      "iteration": 29720,
+      "Ġslurry": 29721,
+      "AML": 29722,
+      "ergetic": 29723,
+      "ĠTHF": 29724,
+      "Ġkilometers": 29725,
+      "fg": 29726,
+      "educ": 29727,
+      "idian": 29728,
+      "Ġpredicates": 29729,
+      "Ġradios": 29730,
+      "ĠPeri": 29731,
+      "ĠShell": 29732,
+      "Ġarcsec": 29733,
+      "Ġstriatal": 29734,
+      "Ġceiling": 29735,
+      "olithic": 29736,
+      "Ġexhaustion": 29737,
+      "PUT": 29738,
+      "thers": 29739,
+      "ymp": 29740,
+      "ĠQian": 29741,
+      "ĠProgressive": 29742,
+      "Ġwel": 29743,
+      "ĠConvention": 29744,
+      "ĠCurie": 29745,
+      "ĠMans": 29746,
+      "ĠNova": 29747,
+      "ĠWells": 29748,
+      "dew": 29749,
+      "Standard": 29750,
+      "realistic": 29751,
+      "transpose": 29752,
+      "serial": 29753,
+      "ĠTx": 29754,
+      "ĠAMR": 29755,
+      "Ġindeterm": 29756,
+      "ĠLiouville": 29757,
+      "hookrightarrow": 29758,
+      "ARs": 29759,
+      "Ġbaseball": 29760,
+      "acious": 29761,
+      "agnetization": 29762,
+      "estimate": 29763,
+      "ĠPAS": 29764,
+      "Ġmeals": 29765,
+      "multiple": 29766,
+      "ĠBiomarkers": 29767,
+      "Wide": 29768,
+      "ĠTomography": 29769,
+      "////////////////////////////////": 29770,
+      "Ġresins": 29771,
+      "Ġanywhere": 29772,
+      "INC": 29773,
+      "ĠTeaching": 29774,
+      "ĠSamuel": 29775,
+      "Ġhallmark": 29776,
+      "ĠThyroid": 29777,
+      "othi": 29778,
+      "Ġconstraining": 29779,
+      "ĠBarrett": 29780,
+      "ĠErrors": 29781,
+      "Cole": 29782,
+      "sharing": 29783,
+      "HDL": 29784,
+      "Effect": 29785,
+      "ĠTolerance": 29786,
+      "Ġstressful": 29787,
+      "ĠBalance": 29788,
+      "ĠTech": 29789,
+      "Ġvalleys": 29790,
+      "setup": 29791,
+      "ĠRadical": 29792,
+      "ĠMacrophages": 29793,
+      "Ġinterrupt": 29794,
+      "Ġdiatom": 29795,
+      "colored": 29796,
+      "Ġpyrid": 29797,
+      "FDG": 29798,
+      "æ": 29799,
+      "Ġreared": 29800,
+      "ĠRating": 29801,
+      "Ġopaque": 29802,
+      "package": 29803,
+      "Ġnasopharyngeal": 29804,
+      "Ġpreconditioning": 29805,
+      "Diptera": 29806,
+      "ĠMing": 29807,
+      "ĠCaro": 29808,
+      "ĠImmunity": 29809,
+      "rifuge": 29810,
+      "ĠObjectives": 29811,
+      "ghan": 29812,
+      "uccin": 29813,
+      "ĠFors": 29814,
+      "ĠFITC": 29815,
+      "Ġseats": 29816,
+      "ĠImpaired": 29817,
+      "Ġreefs": 29818,
+      "emaker": 29819,
+      "Ġoffices": 29820,
+      "Ġaccepting": 29821,
+      "ĠTRAN": 29822,
+      "ĠTargets": 29823,
+      "Ġcorrelator": 29824,
+      "Ġsupercapac": 29825,
+      "inburgh": 29826,
+      "Ġcollider": 29827,
+      "Ġenteric": 29828,
+      "ĠSTRUCTURE": 29829,
+      "Ġminister": 29830,
+      "ĠArchae": 29831,
+      "Loop": 29832,
+      "ĠASA": 29833,
+      "Ġcontacted": 29834,
+      "Ġhistidine": 29835,
+      "folded": 29836,
+      "Search": 29837,
+      "Ġrespects": 29838,
+      "ĠATF": 29839,
+      "Ġtrouble": 29840,
+      "Ġprevailing": 29841,
+      "Cp": 29842,
+      "ĠTCM": 29843,
+      "ĠSpinal": 29844,
+      "Ġguides": 29845,
+      "evitable": 29846,
+      "Ġbrick": 29847,
+      "strings": 29848,
+      "ĠHungary": 29849,
+      "Ġeps": 29850,
+      "entricular": 29851,
+      "Specifically": 29852,
+      "ando": 29853,
+      "issues": 29854,
+      "osomiasis": 29855,
+      "kDa": 29856,
+      "Ġaside": 29857,
+      "Ġadenine": 29858,
+      "Ġmotivate": 29859,
+      "stratig": 29860,
+      "BLE": 29861,
+      "ĠDeposition": 29862,
+      "motor": 29863,
+      "ĠHers": 29864,
+      "Ġnebul": 29865,
+      "ĠBarrier": 29866,
+      "Unlike": 29867,
+      "Ġballistic": 29868,
+      "Ġsouthwestern": 29869,
+      "ĠMontreal": 29870,
+      "Scan": 29871,
+      "Ġmould": 29872,
+      "Ġinterrup": 29873,
+      "smallmatrix": 29874,
+      "Ġelaborated": 29875,
+      "ucks": 29876,
+      "APS": 29877,
+      "ĠConsumption": 29878,
+      "capacity": 29879,
+      "innitus": 29880,
+      "Ġgovernance": 29881,
+      "Ġpalsy": 29882,
+      "Ġsubmission": 29883,
+      "Ġtemple": 29884,
+      "ĠIIA": 29885,
+      "methionine": 29886,
+      "Ġkerat": 29887,
+      "Ġridges": 29888,
+      "Promega": 29889,
+      "cols": 29890,
+      "ISP": 29891,
+      "Ġapnea": 29892,
+      "ĠFlat": 29893,
+      "ĠEpigenetic": 29894,
+      "Ġparish": 29895,
+      "ĠParametric": 29896,
+      "dash": 29897,
+      "future": 29898,
+      "rise": 29899,
+      "Ġcontracting": 29900,
+      "algia": 29901,
+      "Ġgoto": 29902,
+      "stadt": 29903,
+      "Ġfabricate": 29904,
+      "Ġdimerization": 29905,
+      "dump": 29906,
+      "ĠLyn": 29907,
+      "Ġrecycled": 29908,
+      "posedness": 29909,
+      "ĠSensory": 29910,
+      "ïĿ": 29911,
+      "ĠWet": 29912,
+      "Ġdiethyl": 29913,
+      "Ġblades": 29914,
+      "Ġtimed": 29915,
+      "Ġkeyword": 29916,
+      "Ġpolytope": 29917,
+      "ĠGot": 29918,
+      "Ġapproximates": 29919,
+      "Without": 29920,
+      "ĠBere": 29921,
+      "ĠLp": 29922,
+      "oplasty": 29923,
+      "ĠFibr": 29924,
+      "modulated": 29925,
+      "ĠARM": 29926,
+      "Ġunderestimate": 29927,
+      "ĠCBS": 29928,
+      "ĠLectures": 29929,
+      "uncan": 29930,
+      "ĠSeismic": 29931,
+      "Soft": 29932,
+      "Ġzooplankton": 29933,
+      "Ġencephalopathy": 29934,
+      "ĠSSA": 29935,
+      "ĠCros": 29936,
+      "ĠHann": 29937,
+      "Ġshuffle": 29938,
+      "scription": 29939,
+      "ĠRevers": 29940,
+      "Studies": 29941,
+      "Ġsocially": 29942,
+      "Ġsubcl": 29943,
+      "ĠYong": 29944,
+      "ogh": 29945,
+      "Ġïģ³": 29946,
+      "UDY": 29947,
+      "ĠHaar": 29948,
+      "ĠDoctor": 29949,
+      "Ġintakes": 29950,
+      "Ġbarrel": 29951,
+      "ĠTRPV": 29952,
+      "ĠAggreg": 29953,
+      "nyi": 29954,
+      "tuned": 29955,
+      "acquired": 29956,
+      "Ġhook": 29957,
+      "FGF": 29958,
+      "«": 29959,
+      "ĠInjection": 29960,
+      "Ġgravel": 29961,
+      "Ġmicrog": 29962,
+      "Ġmenstrual": 29963,
+      "Feature": 29964,
+      "IRE": 29965,
+      "uu": 29966,
+      "ĠSrc": 29967,
+      "ĠStore": 29968,
+      "Ġinitiator": 29969,
+      "PSO": 29970,
+      "Ġepileptic": 29971,
+      "Ġcingulate": 29972,
+      "IJ": 29973,
+      "Row": 29974,
+      "Ġsinging": 29975,
+      "ĠMethan": 29976,
+      "ĠAldrich": 29977,
+      "Ġtremendous": 29978,
+      "amining": 29979,
+      "Ġtracts": 29980,
+      "Ġâİ£": 29981,
+      "klah": 29982,
+      "Div": 29983,
+      "indol": 29984,
+      "Ġindole": 29985,
+      "exper": 29986,
+      "Ġglycer": 29987,
+      "Ġbenzyl": 29988,
+      "Ġworsening": 29989,
+      "Ġunambiguous": 29990,
+      "uart": 29991,
+      "Ġparsim": 29992,
+      "ricks": 29993,
+      "Ġtrail": 29994,
+      "ĠBlanc": 29995,
+      "Ġaminotransferase": 29996,
+      "ĠDOC": 29997,
+      "Ġfumig": 29998,
+      "idic": 29999,
+      "ĠConsequences": 30000,
+      "Ġacidification": 30001,
+      "ĠCIFAR": 30002,
+      "ĠDatasets": 30003,
+      "ĠAMI": 30004,
+      "Ġexplants": 30005,
+      "ĠDiverse": 30006,
+      "Ġdephasing": 30007,
+      "Ġparliament": 30008,
+      "ipient": 30009,
+      "Ġhoneycomb": 30010,
+      "heavy": 30011,
+      "Ġwatermark": 30012,
+      "MED": 30013,
+      "datasets": 30014,
+      "waters": 30015,
+      "Provid": 30016,
+      "interpret": 30017,
+      "rovirus": 30018,
+      "Io": 30019,
+      "RAD": 30020,
+      "Ġlunar": 30021,
+      "Ġweaning": 30022,
+      "Ġsensorimotor": 30023,
+      "uca": 30024,
+      "Ġinfect": 30025,
+      "ĠUnique": 30026,
+      "GRP": 30027,
+      "QoL": 30028,
+      "ospec": 30029,
+      "Ġforwarding": 30030,
+      "Estim": 30031,
+      "ÅĦski": 30032,
+      "ĠMs": 30033,
+      "achn": 30034,
+      "Ġrota": 30035,
+      "Ġappointment": 30036,
+      "ĠMedal": 30037,
+      "Ġadenovirus": 30038,
+      "quinol": 30039,
+      "Ġdeuterium": 30040,
+      "tep": 30041,
+      "ĠStyle": 30042,
+      "Nd": 30043,
+      "ayama": 30044,
+      "ĠHamm": 30045,
+      "ĠSpecification": 30046,
+      "vability": 30047,
+      "tha": 30048,
+      "Ġjitter": 30049,
+      "Ġâݦ": 30050,
+      "aqu": 30051,
+      "wire": 30052,
+      "Ġclassically": 30053,
+      "Ġsuperpotential": 30054,
+      "ĠSpecim": 30055,
+      "ĠVariance": 30056,
+      "Ġalbums": 30057,
+      "ĠSenior": 30058,
+      "Ġneurotransmitter": 30059,
+      "ĠRecombinant": 30060,
+      "DCS": 30061,
+      "vl": 30062,
+      "Ġpf": 30063,
+      "Ġinevitable": 30064,
+      "ĠNick": 30065,
+      "Ġmanipulating": 30066,
+      "ituximab": 30067,
+      "ceiver": 30068,
+      "ĠBren": 30069,
+      "ĠRace": 30070,
+      "Ġretarded": 30071,
+      "modulin": 30072,
+      "Clinical": 30073,
+      "Ġneurologic": 30074,
+      "ĠRegiment": 30075,
+      "Ġzoom": 30076,
+      "ĠOrthogonal": 30077,
+      "ĠConcerning": 30078,
+      "ĠJurassic": 30079,
+      "ĠArtem": 30080,
+      "ĠMelbourne": 30081,
+      "bins": 30082,
+      "jl": 30083,
+      "Ġinhab": 30084,
+      "Ġsqrt": 30085,
+      "Ġsemisimple": 30086,
+      "astric": 30087,
+      "ĠProxim": 30088,
+      "ĠVariants": 30089,
+      "Ġaesthetic": 30090,
+      "Ġsummarised": 30091,
+      "ĠBecker": 30092,
+      "OCH": 30093,
+      "dale": 30094,
+      "Ġmounting": 30095,
+      "andering": 30096,
+      "Ġsoftmax": 30097,
+      "Ġneuroinflammation": 30098,
+      "Ġesophagus": 30099,
+      "operators": 30100,
+      "ĠADAM": 30101,
+      "Ġviolate": 30102,
+      "ĠPHY": 30103,
+      "ede": 30104,
+      "ĠCher": 30105,
+      "orsal": 30106,
+      "Ġmetamorphic": 30107,
+      "ĠICM": 30108,
+      "ĠAbcam": 30109,
+      "slot": 30110,
+      "serine": 30111,
+      "Ġduplicates": 30112,
+      "ĠMEMS": 30113,
+      "ĠAbl": 30114,
+      "ĠChel": 30115,
+      "ĠAuthority": 30116,
+      "Ġgeo": 30117,
+      "Ġhomeomorphism": 30118,
+      "Ġimmunomodulatory": 30119,
+      "ĠTU": 30120,
+      "ĠKT": 30121,
+      "aterally": 30122,
+      "oxides": 30123,
+      "tebral": 30124,
+      "Ġcataract": 30125,
+      "leaved": 30126,
+      "igu": 30127,
+      "ateur": 30128,
+      "ĠRé": 30129,
+      "Ġdiscoveries": 30130,
+      "boson": 30131,
+      "ocated": 30132,
+      "jpg": 30133,
+      "ĠSato": 30134,
+      "ĠPROP": 30135,
+      "ĠImplement": 30136,
+      "ELISA": 30137,
+      "iqueness": 30138,
+      "Ġsymbion": 30139,
+      "ĠFaraday": 30140,
+      "ĠPPARγ": 30141,
+      "witz": 30142,
+      "reward": 30143,
+      "ĠBush": 30144,
+      "stressed": 30145,
+      "ĠAbor": 30146,
+      "Ġairways": 30147,
+      "Ġinterferometry": 30148,
+      "Circ": 30149,
+      "Ġimmunoprecipitation": 30150,
+      "ĠApache": 30151,
+      "rophosph": 30152,
+      "ĠoC": 30153,
+      "Ġfrog": 30154,
+      "ĠGU": 30155,
+      "ffe": 30156,
+      "ĠStro": 30157,
+      "Ġdodecyl": 30158,
+      "dan": 30159,
+      "folds": 30160,
+      "ĠMust": 30161,
+      "Ġsurroundings": 30162,
+      "Ġcodons": 30163,
+      "onda": 30164,
+      "tb": 30165,
+      "odge": 30166,
+      "avas": 30167,
+      "ĠSeason": 30168,
+      "tude": 30169,
+      "ĠPlasticity": 30170,
+      "ĠHawaii": 30171,
+      "DEG": 30172,
+      "ĠCMD": 30173,
+      "Ġsingleton": 30174,
+      "keley": 30175,
+      "Ġalgebraically": 30176,
+      "Ġnanostructured": 30177,
+      "easible": 30178,
+      "Ġoverlooked": 30179,
+      "ĠPulse": 30180,
+      "romechanical": 30181,
+      "ĠElse": 30182,
+      "Ġexcitons": 30183,
+      "ĠConstrained": 30184,
+      "Ġcohesion": 30185,
+      "Ġrealizing": 30186,
+      "ĠRadiative": 30187,
+      "Ġtrypan": 30188,
+      "xs": 30189,
+      "ĠTas": 30190,
+      "Ġmainstream": 30191,
+      "Ġcompactly": 30192,
+      "growing": 30193,
+      "esc": 30194,
+      "ĠdN": 30195,
+      "ĠSignatures": 30196,
+      "ĠFundamentals": 30197,
+      "Ġexpose": 30198,
+      "ĠRang": 30199,
+      "Ġhanded": 30200,
+      "Ġfunctionalization": 30201,
+      "Ġpassiv": 30202,
+      "altern": 30203,
+      "agul": 30204,
+      "Ġschematically": 30205,
+      "OW": 30206,
+      "ĠÖ": 30207,
+      "ĠPOD": 30208,
+      "Ġhear": 30209,
+      "ymore": 30210,
+      "ĠPremier": 30211,
+      "South": 30212,
+      "Ä«": 30213,
+      "ĠOBS": 30214,
+      "ĠAlg": 30215,
+      "glia": 30216,
+      "ĠTransmembrane": 30217,
+      "Ġspheroids": 30218,
+      "ĠRHS": 30219,
+      "Ġinches": 30220,
+      "ĠKato": 30221,
+      "Ġie": 30222,
+      "ĠCommercial": 30223,
+      "Ġanalytes": 30224,
+      "Ġrisky": 30225,
+      "Ġpiston": 30226,
+      "ĠMarkovian": 30227,
+      "Ġdrama": 30228,
+      "Ġci": 30229,
+      "ĠHistological": 30230,
+      "Ġactuation": 30231,
+      "discrete": 30232,
+      "carbamoyl": 30233,
+      "SMA": 30234,
+      "Ġfeeds": 30235,
+      "Ġneoplasia": 30236,
+      "ĠController": 30237,
+      "been": 30238,
+      "glutamine": 30239,
+      "injected": 30240,
+      "Ġcrab": 30241,
+      "ĠCauses": 30242,
+      "ĠStory": 30243,
+      "Ġvanadium": 30244,
+      "ĠTitan": 30245,
+      "enix": 30246,
+      "assign": 30247,
+      "Ġimmunogenicity": 30248,
+      "ĠApparent": 30249,
+      "Ġenhancers": 30250,
+      "ĠSou": 30251,
+      "alloy": 30252,
+      "mathbin": 30253,
+      "Ġsedation": 30254,
+      "ĠWorkshop": 30255,
+      "gover": 30256,
+      "lst": 30257,
+      "Ġupwelling": 30258,
+      "mez": 30259,
+      "Ġpolypropylene": 30260,
+      "ĠColorectal": 30261,
+      "ĠRelaxation": 30262,
+      "Ġfragile": 30263,
+      "Äĥ": 30264,
+      "Ġsubgraphs": 30265,
+      "theoretical": 30266,
+      "Operator": 30267,
+      "lywood": 30268,
+      "awn": 30269,
+      "ĠPercentage": 30270,
+      "methylation": 30271,
+      "corrhizal": 30272,
+      "Grad": 30273,
+      "dens": 30274,
+      "ĠHα": 30275,
+      "Ġupcoming": 30276,
+      "Ġvirgin": 30277,
+      "Names": 30278,
+      "ĠRyd": 30279,
+      "Ġâݤ": 30280,
+      "phosphorylation": 30281,
+      "renewal": 30282,
+      "Year": 30283,
+      "Init": 30284,
+      "Ġselling": 30285,
+      "ĠMASS": 30286,
+      "rophin": 30287,
+      "ijn": 30288,
+      "Conversely": 30289,
+      "Ġuniversally": 30290,
+      "orhombic": 30291,
+      "Ġunpredictable": 30292,
+      "Fock": 30293,
+      "chair": 30294,
+      "ivas": 30295,
+      "networks": 30296,
+      "Ġterritories": 30297,
+      "thia": 30298,
+      "ĠAmplification": 30299,
+      "March": 30300,
+      "Ġflam": 30301,
+      "ĠChart": 30302,
+      "Ġshortage": 30303,
+      "AMET": 30304,
+      "Ġgrape": 30305,
+      "Ġvoltammetry": 30306,
+      "د": 30307,
+      "ĠSCH": 30308,
+      "Ġepithel": 30309,
+      "ĠChromosome": 30310,
+      "ĠXL": 30311,
+      "ĠPersistent": 30312,
+      "Ġtraveled": 30313,
+      "Ġmeridional": 30314,
+      "Ġfprintf": 30315,
+      "Ġgum": 30316,
+      "visory": 30317,
+      "Unfortunately": 30318,
+      "Ġanteced": 30319,
+      "Ġfrictional": 30320,
+      "DAT": 30321,
+      "acl": 30322,
+      "ĠPregnancy": 30323,
+      "ĠBZ": 30324,
+      "regulatory": 30325,
+      "stimulating": 30326,
+      "Japan": 30327,
+      "machine": 30328,
+      "uti": 30329,
+      "ĠLer": 30330,
+      "Ġnanoflu": 30331,
+      "prototype": 30332,
+      "identification": 30333,
+      "klahoma": 30334,
+      "ĠEmploy": 30335,
+      "Schwarz": 30336,
+      "Ġincorrectly": 30337,
+      "atto": 30338,
+      "rization": 30339,
+      "ismuth": 30340,
+      "Ġiris": 30341,
+      "imentary": 30342,
+      "Ġinflationary": 30343,
+      "Ġoutflows": 30344,
+      "ĠLic": 30345,
+      "oreductase": 30346,
+      "Ġproceeding": 30347,
+      "ĠTAC": 30348,
+      "ĠHTL": 30349,
+      "Ġresides": 30350,
+      "stral": 30351,
+      "ĠTransf": 30352,
+      "Ġdichotom": 30353,
+      "Filter": 30354,
+      "June": 30355,
+      "isure": 30356,
+      "ĠAde": 30357,
+      "Ġijk": 30358,
+      "ĠPhilos": 30359,
+      "Ġstayed": 30360,
+      "Ġtamoxifen": 30361,
+      "Ġasparagine": 30362,
+      "exception": 30363,
+      "Ġaccumulating": 30364,
+      "astro": 30365,
+      "Change": 30366,
+      "uzi": 30367,
+      "Ġlon": 30368,
+      "Instead": 30369,
+      "Ġcentrally": 30370,
+      "ĠDental": 30371,
+      "classified": 30372,
+      "ĠEgyptian": 30373,
+      "Address": 30374,
+      "ĠQuaternary": 30375,
+      "ĠUSP": 30376,
+      "coin": 30377,
+      "Ġembryogenesis": 30378,
+      "ïĢ¨": 30379,
+      "Null": 30380,
+      "ĠMixing": 30381,
+      "intensive": 30382,
+      "Ġnormative": 30383,
+      "ĠLef": 30384,
+      "Ġrumen": 30385,
+      "ĠThai": 30386,
+      "Ġswallow": 30387,
+      "Component": 30388,
+      "Ġrobotics": 30389,
+      "ĠCad": 30390,
+      "ĠCIP": 30391,
+      "ĠAcids": 30392,
+      "ĠOffic": 30393,
+      "urer": 30394,
+      "ĠWick": 30395,
+      "Ġkink": 30396,
+      "ĠScha": 30397,
+      "ĠCharacteristic": 30398,
+      "families": 30399,
+      "ĠGCs": 30400,
+      "ĠOptimizing": 30401,
+      "Ġtimer": 30402,
+      "él": 30403,
+      "jin": 30404,
+      "reversal": 30405,
+      "Ġsandstone": 30406,
+      "HN": 30407,
+      "tk": 30408,
+      "Ġptr": 30409,
+      "Ġmonochromatic": 30410,
+      "Ġfeedforward": 30411,
+      "dington": 30412,
+      "Ġcriticism": 30413,
+      "Ġsig": 30414,
+      "Ġpace": 30415,
+      "ĠTK": 30416,
+      "ĠWas": 30417,
+      "Ġcertificate": 30418,
+      "Ġstuck": 30419,
+      "Ġcorrid": 30420,
+      "Ġlocalisation": 30421,
+      "Ġsilk": 30422,
+      "Ġdigest": 30423,
+      "ĠTemple": 30424,
+      "ĠPosterior": 30425,
+      "Ġcommutator": 30426,
+      "tsch": 30427,
+      "perme": 30428,
+      "ysed": 30429,
+      "Ġmenu": 30430,
+      "Ġmidw": 30431,
+      "ocatalytic": 30432,
+      "Ġppb": 30433,
+      "Types": 30434,
+      "arri": 30435,
+      "ĠLOD": 30436,
+      "Ġloan": 30437,
+      "secret": 30438,
+      "Ġcarbons": 30439,
+      "ĠHolog": 30440,
+      "olipids": 30441,
+      "Ġuplo": 30442,
+      "ĠDNase": 30443,
+      "Ġpuzzle": 30444,
+      "Ġstance": 30445,
+      "ĠManchester": 30446,
+      "ĠDetector": 30447,
+      "ims": 30448,
+      "ĠTerms": 30449,
+      "ĠPGC": 30450,
+      "Ġincidents": 30451,
+      "ieh": 30452,
+      "ĠIDs": 30453,
+      "ĠAhmad": 30454,
+      "Ġnights": 30455,
+      "Ġbiomo": 30456,
+      "ĠMethylation": 30457,
+      "uator": 30458,
+      "resize": 30459,
+      "ĠFinger": 30460,
+      "ĠWo": 30461,
+      "Ġposter": 30462,
+      "Ġsolidification": 30463,
+      "ĠValidity": 30464,
+      "ĠDendritic": 30465,
+      "Ġadherent": 30466,
+      "issions": 30467,
+      "inction": 30468,
+      "Ġantagonistic": 30469,
+      "ĠPreliminaries": 30470,
+      "Ġcoval": 30471,
+      "Ġmovies": 30472,
+      "Ġbudding": 30473,
+      "Kn": 30474,
+      "ĠGit": 30475,
+      "ĠThereafter": 30476,
+      "Ġcapacitive": 30477,
+      "Az": 30478,
+      "ĠTLS": 30479,
+      "Ġinitiates": 30480,
+      "ĠDMR": 30481,
+      "Ġâī«": 30482,
+      "ĠMyocardial": 30483,
+      "ĠRotation": 30484,
+      "CONFIG": 30485,
+      "Ġvowel": 30486,
+      "Ġolivine": 30487,
+      "Hamiltonian": 30488,
+      "Ġstalk": 30489,
+      "Neu": 30490,
+      "Rest": 30491,
+      "anical": 30492,
+      "Ġdst": 30493,
+      "Ġresh": 30494,
+      "Ġexpressive": 30495,
+      "Ġinfectivity": 30496,
+      "oku": 30497,
+      "CTL": 30498,
+      "Frequency": 30499,
+      "Ġpremise": 30500,
+      "Walk": 30501,
+      "ĠâĹ": 30502,
+      "Ġrelapsed": 30503,
+      "tured": 30504,
+      "ĠUML": 30505,
+      "ovan": 30506,
+      "ĠResearchers": 30507,
+      "Ġconveniently": 30508,
+      "usk": 30509,
+      "INIT": 30510,
+      "Eqs": 30511,
+      "Factory": 30512,
+      "Ġunsteady": 30513,
+      "ĠAnsw": 30514,
+      "Ala": 30515,
+      "nitine": 30516,
+      "qp": 30517,
+      "ulous": 30518,
+      "research": 30519,
+      "ĠBrom": 30520,
+      "ĠDemoc": 30521,
+      "configuration": 30522,
+      "ulosic": 30523,
+      "Ġfra": 30524,
+      "Ġgift": 30525,
+      "Third": 30526,
+      "Claim": 30527,
+      "ÄŁ": 30528,
+      "odiazep": 30529,
+      "Ġprox": 30530,
+      "ocystis": 30531,
+      "ĠRPA": 30532,
+      "ĠLikert": 30533,
+      "RMS": 30534,
+      "tech": 30535,
+      "Ġacous": 30536,
+      "TLR": 30537,
+      "buck": 30538,
+      "ĠTherap": 30539,
+      "ussions": 30540,
+      "helor": 30541,
+      "ĠEmotion": 30542,
+      "bird": 30543,
+      "Ġthio": 30544,
+      "Ġquantitation": 30545,
+      "bracket": 30546,
+      "Ġpercept": 30547,
+      "Ġsubcategory": 30548,
+      "Ġlightning": 30549,
+      "Ġhernia": 30550,
+      "Ġneurotrophic": 30551,
+      "SDS": 30552,
+      "ĠAnders": 30553,
+      "Ġslowing": 30554,
+      "strongly": 30555,
+      "ĠCounting": 30556,
+      "ĠIncluding": 30557,
+      "ductions": 30558,
+      "ubated": 30559,
+      "ĠStorm": 30560,
+      "correlated": 30561,
+      "Ġautoantibodies": 30562,
+      "ĠMerg": 30563,
+      "ocer": 30564,
+      "micutes": 30565,
+      "Ġnonlinearities": 30566,
+      "ĠCentury": 30567,
+      "ĠLandscape": 30568,
+      "ĠDerivatives": 30569,
+      "ĠContrary": 30570,
+      "Ġcompile": 30571,
+      "ĠHepatic": 30572,
+      "Ġponds": 30573,
+      "Ġorganize": 30574,
+      "DMSO": 30575,
+      "Position": 30576,
+      "Ġbrach": 30577,
+      "Ġinflat": 30578,
+      "ospace": 30579,
+      "Ġskewness": 30580,
+      "Ġagitation": 30581,
+      "ĠHOMO": 30582,
+      "EU": 30583,
+      "Ġcommented": 30584,
+      "Ġcorpora": 30585,
+      "Ġmalt": 30586,
+      "Hermitian": 30587,
+      "iday": 30588,
+      "ĠHelmholtz": 30589,
+      "roblast": 30590,
+      "ĠCTR": 30591,
+      "unching": 30592,
+      "ĠMond": 30593,
+      "ĠComment": 30594,
+      "Ġosteosarcoma": 30595,
+      "posterior": 30596,
+      "Ġthymus": 30597,
+      "Ġcigarettes": 30598,
+      "NW": 30599,
+      "olem": 30600,
+      "ĠHox": 30601,
+      "ĠNFL": 30602,
+      "ĠAvailable": 30603,
+      "ĠSiber": 30604,
+      "ĠFeld": 30605,
+      "Ġborderline": 30606,
+      "Ġbeats": 30607,
+      "Ġorganised": 30608,
+      "Ġdistinguishes": 30609,
+      "Ġdialog": 30610,
+      "ĠBerger": 30611,
+      "oleic": 30612,
+      "Ġnumbered": 30613,
+      "Ġreachable": 30614,
+      "ĠRobertson": 30615,
+      "ĠChamber": 30616,
+      "ndarray": 30617,
+      "Ġcytoskeletal": 30618,
+      "Ġblending": 30619,
+      "blood": 30620,
+      "Import": 30621,
+      "Ġoverwhelming": 30622,
+      "Ġio": 30623,
+      "Ġoutage": 30624,
+      "ĠScholar": 30625,
+      "placing": 30626,
+      "ĠPolyp": 30627,
+      "Decl": 30628,
+      "ĠMEDLINE": 30629,
+      "ĠKM": 30630,
+      "ĠDAP": 30631,
+      "errors": 30632,
+      "ĠSHR": 30633,
+      "ĠDex": 30634,
+      "ĠGAS": 30635,
+      "ĠGian": 30636,
+      "Ġclinicopathological": 30637,
+      "Ġïģ·": 30638,
+      "ĠPredictions": 30639,
+      "ĠQuadratic": 30640,
+      "Ġarrhythmias": 30641,
+      "arid": 30642,
+      "Ġclothing": 30643,
+      "ĠFracture": 30644,
+      "ĉĠĠĠĠĠ": 30645,
+      "addy": 30646,
+      "ĠAlberta": 30647,
+      "ĠWed": 30648,
+      "phire": 30649,
+      "ĠEncryp": 30650,
+      "ĠLAB": 30651,
+      "ĠFano": 30652,
+      "CTT": 30653,
+      "Ġoryz": 30654,
+      "iliac": 30655,
+      "ĠLiao": 30656,
+      "versus": 30657,
+      "Ġmeso": 30658,
+      "Ġmidpoint": 30659,
+      "Ġstator": 30660,
+      "ĠJenn": 30661,
+      "ovsky": 30662,
+      "Ġuncover": 30663,
+      "erenn": 30664,
+      "ĠMcM": 30665,
+      "âīĪ": 30666,
+      "ĠCircuits": 30667,
+      "Ġfetuses": 30668,
+      "Ġagglomer": 30669,
+      "Ġfb": 30670,
+      "Ġyy": 30671,
+      "atech": 30672,
+      "ARG": 30673,
+      "Ġbaumannii": 30674,
+      "Ġellipsoid": 30675,
+      "Ġloses": 30676,
+      "Ġunve": 30677,
+      "Ġbutt": 30678,
+      "Ġmulticentre": 30679,
+      "iline": 30680,
+      "Ġresort": 30681,
+      "Ġcerebrovascular": 30682,
+      "ĠDecreased": 30683,
+      "jud": 30684,
+      "sus": 30685,
+      "amol": 30686,
+      "constraints": 30687,
+      "Ġteen": 30688,
+      "ĠPassive": 30689,
+      "ĠCaucasian": 30690,
+      "Ġcran": 30691,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 30692,
+      "ün": 30693,
+      "ĠDNMT": 30694,
+      "Ġterror": 30695,
+      "adrenal": 30696,
+      "Ġangiogenic": 30697,
+      "ĠInhibitory": 30698,
+      "prag": 30699,
+      "Ġcob": 30700,
+      "elsh": 30701,
+      "Ġenhancements": 30702,
+      "ĠShaw": 30703,
+      "ĠTakahashi": 30704,
+      "Ġsulphur": 30705,
+      "Ġgravitation": 30706,
+      "ĠPVDF": 30707,
+      "must": 30708,
+      "¢": 30709,
+      "asymptotic": 30710,
+      "elman": 30711,
+      "ĠPros": 30712,
+      "ĠMAD": 30713,
+      "ĠLen": 30714,
+      "therapy": 30715,
+      "efully": 30716,
+      "sulfur": 30717,
+      "ĠTCA": 30718,
+      "additive": 30719,
+      "talk": 30720,
+      "Ġpiglets": 30721,
+      "Ġprospect": 30722,
+      "ecundity": 30723,
+      "ĠXiang": 30724,
+      "handler": 30725,
+      "Ġclath": 30726,
+      "Ġmillimeter": 30727,
+      "jar": 30728,
+      "Ġbiophysical": 30729,
+      "Ġcomplexities": 30730,
+      "ĠHerb": 30731,
+      "Ġrecovers": 30732,
+      "ĠVincent": 30733,
+      "ĠPuerto": 30734,
+      "Earth": 30735,
+      "RAM": 30736,
+      "Ġcables": 30737,
+      "designed": 30738,
+      "ĠOscillation": 30739,
+      "Ġmeiosis": 30740,
+      "Ġfleet": 30741,
+      "ĠHuntington": 30742,
+      "ĠBeg": 30743,
+      "ĠECs": 30744,
+      "ĠAntic": 30745,
+      "Ġpractitioner": 30746,
+      "cultural": 30747,
+      "kat": 30748,
+      "Ġrecoil": 30749,
+      "ĠImplicit": 30750,
+      "Ġsummaries": 30751,
+      "Ġdiscontinued": 30752,
+      "Ġencompassing": 30753,
+      "ĠAltogether": 30754,
+      "ĠDIST": 30755,
+      "Ġconstellation": 30756,
+      "ĠExisting": 30757,
+      "Ġconductors": 30758,
+      "oplasm": 30759,
+      "ĠCosmology": 30760,
+      "Zero": 30761,
+      "ĠInform": 30762,
+      "Ġendangered": 30763,
+      "Ġweapons": 30764,
+      "atype": 30765,
+      "ĠAsc": 30766,
+      "Ġfluence": 30767,
+      "Ġferric": 30768,
+      "ĠLaurent": 30769,
+      "Early": 30770,
+      "Ġsgn": 30771,
+      "ĠHadamard": 30772,
+      "Ġastron": 30773,
+      "Cys": 30774,
+      "ĠThm": 30775,
+      "Ġdece": 30776,
+      "erencing": 30777,
+      "ĠMeans": 30778,
+      "Ġhydrated": 30779,
+      "ÙĪ": 30780,
+      "Ġrigorously": 30781,
+      "Ġambulatory": 30782,
+      "ĠDOI": 30783,
+      "Handle": 30784,
+      "ĠEnterobacteriaceae": 30785,
+      "ĠRQ": 30786,
+      "ĠGFR": 30787,
+      "prote": 30788,
+      "Ġmigrated": 30789,
+      "thening": 30790,
+      "ĠHopkins": 30791,
+      "ĠPsychology": 30792,
+      "igl": 30793,
+      "ĠEDS": 30794,
+      "ĠâĪ¶": 30795,
+      "Ġremotely": 30796,
+      "ĠÂ¥": 30797,
+      "Ġinspiration": 30798,
+      "ĠâĮ¬": 30799,
+      "olian": 30800,
+      "Ġsaliency": 30801,
+      "ĠDog": 30802,
+      "ĠRosa": 30803,
+      "oya": 30804,
+      "Ġoccupies": 30805,
+      "camera": 30806,
+      "Ġdecompression": 30807,
+      "Ġscatt": 30808,
+      "Ġinvestigator": 30809,
+      "Ġcounterex": 30810,
+      "ĠIFNγ": 30811,
+      "ĠPittsburgh": 30812,
+      "Ġadminister": 30813,
+      "negl": 30814,
+      "ussis": 30815,
+      "MPC": 30816,
+      "ĠSwitching": 30817,
+      "Ġcooler": 30818,
+      "Ġbronchi": 30819,
+      "Ġparalle": 30820,
+      "Ġspeckle": 30821,
+      "Ġphysiologic": 30822,
+      "INVAL": 30823,
+      "Ġheterologous": 30824,
+      "|||": 30825,
+      "orghum": 30826,
+      "GAL": 30827,
+      "Ġmalformations": 30828,
+      "Ġweakening": 30829,
+      "Ġpsycho": 30830,
+      "ĠIH": 30831,
+      "Ġcontradictory": 30832,
+      "Ġphonological": 30833,
+      "ĠPerturbation": 30834,
+      "bB": 30835,
+      "ĠNos": 30836,
+      "TRUE": 30837,
+      "folding": 30838,
+      "phenol": 30839,
+      "ĠLSM": 30840,
+      "ĠâĪĹ": 30841,
+      "ĠAngle": 30842,
+      "Ġprovincial": 30843,
+      "FeO": 30844,
+      "ÅĽ": 30845,
+      "ĠIber": 30846,
+      "ressors": 30847,
+      "Ġproliferating": 30848,
+      "zers": 30849,
+      "organism": 30850,
+      "âĨĵ": 30851,
+      "ZO": 30852,
+      "cimg": 30853,
+      "Ġunperturbed": 30854,
+      "Ġjj": 30855,
+      "Ġelectrodynamics": 30856,
+      "ĠEpit": 30857,
+      "NTs": 30858,
+      "ĠBloom": 30859,
+      "Ġlanth": 30860,
+      "aminant": 30861,
+      "ĠSwift": 30862,
+      "European": 30863,
+      "Ġafferent": 30864,
+      "Reduce": 30865,
+      "published": 30866,
+      "ĠFitting": 30867,
+      "ĠFungal": 30868,
+      "Ġtribe": 30869,
+      "recting": 30870,
+      "Ġconjugacy": 30871,
+      "imeters": 30872,
+      "ĠCec": 30873,
+      "ĠKH": 30874,
+      "castle": 30875,
+      "Ġseptal": 30876,
+      "releasing": 30877,
+      "Ġoss": 30878,
+      "Ġ¦": 30879,
+      "ĠMissing": 30880,
+      "ĠFatigue": 30881,
+      "ĠBaseball": 30882,
+      "Ġimmunoblotting": 30883,
+      "Ġoh": 30884,
+      "orations": 30885,
+      "Ġvine": 30886,
+      "azy": 30887,
+      "serum": 30888,
+      "Ġlookup": 30889,
+      "Ġneovascular": 30890,
+      "iah": 30891,
+      "soil": 30892,
+      "Ġairflow": 30893,
+      "ĠSloan": 30894,
+      "him": 30895,
+      "çļ": 30896,
+      "located": 30897,
+      "zantine": 30898,
+      "ĠSuccessful": 30899,
+      "eminal": 30900,
+      "ĠDimensional": 30901,
+      "ĠNSA": 30902,
+      "ĠLogistic": 30903,
+      "emetery": 30904,
+      "Ġbrak": 30905,
+      "antal": 30906,
+      "south": 30907,
+      "Ġprototypes": 30908,
+      "Ġadvised": 30909,
+      "Ġidealized": 30910,
+      "ophytic": 30911,
+      "nbsp": 30912,
+      "Binary": 30913,
+      "Hyp": 30914,
+      "Joh": 30915,
+      "polation": 30916,
+      "Ġpolyvinyl": 30917,
+      "estimated": 30918,
+      "Ġoxytocin": 30919,
+      "ĠLetter": 30920,
+      "ĠImpair": 30921,
+      "Ġenvelopes": 30922,
+      "mainly": 30923,
+      "Ġmys": 30924,
+      "Ġintras": 30925,
+      "Ġbiogenic": 30926,
+      "cysteine": 30927,
+      "Ġuric": 30928,
+      "ĠCyan": 30929,
+      "ryption": 30930,
+      "Ġphotoreceptor": 30931,
+      "ĠToxic": 30932,
+      "ĠGamm": 30933,
+      "Ġcontainment": 30934,
+      "IgG": 30935,
+      "Squ": 30936,
+      "Ġperfused": 30937,
+      "Ġbiosensors": 30938,
+      "Ġmagmatic": 30939,
+      "Rate": 30940,
+      "ĠTf": 30941,
+      "Ġsecrete": 30942,
+      "Ġcriticality": 30943,
+      "Ġcompositionally": 30944,
+      "ĠBruce": 30945,
+      "SZ": 30946,
+      "ĠSport": 30947,
+      "ĠEI": 30948,
+      "Ġdiseased": 30949,
+      "Ġpreschool": 30950,
+      "ĠHarvey": 30951,
+      "ĠPTH": 30952,
+      "Ġbilayers": 30953,
+      "ĠOscillations": 30954,
+      "ĠHonor": 30955,
+      "ĠCCN": 30956,
+      "ĠMOT": 30957,
+      "ĠLloyd": 30958,
+      "Ġtrapez": 30959,
+      "Ġbuds": 30960,
+      "OFFSET": 30961,
+      "Ġmacromolecules": 30962,
+      "Ġbilirubin": 30963,
+      "olly": 30964,
+      "Ġutilities": 30965,
+      "ministered": 30966,
+      "Ġglobe": 30967,
+      "OLOGY": 30968,
+      "ropods": 30969,
+      "ĠMDM": 30970,
+      "ĠPyObject": 30971,
+      "macroph": 30972,
+      "ĠPBMCs": 30973,
+      "ospheres": 30974,
+      "Ġcatastrophic": 30975,
+      "ĠNavigation": 30976,
+      "ĠLSD": 30977,
+      "Ġcream": 30978,
+      "Ġdereg": 30979,
+      "bonded": 30980,
+      "rents": 30981,
+      "Ġpotentiation": 30982,
+      "Ġstro": 30983,
+      "Ġsteeper": 30984,
+      "ulinum": 30985,
+      "Ġperiodontitis": 30986,
+      "arization": 30987,
+      "âĪª": 30988,
+      "amicin": 30989,
+      "Ġmagnetized": 30990,
+      "ĠNutritional": 30991,
+      "Ġaccord": 30992,
+      "gaard": 30993,
+      "FTIR": 30994,
+      "ramethyl": 30995,
+      "ĠGle": 30996,
+      "Mel": 30997,
+      "ĠCTL": 30998,
+      "Ġtranslating": 30999,
+      "Ġautoimmunity": 31000,
+      "olerant": 31001,
+      "triangleq": 31002,
+      "amo": 31003,
+      "Ġvel": 31004,
+      "ĠHCN": 31005,
+      "ĠHamming": 31006,
+      "ĠVenus": 31007,
+      "ĠGad": 31008,
+      "ĠOwing": 31009,
+      "Information": 31010,
+      "ĠSchemes": 31011,
+      "carotene": 31012,
+      "Its": 31013,
+      "anis": 31014,
+      "Ġreplay": 31015,
+      "Ġtouc": 31016,
+      "LECT": 31017,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 31018,
+      "Ġtabulated": 31019,
+      "ĠSchottky": 31020,
+      "Far": 31021,
+      "amation": 31022,
+      "ĠRies": 31023,
+      "Ġexpects": 31024,
+      "ĠInstability": 31025,
+      "Ġsons": 31026,
+      "Ġdeck": 31027,
+      "Ġïģ¥": 31028,
+      "ĠSignature": 31029,
+      "Ġlithosphere": 31030,
+      "WW": 31031,
+      "makers": 31032,
+      "ughters": 31033,
+      "Ġâİ¡": 31034,
+      "ardian": 31035,
+      "à¦": 31036,
+      "Ġaccepts": 31037,
+      "ĠOSA": 31038,
+      "Ġγδ": 31039,
+      "nonumber": 31040,
+      "Select": 31041,
+      "lite": 31042,
+      "ĠAqueous": 31043,
+      "agawa": 31044,
+      "ĠEdinburgh": 31045,
+      "ĠMembranes": 31046,
+      "ĠSIG": 31047,
+      "akia": 31048,
+      "Ġtestes": 31049,
+      "Ġheli": 31050,
+      "++++": 31051,
+      "Ġultrafast": 31052,
+      "Ġmaneuver": 31053,
+      "ĠDate": 31054,
+      "phin": 31055,
+      "ĠKad": 31056,
+      "Ġtransferase": 31057,
+      "Pers": 31058,
+      "Ġtones": 31059,
+      "ĠSGD": 31060,
+      "anto": 31061,
+      "ĠOrange": 31062,
+      "ĠGeography": 31063,
+      "ĠAccumulation": 31064,
+      "aty": 31065,
+      "Ġbeating": 31066,
+      "Ġoverlying": 31067,
+      "ĠNDVI": 31068,
+      "ĠTownship": 31069,
+      "jing": 31070,
+      "ĠNOS": 31071,
+      "player": 31072,
+      "ĠMDD": 31073,
+      "ĠHungarian": 31074,
+      "Ġdw": 31075,
+      "ĠHin": 31076,
+      "Ġvalidating": 31077,
+      "Ġcolorimetric": 31078,
+      "ĠSupersymmetric": 31079,
+      "FUNC": 31080,
+      "gically": 31081,
+      "ofuran": 31082,
+      "-------": 31083,
+      "Ġimping": 31084,
+      "similarity": 31085,
+      "ĠDOX": 31086,
+      "ĠGlo": 31087,
+      "ivirus": 31088,
+      "listed": 31089,
+      "Ġbusy": 31090,
+      "iprofloxacin": 31091,
+      "Ġanxi": 31092,
+      "Ġblunt": 31093,
+      "Ġprocedural": 31094,
+      "Ġunknowns": 31095,
+      "AdS": 31096,
+      "thickness": 31097,
+      "follows": 31098,
+      "closing": 31099,
+      "environmental": 31100,
+      "ĠFeeding": 31101,
+      "unami": 31102,
+      "ende": 31103,
+      "ipine": 31104,
+      "Ġimpacting": 31105,
+      "Ġpenetrating": 31106,
+      "ambia": 31107,
+      "ĠWavelet": 31108,
+      "Ġfilamentous": 31109,
+      "Ġleng": 31110,
+      "ĠSCA": 31111,
+      "ĠEther": 31112,
+      "metall": 31113,
+      "Ġfringe": 31114,
+      "ĠAdjust": 31115,
+      "usz": 31116,
+      "ĠRey": 31117,
+      "ĠBoyd": 31118,
+      "Ġburnout": 31119,
+      "Ġcook": 31120,
+      "Ġnowadays": 31121,
+      "ĠDispersion": 31122,
+      "ĠRodriguez": 31123,
+      "Factor": 31124,
+      "ĠOklahoma": 31125,
+      "Ġunital": 31126,
+      "Ġpredictability": 31127,
+      "Ġlithography": 31128,
+      "ès": 31129,
+      "Willi": 31130,
+      "unal": 31131,
+      "asting": 31132,
+      "correction": 31133,
+      "ĠDed": 31134,
+      "ĠSocio": 31135,
+      "ĠChapman": 31136,
+      "ĠEco": 31137,
+      "Ġoncogene": 31138,
+      "ĠDrive": 31139,
+      "Ġfunnel": 31140,
+      "uis": 31141,
+      "ĠGENER": 31142,
+      "ĠACR": 31143,
+      "Ġworkloads": 31144,
+      "Ġoctahedral": 31145,
+      "vich": 31146,
+      "enburg": 31147,
+      "Ġimproper": 31148,
+      "decoded": 31149,
+      "Ġimmunosorbent": 31150,
+      "Ġinhomogeneity": 31151,
+      "RK": 31152,
+      "onically": 31153,
+      "Ġglycoproteins": 31154,
+      "onics": 31155,
+      "ĠFok": 31156,
+      "ĠBras": 31157,
+      "ĠCalculus": 31158,
+      "ĠMoss": 31159,
+      "ĠRK": 31160,
+      "Ġviolet": 31161,
+      "Ġlymphomas": 31162,
+      "enspace": 31163,
+      "ĠPalae": 31164,
+      "Ġrenin": 31165,
+      "phant": 31166,
+      "ĠRED": 31167,
+      "Ġfaulty": 31168,
+      "Riemann": 31169,
+      "Ãī": 31170,
+      "ĠElli": 31171,
+      "Bol": 31172,
+      "Tn": 31173,
+      "Yang": 31174,
+      "gender": 31175,
+      "Ġdetuning": 31176,
+      "Ġoperon": 31177,
+      "Ġinsecticide": 31178,
+      "esi": 31179,
+      "amon": 31180,
+      "ĠSCD": 31181,
+      "ĠBath": 31182,
+      "ĠâĢĸ": 31183,
+      "ĠGeographic": 31184,
+      "Ġcyclohex": 31185,
+      "ĠConfidence": 31186,
+      "Ġcomet": 31187,
+      "Ġfolate": 31188,
+      "observer": 31189,
+      "Ġvisitors": 31190,
+      "extra": 31191,
+      "ateness": 31192,
+      "ĠSPT": 31193,
+      "arcane": 31194,
+      "Ġholistic": 31195,
+      "semi": 31196,
+      "ĠMild": 31197,
+      "Ġsmear": 31198,
+      "Ġcyclase": 31199,
+      "Ġanymore": 31200,
+      "Ġseagrass": 31201,
+      "Ġconsortium": 31202,
+      "Ġfinishes": 31203,
+      "cyan": 31204,
+      "ductance": 31205,
+      "frost": 31206,
+      "hereafter": 31207,
+      "Ġprescriptions": 31208,
+      "Ġcmd": 31209,
+      "ĠPerceived": 31210,
+      "coordinates": 31211,
+      "Ġstyl": 31212,
+      "ĠBard": 31213,
+      "ĠHoll": 31214,
+      "ĠsiRNAs": 31215,
+      "sugg": 31216,
+      "Ġthr": 31217,
+      "Ġmainland": 31218,
+      "SCH": 31219,
+      "Ġassertions": 31220,
+      "Ġbabies": 31221,
+      "Ġrecapit": 31222,
+      "Tok": 31223,
+      "Ġresected": 31224,
+      "construct": 31225,
+      "Ber": 31226,
+      "Ġcholine": 31227,
+      "Ġunitarity": 31228,
+      "Ġcatalyzes": 31229,
+      "detector": 31230,
+      "ĠSMB": 31231,
+      "tery": 31232,
+      "cluded": 31233,
+      "ĠAbbreviations": 31234,
+      "ĠOliveira": 31235,
+      "LOC": 31236,
+      "zin": 31237,
+      "ĠLorenz": 31238,
+      "Kernel": 31239,
+      "lyn": 31240,
+      "ĠLEP": 31241,
+      "soni": 31242,
+      "Ġseptum": 31243,
+      "TMS": 31244,
+      "Ġunmodified": 31245,
+      "borough": 31246,
+      "ĠAudio": 31247,
+      "Ġdollars": 31248,
+      "CMD": 31249,
+      "Ġnorthwestern": 31250,
+      "Ġpalmit": 31251,
+      "ragalactic": 31252,
+      "ĠMiz": 31253,
+      "FH": 31254,
+      "confidence": 31255,
+      "NEXT": 31256,
+      "ĠAGE": 31257,
+      "ĠEqn": 31258,
+      "ĠClasses": 31259,
+      "Ġmisleading": 31260,
+      "ĠPKA": 31261,
+      "Ġanchored": 31262,
+      "ĠRip": 31263,
+      "phag": 31264,
+      "Ġintubation": 31265,
+      "ĠAngular": 31266,
+      "ĠBEC": 31267,
+      "Thr": 31268,
+      "Ġorganisations": 31269,
+      "Ġcomfortable": 31270,
+      "Ġcommissioned": 31271,
+      "poll": 31272,
+      "ydia": 31273,
+      "instead": 31274,
+      "Ġpassword": 31275,
+      "Ġcompliant": 31276,
+      "ĠPrecipitation": 31277,
+      "ophosphamide": 31278,
+      "usters": 31279,
+      "Ġpneumococcal": 31280,
+      "Ġtomographic": 31281,
+      "tidae": 31282,
+      "ĠFirmicutes": 31283,
+      "bw": 31284,
+      "ĠPDB": 31285,
+      "ĠGPUs": 31286,
+      "ĠPlanar": 31287,
+      "Ġverbose": 31288,
+      "Summary": 31289,
+      "lance": 31290,
+      "ĠEGFP": 31291,
+      "ongru": 31292,
+      "Complex": 31293,
+      "ĠWheat": 31294,
+      "uche": 31295,
+      "ĠMCA": 31296,
+      "ĠProjection": 31297,
+      "Ġstats": 31298,
+      "Ġsummand": 31299,
+      "dimethoxyphenyl": 31300,
+      "ĠABSTRACT": 31301,
+      "Ġcarotenoid": 31302,
+      "Ġbroke": 31303,
+      "ĠDesigning": 31304,
+      "ĠHetero": 31305,
+      "ĠCarlsbad": 31306,
+      "Cov": 31307,
+      "ineral": 31308,
+      "Ġanalyte": 31309,
+      "ĠColeman": 31310,
+      "Ġeigenstate": 31311,
+      "ĠHolland": 31312,
+      "ERSION": 31313,
+      "ĠDak": 31314,
+      "ellers": 31315,
+      "ĠÃĺ": 31316,
+      "missing": 31317,
+      "deposited": 31318,
+      "ĠLincoln": 31319,
+      "anion": 31320,
+      "ĠSPEC": 31321,
+      "Ġfertilizer": 31322,
+      "ĠCPS": 31323,
+      "Ġcofactor": 31324,
+      "Ġtren": 31325,
+      "Ġcalendar": 31326,
+      "Ġyoungest": 31327,
+      "STATUS": 31328,
+      "ĠEXPERIMENTAL": 31329,
+      "Ġsr": 31330,
+      "Ġnl": 31331,
+      "ĠMES": 31332,
+      "Study": 31333,
+      "padding": 31334,
+      "Ġatopic": 31335,
+      "ĠOG": 31336,
+      "Ġentrainment": 31337,
+      "AFM": 31338,
+      "ĠCou": 31339,
+      "Web": 31340,
+      "ĠMicroscopic": 31341,
+      "Ġunambiguously": 31342,
+      "Day": 31343,
+      "yotrophic": 31344,
+      "reous": 31345,
+      "Ġsarcom": 31346,
+      "ĠVAL": 31347,
+      "Ġhindered": 31348,
+      "ĠREM": 31349,
+      "otrexate": 31350,
+      "ocarcin": 31351,
+      "ĠAlk": 31352,
+      "Ġbrevity": 31353,
+      "factual": 31354,
+      "Cer": 31355,
+      "diox": 31356,
+      "ophical": 31357,
+      "Ġlytic": 31358,
+      "Take": 31359,
+      "Ġintend": 31360,
+      "ĠCla": 31361,
+      "Ġasteroid": 31362,
+      "ĠSEP": 31363,
+      "apenem": 31364,
+      "universal": 31365,
+      "Ġoceans": 31366,
+      "Ġmonoid": 31367,
+      "Ġseparator": 31368,
+      "ĠPorous": 31369,
+      "Ġpostoperatively": 31370,
+      "Ġsemin": 31371,
+      "ĠDisplay": 31372,
+      "Ġhydrolase": 31373,
+      "transferases": 31374,
+      "Ġthrombus": 31375,
+      "ĠOv": 31376,
+      "ĠDielectric": 31377,
+      "Ġcompelling": 31378,
+      "assing": 31379,
+      "ĠMAS": 31380,
+      "ullary": 31381,
+      "ĠMori": 31382,
+      "ĠPathogenesis": 31383,
+      "ĠBreaking": 31384,
+      "ĠPLGA": 31385,
+      "cooling": 31386,
+      "§": 31387,
+      "Ġfee": 31388,
+      "Ġreducible": 31389,
+      "Ġdiverge": 31390,
+      "Ġqueues": 31391,
+      "Ġmushroom": 31392,
+      "Ġdeacetylase": 31393,
+      "YFP": 31394,
+      "Ġdisreg": 31395,
+      "ĠArrays": 31396,
+      "processes": 31397,
+      "ĠTransportation": 31398,
+      "Ġundetectable": 31399,
+      "bursts": 31400,
+      "Ġphospholipase": 31401,
+      "Option": 31402,
+      "asin": 31403,
+      "Ġnocturnal": 31404,
+      "tez": 31405,
+      "ĠDisruption": 31406,
+      "oserine": 31407,
+      "behavior": 31408,
+      "ĠTony": 31409,
+      "ĠKot": 31410,
+      "ieval": 31411,
+      "Ġmyofib": 31412,
+      "Ġhalogen": 31413,
+      "ĠCPR": 31414,
+      "ployed": 31415,
+      "ĠPolymers": 31416,
+      "Ġadenoma": 31417,
+      "Ġquartile": 31418,
+      "Ġquaternary": 31419,
+      "ĠIraq": 31420,
+      "Ġsieve": 31421,
+      "Ġintractable": 31422,
+      "Ġfabrics": 31423,
+      "continuum": 31424,
+      "ĠEmergence": 31425,
+      "Pot": 31426,
+      "itism": 31427,
+      "veness": 31428,
+      "hoe": 31429,
+      "Ġredes": 31430,
+      "ĠHRP": 31431,
+      "ploidy": 31432,
+      "picuous": 31433,
+      "ogo": 31434,
+      "ĠGag": 31435,
+      "Ġnominated": 31436,
+      "occupied": 31437,
+      "Ġquench": 31438,
+      "ropolis": 31439,
+      "nucleotide": 31440,
+      "ĠEventually": 31441,
+      "Ñı": 31442,
+      "ĠClock": 31443,
+      "ĠSteady": 31444,
+      "opolymers": 31445,
+      "ĠARE": 31446,
+      "irnov": 31447,
+      "helf": 31448,
+      "blob": 31449,
+      "download": 31450,
+      "PLL": 31451,
+      "UNT": 31452,
+      "predictions": 31453,
+      "Ġoccipital": 31454,
+      "toxic": 31455,
+      "ĠVice": 31456,
+      "Ġangio": 31457,
+      "CuO": 31458,
+      "Ġresistances": 31459,
+      "fflffl": 31460,
+      "Distribution": 31461,
+      "Gre": 31462,
+      "onamide": 31463,
+      "ĠIOP": 31464,
+      "UNEL": 31465,
+      "Ġaids": 31466,
+      "ĠHUV": 31467,
+      "ECM": 31468,
+      "ĠPAD": 31469,
+      "ĠAgNPs": 31470,
+      "Print": 31471,
+      "Ġlamellar": 31472,
+      "ĠUltrason": 31473,
+      "severe": 31474,
+      "ĠAnnotation": 31475,
+      "NIR": 31476,
+      "sgn": 31477,
+      "ĠOften": 31478,
+      "Ġiterate": 31479,
+      "Ġcarriage": 31480,
+      "spherical": 31481,
+      "ĠFrid": 31482,
+      "Ġdiffract": 31483,
+      "ĠBasal": 31484,
+      "Ġunsatisf": 31485,
+      "ĠDysfunction": 31486,
+      "arboxylic": 31487,
+      "ĠCollective": 31488,
+      "Ġdegrading": 31489,
+      "Ġadiposity": 31490,
+      "Ġfifty": 31491,
+      "Ġpars": 31492,
+      "ĠOptimized": 31493,
+      "ocaine": 31494,
+      "Ġbb": 31495,
+      "ĠShip": 31496,
+      "ĠLW": 31497,
+      "Ġtremor": 31498,
+      "Ġã": 31499,
+      "Ġnucleons": 31500,
+      "Ġscientist": 31501,
+      "ĠMish": 31502,
+      "gression": 31503,
+      "ĠMerc": 31504,
+      "ĠFlem": 31505,
+      "Ġcorals": 31506,
+      "Incre": 31507,
+      "ĠDSP": 31508,
+      "Ġdefenses": 31509,
+      "dimer": 31510,
+      "atherine": 31511,
+      "otubes": 31512,
+      "stride": 31513,
+      "ĠAlterations": 31514,
+      "Ġoest": 31515,
+      "ĠBIC": 31516,
+      "Ġradiated": 31517,
+      "Ġketamine": 31518,
+      "Ġdissimilarity": 31519,
+      "ĠAncient": 31520,
+      "ĠHed": 31521,
+      "Ġattr": 31522,
+      "ĠIsa": 31523,
+      "Ġionospheric": 31524,
+      "Ġgovernor": 31525,
+      "ĠEstimated": 31526,
+      "Ġultrathin": 31527,
+      "Update": 31528,
+      "Ġimmunoassay": 31529,
+      "Ġconjectured": 31530,
+      "ĠREF": 31531,
+      "ĠSiegel": 31532,
+      "Adv": 31533,
+      "Mem": 31534,
+      "Ġpups": 31535,
+      "ĠAPPL": 31536,
+      "ecomposable": 31537,
+      "journal": 31538,
+      "ĠRol": 31539,
+      "ĠLob": 31540,
+      "rington": 31541,
+      "Ġnonsingular": 31542,
+      "Ġcitric": 31543,
+      "iones": 31544,
+      "ositis": 31545,
+      "ALY": 31546,
+      "Ġmentions": 31547,
+      "ĠMarkers": 31548,
+      "algebraic": 31549,
+      "Ġflattened": 31550,
+      "Ġmail": 31551,
+      "ĠTGA": 31552,
+      "ĠPMA": 31553,
+      "ĠNaval": 31554,
+      "Ġfacilitation": 31555,
+      "Ġunidentified": 31556,
+      "Ġempathy": 31557,
+      "jectories": 31558,
+      "logits": 31559,
+      "Ġpermanently": 31560,
+      "Ġbottles": 31561,
+      "ĠBengal": 31562,
+      "Ġpeanut": 31563,
+      "Ġcapillaries": 31564,
+      "erents": 31565,
+      "ĠLooking": 31566,
+      "changes": 31567,
+      "ĠMagell": 31568,
+      "ĠCMC": 31569,
+      "ĠVerm": 31570,
+      "Ġsubscales": 31571,
+      "demand": 31572,
+      "orexia": 31573,
+      "Ġachievements": 31574,
+      "ĠRobustness": 31575,
+      "ĠWallace": 31576,
+      "ĠDTT": 31577,
+      "ogels": 31578,
+      "ocker": 31579,
+      "ĠSpike": 31580,
+      "Ġpainter": 31581,
+      "Ġbuses": 31582,
+      "Ġpolluted": 31583,
+      "Ġtort": 31584,
+      "ĠPPP": 31585,
+      "nex": 31586,
+      "extended": 31587,
+      "ucalypt": 31588,
+      "Ġprostatic": 31589,
+      "ĠFCC": 31590,
+      "Ġkick": 31591,
+      "oyal": 31592,
+      "epochs": 31593,
+      "hss": 31594,
+      "yon": 31595,
+      "Ġdans": 31596,
+      "ĠAw": 31597,
+      "Ġadversely": 31598,
+      "Ġaltogether": 31599,
+      "Ġophthalm": 31600,
+      "Ġcpu": 31601,
+      "ĠFRET": 31602,
+      "Ġforensic": 31603,
+      "Ġhotspots": 31604,
+      "Ġpaintings": 31605,
+      "Ġomn": 31606,
+      "ĠpS": 31607,
+      "oglu": 31608,
+      "ofol": 31609,
+      "FTs": 31610,
+      "Ġdermat": 31611,
+      "pragma": 31612,
+      "Ġbump": 31613,
+      "ĠCir": 31614,
+      "aS": 31615,
+      "Ġnaked": 31616,
+      "ĠNLS": 31617,
+      "ĠSpitzer": 31618,
+      "Ġsalvage": 31619,
+      "Ġintuitively": 31620,
+      "Ġcasual": 31621,
+      "Ġfired": 31622,
+      "verages": 31623,
+      "ĠBurden": 31624,
+      "Wang": 31625,
+      "ylem": 31626,
+      "Ġradiographs": 31627,
+      "ĠSchiff": 31628,
+      "OLUTION": 31629,
+      "Cross": 31630,
+      "Ġhints": 31631,
+      "owing": 31632,
+      "ĠStreng": 31633,
+      "ĠANY": 31634,
+      "Ġworry": 31635,
+      "ĠRoger": 31636,
+      "Ġtrabecular": 31637,
+      "Band": 31638,
+      "ĠNec": 31639,
+      "ipes": 31640,
+      "tool": 31641,
+      "ĠILC": 31642,
+      "iÄĩ": 31643,
+      "ocean": 31644,
+      "ĠAri": 31645,
+      "AMA": 31646,
+      "ĠVertex": 31647,
+      "activate": 31648,
+      "Location": 31649,
+      "onts": 31650,
+      "Ġhs": 31651,
+      "Ġslender": 31652,
+      "refring": 31653,
+      "ĠEndogenous": 31654,
+      "adiabatic": 31655,
+      "Ġcryptic": 31656,
+      "Ġeradication": 31657,
+      "ĠKevin": 31658,
+      "Ġmc": 31659,
+      "Ġcardio": 31660,
+      "Ġphosphoryl": 31661,
+      "Witten": 31662,
+      "Ġscl": 31663,
+      "ĠIw": 31664,
+      "ĠMade": 31665,
+      "Ġfounding": 31666,
+      "oflag": 31667,
+      "aline": 31668,
+      "horizontal": 31669,
+      "ĠGeneralization": 31670,
+      "psychiatric": 31671,
+      "ĠDuncan": 31672,
+      "ĠSnO": 31673,
+      "ĠAar": 31674,
+      "Ġgg": 31675,
+      "Ġpremi": 31676,
+      "ĠStrom": 31677,
+      "ĠExplan": 31678,
+      "Ġlethality": 31679,
+      "ÏĤ": 31680,
+      "odo": 31681,
+      "Ġsubscrib": 31682,
+      "ĠSTUDY": 31683,
+      "Ġoutperformed": 31684,
+      "Ġcovalently": 31685,
+      "MHC": 31686,
+      "fail": 31687,
+      "ĠKac": 31688,
+      "EGR": 31689,
+      "ĠTRI": 31690,
+      "robot": 31691,
+      "ĠCandidate": 31692,
+      "ĠTNBC": 31693,
+      "Ġarchaeological": 31694,
+      "Eukary": 31695,
+      "Ġlava": 31696,
+      "dipole": 31697,
+      "Ġuncons": 31698,
+      "Anti": 31699,
+      "Ġprednis": 31700,
+      "ĠRobin": 31701,
+      "Ġstratigraphic": 31702,
+      "Ġ¤": 31703,
+      "Ġfinance": 31704,
+      "ĠStudio": 31705,
+      "render": 31706,
+      "Ġrearing": 31707,
+      "Ġger": 31708,
+      "ĠOpt": 31709,
+      "ĠManifolds": 31710,
+      "Ġdestabil": 31711,
+      "Ġtelomerase": 31712,
+      "Ġpicking": 31713,
+      "Ġamplicon": 31714,
+      "Ġyearly": 31715,
+      "ĠNCC": 31716,
+      "inser": 31717,
+      "ĠEnrichment": 31718,
+      "ĠMicrostructure": 31719,
+      "ĠWarren": 31720,
+      "ophysics": 31721,
+      "Ġfifteen": 31722,
+      "Åij": 31723,
+      "Ġreviewer": 31724,
+      "Ġskilled": 31725,
+      "Ġmagnetoresistance": 31726,
+      "Ġreconfiguration": 31727,
+      "Ġpoet": 31728,
+      "Ġpredetermined": 31729,
+      "Ġcryopres": 31730,
+      "Ġattractors": 31731,
+      "Ġprojectile": 31732,
+      "ĠCrystals": 31733,
+      "ĠMCM": 31734,
+      "ĠXanth": 31735,
+      "Ġclockwise": 31736,
+      "regnant": 31737,
+      "Ġgated": 31738,
+      "ryza": 31739,
+      "ĠProsp": 31740,
+      "adin": 31741,
+      "Ġmolybdenum": 31742,
+      "ĠAlps": 31743,
+      "ĠBald": 31744,
+      "Ġhalluc": 31745,
+      "udo": 31746,
+      "Ġmont": 31747,
+      "ĠFlash": 31748,
+      "Ġpulling": 31749,
+      "ĠLQ": 31750,
+      "ĠWalsh": 31751,
+      "ĠThomson": 31752,
+      "meson": 31753,
+      "Ġintercal": 31754,
+      "Ġelapsed": 31755,
+      "FFFF": 31756,
+      "ĠForecasting": 31757,
+      "à¯": 31758,
+      "ĠLSP": 31759,
+      "endorf": 31760,
+      "Ġxml": 31761,
+      "substrate": 31762,
+      "Mu": 31763,
+      "during": 31764,
+      "oconstr": 31765,
+      "EMA": 31766,
+      "Ġïĥ«": 31767,
+      "ĠDFS": 31768,
+      "ĠVon": 31769,
+      "Ġfathers": 31770,
+      "Ġunco": 31771,
+      "ĠUnderg": 31772,
+      "Ġmultiplexing": 31773,
+      "atra": 31774,
+      "Ġcohesive": 31775,
+      "ĠUI": 31776,
+      "ĠPrev": 31777,
+      "çļĦ": 31778,
+      "cum": 31779,
+      "hf": 31780,
+      "ĠSCN": 31781,
+      "atalysis": 31782,
+      "ĠArsen": 31783,
+      "amping": 31784,
+      "ĠPlastic": 31785,
+      "ĠMadison": 31786,
+      "Ġsupremum": 31787,
+      "ĠCited": 31788,
+      "Ġaren": 31789,
+      "iski": 31790,
+      "inel": 31791,
+      "stro": 31792,
+      "Ġcorrupted": 31793,
+      "Ġglab": 31794,
+      "Ġcardiopulmonary": 31795,
+      "Ġpragmatic": 31796,
+      "CAG": 31797,
+      "Stack": 31798,
+      "thioxo": 31799,
+      "ĠReproductive": 31800,
+      "Ġsteatosis": 31801,
+      "Best": 31802,
+      "ĠBars": 31803,
+      "Ġracing": 31804,
+      "ĠUtah": 31805,
+      "equivalence": 31806,
+      "ĠFifty": 31807,
+      "ĠCytokine": 31808,
+      "Ġutilised": 31809,
+      "horizon": 31810,
+      "ouracil": 31811,
+      "iversary": 31812,
+      "emer": 31813,
+      "ĠQuestions": 31814,
+      "Ġlinkages": 31815,
+      "anchez": 31816,
+      "VV": 31817,
+      "Ġphotodet": 31818,
+      "kowski": 31819,
+      "REST": 31820,
+      "Ġhosting": 31821,
+      "Ġpushing": 31822,
+      "Ġneurotoxicity": 31823,
+      "SQ": 31824,
+      "rst": 31825,
+      "Ġhockey": 31826,
+      "Ġtrips": 31827,
+      "ĠIndoor": 31828,
+      "ematics": 31829,
+      "Ġtransect": 31830,
+      "ĠABI": 31831,
+      "agar": 31832,
+      "âĪļ": 31833,
+      "egenerate": 31834,
+      "ĠQP": 31835,
+      "MID": 31836,
+      "ĠAccept": 31837,
+      "ĠCyber": 31838,
+      "North": 31839,
+      "Ġdθ": 31840,
+      "alla": 31841,
+      "Ġbraid": 31842,
+      "finding": 31843,
+      "alin": 31844,
+      "ĠLST": 31845,
+      "ĠLax": 31846,
+      "udin": 31847,
+      "ĠiNOS": 31848,
+      "convert": 31849,
+      "ACA": 31850,
+      "ĠGuan": 31851,
+      "Ġlymphocytic": 31852,
+      "Ġsyllable": 31853,
+      "ĠTOR": 31854,
+      "ĠSCR": 31855,
+      "ĠAJ": 31856,
+      "Ġoutburst": 31857,
+      "bladder": 31858,
+      "OTA": 31859,
+      "audio": 31860,
+      "chromen": 31861,
+      "ÑģÑĤ": 31862,
+      "Ġgratefully": 31863,
+      "Ġtiling": 31864,
+      "Ġquit": 31865,
+      "shan": 31866,
+      "ĠAccretion": 31867,
+      "Ġnarrowing": 31868,
+      "ĠInduces": 31869,
+      "Mic": 31870,
+      "Ġfuc": 31871,
+      "Ġthalamus": 31872,
+      "ANES": 31873,
+      "Ġquaternion": 31874,
+      "ĠListeria": 31875,
+      "duality": 31876,
+      "hend": 31877,
+      "ande": 31878,
+      "Ġparo": 31879,
+      "Ġinspected": 31880,
+      "question": 31881,
+      "ĠHoney": 31882,
+      "Ġchunks": 31883,
+      "Ġforearm": 31884,
+      "radients": 31885,
+      "ificantly": 31886,
+      "obank": 31887,
+      "Ġsomewhere": 31888,
+      "Ġmonetary": 31889,
+      "ĠLouisiana": 31890,
+      "Ġemulsions": 31891,
+      "Ġprogrammable": 31892,
+      "Ġmanifests": 31893,
+      "ĠMartinez": 31894,
+      "Ġted": 31895,
+      "emen": 31896,
+      "anni": 31897,
+      "Ġoverlaid": 31898,
+      "Ġvirulent": 31899,
+      "Mask": 31900,
+      "ĠUtility": 31901,
+      "Ġwk": 31902,
+      "osexual": 31903,
+      "ĠEarl": 31904,
+      "dar": 31905,
+      "hdr": 31906,
+      "ractors": 31907,
+      "Ġconstructor": 31908,
+      "Ġnascent": 31909,
+      "inzburg": 31910,
+      "ĠCraig": 31911,
+      "Ġplexus": 31912,
+      "reverse": 31913,
+      "ograv": 31914,
+      "tags": 31915,
+      "Ġcalibrate": 31916,
+      "à®": 31917,
+      "Ġhide": 31918,
+      "ĠFol": 31919,
+      "Ġinteracted": 31920,
+      "Ġconfron": 31921,
+      "market": 31922,
+      "Ġsociodemographic": 31923,
+      "ĠLucas": 31924,
+      "ĠMCT": 31925,
+      "ĠRSS": 31926,
+      "Ġmicroplate": 31927,
+      "underst": 31928,
+      "Ital": 31929,
+      "ĠCMR": 31930,
+      "recy": 31931,
+      "ĠPCOS": 31932,
+      "Ġdetoxification": 31933,
+      "Ġsubtree": 31934,
+      "Ġsubsections": 31935,
+      "Ġpropositions": 31936,
+      "Acknowledgements": 31937,
+      "reinforced": 31938,
+      "lis": 31939,
+      "ĠCIR": 31940,
+      "Ġimprinted": 31941,
+      "vium": 31942,
+      "afic": 31943,
+      "Ġchecklist": 31944,
+      "ĠRx": 31945,
+      "ĠEph": 31946,
+      "Ġsolder": 31947,
+      "transformation": 31948,
+      "ĠStrait": 31949,
+      "azar": 31950,
+      "Ġhandler": 31951,
+      "kelet": 31952,
+      "BCL": 31953,
+      "Math": 31954,
+      "Ġwishes": 31955,
+      "uminescent": 31956,
+      "ĠPEC": 31957,
+      "irt": 31958,
+      "ylidene": 31959,
+      "Ġloosely": 31960,
+      "naissance": 31961,
+      "ILs": 31962,
+      "foil": 31963,
+      "ĠGNU": 31964,
+      "ĠKet": 31965,
+      "vix": 31966,
+      "ĠPlain": 31967,
+      "ĠRES": 31968,
+      "Ġparenting": 31969,
+      "ĠConnection": 31970,
+      "Ġrhizosphere": 31971,
+      "oprevalence": 31972,
+      "iatic": 31973,
+      "ĠpA": 31974,
+      "ĠVil": 31975,
+      "setting": 31976,
+      "ĠReLU": 31977,
+      "ĠBOOST": 31978,
+      "Ġappreciate": 31979,
+      "bx": 31980,
+      "orest": 31981,
+      "ologie": 31982,
+      "Ġpalp": 31983,
+      "foo": 31984,
+      "usual": 31985,
+      "Ġquestioned": 31986,
+      "Ġtrigon": 31987,
+      "ĠGFAP": 31988,
+      "ĠKyoto": 31989,
+      "dise": 31990,
+      "antile": 31991,
+      "ück": 31992,
+      "ĠQuantization": 31993,
+      "Ġscler": 31994,
+      "Ġbehalf": 31995,
+      "ĠDuality": 31996,
+      "Ġmagnetically": 31997,
+      "Ġelegant": 31998,
+      "UA": 31999,
+      "epis": 32000,
+      "Ġsubclinical": 32001,
+      "ontrol": 32002,
+      "ĠChemicals": 32003,
+      "Utils": 32004,
+      "Ġlowers": 32005,
+      "extraction": 32006,
+      "Ġamplifiers": 32007,
+      "ĠEntry": 32008,
+      "ĠWORK": 32009,
+      "Ġthrombocytopenia": 32010,
+      "Mil": 32011,
+      "idus": 32012,
+      "embry": 32013,
+      "manager": 32014,
+      "ĠCoordination": 32015,
+      "ĠPhenotypic": 32016,
+      "chunk": 32017,
+      "Ġhypotension": 32018,
+      "Ġcryogenic": 32019,
+      "Ġreactants": 32020,
+      "ĠMMSE": 32021,
+      "Ġcentros": 32022,
+      "ĠButler": 32023,
+      "Ġcavitation": 32024,
+      "ĠLessons": 32025,
+      "estion": 32026,
+      "ĠMIS": 32027,
+      "associ": 32028,
+      "APE": 32029,
+      "ĠEulerian": 32030,
+      "Ġrecreational": 32031,
+      "ĠNeo": 32032,
+      "ĠCDM": 32033,
+      "repeat": 32034,
+      "details": 32035,
+      "Bal": 32036,
+      "STA": 32037,
+      "Ġâīº": 32038,
+      "ĠCamero": 32039,
+      "ĠTelevision": 32040,
+      "Ġworkforce": 32041,
+      "Ġcomputerized": 32042,
+      "Ġextraordinary": 32043,
+      "Ġribonucle": 32044,
+      "Ġhydrophobicity": 32045,
+      "ĠFeasibility": 32046,
+      "Ol": 32047,
+      "Tw": 32048,
+      "ĠMam": 32049,
+      "ĠFAC": 32050,
+      "profit": 32051,
+      "negligible": 32052,
+      "ĠFruit": 32053,
+      "Ġears": 32054,
+      "Ġshearing": 32055,
+      "ĠCorresponding": 32056,
+      "fun": 32057,
+      "ieck": 32058,
+      "mos": 32059,
+      "ĠEMI": 32060,
+      "ĠSometimes": 32061,
+      "Ġfluorine": 32062,
+      "Ġdetergent": 32063,
+      "Ġalg": 32064,
+      "races": 32065,
+      "ivable": 32066,
+      "COMM": 32067,
+      "ĠSwitch": 32068,
+      "Ġstrained": 32069,
+      "virtual": 32070,
+      "Temperature": 32071,
+      "Ġcredible": 32072,
+      "ĠGPCR": 32073,
+      "ĠDebye": 32074,
+      "ĠLit": 32075,
+      "Ġhemic": 32076,
+      "Ġtransducers": 32077,
+      "metast": 32078,
+      "adiene": 32079,
+      "Ġoryzae": 32080,
+      "tn": 32081,
+      "Ġafternoon": 32082,
+      "ĠArabian": 32083,
+      "ĠChromatin": 32084,
+      "Ġxenografts": 32085,
+      "Ġcryptographic": 32086,
+      "Ġaxillary": 32087,
+      "Ġvolunteer": 32088,
+      "ĠNevada": 32089,
+      "Ġpions": 32090,
+      "unknown": 32091,
+      "ĠFU": 32092,
+      "venously": 32093,
+      "radio": 32094,
+      "ĠLabour": 32095,
+      "ĠVillage": 32096,
+      "Ric": 32097,
+      "Ġmetat": 32098,
+      "Ġserotypes": 32099,
+      "regression": 32100,
+      "saturation": 32101,
+      "rera": 32102,
+      "Ġfarther": 32103,
+      "Ġrounding": 32104,
+      "Ġlibitum": 32105,
+      "Ġshuff": 32106,
+      "ĠOw": 32107,
+      "Ġlocalised": 32108,
+      "ĠALG": 32109,
+      "Ġhypertrophic": 32110,
+      "ppm": 32111,
+      "imine": 32112,
+      "ĠAthe": 32113,
+      "Ġanhydro": 32114,
+      "Ġsupramolecular": 32115,
+      "Ġmacros": 32116,
+      "aceted": 32117,
+      "ĠOliv": 32118,
+      "Ġmotivational": 32119,
+      "ĠCave": 32120,
+      "enzie": 32121,
+      "Ġaffiliated": 32122,
+      "Fermi": 32123,
+      "Ġequalities": 32124,
+      "ĠMilan": 32125,
+      "Ġdressed": 32126,
+      "Ġanger": 32127,
+      "ados": 32128,
+      "Ġavg": 32129,
+      "ĠPhon": 32130,
+      "Ġradioactivity": 32131,
+      "ĠEch": 32132,
+      "Ġorganoids": 32133,
+      "Ġïģ§": 32134,
+      "ĠAnthrop": 32135,
+      "lateral": 32136,
+      "Ġalpine": 32137,
+      "Ġaudit": 32138,
+      "WER": 32139,
+      "ĠCSC": 32140,
+      "Ġrankings": 32141,
+      "ĠERR": 32142,
+      "GLER": 32143,
+      "Obviously": 32144,
+      "ĠMadrid": 32145,
+      "obenzene": 32146,
+      "othermia": 32147,
+      "Ġresponsibilities": 32148,
+      "omestic": 32149,
+      "ĠInflation": 32150,
+      "Ġepidemics": 32151,
+      "Ġtaut": 32152,
+      "phos": 32153,
+      "ĠUnless": 32154,
+      "Ġgeomagnetic": 32155,
+      "ĠCFTR": 32156,
+      "veld": 32157,
+      "arietal": 32158,
+      "Ġendotoxin": 32159,
+      "ADP": 32160,
+      "Ġsuppressive": 32161,
+      "randial": 32162,
+      "Ġïĥ©": 32163,
+      "excited": 32164,
+      "ĠInnate": 32165,
+      "ĠLópez": 32166,
+      "omycetes": 32167,
+      "Ġbeautiful": 32168,
+      "irk": 32169,
+      "ĠHwang": 32170,
+      "ĠUSE": 32171,
+      "ÏĢi": 32172,
+      "Record": 32173,
+      "Attribute": 32174,
+      "Ġreacts": 32175,
+      "ĠBund": 32176,
+      "Ġcowork": 32177,
+      "Ġconfluence": 32178,
+      "ĠRegardless": 32179,
+      "Ġmetagenomic": 32180,
+      "MAL": 32181,
+      "Ġaided": 32182,
+      "anga": 32183,
+      "Ġamn": 32184,
+      "ĠICI": 32185,
+      "ĠPML": 32186,
+      "Ġdelivers": 32187,
+      "Ġkeyp": 32188,
+      "Ġbeetles": 32189,
+      "Ġoxidant": 32190,
+      "Immun": 32191,
+      "Ġrhythmic": 32192,
+      "female": 32193,
+      "JC": 32194,
+      "PAD": 32195,
+      "genitor": 32196,
+      "AMS": 32197,
+      "catalytic": 32198,
+      "ĠMom": 32199,
+      "ĠHert": 32200,
+      "adish": 32201,
+      "Ġcontention": 32202,
+      "Ġyolk": 32203,
+      "Ġdemyel": 32204,
+      "Ġsucc": 32205,
+      "Ġtravels": 32206,
+      "Ve": 32207,
+      "ĠFul": 32208,
+      "ĠRif": 32209,
+      "Ġintrons": 32210,
+      "encaps": 32211,
+      "colour": 32212,
+      "Ġhotel": 32213,
+      "Access": 32214,
+      "adoop": 32215,
+      "Ġcoalition": 32216,
+      "ĠMuh": 32217,
+      "ĠLTP": 32218,
+      "autom": 32219,
+      "ĠLak": 32220,
+      "Ġremedi": 32221,
+      "Ġtrailing": 32222,
+      "insulator": 32223,
+      "ĠRelig": 32224,
+      "ĠHudson": 32225,
+      "emics": 32226,
+      "OAc": 32227,
+      "ourt": 32228,
+      "Ġrelic": 32229,
+      "ĠMixture": 32230,
+      "Ġcalorimeter": 32231,
+      "ĠRDF": 32232,
+      "ĠHodgkin": 32233,
+      "Newtonian": 32234,
+      "ĠDelayed": 32235,
+      "ĠNortheast": 32236,
+      "hering": 32237,
+      "Ġhelices": 32238,
+      "Ġprincipally": 32239,
+      "Ġsuspicion": 32240,
+      "Ġextremities": 32241,
+      "Ġdeadline": 32242,
+      "ĠEnterococcus": 32243,
+      "mj": 32244,
+      "Ġhp": 32245,
+      "ĠNAS": 32246,
+      "ouss": 32247,
+      "Ġintramuscular": 32248,
+      "LIN": 32249,
+      "Ġchicks": 32250,
+      "Score": 32251,
+      "Ġfür": 32252,
+      "ĠRSA": 32253,
+      "Ġkr": 32254,
+      "Ġphotography": 32255,
+      "Ġclearing": 32256,
+      "holomorphic": 32257,
+      "them": 32258,
+      "Ġpom": 32259,
+      "ĠLis": 32260,
+      "Ġdiscard": 32261,
+      "Ġguan": 32262,
+      "cx": 32263,
+      "ubov": 32264,
+      "ĠConsistency": 32265,
+      "Ġplei": 32266,
+      "ĠUrinary": 32267,
+      "Ġbreadth": 32268,
+      "EI": 32269,
+      "mechan": 32270,
+      "Ġdq": 32271,
+      "ĠBlast": 32272,
+      "coeff": 32273,
+      "ILD": 32274,
+      "Ġunemployment": 32275,
+      "Arm": 32276,
+      "ĠCn": 32277,
+      "moderate": 32278,
+      "Ġaggress": 32279,
+      "Ġcircumf": 32280,
+      "los": 32281,
+      "Ġbaro": 32282,
+      "velope": 32283,
+      "Ġulcerative": 32284,
+      "Ġhelicase": 32285,
+      "HW": 32286,
+      "KG": 32287,
+      "rion": 32288,
+      "Ġgenotyped": 32289,
+      "Ġarid": 32290,
+      "ĠAndreas": 32291,
+      "Ġthereof": 32292,
+      "ĠOperating": 32293,
+      "ĠNEW": 32294,
+      "ĠAntibacterial": 32295,
+      "ĠDarwin": 32296,
+      "Ġreferee": 32297,
+      "Ġdome": 32298,
+      "agus": 32299,
+      "ĠDMD": 32300,
+      "ATOR": 32301,
+      "Currently": 32302,
+      "ĠInequalities": 32303,
+      "dN": 32304,
+      "olymer": 32305,
+      "empirical": 32306,
+      "ĠBraun": 32307,
+      "FIN": 32308,
+      "ĠOber": 32309,
+      "prone": 32310,
+      "Ġdiminish": 32311,
+      "ĠGraduate": 32312,
+      "ĠTSH": 32313,
+      "ĠHsu": 32314,
+      "oidosis": 32315,
+      "Ġepidural": 32316,
+      "Ġreinforcing": 32317,
+      "Ġtheatre": 32318,
+      "Ġvib": 32319,
+      "ĠHob": 32320,
+      "collection": 32321,
+      "MANGLER": 32322,
+      "ĠHecke": 32323,
+      "Ġtruck": 32324,
+      "Ġmotivates": 32325,
+      "ĠVOC": 32326,
+      "Ġunbound": 32327,
+      "ramid": 32328,
+      "iously": 32329,
+      "ĠFernández": 32330,
+      "ĠFacial": 32331,
+      "oxazol": 32332,
+      "Ġtreadm": 32333,
+      "ĠResid": 32334,
+      "Loader": 32335,
+      "ĠRunning": 32336,
+      "otinib": 32337,
+      "PAC": 32338,
+      "VII": 32339,
+      "iu": 32340,
+      "Ġcite": 32341,
+      "ĠHockey": 32342,
+      "ESC": 32343,
+      "rhoea": 32344,
+      "Ġmacaques": 32345,
+      "Ġmediast": 32346,
+      "atim": 32347,
+      "ĠTMP": 32348,
+      "ĠAGB": 32349,
+      "ĠRup": 32350,
+      "uga": 32351,
+      "Ġassurance": 32352,
+      "pay": 32353,
+      "energies": 32354,
+      "ĠKend": 32355,
+      "tillery": 32356,
+      "Ġanesthetic": 32357,
+      "Window": 32358,
+      "Ġbeverages": 32359,
+      "aguchi": 32360,
+      "ĠFLT": 32361,
+      "ĠBounded": 32362,
+      "ĠPolymerase": 32363,
+      "Sam": 32364,
+      "ĠOrbit": 32365,
+      "Ġseasonality": 32366,
+      "Ġtachycardia": 32367,
+      "esteem": 32368,
+      "ĠPerfect": 32369,
+      "SEC": 32370,
+      "later": 32371,
+      "tale": 32372,
+      "ĠFormally": 32373,
+      "LG": 32374,
+      "zyn": 32375,
+      "Ġmicroalgae": 32376,
+      "Ġindium": 32377,
+      "erennial": 32378,
+      "ĠIPT": 32379,
+      "Ġkj": 32380,
+      "ĠPDA": 32381,
+      "Ġassimil": 32382,
+      "wheel": 32383,
+      "ĠSOS": 32384,
+      "ĠPFC": 32385,
+      "Ġdecoded": 32386,
+      "ATS": 32387,
+      "Ġsocietal": 32388,
+      "Ġdiffeomorphisms": 32389,
+      "Ġtraverse": 32390,
+      "Ġcollateral": 32391,
+      "gives": 32392,
+      "ĠCEN": 32393,
+      "Ġrand": 32394,
+      "Ġherself": 32395,
+      "Ġpayments": 32396,
+      "Ġpsi": 32397,
+      "âIJ£": 32398,
+      "ĠGromov": 32399,
+      "Ġaccidental": 32400,
+      "ĠReality": 32401,
+      "Ġlogistics": 32402,
+      "Ġrobustly": 32403,
+      "ĠSarah": 32404,
+      "NU": 32405,
+      "dates": 32406,
+      "ĠCUR": 32407,
+      "ĠDream": 32408,
+      "Ġdegrades": 32409,
+      "ĠGEO": 32410,
+      "Ġbutterfly": 32411,
+      "Ġpendulum": 32412,
+      "qa": 32413,
+      "Ġaspartate": 32414,
+      "pseudo": 32415,
+      "Ġallosteric": 32416,
+      "derr": 32417,
+      "ĠQoL": 32418,
+      "Agilent": 32419,
+      "ĠHardware": 32420,
+      "ĠCumulative": 32421,
+      "Ġpn": 32422,
+      "quantitative": 32423,
+      "Ġappraisal": 32424,
+      "Ġpolyacrylamide": 32425,
+      "Ġmildly": 32426,
+      "Ġcontraceptive": 32427,
+      "ĠPublished": 32428,
+      "Ġuplift": 32429,
+      "beh": 32430,
+      "Ġadaptor": 32431,
+      "ĠEqual": 32432,
+      "thienyl": 32433,
+      "atched": 32434,
+      "Ġreply": 32435,
+      "Ġupwards": 32436,
+      "Ġautopsy": 32437,
+      "simulation": 32438,
+      "Ġgranite": 32439,
+      "Ġpelvis": 32440,
+      "Ġhatching": 32441,
+      "ĠSPS": 32442,
+      "ĠGEM": 32443,
+      "illiard": 32444,
+      "ĠRetrospective": 32445,
+      "ĠEarthqu": 32446,
+      "ĠInvestigations": 32447,
+      "ĠMerck": 32448,
+      "Ġcholangi": 32449,
+      "Ġinfiltrating": 32450,
+      "Ġoverestimated": 32451,
+      "focused": 32452,
+      "Amin": 32453,
+      "Ġpreeclampsia": 32454,
+      "ospatial": 32455,
+      "ĠTRAIL": 32456,
+      "Pair": 32457,
+      "Ġsubmarine": 32458,
+      "Ġproteolysis": 32459,
+      "Ġcomplements": 32460,
+      "ĠKirch": 32461,
+      "Ġcentrom": 32462,
+      "Ġnap": 32463,
+      "ĠWear": 32464,
+      "Ġpunishment": 32465,
+      "Ġautoregressive": 32466,
+      "Ġcomposer": 32467,
+      "ĠEngel": 32468,
+      "Ġanaemia": 32469,
+      "ĠKronecker": 32470,
+      "ĠDid": 32471,
+      "ĠCarp": 32472,
+      "peer": 32473,
+      "Ġbugs": 32474,
+      "ĠIslamic": 32475,
+      "ithromycin": 32476,
+      "Ġconsec": 32477,
+      "Ġfamiliarity": 32478,
+      "etaxel": 32479,
+      "Ġintensively": 32480,
+      "ĠUpt": 32481,
+      "Ġindica": 32482,
+      "ADA": 32483,
+      "ĠChebyshev": 32484,
+      "Ġhierarchies": 32485,
+      "Ġworthwhile": 32486,
+      "Ġburned": 32487,
+      "ĠHMGB": 32488,
+      "Ġpolygonal": 32489,
+      "brile": 32490,
+      "Ġzoon": 32491,
+      "warning": 32492,
+      "Eukaryota": 32493,
+      "dA": 32494,
+      "ĠRepeated": 32495,
+      "ĠCastro": 32496,
+      "Ġmetropolitan": 32497,
+      "ontinuous": 32498,
+      "ĠBarnes": 32499,
+      "ĠPostoperative": 32500,
+      "Ġcytology": 32501,
+      "Ġspotted": 32502,
+      "versity": 32503,
+      "affine": 32504,
+      "sorted": 32505,
+      "ĠProto": 32506,
+      "ĠDescriptive": 32507,
+      "Ġhitting": 32508,
+      "Ġanalogously": 32509,
+      "feedback": 32510,
+      "Ġspiritual": 32511,
+      "ĠLINE": 32512,
+      "ressin": 32513,
+      "ophthal": 32514,
+      "Ġpolyunsaturated": 32515,
+      "Ġpiper": 32516,
+      "observations": 32517,
+      "ĭ¤": 32518,
+      "irre": 32519,
+      "ĠWNT": 32520,
+      "Ġundifferentiated": 32521,
+      "erald": 32522,
+      "ĠCTC": 32523,
+      "Ġhomomorphisms": 32524,
+      "ĠNeonatal": 32525,
+      "Fin": 32526,
+      "rozen": 32527,
+      "ĠLux": 32528,
+      "Ġmodifier": 32529,
+      "ĠKA": 32530,
+      "osaur": 32531,
+      "Ġinterventional": 32532,
+      "ĠHapl": 32533,
+      "Ġluminance": 32534,
+      "Ġunfortunately": 32535,
+      "Ġsleeping": 32536,
+      "Ġcitrus": 32537,
+      "resonance": 32538,
+      "Ġmoss": 32539,
+      "ulay": 32540,
+      "ĠPenn": 32541,
+      "administration": 32542,
+      "ĠNGF": 32543,
+      "Ġsecured": 32544,
+      "ĠAEs": 32545,
+      "ĠPWM": 32546,
+      "occo": 32547,
+      "obuf": 32548,
+      "Ġphotocurrent": 32549,
+      "ĠScilabDouble": 32550,
+      "April": 32551,
+      "Ġforamin": 32552,
+      "Ġparalysis": 32553,
+      "ĠQuark": 32554,
+      "eqref": 32555,
+      "ĠBrooks": 32556,
+      "ĠCollision": 32557,
+      "War": 32558,
+      "Ġig": 32559,
+      "amylase": 32560,
+      "istered": 32561,
+      "Ġretraction": 32562,
+      "ĠMultiplex": 32563,
+      "ĠMao": 32564,
+      "Common": 32565,
+      "ĠEconomics": 32566,
+      "ĠCriterion": 32567,
+      "ĠCCC": 32568,
+      "ĠLei": 32569,
+      "Ġorthorhombic": 32570,
+      "Ġaliquots": 32571,
+      "Ġstric": 32572,
+      "ĠLenn": 32573,
+      "Ġdisclosure": 32574,
+      "ameth": 32575,
+      "Ġnormalisation": 32576,
+      "Ġphylogen": 32577,
+      "ĠQTLs": 32578,
+      "ĠVersus": 32579,
+      "ĠUtilization": 32580,
+      "yne": 32581,
+      "unted": 32582,
+      "ĠDuff": 32583,
+      "ĠGJ": 32584,
+      "Ġoptimised": 32585,
+      "iformis": 32586,
+      "ĠIncreases": 32587,
+      "ĠFDG": 32588,
+      "ĠBattery": 32589,
+      "Phe": 32590,
+      "ĠCCS": 32591,
+      "Ġchrys": 32592,
+      "ofen": 32593,
+      "Ġmulticomponent": 32594,
+      "discussed": 32595,
+      "bonding": 32596,
+      "oretically": 32597,
+      "ĠAlliance": 32598,
+      "Ġheadquarters": 32599,
+      "ĠGlasgow": 32600,
+      "Ġbout": 32601,
+      "Ġeighth": 32602,
+      "Ġincurred": 32603,
+      "ĠBarry": 32604,
+      "Ġquadric": 32605,
+      "Ġduties": 32606,
+      "Ġmindfulness": 32607,
+      "rastructural": 32608,
+      "Train": 32609,
+      "shitz": 32610,
+      "CDC": 32611,
+      "Ġdyslipidemia": 32612,
+      "Ġalleged": 32613,
+      "Ġbronze": 32614,
+      "Ġattainment": 32615,
+      "QD": 32616,
+      "rombin": 32617,
+      "Ġapolipoprotein": 32618,
+      "owned": 32619,
+      "Ġgeographically": 32620,
+      "working": 32621,
+      "ĠBlind": 32622,
+      "Ġdonation": 32623,
+      "ĠSerge": 32624,
+      "Ġspreads": 32625,
+      "ĠHeterogeneity": 32626,
+      "ĠFré": 32627,
+      "Ġdefer": 32628,
+      "Ġlifts": 32629,
+      "EGFR": 32630,
+      "ĠPortland": 32631,
+      "Ġbrothers": 32632,
+      "ĠTrypanosoma": 32633,
+      "inian": 32634,
+      "Ġpressed": 32635,
+      "Ġtransduced": 32636,
+      "Ġpolyn": 32637,
+      "Ġlisteners": 32638,
+      "boards": 32639,
+      "ĠSustainable": 32640,
+      "alan": 32641,
+      "ĠSullivan": 32642,
+      "Assumption": 32643,
+      "often": 32644,
+      "jp": 32645,
+      "orative": 32646,
+      "plers": 32647,
+      "Ġmodularity": 32648,
+      "ĠHermite": 32649,
+      "Ġhydroxyapatite": 32650,
+      "ĠHirsch": 32651,
+      "Determ": 32652,
+      "facing": 32653,
+      "irradiated": 32654,
+      "Ġharsh": 32655,
+      "Ġtolerate": 32656,
+      "ĠTrap": 32657,
+      "ĠAware": 32658,
+      "otax": 32659,
+      "ATING": 32660,
+      "Ġhistopathology": 32661,
+      "ĠIsraeli": 32662,
+      "clockwise": 32663,
+      "zig": 32664,
+      "ĠJC": 32665,
+      "ĠQuick": 32666,
+      "ĠSLAM": 32667,
+      "Ġfox": 32668,
+      "ĠRav": 32669,
+      "generating": 32670,
+      "Ġhematoxylin": 32671,
+      "yltransferase": 32672,
+      "Ġcorroborated": 32673,
+      "FDR": 32674,
+      "oard": 32675,
+      "Ġequid": 32676,
+      "Ġ»": 32677,
+      "Ġneuropsychological": 32678,
+      "Ġbreakup": 32679,
+      "Ġemphasizing": 32680,
+      "Ġemissivity": 32681,
+      "blocking": 32682,
+      "Ġparall": 32683,
+      "Ġtilting": 32684,
+      "Ġpeng": 32685,
+      "ĠScan": 32686,
+      "Ġionosphere": 32687,
+      "Ġmount": 32688,
+      "forest": 32689,
+      "Ġcallus": 32690,
+      "αβ": 32691,
+      "ĠChristmas": 32692,
+      "ĠMagazine": 32693,
+      "evaluate": 32694,
+      "ĠPag": 32695,
+      "ĠBeat": 32696,
+      "Ġaccumulates": 32697,
+      "Ġcrowding": 32698,
+      "unneling": 32699,
+      "ĠÑ": 32700,
+      "ĠACP": 32701,
+      "geometry": 32702,
+      "MPT": 32703,
+      "Ġpharmacists": 32704,
+      "Ġpullback": 32705,
+      "Ġductility": 32706,
+      "Supervised": 32707,
+      "Ġlymphoblastic": 32708,
+      "pea": 32709,
+      "typical": 32710,
+      "broken": 32711,
+      "Fc": 32712,
+      "Ġlining": 32713,
+      "ĠDum": 32714,
+      "Ġmultiples": 32715,
+      "ów": 32716,
+      "Ġmerits": 32717,
+      "Ġextinct": 32718,
+      "ĠNursing": 32719,
+      "ĠExploiting": 32720,
+      "ĠBhattach": 32721,
+      "July": 32722,
+      "tze": 32723,
+      "thromb": 32724,
+      "teenth": 32725,
+      "Ġtoxicities": 32726,
+      "Ġdenitrification": 32727,
+      "Ġexposition": 32728,
+      "Ġimperf": 32729,
+      "Ġsurname": 32730,
+      "pointer": 32731,
+      "ĠErn": 32732,
+      "ĠAbundance": 32733,
+      "ĠDunn": 32734,
+      "ophora": 32735,
+      "Ġtoolkit": 32736,
+      "Load": 32737,
+      "ĠDerivation": 32738,
+      "could": 32739,
+      "ĠCaspase": 32740,
+      "ĠSprague": 32741,
+      "ĠTrp": 32742,
+      "Ġbrightest": 32743,
+      "illard": 32744,
+      "Ġinterdisciplinary": 32745,
+      "Ġquarant": 32746,
+      "Ġhypersurfaces": 32747,
+      "eliac": 32748,
+      "ĠALMA": 32749,
+      "Ġacrylic": 32750,
+      "Ġgentle": 32751,
+      "Deep": 32752,
+      "ĠPandemic": 32753,
+      "Ġinfeasible": 32754,
+      "Ġradiol": 32755,
+      "ABP": 32756,
+      "Ġmesenteric": 32757,
+      "ylinder": 32758,
+      "packed": 32759,
+      "Ġsomatosensory": 32760,
+      "Ġpave": 32761,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 32762,
+      "Ġpharmacology": 32763,
+      "Ġtanh": 32764,
+      "ĠMtb": 32765,
+      "Ġchimpan": 32766,
+      "Ġautophagic": 32767,
+      "Ġwithdrawn": 32768,
+      "ĠMCC": 32769,
+      "ZF": 32770,
+      "ĠSpl": 32771,
+      "ĠLau": 32772,
+      "Ġbiologic": 32773,
+      "electrons": 32774,
+      "Ġunderestimation": 32775,
+      "Ġcharacterise": 32776,
+      "circular": 32777,
+      "ĠTHEORY": 32778,
+      "Brown": 32779,
+      "FBS": 32780,
+      "Jo": 32781,
+      "dG": 32782,
+      "mars": 32783,
+      "articular": 32784,
+      "ĠPren": 32785,
+      "ĠMSA": 32786,
+      "ĠItem": 32787,
+      "Ġsemidefinite": 32788,
+      "ĠGibson": 32789,
+      "Ġtourism": 32790,
+      "ĠKok": 32791,
+      "Ġexposing": 32792,
+      "Ġintravenously": 32793,
+      "driver": 32794,
+      "ĠFortunately": 32795,
+      "ĠSach": 32796,
+      "Ġcontaminant": 32797,
+      "Ġabrog": 32798,
+      "ĠEmotional": 32799,
+      "VALUE": 32800,
+      "dispersion": 32801,
+      "Jacobi": 32802,
+      "ĠImperial": 32803,
+      "Ion": 32804,
+      "Lin": 32805,
+      "fidelity": 32806,
+      "ĠBirds": 32807,
+      "ĠConcurrent": 32808,
+      "matism": 32809,
+      "coal": 32810,
+      "Ġtq": 32811,
+      "ĠMnO": 32812,
+      "Ġfossils": 32813,
+      "Ġtender": 32814,
+      "Ġrhesus": 32815,
+      "Ġbloom": 32816,
+      "abdominal": 32817,
+      "Ġscalp": 32818,
+      "Ġhomeostatic": 32819,
+      "ĠHunt": 32820,
+      "ĠPharmacokine": 32821,
+      "brown": 32822,
+      "ĠHYP": 32823,
+      "Ġdissociated": 32824,
+      "ĠSoccer": 32825,
+      "ĠInequality": 32826,
+      "maker": 32827,
+      "Ġshade": 32828,
+      "ĠZur": 32829,
+      "observation": 32830,
+      "altered": 32831,
+      "UU": 32832,
+      "Ġtheor": 32833,
+      "epit": 32834,
+      "Ġphylum": 32835,
+      "Ġvigorous": 32836,
+      "ĠACM": 32837,
+      "Ġmethotrexate": 32838,
+      "demographic": 32839,
+      "Ġsingly": 32840,
+      "ĠPhysiology": 32841,
+      "Ġremodelling": 32842,
+      "ĠKrist": 32843,
+      "ropies": 32844,
+      "flows": 32845,
+      "hardness": 32846,
+      "ighteen": 32847,
+      "breve": 32848,
+      "ĠRetinal": 32849,
+      "Ġscintill": 32850,
+      "Ġutterance": 32851,
+      "Ġmonolithic": 32852,
+      "ĠVlad": 32853,
+      "ĠLMC": 32854,
+      "ipt": 32855,
+      "arrows": 32856,
+      "ĠPublishing": 32857,
+      "ĠStreptomyces": 32858,
+      "fal": 32859,
+      "Ġtroposphere": 32860,
+      "Ben": 32861,
+      "candid": 32862,
+      "ĠSic": 32863,
+      "timore": 32864,
+      "Len": 32865,
+      "inen": 32866,
+      "ampered": 32867,
+      "ĠMonth": 32868,
+      "Ġopponent": 32869,
+      "August": 32870,
+      "Ġstaggered": 32871,
+      "centre": 32872,
+      "expect": 32873,
+      "Ġreddening": 32874,
+      "ĠTl": 32875,
+      "hibition": 32876,
+      "Ġmicroparticles": 32877,
+      "ĠIntrac": 32878,
+      "ĠInitialize": 32879,
+      "Ġdictated": 32880,
+      "Dig": 32881,
+      "äº": 32882,
+      "healing": 32883,
+      "ĠdV": 32884,
+      "Ġappetite": 32885,
+      "Ġunusually": 32886,
+      "ĠAstronomy": 32887,
+      "Ġware": 32888,
+      "Ġovercoming": 32889,
+      "Ġcolliders": 32890,
+      "ĠUSING": 32891,
+      "ocarditis": 32892,
+      "Pick": 32893,
+      "Ġdub": 32894,
+      "ĠJason": 32895,
+      "ĠEditor": 32896,
+      "ê³": 32897,
+      "Ġlags": 32898,
+      "Ġcls": 32899,
+      "Ġsurgically": 32900,
+      "ĠPVC": 32901,
+      "particularly": 32902,
+      "Ġredist": 32903,
+      "Ġlogics": 32904,
+      "skii": 32905,
+      "ĠDVD": 32906,
+      "Ġcomply": 32907,
+      "azi": 32908,
+      "ĠInteracts": 32909,
+      "boolean": 32910,
+      "ĠERP": 32911,
+      "ĠErr": 32912,
+      "otranspiration": 32913,
+      "ĠPérez": 32914,
+      "Asp": 32915,
+      "amiliar": 32916,
+      "ĠFetal": 32917,
+      "Ġdeclaration": 32918,
+      "kinson": 32919,
+      "tube": 32920,
+      "Ġphysiologically": 32921,
+      "cue": 32922,
+      "ĠEri": 32923,
+      "Ġenvision": 32924,
+      "external": 32925,
+      "intermediate": 32926,
+      "Ġshopping": 32927,
+      "ĠFras": 32928,
+      "ĠHaj": 32929,
+      "ĠAlger": 32930,
+      "Ġanthropometric": 32931,
+      "Ġcancelled": 32932,
+      "HPV": 32933,
+      "kers": 32934,
+      "afa": 32935,
+      "Ġvulnerabilities": 32936,
+      "electrolyte": 32937,
+      "ĠGonzalez": 32938,
+      "íķĺ": 32939,
+      "qv": 32940,
+      "Ġdeaf": 32941,
+      "Ġbutyrate": 32942,
+      "ĠCoefficient": 32943,
+      "Ġstarburst": 32944,
+      "Ġpolymorph": 32945,
+      "ĠERA": 32946,
+      "ĠMaximal": 32947,
+      "ĠMueller": 32948,
+      "Ġabsorbers": 32949,
+      "Ġarab": 32950,
+      "retions": 32951,
+      "Ġnebula": 32952,
+      "Ġmines": 32953,
+      "ен": 32954,
+      "%%%%%%%%%%%%%%%%": 32955,
+      "Ġbandpass": 32956,
+      "Ġpolyurethane": 32957,
+      "ReLU": 32958,
+      "ĠFerro": 32959,
+      "picillin": 32960,
+      "CAD": 32961,
+      "Ty": 32962,
+      "ĠPCD": 32963,
+      "ĠBAC": 32964,
+      "Ġplanktonic": 32965,
+      "Fer": 32966,
+      "Ġcricket": 32967,
+      "Ġmanure": 32968,
+      "ouns": 32969,
+      "âĪ§": 32970,
+      "Ġtorques": 32971,
+      "mitian": 32972,
+      "Ġtion": 32973,
+      "ĠGarden": 32974,
+      "Ġfolk": 32975,
+      "Ġsuspicious": 32976,
+      "ÃĤ": 32977,
+      "odia": 32978,
+      "istencies": 32979,
+      "ãĢī": 32980,
+      "ĠInvitrogen": 32981,
+      "ĠSUN": 32982,
+      "ĠSuperior": 32983,
+      "Ġdiscontinuation": 32984,
+      "cock": 32985,
+      "knot": 32986,
+      "Ġextens": 32987,
+      "ĠWhitney": 32988,
+      "Ġharbour": 32989,
+      "PID": 32990,
+      "Ġpmol": 32991,
+      "olymph": 32992,
+      "Ġgard": 32993,
+      "ĠOvarian": 32994,
+      "Ġrepressed": 32995,
+      "ĠAlab": 32996,
+      "ĠÃĦ": 32997,
+      "ulex": 32998,
+      "ĠAustrian": 32999,
+      "Ġaflat": 33000,
+      "Ġparathyroid": 33001,
+      "Ġgroupoid": 33002,
+      "Ġdevast": 33003,
+      "ĠKv": 33004,
+      "Ġborrow": 33005,
+      "Ġunconventional": 33006,
+      "Ġborehole": 33007,
+      "ÑĮ": 33008,
+      "ĠDays": 33009,
+      "Ġlexic": 33010,
+      "Nor": 33011,
+      "ĠHerc": 33012,
+      "assays": 33013,
+      "Ġdrawings": 33014,
+      "defin": 33015,
+      "evoked": 33016,
+      "Ġȳ": 33017,
+      "ĠSunday": 33018,
+      "ĠChes": 33019,
+      "considered": 33020,
+      "opedic": 33021,
+      "larger": 33022,
+      "ominant": 33023,
+      "ĠBomb": 33024,
+      "Ġfiss": 33025,
+      "Ġhinge": 33026,
+      "ĠIonic": 33027,
+      "Ġdestro": 33028,
+      "Ġcomplementarity": 33029,
+      "Higgs": 33030,
+      "oria": 33031,
+      "ourcing": 33032,
+      "ĠXin": 33033,
+      "Ġworkspace": 33034,
+      "ĠLigand": 33035,
+      "Ġstruggle": 33036,
+      "ĠImmunohistochemical": 33037,
+      "Ġnick": 33038,
+      "ĠGuard": 33039,
+      "rigid": 33040,
+      "Ġaquaculture": 33041,
+      "Experiment": 33042,
+      "ËĪ": 33043,
+      "tir": 33044,
+      "ĠSMS": 33045,
+      "Ġbevacizumab": 33046,
+      "Ġmodulations": 33047,
+      "Ġgeophysical": 33048,
+      "Properties": 33049,
+      "Ġpainted": 33050,
+      "Ġsanc": 33051,
+      "Ġintimate": 33052,
+      "Ġnail": 33053,
+      "identity": 33054,
+      "Ġdatum": 33055,
+      "anthus": 33056,
+      "Ġdyadic": 33057,
+      "Ġconvincing": 33058,
+      "elem": 33059,
+      "Ġhiding": 33060,
+      "Ġrugby": 33061,
+      "ĠXe": 33062,
+      "ĠIssue": 33063,
+      "Ġvesicular": 33064,
+      "ĠKelvin": 33065,
+      "Ġdistancing": 33066,
+      "echnology": 33067,
+      "afers": 33068,
+      "ĠAuthentic": 33069,
+      "PubMed": 33070,
+      "Ġdeformity": 33071,
+      "ĠChaos": 33072,
+      "ĠShield": 33073,
+      "oxetine": 33074,
+      "ĠWorkers": 33075,
+      "ĠMOI": 33076,
+      "Ġdehydrated": 33077,
+      "ĠGastric": 33078,
+      "Ġmonomials": 33079,
+      "odox": 33080,
+      "ĠDublin": 33081,
+      "Ġleishman": 33082,
+      "Ġplanner": 33083,
+      "circle": 33084,
+      "Ġfractured": 33085,
+      "ĠLocally": 33086,
+      "ĠActions": 33087,
+      "Ġlichen": 33088,
+      "hannel": 33089,
+      "ĠTAG": 33090,
+      "Ġdecisive": 33091,
+      "ĠQM": 33092,
+      "Ġbiomaterials": 33093,
+      "ĠViruses": 33094,
+      "hydroxyphenyl": 33095,
+      "ĠIAA": 33096,
+      "ĠRU": 33097,
+      "violating": 33098,
+      "Ġpockets": 33099,
+      "chant": 33100,
+      "iberg": 33101,
+      "lectomy": 33102,
+      "olerae": 33103,
+      "Ġattracting": 33104,
+      "Ġketone": 33105,
+      "ĠCod": 33106,
+      "Ġmicroarrays": 33107,
+      "ĠMetals": 33108,
+      "benzoyl": 33109,
+      "Ġsemigroups": 33110,
+      "Ġreconstituted": 33111,
+      "sites": 33112,
+      "anabe": 33113,
+      "ĠComposites": 33114,
+      "Ġwildtype": 33115,
+      "Ġleukaemia": 33116,
+      "Ġmurder": 33117,
+      "Ġdentin": 33118,
+      "Hub": 33119,
+      "Orient": 33120,
+      "onn": 33121,
+      "synchron": 33122,
+      "Ġchronically": 33123,
+      "methyleneamino": 33124,
+      "Ġdopant": 33125,
+      "Ġfecundity": 33126,
+      "delete": 33127,
+      "remia": 33128,
+      "ĠNHL": 33129,
+      "itidis": 33130,
+      "Ġcopep": 33131,
+      "XI": 33132,
+      "Ġlocating": 33133,
+      "ĠZIKV": 33134,
+      "hexa": 33135,
+      "ĠFactorization": 33136,
+      "ynchus": 33137,
+      "Methyl": 33138,
+      "hagen": 33139,
+      "ĠPaw": 33140,
+      "neath": 33141,
+      "bsite": 33142,
+      "Ġtrache": 33143,
+      "Bre": 33144,
+      "uw": 33145,
+      "roit": 33146,
+      "Ġreacting": 33147,
+      "ĠBae": 33148,
+      "Ġquotients": 33149,
+      "Ġpins": 33150,
+      "ĠVARI": 33151,
+      "Ġequine": 33152,
+      "ĠRunge": 33153,
+      "Ġcolonial": 33154,
+      "measurement": 33155,
+      "ĠAbbott": 33156,
+      "Ġortho": 33157,
+      "Ġmetaphor": 33158,
+      "benzoic": 33159,
+      "ĠTransformers": 33160,
+      "Lower": 33161,
+      "ĠOVA": 33162,
+      "radial": 33163,
+      "Flag": 33164,
+      "authorbs": 33165,
+      "Ġtreadmill": 33166,
+      "Ġenterica": 33167,
+      "ĠJulia": 33168,
+      "Ġplumes": 33169,
+      "Ġinvoke": 33170,
+      "chloric": 33171,
+      "olino": 33172,
+      "Ġinterruption": 33173,
+      "subunit": 33174,
+      "ĠMDP": 33175,
+      "Ġmanipulator": 33176,
+      "ĠScales": 33177,
+      "ĠHTML": 33178,
+      "ĠFrederick": 33179,
+      "Garc": 33180,
+      "Ġbell": 33181,
+      "ĠRect": 33182,
+      "romised": 33183,
+      "Word": 33184,
+      "oples": 33185,
+      "operated": 33186,
+      "Ġcollects": 33187,
+      "ĠHorizon": 33188,
+      "Ġsafer": 33189,
+      "dup": 33190,
+      "ĠMills": 33191,
+      "ALP": 33192,
+      "Ġexopl": 33193,
+      "ATTR": 33194,
+      "wara": 33195,
+      "ĉĉĉĉĉĉĉ": 33196,
+      "Ġdebug": 33197,
+      "Descriptor": 33198,
+      "statistics": 33199,
+      "ĠCub": 33200,
+      "STER": 33201,
+      "ĠStabilization": 33202,
+      "ĠIRAS": 33203,
+      "Ġconformally": 33204,
+      "Adap": 33205,
+      "ÂŃ": 33206,
+      "ĠQS": 33207,
+      "Ġmicrostrip": 33208,
+      "Ġdelicate": 33209,
+      "Ġpublisher": 33210,
+      "Ġhos": 33211,
+      "ĠSv": 33212,
+      "ĠDesert": 33213,
+      "ĠGuer": 33214,
+      "ĠCapture": 33215,
+      "EBP": 33216,
+      "dust": 33217,
+      "å¤": 33218,
+      "ĠOls": 33219,
+      "Ġsuperscript": 33220,
+      "ĠFluctuations": 33221,
+      "illium": 33222,
+      "Ġcaption": 33223,
+      "Ġconcur": 33224,
+      "Ġquantifies": 33225,
+      "sterdam": 33226,
+      "Ġspiked": 33227,
+      "Nan": 33228,
+      "usin": 33229,
+      "ĠLAN": 33230,
+      "Ġobserves": 33231,
+      "ĠAla": 33232,
+      "ĠIntuitively": 33233,
+      "curr": 33234,
+      "Ġshrinking": 33235,
+      "Ġcompressibility": 33236,
+      "orporeal": 33237,
+      "Ġdebt": 33238,
+      "çĶ": 33239,
+      "ĠTil": 33240,
+      "ĠWAT": 33241,
+      "odyne": 33242,
+      "Ġgateway": 33243,
+      "Ġductile": 33244,
+      "ĠJesus": 33245,
+      "ositol": 33246,
+      "ĠMales": 33247,
+      "Ġsolvation": 33248,
+      "Ġdisagree": 33249,
+      "Ġorthologs": 33250,
+      "San": 33251,
+      "igo": 33252,
+      "Ġphages": 33253,
+      "Ġnegatives": 33254,
+      "Ġinterpre": 33255,
+      "AAA": 33256,
+      "Ġgratings": 33257,
+      "ĠMoll": 33258,
+      "ĠRivers": 33259,
+      "Ġcruzi": 33260,
+      "ĠGenerate": 33261,
+      "ĠBarbara": 33262,
+      "ĠHeritage": 33263,
+      "ĠFluorescent": 33264,
+      "ĠLaws": 33265,
+      "ArrayExpr": 33266,
+      "Ġmultipole": 33267,
+      "Ġsqueezing": 33268,
+      "SPSS": 33269,
+      "lf": 33270,
+      "nlm": 33271,
+      "Ġworn": 33272,
+      "ĠKuz": 33273,
+      "Ġgenesis": 33274,
+      "ĠEmperor": 33275,
+      "volatile": 33276,
+      "Ġsibling": 33277,
+      "ivir": 33278,
+      "oen": 33279,
+      "Ġprotost": 33280,
+      "Ġtransformers": 33281,
+      "ennium": 33282,
+      "Ġproposing": 33283,
+      "Ġbroadcasting": 33284,
+      "QM": 33285,
+      "ĠDependent": 33286,
+      "Ġdisable": 33287,
+      "ĠUAS": 33288,
+      "Ġwarnings": 33289,
+      "Ġarmed": 33290,
+      "Ġjournalist": 33291,
+      "Ġmonoclinic": 33292,
+      "olium": 33293,
+      "aping": 33294,
+      "toon": 33295,
+      "Ġorthodontic": 33296,
+      "ĠNormalization": 33297,
+      "Ġmandible": 33298,
+      "aban": 33299,
+      "ĠWak": 33300,
+      "extend": 33301,
+      "Multiple": 33302,
+      "investig": 33303,
+      "iscal": 33304,
+      "uttered": 33305,
+      "Ġburg": 33306,
+      "decode": 33307,
+      "empor": 33308,
+      "ĠDuration": 33309,
+      "anny": 33310,
+      "oprost": 33311,
+      "ĠRenormalization": 33312,
+      "ĠFUNCTION": 33313,
+      "ytorch": 33314,
+      "Ġsynapt": 33315,
+      "ĠFormat": 33316,
+      "ĠCRT": 33317,
+      "ĠJonathan": 33318,
+      "ĠOFF": 33319,
+      "orr": 33320,
+      "Ġresur": 33321,
+      "Ġcorruption": 33322,
+      "dwelling": 33323,
+      "Ġbackup": 33324,
+      "AGT": 33325,
+      "ĠSafe": 33326,
+      "dorfer": 33327,
+      "Ġataxia": 33328,
+      "Ġparv": 33329,
+      "reader": 33330,
+      "Ġsubtract": 33331,
+      "embolism": 33332,
+      "Ġtinnitus": 33333,
+      "Ġcytomegalovirus": 33334,
+      "Ġdeleting": 33335,
+      "Tex": 33336,
+      "ĠCSS": 33337,
+      "ardt": 33338,
+      "Ġoutgrowth": 33339,
+      "Ġmyocytes": 33340,
+      "digital": 33341,
+      "Ġsubscale": 33342,
+      "uspension": 33343,
+      "Ġhamster": 33344,
+      "Ġinflaton": 33345,
+      "hara": 33346,
+      "urches": 33347,
+      "ĠCLE": 33348,
+      "ĠYas": 33349,
+      "ĠEncoding": 33350,
+      "ĠAuger": 33351,
+      "Ġanastomosis": 33352,
+      "Agent": 33353,
+      "ĠSIL": 33354,
+      "ĠCCT": 33355,
+      "Ġbrine": 33356,
+      "Ġoligo": 33357,
+      "Ġfluoro": 33358,
+      "Ġgallery": 33359,
+      "ddots": 33360,
+      "Ġcilia": 33361,
+      "ĠPPV": 33362,
+      "ĠUTR": 33363,
+      "Ġintertidal": 33364,
+      "ocalized": 33365,
+      "Ġcrowds": 33366,
+      "odor": 33367,
+      "Ġcov": 33368,
+      "Ġnonetheless": 33369,
+      "Ġïģ¤": 33370,
+      "Ġboosted": 33371,
+      "ĠChakra": 33372,
+      "Hal": 33373,
+      "Pear": 33374,
+      "Ġimprec": 33375,
+      "ĠSupplement": 33376,
+      "goal": 33377,
+      "Ġôı¼ģ": 33378,
+      "Ġstall": 33379,
+      "Ġherd": 33380,
+      "smaller": 33381,
+      "Ġreconstructing": 33382,
+      "Ġartefacts": 33383,
+      "Ġteg": 33384,
+      "conventional": 33385,
+      "radical": 33386,
+      "Ġliteral": 33387,
+      "framework": 33388,
+      "iprocal": 33389,
+      "EEG": 33390,
+      "Ġgins": 33391,
+      "odermal": 33392,
+      "ĠAgu": 33393,
+      "ĠTwelve": 33394,
+      "Mul": 33395,
+      "ب": 33396,
+      "irl": 33397,
+      "ĠBelief": 33398,
+      "Ġincont": 33399,
+      "ICC": 33400,
+      "hexane": 33401,
+      "Ġejected": 33402,
+      "ĠPSC": 33403,
+      "ĠHPC": 33404,
+      "ĠVH": 33405,
+      "Ġequivalences": 33406,
+      "plotlib": 33407,
+      "enital": 33408,
+      "rians": 33409,
+      "prov": 33410,
+      "ĠVibr": 33411,
+      "Ġgrammatical": 33412,
+      "bachia": 33413,
+      "acceptable": 33414,
+      "odicity": 33415,
+      "abb": 33416,
+      "Ġherbs": 33417,
+      "Ġpredominance": 33418,
+      "ĠOrientation": 33419,
+      "Ġinvertebrate": 33420,
+      "Ġpelagic": 33421,
+      "country": 33422,
+      "ĠOrigins": 33423,
+      "ĠAdolescents": 33424,
+      "ĠTuning": 33425,
+      "rainian": 33426,
+      "ĠScar": 33427,
+      "Ġlightest": 33428,
+      "Ġemitters": 33429,
+      "ĠTsai": 33430,
+      "ritical": 33431,
+      "ĠExpert": 33432,
+      "authors": 33433,
+      "ECTION": 33434,
+      "ĠSeverity": 33435,
+      "Nam": 33436,
+      "publ": 33437,
+      "ĠAbe": 33438,
+      "Ġnanocrystalline": 33439,
+      "ĠNakamura": 33440,
+      "ĠPec": 33441,
+      "ĠBug": 33442,
+      "Ġsensed": 33443,
+      "ONS": 33444,
+      "ICs": 33445,
+      "Ġelectrochem": 33446,
+      "ĠROM": 33447,
+      "ĠRecruitment": 33448,
+      "Ġ⣩": 33449,
+      "Ġbiomolecules": 33450,
+      "ĠBrac": 33451,
+      "Ġtransposition": 33452,
+      "ĠWP": 33453,
+      "ĠOmega": 33454,
+      "Ġdiagon": 33455,
+      "platelet": 33456,
+      "JM": 33457,
+      "acre": 33458,
+      "ĠASR": 33459,
+      "ĠKath": 33460,
+      "Ġpriv": 33461,
+      "oplasts": 33462,
+      "Samples": 33463,
+      "dF": 33464,
+      "atti": 33465,
+      "ĠSanger": 33466,
+      "ipitated": 33467,
+      "Ġricher": 33468,
+      "ĠGRA": 33469,
+      "Ġplantar": 33470,
+      "Ġfoams": 33471,
+      "Ġmathematic": 33472,
+      "Ġstaphyl": 33473,
+      "ĠUptake": 33474,
+      "Ġcant": 33475,
+      "ĠSZ": 33476,
+      "Ġdismiss": 33477,
+      "Ġselections": 33478,
+      "plitz": 33479,
+      "Ġexemplified": 33480,
+      "Ġtorsional": 33481,
+      "Ev": 33482,
+      "Ġvoters": 33483,
+      "ĠNest": 33484,
+      "yscale": 33485,
+      "Ġspeci": 33486,
+      "Ġpolished": 33487,
+      "Ġlatencies": 33488,
+      "qing": 33489,
+      "Ġonwards": 33490,
+      "llvm": 33491,
+      "theorem": 33492,
+      "logging": 33493,
+      "ĠALK": 33494,
+      "ĠBaum": 33495,
+      "ĠGhosh": 33496,
+      "Ġchairman": 33497,
+      "paired": 33498,
+      "ĠPAP": 33499,
+      "notes": 33500,
+      "olesterolem": 33501,
+      "Ġestuarine": 33502,
+      "ĠTibetan": 33503,
+      "ĠVER": 33504,
+      "Ġchecker": 33505,
+      "FLAGS": 33506,
+      "rolimus": 33507,
+      "ĠMutant": 33508,
+      "Ġspraying": 33509,
+      "ĠChest": 33510,
+      "olinium": 33511,
+      "ĠTriassic": 33512,
+      "Ġlidar": 33513,
+      "Art": 33514,
+      "ĠMilk": 33515,
+      "Ġindecomposable": 33516,
+      "Ġrocket": 33517,
+      "ĠPartners": 33518,
+      "Ġsemantically": 33519,
+      "entinel": 33520,
+      "Large": 33521,
+      "Pen": 33522,
+      "ĠTru": 33523,
+      "Ġheritage": 33524,
+      "ĠMutual": 33525,
+      "ĠChemotherapy": 33526,
+      "Ġdoubles": 33527,
+      "ĠEmbedded": 33528,
+      "itual": 33529,
+      "ĠBPA": 33530,
+      "Ġcholerae": 33531,
+      "ĠInside": 33532,
+      "ĠKatz": 33533,
+      "convergence": 33534,
+      "Ġindividualized": 33535,
+      "kinje": 33536,
+      "Ġdiscovering": 33537,
+      "Ġintricate": 33538,
+      "Ġinland": 33539,
+      "RECT": 33540,
+      "ĠChick": 33541,
+      "ĠSUR": 33542,
+      "Ġyeasts": 33543,
+      "luminosity": 33544,
+      "Ġfain": 33545,
+      "ioni": 33546,
+      "ĠTig": 33547,
+      "ounder": 33548,
+      "Ġdeliber": 33549,
+      "ĠConservative": 33550,
+      "ĠDelhi": 33551,
+      "BER": 33552,
+      "ĠYB": 33553,
+      "oley": 33554,
+      "ĠBeau": 33555,
+      "TEXT": 33556,
+      "Ġsqueezed": 33557,
+      "Ġsocket": 33558,
+      "ĠpT": 33559,
+      "pyrazol": 33560,
+      "coefficients": 33561,
+      "Ġrecruiting": 33562,
+      "Ġducts": 33563,
+      "Ġfoster": 33564,
+      "omeration": 33565,
+      "ĠPSI": 33566,
+      "ĠDup": 33567,
+      "Ġks": 33568,
+      "ĠOptics": 33569,
+      "Ġliterary": 33570,
+      "ĠNiO": 33571,
+      "ĠVEGFR": 33572,
+      "Ġgraviton": 33573,
+      "Ġutterances": 33574,
+      "Ġbrady": 33575,
+      "Ġforty": 33576,
+      "ĠTransplantation": 33577,
+      "Ġagreements": 33578,
+      "Leftrightarrow": 33579,
+      "waves": 33580,
+      "Ġacidosis": 33581,
+      "Ġwooden": 33582,
+      "ĠCytoplasmic": 33583,
+      "safe": 33584,
+      "Ġjumping": 33585,
+      "ennial": 33586,
+      "Various": 33587,
+      "ĠEryth": 33588,
+      "ulins": 33589,
+      "unlock": 33590,
+      "methylated": 33591,
+      "asserstein": 33592,
+      "Ġheterozygosity": 33593,
+      "oxycycl": 33594,
+      "Ġcreativity": 33595,
+      "MPLE": 33596,
+      "inative": 33597,
+      "Ġconvolutions": 33598,
+      "Ġnouns": 33599,
+      "egan": 33600,
+      "ĠAbraham": 33601,
+      "Ġdenser": 33602,
+      "Che": 33603,
+      "lc": 33604,
+      "ĉĉĉĠ": 33605,
+      "Ġsemim": 33606,
+      "ĠOuter": 33607,
+      "Ġcand": 33608,
+      "odule": 33609,
+      "esthesia": 33610,
+      "ĠJoy": 33611,
+      "ĠProtocols": 33612,
+      "ĠCalculated": 33613,
+      "atop": 33614,
+      "ĠFALSE": 33615,
+      "Ġrefin": 33616,
+      "Ġmigrants": 33617,
+      "ĠïĤ´": 33618,
+      "ĠSpecificity": 33619,
+      "ĠFellowship": 33620,
+      "ĠPMT": 33621,
+      "Ġdisclose": 33622,
+      "unches": 33623,
+      "Ġdiatoms": 33624,
+      "corr": 33625,
+      "Ġskyrm": 33626,
+      "Ġrenewal": 33627,
+      "gcd": 33628,
+      "cereb": 33629,
+      "Ġupright": 33630,
+      "Ġmesoscopic": 33631,
+      "hydraz": 33632,
+      "BAS": 33633,
+      "FLO": 33634,
+      "HCC": 33635,
+      "Mouse": 33636,
+      "Ġposet": 33637,
+      "Ġproteinuria": 33638,
+      "Ġreapp": 33639,
+      "ĠNickel": 33640,
+      "Ġstripes": 33641,
+      "Ġripple": 33642,
+      "September": 33643,
+      "odomain": 33644,
+      "ĠPope": 33645,
+      "ĠNons": 33646,
+      "Ġtechnic": 33647,
+      "Ġneutrop": 33648,
+      "descriptor": 33649,
+      "Ġdissipated": 33650,
+      "Ġglaciers": 33651,
+      "ĠHIGH": 33652,
+      "ĠLav": 33653,
+      "retely": 33654,
+      "Ġbackwards": 33655,
+      "Ġcritics": 33656,
+      "ĠExtending": 33657,
+      "bic": 33658,
+      "ĠChao": 33659,
+      "ofibr": 33660,
+      "Ġcounters": 33661,
+      "Ġstreets": 33662,
+      "Ġprosthetic": 33663,
+      "Ġbiodegradation": 33664,
+      "complexity": 33665,
+      "ĠSPL": 33666,
+      "ĠCAC": 33667,
+      "Ġadducts": 33668,
+      "Ġmorphometric": 33669,
+      "ĠMatt": 33670,
+      "Ġinducer": 33671,
+      "Ġastrocyte": 33672,
+      "Ġtriplets": 33673,
+      "Ġpertussis": 33674,
+      "PES": 33675,
+      "idy": 33676,
+      "uncertain": 33677,
+      "Ġhyperparameter": 33678,
+      "ĠInfrastructure": 33679,
+      "ìĿĺ": 33680,
+      "ZW": 33681,
+      "Ġaddr": 33682,
+      "Ġdisrupts": 33683,
+      "Ġoverestimate": 33684,
+      "ĠDYNA": 33685,
+      "Ġvolatiles": 33686,
+      "emerg": 33687,
+      "issue": 33688,
+      "cpp": 33689,
+      "Äħ": 33690,
+      "ĠVIP": 33691,
+      "Ġuve": 33692,
+      "ĠCNV": 33693,
+      "ylethyl": 33694,
+      "onazole": 33695,
+      "ĠHiro": 33696,
+      "Ġcn": 33697,
+      "tik": 33698,
+      "ubted": 33699,
+      "ĠJacobs": 33700,
+      "Ġadvocated": 33701,
+      "ĠBifid": 33702,
+      "material": 33703,
+      "Ġstyrene": 33704,
+      "ĠKeller": 33705,
+      "rocytic": 33706,
+      "pinephrine": 33707,
+      "ĠWritten": 33708,
+      "ĠRecommendation": 33709,
+      "bled": 33710,
+      "ĠBootstrap": 33711,
+      "thirds": 33712,
+      "Ġcaptain": 33713,
+      "equals": 33714,
+      "SRC": 33715,
+      "ĠKentucky": 33716,
+      "Ġeosinophils": 33717,
+      "Average": 33718,
+      "Hi": 33719,
+      "Whe": 33720,
+      "ĠDAT": 33721,
+      "ĠUM": 33722,
+      "Ġtendencies": 33723,
+      "ĠPeterson": 33724,
+      "Ġoccult": 33725,
+      "Ġexhibition": 33726,
+      "ĠINS": 33727,
+      "Ġadipocyte": 33728,
+      "Just": 33729,
+      "hift": 33730,
+      "tensors": 33731,
+      "Ġciliary": 33732,
+      "ipation": 33733,
+      "Ġmotivations": 33734,
+      "Ġwitnessed": 33735,
+      "itches": 33736,
+      "ĠSoy": 33737,
+      "Ġgib": 33738,
+      "eptic": 33739,
+      "ĠKOH": 33740,
+      "Ġïģ¨": 33741,
+      "ĠTorres": 33742,
+      "ο": 33743,
+      "arpo": 33744,
+      "okinase": 33745,
+      "ĠBudd": 33746,
+      "ĠGMM": 33747,
+      "Ġunderpin": 33748,
+      "Ġoptimistic": 33749,
+      "ogeography": 33750,
+      "numerical": 33751,
+      "ogg": 33752,
+      "Ġdisequilibrium": 33753,
+      "Ġswab": 33754,
+      "EDS": 33755,
+      "ĠPDFs": 33756,
+      "ĠSupernova": 33757,
+      "phospho": 33758,
+      "Ġlysosomes": 33759,
+      "galactic": 33760,
+      "ĠPerme": 33761,
+      "Ġfishery": 33762,
+      "ĠBOLD": 33763,
+      "Ġunravel": 33764,
+      "ĠEncryption": 33765,
+      "JP": 33766,
+      "hur": 33767,
+      "Ġdiscount": 33768,
+      "ĠWatanabe": 33769,
+      "ĠRheumat": 33770,
+      "FITC": 33771,
+      "Ġterahertz": 33772,
+      "ĠFont": 33773,
+      "iances": 33774,
+      "ĠAdditive": 33775,
+      "ĠEither": 33776,
+      "metadata": 33777,
+      "amphetamine": 33778,
+      "ĠPalmer": 33779,
+      "Ġleveraging": 33780,
+      "John": 33781,
+      "OCT": 33782,
+      "infer": 33783,
+      "ĠMSD": 33784,
+      "ĠâĪĵ": 33785,
+      "ouver": 33786,
+      "ĠAndersen": 33787,
+      "Ġworlds": 33788,
+      "Ġtori": 33789,
+      "Ġïģ°": 33790,
+      "engineering": 33791,
+      "ĠSquadron": 33792,
+      "Aff": 33793,
+      "åı": 33794,
+      "oxel": 33795,
+      "yletic": 33796,
+      "ĠCharacterizing": 33797,
+      "VT": 33798,
+      "rational": 33799,
+      "eremia": 33800,
+      "Ġcomplexation": 33801,
+      "ĠERα": 33802,
+      "carboxylic": 33803,
+      "ïĤ·": 33804,
+      "Ġgalactose": 33805,
+      "ĠAurora": 33806,
+      "Ġplasminogen": 33807,
+      "uren": 33808,
+      "igne": 33809,
+      "Ġrepaired": 33810,
+      "Ġblockers": 33811,
+      "ĠMNIST": 33812,
+      "Ïħ": 33813,
+      "ĠAxi": 33814,
+      "Ġstadium": 33815,
+      "diethyl": 33816,
+      "âĢİ": 33817,
+      "Ġcyclotron": 33818,
+      "Ġlymphaden": 33819,
+      "Ġvin": 33820,
+      "ĠMayer": 33821,
+      "Ġendometrium": 33822,
+      "ĠSpherical": 33823,
+      "Ġpersu": 33824,
+      "Ġimmortal": 33825,
+      "benzenesulf": 33826,
+      "ĠÅľ": 33827,
+      "Ġbite": 33828,
+      "ugged": 33829,
+      "ĠDiffraction": 33830,
+      "GTG": 33831,
+      "iate": 33832,
+      "Ġtp": 33833,
+      "Ġaber": 33834,
+      "ĠRein": 33835,
+      "Program": 33836,
+      "Style": 33837,
+      "ĠRegularization": 33838,
+      "ĠLeukemia": 33839,
+      "Ġprokaryotic": 33840,
+      "ocomial": 33841,
+      "skb": 33842,
+      "Ġdeviates": 33843,
+      "Ġfuse": 33844,
+      "ĠNull": 33845,
+      "ĠïĥĹ": 33846,
+      "ĠOperational": 33847,
+      "Ġcompressor": 33848,
+      "ĠRydberg": 33849,
+      "Ġfought": 33850,
+      "Ġeco": 33851,
+      "ĠSSP": 33852,
+      "CDs": 33853,
+      "ĠMEK": 33854,
+      "ĠAnisotropic": 33855,
+      "ĠDirection": 33856,
+      "ĠSpectrometry": 33857,
+      "Ġgluten": 33858,
+      "ĠPowell": 33859,
+      "recognized": 33860,
+      "Ġpsychotic": 33861,
+      "Ġhinder": 33862,
+      "Ġaccommodation": 33863,
+      "ĠNorman": 33864,
+      "Qx": 33865,
+      "Ġperiv": 33866,
+      "ĠUnknown": 33867,
+      "Ġjoins": 33868,
+      "ĠMinimization": 33869,
+      "ĠSons": 33870,
+      "ĠCin": 33871,
+      "Ġunavoid": 33872,
+      "ĠPTX": 33873,
+      "Ġcada": 33874,
+      "ĠLuk": 33875,
+      "Ġruling": 33876,
+      "Ġbiphasic": 33877,
+      "ĠComplications": 33878,
+      "ĠDefects": 33879,
+      "Content": 33880,
+      "ĠGregory": 33881,
+      "ĠWerner": 33882,
+      "ĠWeibull": 33883,
+      "eldom": 33884,
+      "Ġactivators": 33885,
+      "GLAPI": 33886,
+      "mathring": 33887,
+      "Ġhens": 33888,
+      "NSC": 33889,
+      "however": 33890,
+      "ĠTME": 33891,
+      "mafrost": 33892,
+      "coefficient": 33893,
+      "ĠInsect": 33894,
+      "ĠROIs": 33895,
+      "ĠBorrel": 33896,
+      "ĠQiu": 33897,
+      "Ġinhaled": 33898,
+      "idate": 33899,
+      "Ġantihypertensive": 33900,
+      "Ġtreats": 33901,
+      "ĠNearly": 33902,
+      "succ": 33903,
+      "ĠOrbital": 33904,
+      "eradish": 33905,
+      "administered": 33906,
+      "ĠÏĤ": 33907,
+      "ĠColony": 33908,
+      "ĠâĮĬ": 33909,
+      "ĠIndonesian": 33910,
+      "ĠBauer": 33911,
+      "ĠKod": 33912,
+      "manned": 33913,
+      "Resistant": 33914,
+      "Ġdaughters": 33915,
+      "ĠPredicted": 33916,
+      "Ġvocab": 33917,
+      "Ġcontrasted": 33918,
+      "margin": 33919,
+      "ĠDirected": 33920,
+      "EDTA": 33921,
+      "Ġsynchrony": 33922,
+      "icki": 33923,
+      "ĠSalv": 33924,
+      "treat": 33925,
+      "incess": 33926,
+      "varnothing": 33927,
+      "Ġhexane": 33928,
+      "Empty": 33929,
+      "Ġgemcitabine": 33930,
+      "omib": 33931,
+      "orepinephrine": 33932,
+      "proc": 33933,
+      "ĠMetS": 33934,
+      "ĠDRAM": 33935,
+      "Ġanticoagulant": 33936,
+      "nom": 33937,
+      "amater": 33938,
+      "ĠLiDAR": 33939,
+      "Ġmobil": 33940,
+      "Ġameliorates": 33941,
+      "niz": 33942,
+      "Ġja": 33943,
+      "Ġemuls": 33944,
+      "ĠZa": 33945,
+      "Ġastronomical": 33946,
+      "ĠAlfred": 33947,
+      "Hilbert": 33948,
+      "ĠKF": 33949,
+      "CRT": 33950,
+      "quadratic": 33951,
+      "Ġdifferentials": 33952,
+      "robacterium": 33953,
+      "ĠHippocampal": 33954,
+      "pull": 33955,
+      "ÄĻ": 33956,
+      "Ġsad": 33957,
+      "allyl": 33958,
+      "Ġhotspot": 33959,
+      "ĠElectronics": 33960,
+      "Ġconstitution": 33961,
+      "itonin": 33962,
+      "اÙĦ": 33963,
+      "Pc": 33964,
+      "Ġrevascular": 33965,
+      "Ġusable": 33966,
+      "ĠScatter": 33967,
+      "Ġgraphically": 33968,
+      "liminf": 33969,
+      "Ġrestaurant": 33970,
+      "ucalyptus": 33971,
+      "ACG": 33972,
+      "Analy": 33973,
+      "ĠMillipore": 33974,
+      "Ġmunicipalities": 33975,
+      "ĠMacrophage": 33976,
+      "Ġmacromolecular": 33977,
+      "License": 33978,
+      "gc": 33979,
+      "Ġlavage": 33980,
+      "ĠAES": 33981,
+      "ĠFCS": 33982,
+      "peritone": 33983,
+      "Ġmeasles": 33984,
+      "TEX": 33985,
+      "ĠVirulence": 33986,
+      "Ġhematoma": 33987,
+      "ĠFres": 33988,
+      "ĠNutrient": 33989,
+      "apar": 33990,
+      "ĠSpot": 33991,
+      "coplasma": 33992,
+      "ĠExpect": 33993,
+      "Ġciprofloxacin": 33994,
+      "phylaxis": 33995,
+      "ĠAtlanta": 33996,
+      "routing": 33997,
+      "arate": 33998,
+      "ĠCis": 33999,
+      "ensure": 34000,
+      "carriers": 34001,
+      "ĠVariant": 34002,
+      "surgical": 34003,
+      "ĠEstimate": 34004,
+      "à¹": 34005,
+      "ĠLiqu": 34006,
+      "Ġamalg": 34007,
+      "Ġbla": 34008,
+      "Ġthematic": 34009,
+      "IRQ": 34010,
+      "ACTION": 34011,
+      "ĠChristi": 34012,
+      "æľ": 34013,
+      "Ġnpy": 34014,
+      "death": 34015,
+      "Ġhairpin": 34016,
+      "Ġmultiplicities": 34017,
+      "Gibco": 34018,
+      "heated": 34019,
+      "afety": 34020,
+      "mutable": 34021,
+      "quarks": 34022,
+      "Sun": 34023,
+      "ql": 34024,
+      "Ġproductions": 34025,
+      "Ġgeology": 34026,
+      "Ġtides": 34027,
+      "atrix": 34028,
+      "Ġadmixture": 34029,
+      "translated": 34030,
+      "ĠAbu": 34031,
+      "nucleus": 34032,
+      "Ġweaknesses": 34033,
+      "Ġflavors": 34034,
+      "ĠLuis": 34035,
+      "ĠPutative": 34036,
+      "sentence": 34037,
+      "ĠMast": 34038,
+      "ĠMPS": 34039,
+      "ĠESS": 34040,
+      "Ġcompose": 34041,
+      "Ġbirefring": 34042,
+      "ĠRamsey": 34043,
+      "ĠCLL": 34044,
+      "Ġlignocell": 34045,
+      "ĠLamin": 34046,
+      "ĠWelsh": 34047,
+      "von": 34048,
+      "Ġpests": 34049,
+      "Ġfiction": 34050,
+      "ĠHRT": 34051,
+      "Ġassure": 34052,
+      "CTs": 34053,
+      "ĠPAHs": 34054,
+      "Ġcryptography": 34055,
+      "enerated": 34056,
+      "Ġops": 34057,
+      "ĠSynerg": 34058,
+      "iginal": 34059,
+      "ĠCraw": 34060,
+      "Ġkne": 34061,
+      "Ġcurvatures": 34062,
+      "Ġlux": 34063,
+      "ĠKenn": 34064,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 34065,
+      "println": 34066,
+      "Ġvertebrae": 34067,
+      "Ġrutile": 34068,
+      "ĠAerosol": 34069,
+      "referred": 34070,
+      "lactamase": 34071,
+      "vehicle": 34072,
+      "adir": 34073,
+      "izards": 34074,
+      "Ġcallback": 34075,
+      "Cluster": 34076,
+      "Ġsilt": 34077,
+      "Ġresearched": 34078,
+      "ĠGenerator": 34079,
+      "ĠRestoration": 34080,
+      "ĠChin": 34081,
+      "ometrical": 34082,
+      "ĠCoefficients": 34083,
+      "rachid": 34084,
+      "Face": 34085,
+      "Men": 34086,
+      "counts": 34087,
+      "Ġpeg": 34088,
+      "Ġecl": 34089,
+      "Ġcomedy": 34090,
+      "ĠLn": 34091,
+      "obuty": 34092,
+      "ĠSharing": 34093,
+      "Ġadequacy": 34094,
+      "urtosis": 34095,
+      "ĠPicard": 34096,
+      "Ġfestival": 34097,
+      "Ġdisposition": 34098,
+      "ĠComplement": 34099,
+      "ĠExclusion": 34100,
+      "Ġdextran": 34101,
+      "mons": 34102,
+      "ĠInterpolation": 34103,
+      "ĠSteven": 34104,
+      "Ġcelebrated": 34105,
+      "ĠhPa": 34106,
+      "ofrequency": 34107,
+      "Ġexceptionally": 34108,
+      "Ġenergetically": 34109,
+      "psychotic": 34110,
+      "Landau": 34111,
+      "Tuple": 34112,
+      "distributions": 34113,
+      "ĠRichards": 34114,
+      "Ġpolyps": 34115,
+      "ĠAbsence": 34116,
+      "Ġceleb": 34117,
+      "XG": 34118,
+      "Ġsimulates": 34119,
+      "mitters": 34120,
+      "Ġheatmap": 34121,
+      "ĠSDN": 34122,
+      "ĠSteps": 34123,
+      "Ġshallower": 34124,
+      "ĠTurbulent": 34125,
+      "YT": 34126,
+      "Ġnal": 34127,
+      "plicative": 34128,
+      "phae": 34129,
+      "ĠLeica": 34130,
+      "ĠAPPRO": 34131,
+      "Ġarrhythmia": 34132,
+      "Ġrewriting": 34133,
+      "Ġunsafe": 34134,
+      "Ġcoworkers": 34135,
+      "ĠGAD": 34136,
+      "ivol": 34137,
+      "Ġdisrupting": 34138,
+      "ĠUltraviolet": 34139,
+      "eree": 34140,
+      "ĠLopez": 34141,
+      "Ġnegation": 34142,
+      "Ġjaponica": 34143,
+      "ecessor": 34144,
+      "ĠPatch": 34145,
+      "Ġsoap": 34146,
+      "ĠYing": 34147,
+      "MSK": 34148,
+      "Ġtracheal": 34149,
+      "icos": 34150,
+      "Ġvp": 34151,
+      "FAIL": 34152,
+      "Ġcatabolism": 34153,
+      "solver": 34154,
+      "font": 34155,
+      "esp": 34156,
+      "ĠZou": 34157,
+      "Ġdarker": 34158,
+      "Ġlysozyme": 34159,
+      "covered": 34160,
+      "Ġmultitude": 34161,
+      "requently": 34162,
+      "Ġmetamorph": 34163,
+      "Ġchapters": 34164,
+      "hh": 34165,
+      "chl": 34166,
+      "redundant": 34167,
+      "acking": 34168,
+      "Ġentail": 34169,
+      "ĠPacket": 34170,
+      "ĠHabitat": 34171,
+      "imedia": 34172,
+      "ĠCof": 34173,
+      "phrase": 34174,
+      "Ġcloth": 34175,
+      "arsal": 34176,
+      "Ġdrums": 34177,
+      "TPUT": 34178,
+      "Args": 34179,
+      "ductory": 34180,
+      "ĠUltimately": 34181,
+      "icates": 34182,
+      "antigen": 34183,
+      "Though": 34184,
+      "ĠFlore": 34185,
+      "probs": 34186,
+      "Ġcirculatory": 34187,
+      "ĠContemporary": 34188,
+      "eplitz": 34189,
+      "Ġhatch": 34190,
+      "rized": 34191,
+      "ĠKop": 34192,
+      "mitting": 34193,
+      "Ġhyperspectral": 34194,
+      "ĠAbst": 34195,
+      "SIM": 34196,
+      "Ġfruitful": 34197,
+      "Ġrecipe": 34198,
+      "Ġimidazole": 34199,
+      "Ġsynonymous": 34200,
+      "Ġattribution": 34201,
+      "ĠMartÃŃnez": 34202,
+      "ĠRodrÃŃguez": 34203,
+      "particular": 34204,
+      "ĠInteracting": 34205,
+      "Conf": 34206,
+      "ORE": 34207,
+      "ĠTMA": 34208,
+      "ucidation": 34209,
+      "Ġbiochemistry": 34210,
+      "ĠLevy": 34211,
+      "Ġconcentrates": 34212,
+      "Ġinductor": 34213,
+      "Ġpyrophosph": 34214,
+      "Ġrespondent": 34215,
+      "Zhang": 34216,
+      "Ġrope": 34217,
+      "Ġdesignation": 34218,
+      "ĠClim": 34219,
+      "Ġconstrains": 34220,
+      "shelf": 34221,
+      "ĠdÏĥ": 34222,
+      "ĠTLC": 34223,
+      "ĠAhar": 34224,
+      "ĠMatch": 34225,
+      "ĠMOL": 34226,
+      "Ġfees": 34227,
+      "wealth": 34228,
+      "Ġhyperactivity": 34229,
+      "ĠBruker": 34230,
+      "ĠFreund": 34231,
+      "dichlorophenyl": 34232,
+      "rero": 34233,
+      "ĠFear": 34234,
+      "dotsc": 34235,
+      "Ġhyg": 34236,
+      "ĠTexture": 34237,
+      "Tak": 34238,
+      "ampled": 34239,
+      "Ġalgeb": 34240,
+      "subt": 34241,
+      "Ġdocumentary": 34242,
+      "ĠJE": 34243,
+      "CNS": 34244,
+      "Ġdeclar": 34245,
+      "Height": 34246,
+      "Ki": 34247,
+      "enoid": 34248,
+      "ĠCervical": 34249,
+      "fractory": 34250,
+      "Ġplanted": 34251,
+      "IFI": 34252,
+      "Ġconceptually": 34253,
+      "Ġfillers": 34254,
+      "icola": 34255,
+      "lean": 34256,
+      "Ġclump": 34257,
+      "Ġwriters": 34258,
+      "Generally": 34259,
+      "Ġost": 34260,
+      "opening": 34261,
+      "CLASS": 34262,
+      "Ġherpesvirus": 34263,
+      "Instit": 34264,
+      "Ġdrinks": 34265,
+      "ĠIntensive": 34266,
+      "Ġmusician": 34267,
+      "Ġanchors": 34268,
+      "Series": 34269,
+      "ĠFAM": 34270,
+      "ĠBott": 34271,
+      "ĠECC": 34272,
+      "Ġinversions": 34273,
+      "Ġacres": 34274,
+      "Ġswabs": 34275,
+      "ĠÍī": 34276,
+      "ĠBerkeley": 34277,
+      "Ġplum": 34278,
+      "Ġempower": 34279,
+      "Ġphotoemission": 34280,
+      "ĠRabi": 34281,
+      "East": 34282,
+      "Taylor": 34283,
+      "OSE": 34284,
+      "Ġdenied": 34285,
+      "ĠHTTP": 34286,
+      "MU": 34287,
+      "hew": 34288,
+      "Ġthri": 34289,
+      "ĠCERN": 34290,
+      "Ġsuffice": 34291,
+      "functionalized": 34292,
+      "Ġcrabs": 34293,
+      "Ġidempotent": 34294,
+      "Ġpostulate": 34295,
+      "ĠCBF": 34296,
+      "discrim": 34297,
+      "Character": 34298,
+      "ĠRecombination": 34299,
+      "Cache": 34300,
+      "omit": 34301,
+      "ĠAda": 34302,
+      "Ġcursor": 34303,
+      "EMT": 34304,
+      "Ġmesoscale": 34305,
+      "guide": 34306,
+      "Hyper": 34307,
+      "Ġht": 34308,
+      "renes": 34309,
+      "ussen": 34310,
+      "whereas": 34311,
+      "Ġintegrator": 34312,
+      "Ġsyncy": 34313,
+      "arous": 34314,
+      "Ġcounteract": 34315,
+      "halose": 34316,
+      "ĠNotation": 34317,
+      "ĠRelevance": 34318,
+      "vf": 34319,
+      "Ġinbred": 34320,
+      "Ġrecirc": 34321,
+      "Ġende": 34322,
+      "Ġpresidential": 34323,
+      "Ġlactose": 34324,
+      "acional": 34325,
+      "ospi": 34326,
+      "ĠVGG": 34327,
+      "oselectivity": 34328,
+      "ĠConfig": 34329,
+      "Ġfingerprints": 34330,
+      "Interface": 34331,
+      "purple": 34332,
+      "etus": 34333,
+      "ĠNin": 34334,
+      "ĠKras": 34335,
+      "ĠReports": 34336,
+      "ĠSeattle": 34337,
+      "ADC": 34338,
+      "Ġlipoproteins": 34339,
+      "cyclohexyl": 34340,
+      "opressin": 34341,
+      "Ġwavefront": 34342,
+      "tetrazol": 34343,
+      "thys": 34344,
+      "Ġdivor": 34345,
+      "aminophen": 34346,
+      "ĠPerry": 34347,
+      "ĠConsiderations": 34348,
+      "ĠHalo": 34349,
+      "Ġreflexive": 34350,
+      "thiazolidin": 34351,
+      "oxycycline": 34352,
+      "CW": 34353,
+      "odim": 34354,
+      "ĠChong": 34355,
+      "Ġequilibrated": 34356,
+      "rime": 34357,
+      "ymology": 34358,
+      "Ġdevoid": 34359,
+      "rigel": 34360,
+      "amatergic": 34361,
+      "Ġidentifications": 34362,
+      "Ġcontrollability": 34363,
+      "ecticut": 34364,
+      "ĠSynchronization": 34365,
+      "ulatus": 34366,
+      "Ġcorrelating": 34367,
+      "Ġmuons": 34368,
+      "Ġcompartmental": 34369,
+      "Ġinhomogeneities": 34370,
+      "Ġevacuation": 34371,
+      "respiratory": 34372,
+      "dimethoxy": 34373,
+      "Ġinterferometric": 34374,
+      "Ġastronomy": 34375,
+      "ZD": 34376,
+      "ĦĦ": 34377,
+      "elia": 34378,
+      "bler": 34379,
+      "Ġpioneering": 34380,
+      "Ġpits": 34381,
+      "Ġmansoni": 34382,
+      "ĠCOND": 34383,
+      "Ġcodeword": 34384,
+      "imura": 34385,
+      "ĠDopamine": 34386,
+      "ĠGiov": 34387,
+      "ĠCameroon": 34388,
+      "Sem": 34389,
+      "dong": 34390,
+      "otto": 34391,
+      "emies": 34392,
+      "Ġinterquartile": 34393,
+      "llbracket": 34394,
+      "otropies": 34395,
+      "Ġhappening": 34396,
+      "ĠPalm": 34397,
+      "Ġstuff": 34398,
+      "Ġparking": 34399,
+      "egal": 34400,
+      "ĠCOP": 34401,
+      "Ġorganizing": 34402,
+      "Ġpolyhedral": 34403,
+      "Ġprovenance": 34404,
+      "Js": 34405,
+      "chains": 34406,
+      "egu": 34407,
+      "mercap": 34408,
+      "leveland": 34409,
+      "Ġerythroid": 34410,
+      "ymptomatic": 34411,
+      "Ġzigzag": 34412,
+      "Ġinferring": 34413,
+      "Ġapprox": 34414,
+      "Ġdownlink": 34415,
+      "ĠDeficiency": 34416,
+      "rbracket": 34417,
+      "ĠTIM": 34418,
+      "STS": 34419,
+      "ainen": 34420,
+      "Ġunloading": 34421,
+      "ĠXP": 34422,
+      "ĠWhilst": 34423,
+      "ĠIDH": 34424,
+      "ĠTIMP": 34425,
+      "rrbracket": 34426,
+      "acities": 34427,
+      "Ġwhale": 34428,
+      "ĠWAR": 34429,
+      "Ġinfl": 34430,
+      "ĠPresentation": 34431,
+      "authorbsnm": 34432,
+      "Ġbactericidal": 34433,
+      "SPEC": 34434,
+      "Ġdysregulated": 34435,
+      "ĠICAM": 34436,
+      "nano": 34437,
+      "Ġwafers": 34438,
+      "ĠMUC": 34439,
+      "Ġalien": 34440,
+      "chke": 34441,
+      "Ġslabs": 34442,
+      "Ġbacking": 34443,
+      "nsis": 34444,
+      "Ġbalances": 34445,
+      "ethane": 34446,
+      "Linked": 34447,
+      "Chen": 34448,
+      "Hymenoptera": 34449,
+      "itations": 34450,
+      "ĠOUT": 34451,
+      "transplant": 34452,
+      "conditioned": 34453,
+      "ĠBenefits": 34454,
+      "Tyr": 34455,
+      "atmosp": 34456,
+      "ĠAdhesion": 34457,
+      "Ġthorac": 34458,
+      "activator": 34459,
+      "Ġphosphatidylinositol": 34460,
+      "Ġreportedly": 34461,
+      "ĠCLASS": 34462,
+      "Ġrenewed": 34463,
+      "ĠPharmacological": 34464,
+      "Ġminimise": 34465,
+      "glucosidase": 34466,
+      "adenosyl": 34467,
+      "Ġovip": 34468,
+      "initializer": 34469,
+      "Ġforage": 34470,
+      "rms": 34471,
+      "ĠImag": 34472,
+      "ĠAnnexin": 34473,
+      "ĠVehicles": 34474,
+      "Ġfles": 34475,
+      "sta": 34476,
+      "ĠGBS": 34477,
+      "ĠChat": 34478,
+      "measurements": 34479,
+      "ĠAuditory": 34480,
+      "Cut": 34481,
+      "Fv": 34482,
+      "Ġmaker": 34483,
+      "application": 34484,
+      "Ġreversing": 34485,
+      "Ġstip": 34486,
+      "Ġfaecalis": 34487,
+      "icycle": 34488,
+      "Ġtrimmed": 34489,
+      "Ġexacerbation": 34490,
+      "Ġtranscranial": 34491,
+      "ĠMomentum": 34492,
+      "Ġfc": 34493,
+      "ĠFOV": 34494,
+      "Ġangina": 34495,
+      "Ġnanostructure": 34496,
+      "Ġantagonism": 34497,
+      "ĠLEDs": 34498,
+      "ìĹIJ": 34499,
+      "Ġfals": 34500,
+      "aporation": 34501,
+      "ĠInvasive": 34502,
+      "ĠKm": 34503,
+      "ertation": 34504,
+      "Ġharness": 34505,
+      "Ġfertile": 34506,
+      "ĠTRUE": 34507,
+      "Ġshelter": 34508,
+      "ĠWolbachia": 34509,
+      "shoot": 34510,
+      "Ġsess": 34511,
+      "ĠHous": 34512,
+      "ĠAce": 34513,
+      "ĠCML": 34514,
+      "Ġproactive": 34515,
+      "Ġshots": 34516,
+      "Ġcoup": 34517,
+      "restling": 34518,
+      "uniformly": 34519,
+      "yam": 34520,
+      "olase": 34521,
+      "ĠICS": 34522,
+      "ĠEbola": 34523,
+      "rolling": 34524,
+      "trunc": 34525,
+      "ĠRepresentatives": 34526,
+      "Ġgrasping": 34527,
+      "ĠAnomaly": 34528,
+      "ĠMine": 34529,
+      "ĠMPO": 34530,
+      "leright": 34531,
+      "Ġinstitute": 34532,
+      "Ġsugarcane": 34533,
+      "ÑĢа": 34534,
+      "Ġoccluded": 34535,
+      "ĠMagellanic": 34536,
+      "BEC": 34537,
+      "Wi": 34538,
+      "oA": 34539,
+      "Ġgapped": 34540,
+      "ĠPRC": 34541,
+      "ĠMAE": 34542,
+      "Ġmusicians": 34543,
+      "ĠSignificantly": 34544,
+      "Ġforthcoming": 34545,
+      "Ġacclimation": 34546,
+      "required": 34547,
+      "verbal": 34548,
+      "ĠFX": 34549,
+      "ĠMLE": 34550,
+      "Ġcompass": 34551,
+      "ĠMultimodal": 34552,
+      "Grant": 34553,
+      "Ġtvb": 34554,
+      "Instruction": 34555,
+      "Ġsenses": 34556,
+      "urbed": 34557,
+      "hamn": 34558,
+      "Ġframed": 34559,
+      "Ġurothel": 34560,
+      "orin": 34561,
+      "seal": 34562,
+      "Ġflasks": 34563,
+      "shops": 34564,
+      "Ġwheels": 34565,
+      "ĠRadon": 34566,
+      "ĠPlanetary": 34567,
+      "Ġhedge": 34568,
+      "Ġdk": 34569,
+      "Ġevidently": 34570,
+      "threads": 34571,
+      "Ġtad": 34572,
+      "elim": 34573,
+      "imov": 34574,
+      "istem": 34575,
+      "andi": 34576,
+      "Ġleisure": 34577,
+      "ostom": 34578,
+      "Ġcaring": 34579,
+      "ĠSmoking": 34580,
+      "Ġcompetitors": 34581,
+      "AFS": 34582,
+      "xl": 34583,
+      "ĠSatur": 34584,
+      "ĠFerg": 34585,
+      "Ġchin": 34586,
+      "ĠCDR": 34587,
+      "ĠSOM": 34588,
+      "osaccharide": 34589,
+      "MODEL": 34590,
+      "ECC": 34591,
+      "Ġdas": 34592,
+      "agonist": 34593,
+      "stery": 34594,
+      "Ġrelays": 34595,
+      "zek": 34596,
+      "Ġneoplasm": 34597,
+      "Chip": 34598,
+      "Ġgill": 34599,
+      "lamed": 34600,
+      "cerning": 34601,
+      "Ġinconsistencies": 34602,
+      "aceans": 34603,
+      "ĠAdri": 34604,
+      "ĠAfghan": 34605,
+      "Ġniches": 34606,
+      "Ġtunnelling": 34607,
+      "gus": 34608,
+      "ĠIan": 34609,
+      "Ġburial": 34610,
+      "Transform": 34611,
+      "ocompatible": 34612,
+      "Ġseldom": 34613,
+      "Ġdisclosed": 34614,
+      "âĪķ": 34615,
+      "Ġrefining": 34616,
+      "Ġtyph": 34617,
+      "Ġcooperate": 34618,
+      "Ġasphalt": 34619,
+      "ĠConstitution": 34620,
+      "flavor": 34621,
+      "Ġwarp": 34622,
+      "ż": 34623,
+      "Ġcraw": 34624,
+      "ĠIndigenous": 34625,
+      "ĠPrevent": 34626,
+      "Ġtrigeminal": 34627,
+      "ĠFriedrich": 34628,
+      "ĠInterferon": 34629,
+      "iosity": 34630,
+      "warm": 34631,
+      "uson": 34632,
+      "Ġunderlies": 34633,
+      "Ġmultiplets": 34634,
+      "ĠSUPER": 34635,
+      "ĠManufacturing": 34636,
+      "Ġvimentin": 34637,
+      "ramine": 34638,
+      "Ġefficacious": 34639,
+      "iced": 34640,
+      "ĠVall": 34641,
+      "othorax": 34642,
+      "Ġaudi": 34643,
+      "Qs": 34644,
+      "ĠPAL": 34645,
+      "ĠHold": 34646,
+      "hattan": 34647,
+      "idding": 34648,
+      "wana": 34649,
+      "Ġpending": 34650,
+      "Ġperennial": 34651,
+      "Ġtouching": 34652,
+      "xpected": 34653,
+      "Distance": 34654,
+      "nav": 34655,
+      "Ġisomeric": 34656,
+      "ĠMCI": 34657,
+      "numbers": 34658,
+      "Ġreverses": 34659,
+      "Ġpolycystic": 34660,
+      "Hem": 34661,
+      "uities": 34662,
+      "optional": 34663,
+      "Ġsubcortical": 34664,
+      "ĠSupply": 34665,
+      "ĠCalder": 34666,
+      "Ġmangrove": 34667,
+      "Ġpads": 34668,
+      "urfaces": 34669,
+      "ĠFaster": 34670,
+      "Ġunderneath": 34671,
+      "Ġprolactin": 34672,
+      "Ġclearer": 34673,
+      "Ġscintillation": 34674,
+      "Ġhumidified": 34675,
+      "ĠWound": 34676,
+      "ĠHPA": 34677,
+      "Ġcollapsing": 34678,
+      "Ġbaryonic": 34679,
+      "ĠMEASU": 34680,
+      "ĠGü": 34681,
+      "Ġdetr": 34682,
+      "Ġsubstituent": 34683,
+      "ĠRomania": 34684,
+      "ĠInvolved": 34685,
+      "Ġduodenal": 34686,
+      "ĠAmp": 34687,
+      "ĠSIS": 34688,
+      "scher": 34689,
+      "auth": 34690,
+      "ĠRespond": 34691,
+      "ĠRanking": 34692,
+      "trip": 34693,
+      "xF": 34694,
+      "istin": 34695,
+      "Ġpauc": 34696,
+      "reflection": 34697,
+      "Ġcornea": 34698,
+      "Ġbolus": 34699,
+      "Ġpivot": 34700,
+      "October": 34701,
+      "ĠSERS": 34702,
+      "ĠXing": 34703,
+      "ANET": 34704,
+      "Chinese": 34705,
+      "ĠMusc": 34706,
+      "Dynamic": 34707,
+      "Mesh": 34708,
+      "Ġdiphosphate": 34709,
+      "Ġconspecific": 34710,
+      "lector": 34711,
+      "ĠEcu": 34712,
+      "ĠCoverage": 34713,
+      "ĠãĢĪ": 34714,
+      "COD": 34715,
+      "among": 34716,
+      "Ġposit": 34717,
+      "imumab": 34718,
+      "ĠpN": 34719,
+      "Ġcoaching": 34720,
+      "exports": 34721,
+      "Ġrealm": 34722,
+      "ĠFerreira": 34723,
+      "Ġnationally": 34724,
+      "Ġturtle": 34725,
+      "ubtedly": 34726,
+      "ĠDraft": 34727,
+      "Ġendl": 34728,
+      "ĠContinuum": 34729,
+      "embeddings": 34730,
+      "Ġá¹½": 34731,
+      "ĠCrime": 34732,
+      "Ġimmigration": 34733,
+      "ĠFilip": 34734,
+      "Ġgarnet": 34735,
+      "Ġobscure": 34736,
+      "ĠTYPE": 34737,
+      "Ġultrastructural": 34738,
+      "caemia": 34739,
+      "ĠSeman": 34740,
+      "rink": 34741,
+      "tiff": 34742,
+      "uccal": 34743,
+      "kee": 34744,
+      "itudinally": 34745,
+      "ĠAlloy": 34746,
+      "ĠAnalyzer": 34747,
+      "continue": 34748,
+      "ĠAlabama": 34749,
+      "QOL": 34750,
+      "Ġpollin": 34751,
+      "Ġcorrespondences": 34752,
+      "ĠResol": 34753,
+      "FIR": 34754,
+      "ulare": 34755,
+      "tawa": 34756,
+      "URCE": 34757,
+      "Ġurbanization": 34758,
+      "zd": 34759,
+      "Ġgloss": 34760,
+      "ERA": 34761,
+      "ĠDetermine": 34762,
+      "Date": 34763,
+      "ĠPSP": 34764,
+      "ĠShig": 34765,
+      "repta": 34766,
+      "ĠGait": 34767,
+      "neutrino": 34768,
+      "Ġpervasive": 34769,
+      "ĠâĢ¢âĢ¢âĢ¢": 34770,
+      "Ġhomozyg": 34771,
+      "Ġadaptively": 34772,
+      "graphic": 34773,
+      "ĠJohnston": 34774,
+      "zt": 34775,
+      "explicit": 34776,
+      "Ġhelmin": 34777,
+      "Ġpes": 34778,
+      "ARF": 34779,
+      "ĠFram": 34780,
+      "ĠAmsterdam": 34781,
+      "Ġlogarithms": 34782,
+      "ĠCreative": 34783,
+      "PageIndex": 34784,
+      "Ġpacing": 34785,
+      "ĠPCS": 34786,
+      "Ġforebrain": 34787,
+      "ĠCTCF": 34788,
+      "decomposition": 34789,
+      "Ġbearings": 34790,
+      "Ġanhydrous": 34791,
+      "Ġcb": 34792,
+      "ĠMON": 34793,
+      "ĠNodes": 34794,
+      "strum": 34795,
+      "ĠJans": 34796,
+      "Ġdelineate": 34797,
+      "Ġdichroism": 34798,
+      "conformal": 34799,
+      "Ġretreat": 34800,
+      "glial": 34801,
+      "Ġnuclease": 34802,
+      "ĠBaltimore": 34803,
+      "Ġpaying": 34804,
+      "Ġboreal": 34805,
+      "tipation": 34806,
+      "Root": 34807,
+      "SQL": 34808,
+      "sources": 34809,
+      "endo": 34810,
+      "ĠOrion": 34811,
+      "Plus": 34812,
+      "ĠDEL": 34813,
+      "ĠThan": 34814,
+      "Ġmonoph": 34815,
+      "Ġreflector": 34816,
+      "Ze": 34817,
+      "ĠLinking": 34818,
+      "sync": 34819,
+      "ĠCREB": 34820,
+      "national": 34821,
+      "urized": 34822,
+      "ĠPeptides": 34823,
+      "ĠBegin": 34824,
+      "borg": 34825,
+      "piperidyl": 34826,
+      "Ġoverestimation": 34827,
+      "RGB": 34828,
+      "TK": 34829,
+      "Ġbeings": 34830,
+      "Ġattains": 34831,
+      "Ġreservation": 34832,
+      "ĠMotivation": 34833,
+      "Ġtrimethyl": 34834,
+      "ĠTerminal": 34835,
+      "Ġintentional": 34836,
+      "Negative": 34837,
+      "ĠCronbach": 34838,
+      "dorferi": 34839,
+      "Daw": 34840,
+      "VAR": 34841,
+      "dP": 34842,
+      "imath": 34843,
+      "overex": 34844,
+      "Ġfibrotic": 34845,
+      "Ġsmartphones": 34846,
+      "Ġontologies": 34847,
+      "Good": 34848,
+      "utively": 34849,
+      "ĠVB": 34850,
+      "SPE": 34851,
+      "ĠMcDonald": 34852,
+      "galaxies": 34853,
+      "Ġbiochar": 34854,
+      "ĠEMS": 34855,
+      "ĠNf": 34856,
+      "orship": 34857,
+      "Ġbackscattering": 34858,
+      "Ġп": 34859,
+      "Ġanthocyanin": 34860,
+      "ĠPhoenix": 34861,
+      "contained": 34862,
+      "ĠPSII": 34863,
+      "hlung": 34864,
+      "ĠLAI": 34865,
+      "Ġlectures": 34866,
+      "Ġdispatch": 34867,
+      "VF": 34868,
+      "ĠMEC": 34869,
+      "ĠWes": 34870,
+      "Ġbackscatter": 34871,
+      "otite": 34872,
+      "ĠSRC": 34873,
+      "Ġcurrency": 34874,
+      "onyms": 34875,
+      "aspartate": 34876,
+      "Ġcoset": 34877,
+      "ĠCPP": 34878,
+      "orbing": 34879,
+      "ĠEmbeddings": 34880,
+      "ĠSurveys": 34881,
+      "Ġneurodevelopmental": 34882,
+      "ĠSRE": 34883,
+      "ĠInterior": 34884,
+      "ĠARDS": 34885,
+      "experiments": 34886,
+      "bromophenyl": 34887,
+      "ĠECL": 34888,
+      "ĠOPE": 34889,
+      "mediation": 34890,
+      "Ġthermoc": 34891,
+      "Ġinterpretable": 34892,
+      "ĠMicrobiome": 34893,
+      "eastern": 34894,
+      "¿": 34895,
+      "ĠTDP": 34896,
+      "athon": 34897,
+      "ĠByzantine": 34898,
+      "anyon": 34899,
+      "Ġepitaxy": 34900,
+      "Ġcriticized": 34901,
+      "Millipore": 34902,
+      "ĠDEP": 34903,
+      "ĠFreedom": 34904,
+      "junctions": 34905,
+      "ĠASM": 34906,
+      "ĠGren": 34907,
+      "Ġsigning": 34908,
+      "Ġconstituting": 34909,
+      "oproterozoic": 34910,
+      "ĠSynech": 34911,
+      "ĠVoice": 34912,
+      "Ġcholecyst": 34913,
+      "bilities": 34914,
+      "online": 34915,
+      "ĠEdd": 34916,
+      "ĠKup": 34917,
+      "ĠLett": 34918,
+      "ĠMarin": 34919,
+      "ĠGoal": 34920,
+      "ĠSYM": 34921,
+      "introduced": 34922,
+      "naphthyl": 34923,
+      "ĠLü": 34924,
+      "Ġmx": 34925,
+      "Ġblu": 34926,
+      "Ġrm": 34927,
+      "ĠDeletion": 34928,
+      "ĠConnecticut": 34929,
+      "Coleoptera": 34930,
+      "try": 34931,
+      "Ġsoot": 34932,
+      "ĠCountries": 34933,
+      "Ġsickle": 34934,
+      "Meta": 34935,
+      "ĠSib": 34936,
+      "ĠHNO": 34937,
+      "ĠUD": 34938,
+      "Ġexpr": 34939,
+      "Ġallowable": 34940,
+      "ĠIndirect": 34941,
+      "tisation": 34942,
+      "Ġadenomas": 34943,
+      "electronics": 34944,
+      "RNN": 34945,
+      "ĠTCF": 34946,
+      "Ġglucagon": 34947,
+      "ĠCitation": 34948,
+      "Ġgamb": 34949,
+      "andez": 34950,
+      "Ġtransmits": 34951,
+      "ajima": 34952,
+      "Ġholonomy": 34953,
+      "ìł": 34954,
+      "actam": 34955,
+      "ĠThreat": 34956,
+      "ĠPearl": 34957,
+      "Ġeruptions": 34958,
+      "ĠImmunohistochemistry": 34959,
+      "Yes": 34960,
+      "patrick": 34961,
+      "Ġama": 34962,
+      "Ġdrew": 34963,
+      "ĠTasks": 34964,
+      "ĠPIM": 34965,
+      "Ġdispat": 34966,
+      "ĠDetroit": 34967,
+      "Ġcoexist": 34968,
+      "arboxylase": 34969,
+      "IBM": 34970,
+      "ĠTUNEL": 34971,
+      "ĠUF": 34972,
+      "ĠANG": 34973,
+      "Ġsarcopenia": 34974,
+      "Ġhaptic": 34975,
+      "Ġcarbonates": 34976,
+      "Ġmitophagy": 34977,
+      "Ġcitizen": 34978,
+      "ĠCONTROL": 34979,
+      "fif": 34980,
+      "Ġwi": 34981,
+      "ĠGLO": 34982,
+      "ensored": 34983,
+      "ĠPara": 34984,
+      "ĠAbdel": 34985,
+      "oietin": 34986,
+      "Ġtoe": 34987,
+      "ĠSQU": 34988,
+      "ĠRag": 34989,
+      "Ġxylem": 34990,
+      "Ġliberal": 34991,
+      "ĠMargaret": 34992,
+      "Wa": 34993,
+      "kp": 34994,
+      "ĠPEM": 34995,
+      "ĠDDR": 34996,
+      "Ġgenotypic": 34997,
+      "ĠYM": 34998,
+      "INGS": 34999,
+      "keras": 35000,
+      "ĠEducational": 35001,
+      "ĠCultures": 35002,
+      "instr": 35003,
+      "ĠFuchs": 35004,
+      "agasc": 35005,
+      "factant": 35006,
+      "Ġtenth": 35007,
+      "ABL": 35008,
+      "Ġpermeable": 35009,
+      "ĠCameron": 35010,
+      "BrN": 35011,
+      "ĠMuller": 35012,
+      "ĠReversible": 35013,
+      "wild": 35014,
+      "Ġfusions": 35015,
+      "osulf": 35016,
+      "ĠEoS": 35017,
+      "ĠKö": 35018,
+      "detected": 35019,
+      "ĠCollagen": 35020,
+      "Ġdescendants": 35021,
+      "election": 35022,
+      "arange": 35023,
+      "Ġbounce": 35024,
+      "Ġcontag": 35025,
+      "Invalid": 35026,
+      "ĠCoating": 35027,
+      "tasks": 35028,
+      "arma": 35029,
+      "ĠKC": 35030,
+      "Ġdiar": 35031,
+      "ĠSuppress": 35032,
+      "Ġfractionated": 35033,
+      "Ġsnail": 35034,
+      "Ġmicrophone": 35035,
+      "ĠScienti": 35036,
+      "Ġchemiluminescence": 35037,
+      "software": 35038,
+      "Ġburgdorferi": 35039,
+      "Ġboot": 35040,
+      "ĠCSCs": 35041,
+      "ĠMSI": 35042,
+      "tsev": 35043,
+      "Ġheater": 35044,
+      "fractal": 35045,
+      "Ġendosomes": 35046,
+      "Uniform": 35047,
+      "Ġathlete": 35048,
+      "ĠDriven": 35049,
+      "Ġvivax": 35050,
+      "Kind": 35051,
+      "satisfies": 35052,
+      "Ġcorticosteroid": 35053,
+      "ĠEstablishment": 35054,
+      "calibration": 35055,
+      "Ġdimeric": 35056,
+      "Ġcereal": 35057,
+      "ĠSupervised": 35058,
+      "ĠSPM": 35059,
+      "MBER": 35060,
+      "Ġhemispheres": 35061,
+      "Ġpercentiles": 35062,
+      "Leu": 35063,
+      "Major": 35064,
+      "Ġexagger": 35065,
+      "ĠdsRNA": 35066,
+      "December": 35067,
+      "ĠZrO": 35068,
+      "Ġasymmetrical": 35069,
+      "ĠVAS": 35070,
+      "ĠJM": 35071,
+      "Ġintegrations": 35072,
+      "Ġhandover": 35073,
+      "Cycl": 35074,
+      "implant": 35075,
+      "Ġquote": 35076,
+      "Ġcyclone": 35077,
+      "ĠStephan": 35078,
+      "ĠFranco": 35079,
+      "Ġawake": 35080,
+      "Ġfeeder": 35081,
+      "CHAR": 35082,
+      "Condition": 35083,
+      "ĠCharl": 35084,
+      "ĠBrigade": 35085,
+      "Ġremediation": 35086,
+      "cig": 35087,
+      "ĠBohr": 35088,
+      "ĠVacuum": 35089,
+      "ĠToxoplasma": 35090,
+      "Ġghrelin": 35091,
+      "ĠTRAF": 35092,
+      "aye": 35093,
+      "Client": 35094,
+      "iliation": 35095,
+      "xyz": 35096,
+      "mingham": 35097,
+      "ĠSUB": 35098,
+      "ïĢł": 35099,
+      "Ġconversions": 35100,
+      "Ġmultipath": 35101,
+      "missive": 35102,
+      "Ġeqn": 35103,
+      "bulk": 35104,
+      "myc": 35105,
+      "Ġexacerbated": 35106,
+      "ت": 35107,
+      "Ġproteinase": 35108,
+      "Ġbuilder": 35109,
+      "ahara": 35110,
+      "Ġinvert": 35111,
+      "ĠReception": 35112,
+      "axanthin": 35113,
+      "Ġprimed": 35114,
+      "Ġcopula": 35115,
+      "Ġproceedings": 35116,
+      "Ġnondegenerate": 35117,
+      "Ġintox": 35118,
+      "Ġneedles": 35119,
+      "lengths": 35120,
+      "Ġtransposon": 35121,
+      "hon": 35122,
+      "ĠTPC": 35123,
+      "pland": 35124,
+      "oxyn": 35125,
+      "ICH": 35126,
+      "Ġintrauterine": 35127,
+      "Ġlaminated": 35128,
+      "ĠOBSERV": 35129,
+      "Match": 35130,
+      "ĠInsur": 35131,
+      "ĠAmyloid": 35132,
+      "Ġwarped": 35133,
+      "ematical": 35134,
+      "ĠPractices": 35135,
+      "ãģ®": 35136,
+      "ĠBrassica": 35137,
+      "Ġhyperthermia": 35138,
+      "Ġdn": 35139,
+      "ĠLIF": 35140,
+      "ĠMetropolitan": 35141,
+      "ĠBrdU": 35142,
+      "impact": 35143,
+      "filtered": 35144,
+      "ĠReagent": 35145,
+      "vp": 35146,
+      "ĠTip": 35147,
+      "ĠProportional": 35148,
+      "Ġbloodstream": 35149,
+      "Simple": 35150,
+      "Ġtyros": 35151,
+      "ĠHenri": 35152,
+      "Ġretrotrans": 35153,
+      "aciens": 35154,
+      "Ġmistakes": 35155,
+      "acylglycerol": 35156,
+      "ĠMirror": 35157,
+      "VERSION": 35158,
+      "vre": 35159,
+      "Ġbact": 35160,
+      "ocked": 35161,
+      "epsis": 35162,
+      "Ġsonication": 35163,
+      "ĠPurkinje": 35164,
+      "Ġmismatches": 35165,
+      "ĠAOD": 35166,
+      "Ġhypergraph": 35167,
+      "ĠMiami": 35168,
+      "ammed": 35169,
+      "Ġconversely": 35170,
+      "ĠGabor": 35171,
+      "ĠGDM": 35172,
+      "Ġcoiled": 35173,
+      "onica": 35174,
+      "Ġevolutions": 35175,
+      "ĠRBM": 35176,
+      "ĠReef": 35177,
+      "ĠAbram": 35178,
+      "ĠPrecise": 35179,
+      "increase": 35180,
+      "ĠPlatelet": 35181,
+      "Generator": 35182,
+      "Arch": 35183,
+      "ĠBened": 35184,
+      "preceq": 35185,
+      "measurable": 35186,
+      "CAS": 35187,
+      "ĠTourn": 35188,
+      "Ġgiants": 35189,
+      "Ġeddies": 35190,
+      "Ġcolumnar": 35191,
+      "aggregation": 35192,
+      "Ġzirconia": 35193,
+      "ducibility": 35194,
+      "Ġservo": 35195,
+      "Ġbeauty": 35196,
+      "Ġheap": 35197,
+      "ĠâĪĴâĪĴâĪĴ": 35198,
+      "Ġconductivities": 35199,
+      "Ġdarkness": 35200,
+      "Ġoccupying": 35201,
+      "ĠClean": 35202,
+      "bash": 35203,
+      "ulans": 35204,
+      "appy": 35205,
+      "ĠMarker": 35206,
+      "runtime": 35207,
+      "Ġhaemoglobin": 35208,
+      "Ġdesktop": 35209,
+      "mis": 35210,
+      "ĠSof": 35211,
+      "osse": 35212,
+      "Ġcomoving": 35213,
+      "Ġclutter": 35214,
+      "ourced": 35215,
+      "Ġsubj": 35216,
+      "arching": 35217,
+      "ĠSolomon": 35218,
+      "locking": 35219,
+      "Ġparap": 35220,
+      "Ġrotator": 35221,
+      "ĠACKNOWLEDGMENTS": 35222,
+      "Ter": 35223,
+      "yster": 35224,
+      "ĠWebb": 35225,
+      "Ġsubsample": 35226,
+      "osilicate": 35227,
+      "Training": 35228,
+      "orpha": 35229,
+      "Ġtimeout": 35230,
+      "otinamide": 35231,
+      "ĠFabry": 35232,
+      "ĠReceiver": 35233,
+      "Ġconjunctiv": 35234,
+      "ĠEcuador": 35235,
+      "ĠIda": 35236,
+      "Ġcasein": 35237,
+      "Ġïģ¸": 35238,
+      "ĠBarn": 35239,
+      "ĠSchools": 35240,
+      "elona": 35241,
+      "dip": 35242,
+      "ĠChrys": 35243,
+      "ICI": 35244,
+      "Ġposteriori": 35245,
+      "Ġbleaching": 35246,
+      "ĠPersonality": 35247,
+      "umbers": 35248,
+      "ĠModes": 35249,
+      "Ġnotification": 35250,
+      "Ġsupine": 35251,
+      "alued": 35252,
+      "keep": 35253,
+      "ĠFranz": 35254,
+      "Ġwounded": 35255,
+      "YL": 35256,
+      "Ġdilemma": 35257,
+      "ĠClara": 35258,
+      "ĠCarroll": 35259,
+      "Ġsickness": 35260,
+      "Ġproxies": 35261,
+      "ecks": 35262,
+      "ĠÏ«": 35263,
+      "Ġplanting": 35264,
+      "Ġciphertext": 35265,
+      "ĠFamilies": 35266,
+      "iesel": 35267,
+      "Ġincongru": 35268,
+      "ĠExcitation": 35269,
+      "Ġconferred": 35270,
+      "ĠButter": 35271,
+      "Impl": 35272,
+      "collision": 35273,
+      "idol": 35274,
+      "Ġacquires": 35275,
+      "ĠOwen": 35276,
+      "SAM": 35277,
+      "ĠGUT": 35278,
+      "lects": 35279,
+      "Ġdeleg": 35280,
+      "Shot": 35281,
+      "Ġanthrac": 35282,
+      "Russian": 35283,
+      "ĠPCE": 35284,
+      "ĠâĥĹ": 35285,
+      "ĠKab": 35286,
+      "NAC": 35287,
+      "Ġargparse": 35288,
+      "ĠViol": 35289,
+      "Ġanticoagulation": 35290,
+      "Ġcredibility": 35291,
+      "Ġrotavirus": 35292,
+      "ĠInvest": 35293,
+      "Ġrecol": 35294,
+      "variety": 35295,
+      "Ġdeformable": 35296,
+      "Ġenergetics": 35297,
+      "Ġconsultations": 35298,
+      "letics": 35299,
+      "ĠFoss": 35300,
+      "ĠLIGO": 35301,
+      "php": 35302,
+      "ĠChal": 35303,
+      "ĠMalawi": 35304,
+      "Ġstrokes": 35305,
+      "horm": 35306,
+      "Ġbs": 35307,
+      "Ġplural": 35308,
+      "strategy": 35309,
+      "Ġmisalignment": 35310,
+      "previous": 35311,
+      "filters": 35312,
+      "ĠDemographics": 35313,
+      "deterministic": 35314,
+      "Ġcyclophosphamide": 35315,
+      "Ġstreak": 35316,
+      "ĠBiosynthesis": 35317,
+      "Ġsubcutaneously": 35318,
+      "jn": 35319,
+      "Ġampicillin": 35320,
+      "ĠChag": 35321,
+      "iformes": 35322,
+      "IFICATION": 35323,
+      "Ġyourself": 35324,
+      "Ġtolerability": 35325,
+      "Ġautocl": 35326,
+      "rhs": 35327,
+      "Ġpupils": 35328,
+      "Ġgauged": 35329,
+      "Lay": 35330,
+      "ĠSanti": 35331,
+      "ĠDBP": 35332,
+      "ĠGary": 35333,
+      "drive": 35334,
+      "Ġtrustworth": 35335,
+      "Ġcontingency": 35336,
+      "Cube": 35337,
+      "Host": 35338,
+      "fu": 35339,
+      "Ġhsa": 35340,
+      "issner": 35341,
+      "ITT": 35342,
+      "ĠSrTiO": 35343,
+      "Ġcounselling": 35344,
+      "integrable": 35345,
+      "Ġunderway": 35346,
+      "Ġstandardised": 35347,
+      "bius": 35348,
+      "Firstly": 35349,
+      "Ġporphyrin": 35350,
+      "Area": 35351,
+      "iw": 35352,
+      "Ġub": 35353,
+      "ĠLynch": 35354,
+      "ĠWBC": 35355,
+      "ilden": 35356,
+      "Ġhomeless": 35357,
+      "Ġmagnetosphere": 35358,
+      "Ġnighttime": 35359,
+      "ncbi": 35360,
+      "Ġdownt": 35361,
+      "lethal": 35362,
+      "Ġinterim": 35363,
+      "ĠResil": 35364,
+      "Ġcontinually": 35365,
+      "ĠImmunofluorescence": 35366,
+      "Design": 35367,
+      "Ġadvocate": 35368,
+      "reptavidin": 35369,
+      "fw": 35370,
+      "story": 35371,
+      "ĠPSS": 35372,
+      "Ġfiled": 35373,
+      "Ġbedrock": 35374,
+      "Ġisoflurane": 35375,
+      "Ġreluct": 35376,
+      "eward": 35377,
+      "ĠIndependence": 35378,
+      "ĠBurkholder": 35379,
+      "Ġcinn": 35380,
+      "Ġcaptive": 35381,
+      "Ġcomposing": 35382,
+      "Ġrestraint": 35383,
+      "Ġquestionable": 35384,
+      "ĠTomato": 35385,
+      "Ġzeroth": 35386,
+      "rins": 35387,
+      "omez": 35388,
+      "Ġglia": 35389,
+      "ĠGlac": 35390,
+      "Independent": 35391,
+      "Ġobjectively": 35392,
+      "pA": 35393,
+      "Ġfavoring": 35394,
+      "ipelago": 35395,
+      "Ġincontinence": 35396,
+      "bium": 35397,
+      "ĠLZ": 35398,
+      "ĠLed": 35399,
+      "hexyl": 35400,
+      "Ġceased": 35401,
+      "Ġoleic": 35402,
+      "ĠImpairment": 35403,
+      "Ñĸ": 35404,
+      "ongo": 35405,
+      "Ġrunner": 35406,
+      "Ġcucumber": 35407,
+      "ĠPerform": 35408,
+      "Ġdoublets": 35409,
+      "Ġeigenfunction": 35410,
+      "Ġ̺": 35411,
+      "ĠHenderson": 35412,
+      "Klein": 35413,
+      "Tab": 35414,
+      "Ġbeer": 35415,
+      "ocom": 35416,
+      "unciation": 35417,
+      "------": 35418,
+      "ĠTSC": 35419,
+      "ogas": 35420,
+      "Ġrud": 35421,
+      "Ġincis": 35422,
+      "ĠLOG": 35423,
+      "FBQ": 35424,
+      "Ġinterconnection": 35425,
+      "î": 35426,
+      "arbox": 35427,
+      "ĠIBS": 35428,
+      "ĠNCT": 35429,
+      "ĠGand": 35430,
+      "Ġyaw": 35431,
+      "ĠTransverse": 35432,
+      "ĠSudan": 35433,
+      "Ġconstriction": 35434,
+      "Hor": 35435,
+      "Ġevasion": 35436,
+      "Ġmeromorphic": 35437,
+      "ĠPBMC": 35438,
+      "IUM": 35439,
+      "reed": 35440,
+      "ĠBö": 35441,
+      "ĠEMB": 35442,
+      "ukh": 35443,
+      "Ġwinners": 35444,
+      "Ġascites": 35445,
+      "Mes": 35446,
+      "Ġeclipse": 35447,
+      "ĠEocene": 35448,
+      "adiazol": 35449,
+      "Ġrecoveries": 35450,
+      "Starting": 35451,
+      "rema": 35452,
+      "ĠÃİ": 35453,
+      "monotonic": 35454,
+      "ĠMeOH": 35455,
+      "ĠFlood": 35456,
+      "Ġwatching": 35457,
+      "GTP": 35458,
+      "iel": 35459,
+      "müller": 35460,
+      "åħ": 35461,
+      "Ġpolyphenol": 35462,
+      "ĠLMI": 35463,
+      "redit": 35464,
+      "therm": 35465,
+      "Ġneurite": 35466,
+      "Quantum": 35467,
+      "rachlor": 35468,
+      "ĠRubin": 35469,
+      "Ġbfnm": 35470,
+      "Are": 35471,
+      "arachn": 35472,
+      "Ġduck": 35473,
+      "ĠTrajectory": 35474,
+      "ĠNitric": 35475,
+      "lv": 35476,
+      "uid": 35477,
+      "imag": 35478,
+      "ĠMULT": 35479,
+      "Ġgenre": 35480,
+      "arie": 35481,
+      "Ġtrifluor": 35482,
+      "ĠCorpus": 35483,
+      "oliosis": 35484,
+      "ĠCCK": 35485,
+      "Kit": 35486,
+      "father": 35487,
+      "Ġtennis": 35488,
+      "itsch": 35489,
+      "HCV": 35490,
+      "lantic": 35491,
+      "ĠAQ": 35492,
+      "izu": 35493,
+      "astatin": 35494,
+      "othio": 35495,
+      "ĠAnatomy": 35496,
+      "Ġáŀ": 35497,
+      "globulin": 35498,
+      "Ġinterpol": 35499,
+      "Ġtunnels": 35500,
+      "Ġnatri": 35501,
+      "imed": 35502,
+      "ĠDew": 35503,
+      "Ġsubscripts": 35504,
+      "tites": 35505,
+      "Ġhistologically": 35506,
+      "Opt": 35507,
+      "xn": 35508,
+      "Ġresampling": 35509,
+      "aney": 35510,
+      "Ġtrast": 35511,
+      "Ġsinensis": 35512,
+      "Ġsenescent": 35513,
+      "Fast": 35514,
+      "Ġhampered": 35515,
+      "Ġblocker": 35516,
+      "ushima": 35517,
+      "Ġhospitalizations": 35518,
+      "Lim": 35519,
+      "oons": 35520,
+      "ÿ": 35521,
+      "ĠAPS": 35522,
+      "ĠYok": 35523,
+      "ĠZam": 35524,
+      "Ġexperimenter": 35525,
+      "ĠDisks": 35526,
+      "Ġà¬": 35527,
+      "ĠScop": 35528,
+      "ĠAph": 35529,
+      "ĠParents": 35530,
+      "ĠPlots": 35531,
+      "ĠCONT": 35532,
+      "ĠÐĪ": 35533,
+      "Ġhomologue": 35534,
+      "ĠCooling": 35535,
+      "reth": 35536,
+      "Ġovari": 35537,
+      "ĠTamil": 35538,
+      "vrule": 35539,
+      "ĠPCP": 35540,
+      "arious": 35541,
+      "Active": 35542,
+      "oprotection": 35543,
+      "ĠAlfv": 35544,
+      "Ġinfra": 35545,
+      "ĠCoherence": 35546,
+      "closures": 35547,
+      "hydroxymethyl": 35548,
+      "EH": 35549,
+      "Ġmaser": 35550,
+      "ĠNIST": 35551,
+      "leck": 35552,
+      "concat": 35553,
+      "Ġtraine": 35554,
+      "Ġmixes": 35555,
+      "Ġribosomes": 35556,
+      "lia": 35557,
+      "puls": 35558,
+      "Ġascer": 35559,
+      "ĠBanks": 35560,
+      "ellin": 35561,
+      "applied": 35562,
+      "Ġclips": 35563,
+      "Ġmetap": 35564,
+      "Ġcopro": 35565,
+      "Ġepidid": 35566,
+      "ĠEpidemiological": 35567,
+      "ĠNicholas": 35568,
+      "ĠKings": 35569,
+      "Ġlarva": 35570,
+      "BsAg": 35571,
+      "ĠSánchez": 35572,
+      "ĠâĪİ": 35573,
+      "AMD": 35574,
+      "ĠHao": 35575,
+      "ĠBillboard": 35576,
+      "ĠAboriginal": 35577,
+      "Ġnylon": 35578,
+      "ĠNAN": 35579,
+      "cores": 35580,
+      "ĠCrop": 35581,
+      "Ġcommittees": 35582,
+      "Ġdihedral": 35583,
+      "ĠJuli": 35584,
+      "ĠAndy": 35585,
+      "hydration": 35586,
+      "corresponds": 35587,
+      "Mut": 35588,
+      "Ġtorn": 35589,
+      "ĠFEV": 35590,
+      "Ġxs": 35591,
+      "amphen": 35592,
+      "Ġsummarization": 35593,
+      "ĠErg": 35594,
+      "ËĨ": 35595,
+      "ĠJunction": 35596,
+      "ancouver": 35597,
+      "ĠExamining": 35598,
+      "ĠMarco": 35599,
+      "Pointer": 35600,
+      "Ġscarcity": 35601,
+      "uncing": 35602,
+      "Ġbijective": 35603,
+      "ĠMaine": 35604,
+      "ĠRHIC": 35605,
+      "Ġtowers": 35606,
+      "Ġgentamicin": 35607,
+      "Ġtonic": 35608,
+      "ĠkT": 35609,
+      "Ġclimbing": 35610,
+      "Ġrecruits": 35611,
+      "ĠHotel": 35612,
+      "ĠJews": 35613,
+      "ĠRUNX": 35614,
+      "Ġaustenite": 35615,
+      "ĠOfficer": 35616,
+      "inent": 35617,
+      "ucc": 35618,
+      "ĠBidirectional": 35619,
+      "Ġmayor": 35620,
+      "ĠAssays": 35621,
+      "ĠERG": 35622,
+      "SNPs": 35623,
+      "dine": 35624,
+      "China": 35625,
+      "starting": 35626,
+      "Ġirrational": 35627,
+      "ĠDIFFE": 35628,
+      "Ġmilliseconds": 35629,
+      "Lik": 35630,
+      "inone": 35631,
+      "ĠâģĦ": 35632,
+      "Ġconspicuous": 35633,
+      "Ġsurplus": 35634,
+      "ĠXiong": 35635,
+      "Ġupgrade": 35636,
+      "Ġtimep": 35637,
+      "ĠÄĮ": 35638,
+      "TeV": 35639,
+      "orbidities": 35640,
+      "invalid": 35641,
+      "Ġvide": 35642,
+      "terra": 35643,
+      "Ġantin": 35644,
+      "emens": 35645,
+      "ocese": 35646,
+      "ĠKI": 35647,
+      "Ġevolutionarily": 35648,
+      "Ker": 35649,
+      "ĠLES": 35650,
+      "clamp": 35651,
+      "Ġslowed": 35652,
+      "glycoprotein": 35653,
+      "entieth": 35654,
+      "Ġabroad": 35655,
+      "Ġinterpolating": 35656,
+      "Ġcatalyze": 35657,
+      "ĠBelgian": 35658,
+      "Ġphotographed": 35659,
+      "Ġpector": 35660,
+      "ĠSIV": 35661,
+      "ĠELECT": 35662,
+      "Ġdesal": 35663,
+      "oneph": 35664,
+      "ĠClos": 35665,
+      "Ġaffordable": 35666,
+      "birds": 35667,
+      "gom": 35668,
+      "Ġrr": 35669,
+      "Ġuni": 35670,
+      "ĠGenus": 35671,
+      "ĠRegge": 35672,
+      "ĠMultidimensional": 35673,
+      "Ġpsychopathology": 35674,
+      "Ġcertification": 35675,
+      "Pattern": 35676,
+      "ĠTower": 35677,
+      "Ġstern": 35678,
+      "Ġsublattice": 35679,
+      "Ġgrat": 35680,
+      "Ġlyrics": 35681,
+      "fmt": 35682,
+      "oceptive": 35683,
+      "ĠdP": 35684,
+      "ĠHolmes": 35685,
+      "Ġbudgets": 35686,
+      "Ġeutectic": 35687,
+      "ĠPv": 35688,
+      "ĠGott": 35689,
+      "Ġdisinfection": 35690,
+      "Ġretinoic": 35691,
+      "ĠObst": 35692,
+      "Ġreplen": 35693,
+      "Ġâĸł": 35694,
+      "Kutta": 35695,
+      "Please": 35696,
+      "ĠCAG": 35697,
+      "ĠStir": 35698,
+      "speaking": 35699,
+      "Ġinsecticides": 35700,
+      "ĠFungi": 35701,
+      "Hod": 35702,
+      "RON": 35703,
+      "coil": 35704,
+      "ĠVisible": 35705,
+      "Ġinception": 35706,
+      "ĠeGFR": 35707,
+      "Ġreionization": 35708,
+      "Ġdomination": 35709,
+      "ĠMetro": 35710,
+      "Ġswept": 35711,
+      "MDS": 35712,
+      "Ġsubsidence": 35713,
+      "ĠFalls": 35714,
+      "ĠDrum": 35715,
+      "ĠConserved": 35716,
+      "ĠMyers": 35717,
+      "Ġadaptability": 35718,
+      "Ġlyophil": 35719,
+      "ulina": 35720,
+      "arelli": 35721,
+      "ocycles": 35722,
+      "ĠSOA": 35723,
+      "ĠdsDNA": 35724,
+      "ĠCEO": 35725,
+      "Ġanchoring": 35726,
+      "Ġdeactivation": 35727,
+      "yler": 35728,
+      "Ġinterestingly": 35729,
+      "Ġiliac": 35730,
+      "ĠBorg": 35731,
+      "ĠPTC": 35732,
+      "ocyanin": 35733,
+      "Ġunused": 35734,
+      "ĠCarrier": 35735,
+      "Which": 35736,
+      "Ġintervening": 35737,
+      "Ġprivile": 35738,
+      "hit": 35739,
+      "Ġcheaper": 35740,
+      "ĠCyclin": 35741,
+      "plying": 35742,
+      "ĠCleveland": 35743,
+      "ĠHahn": 35744,
+      "Ġagglut": 35745,
+      "ĠAnch": 35746,
+      "ĠRedox": 35747,
+      "Will": 35748,
+      "ĠLinn": 35749,
+      "rones": 35750,
+      "ĠNewcastle": 35751,
+      "ĠExpected": 35752,
+      "ĠOpportunities": 35753,
+      "ĠLarger": 35754,
+      "Ġleach": 35755,
+      "Ġpepper": 35756,
+      "Sha": 35757,
+      "sector": 35758,
+      "you": 35759,
+      "Ġreplications": 35760,
+      "cholesterolem": 35761,
+      "ĠInvasion": 35762,
+      "Ġbony": 35763,
+      "ĠHuber": 35764,
+      "thend": 35765,
+      "Ġrealised": 35766,
+      "Ġinvestments": 35767,
+      "Cataly": 35768,
+      "ĠWitt": 35769,
+      "ĠKai": 35770,
+      "Ġetched": 35771,
+      "ĠSTEM": 35772,
+      "Ġexcludes": 35773,
+      "Exec": 35774,
+      "ĠStrongly": 35775,
+      "ĠSymposium": 35776,
+      "ĠTuberculosis": 35777,
+      "ilance": 35778,
+      "ĠRIS": 35779,
+      "apia": 35780,
+      "ensated": 35781,
+      "neb": 35782,
+      "ĠChains": 35783,
+      "Ġenthus": 35784,
+      "quadrup": 35785,
+      "decl": 35786,
+      "Ġbinned": 35787,
+      "Ġsynergistically": 35788,
+      "Ġgauges": 35789,
+      "whether": 35790,
+      "disease": 35791,
+      "Western": 35792,
+      "Ġhypothermia": 35793,
+      "ĠGardner": 35794,
+      "Ġaberration": 35795,
+      "Rod": 35796,
+      "Íĺ": 35797,
+      "Ġfd": 35798,
+      "Ġstood": 35799,
+      "Ġconditionally": 35800,
+      "Ġthrombol": 35801,
+      "PSC": 35802,
+      "Ġmk": 35803,
+      "ĠTER": 35804,
+      "odds": 35805,
+      "ĠKri": 35806,
+      "ĠIVF": 35807,
+      "Ġmites": 35808,
+      "ĠCHE": 35809,
+      "Ġqq": 35810,
+      "ĠInfants": 35811,
+      "ĠCharlot": 35812,
+      "becco": 35813,
+      "etom": 35814,
+      "ĠCDS": 35815,
+      "Ġarchaeal": 35816,
+      "ĠHNSCC": 35817,
+      "Ġmonodromy": 35818,
+      "amphenicol": 35819,
+      "apers": 35820,
+      "reactivity": 35821,
+      "Ġunderm": 35822,
+      "Internal": 35823,
+      "ĠLandsat": 35824,
+      "German": 35825,
+      "Ġcervix": 35826,
+      "idazole": 35827,
+      "ĠSketch": 35828,
+      "ĠLAM": 35829,
+      "ĠNerve": 35830,
+      "ĠTeh": 35831,
+      "Ġmussel": 35832,
+      "з": 35833,
+      "ĠMicroarray": 35834,
+      "wei": 35835,
+      "Ġwhey": 35836,
+      "Ġmixer": 35837,
+      "Ġreconfigurable": 35838,
+      "Ġvasculitis": 35839,
+      "Ġkwargs": 35840,
+      "Ġreus": 35841,
+      "correlations": 35842,
+      "Ġwoody": 35843,
+      "carbonate": 35844,
+      "ectomized": 35845,
+      "Ġretrans": 35846,
+      "Ġcytometric": 35847,
+      "ĠWildlife": 35848,
+      "ĠAnswering": 35849,
+      "Ġpencil": 35850,
+      "ĠDAS": 35851,
+      "akrish": 35852,
+      "CEPT": 35853,
+      "ĠÄĿ": 35854,
+      "ĠPersian": 35855,
+      "converting": 35856,
+      "Ġcater": 35857,
+      "Ġmeanwhile": 35858,
+      "TPA": 35859,
+      "Ġrum": 35860,
+      "ĠGros": 35861,
+      "upe": 35862,
+      "Ġregurg": 35863,
+      "Ġpenalties": 35864,
+      "Positive": 35865,
+      "****************************************************************************": 35866,
+      "XF": 35867,
+      "eenth": 35868,
+      "ĠCory": 35869,
+      "odulation": 35870,
+      "Ġquorum": 35871,
+      "codes": 35872,
+      "aram": 35873,
+      "ĠTSA": 35874,
+      "ĠPn": 35875,
+      "âĪij": 35876,
+      "prison": 35877,
+      "Ġconfidentiality": 35878,
+      "EPS": 35879,
+      "Xiv": 35880,
+      "iensis": 35881,
+      "estones": 35882,
+      "ĠZag": 35883,
+      "Ġpredecessor": 35884,
+      "Ġprize": 35885,
+      "Ġâݨ": 35886,
+      "steroidal": 35887,
+      "opard": 35888,
+      "Ġimpractical": 35889,
+      "Ġdemonstrations": 35890,
+      "Ġpredisposition": 35891,
+      "Ġkk": 35892,
+      "Ġmodifiers": 35893,
+      "Ġpreca": 35894,
+      "Ġexecutes": 35895,
+      "Ġbinning": 35896,
+      "Ġpedig": 35897,
+      "ĠKLF": 35898,
+      "ĠSkeletal": 35899,
+      "ĠCIN": 35900,
+      "atured": 35901,
+      "Ġdecomposes": 35902,
+      "Ġaphid": 35903,
+      "Bern": 35904,
+      "Pur": 35905,
+      "ĠEPO": 35906,
+      "merge": 35907,
+      "ĠCOSM": 35908,
+      "amyloid": 35909,
+      "monia": 35910,
+      "ĠScores": 35911,
+      "ĠRegistration": 35912,
+      "ĠAgrobacterium": 35913,
+      "Ġenterprises": 35914,
+      "locality": 35915,
+      "ĠITO": 35916,
+      "Ġtess": 35917,
+      "Ġfcc": 35918,
+      "ĠNc": 35919,
+      "Ġcoaxial": 35920,
+      "ĠAdvant": 35921,
+      "APC": 35922,
+      "ĠDemand": 35923,
+      "adjust": 35924,
+      "Points": 35925,
+      "Ġheterostructure": 35926,
+      "iffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiff": 35927,
+      "DQ": 35928,
+      "Ġtensions": 35929,
+      "abund": 35930,
+      "ĠHutch": 35931,
+      "brew": 35932,
+      "Ġvitreous": 35933,
+      "ĠEZH": 35934,
+      "Ġmerc": 35935,
+      "Ġdebated": 35936,
+      "Ġpalate": 35937,
+      "ocolate": 35938,
+      "Ġevapotranspiration": 35939,
+      "ĠẼ": 35940,
+      "ĠHoffman": 35941,
+      "ĠGALAXIES": 35942,
+      "CAL": 35943,
+      "caps": 35944,
+      "legal": 35945,
+      "died": 35946,
+      "ĠIsolates": 35947,
+      "Ġaggrav": 35948,
+      "qs": 35949,
+      "ĠICT": 35950,
+      "Ġseals": 35951,
+      "Ġspinel": 35952,
+      "ĠGeor": 35953,
+      "Blue": 35954,
+      "Ġureter": 35955,
+      "spline": 35956,
+      "ĠIntroducing": 35957,
+      "thendieck": 35958,
+      "opper": 35959,
+      "Ġafterglow": 35960,
+      "Ġendosomal": 35961,
+      "Ġrealizes": 35962,
+      "solving": 35963,
+      "Ġmistake": 35964,
+      "ĠAtheros": 35965,
+      "ĠSBS": 35966,
+      "ĠRut": 35967,
+      "exist": 35968,
+      "Prof": 35969,
+      "ĠNeisser": 35970,
+      "MSG": 35971,
+      "ĠEarlier": 35972,
+      "ĠdT": 35973,
+      "ĠSpread": 35974,
+      "ĠReflection": 35975,
+      "secondary": 35976,
+      "approximate": 35977,
+      "Ġnigra": 35978,
+      "Solution": 35979,
+      "anone": 35980,
+      "ĠItems": 35981,
+      "Ġwavelets": 35982,
+      "ĠSoluble": 35983,
+      "Ġcircularly": 35984,
+      "ĠCUDA": 35985,
+      "Ġregenerated": 35986,
+      "SPI": 35987,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 35988,
+      "aturing": 35989,
+      "REQ": 35990,
+      "Ġinteroper": 35991,
+      "reev": 35992,
+      "ONT": 35993,
+      "ischen": 35994,
+      "ĠChoosing": 35995,
+      "phosphorylated": 35996,
+      "áĪ": 35997,
+      "Ġdress": 35998,
+      "ĠConform": 35999,
+      "Ġrememb": 36000,
+      "Ġischaemic": 36001,
+      "Basic": 36002,
+      "ĠPang": 36003,
+      "Ġcrit": 36004,
+      "ĠOrn": 36005,
+      "Ġgm": 36006,
+      "ĠFog": 36007,
+      "ĠBd": 36008,
+      "racheal": 36009,
+      "Ġphenols": 36010,
+      "ĠDistingu": 36011,
+      "Ġâİ©": 36012,
+      "ĠGRBs": 36013,
+      "ĠCeO": 36014,
+      "ĠBiomass": 36015,
+      "Ġaptamer": 36016,
+      "visc": 36017,
+      "hetically": 36018,
+      "Ġsid": 36019,
+      "omeg": 36020,
+      "Ġproportionality": 36021,
+      "ÃŃs": 36022,
+      "toplasmic": 36023,
+      "ĠConnected": 36024,
+      "Ġlaminin": 36025,
+      "strahlung": 36026,
+      "ĠLad": 36027,
+      "TRAN": 36028,
+      "är": 36029,
+      "Ġbasalt": 36030,
+      "ĠCurvature": 36031,
+      "Ġmitigating": 36032,
+      "opaedic": 36033,
+      "ĠMuhammad": 36034,
+      "CAR": 36035,
+      "Gi": 36036,
+      "Ġetch": 36037,
+      "hair": 36038,
+      "Ġpurine": 36039,
+      "Ġbenchmarking": 36040,
+      "reich": 36041,
+      "Ġmethicillin": 36042,
+      "âĪ¥": 36043,
+      "Ġmanages": 36044,
+      "solvent": 36045,
+      "ĠShao": 36046,
+      "hc": 36047,
+      "Ġstruck": 36048,
+      "Ġnucleosome": 36049,
+      "ĠPublication": 36050,
+      "Metric": 36051,
+      "Ġwines": 36052,
+      "ĠMBL": 36053,
+      "ĠHub": 36054,
+      "ĠAssistant": 36055,
+      "Ġreliance": 36056,
+      "Ġrouters": 36057,
+      "ĠHerz": 36058,
+      "ĠTobacco": 36059,
+      "rogram": 36060,
+      "ĠHSD": 36061,
+      "ĠLBP": 36062,
+      "Ġinflection": 36063,
+      "school": 36064,
+      "Ġsponsored": 36065,
+      "ĠCenozoic": 36066,
+      "Ġentertainment": 36067,
+      "atian": 36068,
+      "architecture": 36069,
+      "browse": 36070,
+      "REC": 36071,
+      "isture": 36072,
+      "ĠCholesterol": 36073,
+      "ĠSimplified": 36074,
+      "Ġpolypeptides": 36075,
+      "Ġpunctures": 36076,
+      "arachnoid": 36077,
+      "Self": 36078,
+      "Ġanorexia": 36079,
+      "ĠOle": 36080,
+      "ĉĉĠĠĠĠ": 36081,
+      "GBT": 36082,
+      "Ġcardiomyocyte": 36083,
+      "ĠFloquet": 36084,
+      "analog": 36085,
+      "Ġsensitized": 36086,
+      "ĠCephe": 36087,
+      "catch": 36088,
+      "chial": 36089,
+      "Ġceremony": 36090,
+      "Ġterat": 36091,
+      "Ġameliorate": 36092,
+      "olysin": 36093,
+      "etooth": 36094,
+      "akin": 36095,
+      "haem": 36096,
+      "Ġentropies": 36097,
+      "Ġargu": 36098,
+      "Ġcopied": 36099,
+      "lington": 36100,
+      "ĠHerpes": 36101,
+      "ĠSchwann": 36102,
+      "yk": 36103,
+      "ĠCEA": 36104,
+      "ĠICH": 36105,
+      "Ġwrink": 36106,
+      "Ġrunners": 36107,
+      "Ġgalvan": 36108,
+      "Ġconsolidated": 36109,
+      "ĠâĢ¡": 36110,
+      "ĠClassic": 36111,
+      "Ġepidemiologic": 36112,
+      "ĠDriving": 36113,
+      "Ġtrastuzumab": 36114,
+      "CYP": 36115,
+      "NCT": 36116,
+      "tability": 36117,
+      "Ġslee": 36118,
+      "ĠNeck": 36119,
+      "Ġassesses": 36120,
+      "Ġsymmetrically": 36121,
+      "ĠPotts": 36122,
+      "ĠRibosomal": 36123,
+      "diction": 36124,
+      "gall": 36125,
+      "ĠAcceleration": 36126,
+      "CLA": 36127,
+      "ACTER": 36128,
+      "xed": 36129,
+      "Ġgeriatric": 36130,
+      "threonine": 36131,
+      "Ġabort": 36132,
+      "Ġartem": 36133,
+      "ĠDisney": 36134,
+      "ĠCorrespondence": 36135,
+      "Ġrent": 36136,
+      "ĠNUM": 36137,
+      "ĠChun": 36138,
+      "ĠRecogn": 36139,
+      "Ġcrystallized": 36140,
+      "Ġcontradicting": 36141,
+      "visors": 36142,
+      "malignant": 36143,
+      "rophysiology": 36144,
+      "Infrared": 36145,
+      "gz": 36146,
+      "Ġsublim": 36147,
+      "omatosis": 36148,
+      "osyltransferase": 36149,
+      "Ġholography": 36150,
+      "orenstein": 36151,
+      "¾±": 36152,
+      "ĠSebas": 36153,
+      "accum": 36154,
+      "Upper": 36155,
+      "antenna": 36156,
+      "Ġblur": 36157,
+      "Ġsmell": 36158,
+      "Ġthreefold": 36159,
+      "ĠPlayers": 36160,
+      "Ġalleviated": 36161,
+      "Bin": 36162,
+      "Ġninet": 36163,
+      "ĠDna": 36164,
+      "Ġgeneralizing": 36165,
+      "Ġbreakage": 36166,
+      "ĠMorrison": 36167,
+      "macro": 36168,
+      "Reader": 36169,
+      "ogravimetric": 36170,
+      "Ġdh": 36171,
+      "lew": 36172,
+      "xton": 36173,
+      "Ġdeceleration": 36174,
+      "ĠCorrelated": 36175,
+      "ĠLegion": 36176,
+      "Ġgambling": 36177,
+      "Binding": 36178,
+      "ĠInAs": 36179,
+      "lowering": 36180,
+      "Ġeuthanized": 36181,
+      "ĠDallas": 36182,
+      "ĠDw": 36183,
+      "ĠDijk": 36184,
+      "ĠPolic": 36185,
+      "ĠTIME": 36186,
+      "ĠHEL": 36187,
+      "ĠLanguages": 36188,
+      "Ġparabol": 36189,
+      "porating": 36190,
+      "Ġfrustration": 36191,
+      "μM": 36192,
+      "balls": 36193,
+      "ĠArmstrong": 36194,
+      "Ġcontractility": 36195,
+      "Ġmetalloproteinases": 36196,
+      "americ": 36197,
+      "ĠZak": 36198,
+      "ĠCosts": 36199,
+      "Alex": 36200,
+      "dog": 36201,
+      "pw": 36202,
+      "ĠTight": 36203,
+      "ĠAnterior": 36204,
+      "Ġpeaking": 36205,
+      "Ġnegativity": 36206,
+      "Ġhydride": 36207,
+      "ĠLiv": 36208,
+      "Ġsterilized": 36209,
+      "Ġverbatim": 36210,
+      "Alternatively": 36211,
+      "REQU": 36212,
+      "ĠTyphimurium": 36213,
+      "ĠWeinberg": 36214,
+      "DSC": 36215,
+      "rq": 36216,
+      "Ġcorrug": 36217,
+      "Ġmicrons": 36218,
+      "coord": 36219,
+      "ioid": 36220,
+      "sat": 36221,
+      "Ġflocc": 36222,
+      "ĠAccelerated": 36223,
+      "Ġsixteen": 36224,
+      "absence": 36225,
+      "ĠSpeaker": 36226,
+      "omological": 36227,
+      "ĠApr": 36228,
+      "Ġmatroid": 36229,
+      "tight": 36230,
+      "ogenetically": 36231,
+      "rump": 36232,
+      "ĠInhibits": 36233,
+      "ĠOlympus": 36234,
+      "Ġpossession": 36235,
+      "Ġsupervisor": 36236,
+      "Ġconcise": 36237,
+      "optimized": 36238,
+      "vivo": 36239,
+      "Ġstepped": 36240,
+      "ocyanine": 36241,
+      "Five": 36242,
+      "anas": 36243,
+      "arten": 36244,
+      "ĠCaco": 36245,
+      "Ġsolutes": 36246,
+      "ITAL": 36247,
+      "ĠReddy": 36248,
+      "Ġwarping": 36249,
+      "Ġoligomer": 36250,
+      "Ġcapped": 36251,
+      "Ġvoted": 36252,
+      "ĠRico": 36253,
+      "ĠTrem": 36254,
+      "Ġlime": 36255,
+      "ĠISP": 36256,
+      "ĠLayers": 36257,
+      "skin": 36258,
+      "ranged": 36259,
+      "áz": 36260,
+      "Ġbioactivity": 36261,
+      "Ġdurable": 36262,
+      "Ġhn": 36263,
+      "ĠCAB": 36264,
+      "Ġva": 36265,
+      "ĠUWB": 36266,
+      "ĠStuart": 36267,
+      "Ġlengthy": 36268,
+      "Ġinvasiveness": 36269,
+      "ĠâĩĶ": 36270,
+      "joining": 36271,
+      "ĠRBCs": 36272,
+      "Ġresilient": 36273,
+      "ĠManipulation": 36274,
+      "Germ": 36275,
+      "contribution": 36276,
+      "Ġqualify": 36277,
+      "ĠDashed": 36278,
+      "Ġaccelerations": 36279,
+      "ĠCytochrome": 36280,
+      "Ġcircumstellar": 36281,
+      "cavity": 36282,
+      "Ġanatase": 36283,
+      "ĠDevi": 36284,
+      "Ġpursu": 36285,
+      "ĠMicroRNAs": 36286,
+      "Ġnorthward": 36287,
+      "Ġsunflower": 36288,
+      "ĠEntertainment": 36289,
+      "Pacific": 36290,
+      "ĠHolographic": 36291,
+      "uj": 36292,
+      "erell": 36293,
+      "methanol": 36294,
+      "Surface": 36295,
+      "opositive": 36296,
+      "Ġthreatening": 36297,
+      "Ġtranscend": 36298,
+      "Depend": 36299,
+      "Ġqi": 36300,
+      "tised": 36301,
+      "ĠBristol": 36302,
+      "ummation": 36303,
+      "Ġextractor": 36304,
+      "Ġfavoured": 36305,
+      "ĠPyro": 36306,
+      "ĠEngineers": 36307,
+      "flatten": 36308,
+      "tolerance": 36309,
+      "Ġxt": 36310,
+      "ĠTot": 36311,
+      "Ġtestbed": 36312,
+      "ICU": 36313,
+      "ĠSwarm": 36314,
+      "Ġinternationally": 36315,
+      "Ġantine": 36316,
+      "ĠInsurance": 36317,
+      "bai": 36318,
+      "nh": 36319,
+      "Ñĭ": 36320,
+      "osac": 36321,
+      "ĠLec": 36322,
+      "thor": 36323,
+      "Ġoutermost": 36324,
+      "Ġdoors": 36325,
+      "Ġbiometric": 36326,
+      "glutamate": 36327,
+      "ĠWoods": 36328,
+      "ĠMunich": 36329,
+      "uximab": 36330,
+      "places": 36331,
+      "Ġamyotrophic": 36332,
+      "ĠParam": 36333,
+      "ĠChristensen": 36334,
+      "Age": 36335,
+      "enne": 36336,
+      "Ġanim": 36337,
+      "Ġrecrystallization": 36338,
+      "ĠPropositions": 36339,
+      "Ġsnails": 36340,
+      "Secondly": 36341,
+      "ĠPUFA": 36342,
+      "France": 36343,
+      "Src": 36344,
+      "vitro": 36345,
+      "omass": 36346,
+      "uru": 36347,
+      "ĠLever": 36348,
+      "ectonic": 36349,
+      "embl": 36350,
+      "PCL": 36351,
+      "Ġcoordinator": 36352,
+      "ĠFoxp": 36353,
+      "ĠBirmingham": 36354,
+      "ĠLiberal": 36355,
+      "Ġcruise": 36356,
+      "Ġiθ": 36357,
+      "Ġsymp": 36358,
+      "azaki": 36359,
+      "ĠParse": 36360,
+      "Ġhydrologic": 36361,
+      "Ġprolongation": 36362,
+      "ĠHayes": 36363,
+      "Ġsubmuc": 36364,
+      "Ġagglomeration": 36365,
+      "ARE": 36366,
+      "ĠFMR": 36367,
+      "ĠLomb": 36368,
+      "mathchar": 36369,
+      "Ġstructuring": 36370,
+      "Ġelectrophoretic": 36371,
+      "Ġdiminishing": 36372,
+      "Ġbrake": 36373,
+      "chenko": 36374,
+      "ĠPereira": 36375,
+      "lens": 36376,
+      "Ġbackend": 36377,
+      "Ġillustrations": 36378,
+      "Ġdemanded": 36379,
+      "Ġnoticeably": 36380,
+      "ĠKaiser": 36381,
+      "ĠDavidson": 36382,
+      "Ġbraking": 36383,
+      "Tp": 36384,
+      "Forward": 36385,
+      "μν": 36386,
+      "ĠCdS": 36387,
+      "Ġasteroids": 36388,
+      "Provider": 36389,
+      "ĠEut": 36390,
+      "Ġtril": 36391,
+      "ungs": 36392,
+      "Ġdiving": 36393,
+      "ĠUAVs": 36394,
+      "ĠiPSC": 36395,
+      "iint": 36396,
+      "Ġ×": 36397,
+      "thrombin": 36398,
+      "Ġcoordinating": 36399,
+      "extrem": 36400,
+      "Ġembolization": 36401,
+      "ĠAdip": 36402,
+      "plated": 36403,
+      "ĠHag": 36404,
+      "ĠETS": 36405,
+      "Ġbrood": 36406,
+      "Ang": 36407,
+      "ĠPCV": 36408,
+      "detail": 36409,
+      "RSS": 36410,
+      "bens": 36411,
+      "Ġtier": 36412,
+      "ĠCock": 36413,
+      "Ġgay": 36414,
+      "Ġquint": 36415,
+      "Ġagenda": 36416,
+      "Ġaffairs": 36417,
+      "ĠModerate": 36418,
+      "helical": 36419,
+      "ĠEquivalent": 36420,
+      "Ġproportionally": 36421,
+      "Column": 36422,
+      "FWHM": 36423,
+      "Air": 36424,
+      "Enum": 36425,
+      "ifice": 36426,
+      "arcsec": 36427,
+      "ĠTRIM": 36428,
+      "ĠLabeling": 36429,
+      "QAM": 36430,
+      "pies": 36431,
+      "Ġisotropy": 36432,
+      "ĠGó": 36433,
+      "Ġpointers": 36434,
+      "tigraphy": 36435,
+      "ramers": 36436,
+      "Ġmacaque": 36437,
+      "Ġmisses": 36438,
+      "Ġellipticity": 36439,
+      "presented": 36440,
+      "galactosidase": 36441,
+      "ÉĽ": 36442,
+      "inion": 36443,
+      "Ġmite": 36444,
+      "lll": 36445,
+      "Objective": 36446,
+      "Ġprisoners": 36447,
+      "ĠHercules": 36448,
+      "Ġantis": 36449,
+      "Ġclosures": 36450,
+      "ĠMartian": 36451,
+      "Ġterpen": 36452,
+      "robust": 36453,
+      "Ġsequelae": 36454,
+      "alarial": 36455,
+      "ĠCSA": 36456,
+      "ĠBland": 36457,
+      "ĠGent": 36458,
+      "Ġorphan": 36459,
+      "Ġindent": 36460,
+      "bigwedge": 36461,
+      "Ġdefinable": 36462,
+      "Ġoligosaccharides": 36463,
+      "ĠBattalion": 36464,
+      "Ġisometries": 36465,
+      "azolin": 36466,
+      "ĠShown": 36467,
+      "spectra": 36468,
+      "Visual": 36469,
+      "<<<<<<<<": 36470,
+      "Ġlentiviral": 36471,
+      "othelioma": 36472,
+      "Ġtedious": 36473,
+      "ĠBCI": 36474,
+      "Ġgeologic": 36475,
+      "Ġconsumes": 36476,
+      "ĠAblation": 36477,
+      "least": 36478,
+      "Ġthigh": 36479,
+      "Ġsecrecy": 36480,
+      "covering": 36481,
+      "eiro": 36482,
+      "õ": 36483,
+      "ĠTBS": 36484,
+      "Ġisomerase": 36485,
+      "Ġrecommends": 36486,
+      "ĠVortex": 36487,
+      "ĠBray": 36488,
+      "Ġsubd": 36489,
+      "ĠOptions": 36490,
+      "Ġmetamaterial": 36491,
+      "ĠSquares": 36492,
+      "trap": 36493,
+      "imon": 36494,
+      "Ġhesit": 36495,
+      "Ġabc": 36496,
+      "cessing": 36497,
+      "ĠRET": 36498,
+      "Ġpinned": 36499,
+      "Ġketones": 36500,
+      "Ġwelded": 36501,
+      "ĠMitochondria": 36502,
+      "Ġingested": 36503,
+      "ĠQFT": 36504,
+      "Ġcomparator": 36505,
+      "Ġoxidoreductase": 36506,
+      "Ġtetrad": 36507,
+      "ĠSensitive": 36508,
+      "Ġcatchments": 36509,
+      "Ġrefugees": 36510,
+      "Ġpuberty": 36511,
+      "Arab": 36512,
+      "Ġinterannual": 36513,
+      "scattered": 36514,
+      "ĠMetam": 36515,
+      "Ġcyclization": 36516,
+      "pertures": 36517,
+      "ĠLINC": 36518,
+      "rules": 36519,
+      "ĠPont": 36520,
+      "PTH": 36521,
+      "ĉĉĉĉĉĉĉĉ": 36522,
+      "Santa": 36523,
+      "ĠLNC": 36524,
+      "Ġsubmodular": 36525,
+      "rective": 36526,
+      "Ġtrif": 36527,
+      "Ġsentinel": 36528,
+      "ĠTwin": 36529,
+      "keletons": 36530,
+      "miral": 36531,
+      "aming": 36532,
+      "ĠGay": 36533,
+      "Ġinterspecific": 36534,
+      "Ġrelieve": 36535,
+      "Ġendomorphism": 36536,
+      "ĠExpanding": 36537,
+      "ĠRuntime": 36538,
+      "yang": 36539,
+      "requires": 36540,
+      "odine": 36541,
+      "ometabolic": 36542,
+      "Store": 36543,
+      "planet": 36544,
+      "Ġrenov": 36545,
+      "___": 36546,
+      "adenosine": 36547,
+      "uitive": 36548,
+      "Ġkel": 36549,
+      "ĠProlong": 36550,
+      "ĠAdvance": 36551,
+      "Ġantimicrobials": 36552,
+      "ĠMunicipal": 36553,
+      "ĠNeutrophil": 36554,
+      "FAs": 36555,
+      "ĠFame": 36556,
+      "ibus": 36557,
+      "ETE": 36558,
+      "Ġstepping": 36559,
+      "ĠBlot": 36560,
+      "ĠLaura": 36561,
+      "Ġrocky": 36562,
+      "ĠLima": 36563,
+      "Ġmitigated": 36564,
+      "ĠLambert": 36565,
+      "Ġunexplored": 36566,
+      "Ġtrigonometric": 36567,
+      "pig": 36568,
+      "ĠHeli": 36569,
+      "Ġfinely": 36570,
+      "Ġoxidizing": 36571,
+      "Ġcolonoscopy": 36572,
+      "activities": 36573,
+      "ĠEasy": 36574,
+      "Ġunexplained": 36575,
+      "aky": 36576,
+      "ASM": 36577,
+      "worker": 36578,
+      "ĠCrist": 36579,
+      "ãĢģ": 36580,
+      "ulk": 36581,
+      "ĠSugg": 36582,
+      "ĠMim": 36583,
+      "Ġiterates": 36584,
+      "Ġsulfoxide": 36585,
+      "glucan": 36586,
+      "Ġreactant": 36587,
+      "Ġphagocytic": 36588,
+      "Brain": 36589,
+      "ucted": 36590,
+      "ĠScand": 36591,
+      "ĠCaCO": 36592,
+      "Ġaffiliation": 36593,
+      "Policy": 36594,
+      "ĠInfantry": 36595,
+      "Functional": 36596,
+      "rtimes": 36597,
+      "Ġwond": 36598,
+      "ardment": 36599,
+      "ĠWeil": 36600,
+      "Ġdirectors": 36601,
+      "uffix": 36602,
+      "ĠRuiz": 36603,
+      "ĠPhenomena": 36604,
+      "Ġmicrob": 36605,
+      "cosm": 36606,
+      "Ġutilisation": 36607,
+      "persed": 36608,
+      "Ġconsole": 36609,
+      "ticulate": 36610,
+      "Ġdesens": 36611,
+      "Ġreplicas": 36612,
+      "Ġpluripotency": 36613,
+      "ĠUkrainian": 36614,
+      "Ġhydrolyzed": 36615,
+      "ĠBiodiversity": 36616,
+      "Efficient": 36617,
+      "ĠKash": 36618,
+      "minor": 36619,
+      "Ġconclusive": 36620,
+      "Ġtentative": 36621,
+      "jira": 36622,
+      "Ġmb": 36623,
+      "ĠIPA": 36624,
+      "ĠPis": 36625,
+      "Ġgoverns": 36626,
+      "ĠSouthwest": 36627,
+      "oeba": 36628,
+      "ĠMohammad": 36629,
+      "albumin": 36630,
+      "circles": 36631,
+      "ĠHedge": 36632,
+      "ĠAmph": 36633,
+      "BACK": 36634,
+      "Old": 36635,
+      "histor": 36636,
+      "acular": 36637,
+      "ĠNOR": 36638,
+      "henius": 36639,
+      "visions": 36640,
+      "missibility": 36641,
+      "Ġthromboembolism": 36642,
+      "atized": 36643,
+      "Ġwil": 36644,
+      "awing": 36645,
+      "ASI": 36646,
+      "Ġheterodimer": 36647,
+      "Ġbuffering": 36648,
+      "ĠIdeally": 36649,
+      "ĠEgg": 36650,
+      "ographies": 36651,
+      "ĠAppl": 36652,
+      "ĠCIs": 36653,
+      "meaning": 36654,
+      "ĠSMAD": 36655,
+      "Ġphenylalanine": 36656,
+      "ĠTitanium": 36657,
+      "ĠZariski": 36658,
+      "Ġnymph": 36659,
+      "Ġhired": 36660,
+      "ĠPPC": 36661,
+      "ĠKG": 36662,
+      "ĠGuill": 36663,
+      "oglycans": 36664,
+      "erial": 36665,
+      "Dele": 36666,
+      "ilus": 36667,
+      "ĠFitness": 36668,
+      "Ġwhales": 36669,
+      "grant": 36670,
+      "mostly": 36671,
+      "Ġclimates": 36672,
+      "ĠCampaign": 36673,
+      "MgO": 36674,
+      "Ġepistemic": 36675,
+      "Lipschitz": 36676,
+      "ĠLAT": 36677,
+      "Ġcladding": 36678,
+      "vacuum": 36679,
+      "agglutinin": 36680,
+      "kill": 36681,
+      "Ġsail": 36682,
+      "Ġartistic": 36683,
+      "answ": 36684,
+      "ĠSDF": 36685,
+      "ĠKeith": 36686,
+      "Ġsorafenib": 36687,
+      "Ġgallbladder": 36688,
+      "directory": 36689,
+      "Ġphotoreceptors": 36690,
+      "ĠFokker": 36691,
+      "DU": 36692,
+      "Ġeditors": 36693,
+      "Ġtelecommun": 36694,
+      "ardia": 36695,
+      "ĠPublications": 36696,
+      "Ġscrews": 36697,
+      "ĠMathematica": 36698,
+      "RSV": 36699,
+      "ĠApply": 36700,
+      "ĠSTS": 36701,
+      "ĠMurine": 36702,
+      "Ġdump": 36703,
+      "Ġlingu": 36704,
+      "ĠDixon": 36705,
+      "Ġovercomes": 36706,
+      "ĠPreoperative": 36707,
+      "Ġmigrant": 36708,
+      "Ġbelieves": 36709,
+      "BK": 36710,
+      "actively": 36711,
+      "ĠISC": 36712,
+      "quas": 36713,
+      "Ġalga": 36714,
+      "ichael": 36715,
+      "Ġdisasters": 36716,
+      "Ġpracticed": 36717,
+      "hydrophobic": 36718,
+      "ĠNiño": 36719,
+      "ĠEthanol": 36720,
+      "QE": 36721,
+      "ĠSJ": 36722,
+      "ĠDengue": 36723,
+      "Ġappl": 36724,
+      "ĠYoon": 36725,
+      "enzo": 36726,
+      "IFY": 36727,
+      "Ġchronological": 36728,
+      "erin": 36729,
+      "ĠPeg": 36730,
+      "ĠRelevant": 36731,
+      "Ġqualification": 36732,
+      "evine": 36733,
+      "Ġdendrite": 36734,
+      "DTD": 36735,
+      "cholinesterase": 36736,
+      "watch": 36737,
+      "ĠSanchez": 36738,
+      "Ġwashes": 36739,
+      "Ġpermafrost": 36740,
+      "ĠTertiary": 36741,
+      "Ġsynthesizing": 36742,
+      "Ġexpedition": 36743,
+      "routine": 36744,
+      "ĠSearching": 36745,
+      "ĠSé": 36746,
+      "residual": 36747,
+      "ĠLCD": 36748,
+      "entities": 36749,
+      "Ġendovascular": 36750,
+      "Ġparamount": 36751,
+      "pher": 36752,
+      "Ġstraightforwardly": 36753,
+      "Ġvasodil": 36754,
+      "ĠSchistosoma": 36755,
+      "Ġpermissions": 36756,
+      "centred": 36757,
+      "Ġfrustrated": 36758,
+      "structuring": 36759,
+      "ĠSchl": 36760,
+      "ĠInitiation": 36761,
+      "Ġcuticle": 36762,
+      "Ġforgetting": 36763,
+      "ĠSas": 36764,
+      "ĠSult": 36765,
+      "uno": 36766,
+      "Ġdisintegration": 36767,
+      "ĠVG": 36768,
+      "Ġwards": 36769,
+      "ĠIRE": 36770,
+      "upro": 36771,
+      "Ġsubgen": 36772,
+      "Ġsubclasses": 36773,
+      "ĠStand": 36774,
+      "ĠHeight": 36775,
+      "interpretation": 36776,
+      "Ġglycan": 36777,
+      "ĠSolvent": 36778,
+      "ĠMalignant": 36779,
+      "Ġunsuitable": 36780,
+      "ĠCoxeter": 36781,
+      "Ġspermatogenesis": 36782,
+      "Ġfullerene": 36783,
+      "Fox": 36784,
+      "SOC": 36785,
+      "wet": 36786,
+      "armstadt": 36787,
+      "Ġpropofol": 36788,
+      "indexed": 36789,
+      "Ġsnakes": 36790,
+      "Edit": 36791,
+      "ĠmJy": 36792,
+      "RIB": 36793,
+      "Ġey": 36794,
+      "ĠAlkal": 36795,
+      "Ġtriaxial": 36796,
+      "PSK": 36797,
+      "neo": 36798,
+      "Ġendo": 36799,
+      "Ġglycosides": 36800,
+      "Ġsyllables": 36801,
+      "Ġsorghum": 36802,
+      "loor": 36803,
+      "Ġgeothermal": 36804,
+      "guinal": 36805,
+      "ĠSerbia": 36806,
+      "æĸ": 36807,
+      "ĠSentinel": 36808,
+      "ighters": 36809,
+      "Ġkeyboard": 36810,
+      "Ġbanana": 36811,
+      "granular": 36812,
+      "Ġdeciduous": 36813,
+      "ĠHAR": 36814,
+      "neuron": 36815,
+      "ĠCarn": 36816,
+      "Ġburns": 36817,
+      "Boost": 36818,
+      "ĠDeterministic": 36819,
+      "pipe": 36820,
+      "ĠFAD": 36821,
+      "ĠBovine": 36822,
+      "ĠRou": 36823,
+      "Ġkan": 36824,
+      "autonomous": 36825,
+      "utrients": 36826,
+      "Ġhypothyroidism": 36827,
+      "ĠSINR": 36828,
+      "stret": 36829,
+      "Ġunaltered": 36830,
+      "ĠZika": 36831,
+      "valley": 36832,
+      "Ġlongitudinally": 36833,
+      "Ġfluorescein": 36834,
+      "catheter": 36835,
+      "ĠCongenital": 36836,
+      "Ġpiez": 36837,
+      "Ġabbreviated": 36838,
+      "ĠChlamydia": 36839,
+      "Ġaired": 36840,
+      "Ġqueen": 36841,
+      "Ġinstructive": 36842,
+      "Ġabruptly": 36843,
+      "Ġrecurrences": 36844,
+      "IMP": 36845,
+      "Ġexosome": 36846,
+      "ĠHSCs": 36847,
+      "Writer": 36848,
+      "elis": 36849,
+      "ĠArithmetic": 36850,
+      "enarios": 36851,
+      "Ġligated": 36852,
+      "ĠLocalized": 36853,
+      "ĠFreeman": 36854,
+      "Ġcarniv": 36855,
+      "ĠCereb": 36856,
+      "Ġgrac": 36857,
+      "ĠGond": 36858,
+      "ĠVancouver": 36859,
+      "obox": 36860,
+      "Ġtyped": 36861,
+      "ĠÄ¥": 36862,
+      "Upon": 36863,
+      "Future": 36864,
+      "ENG": 36865,
+      "dead": 36866,
+      "Ġserpent": 36867,
+      "ĠAssignment": 36868,
+      "ĠUpdated": 36869,
+      "Ġhistorian": 36870,
+      "Ġtropospheric": 36871,
+      "Cloud": 36872,
+      "bumin": 36873,
+      "ĠPras": 36874,
+      "ĠBasket": 36875,
+      "ĠâĪĴâĪĴ": 36876,
+      "benzodi": 36877,
+      "ĠTrauma": 36878,
+      "ĠBehaviors": 36879,
+      "Ġpter": 36880,
+      "irradiation": 36881,
+      "Ġspoke": 36882,
+      "ariatric": 36883,
+      "Ġplugin": 36884,
+      "Ġsupersonic": 36885,
+      "Ġdocetaxel": 36886,
+      "itigation": 36887,
+      "Ġdigestibility": 36888,
+      "nem": 36889,
+      "Ġpb": 36890,
+      "ĠCSR": 36891,
+      "Ġfouling": 36892,
+      "Ġrheology": 36893,
+      "Ġfloods": 36894,
+      "Ġgluing": 36895,
+      "agascar": 36896,
+      "jets": 36897,
+      "pti": 36898,
+      "eston": 36899,
+      "ĠKü": 36900,
+      "Ġopenings": 36901,
+      "Ġisolating": 36902,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 36903,
+      "Ġsemiconducting": 36904,
+      "rative": 36905,
+      "ecology": 36906,
+      "urization": 36907,
+      "Ġmultifactorial": 36908,
+      "shadow": 36909,
+      "Ġcrosslinked": 36910,
+      "Ġphyla": 36911,
+      "Ġpremises": 36912,
+      "ĠLOW": 36913,
+      "generalized": 36914,
+      "ĠPolynomials": 36915,
+      "Ġbismuth": 36916,
+      "ĠRoz": 36917,
+      "ĠDecoding": 36918,
+      "ĠClassifier": 36919,
+      "conducting": 36920,
+      "Ġlitterm": 36921,
+      "Mann": 36922,
+      "Ġfant": 36923,
+      "ĠCZ": 36924,
+      "ĠPSNR": 36925,
+      "Ġstarring": 36926,
+      "ĠPolyg": 36927,
+      "ĠHolm": 36928,
+      "rg": 36929,
+      "additional": 36930,
+      "guan": 36931,
+      "professional": 36932,
+      "Ġinquiry": 36933,
+      "ĠPg": 36934,
+      "ĠSchmid": 36935,
+      "Ġheaded": 36936,
+      "chaft": 36937,
+      "ĠExpand": 36938,
+      "Ġcompanions": 36939,
+      "Van": 36940,
+      "ĠSie": 36941,
+      "Ġcanals": 36942,
+      "oredoxin": 36943,
+      "Ġcolliding": 36944,
+      "absolute": 36945,
+      "ĠPhotos": 36946,
+      "ĠLegacy": 36947,
+      "Ġrevascularization": 36948,
+      "ĠPSM": 36949,
+      "Ġexpenses": 36950,
+      "ISMA": 36951,
+      "intervals": 36952,
+      "Ġmulticellular": 36953,
+      "Ġnonsm": 36954,
+      "Ġresemblance": 36955,
+      "Hep": 36956,
+      "Ġwool": 36957,
+      "Ġniger": 36958,
+      "essa": 36959,
+      "asci": 36960,
+      "Ġrotates": 36961,
+      "Ġcompetitions": 36962,
+      "Ġarrivals": 36963,
+      "Ġlutein": 36964,
+      "Ġscholarship": 36965,
+      "Fran": 36966,
+      "Ġreused": 36967,
+      "ĠEquivalence": 36968,
+      "ĠGLUT": 36969,
+      "grading": 36970,
+      "salt": 36971,
+      "Ġcommensal": 36972,
+      "Ġfraud": 36973,
+      "oxib": 36974,
+      "Ġgastroenter": 36975,
+      "Ġrainy": 36976,
+      "Ġasserts": 36977,
+      "Operation": 36978,
+      "Ġflattening": 36979,
+      "Put": 36980,
+      "XB": 36981,
+      "ĠpM": 36982,
+      "Ġconic": 36983,
+      "obtain": 36984,
+      "ĠRober": 36985,
+      "November": 36986,
+      "ĠJP": 36987,
+      "Ġfebrile": 36988,
+      "ĠBarriers": 36989,
+      "================================================================": 36990,
+      "Ġhemicell": 36991,
+      "ĠSCS": 36992,
+      "ĠNem": 36993,
+      "Ġraster": 36994,
+      "clude": 36995,
+      "Ġïģ¦": 36996,
+      "ĠElliott": 36997,
+      "border": 36998,
+      "ĠdÏĨ": 36999,
+      "ribose": 37000,
+      "ĠEnv": 37001,
+      "ĠDiffuse": 37002,
+      "ĠSupersymmetry": 37003,
+      "Pearson": 37004,
+      "FETs": 37005,
+      "yah": 37006,
+      "ulia": 37007,
+      "ĠDwarf": 37008,
+      "ĠHull": 37009,
+      "ĠAttribution": 37010,
+      "Ġrepositories": 37011,
+      "ĠGNSS": 37012,
+      "ĠVectors": 37013,
+      "Ġsuccesses": 37014,
+      "ĠManhattan": 37015,
+      "umbent": 37016,
+      "digit": 37017,
+      "Ġcircumferential": 37018,
+      "Between": 37019,
+      "Deg": 37020,
+      "oue": 37021,
+      "й": 37022,
+      "ĠDere": 37023,
+      "ĠRf": 37024,
+      "Ġride": 37025,
+      "ĠVoc": 37026,
+      "Ġprotest": 37027,
+      "Ġpurpos": 37028,
+      "ĠProofs": 37029,
+      "namese": 37030,
+      "Ġbanking": 37031,
+      "ĠGastrointestinal": 37032,
+      "ĠUnt": 37033,
+      "Ġwhence": 37034,
+      "ĠYue": 37035,
+      "ĠRehabilitation": 37036,
+      "Ġexchanging": 37037,
+      "ĠACTH": 37038,
+      "Ġcapping": 37039,
+      "amido": 37040,
+      "ĠBap": 37041,
+      "Ġplat": 37042,
+      "toString": 37043,
+      "Ġelectroencephal": 37044,
+      "Ġelectrospun": 37045,
+      "Mpc": 37046,
+      "jord": 37047,
+      "onv": 37048,
+      "Ġcraft": 37049,
+      "ĠCCl": 37050,
+      "ĠStrip": 37051,
+      "Ġmeditation": 37052,
+      "oxidative": 37053,
+      "ĠReduce": 37054,
+      "ĠCommonwealth": 37055,
+      "Ġrifamp": 37056,
+      "Flu": 37057,
+      "Ġreanalysis": 37058,
+      "otrich": 37059,
+      "ĠESA": 37060,
+      "Ġjth": 37061,
+      "helin": 37062,
+      "ĠGenotype": 37063,
+      "Ġdiagonalization": 37064,
+      "ĠGabriel": 37065,
+      "Ġquarantine": 37066,
+      "ĠCrab": 37067,
+      "ĠDict": 37068,
+      "accumulation": 37069,
+      "bek": 37070,
+      "ĠDifferentially": 37071,
+      "Ġlactis": 37072,
+      "tetrahydrofuran": 37073,
+      "laser": 37074,
+      "ĠUm": 37075,
+      "Ġmega": 37076,
+      "rme": 37077,
+      "ĠIndians": 37078,
+      "ĠLeonard": 37079,
+      "Ġcommodity": 37080,
+      "Ġfumigatus": 37081,
+      "iou": 37082,
+      "ĠEchin": 37083,
+      "ostream": 37084,
+      "Ġmembran": 37085,
+      "simulations": 37086,
+      "backend": 37087,
+      "ĠOBJECT": 37088,
+      "giving": 37089,
+      "ÅĻ": 37090,
+      "Ġinfective": 37091,
+      "Alg": 37092,
+      "ĠHuh": 37093,
+      "ĠMICR": 37094,
+      "Ġfollowers": 37095,
+      "ferro": 37096,
+      "Ġcyanide": 37097,
+      "Present": 37098,
+      "ĠEND": 37099,
+      "ĠMCs": 37100,
+      "Ġtimeline": 37101,
+      "ĠEmbryonic": 37102,
+      "Identifier": 37103,
+      "Ġinconclusive": 37104,
+      "ĠGammaproteobacteria": 37105,
+      "nets": 37106,
+      "ĠHeating": 37107,
+      "ankar": 37108,
+      "thr": 37109,
+      "ĠKIT": 37110,
+      "ĠChip": 37111,
+      "Ġblob": 37112,
+      "Ġcalculator": 37113,
+      "Ġtextural": 37114,
+      "Ġalloying": 37115,
+      "Application": 37116,
+      "ĠProteomic": 37117,
+      "Ġantidepressants": 37118,
+      "urk": 37119,
+      "Ġcrystallography": 37120,
+      "Ġcredits": 37121,
+      "Ġmussels": 37122,
+      "Tom": 37123,
+      "ĠFST": 37124,
+      "ĠFold": 37125,
+      "ĠHew": 37126,
+      "Ann": 37127,
+      "brook": 37128,
+      "Ġglycolytic": 37129,
+      "Torch": 37130,
+      "Ġvm": 37131,
+      "ĠMare": 37132,
+      "ĠJy": 37133,
+      "Ġheterojunction": 37134,
+      "ĠBorrelia": 37135,
+      "Risk": 37136,
+      "ĠNaturally": 37137,
+      "Ġsupplying": 37138,
+      "signature": 37139,
+      "lk": 37140,
+      "Ġarachid": 37141,
+      "olov": 37142,
+      "ĠSok": 37143,
+      "ĠHö": 37144,
+      "ĠRaz": 37145,
+      "ĠVander": 37146,
+      "Ġdelib": 37147,
+      "Ġmyth": 37148,
+      "Ġmidbrain": 37149,
+      "Ġdeceased": 37150,
+      "ĠSCO": 37151,
+      "ĠThromb": 37152,
+      "Ġcurr": 37153,
+      "Ġsummit": 37154,
+      "miRNAs": 37155,
+      "dimethylamino": 37156,
+      "Ġphotocatalyst": 37157,
+      "verbose": 37158,
+      "gomery": 37159,
+      "Ġwed": 37160,
+      "ĠMate": 37161,
+      "Ġsigni": 37162,
+      "rastructures": 37163,
+      "Ġreciprocity": 37164,
+      "bner": 37165,
+      "mast": 37166,
+      "neck": 37167,
+      "Ġcoins": 37168,
+      "ĠHistogram": 37169,
+      "crit": 37170,
+      "Bbbk": 37171,
+      "AW": 37172,
+      "town": 37173,
+      "displacement": 37174,
+      "ĠNeph": 37175,
+      "separable": 37176,
+      "Ġdiastere": 37177,
+      "ĠMODELS": 37178,
+      "Depth": 37179,
+      "ĠNeisseria": 37180,
+      "pdev": 37181,
+      "uvial": 37182,
+      "ĠBMS": 37183,
+      "ĠDennis": 37184,
+      "Ġrp": 37185,
+      "Ġnanometer": 37186,
+      "rocyt": 37187,
+      "ĠRomanian": 37188,
+      "Ġconceivable": 37189,
+      "COS": 37190,
+      "alveolar": 37191,
+      "astig": 37192,
+      "abwe": 37193,
+      "encode": 37194,
+      "rolactone": 37195,
+      "Ġreadmission": 37196,
+      "intersection": 37197,
+      "Ġamplicons": 37198,
+      "timulated": 37199,
+      "Ġcollapses": 37200,
+      "ochromatin": 37201,
+      "Haw": 37202,
+      "ectrum": 37203,
+      "ftype": 37204,
+      "rica": 37205,
+      "Ġamid": 37206,
+      "MPO": 37207,
+      "ĠExtensions": 37208,
+      "Ġvaric": 37209,
+      "Ġdiminishes": 37210,
+      "Ġcatheters": 37211,
+      "Nodes": 37212,
+      "Ġbbox": 37213,
+      "emination": 37214,
+      "Ġtsunami": 37215,
+      "diagnosis": 37216,
+      "cod": 37217,
+      "qr": 37218,
+      "ĠFen": 37219,
+      "Ġworthy": 37220,
+      "ĠâĩIJ": 37221,
+      "informatic": 37222,
+      "ographer": 37223,
+      "Ġundetected": 37224,
+      "ĠNCAA": 37225,
+      "Ġcarcinogenic": 37226,
+      "RU": 37227,
+      "Ġaneu": 37228,
+      "plitudes": 37229,
+      "keeper": 37230,
+      "ĠÄģ": 37231,
+      "Ġautistic": 37232,
+      "Ġcompromising": 37233,
+      "Ġunimodal": 37234,
+      "Ġrumin": 37235,
+      "apa": 37236,
+      "Ġintolerance": 37237,
+      "Ġdirecting": 37238,
+      "Ġpea": 37239,
+      "Ġcommenced": 37240,
+      "Ġshadowing": 37241,
+      "Center": 37242,
+      "Ġclad": 37243,
+      "Ġblues": 37244,
+      "binits": 37245,
+      "Ġmisclassification": 37246,
+      "ĠFAST": 37247,
+      "Wat": 37248,
+      "ĠmCherry": 37249,
+      "Ġbrig": 37250,
+      "estradiol": 37251,
+      "Ġwavefunctions": 37252,
+      "Ġblooms": 37253,
+      "Ġaccent": 37254,
+      "aji": 37255,
+      "occurring": 37256,
+      "arrest": 37257,
+      "Ġspecialty": 37258,
+      "Ġunconditional": 37259,
+      "Ġsponges": 37260,
+      "Ġdysfunctional": 37261,
+      "ĠNOX": 37262,
+      "Ġultracold": 37263,
+      "Ġmartensite": 37264,
+      "OUS": 37265,
+      "nier": 37266,
+      "isic": 37267,
+      "ĠMatsum": 37268,
+      "Ġleukemic": 37269,
+      "ĠBradley": 37270,
+      "Density": 37271,
+      "ĠSemiconductor": 37272,
+      "ĠCause": 37273,
+      "ĠInset": 37274,
+      "ĠKem": 37275,
+      "ĠUPR": 37276,
+      "para": 37277,
+      "echst": 37278,
+      "ymet": 37279,
+      "Ġagro": 37280,
+      "ĠYY": 37281,
+      "ĠRegeneration": 37282,
+      "Ġancestors": 37283,
+      "ĠTissues": 37284,
+      "Ġsulfuric": 37285,
+      "kd": 37286,
+      "Ġlasing": 37287,
+      "ĠPup": 37288,
+      "aei": 37289,
+      "Ġmammal": 37290,
+      "ĠBradford": 37291,
+      "Ġsegregated": 37292,
+      "isolated": 37293,
+      "ĠCuba": 37294,
+      "Ġblockage": 37295,
+      "Ġseamless": 37296,
+      "Ġperoxisome": 37297,
+      "hui": 37298,
+      "Ġinaug": 37299,
+      "Ġinfecting": 37300,
+      "ĠChampion": 37301,
+      "ĠAttitudes": 37302,
+      "calculate": 37303,
+      "Ġtighter": 37304,
+      "ĠSAC": 37305,
+      "ĠEpi": 37306,
+      "Ġatm": 37307,
+      "Ġphysico": 37308,
+      "Ġnth": 37309,
+      "ĠCanyon": 37310,
+      "Ġseroprevalence": 37311,
+      "Ġhomo": 37312,
+      "ĠUniversit": 37313,
+      "Evaluation": 37314,
+      "ĠAPOE": 37315,
+      "job": 37316,
+      "ĠmK": 37317,
+      "Ġreign": 37318,
+      "abo": 37319,
+      "ĠRugby": 37320,
+      "ĠNets": 37321,
+      "Ġrituximab": 37322,
+      "ativeness": 37323,
+      "Ġphy": 37324,
+      "ornis": 37325,
+      "Ġfeedbacks": 37326,
+      "United": 37327,
+      "Princ": 37328,
+      "imbabwe": 37329,
+      "ĠGirls": 37330,
+      "Ġunavoidable": 37331,
+      "ĠSemantics": 37332,
+      "Break": 37333,
+      "FISH": 37334,
+      "Mix": 37335,
+      "Ġnx": 37336,
+      "ĠBao": 37337,
+      "dimethylphenyl": 37338,
+      "ĠTOF": 37339,
+      "ĠCrown": 37340,
+      "ĠGGA": 37341,
+      "ĠJH": 37342,
+      "Ġsuperstring": 37343,
+      "ĠCRY": 37344,
+      "Ġkindly": 37345,
+      "YN": 37346,
+      "Ġundoped": 37347,
+      "excluding": 37348,
+      "ĠLeo": 37349,
+      "ĠPROPERT": 37350,
+      "peritoneally": 37351,
+      "mant": 37352,
+      "ê°": 37353,
+      "Ġfranch": 37354,
+      "ĠProst": 37355,
+      "DEs": 37356,
+      "Ġcotrans": 37357,
+      "Ġrk": 37358,
+      "Ġgeneralizability": 37359,
+      "Author": 37360,
+      "ĠAndrea": 37361,
+      "ĠConfocal": 37362,
+      "ĠAdipose": 37363,
+      "îĹ": 37364,
+      "erjee": 37365,
+      "Ġanimated": 37366,
+      "ĠFad": 37367,
+      "ĠCorrosion": 37368,
+      "ĠCircadian": 37369,
+      "Ġaccelerators": 37370,
+      "ĠArkansas": 37371,
+      "Ġmars": 37372,
+      "ĠCuc": 37373,
+      "ĠInterfaces": 37374,
+      "Ġretrievals": 37375,
+      "Ġmelanin": 37376,
+      "ĠssDNA": 37377,
+      "vastava": 37378,
+      "Ġallergens": 37379,
+      "bud": 37380,
+      "Ġinaccessible": 37381,
+      "ictions": 37382,
+      "ĠMood": 37383,
+      "inda": 37384,
+      "Ġameric": 37385,
+      "Ġsymbiosis": 37386,
+      "bersome": 37387,
+      "occur": 37388,
+      "ĠMarcus": 37389,
+      "ĠSuperconductivity": 37390,
+      "ĠCort": 37391,
+      "ĠHMS": 37392,
+      "Ġphased": 37393,
+      "ĠJess": 37394,
+      "Ġpropulsion": 37395,
+      "extract": 37396,
+      "Ġsuccinate": 37397,
+      "ĠÖĴ": 37398,
+      "inkel": 37399,
+      "Ġsilence": 37400,
+      "ĠSUV": 37401,
+      "Ġconstituency": 37402,
+      "Ġbacteriophage": 37403,
+      "gem": 37404,
+      "ĠMCL": 37405,
+      "orene": 37406,
+      "ĠGoss": 37407,
+      "ICD": 37408,
+      "Ġglutamic": 37409,
+      "Ġcoexisting": 37410,
+      "STEMS": 37411,
+      "opotential": 37412,
+      "ĠEy": 37413,
+      "ĠLecture": 37414,
+      "ellae": 37415,
+      "Ġimmunoprec": 37416,
+      "Ġtimber": 37417,
+      "ĠVulner": 37418,
+      "Ġaroma": 37419,
+      "Ġsands": 37420,
+      "ĠSpan": 37421,
+      "Ġhern": 37422,
+      "Ġincubating": 37423,
+      "Ġtransmitters": 37424,
+      "ĠHomogeneous": 37425,
+      "ĠConstructing": 37426,
+      "dit": 37427,
+      "Ġtc": 37428,
+      "alass": 37429,
+      "Ġstents": 37430,
+      "ĠMID": 37431,
+      "Ġanoxic": 37432,
+      "Ġprovisions": 37433,
+      "ĠCapac": 37434,
+      "neutron": 37435,
+      "ĠVOCs": 37436,
+      "January": 37437,
+      "VAS": 37438,
+      "once": 37439,
+      "ĠCache": 37440,
+      "opulation": 37441,
+      "ĠVTE": 37442,
+      "Ġinterphase": 37443,
+      "Ġblog": 37444,
+      "ocusing": 37445,
+      "hiro": 37446,
+      "ĠREC": 37447,
+      "Ġanisotropies": 37448,
+      "benef": 37449,
+      "Ġconstipation": 37450,
+      "ĠCanal": 37451,
+      "Ġportrait": 37452,
+      "silyl": 37453,
+      "ĠLinked": 37454,
+      "ĠBowl": 37455,
+      "Ġmonopoles": 37456,
+      "ĠPerez": 37457,
+      "WIN": 37458,
+      "ĠTAP": 37459,
+      "Ġruthenium": 37460,
+      "ĠAdherence": 37461,
+      "ĠEnzymatic": 37462,
+      "Ġspecificities": 37463,
+      "Ġski": 37464,
+      "ĠCST": 37465,
+      "Ġpoetry": 37466,
+      "ATES": 37467,
+      "rama": 37468,
+      "lores": 37469,
+      "ALU": 37470,
+      "Ġvasoconstr": 37471,
+      "Ġgranulocyte": 37472,
+      "ibi": 37473,
+      "Ġopts": 37474,
+      "avesdrop": 37475,
+      "eptin": 37476,
+      "··": 37477,
+      "ĠJeong": 37478,
+      "Ġmedullary": 37479,
+      "ĠDemonstration": 37480,
+      "ĠFIB": 37481,
+      "ĠBRD": 37482,
+      "ĠVV": 37483,
+      "Ġallo": 37484,
+      "Rule": 37485,
+      "Tf": 37486,
+      "Ġunrealistic": 37487,
+      "Ġlatitudinal": 37488,
+      "ROP": 37489,
+      "ĠCorrelates": 37490,
+      "IU": 37491,
+      "ĠPore": 37492,
+      "ocrit": 37493,
+      "ĠKall": 37494,
+      "Ġcharcoal": 37495,
+      "ĠMongolia": 37496,
+      "âĪħ": 37497,
+      "ĠEntity": 37498,
+      "Ġgrams": 37499,
+      "graphene": 37500,
+      "mine": 37501,
+      "entric": 37502,
+      "ĠPp": 37503,
+      "ĠWelfare": 37504,
+      "ĠJets": 37505,
+      "Ġaffirm": 37506,
+      "ĠBelle": 37507,
+      "ĠStrategic": 37508,
+      "APIENTR": 37509,
+      "KH": 37510,
+      "rmann": 37511,
+      "Ġassociating": 37512,
+      "ĠSurviv": 37513,
+      "Ġnicotinic": 37514,
+      "ĠWLAN": 37515,
+      "п": 37516,
+      "Ġtears": 37517,
+      "ĠRevised": 37518,
+      "Ġphosphodies": 37519,
+      "Ġhorseradish": 37520,
+      "ĠLAR": 37521,
+      "took": 37522,
+      "ĠDescent": 37523,
+      "ĠNOx": 37524,
+      "ĠSteiner": 37525,
+      "ĠPermian": 37526,
+      "ĠVenezuela": 37527,
+      "Ġdesiccation": 37528,
+      "DIS": 37529,
+      "ĠMSP": 37530,
+      "Ġpopl": 37531,
+      "rels": 37532,
+      "Ġ": 37533,
+      "Ġlearnt": 37534,
+      "ĠBiofilm": 37535,
+      "ĠPCNA": 37536,
+      "ĠAttribute": 37537,
+      "ĠGrothendieck": 37538,
+      "ĠAdolescent": 37539,
+      "nv": 37540,
+      "stderr": 37541,
+      "obalt": 37542,
+      "ĠYamamoto": 37543,
+      "Ġaliquot": 37544,
+      "rater": 37545,
+      "ĠOre": 37546,
+      "ĠKIR": 37547,
+      "acker": 37548,
+      "Ġïĥ»": 37549,
+      "Ġstratosphere": 37550,
+      "ĠCust": 37551,
+      "respect": 37552,
+      "Ġglutamatergic": 37553,
+      "Ġencourages": 37554,
+      "ctic": 37555,
+      "itched": 37556,
+      "phins": 37557,
+      "Ġsuburb": 37558,
+      "Ġhomeomorphic": 37559,
+      "hexah": 37560,
+      "Ġminiatur": 37561,
+      "CAN": 37562,
+      "ahead": 37563,
+      "ĠBLE": 37564,
+      "ĠRBF": 37565,
+      "Ġacutely": 37566,
+      "Ġ": 37567,
+      "Ġantenn": 37568,
+      "URN": 37569,
+      "ĠGirl": 37570,
+      "Ġbioreactor": 37571,
+      "ĠLeibniz": 37572,
+      "Ġvial": 37573,
+      "ĠLich": 37574,
+      "obac": 37575,
+      "ĠWhenever": 37576,
+      "inhibition": 37577,
+      "Cast": 37578,
+      "Ġstripped": 37579,
+      "ĠAstrophysics": 37580,
+      "presence": 37581,
+      "ĠFloer": 37582,
+      "ipotent": 37583,
+      "dichloro": 37584,
+      "CLE": 37585,
+      "finger": 37586,
+      "onates": 37587,
+      "stri": 37588,
+      "ĠSperm": 37589,
+      "ĠDBS": 37590,
+      "opeptide": 37591,
+      "separation": 37592,
+      "athing": 37593,
+      "mathp": 37594,
+      "ouples": 37595,
+      "Ġentropic": 37596,
+      "Ġswollen": 37597,
+      "Ġdonated": 37598,
+      "Ġsettlements": 37599,
+      "ovenous": 37600,
+      "Perm": 37601,
+      "ĠSard": 37602,
+      "egen": 37603,
+      "ĠAlph": 37604,
+      "ĠCooperation": 37605,
+      "ĠPDAC": 37606,
+      "Final": 37607,
+      "lapse": 37608,
+      "Ġrevol": 37609,
+      "ĠIx": 37610,
+      "ĠLens": 37611,
+      "Ġkth": 37612,
+      "relaxation": 37613,
+      "ClO": 37614,
+      "ichloro": 37615,
+      "Ġwrapper": 37616,
+      "ĠSimultaneously": 37617,
+      "Compute": 37618,
+      "ëĬ": 37619,
+      "implantation": 37620,
+      "ĠVLA": 37621,
+      "heme": 37622,
+      "ĠMayor": 37623,
+      "ĠFacilit": 37624,
+      "Ġbatt": 37625,
+      "immer": 37626,
+      "Ġcurated": 37627,
+      "Ġconfluent": 37628,
+      "generational": 37629,
+      "starts": 37630,
+      "Ġgranulosa": 37631,
+      "arboxylate": 37632,
+      "ĠRiesz": 37633,
+      "Ġtextbook": 37634,
+      "Ġconstitutional": 37635,
+      "ĠPeace": 37636,
+      "ĠCommander": 37637,
+      "Ġobscured": 37638,
+      "vil": 37639,
+      "addition": 37640,
+      "ĠWasserstein": 37641,
+      "coords": 37642,
+      "ĠProbes": 37643,
+      "Ġdelineated": 37644,
+      "TZVP": 37645,
+      "ĠINF": 37646,
+      "Ġdosages": 37647,
+      "Ġoligomerization": 37648,
+      "ĠNADP": 37649,
+      "MKII": 37650,
+      "omin": 37651,
+      "Ġlhs": 37652,
+      "ughen": 37653,
+      "ĠJong": 37654,
+      "ancel": 37655,
+      "letter": 37656,
+      "ĠANC": 37657,
+      "FUNCTION": 37658,
+      "Ġtram": 37659,
+      "Their": 37660,
+      "ĠGenerated": 37661,
+      "Ġpolycyclic": 37662,
+      "Ġculmin": 37663,
+      "Ġrectum": 37664,
+      "Ġceft": 37665,
+      "Ġmetamaterials": 37666,
+      "ĠBiotech": 37667,
+      "Ġmyself": 37668,
+      "Ġunifying": 37669,
+      "Ġeman": 37670,
+      "ĠSinger": 37671,
+      "triangleright": 37672,
+      "omel": 37673,
+      "ĠCFA": 37674,
+      "ocha": 37675,
+      "ĠGSM": 37676,
+      "Ġcentrifuge": 37677,
+      "ĠIndo": 37678,
+      "Ġtransporting": 37679,
+      "LIB": 37680,
+      "Ġoxalate": 37681,
+      "ĠDulbecco": 37682,
+      "Ġali": 37683,
+      "arginal": 37684,
+      "hoo": 37685,
+      "ischem": 37686,
+      "APIENTRYP": 37687,
+      "Apart": 37688,
+      "LDA": 37689,
+      "ensile": 37690,
+      "settings": 37691,
+      "Ġephem": 37692,
+      "ampa": 37693,
+      "Ġduplications": 37694,
+      "ĠWheeler": 37695,
+      "Physical": 37696,
+      "ĠCompletion": 37697,
+      "ĠOrdered": 37698,
+      "Logger": 37699,
+      "Ġinterferences": 37700,
+      "ĠPollution": 37701,
+      "Optimal": 37702,
+      "Sv": 37703,
+      "aicin": 37704,
+      "Ġpicks": 37705,
+      "diversity": 37706,
+      "tigens": 37707,
+      "Ġdimorphism": 37708,
+      "feres": 37709,
+      "ĠRobotic": 37710,
+      "Ġconfirmatory": 37711,
+      "Ġcathodic": 37712,
+      "Ġspirals": 37713,
+      "Ġspruce": 37714,
+      "Lagrange": 37715,
+      "wat": 37716,
+      "ĠAllan": 37717,
+      "denote": 37718,
+      "CID": 37719,
+      "always": 37720,
+      "ithe": 37721,
+      "ĠChim": 37722,
+      "conditional": 37723,
+      "barrier": 37724,
+      "Ġvisualizing": 37725,
+      "Ġïĥ¹": 37726,
+      "Schmidt": 37727,
+      "Ġconventionally": 37728,
+      "ĠQUANT": 37729,
+      "GROUND": 37730,
+      "Ġug": 37731,
+      "ĠCWE": 37732,
+      "ĠInspired": 37733,
+      "Ġbuyer": 37734,
+      "Ġthermost": 37735,
+      "Ġkinematical": 37736,
+      "anolic": 37737,
+      "Ġdif": 37738,
+      "Ġ": 37739,
+      "ĠGeo": 37740,
+      "Examples": 37741,
+      "consistency": 37742,
+      "ĠPalace": 37743,
+      "ĠVaccination": 37744,
+      "Ġnatriuretic": 37745,
+      "YAG": 37746,
+      "ĠCTCs": 37747,
+      "Univers": 37748,
+      "ĠAcknowledgement": 37749,
+      "membered": 37750,
+      "vv": 37751,
+      "ĠSession": 37752,
+      "Ġinstar": 37753,
+      "ĠLevin": 37754,
+      "AVI": 37755,
+      "Ġproliferator": 37756,
+      "oliths": 37757,
+      "ĠTemperatures": 37758,
+      "imming": 37759,
+      "ĠToeplitz": 37760,
+      "ICATIONS": 37761,
+      "ĠIntegrals": 37762,
+      "Ġspliced": 37763,
+      "Dest": 37764,
+      "resulting": 37765,
+      "ĠHope": 37766,
+      "Ġenclosure": 37767,
+      "ieves": 37768,
+      "flav": 37769,
+      "ĠAbdul": 37770,
+      "Ġleishmaniasis": 37771,
+      "ò": 37772,
+      "oskeleton": 37773,
+      "Ġadduct": 37774,
+      "ĠInfluences": 37775,
+      "EQU": 37776,
+      "ĠSitu": 37777,
+      "Ġseas": 37778,
+      "ĠReich": 37779,
+      "cyst": 37780,
+      "ĠEVOLUTION": 37781,
+      "Ġwithstand": 37782,
+      "ĠGinzburg": 37783,
+      "RNAi": 37784,
+      "ĠNonparametric": 37785,
+      "ĠPrincess": 37786,
+      "Ġintravascular": 37787,
+      "UTIONS": 37788,
+      "Ġglutar": 37789,
+      "Ġcoincided": 37790,
+      "ĠSaito": 37791,
+      "pretrained": 37792,
+      "combined": 37793,
+      "ĠTAM": 37794,
+      "Ġalarms": 37795,
+      "Ġcyclooxygenase": 37796,
+      "Ġbn": 37797,
+      "Ġplagi": 37798,
+      "Particle": 37799,
+      "GGG": 37800,
+      "etics": 37801,
+      "amber": 37802,
+      "ABSTRACT": 37803,
+      "ĠExtracts": 37804,
+      "ĉĉĉĠĠĠĠ": 37805,
+      "ĠPhylogeny": 37806,
+      "tow": 37807,
+      "ĠContaining": 37808,
+      "Ġendonuclease": 37809,
+      "incubation": 37810,
+      "Ġofficinal": 37811,
+      "Ġexplosions": 37812,
+      "layout": 37813,
+      "Ġtouchdown": 37814,
+      "ĠRevealed": 37815,
+      "Ġinfiltrate": 37816,
+      "enith": 37817,
+      "timulation": 37818,
+      "ĠKind": 37819,
+      "ervices": 37820,
+      "PDA": 37821,
+      "Ġcereus": 37822,
+      "Env": 37823,
+      "Ġlapa": 37824,
+      "kamp": 37825,
+      "mult": 37826,
+      "enthal": 37827,
+      "ĠGoldstone": 37828,
+      "siRNA": 37829,
+      "strept": 37830,
+      "Qual": 37831,
+      "mother": 37832,
+      "dio": 37833,
+      "Ġinfrequent": 37834,
+      "Ġcyclospor": 37835,
+      "hepatitis": 37836,
+      "thrombotic": 37837,
+      "GST": 37838,
+      "ĠLj": 37839,
+      "ĠUR": 37840,
+      "ofect": 37841,
+      "ĠArrow": 37842,
+      "ethnic": 37843,
+      "ĠBarcelona": 37844,
+      "Care": 37845,
+      "titious": 37846,
+      "Ġeta": 37847,
+      "Ġvirions": 37848,
+      "smash": 37849,
+      "ĠâIJ¤": 37850,
+      "Ġavenues": 37851,
+      "obarb": 37852,
+      "ĠComments": 37853,
+      "Ġanyway": 37854,
+      "afil": 37855,
+      "ĠBea": 37856,
+      "ĠBoys": 37857,
+      "ĠAutomata": 37858,
+      "ĠSuperconducting": 37859,
+      "Pic": 37860,
+      "kHz": 37861,
+      "Ġnorepinephrine": 37862,
+      "ĠGPC": 37863,
+      "Ġunderlined": 37864,
+      "brahim": 37865,
+      "Ġelectrospray": 37866,
+      "Ġsesqu": 37867,
+      "ĠTournament": 37868,
+      "Austr": 37869,
+      "ĠGrowing": 37870,
+      "ĠWebsite": 37871,
+      "LDH": 37872,
+      "covariance": 37873,
+      "several": 37874,
+      "stabilized": 37875,
+      "Ġdecarboxylase": 37876,
+      "Ġremed": 37877,
+      "rhoe": 37878,
+      "ĠSRS": 37879,
+      "ĠTreated": 37880,
+      "ĠMadagascar": 37881,
+      "ĠMagic": 37882,
+      "Ġweapon": 37883,
+      "ĠYoshida": 37884,
+      "Ġhypoglycemia": 37885,
+      "ĠBifidobacterium": 37886,
+      "entitious": 37887,
+      ":::": 37888,
+      "ĠSingles": 37889,
+      "Ġnicely": 37890,
+      "Ġunexpectedly": 37891,
+      "ibles": 37892,
+      "ariae": 37893,
+      "Ġcentroids": 37894,
+      "Ġbroadened": 37895,
+      "ĠJohns": 37896,
+      "ĠBacteroid": 37897,
+      "Ġframing": 37898,
+      "Primary": 37899,
+      "ĠPicture": 37900,
+      "government": 37901,
+      "Ġreq": 37902,
+      "ĠTry": 37903,
+      "ibo": 37904,
+      "Ġliquef": 37905,
+      "osensitivity": 37906,
+      "Ġslaughter": 37907,
+      "ĠDAR": 37908,
+      "Ġlogit": 37909,
+      "Ġpromises": 37910,
+      "Ġlawyer": 37911,
+      "ĠFPG": 37912,
+      "TCP": 37913,
+      "Ġintercalation": 37914,
+      "ĠBoe": 37915,
+      "Ġwideband": 37916,
+      "Ġjudgement": 37917,
+      "romagnets": 37918,
+      "Lastly": 37919,
+      "ĠIschemic": 37920,
+      "IMA": 37921,
+      "food": 37922,
+      "much": 37923,
+      "Ġavenue": 37924,
+      "Ġschistosomiasis": 37925,
+      "ĠExecution": 37926,
+      "DQU": 37927,
+      "GIS": 37928,
+      "kines": 37929,
+      "akage": 37930,
+      "echt": 37931,
+      "ĠScaff": 37932,
+      "ĠStrings": 37933,
+      "ĠMultilevel": 37934,
+      "Ġcumbersome": 37935,
+      "ĠRaymond": 37936,
+      "Ġirregularities": 37937,
+      "ĠAGNs": 37938,
+      "ĠMetastatic": 37939,
+      "ĠIberian": 37940,
+      "Mb": 37941,
+      "RNP": 37942,
+      "hong": 37943,
+      "isinin": 37944,
+      "Ġthirteen": 37945,
+      "ĠFAS": 37946,
+      "Ġsealing": 37947,
+      "Ġapatite": 37948,
+      "Ġserially": 37949,
+      "ĠÅĿ": 37950,
+      "DEL": 37951,
+      "Fo": 37952,
+      "ĠSoph": 37953,
+      "ĠBear": 37954,
+      "ĠJosh": 37955,
+      "reck": 37956,
+      "uller": 37957,
+      "Ġexcursion": 37958,
+      "Ġembodied": 37959,
+      "Ġhybridized": 37960,
+      "ĠLieutenant": 37961,
+      "Period": 37962,
+      "Ġmollus": 37963,
+      "CVD": 37964,
+      "Ren": 37965,
+      "REAM": 37966,
+      "ĠBACK": 37967,
+      "Ġaccreting": 37968,
+      "Ġculturing": 37969,
+      "ĠBurst": 37970,
+      "ĠSegment": 37971,
+      "Ġasterisk": 37972,
+      "ĠIdeal": 37973,
+      "Ġintertw": 37974,
+      "ĠAtoms": 37975,
+      "ĠSTE": 37976,
+      "Ġïģª": 37977,
+      "Ġremarked": 37978,
+      "Ġhairs": 37979,
+      "âľ": 37980,
+      "ĠMetropolis": 37981,
+      "ĠPartially": 37982,
+      "ĠObserver": 37983,
+      "Ġhematologic": 37984,
+      "obilization": 37985,
+      "ĠBergman": 37986,
+      "Ġcartesian": 37987,
+      "Ġclathrin": 37988,
+      "ĠSung": 37989,
+      "Ġration": 37990,
+      "Ġscoliosis": 37991,
+      "ohl": 37992,
+      "mutant": 37993,
+      "NNs": 37994,
+      "ĠRahman": 37995,
+      "ĠSpatially": 37996,
+      "PIP": 37997,
+      "Yb": 37998,
+      "Ġdiaz": 37999,
+      "vertebral": 38000,
+      "adzu": 38001,
+      "alski": 38002,
+      "answer": 38003,
+      "Ġgeochemistry": 38004,
+      "Ġstemming": 38005,
+      "wes": 38006,
+      "oxys": 38007,
+      "Ġmats": 38008,
+      "eva": 38009,
+      "ĠHyperbolic": 38010,
+      "arbage": 38011,
+      "Ġclipping": 38012,
+      "ĠSugar": 38013,
+      "ĠCognition": 38014,
+      "ĠDIV": 38015,
+      "Ġtempt": 38016,
+      "ĠPathogen": 38017,
+      "ĠPedro": 38018,
+      "Ġwak": 38019,
+      "entries": 38020,
+      "ĠGCM": 38021,
+      "projective": 38022,
+      "Ġproficiency": 38023,
+      "ĠKnown": 38024,
+      "Ġlexicon": 38025,
+      "ĠMendelian": 38026,
+      "Ġzoonotic": 38027,
+      "leans": 38028,
+      "ĠTalk": 38029,
+      "Ġkurtosis": 38030,
+      "NAS": 38031,
+      "ĠNowadays": 38032,
+      "ĠLil": 38033,
+      "ĠWMAP": 38034,
+      "Ġdisperse": 38035,
+      "Ġcolloids": 38036,
+      "ebra": 38037,
+      "OMET": 38038,
+      "ĠDCT": 38039,
+      "ĠRise": 38040,
+      "Ġintergenic": 38041,
+      "GTH": 38042,
+      "Ġtapered": 38043,
+      "Markovian": 38044,
+      "Protocol": 38045,
+      "ĠVegetation": 38046,
+      "rats": 38047,
+      "Ġdivalent": 38048,
+      "ĠCrust": 38049,
+      "zyg": 38050,
+      "Ġpigmentation": 38051,
+      "graduate": 38052,
+      "ĠRicc": 38053,
+      "Ġcounterexample": 38054,
+      "Ġsativ": 38055,
+      "Ġls": 38056,
+      "ĠCirculation": 38057,
+      "isotropic": 38058,
+      "ĠENSO": 38059,
+      "Ġtroponin": 38060,
+      "Ġdissolving": 38061,
+      "Ġcosmetic": 38062,
+      "Hf": 38063,
+      "further": 38064,
+      "Ġpanc": 38065,
+      "Ġhops": 38066,
+      "intra": 38067,
+      "ĠZhe": 38068,
+      "ĠReliable": 38069,
+      "ivolumab": 38070,
+      "MX": 38071,
+      "Rab": 38072,
+      "ĠPES": 38073,
+      "ĠBü": 38074,
+      "Ġadhered": 38075,
+      "Ġfluency": 38076,
+      "ĠClaus": 38077,
+      "Ġdelamination": 38078,
+      "Ġguanine": 38079,
+      "ĠMultiscale": 38080,
+      "ĠEquip": 38081,
+      "ĠIllustr": 38082,
+      "Ġtetrahydro": 38083,
+      "fel": 38084,
+      "lists": 38085,
+      "Îŀ": 38086,
+      "emulsion": 38087,
+      "ĠNZ": 38088,
+      "Ġwasn": 38089,
+      "aira": 38090,
+      "Ġarguing": 38091,
+      "miRNA": 38092,
+      "ĠExpressed": 38093,
+      "Ġspectrophotometric": 38094,
+      "Ġileum": 38095,
+      "Ġflames": 38096,
+      "Fit": 38097,
+      "Gon": 38098,
+      "ĠCulex": 38099,
+      "Ġunweighted": 38100,
+      "Ġnanob": 38101,
+      "SHV": 38102,
+      "Ġaligning": 38103,
+      "Ġshuttle": 38104,
+      "Ġchloroquine": 38105,
+      "Ġpyrite": 38106,
+      "ĠRica": 38107,
+      "Ġrift": 38108,
+      "Ġcathepsin": 38109,
+      "ĠPROCESS": 38110,
+      "Pf": 38111,
+      "Raw": 38112,
+      "rayfish": 38113,
+      "SAL": 38114,
+      "collapse": 38115,
+      "................": 38116,
+      "atases": 38117,
+      "Ġworkshops": 38118,
+      "ophile": 38119,
+      "ĠâĬĥ": 38120,
+      "Ġbifurcations": 38121,
+      "Trace": 38122,
+      "Ġpause": 38123,
+      "Ġorbiting": 38124,
+      "oliubov": 38125,
+      "ĠCurtis": 38126,
+      "ĠRevisiting": 38127,
+      "oret": 38128,
+      "Ġinfused": 38129,
+      "luents": 38130,
+      "Ġplastid": 38131,
+      "Ġïģ¹": 38132,
+      "Ġexecutions": 38133,
+      "ĠGraves": 38134,
+      "locally": 38135,
+      "ĠAtmosphere": 38136,
+      "diabetes": 38137,
+      "ĠPradesh": 38138,
+      "ĠCofactor": 38139,
+      "isomorphic": 38140,
+      "Ġbod": 38141,
+      "ĠCBD": 38142,
+      "Ġincap": 38143,
+      "Ġretrovirus": 38144,
+      "Ġlipophilic": 38145,
+      "Ġlinoleic": 38146,
+      "Ġtravelled": 38147,
+      "covalent": 38148,
+      "pick": 38149,
+      "upl": 38150,
+      "ĠPole": 38151,
+      "ĠThym": 38152,
+      "ĠTeich": 38153,
+      "Ġcollaborators": 38154,
+      "Ġinstantons": 38155,
+      "ĠMEGA": 38156,
+      "ĠHepatocellular": 38157,
+      "Ġinfestation": 38158,
+      "ĠPiezo": 38159,
+      "ĠLub": 38160,
+      "ĠNCs": 38161,
+      "Ġnucleoside": 38162,
+      "Ġosteogenesis": 38163,
+      "Eigen": 38164,
+      "RMSE": 38165,
+      "Ġlax": 38166,
+      "ĠKost": 38167,
+      "ĠVero": 38168,
+      "ĠChou": 38169,
+      "electrochemical": 38170,
+      "Ġcompeti": 38171,
+      "chia": 38172,
+      "Ġsubmodule": 38173,
+      "ĠAllow": 38174,
+      "Ġresolvent": 38175,
+      "Ġsweeps": 38176,
+      "Ġsuperconformal": 38177,
+      "pyrrolidine": 38178,
+      "lofen": 38179,
+      "åŃ": 38180,
+      "Ġdeserves": 38181,
+      "ĠZimbabwe": 38182,
+      "azines": 38183,
+      "ĠConsult": 38184,
+      "Ġcastle": 38185,
+      "Ġpharmaceuticals": 38186,
+      "Ġparacrine": 38187,
+      "Ġjejuni": 38188,
+      "Ġarguably": 38189,
+      "ĠeNOS": 38190,
+      "Ġherds": 38191,
+      "Ġvehicular": 38192,
+      "Ġtriangulated": 38193,
+      "Ġîµ": 38194,
+      "ĠGrande": 38195,
+      "Ġanthocyanins": 38196,
+      "ĠDuan": 38197,
+      "ĠVibration": 38198,
+      "Ġtriad": 38199,
+      "Ġhousekeeping": 38200,
+      "Bor": 38201,
+      "Ġpub": 38202,
+      "Ġmalformation": 38203,
+      "glucosamine": 38204,
+      "inhibitory": 38205,
+      "Dirac": 38206,
+      "ĠCSD": 38207,
+      "ĠRotating": 38208,
+      "ĠHTLV": 38209,
+      "Ġdemol": 38210,
+      "infiltr": 38211,
+      "Ġhemolytic": 38212,
+      "Ġcarbapenem": 38213,
+      "Ġluminescent": 38214,
+      "ĠPlanets": 38215,
+      "Ġmellifera": 38216,
+      "Ġcorticosterone": 38217,
+      "ĠAddress": 38218,
+      "Ġhubs": 38219,
+      "omethacin": 38220,
+      "åIJ": 38221,
+      "ĠChampions": 38222,
+      "ĠRevision": 38223,
+      "ĠHerbert": 38224,
+      "Ġambiguities": 38225,
+      "KERN": 38226,
+      "Ġdé": 38227,
+      "Ġlp": 38228,
+      "Ġenvis": 38229,
+      "ĠChol": 38230,
+      "ropin": 38231,
+      "Ġdrone": 38232,
+      "meyer": 38233,
+      "Ġisotype": 38234,
+      "ĠVu": 38235,
+      "ERC": 38236,
+      "Ġversatility": 38237,
+      "Speed": 38238,
+      "Ġaetiology": 38239,
+      "Ġgonadotropin": 38240,
+      "Ġcognate": 38241,
+      "ĠCotton": 38242,
+      "reasonable": 38243,
+      "disable": 38244,
+      "Ġdevastating": 38245,
+      "Pier": 38246,
+      "POL": 38247,
+      "ĠBé": 38248,
+      "incter": 38249,
+      "aluable": 38250,
+      "Ġpolyhedron": 38251,
+      "ĠRelay": 38252,
+      "Ġworkflows": 38253,
+      "FEM": 38254,
+      "inp": 38255,
+      "Ġmph": 38256,
+      "softmax": 38257,
+      "mur": 38258,
+      "vr": 38259,
+      "Ġerent": 38260,
+      "ĠKN": 38261,
+      "Ġstatin": 38262,
+      "Ġflatness": 38263,
+      "ĠArchitectures": 38264,
+      "ĠVeterinary": 38265,
+      "Ġnosocomial": 38266,
+      "Sk": 38267,
+      "XML": 38268,
+      "ĠFos": 38269,
+      "ĠLor": 38270,
+      "Ġradiography": 38271,
+      "ĠBlum": 38272,
+      "ĠDiscrimination": 38273,
+      "Ġpunc": 38274,
+      "Ġexits": 38275,
+      "ĠBilateral": 38276,
+      "msstrahlung": 38277,
+      "Ġcolonized": 38278,
+      "ĠFibrosis": 38279,
+      "Ġchaperones": 38280,
+      "aboratory": 38281,
+      "ĠPersistence": 38282,
+      "Ġlumped": 38283,
+      "Ġrabies": 38284,
+      "ĠBurns": 38285,
+      "Dense": 38286,
+      "ontium": 38287,
+      "acetylation": 38288,
+      "ĠFET": 38289,
+      "Ġhandful": 38290,
+      "biology": 38291,
+      "Ġundesired": 38292,
+      "Limit": 38293,
+      "ĠNBA": 38294,
+      "ĠSeoul": 38295,
+      "APT": 38296,
+      "ĠTransgenic": 38297,
+      "oxygenation": 38298,
+      "Button": 38299,
+      "ĠTreatments": 38300,
+      "ZV": 38301,
+      "isomorphism": 38302,
+      "octa": 38303,
+      "iffe": 38304,
+      "odeoxy": 38305,
+      "Ġorganelle": 38306,
+      "Ġcolloid": 38307,
+      "Ġceramide": 38308,
+      "Ġtqdm": 38309,
+      "GPS": 38310,
+      "ĠISR": 38311,
+      "oclinic": 38312,
+      "ĠLyme": 38313,
+      "Ġepig": 38314,
+      "ĠTrail": 38315,
+      "IPS": 38316,
+      "Ġsorts": 38317,
+      "ĠZebrafish": 38318,
+      "Ġhydroxylase": 38319,
+      "Smirnov": 38320,
+      "Bax": 38321,
+      "ĠDance": 38322,
+      "ĠHors": 38323,
+      "Ġreachability": 38324,
+      "Parallel": 38325,
+      "ĠESBL": 38326,
+      "Ġuplink": 38327,
+      "Ġpostprandial": 38328,
+      "solar": 38329,
+      "itabine": 38330,
+      "ordism": 38331,
+      "Neasy": 38332,
+      "Ġabandon": 38333,
+      "IMI": 38334,
+      "fake": 38335,
+      "statistical": 38336,
+      "ĠCars": 38337,
+      "ibia": 38338,
+      "ĠÃĩ": 38339,
+      "spc": 38340,
+      "MDP": 38341,
+      "tizations": 38342,
+      "International": 38343,
+      "ularis": 38344,
+      "Ġvacuoles": 38345,
+      "KC": 38346,
+      "ĠAPT": 38347,
+      "ĠBt": 38348,
+      "ĠBom": 38349,
+      "ĠGMP": 38350,
+      "Ġpioneer": 38351,
+      "ĠChairman": 38352,
+      "ĠTucker": 38353,
+      "ĠRAF": 38354,
+      "ĠNASH": 38355,
+      "ĠWIT": 38356,
+      "ynyl": 38357,
+      "Ġsupplier": 38358,
+      "ansky": 38359,
+      "Ġdecomposing": 38360,
+      "ĠUVB": 38361,
+      "ophenol": 38362,
+      "Ġbarium": 38363,
+      "ĠSMT": 38364,
+      "otocin": 38365,
+      "lytic": 38366,
+      "ranking": 38367,
+      "ĠDirections": 38368,
+      "Ġinnervation": 38369,
+      "switching": 38370,
+      "dac": 38371,
+      "ĠhT": 38372,
+      "Ġdoctr": 38373,
+      "ĠIncremental": 38374,
+      "ĠEarthquake": 38375,
+      "Has": 38376,
+      "Lee": 38377,
+      "mates": 38378,
+      "proline": 38379,
+      "ĠREE": 38380,
+      "Ġviolates": 38381,
+      "ðx": 38382,
+      "Ġhomogenates": 38383,
+      "Boolean": 38384,
+      "Ġdoxycycline": 38385,
+      "ĠMOF": 38386,
+      "iophen": 38387,
+      "Ġappreciation": 38388,
+      "finals": 38389,
+      "characteristic": 38390,
+      "ĠContinental": 38391,
+      "Bus": 38392,
+      "Esc": 38393,
+      "XP": 38394,
+      "ÛĮ": 38395,
+      "ĠCTA": 38396,
+      "Maxwell": 38397,
+      "Ġarchaea": 38398,
+      "Nik": 38399,
+      "NONE": 38400,
+      "TW": 38401,
+      "tering": 38402,
+      "ĠPerman": 38403,
+      "Ġrestores": 38404,
+      "opathogenic": 38405,
+      "ĠMontgomery": 38406,
+      "Ġglucocorticoids": 38407,
+      "Ġud": 38408,
+      "ĠNuss": 38409,
+      "ĠNé": 38410,
+      "ĠSturm": 38411,
+      "Ġattaching": 38412,
+      "Ġintraperitoneally": 38413,
+      "lasov": 38414,
+      "Ġstellate": 38415,
+      "Ġantiproliferative": 38416,
+      "Ġmicroorganism": 38417,
+      "Ġvisu": 38418,
+      "Ġjudges": 38419,
+      "randomized": 38420,
+      "allowed": 38421,
+      "Ġdeprived": 38422,
+      "development": 38423,
+      "scribed": 38424,
+      "etherian": 38425,
+      "ĠFraser": 38426,
+      "Ram": 38427,
+      "bib": 38428,
+      "Ġliner": 38429,
+      "Ġguns": 38430,
+      "resnet": 38431,
+      "ĠLTR": 38432,
+      "ighting": 38433,
+      "Initi": 38434,
+      "ĠZimm": 38435,
+      "ĠGeology": 38436,
+      "Ġantioxidative": 38437,
+      "Ġmagenta": 38438,
+      "ĠNigerian": 38439,
+      "galaxy": 38440,
+      "ĠMelanoma": 38441,
+      "Found": 38442,
+      "Ġbum": 38443,
+      "ĠTrop": 38444,
+      "ĠDos": 38445,
+      "Ġmetab": 38446,
+      "Ġinvoking": 38447,
+      "ĠSchizophrenia": 38448,
+      "CFG": 38449,
+      "Ġgelation": 38450,
+      "Ġopioids": 38451,
+      "pis": 38452,
+      "Ġchurches": 38453,
+      "Ġcanonically": 38454,
+      "Ġjug": 38455,
+      "Ġacceptors": 38456,
+      "DMEM": 38457,
+      "Ġobliqu": 38458,
+      "ĠMedicare": 38459,
+      "arpoon": 38460,
+      "ZIP": 38461,
+      "oreactive": 38462,
+      "Ġimprinting": 38463,
+      "ĠVinc": 38464,
+      "Ġ¿": 38465,
+      "Ġrestart": 38466,
+      "Ġdentate": 38467,
+      "enzymatic": 38468,
+      "Ġinguinal": 38469,
+      "ĠNt": 38470,
+      "Ġunobserved": 38471,
+      "uctuation": 38472,
+      "Ġbiasing": 38473,
+      "Ġintegrins": 38474,
+      "Ġurl": 38475,
+      "FPGAM": 38476,
+      "ĠCLUST": 38477,
+      "omatology": 38478,
+      "Ġmetallicities": 38479,
+      "Ġintentionally": 38480,
+      "FPGAMGR": 38481,
+      "Typ": 38482,
+      "Ġally": 38483,
+      "Ġcomic": 38484,
+      "ĠLions": 38485,
+      "Ġimputed": 38486,
+      "ĠÃŁ": 38487,
+      "lexia": 38488,
+      "ĠJanus": 38489,
+      "Ġbrass": 38490,
+      "ĠDownloaded": 38491,
+      "BUFF": 38492,
+      "identical": 38493,
+      "Ġpsychiatry": 38494,
+      "CCT": 38495,
+      "ifar": 38496,
+      "ĠMandel": 38497,
+      "Ġoptoelectronic": 38498,
+      "Ġisomerization": 38499,
+      "ĠFant": 38500,
+      "ĠLion": 38501,
+      "ĠLov": 38502,
+      "ĠNaf": 38503,
+      "esta": 38504,
+      "Ġbiocompatible": 38505,
+      "Ġsecretions": 38506,
+      "sci": 38507,
+      "ĠRetro": 38508,
+      "oisomerase": 38509,
+      "ĠSnap": 38510,
+      "Ġsplittings": 38511,
+      "Ġscavenger": 38512,
+      "procedure": 38513,
+      "Dawley": 38514,
+      "ëĭ¤": 38515,
+      "unate": 38516,
+      "ĠDye": 38517,
+      "ĠNEC": 38518,
+      "Ġnanocl": 38519,
+      "Ġplanetes": 38520,
+      "ĠTRPM": 38521,
+      "Ġvoices": 38522,
+      "ĠHierarchy": 38523,
+      "mv": 38524,
+      "Ġlasts": 38525,
+      "Ġhoped": 38526,
+      "Ġmedians": 38527,
+      "ĠAndreev": 38528,
+      "Ġheightened": 38529,
+      "ä»": 38530,
+      "Ġindefinite": 38531,
+      "ĠKamp": 38532,
+      "angel": 38533,
+      "grids": 38534,
+      "archae": 38535,
+      "Ġtherapists": 38536,
+      "ĠMiR": 38537,
+      "Ġnegotiation": 38538,
+      "HSP": 38539,
+      "ĠCustom": 38540,
+      "Ġstria": 38541,
+      "Ġunacceptable": 38542,
+      "retin": 38543,
+      "penet": 38544,
+      "ĠORR": 38545,
+      "ĠLifetime": 38546,
+      "ĠPhosphate": 38547,
+      "Ġtropics": 38548,
+      "ĠWelch": 38549,
+      "ĠPyr": 38550,
+      "Ġamputation": 38551,
+      "ĠArtin": 38552,
+      "ĠCaO": 38553,
+      "Ġconjectures": 38554,
+      "Ġatrium": 38555,
+      "ĠComplementary": 38556,
+      "ĠAluminum": 38557,
+      "Ġmicrow": 38558,
+      "iliated": 38559,
+      "ĠImmuno": 38560,
+      "Ġbinocular": 38561,
+      "ĠWeakly": 38562,
+      "Ġimmunogenic": 38563,
+      "Ġbathym": 38564,
+      "ĠPhenotype": 38565,
+      "Ġsialic": 38566,
+      "Six": 38567,
+      "Ġakin": 38568,
+      "rotor": 38569,
+      "helm": 38570,
+      "CCESS": 38571,
+      "Ġneuroprotection": 38572,
+      "ĠFifth": 38573,
+      "Ġcontingent": 38574,
+      "Ġsketched": 38575,
+      "Imp": 38576,
+      "Ġcached": 38577,
+      "urement": 38578,
+      "ĠBic": 38579,
+      "ĠKah": 38580,
+      "beration": 38581,
+      "atterson": 38582,
+      "Ġglycation": 38583,
+      "Ġinvestors": 38584,
+      "Assisted": 38585,
+      "iales": 38586,
+      "science": 38587,
+      "Ġpilots": 38588,
+      "uscripts": 38589,
+      "MICS": 38590,
+      "Ġorthopedic": 38591,
+      "warfs": 38592,
+      "greater": 38593,
+      "ĠArtery": 38594,
+      "Video": 38595,
+      "Ġarrange": 38596,
+      "avar": 38597,
+      "charges": 38598,
+      "dialdehyde": 38599,
+      "ĠTPA": 38600,
+      "Ġspelling": 38601,
+      "ĠSeiberg": 38602,
+      "Ġnavigate": 38603,
+      "ĠPowder": 38604,
+      "ĠRings": 38605,
+      "ĠChron": 38606,
+      "ĠAtg": 38607,
+      "Ġhomocysteine": 38608,
+      "ĠIdentify": 38609,
+      "Ġoak": 38610,
+      "Ġliability": 38611,
+      "Ġoperands": 38612,
+      "ĠCTD": 38613,
+      "Ġalleviates": 38614,
+      "mA": 38615,
+      "ĠLanger": 38616,
+      "Ġsubmanifolds": 38617,
+      "ĠJag": 38618,
+      "Ġradiance": 38619,
+      "constants": 38620,
+      "ĠMorocco": 38621,
+      "Engine": 38622,
+      "á¸": 38623,
+      "âĤ¬": 38624,
+      "revers": 38625,
+      "PCI": 38626,
+      "unsqueeze": 38627,
+      "oconversion": 38628,
+      "Ġintensified": 38629,
+      "Ġrefinements": 38630,
+      "ofectamine": 38631,
+      "ayas": 38632,
+      "Ġincidental": 38633,
+      "ĠThur": 38634,
+      "Ġoverd": 38635,
+      "Ġbitter": 38636,
+      "Ġignores": 38637,
+      "ан": 38638,
+      "ĠOTU": 38639,
+      "Ġserr": 38640,
+      "aby": 38641,
+      "ĠGCN": 38642,
+      "ĠConsumer": 38643,
+      "Ġconcordant": 38644,
+      "ĠMRC": 38645,
+      "ĠEconomy": 38646,
+      "satisfying": 38647,
+      "Ġbiotinylated": 38648,
+      "Numerical": 38649,
+      "ĠRashba": 38650,
+      "stochastic": 38651,
+      "ĠLal": 38652,
+      "Ġburdens": 38653,
+      "Alloc": 38654,
+      "ĠGraphics": 38655,
+      "ĠLRRK": 38656,
+      "AIC": 38657,
+      "ĠTed": 38658,
+      "ĠSark": 38659,
+      "owl": 38660,
+      "Ġhemost": 38661,
+      "ĠAnat": 38662,
+      "Ġhoming": 38663,
+      "ĠCharlie": 38664,
+      "ĠBruc": 38665,
+      "ihara": 38666,
+      "ingen": 38667,
+      "ĠVern": 38668,
+      "ĠYers": 38669,
+      "Ġids": 38670,
+      "ĠcircRNAs": 38671,
+      "Ġconducive": 38672,
+      "ĠBRST": 38673,
+      "Ġgallium": 38674,
+      "Ġdichotomy": 38675,
+      "Fr": 38676,
+      "etition": 38677,
+      "Ġcesarean": 38678,
+      "olan": 38679,
+      "Ġrn": 38680,
+      "ubstituted": 38681,
+      "ĠLeaves": 38682,
+      "ĠLeader": 38683,
+      "coloring": 38684,
+      "Draw": 38685,
+      "Ġserous": 38686,
+      "Err": 38687,
+      "Ġinnermost": 38688,
+      "ĠHamburg": 38689,
+      "Stor": 38690,
+      "jes": 38691,
+      "Ġtol": 38692,
+      "idade": 38693,
+      "Ġrv": 38694,
+      "ĠInversion": 38695,
+      "Ġmultiphase": 38696,
+      "Ġpseudor": 38697,
+      "ĠGoodman": 38698,
+      "ĠJSON": 38699,
+      "Ġcorridor": 38700,
+      "Ġpork": 38701,
+      "ĠSale": 38702,
+      "ĠNatal": 38703,
+      "Ġattacking": 38704,
+      "ĠSheet": 38705,
+      "Ġstreamwise": 38706,
+      "Ġatomistic": 38707,
+      "Ġfirmly": 38708,
+      "ĠAchie": 38709,
+      "Ġpir": 38710,
+      "ĠIKK": 38711,
+      "ĠFalk": 38712,
+      "ileptic": 38713,
+      "ĠTRPC": 38714,
+      "Ġadhesions": 38715,
+      "HRP": 38716,
+      "Ġpaucity": 38717,
+      "Split": 38718,
+      "UDI": 38719,
+      "ĠSend": 38720,
+      "ĠPine": 38721,
+      "ĠLon": 38722,
+      "ĠLost": 38723,
+      "efer": 38724,
+      "concaten": 38725,
+      "Ġloyal": 38726,
+      "Ġglycop": 38727,
+      "ĠObserving": 38728,
+      "ĠMohamed": 38729,
+      "YR": 38730,
+      "ĠFilters": 38731,
+      "cas": 38732,
+      "pages": 38733,
+      "ĠdA": 38734,
+      "Ġareal": 38735,
+      "adis": 38736,
+      "ĠLHS": 38737,
+      "ĠThereby": 38738,
+      "Ġvisualizations": 38739,
+      "Ġtwistor": 38740,
+      "unitary": 38741,
+      "Ġarchives": 38742,
+      "Ġphenolics": 38743,
+      "hik": 38744,
+      "sson": 38745,
+      "ĠIK": 38746,
+      "ĠStudying": 38747,
+      "Ġtwisting": 38748,
+      "ĠHydrodynamic": 38749,
+      "Ġsplitter": 38750,
+      "Ġurothelial": 38751,
+      "Ġalken": 38752,
+      "ĠGPI": 38753,
+      "Ġcortices": 38754,
+      "Ġcropping": 38755,
+      "Patient": 38756,
+      "ĠChlamyd": 38757,
+      "inberg": 38758,
+      "ĠAircraft": 38759,
+      "cele": 38760,
+      "ectral": 38761,
+      "Ġconferences": 38762,
+      "Ġcreatine": 38763,
+      "alty": 38764,
+      "proportional": 38765,
+      "Ġleptonic": 38766,
+      "Ġovulation": 38767,
+      "uerre": 38768,
+      "tezomib": 38769,
+      "dle": 38770,
+      "initeness": 38771,
+      "ĠSpecimens": 38772,
+      "Ġcoma": 38773,
+      "inephrine": 38774,
+      "Ġepim": 38775,
+      "ĠPercent": 38776,
+      "CoO": 38777,
+      "ĠLoading": 38778,
+      "Ġvenue": 38779,
+      "ĠTNM": 38780,
+      "Ġpacemaker": 38781,
+      "ĠHoffmann": 38782,
+      "Tech": 38783,
+      "nie": 38784,
+      "ĠOrleans": 38785,
+      "Ġmagnetron": 38786,
+      "Ġhospitality": 38787,
+      "ĠNordic": 38788,
+      "oproliferative": 38789,
+      "Ġundoubtedly": 38790,
+      "ĠSrin": 38791,
+      "Ġhumic": 38792,
+      "ĠIntegrative": 38793,
+      "ĠCampus": 38794,
+      "Ġplantarum": 38795,
+      "radiative": 38796,
+      "Ġiterator": 38797,
+      "ĠMesozoic": 38798,
+      "APs": 38799,
+      "carinic": 38800,
+      "Ġcheckpoints": 38801,
+      "ĠïĤ£": 38802,
+      "ĠmAbs": 38803,
+      "ĠLiverpool": 38804,
+      "ìĿ´": 38805,
+      "ĠEcosystem": 38806,
+      "Ġneovascularization": 38807,
+      "Ġdemoc": 38808,
+      "loops": 38809,
+      "ĠSURF": 38810,
+      "Ġpassivation": 38811,
+      "Ġconsecutively": 38812,
+      "ĠAlfvén": 38813,
+      "ĠSSE": 38814,
+      "Ġouts": 38815,
+      "stimulation": 38816,
+      "Ġphilosophical": 38817,
+      "ĠSask": 38818,
+      "Ġflakes": 38819,
+      "Ġfingerprinting": 38820,
+      "Ġbuffalo": 38821,
+      "ĠWikimedia": 38822,
+      "Ġreconstitution": 38823,
+      "Ġepithelia": 38824,
+      "onk": 38825,
+      "eny": 38826,
+      "ĠMQ": 38827,
+      "ĠFork": 38828,
+      "endance": 38829,
+      "Ġgeneralisation": 38830,
+      "Ġpeoples": 38831,
+      "Ġconnector": 38832,
+      "gesia": 38833,
+      "interference": 38834,
+      "Ġcoloration": 38835,
+      "calculation": 38836,
+      "ĠAxial": 38837,
+      "ĠDESIGN": 38838,
+      "Ġrecession": 38839,
+      "Ġdissolve": 38840,
+      "ĠPartitioning": 38841,
+      "QxMD": 38842,
+      "GES": 38843,
+      "Vo": 38844,
+      "khar": 38845,
+      "ĠEAE": 38846,
+      "Ġcoarser": 38847,
+      "Ġposttraumatic": 38848,
+      "Ġsynthesised": 38849,
+      "silica": 38850,
+      "tetrahydropy": 38851,
+      "ĠPorter": 38852,
+      "vark": 38853,
+      "entanyl": 38854,
+      "Ġconve": 38855,
+      "Ġrafts": 38856,
+      "brecht": 38857,
+      "Ġrectifier": 38858,
+      "Ġoroph": 38859,
+      "ĠCEP": 38860,
+      "Ġhistones": 38861,
+      "Ġstandpoint": 38862,
+      "Ġancillary": 38863,
+      "ĠHurricane": 38864,
+      "cro": 38865,
+      "Ġreb": 38866,
+      "ĠiT": 38867,
+      "Ġgeography": 38868,
+      "olarization": 38869,
+      "ĠManaging": 38870,
+      "Ġxylose": 38871,
+      "utherland": 38872,
+      "ĠTaqMan": 38873,
+      "KN": 38874,
+      "Ġtm": 38875,
+      "ĠTAS": 38876,
+      "istle": 38877,
+      "âĢ«": 38878,
+      "Ġmycorrhizal": 38879,
+      "ĠTerrestrial": 38880,
+      "hausen": 38881,
+      "observable": 38882,
+      "Brien": 38883,
+      "Ġneutropenia": 38884,
+      "Taken": 38885,
+      "ĠSMI": 38886,
+      "Ġpolishing": 38887,
+      "Ġphotop": 38888,
+      "Ġthermalization": 38889,
+      "Ġpseudoscalar": 38890,
+      "ĠDominic": 38891,
+      "romyalgia": 38892,
+      "Ġechocardiographic": 38893,
+      "Illumina": 38894,
+      "ĠIPC": 38895,
+      "ĠHuss": 38896,
+      "essive": 38897,
+      "uptake": 38898,
+      "Ġweekend": 38899,
+      "Ġcorroborate": 38900,
+      "ĠTasman": 38901,
+      "herty": 38902,
+      "Ġperine": 38903,
+      "Ġtransports": 38904,
+      "Ġglance": 38905,
+      "retinal": 38906,
+      "Proto": 38907,
+      "igenes": 38908,
+      "Ġprohibited": 38909,
+      "behavioral": 38910,
+      "opherol": 38911,
+      "ë¡": 38912,
+      "ĠNecess": 38913,
+      "obiology": 38914,
+      "okk": 38915,
+      "Ġtraversal": 38916,
+      "ĠAndes": 38917,
+      "Resource": 38918,
+      "olitic": 38919,
+      "ça": 38920,
+      "irie": 38921,
+      "arctan": 38922,
+      "Ġmorphogenetic": 38923,
+      "ĠHui": 38924,
+      "losses": 38925,
+      "Ġfulfilling": 38926,
+      "Ġhurricane": 38927,
+      "ombo": 38928,
+      "Ġgs": 38929,
+      "ĠLv": 38930,
+      "ĠNerv": 38931,
+      "ellosis": 38932,
+      "Ġconfront": 38933,
+      "Ġorthologous": 38934,
+      "Ġwettability": 38935,
+      "Ġcyanobacterial": 38936,
+      "Ġcassava": 38937,
+      "AUT": 38938,
+      "avi": 38939,
+      "hlen": 38940,
+      "ĠSLA": 38941,
+      "Ġconvol": 38942,
+      "Ġintermetallic": 38943,
+      "inside": 38944,
+      "Ġpolarizability": 38945,
+      "Ġensuing": 38946,
+      "Ġchloroplasts": 38947,
+      "lid": 38948,
+      "lips": 38949,
+      "Ġrebound": 38950,
+      "ĠCary": 38951,
+      "ĠLambda": 38952,
+      "ĠViv": 38953,
+      "Ġcalcination": 38954,
+      "ĠÌĨ": 38955,
+      "Ġcounterfactual": 38956,
+      "ĠSilica": 38957,
+      "Referee": 38958,
+      "Ġhomologues": 38959,
+      "ĠSpatiotemporal": 38960,
+      "ĠArrhenius": 38961,
+      "Ġinflamed": 38962,
+      "ĠZambia": 38963,
+      "Ġantipsychotic": 38964,
+      "helper": 38965,
+      "Blood": 38966,
+      "Ġpurchasing": 38967,
+      "ĠSchwinger": 38968,
+      "ĠWilkinson": 38969,
+      "Ġfainter": 38970,
+      "Ġrash": 38971,
+      "ĠJang": 38972,
+      "ĠConductivity": 38973,
+      "ropoda": 38974,
+      "ĠSeq": 38975,
+      "Ġpropolis": 38976,
+      "Ġtubule": 38977,
+      "ĠLieb": 38978,
+      "optimization": 38979,
+      "mounted": 38980,
+      "emes": 38981,
+      "canic": 38982,
+      "oradiotherapy": 38983,
+      "ĠJenkins": 38984,
+      "Nc": 38985,
+      "Together": 38986,
+      "Ġfove": 38987,
+      "Ġmv": 38988,
+      "ĠDefect": 38989,
+      "ät": 38990,
+      "ĠFinance": 38991,
+      "umarin": 38992,
+      "mittance": 38993,
+      "erel": 38994,
+      "ĠFren": 38995,
+      "ĠRhyth": 38996,
+      "ramified": 38997,
+      "Ġhypercholesterolem": 38998,
+      "Ġstimulatory": 38999,
+      "ĠRichmond": 39000,
+      "Ġadvancements": 39001,
+      "bles": 39002,
+      "xu": 39003,
+      "allation": 39004,
+      "Ġintral": 39005,
+      "iterpene": 39006,
+      "Concerning": 39007,
+      "Ġbulky": 39008,
+      "Ġá¾±": 39009,
+      "computation": 39010,
+      "ĠAgarwal": 39011,
+      "Central": 39012,
+      "XPS": 39013,
+      "Ġtalks": 39014,
+      "ĠTap": 39015,
+      "imilar": 39016,
+      "ĠNCI": 39017,
+      "Ġaccused": 39018,
+      "Ġtranscriptomes": 39019,
+      "Ġprovisioning": 39020,
+      "ĠEtOH": 39021,
+      "gm": 39022,
+      "Ġtid": 39023,
+      "ĠPOC": 39024,
+      "ffman": 39025,
+      "ĠIner": 39026,
+      "ĠUB": 39027,
+      "incubated": 39028,
+      "ĠAtrial": 39029,
+      "Ġfourteen": 39030,
+      "ĠAstronomical": 39031,
+      "ĠMiguel": 39032,
+      "ĠKov": 39033,
+      "Ġscipy": 39034,
+      "Ġthermoplastic": 39035,
+      "ĠManuel": 39036,
+      "ĠPromotion": 39037,
+      "ĠAccessed": 39038,
+      "Ġterritorial": 39039,
+      "inas": 39040,
+      "ĠMPs": 39041,
+      "monitoring": 39042,
+      "ĠSimulating": 39043,
+      "Ġpanor": 39044,
+      "Ġrheumatic": 39045,
+      "selectin": 39046,
+      "ĠLaparoscopic": 39047,
+      "HLA": 39048,
+      "ĠYale": 39049,
+      "spread": 39050,
+      "ETS": 39051,
+      "Ġglycans": 39052,
+      "Ġimmigrant": 39053,
+      "Donald": 39054,
+      "ĠCASE": 39055,
+      "ĠHII": 39056,
+      "glomer": 39057,
+      "Ġïĥİ": 39058,
+      "ĠExperiences": 39059,
+      "ĠVietnamese": 39060,
+      "Hodgkin": 39061,
+      "oader": 39062,
+      "heart": 39063,
+      "Ġremedy": 39064,
+      "Ġfacilitators": 39065,
+      "openhagen": 39066,
+      "dodec": 39067,
+      "ĠFriend": 39068,
+      "ĠTouch": 39069,
+      "arms": 39070,
+      "CRs": 39071,
+      "Ġultrahigh": 39072,
+      "ĠDriver": 39073,
+      "GEMENTS": 39074,
+      "ĠOu": 39075,
+      "Ġendocarditis": 39076,
+      "Ġautoencoder": 39077,
+      "Ġich": 39078,
+      "Ġfetch": 39079,
+      "urian": 39080,
+      "ĠORFs": 39081,
+      "Ġpermeabilized": 39082,
+      "ĠWiFi": 39083,
+      "ĠLithuan": 39084,
+      "Structure": 39085,
+      "Ln": 39086,
+      "houses": 39087,
+      "Ġought": 39088,
+      "ĠConcluding": 39089,
+      "Ġanniversary": 39090,
+      "ĠCreation": 39091,
+      "Ġblindness": 39092,
+      "ĠpcDNA": 39093,
+      "ĠSusan": 39094,
+      "ĠBenjamini": 39095,
+      "ĠSentence": 39096,
+      "Ġsnd": 39097,
+      "Ġfins": 39098,
+      "phis": 39099,
+      "ĠModules": 39100,
+      "Ġneuropsychiatric": 39101,
+      "ĠPotassium": 39102,
+      "Ġsacrifice": 39103,
+      "Ġdyspnea": 39104,
+      "Ġdeliberately": 39105,
+      "omegaly": 39106,
+      "Media": 39107,
+      "Temporal": 39108,
+      "Ġshark": 39109,
+      "SCAN": 39110,
+      "splitting": 39111,
+      "Ġmisuse": 39112,
+      "Ġbirefringence": 39113,
+      "ĠÖĴâĨĴ": 39114,
+      "Ġpier": 39115,
+      "Ġnurs": 39116,
+      "ĠSclerosis": 39117,
+      "adhy": 39118,
+      "Ġundetermined": 39119,
+      "Ġcomplementation": 39120,
+      "ĠAffect": 39121,
+      "ĠHamps": 39122,
+      "Ġgob": 39123,
+      "ĠFate": 39124,
+      "ĠHAL": 39125,
+      "ĠKiss": 39126,
+      "Ġmicrobe": 39127,
+      "Ġcarbonaceous": 39128,
+      "Ġliposome": 39129,
+      "ĠUsage": 39130,
+      "Ġquasiparticles": 39131,
+      "Ġcasp": 39132,
+      "ĠNarrow": 39133,
+      "Ġoutlook": 39134,
+      "ĠChord": 39135,
+      "Ġclaiming": 39136,
+      "Ġdiverging": 39137,
+      "ĠBioinformatics": 39138,
+      "ĠPsychiatric": 39139,
+      "ĠMasters": 39140,
+      "Ġllvm": 39141,
+      "ĠIQR": 39142,
+      "phases": 39143,
+      "ĠThy": 39144,
+      "erger": 39145,
+      "ĠDipl": 39146,
+      "SFR": 39147,
+      "Ġcredited": 39148,
+      "ĠTetra": 39149,
+      "âĭ¯": 39150,
+      "Ġamniotic": 39151,
+      "ĠCharlotte": 39152,
+      "Cox": 39153,
+      "Hard": 39154,
+      "article": 39155,
+      "ĠDEA": 39156,
+      "ĠEclipse": 39157,
+      "ĠLMP": 39158,
+      "Ġimprison": 39159,
+      "ĠVarying": 39160,
+      "ESCs": 39161,
+      "ĠTHEO": 39162,
+      "Ġnervosa": 39163,
+      "Ġprecedes": 39164,
+      "Ġgyro": 39165,
+      "ĠWORDS": 39166,
+      "ĠDakota": 39167,
+      "utory": 39168,
+      "ĠEmer": 39169,
+      "adam": 39170,
+      "ĠNah": 39171,
+      "ĠVirgo": 39172,
+      "Setting": 39173,
+      "PQ": 39174,
+      "å®": 39175,
+      "erus": 39176,
+      "Ġcep": 39177,
+      "Ġbd": 39178,
+      "dier": 39179,
+      "Ġimbalanced": 39180,
+      "Ġtimestep": 39181,
+      "än": 39182,
+      "ĠRabbit": 39183,
+      "Ġhamsters": 39184,
+      "Ġmedulla": 39185,
+      "ĠChromatography": 39186,
+      "INPUT": 39187,
+      "Ġlossy": 39188,
+      "Pseud": 39189,
+      "ĠPBL": 39190,
+      "ĠDomestic": 39191,
+      "iau": 39192,
+      "ancell": 39193,
+      "Ġmultilayers": 39194,
+      "Ġsubsidi": 39195,
+      "ĠUtilizing": 39196,
+      "tune": 39197,
+      "rehend": 39198,
+      "arte": 39199,
+      "Ġburs": 39200,
+      "ĠNHE": 39201,
+      "Ġcloseness": 39202,
+      "ĠColour": 39203,
+      "ĠHomo": 39204,
+      "Equations": 39205,
+      "Ġsutures": 39206,
+      "acus": 39207,
+      "Ġknocked": 39208,
+      "Ġsecretary": 39209,
+      "Ġascertained": 39210,
+      "Ġinpatients": 39211,
+      "irts": 39212,
+      "Ġplut": 39213,
+      "ansson": 39214,
+      "rami": 39215,
+      "Ġosteotomy": 39216,
+      "ĠPrimers": 39217,
+      "ĠLegislative": 39218,
+      "ĠCardiology": 39219,
+      "Ġadmitting": 39220,
+      "Ġexcavation": 39221,
+      "ĠHedgehog": 39222,
+      "WG": 39223,
+      "frozen": 39224,
+      "Ġliber": 39225,
+      "ĠICE": 39226,
+      "chosen": 39227,
+      "ĠKohn": 39228,
+      "Stop": 39229,
+      "Phil": 39230,
+      "phagia": 39231,
+      "ĠBCA": 39232,
+      "Ġempt": 39233,
+      "Ġzz": 39234,
+      "opers": 39235,
+      "ĠSixty": 39236,
+      "eckman": 39237,
+      "Ġtransferrin": 39238,
+      "Ġpenalized": 39239,
+      "Being": 39240,
+      "Ġextruded": 39241,
+      "Ġminiature": 39242,
+      "Ġeditorial": 39243,
+      "Ġinterconnect": 39244,
+      "gro": 39245,
+      "kv": 39246,
+      "olen": 39247,
+      "ĠSYSTEMS": 39248,
+      "ĠColonel": 39249,
+      "ĠMediated": 39250,
+      "ĠEMD": 39251,
+      "Ġknife": 39252,
+      "Ġcytogenetic": 39253,
+      "Ġdigitized": 39254,
+      "abinoids": 39255,
+      "arterial": 39256,
+      "Ġdiarrhoea": 39257,
+      "bag": 39258,
+      "Ġbuccal": 39259,
+      "stay": 39260,
+      "ĠLAMP": 39261,
+      "oko": 39262,
+      "ĠPolyt": 39263,
+      "masked": 39264,
+      "ĠTunable": 39265,
+      "Ġcoagul": 39266,
+      "paras": 39267,
+      "Ġterminating": 39268,
+      "ICAg": 39269,
+      "ĠExcellence": 39270,
+      "Ġregurgitation": 39271,
+      "DQUFD": 39272,
+      "Jack": 39273,
+      "Ġapertures": 39274,
+      "ĠIp": 39275,
+      "ĠHCMV": 39276,
+      "ĠGom": 39277,
+      "Ġnucleophilic": 39278,
+      "Ġparenteral": 39279,
+      "TIM": 39280,
+      "oine": 39281,
+      "ĠnT": 39282,
+      "ĠSense": 39283,
+      "ĠFocal": 39284,
+      "ranges": 39285,
+      "Ġhept": 39286,
+      "ĠPlat": 39287,
+      "Ġmyx": 39288,
+      "Ġcodebook": 39289,
+      "Expl": 39290,
+      "ĠRhoA": 39291,
+      "Ġrhinitis": 39292,
+      "ĠErratum": 39293,
+      "Oriented": 39294,
+      "Well": 39295,
+      "doping": 39296,
+      "Ġbup": 39297,
+      "ĠImpedance": 39298,
+      "Ġsubstitutes": 39299,
+      "actorily": 39300,
+      "Ġcollaborations": 39301,
+      "ĠWayne": 39302,
+      "Ġvowels": 39303,
+      "ĠShadow": 39304,
+      "Ġphenology": 39305,
+      "Ġconcurrency": 39306,
+      "having": 39307,
+      "ĠCES": 39308,
+      "ĠFIN": 39309,
+      "ĠLoh": 39310,
+      "oxa": 39311,
+      "ĠAlN": 39312,
+      "ĠAlvarez": 39313,
+      "instit": 39314,
+      "Ġgermplasm": 39315,
+      "ĠBoliv": 39316,
+      "ĠRCP": 39317,
+      "assador": 39318,
+      "Ġesp": 39319,
+      "Ġphenotyping": 39320,
+      "Ġskipping": 39321,
+      "ĠFractal": 39322,
+      "ĠPEDOT": 39323,
+      "wake": 39324,
+      "ĠFIT": 39325,
+      "ĠESD": 39326,
+      "ĠAntif": 39327,
+      "ubiquitin": 39328,
+      "ĠAerial": 39329,
+      "ĠPrognosis": 39330,
+      "ĠRELATED": 39331,
+      "Ġstratigraphy": 39332,
+      "vatron": 39333,
+      "ĠPROPERTIES": 39334,
+      "Ġicon": 39335,
+      "isers": 39336,
+      "Ġwal": 39337,
+      "Ġstamp": 39338,
+      "ĠOptimum": 39339,
+      "Ġoligomeric": 39340,
+      "Ġinnerv": 39341,
+      "YA": 39342,
+      "Abcam": 39343,
+      "Ġvials": 39344,
+      "ĠGrig": 39345,
+      "Ġunaware": 39346,
+      "Ġopera": 39347,
+      "ĠWarner": 39348,
+      "Ġprotonated": 39349,
+      "ĠDRG": 39350,
+      "Ġtroubles": 39351,
+      "Ġpropositional": 39352,
+      "ĠAfghanistan": 39353,
+      "ĠHampshire": 39354,
+      "Gd": 39355,
+      "lung": 39356,
+      "Ġaviation": 39357,
+      "Ġapartment": 39358,
+      "Ġinfusions": 39359,
+      "Ġbroilers": 39360,
+      "ĠDisability": 39361,
+      "ĠRobots": 39362,
+      "Ġdebugging": 39363,
+      "ĠìĿ": 39364,
+      "Wilson": 39365,
+      "uprofen": 39366,
+      "obarbital": 39367,
+      "JB": 39368,
+      "isance": 39369,
+      "itizer": 39370,
+      "MIS": 39371,
+      "ĠARF": 39372,
+      "Ġprostheses": 39373,
+      "Ġdichloromethane": 39374,
+      "mCherry": 39375,
+      "ĠSSS": 39376,
+      "ĠLPA": 39377,
+      "SCF": 39378,
+      "attract": 39379,
+      "Ġcalibrations": 39380,
+      "Ġfibril": 39381,
+      "Ġhaploid": 39382,
+      "usalem": 39383,
+      "ĠNut": 39384,
+      "Ġdeut": 39385,
+      "chronic": 39386,
+      "NAP": 39387,
+      "ĠCytokines": 39388,
+      "rageen": 39389,
+      "ĠCategories": 39390,
+      "rains": 39391,
+      "Ġsummands": 39392,
+      "Ġproliferate": 39393,
+      "rylov": 39394,
+      "Ġpleasure": 39395,
+      "Ġdensit": 39396,
+      "ĠSURVE": 39397,
+      "HIP": 39398,
+      "hall": 39399,
+      "ĠFUS": 39400,
+      "Ġwasting": 39401,
+      "ERY": 39402,
+      "Ġstatins": 39403,
+      "Ġeastward": 39404,
+      "sometimes": 39405,
+      "Ġwrapping": 39406,
+      "ĠTWO": 39407,
+      "vine": 39408,
+      "Ġsacchar": 39409,
+      "Ġamateur": 39410,
+      "ĠÃĽ": 39411,
+      "Ġmyster": 39412,
+      "ĠMyo": 39413,
+      "Ġrhabd": 39414,
+      "ĠProtease": 39415,
+      "Ġcholera": 39416,
+      "ĠGov": 39417,
+      "ĠGCC": 39418,
+      "Ġclays": 39419,
+      "transmission": 39420,
+      "ĠHollywood": 39421,
+      "Ġxenob": 39422,
+      "FLOAT": 39423,
+      "Ġascent": 39424,
+      "Ġsharks": 39425,
+      "Ġinterferes": 39426,
+      "ĠFormer": 39427,
+      "ĠHartmann": 39428,
+      "sha": 39429,
+      "ĠSave": 39430,
+      "Ġparks": 39431,
+      "ĠVenn": 39432,
+      "Ġunions": 39433,
+      "Ġdiscour": 39434,
+      "Ġsuperlattices": 39435,
+      "Ġcoupler": 39436,
+      "proteins": 39437,
+      "ĠStationary": 39438,
+      "ĠEthernet": 39439,
+      "ĠFréchet": 39440,
+      "Ġkines": 39441,
+      "Ġjazz": 39442,
+      "Asn": 39443,
+      "Ġextensional": 39444,
+      "Ġtelomeres": 39445,
+      "Ġpermitting": 39446,
+      "Ġexhausted": 39447,
+      "ĠSphing": 39448,
+      "Turn": 39449,
+      "mind": 39450,
+      "Ġsf": 39451,
+      "ĠHak": 39452,
+      "ranolol": 39453,
+      "portation": 39454,
+      "Consistent": 39455,
+      "Ġventilated": 39456,
+      "ĠDISTRIB": 39457,
+      "Ġtelling": 39458,
+      "Ġmannose": 39459,
+      "ÃŃaz": 39460,
+      "Ġborne": 39461,
+      "Ġintensification": 39462,
+      "Ġenjoyed": 39463,
+      "ĠBruno": 39464,
+      "ĠSaturday": 39465,
+      "Ġcocycle": 39466,
+      "itate": 39467,
+      "Ġgolf": 39468,
+      "approved": 39469,
+      "ĠNikol": 39470,
+      "itri": 39471,
+      "ĠSentiment": 39472,
+      "Ġglow": 39473,
+      "Ġgyp": 39474,
+      "ĠPCT": 39475,
+      "aber": 39476,
+      "ĠWis": 39477,
+      "porum": 39478,
+      "Ġhyphae": 39479,
+      "feas": 39480,
+      "ĠTraits": 39481,
+      "ĠConflicts": 39482,
+      "degrading": 39483,
+      "Raman": 39484,
+      "pharmac": 39485,
+      "Ġimmunocyt": 39486,
+      "ĠBlake": 39487,
+      "Ġpseudoc": 39488,
+      "ĠCharacterisation": 39489,
+      "ĠGalileo": 39490,
+      "Enabl": 39491,
+      "Jy": 39492,
+      "Ġclav": 39493,
+      "Ġϳ": 39494,
+      "Ġcommunicated": 39495,
+      "eutical": 39496,
+      "Ġnanotechnology": 39497,
+      "ĠHassan": 39498,
+      "ĠTec": 39499,
+      "Ġhanging": 39500,
+      "ĠBSD": 39501,
+      "ĠContour": 39502,
+      "Ġfragility": 39503,
+      "Ġdisruptions": 39504,
+      "Ġfiniteness": 39505,
+      "ĠPhilippine": 39506,
+      "nicity": 39507,
+      "Ùĩ": 39508,
+      "ĠCrim": 39509,
+      "ĠCNF": 39510,
+      "ĠISI": 39511,
+      "adapter": 39512,
+      "ĠUCP": 39513,
+      "Ġtextured": 39514,
+      "AAV": 39515,
+      "keto": 39516,
+      "Np": 39517,
+      "counting": 39518,
+      "hynchus": 39519,
+      "Ġprosec": 39520,
+      "ĠAnnot": 39521,
+      "ĠHarbor": 39522,
+      "degrees": 39523,
+      "akar": 39524,
+      "ĠVik": 39525,
+      "bfd": 39526,
+      "Ġdrip": 39527,
+      "ĠCaucas": 39528,
+      "Ġtrench": 39529,
+      "Ġweed": 39530,
+      "Ġdistractor": 39531,
+      "genetic": 39532,
+      "specifically": 39533,
+      "ulfite": 39534,
+      "ĠConsistently": 39535,
+      "Ġbreakfast": 39536,
+      "Ġbullet": 39537,
+      "Ġlegisl": 39538,
+      "ĠTraumatic": 39539,
+      "Ġcollectors": 39540,
+      "ĠBullet": 39541,
+      "ĠMYB": 39542,
+      "ĠPink": 39543,
+      "versive": 39544,
+      "ĠAttem": 39545,
+      "Ġculturally": 39546,
+      "Bell": 39547,
+      "undef": 39548,
+      "vii": 39549,
+      "Ġhistocompatibility": 39550,
+      "letcher": 39551,
+      "ĠStef": 39552,
+      "Amp": 39553,
+      "ĠRid": 39554,
+      "ĠEucl": 39555,
+      "Ġdecryption": 39556,
+      "ĠSpencer": 39557,
+      "ĠBitcoin": 39558,
+      "wic": 39559,
+      "Ġcomplicate": 39560,
+      "ĠProposal": 39561,
+      "ĠÄĪ": 39562,
+      "aviruses": 39563,
+      "ĠFay": 39564,
+      "ĠRd": 39565,
+      "ĠGale": 39566,
+      "ĠMetastasis": 39567,
+      "ĠImprovements": 39568,
+      "©": 39569,
+      "Ġpolyester": 39570,
+      "Ġstratospheric": 39571,
+      "ĠSAH": 39572,
+      "Ġamphip": 39573,
+      "ĠAFP": 39574,
+      "ĠHair": 39575,
+      "ĠEPI": 39576,
+      "ĠUltrast": 39577,
+      "Ġâĭ¯": 39578,
+      "Ġgapless": 39579,
+      "Ham": 39580,
+      "etto": 39581,
+      "Ġthreonine": 39582,
+      "ĠECO": 39583,
+      "Ġia": 39584,
+      "Ġundist": 39585,
+      "Ġradiology": 39586,
+      "Ġsuperlattice": 39587,
+      "ibraries": 39588,
+      "Ġturbid": 39589,
+      "ĠPotentials": 39590,
+      "ĠPipeline": 39591,
+      "Ġwarfarin": 39592,
+      "WISE": 39593,
+      "ĠLid": 39594,
+      "Ġrecurring": 39595,
+      "ĠMono": 39596,
+      "ĠGovern": 39597,
+      "ĠAwareness": 39598,
+      "olab": 39599,
+      "iflora": 39600,
+      "stris": 39601,
+      "INDEX": 39602,
+      "ĠDementia": 39603,
+      "Does": 39604,
+      "wright": 39605,
+      "Íī": 39606,
+      "Ġsb": 39607,
+      "ĠDOM": 39608,
+      "ĠHBsAg": 39609,
+      "clinic": 39610,
+      "ĠExped": 39611,
+      "Ġproteas": 39612,
+      "Ġsterilization": 39613,
+      "ĠBanerjee": 39614,
+      "ĠPersonnel": 39615,
+      "âĮĭ": 39616,
+      "onephritis": 39617,
+      "omite": 39618,
+      "ĠCCF": 39619,
+      "ositi": 39620,
+      "ĠEucalyptus": 39621,
+      "ĠIsotope": 39622,
+      "coli": 39623,
+      "possibility": 39624,
+      "Ġstrontium": 39625,
+      "Ġraref": 39626,
+      "ĠInterstellar": 39627,
+      "kinin": 39628,
+      "ylethanol": 39629,
+      "JT": 39630,
+      "north": 39631,
+      "Ġcensored": 39632,
+      "istive": 39633,
+      "Ġnoticing": 39634,
+      "Ġshipping": 39635,
+      "Embed": 39636,
+      "Observ": 39637,
+      "Ġzeolites": 39638,
+      "ubit": 39639,
+      "Ġflaps": 39640,
+      "Ġdrifts": 39641,
+      "Ġtherapist": 39642,
+      "Ġpollination": 39643,
+      "aliplatin": 39644,
+      "Johnson": 39645,
+      "Ġimperfections": 39646,
+      "NY": 39647,
+      "Ġthalamic": 39648,
+      "ocarb": 39649,
+      "ozotocin": 39650,
+      "Ġtetramer": 39651,
+      "Plas": 39652,
+      "Ġmultichannel": 39653,
+      "ĠInsight": 39654,
+      "opods": 39655,
+      "ĠNacional": 39656,
+      "Ġimatinib": 39657,
+      "actual": 39658,
+      "ĠXOR": 39659,
+      "Ġblight": 39660,
+      "ĠLeading": 39661,
+      "amese": 39662,
+      "ĠAmplitude": 39663,
+      "ĠMonitor": 39664,
+      "ĠNeurological": 39665,
+      "propagating": 39666,
+      "Ġpaddle": 39667,
+      "ĠHarvest": 39668,
+      "Ġodont": 39669,
+      "BUF": 39670,
+      "Ġtactics": 39671,
+      "ĠAnisotropy": 39672,
+      "adip": 39673,
+      "ĠAlpine": 39674,
+      "Ġfeels": 39675,
+      "Ġmedieval": 39676,
+      "Ġelucidation": 39677,
+      "Ġheterotrophic": 39678,
+      "Ġrelaxing": 39679,
+      "Ġhappiness": 39680,
+      "ĠCytotoxicity": 39681,
+      "ĠRANKL": 39682,
+      "Walker": 39683,
+      "mig": 39684,
+      "ĠSSL": 39685,
+      "ĠSepsis": 39686,
+      "ĠGes": 39687,
+      "Ġhydrochloric": 39688,
+      "Ġclarification": 39689,
+      "Ġdisparate": 39690,
+      "tested": 39691,
+      "Ġdatap": 39692,
+      "Ġnovels": 39693,
+      "ĠMicroc": 39694,
+      "ál": 39695,
+      "ĠARC": 39696,
+      "ĠYangtze": 39697,
+      "etomidine": 39698,
+      "ĠMatrigel": 39699,
+      "ihilation": 39700,
+      "ĠcDNAs": 39701,
+      "Ġprostat": 39702,
+      "ĠRailroad": 39703,
+      "UBLE": 39704,
+      "ĠPARTIC": 39705,
+      "ĠSax": 39706,
+      "Ġinsecurity": 39707,
+      "Ġcrushed": 39708,
+      "Ġhalves": 39709,
+      "giant": 39710,
+      "ĠCroatia": 39711,
+      "icyclo": 39712,
+      "ĠUnexpected": 39713,
+      "Ġloneliness": 39714,
+      "anu": 39715,
+      "Ġchampions": 39716,
+      "uberculosis": 39717,
+      "Ġequi": 39718,
+      "Ġaccreted": 39719,
+      "Ġinvading": 39720,
+      "Ġafferents": 39721,
+      "Ġalternation": 39722,
+      "Ġkinet": 39723,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 39724,
+      "ĠMAGNET": 39725,
+      "ĠFIFA": 39726,
+      "zadeh": 39727,
+      "iphenyl": 39728,
+      "ĠKro": 39729,
+      "ĠEvaluate": 39730,
+      "illiant": 39731,
+      "curvature": 39732,
+      "ĠPierce": 39733,
+      "better": 39734,
+      "nos": 39735,
+      "à¥": 39736,
+      "ĠKCN": 39737,
+      "ĠStrand": 39738,
+      "caemic": 39739,
+      "ĠHoechst": 39740,
+      "ĠEXT": 39741,
+      "ĠLLVM": 39742,
+      "BZ": 39743,
+      "tgt": 39744,
+      "ondialdehyde": 39745,
+      "ĠEvid": 39746,
+      "ĠGul": 39747,
+      "Ġmultiplications": 39748,
+      "Ġauth": 39749,
+      "ĠAustral": 39750,
+      "Ġstaying": 39751,
+      "ĠGlutamate": 39752,
+      "Ġstray": 39753,
+      "ĠISA": 39754,
+      "Ġlowland": 39755,
+      "Ġparallels": 39756,
+      "Ġattractiveness": 39757,
+      "Ġelectrospinning": 39758,
+      "Ġportrayed": 39759,
+      "ospecific": 39760,
+      "folate": 39761,
+      "Ġcoeff": 39762,
+      "ĠEstrogen": 39763,
+      "tumour": 39764,
+      "Ġhysterectomy": 39765,
+      "Ġinositol": 39766,
+      "ĠBaz": 39767,
+      "istein": 39768,
+      "Ġcrucially": 39769,
+      "Ġdinoflag": 39770,
+      "ÍĶÍĴ": 39771,
+      "ĠDragon": 39772,
+      "ĠSpor": 39773,
+      "ĠMater": 39774,
+      "ĠHero": 39775,
+      "plicing": 39776,
+      "ĠANT": 39777,
+      "ĠFormic": 39778,
+      "Queue": 39779,
+      "ocarcinomas": 39780,
+      "UPS": 39781,
+      "ĠPc": 39782,
+      "encoders": 39783,
+      "Ġinvaded": 39784,
+      "ĠPhases": 39785,
+      "Ġpostmortem": 39786,
+      "Ġslows": 39787,
+      "ĠMcL": 39788,
+      "ĠVerma": 39789,
+      "ĠViability": 39790,
+      "Ġcompensating": 39791,
+      "Ġclamped": 39792,
+      "jm": 39793,
+      "ĠRiv": 39794,
+      "upon": 39795,
+      "ĠDickinson": 39796,
+      "initiated": 39797,
+      "Ġsider": 39798,
+      "ĠSelen": 39799,
+      "ĠAka": 39800,
+      "idelberg": 39801,
+      "Ġqualifying": 39802,
+      "Ġenforcing": 39803,
+      "otrophs": 39804,
+      "ĠSNAP": 39805,
+      "Ġrust": 39806,
+      "imburs": 39807,
+      "Ġimmunocompromised": 39808,
+      "ĠFleming": 39809,
+      "Ġlizards": 39810,
+      "dialysis": 39811,
+      "ĠUnivariate": 39812,
+      "Ġgasoline": 39813,
+      "Ġtenure": 39814,
+      "Ġsustaining": 39815,
+      "Ġmotone": 39816,
+      "bay": 39817,
+      "wani": 39818,
+      "orestation": 39819,
+      "ĠXII": 39820,
+      "Ġradiofrequency": 39821,
+      "ĠGuided": 39822,
+      "Individual": 39823,
+      "ĠSpectrometer": 39824,
+      "ĠGoing": 39825,
+      "ĠMartins": 39826,
+      "Approxim": 39827,
+      "amak": 39828,
+      "ĠâĪı": 39829,
+      "ĠOmn": 39830,
+      "Ġoutpatients": 39831,
+      "Ġhyperbol": 39832,
+      "ĠPerceptual": 39833,
+      "ĠBurke": 39834,
+      "Boltzmann": 39835,
+      "ĠMd": 39836,
+      "Ġpaw": 39837,
+      "ĠCathedral": 39838,
+      "Ġhyaluron": 39839,
+      "Ġbrachial": 39840,
+      "Ġaflatoxin": 39841,
+      "imo": 39842,
+      "Ġenrol": 39843,
+      "Ġdetonation": 39844,
+      "Ġoverly": 39845,
+      "thest": 39846,
+      "Ġsecondly": 39847,
+      "ĠSchiz": 39848,
+      "ĠIGFBP": 39849,
+      "atechin": 39850,
+      "Ġsaves": 39851,
+      "tiers": 39852,
+      "ĠBates": 39853,
+      "Ġalliance": 39854,
+      "Ġattri": 39855,
+      "Ġastro": 39856,
+      "ĠPathological": 39857,
+      "Ġgambiae": 39858,
+      "Park": 39859,
+      "idable": 39860,
+      "ĠNil": 39861,
+      "ĠJas": 39862,
+      "Ġneeding": 39863,
+      "meier": 39864,
+      "Ġferroptosis": 39865,
+      "ĠGuidance": 39866,
+      "AZ": 39867,
+      "iol": 39868,
+      "Ġacknowledg": 39869,
+      "exual": 39870,
+      "Ġmenopause": 39871,
+      "Ġadjunct": 39872,
+      "capture": 39873,
+      "ĠDeputy": 39874,
+      "Ġbial": 39875,
+      "ifa": 39876,
+      "ĠChitosan": 39877,
+      "ĠTopics": 39878,
+      "ĠPlasmid": 39879,
+      "calculations": 39880,
+      "give": 39881,
+      "responders": 39882,
+      "ulla": 39883,
+      "ĠMoreno": 39884,
+      "Ġcommentary": 39885,
+      "ĠMahm": 39886,
+      "": 39887,
+      "onacci": 39888,
+      "ĠCould": 39889,
+      "ĠTRP": 39890,
+      "seconds": 39891,
+      "GraphPad": 39892,
+      "Little": 39893,
+      "hey": 39894,
+      "Ġalike": 39895,
+      "ĠDias": 39896,
+      "aroo": 39897,
+      "Ġı": 39898,
+      "Ġtaxes": 39899,
+      "phenanth": 39900,
+      "ĠCheung": 39901,
+      "ĠPiet": 39902,
+      "Df": 39903,
+      "GU": 39904,
+      "mectin": 39905,
+      "zee": 39906,
+      "Ġdλ": 39907,
+      "Ġsyntheses": 39908,
+      "ĠáĪ": 39909,
+      "Simulation": 39910,
+      "ĠEleven": 39911,
+      "worms": 39912,
+      "lymphocyte": 39913,
+      "Ġhaemorrhage": 39914,
+      "ĠOwn": 39915,
+      "ĠKant": 39916,
+      "Ġoverse": 39917,
+      "Ġideation": 39918,
+      "ĠHarper": 39919,
+      "Acknowledgments": 39920,
+      "vili": 39921,
+      "yna": 39922,
+      "ĠRecurrence": 39923,
+      "oza": 39924,
+      "Ġhenceforth": 39925,
+      "zees": 39926,
+      "Ġquasic": 39927,
+      "Ġchoroidal": 39928,
+      "Ġantimalarial": 39929,
+      "Ġcoarsening": 39930,
+      "Deb": 39931,
+      "diam": 39932,
+      "ĠWeights": 39933,
+      "Ġbuying": 39934,
+      "Ġmessaging": 39935,
+      "February": 39936,
+      "Extended": 39937,
+      "ĠRossi": 39938,
+      "Ġmistaken": 39939,
+      "Ġutero": 39940,
+      "jas": 39941,
+      "icitis": 39942,
+      "ĠTidal": 39943,
+      "Ġpharyngeal": 39944,
+      "click": 39945,
+      "Ġmyo": 39946,
+      "knock": 39947,
+      "Ġprominence": 39948,
+      "Ġamphiphilic": 39949,
+      "corn": 39950,
+      "Ġonboard": 39951,
+      "ĠDud": 39952,
+      "ĠWoman": 39953,
+      "ĠOutbreak": 39954,
+      "Ġpreferably": 39955,
+      "Ġsketches": 39956,
+      "Sat": 39957,
+      "fixing": 39958,
+      "ĠMey": 39959,
+      "ĠLetters": 39960,
+      "ITIES": 39961,
+      "ĠSDP": 39962,
+      "ĠLNCaP": 39963,
+      "DX": 39964,
+      "Fluor": 39965,
+      "Rv": 39966,
+      "Sect": 39967,
+      "ĠIons": 39968,
+      "Ġtrachom": 39969,
+      "Ġultrastructure": 39970,
+      "qvist": 39971,
+      "rophe": 39972,
+      "Ġreceipt": 39973,
+      "ĠQuint": 39974,
+      "Ġswapping": 39975,
+      "aminidase": 39976,
+      "Ġarchival": 39977,
+      "ĠCreating": 39978,
+      "ĠBarton": 39979,
+      "diagnosed": 39980,
+      "atological": 39981,
+      "olph": 39982,
+      "ĠPFA": 39983,
+      "ĠLAP": 39984,
+      "Ġunphysical": 39985,
+      "eqn": 39986,
+      "Ġquartiles": 39987,
+      "olytica": 39988,
+      "ĠFreed": 39989,
+      "Ġventilator": 39990,
+      "Ġkaryotype": 39991,
+      "Sta": 39992,
+      "still": 39993,
+      "ĠTate": 39994,
+      "urability": 39995,
+      "ĠGron": 39996,
+      "Ġtrimer": 39997,
+      "IPA": 39998,
+      "adeca": 39999,
+      "ĠImplementing": 40000,
+      "sity": 40001,
+      "itr": 40002,
+      "Ġbom": 40003,
+      "Ġnonrelativistic": 40004,
+      "Ġmicelle": 40005,
+      "ĠAdminist": 40006,
+      "Ġelectrolysis": 40007,
+      "harmon": 40008,
+      "OLOGICAL": 40009,
+      "Liter": 40010,
+      "ĠGUI": 40011,
+      "ĠQL": 40012,
+      "months": 40013,
+      "Ġsuperflu": 40014,
+      "cuts": 40015,
+      "Ġelicits": 40016,
+      "Ġmultiplexed": 40017,
+      "overlap": 40018,
+      "Ġcadaver": 40019,
+      "Ġou": 40020,
+      "ĠSheng": 40021,
+      "erea": 40022,
+      "ĠNBC": 40023,
+      "Ġdeter": 40024,
+      "tyrosine": 40025,
+      "ĠParts": 40026,
+      "Ġessay": 40027,
+      "kas": 40028,
+      "itted": 40029,
+      "ĠPZT": 40030,
+      "essler": 40031,
+      "Ġsimulators": 40032,
+      "Ġradiating": 40033,
+      "cutting": 40034,
+      "ĠCalculating": 40035,
+      "THER": 40036,
+      "ĠROCK": 40037,
+      "communic": 40038,
+      "Ġbonus": 40039,
+      "ĠCPA": 40040,
+      "ĠPUR": 40041,
+      "ulton": 40042,
+      "ĠZhi": 40043,
+      "Ġcaloric": 40044,
+      "Ġinterpolate": 40045,
+      "ĠSecretion": 40046,
+      "Ġneurocognitive": 40047,
+      "Ġgadolinium": 40048,
+      "frequencies": 40049,
+      "ĠTract": 40050,
+      "Ġminimax": 40051,
+      "ĠBrock": 40052,
+      "rypsin": 40053,
+      "ĠResonant": 40054,
+      "ĠACKNOWLEDGEMENTS": 40055,
+      "Dom": 40056,
+      "Ġholotype": 40057,
+      "Special": 40058,
+      "Ġimmunoreactive": 40059,
+      "ARNING": 40060,
+      "Panel": 40061,
+      "ĠJohannes": 40062,
+      "RFP": 40063,
+      "zzi": 40064,
+      "ĠPomer": 40065,
+      "Ġtransects": 40066,
+      "Ġpoured": 40067,
+      "EDs": 40068,
+      "ĠCircum": 40069,
+      "Ġabnormally": 40070,
+      "ĠPunj": 40071,
+      "Gol": 40072,
+      "Hop": 40073,
+      "Hex": 40074,
+      "ILE": 40075,
+      "Ġsourced": 40076,
+      "oclase": 40077,
+      "protobuf": 40078,
+      "Ġfrogs": 40079,
+      "ĠOttawa": 40080,
+      "Ġbiogeochemical": 40081,
+      "Ġlentivirus": 40082,
+      "Young": 40083,
+      "ĠIPS": 40084,
+      "assen": 40085,
+      "Ġunrestricted": 40086,
+      "Ġmatplotlib": 40087,
+      "Ġchloramphenicol": 40088,
+      "ĠContextual": 40089,
+      "ĠHawaiian": 40090,
+      "Legend": 40091,
+      "Sparse": 40092,
+      "bore": 40093,
+      "gaussian": 40094,
+      "uke": 40095,
+      "ĠâĢ°": 40096,
+      "retest": 40097,
+      "SSE": 40098,
+      "preting": 40099,
+      "ĠPanama": 40100,
+      "ĠBroadband": 40101,
+      "conjugate": 40102,
+      "Bytes": 40103,
+      "GSH": 40104,
+      "Uns": 40105,
+      "rina": 40106,
+      "Ġdrained": 40107,
+      "Ġscap": 40108,
+      "Ġinvested": 40109,
+      "Ġsatisfactorily": 40110,
+      "Ġherbivores": 40111,
+      "Ġarachidonic": 40112,
+      "ymetrix": 40113,
+      "Ġnect": 40114,
+      "Ġconges": 40115,
+      "ĠMerr": 40116,
+      "ĠMai": 40117,
+      "Chain": 40118,
+      "Ġretrieving": 40119,
+      "Collection": 40120,
+      "ĠMTX": 40121,
+      "ĠFernando": 40122,
+      "hg": 40123,
+      "ĠRams": 40124,
+      "thresh": 40125,
+      "apsules": 40126,
+      "Ġconduit": 40127,
+      "swap": 40128,
+      "Ġblowing": 40129,
+      "ĠNyquist": 40130,
+      "Ġunconscious": 40131,
+      "ĠDIFFERENT": 40132,
+      "Techn": 40133,
+      "hiz": 40134,
+      "îĤ": 40135,
+      "Ġdξ": 40136,
+      "ĠSto": 40137,
+      "ĠFlavon": 40138,
+      "David": 40139,
+      "Ġfiltrate": 40140,
+      "lith": 40141,
+      "ĠWool": 40142,
+      "ĠKnot": 40143,
+      "Ġhalide": 40144,
+      "Ġbioassay": 40145,
+      "ĠGoldberg": 40146,
+      "ĠTrichoderma": 40147,
+      "Ġintraspecific": 40148,
+      "crystall": 40149,
+      "ĠRend": 40150,
+      "ourg": 40151,
+      "Ġundertake": 40152,
+      "ĠEnum": 40153,
+      "infect": 40154,
+      "Ġmidgut": 40155,
+      "attack": 40156,
+      "ĠCircle": 40157,
+      "Ġpleiotropic": 40158,
+      "escent": 40159,
+      "ĠFri": 40160,
+      "philis": 40161,
+      "astings": 40162,
+      "Ġbiogas": 40163,
+      "ĠÄľ": 40164,
+      "Ġaccompany": 40165,
+      "Ġrolled": 40166,
+      "Ġchirp": 40167,
+      "Ġsomatostatin": 40168,
+      "varkappa": 40169,
+      "Scal": 40170,
+      "Ġdrow": 40171,
+      "romed": 40172,
+      "ĠLup": 40173,
+      "ĠLuminosity": 40174,
+      "ĠNig": 40175,
+      "ferromagnetic": 40176,
+      "ĠToy": 40177,
+      "Ġcannabinoid": 40178,
+      "ĠHOX": 40179,
+      "iele": 40180,
+      "ĠCTX": 40181,
+      "Ġhydrop": 40182,
+      "Ġfavorite": 40183,
+      "Ġstretches": 40184,
+      "evaluated": 40185,
+      "ogroups": 40186,
+      "acal": 40187,
+      "ollo": 40188,
+      "Ġgenders": 40189,
+      "ĠGraft": 40190,
+      "Ġincidences": 40191,
+      "Ġreplacements": 40192,
+      "ĠTRUNC": 40193,
+      "CRF": 40194,
+      "Ġequalization": 40195,
+      "ĠRenew": 40196,
+      "Ġplethora": 40197,
+      "ĠEncoder": 40198,
+      "Mit": 40199,
+      "Ġcaches": 40200,
+      "orate": 40201,
+      "endors": 40202,
+      "ĠCaution": 40203,
+      "ĠAbel": 40204,
+      "compression": 40205,
+      "ĠLarsen": 40206,
+      "ĠElimination": 40207,
+      "Ġtester": 40208,
+      "Ġninth": 40209,
+      "ĠLö": 40210,
+      "Ġspiders": 40211,
+      "Ġpoem": 40212,
+      "Ġeducators": 40213,
+      "ĠEnhances": 40214,
+      "destructive": 40215,
+      "Fourier": 40216,
+      "Ġseismicity": 40217,
+      "ĠYunnan": 40218,
+      "Riemannian": 40219,
+      "WID": 40220,
+      "vular": 40221,
+      "ĠBorder": 40222,
+      "Ġcombin": 40223,
+      "singlet": 40224,
+      "ĠEddington": 40225,
+      "ĠTemplate": 40226,
+      "ĠPAX": 40227,
+      "Ġbasalts": 40228,
+      "Enh": 40229,
+      "Ġassistants": 40230,
+      "ĠCascade": 40231,
+      "Ġinbreeding": 40232,
+      "chini": 40233,
+      "Ġupgraded": 40234,
+      "ĠTransit": 40235,
+      "survival": 40236,
+      "Ġinjector": 40237,
+      "ĠPascal": 40238,
+      "DEVICE": 40239,
+      "Ġfost": 40240,
+      "ĠKand": 40241,
+      "Ġextragalactic": 40242,
+      "ependently": 40243,
+      "Ġexcite": 40244,
+      "Ġfulfil": 40245,
+      "Ġriparian": 40246,
+      "Ġuploaded": 40247,
+      "aun": 40248,
+      "lod": 40249,
+      "saving": 40250,
+      "ĠHib": 40251,
+      "ĠEra": 40252,
+      "obese": 40253,
+      "Ġui": 40254,
+      "Ġspectrally": 40255,
+      "keV": 40256,
+      "xxx": 40257,
+      "ĠOtto": 40258,
+      "Ġétale": 40259,
+      "LAT": 40260,
+      "dermal": 40261,
+      "diaz": 40262,
+      "ĠPli": 40263,
+      "Ġlegume": 40264,
+      "Ġinspect": 40265,
+      "Ġthymic": 40266,
+      "ĠHormone": 40267,
+      "áĢ": 40268,
+      "inot": 40269,
+      "ĠShib": 40270,
+      "ĠBCC": 40271,
+      "ĠVital": 40272,
+      "Ġprofits": 40273,
+      "ĠFederated": 40274,
+      "Ġflipped": 40275,
+      "Ġproprietary": 40276,
+      "incorporated": 40277,
+      "Ġbacteremia": 40278,
+      "Ġáŀĩ": 40279,
+      "fins": 40280,
+      "ä½": 40281,
+      "esia": 40282,
+      "ĠHollow": 40283,
+      "geons": 40284,
+      "Ġtrehalose": 40285,
+      "ERO": 40286,
+      "osterol": 40287,
+      "omus": 40288,
+      "ĠCrystall": 40289,
+      "Ġcuration": 40290,
+      "Ġmagnon": 40291,
+      "ĠAmend": 40292,
+      "Ġharb": 40293,
+      "Ġneutrality": 40294,
+      "ĠDelphi": 40295,
+      "Ġnonsense": 40296,
+      "ĠHomeostasis": 40297,
+      "Ġexpenditures": 40298,
+      "Sequential": 40299,
+      "imodular": 40300,
+      "Ġzenith": 40301,
+      "ĠMoran": 40302,
+      "Ġbootstrapping": 40303,
+      "iomy": 40304,
+      "lactic": 40305,
+      "iture": 40306,
+      "Ġnat": 40307,
+      "Ġgab": 40308,
+      "Ġchat": 40309,
+      "regional": 40310,
+      "Ġcrashes": 40311,
+      "ĠAFB": 40312,
+      "Ġcrowded": 40313,
+      "Ġtweet": 40314,
+      "engineered": 40315,
+      "ĠCharged": 40316,
+      "Sche": 40317,
+      "ITIONS": 40318,
+      "ĠCoral": 40319,
+      "ĠEli": 40320,
+      "Ġinverting": 40321,
+      "Ġpedag": 40322,
+      "ĠSanders": 40323,
+      "Meanwhile": 40324,
+      "ĠGriffiths": 40325,
+      "PSCs": 40326,
+      "tize": 40327,
+      "ĠMail": 40328,
+      "Ġundec": 40329,
+      "Ġhermitian": 40330,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 40331,
+      "ĠExplos": 40332,
+      "Ġwestward": 40333,
+      "ĠConfirm": 40334,
+      "Begin": 40335,
+      "Ġfactories": 40336,
+      "ĠPRL": 40337,
+      "shear": 40338,
+      "Header": 40339,
+      "ĠFLAGS": 40340,
+      "anomal": 40341,
+      "ĠQW": 40342,
+      "ĠÌħ": 40343,
+      "oinositi": 40344,
+      "Ġmammography": 40345,
+      "Ġdepositional": 40346,
+      "EXP": 40347,
+      "residue": 40348,
+      "Ġunsatisfactory": 40349,
+      "Aβ": 40350,
+      "MUX": 40351,
+      "Ġstaged": 40352,
+      "ĠMMT": 40353,
+      "ĠKus": 40354,
+      "llo": 40355,
+      "Ġtrainer": 40356,
+      "adden": 40357,
+      "Ġpinch": 40358,
+      "WARE": 40359,
+      "Ġcabinet": 40360,
+      "CSP": 40361,
+      "ecum": 40362,
+      "oteric": 40363,
+      "ĠHav": 40364,
+      "Ġresume": 40365,
+      "Ġnetworked": 40366,
+      "share": 40367,
+      "ĠColle": 40368,
+      "Ġchemotactic": 40369,
+      "ĠGlyc": 40370,
+      "olkit": 40371,
+      "Ġbotulinum": 40372,
+      "ĠNeighborhood": 40373,
+      "mV": 40374,
+      "ĠHQ": 40375,
+      "efaciens": 40376,
+      "gett": 40377,
+      "Ġgeost": 40378,
+      "ĠCDW": 40379,
+      "Ģ̇": 40380,
+      "Ġfloors": 40381,
+      "representing": 40382,
+      "odiode": 40383,
+      "ĠInstance": 40384,
+      "Ġmonodis": 40385,
+      "drying": 40386,
+      "reasing": 40387,
+      "igi": 40388,
+      "Ġgout": 40389,
+      "ĠIEC": 40390,
+      "Ġflush": 40391,
+      "Ġtraded": 40392,
+      "Review": 40393,
+      "ĠïĤ¢": 40394,
+      "Ġà¤": 40395,
+      "Ġabbreviations": 40396,
+      "otherapies": 40397,
+      "Ġindeterminate": 40398,
+      "Ġglutaraldehyde": 40399,
+      "Ġattrition": 40400,
+      "jump": 40401,
+      "inde": 40402,
+      "ĠGri": 40403,
+      "arction": 40404,
+      "TRAIN": 40405,
+      "Ġescaped": 40406,
+      "atement": 40407,
+      "ĠPam": 40408,
+      "ĠGAM": 40409,
+      "productive": 40410,
+      "ĠAmericas": 40411,
+      "agenesis": 40412,
+      "ĠMixtures": 40413,
+      "ĠHooft": 40414,
+      "ĠWindow": 40415,
+      "Ġnodular": 40416,
+      "Ġechin": 40417,
+      "DOF": 40418,
+      "ĠDDT": 40419,
+      "electrical": 40420,
+      "ĠDecentralized": 40421,
+      "Ġcontradict": 40422,
+      "French": 40423,
+      "Ġaustr": 40424,
+      "ĠAPD": 40425,
+      "ĠDIM": 40426,
+      "ĠSten": 40427,
+      "ronomic": 40428,
+      "ĠPolymorphism": 40429,
+      "Ġcocc": 40430,
+      "itings": 40431,
+      "Ġsubcritical": 40432,
+      "ĠUniqueness": 40433,
+      "OPEN": 40434,
+      "rotoxicity": 40435,
+      "GenBank": 40436,
+      "atabases": 40437,
+      "Nets": 40438,
+      "uistic": 40439,
+      "yric": 40440,
+      "ĠSID": 40441,
+      "Ġcooked": 40442,
+      "ĠJudge": 40443,
+      "Ġparameterizations": 40444,
+      "Ġenumerated": 40445,
+      "ĠAsthma": 40446,
+      "Develop": 40447,
+      "Ġcake": 40448,
+      "ĠAges": 40449,
+      "venile": 40450,
+      "Ġflor": 40451,
+      "Ġcouldn": 40452,
+      "detach": 40453,
+      "Ġpipette": 40454,
+      "ĠInstant": 40455,
+      "Ġtentatively": 40456,
+      "ĠINTEGR": 40457,
+      "HQ": 40458,
+      "Mapping": 40459,
+      "cq": 40460,
+      "åĪ": 40461,
+      "SUM": 40462,
+      "fractions": 40463,
+      "ĠClaud": 40464,
+      "Formula": 40465,
+      "Axis": 40466,
+      "ĠBilly": 40467,
+      "ĠMethane": 40468,
+      "ĠIGM": 40469,
+      "cannot": 40470,
+      "س": 40471,
+      "Ġciting": 40472,
+      "ĠDynam": 40473,
+      "Ġlett": 40474,
+      "egler": 40475,
+      "ĠPhysicians": 40476,
+      "xFF": 40477,
+      "Ġoyster": 40478,
+      "ĠTOC": 40479,
+      "Ġsubarachnoid": 40480,
+      "ĠCOM": 40481,
+      "ITER": 40482,
+      "Ġbenzodiazep": 40483,
+      "Ġuncomplicated": 40484,
+      "tillo": 40485,
+      "Carbon": 40486,
+      "atem": 40487,
+      "Ġsel": 40488,
+      "ingo": 40489,
+      "IVITY": 40490,
+      "Ġcavern": 40491,
+      "Ġspacelike": 40492,
+      "ĠCollisions": 40493,
+      "straint": 40494,
+      "Ġmycobacterial": 40495,
+      "Ġtrachomatis": 40496,
+      "Ai": 40497,
+      "mf": 40498,
+      "ĠTric": 40499,
+      "tico": 40500,
+      "ĠElection": 40501,
+      "ĠKDM": 40502,
+      "ĠExosomes": 40503,
+      "afluor": 40504,
+      "Ġformalized": 40505,
+      "ĠELF": 40506,
+      "vphantom": 40507,
+      "ĠSME": 40508,
+      "ichuan": 40509,
+      "ĠVMs": 40510,
+      "Ġrostral": 40511,
+      "ofer": 40512,
+      "ramanian": 40513,
+      "intercal": 40514,
+      "Merck": 40515,
+      "ĠFerguson": 40516,
+      "HU": 40517,
+      "lj": 40518,
+      "Ġrack": 40519,
+      "Ġmicrograph": 40520,
+      "CTS": 40521,
+      "Ġpassively": 40522,
+      "ĠMasses": 40523,
+      "rangians": 40524,
+      "ĠADM": 40525,
+      "ĠProvided": 40526,
+      "ĠVeterans": 40527,
+      "sound": 40528,
+      "gex": 40529,
+      "ĠSpiral": 40530,
+      "Ġfossa": 40531,
+      "Ġthermodynamically": 40532,
+      "ĠStaining": 40533,
+      "Ġkar": 40534,
+      "eflon": 40535,
+      "ĠBruns": 40536,
+      "VAE": 40537,
+      "olyticus": 40538,
+      "Ġintranasal": 40539,
+      "ĠProspects": 40540,
+      "athers": 40541,
+      "Ġnumbering": 40542,
+      "ĠReplacement": 40543,
+      "ĠNoncommutative": 40544,
+      "quisitions": 40545,
+      "ĠSIMD": 40546,
+      "ĠArterial": 40547,
+      "ĠHGF": 40548,
+      "ĠGPP": 40549,
+      "Ġfluvial": 40550,
+      "neri": 40551,
+      "ĠCompressed": 40552,
+      "VIDIA": 40553,
+      "Ġshocked": 40554,
+      "dys": 40555,
+      "invariance": 40556,
+      "enstein": 40557,
+      "ĠSCM": 40558,
+      "ĠDod": 40559,
+      "Ġsho": 40560,
+      "Chlor": 40561,
+      "duino": 40562,
+      "Ġurgently": 40563,
+      "soc": 40564,
+      "etching": 40565,
+      "Ġdiffractive": 40566,
+      "ĠZF": 40567,
+      "Ġhyperplanes": 40568,
+      "Ġmyri": 40569,
+      "Ġferromagnetism": 40570,
+      "filament": 40571,
+      "Ġjustifies": 40572,
+      "fault": 40573,
+      "ĠHTS": 40574,
+      "ĠEPC": 40575,
+      "too": 40576,
+      "ĠTRAP": 40577,
+      "ión": 40578,
+      "rv": 40579,
+      "ĠBPD": 40580,
+      "ĠNod": 40581,
+      "posit": 40582,
+      "ĠConvers": 40583,
+      "Ġzeroes": 40584,
+      "ĠGlen": 40585,
+      "ĠDisturb": 40586,
+      "Ġtableau": 40587,
+      "Ġpseudot": 40588,
+      "ĠCollider": 40589,
+      "Ġadsorbents": 40590,
+      "ĠGrove": 40591,
+      "Ġkingdom": 40592,
+      "Est": 40593,
+      "Xs": 40594,
+      "czyk": 40595,
+      "ĠWille": 40596,
+      "ĠVOL": 40597,
+      "scar": 40598,
+      "ĠAdler": 40599,
+      "ĠOrchestra": 40600,
+      "Ġsparsely": 40601,
+      "glycosylation": 40602,
+      "Lac": 40603,
+      "otions": 40604,
+      "ĠIle": 40605,
+      "Ġbeacon": 40606,
+      "ĠRn": 40607,
+      "ullah": 40608,
+      "Ġtimelike": 40609,
+      "ĠForests": 40610,
+      "Ġupload": 40611,
+      "jit": 40612,
+      "ĠEDM": 40613,
+      "Ġtransplants": 40614,
+      "marker": 40615,
+      "ĠBreeding": 40616,
+      "ÎĶÎĶ": 40617,
+      "Ġfavorably": 40618,
+      "ĠTransformations": 40619,
+      "abeled": 40620,
+      "ĠPolitics": 40621,
+      "episode": 40622,
+      "Ġfut": 40623,
+      "Ġdithi": 40624,
+      "ĠMw": 40625,
+      "Ġtranspiration": 40626,
+      "Ġunlimited": 40627,
+      "ĠAntiv": 40628,
+      "PPV": 40629,
+      "Ġnomogram": 40630,
+      "Ġinvented": 40631,
+      "ĠSchedule": 40632,
+      "allows": 40633,
+      "Ġtransvers": 40634,
+      "Ġworkpiece": 40635,
+      "blacksquare": 40636,
+      "callback": 40637,
+      "ĠAthletic": 40638,
+      "hans": 40639,
+      "poles": 40640,
+      "Ġeavesdrop": 40641,
+      "ĠCone": 40642,
+      "oclim": 40643,
+      "ĠGhost": 40644,
+      "iterations": 40645,
+      "Ġweaken": 40646,
+      "Ġalkaloid": 40647,
+      "Ġveterans": 40648,
+      "Ġprostatectomy": 40649,
+      "Ġbog": 40650,
+      "ĠCed": 40651,
+      "ĠFever": 40652,
+      "ylan": 40653,
+      "archive": 40654,
+      "Ġattackers": 40655,
+      "Making": 40656,
+      "bane": 40657,
+      "ĠPull": 40658,
+      "ĠFLO": 40659,
+      "Ġcoaches": 40660,
+      "ĠVSM": 40661,
+      "okh": 40662,
+      "ĠSpo": 40663,
+      "amilial": 40664,
+      "principle": 40665,
+      "Ġaggressiveness": 40666,
+      "Ġgardens": 40667,
+      "SIG": 40668,
+      "Ġmbar": 40669,
+      ".....": 40670,
+      "Ġoptimizes": 40671,
+      "Ġmorphologic": 40672,
+      "hani": 40673,
+      "Ġgermanium": 40674,
+      "Enabled": 40675,
+      "wb": 40676,
+      "Ġforamen": 40677,
+      "ĠSPA": 40678,
+      "Ġmagnified": 40679,
+      "ĠSlater": 40680,
+      "ĠSyrian": 40681,
+      "Ġtert": 40682,
+      "Ġtraditions": 40683,
+      "Ġoffensive": 40684,
+      "Ġhydrology": 40685,
+      "ergetics": 40686,
+      "Phot": 40687,
+      "Ġperovskites": 40688,
+      "Ġwavenumbers": 40689,
+      "Ġosteoclasts": 40690,
+      "imedean": 40691,
+      "ĠBasketball": 40692,
+      "benzodiox": 40693,
+      "ĠTRUNCATED": 40694,
+      "Ġbishop": 40695,
+      "ĠSgr": 40696,
+      "ĠSatisfaction": 40697,
+      "agnostic": 40698,
+      "numeric": 40699,
+      "Ġnormals": 40700,
+      "Ġhumor": 40701,
+      "Ġcontinents": 40702,
+      "NATION": 40703,
+      "Lagrangian": 40704,
+      "Ġknees": 40705,
+      "ĠIDE": 40706,
+      "adas": 40707,
+      "adia": 40708,
+      "ĠOU": 40709,
+      "onds": 40710,
+      "ĠChaud": 40711,
+      "Ġslicing": 40712,
+      "ĠActor": 40713,
+      "Alt": 40714,
+      "Ġbroadcasts": 40715,
+      "osaurs": 40716,
+      "Ġpickle": 40717,
+      "Ġunfamiliar": 40718,
+      "allus": 40719,
+      "Ġhashing": 40720,
+      "incidence": 40721,
+      "Ġmetabolized": 40722,
+      "Ġhomogeneously": 40723,
+      "ĠFalcon": 40724,
+      "ĠÑģ": 40725,
+      "ĠCere": 40726,
+      "ĠCLA": 40727,
+      "ĠPaste": 40728,
+      "ĠFry": 40729,
+      "ĠDre": 40730,
+      "adult": 40731,
+      "Ġdiscounted": 40732,
+      "sensitized": 40733,
+      "erculous": 40734,
+      "ĠPixel": 40735,
+      "ĠBram": 40736,
+      "allo": 40737,
+      "ipative": 40738,
+      "ipality": 40739,
+      "ĠStrict": 40740,
+      "ĠTrinity": 40741,
+      "ĠClassifiers": 40742,
+      "ĠBasel": 40743,
+      "ĠCurcumin": 40744,
+      "ĠLUMO": 40745,
+      "Ġmediastinal": 40746,
+      "ĠFFA": 40747,
+      "Ġplenty": 40748,
+      "prised": 40749,
+      "Ġprinter": 40750,
+      "Ġcalcare": 40751,
+      "insn": 40752,
+      "ontology": 40753,
+      "Ġgrounding": 40754,
+      "ĠALDH": 40755,
+      "STRING": 40756,
+      "ĠFemales": 40757,
+      "ĠFocusing": 40758,
+      "assessment": 40759,
+      "ĠBluetooth": 40760,
+      "ëĬĶ": 40761,
+      "Ġego": 40762,
+      "ĠDAC": 40763,
+      "onding": 40764,
+      "randa": 40765,
+      "ĠLudwig": 40766,
+      "Ġantecedent": 40767,
+      "ĠErnst": 40768,
+      "dX": 40769,
+      "odeling": 40770,
+      "âĢĭ": 40771,
+      "Inser": 40772,
+      "ognormal": 40773,
+      "ĠTevatron": 40774,
+      "Ġcovariances": 40775,
+      "riging": 40776,
+      "ĠMgSO": 40777,
+      "carbonitrile": 40778,
+      "ĠLoren": 40779,
+      "Ġpolytopes": 40780,
+      "ĠParental": 40781,
+      "Ġunhealthy": 40782,
+      "itherto": 40783,
+      "ĠMotif": 40784,
+      "DataType": 40785,
+      "ĠMIPS": 40786,
+      "ĠPhosphorus": 40787,
+      "MoO": 40788,
+      "ĠPerturbations": 40789,
+      "Ġaphids": 40790,
+      "Ġanhydride": 40791,
+      "ideration": 40792,
+      "ĠMits": 40793,
+      "gravit": 40794,
+      "Ġdestinations": 40795,
+      "Commun": 40796,
+      "Ġtetrahedron": 40797,
+      "implicit": 40798,
+      "Ġassort": 40799,
+      "ĠSubt": 40800,
+      "ĠAcetyl": 40801,
+      "ecium": 40802,
+      "ĠNie": 40803,
+      "Ġoperand": 40804,
+      "ĠScher": 40805,
+      "azoles": 40806,
+      "tlement": 40807,
+      "ĠBlocking": 40808,
+      "Ġbottlenecks": 40809,
+      "ĠOccupational": 40810,
+      "HAS": 40811,
+      "Teller": 40812,
+      "Ġvague": 40813,
+      "esting": 40814,
+      "SNE": 40815,
+      "Ġphotoionization": 40816,
+      "Ġcomplaint": 40817,
+      "uspid": 40818,
+      "owler": 40819,
+      "Ġendocytic": 40820,
+      "Ġflocks": 40821,
+      "epsin": 40822,
+      "colors": 40823,
+      "otopes": 40824,
+      "ĠDepletion": 40825,
+      "ELLAR": 40826,
+      "armed": 40827,
+      "ĠTIR": 40828,
+      "Ġbullying": 40829,
+      "Ġâݧ": 40830,
+      "osporidium": 40831,
+      "Mr": 40832,
+      "ĠCic": 40833,
+      "ogal": 40834,
+      "Ġsectioned": 40835,
+      "Chapter": 40836,
+      "ĠContents": 40837,
+      "ĠPaths": 40838,
+      "ĠExplain": 40839,
+      "computing": 40840,
+      "Ġshrub": 40841,
+      "ĠMalaysian": 40842,
+      "Beta": 40843,
+      "Mad": 40844,
+      "Ros": 40845,
+      "Ġeyel": 40846,
+      "ĠACF": 40847,
+      "ĠMm": 40848,
+      "texture": 40849,
+      "Ġinterpretability": 40850,
+      "ĠTopic": 40851,
+      "Ġbadly": 40852,
+      "ĠmAh": 40853,
+      "Eg": 40854,
+      "RQ": 40855,
+      "pins": 40856,
+      "etc": 40857,
+      "ĠVogel": 40858,
+      "Ġhypoc": 40859,
+      "Ġrunaway": 40860,
+      "Ġpersonally": 40861,
+      "Ġbinders": 40862,
+      "sensory": 40863,
+      "ĠIPv": 40864,
+      "ranked": 40865,
+      "Ġfibrations": 40866,
+      "Ġstrawberry": 40867,
+      "arotomy": 40868,
+      "FLI": 40869,
+      "rator": 40870,
+      "odys": 40871,
+      "iran": 40872,
+      "ĠBead": 40873,
+      "ĠDAM": 40874,
+      "âĪĥ": 40875,
+      "Ġillusion": 40876,
+      "pidium": 40877,
+      "Place": 40878,
+      "Region": 40879,
+      "Ġallocations": 40880,
+      "Ġohmic": 40881,
+      "Ġnf": 40882,
+      "imino": 40883,
+      "ĠBris": 40884,
+      "itizing": 40885,
+      "proper": 40886,
+      "subgroup": 40887,
+      "Ġsalience": 40888,
+      "rangement": 40889,
+      "ĠMeaning": 40890,
+      "Ġbarcode": 40891,
+      "Ġneuropeptide": 40892,
+      "Ġendosperm": 40893,
+      "imab": 40894,
+      "Ġnanod": 40895,
+      "ĠMetz": 40896,
+      "Ġcollocation": 40897,
+      "ĠInfected": 40898,
+      "Ġpackaged": 40899,
+      "ĠADA": 40900,
+      "ĠBarth": 40901,
+      "ĠCNC": 40902,
+      "Ġcascaded": 40903,
+      "ĠStockholm": 40904,
+      "itas": 40905,
+      "ĠMMR": 40906,
+      "ĠDrought": 40907,
+      "Ġpermissible": 40908,
+      "Ġsciatic": 40909,
+      "Ġfringes": 40910,
+      "Ġexecutable": 40911,
+      "Ġstemness": 40912,
+      "ĠEndoscopic": 40913,
+      "aporin": 40914,
+      "TOP": 40915,
+      "eB": 40916,
+      "tur": 40917,
+      "ĠStages": 40918,
+      "anches": 40919,
+      "Ġnonperturbative": 40920,
+      "Ġmaritime": 40921,
+      "Ġcoverslips": 40922,
+      "Ġlagoon": 40923,
+      "Experiments": 40924,
+      "Ġcodewords": 40925,
+      "Encoder": 40926,
+      "das": 40927,
+      "prac": 40928,
+      "Ġpaddy": 40929,
+      "Ġdraining": 40930,
+      "Ġkids": 40931,
+      "Ġenemies": 40932,
+      "Ġmotile": 40933,
+      "Ġslack": 40934,
+      "bees": 40935,
+      "ĠSuppl": 40936,
+      "ĠBarber": 40937,
+      "ĠSPH": 40938,
+      "Ġcrystallite": 40939,
+      "fitted": 40940,
+      "cyclopent": 40941,
+      "ĠRMSD": 40942,
+      "Ġparkinson": 40943,
+      "Ġuncorrected": 40944,
+      "ĠSyntax": 40945,
+      "Ġmultinomial": 40946,
+      "ĠIncorporating": 40947,
+      "akrishnan": 40948,
+      "JL": 40949,
+      "NESS": 40950,
+      "mim": 40951,
+      "ĠTET": 40952,
+      "ĠPorph": 40953,
+      "ĠSchwe": 40954,
+      "Ġcatalogs": 40955,
+      "ĠAuthentication": 40956,
+      "Bro": 40957,
+      "ugar": 40958,
+      "ĠAmpl": 40959,
+      "Ġsapiens": 40960,
+      "ĠGANs": 40961,
+      "Ġnecessitates": 40962,
+      "tg": 40963,
+      "edal": 40964,
+      "ĠRear": 40965,
+      "opeptidase": 40966,
+      "ĠInformed": 40967,
+      "Ġtailor": 40968,
+      "ĠNNLO": 40969,
+      "Ġhemodynamics": 40970,
+      "Sy": 40971,
+      "dating": 40972,
+      "achers": 40973,
+      "ĠTodd": 40974,
+      "ĠMario": 40975,
+      "ĠUGT": 40976,
+      "ĠValent": 40977,
+      "Ġstreamlines": 40978,
+      "Ġwarrants": 40979,
+      "Ġrew": 40980,
+      "ĠMud": 40981,
+      "ĠGK": 40982,
+      "Ġcoke": 40983,
+      "ĠUran": 40984,
+      "Ġgrooves": 40985,
+      "ronate": 40986,
+      "ĠRadius": 40987,
+      "ĠSuite": 40988,
+      "atumoral": 40989,
+      "Workspace": 40990,
+      "ĠSynergistic": 40991,
+      "ĠAtherosclerosis": 40992,
+      "maz": 40993,
+      "argmax": 40994,
+      "shield": 40995,
+      "ontin": 40996,
+      "Ġlistener": 40997,
+      "ocytoma": 40998,
+      "ĠGrav": 40999,
+      "ĠJerusalem": 41000,
+      "pyrrolidin": 41001,
+      "ĠSprings": 41002,
+      "Ġseafloor": 41003,
+      "Ġdips": 41004,
+      "istani": 41005,
+      "obis": 41006,
+      "Ġphotodynamic": 41007,
+      "ADO": 41008,
+      "Ġionisation": 41009,
+      "Ġbarn": 41010,
+      "igenetics": 41011,
+      "Ġeconomies": 41012,
+      "ĠGlacier": 41013,
+      "Ġç": 41014,
+      "imes": 41015,
+      "ĠSasaki": 41016,
+      "chio": 41017,
+      "Ġassisting": 41018,
+      "ostin": 41019,
+      "Ġindiff": 41020,
+      "ĠShot": 41021,
+      "ĠNeuron": 41022,
+      "CDD": 41023,
+      "ĠCONST": 41024,
+      "ĠBSs": 41025,
+      "tas": 41026,
+      "association": 41027,
+      "parg": 41028,
+      "Ġescal": 41029,
+      "exercise": 41030,
+      "ĠAdela": 41031,
+      "Ġmyogenic": 41032,
+      "ĠNOAA": 41033,
+      "yclo": 41034,
+      "linal": 41035,
+      "ĠHut": 41036,
+      "Ġintroductory": 41037,
+      "Ġheterochromatin": 41038,
+      "Ġchemoresistance": 41039,
+      "Ġsimplifications": 41040,
+      "pyridin": 41041,
+      "Ġamyloidosis": 41042,
+      "Ġnanofiber": 41043,
+      "ĠSutton": 41044,
+      "ĠResilience": 41045,
+      "Parent": 41046,
+      "ĠLMS": 41047,
+      "Ġlets": 41048,
+      "ĠElderly": 41049,
+      "Ġirrevers": 41050,
+      "sheets": 41051,
+      "Effects": 41052,
+      "Ġellipses": 41053,
+      "ĠPhilosophy": 41054,
+      "Ġphotographic": 41055,
+      "HEAD": 41056,
+      "Ġreversibility": 41057,
+      "Ġfederated": 41058,
+      "ĠPhosphoserine": 41059,
+      "estimation": 41060,
+      "ĠHumph": 41061,
+      "Json": 41062,
+      "Ġfills": 41063,
+      "Ġverm": 41064,
+      "ĠSeif": 41065,
+      "withstanding": 41066,
+      "ĠYamada": 41067,
+      "eria": 41068,
+      "ĠFLA": 41069,
+      "ĠVick": 41070,
+      "toid": 41071,
+      "annier": 41072,
+      "Ġcancerous": 41073,
+      "PRINT": 41074,
+      "ĠMechanistic": 41075,
+      "Ġdusty": 41076,
+      "ĠAppend": 41077,
+      "ycin": 41078,
+      "Ġazo": 41079,
+      "Ġsizing": 41080,
+      "Ġcrayfish": 41081,
+      "avis": 41082,
+      "ĠAdvent": 41083,
+      "ĠCommunist": 41084,
+      "ĠIMU": 41085,
+      "pixels": 41086,
+      "Hall": 41087,
+      "past": 41088,
+      "ĠRous": 41089,
+      "ressional": 41090,
+      "aird": 41091,
+      "ĠADD": 41092,
+      "Ġsummarizing": 41093,
+      "Ġelectrol": 41094,
+      "Station": 41095,
+      "ĠLyα": 41096,
+      "ĠTMEM": 41097,
+      "Ġpeptidase": 41098,
+      "Dual": 41099,
+      "git": 41100,
+      "ĠBOD": 41101,
+      "ĠRham": 41102,
+      "ĠKak": 41103,
+      "Ġreadiness": 41104,
+      "ĠCompare": 41105,
+      "ĠRamos": 41106,
+      "significantly": 41107,
+      "Ġhairy": 41108,
+      "Ġvasopressin": 41109,
+      "ĠGuideline": 41110,
+      "BNP": 41111,
+      "Ġdirty": 41112,
+      "Ġinfimum": 41113,
+      "ĠAless": 41114,
+      "ĠVolcano": 41115,
+      "Magn": 41116,
+      "YY": 41117,
+      "ughlin": 41118,
+      "boplatin": 41119,
+      "ĠCantor": 41120,
+      "Ġclothes": 41121,
+      "ĠRw": 41122,
+      "Ġuseless": 41123,
+      "ĠKdV": 41124,
+      "operoxidase": 41125,
+      "ĠCorrect": 41126,
+      "Ġfatality": 41127,
+      "ĠRestriction": 41128,
+      "Computer": 41129,
+      "Department": 41130,
+      "Il": 41131,
+      "gang": 41132,
+      "ĠElectroc": 41133,
+      "obaric": 41134,
+      "Phen": 41135,
+      "Ġned": 41136,
+      "adh": 41137,
+      "issing": 41138,
+      "tooth": 41139,
+      "Ġmanuscripts": 41140,
+      "Ġbiotechnology": 41141,
+      "Supp": 41142,
+      "ĠPairwise": 41143,
+      "ĠParsing": 41144,
+      "dH": 41145,
+      "melt": 41146,
+      "rz": 41147,
+      "ĠCatalyst": 41148,
+      "emption": 41149,
+      "Ġshowers": 41150,
+      "BLEM": 41151,
+      "ĠBrothers": 41152,
+      "banon": 41153,
+      "Ġbrachy": 41154,
+      "metallicity": 41155,
+      "ĠCIS": 41156,
+      "ĠCopenhagen": 41157,
+      "Ġelder": 41158,
+      "Ġfinanc": 41159,
+      "odesic": 41160,
+      "Ġdevise": 41161,
+      "Ġsurvives": 41162,
+      "ĠðtÃŀ": 41163,
+      "Ġfascinating": 41164,
+      "Ġparallax": 41165,
+      "HOR": 41166,
+      "yth": 41167,
+      "onins": 41168,
+      "ĠEHR": 41169,
+      "ĠGates": 41170,
+      "obase": 41171,
+      "ĠConway": 41172,
+      "operations": 41173,
+      "manuel": 41174,
+      "ĠAbdominal": 41175,
+      "ĠARG": 41176,
+      "ĠGrö": 41177,
+      "Ġphotosens": 41178,
+      "ĠCoulter": 41179,
+      "ĠJulian": 41180,
+      "ĠLevine": 41181,
+      "Ġsarcoidosis": 41182,
+      "Ġpillars": 41183,
+      "ĠdR": 41184,
+      "ĠWG": 41185,
+      "Ġspeculation": 41186,
+      "anski": 41187,
+      "ĠGaussians": 41188,
+      "Schw": 41189,
+      "ĠNambu": 41190,
+      "parents": 41191,
+      "intrinsic": 41192,
+      "ĠKendall": 41193,
+      "ĠRg": 41194,
+      "Ġprototypical": 41195,
+      "brella": 41196,
+      "Ġtetrap": 41197,
+      "ĠPathophys": 41198,
+      "nmt": 41199,
+      "Ġergodicity": 41200,
+      "ĠYersinia": 41201,
+      "QO": 41202,
+      "ĠIAV": 41203,
+      "Ġchocolate": 41204,
+      "Ġconferring": 41205,
+      "ClNO": 41206,
+      "otia": 41207,
+      "Complete": 41208,
+      "ĠAMPA": 41209,
+      "ïĢŃ": 41210,
+      "Ġḡ": 41211,
+      "ĠiPSCs": 41212,
+      "ĠApparently": 41213,
+      "Ġintoxication": 41214,
+      "ĠFather": 41215,
+      "percent": 41216,
+      "Ġshaker": 41217,
+      "Ġfinancing": 41218,
+      "Ġgenitalia": 41219,
+      "members": 41220,
+      "Ġstagnation": 41221,
+      "hmatic": 41222,
+      "rored": 41223,
+      "Ġconidia": 41224,
+      "ataloader": 41225,
+      "ĠNeil": 41226,
+      "Ġliteratures": 41227,
+      "ĠGlc": 41228,
+      "ĠDevelopments": 41229,
+      "differentiation": 41230,
+      "ĠRevisited": 41231,
+      "nil": 41232,
+      "tom": 41233,
+      "diol": 41234,
+      "ĠAbell": 41235,
+      "Ġplastics": 41236,
+      "ĠLuke": 41237,
+      "adjacent": 41238,
+      "ĠBHs": 41239,
+      "ĠPositioning": 41240,
+      "ør": 41241,
+      "overexpressing": 41242,
+      "Ec": 41243,
+      "Pref": 41244,
+      "orting": 41245,
+      "Ġinch": 41246,
+      "ĠCatherine": 41247,
+      "ĠDMP": 41248,
+      "ĠOe": 41249,
+      "endothelial": 41250,
+      "ICES": 41251,
+      "ĠHadron": 41252,
+      "Ġrevisit": 41253,
+      "ĠPictures": 41254,
+      "ĠKnockdown": 41255,
+      "ilian": 41256,
+      "ĠALA": 41257,
+      "opamine": 41258,
+      "ĠLah": 41259,
+      "climate": 41260,
+      "Ġdistraction": 41261,
+      "arski": 41262,
+      "ĠAccount": 41263,
+      "reflex": 41264,
+      "Ġelongate": 41265,
+      "ĠAmbient": 41266,
+      "Cx": 41267,
+      "Machine": 41268,
+      "ĠJPEG": 41269,
+      "Ġclassifies": 41270,
+      "ĠRegulate": 41271,
+      "oplasia": 41272,
+      "injury": 41273,
+      "neighbors": 41274,
+      "ĠFORMATION": 41275,
+      "FIS": 41276,
+      "Sz": 41277,
+      "ĠOSC": 41278,
+      "Ġassembling": 41279,
+      "Ġintracerebral": 41280,
+      "supers": 41281,
+      "ĠpF": 41282,
+      "Ġheal": 41283,
+      "ĠVries": 41284,
+      "arche": 41285,
+      "ĠDecom": 41286,
+      "ĠDiffic": 41287,
+      "agenta": 41288,
+      "ĠSpirit": 41289,
+      "ĠIntersection": 41290,
+      "Ġendorsed": 41291,
+      "ĠNobel": 41292,
+      "iÏī": 41293,
+      "wu": 41294,
+      "phaly": 41295,
+      "Ġqueu": 41296,
+      "ĠForum": 41297,
+      "lander": 41298,
+      "Ġspectrometric": 41299,
+      "ĠHankel": 41300,
+      "ĠCSE": 41301,
+      "Ġresumed": 41302,
+      "ĠGRE": 41303,
+      "ACES": 41304,
+      "CTA": 41305,
+      "Ġbehaved": 41306,
+      "Monitor": 41307,
+      "ĠNikon": 41308,
+      "ĠCHARACTER": 41309,
+      "ĠSAL": 41310,
+      "ĠIch": 41311,
+      "ĠHSF": 41312,
+      "Ġgenotoxic": 41313,
+      "ificance": 41314,
+      "ĠChiang": 41315,
+      "ĠZen": 41316,
+      "ĠArrows": 41317,
+      "ĠPDT": 41318,
+      "ĠFLASH": 41319,
+      "ocortex": 41320,
+      "onstructing": 41321,
+      "Treatment": 41322,
+      "ĉĠĠĠĠĠĠ": 41323,
+      "edullary": 41324,
+      "ilty": 41325,
+      "indentation": 41326,
+      "Ġamended": 41327,
+      "Ġfled": 41328,
+      "rophication": 41329,
+      "ĠGLM": 41330,
+      "ĠOpera": 41331,
+      "HLH": 41332,
+      "Lite": 41333,
+      "bmod": 41334,
+      "Ġaversion": 41335,
+      "ĠSweet": 41336,
+      "Ġstreptavidin": 41337,
+      "ĠPairs": 41338,
+      "ugos": 41339,
+      "epoxy": 41340,
+      "Ġunspecified": 41341,
+      "Ġmicrochannel": 41342,
+      "ĠVictorian": 41343,
+      "Could": 41344,
+      "informed": 41345,
+      "Ġsits": 41346,
+      "Ġrx": 41347,
+      "Ġnep": 41348,
+      "touch": 41349,
+      "ROI": 41350,
+      "Ġheaders": 41351,
+      "flush": 41352,
+      "ĠPathogenic": 41353,
+      "Ġspacings": 41354,
+      "hetti": 41355,
+      "Ġmotivating": 41356,
+      "Ġstakeholder": 41357,
+      "ĠSymbolic": 41358,
+      "ĠCrani": 41359,
+      "Ġdispute": 41360,
+      "Ġassists": 41361,
+      "indler": 41362,
+      "ĠSpatio": 41363,
+      "ohm": 41364,
+      "Ġextrapolate": 41365,
+      "Ġelaboration": 41366,
+      "ĠGTPases": 41367,
+      "Ġcellulase": 41368,
+      "ĠTuc": 41369,
+      "olide": 41370,
+      "ĠAIM": 41371,
+      "plast": 41372,
+      "ĠBible": 41373,
+      "opoly": 41374,
+      "ubo": 41375,
+      "acean": 41376,
+      "ĠPenrose": 41377,
+      "ĠMapReduce": 41378,
+      "ĠKwon": 41379,
+      "Wall": 41380,
+      "Ġgcd": 41381,
+      "ĠArbitrary": 41382,
+      "Product": 41383,
+      "ĠGitHub": 41384,
+      "Fn": 41385,
+      "Ġck": 41386,
+      "ĠAus": 41387,
+      "Ġgrave": 41388,
+      "Ġmetabolically": 41389,
+      "Ġlisten": 41390,
+      "Ġextractions": 41391,
+      "ĠTrunc": 41392,
+      "ĠRadiology": 41393,
+      "conserving": 41394,
+      "ĠCompositional": 41395,
+      "reporting": 41396,
+      "sustain": 41397,
+      "îĢ": 41398,
+      "ĠObl": 41399,
+      "ĠkN": 41400,
+      "Stre": 41401,
+      "ĠSupergravity": 41402,
+      "ĠPVP": 41403,
+      "Ġcivilian": 41404,
+      "ĠTunnel": 41405,
+      "Ġhelicopter": 41406,
+      "ĠCambrian": 41407,
+      "Ġrg": 41408,
+      "ĠUPF": 41409,
+      "Ġpolyd": 41410,
+      "Ġobservability": 41411,
+      "container": 41412,
+      "nitros": 41413,
+      "ĠCutting": 41414,
+      "Ġdecouple": 41415,
+      "Ġcarboxy": 41416,
+      "crow": 41417,
+      "Ġcx": 41418,
+      "ĠKell": 41419,
+      "Ġprojectors": 41420,
+      "Ġmyocarditis": 41421,
+      "itoneum": 41422,
+      "conditioning": 41423,
+      "ĠETH": 41424,
+      "oyama": 41425,
+      "Ġphosphates": 41426,
+      "ĠSubjective": 41427,
+      "ĠSerre": 41428,
+      "Ġcollagenase": 41429,
+      "Ġvibrating": 41430,
+      "streptomycin": 41431,
+      "zhen": 41432,
+      "Ġcres": 41433,
+      "Ġcull": 41434,
+      "Ġhaven": 41435,
+      "ĠGPL": 41436,
+      "lessness": 41437,
+      "Ġviewpoints": 41438,
+      ",,,": 41439,
+      "ochromic": 41440,
+      "uyama": 41441,
+      "Ġpartnerships": 41442,
+      "LICENSE": 41443,
+      "ĠSIFT": 41444,
+      "resources": 41445,
+      "ĠGos": 41446,
+      "ivac": 41447,
+      "Ġneurogenic": 41448,
+      "Adj": 41449,
+      "Ġaquifers": 41450,
+      "ĠGlycos": 41451,
+      "ĠMatthews": 41452,
+      "ĠFriday": 41453,
+      "ĠpX": 41454,
+      "Ġante": 41455,
+      "ĠFenton": 41456,
+      "ĠEukary": 41457,
+      "ibal": 41458,
+      "ideae": 41459,
+      "Attention": 41460,
+      "ĠPolymerization": 41461,
+      "Ġflipping": 41462,
+      "ĠMediates": 41463,
+      "Ġstationarity": 41464,
+      "Ġechoes": 41465,
+      "alidomide": 41466,
+      "Ġdelineation": 41467,
+      "Ġnaval": 41468,
+      "ĠSomatic": 41469,
+      "Ġstub": 41470,
+      "ĠBever": 41471,
+      "ĠRiz": 41472,
+      "Ġresummation": 41473,
+      "Ġassault": 41474,
+      "Ġpreexisting": 41475,
+      "Ġhypermethylation": 41476,
+      "Ġconserving": 41477,
+      "recorded": 41478,
+      "amn": 41479,
+      "Ġchow": 41480,
+      "ĠKill": 41481,
+      "ĠProduced": 41482,
+      "Ġrefs": 41483,
+      "ĠEnzymes": 41484,
+      "Ġdeepest": 41485,
+      "&&&": 41486,
+      "ĠFRP": 41487,
+      "Ġmilieu": 41488,
+      "Ġirrigated": 41489,
+      "ĠAnatomical": 41490,
+      "Ġdissect": 41491,
+      "iliensis": 41492,
+      "razolo": 41493,
+      "ĠProbable": 41494,
+      "solve": 41495,
+      "confirmed": 41496,
+      "ohydrodynamic": 41497,
+      "library": 41498,
+      "ĠCiti": 41499,
+      "ĠPHA": 41500,
+      "itsky": 41501,
+      "Ġelectrone": 41502,
+      "naive": 41503,
+      "Ġribs": 41504,
+      "ĠPhotonic": 41505,
+      "Ġsubstantia": 41506,
+      "ĠESTIM": 41507,
+      "Ġduodenum": 41508,
+      "ĠChagas": 41509,
+      "ĠSURVEY": 41510,
+      "Press": 41511,
+      "bian": 41512,
+      "¤": 41513,
+      "hei": 41514,
+      "ĠVAR": 41515,
+      "Ġcolocalization": 41516,
+      "Ġpolyl": 41517,
+      "Ġdough": 41518,
+      "Ġmicrocontroller": 41519,
+      "Ġinternalized": 41520,
+      "Ġdischarging": 41521,
+      "ĠChlamydomonas": 41522,
+      "orad": 41523,
+      "itel": 41524,
+      "ĠWend": 41525,
+      "Ġlogits": 41526,
+      "Ġelectrocataly": 41527,
+      "ĠAmar": 41528,
+      "Ġappreciably": 41529,
+      "Ġneurotransmitters": 41530,
+      "formerly": 41531,
+      "cul": 41532,
+      "rata": 41533,
+      "ĠBalk": 41534,
+      "ĠOsm": 41535,
+      "Ġsymptomatology": 41536,
+      "ĠFIELD": 41537,
+      "ĠAPs": 41538,
+      "Ġgymn": 41539,
+      "ĠMMS": 41540,
+      "Ġrefresh": 41541,
+      "ĠIndices": 41542,
+      "Ġimplantable": 41543,
+      "shuffle": 41544,
+      "ĠHeath": 41545,
+      "Ġcrisp": 41546,
+      "Ġdissertation": 41547,
+      "ĠUlt": 41548,
+      "Description": 41549,
+      "ĠOriginally": 41550,
+      "ĠFn": 41551,
+      "ĠFLOW": 41552,
+      "ubility": 41553,
+      "Ġexams": 41554,
+      "ĠShor": 41555,
+      "ĠCdTe": 41556,
+      "psycho": 41557,
+      "Ġdictates": 41558,
+      "Ġparenchymal": 41559,
+      "ĠPretreatment": 41560,
+      "Ġremembered": 41561,
+      "Ġbras": 41562,
+      "otid": 41563,
+      "Ġrecommender": 41564,
+      "Ġflesh": 41565,
+      "pitch": 41566,
+      "inist": 41567,
+      "Ġbtitle": 41568,
+      "Ġlc": 41569,
+      "assigned": 41570,
+      "ĠAdvisory": 41571,
+      "ĠGeneva": 41572,
+      "weighting": 41573,
+      "ĠABTS": 41574,
+      "Ġhexagon": 41575,
+      "ovskite": 41576,
+      "ĠAPIs": 41577,
+      "Ġbolometric": 41578,
+      "Ġmultifaceted": 41579,
+      "iak": 41580,
+      "Ġtn": 41581,
+      "ĠBibli": 41582,
+      "prosy": 41583,
+      "ĠJama": 41584,
+      "Ġinfrastructures": 41585,
+      "ĠShare": 41586,
+      "Ġintrogression": 41587,
+      "transforms": 41588,
+      "Report": 41589,
+      "ĠTRANS": 41590,
+      "ĠEXP": 41591,
+      "ĠCBT": 41592,
+      "ĠUbiquitin": 41593,
+      "ĠThickness": 41594,
+      "pub": 41595,
+      "toxin": 41596,
+      "ĠFriction": 41597,
+      "ĠLAG": 41598,
+      "mails": 41599,
+      "ĠImmediately": 41600,
+      "Ġweakest": 41601,
+      "ĠMRS": 41602,
+      "Ġcalcareous": 41603,
+      "bath": 41604,
+      "Ġcg": 41605,
+      "urational": 41606,
+      "tero": 41607,
+      "ĠInoue": 41608,
+      "Ġinstructor": 41609,
+      "acceptor": 41610,
+      "ĠEvolving": 41611,
+      "ĠLuther": 41612,
+      "Ġresigned": 41613,
+      "ĠChond": 41614,
+      "ERF": 41615,
+      "Ġselector": 41616,
+      "Ġnewspapers": 41617,
+      "GRA": 41618,
+      "Spe": 41619,
+      "VH": 41620,
+      "rA": 41621,
+      "otine": 41622,
+      "ĠFACT": 41623,
+      "composition": 41624,
+      "riding": 41625,
+      "PCM": 41626,
+      "Ġmiddleware": 41627,
+      "ĠGRP": 41628,
+      "Phosph": 41629,
+      "ĠEPIC": 41630,
+      "speech": 41631,
+      "vortex": 41632,
+      "ĠHerschel": 41633,
+      "amis": 41634,
+      "otube": 41635,
+      "ĠGomez": 41636,
+      "comyc": 41637,
+      "ĠPhyto": 41638,
+      "ĠConserv": 41639,
+      "Ġcava": 41640,
+      "arrhyth": 41641,
+      "ĠRestricted": 41642,
+      "ilicity": 41643,
+      "ogap": 41644,
+      "CTP": 41645,
+      "ĠLatino": 41646,
+      "attenuated": 41647,
+      "mobility": 41648,
+      "anen": 41649,
+      "Ġnif": 41650,
+      "ĠVideos": 41651,
+      "ĠSchubert": 41652,
+      "Features": 41653,
+      "opropanol": 41654,
+      "ĠThirdly": 41655,
+      "atula": 41656,
+      "ĠCemetery": 41657,
+      "entist": 41658,
+      "Ġdeli": 41659,
+      "trials": 41660,
+      "Ġgranulation": 41661,
+      "TTG": 41662,
+      "Ġteleost": 41663,
+      "morill": 41664,
+      "orse": 41665,
+      "otypically": 41666,
+      "ĠAbility": 41667,
+      "Amino": 41668,
+      "aqueous": 41669,
+      "ĠpCO": 41670,
+      "econ": 41671,
+      "ĠLign": 41672,
+      "essels": 41673,
+      "Ġformulating": 41674,
+      "ĠToo": 41675,
+      "ĠTranslational": 41676,
+      "ourses": 41677,
+      "ubiquit": 41678,
+      "statistic": 41679,
+      "Ġbursting": 41680,
+      "Ġestuaries": 41681,
+      "ĠNanocom": 41682,
+      "Ġexports": 41683,
+      "Ġê": 41684,
+      "contaminated": 41685,
+      "Ġtubing": 41686,
+      "Ġautomobile": 41687,
+      "Ġmissile": 41688,
+      "Ġhierarchically": 41689,
+      "Ġrepairs": 41690,
+      "ĠImproves": 41691,
+      "ĠEFFECTS": 41692,
+      "QDs": 41693,
+      "roz": 41694,
+      "aric": 41695,
+      "Ġparsed": 41696,
+      "ĠBrink": 41697,
+      "Ġprogressing": 41698,
+      "ĠpermNeigh": 41699,
+      "Agg": 41700,
+      "ZX": 41701,
+      "sink": 41702,
+      "Ġwise": 41703,
+      "etence": 41704,
+      "ĠIc": 41705,
+      "loo": 41706,
+      "meida": 41707,
+      "Ġpolariton": 41708,
+      "ĠConnections": 41709,
+      "Ġhallmarks": 41710,
+      "Longrightarrow": 41711,
+      "Ġtheater": 41712,
+      "esar": 41713,
+      "Ġreimburs": 41714,
+      "Ġlogo": 41715,
+      "Ġexcreted": 41716,
+      "ĠNoisy": 41717,
+      "Ġleaks": 41718,
+      "ĠDaph": 41719,
+      "Ġparagraphs": 41720,
+      "Ġlandslides": 41721,
+      "Ġpreclude": 41722,
+      "Ġcoercivity": 41723,
+      "ĠBurkholderia": 41724,
+      "ĠGómez": 41725,
+      "price": 41726,
+      "Theory": 41727,
+      "surgery": 41728,
+      "fname": 41729,
+      "failure": 41730,
+      "liness": 41731,
+      "refer": 41732,
+      "rique": 41733,
+      "ĠDogs": 41734,
+      "ĠEUV": 41735,
+      "ĠVapor": 41736,
+      "CSR": 41737,
+      "Biggl": 41738,
+      "Constraint": 41739,
+      "gravitational": 41740,
+      "Ġcombinatorics": 41741,
+      "Ġarticulated": 41742,
+      "ĠBaxter": 41743,
+      "ĠUltrasonic": 41744,
+      "LTE": 41745,
+      "lop": 41746,
+      "Ġinteratomic": 41747,
+      "intuitive": 41748,
+      "simplex": 41749,
+      "Ġexperimented": 41750,
+      "organizing": 41751,
+      "ĠOsaka": 41752,
+      "hadron": 41753,
+      "Ġdendrimers": 41754,
+      "ĠElsevier": 41755,
+      "CIP": 41756,
+      "ĠBAP": 41757,
+      "ĠAlonso": 41758,
+      "artet": 41759,
+      "antis": 41760,
+      "Ġextracorporeal": 41761,
+      "Ġpowdered": 41762,
+      "ĠSettings": 41763,
+      "etallic": 41764,
+      "ĠTEC": 41765,
+      "ĠArena": 41766,
+      "Ġanod": 41767,
+      "ĠReagents": 41768,
+      "licenses": 41769,
+      "ĠRemove": 41770,
+      "Ġpronunciation": 41771,
+      "thinspace": 41772,
+      "ĠClinically": 41773,
+      "gative": 41774,
+      "Ġwage": 41775,
+      "ĠHap": 41776,
+      "ĠGrac": 41777,
+      "fft": 41778,
+      "Ġformate": 41779,
+      "infeld": 41780,
+      "ĠQuin": 41781,
+      "Ġglomerul": 41782,
+      "Way": 41783,
+      "Ġkills": 41784,
+      "Ġtransversely": 41785,
+      "variation": 41786,
+      "ennas": 41787,
+      "ĠPLL": 41788,
+      "Ġinstrumented": 41789,
+      "ĠSpark": 41790,
+      "Ġpillar": 41791,
+      "ĠCaval": 41792,
+      "aders": 41793,
+      "issen": 41794,
+      "scene": 41795,
+      "otherm": 41796,
+      "ées": 41797,
+      "Ġpracticing": 41798,
+      "ĠBMSCs": 41799,
+      "ĠFernandez": 41800,
+      "Ġbeside": 41801,
+      "few": 41802,
+      "ĠCru": 41803,
+      "Ġprod": 41804,
+      "anders": 41805,
+      "azoline": 41806,
+      "Ġlegislative": 41807,
+      "balances": 41808,
+      "URL": 41809,
+      "Ġstereotactic": 41810,
+      "Ġtribes": 41811,
+      "Ġá¹¼": 41812,
+      "ĠPANI": 41813,
+      "adreno": 41814,
+      "gotten": 41815,
+      "cranial": 41816,
+      "ĠMick": 41817,
+      "ĠMMC": 41818,
+      "adiazole": 41819,
+      "entiation": 41820,
+      "ĠGln": 41821,
+      "ĠHolstein": 41822,
+      "ĠExplorer": 41823,
+      "Ġosse": 41824,
+      "arthy": 41825,
+      "ĠEVALU": 41826,
+      "adrenaline": 41827,
+      "JJ": 41828,
+      "ĠCMA": 41829,
+      "ĠInactivation": 41830,
+      "ABS": 41831,
+      "ĠSTZ": 41832,
+      "Configuration": 41833,
+      "ĠOlfactory": 41834,
+      "ĠSulfur": 41835,
+      "symbols": 41836,
+      "ĠASTM": 41837,
+      "divergence": 41838,
+      "ĠOCR": 41839,
+      "medical": 41840,
+      "Ġviewer": 41841,
+      "Ġbombardment": 41842,
+      "fair": 41843,
+      "nice": 41844,
+      "elberg": 41845,
+      "ĠGPT": 41846,
+      "ĠKow": 41847,
+      "Ġphotosphere": 41848,
+      "Ġlabile": 41849,
+      "ĠShang": 41850,
+      "Ġhomotopic": 41851,
+      "SVD": 41852,
+      "becomes": 41853,
+      "Ġgonor": 41854,
+      "Ġdeuteron": 41855,
+      "Ġphylogenies": 41856,
+      "ĠSAF": 41857,
+      "rapment": 41858,
+      "ĠCHF": 41859,
+      "Plan": 41860,
+      "ĠLegal": 41861,
+      "ĠFredholm": 41862,
+      "Ġsharper": 41863,
+      "Ġnanorib": 41864,
+      "ĠBuffalo": 41865,
+      "BMD": 41866,
+      "Ġlg": 41867,
+      "osup": 41868,
+      "ĠOPC": 41869,
+      "Ġendophytic": 41870,
+      "ATR": 41871,
+      "ĠExpressions": 41872,
+      "ĠMusical": 41873,
+      "Introduction": 41874,
+      "ĠSLM": 41875,
+      "çois": 41876,
+      "oglycos": 41877,
+      "aglia": 41878,
+      "mussen": 41879,
+      "formans": 41880,
+      "Ġsubstructures": 41881,
+      "ympan": 41882,
+      "hae": 41883,
+      "shi": 41884,
+      "ĠInterpret": 41885,
+      "Ġcatabolic": 41886,
+      "Ġoccupations": 41887,
+      "ĠBifurc": 41888,
+      "Hydroxy": 41889,
+      "ĠKauf": 41890,
+      "sleep": 41891,
+      "amas": 41892,
+      "ĠSf": 41893,
+      "ĠMBP": 41894,
+      "Ġnonalcoholic": 41895,
+      "Ġdiscordant": 41896,
+      "Ġepigen": 41897,
+      "PRI": 41898,
+      "ĠRedshift": 41899,
+      "warn": 41900,
+      "Ġlaptop": 41901,
+      "Ġabrasive": 41902,
+      "îĤĿ": 41903,
+      "lh": 41904,
+      "ĠKnee": 41905,
+      "Ġscrat": 41906,
+      "Ġpoloidal": 41907,
+      "ĠUniv": 41908,
+      "omyosin": 41909,
+      "ĠAugmented": 41910,
+      "Ġtaxonom": 41911,
+      "ZrO": 41912,
+      "Ġphytochemicals": 41913,
+      "iten": 41914,
+      "ĠPatterson": 41915,
+      "thym": 41916,
+      "dihydropy": 41917,
+      "Ġky": 41918,
+      "ĠMetazoa": 41919,
+      "ALLY": 41920,
+      "Ġretinoblastoma": 41921,
+      "concatenate": 41922,
+      "Male": 41923,
+      "Ġomission": 41924,
+      "icher": 41925,
+      "ĠAzer": 41926,
+      "opp": 41927,
+      "pleasant": 41928,
+      "ningham": 41929,
+      "Ġaxially": 41930,
+      "HDFS": 41931,
+      "Ġfictional": 41932,
+      "Ï«": 41933,
+      "Ġnarc": 41934,
+      "Ġundertook": 41935,
+      "Ġmicrocirc": 41936,
+      "ONLY": 41937,
+      "IVER": 41938,
+      "ĠCycles": 41939,
+      "Meas": 41940,
+      "ĠGriffin": 41941,
+      "ĠPliocene": 41942,
+      "ĠpI": 41943,
+      "ĠAviation": 41944,
+      "ĠCategorical": 41945,
+      "ĠNils": 41946,
+      "Ġresidency": 41947,
+      "ĠLaur": 41948,
+      "Ġprefers": 41949,
+      "ĠassertEquals": 41950,
+      "Ġliquor": 41951,
+      "dM": 41952,
+      "osperm": 41953,
+      "ĠFUT": 41954,
+      "AlO": 41955,
+      "Ġcytometer": 41956,
+      "Ġstabilizers": 41957,
+      "Ġpremium": 41958,
+      "Serial": 41959,
+      "ĠWalking": 41960,
+      "íķľ": 41961,
+      "Ġconfronted": 41962,
+      "encapsulated": 41963,
+      "Card": 41964,
+      "ĠSeeds": 41965,
+      "abular": 41966,
+      "ukov": 41967,
+      "Listener": 41968,
+      "Choose": 41969,
+      "ĠSjö": 41970,
+      "Make": 41971,
+      "Ġisoc": 41972,
+      "amount": 41973,
+      "ATC": 41974,
+      "ija": 41975,
+      "Ġsulcus": 41976,
+      "ĠMöbius": 41977,
+      "ĠPrenatal": 41978,
+      "Ġß": 41979,
+      "Ġisochron": 41980,
+      "Ġbeans": 41981,
+      "ĠDens": 41982,
+      "ĠWelling": 41983,
+      "ĠOman": 41984,
+      "Stats": 41985,
+      "ĠValid": 41986,
+      "ĠReward": 41987,
+      "GK": 41988,
+      "Ġâ©": 41989,
+      "Ġelectroporation": 41990,
+      "ĠSNRs": 41991,
+      "Ġgarlic": 41992,
+      "ĠParticipant": 41993,
+      "ĠSplitting": 41994,
+      "ĠMeteorological": 41995,
+      "morillonite": 41996,
+      "Ġoedema": 41997,
+      "ĠDots": 41998,
+      "ĠClare": 41999,
+      "Ġstarter": 42000,
+      "Ġclimatology": 42001,
+      "Ġcommence": 42002,
+      "Ġfallen": 42003,
+      "ĠAuNPs": 42004,
+      "attrs": 42005,
+      "Ġconsultant": 42006,
+      "twisted": 42007,
+      "Solving": 42008,
+      "Ġcoercive": 42009,
+      "ë¡ľ": 42010,
+      "Kar": 42011,
+      "Ġstit": 42012,
+      "ĠSSB": 42013,
+      "ĠIW": 42014,
+      "Ġcanvas": 42015,
+      "pyruvate": 42016,
+      "methylsulfanyl": 42017,
+      "Ġastrophysics": 42018,
+      "ĠTraditionally": 42019,
+      "Ġexcitonic": 42020,
+      "wear": 42021,
+      "ĠTin": 42022,
+      "rosh": 42023,
+      "ĠClient": 42024,
+      "ĠCorrections": 42025,
+      "ĠPopular": 42026,
+      "ĠLiquids": 42027,
+      "finder": 42028,
+      "Ġstran": 42029,
+      "pline": 42030,
+      "orella": 42031,
+      "Ġincur": 42032,
+      "Ġarche": 42033,
+      "Ġmedically": 42034,
+      "Mur": 42035,
+      "peter": 42036,
+      "Ġbeverage": 42037,
+      "ĠNWs": 42038,
+      "Ġfolic": 42039,
+      "Ġspeculative": 42040,
+      "Ġñ": 42041,
+      "Ġribbons": 42042,
+      "ĠPriest": 42043,
+      "Quanti": 42044,
+      "Ġundisturbed": 42045,
+      "atche": 42046,
+      "assi": 42047,
+      "ĠPerforming": 42048,
+      "ĠElong": 42049,
+      "Ġmatchings": 42050,
+      "Ġfranchise": 42051,
+      "gio": 42052,
+      "ĠSarg": 42053,
+      "Ġaboard": 42054,
+      "cyclodextrin": 42055,
+      "ĠTHER": 42056,
+      "Ġsaturate": 42057,
+      "ĠKinematics": 42058,
+      "Ġpeptidoglycan": 42059,
+      "ĠShelf": 42060,
+      "tocopherol": 42061,
+      "Bottom": 42062,
+      "mith": 42063,
+      "rdx": 42064,
+      "zos": 42065,
+      "ĠtRNAs": 42066,
+      "utf": 42067,
+      "ENA": 42068,
+      "Ġlesson": 42069,
+      "Ġpolaron": 42070,
+      "braska": 42071,
+      "Ġathletic": 42072,
+      "Ġscrambled": 42073,
+      "Ġpursuing": 42074,
+      "Ġbodily": 42075,
+      "Ġcac": 42076,
+      "imen": 42077,
+      "ĠIκB": 42078,
+      "ĠRö": 42079,
+      "ĠRFC": 42080,
+      "ĠLPC": 42081,
+      "ĠiÏī": 42082,
+      "Ġdiary": 42083,
+      "Ġqueueing": 42084,
+      "ĠDivergence": 42085,
+      "ĠShigella": 42086,
+      "ĠUltrastruct": 42087,
+      "Ġtriphosphate": 42088,
+      "ĠImplant": 42089,
+      "Ġferrous": 42090,
+      "ĠBurton": 42091,
+      "ĠHertz": 42092,
+      "fabric": 42093,
+      "turing": 42094,
+      "ĠSSM": 42095,
+      "ograd": 42096,
+      "Ġmetazo": 42097,
+      "Chang": 42098,
+      "Ġadipogenesis": 42099,
+      "Ġnuisance": 42100,
+      "Ġanonymity": 42101,
+      "Ġrefrigerator": 42102,
+      "ìľ": 42103,
+      "oction": 42104,
+      "Ġsparing": 42105,
+      "Ġchalcogen": 42106,
+      "Ġobservatory": 42107,
+      "Ġbooster": 42108,
+      "ĠAndré": 42109,
+      "ĠSTO": 42110,
+      "yryl": 42111,
+      "ĠEDX": 42112,
+      "ĠDenver": 42113,
+      "Ġhomogenate": 42114,
+      "Callback": 42115,
+      "aC": 42116,
+      "hours": 42117,
+      "kova": 42118,
+      "ĠAUD": 42119,
+      "Ġspare": 42120,
+      "Ġpartons": 42121,
+      "ĠIntake": 42122,
+      "Ġrecognizable": 42123,
+      "ĠGoldstein": 42124,
+      "Ġstrikingly": 42125,
+      "Ġsanitation": 42126,
+      "Finder": 42127,
+      "Generation": 42128,
+      "boy": 42129,
+      "tam": 42130,
+      "ĠRPM": 42131,
+      "ivious": 42132,
+      "ylak": 42133,
+      "ophiles": 42134,
+      "Ġpriest": 42135,
+      "Ġeasiest": 42136,
+      "Ġdeliveries": 42137,
+      "Elmer": 42138,
+      "Ġzirconium": 42139,
+      "ĠMishra": 42140,
+      "ĠâĶ": 42141,
+      "ĠWDM": 42142,
+      "Ġperid": 42143,
+      "ĠZT": 42144,
+      "Ġlocalizes": 42145,
+      "ĠORs": 42146,
+      "ĠIDO": 42147,
+      "Ġpleasant": 42148,
+      "ĠMWCNTs": 42149,
+      "ĠJimmy": 42150,
+      "ĠYeh": 42151,
+      "gathered": 42152,
+      "kil": 42153,
+      "ĠKappa": 42154,
+      "Ġcartoon": 42155,
+      "ĠHeuristic": 42156,
+      "Ġsz": 42157,
+      "Ġorifice": 42158,
+      "Ġmannit": 42159,
+      "ĠCOMM": 42160,
+      "ICK": 42161,
+      "Ġfarmer": 42162,
+      "ĠSilencing": 42163,
+      "Ġprefixes": 42164,
+      "qc": 42165,
+      "ineurin": 42166,
+      "Ġinflated": 42167,
+      "ĠRez": 42168,
+      "Ġhydrodynamical": 42169,
+      "Ġoscillate": 42170,
+      "Ġpedestrians": 42171,
+      "ĠAngiotensin": 42172,
+      "ĠViscosity": 42173,
+      "Ġoligodendrocytes": 42174,
+      "Ġparotid": 42175,
+      "Layout": 42176,
+      "rageenan": 42177,
+      "Ġè": 42178,
+      "ĠmN": 42179,
+      "Ġdozen": 42180,
+      "exclusion": 42181,
+      "Ġpanic": 42182,
+      "ĠPDI": 42183,
+      "Ġtwentieth": 42184,
+      "ĠElectroph": 42185,
+      "Ġmicrobiology": 42186,
+      "Server": 42187,
+      "ĠParticipation": 42188,
+      "DET": 42189,
+      "Poss": 42190,
+      "ĠNemat": 42191,
+      "ĠNRF": 42192,
+      "arguments": 42193,
+      "Ġamylase": 42194,
+      "Ġargv": 42195,
+      "Ġresolves": 42196,
+      "Ġrevisions": 42197,
+      "Packet": 42198,
+      "Tools": 42199,
+      "YE": 42200,
+      "Ġtire": 42201,
+      "induction": 42202,
+      "asive": 42203,
+      "tonic": 42204,
+      "ém": 42205,
+      "carrying": 42206,
+      "ĠImmunoblot": 42207,
+      "ĠIPF": 42208,
+      "Ġdeteriorated": 42209,
+      "Ġjurisdiction": 42210,
+      "released": 42211,
+      "osmotic": 42212,
+      "Ġrestaurants": 42213,
+      "ï¸": 42214,
+      "ĠNm": 42215,
+      "Ġflips": 42216,
+      "Ġseparability": 42217,
+      "ĠRecursive": 42218,
+      "Ġpasture": 42219,
+      "ĠÄī": 42220,
+      "Ġblastocyst": 42221,
+      "MCP": 42222,
+      "Tb": 42223,
+      "uene": 42224,
+      "esulf": 42225,
+      "essim": 42226,
+      "Ġhen": 42227,
+      "ĠKull": 42228,
+      "ylum": 42229,
+      "arev": 42230,
+      "uests": 42231,
+      "ĠZip": 42232,
+      "Ġboats": 42233,
+      "Command": 42234,
+      "Continu": 42235,
+      "ĠBogoliubov": 42236,
+      "Ġmannitol": 42237,
+      "Know": 42238,
+      "г": 42239,
+      "ĠHack": 42240,
+      "Ġmassively": 42241,
+      "ĠAlloys": 42242,
+      "ĠCDP": 42243,
+      "ĠStereo": 42244,
+      "ĠElectrode": 42245,
+      "Ġisoflav": 42246,
+      "Ġinteroperability": 42247,
+      "ĠAdelaide": 42248,
+      "ĠPPD": 42249,
+      "ĠKou": 42250,
+      "Ġdiap": 42251,
+      "Ġconserve": 42252,
+      "Ġaggregating": 42253,
+      "Gluc": 42254,
+      "Ġîģ": 42255,
+      "Ġgust": 42256,
+      "ĠLeb": 42257,
+      "ETIC": 42258,
+      "ĠConsol": 42259,
+      "ĠMorita": 42260,
+      "Relative": 42261,
+      "Ġpaleo": 42262,
+      "Ġwitnesses": 42263,
+      "ĠLauren": 42264,
+      "azepine": 42265,
+      "ĠTY": 42266,
+      "ĠIdi": 42267,
+      "ĠMent": 42268,
+      "ĠRCA": 42269,
+      "igenin": 42270,
+      "ĠDefence": 42271,
+      "Ġpyrimidine": 42272,
+      "ĠTiN": 42273,
+      "Ġendothelin": 42274,
+      "Ġpandas": 42275,
+      "Ġswallowing": 42276,
+      "Ġcongestive": 42277,
+      "Ġvinc": 42278,
+      "ĠDIP": 42279,
+      "ĠHough": 42280,
+      "Ġzw": 42281,
+      "ĠKimura": 42282,
+      "representations": 42283,
+      "ĠPromote": 42284,
+      "ĠTerry": 42285,
+      "Ġhatched": 42286,
+      "lookup": 42287,
+      "Electron": 42288,
+      "Ġdormancy": 42289,
+      "Ġresign": 42290,
+      "Ġvaluations": 42291,
+      "Ġmakeup": 42292,
+      "ĠAmy": 42293,
+      "CLUD": 42294,
+      "SEP": 42295,
+      "tubule": 42296,
+      "Ġsoldier": 42297,
+      "ĠTz": 42298,
+      "ĠTrump": 42299,
+      "ĠKramer": 42300,
+      "coni": 42301,
+      "Ġengraft": 42302,
+      "Ġvacuole": 42303,
+      "Ġreplicating": 42304,
+      "itonitis": 42305,
+      "ĠBacteri": 42306,
+      "vaccinated": 42307,
+      "olt": 42308,
+      "ĠAhn": 42309,
+      "Ġanem": 42310,
+      "ĠBIT": 42311,
+      "geo": 42312,
+      "Ġmicrogravity": 42313,
+      "ĠShir": 42314,
+      "ĠWorldwide": 42315,
+      "Ġignor": 42316,
+      "ĠËĩ": 42317,
+      "Ġlubrication": 42318,
+      "java": 42319,
+      "vt": 42320,
+      "Ġyl": 42321,
+      "Ġhills": 42322,
+      "ĠFOL": 42323,
+      "Ġbasaltic": 42324,
+      "Neill": 42325,
+      "ĠEthiopian": 42326,
+      "ĠNOTCH": 42327,
+      "ĠMOSFET": 42328,
+      "leaving": 42329,
+      "ĠPter": 42330,
+      "ĠWeld": 42331,
+      "aple": 42332,
+      "Ġsandwic": 42333,
+      "Ġazide": 42334,
+      "ĠStimuli": 42335,
+      "Ġlizard": 42336,
+      "ĠCinc": 42337,
+      "ĠHain": 42338,
+      "icals": 42339,
+      "Ġcontacting": 42340,
+      "ĠMarx": 42341,
+      "Ġpsychotherapy": 42342,
+      "ĠRetin": 42343,
+      "Ġcatheterization": 42344,
+      "ĠNanoparticle": 42345,
+      "ĠTCC": 42346,
+      "vermectin": 42347,
+      "ĠBASE": 42348,
+      "Ġnotor": 42349,
+      "Ġelectronically": 42350,
+      "steroid": 42351,
+      "Ġbilaterally": 42352,
+      "Ġnephritis": 42353,
+      "Ġirritation": 42354,
+      "ĠProlonged": 42355,
+      "Your": 42356,
+      "heuristic": 42357,
+      "urgeon": 42358,
+      "Ġleftmost": 42359,
+      "ĠREVIEW": 42360,
+      "Ġgastrectomy": 42361,
+      "ENTIAL": 42362,
+      "Means": 42363,
+      "ĠDyson": 42364,
+      "Ġbrands": 42365,
+      "yields": 42366,
+      "mercapto": 42367,
+      "rub": 42368,
+      "ouncement": 42369,
+      "errno": 42370,
+      "Ġviewers": 42371,
+      "butan": 42372,
+      "ĠMalay": 42373,
+      "ylindrical": 42374,
+      "Ġprominently": 42375,
+      "Ġescaping": 42376,
+      "Ġquerying": 42377,
+      "Storage": 42378,
+      "Fos": 42379,
+      "codec": 42380,
+      "ĠcM": 42381,
+      "strates": 42382,
+      "glove": 42383,
+      "ĠTrajectories": 42384,
+      "Ġsterol": 42385,
+      "Ġhematopoiesis": 42386,
+      "Ġcuprates": 42387,
+      "Ok": 42388,
+      "daily": 42389,
+      "ĠBIO": 42390,
+      "ĠLICENSE": 42391,
+      "ellations": 42392,
+      "assy": 42393,
+      "SURE": 42394,
+      "Ġepinephrine": 42395,
+      "Ġdownwards": 42396,
+      "corner": 42397,
+      "assertTrue": 42398,
+      "Ġáºij": 42399,
+      "ĠSouza": 42400,
+      "MAG": 42401,
+      "porph": 42402,
+      "Ġeffluents": 42403,
+      "loem": 42404,
+      "oaddition": 42405,
+      "obutyl": 42406,
+      "elestial": 42407,
+      "Fem": 42408,
+      "mpi": 42409,
+      "ĠRs": 42410,
+      "ellates": 42411,
+      "ĠKag": 42412,
+      "Ġuncoupled": 42413,
+      "Ġlegumes": 42414,
+      "Ġomitting": 42415,
+      "û": 42416,
+      "ĠTABLE": 42417,
+      "haled": 42418,
+      "ĠÅģ": 42419,
+      "Ġmisfit": 42420,
+      "Ġmolars": 42421,
+      "otechnological": 42422,
+      "Markov": 42423,
+      "Ġpraised": 42424,
+      "ĠDab": 42425,
+      "ĠVij": 42426,
+      "entilation": 42427,
+      "ĠChatter": 42428,
+      "Ġboiled": 42429,
+      "Ġcatches": 42430,
+      "annotation": 42431,
+      "Signal": 42432,
+      "Ġleverages": 42433,
+      "Ġadvisory": 42434,
+      "song": 42435,
+      "ondition": 42436,
+      "Ġfug": 42437,
+      "raps": 42438,
+      "ĠMCD": 42439,
+      "particip": 42440,
+      "obian": 42441,
+      "Ġcounsel": 42442,
+      "ĠPRP": 42443,
+      "ediol": 42444,
+      "ĠŨ": 42445,
+      "Ġbruce": 42446,
+      "Shanghai": 42447,
+      "DataFrame": 42448,
+      "ĠCorrespondingly": 42449,
+      "Ġacrylamide": 42450,
+      "fellow": 42451,
+      "lob": 42452,
+      "igt": 42453,
+      "ĠCrystallization": 42454,
+      "Ġindomethacin": 42455,
+      "ĠPDR": 42456,
+      "giate": 42457,
+      "ĠPanels": 42458,
+      "complexes": 42459,
+      "ĠNicol": 42460,
+      "Ġfoliar": 42461,
+      "cubic": 42462,
+      "ĠdE": 42463,
+      "ĠCCM": 42464,
+      "plating": 42465,
+      "Ġresistors": 42466,
+      "ĠGaz": 42467,
+      "Ġoverturn": 42468,
+      "Ġrepress": 42469,
+      "Ġpots": 42470,
+      "ĠPIK": 42471,
+      "Ġdermis": 42472,
+      "Represent": 42473,
+      "ĠAndersson": 42474,
+      "Ġretrotranspos": 42475,
+      "ASA": 42476,
+      "Counter": 42477,
+      "Tet": 42478,
+      "imin": 42479,
+      "performed": 42480,
+      "ĠNept": 42481,
+      "Ġheel": 42482,
+      "rold": 42483,
+      "aires": 42484,
+      "Ġreadability": 42485,
+      "Ġbenefited": 42486,
+      "Ġpulsation": 42487,
+      "ĠBalancing": 42488,
+      "Ġevaporator": 42489,
+      "Ġeigens": 42490,
+      "ĠHospit": 42491,
+      "Ġtrails": 42492,
+      "ĠCoordinate": 42493,
+      "accase": 42494,
+      "ĠHRMS": 42495,
+      "signaling": 42496,
+      "ĠNPY": 42497,
+      "Ġameliorated": 42498,
+      "tuples": 42499,
+      "Ġmetasurface": 42500,
+      "ĠFrancesco": 42501,
+      "Ġspoof": 42502,
+      "îŸ": 42503,
+      "Fu": 42504,
+      "JK": 42505,
+      "ej": 42506,
+      "Ġgoss": 42507,
+      "ĠHim": 42508,
+      "Ġprioritized": 42509,
+      "Ġmesothelioma": 42510,
+      "idxs": 42511,
+      "Ġreconnaissance": 42512,
+      "Ġlamps": 42513,
+      "ãĢĤ": 42514,
+      "Ġreformulation": 42515,
+      "Ġdelirium": 42516,
+      "ĠNPR": 42517,
+      "ĠGamb": 42518,
+      "illas": 42519,
+      "-----": 42520,
+      "Ġdrilled": 42521,
+      "ĠGenotyping": 42522,
+      "ĠBlank": 42523,
+      "Ġpropeller": 42524,
+      "Ġcereals": 42525,
+      "ĠAirborne": 42526,
+      "ĠPhotocatalytic": 42527,
+      "ĠCavity": 42528,
+      "Ġdolphins": 42529,
+      "ĠsgRNA": 42530,
+      "understood": 42531,
+      "eous": 42532,
+      "Ġsax": 42533,
+      "olayer": 42534,
+      "ĠPend": 42535,
+      "ĠGET": 42536,
+      "cled": 42537,
+      "Ġü": 42538,
+      "Ġcytosine": 42539,
+      "Ġparallelization": 42540,
+      "MMs": 42541,
+      "ĠOrganisation": 42542,
+      "Models": 42543,
+      "Ġaccommodated": 42544,
+      "Ġcholest": 42545,
+      "Ġinactivity": 42546,
+      "ĠBoss": 42547,
+      "ĠGCS": 42548,
+      "Ġsoaked": 42549,
+      "ĠSecreted": 42550,
+      "Ġvacuolar": 42551,
+      "zoan": 42552,
+      "ĠGreene": 42553,
+      "Ġbolt": 42554,
+      "bj": 42555,
+      "ĠTall": 42556,
+      "Ġstor": 42557,
+      "ĠMob": 42558,
+      "Ġblurred": 42559,
+      "INO": 42560,
+      "ĠMetallic": 42561,
+      "uffs": 42562,
+      "ĠNOTE": 42563,
+      "Ġsonicated": 42564,
+      "obutyric": 42565,
+      "ĠtDCS": 42566,
+      "ĠNes": 42567,
+      "ospir": 42568,
+      "weigh": 42569,
+      "ĠRegulator": 42570,
+      "Ġhemolysis": 42571,
+      "Ġsounding": 42572,
+      "Ġcruciate": 42573,
+      "Ġcapsaicin": 42574,
+      "ĠTyrosine": 42575,
+      "ĠTell": 42576,
+      "ĠPEP": 42577,
+      "ĠRc": 42578,
+      "ĠEating": 42579,
+      "ĠGoals": 42580,
+      "uret": 42581,
+      "Ġearn": 42582,
+      "Ġcolleges": 42583,
+      "Ġchemoattract": 42584,
+      "Ġỹ": 42585,
+      "ĠEchocardi": 42586,
+      "Fort": 42587,
+      "sodium": 42588,
+      "amined": 42589,
+      "ĠNPP": 42590,
+      "ĠKalu": 42591,
+      "Ġdecipher": 42592,
+      "tetramethyl": 42593,
+      "ĠOPN": 42594,
+      "straight": 42595,
+      "Ġtempered": 42596,
+      "ĠHindu": 42597,
+      "ĠOrdinary": 42598,
+      "ĠAChE": 42599,
+      "JNK": 42600,
+      "fos": 42601,
+      "vcpu": 42602,
+      "enamide": 42603,
+      "ĠCrack": 42604,
+      "apical": 42605,
+      "Ġantiserum": 42606,
+      "triplet": 42607,
+      "decision": 42608,
+      "Ġcancels": 42609,
+      "ĠPMN": 42610,
+      "Ġporphy": 42611,
+      "ĠDSA": 42612,
+      "Ġsubmatrix": 42613,
+      "Ġjas": 42614,
+      "Ġreptiles": 42615,
+      "ĠSoon": 42616,
+      "ĠStatistically": 42617,
+      "Ġleveraged": 42618,
+      "Williams": 42619,
+      "FLD": 42620,
+      "folk": 42621,
+      "Ġbang": 42622,
+      "ĠSCL": 42623,
+      "asses": 42624,
+      "Ġtendons": 42625,
+      "founded": 42626,
+      "ĠRicketts": 42627,
+      "inset": 42628,
+      "Ġspun": 42629,
+      "Ġunramified": 42630,
+      "Ġrape": 42631,
+      "ĠZZ": 42632,
+      "ĠNebula": 42633,
+      "Ġthrombotic": 42634,
+      "ĠBoron": 42635,
+      "ĠArgon": 42636,
+      "pooling": 42637,
+      "ĠMarginal": 42638,
+      "Ġfellowship": 42639,
+      "Ġerythropoietin": 42640,
+      "mathpzc": 42641,
+      "xL": 42642,
+      "ĠSik": 42643,
+      "ĠBayer": 42644,
+      "Ġoverdose": 42645,
+      "ĠCOI": 42646,
+      "ĠLesions": 42647,
+      "ĠCompetitive": 42648,
+      "ĠODEs": 42649,
+      "wrap": 42650,
+      "achlor": 42651,
+      "Ġsubordinate": 42652,
+      "ĠYBa": 42653,
+      "headed": 42654,
+      "Ġgrasses": 42655,
+      "Ġbirational": 42656,
+      "ĠJeffrey": 42657,
+      "Ġmolding": 42658,
+      "Ġlidocaine": 42659,
+      "ĠFOXO": 42660,
+      "terminated": 42661,
+      "ĠâĩIJâĩĴ": 42662,
+      "ĠMEL": 42663,
+      "ticulum": 42664,
+      "Ġré": 42665,
+      "Ġclaud": 42666,
+      "Ġjamming": 42667,
+      "Static": 42668,
+      "Ġtributary": 42669,
+      "atet": 42670,
+      "edonia": 42671,
+      "ĠCMP": 42672,
+      "ĠVN": 42673,
+      "represents": 42674,
+      "SOURCE": 42675,
+      "uckland": 42676,
+      "ĠInterests": 42677,
+      "ĠNanoscale": 42678,
+      "oconjug": 42679,
+      "Ġcatalogues": 42680,
+      "ĠActinobacteria": 42681,
+      "Fixed": 42682,
+      "basal": 42683,
+      "Ġantiparallel": 42684,
+      "Ġconfusing": 42685,
+      "Ġmarkings": 42686,
+      "Ġdistinctions": 42687,
+      "ĠHua": 42688,
+      "ĠWatts": 42689,
+      "Ġnanofluid": 42690,
+      "Ġdiffractometer": 42691,
+      "Later": 42692,
+      "migration": 42693,
+      "Ġcoplanar": 42694,
+      "Ġhypomethyl": 42695,
+      "PDS": 42696,
+      "SOs": 42697,
+      "Correspond": 42698,
+      "Ġelucidating": 42699,
+      "IZED": 42700,
+      "EVs": 42701,
+      "gart": 42702,
+      "mTc": 42703,
+      "ĠTUR": 42704,
+      "uracies": 42705,
+      "Ġfollower": 42706,
+      "Ġhaze": 42707,
+      "OUTPUT": 42708,
+      "ĠMyeloid": 42709,
+      "BUFFER": 42710,
+      "Camp": 42711,
+      "anim": 42712,
+      "ĠTES": 42713,
+      "ĠCricket": 42714,
+      "ĠPaired": 42715,
+      "ĠPAGE": 42716,
+      "ĠBid": 42717,
+      "Ġyrs": 42718,
+      "Ġendow": 42719,
+      "ircase": 42720,
+      "ĠSupported": 42721,
+      "Ġleaflet": 42722,
+      "ĠPromoter": 42723,
+      "Ġconvinced": 42724,
+      "liers": 42725,
+      "hera": 42726,
+      "Ġseller": 42727,
+      "agreement": 42728,
+      "Ġunary": 42729,
+      "onstructed": 42730,
+      "Ġrestrained": 42731,
+      "ĠPetersen": 42732,
+      "Analog": 42733,
+      "Ġexacerbations": 42734,
+      "Ġperforated": 42735,
+      "tids": 42736,
+      "ĠMSH": 42737,
+      "oui": 42738,
+      "ĠCori": 42739,
+      "ĠCruc": 42740,
+      "Ġfracturing": 42741,
+      "Ġinfertile": 42742,
+      "ĠPROBLEM": 42743,
+      "ĠFriedmann": 42744,
+      "Ġspectrophotometry": 42745,
+      "ERGY": 42746,
+      "otus": 42747,
+      "proposed": 42748,
+      "ĠMoisture": 42749,
+      "ĠNoether": 42750,
+      "ĠLaunch": 42751,
+      "ĠLearn": 42752,
+      "Ġvena": 42753,
+      "Ġfasci": 42754,
+      "Ġquiescence": 42755,
+      "ĠPrand": 42756,
+      "ĠConvert": 42757,
+      "Ġtriage": 42758,
+      "ANE": 42759,
+      "Ġfeedstock": 42760,
+      "ĠdBm": 42761,
+      "Ġneoformans": 42762,
+      "GSE": 42763,
+      "ĠAPE": 42764,
+      "Ġcardiometabolic": 42765,
+      "Ġmagnetometer": 42766,
+      "Environment": 42767,
+      "Ġfulfills": 42768,
+      "ĠManganese": 42769,
+      "BMP": 42770,
+      "ĠRatios": 42771,
+      "istable": 42772,
+      "assume": 42773,
+      "Ġrespected": 42774,
+      "Ġscars": 42775,
+      "Ġsupporters": 42776,
+      "ĠAugmentation": 42777,
+      "Ġglycosylated": 42778,
+      "ĠUltrafast": 42779,
+      "Ġdemethylation": 42780,
+      "metastatic": 42781,
+      "cylinder": 42782,
+      "Ġhang": 42783,
+      "ĠMAV": 42784,
+      "disjoint": 42785,
+      "pharose": 42786,
+      "ĠLebanon": 42787,
+      "PIs": 42788,
+      "labeling": 42789,
+      "Ġneutralino": 42790,
+      "ĠSOCS": 42791,
+      "xcb": 42792,
+      "ĠTerritory": 42793,
+      "ĠPolicies": 42794,
+      "King": 42795,
+      "Ġallied": 42796,
+      "Ġsaturates": 42797,
+      "muscle": 42798,
+      "odimers": 42799,
+      "Ġbt": 42800,
+      "ĠHang": 42801,
+      "ĠEb": 42802,
+      "Ġchimer": 42803,
+      "Ġnotational": 42804,
+      "Ġcolder": 42805,
+      "ultz": 42806,
+      "transverse": 42807,
+      "HOUT": 42808,
+      "ĠKarls": 42809,
+      "torsion": 42810,
+      "JI": 42811,
+      "Ġmari": 42812,
+      "emon": 42813,
+      "Ġlogarithmically": 42814,
+      "ĠâIJ¦": 42815,
+      "ìĿĦ": 42816,
+      "Ġaeration": 42817,
+      "Ġsoma": 42818,
+      "ĠSomal": 42819,
+      "Ġspoil": 42820,
+      "diver": 42821,
+      "Ġbreakpoints": 42822,
+      "ĠHarmon": 42823,
+      "Ġpharmacologic": 42824,
+      "ĠMosquito": 42825,
+      "ĠModifications": 42826,
+      "Ġadjo": 42827,
+      "ĠPapers": 42828,
+      "generally": 42829,
+      "ïĺ¹": 42830,
+      "TARGET": 42831,
+      "ĠPrix": 42832,
+      "ocaps": 42833,
+      "ĠEin": 42834,
+      "Ġmicrogrid": 42835,
+      "ĠInterplay": 42836,
+      "Ġcopying": 42837,
+      "Alpha": 42838,
+      "ĠSlope": 42839,
+      "ĠLipofectamine": 42840,
+      "highest": 42841,
+      "DRO": 42842,
+      "ĠHipp": 42843,
+      "Ġshaken": 42844,
+      "Ġunderline": 42845,
+      "Ġfilmed": 42846,
+      "maturity": 42847,
+      "icture": 42848,
+      "ILS": 42849,
+      "Span": 42850,
+      "Ġinverters": 42851,
+      "QUE": 42852,
+      "determining": 42853,
+      "Ġeosinophilic": 42854,
+      "DY": 42855,
+      "ĠLID": 42856,
+      "ĠGig": 42857,
+      "Ġintraepithelial": 42858,
+      "NbO": 42859,
+      "freedom": 42860,
+      "Ġassured": 42861,
+      "ĠArche": 42862,
+      "ĠSubstitution": 42863,
+      "ĠSrivastava": 42864,
+      "ĠMozamb": 42865,
+      "Ġaro": 42866,
+      "orc": 42867,
+      "ĠIbrahim": 42868,
+      "ĠDST": 42869,
+      "Ġabl": 42870,
+      "Ġxer": 42871,
+      "ountable": 42872,
+      "Ġlossless": 42873,
+      "Ġconcentrating": 42874,
+      "Ġstains": 42875,
+      "ĠSolve": 42876,
+      "continuity": 42877,
+      "ĠTorr": 42878,
+      "Ġpitfalls": 42879,
+      "bestos": 42880,
+      "Otherwise": 42881,
+      "adhyay": 42882,
+      "bard": 42883,
+      "ĠCAA": 42884,
+      "odetic": 42885,
+      "Ġasthmatic": 42886,
+      "Ġrationality": 42887,
+      "ĠYorkshire": 42888,
+      "neighborhood": 42889,
+      "Ġheroin": 42890,
+      "Ġscatterers": 42891,
+      "ĠHearing": 42892,
+      "ĠEFT": 42893,
+      "ĠNurses": 42894,
+      "ĠGLI": 42895,
+      "ĠZeta": 42896,
+      "ĠNeigh": 42897,
+      "Ġventure": 42898,
+      "Ġtoxicological": 42899,
+      "Ġrolls": 42900,
+      "fv": 42901,
+      "Ġcrick": 42902,
+      "ĠdÏī": 42903,
+      "avia": 42904,
+      "elder": 42905,
+      "Ġinvade": 42906,
+      "extracted": 42907,
+      "MLP": 42908,
+      "ĠPAI": 42909,
+      "ĠMellitus": 42910,
+      "Ġbrucei": 42911,
+      "gpio": 42912,
+      "emotional": 42913,
+      "ĠDale": 42914,
+      "ĠEz": 42915,
+      "Ġtransactivation": 42916,
+      "Ġquantiles": 42917,
+      "Ġnucleosynthesis": 42918,
+      "ĠAmel": 42919,
+      "Ġchromophore": 42920,
+      "Ġliterally": 42921,
+      "bandwidth": 42922,
+      "atohepatitis": 42923,
+      "Ġultrafiltration": 42924,
+      "Martin": 42925,
+      "Ġangioplasty": 42926,
+      "insertion": 42927,
+      "Dan": 42928,
+      "squeeze": 42929,
+      "usr": 42930,
+      "uconazole": 42931,
+      "ĠFAR": 42932,
+      "Ġshadows": 42933,
+      "Ġimitation": 42934,
+      "ĠKann": 42935,
+      "hesi": 42936,
+      "Ġmicellar": 42937,
+      "vester": 42938,
+      "ĠPerse": 42939,
+      "acetamol": 42940,
+      "GRAPH": 42941,
+      "ĠAIPS": 42942,
+      "Ġpromptly": 42943,
+      "anchor": 42944,
+      "Ġischaemia": 42945,
+      "pump": 42946,
+      "Ġmafic": 42947,
+      "Ġlazy": 42948,
+      "ĠCEL": 42949,
+      "ĠGorenstein": 42950,
+      "ĠWGS": 42951,
+      "Ġsignifies": 42952,
+      "Ġsplines": 42953,
+      "determination": 42954,
+      "Ġrelaying": 42955,
+      "piperazine": 42956,
+      "Ġsyncytial": 42957,
+      "ĠAub": 42958,
+      "ĠDX": 42959,
+      "Ġorthotopic": 42960,
+      "ĠLinkage": 42961,
+      "Ġharmony": 42962,
+      "ĠKazakh": 42963,
+      "ĠVladimir": 42964,
+      "Ġpray": 42965,
+      "imolar": 42966,
+      "Ġgrayscale": 42967,
+      "Ġanalyst": 42968,
+      "ĠTransl": 42969,
+      "Ġmeniscus": 42970,
+      "ĠMedica": 42971,
+      "osaurus": 42972,
+      "Ġв": 42973,
+      "Ġinfiltrated": 42974,
+      "Ġâĸ³": 42975,
+      "Ġsaccades": 42976,
+      "Ġdisentangle": 42977,
+      "Hart": 42978,
+      "fined": 42979,
+      "Ġbicycle": 42980,
+      "ository": 42981,
+      "unlikely": 42982,
+      "erephthal": 42983,
+      "ĠLia": 42984,
+      "Ġgroupings": 42985,
+      "Ġcategorize": 42986,
+      "Ġbiogeography": 42987,
+      "ĠAPPROACH": 42988,
+      "ĠNing": 42989,
+      "ĠGrap": 42990,
+      "versa": 42991,
+      "Ġradiologists": 42992,
+      "ĠRecording": 42993,
+      "Ġboiler": 42994,
+      "adders": 42995,
+      "Candid": 42996,
+      "MQ": 42997,
+      "Ġbw": 42998,
+      "ĠSector": 42999,
+      "ĠHIT": 43000,
+      "ĠESCC": 43001,
+      "essence": 43002,
+      "orean": 43003,
+      "estyles": 43004,
+      "SUCCESS": 43005,
+      "nein": 43006,
+      "ultra": 43007,
+      "ramp": 43008,
+      "Thomas": 43009,
+      "ĠPrepar": 43010,
+      "ĠInstitut": 43011,
+      "Ġherbicide": 43012,
+      "ĠChaotic": 43013,
+      "Ġsphincter": 43014,
+      "Ġcompactifications": 43015,
+      "Clear": 43016,
+      "Trp": 43017,
+      "Decoder": 43018,
+      "Ġsapphire": 43019,
+      "ĠIdaho": 43020,
+      "persing": 43021,
+      "chiral": 43022,
+      "ĠDischarge": 43023,
+      "Accordingly": 43024,
+      "ĠArthritis": 43025,
+      "ĠJaneiro": 43026,
+      "nj": 43027,
+      "ĠKd": 43028,
+      "Ġoutlets": 43029,
+      "Ġsusceptibilities": 43030,
+      "Ġdiverged": 43031,
+      "Ġroller": 43032,
+      "sufficient": 43033,
+      "clustering": 43034,
+      "ĠTehran": 43035,
+      "Ġtb": 43036,
+      "blank": 43037,
+      "Ġdigitally": 43038,
+      "Ġnecrotizing": 43039,
+      "FALSE": 43040,
+      "Ġwhor": 43041,
+      "errals": 43042,
+      "ĠMotivated": 43043,
+      "enzae": 43044,
+      "ĠRefinement": 43045,
+      "Ġticket": 43046,
+      "Ġprotrusions": 43047,
+      "ĠDonaldson": 43048,
+      "ĠBeth": 43049,
+      "Ġsputtered": 43050,
+      "Ġautocrine": 43051,
+      "copene": 43052,
+      "Ġcollar": 43053,
+      "Ġuppermost": 43054,
+      "Ġoxygenated": 43055,
+      "Intro": 43056,
+      "âĨIJ": 43057,
+      "ĠHippo": 43058,
+      "Ġdune": 43059,
+      "idines": 43060,
+      "ĠHä": 43061,
+      "Ġregi": 43062,
+      "Ġnois": 43063,
+      "Ġphotodiode": 43064,
+      "ĠFeb": 43065,
+      "mutated": 43066,
+      "ĠCFL": 43067,
+      "stepping": 43068,
+      "Selection": 43069,
+      "ĠWebster": 43070,
+      "ĠHERA": 43071,
+      "indicating": 43072,
+      "Ġtrainees": 43073,
+      "Rot": 43074,
+      "ĠFAK": 43075,
+      "ĠAsn": 43076,
+      "Ġfats": 43077,
+      "foliation": 43078,
+      "Ġarticulation": 43079,
+      "Ġcusps": 43080,
+      "ĠJennifer": 43081,
+      "Ġintimately": 43082,
+      "ĠPing": 43083,
+      "sov": 43084,
+      "oxious": 43085,
+      "hydrate": 43086,
+      "ĠArchives": 43087,
+      "Gonz": 43088,
+      "Ġé": 43089,
+      "Ġchl": 43090,
+      "ĠOLS": 43091,
+      "coph": 43092,
+      "Ġairline": 43093,
+      "Ġfoetal": 43094,
+      "ĠRolling": 43095,
+      "ĠGENERAL": 43096,
+      "ONAL": 43097,
+      "agons": 43098,
+      "ĠDorsal": 43099,
+      "Ġritual": 43100,
+      "butyrate": 43101,
+      "oglut": 43102,
+      "Ġhexa": 43103,
+      "ĠSyria": 43104,
+      "Ġontogeny": 43105,
+      "ĠFBG": 43106,
+      "coverage": 43107,
+      "Ġtachyon": 43108,
+      "ĠPermanent": 43109,
+      "lum": 43110,
+      "Ġsv": 43111,
+      "Ġoo": 43112,
+      "energetic": 43113,
+      "altitude": 43114,
+      "Inc": 43115,
+      "ĠNebraska": 43116,
+      "ĠRESP": 43117,
+      "Ġdysbiosis": 43118,
+      "Ġmarketed": 43119,
+      "oxicillin": 43120,
+      "ĠBroadcast": 43121,
+      "racyclo": 43122,
+      "ĠFifteen": 43123,
+      "ĠNarayan": 43124,
+      "Ġlettuce": 43125,
+      "orea": 43126,
+      "Ġintercepts": 43127,
+      "Ġworkstation": 43128,
+      "ĠPlains": 43129,
+      "CCL": 43130,
+      "Ġorientable": 43131,
+      "ĠBoosting": 43132,
+      "ĠSOI": 43133,
+      "ĠChecking": 43134,
+      "ĠFIFO": 43135,
+      "Ġinsets": 43136,
+      "ĠSRT": 43137,
+      "Ġacrom": 43138,
+      "owner": 43139,
+      "MIX": 43140,
+      "ĠArb": 43141,
+      "Ġfaeces": 43142,
+      "ĠCarlson": 43143,
+      "Ġperivascular": 43144,
+      "infiltrating": 43145,
+      "Ìħ": 43146,
+      "Ġmalle": 43147,
+      "ocate": 43148,
+      "ĠBold": 43149,
+      "unctive": 43150,
+      "excess": 43151,
+      "Ġloosen": 43152,
+      "Ġprioritization": 43153,
+      "Ġannotate": 43154,
+      "Ġgrammars": 43155,
+      "Ġbred": 43156,
+      "Ġexocytosis": 43157,
+      "ĠDahl": 43158,
+      "athyroidism": 43159,
+      "veli": 43160,
+      "Ġopted": 43161,
+      "Ġsmoked": 43162,
+      "ĠPlates": 43163,
+      "EMG": 43164,
+      "ROW": 43165,
+      "IFIC": 43166,
+      "OLS": 43167,
+      "oregulatory": 43168,
+      "Ġwhiskers": 43169,
+      "secretase": 43170,
+      "Ġexaggerated": 43171,
+      "ĠBib": 43172,
+      "deformed": 43173,
+      "Ġzur": 43174,
+      "ropine": 43175,
+      "Ġpairings": 43176,
+      "chromosome": 43177,
+      "Elements": 43178,
+      "priority": 43179,
+      "Ġlyophilized": 43180,
+      "ĠChaudh": 43181,
+      "Wilk": 43182,
+      "ĠCation": 43183,
+      "otta": 43184,
+      "Ġnonconvex": 43185,
+      "Ġdepolymer": 43186,
+      "MMARY": 43187,
+      "Controlled": 43188,
+      "carboxy": 43189,
+      "Ġaugmenting": 43190,
+      "Ġappointments": 43191,
+      "Ġtraversed": 43192,
+      "ĠFletcher": 43193,
+      "Ġexpiratory": 43194,
+      "Ġelephant": 43195,
+      "ĠBlocks": 43196,
+      "ĠFluids": 43197,
+      "walls": 43198,
+      "increased": 43199,
+      "propanamide": 43200,
+      "ĠAkaike": 43201,
+      "ĠCBM": 43202,
+      "ĠEcho": 43203,
+      "admissible": 43204,
+      "Ġdisassembly": 43205,
+      "ĠarXiv": 43206,
+      "icke": 43207,
+      "LIST": 43208,
+      "phenotype": 43209,
+      "ĠProvincial": 43210,
+      "legend": 43211,
+      "PAS": 43212,
+      "rnn": 43213,
+      "sand": 43214,
+      "Ġbariatric": 43215,
+      "ĠPush": 43216,
+      "ĠApoE": 43217,
+      "caprolactone": 43218,
+      "modeling": 43219,
+      "Ġŵ": 43220,
+      "Ġsupercapacitors": 43221,
+      "oron": 43222,
+      "ĠpK": 43223,
+      "strophy": 43224,
+      "ĠSuc": 43225,
+      "unda": 43226,
+      "team": 43227,
+      "Ġitiner": 43228,
+      "Ġswell": 43229,
+      "ĠBioactive": 43230,
+      "ĠIndicators": 43231,
+      "ĠIFT": 43232,
+      "ĠDK": 43233,
+      "Ġcapit": 43234,
+      "shapes": 43235,
+      "Ġtrachea": 43236,
+      "delayed": 43237,
+      "ĠGuangdong": 43238,
+      "Lepid": 43239,
+      "TGA": 43240,
+      "hk": 43241,
+      "olon": 43242,
+      "ogenin": 43243,
+      "ĠAck": 43244,
+      "Ġlogically": 43245,
+      "contributions": 43246,
+      "ĠCleavage": 43247,
+      "hurst": 43248,
+      "bdd": 43249,
+      "STD": 43250,
+      "ĠFut": 43251,
+      "tek": 43252,
+      "ĠInher": 43253,
+      "Ġchemis": 43254,
+      "Ġbreakpoint": 43255,
+      "estimates": 43256,
+      "ĠOttoman": 43257,
+      "ĠNafion": 43258,
+      "WIDTH": 43259,
+      "Ġsizable": 43260,
+      "ĠTsu": 43261,
+      "embolic": 43262,
+      "Ġrightmost": 43263,
+      "ĠCellulose": 43264,
+      "ictionaries": 43265,
+      "ĠMycoplasma": 43266,
+      "ĠBurgers": 43267,
+      "ĠKeplerian": 43268,
+      "UCTION": 43269,
+      "VB": 43270,
+      "Ġbcc": 43271,
+      "raid": 43272,
+      "ENDIX": 43273,
+      "Ġscoping": 43274,
+      "ĠPRI": 43275,
+      "ĠCdSe": 43276,
+      "ĠGreedy": 43277,
+      "ĠHammer": 43278,
+      "ĠBacteroides": 43279,
+      "informative": 43280,
+      "Ġresembled": 43281,
+      "yllium": 43282,
+      "Twenty": 43283,
+      "Ġpounds": 43284,
+      "Ġunpolarized": 43285,
+      "Ġconfigure": 43286,
+      "Ġtranscriptionally": 43287,
+      "Ġmicroscale": 43288,
+      "ĠPutting": 43289,
+      "Ġpyrrol": 43290,
+      "ĠLASSO": 43291,
+      "filtration": 43292,
+      "Ġtech": 43293,
+      "performing": 43294,
+      "Along": 43295,
+      "ĠCTLA": 43296,
+      "Ġauthorization": 43297,
+      "URAL": 43298,
+      "Ġleaky": 43299,
+      "Optical": 43300,
+      "ĠReveal": 43301,
+      "ĠHUVECs": 43302,
+      "Wu": 43303,
+      "custom": 43304,
+      "dible": 43305,
+      "Ġ": 43306,
+      "CDCl": 43307,
+      "Ġemphys": 43308,
+      "Neut": 43309,
+      "collagen": 43310,
+      "necessarily": 43311,
+      "ĠRoots": 43312,
+      "Pose": 43313,
+      "Tu": 43314,
+      "Ġclue": 43315,
+      "Ġperturbing": 43316,
+      "ĠHelium": 43317,
+      "ĠCombustion": 43318,
+      "nitrogen": 43319,
+      "amplified": 43320,
+      "prove": 43321,
+      "ĠSoils": 43322,
+      "normalization": 43323,
+      "ĠCHOP": 43324,
+      "ĠMcLe": 43325,
+      "Ġstrikes": 43326,
+      "Ġcropped": 43327,
+      "ĠKuo": 43328,
+      "Ġvagal": 43329,
+      "Ġdinucleotide": 43330,
+      "ĠIsaac": 43331,
+      "ĠLOX": 43332,
+      "Ġdirectionality": 43333,
+      "Ġchemoradiotherapy": 43334,
+      "calculus": 43335,
+      "ĠMohammed": 43336,
+      "mapped": 43337,
+      "Ġreforms": 43338,
+      "Ġreordering": 43339,
+      "ĠBm": 43340,
+      "ĠESCs": 43341,
+      "ĠNUC": 43342,
+      "thaw": 43343,
+      "Ġnanoporous": 43344,
+      "Ġtrainable": 43345,
+      "ĠATT": 43346,
+      "feats": 43347,
+      "OFDM": 43348,
+      "ĠSHP": 43349,
+      "ĠRichter": 43350,
+      "Ġsprayed": 43351,
+      "ĠJefferson": 43352,
+      "FOX": 43353,
+      "bh": 43354,
+      "otte": 43355,
+      "Ġleiomy": 43356,
+      "ospores": 43357,
+      "specificity": 43358,
+      "ĠRefer": 43359,
+      "ĠHaas": 43360,
+      "Move": 43361,
+      "Materials": 43362,
+      "tec": 43363,
+      "utility": 43364,
+      "entional": 43365,
+      "ĠMPP": 43366,
+      "chond": 43367,
+      "Ġseepage": 43368,
+      "Ġpeach": 43369,
+      "ĠÎĶt": 43370,
+      "embryonic": 43371,
+      "Yan": 43372,
+      "Ġliposomal": 43373,
+      "ĠValencia": 43374,
+      "ĠEndo": 43375,
+      "ĠPAO": 43376,
+      "Ġdialect": 43377,
+      "Ġchondrocyte": 43378,
+      "ĠMillimeter": 43379,
+      "ĠRegularity": 43380,
+      "destroy": 43381,
+      "ĠCondensation": 43382,
+      "Bayes": 43383,
+      "abundance": 43384,
+      "ĠdU": 43385,
+      "ĠSSI": 43386,
+      "ĠHAND": 43387,
+      "Ġconsulted": 43388,
+      "Ġsuppliers": 43389,
+      "Ġdemo": 43390,
+      "registered": 43391,
+      "Ġmicrosomal": 43392,
+      "Ġlambs": 43393,
+      "responsiveness": 43394,
+      "Dy": 43395,
+      "GAS": 43396,
+      "UME": 43397,
+      "Ġaero": 43398,
+      "Ġcalmodulin": 43399,
+      "Ġcalcined": 43400,
+      "Ġinsula": 43401,
+      "ĠMei": 43402,
+      "ĠREAL": 43403,
+      "Ġcontractible": 43404,
+      "ĠEssentially": 43405,
+      "Ġgaming": 43406,
+      "Ġspillover": 43407,
+      "residues": 43408,
+      "âİ": 43409,
+      "ĠEMC": 43410,
+      "ĠSDE": 43411,
+      "ĠSerine": 43412,
+      "ecki": 43413,
+      "ĠPrinceton": 43414,
+      "ĠBACKGROUND": 43415,
+      "masks": 43416,
+      "ĠLom": 43417,
+      "ffield": 43418,
+      "efitinib": 43419,
+      "Ġpatents": 43420,
+      "ĠBez": 43421,
+      "loads": 43422,
+      "Ġgonadal": 43423,
+      "Ġnitrocellulose": 43424,
+      "âĻĤ": 43425,
+      "Ġthrown": 43426,
+      "Ġrectification": 43427,
+      "mina": 43428,
+      "iscid": 43429,
+      "ĠBiobank": 43430,
+      "paramagnetic": 43431,
+      "GSK": 43432,
+      "ĠDerivative": 43433,
+      "criterion": 43434,
+      "ĠMonthly": 43435,
+      "ë¥": 43436,
+      "ĠSichuan": 43437,
+      "Ġimmunologic": 43438,
+      "Ġheterotic": 43439,
+      "ĠMcCl": 43440,
+      "ĠSMART": 43441,
+      "ĠBatteries": 43442,
+      "Ġpremiered": 43443,
+      "Ġcryopreservation": 43444,
+      "Nu": 43445,
+      "valho": 43446,
+      "Ġflotation": 43447,
+      "topological": 43448,
+      "ĠNanjing": 43449,
+      "Ġjuxt": 43450,
+      "ĠFeder": 43451,
+      "Ġprofoundly": 43452,
+      "cad": 43453,
+      "ienced": 43454,
+      "chuk": 43455,
+      "ĠIng": 43456,
+      "ĠKSHV": 43457,
+      "aminobenz": 43458,
+      "ĉĉĉĠĠĠ": 43459,
+      "Ġmetaph": 43460,
+      "ĠEpidemic": 43461,
+      "ĠAssociate": 43462,
+      "Ġsaccade": 43463,
+      "Ġdawn": 43464,
+      "Ġreheating": 43465,
+      "Ġspell": 43466,
+      "fractive": 43467,
+      "ĠToolkit": 43468,
+      "Ġrecognise": 43469,
+      "pathogen": 43470,
+      "Ġophthalmic": 43471,
+      "Ġqueried": 43472,
+      "thens": 43473,
+      "ithine": 43474,
+      "umably": 43475,
+      "Ġstrides": 43476,
+      "haul": 43477,
+      "Ġpassion": 43478,
+      "Ġdysfunctions": 43479,
+      "Byte": 43480,
+      "Ġcaesarean": 43481,
+      "prey": 43482,
+      "ĠHorse": 43483,
+      "ĠGABAA": 43484,
+      "Natural": 43485,
+      "kos": 43486,
+      "inators": 43487,
+      "odings": 43488,
+      "ARRAY": 43489,
+      "Ġunipotent": 43490,
+      "Ġelectromy": 43491,
+      "compart": 43492,
+      "Liu": 43493,
+      "encephalic": 43494,
+      "ĠCOMPAR": 43495,
+      "Ġsymbionts": 43496,
+      "ivacaine": 43497,
+      "OI": 43498,
+      "PVA": 43499,
+      "ĠNVIDIA": 43500,
+      "calibrated": 43501,
+      "Ġquest": 43502,
+      "NAD": 43503,
+      "ĠXyl": 43504,
+      "Ġpharmacist": 43505,
+      "directly": 43506,
+      "Ġquadruple": 43507,
+      "ethanone": 43508,
+      "ĠBulgaria": 43509,
+      "Ġoviposition": 43510,
+      "runs": 43511,
+      "Ġnociceptive": 43512,
+      "Ġasexual": 43513,
+      "SULT": 43514,
+      "Ġwouldn": 43515,
+      "ĠIndustries": 43516,
+      "abilizing": 43517,
+      "ĠCompressive": 43518,
+      "COOH": 43519,
+      "USH": 43520,
+      "kiewicz": 43521,
+      "Ġigneous": 43522,
+      "Ġdisappoint": 43523,
+      "ĠCKM": 43524,
+      "ĠDiagrams": 43525,
+      "ĠFlam": 43526,
+      "ĠGould": 43527,
+      "Ġcoenzyme": 43528,
+      "Ġparan": 43529,
+      "Ġ¶": 43530,
+      "Ġprogrammer": 43531,
+      "ĠTransforming": 43532,
+      "Ġmuscarinic": 43533,
+      "onucleotide": 43534,
+      "FIELD": 43535,
+      "ĠFuji": 43536,
+      "Ġnondec": 43537,
+      "Ġblanket": 43538,
+      "Ġpredisposing": 43539,
+      "ĠTrigger": 43540,
+      "Ġwelcome": 43541,
+      "Family": 43542,
+      "UINT": 43543,
+      "hfill": 43544,
+      "tvb": 43545,
+      "ĠBatt": 43546,
+      "Ġunmet": 43547,
+      "ĠApo": 43548,
+      "otient": 43549,
+      "Ġfundus": 43550,
+      "ĠLearned": 43551,
+      "Ġintrusions": 43552,
+      "Ġsolubilization": 43553,
+      "fundamental": 43554,
+      "ĠSantiago": 43555,
+      "Ġhpi": 43556,
+      "throw": 43557,
+      "ĠInto": 43558,
+      "timeout": 43559,
+      "Ġthickened": 43560,
+      "iasm": 43561,
+      "Ġgravitino": 43562,
+      "branched": 43563,
+      "VIII": 43564,
+      "Ġoch": 43565,
+      "Ġgym": 43566,
+      "ĠKrylov": 43567,
+      "Ġcorrective": 43568,
+      "ĠInstitution": 43569,
+      "Ġcrimes": 43570,
+      "ĠBacteroidetes": 43571,
+      "ĠEhr": 43572,
+      "Ġseated": 43573,
+      "rolizumab": 43574,
+      "Ġfactorized": 43575,
+      "rotational": 43576,
+      "Ġadministrators": 43577,
+      "âĭĨ": 43578,
+      "ineralization": 43579,
+      "lining": 43580,
+      "âĹ": 43581,
+      "urai": 43582,
+      "ĠFAP": 43583,
+      "ĠFisheries": 43584,
+      "ĠESO": 43585,
+      "temper": 43586,
+      "Biggr": 43587,
+      "ĠAlternating": 43588,
+      "twin": 43589,
+      "amatsu": 43590,
+      "Ġintrad": 43591,
+      "overflow": 43592,
+      "Ġcomparability": 43593,
+      "Ġsynoptic": 43594,
+      "USB": 43595,
+      "dbg": 43596,
+      "demonstr": 43597,
+      "ĠAchieving": 43598,
+      "Ġtectonics": 43599,
+      "ĠRandall": 43600,
+      "ĠPrepared": 43601,
+      "Ġsublimation": 43602,
+      "ĠBaj": 43603,
+      "Ġclutch": 43604,
+      "Ġsubdomain": 43605,
+      "Ġflaws": 43606,
+      "influ": 43607,
+      "Ġwidening": 43608,
+      "Ġmelted": 43609,
+      "Ġadministrator": 43610,
+      "Ġsubsidiary": 43611,
+      "ĠPricing": 43612,
+      "ticus": 43613,
+      "ogi": 43614,
+      "ĠAlign": 43615,
+      "ĠADV": 43616,
+      "Ġvastly": 43617,
+      "benchmark": 43618,
+      "Ġprioritize": 43619,
+      "Radi": 43620,
+      "essed": 43621,
+      "Ġsuprac": 43622,
+      "accard": 43623,
+      "Ġbiomimetic": 43624,
+      "ĠIrradiation": 43625,
+      "ĠALGOR": 43626,
+      "Ġpedigree": 43627,
+      "ĠCMT": 43628,
+      "odym": 43629,
+      "ĠVig": 43630,
+      "ĠBiochemistry": 43631,
+      "ĠAccum": 43632,
+      "Indices": 43633,
+      "hardtii": 43634,
+      "ĠRalph": 43635,
+      "Ġruminants": 43636,
+      "iT": 43637,
+      "onau": 43638,
+      "aner": 43639,
+      "planned": 43640,
+      "evers": 43641,
+      "Ġretroviral": 43642,
+      "Ġquantifier": 43643,
+      "ĠExtracting": 43644,
+      "Ġacetylated": 43645,
+      "Orth": 43646,
+      "ĠSenator": 43647,
+      "Ġnanosecond": 43648,
+      "Ġanticipation": 43649,
+      "ĠECMO": 43650,
+      "Ġsemicirc": 43651,
+      "ĠCryptosporidium": 43652,
+      "ĠTARGET": 43653,
+      "Ġapples": 43654,
+      "efield": 43655,
+      "Ġreman": 43656,
+      "Ġserovar": 43657,
+      "ĠTransactions": 43658,
+      "transitions": 43659,
+      "ursions": 43660,
+      "ĠMelatonin": 43661,
+      "Ġcholecystectomy": 43662,
+      "ĠAntiviral": 43663,
+      "hous": 43664,
+      "Ġotol": 43665,
+      "Ġmaj": 43666,
+      "Ġeclip": 43667,
+      "arel": 43668,
+      "ATIONAL": 43669,
+      "MIM": 43670,
+      "ĠCImg": 43671,
+      "ĠEndomet": 43672,
+      "ĠHayashi": 43673,
+      "Ġchimpanzees": 43674,
+      "mbf": 43675,
+      "ĠIPV": 43676,
+      "actoring": 43677,
+      "outside": 43678,
+      "neapolis": 43679,
+      "Ġdiscarding": 43680,
+      "numtype": 43681,
+      "ĠREST": 43682,
+      "Ġflagellar": 43683,
+      "ĠChandrase": 43684,
+      "hofer": 43685,
+      "Ġelectrocardiogram": 43686,
+      "Gb": 43687,
+      "mock": 43688,
+      "oeb": 43689,
+      "ĠSMO": 43690,
+      "ĠMord": 43691,
+      "ĠBoz": 43692,
+      "Ġminors": 43693,
+      "INLINE": 43694,
+      "Ġthermogravimetric": 43695,
+      "ĠMelting": 43696,
+      "ĠNSW": 43697,
+      "Sham": 43698,
+      "lotinib": 43699,
+      "Ġacquisitions": 43700,
+      "taz": 43701,
+      "Ġdefaults": 43702,
+      "Ġoscillates": 43703,
+      "ĠCaption": 43704,
+      "Ġdisruptive": 43705,
+      "Ġsweeping": 43706,
+      "ĠToolbox": 43707,
+      "Ġurethral": 43708,
+      "HBV": 43709,
+      "ĠRCS": 43710,
+      "Ġoxys": 43711,
+      "immuno": 43712,
+      "htm": 43713,
+      "oflavin": 43714,
+      "HIF": 43715,
+      "ĠSBA": 43716,
+      "ĠCPE": 43717,
+      "Ġwhites": 43718,
+      "ĠReactor": 43719,
+      "Ġpurp": 43720,
+      "Ġelectrocatalytic": 43721,
+      "Ġnonex": 43722,
+      "Ġtyphimurium": 43723,
+      "Ġeurop": 43724,
+      "concave": 43725,
+      "macrophage": 43726,
+      "SER": 43727,
+      "Ġlapse": 43728,
+      "Ġanatom": 43729,
+      "ĠRAC": 43730,
+      "tax": 43731,
+      "Ġmins": 43732,
+      "Ġsensu": 43733,
+      "ĠHebrew": 43734,
+      "Ġrealism": 43735,
+      "ĠMicroglia": 43736,
+      "Ġserialized": 43737,
+      "ĠHazard": 43738,
+      "Ġmetamorphosis": 43739,
+      "ĠIRA": 43740,
+      "Ġsmearing": 43741,
+      "Ġphotolysis": 43742,
+      "Ġchildbirth": 43743,
+      "Ġsilenced": 43744,
+      "rawal": 43745,
+      "Ġquadrants": 43746,
+      "butanol": 43747,
+      "Ġstochastically": 43748,
+      "ĠChambers": 43749,
+      "ĠNavarro": 43750,
+      "Ġprocurement": 43751,
+      "cite": 43752,
+      "ĠSle": 43753,
+      "ĠHadoop": 43754,
+      "Ġdelaying": 43755,
+      "Atlantic": 43756,
+      "Spain": 43757,
+      "falfa": 43758,
+      "odialysis": 43759,
+      "ynia": 43760,
+      "Ġplateaus": 43761,
+      "Ġmultimode": 43762,
+      "RESET": 43763,
+      "ĠRocky": 43764,
+      "ĠRodrigues": 43765,
+      "fMRI": 43766,
+      "rint": 43767,
+      "ĠTAL": 43768,
+      "Ġspecular": 43769,
+      "construction": 43770,
+      "ĠAthens": 43771,
+      "Ġcrosslink": 43772,
+      "Ġcountably": 43773,
+      "Ġspreadsheet": 43774,
+      "cribes": 43775,
+      "consistently": 43776,
+      "Ġfloodplain": 43777,
+      "EINVAL": 43778,
+      "Maca": 43779,
+      "Ġei": 43780,
+      "Ġhitherto": 43781,
+      "Ġsemif": 43782,
+      "Ġcontinual": 43783,
+      "ĠHomology": 43784,
+      "Ġphotocatalysts": 43785,
+      "isable": 43786,
+      "ĠHAT": 43787,
+      "Ġpolyhedra": 43788,
+      "ĠMayo": 43789,
+      "Ġlactating": 43790,
+      "sampler": 43791,
+      "Ġappliances": 43792,
+      "TU": 43793,
+      "Ġchess": 43794,
+      "ĠTing": 43795,
+      "Ġinvitation": 43796,
+      "Ġdistributing": 43797,
+      "ashima": 43798,
+      "Ġultral": 43799,
+      "trend": 43800,
+      "Ġrelaxations": 43801,
+      "ĠHelen": 43802,
+      "Ġbedding": 43803,
+      "Ġglandular": 43804,
+      "Ġincrementally": 43805,
+      "Ġconceal": 43806,
+      "claimed": 43807,
+      "ĠEddy": 43808,
+      "Ġmos": 43809,
+      "ĠTube": 43810,
+      "ĠToda": 43811,
+      "raj": 43812,
+      "ĠMü": 43813,
+      "ĠUll": 43814,
+      "Ġune": 43815,
+      "berine": 43816,
+      "Ġpolicym": 43817,
+      "Ġscholarly": 43818,
+      "Ġshoreline": 43819,
+      "Ġaldosterone": 43820,
+      "ĠBrucella": 43821,
+      "THE": 43822,
+      "REAL": 43823,
+      "Ġexome": 43824,
+      "ĠDAB": 43825,
+      "Ġextras": 43826,
+      "Ġbanding": 43827,
+      "ĠSiemens": 43828,
+      "ĠBoost": 43829,
+      "ĠSupernovae": 43830,
+      "ĠTracing": 43831,
+      "Ġascorbate": 43832,
+      "Italy": 43833,
+      "bund": 43834,
+      "Ġdecrement": 43835,
+      "Ġneurophysiological": 43836,
+      "Ġblackbody": 43837,
+      "ĠMcN": 43838,
+      "Ġcompetencies": 43839,
+      "oscape": 43840,
+      "ĠHonours": 43841,
+      "Ġmastitis": 43842,
+      "criteria": 43843,
+      "Ġbiaxial": 43844,
+      "Ġthawed": 43845,
+      "ĠFoll": 43846,
+      "oreceptor": 43847,
+      "Ġinvention": 43848,
+      "ADs": 43849,
+      "Show": 43850,
+      "------------------------------------------------": 43851,
+      "Ġellipsoidal": 43852,
+      "Ġfocussed": 43853,
+      "ĠDat": 43854,
+      "ĠRim": 43855,
+      "ĠLX": 43856,
+      "ĠGER": 43857,
+      "insler": 43858,
+      "Ġtopoisomerase": 43859,
+      "Ġhyperlipidemia": 43860,
+      "Ġmystery": 43861,
+      "Ġnitrification": 43862,
+      "Ġoncogenes": 43863,
+      "ĠFuller": 43864,
+      "ĠBartlett": 43865,
+      "Ġamphibians": 43866,
+      "SST": 43867,
+      "bly": 43868,
+      "leads": 43869,
+      "ecycle": 43870,
+      "ampl": 43871,
+      "ĠPOM": 43872,
+      "ĠDCF": 43873,
+      "strass": 43874,
+      "antibody": 43875,
+      "nonlinear": 43876,
+      "ĠBroadway": 43877,
+      "ĠCatalogue": 43878,
+      "ĠμA": 43879,
+      "Ġacetaminophen": 43880,
+      "Ġcrystallites": 43881,
+      "ĠNanotubes": 43882,
+      "ĠAcknowledgment": 43883,
+      "Ġmetamorphism": 43884,
+      "Ġtwinning": 43885,
+      "ĠAzerbai": 43886,
+      "xA": 43887,
+      "CCC": 43888,
+      "ĠSolids": 43889,
+      "preds": 43890,
+      "ĠMontana": 43891,
+      "WRITE": 43892,
+      "Ratio": 43893,
+      "Ġpunch": 43894,
+      "Ġriding": 43895,
+      "Ġacne": 43896,
+      "ĠUre": 43897,
+      "Ġcorr": 43898,
+      "ĠQOL": 43899,
+      "Ġinsult": 43900,
+      "Ġdominantly": 43901,
+      "Ġsubsamples": 43902,
+      "rews": 43903,
+      "ĠPreservation": 43904,
+      "ĠAffine": 43905,
+      "methanone": 43906,
+      "Ġhedgehog": 43907,
+      "JH": 43908,
+      "Ġlined": 43909,
+      "Ġsten": 43910,
+      "ĠDarmstadt": 43911,
+      "ĠLasso": 43912,
+      "Ġdeproton": 43913,
+      "Ġshoes": 43914,
+      "Ġmotives": 43915,
+      "Ġmicroscop": 43916,
+      "ophthora": 43917,
+      "Ġmacron": 43918,
+      "Ġencouragement": 43919,
+      "acrylic": 43920,
+      "ĠTensorFlow": 43921,
+      "Wrapper": 43922,
+      "oise": 43923,
+      "ayak": 43924,
+      "Ġrepresses": 43925,
+      "Ġpruned": 43926,
+      "ĠClar": 43927,
+      "ĠâĬ²": 43928,
+      "ĠUnderlying": 43929,
+      "Implemented": 43930,
+      "Ġsweat": 43931,
+      "Ġmeteorites": 43932,
+      "Ġtweez": 43933,
+      "Ġprosocial": 43934,
+      "Ġabrasion": 43935,
+      "hail": 43936,
+      "Ġshorth": 43937,
+      "ismatch": 43938,
+      "INTR": 43939,
+      "ĠTrin": 43940,
+      "Ġphysicists": 43941,
+      "ĠPEO": 43942,
+      "ĠMagneto": 43943,
+      "ĠJacobson": 43944,
+      "ĠMMPs": 43945,
+      "ĠIntravenous": 43946,
+      "Ġneurotransmission": 43947,
+      "ĠPneumonia": 43948,
+      "Lind": 43949,
+      "yre": 43950,
+      "Ġmaternity": 43951,
+      "ĠIris": 43952,
+      "riatal": 43953,
+      "ĠâĢļ": 43954,
+      "medetomidine": 43955,
+      "Ġtriterpen": 43956,
+      "ĠManuscript": 43957,
+      "ĠEndoplasmic": 43958,
+      "ĠPotter": 43959,
+      "Ġgerminal": 43960,
+      "Ġphotosystem": 43961,
+      "Guided": 43962,
+      "Ġguitarist": 43963,
+      "ĠChilean": 43964,
+      "ĠSalvador": 43965,
+      "ÉĻ": 43966,
+      "Ġcelestial": 43967,
+      "omand": 43968,
+      "Ġnk": 43969,
+      "Ġvendors": 43970,
+      "ĠPINK": 43971,
+      "ĠInorganic": 43972,
+      "Ġmoderated": 43973,
+      "SUS": 43974,
+      "ĠJoshi": 43975,
+      "ĠStata": 43976,
+      "ikes": 43977,
+      "oye": 43978,
+      "ĠJohnny": 43979,
+      "Leica": 43980,
+      "Ġkaon": 43981,
+      "ĠEquipment": 43982,
+      "Kim": 43983,
+      "gado": 43984,
+      "tures": 43985,
+      "Ġelem": 43986,
+      "ĠAAC": 43987,
+      "cea": 43988,
+      "odality": 43989,
+      "Ġaniline": 43990,
+      "Ġexothermic": 43991,
+      "ĠGunn": 43992,
+      "ĠJU": 43993,
+      "plicable": 43994,
+      "scapes": 43995,
+      "typed": 43996,
+      "Ġinspiratory": 43997,
+      "REGIST": 43998,
+      "ĠBryan": 43999,
+      "Ġanxious": 44000,
+      "ĠCarpenter": 44001,
+      "ĠPharmacokinetics": 44002,
+      "inferior": 44003,
+      "Frag": 44004,
+      "ZY": 44005,
+      "Ġoesophageal": 44006,
+      "ĠSuk": 44007,
+      "ĠPron": 44008,
+      "ĠCDI": 44009,
+      "AGC": 44010,
+      "keywords": 44011,
+      "susceptible": 44012,
+      "Germany": 44013,
+      "MAS": 44014,
+      "iC": 44015,
+      "anmar": 44016,
+      "Ġexiting": 44017,
+      "ĠHale": 44018,
+      "Ġrhamn": 44019,
+      "industrial": 44020,
+      "Ġraft": 44021,
+      "embrolizumab": 44022,
+      "Ġdeploying": 44023,
+      "Ġsalicylic": 44024,
+      "Rn": 44025,
+      "Ġcensor": 44026,
+      "ĠdX": 44027,
+      "Ġforum": 44028,
+      "MSI": 44029,
+      "blad": 44030,
+      "Ġsquir": 44031,
+      "CPP": 44032,
+      "Ġgrapevine": 44033,
+      "ĠRAFT": 44034,
+      "Monte": 44035,
+      "Ġmicroflora": 44036,
+      "rcl": 44037,
+      "Ġdecap": 44038,
+      "ANC": 44039,
+      "Ġbroaden": 44040,
+      "Ġfreed": 44041,
+      "Ġsouthward": 44042,
+      "ĠJacques": 44043,
+      "Ġrequesting": 44044,
+      "ĠAspect": 44045,
+      "arajan": 44046,
+      "Failed": 44047,
+      "fprintf": 44048,
+      "pytest": 44049,
+      "ʹ": 44050,
+      "ĠCm": 44051,
+      "until": 44052,
+      "neiss": 44053,
+      "Ġmonos": 44054,
+      "ospinal": 44055,
+      "olsky": 44056,
+      "contrib": 44057,
+      "Container": 44058,
+      "ĠVolunte": 44059,
+      "ĠAttributes": 44060,
+      "ĠTumour": 44061,
+      "Ġreinhardtii": 44062,
+      "Ġcentromere": 44063,
+      "ĠSymph": 44064,
+      "ĠAo": 44065,
+      "agens": 44066,
+      "pleted": 44067,
+      "ieder": 44068,
+      "Ġactivist": 44069,
+      "ĠAlmeida": 44070,
+      "Ġdisturbing": 44071,
+      "Ġreflexes": 44072,
+      "DSS": 44073,
+      "Ġforwards": 44074,
+      "ronym": 44075,
+      "ĠIntegrity": 44076,
+      "WEEN": 44077,
+      "Ġôı¼Į": 44078,
+      "Ġfaithfully": 44079,
+      "Ġpericardial": 44080,
+      "Japanese": 44081,
+      "ĠCENP": 44082,
+      "Kr": 44083,
+      "Ġdefending": 44084,
+      "Ġzon": 44085,
+      "insensitive": 44086,
+      "Ġlabs": 44087,
+      "ĠCaM": 44088,
+      "ĠEurop": 44089,
+      "MEA": 44090,
+      "BLAST": 44091,
+      "xN": 44092,
+      "alen": 44093,
+      "Ġclk": 44094,
+      "lineage": 44095,
+      "coating": 44096,
+      "Ġtailoring": 44097,
+      "CONTR": 44098,
+      "Ġadrenergic": 44099,
+      "interpreted": 44100,
+      "NIH": 44101,
+      "amoeba": 44102,
+      "ĠCyr": 44103,
+      "Ġtriplicates": 44104,
+      "defining": 44105,
+      "ĠLogan": 44106,
+      "tesy": 44107,
+      "ĠTwist": 44108,
+      "ĠGrammar": 44109,
+      "ĠSustained": 44110,
+      "Ġanharmonic": 44111,
+      "Ġalve": 44112,
+      "Ġruler": 44113,
+      "Ġquanta": 44114,
+      "Ġdirects": 44115,
+      "Ġoffloading": 44116,
+      "ĠGeophysical": 44117,
+      "Require": 44118,
+      "Ġhepatoma": 44119,
+      "Ġfoo": 44120,
+      "ĠGeorges": 44121,
+      "Ġbouts": 44122,
+      "ĠTAK": 44123,
+      "Ġantidiabetic": 44124,
+      "ĠReported": 44125,
+      "presenting": 44126,
+      "ĠLayered": 44127,
+      "RENCE": 44128,
+      "Ġuveitis": 44129,
+      "bage": 44130,
+      "Ġfentanyl": 44131,
+      "ensemble": 44132,
+      "ĠOSCC": 44133,
+      "Ġminers": 44134,
+      "looking": 44135,
+      "ĠBeer": 44136,
+      "precipitation": 44137,
+      "ĠEnterprise": 44138,
+      "Ġserotonergic": 44139,
+      "Ġseesaw": 44140,
+      "ĠAthletics": 44141,
+      "Ġhydrolytic": 44142,
+      "Ġtalent": 44143,
+      "Ġdiscernible": 44144,
+      "FIL": 44145,
+      "lives": 44146,
+      "ĠSales": 44147,
+      "ĠSSc": 44148,
+      "erend": 44149,
+      "clim": 44150,
+      "antid": 44151,
+      "INTS": 44152,
+      "Ġattenuating": 44153,
+      "Ġtwists": 44154,
+      "Ġoxygenase": 44155,
+      "rini": 44156,
+      "Macaulay": 44157,
+      "zm": 44158,
+      "ĠPOT": 44159,
+      "ĠMp": 44160,
+      "ĠHey": 44161,
+      "ĠOVER": 44162,
+      "illion": 44163,
+      "Ġinvaluable": 44164,
+      "Ġantiplatelet": 44165,
+      "Ġmutans": 44166,
+      "Ġgraduates": 44167,
+      "GRAM": 44168,
+      "ispheric": 44169,
+      "Ġincompatibility": 44170,
+      "Ġcarboxylase": 44171,
+      "Ġbiocontrol": 44172,
+      "ĠPhysicochemical": 44173,
+      "ïĻ": 44174,
+      "Ġlae": 44175,
+      "ĠAortic": 44176,
+      "ĠRacing": 44177,
+      "ĠECD": 44178,
+      "ivic": 44179,
+      "Ġelectromechanical": 44180,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 44181,
+      "Ġsteer": 44182,
+      "ĠOutside": 44183,
+      "Ġadenocarcinomas": 44184,
+      "ĠKeep": 44185,
+      "Ġcocon": 44186,
+      "Ġmoderating": 44187,
+      "Ġreformulated": 44188,
+      "Ġfundamentals": 44189,
+      "ĠTesla": 44190,
+      "ĠStirling": 44191,
+      "orated": 44192,
+      "opid": 44193,
+      "Ġparox": 44194,
+      "Ġtrivalent": 44195,
+      "Ġexchangeable": 44196,
+      "Ġdebuted": 44197,
+      "Very": 44198,
+      "reements": 44199,
+      "ĠTomm": 44200,
+      "ĠCyn": 44201,
+      "ĠCatalysts": 44202,
+      "quat": 44203,
+      "ĠFER": 44204,
+      "ĠRum": 44205,
+      "Ġscanners": 44206,
+      "ĠâĨĴâĪŀ": 44207,
+      "otropical": 44208,
+      "Ġvenues": 44209,
+      "estimator": 44210,
+      "Ġemptying": 44211,
+      "GPP": 44212,
+      "VIR": 44213,
+      "Ġcomplicates": 44214,
+      "ĠNIS": 44215,
+      "ĠZhen": 44216,
+      "ĠBlues": 44217,
+      "Ġtextbooks": 44218,
+      "Ġsixty": 44219,
+      "Ġethers": 44220,
+      "Ġfinancially": 44221,
+      "ĠmHealth": 44222,
+      "ĠTut": 44223,
+      "Ġlaryng": 44224,
+      "ĠGs": 44225,
+      "ĠWatch": 44226,
+      "Ġsev": 44227,
+      "Ġital": 44228,
+      "assed": 44229,
+      "Ġ÷": 44230,
+      "ĠConsent": 44231,
+      "Ġnuts": 44232,
+      "vitreal": 44233,
+      "Ġmetaphase": 44234,
+      "Ġtitania": 44235,
+      "Ġfoils": 44236,
+      "Ġgalectin": 44237,
+      "initialize": 44238,
+      "Ġadvisor": 44239,
+      "Ġadministering": 44240,
+      "Bool": 44241,
+      "Ġcem": 44242,
+      "Ġreforming": 44243,
+      "Ġgn": 44244,
+      "ysh": 44245,
+      "Ġattor": 44246,
+      "SCI": 44247,
+      "Exc": 44248,
+      "builder": 44249,
+      "Ġcerium": 44250,
+      "Ġregistries": 44251,
+      "ĠMatsumoto": 44252,
+      "Ġtempting": 44253,
+      "isha": 44254,
+      "Ġreorientation": 44255,
+      "ĠMold": 44256,
+      "ĠRAGE": 44257,
+      "yson": 44258,
+      "Ġunequiv": 44259,
+      "Ġrelocation": 44260,
+      "ĠÃķ": 44261,
+      "ĠReform": 44262,
+      "ĠREQU": 44263,
+      "Ġcommensurate": 44264,
+      "catalog": 44265,
+      "ĠTPS": 44266,
+      "Ġlamb": 44267,
+      "Ġprefactor": 44268,
+      "archy": 44269,
+      "Ġdopants": 44270,
+      "drv": 44271,
+      "ĠPARAMET": 44272,
+      "schedule": 44273,
+      "ochemically": 44274,
+      "ĠeHealth": 44275,
+      "unas": 44276,
+      "ĠPinus": 44277,
+      "ĠHSA": 44278,
+      "Ġinterrelations": 44279,
+      "Ġdepot": 44280,
+      "ĠPlatinum": 44281,
+      "Ġlifelong": 44282,
+      "Ġpersistently": 44283,
+      "ĠParadox": 44284,
+      "ĠConformational": 44285,
+      "esophag": 44286,
+      "ĠAAT": 44287,
+      "plin": 44288,
+      "ĠFCN": 44289,
+      "ĠDt": 44290,
+      "oposide": 44291,
+      "Ġchal": 44292,
+      "Ġhalt": 44293,
+      "ĠDetect": 44294,
+      "Ġdiscriminated": 44295,
+      "ĠLagrangians": 44296,
+      "Appro": 44297,
+      "Ġȧ": 44298,
+      "Ġimpulsivity": 44299,
+      "BAT": 44300,
+      "Chemical": 44301,
+      "gather": 44302,
+      "ĠUNC": 44303,
+      "intron": 44304,
+      "ĠSimulator": 44305,
+      "ĠGla": 44306,
+      "TTT": 44307,
+      "ĠVolatile": 44308,
+      "Ġsubsid": 44309,
+      "ĠBroadcasting": 44310,
+      "Ġstreptozotocin": 44311,
+      "Ġfumar": 44312,
+      "ĠMPEG": 44313,
+      "Ġinfluenzae": 44314,
+      "subjects": 44315,
+      "Ġappropriateness": 44316,
+      "Ġarcmin": 44317,
+      "Ġstranded": 44318,
+      "oylation": 44319,
+      "ĠDEX": 44320,
+      "oviral": 44321,
+      "ĠQuarter": 44322,
+      "colytic": 44323,
+      "Ġfriendship": 44324,
+      "HES": 44325,
+      "loxacin": 44326,
+      "Ġere": 44327,
+      "ĠTrad": 44328,
+      "uristics": 44329,
+      "ĠECT": 44330,
+      "ĠEGCG": 44331,
+      "ĠLRP": 44332,
+      "ĠGAG": 44333,
+      "ĠInP": 44334,
+      "Ġcontempor": 44335,
+      "Ġmicror": 44336,
+      "ierstrass": 44337,
+      "ĠElectrosp": 44338,
+      "needed": 44339,
+      "atmosphere": 44340,
+      "nT": 44341,
+      "Ġbandwidths": 44342,
+      "Ġdiversified": 44343,
+      "ĠAppropriate": 44344,
+      "restore": 44345,
+      "rocnem": 44346,
+      "ĠLaguerre": 44347,
+      "ĠSongs": 44348,
+      "ĠKaluza": 44349,
+      "ĠSymmetries": 44350,
+      "ĠSchmitt": 44351,
+      "Ġbiomolecular": 44352,
+      "scalebox": 44353,
+      "Ġintrahepatic": 44354,
+      "understanding": 44355,
+      "ĠABCG": 44356,
+      "Ġunderestimates": 44357,
+      "ĠStreaming": 44358,
+      "Ġfictitious": 44359,
+      "oplasmosis": 44360,
+      "resident": 44361,
+      "ĠBary": 44362,
+      "ĠComa": 44363,
+      "scrip": 44364,
+      "Ġdegran": 44365,
+      "ĠCaMKII": 44366,
+      "ĠBalmer": 44367,
+      "ĠPlasm": 44368,
+      "Ġchelating": 44369,
+      "ĠParadigm": 44370,
+      "Ġopponents": 44371,
+      "EK": 44372,
+      "Pin": 44373,
+      "Ġmsec": 44374,
+      "adone": 44375,
+      "acht": 44376,
+      "CCG": 44377,
+      "ECO": 44378,
+      "normalize": 44379,
+      "ĠDesigns": 44380,
+      "Ġyellowish": 44381,
+      "glutamyl": 44382,
+      "Ġdomestication": 44383,
+      "Ġmonophyletic": 44384,
+      "dles": 44385,
+      "nested": 44386,
+      "ĠGrace": 44387,
+      "ĠStudios": 44388,
+      "ĠDiscussions": 44389,
+      "ophenoxy": 44390,
+      "Ġveterin": 44391,
+      "Ġendosym": 44392,
+      "uttinger": 44393,
+      "batches": 44394,
+      "ĠFiji": 44395,
+      "ĠRNF": 44396,
+      "ousa": 44397,
+      "ĠKY": 44398,
+      "Ġspectrograph": 44399,
+      "ERIC": 44400,
+      "ĠMyanmar": 44401,
+      "ĠConstraining": 44402,
+      "Ġecologically": 44403,
+      "Ġfrost": 44404,
+      "arboux": 44405,
+      "ĠFibonacci": 44406,
+      "Ġcanceled": 44407,
+      "ĠISSN": 44408,
+      "Rect": 44409,
+      "another": 44410,
+      "ĠMMA": 44411,
+      "OLO": 44412,
+      "ĠTruth": 44413,
+      "Ġorthopaedic": 44414,
+      "Ġtraversing": 44415,
+      "ischemic": 44416,
+      "ĠMozambique": 44417,
+      "ĠMSR": 44418,
+      "apace": 44419,
+      "ĠThread": 44420,
+      "ologia": 44421,
+      "Ġcalm": 44422,
+      "methyltransferase": 44423,
+      "Ġͪ": 44424,
+      "Ġdrove": 44425,
+      "Ġcommanded": 44426,
+      "Ġfeline": 44427,
+      "ĠRush": 44428,
+      "ĠLisa": 44429,
+      "Ġsuperspace": 44430,
+      "arcy": 44431,
+      "ĠRegulated": 44432,
+      "ĠResting": 44433,
+      "causing": 44434,
+      "psychotics": 44435,
+      "qt": 44436,
+      "Ġtulare": 44437,
+      "Ġrelocated": 44438,
+      "Ġrepell": 44439,
+      "Ġpredatory": 44440,
+      "people": 44441,
+      "traits": 44442,
+      "Euclidean": 44443,
+      "FDA": 44444,
+      "XRT": 44445,
+      "pC": 44446,
+      "pand": 44447,
+      "ĠÆ": 44448,
+      "reve": 44449,
+      "Ġbids": 44450,
+      "Ġcousin": 44451,
+      "Ġsubdomains": 44452,
+      "tilb": 44453,
+      "énez": 44454,
+      "linearly": 44455,
+      "oproteins": 44456,
+      "Ġcodec": 44457,
+      "Ġcontraception": 44458,
+      "ĠCdk": 44459,
+      "Ġrailroad": 44460,
+      "Bench": 44461,
+      "rng": 44462,
+      "ĠDLA": 44463,
+      "entile": 44464,
+      "ĠCOCO": 44465,
+      "ĠMatth": 44466,
+      "ĠOverl": 44467,
+      "ĠRoutine": 44468,
+      "Ġmultifocal": 44469,
+      "Ġartefact": 44470,
+      "Ġsculpture": 44471,
+      "cies": 44472,
+      "mate": 44473,
+      "ĠØ": 44474,
+      "urek": 44475,
+      "ĠBend": 44476,
+      "ĠNathan": 44477,
+      "Ġdew": 44478,
+      "ymia": 44479,
+      "azzi": 44480,
+      "ĠErk": 44481,
+      "Ġgraduation": 44482,
+      "Boundary": 44483,
+      "Gra": 44484,
+      "Ġbfd": 44485,
+      "ĠSert": 44486,
+      "Ġovershoot": 44487,
+      "ĠSeo": 44488,
+      "Ġsklearn": 44489,
+      "Ġconservatively": 44490,
+      "piracy": 44491,
+      "Ġlaunching": 44492,
+      "XD": 44493,
+      "arbitrary": 44494,
+      "perone": 44495,
+      "Ġshops": 44496,
+      "competitive": 44497,
+      "ĠPakistani": 44498,
+      "Ġcompetitor": 44499,
+      "biotics": 44500,
+      "raits": 44501,
+      "ĠNoble": 44502,
+      "Ġsubregions": 44503,
+      "ĠJump": 44504,
+      "roller": 44505,
+      "tris": 44506,
+      "Ġmacrol": 44507,
+      "ós": 44508,
+      "ĠPenic": 44509,
+      "Ġmicrosomes": 44510,
+      "Ġimprecise": 44511,
+      "Ġdowntown": 44512,
+      "ĠeQTL": 44513,
+      "ifest": 44514,
+      "ĠMFI": 44515,
+      "Ġrarity": 44516,
+      "âĢĻâĢĻ": 44517,
+      "Ġbelts": 44518,
+      "Ġglycosyl": 44519,
+      "ĠNicolas": 44520,
+      "synthesis": 44521,
+      "Oh": 44522,
+      "hierarch": 44523,
+      "pps": 44524,
+      "anets": 44525,
+      "roads": 44526,
+      "ATIC": 44527,
+      "MIMO": 44528,
+      "ĠContract": 44529,
+      "Leib": 44530,
+      "opyrox": 44531,
+      "Ġinformational": 44532,
+      "Synonyms": 44533,
+      "challenge": 44534,
+      "ĠProximal": 44535,
+      "ĠCrawford": 44536,
+      "Ġisopropyl": 44537,
+      "Ġsubfamilies": 44538,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 44539,
+      "Ġannotators": 44540,
+      "Ġreconcile": 44541,
+      "Ġparsimony": 44542,
+      "Ġcaspases": 44543,
+      "cott": 44544,
+      "environments": 44545,
+      "Ġdrm": 44546,
+      "ĠPIL": 44547,
+      "ĠMec": 44548,
+      "ĠInfer": 44549,
+      "ĠSirt": 44550,
+      "Shell": 44551,
+      "agulants": 44552,
+      "seismic": 44553,
+      "Ġsuburban": 44554,
+      "ĠXXX": 44555,
+      "iodes": 44556,
+      "Ġbackpropagation": 44557,
+      "traditional": 44558,
+      "Ġphotocon": 44559,
+      "ĠMicrobiology": 44560,
+      "QT": 44561,
+      "uridine": 44562,
+      "Ġchop": 44563,
+      "ĠThé": 44564,
+      "Ġprejud": 44565,
+      "Ġencoders": 44566,
+      "collected": 44567,
+      "remark": 44568,
+      "Ġsunspot": 44569,
+      "ĠPhenolic": 44570,
+      "Understanding": 44571,
+      "Ġrejecting": 44572,
+      "Ġromantic": 44573,
+      "Ġcentimeters": 44574,
+      "Ġhallucinations": 44575,
+      "Home": 44576,
+      "casted": 44577,
+      "Ġcw": 44578,
+      "rai": 44579,
+      "ĠDisplacement": 44580,
+      "PHY": 44581,
+      "carbam": 44582,
+      "Ġxenon": 44583,
+      "Ġnarratives": 44584,
+      "Ġdollar": 44585,
+      "Ġdynasty": 44586,
+      "ì§": 44587,
+      "Ġinforming": 44588,
+      "ĠOCD": 44589,
+      "ák": 44590,
+      "Ġoverheads": 44591,
+      "juana": 44592,
+      "ĠKraus": 44593,
+      "fx": 44594,
+      "kaya": 44595,
+      "Ġnid": 44596,
+      "ĠGrab": 44597,
+      "Ġinflores": 44598,
+      "Arc": 44599,
+      "============": 44600,
+      "Ġcondenser": 44601,
+      "Ġnanocar": 44602,
+      "ommens": 44603,
+      "Ġsaturating": 44604,
+      "rece": 44605,
+      "elif": 44606,
+      "ĠALE": 44607,
+      "ĠBub": 44608,
+      "ĠLaf": 44609,
+      "andran": 44610,
+      "Ġpouch": 44611,
+      "roline": 44612,
+      "ACHE": 44613,
+      "CCD": 44614,
+      "Ġcoolant": 44615,
+      "Ġgrasslands": 44616,
+      "ĠSynchronous": 44617,
+      "izziness": 44618,
+      "Ġcetuximab": 44619,
+      "Ġdichotomous": 44620,
+      "roch": 44621,
+      "ĠAuckland": 44622,
+      "obesity": 44623,
+      "ikit": 44624,
+      "Ġoperad": 44625,
+      "ĠOnset": 44626,
+      "Ġbeforehand": 44627,
+      "Ġuncomp": 44628,
+      "USED": 44629,
+      "ubbing": 44630,
+      "ĠSMBH": 44631,
+      "ĠExpedition": 44632,
+      "Ġhib": 44633,
+      "ĠPPR": 44634,
+      "ĠNED": 44635,
+      "udio": 44636,
+      "ĠJal": 44637,
+      "ĠArp": 44638,
+      "ĠBee": 44639,
+      "ĠVarieties": 44640,
+      "Comm": 44641,
+      "About": 44642,
+      "ĠAttachment": 44643,
+      "ODULE": 44644,
+      "Calculate": 44645,
+      "Tan": 44646,
+      "inism": 44647,
+      "Ġara": 44648,
+      "Ġcabin": 44649,
+      "Ġconnexin": 44650,
+      "Ġcomets": 44651,
+      "umptive": 44652,
+      "Ġdestabilization": 44653,
+      "ĠHolt": 44654,
+      "ructose": 44655,
+      "anishi": 44656,
+      "plasticity": 44657,
+      "omycosis": 44658,
+      "ovician": 44659,
+      "________________": 44660,
+      "rar": 44661,
+      "Ġwore": 44662,
+      "udine": 44663,
+      "ĠInvariance": 44664,
+      "Ġperitonitis": 44665,
+      "Ġmetrology": 44666,
+      "Ġcloses": 44667,
+      "Ġcolorless": 44668,
+      "Noise": 44669,
+      "DIO": 44670,
+      "ĠLifshitz": 44671,
+      "zul": 44672,
+      "estive": 44673,
+      "ĠMPA": 44674,
+      "ĠBooth": 44675,
+      "ĠDoll": 44676,
+      "arene": 44677,
+      "geness": 44678,
+      "Ġmolecularly": 44679,
+      "ĠPerkin": 44680,
+      "Ġdosimetry": 44681,
+      "ĠSOFT": 44682,
+      "ĠPyTorch": 44683,
+      "Ġquarters": 44684,
+      "ĠKuhn": 44685,
+      "Ġsplenocytes": 44686,
+      "RW": 44687,
+      "cart": 44688,
+      "leb": 44689,
+      "Ġcondom": 44690,
+      "ĠHoc": 44691,
+      "Ġextents": 44692,
+      "Ġslug": 44693,
+      "ĠSupplementation": 44694,
+      "diffic": 44695,
+      "esterly": 44696,
+      "Yu": 44697,
+      "antigens": 44698,
+      "ĠÃĴ": 44699,
+      "Changes": 44700,
+      "Ġpropylene": 44701,
+      "ĠPrison": 44702,
+      "ĠAlgorithmic": 44703,
+      "Ġtolerances": 44704,
+      "Adam": 44705,
+      "Ġesterase": 44706,
+      "Ġmilder": 44707,
+      "ĠConvection": 44708,
+      "PTR": 44709,
+      "kpc": 44710,
+      "Ġexo": 44711,
+      "ĠFah": 44712,
+      "ĠYFP": 44713,
+      "ĠCRM": 44714,
+      "Ġhepatotoxicity": 44715,
+      "Ġnicotinamide": 44716,
+      "Ġpatchy": 44717,
+      "depends": 44718,
+      "ĠpB": 44719,
+      "Ġeel": 44720,
+      "Ġnv": 44721,
+      "ĠSes": 44722,
+      "ĠHZ": 44723,
+      "Ġimprint": 44724,
+      "epileptic": 44725,
+      "fluctuations": 44726,
+      "Ġformalize": 44727,
+      "chev": 44728,
+      "Ġdipping": 44729,
+      "ĠPyramid": 44730,
+      "Ġholo": 44731,
+      "ĠMTs": 44732,
+      "Ġlaminates": 44733,
+      "Ġwormhole": 44734,
+      "LAP": 44735,
+      "hape": 44736,
+      "Ġak": 44737,
+      "Ġreals": 44738,
+      "Ġbystand": 44739,
+      "Ġinterleaved": 44740,
+      "Ġxz": 44741,
+      "ovy": 44742,
+      "Ġcoprime": 44743,
+      "uclides": 44744,
+      "Ġtrimming": 44745,
+      "MICAL": 44746,
+      "pyrrole": 44747,
+      "Ia": 44748,
+      "NLS": 44749,
+      "Quality": 44750,
+      "takes": 44751,
+      "zinc": 44752,
+      "ĠPione": 44753,
+      "ĠEwing": 44754,
+      "ĠLCA": 44755,
+      "ĠÃĶ": 44756,
+      "ictus": 44757,
+      "Ġcollim": 44758,
+      "Ġphylogenetically": 44759,
+      "ĠKeeping": 44760,
+      "ĠFaith": 44761,
+      "bonds": 44762,
+      "titer": 44763,
+      "Ġsubcategories": 44764,
+      "shaded": 44765,
+      "Ġphotospheric": 44766,
+      "ĠAppearance": 44767,
+      "ĠUniversities": 44768,
+      "Ġglomeruli": 44769,
+      "ĠPrefrontal": 44770,
+      "Ġprivilege": 44771,
+      "iH": 44772,
+      "uya": 44773,
+      "ĠLCL": 44774,
+      "ĠInGaAs": 44775,
+      "Inspired": 44776,
+      "atalog": 44777,
+      "ĠPerceptions": 44778,
+      "ĠNaHCO": 44779,
+      "Ġstreamline": 44780,
+      "trajectory": 44781,
+      "ĠMicrom": 44782,
+      "Ġbedside": 44783,
+      "ĠRomero": 44784,
+      "Ġgaugino": 44785,
+      "DEN": 44786,
+      "Fa": 44787,
+      "Olymp": 44788,
+      "eal": 44789,
+      "uels": 44790,
+      "icylic": 44791,
+      "Ġgod": 44792,
+      "Ġattaining": 44793,
+      "Ġprotests": 44794,
+      "Ġnowhere": 44795,
+      "desorption": 44796,
+      "ĠHydroxy": 44797,
+      "ĠErbB": 44798,
+      "ĠSPAR": 44799,
+      "Ġhinders": 44800,
+      "herenkov": 44801,
+      "KERNEL": 44802,
+      "Ġsect": 44803,
+      "ulong": 44804,
+      "Ġpreprocessed": 44805,
+      "fractional": 44806,
+      "oyage": 44807,
+      "Ġphosphatases": 44808,
+      "Ġcoastline": 44809,
+      "Ġhref": 44810,
+      "ĠSutherland": 44811,
+      "oxone": 44812,
+      "Ġhomomorphic": 44813,
+      "DEM": 44814,
+      "Ġbovis": 44815,
+      "ĠCBP": 44816,
+      "plen": 44817,
+      "ĠBuc": 44818,
+      "ĠGior": 44819,
+      "Ġcompost": 44820,
+      "ĠOracle": 44821,
+      "ĠSphere": 44822,
+      "ĠSchre": 44823,
+      "derivatives": 44824,
+      "lytes": 44825,
+      "ĠYo": 44826,
+      "Ġcyclones": 44827,
+      "ĠMaize": 44828,
+      "Ġunfair": 44829,
+      "Template": 44830,
+      "Ġimpregnation": 44831,
+      "Ġlaparoscopy": 44832,
+      "Ġhamiltonian": 44833,
+      "ignore": 44834,
+      "Ġdisposable": 44835,
+      "earic": 44836,
+      "Ġelectoral": 44837,
+      "ccos": 44838,
+      "ĠShh": 44839,
+      "Ġturbo": 44840,
+      "Ġintrusive": 44841,
+      "Ġprecedence": 44842,
+      "annotated": 44843,
+      "Ġdystonia": 44844,
+      "Fat": 44845,
+      "uins": 44846,
+      "Ġsway": 44847,
+      "arizing": 44848,
+      "illen": 44849,
+      "Ġyi": 44850,
+      "Ġnormed": 44851,
+      "ĠÌĤ": 44852,
+      "ĠExtr": 44853,
+      "ĠProteome": 44854,
+      "Document": 44855,
+      "ĠQUANTUM": 44856,
+      "titi": 44857,
+      "ĠCPC": 44858,
+      "ĠMiles": 44859,
+      "ĠBoc": 44860,
+      "ĠRTS": 44861,
+      "CTX": 44862,
+      "Ġsafegu": 44863,
+      "ĠNormally": 44864,
+      "ĠÃľber": 44865,
+      "onious": 44866,
+      "ĠSCE": 44867,
+      "Ġalfalfa": 44868,
+      "ĠLut": 44869,
+      "Ġcout": 44870,
+      "Ġenlarge": 44871,
+      "ĠEnable": 44872,
+      "Ġvirion": 44873,
+      "ĠShallow": 44874,
+      "definitely": 44875,
+      "ĠColin": 44876,
+      "ĠRetention": 44877,
+      "Ġmimicry": 44878,
+      "################################################################": 44879,
+      "NSCLC": 44880,
+      "Ġgratitude": 44881,
+      "Ġtending": 44882,
+      "ĠIDS": 44883,
+      "eret": 44884,
+      "rican": 44885,
+      "Ġxn": 44886,
+      "ĠYoo": 44887,
+      "Ġoptimise": 44888,
+      "Arrow": 44889,
+      "ĠTransferase": 44890,
+      "PKC": 44891,
+      "ĠGuangzhou": 44892,
+      "ruc": 44893,
+      "yrid": 44894,
+      "isz": 44895,
+      "ĠFIX": 44896,
+      "ĠDatabases": 44897,
+      "astron": 44898,
+      "Ġplayback": 44899,
+      "Ġnarrowly": 44900,
+      "Correlation": 44901,
+      "ĠAffinity": 44902,
+      "Ġfunctorial": 44903,
+      "Ġlectins": 44904,
+      "Ġruptured": 44905,
+      "Display": 44906,
+      "ĠSymptom": 44907,
+      "Ġequidistant": 44908,
+      "ĠRiccati": 44909,
+      "ĠAchievement": 44910,
+      "grand": 44911,
+      "onated": 44912,
+      "ĠdH": 44913,
+      "ĠFID": 44914,
+      "ĠDER": 44915,
+      "ĠCoA": 44916,
+      "Ġgasification": 44917,
+      "ĠCONS": 44918,
+      "Ġaccompanies": 44919,
+      "Ġimpede": 44920,
+      "Ġprecede": 44921,
+      "Ġkitchen": 44922,
+      "progress": 44923,
+      "Ġwiring": 44924,
+      "lerenes": 44925,
+      "ĠGius": 44926,
+      "Ġtransp": 44927,
+      "retrie": 44928,
+      "ijer": 44929,
+      "affer": 44930,
+      "Ġbirthday": 44931,
+      "ĠHald": 44932,
+      "Ġmusculus": 44933,
+      "ĠToken": 44934,
+      "ĠBowel": 44935,
+      "Ġskipped": 44936,
+      "Cha": 44937,
+      "bv": 44938,
+      "ĠBlow": 44939,
+      "Ġpreoperatively": 44940,
+      "Ġglove": 44941,
+      "ĠLeven": 44942,
+      "Ġmesop": 44943,
+      "ĠAuxiliary": 44944,
+      "ensuremath": 44945,
+      "jus": 44946,
+      "Å©": 44947,
+      "Ġvoter": 44948,
+      "ĠHitch": 44949,
+      "proxy": 44950,
+      "ĠKut": 44951,
+      "Ġpoems": 44952,
+      "ĠAngl": 44953,
+      "cera": 44954,
+      "Ġstarred": 44955,
+      "AGES": 44956,
+      "Science": 44957,
+      "Analyses": 44958,
+      "Ġreferees": 44959,
+      "Ġabrogated": 44960,
+      "Ġdesalination": 44961,
+      "ĠPrandtl": 44962,
+      "Pit": 44963,
+      "Ġnatal": 44964,
+      "ogran": 44965,
+      "ystitis": 44966,
+      "Ġdesm": 44967,
+      "Ġcurious": 44968,
+      "Ġdemon": 44969,
+      "uzzi": 44970,
+      "ochondrial": 44971,
+      "ĠTreaty": 44972,
+      "Tracker": 44973,
+      "rhoeae": 44974,
+      "LW": 44975,
+      "furt": 44976,
+      "Ġomp": 44977,
+      "isational": 44978,
+      "Ġmemorial": 44979,
+      "ĠLatency": 44980,
+      "ĠHypot": 44981,
+      "Ġglued": 44982,
+      "exactly": 44983,
+      "Ġcontraind": 44984,
+      "Cancer": 44985,
+      "Ġffi": 44986,
+      "ĠNAA": 44987,
+      "ĠChr": 44988,
+      "egg": 44989,
+      "ĠMotiv": 44990,
+      "Ġlayouts": 44991,
+      "Ġclimb": 44992,
+      "Ġappendicitis": 44993,
+      "CUDA": 44994,
+      "Ġphotoproduction": 44995,
+      "ĠSIP": 44996,
+      "Ġveto": 44997,
+      "perin": 44998,
+      "ĠUnity": 44999,
+      "byear": 45000,
+      "Ġforwarded": 45001,
+      "ĠDominant": 45002,
+      "holz": 45003,
+      "ĠThoracic": 45004,
+      "DEFINE": 45005,
+      "Ġtyrosinase": 45006,
+      "Bad": 45007,
+      "INA": 45008,
+      "fuel": 45009,
+      "Ġgi": 45010,
+      "ĠVIS": 45011,
+      "astolic": 45012,
+      "Ġoxaliplatin": 45013,
+      "effector": 45014,
+      "ĉĉĉĉĠ": 45015,
+      "еÑĢ": 45016,
+      "ĠBaby": 45017,
+      "Ġwashout": 45018,
+      "pituitary": 45019,
+      "NGC": 45020,
+      "Ġdns": 45021,
+      "ĠPoz": 45022,
+      "ĠUz": 45023,
+      "positron": 45024,
+      "ĠElectrons": 45025,
+      "Ġhemangi": 45026,
+      "ĠZnS": 45027,
+      "ĠTEMP": 45028,
+      "ĠExperimentally": 45029,
+      "fluorouracil": 45030,
+      "Ġlaparotomy": 45031,
+      "analyzer": 45032,
+      "ocorticoid": 45033,
+      "ĠIMPL": 45034,
+      "ĠDNNs": 45035,
+      "ĠFresnel": 45036,
+      "Mont": 45037,
+      "Ġtapes": 45038,
+      "ulomb": 45039,
+      "impedance": 45040,
+      "ĠHET": 45041,
+      "atha": 45042,
+      "modulation": 45043,
+      "ĠCortic": 45044,
+      "Ġâľĵ": 45045,
+      "ĠFairness": 45046,
+      "ĠStiff": 45047,
+      "Ġbuttons": 45048,
+      "css": 45049,
+      "Ġandroid": 45050,
+      "elast": 45051,
+      "ĠTeflon": 45052,
+      "ĠMBC": 45053,
+      "ĠJT": 45054,
+      "Ġmultilayered": 45055,
+      "ĠRee": 45056,
+      "uitar": 45057,
+      "ĠPhilips": 45058,
+      "ĠSkip": 45059,
+      "doctoral": 45060,
+      "iyama": 45061,
+      "ĠLeadership": 45062,
+      "ĠCrisis": 45063,
+      "Ġdesensitization": 45064,
+      "vous": 45065,
+      "ĠSPP": 45066,
+      "ĠPGA": 45067,
+      "ĠNever": 45068,
+      "Ġdefeating": 45069,
+      "Ġfibromyalgia": 45070,
+      "ĠMRP": 45071,
+      "ĠABCA": 45072,
+      "ĠLowe": 45073,
+      "Ġeroded": 45074,
+      "Ġaugments": 45075,
+      "ĠBoris": 45076,
+      "Ġnephrectomy": 45077,
+      "ĠSherman": 45078,
+      "Ġrefrigeration": 45079,
+      "ĠHernández": 45080,
+      "Ãĺ": 45081,
+      "ĠTors": 45082,
+      "chus": 45083,
+      "ĠVarg": 45084,
+      "Ġroset": 45085,
+      "CLR": 45086,
+      "DEP": 45087,
+      "Strong": 45088,
+      "Ġcinerea": 45089,
+      "ĠHeinrich": 45090,
+      "Rout": 45091,
+      "odus": 45092,
+      "ĠPhone": 45093,
+      "ĠPerl": 45094,
+      "Ġseasonally": 45095,
+      "holding": 45096,
+      "Ġencephalomyelitis": 45097,
+      "Ġfascia": 45098,
+      "Ġlittermates": 45099,
+      "ĠWITHOUT": 45100,
+      "б": 45101,
+      "Ġalerts": 45102,
+      "ĠKoll": 45103,
+      "ĠUrs": 45104,
+      "elfand": 45105,
+      "ĠRNAP": 45106,
+      "Ġinvariably": 45107,
+      "Ġscintigraphy": 45108,
+      "ĠSebastian": 45109,
+      "kinesia": 45110,
+      "CUR": 45111,
+      "inants": 45112,
+      "ĠpET": 45113,
+      "idial": 45114,
+      "ĠUPLC": 45115,
+      "Ġsuis": 45116,
+      "Ġbasolateral": 45117,
+      "ĠModulates": 45118,
+      "orbic": 45119,
+      "Img": 45120,
+      "Ġparasitism": 45121,
+      "Ġlaminate": 45122,
+      "ogeographic": 45123,
+      "ĠRibeiro": 45124,
+      "ĠGlutathione": 45125,
+      "ĠAberrant": 45126,
+      "Ġsclero": 45127,
+      "ĠDLS": 45128,
+      "ĠRuth": 45129,
+      "Ġrecast": 45130,
+      "recated": 45131,
+      "okie": 45132,
+      "ĠParks": 45133,
+      "Ġfoliations": 45134,
+      "ĠDawson": 45135,
+      "Ġtannins": 45136,
+      "ĠAaron": 45137,
+      "pS": 45138,
+      "itating": 45139,
+      "ĠITC": 45140,
+      "ipients": 45141,
+      "ohy": 45142,
+      "CCs": 45143,
+      "Ġethanolic": 45144,
+      "corhynchus": 45145,
+      "Ġorientational": 45146,
+      "Ġhabituation": 45147,
+      "Ġconversational": 45148,
+      "ĠVentricular": 45149,
+      "Ġintercalated": 45150,
+      "Ġphosphodiesterase": 45151,
+      "ĠSeifert": 45152,
+      "wk": 45153,
+      "algesia": 45154,
+      "Ġstegan": 45155,
+      "ĠLus": 45156,
+      "ophantine": 45157,
+      "Ġcorrects": 45158,
+      "ĠObama": 45159,
+      "latency": 45160,
+      "Ġsonar": 45161,
+      "ORMAL": 45162,
+      "Ġseaweed": 45163,
+      "ĠPowers": 45164,
+      "ĠShapley": 45165,
+      "Lore": 45166,
+      "Ġawa": 45167,
+      "alach": 45168,
+      "ĠFon": 45169,
+      "ensate": 45170,
+      "Ġoptima": 45171,
+      "INF": 45172,
+      "Ġpolygenic": 45173,
+      "Ġmesoderm": 45174,
+      "Conver": 45175,
+      "BRID": 45176,
+      "ĠHelp": 45177,
+      "ĠRasmussen": 45178,
+      "Ġprokaryotes": 45179,
+      "ĠEurasian": 45180,
+      "ĠPermeability": 45181,
+      "Ġnau": 45182,
+      "ĠClem": 45183,
+      "odilation": 45184,
+      "ĠDiaz": 45185,
+      "itious": 45186,
+      "ĠChad": 45187,
+      "ORA": 45188,
+      "ĠSimons": 45189,
+      "ĠDistances": 45190,
+      "Ġastrometric": 45191,
+      "ĠCPUs": 45192,
+      "Ġthioredoxin": 45193,
+      "perturbation": 45194,
+      "Ġdendrimer": 45195,
+      "algal": 45196,
+      "Ġceliac": 45197,
+      "asz": 45198,
+      "ĠPPE": 45199,
+      "qua": 45200,
+      "ĠBoll": 45201,
+      "chr": 45202,
+      "Ġpreview": 45203,
+      "ĠProjections": 45204,
+      "ĠAsians": 45205,
+      "ĠInferring": 45206,
+      "ĠNaive": 45207,
+      "ĠHiggins": 45208,
+      "ĠLocated": 45209,
+      "cardiac": 45210,
+      "ĠLarson": 45211,
+      "hazard": 45212,
+      "ĠScientists": 45213,
+      "Ġpinn": 45214,
+      "ENCY": 45215,
+      "forme": 45216,
+      "chitects": 45217,
+      "ofluorescent": 45218,
+      "ĠPortal": 45219,
+      "Ġpupae": 45220,
+      "interesting": 45221,
+      "įĢ": 45222,
+      "react": 45223,
+      "atos": 45224,
+      "enin": 45225,
+      "tio": 45226,
+      "ĠCapp": 45227,
+      "ĠMau": 45228,
+      "ĠLSC": 45229,
+      "ĠVlasov": 45230,
+      "Ġsubsum": 45231,
+      "Ġdeserve": 45232,
+      "ASD": 45233,
+      "Rece": 45234,
+      "Ġconsonant": 45235,
+      "Ġimpregnated": 45236,
+      "Ġlignocellulosic": 45237,
+      "Ġsows": 45238,
+      "lement": 45239,
+      "ĠTier": 45240,
+      "ĠMEF": 45241,
+      "ĠHugh": 45242,
+      "inck": 45243,
+      "pyrazole": 45244,
+      "ULATIONS": 45245,
+      "ĠALI": 45246,
+      "ĠDrift": 45247,
+      "Ġsolubilized": 45248,
+      "Ġdrafting": 45249,
+      "icyclic": 45250,
+      "Ġredesign": 45251,
+      "Ġdeliberate": 45252,
+      "Ġtapping": 45253,
+      "ĠTomas": 45254,
+      "ĠTunneling": 45255,
+      "ĠCBR": 45256,
+      "Ġanodes": 45257,
+      "ĠLSR": 45258,
+      "ĠNath": 45259,
+      "rosive": 45260,
+      "ĠHeidelberg": 45261,
+      "Ġcrushing": 45262,
+      "ĠShore": 45263,
+      "Ġmalondialdehyde": 45264,
+      "ĠMRD": 45265,
+      "ogloss": 45266,
+      "ncia": 45267,
+      "Ġgranuloma": 45268,
+      "Ġplaintext": 45269,
+      "Ġarteriovenous": 45270,
+      "Ġrifampicin": 45271,
+      "Lepidoptera": 45272,
+      "Oct": 45273,
+      "Ġlone": 45274,
+      "ĠAppe": 45275,
+      "ĠIntermitt": 45276,
+      "compile": 45277,
+      "potentials": 45278,
+      "ĠStandardized": 45279,
+      "Ġventilatory": 45280,
+      "Ġhypercholesterolemia": 45281,
+      "ĠEVALUATION": 45282,
+      "ked": 45283,
+      "xC": 45284,
+      "enos": 45285,
+      "Ġbauthorbsnm": 45286,
+      "ĠRost": 45287,
+      "mathopen": 45288,
+      "Ġcontested": 45289,
+      "Ġros": 45290,
+      "otho": 45291,
+      "Ġemits": 45292,
+      "erozo": 45293,
+      "Ġpropranolol": 45294,
+      "Ġexacerbate": 45295,
+      "Integrating": 45296,
+      "ĠWarsaw": 45297,
+      "Ñĩ": 45298,
+      "refractory": 45299,
+      "ĠMort": 45300,
+      "phosphonate": 45301,
+      "GLT": 45302,
+      "ĠChloride": 45303,
+      "ĠLUAD": 45304,
+      "ĠSQUID": 45305,
+      "ĠOBSERVATIONS": 45306,
+      "Ħĺ": 45307,
+      "agles": 45308,
+      "uger": 45309,
+      "Ġdiffusing": 45310,
+      "ylar": 45311,
+      "Ġantip": 45312,
+      "renormal": 45313,
+      "Ġsheared": 45314,
+      "ĠAndr": 45315,
+      "ymptotics": 45316,
+      "ĠIdentified": 45317,
+      "Ġflexor": 45318,
+      "Liouville": 45319,
+      "ĠCytotoxic": 45320,
+      "Lock": 45321,
+      "donald": 45322,
+      "ĠSHA": 45323,
+      "projected": 45324,
+      "plicial": 45325,
+      "Ġbasics": 45326,
+      "ĠCarvalho": 45327,
+      "Ġheterocyclic": 45328,
+      "Ġfluorophore": 45329,
+      "ĠIntrigu": 45330,
+      "ĠAnnealing": 45331,
+      "Gln": 45332,
+      "Hispanic": 45333,
+      "Ġsaus": 45334,
+      "ĠTCS": 45335,
+      "ĠHAP": 45336,
+      "Ġytt": 45337,
+      "Ġconsulting": 45338,
+      "rects": 45339,
+      "Ġinfall": 45340,
+      "LEV": 45341,
+      "triazole": 45342,
+      "Ġnarrowed": 45343,
+      "Ġamphoteric": 45344,
+      "ĠSorting": 45345,
+      "ĠMoments": 45346,
+      "Ġarabin": 45347,
+      "Ġcoconut": 45348,
+      "ĠIntriguingly": 45349,
+      "Ġpushes": 45350,
+      "Ġmec": 45351,
+      "ĠNair": 45352,
+      "Ġcolistin": 45353,
+      "ĠObtained": 45354,
+      "dfs": 45355,
+      "Ġcompetency": 45356,
+      "WORD": 45357,
+      "ĠAAS": 45358,
+      "ĠBNP": 45359,
+      "ĠHAS": 45360,
+      "ĠLun": 45361,
+      "ĠLnc": 45362,
+      "Ġhydrocephalus": 45363,
+      "Ġhomological": 45364,
+      "Ġcarbonic": 45365,
+      "ĠHiSeq": 45366,
+      "community": 45367,
+      "Ġcephalospor": 45368,
+      "Ġhostile": 45369,
+      "provide": 45370,
+      "Ġskyrmion": 45371,
+      "DAG": 45372,
+      "Ġcnt": 45373,
+      "Ġhay": 45374,
+      "Ġorderings": 45375,
+      "Ġflock": 45376,
+      "HEA": 45377,
+      "ĠNeurom": 45378,
+      "Ġboosts": 45379,
+      "ĠCardinal": 45380,
+      "ĠBachelor": 45381,
+      "Ġdecent": 45382,
+      "ĠYak": 45383,
+      "Ġcalcd": 45384,
+      "ĠBoer": 45385,
+      "Ġtranscriptomics": 45386,
+      "Ġrearranged": 45387,
+      "ĠPolymorphisms": 45388,
+      "ĠPrasad": 45389,
+      "oinositide": 45390,
+      "bars": 45391,
+      "Ġãģ": 45392,
+      "ĠSAA": 45393,
+      "Ġonion": 45394,
+      "agel": 45395,
+      "ĠHp": 45396,
+      "ogrel": 45397,
+      "divisions": 45398,
+      "andan": 45399,
+      "arias": 45400,
+      "Ġcolo": 45401,
+      "ragon": 45402,
+      "Ġschizophren": 45403,
+      "âī¡": 45404,
+      "Ġreplicative": 45405,
+      "Ġdegenerated": 45406,
+      "Ġsteepest": 45407,
+      "Volume": 45408,
+      "IENT": 45409,
+      "Public": 45410,
+      "Ten": 45411,
+      "enberger": 45412,
+      "ĠCoun": 45413,
+      "ĠEpp": 45414,
+      "izo": 45415,
+      "Ġcomplexed": 45416,
+      "Ġferroc": 45417,
+      "kenstein": 45418,
+      "ĠJerry": 45419,
+      "Ġparadoxical": 45420,
+      "xg": 45421,
+      "icer": 45422,
+      "osol": 45423,
+      "Ġannu": 45424,
+      "Ġankyl": 45425,
+      "chung": 45426,
+      "entious": 45427,
+      "Ġpreshe": 45428,
+      "enetic": 45429,
+      "ĠHealing": 45430,
+      "ĠParabolic": 45431,
+      "Ġfigs": 45432,
+      "ĠKinematic": 45433,
+      "Ġobligate": 45434,
+      "ĠLayout": 45435,
+      "Ġtelemedicine": 45436,
+      "ĠLennard": 45437,
+      "pci": 45438,
+      "arone": 45439,
+      "ĠZach": 45440,
+      "Ġprototyping": 45441,
+      "ĠMetagen": 45442,
+      "IMAL": 45443,
+      "conscious": 45444,
+      "Ġquadrilateral": 45445,
+      "ĠUncertainties": 45446,
+      "ĠPrefecture": 45447,
+      "GBM": 45448,
+      "rals": 45449,
+      "alus": 45450,
+      "Ġhopes": 45451,
+      "Ġclicks": 45452,
+      "ĠJD": 45453,
+      "lectance": 45454,
+      "Ġpathologists": 45455,
+      "ussels": 45456,
+      "tisone": 45457,
+      "CPT": 45458,
+      "Ġmiscon": 45459,
+      "ĠNeurode": 45460,
+      "Ġmutagenic": 45461,
+      "ĠMultimedia": 45462,
+      "Original": 45463,
+      "ĠDrake": 45464,
+      "PWM": 45465,
+      "Ġpiles": 45466,
+      "stant": 45467,
+      "ARA": 45468,
+      "ĠRING": 45469,
+      "modifying": 45470,
+      "Ġastrocyt": 45471,
+      "ĠCyst": 45472,
+      "Ġlegends": 45473,
+      "glucuron": 45474,
+      "Ġincompletely": 45475,
+      "ĠConfed": 45476,
+      "ĠDLBCL": 45477,
+      "ĠPapua": 45478,
+      "Ġcontrastive": 45479,
+      "ĠSIMULATION": 45480,
+      "ĠJuvenile": 45481,
+      "aggregated": 45482,
+      "ĠcGMP": 45483,
+      "ictive": 45484,
+      "ĠHNF": 45485,
+      "ĠNPV": 45486,
+      "ĠKoc": 45487,
+      "ometallic": 45488,
+      "mini": 45489,
+      "ĠQuantit": 45490,
+      "ĠCornell": 45491,
+      "Ġdeduction": 45492,
+      "Ġcoinciding": 45493,
+      "ĠIrr": 45494,
+      "Precision": 45495,
+      "Ġginseng": 45496,
+      "ões": 45497,
+      "jer": 45498,
+      "ĠReader": 45499,
+      "ĠByr": 45500,
+      "corrections": 45501,
+      "devices": 45502,
+      "Ġambul": 45503,
+      "Ġpedicle": 45504,
+      "ĠDependency": 45505,
+      "ĠStriking": 45506,
+      "Ġwarehouse": 45507,
+      "Ġrecirculation": 45508,
+      "Ġgonorrhoeae": 45509,
+      "ĠPRES": 45510,
+      "ĠBhar": 45511,
+      "Ġflushing": 45512,
+      "torus": 45513,
+      "ĠIRB": 45514,
+      "glycine": 45515,
+      "Ġmethamphetamine": 45516,
+      "Ġmirrored": 45517,
+      "ĠWilliamson": 45518,
+      "Ġcathodes": 45519,
+      "hydroxylase": 45520,
+      "Radio": 45521,
+      "Ġfurniture": 45522,
+      "ĠRosenberg": 45523,
+      "ĠNSAIDs": 45524,
+      "semiconductor": 45525,
+      "Ġasynchron": 45526,
+      "ĠBerm": 45527,
+      "ĠInten": 45528,
+      "ibe": 45529,
+      "Force": 45530,
+      "pathogenic": 45531,
+      "smokers": 45532,
+      "Ġdiphenyl": 45533,
+      "Ġи": 45534,
+      "Ġstandalone": 45535,
+      "Ġlithospheric": 45536,
+      "Ġtradeoffs": 45537,
+      "Ġantich": 45538,
+      "Ġthymidine": 45539,
+      "ĠMedicinal": 45540,
+      "Ġentrepreneurial": 45541,
+      "Ġtrapezoidal": 45542,
+      "ĠAsynchronous": 45543,
+      "tifying": 45544,
+      "ĠCollapse": 45545,
+      "ĠHEV": 45546,
+      "ĠFrozen": 45547,
+      "ĠTeichmüller": 45548,
+      "rocnemius": 45549,
+      "Ġfern": 45550,
+      "Ġws": 45551,
+      "omol": 45552,
+      "Ġenclosing": 45553,
+      "rapid": 45554,
+      "Ġlogged": 45555,
+      "varvec": 45556,
+      "Ġamplifying": 45557,
+      "differences": 45558,
+      "otonin": 45559,
+      "ĠPromoting": 45560,
+      "ĠFritz": 45561,
+      "Ġattainable": 45562,
+      "Ġaltim": 45563,
+      "ĠOGD": 45564,
+      "Ġthermometer": 45565,
+      "Solver": 45566,
+      "ĠBirk": 45567,
+      "LENBQU": 45568,
+      "ĠGateway": 45569,
+      "Ġengraftment": 45570,
+      "FIF": 45571,
+      "HSD": 45572,
+      "Ġrestructuring": 45573,
+      "ĠTensile": 45574,
+      "ĠCele": 45575,
+      "ylus": 45576,
+      "Ġfeather": 45577,
+      "Ġdrifting": 45578,
+      "ĠPreclinical": 45579,
+      "yrrole": 45580,
+      "Ġcommem": 45581,
+      "Ġfixations": 45582,
+      "Petsc": 45583,
+      "ĠIschemia": 45584,
+      "aA": 45585,
+      "asoro": 45586,
+      "ĠSony": 45587,
+      "ĠUt": 45588,
+      "Ġextensor": 45589,
+      "ĠChau": 45590,
+      "ĠIsotopic": 45591,
+      "ILI": 45592,
+      "CNP": 45593,
+      "ĠDEF": 45594,
+      "Ġmountainous": 45595,
+      "Ġsarcomas": 45596,
+      "ugoslav": 45597,
+      "CALL": 45598,
+      "Sensitive": 45599,
+      "atro": 45600,
+      "Ġuncoupling": 45601,
+      "skew": 45602,
+      "ĠEmissions": 45603,
+      "innati": 45604,
+      "Ġconceptualization": 45605,
+      "Ġowns": 45606,
+      "Ġsquadron": 45607,
+      "ĠStrengths": 45608,
+      "Coh": 45609,
+      "UAL": 45610,
+      "magenta": 45611,
+      "usb": 45612,
+      "ĠSPC": 45613,
+      "cones": 45614,
+      "ĠSelecting": 45615,
+      "ĠParish": 45616,
+      "Ġvalidates": 45617,
+      "ĠÍĹ": 45618,
+      "Ġposteriorly": 45619,
+      "omonad": 45620,
+      "VOL": 45621,
+      "jectivity": 45622,
+      "ĠCLO": 45623,
+      "ĠVTA": 45624,
+      "Ġunpleasant": 45625,
+      "Ġcareers": 45626,
+      "Ġautomorphic": 45627,
+      "ĠNanow": 45628,
+      "Ġasterisks": 45629,
+      "ĠSchulz": 45630,
+      "publication": 45631,
+      "Ġbiv": 45632,
+      "Ġrug": 45633,
+      "recognition": 45634,
+      "Ġreferrals": 45635,
+      "Ġneurones": 45636,
+      "ĠCaffe": 45637,
+      "Connor": 45638,
+      "ĠSheffield": 45639,
+      "unitinib": 45640,
+      "ĠAntagon": 45641,
+      "Ġpneumatic": 45642,
+      "Ġcleaner": 45643,
+      "ĠBAO": 45644,
+      "ĠScilabString": 45645,
+      "neighbour": 45646,
+      "Euler": 45647,
+      "ĠTuple": 45648,
+      "oty": 45649,
+      "dian": 45650,
+      "Ġyoga": 45651,
+      "Ġevanes": 45652,
+      "Ġstarved": 45653,
+      "Ġfluctuate": 45654,
+      "ĠBiomarker": 45655,
+      "Ġimpulses": 45656,
+      "Ġossification": 45657,
+      "Ġdemyelination": 45658,
+      "ĠSAD": 45659,
+      "essing": 45660,
+      "Ġreddish": 45661,
+      "Ġsynth": 45662,
+      "Ġcurvilinear": 45663,
+      "ĠDenis": 45664,
+      "Ġphonetic": 45665,
+      "Ġhammer": 45666,
+      "Ġepidermidis": 45667,
+      "Ġplagioclase": 45668,
+      "Ġĉ": 45669,
+      "Ġwolf": 45670,
+      "osced": 45671,
+      "Ġphotothermal": 45672,
+      "Ġchewing": 45673,
+      "Maximum": 45674,
+      "Ġmismatched": 45675,
+      "ĠFcγ": 45676,
+      "Ġumbrella": 45677,
+      "ĠSiberian": 45678,
+      "arra": 45679,
+      "ipped": 45680,
+      "ympathetic": 45681,
+      "acceleration": 45682,
+      "Ġeigenmodes": 45683,
+      "ĠEquivalently": 45684,
+      "ĠPRISMA": 45685,
+      "conservative": 45686,
+      "ñez": 45687,
+      "Ġvolcanoes": 45688,
+      "Ġtelemetry": 45689,
+      "mile": 45690,
+      "ĠBoch": 45691,
+      "oprim": 45692,
+      "Ġincipient": 45693,
+      "Ġunderstandable": 45694,
+      "atricyclo": 45695,
+      "ĠLogical": 45696,
+      "ĠQueue": 45697,
+      "Ġcryostat": 45698,
+      "definecolor": 45699,
+      "ĠSae": 45700,
+      "Ġarct": 45701,
+      "Ġsoul": 45702,
+      "ĠHistopathological": 45703,
+      "ĠNeurot": 45704,
+      "Ġmethanolic": 45705,
+      "Px": 45706,
+      "ĠTitle": 45707,
+      "otomic": 45708,
+      "ĠEld": 45709,
+      "ĠEMA": 45710,
+      "Ġdebrid": 45711,
+      "timulatory": 45712,
+      "ĠZan": 45713,
+      "Ġnormot": 45714,
+      "Ġfluidity": 45715,
+      "Ġfluidized": 45716,
+      "previously": 45717,
+      "Ġcracked": 45718,
+      "ĠExplaining": 45719,
+      "ĠONE": 45720,
+      "ĠFlora": 45721,
+      "ĠHybridization": 45722,
+      "Ġreticul": 45723,
+      "FK": 45724,
+      "notic": 45725,
+      "ĠnA": 45726,
+      "ĠPab": 45727,
+      "ticum": 45728,
+      "andy": 45729,
+      "ugia": 45730,
+      "ilet": 45731,
+      "MING": 45732,
+      "Ġrests": 45733,
+      "ompact": 45734,
+      "Ġtrackers": 45735,
+      "phosphatase": 45736,
+      "ĠTransfection": 45737,
+      "ĠHospitals": 45738,
+      "acrine": 45739,
+      "ĠDell": 45740,
+      "ĠVAE": 45741,
+      "ĠThroughput": 45742,
+      "hevsky": 45743,
+      "ĠSommer": 45744,
+      "PSA": 45745,
+      "ìļ": 45746,
+      "Ġbush": 45747,
+      "Ġlunch": 45748,
+      "ĠSwe": 45749,
+      "ĠInstruction": 45750,
+      "akami": 45751,
+      "Ġdisinfect": 45752,
+      "Ġcorps": 45753,
+      "ĉĉĠĠ": 45754,
+      "Ġprompts": 45755,
+      "MSH": 45756,
+      "ĠAgrawal": 45757,
+      "Ġlysosome": 45758,
+      "integrin": 45759,
+      "ĠỸ": 45760,
+      "Ġnondecreasing": 45761,
+      "ĠRequest": 45762,
+      "ĠREP": 45763,
+      "occus": 45764,
+      "Ġlagrangian": 45765,
+      "oregulation": 45766,
+      "ол": 45767,
+      "ĠBoson": 45768,
+      "Iso": 45769,
+      "atellites": 45770,
+      "resectable": 45771,
+      "riv": 45772,
+      "Ġdeaminase": 45773,
+      "Ġcoheren": 45774,
+      "Ġdecoy": 45775,
+      "ĠExtinction": 45776,
+      "acetone": 45777,
+      "Ġgovernmental": 45778,
+      "Ġcumulants": 45779,
+      "Ġviscosities": 45780,
+      "Register": 45781,
+      "documented": 45782,
+      "Ġimmortalized": 45783,
+      "DPP": 45784,
+      "Gel": 45785,
+      "bron": 45786,
+      "kow": 45787,
+      "ĠProportion": 45788,
+      "ĠChase": 45789,
+      "ĠClad": 45790,
+      "Ġadapts": 45791,
+      "ĠCAV": 45792,
+      "Ġż": 45793,
+      "Ġpelleted": 45794,
+      "Ġpenguin": 45795,
+      "ĠZhejiang": 45796,
+      "feasible": 45797,
+      "DIV": 45798,
+      "iya": 45799,
+      "Ġthrowing": 45800,
+      "resia": 45801,
+      "ĠNr": 45802,
+      "ESP": 45803,
+      "CDF": 45804,
+      "suppressed": 45805,
+      "Ġtetrachlor": 45806,
+      "Ġaerospace": 45807,
+      "Until": 45808,
+      "Ġpayoffs": 45809,
+      "Ġtownship": 45810,
+      "Ġesterification": 45811,
+      "ĠAchilles": 45812,
+      "Ġracem": 45813,
+      "opyranoside": 45814,
+      "ĠCSM": 45815,
+      "assis": 45816,
+      "Ġsupercell": 45817,
+      "ĠRegime": 45818,
+      "IRA": 45819,
+      "Ġsubsequences": 45820,
+      "ĠPenet": 45821,
+      "ĠAnalytics": 45822,
+      "ĠLVEF": 45823,
+      "Ġbiphenyl": 45824,
+      "Gradient": 45825,
+      "osylation": 45826,
+      "ĠWRF": 45827,
+      "ofs": 45828,
+      "conductors": 45829,
+      "Ġbacked": 45830,
+      "pidal": 45831,
+      "ĠNFAT": 45832,
+      "ĠRemember": 45833,
+      "Ġtelomeric": 45834,
+      "Ġtaurine": 45835,
+      "increases": 45836,
+      "Ġunintended": 45837,
+      "ĠNervous": 45838,
+      "Ras": 45839,
+      "ylyl": 45840,
+      "Ġaestiv": 45841,
+      "ĠSick": 45842,
+      "ĠTheta": 45843,
+      "Ġcliques": 45844,
+      "Ġsofter": 45845,
+      "ĠQRS": 45846,
+      "lliptic": 45847,
+      "ĠImmunotherapy": 45848,
+      "QUF": 45849,
+      "onomously": 45850,
+      "ĠFLU": 45851,
+      "ĠIncorporation": 45852,
+      "ĠFormicidae": 45853,
+      "JR": 45854,
+      "whole": 45855,
+      "Ġcasing": 45856,
+      "Ġnob": 45857,
+      "ĠDou": 45858,
+      "Ġintronic": 45859,
+      "Ġentrapment": 45860,
+      "orbits": 45861,
+      "Ġsalam": 45862,
+      "ĠCRS": 45863,
+      "ĠSwan": 45864,
+      "ĠEdgar": 45865,
+      "Ġconcomitantly": 45866,
+      "atetracyclo": 45867,
+      "ĠAHR": 45868,
+      "ticks": 45869,
+      "ĠBing": 45870,
+      "ĠRift": 45871,
+      "Ġplugging": 45872,
+      "ĠscRNA": 45873,
+      "Ġoutreach": 45874,
+      "inskii": 45875,
+      "Ġcustomary": 45876,
+      "Ġmd": 45877,
+      "ĠOzone": 45878,
+      "ussing": 45879,
+      "others": 45880,
+      "Ġentirety": 45881,
+      "Arth": 45882,
+      "Acet": 45883,
+      "ĠFleet": 45884,
+      "ĠBehavioural": 45885,
+      "ĠQSOs": 45886,
+      "arina": 45887,
+      "Ġprodrug": 45888,
+      "ĠBros": 45889,
+      "ĠWorth": 45890,
+      "Ġyz": 45891,
+      "contig": 45892,
+      "ĠAmorphous": 45893,
+      "ĠErlang": 45894,
+      "Ġhonour": 45895,
+      "ĠâIJ¥": 45896,
+      "Ġinfiltrates": 45897,
+      "ĠIvanov": 45898,
+      "ĠMunicipality": 45899,
+      "ĠDialogue": 45900,
+      "tone": 45901,
+      "Ġpytest": 45902,
+      "iculus": 45903,
+      "ĠGoth": 45904,
+      "ĠXC": 45905,
+      "ĠSUMMARY": 45906,
+      "Ġshrinks": 45907,
+      "Ġinverses": 45908,
+      "iomas": 45909,
+      "robi": 45910,
+      "ĠTPR": 45911,
+      "ĠANA": 45912,
+      "istries": 45913,
+      "Ġregiment": 45914,
+      "indo": 45915,
+      "ĠReproduction": 45916,
+      "loqu": 45917,
+      "inflation": 45918,
+      "ETX": 45919,
+      "Ġïĺ»": 45920,
+      "ĠAPPENDIX": 45921,
+      "Ġworsened": 45922,
+      "Ġpsoriatic": 45923,
+      "Ġmidwives": 45924,
+      "Ġtouched": 45925,
+      "Ëĩ": 45926,
+      "ĠPatric": 45927,
+      "ĠDON": 45928,
+      "ĠLIM": 45929,
+      "akos": 45930,
+      "ĠVie": 45931,
+      "ĠAntit": 45932,
+      "Ġflake": 45933,
+      "ĠSchle": 45934,
+      "ĠCoronal": 45935,
+      "Ġsalary": 45936,
+      "slight": 45937,
+      "ĠCAF": 45938,
+      "Ġsummarise": 45939,
+      "Ġflavus": 45940,
+      "ĠBalanced": 45941,
+      "ĠPHOT": 45942,
+      "Ġmillet": 45943,
+      "Ġurgency": 45944,
+      "ĠGleason": 45945,
+      "ĠMie": 45946,
+      "ĠDp": 45947,
+      "ĠGarg": 45948,
+      "Ġleprosy": 45949,
+      "Ġunoccupied": 45950,
+      "ĠStret": 45951,
+      "ilept": 45952,
+      "ĠChor": 45953,
+      "ibrate": 45954,
+      "ĠÍļ": 45955,
+      "ĠPHB": 45956,
+      "Ġmonoter": 45957,
+      "ĠJavaScript": 45958,
+      "btn": 45959,
+      "ĠPulsar": 45960,
+      "ĠKirchhoff": 45961,
+      "Ġoverseas": 45962,
+      "Ġdephosphorylation": 45963,
+      "ortin": 45964,
+      "ĠPolyakov": 45965,
+      "Ġinsightful": 45966,
+      "ĠPurified": 45967,
+      "Ġanchorage": 45968,
+      "ĠGlycoprotein": 45969,
+      "studies": 45970,
+      "Ġchronology": 45971,
+      "roxine": 45972,
+      "ĠNeptune": 45973,
+      "Ban": 45974,
+      "Ġlion": 45975,
+      "PSD": 45976,
+      "ĠBarr": 45977,
+      "Ġdonkey": 45978,
+      "Ġlikelihoods": 45979,
+      "atchewan": 45980,
+      "otet": 45981,
+      "ospha": 45982,
+      "ticism": 45983,
+      "Ġry": 45984,
+      "asthen": 45985,
+      "rhotic": 45986,
+      "ĠSubgroup": 45987,
+      "yev": 45988,
+      "ĠPatri": 45989,
+      "provides": 45990,
+      "SGD": 45991,
+      "berell": 45992,
+      "vw": 45993,
+      "ĠAACR": 45994,
+      "Ġsmears": 45995,
+      "ODS": 45996,
+      "supplemented": 45997,
+      "ĠEngagement": 45998,
+      "oglobulins": 45999,
+      "Ġirregularly": 46000,
+      "ĠSzeg": 46001,
+      "ĠWolff": 46002,
+      "Ġenantiomers": 46003,
+      "Ġobeying": 46004,
+      "Ġdestroying": 46005,
+      "omially": 46006,
+      "ĠAti": 46007,
+      "ĠGAT": 46008,
+      "ĠInvariants": 46009,
+      "ĠScoring": 46010,
+      "Ġhalides": 46011,
+      "Ġtransformants": 46012,
+      "Ġforested": 46013,
+      "Ġgallic": 46014,
+      "ĠBetti": 46015,
+      "threaded": 46016,
+      "ĠBudget": 46017,
+      "junctive": 46018,
+      "ĠInnovative": 46019,
+      "Ġpositrons": 46020,
+      "Brazil": 46021,
+      "eira": 46022,
+      "Ġlavas": 46023,
+      "ĠLt": 46024,
+      "photo": 46025,
+      "Ġspam": 46026,
+      "Ġih": 46027,
+      "ustering": 46028,
+      "Ġbioluminescence": 46029,
+      "ĠShapes": 46030,
+      "ULTI": 46031,
+      "triangles": 46032,
+      "ĠSMN": 46033,
+      "enhancing": 46034,
+      "ĠReduces": 46035,
+      "ĠTHEOREM": 46036,
+      "Dop": 46037,
+      "ĠdL": 46038,
+      "emptive": 46039,
+      "Ġreminder": 46040,
+      "Ġgonads": 46041,
+      "Ġxylan": 46042,
+      "cultures": 46043,
+      "tles": 46044,
+      "Ġtd": 46045,
+      "Ġerected": 46046,
+      "terone": 46047,
+      "ĠPDC": 46048,
+      "Ġincongruent": 46049,
+      "Ġmembranous": 46050,
+      "pac": 46051,
+      "yless": 46052,
+      "Ġsubalgebras": 46053,
+      "ĠChir": 46054,
+      "ĠZIP": 46055,
+      "autious": 46056,
+      "Ġlightly": 46057,
+      "ĠPhotometric": 46058,
+      "Transfer": 46059,
+      "Ġketo": 46060,
+      "Ġexercised": 46061,
+      "dispersive": 46062,
+      "ĠBETWEEN": 46063,
+      "rou": 46064,
+      "Ġgarbage": 46065,
+      "ĠMaf": 46066,
+      "ĠDoming": 46067,
+      "ĠSubspace": 46068,
+      "ĠMarÃŃa": 46069,
+      "Ġtetrahedra": 46070,
+      "ĠBarker": 46071,
+      "Side": 46072,
+      "bishop": 46073,
+      "iD": 46074,
+      "reversible": 46075,
+      "orman": 46076,
+      "orescein": 46077,
+      "ĠContrib": 46078,
+      "Ġderivatization": 46079,
+      "romeres": 46080,
+      "ĠALD": 46081,
+      "EEK": 46082,
+      "ĠTreating": 46083,
+      "combination": 46084,
+      "ïĺ»": 46085,
+      "restriction": 46086,
+      "supseteq": 46087,
+      "ĠRAPD": 46088,
+      "Ġamendment": 46089,
+      "zynski": 46090,
+      "Ġcaves": 46091,
+      "ilot": 46092,
+      "Ġabundantly": 46093,
+      "на": 46094,
+      "Ġinjectable": 46095,
+      "ĠReinforced": 46096,
+      "ĠWidth": 46097,
+      "ĠHaemophilus": 46098,
+      "ilane": 46099,
+      "props": 46100,
+      "Ġintervertebral": 46101,
+      "Ġscroll": 46102,
+      "Ġamput": 46103,
+      "ĠUnusual": 46104,
+      "Ġstatically": 46105,
+      "Ġsynergies": 46106,
+      "Ġdims": 46107,
+      "plasmic": 46108,
+      "Ġneutralized": 46109,
+      "Selected": 46110,
+      "Ġinherits": 46111,
+      "ĠAutomation": 46112,
+      "Ġprotoplanetary": 46113,
+      "Statement": 46114,
+      "ĠAPOBEC": 46115,
+      "Ġcertificates": 46116,
+      "ĠCitrus": 46117,
+      "quadruplex": 46118,
+      "Nord": 46119,
+      "Ġfran": 46120,
+      "ĠCarcin": 46121,
+      "utan": 46122,
+      "ĠPump": 46123,
+      "ĠBav": 46124,
+      "ĠGras": 46125,
+      "tingales": 46126,
+      "Ġcausally": 46127,
+      "Ġradon": 46128,
+      "Compare": 46129,
+      "Ġclamping": 46130,
+      "irreducible": 46131,
+      "IHC": 46132,
+      "ĠÙ": 46133,
+      "Ġcyp": 46134,
+      "ĠTPP": 46135,
+      "ĠSuff": 46136,
+      "undra": 46137,
+      "ĠVilla": 46138,
+      "Ġrelieved": 46139,
+      "ĠJCM": 46140,
+      "Ġtreaty": 46141,
+      "IGEN": 46142,
+      "ĠDevonian": 46143,
+      "Ġerythropo": 46144,
+      "RAP": 46145,
+      "Ġaversive": 46146,
+      "entate": 46147,
+      "odactyl": 46148,
+      "ĠParal": 46149,
+      "Ġmilled": 46150,
+      "Ġbioinformatic": 46151,
+      "okinetic": 46152,
+      "ĠSTRING": 46153,
+      "ĠPedersen": 46154,
+      "database": 46155,
+      "inorganic": 46156,
+      "Ġdeput": 46157,
+      "Ġneb": 46158,
+      "iped": 46159,
+      "Ġdiffused": 46160,
+      "othione": 46161,
+      "Ġnonstationary": 46162,
+      "Ġundertaking": 46163,
+      "ĠEnabling": 46164,
+      "Ġdenatured": 46165,
+      "Ġloader": 46166,
+      "ĠLyon": 46167,
+      "iparametric": 46168,
+      "Ġmeristem": 46169,
+      "ĠAngiogenesis": 46170,
+      "ĠPulsed": 46171,
+      "Ġexcer": 46172,
+      "ĠDf": 46173,
+      "arches": 46174,
+      "Ġcollide": 46175,
+      "ĠRelational": 46176,
+      "ĠNFκB": 46177,
+      "Metadata": 46178,
+      "ĠAddressing": 46179,
+      "Ġpercussion": 46180,
+      "ĠFlorence": 46181,
+      "Ġnymphs": 46182,
+      "Cn": 46183,
+      "storm": 46184,
+      "ĠGraz": 46185,
+      "composite": 46186,
+      "ĠAdmiral": 46187,
+      "ĠScotia": 46188,
+      "Ġbremsstrahlung": 46189,
+      "apsack": 46190,
+      "Ġminimizers": 46191,
+      "Ġmanageable": 46192,
+      "Ġcarboxylate": 46193,
+      "Ġintermediary": 46194,
+      "ĠBranching": 46195,
+      "scheduler": 46196,
+      "inoculated": 46197,
+      "ĠExtremely": 46198,
+      "Ġantennae": 46199,
+      "ĠTill": 46200,
+      "RESH": 46201,
+      "Ġopacities": 46202,
+      "Ġchemopre": 46203,
+      "Ġadenylate": 46204,
+      "Ġcircumstance": 46205,
+      "ĠHashimoto": 46206,
+      "ÄĽ": 46207,
+      "ceae": 46208,
+      "ĠFm": 46209,
+      "ĠBX": 46210,
+      "Ġmeantime": 46211,
+      "accurate": 46212,
+      "collinear": 46213,
+      "ACTIC": 46214,
+      "ĠSlovenia": 46215,
+      "Fed": 46216,
+      "Kh": 46217,
+      "Tm": 46218,
+      "fork": 46219,
+      "inology": 46220,
+      "lef": 46221,
+      "ĠDCS": 46222,
+      "Ġheritable": 46223,
+      "Ġannouncement": 46224,
+      "Ġbusinessman": 46225,
+      "Ġbortezomib": 46226,
+      "Ġtourist": 46227,
+      "ĠEtymology": 46228,
+      "Ġdoctrine": 46229,
+      "BIN": 46230,
+      "suffix": 46231,
+      "aras": 46232,
+      "ĠSau": 46233,
+      "unboldmath": 46234,
+      "ĠMEP": 46235,
+      "inker": 46236,
+      "Ġoptimism": 46237,
+      "ĠLeuc": 46238,
+      "efulness": 46239,
+      "crust": 46240,
+      "ĠKeys": 46241,
+      "ĠâĻ¦": 46242,
+      "ĠBrandt": 46243,
+      "âĮ¬": 46244,
+      "ĠSeventy": 46245,
+      "Ġnursery": 46246,
+      "Ġdeputy": 46247,
+      "ì": 46248,
+      "onis": 46249,
+      "amus": 46250,
+      "ĠCig": 46251,
+      "Ġexergy": 46252,
+      "ĠFrequent": 46253,
+      "Ġabor": 46254,
+      "ĠJazz": 46255,
+      "Ġstatue": 46256,
+      "ĠScenarios": 46257,
+      "Ġcytological": 46258,
+      "figures": 46259,
+      "MCI": 46260,
+      "dirname": 46261,
+      "Ġcytokinesis": 46262,
+      "delivery": 46263,
+      "ĠBowen": 46264,
+      "Ġflanked": 46265,
+      "Ġregenerating": 46266,
+      "ĠFerrari": 46267,
+      "kiss": 46268,
+      "ĠAval": 46269,
+      "ĠCIT": 46270,
+      "ĠMum": 46271,
+      "ĠLSB": 46272,
+      "ogging": 46273,
+      "Ġunited": 46274,
+      "Ġtritium": 46275,
+      "ontamination": 46276,
+      "coef": 46277,
+      "Ġpropell": 46278,
+      "triple": 46279,
+      "Ġimmense": 46280,
+      "Ġcomplained": 46281,
+      "Ġdielectrics": 46282,
+      "ĠCardiomy": 46283,
+      "Ġflooded": 46284,
+      "ĠCovariance": 46285,
+      "Attendance": 46286,
+      "TMP": 46287,
+      "Ġsob": 46288,
+      "ĠSonic": 46289,
+      "ĠFTS": 46290,
+      "ĠRSD": 46291,
+      "essors": 46292,
+      "ĠWon": 46293,
+      "iffs": 46294,
+      "Ġflowchart": 46295,
+      "ĠElemental": 46296,
+      "Ġìŀ": 46297,
+      "Ġfoliage": 46298,
+      "differentiated": 46299,
+      "ĠGlobular": 46300,
+      "Ġperceptron": 46301,
+      "candidate": 46302,
+      "Social": 46303,
+      "Witt": 46304,
+      "dyn": 46305,
+      "paces": 46306,
+      "ĠmGlu": 46307,
+      "Ġbanned": 46308,
+      "olinite": 46309,
+      "ĠFriends": 46310,
+      "ĠLibraries": 46311,
+      "unces": 46312,
+      "ĠReach": 46313,
+      "ĠSkills": 46314,
+      "Ġrecipes": 46315,
+      "Ġcannula": 46316,
+      "ĠOrthodox": 46317,
+      "ĠCarbohydrate": 46318,
+      "Ġaromatase": 46319,
+      "Åijs": 46320,
+      "Ġemanating": 46321,
+      "elected": 46322,
+      "Ġtense": 46323,
+      "ĠFLC": 46324,
+      "ĠLET": 46325,
+      "herjee": 46326,
+      "Ġsubband": 46327,
+      "ophone": 46328,
+      "ĠActual": 46329,
+      "msgs": 46330,
+      "EMD": 46331,
+      "ISON": 46332,
+      "leyball": 46333,
+      "ĠNiu": 46334,
+      "Ġberries": 46335,
+      "diagnostic": 46336,
+      "NER": 46337,
+      "ĠdΩ": 46338,
+      "percentage": 46339,
+      "ĠHerman": 46340,
+      "ĠGSD": 46341,
+      "Ġsubproblem": 46342,
+      "overall": 46343,
+      "ophor": 46344,
+      "Ġdelocalized": 46345,
+      "account": 46346,
+      "ĠGeographical": 46347,
+      "distances": 46348,
+      "Ġàµ": 46349,
+      "Ġneurotoxic": 46350,
+      "opodia": 46351,
+      "ĠDicer": 46352,
+      "ĠðxÃŀ": 46353,
+      "Ġdunes": 46354,
+      "Ġwhit": 46355,
+      "ĠImmediate": 46356,
+      "Ġ̸": 46357,
+      "Ġadhesives": 46358,
+      "ĠNSs": 46359,
+      "Ġguessing": 46360,
+      "ĠColumbus": 46361,
+      "ĠUrugu": 46362,
+      "behaviour": 46363,
+      "ĠSerbian": 46364,
+      "benzodioxol": 46365,
+      "implementation": 46366,
+      "osensitive": 46367,
+      "ĠFill": 46368,
+      "phage": 46369,
+      "recovery": 46370,
+      "ESR": 46371,
+      "Ġanalysts": 46372,
+      "Ġdissatisfaction": 46373,
+      "banded": 46374,
+      "ĠDepressive": 46375,
+      "ĠRTs": 46376,
+      "Refs": 46377,
+      "millimeter": 46378,
+      "ĠOlsen": 46379,
+      "ampton": 46380,
+      "ĠACA": 46381,
+      "ĠAvian": 46382,
+      "ĠFowler": 46383,
+      "ubini": 46384,
+      "estamps": 46385,
+      "ĠProtest": 46386,
+      "Connection": 46387,
+      "Ġmerchant": 46388,
+      "ĠENC": 46389,
+      "ĠRyu": 46390,
+      "ĠLymphoma": 46391,
+      "ĠLarry": 46392,
+      "Ġjaponicum": 46393,
+      "ĠSymbols": 46394,
+      "Lib": 46395,
+      "VG": 46396,
+      "ĠTav": 46397,
+      "ĠAssim": 46398,
+      "ĠLeung": 46399,
+      "dependency": 46400,
+      "largest": 46401,
+      "ĠDOE": 46402,
+      "Ġaligns": 46403,
+      "oflurane": 46404,
+      "ĠAdjusted": 46405,
+      "Ġpeculiarities": 46406,
+      "decrease": 46407,
+      "ĠPlacement": 46408,
+      "vig": 46409,
+      "zak": 46410,
+      "Ġpenta": 46411,
+      "Ġfres": 46412,
+      "Ġacros": 46413,
+      "Ġsolvability": 46414,
+      "ansions": 46415,
+      "ALA": 46416,
+      "Ġmalfunction": 46417,
+      "ĠGiovanni": 46418,
+      "AOR": 46419,
+      "Had": 46420,
+      "Ġporn": 46421,
+      "undice": 46422,
+      "ĠUi": 46423,
+      "Ġexpelled": 46424,
+      "ĠAnk": 46425,
+      "Ġdiscounting": 46426,
+      "ĠRegulating": 46427,
+      "astery": 46428,
+      "phenylethyl": 46429,
+      "Ġcastration": 46430,
+      "Ġerythromycin": 46431,
+      "Ġbifunctional": 46432,
+      "��": 46433,
+      "ĠAlgeria": 46434,
+      "mess": 46435,
+      "Ġwis": 46436,
+      "ĠTay": 46437,
+      "assumed": 46438,
+      "Ġescalation": 46439,
+      "Ġhydroper": 46440,
+      "Ġcallosum": 46441,
+      "Ġatomization": 46442,
+      "ĠSAW": 46443,
+      "Ġacetylcholinesterase": 46444,
+      "Ġsucceeds": 46445,
+      "Ġphysiotherapy": 46446,
+      "tro": 46447,
+      "Ġmason": 46448,
+      "ĠTMB": 46449,
+      "Ġphant": 46450,
+      "Ġadjusts": 46451,
+      "antha": 46452,
+      "ĠEisenstein": 46453,
+      "Ġshorthand": 46454,
+      "GABA": 46455,
+      "Ġprover": 46456,
+      "Ġpatrol": 46457,
+      "ĠModal": 46458,
+      "ollaries": 46459,
+      "ĠInterfacial": 46460,
+      "ĠCIA": 46461,
+      "attn": 46462,
+      "ĠCryptococcus": 46463,
+      "athecal": 46464,
+      "ĠFreshwater": 46465,
+      "Ġspectrogram": 46466,
+      "opidogrel": 46467,
+      "morphism": 46468,
+      "Ġrelapsing": 46469,
+      "Ġgeneralizable": 46470,
+      "ĠShale": 46471,
+      "ĠTransplant": 46472,
+      "contraction": 46473,
+      "URI": 46474,
+      "ĠPetrov": 46475,
+      "ĠSliding": 46476,
+      "Ġanteriorly": 46477,
+      "Ġquasilinear": 46478,
+      "Ġripples": 46479,
+      "ZP": 46480,
+      "bacterial": 46481,
+      "spr": 46482,
+      "animal": 46483,
+      "Ġreporters": 46484,
+      "ĠBSS": 46485,
+      "ĠDia": 46486,
+      "ĠRSC": 46487,
+      "ounding": 46488,
+      "ITHM": 46489,
+      "logical": 46490,
+      "Ġpolycarbonate": 46491,
+      "Animal": 46492,
+      "umbai": 46493,
+      "Ġarchived": 46494,
+      "ĠDurham": 46495,
+      "âĸĪ": 46496,
+      "ĠVermont": 46497,
+      "Ġpw": 46498,
+      "essen": 46499,
+      "Ġconstexpr": 46500,
+      "ĠPruss": 46501,
+      "Ġsharpness": 46502,
+      "divide": 46503,
+      "primitive": 46504,
+      "Ġacrylate": 46505,
+      "MYC": 46506,
+      "ĠMonday": 46507,
+      "ĠSrinivas": 46508,
+      "Born": 46509,
+      "attice": 46510,
+      "omorpha": 46511,
+      "ĠMERS": 46512,
+      "ĠFactory": 46513,
+      "ĠWN": 46514,
+      "rectile": 46515,
+      "Ġheats": 46516,
+      "UNK": 46517,
+      "Ġsynchronize": 46518,
+      "ĠAttenuation": 46519,
+      "Children": 46520,
+      "Pat": 46521,
+      "pregnant": 46522,
+      "Ġwished": 46523,
+      "Ġthawing": 46524,
+      "ĠBey": 46525,
+      "ĠDÃŃaz": 46526,
+      "Ġleather": 46527,
+      "ĠUnic": 46528,
+      "Ġspecialised": 46529,
+      "Ġcatalytically": 46530,
+      "PLGA": 46531,
+      "hydroxyethyl": 46532,
+      "Ġmagmas": 46533,
+      "Ġpronoun": 46534,
+      "Ġeutrophication": 46535,
+      "ĠWeekly": 46536,
+      "MHD": 46537,
+      "malloc": 46538,
+      "ecologic": 46539,
+      "ilo": 46540,
+      "ĠFrequencies": 46541,
+      "Ġorchestra": 46542,
+      "Ġmetabolomic": 46543,
+      "ĠBlockade": 46544,
+      "Ġasserted": 46545,
+      "ĠLewy": 46546,
+      "Ġalleviating": 46547,
+      "Ġocclusions": 46548,
+      "Ġchoroid": 46549,
+      "technical": 46550,
+      "Ġenvisioned": 46551,
+      "ĠHousing": 46552,
+      "Pn": 46553,
+      "ĠTECH": 46554,
+      "ĠSSH": 46555,
+      "ĠValle": 46556,
+      "ylmethyl": 46557,
+      "Ġphloem": 46558,
+      "ĠProjects": 46559,
+      "button": 46560,
+      "Ġaccelerometers": 46561,
+      "umni": 46562,
+      "ĠHandling": 46563,
+      "Ġvaso": 46564,
+      "permeable": 46565,
+      "Ġcords": 46566,
+      "ĠCf": 46567,
+      "ĠDz": 46568,
+      "Ġeditions": 46569,
+      "Ġhumerus": 46570,
+      "doors": 46571,
+      "Ġdorsolateral": 46572,
+      "Ġaptamers": 46573,
+      "Ġcommodities": 46574,
+      "osperms": 46575,
+      "Ġprednisone": 46576,
+      "IQ": 46577,
+      "Metal": 46578,
+      "tus": 46579,
+      "Ġisotopy": 46580,
+      "ĠTheater": 46581,
+      "iffi": 46582,
+      "Ġyarn": 46583,
+      "deletion": 46584,
+      "ĠQPO": 46585,
+      "Ġmultiobjective": 46586,
+      "Ġurchin": 46587,
+      "Ġpulsations": 46588,
+      "ĠSRP": 46589,
+      "ðtÃŀ": 46590,
+      "glucoside": 46591,
+      "Ġdepartures": 46592,
+      "PyObject": 46593,
+      "ĠBandwidth": 46594,
+      "ĠAcceptance": 46595,
+      "reys": 46596,
+      "ĠION": 46597,
+      "Ġcompuls": 46598,
+      "ĠJW": 46599,
+      "Ġparthen": 46600,
+      "Close": 46601,
+      "ĠBaTiO": 46602,
+      "ñoz": 46603,
+      "aggregate": 46604,
+      "Initially": 46605,
+      "qh": 46606,
+      "ĠCancers": 46607,
+      "opin": 46608,
+      "never": 46609,
+      "isman": 46610,
+      "Ġconstancy": 46611,
+      "Ġtrucks": 46612,
+      "Ġvisualisation": 46613,
+      "ĠIllness": 46614,
+      "Ġsulphide": 46615,
+      "ĠMetabolites": 46616,
+      "Ġoxysporum": 46617,
+      "HPP": 46618,
+      "Ġnoradrenaline": 46619,
+      "Ġcommutativity": 46620,
+      "Quad": 46621,
+      "NiO": 46622,
+      "ĠGetting": 46623,
+      "Ġbait": 46624,
+      "Ġë°": 46625,
+      "Ġmentally": 46626,
+      "Ġauroral": 46627,
+      "ĠDrawing": 46628,
+      "Sin": 46629,
+      "receiver": 46630,
+      "atov": 46631,
+      "isotope": 46632,
+      "Ġisothi": 46633,
+      "ĠSenes": 46634,
+      "ĠACO": 46635,
+      "ĠGCT": 46636,
+      "ysmal": 46637,
+      "ĠVog": 46638,
+      "Ġdistractors": 46639,
+      "Ġconnectedness": 46640,
+      "Ġaccumbens": 46641,
+      "äck": 46642,
+      "hydrated": 46643,
+      "Ġpharmacodynamic": 46644,
+      "Ġmineralogy": 46645,
+      "Ġarthropods": 46646,
+      "Ġmycotoxins": 46647,
+      "Ġbattles": 46648,
+      "ĠSara": 46649,
+      "ĠEIS": 46650,
+      "ĠWinn": 46651,
+      "Ġlimbic": 46652,
+      "WORK": 46653,
+      "Ž": 46654,
+      "Ġeaten": 46655,
+      "ĠTod": 46656,
+      "apillary": 46657,
+      "oxyp": 46658,
+      "ĠNewly": 46659,
+      "Ġcamel": 46660,
+      "arrison": 46661,
+      "ECTOR": 46662,
+      "Ġhopefully": 46663,
+      "ĠHurwitz": 46664,
+      "Ġibuprofen": 46665,
+      "ĠFIRST": 46666,
+      "Ġbistable": 46667,
+      "Ġdismissed": 46668,
+      "gat": 46669,
+      "inogen": 46670,
+      "ĠPON": 46671,
+      "phas": 46672,
+      "ĠKorn": 46673,
+      "Ġpolyaniline": 46674,
+      "ĠMicroscope": 46675,
+      "Ġmucous": 46676,
+      "Ġcollisionless": 46677,
+      "hydrogenase": 46678,
+      "Build": 46679,
+      "pairing": 46680,
+      "ĠWIMP": 46681,
+      "builtin": 46682,
+      "ĠSeparate": 46683,
+      "ĠCunningham": 46684,
+      "ĠNecessary": 46685,
+      "Ġbry": 46686,
+      "ecrosis": 46687,
+      "ĠLSS": 46688,
+      "Ġsyphilis": 46689,
+      "ĠVid": 46690,
+      "Ġcarrot": 46691,
+      "ĠResistant": 46692,
+      "registration": 46693,
+      "Ġmyopathy": 46694,
+      "Ġangry": 46695,
+      "MDR": 46696,
+      "Ġhypothesised": 46697,
+      "ĠVolterra": 46698,
+      "elevation": 46699,
+      "Ġmycobacteria": 46700,
+      "Ġcaudate": 46701,
+      "iidae": 46702,
+      "ĠÇ": 46703,
+      "ĠDich": 46704,
+      "ĠReth": 46705,
+      "ellus": 46706,
+      "chamber": 46707,
+      "shine": 46708,
+      "ochore": 46709,
+      "ĠColumns": 46710,
+      "COUNT": 46711,
+      "Ġïĥ²": 46712,
+      "ĠPrimordial": 46713,
+      "Ġnegotiations": 46714,
+      "stedt": 46715,
+      "RII": 46716,
+      "UES": 46717,
+      "tiques": 46718,
+      "ĠPfe": 46719,
+      "Ġplast": 46720,
+      "pron": 46721,
+      "ĠZw": 46722,
+      "inkler": 46723,
+      "Ġmetabolome": 46724,
+      "EGA": 46725,
+      "ĠSpectrophot": 46726,
+      "Ġubiquity": 46727,
+      "ĠElectrodes": 46728,
+      "Ġchondro": 46729,
+      "DomainIs": 46730,
+      "ĠResidues": 46731,
+      "ĠdnsDomainIs": 46732,
+      "DIC": 46733,
+      "pth": 46734,
+      "Ġaest": 46735,
+      "Ġcient": 46736,
+      "Ġpessim": 46737,
+      "Ġreinst": 46738,
+      "ĠSans": 46739,
+      "endazole": 46740,
+      "ĠUrine": 46741,
+      "Ġsubacute": 46742,
+      "iximab": 46743,
+      "Ġprofitable": 46744,
+      "Ġmaximise": 46745,
+      "ĠDelaware": 46746,
+      "Ġclinicopathologic": 46747,
+      "ThermoFisher": 46748,
+      "FAR": 46749,
+      "RAS": 46750,
+      "witch": 46751,
+      "inactivated": 46752,
+      "enesis": 46753,
+      "unless": 46754,
+      "ĠPanc": 46755,
+      "ĠMTS": 46756,
+      "ĠBast": 46757,
+      "Ġchilling": 46758,
+      "Ġincumbent": 46759,
+      "Ġjelly": 46760,
+      "Ġdistributive": 46761,
+      "Ġcyto": 46762,
+      "schen": 46763,
+      "Ġinducers": 46764,
+      "ĠNonequilibrium": 46765,
+      "ĠRobotics": 46766,
+      "ĠArgentine": 46767,
+      "Ġmeridian": 46768,
+      "Ġhunger": 46769,
+      "Adaptive": 46770,
+      "Ġgor": 46771,
+      "ilepsy": 46772,
+      "Ġnonvanishing": 46773,
+      "Ġpeti": 46774,
+      "ĠMetformin": 46775,
+      "Ġbiomaterial": 46776,
+      "Ġantennal": 46777,
+      "ĠAffective": 46778,
+      "ĠAquatic": 46779,
+      "enediamine": 46780,
+      "ĠSiberia": 46781,
+      "ĠPenicillium": 46782,
+      "Functions": 46783,
+      "Ġlec": 46784,
+      "Ġfeld": 46785,
+      "ĠSpart": 46786,
+      "ĠCement": 46787,
+      "addi": 46788,
+      "sek": 46789,
+      "ĠNp": 46790,
+      "olesky": 46791,
+      "ĠMacroscopic": 46792,
+      "ères": 46793,
+      "Ġcaveat": 46794,
+      "Ġcourtship": 46795,
+      "mice": 46796,
+      "Ġfence": 46797,
+      "Ġmined": 46798,
+      "ulink": 46799,
+      "IDA": 46800,
+      "Ġtruncate": 46801,
+      "ĠCatalan": 46802,
+      "Ġtranst": 46803,
+      "Ġamendments": 46804,
+      "uncertainty": 46805,
+      "Ġoropharyngeal": 46806,
+      "ĠAid": 46807,
+      "oulder": 46808,
+      "ĠIncident": 46809,
+      "ĠáIJ": 46810,
+      "angiogenesis": 46811,
+      "ĠBEH": 46812,
+      "Ġicosa": 46813,
+      "ĠFOXP": 46814,
+      "fragment": 46815,
+      "Ġscintillator": 46816,
+      "JO": 46817,
+      "Law": 46818,
+      "ĠpL": 46819,
+      "Ġetoposide": 46820,
+      "Ġpolyaden": 46821,
+      "Ġhabitual": 46822,
+      "Ġtaxi": 46823,
+      "Ġcumulant": 46824,
+      "Ġhindrance": 46825,
+      "trigger": 46826,
+      "ratios": 46827,
+      "ilio": 46828,
+      "ĠPIR": 46829,
+      "ĠTheod": 46830,
+      "ĠMorton": 46831,
+      "ĠHaf": 46832,
+      "ĠOch": 46833,
+      "ĠExo": 46834,
+      "Ġurtic": 46835,
+      "ĠCFRP": 46836,
+      "Screen": 46837,
+      "Slice": 46838,
+      "Ġmushrooms": 46839,
+      "Ġevanescent": 46840,
+      "Sx": 46841,
+      "ËIJ": 46842,
+      "ìŀ": 46843,
+      "Ġsigm": 46844,
+      "icl": 46845,
+      "Ġguests": 46846,
+      "ĠGIST": 46847,
+      "Ġdeformities": 46848,
+      "polyacrylamide": 46849,
+      "Significant": 46850,
+      "Ġimpressions": 46851,
+      "jmath": 46852,
+      "emoral": 46853,
+      "ĠBn": 46854,
+      "ĠHDR": 46855,
+      "ĠKeck": 46856,
+      "Ġvaline": 46857,
+      "spi": 46858,
+      "iterate": 46859,
+      "Ġsync": 46860,
+      "otiana": 46861,
+      "Interval": 46862,
+      "ĠBrauer": 46863,
+      "Ġsticky": 46864,
+      "ĠNeuroscience": 46865,
+      "Baxter": 46866,
+      "Ġcasts": 46867,
+      "allocation": 46868,
+      "neal": 46869,
+      "Ġbiop": 46870,
+      "Ġrestorations": 46871,
+      "Images": 46872,
+      "mitic": 46873,
+      "ĠElevation": 46874,
+      "Ġabstinence": 46875,
+      "ĠLesser": 46876,
+      "ĠRainfall": 46877,
+      "PAM": 46878,
+      "Wol": 46879,
+      "usch": 46880,
+      "Ġpromisc": 46881,
+      "naïve": 46882,
+      "Ġdeduc": 46883,
+      "accharide": 46884,
+      "Ġnominally": 46885,
+      "ĠExploratory": 46886,
+      "Ġreconciliation": 46887,
+      "linalg": 46888,
+      "TCR": 46889,
+      "Ġsore": 46890,
+      "ĠNab": 46891,
+      "Ġoutgroup": 46892,
+      "Ġmonophosphate": 46893,
+      "insu": 46894,
+      "ĠAddis": 46895,
+      "SPR": 46896,
+      "pointing": 46897,
+      "HERE": 46898,
+      "ĠTechnological": 46899,
+      "Ġcochlea": 46900,
+      "Ġspheroidal": 46901,
+      "ĠBaldwin": 46902,
+      "Feed": 46903,
+      "Ġfusing": 46904,
+      "Ġasper": 46905,
+      "Ġexosomal": 46906,
+      "ĠLinguistic": 46907,
+      "SCA": 46908,
+      "ĠEmpty": 46909,
+      "Ġvacant": 46910,
+      "glycol": 46911,
+      "immunoprecipitation": 46912,
+      "ĠITER": 46913,
+      "SnO": 46914,
+      "patterns": 46915,
+      "continental": 46916,
+      "ĠAccelerating": 46917,
+      "ĠAveraging": 46918,
+      "Ġchemoattractant": 46919,
+      "hb": 46920,
+      "sulph": 46921,
+      "ĠBx": 46922,
+      "Ġcomplicating": 46923,
+      "ĠWare": 46924,
+      "Ġsoaking": 46925,
+      "Ġupregulate": 46926,
+      "---------": 46927,
+      "Ġsemester": 46928,
+      "ĠBrod": 46929,
+      "Ġcascading": 46930,
+      "ĠCastell": 46931,
+      "Ġẽ": 46932,
+      "ĠEQUATIONS": 46933,
+      "Ġparsimonious": 46934,
+      "Ġsorbent": 46935,
+      "Ġeug": 46936,
+      "odin": 46937,
+      "ĠWig": 46938,
+      "ĠThir": 46939,
+      "Ġsolv": 46940,
+      "Ġcarboplatin": 46941,
+      "Ġzebra": 46942,
+      "venient": 46943,
+      "ĠmedRxiv": 46944,
+      "Ġautobi": 46945,
+      "Ġrepeatable": 46946,
+      "Ġmigrations": 46947,
+      "Ġд": 46948,
+      "holonomic": 46949,
+      "Ġmoderator": 46950,
+      "Ġchimera": 46951,
+      "ĠGrassmannian": 46952,
+      "ĠRonald": 46953,
+      "ĠVega": 46954,
+      "astes": 46955,
+      "Ġquotes": 46956,
+      "Ġmonic": 46957,
+      "Ġprecoding": 46958,
+      "ĠAssisted": 46959,
+      "ĠNetworking": 46960,
+      "Ġfabricating": 46961,
+      "Ġbotanical": 46962,
+      "Ġswarms": 46963,
+      "Ġmartensitic": 46964,
+      "elliptic": 46965,
+      "pherd": 46966,
+      "baryon": 46967,
+      "xfe": 46968,
+      "route": 46969,
+      "ĠFIL": 46970,
+      "opies": 46971,
+      "ĠPCBs": 46972,
+      "Ġerasure": 46973,
+      "ĠRemodeling": 46974,
+      "Ġanaer": 46975,
+      "Smad": 46976,
+      "injured": 46977,
+      "Ġimmunocompetent": 46978,
+      "dell": 46979,
+      "failed": 46980,
+      "Ġsinking": 46981,
+      "oracic": 46982,
+      "Ġdred": 46983,
+      "ĠVDR": 46984,
+      "Ġconnectors": 46985,
+      "Ġintratumoral": 46986,
+      "Ġcommutators": 46987,
+      "ĠAleks": 46988,
+      "ĠDicty": 46989,
+      "Ak": 46990,
+      "Ġrecalc": 46991,
+      "Ġisl": 46992,
+      "otrim": 46993,
+      "ncephal": 46994,
+      "ĠRees": 46995,
+      "Ġsteatohepatitis": 46996,
+      "ĠPolarized": 46997,
+      "SBATCH": 46998,
+      "ĠCrossing": 46999,
+      "Accuracy": 47000,
+      "ĠGiardia": 47001,
+      "ĠNovo": 47002,
+      "Ġvigilance": 47003,
+      "Ġphosphatidylcholine": 47004,
+      "ĠUEFA": 47005,
+      "Jim": 47006,
+      "Ġfasted": 47007,
+      "ĠTiny": 47008,
+      "Ġlang": 47009,
+      "issociation": 47010,
+      "Auto": 47011,
+      "ĠNorfolk": 47012,
+      "ĠArms": 47013,
+      "ĠSWI": 47014,
+      "ĠAmbros": 47015,
+      "transfection": 47016,
+      "Oryza": 47017,
+      "harm": 47018,
+      "ĠDs": 47019,
+      "Ġintrag": 47020,
+      "Ġcaller": 47021,
+      "Ġwritings": 47022,
+      "ĠElast": 47023,
+      "ĠMarvel": 47024,
+      "ĠImmunodeficiency": 47025,
+      "ĠMillion": 47026,
+      "Texture": 47027,
+      "ĠIceCube": 47028,
+      "snap": 47029,
+      "Ġenjoys": 47030,
+      "ĠChapel": 47031,
+      "ĠEstablishing": 47032,
+      "Actually": 47033,
+      "Ġphosphorylates": 47034,
+      "Ġchinensis": 47035,
+      "Ġrhabdomy": 47036,
+      "Ġemphysema": 47037,
+      "Middle": 47038,
+      "nant": 47039,
+      "Ñħ": 47040,
+      "Ġtart": 47041,
+      "lowest": 47042,
+      "hemia": 47043,
+      "Ġutilising": 47044,
+      "constit": 47045,
+      "Ġmagmatism": 47046,
+      "оÑĢ": 47047,
+      "ĠHasan": 47048,
+      "dispersed": 47049,
+      "Hear": 47050,
+      "Qt": 47051,
+      "zations": 47052,
+      "alon": 47053,
+      "ĠStau": 47054,
+      "ĠAmer": 47055,
+      "osystems": 47056,
+      "Ġdemarc": 47057,
+      "ĠNeoproterozoic": 47058,
+      "ĠMek": 47059,
+      "ĠDisclosure": 47060,
+      "Ġhematocrit": 47061,
+      "ĠCytoscape": 47062,
+      "Ġramification": 47063,
+      "Ġcommunicative": 47064,
+      "Ġbutterflies": 47065,
+      "Ġantisera": 47066,
+      "Ġaestivum": 47067,
+      "Bra": 47068,
+      "LTP": 47069,
+      "socket": 47070,
+      "ĠCherenkov": 47071,
+      "Ġchlam": 47072,
+      "angial": 47073,
+      "ultured": 47074,
+      "enged": 47075,
+      "ĠClinton": 47076,
+      "Ġmyoblasts": 47077,
+      "ĠCompensation": 47078,
+      "ymmetrically": 47079,
+      "Ġemployer": 47080,
+      "ozol": 47081,
+      "ĠSAXS": 47082,
+      "Ġretinas": 47083,
+      "piperidine": 47084,
+      "XYZ": 47085,
+      "ĠRoughly": 47086,
+      "Prep": 47087,
+      "Ġbinge": 47088,
+      "Ġerect": 47089,
+      "ĠOPER": 47090,
+      "Ġstressor": 47091,
+      "Christ": 47092,
+      "ĠPDZ": 47093,
+      "Ġsubstan": 47094,
+      "ĠSnail": 47095,
+      "Ġlamellae": 47096,
+      "ĠCycling": 47097,
+      "shifting": 47098,
+      "ĠHsieh": 47099,
+      "verify": 47100,
+      "Ġpreimage": 47101,
+      "Ġartillery": 47102,
+      "Ġepil": 47103,
+      "ĠApost": 47104,
+      "Ġhelmet": 47105,
+      "Ġmachined": 47106,
+      "ĠMinneapolis": 47107,
+      "ĠCryp": 47108,
+      "Ġsituational": 47109,
+      "passing": 47110,
+      "quinazolin": 47111,
+      "ĠCroatian": 47112,
+      "Ġstaircase": 47113,
+      "Bonnet": 47114,
+      "NLP": 47115,
+      "cium": 47116,
+      "Ġskeletons": 47117,
+      "Ġoxim": 47118,
+      "orib": 47119,
+      "Ġreticular": 47120,
+      "ĠSLS": 47121,
+      "ĠAromatic": 47122,
+      "ĠKes": 47123,
+      "Ġphor": 47124,
+      "Ġinvocation": 47125,
+      "Ġdozens": 47126,
+      "aively": 47127,
+      "Ġdetectability": 47128,
+      "Ġconcerted": 47129,
+      "yrins": 47130,
+      "ĠProcessor": 47131,
+      "Ġtolerable": 47132,
+      "attached": 47133,
+      "Ġannexin": 47134,
+      "ĠROSAT": 47135,
+      "ĠAlternate": 47136,
+      "ĠWavelength": 47137,
+      "ĠWillis": 47138,
+      "Ġsemicontinuous": 47139,
+      "Ġadvocacy": 47140,
+      "Ġobligation": 47141,
+      "chanter": 47142,
+      "ĠInsertion": 47143,
+      "Ġsymbiont": 47144,
+      "ZM": 47145,
+      "Ġtars": 47146,
+      "rof": 47147,
+      "Ġrevival": 47148,
+      "ĠTST": 47149,
+      "ĠEMP": 47150,
+      "Ġmex": 47151,
+      "ullin": 47152,
+      "ĠAdop": 47153,
+      "ĠDNAs": 47154,
+      "Ġemployers": 47155,
+      "MTs": 47156,
+      "ĠMartÃŃn": 47157,
+      "electrodes": 47158,
+      "ĠMedicaid": 47159,
+      "Ġtgt": 47160,
+      "Ġlognormal": 47161,
+      "ĠFrames": 47162,
+      "Ġpermissive": 47163,
+      "ĠArduino": 47164,
+      "Ġsemilinear": 47165,
+      "ĠAssign": 47166,
+      "ĠPrEP": 47167,
+      "ĠSiamese": 47168,
+      "benzimidazol": 47169,
+      "connectivity": 47170,
+      "ĠPEI": 47171,
+      "Ġbisulfite": 47172,
+      "Ġacetyltransferase": 47173,
+      "Ġswimmer": 47174,
+      "juven": 47175,
+      "Ġjejunum": 47176,
+      "ĠCincinnati": 47177,
+      "tai": 47178,
+      "ĠQI": 47179,
+      "ĠCommut": 47180,
+      "spacing": 47181,
+      "Ġaffords": 47182,
+      "itisation": 47183,
+      "elasticity": 47184,
+      "Ġdragon": 47185,
+      "Ġproteasomal": 47186,
+      "Ġpant": 47187,
+      "ĠNitro": 47188,
+      "Ġspic": 47189,
+      "Ġnanopl": 47190,
+      "ĠAllied": 47191,
+      "Ġthorax": 47192,
+      "ĠFTO": 47193,
+      "ĠJurkat": 47194,
+      "chiatry": 47195,
+      "young": 47196,
+      "directions": 47197,
+      "Ġneocortex": 47198,
+      "ĠKik": 47199,
+      "ango": 47200,
+      "clay": 47201,
+      "iodo": 47202,
+      "Ġabovementioned": 47203,
+      "ĠGuardian": 47204,
+      "Conjecture": 47205,
+      "ĠTrend": 47206,
+      "Ġfertilized": 47207,
+      "ĠSulfate": 47208,
+      "ochronology": 47209,
+      "Ġcraniofacial": 47210,
+      "ĠSaskatchewan": 47211,
+      "QQ": 47212,
+      "hman": 47213,
+      "Ġzym": 47214,
+      "logs": 47215,
+      "Ġïģ®": 47216,
+      "Ġgraduating": 47217,
+      "pinene": 47218,
+      "ĠîĢ": 47219,
+      "Ġetiological": 47220,
+      "ĠComprehension": 47221,
+      "Ġwandering": 47222,
+      "Ġlan": 47223,
+      "Ġsyst": 47224,
+      "returns": 47225,
+      "MOF": 47226,
+      "choalveolar": 47227,
+      "ĠArmen": 47228,
+      "Ġbimetallic": 47229,
+      "ĠPollen": 47230,
+      "Files": 47231,
+      "Ġssp": 47232,
+      "ENSI": 47233,
+      "ĠYus": 47234,
+      "Ġfinest": 47235,
+      "AGEN": 47236,
+      "Ġmicrobiomes": 47237,
+      "Ġpalind": 47238,
+      "Ġpetals": 47239,
+      "ĠRadiotherapy": 47240,
+      "ophenone": 47241,
+      "speaker": 47242,
+      "Ġcopepods": 47243,
+      "Ġkanamycin": 47244,
+      "Ġdegranulation": 47245,
+      "Construct": 47246,
+      "alter": 47247,
+      "ĠFgf": 47248,
+      "ĠNBS": 47249,
+      "ĠIncomplete": 47250,
+      "Ġparcel": 47251,
+      "neau": 47252,
+      "ĠÃIJ": 47253,
+      "ĠCHA": 47254,
+      "Ġduals": 47255,
+      "Ġsilicates": 47256,
+      "ĠGlobally": 47257,
+      "Ġkinesin": 47258,
+      "fid": 47259,
+      "ĠCPD": 47260,
+      "ĠYad": 47261,
+      "Ġdepress": 47262,
+      "ODY": 47263,
+      "ĠHistograms": 47264,
+      "ĠSummarization": 47265,
+      "automatic": 47266,
+      "ĠDomin": 47267,
+      "otransformation": 47268,
+      "Ġventricles": 47269,
+      "Widget": 47270,
+      "ĠPetersburg": 47271,
+      "Ġcholangiocarcinoma": 47272,
+      "Ġnectar": 47273,
+      "PIC": 47274,
+      "Scope": 47275,
+      "Tek": 47276,
+      "nitz": 47277,
+      "ĠPHD": 47278,
+      "Ġspiro": 47279,
+      "ĠCOG": 47280,
+      "ĠDioxide": 47281,
+      "conductivity": 47282,
+      "ĠGranger": 47283,
+      "ĠWearable": 47284,
+      "ĠKenneth": 47285,
+      "CCR": 47286,
+      "LINK": 47287,
+      "ĠÜ": 47288,
+      "retic": 47289,
+      "lya": 47290,
+      "Ġdemocratic": 47291,
+      "Ġradiograph": 47292,
+      "ĠRelax": 47293,
+      "ĠIncubation": 47294,
+      "ĠDenoising": 47295,
+      "COLOR": 47296,
+      "ĠClosure": 47297,
+      "HMM": 47298,
+      "urd": 47299,
+      "rada": 47300,
+      "ĠRv": 47301,
+      "ĠLuz": 47302,
+      "alls": 47303,
+      "Ġmultispectral": 47304,
+      "INED": 47305,
+      "SCN": 47306,
+      "Ġdyslexia": 47307,
+      "Ġsettlers": 47308,
+      "ĠVLSI": 47309,
+      "Ġavid": 47310,
+      "Ġlarynx": 47311,
+      "ĠChess": 47312,
+      "ĠFAA": 47313,
+      "Ġdefender": 47314,
+      "Ġlipolysis": 47315,
+      "ĠElmer": 47316,
+      "ĠAffymetrix": 47317,
+      "Ġrhodamine": 47318,
+      "Morph": 47319,
+      "Site": 47320,
+      "purity": 47321,
+      "ĠÊ": 47322,
+      "ĠTank": 47323,
+      "ĠMiao": 47324,
+      "Ġrecrystall": 47325,
+      "Weyl": 47326,
+      "ĠGuil": 47327,
+      "Ġmisfolded": 47328,
+      "suited": 47329,
+      "ĠApproximations": 47330,
+      "ĠABCB": 47331,
+      "donor": 47332,
+      "GWAS": 47333,
+      "---------------": 47334,
+      "Ġputida": 47335,
+      "Ġimpingement": 47336,
+      "yaml": 47337,
+      "Hill": 47338,
+      "Ġtl": 47339,
+      "agua": 47340,
+      "timing": 47341,
+      "Ġregenerate": 47342,
+      "Ġmultilingual": 47343,
+      "rador": 47344,
+      "classifier": 47345,
+      "ĠJohansson": 47346,
+      "Ġsulfides": 47347,
+      "hammer": 47348,
+      "Ġwalked": 47349,
+      "Ġallocating": 47350,
+      "ĠGustav": 47351,
+      "Ġimmunoprecipitated": 47352,
+      "ĠBrisbane": 47353,
+      "Ġsandwiched": 47354,
+      "ĠChatterjee": 47355,
+      "omandibular": 47356,
+      "Ġosc": 47357,
+      "Ġassass": 47358,
+      "Ġmultistage": 47359,
+      "Ġmultipartite": 47360,
+      "Ġpigmented": 47361,
+      "ĠVisualizing": 47362,
+      "Keys": 47363,
+      "pipeline": 47364,
+      "Ġdubbed": 47365,
+      "Ġcroc": 47366,
+      "ĠDLC": 47367,
+      "ĠRAT": 47368,
+      "ĠNex": 47369,
+      "plica": 47370,
+      "tingham": 47371,
+      "ĠSpider": 47372,
+      "Ġuncle": 47373,
+      "auts": 47374,
+      "ĠHowe": 47375,
+      "Ġarthropod": 47376,
+      "ĠPapad": 47377,
+      "urgy": 47378,
+      "Ġacclim": 47379,
+      "Broad": 47380,
+      "acer": 47381,
+      "vez": 47382,
+      "ĠDivers": 47383,
+      "Ġmodifiable": 47384,
+      "Ġantipsychotics": 47385,
+      "Prog": 47386,
+      "osahexa": 47387,
+      "ambrian": 47388,
+      "ĠIonization": 47389,
+      "ZA": 47390,
+      "oate": 47391,
+      "Ġpays": 47392,
+      "Ġewes": 47393,
+      "Ġbeaches": 47394,
+      "Ġevil": 47395,
+      "ĠCDs": 47396,
+      "naud": 47397,
+      "Ġconformity": 47398,
+      "ĠDMN": 47399,
+      "Ġcollaborate": 47400,
+      "Ġdeteriorate": 47401,
+      "VALID": 47402,
+      "ĠVegas": 47403,
+      "Ġultracent": 47404,
+      "BRA": 47405,
+      "Rub": 47406,
+      "YC": 47407,
+      "fh": 47408,
+      "åľ": 47409,
+      "ĠOWL": 47410,
+      "oseismic": 47411,
+      "oferrin": 47412,
+      "ochthon": 47413,
+      "ĠTNFR": 47414,
+      "smallsetminus": 47415,
+      "ĠArgument": 47416,
+      "Ġgranulocytes": 47417,
+      "Ġramified": 47418,
+      "Ġepiphy": 47419,
+      "fusc": 47420,
+      "ecdot": 47421,
+      "Ġhw": 47422,
+      "ĠNMS": 47423,
+      "ercus": 47424,
+      "Ġtether": 47425,
+      "ĠTrait": 47426,
+      "AgCl": 47427,
+      "ĠNearby": 47428,
+      "Ġhelminth": 47429,
+      "Ġlaevis": 47430,
+      "ĠBAR": 47431,
+      "ĠNancy": 47432,
+      "ĠGyn": 47433,
+      "Ġsecreting": 47434,
+      "Stellar": 47435,
+      "Ġsilhou": 47436,
+      "IMT": 47437,
+      "Ġscaffolding": 47438,
+      "ĠConverter": 47439,
+      "hid": 47440,
+      "Ġnud": 47441,
+      "estrian": 47442,
+      "anno": 47443,
+      "Ġdepiction": 47444,
+      "oremost": 47445,
+      "ĠShand": 47446,
+      "ABCD": 47447,
+      "ĠPDL": 47448,
+      "Ġdysphagia": 47449,
+      "Ġintrat": 47450,
+      "Ġhemip": 47451,
+      "Ġadaptable": 47452,
+      "longmapsto": 47453,
+      "ssbauer": 47454,
+      "ĠMcCarthy": 47455,
+      "ĠAutoimmune": 47456,
+      "ĠCutaneous": 47457,
+      "Inserting": 47458,
+      "Material": 47459,
+      "ĠAa": 47460,
+      "ĠGav": 47461,
+      "Ġmonocular": 47462,
+      "equil": 47463,
+      "ĠGeoff": 47464,
+      "Ġtethered": 47465,
+      "obilized": 47466,
+      "ĠShortly": 47467,
+      "Details": 47468,
+      "Ġrefugee": 47469,
+      "Ġabscisic": 47470,
+      "FBQyx": 47471,
+      "Ġdemocracy": 47472,
+      "crafted": 47473,
+      "difluor": 47474,
+      "yder": 47475,
+      "essment": 47476,
+      "Ġhistopathologic": 47477,
+      "Ġastrocytic": 47478,
+      "Ġwithdrew": 47479,
+      "Ġmoles": 47480,
+      "athic": 47481,
+      "mono": 47482,
+      "manual": 47483,
+      "Ġfoodborne": 47484,
+      "ĠRepository": 47485,
+      "Ġcovert": 47486,
+      "OTE": 47487,
+      "Ġtightness": 47488,
+      "Ġinstantiated": 47489,
+      "Ġwatermarking": 47490,
+      "Ġartemisinin": 47491,
+      "Language": 47492,
+      "OES": 47493,
+      "cant": 47494,
+      "already": 47495,
+      "unts": 47496,
+      "itia": 47497,
+      "ĠKaren": 47498,
+      "Ġalluvial": 47499,
+      "stratigraphy": 47500,
+      "ĠPIV": 47501,
+      "ĠFaces": 47502,
+      "ĠBim": 47503,
+      "applications": 47504,
+      "tails": 47505,
+      "Ġeld": 47506,
+      "IRB": 47507,
+      "ĠINTE": 47508,
+      "ĠNotImplemented": 47509,
+      "Ġmisclassified": 47510,
+      "Ġfertilizers": 47511,
+      "ĠElectricity": 47512,
+      "Ġtributaries": 47513,
+      "ĠDeutsch": 47514,
+      "Ġsleeve": 47515,
+      "fuzzy": 47516,
+      "ĠMTL": 47517,
+      "ĠBres": 47518,
+      "ĠWyn": 47519,
+      "Ġkyr": 47520,
+      "neuronal": 47521,
+      "oxymethyl": 47522,
+      "disorder": 47523,
+      "inches": 47524,
+      "ramidal": 47525,
+      "Ġpolyimide": 47526,
+      "ResNet": 47527,
+      "ĠEdmund": 47528,
+      "Ġdegeneracies": 47529,
+      "utherford": 47530,
+      "Dropout": 47531,
+      "ijĢ": 47532,
+      "Ġvoiced": 47533,
+      "ĠGomes": 47534,
+      "ivities": 47535,
+      "conductance": 47536,
+      "compl": 47537,
+      "vecs": 47538,
+      "Ġtuna": 47539,
+      "ĠKinect": 47540,
+      "Ġconveyed": 47541,
+      "Ġsphingosine": 47542,
+      "bat": 47543,
+      "ĠPurs": 47544,
+      "ounded": 47545,
+      "ĠStam": 47546,
+      "ĠXIII": 47547,
+      "ĠComics": 47548,
+      "MSM": 47549,
+      "SSL": 47550,
+      "Ġperfluor": 47551,
+      "Ġfluorinated": 47552,
+      "folios": 47553,
+      "Ġreposition": 47554,
+      "ĠSerr": 47555,
+      "ĠCors": 47556,
+      "ĠLabs": 47557,
+      "Ġcox": 47558,
+      "ĠAcquired": 47559,
+      "Ġreasoned": 47560,
+      "Genome": 47561,
+      "ĠPiper": 47562,
+      "Ġcompactified": 47563,
+      "Ġherbivore": 47564,
+      "lofenac": 47565,
+      "Ġboss": 47566,
+      "ĠBs": 47567,
+      "ĠEMR": 47568,
+      "Ġshoe": 47569,
+      "Ġcarers": 47570,
+      "Chrom": 47571,
+      "SVP": 47572,
+      "ĠTriangle": 47573,
+      "Ġhematite": 47574,
+      "dorf": 47575,
+      "ĠMovements": 47576,
+      "ĠVesicles": 47577,
+      "Olympus": 47578,
+      "Mol": 47579,
+      "Ġlend": 47580,
+      "uras": 47581,
+      "ĠASE": 47582,
+      "ĠWKB": 47583,
+      "proved": 47584,
+      "ĠKV": 47585,
+      "ĠUART": 47586,
+      "logarithmic": 47587,
+      "ĠADI": 47588,
+      "ĠDoing": 47589,
+      "Ġcease": 47590,
+      "Ġlengthening": 47591,
+      "Ġpyrophosphate": 47592,
+      "Fre": 47593,
+      "ĠCLD": 47594,
+      "ĠMLS": 47595,
+      "ĠPlum": 47596,
+      "Ġpropionate": 47597,
+      "ĠGuatem": 47598,
+      "CKD": 47599,
+      "Ġisos": 47600,
+      "ĠManning": 47601,
+      "neuro": 47602,
+      "OPER": 47603,
+      "ĠWilhelm": 47604,
+      "Ġacademia": 47605,
+      "AChR": 47606,
+      "ĠInertial": 47607,
+      "Occ": 47608,
+      "ujan": 47609,
+      "onas": 47610,
+      "Ġinulin": 47611,
+      "icia": 47612,
+      "andal": 47613,
+      "ĠKahn": 47614,
+      "Ġunmanned": 47615,
+      "ĠCoarse": 47616,
+      "Ġguilty": 47617,
+      "ĠPei": 47618,
+      "ĠLuca": 47619,
+      "ĠFibroblast": 47620,
+      "avian": 47621,
+      "vx": 47622,
+      "Ġdizziness": 47623,
+      "ĠDox": 47624,
+      "ĠHour": 47625,
+      "Ġdecoration": 47626,
+      "Ġverifier": 47627,
+      "rado": 47628,
+      "Ġfootprints": 47629,
+      "Ġdispensable": 47630,
+      "ĠAnaerobic": 47631,
+      "IoT": 47632,
+      "ĠRisks": 47633,
+      "ĠGLS": 47634,
+      "Ġchords": 47635,
+      "oidy": 47636,
+      "Ġneurolog": 47637,
+      "ruh": 47638,
+      "Ġvirtualization": 47639,
+      "Ġprotonation": 47640,
+      "ĠConstantin": 47641,
+      "Ġkeypoints": 47642,
+      "Buck": 47643,
+      "Hopf": 47644,
+      "Much": 47645,
+      "regime": 47646,
+      "Ġpromised": 47647,
+      "aij": 47648,
+      "ĠDesulf": 47649,
+      "ĠFormulas": 47650,
+      "Ġhump": 47651,
+      "lnc": 47652,
+      "ĠSuicide": 47653,
+      "ĠHOMA": 47654,
+      "oglycer": 47655,
+      "ĠProteomics": 47656,
+      "Ġdictate": 47657,
+      "ĠSpermat": 47658,
+      "Fun": 47659,
+      "Ġsag": 47660,
+      "ĠFam": 47661,
+      "eppe": 47662,
+      "ĠJah": 47663,
+      "Ġarisen": 47664,
+      "opharmaceutical": 47665,
+      "SAGE": 47666,
+      "ĠTHIS": 47667,
+      "enhance": 47668,
+      "Ġnapus": 47669,
+      "roe": 47670,
+      "ensch": 47671,
+      "deformation": 47672,
+      "bones": 47673,
+      "ĠErnest": 47674,
+      "irability": 47675,
+      "decom": 47676,
+      "Ġcrustaceans": 47677,
+      "Ġguaranteeing": 47678,
+      "OVAs": 47679,
+      "ĠMulticenter": 47680,
+      "ĠctDNA": 47681,
+      "Ġforaminifera": 47682,
+      "Linn": 47683,
+      "Ġcups": 47684,
+      "esch": 47685,
+      "ĠdF": 47686,
+      "ĠTah": 47687,
+      "pll": 47688,
+      "projects": 47689,
+      "ĠUCI": 47690,
+      "Ġhumanized": 47691,
+      "Ġabsl": 47692,
+      "ĠScho": 47693,
+      "Ġliterals": 47694,
+      "ĠSVR": 47695,
+      "Ġtoxicology": 47696,
+      "pgf": 47697,
+      "ĠIPTG": 47698,
+      "ĠMEASUREM": 47699,
+      "oing": 47700,
+      "ĠPasc": 47701,
+      "ĠBau": 47702,
+      "ĠWannier": 47703,
+      "Ġhypre": 47704,
+      "attributes": 47705,
+      "Ġpreconditioner": 47706,
+      "Writing": 47707,
+      "Ġgypsum": 47708,
+      "yuan": 47709,
+      "Ġupregulates": 47710,
+      "Ġtelec": 47711,
+      "ĠDiscre": 47712,
+      "guard": 47713,
+      "Ġdebates": 47714,
+      "Ġparasitoid": 47715,
+      "Lam": 47716,
+      "tige": 47717,
+      "Ġisopropanol": 47718,
+      "ĠIwas": 47719,
+      "plify": 47720,
+      "indolin": 47721,
+      "ĠApollo": 47722,
+      "Ġlanded": 47723,
+      "Ġbeamline": 47724,
+      "Union": 47725,
+      "Ġreciproc": 47726,
+      "ĠRossby": 47727,
+      "principal": 47728,
+      "Ġdescendant": 47729,
+      "ĠAnalogously": 47730,
+      "Ġderegulation": 47731,
+      "DSM": 47732,
+      "cta": 47733,
+      "Ġrebuilt": 47734,
+      "ĠMund": 47735,
+      "ĠFEC": 47736,
+      "ryn": 47737,
+      "plice": 47738,
+      "ĠYugoslav": 47739,
+      "ĠNorthwestern": 47740,
+      "ĠHomogen": 47741,
+      "ĠLISA": 47742,
+      "Ġinvestor": 47743,
+      "HSA": 47744,
+      "HPO": 47745,
+      "Ġdictionaries": 47746,
+      "ĠCategor": 47747,
+      "Ġcompacted": 47748,
+      "tilled": 47749,
+      "ç»": 47750,
+      "Ġfines": 47751,
+      "urans": 47752,
+      "Ġbetweenness": 47753,
+      "ĠZig": 47754,
+      "schema": 47755,
+      "Ġcommune": 47756,
+      "ĠQuinn": 47757,
+      "Ġanaphylaxis": 47758,
+      "TIES": 47759,
+      "Ġsnowpack": 47760,
+      "ĠDOA": 47761,
+      "agos": 47762,
+      "ĠOdd": 47763,
+      "arde": 47764,
+      "Ġevoke": 47765,
+      "ĠOcular": 47766,
+      "Ġfaulting": 47767,
+      "Ġvolcanism": 47768,
+      "ĠPaleozoic": 47769,
+      "Ġmycelium": 47770,
+      "ĠAdjustment": 47771,
+      "ICT": 47772,
+      "Nov": 47773,
+      "alias": 47774,
+      "ĠTul": 47775,
+      "ĠHh": 47776,
+      "Ġevade": 47777,
+      "ORs": 47778,
+      "Ġstrengthens": 47779,
+      "ĠUSGS": 47780,
+      "Ġlicensing": 47781,
+      "ĠClement": 47782,
+      "ĠPhytophthora": 47783,
+      "rified": 47784,
+      "Ġeighteen": 47785,
+      "Ġtops": 47786,
+      "ĠCLP": 47787,
+      "Ġstabilities": 47788,
+      "ĠPPT": 47789,
+      "ĠBIN": 47790,
+      "ĠRak": 47791,
+      "Ġgenistein": 47792,
+      "volve": 47793,
+      "Ġquicker": 47794,
+      "ĠCaused": 47795,
+      "benefit": 47796,
+      "YB": 47797,
+      "lift": 47798,
+      "Ġhood": 47799,
+      "ĠSCs": 47800,
+      "ofa": 47801,
+      "ĠMicron": 47802,
+      "angiotensin": 47803,
+      "Ġfeathers": 47804,
+      "Ġantiferromagnet": 47805,
+      "DECREF": 47806,
+      "yledons": 47807,
+      "Ġmyriad": 47808,
+      "Ġiz": 47809,
+      "ĠTrough": 47810,
+      "âĪ«": 47811,
+      "hemoglobin": 47812,
+      "ĠEnvelope": 47813,
+      "ĠClick": 47814,
+      "soliton": 47815,
+      "ĠSynchrotron": 47816,
+      "Ġlagged": 47817,
+      "MYB": 47818,
+      "Ġtrophoblast": 47819,
+      "Ġinterrogation": 47820,
+      "onvuls": 47821,
+      "Bac": 47822,
+      "Ġaperiodic": 47823,
+      "Ġgpu": 47824,
+      "Ġpropidium": 47825,
+      "teps": 47826,
+      "ĠKarp": 47827,
+      "ĠVaz": 47828,
+      "ackage": 47829,
+      "onson": 47830,
+      "Instr": 47831,
+      "filer": 47832,
+      "rifugation": 47833,
+      "KOV": 47834,
+      "fourth": 47835,
+      "Ġôı¼IJ": 47836,
+      "hyperbolic": 47837,
+      "schetz": 47838,
+      "Discussion": 47839,
+      "ĠOriented": 47840,
+      "jad": 47841,
+      "Ġauctions": 47842,
+      "usivity": 47843,
+      "ĠCran": 47844,
+      "Ġkd": 47845,
+      "Ġintest": 47846,
+      "rosarcoma": 47847,
+      "ugger": 47848,
+      "ĠILP": 47849,
+      "ĠSTA": 47850,
+      "Ġreversals": 47851,
+      "Ġgrapes": 47852,
+      "ĠPopulus": 47853,
+      "ĠKitaev": 47854,
+      "ĠAVP": 47855,
+      "Previously": 47856,
+      "Ġquadratically": 47857,
+      "ĠLOCAL": 47858,
+      "Bert": 47859,
+      "PED": 47860,
+      "live": 47861,
+      "à¬": 47862,
+      "Ġbidding": 47863,
+      "Ġtoss": 47864,
+      "ento": 47865,
+      "Ġthylak": 47866,
+      "Ġcomprehend": 47867,
+      "Ġdive": 47868,
+      "Ġapplicants": 47869,
+      "ĠÄħ": 47870,
+      "ĠVolcanic": 47871,
+      "adaptation": 47872,
+      "Ġá¹Ģ": 47873,
+      "ĠJanssen": 47874,
+      "Ġadjoining": 47875,
+      "ozolomide": 47876,
+      "CIS": 47877,
+      "dC": 47878,
+      "ducted": 47879,
+      "ĠAnast": 47880,
+      "ĠEmployment": 47881,
+      "ĠEndocrine": 47882,
+      "siloxane": 47883,
+      "Session": 47884,
+      "ĠNarr": 47885,
+      "ĠâĪĴâĪĨ": 47886,
+      "deev": 47887,
+      "othiaz": 47888,
+      "ringing": 47889,
+      "pointed": 47890,
+      "Ġacetylene": 47891,
+      "Ġglobulin": 47892,
+      "packing": 47893,
+      "ĠUses": 47894,
+      "AES": 47895,
+      "Hen": 47896,
+      "ĠSavage": 47897,
+      "ĠCanc": 47898,
+      "isto": 47899,
+      "ĠChromosomal": 47900,
+      "Ġcemented": 47901,
+      "Ġpyrox": 47902,
+      "ĠConstitutive": 47903,
+      "Ġphthalate": 47904,
+      "mechanism": 47905,
+      "Ġcyclosporine": 47906,
+      "PAP": 47907,
+      "arted": 47908,
+      "ĠRDT": 47909,
+      "Ġplains": 47910,
+      "Clone": 47911,
+      "propanol": 47912,
+      "regularity": 47913,
+      "Ġcotangent": 47914,
+      "ĠLeslie": 47915,
+      "ĠNitrate": 47916,
+      "ĠKawasaki": 47917,
+      "ĠPageRank": 47918,
+      "Ġanhydrase": 47919,
+      "ĠKrishna": 47920,
+      "Ġhemicellulose": 47921,
+      "Ġery": 47922,
+      "llis": 47923,
+      "Ġmicrogram": 47924,
+      "ĠDeligne": 47925,
+      "Ġenforces": 47926,
+      "Ġthrombolysis": 47927,
+      "Parse": 47928,
+      "orvastatin": 47929,
+      "Ġmated": 47930,
+      "ĠCrystalline": 47931,
+      "Ġautoradi": 47932,
+      "Ġthermophilic": 47933,
+      "infectious": 47934,
+      "Ġultram": 47935,
+      "ĠMLL": 47936,
+      "ĠFibers": 47937,
+      "Ġulceration": 47938,
+      "omedial": 47939,
+      "stratigraphic": 47940,
+      "Ġtouches": 47941,
+      "rhe": 47942,
+      "Ġtame": 47943,
+      "ĠCulic": 47944,
+      "ARDS": 47945,
+      "chter": 47946,
+      "Ġcounterclockwise": 47947,
+      "Ġcamps": 47948,
+      "VDC": 47949,
+      "Ġmethadone": 47950,
+      "dependently": 47951,
+      "validate": 47952,
+      "Ġprecludes": 47953,
+      "Ġparliamentary": 47954,
+      "ĠINTEREST": 47955,
+      "ĠSerg": 47956,
+      "ĠCBC": 47957,
+      "erella": 47958,
+      "ayi": 47959,
+      "ĠRAB": 47960,
+      "Ġchym": 47961,
+      "Ġnanospheres": 47962,
+      "Ġdiabetics": 47963,
+      "conservation": 47964,
+      "Ġpermeate": 47965,
+      "plotted": 47966,
+      "Ġnaphthalene": 47967,
+      "ĠBonn": 47968,
+      "ĠElectrostatic": 47969,
+      "Ġinventories": 47970,
+      "Gaussianity": 47971,
+      "ĠAdenosine": 47972,
+      "Delay": 47973,
+      "ĠBeginning": 47974,
+      "Ġsided": 47975,
+      "ĠCushing": 47976,
+      "ĠHv": 47977,
+      "Ġcoined": 47978,
+      "ĠAlm": 47979,
+      "scanning": 47980,
+      "fertil": 47981,
+      "Ġαv": 47982,
+      "ĠReactivity": 47983,
+      "Ġproximate": 47984,
+      "dependencies": 47985,
+      "Ġdensification": 47986,
+      "Ġôı¼ij": 47987,
+      "Ġbacteriocin": 47988,
+      "weakly": 47989,
+      "Ġdentistry": 47990,
+      "ĠOriental": 47991,
+      "Ġdormant": 47992,
+      "ĠpC": 47993,
+      "Ġmum": 47994,
+      "REs": 47995,
+      "Ġconval": 47996,
+      "Ġbiota": 47997,
+      "Ġmultilinear": 47998,
+      "ĠPTFE": 47999,
+      "Ġnarrowband": 48000,
+      "ĠRotational": 48001,
+      "Ġhoneybee": 48002,
+      "ĠChlorophyll": 48003,
+      "Baseline": 48004,
+      "Fern": 48005,
+      "Ġlk": 48006,
+      "ĠMash": 48007,
+      "rived": 48008,
+      "ĠBases": 48009,
+      "ĠDah": 48010,
+      "ĠKui": 48011,
+      "ĠÃĵ": 48012,
+      "ĠRecycl": 48013,
+      "AGN": 48014,
+      "PDE": 48015,
+      "Ġclimatological": 48016,
+      "ĠBasically": 48017,
+      "conserved": 48018,
+      "absorbing": 48019,
+      "ĠKoszul": 48020,
+      "oussines": 48021,
+      "Ġmdx": 48022,
+      "ithymia": 48023,
+      "ĠHinton": 48024,
+      "Ġkh": 48025,
+      "Ġadmittance": 48026,
+      "ĠVy": 48027,
+      "Ġextrema": 48028,
+      "Ġcreftype": 48029,
+      "subst": 48030,
+      "Ġbleomycin": 48031,
+      "LINEAR": 48032,
+      "AQ": 48033,
+      "iom": 48034,
+      "Ġnong": 48035,
+      "opian": 48036,
+      "sein": 48037,
+      "udal": 48038,
+      "Ġearning": 48039,
+      "Ġstandardize": 48040,
+      "ĠParticular": 48041,
+      "Ġwavevector": 48042,
+      "dxdy": 48043,
+      "ĠMacDonald": 48044,
+      "ĠEstuary": 48045,
+      "validated": 48046,
+      "ĠHurst": 48047,
+      "ĠMukherjee": 48048,
+      "Ġbivalves": 48049,
+      "Ġjugular": 48050,
+      "Ub": 48051,
+      "vill": 48052,
+      "enough": 48053,
+      "Ġinforms": 48054,
+      "anatomical": 48055,
+      "ulou": 48056,
+      "resa": 48057,
+      "ĠPMC": 48058,
+      "ĠMira": 48059,
+      "ĠRPL": 48060,
+      "ĠSDC": 48061,
+      "Ġhemi": 48062,
+      "MoS": 48063,
+      "ĠFloat": 48064,
+      "Ġocclusal": 48065,
+      "ĠRainbow": 48066,
+      "ĠProviding": 48067,
+      "Ġsupercapacitor": 48068,
+      "osf": 48069,
+      "ĠIRT": 48070,
+      "Ġadm": 48071,
+      "Ġdecoders": 48072,
+      "ĠXR": 48073,
+      "ĠRescue": 48074,
+      "Ġentom": 48075,
+      "Ġmortal": 48076,
+      "Angle": 48077,
+      "India": 48078,
+      "ĠMali": 48079,
+      "Ġinspecting": 48080,
+      "ĠGALAXY": 48081,
+      "ĠEriks": 48082,
+      "YF": 48083,
+      "rings": 48084,
+      "Ġsir": 48085,
+      "Ġgsl": 48086,
+      "ĠBubble": 48087,
+      "ĠDCA": 48088,
+      "ĠWidespread": 48089,
+      "assignment": 48090,
+      "Ġgeomorph": 48091,
+      "ĠPreference": 48092,
+      "COPD": 48093,
+      "processors": 48094,
+      "cutoff": 48095,
+      "ĠFlower": 48096,
+      "phenomen": 48097,
+      "music": 48098,
+      "ĠSlovakia": 48099,
+      "Supporting": 48100,
+      "blow": 48101,
+      "edit": 48102,
+      "ĠTrophy": 48103,
+      "ĠASF": 48104,
+      "ĠMoses": 48105,
+      "Ġindels": 48106,
+      "Ġnonhuman": 48107,
+      "Ġhandic": 48108,
+      "Ġrepairing": 48109,
+      "Ġmicrometer": 48110,
+      "ĠPhilippe": 48111,
+      "Ġexudates": 48112,
+      "ĠâĹĭ": 48113,
+      "Ġamalgam": 48114,
+      "Kin": 48115,
+      "fors": 48116,
+      "fron": 48117,
+      "Ġanabolic": 48118,
+      "ĠEich": 48119,
+      "NAN": 48120,
+      "Ġpseudogap": 48121,
+      "analyzed": 48122,
+      "Ġtackled": 48123,
+      "aginous": 48124,
+      "Ġlubricant": 48125,
+      "Ġradionuclides": 48126,
+      "arrestin": 48127,
+      "oussinesq": 48128,
+      "Lif": 48129,
+      "Î¥": 48130,
+      "received": 48131,
+      "astive": 48132,
+      "ĠPBC": 48133,
+      "Ġamoxicillin": 48134,
+      "copper": 48135,
+      "ubling": 48136,
+      "ophages": 48137,
+      "ĠSeas": 48138,
+      "ĠElite": 48139,
+      "PMMA": 48140,
+      "Ġcholang": 48141,
+      "Depending": 48142,
+      "Ġasbestos": 48143,
+      "ĠFecal": 48144,
+      "ĠRath": 48145,
+      "ĠLey": 48146,
+      "Ġfactored": 48147,
+      "bbles": 48148,
+      "Ġtokenizer": 48149,
+      "Ġofficinalis": 48150,
+      "ĠNUCLE": 48151,
+      "ĠSemicon": 48152,
+      "ĠBous": 48153,
+      "ĠRis": 48154,
+      "Ġloans": 48155,
+      "ACP": 48156,
+      "âĻĢ": 48157,
+      "phosate": 48158,
+      "Ġcherry": 48159,
+      "anan": 48160,
+      "arre": 48161,
+      "ĠCredit": 48162,
+      "isexual": 48163,
+      "ĠActa": 48164,
+      "ĠLetting": 48165,
+      "ĠInfarction": 48166,
+      "ĠAccounting": 48167,
+      "Ġcounterstained": 48168,
+      "Ġaerogel": 48169,
+      "standardized": 48170,
+      "Ġlyase": 48171,
+      "segments": 48172,
+      "Ġbachelor": 48173,
+      "Ġhue": 48174,
+      "ĠNETs": 48175,
+      "Ġunadjusted": 48176,
+      "Ġmicrohardness": 48177,
+      "Ġsinglets": 48178,
+      "ĠSPACE": 48179,
+      "ĠHydraulic": 48180,
+      "METHOD": 48181,
+      "ĠBjör": 48182,
+      "ĠKU": 48183,
+      "Ġrepur": 48184,
+      "Ġradiocarbon": 48185,
+      "Ġheterogeneities": 48186,
+      "Ġgastrocnemius": 48187,
+      "ĠLTD": 48188,
+      "Ġaccidentally": 48189,
+      "Processing": 48190,
+      "Doppler": 48191,
+      "TBI": 48192,
+      "Ġlingual": 48193,
+      "ĠAGS": 48194,
+      "ĠFrontal": 48195,
+      "ĠBrack": 48196,
+      "thema": 48197,
+      "Ġrepresentable": 48198,
+      "Ġpressurized": 48199,
+      "ADR": 48200,
+      "ĠMicrofluid": 48201,
+      "Ġê°": 48202,
+      "Ġreusable": 48203,
+      "Ġvendor": 48204,
+      "aller": 48205,
+      "Ġdiversion": 48206,
+      "FAST": 48207,
+      "ĠKirby": 48208,
+      "ĠStimulus": 48209,
+      "Ġattachments": 48210,
+      "ĠBridging": 48211,
+      "ĠRoberto": 48212,
+      "Ġqueuing": 48213,
+      "tling": 48214,
+      "roots": 48215,
+      "ĠMx": 48216,
+      "ĠMarrow": 48217,
+      "ĠLocus": 48218,
+      "Ġunimportant": 48219,
+      "ergarten": 48220,
+      "ÃŃk": 48221,
+      "ĠPotent": 48222,
+      "ĠBrunswick": 48223,
+      "ĠSCT": 48224,
+      "ĠMour": 48225,
+      "emias": 48226,
+      "ĠNCS": 48227,
+      "chicine": 48228,
+      "ĠOryza": 48229,
+      "Ġwherever": 48230,
+      "ĠXGB": 48231,
+      "COX": 48232,
+      "Ġhydrogenated": 48233,
+      "Ġhydraz": 48234,
+      "ĠPersons": 48235,
+      "Ġframeshift": 48236,
+      "Ġelectrolytic": 48237,
+      "ĠSenegal": 48238,
+      "Ġphagocyt": 48239,
+      "Ġinstantaneously": 48240,
+      "ĠGroundwater": 48241,
+      "Ġimperial": 48242,
+      "ĠRhode": 48243,
+      "ÅĦska": 48244,
+      "ovisual": 48245,
+      "ontsize": 48246,
+      "ĠExplanation": 48247,
+      "Ġempowerment": 48248,
+      "NTA": 48249,
+      "Pu": 48250,
+      "Por": 48251,
+      "Sched": 48252,
+      "eats": 48253,
+      "Ġys": 48254,
+      "inous": 48255,
+      "Ġwilt": 48256,
+      "ĠMov": 48257,
+      "ecton": 48258,
+      "ĠGins": 48259,
+      "introduction": 48260,
+      "inception": 48261,
+      "ĠInterpreting": 48262,
+      "Ġstartup": 48263,
+      "Ġalbino": 48264,
+      "Ġtetras": 48265,
+      "ĠHousehold": 48266,
+      "ĠELM": 48267,
+      "Ġsporulation": 48268,
+      "Ġosmol": 48269,
+      "Bis": 48270,
+      "erule": 48271,
+      "ĠEAR": 48272,
+      "Ġimbalances": 48273,
+      "Ġkt": 48274,
+      "Ġjl": 48275,
+      "gesterone": 48276,
+      "erala": 48277,
+      "ĠPointer": 48278,
+      "ĠHRQoL": 48279,
+      "ĠRiet": 48280,
+      "ĠEscape": 48281,
+      "purified": 48282,
+      "Ġinstantiation": 48283,
+      "matis": 48284,
+      "iona": 48285,
+      "Ġnoxious": 48286,
+      "ĠNog": 48287,
+      "Ġjam": 48288,
+      "ĠAntoni": 48289,
+      "ĠGodd": 48290,
+      "ĠPersonalized": 48291,
+      "Ġpermuted": 48292,
+      "ĠSHE": 48293,
+      "ĠOblast": 48294,
+      "ĠForbes": 48295,
+      "ĠResveratrol": 48296,
+      "ĠFeSe": 48297,
+      "Ġelectrodeposition": 48298,
+      "Ġhomeobox": 48299,
+      "Ġpyogenes": 48300,
+      "Ġviolin": 48301,
+      "Ġisoelectric": 48302,
+      "ĠPPG": 48303,
+      "probably": 48304,
+      "AMPK": 48305,
+      "ĠWolfe": 48306,
+      "Ġultrafine": 48307,
+      "Beyond": 48308,
+      "onat": 48309,
+      "edian": 48310,
+      "ENABLE": 48311,
+      "ĠHAM": 48312,
+      "sout": 48313,
+      "ĠOpinion": 48314,
+      "rinted": 48315,
+      "typing": 48316,
+      "Unknown": 48317,
+      "Ġbuckets": 48318,
+      "Ġintuitionistic": 48319,
+      "algorithms": 48320,
+      "SSC": 48321,
+      "bir": 48322,
+      "ĠPond": 48323,
+      "advert": 48324,
+      "ipin": 48325,
+      "Ġupwind": 48326,
+      "ĠClaire": 48327,
+      "ĠMaturation": 48328,
+      "ĠPrP": 48329,
+      "OPO": 48330,
+      "FORMANCE": 48331,
+      "ĠdM": 48332,
+      "ĠCities": 48333,
+      "Ġinterrelated": 48334,
+      "ĠApparatus": 48335,
+      "Ġprecious": 48336,
+      "criptors": 48337,
+      "Ġpreparedness": 48338,
+      "ĠARCH": 48339,
+      "ĠPathogens": 48340,
+      "HOST": 48341,
+      "ĠGibbons": 48342,
+      "Ġirregularity": 48343,
+      "ĠLipids": 48344,
+      "Ġcfu": 48345,
+      "Ġvasodilation": 48346,
+      "imetre": 48347,
+      "improved": 48348,
+      "mq": 48349,
+      "ĠHens": 48350,
+      "ĠLoci": 48351,
+      "uncredited": 48352,
+      "Ġmultigrid": 48353,
+      "tigo": 48354,
+      "Ġaccountability": 48355,
+      "enchyme": 48356,
+      "Ġdisadvantaged": 48357,
+      "Ġbisphenol": 48358,
+      "Ġtic": 48359,
+      "Ġforks": 48360,
+      "ĠWester": 48361,
+      "ĠVii": 48362,
+      "ĠJere": 48363,
+      "simultaneous": 48364,
+      "ĠGuarant": 48365,
+      "ĠDoyle": 48366,
+      "Ġpotentiates": 48367,
+      "lassified": 48368,
+      "Ġileal": 48369,
+      "Ġvasoconstriction": 48370,
+      "MODULE": 48371,
+      "Nano": 48372,
+      "Wood": 48373,
+      "ĠTAT": 48374,
+      "urious": 48375,
+      "unya": 48376,
+      "Ġinstillation": 48377,
+      "ĠSimmons": 48378,
+      "ĠDirectional": 48379,
+      "Ġmalate": 48380,
+      "Ġplantation": 48381,
+      "Ġunsolved": 48382,
+      "ĠTauri": 48383,
+      "Ġovine": 48384,
+      "Ġkeratinocyte": 48385,
+      "ĠKullback": 48386,
+      "ĠKazakhstan": 48387,
+      "Ġhirs": 48388,
+      "ĠAerobic": 48389,
+      "ĠHai": 48390,
+      "ĠRiley": 48391,
+      "ensible": 48392,
+      "Ġinterplanetary": 48393,
+      "Ġtransits": 48394,
+      "Ġgenerous": 48395,
+      "Ġcalpain": 48396,
+      "Ġappended": 48397,
+      "ĠHydrodynamics": 48398,
+      "Ġcolonize": 48399,
+      "Ġheartbeat": 48400,
+      "Ġmetastas": 48401,
+      "Ġpyreth": 48402,
+      "ĠPAK": 48403,
+      "ĠС": 48404,
+      "multiplet": 48405,
+      "ĠBrady": 48406,
+      "Ġpropria": 48407,
+      "ĠFrontier": 48408,
+      "ĠJoyce": 48409,
+      "ĠPGF": 48410,
+      "ĠMcl": 48411,
+      "recurrent": 48412,
+      "ĠReplacing": 48413,
+      "inference": 48414,
+      "ĠWhitt": 48415,
+      "Ġschooling": 48416,
+      "ĠHarold": 48417,
+      "Ġabstractions": 48418,
+      "âĬķ": 48419,
+      "memcpy": 48420,
+      "Ġmicronucle": 48421,
+      "Ġradionuclide": 48422,
+      "otyl": 48423,
+      "ĠMIF": 48424,
+      "ĠMUS": 48425,
+      "Ġexfoli": 48426,
+      "ĠFamilial": 48427,
+      "Ġclam": 48428,
+      "ONO": 48429,
+      "Ġvanilla": 48430,
+      "Ġpastoris": 48431,
+      "ĠATL": 48432,
+      "ĠBursts": 48433,
+      "Quantitative": 48434,
+      "Ġeliciting": 48435,
+      "Ġgranulomatous": 48436,
+      "Ġbrowsing": 48437,
+      "tracks": 48438,
+      "Ġhij": 48439,
+      "ĠBCP": 48440,
+      "incomp": 48441,
+      "azid": 48442,
+      "ckpt": 48443,
+      "Ġlinkers": 48444,
+      "Ġsquid": 48445,
+      "Ġheadaches": 48446,
+      "ĠMoral": 48447,
+      "Ġstabilisation": 48448,
+      "&&&&": 48449,
+      "ĠSufficient": 48450,
+      "ĠArchaea": 48451,
+      "Ġìł": 48452,
+      "ĠLuciferase": 48453,
+      "Camera": 48454,
+      "expanded": 48455,
+      "Ġmysterious": 48456,
+      "HPS": 48457,
+      "ĠBJ": 48458,
+      "ĠKNN": 48459,
+      "Ġsuperhydrophobic": 48460,
+      "ĠHydrothermal": 48461,
+      "ĠRusso": 48462,
+      "ĠArsenic": 48463,
+      "Ġnormotensive": 48464,
+      "ultimate": 48465,
+      "ĠCMIP": 48466,
+      "examined": 48467,
+      "Ġmicroporous": 48468,
+      "Ġforever": 48469,
+      "ĠSTING": 48470,
+      "IGS": 48471,
+      "ĉĉĉĠĠ": 48472,
+      "Plant": 48473,
+      "Ġcoherently": 48474,
+      "charging": 48475,
+      "Ġinherit": 48476,
+      "alternative": 48477,
+      "ĠBaptist": 48478,
+      "Fm": 48479,
+      "bipy": 48480,
+      "Ġoler": 48481,
+      "ĠSubstit": 48482,
+      "Ġultrap": 48483,
+      "freeze": 48484,
+      "pergill": 48485,
+      "POSE": 48486,
+      "Ġadvertisements": 48487,
+      "ECHAN": 48488,
+      "Bayesian": 48489,
+      "Ġcobordism": 48490,
+      "¸°": 48491,
+      "ĠAER": 48492,
+      "ĠAIP": 48493,
+      "ĠLNA": 48494,
+      "essentially": 48495,
+      "reciprocal": 48496,
+      "ĠAnand": 48497,
+      "Ġsmeared": 48498,
+      "onese": 48499,
+      "ethylamine": 48500,
+      "ĠERS": 48501,
+      "Ġjudicial": 48502,
+      "Ġwoodland": 48503,
+      "ĠGregor": 48504,
+      "Ġtabular": 48505,
+      "avirin": 48506,
+      "mirror": 48507,
+      "Ġjaundice": 48508,
+      "astigotes": 48509,
+      "ĠLGBT": 48510,
+      "ĠNaj": 48511,
+      "Ġsubscheme": 48512,
+      "Ġmultiuser": 48513,
+      "Ġdrains": 48514,
+      "Ġevacuated": 48515,
+      "phosphoryl": 48516,
+      "ĠFeldman": 48517,
+      "ĠTRIzol": 48518,
+      "ĠBLEU": 48519,
+      "aromatic": 48520,
+      "oviÄĩ": 48521,
+      "pion": 48522,
+      "repr": 48523,
+      "roth": 48524,
+      "ĠFES": 48525,
+      "ĠLeeds": 48526,
+      "Ġung": 48527,
+      "obranch": 48528,
+      "Ġpatency": 48529,
+      "ĠScr": 48530,
+      "ĠSimplex": 48531,
+      "pecies": 48532,
+      "Ġbenefici": 48533,
+      "Ġpolymerases": 48534,
+      "ĠCygn": 48535,
+      "octadec": 48536,
+      "Ġpunctured": 48537,
+      "Ġjaponicus": 48538,
+      "ĠFPGAs": 48539,
+      "frown": 48540,
+      "Ġeb": 48541,
+      "utiny": 48542,
+      "ĠPoy": 48543,
+      "ĠBrent": 48544,
+      "ĠBAM": 48545,
+      "ĠHick": 48546,
+      "ĠNPS": 48547,
+      "ĠGDF": 48548,
+      "ĠVIRT": 48549,
+      "Ġinterl": 48550,
+      "ĠscFv": 48551,
+      "Ġteamm": 48552,
+      "Ġparticipatory": 48553,
+      "Ġexistential": 48554,
+      "Ġosteomyelitis": 48555,
+      "Ġpneumothorax": 48556,
+      "stdout": 48557,
+      "Ġsingletons": 48558,
+      "hypothesis": 48559,
+      "stratified": 48560,
+      "USD": 48561,
+      "onasal": 48562,
+      "eris": 48563,
+      "imits": 48564,
+      "ĠICs": 48565,
+      "ĠEncephal": 48566,
+      "izi": 48567,
+      "ĠGradients": 48568,
+      "Ġallop": 48569,
+      "Ġcorp": 48570,
+      "constructed": 48571,
+      "Ġmonument": 48572,
+      "simulator": 48573,
+      "ĠFermions": 48574,
+      "ĠWyoming": 48575,
+      "Ġprednisolone": 48576,
+      "Lang": 48577,
+      "Notes": 48578,
+      "eer": 48579,
+      "Ġfighter": 48580,
+      "entrant": 48581,
+      "ĠNij": 48582,
+      "ĠGPD": 48583,
+      "ĠProl": 48584,
+      "Ġrealisation": 48585,
+      "Ġpackings": 48586,
+      "ĠDiscovering": 48587,
+      "ĠAnglo": 48588,
+      "ĠCassini": 48589,
+      "execute": 48590,
+      "Ġinhabited": 48591,
+      "across": 48592,
+      "ĠCram": 48593,
+      "ĠNBR": 48594,
+      "antes": 48595,
+      "Ġdispersing": 48596,
+      "achandran": 48597,
+      "ĠUND": 48598,
+      "Ġshoulders": 48599,
+      "Ġcrises": 48600,
+      "ustrine": 48601,
+      "Ġpropane": 48602,
+      "UNE": 48603,
+      "brush": 48604,
+      "Ġetiologies": 48605,
+      "Ġshotgun": 48606,
+      "showing": 48607,
+      "ĠPhytochemical": 48608,
+      "ĠMehta": 48609,
+      "orrhea": 48610,
+      "ĠImagery": 48611,
+      "Tre": 48612,
+      "wc": 48613,
+      "Ġeluent": 48614,
+      "ondin": 48615,
+      "ĠAttitude": 48616,
+      "Ġferromagnet": 48617,
+      "Ġcountermeasures": 48618,
+      "Ġalkanes": 48619,
+      "ĠCapillary": 48620,
+      "latent": 48621,
+      "Ġsolubil": 48622,
+      "Viewer": 48623,
+      "ázquez": 48624,
+      "ĠPunjab": 48625,
+      "aas": 48626,
+      "tang": 48627,
+      "Ġimports": 48628,
+      "ĠYounger": 48629,
+      "roughly": 48630,
+      "Weinberg": 48631,
+      "ĠAtkinson": 48632,
+      "bfa": 48633,
+      "MPa": 48634,
+      "steel": 48635,
+      "PCP": 48636,
+      "chlorinated": 48637,
+      "ĠPsychometric": 48638,
+      "Ġpyroptosis": 48639,
+      "Ġwatched": 48640,
+      "ĠPercutaneous": 48641,
+      "RBD": 48642,
+      "VARI": 48643,
+      "atu": 48644,
+      "ĠWake": 48645,
+      "Ġcanyon": 48646,
+      "iparous": 48647,
+      "Ġscall": 48648,
+      "completely": 48649,
+      "interfer": 48650,
+      "ophyceae": 48651,
+      "Ġfatalities": 48652,
+      "czak": 48653,
+      "ĠPathophysiology": 48654,
+      "Lem": 48655,
+      "lach": 48656,
+      "tuary": 48657,
+      "Ġalex": 48658,
+      "Ġsisters": 48659,
+      "Ġpum": 48660,
+      "ĠCatch": 48661,
+      "ĠEber": 48662,
+      "inex": 48663,
+      "phthe": 48664,
+      "Ġboar": 48665,
+      "ĠSoul": 48666,
+      "Ġcatfish": 48667,
+      "Ġcloudy": 48668,
+      "ĠBuilt": 48669,
+      "ophylline": 48670,
+      "ĠRibosome": 48671,
+      "ĠAnomalies": 48672,
+      "YD": 48673,
+      "categorical": 48674,
+      "wor": 48675,
+      "openta": 48676,
+      "ĠLIB": 48677,
+      "Ġrick": 48678,
+      "Ġradiations": 48679,
+      "Ġhypercube": 48680,
+      "Ġmaltreatment": 48681,
+      "ĠîĦĦ": 48682,
+      "dispersity": 48683,
+      "continent": 48684,
+      "Digital": 48685,
+      "ĠCoryneb": 48686,
+      "Ġrevert": 48687,
+      "ĠTEA": 48688,
+      "ĠMLR": 48689,
+      "ĠFCM": 48690,
+      "ĠLamp": 48691,
+      "izabilities": 48692,
+      "Ġcarved": 48693,
+      "ĠMonoclonal": 48694,
+      "Ġpenis": 48695,
+      "ĠMorales": 48696,
+      "Enter": 48697,
+      "esterification": 48698,
+      "Ġcabbage": 48699,
+      "RANTIES": 48700,
+      "Ġdebridement": 48701,
+      "Lead": 48702,
+      "cAMP": 48703,
+      "Ġcesium": 48704,
+      "ĠCubic": 48705,
+      "Ġunimodular": 48706,
+      "ĠExport": 48707,
+      "Ġanalyser": 48708,
+      "denotes": 48709,
+      "Ġradically": 48710,
+      "ĠHistology": 48711,
+      "Ġmelanomas": 48712,
+      "Ġworship": 48713,
+      "ĠHimalayan": 48714,
+      "ĠIntegrable": 48715,
+      "benzenesulfonamide": 48716,
+      "Ġharbored": 48717,
+      "Putting": 48718,
+      "ĠTir": 48719,
+      "ĠUTI": 48720,
+      "centers": 48721,
+      "ĠPluripot": 48722,
+      "Ġharbors": 48723,
+      "Ġcarbam": 48724,
+      "ĠAppalach": 48725,
+      "ĠJoan": 48726,
+      "ĠCommissioner": 48727,
+      "ĠGemini": 48728,
+      "Near": 48729,
+      "OPS": 48730,
+      "QG": 48731,
+      "pytorch": 48732,
+      "staining": 48733,
+      "ĠhCG": 48734,
+      "Ġgavage": 48735,
+      "perhaps": 48736,
+      "ĠGrib": 48737,
+      "ĠZah": 48738,
+      "Ġcomparably": 48739,
+      "ĠBioscience": 48740,
+      "SPL": 48741,
+      "Connell": 48742,
+      "ĠAirway": 48743,
+      "primed": 48744,
+      "Ġsubmucosal": 48745,
+      "Enhanced": 48746,
+      "Ġwisdom": 48747,
+      "VN": 48748,
+      "ĠMumbai": 48749,
+      "rius": 48750,
+      "ĠRGD": 48751,
+      "ĠRNeasy": 48752,
+      "mai": 48753,
+      "ĠADL": 48754,
+      "Ġadoptive": 48755,
+      "Outlined": 48756,
+      "ĠWARRANTIES": 48757,
+      "ĠViolence": 48758,
+      "Ġcaterp": 48759,
+      "Fund": 48760,
+      "dθ": 48761,
+      "ĠPok": 48762,
+      "ĠBenson": 48763,
+      "ĠRIG": 48764,
+      "ĠVs": 48765,
+      "Ġinstants": 48766,
+      "ĠMultidrug": 48767,
+      "PDMS": 48768,
+      "CONST": 48769,
+      "Ġcartridge": 48770,
+      "ĠLifestyle": 48771,
+      "ĠCONDITIONS": 48772,
+      "odysplastic": 48773,
+      "CONTROL": 48774,
+      "LHC": 48775,
+      "tire": 48776,
+      "ĠStain": 48777,
+      "Ġyx": 48778,
+      "Ġjunctional": 48779,
+      "obo": 48780,
+      "annah": 48781,
+      "ĠCPAP": 48782,
+      "Ġsoundness": 48783,
+      "ĠUltimate": 48784,
+      "silicon": 48785,
+      "Ġparalog": 48786,
+      "Events": 48787,
+      "Gas": 48788,
+      "JE": 48789,
+      "ĠJorge": 48790,
+      "Ġoverproduction": 48791,
+      "Ġmaxilla": 48792,
+      "ĠReasons": 48793,
+      "weeks": 48794,
+      "ĠNearest": 48795,
+      "Ġheadspace": 48796,
+      "ĠATC": 48797,
+      "balancing": 48798,
+      "Ġjudging": 48799,
+      "ĠUniversality": 48800,
+      "Ġsinuses": 48801,
+      "Ġretroperitoneal": 48802,
+      "Detection": 48803,
+      "Ġhydrolysate": 48804,
+      "Hoch": 48805,
+      "wrapper": 48806,
+      "ĠpKa": 48807,
+      "Ġlasso": 48808,
+      "ĠAlu": 48809,
+      "ĠAPR": 48810,
+      "ĠDors": 48811,
+      "ĠDarboux": 48812,
+      "ĠRFS": 48813,
+      "ĠKhar": 48814,
+      "ĠThrom": 48815,
+      "Ġdesignate": 48816,
+      "arco": 48817,
+      "Ġthermostat": 48818,
+      "ĠGlacial": 48819,
+      "IFF": 48820,
+      "ĠManifest": 48821,
+      "Ġinterspersed": 48822,
+      "hauser": 48823,
+      "ĠDDX": 48824,
+      "Ġale": 48825,
+      "tides": 48826,
+      "Ġlaccase": 48827,
+      "ĠHered": 48828,
+      "ĠRacial": 48829,
+      "ĠKats": 48830,
+      "ajo": 48831,
+      "ĠCoordinated": 48832,
+      "ĠProbably": 48833,
+      "Ġtitanate": 48834,
+      "SLAM": 48835,
+      "driving": 48836,
+      "ĠEmergent": 48837,
+      "ĠDrives": 48838,
+      "Ġobligations": 48839,
+      "Ġnebulae": 48840,
+      "fried": 48841,
+      "ithin": 48842,
+      "ĠPGD": 48843,
+      "occlusion": 48844,
+      "ĠUH": 48845,
+      "Ġsubroutine": 48846,
+      "oidin": 48847,
+      "Ġannoy": 48848,
+      "ĠVirasoro": 48849,
+      "instances": 48850,
+      "ĠDerby": 48851,
+      "Ġtriangulations": 48852,
+      "Ġcutoffs": 48853,
+      "ĠOrganizational": 48854,
+      "ĠVenk": 48855,
+      "ĠEGTA": 48856,
+      "ĠDeutsche": 48857,
+      "Ġantineut": 48858,
+      "ĠVulnerability": 48859,
+      "iated": 48860,
+      "Ġavium": 48861,
+      "Ġhsp": 48862,
+      "emulsions": 48863,
+      "adherence": 48864,
+      "ĠUPS": 48865,
+      "maph": 48866,
+      "ĠVAP": 48867,
+      "relatively": 48868,
+      "Ġmaxill": 48869,
+      "ophase": 48870,
+      "Threshold": 48871,
+      "ĠSupp": 48872,
+      "ethoprim": 48873,
+      "Ġpenetrated": 48874,
+      "Ġblasting": 48875,
+      "ĠAdvantages": 48876,
+      "BUS": 48877,
+      "olson": 48878,
+      "recv": 48879,
+      "Ġcarnitine": 48880,
+      "tening": 48881,
+      "Ġprovoked": 48882,
+      "various": 48883,
+      "ĠCalab": 48884,
+      "leneck": 48885,
+      "ĠParkin": 48886,
+      "Ġblowup": 48887,
+      "ĠDWI": 48888,
+      "synthesized": 48889,
+      "Ġdisproportionately": 48890,
+      "Ġcardiorespiratory": 48891,
+      "ĠXanthomonas": 48892,
+      "Ġpuncta": 48893,
+      "bddc": 48894,
+      "ĠPACS": 48895,
+      "aseg": 48896,
+      "Ġincurs": 48897,
+      "osta": 48898,
+      "ĠJL": 48899,
+      "ĠWeierstrass": 48900,
+      "oleucine": 48901,
+      "Ġfinals": 48902,
+      "Ġcausation": 48903,
+      "ĠDirective": 48904,
+      "ĠPorto": 48905,
+      "ĠFlores": 48906,
+      "arbonyl": 48907,
+      "----------------------------------------------------------------------------": 48908,
+      "historic": 48909,
+      "Kähler": 48910,
+      "onna": 48911,
+      "Ġcel": 48912,
+      "ĠTBA": 48913,
+      "ĠOphthal": 48914,
+      "Ġsubthreshold": 48915,
+      "Ġlips": 48916,
+      "ĠSubstrates": 48917,
+      "Ġpeninsula": 48918,
+      "Ġadsor": 48919,
+      "Ġdryness": 48920,
+      "masses": 48921,
+      "ème": 48922,
+      "strok": 48923,
+      "ĠExpanded": 48924,
+      "Ġgc": 48925,
+      "ĠGolf": 48926,
+      "Ġcritique": 48927,
+      "ĠÍ©": 48928,
+      "Ġlensed": 48929,
+      "ĠKingma": 48930,
+      "ĠGoldman": 48931,
+      "ĠFacile": 48932,
+      "Carl": 48933,
+      "Ġchondrites": 48934,
+      "ĠCohomology": 48935,
+      "ĠSocioeconomic": 48936,
+      "ĠDominican": 48937,
+      "ĠAzerbaijan": 48938,
+      "ĠAne": 48939,
+      "ĠMidd": 48940,
+      "ĠNed": 48941,
+      "Ġemulate": 48942,
+      "ĠShakes": 48943,
+      "Ġliked": 48944,
+      "Ġbuildup": 48945,
+      "Ġexcessively": 48946,
+      "ĠŶ": 48947,
+      "ĠAdapted": 48948,
+      "Ġauthenticated": 48949,
+      "Ġlocomotive": 48950,
+      "Ġsubmill": 48951,
+      "Ġinterpreter": 48952,
+      "ĠVibrational": 48953,
+      "Rα": 48954,
+      "laden": 48955,
+      "pkl": 48956,
+      "rw": 48957,
+      "yet": 48958,
+      "enzymes": 48959,
+      "Ġwav": 48960,
+      "ĠNMT": 48961,
+      "athion": 48962,
+      "Ġbiotechnological": 48963,
+      "arcs": 48964,
+      "Ġactuated": 48965,
+      "Ġherring": 48966,
+      "ECG": 48967,
+      "OCI": 48968,
+      "Activated": 48969,
+      "Ġparaph": 48970,
+      "Observation": 48971,
+      "ĠEkman": 48972,
+      "ancellor": 48973,
+      "velihood": 48974,
+      "Gauss": 48975,
+      "HAL": 48976,
+      "rdev": 48977,
+      "tbl": 48978,
+      "icits": 48979,
+      "ĠRoux": 48980,
+      "opram": 48981,
+      "Ġseropositive": 48982,
+      "ĠPhysically": 48983,
+      "ĠEdu": 48984,
+      "Ġdocumenting": 48985,
+      "Ġо": 48986,
+      "ĠSmaller": 48987,
+      "chery": 48988,
+      "Ġlanthanide": 48989,
+      "Today": 48990,
+      "ÑĨ": 48991,
+      "Ġotitis": 48992,
+      "ĠCores": 48993,
+      "ifolium": 48994,
+      "ĠZel": 48995,
+      "Ġtimings": 48996,
+      "coarse": 48997,
+      "repair": 48998,
+      "ĠLDPC": 48999,
+      "Ġbowl": 49000,
+      "ĠEpidermal": 49001,
+      "Ġâľ²": 49002,
+      "Ġsynonyms": 49003,
+      "ĠENT": 49004,
+      "Ġbilliard": 49005,
+      "Ġejac": 49006,
+      "ĠBAA": 49007,
+      "Ġscientif": 49008,
+      "Ġγγ": 49009,
+      "Ġcapsular": 49010,
+      "Ġazithromycin": 49011,
+      "Ġcredentials": 49012,
+      "Ġḳ": 49013,
+      "ĠGlioblastoma": 49014,
+      "Ġuncoated": 49015,
+      "Ġhibern": 49016,
+      "ĠTos": 49017,
+      "ĠBaro": 49018,
+      "ĠKass": 49019,
+      "yland": 49020,
+      "ĠXM": 49021,
+      "Ġaggra": 49022,
+      "Ġneutralize": 49023,
+      "licted": 49024,
+      "Ġsoundtrack": 49025,
+      "ĠKnud": 49026,
+      "ensorship": 49027,
+      "empfer": 49028,
+      "ĠHaldane": 49029,
+      "ĠRocks": 49030,
+      "ĠGou": 49031,
+      "ĠOpi": 49032,
+      "ibacterium": 49033,
+      "eptives": 49034,
+      "usta": 49035,
+      "pars": 49036,
+      "ukawa": 49037,
+      "bein": 49038,
+      "elius": 49039,
+      "averaging": 49040,
+      "ĠMWCNT": 49041,
+      "Ġshielded": 49042,
+      "Ġquaternionic": 49043,
+      "BatchNorm": 49044,
+      "Ġdella": 49045,
+      "ĠTp": 49046,
+      "Ġbyproduct": 49047,
+      "ĠGow": 49048,
+      "ĠJO": 49049,
+      "Ġparameterize": 49050,
+      "gler": 49051,
+      "Ġfacult": 49052,
+      "Ġ͵": 49053,
+      "Ġnomination": 49054,
+      "Ġbaths": 49055,
+      "Ġinstallations": 49056,
+      "ĠJustin": 49057,
+      "Ġchampionships": 49058,
+      "tap": 49059,
+      "ĠSanc": 49060,
+      "ĠSusp": 49061,
+      "Ġsubleading": 49062,
+      "Ġdefended": 49063,
+      "Ġbutyl": 49064,
+      "remote": 49065,
+      "Ġcarbides": 49066,
+      "ĠPredicts": 49067,
+      "ĠPriority": 49068,
+      "ĠAntibiotics": 49069,
+      "ĠPUFAs": 49070,
+      "ĠMICs": 49071,
+      "ĠMaximization": 49072,
+      "bare": 49073,
+      "ĠPCN": 49074,
+      "Ġinfested": 49075,
+      "Ġsolenoid": 49076,
+      "Ġagronomic": 49077,
+      "ANGE": 49078,
+      "Rev": 49079,
+      "ĠNKG": 49080,
+      "Ġsaponins": 49081,
+      "Recommend": 49082,
+      "Ġsharpen": 49083,
+      "othioyl": 49084,
+      "suspended": 49085,
+      "atron": 49086,
+      "usage": 49087,
+      "ilter": 49088,
+      "ĠNER": 49089,
+      "CRIPT": 49090,
+      "infections": 49091,
+      "Ġheterosexual": 49092,
+      "Ġmesoc": 49093,
+      "ĠBobby": 49094,
+      "allocate": 49095,
+      "ĠPayne": 49096,
+      "ĠSultan": 49097,
+      "è¡": 49098,
+      "racles": 49099,
+      "ribe": 49100,
+      "ĠDoub": 49101,
+      "ĠPAF": 49102,
+      "communication": 49103,
+      "Ġnineteenth": 49104,
+      "Ġpoplar": 49105,
+      "pgfstrok": 49106,
+      "pgfstrokecolor": 49107,
+      "SLE": 49108,
+      "ecia": 49109,
+      "Ġdetach": 49110,
+      "Ġcharity": 49111,
+      "Ġmonochrom": 49112,
+      "Ġprescribe": 49113,
+      "Ġsupermassive": 49114,
+      "Ġguards": 49115,
+      "Ġcycloaddition": 49116,
+      "Cohen": 49117,
+      "phosphatidyl": 49118,
+      "created": 49119,
+      "ĠElectrodynamics": 49120,
+      "Ġplasmons": 49121,
+      "Ñģк": 49122,
+      "ĠDaphnia": 49123,
+      "easy": 49124,
+      "Ġaq": 49125,
+      "Ġfimb": 49126,
+      "Ġwrest": 49127,
+      "ĠTend": 49128,
+      "hipp": 49129,
+      "Ġorganisational": 49130,
+      "MAE": 49131,
+      "OPY": 49132,
+      "Ġpotentiated": 49133,
+      "Ġbrute": 49134,
+      "omassie": 49135,
+      "aunay": 49136,
+      "luster": 49137,
+      "Ġophi": 49138,
+      "unge": 49139,
+      "ĠPom": 49140,
+      "Ġplague": 49141,
+      "berries": 49142,
+      "Ġinviscid": 49143,
+      "ĠQE": 49144,
+      "incident": 49145,
+      "ximide": 49146,
+      "Ġestrogens": 49147,
+      "ĠTransparent": 49148,
+      "vereign": 49149,
+      "ĠPreferred": 49150,
+      "Ġelastase": 49151,
+      "Civ": 49152,
+      "JF": 49153,
+      "Ku": 49154,
+      "caster": 49155,
+      "Ġspends": 49156,
+      "Ġabstracted": 49157,
+      "otechnical": 49158,
+      "Ġbreeders": 49159,
+      "Ġsyringae": 49160,
+      "Ġclotting": 49161,
+      "African": 49162,
+      "PEC": 49163,
+      "usep": 49164,
+      "Ġstark": 49165,
+      "solete": 49166,
+      "ofovir": 49167,
+      "Ġsensations": 49168,
+      "azawa": 49169,
+      "Ġbiomechanics": 49170,
+      "Ġemergencies": 49171,
+      "Ġspectrometers": 49172,
+      "Ġhemispheric": 49173,
+      "Ġdiscriminatory": 49174,
+      "ĠInspection": 49175,
+      "ndim": 49176,
+      "REP": 49177,
+      "ĠWess": 49178,
+      "Ġhyperalgesia": 49179,
+      "IRC": 49180,
+      "Ġauthorship": 49181,
+      "CPA": 49182,
+      "Ġrotationally": 49183,
+      "ĠPyth": 49184,
+      "ĠYamaguchi": 49185,
+      "Fields": 49186,
+      "Ġenrolment": 49187,
+      "ĠRethinking": 49188,
+      "Gate": 49189,
+      "ìĺ": 49190,
+      "Ġcements": 49191,
+      "ĠTTS": 49192,
+      "ĠFink": 49193,
+      "adus": 49194,
+      "ĠLl": 49195,
+      "Ġimplicate": 49196,
+      "annihilation": 49197,
+      "Ġfeeders": 49198,
+      "ĠPDX": 49199,
+      "ĠFrançois": 49200,
+      "Spearman": 49201,
+      "ĠBenchmarking": 49202,
+      "Forest": 49203,
+      "evidence": 49204,
+      "enoyl": 49205,
+      "olactone": 49206,
+      "cephaly": 49207,
+      "ĠPEA": 49208,
+      "ĠNSE": 49209,
+      "Ġnotified": 49210,
+      "Ġpolyelectrolyte": 49211,
+      "ĠMalik": 49212,
+      "anthine": 49213,
+      "tetrad": 49214,
+      "Ġflagella": 49215,
+      "ãĥ¼": 49216,
+      "orpion": 49217,
+      "Ġbuyers": 49218,
+      "Ġoligodendrocyte": 49219,
+      "ĠNMDAR": 49220,
+      "ĠHarvesting": 49221,
+      "Ġkarst": 49222,
+      "IBD": 49223,
+      "ĠFolk": 49224,
+      "Ġsubcarrier": 49225,
+      "Ġnotices": 49226,
+      "ĠYous": 49227,
+      "awak": 49228,
+      "Ġadversaries": 49229,
+      "Looking": 49230,
+      "Ġthymocytes": 49231,
+      "Ġmeningioma": 49232,
+      "Ġilluminate": 49233,
+      "ĠSPDX": 49234,
+      "Ġimpinging": 49235,
+      "associative": 49236,
+      "Ġtiger": 49237,
+      "leon": 49238,
+      "Ġstature": 49239,
+      "ĠHood": 49240,
+      "ĠRutherford": 49241,
+      "ĠEIT": 49242,
+      "Ġinfantile": 49243,
+      "ĠQubit": 49244,
+      "Ġpacks": 49245,
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ": 49246,
+      "azolam": 49247,
+      "า": 49248,
+      "ĠTunisia": 49249,
+      "dilution": 49250,
+      "Ġsympatric": 49251,
+      "Ġliquefaction": 49252,
+      "porphyrin": 49253,
+      "Gn": 49254,
+      "Rib": 49255,
+      "isothermal": 49256,
+      "apo": 49257,
+      "Ġregressors": 49258,
+      "mani": 49259,
+      "ĠILs": 49260,
+      "oxidants": 49261,
+      "Atom": 49262,
+      "ligo": 49263,
+      "ĠSRAM": 49264,
+      "alcone": 49265,
+      "csr": 49266,
+      "Ġcautious": 49267,
+      "Ġhandlers": 49268,
+      "Ġgastritis": 49269,
+      "ĠSupervision": 49270,
+      "Ġevaporative": 49271,
+      "RUN": 49272,
+      "ĠICG": 49273,
+      "ĠPrague": 49274,
+      "ĠMLC": 49275,
+      "ĠMoney": 49276,
+      "ĠRm": 49277,
+      "ĠECS": 49278,
+      "velt": 49279,
+      "ĠVg": 49280,
+      "Ġbiography": 49281,
+      "Ġministry": 49282,
+      "convolution": 49283,
+      "ogenomics": 49284,
+      "rounding": 49285,
+      "ĠPhag": 49286,
+      "Ġaudiences": 49287,
+      "ĠHCWs": 49288,
+      "Ġblastocysts": 49289,
+      "Ġdiagonals": 49290,
+      "Ġprecautions": 49291,
+      "Íĵ": 49292,
+      "ecewise": 49293,
+      "ĠToxin": 49294,
+      "ĠHapp": 49295,
+      "ĠâĢĭ": 49296,
+      "Ġpopulate": 49297,
+      "mmol": 49298,
+      "ĠPRS": 49299,
+      "Ġreinforces": 49300,
+      "ISTIC": 49301,
+      "ozoites": 49302,
+      "arrival": 49303,
+      "Ġpavement": 49304,
+      "REGISTER": 49305,
+      "ĠGases": 49306,
+      "ĠExhib": 49307,
+      "Ġfactorizations": 49308,
+      "Ġmyopia": 49309,
+      "Ġmovable": 49310,
+      "ĠLIMIT": 49311,
+      "Ġsoleus": 49312,
+      "DOUBLE": 49313,
+      "ĠInputs": 49314,
+      "footnotes": 49315,
+      "BITS": 49316,
+      "ĠCyprus": 49317,
+      "reports": 49318,
+      "ĠPAA": 49319,
+      "istal": 49320,
+      "Ġgroupoids": 49321,
+      "orphin": 49322,
+      "ĠCoordinates": 49323,
+      "boro": 49324,
+      "ĠOslo": 49325,
+      "whenever": 49326,
+      "Ġplausibility": 49327,
+      "ĠFoxO": 49328,
+      "ĠIntrusion": 49329,
+      "Ġsimplices": 49330,
+      "ĠFaso": 49331,
+      "Ġpicosecond": 49332,
+      "ĠAnsatz": 49333,
+      "Importantly": 49334,
+      "ĠHutchinson": 49335,
+      "ovani": 49336,
+      "ĠAsymptotics": 49337,
+      "Ġappra": 49338,
+      "ĠExogenous": 49339,
+      "Ġcaptions": 49340,
+      "ĠAcanth": 49341,
+      "Ġillicit": 49342,
+      "ĠBladder": 49343,
+      "Ġboom": 49344,
+      "ĠSalinity": 49345,
+      "Ġmuscul": 49346,
+      "eptidyl": 49347,
+      "Ġavalanches": 49348,
+      "Helper": 49349,
+      "Ġbivalve": 49350,
+      "Ġreimbursement": 49351,
+      "zzo": 49352,
+      "romatosis": 49353,
+      "Ġparacetamol": 49354,
+      "vio": 49355,
+      "Ġvalpro": 49356,
+      "clamation": 49357,
+      "Ġuu": 49358,
+      "ĠSoC": 49359,
+      "ĠGiac": 49360,
+      "Ġlycopene": 49361,
+      "Flags": 49362,
+      "Ġsticking": 49363,
+      "Ġsqueeze": 49364,
+      "synthetic": 49365,
+      "osahexaenoic": 49366,
+      "mobile": 49367,
+      "vect": 49368,
+      "ĠBaryon": 49369,
+      "Ġnef": 49370,
+      "Ġflatter": 49371,
+      "SSI": 49372,
+      "Ġschw": 49373,
+      "ancreas": 49374,
+      "Buf": 49375,
+      "Solid": 49376,
+      "ĠRIPA": 49377,
+      "Square": 49378,
+      "Ġtranscendental": 49379,
+      "Ġcyn": 49380,
+      "Ġmf": 49381,
+      "ĠSew": 49382,
+      "ĠPIT": 49383,
+      "ocs": 49384,
+      "ĠBash": 49385,
+      "Ġsurprised": 49386,
+      "Ġautonomously": 49387,
+      "Ġlocalizing": 49388,
+      "Ġvisitor": 49389,
+      "Ġpersisting": 49390,
+      "Ġlandfill": 49391,
+      "datetime": 49392,
+      "Ġfiref": 49393,
+      "ĠEngineered": 49394,
+      "ĠSnyder": 49395,
+      "ochromes": 49396,
+      "fractionated": 49397,
+      "GPI": 49398,
+      "Ġwoven": 49399,
+      "ĠTMR": 49400,
+      "Ġforgotten": 49401,
+      "ĠMult": 49402,
+      "ĠBipolar": 49403,
+      "ĠHisp": 49404,
+      "opeptides": 49405,
+      "apamil": 49406,
+      "Ġrouted": 49407,
+      "Ġagn": 49408,
+      "Ġdaylight": 49409,
+      "ĠÍĶ": 49410,
+      "BBB": 49411,
+      "ĠMajority": 49412,
+      "Ġconfounded": 49413,
+      "ĠCaroline": 49414,
+      "ĠShimura": 49415,
+      "ruction": 49416,
+      "Ġtympan": 49417,
+      "acio": 49418,
+      "ĠTFE": 49419,
+      "ĠTutorial": 49420,
+      "Ġallyl": 49421,
+      "ĠFrost": 49422,
+      "ĠRPS": 49423,
+      "ĠWah": 49424,
+      "Ġiy": 49425,
+      "Ġsubproblems": 49426,
+      "Ġairfoil": 49427,
+      "Ġembeds": 49428,
+      "ĠMorning": 49429,
+      "Ġminorities": 49430,
+      "ĠMembership": 49431,
+      "Ġquadriceps": 49432,
+      "yly": 49433,
+      "ĠBodies": 49434,
+      "ĠRAND": 49435,
+      "Ġrationally": 49436,
+      "ĠManifold": 49437,
+      "phosphine": 49438,
+      "considering": 49439,
+      "ezvous": 49440,
+      "ĠKnowing": 49441,
+      "Ġtumorigenic": 49442,
+      "Ġilluminating": 49443,
+      "ĠFernandes": 49444,
+      "polynomials": 49445,
+      "ĠBulgarian": 49446,
+      "ĠBhattacharya": 49447,
+      "ospira": 49448,
+      "Api": 49449,
+      "henne": 49450,
+      "Ġmays": 49451,
+      "ĠArgin": 49452,
+      "interpol": 49453,
+      "ĠAndean": 49454,
+      "ĠPDS": 49455,
+      "ĠCNP": 49456,
+      "Ġtransfusions": 49457,
+      "ĠNanom": 49458,
+      "Ġsynergism": 49459,
+      "Ġbentonite": 49460,
+      "Ġgravitons": 49461,
+      "aquette": 49462,
+      "Ġfissure": 49463,
+      "tandem": 49464,
+      "wash": 49465,
+      "ĠEyes": 49466,
+      "Ġdilepton": 49467,
+      "Ġrectified": 49468,
+      "ĠArist": 49469,
+      "iscible": 49470,
+      "Ġirq": 49471,
+      "Ġligaments": 49472,
+      "security": 49473,
+      "Ġvascularization": 49474,
+      "NaCl": 49475,
+      "ĠStraight": 49476,
+      "ĠLeptin": 49477,
+      "ĠAbundances": 49478,
+      "ĠKEY": 49479,
+      "ĠMothers": 49480,
+      "ĠRenewable": 49481,
+      "Ġmasonry": 49482,
+      "ëı": 49483,
+      "raceutical": 49484,
+      "Ġarity": 49485,
+      "ĠAlves": 49486,
+      "ospectral": 49487,
+      "Ġimmunod": 49488,
+      "Ġmarble": 49489,
+      "Ġcoverings": 49490,
+      "ĠConstants": 49491,
+      "ĠReversal": 49492,
+      "Works": 49493,
+      "ĠNurse": 49494,
+      "ĠAggregate": 49495,
+      "acillin": 49496,
+      "plug": 49497,
+      "Ġjury": 49498,
+      "oneogenesis": 49499,
+      "Ġamoeb": 49500,
+      "aukee": 49501,
+      "Ġphosphoric": 49502,
+      "ĠRemoving": 49503,
+      "Ġworsen": 49504,
+      "ĠESRD": 49505,
+      "ĠHernandez": 49506,
+      "ĠEugene": 49507,
+      "viscosity": 49508,
+      "FID": 49509,
+      "¦": 49510,
+      "ĠÝ": 49511,
+      "ĠStig": 49512,
+      "ĠSING": 49513,
+      "ĠIMRT": 49514,
+      "ĠBq": 49515,
+      "ĠDME": 49516,
+      "ĠHOM": 49517,
+      "physis": 49518,
+      "obes": 49519,
+      "Ġsuperfields": 49520,
+      "Ġargc": 49521,
+      "Ġmaladaptive": 49522,
+      "ĠEditing": 49523,
+      "Ġcondem": 49524,
+      "ubei": 49525,
+      "stimulus": 49526,
+      "Ġabbreviation": 49527,
+      "Haus": 49528,
+      "ĠNeeds": 49529,
+      "Ġadhering": 49530,
+      "ĠVPA": 49531,
+      "ofrontal": 49532,
+      "ĠŪ": 49533,
+      "ĠVerde": 49534,
+      "ĠSlav": 49535,
+      "ĠPropag": 49536,
+      "Ġcongeners": 49537,
+      "Ġtilapia": 49538,
+      "ĠRachel": 49539,
+      "Less": 49540,
+      "Ġmasc": 49541,
+      "entangled": 49542,
+      "ĠDTI": 49543,
+      "atik": 49544,
+      "rolases": 49545,
+      "ĠYen": 49546,
+      "armor": 49547,
+      "ĠDecisions": 49548,
+      "Ġηp": 49549,
+      "Intuitively": 49550,
+      "ĠPharmaceuticals": 49551,
+      "Ju": 49552,
+      "uddin": 49553,
+      "ĠWASP": 49554,
+      "nton": 49555,
+      "Ġbiot": 49556,
+      "ĠZNF": 49557,
+      "Ġcrush": 49558,
+      "ĠParity": 49559,
+      "SIST": 49560,
+      "EVENT": 49561,
+      "ĠSquamous": 49562,
+      "Student": 49563,
+      "Ġgingivalis": 49564,
+      "fused": 49565,
+      "ĠMises": 49566,
+      "ĠFDTD": 49567,
+      "oreceptors": 49568,
+      "Ġdiscretion": 49569,
+      "ORTC": 49570,
+      "MSP": 49571,
+      "Ġexposes": 49572,
+      "Ġchlorinated": 49573,
+      "ĠUpregulation": 49574,
+      "ĠLimb": 49575,
+      "Ġanticor": 49576,
+      "Regular": 49577,
+      "Advanced": 49578,
+      "Xe": 49579,
+      "aghan": 49580,
+      "ĠBLA": 49581,
+      "Ġcoasts": 49582,
+      "ĠThirteen": 49583,
+      "hesin": 49584,
+      "ĠMethanol": 49585,
+      "rotus": 49586,
+      "ĠStephens": 49587,
+      "Book": 49588,
+      "ĠHistorically": 49589,
+      "ĠEmploying": 49590,
+      "Ġcorrugated": 49591,
+      "mercaptoethanol": 49592,
+      "ĠDnmt": 49593,
+      "ĠQueries": 49594,
+      "DRB": 49595,
+      "ĠGRU": 49596,
+      "FLU": 49597,
+      "ĠCarboniferous": 49598,
+      "OBJECT": 49599,
+      "ĠDiscriminative": 49600,
+      "ĠBurgess": 49601,
+      "Ġplanetesimals": 49602,
+      "ĠAmendment": 49603,
+      "ĠStrikingly": 49604,
+      "tric": 49605,
+      "ecure": 49606,
+      "Ġtransposable": 49607,
+      "rolateral": 49608,
+      "Ġhisti": 49609,
+      "egaard": 49610,
+      "Ġskim": 49611,
+      "ĠSPF": 49612,
+      "Statistics": 49613,
+      "Ġintestines": 49614,
+      "feng": 49615,
+      "lain": 49616,
+      "Ġtheat": 49617,
+      "Ġorogen": 49618,
+      "Ġpill": 49619,
+      "odopa": 49620,
+      "Ġcorrelative": 49621,
+      "ACO": 49622,
+      "Ġadjunction": 49623,
+      "ĠCarey": 49624,
+      "Ġteleportation": 49625,
+      "ĠBoundaries": 49626,
+      "ĠGoodfellow": 49627,
+      "ĠLinda": 49628,
+      "ĠPolymeric": 49629,
+      "Ġexertion": 49630,
+      "Ġsteeply": 49631,
+      "Ġprotrusion": 49632,
+      "Ġhyaluronic": 49633,
+      "ĠRochester": 49634,
+      "ENSIONAL": 49635,
+      "Dar": 49636,
+      "fet": 49637,
+      "ĠFSS": 49638,
+      "hemically": 49639,
+      "Ġflashes": 49640,
+      "Ġdeviated": 49641,
+      "feldt": 49642,
+      "Ġsticks": 49643,
+      "Ġoctet": 49644,
+      "Ġgravitationally": 49645,
+      "footnotesize": 49646,
+      "Lex": 49647,
+      "ovi": 49648,
+      "Ġwired": 49649,
+      "ĠSMP": 49650,
+      "ermans": 49651,
+      "Ġunbroken": 49652,
+      "Ġemulation": 49653,
+      "simulated": 49654,
+      "Ġminimality": 49655,
+      "ardiac": 49656,
+      "Ġshipw": 49657,
+      "Genetic": 49658,
+      "ĠHermann": 49659,
+      "ynchronization": 49660,
+      "ĠNapole": 49661,
+      "Ġmonodisperse": 49662,
+      "Rho": 49663,
+      "rists": 49664,
+      "Ġfx": 49665,
+      "ĠFUV": 49666,
+      "ĠGelfand": 49667,
+      "hemispheric": 49668,
+      "ronidazole": 49669,
+      "Ġsupersaturation": 49670,
+      "oudh": 49671,
+      "olitical": 49672,
+      "ĠAiry": 49673,
+      "Ġmanifestly": 49674,
+      "ĠHMG": 49675,
+      "Ġadvertisement": 49676,
+      "ĠBrooklyn": 49677,
+      "Ġparalleled": 49678,
+      "answered": 49679,
+      "ĠNotImplementedError": 49680,
+      "UPD": 49681,
+      "oust": 49682,
+      "ĠTeng": 49683,
+      "Ġfortified": 49684,
+      "ĠSort": 49685,
+      "ENE": 49686,
+      "ĠFris": 49687,
+      "ĠHIS": 49688,
+      "ĠROT": 49689,
+      "ĠNested": 49690,
+      "produce": 49691,
+      "ĠKerala": 49692,
+      "genomic": 49693,
+      "ĠIsab": 49694,
+      "Ġuracil": 49695,
+      "burger": 49696,
+      "ĠLogarithmic": 49697,
+      "Ġsterility": 49698,
+      "Ġunemployed": 49699,
+      "Ġoriental": 49700,
+      "Ko": 49701,
+      "jima": 49702,
+      "ĠCTP": 49703,
+      "ĠLAD": 49704,
+      "Ġconformers": 49705,
+      "CGG": 49706,
+      "Perkin": 49707,
+      "Ġbridged": 49708,
+      "ĠDissociation": 49709,
+      "ĠQiagen": 49710,
+      "Ġwealthy": 49711,
+      "Ġanaesthetic": 49712,
+      "ĠMinimizing": 49713,
+      "Ġacoustics": 49714,
+      "bucket": 49715,
+      "ĠSertoli": 49716,
+      "rath": 49717,
+      "saw": 49718,
+      "Ġgarn": 49719,
+      "ĠTheoretically": 49720,
+      "ticul": 49721,
+      "ĠThinking": 49722,
+      "iker": 49723,
+      "ĠChit": 49724,
+      "Ġtrin": 49725,
+      "ALITY": 49726,
+      "ĠFeO": 49727,
+      "Ġpolymerized": 49728,
+      "Encoding": 49729,
+      "Ġanalgesics": 49730,
+      "ĠLexical": 49731,
+      "Ġmarijuana": 49732,
+      "âĸĪâĸĪ": 49733,
+      "crops": 49734,
+      "entropic": 49735,
+      "olocation": 49736,
+      "ĠPomp": 49737,
+      "Ġcofactors": 49738,
+      "boxtimes": 49739,
+      "ĠArri": 49740,
+      "Angel": 49741,
+      "ĠRequirement": 49742,
+      "Ġmicrolensing": 49743,
+      "ĠTRANSF": 49744,
+      "åº": 49745,
+      "Ġdma": 49746,
+      "Ġrerio": 49747,
+      "undancy": 49748,
+      "antel": 49749,
+      "Ġradiometric": 49750,
+      "ĠSean": 49751,
+      "randn": 49752,
+      "ĠCRL": 49753,
+      "halos": 49754,
+      "ubertal": 49755,
+      "Ġquinone": 49756,
+      "TES": 49757,
+      "ĠdW": 49758,
+      "ĠCGM": 49759,
+      "Ġhealed": 49760,
+      "iane": 49761,
+      "Ġobtainable": 49762,
+      "ĠAdrian": 49763,
+      "Ġlikes": 49764,
+      "ĠMedication": 49765,
+      "Ġcognitively": 49766,
+      "Whether": 49767,
+      "Bob": 49768,
+      "did": 49769,
+      "alcohol": 49770,
+      "Ġnivolumab": 49771,
+      "ĠFY": 49772,
+      "Ġatresia": 49773,
+      "achs": 49774,
+      "ĠKip": 49775,
+      "Ġunigenes": 49776,
+      "ĠJaccard": 49777,
+      "ustri": 49778,
+      "Ġconfine": 49779,
+      "Ġautofluorescence": 49780,
+      "Ġpyg": 49781,
+      "Sea": 49782,
+      "Settings": 49783,
+      "Ġtruncatula": 49784,
+      "Literal": 49785,
+      "Fab": 49786,
+      "Mah": 49787,
+      "Ven": 49788,
+      "Ġtig": 49789,
+      "Ġcher": 49790,
+      "ĠCCI": 49791,
+      "ĠFunk": 49792,
+      "ĠBess": 49793,
+      "ĠNasal": 49794,
+      "iffer": 49795,
+      "Ġobsessive": 49796,
+      "ĠOpening": 49797,
+      "ochondral": 49798,
+      "ĠTRPA": 49799,
+      "ĠRabin": 49800,
+      "Ġtaper": 49801,
+      "Ġdeafness": 49802,
+      "DOS": 49803,
+      "isites": 49804,
+      "anite": 49805,
+      "leost": 49806,
+      "ĠSTP": 49807,
+      "ĠBACE": 49808,
+      "ĠHenn": 49809,
+      "ĠESM": 49810,
+      "Ġsuperfield": 49811,
+      "ĠOrland": 49812,
+      "ĠAMPs": 49813,
+      "ĠHemorrh": 49814,
+      "Ġrescues": 49815,
+      "Ġtourists": 49816,
+      "ĠVLBI": 49817,
+      "Ġneighbourhoods": 49818,
+      "communicable": 49819,
+      "gx": 49820,
+      "ratase": 49821,
+      "ĠNRT": 49822,
+      "Ġobstructions": 49823,
+      "Ġdeforestation": 49824,
+      "Ġqp": 49825,
+      "ĠPhan": 49826,
+      "ĠSTI": 49827,
+      "imento": 49828,
+      "ĠIRI": 49829,
+      "SVs": 49830,
+      "Ġstriped": 49831,
+      "Poinc": 49832,
+      "ĠBedford": 49833,
+      "ĠFragment": 49834,
+      "ĠReligion": 49835,
+      "Ġdrones": 49836,
+      "imulation": 49837,
+      "ĠCet": 49838,
+      "Ġgills": 49839,
+      "ĠNorton": 49840,
+      "ibatch": 49841,
+      "estructive": 49842,
+      "ĠJav": 49843,
+      "ĠϽ": 49844,
+      "Ġmica": 49845,
+      "AGB": 49846,
+      "RAW": 49847,
+      "ĠMyD": 49848,
+      "ctl": 49849,
+      "Ġreversibly": 49850,
+      "Ġsuppressors": 49851,
+      "ĠFAIL": 49852,
+      "ĠFukushima": 49853,
+      "Evidence": 49854,
+      "pink": 49855,
+      "asarray": 49856,
+      "ĠTann": 49857,
+      "Ġloved": 49858,
+      "Ġbiologists": 49859,
+      "Ġendothermic": 49860,
+      "Ġbroker": 49861,
+      "ĠPerkins": 49862,
+      "Ġcategorised": 49863,
+      "ĠSOME": 49864,
+      "hydroxyvitamin": 49865,
+      "rogates": 49866,
+      "ĠAgeing": 49867,
+      "Ġtournaments": 49868,
+      "ĠStromal": 49869,
+      "Ġdeferred": 49870,
+      "ĠSREBP": 49871,
+      "MAD": 49872,
+      "Say": 49873,
+      "cgi": 49874,
+      "phe": 49875,
+      "olini": 49876,
+      "ĠDü": 49877,
+      "Ġdehydro": 49878,
+      "apeptide": 49879,
+      "Ġhes": 49880,
+      "Ġdistally": 49881,
+      "versions": 49882,
+      "Ġmedals": 49883,
+      "Ġflaw": 49884,
+      "Ġduo": 49885,
+      "Ġimpairing": 49886,
+      "toplasts": 49887,
+      "ĠHFILL": 49888,
+      "Ġesculent": 49889,
+      "Classification": 49890,
+      "ĠGriffith": 49891,
+      "ĠWellington": 49892,
+      "Ġattorney": 49893,
+      "Ast": 49894,
+      "kA": 49895,
+      "Ġmilit": 49896,
+      "Ġnite": 49897,
+      "ĠCasp": 49898,
+      "ĠChester": 49899,
+      "ĠMok": 49900,
+      "ĠRAR": 49901,
+      "Ġchr": 49902,
+      "unctor": 49903,
+      "Ġabduction": 49904,
+      "Ġuniv": 49905,
+      "ovars": 49906,
+      "ouk": 49907,
+      "ERICAL": 49908,
+      "éri": 49909,
+      "orbance": 49910,
+      "ĠIdentifies": 49911,
+      "amento": 49912,
+      "Ġparenthesis": 49913,
+      "ĠMEFs": 49914,
+      "Ġabsorbs": 49915,
+      "ĠArrayList": 49916,
+      "Ġcaregiving": 49917,
+      "ĠFILE": 49918,
+      "Ġfeldspar": 49919,
+      "ochthonous": 49920,
+      "Sort": 49921,
+      "jal": 49922,
+      "Ġtantal": 49923,
+      "arabine": 49924,
+      "ĠSaid": 49925,
+      "ĠBCE": 49926,
+      "ĠNGO": 49927,
+      "ynure": 49928,
+      "doteq": 49929,
+      "ĠLeyd": 49930,
+      "modality": 49931,
+      "ĠGeometrical": 49932,
+      "Almost": 49933,
+      "Ġhardened": 49934,
+      "noea": 49935,
+      "news": 49936,
+      "Ġcleanup": 49937,
+      "ĠArmed": 49938,
+      "ĠSnake": 49939,
+      "multiply": 49940,
+      "ĠMillennium": 49941,
+      "ĠSmoothing": 49942,
+      "positely": 49943,
+      "enary": 49944,
+      "isse": 49945,
+      "ĠYuc": 49946,
+      "Ġgeneal": 49947,
+      "Ġsupers": 49948,
+      "Ġhandheld": 49949,
+      "Ġembark": 49950,
+      "ĠBla": 49951,
+      "horst": 49952,
+      "ĠPDGFR": 49953,
+      "Ġcitr": 49954,
+      "Ġcalorie": 49955,
+      "ĠBuddhist": 49956,
+      "Member": 49957,
+      "Ġfears": 49958,
+      "Ġfiscal": 49959,
+      "ĠAIF": 49960,
+      "LOAD": 49961,
+      "peare": 49962,
+      "Ġbitumen": 49963,
+      "Particip": 49964,
+      "ĠIndianapolis": 49965,
+      "ĠAlbum": 49966,
+      "Ġscrutiny": 49967,
+      "acylglycer": 49968,
+      "ĠSakai": 49969,
+      "Ġthermodynamical": 49970,
+      "ZB": 49971,
+      "Ġhpf": 49972,
+      "ĠLIP": 49973,
+      "Ġexpiration": 49974,
+      "tilt": 49975,
+      "Ġflax": 49976,
+      "ĠSelectivity": 49977,
+      "ĠSchol": 49978,
+      "anya": 49979,
+      "orbents": 49980,
+      "Ġincubations": 49981,
+      "Ġmarginals": 49982,
+      "involved": 49983,
+      "Ġenthalpies": 49984,
+      "macrophages": 49985,
+      "constructor": 49986,
+      "ĠRoland": 49987,
+      "ĠPm": 49988,
+      "ĠRY": 49989,
+      "Ġblobs": 49990,
+      "Ġannuli": 49991,
+      "Ġunstimulated": 49992,
+      "ĠPetroleum": 49993,
+      "Ġmerges": 49994,
+      "Ġenveloping": 49995,
+      "ĠInitialization": 49996,
+      "Ġsheds": 49997,
+      "Ġadvisable": 49998,
+      "ylethanolamine": 49999
+    },
+    "merges": [
+      "Ġ t",
+      "i n",
+      "Ġ a",
+      "h e",
+      "o n",
+      "r e",
+      "a t",
+      "Ġt he",
+      "e r",
+      "Ġ s",
+      "Ġ o",
+      "e n",
+      "a l",
+      "Ġ c",
+      "t i",
+      "o r",
+      "e d",
+      "e s",
+      "i s",
+      "Ġ p",
+      "Ġo f",
+      "n d",
+      "Ġ in",
+      "Ġ f",
+      "Ġ w",
+      "Ġ Ġ",
+      "i t",
+      "a n",
+      "r o",
+      "a r",
+      "Ġ d",
+      "Ġ m",
+      "Ġ b",
+      "Ġa nd",
+      "i c",
+      "l e",
+      "in g",
+      "i on",
+      "a s",
+      "Ġ e",
+      "Ġ re",
+      "at ion",
+      "Ġt o",
+      "e l",
+      "en t",
+      "a c",
+      "e t",
+      "e c",
+      "ti on",
+      "o m",
+      "s t",
+      "Ġ T",
+      "Ġ n",
+      "Ġt h",
+      "o l",
+      "u l",
+      "i m",
+      "R E",
+      "i g",
+      "u s",
+      "RE F",
+      "Ġ l",
+      "Ġ h",
+      "u r",
+      "Ġ is",
+      "ĠĠ ĠĠ",
+      "Ġf or",
+      "i d",
+      "a m",
+      "Ġ S",
+      "v e",
+      "i l",
+      "Ġ A",
+      "Ġ C",
+      "Ġ g",
+      "o t",
+      "it h",
+      "l y",
+      "c e",
+      "Ġc on",
+      "o w",
+      "Ġs t",
+      "u t",
+      "o s",
+      "Ġw ith",
+      "o d",
+      "r a",
+      "Ġ v",
+      "Ġp ro",
+      "u m",
+      "Ġ I",
+      "i f",
+      "u c",
+      "t er",
+      "u n",
+      "A R",
+      "S T",
+      "re s",
+      "Ġ on",
+      "E N",
+      "e re",
+      "Ġ P",
+      "ĠT he",
+      "Ġ M",
+      "Ġa s",
+      "AR T",
+      "Ġa n",
+      "EN D",
+      "ST ART",
+      "Ġth at",
+      "q u",
+      "e m",
+      "Ġb e",
+      "Ġe x",
+      "r i",
+      "a b",
+      "it y",
+      "ti c",
+      "v er",
+      "Ġa l",
+      "p l",
+      "t s",
+      "Ġ F",
+      "Ġ â",
+      "u re",
+      "Ġb y",
+      "at e",
+      "a g",
+      "i r",
+      "o c",
+      "p er",
+      "Ġ B",
+      "a y",
+      "Ġ D",
+      "Ġc om",
+      "Ġ H",
+      "at ed",
+      "Ġ R",
+      "Ġa re",
+      "ro m",
+      "Ġ E",
+      "o p",
+      "a d",
+      "s e",
+      "Ġ L",
+      "ig h",
+      "Ġ N",
+      "m ent",
+      "he r",
+      "o g",
+      "a in",
+      "ec t",
+      "u d",
+      "Ġd e",
+      "Ġ r",
+      "Ġa t",
+      "Ġw as",
+      "Ġ us",
+      "Ġre s",
+      "el l",
+      "i z",
+      "in e",
+      "p h",
+      "Ġa c",
+      "es s",
+      "o re",
+      "ic al",
+      "t h",
+      "u nd",
+      "r ac",
+      "Ġw e",
+      "at h",
+      "Ġ G",
+      "Ġf rom",
+      "at i",
+      "u p",
+      "is t",
+      "an t",
+      "Ġo r",
+      "f f",
+      "Ġcom p",
+      "Ġw h",
+      "Ġ W",
+      "c h",
+      "er s",
+      "Ġs p",
+      "or m",
+      "Ġc h",
+      "ation s",
+      "r an",
+      "u b",
+      "t e",
+      "d i",
+      "Ġs h",
+      "g e",
+      "as e",
+      "Ġw ere",
+      "ĠĠĠĠ ĠĠĠĠ",
+      "Ġ Î",
+      "a p",
+      "ĠI n",
+      "a nd",
+      "Ġs e",
+      "v el",
+      "Ġ im",
+      "Ġâ Ī",
+      "en s",
+      "i es",
+      "ic h",
+      "igh t",
+      "d uc",
+      "Ġ O",
+      "Ġ it",
+      "tion s",
+      "en d",
+      "Ġc o",
+      "Ġth is",
+      "Ġc an",
+      "Ġ k",
+      "â Ģ",
+      "le c",
+      "t ed",
+      "Ġm od",
+      "m ath",
+      "Ġcon t",
+      "Ġn e",
+      "Ġp ar",
+      "i b",
+      "ĠĠ Ġ",
+      "Ġ le",
+      "i v",
+      "u g",
+      "en ce",
+      "ig n",
+      "o us",
+      "ent s",
+      "y s",
+      "a ve",
+      "re d",
+      "res s",
+      "ab le",
+      "p or",
+      "al l",
+      "if f",
+      "es t",
+      "Ġa p",
+      "Ġin c",
+      "n t",
+      "ar y",
+      "i ti",
+      "Ġwh ich",
+      "Ġn ot",
+      "f orm",
+      "Ġs y",
+      "Ġa d",
+      "l ow",
+      "a k",
+      "Ġp er",
+      "Ġ he",
+      "p ro",
+      "an ce",
+      "i al",
+      "u e",
+      "Ġ en",
+      "Ġc l",
+      "as s",
+      "i p",
+      "ran s",
+      "Ġo b",
+      "Ġg en",
+      "ti m",
+      "Ġd is",
+      "un c",
+      "Ġin t",
+      "e p",
+      "et w",
+      "Ġd iff",
+      "ac h",
+      "t her",
+      "im e",
+      "ag e",
+      "p le",
+      "il l",
+      "y p",
+      "Ġ K",
+      "ac t",
+      "ar i",
+      "Ġm et",
+      "or s",
+      "Ġh ave",
+      "Ġst ud",
+      "on g",
+      "Ġ U",
+      "Ġp l",
+      "id e",
+      "m a",
+      "he n",
+      "if ic",
+      "om e",
+      "Ġ i",
+      "ul ar",
+      "Ġ V",
+      "al ly",
+      "Ġsh ow",
+      "ri b",
+      "i a",
+      "en ti",
+      "Ġas s",
+      "on d",
+      "f t",
+      "Ġa b",
+      "Ġin ter",
+      "ĠT h",
+      "T he",
+      "st r",
+      "Ġc ell",
+      "c al",
+      "Ġmod el",
+      "at a",
+      "as t",
+      "Ġe ff",
+      "Ġt rans",
+      "at es",
+      "as ed",
+      "o st",
+      "v i",
+      "an g",
+      "o ur",
+      "Ġm e",
+      "ar d",
+      "Ġdiff ere",
+      "Ġp re",
+      "Ġd i",
+      "ĠâĪ Ĵ",
+      "ol og",
+      "u tion",
+      "o und",
+      "ac e",
+      "Ġres ul",
+      "er m",
+      "p os",
+      "he re",
+      "ti ve",
+      "or d",
+      "s o",
+      "st em",
+      "y l",
+      "Ġp h",
+      "Ġ y",
+      "am e",
+      "or k",
+      "ati ve",
+      "Ġ qu",
+      "r ic",
+      "S U",
+      "w o",
+      "Ġ un",
+      "Ġe v",
+      "a re",
+      "# #",
+      "d e",
+      "e en",
+      "ti v",
+      "Ġg ro",
+      "or y",
+      "Ġcon s",
+      "Ġs ub",
+      "t a",
+      "- -",
+      "Ġst r",
+      "b er",
+      "er v",
+      "etw een",
+      "en c",
+      "Ġan al",
+      "in t",
+      "Ġh as",
+      "uc h",
+      "Ġre g",
+      "Ġb etween",
+      "Ġd et",
+      "Ġal l",
+      "c ess",
+      "Ġex p",
+      "ec tion",
+      "Ġâ Ģ",
+      "in d",
+      "at er",
+      "Ġs ign",
+      "p t",
+      "ug h",
+      "it e",
+      "il ity",
+      "Ġus ing",
+      "Ġv al",
+      "Ġ ro",
+      "re e",
+      "Ġre l",
+      "o ut",
+      "Ġf unc",
+      "i tion",
+      "Ġc or",
+      "Ġal so",
+      "Ġt wo",
+      "n e",
+      "Ġ J",
+      "Ġsy stem",
+      "c l",
+      "uc t",
+      "Ġs im",
+      "t ain",
+      "u st",
+      "i ed",
+      "por t",
+      "Ġre c",
+      "Ġres p",
+      "Ġd ata",
+      "r m",
+      "res ent",
+      "ul d",
+      "x t",
+      "Ġ j",
+      "r y",
+      "ac k",
+      "Ġ ra",
+      "p ar",
+      "Ġfor m",
+      "Ġs c",
+      "f rac",
+      "ĠW e",
+      "at ing",
+      "ec h",
+      "h od",
+      "Ġf ol",
+      "in ed",
+      "ĠS t",
+      "u al",
+      "Ġus ed",
+      "Ġon e",
+      "Ġd es",
+      "Ġ Ï",
+      "Ġv ari",
+      "Ġd ist",
+      "Ġn um",
+      "y m",
+      "e w",
+      "re c",
+      "o b",
+      "Ġin f",
+      "Ġa r",
+      "lec t",
+      "l l",
+      "on s",
+      "ĠTh is",
+      "os e",
+      "i le",
+      "pl ay",
+      "e ar",
+      "o x",
+      "u res",
+      "on e",
+      "Ġstud y",
+      "ys is",
+      "Ġfol low",
+      "y le",
+      "rac t",
+      "d is",
+      "Ġp os",
+      "r ight",
+      "Ġth an",
+      "ro s",
+      "a v",
+      "F ig",
+      "Ġt ime",
+      "iz ation",
+      "ul ation",
+      "iz ed",
+      "Ġs ur",
+      "ot h",
+      "Ġo ut",
+      "Ġc ol",
+      "at ure",
+      "i ve",
+      "Ġs ol",
+      "Ġ x",
+      "el d",
+      "Ġo ther",
+      "pl ic",
+      "Ġde f",
+      "er g",
+      "Ġgen er",
+      "el y",
+      "Ġbe en",
+      "Ġinc re",
+      "Ġthe se",
+      "Ġn o",
+      "a x",
+      "st yle",
+      "ar g",
+      "i an",
+      "Ġin d",
+      "Ġs uch",
+      "Ġfunc tion",
+      "t ing",
+      "Ġe qu",
+      "a us",
+      "Ġ und",
+      "math b",
+      "tic al",
+      "Ġh igh",
+      "ra in",
+      "Ġa m",
+      "i eld",
+      "o un",
+      "ress ion",
+      "Ġsp ec",
+      "Ġo p",
+      "Ġd ec",
+      "Ġo ver",
+      "Ġmet hod",
+      "Ġs et",
+      "â Ī",
+      "Ġ if",
+      "di tion",
+      "u es",
+      "ec ts",
+      "dis play",
+      "he m",
+      "Ġp ati",
+      "Ġresul ts",
+      "ol d",
+      "an c",
+      "display style",
+      "Ġe ach",
+      "Ġm ore",
+      "l es",
+      "p r",
+      "ac ter",
+      "Ġthe ir",
+      "Ġac c",
+      "Ġap pro",
+      "is s",
+      "iz e",
+      "Ġin v",
+      "as es",
+      "Ġcell s",
+      "ir st",
+      "l u",
+      "a il",
+      "Ġme as",
+      "Ġl ow",
+      "o v",
+      "t he",
+      "i k",
+      "* *",
+      "e f",
+      "Ġb ut",
+      "he s",
+      "f ter",
+      "Ġdiffere nt",
+      "vel y",
+      "Ġex t",
+      "Ġthe re",
+      "oc i",
+      "Ġpro b",
+      "Ġit s",
+      "r on",
+      "ment s",
+      "Ġa g",
+      "N A",
+      "Ġp o",
+      "ic e",
+      "yp e",
+      "Ġgro up",
+      "âĢ ĵ",
+      "e ver",
+      "ul t",
+      "is m",
+      "ter n",
+      "ab ility",
+      "ion s",
+      "ar k",
+      "Ġn on",
+      "t o",
+      "ĠĠĠĠ ĠĠĠ",
+      "Ġob s",
+      "Ġt re",
+      "al s",
+      "le ft",
+      "ĠP ro",
+      "Ġon ly",
+      "Ġm an",
+      "d er",
+      "Ġp ol",
+      "ur ing",
+      "am et",
+      "ro l",
+      "I n",
+      "y n",
+      "Ġund er",
+      "ĠC h",
+      "Ġw here",
+      "o od",
+      "Ġ X",
+      "n ce",
+      "Ġpar tic",
+      "ect ed",
+      "ĠF ig",
+      "Ġe m",
+      "Ġf act",
+      "ĠA n",
+      "Ġper form",
+      "Ġs o",
+      "Ġanal ysis",
+      "st ract",
+      "he d",
+      "Ġm ay",
+      "at ic",
+      "Ġre p",
+      "te in",
+      "duc ed",
+      "Ġ up",
+      "Ġint o",
+      "Ġnum ber",
+      "Ġo ur",
+      "Ġe t",
+      "e g",
+      "it le",
+      "o ver",
+      "i x",
+      "at or",
+      "ul ti",
+      "Ġinc l",
+      "o uld",
+      "ic i",
+      "b stract",
+      "Ġcomp le",
+      "Ġpati ents",
+      "Ġd o",
+      "Ġex per",
+      "v id",
+      "an ge",
+      "Ġle vel",
+      "Ġpro cess",
+      "math cal",
+      "p s",
+      "Ġsign ific",
+      "Ġs am",
+      "T itle",
+      "Ġb l",
+      "Ġstr uct",
+      "et a",
+      "Ġobs erv",
+      "ra ph",
+      "g r",
+      "Ġac tiv",
+      "Ġf irst",
+      "vel op",
+      "g en",
+      "ib le",
+      "Ġs m",
+      "Ġw ill",
+      "Ġ Q",
+      "Ġmeas ure",
+      "p ut",
+      "Ġl oc",
+      "Ġm o",
+      "ver s",
+      "o f",
+      "t al",
+      "ere d",
+      "ow n",
+      "Ġm at",
+      "iti es",
+      "ti l",
+      "in al",
+      "Ġc ar",
+      "ph a",
+      "Ġb oth",
+      "Ġc ur",
+      "SU B",
+      "it s",
+      "re l",
+      "Ġw hen",
+      "Ġ z",
+      "Ġch ar",
+      "Ġb i",
+      "c ent",
+      "Ġthe n",
+      "is e",
+      "ow ever",
+      "Ġm in",
+      "ĠF or",
+      "Ġ Y",
+      "p tion",
+      "Ġ es",
+      "m un",
+      "Ġincl ud",
+      "is tic",
+      "c on",
+      "Ġob tain",
+      "a red",
+      "duc tion",
+      "Ġsignific ant",
+      "Ġ Z",
+      "Ġp resent",
+      "an n",
+      "Ġ id",
+      "enc y",
+      "Ġv er",
+      "v al",
+      "y d",
+      "ro ugh",
+      "SU P",
+      "f ore",
+      "Ġs ome",
+      "ĠA s",
+      "Ġs up",
+      "Ġa fter",
+      "olog ical",
+      "enti f",
+      "Ġc ase",
+      "Ġs ec",
+      "el f",
+      "Ġde p",
+      "k s",
+      "Ġc al",
+      "v ed",
+      "Ġt em",
+      "Ġus e",
+      "ĠC om",
+      "l am",
+      "in es",
+      "ay s",
+      "Ġg iv",
+      "Ġcons id",
+      "Ġe lect",
+      "ation al",
+      "ĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠ",
+      "i qu",
+      "ti es",
+      "Ġl ine",
+      "Ġs u",
+      "A bstract",
+      "oun t",
+      "Ġde velop",
+      "ĠC on",
+      "olog y",
+      "al pha",
+      "an s",
+      "pr ime",
+      "c c",
+      "og en",
+      "Ġw ork",
+      "v en",
+      "i um",
+      "ec tive",
+      "Ġp a",
+      "t en",
+      "ĠA l",
+      "Ġ ï",
+      "Ġf e",
+      "âĢ Ļ",
+      "enti al",
+      "l ine",
+      "Ġpar amet",
+      "Ġpro tein",
+      "Ġdis c",
+      "f ace",
+      "c es",
+      "Ġw ell",
+      "ur al",
+      "en g",
+      "Ġd uring",
+      "ro w",
+      "an ts",
+      "Ġre m",
+      "form ation",
+      "Ġex am",
+      "Ġm ic",
+      "âĪ Ĵ",
+      "le m",
+      "erg y",
+      "Ġass oci",
+      "Ġ Ã",
+      "ro p",
+      "Ġf ield",
+      "t y",
+      "Ġcl ass",
+      "Ġ u",
+      "i e",
+      "Ġb ec",
+      "Ġexper im",
+      "s p",
+      "Ġp r",
+      "il ar",
+      "ti al",
+      "Ġcon st",
+      "ĠI t",
+      "Ġcont rol",
+      "d a",
+      "Ġm ulti",
+      "iti ve",
+      "ic s",
+      "ur n",
+      "Ġind ic",
+      "Ġf ound",
+      "te xt",
+      "Ġne w",
+      "Ġre f",
+      "g or",
+      "ra p",
+      "Ġdes c",
+      "Ġs ame",
+      "Ġfollow ing",
+      "Ġdist rib",
+      "Fig ure",
+      "il d",
+      "Ġan ti",
+      "etw ork",
+      "o ve",
+      "Ġth rough",
+      "Ġm ost",
+      "c er",
+      "Ġdet erm",
+      "h a",
+      "el ta",
+      "ar ge",
+      "Ġshow n",
+      "in ce",
+      "Ġan y",
+      "re n",
+      "d ot",
+      "r al",
+      "r ation",
+      "am ma",
+      "o id",
+      "Ġm ed",
+      "ens ion",
+      "ar t",
+      "Ġp red",
+      "m et",
+      "mathb b",
+      "ak e",
+      "Ġcal c",
+      "Ġh ig",
+      "Ġth ree",
+      "Ġb ased",
+      "m on",
+      "ar ch",
+      "-- --",
+      "pl es",
+      "ag es",
+      "aus e",
+      "is h",
+      "ti vely",
+      "qu i",
+      "res p",
+      "Ġchar acter",
+      "oc k",
+      "Ġtre at",
+      "Ġpro per",
+      "e x",
+      "Ġsm all",
+      "Ġt erm",
+      "b da",
+      "Ġk n",
+      "od e",
+      "ing s",
+      "Ġexp ression",
+      "Ġm on",
+      "em b",
+      "ut e",
+      "ech n",
+      "h ib",
+      "Ġdi rec",
+      "in ation",
+      "ith m",
+      "ul ated",
+      "Ġc y",
+      "Ġp ot",
+      "Ġor der",
+      "ot e",
+      "ical ly",
+      "Ġval ues",
+      "or t",
+      "ur ther",
+      "ce pt",
+      "yn am",
+      "o ugh",
+      "ech an",
+      "Ġâ ī",
+      "o k",
+      "em ent",
+      "ĠÎ ¼",
+      "Ġes tim",
+      "Ġeff ect",
+      "Ġp ath",
+      "Ġcon f",
+      "Ġap p",
+      "Ġgiv en",
+      "Ġ end",
+      "s et",
+      "Ġg l",
+      "Ġthe y",
+      "n ing",
+      "Ġt est",
+      "Ġtem per",
+      "v es",
+      "Ġval ue",
+      "it ed",
+      "al ity",
+      "Ġl im",
+      "Ġsp ect",
+      "ent ly",
+      "ti t",
+      "Ġse qu",
+      "Ġid entif",
+      "/ /",
+      "ig ma",
+      "Ġen ergy",
+      "in c",
+      "n ess",
+      "ens ity",
+      "Ġprob lem",
+      "yd ro",
+      "ag n",
+      "an e",
+      "re nt",
+      "c om",
+      "j ect",
+      "Ġim port",
+      "ĉ ĉ",
+      "Ġo per",
+      "ol ution",
+      "Ġa ut",
+      "ec tively",
+      "ĠH owever",
+      "h o",
+      "ent al",
+      "Ġs ing",
+      "e y",
+      "m u",
+      "ros s",
+      "ac tion",
+      "ep end",
+      "ĠE x",
+      "vi ous",
+      "Ġstud ies",
+      "s c",
+      "orm al",
+      "Ġh ad",
+      "Ġm ain",
+      "al th",
+      "gor ithm",
+      "Ġf l",
+      "om et",
+      "Ġ Â",
+      ". .",
+      "er r",
+      "Ġpos s",
+      "Ġdiffere n",
+      "Ġobserv ed",
+      "ra y",
+      "Ġpred ic",
+      "Ġgen e",
+      "Ġst ate",
+      "W e",
+      "Ġstruct ure",
+      "Ġre t",
+      "resp ond",
+      "re qu",
+      "il y",
+      "ĠâĪ Ī",
+      "Ġs er",
+      "Ġb ound",
+      "Ġrep resent",
+      "ph i",
+      "Ġtreat ment",
+      "h at",
+      "Ġre qui",
+      "ap p",
+      "um an",
+      "Ġhig her",
+      "Ġl arge",
+      "Ġt ra",
+      "w ard",
+      "Ġobtain ed",
+      "Ġco uld",
+      "ti g",
+      "ĠU n",
+      "Ġdesc rib",
+      "Ġsim ilar",
+      "por ted",
+      "in s",
+      "Ġad dition",
+      "os is",
+      "Ġn etwork",
+      "Ġe le",
+      "p i",
+      "ri x",
+      "Ġr ate",
+      "g an",
+      "ug g",
+      "us s",
+      "Ġm echan",
+      "Ġdis e",
+      "Ġeff ects",
+      "Ġmodel s",
+      "or ph",
+      "ik e",
+      "Ġsec ond",
+      "mathb f",
+      "Ġd ue",
+      "Ġ q",
+      "Ġp res",
+      "Ġt echn",
+      "el s",
+      "Ġcor respond",
+      "Ġassoci ated",
+      "pos ed",
+      "Ġm ass",
+      "ro und",
+      "vi ew",
+      "Ġin s",
+      "ĠâĢ ¢",
+      "di tions",
+      "Ġwh ile",
+      "o le",
+      "Ġl ong",
+      "al u",
+      "Ġc ap",
+      "Ġsur face",
+      "Ġcomple x",
+      "Ġc ent",
+      "Ġcomp ared",
+      "Ġf ind",
+      "arg et",
+      "at ory",
+      "f er",
+      "Ġs ize",
+      "Ġcont ain",
+      "us ion",
+      "u tions",
+      "Ġd em",
+      "E S",
+      "Ġdep end",
+      "at is",
+      "s um",
+      "ff ici",
+      "Ġb as",
+      "lam bda",
+      "i er",
+      "A T",
+      "Ġm ax",
+      "Ġim p",
+      "Ġev alu",
+      "Ġtemper ature",
+      "in k",
+      "ect or",
+      "Ġs cal",
+      "Ġgro w",
+      "ow er",
+      "Ġresp ectively",
+      "le ar",
+      "s h",
+      "ic k",
+      "Ġf il",
+      "ir c",
+      "il on",
+      "r am",
+      "ĠÎ ±",
+      "ific ation",
+      "Ġo cc",
+      "Ġy ear",
+      "Ġs ugg",
+      "Ġra di",
+      "if ied",
+      "ha vi",
+      "Ġwith in",
+      "Ġs ens",
+      "Ġin te",
+      "Ġw ould",
+      "Ġcon cent",
+      "Ġmic ro",
+      "Ġsing le",
+      "ĠS p",
+      "o u",
+      "Ġat t",
+      "Ġs elf",
+      "Ġab out",
+      "eng th",
+      "Ġe l",
+      "ĠR e",
+      "x im",
+      "Ġcon ditions",
+      "ud e",
+      "ĠA t",
+      "w here",
+      "m ed",
+      "Ġne ed",
+      "ir on",
+      "Ġp op",
+      "Ġresul t",
+      "Ġpo int",
+      "Ġl o",
+      "Ġal gorithm",
+      "Ġactiv ity",
+      "le q",
+      "ple ment",
+      "ĠR es",
+      "Ġsy m",
+      "on str",
+      "at ures",
+      "Ġim pro",
+      "f or",
+      "Ġgener al",
+      "it er",
+      "Ġex pl",
+      "## #",
+      "Ġd om",
+      "Ġt ri",
+      "m in",
+      "Ġdistrib ution",
+      "Ġt r",
+      "ĠThe re",
+      "os s",
+      "u ce",
+      "math rm",
+      "ul l",
+      "E R",
+      "re g",
+      "Ġp e",
+      "Ġto tal",
+      "Ġle ad",
+      "= =",
+      "i od",
+      "Ġass um",
+      "Ġch ang",
+      "Ġg ra",
+      "M I",
+      "Ġcomp ut",
+      "Ġcom b",
+      "Ġin formation",
+      "Ġdes ign",
+      "Ġin iti",
+      "Ġf requ",
+      "im ension",
+      "c op",
+      "Ġproper ties",
+      "Ġconsid er",
+      "Ġlevel s",
+      "en e",
+      "Ġt ype",
+      "iv ed",
+      "ĠH e",
+      "epend ent",
+      "Ġap plic",
+      "Ġinv es",
+      "Ġpre vious",
+      "a w",
+      "Ġsp ace",
+      "Ġpro vid",
+      "h yl",
+      "Ġinves tig",
+      "Ġappro ach",
+      "ater ial",
+      "on se",
+      "lec ular",
+      "Ġparamet ers",
+      "Ġph ase",
+      "ul ations",
+      "ub l",
+      "b eta",
+      "Ġa v",
+      "Ġf lu",
+      "Ġpot ential",
+      "ĠThe se",
+      "s igma",
+      "l o",
+      "tim es",
+      "Ġop tim",
+      "is ion",
+      "Ġa ff",
+      "Ġme an",
+      "Ġbe havi",
+      "Ġv ol",
+      "ore m",
+      "ag ne",
+      "Ġdec re",
+      "tion al",
+      "Ġsol ution",
+      "Ġh uman",
+      "g er",
+      "Ġpa per",
+      "Ġcomp ar",
+      "Ġlow er",
+      "and ard",
+      "Ġcor rel",
+      "c ri",
+      "Ġcur rent",
+      "Ġd er",
+      "iss ion",
+      "ĠFig ure",
+      "Ġpro duc",
+      "Ġw ater",
+      "ĠT o",
+      "Ġth ose",
+      "Ġac id",
+      "Ġcan cer",
+      "Ġloc al",
+      "t on",
+      "Ġf low",
+      "Ġreg ion",
+      "Ġhe alth",
+      "Ġimport ant",
+      "og raph",
+      "ab l",
+      "Ġse lec",
+      "Ġg re",
+      "Ġin di",
+      "ad e",
+      "r id",
+      "Ġsh ould",
+      "b ased",
+      "Ġab ove",
+      "l d",
+      "Ġsystem s",
+      "ic ation",
+      "Ġ ed",
+      "Ġt yp",
+      "Ġph ys",
+      "o per",
+      "Ġcomp on",
+      "O N",
+      "Ġsu per",
+      "g a",
+      "hem ical",
+      "is k",
+      "op h",
+      "Ġh y",
+      "Ġanal y",
+      "in u",
+      "Ġt arget",
+      "ĠA d",
+      "Ġp at",
+      "g amma",
+      "Ġsam ples",
+      "Ġs l",
+      "Ġpar t",
+      "old s",
+      "Ġb el",
+      "im um",
+      "ĠI m",
+      "Ġdise ase",
+      "I I",
+      "is ts",
+      "i ver",
+      "Ġperform ance",
+      "ĠĠĠĠĠĠĠĠ ĠĠĠ",
+      "g le",
+      "Ġo x",
+      "nd om",
+      "ĠĠĠĠ Ġ",
+      "Ġbec ause",
+      "ay er",
+      "Ġr ange",
+      "Ġco un",
+      "Ġincre ased",
+      "oc h",
+      "on al",
+      "Ġver y",
+      "Ġd ynam",
+      "an ti",
+      "Ġad d",
+      "Ġin hib",
+      "Ġmethod s",
+      "id ence",
+      "in ical",
+      "ere nce",
+      "iv al",
+      "u le",
+      "Ġfact or",
+      "Ġf in",
+      "in ts",
+      "v iron",
+      "Ġs our",
+      "ver age",
+      "e qu",
+      "Ġe ar",
+      "Ġshow ed",
+      "it es",
+      "Ġperform ed",
+      "Ġre se",
+      "ĠE n",
+      "Ġspec ies",
+      "A C",
+      "ĠC l",
+      "h ip",
+      "til de",
+      "i o",
+      "at ely",
+      "T h",
+      "od y",
+      "Ġincre ase",
+      "ĠP h",
+      "âĢ Ŀ",
+      "Ġshow s",
+      "ĠA c",
+      "Ġp ost",
+      "ord ing",
+      "enc es",
+      "o y",
+      "n er",
+      "Ġresp onse",
+      "Ġocc ur",
+      "r ho",
+      "Ġper iod",
+      "ar s",
+      "Ġre d",
+      "ĠO n",
+      "Ġd ensity",
+      "Ġexam ple",
+      "g et",
+      "Ġre al",
+      "ĠC ount",
+      "ac y",
+      "Ġp ower",
+      "Ġab s",
+      "it al",
+      "Ġpr im",
+      "âĢ IJ",
+      "Ġdef ined",
+      "Ġn ormal",
+      "a j",
+      "Ġin st",
+      "Ġal low",
+      "Ġposs ible",
+      "Ġv is",
+      "Ġre ported",
+      "Ġsign al",
+      "the ta",
+      "Ġd en",
+      "ab les",
+      "Ġde g",
+      "Ġindi vid",
+      "agne tic",
+      "Ġgroup s",
+      "a e",
+      "ar row",
+      "Ġst at",
+      "Ġmechan ism",
+      "os p",
+      "m er",
+      "ot her",
+      "Ġpro t",
+      "Ġc ases",
+      "Ġc r",
+      "Ġt e",
+      "Ġinte gr",
+      "et s",
+      "Ġdevelop ment",
+      "Ġra ndom",
+      "Ġinv ol",
+      "Ġinclud ing",
+      "Ġ err",
+      "gr am",
+      "Ġpartic ular",
+      "ep s",
+      "Ġst andard",
+      "pos ition",
+      "Ġcont rib",
+      "se qu",
+      "Ġman y",
+      "Ġf urther",
+      "Ġsignificant ly",
+      "at ors",
+      "ur b",
+      "Ġag ain",
+      "b ar",
+      "Ġwith out",
+      "Ġse ver",
+      "Ġto p",
+      "re t",
+      "l ed",
+      "Ġmat rix",
+      "Ġspec ific",
+      "ate g",
+      "Ĩ Ĵ",
+      "Ġdirec t",
+      "Ġsam ple",
+      "Ġthe m",
+      "S A",
+      "o int",
+      "Ġro le",
+      "Ġchang es",
+      "rac tion",
+      "Ġs um",
+      "Ġindivid ual",
+      "I N",
+      "Ġim mun",
+      "c ed",
+      "o h",
+      "Ġstr ong",
+      "Ġe p",
+      "Ġline ar",
+      "u ally",
+      "d elta",
+      "w ay",
+      "as ing",
+      "Ġt im",
+      "Ġv i",
+      "is on",
+      "Ġfunc tions",
+      "Ġam ong",
+      "Ġse e",
+      "ere st",
+      "Ġgrow th",
+      "Ġr ati",
+      "ĠS c",
+      "ix ed",
+      "R NA",
+      "e ed",
+      "ta u",
+      "Ġ ent",
+      "Ġd r",
+      "o res",
+      "Ġappro xim",
+      "f ul",
+      "Ġre le",
+      "Ġfact ors",
+      "Ġdisc uss",
+      "Ġph ot",
+      "Ġpro posed",
+      "er o",
+      "ome ga",
+      "Ġf our",
+      "as tic",
+      "Ġyear s",
+      "hes is",
+      "iqu e",
+      "Ġm aterial",
+      "Ġb re",
+      "Ġpro f",
+      "ĠA p",
+      "Ġne g",
+      "Ġb u",
+      "Ġass ess",
+      "ĠâĢ ľ",
+      "Ġv ir",
+      "at ter",
+      "Ġdescrib ed",
+      "istic s",
+      "Ġcomp os",
+      "a z",
+      "str uc",
+      "Ġt um",
+      "par tial",
+      "a f",
+      "Ġwh o",
+      "at al",
+      "Ġdem onstr",
+      "anc es",
+      "y t",
+      "Ġrem ain",
+      "Ġl ess",
+      "Ġpos itive",
+      "om ic",
+      "Ġs ince",
+      "og n",
+      "Ġcon dition",
+      ": :",
+      "Ġdo es",
+      "ti ce",
+      "os ph",
+      "Ġpro v",
+      "ĠC O",
+      "Ġr at",
+      "Ġterm s",
+      "b ox",
+      "Ġt ak",
+      "Ġpat tern",
+      "al e",
+      "Ġn an",
+      "ul es",
+      "Ġm ut",
+      "is hed",
+      "Ġrel ated",
+      "Ġthe ory",
+      "b ol",
+      "c dot",
+      "viron ment",
+      "a ir",
+      "i vers",
+      "ĠA r",
+      "Ġï £",
+      "ress ed",
+      "Ġâī ¤",
+      "ĠM et",
+      "I D",
+      "ul ts",
+      "ĠÎ ²",
+      "Ġd at",
+      "pos e",
+      "Ġor ig",
+      "Ġret urn",
+      "Ġch ange",
+      "Ġl arg",
+      "a u",
+      "ac es",
+      "Ġare a",
+      "Ġgen es",
+      "A S",
+      "Ġh ydro",
+      "Ġcons ist",
+      "m an",
+      "Ġrese arch",
+      "ĠD e",
+      "Ġor gan",
+      "as k",
+      "Ġb ack",
+      "Ġfollow s",
+      "un g",
+      "ro ll",
+      "Ġequ ation",
+      "pl ied",
+      "t r",
+      "Ġcorrespond ing",
+      "od es",
+      "es ted",
+      "Ġrel ations",
+      "n al",
+      "Ġf r",
+      "Ġlim it",
+      "m it",
+      "Ġof f",
+      "ut ed",
+      "Ġr isk",
+      "re ad",
+      "Ġkn own",
+      "pl it",
+      "tiv ity",
+      "Ġsequ ence",
+      "Ġconsid ered",
+      "x i",
+      "ĠM od",
+      "v ity",
+      "Ġn uc",
+      "c le",
+      "ic es",
+      "Ġl ength",
+      "Ġsever al",
+      "s ing",
+      "o ot",
+      "n ot",
+      "Ġst ress",
+      "ĠI f",
+      "C T",
+      "ro ph",
+      "Ġcom mun",
+      "Ġcl ust",
+      "ĠL e",
+      "m e",
+      "ant um",
+      "Ġm emb",
+      "Ġl ab",
+      "Ġev en",
+      "Ġinf lu",
+      "c k",
+      "ĠÃ Ĺ",
+      "Ġl og",
+      "v ing",
+      "es ts",
+      "Ġh is",
+      "an k",
+      "ĠI nd",
+      "ac tions",
+      "ft y",
+      "m od",
+      "Ġre view",
+      "th ough",
+      "Ġeff ici",
+      "Ġm ap",
+      "in fty",
+      "Ġbe ing",
+      "l and",
+      "Ġcl inical",
+      "Ġmeasure d",
+      "er ing",
+      "ĠT able",
+      "Ġs he",
+      "se e",
+      "Ġs ection",
+      "Ġav ail",
+      "om en",
+      "Ġv ers",
+      "Ġd el",
+      "it her",
+      "er ation",
+      "Ġh and",
+      "Ġcont inu",
+      "Ġcon n",
+      "h ors",
+      "ra d",
+      "Ġf am",
+      "Ġle ar",
+      "Ġiniti al",
+      "y stem",
+      "Ġg e",
+      "ĠâĢ ²",
+      "Ġc irc",
+      "Ġp ubl",
+      "ĠI s",
+      "Ġv ia",
+      "Ġcom mon",
+      "if e",
+      "Ġm ark",
+      "Ġe ver",
+      "ar c",
+      "b ig",
+      "er tain",
+      "\\ \\",
+      "v ar",
+      "A s",
+      "ros cop",
+      "Ġa ge",
+      "Ġh ow",
+      "ĠL et",
+      "str uct",
+      "Ġa verage",
+      "v ant",
+      "ĠS h",
+      "imension al",
+      "S C",
+      "ap e",
+      "n u",
+      "Ġl oss",
+      "as on",
+      "id es",
+      "Ġpop ulation",
+      "Ġdom ain",
+      "ind ing",
+      "w e",
+      "A L",
+      "Ġacc ur",
+      "et y",
+      "Ġc aus",
+      "D elta",
+      "rap y",
+      "Ġpro m",
+      "tim e",
+      "Ġint ro",
+      "Ġmulti ple",
+      "Ġconst ant",
+      "pl ing",
+      "in o",
+      "aj or",
+      "i or",
+      "ab ol",
+      "de f",
+      "Ġpo ints",
+      "ver se",
+      "n ame",
+      "ĠS e",
+      "it or",
+      "P ro",
+      "ar m",
+      "Ġt iss",
+      "Ġf ib",
+      "Ġg raph",
+      "Ġc all",
+      "atis f",
+      "Ġcon duc",
+      "de x",
+      "ĠN e",
+      "Ġp ers",
+      "er n",
+      "C R",
+      "ang le",
+      "Ġfrequ ency",
+      "A P",
+      "Ġpresent ed",
+      "am p",
+      "Ġbe fore",
+      "ord s",
+      "Ġin put",
+      "Ġâ ĨĴ",
+      "Ġpartic ip",
+      "O R",
+      "Ġch ild",
+      "Ġc re",
+      "ffici ent",
+      "Ġse par",
+      "ur ation",
+      "Î ±",
+      "Ġex ist",
+      "is ed",
+      "Ġl ight",
+      "im al",
+      "** **",
+      "ĠD NA",
+      "he l",
+      "Ġint erest",
+      "b f",
+      "k e",
+      "Ġcol lec",
+      "Ġt rain",
+      "a i",
+      "ĠP l",
+      "ĠÎ »",
+      "ĠC o",
+      "Ġim age",
+      "Ġh yp",
+      "om a",
+      "Ġwe ight",
+      "Ġc ross",
+      "r t",
+      "Ġdiffere nce",
+      "Ġfe atures",
+      "med i",
+      "t ype",
+      "Ġp ress",
+      "I C",
+      "Ġthe rm",
+      "Ġst ates",
+      "u str",
+      "ti ll",
+      "Ġh ist",
+      "Ġrati o",
+      "ag ing",
+      "ĠA ll",
+      "Ġhe l",
+      "b on",
+      "Ġbehavi or",
+      "Ġp ri",
+      "Ġsy nt",
+      "end ed",
+      "ĠIn t",
+      "t t",
+      "Ġvari ous",
+      "rec t",
+      "Ġpre c",
+      "Ġtim es",
+      "M S",
+      "Ġanaly z",
+      "Ġc are",
+      "m at",
+      "Ġal ong",
+      "Ġp ur",
+      "ati vely",
+      "Ġst ar",
+      "j ects",
+      "i i",
+      "ist ance",
+      "ĠThe n",
+      "A N",
+      "Ġparamet er",
+      "ul ate",
+      "Ġever y",
+      "Ġs atisf",
+      "Ġdeterm ined",
+      "in a",
+      "ran e",
+      "Ġpa ir",
+      "o ol",
+      "T able",
+      "Ġth us",
+      "ogen e",
+      "ĠÏ Ĩ",
+      "Ġpro gram",
+      "as c",
+      "Ġen vironment",
+      "M P",
+      "Ġre ad",
+      "Ġac h",
+      "Ġpres ence",
+      "Ġm ice",
+      "F or",
+      "Ġpro duction",
+      "Ġdifferen ces",
+      "Ġprov ide",
+      "st e",
+      "am es",
+      "ĉ Ġ",
+      "ĠÂ ±",
+      "ro up",
+      "Ġelect ron",
+      "Ġhy per",
+      "b it",
+      "ĠR ec",
+      "Ġv ector",
+      "ub le",
+      "ran gle",
+      "Ġw r",
+      "w ide",
+      "Ġâ Ĭ",
+      "rac k",
+      "ry st",
+      "Ġin j",
+      "eg a",
+      "Ġw he",
+      "ps ilon",
+      "Ġagain st",
+      "Ġdi agn",
+      "Ġh om",
+      "Ġach ie",
+      "n s",
+      "Ġre ce",
+      "---- ----",
+      "Ġavail able",
+      "in f",
+      "Ġs uc",
+      "Ġg u",
+      "Ġm ajor",
+      "ĠTh us",
+      "w are",
+      "Ġsup port",
+      "l or",
+      "Ġexperim ental",
+      "ĠM o",
+      "Ġconcent ration",
+      "tic s",
+      "Ġn ec",
+      "Ġp hen",
+      "s q",
+      "Ġcl os",
+      "s ub",
+      "Ġkn ow",
+      "Ġform ation",
+      "Ġd id",
+      "ous e",
+      "in ary",
+      "ic t",
+      "ĠC D",
+      "Th is",
+      "l ess",
+      "Ġne ar",
+      "Ġimpro ve",
+      "ab il",
+      "Ġre ve",
+      "Ġexperim ents",
+      "i ence",
+      "ul a",
+      "ore d",
+      "Ġ unc",
+      "_ _",
+      "Ġap plied",
+      "Ġre duced",
+      "Ġdet ail",
+      "st and",
+      "Ġch o",
+      "om y",
+      "Ġcalc ulated",
+      "Ġen h",
+      "L ES",
+      "it ro",
+      "Ġresp ons",
+      "Ġ est",
+      "Ġm i",
+      "Ġco e",
+      "ĠThere fore",
+      "ĠM ore",
+      "b l",
+      "anc ed",
+      "um e",
+      "Ġb and",
+      "Ġac t",
+      "Ġe ither",
+      "om es",
+      "ĠG en",
+      "v are",
+      "E T",
+      "re en",
+      "ĠP ar",
+      "ĠS im",
+      "Ġidentif ied",
+      "Ġinter action",
+      "Ġm ade",
+      "Ġsour ce",
+      "ti s",
+      "ot s",
+      "m ega",
+      "Ġs erv",
+      "m s",
+      "al ysis",
+      "v ent",
+      "en se",
+      "g l",
+      "Ġl ines",
+      "Ġapp ear",
+      "ti f",
+      "Ġf ree",
+      "om s",
+      "in ing",
+      "ere n",
+      "Ġch ann",
+      "vare psilon",
+      "s im",
+      "Ġco u",
+      "Â °",
+      "Ġerr or",
+      "Ġqu anti",
+      "ĠE q",
+      "b y",
+      "ĠI I",
+      "te x",
+      "ĠS ch",
+      "sq rt",
+      "oc us",
+      "Ġde v",
+      "qu ad",
+      "ter s",
+      "Ġrelations hip",
+      "ol l",
+      "Ġg o",
+      "Ġw ave",
+      "Ġle ft",
+      "w ays",
+      "h i",
+      "Ġr ight",
+      "ob al",
+      "Ġd own",
+      "u k",
+      "Ġcol l",
+      "Ġm agnetic",
+      "Ġpro g",
+      "dot s",
+      "Ġstr ateg",
+      "b s",
+      "unc tion",
+      "Ġen c",
+      "Ġc lear",
+      "Ġco st",
+      "ge b",
+      "et ter",
+      "MI LES",
+      "lam m",
+      "Ġm ust",
+      "Ġeff ective",
+      "Ġex c",
+      "Ġpl as",
+      "Ġsugg est",
+      "i tions",
+      "Ġle ast",
+      "y ing",
+      "ly ing",
+      "Ġl ik",
+      "O mega",
+      "ak ing",
+      "Ġmax imum",
+      "Ġrel ative",
+      "Ã ©",
+      "Ġacc ording",
+      "i ent",
+      "Ġw ay",
+      "Ġs em",
+      "at ural",
+      "l ike",
+      "res h",
+      "ĠM e",
+      "P s",
+      "ĠT rans",
+      "is c",
+      "Ġp rac",
+      "Ġr un",
+      "Ġcon ver",
+      "Ġs k",
+      "Ġy ield",
+      "ge q",
+      "ab ly",
+      "Ġanti b",
+      "iz ing",
+      "Î ²",
+      "m ission",
+      "Ġn ow",
+      "Ġdet ection",
+      "el oc",
+      "Ġg et",
+      "er t",
+      "Ġvari ables",
+      "Ġop en",
+      "Ġpress ure",
+      "Ġst rain",
+      "um ent",
+      "ĠF urther",
+      "Ġqu antum",
+      "Ġim plement",
+      "Ġear ly",
+      "Ġfr ame",
+      "Ġsh ort",
+      "Ġdr ug",
+      "Ġrequi red",
+      "P S",
+      "Ġm y",
+      "Ġm uch",
+      "Ġm em",
+      "C C",
+      "Ġqu ality",
+      "Ġprotein s",
+      "Ġl ayer",
+      "Ġqu es",
+      "Ġre cept",
+      "Ġhe re",
+      "Ġpro ced",
+      "ure d",
+      "Ġdevelop ed",
+      "Ġpos ition",
+      "r um",
+      "Ġl at",
+      "Ġincre asing",
+      "E M",
+      "Ġmeasure ments",
+      "Ġb en",
+      "Ġis ol",
+      "w h",
+      "T o",
+      "Ġval id",
+      "Ġfunction al",
+      "em ma",
+      ".. .",
+      "or ld",
+      "ri es",
+      "Ġprob ability",
+      "ĠN ew",
+      "Ġm m",
+      "O S",
+      "A D",
+      "ĠÎ ´",
+      "Ġscal e",
+      "ĠF e",
+      "ĠThe orem",
+      "ĠQ u",
+      "Ġcompon ents",
+      "Ġbl ood",
+      "ĠÏ ĥ",
+      "ac c",
+      "Ġb etter",
+      "Ġst ep",
+      "ĠÎ ³",
+      "Ġf ac",
+      "ane ous",
+      "Ġlo ad",
+      "Ġmet abol",
+      "Ġev olution",
+      "s on",
+      "re am",
+      "Ġe as",
+      "ir d",
+      "d imensional",
+      "b or",
+      "Ġm us",
+      "Ġequ ations",
+      "ps i",
+      "ord er",
+      "ol ar",
+      "Ġnum er",
+      "Ġk ey",
+      "or th",
+      "Ġsim ple",
+      "if t",
+      "cal e",
+      "Ġin dex",
+      "ĠâĢ ĵ",
+      "Ġconcent r",
+      "g es",
+      "Ġneg ative",
+      "Ġv eloc",
+      "Ġa x",
+      "ĠE ff",
+      "Ġfin ite",
+      "Ġ ill",
+      "ch ing",
+      "Ġpati ent",
+      "eps ilon",
+      "Ġm en",
+      "Ġc ri",
+      "I S",
+      "C l",
+      "Ġcon cl",
+      "ĠÎ ¸",
+      "ib ility",
+      "Ġsym met",
+      "ent er",
+      "Ġdist ance",
+      "Ġpol ym",
+      "igh ts",
+      "Ġc ult",
+      "Ġpe ak",
+      "Ġac ross",
+      "in ition",
+      "Ġle t",
+      "Ġcon struc",
+      "Ġinclud ed",
+      "Ġh owever",
+      "Ġreg ions",
+      "Ġlear ning",
+      "Ġev idence",
+      "in ally",
+      "Ġne ut",
+      "it ation",
+      "Ġwhe ther",
+      "Ġout put",
+      "ĠS ection",
+      "Ġg ood",
+      "I T",
+      "u ation",
+      "Ġtyp es",
+      "b m",
+      "c os",
+      "w ith",
+      "l im",
+      "o tic",
+      "Ġs till",
+      "Ġd ays",
+      "Ġstud ied",
+      "Ġim ages",
+      "b le",
+      "Ġar g",
+      "line ar",
+      "Ġprocess es",
+      "Ġw id",
+      "Ġtrain ing",
+      "Ġind ependent",
+      "pl ac",
+      "Ġres id",
+      "Ġsuc cess",
+      "Ġnuc le",
+      "G F",
+      "le t",
+      "pl oy",
+      "Ġtum or",
+      "G amma",
+      "Ġthere fore",
+      "r ast",
+      "Ġf ocus",
+      "as h",
+      "Ġbel ow",
+      "ial ly",
+      "Ġcompar ison",
+      "Ġad j",
+      "Ġl ike",
+      "Ġmo lecular",
+      "ri ed",
+      "Ġf it",
+      "ĠD i",
+      "l og",
+      "Ġpl ay",
+      "w ork",
+      "ec tions",
+      "Ġelect ro",
+      "u it",
+      "m ore",
+      "Ġm ight",
+      "Ġanal ys",
+      "Ġme ans",
+      "Ġcorrel ation",
+      "k n",
+      "Ġcont roll",
+      "I V",
+      "C h",
+      "p ec",
+      "ra g",
+      "Ġm agn",
+      "Ġphys ical",
+      "I ON",
+      "Ġreve al",
+      "Ġph osph",
+      "Ġr ates",
+      "Ġlarg er",
+      "Ġs tim",
+      "Ġso ft",
+      "Ġcomp ound",
+      "b e",
+      "ch i",
+      "ĠN o",
+      "Ġimp act",
+      "t or",
+      "Ġprim ary",
+      "oc ial",
+      "Ġapplic ation",
+      "Ġsol utions",
+      "d uce",
+      "Ġcharacter istics",
+      "Ġele ments",
+      "Ġvi ew",
+      "Ġl ater",
+      "ut ure",
+      "Ġfam ily",
+      "ri al",
+      "Ġtrans cri",
+      "or ption",
+      "Ġs w",
+      "C D",
+      "E D",
+      "Ġem b",
+      "Ġz ero",
+      "ol s",
+      "Ġl ife",
+      "ce p",
+      "ĠL i",
+      "th s",
+      "Ġser ies",
+      "Ġa round",
+      "Ġtrans ition",
+      "ĠC or",
+      "ĠâĪ Ĥ",
+      "Ġdat as",
+      "Ġ her",
+      "ĠB y",
+      "A M",
+      "sp ec",
+      "ol es",
+      "ograph y",
+      "t le",
+      "ĠC ar",
+      "al le",
+      "Ġest abl",
+      "ag ement",
+      "Ġsc hem",
+      "g round",
+      "Ġf ail",
+      "Ġexp ected",
+      "Ġrequi re",
+      "ar ray",
+      "Ġexperim ent",
+      "Ġele ment",
+      "Ġne u",
+      "Ġgener ated",
+      "Ġs ite",
+      "ĠCon t",
+      "ĠR NA",
+      "er al",
+      "Ġcont ent",
+      "Ġb acter",
+      "l er",
+      "Ġtrans fer",
+      "ul f",
+      "right arrow",
+      "an y",
+      "ĠS ince",
+      "in duced",
+      "Ġre action",
+      "he ck",
+      "Ġstruct ures",
+      "Ġcoun t",
+      "Ġdeterm ine",
+      "z ym",
+      "ĠB l",
+      "Ġunder stand",
+      "oc al",
+      "Ġsy n",
+      "Ġpol y",
+      "ur y",
+      "Ġb est",
+      "Ġf ixed",
+      "ren g",
+      "Ġc hemical",
+      "Ġtiss ue",
+      "Ġp ul",
+      "Ġbound ary",
+      "is ing",
+      "Ġb ro",
+      "atis tical",
+      "ic ity",
+      "s k",
+      "r ing",
+      "Ġl ast",
+      "Ġchild ren",
+      "r im",
+      "Ġre duction",
+      "Ġsp in",
+      "Ġb ody",
+      "oper ator",
+      "v ari",
+      "Ġd iv",
+      "ym bol",
+      "Ġm al",
+      "Ġsp ati",
+      "a h",
+      "ĠB i",
+      "b ack",
+      "s y",
+      "Ġse en",
+      "ĠW ith",
+      "id s",
+      "plic ations",
+      "Ġnec ess",
+      "Ġs ide",
+      "Ġb rain",
+      "Ġf ew",
+      "Ġapplic ations",
+      "ut es",
+      "ac hes",
+      "Ġac tive",
+      "var phi",
+      "ter m",
+      "Ġm om",
+      "ivers ity",
+      "Ġf inal",
+      "led ge",
+      "Ġdynam ics",
+      "av ing",
+      "er c",
+      "orph ism",
+      "on es",
+      "o ff",
+      "p m",
+      "Ġac tion",
+      "Ġn atural",
+      "ĠG e",
+      "Ġy ou",
+      "le x",
+      "ĠĠĠĠ ĠĠ",
+      "s tit",
+      "Ġg as",
+      "Ġm ake",
+      "Ġin duced",
+      "ĠA fter",
+      "ĠW h",
+      "Ġcompon ent",
+      "Ġinf ection",
+      "ĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠ",
+      "Ġconf ir",
+      "ig en",
+      "ĠS ystem",
+      "tic le",
+      "Ġprovid ed",
+      "tern al",
+      "b ers",
+      "O D",
+      "ĠIn ter",
+      "ot t",
+      "av es",
+      "ĠSt ud",
+      "p y",
+      "Ġres istance",
+      "ĠS ur",
+      "at ch",
+      "Ġd im",
+      "Ġinter p",
+      "Ġcy cl",
+      "on t",
+      "it ing",
+      "A G",
+      "Ġequ ival",
+      "ot ype",
+      "Ġprevious ly",
+      "Ġaddition al",
+      "out h",
+      "Ġim pl",
+      "Ġ ion",
+      "Ġ ir",
+      "Ġc op",
+      "Ġh al",
+      "Ġactiv ation",
+      "l angle",
+      "Ġf ull",
+      "S S",
+      "ĠO p",
+      "id d",
+      "Ġpro of",
+      "Ġproblem s",
+      "Ġtrans form",
+      "Ġinter actions",
+      "Ġsup p",
+      "d es",
+      "ĠR eg",
+      "operator name",
+      "eg in",
+      "Ġc ryst",
+      "Ġincre ases",
+      "ron ic",
+      "Ġad ap",
+      "in ant",
+      "Ġveloc ity",
+      "ĠAs s",
+      "iqu es",
+      "Ġcontinu ous",
+      "ĠCom p",
+      "ĠPro per",
+      "Ġpri or",
+      "or b",
+      "Ġno vel",
+      "Ġbl ock",
+      "Ġvol ume",
+      "Ġreg ard",
+      "omet ry",
+      "E C",
+      "Ġresul ting",
+      "ĠO r",
+      "Ġcar bon",
+      "are nt",
+      "Ġb inding",
+      "i j",
+      "Ġac cess",
+      "Ġwe ak",
+      "Ġun it",
+      "Ġ ide",
+      "\" \"",
+      "Ġc m",
+      "Ġcri tical",
+      "Ġresp ect",
+      "t rans",
+      "Ġâī ¥",
+      "Ġs al",
+      "e ad",
+      "Ġsim ulation",
+      "Ġcap ac",
+      "iti vity",
+      "Ġrec ord",
+      "ra k",
+      "Ġne ur",
+      "on ic",
+      "op le",
+      "Ġm g",
+      "Ġst reng",
+      "er ve",
+      "Ġre duc",
+      "Ġp ass",
+      "ord in",
+      "ex p",
+      "j ective",
+      "ens or",
+      "Ġpartic les",
+      "Ġa ir",
+      "Ġl ink",
+      "ĠÏ Ħ",
+      "Ġl ist",
+      "c in",
+      "ĠO ur",
+      "p ri",
+      "ve re",
+      "ib r",
+      "if orm",
+      "Ġexpl ain",
+      "Ġf em",
+      "Ġu til",
+      "S t",
+      "over line",
+      "Ġof ten",
+      "er y",
+      "op e",
+      "ĠU sing",
+      "b egin",
+      "Ġdifferen ti",
+      "per s",
+      "s elf",
+      "iz es",
+      "Ġconcentr ations",
+      "I R",
+      "ĠS up",
+      "Ġbas is",
+      "Ġinclud e",
+      "ĠB ond",
+      "Ġext rac",
+      "ĠMet hod",
+      "ĠD ata",
+      "ĠD ef",
+      "w n",
+      "Ġnetwork s",
+      "ign ed",
+      "âĢ ¢",
+      "Ġexp ressed",
+      "Ġcont rast",
+      "es is",
+      "c ol",
+      "in ter",
+      "p id",
+      "Ġd ri",
+      "Ġdef ine",
+      "Ġinflu ence",
+      "Ġselec ted",
+      "E L",
+      "Ġcontain ing",
+      "Ġs il",
+      "geb ra",
+      "re at",
+      "b olds",
+      "Ġinvestig ated",
+      "ĠC ol",
+      "ym met",
+      "yt es",
+      "Ġmo lec",
+      "Ġinvol ved",
+      "Ġd ay",
+      "Ġch ain",
+      "ĠMore over",
+      "Ġdi ag",
+      "Ġan g",
+      "Ġlik ely",
+      "Ġspect rum",
+      "Ġder iv",
+      "bolds ymbol",
+      "Ġhel p",
+      "ĠA m",
+      "Ġtre ated",
+      "Ġvari able",
+      "ell ular",
+      "ĠD es",
+      "ap s",
+      "Ġn m",
+      "ĠÏ ģ",
+      "ĠW hen",
+      "Ġhigh ly",
+      "am in",
+      "Ġwh at",
+      "rel ated",
+      "Ġch rom",
+      "Ġsur v",
+      "ĠAn alysis",
+      "Ġs it",
+      "f act",
+      "od ing",
+      "Ġproduc t",
+      "Ġev ents",
+      "r as",
+      "ĠP er",
+      "ma x",
+      "ĠA g",
+      "con t",
+      "ic ro",
+      "Ġad v",
+      "Ġcall ed",
+      "Ġdeg ree",
+      "A B",
+      "T R",
+      "Ġse g",
+      "ĠC an",
+      "Ġdemonstr ated",
+      "w ise",
+      "Ġ ve",
+      "ĠC a",
+      "Ġdet ected",
+      "c o",
+      "Ġder ived",
+      "Ġex hib",
+      "Ġgl obal",
+      "al ax",
+      "ul ating",
+      "A l",
+      "ang u",
+      "b o",
+      "Ġrec om",
+      "Ġfe ature",
+      "d ependent",
+      "Ġro t",
+      "ven tion",
+      "Ġrem ov",
+      "Ġw ind",
+      "Ġaccur acy",
+      "s ize",
+      "Ġsum m",
+      "Ġmeasure ment",
+      "Ġfield s",
+      "ward s",
+      "Ġl iter",
+      "atal y",
+      "ĠSt r",
+      "Ġre port",
+      "Ġcent ral",
+      "Ġs qu",
+      "Ġthe rapy",
+      "he st",
+      "Ġfe ed",
+      "S MILES",
+      "ĠA N",
+      "Ġs ites",
+      "âĢ ²",
+      "our s",
+      "om al",
+      "Ġl ip",
+      "Ġanalyz ed",
+      "ĠÂ °",
+      "Ġwe e",
+      "t em",
+      "Ġan other",
+      "il es",
+      "Ġcomple te",
+      "Ġne xt",
+      "ĠO ne",
+      "b i",
+      "ri p",
+      "st ate",
+      "ĠMod el",
+      "Ġfind ings",
+      "ĠP re",
+      "Ġrec ent",
+      "asc ular",
+      "Ġestim ate",
+      "Ġmechanism s",
+      "ĠRes ults",
+      "Ġparticip ants",
+      "Ġen g",
+      "m ost",
+      "omet ric",
+      "Ġequ al",
+      "Ġro b",
+      "Ġpol ar",
+      "Ġgene tic",
+      "Ġb o",
+      "Ġre st",
+      "ĠÏ Ģ",
+      "Ġrel ation",
+      "Ġques tion",
+      "ep ti",
+      "Ġdiff ic",
+      "em s",
+      "Ġf uture",
+      "if y",
+      "Ġmod e",
+      "Ġmemb rane",
+      "Ġhe at",
+      "A ut",
+      "d ing",
+      "Ġox id",
+      "Ġconf ig",
+      "plic ation",
+      "ĠM on",
+      "alle l",
+      "id ed",
+      "Ġdirec tion",
+      "pl ed",
+      "Ġprovid es",
+      "Ġindic ate",
+      "Ġset s",
+      "Ġtechn ique",
+      "Ġm ac",
+      "Ġhyp ot",
+      "Ġat ten",
+      "Ġev ent",
+      "Ġst age",
+      "Ġn ode",
+      "Ġref erence",
+      "Ġup per",
+      "Ġtechn iques",
+      "Ġgre ater",
+      "Ġdirect ly",
+      "Ġare as",
+      "Ġdis s",
+      "h or",
+      "ĠP ol",
+      "Ġevalu ation",
+      "Ġpattern s",
+      "ĠA bstract",
+      "Ġvir us",
+      "ve y",
+      "P C",
+      "Ġw omen",
+      "ri ent",
+      "Ġplas ma",
+      "Ġpro duced",
+      "ĠÎ µ",
+      "Ġanalys es",
+      "ĠS ub",
+      "Ġset ting",
+      "Ġmom ent",
+      "Ġtherm al",
+      "Ġoptim al",
+      "Ġtak en",
+      "Ġrec ogn",
+      "Ġvari ation",
+      "ĠL emma",
+      "Ġs us",
+      "f rak",
+      "ĠI L",
+      "Ġproced ure",
+      "h ood",
+      "Ġa im",
+      "ar ies",
+      "math frak",
+      "Ġpl ant",
+      "b rid",
+      "e lect",
+      "Ġvis ual",
+      "ur s",
+      "c ence",
+      "Ġf ive",
+      "Ġspati al",
+      "Ġrecept or",
+      "Ġindic ated",
+      "Ġ ess",
+      "Ġconsist ent",
+      "Ġt urn",
+      "tic es",
+      "Ġex ists",
+      "ect ors",
+      "Ġen zym",
+      "mer ic",
+      "Ġno ise",
+      "Ġgro und",
+      "Ġestim ated",
+      "el ine",
+      "Ġchann el",
+      "ti tion",
+      "Ġdiscuss ed",
+      "om er",
+      "ot es",
+      "Ġex act",
+      "ĠS ec",
+      "Ġt ake",
+      "Ġknow ledge",
+      "Ġpro p",
+      "Ġinf lamm",
+      "Ġdo uble",
+      "I t",
+      "Ġcon text",
+      "ĠM ed",
+      "M A",
+      "Ġf at",
+      "am s",
+      "d ata",
+      "and s",
+      "Ġcar di",
+      "ĠFurther more",
+      "oc y",
+      "Ġobserv ations",
+      "app ing",
+      "ĠIn f",
+      "om ial",
+      "Ġpubl ic",
+      "Ġem ploy",
+      "Ġre ason",
+      "y gen",
+      "Ġfollow ed",
+      "Ġam ount",
+      "Ġc ertain",
+      "wh ich",
+      "ot yp",
+      "ĠC ell",
+      "Ġch all",
+      "Ġpartic le",
+      "am bda",
+      "Ġ ens",
+      "Ġpe ople",
+      "a ult",
+      "ĠU nd",
+      "ĠB e",
+      "um in",
+      "roscop y",
+      "M R",
+      "l ation",
+      "Ġrep e",
+      "Ġab le",
+      "ĠS o",
+      "ĠâĪ ŀ",
+      "Ġen ti",
+      "Ġmo ve",
+      "Ġt rac",
+      "C O",
+      "Ġhe ter",
+      "Ġsp eed",
+      "Ġeffici ency",
+      "Ġop tical",
+      "Ġcomb ination",
+      "en ess",
+      "Ġc hem",
+      "L E",
+      "app a",
+      "Ġdecre ase",
+      "Î ¼",
+      "p ed",
+      "n ote",
+      "ĠM ulti",
+      "Ġal tern",
+      "Ġassum e",
+      "ĠF orm",
+      "str ic",
+      "qu e",
+      "Ġis s",
+      "ur rent",
+      "Ġpr inc",
+      "Ġt ask",
+      "op s",
+      "Ġwhere as",
+      "C H",
+      "Ġreveal ed",
+      "Ġcan not",
+      "ac tive",
+      "en z",
+      "Ġf ore",
+      "Ġoper ator",
+      "Ġcol um",
+      "at in",
+      "Ġorig inal",
+      "Ġsmall er",
+      "Ġmaterial s",
+      "h ydro",
+      "Ġcur ve",
+      "Ġselec tion",
+      "ak es",
+      "Ġex pos",
+      "at s",
+      "ĠÏ ī",
+      "Ġp ack",
+      "Ġst ability",
+      "Ġover all",
+      "Ġm orph",
+      "Ġmet ric",
+      "Ġo l",
+      "Ġb ar",
+      "ĠI N",
+      "I M",
+      "c y",
+      "et hyl",
+      "S P",
+      "Ġrespons es",
+      "anc y",
+      "Ġl ay",
+      "spec ific",
+      "Ġv s",
+      "ag ed",
+      "Ġs ocial",
+      "Ġc ut",
+      "I P",
+      "Ġlim ited",
+      "enc ies",
+      "Ġprot oc",
+      "Ġcompos ition",
+      "ĠThe y",
+      "Ġnum bers",
+      "m box",
+      "Ġdecre ased",
+      "v ec",
+      "R O",
+      "Aut hors",
+      "Ġth ick",
+      "Ġco ordin",
+      "Ġm es",
+      "Ġaff ect",
+      "Ġcl ose",
+      "Ġtrans port",
+      "C A",
+      "re te",
+      "c ome",
+      "Ġcollec ted",
+      "ĠF rom",
+      "Ġcontain s",
+      "ch it",
+      "ĠD et",
+      "Ġflu x",
+      "over y",
+      "e u",
+      "a ff",
+      "Ġconduc ted",
+      "Ġcr iter",
+      "Ġliter ature",
+      "Ġmem ory",
+      "Ġsequ ences",
+      "Ġp an",
+      "plic it",
+      "Ġtr ue",
+      "Ġmed ium",
+      "Ġd am",
+      "i re",
+      "c ell",
+      "L et",
+      "ef ul",
+      "ĠA meric",
+      "Ġn odes",
+      "get her",
+      "Ġto gether",
+      "T P",
+      "Ġrat her",
+      "Ġaut hors",
+      "Ġs ch",
+      "Ġprocess ing",
+      "Ġspect ra",
+      "Ġevalu ated",
+      "al k",
+      "Ġred uce",
+      "ĠH igh",
+      "ĠC ons",
+      "Ġcy cle",
+      "or n",
+      "i ers",
+      "Ġpro por",
+      "or ies",
+      "r ate",
+      "Ġh ost",
+      "o oth",
+      "y nt",
+      "Ġsour ces",
+      "Ġindividual s",
+      "Ġacc ount",
+      "ĠAl though",
+      "Ġcor rec",
+      "Ġpl an",
+      "enti ally",
+      "Ġdist inc",
+      "Ġso il",
+      "Ġse arch",
+      "Ġman agement",
+      "Ġvers ion",
+      "âĢ Ķ",
+      "Ġf ig",
+      "ĠN ote",
+      "Ġhe ad",
+      "dition al",
+      "Ġbu ild",
+      "ĠG l",
+      "as is",
+      "g roup",
+      "Ġdis play",
+      "ĠUn iversity",
+      "oot note",
+      "amet er",
+      "min ist",
+      "o pl",
+      "ym ph",
+      "L ambda",
+      "Ġidentif y",
+      "ĠSt ere",
+      "Ġï Ģ",
+      "Ġpro l",
+      "our ce",
+      "ic ial",
+      "Ġsim ulations",
+      "Ġth resh",
+      "p oint",
+      "ear ch",
+      "ell ing",
+      "ĠA cc",
+      "Ġframe work",
+      "Ġstreng th",
+      "ĠA b",
+      "tic les",
+      "Ġc os",
+      "F ootnote",
+      "r u",
+      "osp ital",
+      "Ġst able",
+      "Ġmo tion",
+      "Ġt ested",
+      "Ġt ests",
+      "as ter",
+      "l dots",
+      "C L",
+      "in ite",
+      "Ġspec ial",
+      "== ==",
+      "Ġappro aches",
+      "p ing",
+      "Ġcons um",
+      "S D",
+      "Ġj ust",
+      "k appa",
+      "Ġth ough",
+      "f aces",
+      "Ġra pid",
+      "ens ive",
+      "Ġnecess ary",
+      "Ġt ub",
+      "Ġfor ce",
+      "Ġbl ack",
+      "v olution",
+      "ĠAt om",
+      "ĠH ere",
+      "it ude",
+      "ens ions",
+      "ff er",
+      "r ich",
+      "Ġgiv es",
+      "Ġsh ape",
+      "Ġh ard",
+      "om p",
+      "Ġrepresent ation",
+      "l ing",
+      "ĠD ec",
+      "Ġnumer ical",
+      "Ġpl ace",
+      "Ġlead ing",
+      "Ġben ef",
+      "Ġreg ular",
+      "Ġclust er",
+      "Ġrel atively",
+      "Ġper cent",
+      "Ġaut om",
+      "Ġsym pt",
+      "ib ri",
+      "c hes",
+      "hen yl",
+      "c ar",
+      "Ġill ustr",
+      "por ts",
+      "em ic",
+      "Ġg ive",
+      "Ġcon ven",
+      "lec tion",
+      "ĠĠĠĠĠĠĠĠ ĠĠĠĠ",
+      "ĠA nd",
+      "Ġf ood",
+      "m ic",
+      "ograph ic",
+      "Ġc heck",
+      "Ġab ility",
+      "iqu id",
+      "Ġsub str",
+      "ĠâĪ Ĩ",
+      "Ġed ge",
+      "ĠP D",
+      "Ġclass ification",
+      "Ġsurv ival",
+      "ĠC al",
+      "er ate",
+      "Ġus eful",
+      "Ġcar ried",
+      "Ġint ensity",
+      "H E",
+      "oc enter",
+      "Ġpath way",
+      "Ġdef inition",
+      "Ġschem e",
+      "Ġsub sequ",
+      "ĠF irst",
+      "Ġcon sequ",
+      "ĠD iff",
+      "Ġinhib it",
+      "Ġam plit",
+      "as er",
+      "ĠN etwork",
+      "n ormal",
+      "ĠS T",
+      "Ġsol id",
+      "per im",
+      "com es",
+      "Ġcy t",
+      "od ies",
+      "I F",
+      "ra di",
+      "Ġm or",
+      "Ġc ore",
+      "B S",
+      "**** ****",
+      "Ġsoft ware",
+      "ĠG u",
+      "i red",
+      "id ent",
+      "Ġdiffic ult",
+      "us e",
+      "Ġadd ed",
+      "le y",
+      "Ġcaus ed",
+      "g ence",
+      "Ġb ase",
+      "## ##",
+      "ogen ic",
+      "f rom",
+      "Ġstat us",
+      "Ġassoci ation",
+      "ĠStere ocenter",
+      "Ġg alax",
+      "N O",
+      "angu age",
+      "Ġd imension",
+      "ogen esis",
+      "Ġem ission",
+      "Ġde ath",
+      "ul in",
+      "Ġag re",
+      "t urb",
+      "n abl",
+      "por al",
+      "Ġp or",
+      "Ġcomb ined",
+      "Ġalgorithm s",
+      "C s",
+      "Ġsens itivity",
+      "Ġallow s",
+      "Ġcapac ity",
+      "vers ion",
+      "Ġre stric",
+      "rom e",
+      "Ġexpos ure",
+      "h y",
+      "ann ing",
+      "Ġob ject",
+      "Ġc ode",
+      "f l",
+      "ro duction",
+      "res ents",
+      "r up",
+      "Ġte xt",
+      "ĠM at",
+      "Ġlead s",
+      "Ġres on",
+      "Ġproduc ts",
+      "Ġwh ole",
+      "Ġmat ter",
+      "P hi",
+      "op t",
+      "enc ing",
+      "ffici ents",
+      "n a",
+      "pec ially",
+      "Ġh aving",
+      "rop y",
+      "Ġunc ertain",
+      "en ari",
+      "r ical",
+      "Ġmin im",
+      "Ġorig in",
+      "u per",
+      "ĠN on",
+      "Ġevalu ate",
+      "Pro of",
+      "c ap",
+      "Ġsignal ing",
+      "Ġpolym er",
+      "tic ally",
+      "it ten",
+      "an tit",
+      "Ġus er",
+      "le vel",
+      "Ġmeas ures",
+      "Ġdynam ic",
+      "Ġmon ths",
+      "o ti",
+      "ra nd",
+      "Ġun til",
+      "Ġden ote",
+      "Ġnot e",
+      "Ġmain tain",
+      "Ġk in",
+      "sc ill",
+      "Ġim aging",
+      "Ġp ain",
+      "av y",
+      "Ġm it",
+      "ot he",
+      "Ġreg ul",
+      "kn own",
+      "Ġpl ot",
+      "nabl a",
+      "Ġf raction",
+      "w er",
+      "Ġstrateg y",
+      "Ġgre at",
+      "Ġdatas et",
+      "Ġun ique",
+      "C M",
+      "Ġt w",
+      "h an",
+      "ĠE u",
+      "and id",
+      "Ġback ground",
+      "Ġbro ad",
+      "il t",
+      "Ġimpro ved",
+      "Ġdiagn osis",
+      "i ous",
+      "Ġd ig",
+      "re m",
+      "er a",
+      "Ġex cl",
+      "Ġmet al",
+      "Ġs ix",
+      "Ġmin imum",
+      "us ions",
+      "e e",
+      "Ġcompound s",
+      "Ġas p",
+      "Ġe th",
+      "Ġdet ect",
+      "f erence",
+      "ĠÎ ·",
+      "Ġst atistical",
+      "ati ves",
+      "r is",
+      "Ġthe orem",
+      "ĠO F",
+      "w w",
+      "ar ily",
+      "ce ption",
+      "iv ing",
+      "Ġtest ing",
+      "Ġdiagn os",
+      "Ġrep resents",
+      "S igma",
+      "on ical",
+      "Ġequival ent",
+      "Ġbi om",
+      "Ġsub st",
+      "rain ts",
+      "ĠR ef",
+      "Ġsc ore",
+      "Ġd oc",
+      "Ġimpl ies",
+      "et er",
+      "Ġsynt hesis",
+      "il ibri",
+      "atter ing",
+      "C S",
+      "al se",
+      "Ġneu ro",
+      "Ġal though",
+      "ir us",
+      "met hyl",
+      "Ġtranscri ption",
+      "Ï Ģ",
+      "ĠMo lecular",
+      "Ġc ause",
+      "m ut",
+      "ĠI d",
+      "Î »",
+      "ad d",
+      "Ġpl ac",
+      "Ġag g",
+      "t ure",
+      "Ġl ack",
+      "Ġpredic tion",
+      "ra w",
+      "A n",
+      "Ġ ult",
+      "yn omial",
+      "Ġimmun e",
+      "il i",
+      "Ġpre p",
+      "Î ³",
+      "cl ass",
+      "Ġm ach",
+      "am ple",
+      "Ġres olution",
+      "Ġcou pling",
+      "se ud",
+      "Ġindic ates",
+      "Ġgener ation",
+      "Ġh ar",
+      "Ġf und",
+      "s cale",
+      "Ġe igen",
+      "ĠR el",
+      "ab or",
+      "ĠC H",
+      "e xt",
+      "am m",
+      "Ġcor rect",
+      "Ġsc reen",
+      "Ġstruct ural",
+      "Ġp H",
+      "Ġrele vant",
+      "Ġan gle",
+      "I G",
+      "Ġal gebra",
+      "hel ial",
+      "Ġw orld",
+      "Ġcur ves",
+      "ĠInt roduction",
+      "Ġth ird",
+      "Ġintro duced",
+      "B ig",
+      "n o",
+      "aus s",
+      "sub set",
+      "Ġtrans mission",
+      "Ġprof ile",
+      "ĠÎ ½",
+      "Ġes pecially",
+      "Ġatt rib",
+      "uc tion",
+      "Ġcoe fficients",
+      "Ġremain s",
+      "Ġne igh",
+      "os en",
+      "Ġrel i",
+      "Ġhig hest",
+      "Ġun iform",
+      "Ġf ar",
+      "chit ect",
+      "| |",
+      "Ġappro pri",
+      "ple x",
+      "ĠM ass",
+      "ogene ous",
+      "al es",
+      "Ġref er",
+      "Ġneed ed",
+      "Ġdifferen tial",
+      "ce ed",
+      "$ $",
+      "ynam ic",
+      "Ġse x",
+      "Ġspect ral",
+      "ch ar",
+      "P E",
+      "T S",
+      "Ġapproxim ately",
+      "val ue",
+      "Ġhal f",
+      "end ing",
+      "Ġgra di",
+      "Ġcoe fficient",
+      "ĠPh ys",
+      "Ġcon cer",
+      "Ġlab el",
+      "ir al",
+      "Ġchar ge",
+      "Ġox ygen",
+      "Ġde vi",
+      "Ġinter nal",
+      "Ġexp ans",
+      "lo ad",
+      "ĠS m",
+      "ran g",
+      "C on",
+      "ĠN a",
+      "Ġk e",
+      "Ġdi ab",
+      "ac hed",
+      "Ġloc ation",
+      "Ġvol t",
+      "ĠD isc",
+      "-- -",
+      "oc ytes",
+      "ore tical",
+      "Ġg ain",
+      "Ġmed i",
+      "ym pt",
+      "ot ed",
+      "ĠV al",
+      "Ġcommun ity",
+      "plement ary",
+      "Ġt ree",
+      "ĠT wo",
+      "Ġwh ose",
+      "Ġd one",
+      "am ine",
+      "Ġbi ological",
+      "in ks",
+      "Ġal most",
+      "Ġsl ight",
+      "Ġre pro",
+      "ģ Ħ",
+      "Ġthe rap",
+      "oc ation",
+      "Ġg ly",
+      "ĠE qu",
+      "Ġcol or",
+      "Ġn am",
+      "s ection",
+      "ĠE m",
+      "read y",
+      "H z",
+      "P D",
+      "f unction",
+      "ch ange",
+      "Ġprinc ip",
+      "Ġbec ome",
+      "ĠâĢ ĺ",
+      "Ġco ur",
+      "Ġloc ated",
+      "Ġr ang",
+      "in ity",
+      "Ġinter val",
+      "g in",
+      "Ġinvestig ate",
+      "f ree",
+      "Ġv itro",
+      "Ġsub set",
+      "Ġm ov",
+      "Ġpro ve",
+      "Ġl iver",
+      "ate gor",
+      "et es",
+      "Ġl ymph",
+      "d om",
+      "ĠE lect",
+      "Ġser um",
+      "Ġsc enari",
+      "end s",
+      "ĠF inally",
+      "Ġfil ter",
+      "I L",
+      "Ġab und",
+      "ment ation",
+      "im als",
+      "n um",
+      "enc ed",
+      "Ġproper ty",
+      "mat rix",
+      "ĠCom par",
+      "Ġl and",
+      "ĠCh ar",
+      "ress ive",
+      "ul us",
+      "Ġb one",
+      "E x",
+      "Ġradi ation",
+      "Ġsugg ested",
+      "ĠCom put",
+      "Ġthresh old",
+      "ĠA D",
+      "Ġh or",
+      "Ġin duc",
+      "Ġapproxim ation",
+      "Ġad minist",
+      "Ġor d",
+      "Ġl ung",
+      "Ġrece ived",
+      "Ġn orm",
+      "Ġestim ates",
+      "Ġl aw",
+      "Ġout comes",
+      "ĠP r",
+      "Ġdep th",
+      "Ġel se",
+      "Ġcontrib ution",
+      "he tic",
+      "Ġcons erv",
+      "Ġup on",
+      "Ġde ep",
+      "M D",
+      "Ġm el",
+      "Ġfil m",
+      "ilibri um",
+      "Ġo scill",
+      "ol ved",
+      "Ġbre ast",
+      "C P",
+      "ĠD ist",
+      "ric es",
+      "in ated",
+      "Ġoptim ization",
+      "Ġpredic ted",
+      "s f",
+      "d im",
+      "ĠS N",
+      "Ġav oid",
+      "Ġne ural",
+      "Ġw a",
+      "rop e",
+      "Ġdistrib utions",
+      "ox id",
+      "Ġsm ooth",
+      "p ath",
+      "Ġflu id",
+      "Ġs af",
+      "Ġcho ice",
+      "A A",
+      "Ġmolec ules",
+      "U S",
+      "Ġal ways",
+      "iv o",
+      "Ġreg ression",
+      "Ġsuccess ful",
+      "Ġw all",
+      "oun g",
+      "Ġactiv ities",
+      "Ġdepend ence",
+      "Ġrequi res",
+      "Ġpl ane",
+      "Ġdesign ed",
+      "P I",
+      "d own",
+      "Ġpop ulations",
+      "c or",
+      "medi ate",
+      "Ġd ose",
+      "Ġb ond",
+      "C o",
+      "ĠM an",
+      "Ġdiag ram",
+      "g s",
+      "Ġto ol",
+      "Ġisol ated",
+      "Ġvers us",
+      "ne y",
+      "Ġem erg",
+      "ĠA ut",
+      "a im",
+      "f ield",
+      "Ġexam ined",
+      "Ġs at",
+      "S M",
+      "ĠSp ec",
+      "Ġpar allel",
+      "is ation",
+      "Ġdistinc t",
+      "Ġpredic t",
+      "Ġf er",
+      "Ġunderstand ing",
+      "ĠSim ilar",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ",
+      "ud es",
+      "Ġo rient",
+      "h ic",
+      "u z",
+      "Ġmod ified",
+      "ĠâĪ ¼",
+      "F F",
+      "The re",
+      "Ġtri al",
+      "x y",
+      "ger y",
+      "Ġal ready",
+      "def ine",
+      "m ing",
+      "ĠS D",
+      "Ġmon itor",
+      "Ġp sy",
+      "Ġbec omes",
+      "ist ry",
+      "ĠÎ ĵ",
+      "Ġh um",
+      "ri er",
+      "ess ion",
+      "Ġhist ory",
+      "Ã ¶",
+      "ĠÎ ¾",
+      "Ġestabl ished",
+      "Ġachie ved",
+      "es tern",
+      "Ï Ĩ",
+      "ĠH ence",
+      "Ġassess ment",
+      "ot or",
+      "Ġdescrib e",
+      "och ond",
+      "yl ation",
+      "st s",
+      "sp ace",
+      "Ġdise ases",
+      "j ection",
+      "Ġs low",
+      "Ġnon linear",
+      "p ly",
+      "m l",
+      "Ġemb ed",
+      "com p",
+      "Ġeffici ent",
+      "Ġoper ation",
+      "Ġcont act",
+      "o z",
+      "Ġinv ari",
+      "Ġcent er",
+      "Ġcon c",
+      "wide tilde",
+      "Ġbe am",
+      "Ġclos ed",
+      "ĠMethod s",
+      "Ġch ronic",
+      "al ing",
+      "Ġse vere",
+      "Ġform s",
+      "il it",
+      "s ide",
+      "p en",
+      "Ġb ran",
+      "o ud",
+      "tal ity",
+      "Ġmap s",
+      "ac ts",
+      "O L",
+      "P R",
+      "Ġ Í",
+      "s l",
+      "Ġinst ance",
+      "ul ly",
+      "Ġestim ation",
+      "Ġpl ate",
+      "Ġdev ice",
+      "ĠI II",
+      "s in",
+      "Ġpl ants",
+      "it tle",
+      "Ġpro duce",
+      "Ġhe nce",
+      "Ġn ature",
+      "Ġrele ase",
+      "ĠM in",
+      "ric t",
+      "Ġconn ected",
+      "ott om",
+      "ell ar",
+      "Ġform ed",
+      "Ġm ob",
+      "Ġcomput ed",
+      "Ġ RE",
+      "Ġpol ynomial",
+      "Ġl iquid",
+      "g n",
+      "Ġass ay",
+      "Ġman if",
+      "ĠS i",
+      "re nce",
+      "Ġax is",
+      "V ID",
+      "Ġsign als",
+      "Î ¸",
+      "to k",
+      "d s",
+      "Ġrat s",
+      "Ġt or",
+      "o lecular",
+      "c hed",
+      "Ġdesc ri",
+      "Ġexp on",
+      "Ġper turb",
+      "Ġgl uc",
+      "Ġcolum n",
+      "U L",
+      "Ġmain ly",
+      "Ġm ul",
+      "id er",
+      "ĠC R",
+      "Ġc ataly",
+      "Ġl aser",
+      "tion ed",
+      "d en",
+      "Ġsugg ests",
+      "f ig",
+      "Ġprop ag",
+      "or g",
+      "re p",
+      "Ġcharacter ized",
+      "olog ies",
+      "Ġacc um",
+      "Ġv ary",
+      "Ġcontroll ed",
+      "Ġup d",
+      "ĠB r",
+      "Ġenti re",
+      "Ġ @",
+      "â ģĦ",
+      "Ġ Ì",
+      "Ġdat ab",
+      "an o",
+      "am il",
+      "Ġadj ust",
+      "y e",
+      "p ression",
+      "eren ces",
+      "Ġess ential",
+      "ĠH ydro",
+      "ĠT r",
+      "Ġappropri ate",
+      "Ġform ula",
+      "Ġlat tice",
+      "Ġac ute",
+      "Ġus ually",
+      "it able",
+      "Ġm ar",
+      "Ġμ m",
+      "ĠU SA",
+      "Ġinc ub",
+      "oc ks",
+      "Ġp epti",
+      "idd le",
+      "Ġdec om",
+      "Ġdam age",
+      "Ġgen ome",
+      "Ġm ouse",
+      "c irc",
+      "Ġlay ers",
+      "Ġt rack",
+      "Ġto x",
+      "Ġre plac",
+      "Ġad vant",
+      "iz on",
+      "Ġrecord ed",
+      "Ġst art",
+      "Ġr ank",
+      "s er",
+      "ĠG ene",
+      "auss ian",
+      "ing u",
+      "Ġconst raints",
+      "f low",
+      "Ġm ig",
+      "P L",
+      "Ġinc or",
+      "ap pro",
+      "Ġf ast",
+      "Ġmus cle",
+      "Ġh ome",
+      "e q",
+      "ĠÏ Ī",
+      "Ġstrong ly",
+      "ĠEu rope",
+      "Ġsub jects",
+      "Ġob jects",
+      "t est",
+      "t ered",
+      "ĠWh ile",
+      "Ġsymmet ry",
+      "Ġquanti f",
+      "` `",
+      "Ġbre ak",
+      "ĠEx perim",
+      "Ġmi xt",
+      "< <",
+      "ĠCh ina",
+      "ĠId entif",
+      "Ġaff ected",
+      "Ġsecond ary",
+      "Ġin equ",
+      "in cl",
+      "E G",
+      "F T",
+      "Ġfail ure",
+      "ec tiv",
+      "Ġk m",
+      "Ġsam pling",
+      "Ġexpans ion",
+      "Ġprac tice",
+      "u ations",
+      "ogn itive",
+      "Ġdi et",
+      "Ġtemper atures",
+      "Ġcontrol s",
+      "Ġch osen",
+      "Ġgener ally",
+      "anc er",
+      "Ġdeg rad",
+      "ul i",
+      "s m",
+      "othe rapy",
+      "Ġto wards",
+      "ĠProper ties",
+      "Ġclust ers",
+      "Ġdel ay",
+      "Ġhe p",
+      "P A",
+      "ĠStud y",
+      "antit ative",
+      "Ġclass ical",
+      "ĠZ h",
+      "ĠÎ ©",
+      "ĠB o",
+      "Ġse ed",
+      "ĠStr uct",
+      "Ġtre nd",
+      "i ological",
+      "Ġconfir med",
+      "Ġdistrib uted",
+      "b ial",
+      "Ġn ame",
+      "C N",
+      "val ence",
+      "er ior",
+      "iv en",
+      "n ed",
+      "Ġbehavi our",
+      "as ks",
+      "g ra",
+      "m ark",
+      "Ġerr ors",
+      "ĠR ep",
+      "l ight",
+      "cri pt",
+      "I f",
+      "Ġc andid",
+      "Ġdepend s",
+      "ĠN ational",
+      "Ġh olds",
+      "Ġprotoc ol",
+      "ĠUn ited",
+      "Ġinter face",
+      "Ġexp ect",
+      "Ġï ģ",
+      "ĠH IV",
+      "Ġro ot",
+      "Ġsc attering",
+      "w ords",
+      "Ġobserv ation",
+      "ot op",
+      "Ġoccur s",
+      "our ces",
+      "p ite",
+      "ĠS te",
+      "Ġor th",
+      "Ġst ain",
+      "Ġst eps",
+      "Ġcomp are",
+      "Ġbas ic",
+      "Ġinhib ition",
+      "Ġsympt oms",
+      "ĠHe alth",
+      "Ġpubl ished",
+      "f old",
+      "Ġt un",
+      "Ġv ivo",
+      "Ġrec onstr",
+      "Ġm RNA",
+      "ic y",
+      "Ġhy brid",
+      "y r",
+      "Ġm ixed",
+      "v is",
+      "Ch I",
+      "Ġmed ical",
+      "Ġf rag",
+      "Ġan imals",
+      "Ġimport ance",
+      "Ġeng ine",
+      "ĠC T",
+      "Ġpair s",
+      "Ġb al",
+      "ĠE ar",
+      "her s",
+      "Ġsy nd",
+      "Ġar chitect",
+      "Ġidentif ication",
+      "Ġstrateg ies",
+      "Ġreg ulation",
+      "ĠL a",
+      "r or",
+      "Ġflu ores",
+      "ur ity",
+      "Ġcon cept",
+      "Ġatten tion",
+      "Ġtrans formation",
+      "uc le",
+      "ĠRes earch",
+      "Ġsim pl",
+      "Ġcult ure",
+      "ar ing",
+      "if ically",
+      "p ir",
+      "z e",
+      "P T",
+      "m osp",
+      "Ġsw it",
+      "Ġn or",
+      "Ġenh ance",
+      "Ġenvironment al",
+      "r ary",
+      "ĠM icro",
+      "Ġw ide",
+      "op ath",
+      "au ge",
+      "z eta",
+      "Ġst e",
+      "ĠE l",
+      "Ġw ords",
+      "Ġnuc lear",
+      "Ġl anguage",
+      "Ġdetail s",
+      "op ar",
+      "ĠR ed",
+      "w ater",
+      "Ġc ategor",
+      "Ġf ile",
+      "Ġco ver",
+      "Ġachie ve",
+      "Ã ¡",
+      "um m",
+      "Ġl ig",
+      "Ġsur vey",
+      "Ġext ended",
+      "l ab",
+      "ĠIn c",
+      "Ġdis pers",
+      "Ġrecom m",
+      "ĠB ased",
+      "Ġabs ence",
+      "Ġconstruc tion",
+      "Ġpo or",
+      "Ġvolt age",
+      "Ġcell ular",
+      "Ġmor tality",
+      "Ġshow ing",
+      "Ġprol if",
+      "m p",
+      "Ġneur ons",
+      "Ġsup ported",
+      "Ġpre vent",
+      "el i",
+      "ox y",
+      "ic a",
+      "Ġf ully",
+      "Ġen ough",
+      "o times",
+      "ĠM R",
+      "Ġb ul",
+      "Ġphen omen",
+      "F A",
+      "Ġdec ision",
+      "Ġd ual",
+      "Ġdec ay",
+      "Ġo wn",
+      "Ġus es",
+      "Ġchall eng",
+      "Ġadd ress",
+      "O C",
+      "tiv ation",
+      "Ġm ill",
+      "Ġmod es",
+      "at us",
+      "ic tion",
+      "Ġabs orption",
+      "Ġep it",
+      "Ġconst ra",
+      "Ġagre ement",
+      "ĠA f",
+      "Ġbi as",
+      "ud ed",
+      "Ġpar ts",
+      "Ġv an",
+      "Ġcol on",
+      "Ġex ternal",
+      "Ġthe oretical",
+      "as i",
+      "Ġl es",
+      "abil ities",
+      "L A",
+      "tt ps",
+      "Ġinst ead",
+      "Ġmemb ers",
+      "+ +",
+      "Ġrec ently",
+      "Ġprep ared",
+      "Ġar ticle",
+      "d ay",
+      "Ġext ract",
+      "Ġâ İ",
+      "Ġpath ways",
+      "Ï Ħ",
+      "m id",
+      "or age",
+      "Ġcommun ication",
+      "Ġacc el",
+      "Ġun its",
+      "iti s",
+      "ynt hesis",
+      "Ġamplit ude",
+      "ri e",
+      "ult aneous",
+      "ĠL ear",
+      "ec ause",
+      "d o",
+      "e ff",
+      "Ġex plicit",
+      "Ġcriter ia",
+      "b re",
+      "Ġex ec",
+      "Ġmechan ical",
+      "er os",
+      "ĠCon cl",
+      "ĠE xt",
+      "Ġclass es",
+      "Ġlong er",
+      "Ġcalc ulations",
+      "eu tic",
+      "oci ated",
+      "ar di",
+      "Ġcour se",
+      "Ġpar tial",
+      "Ġsens or",
+      "Ï ĥ",
+      "Ġoper ators",
+      "ĠAmeric an",
+      "Ġm M",
+      "Ġv acc",
+      "oc c",
+      "ic on",
+      "Ġout come",
+      "Ġanal og",
+      "Ġthick ness",
+      "Ġre ach",
+      "Ġassum ed",
+      "end er",
+      "Ġm ale",
+      "S E",
+      "Ġint ra",
+      "Ġimplement ation",
+      "em ia",
+      "Ġenh anced",
+      "b ility",
+      "Ġeas ily",
+      "um p",
+      "Ġcar cin",
+      "os a",
+      "Ġcorrespond s",
+      "ne g",
+      "Ġmagn itude",
+      "con st",
+      "Ġl atter",
+      "Ġrepresent ed",
+      "Ġs ed",
+      "Ġparticular ly",
+      "Ġwr itten",
+      "par t",
+      "Ġo il",
+      "ber g",
+      "ĠB ar",
+      "Ġd ys",
+      "ĠS ome",
+      "ĠM ar",
+      "Ġaltern ative",
+      "ĠG erm",
+      "Ġgener ate",
+      "Ġcon struct",
+      "ian s",
+      "st ream",
+      "Ġe c",
+      "oc hemical",
+      "ib ration",
+      "oper ative",
+      "is ter",
+      "Ġrob ust",
+      "t re",
+      "Ġmodel ing",
+      "or ing",
+      "es e",
+      "d ed",
+      "ide o",
+      "Ġhydro gen",
+      "um ents",
+      "Ġdemonstr ate",
+      "Ġcorrel ated",
+      "Ġsystem atic",
+      "Ġsur gery",
+      "Ġindic ating",
+      "Ġhypot hesis",
+      "y ear",
+      "mit ted",
+      "Ġst ars",
+      "Ġprof iles",
+      "Ġcons ists",
+      "t ri",
+      "Ġdepend ent",
+      "ish ing",
+      "t op",
+      "Ġhe art",
+      "at ically",
+      "Ġinj ury",
+      "Ġqu ad",
+      "Ġwee ks",
+      "ut ing",
+      "ĠT e",
+      "Ġid enti",
+      "Ġgradi ent",
+      "Ġcalc ulation",
+      "Ġ ur",
+      "R T",
+      "z ation",
+      "Ġed uc",
+      "en ing",
+      "P P",
+      "z ed",
+      "us h",
+      "Ġcharacter istic",
+      "Ġstrain s",
+      "et h",
+      "Ġdi vers",
+      "âĪ Ī",
+      "oid s",
+      "ol ic",
+      "Ġinterp ret",
+      "K ey",
+      "Ġatt ack",
+      "p ective",
+      "Ġlab or",
+      "Ġmet ast",
+      "N F",
+      "Ġtiss ues",
+      "Ġradi us",
+      "ĠE ach",
+      "Ġc at",
+      "Ġd on",
+      "Ġele v",
+      "Ġass emb",
+      "r ons",
+      "Ġar bit",
+      "Ġpan el",
+      "Ġg rid",
+      "Ġt able",
+      "roscop ic",
+      "Ġc le",
+      "ĠIn tern",
+      "ob acter",
+      "Ġassum ption",
+      "ĠCO VID",
+      "Ġbound ed",
+      "Ġother s",
+      "Ġsch ool",
+      "Ġh ospital",
+      "lec ted",
+      "ĠC u",
+      "Ã Ĺ",
+      "Ġcomple t",
+      "Ġwid th",
+      "Ġl inks",
+      "p o",
+      "ol low",
+      "Ġn ut",
+      "Ġappear s",
+      "row n",
+      "a ro",
+      "Ġus ers",
+      "Ġcl im",
+      "Ġslight ly",
+      "Ġbl ue",
+      "ra b",
+      "ĠS er",
+      "Ġfig ure",
+      "ĠR ad",
+      "Ġelect ric",
+      "m m",
+      "och astic",
+      "ri ef",
+      "Ġcollec tion",
+      "Ġst em",
+      "Ġg over",
+      "Ġb ur",
+      "Ġtyp ical",
+      "s up",
+      "Ġagg reg",
+      "ra z",
+      "ĉĉ ĉ",
+      "Ġst ation",
+      "Ġar ter",
+      "i vely",
+      "itro gen",
+      "Ġcons tit",
+      "em pt",
+      "ĠEff ect",
+      "Ġdescri ption",
+      "Ġsc ores",
+      "Ġmet hyl",
+      "ĠO b",
+      "ĠSt ates",
+      "Ġs plit",
+      "ĠV ari",
+      "ĠW ang",
+      "Ġc ere",
+      "ĠF ran",
+      "Ġneed s",
+      "ĠF our",
+      "Ġpro ject",
+      "Ġdev ices",
+      "Ġintegr al",
+      "ĠE s",
+      "ymmet ric",
+      "Ġm ess",
+      "Ġpl ays",
+      "ĠLear ning",
+      "Ġover l",
+      "H ere",
+      "ign ment",
+      "Ġdel iver",
+      "ap an",
+      "C E",
+      "Ġg auge",
+      "ĠJ oh",
+      "-------- --------",
+      "Ġunder lying",
+      "Ġth in",
+      "Ġassess ed",
+      "Ġdiff usion",
+      "Ġhe ight",
+      "ĠS w",
+      "Ġd ark",
+      "pr int",
+      "ran ge",
+      "ĠC I",
+      "is es",
+      "l ier",
+      "r ant",
+      "om orphism",
+      "Ġcomp act",
+      "ip s",
+      "ĠN ame",
+      "Ġtechn ology",
+      "ag en",
+      "Ġconfig uration",
+      "Ġd uration",
+      "ĠCl ass",
+      "Ġp ut",
+      "Ġm aking",
+      "Ġas ympt",
+      "a id",
+      "Ġco h",
+      "Ġcomplex ity",
+      "Ġsec tions",
+      "ĠM D",
+      "ĠĠĠĠĠĠĠĠ Ġ",
+      "Ġra d",
+      "Ġsubstr ate",
+      "d d",
+      "Ġan n",
+      "Ġorgan ic",
+      "Ġtak ing",
+      "Ġinclud es",
+      "Ġk ine",
+      "a res",
+      "Ġro w",
+      "ateg ory",
+      "Ġmit ochond",
+      "U T",
+      "Ġsynd rome",
+      "ĠPro b",
+      "re tion",
+      "Ġfl uct",
+      "ĠD is",
+      "Ġtrans l",
+      "pl as",
+      "Ġpsy ch",
+      "Ġsur faces",
+      "Ġdetail ed",
+      "amil ton",
+      "Ġh old",
+      "ĠâĬ Ĺ",
+      "ĠC N",
+      "Ġd il",
+      "ĠO ver",
+      "at form",
+      "Ġver tical",
+      "Ġcomput ation",
+      "Ġp ure",
+      "Ġm akes",
+      "Ġexist ing",
+      "Ġexam ples",
+      "S O",
+      "ord ers",
+      "Ġm ix",
+      "Ġincor por",
+      "Ġre qu",
+      "an tic",
+      "D NA",
+      "Î ´",
+      "Ġcl oud",
+      "ĠT echn",
+      "Ġï ĥ",
+      "em ents",
+      "Ġbas eline",
+      "ste in",
+      "Ġbel ong",
+      "Ġtri als",
+      "Ġhor izon",
+      "Ġphosph or",
+      "Ġan s",
+      "di x",
+      "ro id",
+      "Ġapp ly",
+      "u ed",
+      "ern el",
+      "Ġfem ale",
+      "ic acy",
+      "Ġv ectors",
+      "Ġmat rices",
+      "at ric",
+      "ĠM c",
+      "Ġp y",
+      "Ġch lor",
+      "l en",
+      "Ġclear ly",
+      "st atic",
+      "re f",
+      "ĠS outh",
+      "Ġmed ia",
+      "ĠS he",
+      "ĠB ay",
+      "Ġag ents",
+      "B y",
+      "Ġdifferenti ation",
+      "ist ant",
+      "orph ic",
+      "Ġvari ety",
+      "Ġserv ice",
+      "Ġm apping",
+      "vel ength",
+      "Ġchann els",
+      "Ġcomp ute",
+      "Ġst ream",
+      "ul s",
+      "am ide",
+      "ok ing",
+      "v it",
+      "Ġyield s",
+      "om b",
+      "ĠG aussian",
+      "Ġp en",
+      "un e",
+      "Ġexper ience",
+      "b and",
+      "ĠD o",
+      "math sf",
+      "Ġallow ed",
+      "A r",
+      "R A",
+      "Ġbacter ial",
+      "Ġm iss",
+      "Ġbacter ia",
+      "Ġmoment um",
+      "Ġh ours",
+      "uc k",
+      "ĠPro position",
+      "ber t",
+      "ot rop",
+      "Ġvari ance",
+      "Ġtr ig",
+      "Ġsh ift",
+      "Ġequ ilibrium",
+      "b u",
+      "IN G",
+      "Ġwh ite",
+      "Ġk ind",
+      "Ġj oint",
+      "Ġtem poral",
+      "ĠI V",
+      "ĠAf ric",
+      "Ġsub ject",
+      "ĠP o",
+      "he ad",
+      "id el",
+      "Ġantib ody",
+      "ĠEff ects",
+      "Ġsp e",
+      "Ġsu fficient",
+      "j ected",
+      "re es",
+      "ĠT op",
+      "Ġmut ations",
+      "is ions",
+      "B C",
+      "Ġin duction",
+      "Ġinterest ing",
+      "ell a",
+      "c an",
+      "Ġsus p",
+      "ĠG roup",
+      "Ġextrac ted",
+      "istic ally",
+      "c oh",
+      "m ap",
+      "Ġaccur ate",
+      "Ġto o",
+      "Ġdim ensions",
+      "te gr",
+      "Ġgre en",
+      "ĠR o",
+      "Ġw ild",
+      "Ġlo op",
+      "Ġmet a",
+      "Ġsub stit",
+      "os ome",
+      "Ġsuggest ing",
+      "Ġspec im",
+      "am ental",
+      "im ent",
+      "Ġi j",
+      "Ġcl aim",
+      "Ġaut hor",
+      "Ġfil ms",
+      "Ġcoun ter",
+      "Ġconven tional",
+      "r in",
+      "otyp es",
+      "Ġp ast",
+      "S ince",
+      "medi ated",
+      "reat ment",
+      "Ġext ension",
+      "Ġbi o",
+      "Ġs ent",
+      "h al",
+      "Ġob jective",
+      "Ġar ray",
+      "Ġsu itable",
+      "ĠB ut",
+      "ĠH uman",
+      "or gan",
+      "b ut",
+      "mod el",
+      "S I",
+      "Ġhealth y",
+      "Ġv ac",
+      "Ġl ate",
+      "Ġr ing",
+      "Ġl ittle",
+      "M T",
+      "Ġsqu are",
+      "Ġge ometry",
+      "ĠT HE",
+      "ĠS ing",
+      "j ug",
+      "Ġstud ents",
+      ", ,",
+      "Ġad ult",
+      "Ġcharacter ization",
+      "Ġat mosp",
+      "Ġmonitor ing",
+      "an i",
+      "n et",
+      "ĠP a",
+      "opt osis",
+      "Ġcont in",
+      "ĠS ol",
+      "Ġdatab ase",
+      "im port",
+      "m ann",
+      "ĠPro cess",
+      "ĠC hen",
+      "Ġg ap",
+      "Ġenzym e",
+      "O T",
+      "Ġsim ultaneous",
+      "Ġexist ence",
+      "B P",
+      "ĠJ apan",
+      "oun ts",
+      "Ġt urb",
+      "Ġsp aces",
+      "ĠWe ight",
+      "oph il",
+      "Ġa st",
+      "Ġwr ite",
+      "Ġdiab etes",
+      "ĠC A",
+      "Ġneut ral",
+      "Ġvari ations",
+      "ax on",
+      "Ġbe gin",
+      "und er",
+      "Ġext raction",
+      "ĠP ati",
+      "Ġf ron",
+      "ef ined",
+      "Ġacid s",
+      "Ġserv ices",
+      "Ġs ense",
+      "Ġag ent",
+      "hen s",
+      "elect ric",
+      "val ues",
+      "Ġimprove ment",
+      "here nt",
+      "ac tic",
+      "Ġac et",
+      "cdot s",
+      "Ġam ino",
+      "Ġro om",
+      "Ġexp ress",
+      "Ġex cept",
+      "Ġo ld",
+      "pl ant",
+      "cep ti",
+      "ĠP CR",
+      "ĠE R",
+      "ĠB oth",
+      "ve x",
+      "Ġad ults",
+      "Ġp seud",
+      "Ġal le",
+      "Ġwork s",
+      "Ġconsum ption",
+      "ip her",
+      "c m",
+      "c ast",
+      "Ġnan opar",
+      "Ï ī",
+      "Ġe con",
+      "ynam ics",
+      "Ġal ter",
+      "Ġsk in",
+      "Ġdi ameter",
+      "G C",
+      "ĠS ign",
+      "v ial",
+      "Ġgluc ose",
+      "ĠN orth",
+      "ot ox",
+      "Ġpro te",
+      "d x",
+      "ĠC r",
+      "Ġf ract",
+      "Ġins ide",
+      "Ġst atic",
+      "w id",
+      "Ġst orage",
+      "ĠA L",
+      "ĠM ark",
+      "ĠA T",
+      "Ġsens itive",
+      "Ġad s",
+      "Ġed ges",
+      "an a",
+      "R e",
+      "Ġsumm ar",
+      "ĠAN D",
+      "Ġremain ing",
+      "dition ally",
+      "Ġm id",
+      "ĠThe ory",
+      "M C",
+      "Ġf lex",
+      "ol y",
+      "Ġdegrad ation",
+      "Ġint r",
+      "ot a",
+      "ism s",
+      "Ġam pl",
+      "ĠA re",
+      "Ġwork ing",
+      "Ġdivers ity",
+      "Ġt ensor",
+      "Ġb inary",
+      "\"\" \"",
+      "v als",
+      "Ġhe m",
+      "M L",
+      "Ġμ g",
+      "ne q",
+      "ens ities",
+      "Ġtak es",
+      "Ġch arg",
+      "Ġinter vention",
+      "Ġal b",
+      "Ġqu al",
+      "Ġmen tioned",
+      "Ġon es",
+      "ĠAcc ording",
+      "ill ed",
+      "O H",
+      "S up",
+      "Ġgalax ies",
+      "ail y",
+      "Ġr ule",
+      "Ġc ognitive",
+      "her n",
+      "Ġrecogn ition",
+      "Ġbu ffer",
+      "Ġm arg",
+      "ĠN i",
+      "ĠâĪ ļ",
+      "Ġcl in",
+      "Ġintegr ation",
+      "Ġs in",
+      "ĠAl so",
+      "Ġmach ine",
+      "w r",
+      "id ity",
+      "Ġsubsequ ent",
+      "F e",
+      "Ġn ames",
+      "at her",
+      "ĠC y",
+      "Ġmetabol ism",
+      "Ġre actions",
+      "Ġit er",
+      "Ġnot ed",
+      "Ġcaus es",
+      "ĠH amilton",
+      "g o",
+      "Ġra re",
+      "V A",
+      "ĠM y",
+      "v ol",
+      "as ure",
+      "Ġsignific ance",
+      "ĠN one",
+      "Ġve hic",
+      "S R",
+      "Ġvari ability",
+      "ĠDe velop",
+      "are n",
+      "Ġprom ot",
+      "ard s",
+      "Ġcomput ational",
+      "Ġsh all",
+      "iz ations",
+      "ĠHydro gen",
+      "Ġprolif eration",
+      "Ġcou pled",
+      "ch ron",
+      "Ġconver gence",
+      "Ġg ast",
+      "Ġcalc ulate",
+      "ra ft",
+      "par ation",
+      "her ic",
+      "ĠP C",
+      "pl ate",
+      "p tions",
+      "ĠAl gorithm",
+      "Ġresul ted",
+      "D E",
+      "Ġinvestig ation",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠ",
+      "ol ation",
+      "Ġt asks",
+      "Ġle g",
+      "in ess",
+      "Ġemploy ed",
+      "O n",
+      "Ġexper i",
+      "Ġtra ject",
+      "G A",
+      "Ġpur pose",
+      "ĠN um",
+      "Ġcomplet ely",
+      "th at",
+      "ĠOp tim",
+      "Ġform al",
+      "ec k",
+      "ĠPro tein",
+      "Ġgo al",
+      "Ġthrough out",
+      "Ġconsider ing",
+      "Ġref lect",
+      "tre ated",
+      "or ation",
+      "rib ution",
+      "Ġtherap eutic",
+      "Ġfind ing",
+      "U N",
+      "T hen",
+      "il ities",
+      "Ġun known",
+      "ove red",
+      "Ġver tex",
+      "Ġex change",
+      "Ġdrug s",
+      "ĠC P",
+      "Ġin str",
+      "Ġsymmet ric",
+      "ĠD ep",
+      "Ġconstruc ted",
+      "Ġpre valence",
+      "Ġdecre ases",
+      "Ġmi R",
+      "Ġy et",
+      "Ġb ox",
+      "g raph",
+      "wide hat",
+      "al ian",
+      "u fact",
+      "L R",
+      "cri ption",
+      "Ġn p",
+      "ĠChar acter",
+      "Ġep id",
+      "Î ½",
+      "Ġst ages",
+      "Ġs ay",
+      "ĠD uring",
+      "at ur",
+      "i entif",
+      "ab ric",
+      "Ã ¼",
+      "am ent",
+      "in ations",
+      "Ġsol ar",
+      "Ġdisc rete",
+      "ĠE r",
+      "ĠGen eral",
+      "b al",
+      "ĠC ent",
+      "u el",
+      "Ġmixt ure",
+      "Ġwid ely",
+      "ĠSec ond",
+      "Ġres ources",
+      "ĠAp pro",
+      "ĠI R",
+      "Ġstr ing",
+      "op ro",
+      "Ġin ner",
+      "ĠCom plex",
+      "O P",
+      "Ġat oms",
+      "Ġph ases",
+      "Ġdomain s",
+      "ad a",
+      "Ġcount ries",
+      "ac et",
+      "oci ation",
+      "iz er",
+      "Ġits elf",
+      "Ġmin imal",
+      "ĠCont rol",
+      "tt p",
+      "Ġb ottom",
+      "b all",
+      "ĠM ay",
+      "de v",
+      "n ow",
+      "em ber",
+      "Ġpercent age",
+      "ĠO ther",
+      "om as",
+      "Ġl ed",
+      "R es",
+      "ĠEn g",
+      "k g",
+      "Ġfrequ encies",
+      "k in",
+      "Ġinc idence",
+      "Ġan imal",
+      "Ġad op",
+      "Ġidenti ty",
+      "ĠR T",
+      "Ġy oung",
+      "ist ent",
+      "we ight",
+      "g u",
+      "Ġse ason",
+      "Ġexplain ed",
+      "ĠUnd er",
+      "io tic",
+      "w ell",
+      "Ġmetabol ic",
+      "g ical",
+      "Â ±",
+      "The orem",
+      "ad es",
+      "plic ated",
+      "Ġcontain ed",
+      "Ġs ulf",
+      "Ġco ol",
+      "Ġpers on",
+      "Ï ģ",
+      "Ġp ix",
+      "ĠS al",
+      "l ink",
+      "in i",
+      "t ual",
+      "S H",
+      "g ed",
+      "k y",
+      "as ts",
+      "erc ise",
+      "ĠH ar",
+      "Ġrel ax",
+      "equ iv",
+      "Ġy our",
+      "Ġund erg",
+      "Ġrec overy",
+      "Ġcom m",
+      "Ġden otes",
+      "form ed",
+      "ari a",
+      "e tic",
+      "Ġtum ors",
+      "ĠH y",
+      "Ġmark ers",
+      "Ġplac ed",
+      "ol ute",
+      "Ġw aves",
+      "Ġuncertain ty",
+      "Ġcontrib ute",
+      "ĠH ist",
+      "Ġa ver",
+      "Ġf av",
+      "Ġp ow",
+      "ĠSe e",
+      "Ġte am",
+      "Ġscal es",
+      "ientif ic",
+      "ier arch",
+      "Ġear lier",
+      "Ġsatisf ies",
+      "Ġcryst al",
+      "Ġpre gn",
+      "Ġobs erve",
+      "Ġon line",
+      "Ġcontrib utions",
+      "og ram",
+      "ĠM a",
+      "Ġf rac",
+      "Ġsp read",
+      "Ġon ce",
+      "d et",
+      "Ġresp ond",
+      "Ġpl atform",
+      "Ġinflamm atory",
+      "u tive",
+      "ĠS umm",
+      "pl ace",
+      "Ġ ions",
+      "Ġwind ow",
+      "ax is",
+      "est inal",
+      "Ġdepend ing",
+      "Ġsepar ation",
+      "Ġfor ward",
+      "ĠT i",
+      "Ġgl ass",
+      "Ġac cept",
+      "Ġfeed back",
+      "Ġon to",
+      "M E",
+      "mer c",
+      "unc tional",
+      "Ġap optosis",
+      "ĠProper ty",
+      "Ġintegr ated",
+      "Ġor b",
+      "Ġdevi ation",
+      "Ġantib odies",
+      "Ġremov ed",
+      "Ġlip id",
+      "arm ac",
+      "Ġarbit rary",
+      "ag ger",
+      "Ġemb ry",
+      "Ġg rain",
+      "Ġd rop",
+      "Ġstar ting",
+      "Ġrelationship s",
+      "ĠÏ ĩ",
+      "S F",
+      "Ġsim ply",
+      "Ġfac ilit",
+      "Ġz one",
+      "il s",
+      "Ps i",
+      "Ġinequ ality",
+      "Key words",
+      "Ġto ler",
+      "ed ge",
+      "Ġeas y",
+      "Ġal pha",
+      "Ġper f",
+      "wid th",
+      "in it",
+      "Ġimplement ed",
+      "C F",
+      "os ity",
+      "ocy te",
+      "Ġpropor tion",
+      "re st",
+      "ĠS uper",
+      "Ġpre f",
+      "Ġw ord",
+      "e v",
+      "Ġext ent",
+      "Ġinj ection",
+      "all ed",
+      "ĠAn ti",
+      "Ġb eta",
+      "ĠJ an",
+      "ĠG a",
+      "ĠZh ang",
+      "Ġ iron",
+      "Ġqu antitative",
+      "ro c",
+      "Ġf all",
+      "Ġregard ing",
+      "Ġf ix",
+      "Ġdatas ets",
+      "Ġt end",
+      "Ġscal ar",
+      "Ġresid ual",
+      "Ġrati os",
+      "ĠÎ ¦",
+      "k ing",
+      "Ġinflamm ation",
+      "Ġsing ular",
+      "ĠP ark",
+      "om atic",
+      "unc tions",
+      "Ġw ar",
+      "Í Ĵ",
+      "hem at",
+      "Ġf ace",
+      "ĠH u",
+      "Ġfund amental",
+      "Ġwa velength",
+      "el ing",
+      "ĠS uch",
+      "RNA s",
+      "c t",
+      "Ġid en",
+      "ce an",
+      "ne w",
+      "T ype",
+      "ĠForm ula",
+      "Ġmed ic",
+      "uss ion",
+      "Ġdist ingu",
+      "Ġreson ance",
+      "AT ION",
+      "ine ar",
+      "Ġh yd",
+      "l n",
+      "â ĨĴ",
+      "ĠU p",
+      "Ġact ual",
+      "Ġadap t",
+      "hen e",
+      "Ġm otor",
+      "l ist",
+      "ab it",
+      "I nd",
+      "ot al",
+      "Ġneigh bor",
+      "ĠP T",
+      "gen er",
+      "Ġposs ibility",
+      "erg ies",
+      "Ġse ems",
+      "ĠU S",
+      "Ġim m",
+      "Ġtyp ically",
+      "Ġsim ulated",
+      "ĠSystem s",
+      "ectiv eness",
+      "ry ing",
+      "Ġkin ase",
+      "Ġdecom position",
+      "ater al",
+      "Ġrot ation",
+      "pen dix",
+      "en n",
+      "at t",
+      "v ate",
+      "Ġtarget s",
+      "Ġsit uation",
+      "Ġinvol ve",
+      "Ġcre ated",
+      "hes ized",
+      "Ġal one",
+      "c i",
+      "Ġm L",
+      "Ġdiv ided",
+      "Ġbul k",
+      "o in",
+      "H C",
+      "Ġa rm",
+      "L O",
+      "ill s",
+      "Ġmed ian",
+      "h am",
+      "im er",
+      "f lu",
+      "Ġfib er",
+      "ĠS U",
+      "f ile",
+      "tiv ated",
+      "Ġradi o",
+      "ĠN ames",
+      "p e",
+      "Ġo ste",
+      "Ġel im",
+      "Ġsus cepti",
+      "re hens",
+      "Ġdiscuss ion",
+      "ĠS ep",
+      "Ġarchitect ure",
+      "Ġd est",
+      "t yp",
+      "r ame",
+      "Ġpar tition",
+      "Ġoccur red",
+      "Ġs izes",
+      "cl es",
+      "Ġsc hed",
+      "M olecular",
+      "ĠÎ º",
+      "Ġinv as",
+      "c up",
+      "P CR",
+      "ĠS MILES",
+      "ti ally",
+      "ox ide",
+      "ĠE d",
+      "Ġman ufact",
+      "ĠM aterial",
+      "Ġfl at",
+      "Ġmut ation",
+      "Ġintro duce",
+      "b ound",
+      "Ġdis orders",
+      "reg ulated",
+      "ĠM or",
+      "Ġf alse",
+      "ing er",
+      "ĠT R",
+      "Ġext rem",
+      "w ar",
+      "Ġsym bol",
+      "Ġan omal",
+      "ĠA R",
+      "Ġiss ues",
+      "Ġcoordin ates",
+      "Ġrecept ors",
+      "Ġprog ression",
+      "ĠF l",
+      "ubl ic",
+      "Ġelectron ic",
+      "Ġasp ects",
+      "Ġdoc ument",
+      "f lo",
+      "ĠP red",
+      "Ġgraph s",
+      "Ġtra ditional",
+      "D M",
+      "Ġsaf ety",
+      "ĠD r",
+      "ĠS equ",
+      "Ġcompos ite",
+      "ĠÎ Ľ",
+      "Ġrespons ible",
+      "Ġg ran",
+      "Ġinter mediate",
+      "od ium",
+      "pos ite",
+      "ph ase",
+      "d t",
+      "Ġwee k",
+      "Ġd os",
+      "Ġst abil",
+      "L C",
+      "ĠK ey",
+      "Ġver tices",
+      "Ġcomput er",
+      "ĠCan onical",
+      "Ġinvari ant",
+      "em ark",
+      "b enz",
+      "Ġ ice",
+      "ti le",
+      "z y",
+      "ĠO ut",
+      "Ġmove ment",
+      "Ġsh if",
+      "le ep",
+      "Ġd aily",
+      "Ġpos itions",
+      "Ġh im",
+      "Ġcre ate",
+      "O ur",
+      "Ġrese arc",
+      "Ġprog n",
+      "duc t",
+      "Ġscreen ing",
+      "Ġcho ose",
+      "pro cess",
+      "m al",
+      "Ġlabor atory",
+      "Ġoper ations",
+      "Ġto ols",
+      "olog ic",
+      "q quad",
+      "Ġcommon ly",
+      "Ġv oid",
+      "Ġocc up",
+      "ass ociated",
+      "Ġcorrel ations",
+      "Ġcarcin oma",
+      "l in",
+      "Ġv ideo",
+      "Ġhe avy",
+      "Ġlarg est",
+      "Ġm iddle",
+      "ĉĉ ĉĉ",
+      "ĠB as",
+      "as ons",
+      "id ing",
+      "Ġet c",
+      "ac he",
+      "ĠE val",
+      "i ra",
+      "rom agnetic",
+      "Ġco vari",
+      "L I",
+      "Ġde le",
+      "Ġst ra",
+      "am ples",
+      "od er",
+      "Ġc ategory",
+      "ĠIn stit",
+      "Ġpol icy",
+      "B ased",
+      "ib ly",
+      "Ġdeterm ination",
+      "Ġresp ir",
+      "otrop ic",
+      "Ġol der",
+      "ĠM al",
+      "Ġcy tok",
+      "Ġdeg rees",
+      "a ut",
+      "ill ing",
+      "et ing",
+      "Ġreduc es",
+      "Ġide al",
+      "b inding",
+      "ĠSp ect",
+      "un it",
+      "Ġdi ver",
+      "ĠW orld",
+      "Ġmark ed",
+      "al y",
+      "Ġcomplex es",
+      "ĠSumm ary",
+      "Ġpro pose",
+      "ĠA ustr",
+      "Ġmax im",
+      "Ġro und",
+      "Ġinhib itor",
+      "Ġeff icacy",
+      "act or",
+      "b ur",
+      "Ġtrans f",
+      "ĠG al",
+      "Ġpro ved",
+      "ĠDef ined",
+      "A t",
+      "Ġse lect",
+      "Ġnanopar ticles",
+      "W h",
+      "k en",
+      "ĠS P",
+      "en ge",
+      "Ġdeliver y",
+      "Ġdis order",
+      "ĠIn ChI",
+      "ĠCompar ison",
+      "if ying",
+      "ĠM echan",
+      "Ġconcl ude",
+      "Ġrepe ated",
+      "ell ow",
+      "ĠÃ Ģ",
+      "C I",
+      "ĠH z",
+      "an alysis",
+      "T r",
+      "Ã Ń",
+      "eli hood",
+      "Ġexp and",
+      "ĠDevelop ment",
+      "ĠSt ate",
+      "Ġt et",
+      "ff ic",
+      "Ġp arent",
+      "Ġscenari o",
+      "r s",
+      "ĠW hat",
+      "â ī",
+      "Ġstim ulation",
+      "ĠO bs",
+      "z ero",
+      "Ġman ner",
+      "as hed",
+      "ĠL og",
+      "Ġox ide",
+      "ph osph",
+      "Ġmig ration",
+      "Ġsub group",
+      "ros is",
+      "ip p",
+      "D R",
+      "d ec",
+      "os omal",
+      "Ġseg ment",
+      "ogen ous",
+      "F P",
+      "h and",
+      "ĠSur face",
+      "it z",
+      "Ġcryst all",
+      "th is",
+      "Ġbuild ing",
+      "t ag",
+      "Ġreduc ing",
+      "Ġun s",
+      "Ġrecom b",
+      "Ġc am",
+      "Ġlim its",
+      "oc ardi",
+      "& &",
+      "Ġsepar ate",
+      "Ġsup plement",
+      "ke le",
+      "Ġgra d",
+      "Ġiss ue",
+      "ĠQu antum",
+      "Ġcurrent ly",
+      "Ġqu ite",
+      "E P",
+      "Ġr ules",
+      "Ġwe ights",
+      "u ary",
+      "ill i",
+      "Ġbec ame",
+      "Ã ³",
+      "Ġnormal ized",
+      "ĠNetwork s",
+      "erv ed",
+      "Ġstat istics",
+      "ĠT ime",
+      "ĠU V",
+      "Ġc av",
+      "us ed",
+      "Ġf ish",
+      "Ġmajor ity",
+      "ĠP e",
+      "Ġcoh ort",
+      "Ġsem i",
+      "Ġg ame",
+      "mon ary",
+      "M M",
+      "od ed",
+      "Ġv ent",
+      "Ġaut o",
+      "Ġabund ance",
+      "n ov",
+      "Ġasympt otic",
+      "Ġtreat ments",
+      "ul y",
+      "Ġconstra int",
+      "Ġbe y",
+      "ĠS O",
+      "Ġst d",
+      "Ġdevelop ing",
+      "ĠN ot",
+      "L emma",
+      "Ġapp arent",
+      "Ġcirc uit",
+      "F rom",
+      "ĠEurope an",
+      "Ġsol ve",
+      "ĠÍ ij",
+      "u x",
+      "Ġbey ond",
+      "ep t",
+      "Ġapp e",
+      "requ ency",
+      "Ġvac u",
+      "ĠInd eed",
+      "ĠC hemical",
+      "ĠUnd efined",
+      "N ote",
+      "Ġn ull",
+      "Ġin verse",
+      "Ġnam ely",
+      "Ġshe ar",
+      "m L",
+      "A ll",
+      "R ec",
+      "Ġgeneral ized",
+      "ran es",
+      "ĠT est",
+      "il ing",
+      "Ġfluores cence",
+      "ĠÎ £",
+      "Ġind epend",
+      "d iff",
+      "Ġprovid ing",
+      "p henyl",
+      "h ing",
+      "Ġvir al",
+      "ĠB ecause",
+      "Ġint rac",
+      "ĠH ig",
+      "Ġw ant",
+      "Ġprincip le",
+      "an ol",
+      "Ġh a",
+      "ov ascular",
+      "Ġform er",
+      "Ġestabl ish",
+      "Ġadvant age",
+      "II I",
+      "Ġsequ encing",
+      "Ġproced ures",
+      "t ra",
+      "in dex",
+      "f e",
+      "Ġp i",
+      "Ġob vious",
+      "Ġreg ime",
+      "s ur",
+      "Ġpres ents",
+      "Ġdis plac",
+      "Ġdec l",
+      "ĠAp pendix",
+      "Ġinter act",
+      "land s",
+      "in ate",
+      "om orphic",
+      "Ġlow est",
+      "Ġar tif",
+      "Ġinvol ving",
+      "Ġcom merc",
+      "Ġd op",
+      "Ġcon form",
+      "ĠI g",
+      "rol og",
+      "v ised",
+      "Ġfl o",
+      "Ġcardi ac",
+      "p ts",
+      "r ig",
+      "Ġens ure",
+      "Ġaccum ulation",
+      "Ġent ropy",
+      "Ġide a",
+      "per ature",
+      "Ġques tions",
+      "ĠP R",
+      "Ġstat istically",
+      "d agger",
+      "Ġn itrogen",
+      "sc r",
+      "ĠDisc ussion",
+      "Ġre ports",
+      "Ġpul se",
+      "Ġrequire ments",
+      "Ġcompar ing",
+      "qui red",
+      "l ayer",
+      "Ġspect roscopy",
+      "viron ments",
+      "Ġscal ing",
+      "Ġex posed",
+      "M B",
+      "Î ¾",
+      "Ġh ole",
+      "Ġ á",
+      "Ġsimilar ity",
+      "Ġvari ants",
+      "b ody",
+      "Ġke ep",
+      "ĠC ancer",
+      "ed i",
+      "os omes",
+      "Ç «",
+      "A d",
+      "âĪ ŀ",
+      "mon ic",
+      "g ing",
+      "s plit",
+      "kn ow",
+      "Ġro ugh",
+      "hemat ical",
+      "v ision",
+      "Ġd ed",
+      "Ġcycl es",
+      "Ġfam il",
+      "Ġadminist ration",
+      "et al",
+      "Ġcor on",
+      "Ġinf ections",
+      "Ġmac roph",
+      "atic s",
+      "Ġpredic tions",
+      "is her",
+      "ere nt",
+      "re ted",
+      "incl ude",
+      "Ġclim ate",
+      "s ec",
+      "==== ====",
+      "ĠM S",
+      "Ġcomp e",
+      "r atic",
+      "l ig",
+      "pos es",
+      "Ġpolar ization",
+      "ll ip",
+      "der ived",
+      "Ġrele ased",
+      "Ġconn ection",
+      "l ic",
+      "Ġcol i",
+      "Ġout side",
+      "Ġabs olute",
+      "es ian",
+      "ĠE nd",
+      "ĠO f",
+      "Ġiden tical",
+      "Ġmod ule",
+      "Ġmitochond rial",
+      "Ġadv anced",
+      "ing ly",
+      "form ance",
+      "Ġto ward",
+      "ud ing",
+      "e k",
+      "Ġmean ing",
+      "c rib",
+      "ul ator",
+      "F N",
+      "k ey",
+      "c ons",
+      "Ġapp lying",
+      "is hes",
+      "Ġm amm",
+      "Ġderiv atives",
+      "Ġorient ation",
+      "Ġst ochastic",
+      "ĠA ug",
+      "Ġre nal",
+      "ĠG reen",
+      "Ġcomple ment",
+      "ob l",
+      "pir ical",
+      "or ts",
+      "B M",
+      "Ġex cess",
+      "Ġmorph ology",
+      "Ġs ound",
+      "if ier",
+      "Ġim plications",
+      "ĠDes ign",
+      "appro x",
+      "pro p",
+      "Ġcandid ate",
+      "Ġde pos",
+      "Ġequ ip",
+      "ust ain",
+      "ines e",
+      "et ry",
+      "Ġpot entially",
+      "Ġstra ight",
+      "Ġcr uc",
+      "i ology",
+      "Ġk ernel",
+      "Ġal coh",
+      "idd en",
+      "ret urn",
+      "Ġcorrec tion",
+      "ro t",
+      "Ġmic roscopy",
+      "Ġf oot",
+      "G L",
+      "ĠCell s",
+      "ir th",
+      "y g",
+      "ĠP ath",
+      "out hern",
+      "ĠL ong",
+      "Ġre vers",
+      "Î µ",
+      "ar se",
+      "Ġcere b",
+      "ist ed",
+      "Ġpul s",
+      "Ġdis k",
+      "it ud",
+      "Ġd u",
+      "Ġang ular",
+      "c hem",
+      "l ength",
+      "Ġexact ly",
+      "ro ke",
+      "ut h",
+      "Ġcon d",
+      "ins ic",
+      "Ġr ise",
+      "t ake",
+      "Ġtop ological",
+      "Ġrem ark",
+      "oll ary",
+      "Ġc er",
+      "T E",
+      "n ment",
+      "Ġbu ilt",
+      "Ġf re",
+      "Ġen ergies",
+      "ect ing",
+      "ĠT em",
+      "ra red",
+      "ĠN ow",
+      "ch arge",
+      "Ġloc ations",
+      "Ġbal ance",
+      "Ġl a",
+      "Ġre ached",
+      "lamm atory",
+      "Ġf abric",
+      "ific ations",
+      "Ġdiagnos tic",
+      "Ġmut ant",
+      "ĠN O",
+      "H D",
+      "ĠA B",
+      "Ġdisc rim",
+      "Ġprec ip",
+      "ĠTh ree",
+      "Ġins er",
+      "Ġinf ected",
+      "Ġconst ants",
+      "Î ©",
+      "neg ative",
+      "Ġconf idence",
+      "ĠPati ents",
+      "ollow ing",
+      "ad s",
+      "Ġhyper t",
+      "ĠIntern ational",
+      "D ef",
+      "ari ate",
+      "Ġinter vals",
+      "Ġex ercise",
+      "Ġeduc ation",
+      "Ġremov al",
+      "ther n",
+      "st er",
+      "Ġinte ger",
+      "ĠP A",
+      "Ġk id",
+      "Ġcategor ies",
+      "ĠG iven",
+      "Ġv ascular",
+      "here nce",
+      "math scr",
+      "ĠR et",
+      "Ġins ulin",
+      "tic ip",
+      "ĠC F",
+      "Ġlo ok",
+      "ymmet ry",
+      "Ġfor ces",
+      "ĠPhys ical",
+      "L S",
+      "c are",
+      "Ġh ouse",
+      "Ġind uce",
+      "Ġbel ie",
+      "ri a",
+      "ĠAs sum",
+      "Ġcomput ing",
+      "Ġb us",
+      "âĪ İ",
+      "Ġprac tical",
+      "t rain",
+      "T T",
+      "Ġpl astic",
+      "ĠN or",
+      "Ġfe as",
+      "ĠHamilton ian",
+      "Ġt ail",
+      "ĠZ n",
+      "Ġinterpret ation",
+      "duc ing",
+      "I s",
+      "Ġexam ine",
+      "ul ates",
+      "Ġmat ch",
+      "Ġ Ä",
+      "iv es",
+      "amet ers",
+      "Ġμ M",
+      "Ġexhib it",
+      "Ġn it",
+      "ot o",
+      "ĠCl inical",
+      "erv ation",
+      "ĠAd ditionally",
+      "ar ant",
+      "Ġel astic",
+      "D A",
+      "otop ic",
+      "Ġactiv ated",
+      "Ġt er",
+      "Ġconsequ ence",
+      "Ġend ot",
+      "oph ag",
+      "Ġcompar able",
+      "Ġdom inant",
+      "Î ·",
+      "Ġvalid ation",
+      "I m",
+      "Ġ Å",
+      "Ġle af",
+      "Ġf ung",
+      "tain ing",
+      "Ġun ivers",
+      "Ġph yl",
+      "Ġl ibr",
+      "Ġext ra",
+      "Ġpr int",
+      "medi ately",
+      "Ġmax imal",
+      "id ae",
+      "Ġor al",
+      "b in",
+      "Ġpepti de",
+      "ĠM ax",
+      "ar p",
+      "Ġconcl usion",
+      "Ġsatisf y",
+      "Ġanalyz e",
+      "o is",
+      "Ġinf er",
+      "Ġd raw",
+      "Ġdep ression",
+      "Ġmet all",
+      "Ġpost erior",
+      "Ġpeak s",
+      "s ol",
+      "Ġhorizon tal",
+      "Ġlater al",
+      "ĠO R",
+      "N N",
+      "Ġem o",
+      "P V",
+      "T A",
+      "Ġincub ated",
+      "Ġret rie",
+      "Ġhum ans",
+      "Ġ ri",
+      "Ġs oci",
+      "on ia",
+      "Ġinter ven",
+      "Ġvary ing",
+      "Ġs ti",
+      "ĠIm mun",
+      "Ġon set",
+      "Ġle aves",
+      "Ġother wise",
+      "Ġbl ocks",
+      "Ġass igned",
+      "SC s",
+      "Ġbi os",
+      "Ġmix ing",
+      "ar a",
+      "l i",
+      "Ġde formation",
+      "Ġcost s",
+      "Ġper ipher",
+      "ĠT ra",
+      "Ġat omic",
+      "Ġrandom ly",
+      "Ġarg ument",
+      "Ġit ems",
+      "Ġsu ff",
+      "Ġprob ably",
+      "n ers",
+      "Ġinhibit ors",
+      "Ġbe h",
+      "ĠDe ep",
+      "Ġp ig",
+      "ĠT ype",
+      "ĠM ost",
+      "ur a",
+      "itud inal",
+      "Ġderiv ative",
+      "Ġexpl ore",
+      "ĠIn formation",
+      "Ġg rap",
+      "ĠÎ Ķ",
+      "Ġprog ress",
+      "******** ********",
+      "Ġ ul",
+      "AR S",
+      "or al",
+      "os tic",
+      "C om",
+      "ĠEx ternal",
+      "ĠSt atistical",
+      "ĠR am",
+      "ĠL o",
+      "Ġelect rical",
+      "l ong",
+      "N et",
+      "EN T",
+      "v a",
+      "Ã ¤",
+      "ur ations",
+      "Ġdes ired",
+      "ir ing",
+      "Ġphys ics",
+      "Ġmass es",
+      "k i",
+      "Ġband s",
+      "Ġal k",
+      "ĠSimilar ly",
+      "Ġsur round",
+      "Ġcon vex",
+      "ost er",
+      "Ġlink ed",
+      "Ġfocus ed",
+      "Ġh ot",
+      "Ġmat ching",
+      "Ġoxid ation",
+      "Ġan ten",
+      "m iss",
+      "Ġm ental",
+      "il le",
+      "ici ency",
+      "ĠLi u",
+      "Ġprob e",
+      "ĠEs tim",
+      "Ġindic es",
+      "c he",
+      "ĠR ob",
+      "Ġcon v",
+      "ĠV er",
+      "ap se",
+      "S i",
+      "ph al",
+      "Ġles ions",
+      "Ġmolec ule",
+      "Ġa di",
+      "Ġd ate",
+      "Ġcompos ed",
+      "Ġa ud",
+      "struct ure",
+      "ot on",
+      "in for",
+      "Ġclust ering",
+      "ac ent",
+      "st ar",
+      "P O",
+      "ĠCh inese",
+      "Ġspec ifically",
+      "eren tial",
+      "Ġcap ture",
+      "ĠL ow",
+      "Ġf ine",
+      "Ġfem ales",
+      "ĠH ow",
+      "Ġa er",
+      "v ector",
+      "port un",
+      "form s",
+      "z o",
+      "Ġprec ision",
+      "yp t",
+      "Ġmin utes",
+      "Î º",
+      "Ġoxid ative",
+      "con n",
+      "ens us",
+      "Ġtrac e",
+      "Ġcon jug",
+      "Ġhigh light",
+      "s s",
+      "ĠExperim ental",
+      "ĠTh at",
+      "art ment",
+      "M O",
+      "' '",
+      "omet er",
+      "Ġst op",
+      "Ġ rib",
+      "Ġout er",
+      "r h",
+      "ri pt",
+      "Ġfluct uations",
+      "ob s",
+      "n on",
+      "Ġqu ark",
+      "ĠÃ °",
+      "ĠM ac",
+      "Ġperiod s",
+      "roll ed",
+      "A V",
+      "ĠO c",
+      "ĠIm age",
+      "ĠB el",
+      "Ġpropag ation",
+      "ĠD on",
+      "ww w",
+      "gl ish",
+      "Ġexhib ited",
+      "ogene ity",
+      "ĠB ack",
+      "Ġac tions",
+      "sk i",
+      "ĠAm ong",
+      "Ġb rief",
+      "ri ers",
+      "ĠN F",
+      "pos itive",
+      "sequ ently",
+      "ul ence",
+      "Ġen vironments",
+      "Ġcur v",
+      "om ics",
+      "Ġb it",
+      "Ġg el",
+      "Ġrepresent ations",
+      "Ġa way",
+      "ĠF ield",
+      "ob ic",
+      "C G",
+      "Ġcomp rehens",
+      "Ġh ierarch",
+      "Ġinduc es",
+      "B D",
+      "Ġh app",
+      "Ġe ight",
+      "Ġgra vity",
+      "Ġadap tive",
+      "B L",
+      "gen ic",
+      "Ġin struc",
+      "Ġanaly tical",
+      "ĠO x",
+      "ĠC ON",
+      "Ġsur gical",
+      "Ġd ip",
+      "at o",
+      "Ġrandom ized",
+      "Ġro les",
+      "d ep",
+      "ĠâĪ ĩ",
+      "ch ang",
+      "Ġdispers ion",
+      "Ġsepar ated",
+      "ĠOr gan",
+      "ĠV i",
+      "ĠJoh n",
+      "Ġan not",
+      "Ġres ource",
+      "en ergy",
+      "rel ation",
+      "me an",
+      "ĠB en",
+      "Ġconfir m",
+      "W ith",
+      "Ġinf inite",
+      "ĠSc ience",
+      "Ġsuccessful ly",
+      "Ġlocal ization",
+      "m ode",
+      "h ttps",
+      "geb ras",
+      "idel ines",
+      "Ġeff ectiveness",
+      "h yd",
+      "Ġs aid",
+      "ic o",
+      "Ġtrans itions",
+      "ed ing",
+      "Ġprogram s",
+      "Ġmob ile",
+      "Ġim mediately",
+      "ec tivity",
+      "ĠThe rm",
+      "ogene tic",
+      "Ġse ven",
+      "Ġem ph",
+      "G E",
+      "ne um",
+      "Ġf usion",
+      "lim its",
+      "Ġcalc ium",
+      "ra f",
+      "min us",
+      "Ġt rap",
+      "Ġspecim ens",
+      "anc ing",
+      "ĠM arch",
+      "Ġt en",
+      "Ġfamil ies",
+      "ĠH D",
+      "is ons",
+      "Ġpre paration",
+      "h old",
+      "et her",
+      "ĠV ol",
+      "ĠD ise",
+      "Ġrun ning",
+      "Ġqual it",
+      "Ġeff ectively",
+      "ffici ently",
+      "B I",
+      "Ġden oted",
+      "ĠEqu ation",
+      "Ġdem and",
+      "it ory",
+      "ach ing",
+      "Ġs odium",
+      "Ġrepro duc",
+      "ch o",
+      "Ġb il",
+      "P i",
+      "um b",
+      "Ġreconstr uction",
+      "for ward",
+      "O ne",
+      "Ġcon version",
+      "Ġform ulation",
+      "Ġnear ly",
+      "ĠL ag",
+      "S tr",
+      "ter ior",
+      "Ġoper ating",
+      "and om",
+      "Ġmov ing",
+      "ĠRe view",
+      "// //",
+      "n ai",
+      "p p",
+      "oti de",
+      "lab el",
+      "oc occ",
+      "Ġne ver",
+      "ak er",
+      "Ġdig ital",
+      "B l",
+      "U n",
+      "Ġmem ber",
+      "s el",
+      "Ġpot enti",
+      "Ġcop y",
+      "Ġelect rons",
+      "ch lor",
+      "ann el",
+      "yl ind",
+      "Ġm is",
+      "ĠS et",
+      "Ġnut ri",
+      "Ġdescrib es",
+      "Ġassum ptions",
+      "Ġvir tual",
+      "Ġcoordin ate",
+      "Ġv or",
+      "ĠA rab",
+      "ĠIm p",
+      "Ġde position",
+      "Ġins tit",
+      "Ġrepresent ative",
+      "ever al",
+      "Ġmill ion",
+      "ĠM A",
+      "Ġmal es",
+      "Ġcruc ial",
+      "Ġcol d",
+      "Ġload ing",
+      "Ġtrans lation",
+      "Ġst ead",
+      "ra ys",
+      "Ġchall enge",
+      "ac tivity",
+      "id al",
+      "u ff",
+      "Ġse em",
+      "Ġn ational",
+      "Ġf a",
+      "Ġmin or",
+      "Ġunderg o",
+      "c r",
+      "Ġcap t",
+      "e le",
+      "up le",
+      "ĠM g",
+      "le ge",
+      "G R",
+      "Ġr ig",
+      "Ġar ri",
+      "Ġdet ector",
+      "Ġst rict",
+      "Ġad hes",
+      "Ġse a",
+      "the less",
+      "Ġs leep",
+      "ĠCom mun",
+      "Ġanti oxid",
+      "Ġmark er",
+      "Ġflow s",
+      "anc re",
+      "ĠJan uary",
+      "in put",
+      "U P",
+      "Ġst ored",
+      "ad ing",
+      "iti vely",
+      "Ġsl ope",
+      "Ġshe ll",
+      "Ġelev ated",
+      "il k",
+      "Ġfrequ ently",
+      "Ġb all",
+      "urb an",
+      "Ġm l",
+      "us ive",
+      "ĠA nt",
+      "am ino",
+      "S im",
+      "Ġphys iological",
+      "reg ulation",
+      "es ity",
+      "Ġexpl an",
+      "Ġad en",
+      "re me",
+      "Ġdiff er",
+      "Ġmod ification",
+      "Ġir radi",
+      "H e",
+      "ac ial",
+      "Ġsupp ress",
+      "qu is",
+      "Ġd ry",
+      "er ated",
+      "Ġpro jection",
+      "Ġpo ol",
+      "ple te",
+      "Ġdirec tions",
+      "Ġchang ed",
+      "ĠI ts",
+      "Ġst er",
+      "Ġradi al",
+      "Ġg r",
+      "Ġperiod ic",
+      "Ġb in",
+      "Ġp ip",
+      "m en",
+      "t hen",
+      "p c",
+      "am ily",
+      "ĠD M",
+      "Ġsed iment",
+      "m i",
+      "Ġclos ely",
+      "Ġrep air",
+      "Ġrespir atory",
+      "Ġh orm",
+      "A ns",
+      "d r",
+      "l s",
+      "Ġhom ogeneous",
+      "et ric",
+      "D S",
+      "Ġresid ues",
+      "ĠVal ue",
+      "F s",
+      "Ġwh y",
+      "S p",
+      "Ġc a",
+      "Ġn arrow",
+      "g ent",
+      "Ġb r",
+      "Ġqu asi",
+      "Ġp ict",
+      "m o",
+      "Ġat om",
+      "Ġh abit",
+      "Ġlimit ations",
+      "con duc",
+      "Ġsh ock",
+      "cept or",
+      "ĠDet ection",
+      "S h",
+      "ub e",
+      "Ġe llip",
+      "U R",
+      "Ġstain ing",
+      "Ġrapid ly",
+      "ĠB ur",
+      "ĠB ro",
+      "Ġup take",
+      "Ġchalleng es",
+      "S N",
+      "Ġan is",
+      "Ġbound s",
+      "st ep",
+      "omer ic",
+      "ten tion",
+      "ĠEval uation",
+      "Ġrecomm end",
+      "M e",
+      "Ġmod erate",
+      "ell ed",
+      "Ġt it",
+      "ĠY ang",
+      "Ġph armac",
+      "inf lammatory",
+      "ĠJ une",
+      "Ġsens ors",
+      "ai red",
+      "Ġapproxim ate",
+      "S V",
+      "Ġb und",
+      "r c",
+      "om an",
+      "Ġvis ible",
+      "Ġmeas uring",
+      "og onal",
+      "ĠFour ier",
+      "Ġthe ories",
+      "Ġprof ession",
+      "tain ed",
+      "at as",
+      "ĠInt erest",
+      "par am",
+      "ĠStruct ure",
+      "Ġl iving",
+      "D ata",
+      "ĠS M",
+      "Ġn et",
+      "Ġsimultaneous ly",
+      "cont inu",
+      "Ġsh or",
+      "#### ####",
+      "Ġdecre asing",
+      "Ġrefer red",
+      "g g",
+      "Th us",
+      "Ġd ro",
+      "pr il",
+      "ĠP ers",
+      "Ġenc oding",
+      "Ġar c",
+      "Ġregul atory",
+      "Ġtra ined",
+      "cep ts",
+      "Ġro ut",
+      "ly s",
+      "P ar",
+      "ĠU l",
+      "ĠG raph",
+      "âĪ Ĥ",
+      "Ġir re",
+      "oid al",
+      "Ġex ceed",
+      "Ġmost ly",
+      "ĠP at",
+      "ater nal",
+      "Ġ er",
+      "Ġco verage",
+      "F S",
+      "ĠR ot",
+      "Ġclass ified",
+      "Ġexc itation",
+      "Ġconduc tivity",
+      "Ġcommerc ial",
+      "ĠD el",
+      "ĠP olar",
+      "H R",
+      "Ġtra ffic",
+      "z ing",
+      "Ġsetting s",
+      "Ġincl usion",
+      "Ans wer",
+      "Ġv it",
+      "vit ational",
+      "Ġb ind",
+      "Ġo c",
+      "ĠW estern",
+      "Ġpro sp",
+      "Ġn orth",
+      "it ch",
+      "ĠR iver",
+      "Ġvehic le",
+      "Ġlik elihood",
+      "L D",
+      "Ġin sp",
+      "âĪ Ĩ",
+      "Ġle uk",
+      "ĠB re",
+      "Ġsynt hetic",
+      "ĠGerm any",
+      "ĠThe ir",
+      "t arget",
+      "ĠEn glish",
+      "Ġnot ation",
+      "ĠA TP",
+      "ĠMod els",
+      "Ġab normal",
+      "ĠConcl usions",
+      "Ġoccur rence",
+      "Ġmicro bi",
+      "ĠW ar",
+      "tem ber",
+      "Ġloc ally",
+      "bor n",
+      "Ġbar rier",
+      "Ġexpression s",
+      "ov al",
+      "Ġfl av",
+      "emb le",
+      "Ġdynam ical",
+      "Ġphot on",
+      "app ed",
+      "Ġgl ut",
+      "Ġkine tic",
+      "Ġalcoh ol",
+      "Ġtrans plant",
+      "L P",
+      "Ġdef ault",
+      "Ġop portun",
+      "arg s",
+      "ĠD av",
+      "Ġfron t",
+      "h om",
+      "Ġw ays",
+      "ĠAss ociation",
+      "Ġkid ney",
+      "Ġpropor tional",
+      "W hen",
+      "Ġepit helial",
+      "Ġf resh",
+      "Ġrec all",
+      "Ġenzym es",
+      "b r",
+      "Ġmal ign",
+      "text rm",
+      "ĠU se",
+      "N ow",
+      "ĠL ie",
+      "Ġimp air",
+      "Ġgu arant",
+      "Ġin ver",
+      "Ġtranscri pt",
+      "Ġs ustain",
+      "Ġact ually",
+      "al ities",
+      "ĠM ic",
+      "ĠI C",
+      "ĠMe asure",
+      "Ġï£ ´",
+      "Ġd ensities",
+      "Ġgalax y",
+      "Ġsu fficiently",
+      "Ġor bit",
+      "f ord",
+      "Ġpar tially",
+      "ĠP y",
+      "Ġre verse",
+      "Ġsur ve",
+      "ĠW ork",
+      "Ġas k",
+      "H owever",
+      "Ġsit u",
+      "Ġvacu um",
+      "to ber",
+      "Ġsp ac",
+      "an th",
+      "O r",
+      "ag s",
+      "Ġb ig",
+      "her ical",
+      "er ge",
+      "ell ite",
+      "Ġinvol ves",
+      "ĠV is",
+      "Ġsumm ary",
+      "ĠSup plementary",
+      "ĠC oll",
+      "Ġadj acent",
+      "ont aneous",
+      "ab s",
+      "Ġresearc hers",
+      "k a",
+      "Ġinter n",
+      "Ġmon th",
+      "ĠNe ural",
+      "ap or",
+      "ĠN an",
+      "Ġst ri",
+      "E E",
+      "Ġconsist ing",
+      "Ġupd ate",
+      "Ġphot o",
+      "V al",
+      "s ens",
+      "Ġve get",
+      "B R",
+      "Ġco inc",
+      "ĠJ uly",
+      "til ity",
+      "ĠEx pression",
+      "Ġtop ology",
+      "Ġgrow ing",
+      "ap tic",
+      "uc ed",
+      "Ġperipher al",
+      "en es",
+      "Ġpl ots",
+      "Ġexpl o",
+      "Ġw or",
+      "b a",
+      "ati tis",
+      "i ef",
+      "w ave",
+      "Ġprot ection",
+      "Ġdef ects",
+      "Ġads orption",
+      "Ġsh ared",
+      "Ġst ellar",
+      "ĠB a",
+      "ĠEn ergy",
+      "que ous",
+      "ĠAug ust",
+      "Ġl ys",
+      "Ġpl us",
+      "i rel",
+      "ĠG P",
+      "ĠNe u",
+      "d ist",
+      "g ers",
+      "if er",
+      "is p",
+      "Ġstr at",
+      "ion e",
+      "ĠMaterial s",
+      "Ġl n",
+      "Ġpul monary",
+      "en ed",
+      "pl an",
+      "M od",
+      "Ġorgan ization",
+      "Ġrelax ation",
+      "Ġcor tex",
+      "Ġmod ulation",
+      "og l",
+      "sh ift",
+      "Ġsec urity",
+      "Ġfat ty",
+      "Ġm s",
+      "l ocal",
+      "erg ic",
+      "Ġinter ference",
+      "ins on",
+      "c f",
+      "Ġre asons",
+      "p red",
+      "Ġinterven tions",
+      "Ġj o",
+      "ĠI D",
+      "ĠAre a",
+      "ĠH a",
+      "u its",
+      "out put",
+      "L e",
+      "y cl",
+      "in ted",
+      "Ġnan o",
+      "N C",
+      "ĠC ap",
+      "Ġchang ing",
+      "Ġc ust",
+      "Ġappe ared",
+      "Ġgrow n",
+      "ĠU K",
+      "Ġrad ical",
+      "ĠP ot",
+      "ĠPro gram",
+      "ĠS R",
+      "Ġsh ap",
+      "os cop",
+      "ĠCh ang",
+      "Ġquanti ty",
+      "ĠT axon",
+      "id ation",
+      "Ġadd ing",
+      "ĠLe e",
+      "Ġam ounts",
+      "Ġdes pite",
+      "Ġremain ed",
+      "Ġscenari os",
+      "le ts",
+      "om ing",
+      "Ġcurv ature",
+      "Ġd imensional",
+      "Ġprom ising",
+      "ĠF il",
+      "str ing",
+      "Ġattrib uted",
+      "ym er",
+      "Ġneigh b",
+      "Ġinput s",
+      "Ġmagn et",
+      "Ġtre es",
+      "Ġent er",
+      "r uit",
+      "st able",
+      "to plas",
+      "Ġmess age",
+      "roph ic",
+      "Ġisol ates",
+      "t z",
+      "Ġdisplay ed",
+      "H A",
+      "oc l",
+      "Ġder ive",
+      "Ġsyn chron",
+      "Q U",
+      "Ã ŀ",
+      "Ġexam ination",
+      "Ġde b",
+      "Ġdef in",
+      "Ġf ault",
+      "Ġstead y",
+      "Ġphen otype",
+      "Ġpers pective",
+      "Ġstat ement",
+      "d f",
+      "v oid",
+      "Ġprom ote",
+      "ill ary",
+      "ĠE th",
+      "Ġw alk",
+      "Ġrepresent ing",
+      "Ġgen omic",
+      "ĠG r",
+      "sh ape",
+      "ĠP et",
+      "ĠL ocal",
+      "plic ity",
+      "ĠProb lem",
+      "G S",
+      "Ġcomple ted",
+      "ink ing",
+      "Ġread s",
+      "Ġin de",
+      "ce ived",
+      "ĠP L",
+      "ĠMe an",
+      "ĠSch ool",
+      "Ġbiom ark",
+      "irel ess",
+      "c ut",
+      "os ing",
+      "n el",
+      "ĠA pril",
+      "ĠB al",
+      "Ġadop ted",
+      "Ġcom plications",
+      "Ġassemb ly",
+      "f ort",
+      "h ar",
+      "Ġad oles",
+      "Ġans wer",
+      "Ġcommun ities",
+      "ĠInstit ute",
+      "Ġvari ant",
+      "F inally",
+      "mit te",
+      "Ġrestric ted",
+      "Ġman ip",
+      "at ers",
+      "E X",
+      "Ġd ust",
+      "Ġsupp ly",
+      "Ġper me",
+      "Ġreli able",
+      "ĠRes p",
+      "Ġsub t",
+      "o ks",
+      "Ġpol l",
+      "Ġcan c",
+      "ĠUn it",
+      "Ġendot helial",
+      "d y",
+      "ĠBl ack",
+      "Ġem pirical",
+      "Ġp ort",
+      "op y",
+      "Ġiniti ally",
+      "Ġcond ens",
+      "Ġe ye",
+      "Ġlist ed",
+      "ur rence",
+      "Ġreplac ed",
+      "Ġselec tive",
+      "Ġdist ances",
+      "Ġpar as",
+      "ĠP ost",
+      "ĠSep tember",
+      "Ġmiss ing",
+      "vere x",
+      "E r",
+      "Ġthough t",
+      "th al",
+      "Ġchrom at",
+      "Ġbenef it",
+      "ram es",
+      "ĠSup pose",
+      "Ġsub s",
+      "Ġang i",
+      "or i",
+      "Ġre plic",
+      "Ġschem es",
+      "p re",
+      "pl ane",
+      "Ġs outh",
+      "ag er",
+      "Ġbegin ning",
+      "v ents",
+      "on ent",
+      "i ples",
+      "ĠH er",
+      "Ġspect rom",
+      "Ġden se",
+      "Ġto ok",
+      "iver se",
+      "Ġdist urb",
+      "p ass",
+      "Ġillustr ated",
+      "Ġreve als",
+      "am a",
+      "Ġref lec",
+      "Ġallow ing",
+      "Ġexpon ential",
+      "ous tic",
+      "subset eq",
+      "Ġs n",
+      "Ġ urban",
+      "Ġext end",
+      "Ġass ays",
+      "ric e",
+      "Co V",
+      "quis ition",
+      "r ine",
+      "ĠIn tegr",
+      "f il",
+      "V D",
+      "Ġfib ro",
+      "Ġcomp ens",
+      "ĠIm pro",
+      "ĠĠĠĠĠĠĠĠ ĠĠ",
+      "ĠG R",
+      "Ï Ī",
+      "Ġbas al",
+      "Ġol ig",
+      "H T",
+      "Ġv ess",
+      "uz zy",
+      "Ġposs ibly",
+      "Ġtoler ance",
+      "The ta",
+      "Ġvi ol",
+      "uc lear",
+      "ĠL im",
+      "g el",
+      "Ġmetric s",
+      "ĠM us",
+      "am ination",
+      "Ġelectro de",
+      "Ġpers onal",
+      "Ġcool ing",
+      "Ġac quired",
+      "ĠF unction",
+      "ow s",
+      "oles ter",
+      "D P",
+      "Ġreli ability",
+      "Ġm uc",
+      "ĠOc tober",
+      "Ġg old",
+      "c a",
+      "Ġc ul",
+      "f it",
+      "Ġle m",
+      "Ġexc it",
+      "Ġnucle us",
+      "i ation",
+      "Ġpregn ancy",
+      "Ġsynt hesized",
+      "hem istry",
+      "Ġmemb ranes",
+      "ver t",
+      "ĠK im",
+      "ten ance",
+      "Ġquanti ties",
+      "Ġecon omic",
+      "Ġbenef its",
+      "Ġc ylind",
+      "pl er",
+      "ĠL arge",
+      "Ġengine ering",
+      "ĠE p",
+      "Ġco ating",
+      "ati v",
+      "Ġconduc t",
+      "Ġabs orb",
+      "ĠDec ember",
+      "Ġop posite",
+      "ĠGl obal",
+      "Ġl if",
+      "ĠD ue",
+      "Ġint ake",
+      "od ynamic",
+      "T M",
+      "Ġf ed",
+      "Ġspec ified",
+      "Ġge ometric",
+      "Ġresp ective",
+      "Ġb irth",
+      "ĠComp ound",
+      "Ġstar ted",
+      "Ġm other",
+      "ar r",
+      "Ġprim arily",
+      "Ġp aren",
+      "Ġtub e",
+      "Ġinter s",
+      "Ġgrap hene",
+      "iti al",
+      "ous ly",
+      "Ġcardi ovascular",
+      "Ġe V",
+      "Ġhe ating",
+      "Ġmat hematical",
+      "Ġindepend ently",
+      "B A",
+      "Ġaff ects",
+      "um or",
+      "ĠM P",
+      "ĠD em",
+      "ĠW est",
+      "ĠD om",
+      "it ter",
+      "Ġdis rup",
+      "op ed",
+      "Ġphenomen on",
+      "Ġl umin",
+      "A c",
+      "Ġpre fer",
+      "om ers",
+      "Ġg ender",
+      "ĠG L",
+      "F C",
+      "Ġinde ed",
+      "Ġr ational",
+      "ĠS C",
+      "Ġprincip al",
+      "Ġperf ect",
+      "Ġintro duction",
+      "t es",
+      "Ġpi ec",
+      "Ġc ity",
+      "Ġpop ular",
+      "Ġc oding",
+      "cl er",
+      "ag ue",
+      "ĠH R",
+      "Ġtrack ing",
+      "k er",
+      "Ġphosphor ylation",
+      "Ġpath s",
+      "Ġsol ving",
+      "Ġd y",
+      "Ġplay ed",
+      "Ġprec ise",
+      "ĠS l",
+      "ĠS em",
+      "Ġgener ating",
+      "ĠS un",
+      "Ġcriter ion",
+      "Ġbran ch",
+      "ĠÎ ¶",
+      "ti sh",
+      "S e",
+      "Ġanti gen",
+      "Ġcal ibration",
+      "E s",
+      "ĠI tal",
+      "Ġmass ive",
+      "E n",
+      "N o",
+      "Y P",
+      "y a",
+      "Ġsatisf ying",
+      "Ġqu ick",
+      "H O",
+      "Ġbehavi ors",
+      "icro bial",
+      "Ġam b",
+      "Ġpro ton",
+      "S L",
+      "Ġus ual",
+      "row s",
+      "en ch",
+      "U C",
+      "Ġweight ed",
+      "Ġrec ords",
+      "ĠA C",
+      "G T",
+      "in n",
+      "Ġe q",
+      "ĠW il",
+      "y roid",
+      "Ġset up",
+      "I A",
+      "p ress",
+      "is ely",
+      "Ġent ry",
+      "% %",
+      "ĠS il",
+      "e ast",
+      "ĠE volution",
+      "ĠR andom",
+      "Ġcav ity",
+      "Ġnam ed",
+      "know led",
+      "m ber",
+      "ues tion",
+      "ĠâĪ ©",
+      "g i",
+      "Ġdeterm ining",
+      "t in",
+      "Ġgen us",
+      "Ġtox icity",
+      "oc yt",
+      "Ġperturb ation",
+      "rough t",
+      "ĠB ri",
+      "Ġcar b",
+      "ĠG ra",
+      "ĠF lu",
+      "un s",
+      "Ġdri ven",
+      "Ġb atch",
+      "r if",
+      "P l",
+      "Ġdisplac ement",
+      "ĠC L",
+      "Ġdep ic",
+      "Ġpredic tive",
+      "I nt",
+      "hydro xy",
+      "ti d",
+      "d ri",
+      "Ġp ancre",
+      "Ġdiag onal",
+      "Ġsever ity",
+      "Ġlong itudinal",
+      "ĠE D",
+      "ati ble",
+      "d ir",
+      "ĠAn other",
+      "ĠH el",
+      "v an",
+      "Ġp neum",
+      "Ġspecific ity",
+      "s qu",
+      "Ġ ign",
+      "Ġb ed",
+      "ĠW T",
+      "aw a",
+      "es ter",
+      "Ġk g",
+      "Ġcomp ression",
+      "ever theless",
+      "Ġm ask",
+      "-------- ---",
+      "Ġt ens",
+      "row th",
+      "ĠG o",
+      "Ġf aster",
+      "Ġcan onical",
+      "Ġdeterm in",
+      "ust rial",
+      "ĠEar th",
+      "wh ile",
+      "our nal",
+      "Ġcount ry",
+      "Ġf erm",
+      "r ist",
+      "Ġpro xim",
+      "Ġmicro bial",
+      "Ġext ensive",
+      "Ġch am",
+      "ĠÂ §",
+      "s uch",
+      "w ent",
+      "Ġl ar",
+      "U sing",
+      "ĠP M",
+      "Ġoff set",
+      "ĠP I",
+      "ĠBay esian",
+      "H S",
+      "ĠAfric a",
+      "Ġsuscepti bility",
+      "ĠâĬ Ĥ",
+      "ococc us",
+      "ĠD ir",
+      "Ġb os",
+      "Ġdys function",
+      "ove mber",
+      "Ġunder st",
+      "Ġlarg ely",
+      "ĠC M",
+      "Ġmaintain ed",
+      "Ġposs ess",
+      "Ġexcl uded",
+      "ens is",
+      "ĠD C",
+      "ops is",
+      "Ġtor ch",
+      "id ine",
+      "Ġfore st",
+      "ĠEx act",
+      "ĠStud ies",
+      "iff iff",
+      "ĠC am",
+      "ang ular",
+      "Ġrem ove",
+      "o ir",
+      "av a",
+      "id a",
+      "Ġm ant",
+      "L og",
+      "Ġrang ing",
+      "ro g",
+      "Ġchain s",
+      "Ġ Ç«",
+      "ĠC ase",
+      "ĠA P",
+      "po ints",
+      "Ġtarget ing",
+      "Ġsc ience",
+      "Ġep is",
+      "ĠS oci",
+      "Ġphys ic",
+      "Ġpromot er",
+      "ĠEar ly",
+      "es tic",
+      "tiv es",
+      "Ġassum ing",
+      "ĠM i",
+      "Ġlem ma",
+      "Ġconfig urations",
+      "al ia",
+      "Ġp ay",
+      "r ino",
+      "e b",
+      "Ġvari ed",
+      "oun ted",
+      "Ġinter view",
+      "ĠGe V",
+      "O M",
+      "ogn ition",
+      "Ġenhance ment",
+      "ĠM ach",
+      "pl ies",
+      "O b",
+      "set minus",
+      "Ġintr insic",
+      "Ġcompar isons",
+      "b old",
+      "xi ety",
+      "Ġst roke",
+      "G B",
+      "anc ial",
+      "ste ad",
+      "Ġro ck",
+      "th on",
+      "ĠC urrent",
+      "c at",
+      "Ġgu idelines",
+      "cy cl",
+      "Ġintrac ellular",
+      "one y",
+      "k o",
+      "Ġdirec ted",
+      "rip ts",
+      "Ġtra vel",
+      "Ġl ens",
+      "id i",
+      "ĠAss ess",
+      "Ġd x",
+      "ĠP os",
+      "Ġmethod ology",
+      "Ġpred om",
+      "def ined",
+      "ĠP op",
+      "Ġgover nment",
+      "ell ig",
+      "ph yl",
+      "ol i",
+      "rop ical",
+      "Ġembed ded",
+      "ed om",
+      "crib ed",
+      "ĠDise ase",
+      "Ġmedi ated",
+      "Ġcirc ular",
+      "ĠTop ological",
+      "Ġear th",
+      "ri tis",
+      "g al",
+      "m ass",
+      "Ġcomprehens ive",
+      "ĠA ir",
+      "Ġn erve",
+      "Ġimpl ant",
+      "Ġextrem ely",
+      "ĠS E",
+      "Ġmark et",
+      "Ġconserv ed",
+      "emb rane",
+      "Ġsched ul",
+      "Ġrun s",
+      "P h",
+      "Ġtechn ical",
+      "T L",
+      "Ġregion al",
+      "Ġg erm",
+      "ĠPro t",
+      "Ġb right",
+      "Ġarter y",
+      "Ġmacroph ages",
+      "mitte e",
+      "ĠSing le",
+      "Ġcom e",
+      "w a",
+      "ac char",
+      "ple t",
+      "Ġsens ing",
+      "ros p",
+      "at om",
+      "Ġcomp r",
+      "ĠL u",
+      "Ġavail ability",
+      "pro t",
+      "Ġfit ting",
+      "sel ves",
+      "ĠP rim",
+      "re w",
+      "Ġwas te",
+      "ĠK ing",
+      "p ot",
+      "Ġinstr ument",
+      "ĠY ork",
+      "A F",
+      "an tial",
+      "stand ing",
+      "Ġpl anning",
+      "ust er",
+      "Ġâ Ĩ",
+      "N T",
+      "ic ular",
+      "Ġmel an",
+      "Ġexc ell",
+      "ill er",
+      "ĠL D",
+      "inf o",
+      "Ġsh are",
+      "v as",
+      "Ġl um",
+      "Ġa queous",
+      "Ġqu ery",
+      "Ġm ag",
+      "ult ure",
+      "ĠB er",
+      "Ġof fer",
+      "ĠN MR",
+      "ace ae",
+      "Ġmod ern",
+      "Ġcirc um",
+      "Ġcult ures",
+      "Ġd og",
+      "Ġc ir",
+      "Ġpol i",
+      "Ġchem otherapy",
+      "Ġpl ates",
+      "Ġrestric tion",
+      "st ack",
+      "ĠF low",
+      "ĠB u",
+      "ĠC enter",
+      "Ġpro ceed",
+      "tim icrobial",
+      "s he",
+      "Ġthere by",
+      "Ġkn ock",
+      "Ġdi verse",
+      "ustr y",
+      "Ġst ated",
+      "ĠH ol",
+      "M ore",
+      "Ġconserv ation",
+      "Ġpre vention",
+      "n orm",
+      "Ġp al",
+      "ĠCal c",
+      "Ġcle an",
+      "ĠPl as",
+      "`` `",
+      "per p",
+      "pro d",
+      "Ġâī ¡",
+      "por ter",
+      "Ġtrans ient",
+      "as p",
+      "Ġtarget ed",
+      "ĠP ri",
+      "Sup plementary",
+      "ĠT reatment",
+      "z en",
+      "ĠM art",
+      "ĠF erm",
+      "us cript",
+      "ĠS ynthesis",
+      "Ġcomb inations",
+      "UL L",
+      "Ġwe b",
+      "Ġth rom",
+      "Ġexplicit ly",
+      "an ks",
+      "Ġadapt ation",
+      "ĠSequ ence",
+      "Ġac ts",
+      "Ġrang es",
+      "f s",
+      "b ru",
+      "Ġsystem ic",
+      "Ġste el",
+      "Ġpri vate",
+      "Ġob esity",
+      "ĠP art",
+      "ment ed",
+      "bre ak",
+      "ER T",
+      "Ġfib ers",
+      "Ġis o",
+      "Ġtrans verse",
+      "CT ION",
+      "ĠR i",
+      "it in",
+      "ĠRep resent",
+      "oph ys",
+      "Ġco ast",
+      "Ġal ignment",
+      "AC T",
+      "es ides",
+      "op en",
+      "g ly",
+      "Ġsal t",
+      "unc ed",
+      "ia z",
+      "Ġcos m",
+      "Ġang les",
+      "ĠâĢ ł",
+      "ĠIdentif ication",
+      "he x",
+      "ĠH all",
+      "Ġhep at",
+      "Ġseg ments",
+      "ĠPh ase",
+      "ĠL and",
+      "form ing",
+      "h box",
+      "ic ations",
+      "Ġsubsequ ently",
+      "ĠC ur",
+      "Ġlab els",
+      "vid ence",
+      "ual ity",
+      "Ġhe ld",
+      "em ann",
+      "Ġcam era",
+      "c ing",
+      "ub ic",
+      "ĠS ARS",
+      "ul atory",
+      "kele tal",
+      "ĠInf lu",
+      "ĠInd ia",
+      "am ic",
+      "Ġs and",
+      "Ġcom es",
+      "Ġassoci ations",
+      "Ġcharg ed",
+      "Ġs per",
+      "opro tein",
+      "ii i",
+      "od al",
+      "Ġbound aries",
+      "ti zation",
+      "ĠHe avy",
+      "ĠRe al",
+      "ĠA F",
+      "Ġcontroll er",
+      "Ġantioxid ant",
+      "Ġb ars",
+      "Ġw et",
+      "en er",
+      "ĠComplex ity",
+      "Ġst ack",
+      "There fore",
+      "Ġre plication",
+      "Ġappear ance",
+      "Ġtraject ory",
+      "Ġunderst ood",
+      "Ġd ot",
+      "Ġim ag",
+      "Ġsc anning",
+      "T i",
+      "r uct",
+      "ĠL y",
+      "Ġsp ontaneous",
+      "l at",
+      "om on",
+      "Ġro ots",
+      "Ġl ive",
+      "Ġfin ally",
+      "¿ ½",
+      "Ġappro ved",
+      "ĠAp plications",
+      "ĠP an",
+      "Ġl ost",
+      "Ġsatisf ied",
+      "Ġg amma",
+      "ion al",
+      "Ġimpro ving",
+      "Ġmanif old",
+      "Ġc odes",
+      "b b",
+      "ĠN ovember",
+      "Ġr ich",
+      "N P",
+      "ĠE le",
+      "S B",
+      "Ġde al",
+      "Ġop tions",
+      "Ġcult ured",
+      "Ġv ul",
+      "> >",
+      "ar ithm",
+      "o ys",
+      "The se",
+      "ĠDet erm",
+      "Ġquad ratic",
+      "ĠCom b",
+      "iss on",
+      "ĠPer formance",
+      "Ġex ception",
+      "Ġnucle i",
+      "Ġad verse",
+      "k et",
+      "ĠP al",
+      "ĠM any",
+      "Ġdiff raction",
+      "Ġtrans mit",
+      "Ġphosph ate",
+      "olester ol",
+      "Ġquestion nai",
+      "ĠSe a",
+      "bru ary",
+      "Ġmod elling",
+      "ĠD R",
+      "ol in",
+      "ch mark",
+      "Ġprec isely",
+      "g ans",
+      "v in",
+      "rid ge",
+      "ĠInc re",
+      "Ġneur onal",
+      "Ġâī Ī",
+      "Ġexcell ent",
+      "et ary",
+      "Ġoverl ap",
+      "Ġstrong er",
+      "Ġfract ure",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠ",
+      "Ġclin ic",
+      "ĠL ist",
+      "Ġhist or",
+      "gen eration",
+      "ric hed",
+      "ill us",
+      "ĠÃ ħ",
+      "ĠR ole",
+      "Ġlabel ed",
+      "Ġorth ogonal",
+      "Ġis chem",
+      "Ġinst ability",
+      "lo op",
+      "Ġplot ted",
+      "ĠProcess ing",
+      "ĠT a",
+      "ĠConcl usion",
+      "Ġm agne",
+      "Ġunivers al",
+      "Ġj et",
+      "Ġreg im",
+      "flo at",
+      "Ġc od",
+      "ad j",
+      "bold math",
+      "Ġar rang",
+      "Ġtrend s",
+      "Ġprecip itation",
+      "f requency",
+      "Ġcont rad",
+      "Ġtransfer red",
+      "Ġmain tenance",
+      "Î Ķ",
+      "n p",
+      "ist ence",
+      "he res",
+      "lec tive",
+      "ĠSur vey",
+      "Ġ Ð",
+      "Ġst and",
+      "Ġdisc overy",
+      "ain s",
+      "vers ely",
+      "Ġnumer ous",
+      "yl ated",
+      "Ġembed ding",
+      "Ġcoll abor",
+      "en ame",
+      "im mun",
+      "Ġadjust ed",
+      "i res",
+      "c ur",
+      "Ġvacc ine",
+      "Ġtra its",
+      "Ġmorph ological",
+      "Ġprec urs",
+      "roscop e",
+      "ad i",
+      "ec utive",
+      "u an",
+      "Ġt ract",
+      "ĠP res",
+      "Ġmy el",
+      "Ġad equ",
+      "Ġeth anol",
+      "i h",
+      "Ġmet h",
+      "Ġcoun ts",
+      "Ġqualit ative",
+      "Ġmus ic",
+      "Ġre infor",
+      "A fter",
+      "Ġac quisition",
+      "Ġh ttps",
+      "all ing",
+      "it a",
+      "ic ate",
+      "sc ript",
+      "Ġoptim ized",
+      "ĠH o",
+      "Ġm ild",
+      "opl as",
+      "Ġo verex",
+      "ĠâĪ §",
+      "Ġcol lect",
+      "ĠM ain",
+      "Ġextrac ellular",
+      "Ġan c",
+      "ra wn",
+      "Ġexpl ored",
+      "Ġres erv",
+      "ĠAp plication",
+      "c ase",
+      "Ġmar ine",
+      "ĠĠĠĠĠĠĠĠ ĠĠĠĠĠ",
+      "il ed",
+      "Ġmes h",
+      "ĠMon te",
+      "cl os",
+      "Ġperform ing",
+      "A g",
+      "reg ular",
+      "Ġc atal",
+      "Ġpotenti als",
+      "ant ly",
+      "U RE",
+      "Ġacc omp",
+      "Ġreason able",
+      "Ġpresent ation",
+      "abol ic",
+      "ĠOn ly",
+      "ann ed",
+      "Ġsubst antial",
+      "Ġdiet ary",
+      "Ġsubstr ates",
+      "ap ter",
+      "Ġint estinal",
+      "Ġproduc es",
+      "Pro position",
+      "ro gen",
+      "ĠSt at",
+      "bur g",
+      "ren ch",
+      "text bf",
+      "ystem s",
+      "at able",
+      "ĠV ir",
+      "Ġsol ved",
+      "ic ense",
+      "Ġs ong",
+      "Ġext reme",
+      "pt y",
+      "ĠC ity",
+      "ve red",
+      "ĠMR I",
+      "Ġtw ice",
+      "ĠM n",
+      "Ġm erg",
+      "ac tivation",
+      "Ġn g",
+      "Ġo dd",
+      "Ġatt rac",
+      "Ġatt empt",
+      "Ġsepar ately",
+      "Ġrob ot",
+      "ĠMulti ple",
+      "Ġsc ientific",
+      "ĠP P",
+      "Ġmin eral",
+      "Ġprotoc ols",
+      "Ġsuper ior",
+      "oc amp",
+      "box yl",
+      "Ġuniform ly",
+      "ĠS everal",
+      "Ġm ol",
+      "C or",
+      "under line",
+      "Ġinflu enced",
+      "Ġcur ren",
+      "us ing",
+      "rac e",
+      "ĠN evertheless",
+      "Ġacc om",
+      "Ġgra vitational",
+      "Ġindi rect",
+      "Ġcap able",
+      "Ġanalys ed",
+      "Ġdis charge",
+      "Ġv es",
+      "Ġlig and",
+      "l ik",
+      "Ġs i",
+      "Ġag ed",
+      "Ġcryst als",
+      "Ġspe ech",
+      "Ġcop per",
+      "ĠS an",
+      "ĠA rm",
+      "Ġman uscript",
+      "Ġsec retion",
+      "w edge",
+      "Â ·",
+      "Ġra w",
+      "Ġaim ed",
+      "Ġevolution ary",
+      "Ġconsequ ences",
+      "Ġit em",
+      "Ġw estern",
+      "Ġsol vent",
+      "Ġstim uli",
+      "Ġrequire ment",
+      "h ttp",
+      "ef ore",
+      "ĠAt l",
+      "Ġatmosp heric",
+      "Ġpack age",
+      "Ġmy ocardi",
+      "Ġd ashed",
+      "Ġver ify",
+      "ativ istic",
+      "Ġto m",
+      "av irus",
+      "ak en",
+      "ĠNum er",
+      "Ġadvant ages",
+      "F R",
+      "ĠS elf",
+      "rec ted",
+      "con fig",
+      "Ġit eration",
+      "Ġeigen values",
+      "Ġprob abilities",
+      "F IG",
+      "ĠW ater",
+      "ĠA u",
+      "Ġg ave",
+      "Ġv ar",
+      "ric ular",
+      "opath y",
+      "Ġr h",
+      "ord ance",
+      "Ġw in",
+      "ĠS cale",
+      "Ġann ual",
+      "atas et",
+      "Ġp el",
+      "ĠâĪ ª",
+      "ĠC C",
+      "it ors",
+      "Ġl ith",
+      "Ġchrom osome",
+      "Ġf uel",
+      "Ġmul tiv",
+      "Ġmanufact ure",
+      "l a",
+      "ĠS a",
+      "um es",
+      "ig m",
+      "Ġnan oc",
+      "E GF",
+      "Ġsign ature",
+      "N S",
+      "Ġme et",
+      "Ġf air",
+      "met h",
+      "Ġlocal ized",
+      "ĠCent ral",
+      "de g",
+      "Ġsurround ing",
+      "Ġn one",
+      "ĠM O",
+      "ĠInterest ingly",
+      "Ġmul tic",
+      "ĠK e",
+      "Ġinhib ited",
+      "ĠC are",
+      "ĠOp en",
+      "Ġgl ob",
+      "E A",
+      "ĠF ound",
+      "Ġpix el",
+      "ok e",
+      "R D",
+      "l oc",
+      "ti ous",
+      "Ġdistingu ish",
+      "Ġan terior",
+      "ur ch",
+      "Ġj ud",
+      "ĠP ower",
+      "Ġswit ch",
+      "ĠS yn",
+      "Ġinvolve ment",
+      "uc l",
+      "Ġlibr ary",
+      "ĠCon st",
+      "Ġsp herical",
+      "ĠT NF",
+      "Ġal tered",
+      "v ance",
+      "trans fer",
+      "M s",
+      "ĠO per",
+      "in ement",
+      "se q",
+      "C ons",
+      "ho le",
+      "ĠPh ot",
+      "Ġg ut",
+      "acter ial",
+      "ĠI P",
+      "un t",
+      "Ġn om",
+      "h as",
+      "ĠFe bruary",
+      "Ġpro state",
+      "ĠM L",
+      "h igh",
+      "ĠBack ground",
+      "ul ent",
+      "Ġo cean",
+      "a fter",
+      "ĠO ff",
+      "l oss",
+      "Ġfav or",
+      "Ġwork ers",
+      "Ġh idden",
+      "Ġextrac ts",
+      "raz il",
+      "s ign",
+      "N one",
+      "Ġcolum ns",
+      "Ġfrac tions",
+      "Ġco vered",
+      "ĠS erv",
+      "Ġin form",
+      "b ed",
+      "Ġatt em",
+      "rain ing",
+      "Ġneut ron",
+      "Ġr ice",
+      "Ġmo tif",
+      "Ġartif icial",
+      "Ġinhibit ory",
+      "Ġd t",
+      "AG E",
+      "Ġsam pled",
+      "Ġb atter",
+      "Ġsub jected",
+      "Ġgener ic",
+      "ĠN H",
+      "Ġcontin ue",
+      "ution al",
+      "Ġa ug",
+      "i us",
+      "Ġexec ution",
+      "ĠW illi",
+      "ĠDes pite",
+      "A MI",
+      "Ġcont ents",
+      "ĠS ens",
+      "og ens",
+      "C ol",
+      "Ġf o",
+      "Ġad di",
+      "u ated",
+      "Ġrecomm ended",
+      "ĠS W",
+      "Ġar ch",
+      "ĠY es",
+      "Ġh ol",
+      "atur ally",
+      "ti tive",
+      "Ġc he",
+      "Ġs ector",
+      "ĠDef inition",
+      "Ġcon cepts",
+      "or ous",
+      "sm all",
+      "ers on",
+      "in ator",
+      "ĠM T",
+      "Ġhypert ension",
+      "c ks",
+      "Ġn ative",
+      "Ġt ax",
+      "r yl",
+      "Ġre active",
+      "r b",
+      "duc ible",
+      "om m",
+      "Ġdiagnos ed",
+      "Ġdri ving",
+      "Ġbiom ass",
+      "u ate",
+      "Ġp il",
+      "c alled",
+      "Ġser ve",
+      "Ġinter fer",
+      "ipp ocamp",
+      "Ġalgebra ic",
+      "Ġbe gan",
+      "Ġpict ure",
+      "ind ependent",
+      "Ġutil ized",
+      "go ing",
+      "or a",
+      "n m",
+      "Ġdown stream",
+      "Ġorb ital",
+      "oun tain",
+      "ĠH is",
+      "Ġres ol",
+      "Ġcorrec tions",
+      "on ym",
+      "sc ripts",
+      "Ġsil icon",
+      "Ġc um",
+      "ĠT ri",
+      "Ġpepti des",
+      "Ġrece iving",
+      "Ġstation ary",
+      "Ġμ L",
+      "cler osis",
+      "Ġmod ules",
+      "em a",
+      "ĠAfric an",
+      "struc tion",
+      "Ġf arm",
+      "Ġlear n",
+      "n ode",
+      "Â ®",
+      "Ġsuper conduc",
+      "ĠL inear",
+      "Ġtechn ologies",
+      "Ġnecess arily",
+      "Ġcoron ary",
+      "ĠE ast",
+      "Ġf rames",
+      "Ġseg mentation",
+      "V s",
+      "Ġbehavior al",
+      "Î ĵ",
+      "Ġlog ic",
+      "Ġaccomp an",
+      "tif ied",
+      "han ol",
+      "ĠIn hib",
+      "il ation",
+      "and er",
+      "Ġeff ort",
+      "ĠD en",
+      "D I",
+      "op tim",
+      "term inal",
+      "Ġmob ility",
+      "Ġconsider ation",
+      "O VA",
+      "Ġpar ad",
+      "ox o",
+      "Ġdef iciency",
+      "ult ural",
+      "Ġvalid ity",
+      "Ġord ers",
+      "Ġloc us",
+      "Ġar th",
+      "em at",
+      "Ġfeed ing",
+      "Ġprogram ming",
+      "Ġtem plate",
+      "el ian",
+      "Ġop tion",
+      "ĠF ollowing",
+      "Ġen able",
+      "Ġass ign",
+      "Ġform ul",
+      "p u",
+      "Ġatmosp here",
+      "sl ant",
+      "ĠR uss",
+      "ĠE vidence",
+      "Ġsimilar ly",
+      "Ġc amp",
+      "Ġw ound",
+      "ĠCharacter ization",
+      "ĠP BS",
+      "e es",
+      "ĠDi rect",
+      "ĠS L",
+      "Ġfr uit",
+      "Ġg ate",
+      "it o",
+      "C hem",
+      "Ġcoll ision",
+      "or tic",
+      "Ġpolym orphism",
+      "enz a",
+      "w hat",
+      "Ġexperiment ally",
+      "Ġult ra",
+      "e z",
+      "Ġn erv",
+      "Ġess entially",
+      "ĠAustr alia",
+      "ĠSt andard",
+      "Ġmedic ine",
+      "ad ian",
+      "ĠHig gs",
+      "u ge",
+      "Ġsup ports",
+      "um a",
+      "Ġcom plicated",
+      "d ate",
+      "ophag y",
+      "ĠMark ov",
+      "Ġoccur ring",
+      "opl us",
+      "P ub",
+      "pro b",
+      "ur able",
+      "Ġk ept",
+      "Ġisol ation",
+      "Ġev ol",
+      "ili ary",
+      "Ġreg ist",
+      "Ġh oles",
+      "Ġcl ar",
+      "ip ar",
+      "Ġen rich",
+      "Ġro ute",
+      "ay ers",
+      "edi atric",
+      "Ġpolynomial s",
+      "Ġtri vial",
+      "ĠS am",
+      "vari ant",
+      "Ġfre edom",
+      "pos s",
+      "Ġinf erence",
+      "ol a",
+      "Ġinterp reted",
+      "C a",
+      "em ory",
+      "Ġcent ury",
+      "ĠR em",
+      "ĠW u",
+      "Ġsupp ression",
+      "Ġgener ator",
+      "ĠH om",
+      "Ġvis cos",
+      "Ġpseud o",
+      "ĠCh ild",
+      "ĠS A",
+      "ib er",
+      "Ġequival ence",
+      "if ies",
+      "ĠCons ider",
+      "ol ine",
+      "âī ¤",
+      "Ġde ple",
+      "Ġaver aged",
+      "Ġs outhern",
+      "Ġord ered",
+      "ĠB rown",
+      "Ġmethyl ation",
+      "ĠAd ap",
+      "Ġm aternal",
+      "ond ed",
+      "ĠBe havi",
+      "Ġidentif iers",
+      "Ġprocess ed",
+      "G G",
+      "V I",
+      "Ġch a",
+      "un k",
+      "ĠF unctional",
+      "Ġhydro ph",
+      "Ġfin ancial",
+      "ec ond",
+      "ĠÎ ¨",
+      "Ġemph as",
+      "Ġdef ect",
+      "m ar",
+      "Ġnor thern",
+      "c ore",
+      "Ġadhes ion",
+      "Ġte le",
+      "Ġw arm",
+      "rif ug",
+      "rang ian",
+      "res olution",
+      "Ġhe x",
+      "h bar",
+      "Ġhar monic",
+      "Ġcont rac",
+      "Ġread ing",
+      "Ġeff orts",
+      "ĠO l",
+      "Ġan xiety",
+      "b ul",
+      "T C",
+      "ip id",
+      "R emark",
+      "Ġform ing",
+      "il bert",
+      "am ond",
+      "Ġanaly tic",
+      "ore c",
+      "ch a",
+      "ĠCon sequently",
+      "ĠS u",
+      "for all",
+      "ĠÃ ŀ",
+      "Ġasp ect",
+      "Ġins ights",
+      "ati vity",
+      "io tics",
+      "he imer",
+      "ĠL abor",
+      "Ġa ware",
+      "ĠBri tish",
+      "c hemical",
+      "Ġâ ĭ",
+      "cl usion",
+      "ĠM ich",
+      "Ġgra de",
+      "ĠS EM",
+      "ĠC irc",
+      "hes es",
+      "W L",
+      "Ġen abl",
+      "Ġd end",
+      "Ġind ustry",
+      "Ġimpro ves",
+      "t et",
+      "Ġt el",
+      "Ġwas hed",
+      "Ġshor ter",
+      "Ġinc ident",
+      "ĠAc tivity",
+      "Ġdos es",
+      "ĠB razil",
+      "Ġtransform ations",
+      "Ġform at",
+      "ĠPro of",
+      "Ġl en",
+      "ul ative",
+      "Ġcycl ic",
+      "Ġrec ruit",
+      "pt r",
+      "T H",
+      "Ġrece ive",
+      "ĠNe xt",
+      "ĠEx p",
+      "i ant",
+      "in stein",
+      "S et",
+      "re ne",
+      "Ġge omet",
+      "Ġconsider able",
+      "S o",
+      "ugh t",
+      "Ġpaper s",
+      "ĠC S",
+      "z a",
+      "Ġis omorphism",
+      "ho u",
+      "Ġmut ants",
+      "Ġpor tion",
+      "ĠÃ ¾",
+      "Ġcontinu um",
+      "C u",
+      "ĠComput ed",
+      "Ġcomb ining",
+      "ov a",
+      "ĠN P",
+      "Ġc rack",
+      "Ġsome times",
+      "Ġcontinu ed",
+      "Def inition",
+      "arc in",
+      "ĠC d",
+      "ĠMed ical",
+      "i ences",
+      "ĠC ross",
+      "Ġtranscription al",
+      "ĠZ e",
+      "st d",
+      "if orn",
+      "Ġfail ed",
+      "Ġidentif ying",
+      "Ġm ir",
+      "Ġmetast asis",
+      "O F",
+      "n n",
+      "ĠC ID",
+      "Ġoscill ations",
+      "anc ies",
+      "wr ite",
+      "Ġband width",
+      "Ġtra de",
+      "Ġag ing",
+      "ĠModel ing",
+      "Ġass ert",
+      "Ġcurren ts",
+      "Ġf ire",
+      "ub iqu",
+      "Ġalb um",
+      "Ġfrequ ent",
+      "N ame",
+      "Ġpur ch",
+      "Ġpl ayer",
+      "ĠE sc",
+      "Ġno tion",
+      "Ġintern ational",
+      "ul um",
+      "o ic",
+      "Ġincub ation",
+      "Ġphenomen a",
+      "Ġser ver",
+      "ut er",
+      "Ġv en",
+      "qu in",
+      "Ġhyp ox",
+      "ĠR F",
+      "it on",
+      "Er ror",
+      "Ġhe mat",
+      "Ġthem selves",
+      "Ġper p",
+      "id ual",
+      "Ġpur poses",
+      "m es",
+      "w ing",
+      "ro v",
+      "Ġem iss",
+      "Ġexperi enced",
+      "qu es",
+      "ĠL C",
+      "ĠRec ent",
+      "bo ok",
+      "Ġalk al",
+      "id x",
+      "hy th",
+      "Ġconc rete",
+      "Ġswit ching",
+      "Ġexplan ation",
+      "ird s",
+      "Ġsign s",
+      "Ġob j",
+      "Ġcytok ines",
+      "ub ble",
+      "ad der",
+      "Ġuncertain ties",
+      "Ġprom otes",
+      "Ġcom pl",
+      "Ġsc an",
+      "Ġpr ime",
+      "P H",
+      "Ġheter ogeneous",
+      "ĠY ou",
+      "Al though",
+      "Ġser ious",
+      "Ġdri ve",
+      "Ġheter ogeneity",
+      "ryst all",
+      "Ġo d",
+      "Ġcon volution",
+      "ĠâĬ Ĩ",
+      "ĠSp ace",
+      "Ġgast ric",
+      "ĠSt re",
+      "ĠP V",
+      "b ase",
+      "M et",
+      "Ġloss es",
+      "Ġcyt otox",
+      "Ġcontroll ing",
+      "le ase",
+      "Ġreg ulated",
+      "ĠEng ine",
+      "ĠH ospital",
+      "B r",
+      "on om",
+      "hy de",
+      "st age",
+      "Ġgiv ing",
+      "ĠP en",
+      "ĠSoci ety",
+      "dri ven",
+      "i ang",
+      "Ġmod ifications",
+      "B V",
+      "Ġaccel eration",
+      "Ġm ilk",
+      "on omic",
+      "Ġth ink",
+      "ogl ob",
+      "Ġfeas ible",
+      "n am",
+      "Ġref lection",
+      "ĠPol y",
+      "Ġsummar ized",
+      "F L",
+      "Ġrec t",
+      "Ġpredom inant",
+      "Ġbl ot",
+      "de hyde",
+      "Ġtransform ed",
+      "Ġfacilit ate",
+      "ĠCar lo",
+      "Ġgreat ly",
+      "ĠS ocial",
+      "Ġparen ts",
+      "big g",
+      "ros pective",
+      "Ġprogn osis",
+      "Ġcharacter ize",
+      "Ġconn ectivity",
+      "Ġtraject ories",
+      "ĠS H",
+      "Ġl ies",
+      "Ġcandid ates",
+      "rom y",
+      "Ġs or",
+      "ĠIn s",
+      "Ġth or",
+      "Ġmet als",
+      "ĠS V",
+      "Ġtim ing",
+      "Ġutil ity",
+      "Ġnew ly",
+      "ĠI FN",
+      "Ġaffect ing",
+      "ce ment",
+      "ĠM el",
+      "ĠÌ ģ",
+      "typ es",
+      "lys is",
+      "erc ul",
+      "Ġdist or",
+      "act ors",
+      "ps y",
+      "Ġbo ok",
+      "ĠE ven",
+      "tem perature",
+      "Ġinvas ion",
+      "Ġrecogn ized",
+      "fact or",
+      "N e",
+      "Ġinter section",
+      "Ġcor tical",
+      "n g",
+      "Ġde ploy",
+      "Ġamplit udes",
+      "Ġd a",
+      "ĠG C",
+      "Ġchalleng ing",
+      "Ġpre lim",
+      "G M",
+      "A cc",
+      "Ġfour th",
+      "al c",
+      "ĠP S",
+      "ĠGene tic",
+      "l ock",
+      "err or",
+      "sk ip",
+      "s ime",
+      "Ġan a",
+      "sime q",
+      "Ġcereb ral",
+      "ĠE X",
+      "av ed",
+      "roph y",
+      "id opsis",
+      "Ġbeh ind",
+      "Ġen ables",
+      "Ġind ustrial",
+      "ĠP ac",
+      "Ġdefin itions",
+      "Ġcataly tic",
+      "Ġdiss ip",
+      "erv ical",
+      "Ġcom mut",
+      "Ġrepe at",
+      "Ġch iral",
+      "Ġp ron",
+      "p ol",
+      "Ġgo ing",
+      "Ġmic roscope",
+      "Ġhealth care",
+      "ĠClass ification",
+      "tit ude",
+      "ĠFerm i",
+      "Ġh ttp",
+      "are st",
+      "Ġsupport ing",
+      "Ġw ood",
+      "n ight",
+      "Ġkine tics",
+      "Ġsubset s",
+      "Ġsub unit",
+      "ĠCan ada",
+      "at on",
+      "Ġaccur ately",
+      "Ġres istant",
+      "ĠïĢ ½",
+      "ric tion",
+      "Ġcham ber",
+      "ig ue",
+      "ĠPh il",
+      "Ġrec over",
+      "c s",
+      "Ġsp here",
+      "ĠSpec ifically",
+      "Ġan ne",
+      "Ġiniti ation",
+      "ĠT H",
+      "Ġb ud",
+      "ord ered",
+      "Ġdi electric",
+      "ĠCol lege",
+      "Ġproduc ing",
+      "Ġanten na",
+      "B s",
+      "ĠF rench",
+      "O X",
+      "ĠAmeric a",
+      "ĠâĢ Ķ",
+      "oun ting",
+      "ful ly",
+      "Ġserv ed",
+      "Ġresid ue",
+      "Ġarg uments",
+      "Ġp and",
+      "Ġcomp any",
+      "Ġcondition al",
+      "m ia",
+      "ĠQ CD",
+      "Ġviscos ity",
+      "Ġprosp ective",
+      "as onal",
+      "Ġdom inated",
+      "Ġpen et",
+      "op o",
+      "Ġn ine",
+      "ĠI ll",
+      "ĠVis ual",
+      "Ġfil es",
+      "Ġy east",
+      "Ġthan k",
+      "G N",
+      "re al",
+      "Ġver ified",
+      "ĠInd ian",
+      "Ġsti ff",
+      "rolog ical",
+      "Ġd ram",
+      "Ġt ight",
+      "ĠGerm an",
+      "ĠTechn ology",
+      "ĠAppro ach",
+      "rom atic",
+      "Ġac oustic",
+      "ti an",
+      "os in",
+      "ĠDep artment",
+      "ot ropy",
+      "Ġem pty",
+      "tri vial",
+      "of il",
+      "Ġal gebras",
+      "tex ts",
+      "Ġwe bs",
+      "Ġp ore",
+      "Ġpack et",
+      "T ime",
+      "im g",
+      "on y",
+      "ri tic",
+      "Ġveloc ities",
+      "ĠD ynamics",
+      "Ġcanc ers",
+      "Ġtr unc",
+      "ĠForm ation",
+      "ĠDon or",
+      "ĠM it",
+      "I ST",
+      "Ġconcl uded",
+      "Ġan tag",
+      "ĠSo ft",
+      "app end",
+      "Ġfrag ments",
+      "ĠPro f",
+      "Ġflu or",
+      "ĠJ ac",
+      "ĠS n",
+      "Ġle pt",
+      "Ġsplit ting",
+      "Ġsex ual",
+      "ĠF ore",
+      "ĠGen er",
+      "Ġneighbor hood",
+      "Ġben chmark",
+      "ĠR A",
+      "Ġdiv ision",
+      "iforn ia",
+      "Tr ue",
+      "Ġf uzzy",
+      "Ġt ro",
+      "c ents",
+      "Ġconstit u",
+      "ati al",
+      "as tern",
+      "ĠT im",
+      "Ġper ception",
+      "Ġsubst anti",
+      "Ġmac ro",
+      "Ġout l",
+      "ĠObs erv",
+      "pr ising",
+      "ok ed",
+      "orec tal",
+      "ĠCh o",
+      "ĠDiff erent",
+      "Ġinvestig ations",
+      "Ġconsist ency",
+      "i ents",
+      "ĠF OR",
+      "AS S",
+      "ĠV an",
+      "Ġsit uations",
+      "ĠB R",
+      "Ġinf rared",
+      "ym al",
+      "Ġpix els",
+      "Ġcar rier",
+      "s en",
+      "IN T",
+      "Ġeffici ently",
+      "D T",
+      "ĠEx pl",
+      "ion ic",
+      "Ġn aturally",
+      "Ġpro pos",
+      "Ġgu ide",
+      "Ġconcl usions",
+      "o on",
+      "Ġg rant",
+      "Ġinst ances",
+      "Ġreview ed",
+      "Ġelect romagnetic",
+      "Ġth reat",
+      "ed ia",
+      "ĠOptim ization",
+      "ĠB io",
+      "Ġtrig ger",
+      "ici ent",
+      "otyp ic",
+      "Ġst ret",
+      "Ġan tic",
+      "Ġtox ic",
+      "Ġsp inal",
+      "UP AC",
+      "Ġover view",
+      "o tion",
+      "Ġstraight forward",
+      "Ġpos itively",
+      "as te",
+      "Ġref erences",
+      "ul ose",
+      "ĠG re",
+      "Ġantag on",
+      "Ġshif ts",
+      "Ġd rawn",
+      "ĠWh ite",
+      "Ġfrac tional",
+      "Ġbund le",
+      "Ġexhib its",
+      "Ġreserv oir",
+      "ĠA lex",
+      "Ġaggreg ation",
+      "Ġcirc le",
+      "Ġprac tices",
+      "ĠC oval",
+      "ĠDist ribution",
+      "Ġt ang",
+      "ĠM ut",
+      "Ġreg ulate",
+      "osp here",
+      "i ro",
+      "AMI NO",
+      "v est",
+      "Ġphot os",
+      "Ġev ident",
+      "Ġbus iness",
+      "cont rol",
+      "Ġw orth",
+      "ĠPo isson",
+      "ĠArab idopsis",
+      "ĠT arget",
+      "Ġregul ates",
+      "ĠI r",
+      "ĠAd v",
+      "Ġens emble",
+      "pr ing",
+      "Ġp rice",
+      "ĠF L",
+      "ĠImp act",
+      "Ġevent ually",
+      "in ating",
+      "Ġcent rifug",
+      "f rame",
+      "Ġdiagram s",
+      "Ġt ag",
+      "Ġt ry",
+      "sur face",
+      "ĠIdentif iers",
+      "ra ined",
+      "Ġs ides",
+      "Ġin n",
+      "Ġflex ible",
+      "Ġsat ellite",
+      "Ġaff inity",
+      "Ġsum mer",
+      "G P",
+      "am b",
+      "Ġa qu",
+      "Str ing",
+      "t reatment",
+      "ĠD ynamic",
+      "math op",
+      "Ġno tice",
+      "n es",
+      "row ave",
+      "ves tig",
+      "Ġoutput s",
+      "Ġco herent",
+      "Ġillustr ate",
+      "Ġvalid ated",
+      "ĠSc hem",
+      "Ġask ed",
+      "b atch",
+      "Ġpur ified",
+      "Ġminim ize",
+      "ĠD E",
+      "U M",
+      "c heck",
+      "vari an",
+      "ĠG old",
+      "yl ene",
+      "I O",
+      "Ġch olesterol",
+      "Pub Chem",
+      "ĠK ore",
+      "ĠCount y",
+      "Ġi i",
+      "ĠM AP",
+      "ect omy",
+      "Ġsem antic",
+      "Ġcoll agen",
+      "Ġper ceived",
+      "ich ia",
+      "Ġadminist ered",
+      "con taining",
+      "ran k",
+      "In ChI",
+      "Ġirradi ation",
+      "Ġlog arithm",
+      "Ġg ames",
+      "Ġinj ected",
+      "ĠM Hz",
+      "Ġd ors",
+      "Ġevalu ating",
+      "ĠHy per",
+      "Ġchromat ography",
+      "p hen",
+      "ĠK ar",
+      "Ġan timicrobial",
+      "ri end",
+      "Ġdescrib ing",
+      "Ġw t",
+      "Ġhorm one",
+      "A K",
+      "ĠI UPAC",
+      "G a",
+      "Ġvit amin",
+      "Ġconn ections",
+      "u ous",
+      "ĠL ine",
+      "Ġbenef icial",
+      "c ases",
+      "ic ated",
+      "is ks",
+      "p arent",
+      "I d",
+      "er ies",
+      "r un",
+      "Ġm ind",
+      "it t",
+      "s ulf",
+      "z heimer",
+      "Ġinter f",
+      "V ert",
+      "Ġan th",
+      "olog ous",
+      "ĠL ife",
+      "Ġm ur",
+      "Ġper mut",
+      "ot ing",
+      "Ġneut rino",
+      "Ġb orn",
+      "p matrix",
+      "ĠCal ifornia",
+      "ag ent",
+      "Ġcoll isions",
+      "ĠN S",
+      "Ġh ippocamp",
+      "Ġpow der",
+      "Ġv aries",
+      "Ġepid em",
+      "ĠWe b",
+      "ul er",
+      "Ġinterest ed",
+      "Ġdevelopment al",
+      "Ġlength s",
+      "Ġcol our",
+      "Ġqu as",
+      "ĠR ich",
+      "E q",
+      "Ġinf ants",
+      "ĠP H",
+      "ophil a",
+      "Ġcaus ing",
+      "G e",
+      "mod ule",
+      "I B",
+      "Ġcontrib uted",
+      "ro se",
+      "Ġcy toplas",
+      "---------------- ----------------",
+      "Ġro ad",
+      "s ymmetric",
+      "U s",
+      "Ġweak ly",
+      "ti te",
+      "Ġdef ines",
+      "ĠP E",
+      "Ġmetabol ites",
+      "Ġl ob",
+      "Ġterm inal",
+      "Ġdemonstr ates",
+      "ĠAc ceptor",
+      "ĠC lo",
+      "Ġinfer red",
+      "Ġv ill",
+      "F irst",
+      "Ġneg lig",
+      "Ġw ireless",
+      "A b",
+      "par ticle",
+      "ois otopic",
+      "Ġexc ited",
+      "P M",
+      "Ġcons ecutive",
+      "ĠIs otype",
+      "Ġstim ulus",
+      "ĠM C",
+      "tim ate",
+      "ĠCoval ently",
+      "B onded",
+      "Ġy ellow",
+      "Ġall oy",
+      "d ensity",
+      "Ġfil ters",
+      "Ġampl ification",
+      "Ġw on",
+      "h t",
+      "Ġimp acts",
+      "Ġst aff",
+      "ĠâĪ Ģ",
+      "ĠIs omeric",
+      "Ġsm oking",
+      "Q u",
+      "Ġcapt ured",
+      "h aps",
+      "ĠN ULL",
+      "Ġri ver",
+      "c ount",
+      "Ġmanif est",
+      "Ġdiab etic",
+      "Ġalter ations",
+      "ĠRot atable",
+      "ĠP RO",
+      "ĠMon oisotopic",
+      "Ġï Ĥ",
+      "sp ect",
+      "Ġcataly st",
+      "Ġmodel ed",
+      "Ġp age",
+      "ĠR OS",
+      "ĠCanonical ized",
+      "ĠT w",
+      "Ġa ux",
+      "av age",
+      "ĠRam an",
+      "st o",
+      "per f",
+      "Ġreplac ement",
+      "ĠEn vironment",
+      "Ġac ting",
+      "p ati",
+      "ific ant",
+      "th rough",
+      "Ġsat uration",
+      "Ġt ip",
+      "Ġrec urrence",
+      "ĠHist ory",
+      "Ġprot ective",
+      "Ġbur den",
+      "ad o",
+      "y es",
+      "in st",
+      "A p",
+      "ĠS y",
+      "Ġph on",
+      "ĠâĪ ij",
+      "Ġgen otype",
+      "Ġcovari ance",
+      "Ġquick ly",
+      "ĠD u",
+      "Ġs ug",
+      "Ġdec line",
+      "ĠT B",
+      "Ġstrict ly",
+      "Ġmo ist",
+      "und red",
+      "ĠC B",
+      "ati le",
+      "ĠH F",
+      "Ġar ticles",
+      "Ġp s",
+      "ĠEn h",
+      "ist ing",
+      "Ġbi ology",
+      "Ġb odies",
+      "ĠA k",
+      "ĠNumer ical",
+      "ĠLag rangian",
+      "Ġdisc overed",
+      "Ġv ic",
+      "op es",
+      "Ġfrag ment",
+      "Ġt y",
+      "ism ic",
+      "Ġhep atic",
+      "Ġen riched",
+      "p an",
+      "Ġinflu ences",
+      "ĠL ake",
+      "col or",
+      "Ġenrich ment",
+      "oc hemistry",
+      "Ġun stable",
+      "ĠIg G",
+      "der ly",
+      "Ġe cos",
+      "Ġconcer ning",
+      "ĠR isk",
+      "Ġmarg in",
+      "Ġpath ogenesis",
+      "Ġp ump",
+      "Ġprelim inary",
+      "Ġtum our",
+      "F urther",
+      "az ole",
+      "Ġelectro des",
+      "Ġd ial",
+      "ub es",
+      "ĠN atural",
+      "ĠM ul",
+      "ĠïĢ Ń",
+      "Ġn ic",
+      "Ġimp ed",
+      "on ly",
+      "Ġcompar ative",
+      "rec tion",
+      "ak i",
+      "Ġre nd",
+      "Ġsp arse",
+      "Ġindic ator",
+      "l ocation",
+      "tis m",
+      "ac tivated",
+      "ĠP b",
+      "epti de",
+      "Ġend ogenous",
+      "Ġcent ers",
+      "a o",
+      "s w",
+      "Ġcons ensus",
+      "Ġattrib utes",
+      "Ġsaf e",
+      "Ġbelie ve",
+      "ov irus",
+      "Ġimmun ity",
+      "Ġfit ted",
+      "Ġcontrib utes",
+      "i able",
+      "Ġvirus es",
+      "Ġins ight",
+      "ĠNo vel",
+      "ĠAl zheimer",
+      "cep ted",
+      "ĠP t",
+      "Ġcent re",
+      "n at",
+      "Ġbios ynthesis",
+      "m its",
+      "Ġchem istry",
+      "Ġj us",
+      "an ish",
+      "Ġre frac",
+      "ĠT or",
+      "Ġpan els",
+      "Ġimp ly",
+      "Ġmat ched",
+      "us c",
+      "w ord",
+      "v ae",
+      "ĠSt ar",
+      "s yn",
+      "M at",
+      "Ġapplic able",
+      "ĠP seud",
+      "amp ions",
+      "ĠR en",
+      "Ġus age",
+      "ĠL ight",
+      "p rec",
+      "Ġfib rosis",
+      "Ġreconstr uc",
+      "ĠO N",
+      "ĠG Hz",
+      "G D",
+      "al gebra",
+      "ig er",
+      "Ġdec isions",
+      "inf ected",
+      "knowled g",
+      "Ġexpress ing",
+      "Ġmyocardi al",
+      "ord ination",
+      "Ġprogn ostic",
+      "Ġfibro bl",
+      "Ġaccel er",
+      "ĠAssess ment",
+      "Ġconstra ined",
+      "Ġalle le",
+      "r ide",
+      "Ġrequ est",
+      "abil istic",
+      "te b",
+      "Ġg a",
+      "Ġrec overed",
+      "Ġpro min",
+      "urs es",
+      "ĠH C",
+      "ĠM ur",
+      "ĠEq s",
+      "Ġdef ining",
+      "Ġm er",
+      "im age",
+      "Ġorgan isms",
+      "g rad",
+      "Ġref lected",
+      "el astic",
+      "e ties",
+      "dim ethyl",
+      "EL O",
+      "ra ndom",
+      "ĠDi agn",
+      "ercul osis",
+      "ro b",
+      "Ġmom ents",
+      "ĠE C",
+      "Ġexperi ences",
+      "erv ing",
+      "ĠN C",
+      "Ġvor tex",
+      "g re",
+      "struct ures",
+      "el t",
+      "Ġcar ry",
+      "ĠTh rough",
+      "Ġpre ced",
+      "rast ruct",
+      "it us",
+      "Ġpsych ological",
+      "Ġlimit ing",
+      "t wo",
+      "ĠB ound",
+      "ĠC re",
+      "ĠSm ith",
+      "Ġc ast",
+      "Ġcompe tition",
+      "s ch",
+      "Ġcap ability",
+      "ach ment",
+      "Ġinhib its",
+      "Ã °",
+      "ĠDiff erential",
+      "Ġautom atically",
+      "Ġg est",
+      "Ġw aters",
+      "Ġun iqu",
+      "z er",
+      "E qu",
+      "Ġstudy ing",
+      "Ġdi ed",
+      "Ġo s",
+      "Ġrecomb ination",
+      "unc il",
+      "Ġpath ogen",
+      "GF R",
+      "U V",
+      "en eration",
+      "ĠS ta",
+      "Ġinst ant",
+      "Ġpro ven",
+      "Ġd s",
+      "Ġd amp",
+      "N ext",
+      "ĠY oung",
+      "Ġpower ful",
+      "Ġwr iting",
+      "k l",
+      "Ġcare er",
+      "ĠCor ollary",
+      "N s",
+      "Ġï ¿½",
+      "ĠM il",
+      "Ġb urn",
+      "tic ular",
+      "ond on",
+      "P r",
+      "ĠL in",
+      "ĠJapan ese",
+      "ĠL ab",
+      "Ġst rip",
+      "pro tein",
+      "Ġh our",
+      "angle ment",
+      "angu ages",
+      "r d",
+      "par se",
+      "Ġemiss ions",
+      "H ence",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠ",
+      "Ġj ob",
+      "ĠA S",
+      "Ġax ial",
+      "ĠT ur",
+      "car bon",
+      "M F",
+      "ĠN E",
+      "Ġar ise",
+      "Ġlinear ly",
+      "Ġprol ong",
+      "Ġle ak",
+      "ĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠ",
+      "Ġmov ed",
+      "orb idity",
+      "Ġprofession al",
+      "c ode",
+      "os ine",
+      "Ġpol ic",
+      "Ġbond s",
+      "m ask",
+      "Ġconver ted",
+      "v ille",
+      "ec tious",
+      "par allel",
+      "ĠH al",
+      "ĠT GF",
+      "ment al",
+      "Ġread er",
+      "Ġstandard s",
+      "ag o",
+      "Ġ EN",
+      "Ġst ations",
+      "Ġnormal ization",
+      "ĠÎ ĺ",
+      "ch ain",
+      "W hat",
+      "Ġtom ography",
+      "Ġent ries",
+      "bl ue",
+      "ĠPre vious",
+      "i as",
+      "Ġquestionnai re",
+      "Ġh az",
+      "Ġhom ology",
+      "ver y",
+      "Ġnucle otide",
+      "ĠGen ome",
+      "Ġμ l",
+      "Ġutil ization",
+      "Ġpolym ers",
+      "ro te",
+      "Ġsmall est",
+      "cal c",
+      "Ġs pl",
+      "Ġt ension",
+      "Ġdis continu",
+      "al a",
+      "h ol",
+      "Ġdeterm ines",
+      "Ġpro j",
+      "ĠOver all",
+      "Ġb le",
+      "f o",
+      "Ġprinc iples",
+      "Ġinteract ing",
+      "Ġhard ware",
+      "l ife",
+      "ail s",
+      "Ġdifficult y",
+      "Ġcho ices",
+      "Ġc ard",
+      "Ġl act",
+      "Ġro ll",
+      "Ġquantif ied",
+      "ĠSc ientific",
+      "Ġland sc",
+      "al igned",
+      "Ġcompos ites",
+      "her ichia",
+      "Ġen velop",
+      "iti g",
+      "S te",
+      "Ġcomp et",
+      "Ġimpair ment",
+      "Ġclos ure",
+      "Ġreturn ed",
+      "Ġrece iver",
+      "Ġpe er",
+      "Ġcons ent",
+      "Ġult ras",
+      "Ġphot ons",
+      "Ġsup pose",
+      "Ġpredic ting",
+      "ĠâĬ ķ",
+      "Ġcomp an",
+      "Ġneglig ible",
+      "c urrent",
+      "um ber",
+      "Ġcomp atible",
+      "i op",
+      "ĠStruct ural",
+      "R ef",
+      "Ġs on",
+      "Ġequ ality",
+      "Ġconsist ed",
+      "Ġv ibr",
+      "ou pling",
+      "v ation",
+      "Ġover come",
+      "s uper",
+      "l ict",
+      "Ġpancre atic",
+      "G s",
+      "ap ed",
+      "as al",
+      "w an",
+      "Ġlat ent",
+      "Ġcover ing",
+      "Ġles ion",
+      "i ance",
+      "ĠF T",
+      "wo od",
+      "ject ure",
+      "ĠB C",
+      "link ed",
+      "ĠL aw",
+      "Ġem it",
+      "Ġunc lear",
+      "Ġpre m",
+      "ac ted",
+      "p olar",
+      "c re",
+      "Ġmod ulus",
+      "rop ath",
+      "S ub",
+      "am i",
+      "Ġp ick",
+      "ER R",
+      "Ġmove ments",
+      "N i",
+      "Ġmechan ics",
+      "od ic",
+      "Ġg al",
+      "ĠMan agement",
+      "h ost",
+      "ew ise",
+      "ĠT otal",
+      "ĠInflu ence",
+      "Ġ ubiqu",
+      "roph ys",
+      "Ġcap s",
+      "Ġparticip ant",
+      "Ġpol yp",
+      "t d",
+      "Ġiter ations",
+      "dom inal",
+      "B B",
+      "Ġcharacter s",
+      "Ġdevi ations",
+      "res istant",
+      "Ġmal aria",
+      "Ġrem ote",
+      "h skip",
+      "Ġunder went",
+      "u til",
+      "bl ock",
+      "ucl ide",
+      "Î ¦",
+      "elect ron",
+      "Ġsens ory",
+      "ĠSim ulation",
+      "Ġre ward",
+      "Ġpand emic",
+      "Ġb or",
+      "ynt hetic",
+      "Ġinvas ive",
+      "R F",
+      "ĠSm all",
+      "ĠF isher",
+      "val ent",
+      "ĠM I",
+      "roc ytes",
+      "ĠT E",
+      "Ġst re",
+      "Ġperturb ations",
+      "Ġsim plicity",
+      "ĠG rowth",
+      "ĠÎ ł",
+      "Ġin oc",
+      "ard ing",
+      "at um",
+      "m ulti",
+      "ĠD iv",
+      "an es",
+      "ac illus",
+      "Ġlife time",
+      "ĠH ep",
+      "Ġa z",
+      "us p",
+      "ĠAssum e",
+      "Ġbre aking",
+      "ĠAt t",
+      "ticip ants",
+      "Ġlumin osity",
+      "Ġdon or",
+      "par ams",
+      "oh yd",
+      "Ġpro gen",
+      "ĠP O",
+      "G O",
+      "ĠL eg",
+      "Ġbiomark ers",
+      "Ġr ural",
+      "Ġne on",
+      "gl uc",
+      "ĠP B",
+      "Ġgu id",
+      "Ġc ervical",
+      "p ace",
+      "Ġc ord",
+      "um n",
+      "Ġsub space",
+      "Ġatt ached",
+      "Ġdepos ited",
+      "Ġindic ators",
+      "ĠS F",
+      "qui re",
+      "Ġdiss olved",
+      "r ite",
+      "ĠN A",
+      "Ġj u",
+      "Ġadd ressed",
+      "Ġsupp ressed",
+      "Ġpneum onia",
+      "Ġs ession",
+      "ĠC he",
+      "ĠF er",
+      "Ġacc ordance",
+      "D es",
+      "Ġqu ar",
+      "Ġfit ness",
+      "Ġvi ability",
+      "os h",
+      "Ġphyl ogenetic",
+      "ect in",
+      "p at",
+      "ĠFran ce",
+      "Ġmess ages",
+      "Ġl oci",
+      "Ġconf lict",
+      "Ġrele vance",
+      "Ġinstruc tions",
+      "Ġsome what",
+      "chang ed",
+      "Ġcorrect ly",
+      "oz yg",
+      "av ig",
+      "ĠL at",
+      "Ġo varian",
+      "ĠR emark",
+      "j oint",
+      "ain t",
+      "w est",
+      "s ample",
+      "Ġdiver gence",
+      "Ġh air",
+      "ag onal",
+      "Ġm im",
+      "Ġim mediate",
+      "ĠP ort",
+      "Ġoff ers",
+      "Ġdepic ted",
+      "Ġhydro x",
+      "ĠT ow",
+      "Ġemerg ing",
+      "ou pled",
+      "Ġh undred",
+      "Ġadap ted",
+      "ell er",
+      "ĠRel ations",
+      "et te",
+      "Ġgast ro",
+      "Ġm orphism",
+      "Ġequip ment",
+      "p op",
+      "un ately",
+      "Ġtransplant ation",
+      "if iers",
+      "Ġel derly",
+      "on ucle",
+      "Ġref ers",
+      "ar ial",
+      "ĠCom mittee",
+      "Ġmalign ant",
+      "omon as",
+      "Ġall ocation",
+      "og ether",
+      "Ġnan ot",
+      "pl ot",
+      "ĠM es",
+      "Ġplan ar",
+      "ell s",
+      "s ource",
+      "ow ski",
+      "Ġn a",
+      "Ġcl ock",
+      "Ġamb ient",
+      "oc ene",
+      "Ġfluores cent",
+      "Ġval u",
+      "ĠM agnetic",
+      "Ġde part",
+      "phosph ate",
+      "Ġrough ly",
+      "Ġne ither",
+      "ĠAl tern",
+      "Ġst ay",
+      "Ġsp ot",
+      "ĠE nt",
+      "Ġsecond s",
+      "h ard",
+      "Ġrec urrent",
+      "Ġpat ch",
+      "Ġlimit ation",
+      "ĠD er",
+      "Ġsh arp",
+      "Ġexpect ation",
+      "ĠL ore",
+      "d ict",
+      "R eg",
+      "Ġneut roph",
+      "Ġn ur",
+      "Ġstar ts",
+      "ost asis",
+      "Ġorgan ized",
+      "Ġc DNA",
+      "or ient",
+      "ĠEx ample",
+      "ĠF und",
+      "ay lor",
+      "id ering",
+      "Ġtri ple",
+      "n ic",
+      "Ġattack s",
+      "ĠD ros",
+      "Ã ¨",
+      "ĠE M",
+      "Ġoptim um",
+      "Ġp ull",
+      "Ġc e",
+      "ery th",
+      "Ġr ating",
+      "Ġreproduc tive",
+      "Ġdec ades",
+      "Ġre place",
+      "L ist",
+      "ĠF ast",
+      "Ġred shift",
+      "op sy",
+      "ill a",
+      "do uble",
+      "ter a",
+      "Ġgo als",
+      "ĠS k",
+      "IN E",
+      "Ġbi ochemical",
+      "u int",
+      "Ġf etal",
+      "ĠRi emann",
+      "ur ies",
+      "Ġp p",
+      "Ġsymbol s",
+      "ĠK a",
+      "D i",
+      "ĠG alax",
+      "ĠComp ared",
+      "Ġc asc",
+      "Ġb its",
+      "Ġsc aff",
+      "Ġestim ator",
+      "ĠAd ditional",
+      "Ġimprove ments",
+      "ectiv es",
+      "Ġh ous",
+      "ĠM agn",
+      "Ġmultiv ariate",
+      "Ġag ric",
+      "v o",
+      "ut ter",
+      "ĠAc knowledg",
+      "s u",
+      "Ġam mon",
+      "Ġaim s",
+      "Ġz inc",
+      "Ġel ong",
+      "ĠG O",
+      "Q uestion",
+      "incl uding",
+      "Log P",
+      "Ġint ellig",
+      "Ġcon e",
+      "ĠFound ation",
+      "Ġimp aired",
+      "Ġill ness",
+      "ĠEsc herichia",
+      "Ġabund ant",
+      "s cal",
+      "ens ively",
+      "Ġneg atively",
+      "par ameter",
+      "Ġperme ability",
+      "dom ain",
+      "r ated",
+      "Ġep och",
+      "Ġadoles cents",
+      "Ġdef ic",
+      "ĠEstim ation",
+      "Ġrout ine",
+      "P er",
+      "t ol",
+      "Ġellip tic",
+      "ĠH E",
+      "obl ast",
+      "Ġre aches",
+      "Ġflux es",
+      "Ġs un",
+      "ĠAn aly",
+      "âĢ ľ",
+      "ĠX LogP",
+      "Ġfilter ing",
+      "ri an",
+      "ĠS cal",
+      "Ġp in",
+      "ĠTi O",
+      "im ents",
+      "Ġmarg inal",
+      "Ġrecomb inant",
+      "Ġenc our",
+      "Ġal umin",
+      "Ġt f",
+      "ataly tic",
+      "Ġobserv ational",
+      "Ġgeneral ization",
+      "Ġï£ ¬",
+      "Ġantib iotic",
+      "Ġgener ates",
+      "Ġd B",
+      "S pec",
+      "r ically",
+      "Ġvalu able",
+      "Ġtop ic",
+      "Ġterm in",
+      "Ġsem icon",
+      "Ġquantif ication",
+      "ub b",
+      "Ġkin em",
+      "err ing",
+      "Ġa eros",
+      "p ack",
+      "Ġfew er",
+      "Ġfat igue",
+      "Ġgo es",
+      "Ġn ight",
+      "ĠU s",
+      "âĢ ¬",
+      "ĠPr inc",
+      "Ġsp ring",
+      "Ġconcer ns",
+      "Ġsm art",
+      "Ġsec ret",
+      "Ġmm ol",
+      "Ġbel ief",
+      "D C",
+      "Ġsubstanti ally",
+      "âĪ ĩ",
+      "Ġsubstit ution",
+      "map sto",
+      "sk y",
+      "ill ance",
+      "Ġstud ent",
+      "ok ine",
+      "Ġinter ior",
+      "Ġeigen value",
+      "m y",
+      "Ġclos er",
+      "eren ti",
+      "Ġec ological",
+      "ĠFig ures",
+      "oly tic",
+      "Ġar rays",
+      "ĠC as",
+      "Ġlo ops",
+      "Ġcorrec ted",
+      "Ġr he",
+      "Ġin version",
+      "Ġprefer red",
+      "um ab",
+      "ĠD I",
+      "Ġadequ ate",
+      "ir m",
+      "Ġim plicit",
+      "sh ip",
+      "Ġplay ers",
+      "Ġdelay ed",
+      "Ġw inter",
+      "Ġvul ner",
+      "Ġshap es",
+      "Ġstain ed",
+      "ĠM ajor",
+      "Ġhierarch ical",
+      "ĠD ig",
+      "ers ion",
+      "ĠE fficient",
+      "Ġwall s",
+      "d frac",
+      "Ġclass ifier",
+      "Ġmon ol",
+      "Ġupd ated",
+      "Ġm ature",
+      "ĠL I",
+      "ear ing",
+      "Ġf inger",
+      "oun ter",
+      "ank ton",
+      "Wh ile",
+      "Ġreal istic",
+      "ĠC amp",
+      "Ġf illed",
+      "Ġde ad",
+      "ĠPac ific",
+      "Ï ĩ",
+      "ĠDav id",
+      "Ġaddi tive",
+      "ench ymal",
+      "Ġobs er",
+      "Ġst ere",
+      "Ġultras ound",
+      "ĠPred ic",
+      "Ġend s",
+      "section al",
+      "m as",
+      "om at",
+      "iv ity",
+      "Ġhand le",
+      "Ġmetast atic",
+      "ole t",
+      "r yp",
+      "AC E",
+      "Ġpor ous",
+      "Ġconcer n",
+      "it ored",
+      "Ġcir cles",
+      "Ġemo tional",
+      "g ered",
+      "Ġf riction",
+      "f irst",
+      "oph y",
+      "es cop",
+      "ad ed",
+      "Ġres olved",
+      "ER S",
+      "Ġpath ogens",
+      "Ġgrad ually",
+      "ĠB rain",
+      "x f",
+      "an ium",
+      "a el",
+      "N ew",
+      "Ġcytok ine",
+      "ĠB P",
+      "Ġspecim en",
+      "ole an",
+      "Ġt axon",
+      "Ġsequ ential",
+      "κ B",
+      "ad emic",
+      "pl ings",
+      "~ ~",
+      "erm al",
+      "t ree",
+      "Ġcaus al",
+      "ari an",
+      "Ġc rop",
+      "op ol",
+      "ch annel",
+      "ĠM ex",
+      "Ġcl on",
+      "ĠRec ently",
+      "ĠIn vestig",
+      "Ġrecommend ations",
+      "form at",
+      "ĠM ET",
+      "Ġsent ence",
+      "Ġb p",
+      "ĠG W",
+      "Ġrec ording",
+      "Ġp le",
+      "to tic",
+      "Ġï£ ·",
+      "Ġrang ed",
+      "en tion",
+      "obacter ia",
+      "cep tions",
+      "ĠIm port",
+      "d ynamic",
+      "por ary",
+      "G iven",
+      "Ġturb ulence",
+      "Ġg ram",
+      "Ġequ ally",
+      "c d",
+      "ĠO s",
+      "Ġturn s",
+      "Ġdetect ing",
+      "ati o",
+      "gen erate",
+      "gra de",
+      "Ġcirc ulation",
+      "Ġmanufacture r",
+      "L a",
+      "ĠH ilbert",
+      "T s",
+      "in tegr",
+      "Ġbelong s",
+      "ĠIntern et",
+      "ang l",
+      "ĠâĬ ¥",
+      "ĠDros ophila",
+      "uclide an",
+      "t an",
+      "Ġext ends",
+      "Ġexpand ed",
+      "ill in",
+      "squ are",
+      "ys acchar",
+      "Ġquantif y",
+      "Ġpuls es",
+      "Ġves ic",
+      "ĠN K",
+      "ores cence",
+      "ĠPh osph",
+      "Ġv ision",
+      "ĠHu ang",
+      "ĠResp onse",
+      "h ouse",
+      "ear s",
+      "Ġe g",
+      "Ġac cepted",
+      "ĠT M",
+      "amet ric",
+      "Ġpath ological",
+      "Ġrecruit ment",
+      "AT A",
+      "Ġfig ures",
+      "ĠP ress",
+      "Ġal igned",
+      "Ġpost operative",
+      "ĠMe V",
+      "Ġconsider ably",
+      "Ġconform al",
+      "ĠIs land",
+      "num ber",
+      "Ġautom atic",
+      "Ġs plic",
+      "Ġcyt os",
+      "Ġdesc rip",
+      "ĠS ant",
+      "l ies",
+      "u ity",
+      "it one",
+      "E CT",
+      "ĠB on",
+      "Ġdis app",
+      "bo ard",
+      "or rh",
+      "Ġcalc ulating",
+      "ne e",
+      "ĠMe as",
+      "Ġgen omes",
+      "Ġphot oc",
+      "Ġread ily",
+      "ov ine",
+      "ĠDe v",
+      "Ġsat ur",
+      "Ġkind s",
+      "ĠP K",
+      "Ġro d",
+      "Ġj unction",
+      "ĠH A",
+      "Ġdesign s",
+      "h n",
+      "Ġorder ing",
+      "Ġcosm ological",
+      "Ġpil ot",
+      "Ġcol orectal",
+      "ĠL ondon",
+      "ĠDir ac",
+      "C ont",
+      "ĠW ind",
+      "ĠT re",
+      "id in",
+      "ĠïĢ «",
+      "ilt ration",
+      "More over",
+      "Ġre tention",
+      "tim ately",
+      "hydro gen",
+      "d el",
+      "bol ic",
+      "ĠQu anti",
+      "per iod",
+      "Ġretrie val",
+      "at ase",
+      "end icular",
+      "ulti es",
+      "R S",
+      "N H",
+      "Ġin formed",
+      "Ġfil tered",
+      "m embrane",
+      "Ġstiff ness",
+      "ĠO cean",
+      "ĠS Y",
+      "Ġl ot",
+      "ĠFig s",
+      "Ġans w",
+      "ĠEng land",
+      "ĠAtl antic",
+      "process ing",
+      "Ġdog s",
+      "Ġl ie",
+      "Ġun ion",
+      "ĠT an",
+      "Ġhal o",
+      "Ġcontinuous ly",
+      "B u",
+      "A MP",
+      "ĠAp p",
+      "Ġmoist ure",
+      "Ġth yroid",
+      "Ġaccompan ied",
+      "Ġfol d",
+      "Ġorig inally",
+      "Ġsp an",
+      "ĠF A",
+      "conn ected",
+      "Ġrec urs",
+      "vi an",
+      "ĠEqu ations",
+      "en a",
+      "arcin oma",
+      ".. ..",
+      "Ġdisc rep",
+      "U H",
+      "Ð ¾",
+      "ang er",
+      "Ġmon itored",
+      "Ġinflu enza",
+      "Ġs ure",
+      "bl ack",
+      "o e",
+      "Ġall oc",
+      "Ġhabit at",
+      "op henyl",
+      "Ġvent ricular",
+      "Ġpolic ies",
+      "am ate",
+      "Ġreport ing",
+      "Ġsol uble",
+      "======== ========",
+      "Ġdip ole",
+      "Ġirre ducible",
+      "ĠP rec",
+      "acet yl",
+      "Ġth read",
+      "ĠAppro xim",
+      "Ġm apped",
+      "i pro",
+      "Ġt ropical",
+      "S ch",
+      "ĠAN OVA",
+      "Ġl anguages",
+      "ic ine",
+      "ĠF amily",
+      "f unctions",
+      "E F",
+      "Ġnut rient",
+      "Ġanalyz ing",
+      "ines cence",
+      "Ġthrom b",
+      "Ġk it",
+      "Ġmamm alian",
+      "op totic",
+      "Ġequip ped",
+      "on a",
+      "Ġqu e",
+      "Ġc ame",
+      "Ġsimpl ified",
+      "Ġdec ays",
+      "Ġpass ive",
+      "Ġdele tion",
+      "Ġobtain ing",
+      "Ġmixt ures",
+      "Ġprim ers",
+      "ĠP sy",
+      "os c",
+      "om ent",
+      "Ġchlor ide",
+      "ĠPa ul",
+      "st art",
+      "int estinal",
+      "hel ium",
+      "ar th",
+      "od ot",
+      "Ġf its",
+      "Ġsqu ares",
+      "ĠCar di",
+      "ak a",
+      "rib uted",
+      "Ġinequ alities",
+      "omet hing",
+      "hed ral",
+      "ĠF uture",
+      "Ġgl i",
+      "Ġmetall ic",
+      "Ġfac ilities",
+      "Ġob st",
+      "poss ible",
+      "Ġz ones",
+      "uc id",
+      "Ġdr ift",
+      "d epend",
+      "val ued",
+      "Ġn ons",
+      "Ġworld wide",
+      "Ġtr ust",
+      "Ġso le",
+      "ĠLe vel",
+      "ĠS ha",
+      "Ġregard less",
+      "Ġspectrom etry",
+      "duc tor",
+      "le uk",
+      "Ġsk ills",
+      "Ġincorpor ated",
+      "Ġlear ned",
+      "Ġ ure",
+      "Ġext inc",
+      "OD U",
+      "Ġgrain s",
+      "ater n",
+      "ĠInd ex",
+      "com put",
+      "u a",
+      "Ġcont amination",
+      "ĠA ff",
+      "un ing",
+      "Ġas ymmetric",
+      "Ġopen ing",
+      "Ġb at",
+      "Ġag ree",
+      "IT Y",
+      "ĠChang es",
+      "organ ic",
+      "ĠR ay",
+      "ĠH and",
+      "n i",
+      "in ic",
+      "Ġr isks",
+      "Ġst ock",
+      "Ġnec k",
+      "Ġvol umes",
+      "ĠP rac",
+      "Ġincreasing ly",
+      "S c",
+      "os es",
+      "GF P",
+      "Ġass ignment",
+      "ĠF ed",
+      "osp it",
+      "Ġoverex pression",
+      "Ġm aster",
+      "Ġo pt",
+      "il er",
+      "in variant",
+      "Ġconver ges",
+      "Sim ilar",
+      "n y",
+      "Ġst ore",
+      "Ġelev ation",
+      "Ġco al",
+      "he t",
+      "it em",
+      "PL C",
+      "oh ist",
+      "G en",
+      "ĠC hem",
+      "ĠC ost",
+      "p air",
+      "Ġnumer ically",
+      "Ġpre ference",
+      "ĠN ucle",
+      "ĠB D",
+      "T I",
+      "ĠH yp",
+      "ro y",
+      "T e",
+      "ĠF in",
+      "Ġclaim s",
+      "ib ilities",
+      "Ġlar vae",
+      "im a",
+      "emb ly",
+      "Ġc it",
+      "L L",
+      "Ġsil ica",
+      "ĠV I",
+      "Ġreach ing",
+      "O f",
+      "ĠAustr alian",
+      "t ub",
+      "w orld",
+      "on i",
+      "ĠF P",
+      "Ġbrief ly",
+      "ĠDes cription",
+      "Î ¶",
+      "ch arg",
+      "Ġc is",
+      "ĠC at",
+      "Ġrec ip",
+      "Ġemerg ency",
+      "Ġst rand",
+      "Ġreal ized",
+      "pos ing",
+      "ot ope",
+      "Ġmaintain ing",
+      "ĠCh rist",
+      "Ġcre ating",
+      "Ġembry os",
+      "Ġs keletal",
+      "Ġag es",
+      "rep resent",
+      "C r",
+      "Ġestim ating",
+      "Ġre ar",
+      "ĠY u",
+      "ĠP i",
+      "m g",
+      "Ġflo at",
+      "ĠR oy",
+      "p us",
+      "Ġch ick",
+      "Ġmicrobi ota",
+      "vas ive",
+      "ĠB ern",
+      "ĠPat tern",
+      "l ines",
+      "Ġfl ood",
+      "ĠL ou",
+      "ilit ary",
+      "ros ion",
+      "Ġsurve ys",
+      "F I",
+      "ia e",
+      "Ġse arc",
+      "m ol",
+      "Ġt itle",
+      "ĠMach ine",
+      "Ġcirc uits",
+      "ĠNum ber",
+      "z i",
+      "ĠB MI",
+      "Ġautom ated",
+      "plic ate",
+      "ĠL PS",
+      "Ġelectro chemical",
+      "Ġwebs ite",
+      "Ġanis otropy",
+      "Ġr ings",
+      "Ġin nov",
+      "b its",
+      "w in",
+      "ĠN AD",
+      "Acc ording",
+      "ĠCon n",
+      "ure us",
+      "ĠFe ature",
+      "ĠIn stead",
+      "C omp",
+      "it udes",
+      "M o",
+      "Ġsc ope",
+      "tif ication",
+      "ĠI S",
+      "ĠNe ut",
+      "Ġreg ulating",
+      "c oding",
+      "Ġrow s",
+      "h l",
+      "ĠK n",
+      "ist or",
+      "ampions hip",
+      "Ġpromin ent",
+      "Ġr s",
+      "um atic",
+      "A m",
+      "Ġdifferenti ally",
+      "ug in",
+      "Ġadv ance",
+      "ph ys",
+      "Ġsh aring",
+      "Ġar t",
+      "v acy",
+      "ti tions",
+      "Ġst yle",
+      "Fig ures",
+      "Ġg lu",
+      "Ġvacc ination",
+      "ĠOp tical",
+      "flu id",
+      "ĠF re",
+      "Ġgradi ents",
+      "op hyl",
+      "ĠP ubl",
+      "Ġacc retion",
+      "ĠâĢ² âĢ²",
+      "ress ing",
+      "Ġtrans mitted",
+      "Ġnerv ous",
+      "um ar",
+      "Ġreview s",
+      "Ġgen otypes",
+      "low er",
+      "ĠE V",
+      "Ġcont ract",
+      "ati bility",
+      "Ġchild hood",
+      "Ġon c",
+      "Ġbi ofil",
+      "Ġaut ophagy",
+      "Ġads orb",
+      "ĠSup port",
+      "Ġlig ands",
+      "p ower",
+      "rec tional",
+      "ĠR ap",
+      "sim ilar",
+      "Ġinf arc",
+      "Ġelectro ly",
+      "Ġinc ome",
+      "ar ity",
+      "ĠA v",
+      "er ic",
+      "Ġclin ically",
+      "un ch",
+      "Ġattrib ute",
+      "Ġcomm and",
+      "rib utions",
+      "Ġgly c",
+      "Ġtranscri pts",
+      "ogram s",
+      "Ġassess ing",
+      "F O",
+      "script style",
+      "j i",
+      "ric k",
+      "en vironment",
+      "Ġlaw s",
+      "Ġnorm ally",
+      "Ġdeple tion",
+      "ĠR O",
+      "Ġenc oded",
+      "h ma",
+      "Ġbran ches",
+      "Ġarg s",
+      "oun ger",
+      "or ge",
+      "um ps",
+      "Ġview ed",
+      "Ġult r",
+      "R R",
+      "uls ion",
+      "ĠH or",
+      "Ġf ro",
+      "ĠMeasure ment",
+      "x x",
+      "erm an",
+      "ĠO nce",
+      "Ġorient ed",
+      "ĠP oint",
+      "Ġto wn",
+      "Ġformul as",
+      "S Y",
+      "ĠA M",
+      "Ġconsider ations",
+      "ĠT C",
+      "ĠK it",
+      "Ġact in",
+      "Ġplas mid",
+      "Ġhistor ical",
+      "Ġd ye",
+      "Ġhe ur",
+      "ĠLe ague",
+      "ĠM ad",
+      "Ġgra ft",
+      "Ġsil ver",
+      "O ver",
+      "ĠC os",
+      "ograph ical",
+      "Ġprecurs or",
+      "r us",
+      "Ġregard ed",
+      "ĠH am",
+      "f unctional",
+      "iv eness",
+      "ffici ency",
+      "ig ene",
+      "oc ol",
+      "Ġcum ulative",
+      "Ġse asonal",
+      "Ġm u",
+      "ĠB an",
+      "omy cin",
+      "Ġb ool",
+      "ĠM ag",
+      "ĠAn al",
+      "enti a",
+      "a ign",
+      "Ġfoot ball",
+      "act ing",
+      "Ġreturn s",
+      "ĠT om",
+      "sh aped",
+      "it ance",
+      "ĠExperim ent",
+      "ĠO S",
+      "Ġabs ent",
+      "ran ial",
+      "Ġtherap ies",
+      "O p",
+      "o unced",
+      "AT E",
+      "Val ue",
+      "g reen",
+      "Ġveget ation",
+      "D s",
+      "Ġinc om",
+      "Ã §",
+      "Ġm arrow",
+      "ĠCo uncil",
+      "Ġinv est",
+      "Ġcl ub",
+      "T rans",
+      "dev ice",
+      "Ġv ibration",
+      "ĠX u",
+      "//// ////",
+      "ĠH en",
+      "vi er",
+      "Ġanalog ous",
+      "Ġd elta",
+      "Ġsal ine",
+      "Ġrequ iring",
+      "Ġneur on",
+      "o o",
+      "ĠQ uality",
+      "Ġte ac",
+      "ĠE c",
+      "L i",
+      "Ġpubl ication",
+      "ĠPhys ics",
+      "Ġp pm",
+      "th ase",
+      "Ġcre ation",
+      "ĠA ge",
+      "Ġbelong ing",
+      "Ġion ic",
+      "ĠS I",
+      "u ating",
+      "end if",
+      "ĠC our",
+      "Ð °",
+      "Ġd ots",
+      "Ġe ast",
+      "ar com",
+      "Ġâ ĩ",
+      "Ġr ights",
+      "ess ions",
+      "Ġvers ions",
+      "ĠF ree",
+      "ĠSt ress",
+      "Ġsed iments",
+      "Ġm itig",
+      "Ġb ow",
+      "ĠAc t",
+      "ĠCar bon",
+      "t here",
+      "te en",
+      "Ġphen otypes",
+      "Ġne arest",
+      "ĠPot ential",
+      "Ġde form",
+      "Ġreflec ts",
+      "Ġpart ners",
+      "Ġan est",
+      "Ġad vers",
+      "ĠF actor",
+      "Ġconven ient",
+      "ul os",
+      "ĠP ur",
+      "ĠM er",
+      "Ġfl ag",
+      "Ġtri ang",
+      "Ġseed s",
+      "Ġf if",
+      "ob il",
+      "ĠC K",
+      "men tioned",
+      "Ġv apor",
+      "og ue",
+      "Ġpredic tor",
+      "O ut",
+      "Ġcomple tion",
+      "ĠS eg",
+      "Ġdiff use",
+      "Ġra ised",
+      "Ġco ordination",
+      "Ġsyn aptic",
+      "ĠB or",
+      "ĠB ol",
+      "Ġpolymer ase",
+      "Ġwhe at",
+      "Ġinser tion",
+      "Ġes c",
+      "ĠW al",
+      "Ġdist al",
+      "transfer ase",
+      "Ġinter faces",
+      "Ġins u",
+      "Ġpoor ly",
+      "Ġa ureus",
+      "Ġben z",
+      "Ġun iverse",
+      "ĠInter action",
+      "ĠF rame",
+      "ĠIm aging",
+      "Ġexpl oration",
+      "ĠEngine ering",
+      "ĠB esides",
+      "ti a",
+      "Ġen um",
+      "an ine",
+      "Ġto t",
+      "ĠE duc",
+      "Ġderiv ation",
+      "Ar ray",
+      "yl oid",
+      "ĠAr ch",
+      "is en",
+      "ac ity",
+      "ak ers",
+      "Ġshe et",
+      "ĠE st",
+      "Ġwe ar",
+      "Ġ eryth",
+      "EC K",
+      "hem atics",
+      "Ġarter ial",
+      "cript style",
+      "scripts criptstyle",
+      "echan ical",
+      "Ġparticip ation",
+      "c her",
+      "ur ance",
+      "ĠF R",
+      "ĠC V",
+      "Ġcomplement ary",
+      "ain e",
+      "empt y",
+      "Ġdig es",
+      "Ġexpon ent",
+      "Ġsim ulate",
+      "U E",
+      "Ġantib iotics",
+      "ĠUn ivers",
+      "Ġpath ology",
+      "ther mal",
+      "p a",
+      "Ġstress es",
+      "ĠLabor atory",
+      "N ode",
+      "Ġle ave",
+      "ash ing",
+      "Ġdisc re",
+      "Ġsusp ension",
+      "ree k",
+      "Ġschedul ing",
+      "ĠD A",
+      "ary n",
+      "ĠNa Cl",
+      "st rain",
+      "ST R",
+      "ĠC ong",
+      "ol f",
+      "Ġcal ibr",
+      "ĠOptim al",
+      "Ġ ó",
+      "G l",
+      "ĠR h",
+      "Ġdiffic ulties",
+      "Ġvess els",
+      "Ġas ymmetry",
+      "Ġco herence",
+      "ĠTaxon omy",
+      "Ġp ed",
+      "ĠH ouse",
+      "tit udes",
+      "ĠF ar",
+      "O Y",
+      "Ġconcentr ated",
+      "Ġsign alling",
+      "Ġfung al",
+      "Ġconsist ently",
+      "Ġenh ances",
+      "Ġfore cast",
+      "Ġc ubic",
+      "ĠE P",
+      "Ġparticip ate",
+      "ĠPl ant",
+      "r isk",
+      "A nd",
+      "ad ic",
+      "of lu",
+      "Ġsper m",
+      "ĠCh ris",
+      "N D",
+      "col on",
+      "Ġf aces",
+      "Ġtub erculosis",
+      "ryst al",
+      "flo or",
+      "up s",
+      "Ġg ray",
+      "ĠP ublic",
+      "t ensor",
+      "Ġrig id",
+      "Ġeas tern",
+      "ĠItal y",
+      "Ġsign atures",
+      "Ġshall ow",
+      "ó n",
+      "ĠC e",
+      "Ġpro jects",
+      "Ġro uting",
+      "Ġpredic ts",
+      "ĠFe atures",
+      "ĠDist rict",
+      "Ġcar rying",
+      "ĉ ĠĠĠĠ",
+      "ĠT O",
+      "H M",
+      "d ings",
+      "Ġre normal",
+      "Ġb ring",
+      "p in",
+      "al ed",
+      "Ġcloud s",
+      "n ames",
+      "ox in",
+      "Ġperp endicular",
+      "W T",
+      "ers hip",
+      "Ġrec on",
+      "Ġwork ed",
+      "ĠâĢ «",
+      "rastruct ure",
+      "Ġpo inted",
+      "E V",
+      "ĠT aylor",
+      "Ġhep atitis",
+      "Ġorb its",
+      "ĠF actors",
+      "c ellular",
+      "Ġf ocal",
+      "Ġbo ost",
+      "Ġmic rowave",
+      "ĠPro ject",
+      "B F",
+      "Ġpoli tical",
+      "Ġsupplement ed",
+      "Ġillustr ates",
+      "Ġide as",
+      "ĠDr ug",
+      "ob ile",
+      "ĠH O",
+      "Ġrobust ness",
+      "ros ine",
+      "ĠN ormal",
+      "Ġstim ulated",
+      "Ġimped ance",
+      "fort unately",
+      "zym e",
+      "Ġbar riers",
+      "act ory",
+      "lear ly",
+      "Ġpre print",
+      "sens itive",
+      "Ġturb ulent",
+      "th ing",
+      "Ġbo ard",
+      "Ġp it",
+      "Ġintegr ity",
+      "Ġrot ating",
+      "ud a",
+      "Ġv enti",
+      "ĠSN Ps",
+      "Ġcorrespond ence",
+      "Ġvisual ization",
+      "av ail",
+      "Ġbe ams",
+      "ĠCont inu",
+      "Ġpers istent",
+      "Ġb ath",
+      "Ġmi RNAs",
+      "Ġcust om",
+      "Ġord inary",
+      "Ġgener ators",
+      "Ġb ridge",
+      "Ġdom in",
+      "am y",
+      "Ġlo oking",
+      "t able",
+      "F alse",
+      "Ġsoil s",
+      "Ġmat ches",
+      "Ġprog ressive",
+      "st ates",
+      "ĠSh ort",
+      "Ġco res",
+      "Ġintro ducing",
+      "Ġar rest",
+      "Ġtext ure",
+      "Ġdors al",
+      "Ġd rain",
+      "iz oph",
+      "ĠQ ue",
+      "Ã ±",
+      "dis c",
+      "Ind ex",
+      "Ġext ensively",
+      "Ġplastic ity",
+      "Ġre ally",
+      "ĠEr ror",
+      "Ġsugg es",
+      "Ġconsequ ently",
+      "Ġperform s",
+      "lik ely",
+      "ive red",
+      "Ġtherm odynamic",
+      "Ġk er",
+      "Ġacet ate",
+      "Ġg ets",
+      "leq slant",
+      "Ġpredict ors",
+      "ĠSw ed",
+      "n an",
+      "he ter",
+      "Ġanomal y",
+      "Ġoper ational",
+      "Ġret rospective",
+      "Ġt ends",
+      "ad en",
+      "Ġb order",
+      "Ġmet hanol",
+      "ĠEn ter",
+      "Ġcoll apse",
+      "Ġpurch ased",
+      "D a",
+      "ĠH T",
+      "Ġf ulf",
+      "Ġcr ust",
+      "st one",
+      "Ġpen al",
+      "Ġtun n",
+      "ĠTem perature",
+      "Ġpot ent",
+      "lec ule",
+      "Ġco vers",
+      "Ġbatter y",
+      "Ġbe g",
+      "Ġor gans",
+      "ĠTh omas",
+      "Ġsol ub",
+      "oc rine",
+      "ĠSp in",
+      "Ġinterest s",
+      "d oc",
+      "Ġundergo ing",
+      "u i",
+      "Ġin herent",
+      "Ġintegr als",
+      "ira ble",
+      "as hi",
+      "Ġreg eneration",
+      "Ġinf lation",
+      "man if",
+      "ĠRec ognition",
+      "Ġdisplay s",
+      "An other",
+      "Ġcont amin",
+      "j unction",
+      "Ġcop ies",
+      "MR I",
+      "Ġvehic les",
+      "G et",
+      "Ġper haps",
+      "Ġw est",
+      "Ġint ensive",
+      "Ġs omething",
+      "Ġhypox ia",
+      "Ġcou plings",
+      "Ġfeas ibility",
+      "az ine",
+      "un ic",
+      "in er",
+      "ĠI T",
+      "Ġdist rict",
+      "ĠJ ames",
+      "e val",
+      "Ġplace bo",
+      "a que",
+      "Ġel ucid",
+      "ĠJac ob",
+      "Ġcoun ting",
+      "Ġflex ibility",
+      "Ġper man",
+      "Ġadv ances",
+      "ul ph",
+      "Ġent anglement",
+      "Ġinte gers",
+      "Ġfocus ing",
+      "k ov",
+      "Ġh ospit",
+      "Ġap plies",
+      "Ġc ot",
+      "S m",
+      "ass ium",
+      "Ġdocument ed",
+      "Ġload ed",
+      "Ġre ly",
+      "Ġinf ectious",
+      "Ġprob es",
+      "Ġhighlight ed",
+      "Ġp ediatric",
+      "Ġwe ather",
+      "Ġman ual",
+      "Ġc ation",
+      "Ġinterp olation",
+      "ĠSte p",
+      "ĠK al",
+      "D H",
+      "d b",
+      "izoph ren",
+      "ad er",
+      "car b",
+      "Ġag on",
+      "orph ous",
+      "t ors",
+      "at z",
+      "Ġb if",
+      "Ġcharg es",
+      "ĠAg ain",
+      "Ġb ron",
+      "ĠG over",
+      "Ġmin ing",
+      "a ver",
+      "Ġearth qu",
+      "Ġview s",
+      "Ġsc ene",
+      "par ameters",
+      "Ġbro ken",
+      "T est",
+      "ĠS um",
+      "ĠP rom",
+      "Î Ľ",
+      "Ġcut off",
+      "Ġb irds",
+      "Ġar ising",
+      "ĠA I",
+      "ĠC E",
+      "Ġpron ounced",
+      "asp ase",
+      "Ġint ended",
+      "Ġaff ine",
+      "Ġur ine",
+      "Ġbelie ved",
+      "ĠPrim ary",
+      "ĠCon f",
+      "Ġab dominal",
+      "sp in",
+      "un iform",
+      "ĠSt ochastic",
+      "ĠPro v",
+      "Ġmi RNA",
+      "ĠB ell",
+      "B O",
+      "ĠSoft ware",
+      "ĠT s",
+      "ut ri",
+      "ick ing",
+      "i en",
+      "Ġmic ros",
+      "ĠN R",
+      "Ġleuk emia",
+      "Ġsuper nat",
+      "f amily",
+      "Ġall oys",
+      "ĠP ET",
+      "ĠA bs",
+      "ĠG A",
+      "ĠQu antitative",
+      "L o",
+      "Ġis land",
+      "sec ond",
+      "pec tives",
+      "Ġlat ency",
+      "ang i",
+      "Ġfl ight",
+      "ĠE uclidean",
+      "em y",
+      "ĠBl ood",
+      "leuk in",
+      "L T",
+      "en h",
+      "Ġs we",
+      "Ġunit ary",
+      "ĠRep ublic",
+      "Ġstructure d",
+      "ĠS en",
+      "M n",
+      "cent ric",
+      "Ġtrans genic",
+      "Ġhelp ful",
+      "py x",
+      "Ġhome ostasis",
+      "N a",
+      "Ġpass ed",
+      "Ġe yes",
+      "Ġab stract",
+      "ul se",
+      "Ġmir ror",
+      "Ġregul ator",
+      "Ġmur ine",
+      "load ed",
+      "Ġmod ular",
+      "Ġlandsc ape",
+      "ic ks",
+      "Ġs now",
+      "Ġb ovine",
+      "ell i",
+      "Ġdatab ases",
+      "Ġout break",
+      "l arg",
+      "ĠR un",
+      "B E",
+      "Ġsur prising",
+      "Ġaccept able",
+      "Ġrot ational",
+      "p g",
+      "F E",
+      "w ik",
+      "Ġy ounger",
+      "ash ion",
+      "Ġmic roscopic",
+      "reg ation",
+      "Ġfib r",
+      "ĠPl an",
+      "Ġha pl",
+      "Ġmanif olds",
+      "Ġout per",
+      "Ġcho osing",
+      "e per",
+      "Ġke V",
+      "ĠT yp",
+      "p read",
+      "nt z",
+      "ĠRe port",
+      "ĠMat rix",
+      "Ġint u",
+      "Ġproper ly",
+      "og ly",
+      "oscop ic",
+      "ĠA MP",
+      "ĠB M",
+      "Ġelement ary",
+      "kele ton",
+      "Ġsyn thase",
+      "Ġion ization",
+      "b es",
+      "oph age",
+      "duc es",
+      "acc o",
+      "Ġprot ect",
+      "ĠCo ul",
+      "Ġsp ent",
+      "Ġm and",
+      "Ġh ind",
+      "flu or",
+      "ĠG ood",
+      "Ġdo ing",
+      "Ob ject",
+      "duc ts",
+      "o yl",
+      "chi atric",
+      "Ġo v",
+      "c el",
+      "Ġb ases",
+      "Ġmitochond ria",
+      "p ted",
+      "art z",
+      "Ġb rown",
+      "Ġequ als",
+      "ti ble",
+      "Ġopportun ity",
+      "az ol",
+      "Ġoff icial",
+      "ail ed",
+      "Ġur inary",
+      "ĠH an",
+      "B e",
+      "res ult",
+      "un its",
+      "Ġb ad",
+      "ĠSt ring",
+      "iz able",
+      "con dition",
+      "ĠElect ron",
+      "immun e",
+      "ĠM E",
+      "ha o",
+      "Î £",
+      "ĠM AT",
+      "Ġad opt",
+      "Ġel ic",
+      "Ġsh r",
+      "Ġproxim al",
+      "F D",
+      "ĠS S",
+      "Ġentire ly",
+      "es ium",
+      "ĠE EG",
+      "Ġpa ired",
+      "ĠT P",
+      "ĠD O",
+      "NA L",
+      "ides pread",
+      "Ġmov es",
+      "s ite",
+      "Ġra in",
+      "Ġl ap",
+      "ĠF u",
+      "ĠM eta",
+      "irc raft",
+      "Ġmagne tization",
+      "oper ation",
+      "Ġpro st",
+      "Ste p",
+      "Ġsubgroup s",
+      "ĠS outhern",
+      "Ġat he",
+      "lu or",
+      "ĠTaxon omic",
+      "ĠE instein",
+      "Ġr ace",
+      "ĠK en",
+      "Ġattem pts",
+      "Ġcos mic",
+      "ĠD op",
+      "Ġfix ation",
+      "Ġremov ing",
+      "B T",
+      "Ġlim b",
+      "Ġal ign",
+      "Ġd ried",
+      "d u",
+      "Ġput ative",
+      "uc cess",
+      "per t",
+      "Ġslow ly",
+      "al so",
+      "ol ip",
+      "Ġcl ient",
+      "Ġbas in",
+      "Ġsuscepti ble",
+      "Ġcom ing",
+      "ns on",
+      "ĠN GC",
+      "ass ert",
+      "Ġtens ile",
+      "Ġar ises",
+      "cut aneous",
+      "Ġc aro",
+      "B i",
+      "Ġdiscuss ions",
+      "Ġabnormal ities",
+      "Ġpoll ution",
+      "ĠA x",
+      "Ġload s",
+      "D o",
+      "ia o",
+      "Ġmed ication",
+      "Ġint act",
+      "ĠC X",
+      "Ġbre eding",
+      "ĠUn ion",
+      "ĠB at",
+      "ĠPar ticipants",
+      "ĠReg ulation",
+      "Ġcontrad iction",
+      "Ġint ensities",
+      "ence phal",
+      "ri le",
+      "ĠT LR",
+      "Ġred und",
+      "Ġpers ons",
+      "ĠAr c",
+      "sol id",
+      "l aw",
+      "Res ults",
+      "il ic",
+      "z one",
+      "ocyt osis",
+      "Ġtri angle",
+      "ST M",
+      "ĠV irus",
+      "Ġa id",
+      "so ft",
+      "Ġso on",
+      "exp ected",
+      "Ġan ch",
+      "ĠM u",
+      "ĠS r",
+      "ĠL O",
+      "Ġc ry",
+      "Ġup stream",
+      "ox ic",
+      "math it",
+      "ĠK le",
+      "Ġis otropic",
+      "Ġspati ally",
+      "ĠH ard",
+      "Ġext r",
+      "b as",
+      "e or",
+      "iv il",
+      "y an",
+      "Ġshif ted",
+      "Ġbi opsy",
+      "Ġfe el",
+      "gl ut",
+      "S ize",
+      "Ġ erg",
+      "ĠT er",
+      "Ġdeath s",
+      "bor ne",
+      "Ġrel ativistic",
+      "ĠV EGF",
+      "at ab",
+      "s pring",
+      "res tim",
+      "ĠS earch",
+      "yp henyl",
+      "ec al",
+      "ur c",
+      "Ġl amin",
+      "Ġser ial",
+      "l as",
+      "ĠPro duction",
+      "Ġsoci o",
+      "Ġmod ify",
+      "ĠServ ice",
+      "Ġb ary",
+      "Ġradi ative",
+      "big l",
+      "Ġparad igm",
+      "pati ent",
+      "Ġsp p",
+      "ph one",
+      "Ġ î",
+      "Ġro cks",
+      "ĠMart in",
+      "m n",
+      "Ġflu ids",
+      "ĠIN TR",
+      "od s",
+      "Ġdiv is",
+      "Cons ider",
+      "comp onent",
+      "Ġanomal ies",
+      "Ġk nee",
+      "ĠRelations hip",
+      "a ud",
+      "Ġover night",
+      "Ġra inf",
+      "Ġanne aling",
+      "Ġtre ating",
+      "Ġco arse",
+      "Mod el",
+      "Ġp ose",
+      "Ġocc as",
+      "ĠWilli am",
+      "o or",
+      "Ġadjust ment",
+      "ĠF unctions",
+      "im eter",
+      "Ġdet ectors",
+      "Ġinstit utional",
+      "Ġthrough put",
+      "iv idual",
+      "Ġenti ties",
+      "Ġprolong ed",
+      "Ġsh ip",
+      "Ġpres erved",
+      "ODU CTION",
+      "Ġlog istic",
+      "ĠPred iction",
+      "ti zed",
+      "ĠOr ig",
+      "ĠH em",
+      "onom ous",
+      "######## ########",
+      "ĠGen eration",
+      "b ottom",
+      "ĠK now",
+      "cl inical",
+      "Ġtra uma",
+      "Ġiter ative",
+      "Ġfac ility",
+      "ron t",
+      "ĠB us",
+      "Ġret inal",
+      "Ġcon duction",
+      "Ġcheck ed",
+      "Ġcall s",
+      "olog ists",
+      "C ON",
+      "ĠSc iences",
+      "Ġnon zero",
+      "Ġb rack",
+      "Ġmel ting",
+      "Ġas c",
+      "Ġmen tion",
+      "ĠB L",
+      "Ġver ification",
+      "uk ary",
+      "ĠSp atial",
+      "ĠG ram",
+      "Ġpl aces",
+      "Ġnec rosis",
+      "ĠChild ren",
+      "Ġdel ivered",
+      "Ġres ection",
+      "Ġdetermin istic",
+      "S ection",
+      "Ġmul tim",
+      "D F",
+      "Ġhypot heses",
+      "Ġra ise",
+      "Ġse ismic",
+      "Ġl am",
+      "ĠH CC",
+      "big r",
+      "Ġhe aling",
+      "is y",
+      "Ġoptim ize",
+      "obacter ium",
+      "ed y",
+      "Ġtr uth",
+      "Ġspace time",
+      "Ġchrom atin",
+      "Ġdom estic",
+      "Ġrec ru",
+      "ĠJ ose",
+      "ĠTherm al",
+      "Ġenvelop e",
+      "v able",
+      "Ġinc ons",
+      "Ġn od",
+      "Ð ¸",
+      "Ġcontrib uting",
+      "Ġguarant ee",
+      "ĠP hen",
+      "Ġra b",
+      "M an",
+      "Ġsurve illance",
+      "Ġth ings",
+      "Ġpre v",
+      "ĠNon linear",
+      "Ġg aps",
+      "ay a",
+      "ĠC ri",
+      "Ġcrystall ine",
+      "str ict",
+      "Ġcomput ations",
+      "Ġun able",
+      "h abil",
+      "um ina",
+      "Ġpromot ing",
+      "eg rad",
+      "Ġreg ister",
+      "Ġcross ing",
+      "ul ators",
+      "ĠL anguage",
+      "ĠA A",
+      "Ġin er",
+      "ĠL V",
+      "os an",
+      "Ġcoast al",
+      "Ġbi od",
+      "ĠM OD",
+      "Ġneighb our",
+      "Ġpredominant ly",
+      "ĠNew ton",
+      "ĠStr ateg",
+      "be ing",
+      "Ġ ì",
+      "Ġcap abilities",
+      "Ġun less",
+      "form al",
+      "Ġvess el",
+      "b matrix",
+      "ES S",
+      "Ġrainf all",
+      "Ã £",
+      "Ġpre par",
+      "ax ial",
+      "Ġd ental",
+      "ĠPro te",
+      "Ġwor se",
+      "d oped",
+      "hen tic",
+      "Ġvalid ate",
+      "Z n",
+      "Ġspec ification",
+      "s i",
+      "ĠAn g",
+      "Ġtub es",
+      "ul ic",
+      "ĠAn y",
+      "ĠM ap",
+      "Ġfabric ated",
+      "Ġfor ced",
+      "ĠWil son",
+      "ol ysis",
+      "ĠW ave",
+      "ĠC ast",
+      "Ġast hma",
+      "Ġper i",
+      "ĠC yt",
+      "ast y",
+      "Ġsk y",
+      "rup t",
+      "D ec",
+      "Ġmelan oma",
+      "P ER",
+      "Ġcontinu ity",
+      "B ox",
+      "s ystem",
+      "Ġn avig",
+      "Ġcirc ulating",
+      "Ġcolon y",
+      "less sim",
+      "ad ium",
+      "Ġtet ra",
+      "Ġacc ounts",
+      "Ġpresent ing",
+      "ĠL ik",
+      "Ġres is",
+      "Ġdamp ing",
+      "ĠG ly",
+      "ĠNeu ro",
+      "us er",
+      "Ġcap ital",
+      "ur ate",
+      "ĠM W",
+      "Ġcorrel ates",
+      "ĠG ib",
+      "Ġhapp ens",
+      "Ġg all",
+      "ĠWith in",
+      "Ġcomb ine",
+      "Ġsin us",
+      "ĠK in",
+      "**************** ****************",
+      "M ap",
+      "Ġmat uration",
+      "Ġblock ing",
+      "ĠClo ud",
+      "Ġcont acts",
+      "Ġs ac",
+      "AL L",
+      "ĠR ab",
+      "z z",
+      "ut ch",
+      "Ġcar riers",
+      "ĠSN R",
+      "er b",
+      "Ġprot ected",
+      "rack ing",
+      "radi ent",
+      "Ġattrac tive",
+      "Ġl ag",
+      "Ġop in",
+      "ĠG i",
+      "Ġdef ense",
+      "Ġtun ing",
+      "Ġelect roph",
+      "Ġgreat est",
+      "Ġreconstruc ted",
+      "ĠPop ulation",
+      "M AP",
+      "Ġw rote",
+      "AN D",
+      "ec onomic",
+      "ĠMich ael",
+      "ĠBl ock",
+      "Ġv o",
+      "op rop",
+      "Ġprof iling",
+      "oot st",
+      "ĠAs ian",
+      "Ġoscill ation",
+      "ĠâĨ IJ",
+      "U D",
+      "Ġsign ed",
+      "ĠE uler",
+      "ĠCompar ative",
+      "ĠW here",
+      "ĠJ ack",
+      "Ġpass ing",
+      "Ġvill age",
+      "Ġa u",
+      "ĠNor thern",
+      "ess age",
+      "m atic",
+      "Ġaff ili",
+      "ĠF ac",
+      "Ġoverl apping",
+      "she ll",
+      "Ġobst ac",
+      "Ġbec oming",
+      "enti ve",
+      "Ġeas ier",
+      "init ely",
+      "Ġcent ered",
+      "Ġac ademic",
+      "ann els",
+      "Ġir regular",
+      "Ġproj ections",
+      "Ġpro position",
+      "Ġdiscrim ination",
+      "Ġrem od",
+      "Ġsh oot",
+      "mon th",
+      "ess or",
+      "Ġdiff ers",
+      "ĠT V",
+      "ĠZ hou",
+      "Ġin her",
+      "Ġmach ines",
+      "Ġm ell",
+      "Ġconjug ate",
+      "Ġc oc",
+      "un a",
+      "an yl",
+      "Ġoff ic",
+      "Ġopportun ities",
+      "Ġve in",
+      "ĠCharacter istics",
+      "Ġpath ogenic",
+      "OY SA",
+      "ĠPark inson",
+      "ĠGal actic",
+      "FF FA",
+      "ys es",
+      "UH FFFA",
+      "UHFFFA OYSA",
+      "act in",
+      "Ġun us",
+      "hes ia",
+      "ace u",
+      "ad ow",
+      "os ide",
+      "Ġgly cos",
+      "Ġdil uted",
+      "ĠS ource",
+      "ol ated",
+      "arm aceu",
+      "ant om",
+      "Ġmus c",
+      "Ġaver aging",
+      "Ġvis it",
+      "Ġc atch",
+      "Ġsatisf action",
+      "Ġv on",
+      "val id",
+      "Ġyield ed",
+      "Ġpack ets",
+      "Ġreson ant",
+      "p ret",
+      "ĠG FP",
+      "Ġcut ting",
+      "Ġreplac ing",
+      "az e",
+      "P a",
+      "Ġto day",
+      "Ġdec ided",
+      "il ateral",
+      "im ate",
+      "l ings",
+      "ĠRob ust",
+      "ĠA st",
+      "od ynamics",
+      "Ġlack ing",
+      "izophren ia",
+      "Ġcont raction",
+      "um ann",
+      "ĠS ample",
+      "Ġdi amond",
+      "met hod",
+      "T OR",
+      "Ġcom ments",
+      "se y",
+      "Ġmanufact uring",
+      "ĠD a",
+      "N R",
+      "Ġoper ated",
+      "r ates",
+      "Ġextinc tion",
+      "u vant",
+      "ĠF inite",
+      "Ġlymph ocytes",
+      "b ro",
+      "om ology",
+      "Ġinstr uments",
+      "b ec",
+      "og le",
+      "Ġqu oti",
+      "Ġhyper bolic",
+      "Ġtr im",
+      "Ġp ap",
+      "atur ated",
+      "h aus",
+      "Ġs essions",
+      "Ġcamp aign",
+      "Ġvari eties",
+      "Ġpro jected",
+      "Ġr id",
+      "b one",
+      "Ġanc est",
+      "ĠE T",
+      "ma il",
+      "ĠTrans port",
+      "// /",
+      "ĠAn n",
+      "Ġcompos itions",
+      "ĠINTR ODUCTION",
+      "ĠâĪĴ âĨĴ",
+      "Ġwhen ever",
+      "ĠL ip",
+      "par ts",
+      "Ġis omorphic",
+      "Ġsulf ate",
+      "Ġh op",
+      "Ġg on",
+      "ĠOb ject",
+      "Ġpip eline",
+      "Ġm a",
+      "ĠG as",
+      "Ġtend ency",
+      "ob ject",
+      "Ġparamet ric",
+      "ĠRet urn",
+      "Ġd war",
+      "Ġpress ures",
+      "ĠBi os",
+      "Ġmulti plication",
+      "Ġdim in",
+      "Ġcol ors",
+      "ĠTr ue",
+      "M ax",
+      "ĠD epend",
+      "Ġpair wise",
+      "Ġl ake",
+      "Ġhierarch y",
+      "Ġthresh olds",
+      "ĠAdap tive",
+      "m aking",
+      "Ġcataly sts",
+      "ip al",
+      "Ġeg gs",
+      "Ġw ire",
+      "ophyl l",
+      "ict or",
+      "label ed",
+      "Ġmus cles",
+      "ĠUnder standing",
+      "Ġfib re",
+      "cont rolled",
+      "Ġinvari ance",
+      "Ġc ache",
+      "Ġbos on",
+      "Ġnear by",
+      "ĠW omen",
+      "ĠIn itial",
+      "Ġprob abilistic",
+      "Ġembry onic",
+      "ĠB etween",
+      "Ġcon jecture",
+      "i enti",
+      "t x",
+      "g ens",
+      "anc k",
+      "Ġg ir",
+      "ĠL ower",
+      "Ġhospit als",
+      "brid ge",
+      "Met hod",
+      "Ġthe ta",
+      "j a",
+      "Ġconcept ual",
+      "Ġcol le",
+      "ĠS af",
+      "d ic",
+      "Ġp et",
+      "Ġprim er",
+      "ĠO h",
+      "Ġun treated",
+      "long rightarrow",
+      "Ġl icense",
+      "Ġhel ps",
+      "Ġcle avage",
+      "Ġampl ified",
+      "Ð µ",
+      "Ġaccess ible",
+      "ĠSe lection",
+      "ĠLore ntz",
+      "P y",
+      "Ġpolar ized",
+      "ĠST AT",
+      "mit t",
+      "U p",
+      "Ġon going",
+      "Ġne ph",
+      "e fficient",
+      "ac tiv",
+      "ĠR R",
+      "Ġfunction ing",
+      "ot in",
+      "Ġl ists",
+      "Ġformal ism",
+      "Ġoscill ator",
+      "Ġgastro intestinal",
+      "ootst rap",
+      "ĠAs ia",
+      "ĠD ay",
+      "Ġcomp eting",
+      "ival ent",
+      "Ġbl adder",
+      "Ġh it",
+      "Ġapproxim ations",
+      "ĠE g",
+      "ĠCl ust",
+      "Ġrel ies",
+      "N E",
+      "cop ro",
+      "Ġb ank",
+      "Ġintegr ating",
+      "ĠH ear",
+      "Ġiniti ated",
+      "ac ryl",
+      "ĠB H",
+      "rac ted",
+      "y c",
+      "ĠR a",
+      "Ġremark able",
+      "Ġ Ë",
+      "ten ess",
+      "Ġemploy ing",
+      "ste ine",
+      "Ġï£ Ń",
+      "Ġtransf ected",
+      "Ġinj uries",
+      "ĠB rief",
+      "Ġw idespread",
+      "ĠA K",
+      "IV E",
+      "Ġh arm",
+      "Ġpo le",
+      "Ġanis otropic",
+      "at en",
+      "gen e",
+      "iv ariate",
+      "In ter",
+      "duct ors",
+      "Ġaccom pl",
+      "oglob in",
+      "c ong",
+      "Ġqu eries",
+      "escop e",
+      "ĠH op",
+      "Ġenti ty",
+      "Ġoff ered",
+      "St ate",
+      "ĠExperim ents",
+      "ann er",
+      "ĠW ood",
+      "ard ed",
+      "ag on",
+      "Ġfibrobl asts",
+      "Ġnan os",
+      "Ġper oxid",
+      "Ġev id",
+      "Ġï£ ¸",
+      "Ġre tained",
+      "os qu",
+      "Ġle aving",
+      "Ġf ashion",
+      "Ġn M",
+      "Ġmut ual",
+      "appro xim",
+      "Ġwalk ing",
+      "Ġim possible",
+      "Ġdemonstr ating",
+      "Ġde gener",
+      "ĠA V",
+      "Ġcont rary",
+      "us tion",
+      "ocl onal",
+      "A nal",
+      "Ġperform ances",
+      "Ġcomp rom",
+      "orm s",
+      "Ġbud get",
+      "ĠH aw",
+      "Ġarth ritis",
+      "ob j",
+      "no ise",
+      "Ti O",
+      "och rome",
+      "Ġge odes",
+      "be an",
+      "Ġselec tivity",
+      "ĠF ood",
+      "ugh ter",
+      "Ġpermut ation",
+      "ĠR P",
+      "os al",
+      "Ġadi p",
+      "armaceu tical",
+      "w hen",
+      "ĠT ext",
+      "we ek",
+      "Ġbond ing",
+      "ar b",
+      "oc or",
+      "Ġv oc",
+      "Ġup regulated",
+      "Ġneighb ors",
+      "Ġtra it",
+      "Ġthe ore",
+      "Ġc f",
+      "ĠB erg",
+      "ĠL A",
+      "Ġl as",
+      "un te",
+      "cept ual",
+      "AS E",
+      "Ġischem ic",
+      "Ġb ending",
+      "d ataset",
+      "Ġkeep ing",
+      "Ġar rows",
+      "Ġsubst ances",
+      "Ġn s",
+      "Ġext ending",
+      "ĠR u",
+      "Ġsupplement ation",
+      "cri tical",
+      "ĠT raining",
+      "bul let",
+      "Ġpar a",
+      "ta il",
+      "ĠRef erence",
+      "Ġï£ ¶",
+      "Ġdissip ation",
+      "Ġaux iliary",
+      "ĠCy cl",
+      "s tim",
+      "Ġdil ution",
+      "bu f",
+      "ĠM iss",
+      "Ġul timately",
+      "Ġpow ers",
+      "Ġst ands",
+      "ust ed",
+      "ĠO H",
+      "habil itation",
+      "an aly",
+      "ĠB ra",
+      "ad ding",
+      "Cor ollary",
+      "Ġd rought",
+      "qu ality",
+      "Ġstandard ized",
+      "ĠJ e",
+      "ĠAc id",
+      "Ġm ism",
+      "ĠCh rom",
+      "d raw",
+      "ĠBi om",
+      "ĠSt ability",
+      "Further more",
+      "l ast",
+      "v ic",
+      "Ġab st",
+      "Ġb is",
+      "Ġemerg ence",
+      "Ġg iant",
+      "D e",
+      "ĠS amples",
+      "AB A",
+      "n as",
+      "Ġon t",
+      "Ġev ap",
+      "le vant",
+      "m ain",
+      "ĠR od",
+      "Ġc ros",
+      "it ary",
+      "Ġdo ub",
+      "r ö",
+      "igene tic",
+      "Ġincom plete",
+      "dep th",
+      "ï ģ",
+      "Ġsatur ated",
+      "Ġaeros ol",
+      "As sum",
+      "Ġimmun os",
+      "Ġlip ids",
+      "itone al",
+      "Ġbe aring",
+      "ĠIm plications",
+      "Ġsustain ed",
+      "Ġcompe titive",
+      "Ġmo tivation",
+      "Ġdisturb ance",
+      "rystall ine",
+      "Ġtax a",
+      "Ġdem entia",
+      "Ġconcer ned",
+      "PI O",
+      "hom ogeneous",
+      "ĠE v",
+      "ĠGe orge",
+      "ĠAlgorithm s",
+      "ick el",
+      "us ively",
+      "Ġcor ner",
+      "ĠR est",
+      "Ġinf inity",
+      "ĠTrans form",
+      "hen g",
+      "Ġneuro de",
+      "ol im",
+      "Í ij",
+      "Ġsk ew",
+      "ĠB S",
+      "sc ore",
+      "Y PE",
+      "em an",
+      "el le",
+      "ĠCor relation",
+      "Ġcult ural",
+      "oph osph",
+      "Ġatten uation",
+      "Ġaggreg ate",
+      "Ġam big",
+      "Ġanomal ous",
+      "Ġt ors",
+      "Ġplan et",
+      "ĠN Ps",
+      "h r",
+      "ĠDiv ision",
+      "ĠEduc ation",
+      "lec tic",
+      "Ġb rought",
+      "ĠM orph",
+      "Ġplan es",
+      "Ġsug ar",
+      "Ġdend ritic",
+      "Ġcont our",
+      "Ġcylind er",
+      "p ost",
+      "Ġw ent",
+      "R L",
+      "Ġad mission",
+      "MS E",
+      "I X",
+      "Ġdis joint",
+      "Ġannot ation",
+      "Ġis otope",
+      "Ġμ ν",
+      "Ġelim inate",
+      "Ġre actor",
+      "on ents",
+      "Ġreason ing",
+      "Ġm orbidity",
+      "Ġcor rosion",
+      "other mal",
+      "arc tic",
+      "ĠM B",
+      "ĠZ hao",
+      "Ġhist ological",
+      "Ġsuperconduc ting",
+      "at tered",
+      "Ġhouse hold",
+      "ĠPro p",
+      "Ġass er",
+      "he red",
+      "Ġte ams",
+      "Ġvan ishes",
+      "P re",
+      "am ents",
+      "Ġam orphous",
+      "ĠDeterm ination",
+      "miss ions",
+      "Ġover head",
+      "det erm",
+      "Ġutil izing",
+      "f a",
+      "ip olar",
+      "Ġform ulated",
+      "Ġext rap",
+      "gr id",
+      "Ġhum idity",
+      "ub er",
+      "t umor",
+      "ro us",
+      "Ġdistor tion",
+      "d ynamics",
+      "ĠL oss",
+      "Ġscal ed",
+      "Ġischem ia",
+      "Ġax es",
+      "Ġqu antit",
+      "n it",
+      "ĠReg ion",
+      "ain ed",
+      "Ġf ill",
+      "Ġbran ching",
+      "ĠT iss",
+      "c ross",
+      "Ġplate let",
+      "iffiff iffiff",
+      "ro ps",
+      "lu x",
+      "j oin",
+      "ur acy",
+      "ic ide",
+      "ĠLou is",
+      "Ġï£ «",
+      "Ġstr ings",
+      "ys et",
+      "Ġfac ial",
+      "ĠM MP",
+      "RE S",
+      "Ġhydro lysis",
+      "ĠCan adian",
+      "Ġpro jective",
+      "Ġsc atter",
+      "ur on",
+      "ĠPsy ch",
+      "com plex",
+      "ĠN am",
+      "Ġconc urrent",
+      "ION S",
+      "Ġth ous",
+      "Ġch ance",
+      "Ġplac ement",
+      "Ġaware ness",
+      "Ġt rib",
+      "ĠT ex",
+      "ĠTh ird",
+      "Ġlabel ing",
+      "cer ol",
+      "Ġs aw",
+      "ĠB and",
+      "ĠP ear",
+      "Ġpregn ant",
+      "ĠD own",
+      "pl atin",
+      "S eq",
+      "x e",
+      "ethyl ene",
+      "ĠHig her",
+      "Ġre ality",
+      "ur is",
+      "ĠP AR",
+      "l b",
+      "d ose",
+      "sh if",
+      "ili ar",
+      "t otal",
+      "S W",
+      "Ġval ve",
+      "nd er",
+      "Ð ½",
+      "am ous",
+      "Ġend omet",
+      "LI SA",
+      "Ġfract ures",
+      "Ġfil t",
+      "ro le",
+      "Ġmicro structure",
+      "ĠSN P",
+      "T ER",
+      "ĠZn O",
+      "ov ing",
+      "al i",
+      "ĠG M",
+      "unc t",
+      "Ġext ensions",
+      "exp ression",
+      "Ġesc ape",
+      "ĠM as",
+      "ĠSp anish",
+      "Ġflo or",
+      "ĠCom mon",
+      "otop y",
+      "plement ation",
+      "Ġr hyth",
+      "Ġserv es",
+      "y to",
+      "Ġwavelength s",
+      "empt yset",
+      "ĠH ill",
+      "n or",
+      "ĠElect ro",
+      "Ġde hydrogen",
+      "Ġwh om",
+      "im etric",
+      "ĠR oman",
+      "ĠV e",
+      "âī ¥",
+      "ĠK u",
+      "ĠTrans fer",
+      "Ä ĩ",
+      "ĠT F",
+      "b rain",
+      "copro tein",
+      "ĠG reat",
+      "av en",
+      "ĠInd ividual",
+      "ur i",
+      "Ġfung i",
+      "Ġpar am",
+      "pt on",
+      "s ymmetry",
+      "Ġloc k",
+      "me as",
+      "Ġha em",
+      "Ġh ip",
+      "As s",
+      "eng er",
+      "Ġpot assium",
+      "an al",
+      "ibr ary",
+      "Ġschool s",
+      "n atal",
+      "Ġalle les",
+      "ĠH LA",
+      "ox ygen",
+      "ĠC up",
+      "Ġpure ly",
+      "D O",
+      "Ġch ip",
+      "ô ı",
+      "C ar",
+      "s il",
+      "Ġun likely",
+      "cor respond",
+      "ĠD P",
+      "Ġint ense",
+      "Ġfor cing",
+      "ĠJ ournal",
+      "Ġar row",
+      "ocy an",
+      "Ġcul tiv",
+      "Ġbl ind",
+      "Ġselect ing",
+      "oc arcinoma",
+      "ran ce",
+      "Ġhydroph obic",
+      "clos ed",
+      "Ġens ures",
+      "Ġprom oted",
+      "Ġdetect able",
+      "rane an",
+      "Ġsched ule",
+      "Ġpart ly",
+      "Ġgl and",
+      "Ġco uple",
+      "ĠEm erg",
+      "Ġtrac es",
+      "p oly",
+      "Ġprote ase",
+      "ys tic",
+      "Ġdoc uments",
+      "pos itions",
+      "Ġdri ver",
+      "ti um",
+      "ĠC YP",
+      "cl ose",
+      "ĠRec ep",
+      "Ġper mit",
+      "Ġblock ed",
+      "Ġinvestig ating",
+      "ĠT umor",
+      "ĠB ig",
+      "Ġwave gu",
+      "Ġsubst ance",
+      "Ġweak er",
+      "ĠM ont",
+      "ro vers",
+      "ĠMex ico",
+      "p res",
+      "ĠAc ute",
+      "Ġmicro gl",
+      "ĠE S",
+      "itor ing",
+      "ĠSer ies",
+      "l ights",
+      "Ġhypot hesized",
+      "Ġconstruc ts",
+      "Ġfilt ration",
+      "Bl ack",
+      "Ġun changed",
+      "Ġobserv able",
+      "Ġra y",
+      "b etween",
+      "Ġï£ ¯",
+      "ĠPos ition",
+      "Ġth i",
+      "ĠSystem atic",
+      "Cl ass",
+      "k m",
+      "ĠT ak",
+      "Ġrespond ents",
+      "Ġinn ate",
+      "Ġan t",
+      "Ġconn ecting",
+      "R el",
+      "Ġmanip ulation",
+      "ĠN eg",
+      "N Ps",
+      "ĠDi ab",
+      "ĠAc tive",
+      "ĠG all",
+      "ĠCoul omb",
+      "Ġspac ing",
+      "ĠF lor",
+      "Ġconduct ance",
+      "Ġtrac ks",
+      "ĠZh u",
+      "weight ed",
+      "ro cy",
+      "Ġfat her",
+      "id ium",
+      "struct ured",
+      "ĠT el",
+      "Ġst rom",
+      "ith ub",
+      "cer tain",
+      "B ut",
+      "ĠAc cess",
+      "Ġprevent ing",
+      "rest rial",
+      "ĠCons idering",
+      "tr ue",
+      "Ġhost s",
+      "Ġwor st",
+      "ĠP d",
+      "gre di",
+      "Ġgly col",
+      "Ġst ory",
+      "osqu ito",
+      "par atus",
+      "Ġme eting",
+      "Ġepis ode",
+      "n c",
+      "ĠS and",
+      "Ġu int",
+      "ynam ical",
+      "ur t",
+      "Ġeduc ational",
+      "Ġfocus es",
+      "g t",
+      "ĠH S",
+      "Ġdeterm inant",
+      "Ġlith ium",
+      "ĠDig ital",
+      "Ġguid ance",
+      "Ġprior ity",
+      "Ġpar ty",
+      "or ial",
+      "T wo",
+      "ĠProblem s",
+      "Ġsem an",
+      "ĠCN N",
+      "ĠE pid",
+      "Ġplay ing",
+      "Ġelim ination",
+      "ĠS at",
+      "Ġobj ectives",
+      "p lectic",
+      "Ġcircum st",
+      "ĠG S",
+      "oc ellular",
+      "ot rans",
+      "Ġfind s",
+      "Ġa romatic",
+      "iz ers",
+      "Ġfavor able",
+      "st andard",
+      "ich lor",
+      "mod els",
+      "otyp ing",
+      "Ġstabil ization",
+      "Ġhand ling",
+      "Ġco ated",
+      "e ven",
+      "Ġlet ter",
+      "Z E",
+      "Ġultr ason",
+      "Ġf riend",
+      "Ġsens iti",
+      "Ġatt achment",
+      "Ġap art",
+      "Ġgre y",
+      "Ġa ircraft",
+      "Ġr RNA",
+      "Ġenabl ed",
+      "Ġbu ff",
+      "Ġred ox",
+      "ass isted",
+      "Ġgener ality",
+      "PS S",
+      "Ġe lection",
+      "resp onse",
+      "Ġded icated",
+      "Ġdem ographic",
+      "Ġim posed",
+      "ĠK ir",
+      "ĠRad io",
+      "ĠE LISA",
+      "ga e",
+      "Ġres c",
+      "ĠR ic",
+      "raph ic",
+      "Ġra il",
+      "Ġj ournal",
+      "ol er",
+      "W S",
+      "Ġincorpor ation",
+      "w ind",
+      "Ġaud itory",
+      "A E",
+      "t ask",
+      "Ġp c",
+      "w all",
+      "Ġapp rec",
+      "aterial s",
+      "Ġpart ner",
+      "Ġcollec tive",
+      "Ġsc oring",
+      "ĠFran k",
+      "Ġperman ent",
+      "ĠI ran",
+      "um ination",
+      "M ed",
+      "ĠHy brid",
+      "Ġphen otypic",
+      "Ġdisrup tion",
+      "vi olet",
+      "osp heric",
+      "Ġregim es",
+      "ĠCol or",
+      "ĠPati ent",
+      "Ġf ever",
+      "Ġn n",
+      "Ġvari ational",
+      "ke ys",
+      "Ġdis till",
+      "Ġspect roscopic",
+      "ĠAr chitect",
+      "ac ing",
+      "Ġpro ves",
+      "Ġver teb",
+      "ĠComput er",
+      "Ġexp ensive",
+      "Ġfro zen",
+      "arcom a",
+      "N K",
+      "Ġhist one",
+      "Ġpolymer ization",
+      "Ġto b",
+      "Ġturn ed",
+      "eff ective",
+      "ĠAut hor",
+      "AP I",
+      "Ġdec ade",
+      "ĠRo bert",
+      "Ex ample",
+      "over set",
+      "AB LE",
+      "ĠBehavi or",
+      "f eed",
+      "ĠT ai",
+      "Ġï£ º",
+      "Ġe gg",
+      "Ġc ath",
+      "au x",
+      "ĠJoh nson",
+      "Ġtor que",
+      "Ġpur ification",
+      "Wh ite",
+      "c ious",
+      "ĠS ong",
+      "Ġprecip it",
+      "resh old",
+      "Ġm ilitary",
+      "Ġconv ection",
+      "ĠM iddle",
+      "ĠW he",
+      "Ġ ôı",
+      "al and",
+      "ar ation",
+      "fig ure",
+      "Ġded uce",
+      "chlor o",
+      "c ost",
+      "ithm etic",
+      "ĠItal ian",
+      "miss ible",
+      "ĠCommun ity",
+      "ĠN ature",
+      "Ġdi oxide",
+      "Ġbal anced",
+      "et t",
+      "ST AT",
+      "ild ing",
+      "Ġev olved",
+      "Ġmon ot",
+      "p ur",
+      "Ġpref erences",
+      "ding er",
+      "Ġarg ue",
+      "Ġmo tions",
+      "Ġinf ant",
+      "Ġaccel erated",
+      "Ġobser ver",
+      "Ġfabric ation",
+      "ĠMechan isms",
+      "Ġfunc tor",
+      "Ġhar ves",
+      "r ase",
+      "ĠSpec ial",
+      "Ġdepos its",
+      "Ġr ub",
+      "à ¸",
+      "ĠCP U",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ Ġ",
+      "atom ical",
+      "Ġfin it",
+      "Ġsec ure",
+      "Ġnutri tional",
+      "ren al",
+      "ĠF alse",
+      "Ġshe l",
+      "Ġrecru ited",
+      "am big",
+      "ĠSign aling",
+      "K O",
+      "organ isms",
+      "ĠL T",
+      "el en",
+      "ĠM arc",
+      "ab atic",
+      "Ġt ables",
+      "Ġconf ined",
+      "ĠA z",
+      "Ġproduc tivity",
+      "Ġad herence",
+      "Ġreplic ates",
+      "Ġvir t",
+      "f in",
+      "Ġagric ultural",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠ",
+      "ĠCh ampionship",
+      "and a",
+      "ĠCh urch",
+      "D uring",
+      "Ġinser ted",
+      "igh ter",
+      "Ġx en",
+      "Ġs ave",
+      "Ġtang ent",
+      "ven ous",
+      "Ġconver ge",
+      "Ġdistingu ished",
+      "Ġexpl os",
+      "Ġa ortic",
+      "Ġj ump",
+      "Ġneon atal",
+      "ud den",
+      "Ġslow er",
+      "Ġinfarc tion",
+      "Ġpre vents",
+      "u er",
+      "Ġ eros",
+      "R P",
+      "Ġcontin ues",
+      "OR T",
+      "Ġconsid ers",
+      "ĠN uclear",
+      "ly mp",
+      "Ġacc ounted",
+      "ores is",
+      "Ġneighbor ing",
+      "ĠRich ard",
+      "Ġen for",
+      "ĠCh ronic",
+      "Ġdisc over",
+      "ĠH ong",
+      "cell s",
+      "ĠCh all",
+      "Ġhom ogen",
+      "Ġathe ros",
+      "Ġisol ate",
+      "ĠPlas ma",
+      "ĠD L",
+      "par ametric",
+      "ĠUp per",
+      "H P",
+      "Ġintro duces",
+      "Ġmother s",
+      "Ġatt ract",
+      "Ġexcl usion",
+      "gra vity",
+      "ĠK r",
+      "Ġsp ike",
+      "ĠHe at",
+      "v ival",
+      "ĠRNA s",
+      "b ach",
+      "ator ial",
+      "ĠL td",
+      "on omy",
+      "in vasive",
+      "l ass",
+      "Ġwell s",
+      "Ġimag inary",
+      "Ġcarb ohyd",
+      "od a",
+      "Ġactiv ate",
+      "µ Ħ",
+      "Ġenzym atic",
+      "p es",
+      "Ġstat ements",
+      "Ġapproxim ated",
+      "ĠSal mon",
+      "ophage al",
+      "ĠH PV",
+      "con f",
+      "um at",
+      "Ġsulf ur",
+      "ĠRec all",
+      "Ġch ond",
+      "Ġvi able",
+      "por ation",
+      "Ġcare fully",
+      "tet ra",
+      "Ġlymph oma",
+      "st at",
+      "Ġconserv ative",
+      "atab ase",
+      "m and",
+      "Ġsc ored",
+      "Ġv as",
+      "Ġpri vacy",
+      "onym ous",
+      "Ġlogarithm ic",
+      "ĠE con",
+      "Ġachie ves",
+      "Ġabund ances",
+      "c am",
+      "Ġcy an",
+      "ĠE L",
+      "idel ity",
+      "j o",
+      "Ġan ticip",
+      "re ported",
+      "Ġarrang ement",
+      "iter ranean",
+      "ps is",
+      "ich i",
+      "Ġt a",
+      "um ping",
+      "ĠAc tivation",
+      "Ġmel t",
+      "Ġan no",
+      "og e",
+      "ĠD am",
+      "optim al",
+      "Ġneu rological",
+      "s a",
+      "ĠPar ameters",
+      "off set",
+      "Ġc ement",
+      "Ġinhib iting",
+      "Ġch ose",
+      "itz er",
+      "at tr",
+      "Ġmod er",
+      "ator ies",
+      "Ġte aching",
+      "ĠC ore",
+      "ph thal",
+      "ĠL uc",
+      "Ġin gredi",
+      "Ġclear ance",
+      "Ġachie ving",
+      "t age",
+      "Ġbur st",
+      "vi e",
+      "ĠSp ain",
+      "pt o",
+      "Ġtrans membrane",
+      "Ġsup plementary",
+      "Ġto ken",
+      "Ġobvious ly",
+      "ĠV ector",
+      "Ġdest r",
+      "H OD",
+      "Ġassum es",
+      "Ġpenet ration",
+      "Ġsub jective",
+      "h olds",
+      "ã o",
+      "Ġmo tiv",
+      "Ġprovid ers",
+      "v ascular",
+      "Ġdepart ment",
+      "ock et",
+      "F ile",
+      "Ġbre ath",
+      "ĠB est",
+      "gra ble",
+      "Ġl iqu",
+      "ĠAr g",
+      "ĠB ob",
+      "Ġfrag mentation",
+      "ec tic",
+      "Ġv ital",
+      "s ince",
+      "all oc",
+      "ox yphenyl",
+      "Ġradi otherapy",
+      "ĠSD S",
+      "Ġcyt ometry",
+      "n ucle",
+      "ĠI M",
+      "ĠTe V",
+      "raf ish",
+      "ĠKore a",
+      "Ġstreng then",
+      "Ġb are",
+      "Ġw oman",
+      "Ġrad ar",
+      "Ġplatform s",
+      "ozyg ous",
+      "ĠA h",
+      "Ġsub types",
+      "py rid",
+      "ĠTrans cription",
+      "Ġá º",
+      "ĠMeasure ments",
+      "Ġsurv iv",
+      "ĠN ear",
+      "Ġcasc ade",
+      "out he",
+      "B U",
+      "Ġexpon entially",
+      "Ġhaz ard",
+      "Ġsi RNA",
+      "Ġcell ulose",
+      "Fig s",
+      "Ġdifferenti ated",
+      "Ġim plicated",
+      "met ric",
+      "Ġcorrel ate",
+      "Ġm ission",
+      "Ġmant le",
+      "ĠP hyl",
+      "ĠH art",
+      "Ġg ases",
+      "Ġun ity",
+      "Ġexper t",
+      "Ġchar t",
+      "Ġd ict",
+      "Ġep ile",
+      "Ġoff spring",
+      "Ġemerg ed",
+      "Ġdem ands",
+      "Ġpres um",
+      "orb id",
+      "ĠMed icine",
+      "Ġstream s",
+      "tic ed",
+      "ĠN ic",
+      "Ġf illing",
+      "ĠC ro",
+      "Ġrestric tions",
+      "S ee",
+      "ĠM ill",
+      "Ġparent al",
+      "Ġdetermin ants",
+      "Ġecos ystem",
+      "ĠW all",
+      "ĠM emory",
+      "ple ts",
+      "Ġaggreg ates",
+      "per turb",
+      "Ġresid ents",
+      "AC K",
+      "v ectors",
+      "Ġman ually",
+      "Ġï ĺ",
+      "ĠFrame work",
+      "Ġv ag",
+      "eb rafish",
+      "l ib",
+      "ĠHear t",
+      "ĠAn imal",
+      "Ġwid er",
+      "G ene",
+      "ĠR os",
+      "Ġoper ate",
+      "Ġposs ibilities",
+      "ĠStr ong",
+      "Ġpy ro",
+      "resp ectively",
+      "Ġhybrid ization",
+      "ip edia",
+      "x in",
+      "Ġst om",
+      "f ish",
+      "ĠFor ce",
+      "Ġdim er",
+      "SU L",
+      "el se",
+      "Ġund e",
+      "g ar",
+      "con v",
+      "Ġarri val",
+      "Ġmon oclonal",
+      "I AL",
+      "Ġl y",
+      "Ġsymmet ries",
+      "Ġnur sing",
+      "rac h",
+      "Ġó µĦ",
+      "Ġbi ased",
+      "Ġc ues",
+      "Ġbiomark er",
+      "d ers",
+      "Ġc row",
+      "ern els",
+      "Ġbil ateral",
+      "Ġphys ically",
+      "Ġpat ches",
+      "Ġunc on",
+      "ĠB efore",
+      "def ault",
+      "est yle",
+      "t frac",
+      "ĠC ox",
+      "Ġinf iltration",
+      "Ġconver t",
+      "Ġstreng ths",
+      "ĠS ar",
+      "ig ible",
+      "oc omp",
+      "Ġsti r",
+      "Ġsch izophrenia",
+      "w as",
+      "Ġo w",
+      "et erm",
+      "ĠOr der",
+      "Ġf oss",
+      "Ġline age",
+      "Ġrab bit",
+      "Ġregular ization",
+      "ran ch",
+      "opl astic",
+      "T O",
+      "Ġmeas urable",
+      "Ġm ang",
+      "in itial",
+      "Ġbuild ings",
+      "Ġsystem atically",
+      "Ġferm ions",
+      "Ġlibr aries",
+      "Ġab lation",
+      "ide os",
+      "ĠW i",
+      "ph oton",
+      "ĠTest ing",
+      "ĠComput ing",
+      "ti er",
+      "in et",
+      "Ġprim itive",
+      "Ġcap illary",
+      "Ġsl ip",
+      "ver gence",
+      "rap eutic",
+      "ĠBl ue",
+      "ĠAc ad",
+      "ha i",
+      "ĠL ew",
+      "Ġtri angular",
+      "MS O",
+      "Ġsal inity",
+      "Ġnanoc om",
+      "o a",
+      "Ġhom omorphism",
+      "ĠM M",
+      "Ġres in",
+      "D B",
+      "um inescence",
+      "d ashed",
+      "ĠK h",
+      "qu ark",
+      "emb les",
+      "Ġidentif ies",
+      "Ġfol lic",
+      "Ġmet am",
+      "ĠH erm",
+      "Ġtob acco",
+      "Ġreal ization",
+      "hydro x",
+      "ĠB et",
+      "B ecause",
+      "Ġpiec es",
+      "Ġt alk",
+      "Ġopen ed",
+      "as ome",
+      "Ġsur ge",
+      "Ġfluct uation",
+      "g ithub",
+      "ĠB acter",
+      "Ġbind s",
+      "ĠRap id",
+      "au er",
+      "p H",
+      "emb ed",
+      "ĠD oc",
+      "uch i",
+      "ĠC andid",
+      "Ġrare ly",
+      "Ġm ountain",
+      "ĠF at",
+      "Ġs end",
+      "ov sk",
+      "ĠOrgan ization",
+      "ĠFran c",
+      "ĠO P",
+      "âĪ ¼",
+      "ok es",
+      "ec e",
+      "def icient",
+      "Ġlink age",
+      "od on",
+      "Ġf ly",
+      "Ġt idal",
+      "ĠEx amples",
+      "ĠR out",
+      "Ġaccom mod",
+      "Sup pose",
+      "ad ap",
+      "Ġdi e",
+      "ro ot",
+      "Ġh on",
+      "Ġminim izing",
+      "Ġrough ness",
+      "Ġgr ass",
+      "ent a",
+      "ĠL ang",
+      "ed u",
+      "ĠSim ple",
+      "en ic",
+      "Ġinduc ing",
+      "t f",
+      "Ġcon texts",
+      "ĠGeneral ized",
+      "ĠW nt",
+      "P b",
+      "at omic",
+      "d em",
+      "ĠPre paration",
+      "Ġinsu fficient",
+      "s am",
+      "ĠSpec ies",
+      "ĠS olar",
+      "Ġuns igned",
+      "ĠH ER",
+      "â Ĭ",
+      "Ġpar ity",
+      "Ġnit rate",
+      "ĠC er",
+      "p tic",
+      "id entif",
+      "ge al",
+      "Ġemo tion",
+      "ĠL P",
+      "Ġenh ancing",
+      "Ġmeaning ful",
+      "st ation",
+      "Ġrel ig",
+      "y o",
+      "Ġpers pectives",
+      "Ġsc ans",
+      "ugin osa",
+      "Ġsummar ize",
+      "rel ations",
+      "Ġdist ant",
+      "Ġfunction ality",
+      "Ġde eper",
+      "ol ate",
+      "ĠP or",
+      "graph s",
+      "ĠW a",
+      "ophil ic",
+      "CL US",
+      "ropath y",
+      "Ġc red",
+      "Ġun iversity",
+      "se g",
+      "ve e",
+      "O G",
+      "ĠM en",
+      "ĠCri tical",
+      "ã ģ",
+      "Ġex it",
+      "var theta",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠĠĠĠ",
+      "Ġun f",
+      "Ġpropos al",
+      "Ġty rosine",
+      "oti des",
+      "Ġproxim ity",
+      "Ġbox es",
+      "cat en",
+      "ĠEnvironment al",
+      "bound ed",
+      "down arrow",
+      "Ġfall s",
+      "Ġfer til",
+      "Ġcompr ised",
+      "Ġmell itus",
+      "Ġleak age",
+      "ut y",
+      "Ġchrom osomes",
+      "ĠStat istics",
+      "%% %%",
+      "Ġcomb inator",
+      "Ġk et",
+      "ad vant",
+      "T her",
+      "Ġtop ics",
+      "fl at",
+      "n ia",
+      "ĠSpect ral",
+      "Ġsynchron ization",
+      "var rho",
+      "Ġcolon ies",
+      "ĠF ive",
+      "ag ues",
+      "ĠF C",
+      "ID S",
+      "Ġa ward",
+      "Ġyield ing",
+      "Ġarchitect ures",
+      "ashing ton",
+      "chit z",
+      "per ty",
+      "Ġmod uli",
+      "m oment",
+      "sp eed",
+      "Ġmes enchymal",
+      "op tera",
+      "Ġinc omp",
+      "C ell",
+      "ĠM ice",
+      "Ġg ot",
+      "te ger",
+      "Ġt au",
+      "ĠAd S",
+      "Ġb ill",
+      "Ġdr inking",
+      "uls ive",
+      "Ġknock down",
+      "Ġarm s",
+      "ĠAut om",
+      "ĠIncre ased",
+      "H F",
+      "Ġglob ally",
+      "Ġdop ing",
+      "Ġat h",
+      "ĠC op",
+      "Ġsuccess ive",
+      "UL T",
+      "el ess",
+      "Ġble eding",
+      "Ġfood s",
+      "Ġimmun ohist",
+      "Ġdef inite",
+      "ĠJ ones",
+      "ĠT S",
+      "Ġjo ined",
+      "ĠTow ards",
+      "ĠC s",
+      "Ġun like",
+      "Ġval ence",
+      "d or",
+      "o S",
+      "Ġp ush",
+      "Ġoff ice",
+      "Ġalumin um",
+      "id yl",
+      "idi rectional",
+      "wr itten",
+      "Ġb ubble",
+      "H I",
+      "Ġmarked ly",
+      "ĠT ok",
+      "Ġvesic les",
+      "Ġquoti ent",
+      "Ġrepro duce",
+      "Ġelse where",
+      "ĠMy c",
+      "Ġinf rastructure",
+      "Ġgain ed",
+      "ab el",
+      "ĠS ex",
+      "ĠT ables",
+      "et in",
+      "Ġhom olog",
+      "Ġleg al",
+      "he a",
+      "Ġsoci ety",
+      "Ġman aged",
+      "id ase",
+      "ĠInhib ition",
+      "Ġparas ite",
+      "Ġvol unte",
+      "AT P",
+      "i os",
+      "Ġse psis",
+      "Ġrib osomal",
+      "Ġconf ound",
+      "ĠSta phyl",
+      "aryn geal",
+      "ï Ģ",
+      "com b",
+      "ĠOb jective",
+      "SUL TS",
+      "Ġthor ough",
+      "m t",
+      "Ġc hest",
+      "V ector",
+      "ele ment",
+      "Ġvir ulence",
+      "Ġhem isp",
+      "Ġso ught",
+      "ĠK o",
+      "Ġnutri tion",
+      "ul ing",
+      "ian a",
+      "Ġprot otype",
+      "ĠO nt",
+      "c ine",
+      "Ġdot ted",
+      "Ġob ese",
+      "ount ered",
+      "Ġphysic ians",
+      "Ġmin i",
+      "Ľ ľ",
+      "sp aces",
+      "Ġexcl usively",
+      "ĠCon volution",
+      "Ġc aspase",
+      "ĠL ink",
+      "di v",
+      "ĠRoy al",
+      "h ist",
+      "it ness",
+      "Ġes ter",
+      "Ġconduc ting",
+      "Ġparticip ated",
+      "Ġair way",
+      "Ġaer uginosa",
+      "E xt",
+      "arg ument",
+      "ock ing",
+      "Ġintegr ate",
+      "Ġcont rovers",
+      "ap es",
+      "train ing",
+      "ĠPre valence",
+      "tem p",
+      "b oth",
+      "Ġre activity",
+      "Ġrank ing",
+      "Ġtunn eling",
+      "OD E",
+      "ĠMed iterranean",
+      "Ġreson ances",
+      "M g",
+      "Ġl ib",
+      "ĠH eter",
+      "Ġnot hing",
+      "Ġindic ation",
+      "ĠH M",
+      "ocy tic",
+      "st rand",
+      "Ġcollabor ation",
+      "Ġelectro static",
+      "Ġindepend ence",
+      "h ab",
+      "Ġconf lic",
+      "Ġi od",
+      "in us",
+      "Ġdepend ency",
+      "ĠL am",
+      "Ġexam ining",
+      "Ġoccup ied",
+      "Ġque ue",
+      "ĠB ul",
+      "Ġregist ered",
+      "Ġindivid ually",
+      "R x",
+      "aus al",
+      "V E",
+      "Ġbright ness",
+      "resp ons",
+      "bal ance",
+      "Ġcytotox ic",
+      "f all",
+      "com mut",
+      "IC AL",
+      "ur an",
+      "ain ing",
+      "ra ulic",
+      "res ults",
+      "Ġepis odes",
+      "Y S",
+      "ĠG ar",
+      "Ġsur fact",
+      "dr ug",
+      "Ġc ities",
+      "ĠCh ange",
+      "os ition",
+      "Ġtrig gered",
+      "Ġcytoplas mic",
+      "erv es",
+      "Ġle x",
+      "Ġasymptotic ally",
+      "ph y",
+      "Ġfron tal",
+      "ĠD ensity",
+      "Ġsyn erg",
+      "cy cle",
+      "ĠImpro ved",
+      "Ã ¸",
+      "Ġmon o",
+      "Ġaccum ulated",
+      "orient ed",
+      "b our",
+      "Ġtun nel",
+      "com ing",
+      "Ġap paratus",
+      "Ġenc ountered",
+      "C re",
+      "Ġlet ters",
+      "et ch",
+      "Ġexcess ive",
+      "Ġbiofil m",
+      "Ġrear rang",
+      "Ġpolymorphism s",
+      "er obic",
+      "Ġconn ect",
+      "res olved",
+      "ĠN N",
+      "Ġret ro",
+      "ĠIn iti",
+      "ĠQuanti f",
+      "Ġp up",
+      "T ensor",
+      "Ġsent ences",
+      "l ay",
+      "ran ts",
+      "pl oid",
+      "ĠAnd erson",
+      "Ġdes irable",
+      "st ud",
+      "i ability",
+      "Ġd rying",
+      "ec ess",
+      "Ġd ens",
+      "Ġdescri pt",
+      "ĠË Ĩ",
+      "Ġcl ones",
+      "Ġju ven",
+      "b p",
+      "Ġk il",
+      "H L",
+      "Ġhem orrh",
+      "ĠK i",
+      "H ow",
+      "Ġen erge",
+      "Ġsub section",
+      "ĠS ac",
+      "di al",
+      "Ġcardi omy",
+      "Ġto uch",
+      "d m",
+      "Ġsc ienti",
+      "oid es",
+      "ĠÃ Ĥ",
+      "ysacchar ide",
+      "Ġs clerosis",
+      "ĠZe aland",
+      "in ine",
+      "Ġunus ual",
+      "ĠB A",
+      "ips chitz",
+      "g ap",
+      "ĠDiff erences",
+      "Ġdual ity",
+      "ed ical",
+      "Ġl ign",
+      "Ġfail s",
+      "Ġ lect",
+      "Ġrel ate",
+      "Ġincor rect",
+      "Ġspec ify",
+      "Ġcylind rical",
+      "ĠP F",
+      "ĠL ind",
+      "Ġdet erior",
+      "Ġher b",
+      "d z",
+      "Ġw eld",
+      "Ġnom inal",
+      "cop y",
+      "Ġacet yl",
+      "ht ml",
+      "Ġrecogn ize",
+      "** *",
+      "iti an",
+      "W A",
+      "ĠM N",
+      "ĠF ind",
+      "Ġaut hentic",
+      "per ture",
+      "Ġcytotox icity",
+      "of l",
+      "ĠG et",
+      "Ġcoh omology",
+      "Ġremain der",
+      "Ġexpand ing",
+      "Ġhe av",
+      "oster one",
+      "R ight",
+      "Ġcop ol",
+      "Ġs hed",
+      "Ġcompl iance",
+      "Ġacid ic",
+      "or ic",
+      "Ġam yloid",
+      "Ġevap oration",
+      "d l",
+      "Ġdel ays",
+      "P o",
+      "ĠCH ECK",
+      "tain s",
+      "Ġrevers ed",
+      "ĠMP a",
+      "Ġprocess or",
+      "Ġh all",
+      "ĠL ast",
+      "Ġplas m",
+      "ĠAss ociated",
+      "ĠBas ic",
+      "in os",
+      "Ġsympt om",
+      "ã Ģ",
+      "Ġanth rop",
+      "Ġjud g",
+      "Ġe ti",
+      "k le",
+      "Ġwr ong",
+      "ro om",
+      "Ġdevelop ments",
+      "ĠMax imum",
+      "Ġcoating s",
+      "Ġheur istic",
+      "ron tal",
+      "S ome",
+      "Ġutil ize",
+      "ĠâĪ ħ",
+      "c oll",
+      "ĠRel ated",
+      "Ġde generation",
+      "tem plate",
+      "Ġmod ulated",
+      "Ġparamet ri",
+      "Ġsal iv",
+      "ĠPseud omonas",
+      "Ġanti gens",
+      "Ġhar mon",
+      "ĠL HC",
+      "do i",
+      "ens itive",
+      "ĠNo tice",
+      "ĠM oh",
+      "til age",
+      "AC S",
+      "Ġdiscrep ancy",
+      "Ġsp ik",
+      "Ġre strict",
+      "it rile",
+      "le g",
+      "ĠB ase",
+      "Ġconvolution al",
+      "ĠRes istance",
+      "Ġappear ing",
+      "ĠIm ages",
+      "ĠM ann",
+      "Ġre act",
+      "Ġmacroph age",
+      "Ġwave let",
+      "och rom",
+      "Ġfair ly",
+      "Ġpreced ing",
+      "Ġsp ir",
+      "n etwork",
+      "ĠN ak",
+      "IF T",
+      "Ġag o",
+      "Ġenc ryp",
+      "al d",
+      "ens in",
+      "Ġs ulph",
+      "ĠPol ymer",
+      "ĠAr t",
+      "Ġsub units",
+      "sh ot",
+      "Ġbeg ins",
+      "Ġex er",
+      "pro pto",
+      "Ġn urses",
+      "Ġsuff ices",
+      "Ġgra ded",
+      "ĠR ock",
+      "Ġuniqu ely",
+      "it ol",
+      "Ġsp iral",
+      "Ġthan ks",
+      "char acter",
+      "ĠDist ributed",
+      "ĠC art",
+      "F orm",
+      "Ġform ulations",
+      "iction ary",
+      "Ġspread ing",
+      "Ġsingular ity",
+      "Ġpig s",
+      "it u",
+      "ot rophic",
+      "Ñ Ģ",
+      "Ġsemicon ductor",
+      "Ġd rag",
+      "ne xt",
+      "ma xim",
+      "un n",
+      "Ġarg ued",
+      "pl astic",
+      "Ġdehydrogen ase",
+      "Ġreinfor cement",
+      "ent ral",
+      "ĠD S",
+      "Ġcompan ies",
+      "Ġquanti zation",
+      "ĠD ri",
+      "Ġsimpl er",
+      "Ġradi i",
+      "ĠEth ics",
+      "ĠElect ronic",
+      "t aken",
+      "Ġpharmac ological",
+      "ps on",
+      "Ġpair ing",
+      "Ġn est",
+      "ĠR S",
+      "Ġl ic",
+      "oc on",
+      "Ġobserv ing",
+      "ĠF M",
+      "I ES",
+      "Ġsub mitted",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠ",
+      "Ġno isy",
+      "Ġvan ishing",
+      "ĠTechn ologies",
+      "il st",
+      "ag ic",
+      "Ġembed dings",
+      "Ġpl ans",
+      "re ak",
+      "oc t",
+      "Ġepit helium",
+      "Ġrevers ible",
+      "Ġrequ ests",
+      "V i",
+      "ĠPro g",
+      "meth oxy",
+      "ur ia",
+      "Ġsl ice",
+      "Ġmetast ases",
+      "ĠM ary",
+      "Ġprior i",
+      "Ġexplain s",
+      "ĠS igma",
+      "ĠArm y",
+      "Ġpre y",
+      "K L",
+      "ĠP ass",
+      "Ġrepro duction",
+      "Ġfer mentation",
+      "ul o",
+      "Ġproof s",
+      "ĠAccording ly",
+      "ti st",
+      "ĠïĢ ©",
+      "Ġme at",
+      "Ġpl anned",
+      "Ġangi ogenesis",
+      "W R",
+      "ĠA ust",
+      "Similar ly",
+      "ĠW ashington",
+      "Ġref inement",
+      "Ġembry o",
+      "Ġdiss ociation",
+      "á n",
+      "plas ia",
+      "ĠG ro",
+      "Ġsimilar ities",
+      "Ġsolub ility",
+      "Ġimm obil",
+      "ĠSc ot",
+      "ĠSub sequently",
+      "di vid",
+      "Ġclos est",
+      "ĠW at",
+      "Ġâ Į",
+      "ĠA GN",
+      "Ġpres cribed",
+      "Ġm osquito",
+      "Ġf irm",
+      "Ġde generate",
+      "Ġeth yl",
+      "Ġhar vest",
+      "ĠSpec ific",
+      "Ġcomp artment",
+      "p ublic",
+      "ĠBi ological",
+      "Ġpiec e",
+      "Ġat titudes",
+      "Ġsp ray",
+      "ĠS ix",
+      "Ġprofession als",
+      "Ġsl ot",
+      "Ġretrie ved",
+      "ve ment",
+      "Ġexec uted",
+      "se ed",
+      "Ġout flow",
+      "d istance",
+      "ĠT erm",
+      "ad y",
+      "ĠProv ince",
+      "ĠCent re",
+      "ĠD FT",
+      "Ġs udden",
+      "Ġse iz",
+      "r at",
+      "rom o",
+      "ot echn",
+      "Ġhigh lights",
+      "Ġelectroly te",
+      "ĠAdv anced",
+      "all ow",
+      "p x",
+      "os ed",
+      "sub array",
+      "rac ks",
+      "P RO",
+      "ogen y",
+      "Ġpool ed",
+      "Ġd type",
+      "Ġop posed",
+      "ĠG rand",
+      "Ġdesign ing",
+      "b el",
+      "it ability",
+      "Ġminim ization",
+      "Ġdram atically",
+      "Ġso y",
+      "ag ents",
+      "ĠMet al",
+      "ĠM V",
+      "rib ute",
+      "D D",
+      "it an",
+      "Ġspeed s",
+      "Ġmar ried",
+      "Ġevalu ations",
+      "ĠKing dom",
+      "Ġcl ay",
+      "ĠTiss ue",
+      "left arrow",
+      "Ġcompens ation",
+      "ch ild",
+      "p ool",
+      "up arrow",
+      "ĠDom ain",
+      "spec ies",
+      "Ġmeth ane",
+      "ĠE GFR",
+      "Ġpar ser",
+      "h ave",
+      "Ġneg lected",
+      "f unc",
+      "aps ed",
+      "Ġs ays",
+      "ad ata",
+      "bin om",
+      "C ase",
+      "Ġre porter",
+      "S n",
+      "Ġmaxim ize",
+      "Ġbif urc",
+      "ĠCN S",
+      "ĠO lymp",
+      "Ġdecl are",
+      "Ġenc oder",
+      "Ġab elian",
+      "Ġsingular ities",
+      "Ġe ch",
+      "Î ¨",
+      "Ġpro to",
+      "Ġph ag",
+      "Ġpoly g",
+      "Ġb ott",
+      "Ġadi pose",
+      "u ing",
+      "j k",
+      "uch y",
+      "ĠStud ent",
+      "Ġnan ow",
+      "Ġth ym",
+      "E d",
+      "E nd",
+      "Ġtransform s",
+      "ĠP CA",
+      "k ern",
+      "reg n",
+      "Ġcom ment",
+      "ĠL L",
+      "ell es",
+      "Ġeng agement",
+      "ĠP eter",
+      "IS PR",
+      "ĠCh annel",
+      "in y",
+      "Ġbund les",
+      "A ld",
+      "Ġpublic ations",
+      "T G",
+      "st ra",
+      "Ġf ear",
+      "Ġre tic",
+      "ple ments",
+      "Ġcor pus",
+      "ĠCl uster",
+      "ĠR ate",
+      "Ġsimpl est",
+      "ac ic",
+      "rb rack",
+      "Ġb low",
+      "Ġcomp ress",
+      "ĠD ark",
+      "Ġpsy chiatric",
+      "ĠCon versely",
+      "Ġo wing",
+      "Ġabs or",
+      "ĠH P",
+      "Ġcr ude",
+      "equ al",
+      "ĠAr ray",
+      "ĠRel ative",
+      "Ġcomb ustion",
+      "R ed",
+      "k t",
+      "Ġm A",
+      "Ġt ex",
+      "por ters",
+      "Ġdiffere d",
+      "Ġaud io",
+      "z on",
+      "od i",
+      "Ġmac roscopic",
+      "ac in",
+      "Ġz eros",
+      "Ġfore ign",
+      "Ġd uct",
+      "b ow",
+      "w orth",
+      "ĠRo ad",
+      "re y",
+      "ace ous",
+      "Ġbl ast",
+      "Ġgran ul",
+      "Ġw ing",
+      "Ġannot ated",
+      "ĠF ull",
+      "Ġinflu encing",
+      "v y",
+      "iaz ol",
+      "Ġp itch",
+      "Ġre habilitation",
+      "ĠPri or",
+      "com it",
+      "math tt",
+      "di a",
+      "ĠI on",
+      "Ġab use",
+      "Ġharves ted",
+      "Ġepid emic",
+      "Ġfil ament",
+      "Ġnucle ation",
+      "ĠKnow ledge",
+      "rin os",
+      "Ġb ent",
+      "Ġsqu ared",
+      "Ġhippocamp al",
+      "ĠT G",
+      "AN T",
+      "mod ified",
+      "ari o",
+      "ĠF ace",
+      "Ġgrow s",
+      "Ġfa ults",
+      "v irus",
+      "Ġpartition ing",
+      "air s",
+      "Ġhe aring",
+      "Ġcon gen",
+      "Ġ rip",
+      "ĠColl abor",
+      "Ġinterview s",
+      "Ġh uge",
+      "Ġbreak down",
+      "Ġmonth ly",
+      "ĠCON CLUS",
+      "E ach",
+      "D iff",
+      "Ġrel ay",
+      "ĠM use",
+      "oscop y",
+      "Ġre new",
+      "g b",
+      "Ġb rid",
+      "Ġoutl ined",
+      "or ig",
+      "e at",
+      "ĠWith out",
+      "Ġsp or",
+      "ĠT N",
+      "ĠJ o",
+      "ĠA U",
+      "N ot",
+      "Ġret in",
+      "ĠAn gel",
+      "Ġtri ed",
+      "ey ond",
+      "j e",
+      "ĠRuss ian",
+      "ĠUn fortunately",
+      "ĠMean while",
+      "ograph s",
+      "Ġacc ounting",
+      "ĠA β",
+      "m b",
+      "Ġdop amine",
+      "ĠBrief ly",
+      "ĠF requency",
+      "Mat rix",
+      "ĠJose ph",
+      "Ġexper ts",
+      "Ġdro ps",
+      "ĠRE SULTS",
+      "Ġrect angular",
+      "ath ione",
+      "cent er",
+      "ĠLe ft",
+      "in form",
+      "k ins",
+      "Ġm il",
+      "ĠM ah",
+      "Ġmed ial",
+      "ĠComp any",
+      "Ġpass age",
+      "Ġlead er",
+      "Ġscreen ed",
+      "er i",
+      "pos ites",
+      "r arily",
+      "Ġph one",
+      "ie tic",
+      "Ġexpect ations",
+      "ĠPar ticle",
+      "ĠM ountain",
+      "Ġinter leukin",
+      "Ġfif th",
+      "Ġv ast",
+      "Ġlog ical",
+      "Ġt err",
+      "Ġcre ates",
+      "Ġfinit ely",
+      "Ġsw im",
+      "Ġsupernat ant",
+      "opath ological",
+      "ĠUl tra",
+      "ĠT y",
+      "Ġgra nd",
+      "Ġconstit ute",
+      "olog ist",
+      "ĠBro ad",
+      "aw are",
+      "Ġvic inity",
+      "ag ulation",
+      "uns igned",
+      "ĠS ize",
+      "ĠC ognitive",
+      "Ġsusp ected",
+      "Ġu pl",
+      "Ġauto immune",
+      "ĠS K",
+      "C B",
+      "Ġsl ices",
+      "ĠCh i",
+      "Ġobserv ables",
+      "Ġhippocamp us",
+      "so ver",
+      "Ġfund ing",
+      "Ġcon formation",
+      "ĠQ uestion",
+      "ĠS qu",
+      "ĠW ill",
+      "Ġsc attered",
+      "ir ty",
+      "Ġpl aus",
+      "cor relation",
+      "Ġventi lation",
+      "ĠGen es",
+      "Ġben ign",
+      "Ġheter o",
+      "St atus",
+      "ang led",
+      "Ġb ootstrap",
+      "Ġvacc ines",
+      "Ġmicro organisms",
+      "Ġvis its",
+      "Ġtheorem s",
+      "d rop",
+      "ĠT A",
+      "Ġcycl ing",
+      "Ġspectrom eter",
+      "Ġground water",
+      "Ġnanot ubes",
+      "Ġjo ints",
+      "ĠE ll",
+      "Ġcons ult",
+      "Ġwindow s",
+      "Ġdis ability",
+      "Ġgain s",
+      "Ġdis charg",
+      "Ġhe ated",
+      "Ġa fore",
+      "ary ing",
+      "inc re",
+      "Ġagg ressive",
+      "Ġhe mod",
+      "ari um",
+      "ĠIn st",
+      "v m",
+      "Ġdro plet",
+      "p tive",
+      "vious ly",
+      "Ġst arch",
+      "Ġd f",
+      "os yl",
+      "Ġdon ors",
+      "ĠUn like",
+      "Ġalkal ine",
+      "Ġintellig ence",
+      "a a",
+      "Ġaccept ance",
+      "Ġsl iding",
+      "aps es",
+      "ĠD iss",
+      "ist an",
+      "a uc",
+      "Ġb ins",
+      "Ġmod ulate",
+      "Ġman age",
+      "out s",
+      "Ġs enes",
+      "Ġdifferenti ate",
+      "Ġcoun ted",
+      "AS K",
+      "Ġantib acterial",
+      "Ġent ered",
+      "Ġdis advant",
+      "ĠSalmon ella",
+      "Ġis otopic",
+      "Ġanno unced",
+      "ĠBo ard",
+      "Ġrest oration",
+      "Ġalle vi",
+      "Ġprogram me",
+      "Ġalb umin",
+      "Ġcatal og",
+      "est ine",
+      "Ġdifferent ly",
+      "Ġm olar",
+      "rö dinger",
+      "ĠE vent",
+      "minist ration",
+      "ĠSer um",
+      "RO M",
+      "k w",
+      "b ot",
+      "Ġj ets",
+      "ĠDo uble",
+      "el er",
+      "Ġinf usion",
+      "Ġconsum ed",
+      "ĠI ron",
+      "ĠProcess es",
+      "Ġad mits",
+      "Ġj uris",
+      "ĠPer iod",
+      "Ġremod eling",
+      "alle y",
+      "Ġenabl ing",
+      "Ġback ward",
+      "ĠM id",
+      "bre vi",
+      "Ġclass ify",
+      "Ġcr ypt",
+      "Ġhel ix",
+      "ĠJ iang",
+      "Ġh oney",
+      "ges tion",
+      "x c",
+      "Ġcoinc ides",
+      "ĠD N",
+      "Ġap optotic",
+      "Ġinst all",
+      "ĠR ever",
+      "ĠDop pler",
+      "ic ago",
+      "er als",
+      "Ġp ie",
+      "ĠM ars",
+      "ĠStaphyl ococcus",
+      "Ġnot ing",
+      "Ġgener a",
+      "ĠI o",
+      "Ġh ope",
+      "Ġpres erve",
+      "MA X",
+      "yn chron",
+      "Ġr up",
+      "Ġcompr ising",
+      "ĠW ay",
+      "Ġvi olation",
+      "Q R",
+      "Ġreflect ing",
+      "Ġregular ity",
+      "ĠSi O",
+      "ĠJ un",
+      "Ġcommun ications",
+      "r ating",
+      "Ġfam iliar",
+      "Ġinstant aneous",
+      "Ġcor tic",
+      "Ġapparent ly",
+      "X X",
+      "Ġexcit ations",
+      "ĠA ward",
+      "N um",
+      "ĠU N",
+      "Ġqu bit",
+      "ĠAc tion",
+      "ĠF ried",
+      "Ġelim inated",
+      "Ġasp ir",
+      "h ler",
+      "Ġdec oding",
+      "un ov",
+      "Ġanalog ue",
+      "ul monary",
+      "Ġge ographic",
+      "Ġs ort",
+      "ĠCR C",
+      "Ald rich",
+      "Ġk Da",
+      "ĠN D",
+      "Ġset tle",
+      "ex ists",
+      "Ġstat istic",
+      "ĠB ow",
+      "ĠC G",
+      "Ġorgan izations",
+      "ĠM obile",
+      "Ġinv ent",
+      "Ġincorpor ate",
+      "ĠF ib",
+      "ord an",
+      "Ġcolle agues",
+      "ĠSt ation",
+      "Ġs en",
+      "Ġenc aps",
+      "ĠR H",
+      "rel im",
+      "Ġcarbon ate",
+      "ĠN ether",
+      "m em",
+      "EE E",
+      "Ġafore mentioned",
+      "Ġp ent",
+      "ĠSign al",
+      "Ġsusp ended",
+      "Col or",
+      "Ġsp ins",
+      "Ġpropor tions",
+      "ult y",
+      "Ġen rolled",
+      "ĠT EM",
+      "ĠRecep tor",
+      "Ġpre valent",
+      "l arge",
+      "v s",
+      "Ġtrunc ated",
+      "Ġâĭ ħ",
+      "l m",
+      "an il",
+      "Ġann ih",
+      "ĠGalax y",
+      "er as",
+      "Ġep igenetic",
+      "Ġto oth",
+      "Ġcondens ation",
+      "ĠT ensor",
+      "Ġin organic",
+      "ym ers",
+      "u f",
+      "an ese",
+      "are t",
+      "Ġar ithmetic",
+      "â Ĩ",
+      "Ġt rying",
+      "Ġimplement ing",
+      "x d",
+      "Ġill umination",
+      "el a",
+      "Ġdefic its",
+      "Ġsp ots",
+      "Ġdoes n",
+      "Ġrest ing",
+      "tra ined",
+      "Ġeros ion",
+      "Ġgran ular",
+      "Ġsc ar",
+      "Ġpol len",
+      "l ie",
+      "Ġcon vers",
+      "Ġdisturb ances",
+      "ĠG od",
+      "Ġen larg",
+      "ĠL ate",
+      "yl ase",
+      "Ġfac ts",
+      "ent y",
+      "ĠStre et",
+      "sequ ence",
+      "Ġven ous",
+      "ĠC heck",
+      "ag g",
+      "Ġabsorb ed",
+      "Ġcom mit",
+      "set s",
+      "Ġdest roy",
+      "Ġbow el",
+      "Ġfin ished",
+      "ĠF eed",
+      "Ġdop ed",
+      "ĠAl b",
+      "ĠMit ochond",
+      "Ġtheore tically",
+      "R I",
+      "Ġmet eor",
+      "ĠM G",
+      "Ġn ation",
+      "ĠBas in",
+      "n ik",
+      "Ġdep ths",
+      "ĠMechan ism",
+      "Ġmotif s",
+      "ĠH ay",
+      "Ġmo tivated",
+      "ĠC opy",
+      "ĠE astern",
+      "Ġpers istence",
+      "Ġra ys",
+      "F B",
+      "and em",
+      "l ayers",
+      "ey er",
+      "ĠStre pt",
+      "Ġregist ration",
+      "ĠAnt arctic",
+      "C V",
+      "ĠP ap",
+      "ĠSp e",
+      "Ġsplic ing",
+      "per formance",
+      "Ġseman tics",
+      "Ġloc om",
+      "oblast oma",
+      "Ġm oney",
+      "Ġtrans parent",
+      "Ġh r",
+      "ĠInter actions",
+      "Ġs ap",
+      "Ġbi ases",
+      "Ġte eth",
+      "yn olds",
+      "omet hyl",
+      "Ġm V",
+      "Ġsole ly",
+      "Ġor ange",
+      "bl ast",
+      "ATION S",
+      "c all",
+      "opo ietic",
+      "s ided",
+      "ĠF ox",
+      "ĠV ideo",
+      "Ġinsp ection",
+      "Ġb uck",
+      "hes ize",
+      "p resent",
+      "ĠAnti b",
+      "Ġh am",
+      "al am",
+      "ĠP G",
+      "ĠA E",
+      "Ġj oin",
+      "Ġmon ocytes",
+      "es tiv",
+      "Ġrandom ised",
+      "Ġtransl ocation",
+      "Ġincorpor ating",
+      "Ġprolif er",
+      "Ġod ds",
+      "IT H",
+      "Ġr an",
+      "Ġinstruc tion",
+      "Ġresol ve",
+      "Ġf t",
+      "ĠHe ad",
+      "Ġre agent",
+      "Ġad mitted",
+      "h uman",
+      "pos ure",
+      "ĠCh a",
+      "ĠF r",
+      "Ġbroad cast",
+      "Ġnutri ents",
+      "n ob",
+      "Ġnot able",
+      "ĠI GF",
+      "ĠC learly",
+      "Ġquark s",
+      "Ġe ukary",
+      "ĠAd d",
+      "it osan",
+      "Ġinter active",
+      "it ting",
+      "ĠComput ational",
+      "Ġdiss olution",
+      "ist ribution",
+      "pro duct",
+      "ĠA BC",
+      "olim its",
+      "bi ased",
+      "Ġtrap ped",
+      "P K",
+      "ĠH PLC",
+      "roph ot",
+      "z es",
+      "our se",
+      "ĠH ot",
+      "Ġrec ipro",
+      "n olimits",
+      "ell o",
+      "Ġassess ments",
+      "EN TS",
+      "Ġalter ation",
+      "t w",
+      "Ġcha otic",
+      "ĠL oc",
+      "Ġcat tle",
+      "R ay",
+      "Ġform ally",
+      "le ave",
+      "text style",
+      "Ġvent ral",
+      "ĠWilli ams",
+      "ĠPe ople",
+      "ix ing",
+      "ĠThe rapy",
+      "Ġi ii",
+      "ĠD T",
+      "Ġb ic",
+      "Ġsp heres",
+      "Ġvis c",
+      "Ġestablish ment",
+      "Ġdescrip tions",
+      "ĠA verage",
+      "Ġto ur",
+      "ĠInf ection",
+      "ĠL icense",
+      "Ġprep are",
+      "H s",
+      "f inite",
+      "ri um",
+      "ore g",
+      "ent ry",
+      "Ġdis ks",
+      "Ġelong ation",
+      "c pu",
+      "ĠChar les",
+      "FIG URE",
+      "st on",
+      "ĠObserv ations",
+      "A dd",
+      "ĠT ask",
+      "at omy",
+      "ig ration",
+      "ĠD atabase",
+      "ĠTex as",
+      "Ġph yt",
+      "ll er",
+      "con jug",
+      "onal d",
+      "Ġheav ily",
+      "Ġsp le",
+      "Ġass ist",
+      "ĠC p",
+      "Ġhapp en",
+      "u v",
+      "ĠUn iverse",
+      "ĠG PS",
+      "W E",
+      "X i",
+      "Ġadminist r",
+      "str ong",
+      "Ġmagn itudes",
+      "Ġsimpl ify",
+      "Ġele gans",
+      "es h",
+      "ĠB ody",
+      "ĠNether lands",
+      "Ã ¯",
+      "omet ers",
+      "B o",
+      "F M",
+      "ĠN iger",
+      "pl us",
+      "inst ance",
+      "Ġdist ress",
+      "Or gan",
+      "C as",
+      "Ġsym plectic",
+      "Ġbreak s",
+      "Ñ Ĥ",
+      "Ġferm ion",
+      "em poral",
+      "Ġs omatic",
+      "e vent",
+      "ne ut",
+      "lamm ation",
+      "ĠL ibrary",
+      "Ġmulti plic",
+      "ĠIn str",
+      "et hel",
+      "ur ys",
+      "Ġhelp ed",
+      "Ġcol lege",
+      "Ġcar tilage",
+      "Ġr pm",
+      "w estern",
+      "res is",
+      "Ġlob e",
+      "Q L",
+      "In put",
+      "Ġemph asis",
+      "b est",
+      "Ġtot ally",
+      "ĠMET HOD",
+      "ĠF a",
+      "ĠRed uction",
+      "ici ous",
+      "Ġimplant ation",
+      "pot ential",
+      "prob lem",
+      "Ġobtain s",
+      "ur ons",
+      "Ġconstruct ing",
+      "ĠMus ic",
+      "Ġcan cell",
+      "Ġnew s",
+      "ĠCh apter",
+      "Ġlab elled",
+      "Ġz ebrafish",
+      "ĠSol id",
+      "Ġglut amate",
+      "ĉĉ ĉĉĉ",
+      "Ġch apter",
+      "ĠPres ident",
+      "M in",
+      "Ġat rial",
+      "c p",
+      "f i",
+      "f inal",
+      "Ġto k",
+      "Ġeffect or",
+      "Ġsp ine",
+      "Ġidenti ties",
+      "isc o",
+      "ol is",
+      "ĠC le",
+      "Ġinvari ants",
+      "P ath",
+      "ĠG on",
+      "fact ory",
+      "Ġex ogenous",
+      "ĠMAP K",
+      "Ġansw ers",
+      "Ġget ting",
+      "R s",
+      "I H",
+      "ĠDef ine",
+      "ĠConvolution al",
+      "Ġgeomet rical",
+      "ĠIn put",
+      "Ġ à",
+      "Ġatten uated",
+      "Ġradical s",
+      "ĠAcad emy",
+      "ã ĥ",
+      "ich let",
+      "Ġtor us",
+      "ĠThe oretical",
+      "ĠT D",
+      "Ġan tiv",
+      "on ge",
+      "Ġintra venous",
+      "Ġhyp oth",
+      "Ġwaste water",
+      "ĠF lo",
+      "Ġpor osity",
+      "Ġp all",
+      "ac i",
+      "Ġrecord ings",
+      "Ġe ating",
+      "ĠD W",
+      "un ting",
+      "ĠD im",
+      "Ġemit ted",
+      "ĠJ oint",
+      "of ib",
+      "Ġearthqu ake",
+      "Ġm unic",
+      "Ġreduc tions",
+      "Ġcon junction",
+      "ĠL ocation",
+      "Ġestabl ishing",
+      "ĠMat hematical",
+      "ĠS olution",
+      "bu ffer",
+      "ar in",
+      "ile y",
+      "ĠCom mission",
+      "ĠG ABA",
+      "ĠMuse um",
+      "Ġver b",
+      "lec ules",
+      "inf ection",
+      "Ġins ect",
+      "is er",
+      "Ġprov ision",
+      "Ġagre ed",
+      "Ġaff ord",
+      "the ory",
+      "know ledge",
+      "Pro tein",
+      "Ġk ernels",
+      "Ġd erm",
+      "Ġw ish",
+      "Ġv ox",
+      "S cale",
+      "h u",
+      "Ġcounter parts",
+      "ĠR oss",
+      "Ġun p",
+      "ĠOn line",
+      "Ġtrans porter",
+      "G raph",
+      "Ġ uter",
+      "Ġmin ute",
+      "Ġautom orphism",
+      "il tr",
+      "ĠResp ons",
+      "ĠS ym",
+      "Ġfactor ization",
+      "s em",
+      "Ġmedi ates",
+      "Ġun expected",
+      "Ġorgan ism",
+      "Ġattem pted",
+      "ar an",
+      "ven ue",
+      "ethel ess",
+      "Ġno ticed",
+      "ĠInvestig ation",
+      "Ġcare g",
+      "Ġgroup ed",
+      "or bit",
+      "Ġshort est",
+      "Ġbroad er",
+      "ĠM IM",
+      "ris es",
+      "vel oper",
+      "ĠH i",
+      "Ġk Hz",
+      "Ġbe ads",
+      "Ġph yto",
+      "ĠDo es",
+      "Ġmamm als",
+      "Ġref ined",
+      "vol ume",
+      "S er",
+      "Ġresis tivity",
+      "Ġter restrial",
+      "Ġa xi",
+      "if luor",
+      "ĠÂ £",
+      "Ġv ice",
+      "ĠK el",
+      "V M",
+      "ĠT own",
+      "ad m",
+      "pl ates",
+      "Ġhol omorphic",
+      "ĠR ib",
+      "ĠS B",
+      "ĠTem poral",
+      "s rc",
+      "Ġupd ates",
+      "Ġsee k",
+      "en dix",
+      "ore tic",
+      "war z",
+      "Ġro utes",
+      "Ġstand ing",
+      "ĠÃ ģ",
+      "Ġclass ic",
+      "Ġp ale",
+      "lec tions",
+      "Ġclass ifiers",
+      "Ġpath ophys",
+      "Ġm ounted",
+      "Ġdesign ated",
+      "Ġv ideos",
+      "Ġinc oming",
+      "Ġguarant ees",
+      "Ġparas ites",
+      "ĠB acillus",
+      "f our",
+      "ĠâĪ ¨",
+      "Ġcommut ative",
+      "stack rel",
+      "ĠBan ach",
+      "Ġde aling",
+      "em porary",
+      "M ulti",
+      "ot omy",
+      "re ting",
+      "Ġn ond",
+      "ĠCon ference",
+      "tz mann",
+      "Ġphosphor us",
+      "Ġchemical s",
+      "Ġdis par",
+      "deg ree",
+      "Ġarbit rarily",
+      "rocy te",
+      "Ġpar abolic",
+      "Ġdimension less",
+      "Ġo sm",
+      "Ġphon on",
+      "ti ary",
+      "ĠS ect",
+      "ophys ical",
+      "ĠM apping",
+      "b is",
+      "ĠCommun ication",
+      "Ġmim ic",
+      "Ġregul ators",
+      "Ġneutroph ils",
+      "f n",
+      "ĠImport antly",
+      "Ġm ere",
+      "Ġconfir ms",
+      "ag ram",
+      "Ġatt end",
+      "ung al",
+      "ĠGroup s",
+      "Ġz o",
+      "Ġm outh",
+      "Ġste ep",
+      "Ġprevent ed",
+      "Ġdep ressive",
+      "ac ies",
+      "ĠL S",
+      "Ġnit ric",
+      "Ġvisual ized",
+      "Ġtranscript ome",
+      "Ġga it",
+      "erc ury",
+      "Ġsh ot",
+      "ĠV en",
+      "Ġex chang",
+      "Ġint ention",
+      "ĠT ang",
+      "Ġfav our",
+      "ve olar",
+      "Ġper mission",
+      "Ġhabit ats",
+      "Ġma ize",
+      "inc t",
+      "Ġtele vision",
+      "ryst als",
+      "ĠRad i",
+      "Ġflav on",
+      "Ġcan n",
+      "i ota",
+      "ĠO T",
+      "p ic",
+      "R ad",
+      "ti tial",
+      "ĠOr th",
+      "st ellar",
+      "ĠK ine",
+      "Ġnavig ation",
+      "f ast",
+      "ĠCR ISPR",
+      "Ġkinem atic",
+      "Ġsearch ing",
+      "Ġmic rom",
+      "Ġinst alled",
+      "ĠTai wan",
+      "il a",
+      "r f",
+      "ri age",
+      "pl inary",
+      "Ġe cho",
+      "ra v",
+      "ĠL es",
+      "cre ate",
+      "Ġubiqu it",
+      "Ġprecurs ors",
+      "K E",
+      "Ġdiv ide",
+      "Ġln c",
+      "ĠCon struction",
+      "an ic",
+      "es tim",
+      "is ters",
+      "Ġfe et",
+      "ari ant",
+      "ĠSch w",
+      "Ġexcl ude",
+      "Ġvol can",
+      "ĠOver view",
+      "Ġy r",
+      "ol k",
+      "ĠÂ ©",
+      "ĠF E",
+      "Ġsper mat",
+      "Ġcapac itance",
+      "ĠSch rödinger",
+      "ĠG E",
+      "Ġcalibr ated",
+      "S EM",
+      "Ġlat tices",
+      "pl ier",
+      "Ar g",
+      "ĠN T",
+      "ĠEnh anced",
+      "Ġb rom",
+      "Ġmulti p",
+      "Ġcer tified",
+      "Ġis lands",
+      "Ġcy st",
+      "Ġal titude",
+      "ed ef",
+      "Ġconst rain",
+      "Ġsatisf actory",
+      "Ġspecial ized",
+      "Ġj unctions",
+      "Ġcoron avirus",
+      "ud ge",
+      "ex c",
+      "Ġal t",
+      "ĠB acterial",
+      "Ġse asons",
+      "ĠL M",
+      "Ġhist ogram",
+      "Ġsol vents",
+      "a verage",
+      "Ġcard inal",
+      "ch rom",
+      "py thon",
+      "d ered",
+      "en ia",
+      "ĠG H",
+      "ĠEs s",
+      "__ __",
+      "ĠP ak",
+      "s ized",
+      "ĠH g",
+      "Ġel if",
+      "ĠSchem atic",
+      "Ġcytoplas m",
+      "ĠFor t",
+      "an ia",
+      "Ġcare ful",
+      "ĠD ual",
+      "Ġtransl ated",
+      "Ġn asal",
+      "In v",
+      "Ġda ughter",
+      "Ġemphas ize",
+      "mod ules",
+      "Ġl ives",
+      "Ġhom otopy",
+      "Ġb ot",
+      "Ġdis ordered",
+      "mat o",
+      "S econd",
+      "Ġclaim ed",
+      "add le",
+      "Ġinterf acial",
+      "Ġvisc ous",
+      "Ġdest ination",
+      "ĠPl anck",
+      "Ġabsorb ance",
+      "Ġvol atile",
+      "Ġst orm",
+      "Ġcar boxyl",
+      "ĠB ank",
+      "ĠP ack",
+      "Ġscaff old",
+      "te br",
+      "ip ot",
+      "Ġtum ours",
+      "ĠG ol",
+      "Ġelectroph oresis",
+      "Ġreal ize",
+      "Ġconstitu ents",
+      "S ol",
+      "ĠE very",
+      "Ġmedi ate",
+      "Ġcoinc ide",
+      "Ġexplo it",
+      "Ġmon oton",
+      "me asure",
+      "Ġsup plied",
+      "rac ellular",
+      "Ġfer ro",
+      "Ġpur s",
+      "eren tially",
+      "tr ast",
+      "ĠR B",
+      "Ġdiss em",
+      "as y",
+      "Ġrel ating",
+      "n ull",
+      "u ates",
+      "const ant",
+      "ĠContinu ous",
+      "Ġgeomet ries",
+      "r ust",
+      "ĠS TR",
+      "cl uster",
+      "Ġprogen itor",
+      "ĠC SF",
+      "ĠY am",
+      "ĠRe ynolds",
+      "ĠM Y",
+      "ĠK O",
+      "ĠW alk",
+      "ari able",
+      "ind er",
+      "ĠR ight",
+      "ĠAl gebra",
+      "ĠW ik",
+      "Ġin activation",
+      "t mp",
+      "ac cess",
+      "ĠL ater",
+      "Ġmicrobi ome",
+      "Ġgeodes ic",
+      "Ġre jection",
+      "us es",
+      "Ġhard ness",
+      "Ġhydro dynamic",
+      "Ġvan ish",
+      "Ġpoll ut",
+      "amy cin",
+      "ĠÏ Ń",
+      "ip itation",
+      "Ġaug mented",
+      "ĠT T",
+      "av al",
+      "Ġenc ode",
+      "Ġtox in",
+      "et o",
+      "igh bor",
+      "add r",
+      "Ġdam aged",
+      "o i",
+      "Ġtrans duction",
+      "Ġinter acts",
+      "ÃŃ a",
+      "ĠC all",
+      "ri ends",
+      "ĠMon itoring",
+      "ĠVari ation",
+      "Ġôı ¼",
+      "Ġd ich",
+      "Ġsp ars",
+      "al ign",
+      "Ġan atomical",
+      "Ġcentrifug ed",
+      "ur ally",
+      "ĠZ r",
+      "ĠCar l",
+      "Rec all",
+      "Ġopin ion",
+      "Ġ era",
+      "Ġdrain age",
+      "Ġmicro array",
+      "st atus",
+      "um ental",
+      "Ġcomp rises",
+      "press ure",
+      "Ġprac tition",
+      "m ac",
+      "Ġcon gr",
+      "urn al",
+      "ĠA PI",
+      "ĠL R",
+      "Ġtransf ection",
+      "Ġsl opes",
+      "ĠC ode",
+      "Ġph il",
+      "b ool",
+      "W s",
+      "Ġâ Ļ",
+      "Ġassoci ate",
+      "otox icity",
+      "ra de",
+      "ĠM iller",
+      "ĠÏ ª",
+      "Ġshor ten",
+      "Ġaddition ally",
+      "ĠEff ective",
+      "Ġsuper vised",
+      "Ġel abor",
+      "ĠC ellular",
+      "Ġt ell",
+      "ĠR C",
+      "s ave",
+      "im id",
+      "Ġrating s",
+      "ĠT aking",
+      "Ġappro val",
+      "Ġpenal ty",
+      "K K",
+      "con text",
+      "ak s",
+      "pec ific",
+      "Ġtem por",
+      "Ġup regulation",
+      "V AL",
+      "Ġenc odes",
+      "in in",
+      "Ġnot es",
+      "ĠFore st",
+      "Ġcombinator ial",
+      "ympt otic",
+      "Ġsqu amous",
+      "ĠAs h",
+      "our n",
+      "Ġmyel oid",
+      "el ines",
+      "B io",
+      "Ġbre ed",
+      "ĠR ub",
+      "uz z",
+      "Ġsingle t",
+      "en na",
+      "Ġcri tically",
+      "d ig",
+      "dis ci",
+      "Ġdrop ped",
+      "Ġlip oprotein",
+      "ĠE t",
+      "Ġno v",
+      "op hen",
+      "Ġanc ient",
+      "B ase",
+      "Ġsmooth ing",
+      "iti ves",
+      "p ine",
+      "Ġsol ver",
+      "per m",
+      "ĠH ome",
+      "Ġaz im",
+      "l Vert",
+      "Ġtransport ation",
+      "Ġde x",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠ",
+      "opath ic",
+      "ex perim",
+      "âĢ¢ âĢ¢",
+      "perf usion",
+      "Ġdo i",
+      "ĠL act",
+      "Ġhepat ocellular",
+      "Ġmism atch",
+      "Ġaden ocarcinoma",
+      "ĠP ain",
+      "Ġsp r",
+      "Ġconf inement",
+      "Ġexceed s",
+      "Ġhas h",
+      "ĠCompar ing",
+      "ĠS ensor",
+      "Ġf iring",
+      "k es",
+      "v ir",
+      "ine a",
+      "aff ected",
+      "Ġmod elled",
+      "Ġe ther",
+      "Ġsu ffer",
+      "âĢ² âĢ²",
+      "о Ð",
+      "ĠB ir",
+      "Ä ģ",
+      "Ġsec reted",
+      "Ġcat heter",
+      "Ġy outh",
+      "ex pl",
+      "ĠD ar",
+      "ĠW HO",
+      "Ġfound ation",
+      "Ġhyd raulic",
+      "ĠCa rol",
+      "SS ION",
+      "Ġá ¹",
+      "f eld",
+      "av or",
+      "Ġpass es",
+      "vis iae",
+      "Ġapplic ability",
+      "Ġn ested",
+      "F l",
+      "ĠC atal",
+      "Ġmicro environment",
+      "lab els",
+      "Ġcrystall ization",
+      "In fo",
+      "Ġposition ing",
+      "Ġtriang les",
+      "Ġtr yp",
+      "ĠTrans ition",
+      "Ġset t",
+      "Ġneuro t",
+      "M on",
+      "Ġdro plets",
+      "ĠA RT",
+      "Ġcor ne",
+      "Ġmulti plicity",
+      "Ġec centric",
+      "Ġ iv",
+      "ĠM atter",
+      "lear ning",
+      "elect ro",
+      "ĠWe yl",
+      "Ġdec ide",
+      "ĠW r",
+      "ĠH ierarch",
+      "Ġap ical",
+      "Ġfail ures",
+      "Ġdiges tion",
+      "MI C",
+      "Ġge ographical",
+      "ĠEle ment",
+      "ĠTh ough",
+      "Ġch ron",
+      "lim ited",
+      "ĠDI SC",
+      "ĠArchitect ure",
+      "Ġvibr ational",
+      "ĠVari ous",
+      "Ġdynam ically",
+      "ak ed",
+      "Ġconven ience",
+      "ĠIs ra",
+      "ĠMD A",
+      "i tic",
+      "A u",
+      "Ġass istance",
+      "ven tional",
+      "mid t",
+      "os por",
+      "F ollowing",
+      "Ġinf erior",
+      "Ġn ickel",
+      "ra ine",
+      "p aren",
+      "Ġtit anium",
+      "F ield",
+      "Ġh oc",
+      "ĠCa uchy",
+      "ĠMc C",
+      "ĠSc reen",
+      "Ġneg lect",
+      "class es",
+      "ĠI F",
+      "Ġstrat ified",
+      "ens es",
+      "ĠPl ate",
+      "oz oic",
+      "Ġinstit utions",
+      "ĠTh ose",
+      "Ġgener ations",
+      "trans form",
+      "Ġpar titions",
+      "Rx iv",
+      "ent h",
+      "Ġs tic",
+      "ol ith",
+      "ĠF em",
+      "Ġag ar",
+      "be am",
+      "Ġprot ons",
+      "L U",
+      "Ġwork load",
+      "Ġmin erals",
+      "Ġm t",
+      "ll a",
+      "ĠPh armac",
+      "Ġconver ter",
+      "ĠMechan ical",
+      "Ġflav or",
+      "Ġphosph atase",
+      "Ġsum s",
+      "PC s",
+      "Ġiso forms",
+      "ig roup",
+      "py r",
+      "fe atures",
+      "Ġper c",
+      "Ġcomple teness",
+      "Ġfore sts",
+      "Ġdiv iding",
+      "ĠL ipschitz",
+      "period ic",
+      "Ġrec ycl",
+      "ĠN ag",
+      "Ġtw in",
+      "epti des",
+      "Ġco hor",
+      "Ġsearc hes",
+      "e ated",
+      "H g",
+      "ĠP U",
+      "ĠT ree",
+      "all ic",
+      "P F",
+      "Ġapp endix",
+      "ĠC ov",
+      "Ġcheck ing",
+      "Ġback bone",
+      "Ther mo",
+      "Ġactiv ating",
+      "ĠV ictor",
+      "Ġcri tic",
+      "ĠL em",
+      "group s",
+      "RE G",
+      "ĠO cc",
+      "SC C",
+      "ĠX RD",
+      "ĠVal ues",
+      "Ġsub type",
+      "Ġstret ching",
+      "OR M",
+      "s ome",
+      "Ġfl ip",
+      "Ġphen olic",
+      "Ġk illed",
+      "Ġsequ enced",
+      "usc ular",
+      "ab in",
+      "Ġquad r",
+      "Ġtransl ational",
+      "Ġsol ids",
+      "di rect",
+      "Ġprom otion",
+      "Ġcohor ts",
+      "ĠCl imate",
+      "ĠO ld",
+      "ĠS ir",
+      "g ue",
+      "str ate",
+      "ĠP oss",
+      "Ġrece ives",
+      "ĠVal idation",
+      "uc tive",
+      "Ġcere visiae",
+      "G u",
+      "is is",
+      "ce il",
+      "ĠPear son",
+      "ĠP relim",
+      "ĠG ran",
+      "CS F",
+      "Ġster ile",
+      "oflu orescence",
+      "b ad",
+      "Ġcol ored",
+      "comp ass",
+      "equ ation",
+      "j an",
+      "Ġcondition ing",
+      "Ġvo ice",
+      "Ġmen ing",
+      "Ġgrant ed",
+      "Ġrenormal ization",
+      "ĠLim it",
+      "th i",
+      "Ġa perture",
+      "Ġdos age",
+      "di rected",
+      "ĠBre ast",
+      "oc ular",
+      "b earing",
+      "s al",
+      "asc ul",
+      "uper vised",
+      "Ġmonol ayer",
+      "Ġmembers hip",
+      "ĠW ireless",
+      "sh ow",
+      "ĠMed ia",
+      "ĠV L",
+      "ess el",
+      "Ġdec oder",
+      "ĠM F",
+      "ĠCom position",
+      "ĠCl ark",
+      "P oint",
+      "ĠN ano",
+      "ĠD eg",
+      "N L",
+      "ĠB ox",
+      "Ġexpl oring",
+      "m olecular",
+      "O ther",
+      "ĠDiab etes",
+      "he ight",
+      "Ġkin ases",
+      "Ġadjust ing",
+      "Ġsp orts",
+      "off s",
+      "ĠI EEE",
+      "Ġt il",
+      "ĠInt ra",
+      "Ġplan ets",
+      "ĠEpid em",
+      "Ġto mato",
+      "Ġscaff olds",
+      "ĠMet abol",
+      "ĠGe ometry",
+      "im etry",
+      "ĠT en",
+      "th read",
+      "o hex",
+      "Ġpro poses",
+      "pr im",
+      "ĠPar ty",
+      "Ġquar ter",
+      "ĠSh i",
+      "Ġab err",
+      "ĠIn tr",
+      "Ġdirect or",
+      "aff e",
+      "ĠS us",
+      "ens ors",
+      "E le",
+      "Ġpol es",
+      "Ad ditional",
+      "Ġby pass",
+      "caten in",
+      "Ġunder taken",
+      "im ation",
+      "op or",
+      "Ġpres erving",
+      "Ġmultiple x",
+      "ĠRepresent ative",
+      "s is",
+      "ĠA G",
+      "ach y",
+      "Ġfr uits",
+      "Ġreconstr uct",
+      "ens en",
+      "Ġstrong est",
+      "Ġsc av",
+      "ĠChen g",
+      "ĠCor on",
+      "ĠObs ervation",
+      "ĠA ch",
+      "ĠGe org",
+      "ĠSV M",
+      "ĠC hern",
+      "Ġrevers al",
+      "v ia",
+      "im p",
+      "Ġdeploy ment",
+      "ĠH ad",
+      "Ġcircumst ances",
+      "ob i",
+      "Ġcur ved",
+      "Ind uced",
+      "ĠPos itive",
+      "im b",
+      "ĠPar is",
+      "ĠS tein",
+      "ic z",
+      "ĠC ath",
+      "Ġdraw ing",
+      "t ory",
+      "Ġcontin ental",
+      "Ġquantit atively",
+      "ac erb",
+      "Ġnorm s",
+      "ĠB E",
+      "S everal",
+      "do or",
+      "Ġplate au",
+      "G al",
+      "Ġc ivil",
+      "ĠF ix",
+      "L AB",
+      "oc cal",
+      "Ġsor ted",
+      "ĠâĢ Ŀ",
+      "Ġed iting",
+      "ĠChris tian",
+      "Ġclar ify",
+      "Ġwavegu ide",
+      "b ell",
+      "Ġded uced",
+      "od ec",
+      "utri tion",
+      "Ġcomp ressive",
+      "ĠE U",
+      "ĠReg ression",
+      "Ġrank ed",
+      "Ġestim ators",
+      "Ġab ilities",
+      "Ġbelief s",
+      "th ree",
+      "Ġâĩ Ĵ",
+      "rolog y",
+      "Ġaut onomous",
+      "ĠS z",
+      "sc hem",
+      "ĠAL T",
+      "ĠPattern s",
+      "Ġex on",
+      "Ġlif estyle",
+      "f ill",
+      "ĠC AR",
+      "ĠDom ains",
+      "Ġpa id",
+      "Ġt ab",
+      "ĠCo hen",
+      "air y",
+      "Ġshe ep",
+      "Ġse aw",
+      "ĠK ong",
+      "g as",
+      "Ġres erved",
+      "Ġres il",
+      "Ġob l",
+      "car box",
+      "ĠGover nment",
+      "up per",
+      "ract ing",
+      "Ġg angl",
+      "ĠR V",
+      "Ġbron ch",
+      "Method s",
+      "ĠL iver",
+      "Ġgu ess",
+      "cha romy",
+      "IC E",
+      "Ġcongen ital",
+      "Ġk a",
+      "Ġsp anning",
+      "ĠRec omm",
+      "e a",
+      "Ġcon vention",
+      "Ġshe ets",
+      "Ġtherm o",
+      "Ġqualit atively",
+      "Ġox ides",
+      "Ġcongr u",
+      "ĠJ er",
+      "Ġpres ervation",
+      "ĠB T",
+      "ĠD MSO",
+      "Ġcom plication",
+      "Ġsurviv ors",
+      "Ġreduc t",
+      "Ġdes cent",
+      "Ġsuc rose",
+      "ĠCour t",
+      "Ġmetabol ite",
+      "ĠM ath",
+      "ĠSec urity",
+      "ĠNot ably",
+      "ĠSt em",
+      "Ġdwar f",
+      "b c",
+      "Ġre vis",
+      "ĠK l",
+      "ĠG h",
+      "Ġman ager",
+      "Ġinvest ment",
+      "Ġmo tility",
+      "E m",
+      "ĠM r",
+      "as ic",
+      "ĠB os",
+      "Ġinsp ired",
+      "plac ian",
+      "Ġe ase",
+      "Ġtors ion",
+      "ĠDir ichlet",
+      "Ġsple en",
+      "ag ation",
+      "on ate",
+      "ĠT rial",
+      "Ġturn over",
+      "Ġselec tively",
+      "ĠÍ Ĵ",
+      "ian o",
+      "Ġnon trivial",
+      "i asis",
+      "Ñ ģ",
+      "ĠGu o",
+      "Ġaddress es",
+      "Ġuniqu eness",
+      "Ġwith draw",
+      "ri z",
+      "Ġcomputation ally",
+      "Ġperson ality",
+      "A X",
+      "went y",
+      "Ġgover n",
+      "ber ts",
+      "Ġrob ots",
+      "Ġread y",
+      "Ġdi ets",
+      "l it",
+      "M y",
+      "ĠRe ve",
+      "ĠL os",
+      "inf rared",
+      "Ġint ram",
+      "l ated",
+      "pl ankton",
+      "ĠG rant",
+      "pi per",
+      "Ġanten nas",
+      "Ġb ol",
+      "f p",
+      "ĠV it",
+      "Com par",
+      "ok en",
+      "Ġke ys",
+      "ĠCl ub",
+      "in ery",
+      "ĠF oot",
+      "Ġwarm ing",
+      "m ond",
+      "Ġm iles",
+      "Ġspe aking",
+      "ĠI v",
+      "Ġconform ational",
+      "ĠO k",
+      "Ġun ified",
+      "Ġassemb led",
+      "Ġinver ted",
+      "Ġf elt",
+      "correspond ing",
+      "ĠE CM",
+      "ĠN SC",
+      "Ġind oor",
+      "g ov",
+      "Ġantagon ist",
+      "unc hed",
+      "ĠJ ava",
+      "ĠComb ined",
+      "tiv ities",
+      "Ġaltern ating",
+      "ã Ĥ",
+      "ĠDiagn osis",
+      "Ġdistinc tion",
+      "le igh",
+      "ĠT ogether",
+      "Ġparticip ating",
+      "Ġgl omer",
+      "oc he",
+      "Ġcopy right",
+      "ĠG TP",
+      "ĠV ar",
+      "Ġammon ium",
+      "Ġfacilit ates",
+      "Ġperf usion",
+      "ĠL B",
+      "f ull",
+      "Ġre ti",
+      "ifer ase",
+      "Ġimmunos up",
+      "ĠIm plementation",
+      "Ġpo res",
+      "ĠB B",
+      "ĠB ud",
+      "ĠV O",
+      "ĠV o",
+      "Ġphysic ian",
+      "ĠA UC",
+      "Ġcertain ly",
+      "μ m",
+      "ĠK ol",
+      "Ġw rap",
+      "m iddle",
+      "Ġsil encing",
+      "Ġfresh water",
+      "ig an",
+      "are a",
+      "A I",
+      "Ġmicro tub",
+      "Ġarrang ed",
+      "struc tive",
+      "ĠReg ular",
+      "ĠF ile",
+      "al ks",
+      "Ġpl ain",
+      "Ġinte grable",
+      "ĠM embrane",
+      "ist ors",
+      "Ġaqu atic",
+      "Ġwork flow",
+      "ĠG er",
+      "ul ant",
+      "Ġactiv ates",
+      "T erm",
+      "ĠUp on",
+      "ĠP ut",
+      "V ar",
+      "ĠO D",
+      "hal f",
+      "Ġul cer",
+      "ĠB O",
+      "ĠG y",
+      "ren ces",
+      "Ġpur ity",
+      "Ġarri ve",
+      "ĠSign ificant",
+      "ĠM AC",
+      "ĠOther wise",
+      "o ured",
+      "Ġt an",
+      "ĠR L",
+      "ĠQ TL",
+      "Ġammon ia",
+      "v mode",
+      "Ġmagn esium",
+      "Ġac knowled",
+      "Ġaltern atives",
+      "id ents",
+      "r Vert",
+      "ĠCom plete",
+      "ĠB one",
+      "y er",
+      "ĠB ab",
+      "Ġe ut",
+      "Ġno vo",
+      "disci plinary",
+      "Ġsevere ly",
+      "uk i",
+      "ĠP N",
+      "leave vmode",
+      "cl ip",
+      "ĠSy nd",
+      "ĠMIM O",
+      "ade qu",
+      "ĠArc tic",
+      "ly cer",
+      "RE T",
+      "ens ed",
+      "co ated",
+      "V P",
+      "Ġl akes",
+      "Ġch urch",
+      "Ġhom ologous",
+      "Ġoxid ase",
+      "ĠA ud",
+      "Ġincre ment",
+      "Ġneut rinos",
+      "ar bon",
+      "T YPE",
+      "iz umab",
+      "ut able",
+      "Ġimp lying",
+      "ĠMo tion",
+      "Ġâī ĥ",
+      "Ġp ages",
+      "Ġplaus ible",
+      "ĠN L",
+      "Ġis otop",
+      "ĠH yd",
+      "A tt",
+      "lat tice",
+      "sh ore",
+      "Ġsuc ceed",
+      "Ġsup posed",
+      "ĠTrans mission",
+      "D imensional",
+      "ingu istic",
+      "Ġcont ours",
+      "Ġcon comit",
+      "Ġagre es",
+      "ĠD ani",
+      "qu ar",
+      "Ġsh ield",
+      "Ġo zone",
+      "ĠT et",
+      "lb rack",
+      "Ġw at",
+      "Ġcyt ochrome",
+      "ta iled",
+      "p ix",
+      "Ġco ex",
+      "ĠVi ew",
+      "od ef",
+      "ĠW ild",
+      "ĠL E",
+      "h op",
+      "Ġpoint ing",
+      "unc ture",
+      "Ġec ology",
+      "Ġb ab",
+      "re a",
+      "eg o",
+      "Ġviol ence",
+      "Ġt RNA",
+      "ĠR N",
+      "p ent",
+      "ore l",
+      "ĠPar allel",
+      "Ġdri ves",
+      "nob reak",
+      "Ġh olog",
+      "Ġprob able",
+      "Ġent ering",
+      "Ġs ink",
+      "Ġsw elling",
+      "pro ducing",
+      "âĨĴ âĪŀ",
+      "ĠSaf ety",
+      "Ġanaly se",
+      "ser ies",
+      "Ġdri vers",
+      "K S",
+      "ĠR MS",
+      "Ġgene tics",
+      "ĠF red",
+      "Ġsub m",
+      "Ġscienti sts",
+      "ĠF D",
+      "ĠSol utions",
+      "ĠF ab",
+      "Ġen compass",
+      "commut ative",
+      "Ġadi abatic",
+      "but yl",
+      "PE G",
+      "Ġα β",
+      "ĠSt an",
+      "Ġclust ered",
+      "Ġhold ing",
+      "ĠB eck",
+      "ĠY an",
+      "Ġast er",
+      "Ġecon om",
+      "Ġign ored",
+      "u ro",
+      "yl es",
+      "ubb les",
+      "Ġf ate",
+      "Ġper ceptions",
+      "Ġl in",
+      "é n",
+      "Ġact u",
+      "Ġar sen",
+      "Ġb a",
+      "ep och",
+      "ĠS tim",
+      "Ġmedic ations",
+      "EC s",
+      "ĠMin istry",
+      "ĠPubl isher",
+      "Ġdep ri",
+      "Ġob struction",
+      "ĠmRNA s",
+      "Ġbro ther",
+      "Ġcros sover",
+      "ĠT urb",
+      "t ation",
+      "Ġt ank",
+      "ĠM em",
+      "Ġint estine",
+      "Ġmicrogl ia",
+      "ĠMax well",
+      "Ġjuris dic",
+      "Ġphen yl",
+      "hy per",
+      "um s",
+      "ĠH IF",
+      "ĠS hen",
+      "Ġcheck point",
+      "ĠBrown ian",
+      "Ġâĭ Ĩ",
+      "ĠSt rain",
+      "ĠExt raction",
+      "Ġbatter ies",
+      "ĠP le",
+      "ĠCon ditions",
+      "Ġincons istent",
+      "ĠH ost",
+      "yp ical",
+      "Ġc rops",
+      "al g",
+      "ĠF I",
+      "ant a",
+      "Ġfound ed",
+      "Ġmark s",
+      "dist ribution",
+      "ĠÎ ¹",
+      "Ġh ors",
+      "Ġsn ap",
+      "W M",
+      "Ġmanifest ations",
+      "em pl",
+      "Ġprov ing",
+      "le ading",
+      "ĠA CE",
+      "ĠL ED",
+      "ch annels",
+      "Ġl ift",
+      "F unction",
+      "in ase",
+      "super vised",
+      "ĠU ser",
+      "Ġphys iology",
+      "Ġlink ing",
+      "p ressed",
+      "Ġ iff",
+      "ĠJ im",
+      "Ġglut athione",
+      "ĠT I",
+      "Ġan e",
+      "en osis",
+      "Ġcollec tions",
+      "Ġgenetic ally",
+      "ĠFil ter",
+      "ĠCh icago",
+      "ĠServ ices",
+      "Ġsuper symmetric",
+      "Ġstri king",
+      "Ġir rig",
+      "oc occal",
+      "Ġfib res",
+      "Ġecos ystems",
+      "um ing",
+      "f ly",
+      "Ġlung s",
+      "Ġcovari ates",
+      "Ġlay out",
+      "ĠR aj",
+      "Ġsumm ation",
+      "abl ed",
+      "Ġfre ely",
+      "Ġre vised",
+      "Ġcut s",
+      "ĠIntegr ated",
+      "Ġph armaceutical",
+      "Ġrespir ation",
+      "ĠB ill",
+      "Ġest rogen",
+      "ra int",
+      "Ġpercent ages",
+      "ĠP f",
+      "ĠG F",
+      "methyl ene",
+      "Ġorig ins",
+      "tr im",
+      "mat ch",
+      "it ney",
+      "ĠY e",
+      "Ġalloc ated",
+      "manif old",
+      "ĠT ris",
+      "ĠL ys",
+      "Ġcomp ressed",
+      "ore r",
+      "Ġhim self",
+      "Ġqu in",
+      "ĠAss embly",
+      "sing le",
+      "tem poral",
+      "Ġs oph",
+      "Ġepidem iological",
+      "Ġknock out",
+      "Ġcomp ares",
+      "ĠSens itivity",
+      "Ġgir ls",
+      "ĠV alley",
+      "al id",
+      "ĠSchem e",
+      "ĠCO MP",
+      "Ġrefrac tive",
+      "ĠOff ice",
+      "Ġlat est",
+      "Ġp rices",
+      "car boxyl",
+      "Ġecon omy",
+      "Ġbo oks",
+      "ĠD D",
+      "Ġne oplas",
+      "app ings",
+      "Ġfol ding",
+      "moment um",
+      "pot ent",
+      "Ġpref ix",
+      "ĠRiemann ian",
+      "ĠER K",
+      "ĠPath way",
+      "Ġlar val",
+      "ol or",
+      "Ġat titude",
+      "geq slant",
+      "Ġg ates",
+      "Ġagon ist",
+      "ĠïĢ ¨",
+      "ĠM CF",
+      "ost atic",
+      "m icro",
+      "Ġdo ubl",
+      "ĠPar ameter",
+      "Ġequival ently",
+      "Ġs rc",
+      "M ost",
+      "ĉ ĠĠĠ",
+      "Ġrhe umat",
+      "ĠH um",
+      "reg ion",
+      "Ġwind s",
+      "Ġquad rup",
+      "cal es",
+      "ulf ide",
+      "bal anced",
+      "U nder",
+      "gener ated",
+      "oplas mic",
+      "Ġweight ing",
+      "ĠN ov",
+      "vel oc",
+      "util s",
+      "ĠA CT",
+      "Ġvulner able",
+      "d c",
+      "Ġstrom al",
+      "Ġex acerb",
+      "H V",
+      "Ġperfect ly",
+      "t xt",
+      "di rection",
+      "og on",
+      "Ġb im",
+      "ĠM arg",
+      "it ons",
+      "Ġterm ination",
+      "ed a",
+      "Ġpre treatment",
+      "Ġimportant ly",
+      "Ġd uc",
+      "Ġartif acts",
+      "St ud",
+      "ot ensin",
+      "rel and",
+      "ah n",
+      "Ġdeploy ed",
+      "ĠE F",
+      "ens ing",
+      "ĠC ard",
+      "ĠJ ordan",
+      "ap unov",
+      "Ġanest hesia",
+      "Ġatheros clerosis",
+      "in ner",
+      "struct ural",
+      "ĠAs p",
+      "through put",
+      "ur ities",
+      "Ġin set",
+      "with out",
+      "Ġac quire",
+      "Ġcomb ines",
+      "ĠSh ar",
+      "M ASK",
+      "ĠL iter",
+      "Ġcons cious",
+      "isc ell",
+      "cons istent",
+      "y st",
+      "Ġfil aments",
+      "ĠAl ice",
+      "ĠG round",
+      "Ġm TOR",
+      "vers al",
+      "Ġline ages",
+      "par ticles",
+      "a roscopic",
+      "ĠPro ced",
+      "Ġorient ations",
+      "ĠM ouse",
+      "Ġaccording ly",
+      "Ġsuppress or",
+      "Ġdestr uction",
+      "O V",
+      "ĠProtein s",
+      "PE CT",
+      "Ġc up",
+      "Ġmon omer",
+      "plement al",
+      "Ġneutroph il",
+      "Ġer up",
+      "Ġt ac",
+      "Ġasympt omatic",
+      "ĠEm bed",
+      "ĠRad iation",
+      "ĠG ame",
+      "Ġneed le",
+      "Ġre use",
+      "ĠD utch",
+      "Ġjuven ile",
+      "Ġmoment a",
+      "ĠB ose",
+      "Ġde veloper",
+      "Ġresidual s",
+      "Å ¡",
+      "Ġc ognition",
+      "ĠReg ional",
+      "Y ou",
+      "ĠCon cent",
+      "oc in",
+      "ĠPar tial",
+      "Ġcomplet es",
+      "ĠSing h",
+      "ĠEx c",
+      "ĠIs olation",
+      "ĠStruct ures",
+      "Ġinter mitt",
+      "Ex ception",
+      "Ġanaly tically",
+      "Ġelectric ity",
+      "â ĭ",
+      "Ä į",
+      "Ġprote ome",
+      "Ġ ic",
+      "k al",
+      "inu x",
+      "ĠB eyond",
+      "Ġim plied",
+      "AS H",
+      "Ġcl one",
+      "ĠRuss ia",
+      "ĠH od",
+      "tebr ates",
+      "Ġpro xy",
+      "hold er",
+      "el ve",
+      "Ġval ley",
+      "ut ely",
+      "Ġj obs",
+      "rup tion",
+      "ro ids",
+      "ĠWh y",
+      "ep ing",
+      "ĠY et",
+      "Ġp yl",
+      "Ġb ra",
+      "il ization",
+      "et ers",
+      "Ġad ver",
+      "Ġo ve",
+      "k ernel",
+      "s amples",
+      "ordin ate",
+      "ĠAssum ing",
+      "Ġcontamin ated",
+      "Ġb ipolar",
+      "Ġl ac",
+      "Ġl uc",
+      "Ġcentrifug ation",
+      "B oth",
+      "Ġ nd",
+      "Ġt ib",
+      "B efore",
+      "ĠImmun e",
+      "Ġas h",
+      "Ġcondition ed",
+      "ĠR ank",
+      "N OS",
+      "Ġnanopar ticle",
+      "Ġdepend encies",
+      "Ġhouse holds",
+      "ag ers",
+      "Ġspect rophot",
+      "Ġb ile",
+      "ĠH ans",
+      "ĠAcknowledg ements",
+      "r atio",
+      "ĠSecond ary",
+      "Ġdown regulated",
+      "f ixed",
+      "O bs",
+      "ĠH L",
+      "Ġs ends",
+      "ting s",
+      "Ġf i",
+      "ĠPa per",
+      "Ġultra violet",
+      "ĠB all",
+      "Ġdr astic",
+      "ail ure",
+      "o il",
+      "ex change",
+      "ĠD an",
+      "ĠAut o",
+      "Ġarch ae",
+      "ĠCol lection",
+      "Ġantiv iral",
+      "ĠC hemistry",
+      "Ġf err",
+      "cho ice",
+      "v ac",
+      "ol ipid",
+      "Ġd anger",
+      "ĠL ittle",
+      "Ġde hyd",
+      "Ġoccas ion",
+      "oprop yl",
+      "ab e",
+      "Ġinterfer on",
+      "Ġex port",
+      "on itrile",
+      "p d",
+      "ĠCon text",
+      "ru z",
+      "ĠD ys",
+      "Ġassemb l",
+      "Ġoil s",
+      "Im age",
+      "row ing",
+      "Ġane urys",
+      "Ġliqu ids",
+      "Ġac tively",
+      "Ġev apor",
+      "ĠP resent",
+      "Ġconstit utive",
+      "ĠS ite",
+      "Ġsc ript",
+      "Ġrepe ats",
+      "ĠS IR",
+      "ĠFil m",
+      "ĠSant a",
+      "ĠRepresent ation",
+      "ĠA ma",
+      "ord on",
+      "ĠMo lecule",
+      "Ġgover ning",
+      "ĠSo il",
+      "V er",
+      "Ġphot onic",
+      "tif y",
+      "ĠLew is",
+      "at hered",
+      "Ġcategor ical",
+      "iscell aneous",
+      "up date",
+      "Ġdefic it",
+      "Ġadj uvant",
+      "ĠHen ry",
+      "G roup",
+      "ist ency",
+      "ag raph",
+      "ĠImpro ving",
+      "E l",
+      "Ġfl ame",
+      "rog ate",
+      "om orph",
+      "Ġqu bits",
+      "Ġillustr ation",
+      "ĠFlor ida",
+      "ĠD G",
+      "big cup",
+      "Ġprov ince",
+      "egrad ation",
+      "ĠLand au",
+      "Ġgr ating",
+      "Ġins ects",
+      "Ġd raft",
+      "ĠH b",
+      "Ġs s",
+      "ĠR as",
+      "Ġmuc osa",
+      "Ġhydrox yl",
+      "Ġmod est",
+      "Ġconfir ming",
+      "ĠGalax ies",
+      "G aussian",
+      "ĠRet rie",
+      "Ġrest ored",
+      "m emory",
+      "Ġreinfor ced",
+      "r ific",
+      "Ġass isted",
+      "Ġaffili ations",
+      "R C",
+      "duc er",
+      "ĠInt ellig",
+      "ĠA SD",
+      "mod ium",
+      "Ġo mitted",
+      "ok ers",
+      "Ġgu ided",
+      "Ġgraph ical",
+      "ĠQ ual",
+      "D ue",
+      "Ġn emat",
+      "vari able",
+      "Ġsenes cence",
+      "Ġpip e",
+      "Ġsustain able",
+      "Ġteac her",
+      "Ġth ing",
+      "ĠGP U",
+      "T B",
+      "Ġre form",
+      "Ġref lex",
+      "Ġindic ative",
+      "ab out",
+      "Ġop i",
+      "eff ect",
+      "Ġdispers ed",
+      "k h",
+      "it helial",
+      "ĠT reg",
+      "i pl",
+      "ĠAut omatic",
+      "Ġn itro",
+      "com plete",
+      "Ġbos ons",
+      "Ġp ac",
+      "Ġavoid ing",
+      "is l",
+      "pl asty",
+      "respons ive",
+      "d est",
+      "ĠB rad",
+      "ĠDec ision",
+      "ĠDisc overy",
+      "Ġchick en",
+      "m us",
+      "ĠW ITH",
+      "Ġt ric",
+      "Ġqu artz",
+      "onstr uction",
+      "ĠField s",
+      "Ġass im",
+      "opro t",
+      "Ġguarant eed",
+      "f at",
+      "ic ts",
+      "Ġch ol",
+      "id o",
+      "ĠK L",
+      "Ġch itosan",
+      "ĠN d",
+      "ĠO scill",
+      "Ġevol ve",
+      "c u",
+      "Ġm ast",
+      "Ġam ph",
+      "tor ch",
+      "V is",
+      "enti ty",
+      "ĠAd am",
+      "Ġdev oted",
+      "Ġeth ical",
+      "Ġprem ature",
+      "Ġconsum er",
+      "Ġrecurs ive",
+      "Ġglu on",
+      "Ġmoder ately",
+      "Ġmod alities",
+      "Ġcan al",
+      "for ce",
+      "ĠCh lor",
+      "sl ash",
+      "st en",
+      "Ġcommerc ially",
+      "ong s",
+      "Ġstim ulate",
+      "atin um",
+      "ĠR ail",
+      "Ġconv ective",
+      "Ġarter ies",
+      "in v",
+      "ĠW ol",
+      "ĠL ung",
+      "let es",
+      "raph y",
+      "ĠH I",
+      "Ġgraph ite",
+      "Ġhous ing",
+      "e ach",
+      "Ġcal or",
+      "acet amide",
+      "roc hemical",
+      "Ġhand s",
+      "Ġelucid ate",
+      "ĠCh and",
+      "ro ad",
+      "nov a",
+      "ĠLine age",
+      "Ġr am",
+      "Ġf ight",
+      "Ġrecommend ation",
+      "Ġamong st",
+      "Ġswit ches",
+      "ber ry",
+      "Ġthere in",
+      "al gebras",
+      "ĠT aken",
+      "az z",
+      "Ġf urn",
+      "Ġam el",
+      "Ġteac hers",
+      "ar n",
+      "Ġavoid ed",
+      "Ġaver ages",
+      "am er",
+      "ĠCon dition",
+      "Ġdis location",
+      "ir con",
+      "Ġadoles cent",
+      "Ġt ur",
+      "en v",
+      "Ġz e",
+      "D L",
+      "load ing",
+      "ic idal",
+      "c ategory",
+      "ĠD B",
+      "Ġmuc osal",
+      "ĠR G",
+      "Ġtaxon omic",
+      "Ġmut agen",
+      "ĠSt age",
+      "n ecess",
+      "ĠP erm",
+      "Ġoc clusion",
+      "Ġexplo ited",
+      "Ġana erobic",
+      "ul ed",
+      "Ġwant ed",
+      "ĠComb ining",
+      "Ġsub cutaneous",
+      "Rec omm",
+      "Ġdiscuss es",
+      "Ġcounter part",
+      "ĠF B",
+      "Ġadsorb ed",
+      "d on",
+      "M any",
+      "ĠSwed en",
+      "ĠAnd rew",
+      "enh anced",
+      "Ġdoc tor",
+      "ĠKore an",
+      "ĠS AR",
+      "Ġm ating",
+      "at uration",
+      "ĠL atin",
+      "Ġsor ting",
+      "Ġsk ip",
+      "O s",
+      "Ġw ife",
+      "Ġcom mittee",
+      "l vert",
+      "ĠA CC",
+      "ĠCom m",
+      "Ġsub tle",
+      "ĠSur vival",
+      "b ecause",
+      "Ġfe at",
+      "ĠPort ug",
+      "AR Y",
+      "ĠI SB",
+      "it ron",
+      "Ġs ectors",
+      "Ġadj oint",
+      "ĠAlex ander",
+      "Ġimp urity",
+      "ĠMar ine",
+      "l act",
+      "Ġtrap ping",
+      "Ġgeneral ize",
+      "fil ter",
+      "Ġpolar ity",
+      "Al so",
+      "Ġstabil ized",
+      "ĠVir gin",
+      "Ġst ores",
+      "P AGE",
+      "Ġdraw back",
+      "Ġâİ ª",
+      "j et",
+      "Ġsubstit uted",
+      "L INE",
+      "Ġoutper forms",
+      "Ġterm ed",
+      "Ġweek ly",
+      "Ġpoly c",
+      "Ġf used",
+      "Ġfer romagnetic",
+      "l r",
+      "ell ites",
+      "ĠT urn",
+      "ĠC ulture",
+      "pr ise",
+      "Å Ĥ",
+      "om position",
+      "elf are",
+      "ĠGo ogle",
+      "o arth",
+      "Ġ ë",
+      "Ġm ist",
+      "ĠMat hematics",
+      "S ET",
+      "Ġepoch s",
+      "Ġcont ras",
+      "ish ment",
+      "ĠFirst ly",
+      "Ġdecl ared",
+      "a ur",
+      "ĠP ed",
+      "Ġreplic ate",
+      "Ġel igible",
+      "Ġconc aten",
+      "Ġc ig",
+      "Ġtri plet",
+      "f ound",
+      "ĠC z",
+      "Ġaccompl ished",
+      "Ġgover ned",
+      "on uclear",
+      "ĠN Y",
+      "ĠEth iop",
+      "Ġin ject",
+      "Ġe osin",
+      "ann on",
+      "ol o",
+      "ĠM HC",
+      "Ġpre operative",
+      "Ġd ates",
+      "Ġs igma",
+      "L ong",
+      "ĠRes on",
+      "Ġsympt omatic",
+      "Ġvolunte ers",
+      "Ġco operation",
+      "Ġar r",
+      "Ġclon ed",
+      "Ġd ent",
+      "ĠS ob",
+      "Ġcath ode",
+      "ct x",
+      "Ġ encephal",
+      "Ġp iv",
+      "vi ve",
+      "um etric",
+      "ĠF F",
+      "Ġunde restim",
+      "Ġc oded",
+      "Ġanal ges",
+      "spect ral",
+      "Ġattrac ted",
+      "Ġtw enty",
+      "Ġin active",
+      "Ġvic tim",
+      "Ġhold er",
+      "ogen es",
+      "Ġsuff ering",
+      "re x",
+      "Ġpro phyl",
+      "ĠUnivers al",
+      "Ġden om",
+      "st olic",
+      "ans ion",
+      "SI ZE",
+      "ĠHC V",
+      "Ġtechn ological",
+      "CN N",
+      "en ching",
+      "Ġdeb ris",
+      "ĠBound ary",
+      "link ing",
+      "Ġstop ped",
+      "ĠD ie",
+      "ĠCos m",
+      "Ġturn ing",
+      "Ġgly coprotein",
+      "ĠK umar",
+      "Ġp g",
+      "ĠB Y",
+      "Ġr ising",
+      "ĠR OC",
+      "Des pite",
+      "ĠBo olean",
+      "il der",
+      "Ġexpon ents",
+      "in ters",
+      "print f",
+      "Ġl it",
+      "t rack",
+      "Ġf idelity",
+      "Ġsm oke",
+      "ot emporal",
+      "Ġad missible",
+      "ĠBol tzmann",
+      "T F",
+      "ol ite",
+      "li ament",
+      "Ġcalc ulus",
+      "iti zed",
+      "Ġdiver gent",
+      "Ġcolon ization",
+      "Ġconver gent",
+      "ĠH as",
+      "Ġconsum ers",
+      "Ġmy c",
+      "Ġcon tig",
+      "Ġepidem iology",
+      "é s",
+      "ĠAss oci",
+      "g iven",
+      "Ġwh ilst",
+      "ĠK ur",
+      "Ġreason ably",
+      "Ġaer obic",
+      "se par",
+      "Ġche cks",
+      "ĠSem antic",
+      "Ġserv ing",
+      "ĠAt mosp",
+      "Ġoxid ized",
+      "c oupled",
+      "Ġbio Rxiv",
+      "Ġtun ed",
+      "usp ended",
+      "Ġindirect ly",
+      "ĠC AD",
+      "ĠCurrent ly",
+      "Ġbehavi ours",
+      "ĠPP AR",
+      "r ors",
+      "ere b",
+      "Ġwid ths",
+      "di agonal",
+      "erv ice",
+      "Ġo le",
+      "me ans",
+      "IM E",
+      "ĠT racking",
+      "Ġac knowledge",
+      "ĠH on",
+      "ĠTechn iques",
+      "ĠOx id",
+      "bl ind",
+      "Ġdi ast",
+      "nam ed",
+      "asi tic",
+      "Ġprepar ations",
+      "ĠAr th",
+      "Ġpres erves",
+      "Ġf asc",
+      "Ġwave form",
+      "ĠC rystal",
+      "Ġunc om",
+      "Ġel ast",
+      "Ġfunction ally",
+      "H om",
+      "ĠCo ast",
+      "Ġop tic",
+      "ĠAltern atively",
+      "on yl",
+      "ĠL ig",
+      "al dehyde",
+      "Ġsim ulator",
+      "Ġdram atic",
+      "if era",
+      "Ġexhib iting",
+      "Ġbehaviour al",
+      "th ick",
+      "xt ure",
+      "Ġexec utive",
+      "Ġcondens ate",
+      "ĠOut comes",
+      "T ext",
+      "oin ted",
+      "ĠCopy right",
+      "Ġd c",
+      "od d",
+      "ĠD iversity",
+      "ch ip",
+      "ĠBu ilding",
+      "Ġpuls ed",
+      "har monic",
+      "Ġclinic ians",
+      "d p",
+      "Ġq PCR",
+      "mark s",
+      "Ġapprec i",
+      "ĠL aser",
+      "Ġsize of",
+      "y rene",
+      "Ġco operative",
+      "gener ative",
+      "ĠL ib",
+      "Ġdispers al",
+      "Ġevol ving",
+      "ĠSt atus",
+      "Ġsuper con",
+      "ĠM amm",
+      "Ġinters titial",
+      "isen berg",
+      "Ġâ ľ",
+      "Ġconf ocal",
+      "Ġmod ulates",
+      "h our",
+      "Ġper oxide",
+      "depend ence",
+      "Ġperturb ed",
+      "ill ation",
+      "Ġpl aque",
+      "ĠNe umann",
+      "Ġtrig gers",
+      "om ain",
+      "ĠAd ministration",
+      "ol ia",
+      "ĠM IC",
+      "osa ic",
+      "ĠG B",
+      "text normal",
+      "Ġdomin ance",
+      "ĠEx per",
+      "C AM",
+      "ĠAb out",
+      "ĠG arc",
+      "Ġsummar izes",
+      "A pp",
+      "charomy ces",
+      "tif icial",
+      "Ġgly cerol",
+      "ĠAssum ption",
+      "Ġt ect",
+      "ĠF W",
+      "Ġcot ton",
+      "gen eral",
+      "ĠF ern",
+      "P t",
+      "Ġwork er",
+      "Ġan ion",
+      "gram s",
+      "re q",
+      "Ġlo oks",
+      "Ġimplement ations",
+      "ĠCol umb",
+      "ag i",
+      "ĠAt tention",
+      "ĠTe am",
+      "on ing",
+      "on ential",
+      "tin y",
+      "ĠHigh ly",
+      "text up",
+      "Ġinver tible",
+      "oc ortic",
+      "In f",
+      "ĠOff icial",
+      "ĠMod elling",
+      "Ġincl usions",
+      "Ġbl ank",
+      "Ġs ight",
+      "ĠG amma",
+      "Ġlept on",
+      "Ġpneumonia e",
+      "Ġro tor",
+      "Ġeth nic",
+      "Ġre tain",
+      "v arying",
+      "ĠE B",
+      "Ġast rocytes",
+      "ĠN orm",
+      "Ġnan om",
+      "class ical",
+      "Ġsh adow",
+      "ĠRef erences",
+      "ĠF S",
+      "Ġnon negative",
+      "b ond",
+      "ĠC oh",
+      "Ġnum py",
+      "Ġo ct",
+      "sp an",
+      "rac ts",
+      "Ġnot ably",
+      "Ġsoph istic",
+      "P AR",
+      "Ġhorm ones",
+      "Ġtens ors",
+      "ĠÌ Ħ",
+      "ĠConst raints",
+      "Ġâ IJ",
+      "Ġtrans it",
+      "Ġrun time",
+      "aut hor",
+      "Ġprom pt",
+      "ĠS G",
+      "Ġg rate",
+      "ce mia",
+      "ĠLy apunov",
+      "con vex",
+      "Ġforecast ing",
+      "p ush",
+      "Ġjurisdic tional",
+      "Ã Ģ",
+      "Ġbiom edical",
+      "Ġepile psy",
+      "fe ature",
+      "wik i",
+      "Vi ew",
+      "Ġless er",
+      "Ġconjug ated",
+      "Ġwa iting",
+      "ĠW ord",
+      "I Z",
+      "Ġhydro xy",
+      "Ġdis p",
+      "Ġseed ed",
+      "fit ting",
+      "Ġstrat ification",
+      "Ġend point",
+      "Ġmedi ators",
+      "duc tive",
+      "Ġinj ections",
+      "ĠMicro bi",
+      "Ġins ert",
+      "ĠEm b",
+      "Ġstop ping",
+      "w elling",
+      "Ġirradi ated",
+      "Ġmetall icity",
+      "vin yl",
+      "Ġplasm ids",
+      "R ep",
+      "ĠDiff erenti",
+      "ĠSm art",
+      "ĠIdentif ier",
+      "ĠB F",
+      "rop ic",
+      "Ġkinem atics",
+      "Ġinoc ulated",
+      "C K",
+      "aus es",
+      "ĠReturn s",
+      "re ement",
+      "Ġantic ancer",
+      "Ġspecific ations",
+      "Ġadd s",
+      "Ġst ake",
+      "Ġwhe el",
+      "ü ller",
+      "ĠS on",
+      "Ġrup ture",
+      "Ġsol d",
+      "th an",
+      "Ġinter medi",
+      "ĠN ik",
+      "Ġt uple",
+      "est abl",
+      "Ġnor the",
+      "Ġsuppress es",
+      "Ġf et",
+      "Ġwas hing",
+      "Ġinter play",
+      "Ġregular ly",
+      "EX T",
+      "Ġemploy ees",
+      "y z",
+      "rup ted",
+      "et ts",
+      "ĠU AV",
+      "Ġdifferenti able",
+      "ing e",
+      "MD A",
+      "Ġh o",
+      "Ġt ags",
+      "Ġcomp atibility",
+      "ĠÃ ĥ",
+      "b us",
+      "ĠU C",
+      "Ġtok ens",
+      "Ġcl ients",
+      "Ġpres cription",
+      "ĠÌ Ī",
+      "ĠRe action",
+      "veloc ity",
+      "ĠN LR",
+      "ĠG ast",
+      "ĠPlas modium",
+      "ĠC ut",
+      "Ġn as",
+      "gra ined",
+      "Ġchrom osomal",
+      "Ġpossess es",
+      "Ġm ath",
+      "Ġe lected",
+      "plac ement",
+      "Ġcollect ing",
+      "Ġg els",
+      "ai re",
+      "Ġdeform ations",
+      "ra ise",
+      "Ġfl ank",
+      "sulf anyl",
+      "z ens",
+      "pri ate",
+      "Ġchlor ophyll",
+      "ab i",
+      "avail able",
+      "Ø §",
+      "Ġt ack",
+      "field s",
+      "Ġrich ness",
+      "Ġimpl ants",
+      "ob enz",
+      "id ential",
+      "Ġbill ion",
+      "ut or",
+      "ĠISB N",
+      "Ġins urance",
+      "N ET",
+      "Ġin adequ",
+      "Ġmerg ed",
+      "ĠR ange",
+      "Ġavoid ance",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠ",
+      "ric s",
+      "Ġexcl usive",
+      "L V",
+      "Ġï£ ²",
+      "Ġcategor ized",
+      "Ġultrason ic",
+      "ip e",
+      "ic ans",
+      "ĠA PP",
+      "Ġtra umatic",
+      "B a",
+      "ĠAss ay",
+      "ĠG rid",
+      "ĠClass ical",
+      "ĠD ES",
+      "Ġsoy bean",
+      "Ġtop ography",
+      "ĠCont roll",
+      "Ġemo tions",
+      "Ġcarbohyd rate",
+      "Ġcons ol",
+      "ox yl",
+      "Ġbifurc ation",
+      "Ġco il",
+      "f ind",
+      "Ġw itness",
+      "ĠL F",
+      "th reshold",
+      "Ġaddress ing",
+      "Ġsc rew",
+      "Ġact or",
+      "ĠW ell",
+      "Ġï£ °",
+      "ï ĺ",
+      "ĠD F",
+      "ĠCor poration",
+      "ĠMitochond rial",
+      "Ġk pc",
+      "und ers",
+      "Ġfib rin",
+      "ax el",
+      "Ġpol yt",
+      "Ġshap ed",
+      "re z",
+      "ste resis",
+      "ĠComp rehens",
+      "Ġï£ ³",
+      "d h",
+      "Ġsem ic",
+      "Ġm ot",
+      "ĠDav is",
+      "sk a",
+      "ĠL H",
+      "Ġexpans ions",
+      "ack s",
+      "Ġoptim izing",
+      "e ak",
+      "ĠQ i",
+      "m ul",
+      "og raft",
+      "Ġsu icide",
+      "cal ar",
+      "ĠSc ott",
+      "Ġth inking",
+      "Ġdirec tional",
+      "Ġsurfact ant",
+      "Ġdegrad ed",
+      "Ġregim en",
+      "it ative",
+      "ĠV ersion",
+      "ĠM aster",
+      "ĠSim ulations",
+      "NC BI",
+      "l ip",
+      "Ġre agents",
+      "Ġpost ed",
+      "os us",
+      "Ġlay ered",
+      "ĠSpect rum",
+      "ĠGraph s",
+      "bur st",
+      "Ġl ived",
+      "Ġelement al",
+      "Ġï£ »",
+      "ĠDisc rete",
+      "Ġexcl uding",
+      "Ġorigin ating",
+      "ĠG ames",
+      "continu ous",
+      "AT ED",
+      "Ġpy ram",
+      "lu ent",
+      "Ġtw isted",
+      "ĠN b",
+      "ox icity",
+      "Ġsc r",
+      "Ġf un",
+      "ĠSeg mentation",
+      "Ġphen ol",
+      "Ġmet ers",
+      "ĠE igen",
+      "ĠWe ak",
+      "Ġschem atic",
+      "r one",
+      "Ġphil os",
+      "ti tis",
+      "ĠI reland",
+      "Ġg y",
+      "ĠPT M",
+      "Ġpack ing",
+      "il inear",
+      "z eros",
+      "Ġubiqu itin",
+      "ĠPress ure",
+      "Ġinf iltr",
+      "EN S",
+      "val idation",
+      "Ġpr one",
+      "Ġout line",
+      "h s",
+      "reng th",
+      "Ġat tain",
+      "Ġt we",
+      "Ġt andem",
+      "C an",
+      "Ġlat itude",
+      "uit ary",
+      "Ġvolt ages",
+      "ĠGa o",
+      "Ġpharmac okine",
+      "Ġcontext ual",
+      "Ġx yl",
+      "els on",
+      "ĠMet abolic",
+      "od en",
+      "ti les",
+      "ff icking",
+      "Ġdistill ed",
+      "Ġal ph",
+      "Ġpie zo",
+      "g rowth",
+      "Ġb ore",
+      "Ġredund ant",
+      "Ġdemonstr ation",
+      "Ġi k",
+      "Ġround s",
+      "ĠS ri",
+      "fig uration",
+      "ĠRay leigh",
+      "L ine",
+      "ov ol",
+      "Ġobstac le",
+      "c n",
+      "Ġbio active",
+      "ĠO A",
+      "phys ical",
+      "at idyl",
+      "AC C",
+      "h ow",
+      "Ġresult ant",
+      "ĠH ubble",
+      "ĠV or",
+      "Ġens uring",
+      "Ġannot ations",
+      "ac yl",
+      "stit uted",
+      "ĠAm b",
+      "feed ing",
+      "Ġpresum ably",
+      "Ġblock ade",
+      "Ġs oc",
+      "ĠU rb",
+      "Ġmulti plied",
+      "Ġdiff e",
+      "Ġreflect ance",
+      "ĠKey words",
+      "ĠBay es",
+      "odef iciency",
+      "ĠB inding",
+      "in ely",
+      "ex cept",
+      "ĠUl tr",
+      "ĠBrazil ian",
+      "N umber",
+      "Ġmass less",
+      "ĠCons istent",
+      "Ġcr isis",
+      "og s",
+      "Ġres idence",
+      "Ġim per",
+      "f ts",
+      "Ġcapt ures",
+      "ĠSynd rome",
+      "Ġdimension ality",
+      "j un",
+      "Ġex haus",
+      "ĠMod ern",
+      "Ġperc enti",
+      "Le vel",
+      "ĠRespons es",
+      "Ġla unched",
+      "Ġre pos",
+      "ĠK am",
+      "at ility",
+      "Ġcaro tid",
+      "ro tic",
+      "ĠM and",
+      "U B",
+      "ĠM ixed",
+      "Ġindex es",
+      "Ġcis platin",
+      "ic an",
+      "ion ine",
+      "Ġh ab",
+      "ĠI ce",
+      "ĠG T",
+      "ĠAg g",
+      "ĠLD L",
+      "Ġvolcan ic",
+      "d B",
+      "ĠElect ric",
+      "Ġt mp",
+      "Ġgrid s",
+      "l iquid",
+      "p rom",
+      "ĠG AL",
+      "Ġp estic",
+      "Ġhel ium",
+      "Ġï£ ¹",
+      "ĠD ong",
+      "Ġmagn ification",
+      "k ip",
+      "ĠG rad",
+      "ĠWe i",
+      "ĠPD F",
+      "ĠGl uc",
+      "P ol",
+      "Ġtumor igen",
+      "yr in",
+      "Ġshel f",
+      "ad her",
+      "enti als",
+      "s n",
+      "Ġcultiv ars",
+      "Ġorbit als",
+      "ĠP EG",
+      "ĠAn ne",
+      "en o",
+      "Ġatt ended",
+      "oph ore",
+      "ish op",
+      "Ġf riends",
+      "pos able",
+      "Ġim pose",
+      "Ġend emic",
+      "Ġs ick",
+      "shif ts",
+      "ĠOut put",
+      "L M",
+      "ĠM iscellaneous",
+      "Ġthous ands",
+      "ĠD ataset",
+      "Ġperturb ative",
+      "op rec",
+      "Ġb ene",
+      "Ġre ef",
+      "Ġfoss il",
+      "Ġc ited",
+      "plic ates",
+      "Ġrel ates",
+      "ĠV II",
+      "Ġanti fer",
+      "Ġglass es",
+      "clos ure",
+      "Ġrub ber",
+      "Ġb ird",
+      "Ġsuper symmetry",
+      "Ġmes on",
+      "he ll",
+      "Ġpar ties",
+      "k ar",
+      "ĠH ur",
+      "ĠE A",
+      "ĠSt ars",
+      "oth ing",
+      "h ot",
+      "ill ar",
+      "AS P",
+      "he v",
+      "ï ĥ",
+      "a ques",
+      "Ġcoordin ated",
+      "ĠIs lands",
+      "en able",
+      "Si O",
+      "Ġexception al",
+      "C omb",
+      "ĠL ike",
+      "Ġbroad ly",
+      "ĠB ac",
+      "Ġn il",
+      "ipar tite",
+      "r ations",
+      "Ġre write",
+      "Ġsal ts",
+      "d imension",
+      "ĠVe hic",
+      "Ġhundred s",
+      "ĠU r",
+      "Ġend points",
+      "ĠMOD EL",
+      "ĠH BV",
+      "ĠVir tual",
+      "ĠCon fl",
+      "ĠPrac tice",
+      "ĠAF M",
+      "Ġadvers arial",
+      "Ġdi ameters",
+      "Ġtrans ported",
+      "RE M",
+      "ĠB art",
+      "Ġed ition",
+      "Ġturb ine",
+      "Ġmin us",
+      "otechn ology",
+      "I g",
+      "Ġbig ger",
+      "ab ul",
+      "Ġperoxid ase",
+      "wh ite",
+      "ĠS ed",
+      "di hydro",
+      "Ġseg regation",
+      "Ġreduct ase",
+      "Ġhor iz",
+      "Ġinf initely",
+      "avail ability",
+      "Ġactiv ator",
+      "Ġc ensus",
+      "press ing",
+      "Ġspir it",
+      "con ver",
+      "ĠQuantif ication",
+      "omer ase",
+      "Ġrel apse",
+      "ĠF inal",
+      "Ġover weight",
+      "a per",
+      "Ġformul ae",
+      "r r",
+      "Ġfem oral",
+      "Ġfo am",
+      "o tics",
+      "Ġprovid er",
+      "Ġinstr umental",
+      "Ġadv ice",
+      "Ġoccup ation",
+      "ass embly",
+      "bi as",
+      "ĠN OT",
+      "re stric",
+      "ĠProt ocol",
+      "ĠCandid a",
+      "ĠR hod",
+      "ard en",
+      "f under",
+      "os ens",
+      "Ġpar ams",
+      "f ront",
+      "Ġex erc",
+      "Ġgal actic",
+      "r vert",
+      "Ġim balance",
+      "Ġk illing",
+      "ĠGen omic",
+      "Ġ ip",
+      "Ġc ave",
+      "Ġf alc",
+      "ĠR M",
+      "Ġcar ries",
+      "gl obal",
+      "Ġc ube",
+      "Ġrig orous",
+      "Ġcomput es",
+      "Q P",
+      "Ġexpos ures",
+      "c over",
+      "ological ly",
+      "O per",
+      "Ġp ec",
+      "Ġin homogeneous",
+      "Ġser vers",
+      "alian a",
+      "n b",
+      "Ġexplain ing",
+      "Ġshr ink",
+      "Ġcom orbid",
+      "eth oxy",
+      "outhe ast",
+      "Ġco urses",
+      "ĠN M",
+      "ĠSh ape",
+      "Ġfl ies",
+      "ĠM ir",
+      "Ġpublic ly",
+      "Ġphot ometric",
+      "vers ible",
+      "ole v",
+      "Ġvulner ability",
+      "Ġc ations",
+      "Ġsee king",
+      "U TR",
+      "Ġdecom posed",
+      "Ġh us",
+      "Ġdisapp ear",
+      "Ġenc ounter",
+      "Ġtransform ing",
+      "Ġpolymer ic",
+      "Ġdiscre tization",
+      "otox ic",
+      "ĠI ter",
+      "ĠM ari",
+      "Ġun fold",
+      "ĠAd ult",
+      "ob acillus",
+      "met al",
+      "ber ger",
+      "rap hene",
+      "resp ective",
+      "Ġsur vive",
+      "ov ich",
+      "Ġprot ects",
+      "ĠR og",
+      "Ġimmun otherapy",
+      "ĠD SM",
+      "Ġanalog y",
+      "ĠP ER",
+      "ĠPy thon",
+      "h um",
+      "ĠAd j",
+      "ĠLik ewise",
+      "Ġï£ ®",
+      "Ġstom ach",
+      "Ġin it",
+      "Ġw ires",
+      "Ġingredi ents",
+      "Ġper ceptual",
+      "H and",
+      "B ack",
+      "Ġm ood",
+      "Ġde formed",
+      "ĠRe ad",
+      "Ġrh iz",
+      "ĠOrgan ism",
+      "ĠInd ones",
+      "ann ot",
+      "ict ory",
+      "Ġt ended",
+      "ĠS ound",
+      "ia x",
+      "S r",
+      "ĠT ab",
+      "ĠLa placian",
+      "ol uminescence",
+      "back slash",
+      "i ologic",
+      "Ġtyp ename",
+      "ĠY ear",
+      "D ependent",
+      "Ġsl ides",
+      "Ġsac rific",
+      "Ġconcomit ant",
+      "ops ies",
+      "Big g",
+      "pe ak",
+      "ĠApp lying",
+      "Ġcod on",
+      "ĠSim ultaneous",
+      "ti se",
+      "Ġter tiary",
+      "ĠP oll",
+      "Ġre vision",
+      "RA F",
+      "x mm",
+      "Ġsu ited",
+      "ĠRecomm end",
+      "ĠR y",
+      "Ġs ake",
+      "Ġstret ch",
+      "ĠSam pling",
+      "Ġtub ular",
+      "Ġpar k",
+      "Ġul timate",
+      "Ġl ands",
+      "ĠCr iter",
+      "ass ay",
+      "m or",
+      "Ġd ocking",
+      "Ġgrad ual",
+      "Ġed itor",
+      "Ġpol ice",
+      "aff in",
+      "ĠDe ath",
+      "Ġpromot ers",
+      "ass ic",
+      "Ġwr iter",
+      "ĠVol ume",
+      "is o",
+      "Ġdis ag",
+      "tok en",
+      "Ġster oid",
+      "N on",
+      "ĠMet hyl",
+      "A meric",
+      "d ue",
+      "ĠL ess",
+      "Ġdy st",
+      "ĠStat ement",
+      "ĠT wenty",
+      "Ġaccess ed",
+      "Ġblot ting",
+      "ĠCO PD",
+      "Ġste am",
+      "Ġdescrip tive",
+      "ĠV ery",
+      "Ġcapac ities",
+      "ĠPers onal",
+      "ac id",
+      "ä hler",
+      "estiv al",
+      "Con text",
+      "Ġa str",
+      "Anal ysis",
+      "Ġse pt",
+      "Ġpr inted",
+      "d ual",
+      "am an",
+      "ere r",
+      "Ġweak ness",
+      "ì Ŀ",
+      "ĠTrans lation",
+      "Ġpropag ating",
+      "ĠS ections",
+      "ac a",
+      "Ġconf usion",
+      "I K",
+      "Ġframework s",
+      "Ġsitu ated",
+      "Ġst ays",
+      "n odes",
+      "c hen",
+      "art ments",
+      "Ġfree zing",
+      "w s",
+      "net t",
+      "Ġcontroll ers",
+      "Ġsil ic",
+      "LA ST",
+      "f oot",
+      "ĠDISC U",
+      "R H",
+      "rid ine",
+      "ĠRe v",
+      "per g",
+      "py rim",
+      "fl ags",
+      "ĠGu ide",
+      "Ġspe aker",
+      "tis ol",
+      "re ll",
+      "ĠD EG",
+      "Ġf u",
+      "ĠG ut",
+      "Ġsh ar",
+      "Ġgro ss",
+      "Ġcross es",
+      "wa velength",
+      "ĠAp plied",
+      "ï ve",
+      "ĠH B",
+      "ĠEd ge",
+      "Ġiner tial",
+      "Ġv ocal",
+      "pro duction",
+      "pat hetic",
+      "Ġplan etary",
+      "Ġs ister",
+      "Ġminim a",
+      "Ġlong est",
+      "Ġfl ash",
+      "Ġperiod on",
+      "Ġepid ermal",
+      "Ġflo ating",
+      "G ET",
+      "ĠT ake",
+      "p df",
+      "ĠL iquid",
+      "Ġremark ably",
+      "S ign",
+      "Ġshell s",
+      "oglob ulin",
+      "qu ilibrium",
+      "ĠMo ore",
+      "ĠAd vers",
+      "ĠMyc obacterium",
+      "Inv itrogen",
+      "Ġth aliana",
+      "B Y",
+      "ĠB it",
+      "Ġt s",
+      "Ġsynchron ous",
+      "y x",
+      "Ġpropag ator",
+      "ĠIncre asing",
+      "ipar um",
+      "Ġfree ze",
+      "ĠSe lective",
+      "af e",
+      "Ġstre pt",
+      "ph antom",
+      "ĠGener ally",
+      "Ġaltern ate",
+      "ĠCon vergence",
+      "//////// ////////",
+      "eng ing",
+      "ĠRandom ized",
+      "de velop",
+      "pred ict",
+      "ress or",
+      "Ġmat hematics",
+      "f r",
+      "ĠComput ation",
+      "ĠMal ays",
+      "Ġbreath ing",
+      "Th rough",
+      "ĠS IM",
+      "Ġan ode",
+      "o ad",
+      "ĠAT CC",
+      "Ġconstitu ent",
+      "ĠMeas uring",
+      "Ġf MRI",
+      "Ġan emia",
+      "lies t",
+      "Ġhemisp here",
+      "Ġmaxim a",
+      "Ġtem porary",
+      "Ġd z",
+      "otox in",
+      "C ount",
+      "on ed",
+      "Ã º",
+      "Ġcollabor ative",
+      "Ġk b",
+      "Ġvers a",
+      "ĠSwed ish",
+      "ik a",
+      "Ġdial ysis",
+      "Ġper ovsk",
+      "Ġwill ing",
+      "ĠG reek",
+      "Out put",
+      "Ġsem igroup",
+      "Ġbott len",
+      "ĠGib bs",
+      "d ark",
+      "Ġrheumat oid",
+      "ur ring",
+      "mat ched",
+      "Ġsophistic ated",
+      "Ġcust omer",
+      "tetra hydro",
+      "X Y",
+      "b ug",
+      "Ġmor ning",
+      "ĠC VD",
+      "Ġm appings",
+      "ĠM SCs",
+      "ĠD H",
+      "Ġqu atern",
+      "he alth",
+      "Ä ±",
+      "Ġtem p",
+      "ĠJ ew",
+      "ĠI l",
+      "Ġvor tices",
+      "Ġser ine",
+      "ĠOx ygen",
+      "w eg",
+      "Ġexplan ations",
+      "P G",
+      "Ġc iti",
+      "Ġloc ality",
+      "== =",
+      "ĠTh om",
+      "Ġd airy",
+      "Bl ock",
+      "or dial",
+      "ak ov",
+      "Ġgli oma",
+      "Ġtrans action",
+      "Ġincre mental",
+      "anc he",
+      "R et",
+      "m agnetic",
+      "pyr rol",
+      "ĠP ic",
+      "Ġamel ior",
+      "oxid ant",
+      "rov iral",
+      "or atory",
+      "Ġs av",
+      "ĠSt ream",
+      "Ġsuper f",
+      "ĠIC U",
+      "Ġevid enced",
+      "Ġrepeated ly",
+      "Ġr ated",
+      "ĠP it",
+      "FA ULT",
+      "Ġh at",
+      "ĠCont ent",
+      "Ġiso form",
+      "V ER",
+      "Ġn odal",
+      "Ġschedul ed",
+      "Ġshould er",
+      "Ġt ap",
+      "Ġpor tal",
+      "Ġtra ps",
+      "ae v",
+      "ĠS OD",
+      "em atic",
+      "Ġen j",
+      "Ġretic ulum",
+      "ĠMin ister",
+      "ĠS el",
+      "Ġfall ing",
+      "ro st",
+      "N G",
+      "f d",
+      "n itro",
+      "ĠM ove",
+      "rel ativistic",
+      "eng es",
+      "ĠS ST",
+      "ĠIn v",
+      "Ġfin ish",
+      "ĠPol and",
+      "os econd",
+      "ĠB AL",
+      "oarth ritis",
+      "Ġop tics",
+      "ĠS ky",
+      "Ġadv oc",
+      "Ġhemorrh age",
+      "Ġmod ulating",
+      "n is",
+      "Ġmach inery",
+      "Ġupd ating",
+      "Ġcharacter izing",
+      "ish man",
+      "Ġtem plates",
+      "ĠLa place",
+      "ĠEn s",
+      "Rec ently",
+      "or us",
+      "ar ts",
+      "diff usion",
+      "ĠLevel s",
+      "ag a",
+      "ĠIn j",
+      "ĠL ayer",
+      "Ġrem n",
+      "Ġelastic ity",
+      "Ġmere ly",
+      "Ġf ission",
+      "eng ue",
+      "m ake",
+      "Ġmon op",
+      "Ġure a",
+      "ĠSim on",
+      "mi R",
+      "ĠSecond ly",
+      "ur ic",
+      "ĠVari able",
+      "il is",
+      "Ġmultiplic ative",
+      "ĠNo ise",
+      "Ġswit ched",
+      "Ġnic ot",
+      "Ġeffici encies",
+      "he ma",
+      "Ġapp ointed",
+      "gu ided",
+      "Ġwin ning",
+      "ĠMechan ics",
+      "Ġne o",
+      "ĠBR CA",
+      "ud i",
+      "Ġcontain er",
+      "sh op",
+      "Ġsugges tions",
+      "K B",
+      "Ġsubstit ute",
+      "O x",
+      "V C",
+      "Ġst one",
+      "ann a",
+      "ĠDep ression",
+      "Ġcont emporary",
+      "Ġoutl iers",
+      "qu et",
+      "ĠZ heng",
+      "Ġoc cl",
+      "Ġal veolar",
+      "exp ressing",
+      "Ġcom fort",
+      "Ġign ore",
+      "Am ong",
+      "ĠKle in",
+      "Ġrhyth m",
+      "Ġimm ers",
+      "Ġfa ith",
+      "bl ing",
+      "Ġaug mentation",
+      "ĠPre vention",
+      "Ġhe par",
+      "Ġnot ations",
+      "Ġhemat opoietic",
+      "perf ect",
+      "Ġsh ares",
+      "not in",
+      "Ġpict ures",
+      "ĠAcknowledg ments",
+      "Ġt ick",
+      "Ġun related",
+      "ĠTo ol",
+      "Ġm as",
+      "os ocial",
+      "g est",
+      "us hed",
+      "Ġphosphor ylated",
+      "Ġcer amic",
+      "c ool",
+      "or ylation",
+      "Ġdef icient",
+      "Ġrelax ed",
+      "ĠAnal yses",
+      "ec raft",
+      "Ġret ina",
+      "ĠIn ternal",
+      "Ġsp ite",
+      "Ġrecip ients",
+      "Ġsh ut",
+      "Ġeth ylene",
+      "ĠG ulf",
+      "Ġun affected",
+      "ĠRes ource",
+      "ĠN et",
+      "Ġperp et",
+      "Ġsl ab",
+      "re port",
+      "Ġμm ol",
+      "Ġid x",
+      "Ġsk ill",
+      "ĠInd uction",
+      "Ġmalign ancy",
+      "Ġc v",
+      "Ġdiff ering",
+      "Ġappropri ately",
+      "ij ing",
+      "Ġwar rant",
+      "r ally",
+      "Ġal gae",
+      "we ights",
+      "c asts",
+      "Ġoc ular",
+      "rac ycl",
+      "Ġdomin ates",
+      "Ġle uc",
+      "W here",
+      "ph on",
+      "Ġsocio economic",
+      "itzer land",
+      "Ġresil ience",
+      "Ġneighbour hood",
+      "Ġt one",
+      "psy ch",
+      "ĠOrgan ic",
+      "Ġg ather",
+      "Ġfalc iparum",
+      "Ġengine ered",
+      "ĠAv ail",
+      "inter ing",
+      "Ġclim atic",
+      "ĠEvolution ary",
+      "N MR",
+      "Ġre v",
+      "cent ral",
+      "ĠS in",
+      "Ġdecl ined",
+      "op ausal",
+      "Ġal arm",
+      "Right arrow",
+      "se x",
+      "Ġenerge tic",
+      "ï Ĥ",
+      "Ġdisc s",
+      "Ġol factory",
+      "ur ipot",
+      "spect rum",
+      "sp ot",
+      "Ġhem oglobin",
+      "M ark",
+      "c ov",
+      "ar boxyl",
+      "Ġindic ations",
+      "Ġsal mon",
+      "Ġsearc hed",
+      "Ġend ed",
+      "rolog ic",
+      "r floor",
+      "Ġau tism",
+      "Ġs elen",
+      "ĠH ung",
+      "ĠInf erence",
+      "Ġmamm ary",
+      "l floor",
+      "Ġser oton",
+      "Ġfund ed",
+      "ĠVi et",
+      "Ġri vers",
+      "ĠRe infor",
+      "ur g",
+      "Ġalb icans",
+      "ĠTherm o",
+      "ERR OR",
+      "Ġmut ually",
+      "Ġir r",
+      "ĠR at",
+      "Ġim g",
+      "Ġlymph ocyte",
+      "ĠRef s",
+      "ĠS parse",
+      "hold ers",
+      "F ree",
+      "RE D",
+      "ĠG auss",
+      "Ġcirc adian",
+      "ĠJ in",
+      "Ġconstit utes",
+      "Ġw ors",
+      "Ġfeature d",
+      "oc ent",
+      "le te",
+      "Ġont ology",
+      "Ġbil ayer",
+      "ĠCam bridge",
+      "Ġencryp tion",
+      "rot ron",
+      "et ti",
+      "ĠA er",
+      "Ġcou ples",
+      "ra il",
+      "Ġtw ist",
+      "Ġrid ge",
+      "G AN",
+      "id ers",
+      "SH IFT",
+      "Ġdiff us",
+      "Ġme ant",
+      "ĠSch warz",
+      "S b",
+      "Ġarc s",
+      "No tice",
+      "i y",
+      "Ġem erge",
+      "kw args",
+      "E ff",
+      "E nt",
+      "ion ization",
+      "ch oline",
+      "ust ries",
+      "ac her",
+      "s pl",
+      "pop ulation",
+      "f ol",
+      "Ġquestionnai res",
+      "Ġall ergic",
+      "w ich",
+      "ĠV acc",
+      "Ġat tained",
+      "ĠAn imals",
+      "am ics",
+      "ĠReg arding",
+      "ĠSem i",
+      "Ġgl ac",
+      "ĠEff icacy",
+      "Ġsynerg istic",
+      "IS H",
+      "Ġmaintain s",
+      "Ġsong s",
+      "ĠNeg ative",
+      "am oto",
+      "ĠMod ified",
+      "Ġsepar able",
+      "Ġbin aries",
+      "Ġaccess ibility",
+      "I ter",
+      "d in",
+      "ĠB inary",
+      "equ ilibrium",
+      "Ġc ue",
+      "m agn",
+      "Ġed ema",
+      "ï ¿½",
+      "Ġposition ed",
+      "Ġcharg ing",
+      "Ġun ivariate",
+      "he p",
+      "Ġcl ade",
+      "Ġcy steine",
+      "rac le",
+      "Ġresc ue",
+      "h abit",
+      "ĠDISCU SSION",
+      "Ġdepic ts",
+      "p ole",
+      "Ġst enosis",
+      "Ġv eter",
+      "pr inger",
+      "ĠP ow",
+      "Ġcovari ant",
+      "Ġmod ifying",
+      "Al gorithm",
+      "aver aged",
+      "al o",
+      "res on",
+      "Ġcharacter ised",
+      "Ġn i",
+      "Ġseem ed",
+      "ĠR om",
+      "sh ort",
+      "N V",
+      "Ġfer tility",
+      "ĠM emb",
+      "Ġl ying",
+      "Ġinstit ution",
+      "im ages",
+      "ĠB orel",
+      "fs ys",
+      "c ataly",
+      "Ġsepar ating",
+      "b iotic",
+      "m el",
+      "pg fsys",
+      "ĠJack son",
+      "Ġb ag",
+      "og rap",
+      "prop yl",
+      "ĠProgram ming",
+      "oc ratic",
+      "Ġp ion",
+      "ĠG radient",
+      "Ġsp he",
+      "Ġin line",
+      "Ġdom inate",
+      "Ġsuff ered",
+      "ĠDise ases",
+      "igen ous",
+      "w ill",
+      "Ġam in",
+      "adher in",
+      "ĠT ro",
+      "adj usted",
+      "E W",
+      "Ġde but",
+      "ne a",
+      "ĠD un",
+      "Ġd ictionary",
+      "oper atively",
+      "K A",
+      "be it",
+      "Ġperson nel",
+      "ĠÅ ½",
+      "re view",
+      "int o",
+      "ĠTok yo",
+      "Ġt rop",
+      "Ġvent ric",
+      "ĠMETHOD S",
+      "Ġim plication",
+      "ak is",
+      "ĠC MB",
+      "Ġtransmit ter",
+      "o ichi",
+      "ĠNiger ia",
+      "ĠK on",
+      "Ġbe ar",
+      "ĠK an",
+      "ĠPl ot",
+      "ĠS PSS",
+      "ĠBi ology",
+      "Ġbary on",
+      "Ġmicro RNA",
+      "Ġreproduc ibility",
+      "Ġlact ate",
+      "Ġpolyp hen",
+      "ĠM t",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ",
+      "end it",
+      "Ġhydro thermal",
+      "Ġwe alth",
+      "Ġhad ron",
+      "Ġwhere by",
+      "ell um",
+      "ĠDiff usion",
+      "ĠOrig in",
+      "Ġnonlinear ity",
+      "Ġinform ative",
+      "Ġvis ited",
+      "Ġvirt ually",
+      "ĠT un",
+      "Ġres et",
+      "ĠElect rical",
+      "ĠG lu",
+      "ĠS AM",
+      "ĠI sing",
+      "ĠSt ra",
+      "ond er",
+      "Ġd ies",
+      "Ġrecipro cal",
+      "C heck",
+      "ĠGu idelines",
+      "hest er",
+      "Ġproblem atic",
+      "ĠAt omic",
+      "Ġconcentr ate",
+      "st eps",
+      "j son",
+      "Recomm ended",
+      "ĠScreen ing",
+      "Ġna ive",
+      "Ġpractition ers",
+      "Ġfast ing",
+      "Ġmechan istic",
+      "op tions",
+      "P tr",
+      "IT E",
+      "W ork",
+      "âĢ ĺ",
+      "raf ts",
+      "Ġun w",
+      "Ġannih ilation",
+      "ob jective",
+      "ĠD ynamical",
+      "ad ec",
+      "ĠL ith",
+      "Ġextract ing",
+      "Ġcor al",
+      "ĠSt able",
+      "Ġbackground s",
+      "omorphism s",
+      "ĠâĪ «",
+      "Ġgre w",
+      "In st",
+      "g els",
+      "Ġin hal",
+      "d am",
+      "he im",
+      "benz yl",
+      "Ġpel vic",
+      "Ġdi arr",
+      "Ġdi ode",
+      "Ġem pir",
+      "ĠAl f",
+      "ĠUn certain",
+      "ĠH Cl",
+      "Ġjoint ly",
+      "Ġde par",
+      "Ġmerg ing",
+      "Ġch i",
+      "ap t",
+      "Ġpl t",
+      "Ġid i",
+      "Ġper for",
+      "stit uting",
+      "p age",
+      "ar é",
+      "ind ices",
+      "put ation",
+      "diff erent",
+      "b urn",
+      "Ġsurround ed",
+      "ĠT L",
+      "unt ary",
+      "st rip",
+      "l an",
+      "Ġc ow",
+      "ĠS ab",
+      "ĠGa As",
+      "p f",
+      "Ġes ophageal",
+      "ĠAl t",
+      "Ġhospital ization",
+      "ĠApproxim ation",
+      "Organ ism",
+      "ĠF air",
+      "Ġtrac ing",
+      "Ġpref erentially",
+      "Ġlower ing",
+      "uli ar",
+      "ĠDer iv",
+      "Ġphyto plankton",
+      "omy c",
+      "T hat",
+      "ĠIsra el",
+      "Ġminim ized",
+      "Ġany thing",
+      "r ule",
+      "p ow",
+      "Ġfam ous",
+      "ĠAcc uracy",
+      "Ġphotoc atalytic",
+      "ĠNon etheless",
+      "Ġdivis or",
+      "v b",
+      "Ġcam eras",
+      "ĠW ales",
+      "ĠCont ributions",
+      "Ġdisplac ements",
+      "ĠT am",
+      "Ġvol umetric",
+      "ession al",
+      "Ġcompens ate",
+      "Ġa ce",
+      "tri angle",
+      "bu ff",
+      "Ġnames pace",
+      "Ġbound ing",
+      "ynchron ous",
+      "m d",
+      "Ġimag ery",
+      "it ated",
+      "Ġorigin ated",
+      "ĠBel g",
+      "ĠE CG",
+      "ex isting",
+      "ĠSt okes",
+      "sens itivity",
+      "tid ine",
+      "ĠW M",
+      "Ġmonot one",
+      "Ġproceed s",
+      "ĠClust ering",
+      "ĠIo T",
+      "ern ary",
+      "al amic",
+      "ĠCollabor ation",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠĠĠĠĠ",
+      "OL D",
+      "Î ĺ",
+      "ĠNan opar",
+      "ĠMul tiv",
+      "Ġc ystic",
+      "pi re",
+      "Ġoper ates",
+      "Ġmedi ating",
+      "Ġbene ath",
+      "ob e",
+      "g ate",
+      "Ġo ocytes",
+      "Ġmarg ins",
+      "ymmet ries",
+      "Ġrelig ious",
+      "ĠN it",
+      "Ġcut aneous",
+      "AN S",
+      "Ġdevelop s",
+      "as ia",
+      "ĠRo berts",
+      "a vier",
+      "Ġsim plic",
+      "Ġreveal ing",
+      "UN D",
+      "Ġte a",
+      "Ġl ysis",
+      "Ġaggreg ated",
+      "ĠR GB",
+      "Ġcor ro",
+      "Ġb ir",
+      "in ae",
+      "v d",
+      "Ġcour t",
+      "Ġcontrovers ial",
+      "Ġto w",
+      "Ġhy steresis",
+      "en berg",
+      "Ġent ers",
+      "p ng",
+      "ĠF lex",
+      "Assum e",
+      "ĠB ad",
+      "ĠSimilar ities",
+      "Ex perim",
+      "AT H",
+      "Ġ ut",
+      "ter ms",
+      "ĠM ol",
+      "Ġvis ually",
+      "Ġadop tion",
+      "Ġprint ing",
+      "Ġequ iv",
+      "ĠP ert",
+      "Ġper col",
+      "Ġsome one",
+      "abul ary",
+      "Ġle ver",
+      "ĠH aus",
+      "ic illin",
+      "it ar",
+      "Ġto urn",
+      "Al tern",
+      "Ex p",
+      "~~ ~~",
+      "ĠF o",
+      "Ġab ol",
+      "med ian",
+      "Ġroll ing",
+      "h m",
+      "Ġtel escope",
+      "ĠC av",
+      "Ġseed lings",
+      "in hib",
+      "Ġd in",
+      "Ġimp urities",
+      "Ġampl ifier",
+      "ĠK er",
+      "Ġdimin ished",
+      "P B",
+      "f ib",
+      "ro ck",
+      "ĠB in",
+      "Ġphotos ynthetic",
+      "ĠCr ypt",
+      "Ġpre term",
+      "Ġh its",
+      "Ġfract al",
+      "Ġdisc arded",
+      "Ġend ocrine",
+      "os hi",
+      "Ġmod ulo",
+      "w t",
+      "Ġqu enching",
+      "Ġsound s",
+      "ĠED TA",
+      "re active",
+      "Ġres ist",
+      "ang hai",
+      "Ġn arr",
+      "Ġiniti ate",
+      "ĠS aint",
+      "X R",
+      "Ge V",
+      "ĠInd ependent",
+      "Ġinj ective",
+      "up us",
+      "Ġl inguistic",
+      "Ġanalog ues",
+      "Ġdiss ection",
+      "Ġlas ers",
+      "di ab",
+      "ĠTe le",
+      "Ġc racks",
+      "Ġb rane",
+      "V O",
+      "ĠExt ended",
+      "Ġt ells",
+      "Ġremark s",
+      "ul ting",
+      "ĠB urn",
+      "d L",
+      "ress ible",
+      "ĠCh ap",
+      "Ġs q",
+      "Ġrepro duced",
+      "ĠB cl",
+      "Ġsw arm",
+      "opath ology",
+      "ch rotron",
+      "Ġm ine",
+      "Ġhad ronic",
+      "ĠLocal ization",
+      "ĠM otor",
+      "Ġvisual ize",
+      "Ġc ats",
+      "Ġbal ancing",
+      "ĠSc hed",
+      "Co A",
+      "Ġtherm odynamics",
+      "ĠDiagn ostic",
+      "Ġreli ef",
+      "Ġpos itivity",
+      "Ġh ub",
+      "ĠInf rared",
+      "S ur",
+      "om ed",
+      "Ġop tically",
+      "Ġv ascul",
+      "is ations",
+      "enc oder",
+      "Ġcopol ymer",
+      "Ġrest ore",
+      "Ġiner tia",
+      "ubic in",
+      "Ġeti ology",
+      "ĠSec ret",
+      "ĠC W",
+      "Con st",
+      "ĠBr it",
+      "ĠConst ant",
+      "ĠD IS",
+      "Ġdisc ipl",
+      "b ra",
+      "ĠO ral",
+      "ĠU L",
+      "Ġdel ine",
+      "Ġnucle on",
+      "Ġemploy ment",
+      "ĠR D",
+      "q q",
+      "ĠCarol ina",
+      "ĠG ab",
+      "Ġasser tion",
+      "CM C",
+      "r gb",
+      "F rame",
+      "ĠJ ust",
+      "Ġinoc ulation",
+      "cl uding",
+      "Ġoscill atory",
+      "Ġcanc el",
+      "ĠPo inc",
+      "por a",
+      "ĠJ ul",
+      "ru vate",
+      "Ġpoli tic",
+      "ur us",
+      "ĠAdv ances",
+      "ĠR oot",
+      "th ood",
+      "oxygen ase",
+      "ms g",
+      "Ġk V",
+      "Ġad mit",
+      "Ġrefrac tory",
+      "Ġclon ing",
+      "Ġf atal",
+      "plant ation",
+      "ĠG ir",
+      "Ġt es",
+      "ĠR ho",
+      "oh n",
+      "Ġinnov ation",
+      "Ġs ending",
+      "Ġc able",
+      "Ġnic he",
+      "Ġres erve",
+      "Ġat rophy",
+      "ath an",
+      "ĠÃ ij",
+      "iti zation",
+      "Ġf an",
+      "Ġb ubbles",
+      "ĠTheorem s",
+      "ĠSw itzerland",
+      "ĠHe isenberg",
+      "ĠRed uced",
+      "R a",
+      "Z r",
+      "ĠPoss ible",
+      "U psilon",
+      "ĠAg ric",
+      "el lect",
+      "nd s",
+      "math ds",
+      "at re",
+      "Ġfor aging",
+      "Ġup ward",
+      "id ene",
+      "Ġgl ands",
+      "f ed",
+      "uccess ful",
+      "ĠW olf",
+      "Ġuseful ness",
+      "op orous",
+      "Ġp unct",
+      "ard o",
+      "Ġsy stolic",
+      "ĠTarget ing",
+      "Ġill umin",
+      "Ġpig ment",
+      "Ġsim ulating",
+      "Ġpor tions",
+      "ĠPrinc iples",
+      "ĠHop f",
+      "l ipid",
+      "ĠL U",
+      "ub ation",
+      "ĠAr tificial",
+      "Ġpr ison",
+      "an ing",
+      "ĠG N",
+      "ĠStrateg ies",
+      "ĠP as",
+      "T a",
+      "ĠProb ability",
+      "or um",
+      "Ġs keleton",
+      "Ġcomp artments",
+      "R ead",
+      "Ġco ach",
+      "Ġmod ality",
+      "ĠReg ister",
+      "Ġj e",
+      "Ġhe ights",
+      "in yl",
+      "Ġsub spaces",
+      "ti p",
+      "Ġá ¸",
+      "ĠG I",
+      "Ch ar",
+      "ro genic",
+      "ret t",
+      "eu tics",
+      "Ġadhes ive",
+      "ĠP ier",
+      "Le ft",
+      "id ental",
+      "NA c",
+      "Ġconjug ation",
+      "or ov",
+      "id ge",
+      "im aging",
+      "ĠT W",
+      "Ġpres ident",
+      "ĠO ste",
+      "ass emb",
+      "Ġinter net",
+      "Ġde als",
+      "ĠG AP",
+      "Ġform ulate",
+      "ĠUp date",
+      "ĠRNA i",
+      "cl ero",
+      "Ġpermut ations",
+      "Ġisotop es",
+      "op ic",
+      "ĠQ U",
+      "rom es",
+      "ĠPol icy",
+      "ĠC reek",
+      "ĠWind ows",
+      "Ġm erge",
+      "Ġacc ident",
+      "Ġsuper position",
+      "Ġdeb ate",
+      "Ġdocument ation",
+      "Ġeigen vectors",
+      "s or",
+      "ĠPh oto",
+      "Ġdepos it",
+      "Ġgerm ination",
+      "Ġsub graph",
+      "ĠRec ords",
+      "Ġchem ically",
+      "ĠPredic ting",
+      "ĠK y",
+      "se lective",
+      "yn man",
+      "dis pers",
+      "Ġlum bar",
+      "Ġmus ical",
+      "in ates",
+      "Ġinher ited",
+      "j u",
+      "Ġtrac er",
+      "Ġend ing",
+      "Ġeng aged",
+      "hand ed",
+      "Ġproduc er",
+      "Ġent angled",
+      "ĠD elta",
+      "Ġpiec ewise",
+      "NA ME",
+      "st op",
+      "Ġmut ated",
+      "Ġre cess",
+      "Ġimmun o",
+      "c ancer",
+      "ĠAk t",
+      "it ers",
+      "ĠB MP",
+      "Ġcompan ion",
+      "Ġcommun icate",
+      "Ġh ollow",
+      "Ġp ad",
+      "Ġs ph",
+      "om od",
+      "Ġpar ton",
+      "Ġspontaneous ly",
+      "e ared",
+      "Ġrot ations",
+      "Ġcosm ology",
+      "Ġmore over",
+      "pr inc",
+      "Ġevery where",
+      "b rane",
+      "l ational",
+      "em e",
+      "Ġbeh ave",
+      "um en",
+      "ost on",
+      "ov es",
+      "Ġg ar",
+      "Ġad renal",
+      "ĠEstim ating",
+      "N b",
+      "Ġech ocardi",
+      "Ġemphas ized",
+      "Ġeng ines",
+      "Ġbrack ets",
+      "Ġlead ers",
+      "Ġdistinc tive",
+      "ĠL ymph",
+      "Ġex ert",
+      "Ġinnov ative",
+      "c oupling",
+      "ĠSign ific",
+      "she et",
+      "ĠC over",
+      "ĠC CD",
+      "ĠF all",
+      "stim ulated",
+      "Ġsuper oxide",
+      "Ġpollut ants",
+      "Ġby tes",
+      "ĠL ipid",
+      "Ġtra fficking",
+      "Ġlead ership",
+      "inform atics",
+      "Ġbiod iversity",
+      "ad or",
+      "Ġinter conn",
+      "Ġharmon ics",
+      "Ġseaw ater",
+      "ĠIll umina",
+      "necess ary",
+      "ĠAnt on",
+      "Ġprocess ors",
+      "typ ename",
+      "D et",
+      "pro ton",
+      "Ġsubt raction",
+      "Ġshif ting",
+      "Ġcust omers",
+      "K e",
+      "ĠO B",
+      "aton in",
+      "at ellite",
+      "ĠS US",
+      "ĠCol on",
+      "ĠTim es",
+      "T V",
+      "ĠM ink",
+      "ĠIntegr ation",
+      "Ġprof ound",
+      "IT C",
+      "Ġg ras",
+      "ĠNA SA",
+      "ĠAC K",
+      "radi ol",
+      "ĠM ale",
+      "ĠWork ing",
+      "tic ity",
+      "ilibri a",
+      "bound ary",
+      "ĠR I",
+      "ĠAl i",
+      "car di",
+      "ĠF GF",
+      "b ranes",
+      "Ġbe et",
+      "Ġmiss ed",
+      "S ource",
+      "ĠB ot",
+      "ie ve",
+      "Ġis other",
+      "ne ys",
+      "n l",
+      "or tion",
+      "Ġcool ed",
+      "M V",
+      "Ġo mit",
+      "Ġver bal",
+      "aret te",
+      "Ġconf erence",
+      "Ġtransform er",
+      "Ġre jected",
+      "Ġprogress ively",
+      "ĠTur key",
+      "Ġath letes",
+      "Ġan atomy",
+      "E Q",
+      "Ġdeterior ation",
+      "ĠDi etary",
+      "Ġcor n",
+      "Ġcaps ule",
+      "Ġvibr ations",
+      "Ġoccup ational",
+      "Ġex osomes",
+      "Ġre written",
+      "Ġlign in",
+      "Ġbi opsies",
+      "ĠAdvers arial",
+      "Ġm ercury",
+      "Ġpl atinum",
+      "Ġirre levant",
+      "Ġker atin",
+      "ĠE mission",
+      "Ġeukary otic",
+      "Ġinte g",
+      "Ġkn ot",
+      "Ġser a",
+      "Ġcav ities",
+      "ĠMed i",
+      "Ind eed",
+      "E u",
+      "Ġâ Ł",
+      "Ġsc enes",
+      "Ġlap aroscopic",
+      "Ġsen ior",
+      "ĠD istance",
+      "pred ic",
+      "Ġear liest",
+      "Ġor g",
+      "ĠTh or",
+      "b ury",
+      "obl asts",
+      "Ġp umping",
+      "target ed",
+      "Ġra p",
+      "ĠP il",
+      "Î ł",
+      "Ġneu rom",
+      "o ft",
+      "ost at",
+      "Ġp adding",
+      "Ġconflic ts",
+      "Ġst ems",
+      "ĠSac charomyces",
+      "eng ine",
+      "Ġalk yl",
+      "Ġt ill",
+      "ĠQu ad",
+      "g ood",
+      "ro x",
+      "ĠF uzzy",
+      "Ġrob otic",
+      "ĠDen ote",
+      "ĠN IR",
+      "ĠY uk",
+      "paren cy",
+      "Ġle gs",
+      "yl van",
+      "Ġtight ly",
+      "Ġdec or",
+      "ĠV P",
+      "ĠM un",
+      "at oms",
+      "ĠSil ver",
+      "Ġneurode generative",
+      "Ġrespond ed",
+      "Ġrec ons",
+      "G EN",
+      "ĠF ine",
+      "f c",
+      "Ġpar agraph",
+      "Ġint ens",
+      "Ġalong side",
+      "Ġb rand",
+      "mon ium",
+      "Ġp m",
+      "Ġsimple x",
+      "ĠPrelim inary",
+      "Ġdown regulation",
+      "Ġx y",
+      "ĠM ak",
+      "op ter",
+      "ush ing",
+      "ĠB og",
+      "ox ia",
+      "================ ================",
+      "com mon",
+      "ĠA SS",
+      "ĠHD L",
+      "alam us",
+      "Ġirrig ation",
+      "N M",
+      "Ġf ading",
+      "Ġprev entive",
+      "Ġreli ably",
+      "ĠEthiop ia",
+      "ot hesis",
+      "iz ability",
+      "O B",
+      "Ġtrig lycer",
+      "Ġgest ational",
+      "Ġb esides",
+      "ĠI ii",
+      "ĠZ one",
+      "Ġcop ing",
+      "Ġminor ity",
+      "Ġdepri vation",
+      "Ġhex agonal",
+      "chlor ophenyl",
+      "ĠóµĦ ¨",
+      "Ġg yr",
+      "Ġview ing",
+      "New ton",
+      "ĠHierarch ical",
+      "o L",
+      "ec es",
+      "Ġconcl udes",
+      "Ġfung us",
+      "Ġpyl ori",
+      "Ġobstac les",
+      "th iazol",
+      "conjug ated",
+      "r ass",
+      "Ġl ose",
+      "Ġfor th",
+      "ĠAll en",
+      "opl ast",
+      "ĠProt ection",
+      "Ġintermitt ent",
+      "Ġluc iferase",
+      "ĠM K",
+      "Ġga ug",
+      "ĠF an",
+      "Ġmod al",
+      "ĠEx ercise",
+      "sc attering",
+      "ĠSh im",
+      "Ġexc retion",
+      "Ġat ypical",
+      "Ġmalign ancies",
+      "angl ades",
+      "ĠSpect roscopy",
+      "Ġaden osine",
+      "l if",
+      "Ġnucle ic",
+      "Ġincl ination",
+      "ĠC ass",
+      "Ġeth n",
+      "Ġex empl",
+      "ĠD y",
+      "Ġl ambda",
+      "Ġj ac",
+      "ĠP RE",
+      "Ġrail way",
+      "Ġf le",
+      "Ġreflec tions",
+      "Ġnano structures",
+      "ti sts",
+      "pr ints",
+      "ĠC AT",
+      "Ġs ib",
+      "Ġchlor o",
+      "Ġrecip ient",
+      "op tic",
+      "Ġcoun ty",
+      "Ġnucle otides",
+      "Ġz ircon",
+      "Ġhors es",
+      "ĠM ental",
+      "in line",
+      "ĠNor way",
+      "The y",
+      "Ġmusc ular",
+      "ace tic",
+      "ĠJ u",
+      "Ġcommun ic",
+      "f iles",
+      "f illed",
+      "H B",
+      "Ġreg ulations",
+      "Ġaccum ulate",
+      "ĠPan el",
+      "C y",
+      "ö l",
+      "ĠPak istan",
+      "Ġthor acic",
+      "ĠM PI",
+      "por tion",
+      "Ġinduc tive",
+      "ĠCong ress",
+      "Ġfibrobl ast",
+      "cl ust",
+      "Ġcent res",
+      "ad el",
+      "Ġsubstit utions",
+      "Ġtrunc ation",
+      "r ification",
+      "ok a",
+      "F low",
+      "ĠRed uc",
+      "polar ized",
+      "ib ular",
+      "P e",
+      "ĠA ML",
+      "ĠAg ency",
+      "Ġt ilt",
+      "ubl ished",
+      "Ġdep olar",
+      "Ġbel t",
+      "Ġoptim izer",
+      "EL L",
+      "ĠHand book",
+      "ĠVirgin ia",
+      "s ense",
+      "ĠD ur",
+      "Ġpiezo electric",
+      "Ġaward ed",
+      "ail ing",
+      "P os",
+      "p ref",
+      "ĠSum mer",
+      "ed o",
+      "ĠI de",
+      "ĠB SA",
+      "Ġmon omers",
+      "Ġco agulation",
+      "Ġg am",
+      "Ġhom es",
+      "Ġhead s",
+      "adm ium",
+      "ĠO C",
+      "Ġoccup ancy",
+      "ĠEm pirical",
+      "ĠI i",
+      "Ġch ir",
+      "Ġdegener acy",
+      "Ġflow ers",
+      "Ġsuperconduc tivity",
+      "Ġin versely",
+      "op tical",
+      "w ere",
+      "ĠAs ymptotic",
+      "S ec",
+      "tit le",
+      "pos al",
+      "ĠPro gn",
+      "Ġpos es",
+      "ĠB orn",
+      "Ġcontinu ation",
+      "Ġcul tivated",
+      "enti ment",
+      "Ġman aging",
+      "Ġthromb osis",
+      "a ug",
+      "CN T",
+      "ure a",
+      "Ġsp ind",
+      "ĠWhere as",
+      "ĠPers on",
+      "Ġb ipartite",
+      "Ġres cal",
+      "Ġmark ets",
+      "ph an",
+      "per ties",
+      "Ġferm ionic",
+      "Ġmunic ip",
+      "Ġachie vable",
+      "t ab",
+      "Å į",
+      "ĠRel ation",
+      "T otal",
+      "x ia",
+      "Ġintellig ent",
+      "ĠU T",
+      "ĠD al",
+      "Ġmedic inal",
+      "Ġinadequ ate",
+      "i ently",
+      "ers en",
+      "Ġpre condition",
+      "Ġmethod ological",
+      "Ġcan opy",
+      "Ġbacter ium",
+      "col umn",
+      "C al",
+      "ĠDi ego",
+      "ĠS ak",
+      "ĠComprehens ive",
+      "Ġanti tumor",
+      "Ġflow er",
+      "ĠK han",
+      "Ġmet adata",
+      "Ġphot ore",
+      "ogen icity",
+      "Ġle ague",
+      "ol ating",
+      "Ġprom ise",
+      "ĠP ere",
+      "Ġper mits",
+      "Ġthread s",
+      "ĠD Cs",
+      "ĠCh am",
+      "raz ol",
+      "B ank",
+      "Ġwithdraw al",
+      "Ġapp end",
+      "ot helial",
+      "ĠMeas ures",
+      "Ġguid eline",
+      "Ġmitig ate",
+      "adj oint",
+      "Ġbrack et",
+      "P ad",
+      "M ills",
+      "Bu ffer",
+      "Ġc ass",
+      "h oc",
+      "manif olds",
+      "her ry",
+      "Ġfacilit ated",
+      "E vent",
+      "Ġ È",
+      "ĠC ruz",
+      "ĠB rand",
+      "Ġnecess ity",
+      "burg h",
+      "Ġme V",
+      "Ġc AMP",
+      "O ff",
+      "se lected",
+      "Ġeng age",
+      "Ġredund ancy",
+      "Ġnanocom posites",
+      "s olution",
+      "ons et",
+      "ĠEx posure",
+      "Ġrepe titive",
+      "Ã ł",
+      "ĠR AD",
+      "ĠTur k",
+      "Ġcorne al",
+      "Ġexplo iting",
+      "Ġob structive",
+      "gram ming",
+      "ĠM ED",
+      "Ġmat hem",
+      "Ġconduc tive",
+      "Ġphotos ynthesis",
+      "E instein",
+      "ĠP eng",
+      "M W",
+      "ĠSch midt",
+      "Ġrepe tition",
+      "identif ied",
+      "Ġinj ured",
+      "Ġdef ective",
+      "ĠP el",
+      "Ġcul tivation",
+      "Ġfirst ly",
+      "Ġanalyz er",
+      "Ġstain less",
+      "Ġjo ining",
+      "ĠOxid ative",
+      "Ġph age",
+      "Ġexp endit",
+      "Ġhom ogeneity",
+      "ip le",
+      "ov ic",
+      "Ġcross ed",
+      "ĠTr ust",
+      "ĠF ract",
+      "rophys iological",
+      "Ġbas ically",
+      "Ġco ales",
+      "Ġgra vit",
+      "ful ness",
+      "c ano",
+      "Ġcol itis",
+      "Ġcha os",
+      "carb ons",
+      "O nce",
+      "ĠTow ard",
+      "or f",
+      "top ic",
+      "ĠPl ay",
+      "ĠCor respond",
+      "ĠS leep",
+      "ticular ly",
+      "c umin",
+      "v dots",
+      "ĠR he",
+      "Ġult raf",
+      "Ġtimes cale",
+      "ĠDet ails",
+      "ang les",
+      "Ġsur rogate",
+      "ĠFlu id",
+      "c z",
+      "Ġinitial ization",
+      "ĠTel escope",
+      "r ases",
+      "ĠSt ock",
+      "ĠC ond",
+      "Ġimmun odeficiency",
+      "B el",
+      "os er",
+      "sh own",
+      "Ġk cal",
+      "Equ ation",
+      "prot ective",
+      "Ġcall ing",
+      "Ġanticip ated",
+      "Ġambig uity",
+      "ĠN ode",
+      "ĠG D",
+      "Ġin let",
+      "Ġbre ad",
+      "Ġexceed ed",
+      "Ġimmun ization",
+      "Ġpro hib",
+      "y tic",
+      "Ġbo ys",
+      "t u",
+      "Ġto wer",
+      "L ike",
+      "ĠAn omal",
+      "â Į",
+      "ĠSh ow",
+      "Ġim aged",
+      "Ġequ il",
+      "Ġrend ering",
+      "ob ility",
+      "Ġge ological",
+      "f riend",
+      "ö r",
+      "carbox amide",
+      "ovol ta",
+      "C urrent",
+      "ĠS ti",
+      "ĠM U",
+      "Ġval ued",
+      "Ġpo ison",
+      "Ġprac tically",
+      "Ġrequ ested",
+      "C ode",
+      "Ġbr ings",
+      "Ġdim ethyl",
+      "h yp",
+      "ce mic",
+      "V ol",
+      "qu anti",
+      "Ġex ha",
+      "Ġrespons ibility",
+      "ĠCont rolled",
+      "Ġf ur",
+      "Ġres emb",
+      "ĠK aw",
+      "Ġev oked",
+      "Ġuter ine",
+      "Ð »",
+      "Ġan onymous",
+      "ĠChall enges",
+      "Ġanch or",
+      "ĠAb d",
+      "D er",
+      "Ġtherm ally",
+      "ĠC AP",
+      "obl ot",
+      "ĠF ire",
+      "Ġdiagnos tics",
+      "Ġexec ute",
+      "al is",
+      "ron i",
+      "ĠHar ris",
+      "ĠGon z",
+      "Ġv ig",
+      "ĠProf essor",
+      "Ġinvent ory",
+      "int ensity",
+      "ĠNSC LC",
+      "Ġinterf ere",
+      "ysacchar ides",
+      "Ġreg ener",
+      "ĠAut hors",
+      "Ġtransl ate",
+      "ĠT ests",
+      "ĠL ove",
+      "ĠInd uced",
+      "enn is",
+      "ĠG EN",
+      "Ġolig onucle",
+      "Ġmet er",
+      "s atisf",
+      "hes ion",
+      "Ġtrans porters",
+      "B IT",
+      "ĠCon c",
+      "Ġgl auc",
+      "sc ores",
+      "Ġmerg er",
+      "G H",
+      "Ġst oichi",
+      "ĠX ia",
+      "eff ects",
+      "ĠExpl oring",
+      "dor ff",
+      "Ġcardinal ity",
+      "ĠK az",
+      "f alse",
+      "ĠH SP",
+      "Ġuns upervised",
+      "ingu ish",
+      "isc her",
+      "Ġrel ativity",
+      "on ormal",
+      "oot hed",
+      "ed ges",
+      "ĠI MP",
+      "Ġimp ulse",
+      "ĠColumb ia",
+      "Ġpartic ulate",
+      "ĠSupport ing",
+      "ĠSD SS",
+      "vol tage",
+      "ĠAma zon",
+      "Ġep oxy",
+      "C all",
+      "Big l",
+      "Ġme ets",
+      "Ġequ atorial",
+      "Ġneu ros",
+      "Ġper itoneal",
+      "des c",
+      "input s",
+      "Ġex terior",
+      "ac o",
+      "Ġme al",
+      "ĠDani el",
+      "Ġintu itive",
+      "Ġcoun s",
+      "dep ress",
+      "in is",
+      "ph ot",
+      "ĠA min",
+      "Ġreservoir s",
+      "ĠW hole",
+      "Ġca ud",
+      "Ġbos onic",
+      "Ġread ers",
+      "Ġcr im",
+      "Ġpathophys iology",
+      "arg o",
+      "the se",
+      "inc ome",
+      "Ġiss ued",
+      "Ġhepat ocytes",
+      "ĠC i",
+      "der iv",
+      "up ta",
+      "t uple",
+      "ĠCh an",
+      "Ġauthentic ation",
+      "yg d",
+      "Ġinf in",
+      "Ġaccel erate",
+      "ep tive",
+      "Ġhydro gel",
+      "ask a",
+      "ON E",
+      "Ġfed eral",
+      "ograph ics",
+      "Ġmu on",
+      "Ġsl ide",
+      "Ġellip tical",
+      "at ite",
+      "Ġc c",
+      "ET s",
+      "Ġclar ity",
+      "ocy cl",
+      "is al",
+      "rec tions",
+      "ay an",
+      "row eak",
+      "ĠS OC",
+      "od erm",
+      "t un",
+      "as m",
+      "ĠH ir",
+      "lik elihood",
+      "Ġad ul",
+      "t l",
+      "H igh",
+      "Ġal ters",
+      "plit ude",
+      "ĠRe lease",
+      "Ġharm ful",
+      "l ate",
+      "ound s",
+      "ĠFed eral",
+      "ĠEcon omic",
+      "Ġra bb",
+      "Ġaccommod ate",
+      "em ission",
+      "ĠB ah",
+      "c ox",
+      "ĠMod ulation",
+      "Ġconstruc tions",
+      "ign er",
+      "ĠUrb an",
+      "Ġw ake",
+      "Ġadvers ary",
+      "wik ipedia",
+      "Ġsu ite",
+      "w ick",
+      "exp ressed",
+      "ro d",
+      "K D",
+      "Ġcomput ers",
+      "ĠB anglades",
+      "Ġpers ist",
+      "Ġburn ing",
+      "Ġadministr ative",
+      "Ġpl ug",
+      "ĠRepresent ations",
+      "ĠSc attering",
+      "Ġendomet rial",
+      "Ġdescript ors",
+      "Ġcom mission",
+      "B ar",
+      "igh th",
+      "ĠMar sh",
+      "sam pling",
+      "Ġh ull",
+      "ic in",
+      "Pro b",
+      "Ġnur se",
+      "Ġsh am",
+      "ĠK err",
+      "Ġpref rontal",
+      "Ġfix ing",
+      "O K",
+      "Ġb old",
+      "Ġcor ollary",
+      "cf g",
+      "ĠOx ford",
+      "Ġbor on",
+      "R B",
+      "ĠC ab",
+      "Big r",
+      "ĠPred ict",
+      "Ġpec uliar",
+      "h idden",
+      "is a",
+      "id en",
+      "appro priate",
+      "or h",
+      "ellect ual",
+      "Ġseiz ures",
+      "ass er",
+      "til is",
+      "hand le",
+      "iax ial",
+      "s ym",
+      "Ġcarcin omas",
+      "se a",
+      "sp ired",
+      "Ġab rupt",
+      "t ests",
+      "Ġw elfare",
+      "ĠO il",
+      "ĠLo ad",
+      "FL AG",
+      "ut hal",
+      "Ġfac ing",
+      "Americ an",
+      "L AS",
+      "Ġir respective",
+      "Ġrout inely",
+      "w al",
+      "Ġsettle ment",
+      "ĠA qu",
+      "Ġelectron ics",
+      "Ġhand led",
+      "Ġbiological ly",
+      "sm ooth",
+      "ĠBel ongs",
+      "ti b",
+      "Ġtra v",
+      "p ressive",
+      "ourn als",
+      "Ð º",
+      "fil ename",
+      "Ġhel ical",
+      "Ġbacter i",
+      "Ġsat ellites",
+      "B H",
+      "ent ed",
+      "ĠFoot ball",
+      "Ġï£ ±",
+      "ĠH V",
+      "Ġtri p",
+      "ĠCK D",
+      "ran i",
+      "Ġclean ing",
+      "lim it",
+      "ĠT CP",
+      "Ġsc in",
+      "Ġsl udge",
+      "Ġsymbol ic",
+      "ĠSequ encing",
+      "ad al",
+      "ĠPhil ipp",
+      "IC S",
+      "Ġvag inal",
+      "Ġcommit ment",
+      "ĠA wards",
+      "tr ig",
+      "Ġgu itar",
+      "acet ate",
+      "Ġb et",
+      "Cl N",
+      "Ġagric ulture",
+      "Ġch ief",
+      "Ġem bol",
+      "bu ild",
+      "Ġtex ts",
+      "ĠCo oper",
+      "l ived",
+      "ĠDel ay",
+      "ĠM ode",
+      "y al",
+      "B N",
+      "Ġindex ed",
+      "ex pr",
+      "ER N",
+      "v ens",
+      "Ġpo inter",
+      "c v",
+      "ac on",
+      "t ance",
+      "ĠâĪ Ŀ",
+      "Ġlow ered",
+      "Ġmit otic",
+      "rh osis",
+      "ĠP age",
+      "ü r",
+      "im m",
+      "ĠThe rapeutic",
+      "Ġoste opor",
+      "Ġbil inear",
+      "ĠCath olic",
+      "ĠAltern ative",
+      "oxid ation",
+      "Ġiniti o",
+      "benz o",
+      "ĠA di",
+      "per son",
+      "per itoneal",
+      "ĉĉ Ġ",
+      "Ġatt raction",
+      "Ġdiarr hea",
+      "Ġre n",
+      "ĠI SO",
+      "im ir",
+      "Ġtermin ology",
+      "uk ey",
+      "Ġreson ator",
+      "Ġsubstit uting",
+      "Ġhar bor",
+      "pro vid",
+      "dec ay",
+      "ĠHD AC",
+      "ĠAnaly tical",
+      "Ġpost natal",
+      "Ġund es",
+      "Spec ific",
+      "d ichlor",
+      "AR I",
+      "t ot",
+      "Ġdig it",
+      "op ing",
+      "ĠZ inc",
+      "Ġle thal",
+      "Wh itney",
+      "F i",
+      "qu antum",
+      "ĠF ailure",
+      "Ġsol ves",
+      "ĠSp aces",
+      "ear man",
+      "Ġgo at",
+      "Ġsyn apses",
+      "Ġres uspended",
+      "Ġresid ent",
+      "Ġcomp ac",
+      "Ġcor tisol",
+      "Ġphot ometry",
+      "W P",
+      "se lect",
+      "Ġc ele",
+      "or ubicin",
+      "ĠMul tic",
+      "ĠJe an",
+      "Ġcl ip",
+      "Ġs a",
+      "oc o",
+      "ge ometric",
+      "Ġhel ic",
+      "Ġempir ically",
+      "Ġmicro fluid",
+      "id is",
+      "Ġaut ocor",
+      "W F",
+      "ĠRes pir",
+      "radi ation",
+      "Ġthem es",
+      "Ġt aste",
+      "ric ing",
+      "Ġexam inations",
+      "ĠSens ing",
+      "s ame",
+      "DE FAULT",
+      "Ġphyl ogeny",
+      "h ig",
+      "Ġplate lets",
+      "ĠHist or",
+      "ab a",
+      "Ġresid ential",
+      "Ġun bounded",
+      "and ing",
+      "hed ron",
+      "r ys",
+      "ĠC CR",
+      "Ġcon ce",
+      "Ġpar asitic",
+      "c b",
+      "ĠFe ynman",
+      "ĠKe pler",
+      "Ã ´",
+      "ĠG il",
+      "ĠMAT LAB",
+      "b en",
+      "sc ope",
+      "Ġdiscrim in",
+      "Ġjus tified",
+      "plas ma",
+      "ĠCho i",
+      "Ġro of",
+      "PC A",
+      "ĠT CR",
+      "Ġvox el",
+      "ĠW ard",
+      "Ġunc or",
+      "S tok",
+      "Ġsp ur",
+      "TR A",
+      "Ġdiagnos es",
+      "rophys ical",
+      "ategor ies",
+      "Ġove restim",
+      "Ġstream ing",
+      "ĠRec overy",
+      "Ġevery thing",
+      "LO W",
+      "G ener",
+      "Ġun biased",
+      "Ġvari ances",
+      "comp act",
+      "es pan",
+      "in j",
+      "Ġend oscopic",
+      "Ġide als",
+      "ĠR ice",
+      "ĠKa plan",
+      "Ġf ecal",
+      "fer red",
+      "ĠCy cle",
+      "Ġimplant ed",
+      "Ġw ine",
+      "P ET",
+      "Ġassign ments",
+      "Ġabs ol",
+      "X T",
+      "Ġswim ming",
+      "M N",
+      "ĠGe ometric",
+      "ĠHealth care",
+      "Ġpow ders",
+      "ĠG el",
+      "Ġdown ward",
+      "Ġexceed ing",
+      "ĠHE K",
+      "ly m",
+      "ĠB V",
+      "Ġvis co",
+      "i et",
+      "ĠCO X",
+      "ploy ment",
+      "ins ki",
+      "Ġout door",
+      "ĠLiter ature",
+      "ant ed",
+      "meth oxyphenyl",
+      "ĠMed ium",
+      "Ġd ia",
+      "ail and",
+      "vari ance",
+      "ĠEval uating",
+      "ox acin",
+      "Ġanti f",
+      "Ġpul p",
+      "Ġcorro bor",
+      "ĠO t",
+      "Ġrabb its",
+      "R u",
+      "Ġfunction als",
+      "â ĩ",
+      "Ġimm ersion",
+      "Ġcre atin",
+      "Ġq RT",
+      "Ġcondens ed",
+      "n r",
+      "ĠV A",
+      "h ad",
+      "Ġk ing",
+      "ob le",
+      "Ġexist ed",
+      "Ġthe sis",
+      "ubb ard",
+      "ap optotic",
+      "Ġflow ering",
+      "ĠAdap tation",
+      "ĠKal man",
+      "tr l",
+      "Ġm ent",
+      "ut ation",
+      "ĠCon v",
+      "Ġhist ories",
+      "Ġen anti",
+      "n ell",
+      "on ian",
+      "ĠF abric",
+      "Ġx x",
+      "Ġf ell",
+      "Ġcytos olic",
+      "Ġm ud",
+      "Ġsusp ensions",
+      "ĠMicro bial",
+      "meas ured",
+      "Ġdown load",
+      "Ġinv alid",
+      "Ġcapt uring",
+      "ĠH H",
+      "ĠG ray",
+      "ĠA Z",
+      "ĠN ash",
+      "vi ation",
+      "nai re",
+      "or tium",
+      "yn ch",
+      "amin ergic",
+      "Ġwa it",
+      "S chem",
+      "t race",
+      "ĠV ill",
+      "Ġpo ols",
+      "Ġhypox ic",
+      "x p",
+      "Ġsh aded",
+      "OR Y",
+      "t urn",
+      "inter acting",
+      "Ġdestroy ed",
+      "ak h",
+      "ĠCp G",
+      "dot ted",
+      "ĠTrans cript",
+      "plan ar",
+      "Ġpre clinical",
+      "ĠRe pro",
+      "ĠSur gery",
+      "Stok es",
+      "if def",
+      "Ġdiscrim inate",
+      "ĠG ross",
+      "Ġfl ags",
+      "i ety",
+      "umm y",
+      "Ġtransf ers",
+      "S G",
+      "ĠSc i",
+      "Ġhead er",
+      "ĠFund ing",
+      "Ġdet rim",
+      "Ġinst abilities",
+      "ĠPhyl ogenetic",
+      "ym ethyl",
+      "ĠAssess ing",
+      "RO C",
+      "els en",
+      "Equ al",
+      "Ġc as",
+      "Ġver tically",
+      "Ġvis ibility",
+      "ĠFT IR",
+      "sc rib",
+      "Ġbur sts",
+      "ĠDo ug",
+      "ĠFranc isco",
+      "ĠM SC",
+      "Ġpred is",
+      "establ ished",
+      "Ġfac ed",
+      "ĠW I",
+      "S l",
+      "Ġchar ts",
+      "orth y",
+      "izon tal",
+      "ial ysis",
+      "Ġtun able",
+      "Ġexplos ion",
+      "S w",
+      "T NF",
+      "Ġdiscontinu ous",
+      "ect ure",
+      "ci ences",
+      "mathbb m",
+      "lo ok",
+      "Ġt achy",
+      "Ġb row",
+      "obs erved",
+      "Ġana est",
+      "S al",
+      "q PCR",
+      "Ġse es",
+      "Ġspac ecraft",
+      "Ġsal es",
+      "ĠT rac",
+      "T em",
+      "iv est",
+      "ĠF c",
+      "ĠNew s",
+      "Ġharvest ing",
+      "ĠE G",
+      "p ad",
+      "Ġnanow ires",
+      "Ġpot ato",
+      "pl iers",
+      "on in",
+      "Ġw orm",
+      "s ue",
+      "ti e",
+      "Ġm asks",
+      "Ġth row",
+      "! !",
+      "be havi",
+      "Ġp ine",
+      "og y",
+      "TE ST",
+      "on to",
+      "Ġcreatin ine",
+      "ĠB oston",
+      "Ġch air",
+      "pl oys",
+      "ov en",
+      "Ġent rance",
+      "Ġc och",
+      "Ġdy es",
+      "T or",
+      "ĠPD E",
+      "unders et",
+      "atas ets",
+      "Ġt ernary",
+      "cho ose",
+      "f ive",
+      "chlor ide",
+      "on ium",
+      "Pro perty",
+      "Ġt u",
+      "Ġadequ ately",
+      "romy cin",
+      "Ġco oper",
+      "ï Ľľ",
+      "Ġpap ill",
+      "ĠStrept ococcus",
+      "ĠC Y",
+      "Ġgroup ing",
+      "Ġbi oc",
+      "ĠCardi ac",
+      "ĠBo ok",
+      "re ference",
+      "Ġconfirm ation",
+      "iver y",
+      "Ġwar ning",
+      "pret ation",
+      "Ġl ove",
+      "Ġoscill ators",
+      "s ed",
+      "ĠT X",
+      "il ent",
+      "ĠV as",
+      "Ġcl amp",
+      "Ġa head",
+      "ac s",
+      "Ġdeple ted",
+      "Ġmethod ologies",
+      "m ay",
+      "Ġc affe",
+      "Ġsequ entially",
+      "os acchar",
+      "Ġcompr ise",
+      "Ġc hel",
+      "Ġin acc",
+      "Ġtend on",
+      "S equ",
+      "ough t",
+      "ser ver",
+      "ĠPert urb",
+      "Ġter rain",
+      "cur ve",
+      "ĠAr gent",
+      "T ABLE",
+      "Ġimplicit ly",
+      "Ġenj oy",
+      "ĠS itter",
+      "Ġmic ron",
+      "ĠEv ans",
+      "ns ylvan",
+      "Ġlook ed",
+      "sp e",
+      "vol ving",
+      "ĠL STM",
+      "agne tism",
+      "ĠNot ch",
+      "ĠT al",
+      "ĠDEG s",
+      "lem an",
+      "Ġbo olean",
+      "Ġob ey",
+      "organ ization",
+      "se en",
+      "ĠEn c",
+      "sch ild",
+      "ĠOnt ario",
+      "Ele ment",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠĠ",
+      "m ouse",
+      "Ġpoly ethylene",
+      "Ġace tic",
+      "s ections",
+      "ur onal",
+      "ĠD ick",
+      "Ġk ill",
+      "Ġbroad ening",
+      "Ġfluor ide",
+      "Ġs aved",
+      "Ġde em",
+      "St ream",
+      "ac ed",
+      "ĠJ eff",
+      "Q A",
+      "Ġscal able",
+      "ĠF if",
+      "ĠMin i",
+      "Ġsuper gravity",
+      "Ġcoll oidal",
+      "L Y",
+      "O A",
+      "Ġper ic",
+      "Ġshort ly",
+      "Ġv ap",
+      "Ġspl its",
+      "m ove",
+      "Ġstim ulating",
+      "ĠBe ijing",
+      "Ġp yr",
+      "Ï Ń",
+      "Ġlex ical",
+      "âĢ ł",
+      "Å Ħ",
+      "itor ies",
+      "oler ance",
+      "Ġins ulator",
+      "ĠLe on",
+      "Ġpropag ate",
+      "ĠEle ments",
+      "y en",
+      "Mod ule",
+      "ĠWhe ther",
+      "Ġa ph",
+      "ĠLa ure",
+      "ĠMut ations",
+      "Ġhypert rophy",
+      "Ġocean ic",
+      "ograph ically",
+      "pati ents",
+      "ĠAngel es",
+      "Ġp he",
+      "Ġsqu ee",
+      "Ġcaro ten",
+      "f ine",
+      "Ġsk etch",
+      "Ġans atz",
+      "tit ution",
+      "ĠF us",
+      "ĠS ug",
+      "obacter ial",
+      "Ħ ĥ",
+      "Rel ated",
+      "Ġar tist",
+      "Ġac ryl",
+      "l ined",
+      "raf ted",
+      "ĠQ oS",
+      "ĠF eng",
+      "se arch",
+      "Ġnanot ube",
+      "ĠV M",
+      "ah l",
+      "Ġstr ide",
+      "ĠT ag",
+      "ĠL ar",
+      "Ġdes orption",
+      "d type",
+      "Ġb ug",
+      "Ġcareg ivers",
+      "ĠH un",
+      "ĠPrac tical",
+      "Ġob lig",
+      "re r",
+      "ĠK ang",
+      "ĠPro ducts",
+      "omet h",
+      "ĠHe La",
+      "Ġlabor atories",
+      "n atural",
+      "Ġf ul",
+      "Ġm old",
+      "ab ine",
+      "ĠS pring",
+      "Ġco bal",
+      "Ġhighlight ing",
+      "ĠPre f",
+      "cycl ic",
+      "ĠCONCLUS ION",
+      "ĠS ources",
+      "Ġap ex",
+      "par ser",
+      "ĠLog ic",
+      "Ġp ond",
+      "Ġto ld",
+      "ĠSh ap",
+      "perg illus",
+      "Ġsay ing",
+      "Ġmutagen esis",
+      "Ġmm Hg",
+      "ĠP AN",
+      "Ġsm okers",
+      "od ay",
+      "Ġhere in",
+      "CM V",
+      "ĠP W",
+      "Ġred shifts",
+      "ĠMin im",
+      "ym an",
+      "ull i",
+      "d ense",
+      "Ġarsen ic",
+      "ĠE MT",
+      "og aster",
+      "carboxyl ate",
+      "s ys",
+      "R o",
+      "an ch",
+      "ĠAl pha",
+      "ĠTechn ical",
+      "s v",
+      "Ġb ones",
+      "Ġaccept or",
+      "Ġnew born",
+      "pri vate",
+      "Ġnan or",
+      "ĠSw iss",
+      "a round",
+      "Ġsynt ax",
+      "ĠK ähler",
+      "Ġaer ial",
+      "ĠP ale",
+      "typ edef",
+      "names pace",
+      "Ġconfound ing",
+      "vi Äĩ",
+      "Ġret ard",
+      "Ġz eta",
+      "ĠT um",
+      "is ch",
+      "Ġsulf ide",
+      "ĠT ian",
+      "u y",
+      "Ġintu ition",
+      "Ġphosph olip",
+      "ĠS her",
+      "ric ts",
+      "-------------------------------- --------------------------------",
+      "ok ines",
+      "gluc ose",
+      "tol er",
+      "ifer ative",
+      "ĠFlu or",
+      "Ġencour age",
+      "Ġrespons ive",
+      "perturb ative",
+      "Ġs addle",
+      "l ers",
+      "nd ez",
+      "ĠZ ero",
+      "ĠDi et",
+      "Ġdeveloper s",
+      "S yn",
+      "Ġconf er",
+      "Ġorigin ate",
+      "rop ol",
+      "ha w",
+      "le tion",
+      "ms kip",
+      "Ġb er",
+      "Ġpe at",
+      "v ially",
+      "Ġgran ules",
+      "ĠÌ ĥ",
+      "Ġpl uripot",
+      "Ġassim ilation",
+      "Ġdenom inator",
+      "abil ization",
+      "ĠEpidem iology",
+      "MI N",
+      "eed s",
+      "ĠV R",
+      "E val",
+      "st ore",
+      "ĠBas eline",
+      "Ġc u",
+      "ĠSpect ra",
+      "Ġfraction ation",
+      "Ġplac ing",
+      "Ġbur ied",
+      "el eration",
+      "Ġalkal i",
+      "ĠI U",
+      "C alc",
+      "we ak",
+      "Ġmorphism s",
+      "Ġlig ase",
+      "Ġf s",
+      "Ġutil izes",
+      "Com put",
+      "Ã ¢",
+      "Ġs tig",
+      "rel ative",
+      "Ġimm ature",
+      "ĠF rac",
+      "ap i",
+      "Ġout patient",
+      "Ġachieve ment",
+      "Ġstack ing",
+      "Ġnod ules",
+      "IN D",
+      "ĠGP a",
+      "Ġpercol ation",
+      "m space",
+      "Ġbrain s",
+      "uff le",
+      "ent ropy",
+      "L ab",
+      "Ġstabil ize",
+      "ĠRic ci",
+      "ĠAn timicrobial",
+      "pers onal",
+      "Ġfarm s",
+      "ĠP in",
+      "Ġpor cine",
+      "Ġoccasion ally",
+      "w he",
+      "Ġundergo es",
+      "Ġregim ens",
+      "Ġbl ade",
+      "Ġlinear ized",
+      "Ġdec on",
+      "Ġpack ed",
+      "Ġf ishes",
+      "ĠM end",
+      "Ġapproach ing",
+      "Ġball s",
+      "Ġpro inflammatory",
+      "imer ic",
+      "ĠDirect or",
+      "Ġsol iton",
+      "Ġm osaic",
+      "vi et",
+      "Me an",
+      "ĠP ad",
+      "Ġtri plicate",
+      "sup ported",
+      "Ġcar t",
+      "<< <<",
+      "Ġrem ission",
+      "ase ous",
+      "astic ity",
+      "ĠM ik",
+      "ĠStrateg y",
+      "ram er",
+      "ĠPol ish",
+      "Ġent hal",
+      "Ġheter ozygous",
+      "ĠGra vity",
+      "A x",
+      "Ġorganization al",
+      "Ġmo vie",
+      "Ġexpl oratory",
+      "WL ED",
+      "Ġmo iety",
+      "dec re",
+      "ĠS till",
+      "ĠÂ ¡",
+      "Ġgreen house",
+      "Ġsupercon ductors",
+      "en um",
+      "el in",
+      "Ġoffer ing",
+      "st ad",
+      "ĠT rich",
+      "Ġre pl",
+      "Ġrecycl ing",
+      "ph or",
+      "Ġin elastic",
+      "ock ey",
+      "ĠâĢ Ļ",
+      "Ġsequ el",
+      "E B",
+      "ĠCh ile",
+      "Ġfibr illation",
+      "Ġdis ulfide",
+      "ob tained",
+      "ub in",
+      "Ĥ ¬",
+      "Ġfacilit ating",
+      "Ġhop ping",
+      "Ġmedi ator",
+      "Ġhyd ration",
+      "Ġspars ity",
+      "Ġs ati",
+      "Ġis othermal",
+      "Ġreturn ing",
+      "Ġtravel ing",
+      "Ġin g",
+      "Ġst ent",
+      "Ġcapac itor",
+      "Ġcomprom ise",
+      "ĠS ud",
+      "ĠV ision",
+      "Ġtop ologies",
+      "opol ysaccharide",
+      "ĠPro file",
+      "ĠR ing",
+      "Ġdiscrep ancies",
+      "D is",
+      "AR D",
+      "cc cc",
+      "Ġdirect ory",
+      "ĠCM OS",
+      "ow ed",
+      "ill o",
+      "ĠIns ights",
+      "ĠT ib",
+      "Ġab and",
+      "aro se",
+      "Or der",
+      "ĠÂ ¬",
+      "Ġintrac ranial",
+      "Ġintermedi ates",
+      "Ġhab its",
+      "Ġcar p",
+      "pro perty",
+      "IM AGE",
+      "ĠU k",
+      "Ġhydroph ilic",
+      "W id",
+      "Ġab iotic",
+      "Ġobser vers",
+      "Ġch or",
+      "ĠCons ervation",
+      "ĠEnh ance",
+      "ĠAutom ated",
+      "ĠGl ut",
+      "ir atory",
+      "Ġsp aw",
+      "ĠE fficiency",
+      "v ast",
+      "in iti",
+      "Ġop tional",
+      "ĠScal ing",
+      "if old",
+      "Ġmt DNA",
+      "ĠRec onstruction",
+      "Ġcount able",
+      "ĠGr ass",
+      "D en",
+      "ĠCh ain",
+      "en zyme",
+      "Ġwave forms",
+      "Ġpancre as",
+      "ĠDet ailed",
+      "cm d",
+      "Ġâİ ľ",
+      "Ġmagnet o",
+      "ĠFP GA",
+      "Ġabsol utely",
+      "Ġstim ulates",
+      "ach us",
+      "ĠAr n",
+      "m essage",
+      "ocomp atibility",
+      "H Cl",
+      "ĠF ish",
+      "Ġphenomen ological",
+      "Ġsaliv ary",
+      "ond o",
+      "Ġno tions",
+      "f ur",
+      "U CT",
+      "Ġw ww",
+      "ab et",
+      "ĠS ulf",
+      "F il",
+      "dom inated",
+      "ars er",
+      "Ġpack ages",
+      "Ġsplic e",
+      "F lo",
+      "NO WLED",
+      "x a",
+      "ĠY uan",
+      "Ġacet one",
+      "ĠVit amin",
+      "ĠÎ ŀ",
+      "Ġobs c",
+      "Ġcha per",
+      "Ġm ort",
+      "M AN",
+      "Ġsub tilis",
+      "Ġoptim ality",
+      "Ġcontinu ing",
+      "Ġdu plication",
+      "Ġmultip lying",
+      "Ġimmun ological",
+      "Ġcir rhosis",
+      "h ospital",
+      "ĠProb abilistic",
+      "Ġdele tions",
+      "Ġca ution",
+      "Ġow ner",
+      "ox orubicin",
+      "Ġla unch",
+      "Ġc ure",
+      "th us",
+      "ĠHerm itian",
+      "can onical",
+      "Ġimmun ore",
+      "form in",
+      "Ġbroad band",
+      "part um",
+      "op he",
+      "ĠB eta",
+      "ĠB I",
+      "Ġïĺ º",
+      "Ġj umps",
+      "Ġparad ox",
+      "um ped",
+      "Ġdoc tors",
+      "Ġhospital ized",
+      "Ġwas h",
+      "prec ision",
+      "Ġr uled",
+      "Ġdu plicate",
+      "ant e",
+      "Ġneuro trans",
+      "Ġïĥ §",
+      "Ġthem e",
+      "T aking",
+      "ĠPl ants",
+      "f ollowing",
+      "Ġage ing",
+      "Ġcon gestion",
+      "os arcoma",
+      "Ġrepos itory",
+      "ĠH ess",
+      "ĠC atalytic",
+      "ĠD V",
+      "IN K",
+      "pri v",
+      "ĠAn a",
+      "ĠS LE",
+      "ĠTh ailand",
+      "í ķ",
+      "Ġd uty",
+      "loc ations",
+      "ot er",
+      "Ġlys ine",
+      "Ġind ist",
+      "Ġagon ists",
+      "A ck",
+      "Ġminim ally",
+      "Ġet ching",
+      "ugg ing",
+      "c uda",
+      "nd ef",
+      "Ġref erring",
+      "Ġlys ates",
+      "Ġseroton in",
+      "crib ing",
+      "ĠInter face",
+      "d V",
+      "Ġd urations",
+      "Ġphot od",
+      "Ġd ating",
+      "Ġirre versible",
+      "os idase",
+      "ĠF ROM",
+      "with in",
+      "SN R",
+      "Ġarr hyth",
+      "ĠR atio",
+      "ĠTh in",
+      "cent ered",
+      "Ġsh ocks",
+      "ĠV ers",
+      "Ġnotice able",
+      "Ġf oci",
+      "Ġorth onormal",
+      "Ġâİ Ł",
+      "Ġlum inescence",
+      "ĠSUS Y",
+      "in ternal",
+      "ĠT our",
+      "Ġab brevi",
+      "ĠM AL",
+      "ver tex",
+      "Ġem ploys",
+      "IN S",
+      "Ġimmunohist ochemistry",
+      "Ġhepar in",
+      "Ġidi opathic",
+      "Ġimmobil ized",
+      "is he",
+      "ph th",
+      "th in",
+      "ĠSt orage",
+      "Ġperovsk ite",
+      "Pro t",
+      "ĠDepend ing",
+      "Ġbl ends",
+      "Ġpred ator",
+      "Ġdisplay ing",
+      "Ġvesic le",
+      "ĠK ra",
+      "Ġl ane",
+      "Ġmulti layer",
+      "Ġhom ozygous",
+      "cos h",
+      "Ġsuperf icial",
+      "Ġ il",
+      "ĠK R",
+      "ĠBr un",
+      "ĠE W",
+      "op a",
+      "ĠCart esian",
+      "ĠCy toplas",
+      "ĠPen nsylvan",
+      "b ands",
+      "Ġangi otensin",
+      "ĠLat tice",
+      "G I",
+      "j ee",
+      "Ġenlarg ed",
+      "en ius",
+      "ĠI a",
+      "ou x",
+      "Ġg ent",
+      "Ġcarbon yl",
+      "c hers",
+      "Ġhypot he",
+      "Ġmic rosp",
+      "Ġaff ective",
+      "Ġax ons",
+      "e i",
+      "ypt oph",
+      "ĠJ on",
+      "que ue",
+      "ĠG auge",
+      "men opausal",
+      "ĠD as",
+      "ĠEss ential",
+      "ĠF ault",
+      "ĠB il",
+      "Ġtest osterone",
+      "Ġcham bers",
+      "d ione",
+      "Ġelic ited",
+      "IG N",
+      "Ġantioxid ants",
+      "pop ulations",
+      "Ġov ary",
+      "Ġâ ĸ",
+      "Ġabst raction",
+      "Ġhydro carbons",
+      "Ġrec tal",
+      "Ġtrigger ing",
+      "Ġthorough ly",
+      "R un",
+      "acter ia",
+      "in formation",
+      "ĠB ed",
+      "Ġqu enc",
+      "Ġund ers",
+      "ĠScot land",
+      "Ġre volution",
+      "Ġpit uitary",
+      "Ġanthrop ogenic",
+      "f ocus",
+      "Ġmet han",
+      "Ġinf low",
+      "Ġdef lection",
+      "ĠC ape",
+      "Ġmulti dimensional",
+      "Ġarri ved",
+      "ĠS par",
+      "d v",
+      "Ġc ows",
+      "ĠB h",
+      "Ġj k",
+      "tol yl",
+      "Ġeigen states",
+      "Ġpre processing",
+      "ĠR ain",
+      "ä ¸",
+      "in z",
+      "Ġm n",
+      "RE E",
+      "atric k",
+      "D ev",
+      "Ġfulf illed",
+      "Ġar tic",
+      "Ġreal izations",
+      "ĠComp onent",
+      "ĠW S",
+      "Ġinf o",
+      "pr inted",
+      "at osis",
+      "c ache",
+      "an ov",
+      "ĠT g",
+      "cont ent",
+      "j unc",
+      "ĠCD K",
+      "Ġbeh aves",
+      "ĠK id",
+      "diff erence",
+      "ĠP s",
+      "ĠU g",
+      "Ġstruct urally",
+      "ereb ral",
+      "ĠSur ve",
+      "he al",
+      "on ite",
+      "Ġdele ted",
+      "iti m",
+      "St ar",
+      "ĠSpe ech",
+      "ĠA str",
+      "g radient",
+      "Ġf ellow",
+      "Ġsy ring",
+      "N B",
+      "ĠN B",
+      "Ġcre ep",
+      "Ġlog ging",
+      "Ġint en",
+      "scal ar",
+      "ĠAtmosp heric",
+      "Ġl upus",
+      "Ġiden tically",
+      "process ed",
+      "sign al",
+      "ĠClo str",
+      "anc ers",
+      "Ġd b",
+      "Ġsubs ystem",
+      "s itu",
+      "Ġferro electric",
+      "Ġï Ľľ",
+      "Ġo re",
+      "ĠR b",
+      "ĠMicro soft",
+      "ĠC och",
+      "ĠAc tin",
+      "Ġnerv es",
+      "Ġexper tise",
+      "o tive",
+      "ĠPoinc aré",
+      "ĠR ig",
+      "Ġpsych osocial",
+      "Ġprogen itors",
+      "ĠM yr",
+      "ĠH ug",
+      "Ġbi ogenesis",
+      "Ġincorpor ates",
+      "Ġnever theless",
+      "ĠDec l",
+      "obs erv",
+      "Ġmulti plier",
+      "Ġrespond ing",
+      "ho ff",
+      "Ġimp acted",
+      "Ġsynd romes",
+      "k el",
+      "ĠS ynt",
+      "ĠCon cer",
+      "ĠAmeric ans",
+      "Ġspac ed",
+      "um ption",
+      "ĠThom pson",
+      "ĠJacob ian",
+      "T ra",
+      "e volution",
+      "Ġdid n",
+      "Ġpercenti le",
+      "Ġl id",
+      "equ ivalent",
+      "Ġantic o",
+      "Ġmulti ply",
+      "Ġpen icillin",
+      "Ġrespons iveness",
+      "Ġrun off",
+      "al anine",
+      "squ ares",
+      "ĠIns ulin",
+      "re le",
+      "ĠL if",
+      "ĠMink owski",
+      "Ġbl end",
+      "ĠP and",
+      "Ġtw elve",
+      "Ġhybrid s",
+      "Ġb ass",
+      "inter action",
+      "ĠBanglades h",
+      "Ġop ens",
+      "ĠAr ts",
+      "Ġconc ave",
+      "Ġped est",
+      "Ġf ist",
+      "ĠAd ults",
+      "open ia",
+      "EN CE",
+      "ĠF usion",
+      "Ġmicro c",
+      "ĠSur gical",
+      "yl ate",
+      "Ġpack aging",
+      "OC K",
+      "Q C",
+      "T ri",
+      "sc an",
+      "Ġregard s",
+      "Ġdiscrim inant",
+      "Ġind ustries",
+      "ic us",
+      "ĠWalk er",
+      "Ġpe ers",
+      "sy nt",
+      "Ġhor se",
+      "Ġflow ing",
+      "ur red",
+      "ĠCR P",
+      "ĠCare er",
+      "iffiffiffiff iffiffiffiff",
+      "ĠM SE",
+      "han a",
+      "ĠMor tality",
+      "Ġtumorigen esis",
+      "ĠIs lam",
+      "Ġazim uthal",
+      "w en",
+      "Ġs ys",
+      "az in",
+      "ne ighbor",
+      "Con fig",
+      "the y",
+      "Ġs orption",
+      "Ġsp anned",
+      "Ġview point",
+      "M OD",
+      "Ġth rust",
+      "up lex",
+      "Ġhist ograms",
+      "Ġprogram med",
+      "Ġeth ics",
+      "ect able",
+      "represent ation",
+      "um ns",
+      "Ġstre et",
+      "ĠSob olev",
+      "Ġexc ision",
+      "ĠR ud",
+      "qui res",
+      "Ġown ed",
+      "Ġthous and",
+      "Ġantagon ists",
+      "U ST",
+      "Ġdrastic ally",
+      "ĠóµĦ ©",
+      "ĠD or",
+      "ĠM OS",
+      "p n",
+      "ĠDec re",
+      "D ep",
+      "Ġs intering",
+      "Ġpur ple",
+      "et hanol",
+      "Ġhydro carbon",
+      "ĠF O",
+      "left rightarrow",
+      "Ġimmun ofluorescence",
+      "ĠO M",
+      "Ġmat urity",
+      "Ġearthqu akes",
+      "Ġax on",
+      "Ġprob ed",
+      "OR D",
+      "ĠAD P",
+      "s g",
+      "om ere",
+      "Ġtrans cribed",
+      "M ar",
+      "ĠU til",
+      "ĠI A",
+      "Ġcomp iled",
+      "Ġsuper vision",
+      "ĠX en",
+      "ĠJ ur",
+      "com par",
+      "Ġhypert ensive",
+      "il ized",
+      "ra e",
+      "Con clusion",
+      "'' '",
+      "Do uble",
+      "ĠF as",
+      "Ġins ectic",
+      "ĠPre m",
+      "P ri",
+      "ĠCa o",
+      "ĠQuestion naire",
+      "Ġg athered",
+      "G W",
+      "ĠN V",
+      "ĠLact obacillus",
+      "Ġcycl in",
+      "Ġre ject",
+      "Ġsk ull",
+      "Ġa w",
+      "ĠC old",
+      "Ġmes ons",
+      "b d",
+      "Ġdetrim ental",
+      "ap ore",
+      "now led",
+      "ĠCX CL",
+      "Ġspik es",
+      "Ġt ent",
+      "ĠL ength",
+      "Ġdo or",
+      "Ġfl our",
+      "ustr ation",
+      "He alth",
+      "Ġtrans parency",
+      "Ġdisrup ted",
+      "H y",
+      "over l",
+      "ĠReinfor cement",
+      "cept ors",
+      "ĠK os",
+      "ret roviral",
+      "ĠIN T",
+      "ĠS or",
+      "Ġadop ting",
+      "Ġend oplasmic",
+      "Ġsu it",
+      "Ġopi oid",
+      "Ġintegr in",
+      "aw ay",
+      "Ġtail ored",
+      "ĠS oc",
+      "Ġqu ies",
+      "Ġhus band",
+      "Ġ umb",
+      "ĠC ai",
+      "ĠAs pergillus",
+      "ĠGa N",
+      "Ġdistingu ishing",
+      "Ġextrap olation",
+      "Ġc age",
+      "Ġscav enging",
+      "K F",
+      "T ree",
+      "ĠConfl ict",
+      "UN C",
+      "Ġmang anese",
+      "d ays",
+      "Ã Ł",
+      "ĠL ive",
+      "s d",
+      "ract or",
+      "Ġl ute",
+      "Ġdis similar",
+      "Ġ ib",
+      "ĠV eg",
+      "Ġoccur rences",
+      "Ġbin omial",
+      "Schem e",
+      "Ġt ape",
+      "ĠC ant",
+      "Ġelect rosp",
+      "C d",
+      "m ade",
+      "Ġse vent",
+      "sh ared",
+      "Ġaccess ion",
+      "or p",
+      "D ATA",
+      "le ted",
+      "V ari",
+      "Ġro se",
+      "tag ged",
+      "ĠA th",
+      "Ġed dy",
+      "est one",
+      "Ġes ters",
+      "Ġtyp ing",
+      "ĠStud ents",
+      "y i",
+      "ores istance",
+      "ino is",
+      "Ġgluc ocortic",
+      "i osis",
+      "Ġcor onal",
+      "Ġshe ath",
+      "ĠT rack",
+      "Ġequ ilibria",
+      "amm ing",
+      "Ġp ione",
+      "Ġsc iences",
+      "Ġsuppress ing",
+      "Ġdec o",
+      "if ndef",
+      "H is",
+      "Ġpel let",
+      "L inear",
+      "orb ent",
+      "Ġflat ten",
+      "Ġst raw",
+      "Ġal beit",
+      "ĠPredic tive",
+      "Ġg aze",
+      "Ġhydro ly",
+      "ut her",
+      "od ers",
+      "Ġfl ap",
+      "Ġsimplic ial",
+      "S ystem",
+      "Ġst ressed",
+      "Ġimmun oglobulin",
+      "il ia",
+      "Ġconsum ing",
+      "ĠÃ ©",
+      "gal act",
+      "Ġadul thood",
+      "Ġvor ticity",
+      "ycl ic",
+      "ovolta ic",
+      "ivest ock",
+      "Ġbed s",
+      "ĠPl anning",
+      "Ġparameter ized",
+      "Ġg host",
+      "maxim um",
+      "Ġsuper im",
+      "Ġphysic ochemical",
+      "g p",
+      "ong ue",
+      "Ġprim ordial",
+      "x ff",
+      "ins ula",
+      "M c",
+      "Ġminim izes",
+      "ĠGra vitational",
+      "os oma",
+      "ign ificant",
+      "Ġelucid ated",
+      "Ġsub surface",
+      "sign ificant",
+      "Ġrel atives",
+      "fer roni",
+      "trans f",
+      "Ġtail s",
+      "b eck",
+      "om agnetic",
+      "Ġun necessary",
+      "Ġmon omial",
+      "del ay",
+      "Ġst a",
+      "ĠS uz",
+      "Ġalter ing",
+      "LO G",
+      "ĠL ac",
+      "Ġr anks",
+      "h w",
+      "ĠN ep",
+      "Ġneu ropath",
+      "ĠComp e",
+      "G r",
+      "P ati",
+      "red uce",
+      "ĠMalays ia",
+      "cer al",
+      "Ġmicro bes",
+      "Ġlens ing",
+      "ĠCalc ium",
+      "ĠDeterm in",
+      "ĠCost a",
+      "Ġke eps",
+      "print ing",
+      "ĉĉĉĉ ĉĉ",
+      "ch in",
+      "ex posed",
+      "Ġperiod ically",
+      "Ġrend er",
+      "ĠCardi ovascular",
+      "ent in",
+      "Ġbio availability",
+      "Ġinterpret ations",
+      "ĠC U",
+      "Ġneg oti",
+      "Ġan tim",
+      "Ġdeem ed",
+      "Ġa e",
+      "Ġhal os",
+      "ĠMich igan",
+      "Ġoste oarthritis",
+      "di ag",
+      "ĠB eng",
+      "Ġmet agen",
+      "Ġparameter ization",
+      "di agn",
+      "ĠMat ching",
+      "Ġcatal ysis",
+      "ut s",
+      "Ġdissem ination",
+      "Ġout let",
+      "ĠMo on",
+      "ĠG ST",
+      "sp here",
+      "Ġresearc her",
+      "ambig uation",
+      "Ġra ises",
+      "Ġflavon oids",
+      "ĠMultiv ariate",
+      "Ġac cl",
+      "W I",
+      "Ġn u",
+      "Ġerg odic",
+      "un ique",
+      "atin ib",
+      "Ġresol utions",
+      "Ġhous es",
+      "D EC",
+      "ig hed",
+      "Ġsix th",
+      "Ġpolitic ian",
+      "ap ache",
+      "Ġsol ute",
+      "Ġaug ment",
+      "st ress",
+      "H IV",
+      "ĠS ets",
+      "Ġtrans istors",
+      "qu bit",
+      "am ines",
+      "Ġfarm ers",
+      "Ġn t",
+      "ĠLag range",
+      "Ġveget ables",
+      "Ġpre t",
+      "ĠS ynthetic",
+      "Ġcon es",
+      "Ġmedic ines",
+      "Ġgen omics",
+      "Ġexperi encing",
+      "ag land",
+      "Ġgen ital",
+      "ĠObserv atory",
+      "ĠS kin",
+      "ĠR osen",
+      "ĠBrit ain",
+      "gen ome",
+      "ĠEnt ropy",
+      "Ġr ac",
+      "G o",
+      "Ġw alks",
+      "cript or",
+      "ĠB aker",
+      "ok er",
+      "Ġprop ensity",
+      "Ġpopular ity",
+      "restric ted",
+      "ĠB ert",
+      "b efore",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠĠĠ",
+      "aut o",
+      "R ank",
+      "ĠR CT",
+      "Ġp ocket",
+      "ob ut",
+      "Ġbenz ene",
+      "ĠCN T",
+      "yptoph an",
+      "all is",
+      "ĠRes ources",
+      "ĠBer lin",
+      "Ġsch olar",
+      "gl ob",
+      "ĠSp eed",
+      "ĠX iao",
+      "big gl",
+      "AN CE",
+      "ĠPr ime",
+      "Ph ys",
+      "id ia",
+      "Ġmon oc",
+      "ĠCommun ications",
+      "ĠPrec ision",
+      "ĠPa uli",
+      "Ġinvestig ators",
+      "ĠLi ang",
+      "Ġmeteor ological",
+      "m og",
+      "re ens",
+      "ub ric",
+      "Ġrearrang ement",
+      "or ta",
+      "E lect",
+      "ĠT ukey",
+      "ĠM is",
+      "Ġepid erm",
+      "ĠACK NOWLED",
+      "w art",
+      "Ġexcit on",
+      "Ġassoci ative",
+      "st yrene",
+      "Ġl osing",
+      "ĠO d",
+      "p rep",
+      "ess ation",
+      "Ġattrib utable",
+      "ĠNa vier",
+      "an z",
+      "Ġcorrect ness",
+      "o ints",
+      "ĠR ather",
+      "Ġassemb lies",
+      "Ġbrid ges",
+      "OS S",
+      "M ET",
+      "Ġper m",
+      "Ġauthor ities",
+      "Ġiod ine",
+      "sh ire",
+      "inter val",
+      "epti d",
+      "Ġpot ency",
+      "Ġrenew able",
+      "v ard",
+      "Ġsur jective",
+      "Ġsubsequ ence",
+      "ĠE Vs",
+      "it ching",
+      "Ġgen otyping",
+      "ĠAcc urate",
+      "iop hene",
+      "G ly",
+      "pl ified",
+      "ĠDist inct",
+      "AC H",
+      "Ġspe akers",
+      "hol m",
+      "Ġpro s",
+      "ĠDev ice",
+      "m c",
+      "ĠD ense",
+      "ĠV a",
+      "r ison",
+      "Ġac yl",
+      "ĠPrinc ipal",
+      "ĠV iral",
+      "Ġcos ine",
+      "ĠRes idual",
+      "Ġeff lux",
+      "ĠSub jects",
+      "Ġrect angle",
+      "work ers",
+      "Ġrot ated",
+      "Ġb omb",
+      "ĠRes olution",
+      "ne ar",
+      "ĠÂ ®",
+      "Ġestabl ishes",
+      "am ed",
+      "Ġcompet ence",
+      "G lu",
+      "ĠD end",
+      "ĠH sp",
+      "ens ation",
+      "ĠLe ad",
+      "Ġlog ger",
+      "sin h",
+      "Ġint ellectual",
+      "form er",
+      "C e",
+      "Ġmon ocyte",
+      "ho res",
+      "Ġdiast olic",
+      "Ġlif espan",
+      "ĠSil va",
+      "ar um",
+      "Ġtrans ducer",
+      "Ġout going",
+      "ent ation",
+      "Ġabsorb ing",
+      "it age",
+      "Ġsynt hesize",
+      "Ġfe eling",
+      "as ian",
+      "Ġcer amics",
+      "i ph",
+      "Ġnon local",
+      "P art",
+      "Ġimmers ed",
+      "station ary",
+      "lect ing",
+      "Ġweld ing",
+      "Ġres embles",
+      "ĠK at",
+      "m aster",
+      "Ġinters ect",
+      "ĠO lig",
+      "ĠTre nds",
+      "ag h",
+      "ĠN av",
+      "ĠT u",
+      "Ġep ist",
+      "Ġclin ics",
+      "Ġrepresent atives",
+      "Ġgrate ful",
+      "G PIO",
+      "H H",
+      "Ġun ambig",
+      "t uning",
+      "Ġnew sp",
+      "coh ol",
+      "################ ################",
+      "%%%% %%%%",
+      "represent ed",
+      "oc ic",
+      "ĠF uk",
+      "ĠS und",
+      "has one",
+      "M ode",
+      "ol one",
+      "ĠS b",
+      "Th ree",
+      "L ink",
+      "ce phal",
+      "ĠK ap",
+      "Ġelim inating",
+      "Ġmelan ogaster",
+      "â Ł",
+      "ĠB MD",
+      "IS E",
+      "ĠBat tle",
+      "Ġshrink age",
+      "ĠSe ven",
+      "ĠGl ass",
+      "rom agn",
+      "Ġk l",
+      "ĠOb viously",
+      "pres erving",
+      "ĠPl atform",
+      "ĠÌ ĩ",
+      "om avirus",
+      "ĠE ight",
+      "Ġall erg",
+      "ĠNanopar ticles",
+      "ary l",
+      "Ġpri ors",
+      "pat tern",
+      "Ġlinear ity",
+      "Ġtr uly",
+      "Pro cess",
+      "Ġdesc ending",
+      "ĠVictor ia",
+      "c ond",
+      "ĠI CP",
+      "ores cent",
+      "Ġauthor ity",
+      "Ġm ock",
+      "igm oid",
+      "Ġcomorbid ities",
+      "sim ple",
+      "Ġbl o",
+      "ĠComput e",
+      "Ġgest ation",
+      "achus etts",
+      "Ġph antom",
+      "ĠEd ward",
+      "ĠF BS",
+      "fact ors",
+      "ĠEstim ates",
+      "c lear",
+      "W B",
+      "pro ducts",
+      "num py",
+      "b rief",
+      "Ġsh op",
+      "ĠPol i",
+      "ĠRespir atory",
+      "Ġsurprising ly",
+      "Ġnanocom posite",
+      "divid ual",
+      "Ġholog raphic",
+      "ygd ala",
+      "ro plasty",
+      "ot actic",
+      "ĠPennsylvan ia",
+      "ĠSc ore",
+      "Ob j",
+      "Ġst ories",
+      "Ġmaxim izing",
+      "Ġgel atin",
+      "r ites",
+      "ĠT au",
+      "Ġtryp sin",
+      "Ġ ith",
+      "Ġf aint",
+      "Ġprim ing",
+      "ew orthy",
+      "ĠIn verse",
+      "Ġkn ots",
+      "sh arp",
+      "Ġtrain s",
+      "Ġcred it",
+      "ĠBel ow",
+      "pix el",
+      "Ġspind le",
+      "ĠP ast",
+      "Ġenum erate",
+      "ol ateral",
+      "Ġatt ending",
+      "Ġquanti zed",
+      "Ġhapl otypes",
+      "enc l",
+      "Ġw aven",
+      "Ġfurther more",
+      "Ġchalleng ed",
+      "Ġmanufact ured",
+      "ipher al",
+      "Ġinfin ites",
+      "ĠR and",
+      "Ġst aging",
+      "ag an",
+      "Ġper ox",
+      "tr ifluor",
+      "ĠMc K",
+      "ĠF OX",
+      "ĠL ank",
+      "ĠLu o",
+      "ĠAn th",
+      "ibri o",
+      "y el",
+      "ĠJ i",
+      "ĠI O",
+      "ĠB ridge",
+      "ĠR ow",
+      "Ġcompens ated",
+      "ats u",
+      "Ġhypothe tical",
+      "Ġtermin als",
+      "Ġcobal t",
+      "m ers",
+      "ĠM ang",
+      "N I",
+      "ĠR ac",
+      "AL S",
+      "f en",
+      "ĠU b",
+      "Ġpred ation",
+      "c adherin",
+      "ĠSh anghai",
+      "Ġtri es",
+      "Ġsp ort",
+      "acryl ate",
+      "ĠAlgebra ic",
+      "ain ts",
+      "Ex pr",
+      "Ġand rogen",
+      "Ġw edge",
+      "dis p",
+      "Ġstir red",
+      "ĠA le",
+      "Ġc ock",
+      "F our",
+      "Ġsc anner",
+      "Ġplas mon",
+      "ĠG ender",
+      "ĠRec ord",
+      "ĠInj ury",
+      "obl astic",
+      "ĠFlu orescence",
+      "Ġanti depress",
+      "Ġdefin itive",
+      "Ġrep ression",
+      "ordin ates",
+      "Ġangi ography",
+      "ĠHel ical",
+      "Ġcancell ation",
+      "re lease",
+      "Ġrel ational",
+      "ĠAnd re",
+      "mo lecule",
+      "Ġshap ing",
+      "ĠDen mark",
+      "ĠAL S",
+      "ĠN W",
+      "over rightarrow",
+      "Ġcomb at",
+      "box es",
+      "sub ject",
+      "Ġnanos cale",
+      "Ġcan ine",
+      "Ġs aving",
+      "Ġstrateg ic",
+      "St at",
+      "ĠD ub",
+      "Ġper mitted",
+      "ĠTw itter",
+      "â Ķ",
+      "Ġmem ories",
+      "ĠBus iness",
+      "ad ays",
+      "Ġpool ing",
+      "ĠClust ers",
+      "im ide",
+      "oun ters",
+      "frac tion",
+      "ĠCl iff",
+      "C am",
+      "E ven",
+      "K Y",
+      "k it",
+      "ibr ated",
+      "Ġaccompan ying",
+      "an us",
+      "Ġbu oy",
+      "Ġprolifer ative",
+      "Ġpro c",
+      "Ġstabil izing",
+      "ĠNam ely",
+      "pos p",
+      "so on",
+      "Ġaberr ant",
+      "Ġinter stellar",
+      "Over all",
+      "ĠG n",
+      "ĠFeed back",
+      "Ġo racle",
+      "Ġpre natal",
+      "com mun",
+      "Ġoutbreak s",
+      "Ġfertil ization",
+      "ĠM AG",
+      "Ġsing er",
+      "ĠMic rowave",
+      "ĠPar liament",
+      "cast ing",
+      "Gen eral",
+      "al gorithm",
+      "Ġph rase",
+      "Ġa vian",
+      "ĠP LA",
+      "Ġhard ly",
+      "approxim ately",
+      "AR CH",
+      "Ġtrans c",
+      "Ġdec omp",
+      "cont in",
+      "ĠMil ky",
+      "Ġher pes",
+      "R ange",
+      "O FF",
+      "prising ly",
+      "l x",
+      "ĠAB A",
+      "Ġsh ore",
+      "Ġderiv ing",
+      "Ġpel lets",
+      "nowled g",
+      "I tem",
+      "strand ed",
+      "bu ilt",
+      "Gl c",
+      "qu ist",
+      "ĠSub strate",
+      "Ġtra ditionally",
+      "ĠM ount",
+      "ival ence",
+      "ax ation",
+      "Ġloc ate",
+      "Ġg un",
+      "Ġvoc abulary",
+      "ĠPol ym",
+      "Ġec t",
+      "Ġm ult",
+      "Ġsediment ary",
+      "Ġautocor relation",
+      "ĠS ympt",
+      "Ġterr itory",
+      "Ġexcit atory",
+      "Ġv ote",
+      "Ġhe red",
+      "ace a",
+      "ĠF ocus",
+      "am pling",
+      "ff ee",
+      "Ġprim es",
+      "ĠM aking",
+      "ir s",
+      "MP s",
+      "Ġl itter",
+      "amet hasone",
+      "Ġk J",
+      "Ġsecret ory",
+      "Ġcost ly",
+      "Ġpartners hip",
+      "ĠBacter ia",
+      "Ġperoxid ation",
+      "st roke",
+      "ĠS av",
+      "ĠB W",
+      "Ġconn ects",
+      "Ġam ine",
+      "r il",
+      "Ġbat tle",
+      "ĠN otes",
+      "ĠPro vid",
+      "ĠInstit utional",
+      "Ġpro pri",
+      "f an",
+      "Ġp un",
+      "rom b",
+      "v ities",
+      "ĠC AM",
+      "ĠI sh",
+      "ĠH N",
+      "ĠRec omb",
+      "sc he",
+      "Ġsyn chrotron",
+      "ri k",
+      "syn aptic",
+      "ĠGeorg ia",
+      "? ?",
+      "C Y",
+      "Ġcorrespond ed",
+      "kin ase",
+      "ĠI TS",
+      "Ġpropos als",
+      "Ġbi oge",
+      "ĠE SR",
+      "ĠW en",
+      "ĠJ a",
+      "ĠSe vere",
+      "ĠAd en",
+      "ĠC CL",
+      "Ġse at",
+      "ĠK re",
+      "Ġhelp ing",
+      "Ġn ets",
+      "ĠL ep",
+      "hed ra",
+      "opo ulos",
+      "ĠB ak",
+      "ans as",
+      "Ġref rig",
+      "Ġubiquit ous",
+      "Ġmat ters",
+      "Ġsil icate",
+      "ĠLast ly",
+      "ĠThe ories",
+      "Ġag arose",
+      "big gr",
+      "trans ition",
+      "ĠDec omposition",
+      "b romo",
+      "Ġstake holders",
+      "ĠE E",
+      "On ly",
+      "ĠKen ya",
+      "Ġarg on",
+      "ĠIdentif ying",
+      "Ġtourn ament",
+      "cl ock",
+      "ĠCF U",
+      "ĠBehavi oral",
+      "Ġp od",
+      "Ġtaxon omy",
+      "ĠPro duct",
+      "ĠAl ong",
+      "Ġfamil ial",
+      "Ġdescript or",
+      "v ated",
+      "ĠVari ables",
+      "t p",
+      "Ġgood s",
+      "ĠA ST",
+      "ĠAn is",
+      "Ġspin or",
+      "at tention",
+      "Ġbas ket",
+      "Str uct",
+      "Ġimmunohist ochemical",
+      "eng ers",
+      "C AT",
+      "Ġtang ential",
+      "C ap",
+      "ĠP air",
+      "Ġvisco elastic",
+      "ĠAd s",
+      "Ġglycos ylation",
+      "Ġd ur",
+      "ĠMin imum",
+      "Ġrig idity",
+      "st ats",
+      "till ation",
+      "ĠDisc rim",
+      "ĠLeg end",
+      "Pre vious",
+      "fil m",
+      "Ġalumin ium",
+      "M icro",
+      "in ia",
+      "eg el",
+      "ĠSub cellular",
+      "Ġbottlen eck",
+      "Ġsy ll",
+      "ic le",
+      "Ġshe af",
+      "che ll",
+      "ex ample",
+      "ĠSe lected",
+      "Ġpred ators",
+      "Ġre per",
+      "Ġstr ugg",
+      "ĠM aria",
+      "ly l",
+      "L F",
+      "Ġexerc ises",
+      "ob ium",
+      "IL ITY",
+      "cor rected",
+      "Ġbenchmark s",
+      "ĠT ol",
+      "Ġinter cept",
+      "ĠCalc ulation",
+      "ĠIndones ia",
+      "Ġgli oblastoma",
+      "K M",
+      "ĠSup plemental",
+      "Ġciti zens",
+      "ad ren",
+      "Ġmultim odal",
+      "Ġmosquito es",
+      "iv a",
+      "ĠFind ings",
+      "ĠP ub",
+      "ĠMac roph",
+      "Ack nowledg",
+      "Ġbas ins",
+      "ex act",
+      "Ġgra des",
+      "Ġf ir",
+      "ig a",
+      "ĠPol ynomial",
+      "ĠLong itudinal",
+      "Ġsemicon ductors",
+      "T op",
+      "ip tera",
+      "Ġlack s",
+      "ro graph",
+      "Ġselec ts",
+      "Ġswe et",
+      "Ġb ac",
+      "Ġdown loaded",
+      "ap onic",
+      "ij k",
+      "ot onic",
+      "normal ized",
+      "ĠVari ability",
+      "di vision",
+      "ĠSu pers",
+      "il ab",
+      "H uman",
+      "Ġlept in",
+      "Ġosm otic",
+      "Ġh ur",
+      "ĠSing apore",
+      "ĠO PT",
+      "ĠSo viet",
+      "lit axel",
+      "ret aceous",
+      "ĠOn c",
+      "ĠI X",
+      "ul as",
+      "u ent",
+      "Ġlymph oid",
+      "T c",
+      "Ġrational e",
+      "L ayer",
+      "os ities",
+      "Ġdes ire",
+      "ĠAnn ual",
+      "ub a",
+      "ĠCompound s",
+      "Ġantif ungal",
+      "Ġcation ic",
+      "it ems",
+      "acter ium",
+      "amil ies",
+      "Ġelong ated",
+      "ĠMass achusetts",
+      "ĠIr ish",
+      "ass o",
+      "az o",
+      "ĠBur k",
+      "rob enius",
+      "Ġis instance",
+      "b ion",
+      "Ġgre edy",
+      "Ġnicot ine",
+      "Ġretrie ve",
+      "Ġsym pathetic",
+      "que e",
+      "Ġfol i",
+      "Ġsp utter",
+      "Ġgra ding",
+      "determ ined",
+      "Ġab norm",
+      "Ġman agers",
+      "Ġtop ical",
+      "Ġimm ig",
+      "ĠD NN",
+      "g tr",
+      "Ġdet ections",
+      "ĠOb esity",
+      "s uc",
+      "ĠSc he",
+      "Ġtr unk",
+      "Ġto ugh",
+      "ĠB N",
+      "Ġr u",
+      "ox if",
+      "Ġaim ing",
+      "ĠExt racellular",
+      "Ġhapl otype",
+      "D u",
+      "ĠD ing",
+      "ĠD ol",
+      "Ġhum id",
+      "b rom",
+      "Ġoff line",
+      "Comb ining",
+      "Ġpuls ar",
+      "Ġpar i",
+      "part ate",
+      "im ated",
+      "Ġwaters hed",
+      "acryl amide",
+      "ex ec",
+      "ĠCom posite",
+      "Ġdispers ive",
+      "Ġt ons",
+      "rom etry",
+      "ĠJ ud",
+      "az a",
+      "Ġchick ens",
+      "reg ister",
+      "n z",
+      "U til",
+      "ĠV es",
+      "e V",
+      "ĠR ule",
+      "sub stituted",
+      "Con v",
+      "qu ery",
+      "M ac",
+      "ĠT ar",
+      "im plies",
+      "ĠR ates",
+      "Ġr ins",
+      "Ġtimes cales",
+      "ĠCz ech",
+      "S uch",
+      "res timate",
+      "ĠM b",
+      "ĠFu j",
+      "ĠI MD",
+      "c it",
+      "Ġra ising",
+      ".... ....",
+      "h ome",
+      "as ted",
+      "Ġoc ta",
+      "Ġc admium",
+      "Ġps ori",
+      "role um",
+      "ĠSt ellar",
+      "ĠKin ase",
+      "ĠG ard",
+      "ie u",
+      "ĠMo S",
+      "M G",
+      "ĠG SH",
+      "Ġhaz ards",
+      "Ġn ice",
+      "he ating",
+      "Ġreproduc ible",
+      "gen esis",
+      "ĠIg M",
+      "Ġbe at",
+      "onucle ase",
+      "entral ized",
+      "ĠL é",
+      "Ġd ol",
+      "Ġdeep ly",
+      "rac tive",
+      "Ġgl ial",
+      "i ella",
+      "Ġinitial ized",
+      "ĠMethod ology",
+      "Ġbent hic",
+      "om i",
+      "ĠAl ter",
+      "Or dered",
+      "ĠL IN",
+      "Ġun ilateral",
+      "Ġcortic oster",
+      "L EN",
+      "Ġdil ute",
+      "Ġmetall oprotein",
+      "ab eth",
+      "amp ion",
+      "Ġmor al",
+      "ĠSi C",
+      "Ġquadr ature",
+      "Ġsediment ation",
+      "et e",
+      "ĠF rag",
+      "Ġpeak ed",
+      "Ġmitig ation",
+      "Ġsol di",
+      "Ġdoub ly",
+      "Ġellip so",
+      "Ġlnc RNAs",
+      "Ġâİ ¢",
+      "ĠS ame",
+      "ĠS ustain",
+      "ĠCap acity",
+      "Ġs omat",
+      "Ġtrans istor",
+      "Ġassay ed",
+      "ĠN ur",
+      "to ols",
+      "S ing",
+      "Ġlig ament",
+      "ate ver",
+      "Ġper ce",
+      "hen ce",
+      "U X",
+      "s ent",
+      "EG G",
+      "th ird",
+      "end ers",
+      "the oretic",
+      "Ġre wards",
+      "ut o",
+      "Ġinstall ation",
+      "ĠKine tic",
+      "ĠIn nov",
+      "ĠSol ving",
+      "ĠS ymmetry",
+      "Ġr amp",
+      "Ġneu ropathy",
+      "omer ization",
+      "Ġcat ech",
+      "P red",
+      "ĠB oh",
+      "EM ENT",
+      "Ġarm y",
+      "ĠYuk awa",
+      "Ġalign ments",
+      "ĠDepend ence",
+      "Ġen v",
+      "e an",
+      "s r",
+      "Ġinterp reting",
+      "eloc ity",
+      "Ġpsych ology",
+      "Ġbiofil ms",
+      "Ġeccentric ity",
+      "l ot",
+      "analy tic",
+      "Ġperiod icity",
+      "n ings",
+      "ĠK ent",
+      "fl ag",
+      "Ġm p",
+      "ĠN ich",
+      "hi re",
+      "Ġfl are",
+      "Ġcit rate",
+      "Ġp aste",
+      "Ġdele te",
+      "zym es",
+      "orient ation",
+      "ĠH Y",
+      "Ġcomm ands",
+      "Ġstri ke",
+      "s ymbol",
+      "ĠM ind",
+      "Ġoptim isation",
+      "Ġosteopor osis",
+      "ĠInf lammation",
+      "ĠIntellig ence",
+      "e h",
+      "ut um",
+      "Ġv ec",
+      "ell ation",
+      "ĠBl och",
+      "ĠMajor ana",
+      "en or",
+      "ĠN gu",
+      "Ġde uter",
+      "oped ia",
+      "Ġ utter",
+      "Ġrib osome",
+      "Ġact ors",
+      "elect ronic",
+      "é e",
+      "Ġfeat uring",
+      "ag le",
+      "Ġper in",
+      "ĠC ivil",
+      "Ġpred efined",
+      "l ag",
+      "ĠJ AK",
+      "j amin",
+      "in dividual",
+      "on c",
+      "Ġf ishing",
+      "di tive",
+      "N orm",
+      "ĠSc anning",
+      "van ishing",
+      "Ġc essation",
+      "ĠH ole",
+      "rib utes",
+      "I E",
+      "ĠM pc",
+      "weg ian",
+      "M a",
+      "Ġrevis ited",
+      "ĠPl us",
+      "abil ized",
+      "Ġsc anned",
+      "ĠEx change",
+      "Ġbrom ide",
+      "L ife",
+      "ot roph",
+      "AD S",
+      "âĭ ħ",
+      "Ġoper ative",
+      "ĠB ERT",
+      "Ġpl ume",
+      "Ġpo orer",
+      "Ġtro ut",
+      "Ġmicrotub ule",
+      "Ġphosph atidyl",
+      "radi us",
+      "ĠMus cle",
+      "Ġcarcin ogenesis",
+      "Ġsee ing",
+      "ucle in",
+      "f ollow",
+      "Ġsup plements",
+      "ol ars",
+      "spec ially",
+      "Ġcomple ting",
+      "Ġna ïve",
+      "ĠÏ ©",
+      "clero tic",
+      "D isc",
+      "ĠF estival",
+      "Ġcl ick",
+      "cl usive",
+      "Ġcatal ogue",
+      "Ġap ps",
+      "ĠS ED",
+      "Ġstack ed",
+      "Ġtun e",
+      "ĠDM EM",
+      "Ġaeros ols",
+      "Ġg ear",
+      "ant ine",
+      "ĠSt one",
+      "Ġpos itives",
+      "tri ang",
+      "prob ability",
+      "Ġdec oupling",
+      "ĠÍ ĵ",
+      "ĠV in",
+      "Ġsurv ived",
+      "Ġre plicated",
+      "ut rient",
+      "Ġtemper ate",
+      "Ġens embles",
+      "Ġmultic enter",
+      "Ġg aseous",
+      "ide a",
+      "class ification",
+      "ĠOut come",
+      "cl onal",
+      "Ġdiscontinu ity",
+      "Ġadvantage ous",
+      "Ġdist ricts",
+      "ĠI BM",
+      "inguish able",
+      "Ġcar s",
+      "c ult",
+      "en riched",
+      "arg in",
+      "nov ae",
+      "stead y",
+      "Ġbu y",
+      "pir ation",
+      "Ġpartition ed",
+      "Ġin ability",
+      "p q",
+      "Ġb ull",
+      "od end",
+      "Ġass istant",
+      "Ġlum en",
+      "Ġconver ting",
+      "P Y",
+      "z ol",
+      "ut ors",
+      "ĠNLR P",
+      "app ly",
+      "ĠBon ferroni",
+      "L s",
+      "Ġt ips",
+      "ĠL N",
+      "rol ase",
+      "Ġadv is",
+      "ĠMet ast",
+      "Ġsaliv a",
+      "Ġin habit",
+      "Ġr im",
+      "de bug",
+      "An y",
+      "Ġfor b",
+      "Ġvers atile",
+      "ĠComp act",
+      "v oc",
+      "ĠI so",
+      "ĠJ us",
+      "b odies",
+      "AR M",
+      "ĠGW AS",
+      "he tized",
+      "Ġmicrofluid ic",
+      "Ġacet onitrile",
+      "Ġin hom",
+      "Ġparen ch",
+      "Ġins ensitive",
+      "Ġag ency",
+      "po or",
+      "ĠAn gi",
+      "Ġappro ached",
+      "Ġem ulsion",
+      "Ġvol untary",
+      "ut t",
+      "ĠRec urrent",
+      "ric ulum",
+      "Ã ª",
+      "Ġt all",
+      "ĠDep th",
+      "Ġf f",
+      "ĠInc idence",
+      "Ġmanifest ation",
+      "Ġcomprom ised",
+      "i aceae",
+      "ĠM IT",
+      "otrans fer",
+      "ĠW D",
+      "m ov",
+      "ĠMan ual",
+      "M edi",
+      "Ġinterfer ing",
+      "ĠJacob i",
+      "K T",
+      "Ġs arcoma",
+      "Ġkid neys",
+      "Ġod or",
+      "Ġt i",
+      "yd ay",
+      "alth ough",
+      "vis ible",
+      "Ġd engue",
+      "ĠC AL",
+      "str at",
+      "ĠVari ations",
+      "in ib",
+      "comp onents",
+      "ĠT ob",
+      "ĠAnti oxidant",
+      "Í Ķ",
+      "Ġk iller",
+      "Ġsubt racted",
+      "ĠE vents",
+      "Ġim plements",
+      "ĠG AN",
+      "Ġprophyl axis",
+      "Ġno zz",
+      "Ġsm oothed",
+      "Ġdecay ing",
+      "ĠIniti ally",
+      "Ġuncom mon",
+      "Ġconduc tor",
+      "ĠW OR",
+      "av ity",
+      "ĠX ie",
+      "ĠAc et",
+      "Ġin e",
+      "ĠBe am",
+      "opol ymer",
+      "ĠX ML",
+      "ĠW ide",
+      "Ñ ĥ",
+      "Ġe jection",
+      "B MI",
+      "t c",
+      "ue z",
+      "Ġcereb ellar",
+      "Ġcatch ment",
+      "cox on",
+      "ĠSh annon",
+      "Ġcentral ity",
+      "Ġsaf ely",
+      "pro be",
+      "ĠLabor atories",
+      "Ġn c",
+      "Ġsp her",
+      "Ġprob ing",
+      "ĠLe v",
+      "Ġa f",
+      "ĠM ig",
+      "ĠV ascular",
+      "Ġprogram mes",
+      "Ġcontamin ants",
+      "sequ ent",
+      "Ġbond ed",
+      "integr ation",
+      "b os",
+      "ĠF ew",
+      "ĠIll inois",
+      "S he",
+      "W C",
+      "ĠG PIO",
+      "o C",
+      "ĠM aternal",
+      "erc etin",
+      "ĠMass ive",
+      "Ġen orm",
+      "img ur",
+      "Ġb idirectional",
+      "ĠG raphene",
+      "ins ky",
+      "ĠObs erve",
+      "Ġst ops",
+      "b io",
+      "ĠL ines",
+      "ĠG ill",
+      "Ġeigen vector",
+      "Sp ace",
+      "ĠM ining",
+      "Ġmel atonin",
+      "ĠS ET",
+      "onse qu",
+      "os cale",
+      "ĠR aw",
+      "Ġreview ers",
+      "Ġnan ofib",
+      "t aking",
+      "amm ad",
+      "Ġrecurs ion",
+      "us al",
+      "Ġpos itron",
+      "ĠN IH",
+      "ĠIN TER",
+      "ĠDoc ument",
+      "Ġconstant ly",
+      "Ġunderg one",
+      "Ġelect roweak",
+      "Ġiter atively",
+      "fol io",
+      "Ġsub family",
+      "Ġâİ ¥",
+      "P age",
+      "f erm",
+      "av ir",
+      "Ġag encies",
+      "Ġpol ys",
+      "ĠSqu are",
+      "ym m",
+      "Ġhydro gels",
+      "al most",
+      "ar ter",
+      "Ġan kle",
+      "Ġr ises",
+      "Ġmed ull",
+      "g ated",
+      "Ġmon onuclear",
+      "Ġdiscuss ing",
+      "Ġprof essor",
+      "trans formed",
+      "Ġcol ours",
+      "rag g",
+      "emic on",
+      "Ġsymmet rical",
+      "Ġplac ental",
+      "Ġl i",
+      "Ġstud io",
+      "sequ ences",
+      "Ġt am",
+      "ĠL ap",
+      "ĠCriter ia",
+      "Ġhapp ened",
+      "Ġantifer romagnetic",
+      "ĠHaus dorff",
+      "ĠCONCLUS IONS",
+      "H ER",
+      "V R",
+      "ĠK or",
+      "ĠA PO",
+      "Ġprot ecting",
+      "ĠS OL",
+      "ĠB uck",
+      "ph ia",
+      "ĠMul tim",
+      "on ine",
+      "uls ions",
+      "Ġg p",
+      "benz amide",
+      "ĠNAD PH",
+      "ĠOh io",
+      "ĠM EG",
+      "CO VID",
+      "Ġdisplac ed",
+      "ĠAb b",
+      "Ġbran ched",
+      "ĠN avy",
+      "ĠN rf",
+      "ĠO DE",
+      "ach i",
+      "ĠTrans ient",
+      "Ġcircum ference",
+      "Ġbe es",
+      "ir ation",
+      "Ġfac ulty",
+      "IG HT",
+      "ĠMetabol ism",
+      "M K",
+      "ĠIn equ",
+      "ĠQual itative",
+      "P BS",
+      "ter minus",
+      "k ary",
+      "o vian",
+      "ĠT Hz",
+      "ĠRel iability",
+      "f uran",
+      "Ġcor ners",
+      "Ġattack er",
+      "Ġmar riage",
+      "oprec ipitation",
+      "ĠC ry",
+      "ĠâĬ Ļ",
+      "Ġevol ves",
+      "Ġb an",
+      "Ġdi urnal",
+      "oun ce",
+      "Ġover w",
+      "ĠH off",
+      "Ġextr insic",
+      "am ps",
+      "UL AR",
+      "op her",
+      "Ġlight ing",
+      "Ġarchitect ural",
+      "hes ive",
+      "Ġsav ings",
+      "Ġglauc oma",
+      "oz oa",
+      "ĠO ption",
+      "cont roll",
+      "eck er",
+      "Ġoste ocl",
+      "Ġglyc ine",
+      "anal yses",
+      "ĠAl d",
+      "ĠS yd",
+      "ĠC x",
+      "Ġscal ars",
+      "Ġknow ing",
+      "Ġrem ember",
+      "ĠEmb ry",
+      "T EM",
+      "ĠB ran",
+      "F ORM",
+      "Ġsurv iving",
+      "Ġglob ular",
+      "Ġincl usive",
+      "sc hed",
+      "UT ION",
+      "Ġquadrup ole",
+      "ĠH ubbard",
+      "Ġax onal",
+      "ĠCos mic",
+      "Ġsl ots",
+      "ĠProced ure",
+      "ag in",
+      "ĠLo op",
+      "are r",
+      "Ġbut ter",
+      "Ġhist opathological",
+      "f usion",
+      "AN OVA",
+      "Ġclos ing",
+      "ĠL ord",
+      "ĠB is",
+      "ĠR AM",
+      "ID E",
+      "Ġj ournals",
+      "Ġmon keys",
+      "Ġatten uates",
+      "Ġsegment ed",
+      "T OF",
+      "o tional",
+      "pol ymer",
+      "ĠSha h",
+      "A kt",
+      "W r",
+      "l ov",
+      "Ġpolym orphic",
+      "Ġarrang ements",
+      "U F",
+      "l on",
+      "Ġdep ressed",
+      "NA T",
+      "ĠOper ation",
+      "Î ¹",
+      "ĠR an",
+      "â IJ",
+      "Ġthere after",
+      "Ġmyel oma",
+      "j or",
+      "Ã ¥",
+      "ĠW inter",
+      "pt osis",
+      "D ir",
+      "ver ty",
+      "ĠF inn",
+      "Ġorth olog",
+      "Ġmonoton ically",
+      "Ġtect onic",
+      "ĠG BM",
+      "ĠA O",
+      "Ġgener ative",
+      "C learly",
+      "Ġt ile",
+      "ĠR NN",
+      "Ġground s",
+      "Ga As",
+      "Ġbe e",
+      "ĠB oy",
+      "ĠTranscription al",
+      "ur in",
+      "ot om",
+      "Ġsinus oidal",
+      "ĠA y",
+      "ĠCl inic",
+      "ut orial",
+      "ĠAD C",
+      "ER IAL",
+      "c ation",
+      "ĠAD HD",
+      "cycl ohex",
+      "ĠHaw ai",
+      "ast om",
+      "Ġmorph ologies",
+      "Ġrod ents",
+      "Ġscal ability",
+      "R OS",
+      "a emia",
+      "Ġdecom pose",
+      "Ġpiv otal",
+      "Ġdiffus ivity",
+      "Ġco valent",
+      "ĠK D",
+      "ataly st",
+      "Ġold est",
+      "Ġsu itability",
+      "Ġw ants",
+      "if ts",
+      "ĠDist ributions",
+      "ĠQue en",
+      "l ich",
+      "Ġpar se",
+      "ĠM HD",
+      "Ġrec re",
+      "Ġhydrox ide",
+      "e um",
+      "Ġle v",
+      "Ġrefer ral",
+      "plan es",
+      "ĠEg ypt",
+      "Ġl enti",
+      "Ġtrans actions",
+      "Ġexp ense",
+      "Ġcy sts",
+      "Ġabs cess",
+      "Ġmicro RNAs",
+      "eff ectiveness",
+      "ĠDifferenti ation",
+      "Ġcer tif",
+      "c ience",
+      "ĠRE L",
+      "Ġread out",
+      "ĠQu asi",
+      "Ġround ed",
+      "ot ti",
+      "e fficients",
+      "Ġsynchron ized",
+      "Ġsil ico",
+      "Ġfore casts",
+      "Ġd μ",
+      "Ġex otic",
+      "ĠO CT",
+      "x b",
+      "Ġas ynchronous",
+      "ne z",
+      "chi at",
+      "Ġha emat",
+      "Ġfulf ill",
+      "ĠM ix",
+      "ib li",
+      "f m",
+      "Ġj ava",
+      "sol uble",
+      "Ġincomp ressible",
+      "âĨ ij",
+      "CD M",
+      "Ġdil ation",
+      "L YP",
+      "as hes",
+      "ĠS ports",
+      "Ġfund ament",
+      "ĠSa udi",
+      "Ġen roll",
+      "ĠNa OH",
+      "Ġcrust al",
+      "ĠInstr uments",
+      "Ġïģ ¡",
+      "Res ult",
+      "Ġpref erential",
+      "Ġsug ars",
+      "Ġdim ers",
+      "ĠEmerg ing",
+      "è re",
+      "diab etic",
+      "Ġstrengthen ing",
+      "ep i",
+      "ĠM eg",
+      "ĠY our",
+      "ĠSet ting",
+      "le z",
+      "ĠB ou",
+      "Ġhist ology",
+      "Ġol ive",
+      "ĠDis orders",
+      "Ġdistor ted",
+      "Ġcompet e",
+      "c ens",
+      "ĠA e",
+      "ĠG G",
+      "Ġquantif ying",
+      "Ġa ur",
+      "ĠW right",
+      "Ġsuperconduc tor",
+      "ed s",
+      "st alk",
+      "con cent",
+      "ĠLim ited",
+      "Ġst yles",
+      "des ign",
+      "ĠE llip",
+      "PL A",
+      "mog orov",
+      "ĠR idge",
+      "Ġrandom ization",
+      "a ft",
+      "ic ially",
+      "ĠBi otechnology",
+      "Ġseiz ure",
+      "K I",
+      "AV E",
+      "re ceptor",
+      "Ġgram mar",
+      "Ġcr ime",
+      "n ection",
+      "in ces",
+      "ĠCom pton",
+      "Ġventric le",
+      "Ġred istribution",
+      "yn aptic",
+      "Par ameter",
+      "N ormal",
+      "P ack",
+      "erm ann",
+      "ul ants",
+      "de generate",
+      "ĠNewton ian",
+      "Ġancest ral",
+      "ph rag",
+      "Ġimp ression",
+      "Ġnormal ize",
+      "Ġambig uous",
+      "Ġingredi ent",
+      "ĠCl aim",
+      "Ġcle aved",
+      "ĠAppro aches",
+      "ĠS PECT",
+      "cs v",
+      "ĠReve als",
+      "ĠW aves",
+      "Ġdwar fs",
+      "ĠProg ress",
+      "Ġa orta",
+      "Ġn ig",
+      "ĠAd ams",
+      "ĠM üller",
+      "ĠY ellow",
+      "ĠC ord",
+      "ĠPh ill",
+      "ĠF ormal",
+      "bes gue",
+      "ter min",
+      "r n",
+      "b n",
+      "k ine",
+      "r it",
+      "q i",
+      "ĠRout e",
+      "en ol",
+      "ĠA SC",
+      "ĠP u",
+      "m ill",
+      "um er",
+      "Ġsuper nova",
+      "i ative",
+      "diff erenti",
+      "Ġto lu",
+      "op us",
+      "R M",
+      "Ġpo verty",
+      "ĠX X",
+      "ĠïĤ ¶",
+      "ult ry",
+      "Op tim",
+      "Ġgl acial",
+      "ĠDis pers",
+      "Ġdifferenti ating",
+      "á ndez",
+      "pro ject",
+      "ĠEl iz",
+      "scal ing",
+      "ĠT oll",
+      "Ġnon empty",
+      "Ġpredic ate",
+      "Ġgyr us",
+      "min ute",
+      "â ĸ",
+      "ĠH ind",
+      "ĠL iving",
+      "V S",
+      "pri or",
+      "ĠVer tical",
+      "ark s",
+      "ĠS FR",
+      "ĠViet nam",
+      "comp are",
+      ">> >",
+      "Ġb anks",
+      "Ġse ptic",
+      "ĠB if",
+      "ĠE PS",
+      "ĠInt el",
+      "ĠDis order",
+      "P N",
+      "ĠN ord",
+      "tiv eness",
+      "Ġdr illing",
+      "ĠSub ject",
+      "enari o",
+      "Ġr ms",
+      "ph ones",
+      "h ang",
+      "ĠTechn ique",
+      "Ġcl ot",
+      "Ġinters ections",
+      "Ġan ions",
+      "ab ove",
+      "Ġcl ause",
+      "Ġgen u",
+      "oz o",
+      "rh iz",
+      "Ġlob es",
+      "ĠB ian",
+      "Ġexer ted",
+      "ure th",
+      "rom a",
+      "ĠCh arge",
+      "ĠSyn chron",
+      "Ġcont ing",
+      "othe rapeutic",
+      "gtr sim",
+      "ĠReson ance",
+      "ĠF al",
+      "und le",
+      "Ġdrop out",
+      "ĠPers pective",
+      "OL OG",
+      "at ches",
+      "ĠSequ ences",
+      "Cons idering",
+      "Ġprosp ects",
+      "Ġal iqu",
+      "Ġstr ata",
+      "Ġanalog s",
+      "Ġencour aged",
+      "ĠP ulmonary",
+      "Ġch im",
+      "ĠC FT",
+      "un ar",
+      "iz z",
+      "end ocrine",
+      "ĠC RE",
+      "ĠSt roke",
+      "âĩ Ĵ",
+      "up uncture",
+      "trans lational",
+      "ĠGr iff",
+      "ĠS ter",
+      "erg ed",
+      "ph rine",
+      "Ġl ivestock",
+      "ĠH ash",
+      "Ġdos ing",
+      "Ġplas mas",
+      "ĠCompar isons",
+      "Ġencour aging",
+      "Ġcompar atively",
+      "Ġcharacter isation",
+      "Ġasc ending",
+      "ĠF ixed",
+      "Ġveget able",
+      "es pecially",
+      "ĠL ange",
+      "ĠC oding",
+      "Ġverteb rate",
+      "F W",
+      "ĠOR F",
+      "ĠT ub",
+      "le e",
+      "Ġtim ely",
+      "E p",
+      "ĠâĪĴ âĪŀ",
+      "Ġlip osomes",
+      "Ġextrem al",
+      "ropol itan",
+      "ĠC ay",
+      "ĠB iod",
+      "o ulli",
+      "D ri",
+      "ĠR ats",
+      "Ġcent roid",
+      "osp in",
+      "rosp inal",
+      "Ġsol itons",
+      "por tive",
+      "ĠMc G",
+      "B bb",
+      "Ġpar affin",
+      "lec tively",
+      "Ġmetast able",
+      "Ġdissip ative",
+      "Ġassembl ages",
+      "Ġcol onic",
+      "Ġs ized",
+      "Ġcr yp",
+      "process or",
+      "ç ão",
+      "Ġacknowled ged",
+      "ĠUncertain ty",
+      "ĠInd ustrial",
+      "Ġunc ont",
+      "Ġref ere",
+      "ĠN itrogen",
+      "Ġlif ting",
+      "Ġfor get",
+      "Ġfeel ings",
+      "Ġdig its",
+      "Ġstrat ig",
+      "yp es",
+      "Ġcomposition al",
+      "Ġsupernat ants",
+      "Ġconflic ting",
+      "Ġdisadvant age",
+      "adel phia",
+      "P d",
+      "ĠC oupling",
+      "Ġexpendit ure",
+      "ik i",
+      "des cribed",
+      "ĠRN ase",
+      "ĠCon vex",
+      "ĠB ax",
+      "ung sten",
+      "Ġbo iling",
+      "Ġbas ement",
+      "ocardi al",
+      "hist ory",
+      "int on",
+      "trim ethyl",
+      "Ġgraft ing",
+      "ĠHard y",
+      "ĠFem ale",
+      "ĠF ollow",
+      "ĠE ST",
+      "tis tic",
+      "O pen",
+      "Ġref lux",
+      "ele ments",
+      "Ġpol ysaccharide",
+      "dim s",
+      "ac ency",
+      "Ġbi ore",
+      "cap ac",
+      "Ġoverex pressed",
+      "e ither",
+      "Ġl aid",
+      "Ġinc ision",
+      "Ġass ets",
+      "inf lammation",
+      "Ġreconstruc tions",
+      "Ġglomer ular",
+      "Ġcon vey",
+      "ĠCX CR",
+      "or o",
+      "Ġclass ifying",
+      "Ġcop e",
+      "Ġp d",
+      "lin ic",
+      "Ġch ord",
+      "ĠAg ing",
+      "Ġpal m",
+      "Ġpermit tivity",
+      "ĠRever se",
+      "Ġoff shore",
+      "Ġdoub t",
+      "im oto",
+      "ĠCol omb",
+      "Ġrod ent",
+      "ĠElect rochemical",
+      "ĠImpro vement",
+      "ines cent",
+      "ĠTr iton",
+      "Ġtransf usion",
+      "Ġlocom otion",
+      "Ġdanger ous",
+      "Ġwe ighed",
+      "ĠH SV",
+      "t echn",
+      "ĠDi agram",
+      "Ġpari etal",
+      "s ix",
+      "Ġtit les",
+      "yl on",
+      "Ġheur istics",
+      "Ġj aponic",
+      "Ġtransl ations",
+      "Ġtit ers",
+      "Ġw orms",
+      "ĠD PP",
+      "Ġcytos keleton",
+      "Medi ated",
+      "ari ance",
+      "the l",
+      "Ã ħ",
+      "ĠInf lammatory",
+      "Ġoscill ating",
+      "Ġavoid s",
+      "Def ine",
+      "ĠOlymp ics",
+      "og el",
+      "Ġhe me",
+      "Ġmic rop",
+      "Ġthreat s",
+      "Q CD",
+      "X RD",
+      "ĠC oupled",
+      "Ġl m",
+      "ĠHel ic",
+      "Ġdischarg ed",
+      "Ġro oted",
+      "Ġallevi ate",
+      "Ġcaus ality",
+      "ĠC row",
+      "ĠM ack",
+      "ĠAir port",
+      "Ġchem okine",
+      "Ġl l",
+      "ĠN ar",
+      "omy ces",
+      "eth oxyphenyl",
+      "ĠD aily",
+      "ĠFin land",
+      "Ġh orn",
+      "ĠO rient",
+      "Ġion ized",
+      "ĠY ears",
+      "Ġquas ipar",
+      "Ġper cutaneous",
+      "Ph ase",
+      "Ġfore ground",
+      "ĠA NAL",
+      "Ġincre ments",
+      "st an",
+      "Ġspec ulate",
+      "T X",
+      "Ġp ile",
+      "Ġd ic",
+      "ip y",
+      "wind ow",
+      "neut ral",
+      "ĠAtl as",
+      "ĠM TT",
+      "ĠN y",
+      "ĠV III",
+      "ĠFil ms",
+      "sing ular",
+      "rem ove",
+      "L ength",
+      "ĠRec e",
+      "wa it",
+      "Ġpurch ase",
+      "ĠWik ipedia",
+      "ĠL ars",
+      "Ġsynt actic",
+      "Ġactu ator",
+      "ĠAK T",
+      "ĠB ry",
+      "ĠRes ult",
+      "ĠVari ational",
+      "Ġjudg ment",
+      "J ECT",
+      "xim ab",
+      "Ġtrac ed",
+      "Ġcardiomy opathy",
+      "W N",
+      "ĠRod rig",
+      "b t",
+      "Ġb id",
+      "ac le",
+      "am ura",
+      "Ġep ic",
+      "Ġp uzz",
+      "ĠS ox",
+      "Ġinflu x",
+      "ÃŃ n",
+      "ulos keletal",
+      "D im",
+      "ĠS CC",
+      "ĠR AS",
+      "m r",
+      "U I",
+      "Ġj un",
+      "ĠSp earman",
+      "Ġfair ness",
+      "et z",
+      "ĠP PI",
+      "in ance",
+      "en ko",
+      "Ġgal act",
+      "ö m",
+      "Ġex ceptions",
+      "ĠC retaceous",
+      "M Y",
+      "Res p",
+      "Ġp ep",
+      "ĠOr d",
+      "ST E",
+      "Ġhel icity",
+      "Ġoffic er",
+      "T arget",
+      "ĠNor wegian",
+      "Ġdehyd ration",
+      "ĠSIR T",
+      "ĠRob inson",
+      "ĠBen chmark",
+      "v iral",
+      "Re al",
+      "Ġd oxorubicin",
+      "Ġestim ations",
+      "ĠCa uc",
+      "Ġaddi tives",
+      "m odes",
+      "ĠH end",
+      "Ġacceler ating",
+      "ĠG ordon",
+      "ĠMagn et",
+      "Ġgon ad",
+      "Ġpyro lysis",
+      "coh olic",
+      "ĠPK C",
+      "S AR",
+      "Ġw inding",
+      "ter ious",
+      "ĠMountain s",
+      "ĠS ymbol",
+      "ĠMat the",
+      "ĠSh in",
+      "S cript",
+      "r ug",
+      "Ġm W",
+      "ĠI SM",
+      "ĠN g",
+      "Ġapp oint",
+      "ĠA IDS",
+      "Ġpor ts",
+      "diff erential",
+      "ĠJ es",
+      "ĠN eed",
+      "Ġlens es",
+      "ĠAMP K",
+      "à ¤",
+      "le af",
+      "ĠB ron",
+      "Ġprof it",
+      "L ocal",
+      "ĠEx amination",
+      "ĠCh ief",
+      "Ġopin ions",
+      "ĠR ound",
+      "form ations",
+      "Ġcol linear",
+      "Ġdig ested",
+      "lass ical",
+      "erv ative",
+      "Ġce phal",
+      "Ġdisadvant ages",
+      "Ġïĥ ·",
+      "Ġsubt racting",
+      "Ġwe igh",
+      "B ound",
+      "D G",
+      "Ġinflu ential",
+      "Ġtox ins",
+      "ĠBen jamin",
+      "ĠNum bers",
+      "c rystal",
+      "Ġst ocks",
+      "ĠB our",
+      "ĠComp eting",
+      "Ġac qu",
+      "t RNA",
+      "ĠSep aration",
+      "Ġtag ged",
+      "Ġcon ject",
+      "ĠPr ince",
+      "Ġgra zing",
+      "Ġrele ases",
+      "ĠChall enge",
+      "ATP ase",
+      "Ġe mail",
+      "ins ically",
+      "ĠReg ulatory",
+      "M essage",
+      "Ġsl it",
+      "Ġpolyg on",
+      "Ġdoubl ing",
+      "Ġrece ivers",
+      "Ġtrack ed",
+      "Ġengine er",
+      "st ained",
+      "ĠD anish",
+      "sh ock",
+      "ĠM az",
+      "Ġco ugh",
+      "ĠImmun ohist",
+      "C onsequ",
+      "arm acy",
+      "Ġchem o",
+      "ĠM H",
+      "Ġemerg es",
+      "Ġanne aled",
+      "Ġhypot hesize",
+      "ĠTyp ically",
+      "ĠB ang",
+      "ĠP uls",
+      "Ġgir l",
+      "Ġherb iv",
+      "ĠAN N",
+      "Ġse ism",
+      "ĠCy tok",
+      "ĠThrough out",
+      "Ġadapt ations",
+      "l ang",
+      "Ġcl onal",
+      "um ulation",
+      "ĠUn iform",
+      "Ġh i",
+      "op ent",
+      "Ġbut ton",
+      "ten e",
+      "Ġprote asome",
+      "b red",
+      "ĠN elson",
+      "racycl ine",
+      "ĠD Y",
+      "Ġimmun oblot",
+      "pro l",
+      "Ġp ic",
+      "Ġcomp ilation",
+      "ĠDev ices",
+      "eterm ined",
+      "ĠFranc is",
+      "not ation",
+      "wr iting",
+      "ter ase",
+      "ĠSte phen",
+      "am el",
+      "ĠCh u",
+      "al one",
+      "Ġexha ust",
+      "re levant",
+      "ĠStr at",
+      "Ġparametri zation",
+      "ĠB ull",
+      "ĠRem ote",
+      "incre asing",
+      "Ġd d",
+      "âĢ °",
+      "yroid ism",
+      "il in",
+      "ĠH ip",
+      "IC A",
+      "ĠAp optosis",
+      "Ġmach ining",
+      "LD L",
+      "Ġg em",
+      "ĠF FT",
+      "ĠGu ang",
+      "Ġorigin ates",
+      "d at",
+      "c one",
+      "ĠAd oles",
+      "uc ci",
+      "av oid",
+      "ul pt",
+      "ur ium",
+      "Ġliter acy",
+      "Rec ent",
+      "av g",
+      "Ġinv ited",
+      "ĠPe ak",
+      "ĠEnter obacter",
+      "Ġaneurys m",
+      "ĠMor ris",
+      "ti da",
+      "ĠS ER",
+      "ĠMic hel",
+      "ĠI BD",
+      "ĠN G",
+      "Ġscar ce",
+      "we b",
+      "Ġexpress es",
+      "Ġsc hema",
+      "Ġless ons",
+      "Ġarg inine",
+      "Ġphot ographs",
+      "ĠNe urons",
+      "ĠATP ase",
+      "Ġf iller",
+      "rap ped",
+      "Ġrandom ness",
+      "Ġve ins",
+      "Ġwound s",
+      "ĠA part",
+      "Ġr acial",
+      "Ġnot eworthy",
+      "Ġremov es",
+      "Ġgangl ion",
+      "Ġlamin ar",
+      "ĠS SR",
+      "Ġpol ysaccharides",
+      "Ġbu f",
+      "Ġendot helium",
+      "ĠC AS",
+      "ĠGol gi",
+      "Ġinher itance",
+      "is ite",
+      "CO MP",
+      "Ġp t",
+      "Ġmes hes",
+      "Ġtherap eutics",
+      "O ST",
+      "olin ergic",
+      "U G",
+      "squ ared",
+      "Ġdeg rade",
+      "u um",
+      "Ġret rosp",
+      "L oc",
+      "ĠJ NK",
+      "O ptions",
+      "Ġins ulating",
+      "Ġspec ifies",
+      "Ġo ven",
+      "y y",
+      "ĠCon ver",
+      "Ġdisapp eared",
+      "ĠProgn ostic",
+      "ĠNgu yen",
+      "Ġperipher y",
+      "b ank",
+      "Ġim id",
+      "Ġassign ing",
+      "ĠM ess",
+      "prop an",
+      "i oma",
+      "oly b",
+      "Ġepit ope",
+      "Ġemit ting",
+      "D IR",
+      "yn c",
+      "Ġimpair ments",
+      "ĠMic roscopy",
+      "ĠFW HM",
+      "g ray",
+      "Ġf ing",
+      "uc ial",
+      "plement ed",
+      "e as",
+      "est amp",
+      "Ġcre st",
+      "ĠM os",
+      "Ġneut rons",
+      "Ġbro th",
+      "Ġhead ache",
+      "onge vity",
+      "Ġre ass",
+      "ĠP SF",
+      "ĠB uch",
+      "vis or",
+      "Ġden oting",
+      "in teger",
+      "ou in",
+      "eff icacy",
+      "Ġglut amine",
+      "Ġpick ed",
+      "ĠCamp bell",
+      "ĠK ernel",
+      "Ġsh ips",
+      "l t",
+      "ond yl",
+      "Ġcre di",
+      "Ġpepti d",
+      "ĠEst abl",
+      "b ons",
+      "Ġag gl",
+      "US E",
+      "sup p",
+      "ups ilon",
+      "character ized",
+      "ishe ries",
+      "M ay",
+      "AR C",
+      "Ġro ads",
+      "Ġdepar ture",
+      "ĠMA X",
+      "ĠT RA",
+      "im od",
+      "ĠAl ber",
+      "Ġterm inated",
+      "öl der",
+      "S calar",
+      "h ash",
+      "ĠM SS",
+      "Ġsmooth ness",
+      "Ġres emble",
+      "ĠEff ectiveness",
+      "r x",
+      "ĠE ye",
+      "Ġfa ecal",
+      "Ã ¾",
+      "ĠClostr idium",
+      "ach ine",
+      "ĠBD NF",
+      "Ġc ab",
+      "ĠW ong",
+      "ĠDoug las",
+      "Ġre perfusion",
+      "ĠX i",
+      "Ġconf used",
+      "ĠPhil adelphia",
+      "Ġap ple",
+      "Ġi le",
+      "Ġfav ored",
+      "Ġpl aques",
+      "Ġtri vially",
+      "ĠTyp ical",
+      "Ġcentral ized",
+      "ĠFace book",
+      "Ġnorthe ast",
+      "Ġnorm ality",
+      "ĠT b",
+      "Ġap t",
+      "Ġfac et",
+      "ĠRen al",
+      "cl k",
+      "Ġlig ation",
+      "iff erenti",
+      "Ġput ting",
+      "Ġintr ig",
+      "w alled",
+      "E t",
+      "ĠC ow",
+      "ĠN ations",
+      "Ġcamp us",
+      "ĠKine tics",
+      "ĠMex ican",
+      "ER K",
+      "Ġlat itudes",
+      "ĠR oll",
+      "ĠQ D",
+      "adap tive",
+      "Ġquenc hed",
+      "Ġf ram",
+      "Q i",
+      "Ġt ongue",
+      "ed es",
+      "Ġasc orb",
+      "ĠGluc ose",
+      "our i",
+      "Ġdef eated",
+      "ophil us",
+      "ral ateral",
+      "x rightarrow",
+      "ĠJ up",
+      "ax es",
+      "eg er",
+      "MI T",
+      "ĠM ember",
+      "ĠN u",
+      "Ġtransl oc",
+      "ĠFlu x",
+      "ĠColor ado",
+      "Ġre lying",
+      "at rol",
+      "Ġcontras ts",
+      "cent age",
+      "Ġleuk ocyte",
+      "Ġcoinc idence",
+      "Ġcontrac tions",
+      "og a",
+      "AN N",
+      "ĠAbs orption",
+      "Ret urn",
+      "rep rene",
+      "ba um",
+      "tra umatic",
+      "inc ial",
+      "Ġaut ophag",
+      "Ġalgorithm ic",
+      "rim p",
+      "Ġdiv ides",
+      "ĠR ose",
+      "ĠE ric",
+      "Ġadd iction",
+      "pl ification",
+      "Ġdiff usive",
+      "ĠVehic le",
+      "en erate",
+      "ti sing",
+      "Ġstar vation",
+      "abs orption",
+      "ĠA ra",
+      "Ġgra v",
+      "ĠSub unit",
+      "Ġam ide",
+      "Ġenh ancer",
+      "Ġmer id",
+      "erm ost",
+      "Ġal gal",
+      "ĠQue ens",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠ",
+      "Ġjud ge",
+      "ĠGreen land",
+      "b race",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠĠĠĠĠĠ",
+      "Ġhyper gly",
+      "Ġemerg ent",
+      "F isher",
+      "ĠL as",
+      "Ġsex es",
+      "S ep",
+      "Ġph rases",
+      "ĠSequ ential",
+      "ink i",
+      "Ġaxi oms",
+      "stud y",
+      "Ġt iny",
+      "Ġc d",
+      "cataly zed",
+      "as aki",
+      "ĠW R",
+      "ĠMin imal",
+      "Ġsub cellular",
+      "Ġphosph o",
+      "ES I",
+      "Ġv ow",
+      "Ġsup plies",
+      "oper and",
+      "F ix",
+      "an ian",
+      "wr iter",
+      "âĪ ¶",
+      "Ġwin ner",
+      "ĠP ID",
+      "ĠLe besgue",
+      "Ġsimpl ification",
+      "ĠRelationship s",
+      "Ġautom ata",
+      "ĠCont ribution",
+      "Ġhered itary",
+      "err in",
+      "ĠB LAST",
+      "ae a",
+      "yle th",
+      "ĠT c",
+      "ade h",
+      "adj uvant",
+      "W ave",
+      "c ounter",
+      "ĠG upta",
+      "ĠG hana",
+      "C ho",
+      "Ġour selves",
+      "Ġeven ly",
+      "lym ph",
+      "Ġcereb ellum",
+      "Ġcopol ymers",
+      "mod ular",
+      "Ġhard er",
+      "Ġp lease",
+      "ĠP SD",
+      "Ġlim bs",
+      "Ġexplo itation",
+      "ir y",
+      "Ġperiodon tal",
+      "AT CH",
+      "Ġmal icious",
+      "ĠSl ov",
+      "H Y",
+      "Consequ ently",
+      "ore n",
+      "ĠP are",
+      "ag ine",
+      "ĠRO I",
+      "ĠWh ich",
+      "ĠN ative",
+      "am en",
+      "resh ape",
+      "opl ankton",
+      "Ġartif act",
+      "Ġrh in",
+      "g pu",
+      "Ġund et",
+      "Ġspor adic",
+      "Ġor ally",
+      "Ġstep wise",
+      "ĠCoh ort",
+      "Ġr hod",
+      "c yt",
+      "Ġi err",
+      "Ġmot ors",
+      "ĠIg E",
+      "calc ulated",
+      "ĠChampionship s",
+      "p el",
+      "ĠF err",
+      "Ġis ometric",
+      "n utrition",
+      "Ġuns aturated",
+      "Ġd oll",
+      "ĠR MSE",
+      "Ġsol itary",
+      "approxim ation",
+      "Ġreper to",
+      "s ight",
+      "Ġc ranial",
+      "il ical",
+      "ĠO st",
+      "o ul",
+      "Ġd g",
+      "ĠPro ceed",
+      "Ġmill ing",
+      "s z",
+      "Ġmineral ization",
+      "Ġcig arette",
+      "Ġp orph",
+      "Ġsp ons",
+      "ĠGre ece",
+      "ip ore",
+      "ac cept",
+      "ĠPT SD",
+      "Å «",
+      "Ġc ipher",
+      "Ġfunctional ized",
+      "P oly",
+      "Ġab d",
+      "fl ight",
+      "ĠSyd ney",
+      "Ġdis aster",
+      "ĠH aving",
+      "Ġdies el",
+      "ĠG reg",
+      "Ġsp ans",
+      "ĠSe asonal",
+      "ST EM",
+      "i err",
+      "ĠI B",
+      "Ġle mm",
+      "an um",
+      "ĠB ottom",
+      "Ġse al",
+      "bo ost",
+      "Ġleg end",
+      "b ing",
+      "ab is",
+      "Ġch itin",
+      "Ġmaxim ally",
+      "Ġimmunosup pressive",
+      "âĪĴ âĪĴ",
+      "Ġabol ished",
+      "ig e",
+      "Ġes ophag",
+      "Ġlas ted",
+      "Ġcarbohyd rates",
+      "Ġch ips",
+      "ĠFern and",
+      "f ar",
+      "ĠPo ints",
+      "cal ation",
+      "ĠReg ions",
+      "CH K",
+      "ver atrol",
+      "tr uth",
+      "Ġst range",
+      "Int erest",
+      "s ho",
+      "ĠInd uc",
+      "Ġmig raine",
+      "ĠV ac",
+      "op hores",
+      "Ġerr one",
+      "scripts ize",
+      "ĠNeut ron",
+      "Ġindist inguishable",
+      "ist ine",
+      "Ġhel per",
+      "spec ified",
+      "Ġju ice",
+      "ox al",
+      "ĠJ ung",
+      "Ġmag azine",
+      "Ġtele phone",
+      "ĠPh yt",
+      "Ġ um",
+      "ĠAvail ability",
+      "ĠT ropical",
+      "ĠC ases",
+      "Ġdesc end",
+      "H ar",
+      "âĪ Ĺ",
+      "ĠâĨ ĵ",
+      "K s",
+      "Ġ ê",
+      "ol uble",
+      "Ġch ampionship",
+      "ĠMove ment",
+      "ĠX Y",
+      "kappa B",
+      "year s",
+      "m emb",
+      "qu ine",
+      "Ġlet ting",
+      "Ġbig gest",
+      "Ġc ards",
+      "Ġbi otin",
+      "ĠA ur",
+      "mod al",
+      "Ġvacc inated",
+      "Ġtransl ates",
+      "ĠP AC",
+      "ll i",
+      "re onine",
+      "Ġcur cumin",
+      "ĠCon struct",
+      "Ġconv inc",
+      "ĠN at",
+      "Ġam ygdala",
+      "Ġprot r",
+      "ĠSing ular",
+      "ĠCont act",
+      "k ind",
+      "ĠD aw",
+      "og roup",
+      "ĠK Cl",
+      "Ġhy gi",
+      "eren ced",
+      "Ġsurvey ed",
+      "ĠM ull",
+      "est hetic",
+      "Ġms g",
+      "ĠRe quire",
+      "Ġdistor tions",
+      "Cont rol",
+      "B ERT",
+      "Ġaut onomic",
+      "Ġhorm onal",
+      "Ġstri ps",
+      "Ġt rophic",
+      "if ting",
+      "op od",
+      "ĠSp ontaneous",
+      "Ġlog s",
+      "O PT",
+      "ĠM ot",
+      "ĠG mb",
+      "ah aran",
+      "ĠP OL",
+      "Ġvis ceral",
+      "bl ocks",
+      "Ġsit ting",
+      "Ġs ine",
+      "Ġonc ogenic",
+      "ERR Q",
+      "quin one",
+      "Ġsmart phone",
+      "ĠTan z",
+      "lact am",
+      "ĠSignific ance",
+      "Ġe u",
+      "ĠI SS",
+      "ĠTr ig",
+      "ĠM aj",
+      "ting ale",
+      "Ġdil at",
+      "enn es",
+      "ĠBelg ium",
+      "le v",
+      "ĠCon tr",
+      "ĠGal ois",
+      "ĠComb ination",
+      "ĠTh i",
+      "ĠAust ria",
+      "P rom",
+      "Ġelic it",
+      "bi osis",
+      "Ġlymph atic",
+      "ĠMur ray",
+      "ĠX PS",
+      "Ġcon g",
+      "sc reen",
+      "ti de",
+      "am oyl",
+      "ĠMc D",
+      "Ġreti red",
+      "m ixed",
+      "EL D",
+      "ĠM aps",
+      "ĠV E",
+      "cess ion",
+      "num er",
+      "id ated",
+      "ĠB ishop",
+      "Ġneon ates",
+      "Ġlands l",
+      "ĠFrac tional",
+      "Ġspec ifying",
+      "ĠJ r",
+      "Ġnanow ire",
+      "Ġconsult ation",
+      "l anguage",
+      "Ġp ricing",
+      "ĠLimit ations",
+      "ĠP ediatric",
+      "ĠD imension",
+      "Ġprepar ing",
+      "L ag",
+      "seg ment",
+      "Ġsp end",
+      "at he",
+      "Ġwe ap",
+      "ĠJ os",
+      "tex tit",
+      "output s",
+      "ord ering",
+      "Ġplac enta",
+      "ation ally",
+      "ĠK un",
+      "Ġout standing",
+      "Ġthickness es",
+      "ĠCh IP",
+      "de oxy",
+      "ĠZ o",
+      "ĠDevelop ing",
+      "Ġstring ent",
+      "i ency",
+      "per se",
+      "Ġp end",
+      "ĠDevelopment al",
+      "Ġex tern",
+      "Ġinver ter",
+      "ĠD API",
+      "lec tivity",
+      "Ġtable ts",
+      "Ġprog ester",
+      "Ġïģ Ń",
+      "Ġansw ered",
+      "ent ary",
+      "OR S",
+      "Ġd ir",
+      "Ġdele terious",
+      "Ġdop aminergic",
+      "R andom",
+      "dis s",
+      "Ġmonol ayers",
+      "Ġinteg rand",
+      "ĠComp onents",
+      "ĠP erc",
+      "ag it",
+      "AR N",
+      "es ophageal",
+      "iv an",
+      "ne ider",
+      "ĠStar ting",
+      "P ORT",
+      "y ellow",
+      "Ġreg isters",
+      "pair s",
+      "Ġethn icity",
+      "Ġb oy",
+      "au ti",
+      "Ġchrom ium",
+      "P OS",
+      "v ature",
+      "ay ashi",
+      "Ġin appropriate",
+      "ĠS NA",
+      "D omain",
+      "ĠP rice",
+      "Ġmac ular",
+      "Ġover load",
+      "ĠUn ified",
+      "Ġatt ach",
+      "ĠScot tish",
+      "m aps",
+      "ag l",
+      "em i",
+      "Ġse am",
+      "ĠAnal og",
+      "d ated",
+      "u o",
+      "Ġpl ated",
+      "Ġass et",
+      "Ġsc reens",
+      "Ġspur ious",
+      "B esides",
+      "Ġbas elines",
+      "head s",
+      "Ġco at",
+      "ĠRem oval",
+      "Ġinfinites imal",
+      "ĠTrans formation",
+      "Ġcomm ens",
+      "Flo at",
+      "A UC",
+      "ĠL ay",
+      "Ġint ron",
+      "ĠDet ecting",
+      "ĠHere in",
+      "ĠAssoci ations",
+      "Ġprogester one",
+      "B acteria",
+      "Ġs entiment",
+      "ĠPhen omen",
+      "m atter",
+      "Ġcylind ers",
+      "Ġtolu ene",
+      "Ġspati otemporal",
+      "Ġland ing",
+      "ĠCoron avirus",
+      "ĠBer ry",
+      "ĠB ragg",
+      "Ġreg istry",
+      "Ġenthal py",
+      "tic a",
+      "raz ine",
+      "Ġc argo",
+      "ot ation",
+      "Ġcontrad icts",
+      "Ġpestic ides",
+      "ĠF ischer",
+      "Ġmechan ically",
+      "ĠInter fer",
+      "ĠC yp",
+      "ĠK as",
+      "Ġmet res",
+      "Ġanti retroviral",
+      "Ġtra vers",
+      "se lection",
+      "ĠW A",
+      "Ġdouble t",
+      "m eta",
+      "EN TR",
+      "son ic",
+      "Ġmark ing",
+      "ĠO verex",
+      "Ġpy ruvate",
+      "Ġextr usion",
+      "Ġin gestion",
+      "Ġcoc aine",
+      "ĠF ellow",
+      "CN Ts",
+      "B G",
+      "ĠMorph ological",
+      "Ġdef ence",
+      "ĠY osh",
+      "mit ter",
+      "rystall ization",
+      "STR ACT",
+      "Ġinflamm asome",
+      "ĠG d",
+      "Ġsh aft",
+      "Ġerup tion",
+      "ĠOx ide",
+      "if olds",
+      "ĠG am",
+      "ĠG ap",
+      "com mand",
+      "ĠIg A",
+      "Ġshorten ing",
+      "assemb led",
+      "is opropyl",
+      "Ġal umina",
+      "ĠAT M",
+      "Ġc t",
+      "Ġspin ning",
+      "ĠPet sc",
+      "pref ix",
+      "Ġperpet uity",
+      "P RE",
+      "Ġfr uct",
+      "G Hz",
+      "el ike",
+      "en yl",
+      "Ġwhere in",
+      "U K",
+      "vis ual",
+      "lipid emia",
+      "re duction",
+      "an in",
+      "ol as",
+      "Ġam plic",
+      "ĠS AT",
+      "Ġmod ulator",
+      "for th",
+      "r l",
+      "Ġcre w",
+      "Ġi P",
+      "Ġx i",
+      "AD D",
+      "ĠAlex and",
+      "const rained",
+      "r atory",
+      "Ġk W",
+      "ĠMD R",
+      "Ġlnc RNA",
+      "M ill",
+      "ĠMg O",
+      "circ uit",
+      "Ġpersonal ized",
+      "ĠOper ator",
+      "st ock",
+      "ĠP SA",
+      "ens able",
+      "Ġle an",
+      "y ield",
+      "Ġop acity",
+      "ĠComm ons",
+      "Ġsum med",
+      "uck er",
+      "ec ke",
+      "ep ithelial",
+      "Ġas king",
+      "ues e",
+      "ĠFl av",
+      "Ġl actic",
+      "Ġl ubric",
+      "Ġis n",
+      "reg ions",
+      "sup port",
+      "Bel ow",
+      "ĠN om",
+      "Ġhy al",
+      "ik h",
+      "b an",
+      "ĠB G",
+      "rom eter",
+      "ind ic",
+      "oph aryngeal",
+      "IT ION",
+      "ĠProp agation",
+      "ĠPl ace",
+      "ĠCirc uit",
+      "ĠCO L",
+      "G reen",
+      "I r",
+      "l av",
+      "Ġd S",
+      "ĠM oment",
+      "Ġinduc ible",
+      "Ġdischarg es",
+      "hab di",
+      "ĠExper ience",
+      "Ġs g",
+      "Ġout ward",
+      "Ġport able",
+      "ĠOper ators",
+      "A v",
+      "ĠD Q",
+      "ost atin",
+      "Ġeosin ophil",
+      "Ġstri atum",
+      "ĠCons ensus",
+      "Ġim perfect",
+      "NO T",
+      "ĠDem ocratic",
+      "; ;",
+      "B ody",
+      "di i",
+      "H o",
+      "ĠRail way",
+      "ĠUg anda",
+      "Ġunp aired",
+      "friend ly",
+      "Ġrepro gramming",
+      "Altern ative",
+      "R G",
+      "im et",
+      "ene z",
+      "ĠHyp othesis",
+      "Ġt on",
+      "ĠCom bin",
+      "ĠDel ivery",
+      "L ast",
+      "Ġown ers",
+      "raz ole",
+      "ĠK ob",
+      "Ġform ats",
+      "Ġpoly clonal",
+      "Ġidentif ier",
+      "IL L",
+      "Ġsurge on",
+      "Ġpost p",
+      "ĠGener ative",
+      "ĠM all",
+      "ab c",
+      "ĠH az",
+      "Ġsmooth ly",
+      "Ġcrystall ographic",
+      "ĠF DA",
+      "Ġcoex istence",
+      "ion ized",
+      "Ġcomp iler",
+      "ĠAr ter",
+      "Ġappear ances",
+      "amilton ian",
+      "Ġencaps ulated",
+      "ati a",
+      "w i",
+      "re b",
+      "Ġwa fer",
+      "ub s",
+      "ĠU E",
+      "ĠGS K",
+      "Ġv iv",
+      "Ġflood ing",
+      "ĠG yr",
+      "Ġst ably",
+      "Ġdis locations",
+      "Ġes cap",
+      "ĠPhys iological",
+      "tid al",
+      "ym e",
+      "ĠMax im",
+      "iter ator",
+      "ord ant",
+      "Ġatten tional",
+      "Ġcataly zed",
+      "ĠTr yp",
+      "P IN",
+      "ĠCor relations",
+      "Ġhyd rological",
+      "Ġn ose",
+      "ex port",
+      "Ġde xt",
+      "ĠBen ef",
+      "ĠBios ystems",
+      "ĠP ars",
+      "Ġread ings",
+      "Ġinstrument ation",
+      "ĠI Q",
+      "R IC",
+      "Ġgra fts",
+      "over s",
+      "ĠMed ic",
+      "Ġmon od",
+      "Ġuniform ity",
+      "ĠAT LAS",
+      "Ġmask ed",
+      "R i",
+      "ĠPhys ic",
+      "Ġim posing",
+      "ĠPar ad",
+      "ime tic",
+      "Ġdemand ing",
+      "un ks",
+      "Ġfol ds",
+      "ĠAn c",
+      "Ġvol atility",
+      "Ġbring ing",
+      "ac il",
+      "ĠN MDA",
+      "re duced",
+      "ti i",
+      "Ġnorth west",
+      "ĠB essel",
+      "ven tions",
+      "Ġconsol idation",
+      "Me ier",
+      "Ġmicro f",
+      "Ġqual ified",
+      "Ġins ignificant",
+      "ĠMorph ology",
+      "Ġpoint wise",
+      "Ġlear ns",
+      "Ġgu ard",
+      "CH ECK",
+      "phon on",
+      "ĠEnhance ment",
+      "Ġz onal",
+      "ER G",
+      "St art",
+      "Ġhistor ic",
+      "ĠP ure",
+      "ĠGmb H",
+      "g lu",
+      "Ġpattern ing",
+      "Ġstic k",
+      "umin osity",
+      "D ataset",
+      "Ġover ride",
+      "ĠSte el",
+      "Ġfu els",
+      "m echanical",
+      "Ġaut ologous",
+      "Ġdepart ments",
+      "ĠB lo",
+      "Ġim ported",
+      "Ġrestric tive",
+      "e igen",
+      "ĠR ome",
+      "ĠÌ Ĭ",
+      "Ġepit opes",
+      "Ġlab elling",
+      "Ġown ership",
+      "ĠE specially",
+      "Ġco ffee",
+      "ĠGR B",
+      "H ead",
+      "ĠV ent",
+      "es are",
+      "ĠPar ticles",
+      "UN CTION",
+      "j j",
+      "u ents",
+      "el ic",
+      "ĠT at",
+      "ĠF le",
+      "Ġg ating",
+      "Ġref uge",
+      "Ad ditionally",
+      "Ġrh s",
+      "Ġmay be",
+      "ĠF ang",
+      "Ġad vent",
+      "otransfer ase",
+      "sh ould",
+      "Ġprote omic",
+      "Ġleg itim",
+      "PER IM",
+      "ĠG iant",
+      "Ġgraph ics",
+      "onom ical",
+      "sc atter",
+      "Ġsugges tive",
+      "pl ots",
+      "Ġmulti drug",
+      "Ġabsor ber",
+      "X S",
+      "cons uming",
+      "Ġsustain ability",
+      "op re",
+      "f ix",
+      "Ġvol cano",
+      "ĠTyp es",
+      "ĠCre ate",
+      "Ġcho oses",
+      "Ġstir ring",
+      "Ġsurge ons",
+      "d S",
+      "Ġcharacter izes",
+      "Ġadjust ments",
+      "text tt",
+      "et ra",
+      "Ġclass ifications",
+      "sp ots",
+      "ĠâĻ ¯",
+      "ere x",
+      "de hyd",
+      "ĠBr ig",
+      "ĠSuper conduc",
+      "Ġgran ts",
+      "ĠC en",
+      "ĠY in",
+      "ĠRe actions",
+      "des cription",
+      "trans cription",
+      "import ant",
+      "Ġhemod ynamic",
+      "ĠY i",
+      "ĠGold en",
+      "k k",
+      "al b",
+      "Ġro oms",
+      "Ġseg reg",
+      "Ġsumm ing",
+      "Ġsuccess ion",
+      "Ġfollic ular",
+      "Ġtack le",
+      "D own",
+      "Ġevalu ates",
+      "atic a",
+      "ann ual",
+      "ĠAl bert",
+      "Ġt al",
+      "orb ital",
+      "f ted",
+      "vari ables",
+      "Ġwet land",
+      "outhe astern",
+      "M EM",
+      "ĠBr ill",
+      "ĠS odium",
+      "ĠAlex a",
+      "um ed",
+      "BU G",
+      "ar ine",
+      "Ġre venue",
+      "habdi tis",
+      "Ġdiss ol",
+      "am plitude",
+      "Ġar tists",
+      "Ġnormal ised",
+      "Ġfluct uating",
+      "Ġas par",
+      "ĠF i",
+      "ol ates",
+      "isp anic",
+      "Ġacet ylation",
+      "ĠConcent ration",
+      "Ġth ro",
+      "sh ots",
+      "Ġnarr ative",
+      "ĠWa als",
+      "am monium",
+      "ure au",
+      "-------- ----",
+      "Ġresearc hes",
+      "Ġbab y",
+      "Ġshar ply",
+      "Ù Ħ",
+      "ĠC el",
+      "C X",
+      "um inal",
+      "Ġgerm line",
+      "ĠTransform er",
+      "p seud",
+      "H G",
+      "K a",
+      "ĠS MC",
+      "ĠN utrition",
+      "Ġb arc",
+      "ĠW rite",
+      "Ġprote ases",
+      "Ġswe ep",
+      "ĠKol mogorov",
+      "m orph",
+      "in ducible",
+      "Ġexc iting",
+      "le in",
+      "ĠH ass",
+      "Ġproduc tive",
+      "mes h",
+      "ĠC MS",
+      "Ġhe avier",
+      "Ġmeet ings",
+      "ĠCop per",
+      "Ġvirt ue",
+      "as ant",
+      "ĠD EN",
+      "Ġinherent ly",
+      "ri o",
+      "Ġhous ed",
+      "Ġintra operative",
+      "Ġc rown",
+      "con ditions",
+      "AN G",
+      "YS IS",
+      "im an",
+      "Ġnm ol",
+      "ĠRetrie val",
+      "al gae",
+      "Ġk appa",
+      "de ep",
+      "in ence",
+      "ĠC arcinoma",
+      "Ġchromat ographic",
+      "Ġas cribed",
+      "Ġle verage",
+      "ĠK K",
+      "omy el",
+      "p et",
+      "ĠN J",
+      "com m",
+      "Ġann ually",
+      "g ran",
+      "Ġa val",
+      "ĠN ish",
+      "Ġev ac",
+      "Ġmulti f",
+      "Ġfund s",
+      "enn y",
+      "ĠM ong",
+      "ĠEx ception",
+      "path s",
+      "ym en",
+      "h pp",
+      "Ġrestric ting",
+      "s aturated",
+      "â Ļ",
+      "Ġlear ners",
+      "ĠLank a",
+      "in ities",
+      "ĠG DP",
+      "Ġspec iation",
+      "Ġens ured",
+      "Ġneutral izing",
+      "Ġball oon",
+      "Compar ison",
+      "ĠCal ibration",
+      "ĠInflu enza",
+      "Ġvap our",
+      "X A",
+      "t racking",
+      "ĠI CD",
+      "fluor o",
+      "ĠDam age",
+      "Ġp ra",
+      "Ġcon ceived",
+      "ĠCosm ological",
+      "Ġlo ose",
+      "inos itol",
+      "ĠCliff ord",
+      "ow a",
+      "Ġoffset s",
+      "doc ument",
+      "Ġenorm ous",
+      "Ġphoto electron",
+      "rec ord",
+      "estic ular",
+      "Ġvoc als",
+      "Ġconscious ness",
+      "Ġtre m",
+      "Ġlandsc apes",
+      "ĠFund amental",
+      "teb rate",
+      "Ġverteb ral",
+      "Ġregener ative",
+      "Ġtro posp",
+      "In tegr",
+      "Ġassoci ates",
+      "ov ed",
+      "uss ed",
+      "aw s",
+      "ĠS ide",
+      "Ġinter connected",
+      "Ġsuper family",
+      "ĠCo ok",
+      "load er",
+      "Ġpy thon",
+      "ĠC ounter",
+      "bo oks",
+      "Ġïģ ²",
+      "bre aking",
+      "g y",
+      "Ġcar box",
+      "Ġed ited",
+      "otyp ed",
+      "Ġdu oden",
+      "an ne",
+      "Ġan astom",
+      "gin ate",
+      "ĠBios ciences",
+      "ra ge",
+      "ĠCh iral",
+      "Ġsimpl ifies",
+      "Ġtes tis",
+      "str öm",
+      "ial s",
+      "Ġmic elles",
+      "cor rect",
+      "ĠGene tics",
+      "al ong",
+      "R em",
+      "res istance",
+      "Ġdr ink",
+      "orb ed",
+      "ĠT reat",
+      "ĠS ho",
+      "sh ows",
+      "é r",
+      "Ġmim ics",
+      "occ up",
+      "ec lam",
+      "ON G",
+      "Ġmark eting",
+      "ĠF inding",
+      "Ġendomet ri",
+      "âĶ Ģ",
+      "st rained",
+      "ĠM uch",
+      "Ġex ons",
+      "ĠH il",
+      "T D",
+      "ĠW W",
+      "ĠV ic",
+      "end a",
+      "Ġfact ory",
+      "ĠHep G",
+      "ĠSt atic",
+      "blast oma",
+      "w d",
+      "ra isal",
+      "ĠB asis",
+      "In s",
+      "ĠUn supervised",
+      "el o",
+      "ose lective",
+      "Ġaccompl ish",
+      "ĠP rospective",
+      "Ġuncor related",
+      "ĠG ate",
+      "icy cl",
+      "Ġur gent",
+      "ĠPath ways",
+      "Ġobl ique",
+      "ĠIndividual s",
+      "Ġiniti ative",
+      "Ġcat ast",
+      "j ections",
+      "Ġaut osomal",
+      "ĠPhil ip",
+      "Ġcomprehens ion",
+      "m M",
+      "p ain",
+      "Ġmicro M",
+      "Ġenc ounters",
+      "g oto",
+      "Ġl adder",
+      "Ġoccup y",
+      "ĠSur faces",
+      "D oc",
+      "ug by",
+      "Ġexam ines",
+      "os ynthesis",
+      "ĠK EGG",
+      "gl ass",
+      "sl ice",
+      "prop agation",
+      "str y",
+      "Ġillustr ating",
+      "im i",
+      "Ġsp ores",
+      "Ġast rophysical",
+      "Ġen closed",
+      "Ġinf erences",
+      "Ġbi jection",
+      "Ġever yday",
+      "Ġaltern atively",
+      "re action",
+      "ian ts",
+      "cont act",
+      "Ġg ing",
+      "ĠBi as",
+      "Ġautom aton",
+      "back ground",
+      "Ġneighbour ing",
+      "Ġdet ects",
+      "por ate",
+      "ĠShar ma",
+      "H ydro",
+      "Ġs acc",
+      "ĠF iber",
+      "ĠCh lam",
+      "Ġbuff ers",
+      "App lying",
+      "l ceil",
+      "em ph",
+      "ĠG SE",
+      "met ry",
+      "Ġimmun ost",
+      "ĠHistor ical",
+      "ĠD rag",
+      "Ġtransplant ed",
+      "Ġf rail",
+      "Ġanth ocyan",
+      "in te",
+      "ĠB hat",
+      "ĠO g",
+      "Ġste ering",
+      "benz ene",
+      "******************************** ********************************",
+      "Ġsynt het",
+      "A ct",
+      "Ġc in",
+      "Ġher bal",
+      "Ġd yn",
+      "Ġhyper plasia",
+      "head er",
+      "Ġcalc ulates",
+      "ĠDiff erence",
+      "Ġb ats",
+      "duc tivity",
+      "Ġconform ations",
+      "c ity",
+      "Ġsepar ates",
+      "ĠCD C",
+      "ĠPr ism",
+      "ĠBehavi our",
+      "ĠKel ly",
+      "ĠS ey",
+      "ĠÃ ł",
+      "LE X",
+      "g kin",
+      "st rom",
+      "Ġv om",
+      "ĠW in",
+      "ĠW igner",
+      "Ġcont ralateral",
+      "ĠMin or",
+      "Ġstere o",
+      "ĠApproxim ately",
+      "L ED",
+      "s ay",
+      "ĠJ S",
+      "Ġalcoh ols",
+      "Ġs an",
+      "Ġhard ening",
+      "IF N",
+      "Ġretrosp ectively",
+      "Ġgeneral ised",
+      "Ġtib ial",
+      "ĠWe ek",
+      "Ġar yl",
+      "ĠPen insula",
+      "Ġdeterm inations",
+      "Ġphot ovoltaic",
+      "Ġsugges tion",
+      "J ac",
+      "ĠV itro",
+      "Ġcycl o",
+      "Ġfibro us",
+      "dis ambiguation",
+      "pro gram",
+      "Ġgu est",
+      "ĠD ust",
+      "r ceil",
+      "Ġpow ered",
+      "Ġcardiomy ocytes",
+      "he at",
+      "yl ic",
+      "Ġpresent ations",
+      "Ġtransmit ting",
+      "W D",
+      "add ed",
+      "In itial",
+      "D el",
+      "ĠV elocity",
+      "Ġmo le",
+      "Ġo val",
+      "Ġpl ankton",
+      "the ir",
+      "ĠQ ED",
+      "vol utions",
+      "Ġmand atory",
+      "Ġrep ulsive",
+      "ĉ ĠĠ",
+      "Ġpost ulated",
+      "ĠCor tex",
+      "ĠCar b",
+      "CHK ERRQ",
+      "Ġoverl ay",
+      "ĠF arm",
+      "enor habditis",
+      "Ġpos ed",
+      "Ġinst anti",
+      "Z T",
+      "ĠVisual ization",
+      "ĠGAP DH",
+      "lec om",
+      "och ron",
+      "ĠB j",
+      "ĠT rib",
+      "Ġby te",
+      "Ġsuperim posed",
+      "Ġund i",
+      "Ġacceler ator",
+      "cri ptions",
+      "ĠSm ooth",
+      "Ġz ip",
+      "nes ota",
+      "ĠE FF",
+      "ĠC ole",
+      "ĠB ru",
+      "re nd",
+      "ut z",
+      "Ġdiagn ose",
+      "b asis",
+      "di amond",
+      "ĠIn n",
+      "ĠMed ian",
+      "Ġmarg inally",
+      "Ġlemm as",
+      "rect omy",
+      "Ġdial ogue",
+      "ĠB rid",
+      "Ġ å",
+      "ox ane",
+      "ar is",
+      "Ġmunicip ality",
+      "Ġproduc ers",
+      "Reg arding",
+      "ĠF V",
+      "ide al",
+      "exp onential",
+      "L abel",
+      "ĠF robenius",
+      "Ġe ll",
+      "ĠL TE",
+      "Ġlip ase",
+      "r p",
+      "Ġd m",
+      "ot ri",
+      "cl oud",
+      "ĠAg ent",
+      "M SCs",
+      "os om",
+      "hyd ropy",
+      "ne urons",
+      "Ġsol vable",
+      "duc ting",
+      "Ġrend ered",
+      "Ġattract or",
+      "Ġb rac",
+      "Ã ģ",
+      "Ġhost ed",
+      "ĠO ct",
+      "Ġgu iding",
+      "Ġdiges tive",
+      "j s",
+      "Ġint ent",
+      "flu x",
+      "Ġbios ynthetic",
+      "Ġe lections",
+      "ĠWil coxon",
+      "Ġspectrophot ometer",
+      "Ġimpair s",
+      "Ġabd omen",
+      "k b",
+      "ĠW ho",
+      "ASS ERT",
+      "Ġel uted",
+      "Ġmaxim ization",
+      "Ġcollect or",
+      "ĠPrevious ly",
+      "a q",
+      "am bo",
+      "ĠO z",
+      "C ur",
+      "Ġcaffe ine",
+      "M ass",
+      "p al",
+      "pi ece",
+      "ou ville",
+      "ĠM eyer",
+      "ut a",
+      "ch an",
+      "ĠK S",
+      "om otor",
+      "ĠG PR",
+      "Ġev al",
+      "ĠCo operative",
+      "ogly can",
+      "Ġnozz le",
+      "ĠS hel",
+      "Ġinter change",
+      "Ġunderg rad",
+      "Ġexplan atory",
+      "Ġphag ocytosis",
+      "Ġc tx",
+      "hes s",
+      "Ġunivers ality",
+      "ĠK illing",
+      "ons in",
+      "Ġlast ing",
+      "ĠIm m",
+      "Ġconc ordance",
+      "y ma",
+      "Ġaut umn",
+      "Ġbar ley",
+      "Ġconsequ ent",
+      "is i",
+      "Ġconjug ates",
+      "Ġta ught",
+      "Ġcovari ate",
+      "Ġadoles cence",
+      "Ġvill ages",
+      "Ġeigen functions",
+      "Ġtempor ally",
+      "ĠMin nesota",
+      "y rate",
+      "ies is",
+      "def inite",
+      "Ġalph abet",
+      "ĠY un",
+      "ĠM AR",
+      "Ġse aled",
+      "ron ectin",
+      "ĠSep ar",
+      "n x",
+      "CA A",
+      "Ġrece ption",
+      "uck y",
+      "ĠPT EN",
+      "ĠM organ",
+      "Ġdi odes",
+      "Ġmet formin",
+      "Ġsynt hes",
+      "ĠPar ticip",
+      "ĠJer sey",
+      "Ġamph ib",
+      "c hel",
+      "Ġl amp",
+      "ĠH els",
+      "ĠF N",
+      "Ġexc av",
+      "is econd",
+      "int ro",
+      "Ġnon commutative",
+      "Ġsubs ystems",
+      "sum m",
+      "Ġcontrast ing",
+      "ĠSil icon",
+      "ĠPar tition",
+      "Glc NAc",
+      "Ġdisc ern",
+      "ĠBound s",
+      "ĠR ah",
+      "Ġapproxim ating",
+      "ĠHyper t",
+      "ĠD il",
+      "Ġcompact ness",
+      "Ġca ught",
+      "ĠImpro ve",
+      "ĠTor onto",
+      "ĠBiom ark",
+      "ĠB ag",
+      "ĠIn vent",
+      "Ġelabor ate",
+      "ĠM ott",
+      "AB C",
+      "ĠGra ham",
+      "Ġpo ultry",
+      "ĠCon jecture",
+      "ĠAl gebras",
+      "ĠN LO",
+      "ap sing",
+      "path y",
+      "ĠEliz abeth",
+      "ĠT it",
+      "ĠS CI",
+      "ant on",
+      "Ġv oting",
+      "math rel",
+      "ĠF ord",
+      "ig ibility",
+      "Ġall ergy",
+      "ac oustic",
+      "ĠD yn",
+      "ĠD SC",
+      "ĠG RO",
+      "ĠTh irty",
+      "Ġanalys ing",
+      "ĠEm pire",
+      "f ire",
+      "Ġpath ologic",
+      "Ġpat ent",
+      "Ġhe ard",
+      "ĠF ront",
+      "isc onsin",
+      "hy pert",
+      "uz umab",
+      "ĠMut ation",
+      "Ġb iliary",
+      "Ġsuper fluid",
+      "ĠW C",
+      "ust om",
+      "ĠAc tivities",
+      "Ġpolyp eptide",
+      "he ets",
+      "Ġb orders",
+      "ear ly",
+      "Ġorth ogon",
+      "Ġbul ge",
+      "ï £",
+      "Ġcon ical",
+      "ĠL ept",
+      "Ġelectroly tes",
+      "ĠÂ «",
+      "reg ulating",
+      "Ġviol ated",
+      "â ĺ",
+      "AL T",
+      "ĠWork s",
+      "ĠHep at",
+      "ur gical",
+      "ob ar",
+      "ĠRe active",
+      "poss ibly",
+      "ĠAds orption",
+      "ĠR io",
+      "ano ic",
+      "ĠâĨ ij",
+      "Ġintrig uing",
+      "Ġo m",
+      "her tz",
+      "ĠApproxim ate",
+      "ĠP arent",
+      "Ġco in",
+      "exp and",
+      "Ð ²",
+      "Ġnon parametric",
+      "ex tern",
+      "ae us",
+      "gly cerol",
+      "Ġc p",
+      "Ġbat ches",
+      "Ġnanom aterials",
+      "U se",
+      "ĠV ivo",
+      "R h",
+      "Ġt iles",
+      "Ġdep ict",
+      "Ġsouth west",
+      "ĠCas imir",
+      "lay ered",
+      "ĠLe af",
+      "f em",
+      "b ered",
+      "Ġsub algebra",
+      "Ġdet achment",
+      "ĠLe uk",
+      "ol us",
+      "ĠR ick",
+      "Ġab ortion",
+      "Ġclar ified",
+      "Ġgangl ia",
+      "Q S",
+      "o ising",
+      "ĠFor ward",
+      "ĠPer ipheral",
+      "shif ted",
+      "b ula",
+      "ram olecular",
+      "ĠF EM",
+      "ĠPro ton",
+      "AM E",
+      "Ġsched ules",
+      "Ġa a",
+      "ĠU DP",
+      "st ere",
+      "Ġmorph ine",
+      "Ġspecial ist",
+      "ĠAnd roid",
+      "Id entif",
+      "Ġun expl",
+      "Ġheter ozyg",
+      "Ġf id",
+      "pyrid yl",
+      "ĠW y",
+      "phosph or",
+      "Ġfriend ly",
+      "Ġmic rol",
+      "ĠS plit",
+      "agn er",
+      "crib e",
+      "Ġm oth",
+      "ĠEu ro",
+      "ig s",
+      "ĠCon ditional",
+      "ĠSte wart",
+      "pro perties",
+      "AS C",
+      "ĠTra ditional",
+      "ĠPortug al",
+      "Ġear ned",
+      "Ġcat he",
+      "Cre ate",
+      "ici encies",
+      "Ġsph ing",
+      "x ml",
+      "Ġimmun omod",
+      "Ġcomm ute",
+      "Ġselen ium",
+      "ang es",
+      "ho ok",
+      "den oted",
+      "Ġjus tify",
+      "ĠP ool",
+      "Ġgu inea",
+      "Ġcont ra",
+      "Ġfol ded",
+      "Ġlist ing",
+      "ĠL G",
+      "ĠL ane",
+      "Ġsure ly",
+      "v et",
+      "fluor ophenyl",
+      "Ġcoron a",
+      "ĠAb und",
+      "ĠOb jects",
+      "Ġt rough",
+      "ch t",
+      "Ġdis h",
+      "ith i",
+      "ĠMat lab",
+      "w orm",
+      "Ġprote omics",
+      "Ġinter molecular",
+      "ĠPet ers",
+      "Ġmir rors",
+      "quin oline",
+      "art ens",
+      "ĠJew ish",
+      "k B",
+      "ĠD egradation",
+      "Ġrele asing",
+      "V EGF",
+      "Ġsub populations",
+      "ĠTra ffic",
+      "Ġpro line",
+      "ĠH f",
+      "Ġad ren",
+      "b irth",
+      "Ġs ender",
+      "Ġat las",
+      "Ġwork place",
+      "Ġreflec tivity",
+      "ĠEx istence",
+      "cl s",
+      "Ġfin er",
+      "Ġbreast feeding",
+      "on ectin",
+      "Ġc ogn",
+      "ell ate",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ",
+      "by te",
+      "Ġsk et",
+      "N ULL",
+      "s ystems",
+      "ĠB ranch",
+      "ĠPro posed",
+      "lear n",
+      "Ġtoler ant",
+      "Ġver tebrates",
+      "Ġmulti level",
+      "ĠPA H",
+      "Ġaud ience",
+      "ĠW L",
+      "nit rop",
+      "ĠC t",
+      "Ġsati va",
+      "e ight",
+      "Ġme g",
+      "oc ell",
+      "Ġst ating",
+      "dom inant",
+      "b ytes",
+      "Ġp u",
+      "ĠB atter",
+      "ot axis",
+      "ĠE BV",
+      "Ġnanoc rystals",
+      "Ġmonop ole",
+      "Ġdia phrag",
+      "ĠV el",
+      "Ap pendix",
+      "at ten",
+      "im pl",
+      "Ġland mark",
+      "encl ature",
+      "ĠST AR",
+      "Ġprost agland",
+      "oprot ective",
+      "Ġload ings",
+      "ĠPres ence",
+      "ĠN SF",
+      "ress es",
+      "F U",
+      "il ers",
+      "Ġeryth rocytes",
+      "t rac",
+      "is lation",
+      "ĠN ight",
+      "Ġster oids",
+      "ti z",
+      "ĠD MA",
+      "Ġr ic",
+      "Ġsal ient",
+      "ĠF ur",
+      "spec ial",
+      "Ġbio informatics",
+      "ign ant",
+      "ĠEX PERIM",
+      "avor able",
+      "dis k",
+      "Ġcur riculum",
+      "imid azol",
+      "hig her",
+      "Ġdesign er",
+      "ĠSt rength",
+      "Ġcytos ol",
+      "ĠCh annels",
+      "L and",
+      "s par",
+      "Ex pression",
+      "Ġday time",
+      "merc ial",
+      "v box",
+      "in ar",
+      "ie ving",
+      "ce in",
+      "ĠNC BI",
+      "R AN",
+      "¸ Ģ",
+      "H ig",
+      "ĠD HA",
+      "Ġsub script",
+      "ĠÂ ¢",
+      "or ange",
+      "Ġknow s",
+      "ĠN AF",
+      "pro duced",
+      "ep id",
+      "Ġdex amethasone",
+      "Ġformal dehyde",
+      "yl l",
+      "Ġec topic",
+      "ĠVer ification",
+      "activ ating",
+      "ĠI G",
+      "ĠP av",
+      "Ġtra ding",
+      "Ġgrad uate",
+      "ĠF IR",
+      "enc il",
+      "ever y",
+      "Ġradi ological",
+      "ĠMamm alian",
+      "M ES",
+      "in ium",
+      "ĠS AS",
+      "ĠW H",
+      "Over ride",
+      "ĠSched uling",
+      "ĠB es",
+      "ĠY ao",
+      "Ġgl ad",
+      "ĠStandard s",
+      "Ġprov inces",
+      "en ers",
+      "Ġn r",
+      "Ġtrans pos",
+      "ĠCar ib",
+      "Ġfa una",
+      "um i",
+      "res et",
+      "Ġsup ra",
+      "Ġdiv isions",
+      "Ġbiod egrad",
+      "metric s",
+      "og rafts",
+      "Ġfunc tors",
+      "Ġsup portive",
+      "Ġcaud al",
+      "Ġexer ts",
+      "Ġc ub",
+      "od imer",
+      "Ġair borne",
+      "Ġdeliver ing",
+      "Ġmultiv ariable",
+      "Ġfurn ace",
+      "Ġremn ant",
+      "Ġinc o",
+      "ĠElect romagnetic",
+      "m apping",
+      "Ġdecl ines",
+      "c old",
+      "ĠS eed",
+      "con version",
+      "Ġglyc ogen",
+      "d T",
+      "aw i",
+      "AP P",
+      "H ol",
+      "ataly sts",
+      "ĠSat ellite",
+      "gar is",
+      "c ard",
+      "ĠBre ak",
+      "ĠAgain st",
+      "d dot",
+      "Ġpr uning",
+      "ĠCa enorhabditis",
+      "Ġsucceed ed",
+      "ub ert",
+      "ĠÏ ħ",
+      "ID s",
+      "Ġasympt otics",
+      "Ġauto anti",
+      "ĠScal ar",
+      "Ġnemat ode",
+      "h d",
+      "Ġg yn",
+      "ist ocene",
+      "Ġunderg round",
+      "ĠEth ical",
+      "Ġs ial",
+      "ĠM igration",
+      "cop e",
+      "Ġstig ma",
+      "Ġele ven",
+      "Ġcolor ing",
+      "in itions",
+      "ĠJ ay",
+      "ob a",
+      "ĠL DA",
+      "Ġbuild s",
+      "g ences",
+      "ĠEc ology",
+      "schem e",
+      "ĠUltr as",
+      "Ġmedi ation",
+      "ĠTa q",
+      "Ġf lying",
+      "ĠEqu ilibrium",
+      "ophosph ate",
+      "ĠArgent ina",
+      "ps ia",
+      "tt es",
+      "Ġdispar ity",
+      "Ġadver tis",
+      "agg reg",
+      "I SA",
+      "od em",
+      "ĠR ational",
+      "Ġsil ent",
+      "divid ed",
+      "P an",
+      "J A",
+      "cl aim",
+      "Ġradio active",
+      "Ġp ink",
+      "Ġcon verse",
+      "ĠM ell",
+      "en ib",
+      "rus kal",
+      "sl ope",
+      "hen ol",
+      "ĠP on",
+      "par tition",
+      "SM GR",
+      "tit led",
+      "ĠInter ference",
+      "t osecond",
+      "Ġse q",
+      "Ġtrans itive",
+      "ĠW id",
+      "review ed",
+      "× ¥",
+      "ĠV C",
+      "rec all",
+      "ogene ic",
+      "ĠOverex pression",
+      "Ġcom mitted",
+      "Ġsyn apse",
+      "Sh ort",
+      "ĠNeut ral",
+      "ic les",
+      "IS M",
+      "Ġintr insically",
+      "Ġmicros atellite",
+      "R N",
+      "ĠâĪ ĥ",
+      "det ection",
+      "Ġcod imension",
+      "Ġdrawback s",
+      "ĠTurn er",
+      "Ġsputter ing",
+      "Ġdis mut",
+      "Ġhyp ogly",
+      "Ġspe ak",
+      "J D",
+      "Ġs ul",
+      "Ġperin atal",
+      "Ġin k",
+      "ies t",
+      "Ġoffic ers",
+      "tic k",
+      "Ġre taining",
+      "ĠN ET",
+      "Ġexchang es",
+      "Ġany one",
+      "ĠEnd othelial",
+      "s end",
+      "in jection",
+      "ĠPer u",
+      "Ġcl ades",
+      "uct uations",
+      "Ġsulph ate",
+      "pi o",
+      "Ġphys i",
+      "ĠMi y",
+      "ĠB AS",
+      "ari us",
+      "Ġlip opolysaccharide",
+      "Ġneurode generation",
+      "ĠTurk ish",
+      "Ġo phthal",
+      "Ġac ted",
+      "ent re",
+      "Ġsh aking",
+      "Ġchlor oplast",
+      "ĠS id",
+      "regn ancy",
+      "as ion",
+      "ĠH s",
+      "Ġiniti ating",
+      "Ġflex ural",
+      "Ï ª",
+      "Ġpar ac",
+      "Ġinter layer",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠĠĠĠĠĠĠ",
+      "c ause",
+      "rac tions",
+      "Ġval uation",
+      "SY SMGR",
+      "ĠGarc ia",
+      "ar rays",
+      "Ġcast ing",
+      "ĠP FN",
+      "ĠL anc",
+      "ĠGl ob",
+      "Ġd enti",
+      "Ġport folio",
+      "ĠHol ocene",
+      "ĠMAT ERIAL",
+      "Ġs arc",
+      "L ear",
+      "Ġt in",
+      "ĠC lear",
+      "bel ow",
+      "Ġadv ection",
+      "Ġoverl aps",
+      "Ġarth roplasty",
+      "comput e",
+      "Ġglycol ysis",
+      "he pt",
+      "lor a",
+      "f rames",
+      "ĠH ern",
+      "pro to",
+      "Ġsw ine",
+      "Ġje jun",
+      "Ġrepe ating",
+      "ancre atic",
+      "ĠColl ins",
+      "ĠPrinc iple",
+      "Ġnan of",
+      "Ġadj acency",
+      "Ġsyn ov",
+      "che t",
+      "ĠAl most",
+      "Ġintr usion",
+      "Ġechocardi ography",
+      "lif eration",
+      "Ġquies cent",
+      "ĠM uk",
+      "Ġlife times",
+      "grad ed",
+      "Ġoverw hel",
+      "z el",
+      "Ġnit ride",
+      "Ġdisturb ed",
+      "Ġfast est",
+      "gra bility",
+      "Ġtoler ated",
+      "f rag",
+      "ĠExt ension",
+      "ano ate",
+      "ifer ous",
+      "Ġhydro dynamics",
+      "IO NAL",
+      "ĠT oday",
+      "ĠExp ansion",
+      "Ġven om",
+      "ĠHep atitis",
+      "ñ o",
+      "on ation",
+      "syn uclein",
+      "Ġbasket ball",
+      "cl usions",
+      "Ġsett led",
+      "I QR",
+      "ĠC ra",
+      "Ġautom ation",
+      "ĠHealth y",
+      "ĠPortug uese",
+      "ĠAb elian",
+      "Ġg ad",
+      "ĠH G",
+      "ĠR oth",
+      "Ġcons ume",
+      "F G",
+      "in als",
+      "ĠM CMC",
+      "Ġpregn ancies",
+      "D ES",
+      "por tional",
+      "ĠBi ochemical",
+      "Ġmiss ions",
+      "ĠAnti body",
+      "ĠB CG",
+      "ĠL AS",
+      "mar ine",
+      "D MA",
+      "Ġl ongevity",
+      "ĠD ry",
+      "ĠR ao",
+      "Ġinterfer ometer",
+      "Ġdiscre tized",
+      "osens ory",
+      "s it",
+      "et ta",
+      "tain er",
+      "other wise",
+      "AK T",
+      "ĠFac ulty",
+      "Ġas certain",
+      "ĠSim ulated",
+      "Ġpay load",
+      "O UT",
+      "Ġsuff ers",
+      "Ġt ungsten",
+      "ĠAn xiety",
+      "ĠHeter ogeneous",
+      "ling ual",
+      "Ġphe rom",
+      "b ors",
+      "l inux",
+      "Ġmon key",
+      "Â £",
+      "ur l",
+      "ĠAc ross",
+      "ĠAK I",
+      "Ġop p",
+      "ocal ization",
+      "Ġmorph ogenesis",
+      "g ic",
+      "ĠP CM",
+      "Ġolig omers",
+      "Ġexhaus tive",
+      "ĠG IS",
+      "Ġpr istine",
+      "ĠAc tiv",
+      "ĠSc ilab",
+      "ĠAc oustic",
+      "ĠP ick",
+      "integr al",
+      "Ġphilos ophy",
+      "ĠD eng",
+      "ĠH ab",
+      "sc ape",
+      "ĠEmerg ency",
+      "Ġe pi",
+      "ĠB ET",
+      "ric ket",
+      "Ġann ulus",
+      "Ġlys osomal",
+      "Ġstrand s",
+      "C AP",
+      "ĠAmin o",
+      "ĠSt ri",
+      "epend ence",
+      "Ġfoot print",
+      "ĠFat ty",
+      "ĠN az",
+      "n est",
+      "ĠEx plicit",
+      "plan etary",
+      "le ad",
+      "Ġg rip",
+      "ne ed",
+      "AT T",
+      "ER V",
+      "ĠTarget ed",
+      "CR P",
+      "Ġparam agnetic",
+      "ĠT yr",
+      "ĠMicro RNA",
+      "h line",
+      "g h",
+      "p it",
+      "ĠIs olated",
+      "ject ory",
+      "Ġclean ed",
+      "ost e",
+      "Ġpath ologies",
+      "prop ylene",
+      "ĠRe ason",
+      "ĠIN FO",
+      "RA Y",
+      "Val ues",
+      "Ġal ive",
+      "Ġbi of",
+      "ew icz",
+      "Ġcrack ing",
+      "go ogle",
+      "lock ed",
+      "c rop",
+      "ec a",
+      "ur ane",
+      "SV M",
+      "ut ta",
+      "ĠMet ric",
+      "ĠEn cycl",
+      "ĠMod ule",
+      "Ġwarrant ed",
+      "Ġmulti disciplinary",
+      "ĠEl astic",
+      "lab elled",
+      "ĠSchwarz schild",
+      "ĠP CC",
+      "ma jor",
+      "v ideo",
+      "Ġst oring",
+      "ĠM ake",
+      "ak o",
+      "ĠJ ia",
+      "Ġtor oidal",
+      "ĠH MM",
+      "Ġmask ing",
+      "Ag ain",
+      "Ġneph ropathy",
+      "g f",
+      "Ġdom inating",
+      "er kin",
+      "ĠFabric ation",
+      "ĠF el",
+      "DE F",
+      "c ulture",
+      "ĠI ra",
+      "ĠRE G",
+      "iling ual",
+      "Ġm uss",
+      "pl ain",
+      "z h",
+      "ist on",
+      "ĠÎ ¥",
+      "min imal",
+      "c mp",
+      "Ga N",
+      "Ġmonot onic",
+      "Ġinv olution",
+      "Ġwh atever",
+      "ĠInstr ument",
+      "im ple",
+      "ĠPC I",
+      "ĠNe uronal",
+      "Ġfac ets",
+      "Ġhemod ialysis",
+      "ap atite",
+      "ĠK il",
+      "ont ally",
+      "Ġinser ting",
+      "ĠR IP",
+      "Ġconn ective",
+      "ĠFed eration",
+      "n ut",
+      "ĠG un",
+      "inu ous",
+      "M or",
+      "ĠW isconsin",
+      "Ġmus h",
+      "IT S",
+      "Ġe ject",
+      "ĠB PS",
+      "ĠH orn",
+      "ĠEmbed ding",
+      "Ġr aces",
+      "ĠJ am",
+      "Ġpost ure",
+      "ĠIn vol",
+      "ĠIMD b",
+      "ĠP lease",
+      "pro portion",
+      "ĠInter leukin",
+      "Ġar te",
+      "Ġsub sp",
+      "oder ma",
+      "F ind",
+      "im it",
+      "ĠCl in",
+      "H el",
+      "FI LE",
+      "orig inal",
+      "erv oir",
+      "Ġple ural",
+      "clip se",
+      "enc er",
+      "in aries",
+      "Ġv ictory",
+      "Ġinvestig ates",
+      "ĠImport ance",
+      "ĠM IN",
+      "Ġphon ons",
+      "integr ated",
+      "Ġex changed",
+      "ys tis",
+      "Ġmig rate",
+      "R ob",
+      "el and",
+      "pro of",
+      "ĠIntegr al",
+      "Ġmerg ers",
+      "Ġpolyphen ols",
+      "ĠF ully",
+      "Ġu ro",
+      "Ġhom ogenous",
+      "Ġrecogn izing",
+      "ĠSign als",
+      "v at",
+      "ig ms",
+      "Ġaccur acies",
+      "Sub stituting",
+      "Ġpoison ing",
+      "Ġsh rimp",
+      "ĠH ölder",
+      "ĠTanz ania",
+      "J S",
+      "M ENT",
+      "ĠTop ology",
+      "Ġin vers",
+      "ĠD U",
+      "Ġun iaxial",
+      "ĠS EC",
+      "par ty",
+      "Ġcontroll able",
+      "Ġf um",
+      "os tics",
+      "Ġmanif ested",
+      "Ġpropag ated",
+      "Ġsuff ix",
+      "ĠC AN",
+      "ĠP ret",
+      "ke eping",
+      "Assum ing",
+      "Ġs uture",
+      "Ġp est",
+      "Ġg amet",
+      "ĠAl ignment",
+      "esare an",
+      "t um",
+      "Ġref ine",
+      "Ġpop ulated",
+      "Ġest u",
+      "ĠDef ense",
+      "ĠPri vacy",
+      "ĠWe in",
+      "ĠSen ate",
+      "Ġazim uth",
+      "ĠProf essional",
+      "Ġlab our",
+      "Ġsem inal",
+      "ĠInter vention",
+      "ĠOl der",
+      "A U",
+      "W ind",
+      "d ynamical",
+      "ĠV eter",
+      "aci ón",
+      "Ġco oking",
+      "Ġâī ª",
+      "Ġbe ad",
+      "Ġdens ely",
+      "Ġpall iative",
+      "m ort",
+      "ĠA AV",
+      "ĠR yan",
+      "P rim",
+      "g alax",
+      "mu ir",
+      "st ers",
+      "ĠSal t",
+      "quee ze",
+      "ĠPlate au",
+      "Ġ í",
+      "Ġl ighter",
+      "ord inary",
+      "formal dehyde",
+      "ĠW er",
+      "Ġb ark",
+      "Ġhomogen ized",
+      "Ġpyram idal",
+      "Ġin ert",
+      "ĠA PC",
+      "ĠMic ros",
+      "ĠProte obacteria",
+      "ĠPur ification",
+      "Ġparametri zed",
+      "Ġ ille",
+      "acc uracy",
+      "embed ding",
+      "Ġtough ness",
+      "Ġis ometry",
+      "back s",
+      "ĠF IG",
+      "ĠR on",
+      "ĠE SP",
+      "Ġmicrogl ial",
+      "inter p",
+      "ĠIntegr ating",
+      "ĠReduc ing",
+      "Ġhe arts",
+      "Ġserious ly",
+      "Ġspec ially",
+      "CT RL",
+      "ĠSur prisingly",
+      "Ġhyper plane",
+      "pol ynomial",
+      "Ġrecon c",
+      "Ġpharmacokine tic",
+      "M art",
+      "ĠB right",
+      "m able",
+      "Ġion izing",
+      "Ġtr ich",
+      "zym atic",
+      "Ġlept ons",
+      "et ting",
+      "ĠH ex",
+      "Ġneu rop",
+      "Ġadip ocytes",
+      "Ġro ds",
+      "Ġsuper critical",
+      "Ġsuc cin",
+      "Ġan ter",
+      "ĠN AC",
+      "ĠSub sequent",
+      "IG H",
+      "Ġs outheast",
+      "Ġend owed",
+      "Ġconver ging",
+      "Ġspati o",
+      "Ġcele br",
+      "hel ix",
+      "Ġaccess ions",
+      "Ġimmobil ization",
+      "ĠE Q",
+      "sp atial",
+      "Ġinform al",
+      "Ġd ere",
+      "ĠEn zyme",
+      "ĠB BC",
+      "ĠE PR",
+      "Ġelect rically",
+      "Ġleuk ocytes",
+      "Ġal anine",
+      "Ġmit ogen",
+      "Ġintram olecular",
+      "ĠN I",
+      "Ġpro kary",
+      "IS O",
+      "Ġd odec",
+      "ĠTra de",
+      "ĠD ai",
+      "cc c",
+      "ĠWal ter",
+      "ĠNe ither",
+      "Ġvul garis",
+      "Ġlong itude",
+      "ĠInt ro",
+      "op tion",
+      "ĠQ C",
+      "Ġâ Ŀ",
+      "prot ection",
+      "ĠI MF",
+      "ap rote",
+      "Ġlink er",
+      "Ġfound er",
+      "Ġaspir ation",
+      "clust ers",
+      "ĠP ay",
+      "ĠR oles",
+      "Ġac yclic",
+      "over ing",
+      "Ġrem ind",
+      "ĠT ong",
+      "ĠAt ten",
+      "Ġengine ers",
+      "Ġdys regulation",
+      "ĠFour th",
+      "Ġfil ename",
+      "ĠCo ol",
+      "prot ected",
+      "Ġnil potent",
+      "ĠH K",
+      "cl one",
+      "ĠSt adium",
+      "a is",
+      "os amine",
+      "AB ILITY",
+      "rov ascular",
+      "ĠA H",
+      "ĠCon cept",
+      "Ġcereb rospinal",
+      "ow itz",
+      "Ġresol ving",
+      "Ġw ings",
+      "ĠE GF",
+      "ĠCom mand",
+      "iaz ep",
+      "Ġbe ef",
+      "Ġsp ines",
+      "Ġprior ities",
+      "Ġattempt ing",
+      "Ġtel omere",
+      "B QU",
+      "Ġviol ations",
+      "L B",
+      "om nia",
+      "os m",
+      "ir q",
+      "Ġdivers ification",
+      "al t",
+      "ĠB RAF",
+      "Ġorgan isation",
+      "di e",
+      "Ġaut oreg",
+      "ick ed",
+      "ĠEc ological",
+      "ĠT rain",
+      "ĠP Y",
+      "Ġmusc uloskeletal",
+      "Ġhoriz ons",
+      "Ġo mega",
+      "Ġquas ars",
+      "ep tion",
+      "Ġer ad",
+      "Ġlum inal",
+      "Interest ingly",
+      "Ġpay ment",
+      "c nt",
+      "Ġdi pl",
+      "Ġrecogn ised",
+      "C at",
+      "ĠCh l",
+      "Ġmill ions",
+      "Ġdisappear ance",
+      "G AP",
+      "Ġradi ographic",
+      "Ġpost partum",
+      "develop ed",
+      "x ual",
+      "Ġhe d",
+      "id ered",
+      "ĠC ertain",
+      "Ġdys plasia",
+      "____ ____",
+      "ĠHal f",
+      "Ġas ymmetries",
+      "ĠAl cohol",
+      "S um",
+      "Ġf m",
+      "Ġch ap",
+      "Ġpre treated",
+      "ĠGall ery",
+      "Ġoutper form",
+      "Ġbreed s",
+      "Ġt ied",
+      "Ġdiffe omorphism",
+      "Ġcaus ative",
+      "Ġcollec tively",
+      "Ġsub optimal",
+      "Ġins ulation",
+      "Ġmanip ulate",
+      "Ġkil omet",
+      "Ġrep ulsion",
+      "Ġchloro form",
+      "Ġbe an",
+      "Ġhe ro",
+      "rophys ics",
+      "ĠP eptide",
+      "Ġout lier",
+      "Der ived",
+      "iss er",
+      "ĠInf ant",
+      "sulf onyl",
+      "Ġrecurs ively",
+      "H u",
+      "ĠK oh",
+      "pyrid ine",
+      "Ġs quad",
+      "Ġth irty",
+      "Ġsp oken",
+      "ĠZ ar",
+      "other mic",
+      "Ġcalc ification",
+      "ĠHels inki",
+      "Ġbe ach",
+      "ĠF DR",
+      "Ġprob iotic",
+      "Ġfin ishing",
+      "ymmet rical",
+      "Ġvac ancy",
+      "Ġthrom bo",
+      "Comp ared",
+      "A ST",
+      "st ed",
+      "othe rap",
+      "Ġiod ide",
+      "Ġt t",
+      "al ignment",
+      "Ġmicro vascular",
+      "Ġinitial ize",
+      "ĠANAL YSIS",
+      "Ġtop ographic",
+      "ĠReport ing",
+      "Ġunderestim ated",
+      "put ed",
+      "Ġatheros clerotic",
+      "Qi agen",
+      "g ut",
+      "ĠCor tical",
+      "Ġdisrup t",
+      "es te",
+      "Ġgl ue",
+      "Ġnarrow er",
+      "Ġin patient",
+      "Ġsch olars",
+      "Ġb c",
+      "ĠPsych ological",
+      "ĠHamilton ians",
+      "Ġhon or",
+      "tib ular",
+      "Ġinser tions",
+      "oscop e",
+      "Ġpharmacokine tics",
+      "Ġmathem atically",
+      "Ġfor k",
+      "ip ital",
+      "ĠAr gs",
+      "abol ism",
+      "Ġâİ ł",
+      "ĠRob ot",
+      "ĠC asc",
+      "Ġle aching",
+      "ĠL ack",
+      "Ġend ocytosis",
+      "Ġtr is",
+      "Ġsensiti vities",
+      "Ġlic ensed",
+      "Ġsp onge",
+      "carbon yl",
+      "fe at",
+      "Ġpre cl",
+      "Ġwa ist",
+      "tif ications",
+      "Ġol iv",
+      "b inary",
+      "at ri",
+      "ĠBi ot",
+      "T Z",
+      "Ġf ake",
+      "ĠM osc",
+      "ĠH PS",
+      "ĠVol tage",
+      "Ġâİ Ŀ",
+      "ĠAh med",
+      "ĠSex ual",
+      "dehyd es",
+      "ĠC ot",
+      "Ġmag ma",
+      "oxyl in",
+      "Ð Ī",
+      "amet hyl",
+      "ĠL OS",
+      "di phenyl",
+      "experim ental",
+      "Ġpluripot ent",
+      "agit tal",
+      "w alk",
+      "Ġplas monic",
+      "Ġcontrac ts",
+      "Ġexp ed",
+      "ĠArab ia",
+      "Ġshoot s",
+      "ĠR AN",
+      "ustr ated",
+      "Ġconvex ity",
+      "Ġm J",
+      "ĠAbs olute",
+      "ĠS EL",
+      "MI P",
+      "ĠAct ually",
+      "so le",
+      "Q I",
+      "ĠTGF β",
+      "Ġâİ ŀ",
+      "Ġrearrang ements",
+      "Ġc uring",
+      "exp ensive",
+      "cepti bility",
+      "Ġour s",
+      "ĠKid ney",
+      "Ġassign s",
+      "Ġvox els",
+      "ore al",
+      "Ġeven ing",
+      "h us",
+      "Ġ ãĢ",
+      "or adi",
+      "ĠCor rection",
+      "Ġnanofib ers",
+      "Ġcan tile",
+      "big oplus",
+      "umin ous",
+      "eclam psia",
+      "ĠC ult",
+      "EC H",
+      "at ology",
+      "Ġj i",
+      "cr yp",
+      "ĠAsp ects",
+      "en i",
+      "Ġsem is",
+      "IR S",
+      "ĠP ho",
+      "enc oding",
+      "ĠJus tice",
+      "ococc i",
+      "Ġhypoth alamic",
+      "ract able",
+      "ĠOr b",
+      "Sim ons",
+      "Ġmanip ulated",
+      "att ribute",
+      "on ov",
+      "or ously",
+      "end ar",
+      "ud er",
+      "ins ert",
+      "Ġlys ed",
+      "ĠHod ge",
+      "Ġfootball er",
+      "Dev ice",
+      "ĠLe ast",
+      "Ġstrat um",
+      "Ġmit ral",
+      "Ġs ell",
+      "ĠM uc",
+      "gly cer",
+      "o j",
+      "Ġpathogen icity",
+      "ĠDecl aration",
+      "op ause",
+      "ĠAr ticle",
+      "Ġrins ed",
+      "ĠLé vy",
+      "re ment",
+      "Ġan ts",
+      "ĠD ic",
+      "Ġk Pa",
+      "ur ry",
+      "mo tion",
+      "cl ient",
+      "Ġaccess ory",
+      "Ġdepolar ization",
+      "nam ely",
+      "Ġdispar ities",
+      "Ġfavour able",
+      "ĠTib et",
+      "Ġo ocyte",
+      "ist ration",
+      "Ġun resolved",
+      "cri ptive",
+      "phys ics",
+      "Ġben zo",
+      "Ġcrystall inity",
+      "Ġpay off",
+      "Ġumb ilical",
+      "os il",
+      "ĠSystem ic",
+      "ĠST M",
+      "Ġstabil izer",
+      "U SA",
+      "ĠJ ensen",
+      "A ug",
+      "ĠH at",
+      "AG G",
+      "under brace",
+      "Ġmanip ulations",
+      "ĠM anc",
+      "ned y",
+      "Ġscr atch",
+      "C herry",
+      "osacchar ides",
+      "Ġprecipit ate",
+      "quar ters",
+      "ic ul",
+      "Ġoptim ally",
+      "man y",
+      "Ġneoplas ms",
+      "Ġin ward",
+      "ary ng",
+      "Ġm oll",
+      "ĠW el",
+      "ĠW iley",
+      "Ġnewsp aper",
+      "Ġinhabit ants",
+      "ĠS uccess",
+      "Ġbrid ging",
+      "Ġdis connected",
+      "Ġhygi ene",
+      "D ist",
+      "Ġsc ripts",
+      "Ġmes oporous",
+      "Ġrestric ts",
+      "act one",
+      "Ġaqu ifer",
+      "ĠïĤ ·",
+      "Ġp lex",
+      "Ġpresum ed",
+      "Ġ ips",
+      "ĠM ilitary",
+      "Ġjud ged",
+      "Ġal d",
+      "Ġsequ est",
+      "comp ared",
+      "UL ATION",
+      "adap ted",
+      "Ġinstruc ted",
+      "p ulse",
+      "Ġc usp",
+      "mat ching",
+      "car rier",
+      "Ġenfor ce",
+      "ĠInter view",
+      "ometric s",
+      "Ġnull ptr",
+      "Ġflav our",
+      "ĠPare to",
+      "ĠB ER",
+      "Ġu v",
+      "Ġcr ash",
+      "ĠC ann",
+      "ĠMin eral",
+      "ĠOlymp ic",
+      "Ġpolyc rystalline",
+      "le tt",
+      "T ables",
+      "requ ent",
+      "Ġsed entary",
+      "uns aturated",
+      "ĠBern oulli",
+      "Ġad missions",
+      "itor ial",
+      "ac ute",
+      "Ġad ditions",
+      "we et",
+      "AL E",
+      "ĠMan ip",
+      "tok ens",
+      "prec ed",
+      "d k",
+      "cons ider",
+      "Ġïĺ ¹",
+      "Ġwr ites",
+      "car dia",
+      "ct omy",
+      "omat ous",
+      "S ymbol",
+      "ust en",
+      "Ġprote olytic",
+      "c ategories",
+      "Ġf ic",
+      "Ġsw ing",
+      "Ġpass enger",
+      "Ġoverl apped",
+      "if i",
+      "Ġmut ational",
+      "ĠJoseph son",
+      "Ġreg ret",
+      "ĠAr k",
+      "ĠCF D",
+      "Ġman eu",
+      "enc oded",
+      "texts c",
+      "Ġdecom positions",
+      "ĠDe b",
+      "Ġmand ibular",
+      "d U",
+      "ĠP IC",
+      "Ġtranscript omic",
+      "Ġtel escop",
+      "ĠSant os",
+      "o E",
+      "ĠM CP",
+      "Ġind igenous",
+      "Ġmicrosp heres",
+      "Ġcod ew",
+      "z ip",
+      "Ġfing ers",
+      "Ġcampaign s",
+      "¸Ģ ł",
+      "Ġacc idents",
+      "ĠTo ols",
+      "Pl anck",
+      "Â »",
+      "ed er",
+      "ing ham",
+      "oxid ase",
+      "Ġancest or",
+      "wh ose",
+      "Ġphosph olipid",
+      "Ġconvers ation",
+      "ĠH of",
+      "cor tical",
+      "gly cos",
+      "Ġmanufacture rs",
+      "op ulmonary",
+      "Ġincl ined",
+      "ĠBet he",
+      "Ġsp ending",
+      "ĠFus arium",
+      "u itively",
+      "Ġfem ur",
+      "ĠL inks",
+      "Ġnit rite",
+      "M ain",
+      "Ġfl ora",
+      "ĠPh D",
+      "ĠWr iting",
+      "ĠHess ian",
+      "Ġμ s",
+      "ool s",
+      "Ġvictim s",
+      "ĠR ew",
+      "ans en",
+      "E ar",
+      "Ġor n",
+      "Ġthermo electric",
+      "EN SE",
+      "ĠWeight ed",
+      "h oles",
+      "Ġc en",
+      "Ġac uity",
+      "Ġvac ancies",
+      "ĠDu ke",
+      "Ġpac litaxel",
+      "Ġconver ts",
+      "bour ne",
+      "ĠA CS",
+      "os i",
+      "Ġcrim inal",
+      "ĠI b",
+      "un es",
+      "ĠNan oc",
+      "P ost",
+      "ĠMD S",
+      "Ġecon omics",
+      "Ġthough ts",
+      "Ġneuro protective",
+      "Ġinters ects",
+      "c ers",
+      "at id",
+      "us a",
+      "ĠAn s",
+      "Ġafter wards",
+      "ĠOF DM",
+      "ĠCM V",
+      "ĠC um",
+      "AT G",
+      "ĠImage Net",
+      "ĠAtt ack",
+      "ogene ities",
+      "Ġcouns eling",
+      "ĠCON TR",
+      "á lez",
+      "ĠD h",
+      "ĠG V",
+      "Ġposition al",
+      "Ġg ang",
+      "ĠInter active",
+      "w ig",
+      "ĠT race",
+      "ĠD SS",
+      "Ġsynthet ase",
+      "ĠGal ile",
+      "us ually",
+      "ĠB ass",
+      "ard less",
+      "Ġexec uting",
+      "K P",
+      "ĠNep al",
+      "RE AD",
+      "ĠL ock",
+      "oh ydro",
+      "rot ation",
+      "d il",
+      "roscop ically",
+      "re perfusion",
+      "Ġdis hes",
+      "ĠProceed ings",
+      "ĠN PC",
+      "Ġmon soon",
+      "ĠLem mas",
+      "ĠChand ra",
+      "Ġre actors",
+      "Ġtr yptophan",
+      "ĠV T",
+      "ĠD EM",
+      "Ġleg islation",
+      "m k",
+      "Ġtor ic",
+      "ĠProgram s",
+      "ĠPub Med",
+      "Ġr DNA",
+      "Ġpost s",
+      "Ġâİ Ľ",
+      "Ġshed ding",
+      "toler ant",
+      "Ġv oids",
+      "ĠCarib bean",
+      "C ODE",
+      "T ube",
+      "AL SE",
+      "Ġchlor ine",
+      "Ġco erc",
+      "ĠRh iz",
+      "ĠKir k",
+      "ĠÃ ĸ",
+      "ro ut",
+      "ic ides",
+      "ag u",
+      "ĠK w",
+      "Ġcr u",
+      "Obs erve",
+      "ĠRe vis",
+      "Ġan onym",
+      "Ġpre requ",
+      "ocor tical",
+      "Ġrest aur",
+      "ĠPop ulations",
+      "d st",
+      "Ġfor t",
+      "reg s",
+      "ĠPolar ization",
+      "Ġpancre atitis",
+      "a ph",
+      "th reat",
+      "ft en",
+      "ĠAl aska",
+      "ĠFlex ible",
+      "Ġreperto ire",
+      "k an",
+      "math choice",
+      "Ġmit osis",
+      "Ġe at",
+      "ut in",
+      "Ġr t",
+      "Ġd ummy",
+      "ĠC ys",
+      "ĠG or",
+      "ear chers",
+      "H PLC",
+      "Ġb ay",
+      "ĠNi elsen",
+      "ĠR oc",
+      "ian i",
+      "ic it",
+      "rag ue",
+      "Ġcour ts",
+      "test ing",
+      "Ġampl ify",
+      "Ġtu ples",
+      "prol iferative",
+      "ĠPar as",
+      "Ġmagn ets",
+      "Ġchem okines",
+      "ĠMit chell",
+      "ĠPet ri",
+      "hol tz",
+      "y ch",
+      "mat rices",
+      "Ġcorrec ting",
+      "ĠPC a",
+      "ynam ically",
+      "ĠNAF LD",
+      "Ġeff luent",
+      "it um",
+      "Ġth rows",
+      "ĠGu id",
+      "och romatic",
+      "ĠF ro",
+      "id ad",
+      "rom agnetism",
+      "H erm",
+      "ĠS pi",
+      "ĠQu as",
+      "dom ains",
+      "Ġquad rant",
+      "ĠSO X",
+      "ĠGover nor",
+      "Ġam enable",
+      "he ld",
+      "ĠC ul",
+      "Ġunder water",
+      "ĠK ron",
+      "ĠSp ati",
+      "ano yl",
+      "C U",
+      "ov ir",
+      "Ġdem ographics",
+      "With in",
+      "ĠM é",
+      "texts f",
+      "ĠLab el",
+      "Ġgenu ine",
+      "Ġh ill",
+      "ĠL az",
+      "Ġt esticular",
+      "ĠB row",
+      "IC ATION",
+      "Â ¡",
+      "ĠA IC",
+      "anc omycin",
+      "str ual",
+      "Ġarrest ed",
+      "ĠS om",
+      "ĠI HC",
+      "ĠP ose",
+      "ĠM ö",
+      "ist ar",
+      "ĠP AM",
+      "ĠH CT",
+      "Ġtyp edef",
+      "ĠMor se",
+      "ĠLe ishman",
+      "lim b",
+      "Ġsphe roid",
+      "os ely",
+      "ĠGu inea",
+      "re new",
+      "Ġpsori asis",
+      "ist a",
+      "ĠCh ung",
+      "orth ogonal",
+      "ĠShe ar",
+      "ĠMus lim",
+      "ĠP ict",
+      "In teger",
+      "Ġspac er",
+      "L y",
+      "Ġd ermal",
+      "Ġonc ology",
+      "Ġd p",
+      "Ġphot oluminescence",
+      "reg on",
+      "amin ase",
+      "ĠẠĭ",
+      "Inst ance",
+      "ver b",
+      "Ġmethyl ated",
+      "ĠG em",
+      "ist ently",
+      "ĠMg Cl",
+      "ĠEle vated",
+      "⣠©",
+      "onstr uct",
+      "Ġsnap shot",
+      "en em",
+      "ĠD isk",
+      "Ġhydro static",
+      "Ġïĥ ª",
+      "v or",
+      "ĠI E",
+      "ĠL Y",
+      "OR F",
+      "Ġfo il",
+      "m ale",
+      "Ġdepend ed",
+      "s parse",
+      "Ġmet as",
+      "Ġtext ures",
+      "Ġstack s",
+      "M Hz",
+      "Ġf n",
+      "Ġult rac",
+      "ĠSh ould",
+      "V ec",
+      "n ine",
+      "inf inite",
+      "ĠLaw rence",
+      "ĠInvent ory",
+      "ĠPro state",
+      "Ġgest ure",
+      "ĠSuz uki",
+      "A bs",
+      "ric ane",
+      "ĠPeriod ic",
+      "M yc",
+      "if iable",
+      "Ġin efficient",
+      "Ġcoll apsed",
+      "Ġtopological ly",
+      "Ġprefer able",
+      "Ġbronch ial",
+      "ust on",
+      "Ġflex ion",
+      "our ney",
+      "trans lation",
+      "Ġepit axial",
+      "Ġirradi ance",
+      "Ġneighb ours",
+      "sw itch",
+      "Ġactu ators",
+      "S OD",
+      "m ir",
+      "di es",
+      "ik awa",
+      "ĠAL L",
+      "ĠR SV",
+      "ĠH EP",
+      "Ġend urance",
+      "conn ection",
+      "Ġgest ures",
+      "odon tic",
+      "ĠUn c",
+      "Ġdismut ase",
+      "H aving",
+      "m ix",
+      "Ġneuro genesis",
+      "Ġmyocardi um",
+      "ĠRuss ell",
+      "H ist",
+      "ĠS PI",
+      "tri azol",
+      "ag ulant",
+      "ĠRe quired",
+      "Ġsh RNA",
+      "ĠArth ur",
+      "Ġspaw ning",
+      "d ried",
+      "Ġrec tif",
+      "ĠÃ ī",
+      "Ġoste ogenic",
+      "re place",
+      "Ġgain ing",
+      "Ġneutral ization",
+      "ĠHart ree",
+      "Ġfollic les",
+      "Ġrelig ion",
+      "Ġd uplex",
+      "Ġtrans ients",
+      "amp ed",
+      "Ġmicrotub ules",
+      "int erest",
+      "Ġste els",
+      "B atch",
+      "Ġden aturation",
+      "ĠPhill ips",
+      "Ġqu iet",
+      "ĠB ureau",
+      "ĠR are",
+      "Ġqu ercetin",
+      "a ults",
+      "Ġel ution",
+      "uk a",
+      "ĠInter pretation",
+      "R V",
+      "ĠE SC",
+      "ĠK om",
+      "are ttes",
+      "Ġï ģĦ",
+      "Ġtra dition",
+      "Ġdiss ected",
+      "Ne igh",
+      "Ġshe aves",
+      "Ġbelong ed",
+      "ĠHist oric",
+      "ĠO E",
+      "Ġj son",
+      "lem ma",
+      "ĠY AP",
+      "ode xt",
+      "inter face",
+      "Ġextrem ity",
+      "cross ing",
+      "preced ented",
+      "acc ording",
+      "Ġconstruc tive",
+      "ĠStim ulation",
+      "ĠHF D",
+      "Ġwaven umber",
+      "Ġh rs",
+      "Ġpapill omavirus",
+      "Ġvom iting",
+      "Ġre activation",
+      "omet rically",
+      "ĠDim ensions",
+      "ob jects",
+      "ort on",
+      "ĠMat hem",
+      "ĠOl ive",
+      "Ġcros stalk",
+      "par tite",
+      "opath ies",
+      "ĠCN Ts",
+      "rous al",
+      "Ġcrow d",
+      "ĠLang muir",
+      "ĠT ox",
+      "echan ics",
+      "im us",
+      "ĠSh ock",
+      "tan h",
+      "ĠBrill ouin",
+      "Ġtransf erring",
+      "Ġellip se",
+      "ĠAd dition",
+      "ĠR ural",
+      "Ġgeodes ics",
+      "G EM",
+      "ĠP OS",
+      "ĠM ission",
+      "oc arp",
+      "ĠJ ane",
+      "L ie",
+      "f req",
+      "op ot",
+      "ĠV ibrio",
+      "ĠOb j",
+      "er ts",
+      "ĠTri als",
+      "C FT",
+      "ĠC odes",
+      "μ g",
+      "Ref erence",
+      "ĠF ung",
+      "ĠSup pression",
+      "h og",
+      "Ġresis tive",
+      "C hi",
+      "int ered",
+      "Ġpost menopausal",
+      "St atistical",
+      "ĠEd wards",
+      "Ġs es",
+      "Ġfarm ing",
+      "quar tile",
+      "cool ed",
+      "Ġnan op",
+      "ĠProb ing",
+      "ĠBern ard",
+      "un i",
+      "ie ties",
+      "ĠMark et",
+      "os um",
+      "ĠM essage",
+      "Ġaxi om",
+      "c g",
+      "ĠM oving",
+      "Res olution",
+      "Ġadsorb ent",
+      "Ġmult in",
+      "Ġin effective",
+      "prop ag",
+      "hard t",
+      "S aharan",
+      "W il",
+      "ĠI van",
+      "ir ubin",
+      "Ġtra bec",
+      "all i",
+      "ĠCD Cl",
+      "Ġse w",
+      "ĠIs s",
+      "Ġagg ression",
+      "ĠJ uan",
+      "Ġdispers ions",
+      "Ġaux in",
+      "F ET",
+      "l p",
+      "re ach",
+      "ĠP GE",
+      "che str",
+      "Ġlect ure",
+      "ĠD onald",
+      "sl ip",
+      "ĠHb A",
+      "ĠSec ure",
+      "ĠBe h",
+      "Ġdam ages",
+      "W H",
+      "alk yl",
+      "H a",
+      "ĠTh anks",
+      "Ġsensiti zation",
+      "Ġwat erm",
+      "Ġtw ins",
+      "Ġcultiv ar",
+      "Ġze olite",
+      "V ariable",
+      "ĠB ent",
+      "Ġanti sense",
+      "ĠHans en",
+      "reprene ur",
+      "ĠSN e",
+      "ĠEM G",
+      "Ġre acted",
+      "Ġover flow",
+      "Ġformal in",
+      "ĠUs ually",
+      "olyb den",
+      "Ġac ad",
+      "AT URE",
+      "Ġwavegu ides",
+      "Ġch unk",
+      "Ġmod ifies",
+      "Ġer yt",
+      "ĠZh ong",
+      "Ġgran ule",
+      "Ġc s",
+      "ĠGra de",
+      "Ġland marks",
+      "ur istic",
+      "Ġam ines",
+      "ĠIntr insic",
+      "Ġerrone ous",
+      "Ġlock down",
+      "yp ti",
+      "Ch ild",
+      "Ġunivers ities",
+      "Ġparas it",
+      "Ġign ition",
+      "T im",
+      "ar aj",
+      "ra vel",
+      "ĠL ands",
+      "ĠCirc ular",
+      "Ġrot ate",
+      "Pati ents",
+      "ĠW B",
+      "Ġmyel in",
+      "ĠWe iss",
+      "Ġdip olar",
+      "Ġfollic le",
+      "ĠWat son",
+      "ĠIn cor",
+      "Ġfound ations",
+      "ĠP ip",
+      "Ġpress ing",
+      "Ġforb idden",
+      "av an",
+      "Ġm Ab",
+      "un ion",
+      "ĠF resh",
+      "ĠCor p",
+      "fl oxacin",
+      "co ordinate",
+      "Ġsh unt",
+      "Ġconstit uted",
+      "anil ine",
+      "Ġtwe ets",
+      "ĠCh ow",
+      "Ġmob ilization",
+      "zy k",
+      "E ST",
+      "ne igh",
+      "ĠM eng",
+      "ĠRes Net",
+      "ĠJ et",
+      "Ġlumin ous",
+      "Ġstress ors",
+      "do es",
+      "trifluor omethyl",
+      "Ġconcer t",
+      "ĠCho ice",
+      "ph im",
+      "al coholic",
+      "oc hem",
+      "ilt ered",
+      "Ġpredict able",
+      "Ġt ran",
+      "ĠP ra",
+      "Ġval ves",
+      "Ġaut onomy",
+      "reg ulate",
+      "ĠBe ach",
+      "ĠOnt ology",
+      "Ġis ofl",
+      "Ġqu oted",
+      "ĠL ex",
+      "th y",
+      "Ġcompl aints",
+      "ĠT rees",
+      "Ġop posing",
+      "ĠAcc eler",
+      "con trast",
+      "Ġcompet ed",
+      "O E",
+      "ĠR oche",
+      "iss ance",
+      "Ġpe ace",
+      "ĠA im",
+      "Ġinfer tility",
+      "ĠAntarctic a",
+      "th ien",
+      "S umm",
+      "Ġjudg ments",
+      "am ides",
+      "Ġsp ill",
+      "Ġhere after",
+      "ĠCons tit",
+      "comput er",
+      "Ġbeg un",
+      "ocent ric",
+      "Ġp umps",
+      "med ium",
+      "ch ol",
+      "met allic",
+      "Ġfl ares",
+      "Ġpet roleum",
+      "Ġwith d",
+      "ĠThe atre",
+      "Ġun labeled",
+      "Ġregular ized",
+      "oster ic",
+      "ĠP FS",
+      "Ġun em",
+      "Ġpresent ly",
+      "Ġbuff ered",
+      "aff inity",
+      "ĠDem ographic",
+      "ĠK ondo",
+      "Ġcent uries",
+      "Ġmig ratory",
+      "aryn x",
+      "Ass ociated",
+      "anil ino",
+      "g rown",
+      "ĠEx ecutive",
+      "ĠE k",
+      "ĠH emat",
+      "ĠPl ayer",
+      "ĠCH D",
+      "f lex",
+      "ĠS ever",
+      "alth am",
+      "im pro",
+      "an et",
+      "ocy st",
+      "ĠA ster",
+      "CO L",
+      "ĠSimilar ity",
+      "ĠHow ard",
+      "Ġmultic ast",
+      "ĠEns emble",
+      "ì Ĺ",
+      "ol ys",
+      "ĠGen omics",
+      "Ġreson ators",
+      "Ġfist ula",
+      "on en",
+      "us ers",
+      "Ġhyp o",
+      "rog ens",
+      "Ġmed al",
+      "ĠM IP",
+      "Ġvolt am",
+      "Ġappreci ated",
+      "ĠP é",
+      "ĠGa ia",
+      "Ġbuck ling",
+      "Ġcongru ence",
+      "fur yl",
+      "ĠEp stein",
+      "Ġcasc ades",
+      "g old",
+      "Ġan hyd",
+      "Ġgrad uated",
+      "M emory",
+      "ĠInd ustry",
+      "ĠSch neider",
+      "Ġemploy ee",
+      "ĠCor n",
+      "M AC",
+      "ro ve",
+      "rop od",
+      "s ervice",
+      "ĠOx idation",
+      "Ġenum eration",
+      "m ad",
+      "ĠCl ose",
+      "ĠMod ular",
+      "Ġprogen y",
+      "Ġg t",
+      "read ing",
+      "ĠInd ic",
+      "opath ologic",
+      "ĠPFN GL",
+      "X L",
+      "c is",
+      "ĠM ike",
+      "ĠB BB",
+      "ĠExt reme",
+      "ĠCho ose",
+      "Ġhoriz ontally",
+      "ĠASS ERT",
+      "Ġglucocortic oid",
+      "B ay",
+      "Ġp df",
+      "Ġcontain ers",
+      "ĠL OC",
+      "ĠY ield",
+      "opro te",
+      "Ġfruct ose",
+      "ĠI CC",
+      "Ġdec id",
+      "rim idine",
+      "Ġfrag mented",
+      "Ġisomorphism s",
+      "Ð ¼",
+      "Ġintegr ates",
+      "Ġfib ration",
+      "ĠâĬ ¤",
+      "Ġxen ograft",
+      "nucle on",
+      "ĠC SP",
+      "Ġs ut",
+      "ĠSp ir",
+      "Ġdiss oci",
+      "ĠT BI",
+      "ĠFor ces",
+      "Ġhyper surface",
+      "Ġmy osin",
+      "ĠQueens land",
+      "N eg",
+      "ĠU RL",
+      "b ind",
+      "Ap plied",
+      "ĠD ob",
+      "ĠK E",
+      "Ġmem or",
+      "ĠArab ic",
+      "ĠL ateral",
+      "ĠSt art",
+      "n ose",
+      "ti bility",
+      "as ters",
+      "Ġus ability",
+      "Ġinc enti",
+      "ym n",
+      "ĠAnaly tic",
+      "P et",
+      "ĠM ask",
+      "W orld",
+      "b rand",
+      "Ġelim inates",
+      "Ġmer it",
+      "ĠPhilipp ines",
+      "ĠB CL",
+      "ĠO ri",
+      "Ġparad igms",
+      "ĠIn ters",
+      "riz ona",
+      "Ġcon ception",
+      "Ġrel ied",
+      "ĠJ oe",
+      "ĠAp ple",
+      "Ġlight weight",
+      "mort em",
+      "ol ig",
+      "Ġv iz",
+      "Ġst ones",
+      "Ġkey words",
+      "ĠSecret ary",
+      "T N",
+      "old er",
+      "ĠInt estinal",
+      "Ġpossess ed",
+      "Ġmonoton icity",
+      "em itting",
+      "ĠDef ining",
+      "ĠPar ticularly",
+      "Ġautomorphism s",
+      "Ġeryt hemat",
+      "ĠW aters",
+      "ĠCycl ic",
+      "maxim al",
+      "xt y",
+      "ĠS ad",
+      "Ġur anium",
+      "Ġhypoth alamus",
+      "ĠSU MO",
+      "Ġdeal t",
+      "Ġk its",
+      "Ġpain ting",
+      "ĠS ier",
+      "ch ool",
+      "OD O",
+      "sur faces",
+      "ĠP neum",
+      "organ ized",
+      "ĠC PT",
+      "Ġins oluble",
+      "ĠCo herent",
+      "Ġrecess ive",
+      "Ġb ivariate",
+      "Ġed it",
+      "Ġnation wide",
+      "M ODE",
+      "c hest",
+      "ĠS LC",
+      "Ġintra peritoneal",
+      "ĠDis ordered",
+      "Ġinsu fficiency",
+      "ie v",
+      "iaz ole",
+      "W rite",
+      "ĠD ATA",
+      "tor al",
+      "Ġqual ities",
+      "Ġpossess ing",
+      "ĠM ats",
+      "Ġretin opathy",
+      "ĠB K",
+      "Ġnovel ty",
+      "ce ans",
+      "Ġreserv es",
+      "ĠNAD H",
+      "Ġisother m",
+      "Ġsoldi ers",
+      "p b",
+      "iter pen",
+      "ĠAg ents",
+      "z u",
+      "Ġunw anted",
+      "Ġhyper parameters",
+      "ec an",
+      "ĠS ES",
+      "ĠF G",
+      "ĠN avig",
+      "Ġtriang ulation",
+      "Ġnetwork ing",
+      "Ġpoly styrene",
+      "Ġinduc tively",
+      "brevi ations",
+      "Ġneurom uscular",
+      "ĠL inux",
+      "stud ied",
+      "ĠBe ing",
+      "Ġdef iciencies",
+      "ĠMat rices",
+      "Ġwe aring",
+      "Ġhad rons",
+      "am yl",
+      "Ġdisc ourse",
+      "och lor",
+      "ĠMel an",
+      "ĠL an",
+      "V L",
+      "Ġmunic ipal",
+      "Ġenroll ment",
+      "ĠS ymmetric",
+      "Ġdiscipl ines",
+      "ĠBar on",
+      "Res earch",
+      "Ġmagne tite",
+      "om ide",
+      "polar ization",
+      "le ys",
+      "Ġseem ingly",
+      "hep atic",
+      "Ġcl o",
+      "ĠQu atern",
+      "Ġcompe tit",
+      "R equ",
+      "ga uge",
+      "Ġhydro chloride",
+      "drop out",
+      "pan el",
+      "Ġaspir in",
+      "ĠR UN",
+      "Ġrib bon",
+      "Ġinacc urate",
+      "ĠP all",
+      "duc ers",
+      "Through out",
+      "Ġcell ul",
+      "Ġsusp ect",
+      "Ġalle lic",
+      "Ġsn ake",
+      "ordin ated",
+      "ĠAut ophagy",
+      "Ġe ig",
+      "Ġr if",
+      "ĠKen nedy",
+      "Ġbot tle",
+      "ĠY outh",
+      "aw ed",
+      "linear ity",
+      "uk er",
+      "ĠO X",
+      "ext ension",
+      "Ġw ard",
+      "ĠComplex es",
+      "Ġbios ensor",
+      "ĠCart an",
+      "d n",
+      "Ġs onic",
+      "Ġindex ing",
+      "Ġd v",
+      "rel iable",
+      "p k",
+      "RE NT",
+      "Ġt anks",
+      "ĠH et",
+      "ĠW ing",
+      "ĠCu O",
+      "Ġprint f",
+      "Ġlumin osities",
+      "c ourse",
+      "Ġsc ram",
+      "Ġsam pler",
+      "Ġmulti pliers",
+      "Def ault",
+      "od il",
+      "int r",
+      "sequ encing",
+      "Ġtrans missions",
+      "ĠWh it",
+      "ĠOp portun",
+      "Ġintern ally",
+      "Ġacknowled ges",
+      "ĠE dition",
+      "Ġarter i",
+      "Ġalb edo",
+      "ĠNucle otide",
+      "Ġy es",
+      "ĠRel ativistic",
+      "Ġv otes",
+      "ĠForm ulation",
+      "usc itation",
+      "Ġconcurrent ly",
+      "u in",
+      "Ġnon invasive",
+      "Ġprim ates",
+      "μ l",
+      "Ġsubt ropical",
+      "g un",
+      "ĠS outheast",
+      "ö n",
+      "Ġequ ator",
+      "Ġwork shop",
+      "Ġsch ist",
+      "und ant",
+      "ĠMOD IS",
+      "t ar",
+      "Ġa eg",
+      "Ġplot ting",
+      "ĠD ET",
+      "Man ager",
+      "un ed",
+      "oxif en",
+      "ĠIn ver",
+      "Ġx anth",
+      "ĠSer ver",
+      "Ġstret ched",
+      "Gl obal",
+      "C ore",
+      "ĠWe ber",
+      "y ard",
+      "Ġexpl ores",
+      "ĠBi ography",
+      "SN P",
+      "ĠNeut rino",
+      "Ġkilomet res",
+      "Ġcomm utes",
+      "Ġaccept ability",
+      "ĠAntib odies",
+      "ic ol",
+      "Ġmus eum",
+      "Ġden it",
+      "Ġextrap olated",
+      "Ġacetyl choline",
+      "T oken",
+      "ĠF ock",
+      "ond e",
+      "Ġdiscrimin ative",
+      "ĠM ant",
+      "Ġess ence",
+      "cel and",
+      "ĠCh air",
+      "Ġintegr ative",
+      "ĠS PD",
+      "hen ium",
+      "arbon ate",
+      "B ASE",
+      "reg ulates",
+      "p atch",
+      "Ġd ib",
+      "Ġanti symmetric",
+      "Ġwear able",
+      "Ed ge",
+      "re ts",
+      "Ġperce ive",
+      "ĠMagn esium",
+      "ad ows",
+      "Ġdis posal",
+      "Ġair port",
+      "ause a",
+      "f its",
+      "Ġnec ro",
+      "ĠS IN",
+      "ĠD uc",
+      "ĠRe ading",
+      "b ys",
+      "Ġreflec tive",
+      "h is",
+      "omet ries",
+      "Ġvi rial",
+      "Ġartif icially",
+      "child ren",
+      "ĠUltras ound",
+      "VI EW",
+      "Ġsc ulpt",
+      "Ġsur f",
+      "Ġsex ually",
+      "Ġgeomet rically",
+      "Ġdivis ors",
+      "Ġiniti atives",
+      "acc i",
+      "Ġkeratin ocytes",
+      "a R",
+      "aro t",
+      "Ġïĥ ¨",
+      "comput ed",
+      "ĠTC GA",
+      "psych ological",
+      "ĠM AN",
+      "ĠM PC",
+      "tic ing",
+      "lim iting",
+      "am ins",
+      "Ġsurfact ants",
+      "ĠSer b",
+      "Ġrhyth ms",
+      "ĠRout ing",
+      "w ang",
+      "Ġmicro structures",
+      "oph ytes",
+      "Ġanalges ic",
+      "F OR",
+      "qu al",
+      "Ġpubl ish",
+      "ĠTim ing",
+      "por ous",
+      "rang ing",
+      "er on",
+      "ĠZ i",
+      "ĠMarsh all",
+      "Wid th",
+      "Ġis omers",
+      "ĠÂ ·",
+      "phen oxy",
+      "Ġure th",
+      "ro bl",
+      "Ġmention ing",
+      "o zyme",
+      "ĠL ud",
+      "Ġop position",
+      "Ġaband oned",
+      "Ġrout ines",
+      "ĠH ST",
+      "mut ex",
+      "c oded",
+      "e ating",
+      "ter t",
+      "emicon ductor",
+      "d w",
+      "Ġbary ons",
+      "Ġleuc ine",
+      "ot ron",
+      "Ġend os",
+      "Ġreproduc es",
+      "Ġanalges ia",
+      "Ġimmunore activity",
+      "ĠPre p",
+      "ĠGarc ÃŃa",
+      "Ġinco herent",
+      "an ed",
+      "le pton",
+      "and ra",
+      "ul ae",
+      "ĠH idden",
+      "F V",
+      "Ġgeneral izes",
+      "ĠSte vens",
+      "ĠF oster",
+      "Ġfresh ly",
+      "Ġh f",
+      "Den ote",
+      "o es",
+      "ĠD in",
+      "Ġdet ox",
+      "Ġdec oupled",
+      "Ġsepar ations",
+      "ucle otide",
+      "Ġelect rophysiological",
+      "ĠBAL B",
+      "Q TL",
+      "ĠA Ch",
+      "ĠRe le",
+      "que z",
+      "Mn O",
+      "ect ures",
+      "Ġis cha",
+      "Ġins ulators",
+      "cell ulose",
+      "ĠFL AG",
+      "omb ic",
+      "ĠUs ed",
+      "j iang",
+      "exp ansion",
+      "ĠRep eat",
+      "ĠRes erve",
+      "ab elian",
+      "ĠH unting",
+      "G RO",
+      "ly te",
+      "ĠB ark",
+      "Ġcre ative",
+      "Ġb end",
+      "el erated",
+      "dis h",
+      "Ġhigh way",
+      "Ġcross ings",
+      "j ust",
+      "on o",
+      "ull ivan",
+      "ĠDe ad",
+      "Ġtrade off",
+      "e on",
+      "og ical",
+      "experim ent",
+      "Ġconf ers",
+      "ĠD ot",
+      "Ġco ils",
+      "Ġax ion",
+      "ĠIR S",
+      "ĠÅ ©",
+      "Ġglac ier",
+      "ĠMosc ow",
+      "ĠS pringer",
+      "Ġinv is",
+      "ĠArn old",
+      "Un iversity",
+      "at tern",
+      "per or",
+      "ĠLim its",
+      "Ġincomp atible",
+      "r ather",
+      "ĠT es",
+      "Ġfail ing",
+      "Ġthick ening",
+      "Ġest radiol",
+      "as se",
+      "Ġnecess it",
+      "Ġsacrific ed",
+      "ĠS ear",
+      "ĠNor the",
+      "raise box",
+      "ĠS low",
+      "ĠM unic",
+      "Ġlear ner",
+      "igen ic",
+      "Ġderm atitis",
+      "ut en",
+      "Ġde er",
+      "Ġhist amine",
+      "L at",
+      "M al",
+      "il ly",
+      "Ġge ochemical",
+      "Ġspermat ozoa",
+      "Ġv inyl",
+      "em et",
+      "Ġeffect ors",
+      "ĠEncycl opedia",
+      "Ġord inal",
+      "Ġcontrovers y",
+      "ĠPers pectives",
+      "ovirus es",
+      "mark ed",
+      "ĠS PE",
+      "ĠN utri",
+      "Ġad here",
+      "ĠHigh way",
+      "Ġdistill ation",
+      "MR T",
+      "ple tion",
+      "Ġannih il",
+      "Ġwave function",
+      "Ġconfig ured",
+      "Ġmeth ionine",
+      "L ow",
+      "s ensor",
+      "ĠS now",
+      "S ample",
+      "Ġdef initely",
+      "ĠMet h",
+      "r ypt",
+      "Ġprom pted",
+      "Ġmonol ith",
+      "ĠEn vironments",
+      "t m",
+      "ĠCO D",
+      "or is",
+      "equ ations",
+      "âĺ Ĩ",
+      "ĠNe ighbor",
+      "Ġimag ine",
+      "ĠUs ers",
+      "ĠCam era",
+      "ĠMod ification",
+      "ĠAtt acks",
+      "Ġinhal ation",
+      "á º",
+      "Ġventi l",
+      "ĠN U",
+      "ĠCon trast",
+      "Ġconf ining",
+      "S ervice",
+      "W allis",
+      "ĠA TR",
+      "Ġsub duction",
+      "Ġïģ ¢",
+      "Ġtit ration",
+      "R oche",
+      "v iv",
+      "Ġbe ars",
+      "bol a",
+      "Ġblind ed",
+      "meas ures",
+      "ĠSt ack",
+      "occ urrence",
+      "Ġperme ation",
+      "l ar",
+      "ept ors",
+      "ĠD IF",
+      "cor rhiz",
+      "ĠV isc",
+      "fig urable",
+      "Ġschedul er",
+      "Ġoccas ions",
+      "ambo o",
+      "Ġam p",
+      "g ain",
+      "ĠC it",
+      "Ġpreced ed",
+      "Ġtac tile",
+      "Ġïĥ ¦",
+      "gener ic",
+      "Ġretro grade",
+      "Ġf ans",
+      "Ġf isher",
+      "Ġl ights",
+      "ee per",
+      "Ġundes irable",
+      "w ald",
+      "emb ol",
+      "Ġwr ist",
+      "Ġauthor ized",
+      "Ġchond rocytes",
+      "ĠE PA",
+      "ne u",
+      "ĠOper ations",
+      "Ġche ap",
+      "Ġan ionic",
+      "ĠO regon",
+      "c ot",
+      "re ason",
+      "ex istence",
+      "ĠFin ancial",
+      "olybden um",
+      "c us",
+      "ĠN ON",
+      "Ġlock ed",
+      "B it",
+      "S il",
+      "m ixing",
+      "ĠS ites",
+      "aprote obacteria",
+      "ĠIn ner",
+      "Ġcar c",
+      "Ġbi otic",
+      "ĠFl ag",
+      "Ġmag ic",
+      "kine tic",
+      "ic ted",
+      "Ġbul b",
+      "sup set",
+      "pe z",
+      "deriv ative",
+      "Ġe IF",
+      "ĠR ough",
+      "di rectional",
+      "ex it",
+      "ax y",
+      "xt ures",
+      "phim urium",
+      "ĠT Fs",
+      "ath in",
+      "Ġor ch",
+      "Ġspect ro",
+      "duct ase",
+      "quin olin",
+      "Ġgras p",
+      "Ġpar sing",
+      "Ġdiffic ile",
+      "ĠLD H",
+      "ĠJup iter",
+      "ĠF IF",
+      "ĠPri ze",
+      "Ġinten tions",
+      "s ession",
+      "pow ered",
+      "ĠB am",
+      "ph asic",
+      "Ġign oring",
+      "ĠRichard son",
+      "princ iples",
+      "Ġoffic ially",
+      "C t",
+      "Ġinc on",
+      "ĠReg ulates",
+      "Ġm isc",
+      "ĠE Z",
+      "Ġsyn onym",
+      "Ġunfold ing",
+      "ĠD EC",
+      "ĠR X",
+      "PD F",
+      "Ġbran es",
+      "typ ically",
+      "Ġc ages",
+      "if olia",
+      "ug u",
+      "oll en",
+      "Ġtable t",
+      "ĠS ah",
+      "ĠP VD",
+      "Ġal ert",
+      "Ġformer ly",
+      "ĠKR AS",
+      "s un",
+      "Ġde acetyl",
+      "M er",
+      "Ġskew ed",
+      "ĠPle istocene",
+      "ĠB etter",
+      "ĠH ud",
+      "ĠBro ok",
+      "Ġp ts",
+      "ĠH U",
+      "om o",
+      "ag rass",
+      "Ġenvironment ally",
+      "Ġhon est",
+      "ĠN ine",
+      "Ġpig ments",
+      "l inks",
+      "ĠT OP",
+      "ĠCytoplas m",
+      "G ib",
+      "Ġaccess ing",
+      "mi as",
+      "Ġexplos ive",
+      "Ġres ide",
+      "art an",
+      "Ġtransition al",
+      "Ġun precedented",
+      "Ġ rom",
+      "ĠTNF α",
+      "Ġprecipit ated",
+      "Ġt ie",
+      "IS S",
+      "Ġthick er",
+      "ĠLat ent",
+      "ĠValue Error",
+      "d q",
+      "d ma",
+      "Ġchrom atic",
+      "ĠSub section",
+      "ĠF ACS",
+      "Ġrenormal ized",
+      "P rop",
+      "m TOR",
+      "ĠH CO",
+      "Ġover lo",
+      "bs iella",
+      "yl obacter",
+      "Ġneuro imaging",
+      "Ġassembl age",
+      "Ġexp ands",
+      "Ġî Ī",
+      "ĠF un",
+      "Ġc itation",
+      "IK V",
+      "Ġtro ops",
+      "in istic",
+      "Ġc ubes",
+      "Ġf ont",
+      "ĠH os",
+      "ger ies",
+      "Ġsuccess ively",
+      "Ġdeco herence",
+      "S pringer",
+      "h in",
+      "at ine",
+      "ĠâĪ ¥",
+      "SA S",
+      "é t",
+      "ĠSed iment",
+      "u ously",
+      "ĠW ars",
+      "ind icated",
+      "Ġfl ask",
+      "A IDS",
+      "Ġc ra",
+      "ĠL ot",
+      "Ġprim al",
+      "Ġjus tice",
+      "z ag",
+      "Ġmax illary",
+      "Ġgeneral izations",
+      "uel a",
+      "Ġtag ging",
+      "Ġpup il",
+      "Ġin expensive",
+      "Ġw atch",
+      "ĠA MD",
+      "ĠF ir",
+      "Ġneuro blastoma",
+      "Ġmaxim izes",
+      "ĠObs erved",
+      "mi xture",
+      "Ġopportun istic",
+      "t rial",
+      "ah an",
+      "Ġïģ ¬",
+      "Ġcat ar",
+      "ĠControl s",
+      "ĠNew man",
+      "Ġmicro structural",
+      "bor ns",
+      "Ġoxygen ation",
+      "ĠMac ro",
+      "ĠJ ak",
+      "plic ating",
+      "Ġolig odend",
+      "Ġres orption",
+      "Ġd orm",
+      "Ġsol vers",
+      "ĠK ruskal",
+      "ĠRe volution",
+      "ĠGast ro",
+      "Dri ven",
+      "Ġt iter",
+      "Ġo ri",
+      "ĠP CL",
+      "Ġwet lands",
+      "Ġar ticular",
+      "CC A",
+      "en oic",
+      "Ġt rick",
+      "oper iod",
+      "ĠCoch rane",
+      "ad ay",
+      "ĠC erebral",
+      "Ġmod ulators",
+      "ĠS SC",
+      "Ġactiv ations",
+      "Ġadap ting",
+      "ĠScal able",
+      "n one",
+      "p ip",
+      "Ġpri vi",
+      "ĠPseud o",
+      "Ġdisapp ears",
+      "ĠE ur",
+      "Ġuncon strained",
+      "Ġsub mit",
+      "Ġrep utation",
+      "at ar",
+      "ĠB ai",
+      "ari ans",
+      "ĠInt racellular",
+      "tre es",
+      "Ġwet ting",
+      "ĠFran ces",
+      "Ġel igibility",
+      "fold er",
+      "ĠSta ff",
+      "ok i",
+      "Ġstrengthen ed",
+      "ĠC ob",
+      "ter al",
+      "ĠY east",
+      "by e",
+      "dec oder",
+      "Ġrain bow",
+      "perturb ed",
+      "v c",
+      "Ġsupplement al",
+      "Ġbir ths",
+      "W O",
+      "con c",
+      "stit ution",
+      "hy brid",
+      "Ġk i",
+      "Ġhyp ere",
+      "ĠS MA",
+      "form ula",
+      "Ġund efined",
+      "na phth",
+      "Ġdecl ining",
+      "Ġshield ing",
+      "Y au",
+      "Ġre ver",
+      "ĠW ilk",
+      "Ġdec imal",
+      "H CO",
+      "ang ered",
+      "Ġeryth rocyte",
+      "ĉĉ ĠĠĠ",
+      "n uclear",
+      "Ġabnorm ality",
+      "P res",
+      "Par ticipants",
+      "ĠW agner",
+      "Ġfibr ils",
+      "Ġfet us",
+      "ĠEx press",
+      "requ est",
+      "min imum",
+      "ĠBo oks",
+      "het amine",
+      "us hes",
+      "ĠB ach",
+      "ĠD OS",
+      "lect ric",
+      "ĠTw een",
+      "ĠHug hes",
+      "Ġm artens",
+      "Ġn ematic",
+      "Ġexperiment ation",
+      "ĠPark er",
+      "Ġepis odic",
+      "Ġte lem",
+      "AD E",
+      "col umns",
+      "Ġfundament ally",
+      "en et",
+      "ĠV l",
+      "ear th",
+      "Ġquanti le",
+      "ĠRe plication",
+      "Ġcle ared",
+      "En ergy",
+      "Sm ith",
+      "Ġantidepress ant",
+      "m x",
+      "p mod",
+      "am id",
+      "Ġser otype",
+      "Ġundergrad uate",
+      "ĠA rizona",
+      "Ġp ushed",
+      "ul u",
+      "ĠN IC",
+      "Ġrhe ological",
+      "ome gal",
+      "ĠQ ing",
+      "or ch",
+      "ir med",
+      "ĠQu ery",
+      "Ġsand wich",
+      "Ġclinic ian",
+      "ĠEllip tic",
+      "ĠMe h",
+      "DE V",
+      "ĠDeterm ining",
+      "alc ogen",
+      "b ench",
+      "az ep",
+      "ĠMiss iss",
+      "ti zing",
+      "ĠR BC",
+      "Ġofficial s",
+      "T ag",
+      "k T",
+      "lu ence",
+      "ĠRo om",
+      "Ġlect in",
+      "bar a",
+      "k yl",
+      "ON D",
+      "ĠD ose",
+      "Ġpr ism",
+      "Ġreduc tive",
+      "ĠSpect roscopic",
+      "od ied",
+      "col one",
+      "ĠCON FIG",
+      "Ġbr ittle",
+      "in verse",
+      "ĠB uff",
+      "yt ocin",
+      "Ġform ations",
+      "ĠCon ventional",
+      "pre v",
+      "Ġferr ite",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ Ġ",
+      "Ġadop ts",
+      "ĠMi ocene",
+      "man agement",
+      "ĠCR F",
+      "ĠHel m",
+      "Ġdoubl ed",
+      "ĠEFF ECT",
+      "Ġd ance",
+      "struc tions",
+      "ra it",
+      "if ers",
+      "ell ip",
+      "ut ting",
+      "pro f",
+      "ĠQ in",
+      "Ġab sc",
+      "Ġexplo its",
+      "Ġcy ber",
+      "def inition",
+      "ĠCoron ary",
+      "Ġdet erg",
+      "ĠPer ception",
+      "ĠCur ves",
+      "Ġnemat odes",
+      "Ġlist ening",
+      "Ġcatal ase",
+      "C oll",
+      "r é",
+      "isl ative",
+      "Ġarri ving",
+      "Ġviol ating",
+      "Ð ´",
+      "he tics",
+      "ĠJ ar",
+      "con cept",
+      "Ġbr ush",
+      "immun ity",
+      "Ġfinger print",
+      "res id",
+      "Ġelev ations",
+      "ock ets",
+      "Ġcatech ol",
+      "и Ñ",
+      "Ġprecipit ates",
+      "Ġsoc cer",
+      "ins ulin",
+      "Ġpurs ue",
+      "ĠI CA",
+      "ĠPol ice",
+      "ĠMur phy",
+      "T ask",
+      "ĠC oc",
+      "ĠH abit",
+      "ĠK P",
+      "Ġfl oral",
+      "Ġh un",
+      "Ġhydrogen ation",
+      "Ġsp ong",
+      "Ġch imeric",
+      "ĠK och",
+      "g on",
+      "ĠSch ur",
+      "ĠGre ater",
+      "R X",
+      "Ġc ing",
+      "ĠW altham",
+      "ang ling",
+      "Ġcoun ties",
+      "Ġlam ina",
+      "Ġco uncil",
+      "s ort",
+      "ĠB arc",
+      "ĠD ow",
+      "ĠZ eng",
+      "Ġdev ised",
+      "uit able",
+      "Ġmethyl ene",
+      "Ġsuperior ity",
+      "Ġepiderm is",
+      "Ġp rag",
+      "ĠP ED",
+      "threat ening",
+      "ish i",
+      "Ġe psilon",
+      "add ress",
+      "ENT AL",
+      "ĠB le",
+      "ĠAnton io",
+      "o other",
+      "ĠAg ar",
+      "Ġneighborhood s",
+      "Ġshorten ed",
+      "ST ATE",
+      "ĠSer ial",
+      "M AR",
+      "O U",
+      "Ġencaps ulation",
+      "ĠCons ortium",
+      "D r",
+      "pro file",
+      "Ġem itter",
+      "Ġnec rotic",
+      "ĠAut onomous",
+      "ĠPhosph orylation",
+      "min im",
+      "anth in",
+      "ĠS ph",
+      "ĠG ur",
+      "di hydroxy",
+      "dist ributed",
+      "ĠRP MI",
+      "st ones",
+      "Ġhyper fine",
+      "Ġis let",
+      "ĠS lo",
+      "plet ely",
+      "Ġin activated",
+      "ĠAgric ulture",
+      "Ġtrem end",
+      "Ġevery one",
+      "omp onent",
+      "Zn O",
+      "MP I",
+      "ĠDi amond",
+      "Ġ⣠¨",
+      "C ost",
+      "Ġdis abilities",
+      "in ver",
+      "ĠC ensus",
+      "ech o",
+      "Ġveget ative",
+      "Ġwilling ness",
+      "Ġrec ap",
+      "ĠConst raint",
+      "ĠP atrick",
+      "Ġover t",
+      "Ġmo ieties",
+      "or ax",
+      "ip pi",
+      "Di rect",
+      "Ġcar ies",
+      "Ġlocal ities",
+      "lat tices",
+      "ĠExpl oration",
+      "ĠA W",
+      "Ġloc king",
+      "Ġcoinc ident",
+      "Ġmultim edia",
+      "Ġtempor arily",
+      "ĠC aus",
+      "enc ia",
+      "Ġweather ing",
+      "ĠHelic obacter",
+      "ĠTh ings",
+      "hip s",
+      "m oving",
+      "Ġs igmoid",
+      "is in",
+      "ĠB ec",
+      "Ġmicro grams",
+      "bound s",
+      "ĠCol umn",
+      "Ġcommut ing",
+      "ĠJ en",
+      "Ġhour ly",
+      "M SC",
+      "Ġattend ance",
+      "ĠâIJ £",
+      "ĠE O",
+      "pro g",
+      "Ġrap amycin",
+      "ĠPredict ors",
+      "ĠRetrie ved",
+      "Ġsub species",
+      "Ġderiv es",
+      "ĠÄ ¤",
+      "ĠGener ating",
+      "ann ers",
+      "Ġvol at",
+      "Ġvis iting",
+      "ĠCalc ulations",
+      "ñ a",
+      "Ġdes ert",
+      "Ġexpect ancy",
+      "BM Cs",
+      "ĠExpl o",
+      "Ġtrav elling",
+      "ic um",
+      "Ġsub division",
+      "Ġcross linking",
+      "benz oth",
+      "ĠT on",
+      "RE N",
+      "Ġle th",
+      "rab bit",
+      "ĠAb ove",
+      "ul ted",
+      "Ġcon stric",
+      "J ones",
+      "z hou",
+      "ver n",
+      "ĠL ady",
+      "ĠBu ffer",
+      "ĠControll ing",
+      "Ġmulti scale",
+      "nik ov",
+      "acy cl",
+      "Ġprost hesis",
+      "A f",
+      "ĠCor ps",
+      "struc ted",
+      "G rid",
+      "in ning",
+      "old ing",
+      "Ġthi ol",
+      "ik ov",
+      "âĢ¢âĢ¢ âĢ¢",
+      "Ġgovern ments",
+      "rap ping",
+      "Ġthromb ocyt",
+      "L eg",
+      "R Y",
+      "ĠI celand",
+      "ocy cle",
+      "ĠMem orial",
+      "g ot",
+      "Ġid em",
+      "ĠBu ild",
+      "olip oprotein",
+      "D V",
+      "Ġph thal",
+      "rich ment",
+      "ĠHa em",
+      "Ġansw ering",
+      "ĠI J",
+      "Ġtrans gene",
+      "Ġre named",
+      "ĠImage J",
+      "Ġcass ette",
+      "Ġcoales cence",
+      "Ġcomp action",
+      "Ġwild life",
+      "Ġw ins",
+      "Ġsuper novae",
+      "enter ic",
+      "isp here",
+      "Ġtrack er",
+      "Ġevid ences",
+      "Ġcom orbidity",
+      "ĠR ules",
+      "ph asing",
+      "ĠLange vin",
+      "ĠF it",
+      "Ġpsy chiat",
+      "Ġbreak through",
+      "Ġch olinergic",
+      "ĠMet all",
+      "bre eding",
+      "itin ib",
+      "Ġsol o",
+      "abl ing",
+      "eli ef",
+      "osc ill",
+      "re v",
+      "ary a",
+      "Ġgood ness",
+      "ĠPB E",
+      "Ġa wards",
+      "Ġc rani",
+      "Ġphot ograp",
+      "aren ts",
+      "Ġfix es",
+      "r ÃŃ",
+      "ass uming",
+      "Ġcongru ent",
+      "ĠM other",
+      "ĠN ap",
+      "ĠPro c",
+      "Ġcategor ization",
+      "in ch",
+      "ĠH orm",
+      "ĠInter ventions",
+      "Ġnone quilibrium",
+      "Ġencryp ted",
+      "prim ary",
+      "i ens",
+      "l ac",
+      "ram s",
+      "Ġbo ards",
+      "ĠH ell",
+      "charg ed",
+      "Ġperi operative",
+      "em p",
+      "ĠInvol vement",
+      "R uss",
+      "un ivers",
+      "ĠD J",
+      "Ġdisag reement",
+      "Ġper t",
+      "Ġstrom a",
+      "Ġcalc ite",
+      "Ġrot ary",
+      "Ġmethyl transferase",
+      "Ġancest ry",
+      "ĠW itten",
+      "CR C",
+      "ure tic",
+      "ophy ta",
+      "provid ed",
+      "Ġcorresponding ly",
+      "big cap",
+      "ĠAg ilent",
+      "Ã «",
+      "ro oms",
+      "Ġdis ent",
+      "Ġdil utions",
+      "ĠMy el",
+      "Ġquas ar",
+      "Ġtil ted",
+      "Ġinternal ization",
+      "ĠPri vate",
+      "ĠFried man",
+      "Ġsevent h",
+      "ĠCl osed",
+      "CT C",
+      "g ren",
+      "ĠColomb ia",
+      "od el",
+      "Ġpoli tics",
+      "ĠMSS M",
+      "Ġm ate",
+      "Ġcom mod",
+      "ĠR us",
+      "Ġanest hetized",
+      "t ogether",
+      "ĠB CS",
+      "ew ski",
+      "romagn et",
+      "ĠC un",
+      "Ġcur ative",
+      "Ġim putation",
+      "Ġcarb ide",
+      "D FT",
+      "ns ic",
+      "be e",
+      "Ġspl en",
+      "ĠMary land",
+      "Ġoligonucle otide",
+      "ĠVeg et",
+      "buff ered",
+      "N ational",
+      "le tic",
+      "ĠS yl",
+      "Ġse ab",
+      "ardi al",
+      "Ġport ray",
+      "Ġaberr ations",
+      "Ġst orms",
+      "ĠSh an",
+      "ĠGen Bank",
+      "iss a",
+      "Ġc et",
+      "Ġben ch",
+      "ĠRecommend ations",
+      "Ġtri ples",
+      "Ġïĥ ¥",
+      "ĠNeu ros",
+      "Ġdisc om",
+      "se ason",
+      "ĠEx ec",
+      "chang ing",
+      "Ġarri ves",
+      "H ash",
+      "m RNA",
+      "Ġf ric",
+      "as a",
+      "ob ia",
+      "Ġpost synaptic",
+      "optim izer",
+      "ĠCloud s",
+      "Ġhyper sensitivity",
+      "v acc",
+      "ĠS ig",
+      "ph ilic",
+      "Ġground ed",
+      "ĠW an",
+      "ĠCal abi",
+      "ĠMach ines",
+      "Ġaxis ymmetric",
+      "ĠSte ve",
+      "Ġpull ed",
+      "ĠEx cel",
+      "Ġdiamond s",
+      "K R",
+      "W est",
+      "ĠD est",
+      "Ġann ular",
+      "Ġarch ive",
+      "Ġparench yma",
+      "ĠE H",
+      "ó pez",
+      "Ġunp ublished",
+      "Ġs outheastern",
+      "Ġn ests",
+      "dim ensions",
+      "lat itude",
+      "O rig",
+      "ec ed",
+      "ĠD raw",
+      "red shift",
+      "Ġam yl",
+      "omyel itis",
+      "W hy",
+      "c aro",
+      "i q",
+      "ass ess",
+      "ĠCont in",
+      "Ġchir ality",
+      "mat ical",
+      "Ġchaper one",
+      "Ġendometri osis",
+      "re lu",
+      "Ġconver ged",
+      "bro ad",
+      "ĠIter ative",
+      "Ġvascul ature",
+      "f und",
+      "ĠF ly",
+      "Ġanti genic",
+      "Ġmening itis",
+      "Ġent ails",
+      "hor n",
+      "Ġlocom otor",
+      "iz ard",
+      "Ġun even",
+      "par ity",
+      "pack et",
+      "tub ulin",
+      "Ġsew age",
+      "Ġdec entralized",
+      "Ġgra fted",
+      "Ġse p",
+      "ĠExt ensive",
+      "Ġspl ine",
+      "qu er",
+      "arch it",
+      "Ġprim ate",
+      "Ġïģ ±",
+      "pyrim idin",
+      "ĠS AP",
+      "Ġunder lie",
+      "Ġanalyz es",
+      "ĠC CA",
+      "rec ogn",
+      "IP T",
+      "Diff erent",
+      "ĠTE ST",
+      "Ġunf avorable",
+      "ed ic",
+      "ĠAb normal",
+      "pyrim idine",
+      "ur ine",
+      "embed ded",
+      "var ies",
+      "otrop in",
+      "Ġsem en",
+      "Ġtransmit tance",
+      "Ġab ras",
+      "Ġó ¸Ģł",
+      "Ġtriglycer ide",
+      "b undle",
+      "ĠY b",
+      "ĠCar r",
+      "Ġnam ing",
+      "We ight",
+      "Ġcondens ates",
+      "Ġn os",
+      "am ard",
+      "ver tices",
+      "EL S",
+      "id one",
+      "Ġcont est",
+      "Ġhead ing",
+      "ĠGal erkin",
+      "G V",
+      "ĠGl i",
+      "Ġfer mented",
+      "Ġb ilingual",
+      "Ġt icks",
+      "Ġk ary",
+      "rag al",
+      "ĠA ber",
+      "ĠYou Tube",
+      "UCT URE",
+      "b ranch",
+      "Ø ±",
+      "ĠF H",
+      "on oi",
+      "im otor",
+      "Ġver ifying",
+      "ĠCon ceptual",
+      "ĠDetermin ants",
+      "ur m",
+      "ur onic",
+      "ĠK au",
+      "ĠCon formal",
+      "Ġdrop ping",
+      "ĠFlow s",
+      "glu on",
+      "ag ain",
+      "ĠMR SA",
+      "war f",
+      "Ġemphas izes",
+      "Ent ry",
+      "ĠA SP",
+      "res ol",
+      "vent ricular",
+      "ĠâĨ Ķ",
+      "Ġoverex pressing",
+      "omegal ovirus",
+      "in oc",
+      "SC O",
+      "ĠPAR P",
+      "ĠSch ul",
+      "ĠCam b",
+      "ĠP od",
+      "ĠP un",
+      "ĠCompe tition",
+      "ĠG ATA",
+      "Ġmo on",
+      "Ġput s",
+      "angi ogenic",
+      "ĠRepublic an",
+      "ĠUb iqu",
+      "e ys",
+      "ĠG ong",
+      "arg er",
+      "ĠInter mediate",
+      "Ġinterp olated",
+      "Ġenlarg ement",
+      "Ġin struct",
+      "Ġr c",
+      "di oxo",
+      "ey e",
+      "ĠCar ls",
+      "ĠMeas ured",
+      "ir cles",
+      "ĠR af",
+      "Ġar b",
+      "ex amples",
+      "M i",
+      "ĠS tern",
+      "ĠF K",
+      "Ġmill isecond",
+      "ĠIR F",
+      "ĠEp ithelial",
+      "ed icine",
+      "el es",
+      "s ig",
+      "âĪ Ģ",
+      "ĠWi ener",
+      "b auer",
+      "ous es",
+      "Ġcol oured",
+      "ĠIncre ase",
+      "Ġtriglycer ides",
+      "Ġaeg ypti",
+      "ĠNumer ous",
+      "Ġretard ation",
+      "Ġinter cellular",
+      "ĠKle bsiella",
+      "ĠD ra",
+      "ĠD IC",
+      "ĠTh reshold",
+      "rain ment",
+      "Ġrepro ducing",
+      "Ġul cers",
+      "Ġa rousal",
+      "ĠH ills",
+      "Ġcal ves",
+      "ĠRes ervoir",
+      "ĠRad ar",
+      "Ġpsych osis",
+      "ĠFOR M",
+      "d uration",
+      "ĠAc ademic",
+      "c atal",
+      "oll a",
+      "ol ol",
+      "ĠC ron",
+      "ik o",
+      "Ġextrem es",
+      "ĠTryp an",
+      "Ġb ip",
+      "Ġal ginate",
+      "ĠH och",
+      "ĠBen nett",
+      "ĠH ippocamp",
+      "ĠGe ological",
+      "N evertheless",
+      "ĠH es",
+      "ĠAd ding",
+      "Ġextern ally",
+      "Ġsl ag",
+      "Ġte ach",
+      "ĠStan ley",
+      "controll er",
+      "ĠUn its",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠ",
+      "Ġaer odynamic",
+      "oval ent",
+      "c ube",
+      "Å Ł",
+      "re quire",
+      "romo lecules",
+      "ir teen",
+      "Ġcl auses",
+      "Ġdef eat",
+      "pol icy",
+      "Ġfaith ful",
+      "Ġp q",
+      "ĠTan aka",
+      "ĠE ver",
+      "Ġun predict",
+      "aut y",
+      "ĠGAL AX",
+      "Ġt ide",
+      "ĠFilter ing",
+      "Ġeut han",
+      "mer ce",
+      "DE X",
+      "Ġnest ing",
+      "D N",
+      "IR T",
+      "ĠTh r",
+      "tis sue",
+      "Ġpal ae",
+      "Ï ©",
+      "Ġdil ated",
+      "Ġpin ning",
+      "R b",
+      "ĠS ap",
+      "rag onal",
+      "ĠS PR",
+      "ĠD ial",
+      "Ġac upuncture",
+      "di ameter",
+      "ĠPC B",
+      "Par ameters",
+      "ĠProf iles",
+      "transf ected",
+      "l iter",
+      "ĠR ights",
+      "Ġcontrib utor",
+      "ĠCor rel",
+      "Ġregression s",
+      "Ġsegment al",
+      "Sh ape",
+      "I AN",
+      "ec om",
+      "com ings",
+      "Ġhemorrh agic",
+      "op os",
+      "Ġrefrac tion",
+      "P FC",
+      "pro j",
+      "ov o",
+      "ĠDer ived",
+      "Ġundi rected",
+      "Ġl os",
+      "Ġeng aging",
+      "c ans",
+      "Ġdestr uctive",
+      "P op",
+      "Ġm akers",
+      "ĠW or",
+      "ĠAre as",
+      "vas ion",
+      "Ġpara formaldehyde",
+      "abin oid",
+      "c py",
+      "pro xim",
+      "Ġen amel",
+      "Ġpa ediatric",
+      "ĠChild hood",
+      "Ġp ectin",
+      "ofil m",
+      "Ġcarboxyl ic",
+      "Ġa usten",
+      "Ġun equal",
+      "ĠCount ry",
+      "Ġiter ated",
+      "Ġflank ing",
+      "Ġt raction",
+      "ans on",
+      "isc us",
+      "ĠDav ies",
+      "ra ham",
+      "ter ozoic",
+      "ĠBr ass",
+      "O c",
+      "Ġun ification",
+      "met er",
+      "ĠNe on",
+      "bu ilding",
+      "ic ting",
+      "Ġjus tification",
+      "Pri or",
+      "Ġfir ms",
+      "Ġeduc ated",
+      "Ġinters ecting",
+      "Ġboost ing",
+      "P ass",
+      "m ember",
+      "con tains",
+      "ran o",
+      "rel ax",
+      "ĠCollabor ative",
+      "Ġp x",
+      "Ġseed ing",
+      "cri pts",
+      "ine z",
+      "ome res",
+      "Ġsib lings",
+      "ang ing",
+      "fer t",
+      "Ġrecover ing",
+      "p ure",
+      "Ġs d",
+      "ĠV ul",
+      "ped ance",
+      "Ġfight ing",
+      "S uper",
+      "ĠI to",
+      "Ġper imeter",
+      "ĠInhib itors",
+      "electro de",
+      "en abled",
+      "f b",
+      "ĠP Cs",
+      "Ġn ausea",
+      "ĠCon version",
+      "Ġsl a",
+      "Ġinver tebrates",
+      "ĠBri an",
+      "Ġcontig uous",
+      "ĠACKNOWLED GM",
+      "ur face",
+      "Ġco ars",
+      "ĠLe h",
+      "ĠComp ression",
+      "cy cles",
+      "Ġsin h",
+      "ĠOcc up",
+      "st rength",
+      "Ġcon str",
+      "Ġpestic ide",
+      "Ġb isp",
+      "ĠT n",
+      "Ġparent heses",
+      "deg rad",
+      "Ġhypergly cemia",
+      "P W",
+      "k j",
+      "ec ological",
+      "Ġth y",
+      "Ġele g",
+      "ĠSyn aptic",
+      "scal ed",
+      "ti ty",
+      "Ġequ ity",
+      "Ġblock chain",
+      "ĠLith ium",
+      "Ġsp ark",
+      "Ġen titled",
+      "Ġconven tions",
+      "Arg ument",
+      "Ġre tail",
+      "Ġne oplastic",
+      "Ġdamp ed",
+      "ĠSurve illance",
+      "ĠAn na",
+      "Ġspace times",
+      "ing es",
+      "ah ashi",
+      "ĠInf ections",
+      "Ġneglect ing",
+      "Ġevapor ated",
+      "vast atin",
+      "Ġg h",
+      "ĠN LP",
+      "Ġph ones",
+      "Ġlif ted",
+      "Ġdivis ible",
+      "Ġdur ability",
+      "os ited",
+      "Ġexcit ability",
+      "Ġbuoy ancy",
+      "Ġuncont rolled",
+      "b ran",
+      "ĠP he",
+      "Ġimmun ocomp",
+      "Ġevent ual",
+      "Ġclass room",
+      "Ġmicro graphs",
+      "Ġre charge",
+      "et tes",
+      "ĠD iver",
+      "ĠD all",
+      "Ġmet ac",
+      "Ġneuro endocrine",
+      "top ology",
+      "ĠHaw king",
+      "oms on",
+      "ĠHar ry",
+      "m outh",
+      "Ġdec iding",
+      "Ġunc overed",
+      "Ġgold en",
+      "ĠCast le",
+      "Ġfid ucial",
+      "A ware",
+      "ĠG an",
+      "era hertz",
+      "ĠSat urn",
+      "L N",
+      "Un it",
+      "ĥ Ĺ",
+      "Ġbind er",
+      "IN FO",
+      "ĠTem per",
+      "ip el",
+      "Ġnumer ator",
+      "Ġwebs ites",
+      "Ġthreat ened",
+      "Ġremn ants",
+      "ĠFinn ish",
+      "h of",
+      "med ia",
+      "concent ration",
+      "ĠRe ed",
+      "ĠLeishman ia",
+      "Ġmulti functional",
+      "rac y",
+      "Ġdistrib ute",
+      "ĠDec ay",
+      "Ġgr inding",
+      "L oss",
+      "MP L",
+      "ĠL akes",
+      "ĠQ R",
+      "ĠStruct ured",
+      "ĠMal aria",
+      "Ġflavon oid",
+      "Ġtow ns",
+      "op ia",
+      "ĠV ec",
+      "oth y",
+      "Ġsing les",
+      "Ġpenet rate",
+      "ĠP ig",
+      "ie ved",
+      "Ġderiv ations",
+      "Ġdiscom fort",
+      "af enib",
+      "ĠLegend re",
+      "ĠP ax",
+      "ĠM X",
+      "ĠExt rem",
+      "ĠFore ign",
+      "ĠCour se",
+      "ĠH it",
+      "v age",
+      "Ġcl ique",
+      "Ġcompens atory",
+      "U ser",
+      "Ġdraw s",
+      "ĠProt ective",
+      "Ġalloc ate",
+      "ĠP ant",
+      "Ġd ash",
+      "Ġpar al",
+      "ĠCirc ulating",
+      "ĠHist one",
+      "ĠÅ «",
+      "Ġproj ec",
+      "ĠA AA",
+      "ĠP MS",
+      "gl acial",
+      "ĠMe eting",
+      "ĠAntib iotic",
+      "ategor ical",
+      "Ġatten uate",
+      "P ower",
+      "ow icz",
+      "ĠDef ault",
+      "Ġmar sh",
+      "plas m",
+      "ĠPath ology",
+      "ĠE f",
+      "L ys",
+      "fl ies",
+      "Ġinterview ed",
+      "ĠQ A",
+      "Ġimp uls",
+      "Ġpap illary",
+      "d R",
+      "u h",
+      "ĠJ ing",
+      "Ġrescal ed",
+      "e fficiency",
+      "Ġe f",
+      "ĠE isen",
+      "Ġattack ed",
+      "Ġopt o",
+      "Ġspec ulated",
+      "ha z",
+      "Ġide ally",
+      "ymen optera",
+      "Ġl r",
+      "ĠI z",
+      "res ource",
+      "ĠFac ility",
+      "ĠAc quisition",
+      "Ġpost ural",
+      "auti ful",
+      "Ġging ival",
+      "Ġper taining",
+      "ĠExt ra",
+      "ĠProgram me",
+      "hes us",
+      "ferm ion",
+      "Ġstead ily",
+      "Ġtermin us",
+      "P arser",
+      "ĠIn clusion",
+      "ĠWu han",
+      "Ġrepe titions",
+      "d one",
+      "ĠC ep",
+      "Ġun structured",
+      "ĠCol lectively",
+      "Ġsett ling",
+      "Ġj aw",
+      "ĠUn i",
+      "Ġrest oring",
+      "urt les",
+      "F ull",
+      "Ġdynam o",
+      "IG O",
+      "ĠB AT",
+      "ov á",
+      "ven ues",
+      "ĠPer haps",
+      "sens ing",
+      "ĠI schem",
+      "odem ographic",
+      "S s",
+      "ĠL und",
+      "Ġel ite",
+      "prot ocol",
+      "ĠChrist opher",
+      "bas ic",
+      "Ġp uber",
+      "Ġmagne tism",
+      "v ars",
+      "in ducing",
+      "Ġd ated",
+      "Ġen emy",
+      "ĠSt op",
+      "s ocial",
+      "Ġd ÏĦ",
+      "ĠB un",
+      "Sm all",
+      "pur pose",
+      "Ġh unting",
+      "CP U",
+      "ĠJun ior",
+      "RE L",
+      "Ġcontrac tile",
+      "Ġsilic one",
+      "adren ergic",
+      "b z",
+      "Ġf us",
+      "if ted",
+      "se p",
+      "âĪĴ âĪŀ",
+      "Ġdr um",
+      "-------- --",
+      "ĠTreg s",
+      "it arian",
+      "cent ury",
+      "âĬ ¥",
+      "Num er",
+      "ĠB enz",
+      "Ġcommunic ating",
+      "Ġp aternal",
+      "ĠF GFR",
+      "Ġâ Ĥ¬",
+      "Ġdevi ate",
+      "f re",
+      "Ġmol ten",
+      "Ġstandard ization",
+      "Ġfunctional ities",
+      "ĠPaul o",
+      "Ġbuck et",
+      "ĠConcent rations",
+      "ĠK um",
+      "Ġmim icking",
+      "D rop",
+      "zo a",
+      "ĠNucle i",
+      "b rack",
+      "ec olor",
+      "Ġcar n",
+      "Ġveter inary",
+      "Ġchem otherapeutic",
+      "Ġfer ment",
+      "last ing",
+      "ĠRog ers",
+      "ier i",
+      "Ġconver ters",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠ",
+      "ĠRep air",
+      "Eu rope",
+      "T IME",
+      "Ġt ies",
+      "ĠP IN",
+      "Ġtrib ut",
+      "Ġhomogen ization",
+      "exc itation",
+      "at ization",
+      "ĠR ash",
+      "Ġpre cession",
+      "á s",
+      "Ġspik ing",
+      "ĠGrass mann",
+      "min ister",
+      "Ġfactor ial",
+      "ĠDe ut",
+      "sam pled",
+      "Ġeukary otes",
+      "overl apping",
+      "ag glut",
+      "Ġpres cribing",
+      "Ġc ro",
+      "om echanical",
+      "iz a",
+      "ĠMan ufact",
+      "n ative",
+      "urs ive",
+      "ĠIss ues",
+      "Ġstrept omycin",
+      "en di",
+      "ĠS pr",
+      "ce q",
+      "arg inine",
+      "ix on",
+      "ĠFound ations",
+      "Sing le",
+      "Ġox al",
+      "Ġhyd rate",
+      "Iter ator",
+      "k ii",
+      "amin ated",
+      "Ġspr ings",
+      "ol n",
+      "ĠSet up",
+      "Ġrip ening",
+      "Ġtheore tic",
+      "Ġcf g",
+      "μ L",
+      "G ordon",
+      "S K",
+      "Ġn ations",
+      "Qu ery",
+      "Ù ħ",
+      "Ġf ores",
+      "requ encies",
+      "ĠPh armaceutical",
+      "ĠAll ocation",
+      "otyp ical",
+      "ĠPil ot",
+      "th ora",
+      "ĠV and",
+      "Ġsyring e",
+      "ĠR AP",
+      "rom etric",
+      "Ġïģ ´",
+      "Ġcit ations",
+      "wo uld",
+      "Ġnorthe astern",
+      "compar ison",
+      "l ocus",
+      "et he",
+      "ĠK B",
+      "Ġhomolog s",
+      "Ġencephal itis",
+      "Ġz ig",
+      "Ġinc entive",
+      "Ġconf idential",
+      "Ġves tibular",
+      "ĠOT Us",
+      "Ġsynov ial",
+      "ĠRel ativity",
+      "Ġsub divided",
+      "che z",
+      "Ġlik ewise",
+      "ĠPD MS",
+      "ĠÅ ł",
+      "Ġsoci eties",
+      "ocyan ate",
+      "g ia",
+      "Ġlocal ize",
+      "Ġlact ation",
+      "Ġnod ule",
+      "ĠCO R",
+      "Ġharbor ing",
+      "ĠE QU",
+      "har vest",
+      "Ġband gap",
+      "r k",
+      "Ġres istor",
+      "Ġy e",
+      "ĠAs ymmetric",
+      "Ġpropag ators",
+      "Ġdiagnos ing",
+      "ĠAff airs",
+      "Ġeject a",
+      "Ġis omer",
+      "Ġi x",
+      "Ġfol iation",
+      "Ġcapac itors",
+      "Ġc ad",
+      "ĠNeut roph",
+      "pl iance",
+      "Ġcompress ible",
+      "ĠHun ter",
+      "ĠM Z",
+      "ĠWe ib",
+      "Ġnon coding",
+      "Ġmountain s",
+      "Ġadver tising",
+      "ale z",
+      "b right",
+      "lim sup",
+      "C i",
+      "ĠNe v",
+      "ĠStrain s",
+      "ost omy",
+      "op al",
+      "Ġconcaten ated",
+      "ĠPer f",
+      "CH O",
+      "Ġt urtles",
+      "ĠF ra",
+      "Ġall ogeneic",
+      "Ġuns uccessful",
+      "Y M",
+      "er ver",
+      "Ġc uc",
+      "Ġf ires",
+      "ch art",
+      "Ġinter rupted",
+      "Ġdec ides",
+      "Ġa uction",
+      "ĠUn til",
+      "ĠAT G",
+      "Ġdi am",
+      "magn itude",
+      "Ġd l",
+      "Ver tex",
+      "mon t",
+      "Ġfem tosecond",
+      "Par ams",
+      "Ġlys ate",
+      "is hers",
+      "ĠP AT",
+      "ĠK ev",
+      "ĠKn ock",
+      "Ġgro ove",
+      "L u",
+      "ĠJoh ann",
+      "Ġreplic a",
+      "ĠMATERIAL S",
+      "Ġl ots",
+      "Ġgener ically",
+      "ĠAl tered",
+      "ĠId entity",
+      "Ġunfold ed",
+      "C ES",
+      "ing ular",
+      "ĠF raction",
+      "ĠPro liferation",
+      "ĠVi enna",
+      "ac ia",
+      "pl ess",
+      "ĠSe vent",
+      "Ġturb ines",
+      "lys ine",
+      "Ġperox is",
+      "AR P",
+      "ĠEp is",
+      "ĠSY BR",
+      "Bu ilder",
+      "Ġspher ically",
+      "Ġdef end",
+      "Per formance",
+      "Ġmort ar",
+      "ĠCon cepts",
+      "work s",
+      "Ġreinfor ce",
+      "á ¹",
+      "Ġc us",
+      "ĠC IF",
+      "ĠAgric ultural",
+      "c rystalline",
+      "r ish",
+      "Ġref erenced",
+      "Ġact ress",
+      "Ġbounded ness",
+      "Si C",
+      "ĠÃ ¢",
+      "Ġj ack",
+      "Ġterm inate",
+      "ĠJ A",
+      "ĠKr ish",
+      "M MP",
+      "k x",
+      "ĠP SR",
+      "end l",
+      "W HO",
+      "ĠS ão",
+      "ĠC ultural",
+      "ĠE h",
+      "ul is",
+      "vi k",
+      "pr ises",
+      "ix el",
+      "ĠMet rics",
+      "Ġdiscontinu ities",
+      "ĠU ne",
+      "SC R",
+      "Ġproject ing",
+      "ĠOrig inal",
+      "ĠHum ans",
+      "transcription al",
+      "H K",
+      "ĠJ ain",
+      "atisf action",
+      "mes enchymal",
+      "Ġpyram id",
+      "Ġascorb ic",
+      "g ame",
+      "Ġno un",
+      "otox ins",
+      "p eptide",
+      "Ġglass y",
+      "Ġtalk ing",
+      "D em",
+      "ĠSch ro",
+      "ĠAssum ptions",
+      "Ġð x",
+      "Ġaneurys ms",
+      "M ASS",
+      "ĠH ou",
+      "ex posure",
+      "ĠL LC",
+      "Ġno ises",
+      "CT G",
+      "ĠElement ary",
+      "fl ip",
+      "Ġdys p",
+      "Ġmess enger",
+      "ĠImport ant",
+      "Ġim poses",
+      "Ġorgan elles",
+      "assert Equal",
+      "Ġjus tif",
+      "uc ine",
+      "Ġform ic",
+      "ormal ization",
+      "ĠRad ial",
+      "ĠCur ve",
+      "ĠCro hn",
+      "Ġbrow ser",
+      "Ġeff usion",
+      "Ġhand les",
+      "var sigma",
+      "Ġspecial ists",
+      "Ġpain ful",
+      "Ġerythemat osus",
+      "Ġf en",
+      "nitrop henyl",
+      "Ġleg acy",
+      "ĠQ Ds",
+      "rap per",
+      "Ġmon otherapy",
+      "ĠBel t",
+      "Z Z",
+      "Ġs intered",
+      "en edi",
+      "H b",
+      "t v",
+      "ĠN as",
+      "ov is",
+      "Ġmuc in",
+      "Ġacceler ates",
+      "Ġacqu iring",
+      "l uc",
+      "Ġdil aton",
+      "ĠPit ts",
+      "Ġequiv ariant",
+      "ĠL yman",
+      "ĠY a",
+      "Ġprog ressed",
+      "ĠAfter wards",
+      "ĠCH AR",
+      "D on",
+      "Ġhist ologic",
+      "Ġcircuit ry",
+      "p ene",
+      "op res",
+      "ĠSte fan",
+      "Ġsemic lassical",
+      "m und",
+      "ĠW aste",
+      "B Q",
+      "Ġadip onectin",
+      "Ġun seen",
+      "Ġbiom echanical",
+      "Ġhazard ous",
+      "r uctive",
+      "x yl",
+      "op f",
+      "Ġpr ion",
+      "ĠInf inite",
+      "Ġtrac ers",
+      "ĠHar rison",
+      "Ġfibrin ogen",
+      "Ġhydro lys",
+      "Ġis lets",
+      "Ġparallel ism",
+      "Sp ect",
+      "Ġimper ative",
+      "Ġc ured",
+      "ĠD SB",
+      "ide finite",
+      "ick er",
+      "Ġdiver gences",
+      "ĠShap iro",
+      "ab d",
+      "ĠL um",
+      "ĠV D",
+      "Ġf isheries",
+      "ge on",
+      "cop enia",
+      "ĠCl ay",
+      "Ġmaxim ized",
+      "ĠGre y",
+      "ĠB atch",
+      "Ġinf est",
+      "Ġam ple",
+      "Ġest ate",
+      "ĠSup reme",
+      "A O",
+      "is ia",
+      "ĠSm ad",
+      "Car lo",
+      "ĠSub st",
+      "Ġmon oidal",
+      "Ġnumer ic",
+      "Pl ot",
+      "Ġdyst rophy",
+      "hypert ensive",
+      "Ġst ool",
+      "als y",
+      "Ġche ese",
+      "n ih",
+      "Ġb ought",
+      "ĠS Q",
+      "Ġcl ues",
+      "Ġme iotic",
+      "Ġgo ats",
+      "ĠGTP ase",
+      "Ġrescal ing",
+      "N UM",
+      "ic ing",
+      "ĠÄ Ģ",
+      "Ġpret ty",
+      "lig and",
+      "En glish",
+      "ĠIntellig ent",
+      "E very",
+      "ĠPoli tical",
+      "ent on",
+      "Ġpass ages",
+      "ĠRemark s",
+      "s b",
+      "N etwork",
+      "ĠL RR",
+      "Ġcur l",
+      "urs ion",
+      "ĠA ver",
+      "ĠG LP",
+      "here n",
+      "at an",
+      "IC ENSE",
+      "Ġlate x",
+      "E MI",
+      "qu asi",
+      "ĠO m",
+      "Ġreview ing",
+      "Back ground",
+      "Ġs om",
+      "Ġsnap shots",
+      "b row",
+      "w ho",
+      "ĠT ail",
+      "ĠM SM",
+      "ĠG m",
+      "Ġph i",
+      "ren cy",
+      "separ ated",
+      "Ġg ig",
+      "os ides",
+      "Ġpe an",
+      "Ġappe aling",
+      "P U",
+      "n k",
+      "Ġqu er",
+      "ĠCh arg",
+      "ĠMo lecules",
+      "local ization",
+      "I dx",
+      "l ap",
+      "ĠT ax",
+      "ĠExp onential",
+      "ĠInhib itor",
+      "ĠBiom edical",
+      "ureth ane",
+      "le rene",
+      "rogen esis",
+      "ĠL ai",
+      "ĠAgg regation",
+      "ĠCa Cl",
+      "Ġsens ible",
+      "Ġcon junc",
+      "pa per",
+      "ĠCov id",
+      "ĠProced ures",
+      "Ġk new",
+      "Ġset ae",
+      "ĠAl le",
+      "ĠEx cept",
+      "Ġpres ynaptic",
+      "flow er",
+      "Ġultrason ography",
+      "Ġent ertain",
+      "i ors",
+      "ĠE ry",
+      "ĠIn teger",
+      "Ġrep ressor",
+      "Ġlater ally",
+      "Ġcomplement ed",
+      "T AG",
+      "ĠA round",
+      "ĠL ister",
+      "bit rary",
+      "back ward",
+      "Me V",
+      "Ġwh isk",
+      "AM s",
+      "ĠBul k",
+      "Ġqu iver",
+      "Ġdam aging",
+      "ĠQuantif ying",
+      "Ġsup rem",
+      "t el",
+      "Ġt ear",
+      "ot ers",
+      "vid in",
+      "Ġtub ules",
+      "Ġips ilateral",
+      "is ive",
+      "Ġsuit ably",
+      "ri el",
+      "Ġtub er",
+      "Ġfav ors",
+      "Ġcen tim",
+      "Ġtrans versal",
+      "ĠCH O",
+      "Ġtrim ester",
+      "C AC",
+      "c ognitive",
+      "ĠU TC",
+      "put e",
+      "Ġmid line",
+      "am ers",
+      "eval uation",
+      "D av",
+      "Ġb ags",
+      "tim er",
+      "Ġshort comings",
+      "ĠEr d",
+      "Ġdiscrim inator",
+      "A nt",
+      "s izes",
+      "Ġb ist",
+      "ing ual",
+      "ĠC ategory",
+      "Ġpuls ars",
+      "ĠSchw artz",
+      "ĠD rop",
+      "Sequ ence",
+      "Ġt ann",
+      "ĠSympt oms",
+      "D ict",
+      "ĠB lu",
+      "Sup plemental",
+      "Ġdis abled",
+      "ĠK oz",
+      "Ġinv oked",
+      "ĠC Q",
+      "ĠConn ectivity",
+      "Ġtelescop es",
+      "os o",
+      "Ġphyt ochemical",
+      "Ġorthogon ality",
+      "Ġinvis ible",
+      "ĠS CF",
+      "ĠA void",
+      "ĠH us",
+      "mic ron",
+      "atern ity",
+      "Pro ject",
+      "Ġadv ancing",
+      "ĠLorentz ian",
+      "S a",
+      "t Ãŀ",
+      "ĠU P",
+      "Ġar ts",
+      "Ġz er",
+      "ask et",
+      "Ġappe al",
+      "n ick",
+      "ĠCl oning",
+      "Ġsw ap",
+      "Ġphospholip ids",
+      "b g",
+      "ot hel",
+      "asc o",
+      "T rack",
+      "Ġsub manifold",
+      "Off set",
+      "ĠB ird",
+      "problem s",
+      "D Cs",
+      "Ġd ow",
+      "Ġde ionized",
+      "Ġsub class",
+      "Ġpubl ishing",
+      "ĠCar ter",
+      "Ġsyn ergy",
+      "Ġweak ened",
+      "ĠGl as",
+      "ĠP ie",
+      "hen ko",
+      "Ġsetup s",
+      "ĠBern stein",
+      "ĠÃ ¿",
+      "ĠSh u",
+      "ĠChang ing",
+      "os ov",
+      "ĠMet eor",
+      "in th",
+      "ra h",
+      "par amet",
+      "ren a",
+      "Ġnew borns",
+      "isc he",
+      "rot ating",
+      "Ġconf ident",
+      "f ac",
+      "ĠT err",
+      "Ġline width",
+      "IC P",
+      "thon y",
+      "Ġl anes",
+      "Ġsm oother",
+      "mon y",
+      "ĠCN Ns",
+      "P ort",
+      "Ġtrans iently",
+      "Ġsur geries",
+      "Ġsubm erged",
+      "Ġp uncture",
+      "Ġd ichlor",
+      "Ġsystematic s",
+      "Ġcontig s",
+      "Ġresid ing",
+      "B W",
+      "E O",
+      "G old",
+      "ion ate",
+      "voc ab",
+      "d W",
+      "ST AR",
+      "ĠP LC",
+      "ath i",
+      "ĠInf ectious",
+      "L ight",
+      "á »",
+      "ĠR al",
+      "Ġpropag ates",
+      "ĠLik elihood",
+      "h ill",
+      "cur l",
+      "check point",
+      "ra x",
+      "Ġv ancomycin",
+      "ĠU SD",
+      "ophe les",
+      "Ġfil tr",
+      "Ġstoichi ometry",
+      "âĶĢ âĶĢ",
+      "ĠN ad",
+      "access ible",
+      "Ġto y",
+      "Ġn ude",
+      "ĠS ut",
+      "ess ential",
+      "ĠO L",
+      "Ġper tin",
+      "Ġrec ur",
+      "Ġcap ill",
+      "Ġcomput able",
+      "Ġsuc tion",
+      "Ġsoft ening",
+      "ĠE SI",
+      "Ġmon itors",
+      "Ġpy ridine",
+      "ĠSens ors",
+      "ĠCombin atorial",
+      "at ta",
+      "ĠA MS",
+      "ĠD ul",
+      "ple teness",
+      "E th",
+      "ĠÃ »",
+      "Ġexc ised",
+      "ĠDiab etic",
+      "ĠI owa",
+      "Ġimmunost aining",
+      "Ġillness es",
+      "Ġenum er",
+      "ĠIran ian",
+      "Ġth umb",
+      "orphism s",
+      "Ġlegitim ate",
+      "l g",
+      "ĠS VD",
+      "Ġdes k",
+      "Form at",
+      "B on",
+      "Ġg arden",
+      "Ġinter personal",
+      "Ġel bow",
+      "ĠDem onstr",
+      "Ġnons pecific",
+      "F erm",
+      "ival ently",
+      "phthal ene",
+      "AR GET",
+      "Val id",
+      "Ġsun light",
+      "Ġresc ued",
+      "D AR",
+      "ĠIn variant",
+      "Ġid le",
+      "Ġalkal oids",
+      "scal es",
+      "s es",
+      "ob icity",
+      "be at",
+      "Ġcentrifug al",
+      "analy tical",
+      "p v",
+      "Ġt utorial",
+      "ĠN ation",
+      "gener ator",
+      "Ġcollision al",
+      "ĠC ME",
+      "Ġsc rap",
+      "ĠQ SO",
+      "Ġw ax",
+      "ĠSc enario",
+      "Ġminim izer",
+      "ĠMD PI",
+      "Ġprostagland in",
+      "ol ites",
+      "ocy steine",
+      "Ġcompac tification",
+      "Ġfrail ty",
+      "ops in",
+      "Ġjun ior",
+      "lo ud",
+      "Ġtit led",
+      "Ġeconom ically",
+      "th iophene",
+      "ĠInvestig ating",
+      "ĠE sp",
+      "Ġel usive",
+      "Ġmal ware",
+      "ĠTH P",
+      "imid azole",
+      "Ġre tains",
+      "ĠM IR",
+      "ff l",
+      "j ac",
+      "ĠP ART",
+      "ĠD CM",
+      "trans port",
+      "MAP K",
+      "Prob lem",
+      "S u",
+      "Ġdel im",
+      "Ġpsych ometric",
+      "vit ably",
+      "Ġhyper geometric",
+      "Ġuter us",
+      "Ġanaest hesia",
+      "ĠA venue",
+      "Ġmean ings",
+      "Ġrapid ity",
+      "Ġdend rites",
+      "g rain",
+      "ĠN ile",
+      "Ġfac ies",
+      "Ġpip elines",
+      "ĠCamp ylobacter",
+      "ĠMemb ers",
+      "benzo ate",
+      "Requ est",
+      "Ġp k",
+      "Ġref used",
+      "c aus",
+      "ĠS ay",
+      "l ane",
+      "ĠP SO",
+      "Ġgather ing",
+      "Ġrefrig er",
+      "R CC",
+      "Ġfib ronectin",
+      "hel p",
+      "ĠInt ensity",
+      "CL C",
+      "Q ue",
+      "el ly",
+      "Ġillumin ated",
+      "Ġpedest rian",
+      "ĠM ercury",
+      "Ġafford ed",
+      "Ġpathophys iological",
+      "ĠN GS",
+      "ass a",
+      "Ġend ors",
+      "Ġsens ation",
+      "Ġstream flow",
+      "av in",
+      "ĠGABA ergic",
+      "Ġreti rement",
+      "C ells",
+      "oc a",
+      "Ġoptim izations",
+      "Ġdig raph",
+      "ĠAu tism",
+      "oct urnal",
+      "osc ience",
+      "ĠEll is",
+      "ĠA j",
+      "ĠW SN",
+      "Ġshoot ing",
+      "i per",
+      "î Ħĥ",
+      "ĠWe ather",
+      "Ġrece ptive",
+      "Ġquar tic",
+      "ocycl ic",
+      "P ATH",
+      "size of",
+      "Ġmel ts",
+      "Ġdip oles",
+      "Ġbim odal",
+      "summ ary",
+      "Ġins omnia",
+      "opy ran",
+      "Ġwrap ped",
+      "ĠJos é",
+      "A H",
+      "c ia",
+      "Ġob eys",
+      "ĠK ay",
+      "inter vention",
+      "Ġrout er",
+      "ĠDrug s",
+      "ow ska",
+      "ĠAr r",
+      "ĠCap tain",
+      "ĠT MS",
+      "ad v",
+      "Ġbo at",
+      "Ġtrust ed",
+      "se ver",
+      "ill ars",
+      "ĠMiss ouri",
+      "Ġequival ents",
+      "ĠHar vard",
+      "ĠClark e",
+      "reson ant",
+      "rad y",
+      "trig gered",
+      "Ġc left",
+      "Ġun ic",
+      "Ġbrain stem",
+      "Ġthrom bin",
+      "ĠF light",
+      "Ġsection al",
+      "Ġconcaten ation",
+      "Ġcantile ver",
+      "et on",
+      "Ġdec ode",
+      "of acial",
+      "Ac tion",
+      "ĠIll ustration",
+      "ver tical",
+      "ch all",
+      "ĠReg istry",
+      "M AT",
+      "Ġcons on",
+      "Ġneo adjuvant",
+      "ĠW istar",
+      "ĠIm per",
+      "Ġal titudes",
+      "Ġsub population",
+      "ĠSc ene",
+      "tensor flow",
+      "s low",
+      "Ġh int",
+      "Ġbeam forming",
+      "e in",
+      "Ġimp regn",
+      "ĠRF ID",
+      "ĠAnaly zing",
+      "ĠP ent",
+      "ĠD NS",
+      "ĠG ilbert",
+      "Ġcr ater",
+      "Compar ing",
+      "Ġb f",
+      "Ġfl ights",
+      "Ġmal nutrition",
+      "SM C",
+      "Ġeryth rop",
+      "ĠTum ors",
+      "T x",
+      "Ġis ospin",
+      "ĠK ub",
+      "ik ing",
+      "Ġcorticoster oids",
+      "urs or",
+      "ĠBur g",
+      "in spired",
+      "ĠI gn",
+      "Ġmy cel",
+      "pred iction",
+      "method s",
+      "C opy",
+      "ĠR W",
+      "ĠK night",
+      "Ġdem ethyl",
+      "ì Ħ",
+      "Ġc ili",
+      "Ġb es",
+      "ĠE ck",
+      "Ġdilat ation",
+      "Ġan imation",
+      "ab stract",
+      "Ġcircum vent",
+      "Ġinoc ulum",
+      "S eg",
+      "ĠC aps",
+      "ere rs",
+      "PL S",
+      "ĠPe er",
+      "Ġver ifies",
+      "ateg y",
+      "ogene tics",
+      "Ġoligonucle otides",
+      "rac tical",
+      "Ġdiver ges",
+      "ĠStan ford",
+      "ĠA i",
+      "Ġweigh ing",
+      "T g",
+      "re infor",
+      "ĠA lam",
+      "qu iry",
+      "ĠN ob",
+      "Ġlinear ization",
+      "ĠV enez",
+      "ne xin",
+      "level s",
+      "L ip",
+      "ĠPat el",
+      "ĠMagn itude",
+      "e titive",
+      "ĠE agle",
+      "Ġsp utum",
+      "ĠCO S",
+      "Ġincub ator",
+      "U l",
+      "ĠRec eptors",
+      "ĠSch ott",
+      "GC G",
+      "ĠZe iss",
+      "ĠEnt anglement",
+      "ĠVacc ine",
+      "or ted",
+      "Ġn b",
+      "ĠS j",
+      "ĠM rs",
+      "Ġcal f",
+      "Ġinte grability",
+      "ĠPh oton",
+      "Ġgon dii",
+      "ĠM IL",
+      "Ġal iph",
+      "ĠD ip",
+      "fall s",
+      "c trl",
+      "k u",
+      "et ent",
+      "pl t",
+      "Ġpers isted",
+      "ĠMan ager",
+      "Ġprerequ isite",
+      "f illing",
+      "ĠM EA",
+      "S ym",
+      "ĠG rain",
+      "Ġduct al",
+      "ĠT ODO",
+      "Ġaff inities",
+      "Ġdegener ative",
+      "ĠF itz",
+      "ov ar",
+      "ĠTri ple",
+      "Ġdend rim",
+      "ĠFrank lin",
+      "m ag",
+      "ot ely",
+      "Ġstabil izes",
+      "Ġc ash",
+      "ĠS quad",
+      "Ġch ampion",
+      "PD B",
+      "Ġur g",
+      "Ġalcoh olic",
+      "Ġt ar",
+      "yl ed",
+      "V ersion",
+      "Ġs ale",
+      "ĠM LP",
+      "out er",
+      "Ġsimpl ifying",
+      "ĠExt ract",
+      "Par am",
+      "ĠRest ric",
+      "Ġtract able",
+      "ĠArch ive",
+      "Resp onse",
+      "AD DR",
+      "Ġcommut ation",
+      "R ich",
+      "ĠAndrew s",
+      "Ġosteocl ast",
+      "rom ic",
+      "ĠSh ift",
+      "Ġacceler ometer",
+      "ĠS ent",
+      "Ġch ances",
+      "ost ing",
+      "Ġmeth acrylate",
+      "Ġglu ons",
+      "Ġôı ½",
+      "Ġpolyg ons",
+      "ĠRCT s",
+      "Ġinf ancy",
+      "Ġproceed ed",
+      "ĠHor izontal",
+      "C OR",
+      "Ġc aching",
+      "ĠN HS",
+      "ph obic",
+      "ĠX MM",
+      "Ġmicrobi ological",
+      "G MP",
+      "Ù Ĩ",
+      "ĠT SS",
+      "ĠS ul",
+      "ĠF act",
+      "ĠW E",
+      "Ġcertain ty",
+      "ens itivity",
+      "Ġdecon volution",
+      "ĠG ain",
+      "Ġbl ots",
+      "Ġsee ks",
+      "Ġcos h",
+      "ennes see",
+      "Ġsl ave",
+      "ĠT ran",
+      "Ġtrans pose",
+      "re ated",
+      "Ġsh ading",
+      "ĠB U",
+      "ĠO V",
+      "ĠLo ok",
+      "Ġcomprehens ively",
+      "ĠFred er",
+      "Hand ler",
+      "f ibr",
+      "Ġmiss ense",
+      "target s",
+      "prom oting",
+      "ĠP ep",
+      "var pi",
+      "ĠHar monic",
+      "ĠA IS",
+      "Ġmon ocyt",
+      "Ġthin ning",
+      "Ġpherom one",
+      "W ater",
+      "an ase",
+      "ĠS ang",
+      "Ġsub structure",
+      "w p",
+      "ĠK ansas",
+      "DE BUG",
+      "ĠPro be",
+      "Ġpattern ed",
+      "cle an",
+      "Ġbro iler",
+      "odext rin",
+      "a ided",
+      "op rol",
+      "ubl in",
+      "in um",
+      "Ġan atomic",
+      "Ġpl ating",
+      "ar ro",
+      "uc al",
+      "Ġspeed up",
+      "Ġhaem orrh",
+      "eptid ase",
+      "Ġs agittal",
+      "Ġin tim",
+      "ĠF ISH",
+      "Ġsc arc",
+      "AT CC",
+      "inc or",
+      "Ġser ological",
+      "ent e",
+      "Ġsh ale",
+      "Ġover fitting",
+      "ĠEx cess",
+      "ĠAL P",
+      "P ool",
+      "d ry",
+      "y u",
+      "ĠPM MA",
+      "ĠHyp oxia",
+      "n othing",
+      "chest ra",
+      "colone qq",
+      "Ġb ibli",
+      "ĠEX PECT",
+      "B AL",
+      "et han",
+      "ĠâĪ ĺ",
+      "Ġj ourney",
+      "Ġbi ocompatibility",
+      "P AN",
+      "Ġb on",
+      "ĠR oh",
+      "Ġpolar isation",
+      "Sp in",
+      "id ences",
+      "ĠB CR",
+      "ĠH IP",
+      "ĠTh ick",
+      "Ġrecogn izes",
+      "Ġs ar",
+      "Ġam end",
+      "ques tions",
+      "Ġcareg iver",
+      "ĠMari e",
+      "Ġmetalloprotein ase",
+      "Ġal dehydes",
+      "Ġinter neurons",
+      "Ġtetra hedral",
+      "gue z",
+      "Ġquasipar ticle",
+      "Ġo t",
+      "decre asing",
+      "st re",
+      "Ġphot operiod",
+      "Ġprior iti",
+      "Ġap o",
+      "Ġimmunosup pression",
+      "ĠPier re",
+      "L PS",
+      "Ġcl umps",
+      "ĠPl ane",
+      "Ġturb idity",
+      "Ġpollut ant",
+      "Ġbi och",
+      "ĠT RE",
+      "Ġdesign ers",
+      "Ġrend ers",
+      "Ġrepl aces",
+      "ĠP LS",
+      "Ġhum oral",
+      "B as",
+      "re ira",
+      "ĠA edes",
+      "vit amin",
+      "cur ves",
+      "ocic eptive",
+      "Ġin disp",
+      "Ġox y",
+      "Ġed ible",
+      "ĠMes enchymal",
+      "ĠDeg ree",
+      "Å ¾",
+      "ĠO ak",
+      "ĠBhat t",
+      "on so",
+      "ĠS BP",
+      "ĠA ux",
+      "Ġmar tingale",
+      "ĠMicrobi ota",
+      "g low",
+      "Ġex ud",
+      "ap olis",
+      "Ġsome how",
+      "Ġcent red",
+      "Ch annel",
+      "ĠNormal ized",
+      "il itation",
+      "Ġtranscript ase",
+      "Ġcry o",
+      "predic ted",
+      "ĠD AG",
+      "Ġr f",
+      "end or",
+      "INT ER",
+      "ĠMes h",
+      "ĠFund ament",
+      "y cle",
+      "Ġprim itives",
+      "radi ated",
+      "Ġr ho",
+      "enes ulf",
+      "ĠF SH",
+      "ĠE cos",
+      "local ized",
+      "Ġenter prise",
+      "cephal us",
+      "Ġcarc ass",
+      "A Y",
+      "ec urity",
+      "ĠT MD",
+      "Ġl b",
+      "ĠA eros",
+      "ĠM ER",
+      "At tr",
+      "ĠA CL",
+      "ĠBar b",
+      "c out",
+      "Ġde oxy",
+      "ati os",
+      "Ġpers ists",
+      "Ġviol ent",
+      "Ab elian",
+      "Ġell ips",
+      "ion g",
+      "Ġsuccess or",
+      "ĠGonz ález",
+      "l iving",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠ",
+      "iment in",
+      "Ġcaps ules",
+      "V IS",
+      "ĠP OP",
+      "arithm ic",
+      "O O",
+      "w l",
+      "ino ic",
+      "ĠCent ers",
+      "robl asts",
+      "th ose",
+      "ĠM J",
+      "Ġfron ts",
+      "Ġun int",
+      "Ġfac ile",
+      "co herent",
+      "av our",
+      "cep tive",
+      "ta h",
+      "Ġrelated ness",
+      "d E",
+      "un gen",
+      "## ###",
+      "Ġam phi",
+      "ĠGu y",
+      "st ars",
+      "ect om",
+      "Ġlay ing",
+      "Ġsp ider",
+      "AC s",
+      "Ġseed ling",
+      "Ġdu plicated",
+      "ic he",
+      "ĠM ST",
+      "gr ass",
+      "Ġprophyl actic",
+      "e ks",
+      "Ġl aryngeal",
+      "ĠS per",
+      "ĠW als",
+      "Ġcho lec",
+      "ĠPlan et",
+      "ĠHEP ES",
+      "Ġdi ploid",
+      "const raint",
+      "Py x",
+      "AC h",
+      "ĠCu i",
+      "ĠSh ared",
+      "ĠC and",
+      "ĠG ö",
+      "Ġdet ached",
+      "Ġpass engers",
+      "Ġaliph atic",
+      "Ġp our",
+      "Ġaccess es",
+      "ĠWal d",
+      "Ġdecor ated",
+      "Ġcaroten oids",
+      "ues tions",
+      "ĠImp acts",
+      "S AT",
+      "ar u",
+      "ĠP ir",
+      "ĠCon figuration",
+      "ĠCong o",
+      "ĠL ing",
+      "Ġdes ic",
+      "Ġmac rom",
+      "Ġlack ed",
+      "Ġencompass es",
+      "Ġp umped",
+      "ĠFor ty",
+      "rex ate",
+      "ifferenti ated",
+      "Ġn oble",
+      "Ġrad ion",
+      "Ġimmig rants",
+      "Ġbiodegrad able",
+      "Ġmig rating",
+      "arg v",
+      "CO M",
+      "ĠObserv ational",
+      "Ġcann abis",
+      "y ama",
+      "Ġconcent ric",
+      "Con n",
+      "tal ion",
+      "Ġrespond ers",
+      "uten ant",
+      "ĠT rim",
+      "Ġcontrib utors",
+      "Ġcontrac ted",
+      "ĠXen opus",
+      "Ġlo ud",
+      "ĠEnh ancing",
+      "Ġinfarc t",
+      "Ġo k",
+      "Ġas ks",
+      "rel in",
+      "Ġillustr ative",
+      "vd ash",
+      "d g",
+      "Ġf oc",
+      "Ġl ivers",
+      "ĠO tt",
+      "ĠT SP",
+      "log ger",
+      "depend ing",
+      "Ġdis proportion",
+      "Ġint ric",
+      "Ġimmun ized",
+      "vare z",
+      "Ġsal ic",
+      "ĠInstit utes",
+      "KE Y",
+      "Ġend oscopy",
+      "er k",
+      "el iness",
+      "ĠS ag",
+      "ath yroid",
+      "Ġacid ity",
+      "aro v",
+      "ĠVor onoi",
+      "Experim ental",
+      "Ġg ently",
+      "Me asure",
+      "ïĺ º",
+      "Ġw onder",
+      "ĠP ancreatic",
+      "ĠH ispanic",
+      "ĠE ug",
+      "re ducing",
+      "tain ment",
+      "Ġsur prise",
+      "Ġ æ",
+      "cr iter",
+      "ĠHypert ension",
+      "ti que",
+      "ĠC ris",
+      "comp atible",
+      "ens on",
+      "Ġdistribution al",
+      "ĠN AT",
+      "wid ths",
+      "Ġisother ms",
+      "ĠP rad",
+      "Ġbi odies",
+      "Ġorb ifold",
+      "ĠE OS",
+      "Ġat ax",
+      "ĠB od",
+      "ĠN MD",
+      "Ġmon oxide",
+      "ĠUk raine",
+      "f oli",
+      "ĠD ro",
+      "Ġun available",
+      "Ġbr ighter",
+      "âĬ Ĺ",
+      "ometh ane",
+      "Ġd ream",
+      "Ġsp o",
+      "ĠMa ur",
+      "Ġoccas ional",
+      "Ġincons istency",
+      "ĠT ac",
+      "op ts",
+      "ĠG AB",
+      "ĠTa o",
+      "ĠMatthe w",
+      "Ã ½",
+      "Ġp iano",
+      "ĠR CC",
+      "ĠO K",
+      "ĠK ul",
+      "met han",
+      "ĠPRO C",
+      "Ġconvers ations",
+      "ĠC SI",
+      "ang ent",
+      "ĠX ue",
+      "Ġgraph ic",
+      "den ing",
+      "health y",
+      "Ġf p",
+      "az one",
+      "Ġdiscipl ine",
+      "Ġprogress es",
+      "Ġb amboo",
+      "Ġchar m",
+      "ĠAc tivated",
+      "ĠSh arp",
+      "yn es",
+      "Ġtool box",
+      "Ġhetero structures",
+      "piper azin",
+      "Ġa rose",
+      "ĠInter val",
+      "Ġstrip e",
+      "ĠCh ak",
+      "Ġc uff",
+      "RE SS",
+      "Ġnon uniform",
+      "Ġbeet le",
+      "P rec",
+      "z c",
+      "Th read",
+      "b et",
+      "Ġe e",
+      "ĠOption al",
+      "Ġt roph",
+      "ĠP uer",
+      "ĠF ron",
+      "Ġmultiple t",
+      "Ġcalor imetry",
+      "Ġmonocyt ogenes",
+      "ĠH imal",
+      "Ġdr ill",
+      "AG A",
+      "Ġferr itin",
+      "Ġd pi",
+      "ĠC arm",
+      "Ġg one",
+      "Ġun idirectional",
+      "Ġrem inis",
+      "Ġadjust able",
+      "ĠAust in",
+      "S ARS",
+      "d al",
+      "Ġc ef",
+      "equiv ariant",
+      "bas eline",
+      "Ġspin ors",
+      "ĠPr int",
+      "Ġm ile",
+      "ĠL inc",
+      "mut ation",
+      "Ġmuc us",
+      "ĠH SC",
+      "Ġtherm od",
+      "Ġpain t",
+      "Ġdistinct ly",
+      "ath y",
+      "Ġph armacy",
+      "ĠBul g",
+      "ĠG ang",
+      "hic le",
+      "og an",
+      "ĠJ ian",
+      "ĠIndian a",
+      "Ġinstant on",
+      "Ġpall adium",
+      "f iber",
+      "n py",
+      "ĠU A",
+      "ĠQ T",
+      "cepti ble",
+      "et ine",
+      "ĠH oles",
+      "Ġdepend ences",
+      "Ġthreshold ing",
+      "ĠMain tenance",
+      "Ġparticip ates",
+      "ĠGen omes",
+      "factor ial",
+      "ĠL iber",
+      "ĠTherm odynamic",
+      "Ġe lective",
+      "uc her",
+      "Ġhyper ther",
+      "Ġstom atal",
+      "ĠB irth",
+      "ch olesterol",
+      "Ġnot ch",
+      "Ġsym biotic",
+      "Ġbusiness es",
+      "Ġapprec iable",
+      "Ġspecial ization",
+      "á r",
+      "act yl",
+      "ĠGraph Pad",
+      "os per",
+      "Ġor chestr",
+      "Ġdi hydro",
+      "Ġconcl uding",
+      "CL K",
+      "Ġeq s",
+      "ĠProg ression",
+      "Ġclub s",
+      "ak u",
+      "ev ents",
+      "Ġspl enic",
+      "Ġb unch",
+      "ĠT m",
+      "ĠM obility",
+      "Ġtwo fold",
+      "Ġradi ally",
+      "L STM",
+      "M H",
+      "ĠCo al",
+      "Ġfron tier",
+      "J an",
+      "J un",
+      "ĠSim pson",
+      "Ġabst racts",
+      "P al",
+      "Ġun im",
+      "Ġro bo",
+      "ĠII B",
+      "dep leted",
+      "Ġmorphological ly",
+      "Ġenfor cement",
+      "Ġd well",
+      "Ġst agn",
+      "Ġlim estone",
+      "Ġmicro v",
+      "Ġïĥ ¸",
+      "L uc",
+      "p acs",
+      "cy ano",
+      "Ġintra ocular",
+      "ĠCalc ulate",
+      "Sup port",
+      "SY S",
+      "ĠV S",
+      "CM s",
+      "Const ant",
+      "ĠD j",
+      "Ġun balanced",
+      "Ġrepeat ability",
+      "g ins",
+      "i rect",
+      "ĠM OR",
+      "ĠBa iley",
+      "Ġadvance ment",
+      "Ġpurs uit",
+      "Ġa rom",
+      "pro ced",
+      "ĠIniti ative",
+      "Ġincenti ves",
+      "Ġsur pass",
+      "gen es",
+      "ĠIN D",
+      "L H",
+      "Ġsu icidal",
+      "Ġbiodies el",
+      "x z",
+      "Ù Ĭ",
+      "le a",
+      "ĠAn thony",
+      "Lear ning",
+      "Ġund o",
+      "Ġïĥ º",
+      "ĠCommun ities",
+      "h ua",
+      "iti me",
+      "ĠDe an",
+      "Ġplas min",
+      "ÃŃ nez",
+      "ohyd rate",
+      "Ġneurode velop",
+      "Ġstoichi ometric",
+      "ĠOnc ology",
+      "Ġshow er",
+      "ĠD MS",
+      "W OR",
+      "ĠP IP",
+      "Ġster ic",
+      "mitte es",
+      "ist ol",
+      "ox ins",
+      "no on",
+      "FF T",
+      "Ġá »",
+      "opo iesis",
+      "Ġresemb ling",
+      "ĠB ord",
+      "Ġprob iotics",
+      "ocy sts",
+      "gre y",
+      "ĠCatal og",
+      "IZ ATION",
+      "ill es",
+      "ĠAl an",
+      "ĠÅ ·",
+      "ĠLe ib",
+      "ĠReason ing",
+      "bi ological",
+      "uter ine",
+      "vac izumab",
+      "lecom mun",
+      "ĠW arm",
+      "ep age",
+      "vari ants",
+      "B SA",
+      "Ġïĥ ¶",
+      "Ġhepat ocyte",
+      "ket ch",
+      "Ġstrip ping",
+      "ĠAd verse",
+      "ĠFe as",
+      "Ġïĥ ¯",
+      "P ac",
+      "Ġind entation",
+      "Ġsec ular",
+      "Ġidentif iable",
+      "run ning",
+      "Ġr d",
+      "Ġz yg",
+      "ĠD ictionary",
+      "Ġres veratrol",
+      "ines terase",
+      "Ġtet racycline",
+      "ub les",
+      "Ġthro at",
+      "ĠL amb",
+      "ary on",
+      "ĠS QL",
+      "ĠÃ ľ",
+      "Ġgly cemic",
+      "Ġcompet ent",
+      "ĠAg reement",
+      "oic ed",
+      "Ġconstitu tively",
+      "Ġelectro cardi",
+      "oplas ma",
+      "Ġî Ħĥ",
+      "an ide",
+      "Ġre organization",
+      "Ġun infected",
+      "UT E",
+      "Ġro yal",
+      "ĠS it",
+      "Ġmar ital",
+      "ĠKob ayashi",
+      "B arr",
+      "ĠT ennessee",
+      "ĠChrom at",
+      "ĠD erm",
+      "pro jection",
+      "ĠJ ob",
+      "Ġâī ł",
+      "ĠT rip",
+      "Ġis op",
+      "Ġproject or",
+      "Ġatmosp heres",
+      "Ġperfor ation",
+      "st orage",
+      "ith s",
+      "Ġmon omeric",
+      "ĠUS B",
+      "ĠE ve",
+      "Ġsp ore",
+      "Ġm T",
+      "ox azole",
+      "ĠDe formation",
+      "Ġtext ual",
+      "Ġwar f",
+      "Ġneuropath ic",
+      "prep ared",
+      "Ġbl ended",
+      "ĠHo uston",
+      "**************************************************************** ********",
+      "es ters",
+      "Equ als",
+      "Ġallerg en",
+      "Ġpertin ent",
+      "f acts",
+      "uc tions",
+      "Ġcl ocks",
+      "ĠV ia",
+      "ĠCD F",
+      "Ġest uary",
+      "Ġphenomen ology",
+      "ar us",
+      "AP H",
+      "Ġarg ues",
+      "Ġinser ts",
+      "g ow",
+      "h art",
+      "Ġchem otaxis",
+      "Ġp v",
+      "Ġre in",
+      "ĠG rim",
+      "ĠV F",
+      "Ġeff ic",
+      "ĠProf iling",
+      "Ġan odic",
+      "ĠDEN V",
+      "ĠW it",
+      "ĠSY STEM",
+      "ĠCay ley",
+      "En g",
+      "ĠA QP",
+      "inter actions",
+      "ili arity",
+      "ĠProm otes",
+      "Ġd ams",
+      "ing ton",
+      "ff ff",
+      "Ġint ran",
+      "ĠTurb ulence",
+      "ĠBian chi",
+      "C RE",
+      "ĠN OD",
+      "ap ine",
+      "ĠK ane",
+      "ĠPD GF",
+      "ĠAx is",
+      "ĠC ausal",
+      "ĠPo or",
+      "ĠW ords",
+      "ĠHR V",
+      "Ġcyan obacteria",
+      "Ġreminis cent",
+      "ĠRemark ably",
+      "he et",
+      "@ @",
+      "b il",
+      "Ġdiscrim inating",
+      "ĠBal tic",
+      "ĠQue bec",
+      "Ġdef ensive",
+      "âĪ ©",
+      "k r",
+      "ĠR PE",
+      "see king",
+      "ĠMo vie",
+      "Ġinnov ations",
+      "le pt",
+      "Ġk w",
+      "Ġtib ia",
+      "Ġne at",
+      "yt est",
+      "Ġthin ner",
+      "Ġoste oblasts",
+      "ĠNorth west",
+      "M OS",
+      "ĠP Q",
+      "Ġsp i",
+      "Ġrespond s",
+      "Ġhistor ically",
+      "ĠPack age",
+      "ĠCoast al",
+      "ĠMississ ippi",
+      "ĠP VA",
+      "per ing",
+      "ind ole",
+      "Ġprosp ectively",
+      "ĠHem isphere",
+      "Ġbare ly",
+      "án chez",
+      "ag gered",
+      "yp tian",
+      "ĠG est",
+      "yl ine",
+      "Ġphot ochemical",
+      "os calar",
+      "por ated",
+      "Ġmetabol omics",
+      "Ġoste oblast",
+      "EGF P",
+      "eri atric",
+      "D W",
+      "qu est",
+      "ĠH ave",
+      "Ġsp ondyl",
+      "ĠPrim er",
+      "Ġs inks",
+      "Ġg aussian",
+      "ĠK hal",
+      "En c",
+      "ĠAn opheles",
+      "Th anks",
+      "Ġconstr ued",
+      "ĠU SS",
+      "ĠZe eman",
+      "Ġex ported",
+      "ĠLe vi",
+      "Ġcomm ander",
+      "conn ect",
+      "Ġnom enclature",
+      "there fore",
+      "ul ata",
+      "Ġent repreneur",
+      "Ġneuros cience",
+      "z an",
+      "Ġext ant",
+      "AT IVE",
+      "ope z",
+      "Ġenfor ced",
+      "ĠInnov ation",
+      "ear ance",
+      "Ġimp ressive",
+      "ĠPl ac",
+      "ĠMo z",
+      "ĠSt ark",
+      "Ġri val",
+      "ĠCap ital",
+      "Ġgranular ity",
+      "Ġdiaphrag m",
+      "ut aneous",
+      "ind s",
+      "Ġphot ograph",
+      "Ġrect angles",
+      "T GF",
+      "Ġse af",
+      "Ġm aze",
+      "ĠH W",
+      "Ġcorrel ators",
+      "Ġdistinguish able",
+      "Ġconfound ers",
+      "Ġlandsl ide",
+      "Ġto ll",
+      "Ġwas tes",
+      "ĠW F",
+      "Ġend oc",
+      "Ġcaps id",
+      "ec und",
+      "ĠR BD",
+      "ps in",
+      "Ġobst etric",
+      "Ġnanos heets",
+      "oc ols",
+      "ren s",
+      "ĠSub stituting",
+      "Ġcustom ized",
+      "Ġres uscitation",
+      "Ġtub ulin",
+      "ophy te",
+      "~~~~ ~~~~",
+      "pl ants",
+      "hic illin",
+      "hal o",
+      "ruit ment",
+      "ĠConc rete",
+      "Ġnanor ods",
+      "ĠForm s",
+      "Ġd ying",
+      "dis charge",
+      "Ġwell being",
+      "Ġwar mer",
+      "ĠS SD",
+      "ĠA UT",
+      "ĠCon jug",
+      "Ġjuven iles",
+      "Ġine vitably",
+      "ĠM CS",
+      "appro ach",
+      "ĠM ason",
+      "ĠG ust",
+      "ĠTherm odynamics",
+      "Ġpe el",
+      "ĠTranscript ome",
+      "Ġindisp ensable",
+      "ur gery",
+      "pos ity",
+      "Ġpolar izations",
+      "ĠOther s",
+      "Ġsand y",
+      "Ġgli omas",
+      "Ġpurs ued",
+      "V EL",
+      "Ġr st",
+      "pos ium",
+      "ne arest",
+      "Ġdissem inated",
+      "ĠMY C",
+      "Ġal dehyde",
+      "ĠDiagn ostics",
+      "m ans",
+      "Ġas phal",
+      "ĠSe lect",
+      "ĠRec on",
+      "and ro",
+      "D IM",
+      "Ġf eces",
+      "ill on",
+      "ĠMAL DI",
+      "n f",
+      "ĠE lim",
+      "Ġhapp y",
+      "ĠKar l",
+      "ĠIn ser",
+      "Ġinter rog",
+      "In tern",
+      "Ġtensor flow",
+      "Ġhalo es",
+      "Ġanticip ate",
+      "ĠDPP H",
+      "rÃŃ guez",
+      "H er",
+      "an ate",
+      "Ġd ressing",
+      "ĠH oly",
+      "Ġnew er",
+      "rid es",
+      "plac ed",
+      "inet obacter",
+      "ĠOcc urrence",
+      "ed ema",
+      "ĠI k",
+      "ab ad",
+      "ĠTrans itions",
+      "Ġoutl ines",
+      "Ġcoch lear",
+      "G y",
+      "s uccess",
+      "ĠM EM",
+      "ast ype",
+      "Ġnormal izing",
+      "Ġtermin ates",
+      "Ġsudden ly",
+      "b box",
+      "ĠP ul",
+      "ĠP TP",
+      "ag inal",
+      "Ġpre trained",
+      "Ġun reliable",
+      "ĠGraph ical",
+      "ĠSey fert",
+      "Ġcharacter izations",
+      "Ġt x",
+      "Ġbic arbonate",
+      "math ord",
+      "Ġher itability",
+      "stack exchange",
+      "i ri",
+      "âĢ ĸ",
+      "ip it",
+      "at tle",
+      "Ġare na",
+      "ib a",
+      "ĠA X",
+      "ĠG Ps",
+      "ophil ia",
+      "S EL",
+      "os ystem",
+      "ĠâĬ ¢",
+      "ĠNucle us",
+      "red ited",
+      "AC R",
+      "ĠAnt enna",
+      "ĠCd c",
+      "or ie",
+      "Ġequil ibration",
+      "el ong",
+      "st ability",
+      "ĠSch ist",
+      "Ġinject ing",
+      "h p",
+      "Ġvit amins",
+      "Po isson",
+      "or tal",
+      "ĠÃ Ĭ",
+      "ĠÄ ı",
+      "I ll",
+      "Ġutil s",
+      "оР²",
+      "ĠG rom",
+      ":: ::",
+      "ĠGn RH",
+      "ĠSier ra",
+      "Ġd rafted",
+      "Ġcap ita",
+      "sh ips",
+      "Ġtim estamp",
+      "Ġsubstit uents",
+      "ĠNot able",
+      "ĠPur pose",
+      "in ol",
+      "Ġa i",
+      "Ġf og",
+      "ot one",
+      "ĠPl aces",
+      "bys hev",
+      "ti ology",
+      "ri ption",
+      "Ġy ards",
+      "ĠX I",
+      "Ġtechn ically",
+      "G AM",
+      "ĠA BS",
+      "pl atform",
+      "ĠW O",
+      "PRO C",
+      "Ġrecons tit",
+      "ĠAnomal ous",
+      "ĠBi ol",
+      "St age",
+      "ĠReview s",
+      "Ġrecall ing",
+      "Ġille gal",
+      "l und",
+      "Â ¬",
+      "ut henium",
+      "ĠP es",
+      "Ġo varies",
+      "sol utions",
+      "mass ive",
+      "ĠRA W",
+      "Ġrecon nection",
+      "ĠSus ceptibility",
+      "Ġeconom ical",
+      "cult ured",
+      "ĠSh am",
+      "sq cup",
+      "Ġp ear",
+      "dep osition",
+      "uch s",
+      "ĠS aw",
+      "Ġembol ism",
+      "B ur",
+      "n ar",
+      "ou le",
+      "Ġtex tile",
+      "se ven",
+      "th io",
+      "Ġden oising",
+      "CE P",
+      "Ġubiquit ination",
+      "ĠCarl os",
+      "a P",
+      "Ġfol der",
+      "Ġhemat ological",
+      "il uminescence",
+      "ĠF uel",
+      "ic ion",
+      "ac ulture",
+      "AR B",
+      "ĠTra vel",
+      "F unc",
+      "ac les",
+      "ĠIn te",
+      "Ġvacu a",
+      "Ġcock tail",
+      "ĠIn sp",
+      "Ġcor porate",
+      "Ġdepic ting",
+      "Ġspr int",
+      "ĠmTOR C",
+      "Ġc img",
+      "oc arbon",
+      "ĠD ave",
+      "ĠG b",
+      "ij i",
+      "target ing",
+      "Ġsequest ration",
+      "B ri",
+      "I GF",
+      "Ġanaly tics",
+      "ĠAc inetobacter",
+      "get s",
+      "MP S",
+      "ogl uc",
+      "C ent",
+      "Ġver bs",
+      "Ġinduc tance",
+      "di agram",
+      "Ġrec alled",
+      "Ġcos me",
+      "Ġautom otive",
+      "ĠPD Es",
+      "ĠRe id",
+      "Ġadap ter",
+      "ĠOl iver",
+      "Ġaval anche",
+      "V ir",
+      "ĠT oxicity",
+      "ĠLe u",
+      "Con clusions",
+      "Ġtet ragonal",
+      "ĠDM F",
+      "umann ii",
+      "ĠRequire ments",
+      "t oc",
+      "it é",
+      "Ġcontin ent",
+      "ĠH ank",
+      "ĠDef initions",
+      "GP U",
+      "orig in",
+      "Ġdich ro",
+      "M us",
+      "Ġb ival",
+      "Ġimp ulsive",
+      "Ġassemb le",
+      "Ġpip es",
+      "doc s",
+      "Ġexchang er",
+      "Ġall ograft",
+      "lo yd",
+      "ĠÌ ĭ",
+      "Ġanten atal",
+      "Ġgrass land",
+      "Ġhy stere",
+      "ĠAnti gen",
+      "ĠGener ic",
+      "ĠT uring",
+      "ĠEx cell",
+      "ĠHe in",
+      "aj a",
+      "umin um",
+      "cit abine",
+      "f acial",
+      "iter ation",
+      "Ġsl urry",
+      "AM L",
+      "erge tic",
+      "ĠTH F",
+      "Ġkil ometers",
+      "f g",
+      "ed uc",
+      "id ian",
+      "Ġpredic ates",
+      "Ġradi os",
+      "ĠPer i",
+      "ĠShe ll",
+      "Ġarc sec",
+      "Ġstri atal",
+      "Ġce iling",
+      "olith ic",
+      "Ġexhaus tion",
+      "P UT",
+      "ther s",
+      "ym p",
+      "ĠQ ian",
+      "ĠProg ressive",
+      "Ġw el",
+      "ĠCon vention",
+      "ĠCur ie",
+      "ĠM ans",
+      "ĠN ova",
+      "ĠW ells",
+      "de w",
+      "St andard",
+      "real istic",
+      "trans pose",
+      "ser ial",
+      "ĠT x",
+      "ĠA MR",
+      "Ġind eterm",
+      "ĠLi ouville",
+      "hook rightarrow",
+      "AR s",
+      "Ġbase ball",
+      "ac ious",
+      "agne tization",
+      "es timate",
+      "ĠP AS",
+      "Ġme als",
+      "multi ple",
+      "ĠBiomark ers",
+      "W ide",
+      "ĠTom ography",
+      "//////////////// ////////////////",
+      "Ġres ins",
+      "Ġany where",
+      "IN C",
+      "ĠTe aching",
+      "ĠSam uel",
+      "Ġhall mark",
+      "ĠTh yroid",
+      "oth i",
+      "Ġconst raining",
+      "ĠBar rett",
+      "ĠEr rors",
+      "C ole",
+      "sh aring",
+      "HD L",
+      "Eff ect",
+      "ĠT olerance",
+      "Ġstress ful",
+      "ĠBal ance",
+      "ĠT ech",
+      "Ġval leys",
+      "set up",
+      "ĠRad ical",
+      "ĠMacroph ages",
+      "Ġinter rupt",
+      "Ġdi atom",
+      "col ored",
+      "Ġpy rid",
+      "FD G",
+      "Ã ¦",
+      "Ġre ared",
+      "ĠR ating",
+      "Ġop aque",
+      "pack age",
+      "Ġnas opharyngeal",
+      "Ġprecondition ing",
+      "D iptera",
+      "ĠM ing",
+      "ĠCa ro",
+      "ĠImmun ity",
+      "rif uge",
+      "ĠObj ectives",
+      "g han",
+      "uc cin",
+      "ĠF ors",
+      "ĠF ITC",
+      "Ġse ats",
+      "ĠImp aired",
+      "Ġreef s",
+      "em aker",
+      "Ġoff ices",
+      "Ġaccept ing",
+      "ĠTR AN",
+      "ĠTarget s",
+      "Ġcorrel ator",
+      "Ġsuper capac",
+      "in burgh",
+      "Ġcoll ider",
+      "Ġenter ic",
+      "ĠSTR UCTURE",
+      "Ġmin ister",
+      "ĠArch ae",
+      "Lo op",
+      "ĠA SA",
+      "Ġcont acted",
+      "Ġhis tidine",
+      "fold ed",
+      "S earch",
+      "Ġresp ects",
+      "ĠAT F",
+      "Ġtro uble",
+      "Ġprev ailing",
+      "C p",
+      "ĠT CM",
+      "ĠSp inal",
+      "Ġgu ides",
+      "ev itable",
+      "Ġb rick",
+      "str ings",
+      "ĠHung ary",
+      "Ġe ps",
+      "ent ricular",
+      "Spec ifically",
+      "and o",
+      "iss ues",
+      "osom iasis",
+      "k Da",
+      "Ġas ide",
+      "Ġaden ine",
+      "Ġmotiv ate",
+      "strat ig",
+      "B LE",
+      "ĠDep osition",
+      "m otor",
+      "ĠH ers",
+      "Ġne bul",
+      "ĠBar rier",
+      "Un like",
+      "Ġball istic",
+      "Ġsouth western",
+      "ĠMont real",
+      "S can",
+      "Ġm ould",
+      "Ġinter rup",
+      "small matrix",
+      "Ġelabor ated",
+      "uc ks",
+      "AP S",
+      "ĠCons umption",
+      "cap acity",
+      "inn itus",
+      "Ġgovern ance",
+      "Ġp alsy",
+      "Ġsub mission",
+      "Ġtem ple",
+      "ĠII A",
+      "meth ionine",
+      "Ġker at",
+      "Ġrid ges",
+      "Prom ega",
+      "c ols",
+      "IS P",
+      "Ġap nea",
+      "ĠFl at",
+      "ĠEp igenetic",
+      "Ġpar ish",
+      "ĠPar ametric",
+      "d ash",
+      "f uture",
+      "r ise",
+      "Ġcontract ing",
+      "alg ia",
+      "Ġg oto",
+      "stad t",
+      "Ġfabric ate",
+      "Ġdimer ization",
+      "d ump",
+      "ĠL yn",
+      "Ġrecycl ed",
+      "posed ness",
+      "ĠSens ory",
+      "ï Ŀ",
+      "ĠW et",
+      "Ġdi ethyl",
+      "Ġbl ades",
+      "Ġtim ed",
+      "Ġkey word",
+      "Ġpolyt ope",
+      "ĠG ot",
+      "Ġapproxim ates",
+      "With out",
+      "ĠB ere",
+      "ĠL p",
+      "opl asty",
+      "ĠF ibr",
+      "mod ulated",
+      "ĠAR M",
+      "Ġunde restimate",
+      "ĠC BS",
+      "ĠL ectures",
+      "unc an",
+      "ĠSe ismic",
+      "So ft",
+      "Ġzo oplankton",
+      "Ġencephal opathy",
+      "ĠS SA",
+      "ĠC ros",
+      "ĠH ann",
+      "Ġsh uffle",
+      "sc ription",
+      "ĠRever s",
+      "Stud ies",
+      "Ġsoc ially",
+      "Ġsub cl",
+      "ĠY ong",
+      "og h",
+      "Ġïģ ³",
+      "UD Y",
+      "ĠHa ar",
+      "ĠDoc tor",
+      "Ġint akes",
+      "Ġbar rel",
+      "ĠTR PV",
+      "ĠAgg reg",
+      "ny i",
+      "tun ed",
+      "ac quired",
+      "Ġho ok",
+      "F GF",
+      "Â «",
+      "ĠIn jection",
+      "Ġgra vel",
+      "Ġmicro g",
+      "Ġmen strual",
+      "Fe ature",
+      "I RE",
+      "u u",
+      "ĠS rc",
+      "ĠSt ore",
+      "Ġiniti ator",
+      "PS O",
+      "Ġepile ptic",
+      "Ġcing ulate",
+      "I J",
+      "R ow",
+      "Ġsing ing",
+      "ĠMet han",
+      "ĠAld rich",
+      "Ġtremend ous",
+      "am ining",
+      "Ġtrac ts",
+      "Ġâİ £",
+      "kl ah",
+      "D iv",
+      "ind ol",
+      "Ġind ole",
+      "ex per",
+      "Ġgly cer",
+      "Ġbenz yl",
+      "Ġwors ening",
+      "Ġunambig uous",
+      "u art",
+      "Ġpar sim",
+      "ric ks",
+      "Ġtra il",
+      "ĠBl anc",
+      "Ġamin otransferase",
+      "ĠD OC",
+      "Ġfum ig",
+      "id ic",
+      "ĠCon sequences",
+      "Ġacid ification",
+      "ĠCIF AR",
+      "ĠD atasets",
+      "ĠA MI",
+      "Ġexpl ants",
+      "ĠD iverse",
+      "Ġde phasing",
+      "Ġpar liament",
+      "ip ient",
+      "Ġhoney comb",
+      "he avy",
+      "Ġwaterm ark",
+      "M ED",
+      "d atasets",
+      "w aters",
+      "Pro vid",
+      "inter pret",
+      "rov irus",
+      "I o",
+      "R AD",
+      "Ġl unar",
+      "Ġwe aning",
+      "Ġsensor imotor",
+      "uc a",
+      "Ġinf ect",
+      "ĠUn ique",
+      "GR P",
+      "Q oL",
+      "osp ec",
+      "Ġforward ing",
+      "Es tim",
+      "ÅĦ ski",
+      "ĠM s",
+      "ach n",
+      "Ġro ta",
+      "Ġappoint ment",
+      "ĠMed al",
+      "Ġaden ovirus",
+      "quin ol",
+      "Ġdeuter ium",
+      "te p",
+      "ĠSt yle",
+      "N d",
+      "ay ama",
+      "ĠH amm",
+      "ĠSpec ification",
+      "v ability",
+      "th a",
+      "Ġj itter",
+      "Ġâİ ¦",
+      "a qu",
+      "w ire",
+      "Ġclass ically",
+      "Ġsuper potential",
+      "ĠSpec im",
+      "ĠVari ance",
+      "Ġalbum s",
+      "ĠSen ior",
+      "Ġneurotrans mitter",
+      "ĠRecomb inant",
+      "D CS",
+      "v l",
+      "Ġp f",
+      "Ġin evitable",
+      "ĠN ick",
+      "Ġmanip ulating",
+      "itu ximab",
+      "ce iver",
+      "ĠB ren",
+      "ĠR ace",
+      "Ġret arded",
+      "mod ulin",
+      "Cl inical",
+      "Ġneu rologic",
+      "ĠReg iment",
+      "Ġzo om",
+      "ĠOrth ogonal",
+      "ĠConcer ning",
+      "ĠJur assic",
+      "ĠAr tem",
+      "ĠMel bourne",
+      "b ins",
+      "j l",
+      "Ġin hab",
+      "Ġsq rt",
+      "Ġsemis imple",
+      "ast ric",
+      "ĠPro xim",
+      "ĠVari ants",
+      "Ġa esthetic",
+      "Ġsummar ised",
+      "ĠBeck er",
+      "O CH",
+      "d ale",
+      "Ġm ounting",
+      "and ering",
+      "Ġsoft max",
+      "Ġneuro inflammation",
+      "Ġesophag us",
+      "oper ators",
+      "ĠAD AM",
+      "Ġviol ate",
+      "ĠPH Y",
+      "ed e",
+      "ĠC her",
+      "ors al",
+      "Ġmetam orphic",
+      "ĠI CM",
+      "ĠAb cam",
+      "sl ot",
+      "ser ine",
+      "Ġdu plicates",
+      "ĠME MS",
+      "ĠA bl",
+      "ĠC hel",
+      "ĠAuthor ity",
+      "Ġge o",
+      "Ġhome omorphism",
+      "Ġimmunomod ulatory",
+      "ĠT U",
+      "ĠK T",
+      "ater ally",
+      "ox ides",
+      "teb ral",
+      "Ġcatar act",
+      "le aved",
+      "ig u",
+      "ate ur",
+      "ĠR é",
+      "Ġdiscover ies",
+      "bos on",
+      "oc ated",
+      "j pg",
+      "ĠS ato",
+      "ĠPRO P",
+      "ĠIm plement",
+      "EL ISA",
+      "iqu eness",
+      "Ġsym bion",
+      "ĠFar aday",
+      "ĠPPAR γ",
+      "w itz",
+      "re ward",
+      "ĠB ush",
+      "st ressed",
+      "ĠA bor",
+      "Ġair ways",
+      "Ġinterfer ometry",
+      "C irc",
+      "Ġimmun oprecipitation",
+      "ĠAp ache",
+      "roph osph",
+      "Ġo C",
+      "Ġf rog",
+      "ĠG U",
+      "ff e",
+      "ĠSt ro",
+      "Ġdodec yl",
+      "d an",
+      "f olds",
+      "ĠM ust",
+      "Ġsurround ings",
+      "Ġcod ons",
+      "ond a",
+      "t b",
+      "od ge",
+      "av as",
+      "ĠSe ason",
+      "t ude",
+      "ĠPl asticity",
+      "ĠHawai i",
+      "D EG",
+      "ĠC MD",
+      "Ġsingle ton",
+      "ke ley",
+      "Ġalgebra ically",
+      "Ġnano structured",
+      "eas ible",
+      "Ġoverlo oked",
+      "ĠP ulse",
+      "rom echanical",
+      "ĠEl se",
+      "Ġexcit ons",
+      "ĠConst rained",
+      "Ġco hesion",
+      "Ġreal izing",
+      "ĠRadi ative",
+      "Ġtryp an",
+      "x s",
+      "ĠT as",
+      "Ġmain stream",
+      "Ġcompact ly",
+      "g rowing",
+      "es c",
+      "Ġd N",
+      "ĠSign atures",
+      "ĠFundament als",
+      "Ġex pose",
+      "ĠR ang",
+      "Ġhand ed",
+      "Ġfunctional ization",
+      "Ġpass iv",
+      "al tern",
+      "ag ul",
+      "Ġschem atically",
+      "O W",
+      "Ġ Ö",
+      "ĠP OD",
+      "Ġhe ar",
+      "ym ore",
+      "ĠPrem ier",
+      "S outh",
+      "Ä «",
+      "ĠO BS",
+      "ĠAl g",
+      "gl ia",
+      "ĠTrans membrane",
+      "Ġsphe roids",
+      "ĠR HS",
+      "Ġinc hes",
+      "ĠK ato",
+      "Ġi e",
+      "ĠCom mercial",
+      "Ġanaly tes",
+      "Ġrisk y",
+      "Ġp iston",
+      "ĠMark ovian",
+      "Ġdram a",
+      "Ġc i",
+      "ĠHist ological",
+      "Ġact uation",
+      "disc rete",
+      "carb amoyl",
+      "S MA",
+      "Ġfeed s",
+      "Ġneoplas ia",
+      "ĠControll er",
+      "b een",
+      "glut amine",
+      "in jected",
+      "Ġc rab",
+      "ĠC auses",
+      "ĠSt ory",
+      "Ġvan adium",
+      "ĠT itan",
+      "en ix",
+      "ass ign",
+      "Ġimmun ogenicity",
+      "ĠAp parent",
+      "Ġenh ancers",
+      "ĠS ou",
+      "all oy",
+      "mathb in",
+      "Ġsed ation",
+      "ĠWork shop",
+      "g over",
+      "l st",
+      "Ġup welling",
+      "me z",
+      "Ġpoly propylene",
+      "ĠCol orectal",
+      "ĠRel axation",
+      "Ġfrag ile",
+      "Ä ĥ",
+      "Ġsub graphs",
+      "the oretical",
+      "Oper ator",
+      "ly wood",
+      "aw n",
+      "ĠPer centage",
+      "methyl ation",
+      "corrhiz al",
+      "G rad",
+      "d ens",
+      "ĠH α",
+      "Ġup coming",
+      "Ġvir gin",
+      "N ames",
+      "ĠR yd",
+      "Ġâİ ¤",
+      "phosph orylation",
+      "renew al",
+      "Y ear",
+      "In it",
+      "Ġs elling",
+      "ĠM ASS",
+      "roph in",
+      "ij n",
+      "Con versely",
+      "Ġunivers ally",
+      "orh ombic",
+      "Ġunpredict able",
+      "F ock",
+      "ch air",
+      "iv as",
+      "network s",
+      "Ġterr itories",
+      "th ia",
+      "ĠAm plification",
+      "M arch",
+      "Ġf lam",
+      "ĠCh art",
+      "Ġshort age",
+      "AM ET",
+      "Ġgrap e",
+      "Ġvoltam metry",
+      "Ø ¯",
+      "ĠS CH",
+      "Ġepit hel",
+      "ĠChrom osome",
+      "ĠX L",
+      "ĠPers istent",
+      "Ġtravel ed",
+      "Ġmerid ional",
+      "Ġf printf",
+      "Ġg um",
+      "vis ory",
+      "Un fortunately",
+      "Ġant eced",
+      "Ġfric tional",
+      "D AT",
+      "ac l",
+      "ĠP regnancy",
+      "ĠB Z",
+      "reg ulatory",
+      "stim ulating",
+      "J apan",
+      "m achine",
+      "u ti",
+      "ĠL er",
+      "Ġnan oflu",
+      "prot otype",
+      "identif ication",
+      "klah oma",
+      "ĠEm ploy",
+      "Sch warz",
+      "Ġincorrect ly",
+      "at to",
+      "ri zation",
+      "ism uth",
+      "Ġir is",
+      "iment ary",
+      "Ġinflation ary",
+      "Ġoutflow s",
+      "ĠL ic",
+      "ore ductase",
+      "Ġproceed ing",
+      "ĠT AC",
+      "ĠH TL",
+      "Ġres ides",
+      "str al",
+      "ĠTrans f",
+      "Ġdich otom",
+      "Fil ter",
+      "J une",
+      "is ure",
+      "ĠA de",
+      "Ġij k",
+      "ĠPhil os",
+      "Ġstay ed",
+      "Ġtam oxifen",
+      "Ġaspar agine",
+      "ex ception",
+      "Ġaccum ulating",
+      "ast ro",
+      "Ch ange",
+      "uz i",
+      "Ġl on",
+      "In stead",
+      "Ġcent rally",
+      "ĠD ental",
+      "class ified",
+      "ĠEg yptian",
+      "Add ress",
+      "ĠQuatern ary",
+      "ĠU SP",
+      "co in",
+      "Ġembry ogenesis",
+      "ïĢ ¨",
+      "N ull",
+      "ĠM ixing",
+      "int ensive",
+      "Ġnorm ative",
+      "ĠL ef",
+      "Ġr umen",
+      "ĠTh ai",
+      "Ġsw allow",
+      "Comp onent",
+      "Ġrobo tics",
+      "ĠC ad",
+      "ĠC IP",
+      "ĠAc ids",
+      "ĠO ffic",
+      "ure r",
+      "ĠW ick",
+      "Ġk ink",
+      "ĠSch a",
+      "ĠCharacter istic",
+      "f amilies",
+      "ĠG Cs",
+      "ĠOptim izing",
+      "Ġtim er",
+      "é l",
+      "j in",
+      "re versal",
+      "Ġsand stone",
+      "H N",
+      "t k",
+      "Ġp tr",
+      "Ġmon ochromatic",
+      "Ġfeed forward",
+      "ding ton",
+      "Ġcritic ism",
+      "Ġs ig",
+      "Ġp ace",
+      "ĠT K",
+      "ĠW as",
+      "Ġcertif icate",
+      "Ġst uck",
+      "Ġcor rid",
+      "Ġlocal isation",
+      "Ġsil k",
+      "Ġdig est",
+      "ĠTem ple",
+      "ĠPost erior",
+      "Ġcommut ator",
+      "ts ch",
+      "per me",
+      "ys ed",
+      "Ġmen u",
+      "Ġmid w",
+      "oc atalytic",
+      "Ġpp b",
+      "T ypes",
+      "ar ri",
+      "ĠL OD",
+      "Ġlo an",
+      "sec ret",
+      "Ġcarb ons",
+      "ĠH olog",
+      "olip ids",
+      "Ġupl o",
+      "ĠDN ase",
+      "Ġpuzz le",
+      "Ġst ance",
+      "ĠManc hester",
+      "ĠDet ector",
+      "im s",
+      "ĠTerm s",
+      "ĠP GC",
+      "Ġinc idents",
+      "ie h",
+      "ĠID s",
+      "ĠAh mad",
+      "Ġn ights",
+      "Ġbiom o",
+      "ĠMethyl ation",
+      "u ator",
+      "res ize",
+      "ĠF inger",
+      "ĠW o",
+      "Ġpost er",
+      "Ġsolid ification",
+      "ĠVal idity",
+      "ĠDend ritic",
+      "Ġad herent",
+      "iss ions",
+      "inc tion",
+      "Ġantagon istic",
+      "ĠPrelim inaries",
+      "Ġco val",
+      "Ġmov ies",
+      "Ġbud ding",
+      "K n",
+      "ĠG it",
+      "ĠThere after",
+      "Ġcapac itive",
+      "A z",
+      "ĠT LS",
+      "Ġiniti ates",
+      "ĠD MR",
+      "Ġâī «",
+      "ĠMy ocardial",
+      "ĠRot ation",
+      "CON FIG",
+      "Ġvow el",
+      "Ġoliv ine",
+      "H amiltonian",
+      "Ġst alk",
+      "N eu",
+      "R est",
+      "an ical",
+      "Ġd st",
+      "Ġres h",
+      "Ġexp ressive",
+      "Ġinf ectivity",
+      "ok u",
+      "CT L",
+      "F requency",
+      "Ġprem ise",
+      "W alk",
+      "Ġâ Ĺ",
+      "Ġrel apsed",
+      "t ured",
+      "ĠU ML",
+      "ov an",
+      "ĠRes earchers",
+      "Ġconven iently",
+      "us k",
+      "IN IT",
+      "Eq s",
+      "F actory",
+      "Ġun steady",
+      "ĠAn sw",
+      "Al a",
+      "nit ine",
+      "q p",
+      "ul ous",
+      "res earch",
+      "ĠB rom",
+      "ĠDem oc",
+      "config uration",
+      "ulos ic",
+      "Ġf ra",
+      "Ġg ift",
+      "Th ird",
+      "Cl aim",
+      "Ä Ł",
+      "od iazep",
+      "Ġpro x",
+      "oc ystis",
+      "ĠR PA",
+      "ĠLik ert",
+      "R MS",
+      "t ech",
+      "Ġac ous",
+      "T LR",
+      "b uck",
+      "ĠThe rap",
+      "uss ions",
+      "hel or",
+      "ĠEm otion",
+      "b ird",
+      "Ġth io",
+      "Ġquantit ation",
+      "brack et",
+      "Ġper cept",
+      "Ġsub category",
+      "Ġlight ning",
+      "Ġher nia",
+      "Ġneurot rophic",
+      "SD S",
+      "ĠAnd ers",
+      "Ġslow ing",
+      "strong ly",
+      "ĠC ounting",
+      "ĠIn cluding",
+      "duc tions",
+      "ub ated",
+      "ĠSt orm",
+      "cor related",
+      "Ġautoanti bodies",
+      "ĠM erg",
+      "oc er",
+      "mic utes",
+      "Ġnonlinear ities",
+      "ĠCent ury",
+      "ĠLand scape",
+      "ĠDeriv atives",
+      "ĠContr ary",
+      "Ġcomp ile",
+      "ĠHep atic",
+      "Ġpond s",
+      "Ġorgan ize",
+      "D MSO",
+      "P osition",
+      "Ġb rach",
+      "Ġinf lat",
+      "osp ace",
+      "Ġskew ness",
+      "Ġag itation",
+      "ĠHO MO",
+      "E U",
+      "Ġcom mented",
+      "Ġcor pora",
+      "Ġmal t",
+      "Herm itian",
+      "id ay",
+      "ĠHelm holtz",
+      "ro blast",
+      "ĠC TR",
+      "un ching",
+      "ĠM ond",
+      "ĠCom ment",
+      "Ġoste osarcoma",
+      "post erior",
+      "Ġthym us",
+      "Ġcig arettes",
+      "N W",
+      "o lem",
+      "ĠH ox",
+      "ĠNF L",
+      "ĠAvail able",
+      "ĠS iber",
+      "ĠF eld",
+      "Ġborder line",
+      "Ġbe ats",
+      "Ġorgan ised",
+      "Ġdistingu ishes",
+      "Ġdial og",
+      "ĠBerg er",
+      "ole ic",
+      "Ġnum bered",
+      "Ġreach able",
+      "ĠRoberts on",
+      "ĠCham ber",
+      "nd array",
+      "Ġcytos keletal",
+      "Ġbl ending",
+      "bl ood",
+      "Im port",
+      "Ġoverwhel ming",
+      "Ġi o",
+      "Ġout age",
+      "ĠSch olar",
+      "plac ing",
+      "ĠPol yp",
+      "Dec l",
+      "ĠMED LINE",
+      "ĠK M",
+      "ĠD AP",
+      "err ors",
+      "ĠS HR",
+      "ĠD ex",
+      "ĠG AS",
+      "ĠG ian",
+      "Ġclinic opathological",
+      "Ġïģ ·",
+      "ĠPredic tions",
+      "ĠQuad ratic",
+      "Ġarrhyth mias",
+      "ar id",
+      "Ġcl othing",
+      "ĠFract ure",
+      "ĉ ĠĠĠĠĠ",
+      "add y",
+      "ĠAlber ta",
+      "ĠW ed",
+      "phi re",
+      "ĠEn cryp",
+      "ĠL AB",
+      "ĠF ano",
+      "CT T",
+      "Ġor yz",
+      "ili ac",
+      "ĠL iao",
+      "vers us",
+      "Ġmes o",
+      "Ġmid point",
+      "Ġst ator",
+      "ĠJ enn",
+      "ov sky",
+      "Ġunc over",
+      "eren n",
+      "ĠMc M",
+      "âī Ī",
+      "ĠCirc uits",
+      "Ġfet uses",
+      "Ġaggl omer",
+      "Ġf b",
+      "Ġy y",
+      "at ech",
+      "AR G",
+      "Ġba umannii",
+      "Ġellipso id",
+      "Ġl oses",
+      "Ġun ve",
+      "Ġbut t",
+      "Ġmultic entre",
+      "il ine",
+      "Ġres ort",
+      "Ġcereb rovascular",
+      "ĠDecre ased",
+      "j ud",
+      "s us",
+      "am ol",
+      "const raints",
+      "Ġt een",
+      "ĠPass ive",
+      "ĠCauc asian",
+      "Ġc ran",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠ",
+      "ü n",
+      "ĠDN MT",
+      "Ġt error",
+      "ad renal",
+      "Ġangi ogenic",
+      "ĠInhib itory",
+      "p rag",
+      "Ġco b",
+      "els h",
+      "Ġenhance ments",
+      "ĠSha w",
+      "ĠTak ahashi",
+      "Ġsulph ur",
+      "Ġgrav itation",
+      "ĠPVD F",
+      "m ust",
+      "Â ¢",
+      "as ymptotic",
+      "el man",
+      "ĠP ros",
+      "ĠM AD",
+      "ĠL en",
+      "the rapy",
+      "eful ly",
+      "sulf ur",
+      "ĠT CA",
+      "ad ditive",
+      "tal k",
+      "Ġpig lets",
+      "Ġprosp ect",
+      "ecund ity",
+      "ĠX iang",
+      "hand ler",
+      "Ġcl ath",
+      "Ġmill imeter",
+      "j ar",
+      "Ġbi ophysical",
+      "Ġcomplex ities",
+      "ĠHer b",
+      "Ġrecover s",
+      "ĠVin cent",
+      "ĠPuer to",
+      "E arth",
+      "R AM",
+      "Ġc ables",
+      "des igned",
+      "ĠOscill ation",
+      "Ġme iosis",
+      "Ġfle et",
+      "ĠHunting ton",
+      "ĠB eg",
+      "ĠE Cs",
+      "ĠAn tic",
+      "Ġpractition er",
+      "c ultural",
+      "k at",
+      "Ġrec oil",
+      "ĠIm plicit",
+      "Ġsumm aries",
+      "Ġdiscontinu ed",
+      "Ġencompass ing",
+      "ĠAlt ogether",
+      "ĠD IST",
+      "Ġconst ellation",
+      "ĠEx isting",
+      "Ġconduct ors",
+      "oplas m",
+      "ĠCosm ology",
+      "Z ero",
+      "ĠIn form",
+      "Ġend angered",
+      "Ġweap ons",
+      "at ype",
+      "ĠAs c",
+      "Ġflu ence",
+      "Ġfer ric",
+      "ĠLaure nt",
+      "Ear ly",
+      "Ġs gn",
+      "ĠHad amard",
+      "Ġastr on",
+      "C ys",
+      "ĠTh m",
+      "Ġdec e",
+      "eren cing",
+      "ĠMe ans",
+      "Ġhyd rated",
+      "Ù Ī",
+      "Ġrig orously",
+      "Ġamb ulatory",
+      "ĠDO I",
+      "Hand le",
+      "ĠEnterobacter iaceae",
+      "ĠR Q",
+      "ĠG FR",
+      "pro te",
+      "Ġmig rated",
+      "then ing",
+      "ĠHop kins",
+      "ĠPsych ology",
+      "ig l",
+      "ĠE DS",
+      "ĠâĪ ¶",
+      "Ġrem otely",
+      "ĠÂ ¥",
+      "Ġinsp iration",
+      "ĠâĮ ¬",
+      "ol ian",
+      "Ġsal iency",
+      "ĠD og",
+      "ĠR osa",
+      "oy a",
+      "Ġoccup ies",
+      "cam era",
+      "Ġdecomp ression",
+      "Ġsc att",
+      "Ġinvestig ator",
+      "Ġcount erex",
+      "ĠIFN γ",
+      "ĠPitts burgh",
+      "Ġad minister",
+      "ne gl",
+      "uss is",
+      "MP C",
+      "ĠSw itching",
+      "Ġcool er",
+      "Ġbron chi",
+      "Ġpar alle",
+      "Ġspec kle",
+      "Ġphys iologic",
+      "IN VAL",
+      "Ġheter ologous",
+      "|| |",
+      "org hum",
+      "G AL",
+      "Ġmal formations",
+      "Ġweak ening",
+      "Ġpsych o",
+      "ĠI H",
+      "Ġcontrad ictory",
+      "Ġphon ological",
+      "ĠPerturb ation",
+      "b B",
+      "ĠN os",
+      "TR UE",
+      "fold ing",
+      "phen ol",
+      "ĠL SM",
+      "ĠâĪ Ĺ",
+      "ĠAn gle",
+      "Ġprov incial",
+      "Fe O",
+      "Å Ľ",
+      "ĠI ber",
+      "ress ors",
+      "Ġprolifer ating",
+      "z ers",
+      "organ ism",
+      "âĨ ĵ",
+      "Z O",
+      "c img",
+      "Ġun perturbed",
+      "Ġj j",
+      "Ġelectro dynamics",
+      "ĠEp it",
+      "NT s",
+      "ĠBlo om",
+      "Ġl anth",
+      "am inant",
+      "ĠSw ift",
+      "Europe an",
+      "Ġaff erent",
+      "Red uce",
+      "p ublished",
+      "ĠF itting",
+      "ĠF ungal",
+      "Ġtrib e",
+      "rec ting",
+      "Ġconjug acy",
+      "im eters",
+      "ĠC ec",
+      "ĠK H",
+      "cast le",
+      "Ġsept al",
+      "rele asing",
+      "Ġo ss",
+      "ĠÂ ¦",
+      "ĠMiss ing",
+      "ĠFat igue",
+      "ĠBase ball",
+      "Ġimmunoblot ting",
+      "Ġo h",
+      "or ations",
+      "Ġv ine",
+      "az y",
+      "ser um",
+      "Ġlook up",
+      "Ġne ovascular",
+      "ia h",
+      "so il",
+      "Ġair flow",
+      "ĠSlo an",
+      "h im",
+      "ç ļ",
+      "loc ated",
+      "z antine",
+      "ĠS uccessful",
+      "em inal",
+      "ĠD imensional",
+      "ĠN SA",
+      "ĠLog istic",
+      "emet ery",
+      "Ġb rak",
+      "ant al",
+      "so uth",
+      "Ġprot otypes",
+      "Ġadv ised",
+      "Ġideal ized",
+      "ophy tic",
+      "nb sp",
+      "B inary",
+      "H yp",
+      "J oh",
+      "p olation",
+      "Ġpoly vinyl",
+      "estim ated",
+      "Ġox ytocin",
+      "ĠLet ter",
+      "ĠImp air",
+      "Ġenvelop es",
+      "main ly",
+      "Ġm ys",
+      "Ġint ras",
+      "Ġbi ogenic",
+      "cy steine",
+      "Ġur ic",
+      "ĠCy an",
+      "ryp tion",
+      "Ġphotore ceptor",
+      "ĠT oxic",
+      "ĠG amm",
+      "Ġcontain ment",
+      "Ig G",
+      "S qu",
+      "Ġperf used",
+      "Ġbios ensors",
+      "Ġmag matic",
+      "R ate",
+      "ĠT f",
+      "Ġsec rete",
+      "Ġcritical ity",
+      "Ġcomposition ally",
+      "ĠBr uce",
+      "S Z",
+      "ĠS port",
+      "ĠE I",
+      "Ġdise ased",
+      "Ġpres chool",
+      "ĠHar vey",
+      "ĠPT H",
+      "Ġbil ayers",
+      "ĠOscill ations",
+      "ĠHon or",
+      "ĠC CN",
+      "ĠM OT",
+      "ĠL loyd",
+      "Ġtrap ez",
+      "Ġbud s",
+      "OFF SET",
+      "Ġmac romolecules",
+      "Ġbil irubin",
+      "ol ly",
+      "Ġutil ities",
+      "minist ered",
+      "Ġglob e",
+      "OLOG Y",
+      "rop ods",
+      "ĠMD M",
+      "ĠPy Object",
+      "mac roph",
+      "ĠP BMCs",
+      "osp heres",
+      "Ġcatast rophic",
+      "ĠNavig ation",
+      "ĠL SD",
+      "Ġcre am",
+      "Ġdere g",
+      "b onded",
+      "ren ts",
+      "Ġpotenti ation",
+      "Ġst ro",
+      "Ġst eeper",
+      "ulin um",
+      "Ġperiodon titis",
+      "ar ization",
+      "âĪ ª",
+      "amic in",
+      "Ġmagne tized",
+      "ĠNutri tional",
+      "Ġacc ord",
+      "ga ard",
+      "FT IR",
+      "r amethyl",
+      "ĠG le",
+      "M el",
+      "ĠCT L",
+      "Ġtransl ating",
+      "Ġauto immunity",
+      "oler ant",
+      "triang leq",
+      "am o",
+      "Ġv el",
+      "ĠH CN",
+      "ĠH amming",
+      "ĠVen us",
+      "ĠG ad",
+      "ĠO wing",
+      "In formation",
+      "ĠSchem es",
+      "caro tene",
+      "I ts",
+      "an is",
+      "Ġre play",
+      "Ġto uc",
+      "LE CT",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠ",
+      "Ġtab ulated",
+      "ĠSchott ky",
+      "F ar",
+      "am ation",
+      "ĠR ies",
+      "Ġexp ects",
+      "ĠInst ability",
+      "Ġs ons",
+      "Ġdec k",
+      "Ġïģ ¥",
+      "ĠSign ature",
+      "Ġlith osphere",
+      "W W",
+      "m akers",
+      "ugh ters",
+      "Ġâİ ¡",
+      "ardi an",
+      "à ¦",
+      "Ġac cepts",
+      "ĠO SA",
+      "Ġγ δ",
+      "non umber",
+      "S elect",
+      "l ite",
+      "ĠA queous",
+      "ag awa",
+      "ĠEd inburgh",
+      "ĠMemb ranes",
+      "ĠS IG",
+      "ak ia",
+      "Ġtest es",
+      "Ġhel i",
+      "++ ++",
+      "Ġultraf ast",
+      "Ġmaneu ver",
+      "ĠD ate",
+      "ph in",
+      "ĠK ad",
+      "Ġtransfer ase",
+      "P ers",
+      "Ġt ones",
+      "ĠS GD",
+      "ant o",
+      "ĠO range",
+      "ĠGe ography",
+      "ĠAcc umulation",
+      "at y",
+      "Ġbe ating",
+      "Ġover lying",
+      "ĠND VI",
+      "ĠTown ship",
+      "j ing",
+      "ĠN OS",
+      "play er",
+      "ĠMD D",
+      "ĠHung arian",
+      "Ġd w",
+      "ĠH in",
+      "Ġvalid ating",
+      "Ġcolor imetric",
+      "ĠSupers ymmetric",
+      "F UNC",
+      "g ically",
+      "of uran",
+      "---- ---",
+      "Ġimp ing",
+      "similar ity",
+      "ĠD OX",
+      "ĠG lo",
+      "iv irus",
+      "list ed",
+      "Ġbus y",
+      "ipro floxacin",
+      "Ġan xi",
+      "Ġbl unt",
+      "Ġproced ural",
+      "Ġunknown s",
+      "Ad S",
+      "thick ness",
+      "follow s",
+      "cl osing",
+      "environment al",
+      "ĠFeed ing",
+      "un ami",
+      "end e",
+      "ip ine",
+      "Ġimpact ing",
+      "Ġpenet rating",
+      "amb ia",
+      "ĠWave let",
+      "Ġfilament ous",
+      "Ġl eng",
+      "ĠS CA",
+      "ĠE ther",
+      "met all",
+      "Ġfr inge",
+      "ĠAdj ust",
+      "us z",
+      "ĠR ey",
+      "ĠBo yd",
+      "Ġburn out",
+      "Ġco ok",
+      "Ġnow adays",
+      "ĠDispers ion",
+      "ĠRodrig uez",
+      "F actor",
+      "ĠO klahoma",
+      "Ġun ital",
+      "Ġpredict ability",
+      "Ġlith ography",
+      "è s",
+      "W illi",
+      "un al",
+      "ast ing",
+      "cor rection",
+      "ĠD ed",
+      "ĠSoci o",
+      "ĠChap man",
+      "ĠE co",
+      "Ġonc ogene",
+      "ĠDri ve",
+      "Ġfun nel",
+      "u is",
+      "ĠGEN ER",
+      "ĠA CR",
+      "Ġworkload s",
+      "Ġocta hedral",
+      "v ich",
+      "en burg",
+      "Ġimpro per",
+      "dec oded",
+      "Ġimmunos orbent",
+      "Ġinhom ogeneity",
+      "R K",
+      "on ically",
+      "Ġglycoprotein s",
+      "on ics",
+      "ĠF ok",
+      "ĠB ras",
+      "ĠCalc ulus",
+      "ĠM oss",
+      "ĠR K",
+      "Ġvi olet",
+      "Ġlymph omas",
+      "ens pace",
+      "ĠPal ae",
+      "Ġren in",
+      "ph ant",
+      "ĠRE D",
+      "Ġfault y",
+      "Ri emann",
+      "Ã ī",
+      "ĠEll i",
+      "B ol",
+      "T n",
+      "Y ang",
+      "g ender",
+      "Ġdet uning",
+      "Ġoper on",
+      "Ġinsectic ide",
+      "es i",
+      "am on",
+      "ĠS CD",
+      "ĠB ath",
+      "ĠâĢ ĸ",
+      "ĠGe ographic",
+      "Ġcycl ohex",
+      "ĠConf idence",
+      "Ġcom et",
+      "Ġfol ate",
+      "ob server",
+      "Ġvis itors",
+      "ext ra",
+      "at eness",
+      "ĠS PT",
+      "arc ane",
+      "Ġhol istic",
+      "sem i",
+      "ĠM ild",
+      "Ġsm ear",
+      "Ġcycl ase",
+      "Ġan ymore",
+      "Ġse agrass",
+      "Ġcons ortium",
+      "Ġfin ishes",
+      "cy an",
+      "duct ance",
+      "f rost",
+      "here after",
+      "Ġpres criptions",
+      "Ġcm d",
+      "ĠPer ceived",
+      "co ordinates",
+      "Ġst yl",
+      "ĠB ard",
+      "ĠH oll",
+      "Ġsi RNAs",
+      "s ugg",
+      "Ġth r",
+      "Ġmain land",
+      "SC H",
+      "Ġasser tions",
+      "Ġbab ies",
+      "Ġrecap it",
+      "T ok",
+      "Ġres ected",
+      "con struct",
+      "B er",
+      "Ġch oline",
+      "Ġunit arity",
+      "Ġcataly zes",
+      "det ector",
+      "ĠS MB",
+      "ter y",
+      "cl uded",
+      "ĠAb breviations",
+      "ĠOlive ira",
+      "L OC",
+      "z in",
+      "ĠLore nz",
+      "K ernel",
+      "ly n",
+      "ĠL EP",
+      "son i",
+      "Ġsept um",
+      "T MS",
+      "Ġun modified",
+      "bor ough",
+      "ĠAud io",
+      "Ġdoll ars",
+      "CM D",
+      "Ġnorth western",
+      "Ġpal mit",
+      "ragal actic",
+      "ĠM iz",
+      "F H",
+      "conf idence",
+      "N EXT",
+      "ĠA GE",
+      "ĠEq n",
+      "ĠClass es",
+      "Ġmis leading",
+      "ĠPK A",
+      "Ġanch ored",
+      "ĠR ip",
+      "ph ag",
+      "Ġint ubation",
+      "ĠAng ular",
+      "ĠB EC",
+      "Th r",
+      "Ġorgan isations",
+      "Ġcomfort able",
+      "Ġcommission ed",
+      "p oll",
+      "y dia",
+      "in stead",
+      "Ġpass word",
+      "Ġcompl iant",
+      "ĠPrec ipitation",
+      "ophosph amide",
+      "ust ers",
+      "Ġpneum ococcal",
+      "Ġtom ographic",
+      "tida e",
+      "ĠFir micutes",
+      "b w",
+      "ĠPD B",
+      "ĠGP Us",
+      "ĠPlan ar",
+      "Ġverb ose",
+      "Summ ary",
+      "l ance",
+      "ĠE GFP",
+      "ong ru",
+      "Com plex",
+      "ĠWhe at",
+      "uc he",
+      "ĠM CA",
+      "ĠPro jection",
+      "Ġstat s",
+      "Ġsumm and",
+      "dim ethoxyphenyl",
+      "ĠAB STRACT",
+      "Ġcaroten oid",
+      "Ġbro ke",
+      "ĠDesign ing",
+      "ĠHet ero",
+      "ĠCarls bad",
+      "C ov",
+      "in eral",
+      "Ġanaly te",
+      "ĠCo leman",
+      "Ġeigen state",
+      "ĠHol land",
+      "ERS ION",
+      "ĠD ak",
+      "ell ers",
+      "ĠÃ ĺ",
+      "miss ing",
+      "dep osited",
+      "ĠLinc oln",
+      "an ion",
+      "ĠSP EC",
+      "Ġfertil izer",
+      "ĠC PS",
+      "Ġco factor",
+      "Ġtre n",
+      "Ġcal endar",
+      "Ġyoung est",
+      "STAT US",
+      "ĠEXPERIM ENTAL",
+      "Ġs r",
+      "Ġn l",
+      "ĠM ES",
+      "Stud y",
+      "p adding",
+      "Ġat opic",
+      "ĠO G",
+      "Ġent rainment",
+      "AF M",
+      "ĠC ou",
+      "We b",
+      "ĠMic roscopic",
+      "Ġunambig uously",
+      "D ay",
+      "y otrophic",
+      "re ous",
+      "Ġs arcom",
+      "ĠV AL",
+      "Ġhind ered",
+      "ĠRE M",
+      "ot rexate",
+      "oc arcin",
+      "ĠAl k",
+      "Ġbre vity",
+      "fact ual",
+      "C er",
+      "di ox",
+      "oph ical",
+      "Ġly tic",
+      "T ake",
+      "Ġint end",
+      "ĠCl a",
+      "Ġaster oid",
+      "ĠS EP",
+      "ap enem",
+      "univers al",
+      "Ġo ceans",
+      "Ġmon oid",
+      "Ġsepar ator",
+      "ĠP orous",
+      "Ġpost operatively",
+      "Ġsem in",
+      "ĠDis play",
+      "Ġhyd rolase",
+      "transfer ases",
+      "Ġthromb us",
+      "ĠO v",
+      "ĠDie lectric",
+      "Ġcomp elling",
+      "ass ing",
+      "ĠM AS",
+      "ull ary",
+      "ĠMor i",
+      "ĠPath ogenesis",
+      "ĠBre aking",
+      "ĠPL GA",
+      "cool ing",
+      "Â §",
+      "Ġfe e",
+      "Ġreduc ible",
+      "Ġdiver ge",
+      "Ġque ues",
+      "Ġmush room",
+      "Ġdeacetyl ase",
+      "Y FP",
+      "Ġdis reg",
+      "ĠAr rays",
+      "process es",
+      "ĠTransport ation",
+      "Ġundet ectable",
+      "bur sts",
+      "Ġphospholip ase",
+      "O ption",
+      "as in",
+      "Ġn octurnal",
+      "te z",
+      "ĠDis ruption",
+      "oser ine",
+      "behavi or",
+      "ĠT ony",
+      "ĠK ot",
+      "ie val",
+      "Ġmy ofib",
+      "Ġhal ogen",
+      "ĠC PR",
+      "ploy ed",
+      "ĠPol ymers",
+      "Ġaden oma",
+      "Ġquar tile",
+      "Ġquatern ary",
+      "ĠIra q",
+      "Ġs ieve",
+      "Ġint ractable",
+      "Ġfabric s",
+      "continu um",
+      "ĠEmerg ence",
+      "P ot",
+      "iti sm",
+      "ven ess",
+      "ho e",
+      "Ġred es",
+      "ĠHR P",
+      "ploid y",
+      "pic uous",
+      "og o",
+      "ĠG ag",
+      "Ġnom inated",
+      "occup ied",
+      "Ġqu ench",
+      "rop olis",
+      "nucle otide",
+      "ĠEvent ually",
+      "Ñ ı",
+      "ĠCl ock",
+      "ĠSte ady",
+      "opol ymers",
+      "ĠA RE",
+      "ir nov",
+      "hel f",
+      "bl ob",
+      "down load",
+      "PL L",
+      "UN T",
+      "predic tions",
+      "Ġocc ipital",
+      "t oxic",
+      "ĠV ice",
+      "Ġang io",
+      "Cu O",
+      "Ġresist ances",
+      "ffl ffl",
+      "D istribution",
+      "G re",
+      "on amide",
+      "ĠI OP",
+      "UN EL",
+      "Ġa ids",
+      "ĠH UV",
+      "EC M",
+      "ĠP AD",
+      "ĠAg NPs",
+      "Pr int",
+      "Ġlam ellar",
+      "ĠUltr ason",
+      "se vere",
+      "ĠAn notation",
+      "N IR",
+      "s gn",
+      "ĠO ften",
+      "Ġit erate",
+      "Ġcar riage",
+      "sp herical",
+      "ĠF rid",
+      "Ġdiff ract",
+      "ĠBas al",
+      "Ġuns atisf",
+      "ĠDys function",
+      "arboxyl ic",
+      "ĠCol lective",
+      "Ġdegrad ing",
+      "Ġadi posity",
+      "Ġfif ty",
+      "Ġpar s",
+      "ĠOptim ized",
+      "oc aine",
+      "Ġb b",
+      "ĠS hip",
+      "ĠL W",
+      "Ġtre mor",
+      "ĠÃ £",
+      "Ġnucle ons",
+      "Ġscienti st",
+      "ĠM ish",
+      "g ression",
+      "ĠM erc",
+      "ĠF lem",
+      "Ġcor als",
+      "In cre",
+      "ĠD SP",
+      "Ġdef enses",
+      "dim er",
+      "ather ine",
+      "ot ubes",
+      "str ide",
+      "ĠAlter ations",
+      "Ġo est",
+      "ĠB IC",
+      "Ġradi ated",
+      "Ġket amine",
+      "Ġdissimilar ity",
+      "ĠAnc ient",
+      "ĠH ed",
+      "Ġatt r",
+      "ĠIs a",
+      "Ġion ospheric",
+      "Ġgover nor",
+      "ĠEstim ated",
+      "Ġultr athin",
+      "Up date",
+      "Ġimmuno assay",
+      "Ġconject ured",
+      "Ġ REF",
+      "ĠSi egel",
+      "Ad v",
+      "M em",
+      "Ġp ups",
+      "ĠAP PL",
+      "ecom posable",
+      "j ournal",
+      "ĠR ol",
+      "ĠL ob",
+      "ring ton",
+      "Ġnons ingular",
+      "Ġcit ric",
+      "ion es",
+      "os itis",
+      "AL Y",
+      "Ġmen tions",
+      "ĠMark ers",
+      "algebra ic",
+      "Ġflatten ed",
+      "Ġm ail",
+      "ĠT GA",
+      "ĠP MA",
+      "ĠN aval",
+      "Ġfac ilitation",
+      "Ġun identified",
+      "Ġem pathy",
+      "ject ories",
+      "log its",
+      "Ġperman ently",
+      "Ġbott les",
+      "ĠBeng al",
+      "Ġpean ut",
+      "Ġcapill aries",
+      "eren ts",
+      "ĠLo oking",
+      "chang es",
+      "ĠMag ell",
+      "ĠC MC",
+      "ĠV erm",
+      "Ġsubs cales",
+      "dem and",
+      "ore xia",
+      "Ġachieve ments",
+      "ĠRobust ness",
+      "ĠWall ace",
+      "ĠD TT",
+      "og els",
+      "ock er",
+      "ĠSp ike",
+      "Ġpain ter",
+      "Ġbus es",
+      "Ġpoll uted",
+      "Ġt ort",
+      "ĠP PP",
+      "ne x",
+      "ext ended",
+      "ucal ypt",
+      "Ġpro static",
+      "ĠF CC",
+      "Ġk ick",
+      "oy al",
+      "epoch s",
+      "h ss",
+      "y on",
+      "Ġd ans",
+      "ĠA w",
+      "Ġad versely",
+      "Ġalt ogether",
+      "Ġophthal m",
+      "Ġc pu",
+      "ĠF RET",
+      "Ġfore nsic",
+      "Ġhot spots",
+      "Ġpain tings",
+      "Ġo mn",
+      "Ġp S",
+      "og lu",
+      "of ol",
+      "FT s",
+      "Ġderm at",
+      "prag ma",
+      "Ġb ump",
+      "ĠC ir",
+      "a S",
+      "Ġn aked",
+      "ĠN LS",
+      "ĠSp itzer",
+      "Ġsal vage",
+      "Ġintu itively",
+      "Ġcas ual",
+      "Ġf ired",
+      "ver ages",
+      "ĠBur den",
+      "W ang",
+      "yle m",
+      "Ġradi ographs",
+      "ĠSch iff",
+      "OL UTION",
+      "C ross",
+      "Ġh ints",
+      "ow ing",
+      "ĠSt reng",
+      "ĠAN Y",
+      "Ġwor ry",
+      "ĠRog er",
+      "Ġtrabec ular",
+      "B and",
+      "ĠN ec",
+      "ip es",
+      "to ol",
+      "ĠIL C",
+      "i Äĩ",
+      "o cean",
+      "ĠA ri",
+      "AM A",
+      "ĠVer tex",
+      "activ ate",
+      "L ocation",
+      "on ts",
+      "Ġh s",
+      "Ġsl ender",
+      "ref ring",
+      "ĠEnd ogenous",
+      "adi abatic",
+      "Ġcryp tic",
+      "Ġerad ication",
+      "ĠKev in",
+      "Ġm c",
+      "Ġcardi o",
+      "Ġphosphor yl",
+      "W itten",
+      "Ġs cl",
+      "ĠI w",
+      "ĠM ade",
+      "Ġfound ing",
+      "ofl ag",
+      "al ine",
+      "hor izontal",
+      "ĠGeneral ization",
+      "psy chiatric",
+      "ĠD uncan",
+      "ĠSn O",
+      "ĠA ar",
+      "Ġg g",
+      "Ġpre mi",
+      "ĠSt rom",
+      "ĠEx plan",
+      "Ġleth ality",
+      "Ï Ĥ",
+      "od o",
+      "Ġsub scrib",
+      "ĠST UDY",
+      "Ġoutper formed",
+      "Ġcoval ently",
+      "M HC",
+      "f ail",
+      "ĠK ac",
+      "EG R",
+      "ĠTR I",
+      "rob ot",
+      "ĠCandid ate",
+      "ĠTN BC",
+      "Ġarchae ological",
+      "E ukary",
+      "Ġl ava",
+      "di pole",
+      "Ġunc ons",
+      "An ti",
+      "Ġpred nis",
+      "ĠRob in",
+      "Ġstratig raphic",
+      "ĠÂ ¤",
+      "Ġfin ance",
+      "ĠStud io",
+      "re nder",
+      "Ġre aring",
+      "Ġg er",
+      "ĠO pt",
+      "ĠMan ifolds",
+      "Ġdest abil",
+      "Ġtel omerase",
+      "Ġpick ing",
+      "Ġamplic on",
+      "Ġyear ly",
+      "ĠN CC",
+      "ins er",
+      "ĠEn richment",
+      "ĠMicro structure",
+      "ĠWar ren",
+      "ophys ics",
+      "Ġfif teen",
+      "Å ij",
+      "Ġreview er",
+      "Ġsk illed",
+      "Ġmagnet oresistance",
+      "Ġrecon figuration",
+      "Ġpo et",
+      "Ġpred etermined",
+      "Ġcry opres",
+      "Ġattract ors",
+      "Ġprojec tile",
+      "ĠC rystals",
+      "ĠM CM",
+      "ĠX anth",
+      "Ġclock wise",
+      "regn ant",
+      "Ġg ated",
+      "ry za",
+      "ĠP rosp",
+      "ad in",
+      "Ġm olybdenum",
+      "ĠAl ps",
+      "ĠBal d",
+      "Ġhall uc",
+      "ud o",
+      "Ġmon t",
+      "ĠFl ash",
+      "Ġpull ing",
+      "ĠL Q",
+      "ĠWals h",
+      "ĠTh omson",
+      "mes on",
+      "Ġinter cal",
+      "Ġel apsed",
+      "FF FF",
+      "ĠFore casting",
+      "à ¯",
+      "ĠL SP",
+      "end orf",
+      "Ġx ml",
+      "sub strate",
+      "M u",
+      "d uring",
+      "oc onstr",
+      "EM A",
+      "Ġïĥ «",
+      "ĠD FS",
+      "ĠV on",
+      "Ġfat hers",
+      "Ġunc o",
+      "ĠUnd erg",
+      "Ġmultiplex ing",
+      "at ra",
+      "Ġco hesive",
+      "ĠU I",
+      "ĠPre v",
+      "çļ Ħ",
+      "c um",
+      "h f",
+      "ĠS CN",
+      "atal ysis",
+      "ĠAr sen",
+      "amp ing",
+      "ĠPl astic",
+      "ĠMad ison",
+      "Ġsuprem um",
+      "ĠC ited",
+      "Ġare n",
+      "isk i",
+      "in el",
+      "st ro",
+      "Ġcor rupted",
+      "Ġgl ab",
+      "Ġcardi opulmonary",
+      "Ġprag matic",
+      "C AG",
+      "St ack",
+      "thi oxo",
+      "ĠRepro ductive",
+      "Ġste atosis",
+      "B est",
+      "ĠB ars",
+      "Ġr acing",
+      "ĠU tah",
+      "equ ivalence",
+      "ĠFif ty",
+      "ĠCytok ine",
+      "Ġutil ised",
+      "hor izon",
+      "our acil",
+      "ivers ary",
+      "em er",
+      "ĠQ uestions",
+      "Ġlink ages",
+      "anche z",
+      "V V",
+      "Ġphotod et",
+      "k owski",
+      "RE ST",
+      "Ġhost ing",
+      "Ġpush ing",
+      "Ġneurot oxicity",
+      "S Q",
+      "r st",
+      "Ġh ockey",
+      "Ġtri ps",
+      "ĠInd oor",
+      "em atics",
+      "Ġtrans ect",
+      "ĠAB I",
+      "ag ar",
+      "âĪ ļ",
+      "eg enerate",
+      "ĠQ P",
+      "MI D",
+      "ĠAc cept",
+      "ĠCy ber",
+      "N orth",
+      "Ġd θ",
+      "all a",
+      "Ġbra id",
+      "f inding",
+      "al in",
+      "ĠL ST",
+      "ĠL ax",
+      "ud in",
+      "Ġi NOS",
+      "con vert",
+      "AC A",
+      "ĠGu an",
+      "Ġlymph ocytic",
+      "Ġsyll able",
+      "ĠT OR",
+      "ĠS CR",
+      "ĠA J",
+      "Ġout burst",
+      "bl adder",
+      "OT A",
+      "aud io",
+      "chrom en",
+      "Ñģ ÑĤ",
+      "Ġgrate fully",
+      "Ġt iling",
+      "Ġqu it",
+      "sh an",
+      "ĠAcc retion",
+      "Ġnarrow ing",
+      "ĠInduc es",
+      "M ic",
+      "Ġf uc",
+      "Ġth alamus",
+      "AN ES",
+      "Ġquatern ion",
+      "ĠLister ia",
+      "d uality",
+      "he nd",
+      "and e",
+      "Ġpa ro",
+      "Ġinsp ected",
+      "ques tion",
+      "ĠH oney",
+      "Ġch unks",
+      "Ġfore arm",
+      "radi ents",
+      "ific antly",
+      "ob ank",
+      "Ġsome where",
+      "Ġmon etary",
+      "ĠLouis iana",
+      "Ġem ulsions",
+      "Ġprogram mable",
+      "Ġmanif ests",
+      "ĠMart inez",
+      "Ġt ed",
+      "em en",
+      "ann i",
+      "Ġoverl aid",
+      "Ġvir ulent",
+      "M ask",
+      "ĠU tility",
+      "Ġw k",
+      "ose xual",
+      "ĠEar l",
+      "d ar",
+      "h dr",
+      "ract ors",
+      "Ġconstruct or",
+      "Ġnas cent",
+      "inz burg",
+      "ĠCra ig",
+      "Ġplex us",
+      "re verse",
+      "og rav",
+      "tag s",
+      "Ġcalibr ate",
+      "à ®",
+      "Ġh ide",
+      "ĠF ol",
+      "Ġinter acted",
+      "Ġconf ron",
+      "mark et",
+      "Ġsoci odemographic",
+      "ĠLuc as",
+      "ĠM CT",
+      "ĠR SS",
+      "Ġmicro plate",
+      "under st",
+      "I tal",
+      "ĠC MR",
+      "rec y",
+      "ĠPC OS",
+      "Ġdetox ification",
+      "Ġsubt ree",
+      "Ġsubs ections",
+      "Ġpropos itions",
+      "Acknowledg ements",
+      "reinfor ced",
+      "l is",
+      "ĠC IR",
+      "Ġim printed",
+      "vi um",
+      "af ic",
+      "Ġcheck list",
+      "ĠR x",
+      "ĠE ph",
+      "Ġsol der",
+      "trans formation",
+      "ĠStra it",
+      "az ar",
+      "Ġhand ler",
+      "ke let",
+      "B CL",
+      "M ath",
+      "Ġw ishes",
+      "um inescent",
+      "ĠP EC",
+      "ir t",
+      "yl idene",
+      "Ġlo osely",
+      "na issance",
+      "IL s",
+      "fo il",
+      "ĠGN U",
+      "ĠK et",
+      "vi x",
+      "ĠPl ain",
+      "ĠRE S",
+      "Ġparent ing",
+      "ĠConn ection",
+      "Ġrhiz osphere",
+      "opre valence",
+      "i atic",
+      "Ġp A",
+      "ĠV il",
+      "set ting",
+      "ĠRe LU",
+      "ĠBO OST",
+      "Ġappreci ate",
+      "b x",
+      "ore st",
+      "olog ie",
+      "Ġpal p",
+      "fo o",
+      "us ual",
+      "Ġquestion ed",
+      "Ġtrig on",
+      "ĠGF AP",
+      "ĠKy oto",
+      "dis e",
+      "anti le",
+      "ü ck",
+      "ĠQuanti zation",
+      "Ġs cler",
+      "Ġbe half",
+      "ĠD uality",
+      "Ġmagnetic ally",
+      "Ġeleg ant",
+      "U A",
+      "ep is",
+      "Ġsub clinical",
+      "ont rol",
+      "ĠChemical s",
+      "Util s",
+      "Ġlow ers",
+      "ext raction",
+      "Ġampl ifiers",
+      "ĠEnt ry",
+      "ĠWOR K",
+      "Ġthrombocyt openia",
+      "M il",
+      "id us",
+      "emb ry",
+      "man ager",
+      "ĠCo ordination",
+      "ĠPhen otypic",
+      "ch unk",
+      "Ġhypot ension",
+      "Ġcry ogenic",
+      "Ġreact ants",
+      "ĠM MSE",
+      "Ġcent ros",
+      "ĠBut ler",
+      "Ġcav itation",
+      "ĠLess ons",
+      "es tion",
+      "ĠM IS",
+      "ass oci",
+      "AP E",
+      "ĠEuler ian",
+      "Ġrecre ational",
+      "ĠNe o",
+      "ĠCD M",
+      "rep eat",
+      "det ails",
+      "B al",
+      "ST A",
+      "Ġâī º",
+      "ĠCam ero",
+      "ĠTele vision",
+      "Ġwork force",
+      "Ġcomputer ized",
+      "Ġextra ordinary",
+      "Ġrib onucle",
+      "Ġhydroph obicity",
+      "ĠFeas ibility",
+      "O l",
+      "T w",
+      "ĠM am",
+      "ĠF AC",
+      "pro fit",
+      "negl igible",
+      "ĠF ruit",
+      "Ġear s",
+      "Ġshe aring",
+      "ĠCorrespond ing",
+      "f un",
+      "i eck",
+      "m os",
+      "ĠE MI",
+      "ĠSome times",
+      "Ġfluor ine",
+      "Ġdeterg ent",
+      "Ġal g",
+      "rac es",
+      "iv able",
+      "CO MM",
+      "ĠSw itch",
+      "Ġstra ined",
+      "vir tual",
+      "Tem perature",
+      "Ġcredi ble",
+      "ĠG PCR",
+      "ĠDe bye",
+      "ĠL it",
+      "Ġhe mic",
+      "Ġtrans ducers",
+      "met ast",
+      "adi ene",
+      "Ġoryz ae",
+      "t n",
+      "Ġafter noon",
+      "ĠArab ian",
+      "ĠChrom atin",
+      "Ġxen ografts",
+      "Ġcrypt ographic",
+      "Ġax illary",
+      "Ġvolunte er",
+      "ĠNev ada",
+      "Ġp ions",
+      "un known",
+      "ĠF U",
+      "ven ously",
+      "radi o",
+      "ĠLab our",
+      "ĠVill age",
+      "R ic",
+      "Ġmet at",
+      "Ġser otypes",
+      "reg ression",
+      "s aturation",
+      "re ra",
+      "Ġfar ther",
+      "Ġround ing",
+      "Ġlib itum",
+      "Ġsh uff",
+      "ĠO w",
+      "Ġlocal ised",
+      "ĠAL G",
+      "Ġhypert rophic",
+      "p pm",
+      "im ine",
+      "ĠA the",
+      "Ġan hydro",
+      "Ġsup ramolecular",
+      "Ġmac ros",
+      "acet ed",
+      "ĠOl iv",
+      "Ġmotiv ational",
+      "ĠC ave",
+      "enz ie",
+      "Ġaffili ated",
+      "Ferm i",
+      "Ġequal ities",
+      "ĠMil an",
+      "Ġd ressed",
+      "Ġan ger",
+      "ad os",
+      "Ġav g",
+      "ĠPh on",
+      "Ġradio activity",
+      "ĠE ch",
+      "Ġorgan oids",
+      "Ġïģ §",
+      "ĠAnth rop",
+      "l ateral",
+      "Ġal pine",
+      "Ġaud it",
+      "W ER",
+      "ĠC SC",
+      "Ġrank ings",
+      "ĠER R",
+      "GL ER",
+      "Ob viously",
+      "ĠMad rid",
+      "obenz ene",
+      "other mia",
+      "Ġrespons ibilities",
+      "omes tic",
+      "ĠInf lation",
+      "Ġepidem ics",
+      "Ġt aut",
+      "ph os",
+      "ĠUn less",
+      "Ġge omagnetic",
+      "ĠCF TR",
+      "vel d",
+      "ari etal",
+      "Ġend otoxin",
+      "AD P",
+      "Ġsupp ressive",
+      "rand ial",
+      "Ġïĥ ©",
+      "exc ited",
+      "ĠInn ate",
+      "ĠL ópez",
+      "omyc etes",
+      "Ġbe autiful",
+      "ir k",
+      "ĠH wang",
+      "ĠU SE",
+      "ÏĢ i",
+      "Rec ord",
+      "Att ribute",
+      "Ġre acts",
+      "ĠB und",
+      "Ġcow ork",
+      "Ġconf luence",
+      "ĠReg ardless",
+      "Ġmetagen omic",
+      "M AL",
+      "Ġa ided",
+      "ang a",
+      "Ġam n",
+      "ĠI CI",
+      "ĠP ML",
+      "Ġdel ivers",
+      "Ġke yp",
+      "Ġbeet les",
+      "Ġoxid ant",
+      "Im mun",
+      "Ġrhyth mic",
+      "fem ale",
+      "J C",
+      "P AD",
+      "gen itor",
+      "A MS",
+      "c atalytic",
+      "ĠM om",
+      "ĠH ert",
+      "ad ish",
+      "Ġcont ention",
+      "Ġy olk",
+      "Ġdem yel",
+      "Ġsuc c",
+      "Ġtravel s",
+      "V e",
+      "ĠF ul",
+      "ĠR if",
+      "Ġint rons",
+      "enc aps",
+      "col our",
+      "Ġhot el",
+      "Ac cess",
+      "ado op",
+      "Ġcoal ition",
+      "ĠMu h",
+      "ĠL TP",
+      "aut om",
+      "ĠL ak",
+      "Ġrem edi",
+      "Ġtra iling",
+      "ins ulator",
+      "ĠRel ig",
+      "ĠHud son",
+      "em ics",
+      "O Ac",
+      "our t",
+      "Ġrel ic",
+      "ĠMi xture",
+      "Ġcalor imeter",
+      "ĠR DF",
+      "ĠHod gkin",
+      "Newton ian",
+      "ĠDelay ed",
+      "ĠNorthe ast",
+      "her ing",
+      "Ġhel ices",
+      "Ġprincip ally",
+      "Ġsusp icion",
+      "Ġextrem ities",
+      "Ġdead line",
+      "ĠEnter ococcus",
+      "m j",
+      "Ġh p",
+      "ĠN AS",
+      "ous s",
+      "Ġintram uscular",
+      "L IN",
+      "Ġch icks",
+      "S core",
+      "Ġf ür",
+      "ĠR SA",
+      "Ġk r",
+      "Ġphot ography",
+      "Ġclear ing",
+      "hol omorphic",
+      "t hem",
+      "Ġp om",
+      "ĠL is",
+      "Ġdisc ard",
+      "Ġgu an",
+      "c x",
+      "ub ov",
+      "ĠCons istency",
+      "Ġple i",
+      "ĠUr inary",
+      "Ġbread th",
+      "E I",
+      "m echan",
+      "Ġd q",
+      "ĠBl ast",
+      "co eff",
+      "IL D",
+      "Ġunem ployment",
+      "A rm",
+      "ĠC n",
+      "mod erate",
+      "Ġagg ress",
+      "Ġcircum f",
+      "l os",
+      "Ġb aro",
+      "velop e",
+      "Ġulcer ative",
+      "Ġhelic ase",
+      "H W",
+      "K G",
+      "r ion",
+      "Ġgen otyped",
+      "Ġar id",
+      "ĠAndre as",
+      "Ġthere of",
+      "ĠOper ating",
+      "ĠNE W",
+      "ĠAntib acterial",
+      "ĠDar win",
+      "Ġrefere e",
+      "Ġd ome",
+      "ag us",
+      "ĠD MD",
+      "AT OR",
+      "Current ly",
+      "ĠInequ alities",
+      "d N",
+      "ol ymer",
+      "em pirical",
+      "ĠBra un",
+      "F IN",
+      "ĠO ber",
+      "pr one",
+      "Ġdimin ish",
+      "ĠGrad uate",
+      "ĠT SH",
+      "ĠH su",
+      "oid osis",
+      "Ġepid ural",
+      "Ġreinfor cing",
+      "Ġthe atre",
+      "Ġv ib",
+      "ĠH ob",
+      "col lection",
+      "MAN GLER",
+      "ĠH ecke",
+      "Ġtr uck",
+      "Ġmotiv ates",
+      "ĠV OC",
+      "Ġun bound",
+      "ram id",
+      "ious ly",
+      "ĠFern ández",
+      "ĠF acial",
+      "ox azol",
+      "Ġtre adm",
+      "ĠRes id",
+      "Lo ader",
+      "ĠRun ning",
+      "otin ib",
+      "P AC",
+      "V II",
+      "i u",
+      "Ġc ite",
+      "ĠH ockey",
+      "ES C",
+      "rho ea",
+      "Ġmac aques",
+      "Ġmedi ast",
+      "at im",
+      "ĠT MP",
+      "ĠA GB",
+      "ĠR up",
+      "ug a",
+      "Ġass urance",
+      "p ay",
+      "en ergies",
+      "ĠK end",
+      "till ery",
+      "Ġanest hetic",
+      "Wind ow",
+      "Ġbe verages",
+      "ag uchi",
+      "ĠFL T",
+      "ĠBound ed",
+      "ĠPolymer ase",
+      "S am",
+      "ĠOr bit",
+      "Ġseason ality",
+      "Ġtachy cardia",
+      "este em",
+      "ĠPerf ect",
+      "S EC",
+      "l ater",
+      "tal e",
+      "ĠForm ally",
+      "L G",
+      "z yn",
+      "Ġmicro algae",
+      "Ġindi um",
+      "erenn ial",
+      "ĠI PT",
+      "Ġk j",
+      "ĠPD A",
+      "Ġassim il",
+      "whe el",
+      "ĠS OS",
+      "ĠP FC",
+      "Ġdec oded",
+      "AT S",
+      "Ġsoci etal",
+      "Ġdiffe omorphisms",
+      "Ġtra verse",
+      "Ġcoll ateral",
+      "g ives",
+      "ĠC EN",
+      "Ġra nd",
+      "Ġher self",
+      "Ġpay ments",
+      "Ġps i",
+      "âIJ £",
+      "ĠGrom ov",
+      "Ġacc idental",
+      "ĠRe ality",
+      "Ġlog istics",
+      "Ġrobust ly",
+      "ĠSar ah",
+      "N U",
+      "d ates",
+      "ĠC UR",
+      "ĠD ream",
+      "Ġdegrad es",
+      "ĠGE O",
+      "Ġbutter fly",
+      "Ġpend ulum",
+      "q a",
+      "Ġas partate",
+      "pseud o",
+      "Ġall osteric",
+      "der r",
+      "ĠQ oL",
+      "Ag ilent",
+      "ĠHard ware",
+      "ĠCum ulative",
+      "Ġp n",
+      "qu antitative",
+      "Ġapp raisal",
+      "Ġpoly acrylamide",
+      "Ġmild ly",
+      "Ġcontrac eptive",
+      "ĠPubl ished",
+      "Ġupl ift",
+      "be h",
+      "Ġadap tor",
+      "ĠEqu al",
+      "thien yl",
+      "at ched",
+      "Ġrep ly",
+      "Ġup wards",
+      "Ġaut opsy",
+      "sim ulation",
+      "Ġgran ite",
+      "Ġpel vis",
+      "Ġhat ching",
+      "ĠS PS",
+      "ĠG EM",
+      "illi ard",
+      "ĠRet rospective",
+      "ĠEarth qu",
+      "ĠInvestig ations",
+      "ĠMer ck",
+      "Ġchol angi",
+      "Ġinfiltr ating",
+      "Ġoverestim ated",
+      "focus ed",
+      "A min",
+      "Ġpre eclampsia",
+      "osp atial",
+      "ĠTRA IL",
+      "P air",
+      "Ġsub marine",
+      "Ġprote olysis",
+      "Ġcomple ments",
+      "ĠKir ch",
+      "Ġcent rom",
+      "Ġn ap",
+      "ĠWe ar",
+      "Ġpun ishment",
+      "Ġautoreg ressive",
+      "Ġcompos er",
+      "ĠEng el",
+      "Ġana emia",
+      "ĠKron ecker",
+      "ĠD id",
+      "ĠCar p",
+      "pe er",
+      "Ġbug s",
+      "ĠIslam ic",
+      "ith romycin",
+      "Ġcons ec",
+      "Ġfam iliarity",
+      "et axel",
+      "Ġint ensively",
+      "ĠU pt",
+      "Ġindic a",
+      "AD A",
+      "ĠChe byshev",
+      "Ġhierarch ies",
+      "Ġworth while",
+      "Ġburn ed",
+      "ĠHM GB",
+      "Ġpolyg onal",
+      "b rile",
+      "Ġz oon",
+      "war ning",
+      "Eukary ota",
+      "d A",
+      "ĠRep eated",
+      "ĠCast ro",
+      "Ġmet ropolitan",
+      "ont inuous",
+      "ĠBar nes",
+      "ĠPost operative",
+      "Ġcyt ology",
+      "Ġspot ted",
+      "vers ity",
+      "aff ine",
+      "sor ted",
+      "ĠPro to",
+      "ĠDes criptive",
+      "Ġhit ting",
+      "Ġanalog ously",
+      "feed back",
+      "Ġspirit ual",
+      "ĠL INE",
+      "ress in",
+      "oph thal",
+      "Ġpoly unsaturated",
+      "Ġpi per",
+      "observ ations",
+      "ĭ ¤",
+      "ir re",
+      "ĠW NT",
+      "Ġund ifferentiated",
+      "eral d",
+      "ĠCT C",
+      "Ġhomomorphism s",
+      "ĠNeon atal",
+      "F in",
+      "ro zen",
+      "ĠL ux",
+      "Ġmod ifier",
+      "ĠK A",
+      "osa ur",
+      "Ġinterven tional",
+      "ĠHa pl",
+      "Ġlumin ance",
+      "Ġun fortunately",
+      "Ġsleep ing",
+      "Ġcit rus",
+      "reson ance",
+      "Ġm oss",
+      "ul ay",
+      "ĠP enn",
+      "ad ministration",
+      "ĠN GF",
+      "Ġsec ured",
+      "ĠA Es",
+      "ĠP WM",
+      "oc co",
+      "ob uf",
+      "Ġphotoc urrent",
+      "ĠScilab Double",
+      "A pril",
+      "Ġfor amin",
+      "Ġpar alysis",
+      "ĠQu ark",
+      "eq ref",
+      "ĠBro oks",
+      "ĠColl ision",
+      "W ar",
+      "Ġ ig",
+      "am ylase",
+      "ist ered",
+      "Ġret raction",
+      "ĠMulti plex",
+      "ĠMa o",
+      "Com mon",
+      "ĠEcon omics",
+      "ĠCriter ion",
+      "ĠC CC",
+      "ĠLe i",
+      "Ġorth orhombic",
+      "Ġaliqu ots",
+      "Ġst ric",
+      "ĠL enn",
+      "Ġdis closure",
+      "amet h",
+      "Ġnormal isation",
+      "Ġphyl ogen",
+      "ĠQTL s",
+      "ĠVers us",
+      "ĠUtil ization",
+      "y ne",
+      "un ted",
+      "ĠD uff",
+      "ĠG J",
+      "Ġoptim ised",
+      "iform is",
+      "ĠIncre ases",
+      "ĠFD G",
+      "ĠBatter y",
+      "P he",
+      "ĠC CS",
+      "Ġch rys",
+      "of en",
+      "Ġmultic omponent",
+      "disc ussed",
+      "bond ing",
+      "ore tically",
+      "ĠAll iance",
+      "Ġhead quarters",
+      "ĠGlas gow",
+      "Ġb out",
+      "Ġe ighth",
+      "Ġinc urred",
+      "ĠBar ry",
+      "Ġquad ric",
+      "Ġdu ties",
+      "Ġmind fulness",
+      "rastruct ural",
+      "T rain",
+      "sh itz",
+      "CD C",
+      "Ġdys lipidemia",
+      "Ġalle ged",
+      "Ġbron ze",
+      "Ġattain ment",
+      "Q D",
+      "rom bin",
+      "Ġap olipoprotein",
+      "own ed",
+      "Ġge ographically",
+      "work ing",
+      "ĠBl ind",
+      "Ġdon ation",
+      "ĠSer ge",
+      "Ġspread s",
+      "ĠHeter ogeneity",
+      "ĠFr é",
+      "Ġdef er",
+      "Ġlif ts",
+      "EGF R",
+      "ĠPort land",
+      "Ġbrother s",
+      "ĠTrypan osoma",
+      "in ian",
+      "Ġp ressed",
+      "Ġtrans duced",
+      "Ġpol yn",
+      "Ġlist eners",
+      "bo ards",
+      "ĠSustain able",
+      "al an",
+      "ĠS ullivan",
+      "Assum ption",
+      "oft en",
+      "j p",
+      "or ative",
+      "pl ers",
+      "Ġmodular ity",
+      "ĠHerm ite",
+      "Ġhydroxy apatite",
+      "ĠHir sch",
+      "D eterm",
+      "f acing",
+      "ir radiated",
+      "Ġhar sh",
+      "Ġtoler ate",
+      "ĠT rap",
+      "ĠA ware",
+      "ot ax",
+      "AT ING",
+      "Ġhist opathology",
+      "ĠIsra eli",
+      "clock wise",
+      "z ig",
+      "ĠJ C",
+      "ĠQu ick",
+      "ĠSL AM",
+      "Ġf ox",
+      "ĠR av",
+      "gener ating",
+      "Ġhemat oxylin",
+      "yl transferase",
+      "Ġcorrobor ated",
+      "F DR",
+      "o ard",
+      "Ġequ id",
+      "ĠÂ »",
+      "Ġneuro psychological",
+      "Ġbreak up",
+      "Ġemphas izing",
+      "Ġemiss ivity",
+      "block ing",
+      "Ġpar all",
+      "Ġtil ting",
+      "Ġp eng",
+      "ĠSc an",
+      "Ġion osphere",
+      "Ġm ount",
+      "fore st",
+      "Ġcall us",
+      "α β",
+      "ĠChrist mas",
+      "ĠMag azine",
+      "eval uate",
+      "ĠP ag",
+      "ĠBe at",
+      "Ġaccum ulates",
+      "Ġcrow ding",
+      "unn eling",
+      "Ġ Ñ",
+      "ĠA CP",
+      "ge ometry",
+      "MP T",
+      "Ġpharmac ists",
+      "Ġpull back",
+      "Ġduc tility",
+      "S upervised",
+      "Ġlymph oblastic",
+      "pe a",
+      "typ ical",
+      "bro ken",
+      "F c",
+      "Ġl ining",
+      "ĠD um",
+      "Ġmulti ples",
+      "ó w",
+      "Ġmer its",
+      "Ġextinc t",
+      "ĠNur sing",
+      "ĠExplo iting",
+      "ĠBhatt ach",
+      "J uly",
+      "t ze",
+      "th romb",
+      "te enth",
+      "Ġtoxic ities",
+      "Ġdenit rification",
+      "Ġex position",
+      "Ġim perf",
+      "Ġsur name",
+      "po inter",
+      "ĠEr n",
+      "ĠAbund ance",
+      "ĠD unn",
+      "oph ora",
+      "Ġtool kit",
+      "Lo ad",
+      "ĠDeriv ation",
+      "c ould",
+      "ĠC aspase",
+      "ĠSp rague",
+      "ĠTr p",
+      "Ġbright est",
+      "ill ard",
+      "Ġinter disciplinary",
+      "Ġqu arant",
+      "Ġhyper surfaces",
+      "eli ac",
+      "ĠAL MA",
+      "Ġacryl ic",
+      "Ġgent le",
+      "De ep",
+      "ĠPand emic",
+      "Ġinf easible",
+      "Ġradi ol",
+      "AB P",
+      "Ġmes enteric",
+      "ylind er",
+      "pack ed",
+      "Ġsomat osensory",
+      "Ġp ave",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠ",
+      "Ġpharmac ology",
+      "Ġtan h",
+      "ĠMt b",
+      "Ġchim pan",
+      "Ġautophag ic",
+      "Ġwithd rawn",
+      "ĠM CC",
+      "Z F",
+      "ĠS pl",
+      "ĠL au",
+      "Ġbi ologic",
+      "elect rons",
+      "Ġunderestim ation",
+      "Ġcharacter ise",
+      "circ ular",
+      "ĠTHE ORY",
+      "B rown",
+      "F BS",
+      "J o",
+      "d G",
+      "m ars",
+      "ar ticular",
+      "ĠP ren",
+      "ĠM SA",
+      "ĠIt em",
+      "Ġsem idefinite",
+      "ĠGib son",
+      "Ġtour ism",
+      "ĠK ok",
+      "Ġexpos ing",
+      "Ġintra venously",
+      "dri ver",
+      "ĠFort unately",
+      "ĠS ach",
+      "Ġcont aminant",
+      "Ġab rog",
+      "ĠEm otional",
+      "VAL UE",
+      "dispers ion",
+      "Jac obi",
+      "ĠImper ial",
+      "I on",
+      "L in",
+      "f idelity",
+      "ĠB irds",
+      "ĠCon current",
+      "mat ism",
+      "co al",
+      "Ġt q",
+      "ĠMn O",
+      "Ġfoss ils",
+      "Ġt ender",
+      "Ġr hesus",
+      "Ġblo om",
+      "ab dominal",
+      "Ġscal p",
+      "Ġhome ostatic",
+      "ĠH unt",
+      "ĠPharmac okine",
+      "b rown",
+      "ĠH YP",
+      "Ġdiss ociated",
+      "ĠSoc cer",
+      "ĠInequ ality",
+      "m aker",
+      "Ġsh ade",
+      "ĠZ ur",
+      "obs ervation",
+      "al tered",
+      "U U",
+      "Ġthe or",
+      "ep it",
+      "Ġphyl um",
+      "Ġvig orous",
+      "ĠA CM",
+      "Ġmeth otrexate",
+      "dem ographic",
+      "Ġsing ly",
+      "ĠPhys iology",
+      "Ġremod elling",
+      "ĠK rist",
+      "rop ies",
+      "flow s",
+      "hard ness",
+      "igh teen",
+      "bre ve",
+      "ĠRet inal",
+      "Ġscin till",
+      "Ġutter ance",
+      "Ġmonolith ic",
+      "ĠVl ad",
+      "ĠL MC",
+      "ip t",
+      "arrow s",
+      "ĠPubl ishing",
+      "ĠStrept omyces",
+      "f al",
+      "Ġtroposp here",
+      "B en",
+      "c andid",
+      "ĠS ic",
+      "tim ore",
+      "L en",
+      "in en",
+      "amp ered",
+      "ĠMon th",
+      "Ġopp onent",
+      "Aug ust",
+      "Ġst aggered",
+      "cent re",
+      "exp ect",
+      "Ġred dening",
+      "ĠT l",
+      "hib ition",
+      "Ġmicro particles",
+      "ĠInt rac",
+      "ĠInitial ize",
+      "Ġdict ated",
+      "D ig",
+      "ä º",
+      "he aling",
+      "Ġd V",
+      "Ġappe tite",
+      "Ġunus ually",
+      "ĠAstr onomy",
+      "Ġw are",
+      "Ġover coming",
+      "Ġcoll iders",
+      "ĠUS ING",
+      "ocardi tis",
+      "P ick",
+      "Ġd ub",
+      "ĠJ ason",
+      "ĠEd itor",
+      "ê ³",
+      "Ġl ags",
+      "Ġcl s",
+      "Ġsur gically",
+      "ĠPV C",
+      "par ticularly",
+      "Ġred ist",
+      "Ġlog ics",
+      "sk ii",
+      "ĠD VD",
+      "Ġcomp ly",
+      "az i",
+      "ĠInter acts",
+      "bo olean",
+      "ĠER P",
+      "ĠEr r",
+      "otrans piration",
+      "ĠPé rez",
+      "A sp",
+      "am iliar",
+      "ĠF etal",
+      "Ġdecl aration",
+      "k inson",
+      "t ube",
+      "Ġphysiological ly",
+      "c ue",
+      "ĠE ri",
+      "Ġen vision",
+      "ex ternal",
+      "inter mediate",
+      "Ġshop ping",
+      "ĠF ras",
+      "ĠH aj",
+      "ĠAl ger",
+      "Ġanthrop ometric",
+      "Ġcancell ed",
+      "H PV",
+      "k ers",
+      "af a",
+      "Ġvulner abilities",
+      "electro lyte",
+      "ĠGonz alez",
+      "íķ ĺ",
+      "q v",
+      "Ġde af",
+      "Ġbut yrate",
+      "ĠCo efficient",
+      "Ġstar burst",
+      "Ġpolym orph",
+      "ĠE RA",
+      "ĠMax imal",
+      "ĠMu eller",
+      "Ġabsor bers",
+      "Ġa rab",
+      "re tions",
+      "Ġne bula",
+      "Ġmin es",
+      "е н",
+      "%%%%%%%% %%%%%%%%",
+      "Ġband pass",
+      "Ġpoly urethane",
+      "Re LU",
+      "ĠFer ro",
+      "pic illin",
+      "C AD",
+      "T y",
+      "ĠP CD",
+      "ĠB AC",
+      "Ġplankton ic",
+      "F er",
+      "Ġc ricket",
+      "Ġman ure",
+      "oun s",
+      "âĪ §",
+      "Ġtor ques",
+      "m itian",
+      "Ġt ion",
+      "ĠG arden",
+      "Ġfol k",
+      "Ġsusp icious",
+      "Ã Ĥ",
+      "od ia",
+      "ist encies",
+      "ãĢ ī",
+      "ĠInv itrogen",
+      "ĠS UN",
+      "ĠSuper ior",
+      "Ġdiscontinu ation",
+      "c ock",
+      "k not",
+      "Ġext ens",
+      "ĠWh itney",
+      "Ġhar bour",
+      "P ID",
+      "Ġp mol",
+      "ol ymph",
+      "Ġg ard",
+      "ĠO varian",
+      "Ġrep ressed",
+      "ĠAl ab",
+      "ĠÃ Ħ",
+      "ule x",
+      "ĠAust rian",
+      "Ġa flat",
+      "Ġpar athyroid",
+      "Ġgroup oid",
+      "Ġdev ast",
+      "ĠK v",
+      "Ġbor row",
+      "Ġuncon ventional",
+      "Ġbore hole",
+      "Ñ Į",
+      "ĠD ays",
+      "Ġlex ic",
+      "N or",
+      "ĠH erc",
+      "ass ays",
+      "Ġdraw ings",
+      "def in",
+      "ev oked",
+      "ĠÈ ³",
+      "ĠSund ay",
+      "ĠC hes",
+      "cons idered",
+      "oped ic",
+      "larg er",
+      "om inant",
+      "ĠB omb",
+      "Ġf iss",
+      "Ġh inge",
+      "ĠI onic",
+      "Ġdest ro",
+      "Ġcomplement arity",
+      "Hig gs",
+      "or ia",
+      "our cing",
+      "ĠX in",
+      "Ġwork space",
+      "ĠLig and",
+      "Ġstrugg le",
+      "ĠImmunohist ochemical",
+      "Ġn ick",
+      "ĠGu ard",
+      "rig id",
+      "Ġaqu aculture",
+      "Experim ent",
+      "Ë Ī",
+      "ti r",
+      "ĠS MS",
+      "Ġbe vacizumab",
+      "Ġmod ulations",
+      "Ġge ophysical",
+      "Pro perties",
+      "Ġpain ted",
+      "Ġs anc",
+      "Ġin timate",
+      "Ġn ail",
+      "id entity",
+      "Ġdat um",
+      "anth us",
+      "Ġdy adic",
+      "Ġconvinc ing",
+      "e lem",
+      "Ġh iding",
+      "Ġr ugby",
+      "ĠX e",
+      "ĠIs sue",
+      "Ġves icular",
+      "ĠKel vin",
+      "Ġdist ancing",
+      "echn ology",
+      "af ers",
+      "ĠAut hentic",
+      "Pub Med",
+      "Ġdeform ity",
+      "ĠCha os",
+      "ĠSh ield",
+      "ox etine",
+      "ĠWork ers",
+      "ĠMO I",
+      "Ġdehyd rated",
+      "ĠGast ric",
+      "Ġmonomial s",
+      "od ox",
+      "ĠD ublin",
+      "Ġle ishman",
+      "Ġpl anner",
+      "circ le",
+      "Ġfract ured",
+      "ĠLoc ally",
+      "ĠAc tions",
+      "Ġlic hen",
+      "h annel",
+      "ĠT AG",
+      "Ġdec isive",
+      "ĠQ M",
+      "Ġbiom aterials",
+      "ĠVirus es",
+      "hydrox yphenyl",
+      "ĠI AA",
+      "ĠR U",
+      "vi olating",
+      "Ġp ockets",
+      "ch ant",
+      "ib erg",
+      "lect omy",
+      "oler ae",
+      "Ġattract ing",
+      "Ġket one",
+      "ĠC od",
+      "Ġmicro arrays",
+      "ĠMet als",
+      "benz oyl",
+      "Ġsemigroup s",
+      "Ġreconstit uted",
+      "s ites",
+      "an abe",
+      "ĠCom posites",
+      "Ġwild type",
+      "Ġleuk aemia",
+      "Ġmur der",
+      "Ġdent in",
+      "H ub",
+      "O rient",
+      "on n",
+      "syn chron",
+      "Ġchron ically",
+      "methylene amino",
+      "Ġdop ant",
+      "Ġf ecundity",
+      "de lete",
+      "rem ia",
+      "ĠNH L",
+      "iti dis",
+      "Ġcop ep",
+      "X I",
+      "Ġloc ating",
+      "ĠZ IKV",
+      "hex a",
+      "ĠFactor ization",
+      "ynch us",
+      "M ethyl",
+      "h agen",
+      "ĠP aw",
+      "ne ath",
+      "bs ite",
+      "Ġtrac he",
+      "B re",
+      "u w",
+      "ro it",
+      "Ġre acting",
+      "ĠB ae",
+      "Ġquoti ents",
+      "Ġp ins",
+      "ĠV ARI",
+      "Ġequ ine",
+      "ĠRun ge",
+      "Ġcolon ial",
+      "measure ment",
+      "ĠAbb ott",
+      "Ġorth o",
+      "Ġmeta phor",
+      "benz oic",
+      "ĠTransform ers",
+      "L ower",
+      "ĠO VA",
+      "radi al",
+      "Fl ag",
+      "author bs",
+      "Ġtreadm ill",
+      "Ġenter ica",
+      "ĠJul ia",
+      "Ġpl umes",
+      "Ġinv oke",
+      "chlor ic",
+      "ol ino",
+      "Ġinter ruption",
+      "sub unit",
+      "ĠMD P",
+      "Ġmanip ulator",
+      "ĠScal es",
+      "ĠHT ML",
+      "ĠFreder ick",
+      "G arc",
+      "Ġb ell",
+      "ĠR ect",
+      "rom ised",
+      "W ord",
+      "o ples",
+      "oper ated",
+      "Ġcollec ts",
+      "ĠHor izon",
+      "Ġsa fer",
+      "d up",
+      "ĠM ills",
+      "AL P",
+      "Ġex opl",
+      "AT TR",
+      "war a",
+      "ĉĉĉĉ ĉĉĉ",
+      "Ġdeb ug",
+      "Des criptor",
+      "stat istics",
+      "ĠC ub",
+      "ST ER",
+      "ĠSt abilization",
+      "ĠIR AS",
+      "Ġconform ally",
+      "Ad ap",
+      "Â Ń",
+      "ĠQ S",
+      "Ġmicro strip",
+      "Ġdel icate",
+      "Ġpubl isher",
+      "Ġh os",
+      "ĠS v",
+      "ĠDes ert",
+      "ĠGu er",
+      "ĠCap ture",
+      "E BP",
+      "d ust",
+      "å ¤",
+      "ĠO ls",
+      "Ġsuper script",
+      "ĠFl uctuations",
+      "ill ium",
+      "Ġcap tion",
+      "Ġconc ur",
+      "Ġquantif ies",
+      "ster dam",
+      "Ġspik ed",
+      "N an",
+      "us in",
+      "ĠL AN",
+      "Ġobserv es",
+      "ĠAl a",
+      "ĠInt uitively",
+      "cur r",
+      "Ġshr inking",
+      "Ġcompress ibility",
+      "orp oreal",
+      "Ġdeb t",
+      "ç Ķ",
+      "ĠT il",
+      "ĠW AT",
+      "ody ne",
+      "Ġgate way",
+      "Ġduc tile",
+      "ĠJes us",
+      "os itol",
+      "ĠM ales",
+      "Ġsol vation",
+      "Ġdisag ree",
+      "Ġortholog s",
+      "S an",
+      "ig o",
+      "Ġph ages",
+      "Ġneg atives",
+      "Ġinterp re",
+      "AA A",
+      "Ġgrating s",
+      "ĠM oll",
+      "ĠR ivers",
+      "Ġcr uzi",
+      "ĠGen erate",
+      "ĠBar bara",
+      "ĠHer itage",
+      "ĠFlu orescent",
+      "ĠLaw s",
+      "Array Expr",
+      "Ġmultip ole",
+      "Ġsquee zing",
+      "S PSS",
+      "l f",
+      "n lm",
+      "Ġw orn",
+      "ĠK uz",
+      "Ġgenes is",
+      "ĠEm peror",
+      "vol atile",
+      "Ġsib ling",
+      "iv ir",
+      "o en",
+      "Ġprot ost",
+      "Ġtransform ers",
+      "enn ium",
+      "Ġpropos ing",
+      "Ġbroadcast ing",
+      "Q M",
+      "ĠD ependent",
+      "Ġdis able",
+      "ĠU AS",
+      "Ġwar nings",
+      "Ġarm ed",
+      "Ġjournal ist",
+      "Ġmonoc linic",
+      "ol ium",
+      "ap ing",
+      "to on",
+      "Ġorth odontic",
+      "ĠNormal ization",
+      "Ġmand ible",
+      "ab an",
+      "ĠW ak",
+      "ext end",
+      "Multi ple",
+      "in vestig",
+      "is cal",
+      "ut tered",
+      "Ġbur g",
+      "dec ode",
+      "em por",
+      "ĠD uration",
+      "ann y",
+      "opro st",
+      "ĠRen ormalization",
+      "ĠF UNCTION",
+      "yt orch",
+      "Ġsyn apt",
+      "ĠForm at",
+      "ĠCR T",
+      "ĠJon athan",
+      "ĠOF F",
+      "or r",
+      "Ġres ur",
+      "Ġcor ruption",
+      "d welling",
+      "Ġback up",
+      "AG T",
+      "ĠSaf e",
+      "dor fer",
+      "Ġatax ia",
+      "Ġpar v",
+      "read er",
+      "Ġsubt ract",
+      "embol ism",
+      "Ġt innitus",
+      "Ġcyt omegalovirus",
+      "Ġdele ting",
+      "T ex",
+      "ĠC SS",
+      "ard t",
+      "Ġout growth",
+      "Ġmy ocytes",
+      "dig ital",
+      "Ġsub scale",
+      "usp ension",
+      "Ġham ster",
+      "Ġinflat on",
+      "h ara",
+      "ur ches",
+      "ĠC LE",
+      "ĠY as",
+      "ĠEn coding",
+      "ĠAug er",
+      "Ġanastom osis",
+      "A gent",
+      "ĠS IL",
+      "ĠC CT",
+      "Ġbr ine",
+      "Ġolig o",
+      "Ġfluor o",
+      "Ġgall ery",
+      "d dots",
+      "Ġc ilia",
+      "ĠP PV",
+      "ĠU TR",
+      "Ġinter tidal",
+      "ocal ized",
+      "Ġcrow ds",
+      "od or",
+      "Ġco v",
+      "Ġnon etheless",
+      "Ġïģ ¤",
+      "Ġboost ed",
+      "ĠChak ra",
+      "H al",
+      "P ear",
+      "Ġimp rec",
+      "ĠSup plement",
+      "go al",
+      "Ġôı¼ ģ",
+      "Ġst all",
+      "Ġher d",
+      "small er",
+      "Ġreconstruct ing",
+      "Ġarte facts",
+      "Ġt eg",
+      "con ventional",
+      "rad ical",
+      "Ġliter al",
+      "frame work",
+      "ipro cal",
+      "E EG",
+      "Ġg ins",
+      "od ermal",
+      "ĠAg u",
+      "ĠTw elve",
+      "M ul",
+      "Ø ¨",
+      "ir l",
+      "ĠB elief",
+      "Ġinc ont",
+      "IC C",
+      "hex ane",
+      "Ġe jected",
+      "ĠP SC",
+      "ĠH PC",
+      "ĠV H",
+      "Ġequival ences",
+      "plot lib",
+      "en ital",
+      "ri ans",
+      "pro v",
+      "ĠV ibr",
+      "Ġgram matical",
+      "bach ia",
+      "accept able",
+      "od icity",
+      "ab b",
+      "Ġher bs",
+      "Ġpredom inance",
+      "ĠOrient ation",
+      "Ġinver tebrate",
+      "Ġpel agic",
+      "count ry",
+      "ĠOrig ins",
+      "ĠAdoles cents",
+      "ĠT uning",
+      "rain ian",
+      "ĠSc ar",
+      "Ġlight est",
+      "Ġemit ters",
+      "ĠTs ai",
+      "ri tical",
+      "ĠEx pert",
+      "aut hors",
+      "E CTION",
+      "ĠSever ity",
+      "N am",
+      "p ubl",
+      "ĠA be",
+      "Ġnanoc rystalline",
+      "ĠNak amura",
+      "ĠP ec",
+      "ĠB ug",
+      "Ġsens ed",
+      "ON S",
+      "IC s",
+      "Ġelectro chem",
+      "ĠR OM",
+      "ĠRec ruitment",
+      "Ġ⣠©",
+      "Ġbiomo lecules",
+      "ĠB rac",
+      "Ġtrans position",
+      "ĠW P",
+      "ĠO mega",
+      "Ġdiag on",
+      "plate let",
+      "J M",
+      "ac re",
+      "ĠA SR",
+      "ĠK ath",
+      "Ġpri v",
+      "opl asts",
+      "S amples",
+      "d F",
+      "at ti",
+      "ĠS anger",
+      "ip itated",
+      "Ġric her",
+      "ĠG RA",
+      "Ġplant ar",
+      "Ġfo ams",
+      "Ġmathem atic",
+      "Ġsta phyl",
+      "ĠUpt ake",
+      "Ġc ant",
+      "ĠS Z",
+      "Ġdis miss",
+      "Ġselec tions",
+      "plit z",
+      "Ġexempl ified",
+      "Ġtors ional",
+      "E v",
+      "Ġv oters",
+      "ĠN est",
+      "ys cale",
+      "Ġspec i",
+      "Ġpol ished",
+      "Ġlat encies",
+      "q ing",
+      "Ġon wards",
+      "ll vm",
+      "the orem",
+      "log ging",
+      "ĠAL K",
+      "ĠBa um",
+      "ĠGh osh",
+      "Ġchair man",
+      "p aired",
+      "ĠP AP",
+      "not es",
+      "olester olem",
+      "Ġestu arine",
+      "ĠTibet an",
+      "ĠV ER",
+      "Ġcheck er",
+      "FLAG S",
+      "rol imus",
+      "ĠMut ant",
+      "Ġspray ing",
+      "ĠC hest",
+      "olin ium",
+      "ĠTri assic",
+      "Ġlid ar",
+      "A rt",
+      "ĠM ilk",
+      "Ġind ecomposable",
+      "Ġrock et",
+      "ĠPart ners",
+      "Ġseman tically",
+      "entin el",
+      "L arge",
+      "P en",
+      "ĠT ru",
+      "Ġher itage",
+      "ĠMut ual",
+      "ĠChem otherapy",
+      "Ġdoub les",
+      "ĠEmbed ded",
+      "it ual",
+      "ĠB PA",
+      "Ġch olerae",
+      "ĠIn side",
+      "ĠK atz",
+      "con vergence",
+      "Ġindividual ized",
+      "kin je",
+      "Ġdiscover ing",
+      "Ġintric ate",
+      "Ġin land",
+      "RE CT",
+      "ĠCh ick",
+      "ĠSU R",
+      "Ġye asts",
+      "l uminosity",
+      "Ġf ain",
+      "ion i",
+      "ĠT ig",
+      "ound er",
+      "Ġdel iber",
+      "ĠCons ervative",
+      "ĠDel hi",
+      "B ER",
+      "ĠY B",
+      "ole y",
+      "ĠBe au",
+      "TE XT",
+      "Ġsquee zed",
+      "Ġs ocket",
+      "Ġp T",
+      "py razol",
+      "co efficients",
+      "Ġrecruit ing",
+      "Ġduc ts",
+      "Ġf oster",
+      "om eration",
+      "ĠP SI",
+      "ĠD up",
+      "Ġk s",
+      "ĠOp tics",
+      "Ġliter ary",
+      "ĠNi O",
+      "ĠVEGF R",
+      "Ġgravit on",
+      "Ġutter ances",
+      "Ġb rady",
+      "Ġfor ty",
+      "ĠTrans plantation",
+      "Ġagre ements",
+      "Left rightarrow",
+      "w aves",
+      "Ġacid osis",
+      "Ġwood en",
+      "ĠCytoplas mic",
+      "s afe",
+      "Ġj umping",
+      "enn ial",
+      "Vari ous",
+      "ĠEry th",
+      "ul ins",
+      "un lock",
+      "methyl ated",
+      "asser stein",
+      "Ġheterozyg osity",
+      "oxy cycl",
+      "Ġcre ativity",
+      "MP LE",
+      "in ative",
+      "Ġcon volutions",
+      "Ġno uns",
+      "eg an",
+      "ĠAb raham",
+      "Ġdens er",
+      "C he",
+      "l c",
+      "ĉĉ ĉĠ",
+      "Ġsem im",
+      "ĠOut er",
+      "Ġc and",
+      "od ule",
+      "est hesia",
+      "ĠJ oy",
+      "ĠProt ocols",
+      "ĠCalc ulated",
+      "at op",
+      "ĠF ALSE",
+      "Ġref in",
+      "Ġmig rants",
+      "ĠïĤ ´",
+      "ĠSpecific ity",
+      "ĠFellow ship",
+      "ĠP MT",
+      "Ġdis close",
+      "unc hes",
+      "Ġdi atoms",
+      "cor r",
+      "Ġsky rm",
+      "Ġrenew al",
+      "g cd",
+      "ce reb",
+      "Ġup right",
+      "Ġmes oscopic",
+      "hyd raz",
+      "B AS",
+      "F LO",
+      "H CC",
+      "M ouse",
+      "Ġpos et",
+      "Ġprotein uria",
+      "Ġre app",
+      "ĠN ickel",
+      "Ġstrip es",
+      "Ġrip ple",
+      "Sep tember",
+      "od omain",
+      "ĠP ope",
+      "ĠN ons",
+      "Ġtechn ic",
+      "Ġneut rop",
+      "des criptor",
+      "Ġdissip ated",
+      "Ġglac iers",
+      "ĠH IGH",
+      "ĠL av",
+      "ret ely",
+      "Ġback wards",
+      "Ġcri tics",
+      "ĠExt ending",
+      "b ic",
+      "ĠCh ao",
+      "of ibr",
+      "Ġcoun ters",
+      "Ġstre ets",
+      "Ġprost hetic",
+      "Ġbiod egradation",
+      "complex ity",
+      "ĠS PL",
+      "ĠC AC",
+      "Ġad ducts",
+      "Ġmorph ometric",
+      "ĠMat t",
+      "Ġinduc er",
+      "Ġast rocyte",
+      "Ġtriple ts",
+      "Ġpert ussis",
+      "P ES",
+      "id y",
+      "unc ertain",
+      "Ġhyper parameter",
+      "ĠInf rastructure",
+      "ìĿ ĺ",
+      "Z W",
+      "Ġadd r",
+      "Ġdisrup ts",
+      "Ġove restimate",
+      "ĠDY NA",
+      "Ġvolat iles",
+      "em erg",
+      "iss ue",
+      "c pp",
+      "Ä ħ",
+      "ĠV IP",
+      "Ġu ve",
+      "ĠCN V",
+      "yleth yl",
+      "on azole",
+      "ĠH iro",
+      "Ġc n",
+      "ti k",
+      "ub ted",
+      "ĠJac obs",
+      "Ġadvoc ated",
+      "ĠBif id",
+      "m aterial",
+      "Ġst yrene",
+      "ĠK eller",
+      "rocy tic",
+      "pine phrine",
+      "ĠWr itten",
+      "ĠRecommend ation",
+      "b led",
+      "ĠB ootstrap",
+      "th irds",
+      "Ġcap tain",
+      "equ als",
+      "SR C",
+      "ĠKent ucky",
+      "Ġeosinophil s",
+      "A verage",
+      "H i",
+      "W he",
+      "ĠD AT",
+      "ĠU M",
+      "Ġtend encies",
+      "ĠPet erson",
+      "Ġocc ult",
+      "Ġexhib ition",
+      "ĠIN S",
+      "Ġadip ocyte",
+      "J ust",
+      "h ift",
+      "t ensors",
+      "Ġc iliary",
+      "ip ation",
+      "Ġmotiv ations",
+      "Ġwitness ed",
+      "it ches",
+      "ĠS oy",
+      "Ġg ib",
+      "ep tic",
+      "ĠK OH",
+      "Ġïģ ¨",
+      "ĠTor res",
+      "Î ¿",
+      "ar po",
+      "ok inase",
+      "ĠBud d",
+      "ĠG MM",
+      "Ġunder pin",
+      "Ġoptim istic",
+      "oge ography",
+      "numer ical",
+      "og g",
+      "Ġdise quilibrium",
+      "Ġsw ab",
+      "ED S",
+      "ĠPD Fs",
+      "ĠSuper nova",
+      "phosph o",
+      "Ġlys osomes",
+      "gal actic",
+      "ĠPerm e",
+      "Ġfisher y",
+      "ĠB OLD",
+      "Ġun ravel",
+      "ĠEncryp tion",
+      "J P",
+      "h ur",
+      "Ġdisc ount",
+      "ĠWat anabe",
+      "ĠRhe umat",
+      "F ITC",
+      "Ġt erahertz",
+      "ĠF ont",
+      "ian ces",
+      "ĠAd ditive",
+      "ĠE ither",
+      "met adata",
+      "amp hetamine",
+      "ĠPal mer",
+      "Ġlever aging",
+      "J ohn",
+      "O CT",
+      "in fer",
+      "ĠM SD",
+      "ĠâĪ ĵ",
+      "ou ver",
+      "ĠAnd ersen",
+      "Ġworld s",
+      "Ġtor i",
+      "Ġïģ °",
+      "engine ering",
+      "ĠSquad ron",
+      "A ff",
+      "å ı",
+      "ox el",
+      "yle tic",
+      "ĠCharacter izing",
+      "V T",
+      "r ational",
+      "ere mia",
+      "Ġcomplex ation",
+      "ĠER α",
+      "carboxyl ic",
+      "ïĤ ·",
+      "Ġgalact ose",
+      "ĠAur ora",
+      "Ġplasmin ogen",
+      "ure n",
+      "ign e",
+      "Ġrep aired",
+      "Ġblock ers",
+      "ĠMN IST",
+      "Ï ħ",
+      "ĠA xi",
+      "Ġst adium",
+      "di ethyl",
+      "âĢ İ",
+      "Ġcycl otron",
+      "Ġlymph aden",
+      "Ġv in",
+      "ĠM ayer",
+      "Ġendomet rium",
+      "ĠSp herical",
+      "Ġpers u",
+      "Ġimm ortal",
+      "benz enesulf",
+      "ĠÅ ľ",
+      "Ġb ite",
+      "ugg ed",
+      "ĠDiff raction",
+      "GT G",
+      "i ate",
+      "Ġt p",
+      "Ġab er",
+      "ĠRe in",
+      "Pro gram",
+      "St yle",
+      "ĠRegular ization",
+      "ĠLeuk emia",
+      "Ġprokary otic",
+      "oc omial",
+      "sk b",
+      "Ġdevi ates",
+      "Ġf use",
+      "ĠN ull",
+      "Ġïĥ Ĺ",
+      "ĠOper ational",
+      "Ġcompress or",
+      "ĠRyd berg",
+      "Ġf ought",
+      "Ġe co",
+      "ĠS SP",
+      "CD s",
+      "ĠME K",
+      "ĠAnis otropic",
+      "ĠDi rection",
+      "ĠSpect rometry",
+      "Ġglut en",
+      "ĠPow ell",
+      "recogn ized",
+      "Ġpsych otic",
+      "Ġhind er",
+      "Ġaccommod ation",
+      "ĠNorm an",
+      "Q x",
+      "Ġper iv",
+      "ĠUn known",
+      "Ġjo ins",
+      "ĠMinim ization",
+      "ĠS ons",
+      "ĠC in",
+      "Ġun avoid",
+      "ĠPT X",
+      "Ġc ada",
+      "ĠL uk",
+      "Ġr uling",
+      "Ġbi phasic",
+      "ĠCom plications",
+      "ĠDef ects",
+      "Cont ent",
+      "ĠGreg ory",
+      "ĠWer ner",
+      "ĠWeib ull",
+      "eld om",
+      "Ġactiv ators",
+      "GL API",
+      "math ring",
+      "Ġhe ns",
+      "N SC",
+      "h owever",
+      "ĠT ME",
+      "ma frost",
+      "co efficient",
+      "ĠIns ect",
+      "ĠRO Is",
+      "ĠBor rel",
+      "ĠQi u",
+      "Ġinhal ed",
+      "id ate",
+      "Ġanti hypertensive",
+      "Ġtreat s",
+      "ĠNear ly",
+      "suc c",
+      "ĠOrb ital",
+      "er adish",
+      "ad ministered",
+      "ĠÏ Ĥ",
+      "ĠCol ony",
+      "ĠâĮ Ĭ",
+      "ĠIndones ian",
+      "ĠB auer",
+      "ĠK od",
+      "mann ed",
+      "Res istant",
+      "Ġda ughters",
+      "ĠPredic ted",
+      "Ġvoc ab",
+      "Ġcontras ted",
+      "m argin",
+      "ĠDi rected",
+      "ED TA",
+      "Ġsynchron y",
+      "ick i",
+      "ĠSal v",
+      "t reat",
+      "in cess",
+      "var nothing",
+      "Ġhex ane",
+      "Em pty",
+      "Ġgem citabine",
+      "om ib",
+      "ore pinephrine",
+      "pro c",
+      "ĠMet S",
+      "ĠDR AM",
+      "Ġantico agulant",
+      "n om",
+      "am ater",
+      "ĠLi DAR",
+      "Ġmob il",
+      "Ġamelior ates",
+      "n iz",
+      "Ġj a",
+      "Ġem uls",
+      "ĠZ a",
+      "Ġastr onomical",
+      "ĠAlf red",
+      "H ilbert",
+      "ĠK F",
+      "CR T",
+      "quad ratic",
+      "Ġdifferenti als",
+      "rob acterium",
+      "ĠHippocamp al",
+      "p ull",
+      "Ä Ļ",
+      "Ġs ad",
+      "ally l",
+      "Ġhot spot",
+      "ĠElectron ics",
+      "Ġconstit ution",
+      "iton in",
+      "ا ÙĦ",
+      "P c",
+      "Ġre vascular",
+      "Ġus able",
+      "ĠSc atter",
+      "Ġgraph ically",
+      "lim inf",
+      "Ġrestaur ant",
+      "ucalypt us",
+      "AC G",
+      "Anal y",
+      "ĠMill ipore",
+      "Ġmunicip alities",
+      "ĠMacroph age",
+      "Ġmacrom olecular",
+      "L icense",
+      "g c",
+      "Ġl avage",
+      "ĠA ES",
+      "ĠF CS",
+      "per itone",
+      "Ġmeas les",
+      "TE X",
+      "ĠVir ulence",
+      "Ġhemat oma",
+      "ĠF res",
+      "ĠN utrient",
+      "ap ar",
+      "ĠSp ot",
+      "co plasma",
+      "ĠExp ect",
+      "Ġc iprofloxacin",
+      "phyl axis",
+      "ĠAtl anta",
+      "ro uting",
+      "ar ate",
+      "ĠC is",
+      "ens ure",
+      "car riers",
+      "ĠVari ant",
+      "sur gical",
+      "ĠEstim ate",
+      "à ¹",
+      "ĠL iqu",
+      "Ġam alg",
+      "Ġbl a",
+      "Ġthem atic",
+      "IR Q",
+      "ACT ION",
+      "ĠChris ti",
+      "æ ľ",
+      "Ġn py",
+      "de ath",
+      "Ġhair pin",
+      "Ġmultiplic ities",
+      "Gib co",
+      "he ated",
+      "af ety",
+      "mut able",
+      "quark s",
+      "S un",
+      "q l",
+      "Ġproduc tions",
+      "Ġge ology",
+      "Ġt ides",
+      "at rix",
+      "Ġad mixture",
+      "trans lated",
+      "ĠAb u",
+      "nucle us",
+      "Ġweakness es",
+      "Ġflav ors",
+      "ĠLu is",
+      "ĠPut ative",
+      "sent ence",
+      "ĠM ast",
+      "ĠM PS",
+      "ĠE SS",
+      "Ġcomp ose",
+      "Ġbi refring",
+      "ĠRam sey",
+      "ĠCL L",
+      "Ġlign ocell",
+      "ĠL amin",
+      "ĠW elsh",
+      "v on",
+      "Ġp ests",
+      "Ġf iction",
+      "ĠH RT",
+      "Ġass ure",
+      "CT s",
+      "ĠPA Hs",
+      "Ġcrypt ography",
+      "en erated",
+      "Ġop s",
+      "ĠSyn erg",
+      "ig inal",
+      "ĠC raw",
+      "Ġk ne",
+      "Ġcurv atures",
+      "Ġl ux",
+      "ĠK enn",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠĠĠĠ",
+      "print ln",
+      "Ġverteb rae",
+      "Ġru tile",
+      "ĠAeros ol",
+      "re ferred",
+      "lactam ase",
+      "ve hicle",
+      "ad ir",
+      "iz ards",
+      "Ġcall back",
+      "Cl uster",
+      "Ġsil t",
+      "Ġresearc hed",
+      "ĠGener ator",
+      "ĠRest oration",
+      "ĠCh in",
+      "omet rical",
+      "ĠCo efficients",
+      "rach id",
+      "F ace",
+      "M en",
+      "c ounts",
+      "Ġp eg",
+      "Ġe cl",
+      "Ġcom edy",
+      "ĠL n",
+      "ob uty",
+      "ĠSh aring",
+      "Ġadequ acy",
+      "urt osis",
+      "ĠPic ard",
+      "Ġf estival",
+      "Ġdis position",
+      "ĠCom plement",
+      "ĠEx clusion",
+      "Ġdext ran",
+      "m ons",
+      "ĠInter polation",
+      "ĠSte ven",
+      "Ġcelebr ated",
+      "Ġh Pa",
+      "of requency",
+      "Ġexception ally",
+      "Ġenerge tically",
+      "psych otic",
+      "Land au",
+      "T uple",
+      "dist ributions",
+      "ĠRich ards",
+      "Ġpolyp s",
+      "ĠAbs ence",
+      "Ġcele b",
+      "X G",
+      "Ġsim ulates",
+      "mit ters",
+      "Ġheat map",
+      "ĠSD N",
+      "ĠSte ps",
+      "Ġshall ower",
+      "ĠTurb ulent",
+      "Y T",
+      "Ġn al",
+      "plic ative",
+      "pha e",
+      "ĠLe ica",
+      "ĠAPP RO",
+      "Ġarrhyth mia",
+      "Ġre writing",
+      "Ġuns afe",
+      "Ġcowork ers",
+      "ĠG AD",
+      "iv ol",
+      "Ġdisrup ting",
+      "ĠUltra violet",
+      "ere e",
+      "ĠL opez",
+      "Ġneg ation",
+      "Ġjaponic a",
+      "ec essor",
+      "ĠP atch",
+      "Ġso ap",
+      "ĠY ing",
+      "MS K",
+      "Ġtrac heal",
+      "ic os",
+      "Ġv p",
+      "FA IL",
+      "Ġcat abolism",
+      "sol ver",
+      "f ont",
+      "es p",
+      "ĠZ ou",
+      "Ġdark er",
+      "Ġlys ozyme",
+      "c overed",
+      "Ġmulti tude",
+      "requ ently",
+      "Ġmetam orph",
+      "Ġchap ters",
+      "h h",
+      "ch l",
+      "red undant",
+      "ack ing",
+      "Ġent ail",
+      "ĠPack et",
+      "ĠHabit at",
+      "im edia",
+      "ĠC of",
+      "ph rase",
+      "Ġcl oth",
+      "ars al",
+      "Ġdr ums",
+      "TP UT",
+      "Ar gs",
+      "duct ory",
+      "ĠUl timately",
+      "ic ates",
+      "anti gen",
+      "Th ough",
+      "ĠFl ore",
+      "pro bs",
+      "Ġcirc ulatory",
+      "ĠCont emporary",
+      "e plitz",
+      "Ġh atch",
+      "ri zed",
+      "ĠK op",
+      "mit ting",
+      "Ġhyper spectral",
+      "ĠAb st",
+      "S IM",
+      "Ġfruit ful",
+      "Ġrecip e",
+      "Ġimid azole",
+      "Ġsyn onymous",
+      "Ġattrib ution",
+      "ĠMart ÃŃnez",
+      "ĠRod rÃŃguez",
+      "par ticular",
+      "ĠInter acting",
+      "Con f",
+      "O RE",
+      "ĠT MA",
+      "uc idation",
+      "Ġbi ochemistry",
+      "ĠLe vy",
+      "Ġconcentr ates",
+      "Ġinduc tor",
+      "Ġpy rophosph",
+      "Ġrespond ent",
+      "Z hang",
+      "Ġro pe",
+      "Ġdesign ation",
+      "ĠCl im",
+      "Ġconstrain s",
+      "s helf",
+      "Ġd Ïĥ",
+      "ĠT LC",
+      "ĠA har",
+      "ĠM atch",
+      "ĠM OL",
+      "Ġfe es",
+      "we alth",
+      "Ġhyper activity",
+      "ĠBr uker",
+      "ĠFre und",
+      "dichlor ophenyl",
+      "re ro",
+      "ĠF ear",
+      "dot sc",
+      "Ġhy g",
+      "ĠText ure",
+      "T ak",
+      "am pled",
+      "Ġal geb",
+      "sub t",
+      "Ġdocument ary",
+      "ĠJ E",
+      "CN S",
+      "Ġdecl ar",
+      "He ight",
+      "K i",
+      "en oid",
+      "ĠC ervical",
+      "frac tory",
+      "Ġplant ed",
+      "IF I",
+      "Ġconcept ually",
+      "Ġfill ers",
+      "ic ola",
+      "le an",
+      "Ġcl ump",
+      "Ġwr iters",
+      "Gener ally",
+      "Ġo st",
+      "op ening",
+      "CL ASS",
+      "Ġherpes virus",
+      "In stit",
+      "Ġdr inks",
+      "ĠInt ensive",
+      "Ġmusic ian",
+      "Ġanch ors",
+      "S eries",
+      "ĠF AM",
+      "ĠB ott",
+      "ĠE CC",
+      "Ġinvers ions",
+      "Ġac res",
+      "Ġsw abs",
+      "ĠÍ ī",
+      "ĠBer keley",
+      "Ġpl um",
+      "Ġem power",
+      "Ġphoto emission",
+      "ĠRab i",
+      "E ast",
+      "T aylor",
+      "OS E",
+      "Ġden ied",
+      "ĠHT TP",
+      "M U",
+      "he w",
+      "Ġth ri",
+      "ĠC ERN",
+      "Ġsuff ice",
+      "functional ized",
+      "Ġcra bs",
+      "Ġidem potent",
+      "Ġpost ulate",
+      "ĠCB F",
+      "disc rim",
+      "Char acter",
+      "ĠRecomb ination",
+      "C ache",
+      "om it",
+      "ĠA da",
+      "Ġcur sor",
+      "EM T",
+      "Ġmes oscale",
+      "gu ide",
+      "Hy per",
+      "Ġh t",
+      "ren es",
+      "uss en",
+      "where as",
+      "Ġintegr ator",
+      "Ġsyn cy",
+      "aro us",
+      "Ġcounter act",
+      "hal ose",
+      "ĠNot ation",
+      "ĠRele vance",
+      "v f",
+      "Ġin bred",
+      "Ġrec irc",
+      "Ġend e",
+      "Ġpres idential",
+      "Ġlact ose",
+      "ac ional",
+      "os pi",
+      "ĠV GG",
+      "ose lectivity",
+      "ĠCon fig",
+      "Ġfinger prints",
+      "Inter face",
+      "pur ple",
+      "et us",
+      "ĠN in",
+      "ĠK ras",
+      "ĠRe ports",
+      "ĠSe attle",
+      "AD C",
+      "Ġlipoprotein s",
+      "cyclohex yl",
+      "op ressin",
+      "Ġwave front",
+      "tet razol",
+      "th ys",
+      "Ġdiv or",
+      "amin ophen",
+      "ĠPer ry",
+      "ĠConsider ations",
+      "ĠHal o",
+      "Ġreflex ive",
+      "thiazol idin",
+      "oxycycl ine",
+      "C W",
+      "od im",
+      "ĠCh ong",
+      "Ġequil ibrated",
+      "r ime",
+      "ym ology",
+      "Ġdev oid",
+      "rig el",
+      "amater gic",
+      "Ġidentif ications",
+      "Ġcontroll ability",
+      "ectic ut",
+      "ĠSynchron ization",
+      "ul atus",
+      "Ġcorrel ating",
+      "Ġmu ons",
+      "Ġcompartment al",
+      "Ġinhom ogeneities",
+      "Ġevac uation",
+      "resp iratory",
+      "dim ethoxy",
+      "Ġinterfer ometric",
+      "Ġastr onomy",
+      "Z D",
+      "Ħ Ħ",
+      "el ia",
+      "bl er",
+      "Ġpione ering",
+      "Ġp its",
+      "Ġman soni",
+      "ĠCON D",
+      "Ġcodew ord",
+      "im ura",
+      "ĠDop amine",
+      "ĠGi ov",
+      "ĠCamero on",
+      "S em",
+      "d ong",
+      "ot to",
+      "em ies",
+      "Ġinter quartile",
+      "ll bracket",
+      "otrop ies",
+      "Ġhapp ening",
+      "ĠPal m",
+      "Ġst uff",
+      "Ġpar king",
+      "eg al",
+      "ĠCO P",
+      "Ġorgan izing",
+      "Ġpoly hedral",
+      "Ġproven ance",
+      "J s",
+      "ch ains",
+      "eg u",
+      "mer cap",
+      "level and",
+      "Ġeryth roid",
+      "ympt omatic",
+      "Ġzig zag",
+      "Ġinf erring",
+      "Ġappro x",
+      "Ġdown link",
+      "ĠDef iciency",
+      "rbrack et",
+      "ĠT IM",
+      "ST S",
+      "ain en",
+      "Ġun loading",
+      "ĠX P",
+      "ĠWh ilst",
+      "ĠID H",
+      "ĠTI MP",
+      "r rbracket",
+      "ac ities",
+      "Ġwh ale",
+      "ĠW AR",
+      "Ġinf l",
+      "ĠPresent ation",
+      "authorbs nm",
+      "Ġbacter icidal",
+      "SP EC",
+      "Ġdys regulated",
+      "ĠIC AM",
+      "n ano",
+      "Ġw afers",
+      "ĠM UC",
+      "Ġal ien",
+      "ch ke",
+      "Ġsl abs",
+      "Ġback ing",
+      "ns is",
+      "Ġbal ances",
+      "eth ane",
+      "Link ed",
+      "C hen",
+      "H ymenoptera",
+      "it ations",
+      "ĠO UT",
+      "trans plant",
+      "condition ed",
+      "ĠBenef its",
+      "T yr",
+      "at mosp",
+      "ĠAd hesion",
+      "Ġthor ac",
+      "activ ator",
+      "Ġphosphatidyl inositol",
+      "Ġreported ly",
+      "ĠCL ASS",
+      "Ġrenew ed",
+      "ĠPharmac ological",
+      "Ġminim ise",
+      "gluc osidase",
+      "aden osyl",
+      "Ġov ip",
+      "initial izer",
+      "Ġfor age",
+      "rm s",
+      "ĠIm ag",
+      "ĠAnne xin",
+      "ĠVehic les",
+      "Ġf les",
+      "st a",
+      "ĠG BS",
+      "ĠCh at",
+      "measure ments",
+      "ĠAud itory",
+      "C ut",
+      "F v",
+      "Ġm aker",
+      "ap plication",
+      "Ġrevers ing",
+      "Ġsti p",
+      "Ġfaecal is",
+      "icy cle",
+      "Ġtrim med",
+      "Ġexacerb ation",
+      "Ġtransc ranial",
+      "ĠMoment um",
+      "Ġf c",
+      "ĠF OV",
+      "Ġang ina",
+      "Ġnano structure",
+      "Ġantagon ism",
+      "ĠLED s",
+      "ìĹ IJ",
+      "Ġf als",
+      "ap oration",
+      "ĠIn vasive",
+      "ĠK m",
+      "ert ation",
+      "Ġhar ness",
+      "Ġfer tile",
+      "ĠTR UE",
+      "Ġshel ter",
+      "ĠWol bachia",
+      "sho ot",
+      "Ġs ess",
+      "ĠH ous",
+      "ĠA ce",
+      "ĠC ML",
+      "Ġpro active",
+      "Ġsh ots",
+      "Ġco up",
+      "rest ling",
+      "uniform ly",
+      "y am",
+      "ol ase",
+      "ĠI CS",
+      "ĠE bola",
+      "roll ing",
+      "tr unc",
+      "ĠRepresent atives",
+      "Ġgras ping",
+      "ĠAnomal y",
+      "ĠM ine",
+      "ĠM PO",
+      "ler ight",
+      "Ġinstit ute",
+      "Ġsug arcane",
+      "ÑĢ а",
+      "Ġoccl uded",
+      "ĠMagell anic",
+      "B EC",
+      "W i",
+      "o A",
+      "Ġg apped",
+      "ĠPR C",
+      "ĠMA E",
+      "Ġmusic ians",
+      "ĠSignific antly",
+      "Ġforth coming",
+      "Ġaccl imation",
+      "re quired",
+      "ver bal",
+      "ĠF X",
+      "ĠM LE",
+      "Ġcomp ass",
+      "ĠMultim odal",
+      "G rant",
+      "Ġt vb",
+      "In struction",
+      "Ġsens es",
+      "urb ed",
+      "ham n",
+      "Ġfram ed",
+      "Ġuro thel",
+      "or in",
+      "se al",
+      "Ġfl asks",
+      "sh ops",
+      "Ġwhe els",
+      "ĠRad on",
+      "ĠPlan etary",
+      "Ġhed ge",
+      "Ġd k",
+      "Ġevid ently",
+      "thread s",
+      "Ġt ad",
+      "el im",
+      "im ov",
+      "ist em",
+      "and i",
+      "Ġle isure",
+      "ost om",
+      "Ġcar ing",
+      "ĠSm oking",
+      "Ġcompetit ors",
+      "A FS",
+      "x l",
+      "ĠS atur",
+      "ĠF erg",
+      "Ġch in",
+      "ĠCD R",
+      "ĠSO M",
+      "osacchar ide",
+      "MOD EL",
+      "E CC",
+      "Ġd as",
+      "agon ist",
+      "st ery",
+      "Ġrel ays",
+      "ze k",
+      "Ġneoplas m",
+      "C hip",
+      "Ġg ill",
+      "lam ed",
+      "cer ning",
+      "Ġincons istencies",
+      "ace ans",
+      "ĠAd ri",
+      "ĠAf ghan",
+      "Ġnic hes",
+      "Ġtunn elling",
+      "g us",
+      "ĠI an",
+      "Ġbur ial",
+      "Trans form",
+      "ocomp atible",
+      "Ġs eldom",
+      "Ġdis closed",
+      "âĪ ķ",
+      "Ġref ining",
+      "Ġty ph",
+      "Ġcooper ate",
+      "Ġasphal t",
+      "ĠCons titution",
+      "fl avor",
+      "Ġwar p",
+      "Å ¼",
+      "Ġc raw",
+      "ĠInd igenous",
+      "ĠPre vent",
+      "Ġtrig eminal",
+      "ĠFried rich",
+      "ĠInterfer on",
+      "i osity",
+      "w arm",
+      "us on",
+      "Ġunder lies",
+      "Ġmultiple ts",
+      "ĠSU PER",
+      "ĠManufact uring",
+      "Ġv imentin",
+      "ram ine",
+      "Ġeffic acious",
+      "ic ed",
+      "ĠV all",
+      "oth orax",
+      "Ġaud i",
+      "Q s",
+      "ĠP AL",
+      "ĠH old",
+      "hat tan",
+      "idd ing",
+      "w ana",
+      "Ġp ending",
+      "Ġp erennial",
+      "Ġtouch ing",
+      "xp ected",
+      "D istance",
+      "n av",
+      "Ġis omeric",
+      "ĠM CI",
+      "num bers",
+      "Ġrevers es",
+      "Ġpolyc ystic",
+      "H em",
+      "u ities",
+      "op tional",
+      "Ġsub cortical",
+      "ĠSup ply",
+      "ĠCal der",
+      "Ġmang rove",
+      "Ġp ads",
+      "ur faces",
+      "ĠF aster",
+      "Ġunder neath",
+      "Ġprol actin",
+      "Ġcle arer",
+      "Ġscin tillation",
+      "Ġhumid ified",
+      "ĠW ound",
+      "ĠH PA",
+      "Ġcoll apsing",
+      "Ġbary onic",
+      "ĠMEA SU",
+      "ĠG ü",
+      "Ġdet r",
+      "Ġsubstit uent",
+      "ĠRoman ia",
+      "ĠInv olved",
+      "Ġduoden al",
+      "ĠAm p",
+      "ĠS IS",
+      "sc her",
+      "aut h",
+      "ĠResp ond",
+      "ĠRank ing",
+      "t rip",
+      "x F",
+      "ist in",
+      "Ġpa uc",
+      "ref lection",
+      "Ġcorne a",
+      "Ġbol us",
+      "Ġpiv ot",
+      "Oc tober",
+      "ĠS ERS",
+      "ĠX ing",
+      "AN ET",
+      "Ch inese",
+      "ĠMus c",
+      "D ynamic",
+      "M esh",
+      "Ġdi phosphate",
+      "Ġcons pecific",
+      "lect or",
+      "ĠEc u",
+      "ĠCover age",
+      "ĠãĢ Ī",
+      "C OD",
+      "am ong",
+      "Ġpos it",
+      "imum ab",
+      "Ġp N",
+      "Ġco aching",
+      "ex ports",
+      "Ġreal m",
+      "ĠFer reira",
+      "Ġnation ally",
+      "Ġtur tle",
+      "ubted ly",
+      "ĠD raft",
+      "Ġend l",
+      "ĠContinu um",
+      "embed dings",
+      "Ġá¹ ½",
+      "ĠCr ime",
+      "Ġimm igration",
+      "ĠFil ip",
+      "Ġgar net",
+      "Ġobsc ure",
+      "ĠT YPE",
+      "Ġult rastructural",
+      "ca emia",
+      "ĠSem an",
+      "r ink",
+      "ti ff",
+      "uc cal",
+      "ke e",
+      "itud inally",
+      "ĠAll oy",
+      "ĠAnaly zer",
+      "contin ue",
+      "ĠAlab ama",
+      "Q OL",
+      "Ġpol lin",
+      "Ġcorrespond ences",
+      "ĠRes ol",
+      "F IR",
+      "ul are",
+      "ta wa",
+      "UR CE",
+      "Ġurban ization",
+      "z d",
+      "Ġgl oss",
+      "ER A",
+      "ĠDeterm ine",
+      "D ate",
+      "ĠP SP",
+      "ĠSh ig",
+      "rep ta",
+      "ĠGa it",
+      "neut rino",
+      "Ġper vasive",
+      "ĠâĢ¢ âĢ¢âĢ¢",
+      "Ġhom ozyg",
+      "Ġadap tively",
+      "graph ic",
+      "ĠJohn ston",
+      "z t",
+      "ex plicit",
+      "Ġhel min",
+      "Ġp es",
+      "AR F",
+      "ĠF ram",
+      "ĠAm sterdam",
+      "Ġlogarithm s",
+      "ĠCre ative",
+      "Page Index",
+      "Ġp acing",
+      "ĠP CS",
+      "Ġfore brain",
+      "ĠCT CF",
+      "dec omposition",
+      "Ġbear ings",
+      "Ġanhydro us",
+      "Ġc b",
+      "ĠM ON",
+      "ĠN odes",
+      "str um",
+      "ĠJ ans",
+      "Ġdeline ate",
+      "Ġdichro ism",
+      "con formal",
+      "Ġret reat",
+      "gl ial",
+      "Ġnucle ase",
+      "ĠBal timore",
+      "Ġpay ing",
+      "Ġbore al",
+      "tip ation",
+      "R oot",
+      "S QL",
+      "s ources",
+      "end o",
+      "ĠOr ion",
+      "Pl us",
+      "ĠD EL",
+      "ĠTh an",
+      "Ġmon oph",
+      "Ġreflect or",
+      "Z e",
+      "ĠL inking",
+      "syn c",
+      "ĠCRE B",
+      "n ational",
+      "ur ized",
+      "ĠP eptides",
+      "ĠB egin",
+      "bor g",
+      "piper idyl",
+      "Ġoverestim ation",
+      "R GB",
+      "T K",
+      "Ġbe ings",
+      "Ġat tains",
+      "Ġres ervation",
+      "ĠMo tivation",
+      "Ġtrim ethyl",
+      "ĠTerm inal",
+      "Ġinten tional",
+      "Neg ative",
+      "ĠCron bach",
+      "dorfer i",
+      "D aw",
+      "V AR",
+      "d P",
+      "im ath",
+      "ove rex",
+      "Ġfibro tic",
+      "Ġsmart phones",
+      "Ġont ologies",
+      "G ood",
+      "u tively",
+      "ĠV B",
+      "SP E",
+      "ĠMcD onald",
+      "galax ies",
+      "Ġbioch ar",
+      "ĠE MS",
+      "ĠN f",
+      "ors hip",
+      "Ġback scattering",
+      "ĠÐ ¿",
+      "Ġanthocyan in",
+      "ĠPho enix",
+      "con tained",
+      "ĠPS II",
+      "hl ung",
+      "ĠLA I",
+      "Ġlect ures",
+      "Ġdisp atch",
+      "V F",
+      "ĠM EC",
+      "ĠW es",
+      "Ġback scatter",
+      "oti te",
+      "ĠSR C",
+      "Ġcurren cy",
+      "onym s",
+      "as partate",
+      "Ġcos et",
+      "ĠC PP",
+      "orb ing",
+      "ĠEmbed dings",
+      "ĠSurve ys",
+      "Ġneurodevelop mental",
+      "ĠS RE",
+      "ĠInter ior",
+      "ĠAR DS",
+      "experim ents",
+      "brom ophenyl",
+      "ĠE CL",
+      "ĠO PE",
+      "medi ation",
+      "Ġtherm oc",
+      "Ġinterpret able",
+      "ĠMicrobi ome",
+      "e astern",
+      "Â ¿",
+      "ĠT DP",
+      "ath on",
+      "ĠBy zantine",
+      "any on",
+      "Ġepit axy",
+      "Ġcritic ized",
+      "Mill ipore",
+      "ĠD EP",
+      "ĠFre edom",
+      "j unctions",
+      "ĠA SM",
+      "ĠG ren",
+      "Ġsign ing",
+      "Ġconstit uting",
+      "opro terozoic",
+      "ĠSyn ech",
+      "ĠVo ice",
+      "Ġcholec yst",
+      "b ilities",
+      "on line",
+      "ĠE dd",
+      "ĠK up",
+      "ĠLet t",
+      "ĠMar in",
+      "ĠGo al",
+      "ĠSY M",
+      "intro duced",
+      "naphth yl",
+      "ĠL ü",
+      "Ġm x",
+      "Ġb lu",
+      "Ġr m",
+      "ĠDe letion",
+      "ĠConn ecticut",
+      "Cole optera",
+      "t ry",
+      "Ġso ot",
+      "ĠCount ries",
+      "Ġsick le",
+      "M eta",
+      "ĠS ib",
+      "ĠH NO",
+      "ĠU D",
+      "Ġexp r",
+      "Ġallow able",
+      "ĠInd irect",
+      "tis ation",
+      "Ġaden omas",
+      "electron ics",
+      "R NN",
+      "ĠT CF",
+      "Ġgluc agon",
+      "ĠC itation",
+      "Ġg amb",
+      "and ez",
+      "Ġtrans mits",
+      "aj ima",
+      "Ġhol onomy",
+      "ì ł",
+      "act am",
+      "ĠTh reat",
+      "ĠPear l",
+      "Ġerup tions",
+      "ĠImmunohist ochemistry",
+      "Y es",
+      "p atrick",
+      "Ġa ma",
+      "Ġd rew",
+      "ĠT asks",
+      "ĠP IM",
+      "Ġdis pat",
+      "ĠDet roit",
+      "Ġcoex ist",
+      "arboxyl ase",
+      "I BM",
+      "ĠT UNEL",
+      "ĠU F",
+      "ĠAN G",
+      "Ġsar copenia",
+      "Ġh aptic",
+      "Ġcarbon ates",
+      "Ġmit ophagy",
+      "Ġciti zen",
+      "ĠCONTR OL",
+      "f if",
+      "Ġw i",
+      "ĠG LO",
+      "ens ored",
+      "ĠPar a",
+      "ĠAb del",
+      "oi etin",
+      "Ġto e",
+      "ĠS QU",
+      "ĠR ag",
+      "Ġx ylem",
+      "Ġlib eral",
+      "ĠMarg aret",
+      "W a",
+      "k p",
+      "ĠP EM",
+      "ĠD DR",
+      "Ġgen otypic",
+      "ĠY M",
+      "ING S",
+      "ker as",
+      "ĠEduc ational",
+      "ĠCult ures",
+      "in str",
+      "ĠF uchs",
+      "ag asc",
+      "fact ant",
+      "Ġt enth",
+      "AB L",
+      "Ġperme able",
+      "ĠCam eron",
+      "Br N",
+      "ĠMull er",
+      "ĠRevers ible",
+      "w ild",
+      "Ġf usions",
+      "os ulf",
+      "ĠE oS",
+      "ĠK ö",
+      "det ected",
+      "ĠColl agen",
+      "Ġdescend ants",
+      "e lection",
+      "ar ange",
+      "Ġb ounce",
+      "Ġcont ag",
+      "In valid",
+      "ĠCo ating",
+      "t asks",
+      "ar ma",
+      "ĠK C",
+      "Ġdi ar",
+      "ĠSup press",
+      "Ġfraction ated",
+      "Ġsn ail",
+      "Ġmicro phone",
+      "ĠSc ienti",
+      "Ġchem iluminescence",
+      "soft ware",
+      "Ġburg dorferi",
+      "Ġb oot",
+      "ĠC SCs",
+      "ĠM SI",
+      "ts ev",
+      "Ġhe ater",
+      "frac tal",
+      "Ġend osomes",
+      "Un iform",
+      "Ġath lete",
+      "ĠDri ven",
+      "Ġviv ax",
+      "K ind",
+      "satisf ies",
+      "Ġcorticoster oid",
+      "ĠEstabl ishment",
+      "cal ibration",
+      "Ġdim eric",
+      "Ġcere al",
+      "ĠSuper vised",
+      "ĠSP M",
+      "MB ER",
+      "Ġhemisp heres",
+      "Ġpercenti les",
+      "L eu",
+      "M ajor",
+      "Ġex agger",
+      "Ġds RNA",
+      "Dec ember",
+      "ĠZr O",
+      "Ġas ymmetrical",
+      "ĠV AS",
+      "ĠJ M",
+      "Ġintegr ations",
+      "Ġhand over",
+      "C ycl",
+      "im plant",
+      "Ġqu ote",
+      "Ġcycl one",
+      "ĠSte phan",
+      "ĠFran co",
+      "Ġaw ake",
+      "Ġfeed er",
+      "CH AR",
+      "Con dition",
+      "ĠChar l",
+      "ĠBrig ade",
+      "Ġremedi ation",
+      "c ig",
+      "ĠBoh r",
+      "ĠVac uum",
+      "ĠTox oplasma",
+      "Ġgh relin",
+      "ĠT RAF",
+      "ay e",
+      "Cl ient",
+      "ili ation",
+      "xy z",
+      "ming ham",
+      "ĠSU B",
+      "ïĢ ł",
+      "Ġconvers ions",
+      "Ġmulti path",
+      "miss ive",
+      "Ġeq n",
+      "bul k",
+      "my c",
+      "Ġexacerb ated",
+      "Ø ª",
+      "Ġprotein ase",
+      "Ġbu ilder",
+      "ah ara",
+      "Ġinver t",
+      "ĠRecep tion",
+      "ax anthin",
+      "Ġprim ed",
+      "Ġcop ula",
+      "Ġproceed ings",
+      "Ġnond egenerate",
+      "Ġint ox",
+      "Ġneed les",
+      "length s",
+      "Ġtranspos on",
+      "h on",
+      "ĠT PC",
+      "pl and",
+      "ox yn",
+      "IC H",
+      "Ġintra uterine",
+      "Ġlamin ated",
+      "ĠOBS ERV",
+      "M atch",
+      "ĠIn sur",
+      "ĠAm yloid",
+      "Ġwar ped",
+      "emat ical",
+      "ĠPrac tices",
+      "ãģ ®",
+      "ĠBrass ica",
+      "Ġhyperther mia",
+      "Ġd n",
+      "ĠL IF",
+      "ĠMet ropolitan",
+      "ĠBr dU",
+      "imp act",
+      "f iltered",
+      "ĠRe agent",
+      "v p",
+      "ĠT ip",
+      "ĠPro portional",
+      "Ġblood stream",
+      "Sim ple",
+      "Ġty ros",
+      "ĠHen ri",
+      "Ġretro trans",
+      "aci ens",
+      "Ġmist akes",
+      "acyl glycerol",
+      "ĠMir ror",
+      "V ERSION",
+      "v re",
+      "Ġb act",
+      "ock ed",
+      "eps is",
+      "Ġson ication",
+      "ĠPur kinje",
+      "Ġmism atches",
+      "ĠA OD",
+      "Ġhyper graph",
+      "ĠMi ami",
+      "am med",
+      "Ġcon versely",
+      "ĠG abor",
+      "ĠG DM",
+      "Ġco iled",
+      "onic a",
+      "Ġevol utions",
+      "ĠR BM",
+      "ĠRe ef",
+      "ĠAb ram",
+      "ĠPrec ise",
+      "incre ase",
+      "ĠPlate let",
+      "Gener ator",
+      "Ar ch",
+      "ĠBen ed",
+      "pre ceq",
+      "meas urable",
+      "C AS",
+      "ĠT ourn",
+      "Ġg iants",
+      "Ġed dies",
+      "Ġcolumn ar",
+      "agg regation",
+      "Ġzircon ia",
+      "duc ibility",
+      "Ġserv o",
+      "Ġbe auty",
+      "Ġhe ap",
+      "ĠâĪĴ âĪĴâĪĴ",
+      "Ġconduc tivities",
+      "Ġdark ness",
+      "Ġoccup ying",
+      "ĠCle an",
+      "b ash",
+      "ul ans",
+      "app y",
+      "ĠMark er",
+      "run time",
+      "Ġhaem oglobin",
+      "Ġdesk top",
+      "m is",
+      "ĠS of",
+      "os se",
+      "Ġcom oving",
+      "Ġcl utter",
+      "our ced",
+      "Ġsub j",
+      "arch ing",
+      "ĠSol omon",
+      "lock ing",
+      "Ġpar ap",
+      "Ġrot ator",
+      "ĠACKNOWLEDGM ENTS",
+      "T er",
+      "y ster",
+      "ĠWe bb",
+      "Ġsubs ample",
+      "osil icate",
+      "T raining",
+      "or pha",
+      "Ġtime out",
+      "otin amide",
+      "ĠFab ry",
+      "ĠRece iver",
+      "Ġconjunc tiv",
+      "ĠEcu ador",
+      "ĠI da",
+      "Ġcase in",
+      "Ġïģ ¸",
+      "ĠBar n",
+      "ĠSchool s",
+      "el ona",
+      "di p",
+      "ĠCh rys",
+      "IC I",
+      "Ġposterior i",
+      "Ġble aching",
+      "ĠPerson ality",
+      "um bers",
+      "ĠM odes",
+      "Ġno tification",
+      "Ġsup ine",
+      "alu ed",
+      "ke ep",
+      "ĠFran z",
+      "Ġwound ed",
+      "Y L",
+      "Ġdi lemma",
+      "ĠCl ara",
+      "ĠCar roll",
+      "Ġsick ness",
+      "Ġprox ies",
+      "ec ks",
+      "ĠÏ «",
+      "Ġplant ing",
+      "Ġcipher text",
+      "ĠF amilies",
+      "ies el",
+      "Ġinc ongru",
+      "ĠExc itation",
+      "Ġconfer red",
+      "ĠBut ter",
+      "Im pl",
+      "coll ision",
+      "id ol",
+      "Ġac quires",
+      "ĠO wen",
+      "SA M",
+      "ĠG UT",
+      "lec ts",
+      "Ġdele g",
+      "Sh ot",
+      "Ġanth rac",
+      "Russ ian",
+      "ĠP CE",
+      "Ġâ ĥĹ",
+      "ĠK ab",
+      "NA C",
+      "Ġarg parse",
+      "ĠVi ol",
+      "Ġantico agulation",
+      "Ġcredi bility",
+      "Ġrota virus",
+      "ĠIn vest",
+      "Ġrec ol",
+      "vari ety",
+      "Ġdeform able",
+      "Ġenerge tics",
+      "Ġconsult ations",
+      "le tics",
+      "ĠF oss",
+      "ĠL IGO",
+      "ph p",
+      "ĠCh al",
+      "ĠMal awi",
+      "Ġstro kes",
+      "h orm",
+      "Ġb s",
+      "Ġpl ural",
+      "str ategy",
+      "Ġmis alignment",
+      "pre vious",
+      "fil ters",
+      "ĠDem ographics",
+      "determ inistic",
+      "Ġcycl ophosphamide",
+      "Ġstre ak",
+      "ĠBios ynthesis",
+      "Ġsubcutaneous ly",
+      "j n",
+      "Ġam picillin",
+      "ĠCh ag",
+      "iform es",
+      "IF ICATION",
+      "Ġyour self",
+      "Ġtoler ability",
+      "Ġaut ocl",
+      "rh s",
+      "Ġpup ils",
+      "Ġgaug ed",
+      "L ay",
+      "ĠS anti",
+      "ĠD BP",
+      "ĠG ary",
+      "dri ve",
+      "Ġtrust worth",
+      "Ġconting ency",
+      "C ube",
+      "H ost",
+      "f u",
+      "Ġh sa",
+      "iss ner",
+      "IT T",
+      "ĠSr TiO",
+      "Ġcouns elling",
+      "inte grable",
+      "Ġunder way",
+      "Ġstandard ised",
+      "bi us",
+      "First ly",
+      "Ġporph yrin",
+      "A rea",
+      "i w",
+      "Ġ ub",
+      "ĠL ynch",
+      "ĠW BC",
+      "ild en",
+      "Ġhom eless",
+      "Ġmagnet osphere",
+      "Ġnight time",
+      "nc bi",
+      "Ġdow nt",
+      "le thal",
+      "Ġinter im",
+      "ĠRes il",
+      "Ġcontinu ally",
+      "ĠImmun ofluorescence",
+      "Des ign",
+      "Ġadvoc ate",
+      "repta vidin",
+      "f w",
+      "st ory",
+      "ĠP SS",
+      "Ġfil ed",
+      "Ġbed rock",
+      "Ġisofl urane",
+      "Ġrel uct",
+      "ew ard",
+      "ĠInd ependence",
+      "ĠBurk holder",
+      "Ġc inn",
+      "Ġcap tive",
+      "Ġcompos ing",
+      "Ġrest raint",
+      "Ġquestion able",
+      "ĠTom ato",
+      "Ġzer oth",
+      "r ins",
+      "ome z",
+      "Ġgl ia",
+      "ĠGl ac",
+      "Ind ependent",
+      "Ġobj ectively",
+      "p A",
+      "Ġfav oring",
+      "ipel ago",
+      "Ġincont inence",
+      "b ium",
+      "ĠL Z",
+      "ĠL ed",
+      "hex yl",
+      "Ġce ased",
+      "Ġole ic",
+      "ĠImpair ment",
+      "Ñ ĸ",
+      "ong o",
+      "Ġrun ner",
+      "Ġcuc umber",
+      "ĠPer form",
+      "Ġdouble ts",
+      "Ġeigen function",
+      "ĠÌ º",
+      "ĠHend erson",
+      "K lein",
+      "T ab",
+      "Ġbe er",
+      "oc om",
+      "unc iation",
+      "---- --",
+      "ĠT SC",
+      "og as",
+      "Ġr ud",
+      "Ġinc is",
+      "ĠLO G",
+      "FB Q",
+      "Ġinterconn ection",
+      "Ã ®",
+      "ar box",
+      "ĠI BS",
+      "ĠN CT",
+      "ĠG and",
+      "Ġy aw",
+      "ĠTrans verse",
+      "ĠSud an",
+      "Ġconstric tion",
+      "H or",
+      "Ġev asion",
+      "Ġmer omorphic",
+      "ĠPB MC",
+      "I UM",
+      "re ed",
+      "ĠB ö",
+      "ĠE MB",
+      "uk h",
+      "Ġwin ners",
+      "Ġasc ites",
+      "M es",
+      "Ġe clipse",
+      "ĠE ocene",
+      "ad iazol",
+      "Ġrecover ies",
+      "Star ting",
+      "re ma",
+      "ĠÃ İ",
+      "mon otonic",
+      "ĠMe OH",
+      "ĠFl ood",
+      "Ġwat ching",
+      "G TP",
+      "i el",
+      "m üller",
+      "å ħ",
+      "Ġpolyphen ol",
+      "ĠL MI",
+      "red it",
+      "ther m",
+      "Ġneur ite",
+      "Qu antum",
+      "rach lor",
+      "ĠRub in",
+      "Ġbf nm",
+      "A re",
+      "ar achn",
+      "Ġd uck",
+      "ĠTra jectory",
+      "ĠNit ric",
+      "l v",
+      "u id",
+      "im ag",
+      "ĠM ULT",
+      "Ġgen re",
+      "ari e",
+      "Ġtr ifluor",
+      "ĠCor pus",
+      "oli osis",
+      "ĠCC K",
+      "K it",
+      "f ather",
+      "Ġt ennis",
+      "its ch",
+      "HC V",
+      "l antic",
+      "ĠA Q",
+      "iz u",
+      "ast atin",
+      "oth io",
+      "ĠAn atomy",
+      "Ġá ŀ",
+      "glob ulin",
+      "Ġinterp ol",
+      "Ġtunn els",
+      "Ġn atri",
+      "im ed",
+      "ĠD ew",
+      "Ġsub scripts",
+      "tit es",
+      "Ġhistological ly",
+      "O pt",
+      "x n",
+      "Ġres ampling",
+      "ane y",
+      "Ġtr ast",
+      "Ġsin ensis",
+      "Ġsenes cent",
+      "F ast",
+      "Ġh ampered",
+      "Ġblock er",
+      "ush ima",
+      "Ġhospital izations",
+      "L im",
+      "o ons",
+      "Ã ¿",
+      "ĠA PS",
+      "ĠY ok",
+      "ĠZ am",
+      "Ġexperim enter",
+      "ĠDis ks",
+      "Ġà ¬",
+      "ĠS cop",
+      "ĠA ph",
+      "ĠP arents",
+      "ĠPl ots",
+      "ĠCON T",
+      "ĠÐ Ī",
+      "Ġhomolog ue",
+      "ĠCool ing",
+      "re th",
+      "Ġo vari",
+      "ĠT amil",
+      "v rule",
+      "ĠP CP",
+      "ari ous",
+      "Ac tive",
+      "oprot ection",
+      "ĠAlf v",
+      "Ġinf ra",
+      "ĠCo herence",
+      "clos ures",
+      "hydrox ymethyl",
+      "E H",
+      "Ġm aser",
+      "ĠN IST",
+      "lec k",
+      "con cat",
+      "Ġtra ine",
+      "Ġmix es",
+      "Ġrib osomes",
+      "l ia",
+      "p uls",
+      "Ġas cer",
+      "ĠB anks",
+      "ell in",
+      "ap plied",
+      "Ġcl ips",
+      "Ġmet ap",
+      "Ġcop ro",
+      "Ġepid id",
+      "ĠEpidem iological",
+      "ĠNich olas",
+      "ĠK ings",
+      "Ġlar va",
+      "Bs Ag",
+      "ĠS ánchez",
+      "ĠâĪ İ",
+      "AM D",
+      "ĠHa o",
+      "ĠBill board",
+      "ĠAbor iginal",
+      "Ġn ylon",
+      "ĠN AN",
+      "c ores",
+      "ĠC rop",
+      "Ġcom mittees",
+      "Ġdi hedral",
+      "ĠJ uli",
+      "ĠAnd y",
+      "hyd ration",
+      "correspond s",
+      "M ut",
+      "Ġt orn",
+      "ĠF EV",
+      "Ġx s",
+      "amp hen",
+      "Ġsummar ization",
+      "ĠEr g",
+      "Ë Ĩ",
+      "ĠJ unction",
+      "anc ouver",
+      "ĠEx amining",
+      "ĠMarc o",
+      "Po inter",
+      "Ġscarc ity",
+      "unc ing",
+      "Ġbi jective",
+      "ĠMain e",
+      "ĠRH IC",
+      "Ġtow ers",
+      "Ġgent amicin",
+      "Ġt onic",
+      "Ġk T",
+      "Ġclim bing",
+      "Ġrecru its",
+      "ĠHot el",
+      "ĠJew s",
+      "ĠRUN X",
+      "Ġausten ite",
+      "ĠOffic er",
+      "in ent",
+      "uc c",
+      "ĠB idirectional",
+      "Ġmay or",
+      "ĠAss ays",
+      "ĠER G",
+      "SN Ps",
+      "d ine",
+      "Ch ina",
+      "star ting",
+      "Ġirr ational",
+      "ĠDIF FE",
+      "Ġmillisecond s",
+      "L ik",
+      "in one",
+      "Ġâ ģĦ",
+      "Ġcons picuous",
+      "Ġsur plus",
+      "ĠX iong",
+      "Ġup grade",
+      "Ġtim ep",
+      "ĠÄ Į",
+      "Te V",
+      "orbid ities",
+      "in valid",
+      "Ġv ide",
+      "ter ra",
+      "Ġan tin",
+      "em ens",
+      "oc ese",
+      "ĠK I",
+      "Ġevolution arily",
+      "K er",
+      "ĠL ES",
+      "cl amp",
+      "Ġslow ed",
+      "gly coprotein",
+      "enti eth",
+      "Ġab road",
+      "Ġinterp olating",
+      "Ġcataly ze",
+      "ĠBelg ian",
+      "Ġphotograp hed",
+      "Ġp ector",
+      "ĠS IV",
+      "ĠE LECT",
+      "Ġdes al",
+      "one ph",
+      "ĠCl os",
+      "Ġafford able",
+      "b irds",
+      "g om",
+      "Ġr r",
+      "Ġun i",
+      "ĠGen us",
+      "ĠReg ge",
+      "ĠMulti dimensional",
+      "Ġpsych opathology",
+      "Ġcer tification",
+      "P attern",
+      "ĠT ower",
+      "Ġst ern",
+      "Ġsub lattice",
+      "Ġgr at",
+      "Ġly rics",
+      "f mt",
+      "o ceptive",
+      "Ġd P",
+      "ĠHol mes",
+      "Ġbudget s",
+      "Ġeut ectic",
+      "ĠP v",
+      "ĠG ott",
+      "Ġdis infection",
+      "Ġret inoic",
+      "ĠOb st",
+      "Ġrepl en",
+      "Ġâĸ ł",
+      "K utta",
+      "P lease",
+      "ĠC AG",
+      "ĠSti r",
+      "spe aking",
+      "Ġinsectic ides",
+      "ĠFung i",
+      "H od",
+      "R ON",
+      "co il",
+      "ĠVis ible",
+      "Ġin ception",
+      "Ġe GFR",
+      "Ġre ionization",
+      "Ġdom ination",
+      "ĠMet ro",
+      "Ġsw ept",
+      "MD S",
+      "Ġsubs idence",
+      "ĠFall s",
+      "ĠD rum",
+      "ĠCons erved",
+      "ĠMy ers",
+      "Ġadapt ability",
+      "Ġly ophil",
+      "ul ina",
+      "are lli",
+      "ocy cles",
+      "ĠSO A",
+      "Ġds DNA",
+      "ĠCE O",
+      "Ġanch oring",
+      "Ġde activation",
+      "yl er",
+      "Ġinteresting ly",
+      "Ġ iliac",
+      "ĠB org",
+      "ĠPT C",
+      "ocyan in",
+      "Ġun used",
+      "ĠCar rier",
+      "Wh ich",
+      "Ġinterven ing",
+      "Ġprivi le",
+      "h it",
+      "Ġche aper",
+      "ĠCycl in",
+      "p lying",
+      "ĠC leveland",
+      "ĠH ahn",
+      "Ġag glut",
+      "ĠAn ch",
+      "ĠRed ox",
+      "W ill",
+      "ĠL inn",
+      "ron es",
+      "ĠNew castle",
+      "ĠExp ected",
+      "ĠOpportun ities",
+      "ĠL arger",
+      "Ġle ach",
+      "Ġpep per",
+      "S ha",
+      "s ector",
+      "y ou",
+      "Ġre plications",
+      "ch olesterolem",
+      "ĠIn vasion",
+      "Ġb ony",
+      "ĠH uber",
+      "the nd",
+      "Ġreal ised",
+      "Ġinvest ments",
+      "C ataly",
+      "ĠW itt",
+      "ĠK ai",
+      "Ġet ched",
+      "ĠST EM",
+      "Ġexcl udes",
+      "Ex ec",
+      "ĠStrong ly",
+      "ĠSym posium",
+      "ĠTub erculosis",
+      "il ance",
+      "ĠR IS",
+      "ap ia",
+      "ens ated",
+      "ne b",
+      "ĠCh ains",
+      "Ġent hus",
+      "quad rup",
+      "dec l",
+      "Ġbin ned",
+      "Ġsynerg istically",
+      "Ġgaug es",
+      "whe ther",
+      "dise ase",
+      "W estern",
+      "Ġhyp othermia",
+      "ĠGard ner",
+      "Ġaber ration",
+      "R od",
+      "Í ĺ",
+      "Ġf d",
+      "Ġst ood",
+      "Ġcondition ally",
+      "Ġthrom bol",
+      "P SC",
+      "Ġm k",
+      "ĠT ER",
+      "od ds",
+      "ĠK ri",
+      "ĠIV F",
+      "Ġm ites",
+      "ĠC HE",
+      "Ġq q",
+      "ĠInf ants",
+      "ĠChar lot",
+      "bec co",
+      "et om",
+      "ĠCD S",
+      "Ġarchae al",
+      "ĠHN SCC",
+      "Ġmonod romy",
+      "amphen icol",
+      "a pers",
+      "re activity",
+      "Ġund erm",
+      "In ternal",
+      "ĠLands at",
+      "G erman",
+      "Ġcer vix",
+      "id azole",
+      "ĠS ketch",
+      "ĠL AM",
+      "ĠN erve",
+      "ĠTe h",
+      "Ġmuss el",
+      "Ð ·",
+      "ĠMicro array",
+      "we i",
+      "Ġwhe y",
+      "Ġmix er",
+      "Ġrecon figurable",
+      "Ġvascul itis",
+      "Ġkw args",
+      "Ġre us",
+      "cor relations",
+      "Ġwood y",
+      "carbon ate",
+      "ectom ized",
+      "Ġret rans",
+      "Ġcyt ometric",
+      "ĠWild life",
+      "ĠAnsw ering",
+      "Ġp encil",
+      "ĠD AS",
+      "ak rish",
+      "CE PT",
+      "ĠÄ Ŀ",
+      "ĠPers ian",
+      "conver ting",
+      "Ġc ater",
+      "Ġmean while",
+      "TP A",
+      "Ġr um",
+      "ĠG ros",
+      "up e",
+      "Ġreg urg",
+      "Ġpenal ties",
+      "Pos itive",
+      "************************************************************************ ****",
+      "X F",
+      "e enth",
+      "ĠC ory",
+      "od ulation",
+      "Ġqu orum",
+      "c odes",
+      "ar am",
+      "ĠT SA",
+      "ĠP n",
+      "âĪ ij",
+      "pr ison",
+      "Ġconfidential ity",
+      "E PS",
+      "X iv",
+      "i ensis",
+      "est ones",
+      "ĠZ ag",
+      "Ġpred ecessor",
+      "Ġpri ze",
+      "Ġâİ ¨",
+      "ster oidal",
+      "op ard",
+      "Ġimp ractical",
+      "Ġdemonstr ations",
+      "Ġpredis position",
+      "Ġk k",
+      "Ġmod ifiers",
+      "Ġprec a",
+      "Ġexec utes",
+      "Ġbin ning",
+      "Ġped ig",
+      "ĠKL F",
+      "ĠS keletal",
+      "ĠC IN",
+      "ature d",
+      "Ġdecom poses",
+      "Ġaph id",
+      "B ern",
+      "P ur",
+      "ĠE PO",
+      "mer ge",
+      "ĠCO SM",
+      "am yloid",
+      "mon ia",
+      "ĠSc ores",
+      "ĠReg istration",
+      "ĠAg robacterium",
+      "Ġenter prises",
+      "loc ality",
+      "ĠIT O",
+      "Ġt ess",
+      "Ġf cc",
+      "ĠN c",
+      "Ġco axial",
+      "ĠAd vant",
+      "AP C",
+      "ĠDem and",
+      "adj ust",
+      "Po ints",
+      "Ġhetero structure",
+      "iffiffiffiffiffiffiffiff iffiffiffiffiffiffiffiff",
+      "D Q",
+      "Ġt ensions",
+      "ab und",
+      "ĠH utch",
+      "bre w",
+      "Ġvit reous",
+      "ĠEZ H",
+      "Ġm erc",
+      "Ġdeb ated",
+      "Ġpal ate",
+      "ocol ate",
+      "Ġevap otranspiration",
+      "ĠẠ¼",
+      "ĠHoff man",
+      "ĠGALAX IES",
+      "C AL",
+      "c aps",
+      "le gal",
+      "di ed",
+      "ĠIs olates",
+      "Ġagg rav",
+      "q s",
+      "ĠI CT",
+      "Ġse als",
+      "Ġspin el",
+      "ĠGe or",
+      "Bl ue",
+      "Ġure ter",
+      "spl ine",
+      "ĠIntro ducing",
+      "thend ieck",
+      "op per",
+      "Ġafter glow",
+      "Ġend osomal",
+      "Ġreal izes",
+      "sol ving",
+      "Ġmist ake",
+      "ĠAthe ros",
+      "ĠS BS",
+      "ĠR ut",
+      "ex ist",
+      "Pro f",
+      "ĠNe isser",
+      "MS G",
+      "ĠEar lier",
+      "Ġd T",
+      "ĠSp read",
+      "ĠRef lection",
+      "second ary",
+      "approxim ate",
+      "Ġnig ra",
+      "S olution",
+      "an one",
+      "ĠIt ems",
+      "Ġwave lets",
+      "ĠSol uble",
+      "Ġcircular ly",
+      "ĠCU DA",
+      "Ġreg enerated",
+      "SP I",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠĠ",
+      "at uring",
+      "RE Q",
+      "Ġinter oper",
+      "ree v",
+      "ON T",
+      "isc hen",
+      "ĠCho osing",
+      "phosphor ylated",
+      "á Ī",
+      "Ġd ress",
+      "ĠCon form",
+      "Ġrem emb",
+      "Ġischa emic",
+      "B asic",
+      "ĠP ang",
+      "Ġcr it",
+      "ĠOr n",
+      "Ġg m",
+      "ĠF og",
+      "ĠB d",
+      "rac heal",
+      "Ġphen ols",
+      "ĠDist ingu",
+      "Ġâİ ©",
+      "ĠGR Bs",
+      "ĠCe O",
+      "ĠBiom ass",
+      "Ġapt amer",
+      "v isc",
+      "he tically",
+      "Ġs id",
+      "ome g",
+      "Ġproportion ality",
+      "ÃŃ s",
+      "toplas mic",
+      "ĠConn ected",
+      "Ġlamin in",
+      "stra hlung",
+      "ĠL ad",
+      "TR AN",
+      "ä r",
+      "Ġbasal t",
+      "ĠCur vature",
+      "Ġmitig ating",
+      "opa edic",
+      "ĠMuh ammad",
+      "C AR",
+      "G i",
+      "Ġet ch",
+      "ha ir",
+      "Ġpur ine",
+      "Ġbenchmark ing",
+      "re ich",
+      "Ġmet hicillin",
+      "âĪ ¥",
+      "Ġman ages",
+      "sol vent",
+      "ĠSha o",
+      "h c",
+      "Ġstr uck",
+      "Ġnucle osome",
+      "ĠPubl ication",
+      "M etric",
+      "Ġw ines",
+      "ĠM BL",
+      "ĠH ub",
+      "ĠAss istant",
+      "Ġreli ance",
+      "Ġrout ers",
+      "ĠHer z",
+      "ĠTob acco",
+      "ro gram",
+      "ĠH SD",
+      "ĠL BP",
+      "Ġinf lection",
+      "sch ool",
+      "Ġspons ored",
+      "ĠCen ozoic",
+      "Ġentertain ment",
+      "ati an",
+      "archit ecture",
+      "brow se",
+      "RE C",
+      "ist ure",
+      "ĠCh olesterol",
+      "ĠSim plified",
+      "Ġpolyp eptides",
+      "Ġpunct ures",
+      "arachn oid",
+      "S elf",
+      "Ġan orexia",
+      "ĠO le",
+      "ĉĉ ĠĠĠĠ",
+      "GB T",
+      "Ġcardiomy ocyte",
+      "ĠFlo quet",
+      "anal og",
+      "Ġsensiti zed",
+      "ĠCep he",
+      "c atch",
+      "ch ial",
+      "Ġcere mony",
+      "Ġter at",
+      "Ġamelior ate",
+      "olys in",
+      "et ooth",
+      "ak in",
+      "ha em",
+      "Ġent ropies",
+      "Ġarg u",
+      "Ġcop ied",
+      "ling ton",
+      "ĠHer pes",
+      "ĠSchw ann",
+      "y k",
+      "ĠC EA",
+      "ĠI CH",
+      "Ġwr ink",
+      "Ġrun ners",
+      "Ġgal van",
+      "Ġconsol idated",
+      "ĠâĢ ¡",
+      "ĠClass ic",
+      "Ġepidem iologic",
+      "ĠDri ving",
+      "Ġtrast uzumab",
+      "C YP",
+      "N CT",
+      "t ability",
+      "Ġs lee",
+      "ĠN eck",
+      "Ġassess es",
+      "Ġsymmet rically",
+      "ĠPot ts",
+      "ĠRib osomal",
+      "d iction",
+      "g all",
+      "ĠAcc eleration",
+      "CL A",
+      "ACT ER",
+      "x ed",
+      "Ġg eriatric",
+      "th reonine",
+      "Ġab ort",
+      "Ġar tem",
+      "ĠDis ney",
+      "ĠCorrespond ence",
+      "Ġre nt",
+      "ĠN UM",
+      "ĠCh un",
+      "ĠRec ogn",
+      "Ġcrystall ized",
+      "Ġcontrad icting",
+      "vis ors",
+      "mal ignant",
+      "rophys iology",
+      "Inf rared",
+      "g z",
+      "Ġsub lim",
+      "omat osis",
+      "osyl transferase",
+      "Ġholog raphy",
+      "oren stein",
+      "¾ ±",
+      "ĠSe bas",
+      "acc um",
+      "Up per",
+      "ant enna",
+      "Ġbl ur",
+      "Ġsm ell",
+      "Ġthree fold",
+      "ĠPl ayers",
+      "Ġallevi ated",
+      "B in",
+      "Ġn inet",
+      "ĠD na",
+      "Ġgeneral izing",
+      "Ġbreak age",
+      "ĠMor rison",
+      "mac ro",
+      "Read er",
+      "ograv imetric",
+      "Ġd h",
+      "le w",
+      "xt on",
+      "Ġdec eleration",
+      "ĠCor related",
+      "ĠLeg ion",
+      "Ġgam bling",
+      "B inding",
+      "ĠIn As",
+      "low ering",
+      "Ġeuthan ized",
+      "ĠDall as",
+      "ĠD w",
+      "ĠDi jk",
+      "ĠPol ic",
+      "ĠT IME",
+      "ĠH EL",
+      "ĠL anguages",
+      "Ġpar abol",
+      "por ating",
+      "Ġfr ustration",
+      "μ M",
+      "ball s",
+      "ĠArm strong",
+      "Ġcontrac tility",
+      "Ġmetalloprotein ases",
+      "am eric",
+      "ĠZ ak",
+      "ĠCost s",
+      "A lex",
+      "d og",
+      "p w",
+      "ĠT ight",
+      "ĠAn terior",
+      "Ġpe aking",
+      "Ġneg ativity",
+      "Ġhyd ride",
+      "ĠL iv",
+      "Ġster ilized",
+      "Ġverb atim",
+      "Altern atively",
+      "RE QU",
+      "ĠTy phimurium",
+      "ĠWein berg",
+      "D SC",
+      "r q",
+      "Ġcor rug",
+      "Ġmic rons",
+      "co ord",
+      "i oid",
+      "s at",
+      "Ġfl occ",
+      "ĠAcc elerated",
+      "Ġsix teen",
+      "abs ence",
+      "ĠSpe aker",
+      "om ological",
+      "ĠA pr",
+      "Ġmat roid",
+      "tig ht",
+      "ogene tically",
+      "rum p",
+      "ĠInhib its",
+      "ĠOlymp us",
+      "Ġposs ession",
+      "Ġsuper visor",
+      "Ġconc ise",
+      "optim ized",
+      "v ivo",
+      "Ġstep ped",
+      "ocy anine",
+      "F ive",
+      "an as",
+      "ar ten",
+      "ĠC aco",
+      "Ġsol utes",
+      "IT AL",
+      "ĠRed dy",
+      "Ġwar ping",
+      "Ġolig omer",
+      "Ġc apped",
+      "Ġv oted",
+      "ĠR ico",
+      "ĠT rem",
+      "Ġl ime",
+      "ĠI SP",
+      "ĠL ayers",
+      "sk in",
+      "rang ed",
+      "á z",
+      "Ġbio activity",
+      "Ġd urable",
+      "Ġh n",
+      "ĠC AB",
+      "Ġv a",
+      "ĠU WB",
+      "ĠSt uart",
+      "Ġlength y",
+      "Ġinvas iveness",
+      "Ġâĩ Ķ",
+      "jo ining",
+      "ĠRB Cs",
+      "Ġresil ient",
+      "ĠManip ulation",
+      "G erm",
+      "cont ribution",
+      "Ġqual ify",
+      "ĠD ashed",
+      "Ġacceler ations",
+      "ĠCyt ochrome",
+      "Ġcircumst ellar",
+      "c avity",
+      "Ġan atase",
+      "ĠDe vi",
+      "Ġpur su",
+      "ĠMicro RNAs",
+      "Ġnorth ward",
+      "Ġsun flower",
+      "ĠEnter tainment",
+      "Pac ific",
+      "ĠHolog raphic",
+      "u j",
+      "ere ll",
+      "met hanol",
+      "Sur face",
+      "opos itive",
+      "Ġthreat ening",
+      "Ġtransc end",
+      "D epend",
+      "Ġq i",
+      "tis ed",
+      "ĠBr istol",
+      "umm ation",
+      "Ġextract or",
+      "Ġfav oured",
+      "ĠPy ro",
+      "ĠEngine ers",
+      "flat ten",
+      "toler ance",
+      "Ġ xt",
+      "ĠT ot",
+      "Ġtest bed",
+      "IC U",
+      "ĠSw arm",
+      "Ġintern ationally",
+      "Ġant ine",
+      "ĠInsur ance",
+      "b ai",
+      "n h",
+      "Ñ ĭ",
+      "os ac",
+      "ĠL ec",
+      "th or",
+      "Ġout ermost",
+      "Ġdo ors",
+      "Ġbi ometric",
+      "glut amate",
+      "ĠWood s",
+      "ĠMun ich",
+      "u ximab",
+      "pl aces",
+      "Ġam yotrophic",
+      "ĠPar am",
+      "ĠChrist ensen",
+      "A ge",
+      "en ne",
+      "Ġan im",
+      "Ġrec rystallization",
+      "ĠPro positions",
+      "Ġsn ails",
+      "Second ly",
+      "ĠPU FA",
+      "F rance",
+      "S rc",
+      "v itro",
+      "om ass",
+      "ur u",
+      "ĠL ever",
+      "ect onic",
+      "emb l",
+      "PC L",
+      "Ġcoordin ator",
+      "ĠFox p",
+      "ĠBir mingham",
+      "ĠLib eral",
+      "Ġcru ise",
+      "Ġi θ",
+      "Ġsym p",
+      "az aki",
+      "ĠPar se",
+      "Ġhyd rologic",
+      "Ġprolong ation",
+      "ĠHay es",
+      "Ġsubm uc",
+      "Ġaggl omeration",
+      "A RE",
+      "ĠF MR",
+      "ĠL omb",
+      "math char",
+      "Ġstruct uring",
+      "Ġelectroph oretic",
+      "Ġdimin ishing",
+      "Ġbra ke",
+      "chen ko",
+      "ĠPere ira",
+      "l ens",
+      "Ġback end",
+      "Ġillustr ations",
+      "Ġdemand ed",
+      "Ġnotice ably",
+      "ĠKa iser",
+      "ĠDavid son",
+      "Ġbrak ing",
+      "T p",
+      "For ward",
+      "μ ν",
+      "ĠCd S",
+      "Ġaster oids",
+      "Provid er",
+      "ĠE ut",
+      "Ġtr il",
+      "ung s",
+      "Ġdiv ing",
+      "ĠUAV s",
+      "ĠiP SC",
+      "i int",
+      "Ġ ×",
+      "th rombin",
+      "Ġcoordin ating",
+      "ext rem",
+      "Ġembol ization",
+      "ĠAdi p",
+      "pl ated",
+      "ĠH ag",
+      "ĠE TS",
+      "Ġbro od",
+      "An g",
+      "ĠPC V",
+      "det ail",
+      "R SS",
+      "b ens",
+      "Ġt ier",
+      "ĠC ock",
+      "Ġg ay",
+      "Ġqu int",
+      "Ġag enda",
+      "Ġaff airs",
+      "ĠMod erate",
+      "hel ical",
+      "ĠEqu ivalent",
+      "Ġproportion ally",
+      "Col umn",
+      "FW HM",
+      "A ir",
+      "E num",
+      "ific e",
+      "arc sec",
+      "ĠTR IM",
+      "ĠLab eling",
+      "Q AM",
+      "p ies",
+      "Ġis otropy",
+      "ĠG ó",
+      "Ġpo inters",
+      "tig raphy",
+      "ram ers",
+      "Ġmac aque",
+      "Ġmiss es",
+      "Ġelliptic ity",
+      "present ed",
+      "galact osidase",
+      "É Ľ",
+      "in ion",
+      "Ġm ite",
+      "ll l",
+      "Ob jective",
+      "Ġprison ers",
+      "ĠHerc ules",
+      "Ġanti s",
+      "Ġclos ures",
+      "ĠMar tian",
+      "Ġter pen",
+      "rob ust",
+      "Ġsequel ae",
+      "al arial",
+      "ĠC SA",
+      "ĠB land",
+      "ĠG ent",
+      "Ġor phan",
+      "Ġind ent",
+      "big wedge",
+      "Ġdefin able",
+      "Ġolig osaccharides",
+      "ĠBat talion",
+      "Ġis ometries",
+      "az olin",
+      "ĠSh own",
+      "spect ra",
+      "Vis ual",
+      "<<<< <<<<",
+      "Ġlenti viral",
+      "othel ioma",
+      "Ġted ious",
+      "ĠB CI",
+      "Ġge ologic",
+      "Ġconsum es",
+      "ĠAbl ation",
+      "le ast",
+      "Ġth igh",
+      "Ġsec recy",
+      "cover ing",
+      "e iro",
+      "Ã µ",
+      "ĠT BS",
+      "Ġis omerase",
+      "Ġrecomm ends",
+      "ĠVor tex",
+      "ĠB ray",
+      "Ġsub d",
+      "ĠOp tions",
+      "Ġmetam aterial",
+      "ĠSqu ares",
+      "t rap",
+      "im on",
+      "Ġhe sit",
+      "Ġab c",
+      "cess ing",
+      "ĠRE T",
+      "Ġpin ned",
+      "Ġket ones",
+      "Ġweld ed",
+      "ĠMitochond ria",
+      "Ġing ested",
+      "ĠQ FT",
+      "Ġcompar ator",
+      "Ġoxid oreductase",
+      "Ġtet rad",
+      "ĠSens itive",
+      "Ġcatch ments",
+      "Ġrefuge es",
+      "Ġpuber ty",
+      "A rab",
+      "Ġinter annual",
+      "sc attered",
+      "ĠMet am",
+      "Ġcycl ization",
+      "pert ures",
+      "ĠLIN C",
+      "r ules",
+      "ĠP ont",
+      "PT H",
+      "ĉĉĉĉ ĉĉĉĉ",
+      "S anta",
+      "ĠL NC",
+      "Ġsub modular",
+      "rec tive",
+      "Ġtr if",
+      "Ġsent inel",
+      "ĠTw in",
+      "kelet ons",
+      "m iral",
+      "am ing",
+      "ĠG ay",
+      "Ġinter specific",
+      "Ġrel ieve",
+      "Ġend omorphism",
+      "ĠExp anding",
+      "ĠRun time",
+      "y ang",
+      "re quires",
+      "od ine",
+      "omet abolic",
+      "St ore",
+      "plan et",
+      "Ġre nov",
+      "__ _",
+      "aden osine",
+      "u itive",
+      "Ġk el",
+      "ĠPro long",
+      "ĠAd vance",
+      "Ġantimicrobial s",
+      "ĠMunic ipal",
+      "ĠNeutroph il",
+      "F As",
+      "ĠF ame",
+      "ib us",
+      "ET E",
+      "Ġstep ping",
+      "ĠBl ot",
+      "ĠLa ura",
+      "Ġrock y",
+      "ĠLim a",
+      "Ġmitig ated",
+      "ĠLam bert",
+      "Ġunexpl ored",
+      "Ġtrigon ometric",
+      "p ig",
+      "ĠH eli",
+      "Ġfin ely",
+      "Ġoxid izing",
+      "Ġcolon oscopy",
+      "activ ities",
+      "ĠE asy",
+      "Ġunexpl ained",
+      "ak y",
+      "AS M",
+      "work er",
+      "ĠCr ist",
+      "ãĢ ģ",
+      "ul k",
+      "ĠS ugg",
+      "ĠM im",
+      "Ġiter ates",
+      "Ġsulf oxide",
+      "gluc an",
+      "Ġreact ant",
+      "Ġphag ocytic",
+      "B rain",
+      "uc ted",
+      "ĠSc and",
+      "ĠCa CO",
+      "Ġaffili ation",
+      "Pol icy",
+      "ĠInfant ry",
+      "F unctional",
+      "r times",
+      "Ġw ond",
+      "ard ment",
+      "ĠWe il",
+      "Ġdirect ors",
+      "uff ix",
+      "ĠRu iz",
+      "ĠPhenomen a",
+      "Ġmicro b",
+      "cos m",
+      "Ġutil isation",
+      "pers ed",
+      "Ġcon sole",
+      "tic ulate",
+      "Ġdes ens",
+      "Ġreplic as",
+      "Ġpluripot ency",
+      "ĠUk rainian",
+      "Ġhydroly zed",
+      "ĠBiod iversity",
+      "E fficient",
+      "ĠK ash",
+      "min or",
+      "Ġconcl usive",
+      "Ġtent ative",
+      "j ira",
+      "Ġm b",
+      "ĠI PA",
+      "ĠP is",
+      "Ġgover ns",
+      "ĠSouth west",
+      "oe ba",
+      "ĠMoh ammad",
+      "alb umin",
+      "c ircles",
+      "ĠH edge",
+      "ĠAm ph",
+      "B ACK",
+      "O ld",
+      "h istor",
+      "ac ular",
+      "ĠN OR",
+      "hen ius",
+      "vis ions",
+      "miss ibility",
+      "Ġthrombo embolism",
+      "at ized",
+      "Ġw il",
+      "aw ing",
+      "AS I",
+      "Ġheter odimer",
+      "Ġbuff ering",
+      "ĠIde ally",
+      "ĠE gg",
+      "ograph ies",
+      "ĠAp pl",
+      "ĠCI s",
+      "mean ing",
+      "ĠSM AD",
+      "Ġphenyl alanine",
+      "ĠTit anium",
+      "ĠZar iski",
+      "Ġn ymph",
+      "Ġh ired",
+      "ĠP PC",
+      "ĠK G",
+      "ĠGu ill",
+      "ogly cans",
+      "er ial",
+      "D ele",
+      "il us",
+      "ĠF itness",
+      "Ġwh ales",
+      "gr ant",
+      "most ly",
+      "Ġclim ates",
+      "ĠCamp aign",
+      "Mg O",
+      "Ġepist emic",
+      "L ipschitz",
+      "ĠL AT",
+      "Ġcl adding",
+      "vac uum",
+      "agglut inin",
+      "k ill",
+      "Ġs ail",
+      "Ġar tistic",
+      "ans w",
+      "ĠSD F",
+      "ĠKe ith",
+      "Ġsor afenib",
+      "Ġgall bladder",
+      "direct ory",
+      "Ġphotore ceptors",
+      "ĠFok ker",
+      "D U",
+      "Ġed itors",
+      "Ġte lecommun",
+      "ardi a",
+      "ĠPublic ations",
+      "Ġscrew s",
+      "ĠMathem atica",
+      "R SV",
+      "ĠAp ply",
+      "ĠST S",
+      "ĠMur ine",
+      "Ġd ump",
+      "Ġl ingu",
+      "ĠD ixon",
+      "Ġover comes",
+      "ĠPre operative",
+      "Ġmig rant",
+      "Ġbelie ves",
+      "B K",
+      "ac tively",
+      "ĠI SC",
+      "qu as",
+      "Ġal ga",
+      "ich ael",
+      "Ġdis asters",
+      "Ġprac ticed",
+      "hydro phobic",
+      "ĠNi ño",
+      "ĠEth anol",
+      "Q E",
+      "ĠS J",
+      "ĠD engue",
+      "Ġap pl",
+      "ĠY oon",
+      "enz o",
+      "IF Y",
+      "Ġchron ological",
+      "er in",
+      "ĠP eg",
+      "ĠRe levant",
+      "Ġqual ification",
+      "ev ine",
+      "Ġdend rite",
+      "DT D",
+      "chol inesterase",
+      "w atch",
+      "ĠS anchez",
+      "Ġwas hes",
+      "Ġper mafrost",
+      "ĠTer tiary",
+      "Ġsynthes izing",
+      "Ġexped ition",
+      "rout ine",
+      "ĠSear ching",
+      "ĠS é",
+      "res idual",
+      "ĠL CD",
+      "enti ties",
+      "Ġend ovascular",
+      "Ġparam ount",
+      "p her",
+      "Ġstraightforward ly",
+      "Ġvas odil",
+      "ĠSchist osoma",
+      "Ġper missions",
+      "cent red",
+      "Ġfr ustrated",
+      "struct uring",
+      "ĠSch l",
+      "ĠIniti ation",
+      "Ġcu ticle",
+      "Ġforget ting",
+      "ĠS as",
+      "ĠS ult",
+      "un o",
+      "Ġdis integration",
+      "ĠV G",
+      "Ġw ards",
+      "ĠI RE",
+      "up ro",
+      "Ġsub gen",
+      "Ġsub classes",
+      "ĠSt and",
+      "ĠHe ight",
+      "inter pretation",
+      "Ġgly can",
+      "ĠSol vent",
+      "ĠMal ignant",
+      "Ġuns uitable",
+      "ĠCox eter",
+      "Ġspermat ogenesis",
+      "Ġful lerene",
+      "F ox",
+      "S OC",
+      "w et",
+      "arm stadt",
+      "Ġprop ofol",
+      "index ed",
+      "Ġsn akes",
+      "Ed it",
+      "ĠmJ y",
+      "R IB",
+      "Ġe y",
+      "ĠAl kal",
+      "Ġtri axial",
+      "PS K",
+      "ne o",
+      "Ġend o",
+      "Ġglycos ides",
+      "Ġsyll ables",
+      "Ġs orghum",
+      "lo or",
+      "Ġge othermal",
+      "gu inal",
+      "ĠSerb ia",
+      "æ ĸ",
+      "ĠS entinel",
+      "igh ters",
+      "Ġkey board",
+      "Ġban ana",
+      "gran ular",
+      "Ġdecid uous",
+      "ĠH AR",
+      "ne uron",
+      "ĠCar n",
+      "Ġburn s",
+      "Bo ost",
+      "ĠDetermin istic",
+      "p ipe",
+      "ĠF AD",
+      "ĠB ovine",
+      "ĠR ou",
+      "Ġk an",
+      "aut onomous",
+      "utri ents",
+      "Ġhypoth yroidism",
+      "ĠSIN R",
+      "st ret",
+      "Ġun altered",
+      "ĠZ ika",
+      "val ley",
+      "Ġlong itudinally",
+      "Ġfluores cein",
+      "cat heter",
+      "ĠCong enital",
+      "Ġpie z",
+      "Ġabbrevi ated",
+      "ĠChlam ydia",
+      "Ġa ired",
+      "Ġqu een",
+      "Ġinstruc tive",
+      "Ġabrupt ly",
+      "Ġrecur rences",
+      "I MP",
+      "Ġex osome",
+      "ĠH SCs",
+      "Wr iter",
+      "el is",
+      "ĠAr ithmetic",
+      "enari os",
+      "Ġlig ated",
+      "ĠLocal ized",
+      "ĠFre eman",
+      "Ġcarn iv",
+      "ĠC ereb",
+      "Ġg rac",
+      "ĠG ond",
+      "ĠV ancouver",
+      "ob ox",
+      "Ġtyp ed",
+      "ĠÄ ¥",
+      "Up on",
+      "F uture",
+      "EN G",
+      "de ad",
+      "Ġser pent",
+      "ĠAss ignment",
+      "ĠUp dated",
+      "Ġhistor ian",
+      "Ġtroposp heric",
+      "C loud",
+      "b umin",
+      "ĠP ras",
+      "ĠB asket",
+      "ĠâĪĴ âĪĴ",
+      "benz odi",
+      "ĠTra uma",
+      "ĠBehavi ors",
+      "Ġp ter",
+      "ir radiation",
+      "Ġsp oke",
+      "ari atric",
+      "Ġpl ugin",
+      "Ġsuper sonic",
+      "Ġdoc etaxel",
+      "itig ation",
+      "Ġdiges tibility",
+      "n em",
+      "Ġp b",
+      "ĠC SR",
+      "Ġfo uling",
+      "Ġrhe ology",
+      "Ġflood s",
+      "Ġglu ing",
+      "agasc ar",
+      "j ets",
+      "p ti",
+      "est on",
+      "ĠK ü",
+      "Ġopen ings",
+      "Ġisol ating",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠĠĠ",
+      "Ġsemicon ducting",
+      "r ative",
+      "ec ology",
+      "ur ization",
+      "Ġmulti factorial",
+      "sh adow",
+      "Ġcross linked",
+      "Ġphyl a",
+      "Ġprem ises",
+      "ĠLO W",
+      "general ized",
+      "ĠPolynomial s",
+      "Ġb ismuth",
+      "ĠR oz",
+      "ĠDec oding",
+      "ĠClass ifier",
+      "conduc ting",
+      "Ġlit term",
+      "M ann",
+      "Ġf ant",
+      "ĠC Z",
+      "ĠP SNR",
+      "Ġstar ring",
+      "ĠPol yg",
+      "ĠHol m",
+      "r g",
+      "ad ditional",
+      "gu an",
+      "prof essional",
+      "Ġin quiry",
+      "ĠP g",
+      "ĠSch mid",
+      "Ġhead ed",
+      "cha ft",
+      "ĠExp and",
+      "Ġcompan ions",
+      "V an",
+      "ĠS ie",
+      "Ġcan als",
+      "ored oxin",
+      "Ġcoll iding",
+      "abs olute",
+      "ĠPhot os",
+      "ĠLeg acy",
+      "Ġrevascular ization",
+      "ĠP SM",
+      "Ġexp enses",
+      "IS MA",
+      "inter vals",
+      "Ġmultic ellular",
+      "Ġnons m",
+      "Ġresemb lance",
+      "H ep",
+      "Ġw ool",
+      "Ġn iger",
+      "ess a",
+      "asc i",
+      "Ġrot ates",
+      "Ġcompe titions",
+      "Ġarri vals",
+      "Ġlute in",
+      "Ġscholar ship",
+      "F ran",
+      "Ġre used",
+      "ĠEqu ivalence",
+      "ĠGL UT",
+      "grad ing",
+      "sal t",
+      "Ġcommens al",
+      "Ġfra ud",
+      "ox ib",
+      "Ġgastro enter",
+      "Ġrain y",
+      "Ġasser ts",
+      "Oper ation",
+      "Ġflatten ing",
+      "P ut",
+      "X B",
+      "Ġp M",
+      "Ġcon ic",
+      "ob tain",
+      "ĠRo ber",
+      "N ovember",
+      "ĠJ P",
+      "Ġfe brile",
+      "ĠBar riers",
+      "================================ ================================",
+      "Ġhemic ell",
+      "ĠS CS",
+      "ĠN em",
+      "Ġr aster",
+      "cl ude",
+      "Ġïģ ¦",
+      "ĠElli ott",
+      "b order",
+      "Ġd ÏĨ",
+      "rib ose",
+      "ĠEn v",
+      "ĠDiff use",
+      "ĠSupers ymmetry",
+      "Pear son",
+      "F ETs",
+      "y ah",
+      "ul ia",
+      "ĠD warf",
+      "ĠH ull",
+      "ĠAtt ribution",
+      "Ġrepos itories",
+      "ĠGN SS",
+      "ĠV ectors",
+      "Ġsuccess es",
+      "ĠMan hattan",
+      "umb ent",
+      "dig it",
+      "Ġcircumf erential",
+      "B etween",
+      "D eg",
+      "o ue",
+      "Ð ¹",
+      "ĠD ere",
+      "ĠR f",
+      "Ġr ide",
+      "ĠV oc",
+      "Ġprot est",
+      "Ġpur pos",
+      "ĠProof s",
+      "names e",
+      "Ġbank ing",
+      "ĠGastro intestinal",
+      "ĠU nt",
+      "Ġwhen ce",
+      "ĠY ue",
+      "ĠRe habilitation",
+      "Ġexchang ing",
+      "ĠACT H",
+      "Ġc apping",
+      "am ido",
+      "ĠB ap",
+      "Ġpl at",
+      "to String",
+      "Ġelectro encephal",
+      "Ġelectrosp un",
+      "M pc",
+      "j ord",
+      "on v",
+      "Ġc raft",
+      "ĠC Cl",
+      "ĠSt rip",
+      "Ġmed itation",
+      "oxid ative",
+      "ĠRed uce",
+      "ĠCommon wealth",
+      "Ġrif amp",
+      "F lu",
+      "Ġre analysis",
+      "ot rich",
+      "ĠE SA",
+      "Ġj th",
+      "hel in",
+      "ĠGen otype",
+      "Ġdiagonal ization",
+      "ĠGab riel",
+      "Ġquarant ine",
+      "ĠC rab",
+      "ĠD ict",
+      "acc umulation",
+      "be k",
+      "ĠDiff erentially",
+      "Ġlac tis",
+      "tetrahydro furan",
+      "l aser",
+      "ĠU m",
+      "Ġme ga",
+      "rm e",
+      "ĠInd ians",
+      "ĠLeon ard",
+      "Ġcommod ity",
+      "Ġfumig atus",
+      "i ou",
+      "ĠE chin",
+      "ost ream",
+      "Ġmemb ran",
+      "sim ulations",
+      "back end",
+      "ĠOB JECT",
+      "g iving",
+      "Å Ļ",
+      "Ġinf ective",
+      "Al g",
+      "ĠHu h",
+      "ĠMI CR",
+      "Ġfollow ers",
+      "fer ro",
+      "Ġcyan ide",
+      "P resent",
+      "Ġ END",
+      "ĠM Cs",
+      "Ġtim eline",
+      "ĠEmbry onic",
+      "Identif ier",
+      "Ġincon clusive",
+      "ĠGamm aproteobacteria",
+      "n ets",
+      "ĠHe ating",
+      "ank ar",
+      "th r",
+      "ĠK IT",
+      "ĠCh ip",
+      "Ġbl ob",
+      "Ġcalc ulator",
+      "Ġtext ural",
+      "Ġalloy ing",
+      "Ap plication",
+      "ĠProte omic",
+      "Ġantidepress ants",
+      "ur k",
+      "Ġcrystall ography",
+      "Ġcred its",
+      "Ġmuss els",
+      "T om",
+      "ĠF ST",
+      "ĠF old",
+      "ĠH ew",
+      "An n",
+      "bro ok",
+      "Ġglycol ytic",
+      "Tor ch",
+      "Ġv m",
+      "ĠM are",
+      "ĠJ y",
+      "Ġhetero junction",
+      "ĠBorrel ia",
+      "R isk",
+      "ĠN aturally",
+      "Ġsupp lying",
+      "sign ature",
+      "l k",
+      "Ġa rachid",
+      "ol ov",
+      "ĠS ok",
+      "ĠH ö",
+      "ĠR az",
+      "ĠV ander",
+      "Ġdel ib",
+      "Ġmy th",
+      "Ġmid brain",
+      "Ġdece ased",
+      "ĠS CO",
+      "ĠTh romb",
+      "Ġcur r",
+      "Ġsum mit",
+      "mi RNAs",
+      "dimethyl amino",
+      "Ġphotoc atalyst",
+      "verb ose",
+      "gom ery",
+      "Ġw ed",
+      "ĠM ate",
+      "Ġsign i",
+      "rastruct ures",
+      "Ġrecipro city",
+      "b ner",
+      "m ast",
+      "n eck",
+      "Ġco ins",
+      "ĠHist ogram",
+      "cr it",
+      "Bbb k",
+      "A W",
+      "t own",
+      "dis placement",
+      "ĠNe ph",
+      "separ able",
+      "Ġdiast ere",
+      "ĠMODEL S",
+      "Dep th",
+      "ĠNeisser ia",
+      "p dev",
+      "u vial",
+      "ĠB MS",
+      "ĠD ennis",
+      "Ġr p",
+      "Ġnan ometer",
+      "roc yt",
+      "ĠRoman ian",
+      "Ġconce ivable",
+      "C OS",
+      "al veolar",
+      "as tig",
+      "ab we",
+      "enc ode",
+      "rol actone",
+      "Ġread mission",
+      "inters ection",
+      "Ġamplic ons",
+      "tim ulated",
+      "Ġcoll apses",
+      "ochrom atin",
+      "H aw",
+      "ect rum",
+      "ft ype",
+      "ric a",
+      "Ġam id",
+      "MP O",
+      "ĠExt ensions",
+      "Ġvar ic",
+      "Ġdimin ishes",
+      "Ġcathe ters",
+      "N odes",
+      "Ġb box",
+      "em ination",
+      "Ġts unami",
+      "diagn osis",
+      "c od",
+      "q r",
+      "ĠF en",
+      "Ġworth y",
+      "Ġâĩ IJ",
+      "inform atic",
+      "ograp her",
+      "Ġundet ected",
+      "ĠN CAA",
+      "Ġcarcin ogenic",
+      "R U",
+      "Ġan eu",
+      "plit udes",
+      "ke eper",
+      "ĠÄ ģ",
+      "Ġau tistic",
+      "Ġcomprom ising",
+      "Ġunim odal",
+      "Ġr umin",
+      "ap a",
+      "Ġint olerance",
+      "Ġdirec ting",
+      "Ġpe a",
+      "Ġcomm enced",
+      "Ġshadow ing",
+      "C enter",
+      "Ġcl ad",
+      "Ġbl ues",
+      "bin its",
+      "Ġmis classification",
+      "ĠFA ST",
+      "W at",
+      "Ġm Cherry",
+      "Ġb rig",
+      "est radiol",
+      "Ġwave functions",
+      "Ġblo oms",
+      "Ġacc ent",
+      "aj i",
+      "occ urring",
+      "ar rest",
+      "Ġspecial ty",
+      "Ġuncon ditional",
+      "Ġspong es",
+      "Ġdys functional",
+      "ĠNO X",
+      "Ġultrac old",
+      "Ġmartens ite",
+      "O US",
+      "n ier",
+      "is ic",
+      "ĠMat sum",
+      "Ġleuk emic",
+      "ĠBrad ley",
+      "D ensity",
+      "ĠS emiconductor",
+      "ĠC ause",
+      "ĠIn set",
+      "ĠK em",
+      "ĠU PR",
+      "par a",
+      "ech st",
+      "ym et",
+      "Ġag ro",
+      "ĠY Y",
+      "ĠReg eneration",
+      "Ġancest ors",
+      "ĠTiss ues",
+      "Ġsulfur ic",
+      "k d",
+      "Ġl asing",
+      "ĠP up",
+      "ae i",
+      "Ġmamm al",
+      "ĠBrad ford",
+      "Ġsegreg ated",
+      "is olated",
+      "ĠC uba",
+      "Ġblock age",
+      "Ġseam less",
+      "Ġperoxis ome",
+      "h ui",
+      "Ġin aug",
+      "Ġinf ecting",
+      "ĠCh ampion",
+      "ĠAt titudes",
+      "calc ulate",
+      "Ġt ighter",
+      "ĠS AC",
+      "ĠE pi",
+      "Ġat m",
+      "Ġphys ico",
+      "Ġn th",
+      "ĠC anyon",
+      "Ġser oprevalence",
+      "Ġhom o",
+      "ĠUnivers it",
+      "Eval uation",
+      "ĠAPO E",
+      "j ob",
+      "Ġm K",
+      "Ġre ign",
+      "ab o",
+      "ĠR ugby",
+      "ĠN ets",
+      "Ġr ituximab",
+      "ati veness",
+      "Ġph y",
+      "orn is",
+      "Ġfeedback s",
+      "Un ited",
+      "Pr inc",
+      "imb abwe",
+      "ĠGir ls",
+      "Ġunavoid able",
+      "ĠSeman tics",
+      "B reak",
+      "F ISH",
+      "M ix",
+      "Ġn x",
+      "ĠBa o",
+      "dimethyl phenyl",
+      "ĠT OF",
+      "ĠC rown",
+      "ĠG GA",
+      "ĠJ H",
+      "Ġsuper string",
+      "ĠCR Y",
+      "Ġkind ly",
+      "Y N",
+      "Ġund oped",
+      "ex cluding",
+      "ĠLe o",
+      "ĠPROP ERT",
+      "peritone ally",
+      "m ant",
+      "ê °",
+      "Ġf ranch",
+      "ĠPro st",
+      "DE s",
+      "Ġcot rans",
+      "Ġr k",
+      "Ġgeneral izability",
+      "Aut hor",
+      "ĠAnd rea",
+      "ĠConf ocal",
+      "ĠAdi pose",
+      "î Ĺ",
+      "er jee",
+      "Ġan imated",
+      "ĠF ad",
+      "ĠCor rosion",
+      "ĠCirc adian",
+      "Ġacceler ators",
+      "ĠArk ansas",
+      "Ġm ars",
+      "ĠC uc",
+      "ĠInter faces",
+      "Ġretrie vals",
+      "Ġmelan in",
+      "Ġss DNA",
+      "vast ava",
+      "Ġallerg ens",
+      "b ud",
+      "Ġin accessible",
+      "ic tions",
+      "ĠM ood",
+      "ind a",
+      "Ġam eric",
+      "Ġsym biosis",
+      "bers ome",
+      "occ ur",
+      "ĠMarc us",
+      "ĠSuperconduc tivity",
+      "ĠC ort",
+      "ĠH MS",
+      "Ġph ased",
+      "ĠJ ess",
+      "Ġprop ulsion",
+      "ext ract",
+      "Ġsuccin ate",
+      "ĠÖ Ĵ",
+      "ink el",
+      "Ġsil ence",
+      "ĠSU V",
+      "Ġconstitu ency",
+      "Ġbacteri ophage",
+      "g em",
+      "ĠM CL",
+      "ore ne",
+      "ĠG oss",
+      "IC D",
+      "Ġglut amic",
+      "Ġcoex isting",
+      "STE MS",
+      "opot ential",
+      "ĠE y",
+      "ĠL ecture",
+      "ell ae",
+      "Ġimmun oprec",
+      "Ġtim ber",
+      "ĠVul ner",
+      "Ġa roma",
+      "Ġs ands",
+      "ĠSp an",
+      "Ġher n",
+      "Ġincub ating",
+      "Ġtransmit ters",
+      "ĠHom ogeneous",
+      "ĠConstruct ing",
+      "d it",
+      "Ġt c",
+      "al ass",
+      "Ġst ents",
+      "ĠM ID",
+      "Ġan oxic",
+      "Ġprov isions",
+      "ĠCap ac",
+      "neut ron",
+      "ĠVO Cs",
+      "Jan uary",
+      "V AS",
+      "on ce",
+      "ĠC ache",
+      "op ulation",
+      "ĠV TE",
+      "Ġinter phase",
+      "Ġbl og",
+      "ocus ing",
+      "hi ro",
+      "ĠRE C",
+      "Ġanis otropies",
+      "ben ef",
+      "Ġcons tipation",
+      "ĠCan al",
+      "Ġport rait",
+      "sil yl",
+      "ĠLink ed",
+      "ĠBow l",
+      "Ġmonop oles",
+      "ĠPere z",
+      "W IN",
+      "ĠT AP",
+      "Ġr uthenium",
+      "ĠAd herence",
+      "ĠEn zymatic",
+      "Ġspecific ities",
+      "Ġsk i",
+      "ĠC ST",
+      "Ġpo etry",
+      "AT ES",
+      "ram a",
+      "lo res",
+      "AL U",
+      "Ġvas oconstr",
+      "Ġgranul ocyte",
+      "ib i",
+      "Ġop ts",
+      "aves drop",
+      "ept in",
+      "· ·",
+      "ĠJe ong",
+      "Ġmedull ary",
+      "ĠDemonstr ation",
+      "ĠF IB",
+      "ĠB RD",
+      "ĠV V",
+      "Ġall o",
+      "R ule",
+      "T f",
+      "Ġun realistic",
+      "Ġlat itudinal",
+      "RO P",
+      "ĠCorrel ates",
+      "I U",
+      "ĠP ore",
+      "oc rit",
+      "ĠK all",
+      "Ġchar coal",
+      "ĠMong olia",
+      "âĪ ħ",
+      "ĠEn tity",
+      "Ġgram s",
+      "g raphene",
+      "m ine",
+      "ent ric",
+      "ĠP p",
+      "ĠW elfare",
+      "ĠJ ets",
+      "Ġaff irm",
+      "ĠBel le",
+      "ĠStrateg ic",
+      "API ENTR",
+      "K H",
+      "rm ann",
+      "Ġassoci ating",
+      "ĠSur viv",
+      "Ġnicot inic",
+      "ĠWL AN",
+      "Ð ¿",
+      "Ġt ears",
+      "ĠRe vised",
+      "Ġphosph odies",
+      "Ġhors eradish",
+      "ĠL AR",
+      "to ok",
+      "ĠDes cent",
+      "ĠNO x",
+      "ĠStein er",
+      "ĠPerm ian",
+      "ĠVenez uela",
+      "Ġdesic cation",
+      "D IS",
+      "ĠM SP",
+      "Ġpo pl",
+      "rel s",
+      "Ġï£ ½",
+      "Ġlear nt",
+      "ĠBi ofilm",
+      "ĠPC NA",
+      "ĠAtt ribute",
+      "ĠGro thendieck",
+      "ĠAdoles cent",
+      "n v",
+      "st derr",
+      "obal t",
+      "ĠYam amoto",
+      "Ġaliqu ot",
+      "r ater",
+      "ĠO re",
+      "ĠK IR",
+      "ack er",
+      "Ġïĥ »",
+      "Ġstrat osphere",
+      "ĠC ust",
+      "resp ect",
+      "Ġglut amatergic",
+      "Ġencour ages",
+      "c tic",
+      "it ched",
+      "ph ins",
+      "Ġsub urb",
+      "Ġhome omorphic",
+      "hex ah",
+      "Ġmini atur",
+      "C AN",
+      "a head",
+      "ĠB LE",
+      "ĠR BF",
+      "Ġac utely",
+      "Ġï£ ¾",
+      "Ġanten n",
+      "UR N",
+      "ĠGir l",
+      "Ġbiore actor",
+      "ĠLeib niz",
+      "Ġv ial",
+      "ĠL ich",
+      "ob ac",
+      "ĠWhen ever",
+      "inhib ition",
+      "C ast",
+      "Ġstrip ped",
+      "ĠAst rophysics",
+      "pres ence",
+      "ĠFlo er",
+      "ipot ent",
+      "dichlor o",
+      "C LE",
+      "f inger",
+      "on ates",
+      "st ri",
+      "ĠS perm",
+      "ĠD BS",
+      "op eptide",
+      "se paration",
+      "ath ing",
+      "math p",
+      "ou ples",
+      "Ġent ropic",
+      "Ġsw ollen",
+      "Ġdon ated",
+      "Ġsettle ments",
+      "oven ous",
+      "P erm",
+      "ĠS ard",
+      "eg en",
+      "ĠAl ph",
+      "ĠCo operation",
+      "ĠPD AC",
+      "F inal",
+      "l apse",
+      "Ġre vol",
+      "ĠI x",
+      "ĠL ens",
+      "Ġk th",
+      "rel axation",
+      "Cl O",
+      "ichlor o",
+      "Ġwrap per",
+      "ĠSimultaneous ly",
+      "Comput e",
+      "ë Ĭ",
+      "im plantation",
+      "ĠV LA",
+      "hem e",
+      "ĠMay or",
+      "ĠFac ilit",
+      "Ġb att",
+      "im mer",
+      "Ġcur ated",
+      "Ġconf luent",
+      "gener ational",
+      "star ts",
+      "Ġgranul osa",
+      "arboxyl ate",
+      "ĠRies z",
+      "Ġtext book",
+      "Ġconstit utional",
+      "ĠPe ace",
+      "ĠComm ander",
+      "Ġobsc ured",
+      "v il",
+      "ad dition",
+      "ĠW asserstein",
+      "co ords",
+      "ĠProb es",
+      "Ġdeline ated",
+      "TZ VP",
+      "ĠIN F",
+      "Ġdos ages",
+      "Ġolig omerization",
+      "ĠNAD P",
+      "MK II",
+      "om in",
+      "Ġl hs",
+      "ug hen",
+      "ĠJ ong",
+      "anc el",
+      "let ter",
+      "ĠAN C",
+      "F UNCTION",
+      "Ġt ram",
+      "The ir",
+      "ĠGen erated",
+      "Ġpoly cyclic",
+      "Ġcul min",
+      "Ġrect um",
+      "Ġce ft",
+      "Ġmetam aterials",
+      "ĠBiot ech",
+      "Ġmys elf",
+      "Ġun ifying",
+      "Ġem an",
+      "ĠSing er",
+      "triang leright",
+      "om el",
+      "ĠC FA",
+      "oc ha",
+      "ĠG SM",
+      "Ġcent rifuge",
+      "ĠInd o",
+      "Ġtransport ing",
+      "LI B",
+      "Ġoxal ate",
+      "ĠDul becco",
+      "Ġal i",
+      "arg inal",
+      "ho o",
+      "isc hem",
+      "APIENTR YP",
+      "A part",
+      "L DA",
+      "ens ile",
+      "set tings",
+      "Ġep hem",
+      "amp a",
+      "Ġdu plications",
+      "ĠWhe eler",
+      "Phys ical",
+      "ĠCom pletion",
+      "ĠOr dered",
+      "Log ger",
+      "Ġinterf erences",
+      "ĠPoll ution",
+      "Optim al",
+      "S v",
+      "a icin",
+      "Ġp icks",
+      "di versity",
+      "tig ens",
+      "Ġdim orphism",
+      "fe res",
+      "ĠRob otic",
+      "Ġconfirm atory",
+      "Ġcath odic",
+      "Ġspir als",
+      "Ġspr uce",
+      "Lag range",
+      "w at",
+      "ĠAll an",
+      "den ote",
+      "C ID",
+      "al ways",
+      "it he",
+      "ĠCh im",
+      "con ditional",
+      "bar rier",
+      "Ġvisual izing",
+      "Ġïĥ ¹",
+      "Sch midt",
+      "Ġconvention ally",
+      "ĠQU ANT",
+      "GRO UND",
+      "Ġ ug",
+      "ĠC WE",
+      "ĠIn spired",
+      "Ġbu yer",
+      "Ġtherm ost",
+      "Ġkin ematical",
+      "an olic",
+      "Ġd if",
+      "Ġï£ ¼",
+      "ĠGe o",
+      "Ex amples",
+      "cons istency",
+      "ĠPal ace",
+      "ĠVacc ination",
+      "Ġnatri uretic",
+      "Y AG",
+      "ĠCT Cs",
+      "Un ivers",
+      "ĠAcknowledg ement",
+      "memb ered",
+      "v v",
+      "ĠS ession",
+      "Ġinst ar",
+      "ĠLe vin",
+      "AV I",
+      "Ġprolifer ator",
+      "olith s",
+      "ĠTemper atures",
+      "im ming",
+      "ĠTo eplitz",
+      "IC ATIONS",
+      "ĠIntegr als",
+      "Ġsplic ed",
+      "D est",
+      "res ulting",
+      "ĠH ope",
+      "Ġen closure",
+      "ie ves",
+      "fl av",
+      "ĠAbd ul",
+      "Ġleishman iasis",
+      "Ã ²",
+      "os keleton",
+      "Ġad duct",
+      "ĠInflu ences",
+      "E QU",
+      "ĠS itu",
+      "Ġse as",
+      "ĠRe ich",
+      "cy st",
+      "ĠEV OLUTION",
+      "Ġwith stand",
+      "ĠG inzburg",
+      "RNA i",
+      "ĠNon parametric",
+      "ĠPr incess",
+      "Ġintra vascular",
+      "UT IONS",
+      "Ġglut ar",
+      "Ġcoinc ided",
+      "ĠSa ito",
+      "pret rained",
+      "comb ined",
+      "ĠT AM",
+      "Ġalarm s",
+      "Ġcyclo oxygenase",
+      "Ġb n",
+      "Ġpl agi",
+      "Par ticle",
+      "GG G",
+      "e tics",
+      "am ber",
+      "AB STRACT",
+      "ĠExt racts",
+      "ĉĉĉ ĠĠĠĠ",
+      "ĠPhyl ogeny",
+      "t ow",
+      "ĠCon taining",
+      "Ġend onuclease",
+      "inc ubation",
+      "Ġoffic inal",
+      "Ġexplos ions",
+      "lay out",
+      "Ġtouch down",
+      "ĠReve aled",
+      "Ġinfiltr ate",
+      "en ith",
+      "tim ulation",
+      "ĠK ind",
+      "erv ices",
+      "PD A",
+      "Ġcere us",
+      "En v",
+      "Ġlap a",
+      "k amp",
+      "m ult",
+      "ent hal",
+      "ĠGold stone",
+      "si RNA",
+      "stre pt",
+      "Q ual",
+      "m other",
+      "di o",
+      "Ġinf requent",
+      "Ġcycl ospor",
+      "hep atitis",
+      "thromb otic",
+      "G ST",
+      "ĠL j",
+      "ĠU R",
+      "of ect",
+      "ĠAr row",
+      "eth nic",
+      "ĠBarc elona",
+      "C are",
+      "ti tious",
+      "Ġe ta",
+      "Ġvir ions",
+      "sm ash",
+      "ĠâIJ ¤",
+      "Ġa venues",
+      "ob arb",
+      "ĠCom ments",
+      "Ġany way",
+      "af il",
+      "ĠBe a",
+      "ĠBo ys",
+      "ĠAutom ata",
+      "ĠSuperconduc ting",
+      "P ic",
+      "k Hz",
+      "Ġn orepinephrine",
+      "ĠG PC",
+      "Ġunder lined",
+      "bra him",
+      "Ġelectrosp ray",
+      "Ġses qu",
+      "ĠTourn ament",
+      "A ustr",
+      "ĠG rowing",
+      "ĠWe bsite",
+      "LD H",
+      "cov ariance",
+      "sever al",
+      "st abilized",
+      "Ġdec arboxylase",
+      "Ġrem ed",
+      "rho e",
+      "ĠSR S",
+      "ĠTre ated",
+      "ĠMad agascar",
+      "ĠMag ic",
+      "Ġweap on",
+      "ĠYosh ida",
+      "Ġhypogly cemia",
+      "ĠBifid obacterium",
+      "enti tious",
+      ":: :",
+      "ĠSing les",
+      "Ġnic ely",
+      "Ġunexpected ly",
+      "ib les",
+      "ari ae",
+      "Ġcent roids",
+      "Ġbroad ened",
+      "ĠJoh ns",
+      "ĠBacter oid",
+      "Ġfram ing",
+      "Prim ary",
+      "ĠPict ure",
+      "gover nment",
+      "Ġre q",
+      "ĠT ry",
+      "ib o",
+      "Ġliqu ef",
+      "osens itivity",
+      "Ġsla ughter",
+      "ĠD AR",
+      "Ġlog it",
+      "Ġprom ises",
+      "Ġlaw yer",
+      "ĠFP G",
+      "T CP",
+      "Ġinter calation",
+      "ĠBo e",
+      "Ġwide band",
+      "Ġjudg ement",
+      "romagn ets",
+      "Last ly",
+      "ĠIschem ic",
+      "I MA",
+      "f ood",
+      "m uch",
+      "Ġa venue",
+      "Ġschist osomiasis",
+      "ĠExec ution",
+      "D QU",
+      "G IS",
+      "k ines",
+      "ak age",
+      "ech t",
+      "ĠSc aff",
+      "ĠStr ings",
+      "ĠMulti level",
+      "Ġcum bersome",
+      "ĠRay mond",
+      "Ġirregular ities",
+      "ĠAGN s",
+      "ĠMetast atic",
+      "ĠIber ian",
+      "M b",
+      "R NP",
+      "h ong",
+      "is inin",
+      "Ġth irteen",
+      "ĠF AS",
+      "Ġse aling",
+      "Ġap atite",
+      "Ġser ially",
+      "ĠÅ Ŀ",
+      "D EL",
+      "F o",
+      "ĠS oph",
+      "ĠB ear",
+      "ĠJ osh",
+      "rec k",
+      "ull er",
+      "Ġexc ursion",
+      "Ġemb odied",
+      "Ġhybrid ized",
+      "ĠLie utenant",
+      "Per iod",
+      "Ġmoll us",
+      "C VD",
+      "R en",
+      "RE AM",
+      "ĠB ACK",
+      "Ġacc reting",
+      "Ġcult uring",
+      "ĠBur st",
+      "ĠSeg ment",
+      "Ġaster isk",
+      "ĠIde al",
+      "Ġinter tw",
+      "ĠAt oms",
+      "ĠST E",
+      "Ġïģ ª",
+      "Ġremark ed",
+      "Ġhair s",
+      "â ľ",
+      "ĠMet ropolis",
+      "ĠPar tially",
+      "ĠObs erver",
+      "Ġhemat ologic",
+      "obil ization",
+      "ĠBerg man",
+      "Ġcart esian",
+      "Ġclath rin",
+      "ĠS ung",
+      "Ġr ation",
+      "Ġsc oliosis",
+      "oh l",
+      "mut ant",
+      "NN s",
+      "ĠRah man",
+      "ĠSpati ally",
+      "P IP",
+      "Y b",
+      "Ġd iaz",
+      "ver tebral",
+      "ad zu",
+      "als ki",
+      "ans wer",
+      "Ġge ochemistry",
+      "Ġstem ming",
+      "w es",
+      "ox ys",
+      "Ġmat s",
+      "ev a",
+      "ĠHyper bolic",
+      "arb age",
+      "Ġclip ping",
+      "ĠSug ar",
+      "ĠC ognition",
+      "ĠD IV",
+      "Ġtem pt",
+      "ĠPath ogen",
+      "ĠPed ro",
+      "Ġw ak",
+      "ent ries",
+      "ĠG CM",
+      "pro jective",
+      "Ġprof iciency",
+      "ĠKn own",
+      "Ġlex icon",
+      "ĠMend elian",
+      "Ġzoon otic",
+      "le ans",
+      "ĠT alk",
+      "Ġk urtosis",
+      "NA S",
+      "ĠNow adays",
+      "ĠL il",
+      "ĠW MAP",
+      "Ġdis perse",
+      "Ġcoll oids",
+      "eb ra",
+      "OM ET",
+      "ĠD CT",
+      "ĠR ise",
+      "Ġinter genic",
+      "GT H",
+      "Ġtap ered",
+      "Mark ovian",
+      "Prot ocol",
+      "ĠVeget ation",
+      "r ats",
+      "Ġd ivalent",
+      "ĠCr ust",
+      "zy g",
+      "Ġpig mentation",
+      "grad uate",
+      "ĠRic c",
+      "Ġcounterex ample",
+      "Ġs ativ",
+      "Ġl s",
+      "ĠCirc ulation",
+      "is otropic",
+      "ĠEN SO",
+      "Ġtrop onin",
+      "Ġdissol ving",
+      "Ġcosme tic",
+      "H f",
+      "f urther",
+      "Ġp anc",
+      "Ġh ops",
+      "int ra",
+      "ĠZ he",
+      "ĠRel iable",
+      "ivol umab",
+      "M X",
+      "R ab",
+      "ĠP ES",
+      "ĠB ü",
+      "Ġad hered",
+      "Ġflu ency",
+      "ĠCl aus",
+      "Ġdel amination",
+      "Ġgu anine",
+      "ĠMulti scale",
+      "ĠEqu ip",
+      "ĠIll ustr",
+      "Ġtetra hydro",
+      "f el",
+      "l ists",
+      "Î ŀ",
+      "em ulsion",
+      "ĠN Z",
+      "Ġwas n",
+      "ai ra",
+      "Ġarg uing",
+      "mi RNA",
+      "ĠExp ressed",
+      "Ġspectrophot ometric",
+      "Ġile um",
+      "Ġflam es",
+      "F it",
+      "G on",
+      "ĠC ulex",
+      "Ġun weighted",
+      "Ġnan ob",
+      "SH V",
+      "Ġalign ing",
+      "Ġshut tle",
+      "Ġchloro quine",
+      "Ġpyr ite",
+      "ĠR ica",
+      "Ġr ift",
+      "Ġcathe psin",
+      "ĠPROC ESS",
+      "P f",
+      "R aw",
+      "ray fish",
+      "SA L",
+      "coll apse",
+      "........ ........",
+      "at ases",
+      "Ġwork shops",
+      "oph ile",
+      "ĠâĬ ĥ",
+      "Ġbifurc ations",
+      "T race",
+      "Ġp ause",
+      "Ġorbit ing",
+      "oli ubov",
+      "ĠCur tis",
+      "ĠRevis iting",
+      "ore t",
+      "Ġinf used",
+      "lu ents",
+      "Ġplas tid",
+      "Ġïģ ¹",
+      "Ġexec utions",
+      "ĠGra ves",
+      "loc ally",
+      "ĠAtmosp here",
+      "diab etes",
+      "ĠPrad esh",
+      "ĠCof actor",
+      "is omorphic",
+      "Ġb od",
+      "ĠC BD",
+      "Ġinc ap",
+      "Ġret rovirus",
+      "Ġlip ophilic",
+      "Ġlin oleic",
+      "Ġtrav elled",
+      "c ovalent",
+      "p ick",
+      "u pl",
+      "ĠP ole",
+      "ĠTh ym",
+      "ĠTe ich",
+      "Ġcollabor ators",
+      "Ġinstant ons",
+      "ĠMEG A",
+      "ĠHepat ocellular",
+      "Ġinfest ation",
+      "ĠPie zo",
+      "ĠL ub",
+      "ĠN Cs",
+      "Ġnucle oside",
+      "Ġoste ogenesis",
+      "E igen",
+      "R MSE",
+      "Ġl ax",
+      "ĠK ost",
+      "ĠV ero",
+      "ĠCh ou",
+      "elect rochemical",
+      "Ġcompe ti",
+      "ch ia",
+      "Ġsub module",
+      "ĠAl low",
+      "Ġresol vent",
+      "Ġswe eps",
+      "Ġsupercon formal",
+      "pyrrol idine",
+      "l ofen",
+      "å Ń",
+      "Ġdes erves",
+      "ĠZ imbabwe",
+      "az ines",
+      "ĠCons ult",
+      "Ġcast le",
+      "Ġpharmaceutical s",
+      "Ġparac rine",
+      "Ġjejun i",
+      "Ġargu ably",
+      "Ġe NOS",
+      "Ġher ds",
+      "Ġvehic ular",
+      "Ġtriang ulated",
+      "Ġî µ",
+      "ĠGrand e",
+      "Ġanthocyan ins",
+      "ĠD uan",
+      "ĠV ibration",
+      "Ġtri ad",
+      "Ġhouse keeping",
+      "B or",
+      "Ġp ub",
+      "Ġmal formation",
+      "gluc osamine",
+      "inhib itory",
+      "Dir ac",
+      "ĠC SD",
+      "ĠRot ating",
+      "ĠHTL V",
+      "Ġdem ol",
+      "inf iltr",
+      "Ġhem olytic",
+      "Ġcarb apenem",
+      "Ġlum inescent",
+      "ĠPlan ets",
+      "Ġmell ifera",
+      "Ġcortic osterone",
+      "ĠAdd ress",
+      "Ġhub s",
+      "ometh acin",
+      "å IJ",
+      "ĠCh ampions",
+      "ĠRe vision",
+      "ĠHer bert",
+      "Ġambig uities",
+      "K ERN",
+      "Ġd é",
+      "Ġl p",
+      "Ġen vis",
+      "ĠCh ol",
+      "rop in",
+      "Ġdr one",
+      "m eyer",
+      "Ġis otype",
+      "ĠV u",
+      "ER C",
+      "Ġvers atility",
+      "Sp eed",
+      "Ġae tiology",
+      "Ġgonad otropin",
+      "Ġcogn ate",
+      "ĠCot ton",
+      "reason able",
+      "dis able",
+      "Ġdevast ating",
+      "P ier",
+      "P OL",
+      "ĠB é",
+      "inc ter",
+      "alu able",
+      "Ġpoly hedron",
+      "ĠRel ay",
+      "Ġworkflow s",
+      "F EM",
+      "in p",
+      "Ġm ph",
+      "soft max",
+      "m ur",
+      "v r",
+      "Ġe rent",
+      "ĠK N",
+      "Ġstat in",
+      "Ġflat ness",
+      "ĠArchitect ures",
+      "ĠVeter inary",
+      "Ġnos ocomial",
+      "S k",
+      "X ML",
+      "ĠF os",
+      "ĠL or",
+      "Ġradi ography",
+      "ĠBl um",
+      "ĠDiscrim ination",
+      "Ġp unc",
+      "Ġex its",
+      "ĠB ilateral",
+      "ms strahlung",
+      "Ġcolon ized",
+      "ĠFib rosis",
+      "Ġchaper ones",
+      "abor atory",
+      "ĠPers istence",
+      "Ġlum ped",
+      "Ġrab ies",
+      "ĠBurn s",
+      "D ense",
+      "on tium",
+      "acet ylation",
+      "ĠF ET",
+      "Ġhand ful",
+      "bi ology",
+      "Ġundes ired",
+      "L imit",
+      "ĠN BA",
+      "ĠSe oul",
+      "AP T",
+      "ĠTrans genic",
+      "oxygen ation",
+      "But ton",
+      "ĠTreat ments",
+      "Z V",
+      "is omorphism",
+      "oc ta",
+      "iff e",
+      "ode oxy",
+      "Ġorgan elle",
+      "Ġcoll oid",
+      "Ġcer amide",
+      "Ġtq dm",
+      "G PS",
+      "ĠI SR",
+      "oc linic",
+      "ĠL yme",
+      "Ġep ig",
+      "ĠTra il",
+      "I PS",
+      "Ġs orts",
+      "ĠZ ebrafish",
+      "Ġhydrox ylase",
+      "Sm irnov",
+      "B ax",
+      "ĠD ance",
+      "ĠH ors",
+      "Ġreach ability",
+      "Par allel",
+      "ĠES BL",
+      "Ġupl ink",
+      "Ġpostp randial",
+      "s olar",
+      "it abine",
+      "ord ism",
+      "Ne asy",
+      "Ġaband on",
+      "I MI",
+      "f ake",
+      "st atistical",
+      "ĠC ars",
+      "ib ia",
+      "ĠÃ ĩ",
+      "sp c",
+      "MD P",
+      "tiz ations",
+      "Intern ational",
+      "ular is",
+      "Ġvacu oles",
+      "K C",
+      "ĠA PT",
+      "ĠB t",
+      "ĠB om",
+      "ĠG MP",
+      "Ġpione er",
+      "ĠChair man",
+      "ĠT ucker",
+      "ĠR AF",
+      "ĠN ASH",
+      "ĠW IT",
+      "yn yl",
+      "Ġsup plier",
+      "ans ky",
+      "Ġdecom posing",
+      "ĠUV B",
+      "ophen ol",
+      "Ġb arium",
+      "ĠS MT",
+      "ot ocin",
+      "ly tic",
+      "ran king",
+      "ĠDi rections",
+      "Ġinn ervation",
+      "sw itching",
+      "d ac",
+      "Ġh T",
+      "Ġdoc tr",
+      "ĠIncre mental",
+      "ĠEarthqu ake",
+      "H as",
+      "L ee",
+      "m ates",
+      "pro line",
+      "ĠRE E",
+      "Ġviol ates",
+      "ð x",
+      "Ġhomogen ates",
+      "Bo olean",
+      "Ġd oxycycline",
+      "ĠMO F",
+      "iop hen",
+      "Ġapprec iation",
+      "fin als",
+      "character istic",
+      "ĠContin ental",
+      "B us",
+      "E sc",
+      "X P",
+      "Û Į",
+      "ĠCT A",
+      "Max well",
+      "Ġarchae a",
+      "N ik",
+      "N ONE",
+      "T W",
+      "ter ing",
+      "ĠP erman",
+      "Ġrest ores",
+      "opath ogenic",
+      "ĠMont gomery",
+      "Ġglucocortic oids",
+      "Ġ ud",
+      "ĠN uss",
+      "ĠN é",
+      "ĠSt urm",
+      "Ġatt aching",
+      "Ġintra peritoneally",
+      "las ov",
+      "Ġst ellate",
+      "Ġanti proliferative",
+      "Ġmicro organism",
+      "Ġvis u",
+      "Ġjud ges",
+      "random ized",
+      "allow ed",
+      "Ġdepri ved",
+      "develop ment",
+      "scrib ed",
+      "ethe rian",
+      "ĠFras er",
+      "R am",
+      "b ib",
+      "Ġl iner",
+      "Ġg uns",
+      "res net",
+      "ĠL TR",
+      "ight ing",
+      "In iti",
+      "ĠZ imm",
+      "ĠGe ology",
+      "Ġantioxid ative",
+      "Ġmag enta",
+      "ĠNiger ian",
+      "galax y",
+      "ĠMelan oma",
+      "F ound",
+      "Ġb um",
+      "ĠT rop",
+      "ĠD os",
+      "Ġmet ab",
+      "Ġinv oking",
+      "ĠSch izophrenia",
+      "CF G",
+      "Ġgel ation",
+      "Ġopi oids",
+      "p is",
+      "Ġch urches",
+      "Ġcan onically",
+      "Ġj ug",
+      "Ġaccept ors",
+      "DM EM",
+      "Ġobl iqu",
+      "ĠMedic are",
+      "arpo on",
+      "Z IP",
+      "ore active",
+      "Ġim printing",
+      "ĠV inc",
+      "ĠÂ ¿",
+      "Ġrest art",
+      "Ġdent ate",
+      "en zymatic",
+      "Ġin guinal",
+      "ĠN t",
+      "Ġun observed",
+      "uct uation",
+      "Ġbi asing",
+      "Ġintegr ins",
+      "Ġur l",
+      "FP GAM",
+      "ĠCL UST",
+      "omat ology",
+      "Ġmetallic ities",
+      "Ġintention ally",
+      "FPGAM GR",
+      "T yp",
+      "Ġal ly",
+      "Ġcom ic",
+      "ĠL ions",
+      "Ġim puted",
+      "ĠÃ Ł",
+      "lex ia",
+      "ĠJan us",
+      "Ġbr ass",
+      "ĠDown loaded",
+      "BU FF",
+      "iden tical",
+      "Ġpsychiat ry",
+      "C CT",
+      "if ar",
+      "ĠMand el",
+      "Ġopto electronic",
+      "Ġis omerization",
+      "ĠF ant",
+      "ĠL ion",
+      "ĠL ov",
+      "ĠN af",
+      "est a",
+      "Ġbi ocompatible",
+      "Ġsec retions",
+      "sc i",
+      "ĠRet ro",
+      "ois omerase",
+      "ĠSn ap",
+      "Ġsplitting s",
+      "Ġscav enger",
+      "proced ure",
+      "Daw ley",
+      "ë ĭ¤",
+      "un ate",
+      "ĠD ye",
+      "ĠN EC",
+      "Ġnan ocl",
+      "Ġplan etes",
+      "ĠTR PM",
+      "Ġvo ices",
+      "ĠHierarch y",
+      "m v",
+      "Ġl asts",
+      "Ġh oped",
+      "Ġmed ians",
+      "ĠAnd reev",
+      "Ġheight ened",
+      "ä »",
+      "Ġin definite",
+      "ĠK amp",
+      "ang el",
+      "gr ids",
+      "arch ae",
+      "Ġtherap ists",
+      "ĠMi R",
+      "Ġnegoti ation",
+      "H SP",
+      "ĠC ustom",
+      "Ġst ria",
+      "Ġun acceptable",
+      "ret in",
+      "pen et",
+      "ĠOR R",
+      "ĠLife time",
+      "ĠPhosph ate",
+      "Ġtrop ics",
+      "ĠWel ch",
+      "ĠP yr",
+      "Ġam putation",
+      "ĠAr tin",
+      "ĠCa O",
+      "Ġconject ures",
+      "Ġat rium",
+      "ĠCom plementary",
+      "ĠAl uminum",
+      "Ġmic row",
+      "ili ated",
+      "ĠImmun o",
+      "Ġbin ocular",
+      "ĠWeak ly",
+      "Ġimmun ogenic",
+      "Ġbath ym",
+      "ĠPhen otype",
+      "Ġsial ic",
+      "S ix",
+      "Ġa kin",
+      "ro tor",
+      "hel m",
+      "CC ESS",
+      "Ġneuro protection",
+      "ĠFif th",
+      "Ġconting ent",
+      "Ġsket ched",
+      "I mp",
+      "Ġc ached",
+      "ure ment",
+      "ĠB ic",
+      "ĠK ah",
+      "ber ation",
+      "atter son",
+      "Ġglyc ation",
+      "Ġinvest ors",
+      "Ass isted",
+      "ial es",
+      "sc ience",
+      "Ġpil ots",
+      "us cripts",
+      "MI CS",
+      "Ġorth opedic",
+      "war fs",
+      "gre ater",
+      "ĠArter y",
+      "V ideo",
+      "Ġar range",
+      "av ar",
+      "charg es",
+      "dial dehyde",
+      "ĠT PA",
+      "Ġsp elling",
+      "ĠSe iberg",
+      "Ġnavig ate",
+      "ĠPow der",
+      "ĠR ings",
+      "ĠCh ron",
+      "ĠAt g",
+      "Ġhom ocysteine",
+      "ĠIdentif y",
+      "Ġo ak",
+      "Ġl iability",
+      "Ġoper ands",
+      "ĠCT D",
+      "Ġallevi ates",
+      "m A",
+      "ĠL anger",
+      "Ġsub manifolds",
+      "ĠJ ag",
+      "Ġradi ance",
+      "const ants",
+      "ĠMor occo",
+      "Eng ine",
+      "á ¸",
+      "â Ĥ¬",
+      "re vers",
+      "PC I",
+      "uns queeze",
+      "ocon version",
+      "Ġintens ified",
+      "Ġrefin ements",
+      "ofect amine",
+      "ay as",
+      "Ġinc idental",
+      "ĠTh ur",
+      "Ġover d",
+      "Ġbit ter",
+      "Ġign ores",
+      "а н",
+      "ĠOT U",
+      "Ġs err",
+      "ab y",
+      "ĠG CN",
+      "ĠCons umer",
+      "Ġconc ordant",
+      "ĠMR C",
+      "ĠEcon omy",
+      "satisf ying",
+      "Ġbiotin ylated",
+      "Numer ical",
+      "ĠRash ba",
+      "st ochastic",
+      "ĠL al",
+      "Ġbur dens",
+      "All oc",
+      "ĠGraph ics",
+      "ĠLRR K",
+      "A IC",
+      "ĠT ed",
+      "ĠS ark",
+      "ow l",
+      "Ġhe most",
+      "ĠAn at",
+      "Ġhom ing",
+      "ĠChar lie",
+      "ĠBr uc",
+      "ih ara",
+      "ing en",
+      "ĠV ern",
+      "ĠY ers",
+      "Ġid s",
+      "Ġcirc RNAs",
+      "Ġconduc ive",
+      "ĠBR ST",
+      "Ġgall ium",
+      "Ġdich otomy",
+      "F r",
+      "e tition",
+      "Ġc esarean",
+      "ol an",
+      "Ġr n",
+      "ub stituted",
+      "ĠLe aves",
+      "ĠLe ader",
+      "col oring",
+      "D raw",
+      "Ġser ous",
+      "Er r",
+      "Ġinn ermost",
+      "ĠHam burg",
+      "S tor",
+      "j es",
+      "Ġto l",
+      "id ade",
+      "Ġr v",
+      "ĠIn version",
+      "Ġmulti phase",
+      "Ġpseud or",
+      "ĠGood man",
+      "ĠJS ON",
+      "Ġcorrid or",
+      "Ġp ork",
+      "ĠS ale",
+      "ĠN atal",
+      "Ġattack ing",
+      "ĠShe et",
+      "Ġstream wise",
+      "Ġatom istic",
+      "Ġfirm ly",
+      "ĠAch ie",
+      "Ġp ir",
+      "ĠI KK",
+      "ĠF alk",
+      "ile ptic",
+      "ĠTR PC",
+      "Ġadhes ions",
+      "HR P",
+      "Ġpauc ity",
+      "S plit",
+      "U DI",
+      "ĠS end",
+      "ĠP ine",
+      "ĠL on",
+      "ĠL ost",
+      "ef er",
+      "con caten",
+      "Ġlo yal",
+      "Ġgly cop",
+      "ĠObserv ing",
+      "ĠMoh amed",
+      "Y R",
+      "ĠFil ters",
+      "c as",
+      "p ages",
+      "Ġd A",
+      "Ġare al",
+      "ad is",
+      "ĠL HS",
+      "ĠThere by",
+      "Ġvisual izations",
+      "Ġtw istor",
+      "unit ary",
+      "Ġarch ives",
+      "Ġphenol ics",
+      "h ik",
+      "s son",
+      "ĠI K",
+      "ĠStud ying",
+      "Ġtw isting",
+      "ĠHydro dynamic",
+      "Ġsplit ter",
+      "Ġurothel ial",
+      "Ġal ken",
+      "ĠG PI",
+      "Ġcor tices",
+      "Ġcrop ping",
+      "Pati ent",
+      "ĠChlam yd",
+      "in berg",
+      "ĠA ircraft",
+      "ce le",
+      "ect ral",
+      "Ġconf erences",
+      "Ġcre atine",
+      "al ty",
+      "pro portional",
+      "Ġlept onic",
+      "Ġov ulation",
+      "uer re",
+      "tez omib",
+      "d le",
+      "init eness",
+      "ĠSpecim ens",
+      "Ġcom a",
+      "ine phrine",
+      "Ġep im",
+      "ĠPer cent",
+      "Co O",
+      "ĠLo ading",
+      "Ġven ue",
+      "ĠTN M",
+      "Ġpac emaker",
+      "ĠHoff mann",
+      "T ech",
+      "n ie",
+      "ĠOr leans",
+      "Ġmagnet ron",
+      "Ġhospit ality",
+      "ĠNord ic",
+      "oprol iferative",
+      "Ġundo ubtedly",
+      "ĠS rin",
+      "Ġhum ic",
+      "ĠIntegr ative",
+      "ĠCamp us",
+      "Ġplant arum",
+      "radi ative",
+      "Ġiter ator",
+      "ĠMes ozoic",
+      "AP s",
+      "car inic",
+      "Ġcheck points",
+      "ĠïĤ £",
+      "ĠmA bs",
+      "ĠLiver pool",
+      "ìĿ ´",
+      "ĠEcos ystem",
+      "Ġneovascular ization",
+      "Ġdem oc",
+      "lo ops",
+      "ĠSU RF",
+      "Ġpassiv ation",
+      "Ġconsec utively",
+      "ĠAlfv én",
+      "ĠS SE",
+      "Ġout s",
+      "stim ulation",
+      "Ġphilos ophical",
+      "ĠS ask",
+      "Ġfl akes",
+      "Ġfinger printing",
+      "Ġbuff alo",
+      "ĠWik imedia",
+      "Ġrecons titution",
+      "Ġepithel ia",
+      "on k",
+      "en y",
+      "ĠM Q",
+      "ĠF ork",
+      "end ance",
+      "Ġgeneral isation",
+      "Ġpe oples",
+      "Ġconn ector",
+      "ges ia",
+      "inter ference",
+      "Ġcolor ation",
+      "calc ulation",
+      "ĠAx ial",
+      "ĠDES IGN",
+      "Ġrecess ion",
+      "Ġdissol ve",
+      "ĠPartition ing",
+      "Qx MD",
+      "G ES",
+      "V o",
+      "k har",
+      "ĠE AE",
+      "Ġco arser",
+      "Ġpost traumatic",
+      "Ġsynthesis ed",
+      "sil ica",
+      "tetra hydropy",
+      "ĠPor ter",
+      "v ark",
+      "ent anyl",
+      "Ġcon ve",
+      "Ġra fts",
+      "bre cht",
+      "Ġrectif ier",
+      "Ġo roph",
+      "ĠC EP",
+      "Ġhist ones",
+      "Ġstand point",
+      "Ġanc illary",
+      "ĠHur ricane",
+      "c ro",
+      "Ġre b",
+      "Ġi T",
+      "Ġge ography",
+      "olar ization",
+      "ĠMan aging",
+      "Ġxyl ose",
+      "uther land",
+      "ĠTaq Man",
+      "K N",
+      "Ġt m",
+      "ĠT AS",
+      "ist le",
+      "âĢ «",
+      "Ġmy corrhizal",
+      "ĠTer restrial",
+      "haus en",
+      "observ able",
+      "Bri en",
+      "Ġneutrop enia",
+      "T aken",
+      "ĠS MI",
+      "Ġpol ishing",
+      "Ġphot op",
+      "Ġthermal ization",
+      "Ġpseud oscalar",
+      "ĠDom inic",
+      "romy algia",
+      "Ġechocardi ographic",
+      "Ill umina",
+      "ĠI PC",
+      "ĠH uss",
+      "ess ive",
+      "up take",
+      "Ġweek end",
+      "Ġcorrobor ate",
+      "ĠTas man",
+      "her ty",
+      "Ġper ine",
+      "Ġtrans ports",
+      "Ġgl ance",
+      "ret inal",
+      "Pro to",
+      "igen es",
+      "Ġprohib ited",
+      "behavi oral",
+      "ophe rol",
+      "ë ¡",
+      "ĠN ecess",
+      "ob iology",
+      "ok k",
+      "Ġtra versal",
+      "ĠAnd es",
+      "Res ource",
+      "oli tic",
+      "ç a",
+      "i rie",
+      "arc tan",
+      "Ġmorph ogenetic",
+      "ĠHu i",
+      "loss es",
+      "Ġfulf illing",
+      "Ġhur ricane",
+      "om bo",
+      "Ġg s",
+      "ĠL v",
+      "ĠN erv",
+      "ell osis",
+      "Ġconf ront",
+      "Ġorth ologous",
+      "Ġwet tability",
+      "Ġcyan obacterial",
+      "Ġcass ava",
+      "A UT",
+      "a vi",
+      "h len",
+      "ĠS LA",
+      "Ġcon vol",
+      "Ġinter metallic",
+      "ins ide",
+      "Ġpolar izability",
+      "Ġens uing",
+      "Ġchlor oplasts",
+      "l id",
+      "l ips",
+      "Ġre bound",
+      "ĠC ary",
+      "ĠL ambda",
+      "ĠV iv",
+      "Ġcalc ination",
+      "ĠÌ Ĩ",
+      "Ġcounter factual",
+      "ĠSil ica",
+      "Ref eree",
+      "Ġhomolog ues",
+      "ĠSpati otemporal",
+      "ĠArr henius",
+      "Ġinf lamed",
+      "ĠZ ambia",
+      "Ġanti psychotic",
+      "hel per",
+      "Bl ood",
+      "Ġpurch asing",
+      "ĠSchw inger",
+      "ĠWilk inson",
+      "Ġfain ter",
+      "Ġr ash",
+      "ĠJ ang",
+      "ĠCon ductivity",
+      "rop oda",
+      "ĠSe q",
+      "Ġprop olis",
+      "Ġtub ule",
+      "ĠLie b",
+      "optim ization",
+      "m ounted",
+      "em es",
+      "can ic",
+      "oradi otherapy",
+      "ĠJen kins",
+      "N c",
+      "T ogether",
+      "Ġf ove",
+      "Ġm v",
+      "ĠDef ect",
+      "ä t",
+      "ĠFin ance",
+      "umar in",
+      "mitt ance",
+      "ere l",
+      "ĠF ren",
+      "ĠR hyth",
+      "ram ified",
+      "Ġhyper cholesterolem",
+      "Ġstim ulatory",
+      "ĠRich mond",
+      "Ġadvance ments",
+      "b les",
+      "x u",
+      "all ation",
+      "Ġint ral",
+      "iter pene",
+      "Con cerning",
+      "Ġbul ky",
+      "Ġá ¾±",
+      "comput ation",
+      "ĠAgar wal",
+      "C entral",
+      "X PS",
+      "Ġt alks",
+      "ĠT ap",
+      "im ilar",
+      "ĠN CI",
+      "Ġacc used",
+      "Ġtranscript omes",
+      "Ġprovision ing",
+      "ĠEt OH",
+      "g m",
+      "Ġt id",
+      "ĠP OC",
+      "ff man",
+      "ĠIn er",
+      "ĠU B",
+      "inc ubated",
+      "ĠAt rial",
+      "Ġfour teen",
+      "ĠAstr onomical",
+      "ĠMig uel",
+      "ĠK ov",
+      "Ġsc ipy",
+      "Ġtherm oplastic",
+      "ĠMan uel",
+      "ĠProm otion",
+      "ĠAccess ed",
+      "Ġterr itorial",
+      "in as",
+      "ĠM Ps",
+      "mon itoring",
+      "ĠSim ulating",
+      "Ġpan or",
+      "Ġrhe umatic",
+      "select in",
+      "ĠLap aroscopic",
+      "H LA",
+      "ĠY ale",
+      "sp read",
+      "ET S",
+      "Ġglyc ans",
+      "Ġimmig rant",
+      "D onald",
+      "ĠC ASE",
+      "ĠH II",
+      "gl omer",
+      "Ġïĥ İ",
+      "ĠExper iences",
+      "ĠViet namese",
+      "Hod gkin",
+      "o ader",
+      "he art",
+      "Ġrem edy",
+      "Ġfacilit ators",
+      "open hagen",
+      "d odec",
+      "ĠF riend",
+      "ĠTo uch",
+      "arm s",
+      "CR s",
+      "Ġultra high",
+      "ĠDri ver",
+      "GEM ENTS",
+      "ĠO u",
+      "Ġend ocarditis",
+      "Ġauto encoder",
+      "Ġ ich",
+      "Ġf etch",
+      "ur ian",
+      "ĠOR Fs",
+      "Ġperme abilized",
+      "ĠWi Fi",
+      "ĠLith uan",
+      "Struct ure",
+      "L n",
+      "h ouses",
+      "Ġo ught",
+      "ĠConcl uding",
+      "Ġann iversary",
+      "ĠCre ation",
+      "Ġblind ness",
+      "Ġpc DNA",
+      "ĠSus an",
+      "ĠBenjamin i",
+      "ĠSent ence",
+      "Ġs nd",
+      "Ġf ins",
+      "ph is",
+      "ĠMod ules",
+      "Ġneuro psychiatric",
+      "ĠPot assium",
+      "Ġsacrific e",
+      "Ġdysp nea",
+      "Ġdeliber ately",
+      "omeg aly",
+      "M edia",
+      "T emporal",
+      "Ġsh ark",
+      "SC AN",
+      "split ting",
+      "Ġmis use",
+      "Ġbirefring ence",
+      "ĠÖĴ âĨĴ",
+      "Ġp ier",
+      "Ġn urs",
+      "ĠS clerosis",
+      "ad hy",
+      "Ġund etermined",
+      "Ġcomple mentation",
+      "ĠAff ect",
+      "ĠHam ps",
+      "Ġg ob",
+      "ĠF ate",
+      "ĠH AL",
+      "ĠK iss",
+      "Ġmicro be",
+      "Ġcarbon aceous",
+      "Ġlip osome",
+      "ĠUs age",
+      "Ġquasipar ticles",
+      "Ġc asp",
+      "ĠN arrow",
+      "Ġout look",
+      "ĠCh ord",
+      "Ġclaim ing",
+      "Ġdiver ging",
+      "ĠBio informatics",
+      "ĠPsy chiatric",
+      "ĠMas ters",
+      "Ġll vm",
+      "ĠI QR",
+      "ph ases",
+      "ĠTh y",
+      "erg er",
+      "ĠDi pl",
+      "SF R",
+      "Ġcred ited",
+      "ĠTet ra",
+      "âĭ ¯",
+      "Ġamn iotic",
+      "ĠCharlot te",
+      "C ox",
+      "H ard",
+      "ar ticle",
+      "ĠD EA",
+      "ĠE clipse",
+      "ĠL MP",
+      "Ġim prison",
+      "ĠV arying",
+      "ES Cs",
+      "ĠTHE O",
+      "Ġnerv osa",
+      "Ġpreced es",
+      "Ġgy ro",
+      "ĠWOR DS",
+      "ĠDak ota",
+      "ut ory",
+      "ĠE mer",
+      "ad am",
+      "ĠN ah",
+      "ĠVir go",
+      "Set ting",
+      "P Q",
+      "å ®",
+      "er us",
+      "Ġc ep",
+      "Ġb d",
+      "di er",
+      "Ġim balanced",
+      "Ġtimes tep",
+      "ä n",
+      "ĠRab bit",
+      "Ġham sters",
+      "Ġmedull a",
+      "ĠChromat ography",
+      "IN PUT",
+      "Ġloss y",
+      "P seud",
+      "ĠP BL",
+      "ĠD omestic",
+      "ia u",
+      "anc ell",
+      "Ġmulti layers",
+      "Ġsubs idi",
+      "ĠUtil izing",
+      "t une",
+      "re hend",
+      "ar te",
+      "Ġb urs",
+      "ĠN HE",
+      "Ġclos eness",
+      "ĠCol our",
+      "ĠHom o",
+      "Equ ations",
+      "Ġsut ures",
+      "ac us",
+      "Ġknock ed",
+      "Ġsecret ary",
+      "Ġascer tained",
+      "Ġin patients",
+      "ir ts",
+      "Ġpl ut",
+      "ans son",
+      "ram i",
+      "Ġoste otomy",
+      "ĠPrim ers",
+      "ĠLeg islative",
+      "ĠCardi ology",
+      "Ġadmit ting",
+      "Ġexcav ation",
+      "ĠHedge hog",
+      "W G",
+      "f rozen",
+      "Ġl iber",
+      "ĠI CE",
+      "ch osen",
+      "ĠK ohn",
+      "St op",
+      "Ph il",
+      "phag ia",
+      "ĠB CA",
+      "Ġem pt",
+      "Ġz z",
+      "oper s",
+      "ĠSi xty",
+      "eck man",
+      "Ġtransf errin",
+      "Ġpenal ized",
+      "Be ing",
+      "Ġextr uded",
+      "Ġmini ature",
+      "Ġeditor ial",
+      "Ġinterconn ect",
+      "g ro",
+      "k v",
+      "ol en",
+      "ĠSY STEMS",
+      "ĠColon el",
+      "ĠMedi ated",
+      "ĠE MD",
+      "Ġkn ife",
+      "Ġcyt ogenetic",
+      "Ġdig itized",
+      "abin oids",
+      "arter ial",
+      "Ġdiar rhoea",
+      "b ag",
+      "Ġb uccal",
+      "st ay",
+      "ĠL AMP",
+      "ok o",
+      "ĠPol yt",
+      "mask ed",
+      "ĠTun able",
+      "Ġco agul",
+      "par as",
+      "Ġterm inating",
+      "IC Ag",
+      "ĠExcell ence",
+      "Ġregurg itation",
+      "DQU FD",
+      "J ack",
+      "Ġa pertures",
+      "ĠI p",
+      "ĠH CMV",
+      "ĠG om",
+      "Ġnucle ophilic",
+      "Ġparen teral",
+      "T IM",
+      "o ine",
+      "Ġn T",
+      "ĠS ense",
+      "ĠF ocal",
+      "ran ges",
+      "Ġhe pt",
+      "ĠPl at",
+      "Ġmy x",
+      "Ġcode book",
+      "Ex pl",
+      "ĠRho A",
+      "Ġrhin itis",
+      "ĠErr atum",
+      "Orient ed",
+      "W ell",
+      "d oping",
+      "Ġb up",
+      "ĠIm pedance",
+      "Ġsubstit utes",
+      "actor ily",
+      "Ġcollabor ations",
+      "ĠWay ne",
+      "Ġvow els",
+      "ĠSh adow",
+      "Ġphen ology",
+      "Ġconcur rency",
+      "h aving",
+      "ĠC ES",
+      "ĠF IN",
+      "ĠL oh",
+      "ox a",
+      "ĠAl N",
+      "ĠAl varez",
+      "ins tit",
+      "Ġgerm plasm",
+      "ĠBol iv",
+      "ĠR CP",
+      "ass ador",
+      "Ġes p",
+      "Ġphen otyping",
+      "Ġskip ping",
+      "ĠFract al",
+      "ĠPED OT",
+      "w ake",
+      "ĠF IT",
+      "ĠE SD",
+      "ĠAn tif",
+      "ubiqu itin",
+      "ĠAer ial",
+      "ĠProgn osis",
+      "ĠREL ATED",
+      "Ġstratig raphy",
+      "vat ron",
+      "ĠPROPERT IES",
+      "Ġ icon",
+      "is ers",
+      "Ġw al",
+      "Ġst amp",
+      "ĠOptim um",
+      "Ġolig omeric",
+      "Ġinn erv",
+      "Y A",
+      "Ab cam",
+      "Ġv ials",
+      "ĠG rig",
+      "Ġun aware",
+      "Ġoper a",
+      "ĠWar ner",
+      "Ġproton ated",
+      "ĠDR G",
+      "Ġtro ubles",
+      "Ġproposition al",
+      "ĠAfghan istan",
+      "ĠHamps hire",
+      "G d",
+      "l ung",
+      "Ġa viation",
+      "Ġap artment",
+      "Ġinf usions",
+      "Ġbro ilers",
+      "ĠDis ability",
+      "ĠRob ots",
+      "Ġdeb ugging",
+      "Ġì Ŀ",
+      "Wil son",
+      "upro fen",
+      "obarb ital",
+      "J B",
+      "is ance",
+      "iti zer",
+      "MI S",
+      "ĠAR F",
+      "Ġprost heses",
+      "Ġdichlor omethane",
+      "m Cherry",
+      "ĠS SS",
+      "ĠL PA",
+      "SC F",
+      "att ract",
+      "Ġcalibr ations",
+      "Ġfibr il",
+      "Ġhapl oid",
+      "usal em",
+      "ĠN ut",
+      "Ġde ut",
+      "ch ronic",
+      "NA P",
+      "ĠCytok ines",
+      "rage en",
+      "ĠC ategories",
+      "rain s",
+      "Ġsumm ands",
+      "Ġprolif erate",
+      "ryl ov",
+      "Ġple asure",
+      "Ġdens it",
+      "ĠSUR VE",
+      "H IP",
+      "h all",
+      "ĠF US",
+      "Ġwas ting",
+      "ER Y",
+      "Ġstat ins",
+      "Ġeast ward",
+      "some times",
+      "Ġwrap ping",
+      "ĠTW O",
+      "v ine",
+      "Ġs acchar",
+      "Ġam ateur",
+      "ĠÃ Ľ",
+      "Ġmy ster",
+      "ĠMy o",
+      "Ġrh abd",
+      "ĠProte ase",
+      "Ġchol era",
+      "ĠG ov",
+      "ĠG CC",
+      "Ġcl ays",
+      "trans mission",
+      "ĠHol lywood",
+      "Ġxen ob",
+      "FLO AT",
+      "Ġas cent",
+      "Ġsh arks",
+      "Ġinter feres",
+      "ĠForm er",
+      "ĠHart mann",
+      "s ha",
+      "ĠS ave",
+      "Ġpar ks",
+      "ĠV enn",
+      "Ġun ions",
+      "Ġdisc our",
+      "Ġsuper lattices",
+      "Ġcou pler",
+      "protein s",
+      "ĠStation ary",
+      "ĠEther net",
+      "ĠFré chet",
+      "Ġk ines",
+      "Ġj azz",
+      "As n",
+      "Ġextension al",
+      "Ġtel omeres",
+      "Ġpermit ting",
+      "Ġexha usted",
+      "ĠSph ing",
+      "T urn",
+      "m ind",
+      "Ġs f",
+      "ĠH ak",
+      "ran olol",
+      "port ation",
+      "Cons istent",
+      "Ġventi lated",
+      "ĠDIST RIB",
+      "Ġt elling",
+      "Ġman nose",
+      "ÃŃ az",
+      "Ġbor ne",
+      "Ġintens ification",
+      "Ġenjoy ed",
+      "ĠBrun o",
+      "ĠSatur day",
+      "Ġc ocycle",
+      "it ate",
+      "Ġg olf",
+      "appro ved",
+      "ĠNik ol",
+      "it ri",
+      "ĠS entiment",
+      "Ġg low",
+      "Ġg yp",
+      "ĠP CT",
+      "ab er",
+      "ĠW is",
+      "por um",
+      "Ġhy phae",
+      "fe as",
+      "ĠTra its",
+      "ĠConfl icts",
+      "degrad ing",
+      "R aman",
+      "ph armac",
+      "Ġimmun ocyt",
+      "ĠBl ake",
+      "Ġpseud oc",
+      "ĠCharacter isation",
+      "ĠGalile o",
+      "E nabl",
+      "J y",
+      "Ġcl av",
+      "ĠÏ ³",
+      "Ġcommun icated",
+      "eu tical",
+      "Ġnanot echnology",
+      "ĠHass an",
+      "ĠT ec",
+      "Ġh anging",
+      "ĠB SD",
+      "ĠCont our",
+      "Ġfrag ility",
+      "Ġdisrup tions",
+      "Ġfinit eness",
+      "ĠPhilipp ine",
+      "n icity",
+      "Ù ĩ",
+      "ĠC rim",
+      "ĠC NF",
+      "ĠI SI",
+      "ad apter",
+      "ĠU CP",
+      "Ġtext ured",
+      "AA V",
+      "ket o",
+      "N p",
+      "c ounting",
+      "h ynchus",
+      "Ġpro sec",
+      "ĠAn not",
+      "ĠHar bor",
+      "deg rees",
+      "ak ar",
+      "ĠV ik",
+      "bf d",
+      "Ġdri p",
+      "ĠCauc as",
+      "Ġtren ch",
+      "Ġwe ed",
+      "Ġdist ractor",
+      "gen etic",
+      "spec ifically",
+      "ulf ite",
+      "ĠCons istently",
+      "Ġbreak fast",
+      "Ġbul let",
+      "Ġleg isl",
+      "ĠTra umatic",
+      "Ġcollect ors",
+      "ĠBul let",
+      "ĠMY B",
+      "ĠP ink",
+      "vers ive",
+      "ĠAt tem",
+      "Ġcult urally",
+      "B ell",
+      "und ef",
+      "vi i",
+      "Ġhist ocompatibility",
+      "let cher",
+      "ĠSte f",
+      "A mp",
+      "ĠR id",
+      "ĠE ucl",
+      "Ġdec ryption",
+      "ĠSp encer",
+      "ĠBit coin",
+      "w ic",
+      "Ġcom plicate",
+      "ĠPro posal",
+      "ĠÄ Ī",
+      "avirus es",
+      "ĠF ay",
+      "ĠR d",
+      "ĠG ale",
+      "ĠMetast asis",
+      "ĠImprove ments",
+      "Â ©",
+      "Ġpoly ester",
+      "Ġstrat ospheric",
+      "ĠSA H",
+      "Ġamph ip",
+      "ĠA FP",
+      "ĠH air",
+      "ĠE PI",
+      "ĠUl trast",
+      "Ġâĭ ¯",
+      "Ġga pless",
+      "H am",
+      "et to",
+      "Ġth reonine",
+      "ĠE CO",
+      "Ġi a",
+      "Ġund ist",
+      "Ġradi ology",
+      "Ġsuper lattice",
+      "ibr aries",
+      "Ġturb id",
+      "ĠPot entials",
+      "ĠPip eline",
+      "Ġwarf arin",
+      "W ISE",
+      "ĠL id",
+      "Ġrec urring",
+      "ĠMon o",
+      "ĠGover n",
+      "ĠAware ness",
+      "ol ab",
+      "if lora",
+      "str is",
+      "IN DEX",
+      "ĠDem entia",
+      "Do es",
+      "w right",
+      "Í ī",
+      "Ġs b",
+      "ĠD OM",
+      "ĠH BsAg",
+      "cl inic",
+      "ĠEx ped",
+      "Ġprote as",
+      "Ġster ilization",
+      "ĠBan erjee",
+      "ĠPerson nel",
+      "âĮ ĭ",
+      "oneph ritis",
+      "om ite",
+      "ĠC CF",
+      "os iti",
+      "ĠE ucalyptus",
+      "ĠIs otope",
+      "col i",
+      "poss ibility",
+      "Ġstr ontium",
+      "Ġra ref",
+      "ĠInter stellar",
+      "kin in",
+      "yleth anol",
+      "J T",
+      "n orth",
+      "Ġc ensored",
+      "is tive",
+      "Ġno ticing",
+      "Ġship ping",
+      "Em bed",
+      "Obs erv",
+      "Ġze olites",
+      "ub it",
+      "Ġfl aps",
+      "Ġdr ifts",
+      "Ġtherap ist",
+      "Ġpoll ination",
+      "ali platin",
+      "Joh nson",
+      "Ġimperf ections",
+      "N Y",
+      "Ġth alamic",
+      "oc arb",
+      "oz otocin",
+      "Ġtet ramer",
+      "Pl as",
+      "Ġmultic hannel",
+      "ĠIns ight",
+      "op ods",
+      "ĠN acional",
+      "Ġim atinib",
+      "act ual",
+      "ĠX OR",
+      "Ġbl ight",
+      "ĠLe ading",
+      "ames e",
+      "ĠAm plitude",
+      "ĠMon itor",
+      "ĠNeu rological",
+      "propag ating",
+      "Ġp addle",
+      "ĠHar vest",
+      "Ġod ont",
+      "BU F",
+      "Ġtac tics",
+      "ĠAnis otropy",
+      "ad ip",
+      "ĠAl pine",
+      "Ġfe els",
+      "Ġmed ieval",
+      "Ġel ucidation",
+      "Ġheter otrophic",
+      "Ġrelax ing",
+      "Ġhapp iness",
+      "ĠCyt otoxicity",
+      "ĠRAN KL",
+      "Walk er",
+      "m ig",
+      "ĠS SL",
+      "ĠS epsis",
+      "ĠG es",
+      "Ġhydro chloric",
+      "Ġclar ification",
+      "Ġdispar ate",
+      "t ested",
+      "Ġdat ap",
+      "Ġnovel s",
+      "ĠMicro c",
+      "á l",
+      "ĠAR C",
+      "ĠYang tze",
+      "etom idine",
+      "ĠMat rigel",
+      "ih ilation",
+      "ĠcDNA s",
+      "Ġprost at",
+      "ĠRail road",
+      "UB LE",
+      "ĠPART IC",
+      "ĠS ax",
+      "Ġins ecurity",
+      "Ġcr ushed",
+      "Ġhal ves",
+      "gi ant",
+      "ĠCro atia",
+      "icycl o",
+      "ĠUne xpected",
+      "Ġlon eliness",
+      "an u",
+      "Ġch ampions",
+      "ub erculosis",
+      "Ġequ i",
+      "Ġacc reted",
+      "Ġinv ading",
+      "Ġaff erents",
+      "Ġaltern ation",
+      "Ġkin et",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠĠĠĠĠ",
+      "ĠMAG NET",
+      "ĠFIF A",
+      "z adeh",
+      "ip henyl",
+      "ĠK ro",
+      "ĠEval uate",
+      "illi ant",
+      "cur vature",
+      "ĠPier ce",
+      "b etter",
+      "n os",
+      "à ¥",
+      "ĠK CN",
+      "ĠSt rand",
+      "ca emic",
+      "ĠHo echst",
+      "ĠEX T",
+      "ĠLL VM",
+      "B Z",
+      "t gt",
+      "on dialdehyde",
+      "ĠE vid",
+      "ĠG ul",
+      "Ġmulti plications",
+      "Ġaut h",
+      "ĠAustr al",
+      "Ġstay ing",
+      "ĠGlut amate",
+      "Ġst ray",
+      "ĠI SA",
+      "Ġlow land",
+      "Ġparallel s",
+      "Ġattrac tiveness",
+      "Ġelectrosp inning",
+      "Ġportray ed",
+      "ospec ific",
+      "f olate",
+      "Ġcoe ff",
+      "ĠEst rogen",
+      "tum our",
+      "Ġhystere ctomy",
+      "Ġin ositol",
+      "ĠB az",
+      "ist ein",
+      "Ġcruc ially",
+      "Ġdin oflag",
+      "ÍĶ ÍĴ",
+      "ĠDrag on",
+      "ĠS por",
+      "ĠM ater",
+      "ĠH ero",
+      "plic ing",
+      "ĠAN T",
+      "ĠForm ic",
+      "Que ue",
+      "ocarcin omas",
+      "U PS",
+      "ĠP c",
+      "enc oders",
+      "Ġinv aded",
+      "ĠPh ases",
+      "Ġpost mortem",
+      "Ġslow s",
+      "ĠMc L",
+      "ĠVer ma",
+      "ĠVi ability",
+      "Ġcompens ating",
+      "Ġclamp ed",
+      "j m",
+      "ĠR iv",
+      "up on",
+      "ĠDick inson",
+      "initi ated",
+      "Ġs ider",
+      "ĠS elen",
+      "ĠA ka",
+      "idel berg",
+      "Ġqual ifying",
+      "Ġenfor cing",
+      "otroph s",
+      "ĠSNA P",
+      "Ġr ust",
+      "imb urs",
+      "Ġimmunocomp romised",
+      "ĠFlem ing",
+      "Ġl izards",
+      "di alysis",
+      "ĠUn ivariate",
+      "Ġgas oline",
+      "Ġten ure",
+      "Ġsustain ing",
+      "Ġmot one",
+      "b ay",
+      "w ani",
+      "ore station",
+      "ĠX II",
+      "Ġradi ofrequency",
+      "ĠGu ided",
+      "Ind ividual",
+      "ĠSpect rometer",
+      "ĠGo ing",
+      "ĠMart ins",
+      "Ap proxim",
+      "am ak",
+      "ĠâĪ ı",
+      "ĠO mn",
+      "Ġout patients",
+      "Ġhyper bol",
+      "ĠPer ceptual",
+      "ĠBur ke",
+      "Bol tzmann",
+      "ĠM d",
+      "Ġpa w",
+      "ĠCat hedral",
+      "Ġhyal uron",
+      "Ġbrach ial",
+      "Ġaflat oxin",
+      "im o",
+      "Ġen rol",
+      "Ġdet onation",
+      "Ġover ly",
+      "the st",
+      "Ġsecond ly",
+      "ĠSch iz",
+      "ĠIGF BP",
+      "atech in",
+      "Ġs aves",
+      "ti ers",
+      "ĠB ates",
+      "Ġall iance",
+      "Ġatt ri",
+      "Ġast ro",
+      "ĠPath ological",
+      "Ġgamb iae",
+      "P ark",
+      "id able",
+      "ĠN il",
+      "ĠJ as",
+      "Ġneed ing",
+      "me ier",
+      "Ġferro ptosis",
+      "ĠGuid ance",
+      "A Z",
+      "i ol",
+      "Ġac knowledg",
+      "ex ual",
+      "Ġmen opause",
+      "Ġadj unct",
+      "cap ture",
+      "ĠDep uty",
+      "Ġb ial",
+      "if a",
+      "ĠCh itosan",
+      "ĠTop ics",
+      "ĠPlas mid",
+      "calc ulations",
+      "g ive",
+      "respond ers",
+      "ull a",
+      "ĠMore no",
+      "Ġcomment ary",
+      "ĠMah m",
+      "ï£ ±",
+      "on acci",
+      "ĠC ould",
+      "ĠTR P",
+      "second s",
+      "Graph Pad",
+      "L ittle",
+      "he y",
+      "Ġal ike",
+      "ĠDi as",
+      "aro o",
+      "ĠÄ ±",
+      "Ġtax es",
+      "phen anth",
+      "ĠChe ung",
+      "ĠPi et",
+      "D f",
+      "G U",
+      "m ectin",
+      "z ee",
+      "Ġd λ",
+      "Ġsynt heses",
+      "Ġá Ī",
+      "Sim ulation",
+      "ĠEle ven",
+      "w orms",
+      "lymph ocyte",
+      "Ġhaemorrh age",
+      "ĠO wn",
+      "ĠK ant",
+      "Ġover se",
+      "Ġide ation",
+      "ĠHar per",
+      "Acknowledg ments",
+      "v ili",
+      "yn a",
+      "ĠRec urrence",
+      "oz a",
+      "Ġhence forth",
+      "ze es",
+      "Ġquas ic",
+      "Ġchor oidal",
+      "Ġantim alarial",
+      "Ġcoars ening",
+      "D eb",
+      "di am",
+      "ĠWe ights",
+      "Ġbu ying",
+      "Ġmess aging",
+      "Fe bruary",
+      "Ext ended",
+      "ĠRoss i",
+      "Ġmist aken",
+      "Ġut ero",
+      "j as",
+      "ic itis",
+      "ĠT idal",
+      "Ġph aryngeal",
+      "cl ick",
+      "Ġmy o",
+      "kn ock",
+      "Ġpromin ence",
+      "Ġamphi philic",
+      "c orn",
+      "Ġon board",
+      "ĠD ud",
+      "ĠW oman",
+      "ĠOut break",
+      "Ġprefer ably",
+      "Ġsket ches",
+      "S at",
+      "f ixing",
+      "ĠM ey",
+      "ĠLet ters",
+      "IT IES",
+      "ĠSD P",
+      "ĠLNC aP",
+      "D X",
+      "F luor",
+      "R v",
+      "S ect",
+      "ĠI ons",
+      "Ġtrac hom",
+      "Ġult rastructure",
+      "qv ist",
+      "rop he",
+      "Ġrece ipt",
+      "ĠQu int",
+      "Ġsw apping",
+      "amin idase",
+      "Ġarch ival",
+      "ĠCre ating",
+      "ĠBart on",
+      "diagn osed",
+      "at ological",
+      "ol ph",
+      "ĠP FA",
+      "ĠL AP",
+      "Ġun physical",
+      "eq n",
+      "Ġquar tiles",
+      "olytic a",
+      "ĠFre ed",
+      "Ġventil ator",
+      "Ġkary otype",
+      "S ta",
+      "s till",
+      "ĠT ate",
+      "ur ability",
+      "ĠG ron",
+      "Ġtr imer",
+      "IP A",
+      "adec a",
+      "ĠImplement ing",
+      "s ity",
+      "it r",
+      "Ġb om",
+      "Ġnon relativistic",
+      "Ġmic elle",
+      "ĠAd minist",
+      "Ġelectro lysis",
+      "har mon",
+      "OLOG ICAL",
+      "L iter",
+      "ĠG UI",
+      "ĠQ L",
+      "mon ths",
+      "Ġsuper flu",
+      "cut s",
+      "Ġelic its",
+      "Ġmultiplex ed",
+      "overl ap",
+      "Ġcada ver",
+      "Ġo u",
+      "ĠS heng",
+      "ere a",
+      "ĠN BC",
+      "Ġdet er",
+      "ty rosine",
+      "ĠPar ts",
+      "Ġess ay",
+      "k as",
+      "it ted",
+      "ĠP ZT",
+      "ess ler",
+      "Ġsim ulators",
+      "Ġradi ating",
+      "cut ting",
+      "ĠCalc ulating",
+      "TH ER",
+      "ĠROC K",
+      "commun ic",
+      "Ġbon us",
+      "ĠC PA",
+      "ĠP UR",
+      "ult on",
+      "ĠZ hi",
+      "Ġcal oric",
+      "Ġinterp olate",
+      "ĠSec retion",
+      "Ġneuro cognitive",
+      "Ġgad olinium",
+      "f requencies",
+      "ĠT ract",
+      "Ġminim ax",
+      "ĠBro ck",
+      "ryp sin",
+      "ĠReson ant",
+      "ĠACKNOWLED GEMENTS",
+      "D om",
+      "Ġhol otype",
+      "Spec ial",
+      "Ġimmunore active",
+      "ARN ING",
+      "Pan el",
+      "ĠJohann es",
+      "R FP",
+      "z zi",
+      "ĠP omer",
+      "Ġtrans ects",
+      "Ġpo ured",
+      "ED s",
+      "ĠCirc um",
+      "Ġabnorm ally",
+      "ĠPun j",
+      "G ol",
+      "H op",
+      "H ex",
+      "I LE",
+      "Ġsour ced",
+      "ocl ase",
+      "prot obuf",
+      "Ġfro gs",
+      "ĠOt tawa",
+      "Ġbioge ochemical",
+      "Ġlenti virus",
+      "Y oung",
+      "ĠI PS",
+      "ass en",
+      "Ġun restricted",
+      "Ġmat plotlib",
+      "Ġchlor amphenicol",
+      "ĠContext ual",
+      "ĠHawai ian",
+      "Leg end",
+      "S parse",
+      "b ore",
+      "g aussian",
+      "u ke",
+      "ĠâĢ °",
+      "ret est",
+      "SS E",
+      "pre ting",
+      "ĠPan ama",
+      "ĠBroad band",
+      "conjug ate",
+      "B ytes",
+      "G SH",
+      "U ns",
+      "r ina",
+      "Ġd rained",
+      "Ġsc ap",
+      "Ġinves ted",
+      "Ġsatisf actorily",
+      "Ġherbiv ores",
+      "Ġarachid onic",
+      "ymet rix",
+      "Ġn ect",
+      "Ġcon ges",
+      "ĠM err",
+      "ĠM ai",
+      "Ch ain",
+      "Ġretrie ving",
+      "Col lection",
+      "ĠMT X",
+      "ĠFernand o",
+      "h g",
+      "ĠR ams",
+      "th resh",
+      "aps ules",
+      "Ġcond uit",
+      "sw ap",
+      "Ġblow ing",
+      "ĠNy quist",
+      "Ġuncons cious",
+      "ĠDIFFE RENT",
+      "T echn",
+      "h iz",
+      "î Ĥ",
+      "Ġd ξ",
+      "ĠSt o",
+      "ĠFlav on",
+      "Dav id",
+      "Ġfiltr ate",
+      "l ith",
+      "ĠW ool",
+      "ĠK not",
+      "Ġhal ide",
+      "Ġbio assay",
+      "ĠGold berg",
+      "ĠTrich oderma",
+      "Ġintras pecific",
+      "c rystall",
+      "ĠR end",
+      "our g",
+      "Ġunder take",
+      "ĠEn um",
+      "inf ect",
+      "Ġmid gut",
+      "att ack",
+      "ĠCirc le",
+      "Ġplei otropic",
+      "es cent",
+      "ĠF ri",
+      "ph ilis",
+      "ast ings",
+      "Ġbi ogas",
+      "ĠÄ ľ",
+      "Ġaccomp any",
+      "Ġroll ed",
+      "Ġchir p",
+      "Ġsomat ostatin",
+      "vark appa",
+      "S cal",
+      "Ġd row",
+      "rom ed",
+      "ĠL up",
+      "ĠL uminosity",
+      "ĠN ig",
+      "fer romagnetic",
+      "ĠTo y",
+      "Ġcann abinoid",
+      "ĠH OX",
+      "ie le",
+      "ĠCT X",
+      "Ġhyd rop",
+      "Ġfavor ite",
+      "Ġstret ches",
+      "eval uated",
+      "ogroup s",
+      "ac al",
+      "ol lo",
+      "Ġg enders",
+      "ĠG raft",
+      "Ġinc idences",
+      "Ġreplac ements",
+      "ĠTR UNC",
+      "CR F",
+      "Ġequal ization",
+      "ĠRen ew",
+      "Ġple thora",
+      "ĠEnc oder",
+      "M it",
+      "Ġc aches",
+      "or ate",
+      "end ors",
+      "ĠCa ution",
+      "ĠAb el",
+      "comp ression",
+      "ĠLars en",
+      "ĠElim ination",
+      "Ġt ester",
+      "Ġn inth",
+      "ĠL ö",
+      "Ġsp iders",
+      "Ġpo em",
+      "Ġeduc ators",
+      "ĠEnh ances",
+      "dest ructive",
+      "Four ier",
+      "Ġseism icity",
+      "ĠYun nan",
+      "Riemann ian",
+      "W ID",
+      "v ular",
+      "ĠB order",
+      "Ġcomb in",
+      "sing let",
+      "ĠEd dington",
+      "ĠTem plate",
+      "ĠPA X",
+      "Ġbasal ts",
+      "En h",
+      "Ġassist ants",
+      "ĠCasc ade",
+      "Ġin breeding",
+      "ch ini",
+      "Ġup graded",
+      "ĠTrans it",
+      "sur vival",
+      "Ġinject or",
+      "ĠPas cal",
+      "DEV ICE",
+      "Ġf ost",
+      "ĠK and",
+      "Ġext ragalactic",
+      "epend ently",
+      "Ġexc ite",
+      "Ġfulf il",
+      "Ġrip arian",
+      "Ġuplo aded",
+      "a un",
+      "l od",
+      "s aving",
+      "ĠH ib",
+      "ĠE ra",
+      "ob ese",
+      "Ġu i",
+      "Ġspect rally",
+      "ke V",
+      "xx x",
+      "ĠOt to",
+      "Ġé tale",
+      "L AT",
+      "d ermal",
+      "di az",
+      "ĠPl i",
+      "Ġleg ume",
+      "Ġinsp ect",
+      "Ġthym ic",
+      "ĠHorm one",
+      "á Ģ",
+      "in ot",
+      "ĠS hib",
+      "ĠB CC",
+      "ĠV ital",
+      "Ġprof its",
+      "ĠFed erated",
+      "Ġflip ped",
+      "Ġpropri etary",
+      "incor porated",
+      "Ġbact eremia",
+      "Ġáŀ ĩ",
+      "f ins",
+      "ä ½",
+      "es ia",
+      "ĠH ollow",
+      "ge ons",
+      "Ġtre halose",
+      "ER O",
+      "oster ol",
+      "om us",
+      "ĠC rystall",
+      "Ġcur ation",
+      "Ġmagn on",
+      "ĠAm end",
+      "Ġhar b",
+      "Ġneutral ity",
+      "ĠDel phi",
+      "Ġnons ense",
+      "ĠHome ostasis",
+      "Ġexpendit ures",
+      "Sequ ential",
+      "imod ular",
+      "Ġz enith",
+      "ĠMor an",
+      "Ġbootstrap ping",
+      "i omy",
+      "l actic",
+      "it ure",
+      "Ġn at",
+      "Ġg ab",
+      "Ġch at",
+      "reg ional",
+      "Ġcr ashes",
+      "ĠAF B",
+      "Ġcrow ded",
+      "Ġtwe et",
+      "engine ered",
+      "ĠCharg ed",
+      "S che",
+      "IT IONS",
+      "ĠCor al",
+      "ĠEl i",
+      "Ġinver ting",
+      "Ġped ag",
+      "ĠSand ers",
+      "Mean while",
+      "ĠGriff iths",
+      "P SCs",
+      "ti ze",
+      "ĠM ail",
+      "Ġund ec",
+      "Ġher mitian",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ",
+      "ĠExpl os",
+      "Ġwest ward",
+      "ĠConf irm",
+      "B egin",
+      "Ġfactor ies",
+      "ĠPR L",
+      "she ar",
+      "Head er",
+      "ĠFLAG S",
+      "an omal",
+      "ĠQ W",
+      "ĠÌ ħ",
+      "oin ositi",
+      "Ġmamm ography",
+      "Ġdeposition al",
+      "EX P",
+      "resid ue",
+      "Ġunsatisf actory",
+      "A β",
+      "M UX",
+      "Ġst aged",
+      "ĠM MT",
+      "ĠK us",
+      "ll o",
+      "Ġtrain er",
+      "add en",
+      "Ġpin ch",
+      "WA RE",
+      "Ġcab inet",
+      "C SP",
+      "ec um",
+      "ot eric",
+      "ĠH av",
+      "Ġres ume",
+      "Ġnetwork ed",
+      "sh are",
+      "ĠCol le",
+      "Ġchem otactic",
+      "ĠGly c",
+      "olk it",
+      "Ġbot ulinum",
+      "ĠNeighbor hood",
+      "m V",
+      "ĠH Q",
+      "ef aciens",
+      "get t",
+      "Ġge ost",
+      "ĠCD W",
+      "ĠÌ §",
+      "Ġflo ors",
+      "represent ing",
+      "odi ode",
+      "ĠInst ance",
+      "Ġmonod is",
+      "d rying",
+      "re asing",
+      "ig i",
+      "Ġg out",
+      "ĠI EC",
+      "Ġfl ush",
+      "Ġtra ded",
+      "Re view",
+      "ĠïĤ ¢",
+      "Ġà ¤",
+      "Ġabbrevi ations",
+      "otherap ies",
+      "Ġindeterm inate",
+      "Ġglutar aldehyde",
+      "Ġattri tion",
+      "j ump",
+      "in de",
+      "ĠG ri",
+      "arc tion",
+      "TRA IN",
+      "Ġescap ed",
+      "at ement",
+      "ĠP am",
+      "ĠG AM",
+      "pro ductive",
+      "ĠAmeric as",
+      "agen esis",
+      "ĠMi xtures",
+      "ĠHo oft",
+      "ĠWind ow",
+      "Ġnod ular",
+      "Ġech in",
+      "D OF",
+      "ĠD DT",
+      "elect rical",
+      "ĠDec entralized",
+      "Ġcontrad ict",
+      "F rench",
+      "Ġa ustr",
+      "ĠA PD",
+      "ĠD IM",
+      "ĠSt en",
+      "ron omic",
+      "ĠPolym orphism",
+      "Ġc occ",
+      "it ings",
+      "Ġsub critical",
+      "ĠUn iqueness",
+      "OP EN",
+      "rot oxicity",
+      "Gen Bank",
+      "atab ases",
+      "N ets",
+      "u istic",
+      "y ric",
+      "ĠS ID",
+      "Ġco oked",
+      "ĠJ udge",
+      "Ġparameter izations",
+      "Ġenum erated",
+      "ĠAst hma",
+      "De velop",
+      "Ġc ake",
+      "ĠA ges",
+      "ven ile",
+      "Ġfl or",
+      "Ġcould n",
+      "det ach",
+      "Ġpip ette",
+      "ĠInst ant",
+      "Ġtent atively",
+      "ĠINT EGR",
+      "H Q",
+      "M apping",
+      "c q",
+      "å Ī",
+      "SU M",
+      "frac tions",
+      "ĠCl aud",
+      "Form ula",
+      "Ax is",
+      "ĠBil ly",
+      "ĠMeth ane",
+      "ĠI GM",
+      "c annot",
+      "Ø ³",
+      "Ġc iting",
+      "ĠD ynam",
+      "Ġle tt",
+      "eg ler",
+      "ĠPhysic ians",
+      "x FF",
+      "Ġo yster",
+      "ĠT OC",
+      "Ġsub arachnoid",
+      "ĠCO M",
+      "IT ER",
+      "Ġbenz odiazep",
+      "Ġuncom plicated",
+      "till o",
+      "Car bon",
+      "at em",
+      "Ġs el",
+      "ing o",
+      "IV ITY",
+      "Ġca vern",
+      "Ġspac elike",
+      "ĠColl isions",
+      "stra int",
+      "Ġmyc obacterial",
+      "Ġtrachom atis",
+      "A i",
+      "m f",
+      "ĠT ric",
+      "tic o",
+      "ĠE lection",
+      "ĠK DM",
+      "ĠEx osomes",
+      "af luor",
+      "Ġformal ized",
+      "ĠEL F",
+      "v phantom",
+      "ĠS ME",
+      "ich uan",
+      "ĠV Ms",
+      "Ġro stral",
+      "of er",
+      "ram anian",
+      "inter cal",
+      "Mer ck",
+      "ĠFerg uson",
+      "H U",
+      "l j",
+      "Ġr ack",
+      "Ġmicro graph",
+      "CT S",
+      "Ġpass ively",
+      "ĠMass es",
+      "rang ians",
+      "ĠAD M",
+      "ĠProvid ed",
+      "ĠVeter ans",
+      "s ound",
+      "ge x",
+      "ĠSp iral",
+      "Ġfoss a",
+      "Ġthermod ynamically",
+      "ĠS taining",
+      "Ġk ar",
+      "ef lon",
+      "ĠBr uns",
+      "VA E",
+      "olytic us",
+      "Ġintran asal",
+      "ĠProsp ects",
+      "at hers",
+      "Ġnumber ing",
+      "ĠRe placement",
+      "ĠNon commutative",
+      "quis itions",
+      "ĠSIM D",
+      "ĠArter ial",
+      "ĠH GF",
+      "ĠG PP",
+      "Ġflu vial",
+      "ner i",
+      "ĠComp ressed",
+      "VID IA",
+      "Ġshock ed",
+      "d ys",
+      "in variance",
+      "en stein",
+      "ĠS CM",
+      "ĠD od",
+      "Ġsh o",
+      "Ch lor",
+      "du ino",
+      "Ġurg ently",
+      "s oc",
+      "et ching",
+      "Ġdiff ractive",
+      "ĠZ F",
+      "Ġhyper planes",
+      "Ġmy ri",
+      "Ġfer romagnetism",
+      "fil ament",
+      "Ġjustif ies",
+      "f ault",
+      "ĠH TS",
+      "ĠE PC",
+      "to o",
+      "ĠTR AP",
+      "i ón",
+      "r v",
+      "ĠB PD",
+      "ĠN od",
+      "pos it",
+      "ĠCon vers",
+      "Ġzero es",
+      "ĠGl en",
+      "ĠDist urb",
+      "Ġtable au",
+      "Ġpseud ot",
+      "ĠColl ider",
+      "Ġadsorb ents",
+      "ĠGro ve",
+      "Ġking dom",
+      "E st",
+      "X s",
+      "c zyk",
+      "ĠW ille",
+      "ĠV OL",
+      "sc ar",
+      "ĠAd ler",
+      "ĠOr chestra",
+      "Ġspars ely",
+      "glycos ylation",
+      "L ac",
+      "o tions",
+      "ĠI le",
+      "Ġbe acon",
+      "ĠR n",
+      "ull ah",
+      "Ġtim elike",
+      "ĠFore sts",
+      "Ġupl oad",
+      "j it",
+      "ĠE DM",
+      "Ġtrans plants",
+      "mark er",
+      "ĠBre eding",
+      "ÎĶ ÎĶ",
+      "Ġfavor ably",
+      "ĠTransform ations",
+      "abel ed",
+      "ĠPoli tics",
+      "epis ode",
+      "Ġf ut",
+      "Ġd ithi",
+      "ĠM w",
+      "Ġtrans piration",
+      "Ġun limited",
+      "ĠAn tiv",
+      "PP V",
+      "Ġnom ogram",
+      "Ġinvent ed",
+      "ĠSched ule",
+      "all ows",
+      "Ġtrans vers",
+      "Ġwork piece",
+      "black square",
+      "call back",
+      "ĠAth letic",
+      "h ans",
+      "p oles",
+      "Ġe avesdrop",
+      "ĠC one",
+      "oc lim",
+      "ĠG host",
+      "iter ations",
+      "Ġweak en",
+      "Ġalkal oid",
+      "Ġveter ans",
+      "Ġprostat ectomy",
+      "Ġb og",
+      "ĠC ed",
+      "ĠF ever",
+      "yl an",
+      "arch ive",
+      "Ġattack ers",
+      "M aking",
+      "b ane",
+      "ĠP ull",
+      "ĠF LO",
+      "Ġco aches",
+      "ĠV SM",
+      "ok h",
+      "ĠSp o",
+      "amil ial",
+      "princ iple",
+      "Ġaggress iveness",
+      "Ġgard ens",
+      "S IG",
+      "Ġm bar",
+      ".. ...",
+      "Ġoptim izes",
+      "Ġmorph ologic",
+      "han i",
+      "Ġgerm anium",
+      "Enabl ed",
+      "w b",
+      "Ġfor amen",
+      "ĠS PA",
+      "Ġmagn ified",
+      "ĠSl ater",
+      "ĠSy rian",
+      "Ġt ert",
+      "Ġtra ditions",
+      "Ġoff ensive",
+      "Ġhyd rology",
+      "erge tics",
+      "Ph ot",
+      "Ġperovsk ites",
+      "Ġwaven umbers",
+      "Ġosteocl asts",
+      "imed ean",
+      "ĠBasket ball",
+      "benzodi ox",
+      "ĠTRUNC ATED",
+      "Ġb ishop",
+      "ĠS gr",
+      "ĠS atisfaction",
+      "agn ostic",
+      "num eric",
+      "Ġnorm als",
+      "Ġhum or",
+      "Ġcontin ents",
+      "NAT ION",
+      "Lag rangian",
+      "Ġkne es",
+      "ĠI DE",
+      "ad as",
+      "ad ia",
+      "ĠO U",
+      "ond s",
+      "ĠCh aud",
+      "Ġsl icing",
+      "ĠAc tor",
+      "Al t",
+      "Ġbroad casts",
+      "osa urs",
+      "Ġpick le",
+      "Ġunf amiliar",
+      "all us",
+      "Ġhas hing",
+      "inc idence",
+      "Ġmetabol ized",
+      "Ġhomogeneous ly",
+      "ĠFal con",
+      "Ġ Ñģ",
+      "ĠC ere",
+      "ĠC LA",
+      "ĠP aste",
+      "ĠF ry",
+      "ĠD re",
+      "ad ult",
+      "Ġdisc ounted",
+      "sens itized",
+      "ercul ous",
+      "ĠP ixel",
+      "ĠB ram",
+      "all o",
+      "ip ative",
+      "ip ality",
+      "ĠSt rict",
+      "ĠTr inity",
+      "ĠClass ifiers",
+      "ĠBas el",
+      "ĠCur cumin",
+      "ĠLU MO",
+      "Ġmediast inal",
+      "ĠF FA",
+      "Ġpl enty",
+      "pr ised",
+      "Ġpr inter",
+      "Ġcalc are",
+      "ins n",
+      "ont ology",
+      "Ġground ing",
+      "ĠAL DH",
+      "STR ING",
+      "ĠFem ales",
+      "ĠFocus ing",
+      "assess ment",
+      "ĠBlu etooth",
+      "ëĬ Ķ",
+      "Ġe go",
+      "ĠD AC",
+      "ond ing",
+      "rand a",
+      "ĠLud wig",
+      "Ġanteced ent",
+      "ĠErn st",
+      "d X",
+      "od eling",
+      "âĢ ĭ",
+      "In ser",
+      "ogn ormal",
+      "ĠTe vatron",
+      "Ġcovari ances",
+      "rig ing",
+      "ĠMg SO",
+      "carbon itrile",
+      "ĠLore n",
+      "Ġpolyt opes",
+      "ĠParent al",
+      "Ġun healthy",
+      "ither to",
+      "ĠMo tif",
+      "Data Type",
+      "ĠMI PS",
+      "ĠPhosph orus",
+      "Mo O",
+      "ĠPerturb ations",
+      "Ġaph ids",
+      "Ġanhyd ride",
+      "id eration",
+      "ĠM its",
+      "gra vit",
+      "Ġdest inations",
+      "Com mun",
+      "Ġtetra hedron",
+      "im plicit",
+      "Ġass ort",
+      "ĠSub t",
+      "ĠAcet yl",
+      "ec ium",
+      "ĠN ie",
+      "Ġoper and",
+      "ĠSc her",
+      "az oles",
+      "tle ment",
+      "ĠBlock ing",
+      "Ġbottlen ecks",
+      "ĠOccup ational",
+      "H AS",
+      "T eller",
+      "Ġv ague",
+      "est ing",
+      "SN E",
+      "Ġphoto ionization",
+      "Ġcompl aint",
+      "us pid",
+      "ow ler",
+      "Ġend ocytic",
+      "Ġfl ocks",
+      "eps in",
+      "col ors",
+      "otop es",
+      "ĠDep letion",
+      "ELL AR",
+      "ar med",
+      "ĠT IR",
+      "Ġbul lying",
+      "Ġâİ §",
+      "ospor idium",
+      "M r",
+      "ĠC ic",
+      "og al",
+      "Ġsection ed",
+      "Ch apter",
+      "ĠCont ents",
+      "ĠPath s",
+      "ĠExpl ain",
+      "comput ing",
+      "Ġshr ub",
+      "ĠMalays ian",
+      "B eta",
+      "M ad",
+      "R os",
+      "Ġe yel",
+      "ĠA CF",
+      "ĠM m",
+      "text ure",
+      "Ġinterpret ability",
+      "ĠTop ic",
+      "Ġbad ly",
+      "ĠmA h",
+      "E g",
+      "R Q",
+      "p ins",
+      "et c",
+      "ĠV ogel",
+      "Ġhyp oc",
+      "Ġrun away",
+      "Ġperson ally",
+      "Ġbind ers",
+      "sens ory",
+      "ĠIP v",
+      "rank ed",
+      "Ġfibr ations",
+      "Ġstraw berry",
+      "arot omy",
+      "F LI",
+      "r ator",
+      "od ys",
+      "ir an",
+      "ĠB ead",
+      "ĠD AM",
+      "âĪ ĥ",
+      "Ġill usion",
+      "pid ium",
+      "Pl ace",
+      "Reg ion",
+      "Ġalloc ations",
+      "Ġoh mic",
+      "Ġn f",
+      "im ino",
+      "ĠB ris",
+      "iti zing",
+      "pro per",
+      "sub group",
+      "Ġsal ience",
+      "rang ement",
+      "ĠMean ing",
+      "Ġbarc ode",
+      "Ġneurop eptide",
+      "Ġendos perm",
+      "im ab",
+      "Ġnan od",
+      "ĠMet z",
+      "Ġcoll ocation",
+      "ĠInf ected",
+      "Ġpack aged",
+      "ĠAD A",
+      "ĠBar th",
+      "ĠCN C",
+      "Ġcasc aded",
+      "ĠStock holm",
+      "it as",
+      "ĠM MR",
+      "ĠD rought",
+      "Ġper missible",
+      "Ġsc iatic",
+      "Ġfr inges",
+      "Ġexec utable",
+      "Ġstem ness",
+      "ĠEnd oscopic",
+      "apor in",
+      "T OP",
+      "e B",
+      "t ur",
+      "ĠSt ages",
+      "anc hes",
+      "Ġnon perturbative",
+      "Ġmar itime",
+      "Ġcovers lips",
+      "Ġlag oon",
+      "Experim ents",
+      "Ġcodew ords",
+      "Enc oder",
+      "d as",
+      "p rac",
+      "Ġp addy",
+      "Ġd raining",
+      "Ġk ids",
+      "Ġen emies",
+      "Ġmo tile",
+      "Ġsl ack",
+      "be es",
+      "ĠSup pl",
+      "ĠBar ber",
+      "ĠSP H",
+      "Ġcrystall ite",
+      "fit ted",
+      "cycl opent",
+      "ĠRMS D",
+      "Ġpark inson",
+      "Ġuncor rected",
+      "ĠSynt ax",
+      "Ġmultin omial",
+      "ĠIncor porating",
+      "akrish nan",
+      "J L",
+      "N ESS",
+      "m im",
+      "ĠT ET",
+      "ĠP orph",
+      "ĠSch we",
+      "Ġcatalog s",
+      "ĠAuthentic ation",
+      "B ro",
+      "ug ar",
+      "ĠAm pl",
+      "Ġsap iens",
+      "ĠGAN s",
+      "Ġnecessit ates",
+      "t g",
+      "ed al",
+      "ĠR ear",
+      "op eptidase",
+      "ĠIn formed",
+      "Ġtail or",
+      "ĠNN LO",
+      "Ġhemod ynamics",
+      "S y",
+      "d ating",
+      "ac hers",
+      "ĠT odd",
+      "ĠM ario",
+      "ĠU GT",
+      "ĠVal ent",
+      "Ġstream lines",
+      "Ġwar rants",
+      "Ġre w",
+      "ĠM ud",
+      "ĠG K",
+      "Ġco ke",
+      "ĠU ran",
+      "Ġgro oves",
+      "ron ate",
+      "ĠRad ius",
+      "ĠSu ite",
+      "atum oral",
+      "Work space",
+      "ĠSynerg istic",
+      "ĠAtheros clerosis",
+      "ma z",
+      "arg max",
+      "sh ield",
+      "ont in",
+      "Ġlist ener",
+      "ocyt oma",
+      "ĠGra v",
+      "ĠJer usalem",
+      "pyrrol idin",
+      "ĠSpr ings",
+      "Ġseaf loor",
+      "Ġd ips",
+      "ist ani",
+      "ob is",
+      "Ġphot odynamic",
+      "AD O",
+      "Ġion isation",
+      "Ġbar n",
+      "igene tics",
+      "Ġeconom ies",
+      "ĠGlac ier",
+      "Ġ ç",
+      "im es",
+      "ĠS asaki",
+      "ch io",
+      "Ġass isting",
+      "ost in",
+      "Ġind iff",
+      "ĠSh ot",
+      "ĠNe uron",
+      "CD D",
+      "ĠCON ST",
+      "ĠBS s",
+      "t as",
+      "ass ociation",
+      "par g",
+      "Ġes cal",
+      "ex ercise",
+      "ĠAd ela",
+      "Ġmy ogenic",
+      "ĠNO AA",
+      "ycl o",
+      "l inal",
+      "ĠH ut",
+      "Ġintro ductory",
+      "Ġheter ochromatin",
+      "Ġchem oresistance",
+      "Ġsimpl ifications",
+      "pyrid in",
+      "Ġamyloid osis",
+      "Ġnanof iber",
+      "ĠSut ton",
+      "ĠResil ience",
+      "P arent",
+      "ĠL MS",
+      "Ġle ts",
+      "ĠEl derly",
+      "Ġirre vers",
+      "she ets",
+      "Eff ects",
+      "Ġellips es",
+      "ĠPhilos ophy",
+      "Ġphot ographic",
+      "HE AD",
+      "Ġrevers ibility",
+      "Ġfed erated",
+      "ĠPhosph oserine",
+      "estim ation",
+      "ĠHum ph",
+      "J son",
+      "Ġf ills",
+      "Ġv erm",
+      "ĠSe if",
+      "with standing",
+      "ĠYam ada",
+      "er ia",
+      "ĠF LA",
+      "ĠV ick",
+      "to id",
+      "ann ier",
+      "Ġcancer ous",
+      "PR INT",
+      "ĠMechan istic",
+      "Ġdust y",
+      "ĠApp end",
+      "y cin",
+      "Ġa zo",
+      "Ġs izing",
+      "Ġc rayfish",
+      "av is",
+      "ĠAd vent",
+      "ĠCommun ist",
+      "ĠIM U",
+      "pix els",
+      "H all",
+      "p ast",
+      "ĠR ous",
+      "ression al",
+      "air d",
+      "ĠAD D",
+      "Ġsummar izing",
+      "Ġelect rol",
+      "St ation",
+      "ĠLy α",
+      "ĠTM EM",
+      "Ġpeptid ase",
+      "D ual",
+      "g it",
+      "ĠB OD",
+      "ĠR ham",
+      "ĠK ak",
+      "Ġread iness",
+      "ĠComp are",
+      "ĠRam os",
+      "sign ificantly",
+      "Ġhair y",
+      "Ġvas opressin",
+      "ĠGuid eline",
+      "B NP",
+      "Ġd irty",
+      "Ġinf imum",
+      "ĠAl ess",
+      "ĠVol cano",
+      "M agn",
+      "Y Y",
+      "ugh lin",
+      "bo platin",
+      "ĠCant or",
+      "Ġclot hes",
+      "ĠR w",
+      "Ġus eless",
+      "ĠK dV",
+      "oper oxidase",
+      "ĠCor rect",
+      "Ġfat ality",
+      "ĠRest riction",
+      "Comput er",
+      "Dep artment",
+      "I l",
+      "g ang",
+      "ĠElect roc",
+      "obar ic",
+      "P hen",
+      "Ġn ed",
+      "ad h",
+      "iss ing",
+      "to oth",
+      "Ġman uscripts",
+      "Ġbi otechnology",
+      "Sup p",
+      "ĠPair wise",
+      "ĠPars ing",
+      "d H",
+      "m elt",
+      "r z",
+      "ĠC atalyst",
+      "em ption",
+      "Ġshow ers",
+      "BL EM",
+      "ĠBro thers",
+      "ban on",
+      "Ġbrac hy",
+      "metall icity",
+      "ĠC IS",
+      "ĠC openhagen",
+      "Ġel der",
+      "Ġfin anc",
+      "odes ic",
+      "Ġdev ise",
+      "Ġsurv ives",
+      "Ġð tÃŀ",
+      "Ġfasc inating",
+      "Ġparall ax",
+      "H OR",
+      "y th",
+      "on ins",
+      "ĠE HR",
+      "ĠG ates",
+      "ob ase",
+      "ĠCon way",
+      "oper ations",
+      "man uel",
+      "ĠAb dominal",
+      "ĠAR G",
+      "ĠGr ö",
+      "Ġphotos ens",
+      "ĠCoul ter",
+      "ĠJul ian",
+      "ĠLev ine",
+      "Ġsarc oidosis",
+      "Ġp illars",
+      "Ġd R",
+      "ĠW G",
+      "Ġspec ulation",
+      "ans ki",
+      "ĠGaussian s",
+      "Sch w",
+      "ĠNam bu",
+      "paren ts",
+      "intr insic",
+      "ĠKend all",
+      "ĠR g",
+      "Ġprot otypical",
+      "bre lla",
+      "Ġtet rap",
+      "ĠPath ophys",
+      "nm t",
+      "Ġerg odicity",
+      "ĠYers inia",
+      "Q O",
+      "ĠI AV",
+      "Ġch ocolate",
+      "Ġconf erring",
+      "Cl NO",
+      "oti a",
+      "Com plete",
+      "ĠAMP A",
+      "ïĢ Ń",
+      "ĠḠ¡",
+      "ĠiP SCs",
+      "ĠApparent ly",
+      "Ġintox ication",
+      "ĠF ather",
+      "per cent",
+      "Ġsh aker",
+      "Ġfin ancing",
+      "Ġgenital ia",
+      "memb ers",
+      "Ġstagn ation",
+      "h matic",
+      "ro red",
+      "Ġcon idia",
+      "atal oader",
+      "ĠNe il",
+      "Ġliter atures",
+      "ĠGl c",
+      "ĠDevelop ments",
+      "differenti ation",
+      "ĠRevis ited",
+      "n il",
+      "t om",
+      "di ol",
+      "ĠAb ell",
+      "Ġplastic s",
+      "ĠLu ke",
+      "adj acent",
+      "ĠBH s",
+      "ĠPosition ing",
+      "ø r",
+      "overex pressing",
+      "E c",
+      "P ref",
+      "or ting",
+      "Ġin ch",
+      "ĠC atherine",
+      "ĠD MP",
+      "ĠO e",
+      "end othelial",
+      "IC ES",
+      "ĠHad ron",
+      "Ġrevis it",
+      "ĠPict ures",
+      "ĠKnock down",
+      "il ian",
+      "ĠA LA",
+      "op amine",
+      "ĠL ah",
+      "cl imate",
+      "Ġdist raction",
+      "ars ki",
+      "ĠAcc ount",
+      "ref lex",
+      "Ġelong ate",
+      "ĠAmb ient",
+      "C x",
+      "M achine",
+      "ĠJ PEG",
+      "Ġclass ifies",
+      "ĠReg ulate",
+      "oplas ia",
+      "inj ury",
+      "neigh bors",
+      "ĠFORM ATION",
+      "F IS",
+      "S z",
+      "ĠO SC",
+      "Ġassemb ling",
+      "Ġintrac erebral",
+      "su pers",
+      "Ġp F",
+      "Ġhe al",
+      "ĠV ries",
+      "arc he",
+      "ĠDec om",
+      "ĠDiff ic",
+      "agent a",
+      "ĠSpir it",
+      "ĠInters ection",
+      "Ġendors ed",
+      "ĠNob el",
+      "i Ïī",
+      "w u",
+      "ph aly",
+      "Ġqu eu",
+      "ĠFor um",
+      "land er",
+      "Ġspectrom etric",
+      "ĠHank el",
+      "ĠC SE",
+      "Ġres umed",
+      "ĠG RE",
+      "AC ES",
+      "CT A",
+      "Ġbeh aved",
+      "Mon itor",
+      "ĠNik on",
+      "ĠCHAR ACTER",
+      "ĠS AL",
+      "ĠI ch",
+      "ĠH SF",
+      "Ġgen otoxic",
+      "ific ance",
+      "ĠCh iang",
+      "ĠZ en",
+      "ĠAr rows",
+      "ĠPD T",
+      "ĠFL ASH",
+      "ocor tex",
+      "onstruct ing",
+      "T reatment",
+      "ĉ ĠĠĠĠĠĠ",
+      "ed ullary",
+      "il ty",
+      "ind entation",
+      "Ġam ended",
+      "Ġfl ed",
+      "roph ication",
+      "ĠGL M",
+      "ĠOper a",
+      "HL H",
+      "L ite",
+      "b mod",
+      "Ġa version",
+      "ĠS weet",
+      "Ġst reptavidin",
+      "ĠP airs",
+      "ug os",
+      "ep oxy",
+      "Ġun specified",
+      "Ġmicro channel",
+      "ĠVictor ian",
+      "C ould",
+      "in formed",
+      "Ġs its",
+      "Ġr x",
+      "Ġne p",
+      "to uch",
+      "RO I",
+      "Ġhead ers",
+      "fl ush",
+      "ĠPath ogenic",
+      "Ġspac ings",
+      "het ti",
+      "Ġmotiv ating",
+      "Ġstake holder",
+      "ĠSymbol ic",
+      "ĠC rani",
+      "Ġdis pute",
+      "Ġass ists",
+      "ind ler",
+      "ĠSp atio",
+      "oh m",
+      "Ġextrap olate",
+      "Ġelabor ation",
+      "ĠGTP ases",
+      "Ġcellul ase",
+      "ĠT uc",
+      "ol ide",
+      "ĠA IM",
+      "pl ast",
+      "ĠB ible",
+      "op oly",
+      "ub o",
+      "ace an",
+      "ĠPen rose",
+      "ĠMap Reduce",
+      "ĠKw on",
+      "W all",
+      "Ġg cd",
+      "ĠAr bitrary",
+      "Pro duct",
+      "ĠGit Hub",
+      "F n",
+      "Ġc k",
+      "ĠA us",
+      "Ġgra ve",
+      "Ġmetabol ically",
+      "Ġlist en",
+      "Ġextrac tions",
+      "ĠTr unc",
+      "ĠRad iology",
+      "cons erving",
+      "ĠComposition al",
+      "report ing",
+      "s ustain",
+      "î Ģ",
+      "ĠO bl",
+      "Ġk N",
+      "St re",
+      "ĠSuper gravity",
+      "ĠPV P",
+      "Ġcivil ian",
+      "ĠTun nel",
+      "Ġhelic opter",
+      "ĠCamb rian",
+      "Ġr g",
+      "ĠU PF",
+      "Ġpol yd",
+      "Ġobserv ability",
+      "con tainer",
+      "nit ros",
+      "ĠCut ting",
+      "Ġdeco uple",
+      "Ġcarbox y",
+      "c row",
+      "Ġc x",
+      "ĠK ell",
+      "Ġproject ors",
+      "Ġmyocardi tis",
+      "itone um",
+      "condition ing",
+      "ĠE TH",
+      "oy ama",
+      "Ġphosph ates",
+      "ĠSub jective",
+      "ĠSer re",
+      "Ġcollagen ase",
+      "Ġvibr ating",
+      "strept omycin",
+      "z hen",
+      "Ġc res",
+      "Ġc ull",
+      "Ġh aven",
+      "ĠG PL",
+      "less ness",
+      "Ġview points",
+      ",, ,",
+      "ochrom ic",
+      "uy ama",
+      "Ġpartnership s",
+      "L ICENSE",
+      "ĠS IFT",
+      "res ources",
+      "ĠG os",
+      "iv ac",
+      "Ġneuro genic",
+      "Ad j",
+      "Ġaqu ifers",
+      "ĠGly cos",
+      "ĠMatthe ws",
+      "ĠFrid ay",
+      "Ġp X",
+      "Ġan te",
+      "ĠF enton",
+      "ĠE ukary",
+      "ib al",
+      "ide ae",
+      "At tention",
+      "ĠPolymer ization",
+      "Ġflip ping",
+      "ĠMedi ates",
+      "Ġstation arity",
+      "Ġecho es",
+      "alid omide",
+      "Ġdeline ation",
+      "Ġn aval",
+      "ĠS omatic",
+      "Ġst ub",
+      "ĠB ever",
+      "ĠR iz",
+      "Ġres ummation",
+      "Ġass ault",
+      "Ġpre existing",
+      "Ġhyper methylation",
+      "Ġconserv ing",
+      "record ed",
+      "am n",
+      "Ġch ow",
+      "ĠK ill",
+      "ĠPro duced",
+      "Ġref s",
+      "ĠEn zymes",
+      "Ġdeep est",
+      "&& &",
+      "ĠFR P",
+      "Ġmil ieu",
+      "Ġirrig ated",
+      "ĠAn atomical",
+      "Ġdiss ect",
+      "ili ensis",
+      "raz olo",
+      "ĠProb able",
+      "sol ve",
+      "conf irmed",
+      "ohydro dynamic",
+      "l ibrary",
+      "ĠC iti",
+      "ĠP HA",
+      "its ky",
+      "Ġelect rone",
+      "na ive",
+      "Ġrib s",
+      "ĠPhot onic",
+      "Ġsubstanti a",
+      "ĠEST IM",
+      "Ġduoden um",
+      "ĠChag as",
+      "ĠSURVE Y",
+      "P ress",
+      "b ian",
+      "Â ¤",
+      "he i",
+      "ĠV AR",
+      "Ġcol ocalization",
+      "Ġpol yl",
+      "Ġdo ugh",
+      "Ġmicro controller",
+      "Ġinternal ized",
+      "Ġdischarg ing",
+      "ĠChlamyd omonas",
+      "or ad",
+      "it el",
+      "ĠW end",
+      "Ġlog its",
+      "Ġelectro cataly",
+      "ĠAm ar",
+      "Ġappreci ably",
+      "Ġneurotrans mitters",
+      "former ly",
+      "c ul",
+      "r ata",
+      "ĠB alk",
+      "ĠO sm",
+      "Ġsympt omatology",
+      "ĠFI ELD",
+      "ĠA Ps",
+      "Ġg ymn",
+      "ĠM MS",
+      "Ġref resh",
+      "ĠInd ices",
+      "Ġimplant able",
+      "sh uffle",
+      "ĠHe ath",
+      "Ġcr isp",
+      "Ġdiss ertation",
+      "ĠUl t",
+      "Des cription",
+      "ĠOrig inally",
+      "ĠF n",
+      "ĠF LOW",
+      "ub ility",
+      "Ġexam s",
+      "ĠSh or",
+      "ĠCd Te",
+      "psy cho",
+      "Ġdict ates",
+      "Ġparench ymal",
+      "ĠPret reatment",
+      "Ġrememb ered",
+      "Ġb ras",
+      "oti d",
+      "Ġrecomm ender",
+      "Ġfles h",
+      "p itch",
+      "in ist",
+      "Ġb title",
+      "Ġl c",
+      "ass igned",
+      "ĠAd visory",
+      "ĠGene va",
+      "weight ing",
+      "ĠAB TS",
+      "Ġhex agon",
+      "ovsk ite",
+      "ĠAPI s",
+      "Ġbol ometric",
+      "Ġmultif aceted",
+      "i ak",
+      "Ġt n",
+      "ĠB ibli",
+      "pro sy",
+      "ĠJ ama",
+      "Ġinf rastructures",
+      "ĠSh are",
+      "Ġintro gression",
+      "trans forms",
+      "Re port",
+      "ĠTR ANS",
+      "ĠEX P",
+      "ĠCB T",
+      "ĠUbiqu itin",
+      "ĠThick ness",
+      "p ub",
+      "t oxin",
+      "ĠF riction",
+      "ĠL AG",
+      "ma ils",
+      "ĠIm mediately",
+      "Ġweak est",
+      "ĠMR S",
+      "Ġcalcare ous",
+      "b ath",
+      "Ġc g",
+      "ur ational",
+      "ter o",
+      "ĠIn oue",
+      "Ġinstruct or",
+      "ac ceptor",
+      "ĠE volving",
+      "ĠL uther",
+      "Ġres igned",
+      "ĠCh ond",
+      "ER F",
+      "Ġselect or",
+      "Ġnewsp apers",
+      "G RA",
+      "S pe",
+      "V H",
+      "r A",
+      "ot ine",
+      "ĠF ACT",
+      "com position",
+      "rid ing",
+      "PC M",
+      "Ġmiddle ware",
+      "ĠGR P",
+      "Ph osph",
+      "ĠEP IC",
+      "spe ech",
+      "vor tex",
+      "ĠHers chel",
+      "am is",
+      "ot ube",
+      "ĠG omez",
+      "com yc",
+      "ĠPh yto",
+      "ĠCons erv",
+      "Ġcav a",
+      "arr hyth",
+      "ĠRestric ted",
+      "il icity",
+      "og ap",
+      "CT P",
+      "ĠLat ino",
+      "atten uated",
+      "m obility",
+      "an en",
+      "Ġn if",
+      "ĠV ideos",
+      "ĠSch ubert",
+      "Fe atures",
+      "oprop anol",
+      "ĠThird ly",
+      "at ula",
+      "ĠC emetery",
+      "enti st",
+      "Ġdel i",
+      "tri als",
+      "Ġgran ulation",
+      "TT G",
+      "Ġtele ost",
+      "mor ill",
+      "or se",
+      "otyp ically",
+      "ĠAb ility",
+      "Amin o",
+      "a queous",
+      "Ġp CO",
+      "ec on",
+      "ĠL ign",
+      "ess els",
+      "Ġform ulating",
+      "ĠTo o",
+      "ĠTrans lational",
+      "ours es",
+      "ubiqu it",
+      "stat istic",
+      "Ġburst ing",
+      "Ġestu aries",
+      "ĠNanoc om",
+      "Ġex ports",
+      "ĠÃ ª",
+      "cont aminated",
+      "Ġtub ing",
+      "Ġautom obile",
+      "Ġmiss ile",
+      "Ġhierarch ically",
+      "Ġrepair s",
+      "ĠImpro ves",
+      "ĠEFFECT S",
+      "Q Ds",
+      "ro z",
+      "ar ic",
+      "Ġpar sed",
+      "ĠBr ink",
+      "Ġprogress ing",
+      "Ġperm Neigh",
+      "A gg",
+      "Z X",
+      "s ink",
+      "Ġw ise",
+      "et ence",
+      "ĠI c",
+      "lo o",
+      "me ida",
+      "Ġpolar iton",
+      "ĠConn ections",
+      "Ġhall marks",
+      "Long rightarrow",
+      "Ġthe ater",
+      "es ar",
+      "Ġre imburs",
+      "Ġlog o",
+      "Ġexc reted",
+      "ĠNo isy",
+      "Ġleak s",
+      "ĠDa ph",
+      "Ġparagraph s",
+      "Ġlandsl ides",
+      "Ġprecl ude",
+      "Ġcoerc ivity",
+      "ĠBurkholder ia",
+      "ĠGó mez",
+      "p rice",
+      "The ory",
+      "sur gery",
+      "f name",
+      "f ailure",
+      "l iness",
+      "re fer",
+      "ri que",
+      "ĠD ogs",
+      "ĠE UV",
+      "ĠV apor",
+      "CS R",
+      "Big gl",
+      "Con straint",
+      "gra vitational",
+      "Ġcombinator ics",
+      "Ġartic ulated",
+      "ĠBax ter",
+      "ĠUltrason ic",
+      "L TE",
+      "l op",
+      "Ġinter atomic",
+      "int uitive",
+      "sim plex",
+      "Ġexperiment ed",
+      "organ izing",
+      "ĠOs aka",
+      "had ron",
+      "Ġdendrim ers",
+      "ĠElse vier",
+      "C IP",
+      "ĠB AP",
+      "ĠAl onso",
+      "art et",
+      "anti s",
+      "Ġextrac orporeal",
+      "Ġpow dered",
+      "ĠSet tings",
+      "et allic",
+      "ĠT EC",
+      "ĠA rena",
+      "Ġan od",
+      "ĠRe agents",
+      "lic enses",
+      "ĠRem ove",
+      "Ġpron unciation",
+      "thin space",
+      "ĠClin ically",
+      "g ative",
+      "Ġw age",
+      "ĠH ap",
+      "ĠG rac",
+      "ff t",
+      "Ġform ate",
+      "inf eld",
+      "ĠQu in",
+      "Ġglomer ul",
+      "W ay",
+      "Ġk ills",
+      "Ġtrans versely",
+      "vari ation",
+      "enn as",
+      "ĠPL L",
+      "Ġinstrument ed",
+      "ĠSpar k",
+      "Ġp illar",
+      "ĠC aval",
+      "ad ers",
+      "iss en",
+      "sc ene",
+      "other m",
+      "é es",
+      "Ġprac ticing",
+      "ĠBM SCs",
+      "ĠFernand ez",
+      "Ġbes ide",
+      "f ew",
+      "ĠC ru",
+      "Ġpro d",
+      "and ers",
+      "az oline",
+      "Ġleg islative",
+      "bal ances",
+      "UR L",
+      "Ġstere otactic",
+      "Ġtrib es",
+      "Ġá¹ ¼",
+      "ĠPAN I",
+      "adren o",
+      "got ten",
+      "c ranial",
+      "ĠM ick",
+      "ĠM MC",
+      "ad iazole",
+      "enti ation",
+      "ĠGl n",
+      "ĠHol stein",
+      "ĠExpl orer",
+      "Ġos se",
+      "arth y",
+      "ĠEV ALU",
+      "adrenal ine",
+      "J J",
+      "ĠC MA",
+      "ĠIn activation",
+      "AB S",
+      "ĠST Z",
+      "Con figuration",
+      "ĠOl factory",
+      "ĠSulf ur",
+      "symbol s",
+      "ĠA STM",
+      "di vergence",
+      "ĠO CR",
+      "med ical",
+      "Ġview er",
+      "Ġbomb ardment",
+      "f air",
+      "n ice",
+      "el berg",
+      "ĠG PT",
+      "ĠK ow",
+      "Ġphot osphere",
+      "Ġlab ile",
+      "ĠSh ang",
+      "Ġhom otopic",
+      "SV D",
+      "bec omes",
+      "Ġgon or",
+      "Ġdeuter on",
+      "Ġphylogen ies",
+      "ĠS AF",
+      "rap ment",
+      "ĠCH F",
+      "Pl an",
+      "ĠLeg al",
+      "ĠFred holm",
+      "Ġshar per",
+      "Ġnanor ib",
+      "ĠBuff alo",
+      "B MD",
+      "Ġl g",
+      "os up",
+      "ĠO PC",
+      "Ġend ophytic",
+      "AT R",
+      "ĠExpression s",
+      "ĠMus ical",
+      "Int roduction",
+      "ĠSL M",
+      "ç ois",
+      "ogly cos",
+      "agl ia",
+      "m ussen",
+      "form ans",
+      "Ġsub structures",
+      "ym pan",
+      "ha e",
+      "sh i",
+      "ĠInter pret",
+      "Ġcat abolic",
+      "Ġoccup ations",
+      "ĠBif urc",
+      "Hydro xy",
+      "ĠKau f",
+      "s leep",
+      "am as",
+      "ĠS f",
+      "ĠM BP",
+      "Ġnon alcoholic",
+      "Ġdisc ordant",
+      "Ġep igen",
+      "PR I",
+      "ĠRed shift",
+      "war n",
+      "Ġlap top",
+      "Ġabras ive",
+      "îĤ Ŀ",
+      "l h",
+      "ĠK nee",
+      "Ġsc rat",
+      "Ġpol oidal",
+      "ĠUn iv",
+      "omy osin",
+      "ĠAug mented",
+      "Ġtaxon om",
+      "Zr O",
+      "Ġphytochemical s",
+      "it en",
+      "ĠP atterson",
+      "th ym",
+      "di hydropy",
+      "Ġk y",
+      "ĠMeta zoa",
+      "ALL Y",
+      "Ġretin oblastoma",
+      "concaten ate",
+      "M ale",
+      "Ġo mission",
+      "ic her",
+      "ĠA zer",
+      "op p",
+      "ple asant",
+      "ning ham",
+      "Ġax ially",
+      "HD FS",
+      "Ġfic tional",
+      "Ï «",
+      "Ġn arc",
+      "Ġunder took",
+      "Ġmicro circ",
+      "ON LY",
+      "IV ER",
+      "ĠCy cles",
+      "Me as",
+      "ĠGriff in",
+      "ĠPli ocene",
+      "Ġp I",
+      "ĠA viation",
+      "ĠC ategorical",
+      "ĠN ils",
+      "Ġresid ency",
+      "ĠLa ur",
+      "Ġpref ers",
+      "Ġassert Equals",
+      "Ġliqu or",
+      "d M",
+      "os perm",
+      "ĠF UT",
+      "Al O",
+      "Ġcyt ometer",
+      "Ġstabil izers",
+      "Ġprem ium",
+      "Ser ial",
+      "ĠWalk ing",
+      "íķ ľ",
+      "Ġconfron ted",
+      "encaps ulated",
+      "C ard",
+      "ĠS eeds",
+      "ab ular",
+      "uk ov",
+      "List ener",
+      "Cho ose",
+      "ĠSj ö",
+      "M ake",
+      "Ġis oc",
+      "am ount",
+      "AT C",
+      "ij a",
+      "Ġsul cus",
+      "ĠMö bius",
+      "ĠPren atal",
+      "Ġ ß",
+      "Ġis ochron",
+      "Ġbe ans",
+      "ĠD ens",
+      "ĠW elling",
+      "ĠO man",
+      "St ats",
+      "ĠVal id",
+      "ĠRew ard",
+      "G K",
+      "Ġâ ©",
+      "Ġelectro poration",
+      "ĠSNR s",
+      "Ġgar lic",
+      "ĠParticip ant",
+      "ĠSplit ting",
+      "ĠMeteor ological",
+      "morill onite",
+      "Ġo edema",
+      "ĠD ots",
+      "ĠCl are",
+      "Ġstar ter",
+      "Ġclim atology",
+      "Ġcomm ence",
+      "Ġfall en",
+      "ĠAu NPs",
+      "attr s",
+      "Ġconsult ant",
+      "tw isted",
+      "Sol ving",
+      "Ġcoerc ive",
+      "ë¡ ľ",
+      "K ar",
+      "Ġs tit",
+      "ĠS SB",
+      "ĠI W",
+      "Ġcan vas",
+      "py ruvate",
+      "methyl sulfanyl",
+      "Ġast rophysics",
+      "ĠTra ditionally",
+      "Ġexcit onic",
+      "w ear",
+      "ĠT in",
+      "ros h",
+      "ĠCl ient",
+      "ĠCor rections",
+      "ĠPop ular",
+      "ĠLiqu ids",
+      "f inder",
+      "Ġst ran",
+      "pl ine",
+      "ore lla",
+      "Ġinc ur",
+      "Ġar che",
+      "Ġmed ically",
+      "M ur",
+      "p eter",
+      "Ġbe verage",
+      "ĠN Ws",
+      "Ġfol ic",
+      "Ġspec ulative",
+      "ĠÃ ±",
+      "Ġrib bons",
+      "ĠPri est",
+      "Qu anti",
+      "Ġundist urbed",
+      "at che",
+      "ass i",
+      "ĠPer forming",
+      "ĠEl ong",
+      "Ġmatch ings",
+      "Ġfranch ise",
+      "g io",
+      "ĠS arg",
+      "Ġab oard",
+      "cycl odextrin",
+      "ĠTH ER",
+      "Ġsatur ate",
+      "ĠKin ematics",
+      "Ġpeptid oglycan",
+      "ĠShel f",
+      "toc opherol",
+      "B ottom",
+      "m ith",
+      "r dx",
+      "z os",
+      "Ġt RNAs",
+      "ut f",
+      "EN A",
+      "Ġless on",
+      "Ġpolar on",
+      "br aska",
+      "Ġath letic",
+      "Ġscram bled",
+      "Ġpursu ing",
+      "Ġbod ily",
+      "Ġc ac",
+      "im en",
+      "ĠI κB",
+      "ĠR ö",
+      "ĠR FC",
+      "ĠL PC",
+      "Ġi Ïī",
+      "Ġdi ary",
+      "Ġqueue ing",
+      "ĠDiver gence",
+      "ĠShig ella",
+      "ĠUltrast ruct",
+      "Ġtri phosphate",
+      "ĠIm plant",
+      "Ġfer rous",
+      "ĠBur ton",
+      "ĠHert z",
+      "f abric",
+      "t uring",
+      "ĠS SM",
+      "og rad",
+      "Ġmet azo",
+      "Ch ang",
+      "Ġadip ogenesis",
+      "Ġnu isance",
+      "Ġanonym ity",
+      "Ġrefriger ator",
+      "ì ľ",
+      "oc tion",
+      "Ġsp aring",
+      "Ġch alcogen",
+      "Ġobserv atory",
+      "Ġbo oster",
+      "ĠAnd ré",
+      "ĠST O",
+      "yr yl",
+      "ĠED X",
+      "ĠDen ver",
+      "Ġhomogen ate",
+      "Call back",
+      "a C",
+      "h ours",
+      "k ova",
+      "ĠA UD",
+      "Ġsp are",
+      "Ġpart ons",
+      "ĠInt ake",
+      "Ġrecogn izable",
+      "ĠGold stein",
+      "Ġstriking ly",
+      "Ġsan itation",
+      "F inder",
+      "G eneration",
+      "b oy",
+      "t am",
+      "ĠR PM",
+      "iv ious",
+      "yl ak",
+      "oph iles",
+      "Ġpri est",
+      "Ġeas iest",
+      "Ġdeliver ies",
+      "El mer",
+      "Ġzircon ium",
+      "ĠMish ra",
+      "Ġâ Ķ",
+      "ĠW DM",
+      "Ġper id",
+      "ĠZ T",
+      "Ġlocal izes",
+      "ĠOR s",
+      "ĠID O",
+      "Ġple asant",
+      "ĠMW CNTs",
+      "ĠJim my",
+      "ĠYe h",
+      "g athered",
+      "k il",
+      "ĠK appa",
+      "Ġcar toon",
+      "ĠHe uristic",
+      "Ġs z",
+      "Ġor ifice",
+      "Ġman nit",
+      "ĠCO MM",
+      "IC K",
+      "Ġfar mer",
+      "ĠSil encing",
+      "Ġprefix es",
+      "q c",
+      "ine urin",
+      "Ġinf lated",
+      "ĠRe z",
+      "Ġhydro dynamical",
+      "Ġoscill ate",
+      "Ġpedest rians",
+      "ĠAngi otensin",
+      "ĠVisc osity",
+      "Ġoligodend rocytes",
+      "Ġparo tid",
+      "Lay out",
+      "rageen an",
+      "Ġ è",
+      "Ġm N",
+      "Ġdo zen",
+      "ex clusion",
+      "Ġpan ic",
+      "ĠPD I",
+      "Ġtw entieth",
+      "ĠElect roph",
+      "Ġmicrobi ology",
+      "Ser ver",
+      "ĠParticip ation",
+      "D ET",
+      "P oss",
+      "ĠN emat",
+      "ĠN RF",
+      "arg uments",
+      "Ġam ylase",
+      "Ġarg v",
+      "Ġresol ves",
+      "Ġrevis ions",
+      "Pack et",
+      "T ools",
+      "Y E",
+      "Ġt ire",
+      "in duction",
+      "as ive",
+      "ton ic",
+      "é m",
+      "car rying",
+      "ĠImmun oblot",
+      "ĠIP F",
+      "Ġdeterior ated",
+      "Ġjuris diction",
+      "rele ased",
+      "osm otic",
+      "Ġrestaur ants",
+      "ï ¸",
+      "ĠN m",
+      "Ġfl ips",
+      "Ġsepar ability",
+      "ĠRec ursive",
+      "Ġpast ure",
+      "ĠÄ ī",
+      "Ġblast ocyst",
+      "M CP",
+      "T b",
+      "u ene",
+      "es ulf",
+      "ess im",
+      "Ġhe n",
+      "ĠK ull",
+      "yl um",
+      "are v",
+      "ues ts",
+      "ĠZ ip",
+      "Ġbo ats",
+      "Com mand",
+      "Cont inu",
+      "ĠBog oliubov",
+      "Ġmannit ol",
+      "K now",
+      "Ð ³",
+      "ĠH ack",
+      "Ġmass ively",
+      "ĠAll oys",
+      "ĠCD P",
+      "ĠStere o",
+      "ĠElectro de",
+      "Ġisofl av",
+      "Ġinteroper ability",
+      "ĠAdela ide",
+      "ĠP PD",
+      "ĠK ou",
+      "Ġdi ap",
+      "Ġcons erve",
+      "Ġaggreg ating",
+      "Gl uc",
+      "Ġî ģ",
+      "Ġg ust",
+      "ĠLe b",
+      "ET IC",
+      "ĠCons ol",
+      "ĠMor ita",
+      "Rel ative",
+      "Ġpale o",
+      "Ġwitness es",
+      "ĠLaure n",
+      "azep ine",
+      "ĠT Y",
+      "ĠI di",
+      "ĠM ent",
+      "ĠR CA",
+      "igen in",
+      "ĠDef ence",
+      "Ġpy rimidine",
+      "ĠTi N",
+      "Ġendot helin",
+      "Ġpand as",
+      "Ġswallow ing",
+      "Ġconges tive",
+      "Ġv inc",
+      "ĠD IP",
+      "ĠH ough",
+      "Ġz w",
+      "ĠKim ura",
+      "represent ations",
+      "ĠProm ote",
+      "ĠTer ry",
+      "Ġhat ched",
+      "look up",
+      "Elect ron",
+      "Ġdorm ancy",
+      "Ġres ign",
+      "Ġval uations",
+      "Ġmake up",
+      "ĠAm y",
+      "CL UD",
+      "SE P",
+      "tub ule",
+      "Ġsoldi er",
+      "ĠT z",
+      "ĠT rump",
+      "ĠK ramer",
+      "con i",
+      "Ġeng raft",
+      "Ġvacu ole",
+      "Ġreplic ating",
+      "iton itis",
+      "ĠBacter i",
+      "vacc inated",
+      "ol t",
+      "ĠA hn",
+      "Ġan em",
+      "ĠB IT",
+      "ge o",
+      "Ġmicro gravity",
+      "ĠSh ir",
+      "ĠWorld wide",
+      "Ġign or",
+      "ĠË ĩ",
+      "Ġlubric ation",
+      "j ava",
+      "v t",
+      "Ġ yl",
+      "Ġh ills",
+      "ĠF OL",
+      "Ġbasal tic",
+      "Ne ill",
+      "ĠEthiop ian",
+      "ĠNOT CH",
+      "ĠMOS FET",
+      "le aving",
+      "ĠP ter",
+      "ĠW eld",
+      "ap le",
+      "Ġsand wic",
+      "Ġaz ide",
+      "ĠStim uli",
+      "Ġl izard",
+      "ĠC inc",
+      "ĠH ain",
+      "ical s",
+      "Ġcontact ing",
+      "ĠMar x",
+      "Ġpsych otherapy",
+      "ĠRet in",
+      "Ġcatheter ization",
+      "ĠNanopar ticle",
+      "ĠT CC",
+      "ver mectin",
+      "ĠB ASE",
+      "Ġnot or",
+      "Ġelectron ically",
+      "ster oid",
+      "Ġbil aterally",
+      "Ġneph ritis",
+      "Ġirr itation",
+      "ĠProlong ed",
+      "Y our",
+      "he uristic",
+      "ur geon",
+      "Ġleft most",
+      "ĠRE VIEW",
+      "Ġgast rectomy",
+      "ENT IAL",
+      "Me ans",
+      "ĠDys on",
+      "Ġbrand s",
+      "yield s",
+      "mercap to",
+      "r ub",
+      "oun cement",
+      "err no",
+      "Ġview ers",
+      "but an",
+      "ĠMal ay",
+      "ylind rical",
+      "Ġpromin ently",
+      "Ġescap ing",
+      "Ġquer ying",
+      "Stor age",
+      "F os",
+      "c odec",
+      "Ġc M",
+      "str ates",
+      "gl ove",
+      "ĠTra jectories",
+      "Ġster ol",
+      "Ġhemat opoiesis",
+      "Ġcup rates",
+      "O k",
+      "d aily",
+      "ĠB IO",
+      "ĠL ICENSE",
+      "ell ations",
+      "ass y",
+      "SU RE",
+      "Ġep inephrine",
+      "Ġdown wards",
+      "cor ner",
+      "assert True",
+      "ĠẠij",
+      "ĠSou za",
+      "M AG",
+      "por ph",
+      "Ġeff luents",
+      "lo em",
+      "oad dition",
+      "obut yl",
+      "eles tial",
+      "F em",
+      "m pi",
+      "ĠR s",
+      "ell ates",
+      "ĠK ag",
+      "Ġunc oupled",
+      "Ġleg umes",
+      "Ġomit ting",
+      "Ã »",
+      "ĠT ABLE",
+      "hal ed",
+      "ĠÅ ģ",
+      "Ġmis fit",
+      "Ġmol ars",
+      "otechn ological",
+      "Mark ov",
+      "Ġpra ised",
+      "ĠD ab",
+      "ĠV ij",
+      "enti lation",
+      "ĠCh atter",
+      "Ġbo iled",
+      "Ġcat ches",
+      "annot ation",
+      "Sign al",
+      "Ġlever ages",
+      "Ġadvis ory",
+      "s ong",
+      "on dition",
+      "Ġf ug",
+      "ra ps",
+      "ĠM CD",
+      "par ticip",
+      "ob ian",
+      "Ġcoun sel",
+      "ĠPR P",
+      "edi ol",
+      "ĠÅ ¨",
+      "Ġbr uce",
+      "Sh anghai",
+      "Data Frame",
+      "ĠCorrespond ingly",
+      "Ġacryl amide",
+      "f ellow",
+      "l ob",
+      "ig t",
+      "ĠC rystallization",
+      "Ġind omethacin",
+      "ĠPD R",
+      "gi ate",
+      "ĠPan els",
+      "complex es",
+      "ĠNic ol",
+      "Ġfoli ar",
+      "c ubic",
+      "Ġd E",
+      "ĠC CM",
+      "pl ating",
+      "Ġres istors",
+      "ĠG az",
+      "Ġover turn",
+      "Ġrep ress",
+      "Ġpot s",
+      "ĠPI K",
+      "Ġderm is",
+      "Rep resent",
+      "ĠAnders son",
+      "Ġretrotrans pos",
+      "A SA",
+      "C ounter",
+      "T et",
+      "im in",
+      "per formed",
+      "ĠN ept",
+      "Ġhe el",
+      "rol d",
+      "ai res",
+      "Ġread ability",
+      "Ġbenef ited",
+      "Ġpuls ation",
+      "ĠBal ancing",
+      "Ġevapor ator",
+      "Ġeig ens",
+      "ĠH ospit",
+      "Ġtra ils",
+      "ĠCo ordinate",
+      "acc ase",
+      "ĠHR MS",
+      "sign aling",
+      "ĠNP Y",
+      "Ġamelior ated",
+      "tu ples",
+      "Ġmetas urface",
+      "ĠFrances co",
+      "Ġspo of",
+      "îĹ ¸",
+      "F u",
+      "J K",
+      "e j",
+      "Ġg oss",
+      "ĠH im",
+      "Ġprior itized",
+      "Ġmes othelioma",
+      "idx s",
+      "Ġrecon naissance",
+      "Ġlam ps",
+      "ãĢ Ĥ",
+      "Ġreform ulation",
+      "Ġdeli rium",
+      "ĠN PR",
+      "ĠG amb",
+      "ill as",
+      "---- -",
+      "Ġdr illed",
+      "ĠGen otyping",
+      "ĠBl ank",
+      "Ġprop eller",
+      "Ġcere als",
+      "ĠAir borne",
+      "ĠPhot ocatalytic",
+      "ĠCav ity",
+      "Ġdol phins",
+      "Ġsg RNA",
+      "underst ood",
+      "e ous",
+      "Ġs ax",
+      "ol ayer",
+      "ĠP end",
+      "ĠG ET",
+      "cl ed",
+      "ĠÃ ¼",
+      "Ġcyt osine",
+      "Ġparallel ization",
+      "MM s",
+      "ĠOrgan isation",
+      "Mod els",
+      "Ġaccommod ated",
+      "Ġchol est",
+      "Ġin activity",
+      "ĠB oss",
+      "ĠG CS",
+      "Ġso aked",
+      "ĠSec reted",
+      "Ġvacu olar",
+      "zo an",
+      "ĠGre ene",
+      "Ġbol t",
+      "b j",
+      "ĠT all",
+      "Ġst or",
+      "ĠM ob",
+      "Ġbl urred",
+      "IN O",
+      "ĠMet allic",
+      "uff s",
+      "ĠNOT E",
+      "Ġsonic ated",
+      "obuty ric",
+      "Ġt DCS",
+      "ĠN es",
+      "osp ir",
+      "we igh",
+      "ĠReg ulator",
+      "Ġhem olysis",
+      "Ġsound ing",
+      "Ġcruc iate",
+      "Ġcaps aicin",
+      "ĠTy rosine",
+      "ĠT ell",
+      "ĠP EP",
+      "ĠR c",
+      "ĠE ating",
+      "ĠGo als",
+      "ure t",
+      "Ġear n",
+      "Ġcolle ges",
+      "Ġchemo attract",
+      "Ġá» ¹",
+      "ĠEch ocardi",
+      "F ort",
+      "s odium",
+      "am ined",
+      "ĠN PP",
+      "ĠK alu",
+      "Ġdec ipher",
+      "tet ramethyl",
+      "ĠOP N",
+      "stra ight",
+      "Ġtemp ered",
+      "ĠHind u",
+      "ĠOrd inary",
+      "ĠACh E",
+      "J NK",
+      "f os",
+      "v cpu",
+      "en amide",
+      "ĠC rack",
+      "ap ical",
+      "Ġanti serum",
+      "tri plet",
+      "dec ision",
+      "Ġcanc els",
+      "ĠPM N",
+      "Ġporph y",
+      "ĠD SA",
+      "Ġsub matrix",
+      "Ġj as",
+      "Ġrep tiles",
+      "ĠSo on",
+      "ĠStat istically",
+      "Ġlever aged",
+      "Willi ams",
+      "F LD",
+      "f olk",
+      "Ġb ang",
+      "ĠS CL",
+      "ass es",
+      "Ġtend ons",
+      "found ed",
+      "ĠRick etts",
+      "in set",
+      "Ġsp un",
+      "Ġun ramified",
+      "Ġra pe",
+      "ĠZ Z",
+      "ĠNe bula",
+      "Ġthromb otic",
+      "ĠBor on",
+      "ĠArg on",
+      "pool ing",
+      "ĠMarg inal",
+      "Ġfellow ship",
+      "Ġerythrop oietin",
+      "mathp zc",
+      "x L",
+      "ĠS ik",
+      "ĠB ayer",
+      "Ġover dose",
+      "ĠCO I",
+      "ĠLes ions",
+      "ĠCompe titive",
+      "ĠODE s",
+      "w rap",
+      "ach lor",
+      "Ġsub ordinate",
+      "ĠY Ba",
+      "head ed",
+      "Ġgrass es",
+      "Ġbir ational",
+      "ĠJeff rey",
+      "Ġmold ing",
+      "Ġlid ocaine",
+      "ĠFOX O",
+      "termin ated",
+      "ĠâĩIJ âĩĴ",
+      "ĠM EL",
+      "tic ulum",
+      "Ġr é",
+      "Ġcl aud",
+      "Ġj amming",
+      "St atic",
+      "Ġtribut ary",
+      "at et",
+      "ed onia",
+      "ĠC MP",
+      "ĠV N",
+      "rep resents",
+      "SO URCE",
+      "uck land",
+      "ĠInterest s",
+      "ĠNan oscale",
+      "ocon jug",
+      "Ġcatalog ues",
+      "ĠActin obacteria",
+      "F ixed",
+      "b asal",
+      "Ġanti parallel",
+      "Ġconf using",
+      "Ġmark ings",
+      "Ġdistinc tions",
+      "ĠHu a",
+      "ĠWat ts",
+      "Ġnanoflu id",
+      "Ġdiffract ometer",
+      "L ater",
+      "m igration",
+      "Ġco planar",
+      "Ġhyp omethyl",
+      "PD S",
+      "SO s",
+      "Cor respond",
+      "Ġelucid ating",
+      "IZ ED",
+      "E Vs",
+      "g art",
+      "m Tc",
+      "ĠT UR",
+      "ur acies",
+      "Ġfollow er",
+      "Ġhaz e",
+      "OU TPUT",
+      "ĠMyel oid",
+      "BUFF ER",
+      "C amp",
+      "an im",
+      "ĠT ES",
+      "ĠC ricket",
+      "ĠP aired",
+      "ĠP AGE",
+      "ĠB id",
+      "Ġy rs",
+      "Ġend ow",
+      "irc ase",
+      "ĠSup ported",
+      "Ġleaf let",
+      "ĠProm oter",
+      "Ġconvinc ed",
+      "l iers",
+      "he ra",
+      "Ġs eller",
+      "ag reement",
+      "Ġun ary",
+      "onstr ucted",
+      "Ġrest rained",
+      "ĠPet ersen",
+      "Anal og",
+      "Ġexacerb ations",
+      "Ġperfor ated",
+      "ti ds",
+      "ĠM SH",
+      "ou i",
+      "ĠCor i",
+      "ĠCr uc",
+      "Ġfract uring",
+      "Ġinfer tile",
+      "ĠPRO BLEM",
+      "ĠFried mann",
+      "Ġspectrophot ometry",
+      "ERG Y",
+      "ot us",
+      "pro posed",
+      "ĠMo isture",
+      "ĠNo ether",
+      "ĠLa unch",
+      "ĠLear n",
+      "Ġven a",
+      "Ġfasc i",
+      "Ġquies cence",
+      "ĠP rand",
+      "ĠCon vert",
+      "Ġtri age",
+      "AN E",
+      "Ġfeed stock",
+      "ĠdB m",
+      "Ġneo formans",
+      "G SE",
+      "ĠA PE",
+      "Ġcardi ometabolic",
+      "Ġmagnet ometer",
+      "En vironment",
+      "Ġfulf ills",
+      "ĠMang anese",
+      "B MP",
+      "ĠR atios",
+      "ist able",
+      "ass ume",
+      "Ġresp ected",
+      "Ġsc ars",
+      "Ġsup porters",
+      "ĠAug mentation",
+      "Ġglycos ylated",
+      "ĠUltra fast",
+      "Ġdemethyl ation",
+      "metast atic",
+      "c ylinder",
+      "Ġh ang",
+      "ĠM AV",
+      "dis joint",
+      "pha rose",
+      "ĠLe banon",
+      "PI s",
+      "lab eling",
+      "Ġneutral ino",
+      "ĠSO CS",
+      "xc b",
+      "ĠTerr itory",
+      "ĠPolic ies",
+      "K ing",
+      "Ġall ied",
+      "Ġsatur ates",
+      "mus cle",
+      "odim ers",
+      "Ġb t",
+      "ĠH ang",
+      "ĠE b",
+      "Ġch imer",
+      "Ġnot ational",
+      "Ġcol der",
+      "ult z",
+      "trans verse",
+      "HO UT",
+      "ĠKar ls",
+      "tors ion",
+      "J I",
+      "Ġm ari",
+      "em on",
+      "Ġlogarithm ically",
+      "ĠâIJ ¦",
+      "ìĿ Ħ",
+      "Ġa eration",
+      "Ġs oma",
+      "ĠS omal",
+      "Ġsp oil",
+      "di ver",
+      "Ġbreak points",
+      "ĠHar mon",
+      "Ġpharmac ologic",
+      "ĠM osquito",
+      "ĠMod ifications",
+      "Ġadj o",
+      "ĠPa pers",
+      "gener ally",
+      "ïĺ ¹",
+      "T ARGET",
+      "ĠP rix",
+      "oc aps",
+      "ĠE in",
+      "Ġmicro grid",
+      "ĠInter play",
+      "Ġcop ying",
+      "Al pha",
+      "ĠSl ope",
+      "ĠLip ofectamine",
+      "hig hest",
+      "D RO",
+      "ĠH ipp",
+      "Ġsh aken",
+      "Ġunder line",
+      "Ġfil med",
+      "mat urity",
+      "ict ure",
+      "IL S",
+      "Sp an",
+      "Ġinver ters",
+      "QU E",
+      "determ ining",
+      "Ġeosin ophilic",
+      "D Y",
+      "ĠL ID",
+      "ĠG ig",
+      "Ġintra epithelial",
+      "Nb O",
+      "fre edom",
+      "Ġass ured",
+      "ĠAr che",
+      "ĠSub stitution",
+      "ĠSri vastava",
+      "ĠMoz amb",
+      "Ġa ro",
+      "or c",
+      "ĠI brahim",
+      "ĠD ST",
+      "Ġab l",
+      "Ġx er",
+      "ount able",
+      "Ġloss less",
+      "Ġconcentr ating",
+      "Ġstain s",
+      "ĠSol ve",
+      "continu ity",
+      "ĠTor r",
+      "Ġpit falls",
+      "best os",
+      "Other wise",
+      "adhy ay",
+      "b ard",
+      "ĠC AA",
+      "ode tic",
+      "Ġast hmatic",
+      "Ġrational ity",
+      "ĠYork shire",
+      "neighbor hood",
+      "Ġhero in",
+      "Ġscatt erers",
+      "ĠH earing",
+      "ĠE FT",
+      "ĠN urses",
+      "ĠG LI",
+      "ĠZ eta",
+      "ĠNe igh",
+      "Ġvent ure",
+      "Ġtoxic ological",
+      "Ġroll s",
+      "f v",
+      "Ġc rick",
+      "Ġd Ïī",
+      "av ia",
+      "eld er",
+      "Ġinv ade",
+      "ext racted",
+      "ML P",
+      "ĠPA I",
+      "ĠMell itus",
+      "Ġbruce i",
+      "g pio",
+      "em otional",
+      "ĠD ale",
+      "ĠE z",
+      "Ġtrans activation",
+      "Ġquanti les",
+      "Ġnucle osynthesis",
+      "ĠAm el",
+      "Ġchrom ophore",
+      "Ġliter ally",
+      "band width",
+      "ato hepatitis",
+      "Ġultraf iltration",
+      "Mart in",
+      "Ġangio plasty",
+      "inser tion",
+      "D an",
+      "s queeze",
+      "us r",
+      "uc onazole",
+      "ĠF AR",
+      "Ġsh adows",
+      "Ġim itation",
+      "ĠK ann",
+      "hes i",
+      "Ġmic ellar",
+      "ves ter",
+      "ĠPer se",
+      "acet amol",
+      "GR APH",
+      "ĠAI PS",
+      "Ġprompt ly",
+      "anch or",
+      "Ġischa emia",
+      "p ump",
+      "Ġm afic",
+      "Ġl azy",
+      "ĠC EL",
+      "ĠG orenstein",
+      "ĠW GS",
+      "Ġsign ifies",
+      "Ġspl ines",
+      "determ ination",
+      "Ġrelay ing",
+      "piper azine",
+      "Ġsyncy tial",
+      "ĠA ub",
+      "ĠD X",
+      "Ġorth otopic",
+      "ĠLink age",
+      "Ġharmon y",
+      "ĠKaz akh",
+      "ĠVlad imir",
+      "Ġp ray",
+      "im olar",
+      "Ġgra yscale",
+      "Ġanaly st",
+      "ĠTrans l",
+      "Ġmen iscus",
+      "ĠMed ica",
+      "osa urus",
+      "ĠÐ ²",
+      "Ġinfiltr ated",
+      "Ġâĸ ³",
+      "Ġsacc ades",
+      "Ġdisent angle",
+      "H art",
+      "f ined",
+      "Ġb icycle",
+      "os itory",
+      "un likely",
+      "ere phthal",
+      "ĠL ia",
+      "Ġgroup ings",
+      "Ġcategor ize",
+      "Ġbioge ography",
+      "ĠAPPRO ACH",
+      "ĠN ing",
+      "ĠG rap",
+      "vers a",
+      "Ġradi ologists",
+      "ĠRec ording",
+      "Ġbo iler",
+      "add ers",
+      "C andid",
+      "M Q",
+      "Ġb w",
+      "ĠS ector",
+      "ĠH IT",
+      "ĠE SCC",
+      "ess ence",
+      "ore an",
+      "est yles",
+      "SU CCESS",
+      "ne in",
+      "ult ra",
+      "ram p",
+      "Th omas",
+      "ĠPre par",
+      "ĠInstit ut",
+      "Ġherb icide",
+      "ĠCha otic",
+      "Ġsph incter",
+      "Ġcompac tifications",
+      "C lear",
+      "Tr p",
+      "Dec oder",
+      "Ġsap phire",
+      "ĠIda ho",
+      "per sing",
+      "ch iral",
+      "ĠDis charge",
+      "According ly",
+      "ĠArth ritis",
+      "ĠJane iro",
+      "n j",
+      "ĠK d",
+      "Ġout lets",
+      "Ġsuscepti bilities",
+      "Ġdiver ged",
+      "Ġroll er",
+      "su fficient",
+      "clust ering",
+      "ĠTeh ran",
+      "Ġt b",
+      "bl ank",
+      "Ġdigit ally",
+      "Ġnecro tizing",
+      "F ALSE",
+      "Ġwh or",
+      "err als",
+      "ĠMo tivated",
+      "enz ae",
+      "ĠRef inement",
+      "Ġtick et",
+      "Ġprotr usions",
+      "ĠDonald son",
+      "ĠB eth",
+      "Ġsp uttered",
+      "Ġaut ocrine",
+      "cop ene",
+      "Ġcoll ar",
+      "Ġupper most",
+      "Ġoxygen ated",
+      "Int ro",
+      "âĨ IJ",
+      "ĠHip po",
+      "Ġd une",
+      "id ines",
+      "ĠH ä",
+      "Ġreg i",
+      "Ġno is",
+      "Ġphot odiode",
+      "ĠFe b",
+      "mut ated",
+      "ĠCF L",
+      "step ping",
+      "Se lection",
+      "ĠWeb ster",
+      "ĠHER A",
+      "indic ating",
+      "Ġtraine es",
+      "R ot",
+      "ĠF AK",
+      "ĠAs n",
+      "Ġfat s",
+      "fol iation",
+      "Ġartic ulation",
+      "Ġcus ps",
+      "ĠJenn ifer",
+      "Ġin timately",
+      "ĠP ing",
+      "so v",
+      "ox ious",
+      "hyd rate",
+      "ĠArch ives",
+      "Gon z",
+      "Ġ é",
+      "Ġch l",
+      "ĠO LS",
+      "cop h",
+      "Ġair line",
+      "Ġfo etal",
+      "ĠRoll ing",
+      "ĠGENER AL",
+      "O NAL",
+      "ag ons",
+      "ĠD orsal",
+      "Ġr itual",
+      "but yrate",
+      "ogl ut",
+      "Ġhex a",
+      "ĠSy ria",
+      "Ġont ogeny",
+      "ĠFB G",
+      "cover age",
+      "Ġtachy on",
+      "ĠPerman ent",
+      "l um",
+      "Ġs v",
+      "Ġo o",
+      "en ergetic",
+      "al titude",
+      "In c",
+      "ĠNe braska",
+      "ĠRE SP",
+      "Ġdys biosis",
+      "Ġmarket ed",
+      "oxic illin",
+      "ĠBroad cast",
+      "racycl o",
+      "ĠFif teen",
+      "ĠNar ayan",
+      "Ġlett uce",
+      "ore a",
+      "Ġinter cepts",
+      "Ġwork station",
+      "ĠPl ains",
+      "CC L",
+      "Ġorient able",
+      "ĠBo osting",
+      "ĠSO I",
+      "ĠCheck ing",
+      "ĠFIF O",
+      "Ġin sets",
+      "ĠS RT",
+      "Ġac rom",
+      "own er",
+      "MI X",
+      "ĠAr b",
+      "Ġfa eces",
+      "ĠCarl son",
+      "Ġperiv ascular",
+      "infiltr ating",
+      "Ì ħ",
+      "Ġm alle",
+      "oc ate",
+      "ĠB old",
+      "unc tive",
+      "ex cess",
+      "Ġlo osen",
+      "Ġprior itization",
+      "Ġannot ate",
+      "Ġgram mars",
+      "Ġb red",
+      "Ġex ocytosis",
+      "ĠD ahl",
+      "ath yroidism",
+      "vel i",
+      "Ġop ted",
+      "Ġsm oked",
+      "ĠPl ates",
+      "EM G",
+      "RO W",
+      "IF IC",
+      "OL S",
+      "oreg ulatory",
+      "Ġwhisk ers",
+      "secret ase",
+      "Ġexagger ated",
+      "ĠB ib",
+      "de formed",
+      "Ġz ur",
+      "rop ine",
+      "Ġpair ings",
+      "chrom osome",
+      "Ele ments",
+      "prior ity",
+      "Ġlyophil ized",
+      "ĠChaud h",
+      "W ilk",
+      "ĠC ation",
+      "ot ta",
+      "Ġnon convex",
+      "Ġdep olymer",
+      "MM ARY",
+      "Cont rolled",
+      "carbox y",
+      "Ġaugment ing",
+      "Ġappoint ments",
+      "Ġtravers ed",
+      "ĠF letcher",
+      "Ġexp iratory",
+      "Ġele phant",
+      "ĠBl ocks",
+      "ĠFlu ids",
+      "wall s",
+      "incre ased",
+      "propan amide",
+      "ĠAka ike",
+      "ĠC BM",
+      "ĠE cho",
+      "ad missible",
+      "Ġdis assembly",
+      "Ġar Xiv",
+      "ick e",
+      "LI ST",
+      "phen otype",
+      "ĠProv incial",
+      "leg end",
+      "P AS",
+      "r nn",
+      "s and",
+      "Ġb ariatric",
+      "ĠP ush",
+      "ĠAp oE",
+      "cap rolactone",
+      "model ing",
+      "ĠÅ µ",
+      "Ġsupercapac itors",
+      "or on",
+      "Ġp K",
+      "st rophy",
+      "ĠS uc",
+      "und a",
+      "te am",
+      "Ġit iner",
+      "Ġsw ell",
+      "ĠBio active",
+      "ĠIndic ators",
+      "ĠI FT",
+      "ĠD K",
+      "Ġcap it",
+      "sh apes",
+      "Ġtrac hea",
+      "delay ed",
+      "ĠGuang dong",
+      "L epid",
+      "T GA",
+      "h k",
+      "ol on",
+      "ogen in",
+      "ĠAc k",
+      "Ġlog ically",
+      "cont ributions",
+      "ĠCle avage",
+      "hur st",
+      "b dd",
+      "ST D",
+      "ĠF ut",
+      "te k",
+      "ĠIn her",
+      "Ġchem is",
+      "Ġbreak point",
+      "estim ates",
+      "ĠOtt oman",
+      "ĠNaf ion",
+      "WID TH",
+      "Ġs izable",
+      "ĠT su",
+      "emb olic",
+      "Ġright most",
+      "ĠCell ulose",
+      "iction aries",
+      "ĠMy coplasma",
+      "ĠBur gers",
+      "ĠKepler ian",
+      "U CTION",
+      "V B",
+      "Ġb cc",
+      "ra id",
+      "END IX",
+      "Ġsc oping",
+      "ĠPR I",
+      "ĠCd Se",
+      "ĠGre edy",
+      "ĠHam mer",
+      "ĠBacter oides",
+      "inform ative",
+      "Ġresemb led",
+      "yll ium",
+      "T wenty",
+      "Ġp ounds",
+      "Ġun polarized",
+      "Ġconfig ure",
+      "Ġtranscription ally",
+      "Ġmicros cale",
+      "ĠPut ting",
+      "Ġpyr rol",
+      "ĠLAS SO",
+      "f iltration",
+      "Ġt ech",
+      "per forming",
+      "Al ong",
+      "ĠCT LA",
+      "Ġauthor ization",
+      "UR AL",
+      "Ġleak y",
+      "Op tical",
+      "ĠReve al",
+      "ĠHUV ECs",
+      "W u",
+      "c ustom",
+      "di ble",
+      "Ġï£ ¦",
+      "CD Cl",
+      "Ġemph ys",
+      "Ne ut",
+      "coll agen",
+      "necess arily",
+      "ĠRoot s",
+      "P ose",
+      "T u",
+      "Ġcl ue",
+      "Ġperturb ing",
+      "ĠHel ium",
+      "ĠComb ustion",
+      "n itrogen",
+      "am plified",
+      "pro ve",
+      "ĠSo ils",
+      "normal ization",
+      "ĠCH OP",
+      "ĠMc Le",
+      "Ġstri kes",
+      "Ġcrop ped",
+      "ĠKu o",
+      "Ġvag al",
+      "Ġdin ucleotide",
+      "ĠIsa ac",
+      "ĠL OX",
+      "Ġdirection ality",
+      "Ġchem oradiotherapy",
+      "calc ulus",
+      "ĠMoh ammed",
+      "m apped",
+      "Ġre forms",
+      "Ġre ordering",
+      "ĠB m",
+      "ĠE SCs",
+      "ĠN UC",
+      "th aw",
+      "Ġnan oporous",
+      "Ġtrain able",
+      "ĠAT T",
+      "fe ats",
+      "OF DM",
+      "ĠSH P",
+      "ĠRich ter",
+      "Ġspray ed",
+      "ĠJeff erson",
+      "F OX",
+      "b h",
+      "ot te",
+      "Ġle iomy",
+      "osp ores",
+      "specific ity",
+      "ĠRef er",
+      "ĠHa as",
+      "M ove",
+      "M aterials",
+      "t ec",
+      "u tility",
+      "en tional",
+      "ĠM PP",
+      "ch ond",
+      "Ġse epage",
+      "Ġpe ach",
+      "ĠÎĶ t",
+      "embry onic",
+      "Y an",
+      "Ġlip osomal",
+      "ĠVal encia",
+      "ĠEnd o",
+      "ĠPA O",
+      "Ġdial ect",
+      "Ġchond rocyte",
+      "ĠMill imeter",
+      "ĠRegular ity",
+      "dest roy",
+      "ĠCond ensation",
+      "Bay es",
+      "abund ance",
+      "Ġd U",
+      "ĠS SI",
+      "ĠH AND",
+      "Ġcons ulted",
+      "Ġsup pliers",
+      "Ġdem o",
+      "reg istered",
+      "Ġmicros omal",
+      "Ġlam bs",
+      "respons iveness",
+      "D y",
+      "G AS",
+      "U ME",
+      "Ġa ero",
+      "Ġcal modulin",
+      "Ġcalc ined",
+      "Ġins ula",
+      "ĠMe i",
+      "ĠRE AL",
+      "Ġcontrac tible",
+      "ĠEss entially",
+      "Ġgam ing",
+      "Ġspill over",
+      "resid ues",
+      "â İ",
+      "ĠE MC",
+      "ĠSD E",
+      "ĠSer ine",
+      "eck i",
+      "ĠPrinc eton",
+      "ĠBACK GROUND",
+      "m asks",
+      "ĠL om",
+      "ff ield",
+      "ef itinib",
+      "Ġpat ents",
+      "ĠBe z",
+      "load s",
+      "Ġgon adal",
+      "Ġnitro cellulose",
+      "âĻ Ĥ",
+      "Ġth rown",
+      "Ġrec tification",
+      "min a",
+      "isc id",
+      "ĠBi obank",
+      "param agnetic",
+      "GS K",
+      "ĠDeriv ative",
+      "criter ion",
+      "ĠMonth ly",
+      "ë ¥",
+      "ĠS ichuan",
+      "Ġimmun ologic",
+      "Ġheter otic",
+      "ĠMc Cl",
+      "ĠSM ART",
+      "ĠBatter ies",
+      "Ġpremi ered",
+      "Ġcryopres ervation",
+      "N u",
+      "val ho",
+      "Ġfl otation",
+      "top ological",
+      "ĠNan jing",
+      "Ġju xt",
+      "ĠFed er",
+      "Ġprofound ly",
+      "c ad",
+      "i enced",
+      "ch uk",
+      "ĠIn g",
+      "ĠK SHV",
+      "amin obenz",
+      "ĉĉĉ ĠĠĠ",
+      "Ġmeta ph",
+      "ĠEpid emic",
+      "ĠAssoci ate",
+      "Ġsacc ade",
+      "Ġd awn",
+      "Ġre heating",
+      "Ġsp ell",
+      "frac tive",
+      "ĠTo olkit",
+      "Ġrecogn ise",
+      "path ogen",
+      "Ġophthal mic",
+      "Ġquer ied",
+      "t hens",
+      "ith ine",
+      "um ably",
+      "Ġstr ides",
+      "ha ul",
+      "Ġpass ion",
+      "Ġdys functions",
+      "By te",
+      "Ġca esarean",
+      "pre y",
+      "ĠHor se",
+      "ĠGAB AA",
+      "N atural",
+      "k os",
+      "in ators",
+      "od ings",
+      "AR RAY",
+      "Ġun ipotent",
+      "Ġelect romy",
+      "com part",
+      "Li u",
+      "encephal ic",
+      "ĠCOMP AR",
+      "Ġsymbion ts",
+      "ivac aine",
+      "O I",
+      "P VA",
+      "ĠN VIDIA",
+      "cal ibrated",
+      "Ġqu est",
+      "NA D",
+      "ĠX yl",
+      "Ġpharmac ist",
+      "direct ly",
+      "Ġquadrup le",
+      "ethan one",
+      "ĠBulg aria",
+      "Ġovip osition",
+      "r uns",
+      "Ġn ociceptive",
+      "Ġas exual",
+      "SU LT",
+      "Ġwould n",
+      "ĠInd ustries",
+      "abil izing",
+      "ĠComp ressive",
+      "CO OH",
+      "US H",
+      "ki ewicz",
+      "Ġign eous",
+      "Ġdisapp oint",
+      "ĠCK M",
+      "ĠDiagram s",
+      "ĠF lam",
+      "ĠG ould",
+      "Ġco enzyme",
+      "Ġpar an",
+      "ĠÂ ¶",
+      "Ġprogram mer",
+      "ĠTrans forming",
+      "Ġmus carinic",
+      "onucle otide",
+      "FI ELD",
+      "ĠFu ji",
+      "Ġnond ec",
+      "Ġblank et",
+      "Ġpredis posing",
+      "ĠTrig ger",
+      "Ġwel come",
+      "F amily",
+      "U INT",
+      "h fill",
+      "t vb",
+      "ĠB att",
+      "Ġun met",
+      "ĠAp o",
+      "oti ent",
+      "Ġfund us",
+      "ĠLear ned",
+      "Ġintr usions",
+      "Ġsolub ilization",
+      "fund amental",
+      "ĠSanti ago",
+      "Ġh pi",
+      "th row",
+      "ĠIn to",
+      "time out",
+      "Ġthick ened",
+      "ias m",
+      "Ġgravit ino",
+      "bran ched",
+      "V III",
+      "Ġo ch",
+      "Ġg ym",
+      "ĠK rylov",
+      "Ġcorrec tive",
+      "ĠInstit ution",
+      "Ġcrim es",
+      "ĠBacteroid etes",
+      "ĠE hr",
+      "Ġse ated",
+      "rol izumab",
+      "Ġfactor ized",
+      "rot ational",
+      "Ġadministr ators",
+      "âĭ Ĩ",
+      "ineral ization",
+      "l ining",
+      "â Ĺ",
+      "ur ai",
+      "ĠF AP",
+      "ĠF isheries",
+      "ĠE SO",
+      "tem per",
+      "Big gr",
+      "ĠAltern ating",
+      "t win",
+      "am atsu",
+      "Ġint rad",
+      "over flow",
+      "Ġcompar ability",
+      "Ġsyn optic",
+      "US B",
+      "db g",
+      "dem onstr",
+      "ĠAch ieving",
+      "Ġtect onics",
+      "ĠRand all",
+      "ĠPrep ared",
+      "Ġsublim ation",
+      "ĠB aj",
+      "Ġcl utch",
+      "Ġsub domain",
+      "Ġfl aws",
+      "inf lu",
+      "Ġwid ening",
+      "Ġmel ted",
+      "Ġadministr ator",
+      "Ġsubsidi ary",
+      "ĠP ricing",
+      "tic us",
+      "og i",
+      "ĠAl ign",
+      "ĠAD V",
+      "Ġvast ly",
+      "bench mark",
+      "Ġprioriti ze",
+      "R adi",
+      "ess ed",
+      "Ġsup rac",
+      "acc ard",
+      "Ġbiom imetic",
+      "ĠIr radiation",
+      "ĠALG OR",
+      "Ġpedig ree",
+      "ĠC MT",
+      "od ym",
+      "ĠV ig",
+      "ĠBi ochemistry",
+      "ĠAcc um",
+      "Ind ices",
+      "hard tii",
+      "ĠRal ph",
+      "Ġrumin ants",
+      "i T",
+      "on au",
+      "an er",
+      "pl anned",
+      "ever s",
+      "Ġret roviral",
+      "Ġquantif ier",
+      "ĠExt racting",
+      "Ġacet ylated",
+      "Or th",
+      "ĠSen ator",
+      "Ġnanos econd",
+      "Ġanticip ation",
+      "ĠECM O",
+      "Ġsemic irc",
+      "ĠCrypt osporidium",
+      "ĠT ARGET",
+      "Ġap ples",
+      "ef ield",
+      "Ġrem an",
+      "Ġser ovar",
+      "ĠTrans actions",
+      "trans itions",
+      "urs ions",
+      "ĠMel atonin",
+      "Ġcholecyst ectomy",
+      "ĠAntiv iral",
+      "h ous",
+      "Ġo tol",
+      "Ġm aj",
+      "Ġe clip",
+      "are l",
+      "AT IONAL",
+      "MI M",
+      "ĠCI mg",
+      "ĠEnd omet",
+      "ĠHay ashi",
+      "Ġchimpan zees",
+      "m bf",
+      "ĠI PV",
+      "act oring",
+      "out side",
+      "ne apolis",
+      "Ġdisc arding",
+      "num type",
+      "ĠRE ST",
+      "Ġflag ellar",
+      "ĠChand rase",
+      "hof er",
+      "Ġelectrocardi ogram",
+      "G b",
+      "m ock",
+      "o eb",
+      "ĠS MO",
+      "ĠM ord",
+      "ĠB oz",
+      "Ġmin ors",
+      "IN LINE",
+      "Ġtherm ogravimetric",
+      "ĠMel ting",
+      "ĠNS W",
+      "S ham",
+      "l otinib",
+      "Ġac quisitions",
+      "ta z",
+      "Ġdef aults",
+      "Ġoscill ates",
+      "ĠCap tion",
+      "Ġdisrup tive",
+      "Ġswe eping",
+      "ĠTool box",
+      "Ġureth ral",
+      "H BV",
+      "ĠR CS",
+      "Ġox ys",
+      "immun o",
+      "ht m",
+      "ofl avin",
+      "H IF",
+      "ĠS BA",
+      "ĠC PE",
+      "Ġwh ites",
+      "ĠRe actor",
+      "Ġpur p",
+      "Ġelectro catalytic",
+      "Ġnone x",
+      "Ġty phimurium",
+      "Ġeu rop",
+      "conc ave",
+      "macroph age",
+      "S ER",
+      "Ġl apse",
+      "Ġan atom",
+      "ĠR AC",
+      "ta x",
+      "Ġmin s",
+      "Ġsens u",
+      "ĠHe brew",
+      "Ġreal ism",
+      "ĠMicro glia",
+      "Ġserial ized",
+      "ĠHaz ard",
+      "Ġmetamorph osis",
+      "ĠI RA",
+      "Ġsm earing",
+      "Ġphot olysis",
+      "Ġchild birth",
+      "Ġsil enced",
+      "raw al",
+      "Ġquad rants",
+      "but anol",
+      "Ġstochastic ally",
+      "ĠCham bers",
+      "ĠNav arro",
+      "Ġproc urement",
+      "c ite",
+      "ĠS le",
+      "ĠH adoop",
+      "Ġdelay ing",
+      "At lantic",
+      "Sp ain",
+      "fal fa",
+      "od ialysis",
+      "yn ia",
+      "Ġplate aus",
+      "Ġmultim ode",
+      "RES ET",
+      "ĠRock y",
+      "ĠRodrig ues",
+      "f MRI",
+      "r int",
+      "ĠT AL",
+      "Ġspec ular",
+      "con struction",
+      "ĠAt hens",
+      "Ġcross link",
+      "Ġcount ably",
+      "Ġspread sheet",
+      "crib es",
+      "cons istently",
+      "Ġflood plain",
+      "E INVAL",
+      "M aca",
+      "Ġe i",
+      "Ġh itherto",
+      "Ġsem if",
+      "Ġcontin ual",
+      "ĠHom ology",
+      "Ġphotoc atalysts",
+      "is able",
+      "ĠH AT",
+      "Ġpoly hedra",
+      "ĠMay o",
+      "Ġlact ating",
+      "sam pler",
+      "Ġappl iances",
+      "T U",
+      "Ġc hess",
+      "ĠT ing",
+      "Ġinv itation",
+      "Ġdistrib uting",
+      "ash ima",
+      "Ġult ral",
+      "tre nd",
+      "Ġrelax ations",
+      "ĠHel en",
+      "Ġbed ding",
+      "Ġgland ular",
+      "Ġincrement ally",
+      "Ġconce al",
+      "claim ed",
+      "ĠEdd y",
+      "Ġm os",
+      "ĠT ube",
+      "ĠT oda",
+      "ra j",
+      "ĠM ü",
+      "ĠU ll",
+      "Ġun e",
+      "ber ine",
+      "Ġpolic ym",
+      "Ġscholar ly",
+      "Ġshore line",
+      "Ġald osterone",
+      "ĠBruc ella",
+      "T HE",
+      "RE AL",
+      "Ġex ome",
+      "ĠD AB",
+      "Ġext ras",
+      "Ġband ing",
+      "ĠSi emens",
+      "ĠBo ost",
+      "ĠSuper novae",
+      "ĠTrac ing",
+      "Ġascorb ate",
+      "Ital y",
+      "b und",
+      "Ġdecre ment",
+      "Ġneu rophysiological",
+      "Ġblack body",
+      "ĠMc N",
+      "Ġcompet encies",
+      "osc ape",
+      "ĠHon ours",
+      "Ġmas titis",
+      "criter ia",
+      "Ġb iaxial",
+      "Ġth awed",
+      "ĠF oll",
+      "ore ceptor",
+      "Ġinv ention",
+      "AD s",
+      "Sh ow",
+      "-------------------------------- ----------------",
+      "Ġellipso idal",
+      "Ġfoc ussed",
+      "ĠD at",
+      "ĠR im",
+      "ĠL X",
+      "ĠG ER",
+      "ins ler",
+      "Ġtop oisomerase",
+      "Ġhyper lipidemia",
+      "Ġmy stery",
+      "Ġnit rification",
+      "Ġonc ogenes",
+      "ĠFull er",
+      "ĠBart lett",
+      "Ġamphib ians",
+      "S ST",
+      "b ly",
+      "le ads",
+      "ec ycle",
+      "am pl",
+      "ĠP OM",
+      "ĠD CF",
+      "str ass",
+      "anti body",
+      "non linear",
+      "ĠBroad way",
+      "ĠCatal ogue",
+      "Ġμ A",
+      "Ġacet aminophen",
+      "Ġcrystall ites",
+      "ĠNan otubes",
+      "ĠAcknowledg ment",
+      "Ġmetam orphism",
+      "Ġtwin ning",
+      "ĠAzer bai",
+      "x A",
+      "CC C",
+      "ĠSol ids",
+      "pred s",
+      "ĠMont ana",
+      "WR ITE",
+      "R atio",
+      "Ġp unch",
+      "Ġr iding",
+      "Ġac ne",
+      "ĠU re",
+      "Ġcor r",
+      "ĠQ OL",
+      "Ġins ult",
+      "Ġdominant ly",
+      "Ġsubs amples",
+      "rew s",
+      "ĠPres ervation",
+      "ĠAff ine",
+      "methan one",
+      "Ġhedge hog",
+      "J H",
+      "Ġl ined",
+      "Ġst en",
+      "ĠD armstadt",
+      "ĠL asso",
+      "Ġde proton",
+      "Ġsh oes",
+      "Ġmo tives",
+      "Ġmic roscop",
+      "oph thora",
+      "Ġmac ron",
+      "Ġencour agement",
+      "acryl ic",
+      "ĠTensor Flow",
+      "W rapper",
+      "o ise",
+      "ay ak",
+      "Ġrep resses",
+      "Ġpr uned",
+      "ĠCl ar",
+      "ĠâĬ ²",
+      "ĠUnder lying",
+      "Im plemented",
+      "Ġswe at",
+      "Ġmeteor ites",
+      "Ġtwe ez",
+      "Ġpros ocial",
+      "Ġabras ion",
+      "h ail",
+      "Ġsh orth",
+      "ism atch",
+      "IN TR",
+      "ĠTr in",
+      "Ġphysic ists",
+      "ĠPE O",
+      "ĠMagn eto",
+      "ĠJacob son",
+      "ĠMMP s",
+      "ĠIntra venous",
+      "Ġneurotrans mission",
+      "ĠPneum onia",
+      "L ind",
+      "y re",
+      "Ġm aternity",
+      "ĠI ris",
+      "ri atal",
+      "ĠâĢ ļ",
+      "med etomidine",
+      "Ġtr iterpen",
+      "ĠMan uscript",
+      "ĠEnd oplasmic",
+      "ĠPot ter",
+      "Ġgerm inal",
+      "Ġphotos ystem",
+      "Gu ided",
+      "Ġguitar ist",
+      "ĠChile an",
+      "ĠSalv ador",
+      "É Ļ",
+      "Ġc elestial",
+      "om and",
+      "Ġn k",
+      "Ġv endors",
+      "ĠP INK",
+      "ĠIn organic",
+      "Ġmod erated",
+      "SU S",
+      "ĠJ oshi",
+      "ĠSt ata",
+      "ik es",
+      "oy e",
+      "ĠJohn ny",
+      "Le ica",
+      "Ġka on",
+      "ĠEquip ment",
+      "K im",
+      "g ado",
+      "t ures",
+      "Ġe lem",
+      "ĠA AC",
+      "ce a",
+      "od ality",
+      "Ġan iline",
+      "Ġex othermic",
+      "ĠG unn",
+      "ĠJ U",
+      "plic able",
+      "sc apes",
+      "typ ed",
+      "Ġinsp iratory",
+      "REG IST",
+      "ĠBry an",
+      "Ġanxi ous",
+      "ĠCarp enter",
+      "ĠPharmacokine tics",
+      "infer ior",
+      "F rag",
+      "Z Y",
+      "Ġo esophageal",
+      "ĠS uk",
+      "ĠP ron",
+      "ĠCD I",
+      "AG C",
+      "key words",
+      "sus ceptible",
+      "Germ any",
+      "M AS",
+      "i C",
+      "an mar",
+      "Ġex iting",
+      "ĠH ale",
+      "Ġr hamn",
+      "ind ustrial",
+      "Ġra ft",
+      "emb rolizumab",
+      "Ġdeploy ing",
+      "Ġsalic ylic",
+      "R n",
+      "Ġc ensor",
+      "Ġd X",
+      "Ġfor um",
+      "MS I",
+      "bl ad",
+      "Ġsqu ir",
+      "CP P",
+      "Ġgrap evine",
+      "ĠRA FT",
+      "Mon te",
+      "Ġmicrof lora",
+      "r cl",
+      "Ġdec ap",
+      "AN C",
+      "Ġbroad en",
+      "Ġfre ed",
+      "Ġsouth ward",
+      "ĠJac ques",
+      "Ġrequest ing",
+      "ĠAsp ect",
+      "araj an",
+      "F ailed",
+      "f printf",
+      "p ytest",
+      "Ê ¹",
+      "ĠC m",
+      "un til",
+      "ne iss",
+      "Ġmon os",
+      "osp inal",
+      "ols ky",
+      "cont rib",
+      "Con tainer",
+      "ĠVol unte",
+      "ĠAtt ributes",
+      "ĠTum our",
+      "Ġrein hardtii",
+      "Ġcentrom ere",
+      "ĠS ymph",
+      "ĠA o",
+      "ag ens",
+      "ple ted",
+      "ied er",
+      "Ġactiv ist",
+      "ĠAl meida",
+      "Ġdisturb ing",
+      "Ġreflex es",
+      "D SS",
+      "Ġfor wards",
+      "ron ym",
+      "ĠIntegr ity",
+      "WE EN",
+      "Ġôı¼ Į",
+      "Ġfaith fully",
+      "Ġperic ardial",
+      "Japan ese",
+      "ĠCEN P",
+      "K r",
+      "Ġdef ending",
+      "Ġz on",
+      "ins ensitive",
+      "Ġlab s",
+      "ĠCa M",
+      "ĠEu rop",
+      "ME A",
+      "B LAST",
+      "x N",
+      "al en",
+      "Ġcl k",
+      "line age",
+      "co ating",
+      "Ġtail oring",
+      "CON TR",
+      "Ġadren ergic",
+      "interp reted",
+      "N IH",
+      "am oeba",
+      "ĠC yr",
+      "Ġtri plicates",
+      "def ining",
+      "ĠLog an",
+      "tes y",
+      "ĠTw ist",
+      "ĠGram mar",
+      "ĠSustain ed",
+      "Ġan harmonic",
+      "Ġal ve",
+      "Ġr uler",
+      "Ġqu anta",
+      "Ġdirec ts",
+      "Ġoff loading",
+      "ĠGe ophysical",
+      "Re quire",
+      "Ġhepat oma",
+      "Ġfo o",
+      "ĠGeor ges",
+      "Ġb outs",
+      "ĠT AK",
+      "Ġanti diabetic",
+      "ĠRe ported",
+      "present ing",
+      "ĠLay ered",
+      "REN CE",
+      "Ġuve itis",
+      "b age",
+      "Ġf entanyl",
+      "ens emble",
+      "ĠO SCC",
+      "Ġmin ers",
+      "lo oking",
+      "ĠBe er",
+      "prec ipitation",
+      "ĠEnter prise",
+      "Ġseroton ergic",
+      "Ġsees aw",
+      "ĠAth letics",
+      "Ġhydroly tic",
+      "Ġtal ent",
+      "Ġdiscern ible",
+      "F IL",
+      "l ives",
+      "ĠS ales",
+      "ĠS Sc",
+      "ere nd",
+      "cl im",
+      "anti d",
+      "IN TS",
+      "Ġatten uating",
+      "Ġtw ists",
+      "Ġoxygen ase",
+      "rin i",
+      "Maca ulay",
+      "z m",
+      "ĠP OT",
+      "ĠM p",
+      "ĠH ey",
+      "ĠO VER",
+      "ill ion",
+      "Ġinv aluable",
+      "Ġanti platelet",
+      "Ġmut ans",
+      "Ġgrad uates",
+      "GR AM",
+      "isp heric",
+      "Ġincomp atibility",
+      "Ġcarboxyl ase",
+      "Ġbioc ontrol",
+      "ĠPhysic ochemical",
+      "ï Ļ",
+      "Ġl ae",
+      "ĠA ortic",
+      "ĠR acing",
+      "ĠE CD",
+      "iv ic",
+      "Ġelect romechanical",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠĠĠĠĠĠ",
+      "Ġste er",
+      "ĠOut side",
+      "Ġaden ocarcinomas",
+      "ĠKe ep",
+      "Ġcoc on",
+      "Ġmoder ating",
+      "Ġreform ulated",
+      "Ġfundament als",
+      "ĠTes la",
+      "ĠStir ling",
+      "or ated",
+      "op id",
+      "Ġpa rox",
+      "Ġtri valent",
+      "Ġexchange able",
+      "Ġdeb uted",
+      "V ery",
+      "re ements",
+      "ĠT omm",
+      "ĠC yn",
+      "ĠC atalysts",
+      "qu at",
+      "ĠF ER",
+      "ĠR um",
+      "Ġsc anners",
+      "ĠâĨĴ âĪŀ",
+      "otrop ical",
+      "Ġven ues",
+      "estim ator",
+      "Ġempt ying",
+      "G PP",
+      "V IR",
+      "Ġcom plicates",
+      "ĠN IS",
+      "ĠZ hen",
+      "ĠBl ues",
+      "Ġtext books",
+      "Ġsi xty",
+      "Ġether s",
+      "Ġfinanc ially",
+      "Ġm Health",
+      "ĠT ut",
+      "Ġl aryng",
+      "ĠG s",
+      "ĠW atch",
+      "Ġse v",
+      "Ġit al",
+      "ass ed",
+      "ĠÃ ·",
+      "ĠCons ent",
+      "Ġnut s",
+      "vit real",
+      "Ġmeta phase",
+      "Ġtit ania",
+      "Ġfo ils",
+      "Ġgal ectin",
+      "initial ize",
+      "Ġadvis or",
+      "Ġadminister ing",
+      "B ool",
+      "Ġc em",
+      "Ġre forming",
+      "Ġg n",
+      "ys h",
+      "Ġatt or",
+      "SC I",
+      "Ex c",
+      "bu ilder",
+      "Ġcer ium",
+      "Ġregist ries",
+      "ĠMatsum oto",
+      "Ġtempt ing",
+      "is ha",
+      "Ġre orientation",
+      "ĠM old",
+      "ĠR AGE",
+      "ys on",
+      "Ġun equiv",
+      "Ġrel ocation",
+      "ĠÃ ķ",
+      "ĠRe form",
+      "ĠRE QU",
+      "Ġcommens urate",
+      "catal og",
+      "ĠT PS",
+      "Ġl amb",
+      "Ġpre factor",
+      "arch y",
+      "Ġdop ants",
+      "dr v",
+      "ĠPAR AMET",
+      "sched ule",
+      "ochem ically",
+      "Ġe Health",
+      "un as",
+      "ĠP inus",
+      "ĠH SA",
+      "Ġinter relations",
+      "Ġdep ot",
+      "ĠPl atinum",
+      "Ġlif elong",
+      "Ġpersist ently",
+      "ĠParad ox",
+      "ĠConform ational",
+      "es ophag",
+      "ĠA AT",
+      "pl in",
+      "ĠF CN",
+      "ĠD t",
+      "op oside",
+      "Ġch al",
+      "Ġhal t",
+      "ĠDet ect",
+      "Ġdiscrim inated",
+      "ĠLag rangians",
+      "Ap pro",
+      "ĠÈ §",
+      "Ġimpuls ivity",
+      "B AT",
+      "C hemical",
+      "g ather",
+      "ĠU NC",
+      "int ron",
+      "ĠSim ulator",
+      "ĠGl a",
+      "TT T",
+      "ĠVol atile",
+      "Ġsubs id",
+      "ĠBroad casting",
+      "Ġstrept ozotocin",
+      "Ġf umar",
+      "ĠM PEG",
+      "Ġinflu enzae",
+      "sub jects",
+      "Ġappropri ateness",
+      "Ġarc min",
+      "Ġstrand ed",
+      "o ylation",
+      "ĠD EX",
+      "ov iral",
+      "ĠQu arter",
+      "col ytic",
+      "Ġfriend ship",
+      "H ES",
+      "l oxacin",
+      "Ġe re",
+      "ĠT rad",
+      "ur istics",
+      "ĠE CT",
+      "ĠE GCG",
+      "ĠL RP",
+      "ĠG AG",
+      "ĠIn P",
+      "Ġcont empor",
+      "Ġmic ror",
+      "ier strass",
+      "ĠElect rosp",
+      "need ed",
+      "atmosp here",
+      "n T",
+      "Ġband widths",
+      "Ġdivers ified",
+      "ĠAppro priate",
+      "rest ore",
+      "roc nem",
+      "ĠLag uerre",
+      "ĠSong s",
+      "ĠKalu za",
+      "ĠS ymmetries",
+      "ĠSch mitt",
+      "Ġbiom olecular",
+      "scale box",
+      "Ġintra hepatic",
+      "under standing",
+      "ĠABC G",
+      "Ġunderestim ates",
+      "ĠStream ing",
+      "Ġfic titious",
+      "oplasm osis",
+      "res ident",
+      "ĠB ary",
+      "ĠCom a",
+      "sc rip",
+      "Ġdeg ran",
+      "ĠCa MKII",
+      "ĠBal mer",
+      "ĠPlas m",
+      "Ġchel ating",
+      "ĠParad igm",
+      "Ġopp onents",
+      "E K",
+      "P in",
+      "Ġm sec",
+      "ad one",
+      "ach t",
+      "CC G",
+      "EC O",
+      "normal ize",
+      "ĠDesign s",
+      "Ġyellow ish",
+      "glut amyl",
+      "Ġdomestic ation",
+      "Ġmonoph yletic",
+      "d les",
+      "n ested",
+      "ĠG race",
+      "ĠStud ios",
+      "ĠDisc ussions",
+      "ophen oxy",
+      "Ġveter in",
+      "Ġendos ym",
+      "utting er",
+      "b atches",
+      "ĠF iji",
+      "ĠR NF",
+      "ous a",
+      "ĠK Y",
+      "Ġspect rograph",
+      "ER IC",
+      "ĠMy anmar",
+      "ĠConst raining",
+      "Ġecological ly",
+      "Ġfro st",
+      "arb oux",
+      "ĠFib onacci",
+      "Ġcancel ed",
+      "ĠISS N",
+      "R ect",
+      "an other",
+      "ĠM MA",
+      "OL O",
+      "ĠTr uth",
+      "Ġorth opaedic",
+      "Ġtravers ing",
+      "ischem ic",
+      "ĠMozamb ique",
+      "ĠM SR",
+      "ap ace",
+      "ĠTh read",
+      "olog ia",
+      "Ġcal m",
+      "methyl transferase",
+      "ĠÍ ª",
+      "Ġdro ve",
+      "Ġcommand ed",
+      "Ġf eline",
+      "ĠR ush",
+      "ĠL isa",
+      "Ġsuper space",
+      "arc y",
+      "ĠReg ulated",
+      "ĠRest ing",
+      "caus ing",
+      "psycho tics",
+      "q t",
+      "Ġt ulare",
+      "Ġrel ocated",
+      "Ġrep ell",
+      "Ġpred atory",
+      "pe ople",
+      "tra its",
+      "E uclidean",
+      "F DA",
+      "X RT",
+      "p C",
+      "p and",
+      "Ġ Æ",
+      "re ve",
+      "Ġb ids",
+      "Ġco usin",
+      "Ġsub domains",
+      "til b",
+      "é nez",
+      "linear ly",
+      "oprotein s",
+      "Ġcod ec",
+      "Ġcontrac eption",
+      "ĠCd k",
+      "Ġrail road",
+      "B ench",
+      "r ng",
+      "ĠD LA",
+      "enti le",
+      "ĠCO CO",
+      "ĠMat th",
+      "ĠOver l",
+      "ĠRout ine",
+      "Ġmultif ocal",
+      "Ġarte fact",
+      "Ġsculpt ure",
+      "c ies",
+      "m ate",
+      "Ġ Ø",
+      "ure k",
+      "ĠB end",
+      "ĠN athan",
+      "Ġde w",
+      "ym ia",
+      "az zi",
+      "ĠEr k",
+      "Ġgrad uation",
+      "Bound ary",
+      "G ra",
+      "Ġb fd",
+      "ĠS ert",
+      "Ġover shoot",
+      "ĠSe o",
+      "Ġsk learn",
+      "Ġconserv atively",
+      "pir acy",
+      "Ġla unching",
+      "X D",
+      "ar bitrary",
+      "per one",
+      "Ġsh ops",
+      "comp etitive",
+      "ĠPak istani",
+      "Ġcompetit or",
+      "b iotics",
+      "ra its",
+      "ĠN oble",
+      "Ġsub regions",
+      "ĠJ ump",
+      "roll er",
+      "tr is",
+      "Ġmac rol",
+      "ó s",
+      "ĠPen ic",
+      "Ġmicros omes",
+      "Ġimprec ise",
+      "Ġdownt own",
+      "Ġe QTL",
+      "if est",
+      "ĠM FI",
+      "Ġr arity",
+      "âĢĻ âĢĻ",
+      "Ġbel ts",
+      "Ġglycos yl",
+      "ĠNic olas",
+      "synt hesis",
+      "O h",
+      "h ierarch",
+      "p ps",
+      "an ets",
+      "ro ads",
+      "AT IC",
+      "MI MO",
+      "ĠCont ract",
+      "Le ib",
+      "opy rox",
+      "Ġinform ational",
+      "Syn onyms",
+      "chall enge",
+      "ĠProxim al",
+      "ĠCraw ford",
+      "Ġis opropyl",
+      "Ġsub families",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠĠĠĠĠĠĠ",
+      "Ġannot ators",
+      "Ġreconc ile",
+      "Ġparsim ony",
+      "Ġcasp ases",
+      "c ott",
+      "en vironments",
+      "Ġd rm",
+      "ĠP IL",
+      "ĠM ec",
+      "ĠIn fer",
+      "ĠSir t",
+      "S hell",
+      "ag ulants",
+      "se ismic",
+      "Ġsub urban",
+      "ĠX XX",
+      "iod es",
+      "Ġback propagation",
+      "tra ditional",
+      "Ġphotoc on",
+      "ĠMicrobi ology",
+      "Q T",
+      "ur idine",
+      "Ġch op",
+      "ĠTh é",
+      "Ġpre jud",
+      "Ġenc oders",
+      "col lected",
+      "rem ark",
+      "Ġsun spot",
+      "ĠPhen olic",
+      "Under standing",
+      "Ġreject ing",
+      "Ġrom antic",
+      "Ġcentim eters",
+      "Ġhalluc inations",
+      "H ome",
+      "c asted",
+      "Ġc w",
+      "ra i",
+      "ĠDis placement",
+      "PH Y",
+      "carb am",
+      "Ġxen on",
+      "Ġnarr atives",
+      "Ġdoll ar",
+      "Ġdyn asty",
+      "ì §",
+      "Ġin forming",
+      "ĠO CD",
+      "á k",
+      "Ġoverhead s",
+      "ju ana",
+      "ĠKra us",
+      "f x",
+      "k aya",
+      "Ġn id",
+      "ĠG rab",
+      "Ġinf lores",
+      "Ar c",
+      "======== ====",
+      "Ġcondens er",
+      "Ġnanoc ar",
+      "omm ens",
+      "Ġsatur ating",
+      "re ce",
+      "el if",
+      "ĠA LE",
+      "ĠB ub",
+      "ĠL af",
+      "and ran",
+      "Ġpo uch",
+      "rol ine",
+      "AC HE",
+      "CC D",
+      "Ġcool ant",
+      "Ġgrass lands",
+      "ĠSynchron ous",
+      "izz iness",
+      "Ġcet uximab",
+      "Ġdichotom ous",
+      "ro ch",
+      "ĠA uckland",
+      "ob esity",
+      "ik it",
+      "Ġoper ad",
+      "ĠOn set",
+      "Ġbefore hand",
+      "Ġunc omp",
+      "US ED",
+      "ubb ing",
+      "ĠSMB H",
+      "ĠExped ition",
+      "Ġh ib",
+      "ĠP PR",
+      "ĠN ED",
+      "ud io",
+      "ĠJ al",
+      "ĠAr p",
+      "ĠBe e",
+      "ĠVari eties",
+      "Com m",
+      "Ab out",
+      "ĠAtt achment",
+      "ODU LE",
+      "Calc ulate",
+      "T an",
+      "in ism",
+      "Ġa ra",
+      "Ġc abin",
+      "Ġcon nexin",
+      "Ġcom ets",
+      "ump tive",
+      "Ġdest abilization",
+      "ĠHol t",
+      "ruct ose",
+      "anish i",
+      "plastic ity",
+      "omyc osis",
+      "ovic ian",
+      "________ ________",
+      "r ar",
+      "Ġw ore",
+      "ud ine",
+      "ĠIn variance",
+      "Ġper itonitis",
+      "Ġmet rology",
+      "Ġclos es",
+      "Ġcolor less",
+      "No ise",
+      "DI O",
+      "ĠLif shitz",
+      "z ul",
+      "es tive",
+      "ĠM PA",
+      "ĠB ooth",
+      "ĠD oll",
+      "are ne",
+      "gen ess",
+      "Ġmolecular ly",
+      "ĠPer kin",
+      "Ġdos imetry",
+      "ĠSO FT",
+      "ĠPy Torch",
+      "Ġquar ters",
+      "ĠKu hn",
+      "Ġsplen ocytes",
+      "R W",
+      "c art",
+      "le b",
+      "Ġcon dom",
+      "ĠH oc",
+      "Ġext ents",
+      "Ġsl ug",
+      "ĠSup plementation",
+      "diff ic",
+      "ester ly",
+      "Y u",
+      "an tigens",
+      "ĠÃ Ĵ",
+      "Ch anges",
+      "Ġprop ylene",
+      "ĠPr ison",
+      "ĠAlgorithm ic",
+      "Ġtoler ances",
+      "Ad am",
+      "Ġester ase",
+      "Ġmil der",
+      "ĠConv ection",
+      "P TR",
+      "k pc",
+      "Ġex o",
+      "ĠF ah",
+      "ĠY FP",
+      "ĠCR M",
+      "Ġhepat otoxicity",
+      "Ġnic otinamide",
+      "Ġpatch y",
+      "depend s",
+      "Ġp B",
+      "Ġe el",
+      "Ġn v",
+      "ĠS es",
+      "ĠH Z",
+      "Ġim print",
+      "ep ileptic",
+      "fl uctuations",
+      "Ġformal ize",
+      "che v",
+      "Ġdip ping",
+      "ĠPy ramid",
+      "Ġhol o",
+      "ĠMT s",
+      "Ġlamin ates",
+      "Ġworm hole",
+      "L AP",
+      "h ape",
+      "Ġa k",
+      "Ġre als",
+      "Ġby stand",
+      "Ġinter leaved",
+      "Ġx z",
+      "ov y",
+      "Ġcop rime",
+      "ucl ides",
+      "Ġtrim ming",
+      "MIC AL",
+      "pyr role",
+      "I a",
+      "N LS",
+      "Q uality",
+      "t akes",
+      "z inc",
+      "ĠP ione",
+      "ĠE wing",
+      "ĠL CA",
+      "ĠÃ Ķ",
+      "ict us",
+      "Ġcoll im",
+      "Ġphyl ogenetically",
+      "ĠKe eping",
+      "ĠFa ith",
+      "bond s",
+      "ti ter",
+      "Ġsub categories",
+      "sh aded",
+      "Ġphot ospheric",
+      "ĠApp earance",
+      "ĠUnivers ities",
+      "Ġglomer uli",
+      "ĠPref rontal",
+      "Ġprivi lege",
+      "i H",
+      "u ya",
+      "ĠL CL",
+      "ĠIn GaAs",
+      "In spired",
+      "atal og",
+      "ĠPer ceptions",
+      "ĠNa HCO",
+      "Ġstream line",
+      "tra jectory",
+      "ĠMic rom",
+      "Ġbed side",
+      "ĠRom ero",
+      "Ġgaug ino",
+      "D EN",
+      "F a",
+      "O lymp",
+      "e al",
+      "u els",
+      "ic ylic",
+      "Ġg od",
+      "Ġat taining",
+      "Ġprot ests",
+      "Ġnow here",
+      "des orption",
+      "ĠHydro xy",
+      "ĠEr bB",
+      "ĠSP AR",
+      "Ġhind ers",
+      "heren kov",
+      "KERN EL",
+      "Ġs ect",
+      "ul ong",
+      "Ġpre processed",
+      "frac tional",
+      "oy age",
+      "Ġphosph atases",
+      "Ġcoast line",
+      "Ġh ref",
+      "ĠS utherland",
+      "ox one",
+      "Ġhom omorphic",
+      "D EM",
+      "Ġb ovis",
+      "ĠC BP",
+      "pl en",
+      "ĠB uc",
+      "ĠG ior",
+      "Ġcomp ost",
+      "ĠO racle",
+      "ĠSp here",
+      "ĠSch re",
+      "deriv atives",
+      "ly tes",
+      "ĠY o",
+      "Ġcycl ones",
+      "ĠMa ize",
+      "Ġunf air",
+      "Tem plate",
+      "Ġimpregn ation",
+      "Ġlapa roscopy",
+      "Ġh amiltonian",
+      "ign ore",
+      "Ġdis posable",
+      "ear ic",
+      "Ġelect oral",
+      "cc os",
+      "ĠSh h",
+      "Ġturb o",
+      "Ġintr usive",
+      "Ġpreced ence",
+      "annot ated",
+      "Ġdyst onia",
+      "F at",
+      "u ins",
+      "Ġs way",
+      "ar izing",
+      "ill en",
+      "Ġy i",
+      "Ġnorm ed",
+      "ĠÌ Ĥ",
+      "ĠExt r",
+      "ĠProte ome",
+      "Doc ument",
+      "ĠQUANT UM",
+      "ti ti",
+      "ĠC PC",
+      "ĠM iles",
+      "ĠB oc",
+      "ĠR TS",
+      "CT X",
+      "Ġsaf egu",
+      "ĠNorm ally",
+      "ĠÃľ ber",
+      "on ious",
+      "ĠS CE",
+      "Ġal falfa",
+      "ĠL ut",
+      "Ġco ut",
+      "Ġen large",
+      "ĠEn able",
+      "Ġvir ion",
+      "ĠSh allow",
+      "def initely",
+      "ĠCol in",
+      "ĠRet ention",
+      "Ġmimic ry",
+      "################################ ################################",
+      "NSC LC",
+      "Ġgrat itude",
+      "Ġt ending",
+      "ĠI DS",
+      "ere t",
+      "ric an",
+      "Ġx n",
+      "ĠY oo",
+      "Ġoptim ise",
+      "Ar row",
+      "ĠTransfer ase",
+      "PK C",
+      "ĠGuang zhou",
+      "r uc",
+      "y rid",
+      "is z",
+      "ĠF IX",
+      "ĠD atabases",
+      "ast ron",
+      "Ġplay back",
+      "Ġnarrow ly",
+      "Cor relation",
+      "ĠAff inity",
+      "Ġfunctor ial",
+      "Ġlect ins",
+      "Ġrup tured",
+      "Dis play",
+      "ĠSympt om",
+      "Ġequid istant",
+      "ĠRicc ati",
+      "ĠAchie vement",
+      "g rand",
+      "on ated",
+      "Ġd H",
+      "ĠF ID",
+      "ĠD ER",
+      "ĠCo A",
+      "Ġgas ification",
+      "ĠCON S",
+      "Ġaccompan ies",
+      "Ġimped e",
+      "Ġpreced e",
+      "Ġkit chen",
+      "prog ress",
+      "Ġw iring",
+      "le renes",
+      "ĠG ius",
+      "Ġtrans p",
+      "ret rie",
+      "ij er",
+      "aff er",
+      "Ġbirth day",
+      "ĠHal d",
+      "Ġmusc ulus",
+      "ĠTok en",
+      "ĠBow el",
+      "Ġskip ped",
+      "C ha",
+      "b v",
+      "ĠB low",
+      "Ġpre operatively",
+      "Ġgl ove",
+      "ĠLe ven",
+      "Ġmes op",
+      "ĠAux iliary",
+      "ensure math",
+      "j us",
+      "Å ©",
+      "Ġv oter",
+      "ĠH itch",
+      "pro xy",
+      "ĠK ut",
+      "Ġpo ems",
+      "ĠAn gl",
+      "cer a",
+      "Ġstar red",
+      "AG ES",
+      "Sc ience",
+      "Anal yses",
+      "Ġrefere es",
+      "Ġabrog ated",
+      "Ġdesal ination",
+      "ĠPrand tl",
+      "P it",
+      "Ġn atal",
+      "og ran",
+      "ys titis",
+      "Ġdes m",
+      "Ġcur ious",
+      "Ġdem on",
+      "uz zi",
+      "ochond rial",
+      "ĠTreat y",
+      "Track er",
+      "rhoe ae",
+      "L W",
+      "f urt",
+      "Ġo mp",
+      "is ational",
+      "Ġmem orial",
+      "ĠLat ency",
+      "ĠHyp ot",
+      "Ġglu ed",
+      "exact ly",
+      "Ġcontra ind",
+      "C ancer",
+      "Ġf fi",
+      "ĠN AA",
+      "ĠCh r",
+      "eg g",
+      "ĠMo tiv",
+      "Ġlay outs",
+      "Ġclim b",
+      "Ġappend icitis",
+      "CU DA",
+      "Ġphotop roduction",
+      "ĠS IP",
+      "Ġv eto",
+      "per in",
+      "ĠUn ity",
+      "by ear",
+      "Ġforward ed",
+      "ĠDom inant",
+      "hol z",
+      "ĠThor acic",
+      "DEF INE",
+      "Ġtyros inase",
+      "B ad",
+      "I NA",
+      "f uel",
+      "Ġg i",
+      "ĠV IS",
+      "ast olic",
+      "Ġox aliplatin",
+      "eff ector",
+      "ĉĉĉĉ Ġ",
+      "е ÑĢ",
+      "ĠBab y",
+      "Ġwash out",
+      "pit uitary",
+      "N GC",
+      "Ġd ns",
+      "ĠP oz",
+      "ĠU z",
+      "pos itron",
+      "ĠElect rons",
+      "Ġhem angi",
+      "ĠZn S",
+      "ĠTE MP",
+      "ĠExperiment ally",
+      "fluor ouracil",
+      "Ġlap arotomy",
+      "analy zer",
+      "ocortic oid",
+      "ĠIMP L",
+      "ĠDNN s",
+      "ĠFres nel",
+      "M ont",
+      "Ġt apes",
+      "ul omb",
+      "im pedance",
+      "ĠH ET",
+      "ath a",
+      "mod ulation",
+      "ĠCor tic",
+      "Ġâľ ĵ",
+      "ĠFair ness",
+      "ĠSti ff",
+      "Ġbutt ons",
+      "c ss",
+      "Ġand roid",
+      "el ast",
+      "ĠT eflon",
+      "ĠM BC",
+      "ĠJ T",
+      "Ġmulti layered",
+      "ĠRe e",
+      "uit ar",
+      "ĠPhil ips",
+      "ĠSk ip",
+      "doc toral",
+      "iy ama",
+      "ĠLead ership",
+      "ĠCris is",
+      "Ġdesens itization",
+      "v ous",
+      "ĠS PP",
+      "ĠP GA",
+      "ĠN ever",
+      "Ġdef eating",
+      "Ġfib romyalgia",
+      "ĠMR P",
+      "ĠAB CA",
+      "ĠLow e",
+      "Ġer oded",
+      "Ġaug ments",
+      "ĠBor is",
+      "Ġneph rectomy",
+      "ĠSher man",
+      "Ġrefrig eration",
+      "ĠHern ández",
+      "Ã ĺ",
+      "ĠT ors",
+      "ch us",
+      "ĠV arg",
+      "Ġro set",
+      "CL R",
+      "DE P",
+      "Str ong",
+      "Ġcin erea",
+      "ĠHein rich",
+      "R out",
+      "od us",
+      "ĠPh one",
+      "ĠPer l",
+      "Ġseason ally",
+      "hold ing",
+      "Ġencephal omyelitis",
+      "Ġfasc ia",
+      "Ġlitterm ates",
+      "ĠWIT HOUT",
+      "Ð ±",
+      "Ġal erts",
+      "ĠK oll",
+      "ĠU rs",
+      "elf and",
+      "ĠRNA P",
+      "Ġinvari ably",
+      "Ġscin tigraphy",
+      "ĠSebas tian",
+      "kines ia",
+      "C UR",
+      "in ants",
+      "Ġp ET",
+      "id ial",
+      "ĠU PLC",
+      "Ġsu is",
+      "Ġbas olateral",
+      "ĠMod ulates",
+      "orb ic",
+      "Im g",
+      "Ġparas itism",
+      "Ġlamin ate",
+      "oge ographic",
+      "ĠRib eiro",
+      "ĠGlut athione",
+      "ĠAber rant",
+      "Ġs clero",
+      "ĠD LS",
+      "ĠR uth",
+      "Ġrec ast",
+      "rec ated",
+      "ok ie",
+      "ĠPark s",
+      "Ġfoli ations",
+      "ĠDaw son",
+      "Ġtann ins",
+      "ĠAar on",
+      "p S",
+      "it ating",
+      "ĠI TC",
+      "ip ients",
+      "oh y",
+      "CC s",
+      "Ġeth anolic",
+      "cor hynchus",
+      "Ġorient ational",
+      "Ġhabit uation",
+      "Ġconvers ational",
+      "ĠVent ricular",
+      "Ġintercal ated",
+      "Ġphosphodies terase",
+      "ĠSeif ert",
+      "w k",
+      "al gesia",
+      "Ġst egan",
+      "ĠL us",
+      "oph antine",
+      "Ġcorrec ts",
+      "ĠOb ama",
+      "lat ency",
+      "Ġson ar",
+      "ORM AL",
+      "Ġseaw eed",
+      "ĠPow ers",
+      "ĠShap ley",
+      "L ore",
+      "Ġa wa",
+      "al ach",
+      "ĠF on",
+      "ens ate",
+      "Ġoptim a",
+      "IN F",
+      "Ġpoly genic",
+      "Ġmes oderm",
+      "Con ver",
+      "BR ID",
+      "ĠHel p",
+      "ĠRas mussen",
+      "Ġprokary otes",
+      "ĠEur asian",
+      "ĠPerme ability",
+      "Ġn au",
+      "ĠC lem",
+      "od ilation",
+      "ĠD iaz",
+      "iti ous",
+      "ĠCh ad",
+      "OR A",
+      "ĠSim ons",
+      "ĠDist ances",
+      "Ġast rometric",
+      "ĠCP Us",
+      "Ġthi oredoxin",
+      "perturb ation",
+      "Ġdendrim er",
+      "al gal",
+      "Ġc eliac",
+      "as z",
+      "ĠP PE",
+      "qu a",
+      "ĠB oll",
+      "ch r",
+      "Ġpre view",
+      "ĠPro jections",
+      "ĠAs ians",
+      "ĠInf erring",
+      "ĠNa ive",
+      "ĠHig gins",
+      "ĠLoc ated",
+      "cardi ac",
+      "ĠLars on",
+      "haz ard",
+      "ĠScienti sts",
+      "Ġp inn",
+      "EN CY",
+      "form e",
+      "chit ects",
+      "oflu orescent",
+      "ĠPor tal",
+      "Ġpup ae",
+      "interest ing",
+      "į Ģ",
+      "re act",
+      "at os",
+      "en in",
+      "ti o",
+      "ĠC app",
+      "ĠM au",
+      "ĠL SC",
+      "ĠV lasov",
+      "Ġsub sum",
+      "Ġdes erve",
+      "AS D",
+      "Rec e",
+      "Ġconson ant",
+      "Ġimpregn ated",
+      "Ġlignocell ulosic",
+      "Ġs ows",
+      "le ment",
+      "ĠT ier",
+      "ĠM EF",
+      "ĠH ugh",
+      "inc k",
+      "py razole",
+      "UL ATIONS",
+      "ĠAL I",
+      "ĠDr ift",
+      "Ġsolub ilized",
+      "Ġdraft ing",
+      "icycl ic",
+      "Ġredes ign",
+      "Ġdelib erate",
+      "Ġt apping",
+      "ĠT omas",
+      "ĠT unneling",
+      "ĠC BR",
+      "Ġan odes",
+      "ĠL SR",
+      "ĠN ath",
+      "ros ive",
+      "ĠHe idelberg",
+      "Ġcr ushing",
+      "ĠSh ore",
+      "Ġmal ondialdehyde",
+      "ĠMR D",
+      "ogl oss",
+      "nc ia",
+      "Ġgranul oma",
+      "Ġplain text",
+      "Ġarteri ovenous",
+      "Ġrifamp icin",
+      "Lepid optera",
+      "O ct",
+      "Ġl one",
+      "ĠAp pe",
+      "ĠInter mitt",
+      "comp ile",
+      "pot entials",
+      "ĠStandard ized",
+      "Ġventil atory",
+      "Ġhypercholesterolem ia",
+      "ĠEVALU ATION",
+      "k ed",
+      "x C",
+      "en os",
+      "Ġb authorbsnm",
+      "ĠR ost",
+      "math open",
+      "Ġcont ested",
+      "Ġro s",
+      "oth o",
+      "Ġem its",
+      "ero zo",
+      "Ġprop ranolol",
+      "Ġexacerb ate",
+      "Integr ating",
+      "ĠWars aw",
+      "Ñ ĩ",
+      "re fractory",
+      "ĠM ort",
+      "phosph onate",
+      "GL T",
+      "ĠChlor ide",
+      "ĠLU AD",
+      "ĠSQU ID",
+      "ĠOBSERV ATIONS",
+      "Ħ ĺ",
+      "ag les",
+      "ug er",
+      "Ġdiff using",
+      "yl ar",
+      "Ġanti p",
+      "ren ormal",
+      "Ġshe ared",
+      "ĠAnd r",
+      "ympt otics",
+      "ĠIdentif ied",
+      "Ġflex or",
+      "Li ouville",
+      "ĠCyt otoxic",
+      "L ock",
+      "d onald",
+      "ĠS HA",
+      "pro jected",
+      "plic ial",
+      "Ġbas ics",
+      "ĠCar valho",
+      "Ġheter ocyclic",
+      "Ġfluor ophore",
+      "ĠIntr igu",
+      "ĠAnne aling",
+      "G ln",
+      "H ispanic",
+      "Ġs aus",
+      "ĠT CS",
+      "ĠH AP",
+      "Ġy tt",
+      "Ġcons ulting",
+      "rec ts",
+      "Ġinf all",
+      "LE V",
+      "tri azole",
+      "Ġnarrow ed",
+      "Ġamph oteric",
+      "ĠSor ting",
+      "ĠMom ents",
+      "Ġarab in",
+      "Ġcocon ut",
+      "ĠIntrigu ingly",
+      "Ġp ushes",
+      "Ġm ec",
+      "ĠN air",
+      "Ġcol istin",
+      "ĠOb tained",
+      "df s",
+      "Ġcompet ency",
+      "W ORD",
+      "ĠA AS",
+      "ĠB NP",
+      "ĠH AS",
+      "ĠL un",
+      "ĠL nc",
+      "Ġhydro cephalus",
+      "Ġhom ological",
+      "Ġcarbon ic",
+      "ĠHi Seq",
+      "commun ity",
+      "Ġcephal ospor",
+      "Ġhos tile",
+      "prov ide",
+      "Ġskyrm ion",
+      "D AG",
+      "Ġc nt",
+      "Ġh ay",
+      "Ġorder ings",
+      "Ġfl ock",
+      "HE A",
+      "ĠNeu rom",
+      "Ġboost s",
+      "ĠCard inal",
+      "ĠBac helor",
+      "Ġdec ent",
+      "ĠY ak",
+      "Ġcalc d",
+      "ĠBo er",
+      "Ġtranscript omics",
+      "Ġrearrang ed",
+      "ĠPolym orphisms",
+      "ĠPras ad",
+      "oinositi de",
+      "b ars",
+      "Ġ ãģ",
+      "ĠS AA",
+      "Ġon ion",
+      "ag el",
+      "ĠH p",
+      "og rel",
+      "di visions",
+      "and an",
+      "ari as",
+      "Ġcol o",
+      "rag on",
+      "Ġsch izophren",
+      "âī ¡",
+      "Ġreplic ative",
+      "Ġdegener ated",
+      "Ġsteep est",
+      "Vol ume",
+      "I ENT",
+      "P ublic",
+      "T en",
+      "en berger",
+      "ĠC oun",
+      "ĠE pp",
+      "iz o",
+      "Ġcomplex ed",
+      "Ġfer roc",
+      "ken stein",
+      "ĠJer ry",
+      "Ġparadox ical",
+      "x g",
+      "ic er",
+      "os ol",
+      "Ġan nu",
+      "Ġan kyl",
+      "ch ung",
+      "enti ous",
+      "Ġpres he",
+      "ene tic",
+      "ĠHe aling",
+      "ĠPar abolic",
+      "Ġfig s",
+      "ĠKin ematic",
+      "Ġoblig ate",
+      "ĠLay out",
+      "Ġtelem edicine",
+      "ĠLenn ard",
+      "p ci",
+      "ar one",
+      "ĠZ ach",
+      "Ġprot otyping",
+      "ĠMet agen",
+      "IM AL",
+      "cons cious",
+      "Ġquadr ilateral",
+      "ĠUncertain ties",
+      "ĠPref ecture",
+      "G BM",
+      "r als",
+      "al us",
+      "Ġh opes",
+      "Ġcl icks",
+      "ĠJ D",
+      "lect ance",
+      "Ġpath ologists",
+      "uss els",
+      "tis one",
+      "CP T",
+      "Ġmis con",
+      "ĠNeuro de",
+      "Ġmutagen ic",
+      "ĠMultim edia",
+      "Orig inal",
+      "ĠDra ke",
+      "P WM",
+      "Ġp iles",
+      "st ant",
+      "AR A",
+      "ĠR ING",
+      "mod ifying",
+      "Ġast rocyt",
+      "ĠCy st",
+      "Ġleg ends",
+      "gluc uron",
+      "Ġincom pletely",
+      "ĠConf ed",
+      "ĠDL BCL",
+      "ĠPap ua",
+      "Ġcontras tive",
+      "ĠSIM ULATION",
+      "ĠJu venile",
+      "aggreg ated",
+      "Ġc GMP",
+      "ic tive",
+      "ĠH NF",
+      "ĠN PV",
+      "ĠK oc",
+      "omet allic",
+      "min i",
+      "ĠQu antit",
+      "ĠCor nell",
+      "Ġded uction",
+      "Ġcoinc iding",
+      "ĠIr r",
+      "Prec ision",
+      "Ġgins eng",
+      "õ es",
+      "j er",
+      "ĠRe ader",
+      "ĠBy r",
+      "cor rections",
+      "dev ices",
+      "Ġamb ul",
+      "Ġped icle",
+      "ĠDepend ency",
+      "ĠStri king",
+      "Ġware house",
+      "Ġrecirc ulation",
+      "Ġgonor rhoeae",
+      "ĠP RES",
+      "ĠB har",
+      "Ġfl ushing",
+      "tor us",
+      "ĠIR B",
+      "gly cine",
+      "Ġmeth amphetamine",
+      "Ġmir rored",
+      "ĠWilliam son",
+      "Ġcath odes",
+      "hydrox ylase",
+      "Rad io",
+      "Ġfurn iture",
+      "ĠRosen berg",
+      "ĠNSA IDs",
+      "s emiconductor",
+      "Ġas ynchron",
+      "ĠB erm",
+      "ĠIn ten",
+      "ib e",
+      "For ce",
+      "path ogenic",
+      "sm okers",
+      "Ġdip henyl",
+      "ĠÐ ¸",
+      "Ġstand alone",
+      "Ġlith ospheric",
+      "Ġtrade offs",
+      "Ġantic h",
+      "Ġthym idine",
+      "ĠMedic inal",
+      "Ġentrepreneur ial",
+      "Ġtrapez oidal",
+      "ĠAs ynchronous",
+      "tif ying",
+      "ĠColl apse",
+      "ĠHE V",
+      "ĠFro zen",
+      "ĠTeich müller",
+      "rocnem ius",
+      "Ġf ern",
+      "Ġw s",
+      "om ol",
+      "Ġen closing",
+      "rap id",
+      "Ġlog ged",
+      "var vec",
+      "Ġampl ifying",
+      "diff erences",
+      "oton in",
+      "ĠProm oting",
+      "ĠFr itz",
+      "Ġattain able",
+      "Ġal tim",
+      "ĠO GD",
+      "Ġtherm ometer",
+      "Sol ver",
+      "ĠBir k",
+      "LEN BQU",
+      "ĠGate way",
+      "Ġengraft ment",
+      "F IF",
+      "H SD",
+      "Ġre structuring",
+      "ĠT ensile",
+      "ĠC ele",
+      "yl us",
+      "Ġfe ather",
+      "Ġdr ifting",
+      "ĠPre clinical",
+      "yr role",
+      "Ġcomm em",
+      "Ġfix ations",
+      "Pet sc",
+      "ĠIschem ia",
+      "a A",
+      "as oro",
+      "ĠS ony",
+      "ĠU t",
+      "Ġext ensor",
+      "ĠCh au",
+      "ĠIs otopic",
+      "IL I",
+      "CN P",
+      "ĠDE F",
+      "Ġmountain ous",
+      "Ġsarcom as",
+      "ugos lav",
+      "C ALL",
+      "S ensitive",
+      "at ro",
+      "Ġunc oupling",
+      "sk ew",
+      "ĠEm issions",
+      "inn ati",
+      "Ġconceptual ization",
+      "Ġow ns",
+      "Ġsquad ron",
+      "ĠStreng ths",
+      "C oh",
+      "U AL",
+      "m agenta",
+      "us b",
+      "ĠS PC",
+      "con es",
+      "ĠSe lecting",
+      "ĠPar ish",
+      "Ġvalid ates",
+      "ĠÍ Ĺ",
+      "Ġposterior ly",
+      "omon ad",
+      "V OL",
+      "j ectivity",
+      "ĠC LO",
+      "ĠV TA",
+      "Ġun pleasant",
+      "Ġcare ers",
+      "Ġautom orphic",
+      "ĠNan ow",
+      "Ġaster isks",
+      "ĠSchul z",
+      "publ ication",
+      "Ġb iv",
+      "Ġr ug",
+      "rec ognition",
+      "Ġref errals",
+      "Ġneur ones",
+      "ĠCa ffe",
+      "Con nor",
+      "ĠShe ffield",
+      "unit inib",
+      "ĠAnt agon",
+      "Ġpneum atic",
+      "Ġclean er",
+      "ĠBA O",
+      "ĠScilab String",
+      "neigh bour",
+      "E uler",
+      "ĠT uple",
+      "ot y",
+      "di an",
+      "Ġy oga",
+      "Ġev anes",
+      "Ġstar ved",
+      "Ġfluct uate",
+      "ĠBiomark er",
+      "Ġimpuls es",
+      "Ġoss ification",
+      "Ġdemyel ination",
+      "ĠS AD",
+      "ess ing",
+      "Ġred dish",
+      "Ġsyn th",
+      "Ġcurv ilinear",
+      "ĠDen is",
+      "Ġphone tic",
+      "Ġham mer",
+      "Ġepiderm idis",
+      "Ġplagi oclase",
+      "Ġ ĉ",
+      "Ġw olf",
+      "os ced",
+      "Ġphot othermal",
+      "Ġche wing",
+      "Max imum",
+      "Ġmism atched",
+      "ĠFc γ",
+      "Ġum brella",
+      "ĠSiber ian",
+      "ar ra",
+      "ip ped",
+      "ym pathetic",
+      "acc eleration",
+      "Ġeigen modes",
+      "ĠEqu ivalently",
+      "ĠPR ISMA",
+      "cons ervative",
+      "ñ ez",
+      "Ġvolcano es",
+      "Ġtelem etry",
+      "m ile",
+      "ĠB och",
+      "op rim",
+      "Ġinc ipient",
+      "Ġunderstand able",
+      "atric yclo",
+      "ĠLog ical",
+      "ĠQue ue",
+      "Ġcry ostat",
+      "defin ecolor",
+      "ĠS ae",
+      "Ġar ct",
+      "Ġso ul",
+      "ĠHist opathological",
+      "ĠNeu rot",
+      "Ġmethan olic",
+      "P x",
+      "ĠT itle",
+      "ot omic",
+      "ĠE ld",
+      "ĠE MA",
+      "Ġde brid",
+      "tim ulatory",
+      "ĠZ an",
+      "Ġnorm ot",
+      "Ġfluid ity",
+      "Ġfluid ized",
+      "pre viously",
+      "Ġcrack ed",
+      "ĠExpl aining",
+      "ĠON E",
+      "ĠFlor a",
+      "ĠHybrid ization",
+      "Ġretic ul",
+      "F K",
+      "n otic",
+      "Ġn A",
+      "ĠP ab",
+      "tic um",
+      "and y",
+      "ug ia",
+      "ile t",
+      "MI NG",
+      "Ġrest s",
+      "omp act",
+      "Ġtrack ers",
+      "phosph atase",
+      "ĠTransf ection",
+      "ĠHospit als",
+      "ac rine",
+      "ĠD ell",
+      "ĠV AE",
+      "ĠThrough put",
+      "hev sky",
+      "ĠSom mer",
+      "P SA",
+      "ì ļ",
+      "Ġb ush",
+      "Ġl unch",
+      "ĠS we",
+      "ĠIn struction",
+      "ak ami",
+      "Ġdis infect",
+      "Ġcor ps",
+      "ĉĉ ĠĠ",
+      "Ġprom pts",
+      "MS H",
+      "ĠAg rawal",
+      "Ġlys osome",
+      "integr in",
+      "Ġá» ¸",
+      "Ġnondec reasing",
+      "ĠRe quest",
+      "ĠRE P",
+      "occ us",
+      "Ġlag rangian",
+      "oreg ulation",
+      "оР»",
+      "ĠBos on",
+      "I so",
+      "at ellites",
+      "res ectable",
+      "ri v",
+      "Ġde aminase",
+      "Ġco heren",
+      "Ġdec oy",
+      "ĠExt inction",
+      "acet one",
+      "Ġgovernment al",
+      "Ġcum ulants",
+      "Ġviscos ities",
+      "Reg ister",
+      "document ed",
+      "Ġimmortal ized",
+      "D PP",
+      "G el",
+      "b ron",
+      "k ow",
+      "ĠPro portion",
+      "ĠCh ase",
+      "ĠCl ad",
+      "Ġadap ts",
+      "ĠCA V",
+      "ĠÅ ¼",
+      "Ġpel leted",
+      "Ġpeng uin",
+      "ĠZhe jiang",
+      "feas ible",
+      "D IV",
+      "i ya",
+      "Ġth rowing",
+      "res ia",
+      "ĠN r",
+      "ES P",
+      "CD F",
+      "sup pressed",
+      "Ġtet rachlor",
+      "Ġaer ospace",
+      "Un til",
+      "Ġpay offs",
+      "Ġtown ship",
+      "Ġester ification",
+      "ĠAch illes",
+      "Ġrac em",
+      "opyran oside",
+      "ĠC SM",
+      "ass is",
+      "Ġsuper cell",
+      "ĠReg ime",
+      "IR A",
+      "Ġsubsequ ences",
+      "ĠPen et",
+      "ĠAnaly tics",
+      "ĠLV EF",
+      "Ġbip henyl",
+      "G radient",
+      "os ylation",
+      "ĠW RF",
+      "of s",
+      "con ductors",
+      "Ġback ed",
+      "pid al",
+      "ĠNF AT",
+      "ĠRem ember",
+      "Ġtel omeric",
+      "Ġta urine",
+      "incre ases",
+      "Ġunint ended",
+      "ĠNerv ous",
+      "R as",
+      "y lyl",
+      "Ġa estiv",
+      "ĠS ick",
+      "ĠThe ta",
+      "Ġcl iques",
+      "Ġso fter",
+      "ĠQ RS",
+      "llip tic",
+      "ĠImmun otherapy",
+      "QU F",
+      "onom ously",
+      "ĠFL U",
+      "ĠIncor poration",
+      "ĠFormic idae",
+      "J R",
+      "w hole",
+      "Ġc asing",
+      "Ġn ob",
+      "ĠD ou",
+      "Ġint ronic",
+      "Ġent rapment",
+      "orb its",
+      "Ġsal am",
+      "ĠCR S",
+      "ĠSw an",
+      "ĠEd gar",
+      "Ġconcomit antly",
+      "atet racyclo",
+      "ĠA HR",
+      "tic ks",
+      "ĠB ing",
+      "ĠR ift",
+      "Ġpl ugging",
+      "Ġsc RNA",
+      "Ġout reach",
+      "ins kii",
+      "Ġcustom ary",
+      "Ġm d",
+      "ĠO zone",
+      "uss ing",
+      "other s",
+      "Ġentire ty",
+      "Ar th",
+      "Ac et",
+      "ĠFle et",
+      "ĠBehaviour al",
+      "ĠQSO s",
+      "ar ina",
+      "Ġpro drug",
+      "ĠB ros",
+      "ĠW orth",
+      "Ġy z",
+      "con tig",
+      "ĠAm orphous",
+      "ĠEr lang",
+      "Ġhon our",
+      "ĠâIJ ¥",
+      "Ġinfiltr ates",
+      "ĠIvan ov",
+      "ĠMunic ipality",
+      "ĠDial ogue",
+      "t one",
+      "Ġp ytest",
+      "ic ulus",
+      "ĠG oth",
+      "ĠX C",
+      "ĠSU MMARY",
+      "Ġshr inks",
+      "Ġinvers es",
+      "i omas",
+      "ro bi",
+      "ĠT PR",
+      "ĠA NA",
+      "ist ries",
+      "Ġreg iment",
+      "ind o",
+      "ĠRe production",
+      "lo qu",
+      "inf lation",
+      "ET X",
+      "Ġïĺ »",
+      "ĠAPP ENDIX",
+      "Ġwors ened",
+      "Ġpsori atic",
+      "Ġmidw ives",
+      "Ġtouc hed",
+      "Ë ĩ",
+      "ĠP atric",
+      "ĠD ON",
+      "ĠL IM",
+      "ak os",
+      "ĠV ie",
+      "ĠAn tit",
+      "Ġfl ake",
+      "ĠSch le",
+      "ĠCor onal",
+      "Ġsal ary",
+      "sl ight",
+      "ĠCA F",
+      "Ġsummar ise",
+      "Ġflav us",
+      "ĠBal anced",
+      "ĠPH OT",
+      "Ġmil let",
+      "Ġurg ency",
+      "ĠGle ason",
+      "ĠM ie",
+      "ĠD p",
+      "ĠG arg",
+      "Ġle prosy",
+      "Ġun occupied",
+      "ĠSt ret",
+      "ile pt",
+      "ĠCh or",
+      "ibr ate",
+      "ĠÍ ļ",
+      "ĠPH B",
+      "Ġmonot er",
+      "ĠJava Script",
+      "bt n",
+      "ĠPuls ar",
+      "ĠKirch hoff",
+      "Ġoverse as",
+      "Ġde phosphorylation",
+      "ort in",
+      "ĠPoly akov",
+      "Ġinsight ful",
+      "ĠPur ified",
+      "Ġanch orage",
+      "ĠGly coprotein",
+      "stud ies",
+      "Ġchron ology",
+      "rox ine",
+      "ĠNept une",
+      "B an",
+      "Ġl ion",
+      "PS D",
+      "ĠBar r",
+      "Ġdon key",
+      "Ġlikelihood s",
+      "atche wan",
+      "ot et",
+      "os pha",
+      "tic ism",
+      "Ġr y",
+      "ast hen",
+      "rho tic",
+      "ĠSub group",
+      "ye v",
+      "ĠPat ri",
+      "provid es",
+      "S GD",
+      "b erell",
+      "v w",
+      "ĠA ACR",
+      "Ġsm ears",
+      "OD S",
+      "sup plemented",
+      "ĠEng agement",
+      "oglob ulins",
+      "Ġirregular ly",
+      "ĠSz eg",
+      "ĠWol ff",
+      "Ġenanti omers",
+      "Ġobey ing",
+      "Ġdestro ying",
+      "om ially",
+      "ĠA ti",
+      "ĠG AT",
+      "ĠIn variants",
+      "ĠSc oring",
+      "Ġhal ides",
+      "Ġtransform ants",
+      "Ġforest ed",
+      "Ġgall ic",
+      "ĠBet ti",
+      "thread ed",
+      "ĠBud get",
+      "junc tive",
+      "ĠInnov ative",
+      "Ġposit rons",
+      "B razil",
+      "e ira",
+      "Ġl avas",
+      "ĠL t",
+      "ph oto",
+      "Ġsp am",
+      "Ġi h",
+      "ust ering",
+      "Ġbi oluminescence",
+      "ĠSh apes",
+      "UL TI",
+      "tri angles",
+      "ĠSM N",
+      "enh ancing",
+      "ĠReduc es",
+      "ĠTHEO REM",
+      "D op",
+      "Ġd L",
+      "em ptive",
+      "Ġrem inder",
+      "Ġgon ads",
+      "Ġxyl an",
+      "cult ures",
+      "t les",
+      "Ġt d",
+      "Ġe rected",
+      "ter one",
+      "ĠPD C",
+      "Ġincongru ent",
+      "Ġmembran ous",
+      "p ac",
+      "yl ess",
+      "Ġsub algebras",
+      "ĠCh ir",
+      "ĠZ IP",
+      "au tious",
+      "Ġlight ly",
+      "ĠPhot ometric",
+      "Trans fer",
+      "Ġket o",
+      "Ġexerc ised",
+      "dispers ive",
+      "ĠBET WEEN",
+      "ro u",
+      "Ġg arbage",
+      "ĠM af",
+      "ĠD oming",
+      "ĠSub space",
+      "ĠMar ÃŃa",
+      "Ġtetra hedra",
+      "ĠBark er",
+      "S ide",
+      "b ishop",
+      "i D",
+      "re versible",
+      "orm an",
+      "ores cein",
+      "ĠCont rib",
+      "Ġderiv atization",
+      "rome res",
+      "ĠAL D",
+      "EE K",
+      "ĠTre ating",
+      "comb ination",
+      "ïĺ »",
+      "restric tion",
+      "supset eq",
+      "ĠRAP D",
+      "Ġamend ment",
+      "zyn ski",
+      "Ġc aves",
+      "il ot",
+      "Ġabund antly",
+      "н а",
+      "Ġinject able",
+      "ĠReinfor ced",
+      "ĠWid th",
+      "ĠHaem ophilus",
+      "il ane",
+      "pro ps",
+      "Ġinter vertebral",
+      "Ġsc roll",
+      "Ġam put",
+      "ĠUn usual",
+      "Ġstat ically",
+      "Ġsyn ergies",
+      "Ġdim s",
+      "plas mic",
+      "Ġneutral ized",
+      "Se lected",
+      "Ġinher its",
+      "ĠAutom ation",
+      "Ġproto planetary",
+      "Stat ement",
+      "ĠAPO BEC",
+      "Ġcertif icates",
+      "ĠCit rus",
+      "quadrup lex",
+      "N ord",
+      "Ġf ran",
+      "ĠC arcin",
+      "ut an",
+      "ĠP ump",
+      "ĠB av",
+      "ĠG ras",
+      "ting ales",
+      "Ġcaus ally",
+      "Ġrad on",
+      "Comp are",
+      "Ġclamp ing",
+      "irre ducible",
+      "I HC",
+      "Ġ Ù",
+      "Ġc yp",
+      "ĠT PP",
+      "ĠS uff",
+      "und ra",
+      "ĠV illa",
+      "Ġrel ieved",
+      "ĠJ CM",
+      "Ġtreat y",
+      "IG EN",
+      "ĠDev onian",
+      "Ġerythrop o",
+      "R AP",
+      "Ġa versive",
+      "ent ate",
+      "od actyl",
+      "ĠPar al",
+      "Ġmill ed",
+      "Ġbio informatic",
+      "okine tic",
+      "ĠSTR ING",
+      "ĠPed ersen",
+      "d atabase",
+      "in organic",
+      "Ġde put",
+      "Ġne b",
+      "ip ed",
+      "Ġdiff used",
+      "oth ione",
+      "Ġnon stationary",
+      "Ġunder taking",
+      "ĠEn abling",
+      "Ġden atured",
+      "Ġload er",
+      "ĠLy on",
+      "ipar ametric",
+      "Ġmer istem",
+      "ĠAngi ogenesis",
+      "ĠPuls ed",
+      "Ġex cer",
+      "ĠD f",
+      "arc hes",
+      "Ġcoll ide",
+      "ĠRel ational",
+      "ĠNF κB",
+      "Met adata",
+      "ĠAdd ressing",
+      "Ġperc ussion",
+      "ĠFlore nce",
+      "Ġnymph s",
+      "C n",
+      "st orm",
+      "ĠG raz",
+      "com posite",
+      "ĠAd miral",
+      "ĠSc otia",
+      "Ġbre msstrahlung",
+      "aps ack",
+      "Ġminim izers",
+      "Ġmanage able",
+      "Ġcarboxyl ate",
+      "Ġintermedi ary",
+      "ĠBran ching",
+      "sched uler",
+      "inoc ulated",
+      "ĠExtrem ely",
+      "Ġantenn ae",
+      "ĠT ill",
+      "RE SH",
+      "Ġop acities",
+      "Ġchem opre",
+      "Ġaden ylate",
+      "Ġcircumst ance",
+      "ĠHash imoto",
+      "Ä Ľ",
+      "ce ae",
+      "ĠF m",
+      "ĠB X",
+      "Ġmean time",
+      "acc urate",
+      "col linear",
+      "ACT IC",
+      "ĠSlov enia",
+      "F ed",
+      "K h",
+      "T m",
+      "f ork",
+      "in ology",
+      "le f",
+      "ĠD CS",
+      "Ġher itable",
+      "Ġann ouncement",
+      "Ġbusiness man",
+      "Ġbor tezomib",
+      "Ġtour ist",
+      "ĠEt ymology",
+      "Ġdoctr ine",
+      "B IN",
+      "s uffix",
+      "ar as",
+      "ĠS au",
+      "un boldmath",
+      "ĠM EP",
+      "ink er",
+      "Ġoptim ism",
+      "ĠLe uc",
+      "eful ness",
+      "cr ust",
+      "ĠKe ys",
+      "ĠâĻ ¦",
+      "ĠBrand t",
+      "âĮ ¬",
+      "ĠSevent y",
+      "Ġnurs ery",
+      "Ġdeput y",
+      "Ã ¬",
+      "on is",
+      "am us",
+      "ĠC ig",
+      "Ġex ergy",
+      "ĠF requent",
+      "Ġab or",
+      "ĠJ azz",
+      "Ġstat ue",
+      "ĠSc enarios",
+      "Ġcyt ological",
+      "fig ures",
+      "MC I",
+      "dir name",
+      "Ġcytokines is",
+      "del ivery",
+      "ĠBow en",
+      "Ġflank ed",
+      "Ġregener ating",
+      "ĠFerr ari",
+      "k iss",
+      "ĠA val",
+      "ĠC IT",
+      "ĠM um",
+      "ĠL SB",
+      "og ging",
+      "Ġun ited",
+      "Ġtri tium",
+      "ont amination",
+      "co ef",
+      "Ġprop ell",
+      "tri ple",
+      "Ġimm ense",
+      "Ġcompl ained",
+      "Ġdielectric s",
+      "ĠCardi omy",
+      "Ġflood ed",
+      "ĠCov ariance",
+      "Att endance",
+      "T MP",
+      "Ġs ob",
+      "ĠS onic",
+      "ĠF TS",
+      "ĠR SD",
+      "ess ors",
+      "ĠW on",
+      "iff s",
+      "Ġflow chart",
+      "ĠEle mental",
+      "Ġì ŀ",
+      "Ġfoli age",
+      "differenti ated",
+      "ĠGlob ular",
+      "Ġpercept ron",
+      "candid ate",
+      "S ocial",
+      "W itt",
+      "d yn",
+      "p aces",
+      "Ġm Glu",
+      "Ġb anned",
+      "ol inite",
+      "ĠF riends",
+      "ĠL ibraries",
+      "unc es",
+      "ĠRe ach",
+      "ĠSk ills",
+      "Ġrecip es",
+      "Ġcann ula",
+      "ĠOrth odox",
+      "ĠCarb ohydrate",
+      "Ġarom atase",
+      "Åij s",
+      "Ġeman ating",
+      "e lected",
+      "Ġt ense",
+      "ĠF LC",
+      "ĠL ET",
+      "her jee",
+      "Ġsub band",
+      "oph one",
+      "ĠAc tual",
+      "ms gs",
+      "EM D",
+      "IS ON",
+      "ley ball",
+      "ĠNi u",
+      "Ġber ries",
+      "diagn ostic",
+      "N ER",
+      "Ġd Ω",
+      "per centage",
+      "ĠH erman",
+      "ĠG SD",
+      "Ġsub problem",
+      "over all",
+      "oph or",
+      "Ġdel ocalized",
+      "acc ount",
+      "ĠGe ographical",
+      "dist ances",
+      "Ġà µ",
+      "Ġneurot oxic",
+      "opod ia",
+      "ĠDic er",
+      "Ġðx Ãŀ",
+      "Ġd unes",
+      "Ġwh it",
+      "ĠIm mediate",
+      "ĠÌ ¸",
+      "Ġadhes ives",
+      "ĠNS s",
+      "Ġguess ing",
+      "ĠColumb us",
+      "ĠUr ugu",
+      "behavi our",
+      "ĠSerb ian",
+      "benzodiox ol",
+      "im plementation",
+      "os ensitive",
+      "ĠF ill",
+      "ph age",
+      "rec overy",
+      "ES R",
+      "Ġanaly sts",
+      "Ġdiss atisfaction",
+      "band ed",
+      "ĠDep ressive",
+      "ĠRT s",
+      "Ref s",
+      "mill imeter",
+      "ĠOls en",
+      "am pton",
+      "ĠA CA",
+      "ĠA vian",
+      "ĠF owler",
+      "ub ini",
+      "est amps",
+      "ĠPro test",
+      "Con nection",
+      "Ġmer chant",
+      "ĠEN C",
+      "ĠRy u",
+      "ĠLymph oma",
+      "ĠLar ry",
+      "Ġjaponic um",
+      "ĠSymbol s",
+      "L ib",
+      "V G",
+      "ĠT av",
+      "ĠAs sim",
+      "ĠLe ung",
+      "depend ency",
+      "larg est",
+      "ĠDO E",
+      "Ġalign s",
+      "ofl urane",
+      "ĠAdj usted",
+      "Ġpeculiar ities",
+      "decre ase",
+      "ĠPlac ement",
+      "v ig",
+      "z ak",
+      "Ġp enta",
+      "Ġf res",
+      "Ġac ros",
+      "Ġsol vability",
+      "ans ions",
+      "AL A",
+      "Ġmal function",
+      "ĠGiov anni",
+      "A OR",
+      "H ad",
+      "Ġp orn",
+      "und ice",
+      "ĠU i",
+      "Ġexp elled",
+      "ĠAn k",
+      "Ġdisc ounting",
+      "ĠReg ulating",
+      "aster y",
+      "phen ylethyl",
+      "Ġcast ration",
+      "Ġeryth romycin",
+      "Ġbif unctional",
+      "� �",
+      "ĠAlger ia",
+      "m ess",
+      "Ġw is",
+      "ĠT ay",
+      "ass umed",
+      "Ġes calation",
+      "Ġhydro per",
+      "Ġcall osum",
+      "Ġatom ization",
+      "ĠSA W",
+      "Ġacetyl cholinesterase",
+      "Ġsucceed s",
+      "Ġphysi otherapy",
+      "t ro",
+      "Ġm ason",
+      "ĠT MB",
+      "Ġph ant",
+      "Ġadjust s",
+      "anth a",
+      "ĠEisen stein",
+      "Ġshorth and",
+      "G ABA",
+      "Ġpro ver",
+      "Ġpat rol",
+      "ĠMod al",
+      "oll aries",
+      "ĠInter facial",
+      "ĠCI A",
+      "att n",
+      "ĠCrypt ococcus",
+      "athe cal",
+      "ĠFresh water",
+      "Ġspectro gram",
+      "opid ogrel",
+      "m orphism",
+      "Ġrel apsing",
+      "Ġgeneral izable",
+      "ĠSh ale",
+      "ĠTrans plant",
+      "cont raction",
+      "UR I",
+      "ĠPet rov",
+      "ĠSl iding",
+      "Ġanterior ly",
+      "Ġquas ilinear",
+      "Ġrip ples",
+      "Z P",
+      "b acterial",
+      "s pr",
+      "an imal",
+      "Ġre porters",
+      "ĠB SS",
+      "ĠD ia",
+      "ĠR SC",
+      "ound ing",
+      "IT HM",
+      "log ical",
+      "Ġpoly carbonate",
+      "An imal",
+      "umb ai",
+      "Ġarch ived",
+      "ĠDur ham",
+      "âĸ Ī",
+      "ĠVerm ont",
+      "Ġp w",
+      "ess en",
+      "Ġconst expr",
+      "ĠPr uss",
+      "Ġsharp ness",
+      "div ide",
+      "prim itive",
+      "Ġacryl ate",
+      "MY C",
+      "ĠMond ay",
+      "ĠSrin ivas",
+      "B orn",
+      "at tice",
+      "om orpha",
+      "ĠM ERS",
+      "ĠF actory",
+      "ĠW N",
+      "rec tile",
+      "Ġheat s",
+      "UN K",
+      "Ġsynchron ize",
+      "ĠAtten uation",
+      "Child ren",
+      "P at",
+      "p regnant",
+      "Ġw ished",
+      "Ġth awing",
+      "ĠB ey",
+      "ĠD ÃŃaz",
+      "Ġle ather",
+      "ĠUn ic",
+      "Ġspecial ised",
+      "Ġcataly tically",
+      "PL GA",
+      "hydroxy ethyl",
+      "Ġmag mas",
+      "Ġpron oun",
+      "Ġeut rophication",
+      "ĠWeek ly",
+      "M HD",
+      "m alloc",
+      "ec ologic",
+      "il o",
+      "ĠF requencies",
+      "Ġor chestra",
+      "Ġmetabol omic",
+      "ĠBlock ade",
+      "Ġasser ted",
+      "ĠLew y",
+      "Ġallevi ating",
+      "Ġoccl usions",
+      "Ġchor oid",
+      "techn ical",
+      "Ġenvision ed",
+      "ĠHous ing",
+      "P n",
+      "ĠT ECH",
+      "ĠS SH",
+      "ĠV alle",
+      "yl methyl",
+      "Ġph loem",
+      "ĠPro jects",
+      "but ton",
+      "Ġacceler ometers",
+      "umn i",
+      "ĠHand ling",
+      "Ġvas o",
+      "perme able",
+      "Ġc ords",
+      "ĠC f",
+      "ĠD z",
+      "Ġed itions",
+      "Ġhum erus",
+      "do ors",
+      "Ġdors olateral",
+      "Ġapt amers",
+      "Ġcommod ities",
+      "osper ms",
+      "Ġprednis one",
+      "I Q",
+      "M etal",
+      "t us",
+      "Ġis otopy",
+      "ĠThe ater",
+      "iff i",
+      "Ġy arn",
+      "de letion",
+      "ĠQ PO",
+      "Ġmulti objective",
+      "Ġur chin",
+      "Ġpuls ations",
+      "ĠSR P",
+      "ð tÃŀ",
+      "gluc oside",
+      "Ġdepart ures",
+      "Py Object",
+      "ĠBand width",
+      "ĠAccept ance",
+      "re ys",
+      "ĠI ON",
+      "Ġcomp uls",
+      "ĠJ W",
+      "Ġpart hen",
+      "Cl ose",
+      "ĠBa TiO",
+      "ñ oz",
+      "aggreg ate",
+      "Initi ally",
+      "q h",
+      "ĠC ancers",
+      "op in",
+      "ne ver",
+      "ism an",
+      "Ġconst ancy",
+      "Ġtr ucks",
+      "Ġvisual isation",
+      "ĠIll ness",
+      "Ġsulph ide",
+      "ĠMetabol ites",
+      "Ġoxys porum",
+      "H PP",
+      "Ġnor adrenaline",
+      "Ġcommut ativity",
+      "Qu ad",
+      "Ni O",
+      "ĠGet ting",
+      "Ġba it",
+      "Ġë °",
+      "Ġment ally",
+      "Ġaur oral",
+      "ĠDraw ing",
+      "S in",
+      "re ceiver",
+      "at ov",
+      "is otope",
+      "Ġis othi",
+      "ĠS enes",
+      "ĠA CO",
+      "ĠG CT",
+      "ys mal",
+      "ĠV og",
+      "Ġdist ractors",
+      "Ġconnected ness",
+      "Ġaccum bens",
+      "ä ck",
+      "hyd rated",
+      "Ġpharmac odynamic",
+      "Ġmineral ogy",
+      "Ġarth ropods",
+      "Ġmyc otoxins",
+      "Ġbatt les",
+      "ĠS ara",
+      "ĠE IS",
+      "ĠW inn",
+      "Ġlimb ic",
+      "WOR K",
+      "Å ½",
+      "Ġe aten",
+      "ĠT od",
+      "ap illary",
+      "ox yp",
+      "ĠNew ly",
+      "Ġcam el",
+      "arr ison",
+      "ECT OR",
+      "Ġhop efully",
+      "ĠHur witz",
+      "Ġib uprofen",
+      "ĠFIR ST",
+      "Ġbist able",
+      "Ġdismiss ed",
+      "g at",
+      "in ogen",
+      "ĠP ON",
+      "ph as",
+      "ĠK orn",
+      "Ġpoly aniline",
+      "ĠMic roscope",
+      "Ġmuc ous",
+      "Ġcollision less",
+      "hydrogen ase",
+      "Bu ild",
+      "pair ing",
+      "ĠWI MP",
+      "built in",
+      "ĠSepar ate",
+      "ĠCun ningham",
+      "ĠNecess ary",
+      "Ġb ry",
+      "ec rosis",
+      "ĠL SS",
+      "Ġsy philis",
+      "ĠV id",
+      "Ġcar rot",
+      "ĠRes istant",
+      "reg istration",
+      "Ġmy opathy",
+      "Ġang ry",
+      "MD R",
+      "Ġhypothesis ed",
+      "ĠVol terra",
+      "ele vation",
+      "Ġmyc obacteria",
+      "Ġcaud ate",
+      "i idae",
+      "Ġ Ç",
+      "ĠD ich",
+      "ĠR eth",
+      "ell us",
+      "ch amber",
+      "sh ine",
+      "och ore",
+      "ĠCol umns",
+      "CO UNT",
+      "Ġïĥ ²",
+      "ĠPrim ordial",
+      "Ġnegoti ations",
+      "sted t",
+      "R II",
+      "U ES",
+      "ti ques",
+      "ĠP fe",
+      "Ġpl ast",
+      "pr on",
+      "ĠZ w",
+      "ink ler",
+      "Ġmetabol ome",
+      "EG A",
+      "ĠSpect rophot",
+      "Ġubiqu ity",
+      "ĠElectro des",
+      "Ġchond ro",
+      "Domain Is",
+      "ĠResid ues",
+      "Ġdns DomainIs",
+      "D IC",
+      "p th",
+      "Ġa est",
+      "Ġc ient",
+      "Ġp essim",
+      "Ġre inst",
+      "ĠS ans",
+      "end azole",
+      "ĠU rine",
+      "Ġsub acute",
+      "ix imab",
+      "Ġprof itable",
+      "Ġmaxim ise",
+      "ĠDel aware",
+      "Ġclinic opathologic",
+      "Thermo Fisher",
+      "F AR",
+      "R AS",
+      "w itch",
+      "in activated",
+      "en esis",
+      "un less",
+      "ĠP anc",
+      "ĠM TS",
+      "ĠB ast",
+      "Ġch illing",
+      "Ġinc umbent",
+      "Ġj elly",
+      "Ġdistrib utive",
+      "Ġcy to",
+      "sc hen",
+      "Ġinduc ers",
+      "ĠNone quilibrium",
+      "ĠRob otics",
+      "ĠArgent ine",
+      "Ġmerid ian",
+      "Ġhun ger",
+      "Adap tive",
+      "Ġg or",
+      "ile psy",
+      "Ġnon vanishing",
+      "Ġpe ti",
+      "ĠMet formin",
+      "Ġbiom aterial",
+      "Ġanten nal",
+      "ĠAff ective",
+      "ĠAqu atic",
+      "enedi amine",
+      "ĠSiber ia",
+      "ĠPenic illium",
+      "F unctions",
+      "Ġ lec",
+      "Ġf eld",
+      "ĠS part",
+      "ĠC ement",
+      "ad di",
+      "se k",
+      "ĠN p",
+      "oles ky",
+      "ĠMac roscopic",
+      "è res",
+      "Ġcave at",
+      "Ġcourts hip",
+      "m ice",
+      "Ġf ence",
+      "Ġm ined",
+      "ul ink",
+      "ID A",
+      "Ġtrunc ate",
+      "ĠCatal an",
+      "Ġtran st",
+      "Ġamend ments",
+      "uncertain ty",
+      "Ġoroph aryngeal",
+      "ĠA id",
+      "ould er",
+      "ĠInc ident",
+      "Ġá IJ",
+      "angi ogenesis",
+      "ĠBE H",
+      "Ġic osa",
+      "ĠFOX P",
+      "frag ment",
+      "Ġscintill ator",
+      "J O",
+      "L aw",
+      "Ġp L",
+      "Ġet oposide",
+      "Ġpoly aden",
+      "Ġhabit ual",
+      "Ġtax i",
+      "Ġcum ulant",
+      "Ġhind rance",
+      "trig ger",
+      "r atios",
+      "il io",
+      "ĠP IR",
+      "ĠThe od",
+      "ĠM orton",
+      "ĠH af",
+      "ĠO ch",
+      "ĠEx o",
+      "Ġur tic",
+      "ĠCF RP",
+      "Sc reen",
+      "Sl ice",
+      "Ġmush rooms",
+      "Ġevanes cent",
+      "S x",
+      "Ë IJ",
+      "ì ŀ",
+      "Ġs igm",
+      "ic l",
+      "Ġg uests",
+      "ĠG IST",
+      "Ġdeform ities",
+      "poly acrylamide",
+      "Sign ificant",
+      "Ġimpression s",
+      "j math",
+      "em oral",
+      "ĠB n",
+      "ĠH DR",
+      "ĠK eck",
+      "Ġval ine",
+      "sp i",
+      "iter ate",
+      "Ġsyn c",
+      "oti ana",
+      "Inter val",
+      "ĠBra uer",
+      "Ġstic ky",
+      "ĠNeuros cience",
+      "Bax ter",
+      "Ġc asts",
+      "all ocation",
+      "ne al",
+      "Ġbi op",
+      "Ġrest orations",
+      "Im ages",
+      "mi tic",
+      "ĠEle vation",
+      "Ġabst inence",
+      "ĠLess er",
+      "ĠRain fall",
+      "P AM",
+      "W ol",
+      "us ch",
+      "Ġprom isc",
+      "na ïve",
+      "Ġded uc",
+      "acchar ide",
+      "Ġnom inally",
+      "ĠExpl oratory",
+      "Ġreconc iliation",
+      "linal g",
+      "T CR",
+      "Ġs ore",
+      "ĠN ab",
+      "Ġout group",
+      "Ġmon ophosphate",
+      "ins u",
+      "ĠAd dis",
+      "SP R",
+      "point ing",
+      "HE RE",
+      "ĠTechn ological",
+      "Ġcoch lea",
+      "Ġspheroid al",
+      "ĠBald win",
+      "F eed",
+      "Ġf using",
+      "Ġas per",
+      "Ġex osomal",
+      "ĠL inguistic",
+      "SC A",
+      "ĠEm pty",
+      "Ġvac ant",
+      "gly col",
+      "immun oprecipitation",
+      "ĠIT ER",
+      "Sn O",
+      "pattern s",
+      "contin ental",
+      "ĠAcceler ating",
+      "ĠAver aging",
+      "Ġchemoattract ant",
+      "h b",
+      "s ulph",
+      "ĠB x",
+      "Ġcom plicating",
+      "ĠW are",
+      "Ġso aking",
+      "Ġup regulate",
+      "-------- -",
+      "Ġsem ester",
+      "ĠBro d",
+      "Ġcasc ading",
+      "ĠCast ell",
+      "ĠẠ½",
+      "ĠEQU ATIONS",
+      "Ġparsim onious",
+      "Ġs orbent",
+      "Ġe ug",
+      "od in",
+      "ĠW ig",
+      "ĠTh ir",
+      "Ġsol v",
+      "Ġcar boplatin",
+      "Ġz ebra",
+      "ven ient",
+      "Ġmed Rxiv",
+      "Ġaut obi",
+      "Ġrepe atable",
+      "Ġmig rations",
+      "ĠÐ ´",
+      "hol onomic",
+      "Ġmoder ator",
+      "Ġchim era",
+      "ĠGrassmann ian",
+      "ĠR onald",
+      "ĠV ega",
+      "ast es",
+      "Ġqu otes",
+      "Ġmon ic",
+      "Ġprec oding",
+      "ĠAss isted",
+      "ĠNetwork ing",
+      "Ġfabric ating",
+      "Ġbot anical",
+      "Ġswarm s",
+      "Ġmartens itic",
+      "ellip tic",
+      "pher d",
+      "b aryon",
+      "x fe",
+      "ro ute",
+      "ĠF IL",
+      "op ies",
+      "ĠPC Bs",
+      "Ġer asure",
+      "ĠRem odeling",
+      "Ġana er",
+      "Sm ad",
+      "inj ured",
+      "Ġimmunocomp etent",
+      "d ell",
+      "f ailed",
+      "Ġs inking",
+      "or acic",
+      "Ġd red",
+      "ĠV DR",
+      "Ġconn ectors",
+      "Ġintr atumoral",
+      "Ġcommut ators",
+      "ĠAle ks",
+      "ĠDic ty",
+      "A k",
+      "Ġre calc",
+      "Ġis l",
+      "ot rim",
+      "nce phal",
+      "ĠRe es",
+      "Ġste atohepatitis",
+      "ĠPolar ized",
+      "SB ATCH",
+      "ĠCross ing",
+      "Acc uracy",
+      "ĠGi ardia",
+      "ĠNov o",
+      "Ġvig ilance",
+      "Ġphosphatidyl choline",
+      "ĠUE FA",
+      "J im",
+      "Ġf asted",
+      "ĠT iny",
+      "Ġl ang",
+      "iss ociation",
+      "Aut o",
+      "ĠNor folk",
+      "ĠArm s",
+      "ĠSW I",
+      "ĠAmb ros",
+      "transf ection",
+      "O ryza",
+      "h arm",
+      "ĠD s",
+      "Ġint rag",
+      "Ġcall er",
+      "Ġwr itings",
+      "ĠEl ast",
+      "ĠMar vel",
+      "ĠImmun odeficiency",
+      "ĠMill ion",
+      "Text ure",
+      "ĠIce Cube",
+      "sn ap",
+      "Ġenj oys",
+      "ĠChap el",
+      "ĠEstabl ishing",
+      "Act ually",
+      "Ġphosphoryl ates",
+      "Ġchin ensis",
+      "Ġrhabd omy",
+      "Ġemphys ema",
+      "M iddle",
+      "n ant",
+      "Ñ ħ",
+      "Ġt art",
+      "low est",
+      "hem ia",
+      "Ġutil ising",
+      "cons tit",
+      "Ġmag matism",
+      "о ÑĢ",
+      "ĠHas an",
+      "dispers ed",
+      "H ear",
+      "Q t",
+      "z ations",
+      "al on",
+      "ĠS tau",
+      "ĠA mer",
+      "os ystems",
+      "Ġdem arc",
+      "ĠNe oproterozoic",
+      "ĠMe k",
+      "ĠDis closure",
+      "Ġhemat ocrit",
+      "ĠCyt oscape",
+      "Ġram ification",
+      "Ġcommunic ative",
+      "Ġbutter flies",
+      "Ġantis era",
+      "Ġaestiv um",
+      "B ra",
+      "L TP",
+      "s ocket",
+      "ĠC herenkov",
+      "Ġch lam",
+      "ang ial",
+      "ult ured",
+      "eng ed",
+      "ĠCl inton",
+      "Ġmy oblasts",
+      "ĠComp ensation",
+      "ymmet rically",
+      "Ġemploy er",
+      "oz ol",
+      "ĠSA XS",
+      "Ġretin as",
+      "piper idine",
+      "XY Z",
+      "ĠRough ly",
+      "P rep",
+      "Ġb inge",
+      "Ġe rect",
+      "ĠO PER",
+      "Ġstress or",
+      "Ch rist",
+      "ĠPD Z",
+      "Ġsubst an",
+      "ĠSn ail",
+      "Ġlam ellae",
+      "ĠCycl ing",
+      "shif ting",
+      "ĠHs ieh",
+      "ver ify",
+      "Ġpre image",
+      "Ġar tillery",
+      "Ġep il",
+      "ĠAp ost",
+      "Ġhel met",
+      "Ġmach ined",
+      "ĠMin neapolis",
+      "ĠCr yp",
+      "Ġsitu ational",
+      "pass ing",
+      "quin azolin",
+      "ĠCro atian",
+      "Ġsta ircase",
+      "Bon net",
+      "N LP",
+      "c ium",
+      "Ġs keletons",
+      "Ġo xim",
+      "or ib",
+      "Ġre ticular",
+      "ĠS LS",
+      "ĠA romatic",
+      "ĠK es",
+      "Ġph or",
+      "Ġinv ocation",
+      "Ġdo zens",
+      "ai vely",
+      "Ġdetect ability",
+      "Ġconcer ted",
+      "yr ins",
+      "ĠProcess or",
+      "Ġtoler able",
+      "att ached",
+      "Ġanne xin",
+      "ĠROS AT",
+      "ĠAltern ate",
+      "ĠWa velength",
+      "ĠWill is",
+      "Ġsemic ontinuous",
+      "Ġadvoc acy",
+      "Ġoblig ation",
+      "chan ter",
+      "ĠInser tion",
+      "Ġsymbion t",
+      "Z M",
+      "Ġt ars",
+      "ro f",
+      "Ġre vival",
+      "ĠT ST",
+      "ĠE MP",
+      "Ġme x",
+      "ull in",
+      "ĠAd op",
+      "ĠDNA s",
+      "Ġemploy ers",
+      "MT s",
+      "ĠMart ÃŃn",
+      "electro des",
+      "ĠMedica id",
+      "Ġt gt",
+      "Ġl ognormal",
+      "ĠF rames",
+      "Ġper missive",
+      "ĠAr duino",
+      "Ġsem ilinear",
+      "ĠAss ign",
+      "ĠPr EP",
+      "ĠSi amese",
+      "benz imidazol",
+      "conn ectivity",
+      "ĠPE I",
+      "Ġbis ulfite",
+      "Ġacetyl transferase",
+      "Ġswim mer",
+      "ju ven",
+      "Ġjejun um",
+      "ĠCinc innati",
+      "ta i",
+      "ĠQ I",
+      "ĠCom mut",
+      "sp acing",
+      "Ġaff ords",
+      "itis ation",
+      "elastic ity",
+      "Ġdrag on",
+      "Ġproteas omal",
+      "Ġp ant",
+      "ĠN itro",
+      "Ġsp ic",
+      "Ġnan opl",
+      "ĠAll ied",
+      "Ġthor ax",
+      "ĠFT O",
+      "ĠJur kat",
+      "chiat ry",
+      "y oung",
+      "di rections",
+      "Ġne ocortex",
+      "ĠK ik",
+      "ang o",
+      "cl ay",
+      "iod o",
+      "Ġabove mentioned",
+      "ĠGu ardian",
+      "Con jecture",
+      "ĠTre nd",
+      "Ġfertil ized",
+      "ĠSulf ate",
+      "ochron ology",
+      "Ġcrani ofacial",
+      "ĠSask atchewan",
+      "Q Q",
+      "h man",
+      "Ġz ym",
+      "log s",
+      "Ġïģ ®",
+      "Ġgrad uating",
+      "pin ene",
+      "Ġî Ģ",
+      "Ġeti ological",
+      "ĠComprehens ion",
+      "Ġw andering",
+      "Ġl an",
+      "Ġsy st",
+      "return s",
+      "MO F",
+      "cho alveolar",
+      "ĠArm en",
+      "Ġbim etallic",
+      "ĠPoll en",
+      "F iles",
+      "Ġs sp",
+      "EN SI",
+      "ĠY us",
+      "Ġfin est",
+      "AG EN",
+      "Ġmicrobi omes",
+      "Ġpal ind",
+      "Ġpet als",
+      "ĠRadi otherapy",
+      "ophen one",
+      "spe aker",
+      "Ġcopep ods",
+      "Ġkan amycin",
+      "Ġdegran ulation",
+      "C onstruct",
+      "al ter",
+      "ĠF gf",
+      "ĠN BS",
+      "ĠIn complete",
+      "Ġpar cel",
+      "ne au",
+      "ĠÃ IJ",
+      "ĠCH A",
+      "Ġdual s",
+      "Ġsilic ates",
+      "ĠGlob ally",
+      "Ġkines in",
+      "f id",
+      "ĠC PD",
+      "ĠY ad",
+      "Ġdep ress",
+      "OD Y",
+      "ĠHist ograms",
+      "ĠSumm arization",
+      "aut omatic",
+      "ĠDom in",
+      "otrans formation",
+      "Ġventric les",
+      "Wid get",
+      "ĠPeters burg",
+      "Ġcholangi ocarcinoma",
+      "Ġnect ar",
+      "P IC",
+      "S cope",
+      "T ek",
+      "n itz",
+      "ĠP HD",
+      "Ġsp iro",
+      "ĠCO G",
+      "ĠDi oxide",
+      "conduc tivity",
+      "ĠGran ger",
+      "ĠWear able",
+      "ĠKenn eth",
+      "C CR",
+      "L INK",
+      "Ġ Ü",
+      "re tic",
+      "ly a",
+      "Ġdem ocratic",
+      "Ġradi ograph",
+      "ĠRel ax",
+      "ĠInc ubation",
+      "ĠDen oising",
+      "COL OR",
+      "ĠClos ure",
+      "H MM",
+      "ur d",
+      "ra da",
+      "ĠR v",
+      "ĠL uz",
+      "all s",
+      "Ġmulti spectral",
+      "IN ED",
+      "SC N",
+      "Ġdys lexia",
+      "Ġsett lers",
+      "ĠVL SI",
+      "Ġa vid",
+      "Ġl arynx",
+      "ĠC hess",
+      "ĠF AA",
+      "Ġdef ender",
+      "Ġlip olysis",
+      "ĠEl mer",
+      "ĠAff ymetrix",
+      "Ġrhod amine",
+      "M orph",
+      "S ite",
+      "p urity",
+      "Ġ Ê",
+      "ĠT ank",
+      "ĠM iao",
+      "Ġrec rystall",
+      "We yl",
+      "ĠGu il",
+      "Ġmis folded",
+      "su ited",
+      "ĠApproxim ations",
+      "ĠABC B",
+      "don or",
+      "GW AS",
+      "------------ ---",
+      "Ġpu tida",
+      "Ġimping ement",
+      "yam l",
+      "H ill",
+      "Ġt l",
+      "ag ua",
+      "tim ing",
+      "Ġreg enerate",
+      "Ġmulti lingual",
+      "rad or",
+      "class ifier",
+      "ĠJoh ansson",
+      "Ġsulf ides",
+      "ham mer",
+      "Ġwalk ed",
+      "Ġalloc ating",
+      "ĠGust av",
+      "Ġimmunoprec ipitated",
+      "ĠBris bane",
+      "Ġsandwic hed",
+      "ĠChatter jee",
+      "omand ibular",
+      "Ġo sc",
+      "Ġass ass",
+      "Ġmulti stage",
+      "Ġmulti partite",
+      "Ġpig mented",
+      "ĠVisual izing",
+      "Ke ys",
+      "pip eline",
+      "Ġdub bed",
+      "Ġc roc",
+      "ĠD LC",
+      "ĠR AT",
+      "ĠN ex",
+      "plic a",
+      "ting ham",
+      "ĠSp ider",
+      "Ġunc le",
+      "aut s",
+      "ĠHow e",
+      "Ġarth ropod",
+      "ĠPap ad",
+      "urg y",
+      "Ġaccl im",
+      "B road",
+      "ac er",
+      "ve z",
+      "ĠD ivers",
+      "Ġmod ifiable",
+      "Ġanti psychotics",
+      "Pro g",
+      "osa hexa",
+      "amb rian",
+      "ĠIon ization",
+      "Z A",
+      "o ate",
+      "Ġp ays",
+      "Ġe wes",
+      "Ġbe aches",
+      "Ġev il",
+      "ĠCD s",
+      "na ud",
+      "Ġconform ity",
+      "ĠDM N",
+      "Ġcollabor ate",
+      "Ġdeterior ate",
+      "VAL ID",
+      "ĠVeg as",
+      "Ġultrac ent",
+      "B RA",
+      "R ub",
+      "Y C",
+      "f h",
+      "å ľ",
+      "ĠO WL",
+      "ose ismic",
+      "of errin",
+      "och thon",
+      "ĠTNF R",
+      "small setminus",
+      "ĠArg ument",
+      "Ġgranul ocytes",
+      "Ġram ified",
+      "Ġepi phy",
+      "f usc",
+      "ec dot",
+      "Ġh w",
+      "ĠN MS",
+      "erc us",
+      "Ġtet her",
+      "ĠTra it",
+      "Ag Cl",
+      "ĠNear by",
+      "Ġhelmin th",
+      "Ġlae vis",
+      "ĠB AR",
+      "ĠN ancy",
+      "ĠG yn",
+      "Ġsec reting",
+      "St ellar",
+      "Ġsil hou",
+      "IM T",
+      "Ġscaffold ing",
+      "ĠConver ter",
+      "h id",
+      "Ġn ud",
+      "est rian",
+      "ann o",
+      "Ġdep iction",
+      "orem ost",
+      "ĠSh and",
+      "AB CD",
+      "ĠPD L",
+      "Ġdys phagia",
+      "Ġintr at",
+      "Ġhem ip",
+      "Ġadapt able",
+      "long mapsto",
+      "ss bauer",
+      "ĠMcC arthy",
+      "ĠAuto immune",
+      "ĠCut aneous",
+      "Inser ting",
+      "M aterial",
+      "ĠA a",
+      "ĠG av",
+      "Ġmon ocular",
+      "equ il",
+      "ĠGe off",
+      "Ġtet hered",
+      "obil ized",
+      "ĠShort ly",
+      "Det ails",
+      "Ġrefuge e",
+      "Ġabsc isic",
+      "FBQ yx",
+      "Ġdemoc racy",
+      "c rafted",
+      "d ifluor",
+      "y der",
+      "ess ment",
+      "Ġhist opathologic",
+      "Ġast rocytic",
+      "Ġwithd rew",
+      "Ġm oles",
+      "ath ic",
+      "mon o",
+      "man ual",
+      "Ġfood borne",
+      "ĠRep ository",
+      "Ġcover t",
+      "OT E",
+      "Ġtight ness",
+      "Ġinstanti ated",
+      "Ġwatermark ing",
+      "Ġartem isinin",
+      "L anguage",
+      "O ES",
+      "c ant",
+      "al ready",
+      "un ts",
+      "iti a",
+      "ĠK aren",
+      "Ġall uvial",
+      "stratig raphy",
+      "ĠP IV",
+      "ĠF aces",
+      "ĠB im",
+      "ap plications",
+      "ta ils",
+      "Ġel d",
+      "IR B",
+      "ĠIN TE",
+      "ĠNot Implemented",
+      "Ġmis classified",
+      "Ġfertil izers",
+      "ĠElectric ity",
+      "Ġtribut aries",
+      "ĠDeut sch",
+      "Ġslee ve",
+      "f uzzy",
+      "ĠM TL",
+      "ĠB res",
+      "ĠW yn",
+      "Ġk yr",
+      "ne uronal",
+      "ox ymethyl",
+      "dis order",
+      "inc hes",
+      "ram idal",
+      "Ġpoly imide",
+      "Res Net",
+      "ĠEd mund",
+      "Ġdegener acies",
+      "uther ford",
+      "Drop out",
+      "ij Ģ",
+      "Ġv oiced",
+      "ĠG omes",
+      "iv ities",
+      "con ductance",
+      "com pl",
+      "vec s",
+      "Ġtun a",
+      "ĠKin ect",
+      "Ġconvey ed",
+      "Ġsphing osine",
+      "b at",
+      "ĠP urs",
+      "ound ed",
+      "ĠSt am",
+      "ĠX III",
+      "ĠCom ics",
+      "MS M",
+      "SS L",
+      "Ġperf luor",
+      "Ġfluor inated",
+      "foli os",
+      "Ġre position",
+      "ĠS err",
+      "ĠC ors",
+      "ĠL abs",
+      "Ġco x",
+      "ĠAc quired",
+      "Ġreason ed",
+      "Gen ome",
+      "ĠPi per",
+      "Ġcompac tified",
+      "Ġherbiv ore",
+      "lofen ac",
+      "Ġb oss",
+      "ĠB s",
+      "ĠE MR",
+      "Ġsh oe",
+      "Ġcare rs",
+      "Ch rom",
+      "SV P",
+      "ĠTri angle",
+      "Ġhemat ite",
+      "dor f",
+      "ĠMove ments",
+      "ĠVes icles",
+      "Olymp us",
+      "M ol",
+      "Ġl end",
+      "ur as",
+      "ĠA SE",
+      "ĠW KB",
+      "pro ved",
+      "ĠK V",
+      "ĠU ART",
+      "log arithmic",
+      "ĠAD I",
+      "ĠDo ing",
+      "Ġce ase",
+      "Ġleng thening",
+      "Ġpyrophosph ate",
+      "F re",
+      "ĠC LD",
+      "ĠM LS",
+      "ĠPl um",
+      "Ġprop ionate",
+      "ĠGu atem",
+      "CK D",
+      "Ġis os",
+      "ĠM anning",
+      "ne uro",
+      "OP ER",
+      "ĠWil helm",
+      "Ġacad emia",
+      "ACh R",
+      "ĠIner tial",
+      "O cc",
+      "u jan",
+      "on as",
+      "Ġin ulin",
+      "ic ia",
+      "and al",
+      "ĠK ahn",
+      "Ġun manned",
+      "ĠCo arse",
+      "Ġgu ilty",
+      "ĠPe i",
+      "ĠLuc a",
+      "ĠFib roblast",
+      "a vian",
+      "v x",
+      "Ġd izziness",
+      "ĠD ox",
+      "ĠH our",
+      "Ġdec oration",
+      "Ġver ifier",
+      "rad o",
+      "Ġfoot prints",
+      "Ġdisp ensable",
+      "ĠAna erobic",
+      "Io T",
+      "ĠR isks",
+      "ĠG LS",
+      "Ġch ords",
+      "oid y",
+      "Ġneu rolog",
+      "ru h",
+      "Ġvirtual ization",
+      "Ġproton ation",
+      "ĠConstant in",
+      "Ġkeyp oints",
+      "B uck",
+      "H opf",
+      "M uch",
+      "reg ime",
+      "Ġprom ised",
+      "ai j",
+      "ĠDes ulf",
+      "ĠForm ulas",
+      "Ġhum p",
+      "ln c",
+      "ĠSu icide",
+      "ĠHO MA",
+      "ogly cer",
+      "ĠProte omics",
+      "Ġdict ate",
+      "ĠSper mat",
+      "F un",
+      "Ġs ag",
+      "ĠF am",
+      "ep pe",
+      "ĠJ ah",
+      "Ġar isen",
+      "oph armaceutical",
+      "SA GE",
+      "ĠTH IS",
+      "enh ance",
+      "Ġnap us",
+      "ro e",
+      "ens ch",
+      "de formation",
+      "bon es",
+      "ĠEr nest",
+      "ira bility",
+      "dec om",
+      "Ġcrust aceans",
+      "Ġguarantee ing",
+      "OV As",
+      "ĠMultic enter",
+      "Ġct DNA",
+      "Ġforamin ifera",
+      "L inn",
+      "Ġc ups",
+      "es ch",
+      "Ġd F",
+      "ĠT ah",
+      "pl l",
+      "pro jects",
+      "ĠU CI",
+      "Ġhuman ized",
+      "Ġabs l",
+      "ĠSch o",
+      "Ġliter als",
+      "ĠSV R",
+      "Ġtoxic ology",
+      "pg f",
+      "ĠIPT G",
+      "ĠMEASU REM",
+      "o ing",
+      "ĠP asc",
+      "ĠB au",
+      "ĠW annier",
+      "Ġhyp re",
+      "att ributes",
+      "Ġprecondition er",
+      "Wr iting",
+      "Ġgyp sum",
+      "y uan",
+      "Ġup regulates",
+      "Ġte lec",
+      "ĠDisc re",
+      "gu ard",
+      "Ġdeb ates",
+      "Ġparasit oid",
+      "L am",
+      "ti ge",
+      "Ġis opropanol",
+      "ĠI was",
+      "pl ify",
+      "ind olin",
+      "ĠAp ollo",
+      "Ġland ed",
+      "Ġbeam line",
+      "Un ion",
+      "Ġrecipro c",
+      "ĠRoss by",
+      "princ ipal",
+      "Ġdescend ant",
+      "ĠAnalog ously",
+      "Ġdereg ulation",
+      "D SM",
+      "c ta",
+      "Ġre built",
+      "ĠM und",
+      "ĠF EC",
+      "ry n",
+      "plic e",
+      "ĠY ugoslav",
+      "ĠNorth western",
+      "ĠHom ogen",
+      "ĠLI SA",
+      "Ġinvest or",
+      "H SA",
+      "H PO",
+      "Ġd ictionaries",
+      "ĠC ategor",
+      "Ġcomp acted",
+      "till ed",
+      "ç »",
+      "Ġf ines",
+      "ur ans",
+      "Ġbetween ness",
+      "ĠZ ig",
+      "sc hema",
+      "Ġcommun e",
+      "ĠQu inn",
+      "Ġana phylaxis",
+      "TI ES",
+      "Ġsnow pack",
+      "ĠDO A",
+      "ag os",
+      "ĠO dd",
+      "ard e",
+      "Ġev oke",
+      "ĠOc ular",
+      "Ġfa ulting",
+      "Ġvolcan ism",
+      "ĠPale ozoic",
+      "Ġmycel ium",
+      "ĠAdjust ment",
+      "I CT",
+      "N ov",
+      "al ias",
+      "ĠT ul",
+      "ĠH h",
+      "Ġev ade",
+      "OR s",
+      "Ġstreng thens",
+      "ĠUS GS",
+      "Ġlic ensing",
+      "ĠCle ment",
+      "ĠPhyt ophthora",
+      "r ified",
+      "Ġe ighteen",
+      "Ġto ps",
+      "ĠC LP",
+      "Ġst abilities",
+      "ĠP PT",
+      "ĠB IN",
+      "ĠR ak",
+      "Ġgen istein",
+      "vol ve",
+      "Ġquick er",
+      "ĠCaus ed",
+      "benef it",
+      "Y B",
+      "l ift",
+      "Ġh ood",
+      "ĠS Cs",
+      "of a",
+      "ĠMic ron",
+      "angi otensin",
+      "Ġfeat hers",
+      "Ġantifer romagnet",
+      "DEC REF",
+      "yled ons",
+      "Ġmyri ad",
+      "Ġ iz",
+      "ĠT rough",
+      "âĪ «",
+      "hem oglobin",
+      "ĠEn velope",
+      "ĠCl ick",
+      "sol iton",
+      "ĠSyn chrotron",
+      "Ġlag ged",
+      "MY B",
+      "Ġtroph oblast",
+      "Ġinterrog ation",
+      "onv uls",
+      "B ac",
+      "Ġa periodic",
+      "Ġg pu",
+      "Ġpro pidium",
+      "te ps",
+      "ĠK arp",
+      "ĠV az",
+      "ack age",
+      "ons on",
+      "In str",
+      "fil er",
+      "rifug ation",
+      "KO V",
+      "four th",
+      "Ġôı¼ IJ",
+      "hyper bolic",
+      "sche tz",
+      "Disc ussion",
+      "ĠOrient ed",
+      "j ad",
+      "Ġa uctions",
+      "us ivity",
+      "ĠC ran",
+      "Ġk d",
+      "Ġint est",
+      "ros arcoma",
+      "ugg er",
+      "ĠIL P",
+      "ĠST A",
+      "Ġrevers als",
+      "Ġgrap es",
+      "ĠPop ulus",
+      "ĠKit aev",
+      "ĠAV P",
+      "Pre viously",
+      "Ġquadr atically",
+      "ĠLOC AL",
+      "B ert",
+      "P ED",
+      "l ive",
+      "à ¬",
+      "Ġb idding",
+      "Ġto ss",
+      "ent o",
+      "Ġth ylak",
+      "Ġcomp rehend",
+      "Ġdi ve",
+      "Ġapplic ants",
+      "ĠÄ ħ",
+      "ĠVol canic",
+      "adap tation",
+      "Ġá¹ Ģ",
+      "ĠJans sen",
+      "Ġadjo ining",
+      "ozol omide",
+      "C IS",
+      "d C",
+      "duc ted",
+      "ĠAn ast",
+      "ĠEm ployment",
+      "ĠEnd ocrine",
+      "sil oxane",
+      "S ession",
+      "ĠN arr",
+      "ĠâĪĴ âĪĨ",
+      "de ev",
+      "oth iaz",
+      "ring ing",
+      "po inted",
+      "Ġacet ylene",
+      "Ġglob ulin",
+      "pack ing",
+      "ĠUs es",
+      "A ES",
+      "H en",
+      "ĠS avage",
+      "ĠC anc",
+      "ist o",
+      "ĠChrom osomal",
+      "Ġcement ed",
+      "Ġpyro x",
+      "ĠConstit utive",
+      "Ġphthal ate",
+      "mechan ism",
+      "Ġcyclospor ine",
+      "P AP",
+      "ar ted",
+      "ĠR DT",
+      "Ġpl ains",
+      "Cl one",
+      "prop anol",
+      "regular ity",
+      "Ġcot angent",
+      "ĠLes lie",
+      "ĠNit rate",
+      "ĠKaw asaki",
+      "ĠPage Rank",
+      "Ġanhyd rase",
+      "ĠKrish na",
+      "Ġhemicell ulose",
+      "Ġ ery",
+      "ll is",
+      "Ġmicro gram",
+      "ĠDel igne",
+      "Ġenfor ces",
+      "Ġthrombol ysis",
+      "P arse",
+      "or vastatin",
+      "Ġm ated",
+      "ĠC rystalline",
+      "Ġaut oradi",
+      "Ġtherm ophilic",
+      "inf ectious",
+      "Ġult ram",
+      "ĠML L",
+      "ĠFib ers",
+      "Ġulcer ation",
+      "omed ial",
+      "stratig raphic",
+      "Ġtouc hes",
+      "r he",
+      "Ġt ame",
+      "ĠC ulic",
+      "AR DS",
+      "ch ter",
+      "Ġcounter clockwise",
+      "Ġcam ps",
+      "VD C",
+      "Ġmeth adone",
+      "depend ently",
+      "valid ate",
+      "Ġprecl udes",
+      "Ġparliament ary",
+      "ĠINTE REST",
+      "ĠS erg",
+      "ĠC BC",
+      "ere lla",
+      "ay i",
+      "ĠR AB",
+      "Ġch ym",
+      "Ġnan ospheres",
+      "Ġdiab etics",
+      "cons ervation",
+      "Ġperme ate",
+      "plot ted",
+      "Ġna phthalene",
+      "ĠBon n",
+      "ĠElectro static",
+      "Ġinvent ories",
+      "Gaussian ity",
+      "ĠAden osine",
+      "Del ay",
+      "ĠBegin ning",
+      "Ġs ided",
+      "ĠC ushing",
+      "ĠH v",
+      "Ġco ined",
+      "ĠAl m",
+      "sc anning",
+      "fer til",
+      "Ġα v",
+      "ĠRe activity",
+      "Ġproxim ate",
+      "depend encies",
+      "Ġdens ification",
+      "Ġôı¼ ij",
+      "Ġbacteri ocin",
+      "weak ly",
+      "Ġdenti stry",
+      "ĠOri ental",
+      "Ġdorm ant",
+      "Ġp C",
+      "Ġm um",
+      "RE s",
+      "Ġcon val",
+      "Ġbi ota",
+      "Ġmulti linear",
+      "ĠPT FE",
+      "Ġnarrow band",
+      "ĠRot ational",
+      "Ġhoney bee",
+      "ĠChlor ophyll",
+      "Bas eline",
+      "F ern",
+      "Ġl k",
+      "ĠM ash",
+      "ri ved",
+      "ĠB ases",
+      "ĠD ah",
+      "ĠK ui",
+      "ĠÃ ĵ",
+      "ĠRec ycl",
+      "AG N",
+      "PD E",
+      "Ġclim atological",
+      "ĠBas ically",
+      "cons erved",
+      "abs orbing",
+      "ĠKos zul",
+      "ouss ines",
+      "Ġm dx",
+      "ith ymia",
+      "ĠH inton",
+      "Ġk h",
+      "Ġad mittance",
+      "ĠV y",
+      "Ġext rema",
+      "Ġcre ftype",
+      "sub st",
+      "Ġble omycin",
+      "LINE AR",
+      "A Q",
+      "i om",
+      "Ġn ong",
+      "op ian",
+      "se in",
+      "ud al",
+      "Ġear ning",
+      "Ġstandard ize",
+      "ĠPar ticular",
+      "Ġwave vector",
+      "dx dy",
+      "ĠMac Donald",
+      "ĠEst uary",
+      "valid ated",
+      "ĠHur st",
+      "ĠMuk herjee",
+      "Ġbival ves",
+      "Ġjug ular",
+      "U b",
+      "v ill",
+      "en ough",
+      "Ġin forms",
+      "an atomical",
+      "ul ou",
+      "res a",
+      "ĠP MC",
+      "ĠM ira",
+      "ĠR PL",
+      "ĠSD C",
+      "Ġhem i",
+      "Mo S",
+      "ĠFlo at",
+      "Ġoccl usal",
+      "ĠRain bow",
+      "ĠProvid ing",
+      "Ġsupercapac itor",
+      "os f",
+      "ĠI RT",
+      "Ġad m",
+      "Ġdec oders",
+      "ĠX R",
+      "ĠRes cue",
+      "Ġent om",
+      "Ġmor tal",
+      "An gle",
+      "Ind ia",
+      "ĠMal i",
+      "Ġinsp ecting",
+      "ĠGALAX Y",
+      "ĠEri ks",
+      "Y F",
+      "r ings",
+      "Ġs ir",
+      "Ġg sl",
+      "ĠB ubble",
+      "ĠD CA",
+      "ĠW idespread",
+      "ass ignment",
+      "Ġge omorph",
+      "ĠPre ference",
+      "CO PD",
+      "process ors",
+      "cut off",
+      "ĠFlow er",
+      "phen omen",
+      "mus ic",
+      "ĠSlov akia",
+      "Support ing",
+      "b low",
+      "ed it",
+      "ĠT rophy",
+      "ĠA SF",
+      "ĠM oses",
+      "Ġind els",
+      "Ġnon human",
+      "Ġhand ic",
+      "Ġrepair ing",
+      "Ġmicrom eter",
+      "ĠPhilip pe",
+      "Ġexud ates",
+      "ĠâĹ ĭ",
+      "Ġamalg am",
+      "K in",
+      "f ors",
+      "f ron",
+      "Ġan abolic",
+      "ĠE ich",
+      "NA N",
+      "Ġpseud ogap",
+      "analy zed",
+      "Ġtack led",
+      "agin ous",
+      "Ġlubric ant",
+      "Ġradion uclides",
+      "arrest in",
+      "oussines q",
+      "L if",
+      "Î ¥",
+      "re ceived",
+      "as tive",
+      "ĠP BC",
+      "Ġam oxicillin",
+      "cop per",
+      "ubl ing",
+      "oph ages",
+      "ĠSe as",
+      "ĠEl ite",
+      "PM MA",
+      "Ġchol ang",
+      "Depend ing",
+      "Ġas bestos",
+      "ĠF ecal",
+      "ĠR ath",
+      "ĠL ey",
+      "Ġfact ored",
+      "bb les",
+      "Ġtoken izer",
+      "Ġofficinal is",
+      "ĠNUC LE",
+      "ĠS emicon",
+      "ĠB ous",
+      "ĠR is",
+      "Ġlo ans",
+      "AC P",
+      "âĻ Ģ",
+      "phos ate",
+      "Ġc herry",
+      "an an",
+      "ar re",
+      "ĠC redit",
+      "ise xual",
+      "ĠAc ta",
+      "ĠLet ting",
+      "ĠInf arction",
+      "ĠAcc ounting",
+      "Ġcounter stained",
+      "Ġaer ogel",
+      "standard ized",
+      "Ġly ase",
+      "seg ments",
+      "Ġbac helor",
+      "Ġh ue",
+      "ĠN ETs",
+      "Ġun adjusted",
+      "Ġmicro hardness",
+      "Ġsingle ts",
+      "ĠSP ACE",
+      "ĠHyd raulic",
+      "MET HOD",
+      "ĠBj ör",
+      "ĠK U",
+      "Ġrep ur",
+      "Ġradi ocarbon",
+      "Ġheter ogeneities",
+      "Ġgast rocnemius",
+      "ĠLT D",
+      "Ġaccident ally",
+      "Process ing",
+      "Dop pler",
+      "T BI",
+      "Ġl ingual",
+      "ĠA GS",
+      "ĠF rontal",
+      "ĠB rack",
+      "the ma",
+      "Ġrepresent able",
+      "Ġpress urized",
+      "AD R",
+      "ĠMicro fluid",
+      "Ġê °",
+      "Ġreus able",
+      "Ġv endor",
+      "all er",
+      "Ġdi version",
+      "FA ST",
+      "ĠKir by",
+      "ĠStim ulus",
+      "Ġattach ments",
+      "ĠBrid ging",
+      "ĠRober to",
+      "Ġqueu ing",
+      "t ling",
+      "ro ots",
+      "ĠM x",
+      "ĠM arrow",
+      "ĠL ocus",
+      "Ġun important",
+      "erg arten",
+      "ÃŃ k",
+      "ĠPot ent",
+      "ĠBruns wick",
+      "ĠS CT",
+      "ĠM our",
+      "em ias",
+      "ĠN CS",
+      "ch icine",
+      "ĠO ryza",
+      "Ġwhere ver",
+      "ĠX GB",
+      "CO X",
+      "Ġhydrogen ated",
+      "Ġhyd raz",
+      "ĠPers ons",
+      "Ġframes hift",
+      "Ġelectroly tic",
+      "ĠSen egal",
+      "Ġphag ocyt",
+      "Ġinstantaneous ly",
+      "ĠGround water",
+      "Ġimper ial",
+      "ĠRhod e",
+      "ÅĦ ska",
+      "ovis ual",
+      "onts ize",
+      "ĠExplan ation",
+      "Ġempower ment",
+      "N TA",
+      "P u",
+      "P or",
+      "S ched",
+      "e ats",
+      "Ġ ys",
+      "in ous",
+      "Ġw ilt",
+      "ĠM ov",
+      "ect on",
+      "ĠG ins",
+      "int roduction",
+      "ince ption",
+      "ĠInter preting",
+      "Ġstart up",
+      "Ġalb ino",
+      "Ġtet ras",
+      "ĠHouse hold",
+      "ĠEL M",
+      "Ġspor ulation",
+      "Ġosm ol",
+      "B is",
+      "er ule",
+      "ĠE AR",
+      "Ġim balances",
+      "Ġk t",
+      "Ġj l",
+      "ges terone",
+      "eral a",
+      "ĠPo inter",
+      "ĠHR QoL",
+      "ĠRi et",
+      "ĠEsc ape",
+      "pur ified",
+      "Ġinstanti ation",
+      "m atis",
+      "ion a",
+      "Ġn oxious",
+      "ĠN og",
+      "Ġj am",
+      "ĠAnt oni",
+      "ĠGod d",
+      "ĠPersonal ized",
+      "Ġperm uted",
+      "ĠS HE",
+      "ĠO blast",
+      "ĠFor bes",
+      "ĠRes veratrol",
+      "ĠFe Se",
+      "Ġelectro deposition",
+      "Ġhome obox",
+      "Ġpy ogenes",
+      "Ġviol in",
+      "Ġiso electric",
+      "ĠPP G",
+      "prob ably",
+      "AMP K",
+      "ĠWol fe",
+      "Ġultraf ine",
+      "B eyond",
+      "on at",
+      "ed ian",
+      "EN ABLE",
+      "ĠH AM",
+      "so ut",
+      "ĠOp inion",
+      "rin ted",
+      "typ ing",
+      "Un known",
+      "Ġbuck ets",
+      "Ġintuition istic",
+      "algorithm s",
+      "S SC",
+      "b ir",
+      "ĠP ond",
+      "ad vert",
+      "ip in",
+      "Ġup wind",
+      "ĠCl aire",
+      "ĠMat uration",
+      "ĠPr P",
+      "OP O",
+      "FORM ANCE",
+      "Ġd M",
+      "ĠC ities",
+      "Ġinter related",
+      "ĠAp paratus",
+      "Ġprec ious",
+      "cript ors",
+      "Ġprepared ness",
+      "ĠAR CH",
+      "ĠPath ogens",
+      "HO ST",
+      "ĠGib bons",
+      "Ġirregular ity",
+      "ĠLip ids",
+      "Ġcf u",
+      "Ġvas odilation",
+      "imet re",
+      "impro ved",
+      "m q",
+      "ĠH ens",
+      "ĠL oci",
+      "unc redited",
+      "Ġmulti grid",
+      "tig o",
+      "Ġaccount ability",
+      "ench yme",
+      "Ġdisadvant aged",
+      "Ġbisp henol",
+      "Ġt ic",
+      "Ġfor ks",
+      "ĠW ester",
+      "ĠV ii",
+      "ĠJ ere",
+      "sim ultaneous",
+      "ĠGu arant",
+      "ĠDo yle",
+      "Ġpotenti ates",
+      "lass ified",
+      "Ġile al",
+      "Ġvasoconstr iction",
+      "M ODULE",
+      "N ano",
+      "W ood",
+      "ĠT AT",
+      "ur ious",
+      "un ya",
+      "Ġins tillation",
+      "ĠSim mons",
+      "ĠDi rectional",
+      "Ġmal ate",
+      "Ġplant ation",
+      "Ġuns olved",
+      "ĠTa uri",
+      "Ġov ine",
+      "Ġkeratin ocyte",
+      "ĠKull back",
+      "ĠKazakh stan",
+      "Ġh irs",
+      "ĠA erobic",
+      "ĠH ai",
+      "ĠR iley",
+      "ens ible",
+      "Ġinter planetary",
+      "Ġtrans its",
+      "Ġgener ous",
+      "Ġcal pain",
+      "Ġapp ended",
+      "ĠHydro dynamics",
+      "Ġcolon ize",
+      "Ġheart beat",
+      "Ġmetast as",
+      "Ġpy reth",
+      "ĠPA K",
+      "ĠÐ ¡",
+      "multi plet",
+      "ĠBrad y",
+      "Ġpropri a",
+      "ĠFron tier",
+      "ĠJoy ce",
+      "ĠP GF",
+      "ĠM cl",
+      "rec urrent",
+      "ĠRe placing",
+      "inf erence",
+      "ĠWh itt",
+      "Ġschool ing",
+      "ĠHa rold",
+      "Ġabst ractions",
+      "âĬ ķ",
+      "mem cpy",
+      "Ġmicron ucle",
+      "Ġradion uclide",
+      "ot yl",
+      "ĠM IF",
+      "ĠM US",
+      "Ġex foli",
+      "ĠF amilial",
+      "Ġcl am",
+      "ON O",
+      "Ġvan illa",
+      "Ġpast oris",
+      "ĠAT L",
+      "ĠBur sts",
+      "Qu antitative",
+      "Ġelic iting",
+      "Ġgranul omatous",
+      "Ġbrow sing",
+      "t racks",
+      "Ġh ij",
+      "ĠB CP",
+      "inc omp",
+      "az id",
+      "ck pt",
+      "Ġlink ers",
+      "Ġsqu id",
+      "Ġhead aches",
+      "ĠMor al",
+      "Ġstabil isation",
+      "&& &&",
+      "ĠSu fficient",
+      "ĠArch aea",
+      "Ġì ł",
+      "ĠLuc iferase",
+      "Cam era",
+      "expand ed",
+      "Ġmyster ious",
+      "H PS",
+      "ĠB J",
+      "ĠK NN",
+      "Ġsuper hydrophobic",
+      "ĠHydro thermal",
+      "ĠRus so",
+      "ĠArsen ic",
+      "Ġnormot ensive",
+      "ul timate",
+      "ĠC MIP",
+      "ex amined",
+      "Ġmicro porous",
+      "Ġfore ver",
+      "ĠST ING",
+      "IG S",
+      "ĉĉĉ ĠĠ",
+      "Pl ant",
+      "Ġcoherent ly",
+      "charg ing",
+      "Ġinher it",
+      "altern ative",
+      "ĠBap tist",
+      "F m",
+      "b ipy",
+      "Ġo ler",
+      "ĠSub stit",
+      "Ġult rap",
+      "free ze",
+      "perg ill",
+      "POS E",
+      "Ġadvertis ements",
+      "ECH AN",
+      "Bay esian",
+      "Ġcob ordism",
+      "¸ °",
+      "ĠA ER",
+      "ĠA IP",
+      "ĠL NA",
+      "ess entially",
+      "rec iprocal",
+      "ĠAn and",
+      "Ġsm eared",
+      "ones e",
+      "ethyl amine",
+      "ĠER S",
+      "Ġjud icial",
+      "Ġwood land",
+      "ĠGre gor",
+      "Ġtab ular",
+      "avir in",
+      "mir ror",
+      "Ġja undice",
+      "astig otes",
+      "ĠL GBT",
+      "ĠN aj",
+      "Ġsub scheme",
+      "Ġmulti user",
+      "Ġdrain s",
+      "Ġevac uated",
+      "phosphor yl",
+      "ĠFeld man",
+      "ĠTRI zol",
+      "ĠBLE U",
+      "a romatic",
+      "o viÄĩ",
+      "p ion",
+      "re pr",
+      "ro th",
+      "ĠF ES",
+      "ĠL eeds",
+      "Ġun g",
+      "ob ranch",
+      "Ġpat ency",
+      "ĠSc r",
+      "ĠSim plex",
+      "pec ies",
+      "Ġbenef ici",
+      "Ġpolymer ases",
+      "ĠCy gn",
+      "oct adec",
+      "Ġpunct ured",
+      "Ġjaponic us",
+      "ĠFPG As",
+      "f rown",
+      "Ġe b",
+      "ut iny",
+      "ĠP oy",
+      "ĠB rent",
+      "ĠB AM",
+      "ĠH ick",
+      "ĠN PS",
+      "ĠG DF",
+      "ĠV IRT",
+      "Ġinter l",
+      "Ġsc Fv",
+      "Ġte amm",
+      "Ġparticip atory",
+      "Ġexist ential",
+      "Ġoste omyelitis",
+      "Ġpneum othorax",
+      "std out",
+      "Ġsinglet ons",
+      "hyp othesis",
+      "strat ified",
+      "U SD",
+      "on asal",
+      "er is",
+      "im its",
+      "ĠI Cs",
+      "ĠE ncephal",
+      "iz i",
+      "ĠG radients",
+      "Ġall op",
+      "Ġcor p",
+      "con structed",
+      "Ġmon ument",
+      "sim ulator",
+      "ĠFerm ions",
+      "ĠWy oming",
+      "Ġprednis olone",
+      "L ang",
+      "N otes",
+      "e er",
+      "Ġf ighter",
+      "ent rant",
+      "ĠN ij",
+      "ĠG PD",
+      "ĠPro l",
+      "Ġreal isation",
+      "Ġpack ings",
+      "ĠDisc overing",
+      "ĠAng lo",
+      "ĠCass ini",
+      "exec ute",
+      "Ġinhab ited",
+      "ac ross",
+      "ĠC ram",
+      "ĠN BR",
+      "ant es",
+      "Ġdis persing",
+      "ach andran",
+      "ĠU ND",
+      "Ġshould ers",
+      "Ġcr ises",
+      "ustr ine",
+      "Ġprop ane",
+      "UN E",
+      "br ush",
+      "Ġeti ologies",
+      "Ġshot gun",
+      "show ing",
+      "ĠPhyt ochemical",
+      "ĠMeh ta",
+      "orr hea",
+      "ĠImag ery",
+      "T re",
+      "w c",
+      "Ġe luent",
+      "ond in",
+      "ĠAt titude",
+      "Ġfer romagnet",
+      "Ġcounter measures",
+      "Ġalk anes",
+      "ĠCap illary",
+      "lat ent",
+      "Ġsolub il",
+      "View er",
+      "áz quez",
+      "ĠPunj ab",
+      "a as",
+      "t ang",
+      "Ġim ports",
+      "ĠY ounger",
+      "rough ly",
+      "We inberg",
+      "ĠAt kinson",
+      "bf a",
+      "MP a",
+      "ste el",
+      "PC P",
+      "chlor inated",
+      "ĠPsych ometric",
+      "Ġpyro ptosis",
+      "Ġwat ched",
+      "ĠPerc utaneous",
+      "R BD",
+      "V ARI",
+      "at u",
+      "ĠW ake",
+      "Ġcan yon",
+      "ip arous",
+      "Ġsc all",
+      "com pletely",
+      "inter fer",
+      "ophy ceae",
+      "Ġfatal ities",
+      "cz ak",
+      "ĠPathophys iology",
+      "L em",
+      "l ach",
+      "t uary",
+      "Ġa lex",
+      "Ġs isters",
+      "Ġp um",
+      "ĠC atch",
+      "ĠE ber",
+      "ine x",
+      "ph the",
+      "Ġbo ar",
+      "ĠSo ul",
+      "Ġcat fish",
+      "Ġcloud y",
+      "ĠBu ilt",
+      "ophyl line",
+      "ĠRib osome",
+      "ĠAnomal ies",
+      "Y D",
+      "c ategorical",
+      "w or",
+      "op enta",
+      "ĠL IB",
+      "Ġr ick",
+      "Ġradi ations",
+      "Ġhyper cube",
+      "Ġmal treatment",
+      "Ġî ĦĦ",
+      "dispers ity",
+      "contin ent",
+      "Dig ital",
+      "ĠCory neb",
+      "Ġre vert",
+      "ĠT EA",
+      "ĠM LR",
+      "ĠF CM",
+      "ĠL amp",
+      "iz abilities",
+      "Ġcar ved",
+      "ĠMon oclonal",
+      "Ġpen is",
+      "ĠMor ales",
+      "En ter",
+      "ester ification",
+      "Ġcab bage",
+      "RAN TIES",
+      "Ġdebrid ement",
+      "L ead",
+      "c AMP",
+      "Ġc esium",
+      "ĠC ubic",
+      "Ġun imodular",
+      "ĠEx port",
+      "Ġanalys er",
+      "den otes",
+      "Ġrad ically",
+      "ĠHist ology",
+      "Ġmelan omas",
+      "Ġwors hip",
+      "ĠHimal ayan",
+      "ĠInte grable",
+      "benzenesulf onamide",
+      "Ġharb ored",
+      "P utting",
+      "ĠT ir",
+      "ĠU TI",
+      "cent ers",
+      "ĠPl uripot",
+      "Ġhar bors",
+      "Ġcarb am",
+      "ĠApp alach",
+      "ĠJo an",
+      "ĠCommission er",
+      "ĠGem ini",
+      "N ear",
+      "O PS",
+      "Q G",
+      "p ytorch",
+      "st aining",
+      "Ġh CG",
+      "Ġg avage",
+      "per haps",
+      "ĠG rib",
+      "ĠZ ah",
+      "Ġcompar ably",
+      "ĠBi oscience",
+      "SP L",
+      "Con nell",
+      "ĠAir way",
+      "prim ed",
+      "Ġsubmuc osal",
+      "Enh anced",
+      "Ġwis dom",
+      "V N",
+      "ĠM umbai",
+      "ri us",
+      "ĠR GD",
+      "ĠR Neasy",
+      "ma i",
+      "ĠAD L",
+      "Ġadop tive",
+      "Out lined",
+      "ĠWAR RANTIES",
+      "ĠViol ence",
+      "Ġcater p",
+      "F und",
+      "d θ",
+      "ĠP ok",
+      "ĠB enson",
+      "ĠR IG",
+      "ĠV s",
+      "Ġinst ants",
+      "ĠMulti drug",
+      "PD MS",
+      "CON ST",
+      "Ġcart ridge",
+      "ĠLif estyle",
+      "ĠCOND ITIONS",
+      "odys plastic",
+      "CONTR OL",
+      "L HC",
+      "ti re",
+      "ĠS tain",
+      "Ġy x",
+      "Ġj unctional",
+      "ob o",
+      "ann ah",
+      "ĠCP AP",
+      "Ġsound ness",
+      "ĠUl timate",
+      "sil icon",
+      "Ġparal og",
+      "E vents",
+      "G as",
+      "J E",
+      "ĠJ orge",
+      "Ġover production",
+      "Ġmax illa",
+      "ĠRe asons",
+      "we eks",
+      "ĠNe arest",
+      "Ġhead space",
+      "ĠAT C",
+      "bal ancing",
+      "Ġjud ging",
+      "ĠUnivers ality",
+      "Ġsinus es",
+      "Ġretro peritoneal",
+      "Det ection",
+      "Ġhydrolys ate",
+      "H och",
+      "w rapper",
+      "Ġp Ka",
+      "Ġl asso",
+      "ĠA lu",
+      "ĠA PR",
+      "ĠD ors",
+      "ĠD arboux",
+      "ĠR FS",
+      "ĠK har",
+      "ĠTh rom",
+      "Ġdesign ate",
+      "arc o",
+      "Ġtherm ostat",
+      "ĠGl acial",
+      "IF F",
+      "ĠMan ifest",
+      "Ġinters persed",
+      "haus er",
+      "ĠDD X",
+      "Ġa le",
+      "ti des",
+      "Ġl accase",
+      "ĠH ered",
+      "ĠR acial",
+      "ĠK ats",
+      "aj o",
+      "ĠCo ordinated",
+      "ĠProb ably",
+      "Ġtit anate",
+      "SL AM",
+      "dri ving",
+      "ĠEmerg ent",
+      "ĠDri ves",
+      "Ġoblig ations",
+      "Ġnebul ae",
+      "f ried",
+      "ith in",
+      "ĠP GD",
+      "oc clusion",
+      "ĠU H",
+      "Ġsub routine",
+      "oid in",
+      "Ġann oy",
+      "ĠVir asoro",
+      "inst ances",
+      "ĠDer by",
+      "Ġtriang ulations",
+      "Ġcutoff s",
+      "ĠOrganization al",
+      "ĠVen k",
+      "ĠEG TA",
+      "ĠDeut sche",
+      "Ġantine ut",
+      "ĠVulner ability",
+      "i ated",
+      "Ġa vium",
+      "Ġh sp",
+      "em ulsions",
+      "ad herence",
+      "ĠU PS",
+      "ma ph",
+      "ĠV AP",
+      "rel atively",
+      "Ġmax ill",
+      "oph ase",
+      "Th reshold",
+      "ĠSup p",
+      "eth oprim",
+      "Ġpenet rated",
+      "Ġblast ing",
+      "ĠAdvant ages",
+      "B US",
+      "ol son",
+      "rec v",
+      "Ġcar nitine",
+      "ten ing",
+      "Ġprov oked",
+      "vari ous",
+      "ĠCal ab",
+      "len eck",
+      "ĠPark in",
+      "Ġblow up",
+      "ĠDW I",
+      "synt hesized",
+      "Ġdisproportion ately",
+      "Ġcardio respiratory",
+      "ĠXanth omonas",
+      "Ġpunc ta",
+      "bdd c",
+      "ĠP ACS",
+      "ase g",
+      "Ġinc urs",
+      "ost a",
+      "ĠJ L",
+      "ĠWe ierstrass",
+      "ole ucine",
+      "Ġfin als",
+      "Ġcaus ation",
+      "ĠDi rective",
+      "ĠPor to",
+      "ĠFlo res",
+      "arbon yl",
+      "---------------------------------------------------------------- ------------",
+      "histor ic",
+      "K ähler",
+      "on na",
+      "Ġc el",
+      "ĠT BA",
+      "ĠO phthal",
+      "Ġsub threshold",
+      "Ġlip s",
+      "ĠSub strates",
+      "Ġpen insula",
+      "Ġads or",
+      "Ġdry ness",
+      "mass es",
+      "è me",
+      "stro k",
+      "ĠExpand ed",
+      "Ġg c",
+      "ĠG olf",
+      "Ġcri tique",
+      "ĠÍ ©",
+      "Ġlens ed",
+      "ĠKing ma",
+      "ĠGold man",
+      "ĠFac ile",
+      "Car l",
+      "Ġchond rites",
+      "ĠCoh omology",
+      "ĠSocio economic",
+      "ĠDominic an",
+      "ĠAzerbai jan",
+      "ĠA ne",
+      "ĠM idd",
+      "ĠN ed",
+      "Ġem ulate",
+      "ĠSh akes",
+      "Ġlik ed",
+      "Ġbuild up",
+      "Ġexcess ively",
+      "ĠÅ ¶",
+      "ĠAdap ted",
+      "Ġauthentic ated",
+      "Ġlocom otive",
+      "Ġsubm ill",
+      "Ġinterpre ter",
+      "ĠVibr ational",
+      "R α",
+      "l aden",
+      "p kl",
+      "r w",
+      "y et",
+      "en zymes",
+      "Ġw av",
+      "ĠN MT",
+      "ath ion",
+      "Ġbi otechnological",
+      "arc s",
+      "Ġact uated",
+      "Ġher ring",
+      "EC G",
+      "OC I",
+      "Ac tivated",
+      "Ġpara ph",
+      "Obs ervation",
+      "ĠEk man",
+      "ancell or",
+      "veli hood",
+      "G auss",
+      "H AL",
+      "r dev",
+      "t bl",
+      "ic its",
+      "ĠR oux",
+      "op ram",
+      "Ġser opositive",
+      "ĠPhys ically",
+      "ĠEd u",
+      "Ġdocument ing",
+      "ĠÐ ¾",
+      "ĠSmall er",
+      "cher y",
+      "Ġlanth anide",
+      "T oday",
+      "Ñ Ĩ",
+      "Ġo titis",
+      "ĠC ores",
+      "if olium",
+      "ĠZ el",
+      "Ġtim ings",
+      "co arse",
+      "rep air",
+      "ĠLD PC",
+      "Ġbow l",
+      "ĠEpid ermal",
+      "Ġâľ ²",
+      "Ġsynonym s",
+      "Ġ ENT",
+      "Ġb illiard",
+      "Ġe jac",
+      "ĠB AA",
+      "Ġsc ientif",
+      "Ġγ γ",
+      "Ġcaps ular",
+      "Ġaz ithromycin",
+      "Ġcred entials",
+      "ĠḠł",
+      "ĠGli oblastoma",
+      "Ġunco ated",
+      "Ġhib ern",
+      "ĠT os",
+      "ĠB aro",
+      "ĠK ass",
+      "yl and",
+      "ĠX M",
+      "Ġagg ra",
+      "Ġneutral ize",
+      "lic ted",
+      "Ġsound track",
+      "ĠKn ud",
+      "ensors hip",
+      "emp fer",
+      "ĠHald ane",
+      "ĠR ocks",
+      "ĠG ou",
+      "ĠO pi",
+      "ib acterium",
+      "ep tives",
+      "ust a",
+      "par s",
+      "uk awa",
+      "be in",
+      "eli us",
+      "aver aging",
+      "ĠMW CNT",
+      "Ġshield ed",
+      "Ġquatern ionic",
+      "Batch Norm",
+      "Ġd ella",
+      "ĠT p",
+      "Ġby product",
+      "ĠG ow",
+      "ĠJ O",
+      "Ġparameter ize",
+      "gl er",
+      "Ġfac ult",
+      "ĠÍ µ",
+      "Ġnom ination",
+      "Ġbath s",
+      "Ġinstall ations",
+      "ĠJust in",
+      "Ġchampionship s",
+      "t ap",
+      "ĠS anc",
+      "ĠS usp",
+      "Ġsub leading",
+      "Ġdef ended",
+      "Ġbut yl",
+      "rem ote",
+      "Ġcarb ides",
+      "ĠPredic ts",
+      "ĠPrior ity",
+      "ĠAntib iotics",
+      "ĠPU FAs",
+      "ĠMIC s",
+      "ĠMaxim ization",
+      "b are",
+      "ĠP CN",
+      "Ġinf ested",
+      "Ġsol enoid",
+      "Ġag ronomic",
+      "AN GE",
+      "Re v",
+      "ĠNK G",
+      "Ġsap onins",
+      "Recomm end",
+      "Ġshar pen",
+      "othio yl",
+      "s uspended",
+      "at ron",
+      "us age",
+      "il ter",
+      "ĠN ER",
+      "CR IPT",
+      "inf ections",
+      "Ġheter osexual",
+      "Ġmes oc",
+      "ĠBob by",
+      "alloc ate",
+      "ĠPay ne",
+      "ĠSult an",
+      "è ¡",
+      "rac les",
+      "rib e",
+      "ĠDo ub",
+      "ĠPA F",
+      "commun ication",
+      "Ġninet eenth",
+      "Ġpopl ar",
+      "pgf strok",
+      "pgfstrok ecolor",
+      "S LE",
+      "ec ia",
+      "Ġdet ach",
+      "Ġchar ity",
+      "Ġmon ochrom",
+      "Ġpres cribe",
+      "Ġsuper massive",
+      "Ġgu ards",
+      "Ġcycl oaddition",
+      "Co hen",
+      "phosph atidyl",
+      "cre ated",
+      "ĠElectro dynamics",
+      "Ġplasm ons",
+      "Ñģ к",
+      "ĠDaph nia",
+      "e asy",
+      "Ġa q",
+      "Ġf imb",
+      "Ġw rest",
+      "ĠT end",
+      "hip p",
+      "Ġorgan isational",
+      "MA E",
+      "OP Y",
+      "Ġpotenti ated",
+      "Ġbr ute",
+      "omass ie",
+      "aun ay",
+      "l uster",
+      "Ġo phi",
+      "un ge",
+      "ĠP om",
+      "Ġpl ague",
+      "ber ries",
+      "Ġinv iscid",
+      "ĠQ E",
+      "inc ident",
+      "xim ide",
+      "Ġest rogens",
+      "ĠTrans parent",
+      "vere ign",
+      "ĠPre ferred",
+      "Ġelast ase",
+      "C iv",
+      "J F",
+      "K u",
+      "c aster",
+      "Ġsp ends",
+      "Ġabst racted",
+      "otechn ical",
+      "Ġbreed ers",
+      "Ġsyring ae",
+      "Ġclot ting",
+      "Af rican",
+      "P EC",
+      "us ep",
+      "Ġst ark",
+      "so lete",
+      "of ovir",
+      "Ġsens ations",
+      "az awa",
+      "Ġbiom echanics",
+      "Ġemerg encies",
+      "Ġspectrom eters",
+      "Ġhemisp heric",
+      "Ġdiscrimin atory",
+      "ĠInsp ection",
+      "nd im",
+      "RE P",
+      "ĠW ess",
+      "Ġhyper algesia",
+      "IR C",
+      "Ġauthors hip",
+      "CP A",
+      "Ġrotation ally",
+      "ĠPy th",
+      "ĠYam aguchi",
+      "Field s",
+      "Ġenrol ment",
+      "ĠReth inking",
+      "G ate",
+      "ì ĺ",
+      "Ġc ements",
+      "ĠT TS",
+      "ĠF ink",
+      "ad us",
+      "ĠL l",
+      "Ġim plicate",
+      "ann ihilation",
+      "Ġfeed ers",
+      "ĠPD X",
+      "ĠFran çois",
+      "Sp earman",
+      "ĠBenchmark ing",
+      "F orest",
+      "e vidence",
+      "en oyl",
+      "ol actone",
+      "ce phaly",
+      "ĠP EA",
+      "ĠN SE",
+      "Ġno tified",
+      "Ġpoly electrolyte",
+      "ĠMal ik",
+      "anth ine",
+      "tet rad",
+      "Ġflag ella",
+      "ãĥ ¼",
+      "orp ion",
+      "Ġbuy ers",
+      "Ġoligodend rocyte",
+      "ĠNMD AR",
+      "ĠHarvest ing",
+      "Ġkar st",
+      "I BD",
+      "ĠF olk",
+      "Ġsub carrier",
+      "Ġno tices",
+      "ĠY ous",
+      "aw ak",
+      "Ġadvers aries",
+      "Lo oking",
+      "Ġthym ocytes",
+      "Ġmening ioma",
+      "Ġillumin ate",
+      "ĠSPD X",
+      "Ġimping ing",
+      "associ ative",
+      "Ġt iger",
+      "le on",
+      "Ġst ature",
+      "ĠH ood",
+      "ĠR utherford",
+      "ĠE IT",
+      "Ġinf antile",
+      "ĠQ ubit",
+      "Ġpack s",
+      "ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ ĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠĠ",
+      "azol am",
+      "ภ²",
+      "ĠTun isia",
+      "dil ution",
+      "Ġsymp atric",
+      "Ġliquef action",
+      "porph yrin",
+      "G n",
+      "R ib",
+      "is othermal",
+      "ap o",
+      "Ġreg ressors",
+      "man i",
+      "ĠIL s",
+      "oxid ants",
+      "At om",
+      "lig o",
+      "ĠSR AM",
+      "alc one",
+      "cs r",
+      "Ġc autious",
+      "Ġhand lers",
+      "Ġgast ritis",
+      "ĠSuper vision",
+      "Ġevapor ative",
+      "R UN",
+      "ĠI CG",
+      "ĠP rague",
+      "ĠM LC",
+      "ĠM oney",
+      "ĠR m",
+      "ĠE CS",
+      "vel t",
+      "ĠV g",
+      "Ġbi ography",
+      "Ġmin istry",
+      "con volution",
+      "ogen omics",
+      "round ing",
+      "ĠPh ag",
+      "Ġaud iences",
+      "ĠHC Ws",
+      "Ġblast ocysts",
+      "Ġdiagon als",
+      "Ġpreca utions",
+      "Í ĵ",
+      "ec ewise",
+      "ĠT oxin",
+      "ĠH app",
+      "ĠâĢ ĭ",
+      "Ġpop ulate",
+      "mm ol",
+      "ĠPR S",
+      "Ġreinfor ces",
+      "IST IC",
+      "ozo ites",
+      "arri val",
+      "Ġpave ment",
+      "REGIST ER",
+      "ĠG ases",
+      "ĠEx hib",
+      "Ġfactor izations",
+      "Ġmy opia",
+      "Ġmov able",
+      "ĠLI MIT",
+      "Ġsole us",
+      "DO UBLE",
+      "ĠInput s",
+      "foot notes",
+      "BIT S",
+      "ĠCyp rus",
+      "re ports",
+      "ĠP AA",
+      "ist al",
+      "Ġgroup oids",
+      "orph in",
+      "ĠCo ordinates",
+      "bor o",
+      "ĠOs lo",
+      "when ever",
+      "Ġplaus ibility",
+      "ĠFox O",
+      "ĠIntr usion",
+      "Ġsimplic es",
+      "ĠFas o",
+      "Ġpic osecond",
+      "ĠAns atz",
+      "Import antly",
+      "ĠHutch inson",
+      "ov ani",
+      "ĠAs ymptotics",
+      "Ġapp ra",
+      "ĠEx ogenous",
+      "Ġcap tions",
+      "ĠAc anth",
+      "Ġill icit",
+      "ĠBl adder",
+      "Ġbo om",
+      "ĠSal inity",
+      "Ġmusc ul",
+      "eptid yl",
+      "Ġaval anches",
+      "Hel per",
+      "Ġbival ve",
+      "Ġreimburs ement",
+      "z zo",
+      "rom atosis",
+      "Ġpar acetamol",
+      "vi o",
+      "Ġval pro",
+      "cl amation",
+      "Ġu u",
+      "ĠSo C",
+      "ĠGi ac",
+      "Ġly copene",
+      "Fl ags",
+      "Ġstic king",
+      "Ġsquee ze",
+      "synt hetic",
+      "osahexa enoic",
+      "m obile",
+      "v ect",
+      "ĠB aryon",
+      "Ġne f",
+      "Ġfl atter",
+      "SS I",
+      "Ġsch w",
+      "ancre as",
+      "Bu f",
+      "Sol id",
+      "ĠRIP A",
+      "Squ are",
+      "Ġtranscend ental",
+      "Ġc yn",
+      "Ġm f",
+      "ĠS ew",
+      "ĠP IT",
+      "oc s",
+      "ĠB ash",
+      "Ġsur prised",
+      "Ġaut onomously",
+      "Ġlocal izing",
+      "Ġvis itor",
+      "Ġpers isting",
+      "Ġland fill",
+      "date time",
+      "Ġfire f",
+      "ĠEngine ered",
+      "ĠSn yder",
+      "ochrom es",
+      "fraction ated",
+      "G PI",
+      "Ġw oven",
+      "ĠT MR",
+      "Ġfor gotten",
+      "ĠM ult",
+      "ĠB ipolar",
+      "ĠH isp",
+      "op eptides",
+      "ap amil",
+      "Ġro uted",
+      "Ġag n",
+      "Ġday light",
+      "ĠÍ Ķ",
+      "BB B",
+      "ĠMajor ity",
+      "Ġconfound ed",
+      "ĠCarol ine",
+      "ĠShim ura",
+      "r uction",
+      "Ġt ympan",
+      "ac io",
+      "ĠT FE",
+      "ĠT utorial",
+      "Ġal lyl",
+      "ĠF rost",
+      "ĠR PS",
+      "ĠW ah",
+      "Ġi y",
+      "Ġsub problems",
+      "Ġair foil",
+      "Ġembed s",
+      "ĠMor ning",
+      "Ġminor ities",
+      "ĠMemb ership",
+      "Ġquadric eps",
+      "y ly",
+      "ĠB odies",
+      "ĠR AND",
+      "Ġr ationally",
+      "ĠMan ifold",
+      "phosph ine",
+      "cons idering",
+      "ez vous",
+      "ĠKnow ing",
+      "Ġtumorigen ic",
+      "Ġillumin ating",
+      "ĠFernand es",
+      "polynomial s",
+      "ĠBulg arian",
+      "ĠBhattach arya",
+      "ospi ra",
+      "A pi",
+      "hen ne",
+      "Ġmay s",
+      "ĠAr gin",
+      "inter pol",
+      "ĠAnd ean",
+      "ĠPD S",
+      "ĠCN P",
+      "Ġtransf usions",
+      "ĠNan om",
+      "Ġsynerg ism",
+      "Ġbent onite",
+      "Ġgravit ons",
+      "aqu ette",
+      "Ġfiss ure",
+      "t andem",
+      "w ash",
+      "ĠE yes",
+      "Ġdi lepton",
+      "Ġrec tified",
+      "ĠAr ist",
+      "isc ible",
+      "Ġir q",
+      "Ġlig aments",
+      "sec urity",
+      "Ġvascular ization",
+      "Na Cl",
+      "ĠStra ight",
+      "ĠLept in",
+      "ĠAbund ances",
+      "ĠKE Y",
+      "ĠMother s",
+      "ĠRenew able",
+      "Ġmason ry",
+      "ë ı",
+      "rac eutical",
+      "Ġar ity",
+      "ĠAl ves",
+      "osp ectral",
+      "Ġimmun od",
+      "Ġmar ble",
+      "Ġcover ings",
+      "ĠConst ants",
+      "ĠRever sal",
+      "Work s",
+      "ĠNur se",
+      "ĠAggreg ate",
+      "ac illin",
+      "pl ug",
+      "Ġj ury",
+      "one ogenesis",
+      "Ġam oeb",
+      "au kee",
+      "Ġphosphor ic",
+      "ĠRem oving",
+      "Ġwors en",
+      "ĠESR D",
+      "ĠHern andez",
+      "ĠEug ene",
+      "visc osity",
+      "F ID",
+      "Â ¦",
+      "Ġ Ý",
+      "ĠS tig",
+      "ĠS ING",
+      "ĠI MRT",
+      "ĠB q",
+      "ĠD ME",
+      "ĠH OM",
+      "ph ysis",
+      "ob es",
+      "Ġsuper fields",
+      "Ġarg c",
+      "Ġmal adaptive",
+      "ĠEd iting",
+      "Ġcond em",
+      "ube i",
+      "stim ulus",
+      "Ġabbrevi ation",
+      "H aus",
+      "ĠN eeds",
+      "Ġad hering",
+      "ĠV PA",
+      "of rontal",
+      "ĠÅ ª",
+      "ĠVer de",
+      "ĠSl av",
+      "ĠProp ag",
+      "Ġcongen ers",
+      "Ġtil apia",
+      "ĠRac hel",
+      "L ess",
+      "Ġm asc",
+      "ent angled",
+      "ĠD TI",
+      "ati k",
+      "rol ases",
+      "ĠY en",
+      "arm or",
+      "ĠDec isions",
+      "Ġη p",
+      "Int uitively",
+      "ĠPharmaceutical s",
+      "J u",
+      "ud din",
+      "ĠW ASP",
+      "nt on",
+      "Ġbi ot",
+      "ĠZ NF",
+      "Ġcr ush",
+      "ĠPar ity",
+      "SI ST",
+      "EV ENT",
+      "ĠSqu amous",
+      "Stud ent",
+      "Ġgingival is",
+      "f used",
+      "ĠM ises",
+      "ĠF DTD",
+      "ore ceptors",
+      "Ġdisc retion",
+      "OR TC",
+      "MS P",
+      "Ġexpos es",
+      "Ġchlor inated",
+      "ĠUp regulation",
+      "ĠLim b",
+      "Ġantic or",
+      "Reg ular",
+      "Adv anced",
+      "X e",
+      "ag han",
+      "ĠB LA",
+      "Ġco asts",
+      "ĠTh irteen",
+      "hes in",
+      "ĠMet hanol",
+      "rot us",
+      "ĠStep hens",
+      "Bo ok",
+      "ĠHistor ically",
+      "ĠEmploy ing",
+      "Ġcorrug ated",
+      "mercapto ethanol",
+      "ĠD nmt",
+      "ĠQu eries",
+      "DR B",
+      "ĠGR U",
+      "FL U",
+      "ĠCarbon iferous",
+      "OB JECT",
+      "ĠDiscrim inative",
+      "ĠBurg ess",
+      "Ġplanetes imals",
+      "ĠAmend ment",
+      "ĠStriking ly",
+      "t ric",
+      "ec ure",
+      "Ġtrans posable",
+      "rol ateral",
+      "Ġhis ti",
+      "ega ard",
+      "Ġsk im",
+      "ĠSP F",
+      "Stat istics",
+      "Ġintest ines",
+      "f eng",
+      "l ain",
+      "Ġthe at",
+      "Ġo rogen",
+      "Ġp ill",
+      "od opa",
+      "Ġcorrel ative",
+      "AC O",
+      "Ġadj unction",
+      "ĠCare y",
+      "Ġtele portation",
+      "ĠBound aries",
+      "ĠGood fellow",
+      "ĠLind a",
+      "ĠPolymer ic",
+      "Ġexer tion",
+      "Ġsteep ly",
+      "Ġprotr usion",
+      "Ġhyal uronic",
+      "ĠRoc hester",
+      "ENSI ONAL",
+      "D ar",
+      "f et",
+      "ĠF SS",
+      "hem ically",
+      "Ġfl ashes",
+      "Ġdevi ated",
+      "feld t",
+      "Ġstic ks",
+      "Ġoct et",
+      "Ġgravitation ally",
+      "footnotes ize",
+      "L ex",
+      "o vi",
+      "Ġw ired",
+      "ĠS MP",
+      "erm ans",
+      "Ġun broken",
+      "Ġem ulation",
+      "sim ulated",
+      "Ġminim ality",
+      "ardi ac",
+      "Ġship w",
+      "Gene tic",
+      "ĠHerm ann",
+      "ynchron ization",
+      "ĠNap ole",
+      "Ġmonodis perse",
+      "R ho",
+      "r ists",
+      "Ġf x",
+      "ĠF UV",
+      "ĠG elfand",
+      "hem ispheric",
+      "ron idazole",
+      "Ġsuper saturation",
+      "oud h",
+      "oli tical",
+      "ĠAir y",
+      "Ġmanifest ly",
+      "ĠHM G",
+      "Ġadvertis ement",
+      "ĠBrook lyn",
+      "Ġparalle led",
+      "answ ered",
+      "ĠNotImplemented Error",
+      "U PD",
+      "o ust",
+      "ĠT eng",
+      "Ġfor tified",
+      "ĠS ort",
+      "EN E",
+      "ĠF ris",
+      "ĠH IS",
+      "ĠR OT",
+      "ĠN ested",
+      "pro duce",
+      "ĠK erala",
+      "gen omic",
+      "ĠIs ab",
+      "Ġur acil",
+      "bur ger",
+      "ĠLog arithmic",
+      "Ġster ility",
+      "Ġunem ployed",
+      "Ġori ental",
+      "K o",
+      "j ima",
+      "ĠC TP",
+      "ĠL AD",
+      "Ġconform ers",
+      "CG G",
+      "Per kin",
+      "Ġbrid ged",
+      "ĠDiss ociation",
+      "ĠQi agen",
+      "Ġwealth y",
+      "Ġanaest hetic",
+      "ĠMinim izing",
+      "Ġacous tics",
+      "buck et",
+      "ĠSert oli",
+      "r ath",
+      "s aw",
+      "Ġg arn",
+      "ĠThe oretically",
+      "tic ul",
+      "ĠTh inking",
+      "ik er",
+      "ĠCh it",
+      "Ġtr in",
+      "AL ITY",
+      "ĠFe O",
+      "Ġpolymer ized",
+      "En coding",
+      "Ġanalges ics",
+      "ĠLex ical",
+      "Ġmari juana",
+      "âĸĪ âĸĪ",
+      "c rops",
+      "ent ropic",
+      "ol ocation",
+      "ĠP omp",
+      "Ġco factors",
+      "box times",
+      "ĠAr ri",
+      "An gel",
+      "ĠRequire ment",
+      "Ġmicrol ensing",
+      "ĠTRAN SF",
+      "å º",
+      "Ġd ma",
+      "Ġre rio",
+      "und ancy",
+      "ant el",
+      "Ġradi ometric",
+      "ĠSe an",
+      "rand n",
+      "ĠCR L",
+      "hal os",
+      "uber tal",
+      "Ġquin one",
+      "T ES",
+      "Ġd W",
+      "ĠC GM",
+      "Ġhe aled",
+      "ian e",
+      "Ġobtain able",
+      "ĠAd rian",
+      "Ġlik es",
+      "ĠMed ication",
+      "Ġcogn itively",
+      "Whe ther",
+      "B ob",
+      "d id",
+      "al cohol",
+      "Ġn ivolumab",
+      "ĠF Y",
+      "Ġat resia",
+      "ach s",
+      "ĠK ip",
+      "Ġun igenes",
+      "ĠJ accard",
+      "ust ri",
+      "Ġconf ine",
+      "Ġaut ofluorescence",
+      "Ġpy g",
+      "Se a",
+      "Set tings",
+      "Ġtrunc atula",
+      "Liter al",
+      "F ab",
+      "M ah",
+      "V en",
+      "Ġt ig",
+      "Ġc her",
+      "ĠC CI",
+      "ĠF unk",
+      "ĠB ess",
+      "ĠN asal",
+      "iff er",
+      "Ġobs essive",
+      "ĠOp ening",
+      "ochond ral",
+      "ĠTR PA",
+      "ĠRab in",
+      "Ġta per",
+      "Ġdeaf ness",
+      "D OS",
+      "is ites",
+      "an ite",
+      "le ost",
+      "ĠS TP",
+      "ĠB ACE",
+      "ĠH enn",
+      "ĠE SM",
+      "Ġsuper field",
+      "ĠOr land",
+      "ĠAMP s",
+      "ĠHem orrh",
+      "Ġresc ues",
+      "Ġtour ists",
+      "ĠVL BI",
+      "Ġneighbourhood s",
+      "communic able",
+      "g x",
+      "r atase",
+      "ĠN RT",
+      "Ġob structions",
+      "Ġdef orestation",
+      "Ġq p",
+      "ĠPh an",
+      "ĠST I",
+      "iment o",
+      "ĠIR I",
+      "SV s",
+      "Ġstrip ed",
+      "Po inc",
+      "ĠBed ford",
+      "ĠFrag ment",
+      "ĠRelig ion",
+      "Ġd rones",
+      "im ulation",
+      "ĠC et",
+      "Ġg ills",
+      "ĠN orton",
+      "ib atch",
+      "est ructive",
+      "ĠJ av",
+      "ĠÏ ½",
+      "Ġmic a",
+      "AG B",
+      "RA W",
+      "ĠMy D",
+      "ct l",
+      "Ġrevers ibly",
+      "Ġsuppress ors",
+      "ĠFA IL",
+      "ĠFuk ushima",
+      "E vidence",
+      "p ink",
+      "as array",
+      "ĠT ann",
+      "Ġl oved",
+      "Ġbi ologists",
+      "Ġend othermic",
+      "Ġbro ker",
+      "ĠPer kins",
+      "Ġcategor ised",
+      "ĠSO ME",
+      "hydroxy vitamin",
+      "rog ates",
+      "ĠAge ing",
+      "Ġtourn aments",
+      "ĠStrom al",
+      "Ġdefer red",
+      "ĠSRE BP",
+      "M AD",
+      "S ay",
+      "c gi",
+      "p he",
+      "ol ini",
+      "ĠD ü",
+      "Ġde hydro",
+      "ap eptide",
+      "Ġhe s",
+      "Ġdist ally",
+      "vers ions",
+      "Ġmed als",
+      "Ġfl aw",
+      "Ġdu o",
+      "Ġimpair ing",
+      "toplas ts",
+      "ĠHF ILL",
+      "Ġesc ulent",
+      "Class ification",
+      "ĠGriff ith",
+      "ĠWelling ton",
+      "Ġattor ney",
+      "A st",
+      "k A",
+      "Ġm ilit",
+      "Ġn ite",
+      "ĠC asp",
+      "ĠC hester",
+      "ĠM ok",
+      "ĠR AR",
+      "Ġch r",
+      "unc tor",
+      "Ġab duction",
+      "Ġun iv",
+      "ov ars",
+      "ou k",
+      "ER ICAL",
+      "é ri",
+      "orb ance",
+      "ĠIdentif ies",
+      "ament o",
+      "Ġparent hesis",
+      "ĠME Fs",
+      "Ġabsor bs",
+      "ĠArray List",
+      "Ġcareg iving",
+      "ĠFI LE",
+      "Ġfeld spar",
+      "ochthon ous",
+      "S ort",
+      "j al",
+      "Ġt antal",
+      "ar abine",
+      "ĠS aid",
+      "ĠB CE",
+      "ĠN GO",
+      "yn ure",
+      "dot eq",
+      "ĠLe yd",
+      "mod ality",
+      "ĠGe ometrical",
+      "Al most",
+      "Ġhard ened",
+      "no ea",
+      "new s",
+      "Ġclean up",
+      "ĠArm ed",
+      "ĠSn ake",
+      "multi ply",
+      "ĠMill ennium",
+      "ĠSmooth ing",
+      "posit ely",
+      "en ary",
+      "is se",
+      "ĠY uc",
+      "Ġgene al",
+      "Ġsuper s",
+      "Ġhand held",
+      "Ġemb ark",
+      "ĠBl a",
+      "hor st",
+      "ĠPD GFR",
+      "Ġcit r",
+      "Ġcalor ie",
+      "ĠBudd hist",
+      "M ember",
+      "Ġf ears",
+      "Ġf iscal",
+      "ĠA IF",
+      "LO AD",
+      "pe are",
+      "Ġbit umen",
+      "Par ticip",
+      "ĠIndian apolis",
+      "ĠAlb um",
+      "Ġscr utiny",
+      "acyl glycer",
+      "ĠSak ai",
+      "Ġthermod ynamical",
+      "Z B",
+      "Ġh pf",
+      "ĠL IP",
+      "Ġexp iration",
+      "til t",
+      "Ġfl ax",
+      "ĠSe lectivity",
+      "ĠSch ol",
+      "any a",
+      "orb ents",
+      "Ġincub ations",
+      "Ġmargin als",
+      "inv olved",
+      "Ġenthal pies",
+      "macroph ages",
+      "construct or",
+      "ĠRol and",
+      "ĠP m",
+      "ĠR Y",
+      "Ġbl obs",
+      "Ġann uli",
+      "Ġuns timulated",
+      "ĠPet roleum",
+      "Ġmerg es",
+      "Ġenvelop ing",
+      "ĠInitial ization",
+      "Ġshed s",
+      "Ġadvis able",
+      "ylethanol amine"
+    ]
+  }
+}
diff --git a/ppocr/utils/dict/unimernet_tokenizer/tokenizer_config.json b/ppocr/utils/dict/unimernet_tokenizer/tokenizer_config.json
new file mode 100644
index 0000000..75dad5a
--- /dev/null
+++ b/ppocr/utils/dict/unimernet_tokenizer/tokenizer_config.json
@@ -0,0 +1,205 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<pad>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "4": {
+      "content": "[START_REF]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "5": {
+      "content": "[END_REF]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "6": {
+      "content": "[IMAGE]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "7": {
+      "content": "<fragments>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "8": {
+      "content": "</fragments>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "9": {
+      "content": "<work>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "10": {
+      "content": "</work>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "11": {
+      "content": "[START_SUP]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "12": {
+      "content": "[END_SUP]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "13": {
+      "content": "[START_SUB]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "14": {
+      "content": "[END_SUB]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "15": {
+      "content": "[START_DNA]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "16": {
+      "content": "[END_DNA]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "17": {
+      "content": "[START_AMINO]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "18": {
+      "content": "[END_AMINO]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "19": {
+      "content": "[START_SMILES]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "20": {
+      "content": "[END_SMILES]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "21": {
+      "content": "[START_I_SMILES]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "22": {
+      "content": "[END_I_SMILES]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [],
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "max_length": 4096,
+  "model_max_length": 768,
+  "pad_to_multiple_of": null,
+  "pad_token": "<pad>",
+  "pad_token_type_id": 0,
+  "padding_side": "right",
+  "processor_class": "VariableDonutProcessor",
+  "stride": 0,
+  "tokenizer_class": "NougatTokenizer",
+  "truncation_side": "right",
+  "truncation_strategy": "longest_first",
+  "unk_token": "<unk>",
+  "vocab_file": null
+}
diff --git a/ppocr/utils/dict/ur_dict.txt b/ppocr/utils/dict/ur_dict.txt
new file mode 100644
index 0000000..c06786a
--- /dev/null
+++ b/ppocr/utils/dict/ur_dict.txt
@@ -0,0 +1,137 @@
+u
+r
+_
+i
+m
+g
+/
+3
+I
+L
+S
+V
+R
+C
+2
+0
+1
+v
+a
+l
+9
+7
+8
+.
+j
+p
+	
+چ
+ٹ
+پ
+ا
+ئ
+ی
+ے
+4
+6
+و
+ل
+ن
+ڈ
+ھ
+ک
+ت
+ش
+ف
+ق
+ر
+د
+5
+ب
+ج
+خ
+ہ
+س
+ز
+غ
+ڑ
+ں
+آ
+م
+ؤ
+ط
+ص
+ح
+ع
+گ
+ث
+ض
+ذ
+ۓ
+ِ
+ء
+ظ
+ً
+ي
+ُ
+ۃ
+أ
+ٰ
+ە
+ژ
+ۂ
+ة
+ّ
+ك
+ه
+s
+c
+e
+n
+w
+o
+d
+t
+D
+M
+T
+U
+E
+b
+P
+h
+y
+W
+H
+A
+x
+B
+O
+N
+G
+Y
+Q
+F
+k
+K
+q
+J
+Z
+f
+z
+X
+'
+@
+&
+!
+,
+:
+$
+-
+#
+?
+%
+é
++
+(
+É
diff --git a/ppocr/utils/dict/vi_dict.txt b/ppocr/utils/dict/vi_dict.txt
new file mode 100644
index 0000000..5e8f8c6
--- /dev/null
+++ b/ppocr/utils/dict/vi_dict.txt
@@ -0,0 +1,113 @@
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+P
+Q
+R
+S
+T
+U
+V
+W
+a
+b
+c
+d
+e
+g
+h
+i
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+à
+á
+â
+ã
+è
+é
+ê
+ì
+í
+ò
+ó
+ô
+õ
+ù
+ú
+ý
+ă
+Đ
+đ
+ĩ
+ũ
+ơ
+ư
+ạ
+ả
+ấ
+ầ
+ẩ
+ẫ
+ậ
+ắ
+ằ
+ẳ
+ẵ
+ặ
+ẹ
+ẻ
+ẽ
+ế
+ề
+ể
+ễ
+ệ
+ỉ
+ị
+ọ
+ỏ
+ố
+ồ
+ổ
+ỗ
+ộ
+ớ
+ờ
+ở
+ỡ
+ợ
+ụ
+ủ
+ứ
+ừ
+ử
+ữ
+ự
+ỳ
+ỵ
+ỷ
+ỹ
diff --git a/ppocr/utils/dict/xi_dict.txt b/ppocr/utils/dict/xi_dict.txt
new file mode 100644
index 0000000..f195f1e
--- /dev/null
+++ b/ppocr/utils/dict/xi_dict.txt
@@ -0,0 +1,110 @@
+x
+i
+_
+m
+g
+/
+1
+0
+I
+L
+S
+V
+R
+C
+2
+v
+a
+l
+3
+6
+4
+5
+.
+j
+p
+	
+Q
+u
+e
+r
+o
+8
+7
+n
+c
+9
+t
+b
+é
+q
+d
+ó
+y
+F
+s
+,
+O
+í
+T
+f
+"
+U
+M
+h
+:
+P
+H
+A
+E
+D
+z
+N
+á
+ñ
+ú
+%
+;
+è
++
+Y
+-
+B
+G
+(
+)
+¿
+?
+w
+¡
+!
+X
+É
+K
+k
+Á
+ü
+Ú
+«
+»
+J
+'
+ö
+W
+Z
+º
+Ö
+­
+[
+]
+Ç
+ç
+à
+ä
+û
+ò
+Í
+ê
+ô
+ø
+ª
diff --git a/ppocr/utils/dict90.txt b/ppocr/utils/dict90.txt
new file mode 100644
index 0000000..46b8795
--- /dev/null
+++ b/ppocr/utils/dict90.txt
@@ -0,0 +1,90 @@
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+:
+;
+<
+=
+>
+?
+@
+[
+\
+]
+_
+`
+~
diff --git a/ppocr/utils/e2e_metric/Deteval.py b/ppocr/utils/e2e_metric/Deteval.py
new file mode 100755
index 0000000..f4767e1
--- /dev/null
+++ b/ppocr/utils/e2e_metric/Deteval.py
@@ -0,0 +1,852 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import json
+import numpy as np
+import scipy.io as io
+
+from ppocr.utils.utility import check_install
+
+from ppocr.utils.e2e_metric.polygon_fast import iod, area_of_intersection, area
+
+
+def get_socre_A(gt_dir, pred_dict):
+    allInputs = 1
+
+    def input_reading_mod(pred_dict):
+        """This helper reads input from txt files"""
+        det = []
+        n = len(pred_dict)
+        for i in range(n):
+            points = pred_dict[i]["points"]
+            text = pred_dict[i]["texts"]
+            point = ",".join(
+                map(
+                    str,
+                    points.reshape(
+                        -1,
+                    ),
+                )
+            )
+            det.append([point, text])
+        return det
+
+    def gt_reading_mod(gt_dict):
+        """This helper reads groundtruths from mat files"""
+        gt = []
+        n = len(gt_dict)
+        for i in range(n):
+            points = gt_dict[i]["points"].tolist()
+            h = len(points)
+            text = gt_dict[i]["text"]
+            xx = [
+                np.array(["x:"], dtype="<U2"),
+                0,
+                np.array(["y:"], dtype="<U2"),
+                0,
+                np.array(["#"], dtype="<U1"),
+                np.array(["#"], dtype="<U1"),
+            ]
+            t_x, t_y = [], []
+            for j in range(h):
+                t_x.append(points[j][0])
+                t_y.append(points[j][1])
+            xx[1] = np.array([t_x], dtype="int16")
+            xx[3] = np.array([t_y], dtype="int16")
+            if text != "":
+                xx[4] = np.array([text], dtype="U{}".format(len(text)))
+                xx[5] = np.array(["c"], dtype="<U1")
+            gt.append(xx)
+        return gt
+
+    def detection_filtering(detections, groundtruths, threshold=0.5):
+        for gt_id, gt in enumerate(groundtruths):
+            if (gt[5] == "#") and (gt[1].shape[1] > 1):
+                gt_x = list(map(int, np.squeeze(gt[1])))
+                gt_y = list(map(int, np.squeeze(gt[3])))
+                for det_id, detection in enumerate(detections):
+                    detection_orig = detection
+                    detection = [float(x) for x in detection[0].split(",")]
+                    detection = list(map(int, detection))
+                    det_x = detection[0::2]
+                    det_y = detection[1::2]
+                    det_gt_iou = iod(det_x, det_y, gt_x, gt_y)
+                    if det_gt_iou > threshold:
+                        detections[det_id] = []
+
+                detections[:] = [item for item in detections if item != []]
+        return detections
+
+    def sigma_calculation(det_x, det_y, gt_x, gt_y):
+        """
+        sigma = inter_area / gt_area
+        """
+        return np.round(
+            (area_of_intersection(det_x, det_y, gt_x, gt_y) / area(gt_x, gt_y)), 2
+        )
+
+    def tau_calculation(det_x, det_y, gt_x, gt_y):
+        if area(det_x, det_y) == 0.0:
+            return 0
+        return np.round(
+            (area_of_intersection(det_x, det_y, gt_x, gt_y) / area(det_x, det_y)), 2
+        )
+
+    ##############################Initialization###################################
+    # global_sigma = []
+    # global_tau = []
+    # global_pred_str = []
+    # global_gt_str = []
+    ###############################################################################
+
+    for input_id in range(allInputs):
+        if (
+            (input_id != ".DS_Store")
+            and (input_id != "Pascal_result.txt")
+            and (input_id != "Pascal_result_curved.txt")
+            and (input_id != "Pascal_result_non_curved.txt")
+            and (input_id != "Deteval_result.txt")
+            and (input_id != "Deteval_result_curved.txt")
+            and (input_id != "Deteval_result_non_curved.txt")
+        ):
+            detections = input_reading_mod(pred_dict)
+            groundtruths = gt_reading_mod(gt_dir)
+            detections = detection_filtering(
+                detections, groundtruths
+            )  # filters detections overlapping with DC area
+            dc_id = []
+            for i in range(len(groundtruths)):
+                if groundtruths[i][5] == "#":
+                    dc_id.append(i)
+            cnt = 0
+            for a in dc_id:
+                num = a - cnt
+                del groundtruths[num]
+                cnt += 1
+
+            local_sigma_table = np.zeros((len(groundtruths), len(detections)))
+            local_tau_table = np.zeros((len(groundtruths), len(detections)))
+            local_pred_str = {}
+            local_gt_str = {}
+
+            for gt_id, gt in enumerate(groundtruths):
+                if len(detections) > 0:
+                    for det_id, detection in enumerate(detections):
+                        detection_orig = detection
+                        detection = [float(x) for x in detection[0].split(",")]
+                        detection = list(map(int, detection))
+                        pred_seq_str = detection_orig[1].strip()
+                        det_x = detection[0::2]
+                        det_y = detection[1::2]
+                        gt_x = list(map(int, np.squeeze(gt[1])))
+                        gt_y = list(map(int, np.squeeze(gt[3])))
+                        gt_seq_str = str(gt[4].tolist()[0])
+
+                        local_sigma_table[gt_id, det_id] = sigma_calculation(
+                            det_x, det_y, gt_x, gt_y
+                        )
+                        local_tau_table[gt_id, det_id] = tau_calculation(
+                            det_x, det_y, gt_x, gt_y
+                        )
+                        local_pred_str[det_id] = pred_seq_str
+                        local_gt_str[gt_id] = gt_seq_str
+
+            global_sigma = local_sigma_table
+            global_tau = local_tau_table
+            global_pred_str = local_pred_str
+            global_gt_str = local_gt_str
+
+    single_data = {}
+    single_data["sigma"] = global_sigma
+    single_data["global_tau"] = global_tau
+    single_data["global_pred_str"] = global_pred_str
+    single_data["global_gt_str"] = global_gt_str
+    return single_data
+
+
+def get_socre_B(gt_dir, img_id, pred_dict):
+    allInputs = 1
+
+    def input_reading_mod(pred_dict):
+        """This helper reads input from txt files"""
+        det = []
+        n = len(pred_dict)
+        for i in range(n):
+            points = pred_dict[i]["points"]
+            text = pred_dict[i]["texts"]
+            point = ",".join(
+                map(
+                    str,
+                    points.reshape(
+                        -1,
+                    ),
+                )
+            )
+            det.append([point, text])
+        return det
+
+    def gt_reading_mod(gt_dir, gt_id):
+        gt = io.loadmat("%s/poly_gt_img%s.mat" % (gt_dir, gt_id))
+        gt = gt["polygt"]
+        return gt
+
+    def detection_filtering(detections, groundtruths, threshold=0.5):
+        for gt_id, gt in enumerate(groundtruths):
+            if (gt[5] == "#") and (gt[1].shape[1] > 1):
+                gt_x = list(map(int, np.squeeze(gt[1])))
+                gt_y = list(map(int, np.squeeze(gt[3])))
+                for det_id, detection in enumerate(detections):
+                    detection_orig = detection
+                    detection = [float(x) for x in detection[0].split(",")]
+                    detection = list(map(int, detection))
+                    det_x = detection[0::2]
+                    det_y = detection[1::2]
+                    det_gt_iou = iod(det_x, det_y, gt_x, gt_y)
+                    if det_gt_iou > threshold:
+                        detections[det_id] = []
+
+                detections[:] = [item for item in detections if item != []]
+        return detections
+
+    def sigma_calculation(det_x, det_y, gt_x, gt_y):
+        """
+        sigma = inter_area / gt_area
+        """
+        return np.round(
+            (area_of_intersection(det_x, det_y, gt_x, gt_y) / area(gt_x, gt_y)), 2
+        )
+
+    def tau_calculation(det_x, det_y, gt_x, gt_y):
+        if area(det_x, det_y) == 0.0:
+            return 0
+        return np.round(
+            (area_of_intersection(det_x, det_y, gt_x, gt_y) / area(det_x, det_y)), 2
+        )
+
+    ##############################Initialization###################################
+    # global_sigma = []
+    # global_tau = []
+    # global_pred_str = []
+    # global_gt_str = []
+    ###############################################################################
+
+    for input_id in range(allInputs):
+        if (
+            (input_id != ".DS_Store")
+            and (input_id != "Pascal_result.txt")
+            and (input_id != "Pascal_result_curved.txt")
+            and (input_id != "Pascal_result_non_curved.txt")
+            and (input_id != "Deteval_result.txt")
+            and (input_id != "Deteval_result_curved.txt")
+            and (input_id != "Deteval_result_non_curved.txt")
+        ):
+            detections = input_reading_mod(pred_dict)
+            groundtruths = gt_reading_mod(gt_dir, img_id).tolist()
+            detections = detection_filtering(
+                detections, groundtruths
+            )  # filters detections overlapping with DC area
+            dc_id = []
+            for i in range(len(groundtruths)):
+                if groundtruths[i][5] == "#":
+                    dc_id.append(i)
+            cnt = 0
+            for a in dc_id:
+                num = a - cnt
+                del groundtruths[num]
+                cnt += 1
+
+            local_sigma_table = np.zeros((len(groundtruths), len(detections)))
+            local_tau_table = np.zeros((len(groundtruths), len(detections)))
+            local_pred_str = {}
+            local_gt_str = {}
+
+            for gt_id, gt in enumerate(groundtruths):
+                if len(detections) > 0:
+                    for det_id, detection in enumerate(detections):
+                        detection_orig = detection
+                        detection = [float(x) for x in detection[0].split(",")]
+                        detection = list(map(int, detection))
+                        pred_seq_str = detection_orig[1].strip()
+                        det_x = detection[0::2]
+                        det_y = detection[1::2]
+                        gt_x = list(map(int, np.squeeze(gt[1])))
+                        gt_y = list(map(int, np.squeeze(gt[3])))
+                        gt_seq_str = str(gt[4].tolist()[0])
+
+                        local_sigma_table[gt_id, det_id] = sigma_calculation(
+                            det_x, det_y, gt_x, gt_y
+                        )
+                        local_tau_table[gt_id, det_id] = tau_calculation(
+                            det_x, det_y, gt_x, gt_y
+                        )
+                        local_pred_str[det_id] = pred_seq_str
+                        local_gt_str[gt_id] = gt_seq_str
+
+            global_sigma = local_sigma_table
+            global_tau = local_tau_table
+            global_pred_str = local_pred_str
+            global_gt_str = local_gt_str
+
+    single_data = {}
+    single_data["sigma"] = global_sigma
+    single_data["global_tau"] = global_tau
+    single_data["global_pred_str"] = global_pred_str
+    single_data["global_gt_str"] = global_gt_str
+    return single_data
+
+
+def get_score_C(gt_label, text, pred_bboxes):
+    """
+    get score for CentripetalText (CT) prediction.
+    """
+    check_install("Polygon", "Polygon3")
+    import Polygon as plg
+
+    def gt_reading_mod(gt_label, text):
+        """This helper reads groundtruths from mat files"""
+        groundtruths = []
+        nbox = len(gt_label)
+        for i in range(nbox):
+            label = {"transcription": text[i][0], "points": gt_label[i].numpy()}
+            groundtruths.append(label)
+
+        return groundtruths
+
+    def get_union(pD, pG):
+        areaA = pD.area()
+        areaB = pG.area()
+        return areaA + areaB - get_intersection(pD, pG)
+
+    def get_intersection(pD, pG):
+        pInt = pD & pG
+        if len(pInt) == 0:
+            return 0
+        return pInt.area()
+
+    def detection_filtering(detections, groundtruths, threshold=0.5):
+        for gt in groundtruths:
+            point_num = gt["points"].shape[1] // 2
+            if gt["transcription"] == "###" and (point_num > 1):
+                gt_p = np.array(gt["points"]).reshape(point_num, 2).astype("int32")
+                gt_p = plg.Polygon(gt_p)
+
+                for det_id, detection in enumerate(detections):
+                    det_y = detection[0::2]
+                    det_x = detection[1::2]
+
+                    det_p = np.concatenate((np.array(det_x), np.array(det_y)))
+                    det_p = det_p.reshape(2, -1).transpose()
+                    det_p = plg.Polygon(det_p)
+
+                    try:
+                        det_gt_iou = get_intersection(det_p, gt_p) / det_p.area()
+                    except:
+                        print(det_x, det_y, gt_p)
+                    if det_gt_iou > threshold:
+                        detections[det_id] = []
+
+                detections[:] = [item for item in detections if item != []]
+        return detections
+
+    def sigma_calculation(det_p, gt_p):
+        """
+        sigma = inter_area / gt_area
+        """
+        if gt_p.area() == 0.0:
+            return 0
+        return get_intersection(det_p, gt_p) / gt_p.area()
+
+    def tau_calculation(det_p, gt_p):
+        """
+        tau = inter_area / det_area
+        """
+        if det_p.area() == 0.0:
+            return 0
+        return get_intersection(det_p, gt_p) / det_p.area()
+
+    detections = []
+
+    for item in pred_bboxes:
+        detections.append(item[:, ::-1].reshape(-1))
+
+    groundtruths = gt_reading_mod(gt_label, text)
+
+    detections = detection_filtering(
+        detections, groundtruths
+    )  # filters detections overlapping with DC area
+
+    for idx in range(len(groundtruths) - 1, -1, -1):
+        # NOTE: source code use 'orin' to indicate '#', here we use 'anno',
+        # which may cause slight drop in fscore, about 0.12
+        if groundtruths[idx]["transcription"] == "###":
+            groundtruths.pop(idx)
+
+    local_sigma_table = np.zeros((len(groundtruths), len(detections)))
+    local_tau_table = np.zeros((len(groundtruths), len(detections)))
+
+    for gt_id, gt in enumerate(groundtruths):
+        if len(detections) > 0:
+            for det_id, detection in enumerate(detections):
+                point_num = gt["points"].shape[1] // 2
+
+                gt_p = np.array(gt["points"]).reshape(point_num, 2).astype("int32")
+                gt_p = plg.Polygon(gt_p)
+
+                det_y = detection[0::2]
+                det_x = detection[1::2]
+
+                det_p = np.concatenate((np.array(det_x), np.array(det_y)))
+
+                det_p = det_p.reshape(2, -1).transpose()
+                det_p = plg.Polygon(det_p)
+
+                local_sigma_table[gt_id, det_id] = sigma_calculation(det_p, gt_p)
+                local_tau_table[gt_id, det_id] = tau_calculation(det_p, gt_p)
+
+    data = {}
+    data["sigma"] = local_sigma_table
+    data["global_tau"] = local_tau_table
+    data["global_pred_str"] = ""
+    data["global_gt_str"] = ""
+    return data
+
+
+def combine_results(all_data, rec_flag=True):
+    tr = 0.7
+    tp = 0.6
+    fsc_k = 0.8
+    k = 2
+    global_sigma = []
+    global_tau = []
+    global_pred_str = []
+    global_gt_str = []
+
+    for data in all_data:
+        global_sigma.append(data["sigma"])
+        global_tau.append(data["global_tau"])
+        global_pred_str.append(data["global_pred_str"])
+        global_gt_str.append(data["global_gt_str"])
+
+    global_accumulative_recall = 0
+    global_accumulative_precision = 0
+    total_num_gt = 0
+    total_num_det = 0
+    hit_str_count = 0
+    hit_count = 0
+
+    def one_to_one(
+        local_sigma_table,
+        local_tau_table,
+        local_accumulative_recall,
+        local_accumulative_precision,
+        global_accumulative_recall,
+        global_accumulative_precision,
+        gt_flag,
+        det_flag,
+        idy,
+        rec_flag,
+    ):
+        hit_str_num = 0
+        for gt_id in range(num_gt):
+            gt_matching_qualified_sigma_candidates = np.where(
+                local_sigma_table[gt_id, :] > tr
+            )
+            gt_matching_num_qualified_sigma_candidates = (
+                gt_matching_qualified_sigma_candidates[0].shape[0]
+            )
+            gt_matching_qualified_tau_candidates = np.where(
+                local_tau_table[gt_id, :] > tp
+            )
+            gt_matching_num_qualified_tau_candidates = (
+                gt_matching_qualified_tau_candidates[0].shape[0]
+            )
+
+            det_matching_qualified_sigma_candidates = np.where(
+                local_sigma_table[:, gt_matching_qualified_sigma_candidates[0]] > tr
+            )
+            det_matching_num_qualified_sigma_candidates = (
+                det_matching_qualified_sigma_candidates[0].shape[0]
+            )
+            det_matching_qualified_tau_candidates = np.where(
+                local_tau_table[:, gt_matching_qualified_tau_candidates[0]] > tp
+            )
+            det_matching_num_qualified_tau_candidates = (
+                det_matching_qualified_tau_candidates[0].shape[0]
+            )
+
+            if (
+                (gt_matching_num_qualified_sigma_candidates == 1)
+                and (gt_matching_num_qualified_tau_candidates == 1)
+                and (det_matching_num_qualified_sigma_candidates == 1)
+                and (det_matching_num_qualified_tau_candidates == 1)
+            ):
+                global_accumulative_recall = global_accumulative_recall + 1.0
+                global_accumulative_precision = global_accumulative_precision + 1.0
+                local_accumulative_recall = local_accumulative_recall + 1.0
+                local_accumulative_precision = local_accumulative_precision + 1.0
+
+                gt_flag[0, gt_id] = 1
+                matched_det_id = np.where(local_sigma_table[gt_id, :] > tr)
+                # recg start
+                if rec_flag:
+                    gt_str_cur = global_gt_str[idy][gt_id]
+                    pred_str_cur = global_pred_str[idy][matched_det_id[0].tolist()[0]]
+                    if pred_str_cur == gt_str_cur:
+                        hit_str_num += 1
+                    else:
+                        if pred_str_cur.lower() == gt_str_cur.lower():
+                            hit_str_num += 1
+                # recg end
+                det_flag[0, matched_det_id] = 1
+        return (
+            local_accumulative_recall,
+            local_accumulative_precision,
+            global_accumulative_recall,
+            global_accumulative_precision,
+            gt_flag,
+            det_flag,
+            hit_str_num,
+        )
+
+    def one_to_many(
+        local_sigma_table,
+        local_tau_table,
+        local_accumulative_recall,
+        local_accumulative_precision,
+        global_accumulative_recall,
+        global_accumulative_precision,
+        gt_flag,
+        det_flag,
+        idy,
+        rec_flag,
+    ):
+        hit_str_num = 0
+        for gt_id in range(num_gt):
+            # skip the following if the groundtruth was matched
+            if gt_flag[0, gt_id] > 0:
+                continue
+
+            non_zero_in_sigma = np.where(local_sigma_table[gt_id, :] > 0)
+            num_non_zero_in_sigma = non_zero_in_sigma[0].shape[0]
+
+            if num_non_zero_in_sigma >= k:
+                ####search for all detections that overlaps with this groundtruth
+                qualified_tau_candidates = np.where(
+                    (local_tau_table[gt_id, :] >= tp) & (det_flag[0, :] == 0)
+                )
+                num_qualified_tau_candidates = qualified_tau_candidates[0].shape[0]
+
+                if num_qualified_tau_candidates == 1:
+                    if (local_tau_table[gt_id, qualified_tau_candidates] >= tp) and (
+                        local_sigma_table[gt_id, qualified_tau_candidates] >= tr
+                    ):
+                        # became an one-to-one case
+                        global_accumulative_recall = global_accumulative_recall + 1.0
+                        global_accumulative_precision = (
+                            global_accumulative_precision + 1.0
+                        )
+                        local_accumulative_recall = local_accumulative_recall + 1.0
+                        local_accumulative_precision = (
+                            local_accumulative_precision + 1.0
+                        )
+
+                        gt_flag[0, gt_id] = 1
+                        det_flag[0, qualified_tau_candidates] = 1
+                        # recg start
+                        if rec_flag:
+                            gt_str_cur = global_gt_str[idy][gt_id]
+                            pred_str_cur = global_pred_str[idy][
+                                qualified_tau_candidates[0].tolist()[0]
+                            ]
+                            if pred_str_cur == gt_str_cur:
+                                hit_str_num += 1
+                            else:
+                                if pred_str_cur.lower() == gt_str_cur.lower():
+                                    hit_str_num += 1
+                        # recg end
+                elif np.sum(local_sigma_table[gt_id, qualified_tau_candidates]) >= tr:
+                    gt_flag[0, gt_id] = 1
+                    det_flag[0, qualified_tau_candidates] = 1
+                    # recg start
+                    if rec_flag:
+                        gt_str_cur = global_gt_str[idy][gt_id]
+                        pred_str_cur = global_pred_str[idy][
+                            qualified_tau_candidates[0].tolist()[0]
+                        ]
+                        if pred_str_cur == gt_str_cur:
+                            hit_str_num += 1
+                        else:
+                            if pred_str_cur.lower() == gt_str_cur.lower():
+                                hit_str_num += 1
+                    # recg end
+
+                    global_accumulative_recall = global_accumulative_recall + fsc_k
+                    global_accumulative_precision = (
+                        global_accumulative_precision
+                        + num_qualified_tau_candidates * fsc_k
+                    )
+
+                    local_accumulative_recall = local_accumulative_recall + fsc_k
+                    local_accumulative_precision = (
+                        local_accumulative_precision
+                        + num_qualified_tau_candidates * fsc_k
+                    )
+
+        return (
+            local_accumulative_recall,
+            local_accumulative_precision,
+            global_accumulative_recall,
+            global_accumulative_precision,
+            gt_flag,
+            det_flag,
+            hit_str_num,
+        )
+
+    def many_to_one(
+        local_sigma_table,
+        local_tau_table,
+        local_accumulative_recall,
+        local_accumulative_precision,
+        global_accumulative_recall,
+        global_accumulative_precision,
+        gt_flag,
+        det_flag,
+        idy,
+        rec_flag,
+    ):
+        hit_str_num = 0
+        for det_id in range(num_det):
+            # skip the following if the detection was matched
+            if det_flag[0, det_id] > 0:
+                continue
+
+            non_zero_in_tau = np.where(local_tau_table[:, det_id] > 0)
+            num_non_zero_in_tau = non_zero_in_tau[0].shape[0]
+
+            if num_non_zero_in_tau >= k:
+                ####search for all detections that overlaps with this groundtruth
+                qualified_sigma_candidates = np.where(
+                    (local_sigma_table[:, det_id] >= tp) & (gt_flag[0, :] == 0)
+                )
+                num_qualified_sigma_candidates = qualified_sigma_candidates[0].shape[0]
+
+                if num_qualified_sigma_candidates == 1:
+                    if (local_tau_table[qualified_sigma_candidates, det_id] >= tp) and (
+                        local_sigma_table[qualified_sigma_candidates, det_id] >= tr
+                    ):
+                        # became an one-to-one case
+                        global_accumulative_recall = global_accumulative_recall + 1.0
+                        global_accumulative_precision = (
+                            global_accumulative_precision + 1.0
+                        )
+                        local_accumulative_recall = local_accumulative_recall + 1.0
+                        local_accumulative_precision = (
+                            local_accumulative_precision + 1.0
+                        )
+
+                        gt_flag[0, qualified_sigma_candidates] = 1
+                        det_flag[0, det_id] = 1
+                        # recg start
+                        if rec_flag:
+                            pred_str_cur = global_pred_str[idy][det_id]
+                            gt_len = len(qualified_sigma_candidates[0])
+                            for idx in range(gt_len):
+                                ele_gt_id = qualified_sigma_candidates[0].tolist()[idx]
+                                if ele_gt_id not in global_gt_str[idy]:
+                                    continue
+                                gt_str_cur = global_gt_str[idy][ele_gt_id]
+                                if pred_str_cur == gt_str_cur:
+                                    hit_str_num += 1
+                                    break
+                                else:
+                                    if pred_str_cur.lower() == gt_str_cur.lower():
+                                        hit_str_num += 1
+                                    break
+                        # recg end
+                elif np.sum(local_tau_table[qualified_sigma_candidates, det_id]) >= tp:
+                    det_flag[0, det_id] = 1
+                    gt_flag[0, qualified_sigma_candidates] = 1
+                    # recg start
+                    if rec_flag:
+                        pred_str_cur = global_pred_str[idy][det_id]
+                        gt_len = len(qualified_sigma_candidates[0])
+                        for idx in range(gt_len):
+                            ele_gt_id = qualified_sigma_candidates[0].tolist()[idx]
+                            if ele_gt_id not in global_gt_str[idy]:
+                                continue
+                            gt_str_cur = global_gt_str[idy][ele_gt_id]
+                            if pred_str_cur == gt_str_cur:
+                                hit_str_num += 1
+                                break
+                            else:
+                                if pred_str_cur.lower() == gt_str_cur.lower():
+                                    hit_str_num += 1
+                                    break
+                    # recg end
+
+                    global_accumulative_recall = (
+                        global_accumulative_recall
+                        + num_qualified_sigma_candidates * fsc_k
+                    )
+                    global_accumulative_precision = (
+                        global_accumulative_precision + fsc_k
+                    )
+
+                    local_accumulative_recall = (
+                        local_accumulative_recall
+                        + num_qualified_sigma_candidates * fsc_k
+                    )
+                    local_accumulative_precision = local_accumulative_precision + fsc_k
+        return (
+            local_accumulative_recall,
+            local_accumulative_precision,
+            global_accumulative_recall,
+            global_accumulative_precision,
+            gt_flag,
+            det_flag,
+            hit_str_num,
+        )
+
+    for idx in range(len(global_sigma)):
+        local_sigma_table = np.array(global_sigma[idx])
+        local_tau_table = global_tau[idx]
+
+        num_gt = local_sigma_table.shape[0]
+        num_det = local_sigma_table.shape[1]
+
+        total_num_gt = total_num_gt + num_gt
+        total_num_det = total_num_det + num_det
+
+        local_accumulative_recall = 0
+        local_accumulative_precision = 0
+        gt_flag = np.zeros((1, num_gt))
+        det_flag = np.zeros((1, num_det))
+
+        #######first check for one-to-one case##########
+        (
+            local_accumulative_recall,
+            local_accumulative_precision,
+            global_accumulative_recall,
+            global_accumulative_precision,
+            gt_flag,
+            det_flag,
+            hit_str_num,
+        ) = one_to_one(
+            local_sigma_table,
+            local_tau_table,
+            local_accumulative_recall,
+            local_accumulative_precision,
+            global_accumulative_recall,
+            global_accumulative_precision,
+            gt_flag,
+            det_flag,
+            idx,
+            rec_flag,
+        )
+
+        hit_str_count += hit_str_num
+        #######then check for one-to-many case##########
+        (
+            local_accumulative_recall,
+            local_accumulative_precision,
+            global_accumulative_recall,
+            global_accumulative_precision,
+            gt_flag,
+            det_flag,
+            hit_str_num,
+        ) = one_to_many(
+            local_sigma_table,
+            local_tau_table,
+            local_accumulative_recall,
+            local_accumulative_precision,
+            global_accumulative_recall,
+            global_accumulative_precision,
+            gt_flag,
+            det_flag,
+            idx,
+            rec_flag,
+        )
+        hit_str_count += hit_str_num
+        #######then check for many-to-one case##########
+        (
+            local_accumulative_recall,
+            local_accumulative_precision,
+            global_accumulative_recall,
+            global_accumulative_precision,
+            gt_flag,
+            det_flag,
+            hit_str_num,
+        ) = many_to_one(
+            local_sigma_table,
+            local_tau_table,
+            local_accumulative_recall,
+            local_accumulative_precision,
+            global_accumulative_recall,
+            global_accumulative_precision,
+            gt_flag,
+            det_flag,
+            idx,
+            rec_flag,
+        )
+        hit_str_count += hit_str_num
+
+    try:
+        recall = global_accumulative_recall / total_num_gt
+    except ZeroDivisionError:
+        recall = 0
+
+    try:
+        precision = global_accumulative_precision / total_num_det
+    except ZeroDivisionError:
+        precision = 0
+
+    try:
+        f_score = 2 * precision * recall / (precision + recall)
+    except ZeroDivisionError:
+        f_score = 0
+
+    try:
+        seqerr = 1 - float(hit_str_count) / global_accumulative_recall
+    except ZeroDivisionError:
+        seqerr = 1
+
+    try:
+        recall_e2e = float(hit_str_count) / total_num_gt
+    except ZeroDivisionError:
+        recall_e2e = 0
+
+    try:
+        precision_e2e = float(hit_str_count) / total_num_det
+    except ZeroDivisionError:
+        precision_e2e = 0
+
+    try:
+        f_score_e2e = 2 * precision_e2e * recall_e2e / (precision_e2e + recall_e2e)
+    except ZeroDivisionError:
+        f_score_e2e = 0
+
+    final = {
+        "total_num_gt": total_num_gt,
+        "total_num_det": total_num_det,
+        "global_accumulative_recall": global_accumulative_recall,
+        "hit_str_count": hit_str_count,
+        "recall": recall,
+        "precision": precision,
+        "f_score": f_score,
+        "seqerr": seqerr,
+        "recall_e2e": recall_e2e,
+        "precision_e2e": precision_e2e,
+        "f_score_e2e": f_score_e2e,
+    }
+    return final
diff --git a/ppocr/utils/e2e_metric/polygon_fast.py b/ppocr/utils/e2e_metric/polygon_fast.py
new file mode 100755
index 0000000..e2a08c6
--- /dev/null
+++ b/ppocr/utils/e2e_metric/polygon_fast.py
@@ -0,0 +1,84 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import numpy as np
+from shapely.geometry import Polygon
+
+"""
+:param det_x: [1, N] Xs of detection's vertices
+:param det_y: [1, N] Ys of detection's vertices
+:param gt_x: [1, N] Xs of groundtruth's vertices
+:param gt_y: [1, N] Ys of groundtruth's vertices
+
+##############
+All the calculation of 'AREA' in this script is handled by:
+1) First generating a binary mask with the polygon area filled up with 1's
+2) Summing up all the 1's
+"""
+
+
+def area(x, y):
+    polygon = Polygon(np.stack([x, y], axis=1))
+    return float(polygon.area)
+
+
+def approx_area_of_intersection(det_x, det_y, gt_x, gt_y):
+    """
+    This helper determine if both polygons are intersecting with each others with an approximation method.
+    Area of intersection represented by the minimum bounding rectangular [xmin, ymin, xmax, ymax]
+    """
+    det_ymax = np.max(det_y)
+    det_xmax = np.max(det_x)
+    det_ymin = np.min(det_y)
+    det_xmin = np.min(det_x)
+
+    gt_ymax = np.max(gt_y)
+    gt_xmax = np.max(gt_x)
+    gt_ymin = np.min(gt_y)
+    gt_xmin = np.min(gt_x)
+
+    all_min_ymax = np.minimum(det_ymax, gt_ymax)
+    all_max_ymin = np.maximum(det_ymin, gt_ymin)
+
+    intersect_heights = np.maximum(0.0, (all_min_ymax - all_max_ymin))
+
+    all_min_xmax = np.minimum(det_xmax, gt_xmax)
+    all_max_xmin = np.maximum(det_xmin, gt_xmin)
+    intersect_widths = np.maximum(0.0, (all_min_xmax - all_max_xmin))
+
+    return intersect_heights * intersect_widths
+
+
+def area_of_intersection(det_x, det_y, gt_x, gt_y):
+    p1 = Polygon(np.stack([det_x, det_y], axis=1)).buffer(0)
+    p2 = Polygon(np.stack([gt_x, gt_y], axis=1)).buffer(0)
+    return float(p1.intersection(p2).area)
+
+
+def area_of_union(det_x, det_y, gt_x, gt_y):
+    p1 = Polygon(np.stack([det_x, det_y], axis=1)).buffer(0)
+    p2 = Polygon(np.stack([gt_x, gt_y], axis=1)).buffer(0)
+    return float(p1.union(p2).area)
+
+
+def iou(det_x, det_y, gt_x, gt_y):
+    return area_of_intersection(det_x, det_y, gt_x, gt_y) / (
+        area_of_union(det_x, det_y, gt_x, gt_y) + 1.0
+    )
+
+
+def iod(det_x, det_y, gt_x, gt_y):
+    """
+    This helper determine the fraction of intersection area over detection area
+    """
+    return area_of_intersection(det_x, det_y, gt_x, gt_y) / (area(det_x, det_y) + 1.0)
diff --git a/ppocr/utils/e2e_utils/extract_batchsize.py b/ppocr/utils/e2e_utils/extract_batchsize.py
new file mode 100644
index 0000000..f1ab77b
--- /dev/null
+++ b/ppocr/utils/e2e_utils/extract_batchsize.py
@@ -0,0 +1,88 @@
+import paddle
+import numpy as np
+import copy
+
+
+def org_tcl_rois(batch_size, pos_lists, pos_masks, label_lists, tcl_bs):
+    """ """
+    pos_lists_, pos_masks_, label_lists_ = [], [], []
+    img_bs = batch_size
+    ngpu = int(batch_size / img_bs)
+    img_ids = np.array(pos_lists, dtype=np.int32)[:, 0, 0].copy()
+    pos_lists_split, pos_masks_split, label_lists_split = [], [], []
+    for i in range(ngpu):
+        pos_lists_split.append([])
+        pos_masks_split.append([])
+        label_lists_split.append([])
+
+    for i in range(img_ids.shape[0]):
+        img_id = img_ids[i]
+        gpu_id = int(img_id / img_bs)
+        img_id = img_id % img_bs
+        pos_list = pos_lists[i].copy()
+        pos_list[:, 0] = img_id
+        pos_lists_split[gpu_id].append(pos_list)
+        pos_masks_split[gpu_id].append(pos_masks[i].copy())
+        label_lists_split[gpu_id].append(copy.deepcopy(label_lists[i]))
+    # repeat or delete
+    for i in range(ngpu):
+        vp_len = len(pos_lists_split[i])
+        if vp_len <= tcl_bs:
+            for j in range(0, tcl_bs - vp_len):
+                pos_list = pos_lists_split[i][j].copy()
+                pos_lists_split[i].append(pos_list)
+                pos_mask = pos_masks_split[i][j].copy()
+                pos_masks_split[i].append(pos_mask)
+                label_list = copy.deepcopy(label_lists_split[i][j])
+                label_lists_split[i].append(label_list)
+        else:
+            for j in range(0, vp_len - tcl_bs):
+                c_len = len(pos_lists_split[i])
+                pop_id = np.random.permutation(c_len)[0]
+                pos_lists_split[i].pop(pop_id)
+                pos_masks_split[i].pop(pop_id)
+                label_lists_split[i].pop(pop_id)
+    # merge
+    for i in range(ngpu):
+        pos_lists_.extend(pos_lists_split[i])
+        pos_masks_.extend(pos_masks_split[i])
+        label_lists_.extend(label_lists_split[i])
+    return pos_lists_, pos_masks_, label_lists_
+
+
+def pre_process(
+    label_list, pos_list, pos_mask, max_text_length, max_text_nums, pad_num, tcl_bs
+):
+    label_list = label_list.numpy()
+    batch, _, _, _ = label_list.shape
+    pos_list = pos_list.numpy()
+    pos_mask = pos_mask.numpy()
+    pos_list_t = []
+    pos_mask_t = []
+    label_list_t = []
+    for i in range(batch):
+        for j in range(max_text_nums):
+            if pos_mask[i, j].any():
+                pos_list_t.append(pos_list[i][j])
+                pos_mask_t.append(pos_mask[i][j])
+                label_list_t.append(label_list[i][j])
+    pos_list, pos_mask, label_list = org_tcl_rois(
+        batch, pos_list_t, pos_mask_t, label_list_t, tcl_bs
+    )
+    label = []
+    tt = [l.tolist() for l in label_list]
+    for i in range(tcl_bs):
+        k = 0
+        for j in range(max_text_length):
+            if tt[i][j][0] != pad_num:
+                k += 1
+            else:
+                break
+        label.append(k)
+    label = paddle.to_tensor(label)
+    label = paddle.cast(label, dtype="int64")
+    pos_list = paddle.to_tensor(pos_list)
+    pos_mask = paddle.to_tensor(pos_mask)
+    label_list = paddle.squeeze(paddle.to_tensor(label_list), axis=2)
+    label_list = paddle.cast(label_list, dtype="int32")
+    return pos_list, pos_mask, label_list, label
diff --git a/ppocr/utils/e2e_utils/extract_textpoint_fast.py b/ppocr/utils/e2e_utils/extract_textpoint_fast.py
new file mode 100644
index 0000000..67a89e2
--- /dev/null
+++ b/ppocr/utils/e2e_utils/extract_textpoint_fast.py
@@ -0,0 +1,523 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Contains various CTC decoders."""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import cv2
+import math
+
+import numpy as np
+from itertools import groupby
+from skimage.morphology._skeletonize import thin
+
+
+def get_dict(character_dict_path):
+    character_str = ""
+    with open(character_dict_path, "rb") as fin:
+        lines = fin.readlines()
+        for line in lines:
+            line = line.decode("utf-8").strip("\n").strip("\r\n")
+            character_str += line
+        dict_character = list(character_str)
+    return dict_character
+
+
+def softmax(logits):
+    """
+    logits: N x d
+    """
+    max_value = np.max(logits, axis=1, keepdims=True)
+    exp = np.exp(logits - max_value)
+    exp_sum = np.sum(exp, axis=1, keepdims=True)
+    dist = exp / exp_sum
+    return dist
+
+
+def get_keep_pos_idxs(labels, remove_blank=None):
+    """
+    Remove duplicate and get pos idxs of keep items.
+    The value of keep_blank should be [None, 95].
+    """
+    duplicate_len_list = []
+    keep_pos_idx_list = []
+    keep_char_idx_list = []
+    for k, v_ in groupby(labels):
+        current_len = len(list(v_))
+        if k != remove_blank:
+            current_idx = int(sum(duplicate_len_list) + current_len // 2)
+            keep_pos_idx_list.append(current_idx)
+            keep_char_idx_list.append(k)
+        duplicate_len_list.append(current_len)
+    return keep_char_idx_list, keep_pos_idx_list
+
+
+def remove_blank(labels, blank=0):
+    new_labels = [x for x in labels if x != blank]
+    return new_labels
+
+
+def insert_blank(labels, blank=0):
+    new_labels = [blank]
+    for l in labels:
+        new_labels += [l, blank]
+    return new_labels
+
+
+def ctc_greedy_decoder(probs_seq, blank=95, keep_blank_in_idxs=True):
+    """
+    CTC greedy (best path) decoder.
+    """
+    raw_str = np.argmax(np.array(probs_seq), axis=1)
+    remove_blank_in_pos = None if keep_blank_in_idxs else blank
+    dedup_str, keep_idx_list = get_keep_pos_idxs(
+        raw_str, remove_blank=remove_blank_in_pos
+    )
+    dst_str = remove_blank(dedup_str, blank=blank)
+    return dst_str, keep_idx_list
+
+
+def instance_ctc_greedy_decoder(
+    gather_info, logits_map, pts_num=4, point_gather_mode=None
+):
+    _, _, C = logits_map.shape
+    if point_gather_mode == "align":
+        insert_num = 0
+        gather_info = np.array(gather_info)
+        length = len(gather_info) - 1
+        for index in range(length):
+            stride_y = np.abs(
+                gather_info[index + insert_num][0]
+                - gather_info[index + 1 + insert_num][0]
+            )
+            stride_x = np.abs(
+                gather_info[index + insert_num][1]
+                - gather_info[index + 1 + insert_num][1]
+            )
+            max_points = int(max(stride_x, stride_y))
+            stride = (
+                gather_info[index + insert_num] - gather_info[index + 1 + insert_num]
+            ) / (max_points)
+            insert_num_temp = max_points - 1
+
+            for i in range(int(insert_num_temp)):
+                insert_value = gather_info[index + insert_num] - (i + 1) * stride
+                insert_index = index + i + 1 + insert_num
+                gather_info = np.insert(gather_info, insert_index, insert_value, axis=0)
+            insert_num += insert_num_temp
+        gather_info = gather_info.tolist()
+    else:
+        pass
+    ys, xs = zip(*gather_info)
+    logits_seq = logits_map[list(ys), list(xs)]
+    probs_seq = logits_seq
+    labels = np.argmax(probs_seq, axis=1)
+    dst_str = [k for k, v_ in groupby(labels) if k != C - 1]
+    detal = len(gather_info) // (pts_num - 1)
+    keep_idx_list = [0] + [detal * (i + 1) for i in range(pts_num - 2)] + [-1]
+    keep_gather_list = [gather_info[idx] for idx in keep_idx_list]
+    return dst_str, keep_gather_list
+
+
+def ctc_decoder_for_image(
+    gather_info_list, logits_map, Lexicon_Table, pts_num=6, point_gather_mode=None
+):
+    """
+    CTC decoder using multiple processes.
+    """
+    decoder_str = []
+    decoder_xys = []
+    for gather_info in gather_info_list:
+        if len(gather_info) < pts_num:
+            continue
+        dst_str, xys_list = instance_ctc_greedy_decoder(
+            gather_info,
+            logits_map,
+            pts_num=pts_num,
+            point_gather_mode=point_gather_mode,
+        )
+        dst_str_readable = "".join([Lexicon_Table[idx] for idx in dst_str])
+        if len(dst_str_readable) < 2:
+            continue
+        decoder_str.append(dst_str_readable)
+        decoder_xys.append(xys_list)
+    return decoder_str, decoder_xys
+
+
+def sort_with_direction(pos_list, f_direction):
+    """
+    f_direction: h x w x 2
+    pos_list: [[y, x], [y, x], [y, x] ...]
+    """
+
+    def sort_part_with_direction(pos_list, point_direction):
+        pos_list = np.array(pos_list).reshape(-1, 2)
+        point_direction = np.array(point_direction).reshape(-1, 2)
+        average_direction = np.mean(point_direction, axis=0, keepdims=True)
+        pos_proj_leng = np.sum(pos_list * average_direction, axis=1)
+        sorted_list = pos_list[np.argsort(pos_proj_leng)].tolist()
+        sorted_direction = point_direction[np.argsort(pos_proj_leng)].tolist()
+        return sorted_list, sorted_direction
+
+    pos_list = np.array(pos_list).reshape(-1, 2)
+    point_direction = f_direction[pos_list[:, 0], pos_list[:, 1]]  # x, y
+    point_direction = point_direction[:, ::-1]  # x, y -> y, x
+    sorted_point, sorted_direction = sort_part_with_direction(pos_list, point_direction)
+
+    point_num = len(sorted_point)
+    if point_num >= 16:
+        middle_num = point_num // 2
+        first_part_point = sorted_point[:middle_num]
+        first_point_direction = sorted_direction[:middle_num]
+        sorted_fist_part_point, sorted_fist_part_direction = sort_part_with_direction(
+            first_part_point, first_point_direction
+        )
+
+        last_part_point = sorted_point[middle_num:]
+        last_point_direction = sorted_direction[middle_num:]
+        sorted_last_part_point, sorted_last_part_direction = sort_part_with_direction(
+            last_part_point, last_point_direction
+        )
+        sorted_point = sorted_fist_part_point + sorted_last_part_point
+        sorted_direction = sorted_fist_part_direction + sorted_last_part_direction
+
+    return sorted_point, np.array(sorted_direction)
+
+
+def add_id(pos_list, image_id=0):
+    """
+    Add id for gather feature, for inference.
+    """
+    new_list = []
+    for item in pos_list:
+        new_list.append((image_id, item[0], item[1]))
+    return new_list
+
+
+def sort_and_expand_with_direction(pos_list, f_direction):
+    """
+    f_direction: h x w x 2
+    pos_list: [[y, x], [y, x], [y, x] ...]
+    """
+    h, w, _ = f_direction.shape
+    sorted_list, point_direction = sort_with_direction(pos_list, f_direction)
+
+    point_num = len(sorted_list)
+    sub_direction_len = max(point_num // 3, 2)
+    left_direction = point_direction[:sub_direction_len, :]
+    right_dirction = point_direction[point_num - sub_direction_len :, :]
+
+    left_average_direction = -np.mean(left_direction, axis=0, keepdims=True)
+    left_average_len = np.linalg.norm(left_average_direction)
+    left_start = np.array(sorted_list[0])
+    left_step = left_average_direction / (left_average_len + 1e-6)
+
+    right_average_direction = np.mean(right_dirction, axis=0, keepdims=True)
+    right_average_len = np.linalg.norm(right_average_direction)
+    right_step = right_average_direction / (right_average_len + 1e-6)
+    right_start = np.array(sorted_list[-1])
+
+    append_num = max(int((left_average_len + right_average_len) / 2.0 * 0.15), 1)
+    left_list = []
+    right_list = []
+    for i in range(append_num):
+        ly, lx = (
+            np.round(left_start + left_step * (i + 1))
+            .flatten()
+            .astype("int32")
+            .tolist()
+        )
+        if ly < h and lx < w and (ly, lx) not in left_list:
+            left_list.append((ly, lx))
+        ry, rx = (
+            np.round(right_start + right_step * (i + 1))
+            .flatten()
+            .astype("int32")
+            .tolist()
+        )
+        if ry < h and rx < w and (ry, rx) not in right_list:
+            right_list.append((ry, rx))
+
+    all_list = left_list[::-1] + sorted_list + right_list
+    return all_list
+
+
+def sort_and_expand_with_direction_v2(pos_list, f_direction, binary_tcl_map):
+    """
+    f_direction: h x w x 2
+    pos_list: [[y, x], [y, x], [y, x] ...]
+    binary_tcl_map: h x w
+    """
+    h, w, _ = f_direction.shape
+    sorted_list, point_direction = sort_with_direction(pos_list, f_direction)
+
+    point_num = len(sorted_list)
+    sub_direction_len = max(point_num // 3, 2)
+    left_direction = point_direction[:sub_direction_len, :]
+    right_dirction = point_direction[point_num - sub_direction_len :, :]
+
+    left_average_direction = -np.mean(left_direction, axis=0, keepdims=True)
+    left_average_len = np.linalg.norm(left_average_direction)
+    left_start = np.array(sorted_list[0])
+    left_step = left_average_direction / (left_average_len + 1e-6)
+
+    right_average_direction = np.mean(right_dirction, axis=0, keepdims=True)
+    right_average_len = np.linalg.norm(right_average_direction)
+    right_step = right_average_direction / (right_average_len + 1e-6)
+    right_start = np.array(sorted_list[-1])
+
+    append_num = max(int((left_average_len + right_average_len) / 2.0 * 0.15), 1)
+    max_append_num = 2 * append_num
+
+    left_list = []
+    right_list = []
+    for i in range(max_append_num):
+        ly, lx = (
+            np.round(left_start + left_step * (i + 1))
+            .flatten()
+            .astype("int32")
+            .tolist()
+        )
+        if ly < h and lx < w and (ly, lx) not in left_list:
+            if binary_tcl_map[ly, lx] > 0.5:
+                left_list.append((ly, lx))
+            else:
+                break
+
+    for i in range(max_append_num):
+        ry, rx = (
+            np.round(right_start + right_step * (i + 1))
+            .flatten()
+            .astype("int32")
+            .tolist()
+        )
+        if ry < h and rx < w and (ry, rx) not in right_list:
+            if binary_tcl_map[ry, rx] > 0.5:
+                right_list.append((ry, rx))
+            else:
+                break
+
+    all_list = left_list[::-1] + sorted_list + right_list
+    return all_list
+
+
+def point_pair2poly(point_pair_list):
+    """
+    Transfer vertical point_pairs into poly point in clockwise.
+    """
+    point_num = len(point_pair_list) * 2
+    point_list = [0] * point_num
+    for idx, point_pair in enumerate(point_pair_list):
+        point_list[idx] = point_pair[0]
+        point_list[point_num - 1 - idx] = point_pair[1]
+    return np.array(point_list).reshape(-1, 2)
+
+
+def shrink_quad_along_width(quad, begin_width_ratio=0.0, end_width_ratio=1.0):
+    ratio_pair = np.array([[begin_width_ratio], [end_width_ratio]], dtype=np.float32)
+    p0_1 = quad[0] + (quad[1] - quad[0]) * ratio_pair
+    p3_2 = quad[3] + (quad[2] - quad[3]) * ratio_pair
+    return np.array([p0_1[0], p0_1[1], p3_2[1], p3_2[0]])
+
+
+def expand_poly_along_width(poly, shrink_ratio_of_width=0.3):
+    """
+    expand poly along width.
+    """
+    point_num = poly.shape[0]
+    left_quad = np.array([poly[0], poly[1], poly[-2], poly[-1]], dtype=np.float32)
+    left_ratio = (
+        -shrink_ratio_of_width
+        * np.linalg.norm(left_quad[0] - left_quad[3])
+        / (np.linalg.norm(left_quad[0] - left_quad[1]) + 1e-6)
+    )
+    left_quad_expand = shrink_quad_along_width(left_quad, left_ratio, 1.0)
+    right_quad = np.array(
+        [
+            poly[point_num // 2 - 2],
+            poly[point_num // 2 - 1],
+            poly[point_num // 2],
+            poly[point_num // 2 + 1],
+        ],
+        dtype=np.float32,
+    )
+    right_ratio = 1.0 + shrink_ratio_of_width * np.linalg.norm(
+        right_quad[0] - right_quad[3]
+    ) / (np.linalg.norm(right_quad[0] - right_quad[1]) + 1e-6)
+    right_quad_expand = shrink_quad_along_width(right_quad, 0.0, right_ratio)
+    poly[0] = left_quad_expand[0]
+    poly[-1] = left_quad_expand[-1]
+    poly[point_num // 2 - 1] = right_quad_expand[1]
+    poly[point_num // 2] = right_quad_expand[2]
+    return poly
+
+
+def restore_poly(
+    instance_yxs_list, seq_strs, p_border, ratio_w, ratio_h, src_w, src_h, valid_set
+):
+    poly_list = []
+    keep_str_list = []
+    for yx_center_line, keep_str in zip(instance_yxs_list, seq_strs):
+        if len(keep_str) < 2:
+            print("--> too short, {}".format(keep_str))
+            continue
+
+        offset_expand = 1.0
+        if valid_set == "totaltext":
+            offset_expand = 1.2
+
+        point_pair_list = []
+        for y, x in yx_center_line:
+            offset = p_border[:, y, x].reshape(2, 2) * offset_expand
+            ori_yx = np.array([y, x], dtype=np.float32)
+            point_pair = (
+                (ori_yx + offset)[:, ::-1]
+                * 4.0
+                / np.array([ratio_w, ratio_h]).reshape(-1, 2)
+            )
+            point_pair_list.append(point_pair)
+
+        detected_poly = point_pair2poly(point_pair_list)
+        detected_poly = expand_poly_along_width(
+            detected_poly, shrink_ratio_of_width=0.2
+        )
+        detected_poly[:, 0] = np.clip(detected_poly[:, 0], a_min=0, a_max=src_w)
+        detected_poly[:, 1] = np.clip(detected_poly[:, 1], a_min=0, a_max=src_h)
+
+        keep_str_list.append(keep_str)
+        if valid_set == "partvgg":
+            middle_point = len(detected_poly) // 2
+            detected_poly = detected_poly[[0, middle_point - 1, middle_point, -1], :]
+            poly_list.append(detected_poly)
+        elif valid_set == "totaltext":
+            poly_list.append(detected_poly)
+        else:
+            print("--> Not supported format.")
+            exit(-1)
+    return poly_list, keep_str_list
+
+
+def generate_pivot_list_fast(
+    p_score,
+    p_char_maps,
+    f_direction,
+    Lexicon_Table,
+    score_thresh=0.5,
+    point_gather_mode=None,
+):
+    """
+    return center point and end point of TCL instance; filter with the char maps;
+    """
+    p_score = p_score[0]
+    f_direction = f_direction.transpose(1, 2, 0)
+    p_tcl_map = (p_score > score_thresh) * 1.0
+    skeleton_map = thin(p_tcl_map.astype(np.uint8))
+    instance_count, instance_label_map = cv2.connectedComponents(
+        skeleton_map.astype(np.uint8), connectivity=8
+    )
+
+    # get TCL Instance
+    all_pos_yxs = []
+    if instance_count > 0:
+        for instance_id in range(1, instance_count):
+            pos_list = []
+            ys, xs = np.where(instance_label_map == instance_id)
+            pos_list = list(zip(ys, xs))
+
+            if len(pos_list) < 3:
+                continue
+
+            pos_list_sorted = sort_and_expand_with_direction_v2(
+                pos_list, f_direction, p_tcl_map
+            )
+            all_pos_yxs.append(pos_list_sorted)
+
+    p_char_maps = p_char_maps.transpose([1, 2, 0])
+    decoded_str, keep_yxs_list = ctc_decoder_for_image(
+        all_pos_yxs,
+        logits_map=p_char_maps,
+        Lexicon_Table=Lexicon_Table,
+        point_gather_mode=point_gather_mode,
+    )
+    return keep_yxs_list, decoded_str
+
+
+def extract_main_direction(pos_list, f_direction):
+    """
+    f_direction: h x w x 2
+    pos_list: [[y, x], [y, x], [y, x] ...]
+    """
+    pos_list = np.array(pos_list)
+    point_direction = f_direction[pos_list[:, 0], pos_list[:, 1]]
+    point_direction = point_direction[:, ::-1]  # x, y -> y, x
+    average_direction = np.mean(point_direction, axis=0, keepdims=True)
+    average_direction = average_direction / (np.linalg.norm(average_direction) + 1e-6)
+    return average_direction
+
+
+def sort_by_direction_with_image_id_deprecated(pos_list, f_direction):
+    """
+    f_direction: h x w x 2
+    pos_list: [[id, y, x], [id, y, x], [id, y, x] ...]
+    """
+    pos_list_full = np.array(pos_list).reshape(-1, 3)
+    pos_list = pos_list_full[:, 1:]
+    point_direction = f_direction[pos_list[:, 0], pos_list[:, 1]]  # x, y
+    point_direction = point_direction[:, ::-1]  # x, y -> y, x
+    average_direction = np.mean(point_direction, axis=0, keepdims=True)
+    pos_proj_leng = np.sum(pos_list * average_direction, axis=1)
+    sorted_list = pos_list_full[np.argsort(pos_proj_leng)].tolist()
+    return sorted_list
+
+
+def sort_by_direction_with_image_id(pos_list, f_direction):
+    """
+    f_direction: h x w x 2
+    pos_list: [[y, x], [y, x], [y, x] ...]
+    """
+
+    def sort_part_with_direction(pos_list_full, point_direction):
+        pos_list_full = np.array(pos_list_full).reshape(-1, 3)
+        pos_list = pos_list_full[:, 1:]
+        point_direction = np.array(point_direction).reshape(-1, 2)
+        average_direction = np.mean(point_direction, axis=0, keepdims=True)
+        pos_proj_leng = np.sum(pos_list * average_direction, axis=1)
+        sorted_list = pos_list_full[np.argsort(pos_proj_leng)].tolist()
+        sorted_direction = point_direction[np.argsort(pos_proj_leng)].tolist()
+        return sorted_list, sorted_direction
+
+    pos_list = np.array(pos_list).reshape(-1, 3)
+    point_direction = f_direction[pos_list[:, 1], pos_list[:, 2]]  # x, y
+    point_direction = point_direction[:, ::-1]  # x, y -> y, x
+    sorted_point, sorted_direction = sort_part_with_direction(pos_list, point_direction)
+
+    point_num = len(sorted_point)
+    if point_num >= 16:
+        middle_num = point_num // 2
+        first_part_point = sorted_point[:middle_num]
+        first_point_direction = sorted_direction[:middle_num]
+        sorted_fist_part_point, sorted_fist_part_direction = sort_part_with_direction(
+            first_part_point, first_point_direction
+        )
+
+        last_part_point = sorted_point[middle_num:]
+        last_point_direction = sorted_direction[middle_num:]
+        sorted_last_part_point, sorted_last_part_direction = sort_part_with_direction(
+            last_part_point, last_point_direction
+        )
+        sorted_point = sorted_fist_part_point + sorted_last_part_point
+        sorted_direction = sorted_fist_part_direction + sorted_last_part_direction
+
+    return sorted_point
diff --git a/ppocr/utils/e2e_utils/extract_textpoint_slow.py b/ppocr/utils/e2e_utils/extract_textpoint_slow.py
new file mode 100644
index 0000000..2beff89
--- /dev/null
+++ b/ppocr/utils/e2e_utils/extract_textpoint_slow.py
@@ -0,0 +1,624 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Contains various CTC decoders."""
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import cv2
+import math
+
+import numpy as np
+from itertools import groupby
+from skimage.morphology._skeletonize import thin
+
+
+def get_dict(character_dict_path):
+    character_str = ""
+    with open(character_dict_path, "rb") as fin:
+        lines = fin.readlines()
+        for line in lines:
+            line = line.decode("utf-8").strip("\n").strip("\r\n")
+            character_str += line
+        dict_character = list(character_str)
+    return dict_character
+
+
+def point_pair2poly(point_pair_list):
+    """
+    Transfer vertical point_pairs into poly point in clockwise.
+    """
+    pair_length_list = []
+    for point_pair in point_pair_list:
+        pair_length = np.linalg.norm(point_pair[0] - point_pair[1])
+        pair_length_list.append(pair_length)
+    pair_length_list = np.array(pair_length_list)
+    pair_info = (
+        pair_length_list.max(),
+        pair_length_list.min(),
+        pair_length_list.mean(),
+    )
+
+    point_num = len(point_pair_list) * 2
+    point_list = [0] * point_num
+    for idx, point_pair in enumerate(point_pair_list):
+        point_list[idx] = point_pair[0]
+        point_list[point_num - 1 - idx] = point_pair[1]
+    return np.array(point_list).reshape(-1, 2), pair_info
+
+
+def shrink_quad_along_width(quad, begin_width_ratio=0.0, end_width_ratio=1.0):
+    """
+    Generate shrink_quad_along_width.
+    """
+    ratio_pair = np.array([[begin_width_ratio], [end_width_ratio]], dtype=np.float32)
+    p0_1 = quad[0] + (quad[1] - quad[0]) * ratio_pair
+    p3_2 = quad[3] + (quad[2] - quad[3]) * ratio_pair
+    return np.array([p0_1[0], p0_1[1], p3_2[1], p3_2[0]])
+
+
+def expand_poly_along_width(poly, shrink_ratio_of_width=0.3):
+    """
+    expand poly along width.
+    """
+    point_num = poly.shape[0]
+    left_quad = np.array([poly[0], poly[1], poly[-2], poly[-1]], dtype=np.float32)
+    left_ratio = (
+        -shrink_ratio_of_width
+        * np.linalg.norm(left_quad[0] - left_quad[3])
+        / (np.linalg.norm(left_quad[0] - left_quad[1]) + 1e-6)
+    )
+    left_quad_expand = shrink_quad_along_width(left_quad, left_ratio, 1.0)
+    right_quad = np.array(
+        [
+            poly[point_num // 2 - 2],
+            poly[point_num // 2 - 1],
+            poly[point_num // 2],
+            poly[point_num // 2 + 1],
+        ],
+        dtype=np.float32,
+    )
+    right_ratio = 1.0 + shrink_ratio_of_width * np.linalg.norm(
+        right_quad[0] - right_quad[3]
+    ) / (np.linalg.norm(right_quad[0] - right_quad[1]) + 1e-6)
+    right_quad_expand = shrink_quad_along_width(right_quad, 0.0, right_ratio)
+    poly[0] = left_quad_expand[0]
+    poly[-1] = left_quad_expand[-1]
+    poly[point_num // 2 - 1] = right_quad_expand[1]
+    poly[point_num // 2] = right_quad_expand[2]
+    return poly
+
+
+def softmax(logits):
+    """
+    logits: N x d
+    """
+    max_value = np.max(logits, axis=1, keepdims=True)
+    exp = np.exp(logits - max_value)
+    exp_sum = np.sum(exp, axis=1, keepdims=True)
+    dist = exp / exp_sum
+    return dist
+
+
+def get_keep_pos_idxs(labels, remove_blank=None):
+    """
+    Remove duplicate and get pos idxs of keep items.
+    The value of keep_blank should be [None, 95].
+    """
+    duplicate_len_list = []
+    keep_pos_idx_list = []
+    keep_char_idx_list = []
+    for k, v_ in groupby(labels):
+        current_len = len(list(v_))
+        if k != remove_blank:
+            current_idx = int(sum(duplicate_len_list) + current_len // 2)
+            keep_pos_idx_list.append(current_idx)
+            keep_char_idx_list.append(k)
+        duplicate_len_list.append(current_len)
+    return keep_char_idx_list, keep_pos_idx_list
+
+
+def remove_blank(labels, blank=0):
+    new_labels = [x for x in labels if x != blank]
+    return new_labels
+
+
+def insert_blank(labels, blank=0):
+    new_labels = [blank]
+    for l in labels:
+        new_labels += [l, blank]
+    return new_labels
+
+
+def ctc_greedy_decoder(probs_seq, blank=95, keep_blank_in_idxs=True):
+    """
+    CTC greedy (best path) decoder.
+    """
+    raw_str = np.argmax(np.array(probs_seq), axis=1)
+    remove_blank_in_pos = None if keep_blank_in_idxs else blank
+    dedup_str, keep_idx_list = get_keep_pos_idxs(
+        raw_str, remove_blank=remove_blank_in_pos
+    )
+    dst_str = remove_blank(dedup_str, blank=blank)
+    return dst_str, keep_idx_list
+
+
+def instance_ctc_greedy_decoder(gather_info, logits_map, keep_blank_in_idxs=True):
+    """
+    gather_info: [[x, y], [x, y] ...]
+    logits_map: H x W X (n_chars + 1)
+    """
+    _, _, C = logits_map.shape
+    ys, xs = zip(*gather_info)
+    logits_seq = logits_map[list(ys), list(xs)]  # n x 96
+    probs_seq = softmax(logits_seq)
+    dst_str, keep_idx_list = ctc_greedy_decoder(
+        probs_seq, blank=C - 1, keep_blank_in_idxs=keep_blank_in_idxs
+    )
+    keep_gather_list = [gather_info[idx] for idx in keep_idx_list]
+    return dst_str, keep_gather_list
+
+
+def ctc_decoder_for_image(gather_info_list, logits_map, keep_blank_in_idxs=True):
+    """
+    CTC decoder using multiple processes.
+    """
+    decoder_results = []
+    for gather_info in gather_info_list:
+        res = instance_ctc_greedy_decoder(
+            gather_info, logits_map, keep_blank_in_idxs=keep_blank_in_idxs
+        )
+        decoder_results.append(res)
+    return decoder_results
+
+
+def sort_with_direction(pos_list, f_direction):
+    """
+    f_direction: h x w x 2
+    pos_list: [[y, x], [y, x], [y, x] ...]
+    """
+
+    def sort_part_with_direction(pos_list, point_direction):
+        pos_list = np.array(pos_list).reshape(-1, 2)
+        point_direction = np.array(point_direction).reshape(-1, 2)
+        average_direction = np.mean(point_direction, axis=0, keepdims=True)
+        pos_proj_leng = np.sum(pos_list * average_direction, axis=1)
+        sorted_list = pos_list[np.argsort(pos_proj_leng)].tolist()
+        sorted_direction = point_direction[np.argsort(pos_proj_leng)].tolist()
+        return sorted_list, sorted_direction
+
+    pos_list = np.array(pos_list).reshape(-1, 2)
+    point_direction = f_direction[pos_list[:, 0], pos_list[:, 1]]  # x, y
+    point_direction = point_direction[:, ::-1]  # x, y -> y, x
+    sorted_point, sorted_direction = sort_part_with_direction(pos_list, point_direction)
+
+    point_num = len(sorted_point)
+    if point_num >= 16:
+        middle_num = point_num // 2
+        first_part_point = sorted_point[:middle_num]
+        first_point_direction = sorted_direction[:middle_num]
+        sorted_fist_part_point, sorted_fist_part_direction = sort_part_with_direction(
+            first_part_point, first_point_direction
+        )
+
+        last_part_point = sorted_point[middle_num:]
+        last_point_direction = sorted_direction[middle_num:]
+        sorted_last_part_point, sorted_last_part_direction = sort_part_with_direction(
+            last_part_point, last_point_direction
+        )
+        sorted_point = sorted_fist_part_point + sorted_last_part_point
+        sorted_direction = sorted_fist_part_direction + sorted_last_part_direction
+
+    return sorted_point, np.array(sorted_direction)
+
+
+def add_id(pos_list, image_id=0):
+    """
+    Add id for gather feature, for inference.
+    """
+    new_list = []
+    for item in pos_list:
+        new_list.append((image_id, item[0], item[1]))
+    return new_list
+
+
+def sort_and_expand_with_direction(pos_list, f_direction):
+    """
+    f_direction: h x w x 2
+    pos_list: [[y, x], [y, x], [y, x] ...]
+    """
+    h, w, _ = f_direction.shape
+    sorted_list, point_direction = sort_with_direction(pos_list, f_direction)
+
+    # expand along
+    point_num = len(sorted_list)
+    sub_direction_len = max(point_num // 3, 2)
+    left_direction = point_direction[:sub_direction_len, :]
+    right_dirction = point_direction[point_num - sub_direction_len :, :]
+
+    left_average_direction = -np.mean(left_direction, axis=0, keepdims=True)
+    left_average_len = np.linalg.norm(left_average_direction)
+    left_start = np.array(sorted_list[0])
+    left_step = left_average_direction / (left_average_len + 1e-6)
+
+    right_average_direction = np.mean(right_dirction, axis=0, keepdims=True)
+    right_average_len = np.linalg.norm(right_average_direction)
+    right_step = right_average_direction / (right_average_len + 1e-6)
+    right_start = np.array(sorted_list[-1])
+
+    append_num = max(int((left_average_len + right_average_len) / 2.0 * 0.15), 1)
+    left_list = []
+    right_list = []
+    for i in range(append_num):
+        ly, lx = (
+            np.round(left_start + left_step * (i + 1))
+            .flatten()
+            .astype("int32")
+            .tolist()
+        )
+        if ly < h and lx < w and (ly, lx) not in left_list:
+            left_list.append((ly, lx))
+        ry, rx = (
+            np.round(right_start + right_step * (i + 1))
+            .flatten()
+            .astype("int32")
+            .tolist()
+        )
+        if ry < h and rx < w and (ry, rx) not in right_list:
+            right_list.append((ry, rx))
+
+    all_list = left_list[::-1] + sorted_list + right_list
+    return all_list
+
+
+def sort_and_expand_with_direction_v2(pos_list, f_direction, binary_tcl_map):
+    """
+    f_direction: h x w x 2
+    pos_list: [[y, x], [y, x], [y, x] ...]
+    binary_tcl_map: h x w
+    """
+    h, w, _ = f_direction.shape
+    sorted_list, point_direction = sort_with_direction(pos_list, f_direction)
+
+    # expand along
+    point_num = len(sorted_list)
+    sub_direction_len = max(point_num // 3, 2)
+    left_direction = point_direction[:sub_direction_len, :]
+    right_dirction = point_direction[point_num - sub_direction_len :, :]
+
+    left_average_direction = -np.mean(left_direction, axis=0, keepdims=True)
+    left_average_len = np.linalg.norm(left_average_direction)
+    left_start = np.array(sorted_list[0])
+    left_step = left_average_direction / (left_average_len + 1e-6)
+
+    right_average_direction = np.mean(right_dirction, axis=0, keepdims=True)
+    right_average_len = np.linalg.norm(right_average_direction)
+    right_step = right_average_direction / (right_average_len + 1e-6)
+    right_start = np.array(sorted_list[-1])
+
+    append_num = max(int((left_average_len + right_average_len) / 2.0 * 0.15), 1)
+    max_append_num = 2 * append_num
+
+    left_list = []
+    right_list = []
+    for i in range(max_append_num):
+        ly, lx = (
+            np.round(left_start + left_step * (i + 1))
+            .flatten()
+            .astype("int32")
+            .tolist()
+        )
+        if ly < h and lx < w and (ly, lx) not in left_list:
+            if binary_tcl_map[ly, lx] > 0.5:
+                left_list.append((ly, lx))
+            else:
+                break
+
+    for i in range(max_append_num):
+        ry, rx = (
+            np.round(right_start + right_step * (i + 1))
+            .flatten()
+            .astype("int32")
+            .tolist()
+        )
+        if ry < h and rx < w and (ry, rx) not in right_list:
+            if binary_tcl_map[ry, rx] > 0.5:
+                right_list.append((ry, rx))
+            else:
+                break
+
+    all_list = left_list[::-1] + sorted_list + right_list
+    return all_list
+
+
+def generate_pivot_list_curved(
+    p_score,
+    p_char_maps,
+    f_direction,
+    score_thresh=0.5,
+    is_expand=True,
+    is_backbone=False,
+    image_id=0,
+):
+    """
+    return center point and end point of TCL instance; filter with the char maps;
+    """
+    p_score = p_score[0]
+    f_direction = f_direction.transpose(1, 2, 0)
+    p_tcl_map = (p_score > score_thresh) * 1.0
+    skeleton_map = thin(p_tcl_map)
+    instance_count, instance_label_map = cv2.connectedComponents(
+        skeleton_map.astype(np.uint8), connectivity=8
+    )
+
+    # get TCL Instance
+    all_pos_yxs = []
+    center_pos_yxs = []
+    end_points_yxs = []
+    instance_center_pos_yxs = []
+    pred_strs = []
+    if instance_count > 0:
+        for instance_id in range(1, instance_count):
+            pos_list = []
+            ys, xs = np.where(instance_label_map == instance_id)
+            pos_list = list(zip(ys, xs))
+
+            ### FIX-ME, eliminate outlier
+            if len(pos_list) < 3:
+                continue
+
+            if is_expand:
+                pos_list_sorted = sort_and_expand_with_direction_v2(
+                    pos_list, f_direction, p_tcl_map
+                )
+            else:
+                pos_list_sorted, _ = sort_with_direction(pos_list, f_direction)
+            all_pos_yxs.append(pos_list_sorted)
+
+    # use decoder to filter background points.
+    p_char_maps = p_char_maps.transpose([1, 2, 0])
+    decode_res = ctc_decoder_for_image(
+        all_pos_yxs, logits_map=p_char_maps, keep_blank_in_idxs=True
+    )
+    for decoded_str, keep_yxs_list in decode_res:
+        if is_backbone:
+            keep_yxs_list_with_id = add_id(keep_yxs_list, image_id=image_id)
+            instance_center_pos_yxs.append(keep_yxs_list_with_id)
+            pred_strs.append(decoded_str)
+        else:
+            end_points_yxs.extend((keep_yxs_list[0], keep_yxs_list[-1]))
+            center_pos_yxs.extend(keep_yxs_list)
+
+    if is_backbone:
+        return pred_strs, instance_center_pos_yxs
+    else:
+        return center_pos_yxs, end_points_yxs
+
+
+def generate_pivot_list_horizontal(
+    p_score, p_char_maps, f_direction, score_thresh=0.5, is_backbone=False, image_id=0
+):
+    """
+    return center point and end point of TCL instance; filter with the char maps;
+    """
+    p_score = p_score[0]
+    f_direction = f_direction.transpose(1, 2, 0)
+    p_tcl_map_bi = (p_score > score_thresh) * 1.0
+    instance_count, instance_label_map = cv2.connectedComponents(
+        p_tcl_map_bi.astype(np.uint8), connectivity=8
+    )
+
+    # get TCL Instance
+    all_pos_yxs = []
+    center_pos_yxs = []
+    end_points_yxs = []
+    instance_center_pos_yxs = []
+
+    if instance_count > 0:
+        for instance_id in range(1, instance_count):
+            pos_list = []
+            ys, xs = np.where(instance_label_map == instance_id)
+            pos_list = list(zip(ys, xs))
+
+            ### FIX-ME, eliminate outlier
+            if len(pos_list) < 5:
+                continue
+
+            # add rule here
+            main_direction = extract_main_direction(pos_list, f_direction)  # y x
+            reference_directin = np.array([0, 1]).reshape([-1, 2])  # y x
+            is_h_angle = abs(np.sum(main_direction * reference_directin)) < math.cos(
+                math.pi / 180 * 70
+            )
+
+            point_yxs = np.array(pos_list)
+            max_y, max_x = np.max(point_yxs, axis=0)
+            min_y, min_x = np.min(point_yxs, axis=0)
+            is_h_len = (max_y - min_y) < 1.5 * (max_x - min_x)
+
+            pos_list_final = []
+            if is_h_len:
+                xs = np.unique(xs)
+                for x in xs:
+                    ys = instance_label_map[:, x].copy().reshape((-1,))
+                    y = int(np.where(ys == instance_id)[0].mean())
+                    pos_list_final.append((y, x))
+            else:
+                ys = np.unique(ys)
+                for y in ys:
+                    xs = instance_label_map[y, :].copy().reshape((-1,))
+                    x = int(np.where(xs == instance_id)[0].mean())
+                    pos_list_final.append((y, x))
+
+            pos_list_sorted, _ = sort_with_direction(pos_list_final, f_direction)
+            all_pos_yxs.append(pos_list_sorted)
+
+    # use decoder to filter background points.
+    p_char_maps = p_char_maps.transpose([1, 2, 0])
+    decode_res = ctc_decoder_for_image(
+        all_pos_yxs, logits_map=p_char_maps, keep_blank_in_idxs=True
+    )
+    for decoded_str, keep_yxs_list in decode_res:
+        if is_backbone:
+            keep_yxs_list_with_id = add_id(keep_yxs_list, image_id=image_id)
+            instance_center_pos_yxs.append(keep_yxs_list_with_id)
+        else:
+            end_points_yxs.extend((keep_yxs_list[0], keep_yxs_list[-1]))
+            center_pos_yxs.extend(keep_yxs_list)
+
+    if is_backbone:
+        return instance_center_pos_yxs
+    else:
+        return center_pos_yxs, end_points_yxs
+
+
+def generate_pivot_list_slow(
+    p_score,
+    p_char_maps,
+    f_direction,
+    score_thresh=0.5,
+    is_backbone=False,
+    is_curved=True,
+    image_id=0,
+):
+    """
+    Warp all the function together.
+    """
+    if is_curved:
+        return generate_pivot_list_curved(
+            p_score,
+            p_char_maps,
+            f_direction,
+            score_thresh=score_thresh,
+            is_expand=True,
+            is_backbone=is_backbone,
+            image_id=image_id,
+        )
+    else:
+        return generate_pivot_list_horizontal(
+            p_score,
+            p_char_maps,
+            f_direction,
+            score_thresh=score_thresh,
+            is_backbone=is_backbone,
+            image_id=image_id,
+        )
+
+
+# for refine module
+def extract_main_direction(pos_list, f_direction):
+    """
+    f_direction: h x w x 2
+    pos_list: [[y, x], [y, x], [y, x] ...]
+    """
+    pos_list = np.array(pos_list)
+    point_direction = f_direction[pos_list[:, 0], pos_list[:, 1]]
+    point_direction = point_direction[:, ::-1]  # x, y -> y, x
+    average_direction = np.mean(point_direction, axis=0, keepdims=True)
+    average_direction = average_direction / (np.linalg.norm(average_direction) + 1e-6)
+    return average_direction
+
+
+def sort_by_direction_with_image_id_deprecated(pos_list, f_direction):
+    """
+    f_direction: h x w x 2
+    pos_list: [[id, y, x], [id, y, x], [id, y, x] ...]
+    """
+    pos_list_full = np.array(pos_list).reshape(-1, 3)
+    pos_list = pos_list_full[:, 1:]
+    point_direction = f_direction[pos_list[:, 0], pos_list[:, 1]]  # x, y
+    point_direction = point_direction[:, ::-1]  # x, y -> y, x
+    average_direction = np.mean(point_direction, axis=0, keepdims=True)
+    pos_proj_leng = np.sum(pos_list * average_direction, axis=1)
+    sorted_list = pos_list_full[np.argsort(pos_proj_leng)].tolist()
+    return sorted_list
+
+
+def sort_by_direction_with_image_id(pos_list, f_direction):
+    """
+    f_direction: h x w x 2
+    pos_list: [[y, x], [y, x], [y, x] ...]
+    """
+
+    def sort_part_with_direction(pos_list_full, point_direction):
+        pos_list_full = np.array(pos_list_full).reshape(-1, 3)
+        pos_list = pos_list_full[:, 1:]
+        point_direction = np.array(point_direction).reshape(-1, 2)
+        average_direction = np.mean(point_direction, axis=0, keepdims=True)
+        pos_proj_leng = np.sum(pos_list * average_direction, axis=1)
+        sorted_list = pos_list_full[np.argsort(pos_proj_leng)].tolist()
+        sorted_direction = point_direction[np.argsort(pos_proj_leng)].tolist()
+        return sorted_list, sorted_direction
+
+    pos_list = np.array(pos_list).reshape(-1, 3)
+    point_direction = f_direction[pos_list[:, 1], pos_list[:, 2]]  # x, y
+    point_direction = point_direction[:, ::-1]  # x, y -> y, x
+    sorted_point, sorted_direction = sort_part_with_direction(pos_list, point_direction)
+
+    point_num = len(sorted_point)
+    if point_num >= 16:
+        middle_num = point_num // 2
+        first_part_point = sorted_point[:middle_num]
+        first_point_direction = sorted_direction[:middle_num]
+        sorted_fist_part_point, sorted_fist_part_direction = sort_part_with_direction(
+            first_part_point, first_point_direction
+        )
+
+        last_part_point = sorted_point[middle_num:]
+        last_point_direction = sorted_direction[middle_num:]
+        sorted_last_part_point, sorted_last_part_direction = sort_part_with_direction(
+            last_part_point, last_point_direction
+        )
+        sorted_point = sorted_fist_part_point + sorted_last_part_point
+        sorted_direction = sorted_fist_part_direction + sorted_last_part_direction
+
+    return sorted_point
+
+
+def generate_pivot_list_tt_inference(
+    p_score,
+    p_char_maps,
+    f_direction,
+    score_thresh=0.5,
+    is_backbone=False,
+    is_curved=True,
+    image_id=0,
+):
+    """
+    return center point and end point of TCL instance; filter with the char maps;
+    """
+    p_score = p_score[0]
+    f_direction = f_direction.transpose(1, 2, 0)
+    p_tcl_map = (p_score > score_thresh) * 1.0
+    skeleton_map = thin(p_tcl_map)
+    instance_count, instance_label_map = cv2.connectedComponents(
+        skeleton_map.astype(np.uint8), connectivity=8
+    )
+
+    # get TCL Instance
+    all_pos_yxs = []
+    if instance_count > 0:
+        for instance_id in range(1, instance_count):
+            pos_list = []
+            ys, xs = np.where(instance_label_map == instance_id)
+            pos_list = list(zip(ys, xs))
+            ### FIX-ME, eliminate outlier
+            if len(pos_list) < 3:
+                continue
+            pos_list_sorted = sort_and_expand_with_direction_v2(
+                pos_list, f_direction, p_tcl_map
+            )
+            pos_list_sorted_with_id = add_id(pos_list_sorted, image_id=image_id)
+            all_pos_yxs.append(pos_list_sorted_with_id)
+    return all_pos_yxs
diff --git a/ppocr/utils/e2e_utils/pgnet_pp_utils.py b/ppocr/utils/e2e_utils/pgnet_pp_utils.py
new file mode 100644
index 0000000..71379e5
--- /dev/null
+++ b/ppocr/utils/e2e_utils/pgnet_pp_utils.py
@@ -0,0 +1,179 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+import paddle
+import os
+import sys
+
+__dir__ = os.path.dirname(__file__)
+sys.path.append(__dir__)
+sys.path.append(os.path.join(__dir__, ".."))
+from extract_textpoint_slow import *
+from extract_textpoint_fast import generate_pivot_list_fast, restore_poly
+
+
+class PGNet_PostProcess(object):
+    # two different post-process
+    def __init__(
+        self,
+        character_dict_path,
+        valid_set,
+        score_thresh,
+        outs_dict,
+        shape_list,
+        point_gather_mode=None,
+    ):
+        self.Lexicon_Table = get_dict(character_dict_path)
+        self.valid_set = valid_set
+        self.score_thresh = score_thresh
+        self.outs_dict = outs_dict
+        self.shape_list = shape_list
+        self.point_gather_mode = point_gather_mode
+
+    def pg_postprocess_fast(self):
+        p_score = self.outs_dict["f_score"]
+        p_border = self.outs_dict["f_border"]
+        p_char = self.outs_dict["f_char"]
+        p_direction = self.outs_dict["f_direction"]
+        if isinstance(p_score, paddle.Tensor):
+            p_score = p_score[0].numpy()
+            p_border = p_border[0].numpy()
+            p_direction = p_direction[0].numpy()
+            p_char = p_char[0].numpy()
+        else:
+            p_score = p_score[0]
+            p_border = p_border[0]
+            p_direction = p_direction[0]
+            p_char = p_char[0]
+
+        src_h, src_w, ratio_h, ratio_w = self.shape_list[0]
+        instance_yxs_list, seq_strs = generate_pivot_list_fast(
+            p_score,
+            p_char,
+            p_direction,
+            self.Lexicon_Table,
+            score_thresh=self.score_thresh,
+            point_gather_mode=self.point_gather_mode,
+        )
+        poly_list, keep_str_list = restore_poly(
+            instance_yxs_list,
+            seq_strs,
+            p_border,
+            ratio_w,
+            ratio_h,
+            src_w,
+            src_h,
+            self.valid_set,
+        )
+        data = {
+            "points": poly_list,
+            "texts": keep_str_list,
+        }
+        return data
+
+    def pg_postprocess_slow(self):
+        p_score = self.outs_dict["f_score"]
+        p_border = self.outs_dict["f_border"]
+        p_char = self.outs_dict["f_char"]
+        p_direction = self.outs_dict["f_direction"]
+        if isinstance(p_score, paddle.Tensor):
+            p_score = p_score[0].numpy()
+            p_border = p_border[0].numpy()
+            p_direction = p_direction[0].numpy()
+            p_char = p_char[0].numpy()
+        else:
+            p_score = p_score[0]
+            p_border = p_border[0]
+            p_direction = p_direction[0]
+            p_char = p_char[0]
+        src_h, src_w, ratio_h, ratio_w = self.shape_list[0]
+        is_curved = self.valid_set == "totaltext"
+        char_seq_idx_set, instance_yxs_list = generate_pivot_list_slow(
+            p_score,
+            p_char,
+            p_direction,
+            score_thresh=self.score_thresh,
+            is_backbone=True,
+            is_curved=is_curved,
+        )
+        seq_strs = []
+        for char_idx_set in char_seq_idx_set:
+            pr_str = "".join([self.Lexicon_Table[pos] for pos in char_idx_set])
+            seq_strs.append(pr_str)
+        poly_list = []
+        keep_str_list = []
+        all_point_list = []
+        all_point_pair_list = []
+        for yx_center_line, keep_str in zip(instance_yxs_list, seq_strs):
+            if len(yx_center_line) == 1:
+                yx_center_line.append(yx_center_line[-1])
+
+            offset_expand = 1.0
+            if self.valid_set == "totaltext":
+                offset_expand = 1.2
+
+            point_pair_list = []
+            for batch_id, y, x in yx_center_line:
+                offset = p_border[:, y, x].reshape(2, 2)
+                if offset_expand != 1.0:
+                    offset_length = np.linalg.norm(offset, axis=1, keepdims=True)
+                    expand_length = np.clip(
+                        offset_length * (offset_expand - 1), a_min=0.5, a_max=3.0
+                    )
+                    offset_detal = offset / offset_length * expand_length
+                    offset = offset + offset_detal
+                ori_yx = np.array([y, x], dtype=np.float32)
+                point_pair = (
+                    (ori_yx + offset)[:, ::-1]
+                    * 4.0
+                    / np.array([ratio_w, ratio_h]).reshape(-1, 2)
+                )
+                point_pair_list.append(point_pair)
+
+                all_point_list.append(
+                    [int(round(x * 4.0 / ratio_w)), int(round(y * 4.0 / ratio_h))]
+                )
+                all_point_pair_list.append(point_pair.round().astype(np.int32).tolist())
+
+            detected_poly, pair_length_info = point_pair2poly(point_pair_list)
+            detected_poly = expand_poly_along_width(
+                detected_poly, shrink_ratio_of_width=0.2
+            )
+            detected_poly[:, 0] = np.clip(detected_poly[:, 0], a_min=0, a_max=src_w)
+            detected_poly[:, 1] = np.clip(detected_poly[:, 1], a_min=0, a_max=src_h)
+
+            if len(keep_str) < 2:
+                continue
+
+            keep_str_list.append(keep_str)
+            detected_poly = np.round(detected_poly).astype("int32")
+            if self.valid_set == "partvgg":
+                middle_point = len(detected_poly) // 2
+                detected_poly = detected_poly[
+                    [0, middle_point - 1, middle_point, -1], :
+                ]
+                poly_list.append(detected_poly)
+            elif self.valid_set == "totaltext":
+                poly_list.append(detected_poly)
+            else:
+                print("--> Not supported format.")
+                exit(-1)
+        data = {
+            "points": poly_list,
+            "texts": keep_str_list,
+        }
+        return data
diff --git a/ppocr/utils/e2e_utils/visual.py b/ppocr/utils/e2e_utils/visual.py
new file mode 100644
index 0000000..e321827
--- /dev/null
+++ b/ppocr/utils/e2e_utils/visual.py
@@ -0,0 +1,167 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import numpy as np
+import cv2
+import time
+
+
+def resize_image(im, max_side_len=512):
+    """
+    resize image to a size multiple of max_stride which is required by the network
+    :param im: the resized image
+    :param max_side_len: limit of max image size to avoid out of memory in gpu
+    :return: the resized image and the resize ratio
+    """
+    h, w, _ = im.shape
+
+    resize_w = w
+    resize_h = h
+
+    if resize_h > resize_w:
+        ratio = float(max_side_len) / resize_h
+    else:
+        ratio = float(max_side_len) / resize_w
+
+    resize_h = int(resize_h * ratio)
+    resize_w = int(resize_w * ratio)
+
+    max_stride = 128
+    resize_h = (resize_h + max_stride - 1) // max_stride * max_stride
+    resize_w = (resize_w + max_stride - 1) // max_stride * max_stride
+    im = cv2.resize(im, (int(resize_w), int(resize_h)))
+    ratio_h = resize_h / float(h)
+    ratio_w = resize_w / float(w)
+
+    return im, (ratio_h, ratio_w)
+
+
+def resize_image_min(im, max_side_len=512):
+    """ """
+    h, w, _ = im.shape
+
+    resize_w = w
+    resize_h = h
+
+    if resize_h < resize_w:
+        ratio = float(max_side_len) / resize_h
+    else:
+        ratio = float(max_side_len) / resize_w
+
+    resize_h = int(resize_h * ratio)
+    resize_w = int(resize_w * ratio)
+
+    max_stride = 128
+    resize_h = (resize_h + max_stride - 1) // max_stride * max_stride
+    resize_w = (resize_w + max_stride - 1) // max_stride * max_stride
+    im = cv2.resize(im, (int(resize_w), int(resize_h)))
+    ratio_h = resize_h / float(h)
+    ratio_w = resize_w / float(w)
+    return im, (ratio_h, ratio_w)
+
+
+def resize_image_for_totaltext(im, max_side_len=512):
+    """ """
+    h, w, _ = im.shape
+
+    resize_w = w
+    resize_h = h
+    ratio = 1.25
+    if h * ratio > max_side_len:
+        ratio = float(max_side_len) / resize_h
+
+    resize_h = int(resize_h * ratio)
+    resize_w = int(resize_w * ratio)
+
+    max_stride = 128
+    resize_h = (resize_h + max_stride - 1) // max_stride * max_stride
+    resize_w = (resize_w + max_stride - 1) // max_stride * max_stride
+    im = cv2.resize(im, (int(resize_w), int(resize_h)))
+    ratio_h = resize_h / float(h)
+    ratio_w = resize_w / float(w)
+    return im, (ratio_h, ratio_w)
+
+
+def point_pair2poly(point_pair_list):
+    """
+    Transfer vertical point_pairs into poly point in clockwise.
+    """
+    pair_length_list = []
+    for point_pair in point_pair_list:
+        pair_length = np.linalg.norm(point_pair[0] - point_pair[1])
+        pair_length_list.append(pair_length)
+    pair_length_list = np.array(pair_length_list)
+    pair_info = (
+        pair_length_list.max(),
+        pair_length_list.min(),
+        pair_length_list.mean(),
+    )
+
+    point_num = len(point_pair_list) * 2
+    point_list = [0] * point_num
+    for idx, point_pair in enumerate(point_pair_list):
+        point_list[idx] = point_pair[0]
+        point_list[point_num - 1 - idx] = point_pair[1]
+    return np.array(point_list).reshape(-1, 2), pair_info
+
+
+def shrink_quad_along_width(quad, begin_width_ratio=0.0, end_width_ratio=1.0):
+    """
+    Generate shrink_quad_along_width.
+    """
+    ratio_pair = np.array([[begin_width_ratio], [end_width_ratio]], dtype=np.float32)
+    p0_1 = quad[0] + (quad[1] - quad[0]) * ratio_pair
+    p3_2 = quad[3] + (quad[2] - quad[3]) * ratio_pair
+    return np.array([p0_1[0], p0_1[1], p3_2[1], p3_2[0]])
+
+
+def expand_poly_along_width(poly, shrink_ratio_of_width=0.3):
+    """
+    expand poly along width.
+    """
+    point_num = poly.shape[0]
+    left_quad = np.array([poly[0], poly[1], poly[-2], poly[-1]], dtype=np.float32)
+    left_ratio = (
+        -shrink_ratio_of_width
+        * np.linalg.norm(left_quad[0] - left_quad[3])
+        / (np.linalg.norm(left_quad[0] - left_quad[1]) + 1e-6)
+    )
+    left_quad_expand = shrink_quad_along_width(left_quad, left_ratio, 1.0)
+    right_quad = np.array(
+        [
+            poly[point_num // 2 - 2],
+            poly[point_num // 2 - 1],
+            poly[point_num // 2],
+            poly[point_num // 2 + 1],
+        ],
+        dtype=np.float32,
+    )
+    right_ratio = 1.0 + shrink_ratio_of_width * np.linalg.norm(
+        right_quad[0] - right_quad[3]
+    ) / (np.linalg.norm(right_quad[0] - right_quad[1]) + 1e-6)
+    right_quad_expand = shrink_quad_along_width(right_quad, 0.0, right_ratio)
+    poly[0] = left_quad_expand[0]
+    poly[-1] = left_quad_expand[-1]
+    poly[point_num // 2 - 1] = right_quad_expand[1]
+    poly[point_num // 2] = right_quad_expand[2]
+    return poly
+
+
+def norm2(x, axis=None):
+    if axis:
+        return np.sqrt(np.sum(x**2, axis=axis))
+    return np.sqrt(np.sum(x**2))
+
+
+def cos(p1, p2):
+    return (p1 * p2).sum() / (norm2(p1) * norm2(p2))
diff --git a/ppocr/utils/ema.py b/ppocr/utils/ema.py
new file mode 100644
index 0000000..b127f67
--- /dev/null
+++ b/ppocr/utils/ema.py
@@ -0,0 +1,123 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import math
+
+import paddle
+
+
+class ModelEMA:
+    """Exponential Moving Average for model parameters.
+
+    Maintains shadow copies of model parameters and updates them with:
+        ema_param = decay * ema_param + (1 - decay) * cur_param
+
+    Reference: PaddleDetection ppdet/optimizer/ema.py
+
+    Args:
+        model (nn.Layer): The model whose parameters will be averaged.
+        decay (float): EMA decay coefficient. Default: 0.9998.
+        gamma (int): Warmup parameter for threshold/exponential decay.
+            Default: 2000.
+        ema_decay_type (str): Decay schedule type, one of
+            'threshold' (default), 'exponential', 'normal'.
+        ema_filter_no_grad (bool): If True, parameters with
+            stop_gradient=True (e.g. frozen Teacher in distillation)
+            are excluded from EMA and pass through unchanged.
+            BN running stats are kept even if no-grad. Default: False.
+    """
+
+    def __init__(
+        self,
+        model,
+        decay=0.9998,
+        gamma=2000,
+        ema_decay_type="threshold",
+        ema_filter_no_grad=False,
+    ):
+        self.decay = decay
+        self.gamma = gamma
+        self.ema_decay_type = ema_decay_type
+        self.step = 0
+        self._decay = decay
+
+        # Build black list: frozen params (excluding BN running stats)
+        self.ema_black_list = set()
+        if ema_filter_no_grad:
+            bn_state_names = set()
+            for name, layer in model.named_sublayers():
+                if isinstance(layer, (paddle.nn.BatchNorm2D, paddle.nn.BatchNorm1D)):
+                    prefix = name + "." if name else ""
+                    bn_state_names.add(prefix + "_mean")
+                    bn_state_names.add(prefix + "_variance")
+            for n, p in model.named_parameters():
+                if p.stop_gradient and n not in bn_state_names:
+                    self.ema_black_list.add(n)
+
+        # Initialize shadow weights
+        self.state_dict = {}
+        for k, v in model.state_dict().items():
+            if k in self.ema_black_list:
+                self.state_dict[k] = v.clone()
+            else:
+                self.state_dict[k] = paddle.zeros_like(v).astype("float32")
+
+    def _get_decay(self):
+        if self.ema_decay_type == "threshold":
+            return min(self.decay, (1 + self.step) / (10 + self.step))
+        elif self.ema_decay_type == "exponential":
+            return self.decay * (1 - math.exp(-(self.step + 1) / self.gamma))
+        else:  # normal
+            return self.decay
+
+    def update(self, model):
+        """Update shadow weights with current model parameters."""
+        decay = self._get_decay()
+        self._decay = decay
+        model_dict = model.state_dict()
+        for k, v in self.state_dict.items():
+            if k not in self.ema_black_list and k in model_dict:
+                v = decay * v + (1 - decay) * model_dict[k].astype("float32")
+                v.stop_gradient = True
+                self.state_dict[k] = v
+        self.step += 1
+
+    def apply(self):
+        """Return bias-corrected EMA state dict for eval/save.
+
+        Does NOT modify internal state.
+        """
+        if self.step == 0:
+            return {k: v.clone() for k, v in self.state_dict.items()}
+        state = {}
+        for k, v in self.state_dict.items():
+            if k in self.ema_black_list:
+                state[k] = v
+            else:
+                if self.ema_decay_type != "exponential":
+                    # threshold / normal need bias-correction
+                    v = v / (1 - self._decay**self.step)
+                v = v.clone()
+                v.stop_gradient = True
+                state[k] = v
+        return state
+
+    def state_dict_for_save(self):
+        """Return serializable dict for checkpoint."""
+        return {"ema_state": self.state_dict, "step": self.step}
+
+    def set_state_dict(self, d):
+        """Restore from checkpoint."""
+        self.state_dict = d["ema_state"]
+        self.step = d["step"]
diff --git a/ppocr/utils/en_dict.txt b/ppocr/utils/en_dict.txt
new file mode 100644
index 0000000..7677d31
--- /dev/null
+++ b/ppocr/utils/en_dict.txt
@@ -0,0 +1,95 @@
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+<
+=
+>
+?
+@
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+Z
+[
+\
+]
+^
+_
+`
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+{
+|
+}
+~
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+ 
diff --git a/ppocr/utils/export_model.py b/ppocr/utils/export_model.py
new file mode 100644
index 0000000..e5edb04
--- /dev/null
+++ b/ppocr/utils/export_model.py
@@ -0,0 +1,547 @@
+# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import yaml
+import json
+import copy
+import shutil
+import paddle
+import paddle.nn as nn
+from paddle.jit import to_static
+
+from collections import OrderedDict
+from packaging import version
+from argparse import ArgumentParser, RawDescriptionHelpFormatter
+from ppocr.modeling.architectures import build_model
+from ppocr.postprocess import build_post_process
+from ppocr.utils.save_load import load_model
+from ppocr.utils.logging import get_logger
+
+
+def represent_dictionary_order(self, dict_data):
+    return self.represent_mapping("tag:yaml.org,2002:map", dict_data.items())
+
+
+def setup_orderdict():
+    yaml.add_representer(OrderedDict, represent_dictionary_order)
+
+
+def dump_infer_config(config, path, logger):
+    setup_orderdict()
+    infer_cfg = OrderedDict()
+    if not os.path.exists(os.path.dirname(path)):
+        os.makedirs(os.path.dirname(path))
+    model_name = None
+    if config["Global"].get("model_name", None):
+        model_name = config["Global"]["model_name"]
+        infer_cfg["Global"] = {"model_name": model_name}
+    if config["Global"].get("uniform_output_enabled", True):
+        arch_config = config["Architecture"]
+        if arch_config["algorithm"] in ["SVTR_LCNet", "SVTR_HGNet"]:
+            common_dynamic_shapes = {
+                "x": [[1, 3, 48, 160], [1, 3, 48, 320], [8, 3, 48, 3200]]
+            }
+        elif arch_config["model_type"] == "det":
+            common_dynamic_shapes = {
+                "x": [[1, 3, 32, 32], [1, 3, 736, 736], [1, 3, 4000, 4000]]
+            }
+        elif arch_config["algorithm"] == "SLANet":
+            if model_name == "SLANet_plus":
+                common_dynamic_shapes = {
+                    "x": [[1, 3, 32, 32], [1, 3, 64, 448], [1, 3, 488, 488]]
+                }
+            else:
+                common_dynamic_shapes = {
+                    "x": [[1, 3, 32, 32], [1, 3, 64, 448], [8, 3, 488, 488]]
+                }
+        elif arch_config["algorithm"] == "SLANeXt":
+            common_dynamic_shapes = {
+                "x": [[1, 3, 512, 512], [1, 3, 512, 512], [1, 3, 512, 512]]
+            }
+        elif arch_config["algorithm"] == "LaTeXOCR":
+            common_dynamic_shapes = {
+                "x": [[1, 1, 32, 32], [1, 1, 64, 448], [1, 1, 192, 672]]
+            }
+        elif arch_config["algorithm"] == "UniMERNet":
+            common_dynamic_shapes = {
+                "x": [[1, 1, 192, 672], [1, 1, 192, 672], [8, 1, 192, 672]]
+            }
+        elif arch_config["algorithm"] in ["PP-FormulaNet-L", "PP-FormulaNet_plus-L"]:
+            common_dynamic_shapes = {
+                "x": [[1, 1, 768, 768], [1, 1, 768, 768], [8, 1, 768, 768]]
+            }
+        elif arch_config["algorithm"] in [
+            "PP-FormulaNet-S",
+            "PP-FormulaNet_plus-S",
+            "PP-FormulaNet_plus-M",
+        ]:
+            common_dynamic_shapes = {
+                "x": [[1, 1, 384, 384], [1, 1, 384, 384], [8, 1, 384, 384]]
+            }
+        else:
+            common_dynamic_shapes = None
+
+        backend_keys = ["paddle_infer", "tensorrt"]
+        hpi_config = {
+            "backend_configs": {
+                key: {
+                    (
+                        "dynamic_shapes" if key == "tensorrt" else "trt_dynamic_shapes"
+                    ): common_dynamic_shapes
+                }
+                for key in backend_keys
+            }
+        }
+        if common_dynamic_shapes:
+            infer_cfg["Hpi"] = hpi_config
+
+    infer_cfg["PreProcess"] = {"transform_ops": config["Eval"]["dataset"]["transforms"]}
+    postprocess = OrderedDict()
+    for k, v in config["PostProcess"].items():
+        if config["Architecture"].get("algorithm") in [
+            "LaTeXOCR",
+            "UniMERNet",
+            "PP-FormulaNet-L",
+            "PP-FormulaNet-S",
+            "PP-FormulaNet_plus-L",
+            "PP-FormulaNet_plus-M",
+            "PP-FormulaNet_plus-S",
+        ]:
+            if k != "rec_char_dict_path":
+                postprocess[k] = v
+        else:
+            postprocess[k] = v
+
+    if config["Architecture"].get("algorithm") in ["LaTeXOCR"]:
+        tokenizer_file = config["Global"].get("rec_char_dict_path")
+        if tokenizer_file is not None:
+            with open(tokenizer_file, encoding="utf-8") as tokenizer_config_handle:
+                character_dict = json.load(tokenizer_config_handle)
+                postprocess["character_dict"] = character_dict
+    elif config["Architecture"].get("algorithm") in [
+        "UniMERNet",
+        "PP-FormulaNet-L",
+        "PP-FormulaNet-S",
+        "PP-FormulaNet_plus-L",
+        "PP-FormulaNet_plus-M",
+        "PP-FormulaNet_plus-S",
+    ]:
+        tokenizer_file = config["Global"].get("rec_char_dict_path")
+        fast_tokenizer_file = os.path.join(tokenizer_file, "tokenizer.json")
+        tokenizer_config_file = os.path.join(tokenizer_file, "tokenizer_config.json")
+        postprocess["character_dict"] = {}
+        if fast_tokenizer_file is not None:
+            with open(fast_tokenizer_file, encoding="utf-8") as tokenizer_config_handle:
+                character_dict = json.load(tokenizer_config_handle)
+                postprocess["character_dict"]["fast_tokenizer_file"] = character_dict
+        if tokenizer_config_file is not None:
+            with open(
+                tokenizer_config_file, encoding="utf-8"
+            ) as tokenizer_config_handle:
+                character_dict = json.load(tokenizer_config_handle)
+                postprocess["character_dict"]["tokenizer_config_file"] = character_dict
+    else:
+        if config["Global"].get("character_dict_path") is not None:
+            with open(config["Global"]["character_dict_path"], encoding="utf-8") as f:
+                lines = f.readlines()
+                character_dict = [line.strip("\n") for line in lines]
+            postprocess["character_dict"] = character_dict
+
+    infer_cfg["PostProcess"] = postprocess
+
+    with open(path, "w", encoding="utf-8") as f:
+        yaml.dump(infer_cfg, f, default_flow_style=False, allow_unicode=True)
+    logger.info("Export inference config file to {}".format(os.path.join(path)))
+
+
+def dynamic_to_static(model, arch_config, logger, input_shape=None):
+    if arch_config["algorithm"] == "SRN":
+        max_text_length = arch_config["Head"]["max_text_length"]
+        other_shape = [
+            paddle.static.InputSpec(shape=[None, 1, 64, 256], dtype="float32"),
+            [
+                paddle.static.InputSpec(shape=[None, 256, 1], dtype="int64"),
+                paddle.static.InputSpec(
+                    shape=[None, max_text_length, 1], dtype="int64"
+                ),
+                paddle.static.InputSpec(
+                    shape=[None, 8, max_text_length, max_text_length], dtype="int64"
+                ),
+                paddle.static.InputSpec(
+                    shape=[None, 8, max_text_length, max_text_length], dtype="int64"
+                ),
+            ],
+        ]
+        model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] == "SAR":
+        other_shape = [
+            paddle.static.InputSpec(shape=[None, 3, 48, 160], dtype="float32"),
+            [paddle.static.InputSpec(shape=[None], dtype="float32")],
+        ]
+        model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] in ["SVTR_LCNet", "SVTR_HGNet"]:
+        other_shape = [
+            paddle.static.InputSpec(shape=[None, 3, 48, -1], dtype="float32"),
+        ]
+        model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] in ["SVTR", "CPPD"]:
+        other_shape = [
+            paddle.static.InputSpec(shape=[None] + input_shape, dtype="float32"),
+        ]
+        model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] == "PREN":
+        other_shape = [
+            paddle.static.InputSpec(shape=[None, 3, 64, 256], dtype="float32"),
+        ]
+        model = to_static(model, input_spec=other_shape)
+    elif arch_config["model_type"] == "sr":
+        other_shape = [
+            paddle.static.InputSpec(shape=[None, 3, 16, 64], dtype="float32")
+        ]
+        model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] == "ViTSTR":
+        other_shape = [
+            paddle.static.InputSpec(shape=[None, 1, 224, 224], dtype="float32"),
+        ]
+        model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] == "ABINet":
+        if not input_shape:
+            input_shape = [3, 32, 128]
+        other_shape = [
+            paddle.static.InputSpec(shape=[None] + input_shape, dtype="float32"),
+        ]
+        model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] in ["NRTR", "SPIN", "RFL"]:
+        other_shape = [
+            paddle.static.InputSpec(shape=[None, 1, 32, 100], dtype="float32"),
+        ]
+        model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] in ["SATRN"]:
+        other_shape = [
+            paddle.static.InputSpec(shape=[None, 3, 32, 100], dtype="float32"),
+        ]
+        model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] == "VisionLAN":
+        other_shape = [
+            paddle.static.InputSpec(shape=[None, 3, 64, 256], dtype="float32"),
+        ]
+        model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] == "RobustScanner":
+        max_text_length = arch_config["Head"]["max_text_length"]
+        other_shape = [
+            paddle.static.InputSpec(shape=[None, 3, 48, 160], dtype="float32"),
+            [
+                paddle.static.InputSpec(
+                    shape=[
+                        None,
+                    ],
+                    dtype="float32",
+                ),
+                paddle.static.InputSpec(shape=[None, max_text_length], dtype="int64"),
+            ],
+        ]
+        model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] == "CAN":
+        other_shape = [
+            [
+                paddle.static.InputSpec(shape=[None, 1, None, None], dtype="float32"),
+                paddle.static.InputSpec(shape=[None, 1, None, None], dtype="float32"),
+                paddle.static.InputSpec(
+                    shape=[None, arch_config["Head"]["max_text_length"]], dtype="int64"
+                ),
+            ]
+        ]
+        model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] == "LaTeXOCR":
+        other_shape = [
+            paddle.static.InputSpec(shape=[None, 1, None, None], dtype="float32"),
+        ]
+        model = to_static(model, input_spec=other_shape)
+    elif arch_config["algorithm"] == "UniMERNet":
+        model = paddle.jit.to_static(
+            model,
+            input_spec=[
+                paddle.static.InputSpec(shape=[-1, 1, 192, 672], dtype="float32")
+            ],
+            full_graph=True,
+        )
+    elif arch_config["algorithm"] == "SLANeXt":
+        model = paddle.jit.to_static(
+            model,
+            input_spec=[
+                paddle.static.InputSpec(shape=[-1, 3, 512, 512], dtype="float32")
+            ],
+            full_graph=True,
+        )
+    elif arch_config["algorithm"] in ["PP-FormulaNet-L", "PP-FormulaNet_plus-L"]:
+        model = paddle.jit.to_static(
+            model,
+            input_spec=[
+                paddle.static.InputSpec(shape=[-1, 1, 768, 768], dtype="float32")
+            ],
+            full_graph=True,
+        )
+    elif arch_config["algorithm"] in [
+        "PP-FormulaNet-S",
+        "PP-FormulaNet_plus-S",
+        "PP-FormulaNet_plus-M",
+    ]:
+        model = paddle.jit.to_static(
+            model,
+            input_spec=[
+                paddle.static.InputSpec(shape=[-1, 1, 384, 384], dtype="float32")
+            ],
+            full_graph=True,
+        )
+
+    elif arch_config["algorithm"] in ["LayoutLM", "LayoutLMv2", "LayoutXLM"]:
+        input_spec = [
+            paddle.static.InputSpec(shape=[None, 512], dtype="int64"),  # input_ids
+            paddle.static.InputSpec(shape=[None, 512, 4], dtype="int64"),  # bbox
+            paddle.static.InputSpec(shape=[None, 512], dtype="int64"),  # attention_mask
+            paddle.static.InputSpec(shape=[None, 512], dtype="int64"),  # token_type_ids
+            paddle.static.InputSpec(shape=[None, 3, 224, 224], dtype="int64"),  # image
+        ]
+        if "Re" in arch_config["Backbone"]["name"]:
+            input_spec.extend(
+                [
+                    paddle.static.InputSpec(
+                        shape=[None, 512, 3], dtype="int64"
+                    ),  # entities
+                    paddle.static.InputSpec(
+                        shape=[None, None, 2], dtype="int64"
+                    ),  # relations
+                ]
+            )
+        if model.backbone.use_visual_backbone is False:
+            input_spec.pop(4)
+        model = to_static(model, input_spec=[input_spec])
+    else:
+        infer_shape = [3, -1, -1]
+        if arch_config["model_type"] == "rec":
+            infer_shape = [3, 32, -1]  # for rec model, H must be 32
+            if (
+                "Transform" in arch_config
+                and arch_config["Transform"] is not None
+                and arch_config["Transform"]["name"] == "TPS"
+            ):
+                logger.info(
+                    "When there is tps in the network, variable length input is not supported, and the input size needs to be the same as during training"
+                )
+                infer_shape[-1] = 100
+        elif arch_config["model_type"] == "table":
+            infer_shape = [3, 488, 488]
+            if arch_config["algorithm"] == "TableMaster":
+                infer_shape = [3, 480, 480]
+            if arch_config["algorithm"] == "SLANet":
+                infer_shape = [3, -1, -1]
+        model = to_static(
+            model,
+            input_spec=[
+                paddle.static.InputSpec(shape=[None] + infer_shape, dtype="float32")
+            ],
+        )
+
+    if arch_config["model_type"] != "sr":
+        for layer in model.sublayers():
+            if hasattr(layer, "rep") and not getattr(layer, "is_repped", False):
+                layer.rep()
+    return model
+
+
+def export_single_model(
+    model,
+    arch_config,
+    save_path,
+    logger,
+    yaml_path,
+    config,
+    input_shape=None,
+    quanter=None,
+):
+
+    model = dynamic_to_static(model, arch_config, logger, input_shape)
+
+    if quanter is None:
+        try:
+            import encryption  # Attempt to import the encryption module for AIStudio's encryption model
+        except (
+            ModuleNotFoundError
+        ):  # Encryption is not needed if the module cannot be imported
+            print("Skipping import of the encryption module")
+        paddle_version = version.parse(paddle.__version__)
+        if config["Global"].get("export_with_pir", True):
+            assert (
+                paddle_version >= version.parse("3.0.0b2")
+                or paddle_version == version.parse("0.0.0")
+            ) and os.environ.get("FLAGS_enable_pir_api", None) not in ["0", "False"]
+            paddle.jit.save(model, save_path)
+        else:
+            if paddle_version >= version.parse(
+                "3.0.0b2"
+            ) or paddle_version == version.parse("0.0.0"):
+                model.forward.rollback()
+                with paddle.pir_utils.OldIrGuard():
+                    model = dynamic_to_static(model, arch_config, logger, input_shape)
+                    paddle.jit.save(model, save_path)
+            else:
+                paddle.jit.save(model, save_path)
+    else:
+        quanter.save_quantized_model(model, save_path)
+    logger.info("inference model is saved to {}".format(save_path))
+    return
+
+
+def convert_bn(model):
+    for n, m in model.named_children():
+        if isinstance(m, nn.SyncBatchNorm):
+            bn = nn.BatchNorm2D(
+                m._num_features, m._momentum, m._epsilon, m._weight_attr, m._bias_attr
+            )
+            bn.set_dict(m.state_dict())
+            setattr(model, n, bn)
+        else:
+            convert_bn(m)
+
+
+def export(config, base_model=None, save_path=None):
+    if paddle.distributed.get_rank() != 0:
+        return
+    logger = get_logger()
+    # build post process
+    post_process_class = build_post_process(config["PostProcess"], config["Global"])
+
+    # build model
+    # for rec algorithm
+    if hasattr(post_process_class, "character"):
+        char_num = len(getattr(post_process_class, "character"))
+        if config["Architecture"]["algorithm"] in [
+            "Distillation",
+        ]:  # distillation model
+            for key in config["Architecture"]["Models"]:
+                if (
+                    config["Architecture"]["Models"][key]["Head"]["name"] == "MultiHead"
+                ):  # multi head
+                    out_channels_list = {}
+                    if config["PostProcess"]["name"] == "DistillationSARLabelDecode":
+                        char_num = char_num - 2
+                    if config["PostProcess"]["name"] == "DistillationNRTRLabelDecode":
+                        char_num = char_num - 3
+                    out_channels_list["CTCLabelDecode"] = char_num
+                    out_channels_list["SARLabelDecode"] = char_num + 2
+                    out_channels_list["NRTRLabelDecode"] = char_num + 3
+                    config["Architecture"]["Models"][key]["Head"][
+                        "out_channels_list"
+                    ] = out_channels_list
+                else:
+                    config["Architecture"]["Models"][key]["Head"][
+                        "out_channels"
+                    ] = char_num
+                # just one final tensor needs to exported for inference
+                config["Architecture"]["Models"][key]["return_all_feats"] = False
+        elif config["Architecture"]["Head"]["name"] == "MultiHead":  # multi head
+            out_channels_list = {}
+            char_num = len(getattr(post_process_class, "character"))
+            if config["PostProcess"]["name"] == "SARLabelDecode":
+                char_num = char_num - 2
+            if config["PostProcess"]["name"] == "NRTRLabelDecode":
+                char_num = char_num - 3
+            out_channels_list["CTCLabelDecode"] = char_num
+            out_channels_list["SARLabelDecode"] = char_num + 2
+            out_channels_list["NRTRLabelDecode"] = char_num + 3
+            config["Architecture"]["Head"]["out_channels_list"] = out_channels_list
+        else:  # base rec model
+            config["Architecture"]["Head"]["out_channels"] = char_num
+
+    # for sr algorithm
+    if config["Architecture"]["model_type"] == "sr":
+        config["Architecture"]["Transform"]["infer_mode"] = True
+
+    # for latexocr algorithm
+    if config["Architecture"].get("algorithm") in ["LaTeXOCR"]:
+        config["Architecture"]["Backbone"]["is_predict"] = True
+        config["Architecture"]["Backbone"]["is_export"] = True
+        config["Architecture"]["Head"]["is_export"] = True
+    if config["Architecture"].get("algorithm") in ["UniMERNet"]:
+        config["Architecture"]["Backbone"]["is_export"] = True
+        config["Architecture"]["Head"]["is_export"] = True
+    if config["Architecture"].get("algorithm") in [
+        "PP-FormulaNet-S",
+        "PP-FormulaNet-L",
+        "PP-FormulaNet_plus-S",
+        "PP-FormulaNet_plus-M",
+        "PP-FormulaNet_plus-L",
+    ]:
+        config["Architecture"]["Head"]["is_export"] = True
+    if base_model is not None:
+        model = base_model
+        if isinstance(model, paddle.DataParallel):
+            model = copy.deepcopy(model._layers)
+        else:
+            model = copy.deepcopy(model)
+    else:
+        model = build_model(config["Architecture"])
+        load_model(config, model, model_type=config["Architecture"]["model_type"])
+    convert_bn(model)
+    model.eval()
+
+    if not save_path:
+        save_path = config["Global"]["save_inference_dir"]
+    yaml_path = os.path.join(save_path, "inference.yml")
+
+    arch_config = config["Architecture"]
+
+    if (
+        arch_config["algorithm"] in ["SVTR", "CPPD"]
+        and arch_config["Head"]["name"] != "MultiHead"
+    ):
+        input_shape = config["Eval"]["dataset"]["transforms"][-2]["SVTRRecResizeImg"][
+            "image_shape"
+        ]
+    elif arch_config["algorithm"].lower() == "ABINet".lower():
+        rec_rs = [
+            c
+            for c in config["Eval"]["dataset"]["transforms"]
+            if "ABINetRecResizeImg" in c
+        ]
+        input_shape = rec_rs[0]["ABINetRecResizeImg"]["image_shape"] if rec_rs else None
+    else:
+        input_shape = None
+    dump_infer_config(config, yaml_path, logger)
+    if arch_config["algorithm"] in [
+        "Distillation",
+    ]:  # distillation model
+        archs = list(arch_config["Models"].values())
+        for idx, name in enumerate(model.model_name_list):
+            sub_model_save_path = os.path.join(save_path, name, "inference")
+            export_single_model(
+                model.model_list[idx],
+                archs[idx],
+                sub_model_save_path,
+                logger,
+                yaml_path,
+                config,
+            )
+    else:
+        save_path = os.path.join(save_path, "inference")
+        export_single_model(
+            model,
+            arch_config,
+            save_path,
+            logger,
+            yaml_path,
+            config,
+            input_shape=input_shape,
+        )
diff --git a/ppocr/utils/formula_utils/math_txt2pkl.py b/ppocr/utils/formula_utils/math_txt2pkl.py
new file mode 100644
index 0000000..beafc13
--- /dev/null
+++ b/ppocr/utils/formula_utils/math_txt2pkl.py
@@ -0,0 +1,74 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pickle
+from tqdm import tqdm
+import os
+import math
+from paddle.utils import try_import
+from collections import defaultdict
+import glob
+from os.path import join
+import argparse
+
+
+def txt2pickle(images, equations, save_dir):
+    imagesize = try_import("imagesize")
+    save_p = os.path.join(save_dir, "latexocr_{}.pkl".format(images.split("/")[-1]))
+    min_dimensions = (32, 32)
+    max_dimensions = (672, 192)
+    max_length = 512
+    data = defaultdict(lambda: [])
+    if images is not None and equations is not None:
+        images_list = [
+            path.replace("\\", "/") for path in glob.glob(join(images, "*.png"))
+        ]
+        indices = [int(os.path.basename(img).split(".")[0]) for img in images_list]
+        eqs = open(equations, "r").read().split("\n")
+        for i, im in tqdm(enumerate(images_list), total=len(images_list)):
+            width, height = imagesize.get(im)
+            if (
+                min_dimensions[0] <= width <= max_dimensions[0]
+                and min_dimensions[1] <= height <= max_dimensions[1]
+            ):
+                divide_h = math.ceil(height / 16) * 16
+                divide_w = math.ceil(width / 16) * 16
+                im = os.path.basename(im)
+                data[(divide_w, divide_h)].append((eqs[indices[i]], im))
+        data = dict(data)
+        with open(save_p, "wb") as file:
+            pickle.dump(data, file)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--image_dir",
+        type=str,
+        default=".",
+        help="Input_label or input path to be converted",
+    )
+    parser.add_argument(
+        "--mathtxt_path",
+        type=str,
+        default=".",
+        help="Input_label or input path to be converted",
+    )
+    parser.add_argument(
+        "--output_dir", type=str, default="out_label.txt", help="Output file name"
+    )
+
+    args = parser.parse_args()
+    txt2pickle(args.image_dir, args.mathtxt_path, args.output_dir)
diff --git a/ppocr/utils/formula_utils/unimernet_data_convert.py b/ppocr/utils/formula_utils/unimernet_data_convert.py
new file mode 100644
index 0000000..fed8a7f
--- /dev/null
+++ b/ppocr/utils/formula_utils/unimernet_data_convert.py
@@ -0,0 +1,107 @@
+# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import cv2
+import glob
+import argparse
+from os.path import join
+from tqdm import tqdm
+
+
+def latexocr2paddleocr_train(image_path, math_unimernet_file, math_hwe_file, save_path):
+    convert_f = open(save_path, "w")
+    sub_dir = "UniMER-1M/images"
+    img_sub_dir = os.path.join(image_path, sub_dir)
+    with open(math_unimernet_file, "r") as f:
+        lines = f.readlines()
+        formula_num = len(lines)
+        for i, line in tqdm(enumerate(lines), total=formula_num):
+            image_name = "{0:07d}.png".format(i)
+            math_gt = line.strip()
+            image_p = os.path.join(img_sub_dir, image_name)
+            img_name_subdir = os.path.join(sub_dir, image_name)
+            if os.path.exists(image_p):
+                convert_f.writelines("{}\t{}\n".format(img_name_subdir, math_gt))
+
+    sub_dir = "HME100K/train_images"
+    img_sub_dir = os.path.join(image_path, sub_dir)
+    with open(math_hwe_file, "r") as f:
+        lines = f.readlines()
+        formula_num = len(lines)
+        for i, line in tqdm(enumerate(lines), total=formula_num):
+            img_name, math_gt = line.strip().split("\t")
+            image_path = os.path.join(img_sub_dir, img_name)
+            img_name_subdir = os.path.join(sub_dir, img_name)
+            convert_f.writelines("{}\t{}\n".format(img_name_subdir, math_gt))
+
+    convert_f.close()
+
+
+def unimernet2paddleocr_test(image_path, math_file, save_path):
+    convert_f = open(save_path, "w")
+    with open(math_file, "r") as f:
+        # load maths which
+        lines = f.readlines()
+        formula_num = len(lines)
+        for i, line in tqdm(enumerate(lines), total=formula_num):
+            image_name = "{0:07d}.png".format(i)
+            math_gt = line.strip()
+            image_p = os.path.join(image_path, image_name)
+            if os.path.exists(image_p):
+                convert_f.writelines("{}\t{}\n".format(image_name, math_gt))
+    convert_f.close()
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "--image_dir",
+        type=str,
+        default=".",
+        help="Input_label or input path to be converted",
+    )
+    parser.add_argument(
+        "--unimernet_txt_path",
+        type=str,
+        default="",
+        help="Input_label or input path to be converted",
+    )
+    parser.add_argument(
+        "--hme100k_txt_path",
+        type=str,
+        default="",
+        help="Input_label or input path to be converted",
+    )
+    parser.add_argument(
+        "--output_path", type=str, default="out_label.txt", help="Output file name"
+    )
+    parser.add_argument(
+        "--datatype", type=str, default="out_label.txt", help="datatype"
+    )
+    args = parser.parse_args()
+    if args.datatype == "unimernet_train":
+        latexocr2paddleocr_train(
+            args.image_dir,
+            args.unimernet_txt_path,
+            args.hme100k_txt_path,
+            args.output_path,
+        )
+    elif args.datatype == "unimernet_test":
+        unimernet2paddleocr_test(
+            args.image_dir, args.unimernet_txt_path, args.output_path
+        )
+    else:
+        raise NotImplementedError("the datatype is not supported")
diff --git a/ppocr/utils/gen_label.py b/ppocr/utils/gen_label.py
new file mode 100644
index 0000000..1634b38
--- /dev/null
+++ b/ppocr/utils/gen_label.py
@@ -0,0 +1,82 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import argparse
+import json
+
+
+def gen_rec_label(input_path, out_label):
+    with open(out_label, "w") as out_file:
+        with open(input_path, "r") as f:
+            for line in f.readlines():
+                tmp = line.strip("\n").replace(" ", "").split(",")
+                img_path, label = tmp[0], tmp[1]
+                label = label.replace('"', "")
+                out_file.write(img_path + "\t" + label + "\n")
+
+
+def gen_det_label(root_path, input_dir, out_label):
+    with open(out_label, "w") as out_file:
+        for label_file in os.listdir(input_dir):
+            img_path = os.path.join(root_path, label_file[3:-4] + ".jpg")
+            label = []
+            with open(
+                os.path.join(input_dir, label_file), "r", encoding="utf-8-sig"
+            ) as f:
+                for line in f.readlines():
+                    tmp = line.strip("\n\r").replace("\xef\xbb\xbf", "").split(",")
+                    points = tmp[:8]
+                    s = []
+                    for i in range(0, len(points), 2):
+                        b = points[i : i + 2]
+                        b = [int(t) for t in b]
+                        s.append(b)
+                    result = {"transcription": tmp[8], "points": s}
+                    label.append(result)
+
+            out_file.write(
+                img_path + "\t" + json.dumps(label, ensure_ascii=False) + "\n"
+            )
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--mode",
+        type=str,
+        default="rec",
+        help="Generate rec_label or det_label, can be set rec or det",
+    )
+    parser.add_argument(
+        "--root_path",
+        type=str,
+        default=".",
+        help="The root directory of images.Only takes effect when mode=det ",
+    )
+    parser.add_argument(
+        "--input_path",
+        type=str,
+        default=".",
+        help="Input_label or input path to be converted",
+    )
+    parser.add_argument(
+        "--output_label", type=str, default="out_label.txt", help="Output file name"
+    )
+
+    args = parser.parse_args()
+    if args.mode == "rec":
+        print("Generate rec label")
+        gen_rec_label(args.input_path, args.output_label)
+    elif args.mode == "det":
+        gen_det_label(args.root_path, args.input_path, args.output_label)
diff --git a/ppocr/utils/ic15_dict.txt b/ppocr/utils/ic15_dict.txt
new file mode 100644
index 0000000..7104368
--- /dev/null
+++ b/ppocr/utils/ic15_dict.txt
@@ -0,0 +1,36 @@
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
diff --git a/ppocr/utils/iou.py b/ppocr/utils/iou.py
new file mode 100644
index 0000000..cb77c34
--- /dev/null
+++ b/ppocr/utils/iou.py
@@ -0,0 +1,54 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/whai362/PSENet/blob/python3/models/loss/iou.py
+"""
+
+import paddle
+
+EPS = 1e-6
+
+
+def iou_single(a, b, mask, n_class):
+    valid = mask == 1
+    a = a.masked_select(valid)
+    b = b.masked_select(valid)
+    miou = []
+    for i in range(n_class):
+        if a.shape == [0] and a.shape == b.shape:
+            inter = paddle.to_tensor(0.0)
+            union = paddle.to_tensor(0.0)
+        else:
+            inter = ((a == i).logical_and(b == i)).astype("float32")
+            union = ((a == i).logical_or(b == i)).astype("float32")
+        miou.append(paddle.sum(inter) / (paddle.sum(union) + EPS))
+    miou = sum(miou) / len(miou)
+    return miou
+
+
+def iou(a, b, mask, n_class=2, reduce=True):
+    batch_size = a.shape[0]
+
+    a = a.reshape([batch_size, -1])
+    b = b.reshape([batch_size, -1])
+    mask = mask.reshape([batch_size, -1])
+
+    iou = paddle.zeros((batch_size,), dtype="float32")
+    for i in range(batch_size):
+        iou[i] = iou_single(a[i], b[i], mask[i], n_class)
+
+    if reduce:
+        iou = paddle.mean(iou)
+    return iou
diff --git a/ppocr/utils/loggers/__init__.py b/ppocr/utils/loggers/__init__.py
new file mode 100644
index 0000000..a73351f
--- /dev/null
+++ b/ppocr/utils/loggers/__init__.py
@@ -0,0 +1,2 @@
+from .wandb_logger import WandbLogger
+from .loggers import Loggers
diff --git a/ppocr/utils/loggers/base_logger.py b/ppocr/utils/loggers/base_logger.py
new file mode 100644
index 0000000..0aa132d
--- /dev/null
+++ b/ppocr/utils/loggers/base_logger.py
@@ -0,0 +1,16 @@
+import os
+from abc import ABC, abstractmethod
+
+
+class BaseLogger(ABC):
+    def __init__(self, save_dir):
+        self.save_dir = save_dir
+        os.makedirs(self.save_dir, exist_ok=True)
+
+    @abstractmethod
+    def log_metrics(self, metrics, prefix=None):
+        pass
+
+    @abstractmethod
+    def close(self):
+        pass
diff --git a/ppocr/utils/loggers/loggers.py b/ppocr/utils/loggers/loggers.py
new file mode 100644
index 0000000..a14dbcb
--- /dev/null
+++ b/ppocr/utils/loggers/loggers.py
@@ -0,0 +1,19 @@
+from .wandb_logger import WandbLogger
+
+
+class Loggers(object):
+    def __init__(self, loggers):
+        super().__init__()
+        self.loggers = loggers
+
+    def log_metrics(self, metrics, prefix=None, step=None):
+        for logger in self.loggers:
+            logger.log_metrics(metrics, prefix=prefix, step=step)
+
+    def log_model(self, is_best, prefix, metadata=None):
+        for logger in self.loggers:
+            logger.log_model(is_best=is_best, prefix=prefix, metadata=metadata)
+
+    def close(self):
+        for logger in self.loggers:
+            logger.close()
diff --git a/ppocr/utils/loggers/wandb_logger.py b/ppocr/utils/loggers/wandb_logger.py
new file mode 100644
index 0000000..44cbfc1
--- /dev/null
+++ b/ppocr/utils/loggers/wandb_logger.py
@@ -0,0 +1,84 @@
+import os
+from .base_logger import BaseLogger
+from ppocr.utils.logging import get_logger
+
+
+class WandbLogger(BaseLogger):
+    def __init__(
+        self,
+        project=None,
+        name=None,
+        id=None,
+        entity=None,
+        save_dir=None,
+        config=None,
+        **kwargs,
+    ):
+        try:
+            import wandb
+
+            self.wandb = wandb
+        except ModuleNotFoundError:
+            raise ModuleNotFoundError("Please install wandb using `pip install wandb`")
+
+        self.project = project
+        self.name = name
+        self.id = id
+        self.save_dir = save_dir
+        self.config = config
+        self.kwargs = kwargs
+        self.entity = entity
+        self._run = None
+        self._wandb_init = dict(
+            project=self.project,
+            name=self.name,
+            id=self.id,
+            entity=self.entity,
+            dir=self.save_dir,
+            resume="allow",
+        )
+        self._wandb_init.update(**kwargs)
+        self.logger = get_logger()
+
+        _ = self.run
+
+        if self.config:
+            self.run.config.update(self.config)
+
+    @property
+    def run(self):
+        if self._run is None:
+            if self.wandb.run is not None:
+                self.logger.info(
+                    "There is a wandb run already in progress "
+                    "and newly created instances of `WandbLogger` will reuse"
+                    " this run. If this is not desired, call `wandb.finish()`"
+                    "before instantiating `WandbLogger`."
+                )
+                self._run = self.wandb.run
+            else:
+                self._run = self.wandb.init(**self._wandb_init)
+        return self._run
+
+    def log_metrics(self, metrics, prefix=None, step=None):
+        if not prefix:
+            prefix = ""
+        updated_metrics = {prefix.lower() + "/" + k: v for k, v in metrics.items()}
+
+        self.run.log(updated_metrics, step=step)
+
+    def log_model(self, is_best, prefix, metadata=None):
+        model_path = os.path.join(self.save_dir, prefix + ".pdparams")
+        artifact = self.wandb.Artifact(
+            "model-{}".format(self.run.id), type="model", metadata=metadata
+        )
+        artifact.add_file(model_path, name="model_ckpt.pdparams")
+
+        aliases = [prefix]
+        if is_best:
+            aliases.append("best")
+
+        self.run.log_artifact(artifact, aliases=aliases)
+
+    def close(self):
+        self.run.finish()
diff --git a/ppocr/utils/logging.py b/ppocr/utils/logging.py
new file mode 100644
index 0000000..c460936
--- /dev/null
+++ b/ppocr/utils/logging.py
@@ -0,0 +1,78 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/WenmuZhou/PytorchOCR/blob/master/torchocr/utils/logging.py
+"""
+
+import os
+import sys
+import logging
+import functools
+import paddle.distributed as dist
+
+logger_initialized = {}
+
+
+@functools.lru_cache()
+def get_logger(name="ppocr", log_file=None, log_level=logging.DEBUG, log_ranks="0"):
+    """Initialize and get a logger by name.
+    If the logger has not been initialized, this method will initialize the
+    logger by adding one or two handlers, otherwise the initialized logger will
+    be directly returned. During initialization, a StreamHandler will always be
+    added. If `log_file` is specified a FileHandler will also be added.
+    Args:
+        name (str): Logger name.
+        log_file (str | None): The log filename. If specified, a FileHandler
+            will be added to the logger.
+        log_level (int): The logger level. Note that only the process of
+            rank 0 is affected, and other processes will set the level to
+            "Error" thus be silent most of the time.
+        log_ranks (str): The ids of gpu to log which are separated by "," when more than 1, "0" by default.
+    Returns:
+        logging.Logger: The expected logger.
+    """
+    logger = logging.getLogger(name)
+    if name in logger_initialized:
+        return logger
+    for logger_name in logger_initialized:
+        if name.startswith(logger_name):
+            return logger
+
+    formatter = logging.Formatter(
+        "[%(asctime)s] %(name)s %(levelname)s: %(message)s", datefmt="%Y/%m/%d %H:%M:%S"
+    )
+
+    stream_handler = logging.StreamHandler(stream=sys.stdout)
+    stream_handler.setFormatter(formatter)
+    logger.addHandler(stream_handler)
+    if log_file is not None and dist.get_rank() == 0:
+        log_file_folder = os.path.split(log_file)[0]
+        os.makedirs(log_file_folder, exist_ok=True)
+        file_handler = logging.FileHandler(log_file, "a")
+        file_handler.setFormatter(formatter)
+        logger.addHandler(file_handler)
+
+    if isinstance(log_ranks, str):
+        log_ranks = [int(i) for i in log_ranks.split(",")]
+    elif isinstance(log_ranks, int):
+        log_ranks = [log_ranks]
+
+    if dist.get_rank() in log_ranks:
+        logger.setLevel(log_level)
+    else:
+        logger.setLevel(logging.ERROR)
+    logger_initialized[name] = True
+    logger.propagate = False
+    return logger
diff --git a/ppocr/utils/network.py b/ppocr/utils/network.py
new file mode 100644
index 0000000..761338a
--- /dev/null
+++ b/ppocr/utils/network.py
@@ -0,0 +1,155 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import sys
+import time
+import shutil
+import tarfile
+import requests
+import os.path as osp
+import paddle.distributed as dist
+from tqdm import tqdm
+
+from ppocr.utils.logging import get_logger
+
+MODELS_DIR = os.path.join(
+    os.environ.get("PADDLE_OCR_BASE_DIR", os.path.expanduser("~/.paddleocr/")), "models"
+)
+DOWNLOAD_RETRY_LIMIT = 3
+
+
+def download_with_progressbar(url, save_path):
+    logger = get_logger()
+    if save_path and os.path.exists(save_path):
+        logger.info(f"Path {save_path} already exists. Skipping...")
+        return
+    else:
+        # Mainly used to solve the problem of downloading data from different
+        # machines in the case of multiple machines. Different nodes will download
+        # data, and the same node will only download data once.
+        if dist.get_rank() == 0:
+            _download(url, save_path)
+        else:
+            while not os.path.exists(save_path):
+                time.sleep(1)
+
+
+def _download(url, save_path):
+    """
+    Download from url, save to path.
+
+    url (str): download url
+    save_path (str): download to given path
+    """
+    logger = get_logger()
+
+    fname = osp.split(url)[-1]
+    retry_cnt = 0
+
+    while not osp.exists(save_path):
+        if retry_cnt < DOWNLOAD_RETRY_LIMIT:
+            retry_cnt += 1
+        else:
+            raise RuntimeError(
+                "Download from {} failed. " "Retry limit reached".format(url)
+            )
+
+        try:
+            req = requests.get(url, stream=True)
+        except Exception as e:  # requests.exceptions.ConnectionError
+            logger.info(
+                "Downloading {} from {} failed {} times with exception {}".format(
+                    fname, url, retry_cnt + 1, str(e)
+                )
+            )
+            time.sleep(1)
+            continue
+
+        if req.status_code != 200:
+            raise RuntimeError(
+                "Downloading from {} failed with code "
+                "{}!".format(url, req.status_code)
+            )
+
+        # For protecting download interrupted, download to
+        # tmp_file firstly, move tmp_file to save_path
+        # after download finished
+        tmp_file = save_path + ".tmp"
+        total_size = req.headers.get("content-length")
+        with open(tmp_file, "wb") as f:
+            if total_size:
+                with tqdm(total=(int(total_size) + 1023) // 1024) as pbar:
+                    for chunk in req.iter_content(chunk_size=1024):
+                        f.write(chunk)
+                        pbar.update(1)
+            else:
+                for chunk in req.iter_content(chunk_size=1024):
+                    if chunk:
+                        f.write(chunk)
+        shutil.move(tmp_file, save_path)
+
+    return save_path
+
+
+def maybe_download(model_storage_directory, url):
+    # using custom model
+    tar_file_name_list = [".pdiparams", ".pdiparams.info", ".pdmodel"]
+    if not os.path.exists(
+        os.path.join(model_storage_directory, "inference.pdiparams")
+    ) or not os.path.exists(os.path.join(model_storage_directory, "inference.pdmodel")):
+        assert url.endswith(".tar"), "Only supports tar compressed package"
+        tmp_path = os.path.join(model_storage_directory, url.split("/")[-1])
+        print("download {} to {}".format(url, tmp_path))
+        os.makedirs(model_storage_directory, exist_ok=True)
+        download_with_progressbar(url, tmp_path)
+        with tarfile.open(tmp_path, "r") as tarObj:
+            for member in tarObj.getmembers():
+                filename = None
+                for tar_file_name in tar_file_name_list:
+                    if member.name.endswith(tar_file_name):
+                        filename = "inference" + tar_file_name
+                if filename is None:
+                    continue
+                file = tarObj.extractfile(member)
+                with open(os.path.join(model_storage_directory, filename), "wb") as f:
+                    f.write(file.read())
+        os.remove(tmp_path)
+
+
+def maybe_download_params(model_path):
+    if os.path.exists(model_path) or not is_link(model_path):
+        return model_path
+    else:
+        url = model_path
+    tmp_path = os.path.join(MODELS_DIR, url.split("/")[-1])
+    print("download {} to {}".format(url, tmp_path))
+    os.makedirs(MODELS_DIR, exist_ok=True)
+    download_with_progressbar(url, tmp_path)
+    return tmp_path
+
+
+def is_link(s):
+    return s is not None and s.startswith("http")
+
+
+def confirm_model_dir_url(model_dir, default_model_dir, default_url):
+    url = default_url
+    if model_dir is None or is_link(model_dir):
+        if is_link(model_dir):
+            url = model_dir
+        file_name = url.split("/")[-1][:-4]
+        model_dir = default_model_dir
+        model_dir = os.path.join(model_dir, file_name)
+    return model_dir, url
diff --git a/ppocr/utils/poly_nms.py b/ppocr/utils/poly_nms.py
new file mode 100644
index 0000000..c3a1338
--- /dev/null
+++ b/ppocr/utils/poly_nms.py
@@ -0,0 +1,146 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import numpy as np
+from shapely.geometry import Polygon
+
+
+def points2polygon(points):
+    """Convert k points to 1 polygon.
+
+    Args:
+        points (ndarray or list): A ndarray or a list of shape (2k)
+            that indicates k points.
+
+    Returns:
+        polygon (Polygon): A polygon object.
+    """
+    if isinstance(points, list):
+        points = np.array(points)
+
+    assert isinstance(points, np.ndarray)
+    assert (points.size % 2 == 0) and (points.size >= 8)
+
+    point_mat = points.reshape([-1, 2])
+    return Polygon(point_mat)
+
+
+def poly_intersection(poly_det, poly_gt, buffer=0.0001):
+    """Calculate the intersection area between two polygon.
+
+    Args:
+        poly_det (Polygon): A polygon predicted by detector.
+        poly_gt (Polygon): A gt polygon.
+
+    Returns:
+        intersection_area (float): The intersection area between two polygons.
+    """
+    assert isinstance(poly_det, Polygon)
+    assert isinstance(poly_gt, Polygon)
+
+    if buffer == 0:
+        poly_inter = poly_det & poly_gt
+    else:
+        poly_inter = poly_det.buffer(buffer) & poly_gt.buffer(buffer)
+    return poly_inter.area, poly_inter
+
+
+def poly_union(poly_det, poly_gt):
+    """Calculate the union area between two polygon.
+
+    Args:
+        poly_det (Polygon): A polygon predicted by detector.
+        poly_gt (Polygon): A gt polygon.
+
+    Returns:
+        union_area (float): The union area between two polygons.
+    """
+    assert isinstance(poly_det, Polygon)
+    assert isinstance(poly_gt, Polygon)
+
+    area_det = poly_det.area
+    area_gt = poly_gt.area
+    area_inters, _ = poly_intersection(poly_det, poly_gt)
+    return area_det + area_gt - area_inters
+
+
+def valid_boundary(x, with_score=True):
+    num = len(x)
+    if num < 8:
+        return False
+    if num % 2 == 0 and (not with_score):
+        return True
+    if num % 2 == 1 and with_score:
+        return True
+
+    return False
+
+
+def boundary_iou(src, target):
+    """Calculate the IOU between two boundaries.
+
+    Args:
+       src (list): Source boundary.
+       target (list): Target boundary.
+
+    Returns:
+       iou (float): The iou between two boundaries.
+    """
+    assert valid_boundary(src, False)
+    assert valid_boundary(target, False)
+    src_poly = points2polygon(src)
+    target_poly = points2polygon(target)
+
+    return poly_iou(src_poly, target_poly)
+
+
+def poly_iou(poly_det, poly_gt):
+    """Calculate the IOU between two polygons.
+
+    Args:
+        poly_det (Polygon): A polygon predicted by detector.
+        poly_gt (Polygon): A gt polygon.
+
+    Returns:
+        iou (float): The IOU between two polygons.
+    """
+    assert isinstance(poly_det, Polygon)
+    assert isinstance(poly_gt, Polygon)
+    area_inters, _ = poly_intersection(poly_det, poly_gt)
+    area_union = poly_union(poly_det, poly_gt)
+    if area_union == 0:
+        return 0.0
+    return area_inters / area_union
+
+
+def poly_nms(polygons, threshold):
+    assert isinstance(polygons, list)
+
+    polygons = np.array(sorted(polygons, key=lambda x: x[-1]))
+
+    keep_poly = []
+    index = [i for i in range(polygons.shape[0])]
+
+    while len(index) > 0:
+        keep_poly.append(polygons[index[-1]].tolist())
+        A = polygons[index[-1]][:-1]
+        index = np.delete(index, -1)
+        iou_list = np.zeros((len(index),))
+        for i in range(len(index)):
+            B = polygons[index[i]][:-1]
+            iou_list[i] = boundary_iou(A, B)
+        remove_index = np.where(iou_list > threshold)
+        index = np.delete(index, remove_index)
+
+    return keep_poly
diff --git a/ppocr/utils/ppocr_keys_v1.txt b/ppocr/utils/ppocr_keys_v1.txt
new file mode 100644
index 0000000..b75af21
--- /dev/null
+++ b/ppocr/utils/ppocr_keys_v1.txt
@@ -0,0 +1,6623 @@
+'
+疗
+绚
+诚
+娇
+溜
+题
+贿
+者
+廖
+更
+纳
+加
+奉
+公
+一
+就
+汴
+计
+与
+路
+房
+原
+妇
+2
+0
+8
+-
+7
+其
+>
+:
+]
+,
+，
+骑
+刈
+全
+消
+昏
+傈
+安
+久
+钟
+嗅
+不
+影
+处
+驽
+蜿
+资
+关
+椤
+地
+瘸
+专
+问
+忖
+票
+嫉
+炎
+韵
+要
+月
+田
+节
+陂
+鄙
+捌
+备
+拳
+伺
+眼
+网
+盎
+大
+傍
+心
+东
+愉
+汇
+蹿
+科
+每
+业
+里
+航
+晏
+字
+平
+录
+先
+1
+3
+彤
+鲶
+产
+稍
+督
+腴
+有
+象
+岳
+注
+绍
+在
+泺
+文
+定
+核
+名
+水
+过
+理
+让
+偷
+率
+等
+这
+发
+”
+为
+含
+肥
+酉
+相
+鄱
+七
+编
+猥
+锛
+日
+镀
+蒂
+掰
+倒
+辆
+栾
+栗
+综
+涩
+州
+雌
+滑
+馀
+了
+机
+块
+司
+宰
+甙
+兴
+矽
+抚
+保
+用
+沧
+秩
+如
+收
+息
+滥
+页
+疑
+埠
+!
+！
+姥
+异
+橹
+钇
+向
+下
+跄
+的
+椴
+沫
+国
+绥
+獠
+报
+开
+民
+蜇
+何
+分
+凇
+长
+讥
+藏
+掏
+施
+羽
+中
+讲
+派
+嘟
+人
+提
+浼
+间
+世
+而
+古
+多
+倪
+唇
+饯
+控
+庚
+首
+赛
+蜓
+味
+断
+制
+觉
+技
+替
+艰
+溢
+潮
+夕
+钺
+外
+摘
+枋
+动
+双
+单
+啮
+户
+枇
+确
+锦
+曜
+杜
+或
+能
+效
+霜
+盒
+然
+侗
+电
+晁
+放
+步
+鹃
+新
+杖
+蜂
+吒
+濂
+瞬
+评
+总
+隍
+对
+独
+合
+也
+是
+府
+青
+天
+诲
+墙
+组
+滴
+级
+邀
+帘
+示
+已
+时
+骸
+仄
+泅
+和
+遨
+店
+雇
+疫
+持
+巍
+踮
+境
+只
+亨
+目
+鉴
+崤
+闲
+体
+泄
+杂
+作
+般
+轰
+化
+解
+迂
+诿
+蛭
+璀
+腾
+告
+版
+服
+省
+师
+小
+规
+程
+线
+海
+办
+引
+二
+桧
+牌
+砺
+洄
+裴
+修
+图
+痫
+胡
+许
+犊
+事
+郛
+基
+柴
+呼
+食
+研
+奶
+律
+蛋
+因
+葆
+察
+戏
+褒
+戒
+再
+李
+骁
+工
+貂
+油
+鹅
+章
+啄
+休
+场
+给
+睡
+纷
+豆
+器
+捎
+说
+敏
+学
+会
+浒
+设
+诊
+格
+廓
+查
+来
+霓
+室
+溆
+￠
+诡
+寥
+焕
+舜
+柒
+狐
+回
+戟
+砾
+厄
+实
+翩
+尿
+五
+入
+径
+惭
+喹
+股
+宇
+篝
+|
+;
+美
+期
+云
+九
+祺
+扮
+靠
+锝
+槌
+系
+企
+酰
+阊
+暂
+蚕
+忻
+豁
+本
+羹
+执
+条
+钦
+H
+獒
+限
+进
+季
+楦
+于
+芘
+玖
+铋
+茯
+未
+答
+粘
+括
+样
+精
+欠
+矢
+甥
+帷
+嵩
+扣
+令
+仔
+风
+皈
+行
+支
+部
+蓉
+刮
+站
+蜡
+救
+钊
+汗
+松
+嫌
+成
+可
+.
+鹤
+院
+从
+交
+政
+怕
+活
+调
+球
+局
+验
+髌
+第
+韫
+谗
+串
+到
+圆
+年
+米
+/
+*
+友
+忿
+检
+区
+看
+自
+敢
+刃
+个
+兹
+弄
+流
+留
+同
+没
+齿
+星
+聆
+轼
+湖
+什
+三
+建
+蛔
+儿
+椋
+汕
+震
+颧
+鲤
+跟
+力
+情
+璺
+铨
+陪
+务
+指
+族
+训
+滦
+鄣
+濮
+扒
+商
+箱
+十
+召
+慷
+辗
+所
+莞
+管
+护
+臭
+横
+硒
+嗓
+接
+侦
+六
+露
+党
+馋
+驾
+剖
+高
+侬
+妪
+幂
+猗
+绺
+骐
+央
+酐
+孝
+筝
+课
+徇
+缰
+门
+男
+西
+项
+句
+谙
+瞒
+秃
+篇
+教
+碲
+罚
+声
+呐
+景
+前
+富
+嘴
+鳌
+稀
+免
+朋
+啬
+睐
+去
+赈
+鱼
+住
+肩
+愕
+速
+旁
+波
+厅
+健
+茼
+厥
+鲟
+谅
+投
+攸
+炔
+数
+方
+击
+呋
+谈
+绩
+别
+愫
+僚
+躬
+鹧
+胪
+炳
+招
+喇
+膨
+泵
+蹦
+毛
+结
+5
+4
+谱
+识
+陕
+粽
+婚
+拟
+构
+且
+搜
+任
+潘
+比
+郢
+妨
+醪
+陀
+桔
+碘
+扎
+选
+哈
+骷
+楷
+亿
+明
+缆
+脯
+监
+睫
+逻
+婵
+共
+赴
+淝
+凡
+惦
+及
+达
+揖
+谩
+澹
+减
+焰
+蛹
+番
+祁
+柏
+员
+禄
+怡
+峤
+龙
+白
+叽
+生
+闯
+起
+细
+装
+谕
+竟
+聚
+钙
+上
+导
+渊
+按
+艾
+辘
+挡
+耒
+盹
+饪
+臀
+记
+邮
+蕙
+受
+各
+医
+搂
+普
+滇
+朗
+茸
+带
+翻
+酚
+(
+光
+堤
+墟
+蔷
+万
+幻
+〓
+瑙
+辈
+昧
+盏
+亘
+蛀
+吉
+铰
+请
+子
+假
+闻
+税
+井
+诩
+哨
+嫂
+好
+面
+琐
+校
+馊
+鬣
+缂
+营
+访
+炖
+占
+农
+缀
+否
+经
+钚
+棵
+趟
+张
+亟
+吏
+茶
+谨
+捻
+论
+迸
+堂
+玉
+信
+吧
+瞠
+乡
+姬
+寺
+咬
+溏
+苄
+皿
+意
+赉
+宝
+尔
+钰
+艺
+特
+唳
+踉
+都
+荣
+倚
+登
+荐
+丧
+奇
+涵
+批
+炭
+近
+符
+傩
+感
+道
+着
+菊
+虹
+仲
+众
+懈
+濯
+颞
+眺
+南
+释
+北
+缝
+标
+既
+茗
+整
+撼
+迤
+贲
+挎
+耱
+拒
+某
+妍
+卫
+哇
+英
+矶
+藩
+治
+他
+元
+领
+膜
+遮
+穗
+蛾
+飞
+荒
+棺
+劫
+么
+市
+火
+温
+拈
+棚
+洼
+转
+果
+奕
+卸
+迪
+伸
+泳
+斗
+邡
+侄
+涨
+屯
+萋
+胭
+氡
+崮
+枞
+惧
+冒
+彩
+斜
+手
+豚
+随
+旭
+淑
+妞
+形
+菌
+吲
+沱
+争
+驯
+歹
+挟
+兆
+柱
+传
+至
+包
+内
+响
+临
+红
+功
+弩
+衡
+寂
+禁
+老
+棍
+耆
+渍
+织
+害
+氵
+渑
+布
+载
+靥
+嗬
+虽
+苹
+咨
+娄
+库
+雉
+榜
+帜
+嘲
+套
+瑚
+亲
+簸
+欧
+边
+6
+腿
+旮
+抛
+吹
+瞳
+得
+镓
+梗
+厨
+继
+漾
+愣
+憨
+士
+策
+窑
+抑
+躯
+襟
+脏
+参
+贸
+言
+干
+绸
+鳄
+穷
+藜
+音
+折
+详
+)
+举
+悍
+甸
+癌
+黎
+谴
+死
+罩
+迁
+寒
+驷
+袖
+媒
+蒋
+掘
+模
+纠
+恣
+观
+祖
+蛆
+碍
+位
+稿
+主
+澧
+跌
+筏
+京
+锏
+帝
+贴
+证
+糠
+才
+黄
+鲸
+略
+炯
+饱
+四
+出
+园
+犀
+牧
+容
+汉
+杆
+浈
+汰
+瑷
+造
+虫
+瘩
+怪
+驴
+济
+应
+花
+沣
+谔
+夙
+旅
+价
+矿
+以
+考
+s
+u
+呦
+晒
+巡
+茅
+准
+肟
+瓴
+詹
+仟
+褂
+译
+桌
+混
+宁
+怦
+郑
+抿
+些
+余
+鄂
+饴
+攒
+珑
+群
+阖
+岔
+琨
+藓
+预
+环
+洮
+岌
+宀
+杲
+瀵
+最
+常
+囡
+周
+踊
+女
+鼓
+袭
+喉
+简
+范
+薯
+遐
+疏
+粱
+黜
+禧
+法
+箔
+斤
+遥
+汝
+奥
+直
+贞
+撑
+置
+绱
+集
+她
+馅
+逗
+钧
+橱
+魉
+[
+恙
+躁
+唤
+9
+旺
+膘
+待
+脾
+惫
+购
+吗
+依
+盲
+度
+瘿
+蠖
+俾
+之
+镗
+拇
+鲵
+厝
+簧
+续
+款
+展
+啃
+表
+剔
+品
+钻
+腭
+损
+清
+锶
+统
+涌
+寸
+滨
+贪
+链
+吠
+冈
+伎
+迥
+咏
+吁
+览
+防
+迅
+失
+汾
+阔
+逵
+绀
+蔑
+列
+川
+凭
+努
+熨
+揪
+利
+俱
+绉
+抢
+鸨
+我
+即
+责
+膦
+易
+毓
+鹊
+刹
+玷
+岿
+空
+嘞
+绊
+排
+术
+估
+锷
+违
+们
+苟
+铜
+播
+肘
+件
+烫
+审
+鲂
+广
+像
+铌
+惰
+铟
+巳
+胍
+鲍
+康
+憧
+色
+恢
+想
+拷
+尤
+疳
+知
+S
+Y
+F
+D
+A
+峄
+裕
+帮
+握
+搔
+氐
+氘
+难
+墒
+沮
+雨
+叁
+缥
+悴
+藐
+湫
+娟
+苑
+稠
+颛
+簇
+后
+阕
+闭
+蕤
+缚
+怎
+佞
+码
+嘤
+蔡
+痊
+舱
+螯
+帕
+赫
+昵
+升
+烬
+岫
+、
+疵
+蜻
+髁
+蕨
+隶
+烛
+械
+丑
+盂
+梁
+强
+鲛
+由
+拘
+揉
+劭
+龟
+撤
+钩
+呕
+孛
+费
+妻
+漂
+求
+阑
+崖
+秤
+甘
+通
+深
+补
+赃
+坎
+床
+啪
+承
+吼
+量
+暇
+钼
+烨
+阂
+擎
+脱
+逮
+称
+P
+神
+属
+矗
+华
+届
+狍
+葑
+汹
+育
+患
+窒
+蛰
+佼
+静
+槎
+运
+鳗
+庆
+逝
+曼
+疱
+克
+代
+官
+此
+麸
+耧
+蚌
+晟
+例
+础
+榛
+副
+测
+唰
+缢
+迹
+灬
+霁
+身
+岁
+赭
+扛
+又
+菡
+乜
+雾
+板
+读
+陷
+徉
+贯
+郁
+虑
+变
+钓
+菜
+圾
+现
+琢
+式
+乐
+维
+渔
+浜
+左
+吾
+脑
+钡
+警
+T
+啵
+拴
+偌
+漱
+湿
+硕
+止
+骼
+魄
+积
+燥
+联
+踢
+玛
+则
+窿
+见
+振
+畿
+送
+班
+钽
+您
+赵
+刨
+印
+讨
+踝
+籍
+谡
+舌
+崧
+汽
+蔽
+沪
+酥
+绒
+怖
+财
+帖
+肱
+私
+莎
+勋
+羔
+霸
+励
+哼
+帐
+将
+帅
+渠
+纪
+婴
+娩
+岭
+厘
+滕
+吻
+伤
+坝
+冠
+戊
+隆
+瘁
+介
+涧
+物
+黍
+并
+姗
+奢
+蹑
+掣
+垸
+锴
+命
+箍
+捉
+病
+辖
+琰
+眭
+迩
+艘
+绌
+繁
+寅
+若
+毋
+思
+诉
+类
+诈
+燮
+轲
+酮
+狂
+重
+反
+职
+筱
+县
+委
+磕
+绣
+奖
+晋
+濉
+志
+徽
+肠
+呈
+獐
+坻
+口
+片
+碰
+几
+村
+柿
+劳
+料
+获
+亩
+惕
+晕
+厌
+号
+罢
+池
+正
+鏖
+煨
+家
+棕
+复
+尝
+懋
+蜥
+锅
+岛
+扰
+队
+坠
+瘾
+钬
+@
+卧
+疣
+镇
+譬
+冰
+彷
+频
+黯
+据
+垄
+采
+八
+缪
+瘫
+型
+熹
+砰
+楠
+襁
+箐
+但
+嘶
+绳
+啤
+拍
+盥
+穆
+傲
+洗
+盯
+塘
+怔
+筛
+丿
+台
+恒
+喂
+葛
+永
+￥
+烟
+酒
+桦
+书
+砂
+蚝
+缉
+态
+瀚
+袄
+圳
+轻
+蛛
+超
+榧
+遛
+姒
+奘
+铮
+右
+荽
+望
+偻
+卡
+丶
+氰
+附
+做
+革
+索
+戚
+坨
+桷
+唁
+垅
+榻
+岐
+偎
+坛
+莨
+山
+殊
+微
+骇
+陈
+爨
+推
+嗝
+驹
+澡
+藁
+呤
+卤
+嘻
+糅
+逛
+侵
+郓
+酌
+德
+摇
+※
+鬃
+被
+慨
+殡
+羸
+昌
+泡
+戛
+鞋
+河
+宪
+沿
+玲
+鲨
+翅
+哽
+源
+铅
+语
+照
+邯
+址
+荃
+佬
+顺
+鸳
+町
+霭
+睾
+瓢
+夸
+椁
+晓
+酿
+痈
+咔
+侏
+券
+噎
+湍
+签
+嚷
+离
+午
+尚
+社
+锤
+背
+孟
+使
+浪
+缦
+潍
+鞅
+军
+姹
+驶
+笑
+鳟
+鲁
+》
+孽
+钜
+绿
+洱
+礴
+焯
+椰
+颖
+囔
+乌
+孔
+巴
+互
+性
+椽
+哞
+聘
+昨
+早
+暮
+胶
+炀
+隧
+低
+彗
+昝
+铁
+呓
+氽
+藉
+喔
+癖
+瑗
+姨
+权
+胱
+韦
+堑
+蜜
+酋
+楝
+砝
+毁
+靓
+歙
+锲
+究
+屋
+喳
+骨
+辨
+碑
+武
+鸠
+宫
+辜
+烊
+适
+坡
+殃
+培
+佩
+供
+走
+蜈
+迟
+翼
+况
+姣
+凛
+浔
+吃
+飘
+债
+犟
+金
+促
+苛
+崇
+坂
+莳
+畔
+绂
+兵
+蠕
+斋
+根
+砍
+亢
+欢
+恬
+崔
+剁
+餐
+榫
+快
+扶
+‖
+濒
+缠
+鳜
+当
+彭
+驭
+浦
+篮
+昀
+锆
+秸
+钳
+弋
+娣
+瞑
+夷
+龛
+苫
+拱
+致
+%
+嵊
+障
+隐
+弑
+初
+娓
+抉
+汩
+累
+蓖
+"
+唬
+助
+苓
+昙
+押
+毙
+破
+城
+郧
+逢
+嚏
+獭
+瞻
+溱
+婿
+赊
+跨
+恼
+璧
+萃
+姻
+貉
+灵
+炉
+密
+氛
+陶
+砸
+谬
+衔
+点
+琛
+沛
+枳
+层
+岱
+诺
+脍
+榈
+埂
+征
+冷
+裁
+打
+蹴
+素
+瘘
+逞
+蛐
+聊
+激
+腱
+萘
+踵
+飒
+蓟
+吆
+取
+咙
+簋
+涓
+矩
+曝
+挺
+揣
+座
+你
+史
+舵
+焱
+尘
+苏
+笈
+脚
+溉
+榨
+诵
+樊
+邓
+焊
+义
+庶
+儋
+蟋
+蒲
+赦
+呷
+杞
+诠
+豪
+还
+试
+颓
+茉
+太
+除
+紫
+逃
+痴
+草
+充
+鳕
+珉
+祗
+墨
+渭
+烩
+蘸
+慕
+璇
+镶
+穴
+嵘
+恶
+骂
+险
+绋
+幕
+碉
+肺
+戳
+刘
+潞
+秣
+纾
+潜
+銮
+洛
+须
+罘
+销
+瘪
+汞
+兮
+屉
+r
+林
+厕
+质
+探
+划
+狸
+殚
+善
+煊
+烹
+〒
+锈
+逯
+宸
+辍
+泱
+柚
+袍
+远
+蹋
+嶙
+绝
+峥
+娥
+缍
+雀
+徵
+认
+镱
+谷
+=
+贩
+勉
+撩
+鄯
+斐
+洋
+非
+祚
+泾
+诒
+饿
+撬
+威
+晷
+搭
+芍
+锥
+笺
+蓦
+候
+琊
+档
+礁
+沼
+卵
+荠
+忑
+朝
+凹
+瑞
+头
+仪
+弧
+孵
+畏
+铆
+突
+衲
+车
+浩
+气
+茂
+悖
+厢
+枕
+酝
+戴
+湾
+邹
+飚
+攘
+锂
+写
+宵
+翁
+岷
+无
+喜
+丈
+挑
+嗟
+绛
+殉
+议
+槽
+具
+醇
+淞
+笃
+郴
+阅
+饼
+底
+壕
+砚
+弈
+询
+缕
+庹
+翟
+零
+筷
+暨
+舟
+闺
+甯
+撞
+麂
+茌
+蔼
+很
+珲
+捕
+棠
+角
+阉
+媛
+娲
+诽
+剿
+尉
+爵
+睬
+韩
+诰
+匣
+危
+糍
+镯
+立
+浏
+阳
+少
+盆
+舔
+擘
+匪
+申
+尬
+铣
+旯
+抖
+赘
+瓯
+居
+ˇ
+哮
+游
+锭
+茏
+歌
+坏
+甚
+秒
+舞
+沙
+仗
+劲
+潺
+阿
+燧
+郭
+嗖
+霏
+忠
+材
+奂
+耐
+跺
+砀
+输
+岖
+媳
+氟
+极
+摆
+灿
+今
+扔
+腻
+枝
+奎
+药
+熄
+吨
+话
+q
+额
+慑
+嘌
+协
+喀
+壳
+埭
+视
+著
+於
+愧
+陲
+翌
+峁
+颅
+佛
+腹
+聋
+侯
+咎
+叟
+秀
+颇
+存
+较
+罪
+哄
+岗
+扫
+栏
+钾
+羌
+己
+璨
+枭
+霉
+煌
+涸
+衿
+键
+镝
+益
+岢
+奏
+连
+夯
+睿
+冥
+均
+糖
+狞
+蹊
+稻
+爸
+刿
+胥
+煜
+丽
+肿
+璃
+掸
+跚
+灾
+垂
+樾
+濑
+乎
+莲
+窄
+犹
+撮
+战
+馄
+软
+络
+显
+鸢
+胸
+宾
+妲
+恕
+埔
+蝌
+份
+遇
+巧
+瞟
+粒
+恰
+剥
+桡
+博
+讯
+凯
+堇
+阶
+滤
+卖
+斌
+骚
+彬
+兑
+磺
+樱
+舷
+两
+娱
+福
+仃
+差
+找
+桁
+÷
+净
+把
+阴
+污
+戬
+雷
+碓
+蕲
+楚
+罡
+焖
+抽
+妫
+咒
+仑
+闱
+尽
+邑
+菁
+爱
+贷
+沥
+鞑
+牡
+嗉
+崴
+骤
+塌
+嗦
+订
+拮
+滓
+捡
+锻
+次
+坪
+杩
+臃
+箬
+融
+珂
+鹗
+宗
+枚
+降
+鸬
+妯
+阄
+堰
+盐
+毅
+必
+杨
+崃
+俺
+甬
+状
+莘
+货
+耸
+菱
+腼
+铸
+唏
+痤
+孚
+澳
+懒
+溅
+翘
+疙
+杷
+淼
+缙
+骰
+喊
+悉
+砻
+坷
+艇
+赁
+界
+谤
+纣
+宴
+晃
+茹
+归
+饭
+梢
+铡
+街
+抄
+肼
+鬟
+苯
+颂
+撷
+戈
+炒
+咆
+茭
+瘙
+负
+仰
+客
+琉
+铢
+封
+卑
+珥
+椿
+镧
+窨
+鬲
+寿
+御
+袤
+铃
+萎
+砖
+餮
+脒
+裳
+肪
+孕
+嫣
+馗
+嵇
+恳
+氯
+江
+石
+褶
+冢
+祸
+阻
+狈
+羞
+银
+靳
+透
+咳
+叼
+敷
+芷
+啥
+它
+瓤
+兰
+痘
+懊
+逑
+肌
+往
+捺
+坊
+甩
+呻
+〃
+沦
+忘
+膻
+祟
+菅
+剧
+崆
+智
+坯
+臧
+霍
+墅
+攻
+眯
+倘
+拢
+骠
+铐
+庭
+岙
+瓠
+′
+缺
+泥
+迢
+捶
+?
+？
+郏
+喙
+掷
+沌
+纯
+秘
+种
+听
+绘
+固
+螨
+团
+香
+盗
+妒
+埚
+蓝
+拖
+旱
+荞
+铀
+血
+遏
+汲
+辰
+叩
+拽
+幅
+硬
+惶
+桀
+漠
+措
+泼
+唑
+齐
+肾
+念
+酱
+虚
+屁
+耶
+旗
+砦
+闵
+婉
+馆
+拭
+绅
+韧
+忏
+窝
+醋
+葺
+顾
+辞
+倜
+堆
+辋
+逆
+玟
+贱
+疾
+董
+惘
+倌
+锕
+淘
+嘀
+莽
+俭
+笏
+绑
+鲷
+杈
+择
+蟀
+粥
+嗯
+驰
+逾
+案
+谪
+褓
+胫
+哩
+昕
+颚
+鲢
+绠
+躺
+鹄
+崂
+儒
+俨
+丝
+尕
+泌
+啊
+萸
+彰
+幺
+吟
+骄
+苣
+弦
+脊
+瑰
+〈
+诛
+镁
+析
+闪
+剪
+侧
+哟
+框
+螃
+守
+嬗
+燕
+狭
+铈
+缮
+概
+迳
+痧
+鲲
+俯
+售
+笼
+痣
+扉
+挖
+满
+咋
+援
+邱
+扇
+歪
+便
+玑
+绦
+峡
+蛇
+叨
+〖
+泽
+胃
+斓
+喋
+怂
+坟
+猪
+该
+蚬
+炕
+弥
+赞
+棣
+晔
+娠
+挲
+狡
+创
+疖
+铕
+镭
+稷
+挫
+弭
+啾
+翔
+粉
+履
+苘
+哦
+楼
+秕
+铂
+土
+锣
+瘟
+挣
+栉
+习
+享
+桢
+袅
+磨
+桂
+谦
+延
+坚
+蔚
+噗
+署
+谟
+猬
+钎
+恐
+嬉
+雒
+倦
+衅
+亏
+璩
+睹
+刻
+殿
+王
+算
+雕
+麻
+丘
+柯
+骆
+丸
+塍
+谚
+添
+鲈
+垓
+桎
+蚯
+芥
+予
+飕
+镦
+谌
+窗
+醚
+菀
+亮
+搪
+莺
+蒿
+羁
+足
+J
+真
+轶
+悬
+衷
+靛
+翊
+掩
+哒
+炅
+掐
+冼
+妮
+l
+谐
+稚
+荆
+擒
+犯
+陵
+虏
+浓
+崽
+刍
+陌
+傻
+孜
+千
+靖
+演
+矜
+钕
+煽
+杰
+酗
+渗
+伞
+栋
+俗
+泫
+戍
+罕
+沾
+疽
+灏
+煦
+芬
+磴
+叱
+阱
+榉
+湃
+蜀
+叉
+醒
+彪
+租
+郡
+篷
+屎
+良
+垢
+隗
+弱
+陨
+峪
+砷
+掴
+颁
+胎
+雯
+绵
+贬
+沐
+撵
+隘
+篙
+暖
+曹
+陡
+栓
+填
+臼
+彦
+瓶
+琪
+潼
+哪
+鸡
+摩
+啦
+俟
+锋
+域
+耻
+蔫
+疯
+纹
+撇
+毒
+绶
+痛
+酯
+忍
+爪
+赳
+歆
+嘹
+辕
+烈
+册
+朴
+钱
+吮
+毯
+癜
+娃
+谀
+邵
+厮
+炽
+璞
+邃
+丐
+追
+词
+瓒
+忆
+轧
+芫
+谯
+喷
+弟
+半
+冕
+裙
+掖
+墉
+绮
+寝
+苔
+势
+顷
+褥
+切
+衮
+君
+佳
+嫒
+蚩
+霞
+佚
+洙
+逊
+镖
+暹
+唛
+&
+殒
+顶
+碗
+獗
+轭
+铺
+蛊
+废
+恹
+汨
+崩
+珍
+那
+杵
+曲
+纺
+夏
+薰
+傀
+闳
+淬
+姘
+舀
+拧
+卷
+楂
+恍
+讪
+厩
+寮
+篪
+赓
+乘
+灭
+盅
+鞣
+沟
+慎
+挂
+饺
+鼾
+杳
+树
+缨
+丛
+絮
+娌
+臻
+嗳
+篡
+侩
+述
+衰
+矛
+圈
+蚜
+匕
+筹
+匿
+濞
+晨
+叶
+骋
+郝
+挚
+蚴
+滞
+增
+侍
+描
+瓣
+吖
+嫦
+蟒
+匾
+圣
+赌
+毡
+癞
+恺
+百
+曳
+需
+篓
+肮
+庖
+帏
+卿
+驿
+遗
+蹬
+鬓
+骡
+歉
+芎
+胳
+屐
+禽
+烦
+晌
+寄
+媾
+狄
+翡
+苒
+船
+廉
+终
+痞
+殇
+々
+畦
+饶
+改
+拆
+悻
+萄
+￡
+瓿
+乃
+訾
+桅
+匮
+溧
+拥
+纱
+铍
+骗
+蕃
+龋
+缬
+父
+佐
+疚
+栎
+醍
+掳
+蓄
+x
+惆
+颜
+鲆
+榆
+〔
+猎
+敌
+暴
+谥
+鲫
+贾
+罗
+玻
+缄
+扦
+芪
+癣
+落
+徒
+臾
+恿
+猩
+托
+邴
+肄
+牵
+春
+陛
+耀
+刊
+拓
+蓓
+邳
+堕
+寇
+枉
+淌
+啡
+湄
+兽
+酷
+萼
+碚
+濠
+萤
+夹
+旬
+戮
+梭
+琥
+椭
+昔
+勺
+蜊
+绐
+晚
+孺
+僵
+宣
+摄
+冽
+旨
+萌
+忙
+蚤
+眉
+噼
+蟑
+付
+契
+瓜
+悼
+颡
+壁
+曾
+窕
+颢
+澎
+仿
+俑
+浑
+嵌
+浣
+乍
+碌
+褪
+乱
+蔟
+隙
+玩
+剐
+葫
+箫
+纲
+围
+伐
+决
+伙
+漩
+瑟
+刑
+肓
+镳
+缓
+蹭
+氨
+皓
+典
+畲
+坍
+铑
+檐
+塑
+洞
+倬
+储
+胴
+淳
+戾
+吐
+灼
+惺
+妙
+毕
+珐
+缈
+虱
+盖
+羰
+鸿
+磅
+谓
+髅
+娴
+苴
+唷
+蚣
+霹
+抨
+贤
+唠
+犬
+誓
+逍
+庠
+逼
+麓
+籼
+釉
+呜
+碧
+秧
+氩
+摔
+霄
+穸
+纨
+辟
+妈
+映
+完
+牛
+缴
+嗷
+炊
+恩
+荔
+茆
+掉
+紊
+慌
+莓
+羟
+阙
+萁
+磐
+另
+蕹
+辱
+鳐
+湮
+吡
+吩
+唐
+睦
+垠
+舒
+圜
+冗
+瞿
+溺
+芾
+囱
+匠
+僳
+汐
+菩
+饬
+漓
+黑
+霰
+浸
+濡
+窥
+毂
+蒡
+兢
+驻
+鹉
+芮
+诙
+迫
+雳
+厂
+忐
+臆
+猴
+鸣
+蚪
+栈
+箕
+羡
+渐
+莆
+捍
+眈
+哓
+趴
+蹼
+埕
+嚣
+骛
+宏
+淄
+斑
+噜
+严
+瑛
+垃
+椎
+诱
+压
+庾
+绞
+焘
+廿
+抡
+迄
+棘
+夫
+纬
+锹
+眨
+瞌
+侠
+脐
+竞
+瀑
+孳
+骧
+遁
+姜
+颦
+荪
+滚
+萦
+伪
+逸
+粳
+爬
+锁
+矣
+役
+趣
+洒
+颔
+诏
+逐
+奸
+甭
+惠
+攀
+蹄
+泛
+尼
+拼
+阮
+鹰
+亚
+颈
+惑
+勒
+〉
+际
+肛
+爷
+刚
+钨
+丰
+养
+冶
+鲽
+辉
+蔻
+画
+覆
+皴
+妊
+麦
+返
+醉
+皂
+擀
+〗
+酶
+凑
+粹
+悟
+诀
+硖
+港
+卜
+z
+杀
+涕
+±
+舍
+铠
+抵
+弛
+段
+敝
+镐
+奠
+拂
+轴
+跛
+袱
+e
+t
+沉
+菇
+俎
+薪
+峦
+秭
+蟹
+历
+盟
+菠
+寡
+液
+肢
+喻
+染
+裱
+悱
+抱
+氙
+赤
+捅
+猛
+跑
+氮
+谣
+仁
+尺
+辊
+窍
+烙
+衍
+架
+擦
+倏
+璐
+瑁
+币
+楞
+胖
+夔
+趸
+邛
+惴
+饕
+虔
+蝎
+§
+哉
+贝
+宽
+辫
+炮
+扩
+饲
+籽
+魏
+菟
+锰
+伍
+猝
+末
+琳
+哚
+蛎
+邂
+呀
+姿
+鄞
+却
+歧
+仙
+恸
+椐
+森
+牒
+寤
+袒
+婆
+虢
+雅
+钉
+朵
+贼
+欲
+苞
+寰
+故
+龚
+坭
+嘘
+咫
+礼
+硷
+兀
+睢
+汶
+’
+铲
+烧
+绕
+诃
+浃
+钿
+哺
+柜
+讼
+颊
+璁
+腔
+洽
+咐
+脲
+簌
+筠
+镣
+玮
+鞠
+谁
+兼
+姆
+挥
+梯
+蝴
+谘
+漕
+刷
+躏
+宦
+弼
+b
+垌
+劈
+麟
+莉
+揭
+笙
+渎
+仕
+嗤
+仓
+配
+怏
+抬
+错
+泯
+镊
+孰
+猿
+邪
+仍
+秋
+鼬
+壹
+歇
+吵
+炼
+<
+尧
+射
+柬
+廷
+胧
+霾
+凳
+隋
+肚
+浮
+梦
+祥
+株
+堵
+退
+L
+鹫
+跎
+凶
+毽
+荟
+炫
+栩
+玳
+甜
+沂
+鹿
+顽
+伯
+爹
+赔
+蛴
+徐
+匡
+欣
+狰
+缸
+雹
+蟆
+疤
+默
+沤
+啜
+痂
+衣
+禅
+w
+i
+h
+辽
+葳
+黝
+钗
+停
+沽
+棒
+馨
+颌
+肉
+吴
+硫
+悯
+劾
+娈
+马
+啧
+吊
+悌
+镑
+峭
+帆
+瀣
+涉
+咸
+疸
+滋
+泣
+翦
+拙
+癸
+钥
+蜒
++
+尾
+庄
+凝
+泉
+婢
+渴
+谊
+乞
+陆
+锉
+糊
+鸦
+淮
+I
+B
+N
+晦
+弗
+乔
+庥
+葡
+尻
+席
+橡
+傣
+渣
+拿
+惩
+麋
+斛
+缃
+矮
+蛏
+岘
+鸽
+姐
+膏
+催
+奔
+镒
+喱
+蠡
+摧
+钯
+胤
+柠
+拐
+璋
+鸥
+卢
+荡
+倾
+^
+_
+珀
+逄
+萧
+塾
+掇
+贮
+笆
+聂
+圃
+冲
+嵬
+M
+滔
+笕
+值
+炙
+偶
+蜱
+搐
+梆
+汪
+蔬
+腑
+鸯
+蹇
+敞
+绯
+仨
+祯
+谆
+梧
+糗
+鑫
+啸
+豺
+囹
+猾
+巢
+柄
+瀛
+筑
+踌
+沭
+暗
+苁
+鱿
+蹉
+脂
+蘖
+牢
+热
+木
+吸
+溃
+宠
+序
+泞
+偿
+拜
+檩
+厚
+朐
+毗
+螳
+吞
+媚
+朽
+担
+蝗
+橘
+畴
+祈
+糟
+盱
+隼
+郜
+惜
+珠
+裨
+铵
+焙
+琚
+唯
+咚
+噪
+骊
+丫
+滢
+勤
+棉
+呸
+咣
+淀
+隔
+蕾
+窈
+饨
+挨
+煅
+短
+匙
+粕
+镜
+赣
+撕
+墩
+酬
+馁
+豌
+颐
+抗
+酣
+氓
+佑
+搁
+哭
+递
+耷
+涡
+桃
+贻
+碣
+截
+瘦
+昭
+镌
+蔓
+氚
+甲
+猕
+蕴
+蓬
+散
+拾
+纛
+狼
+猷
+铎
+埋
+旖
+矾
+讳
+囊
+糜
+迈
+粟
+蚂
+紧
+鲳
+瘢
+栽
+稼
+羊
+锄
+斟
+睁
+桥
+瓮
+蹙
+祉
+醺
+鼻
+昱
+剃
+跳
+篱
+跷
+蒜
+翎
+宅
+晖
+嗑
+壑
+峻
+癫
+屏
+狠
+陋
+袜
+途
+憎
+祀
+莹
+滟
+佶
+溥
+臣
+约
+盛
+峰
+磁
+慵
+婪
+拦
+莅
+朕
+鹦
+粲
+裤
+哎
+疡
+嫖
+琵
+窟
+堪
+谛
+嘉
+儡
+鳝
+斩
+郾
+驸
+酊
+妄
+胜
+贺
+徙
+傅
+噌
+钢
+栅
+庇
+恋
+匝
+巯
+邈
+尸
+锚
+粗
+佟
+蛟
+薹
+纵
+蚊
+郅
+绢
+锐
+苗
+俞
+篆
+淆
+膀
+鲜
+煎
+诶
+秽
+寻
+涮
+刺
+怀
+噶
+巨
+褰
+魅
+灶
+灌
+桉
+藕
+谜
+舸
+薄
+搀
+恽
+借
+牯
+痉
+渥
+愿
+亓
+耘
+杠
+柩
+锔
+蚶
+钣
+珈
+喘
+蹒
+幽
+赐
+稗
+晤
+莱
+泔
+扯
+肯
+菪
+裆
+腩
+豉
+疆
+骜
+腐
+倭
+珏
+唔
+粮
+亡
+润
+慰
+伽
+橄
+玄
+誉
+醐
+胆
+龊
+粼
+塬
+陇
+彼
+削
+嗣
+绾
+芽
+妗
+垭
+瘴
+爽
+薏
+寨
+龈
+泠
+弹
+赢
+漪
+猫
+嘧
+涂
+恤
+圭
+茧
+烽
+屑
+痕
+巾
+赖
+荸
+凰
+腮
+畈
+亵
+蹲
+偃
+苇
+澜
+艮
+换
+骺
+烘
+苕
+梓
+颉
+肇
+哗
+悄
+氤
+涠
+葬
+屠
+鹭
+植
+竺
+佯
+诣
+鲇
+瘀
+鲅
+邦
+移
+滁
+冯
+耕
+癔
+戌
+茬
+沁
+巩
+悠
+湘
+洪
+痹
+锟
+循
+谋
+腕
+鳃
+钠
+捞
+焉
+迎
+碱
+伫
+急
+榷
+奈
+邝
+卯
+辄
+皲
+卟
+醛
+畹
+忧
+稳
+雄
+昼
+缩
+阈
+睑
+扌
+耗
+曦
+涅
+捏
+瞧
+邕
+淖
+漉
+铝
+耦
+禹
+湛
+喽
+莼
+琅
+诸
+苎
+纂
+硅
+始
+嗨
+傥
+燃
+臂
+赅
+嘈
+呆
+贵
+屹
+壮
+肋
+亍
+蚀
+卅
+豹
+腆
+邬
+迭
+浊
+}
+童
+螂
+捐
+圩
+勐
+触
+寞
+汊
+壤
+荫
+膺
+渌
+芳
+懿
+遴
+螈
+泰
+蓼
+蛤
+茜
+舅
+枫
+朔
+膝
+眙
+避
+梅
+判
+鹜
+璜
+牍
+缅
+垫
+藻
+黔
+侥
+惚
+懂
+踩
+腰
+腈
+札
+丞
+唾
+慈
+顿
+摹
+荻
+琬
+~
+斧
+沈
+滂
+胁
+胀
+幄
+莜
+Z
+匀
+鄄
+掌
+绰
+茎
+焚
+赋
+萱
+谑
+汁
+铒
+瞎
+夺
+蜗
+野
+娆
+冀
+弯
+篁
+懵
+灞
+隽
+芡
+脘
+俐
+辩
+芯
+掺
+喏
+膈
+蝈
+觐
+悚
+踹
+蔗
+熠
+鼠
+呵
+抓
+橼
+峨
+畜
+缔
+禾
+崭
+弃
+熊
+摒
+凸
+拗
+穹
+蒙
+抒
+祛
+劝
+闫
+扳
+阵
+醌
+踪
+喵
+侣
+搬
+仅
+荧
+赎
+蝾
+琦
+买
+婧
+瞄
+寓
+皎
+冻
+赝
+箩
+莫
+瞰
+郊
+笫
+姝
+筒
+枪
+遣
+煸
+袋
+舆
+痱
+涛
+母
+〇
+启
+践
+耙
+绲
+盘
+遂
+昊
+搞
+槿
+诬
+纰
+泓
+惨
+檬
+亻
+越
+C
+o
+憩
+熵
+祷
+钒
+暧
+塔
+阗
+胰
+咄
+娶
+魔
+琶
+钞
+邻
+扬
+杉
+殴
+咽
+弓
+〆
+髻
+】
+吭
+揽
+霆
+拄
+殖
+脆
+彻
+岩
+芝
+勃
+辣
+剌
+钝
+嘎
+甄
+佘
+皖
+伦
+授
+徕
+憔
+挪
+皇
+庞
+稔
+芜
+踏
+溴
+兖
+卒
+擢
+饥
+鳞
+煲
+‰
+账
+颗
+叻
+斯
+捧
+鳍
+琮
+讹
+蛙
+纽
+谭
+酸
+兔
+莒
+睇
+伟
+觑
+羲
+嗜
+宜
+褐
+旎
+辛
+卦
+诘
+筋
+鎏
+溪
+挛
+熔
+阜
+晰
+鳅
+丢
+奚
+灸
+呱
+献
+陉
+黛
+鸪
+甾
+萨
+疮
+拯
+洲
+疹
+辑
+叙
+恻
+谒
+允
+柔
+烂
+氏
+逅
+漆
+拎
+惋
+扈
+湟
+纭
+啕
+掬
+擞
+哥
+忽
+涤
+鸵
+靡
+郗
+瓷
+扁
+廊
+怨
+雏
+钮
+敦
+E
+懦
+憋
+汀
+拚
+啉
+腌
+岸
+f
+痼
+瞅
+尊
+咀
+眩
+飙
+忌
+仝
+迦
+熬
+毫
+胯
+篑
+茄
+腺
+凄
+舛
+碴
+锵
+诧
+羯
+後
+漏
+汤
+宓
+仞
+蚁
+壶
+谰
+皑
+铄
+棰
+罔
+辅
+晶
+苦
+牟
+闽
+\
+烃
+饮
+聿
+丙
+蛳
+朱
+煤
+涔
+鳖
+犁
+罐
+荼
+砒
+淦
+妤
+黏
+戎
+孑
+婕
+瑾
+戢
+钵
+枣
+捋
+砥
+衩
+狙
+桠
+稣
+阎
+肃
+梏
+诫
+孪
+昶
+婊
+衫
+嗔
+侃
+塞
+蜃
+樵
+峒
+貌
+屿
+欺
+缫
+阐
+栖
+诟
+珞
+荭
+吝
+萍
+嗽
+恂
+啻
+蜴
+磬
+峋
+俸
+豫
+谎
+徊
+镍
+韬
+魇
+晴
+U
+囟
+猜
+蛮
+坐
+囿
+伴
+亭
+肝
+佗
+蝠
+妃
+胞
+滩
+榴
+氖
+垩
+苋
+砣
+扪
+馏
+姓
+轩
+厉
+夥
+侈
+禀
+垒
+岑
+赏
+钛
+辐
+痔
+披
+纸
+碳
+“
+坞
+蠓
+挤
+荥
+沅
+悔
+铧
+帼
+蒌
+蝇
+a
+p
+y
+n
+g
+哀
+浆
+瑶
+凿
+桶
+馈
+皮
+奴
+苜
+佤
+伶
+晗
+铱
+炬
+优
+弊
+氢
+恃
+甫
+攥
+端
+锌
+灰
+稹
+炝
+曙
+邋
+亥
+眶
+碾
+拉
+萝
+绔
+捷
+浍
+腋
+姑
+菖
+凌
+涞
+麽
+锢
+桨
+潢
+绎
+镰
+殆
+锑
+渝
+铬
+困
+绽
+觎
+匈
+糙
+暑
+裹
+鸟
+盔
+肽
+迷
+綦
+『
+亳
+佝
+俘
+钴
+觇
+骥
+仆
+疝
+跪
+婶
+郯
+瀹
+唉
+脖
+踞
+针
+晾
+忒
+扼
+瞩
+叛
+椒
+疟
+嗡
+邗
+肆
+跆
+玫
+忡
+捣
+咧
+唆
+艄
+蘑
+潦
+笛
+阚
+沸
+泻
+掊
+菽
+贫
+斥
+髂
+孢
+镂
+赂
+麝
+鸾
+屡
+衬
+苷
+恪
+叠
+希
+粤
+爻
+喝
+茫
+惬
+郸
+绻
+庸
+撅
+碟
+宄
+妹
+膛
+叮
+饵
+崛
+嗲
+椅
+冤
+搅
+咕
+敛
+尹
+垦
+闷
+蝉
+霎
+勰
+败
+蓑
+泸
+肤
+鹌
+幌
+焦
+浠
+鞍
+刁
+舰
+乙
+竿
+裔
+。
+茵
+函
+伊
+兄
+丨
+娜
+匍
+謇
+莪
+宥
+似
+蝽
+翳
+酪
+翠
+粑
+薇
+祢
+骏
+赠
+叫
+Q
+噤
+噻
+竖
+芗
+莠
+潭
+俊
+羿
+耜
+O
+郫
+趁
+嗪
+囚
+蹶
+芒
+洁
+笋
+鹑
+敲
+硝
+啶
+堡
+渲
+揩
+』
+携
+宿
+遒
+颍
+扭
+棱
+割
+萜
+蔸
+葵
+琴
+捂
+饰
+衙
+耿
+掠
+募
+岂
+窖
+涟
+蔺
+瘤
+柞
+瞪
+怜
+匹
+距
+楔
+炜
+哆
+秦
+缎
+幼
+茁
+绪
+痨
+恨
+楸
+娅
+瓦
+桩
+雪
+嬴
+伏
+榔
+妥
+铿
+拌
+眠
+雍
+缇
+‘
+卓
+搓
+哌
+觞
+噩
+屈
+哧
+髓
+咦
+巅
+娑
+侑
+淫
+膳
+祝
+勾
+姊
+莴
+胄
+疃
+薛
+蜷
+胛
+巷
+芙
+芋
+熙
+闰
+勿
+窃
+狱
+剩
+钏
+幢
+陟
+铛
+慧
+靴
+耍
+k
+浙
+浇
+飨
+惟
+绗
+祜
+澈
+啼
+咪
+磷
+摞
+诅
+郦
+抹
+跃
+壬
+吕
+肖
+琏
+颤
+尴
+剡
+抠
+凋
+赚
+泊
+津
+宕
+殷
+倔
+氲
+漫
+邺
+涎
+怠
+$
+垮
+荬
+遵
+俏
+叹
+噢
+饽
+蜘
+孙
+筵
+疼
+鞭
+羧
+牦
+箭
+潴
+c
+眸
+祭
+髯
+啖
+坳
+愁
+芩
+驮
+倡
+巽
+穰
+沃
+胚
+怒
+凤
+槛
+剂
+趵
+嫁
+v
+邢
+灯
+鄢
+桐
+睽
+檗
+锯
+槟
+婷
+嵋
+圻
+诗
+蕈
+颠
+遭
+痢
+芸
+怯
+馥
+竭
+锗
+徜
+恭
+遍
+籁
+剑
+嘱
+苡
+龄
+僧
+桑
+潸
+弘
+澶
+楹
+悲
+讫
+愤
+腥
+悸
+谍
+椹
+呢
+桓
+葭
+攫
+阀
+翰
+躲
+敖
+柑
+郎
+笨
+橇
+呃
+魁
+燎
+脓
+葩
+磋
+垛
+玺
+狮
+沓
+砜
+蕊
+锺
+罹
+蕉
+翱
+虐
+闾
+巫
+旦
+茱
+嬷
+枯
+鹏
+贡
+芹
+汛
+矫
+绁
+拣
+禺
+佃
+讣
+舫
+惯
+乳
+趋
+疲
+挽
+岚
+虾
+衾
+蠹
+蹂
+飓
+氦
+铖
+孩
+稞
+瑜
+壅
+掀
+勘
+妓
+畅
+髋
+W
+庐
+牲
+蓿
+榕
+练
+垣
+唱
+邸
+菲
+昆
+婺
+穿
+绡
+麒
+蚱
+掂
+愚
+泷
+涪
+漳
+妩
+娉
+榄
+讷
+觅
+旧
+藤
+煮
+呛
+柳
+腓
+叭
+庵
+烷
+阡
+罂
+蜕
+擂
+猖
+咿
+媲
+脉
+【
+沏
+貅
+黠
+熏
+哲
+烁
+坦
+酵
+兜
+×
+潇
+撒
+剽
+珩
+圹
+乾
+摸
+樟
+帽
+嗒
+襄
+魂
+轿
+憬
+锡
+〕
+喃
+皆
+咖
+隅
+脸
+残
+泮
+袂
+鹂
+珊
+囤
+捆
+咤
+误
+徨
+闹
+淙
+芊
+淋
+怆
+囗
+拨
+梳
+渤
+R
+G
+绨
+蚓
+婀
+幡
+狩
+麾
+谢
+唢
+裸
+旌
+伉
+纶
+裂
+驳
+砼
+咛
+澄
+樨
+蹈
+宙
+澍
+倍
+貔
+操
+勇
+蟠
+摈
+砧
+虬
+够
+缁
+悦
+藿
+撸
+艹
+摁
+淹
+豇
+虎
+榭
+ˉ
+吱
+d
+°
+喧
+荀
+踱
+侮
+奋
+偕
+饷
+犍
+惮
+坑
+璎
+徘
+宛
+妆
+袈
+倩
+窦
+昂
+荏
+乖
+K
+怅
+撰
+鳙
+牙
+袁
+酞
+X
+痿
+琼
+闸
+雁
+趾
+荚
+虻
+涝
+《
+杏
+韭
+偈
+烤
+绫
+鞘
+卉
+症
+遢
+蓥
+诋
+杭
+荨
+匆
+竣
+簪
+辙
+敕
+虞
+丹
+缭
+咩
+黟
+m
+淤
+瑕
+咂
+铉
+硼
+茨
+嶂
+痒
+畸
+敬
+涿
+粪
+窘
+熟
+叔
+嫔
+盾
+忱
+裘
+憾
+梵
+赡
+珙
+咯
+娘
+庙
+溯
+胺
+葱
+痪
+摊
+荷
+卞
+乒
+髦
+寐
+铭
+坩
+胗
+枷
+爆
+溟
+嚼
+羚
+砬
+轨
+惊
+挠
+罄
+竽
+菏
+氧
+浅
+楣
+盼
+枢
+炸
+阆
+杯
+谏
+噬
+淇
+渺
+俪
+秆
+墓
+泪
+跻
+砌
+痰
+垡
+渡
+耽
+釜
+讶
+鳎
+煞
+呗
+韶
+舶
+绷
+鹳
+缜
+旷
+铊
+皱
+龌
+檀
+霖
+奄
+槐
+艳
+蝶
+旋
+哝
+赶
+骞
+蚧
+腊
+盈
+丁
+`
+蜚
+矸
+蝙
+睨
+嚓
+僻
+鬼
+醴
+夜
+彝
+磊
+笔
+拔
+栀
+糕
+厦
+邰
+纫
+逭
+纤
+眦
+膊
+馍
+躇
+烯
+蘼
+冬
+诤
+暄
+骶
+哑
+瘠
+」
+臊
+丕
+愈
+咱
+螺
+擅
+跋
+搏
+硪
+谄
+笠
+淡
+嘿
+骅
+谧
+鼎
+皋
+姚
+歼
+蠢
+驼
+耳
+胬
+挝
+涯
+狗
+蒽
+孓
+犷
+凉
+芦
+箴
+铤
+孤
+嘛
+坤
+V
+茴
+朦
+挞
+尖
+橙
+诞
+搴
+碇
+洵
+浚
+帚
+蜍
+漯
+柘
+嚎
+讽
+芭
+荤
+咻
+祠
+秉
+跖
+埃
+吓
+糯
+眷
+馒
+惹
+娼
+鲑
+嫩
+讴
+轮
+瞥
+靶
+褚
+乏
+缤
+宋
+帧
+删
+驱
+碎
+扑
+俩
+俄
+偏
+涣
+竹
+噱
+皙
+佰
+渚
+唧
+斡
+#
+镉
+刀
+崎
+筐
+佣
+夭
+贰
+肴
+峙
+哔
+艿
+匐
+牺
+镛
+缘
+仡
+嫡
+劣
+枸
+堀
+梨
+簿
+鸭
+蒸
+亦
+稽
+浴
+{
+衢
+束
+槲
+j
+阁
+揍
+疥
+棋
+潋
+聪
+窜
+乓
+睛
+插
+冉
+阪
+苍
+搽
+「
+蟾
+螟
+幸
+仇
+樽
+撂
+慢
+跤
+幔
+俚
+淅
+覃
+觊
+溶
+妖
+帛
+侨
+曰
+妾
+泗
+·
+：
+瀘
+風
+Ë
+（
+）
+∶
+紅
+紗
+瑭
+雲
+頭
+鶏
+財
+許
+•
+¥
+樂
+焗
+麗
+—
+；
+滙
+東
+榮
+繪
+興
+…
+門
+業
+π
+楊
+國
+顧
+é
+盤
+寳
+Λ
+龍
+鳳
+島
+誌
+緣
+結
+銭
+萬
+勝
+祎
+璟
+優
+歡
+臨
+時
+購
+＝
+★
+藍
+昇
+鐵
+觀
+勅
+農
+聲
+畫
+兿
+術
+發
+劉
+記
+專
+耑
+園
+書
+壴
+種
+Ο
+●
+褀
+號
+銀
+匯
+敟
+锘
+葉
+橪
+廣
+進
+蒄
+鑽
+阝
+祙
+貢
+鍋
+豊
+夬
+喆
+團
+閣
+開
+燁
+賓
+館
+酡
+沔
+順
+＋
+硚
+劵
+饸
+陽
+車
+湓
+復
+萊
+氣
+軒
+華
+堃
+迮
+纟
+戶
+馬
+學
+裡
+電
+嶽
+獨
+マ
+シ
+サ
+ジ
+燘
+袪
+環
+❤
+臺
+灣
+専
+賣
+孖
+聖
+攝
+線
+▪
+α
+傢
+俬
+夢
+達
+莊
+喬
+貝
+薩
+劍
+羅
+壓
+棛
+饦
+尃
+璈
+囍
+醫
+Ｇ
+Ｉ
+Ａ
+＃
+Ｎ
+鷄
+髙
+嬰
+啓
+約
+隹
+潔
+賴
+藝
+～
+寶
+籣
+麺
+　
+嶺
+√
+義
+網
+峩
+長
+∧
+魚
+機
+構
+②
+鳯
+偉
+Ｌ
+Ｂ
+㙟
+畵
+鴿
+＇
+詩
+溝
+嚞
+屌
+藔
+佧
+玥
+蘭
+織
+１
+３
+９
+０
+７
+點
+砭
+鴨
+鋪
+銘
+廳
+弍
+‧
+創
+湯
+坶
+℃
+卩
+骝
+＆
+烜
+荘
+當
+潤
+扞
+係
+懷
+碶
+钅
+蚨
+讠
+☆
+叢
+爲
+埗
+涫
+塗
+→
+楽
+現
+鯨
+愛
+瑪
+鈺
+忄
+悶
+藥
+飾
+樓
+視
+孬
+ㆍ
+燚
+苪
+師
+①
+丼
+锽
+│
+韓
+標
+è
+兒
+閏
+匋
+張
+漢
+Ü
+髪
+會
+閑
+檔
+習
+裝
+の
+峯
+菘
+輝
+И
+雞
+釣
+億
+浐
+Ｋ
+Ｏ
+Ｒ
+８
+Ｈ
+Ｅ
+Ｐ
+Ｔ
+Ｗ
+Ｄ
+Ｓ
+Ｃ
+Ｍ
+Ｆ
+姌
+饹
+»
+晞
+廰
+ä
+嵯
+鷹
+負
+飲
+絲
+冚
+楗
+澤
+綫
+區
+❋
+←
+質
+靑
+揚
+③
+滬
+統
+産
+協
+﹑
+乸
+畐
+經
+運
+際
+洺
+岽
+為
+粵
+諾
+崋
+豐
+碁
+ɔ
+Ｖ
+２
+６
+齋
+誠
+訂
+´
+勑
+雙
+陳
+無
+í
+泩
+媄
+夌
+刂
+ｉ
+ｃ
+ｔ
+ｏ
+ｒ
+ａ
+嘢
+耄
+燴
+暃
+壽
+媽
+靈
+抻
+體
+唻
+É
+冮
+甹
+鎮
+錦
+ʌ
+蜛
+蠄
+尓
+駕
+戀
+飬
+逹
+倫
+貴
+極
+Я
+Й
+寬
+磚
+嶪
+郎
+職
+｜
+間
+ｎ
+ｄ
+剎
+伈
+課
+飛
+橋
+瘊
+№
+譜
+骓
+圗
+滘
+縣
+粿
+咅
+養
+濤
+彳
+®
+％
+Ⅱ
+啰
+㴪
+見
+矞
+薬
+糁
+邨
+鲮
+顔
+罱
+З
+選
+話
+贏
+氪
+俵
+競
+瑩
+繡
+枱
+β
+綉
+á
+獅
+爾
+™
+麵
+戋
+淩
+徳
+個
+劇
+場
+務
+簡
+寵
+ｈ
+實
+膠
+轱
+圖
+築
+嘣
+樹
+㸃
+營
+耵
+孫
+饃
+鄺
+飯
+麯
+遠
+輸
+坫
+孃
+乚
+閃
+鏢
+㎡
+題
+廠
+關
+↑
+爺
+將
+軍
+連
+篦
+覌
+參
+箸
+－
+窠
+棽
+寕
+夀
+爰
+歐
+呙
+閥
+頡
+熱
+雎
+垟
+裟
+凬
+勁
+帑
+馕
+夆
+疌
+枼
+馮
+貨
+蒤
+樸
+彧
+旸
+靜
+龢
+暢
+㐱
+鳥
+珺
+鏡
+灡
+爭
+堷
+廚
+Ó
+騰
+診
+┅
+蘇
+褔
+凱
+頂
+豕
+亞
+帥
+嘬
+⊥
+仺
+桖
+複
+饣
+絡
+穂
+顏
+棟
+納
+▏
+濟
+親
+設
+計
+攵
+埌
+烺
+ò
+頤
+燦
+蓮
+撻
+節
+講
+濱
+濃
+娽
+洳
+朿
+燈
+鈴
+護
+膚
+铔
+過
+補
+Ｚ
+Ｕ
+５
+４
+坋
+闿
+䖝
+餘
+缐
+铞
+貿
+铪
+桼
+趙
+鍊
+［
+㐂
+垚
+菓
+揸
+捲
+鐘
+滏
+𣇉
+爍
+輪
+燜
+鴻
+鮮
+動
+鹞
+鷗
+丄
+慶
+鉌
+翥
+飮
+腸
+⇋
+漁
+覺
+來
+熘
+昴
+翏
+鲱
+圧
+鄉
+萭
+頔
+爐
+嫚
+г
+貭
+類
+聯
+幛
+輕
+訓
+鑒
+夋
+锨
+芃
+珣
+䝉
+扙
+嵐
+銷
+處
+ㄱ
+語
+誘
+苝
+歸
+儀
+燒
+楿
+內
+粢
+葒
+奧
+麥
+礻
+滿
+蠔
+穵
+瞭
+態
+鱬
+榞
+硂
+鄭
+黃
+煙
+祐
+奓
+逺
+＊
+瑄
+獲
+聞
+薦
+讀
+這
+樣
+決
+問
+啟
+們
+執
+説
+轉
+單
+隨
+唘
+帶
+倉
+庫
+還
+贈
+尙
+皺
+■
+餅
+產
+○
+∈
+報
+狀
+楓
+賠
+琯
+嗮
+禮
+｀
+傳
+＞
+≤
+嗞
+Φ
+≥
+換
+咭
+∣
+↓
+曬
+ε
+応
+寫
+″
+終
+様
+純
+費
+療
+聨
+凍
+壐
+郵
+ü
+黒
+∫
+製
+塊
+調
+軽
+確
+撃
+級
+馴
+Ⅲ
+涇
+繹
+數
+碼
+證
+狒
+処
+劑
+＜
+晧
+賀
+衆
+］
+櫥
+兩
+陰
+絶
+對
+鯉
+憶
+◎
+ｐ
+ｅ
+Ｙ
+蕒
+煖
+頓
+測
+試
+鼽
+僑
+碩
+妝
+帯
+≈
+鐡
+舖
+權
+喫
+倆
+ˋ
+該
+悅
+ā
+俫
+．
+ｆ
+ｓ
+ｂ
+ｍ
+ｋ
+ｇ
+ｕ
+ｊ
+貼
+淨
+濕
+針
+適
+備
+ｌ
+／
+給
+謢
+強
+觸
+衛
+與
+⊙
+＄
+緯
+變
+⑴
+⑵
+⑶
+㎏
+殺
+∩
+幚
+─
+價
+▲
+離
+ú
+ó
+飄
+烏
+関
+閟
+﹝
+﹞
+邏
+輯
+鍵
+驗
+訣
+導
+歷
+屆
+層
+▼
+儱
+錄
+熳
+ē
+艦
+吋
+錶
+辧
+飼
+顯
+④
+禦
+販
+気
+対
+枰
+閩
+紀
+幹
+瞓
+貊
+淚
+△
+眞
+墊
+Ω
+獻
+褲
+縫
+緑
+亜
+鉅
+餠
+｛
+｝
+◆
+蘆
+薈
+█
+◇
+溫
+彈
+晳
+粧
+犸
+穩
+訊
+崬
+凖
+熥
+П
+舊
+條
+紋
+圍
+Ⅳ
+筆
+尷
+難
+雜
+錯
+綁
+識
+頰
+鎖
+艶
+□
+殁
+殼
+⑧
+├
+▕
+鵬
+ǐ
+ō
+ǒ
+糝
+綱
+▎
+μ
+盜
+饅
+醬
+籤
+蓋
+釀
+鹽
+據
+à
+ɡ
+辦
+◥
+彐
+┌
+婦
+獸
+鲩
+伱
+ī
+蒟
+蒻
+齊
+袆
+腦
+寧
+凈
+妳
+煥
+詢
+偽
+謹
+啫
+鯽
+騷
+鱸
+損
+傷
+鎻
+髮
+買
+冏
+儥
+両
+﹢
+∞
+載
+喰
+ｚ
+羙
+悵
+燙
+曉
+員
+組
+徹
+艷
+痠
+鋼
+鼙
+縮
+細
+嚒
+爯
+≠
+維
+＂
+鱻
+壇
+厍
+帰
+浥
+犇
+薡
+軎
+²
+應
+醜
+刪
+緻
+鶴
+賜
+噁
+軌
+尨
+镔
+鷺
+槗
+彌
+葚
+濛
+請
+溇
+緹
+賢
+訪
+獴
+瑅
+資
+縤
+陣
+蕟
+栢
+韻
+祼
+恁
+伢
+謝
+劃
+涑
+總
+衖
+踺
+砋
+凉
+籃
+駿
+苼
+瘋
+昽
+紡
+驊
+腎
+﹗
+響
+杋
+剛
+嚴
+禪
+歓
+槍
+傘
+檸
+檫
+炣
+勢
+鏜
+鎢
+銑
+尐
+減
+奪
+惡
+θ
+僮
+婭
+臘
+ū
+ì
+殻
+鉄
+∑
+蛲
+焼
+緖
+續
+紹
+懮
diff --git a/ppocr/utils/profiler.py b/ppocr/utils/profiler.py
new file mode 100644
index 0000000..e4e3e05
--- /dev/null
+++ b/ppocr/utils/profiler.py
@@ -0,0 +1,130 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import sys
+import paddle.profiler as profiler
+
+# A global variable to record the number of calling times for profiler
+# functions. It is used to specify the tracing range of training steps.
+_profiler_step_id = 0
+
+# A global variable to avoid parsing from string every time.
+_profiler_options = None
+_prof = None
+
+
+class ProfilerOptions(object):
+    """
+    Use a string to initialize a ProfilerOptions.
+    The string should be in the format: "key1=value1;key2=value;key3=value3".
+    For example:
+      "profile_path=model.profile"
+      "batch_range=[50, 60]; profile_path=model.profile"
+      "batch_range=[50, 60]; tracer_option=OpDetail; profile_path=model.profile"
+
+    ProfilerOptions supports following key-value pair:
+      batch_range      - a integer list, e.g. [100, 110].
+      state            - a string, the optional values are 'CPU', 'GPU' or 'All'.
+      sorted_key       - a string, the optional values are 'calls', 'total',
+                         'max', 'min' or 'ave.
+      tracer_option    - a string, the optional values are 'Default', 'OpDetail',
+                         'AllOpDetail'.
+      profile_path     - a string, the path to save the serialized profile data,
+                         which can be used to generate a timeline.
+      exit_on_finished - a boolean.
+    """
+
+    def __init__(self, options_str):
+        assert isinstance(options_str, str)
+
+        self._options = {
+            "batch_range": [10, 20],
+            "state": "All",
+            "sorted_key": "total",
+            "tracer_option": "Default",
+            "profile_path": "/tmp/profile",
+            "exit_on_finished": True,
+            "timer_only": True,
+        }
+        self._parse_from_string(options_str)
+
+    def _parse_from_string(self, options_str):
+        for kv in options_str.replace(" ", "").split(";"):
+            key, value = kv.split("=")
+            if key == "batch_range":
+                value_list = value.replace("[", "").replace("]", "").split(",")
+                value_list = list(map(int, value_list))
+                if (
+                    len(value_list) >= 2
+                    and value_list[0] >= 0
+                    and value_list[1] > value_list[0]
+                ):
+                    self._options[key] = value_list
+            elif key == "exit_on_finished":
+                self._options[key] = value.lower() in ("yes", "true", "t", "1")
+            elif key in ["state", "sorted_key", "tracer_option", "profile_path"]:
+                self._options[key] = value
+            elif key == "timer_only":
+                self._options[key] = value
+
+    def __getitem__(self, name):
+        if self._options.get(name, None) is None:
+            raise ValueError("ProfilerOptions does not have an option named %s." % name)
+        return self._options[name]
+
+
+def add_profiler_step(options_str=None):
+    """
+    Enable the operator-level timing using PaddlePaddle's profiler.
+    The profiler uses a independent variable to count the profiler steps.
+    One call of this function is treated as a profiler step.
+    Args:
+      profiler_options - a string to initialize the ProfilerOptions.
+                         Default is None, and the profiler is disabled.
+    """
+    if options_str is None:
+        return
+
+    global _prof
+    global _profiler_step_id
+    global _profiler_options
+
+    if _profiler_options is None:
+        _profiler_options = ProfilerOptions(options_str)
+    # profile : https://www.paddlepaddle.org.cn/documentation/docs/zh/guides/performance_improving/profiling_model.html#chakanxingnengshujudetongjibiaodan
+    # timer_only = True  only the model's throughput and time overhead are displayed
+    # timer_only = False calling summary can print a statistical form that presents performance data from different perspectives.
+    # timer_only = False the output Timeline information can be found in the profiler_log directory
+    if _prof is None:
+        _timer_only = str(_profiler_options["timer_only"]) == str(True)
+        _prof = profiler.Profiler(
+            scheduler=(
+                _profiler_options["batch_range"][0],
+                _profiler_options["batch_range"][1],
+            ),
+            on_trace_ready=profiler.export_chrome_tracing("./profiler_log"),
+            timer_only=_timer_only,
+        )
+        _prof.start()
+    else:
+        _prof.step()
+
+    if _profiler_step_id == _profiler_options["batch_range"][1]:
+        _prof.stop()
+        _prof.summary(op_detail=True, thread_sep=False, time_unit="ms")
+        _prof = None
+        if _profiler_options["exit_on_finished"]:
+            sys.exit(0)
+
+    _profiler_step_id += 1
diff --git a/ppocr/utils/save_load.py b/ppocr/utils/save_load.py
new file mode 100644
index 0000000..73e79bc
--- /dev/null
+++ b/ppocr/utils/save_load.py
@@ -0,0 +1,439 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import errno
+import os
+import pickle
+import json
+from packaging import version
+
+import paddle
+
+from ppocr.utils.logging import get_logger
+from ppocr.utils.network import maybe_download_params
+
+try:
+    import encryption  # Attempt to import the encryption module for AIStudio's encryption model
+
+    encrypted = encryption.is_encryption_needed()
+except ImportError:
+    print("Skipping import of the encryption module.")
+    encrypted = False  # Encryption is not needed if the module cannot be imported
+
+__all__ = ["load_model"]
+
+
+# just to determine the inference model file format
+def get_FLAGS_json_format_model():
+    # json format by default
+    return os.environ.get("FLAGS_json_format_model", "1").lower() in ("1", "true", "t")
+
+
+FLAGS_json_format_model = get_FLAGS_json_format_model()
+
+
+def _mkdir_if_not_exist(path, logger):
+    """
+    mkdir if not exists, ignore the exception when multiprocess mkdir together
+    """
+    if not os.path.exists(path):
+        try:
+            os.makedirs(path)
+        except OSError as e:
+            if e.errno == errno.EEXIST and os.path.isdir(path):
+                logger.warning(
+                    "be happy if some process has already created {}".format(path)
+                )
+            else:
+                raise OSError("Failed to mkdir {}".format(path))
+
+
+def load_model(config, model, optimizer=None, model_type="det", ema=None):
+    """
+    load model from checkpoint or pretrained_model
+    """
+    logger = get_logger()
+    global_config = config["Global"]
+    checkpoints = global_config.get("checkpoints")
+    pretrained_model = global_config.get("pretrained_model")
+    best_model_dict = {}
+    is_float16 = False
+    is_nlp_model = model_type == "kie" and config["Architecture"]["algorithm"] not in [
+        "SDMGR"
+    ]
+
+    if is_nlp_model is True:
+        # NOTE: for kie model dsitillation, resume training is not supported now
+        if config["Architecture"]["algorithm"] in ["Distillation"]:
+            return best_model_dict
+        checkpoints = config["Architecture"]["Backbone"]["checkpoints"]
+        # load kie method metric
+        if checkpoints:
+            if os.path.exists(os.path.join(checkpoints, "metric.states")):
+                with open(os.path.join(checkpoints, "metric.states"), "rb") as f:
+                    states_dict = pickle.load(f, encoding="latin1")
+                best_model_dict = states_dict.get("best_model_dict", {})
+                if "epoch" in states_dict:
+                    best_model_dict["start_epoch"] = states_dict["epoch"] + 1
+            logger.info("resume from {}".format(checkpoints))
+
+            if optimizer is not None:
+                if checkpoints[-1] in ["/", "\\"]:
+                    checkpoints = checkpoints[:-1]
+                if os.path.exists(checkpoints + ".pdopt"):
+                    optim_dict = paddle.load(checkpoints + ".pdopt")
+                    optimizer.set_state_dict(optim_dict)
+                else:
+                    logger.warning(
+                        "{}.pdopt is not exists, params of optimizer is not loaded".format(
+                            checkpoints
+                        )
+                    )
+
+        return best_model_dict
+
+    if checkpoints:
+        if checkpoints.endswith(".pdparams"):
+            checkpoints = checkpoints.replace(".pdparams", "")
+        assert os.path.exists(
+            checkpoints + ".pdparams"
+        ), "The {}.pdparams does not exists!".format(checkpoints)
+
+        # load params from trained model
+        params = paddle.load(checkpoints + ".pdparams")
+        state_dict = model.state_dict()
+        new_state_dict = {}
+        for key, value in state_dict.items():
+            if key not in params:
+                logger.warning(
+                    "{} not in loaded params {} !".format(key, params.keys())
+                )
+                continue
+            pre_value = params[key]
+            if pre_value.dtype == paddle.float16:
+                is_float16 = True
+            if pre_value.dtype != value.dtype:
+                pre_value = pre_value.astype(value.dtype)
+            if list(value.shape) == list(pre_value.shape):
+                new_state_dict[key] = pre_value
+            else:
+                logger.warning(
+                    "The shape of model params {} {} not matched with loaded params shape {} !".format(
+                        key, value.shape, pre_value.shape
+                    )
+                )
+        model.set_state_dict(new_state_dict)
+        if is_float16:
+            logger.info(
+                "The parameter type is float16, which is converted to float32 when loading"
+            )
+        if optimizer is not None:
+            if os.path.exists(checkpoints + ".pdopt"):
+                optim_dict = paddle.load(checkpoints + ".pdopt")
+                optimizer.set_state_dict(optim_dict)
+            else:
+                logger.warning(
+                    "{}.pdopt is not exists, params of optimizer is not loaded".format(
+                        checkpoints
+                    )
+                )
+
+        if os.path.exists(checkpoints + ".states"):
+            with open(checkpoints + ".states", "rb") as f:
+                states_dict = pickle.load(f, encoding="latin1")
+            best_model_dict = states_dict.get("best_model_dict", {})
+            best_model_dict["acc"] = 0.0
+            if "epoch" in states_dict:
+                best_model_dict["start_epoch"] = states_dict["epoch"] + 1
+        logger.info("resume from {}".format(checkpoints))
+
+        # Restore EMA state if available
+        if ema is not None:
+            pdema_path = checkpoints + ".pdema"
+            if os.path.exists(pdema_path):
+                ema_data = paddle.load(pdema_path)
+                # .pdparams contains EMA weights; restore original training weights
+                if "train_state" in ema_data:
+                    train_sd = ema_data["train_state"]
+                    cur_sd = model.state_dict()
+                    for k in cur_sd:
+                        if k in train_sd:
+                            if list(cur_sd[k].shape) == list(train_sd[k].shape):
+                                cur_sd[k] = train_sd[k]
+                    model.set_state_dict(cur_sd)
+                    logger.info(
+                        "EMA: restored training weights from {}".format(pdema_path)
+                    )
+                # Restore EMA shadow weights + step
+                if "ema_state" in ema_data and "step" in ema_data:
+                    ema.set_state_dict(ema_data)
+                    logger.info(
+                        "EMA: restored shadow weights (step={}) from {}".format(
+                            ema_data["step"], pdema_path
+                        )
+                    )
+
+    elif pretrained_model:
+        is_float16 = load_pretrained_params(model, pretrained_model)
+    else:
+        logger.info("train from scratch")
+    best_model_dict["is_float16"] = is_float16
+    return best_model_dict
+
+
+def load_pretrained_params(model, path):
+    logger = get_logger()
+    path = maybe_download_params(path)
+    if path.endswith(".pdparams"):
+        path = path.replace(".pdparams", "")
+    assert os.path.exists(
+        path + ".pdparams"
+    ), "The {}.pdparams does not exists!".format(path)
+
+    params = paddle.load(path + ".pdparams")
+
+    state_dict = model.state_dict()
+
+    new_state_dict = {}
+    is_float16 = False
+
+    for k1 in params.keys():
+        if k1 not in state_dict.keys():
+            logger.warning("The pretrained params {} not in model".format(k1))
+        else:
+            if params[k1].dtype == paddle.float16:
+                is_float16 = True
+            if params[k1].dtype != state_dict[k1].dtype:
+                params[k1] = params[k1].astype(state_dict[k1].dtype)
+            if list(state_dict[k1].shape) == list(params[k1].shape):
+                new_state_dict[k1] = params[k1]
+            else:
+                logger.warning(
+                    "The shape of model params {} {} not matched with loaded params {} {} !".format(
+                        k1, state_dict[k1].shape, k1, params[k1].shape
+                    )
+                )
+
+    model.set_state_dict(new_state_dict)
+    if is_float16:
+        logger.info(
+            "The parameter type is float16, which is converted to float32 when loading"
+        )
+    logger.info("load pretrain successful from {}".format(path))
+    return is_float16
+
+
+def save_model(
+    model,
+    optimizer,
+    model_path,
+    logger,
+    config,
+    is_best=False,
+    prefix="ppocr",
+    ema=None,
+    train_state=None,
+    **kwargs,
+):
+    """
+    save model to the target path
+    """
+    _mkdir_if_not_exist(model_path, logger)
+    model_prefix = os.path.join(model_path, prefix)
+
+    if prefix == "best_accuracy":
+        best_model_path = os.path.join(model_path, "best_model")
+        _mkdir_if_not_exist(best_model_path, logger)
+
+    paddle.save(optimizer.state_dict(), model_prefix + ".pdopt")
+    if prefix == "best_accuracy":
+        paddle.save(
+            optimizer.state_dict(), os.path.join(best_model_path, "model.pdopt")
+        )
+
+    is_nlp_model = config["Architecture"]["model_type"] == "kie" and config[
+        "Architecture"
+    ]["algorithm"] not in ["SDMGR"]
+    if is_nlp_model is not True:
+        paddle.save(model.state_dict(), model_prefix + ".pdparams")
+        metric_prefix = model_prefix
+
+        # Save EMA state for training resumption
+        if ema is not None and train_state is not None:
+            paddle.save(
+                {
+                    "train_state": train_state,
+                    "ema_state": ema.state_dict,
+                    "step": ema.step,
+                },
+                model_prefix + ".pdema",
+            )
+
+        if prefix == "best_accuracy":
+            paddle.save(
+                model.state_dict(), os.path.join(best_model_path, "model.pdparams")
+            )
+            if ema is not None and train_state is not None:
+                paddle.save(
+                    {
+                        "train_state": train_state,
+                        "ema_state": ema.state_dict,
+                        "step": ema.step,
+                    },
+                    os.path.join(best_model_path, "model.pdema"),
+                )
+
+    else:  # for kie system, we follow the save/load rules in NLP
+        if config["Global"]["distributed"]:
+            arch = model._layers
+        else:
+            arch = model
+        if config["Architecture"]["algorithm"] in ["Distillation"]:
+            arch = arch.Student
+        arch.backbone.model.save_pretrained(model_prefix)
+        metric_prefix = os.path.join(model_prefix, "metric")
+
+        if prefix == "best_accuracy":
+            arch.backbone.model.save_pretrained(best_model_path)
+
+    save_model_info = kwargs.pop("save_model_info", False)
+    if save_model_info:
+        with open(os.path.join(model_path, f"{prefix}.info.json"), "w") as f:
+            json.dump(kwargs, f)
+        logger.info("Already save model info in {}".format(model_path))
+        if prefix != "latest":
+            done_flag = kwargs.pop("done_flag", False)
+            update_train_results(config, prefix, save_model_info, done_flag=done_flag)
+
+    # save metric and config
+    with open(metric_prefix + ".states", "wb") as f:
+        pickle.dump(kwargs, f, protocol=2)
+    if is_best:
+        logger.info("save best model is to {}".format(model_prefix))
+    else:
+        logger.info("save model in {}".format(model_prefix))
+
+
+def update_train_results(config, prefix, metric_info, done_flag=False, last_num=5):
+    if paddle.distributed.get_rank() != 0:
+        return
+
+    assert last_num >= 1
+    train_results_path = os.path.join(
+        config["Global"]["save_model_dir"], "train_result.json"
+    )
+    save_model_tag = ["pdparams", "pdopt", "pdstates"]
+    paddle_version = version.parse(paddle.__version__)
+    if FLAGS_json_format_model or paddle_version >= version.parse("3.0.0"):
+        save_inference_files = {
+            "inference_config": "inference.yml",
+            "pdmodel": "inference.json",
+            "pdiparams": "inference.pdiparams",
+        }
+    else:
+        save_inference_files = {
+            "inference_config": "inference.yml",
+            "pdmodel": "inference.pdmodel",
+            "pdiparams": "inference.pdiparams",
+            "pdiparams.info": "inference.pdiparams.info",
+        }
+    if os.path.exists(train_results_path):
+        with open(train_results_path, "r") as fp:
+            train_results = json.load(fp)
+    else:
+        train_results = {}
+        train_results["model_name"] = config["Global"]["model_name"]
+        label_dict_path = config["Global"].get("character_dict_path", "")
+        if label_dict_path != "":
+            label_dict_path = os.path.abspath(label_dict_path)
+            if not os.path.exists(label_dict_path):
+                label_dict_path = ""
+        train_results["label_dict"] = label_dict_path
+        train_results["train_log"] = "train.log"
+        train_results["visualdl_log"] = ""
+        train_results["config"] = "config.yaml"
+        train_results["models"] = {}
+        for i in range(1, last_num + 1):
+            train_results["models"][f"last_{i}"] = {}
+        train_results["models"]["best"] = {}
+    train_results["done_flag"] = done_flag
+    if "best" in prefix:
+        if "acc" in metric_info["metric"]:
+            metric_score = metric_info["metric"]["acc"]
+        elif "precision" in metric_info["metric"]:
+            metric_score = metric_info["metric"]["precision"]
+        elif "exp_rate" in metric_info["metric"]:
+            metric_score = metric_info["metric"]["exp_rate"]
+        else:
+            raise ValueError("No metric score found.")
+        train_results["models"]["best"]["score"] = metric_score
+        for tag in save_model_tag:
+            if tag == "pdparams" and encrypted:
+                train_results["models"]["best"][tag] = os.path.join(
+                    prefix,
+                    (
+                        f"{prefix}.encrypted.{tag}"
+                        if tag != "pdstates"
+                        else f"{prefix}.states"
+                    ),
+                )
+            else:
+                train_results["models"]["best"][tag] = os.path.join(
+                    prefix,
+                    f"{prefix}.{tag}" if tag != "pdstates" else f"{prefix}.states",
+                )
+        for key in save_inference_files:
+            train_results["models"]["best"][key] = os.path.join(
+                prefix, "inference", save_inference_files[key]
+            )
+    else:
+        for i in range(last_num - 1, 0, -1):
+            train_results["models"][f"last_{i + 1}"] = train_results["models"][
+                f"last_{i}"
+            ].copy()
+        if "acc" in metric_info["metric"]:
+            metric_score = metric_info["metric"]["acc"]
+        elif "precision" in metric_info["metric"]:
+            metric_score = metric_info["metric"]["precision"]
+        elif "exp_rate" in metric_info["metric"]:
+            metric_score = metric_info["metric"]["exp_rate"]
+        else:
+            metric_score = 0
+        train_results["models"][f"last_{1}"]["score"] = metric_score
+        for tag in save_model_tag:
+            if tag == "pdparams" and encrypted:
+                train_results["models"][f"last_{1}"][tag] = os.path.join(
+                    prefix,
+                    (
+                        f"{prefix}.encrypted.{tag}"
+                        if tag != "pdstates"
+                        else f"{prefix}.states"
+                    ),
+                )
+            else:
+                train_results["models"][f"last_{1}"][tag] = os.path.join(
+                    prefix,
+                    f"{prefix}.{tag}" if tag != "pdstates" else f"{prefix}.states",
+                )
+        for key in save_inference_files:
+            train_results["models"][f"last_{1}"][key] = os.path.join(
+                prefix, "inference", save_inference_files[key]
+            )
+
+    with open(train_results_path, "w") as fp:
+        json.dump(train_results, fp)
diff --git a/ppocr/utils/stats.py b/ppocr/utils/stats.py
new file mode 100755
index 0000000..6dd8c58
--- /dev/null
+++ b/ppocr/utils/stats.py
@@ -0,0 +1,70 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import collections
+import numpy as np
+import datetime
+
+__all__ = ["TrainingStats", "Time"]
+
+
+class SmoothedValue(object):
+    """Track a series of values and provide access to smoothed values over a
+    window or the global series average.
+    """
+
+    def __init__(self, window_size):
+        self.deque = collections.deque(maxlen=window_size)
+
+    def add_value(self, value):
+        self.deque.append(value)
+
+    def get_median_value(self):
+        return np.median(self.deque)
+
+
+def Time():
+    return datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S.%f")
+
+
+class TrainingStats(object):
+    def __init__(self, window_size, stats_keys):
+        self.window_size = window_size
+        self.smoothed_losses_and_metrics = {
+            key: SmoothedValue(window_size) for key in stats_keys
+        }
+
+    def update(self, stats):
+        for k, v in stats.items():
+            if k not in self.smoothed_losses_and_metrics:
+                self.smoothed_losses_and_metrics[k] = SmoothedValue(self.window_size)
+            self.smoothed_losses_and_metrics[k].add_value(v)
+
+    def get(self, extras=None):
+        stats = collections.OrderedDict()
+        if extras:
+            for k, v in extras.items():
+                stats[k] = v
+        for k, v in self.smoothed_losses_and_metrics.items():
+            stats[k] = round(v.get_median_value(), 6)
+
+        return stats
+
+    def log(self, extras=None):
+        d = self.get(extras)
+        strs = []
+        for k, v in d.items():
+            strs.append("{}: {:x<6f}".format(k, v))
+        strs = ", ".join(strs)
+        return strs
diff --git a/ppocr/utils/utility.py b/ppocr/utils/utility.py
new file mode 100644
index 0000000..6898ed4
--- /dev/null
+++ b/ppocr/utils/utility.py
@@ -0,0 +1,216 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import logging
+import os
+import cv2
+import random
+import numpy as np
+import paddle
+import importlib.util
+import sys
+import subprocess
+
+
+def print_dict(d, logger, delimiter=0):
+    """
+    Recursively visualize a dict and
+    indenting acrrording by the relationship of keys.
+    """
+    for k, v in sorted(d.items()):
+        if isinstance(v, dict):
+            logger.info("{}{} : ".format(delimiter * " ", str(k)))
+            print_dict(v, logger, delimiter + 4)
+        elif isinstance(v, list) and len(v) >= 1 and isinstance(v[0], dict):
+            logger.info("{}{} : ".format(delimiter * " ", str(k)))
+            for value in v:
+                print_dict(value, logger, delimiter + 4)
+        else:
+            logger.info("{}{} : {}".format(delimiter * " ", k, v))
+
+
+def get_check_global_params(mode):
+    check_params = [
+        "use_gpu",
+        "max_text_length",
+        "image_shape",
+        "image_shape",
+        "character_type",
+        "loss_type",
+    ]
+    if mode == "train_eval":
+        check_params = check_params + [
+            "train_batch_size_per_card",
+            "test_batch_size_per_card",
+        ]
+    elif mode == "test":
+        check_params = check_params + ["test_batch_size_per_card"]
+    return check_params
+
+
+def _check_image_file(path):
+    img_end = {"jpg", "bmp", "png", "jpeg", "rgb", "tif", "tiff", "gif", "pdf"}
+    return any([path.lower().endswith(e) for e in img_end])
+
+
+def get_image_file_list(img_file, infer_list=None):
+    imgs_lists = []
+    if infer_list and not os.path.exists(infer_list):
+        raise Exception("not found infer list {}".format(infer_list))
+    if infer_list:
+        with open(infer_list, "r") as f:
+            lines = f.readlines()
+        for line in lines:
+            image_path = line.strip().split("\t")[0]
+            image_path = os.path.join(img_file, image_path)
+            imgs_lists.append(image_path)
+    else:
+        if img_file is None or not os.path.exists(img_file):
+            raise Exception("not found any img file in {}".format(img_file))
+
+        img_end = {"jpg", "bmp", "png", "jpeg", "rgb", "tif", "tiff", "gif", "pdf"}
+        if os.path.isfile(img_file) and _check_image_file(img_file):
+            imgs_lists.append(img_file)
+        elif os.path.isdir(img_file):
+            for single_file in os.listdir(img_file):
+                file_path = os.path.join(img_file, single_file)
+                if os.path.isfile(file_path) and _check_image_file(file_path):
+                    imgs_lists.append(file_path)
+
+    if len(imgs_lists) == 0:
+        raise Exception("not found any img file in {}".format(img_file))
+    imgs_lists = sorted(imgs_lists)
+    return imgs_lists
+
+
+def binarize_img(img):
+    if len(img.shape) == 3 and img.shape[2] == 3:
+        gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)  # conversion to grayscale image
+        # use cv2 threshold binarization
+        _, gray = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
+        img = cv2.cvtColor(gray, cv2.COLOR_GRAY2BGR)
+    return img
+
+
+def alpha_to_color(img, alpha_color=(255, 255, 255)):
+    if len(img.shape) == 3 and img.shape[2] == 4:
+        B, G, R, A = cv2.split(img)
+        alpha = A / 255
+
+        R = (alpha_color[0] * (1 - alpha) + R * alpha).astype(np.uint8)
+        G = (alpha_color[1] * (1 - alpha) + G * alpha).astype(np.uint8)
+        B = (alpha_color[2] * (1 - alpha) + B * alpha).astype(np.uint8)
+
+        img = cv2.merge((B, G, R))
+    return img
+
+
+def check_and_read(img_path):
+    if os.path.basename(img_path)[-3:].lower() == "gif":
+        gif = cv2.VideoCapture(img_path)
+        ret, frame = gif.read()
+        if not ret:
+            logger = logging.getLogger("ppocr")
+            logger.info("Cannot read {}. This gif image maybe corrupted.")
+            return None, False
+        if len(frame.shape) == 2 or frame.shape[-1] == 1:
+            frame = cv2.cvtColor(frame, cv2.COLOR_GRAY2RGB)
+        imgvalue = frame[:, :, ::-1]
+        return imgvalue, True, False
+    elif os.path.basename(img_path)[-3:].lower() == "pdf":
+        from paddle.utils import try_import
+
+        fitz = try_import("fitz")
+        from PIL import Image
+
+        imgs = []
+        with fitz.open(img_path) as pdf:
+            for pg in range(0, pdf.page_count):
+                page = pdf[pg]
+                mat = fitz.Matrix(2, 2)
+                pm = page.get_pixmap(matrix=mat, alpha=False)
+
+                # if width or height > 2000 pixels, don't enlarge the image
+                if pm.width > 2000 or pm.height > 2000:
+                    pm = page.get_pixmap(matrix=fitz.Matrix(1, 1), alpha=False)
+
+                img = Image.frombytes("RGB", [pm.width, pm.height], pm.samples)
+                img = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
+                imgs.append(img)
+            return imgs, False, True
+    return None, False, False
+
+
+def load_vqa_bio_label_maps(label_map_path):
+    with open(label_map_path, "r", encoding="utf-8") as fin:
+        lines = fin.readlines()
+    old_lines = [line.strip() for line in lines]
+    lines = ["O"]
+    for line in old_lines:
+        # "O" has already been in lines
+        if line.upper() in ["OTHER", "OTHERS", "IGNORE"]:
+            continue
+        lines.append(line)
+    labels = ["O"]
+    for line in lines[1:]:
+        labels.append("B-" + line)
+        labels.append("I-" + line)
+    label2id_map = {label.upper(): idx for idx, label in enumerate(labels)}
+    id2label_map = {idx: label.upper() for idx, label in enumerate(labels)}
+    return label2id_map, id2label_map
+
+
+def set_seed(seed=1024):
+    random.seed(seed)
+    np.random.seed(seed)
+    paddle.seed(seed)
+
+
+def check_install(module_name, install_name):
+    spec = importlib.util.find_spec(module_name)
+    if spec is None:
+        print(f"Warning! The {module_name} module is NOT installed")
+        print(
+            f"Try install {module_name} module automatically. You can also try to install manually by pip install {install_name}."
+        )
+        python = sys.executable
+        try:
+            subprocess.check_call(
+                [python, "-m", "pip", "install", install_name],
+                stdout=subprocess.DEVNULL,
+            )
+            print(f"The {module_name} module is now installed")
+        except subprocess.CalledProcessError as exc:
+            raise Exception(f"Install {module_name} failed, please install manually")
+    else:
+        print(f"{module_name} has been installed.")
+
+
+class AverageMeter:
+    def __init__(self):
+        self.reset()
+
+    def reset(self):
+        """reset"""
+        self.val = 0
+        self.avg = 0
+        self.sum = 0
+        self.count = 0
+
+    def update(self, val, n=1):
+        """update"""
+        self.val = val
+        self.sum += val * n
+        self.count += n
+        self.avg = self.sum / self.count
diff --git a/ppocr/utils/visual.py b/ppocr/utils/visual.py
new file mode 100644
index 0000000..1a49c31
--- /dev/null
+++ b/ppocr/utils/visual.py
@@ -0,0 +1,142 @@
+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import cv2
+import os
+import numpy as np
+import PIL
+from PIL import Image, ImageDraw, ImageFont
+
+
+def draw_ser_results(
+    image, ocr_results, font_path="doc/fonts/simfang.ttf", font_size=14
+):
+    np.random.seed(2021)
+    color = (
+        np.random.permutation(range(255)),
+        np.random.permutation(range(255)),
+        np.random.permutation(range(255)),
+    )
+    color_map = {
+        idx: (color[0][idx], color[1][idx], color[2][idx]) for idx in range(1, 255)
+    }
+    if isinstance(image, np.ndarray):
+        image = Image.fromarray(image)
+    elif isinstance(image, str) and os.path.isfile(image):
+        image = Image.open(image).convert("RGB")
+    img_new = image.copy()
+    draw = ImageDraw.Draw(img_new)
+
+    font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
+    for ocr_info in ocr_results:
+        if ocr_info["pred_id"] not in color_map:
+            continue
+        color = color_map[ocr_info["pred_id"]]
+        text = "{}: {}".format(ocr_info["pred"], ocr_info["transcription"])
+
+        if "bbox" in ocr_info:
+            # draw with ocr engine
+            bbox = ocr_info["bbox"]
+        else:
+            # draw with ocr groundtruth
+            bbox = trans_poly_to_bbox(ocr_info["points"])
+        draw_box_txt(bbox, text, draw, font, font_size, color)
+
+    img_new = Image.blend(image, img_new, 0.7)
+    return np.array(img_new)
+
+
+def draw_box_txt(bbox, text, draw, font, font_size, color):
+    # draw ocr results outline
+    bbox = ((bbox[0], bbox[1]), (bbox[2], bbox[3]))
+    draw.rectangle(bbox, fill=color)
+
+    # draw ocr results
+    if int(PIL.__version__.split(".")[0]) < 10:
+        tw = font.getsize(text)[0]
+        th = font.getsize(text)[1]
+    else:
+        left, top, right, bottom = font.getbbox(text)
+        tw, th = right - left, bottom - top
+
+    start_y = max(0, bbox[0][1] - th)
+    draw.rectangle(
+        [(bbox[0][0] + 1, start_y), (bbox[0][0] + tw + 1, start_y + th)],
+        fill=(0, 0, 255),
+    )
+    draw.text((bbox[0][0] + 1, start_y), text, fill=(255, 255, 255), font=font)
+
+
+def trans_poly_to_bbox(poly):
+    x1 = np.min([p[0] for p in poly])
+    x2 = np.max([p[0] for p in poly])
+    y1 = np.min([p[1] for p in poly])
+    y2 = np.max([p[1] for p in poly])
+    return [x1, y1, x2, y2]
+
+
+def draw_re_results(image, result, font_path="doc/fonts/simfang.ttf", font_size=18):
+    np.random.seed(0)
+    if isinstance(image, np.ndarray):
+        image = Image.fromarray(image)
+    elif isinstance(image, str) and os.path.isfile(image):
+        image = Image.open(image).convert("RGB")
+    img_new = image.copy()
+    draw = ImageDraw.Draw(img_new)
+
+    font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
+    color_head = (0, 0, 255)
+    color_tail = (255, 0, 0)
+    color_line = (0, 255, 0)
+
+    for ocr_info_head, ocr_info_tail in result:
+        draw_box_txt(
+            ocr_info_head["bbox"],
+            ocr_info_head["transcription"],
+            draw,
+            font,
+            font_size,
+            color_head,
+        )
+        draw_box_txt(
+            ocr_info_tail["bbox"],
+            ocr_info_tail["transcription"],
+            draw,
+            font,
+            font_size,
+            color_tail,
+        )
+
+        center_head = (
+            (ocr_info_head["bbox"][0] + ocr_info_head["bbox"][2]) // 2,
+            (ocr_info_head["bbox"][1] + ocr_info_head["bbox"][3]) // 2,
+        )
+        center_tail = (
+            (ocr_info_tail["bbox"][0] + ocr_info_tail["bbox"][2]) // 2,
+            (ocr_info_tail["bbox"][1] + ocr_info_tail["bbox"][3]) // 2,
+        )
+
+        draw.line([center_head, center_tail], fill=color_line, width=5)
+
+    img_new = Image.blend(image, img_new, 0.5)
+    return np.array(img_new)
+
+
+def draw_rectangle(img_path, boxes):
+    boxes = np.array(boxes)
+    img = cv2.imread(img_path)
+    img_show = img.copy()
+    for box in boxes.astype(int):
+        x1, y1, x2, y2 = box
+        cv2.rectangle(img_show, (x1, y1), (x2, y2), (255, 0, 0), 2)
+    return img_show
diff --git a/ppstructure/README.md b/ppstructure/README.md
new file mode 100644
index 0000000..7f169e4
--- /dev/null
+++ b/ppstructure/README.md
@@ -0,0 +1,16 @@
+
+## PP-Structure Upgrade Considerations
+
+Since its launch in 2021, the PP-Structure series models have undergone three generations of updates and iterations. Currently, the latest version is the third generation (PP-StructureV3). All the code and documentation on this page are based on the second generation (PP-StructureV2). **This directory will be removed at an appropriate time in the future, and maintenance for the second generation will be discontinued.**
+
+The third generation, PP-StructureV3, offers more powerful document analysis capabilities, and we strongly recommend users to use this version via the integrated wheel package. For a detailed introduction to PP-StructureV3, please refer to the [PP-StructureV3 Introduction Document](https://www.paddleocr.ai/latest/en/version3.x/algorithm/PP-StructureV3/PP-StructureV3.html); for specific usage instructions, please see the [PP-StructureV3 Usage Guide](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PP-StructureV3.html).
+
+In addition, we will discontinue maintenance of the KIE capability provided in the second generation. We recommend that you use the PP-ChatOCRv4 model in PaddleOCR 3.x, which offers more accurate key information extraction. For a detailed introduction to PP-ChatOCRv4, please refer to the [PP-ChatOCRv4 Introduction Document](https://www.paddleocr.ai/latest/en/version3.x/algorithm/PP-ChatOCRv4/PP-ChatOCRv4.html); for specific usage instructions, please see the [PP-ChatOCRv4 Usage Document](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PP-ChatOCRv4.html).
+
+## PP-Structure 升级事项
+
+PP-Structure 系列模型自 2021 年推出以来，已经历三代版本的更新迭代。目前，最新版本为第三代（PP-StructureV3）。本页面的所有代码与文档均基于第二代（PP-StructureV2），**后续将在适当时机删除本目录，并停止对第二代的维护。**
+
+第三代 PP-StructureV3 提供了更为强大的文档解析能力，我们更推荐用户通过集成 wheel 包来使用该版本。关于 PP-StructureV3 的详细介绍，请参考[PP-StructureV3 介绍文档](https://www.paddleocr.ai/latest/version3.x/algorithm/PP-StructureV3/PP-StructureV3.html)；具体的使用方法可参见[PP-StructureV3 使用文档](https://www.paddleocr.ai/latest/version3.x/pipeline_usage/PP-StructureV3.html)。
+
+另外，在第二代提供的 KIE 能力，我们也将停止维护，推荐大家使用 PaddleOCR 3.x 版本的 PP-ChatOCRv4 模型方案，关键信息抽取更准确。关于 PP-ChatOCRv4 的详细介绍，请参考[PP-ChatOCRv4 介绍文档](https://www.paddleocr.ai/latest/version3.x/algorithm/PP-ChatOCRv4/PP-ChatOCRv4.html)；具体的使用方法可参见[PP-ChatOCRv4 使用文档](https://www.paddleocr.ai/latest/version3.x/pipeline_usage/PP-ChatOCRv4.html)。
diff --git a/ppstructure/__init__.py b/ppstructure/__init__.py
new file mode 100644
index 0000000..1d11e26
--- /dev/null
+++ b/ppstructure/__init__.py
@@ -0,0 +1,13 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/ppstructure/kie/README.md b/ppstructure/kie/README.md
new file mode 100644
index 0000000..685ca9f
--- /dev/null
+++ b/ppstructure/kie/README.md
@@ -0,0 +1,289 @@
+English | [简体中文](README_ch.md)
+
+# Key Information Extraction (KIE)
+
+- [1. Introduction](#1-introduction)
+- [2. Performance](#2-performance)
+- [3. Visualization](#3-visualization)
+  - [3.1 SER](#31-ser)
+  - [3.2 RE](#32-re)
+- [4. Usage](#4-usage)
+  - [4.1 Prepare for the environment](#41-prepare-for-the-environment)
+  - [4.2 Quick start](#42-quick-start)
+  - [4.3 More](#43-more)
+- [5. Reference](#5-reference)
+- [6. License](#6-license)
+
+
+## 1. Introduction
+
+Key information extraction (KIE) refers to extracting key information from text or images. As downstream task of OCR, the key information extraction task of document image has many practical application scenarios, such as form recognition, ticket information extraction, ID card information extraction, etc.
+
+PP-Structure conducts research based on the LayoutXLM multi-modal, and proposes the VI-LayoutXLM, which gets rid of visual features when finetuning the downstream tasks. An textline sorting method is also utilized to fit in reading order. What's more, UDML knowledge distillation is used for higher accuracy. Finally, the accuracy and inference speed of VI-LayoutXLM surpass those of LayoutXLM.
+
+The main features of the key information extraction module in PP-Structure are as follows.
+
+
+- Integrate multi-modal methods such as [LayoutXLM](https://arxiv.org/pdf/2104.08836.pdf), VI-LayoutXLM, and PP-OCR inference engine.
+- Supports Semantic Entity Recognition (SER) and Relation Extraction (RE) tasks based on multimodal methods. Based on the SER task, the text recognition and classification in the image can be completed; based on the RE task, the relationship extraction of the text content in the image can be completed, such as judging the problem pair (pair).
+- Supports custom training for SER tasks and RE tasks.
+- Supports end-to-end system prediction and evaluation of OCR+SER.
+- Supports end-to-end system prediction of OCR+SER+RE.
+- Support SER model export and inference using PaddleInference.
+
+
+## 2. Performance
+
+We evaluate the methods on the Chinese dataset of [XFUND](https://github.com/doc-analysis/XFUND), and the performance is as follows
+
+|Model | Backbone | Task | Config file | Hmean | Inference time (ms) | Download link|
+| --- | --- |  --- | --- | --- | --- | --- |
+|VI-LayoutXLM| VI-LayoutXLM-base | SER | [ser_vi_layoutxlm_xfund_zh_udml.yml](../../configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh_udml.yml)|**93.19%**| 15.49|[trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar)|
+|LayoutXLM| LayoutXLM-base | SER | [ser_layoutxlm_xfund_zh.yml](../../configs/kie/layoutlm_series/ser_layoutxlm_xfund_zh.yml)|90.38%| 19.49 | [trained model](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar)|
+|VI-LayoutXLM| VI-LayoutXLM-base | RE | [re_vi_layoutxlm_xfund_zh_udml.yml](../../configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh_udml.yml)|**83.92%**| 15.49|[trained model](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar)|
+|LayoutXLM| LayoutXLM-base | RE | [re_layoutxlm_xfund_zh.yml](../../configs/kie/layoutlm_series/re_layoutxlm_xfund_zh.yml)|74.83%| 19.49|[trained model](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar)|
+
+
+* Note：Inference environment：V100 GPU + cuda10.2 + cudnn8.1.1 + TensorRT 7.2.3.4，tested using fp16.
+
+For more KIE models in PaddleOCR, please refer to [KIE model zoo](../../doc/doc_en/algorithm_overview_en.md).
+
+
+## 3. Visualization
+
+There are two main solutions to the key information extraction task based on VI-LayoutXLM series model.
+
+(1) Text detection + text recognition + semantic entity recognition (SER)
+
+(2) Text detection + text recognition + semantic entity recognition (SER) + relationship extraction (RE)
+
+
+The following images are demo results of the SER and RE models. For more detailed introduction to the above solutions, please refer to [KIE Guide](./how_to_do_kie.md).
+
+### 3.1 SER
+
+Demo results for SER task are as follows.
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185539141-68e71c75-5cf7-4529-b2ca-219d29fa5f68.jpg" width="600">
+</div>
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808.jpg" width="600">
+</div>
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a.png" width="600">
+</div>
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185539735-37b5c2ef-629d-43fe-9abb-44bb717ef7ee.jpg" width="600">
+</div>
+
+
+
+**Note:** test pictures are from [xfund dataset](https://github.com/doc-analysis/XFUND), [invoice dataset](https://aistudio.baidu.com/aistudio/datasetdetail/165561) and a composite ID card dataset.
+
+
+Boxes of different colors in the image represent different categories.
+
+The invoice and application form images have three categories: `request`, `answer` and `header`. The `question` and `answer` can be used to extract the relationship.
+
+For the ID card image, the model can directly identify the key information such as `name`, `gender`, `nationality`, so that the subsequent relationship extraction process is not required, and the key information extraction task can be completed using only one model.
+
+### 3.2 RE
+
+Demo results for RE task are as follows.
+
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb.jpg" width="600">
+</div>
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f.jpg" width="600">
+</div>
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185540291-f64e5daf-6d42-4e7c-bbbb-471e3fac4fcc.png" width="600">
+</div>
+
+Red boxes are questions, blue boxes are answers. The green lines means the two connected objects are a pair.
+
+
+## 4. Usage
+
+### 4.1 Prepare for the environment
+
+
+Use the following command to install KIE dependencies.
+
+
+```bash
+git clone https://github.com/PaddlePaddle/PaddleOCR.git
+cd PaddleOCR
+pip install -r requirements.txt
+pip install -r ppstructure/kie/requirements.txt
+# 安装PaddleOCR引擎用于预测
+pip install "paddleocr<3.0"
+```
+
+NOTE: For KIE tasks, it is necessary to downgrade the Paddle framework version (Paddle<2.6) and the PaddleNLP version (PaddleNLP<2.6).
+
+The visualized results of SER are saved in the `./output` folder by default. Examples of results are as follows.
+
+
+<div align="center">
+    <img src="../../ppstructure/docs/kie/result_ser/zh_val_42_ser.jpg" width="800">
+</div>
+
+
+### 4.2 Quick start
+
+Here we use XFUND dataset to quickly experience the SER model and RE model.
+
+
+#### 4.2.1 Prepare for the dataset
+
+```bash
+mkdir train_data
+cd train_data
+# download and uncompress the dataset
+wget https://paddleocr.bj.bcebos.com/ppstructure/dataset/XFUND.tar && tar -xf XFUND.tar
+cd ..
+```
+
+#### 4.2.2 Predict images using the trained model
+
+Use the following command to download the models.
+
+```bash
+mkdir pretrained_model
+cd pretrained_model
+# download and uncompress the SER trained model
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar && tar -xf ser_vi_layoutxlm_xfund_pretrained.tar
+
+# download and uncompress the RE trained model
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar && tar -xf re_vi_layoutxlm_xfund_pretrained.tar
+```
+
+
+If you want to use OCR engine to obtain end-to-end prediction results, you can use the following command to predict.
+
+```bash
+# just predict using SER trained model
+python3 tools/infer_kie_token_ser.py \
+  -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrained_model/ser_vi_layoutxlm_xfund_pretrained/best_accuracy \
+  Global.infer_img=./ppstructure/docs/kie/input/zh_val_42.jpg
+
+# predict using SER and RE trained model at the same time
+python3 ./tools/infer_kie_token_ser_re.py \
+  -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrained_model/re_vi_layoutxlm_xfund_pretrained/best_accuracy \
+  Global.infer_img=./train_data/XFUND/zh_val/image/zh_val_42.jpg \
+  -c_ser configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o_ser Architecture.Backbone.checkpoints=./pretrained_model/ser_vi_layoutxlm_xfund_pretrained/best_accuracy
+```
+
+The visual result images and the predicted text file will be saved in the `Global.save_res_path` directory.
+
+If you want to use a custom ocr model, you can set it through the following fields
+- `Global.kie_det_model_dir`: the detection inference model path
+- `Global.kie_rec_model_dir`: the recognition inference model path
+
+
+If you want to load the text detection and recognition results collected before, you can use the following command to predict.
+
+```bash
+# just predict using SER trained model
+python3 tools/infer_kie_token_ser.py \
+  -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrained_model/ser_vi_layoutxlm_xfund_pretrained/best_accuracy \
+  Global.infer_img=./train_data/XFUND/zh_val/val.json \
+  Global.infer_mode=False
+
+# predict using SER and RE trained model at the same time
+python3 ./tools/infer_kie_token_ser_re.py \
+  -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrained_model/re_vi_layoutxlm_xfund_pretrained/best_accuracy \
+  Global.infer_img=./train_data/XFUND/zh_val/val.json \
+  Global.infer_mode=False \
+  -c_ser configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o_ser Architecture.Backbone.checkpoints=./pretrained_model/ser_vi_layoutxlm_xfund_pretrained/best_accuracy
+```
+
+#### 4.2.3 Inference using PaddleInference
+
+Firstly, download the inference SER inference model.
+
+```bash
+mkdir inference
+cd inference
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_infer.tar && tar -xf ser_vi_layoutxlm_xfund_infer.tar
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_infer.tar && tar -xf re_vi_layoutxlm_xfund_infer.tar
+cd ..
+```
+
+- SER
+
+Use the following command for inference.
+
+
+```bash
+cd ppstructure
+python3 kie/predict_kie_token_ser.py \
+  --kie_algorithm=LayoutXLM \
+  --ser_model_dir=../inference/ser_vi_layoutxlm_xfund_infer \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx"
+```
+
+The visual results and text file will be saved in directory `output`.
+
+- RE
+
+Use the following command for inference.
+
+
+```bash
+cd ppstructure
+python3 kie/predict_kie_token_ser_re.py \
+  --kie_algorithm=LayoutXLM \
+  --re_model_dir=../inference/re_vi_layoutxlm_xfund_infer \
+  --ser_model_dir=../inference/ser_vi_layoutxlm_xfund_infer \
+  --use_visual_backbone=False \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx"
+```
+
+The visual results and text file will be saved in directory `output`.
+
+If you want to use a custom ocr model, you can set it through the following fields
+- `--det_model_dir`: the detection inference model path
+- `--rec_model_dir`: the recognition inference model path
+
+### 4.3 More
+
+For training, evaluation and inference tutorial for KIE models, please refer to [KIE doc](../../doc/doc_en/kie_en.md).
+
+For training, evaluation and inference tutorial for text detection models, please refer to [text detection doc](../../doc/doc_en/detection_en.md).
+
+For training, evaluation and inference tutorial for text recognition models, please refer to [text recognition doc](../../doc/doc_en/recognition_en.md).
+
+To complete the key information extraction task in your own scenario from data preparation to model selection, please refer to: [Guide to End-to-end KIE](./how_to_do_kie_en.md)。
+
+
+## 5. Reference
+
+- LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding, https://arxiv.org/pdf/2104.08836.pdf
+- microsoft/unilm/layoutxlm, https://github.com/microsoft/unilm/tree/master/layoutxlm
+- XFUND dataset, https://github.com/doc-analysis/XFUND
+
+## 6. License
+
+The content of this project itself is licensed under the [Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)](https://creativecommons.org/licenses/by-nc-sa/4.0/)
diff --git a/ppstructure/kie/README_ch.md b/ppstructure/kie/README_ch.md
new file mode 100644
index 0000000..e4b1a62
--- /dev/null
+++ b/ppstructure/kie/README_ch.md
@@ -0,0 +1,271 @@
+[English](README.md) | 简体中文
+
+# 关键信息抽取
+
+- [1. 简介](#1-简介)
+- [2. 精度与性能](#2-精度与性能)
+- [3. 效果演示](#3-效果演示)
+  - [3.1 SER](#31-ser)
+  - [3.2 RE](#32-re)
+- [4. 使用](#4-使用)
+  - [4.1 准备环境](#41-准备环境)
+  - [4.2 快速开始](#42-快速开始)
+  - [4.3 更多](#43-更多)
+- [5. 参考链接](#5-参考链接)
+- [6. License](#6-License)
+
+
+## 1. 简介
+
+关键信息抽取 (Key Information Extraction, KIE)指的是是从文本或者图像中，抽取出关键的信息。针对文档图像的关键信息抽取任务作为OCR的下游任务，存在非常多的实际应用场景，如表单识别、车票信息抽取、身份证信息抽取等。
+
+PP-Structure 基于 LayoutXLM 文档多模态系列方法进行研究与优化，设计了视觉特征无关的多模态模型结构VI-LayoutXLM，同时引入符合阅读顺序的文本行排序方法以及UDML联合互学习蒸馏方法，最终在精度与速度均超越LayoutXLM。
+
+PP-Structure中关键信息抽取模块的主要特性如下：
+
+- 集成[LayoutXLM](https://arxiv.org/pdf/2104.08836.pdf)、VI-LayoutXLM等多模态模型以及PP-OCR预测引擎。
+- 支持基于多模态方法的语义实体识别 (Semantic Entity Recognition, SER) 以及关系抽取 (Relation Extraction, RE) 任务。基于 SER 任务，可以完成对图像中的文本识别与分类；基于 RE 任务，可以完成对图象中的文本内容的关系提取，如判断问题对(pair)。
+- 支持SER任务和RE任务的自定义训练。
+- 支持OCR+SER的端到端系统预测与评估。
+- 支持OCR+SER+RE的端到端系统预测。
+- 支持SER模型的动转静导出与基于PaddleInfernece的模型推理。
+
+
+## 2. 精度与性能
+
+
+我们在 [XFUND](https://github.com/doc-analysis/XFUND) 的中文数据集上对算法进行了评估，SER与RE上的任务性能如下
+
+|模型|骨干网络|任务|配置文件|hmean|预测耗时(ms)|下载链接|
+| --- | --- |  --- | --- | --- | --- | --- |
+|VI-LayoutXLM| VI-LayoutXLM-base | SER | [ser_vi_layoutxlm_xfund_zh_udml.yml](../../configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh_udml.yml)|**93.19%**| 15.49|[训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar)|
+|LayoutXLM| LayoutXLM-base | SER | [ser_layoutxlm_xfund_zh.yml](../../configs/kie/layoutlm_series/ser_layoutxlm_xfund_zh.yml)|90.38%| 19.49 | [训练模型](https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar)|
+|VI-LayoutXLM| VI-LayoutXLM-base | RE | [re_vi_layoutxlm_xfund_zh_udml.yml](../../configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh_udml.yml)|**83.92%**| 15.49|[训练模型](https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar)|
+|LayoutXLM| LayoutXLM-base | RE | [re_layoutxlm_xfund_zh.yml](../../configs/kie/layoutlm_series/re_layoutxlm_xfund_zh.yml)|74.83%| 19.49|[训练模型](https://paddleocr.bj.bcebos.com/pplayout/re_LayoutXLM_xfun_zh.tar)|
+
+
+* 注：预测耗时测试条件：V100 GPU + cuda10.2 + cudnn8.1.1 + TensorRT 7.2.3.4，使用FP16进行测试。
+
+更多关于PaddleOCR中关键信息抽取模型的介绍，请参考[关键信息抽取模型库](../../doc/doc_ch/algorithm_overview.md)。
+
+
+## 3. 效果演示
+
+基于多模态模型的关键信息抽取任务有2种主要的解决方案。
+
+（1）文本检测 + 文本识别 + 语义实体识别(SER)
+（2）文本检测 + 文本识别 + 语义实体识别(SER) + 关系抽取(RE)
+
+下面给出SER与RE任务的示例效果，关于上述解决方案的详细介绍，请参考[关键信息抽取全流程指南](./how_to_do_kie.md)。
+
+### 3.1 SER
+
+对于SER任务，效果如下所示。
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185539141-68e71c75-5cf7-4529-b2ca-219d29fa5f68.jpg" width="600">
+</div>
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185310636-6ce02f7c-790d-479f-b163-ea97a5a04808.jpg" width="600">
+</div>
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185539517-ccf2372a-f026-4a7c-ad28-c741c770f60a.png" width="600">
+</div>
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185539735-37b5c2ef-629d-43fe-9abb-44bb717ef7ee.jpg" width="600">
+</div>
+
+**注意：** 测试图片来源于[XFUND数据集](https://github.com/doc-analysis/XFUND)、[发票数据集](https://aistudio.baidu.com/aistudio/datasetdetail/165561)以及合成的身份证数据集。
+
+
+图中不同颜色的框表示不同的类别。
+
+图中的发票以及申请表图像，有`QUESTION`, `ANSWER`, `HEADER` 3种类别，识别的`QUESTION`, `ANSWER`可以用于后续的问题与答案的关系抽取。
+
+图中的身份证图像，则直接识别出其中的`姓名`、`性别`、`民族`等关键信息，这样就无需后续的关系抽取过程，一个模型即可完成关键信息抽取。
+
+
+### 3.2 RE
+
+对于RE任务，效果如下所示。
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185393805-c67ff571-cf7e-4217-a4b0-8b396c4f22bb.jpg" width="600">
+</div>
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185540080-0431e006-9235-4b6d-b63d-0b3c6e1de48f.jpg" width="600">
+</div>
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185540291-f64e5daf-6d42-4e7c-bbbb-471e3fac4fcc.png" width="600">
+</div>
+
+
+红色框是问题，蓝色框是答案。绿色线条表示连接的两端为一个key-value的pair。
+
+## 4. 使用
+
+### 4.1 准备环境
+
+使用下面的命令安装运行SER与RE关键信息抽取的依赖。
+
+```bash
+git clone https://github.com/PaddlePaddle/PaddleOCR.git
+cd PaddleOCR
+pip install -r requirements.txt
+pip install -r ppstructure/kie/requirements.txt
+# 安装PaddleOCR引擎用于预测
+pip install "paddleocr<3.0"
+```
+
+NOTE: 对于KIE任务需要降低Paddle框架版本（Paddle<2.6），和PaddleNLP版本（PaddleNLP<2.6）。
+
+### 4.2 快速开始
+
+下面XFUND数据集，快速体验SER模型与RE模型。
+
+#### 4.2.1 准备数据
+
+```bash
+mkdir train_data
+cd train_data
+# 下载与解压数据
+wget https://paddleocr.bj.bcebos.com/ppstructure/dataset/XFUND.tar && tar -xf XFUND.tar
+cd ..
+```
+
+#### 4.2.2 基于动态图的预测
+
+首先下载模型。
+
+```bash
+mkdir pretrained_model
+cd pretrained_model
+# 下载并解压SER预训练模型
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar && tar -xf ser_vi_layoutxlm_xfund_pretrained.tar
+
+# 下载并解压RE预训练模型
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_pretrained.tar && tar -xf re_vi_layoutxlm_xfund_pretrained.tar
+```
+
+如果希望使用OCR引擎，获取端到端的预测结果，可以使用下面的命令进行预测。
+
+```bash
+# 仅预测SER模型
+python3 tools/infer_kie_token_ser.py \
+  -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrained_model/ser_vi_layoutxlm_xfund_pretrained/best_accuracy \
+  Global.infer_img=./ppstructure/docs/kie/input/zh_val_42.jpg
+
+# SER + RE模型串联
+python3 ./tools/infer_kie_token_ser_re.py \
+  -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrained_model/re_vi_layoutxlm_xfund_pretrained/best_accuracy \
+  Global.infer_img=./train_data/XFUND/zh_val/image/zh_val_42.jpg \
+  -c_ser configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o_ser Architecture.Backbone.checkpoints=./pretrained_model/ser_vi_layoutxlm_xfund_pretrained/best_accuracy
+```
+
+`Global.save_res_path`目录中会保存可视化的结果图像以及预测的文本文件。
+
+如果想使用自定义OCR模型，可通过如下字段进行设置
+- `Global.kie_det_model_dir`: 设置检测inference模型地址
+- `Global.kie_rec_model_dir`: 设置识别inference模型地址
+
+
+如果希望加载标注好的文本检测与识别结果，仅预测可以使用下面的命令进行预测。
+
+```bash
+# 仅预测SER模型
+python3 tools/infer_kie_token_ser.py \
+  -c configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrained_model/ser_vi_layoutxlm_xfund_pretrained/best_accuracy \
+  Global.infer_img=./train_data/XFUND/zh_val/val.json \
+  Global.infer_mode=False
+
+# SER + RE模型串联
+python3 ./tools/infer_kie_token_ser_re.py \
+  -c configs/kie/vi_layoutxlm/re_vi_layoutxlm_xfund_zh.yml \
+  -o Architecture.Backbone.checkpoints=./pretrained_model/re_vi_layoutxlm_xfund_pretrained/best_accuracy \
+  Global.infer_img=./train_data/XFUND/zh_val/val.json \
+  Global.infer_mode=False \
+  -c_ser configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml \
+  -o_ser Architecture.Backbone.checkpoints=./pretrained_model/ser_vi_layoutxlm_xfund_pretrained/best_accuracy
+```
+
+#### 4.2.3 基于PaddleInference的预测
+
+首先下载SER和RE的推理模型。
+
+```bash
+mkdir inference
+cd inference
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_infer.tar && tar -xf ser_vi_layoutxlm_xfund_infer.tar
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/re_vi_layoutxlm_xfund_infer.tar && tar -xf re_vi_layoutxlm_xfund_infer.tar
+cd ..
+```
+
+- SER
+
+执行下面的命令进行预测。
+
+```bash
+cd ppstructure
+python3 kie/predict_kie_token_ser.py \
+  --kie_algorithm=LayoutXLM \
+  --ser_model_dir=../inference/ser_vi_layoutxlm_xfund_infer \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx"
+```
+
+可视化结果保存在`output`目录下。
+
+- RE
+
+执行下面的命令进行预测。
+
+```bash
+cd ppstructure
+python3 kie/predict_kie_token_ser_re.py \
+  --kie_algorithm=LayoutXLM \
+  --re_model_dir=../inference/re_vi_layoutxlm_xfund_infer \
+  --ser_model_dir=../inference/ser_vi_layoutxlm_xfund_infer \
+  --use_visual_backbone=False \
+  --image_dir=./docs/kie/input/zh_val_42.jpg \
+  --ser_dict_path=../train_data/XFUND/class_list_xfun.txt \
+  --vis_font_path=../doc/fonts/simfang.ttf \
+  --ocr_order_method="tb-yx"
+```
+
+可视化结果保存在`output`目录下。
+
+如果想使用自定义OCR模型，可通过如下字段进行设置
+- `--det_model_dir`: 设置检测inference模型地址
+- `--rec_model_dir`: 设置识别inference模型地址
+
+### 4.3 更多
+
+关于KIE模型的训练评估与推理，请参考：[关键信息抽取教程](../../doc/doc_ch/kie.md)。
+
+关于文本检测模型的训练评估与推理，请参考：[文本检测教程](../../doc/doc_ch/detection.md)。
+
+关于文本识别模型的训练评估与推理，请参考：[文本识别教程](../../doc/doc_ch/recognition.md)。
+
+关于怎样在自己的场景中完成关键信息抽取任务，请参考：[关键信息抽取全流程指南](./how_to_do_kie.md)。
+
+
+## 5. 参考链接
+
+- LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding, https://arxiv.org/pdf/2104.08836.pdf
+- microsoft/unilm/layoutxlm, https://github.com/microsoft/unilm/tree/master/layoutxlm
+- XFUND dataset, https://github.com/doc-analysis/XFUND
+
+## 6. License
+
+The content of this project itself is licensed under the [Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)](https://creativecommons.org/licenses/by-nc-sa/4.0/)
diff --git a/ppstructure/kie/how_to_do_kie.md b/ppstructure/kie/how_to_do_kie.md
new file mode 100644
index 0000000..b2679fb
--- /dev/null
+++ b/ppstructure/kie/how_to_do_kie.md
@@ -0,0 +1,168 @@
+
+# 怎样完成基于图像数据的信息抽取任务
+
+- [1. 简介](#1-简介)
+  - [1.1 背景](#11-背景)
+  - [1.2 主流方法](#12-主流方法)
+- [2. 关键信息抽取任务流程](#2-关键信息抽取任务流程)
+  - [2.1 训练OCR模型](#21-训练OCR模型)
+  - [2.2 训练KIE模型](#22-训练KIE模型)
+- [3. 参考文献](#3-参考文献)
+
+
+## 1. 简介
+
+### 1.1 背景
+
+关键信息抽取 (Key Information Extraction, KIE)指的是是从文本或者图像中，抽取出关键的信息。针对文档图像的关键信息抽取任务作为OCR的下游任务，存在非常多的实际应用场景，如表单识别、车票信息抽取、身份证信息抽取等。然而，使用人力从这些文档图像中提取或者收集关键信息耗时费力，怎样自动化融合图像中的视觉、布局、文字等特征并完成关键信息抽取是一个价值与挑战并存的问题。
+
+对于特定场景的文档图像，其中的关键信息位置、版式等较为固定，因此在研究早期有很多基于模板匹配的方法进行关键信息的抽取，考虑到其流程较为简单，该方法仍然被广泛应用在目前的很多场景中。但是这种基于模板匹配的方法在应用到不同的场景中时，需要耗费大量精力去调整与适配模板，迁移成本较高。
+
+文档图像中的KIE一般包含2个子任务，示意图如下图所示。
+
+* （1）SER: 语义实体识别 (Semantic Entity Recognition)，对每一个检测到的文本进行分类，如将其分为姓名，身份证。如下图中的黑色框和红色框。
+* （2）RE: 关系抽取 (Relation Extraction)，对每一个检测到的文本进行分类，如将其分为问题 (key) 和答案 (value) 。然后对每一个问题找到对应的答案，相当于完成key-value的匹配过程。如下图中的红色框和黑色框分别代表问题和答案，黄色线代表问题和答案之间的对应关系。
+
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/184588654-d87f54f3-13ab-42c4-afc0-da79bead3f14.png" width="800">
+</div>
+
+
+### 1.2 基于深度学习的主流方法
+
+一般的KIE方法基于命名实体识别(Named Entity Recognition,NER)来展开研究，但是此类方法仅使用了文本信息而忽略了位置与视觉特征信息，因此精度受限。近几年大多学者开始融合多个模态的输入信息，进行特征融合，并对多模态信息进行处理，从而提升KIE的精度。主要方法有以下几种
+
+* （1）基于Grid的方法：此类方法主要关注图像层面多模态信息的融合，文本大多大多为字符粒度，对文本与结构结构信息的嵌入方式较为简单，如Chargrid[1]等算法。
+* （2）基于Token的方法：此类方法参考NLP中的BERT等方法，将位置、视觉等特征信息共同编码到多模态模型中，并且在大规模数据集上进行预训练，从而在下游任务中，仅需要少量的标注数据便可以获得很好的效果。如LayoutLM[2], LayoutLMv2[3], LayoutXLM[4], StrucText[5]等算法。
+* （3）基于GCN的方法：此类方法尝试学习图像、文字之间的结构信息，从而可以解决开集信息抽取的问题（训练集中没有见过的模板），如GCN[6]、SDMGR[7]等算法。
+* （4）基于End-to-end的方法：此类方法将现有的OCR文字识别以及KIE信息抽取2个任务放在一个统一的网络中进行共同学习，并在学习过程中相互加强。如Trie[8]等算法。
+
+更多关于该系列算法的详细介绍，请参考“动手学OCR·十讲”课程的课节六部分：[文档分析理论与实践](https://aistudio.baidu.com/aistudio/education/group/info/25207)。
+
+## 2. 关键信息抽取任务流程
+
+PaddleOCR中实现了LayoutXLM等算法（基于Token），同时，在PP-StructureV2中，对LayoutXLM多模态预训练模型的网络结构进行简化，去除了其中的Visual backbone部分，设计了视觉无关的VI-LayoutXLM模型，同时引入符合人类阅读顺序的排序逻辑以及UDML知识蒸馏策略，最终同时提升了关键信息抽取模型的精度与推理速度。
+
+下面介绍怎样基于PaddleOCR完成关键信息抽取任务。
+
+在非End-to-end的KIE方法中，完成关键信息抽取，至少需要**2个步骤**：首先使用OCR模型，完成文字位置与内容的提取，然后使用KIE模型，根据图像、文字位置以及文字内容，提取出其中的关键信息。
+
+### 2.1 训练OCR模型
+
+#### 2.1.1 文本检测
+
+**（1）数据**
+
+PaddleOCR中提供的模型大多数为通用模型，在进行文本检测的过程中，相邻文本行的检测一般是根据位置的远近进行区分，如上图，使用PP-OCRv3通用中英文检测模型进行文本检测时，容易将”民族“与“汉”这2个代表不同的字段检测到一起，从而增加后续KIE任务的难度。因此建议在做KIE任务的过程中，首先训练一个针对该文档数据集的检测模型。
+
+在数据标注时，关键信息的标注需要隔开，比上图中的 “民族汉” 3个字相隔较近，此时需要将”民族“与”汉“标注为2个文本检测框，否则会增加后续KIE任务的难度。
+
+对于下游任务，一般来说，`200~300`张的文本训练数据即可保证基本的训练效果，如果没有太多的先验知识，可以先标注 **`200~300`** 张图片，进行后续文本检测模型的训练。
+
+
+**（2）模型**
+
+在模型选择方面，推荐使用PP-OCRv3_det，关于更多关于检测模型的训练方法介绍，请参考：[OCR文本检测模型训练教程](../../doc/doc_ch/detection.md)与[PP-OCRv3 文本检测模型训练教程](../../doc/doc_ch/PPOCRv3_det_train.md)。
+
+#### 2.1.2 文本识别
+
+相对自然场景，文档图像中的文本内容识别难度一般相对较低（背景相对不太复杂），因此**优先建议**尝试PaddleOCR中提供的PP-OCRv3通用文本识别模型([PP-OCRv3模型库链接](../../doc/doc_ch/models_list.md))。
+
+**（1）数据**
+
+然而，在部分文档场景中也会存在一些挑战，如身份证场景中存在着罕见字，在发票等场景中的字体比较特殊，这些问题都会增加文本识别的难度，此时如果希望保证或者进一步提升模型的精度，建议基于特定文档场景的文本识别数据集，加载PP-OCRv3模型进行微调。
+
+在模型微调的过程中，建议准备至少`5000`张垂类场景的文本识别图像，可以保证基本的模型微调效果。如果希望提升模型的精度与泛化能力，可以合成更多与该场景类似的文本识别数据，从公开数据集中收集通用真实文本识别数据，一并添加到该场景的文本识别训练任务过程中。在训练过程中，建议每个epoch的真实垂类数据、合成数据、通用数据比例在`1:1:1`左右，这可以通过设置不同数据源的采样比例进行控制。如有3个训练文本文件，分别包含1W、2W、5W条数据，那么可以在配置文件中设置数据如下：
+
+```yml
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/train_list_1W.txt
+    - ./train_data/train_list_2W.txt
+    - ./train_data/train_list_5W.txt
+    ratio_list: [1.0, 0.5, 0.2]
+    ...
+```
+
+**（2）模型**
+
+在模型选择方面，推荐使用通用中英文文本识别模型PP-OCRv3_rec，关于更多关于文本识别模型的训练方法介绍，请参考：[OCR文本识别模型训练教程](../../doc/doc_ch/recognition.md)与[PP-OCRv3文本识别模型库与配置文件](../../doc/doc_ch/models_list.md)。
+
+### 2.2 训练KIE模型
+
+对于识别得到的文字进行关键信息抽取，有2种主要的方法。
+
+（1）直接使用SER，获取关键信息的类别：如身份证场景中，将“姓名“与”张三“分别标记为`name_key`与`name_value`。最终识别得到的类别为`name_value`对应的**文本字段**即为我们所需要的关键信息。
+
+（2）联合SER与RE进行使用：这种方法中，首先使用SER，获取图像文字内容中所有的key与value，然后使用RE方法，对所有的key与value进行配对，找到映射关系，从而完成关键信息的抽取。
+
+#### 2.2.1 SER
+
+以身份证场景为例， 关键信息一般包含`姓名`、`性别`、`民族`等，我们直接将对应的字段标注为特定的类别即可，如下图所示。
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/184526682-8b810397-5a93-4395-93da-37b8b8494c41.png" width="500">
+</div>
+
+**注意：**
+
+- 标注过程中，对于无关于KIE关键信息的文本内容，均需要将其标注为`other`类别，相当于背景信息。如在身份证场景中，如果我们不关注性别信息，那么可以将“性别”与“男”这2个字段的类别均标注为`other`。
+- 标注过程中，需要以**文本行**为单位进行标注，无需标注单个字符的位置信息。
+
+数据量方面，一般来说，对于比较固定的场景，**50张**左右的训练图片即可达到可以接受的效果，可以使用[PPOCRLabel](https://github.com/PFCCLab/PPOCRLabel/blob/main/README_ch.md)完成KIE的标注过程。
+
+模型方面，推荐使用PP-StructureV2中提出的VI-LayoutXLM模型，它基于LayoutXLM模型进行改进，去除其中的视觉特征提取模块，在精度基本无损的情况下，进一步提升了模型推理速度。更多教程请参考：[VI-LayoutXLM算法介绍](../../doc/doc_ch/algorithm_kie_vi_layoutxlm.md)与[KIE关键信息抽取使用教程](../../doc/doc_ch/kie.md)。
+
+
+#### 2.2.2 SER + RE
+
+该过程主要包含SER与RE 2个过程。SER阶段主要用于识别出文档图像中的所有key与value，RE阶段主要用于对所有的key与value进行匹配。
+
+以身份证场景为例， 关键信息一般包含`姓名`、`性别`、`民族`等关键信息，在SER阶段，我们需要识别所有的question (key) 与answer (value) 。标注如下所示。每个字段的类别信息（`label`字段）可以是question、answer或者other（与待抽取的关键信息无关的字段）
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/184526785-c3d2d310-cd57-4d31-b933-912716b29856.jpg" width="500">
+</div>
+
+
+在RE阶段，需要标注每个字段的的id与连接信息，如下图所示。
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/184528728-626f77eb-fd9f-4709-a7dc-5411cc417dab.jpg" width="500">
+</div>
+
+每个文本行字段中，需要添加`id`与`linking`字段信息，`id`记录该文本行的唯一标识，同一张图片中的不同文本内容不能重复，`linking`是一个列表，记录了不同文本之间的连接信息。如字段“出生”的id为0，字段“1996年1月11日”的id为1，那么它们均有[[0, 1]]的`linking`标注，表示该id=0与id=1的字段构成key-value的关系（姓名、性别等字段类似，此处不再一一赘述）。
+
+
+**注意：**
+
+- 标注过程中，如果value是多个字符，那么linking中可以新增一个key-value对，如`[[0, 1], [0, 2]]`
+
+
+数据量方面，一般来说，对于比较固定的场景，**50张**左右的训练图片即可达到可以接受的效果，可以使用PPOCRLabel完成KIE的标注过程。
+
+模型方面，推荐使用PP-StructureV2中提出的VI-LayoutXLM模型，它基于LayoutXLM模型进行改进，去除其中的视觉特征提取模块，在精度基本无损的情况下，进一步提升了模型推理速度。更多教程请参考：[VI-LayoutXLM算法介绍](../../doc/doc_ch/algorithm_kie_vi_layoutxlm.md)与[KIE关键信息抽取使用教程](../../doc/doc_ch/kie.md)。
+
+
+## 3. 参考文献
+
+
+[1] Katti A R, Reisswig C, Guder C, et al. Chargrid: Towards understanding 2d documents[J]. arXiv preprint arXiv:1809.08799, 2018.
+
+[2] Xu Y, Li M, Cui L, et al. Layoutlm: Pre-training of text and layout for document image understanding[C]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2020: 1192-1200.
+
+[3] Xu Y, Xu Y, Lv T, et al. LayoutLMv2: Multi-modal pre-training for visually-rich document understanding[J]. arXiv preprint arXiv:2012.14740, 2020.
+
+[4]: Xu Y, Lv T, Cui L, et al. Layoutxlm: Multimodal pre-training for multilingual visually-rich document understanding[J]. arXiv preprint arXiv:2104.08836, 2021.
+
+[5] Li Y, Qian Y, Yu Y, et al. StrucTexT: Structured Text Understanding with Multi-Modal Transformers[C]//Proceedings of the 29th ACM International Conference on Multimedia. 2021: 1912-1920.
+
+[6] Liu X, Gao F, Zhang Q, et al. Graph convolution for multimodal information extraction from visually rich documents[J]. arXiv preprint arXiv:1903.11279, 2019.
+
+[7] Sun H, Kuang Z, Yue X, et al. Spatial Dual-Modality Graph Reasoning for Key Information Extraction[J]. arXiv preprint arXiv:2103.14470, 2021.
+
+[8] Zhang P, Xu Y, Cheng Z, et al. Trie: End-to-end text reading and information extraction for document understanding[C]//Proceedings of the 28th ACM International Conference on Multimedia. 2020: 1413-1422.
diff --git a/ppstructure/kie/how_to_do_kie_en.md b/ppstructure/kie/how_to_do_kie_en.md
new file mode 100644
index 0000000..a86a9f8
--- /dev/null
+++ b/ppstructure/kie/how_to_do_kie_en.md
@@ -0,0 +1,179 @@
+
+# Key Information Extraction Pipeline
+
+- [1. Introduction](#1-Introduction)
+  - [1.1 Background](#11-Background)
+  - [1.2 Mainstream Deep-learning Solutions](#12-Mainstream-Deep-learning-Solutions)
+- [2. KIE Pipeline](#2-KIE-Pipeline)
+  - [2.1 Train OCR Models](#21-Train-OCR-Models)
+  - [2.2 Train KIE Models](#22-Train-KIE-Models)
+- [3. Reference](#3-Reference)
+
+
+## 1. Introduction
+
+### 1.1 Background
+
+Key information extraction (KIE) refers to extracting key information from text or images. As the downstream task of OCR, KIE of document image has many practical application scenarios, such as form recognition, ticket information extraction, ID card information extraction, etc. However, it is time-consuming and laborious to extract  key information from these document images by manpower. It's challengable but also valuable to combine multi-modal features (visual, layout, text, etc) together and complete KIE tasks.
+
+For the document images in a specific scene, the position and layout of the key information are relatively fixed. Therefore, in the early stage of the research, there are many methods based on template matching to extract the key information. This method is still widely used in many simple scenarios at present. However, it takes long time to adjut the template for different scenarios.
+
+
+The KIE in the document image generally contains 2 subtasks, which is as shown follows.
+
+* (1) SER: semantic entity recognition, which classifies each detected textline, such as dividing it into name and ID No. As shown in the red boxes in the following figure.
+
+* (2) RE: relationship extraction, which matches the question and answer based on SER results. As shown in the figure below, the yellow arrows match the question and answer.
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185726510-faba470d-2c79-4784-b8da-6c1aa5af9572.png" width="800">
+</div>
+
+
+
+### 1.2 Mainstream Deep-learning Solutions
+
+General KIE methods are based on Named Entity Recognition (NER), but such methods only use text information and ignore location and visual feature information, which leads to limited accuracy. In recent years, most scholars have started to combine mutil-modal features to improve the accuracy of KIE model. The main methods are as follows:
+
+* (1) Grid based methods. These methods mainly focus on the fusion of multi-modal information at the image level. Most texts are of character granularity. The text and structure information embedding method is simple, such as the algorithm of chargrid [1].
+
+* (2) Token based methods. These methods refer to the NLP methods such as Bert, which encode the position, vision and other feature information into the multi-modal model, and conduct pre-training on large-scale datasets, so that in downstream tasks, only a small amount of annotation data is required to obtain excellent results. The representative algorithms are layoutlm [2], layoutlmv2 [3], layoutxlm [4], structext [5], etc.
+
+* (3) GCN based methods. These methods try to learn the structural information between images and characters, so as to solve the problem of extracting open set information (templates not seen in the training set), such as GCN [6], SDMGR [7] and other algorithms.
+
+* (4) End to end based methods: these methods put the existing OCR character recognition and KIE information extraction tasks into a unified network for common learning, and strengthen each other in the learning process. Such as TRIE [8].
+
+
+For more detailed introduction of the algorithms, please refer to Chapter 6 of [Diving into OCR](https://aistudio.baidu.com/aistudio/education/group/info/25207).
+
+## 2. KIE Pipeline
+
+Token based methods such as LayoutXLM are implemented in PaddleOCR. What's more, in PP-StructureV2, we simplify the LayoutXLM model and proposed VI-LayoutXLM, in which the visual feature extraction module is removed for speed-up. The textline sorting strategy conforming to the human reading order and UDML knowledge distillation strategy are utilized for higher model accuracy.
+
+
+In the non end-to-end KIE method, KIE needs at least **2 steps**. Firstly, the OCR model is used to extract the text and its position. Secondly, the KIE model is used to extract the key information according to the image, text position and text content.
+
+
+### 2.1 Train OCR Models
+
+#### 2.1.1 Text Detection
+
+**(1) Data**
+
+Most of the models provided in PaddleOCR are general models. In the process of text detection, the detection of adjacent text lines is generally based on the distance of the position. As shown in the figure above, when using PP-OCRv3 general English detection model for text detection, it is easy to detect the two fields representing different properties as one. Therefore, it is suggested to finetune a detection model according to your scenario firstly during the KIE task.
+
+
+During data annotation, the different key information needs to be separated. Otherwise, it will increase the difficulty of subsequent KIE tasks.
+
+For downstream tasks, generally speaking, `200~300` training images can guarantee the basic training effect. If there is not too much prior knowledge, **`200~300`** images can be labeled firstly for subsequent text detection model training.
+
+**(2) Model**
+
+In terms of model selection, PP-OCRv3 detection model is recommended. For more information about the training methods of the detection model, please refer to: [Text detection tutorial](../../doc/doc_en/detection_en.md) and [PP-OCRv3 detection model tutorial](../../doc/doc_ch/PPOCRv3_det_train.md).
+
+#### 2.1.2 Text recognition
+
+
+Compared with the natural scene, the text recognition in the document image is generally relatively easier (the background is not too complex), so **it is suggested to** try the PP-OCRv3 general text recognition model provided in PaddleOCR ([PP-OCRv3 model list](../../doc/doc_en/models_list_en.md))
+
+
+**(1) Data**
+
+However, there are also some challenges in some document scenarios, such as rare words in ID card scenarios and special fonts in invoice and other scenarios. These problems will increase the difficulty of text recognition. At this time, if you want to ensure or further improve the model accuracy, it is recommended to load PP-OCRv3 model based on the text recognition dataset of specific document scenarios for finetuning.
+
+In the process of model finetuning, it is recommended to prepare at least `5000` vertical scene text recognition images to ensure the basic model fine-tuning effect. If you want to improve the accuracy and generalization ability of the model, you can synthesize more text recognition images similar to the scene, collect general real text recognition data from the public data set, and add them to the text recognition training process. In the training process, it is suggested that the ratio of real data, synthetic data and general data of each epoch should be around `1:1:1`, which can be controlled by setting the sampling ratio of different data sources. If there are 3 training text files, including 10k, 20k and 50k pieces of data respectively, the data can be set in the configuration file as follows:
+
+```yml
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list:
+    - ./train_data/train_list_10k.txt
+    - ./train_data/train_list_10k.txt
+    - ./train_data/train_list_50k.txt
+    ratio_list: [1.0, 0.5, 0.2]
+    ...
+```
+
+**(2) Model**
+
+In terms of model selection, PP-OCRv3 recognition model is recommended. For more information about the training methods of the recognition model, please refer to: [Text recognition tutorial](../../doc/doc_en/recognition_en.md) and [PP-OCRv3 model list](../../doc/doc_en/models_list_en.md).
+
+
+### 2.2 Train KIE Models
+
+There are two main methods to extract the key information from the recognized texts.
+
+(1) Directly use SER model to obtain the key information category. For example, in the ID card scenario, we mark "name" and "Geoff Sample" as "name_key" and "name_value", respectively. The **text field** corresponding to the category "name_value" finally identified is the key information we need.
+
+(2) Joint use SER and RE models. For this case, we firstly use SER model to obtain all questions (keys) and questions (values) for the image text, and then use RE model to match all keys and values to find the relationship, so as to complete the extraction of key information.
+
+#### 2.2.1 SER
+
+Take the ID card scenario as an example. The key information generally includes `name`, `DOB`, etc. We can directly mark the corresponding fields as specific categories, as shown in the following figure.
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185728456-dc396f47-0880-4279-9c7c-c99601bf16a7.png" width="500">
+</div>
+
+**Note:**
+
+- In the labeling process, text content without key information about KIE shall be labeled as`other`, which is equivalent to background information. For example, in the ID card scenario, if we do not pay attention to `DOB` information, we can mark the categories of `DOB` and `Area manager` as `other`.
+- In the annotation process of, it is required to annotate the **textline** position rather than the character.
+
+
+In terms of data, generally speaking, for relatively fixed scenes, **50** training images can achieve acceptable effects. You can refer to [PPOCRLabel](https://github.com/PFCCLab/PPOCRLabel/blob/main/README.md) for finish the labeling process.
+
+In terms of model, it is recommended to use the VI-layoutXLM model proposed in PP-StructureV2. It is improved based on the LayoutXLM model, removing the visual feature extraction module, and further improving the model inference speed without the significant reduction on model accuracy. For more tutorials, please refer to [VI-LayoutXLM introduction](../../doc/doc_en/algorithm_kie_vi_layoutxlm_en.md) and [KIE tutorial](../../doc/doc_en/kie_en.md).
+
+
+#### 2.2.2 SER + RE
+
+The SER model is mainly used to identify all keys and values in the document image, and the RE model is mainly used to match all keys and values.
+
+Taking the ID card scenario as an example, the key information generally includes key information such as `name`, `DOB`, etc. in the SER stage, we need to identify all questions (keys) and answers (values). The demo annotation is as follows. All keys can be annotated as `question`, and all values can be annotated as `answer`.
+
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185728881-b6055e01-034c-4584-aaa6-97c9c25fb61b.png" width="500">
+</div>
+
+
+In the RE stage, the ID and connection information of each field need to be marked, as shown in the following figure.
+
+<div align="center">
+    <img src="https://user-images.githubusercontent.com/14270174/185728948-a4208013-5038-4025-9a93-0c6d51447488.png" width="500">
+</div>
+
+For each textline, you need to add 'ID' and 'linking' field information. The 'ID' records the unique identifier of the textline. Different text contents in the same images cannot be repeated. The 'linking' is a list that records the connection information between different texts. If the ID of the field "name" is 0 and the ID of the field "Geoff Sample" is 1, then they all have [[0, 1]] 'linking' marks, indicating that the fields with `id=0` and `id=1` form a key value relationship (the fields such as DOB and Expires are similar, and will not be repeated here).
+
+
+**Note:**
+
+During annotation, if value is multiple text lines, a key-value pair can be added in linking, such as `[[0, 1], [0, 2]]`.
+
+In terms of data, generally speaking, for relatively fixed scenes, about **50** training images can achieve acceptable effects.
+
+In terms of model, it is recommended to use the VI-layoutXLM model proposed in PP-StructureV2. It is improved based on the LayoutXLM model, removing the visual feature extraction module, and further improving the model inference speed without the significant reduction on model accuracy. For more tutorials, please refer to [VI-LayoutXLM introduction](../../doc/doc_en/algorithm_kie_vi_layoutxlm_en.md) and [KIE tutorial](../../doc/doc_en/kie_en.md).
+
+
+
+## 3. Reference
+
+
+[1] Katti A R, Reisswig C, Guder C, et al. Chargrid: Towards understanding 2d documents[J]. arXiv preprint arXiv:1809.08799, 2018.
+
+[2] Xu Y, Li M, Cui L, et al. Layoutlm: Pre-training of text and layout for document image understanding[C]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2020: 1192-1200.
+
+[3] Xu Y, Xu Y, Lv T, et al. LayoutLMv2: Multi-modal pre-training for visually-rich document understanding[J]. arXiv preprint arXiv:2012.14740, 2020.
+
+[4]: Xu Y, Lv T, Cui L, et al. Layoutxlm: Multimodal pre-training for multilingual visually-rich document understanding[J]. arXiv preprint arXiv:2104.08836, 2021.
+
+[5] Li Y, Qian Y, Yu Y, et al. StrucTexT: Structured Text Understanding with Multi-Modal Transformers[C]//Proceedings of the 29th ACM International Conference on Multimedia. 2021: 1912-1920.
+
+[6] Liu X, Gao F, Zhang Q, et al. Graph convolution for multimodal information extraction from visually rich documents[J]. arXiv preprint arXiv:1903.11279, 2019.
+
+[7] Sun H, Kuang Z, Yue X, et al. Spatial Dual-Modality Graph Reasoning for Key Information Extraction[J]. arXiv preprint arXiv:2103.14470, 2021.
+
+[8] Zhang P, Xu Y, Cheng Z, et al. Trie: End-to-end text reading and information extraction for document understanding[C]//Proceedings of the 28th ACM International Conference on Multimedia. 2020: 1413-1422.
diff --git a/ppstructure/kie/predict_kie_token_ser.py b/ppstructure/kie/predict_kie_token_ser.py
new file mode 100644
index 0000000..9508e60
--- /dev/null
+++ b/ppstructure/kie/predict_kie_token_ser.py
@@ -0,0 +1,191 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "../..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+
+import cv2
+import json
+import numpy as np
+import time
+
+import tools.infer.utility as utility
+from ppocr.data import create_operators, transform
+from ppocr.postprocess import build_post_process
+from ppocr.utils.logging import get_logger
+from ppocr.utils.visual import draw_ser_results
+from ppocr.utils.utility import get_image_file_list, check_and_read
+from ppstructure.utility import parse_args
+
+from paddleocr import PaddleOCR
+
+logger = get_logger()
+
+
+class SerPredictor(object):
+    def __init__(self, args):
+        self.args = args
+        self.ocr_engine = PaddleOCR(
+            use_angle_cls=args.use_angle_cls,
+            det_model_dir=args.det_model_dir,
+            rec_model_dir=args.rec_model_dir,
+            show_log=False,
+            use_gpu=args.use_gpu,
+        )
+
+        pre_process_list = [
+            {
+                "VQATokenLabelEncode": {
+                    "algorithm": args.kie_algorithm,
+                    "class_path": args.ser_dict_path,
+                    "contains_re": False,
+                    "ocr_engine": self.ocr_engine,
+                    "order_method": args.ocr_order_method,
+                }
+            },
+            {"VQATokenPad": {"max_seq_len": 512, "return_attention_mask": True}},
+            {"VQASerTokenChunk": {"max_seq_len": 512, "return_attention_mask": True}},
+            {"Resize": {"size": [224, 224]}},
+            {
+                "NormalizeImage": {
+                    "std": [58.395, 57.12, 57.375],
+                    "mean": [123.675, 116.28, 103.53],
+                    "scale": "1",
+                    "order": "hwc",
+                }
+            },
+            {"ToCHWImage": None},
+            {
+                "KeepKeys": {
+                    "keep_keys": [
+                        "input_ids",
+                        "bbox",
+                        "attention_mask",
+                        "token_type_ids",
+                        "image",
+                        "labels",
+                        "segment_offset_id",
+                        "ocr_info",
+                        "entities",
+                    ]
+                }
+            },
+        ]
+        postprocess_params = {
+            "name": "VQASerTokenLayoutLMPostProcess",
+            "class_path": args.ser_dict_path,
+        }
+
+        self.preprocess_op = create_operators(pre_process_list, {"infer_mode": True})
+        self.postprocess_op = build_post_process(postprocess_params)
+        (
+            self.predictor,
+            self.input_tensor,
+            self.output_tensors,
+            self.config,
+        ) = utility.create_predictor(args, "ser", logger)
+
+    def __call__(self, img):
+        ori_im = img.copy()
+        data = {"image": img}
+        data = transform(data, self.preprocess_op)
+        if data[0] is None:
+            return None, 0
+        starttime = time.time()
+
+        for idx in range(len(data)):
+            if isinstance(data[idx], np.ndarray):
+                data[idx] = np.expand_dims(data[idx], axis=0)
+            else:
+                data[idx] = [data[idx]]
+        if self.args.use_onnx:
+            input_tensor = {
+                name: data[idx] for idx, name in enumerate(self.input_tensor)
+            }
+            self.output_tensors = self.predictor.run(None, input_tensor)
+        else:
+            for idx in range(len(self.input_tensor)):
+                self.input_tensor[idx].copy_from_cpu(data[idx])
+
+            self.predictor.run()
+
+        outputs = []
+        for output_tensor in self.output_tensors:
+            output = (
+                output_tensor if self.args.use_onnx else output_tensor.copy_to_cpu()
+            )
+            outputs.append(output)
+        preds = outputs[0]
+
+        post_result = self.postprocess_op(
+            preds, segment_offset_ids=data[6], ocr_infos=data[7]
+        )
+        elapse = time.time() - starttime
+        return post_result, data, elapse
+
+
+def main(args):
+    image_file_list = get_image_file_list(args.image_dir)
+    ser_predictor = SerPredictor(args)
+    count = 0
+    total_time = 0
+
+    os.makedirs(args.output, exist_ok=True)
+    with open(
+        os.path.join(args.output, "infer.txt"), mode="w", encoding="utf-8"
+    ) as f_w:
+        for image_file in image_file_list:
+            img, flag, _ = check_and_read(image_file)
+            if not flag:
+                img = cv2.imread(image_file)
+                img = img[:, :, ::-1]
+            if img is None:
+                logger.info("error in loading image:{}".format(image_file))
+                continue
+            ser_res, _, elapse = ser_predictor(img)
+            ser_res = ser_res[0]
+
+            res_str = "{}\t{}\n".format(
+                image_file,
+                json.dumps(
+                    {
+                        "ocr_info": ser_res,
+                    },
+                    ensure_ascii=False,
+                ),
+            )
+            f_w.write(res_str)
+
+            img_res = draw_ser_results(
+                image_file,
+                ser_res,
+                font_path=args.vis_font_path,
+            )
+
+            img_save_path = os.path.join(args.output, os.path.basename(image_file))
+            cv2.imwrite(img_save_path, img_res)
+            logger.info("save vis result to {}".format(img_save_path))
+            if count > 0:
+                total_time += elapse
+            count += 1
+            logger.info("Predict time of {}: {}".format(image_file, elapse))
+
+
+if __name__ == "__main__":
+    main(parse_args())
diff --git a/ppstructure/kie/predict_kie_token_ser_re.py b/ppstructure/kie/predict_kie_token_ser_re.py
new file mode 100644
index 0000000..ebf5988
--- /dev/null
+++ b/ppstructure/kie/predict_kie_token_ser_re.py
@@ -0,0 +1,144 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "../..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+
+import cv2
+import json
+import numpy as np
+import time
+
+import tools.infer.utility as utility
+from tools.infer_kie_token_ser_re import make_input
+from ppocr.postprocess import build_post_process
+from ppocr.utils.logging import get_logger
+from ppocr.utils.visual import draw_ser_results, draw_re_results
+from ppocr.utils.utility import get_image_file_list, check_and_read
+from ppstructure.utility import parse_args
+from ppstructure.kie.predict_kie_token_ser import SerPredictor
+
+logger = get_logger()
+
+
+class SerRePredictor(object):
+    def __init__(self, args):
+        self.use_visual_backbone = args.use_visual_backbone
+        self.ser_engine = SerPredictor(args)
+        if args.re_model_dir is not None:
+            postprocess_params = {"name": "VQAReTokenLayoutLMPostProcess"}
+            self.postprocess_op = build_post_process(postprocess_params)
+            (
+                self.predictor,
+                self.input_tensor,
+                self.output_tensors,
+                self.config,
+            ) = utility.create_predictor(args, "re", logger)
+        else:
+            self.predictor = None
+
+    def __call__(self, img):
+        starttime = time.time()
+        ser_results, ser_inputs, ser_elapse = self.ser_engine(img)
+        if self.predictor is None:
+            return ser_results, ser_elapse
+
+        re_input, entity_idx_dict_batch = make_input(ser_inputs, ser_results)
+        if self.use_visual_backbone == False:
+            re_input.pop(4)
+        for idx in range(len(self.input_tensor)):
+            self.input_tensor[idx].copy_from_cpu(re_input[idx])
+
+        self.predictor.run()
+        outputs = []
+        for output_tensor in self.output_tensors:
+            output = output_tensor.copy_to_cpu()
+            outputs.append(output)
+        preds = dict(
+            loss=outputs[1],
+            pred_relations=outputs[2],
+            hidden_states=outputs[0],
+        )
+
+        post_result = self.postprocess_op(
+            preds, ser_results=ser_results, entity_idx_dict_batch=entity_idx_dict_batch
+        )
+
+        elapse = time.time() - starttime
+        return post_result, elapse
+
+
+def main(args):
+    image_file_list = get_image_file_list(args.image_dir)
+    ser_re_predictor = SerRePredictor(args)
+    count = 0
+    total_time = 0
+
+    os.makedirs(args.output, exist_ok=True)
+    with open(
+        os.path.join(args.output, "infer.txt"), mode="w", encoding="utf-8"
+    ) as f_w:
+        for image_file in image_file_list:
+            img, flag, _ = check_and_read(image_file)
+            if not flag:
+                img = cv2.imread(image_file)
+                img = img[:, :, ::-1]
+            if img is None:
+                logger.info("error in loading image:{}".format(image_file))
+                continue
+            re_res, elapse = ser_re_predictor(img)
+            re_res = re_res[0]
+
+            res_str = "{}\t{}\n".format(
+                image_file,
+                json.dumps(
+                    {
+                        "ocr_info": re_res,
+                    },
+                    ensure_ascii=False,
+                ),
+            )
+            f_w.write(res_str)
+            if ser_re_predictor.predictor is not None:
+                img_res = draw_re_results(
+                    image_file, re_res, font_path=args.vis_font_path
+                )
+                img_save_path = os.path.join(
+                    args.output,
+                    os.path.splitext(os.path.basename(image_file))[0] + "_ser_re.jpg",
+                )
+            else:
+                img_res = draw_ser_results(
+                    image_file, re_res, font_path=args.vis_font_path
+                )
+                img_save_path = os.path.join(
+                    args.output,
+                    os.path.splitext(os.path.basename(image_file))[0] + "_ser.jpg",
+                )
+
+            cv2.imwrite(img_save_path, img_res)
+            logger.info("save vis result to {}".format(img_save_path))
+            if count > 0:
+                total_time += elapse
+            count += 1
+            logger.info("Predict time of {}: {}".format(image_file, elapse))
+
+
+if __name__ == "__main__":
+    main(parse_args())
diff --git a/ppstructure/kie/requirements.txt b/ppstructure/kie/requirements.txt
new file mode 100644
index 0000000..500026e
--- /dev/null
+++ b/ppstructure/kie/requirements.txt
@@ -0,0 +1,7 @@
+sentencepiece
+yacs
+seqeval
+pypandoc
+attrdict3
+python_docx
+paddlenlp==2.5.2
diff --git a/ppstructure/kie/tools/eval_with_label_end2end.py b/ppstructure/kie/tools/eval_with_label_end2end.py
new file mode 100644
index 0000000..f97da12
--- /dev/null
+++ b/ppstructure/kie/tools/eval_with_label_end2end.py
@@ -0,0 +1,259 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import re
+import sys
+import shapely
+from shapely.geometry import Polygon
+import numpy as np
+from collections import defaultdict
+import operator
+from rapidfuzz.distance import Levenshtein
+import argparse
+import json
+import copy
+
+
+def parse_ser_results_fp(fp, fp_type="gt", ignore_background=True):
+    # img/zh_val_0.jpg        {
+    #     "height": 3508,
+    #     "width": 2480,
+    #     "ocr_info": [
+    #         {"text": "Maribyrnong", "label": "other", "bbox": [1958, 144, 2184, 198]},
+    #         {"text": "CITYCOUNCIL", "label": "other", "bbox": [2052, 183, 2171, 214]},
+    #     ]
+    assert fp_type in ["gt", "pred"]
+    key = "label" if fp_type == "gt" else "pred"
+    res_dict = dict()
+    with open(fp, "r", encoding="utf-8") as fin:
+        lines = fin.readlines()
+
+    for _, line in enumerate(lines):
+        img_path, info = line.strip().split("\t")
+        # get key
+        image_name = os.path.basename(img_path)
+        res_dict[image_name] = []
+        # get infos
+        json_info = json.loads(info)
+        for single_ocr_info in json_info["ocr_info"]:
+            label = single_ocr_info[key].upper()
+            if label in ["O", "OTHERS", "OTHER"]:
+                label = "O"
+            if ignore_background and label == "O":
+                continue
+            single_ocr_info["label"] = label
+            res_dict[image_name].append(copy.deepcopy(single_ocr_info))
+    return res_dict
+
+
+def polygon_from_str(polygon_points):
+    """
+    Create a shapely polygon object from gt or dt line.
+    """
+    polygon_points = np.array(polygon_points).reshape(4, 2)
+    polygon = Polygon(polygon_points).convex_hull
+    return polygon
+
+
+def polygon_iou(poly1, poly2):
+    """
+    Intersection over union between two shapely polygons.
+    """
+    if not poly1.intersects(poly2):  # this test is fast and can accelerate calculation
+        iou = 0
+    else:
+        try:
+            inter_area = poly1.intersection(poly2).area
+            union_area = poly1.area + poly2.area - inter_area
+            iou = float(inter_area) / union_area
+        except shapely.geos.TopologicalError:
+            # except Exception as e:
+            #     print(e)
+            print("shapely.geos.TopologicalError occurred, iou set to 0")
+            iou = 0
+    return iou
+
+
+def ed(args, str1, str2):
+    if args.ignore_space:
+        str1 = str1.replace(" ", "")
+        str2 = str2.replace(" ", "")
+    if args.ignore_case:
+        str1 = str1.lower()
+        str2 = str2.lower()
+    return Levenshtein.distance(str1, str2)
+
+
+def convert_bbox_to_polygon(bbox):
+    """
+    bbox  : [x1, y1, x2, y2]
+    output: [[x1, y1], [x2, y2], [x3, y3], [x4, y4]]
+    """
+    xmin, ymin, xmax, ymax = bbox
+    poly = [[xmin, ymin], [xmax, ymin], [xmax, ymax], [xmin, ymax]]
+    return poly
+
+
+def eval_e2e(args):
+    # gt
+    gt_results = parse_ser_results_fp(args.gt_json_path, "gt", args.ignore_background)
+    # pred
+    dt_results = parse_ser_results_fp(
+        args.pred_json_path, "pred", args.ignore_background
+    )
+    iou_thresh = args.iou_thres
+    num_gt_chars = 0
+    gt_count = 0
+    dt_count = 0
+    hit = 0
+    ed_sum = 0
+
+    for img_name in dt_results:
+        gt_info = gt_results[img_name]
+        gt_count += len(gt_info)
+
+        dt_info = dt_results[img_name]
+        dt_count += len(dt_info)
+
+        dt_match = [False] * len(dt_info)
+        gt_match = [False] * len(gt_info)
+
+        all_ious = defaultdict(tuple)
+        # gt: {text, label, bbox or poly}
+        for index_gt, gt in enumerate(gt_info):
+            if "poly" not in gt:
+                gt["poly"] = convert_bbox_to_polygon(gt["bbox"])
+            gt_poly = polygon_from_str(gt["poly"])
+            for index_dt, dt in enumerate(dt_info):
+                if "poly" not in dt:
+                    dt["poly"] = convert_bbox_to_polygon(dt["bbox"])
+                dt_poly = polygon_from_str(dt["poly"])
+                iou = polygon_iou(dt_poly, gt_poly)
+                if iou >= iou_thresh:
+                    all_ious[(index_gt, index_dt)] = iou
+        sorted_ious = sorted(all_ious.items(), key=operator.itemgetter(1), reverse=True)
+        sorted_gt_dt_pairs = [item[0] for item in sorted_ious]
+
+        # matched gt and dt
+        for gt_dt_pair in sorted_gt_dt_pairs:
+            index_gt, index_dt = gt_dt_pair
+            if gt_match[index_gt] == False and dt_match[index_dt] == False:
+                gt_match[index_gt] = True
+                dt_match[index_dt] = True
+                # ocr rec results
+                gt_text = gt_info[index_gt]["text"]
+                dt_text = dt_info[index_dt]["text"]
+
+                # ser results
+                gt_label = gt_info[index_gt]["label"]
+                dt_label = dt_info[index_dt]["pred"]
+
+                if True:  # ignore_masks[index_gt] == '0':
+                    ed_sum += ed(args, gt_text, dt_text)
+                    num_gt_chars += len(gt_text)
+                    if gt_text == dt_text:
+                        if args.ignore_ser_prediction or gt_label == dt_label:
+                            hit += 1
+
+        # unmatched dt
+        for tindex, dt_match_flag in enumerate(dt_match):
+            if dt_match_flag == False:
+                dt_text = dt_info[tindex]["text"]
+                gt_text = ""
+                ed_sum += ed(args, dt_text, gt_text)
+
+        # unmatched gt
+        for tindex, gt_match_flag in enumerate(gt_match):
+            if gt_match_flag == False:
+                dt_text = ""
+                gt_text = gt_info[tindex]["text"]
+                ed_sum += ed(args, gt_text, dt_text)
+                num_gt_chars += len(gt_text)
+
+    eps = 1e-9
+    print("config: ", args)
+    print("hit, dt_count, gt_count", hit, dt_count, gt_count)
+    precision = hit / (dt_count + eps)
+    recall = hit / (gt_count + eps)
+    fmeasure = 2.0 * precision * recall / (precision + recall + eps)
+    avg_edit_dist_img = ed_sum / len(gt_results)
+    avg_edit_dist_field = ed_sum / (gt_count + eps)
+    character_acc = 1 - ed_sum / (num_gt_chars + eps)
+
+    print("character_acc: %.2f" % (character_acc * 100) + "%")
+    print("avg_edit_dist_field: %.2f" % (avg_edit_dist_field))
+    print("avg_edit_dist_img: %.2f" % (avg_edit_dist_img))
+    print("precision: %.2f" % (precision * 100) + "%")
+    print("recall: %.2f" % (recall * 100) + "%")
+    print("fmeasure: %.2f" % (fmeasure * 100) + "%")
+
+    return
+
+
+def parse_args():
+    """ """
+
+    def str2bool(v):
+        return v.lower() in ("true", "t", "1")
+
+    parser = argparse.ArgumentParser()
+    ## Required parameters
+    parser.add_argument(
+        "--gt_json_path",
+        default=None,
+        type=str,
+        required=True,
+    )
+    parser.add_argument(
+        "--pred_json_path",
+        default=None,
+        type=str,
+        required=True,
+    )
+
+    parser.add_argument("--iou_thres", default=0.5, type=float)
+
+    parser.add_argument(
+        "--ignore_case",
+        default=False,
+        type=str2bool,
+        help="whether to do lower case for the strs",
+    )
+
+    parser.add_argument(
+        "--ignore_space", default=True, type=str2bool, help="whether to ignore space"
+    )
+
+    parser.add_argument(
+        "--ignore_background",
+        default=True,
+        type=str2bool,
+        help="whether to ignore other label",
+    )
+
+    parser.add_argument(
+        "--ignore_ser_prediction",
+        default=False,
+        type=str2bool,
+        help="whether to ignore ocr pred results",
+    )
+
+    args = parser.parse_args()
+    return args
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    eval_e2e(args)
diff --git a/ppstructure/kie/tools/trans_funsd_label.py b/ppstructure/kie/tools/trans_funsd_label.py
new file mode 100644
index 0000000..104cb68
--- /dev/null
+++ b/ppstructure/kie/tools/trans_funsd_label.py
@@ -0,0 +1,166 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import json
+import os
+import sys
+import cv2
+import numpy as np
+from copy import deepcopy
+
+
+def trans_poly_to_bbox(poly):
+    x1 = np.min([p[0] for p in poly])
+    x2 = np.max([p[0] for p in poly])
+    y1 = np.min([p[1] for p in poly])
+    y2 = np.max([p[1] for p in poly])
+    return [x1, y1, x2, y2]
+
+
+def get_outer_poly(bbox_list):
+    x1 = min([bbox[0] for bbox in bbox_list])
+    y1 = min([bbox[1] for bbox in bbox_list])
+    x2 = max([bbox[2] for bbox in bbox_list])
+    y2 = max([bbox[3] for bbox in bbox_list])
+    return [[x1, y1], [x2, y1], [x2, y2], [x1, y2]]
+
+
+def load_funsd_label(image_dir, anno_dir):
+    imgs = os.listdir(image_dir)
+    annos = os.listdir(anno_dir)
+
+    imgs = [img.replace(".png", "") for img in imgs]
+    annos = [anno.replace(".json", "") for anno in annos]
+
+    fn_info_map = dict()
+    for anno_fn in annos:
+        res = []
+        with open(os.path.join(anno_dir, anno_fn + ".json"), "r") as fin:
+            infos = json.load(fin)
+            infos = infos["form"]
+            old_id2new_id_map = dict()
+            global_new_id = 0
+            for info in infos:
+                if info["text"] is None:
+                    continue
+                words = info["words"]
+                if len(words) <= 0:
+                    continue
+                word_idx = 1
+                curr_bboxes = [words[0]["box"]]
+                curr_texts = [words[0]["text"]]
+                while word_idx < len(words):
+                    # switch to a new link
+                    if words[word_idx]["box"][0] + 10 <= words[word_idx - 1]["box"][2]:
+                        if len("".join(curr_texts[0])) > 0:
+                            res.append(
+                                {
+                                    "transcription": " ".join(curr_texts),
+                                    "label": info["label"],
+                                    "points": get_outer_poly(curr_bboxes),
+                                    "linking": info["linking"],
+                                    "id": global_new_id,
+                                }
+                            )
+                            if info["id"] not in old_id2new_id_map:
+                                old_id2new_id_map[info["id"]] = []
+                            old_id2new_id_map[info["id"]].append(global_new_id)
+                            global_new_id += 1
+                        curr_bboxes = [words[word_idx]["box"]]
+                        curr_texts = [words[word_idx]["text"]]
+                    else:
+                        curr_bboxes.append(words[word_idx]["box"])
+                        curr_texts.append(words[word_idx]["text"])
+                    word_idx += 1
+                if len("".join(curr_texts[0])) > 0:
+                    res.append(
+                        {
+                            "transcription": " ".join(curr_texts),
+                            "label": info["label"],
+                            "points": get_outer_poly(curr_bboxes),
+                            "linking": info["linking"],
+                            "id": global_new_id,
+                        }
+                    )
+                    if info["id"] not in old_id2new_id_map:
+                        old_id2new_id_map[info["id"]] = []
+                    old_id2new_id_map[info["id"]].append(global_new_id)
+                    global_new_id += 1
+            res = sorted(res, key=lambda r: (r["points"][0][1], r["points"][0][0]))
+            for i in range(len(res) - 1):
+                for j in range(i, 0, -1):
+                    if abs(
+                        res[j + 1]["points"][0][1] - res[j]["points"][0][1]
+                    ) < 20 and (res[j + 1]["points"][0][0] < res[j]["points"][0][0]):
+                        tmp = deepcopy(res[j])
+                        res[j] = deepcopy(res[j + 1])
+                        res[j + 1] = deepcopy(tmp)
+                    else:
+                        break
+            # re-generate unique ids
+            for idx, r in enumerate(res):
+                new_links = []
+                for link in r["linking"]:
+                    # illegal links will be removed
+                    if (
+                        link[0] not in old_id2new_id_map
+                        or link[1] not in old_id2new_id_map
+                    ):
+                        continue
+                    for src in old_id2new_id_map[link[0]]:
+                        for dst in old_id2new_id_map[link[1]]:
+                            new_links.append([src, dst])
+                res[idx]["linking"] = deepcopy(new_links)
+
+            fn_info_map[anno_fn] = res
+
+    return fn_info_map
+
+
+def main():
+    test_image_dir = "train_data/FUNSD/testing_data/images/"
+    test_anno_dir = "train_data/FUNSD/testing_data/annotations/"
+    test_output_dir = "train_data/FUNSD/test.json"
+
+    fn_info_map = load_funsd_label(test_image_dir, test_anno_dir)
+    with open(test_output_dir, "w") as fout:
+        for fn in fn_info_map:
+            fout.write(
+                fn
+                + ".png"
+                + "\t"
+                + json.dumps(fn_info_map[fn], ensure_ascii=False)
+                + "\n"
+            )
+
+    train_image_dir = "train_data/FUNSD/training_data/images/"
+    train_anno_dir = "train_data/FUNSD/training_data/annotations/"
+    train_output_dir = "train_data/FUNSD/train.json"
+
+    fn_info_map = load_funsd_label(train_image_dir, train_anno_dir)
+    with open(train_output_dir, "w") as fout:
+        for fn in fn_info_map:
+            fout.write(
+                fn
+                + ".png"
+                + "\t"
+                + json.dumps(fn_info_map[fn], ensure_ascii=False)
+                + "\n"
+            )
+    print("====ok====")
+    return
+
+
+if __name__ == "__main__":
+    main()
diff --git a/ppstructure/kie/tools/trans_xfun_data.py b/ppstructure/kie/tools/trans_xfun_data.py
new file mode 100644
index 0000000..4e83b5a
--- /dev/null
+++ b/ppstructure/kie/tools/trans_xfun_data.py
@@ -0,0 +1,64 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import json
+
+
+def transfer_xfun_data(json_path=None, output_file=None):
+    with open(json_path, "r", encoding="utf-8") as fin:
+        lines = fin.readlines()
+
+    json_info = json.loads(lines[0])
+    documents = json_info["documents"]
+    with open(output_file, "w", encoding="utf-8") as fout:
+        for idx, document in enumerate(documents):
+            label_info = []
+            img_info = document["img"]
+            document = document["document"]
+            image_path = img_info["fname"]
+
+            for doc in document:
+                x1, y1, x2, y2 = doc["box"]
+                points = [[x1, y1], [x2, y1], [x2, y2], [x1, y2]]
+                label_info.append(
+                    {
+                        "transcription": doc["text"],
+                        "label": doc["label"],
+                        "points": points,
+                        "id": doc["id"],
+                        "linking": doc["linking"],
+                    }
+                )
+
+            fout.write(
+                image_path + "\t" + json.dumps(label_info, ensure_ascii=False) + "\n"
+            )
+
+    print("===ok====")
+
+
+def parser_args():
+    import argparse
+
+    parser = argparse.ArgumentParser(description="args for paddleserving")
+    parser.add_argument(
+        "--ori_gt_path", type=str, required=True, help="origin xfun gt path"
+    )
+    parser.add_argument("--output_path", type=str, required=True, help="path to save")
+    args = parser.parse_args()
+    return args
+
+
+args = parser_args()
+transfer_xfun_data(args.ori_gt_path, args.output_path)
diff --git a/ppstructure/layout/README.md b/ppstructure/layout/README.md
new file mode 100644
index 0000000..6830f8e
--- /dev/null
+++ b/ppstructure/layout/README.md
@@ -0,0 +1,470 @@
+English | [简体中文](README_ch.md)
+
+# Layout analysis
+
+- [1. Introduction](#1-Introduction)
+- [2. Quick start](#2-Quick-start)
+- [3. Install](#3-Install)
+  - [3.1 Install PaddlePaddle](#31-Install-paddlepaddle)
+  - [3.2 Install PaddleDetection](#32-Install-paddledetection)
+- [4. Data preparation](#4-Data-preparation)
+  - [4.1 English data set](#41-English-data-set)
+  - [4.2 More datasets](#42-More-datasets)
+- [5. Start training](#5-Start-training)
+  - [5.1 Train](#51-Train)
+  - [5.2 FGD Distillation training](#52-Fgd-distillation-training)
+- [6. Model evaluation and prediction](#6-Model-evaluation-and-prediction)
+  - [6.1 Indicator evaluation](#61-Indicator-evaluation)
+  - [6.2 Test layout analysis results](#62-Test-layout-analysis-results)
+- [7. Model export and inference](#7-Model-export-and-inference)
+  - [7.1 Model export](#71-Model-export)
+  - [7.2 Model inference](#72-Model-inference)
+
+
+## 1. Introduction
+
+Layout analysis refers to the regional division of documents in the form of pictures and the positioning of key areas, such as text, title, table, picture, etc. The layout analysis algorithm is based on the lightweight model PP-picodet of [PaddleDetection]( https://github.com/PaddlePaddle/PaddleDetection ), including English layout analysis, Chinese layout analysis and table layout analysis models.  English layout analysis models can detect document layout elements such as text, title, table, figure, list. Chinese layout analysis models can detect document layout elements such as text, figure, figure caption, table, table caption, header, footer, reference, and equation. Table layout analysis models can detect table regions.
+
+<div align="center">
+    <img src="../docs/layout/layout.png" width="800">
+</div>
+
+## 2. Quick start
+PP-Structure currently provides layout analysis models in Chinese, English and table documents. For the model link, see [models_list](../docs/models_list_en.md). The whl package is also provided for quick use, see [quickstart](../docs/quickstart_en.md) for details.
+
+## 3. Install
+
+### 3.1. Install PaddlePaddle
+
+- **（1) Install PaddlePaddle**
+
+```bash
+python3 -m pip install --upgrade pip
+
+# GPU Install
+python3 -m pip install "paddlepaddle-gpu>=2.3" -i https://mirror.baidu.com/pypi/simple
+
+# CPU Install
+python3 -m pip install "paddlepaddle>=2.3" -i https://mirror.baidu.com/pypi/simple
+```
+For more requirements, please refer to the instructions in the [Install file](https://www.paddlepaddle.org.cn/install/quick)。
+
+### 3.2. Install PaddleDetection
+
+- **（1）Download PaddleDetection Source code**
+
+```bash
+git clone https://github.com/PaddlePaddle/PaddleDetection.git
+```
+
+- **（2）Install third-party libraries**
+
+```bash
+cd PaddleDetection
+python3 -m pip install -r requirements.txt
+```
+
+## 4. Data preparation
+
+If you want to experience the prediction process directly, you can skip data preparation and download the pre-training model.
+
+### 4.1. English data set
+
+Download document analysis data set [PubLayNet](https://developer.ibm.com/exchanges/data/all/publaynet/)（Dataset 96G），contains 5 classes：`{0: "Text", 1: "Title", 2: "List", 3:"Table", 4:"Figure"}`
+
+```
+# Download data
+wget https://dax-cdn.cdn.appdomain.cloud/dax-publaynet/1.0.0/publaynet.tar.gz
+# Decompress data
+tar -xvf publaynet.tar.gz
+```
+
+Uncompressed **directory structure：**
+
+```
+|-publaynet
+  |- test
+     |- PMC1277013_00004.jpg
+     |- PMC1291385_00002.jpg
+     | ...
+  |- train.json
+  |- train
+     |- PMC1291385_00002.jpg
+     |- PMC1277013_00004.jpg
+     | ...
+  |- val.json
+  |- val
+     |- PMC538274_00004.jpg
+     |- PMC539300_00004.jpg
+     | ...
+```
+
+**data distribution：**
+
+| File or Folder | Description    | num     |
+| :------------- | :------------- | ------- |
+| `train/`       | Training set pictures     | 335,703 |
+| `val/`         | Verification set pictures     | 11,245  |
+| `test/`        | Test set pictures     | 11,405  |
+| `train.json`   | Training set annotation files | -       |
+| `val.json`     | Validation set dimension files | -       |
+
+**Data Annotation**
+
+The JSON file contains the annotations of all images, and the data is stored in a dictionary nested manner.Contains the following keys：
+
+- info，represents the dimension file info。
+
+- licenses，represents the dimension file licenses。
+
+- images，represents the list of image information in the annotation file，each element is the information of an image。The information of one of the images is as follows:
+
+  ```
+  {
+      'file_name': 'PMC4055390_00006.jpg',    # file_name
+      'height': 601,                      # image height
+      'width': 792,                       # image width
+      'id': 341427                        # image id
+  }
+  ```
+
+- annotations， represents the list of annotation information of the target object in the annotation file，each element is the annotation information of a target object。The following is the annotation information of one of the target objects:
+
+  ```
+  {
+
+      'segmentation':             # Segmentation annotation of objects
+      'area': 60518.099043117836, # Area of object
+      'iscrowd': 0,               # iscrowd
+      'image_id': 341427,         # image id
+      'bbox': [50.58, 490.86, 240.15, 252.16], # bbox [x1,y1,w,h]
+      'category_id': 1,           # category_id
+      'id': 3322348               # image id
+  }
+  ```
+
+### 4.2. More datasets
+
+We provide CDLA(Chinese layout analysis), TableBank(Table layout analysis)etc. data set download links，process to the JSON format of the above annotation file，that is, the training can be conducted in the same way。
+
+| dataset                                                      | 简介                                                         |
+| ------------------------------------------------------------ | ------------------------------------------------------------ |
+| [cTDaR2019_cTDaR](https://cndplab-founder.github.io/cTDaR2019/) | For form detection (TRACKA) and form identification (TRACKB).Image types include historical data sets (beginning with cTDaR_t0, such as CTDAR_T00872.jpg) and modern data sets (beginning with cTDaR_t1, CTDAR_T10482.jpg). |
+| [IIIT-AR-13K](http://cvit.iiit.ac.in/usodi/iiitar13k.php)    | Data sets constructed by manually annotating figures or pages from publicly available annual reports, containing 5 categories:table, figure, natural image, logo, and signature. |
+| [TableBank](https://github.com/doc-analysis/TableBank)       | For table detection and recognition of large datasets, including Word and Latex document formats |
+| [CDLA](https://github.com/buptlihang/CDLA)                   | Chinese document layout analysis data set, for Chinese literature (paper) scenarios, including 10 categories:Text, Title, Figure, Figure caption, Table, Table caption, Header, Footer, Reference, Equation |
+| [DocBank](https://github.com/doc-analysis/DocBank)           | Large-scale dataset (500K document pages) constructed using weakly supervised methods for document layout analysis, containing 12 categories:Author, Caption, Date, Equation, Figure, Footer, List, Paragraph, Reference, Section, Table, Title |
+
+
+## 5. Start training
+
+Training scripts, evaluation scripts, and prediction scripts are provided, and the PubLayNet pre-training model is used as an example in this section.
+
+If you do not want training and directly experience the following process of model evaluation, prediction, motion to static, and inference, you can download the provided pre-trained model (PubLayNet dataset) and skip this part.
+
+```
+mkdir pretrained_model
+cd pretrained_model
+# Download PubLayNet pre-training model（Direct experience model evaluates, predicts, and turns static）
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout.pdparams
+# Download the PubLaynet inference model（Direct experience model reasoning）
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar
+```
+
+If the test image is Chinese, the pre-trained model of Chinese CDLA dataset can be downloaded to identify 10 types of document regions：Table, Figure, Figure caption, Table, Table caption, Header, Footer, Reference, Equation，Download the training model and inference model of Model 'picodet_lcnet_x1_0_fgd_layout_cdla' in [layout analysis model](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppstructure/docs/models_list.md)。If only the table area in the image is detected, you can download the pre-trained model of the table dataset, and download the training model and inference model of the 'picodet_LCnet_x1_0_FGd_layout_table' model in [Layout Analysis model](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppstructure/docs/models_list.md)
+
+### 5.1. Train
+
+Start training with the PaddleDetection [layout analysis profile](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/picodet/legacy_model/application/layout_analysis)
+
+* Modify Profile
+
+If you want to train your own data set, you need to modify the data configuration and the number of categories in the configuration file.
+
+
+Using 'configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml' as an example, the change is as follows:
+
+```yaml
+metric: COCO
+# Number of categories
+num_classes: 5
+
+TrainDataset:
+  !COCODataSet
+    # Modify to your own training data directory
+    image_dir: train
+    # Modify to your own training data label file
+    anno_path: train.json
+    # Modify to your own training data root directory
+    dataset_dir: /root/publaynet/
+    data_fields: ['image', 'gt_bbox', 'gt_class', 'is_crowd']
+
+EvalDataset:
+  !COCODataSet
+    # Modify to your own validation data directory
+    image_dir: val
+    # Modify to your own validation data label file
+    anno_path: val.json
+    # Modify to your own validation data root
+    dataset_dir: /root/publaynet/
+
+TestDataset:
+  !ImageFolder
+    # Modify to your own test data label file
+    anno_path: /root/publaynet/val.json
+```
+
+* Start training. During training, PP picodet pre training model will be downloaded by default. There is no need to download in advance.
+
+```bash
+# GPU training supports single-card and multi-card training
+# The training log is automatically saved to the log directory
+
+# Single card training
+export CUDA_VISIBLE_DEVICES=0
+python3 tools/train.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --eval
+
+# Multi-card training, with the -- GPUS parameter specifying the card number
+export CUDA_VISIBLE_DEVICES=0,1,2,3
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --eval
+```
+
+**Attention：**If the video memory is out during training, adjust Batch_size in TrainReader and base_LR in LearningRate. The published config is obtained by 8-card training. If the number of GPU cards is changed to 1, then the base_LR needs to be reduced by 8 times.
+
+After starting training normally, you will see the following log output:
+
+```
+[08/15 04:02:30] ppdet.utils.checkpoint INFO: Finish loading model weights: /root/.cache/paddle/weights/LCNet_x1_0_pretrained.pdparams
+[08/15 04:02:46] ppdet.engine INFO: Epoch: [0] [   0/1929] learning_rate: 0.040000 loss_vfl: 1.216707 loss_bbox: 1.142163 loss_dfl: 0.544196 loss: 2.903065 eta: 17 days, 13:50:26 batch_cost: 15.7452 data_cost: 2.9112 ips: 1.5243 images/s
+[08/15 04:03:19] ppdet.engine INFO: Epoch: [0] [  20/1929] learning_rate: 0.064000 loss_vfl: 1.180627 loss_bbox: 0.939552 loss_dfl: 0.442436 loss: 2.628206 eta: 2 days, 12:18:53 batch_cost: 1.5770 data_cost: 0.0008 ips: 15.2184 images/s
+[08/15 04:03:47] ppdet.engine INFO: Epoch: [0] [  40/1929] learning_rate: 0.088000 loss_vfl: 0.543321 loss_bbox: 1.071401 loss_dfl: 0.457817 loss: 2.057003 eta: 2 days, 0:07:03 batch_cost: 1.3190 data_cost: 0.0007 ips: 18.1954 images/s
+[08/15 04:04:12] ppdet.engine INFO: Epoch: [0] [  60/1929] learning_rate: 0.112000 loss_vfl: 0.630989 loss_bbox: 0.859183 loss_dfl: 0.384702 loss: 1.883143 eta: 1 day, 19:01:29 batch_cost: 1.2177 data_cost: 0.0006 ips: 19.7087 images/s
+```
+
+- `--eval` indicates that the best model is saved as `output/picodet_lcnet_x1_0_layout/best_accuracy`  by default during the evaluation process 。
+
+**Note that the configuration file for prediction / evaluation must be consistent with the training.**
+
+### 5.2. FGD Distillation Training
+
+PaddleDetection supports FGD-based [Focal and Global Knowledge Distillation for Detectors]( https://arxiv.org/abs/2111.11837v1)  The training process of the target detection model of distillation, FGD distillation is divided into two parts `Focal` and `Global`.     `Focal` Distillation separates the foreground and background of the image, allowing the student model to focus on the key pixels of the foreground and background features of the teacher model respectively;`  Global`Distillation section reconstructs the relationships between different pixels and transfers them from the teacher to the student to compensate for the global information lost in `Focal`Distillation.
+
+Change the dataset and modify the data configuration and number of categories in the [TODO] configuration, referring to 4.1. Start training:
+
+```bash
+# Single Card Training
+export CUDA_VISIBLE_DEVICES=0
+python3 tools/train.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \
+    --eval
+```
+
+- `-c`: Specify the model configuration file.
+- `--slim_config`:  Specify the compression policy profile.
+
+## 6. Model evaluation and prediction
+
+### 6.1. Indicator evaluation
+
+ Model parameters in training are saved by default in `output/picodet_ Lcnet_ X1_ 0_ Under the layout` directory. When evaluating indicators, you need to set `weights` to point to the saved parameter file.Assessment datasets can be accessed via `configs/picodet/legacy_ Model/application/layout_ Analysis/picodet_ Lcnet_ X1_ 0_ Layout. Yml` . Modify `EvalDataset`  : `img_dir`,`anno_ Path`and`dataset_dir` setting.
+
+```bash
+# GPU evaluation, weights as weights to be measured
+python3 tools/eval.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    -o weights=./output/picodet_lcnet_x1_0_layout/best_model
+```
+
+The following information will be printed out, such as mAP, AP0.5, etc.
+
+```py
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.935
+ Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.979
+ Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.956
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.404
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.782
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.969
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.539
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.938
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.949
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.495
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.818
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.978
+[08/15 07:07:09] ppdet.engine INFO: Total sample number: 11245, averge FPS: 24.405059207157436
+[08/15 07:07:09] ppdet.engine INFO: Best test bbox ap is 0.935.
+```
+
+If you use the provided pre-training model for evaluation or the FGD distillation training model, replace the `weights` model path and execute the following command for evaluation:
+
+```
+python3 tools/eval.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \
+    -o weights=output/picodet_lcnet_x2_5_layout/best_model
+```
+
+- `-c`: Specify the model configuration file.
+- `--slim_config`:  Specify the distillation policy profile.
+- `-o weights`: Specify the model path trained by the distillation algorithm.
+
+### 6.2. Test Layout Analysis Results
+
+
+The profile predicted to be used must be consistent with the training, for example, if you pass `python3 tools/train'. Py-c configs/picodet/legacy_ Model/application/layout_ Analysis/picodet_ Lcnet_ X1_ 0_ Layout. Yml` completed the training process for the model.
+
+With  trained PaddleDetection model, you can use the following commands to make model predictions.
+
+```bash
+python3 tools/infer.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    -o weights='output/picodet_lcnet_x1_0_layout/best_model.pdparams' \
+    --infer_img='docs/images/layout.jpg' \
+    --output_dir=output_dir/ \
+    --draw_threshold=0.5
+```
+
+- `--infer_img`:  Reasoning for a single picture can also be done via `--infer_ Dir`Inform all pictures in the file.
+- `--output_dir`:  Specify the path to save the visualization results.
+- `--draw_threshold`:Specify the NMS threshold for drawing the result box.
+
+If you use the provided pre-training model for prediction or the FGD distillation training model, change the `weights` model path and execute the following command to make the prediction:
+
+```
+python3 tools/infer.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \
+    -o weights='output/picodet_lcnet_x2_5_layout/best_model.pdparams' \
+    --infer_img='docs/images/layout.jpg' \
+    --output_dir=output_dir/ \
+    --draw_threshold=0.5
+```
+
+
+## 7. Model Export and Inference
+
+
+### 7.1 Model Export
+
+The inference model (the model saved by `paddle.jit.save`) is generally a solidified model saved after the model training is completed, and is mostly used to give prediction in deployment.
+
+The model saved during the training process is the checkpoints model, which saves the parameters of the model and is mostly used to resume training.
+
+Compared with the checkpoints model, the inference model will additionally save the structural information of the model. Therefore, it is easier to deploy because the model structure and model parameters are already solidified in the inference model file, and is suitable for integration with actual systems.
+
+Layout analysis model to inference model steps are as follows：
+
+```bash
+python3 tools/export_model.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    -o weights=output/picodet_lcnet_x1_0_layout/best_model \
+    --output_dir=output_inference/
+```
+
+* If no post-export processing is required, specify：`-o export.benchmark=True`（If -o already exists, delete -o here）
+* If you do not need to export NMS, specify：`-o export.nms=False`
+
+After successful conversion, there are three files in the directory:
+
+```
+output_inference/picodet_lcnet_x1_0_layout/
+    ├── model.pdiparams         # inference Parameter file for model
+    ├── model.pdiparams.info    # inference Model parameter information, ignorable
+    └── model.pdmodel           # inference Model Structure File for Model
+```
+
+If you change the `weights` model path using the provided pre-training model to the Inference model, or using the FGD distillation training model, the model to inference model steps are as follows:
+
+```bash
+python3 tools/export_model.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \
+    -o weights=./output/picodet_lcnet_x2_5_layout/best_model \
+    --output_dir=output_inference/
+```
+
+### 7.2 Model inference
+
+Replace model_with the provided inference training model for inference or the FGD distillation training `model_dir`Inference model path, execute the following commands for inference:
+
+```bash
+python3 deploy/python/infer.py \
+    --model_dir=output_inference/picodet_lcnet_x1_0_layout/ \
+    --image_file=docs/images/layout.jpg \
+    --device=CPU
+```
+
+- --device：Specify the GPU or CPU device
+
+When model inference is complete, you will see the following log output:
+
+```
+------------------------------------------
+-----------  Model Configuration -----------
+Model Arch: PicoDet
+Transform Order:
+--transform op: Resize
+--transform op: NormalizeImage
+--transform op: Permute
+--transform op: PadStride
+--------------------------------------------
+class_id:0, confidence:0.9921, left_top:[20.18,35.66],right_bottom:[341.58,600.99]
+class_id:0, confidence:0.9914, left_top:[19.77,611.42],right_bottom:[341.48,901.82]
+class_id:0, confidence:0.9904, left_top:[369.36,375.10],right_bottom:[691.29,600.59]
+class_id:0, confidence:0.9835, left_top:[369.60,608.60],right_bottom:[691.38,736.72]
+class_id:0, confidence:0.9830, left_top:[369.58,805.38],right_bottom:[690.97,901.80]
+class_id:0, confidence:0.9716, left_top:[383.68,271.44],right_bottom:[688.93,335.39]
+class_id:0, confidence:0.9452, left_top:[370.82,34.48],right_bottom:[688.10,63.54]
+class_id:1, confidence:0.8712, left_top:[370.84,771.03],right_bottom:[519.30,789.13]
+class_id:3, confidence:0.9856, left_top:[371.28,67.85],right_bottom:[685.73,267.72]
+save result to: output/layout.jpg
+Test iter 0
+------------------ Inference Time Info ----------------------
+total_time(ms): 2196.0, img_num: 1
+average latency time(ms): 2196.00, QPS: 0.455373
+preprocess_time(ms): 2172.50, inference_time(ms): 11.90, postprocess_time(ms): 11.60
+```
+
+- Model：model structure
+- Transform Order：Preprocessing operation
+- class_id, confidence, left_top, right_bottom：Indicates category id, confidence level, upper left coordinate, lower right coordinate, respectively
+- save result to：Save path of visual layout analysis results, default save to ./output folder
+- inference time info：Inference time, where preprocess_time represents the preprocessing time, Inference_time represents the model prediction time, and postprocess_time represents the post-processing time
+
+The result of visualization layout is shown in the following figure
+
+<div align="center">
+    <img src="../docs/layout/layout_res.jpg" width="800">
+</div>
+
+
+
+## Citations
+
+```
+@inproceedings{zhong2019publaynet,
+  title={PubLayNet: largest dataset ever for document layout analysis},
+  author={Zhong, Xu and Tang, Jianbin and Yepes, Antonio Jimeno},
+  booktitle={2019 International Conference on Document Analysis and Recognition (ICDAR)},
+  year={2019},
+  volume={},
+  number={},
+  pages={1015-1022},
+  doi={10.1109/ICDAR.2019.00166},
+  ISSN={1520-5363},
+  month={Sep.},
+  organization={IEEE}
+}
+
+@inproceedings{yang2022focal,
+  title={Focal and global knowledge distillation for detectors},
+  author={Yang, Zhendong and Li, Zhe and Jiang, Xiaohu and Gong, Yuan and Yuan, Zehuan and Zhao, Danpei and Yuan, Chun},
+  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
+  pages={4643--4652},
+  year={2022}
+}
+```
diff --git a/ppstructure/layout/README_ch.md b/ppstructure/layout/README_ch.md
new file mode 100644
index 0000000..adef46d
--- /dev/null
+++ b/ppstructure/layout/README_ch.md
@@ -0,0 +1,469 @@
+简体中文 | [English](README.md)
+
+# 版面分析
+
+- [1. 简介](#1-简介)
+- [2. 快速开始](#2-快速开始)
+- [3. 安装](#3-安装)
+  - [3.1 安装PaddlePaddle](#31-安装paddlepaddle)
+  - [3.2 安装PaddleDetection](#32-安装paddledetection)
+- [4. 数据准备](#4-数据准备)
+  - [4.1 英文数据集](#41-英文数据集)
+  - [4.2 更多数据集](#42-更多数据集)
+- [5. 开始训练](#5-开始训练)
+  - [5.1 启动训练](#51-启动训练)
+  - [5.2 FGD蒸馏训练](#52-fgd蒸馏训练)
+- [6. 模型评估与预测](#6-模型评估与预测)
+  - [6.1 指标评估](#61-指标评估)
+  - [6.2 测试版面分析结果](#62-测试版面分析结果)
+- [7 模型导出与预测](#7-模型导出与预测)
+  - [7.1 模型导出](#71-模型导出)
+  - [7.2 模型推理](#72-模型推理)
+
+## 1. 简介
+
+版面分析指的是对图片形式的文档进行区域划分，定位其中的关键区域，如文字、标题、表格、图片等。版面分析算法基于[PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection)的轻量模型PP-PicoDet进行开发，包含英文、中文、表格版面分析3类模型。其中，英文模型支持Text、Title、Tale、Figure、List5类区域的检测，中文模型支持Text、Title、Figure、Figure caption、Table、Table caption、Header、Footer、Reference、Equation10类区域的检测，表格版面分析支持Table区域的检测，版面分析效果如下图所示：
+
+<div align="center">
+    <img src="../docs/layout/layout.png" width="800">
+</div>
+
+## 2. 快速开始
+
+PP-Structure目前提供了中文、英文、表格三类文档版面分析模型，模型链接见 [models_list](../docs/models_list.md#1-版面分析模型)。也提供了whl包的形式方便快速使用，详见 [quickstart](../docs/quickstart.md)。
+
+
+## 3. 安装
+
+### 3.1. 安装PaddlePaddle
+
+- **（1) 安装PaddlePaddle**
+
+```bash
+python3 -m pip install --upgrade pip
+
+# GPU安装
+python3 -m pip install "paddlepaddle-gpu>=2.3" -i https://mirror.baidu.com/pypi/simple
+
+# CPU安装
+python3 -m pip install "paddlepaddle>=2.3" -i https://mirror.baidu.com/pypi/simple
+```
+更多需求，请参照[安装文档](https://www.paddlepaddle.org.cn/install/quick)中的说明进行操作。
+
+### 3.2. 安装PaddleDetection
+
+- **（1）下载PaddleDetection源码**
+
+```bash
+git clone https://github.com/PaddlePaddle/PaddleDetection.git
+```
+
+- **（2）安装其他依赖**
+
+```bash
+cd PaddleDetection
+python3 -m pip install -r requirements.txt
+```
+
+## 4. 数据准备
+
+如果希望直接体验预测过程，可以跳过数据准备，下载我们提供的预训练模型。
+
+### 4.1. 英文数据集
+
+下载文档分析数据集[PubLayNet](https://developer.ibm.com/exchanges/data/all/publaynet/)（数据集96G），包含5个类：`{0: "Text", 1: "Title", 2: "List", 3:"Table", 4:"Figure"}`
+
+```
+# 下载数据
+wget https://dax-cdn.cdn.appdomain.cloud/dax-publaynet/1.0.0/publaynet.tar.gz
+# 解压数据
+tar -xvf publaynet.tar.gz
+```
+
+解压之后的**目录结构：**
+
+```
+|-publaynet
+  |- test
+     |- PMC1277013_00004.jpg
+     |- PMC1291385_00002.jpg
+     | ...
+  |- train.json
+  |- train
+     |- PMC1291385_00002.jpg
+     |- PMC1277013_00004.jpg
+     | ...
+  |- val.json
+  |- val
+     |- PMC538274_00004.jpg
+     |- PMC539300_00004.jpg
+     | ...
+```
+
+**数据分布：**
+
+| File or Folder | Description    | num     |
+| :------------- | :------------- | ------- |
+| `train/`       | 训练集图片     | 335,703 |
+| `val/`         | 验证集图片     | 11,245  |
+| `test/`        | 测试集图片     | 11,405  |
+| `train.json`   | 训练集标注文件 | -       |
+| `val.json`     | 验证集标注文件 | -       |
+
+**标注格式：**
+
+json文件包含所有图像的标注，数据以字典嵌套的方式存放，包含以下key：
+
+- info，表示标注文件info。
+
+- licenses，表示标注文件licenses。
+
+- images，表示标注文件中图像信息列表，每个元素是一张图像的信息。如下为其中一张图像的信息：
+
+  ```
+  {
+      'file_name': 'PMC4055390_00006.jpg',    # file_name
+      'height': 601,                      # image height
+      'width': 792,                       # image width
+      'id': 341427                        # image id
+  }
+  ```
+
+- annotations，表示标注文件中目标物体的标注信息列表，每个元素是一个目标物体的标注信息。如下为其中一个目标物体的标注信息：
+
+  ```
+  {
+
+      'segmentation':             # 物体的分割标注
+      'area': 60518.099043117836, # 物体的区域面积
+      'iscrowd': 0,               # iscrowd
+      'image_id': 341427,         # image id
+      'bbox': [50.58, 490.86, 240.15, 252.16], # bbox [x1,y1,w,h]
+      'category_id': 1,           # category_id
+      'id': 3322348               # image id
+  }
+  ```
+
+### 4.2. 更多数据集
+
+我们提供了CDLA(中文版面分析)、TableBank(表格版面分析)等数据集的下连接，处理为上述标注文件json格式，即可以按相同方式进行训练。
+
+| dataset                                                      | 简介                                                         |
+| ------------------------------------------------------------ | ------------------------------------------------------------ |
+| [cTDaR2019_cTDaR](https://cndplab-founder.github.io/cTDaR2019/) | 用于表格检测(TRACKA)和表格识别(TRACKB)。图片类型包含历史数据集(以cTDaR_t0开头，如cTDaR_t00872.jpg)和现代数据集(以cTDaR_t1开头，cTDaR_t10482.jpg)。 |
+| [IIIT-AR-13K](http://cvit.iiit.ac.in/usodi/iiitar13k.php)    | 手动注释公开的年度报告中的图形或页面而构建的数据集，包含5类：table, figure, natural image, logo, and signature |
+| [CDLA](https://github.com/buptlihang/CDLA)                   | 中文文档版面分析数据集，面向中文文献类（论文）场景，包含10类：Text、Title、Figure、Figure caption、Table、Table caption、Header、Footer、Reference、Equation |
+| [TableBank](https://github.com/doc-analysis/TableBank)       | 用于表格检测和识别大型数据集，包含Word和Latex2种文档格式     |
+| [DocBank](https://github.com/doc-analysis/DocBank)           | 使用弱监督方法构建的大规模数据集(500K文档页面)，用于文档布局分析，包含12类：Author、Caption、Date、Equation、Figure、Footer、List、Paragraph、Reference、Section、Table、Title |
+
+
+## 5. 开始训练
+
+提供了训练脚本、评估脚本和预测脚本，本节将以PubLayNet预训练模型为例进行讲解。
+
+如果不希望训练，直接体验后面的模型评估、预测、动转静、推理的流程，可以下载提供的预训练模型(PubLayNet数据集)，并跳过5.1和5.2。
+
+```
+mkdir pretrained_model
+cd pretrained_model
+# 下载PubLayNet预训练模型（直接体验模型评估、预测、动转静）
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout.pdparams
+# 下载PubLaynet推理模型（直接体验模型推理）
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar
+```
+
+如果测试图片为中文，可以下载中文CDLA数据集的预训练模型，识别10类文档区域：Table、Figure、Figure caption、Table、Table caption、Header、Footer、Reference、Equation，在[版面分析模型](../docs/models_list.md)中下载`picodet_lcnet_x1_0_fgd_layout_cdla`模型的训练模型和推理模型。如果只检测图片中的表格区域，可以下载表格数据集的预训练模型，在[版面分析模型](../docs/models_list.md)中下载`picodet_lcnet_x1_0_fgd_layout_table`模型的训练模型和推理模型。
+
+### 5.1. 启动训练
+
+使用PaddleDetection[版面分析配置文件](https://github.com/PaddlePaddle/PaddleDetection/tree/release/2.5/configs/picodet/legacy_model/application/layout_analysis)启动训练
+
+* 修改配置文件
+
+如果你希望训练自己的数据集，需要修改配置文件中的数据配置、类别数。
+
+
+以`configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml` 为例，修改的内容如下所示。
+
+```yaml
+metric: COCO
+# 类别数
+num_classes: 5
+
+TrainDataset:
+  !COCODataSet
+    # 修改为你自己的训练数据目录
+    image_dir: train
+    # 修改为你自己的训练数据标签文件
+    anno_path: train.json
+    # 修改为你自己的训练数据根目录
+    dataset_dir: /root/publaynet/
+    data_fields: ['image', 'gt_bbox', 'gt_class', 'is_crowd']
+
+EvalDataset:
+  !COCODataSet
+    # 修改为你自己的验证数据目录
+    image_dir: val
+    # 修改为你自己的验证数据标签文件
+    anno_path: val.json
+    # 修改为你自己的验证数据根目录
+    dataset_dir: /root/publaynet/
+
+TestDataset:
+  !ImageFolder
+    # 修改为你自己的测试数据标签文件
+    anno_path: /root/publaynet/val.json
+```
+
+* 开始训练，在训练时，会默认下载PP-PicoDet预训练模型，这里无需预先下载。
+
+```bash
+# GPU训练 支持单卡，多卡训练
+# 训练日志会自动保存到 log 目录中
+
+# 单卡训练
+export CUDA_VISIBLE_DEVICES=0
+python3 tools/train.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --eval
+
+# 多卡训练，通过--gpus参数指定卡号
+export CUDA_VISIBLE_DEVICES=0,1,2,3
+python3 -m paddle.distributed.launch --gpus '0,1,2,3'  tools/train.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --eval
+```
+
+**注意：**如果训练时显存out memory，将TrainReader中batch_size调小，同时LearningRate中base_lr等比例减小。发布的config均由8卡训练得到，如果改变GPU卡数为1，那么base_lr需要减小8倍。
+
+正常启动训练后，会看到以下log输出：
+
+```
+[08/15 04:02:30] ppdet.utils.checkpoint INFO: Finish loading model weights: /root/.cache/paddle/weights/LCNet_x1_0_pretrained.pdparams
+[08/15 04:02:46] ppdet.engine INFO: Epoch: [0] [   0/1929] learning_rate: 0.040000 loss_vfl: 1.216707 loss_bbox: 1.142163 loss_dfl: 0.544196 loss: 2.903065 eta: 17 days, 13:50:26 batch_cost: 15.7452 data_cost: 2.9112 ips: 1.5243 images/s
+[08/15 04:03:19] ppdet.engine INFO: Epoch: [0] [  20/1929] learning_rate: 0.064000 loss_vfl: 1.180627 loss_bbox: 0.939552 loss_dfl: 0.442436 loss: 2.628206 eta: 2 days, 12:18:53 batch_cost: 1.5770 data_cost: 0.0008 ips: 15.2184 images/s
+[08/15 04:03:47] ppdet.engine INFO: Epoch: [0] [  40/1929] learning_rate: 0.088000 loss_vfl: 0.543321 loss_bbox: 1.071401 loss_dfl: 0.457817 loss: 2.057003 eta: 2 days, 0:07:03 batch_cost: 1.3190 data_cost: 0.0007 ips: 18.1954 images/s
+[08/15 04:04:12] ppdet.engine INFO: Epoch: [0] [  60/1929] learning_rate: 0.112000 loss_vfl: 0.630989 loss_bbox: 0.859183 loss_dfl: 0.384702 loss: 1.883143 eta: 1 day, 19:01:29 batch_cost: 1.2177 data_cost: 0.0006 ips: 19.7087 images/s
+```
+
+- `--eval`表示训练的同时，进行评估， 评估过程中默认将最佳模型，保存为 `output/picodet_lcnet_x1_0_layout/best_accuracy` 。
+
+**注意，预测/评估时的配置文件请务必与训练一致。**
+
+### 5.2. FGD蒸馏训练
+
+PaddleDetection支持了基于FGD([Focal and Global Knowledge Distillation for Detectors](https://arxiv.org/abs/2111.11837v1))蒸馏的目标检测模型训练过程，FGD蒸馏分为两个部分`Focal`和`Global`。`Focal`蒸馏分离图像的前景和背景，让学生模型分别关注教师模型的前景和背景部分特征的关键像素；`Global`蒸馏部分重建不同像素之间的关系并将其从教师转移到学生，以补偿`Focal`蒸馏中丢失的全局信息。
+
+更换数据集，修改【TODO】配置中的数据配置、类别数，具体可以参考4.1。启动训练：
+
+```bash
+# 单卡训练
+export CUDA_VISIBLE_DEVICES=0
+python3 tools/train.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \
+    --eval
+```
+
+- `-c`: 指定模型配置文件。
+- `--slim_config`: 指定压缩策略配置文件。
+
+## 6. 模型评估与预测
+
+### 6.1. 指标评估
+
+训练中模型参数默认保存在`output/picodet_lcnet_x1_0_layout`目录下。在评估指标时，需要设置`weights`指向保存的参数文件。评估数据集可以通过 `configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml`  修改`EvalDataset`中的 `image_dir`、`anno_path`和`dataset_dir` 设置。
+
+```bash
+# GPU 评估， weights 为待测权重
+python3 tools/eval.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    -o weights=./output/picodet_lcnet_x1_0_layout/best_model
+```
+
+会输出以下信息，打印出mAP、AP0.5等信息。
+
+```py
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.935
+ Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.979
+ Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.956
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.404
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.782
+ Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.969
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.539
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.938
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.949
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.495
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.818
+ Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.978
+[08/15 07:07:09] ppdet.engine INFO: Total sample number: 11245, averge FPS: 24.405059207157436
+[08/15 07:07:09] ppdet.engine INFO: Best test bbox ap is 0.935.
+```
+
+若使用**提供的预训练模型进行评估**，或使用**FGD蒸馏训练的模型**，更换`weights`模型路径，执行如下命令进行评估：
+
+```
+python3 tools/eval.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \
+    -o weights=output/picodet_lcnet_x2_5_layout/best_model
+```
+
+- `-c`: 指定模型配置文件。
+- `--slim_config`: 指定蒸馏策略配置文件。
+- `-o weights`: 指定蒸馏算法训好的模型路径。
+
+### 6.2 测试版面分析结果
+
+
+预测使用的配置文件必须与训练一致，如您通过 `python3 tools/train.py -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml` 完成了模型的训练过程。
+
+使用 PaddleDetection 训练好的模型，您可以使用如下命令进行模型预测。
+
+```bash
+python3 tools/infer.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    -o weights='output/picodet_lcnet_x1_0_layout/best_model.pdparams' \
+    --infer_img='docs/images/layout.jpg' \
+    --output_dir=output_dir/ \
+    --draw_threshold=0.5
+```
+
+- `--infer_img`: 推理单张图片，也可以通过`--infer_dir`推理文件中的所有图片。
+- `--output_dir`: 指定可视化结果保存路径。
+- `--draw_threshold`:指定绘制结果框的NMS阈值。
+
+若使用**提供的预训练模型进行预测**，或使用**FGD蒸馏训练的模型**，更换`weights`模型路径，执行如下命令进行预测：
+
+```
+python3 tools/infer.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \
+    -o weights='output/picodet_lcnet_x2_5_layout/best_model.pdparams' \
+    --infer_img='docs/images/layout.jpg' \
+    --output_dir=output_dir/ \
+    --draw_threshold=0.5
+```
+
+
+## 7. 模型导出与预测
+
+
+### 7.1 模型导出
+
+inference 模型（`paddle.jit.save`保存的模型） 一般是模型训练，把模型结构和模型参数保存在文件中的固化模型，多用于预测部署场景。 训练过程中保存的模型是checkpoints模型，保存的只有模型的参数，多用于恢复训练等。 与checkpoints模型相比，inference 模型会额外保存模型的结构信息，在预测部署、加速推理上性能优越，灵活方便，适合于实际系统集成。
+
+版面分析模型转inference模型步骤如下：
+
+```bash
+python3 tools/export_model.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    -o weights=output/picodet_lcnet_x1_0_layout/best_model \
+    --output_dir=output_inference/
+```
+
+* 如无需导出后处理，请指定：`-o export.benchmark=True`（如果-o已出现过，此处删掉-o）
+* 如无需导出NMS，请指定：`-o export.nms=False`
+
+转换成功后，在目录下有三个文件：
+
+```
+output_inference/picodet_lcnet_x1_0_layout/
+    ├── model.pdiparams         # inference模型的参数文件
+    ├── model.pdiparams.info    # inference模型的参数信息，可忽略
+    └── model.pdmodel           # inference模型的模型结构文件
+```
+
+若使用**提供的预训练模型转Inference模型**，或使用**FGD蒸馏训练的模型**，更换`weights`模型路径，模型转inference模型步骤如下：
+
+```bash
+python3 tools/export_model.py \
+    -c configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x1_0_layout.yml \
+    --slim_config configs/picodet/legacy_model/application/layout_analysis/picodet_lcnet_x2_5_layout.yml \
+    -o weights=./output/picodet_lcnet_x2_5_layout/best_model \
+    --output_dir=output_inference/
+```
+
+
+
+### 7.2 模型推理
+
+若使用**提供的推理训练模型推理**，或使用**FGD蒸馏训练的模型**，更换`model_dir`推理模型路径，执行如下命令进行推理：
+
+```bash
+python3 deploy/python/infer.py \
+    --model_dir=output_inference/picodet_lcnet_x1_0_layout/ \
+    --image_file=docs/images/layout.jpg \
+    --device=CPU
+```
+
+- --device：指定GPU、CPU设备
+
+模型推理完成，会看到以下log输出
+
+```
+------------------------------------------
+-----------  Model Configuration -----------
+Model Arch: PicoDet
+Transform Order:
+--transform op: Resize
+--transform op: NormalizeImage
+--transform op: Permute
+--transform op: PadStride
+--------------------------------------------
+class_id:0, confidence:0.9921, left_top:[20.18,35.66],right_bottom:[341.58,600.99]
+class_id:0, confidence:0.9914, left_top:[19.77,611.42],right_bottom:[341.48,901.82]
+class_id:0, confidence:0.9904, left_top:[369.36,375.10],right_bottom:[691.29,600.59]
+class_id:0, confidence:0.9835, left_top:[369.60,608.60],right_bottom:[691.38,736.72]
+class_id:0, confidence:0.9830, left_top:[369.58,805.38],right_bottom:[690.97,901.80]
+class_id:0, confidence:0.9716, left_top:[383.68,271.44],right_bottom:[688.93,335.39]
+class_id:0, confidence:0.9452, left_top:[370.82,34.48],right_bottom:[688.10,63.54]
+class_id:1, confidence:0.8712, left_top:[370.84,771.03],right_bottom:[519.30,789.13]
+class_id:3, confidence:0.9856, left_top:[371.28,67.85],right_bottom:[685.73,267.72]
+save result to: output/layout.jpg
+Test iter 0
+------------------ Inference Time Info ----------------------
+total_time(ms): 2196.0, img_num: 1
+average latency time(ms): 2196.00, QPS: 0.455373
+preprocess_time(ms): 2172.50, inference_time(ms): 11.90, postprocess_time(ms): 11.60
+```
+
+- Model：模型结构
+- Transform Order：预处理操作
+- class_id、confidence、left_top、right_bottom：分别表示类别id、置信度、左上角坐标、右下角坐标
+- save result to：可视化版面分析结果保存路径，默认保存到`./output`文件夹
+- Inference Time Info：推理时间，其中preprocess_time表示预处理耗时，inference_time表示模型预测耗时，postprocess_time表示后处理耗时
+
+可视化版面结果如下图所示
+
+<div align="center">
+    <img src="../docs/layout/layout_res.jpg" width="800">
+</div>
+
+
+
+## Citations
+
+```
+@inproceedings{zhong2019publaynet,
+  title={PubLayNet: largest dataset ever for document layout analysis},
+  author={Zhong, Xu and Tang, Jianbin and Yepes, Antonio Jimeno},
+  booktitle={2019 International Conference on Document Analysis and Recognition (ICDAR)},
+  year={2019},
+  volume={},
+  number={},
+  pages={1015-1022},
+  doi={10.1109/ICDAR.2019.00166},
+  ISSN={1520-5363},
+  month={Sep.},
+  organization={IEEE}
+}
+
+@inproceedings{yang2022focal,
+  title={Focal and global knowledge distillation for detectors},
+  author={Yang, Zhendong and Li, Zhe and Jiang, Xiaohu and Gong, Yuan and Yuan, Zehuan and Zhao, Danpei and Yuan, Chun},
+  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
+  pages={4643--4652},
+  year={2022}
+}
+```
diff --git a/ppstructure/layout/__init__.py b/ppstructure/layout/__init__.py
new file mode 100644
index 0000000..1d11e26
--- /dev/null
+++ b/ppstructure/layout/__init__.py
@@ -0,0 +1,13 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/ppstructure/layout/predict_layout.py b/ppstructure/layout/predict_layout.py
new file mode 100755
index 0000000..10bb117
--- /dev/null
+++ b/ppstructure/layout/predict_layout.py
@@ -0,0 +1,143 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "../..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+
+import cv2
+import numpy as np
+import time
+
+import tools.infer.utility as utility
+from ppocr.data import create_operators, transform
+from ppocr.postprocess import build_post_process
+from ppocr.utils.logging import get_logger
+from ppocr.utils.utility import get_image_file_list, check_and_read
+from ppstructure.utility import parse_args
+from picodet_postprocess import PicoDetPostProcess
+
+logger = get_logger()
+
+
+class LayoutPredictor(object):
+    def __init__(self, args):
+        pre_process_list = [
+            {"Resize": {"size": [800, 608]}},
+            {
+                "NormalizeImage": {
+                    "std": [0.229, 0.224, 0.225],
+                    "mean": [0.485, 0.456, 0.406],
+                    "scale": "1./255.",
+                    "order": "hwc",
+                }
+            },
+            {"ToCHWImage": None},
+            {"KeepKeys": {"keep_keys": ["image"]}},
+        ]
+        postprocess_params = {
+            "name": "PicoDetPostProcess",
+            "layout_dict_path": args.layout_dict_path,
+            "score_threshold": args.layout_score_threshold,
+            "nms_threshold": args.layout_nms_threshold,
+        }
+
+        self.preprocess_op = create_operators(pre_process_list)
+        self.postprocess_op = build_post_process(postprocess_params)
+        (
+            self.predictor,
+            self.input_tensor,
+            self.output_tensors,
+            self.config,
+        ) = utility.create_predictor(args, "layout", logger)
+        self.use_onnx = args.use_onnx
+
+    def __call__(self, img):
+        ori_im = img.copy()
+        data = {"image": img}
+        data = transform(data, self.preprocess_op)
+        img = data[0]
+
+        if img is None:
+            return None, 0
+
+        img = np.expand_dims(img, axis=0)
+        img = img.copy()
+
+        preds, elapse = 0, 1
+        starttime = time.time()
+
+        np_score_list, np_boxes_list = [], []
+        if self.use_onnx:
+            input_dict = {}
+            input_dict[self.input_tensor.name] = img
+            outputs = self.predictor.run(self.output_tensors, input_dict)
+            num_outs = int(len(outputs) / 2)
+            for out_idx in range(num_outs):
+                np_score_list.append(outputs[out_idx])
+                np_boxes_list.append(outputs[out_idx + num_outs])
+        else:
+            self.input_tensor.copy_from_cpu(img)
+            self.predictor.run()
+            output_names = self.predictor.get_output_names()
+            num_outs = int(len(output_names) / 2)
+            for out_idx in range(num_outs):
+                np_score_list.append(
+                    self.predictor.get_output_handle(
+                        output_names[out_idx]
+                    ).copy_to_cpu()
+                )
+                np_boxes_list.append(
+                    self.predictor.get_output_handle(
+                        output_names[out_idx + num_outs]
+                    ).copy_to_cpu()
+                )
+        preds = dict(boxes=np_score_list, boxes_num=np_boxes_list)
+
+        post_preds = self.postprocess_op(ori_im, img, preds)
+        elapse = time.time() - starttime
+        return post_preds, elapse
+
+
+def main(args):
+    image_file_list = get_image_file_list(args.image_dir)
+    layout_predictor = LayoutPredictor(args)
+    count = 0
+    total_time = 0
+
+    repeats = 50
+    for image_file in image_file_list:
+        img, flag, _ = check_and_read(image_file)
+        if not flag:
+            img = cv2.imread(image_file)
+        if img is None:
+            logger.info("error in loading image:{}".format(image_file))
+            continue
+
+        layout_res, elapse = layout_predictor(img)
+
+        logger.info("result: {}".format(layout_res))
+
+        if count > 0:
+            total_time += elapse
+        count += 1
+        logger.info("Predict time of {}: {}".format(image_file, elapse))
+
+
+if __name__ == "__main__":
+    main(parse_args())
diff --git a/ppstructure/pdf2word/README.md b/ppstructure/pdf2word/README.md
new file mode 100644
index 0000000..783555a
--- /dev/null
+++ b/ppstructure/pdf2word/README.md
@@ -0,0 +1,49 @@
+# PDF2WORD
+
+PDF2Word是PaddleOCR社区开发者 [whjdark](https://github.com/whjdark) 基于PP-StructureV2版面分析与恢复模型实现的PDF转换Word应用程序，提供可直接安装的exe应用程序，**方便Windows用户免环境配置运行**
+
+## 1.使用
+
+### 应用程序
+
+1. 下载与安装：针对Windows用户，根据[软件下载]()一节下载软件后，运行 `pdf2word.exe` 。若您下载的是lite版本，安装过程中会在线下载环境依赖、模型等必要资源，安装时间较长，请确保网络畅通。serve版本打包了相关依赖，安装时间较短，可按需下载。
+
+2. 转换：由于PP-Structure根据中英文数据分别进行适配，在转换相应文件时可**根据文档语言进行相应选择**。
+
+### 脚本运行
+
+3. 打开结果：点击`显示结果`，即可打开转换完成后的文件夹
+
+> 注意：
+>
+> - 初次安装程序根据不同设备需要等待1-2分钟不等
+> - 使用Office与WPS打开的Word结果会出现不同，推荐以Office为准
+> - 本程序使用 [QPT](https://github.com/QPT-Family/QPT) 进行应用程序打包，感谢 [GT-ZhangAcer](https://github.com/GT-ZhangAcer) 对打包过程的支持
+> - 应用程序仅支持正版win10，11系统，不支持盗版Windows系统，若在安装过程中出现报错或缺少依赖，推荐直接使用 `paddleocr` whl包应用PDF2Word功能，详情可查看[链接](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/ppstructure/docs/quickstart.md)
+
+### 脚本启动界面
+
+首次运行需要将切换路径到PaddleOCR文件目录 ，然后运行代码
+
+```
+cd ./ppstructure/pdf2word
+python pdf2word.py
+```
+
+### PaddleOCR whl包
+
+针对Linux、Mac用户或已经拥有Python环境的用户，**推荐安装 `paddleocr` whl包直接应用PDF2Word功能**，详情可查看[链接](https://github.com/PaddlePaddle/PaddleOCR/blob/release/2.6/ppstructure/docs/quickstart.md)
+
+<a name="download"></a>
+
+## 2.软件下载
+
+如需获取已打包程序，可以扫描下方二维码，关注公众号填写问卷后，加入PaddleOCR官方交流群免费获取20G OCR学习大礼包，内含OCR场景应用集合（包含数码管、液晶屏、车牌、高精度SVTR模型等7个垂类模型）、《动手学OCR》电子书、课程回放视频、前沿论文等重磅资料
+
+<div align="center">
+<img src="https://user-images.githubusercontent.com/50011306/186369636-35f2008b-df5a-4784-b1f5-cebebcb2b7a5.jpg"  width = "150" height = "150" />
+</div>
+
+## 3.版本说明
+
+v0.2版：新加入PDF解析功能，仅提供full版本，打包了所有依赖包与模型文件，尽可能避免安装失败问题。若仍然安装失败，推荐使用 `paddleocr` whl包
diff --git a/ppstructure/pdf2word/icons/chinese.png b/ppstructure/pdf2word/icons/chinese.png
new file mode 100644
index 0000000..328e2ff
Binary files /dev/null and b/ppstructure/pdf2word/icons/chinese.png differ
diff --git a/ppstructure/pdf2word/icons/english.png b/ppstructure/pdf2word/icons/english.png
new file mode 100644
index 0000000..536c4a9
Binary files /dev/null and b/ppstructure/pdf2word/icons/english.png differ
diff --git a/ppstructure/pdf2word/icons/folder-open.png b/ppstructure/pdf2word/icons/folder-open.png
new file mode 100644
index 0000000..ab5f55f
Binary files /dev/null and b/ppstructure/pdf2word/icons/folder-open.png differ
diff --git a/ppstructure/pdf2word/icons/folder-plus.png b/ppstructure/pdf2word/icons/folder-plus.png
new file mode 100644
index 0000000..01ce6c1
Binary files /dev/null and b/ppstructure/pdf2word/icons/folder-plus.png differ
diff --git a/ppstructure/pdf2word/pdf2word.py b/ppstructure/pdf2word/pdf2word.py
new file mode 100644
index 0000000..668f04e
--- /dev/null
+++ b/ppstructure/pdf2word/pdf2word.py
@@ -0,0 +1,538 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import sys
+import tarfile
+import os
+import time
+import datetime
+import functools
+import cv2
+import platform
+import subprocess
+import numpy as np
+from paddle.utils import try_import
+
+fitz = try_import("fitz")
+from PIL import Image
+from qtpy.QtWidgets import (
+    QApplication,
+    QWidget,
+    QPushButton,
+    QProgressBar,
+    QGridLayout,
+    QMessageBox,
+    QLabel,
+    QFileDialog,
+    QCheckBox,
+)
+from qtpy.QtCore import Signal, QThread, QObject
+from qtpy.QtGui import QImage, QPixmap, QIcon
+
+file = os.path.dirname(os.path.abspath(__file__))
+root = os.path.abspath(os.path.join(file, "../../"))
+sys.path.append(file)
+sys.path.insert(0, root)
+
+from ppstructure.predict_system import StructureSystem, save_structure_res
+from ppstructure.utility import parse_args, draw_structure_result
+from ppocr.utils.network import download_with_progressbar
+from ppstructure.recovery.recovery_to_doc import sorted_layout_boxes, convert_info_docx
+
+# from ScreenShotWidget import ScreenShotWidget
+
+__APPNAME__ = "pdf2word"
+__VERSION__ = "0.2.2"
+
+URLs_EN = {
+    # 下载超英文轻量级PP-OCRv3模型的检测模型并解压
+    "en_PP-OCRv3_det_infer": "https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar",
+    # 下载英文轻量级PP-OCRv3模型的识别模型并解压
+    "en_PP-OCRv3_rec_infer": "https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_infer.tar",
+    # 下载超轻量级英文表格英文模型并解压
+    "en_ppstructure_mobile_v2.0_SLANet_infer": "https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/en_ppstructure_mobile_v2.0_SLANet_infer.tar",
+    # 英文版面分析模型
+    "picodet_lcnet_x1_0_fgd_layout_infer": "https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar",
+}
+DICT_EN = {
+    "rec_char_dict_path": "en_dict.txt",
+    "layout_dict_path": "layout_publaynet_dict.txt",
+}
+
+URLs_CN = {
+    # 下载超中文轻量级PP-OCRv3模型的检测模型并解压
+    "cn_PP-OCRv3_det_infer": "https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_infer.tar",
+    # 下载中文轻量级PP-OCRv3模型的识别模型并解压
+    "cn_PP-OCRv3_rec_infer": "https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_infer.tar",
+    # 下载超轻量级英文表格英文模型并解压
+    "cn_ppstructure_mobile_v2.0_SLANet_infer": "https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/en_ppstructure_mobile_v2.0_SLANet_infer.tar",
+    # 中文版面分析模型
+    "picodet_lcnet_x1_0_fgd_layout_cdla_infer": "https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_cdla_infer.tar",
+}
+DICT_CN = {
+    "rec_char_dict_path": "ppocr_keys_v1.txt",
+    "layout_dict_path": "layout_cdla_dict.txt",
+}
+
+
+def QImageToCvMat(incomingImage) -> np.array:
+    """
+    Converts a QImage into an opencv MAT format
+    """
+
+    incomingImage = incomingImage.convertToFormat(QImage.Format.Format_RGBA8888)
+
+    width = incomingImage.width()
+    height = incomingImage.height()
+
+    ptr = incomingImage.bits()
+    ptr.setsize(height * width * 4)
+    arr = np.frombuffer(ptr, np.uint8).reshape((height, width, 4))
+    return arr
+
+
+def readImage(image_file) -> list:
+    if os.path.basename(image_file)[-3:] == "pdf":
+        imgs = []
+        with fitz.open(image_file) as pdf:
+            for pg in range(0, pdf.pageCount):
+                page = pdf[pg]
+                mat = fitz.Matrix(2, 2)
+                pm = page.getPixmap(matrix=mat, alpha=False)
+
+                # if width or height > 2000 pixels, don't enlarge the image
+                if pm.width > 2000 or pm.height > 2000:
+                    pm = page.getPixmap(matrix=fitz.Matrix(1, 1), alpha=False)
+
+                img = Image.frombytes("RGB", [pm.width, pm.height], pm.samples)
+                img = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)
+                imgs.append(img)
+    else:
+        img = cv2.imread(image_file, cv2.IMREAD_COLOR)
+        if img is not None:
+            imgs = [img]
+
+    return imgs
+
+
+class Worker(QThread):
+    progressBarValue = Signal(int)
+    progressBarRange = Signal(int)
+    endsignal = Signal()
+    exceptedsignal = Signal(str)  # 发送一个异常信号
+    loopFlag = True
+
+    def __init__(self, predictors, save_pdf, vis_font_path, use_pdf2docx_api):
+        super(Worker, self).__init__()
+        self.predictors = predictors
+        self.save_pdf = save_pdf
+        self.vis_font_path = vis_font_path
+        self.lang = "EN"
+        self.imagePaths = []
+        self.use_pdf2docx_api = use_pdf2docx_api
+        self.outputDir = None
+        self.totalPageCnt = 0
+        self.pageCnt = 0
+        self.setStackSize(1024 * 1024)
+
+    def setImagePath(self, imagePaths):
+        self.imagePaths = imagePaths
+
+    def setLang(self, lang):
+        self.lang = lang
+
+    def setOutputDir(self, outputDir):
+        self.outputDir = outputDir
+
+    def setPDFParser(self, enabled):
+        self.use_pdf2docx_api = enabled
+
+    def resetPageCnt(self):
+        self.pageCnt = 0
+
+    def resetTotalPageCnt(self):
+        self.totalPageCnt = 0
+
+    def ppocrPrecitor(self, imgs, img_name):
+        all_res = []
+        # update progress bar ranges
+        self.totalPageCnt += len(imgs)
+        self.progressBarRange.emit(self.totalPageCnt)
+        # processing pages
+        for index, img in enumerate(imgs):
+            res, time_dict = self.predictors[self.lang](img)
+
+            # save output
+            save_structure_res(res, self.outputDir, img_name)
+            # draw_img = draw_structure_result(img, res, self.vis_font_path)
+            # img_save_path = os.path.join(self.outputDir, img_name, 'show_{}.jpg'.format(index))
+            # if res != []:
+            #     cv2.imwrite(img_save_path, draw_img)
+
+            # recovery
+            h, w, _ = img.shape
+            res = sorted_layout_boxes(res, w)
+            all_res += res
+            self.pageCnt += 1
+            self.progressBarValue.emit(self.pageCnt)
+
+        if all_res != []:
+            try:
+                convert_info_docx(imgs, all_res, self.outputDir, img_name)
+            except Exception as ex:
+                print(
+                    "error in layout recovery image:{}, err msg: {}".format(
+                        img_name, ex
+                    )
+                )
+        print("Predict time : {:.3f}s".format(time_dict["all"]))
+        print("result save to {}".format(self.outputDir))
+
+    def run(self):
+        self.resetPageCnt()
+        self.resetTotalPageCnt()
+        try:
+            os.makedirs(self.outputDir, exist_ok=True)
+            for i, image_file in enumerate(self.imagePaths):
+                if not self.loopFlag:
+                    break
+                # using use_pdf2docx_api for PDF parsing
+                if self.use_pdf2docx_api and os.path.basename(image_file)[-3:] == "pdf":
+                    try_import("pdf2docx")
+                    from pdf2docx.converter import Converter
+
+                    self.totalPageCnt += 1
+                    self.progressBarRange.emit(self.totalPageCnt)
+                    print("===============using use_pdf2docx_api===============")
+                    img_name = os.path.basename(image_file).split(".")[0]
+                    docx_file = os.path.join(self.outputDir, "{}.docx".format(img_name))
+                    cv = Converter(image_file)
+                    cv.convert(docx_file)
+                    cv.close()
+                    print("docx save to {}".format(docx_file))
+                    self.pageCnt += 1
+                    self.progressBarValue.emit(self.pageCnt)
+                else:
+                    # using PPOCR for PDF/Image parsing
+                    imgs = readImage(image_file)
+                    if len(imgs) == 0:
+                        continue
+                    img_name = os.path.basename(image_file).split(".")[0]
+                    os.makedirs(os.path.join(self.outputDir, img_name), exist_ok=True)
+                    self.ppocrPrecitor(imgs, img_name)
+                # file processed
+            self.endsignal.emit()
+            # self.exec()
+        except Exception as e:
+            self.exceptedsignal.emit(str(e))  # 将异常发送给UI进程
+
+
+class APP_Image2Doc(QWidget):
+    def __init__(self):
+        super().__init__()
+        # self.setFixedHeight(100)
+        # self.setFixedWidth(520)
+
+        # settings
+        self.imagePaths = []
+        # self.screenShotWg = ScreenShotWidget()
+        self.screenShot = None
+        self.save_pdf = False
+        self.output_dir = None
+        self.vis_font_path = os.path.join(root, "doc", "fonts", "simfang.ttf")
+        self.use_pdf2docx_api = False
+
+        # ProgressBar
+        self.pb = QProgressBar()
+        self.pb.setRange(0, 100)
+        self.pb.setValue(0)
+
+        # 初始化界面
+        self.setupUi()
+
+        # 下载模型
+        self.downloadModels(URLs_EN)
+        self.downloadModels(URLs_CN)
+
+        # 初始化模型
+        predictors = {
+            "EN": self.initPredictor("EN"),
+            "CN": self.initPredictor("CN"),
+        }
+
+        # 设置工作进程
+        self._thread = Worker(
+            predictors, self.save_pdf, self.vis_font_path, self.use_pdf2docx_api
+        )
+        self._thread.progressBarValue.connect(self.handleProgressBarUpdateSingal)
+        self._thread.endsignal.connect(self.handleEndsignalSignal)
+        # self._thread.finished.connect(QObject.deleteLater)
+        self._thread.progressBarRange.connect(self.handleProgressBarRangeSingal)
+        self._thread.exceptedsignal.connect(self.handleThreadException)
+        self.time_start = 0  # save start time
+
+    def setupUi(self):
+        self.setObjectName("MainWindow")
+        self.setWindowTitle(__APPNAME__ + " " + __VERSION__)
+
+        layout = QGridLayout()
+
+        self.openFileButton = QPushButton("打开文件")
+        self.openFileButton.setIcon(QIcon(QPixmap("./icons/folder-plus.png")))
+        layout.addWidget(self.openFileButton, 0, 0, 1, 1)
+        self.openFileButton.clicked.connect(self.handleOpenFileSignal)
+
+        # screenShotButton = QPushButton("截图识别")
+        # layout.addWidget(screenShotButton, 0, 1, 1, 1)
+        # screenShotButton.clicked.connect(self.screenShotSlot)
+        # screenShotButton.setEnabled(False) # temporarily disenble
+
+        self.startCNButton = QPushButton("中文转换")
+        self.startCNButton.setIcon(QIcon(QPixmap("./icons/chinese.png")))
+        layout.addWidget(self.startCNButton, 0, 1, 1, 1)
+        self.startCNButton.clicked.connect(
+            functools.partial(self.handleStartSignal, "CN", False)
+        )
+
+        self.startENButton = QPushButton("英文转换")
+        self.startENButton.setIcon(QIcon(QPixmap("./icons/english.png")))
+        layout.addWidget(self.startENButton, 0, 2, 1, 1)
+        self.startENButton.clicked.connect(
+            functools.partial(self.handleStartSignal, "EN", False)
+        )
+
+        self.PDFParserButton = QPushButton("PDF解析", self)
+        layout.addWidget(self.PDFParserButton, 0, 3, 1, 1)
+        self.PDFParserButton.clicked.connect(
+            functools.partial(self.handleStartSignal, "CN", True)
+        )
+
+        self.showResultButton = QPushButton("显示结果")
+        self.showResultButton.setIcon(QIcon(QPixmap("./icons/folder-open.png")))
+        layout.addWidget(self.showResultButton, 0, 4, 1, 1)
+        self.showResultButton.clicked.connect(self.handleShowResultSignal)
+
+        # ProgressBar
+        layout.addWidget(self.pb, 2, 0, 1, 5)
+        # time estimate label
+        self.timeEstLabel = QLabel(("Time Left: --"))
+        layout.addWidget(self.timeEstLabel, 3, 0, 1, 5)
+
+        self.setLayout(layout)
+
+    def downloadModels(self, URLs):
+        # using custom model
+        tar_file_name_list = [
+            "inference.pdiparams",
+            "inference.pdiparams.info",
+            "inference.pdmodel",
+            "model.pdiparams",
+            "model.pdiparams.info",
+            "model.pdmodel",
+        ]
+        model_path = os.path.join(root, "inference")
+        os.makedirs(model_path, exist_ok=True)
+
+        # download and unzip models
+        for name in URLs.keys():
+            url = URLs[name]
+            print("Try downloading file: {}".format(url))
+            tarname = url.split("/")[-1]
+            tarpath = os.path.join(model_path, tarname)
+            if os.path.exists(tarpath):
+                print("File have already exist. skip")
+            else:
+                try:
+                    download_with_progressbar(url, tarpath)
+                except Exception as e:
+                    print("Error occurred when downloading file, error message:")
+                    print(e)
+
+            # unzip model tar
+            try:
+                with tarfile.open(tarpath, "r") as tarObj:
+                    storage_dir = os.path.join(model_path, name)
+                    os.makedirs(storage_dir, exist_ok=True)
+                    for member in tarObj.getmembers():
+                        filename = None
+                        for tar_file_name in tar_file_name_list:
+                            if tar_file_name in member.name:
+                                filename = tar_file_name
+                        if filename is None:
+                            continue
+                        file = tarObj.extractfile(member)
+                        with open(os.path.join(storage_dir, filename), "wb") as f:
+                            f.write(file.read())
+            except Exception as e:
+                print("Error occurred when unziping file, error message:")
+                print(e)
+
+    def initPredictor(self, lang="EN"):
+        # init predictor args
+        args = parse_args()
+        args.table_max_len = 488
+        args.ocr = True
+        args.recovery = True
+        args.save_pdf = self.save_pdf
+        args.table_char_dict_path = os.path.join(
+            root, "ppocr", "utils", "dict", "table_structure_dict.txt"
+        )
+        if lang == "EN":
+            args.det_model_dir = os.path.join(
+                root, "inference", "en_PP-OCRv3_det_infer"  # 此处从这里找到模型存放位置
+            )
+            args.rec_model_dir = os.path.join(
+                root, "inference", "en_PP-OCRv3_rec_infer"
+            )
+            args.table_model_dir = os.path.join(
+                root, "inference", "en_ppstructure_mobile_v2.0_SLANet_infer"
+            )
+            args.output = os.path.join(root, "output")  # 结果保存路径
+            args.layout_model_dir = os.path.join(
+                root, "inference", "picodet_lcnet_x1_0_fgd_layout_infer"
+            )
+            lang_dict = DICT_EN
+        elif lang == "CN":
+            args.det_model_dir = os.path.join(
+                root, "inference", "cn_PP-OCRv3_det_infer"  # 此处从这里找到模型存放位置
+            )
+            args.rec_model_dir = os.path.join(
+                root, "inference", "cn_PP-OCRv3_rec_infer"
+            )
+            args.table_model_dir = os.path.join(
+                root, "inference", "cn_ppstructure_mobile_v2.0_SLANet_infer"
+            )
+            args.output = os.path.join(root, "output")  # 结果保存路径
+            args.layout_model_dir = os.path.join(
+                root, "inference", "picodet_lcnet_x1_0_fgd_layout_cdla_infer"
+            )
+            lang_dict = DICT_CN
+        else:
+            raise ValueError("Unsupported language")
+        args.rec_char_dict_path = os.path.join(
+            root, "ppocr", "utils", lang_dict["rec_char_dict_path"]
+        )
+        args.layout_dict_path = os.path.join(
+            root, "ppocr", "utils", "dict", "layout_dict", lang_dict["layout_dict_path"]
+        )
+        # init predictor
+        return StructureSystem(args)
+
+    def handleOpenFileSignal(self):
+        """
+        可以多选图像文件
+        """
+        selectedFiles = QFileDialog.getOpenFileNames(
+            self, "多文件选择", "/", "图片文件 (*.png *.jpeg *.jpg *.bmp *.pdf)"
+        )[0]
+        if len(selectedFiles) > 0:
+            self.imagePaths = selectedFiles
+            self.screenShot = None  # discard screenshot temp image
+            self.pb.setValue(0)
+
+    # def screenShotSlot(self):
+    #     '''
+    #     选定图像文件和截图的转换过程只能同时进行一个
+    #     截图只能同时转换一个
+    #     '''
+    #     self.screenShotWg.start()
+    #     if self.screenShotWg.captureImage:
+    #         self.screenShot = self.screenShotWg.captureImage
+    #         self.imagePaths.clear() # discard openfile temp list
+    #         self.pb.setRange(0, 1)
+    #         self.pb.setValue(0)
+
+    def handleStartSignal(self, lang="EN", pdfParser=False):
+        if self.screenShot:  # for screenShot
+            img_name = "screenshot_" + time.strftime("%Y%m%d%H%M%S", time.localtime())
+            image = QImageToCvMat(self.screenShot)
+            self.predictAndSave(image, img_name, lang)
+            # update Progress Bar
+            self.pb.setValue(1)
+            QMessageBox.information(self, "Information", "文档提取完成")
+        elif len(self.imagePaths) > 0:  # for image file selection
+            # Must set image path list and language before start
+            self.output_dir = os.path.join(
+                os.path.dirname(self.imagePaths[0]), "output"
+            )  # output_dir should be same as imagepath
+            self._thread.setOutputDir(self.output_dir)
+            self._thread.setImagePath(self.imagePaths)
+            self._thread.setLang(lang)
+            self._thread.setPDFParser(pdfParser)
+            # disable buttons
+            self.openFileButton.setEnabled(False)
+            self.startCNButton.setEnabled(False)
+            self.startENButton.setEnabled(False)
+            self.PDFParserButton.setEnabled(False)
+            # 启动工作进程
+            self._thread.start()
+            self.time_start = time.time()  # log start time
+            QMessageBox.information(self, "Information", "开始转换")
+        else:
+            QMessageBox.warning(self, "Information", "请选择要识别的文件或截图")
+
+    def handleShowResultSignal(self):
+        if self.output_dir is None:
+            return
+        if os.path.exists(self.output_dir):
+            if platform.system() == "Windows":
+                os.startfile(self.output_dir)
+            else:
+                subprocess.check_call(["open", os.path.normpath(self.output_dir)])
+        else:
+            QMessageBox.information(self, "Information", "输出文件不存在")
+
+    def handleProgressBarUpdateSingal(self, i):
+        self.pb.setValue(i)
+        # calculate time left of recognition
+        lenbar = self.pb.maximum()
+        avg_time = (
+            time.time() - self.time_start
+        ) / i  # Use average time to prevent time fluctuations
+        time_left = str(datetime.timedelta(seconds=avg_time * (lenbar - i))).split(".")[
+            0
+        ]  # Remove microseconds
+        self.timeEstLabel.setText(f"Time Left: {time_left}")  # show time left
+
+    def handleProgressBarRangeSingal(self, max):
+        self.pb.setRange(0, max)
+
+    def handleEndsignalSignal(self):
+        # enable buttons
+        self.openFileButton.setEnabled(True)
+        self.startCNButton.setEnabled(True)
+        self.startENButton.setEnabled(True)
+        self.PDFParserButton.setEnabled(True)
+        QMessageBox.information(self, "Information", "转换结束")
+
+    def handleCBChangeSignal(self):
+        self._thread.setPDFParser(self.checkBox.isChecked())
+
+    def handleThreadException(self, message):
+        self._thread.quit()
+        QMessageBox.information(self, "Error", message)
+
+
+def main():
+    app = QApplication(sys.argv)
+
+    window = APP_Image2Doc()  # 创建对象
+    window.show()  # 全屏显示窗口
+
+    QApplication.processEvents()
+    sys.exit(app.exec())
+
+
+if __name__ == "__main__":
+    main()
diff --git a/ppstructure/predict_system.py b/ppstructure/predict_system.py
new file mode 100644
index 0000000..152e989
--- /dev/null
+++ b/ppstructure/predict_system.py
@@ -0,0 +1,415 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import sys
+import subprocess
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "../")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+import cv2
+import json
+import numpy as np
+import time
+import logging
+from copy import deepcopy
+
+from paddle.utils import try_import
+from ppocr.utils.utility import get_image_file_list, check_and_read
+from ppocr.utils.logging import get_logger
+from ppocr.utils.visual import draw_ser_results, draw_re_results
+from tools.infer.predict_system import TextSystem
+from tools.infer.predict_rec import TextRecognizer
+from ppstructure.layout.predict_layout import LayoutPredictor
+from ppstructure.table.predict_table import TableSystem, to_excel
+from ppstructure.utility import parse_args, draw_structure_result, cal_ocr_word_box
+
+logger = get_logger()
+
+
+class StructureSystem(object):
+    def __init__(self, args):
+        self.mode = args.mode
+        self.recovery = args.recovery
+
+        self.image_orientation_predictor = None
+        if args.image_orientation:
+            import paddleclas
+
+            self.image_orientation_predictor = paddleclas.PaddleClas(
+                model_name="text_image_orientation"
+            )
+
+        if self.mode == "structure":
+            if not args.show_log:
+                logger.setLevel(logging.INFO)
+            if args.layout == False and args.ocr == True:
+                args.ocr = False
+                logger.warning(
+                    "When args.layout is false, args.ocr is automatically set to false"
+                )
+            # init model
+            self.layout_predictor = None
+            self.text_system = None
+            self.table_system = None
+            self.formula_system = None
+            if args.layout:
+                self.layout_predictor = LayoutPredictor(args)
+                if args.ocr:
+                    self.text_system = TextSystem(args)
+            if args.table:
+                if self.text_system is not None:
+                    self.table_system = TableSystem(
+                        args,
+                        self.text_system.text_detector,
+                        self.text_system.text_recognizer,
+                    )
+                else:
+                    self.table_system = TableSystem(args)
+            if args.formula:
+                args_formula = deepcopy(args)
+                args_formula.rec_algorithm = args.formula_algorithm
+                args_formula.rec_model_dir = args.formula_model_dir
+                args_formula.rec_char_dict_path = args.formula_char_dict_path
+                args_formula.rec_batch_num = args.formula_batch_num
+                self.formula_system = TextRecognizer(args_formula)
+
+        elif self.mode == "kie":
+            from ppstructure.kie.predict_kie_token_ser_re import SerRePredictor
+
+            self.kie_predictor = SerRePredictor(args)
+
+        self.return_word_box = args.return_word_box
+
+    def __call__(self, img, return_ocr_result_in_table=False, img_idx=0):
+        time_dict = {
+            "image_orientation": 0,
+            "layout": 0,
+            "table": 0,
+            "table_match": 0,
+            "formula": 0,
+            "det": 0,
+            "rec": 0,
+            "kie": 0,
+            "all": 0,
+        }
+        start = time.time()
+
+        if self.image_orientation_predictor is not None:
+            tic = time.time()
+            cls_result = self.image_orientation_predictor.predict(input_data=img)
+            cls_res = next(cls_result)
+            angle = cls_res[0]["label_names"][0]
+            cv_rotate_code = {
+                "90": cv2.ROTATE_90_COUNTERCLOCKWISE,
+                "180": cv2.ROTATE_180,
+                "270": cv2.ROTATE_90_CLOCKWISE,
+            }
+            if angle in cv_rotate_code:
+                img = cv2.rotate(img, cv_rotate_code[angle])
+            toc = time.time()
+            time_dict["image_orientation"] = toc - tic
+
+        if self.mode == "structure":
+            ori_im = img.copy()
+            if self.layout_predictor is not None:
+                layout_res, elapse = self.layout_predictor(img)
+                time_dict["layout"] += elapse
+            else:
+                h, w = ori_im.shape[:2]
+                layout_res = [dict(bbox=None, label="table", score=0.0)]
+
+            # As reported in issues such as #10270 and #11665, the old
+            # implementation, which recognizes texts from the layout regions,
+            # has problems with OCR recognition accuracy.
+            #
+            # To enhance the OCR recognition accuracy, we implement a patch fix
+            # that first use text_system to detect and recognize all text information
+            # and then filter out relevant texts according to the layout regions.
+            text_res = None
+            if self.text_system is not None:
+                text_res, ocr_time_dict = self._predict_text(img)
+                time_dict["det"] += ocr_time_dict["det"]
+                time_dict["rec"] += ocr_time_dict["rec"]
+
+            res_list = []
+            for region in layout_res:
+                res = ""
+                if region["bbox"] is not None:
+                    x1, y1, x2, y2 = region["bbox"]
+                    x1, y1, x2, y2 = int(x1), int(y1), int(x2), int(y2)
+                    roi_img = ori_im[y1:y2, x1:x2, :]
+                else:
+                    x1, y1, x2, y2 = 0, 0, w, h
+                    roi_img = ori_im
+                bbox = [x1, y1, x2, y2]
+
+                if region["label"] == "table":
+                    if self.table_system is not None:
+                        res, table_time_dict = self.table_system(
+                            roi_img, return_ocr_result_in_table
+                        )
+                        time_dict["table"] += table_time_dict["table"]
+                        time_dict["table_match"] += table_time_dict["match"]
+                        time_dict["det"] += table_time_dict["det"]
+                        time_dict["rec"] += table_time_dict["rec"]
+
+                elif region["label"] == "equation" and self.formula_system is not None:
+                    latex_res, formula_time = self.formula_system([roi_img])
+                    time_dict["formula"] += formula_time
+                    res = {"latex": latex_res[0]}
+
+                else:
+                    if text_res is not None:
+                        # Filter the text results whose regions intersect with the current layout bbox.
+                        res = self._filter_text_res(text_res, bbox)
+
+                res_list.append(
+                    {
+                        "type": region["label"].lower(),
+                        "bbox": bbox,
+                        "img": roi_img,
+                        "res": res,
+                        "img_idx": img_idx,
+                        "score": region["score"],
+                    }
+                )
+
+            end = time.time()
+            time_dict["all"] = end - start
+            return res_list, time_dict
+
+        elif self.mode == "kie":
+            re_res, elapse = self.kie_predictor(img)
+            time_dict["kie"] = elapse
+            time_dict["all"] = elapse
+            return re_res[0], time_dict
+
+        return None, None
+
+    def _predict_text(self, img):
+        filter_boxes, filter_rec_res, ocr_time_dict = self.text_system(img)
+
+        # remove style char,
+        # when using the recognition model trained on the PubtabNet dataset,
+        # it will recognize the text format in the table, such as <b>
+        style_token = [
+            "<strike>",
+            "<strike>",
+            "<sup>",
+            "</sub>",
+            "<b>",
+            "</b>",
+            "<sub>",
+            "</sup>",
+            "<overline>",
+            "</overline>",
+            "<underline>",
+            "</underline>",
+            "<i>",
+            "</i>",
+        ]
+        res = []
+        for box, rec_res in zip(filter_boxes, filter_rec_res):
+            rec_str, rec_conf = rec_res[0], rec_res[1]
+            for token in style_token:
+                if token in rec_str:
+                    rec_str = rec_str.replace(token, "")
+            if self.return_word_box:
+                word_box_content_list, word_box_list = cal_ocr_word_box(
+                    rec_str, box, rec_res[2]
+                )
+                res.append(
+                    {
+                        "text": rec_str,
+                        "confidence": float(rec_conf),
+                        "text_region": box.tolist(),
+                        "text_word": word_box_content_list,
+                        "text_word_region": word_box_list,
+                    }
+                )
+            else:
+                res.append(
+                    {
+                        "text": rec_str,
+                        "confidence": float(rec_conf),
+                        "text_region": box.tolist(),
+                    }
+                )
+        return res, ocr_time_dict
+
+    def _filter_text_res(self, text_res, bbox):
+        res = []
+        for r in text_res:
+            box = r["text_region"]
+            rect = box[0][0], box[0][1], box[2][0], box[2][1]
+            if self._has_intersection(bbox, rect):
+                res.append(r)
+        return res
+
+    def _has_intersection(self, rect1, rect2):
+        x_min1, y_min1, x_max1, y_max1 = rect1
+        x_min2, y_min2, x_max2, y_max2 = rect2
+        if x_min1 > x_max2 or x_max1 < x_min2:
+            return False
+        if y_min1 > y_max2 or y_max1 < y_min2:
+            return False
+        return True
+
+
+def save_structure_res(res, save_folder, img_name, img_idx=0):
+    excel_save_folder = os.path.join(save_folder, img_name)
+    os.makedirs(excel_save_folder, exist_ok=True)
+    res_cp = deepcopy(res)
+    # save res
+    with open(
+        os.path.join(excel_save_folder, "res_{}.txt".format(img_idx)),
+        "w",
+        encoding="utf8",
+    ) as f:
+        for region in res_cp:
+            roi_img = region.pop("img")
+            f.write("{}\n".format(json.dumps(region)))
+
+            if (
+                region["type"].lower() == "table"
+                and len(region["res"]) > 0
+                and "html" in region["res"]
+            ):
+                excel_path = os.path.join(
+                    excel_save_folder, "{}_{}.xlsx".format(region["bbox"], img_idx)
+                )
+                to_excel(region["res"]["html"], excel_path)
+            elif region["type"].lower() == "figure":
+                img_path = os.path.join(
+                    excel_save_folder, "{}_{}.jpg".format(region["bbox"], img_idx)
+                )
+                cv2.imwrite(img_path, roi_img)
+
+
+def main(args):
+    image_file_list = get_image_file_list(args.image_dir)
+    image_file_list = image_file_list
+    image_file_list = image_file_list[args.process_id :: args.total_process_num]
+
+    if not args.use_pdf2docx_api:
+        structure_sys = StructureSystem(args)
+        save_folder = os.path.join(args.output, structure_sys.mode)
+        os.makedirs(save_folder, exist_ok=True)
+    img_num = len(image_file_list)
+
+    for i, image_file in enumerate(image_file_list):
+        logger.info("[{}/{}] {}".format(i, img_num, image_file))
+        img, flag_gif, flag_pdf = check_and_read(image_file)
+        img_name = os.path.basename(image_file).split(".")[0]
+
+        if args.recovery and args.use_pdf2docx_api and flag_pdf:
+            try_import("pdf2docx")
+            from pdf2docx.converter import Converter
+
+            os.makedirs(args.output, exist_ok=True)
+            docx_file = os.path.join(args.output, "{}_api.docx".format(img_name))
+            cv = Converter(image_file)
+            cv.convert(docx_file)
+            cv.close()
+            logger.info("docx save to {}".format(docx_file))
+            continue
+
+        if not flag_gif and not flag_pdf:
+            img = cv2.imread(image_file)
+
+        if not flag_pdf:
+            if img is None:
+                logger.error("error in loading image:{}".format(image_file))
+                continue
+            imgs = [img]
+        else:
+            imgs = img
+
+        all_res = []
+        for index, img in enumerate(imgs):
+            res, time_dict = structure_sys(img, img_idx=index)
+            img_save_path = os.path.join(
+                save_folder, img_name, "show_{}.jpg".format(index)
+            )
+            os.makedirs(os.path.join(save_folder, img_name), exist_ok=True)
+            if structure_sys.mode == "structure" and res != []:
+                draw_img = draw_structure_result(img, res, args.vis_font_path)
+                save_structure_res(res, save_folder, img_name, index)
+            elif structure_sys.mode == "kie":
+                if structure_sys.kie_predictor.predictor is not None:
+                    draw_img = draw_re_results(img, res, font_path=args.vis_font_path)
+                else:
+                    draw_img = draw_ser_results(img, res, font_path=args.vis_font_path)
+
+                with open(
+                    os.path.join(save_folder, img_name, "res_{}_kie.txt".format(index)),
+                    "w",
+                    encoding="utf8",
+                ) as f:
+                    res_str = "{}\t{}\n".format(
+                        image_file, json.dumps({"ocr_info": res}, ensure_ascii=False)
+                    )
+                    f.write(res_str)
+            if res != []:
+                cv2.imwrite(img_save_path, draw_img)
+                logger.info("result save to {}".format(img_save_path))
+            if args.recovery and res != []:
+                from ppstructure.recovery.recovery_to_doc import (
+                    sorted_layout_boxes,
+                    convert_info_docx,
+                )
+                from ppstructure.recovery.recovery_to_markdown import (
+                    convert_info_markdown,
+                )
+
+                h, w, _ = img.shape
+                res = sorted_layout_boxes(res, w)
+                all_res += res
+
+        if args.recovery and all_res != []:
+            try:
+                convert_info_docx(img, all_res, save_folder, img_name)
+                if args.recovery_to_markdown:
+                    convert_info_markdown(all_res, save_folder, img_name)
+            except Exception as ex:
+                logger.error(
+                    "error in layout recovery image:{}, err msg: {}".format(
+                        image_file, ex
+                    )
+                )
+                continue
+        logger.info("Predict time : {:.3f}s".format(time_dict["all"]))
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    if args.use_mp:
+        p_list = []
+        total_process_num = args.total_process_num
+        for process_id in range(total_process_num):
+            cmd = (
+                [sys.executable, "-u"]
+                + sys.argv
+                + ["--process_id={}".format(process_id), "--use_mp={}".format(False)]
+            )
+            p = subprocess.Popen(cmd, stdout=sys.stdout, stderr=sys.stdout)
+            p_list.append(p)
+        for p in p_list:
+            p.wait()
+    else:
+        main(args)
diff --git a/ppstructure/recovery/README.md b/ppstructure/recovery/README.md
new file mode 100644
index 0000000..04158b9
--- /dev/null
+++ b/ppstructure/recovery/README.md
@@ -0,0 +1,213 @@
+English | [简体中文](README_ch.md)
+
+# Layout Recovery
+
+- [1. Introduction](#1)
+- [2. Install](#2)
+    - [2.1 Install PaddlePaddle](#2.1)
+    - [2.2 Install PaddleOCR](#2.2)
+- [3. Quick Start using standard PDF parse](#3)
+- [4. Quick Start using image format PDF parse ](#4)
+    - [4.1 Download models](#4.1)
+    - [4.2 Layout recovery](#4.2)
+- [5. More](#5)
+
+<a name="1"></a>
+
+## 1. Introduction
+
+The layout recovery module is used to restore the image or pdf to an
+editable Word file consistent with the original image layout.
+
+Two layout recovery methods are provided, you can choose by PDF format:
+
+- **Standard PDF parse(the input is standard PDF)**: Python based PDF to word library [pdf2docx] (https://github.com/dothinking/pdf2docx) is optimized, the method extracts data from PDF with PyMuPDF, then parse layout with rule, finally, generate docx with python-docx.
+
+- **Image format PDF parse(the input can be standard PDF or image format PDF)**: Layout recovery combines [layout analysis](../layout/README.md)、[table recognition](../table/README.md) to better recover images, tables, titles, etc. supports input files in PDF and document image formats in Chinese and English.
+
+The input formats and application scenarios of the two methods are as follows:
+
+|  method   | input formats |                      application scenarios/problem                       |
+| :-----: | :----------: | :----------------------------------------------------------: |
+| Standard PDF parse |     pdf      | Advantages: Better recovery for non-paper documents, each page remains on the same page after restoration<br>Disadvantages: English characters in some Chinese documents are garbled, some contents are still beyond the current page, the whole page content is restored to the table format, and the recovery effect of some pictures is not good |
+| Image format PDF parse( |  pdf、picture   | Advantages: More suitable for paper document content recovery,  OCR recognition effect is more good<br>Disadvantages: Currently, the recovery is based on rules, the effect of content typesetting (spacing, fonts, etc.) need to be further improved, and the effect of layout recovery depends on layout analysis |
+
+The following figure shows the effect of restoring the layout of documents by using PDF parse:
+
+<div align="center">
+<img src="https://user-images.githubusercontent.com/19808900/195319853-045123c9-f542-4596-b4e4-6081708dfc56.png"  width = "700" />
+</div>
+
+The following figures show the effect of restoring the layout of English and Chinese documents by using OCR technique:
+
+<div align="center">
+<img src="../docs/recovery/recovery.jpg"  width = "700" />
+</div>
+
+<div align="center">
+<img src="../docs/recovery/recovery_ch.jpg"  width = "800" />
+</div>
+
+
+<a name="2"></a>
+
+## 2. Install
+
+<a name="2.1"></a>
+
+### 2.1 Install PaddlePaddle
+
+```bash
+python3 -m pip install --upgrade pip
+
+# If you have cuda9 or cuda10 installed on your machine, please run the following command to install
+python3 -m pip install "paddlepaddle-gpu" -i https://mirror.baidu.com/pypi/simple
+
+# CPU installation
+python3 -m pip install "paddlepaddle" -i https://mirror.baidu.com/pypi/simple
+````
+
+For more requirements, please refer to the instructions in [Installation Documentation](https://www.paddlepaddle.org.cn/en/install/quick?docurl=/documentation/docs/en/install/pip/macos-pip_en.html).
+
+<a name="2.2"></a>
+
+### 2.2 Install PaddleOCR
+
+- **(1) Download source code**
+
+```bash
+[Recommended] git clone https://github.com/PaddlePaddle/PaddleOCR
+
+# If the pull cannot be successful due to network problems, you can also choose to use the hosting on the code cloud:
+git clone https://gitee.com/paddlepaddle/PaddleOCR
+
+# Note: Code cloud hosting code may not be able to synchronize the update of this github project in real time, there is a delay of 3 to 5 days, please use the recommended method first.
+````
+
+- **(2) Install recovery `requirements`**
+
+The layout restoration is exported as docx files, so python-docx API need to be installed, and PyMuPDF api([requires Python >= 3.7](https://pypi.org/project/PyMuPDF/)) need to be installed to process the input files in pdf format.
+
+Install all the libraries by running the following command:
+
+```bash
+python3 -m pip install -r ppstructure/recovery/requirements.txt
+````
+
+ And if using pdf parse method, we need to install pdf2docx api.
+
+```bash
+wget https://paddleocr.bj.bcebos.com/whl/pdf2docx-0.0.0-py3-none-any.whl
+pip3 install pdf2docx-0.0.0-py3-none-any.whl
+```
+
+<a name="3"></a>
+
+## 3. Quick Start using standard PDF parse
+
+`use_pdf2docx_api` use PDF parse for layout recovery, The whl package is also provided  for quick use, follow the above code, for more information please refer to [quickstart](../docs/quickstart_en.md) for details.
+
+```bash
+# install paddleocr
+pip3 install "paddleocr>=2.6"
+paddleocr --image_dir=ppstructure/docs/recovery/UnrealText.pdf --type=structure --recovery=true --use_pdf2docx_api=true
+```
+
+Command line:
+
+```bash
+python3 predict_system.py \
+    --image_dir=ppstructure/docs/recovery/UnrealText.pdf \
+    --recovery=True \
+    --use_pdf2docx_api=True \
+    --output=../output/
+```
+
+<a name="4"></a>
+## 4. Quick Start using image format PDF parse
+
+Through layout analysis, we divided the image/PDF documents into regions, located the key regions, such as text, table, picture, etc., and recorded the location, category, and regional pixel value information of each region. Different regions are processed separately, where:
+
+- OCR detection and recognition is performed in the text area, and the coordinates of the OCR detection box and the text content information are added on the basis of the previous information
+
+- The table area identifies tables and records html and text information of tables
+- Save the image directly
+
+We can restore the test picture through the layout information, OCR detection and recognition structure, table information, and saved pictures.
+
+The whl package is also provided  for quick use, follow the above code, for more information please refer to [quickstart](../docs/quickstart_en.md) for details.
+
+```bash
+paddleocr --image_dir=ppstructure/docs/table/1.png --type=structure --recovery=true --lang='en'
+```
+
+<a name="4.1"></a>
+### 4.1 Download models
+
+If input is English document, download English models:
+
+```bash
+cd PaddleOCR/ppstructure
+
+# download model
+mkdir inference && cd inference
+# Download the detection model of the ultra-lightweight English PP-OCRv3 model and unzip it
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar && tar xf en_PP-OCRv3_det_infer.tar
+# Download the recognition model of the ultra-lightweight English PP-OCRv3 model and unzip it
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_infer.tar && tar xf en_PP-OCRv3_rec_infer.tar
+# Download the ultra-lightweight English table inch model and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/en_ppstructure_mobile_v2.0_SLANet_infer.tar
+tar xf en_ppstructure_mobile_v2.0_SLANet_infer.tar
+# Download the layout model of publaynet dataset and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar
+tar xf picodet_lcnet_x1_0_fgd_layout_infer.tar
+cd ..
+```
+If input is Chinese document，download Chinese models:
+[Chinese and English ultra-lightweight PP-OCRv3 model](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/README.md#pp-ocr-series-model-listupdate-on-september-8th)、[table recognition model](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppstructure/docs/models_list.md#22-表格识别模型)、[layout analysis model](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppstructure/docs/models_list.md#1-版面分析模型)
+
+<a name="4.2"></a>
+### 4.2 Layout recovery
+
+
+```bash
+python3 predict_system.py \
+    --image_dir=./docs/table/1.png \
+    --det_model_dir=inference/en_PP-OCRv3_det_infer \
+    --rec_model_dir=inference/en_PP-OCRv3_rec_infer \
+    --rec_char_dict_path=../ppocr/utils/en_dict.txt \
+    --table_model_dir=inference/en_ppstructure_mobile_v2.0_SLANet_infer \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt \
+    --layout_model_dir=inference/picodet_lcnet_x1_0_fgd_layout_infer \
+    --layout_dict_path=../ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt \
+    --vis_font_path=../doc/fonts/simfang.ttf \
+    --recovery=True \
+    --output=../output/
+```
+
+After running, the docx of each picture will be saved in the directory specified by the output field
+
+Field：
+
+- image_dir：test file， can be picture, picture directory, pdf file, pdf file directory
+- det_model_dir：OCR detection model path
+- rec_model_dir：OCR recognition model path
+- rec_char_dict_path：OCR recognition dict path. If the Chinese model is used, change to "../ppocr/utils/ppocr_keys_v1.txt". And if you trained the model on your own dataset, change to the trained dictionary
+- table_model_dir：table recognition model path
+- table_char_dict_path：table recognition dict path. If the Chinese model is used, no need to change
+- layout_model_dir：layout analysis model path
+- layout_dict_path：layout analysis dict path. If the Chinese model is used, change to "../ppocr/utils/dict/layout_dict/layout_cdla_dict.txt"
+- recovery：whether to enable layout of recovery, default False
+- output：save the recovery result path
+
+<a name="5"></a>
+
+## 5. More
+
+For training, evaluation and inference tutorial for text detection models, please refer to [text detection doc](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_en/detection_en.md).
+
+For training, evaluation and inference tutorial for text recognition models, please refer to [text recognition doc](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_en/recognition_en.md).
+
+For training, evaluation and inference tutorial for layout analysis models, please refer to [layout analysis doc](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppstructure/layout/README.md)
+
+For training, evaluation and inference tutorial for table recognition models, please refer to [table recognition doc](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppstructure/table/README.md)
diff --git a/ppstructure/recovery/README_ch.md b/ppstructure/recovery/README_ch.md
new file mode 100644
index 0000000..01f843f
--- /dev/null
+++ b/ppstructure/recovery/README_ch.md
@@ -0,0 +1,222 @@
+[English](README.md) | 简体中文
+
+# 版面恢复
+
+- [1. 简介](#1)
+- [2. 安装](#2)
+  - [2.1 安装PaddlePaddle](#2.1)
+  - [2.2 安装PaddleOCR](#2.2)
+- [3.使用标准PDF解析进行版面恢复](#3)
+- [4. 使用图片格式PDF解析进行版面恢复](#4)
+  - [4.1 下载模型](#4.1)
+  - [4.2 版面恢复](#4.2)
+- [5. 更多](#5)
+
+<a name="1"></a>
+
+## 1. 简介
+
+版面恢复就是将输入的图片、pdf内容仍然像原文档那样排列着，段落不变、顺序不变的输出到word文档中等。
+
+提供了2种版面恢复方法，可根据输入PDF的格式进行选择：
+
+- **标准PDF解析(输入须为标准PDF)**：基于Python的pdf转word库[pdf2docx](https://github.com/dothinking/pdf2docx)进行优化，该方法通过PyMuPDF获取页面元素，然后利用规则解析章节、段落、表格等布局及样式，最后通过python-docx将解析的内容元素重建到word文档中。
+- **图片格式PDF解析(输入可为标准PDF或图片格式PDF)**：结合[版面分析](../layout/README_ch.md)、[表格识别](../table/README_ch.md)技术，从而更好地恢复图片、表格、标题等内容，支持中、英文pdf文档、文档图片格式的输入文件。
+
+2种方法输入格式、适用场景如下：
+
+|      方法       | 支持输入文件 |                      适用场景/存在问题                       |
+| :-------------: | :----------: | :----------------------------------------------------------: |
+|   标准PDF解析   |     pdf      | 优点：非论文文档恢复效果更优、每一页内容恢复后仍在同一页<br>缺点：有些中文文档中的英文乱码、仍存在内容超出当前页面的情况、整页内容恢复为表格格式、部分图片恢复效果不佳 |
+| 图片格式PDF解析 |  pdf、图片   | 优点：更适合论文文档正文内容的恢复、中英文文档OCR识别效果好<br>缺点：目前内容恢复基于规则，内容排版效果(间距、字体等)待进一步提升、版面恢复效果依赖于版面分析效果 |
+
+下图展示了通过PDF解析版面恢复效果：
+
+<div align="center">
+<img src="https://user-images.githubusercontent.com/19808900/195319840-68fc60ec-ea66-4095-b734-0ec115860341.png"  width = "700" />
+</div>
+
+下图分别展示了通过OCR技术，英文文档和中文文档版面恢复的效果：
+
+<div align="center">
+<img src="../docs/recovery/recovery.jpg"  width = "700" />
+</div>
+<div align="center">
+<img src="../docs/recovery/recovery_ch.jpg"  width = "800" />
+</div>
+<a name="2"></a>
+
+## 2. 安装
+
+<a name="2.1"></a>
+
+### 2.1 安装PaddlePaddle
+
+```bash
+python3 -m pip install --upgrade pip
+
+# 您的机器安装的是CUDA9或CUDA10，请运行以下命令安装
+python3 -m pip install "paddlepaddle-gpu" -i https://mirror.baidu.com/pypi/simple
+
+# 您的机器是CPU，请运行以下命令安装
+python3 -m pip install "paddlepaddle" -i https://mirror.baidu.com/pypi/simple
+
+```
+
+更多需求，请参照[安装文档](https://www.paddlepaddle.org.cn/install/quick)中的说明进行操作。
+
+<a name="2.2"></a>
+
+### 2.2 安装PaddleOCR
+
+- **（1）下载版面恢复源码**
+
+```bash
+【推荐】git clone https://github.com/PaddlePaddle/PaddleOCR
+
+# 如果因为网络问题无法pull成功，也可选择使用码云上的托管：
+git clone https://gitee.com/paddlepaddle/PaddleOCR
+
+# 注：码云托管代码可能无法实时同步本github项目更新，存在3~5天延时，请优先使用推荐方式。
+```
+
+- **（2）安装recovery的`requirements`**
+
+版面恢复导出为docx文件，所以需要安装Python处理word文档的python-docx API，同时处理pdf格式的输入文件，需要安装PyMuPDF API([要求Python >= 3.7](https://pypi.org/project/PyMuPDF/))。
+
+通过如下命令安装全部库：
+
+```bash
+python3 -m pip install -r ppstructure/recovery/requirements.txt
+```
+
+使用pdf2docx库解析的方式恢复文档需要安装优化的pdf2docx。
+
+```bash
+wget https://paddleocr.bj.bcebos.com/whl/pdf2docx-0.0.0-py3-none-any.whl
+pip3 install pdf2docx-0.0.0-py3-none-any.whl
+```
+
+<a name="3"></a>
+
+## 3.使用标准PDF解析进行版面恢复
+
+`use_pdf2docx_api`表示使用PDF解析的方式进行版面恢复，通过whl包的形式方便快速使用，代码如下，更多信息详见 [quickstart](../docs/quickstart.md)。
+
+```bash
+# 安装 paddleocr，推荐使用2.6版本
+pip3 install "paddleocr>=2.6"
+paddleocr --image_dir=ppstructure/docs/recovery/UnrealText.pdf --type=structure --recovery=true --use_pdf2docx_api=true
+```
+
+通过命令行的方式：
+
+```bash
+python3 predict_system.py \
+    --image_dir=ppstructure/docs/recovery/UnrealText.pdf \
+    --recovery=True \
+    --use_pdf2docx_api=True \
+    --output=../output/
+```
+
+<a name="4"></a>
+
+## 4.使用图片格式PDF解析进行版面恢复
+
+我们通过版面分析对图片/pdf形式的文档进行区域划分，定位其中的关键区域，如文字、表格、图片等，记录每个区域的位置、类别、区域像素值信息。对不同的区域分别处理，其中：
+
+- 文字区域直接进行OCR检测和识别，在之前信息基础上增加OCR检测框坐标和文本内容信息
+
+- 表格区域进行表格识别，记录表格html和文字信息
+- 图片直接保存
+
+我们通过版面信息、OCR检测和识别结构、表格信息、保存的图片，对测试图片进行恢复即可。
+
+提供如下代码实现版面恢复，也提供了whl包的形式方便快速使用，代码如下，更多信息详见 [quickstart](../docs/quickstart.md)。
+
+```bash
+# 安装 paddleocr，推荐使用2.6版本
+pip3 install "paddleocr>=2.6"
+# 中文测试图
+paddleocr --image_dir=ppstructure/docs/table/1.png --type=structure --recovery=true
+# 英文测试图
+paddleocr --image_dir=ppstructure/docs/table/1.png --type=structure --recovery=true --lang='en'
+# pdf测试文件
+paddleocr --image_dir=ppstructure/docs/recovery/UnrealText.pdf --type=structure --recovery=true --lang='en'
+```
+
+<a name="4.1"></a>
+
+### 4.1 下载模型
+
+如果输入为英文文档类型，下载OCR检测和识别、版面分析、表格识别的英文模型
+
+```bash
+cd PaddleOCR/ppstructure
+
+# 下载模型
+mkdir inference && cd inference
+# 下载英文超轻量PP-OCRv3检测模型并解压
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_det_infer.tar && tar xf en_PP-OCRv3_det_infer.tar
+# 下载英文超轻量PP-OCRv3识别模型并解压
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/english/en_PP-OCRv3_rec_infer.tar && tar xf en_PP-OCRv3_rec_infer.tar
+# 下载英文表格识别模型并解压
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/en_ppstructure_mobile_v2.0_SLANet_infer.tar
+tar xf en_ppstructure_mobile_v2.0_SLANet_infer.tar
+# 下载英文版面分析模型
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/layout/picodet_lcnet_x1_0_fgd_layout_infer.tar
+tar xf picodet_lcnet_x1_0_fgd_layout_infer.tar
+cd ..
+```
+
+如果输入为中文文档类型，在下述链接中下载中文模型即可：
+
+[PP-OCRv3中英文超轻量文本检测和识别模型](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/README_ch.md#pp-ocr%E7%B3%BB%E5%88%97%E6%A8%A1%E5%9E%8B%E5%88%97%E8%A1%A8%E6%9B%B4%E6%96%B0%E4%B8%AD)、[表格识别模型](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppstructure/docs/models_list.md#22-表格识别模型)、[版面分析模型](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppstructure/docs/models_list.md#1-版面分析模型)
+
+<a name="4.2"></a>
+
+### 4.2 版面恢复
+
+使用下载的模型恢复给定文档的版面，以英文模型为例，执行如下命令：
+
+```bash
+python3 predict_system.py \
+    --image_dir=./docs/table/1.png \
+    --det_model_dir=inference/en_PP-OCRv3_det_infer \
+    --rec_model_dir=inference/en_PP-OCRv3_rec_infer \
+    --rec_char_dict_path=../ppocr/utils/en_dict.txt \
+    --table_model_dir=inference/en_ppstructure_mobile_v2.0_SLANet_infer \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt \
+    --layout_model_dir=inference/picodet_lcnet_x1_0_fgd_layout_infer \
+    --layout_dict_path=../ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt \
+    --vis_font_path=../doc/fonts/simfang.ttf \
+    --recovery=True \
+    --output=../output/
+```
+
+运行完成后，恢复版面的docx文档会保存到`output`字段指定的目录下
+
+字段含义：
+
+- image_dir：测试文件，可以是图片、图片目录、pdf文件、pdf文件目录
+- det_model_dir：OCR检测模型路径
+- rec_model_dir：OCR识别模型路径
+- rec_char_dict_path：OCR识别字典，如果更换为中文模型，需要更改为"../ppocr/utils/ppocr_keys_v1.txt"，如果您在自己的数据集上训练的模型，则更改为训练的字典的文件
+- table_model_dir：表格识别模型路径
+- table_char_dict_path：表格识别字典，如果更换为中文模型，不需要更换字典
+- layout_model_dir：版面分析模型路径
+- layout_dict_path：版面分析字典，如果更换为中文模型，需要更改为"../ppocr/utils/dict/layout_dict/layout_cdla_dict.txt"
+- recovery：是否进行版面恢复，默认False
+- output：版面恢复结果保存路径
+
+<a name="5"></a>
+
+## 5. 更多
+
+关于OCR检测模型的训练评估与推理，请参考：[文本检测教程](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/detection.md)
+
+关于OCR识别模型的训练评估与推理，请参考：[文本识别教程](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/recognition.md)
+
+关于版面分析模型的训练评估与推理，请参考：[版面分析教程](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppstructure/layout/README_ch.md)
+
+关于表格识别模型的训练评估与推理，请参考：[表格识别教程](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/ppstructure/table/README_ch.md)
diff --git a/ppstructure/recovery/__init__.py b/ppstructure/recovery/__init__.py
new file mode 100644
index 0000000..1d11e26
--- /dev/null
+++ b/ppstructure/recovery/__init__.py
@@ -0,0 +1,13 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/ppstructure/recovery/recovery_to_doc.py b/ppstructure/recovery/recovery_to_doc.py
new file mode 100644
index 0000000..9ee3852
--- /dev/null
+++ b/ppstructure/recovery/recovery_to_doc.py
@@ -0,0 +1,155 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+from copy import deepcopy
+
+from docx import Document
+from docx import shared
+from docx.enum.text import WD_ALIGN_PARAGRAPH
+from docx.enum.section import WD_SECTION
+from docx.oxml.ns import qn
+from docx.enum.table import WD_TABLE_ALIGNMENT
+
+from ppstructure.recovery.table_process import HtmlToDocx
+
+from ppocr.utils.logging import get_logger
+
+logger = get_logger()
+
+
+def convert_info_docx(img, res, save_folder, img_name):
+    doc = Document()
+    doc.styles["Normal"].font.name = "Times New Roman"
+    doc.styles["Normal"]._element.rPr.rFonts.set(qn("w:eastAsia"), "宋体")
+    doc.styles["Normal"].font.size = shared.Pt(6.5)
+
+    flag = 1
+    for i, region in enumerate(res):
+        if not region["res"] and region["type"].lower() != "figure":
+            continue
+        img_idx = region["img_idx"]
+        if flag == 2 and region["layout"] == "single":
+            section = doc.add_section(WD_SECTION.CONTINUOUS)
+            section._sectPr.xpath("./w:cols")[0].set(qn("w:num"), "1")
+            flag = 1
+        elif flag == 1 and region["layout"] == "double":
+            section = doc.add_section(WD_SECTION.CONTINUOUS)
+            section._sectPr.xpath("./w:cols")[0].set(qn("w:num"), "2")
+            flag = 2
+
+        if region["type"].lower() == "figure":
+            excel_save_folder = os.path.join(save_folder, img_name)
+            img_path = os.path.join(
+                excel_save_folder, "{}_{}.jpg".format(region["bbox"], img_idx)
+            )
+            paragraph_pic = doc.add_paragraph()
+            paragraph_pic.alignment = WD_ALIGN_PARAGRAPH.CENTER
+            run = paragraph_pic.add_run("")
+            if flag == 1:
+                run.add_picture(img_path, width=shared.Inches(5))
+            elif flag == 2:
+                run.add_picture(img_path, width=shared.Inches(2))
+        elif region["type"].lower() == "title":
+            doc.add_heading(region["res"][0]["text"])
+        elif region["type"].lower() == "table":
+            parser = HtmlToDocx()
+            parser.table_style = "TableGrid"
+            parser.handle_table(region["res"]["html"], doc)
+        elif region["type"] == "equation" and "latex" in region["res"]:
+            pass
+        else:
+            paragraph = doc.add_paragraph()
+            paragraph_format = paragraph.paragraph_format
+            for i, line in enumerate(region["res"]):
+                if i == 0:
+                    paragraph_format.first_line_indent = shared.Inches(0.25)
+                text_run = paragraph.add_run(line["text"] + " ")
+                text_run.font.size = shared.Pt(10)
+
+    # save to docx
+    docx_path = os.path.join(save_folder, "{}_ocr.docx".format(img_name))
+    doc.save(docx_path)
+    logger.info("docx save to {}".format(docx_path))
+
+
+def sorted_layout_boxes(res, w):
+    """
+    Sort text boxes in order from top to bottom, left to right
+    args:
+        res(list):ppstructure results
+    return:
+        sorted results(list)
+    """
+    num_boxes = len(res)
+    if num_boxes == 1:
+        res[0]["layout"] = "single"
+        return res
+
+    sorted_boxes = sorted(res, key=lambda x: (x["bbox"][1], x["bbox"][0]))
+    _boxes = list(sorted_boxes)
+
+    new_res = []
+    res_left = []
+    res_right = []
+    i = 0
+
+    while True:
+        if i >= num_boxes:
+            break
+        if i == num_boxes - 1:
+            if (
+                _boxes[i]["bbox"][1] > _boxes[i - 1]["bbox"][3]
+                and _boxes[i]["bbox"][0] < w / 2
+                and _boxes[i]["bbox"][2] > w / 2
+            ):
+                new_res += res_left
+                new_res += res_right
+                _boxes[i]["layout"] = "single"
+                new_res.append(_boxes[i])
+            else:
+                if _boxes[i]["bbox"][2] > w / 2:
+                    _boxes[i]["layout"] = "double"
+                    res_right.append(_boxes[i])
+                    new_res += res_left
+                    new_res += res_right
+                elif _boxes[i]["bbox"][0] < w / 2:
+                    _boxes[i]["layout"] = "double"
+                    res_left.append(_boxes[i])
+                    new_res += res_left
+                    new_res += res_right
+            res_left = []
+            res_right = []
+            break
+        elif _boxes[i]["bbox"][0] < w / 4 and _boxes[i]["bbox"][2] < 3 * w / 4:
+            _boxes[i]["layout"] = "double"
+            res_left.append(_boxes[i])
+            i += 1
+        elif _boxes[i]["bbox"][0] > w / 4 and _boxes[i]["bbox"][2] > w / 2:
+            _boxes[i]["layout"] = "double"
+            res_right.append(_boxes[i])
+            i += 1
+        else:
+            new_res += res_left
+            new_res += res_right
+            _boxes[i]["layout"] = "single"
+            new_res.append(_boxes[i])
+            res_left = []
+            res_right = []
+            i += 1
+    if res_left:
+        new_res += res_left
+    if res_right:
+        new_res += res_right
+    return new_res
diff --git a/ppstructure/recovery/recovery_to_markdown.py b/ppstructure/recovery/recovery_to_markdown.py
new file mode 100644
index 0000000..8ef3370
--- /dev/null
+++ b/ppstructure/recovery/recovery_to_markdown.py
@@ -0,0 +1,187 @@
+# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import re
+
+from ppocr.utils.logging import get_logger
+
+logger = get_logger()
+
+
+def check_merge_method(in_region):
+    """Select the function to merge paragraph.
+
+    Determine the paragraph merging method based on the positional
+    relationship between the text bbox and the first line of text in the text bbox.
+
+    Args:
+        in_region: Elements with text type in the layout result.
+
+    Returns:
+        Merge the functions of paragraph, convert_text_space_head or convert_text_space_tail.
+    """
+    text_bbox = in_region["bbox"]
+    text_x1 = text_bbox[0]
+    frist_line_box = in_region["res"][0]["text_region"]
+    point_1 = frist_line_box[0]
+    point_2 = frist_line_box[2]
+    frist_line_x1 = point_1[0]
+    frist_line_height = abs(point_2[1] - point_1[1])
+    x1_distance = frist_line_x1 - text_x1
+    return (
+        convert_text_space_head
+        if x1_distance > frist_line_height
+        else convert_text_space_tail
+    )
+
+
+def convert_text_space_head(in_region):
+    """The function to merge paragraph.
+
+    The sign of dividing paragraph is that there are two spaces at the beginning.
+
+    Args:
+        in_region: Elements with text type in the layout result.
+
+    Returns:
+        The text content of the current text box.
+    """
+    text = ""
+    pre_x = None
+    frist_line = True
+    for i, res in enumerate(in_region["res"]):
+        point1 = res["text_region"][0]
+        point2 = res["text_region"][2]
+        h = point2[1] - point1[1]
+
+        if i == 0:
+            text += res["text"]
+            pre_x = point1[0]
+            continue
+
+        x1 = point1[0]
+        if frist_line:
+            if abs(pre_x - x1) < h:
+                text += "\n\n"
+                text += res["text"]
+                frist_line = True
+            else:
+                text += res["text"]
+                frist_line = False
+        else:
+            same_paragh = abs(pre_x - x1) < h
+            if same_paragh:
+                text += res["text"]
+                frist_line = False
+            else:
+                text += "\n\n"
+                text += res["text"]
+                frist_line = True
+        pre_x = x1
+    return text
+
+
+def convert_text_space_tail(in_region):
+    """The function to merge paragraph.
+
+    The symbol for dividing paragraph is a space at the end.
+
+    Args:
+        in_region: Elements with text type in the layout result.
+
+    Returns:
+        The text content of the current text box.
+    """
+    text = ""
+    frist_line = True
+    text_bbox = in_region["bbox"]
+    width = text_bbox[2] - text_bbox[0]
+    for i, res in enumerate(in_region["res"]):
+        point1 = res["text_region"][0]
+        point2 = res["text_region"][2]
+        row_width = point2[0] - point1[0]
+        row_height = point2[1] - point1[1]
+        full_row_threshold = width - row_height
+        is_full = row_width >= full_row_threshold
+
+        if frist_line:
+            text += "\n\n"
+            text += res["text"]
+        else:
+            text += res["text"]
+
+        frist_line = not is_full
+    return text
+
+
+def convert_info_markdown(res, save_folder, img_name):
+    """Save the recognition result as a markdown file.
+
+    Args:
+        res: Recognition result
+        save_folder: Folder to save the markdown file
+        img_name: PDF file or image file name
+
+    Returns:
+        None
+    """
+
+    def replace_special_char(content):
+        special_chars = ["*", "`", "~", "$"]
+        for char in special_chars:
+            content = content.replace(char, "\\" + char)
+        return content
+
+    markdown_string = []
+
+    for i, region in enumerate(res):
+        if not region["res"] and region["type"].lower() != "figure":
+            continue
+        img_idx = region["img_idx"]
+
+        if region["type"].lower() == "figure":
+            img_file_name = "{}_{}.jpg".format(region["bbox"], img_idx)
+            markdown_string.append(
+                f"""<div align="center">\n\t<img src="{img_name+"/"+img_file_name}">\n</div>"""
+            )
+        elif region["type"].lower() == "title":
+            markdown_string.append(
+                f"""# {region['res'][0]['text']}"""
+                + "".join(
+                    [" " + one_region["text"] for one_region in region["res"][1:]]
+                )
+            )
+        elif region["type"].lower() == "table":
+            markdown_string.append(region["res"]["html"])
+        elif region["type"].lower() == "header" or region["type"].lower() == "footer":
+            pass
+        elif region["type"].lower() == "equation" and "latex" in region["res"]:
+            markdown_string.append(f"""$${region["res"]["latex"]}$$""")
+        elif region["type"].lower() == "text":
+            merge_func = check_merge_method(region)
+            # logger.warning(f"use merge method:{merge_func.__name__}")
+            markdown_string.append(replace_special_char(merge_func(region)))
+        else:
+            string = ""
+            for line in region["res"]:
+                string += line["text"] + " "
+            markdown_string.append(string)
+
+    md_path = os.path.join(save_folder, "{}_ocr.md".format(img_name))
+    markdown_string = "\n\n".join(markdown_string)
+    markdown_string = re.sub(r"\n{3,}", "\n\n", markdown_string)
+    with open(md_path, "w", encoding="utf-8") as f:
+        f.write(markdown_string)
+    logger.info("markdown save to {}".format(md_path))
diff --git a/ppstructure/recovery/requirements.txt b/ppstructure/recovery/requirements.txt
new file mode 100644
index 0000000..86ec03e
--- /dev/null
+++ b/ppstructure/recovery/requirements.txt
@@ -0,0 +1,4 @@
+python-docx
+beautifulsoup4
+fonttools>=4.43.0
+fire>=0.3.0
diff --git a/ppstructure/recovery/table_process.py b/ppstructure/recovery/table_process.py
new file mode 100644
index 0000000..f5e01e6
--- /dev/null
+++ b/ppstructure/recovery/table_process.py
@@ -0,0 +1,325 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from: https://github.com/weizwx/html2docx/blob/master/htmldocx/h2d.py
+"""
+
+import re
+import docx
+from docx import Document
+from bs4 import BeautifulSoup
+from html.parser import HTMLParser
+
+
+def get_table_rows(table_soup):
+    table_row_selectors = [
+        "table > tr",
+        "table > thead > tr",
+        "table > tbody > tr",
+        "table > tfoot > tr",
+    ]
+    # If there's a header, body, footer or direct child tr tags, add row dimensions from there
+    return table_soup.select(", ".join(table_row_selectors), recursive=False)
+
+
+def get_table_columns(row):
+    # Get all columns for the specified row tag.
+    return row.find_all(["th", "td"], recursive=False) if row else []
+
+
+def get_table_dimensions(table_soup):
+    # Get rows for the table
+    rows = get_table_rows(table_soup)
+    # Table is either empty or has non-direct children between table and tr tags
+    # Thus the row dimensions and column dimensions are assumed to be 0
+
+    cols = get_table_columns(rows[0]) if rows else []
+    # Add colspan calculation column number
+    col_count = 0
+    for col in cols:
+        colspan = col.attrs.get("colspan", 1)
+        col_count += int(colspan)
+
+    return rows, col_count
+
+
+def get_cell_html(soup):
+    # Returns string of td element with opening and closing <td> tags removed
+    # Cannot use find_all as it only finds element tags and does not find text which
+    # is not inside an element
+    return " ".join([str(i) for i in soup.contents])
+
+
+def delete_paragraph(paragraph):
+    # https://github.com/python-openxml/python-docx/issues/33#issuecomment-77661907
+    p = paragraph._element
+    p.getparent().remove(p)
+    p._p = p._element = None
+
+
+def remove_whitespace(string, leading=False, trailing=False):
+    """Remove white space from a string.
+    Args:
+        string(str): The string to remove white space from.
+        leading(bool, optional): Remove leading new lines when True.
+        trailing(bool, optional): Remove trailing new lines when False.
+    Returns:
+        str: The input string with new line characters removed and white space squashed.
+    Examples:
+        Single or multiple new line characters are replaced with space.
+            >>> remove_whitespace("abc\\ndef")
+            'abc def'
+            >>> remove_whitespace("abc\\n\\n\\ndef")
+            'abc def'
+        New line characters surrounded by white space are replaced with a single space.
+            >>> remove_whitespace("abc \\n \\n \\n def")
+            'abc def'
+            >>> remove_whitespace("abc  \\n  \\n  \\n  def")
+            'abc def'
+        Leading and trailing new lines are replaced with a single space.
+            >>> remove_whitespace("\\nabc")
+            ' abc'
+            >>> remove_whitespace("  \\n  abc")
+            ' abc'
+            >>> remove_whitespace("abc\\n")
+            'abc '
+            >>> remove_whitespace("abc  \\n  ")
+            'abc '
+        Use ``leading=True`` to remove leading new line characters, including any surrounding
+        white space:
+            >>> remove_whitespace("\\nabc", leading=True)
+            'abc'
+            >>> remove_whitespace("  \\n  abc", leading=True)
+            'abc'
+        Use ``trailing=True`` to remove trailing new line characters, including any surrounding
+        white space:
+            >>> remove_whitespace("abc  \\n  ", trailing=True)
+            'abc'
+    """
+    # Remove any leading new line characters along with any surrounding white space
+    if leading:
+        string = re.sub(r"^\s*\n+\s*", "", string)
+
+    # Remove any trailing new line characters along with any surrounding white space
+    if trailing:
+        string = re.sub(r"\s*\n+\s*$", "", string)
+
+    # Replace new line characters and absorb any surrounding space.
+    string = re.sub(r"\s*\n\s*", " ", string)
+    # TODO need some way to get rid of extra spaces in e.g. text <span>   </span>  text
+    return re.sub(r"\s+", " ", string)
+
+
+font_styles = {
+    "b": "bold",
+    "strong": "bold",
+    "em": "italic",
+    "i": "italic",
+    "u": "underline",
+    "s": "strike",
+    "sup": "superscript",
+    "sub": "subscript",
+    "th": "bold",
+}
+
+font_names = {
+    "code": "Courier",
+    "pre": "Courier",
+}
+
+
+class HtmlToDocx(HTMLParser):
+    def __init__(self):
+        super().__init__()
+        self.options = {
+            "fix-html": True,
+            "images": True,
+            "tables": True,
+            "styles": True,
+        }
+        self.table_row_selectors = [
+            "table > tr",
+            "table > thead > tr",
+            "table > tbody > tr",
+            "table > tfoot > tr",
+        ]
+        self.table_style = None
+        self.paragraph_style = None
+
+    def set_initial_attrs(self, document=None):
+        self.tags = {
+            "span": [],
+            "list": [],
+        }
+        if document:
+            self.doc = document
+        else:
+            self.doc = Document()
+        self.bs = self.options["fix-html"]  # whether or not to clean with BeautifulSoup
+        self.document = self.doc
+        self.include_tables = True  # TODO add this option back in?
+        self.include_images = self.options["images"]
+        self.include_styles = self.options["styles"]
+        self.paragraph = None
+        self.skip = False
+        self.skip_tag = None
+        self.instances_to_skip = 0
+
+    def copy_settings_from(self, other):
+        """Copy settings from another instance of HtmlToDocx"""
+        self.table_style = other.table_style
+        self.paragraph_style = other.paragraph_style
+
+    def ignore_nested_tables(self, tables_soup):
+        """
+        Returns array containing only the highest level tables
+        Operates on the assumption that bs4 returns child elements immediately after
+        the parent element in `find_all`. If this changes in the future, this method will need to be updated
+        :return:
+        """
+        new_tables = []
+        nest = 0
+        for table in tables_soup:
+            if nest:
+                nest -= 1
+                continue
+            new_tables.append(table)
+            nest = len(table.find_all("table"))
+        return new_tables
+
+    def get_tables(self):
+        if not hasattr(self, "soup"):
+            self.include_tables = False
+            return
+            # find other way to do it, or require this dependency?
+        self.tables = self.ignore_nested_tables(self.soup.find_all("table"))
+        self.table_no = 0
+
+    def run_process(self, html):
+        if self.bs and BeautifulSoup:
+            self.soup = BeautifulSoup(html, "html.parser")
+            html = str(self.soup)
+        if self.include_tables:
+            self.get_tables()
+        self.feed(html)
+
+    def add_html_to_cell(self, html, cell):
+        if not isinstance(cell, docx.table._Cell):
+            raise ValueError("Second argument needs to be a %s" % docx.table._Cell)
+        unwanted_paragraph = cell.paragraphs[0]
+        if unwanted_paragraph.text == "":
+            delete_paragraph(unwanted_paragraph)
+        self.set_initial_attrs(cell)
+        self.run_process(html)
+        # cells must end with a paragraph or will get message about corrupt file
+        # https://stackoverflow.com/a/29287121
+        if not self.doc.paragraphs:
+            self.doc.add_paragraph("")
+
+    def apply_paragraph_style(self, style=None):
+        try:
+            if style:
+                self.paragraph.style = style
+            elif self.paragraph_style:
+                self.paragraph.style = self.paragraph_style
+        except KeyError as e:
+            raise ValueError(f"Unable to apply style {self.paragraph_style}.") from e
+
+    def handle_table(self, html, doc):
+        """
+        To handle nested tables, we will parse tables manually as follows:
+        Get table soup
+        Create docx table
+        Iterate over soup and fill docx table with new instances of this parser
+        Tell HTMLParser to ignore any tags until the corresponding closing table tag
+        """
+        table_soup = BeautifulSoup(html, "html.parser")
+        rows, cols_len = get_table_dimensions(table_soup)
+        table = doc.add_table(len(rows), cols_len)
+        table.style = doc.styles["Table Grid"]
+
+        num_rows = len(table.rows)
+        num_cols = len(table.columns)
+
+        cell_row = 0
+        for index, row in enumerate(rows):
+            cols = get_table_columns(row)
+            cell_col = 0
+            for col in cols:
+                colspan = int(col.attrs.get("colspan", 1))
+                rowspan = int(col.attrs.get("rowspan", 1))
+
+                cell_html = get_cell_html(col)
+                if col.name == "th":
+                    cell_html = "<b>%s</b>" % cell_html
+
+                if cell_row >= num_rows or cell_col >= num_cols:
+                    continue
+
+                docx_cell = table.cell(cell_row, cell_col)
+
+                while docx_cell.text != "":  # Skip the merged cell
+                    cell_col += 1
+                    docx_cell = table.cell(cell_row, cell_col)
+
+                cell_to_merge = table.cell(
+                    cell_row + rowspan - 1, cell_col + colspan - 1
+                )
+                if docx_cell != cell_to_merge:
+                    docx_cell.merge(cell_to_merge)
+
+                child_parser = HtmlToDocx()
+                child_parser.copy_settings_from(self)
+                child_parser.add_html_to_cell(cell_html or " ", docx_cell)
+
+                cell_col += colspan
+            cell_row += 1
+
+    def handle_data(self, data):
+        if self.skip:
+            return
+
+        # Only remove white space if we're not in a pre block.
+        if "pre" not in self.tags:
+            # remove leading and trailing whitespace in all instances
+            data = remove_whitespace(data, True, True)
+
+        if not self.paragraph:
+            self.paragraph = self.doc.add_paragraph()
+            self.apply_paragraph_style()
+
+        # There can only be one nested link in a valid html document
+        # You cannot have interactive content in an A tag, this includes links
+        # https://html.spec.whatwg.org/#interactive-content
+        link = self.tags.get("a")
+        if link:
+            self.handle_link(link["href"], data)
+        else:
+            # If there's a link, dont put the data directly in the run
+            self.run = self.paragraph.add_run(data)
+            spans = self.tags["span"]
+            for span in spans:
+                if "style" in span:
+                    style = self.parse_dict_string(span["style"])
+                    self.add_styles_to_run(style)
+
+            # add font style and name
+            for tag in self.tags:
+                if tag in font_styles:
+                    font_style = font_styles[tag]
+                    setattr(self.run.font, font_style, True)
+
+                if tag in font_names:
+                    font_name = font_names[tag]
+                    self.run.font.name = font_name
diff --git a/ppstructure/table/README.md b/ppstructure/table/README.md
new file mode 100644
index 0000000..a80514f
--- /dev/null
+++ b/ppstructure/table/README.md
@@ -0,0 +1,159 @@
+English | [简体中文](README_ch.md)
+
+# Table Recognition
+
+- [1. pipeline](#1-pipeline)
+- [2. Performance](#2-performance)
+- [3. Result](#3-result)
+- [4. How to use](#4-how-to-use)
+  - [4.1 Quick start](#41-quick-start)
+  - [4.2 Training, Evaluation and Inference](#42-training-evaluation-and-inference)
+  - [4.3 Calculate TEDS](#43-calculate-teds)
+- [5. Reference](#5-reference)
+
+
+## 1. pipeline
+The table recognition mainly contains three models
+1. Single line text detection-DB
+2. Single line text recognition-CRNN
+3. Table structure and cell coordinate prediction-SLANet
+
+The table recognition flow chart is as follows
+
+![tableocr_pipeline](../docs/table/tableocr_pipeline_en.jpg)
+
+1. The coordinates of single-line text is detected by DB model, and then sends it to the recognition model to get the recognition result.
+2. The table structure and cell coordinates is predicted by SLANet model.
+3. The recognition result of the cell is combined by the coordinates, recognition result of the single line and the coordinates of the cell.
+4. The cell recognition result and the table structure together construct the html string of the table.
+
+## 2. Performance
+We evaluated the algorithm on the PubTabNet<sup>[1]</sup> eval dataset, and the performance is as follows:
+
+|Method|Acc|[TEDS(Tree-Edit-Distance-based Similarity)](https://github.com/ibm-aur-nlp/PubTabNet/tree/master/src)|Speed|
+| --- | --- | --- | ---|
+| EDD<sup>[2]</sup> |x| 88.30% |x|
+| TableRec-RARE(ours) | 71.73%| 93.88% |779ms|
+| SLANet(ours) | 76.31%|    95.89%|766ms|
+
+The performance indicators are explained as follows:
+- Acc: The accuracy of the table structure in each image, a wrong token is considered an error.
+- TEDS: The accuracy of the model's restoration of table information. This indicator evaluates not only the table structure, but also the text content in the table.
+- Speed: The inference speed of a single image when the model runs on the CPU machine and MKL is enabled.
+
+## 3. Result
+
+![](../docs/imgs/table_ch_result1.jpg)
+![](../docs/imgs/table_ch_result2.jpg)
+![](../docs/imgs/table_ch_result3.jpg)
+
+## 4. How to use
+
+### 4.1 Quick start
+
+PP-Structure currently provides table recognition models in both Chinese and English. For the model link, see [models_list](../docs/models_list.md). The whl package is also provided for quick use, see [quickstart](../docs/quickstart_en.md) for details.
+
+The following takes the Chinese table recognition model as an example to introduce how to recognize a table.
+
+Use the following commands to quickly complete the identification of a table.
+
+```python
+cd PaddleOCR/ppstructure
+
+# download model
+mkdir inference && cd inference
+# Download the PP-OCRv3 text detection model and unzip it
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar && tar xf PP-OCRv3_mobile_det_infer.tar
+# Download the PP-OCRv3 text recognition model and unzip it
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_infer.tar && tar xf ch_PP-OCRv3_rec_infer.tar
+# Download the PP-StructureV2 form recognition model and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/ch_ppstructure_mobile_v2.0_SLANet_infer.tar && tar xf ch_ppstructure_mobile_v2.0_SLANet_infer.tar
+cd ..
+# run
+python3 table/predict_table.py \
+    --det_model_dir=inference/PP-OCRv3_mobile_det_infer \
+    --rec_model_dir=inference/ch_PP-OCRv3_rec_infer  \
+    --table_model_dir=inference/ch_ppstructure_mobile_v2.0_SLANet_infer \
+    --rec_char_dict_path=../ppocr/utils/ppocr_keys_v1.txt \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict_ch.txt \
+    --image_dir=docs/table/table.jpg \
+    --output=../output/table
+
+```
+
+After the operation is completed, the excel table of each image will be saved to the directory specified by the output field, and an html file will be produced in the directory to visually view the cell coordinates and the recognized table.
+
+**NOTE**
+1. If you want to use the English table recognition model, you need to download the English text detection and recognition model and the English table recognition model in [models_list](../docs/models_list_en.md), and replace `table_structure_dict_ch.txt` with `table_structure_dict.txt`.
+2. To use the TableRec-RARE model, you need to replace `table_structure_dict_ch.txt` with `table_structure_dict.txt`, and add parameter `--merge_no_span_structure=False`
+
+### 4.2 Training, Evaluation and Inference
+
+The training, evaluation and inference process of the text detection model can be referred to [detection](../../doc/doc_en/detection_en.md)
+
+The training, evaluation and inference process of the text recognition model can be referred to [recognition](../../doc/doc_en/recognition_en.md)
+
+The training, evaluation and inference process of the table recognition model can be referred to [table_recognition](../../doc/doc_en/table_recognition_en.md)
+
+### 4.3 Calculate TEDS
+
+The table uses [TEDS(Tree-Edit-Distance-based Similarity)](https://github.com/ibm-aur-nlp/PubTabNet/tree/master/src) as the evaluation metric of the model. Before the model evaluation, the three models in the pipeline need to be exported as inference models (we have provided them), and the gt for evaluation needs to be prepared. Examples of gt are as follows:
+```txt
+PMC5755158_010_01.png    <html><body><table><thead><tr><td></td><td><b>Weaning</b></td><td><b>Week 15</b></td><td><b>Off-test</b></td></tr></thead><tbody><tr><td>Weaning</td><td>–</td><td>–</td><td>–</td></tr><tr><td>Week 15</td><td>–</td><td>0.17 ± 0.08</td><td>0.16 ± 0.03</td></tr><tr><td>Off-test</td><td>–</td><td>0.80 ± 0.24</td><td>0.19 ± 0.09</td></tr></tbody></table></body></html>
+```
+Each line in gt consists of the file name and the html string of the table. The file name and the html string of the table are separated by `\t`.
+
+You can also use the following command to generate an evaluation gt file from the annotation file:
+```python
+python3 ppstructure/table/convert_label2html.py --ori_gt_path /path/to/your_label_file --save_path /path/to/save_file
+```
+
+Use the following command to evaluate. After the evaluation is completed, the teds indicator will be output.
+```python
+python3 table/eval_table.py \
+    --det_model_dir=path/to/det_model_dir \
+    --rec_model_dir=path/to/rec_model_dir \
+    --table_model_dir=path/to/table_model_dir \
+    --image_dir=docs/table/table.jpg \
+    --rec_char_dict_path=../ppocr/utils/dict/table_dict.txt \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt \
+    --det_limit_side_len=736 \
+    --det_limit_type=min \
+    --gt_path=path/to/gt.txt
+```
+
+Evaluate on the PubLatNet dataset using the English model
+
+```bash
+cd PaddleOCR/ppstructure
+# Download the model
+mkdir inference && cd inference
+# Download the text detection model trained on the PubTabNet dataset and unzip it
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar
+# Download the text recognition model trained on the PubTabNet dataset and unzip it
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar
+# Download the table recognition model trained on the PubTabNet dataset and unzip it
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/en_ppstructure_mobile_v2.0_SLANet_infer.tar && tar xf en_ppstructure_mobile_v2.0_SLANet_infer.tar
+cd ..
+
+python3 table/eval_table.py \
+    --det_model_dir=inference/en_ppocr_mobile_v2.0_table_det_infer \
+    --rec_model_dir=inference/en_ppocr_mobile_v2.0_table_rec_infer \
+    --table_model_dir=inference/en_ppstructure_mobile_v2.0_SLANet_infer \
+    --image_dir=train_data/table/pubtabnet/val/ \
+    --rec_char_dict_path=../ppocr/utils/dict/table_dict.txt \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt \
+    --det_limit_side_len=736 \
+    --det_limit_type=min \
+    --rec_image_shape=3,32,320 \
+    --gt_path=path/to/gt.txt
+```
+
+output is
+```bash
+teds: 95.89
+```
+
+## 5. Reference
+1. https://github.com/ibm-aur-nlp/PubTabNet
+2. https://arxiv.org/pdf/1911.10683
diff --git a/ppstructure/table/README_ch.md b/ppstructure/table/README_ch.md
new file mode 100644
index 0000000..2211f79
--- /dev/null
+++ b/ppstructure/table/README_ch.md
@@ -0,0 +1,163 @@
+[English](README.md) | 简体中文
+
+# 表格识别
+
+- [1. 表格识别 pipeline](#1-表格识别-pipeline)
+- [2. 性能](#2-性能)
+- [3. 效果演示](#3-效果演示)
+- [4. 使用](#4-使用)
+  - [4.1 快速开始](#41-快速开始)
+  - [4.2 模型训练、评估与推理](#42-模型训练评估与推理)
+  - [4.3 计算TEDS](#43-计算teds)
+- [5. Reference](#5-reference)
+
+
+## 1. 表格识别 pipeline
+
+表格识别主要包含三个模型
+1. 单行文本检测-DB
+2. 单行文本识别-CRNN
+3. 表格结构和cell坐标预测-SLANet
+
+具体流程图如下
+
+![tableocr_pipeline](../docs/table/tableocr_pipeline.jpg)
+
+流程说明:
+
+1. 图片由单行文字检测模型检测到单行文字的坐标，然后送入识别模型拿到识别结果。
+2. 图片由SLANet模型拿到表格的结构信息和单元格的坐标信息。
+3. 由单行文字的坐标、识别结果和单元格的坐标一起组合出单元格的识别结果。
+4. 单元格的识别结果和表格结构一起构造表格的html字符串。
+
+
+## 2. 性能
+
+我们在 PubTabNet<sup>[1]</sup> 评估数据集上对算法进行了评估，性能如下
+
+
+|算法|Acc|[TEDS(Tree-Edit-Distance-based Similarity)](https://github.com/ibm-aur-nlp/PubTabNet/tree/master/src)|Speed|
+| --- | --- | --- | ---|
+| EDD<sup>[2]</sup> |x| 88.30% |x|
+| TableRec-RARE(ours) | 71.73%| 93.88% |779ms|
+| SLANet(ours) |76.31%| 95.89%|766ms|
+
+性能指标解释如下：
+- Acc: 模型对每张图像里表格结构的识别准确率，错一个token就算错误。
+- TEDS: 模型对表格信息还原的准确度，此指标评价内容不仅包含表格结构，还包含表格内的文字内容。
+- Speed: 模型在CPU机器上，开启MKL的情况下，单张图片的推理速度。
+
+## 3. 效果演示
+
+![](../docs/imgs/table_ch_result1.jpg)
+![](../docs/imgs/table_ch_result2.jpg)
+![](../docs/imgs/table_ch_result3.jpg)
+
+## 4. 使用
+
+### 4.1 快速开始
+
+PP-Structure目前提供了中英文两种语言的表格识别模型，模型链接见 [models_list](../docs/models_list.md)。也提供了whl包的形式方便快速使用，详见 [quickstart](../docs/quickstart.md)。
+
+下面以中文表格识别模型为例，介绍如何识别一张表格。
+
+使用如下命令即可快速完成一张表格的识别。
+```python
+cd PaddleOCR/ppstructure
+
+# 下载模型
+mkdir inference && cd inference
+# 下载PP-OCRv3文本检测模型并解压
+wget https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar && tar xf PP-OCRv3_mobile_det_infer.tar
+# 下载PP-OCRv3文本识别模型并解压
+wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_infer.tar && tar xf ch_PP-OCRv3_rec_infer.tar
+# 下载PP-StructureV2中文表格识别模型并解压
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/ch_ppstructure_mobile_v2.0_SLANet_infer.tar && tar xf ch_ppstructure_mobile_v2.0_SLANet_infer.tar
+cd ..
+# 执行表格识别
+python table/predict_table.py \
+    --det_model_dir=inference/PP-OCRv3_mobile_det_infer \
+    --rec_model_dir=inference/ch_PP-OCRv3_rec_infer  \
+    --table_model_dir=inference/ch_ppstructure_mobile_v2.0_SLANet_infer \
+    --rec_char_dict_path=../ppocr/utils/ppocr_keys_v1.txt \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict_ch.txt \
+    --image_dir=docs/table/table.jpg \
+    --output=../output/table
+```
+运行完成后，每张图片的excel表格会保存到output字段指定的目录下，同时在该目录下回生产一个html文件，用于可视化查看单元格坐标和识别的表格。
+
+**NOTE**
+1. 如果想使用英文模型，需要在 [models_list](../docs/models_list.md) 中下载英文文字检测识别模型和英文表格识别模型，同时替换`table_structure_dict_ch.txt`为`table_structure_dict.txt`即可。
+2. 如需使用TableRec-RARE模型，需要替换`table_structure_dict_ch.txt`为`table_structure_dict.txt`，同时参数`--merge_no_span_structure=False`
+
+### 4.2 模型训练、评估与推理
+
+文本检测模型的训练、评估和推理流程可参考 [detection](../../doc/doc_ch/detection.md)
+
+文本识别模型的训练、评估和推理流程可参考 [recognition](../../doc/doc_ch/recognition.md)
+
+表格识别模型的训练、评估和推理流程可参考 [table_recognition](../../doc/doc_ch/table_recognition.md)
+
+### 4.3 计算TEDS
+
+表格使用 [TEDS(Tree-Edit-Distance-based Similarity)](https://github.com/ibm-aur-nlp/PubTabNet/tree/master/src) 作为模型的评估指标。在进行模型评估之前，需要将pipeline中的三个模型分别导出为inference模型(我们已经提供好)，还需要准备评估的gt， gt示例如下:
+```txt
+PMC5755158_010_01.png    <html><body><table><thead><tr><td></td><td><b>Weaning</b></td><td><b>Week 15</b></td><td><b>Off-test</b></td></tr></thead><tbody><tr><td>Weaning</td><td>–</td><td>–</td><td>–</td></tr><tr><td>Week 15</td><td>–</td><td>0.17 ± 0.08</td><td>0.16 ± 0.03</td></tr><tr><td>Off-test</td><td>–</td><td>0.80 ± 0.24</td><td>0.19 ± 0.09</td></tr></tbody></table></body></html>
+```
+gt每一行都由文件名和表格的html字符串组成，文件名和表格的html字符串之间使用`\t`分隔。
+
+也可使用如下命令，由标注文件生成评估的gt文件：
+```python
+python3 ppstructure/table/convert_label2html.py --ori_gt_path /path/to/your_label_file --save_path /path/to/save_file
+```
+
+准备完成后使用如下命令进行评估，评估完成后会输出teds指标。
+```python
+cd PaddleOCR/ppstructure
+python3 table/eval_table.py \
+    --det_model_dir=path/to/det_model_dir \
+    --rec_model_dir=path/to/rec_model_dir \
+    --table_model_dir=path/to/table_model_dir \
+    --image_dir=docs/table/table.jpg \
+    --rec_char_dict_path=../ppocr/utils/dict/table_dict.txt \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt \
+    --det_limit_side_len=736 \
+    --det_limit_type=min \
+    --gt_path=path/to/gt.txt
+```
+
+如使用英文表格识别模型在PubLatNet数据集上进行评估
+
+```bash
+cd PaddleOCR/ppstructure
+# 下载模型
+mkdir inference && cd inference
+# 下载基于PubTabNet数据集训练的文本检测模型并解压
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar
+# 下载基于PubTabNet数据集训练的文本识别模型并解压
+wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar
+# 下载基于PubTabNet数据集训练的表格识别模型并解压
+wget https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/en_ppstructure_mobile_v2.0_SLANet_infer.tar && tar xf en_ppstructure_mobile_v2.0_SLANet_infer.tar
+cd ..
+
+python3 table/eval_table.py \
+    --det_model_dir=inference/en_ppocr_mobile_v2.0_table_det_infer \
+    --rec_model_dir=inference/en_ppocr_mobile_v2.0_table_rec_infer \
+    --table_model_dir=inference/en_ppstructure_mobile_v2.0_SLANet_infer \
+    --image_dir=train_data/table/pubtabnet/val/ \
+    --rec_char_dict_path=../ppocr/utils/dict/table_dict.txt \
+    --table_char_dict_path=../ppocr/utils/dict/table_structure_dict.txt \
+    --det_limit_side_len=736 \
+    --det_limit_type=min \
+    --rec_image_shape=3,32,320 \
+    --gt_path=path/to/gt.txt
+```
+
+将会输出
+```bash
+teds: 95.89
+```
+
+## 5. Reference
+1. https://github.com/ibm-aur-nlp/PubTabNet
+2. https://arxiv.org/pdf/1911.10683
diff --git a/ppstructure/table/__init__.py b/ppstructure/table/__init__.py
new file mode 100644
index 0000000..1d11e26
--- /dev/null
+++ b/ppstructure/table/__init__.py
@@ -0,0 +1,13 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/ppstructure/table/convert_label2html.py b/ppstructure/table/convert_label2html.py
new file mode 100644
index 0000000..91611dd
--- /dev/null
+++ b/ppstructure/table/convert_label2html.py
@@ -0,0 +1,102 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+convert table label to html
+"""
+
+import json
+import argparse
+from tqdm import tqdm
+
+
+def save_pred_txt(key, val, tmp_file_path):
+    with open(tmp_file_path, "a+", encoding="utf-8") as f:
+        f.write("{}\t{}\n".format(key, val))
+
+
+def skip_char(text, sp_char_list):
+    """
+    skip empty cell
+    @param text: text in cell
+    @param sp_char_list: style char and special code
+    @return:
+    """
+    for sp_char in sp_char_list:
+        text = text.replace(sp_char, "")
+    return text
+
+
+def gen_html(img):
+    """
+    Formats HTML code from tokenized annotation of img
+    """
+    html_code = img["html"]["structure"]["tokens"].copy()
+    to_insert = [i for i, tag in enumerate(html_code) if tag in ("<td>", ">")]
+    for i, cell in zip(to_insert[::-1], img["html"]["cells"][::-1]):
+        if cell["tokens"]:
+            text = "".join(cell["tokens"])
+            # skip empty text
+            sp_char_list = ["<b>", "</b>", "\u2028", " ", "<i>", "</i>"]
+            text_remove_style = skip_char(text, sp_char_list)
+            if len(text_remove_style) == 0:
+                continue
+            html_code.insert(i + 1, text)
+    html_code = "".join(html_code)
+    html_code = "<html><body><table>{}</table></body></html>".format(html_code)
+    return html_code
+
+
+def load_gt_data(gt_path):
+    """
+    load gt
+    @param gt_path:
+    @return:
+    """
+    data_list = {}
+    with open(gt_path, "rb") as f:
+        lines = f.readlines()
+        for line in tqdm(lines):
+            data_line = line.decode("utf-8").strip("\n")
+            info = json.loads(data_line)
+            data_list[info["filename"]] = info
+    return data_list
+
+
+def convert(origin_gt_path, save_path):
+    """
+    gen html from label file
+    @param origin_gt_path:
+    @param save_path:
+    @return:
+    """
+    data_dict = load_gt_data(origin_gt_path)
+    for img_name, gt in tqdm(data_dict.items()):
+        html = gen_html(gt)
+        save_pred_txt(img_name, html, save_path)
+    print("convert finish")
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(description="args for paddleserving")
+    parser.add_argument("--ori_gt_path", type=str, required=True, help="label gt path")
+    parser.add_argument(
+        "--save_path", type=str, required=True, help="path to save file"
+    )
+    args = parser.parse_args()
+    return args
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    convert(args.ori_gt_path, args.save_path)
diff --git a/ppstructure/table/eval_table.py b/ppstructure/table/eval_table.py
new file mode 100755
index 0000000..b9e4661
--- /dev/null
+++ b/ppstructure/table/eval_table.py
@@ -0,0 +1,107 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "../..")))
+
+import cv2
+import pickle
+import paddle
+from tqdm import tqdm
+from ppstructure.table.table_metric import TEDS
+from ppstructure.table.predict_table import TableSystem
+from ppstructure.utility import init_args
+from ppocr.utils.logging import get_logger
+
+logger = get_logger()
+
+
+def parse_args():
+    parser = init_args()
+    parser.add_argument("--gt_path", type=str)
+    return parser.parse_args()
+
+
+def load_txt(txt_path):
+    pred_html_dict = {}
+    if not os.path.exists(txt_path):
+        return pred_html_dict
+    with open(txt_path, encoding="utf-8") as f:
+        lines = f.readlines()
+        for line in lines:
+            line = line.strip().split("\t")
+            img_name, pred_html = line
+            pred_html_dict[img_name] = pred_html
+    return pred_html_dict
+
+
+def load_result(path):
+    data = {}
+    if os.path.exists(path):
+        data = pickle.load(open(path, "rb"))
+    return data
+
+
+def save_result(path, data):
+    old_data = load_result(path)
+    old_data.update(data)
+    with open(path, "wb") as f:
+        pickle.dump(old_data, f)
+
+
+def main(gt_path, img_root, args):
+    os.makedirs(args.output, exist_ok=True)
+    # init TableSystem
+    text_sys = TableSystem(args)
+    # load gt and preds html result
+    gt_html_dict = load_txt(gt_path)
+
+    ocr_result = load_result(os.path.join(args.output, "ocr.pickle"))
+    structure_result = load_result(os.path.join(args.output, "structure.pickle"))
+
+    pred_htmls = []
+    gt_htmls = []
+    for img_name, gt_html in tqdm(gt_html_dict.items()):
+        img = cv2.imread(os.path.join(img_root, img_name))
+        # run ocr and save result
+        if img_name not in ocr_result:
+            dt_boxes, rec_res, _, _ = text_sys._ocr(img)
+            ocr_result[img_name] = [dt_boxes, rec_res]
+            save_result(os.path.join(args.output, "ocr.pickle"), ocr_result)
+        # run structure and save result
+        if img_name not in structure_result:
+            structure_res, _ = text_sys._structure(img)
+            structure_result[img_name] = structure_res
+            save_result(os.path.join(args.output, "structure.pickle"), structure_result)
+        dt_boxes, rec_res = ocr_result[img_name]
+        structure_res = structure_result[img_name]
+        # match ocr and structure
+        pred_html = text_sys.match(structure_res, dt_boxes, rec_res)
+
+        pred_htmls.append(pred_html)
+        gt_htmls.append(gt_html)
+
+    # compute teds
+    teds = TEDS(n_jobs=16)
+    scores = teds.batch_evaluate_html(gt_htmls, pred_htmls)
+    logger.info("teds: {}".format(sum(scores) / len(scores)))
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    main(args.gt_path, args.image_dir, args)
diff --git a/ppstructure/table/matcher.py b/ppstructure/table/matcher.py
new file mode 100755
index 0000000..ae32b4b
--- /dev/null
+++ b/ppstructure/table/matcher.py
@@ -0,0 +1,206 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import numpy as np
+from ppstructure.table.table_master_match import deal_eb_token, deal_bb
+import html
+
+
+def distance(box_1, box_2):
+    x1, y1, x2, y2 = box_1
+    x3, y3, x4, y4 = box_2
+    dis = abs(x3 - x1) + abs(y3 - y1) + abs(x4 - x2) + abs(y4 - y2)
+    dis_2 = abs(x3 - x1) + abs(y3 - y1)
+    dis_3 = abs(x4 - x2) + abs(y4 - y2)
+    return dis + min(dis_2, dis_3)
+
+
+def compute_iou(rec1, rec2):
+    """
+    computing IoU
+    :param rec1: (y0, x0, y1, x1), which reflects
+            (top, left, bottom, right)
+    :param rec2: (y0, x0, y1, x1)
+    :return: scala value of IoU
+    """
+    # computing area of each rectangles
+    S_rec1 = (rec1[2] - rec1[0]) * (rec1[3] - rec1[1])
+    S_rec2 = (rec2[2] - rec2[0]) * (rec2[3] - rec2[1])
+
+    # computing the sum_area
+    sum_area = S_rec1 + S_rec2
+
+    # find the each edge of intersect rectangle
+    left_line = max(rec1[1], rec2[1])
+    right_line = min(rec1[3], rec2[3])
+    top_line = max(rec1[0], rec2[0])
+    bottom_line = min(rec1[2], rec2[2])
+
+    # judge if there is an intersect
+    if left_line >= right_line or top_line >= bottom_line:
+        return 0.0
+    else:
+        intersect = (right_line - left_line) * (bottom_line - top_line)
+        return (intersect / (sum_area - intersect)) * 1.0
+
+
+class TableMatch:
+    def __init__(self, filter_ocr_result=False, use_master=False):
+        self.filter_ocr_result = filter_ocr_result
+        self.use_master = use_master
+
+    def __call__(self, structure_res, dt_boxes, rec_res):
+        pred_structures, pred_bboxes = structure_res
+        if self.filter_ocr_result:
+            dt_boxes, rec_res = self._filter_ocr_result(pred_bboxes, dt_boxes, rec_res)
+        matched_index = self.match_result(dt_boxes, pred_bboxes)
+        if self.use_master:
+            pred_html, pred = self.get_pred_html_master(
+                pred_structures, matched_index, rec_res
+            )
+        else:
+            pred_html, pred = self.get_pred_html(
+                pred_structures, matched_index, rec_res
+            )
+        return pred_html
+
+    def match_result(self, dt_boxes, pred_bboxes):
+        matched = {}
+        for i, gt_box in enumerate(dt_boxes):
+            distances = []
+            for j, pred_box in enumerate(pred_bboxes):
+                if len(pred_box) == 8:
+                    pred_box = [
+                        np.min(pred_box[0::2]),
+                        np.min(pred_box[1::2]),
+                        np.max(pred_box[0::2]),
+                        np.max(pred_box[1::2]),
+                    ]
+                distances.append(
+                    (distance(gt_box, pred_box), 1.0 - compute_iou(gt_box, pred_box))
+                )  # compute iou and l1 distance
+            sorted_distances = distances.copy()
+            # select det box by iou and l1 distance
+            sorted_distances = sorted(
+                sorted_distances, key=lambda item: (item[1], item[0])
+            )
+            if distances.index(sorted_distances[0]) not in matched.keys():
+                matched[distances.index(sorted_distances[0])] = [i]
+            else:
+                matched[distances.index(sorted_distances[0])].append(i)
+        return matched
+
+    def get_pred_html(self, pred_structures, matched_index, ocr_contents):
+        end_html = []
+        td_index = 0
+        for tag in pred_structures:
+            if "</td>" in tag:
+                if "<td></td>" == tag:
+                    end_html.extend("<td>")
+                if td_index in matched_index.keys():
+                    b_with = False
+                    if (
+                        "<b>" in ocr_contents[matched_index[td_index][0]]
+                        and len(matched_index[td_index]) > 1
+                    ):
+                        b_with = True
+                        end_html.extend("<b>")
+                    for i, td_index_index in enumerate(matched_index[td_index]):
+                        content = ocr_contents[td_index_index][0]
+                        if len(matched_index[td_index]) > 1:
+                            if len(content) == 0:
+                                continue
+                            if content[0] == " ":
+                                content = content[1:]
+                            if "<b>" in content:
+                                content = content[3:]
+                            if "</b>" in content:
+                                content = content[:-4]
+                            if len(content) == 0:
+                                continue
+                            if (
+                                i != len(matched_index[td_index]) - 1
+                                and " " != content[-1]
+                            ):
+                                content += " "
+                        # escape content
+                        content = html.escape(content)
+                        end_html.extend(content)
+                    if b_with:
+                        end_html.extend("</b>")
+                if "<td></td>" == tag:
+                    end_html.append("</td>")
+                else:
+                    end_html.append(tag)
+                td_index += 1
+            else:
+                end_html.append(tag)
+        return "".join(end_html), end_html
+
+    def get_pred_html_master(self, pred_structures, matched_index, ocr_contents):
+        end_html = []
+        td_index = 0
+        for token in pred_structures:
+            if "</td>" in token:
+                txt = ""
+                b_with = False
+                if td_index in matched_index.keys():
+                    if (
+                        "<b>" in ocr_contents[matched_index[td_index][0]]
+                        and len(matched_index[td_index]) > 1
+                    ):
+                        b_with = True
+                    for i, td_index_index in enumerate(matched_index[td_index]):
+                        content = ocr_contents[td_index_index][0]
+                        if len(matched_index[td_index]) > 1:
+                            if len(content) == 0:
+                                continue
+                            if content[0] == " ":
+                                content = content[1:]
+                            if "<b>" in content:
+                                content = content[3:]
+                            if "</b>" in content:
+                                content = content[:-4]
+                            if len(content) == 0:
+                                continue
+                            if (
+                                i != len(matched_index[td_index]) - 1
+                                and " " != content[-1]
+                            ):
+                                content += " "
+                        txt += content
+                if b_with:
+                    txt = "<b>{}</b>".format(txt)
+                if "<td></td>" == token:
+                    token = "<td>{}</td>".format(txt)
+                else:
+                    token = "{}</td>".format(txt)
+                td_index += 1
+            token = deal_eb_token(token)
+            end_html.append(token)
+        html = "".join(end_html)
+        html = deal_bb(html)
+        return html, end_html
+
+    def _filter_ocr_result(self, pred_bboxes, dt_boxes, rec_res):
+        y1 = pred_bboxes[:, 1::2].min()
+        new_dt_boxes = []
+        new_rec_res = []
+
+        for box, rec in zip(dt_boxes, rec_res):
+            if np.max(box[1::2]) < y1:
+                continue
+            new_dt_boxes.append(box)
+            new_rec_res.append(rec)
+        return new_dt_boxes, new_rec_res
diff --git a/ppstructure/table/predict_structure.py b/ppstructure/table/predict_structure.py
new file mode 100755
index 0000000..3881f9f
--- /dev/null
+++ b/ppstructure/table/predict_structure.py
@@ -0,0 +1,207 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "../..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+
+import cv2
+import numpy as np
+import time
+import json
+
+import tools.infer.utility as utility
+from ppocr.data import create_operators, transform
+from ppocr.postprocess import build_post_process
+from ppocr.utils.logging import get_logger
+from ppocr.utils.utility import get_image_file_list, check_and_read
+from ppocr.utils.visual import draw_rectangle
+from ppstructure.utility import parse_args
+
+logger = get_logger()
+
+
+def build_pre_process_list(args):
+    resize_op = {
+        "ResizeTableImage": {
+            "max_len": args.table_max_len,
+        }
+    }
+    pad_op = {"PaddingTableImage": {"size": [args.table_max_len, args.table_max_len]}}
+    normalize_op = {
+        "NormalizeImage": {
+            "std": (
+                [0.229, 0.224, 0.225]
+                if args.table_algorithm not in ["TableMaster"]
+                else [0.5, 0.5, 0.5]
+            ),
+            "mean": (
+                [0.485, 0.456, 0.406]
+                if args.table_algorithm not in ["TableMaster"]
+                else [0.5, 0.5, 0.5]
+            ),
+            "scale": "1./255.",
+            "order": "hwc",
+        }
+    }
+    to_chw_op = {"ToCHWImage": None}
+    keep_keys_op = {"KeepKeys": {"keep_keys": ["image", "shape"]}}
+    if args.table_algorithm not in ["TableMaster"]:
+        pre_process_list = [resize_op, normalize_op, pad_op, to_chw_op, keep_keys_op]
+    else:
+        pre_process_list = [resize_op, pad_op, normalize_op, to_chw_op, keep_keys_op]
+    return pre_process_list
+
+
+class TableStructurer(object):
+    def __init__(self, args):
+        self.args = args
+        self.use_onnx = args.use_onnx
+        pre_process_list = build_pre_process_list(args)
+        if args.table_algorithm not in ["TableMaster"]:
+            postprocess_params = {
+                "name": "TableLabelDecode",
+                "character_dict_path": args.table_char_dict_path,
+                "merge_no_span_structure": args.merge_no_span_structure,
+            }
+        else:
+            postprocess_params = {
+                "name": "TableMasterLabelDecode",
+                "character_dict_path": args.table_char_dict_path,
+                "box_shape": "pad",
+                "merge_no_span_structure": args.merge_no_span_structure,
+            }
+
+        self.preprocess_op = create_operators(pre_process_list)
+        self.postprocess_op = build_post_process(postprocess_params)
+        (
+            self.predictor,
+            self.input_tensor,
+            self.output_tensors,
+            self.config,
+        ) = utility.create_predictor(args, "table", logger)
+
+        if args.benchmark:
+            import auto_log
+
+            pid = os.getpid()
+            gpu_id = utility.get_infer_gpuid()
+            self.autolog = auto_log.AutoLogger(
+                model_name="table",
+                model_precision=args.precision,
+                batch_size=1,
+                data_shape="dynamic",
+                save_path=None,  # args.save_log_path,
+                inference_config=self.config,
+                pids=pid,
+                process_name=None,
+                gpu_ids=gpu_id if args.use_gpu else None,
+                time_keys=["preprocess_time", "inference_time", "postprocess_time"],
+                warmup=0,
+                logger=logger,
+            )
+
+    def __call__(self, img):
+        starttime = time.time()
+        if self.args.benchmark:
+            self.autolog.times.start()
+
+        ori_im = img.copy()
+        data = {"image": img}
+        data = transform(data, self.preprocess_op)
+        img = data[0]
+        if img is None:
+            return None, 0
+        img = np.expand_dims(img, axis=0)
+        img = img.copy()
+        if self.args.benchmark:
+            self.autolog.times.stamp()
+        if self.use_onnx:
+            input_dict = {}
+            input_dict[self.input_tensor.name] = img
+            outputs = self.predictor.run(self.output_tensors, input_dict)
+        else:
+            self.input_tensor.copy_from_cpu(img)
+            self.predictor.run()
+            outputs = []
+            for output_tensor in self.output_tensors:
+                output = output_tensor.copy_to_cpu()
+                outputs.append(output)
+            if self.args.benchmark:
+                self.autolog.times.stamp()
+
+        preds = {}
+        preds["structure_probs"] = outputs[1]
+        preds["loc_preds"] = outputs[0]
+
+        shape_list = np.expand_dims(data[-1], axis=0)
+        post_result = self.postprocess_op(preds, [shape_list])
+
+        structure_str_list = post_result["structure_batch_list"][0]
+        bbox_list = post_result["bbox_batch_list"][0]
+        structure_str_list = structure_str_list[0]
+        structure_str_list = (
+            ["<html>", "<body>", "<table>"]
+            + structure_str_list
+            + ["</table>", "</body>", "</html>"]
+        )
+        elapse = time.time() - starttime
+        if self.args.benchmark:
+            self.autolog.times.end(stamp=True)
+        return (structure_str_list, bbox_list), elapse
+
+
+def main(args):
+    image_file_list = get_image_file_list(args.image_dir)
+    table_structurer = TableStructurer(args)
+    count = 0
+    total_time = 0
+    os.makedirs(args.output, exist_ok=True)
+    with open(
+        os.path.join(args.output, "infer.txt"), mode="w", encoding="utf-8"
+    ) as f_w:
+        for image_file in image_file_list:
+            img, flag, _ = check_and_read(image_file)
+            if not flag:
+                img = cv2.imread(image_file)
+            if img is None:
+                logger.info("error in loading image:{}".format(image_file))
+                continue
+            structure_res, elapse = table_structurer(img)
+            structure_str_list, bbox_list = structure_res
+            bbox_list_str = json.dumps(bbox_list.tolist())
+            logger.info("result: {}, {}".format(structure_str_list, bbox_list_str))
+            f_w.write("result: {}, {}\n".format(structure_str_list, bbox_list_str))
+
+            if len(bbox_list) > 0 and len(bbox_list[0]) == 4:
+                img = draw_rectangle(image_file, bbox_list)
+            else:
+                img = utility.draw_boxes(img, bbox_list)
+            img_save_path = os.path.join(args.output, os.path.basename(image_file))
+            cv2.imwrite(img_save_path, img)
+            logger.info("save vis result to {}".format(img_save_path))
+            if count > 0:
+                total_time += elapse
+            count += 1
+            logger.info("Predict time of {}: {}".format(image_file, elapse))
+    if args.benchmark:
+        table_structurer.autolog.report()
+
+
+if __name__ == "__main__":
+    main(parse_args())
diff --git a/ppstructure/table/predict_table.py b/ppstructure/table/predict_table.py
new file mode 100644
index 0000000..e1880cd
--- /dev/null
+++ b/ppstructure/table/predict_table.py
@@ -0,0 +1,241 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "..")))
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "../..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+import cv2
+import copy
+import logging
+import numpy as np
+import time
+import tools.infer.predict_rec as predict_rec
+import tools.infer.predict_det as predict_det
+import tools.infer.utility as utility
+from tools.infer.predict_system import sorted_boxes
+from ppocr.utils.utility import get_image_file_list, check_and_read
+from ppocr.utils.logging import get_logger
+from ppstructure.table.matcher import TableMatch
+from ppstructure.table.table_master_match import TableMasterMatcher
+from ppstructure.utility import parse_args
+import ppstructure.table.predict_structure as predict_strture
+
+logger = get_logger()
+
+
+def expand(pix, det_box, shape):
+    x0, y0, x1, y1 = det_box
+    #     print(shape)
+    h, w, c = shape
+    tmp_x0 = x0 - pix
+    tmp_x1 = x1 + pix
+    tmp_y0 = y0 - pix
+    tmp_y1 = y1 + pix
+    x0_ = tmp_x0 if tmp_x0 >= 0 else 0
+    x1_ = tmp_x1 if tmp_x1 <= w else w
+    y0_ = tmp_y0 if tmp_y0 >= 0 else 0
+    y1_ = tmp_y1 if tmp_y1 <= h else h
+    return x0_, y0_, x1_, y1_
+
+
+class TableSystem(object):
+    def __init__(self, args, text_detector=None, text_recognizer=None):
+        self.args = args
+        if not args.show_log:
+            logger.setLevel(logging.INFO)
+        benchmark_tmp = False
+        if args.benchmark:
+            benchmark_tmp = args.benchmark
+            args.benchmark = False
+        self.text_detector = (
+            predict_det.TextDetector(copy.deepcopy(args))
+            if text_detector is None
+            else text_detector
+        )
+        self.text_recognizer = (
+            predict_rec.TextRecognizer(copy.deepcopy(args))
+            if text_recognizer is None
+            else text_recognizer
+        )
+        if benchmark_tmp:
+            args.benchmark = True
+        self.table_structurer = predict_strture.TableStructurer(args)
+        if args.table_algorithm in ["TableMaster"]:
+            self.match = TableMasterMatcher()
+        else:
+            self.match = TableMatch(filter_ocr_result=True)
+
+        (
+            self.predictor,
+            self.input_tensor,
+            self.output_tensors,
+            self.config,
+        ) = utility.create_predictor(args, "table", logger)
+
+    def __call__(self, img, return_ocr_result_in_table=False):
+        result = dict()
+        time_dict = {"det": 0, "rec": 0, "table": 0, "all": 0, "match": 0}
+        start = time.time()
+        structure_res, elapse = self._structure(copy.deepcopy(img))
+        result["cell_bbox"] = structure_res[1].tolist()
+        time_dict["table"] = elapse
+
+        dt_boxes, rec_res, det_elapse, rec_elapse = self._ocr(copy.deepcopy(img))
+        time_dict["det"] = det_elapse
+        time_dict["rec"] = rec_elapse
+
+        if return_ocr_result_in_table:
+            result["boxes"] = [x.tolist() for x in dt_boxes]
+            result["rec_res"] = rec_res
+
+        tic = time.time()
+        pred_html = self.match(structure_res, dt_boxes, rec_res)
+        toc = time.time()
+        time_dict["match"] = toc - tic
+        result["html"] = pred_html
+        end = time.time()
+        time_dict["all"] = end - start
+        return result, time_dict
+
+    def _structure(self, img):
+        structure_res, elapse = self.table_structurer(copy.deepcopy(img))
+        return structure_res, elapse
+
+    def _ocr(self, img):
+        h, w = img.shape[:2]
+        dt_boxes, det_elapse = self.text_detector(copy.deepcopy(img))
+        dt_boxes = sorted_boxes(dt_boxes)
+
+        r_boxes = []
+        for box in dt_boxes:
+            x_min = max(0, box[:, 0].min() - 1)
+            x_max = min(w, box[:, 0].max() + 1)
+            y_min = max(0, box[:, 1].min() - 1)
+            y_max = min(h, box[:, 1].max() + 1)
+            box = [x_min, y_min, x_max, y_max]
+            r_boxes.append(box)
+        dt_boxes = np.array(r_boxes)
+        logger.debug("dt_boxes num : {}, elapse : {}".format(len(dt_boxes), det_elapse))
+        if dt_boxes is None:
+            return None, None
+
+        img_crop_list = []
+        for i in range(len(dt_boxes)):
+            det_box = dt_boxes[i]
+            x0, y0, x1, y1 = expand(2, det_box, img.shape)
+            text_rect = img[int(y0) : int(y1), int(x0) : int(x1), :]
+            img_crop_list.append(text_rect)
+        rec_res, rec_elapse = self.text_recognizer(img_crop_list)
+        logger.debug("rec_res num  : {}, elapse : {}".format(len(rec_res), rec_elapse))
+        return dt_boxes, rec_res, det_elapse, rec_elapse
+
+
+def to_excel(html_table, excel_path):
+    from tablepyxl import tablepyxl
+
+    tablepyxl.document_to_xl(html_table, excel_path)
+
+
+def main(args):
+    image_file_list = get_image_file_list(args.image_dir)
+    image_file_list = image_file_list[args.process_id :: args.total_process_num]
+    os.makedirs(args.output, exist_ok=True)
+
+    table_sys = TableSystem(args)
+    img_num = len(image_file_list)
+
+    f_html = open(os.path.join(args.output, "show.html"), mode="w", encoding="utf-8")
+    f_html.write("<html>\n<body>\n")
+    f_html.write('<table border="1">\n')
+    f_html.write(
+        '<meta http-equiv="Content-Type" content="text/html; charset=utf-8" />'
+    )
+    f_html.write("<tr>\n")
+    f_html.write("<td>img name\n")
+    f_html.write("<td>ori image</td>")
+    f_html.write("<td>table html</td>")
+    f_html.write("<td>cell box</td>")
+    f_html.write("</tr>\n")
+
+    for i, image_file in enumerate(image_file_list):
+        logger.info("[{}/{}] {}".format(i, img_num, image_file))
+        img, flag, _ = check_and_read(image_file)
+        excel_path = os.path.join(
+            args.output, os.path.basename(image_file).split(".")[0] + ".xlsx"
+        )
+        if not flag:
+            img = cv2.imread(image_file)
+        if img is None:
+            logger.error("error in loading image:{}".format(image_file))
+            continue
+        starttime = time.time()
+        pred_res, _ = table_sys(img)
+        pred_html = pred_res["html"]
+        logger.info(pred_html)
+        to_excel(pred_html, excel_path)
+        logger.info("excel saved to {}".format(excel_path))
+        elapse = time.time() - starttime
+        logger.info("Predict time : {:.3f}s".format(elapse))
+
+        if len(pred_res["cell_bbox"]) > 0 and len(pred_res["cell_bbox"][0]) == 4:
+            img = predict_strture.draw_rectangle(image_file, pred_res["cell_bbox"])
+        else:
+            img = utility.draw_boxes(img, pred_res["cell_bbox"])
+        img_save_path = os.path.join(args.output, os.path.basename(image_file))
+        cv2.imwrite(img_save_path, img)
+
+        f_html.write("<tr>\n")
+        f_html.write(f"<td> {os.path.basename(image_file)} <br/>\n")
+        f_html.write(f'<td><img src="{image_file}" width=640></td>\n')
+        f_html.write(
+            '<td><table  border="1">'
+            + pred_html.replace("<html><body><table>", "").replace(
+                "</table></body></html>", ""
+            )
+            + "</table></td>\n"
+        )
+        f_html.write(f'<td><img src="{os.path.basename(image_file)}" width=640></td>\n')
+        f_html.write("</tr>\n")
+    f_html.write("</table>\n")
+    f_html.close()
+
+    if args.benchmark:
+        table_sys.table_structurer.autolog.report()
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    if args.use_mp:
+        import subprocess
+
+        p_list = []
+        total_process_num = args.total_process_num
+        for process_id in range(total_process_num):
+            cmd = (
+                [sys.executable, "-u"]
+                + sys.argv
+                + ["--process_id={}".format(process_id), "--use_mp={}".format(False)]
+            )
+            p = subprocess.Popen(cmd, stdout=sys.stdout, stderr=sys.stdout)
+            p_list.append(p)
+        for p in p_list:
+            p.wait()
+    else:
+        main(args)
diff --git a/ppstructure/table/table_master_match.py b/ppstructure/table/table_master_match.py
new file mode 100644
index 0000000..54b4506
--- /dev/null
+++ b/ppstructure/table/table_master_match.py
@@ -0,0 +1,995 @@
+# copyright (c) 2022 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+This code is refer from:
+https://github.com/JiaquanYe/TableMASTER-mmocr/blob/master/table_recognition/match.py
+"""
+
+import os
+import re
+import cv2
+import glob
+import copy
+import math
+import pickle
+import numpy as np
+
+from shapely.geometry import Polygon, MultiPoint
+
+"""
+Useful function in matching.
+"""
+
+
+def remove_empty_bboxes(bboxes):
+    """
+    remove [0., 0., 0., 0.] in structure master bboxes.
+    len(bboxes.shape) must be 2.
+    :param bboxes:
+    :return:
+    """
+    new_bboxes = []
+    for bbox in bboxes:
+        if sum(bbox) == 0.0:
+            continue
+        new_bboxes.append(bbox)
+    return np.array(new_bboxes)
+
+
+def xywh2xyxy(bboxes):
+    if len(bboxes.shape) == 1:
+        new_bboxes = np.empty_like(bboxes)
+        new_bboxes[0] = bboxes[0] - bboxes[2] / 2
+        new_bboxes[1] = bboxes[1] - bboxes[3] / 2
+        new_bboxes[2] = bboxes[0] + bboxes[2] / 2
+        new_bboxes[3] = bboxes[1] + bboxes[3] / 2
+        return new_bboxes
+    elif len(bboxes.shape) == 2:
+        new_bboxes = np.empty_like(bboxes)
+        new_bboxes[:, 0] = bboxes[:, 0] - bboxes[:, 2] / 2
+        new_bboxes[:, 1] = bboxes[:, 1] - bboxes[:, 3] / 2
+        new_bboxes[:, 2] = bboxes[:, 0] + bboxes[:, 2] / 2
+        new_bboxes[:, 3] = bboxes[:, 1] + bboxes[:, 3] / 2
+        return new_bboxes
+    else:
+        raise ValueError
+
+
+def xyxy2xywh(bboxes):
+    if len(bboxes.shape) == 1:
+        new_bboxes = np.empty_like(bboxes)
+        new_bboxes[0] = bboxes[0] + (bboxes[2] - bboxes[0]) / 2
+        new_bboxes[1] = bboxes[1] + (bboxes[3] - bboxes[1]) / 2
+        new_bboxes[2] = bboxes[2] - bboxes[0]
+        new_bboxes[3] = bboxes[3] - bboxes[1]
+        return new_bboxes
+    elif len(bboxes.shape) == 2:
+        new_bboxes = np.empty_like(bboxes)
+        new_bboxes[:, 0] = bboxes[:, 0] + (bboxes[:, 2] - bboxes[:, 0]) / 2
+        new_bboxes[:, 1] = bboxes[:, 1] + (bboxes[:, 3] - bboxes[:, 1]) / 2
+        new_bboxes[:, 2] = bboxes[:, 2] - bboxes[:, 0]
+        new_bboxes[:, 3] = bboxes[:, 3] - bboxes[:, 1]
+        return new_bboxes
+    else:
+        raise ValueError
+
+
+def pickle_load(path, prefix="end2end"):
+    if os.path.isfile(path):
+        data = pickle.load(open(path, "rb"))
+    elif os.path.isdir(path):
+        data = dict()
+        search_path = os.path.join(path, "{}_*.pkl".format(prefix))
+        pkls = glob.glob(search_path)
+        for pkl in pkls:
+            this_data = pickle.load(open(pkl, "rb"))
+            data.update(this_data)
+    else:
+        raise ValueError
+    return data
+
+
+def convert_coord(xyxy):
+    """
+    Convert two points format to four points format.
+    :param xyxy:
+    :return:
+    """
+    new_bbox = np.zeros([4, 2], dtype=np.float32)
+    new_bbox[0, 0], new_bbox[0, 1] = xyxy[0], xyxy[1]
+    new_bbox[1, 0], new_bbox[1, 1] = xyxy[2], xyxy[1]
+    new_bbox[2, 0], new_bbox[2, 1] = xyxy[2], xyxy[3]
+    new_bbox[3, 0], new_bbox[3, 1] = xyxy[0], xyxy[3]
+    return new_bbox
+
+
+def cal_iou(bbox1, bbox2):
+    bbox1_poly = Polygon(bbox1).convex_hull
+    bbox2_poly = Polygon(bbox2).convex_hull
+    union_poly = np.concatenate((bbox1, bbox2))
+
+    if not bbox1_poly.intersects(bbox2_poly):
+        iou = 0
+    else:
+        inter_area = bbox1_poly.intersection(bbox2_poly).area
+        union_area = MultiPoint(union_poly).convex_hull.area
+        if union_area == 0:
+            iou = 0
+        else:
+            iou = float(inter_area) / union_area
+    return iou
+
+
+def cal_distance(p1, p2):
+    delta_x = p1[0] - p2[0]
+    delta_y = p1[1] - p2[1]
+    d = math.sqrt((delta_x**2) + (delta_y**2))
+    return d
+
+
+def is_inside(center_point, corner_point):
+    """
+    Find if center_point inside the bbox(corner_point) or not.
+    :param center_point: center point (x, y)
+    :param corner_point: corner point ((x1,y1),(x2,y2))
+    :return:
+    """
+    x_flag = False
+    y_flag = False
+    if (center_point[0] >= corner_point[0][0]) and (
+        center_point[0] <= corner_point[1][0]
+    ):
+        x_flag = True
+    if (center_point[1] >= corner_point[0][1]) and (
+        center_point[1] <= corner_point[1][1]
+    ):
+        y_flag = True
+    if x_flag and y_flag:
+        return True
+    else:
+        return False
+
+
+def find_no_match(match_list, all_end2end_nums, type="end2end"):
+    """
+    Find out no match end2end bbox in previous match list.
+    :param match_list: matching pairs.
+    :param all_end2end_nums: numbers of end2end_xywh
+    :param type: 'end2end' corresponding to idx 0, 'master' corresponding to idx 1.
+    :return: no match pse bbox index list
+    """
+    if type == "end2end":
+        idx = 0
+    elif type == "master":
+        idx = 1
+    else:
+        raise ValueError
+
+    no_match_indexs = []
+    # m[0] is end2end index m[1] is master index
+    matched_bbox_indexs = [m[idx] for m in match_list]
+    for n in range(all_end2end_nums):
+        if n not in matched_bbox_indexs:
+            no_match_indexs.append(n)
+    return no_match_indexs
+
+
+def is_abs_lower_than_threshold(this_bbox, target_bbox, threshold=3):
+    # only consider y axis, for grouping in row.
+    delta = abs(this_bbox[1] - target_bbox[1])
+    if delta < threshold:
+        return True
+    else:
+        return False
+
+
+def sort_line_bbox(g, bg):
+    """
+    Sorted the bbox in the same line(group)
+    compare coord 'x' value, where 'y' value is closed in the same group.
+    :param g: index in the same group
+    :param bg: bbox in the same group
+    :return:
+    """
+
+    xs = [bg_item[0] for bg_item in bg]
+    xs_sorted = sorted(xs)
+
+    g_sorted = [None] * len(xs_sorted)
+    bg_sorted = [None] * len(xs_sorted)
+    for g_item, bg_item in zip(g, bg):
+        idx = xs_sorted.index(bg_item[0])
+        bg_sorted[idx] = bg_item
+        g_sorted[idx] = g_item
+
+    return g_sorted, bg_sorted
+
+
+def flatten(sorted_groups, sorted_bbox_groups):
+    idxs = []
+    bboxes = []
+    for group, bbox_group in zip(sorted_groups, sorted_bbox_groups):
+        for g, bg in zip(group, bbox_group):
+            idxs.append(g)
+            bboxes.append(bg)
+    return idxs, bboxes
+
+
+def sort_bbox(end2end_xywh_bboxes, no_match_end2end_indexes):
+    """
+    This function will group the render end2end bboxes in row.
+    :param end2end_xywh_bboxes:
+    :param no_match_end2end_indexes:
+    :return:
+    """
+    groups = []
+    bbox_groups = []
+    for index, end2end_xywh_bbox in zip(no_match_end2end_indexes, end2end_xywh_bboxes):
+        this_bbox = end2end_xywh_bbox
+        if len(groups) == 0:
+            groups.append([index])
+            bbox_groups.append([this_bbox])
+        else:
+            flag = False
+            for g, bg in zip(groups, bbox_groups):
+                # this_bbox is belong to bg's row or not
+                if is_abs_lower_than_threshold(this_bbox, bg[0]):
+                    g.append(index)
+                    bg.append(this_bbox)
+                    flag = True
+                    break
+            if not flag:
+                # this_bbox is not belong to bg's row, create a row.
+                groups.append([index])
+                bbox_groups.append([this_bbox])
+
+    # sorted bboxes in a group
+    tmp_groups, tmp_bbox_groups = [], []
+    for g, bg in zip(groups, bbox_groups):
+        g_sorted, bg_sorted = sort_line_bbox(g, bg)
+        tmp_groups.append(g_sorted)
+        tmp_bbox_groups.append(bg_sorted)
+
+    # sorted groups, sort by coord y's value.
+    sorted_groups = [None] * len(tmp_groups)
+    sorted_bbox_groups = [None] * len(tmp_bbox_groups)
+    ys = [bg[0][1] for bg in tmp_bbox_groups]
+    sorted_ys = sorted(ys)
+    for g, bg in zip(tmp_groups, tmp_bbox_groups):
+        idx = sorted_ys.index(bg[0][1])
+        sorted_groups[idx] = g
+        sorted_bbox_groups[idx] = bg
+
+    # flatten, get final result
+    end2end_sorted_idx_list, end2end_sorted_bbox_list = flatten(
+        sorted_groups, sorted_bbox_groups
+    )
+
+    return (
+        end2end_sorted_idx_list,
+        end2end_sorted_bbox_list,
+        sorted_groups,
+        sorted_bbox_groups,
+    )
+
+
+def get_bboxes_list(end2end_result, structure_master_result):
+    """
+    This function is use to convert end2end results and structure master results to
+    List of xyxy bbox format and List of xywh bbox format
+    :param end2end_result: bbox's format is xyxy
+    :param structure_master_result: bbox's format is xywh
+    :return: 4 kind list of bbox ()
+    """
+    # end2end
+    end2end_xyxy_list = []
+    end2end_xywh_list = []
+    for end2end_item in end2end_result:
+        src_bbox = end2end_item["bbox"]
+        end2end_xyxy_list.append(src_bbox)
+        xywh_bbox = xyxy2xywh(src_bbox)
+        end2end_xywh_list.append(xywh_bbox)
+    end2end_xyxy_bboxes = np.array(end2end_xyxy_list)
+    end2end_xywh_bboxes = np.array(end2end_xywh_list)
+
+    # structure master
+    src_bboxes = structure_master_result["bbox"]
+    src_bboxes = remove_empty_bboxes(src_bboxes)
+    structure_master_xyxy_bboxes = src_bboxes
+    xywh_bbox = xyxy2xywh(src_bboxes)
+    structure_master_xywh_bboxes = xywh_bbox
+
+    return (
+        end2end_xyxy_bboxes,
+        end2end_xywh_bboxes,
+        structure_master_xywh_bboxes,
+        structure_master_xyxy_bboxes,
+    )
+
+
+def center_rule_match(end2end_xywh_bboxes, structure_master_xyxy_bboxes):
+    """
+    Judge end2end Bbox's center point is inside structure master Bbox or not,
+    if end2end Bbox's center is in structure master Bbox, get matching pair.
+    :param end2end_xywh_bboxes:
+    :param structure_master_xyxy_bboxes:
+    :return: match pairs list, e.g. [[0,1], [1,2], ...]
+    """
+    match_pairs_list = []
+    for i, end2end_xywh in enumerate(end2end_xywh_bboxes):
+        for j, master_xyxy in enumerate(structure_master_xyxy_bboxes):
+            x_end2end, y_end2end = end2end_xywh[0], end2end_xywh[1]
+            x_master1, y_master1, x_master2, y_master2 = (
+                master_xyxy[0],
+                master_xyxy[1],
+                master_xyxy[2],
+                master_xyxy[3],
+            )
+            center_point_end2end = (x_end2end, y_end2end)
+            corner_point_master = ((x_master1, y_master1), (x_master2, y_master2))
+            if is_inside(center_point_end2end, corner_point_master):
+                match_pairs_list.append([i, j])
+    return match_pairs_list
+
+
+def iou_rule_match(
+    end2end_xyxy_bboxes, end2end_xyxy_indexes, structure_master_xyxy_bboxes
+):
+    """
+    Use iou to find matching list.
+    choose max iou value bbox as match pair.
+    :param end2end_xyxy_bboxes:
+    :param end2end_xyxy_indexes: original end2end indexes.
+    :param structure_master_xyxy_bboxes:
+    :return: match pairs list, e.g. [[0,1], [1,2], ...]
+    """
+    match_pair_list = []
+    for end2end_xyxy_index, end2end_xyxy in zip(
+        end2end_xyxy_indexes, end2end_xyxy_bboxes
+    ):
+        max_iou = 0
+        max_match = [None, None]
+        for j, master_xyxy in enumerate(structure_master_xyxy_bboxes):
+            end2end_4xy = convert_coord(end2end_xyxy)
+            master_4xy = convert_coord(master_xyxy)
+            iou = cal_iou(end2end_4xy, master_4xy)
+            if iou > max_iou:
+                max_match[0], max_match[1] = end2end_xyxy_index, j
+                max_iou = iou
+
+        if max_match[0] is None:
+            # no match
+            continue
+        match_pair_list.append(max_match)
+    return match_pair_list
+
+
+def distance_rule_match(end2end_indexes, end2end_bboxes, master_indexes, master_bboxes):
+    """
+    Get matching between no-match end2end bboxes and no-match master bboxes.
+    Use min distance to match.
+    This rule will only run (no-match end2end nums > 0) and (no-match master nums > 0)
+    It will Return master_bboxes_nums match-pairs.
+    :param end2end_indexes:
+    :param end2end_bboxes:
+    :param master_indexes:
+    :param master_bboxes:
+    :return: match_pairs list, e.g. [[0,1], [1,2], ...]
+    """
+    min_match_list = []
+    for j, master_bbox in zip(master_indexes, master_bboxes):
+        min_distance = np.inf
+        min_match = [0, 0]  # i, j
+        for i, end2end_bbox in zip(end2end_indexes, end2end_bboxes):
+            x_end2end, y_end2end = end2end_bbox[0], end2end_bbox[1]
+            x_master, y_master = master_bbox[0], master_bbox[1]
+            end2end_point = (x_end2end, y_end2end)
+            master_point = (x_master, y_master)
+            dist = cal_distance(master_point, end2end_point)
+            if dist < min_distance:
+                min_match[0], min_match[1] = i, j
+                min_distance = dist
+        min_match_list.append(min_match)
+    return min_match_list
+
+
+def extra_match(no_match_end2end_indexes, master_bbox_nums):
+    """
+    This function will create some virtual master bboxes,
+    and get match with the no match end2end indexes.
+    :param no_match_end2end_indexes:
+    :param master_bbox_nums:
+    :return:
+    """
+    end_nums = len(no_match_end2end_indexes) + master_bbox_nums
+    extra_match_list = []
+    for i in range(master_bbox_nums, end_nums):
+        end2end_index = no_match_end2end_indexes[i - master_bbox_nums]
+        extra_match_list.append([end2end_index, i])
+    return extra_match_list
+
+
+def get_match_dict(match_list):
+    """
+    Convert match_list to a dict, where key is master bbox's index, value is end2end bbox index.
+    :param match_list:
+    :return:
+    """
+    match_dict = dict()
+    for match_pair in match_list:
+        end2end_index, master_index = match_pair[0], match_pair[1]
+        if master_index not in match_dict.keys():
+            match_dict[master_index] = [end2end_index]
+        else:
+            match_dict[master_index].append(end2end_index)
+    return match_dict
+
+
+def deal_successive_space(text):
+    """
+    deal successive space character for text
+    1. Replace ' '*3 with '<space>' which is real space is text
+    2. Remove ' ', which is split token, not true space
+    3. Replace '<space>' with ' ', to get real text
+    :param text:
+    :return:
+    """
+    text = text.replace(" " * 3, "<space>")
+    text = text.replace(" ", "")
+    text = text.replace("<space>", " ")
+    return text
+
+
+def reduce_repeat_bb(text_list, break_token):
+    """
+    convert ['<b>Local</b>', '<b>government</b>', '<b>unit</b>'] to ['<b>Local government unit</b>']
+    PS: maybe style <i>Local</i> is also exist, too. it can be processed like this.
+    :param text_list:
+    :param break_token:
+    :return:
+    """
+    count = 0
+    for text in text_list:
+        if text.startswith("<b>"):
+            count += 1
+    if count == len(text_list):
+        new_text_list = []
+        for text in text_list:
+            text = text.replace("<b>", "").replace("</b>", "")
+            new_text_list.append(text)
+        return ["<b>" + break_token.join(new_text_list) + "</b>"]
+    else:
+        return text_list
+
+
+def get_match_text_dict(match_dict, end2end_info, break_token=" "):
+    match_text_dict = dict()
+    for master_index, end2end_index_list in match_dict.items():
+        text_list = [
+            end2end_info[end2end_index]["text"] for end2end_index in end2end_index_list
+        ]
+        text_list = reduce_repeat_bb(text_list, break_token)
+        text = break_token.join(text_list)
+        match_text_dict[master_index] = text
+    return match_text_dict
+
+
+def merge_span_token(master_token_list):
+    """
+    Merge the span style token (row span or col span).
+    :param master_token_list:
+    :return:
+    """
+    new_master_token_list = []
+    pointer = 0
+    if master_token_list[-1] != "</tbody>":
+        master_token_list.append("</tbody>")
+    while master_token_list[pointer] != "</tbody>":
+        try:
+            if master_token_list[pointer] == "<td":
+                if master_token_list[pointer + 1].startswith(
+                    " colspan="
+                ) or master_token_list[pointer + 1].startswith(" rowspan="):
+                    """
+                    example:
+                    pattern <td colspan="3">
+                    '<td' + 'colspan=" "' + '>' + '</td>'
+                    """
+                    tmp = "".join(master_token_list[pointer : pointer + 3 + 1])
+                    pointer += 4
+                    new_master_token_list.append(tmp)
+
+                elif master_token_list[pointer + 2].startswith(
+                    " colspan="
+                ) or master_token_list[pointer + 2].startswith(" rowspan="):
+                    """
+                    example:
+                    pattern <td rowspan="2" colspan="3">
+                    '<td' + 'rowspan=" "' + 'colspan=" "' + '>' + '</td>'
+                    """
+                    tmp = "".join(master_token_list[pointer : pointer + 4 + 1])
+                    pointer += 5
+                    new_master_token_list.append(tmp)
+
+                else:
+                    new_master_token_list.append(master_token_list[pointer])
+                    pointer += 1
+            else:
+                new_master_token_list.append(master_token_list[pointer])
+                pointer += 1
+        except:
+            print("Break in merge...")
+            break
+    new_master_token_list.append("</tbody>")
+
+    return new_master_token_list
+
+
+def deal_eb_token(master_token):
+    """
+    post process with <eb></eb>, <eb1></eb1>, ...
+    emptyBboxTokenDict = {
+        "[]": '<eb></eb>',
+        "[' ']": '<eb1></eb1>',
+        "['<b>', ' ', '</b>']": '<eb2></eb2>',
+        "['\\u2028', '\\u2028']": '<eb3></eb3>',
+        "['<sup>', ' ', '</sup>']": '<eb4></eb4>',
+        "['<b>', '</b>']": '<eb5></eb5>',
+        "['<i>', ' ', '</i>']": '<eb6></eb6>',
+        "['<b>', '<i>', '</i>', '</b>']": '<eb7></eb7>',
+        "['<b>', '<i>', ' ', '</i>', '</b>']": '<eb8></eb8>',
+        "['<i>', '</i>']": '<eb9></eb9>',
+        "['<b>', ' ', '\\u2028', ' ', '\\u2028', ' ', '</b>']": '<eb10></eb10>',
+    }
+    :param master_token:
+    :return:
+    """
+    master_token = master_token.replace("<eb></eb>", "<td></td>")
+    master_token = master_token.replace("<eb1></eb1>", "<td> </td>")
+    master_token = master_token.replace("<eb2></eb2>", "<td><b> </b></td>")
+    master_token = master_token.replace("<eb3></eb3>", "<td>\u2028\u2028</td>")
+    master_token = master_token.replace("<eb4></eb4>", "<td><sup> </sup></td>")
+    master_token = master_token.replace("<eb5></eb5>", "<td><b></b></td>")
+    master_token = master_token.replace("<eb6></eb6>", "<td><i> </i></td>")
+    master_token = master_token.replace("<eb7></eb7>", "<td><b><i></i></b></td>")
+    master_token = master_token.replace("<eb8></eb8>", "<td><b><i> </i></b></td>")
+    master_token = master_token.replace("<eb9></eb9>", "<td><i></i></td>")
+    master_token = master_token.replace(
+        "<eb10></eb10>", "<td><b> \u2028 \u2028 </b></td>"
+    )
+    return master_token
+
+
+def insert_text_to_token(master_token_list, match_text_dict):
+    """
+    Insert OCR text result to structure token.
+    :param master_token_list:
+    :param match_text_dict:
+    :return:
+    """
+    master_token_list = merge_span_token(master_token_list)
+    merged_result_list = []
+    text_count = 0
+    for master_token in master_token_list:
+        if master_token.startswith("<td"):
+            if text_count > len(match_text_dict) - 1:
+                text_count += 1
+                continue
+            elif text_count not in match_text_dict.keys():
+                text_count += 1
+                continue
+            else:
+                master_token = master_token.replace(
+                    "><", ">{}<".format(match_text_dict[text_count])
+                )
+                text_count += 1
+        master_token = deal_eb_token(master_token)
+        merged_result_list.append(master_token)
+
+    return "".join(merged_result_list)
+
+
+def deal_isolate_span(thead_part):
+    """
+    Deal with isolate span cases in this function.
+    It causes by wrong prediction in structure recognition model.
+    eg. predict <td rowspan="2"></td> to <td></td> rowspan="2"></b></td>.
+    :param thead_part:
+    :return:
+    """
+    # 1. find out isolate span tokens.
+    isolate_pattern = (
+        '<td></td> rowspan="(\d)+" colspan="(\d)+"></b></td>|'
+        '<td></td> colspan="(\d)+" rowspan="(\d)+"></b></td>|'
+        '<td></td> rowspan="(\d)+"></b></td>|'
+        '<td></td> colspan="(\d)+"></b></td>'
+    )
+    isolate_iter = re.finditer(isolate_pattern, thead_part)
+    isolate_list = [i.group() for i in isolate_iter]
+
+    # 2. find out span number, by step 1 results.
+    span_pattern = (
+        ' rowspan="(\d)+" colspan="(\d)+"|'
+        ' colspan="(\d)+" rowspan="(\d)+"|'
+        ' rowspan="(\d)+"|'
+        ' colspan="(\d)+"'
+    )
+    corrected_list = []
+    for isolate_item in isolate_list:
+        span_part = re.search(span_pattern, isolate_item)
+        spanStr_in_isolateItem = span_part.group()
+        # 3. merge the span number into the span token format string.
+        if spanStr_in_isolateItem is not None:
+            corrected_item = "<td{}></td>".format(spanStr_in_isolateItem)
+            corrected_list.append(corrected_item)
+        else:
+            corrected_list.append(None)
+
+    # 4. replace original isolated token.
+    for corrected_item, isolate_item in zip(corrected_list, isolate_list):
+        if corrected_item is not None:
+            thead_part = thead_part.replace(isolate_item, corrected_item)
+        else:
+            pass
+    return thead_part
+
+
+def deal_duplicate_bb(thead_part):
+    """
+    Deal duplicate <b> or </b> after replace.
+    Keep one <b></b> in a <td></td> token.
+    :param thead_part:
+    :return:
+    """
+    # 1. find out <td></td> in <thead></thead>.
+    td_pattern = (
+        '<td rowspan="(\d)+" colspan="(\d)+">(.+?)</td>|'
+        '<td colspan="(\d)+" rowspan="(\d)+">(.+?)</td>|'
+        '<td rowspan="(\d)+">(.+?)</td>|'
+        '<td colspan="(\d)+">(.+?)</td>|'
+        "<td>(.*?)</td>"
+    )
+    td_iter = re.finditer(td_pattern, thead_part)
+    td_list = [t.group() for t in td_iter]
+
+    # 2. is multiply <b></b> in <td></td> or not?
+    new_td_list = []
+    for td_item in td_list:
+        if td_item.count("<b>") > 1 or td_item.count("</b>") > 1:
+            # multiply <b></b> in <td></td> case.
+            # 1. remove all <b></b>
+            td_item = td_item.replace("<b>", "").replace("</b>", "")
+            # 2. replace <tb> -> <tb><b>, </tb> -> </b></tb>.
+            td_item = td_item.replace("<td>", "<td><b>").replace("</td>", "</b></td>")
+            new_td_list.append(td_item)
+        else:
+            new_td_list.append(td_item)
+
+    # 3. replace original thead part.
+    for td_item, new_td_item in zip(td_list, new_td_list):
+        thead_part = thead_part.replace(td_item, new_td_item)
+    return thead_part
+
+
+def deal_bb(result_token):
+    """
+    In our opinion, <b></b> always occurs in <thead></thead> text's context.
+    This function will find out all tokens in <thead></thead> and insert <b></b> by manual.
+    :param result_token:
+    :return:
+    """
+    # find out <thead></thead> parts.
+    thead_pattern = "<thead>(.*?)</thead>"
+    if re.search(thead_pattern, result_token) is None:
+        return result_token
+    thead_part = re.search(thead_pattern, result_token).group()
+    origin_thead_part = copy.deepcopy(thead_part)
+
+    # check "rowspan" or "colspan" occur in <thead></thead> parts or not .
+    span_pattern = '<td rowspan="(\d)+" colspan="(\d)+">|<td colspan="(\d)+" rowspan="(\d)+">|<td rowspan="(\d)+">|<td colspan="(\d)+">'
+    span_iter = re.finditer(span_pattern, thead_part)
+    span_list = [s.group() for s in span_iter]
+    has_span_in_head = True if len(span_list) > 0 else False
+
+    if not has_span_in_head:
+        # <thead></thead> not include "rowspan" or "colspan" branch 1.
+        # 1. replace <td> to <td><b>, and </td> to </b></td>
+        # 2. it is possible to predict text include <b> or </b> by Text-line recognition,
+        #    so we replace <b><b> to <b>, and </b></b> to </b>
+        thead_part = (
+            thead_part.replace("<td>", "<td><b>")
+            .replace("</td>", "</b></td>")
+            .replace("<b><b>", "<b>")
+            .replace("</b></b>", "</b>")
+        )
+    else:
+        # <thead></thead> include "rowspan" or "colspan" branch 2.
+        # Firstly, we deal rowspan or colspan cases.
+        # 1. replace > to ><b>
+        # 2. replace </td> to </b></td>
+        # 3. it is possible to predict text include <b> or </b> by Text-line recognition,
+        #    so we replace <b><b> to <b>, and </b><b> to </b>
+
+        # Secondly, deal ordinary cases like branch 1
+
+        # replace ">" to "<b>"
+        replaced_span_list = []
+        for sp in span_list:
+            replaced_span_list.append(sp.replace(">", "><b>"))
+        for sp, rsp in zip(span_list, replaced_span_list):
+            thead_part = thead_part.replace(sp, rsp)
+
+        # replace "</td>" to "</b></td>"
+        thead_part = thead_part.replace("</td>", "</b></td>")
+
+        # remove duplicated <b> by re.sub
+        mb_pattern = "(<b>)+"
+        single_b_string = "<b>"
+        thead_part = re.sub(mb_pattern, single_b_string, thead_part)
+
+        mgb_pattern = "(</b>)+"
+        single_gb_string = "</b>"
+        thead_part = re.sub(mgb_pattern, single_gb_string, thead_part)
+
+        # ordinary cases like branch 1
+        thead_part = thead_part.replace("<td>", "<td><b>").replace("<b><b>", "<b>")
+
+    # convert <tb><b></b></tb> back to <tb></tb>, empty cell has no <b></b>.
+    # but space cell(<tb> </tb>)  is suitable for <td><b> </b></td>
+    thead_part = thead_part.replace("<td><b></b></td>", "<td></td>")
+    # deal with duplicated <b></b>
+    thead_part = deal_duplicate_bb(thead_part)
+    # deal with isolate span tokens, which causes by wrong predict by structure prediction.
+    # eg.PMC5994107_011_00.png
+    thead_part = deal_isolate_span(thead_part)
+    # replace original result with new thead part.
+    result_token = result_token.replace(origin_thead_part, thead_part)
+    return result_token
+
+
+class Matcher:
+    def __init__(self, end2end_file, structure_master_file):
+        """
+        This class process the end2end results and structure recognition results.
+        :param end2end_file: end2end results predict by end2end inference.
+        :param structure_master_file: structure recognition results predict by structure master inference.
+        """
+        self.end2end_file = end2end_file
+        self.structure_master_file = structure_master_file
+        self.end2end_results = pickle_load(end2end_file, prefix="end2end")
+        self.structure_master_results = pickle_load(
+            structure_master_file, prefix="structure"
+        )
+
+    def match(self):
+        """
+        Match process:
+        pre-process : convert end2end and structure master results to xyxy, xywh ndnarray format.
+        1. Use pseBbox is inside masterBbox judge rule
+        2. Use iou between pseBbox and masterBbox rule
+        3. Use min distance of center point rule
+        :return:
+        """
+        match_results = dict()
+        for idx, (file_name, end2end_result) in enumerate(self.end2end_results.items()):
+            match_list = []
+            if file_name not in self.structure_master_results:
+                continue
+            structure_master_result = self.structure_master_results[file_name]
+            (
+                end2end_xyxy_bboxes,
+                end2end_xywh_bboxes,
+                structure_master_xywh_bboxes,
+                structure_master_xyxy_bboxes,
+            ) = get_bboxes_list(end2end_result, structure_master_result)
+
+            # rule 1: center rule
+            center_rule_match_list = center_rule_match(
+                end2end_xywh_bboxes, structure_master_xyxy_bboxes
+            )
+            match_list.extend(center_rule_match_list)
+
+            # rule 2: iou rule
+            # firstly, find not match index in previous step.
+            center_no_match_end2end_indexs = find_no_match(
+                match_list, len(end2end_xywh_bboxes), type="end2end"
+            )
+            if len(center_no_match_end2end_indexs) > 0:
+                center_no_match_end2end_xyxy = end2end_xyxy_bboxes[
+                    center_no_match_end2end_indexs
+                ]
+                # secondly, iou rule match
+                iou_rule_match_list = iou_rule_match(
+                    center_no_match_end2end_xyxy,
+                    center_no_match_end2end_indexs,
+                    structure_master_xyxy_bboxes,
+                )
+                match_list.extend(iou_rule_match_list)
+
+            # rule 3: distance rule
+            # match between no-match end2end bboxes and no-match master bboxes.
+            # it will return master_bboxes_nums match-pairs.
+            # firstly, find not match index in previous step.
+            centerIou_no_match_end2end_indexs = find_no_match(
+                match_list, len(end2end_xywh_bboxes), type="end2end"
+            )
+            centerIou_no_match_master_indexs = find_no_match(
+                match_list, len(structure_master_xywh_bboxes), type="master"
+            )
+            if (
+                len(centerIou_no_match_master_indexs) > 0
+                and len(centerIou_no_match_end2end_indexs) > 0
+            ):
+                centerIou_no_match_end2end_xywh = end2end_xywh_bboxes[
+                    centerIou_no_match_end2end_indexs
+                ]
+                centerIou_no_match_master_xywh = structure_master_xywh_bboxes[
+                    centerIou_no_match_master_indexs
+                ]
+                distance_match_list = distance_rule_match(
+                    centerIou_no_match_end2end_indexs,
+                    centerIou_no_match_end2end_xywh,
+                    centerIou_no_match_master_indexs,
+                    centerIou_no_match_master_xywh,
+                )
+                match_list.extend(distance_match_list)
+
+            # TODO:
+            # The render no-match pseBbox, insert the last
+            # After step3 distance rule, a master bbox at least match one end2end bbox.
+            # But end2end bbox maybe overmuch, because numbers of master bbox will cut by max length.
+            # For these render end2end bboxes, we will make some virtual master bboxes, and get matching.
+            # The above extra insert bboxes will be further processed in "formatOutput" function.
+            # After this operation, it will increase TEDS score.
+            no_match_end2end_indexes = find_no_match(
+                match_list, len(end2end_xywh_bboxes), type="end2end"
+            )
+            if len(no_match_end2end_indexes) > 0:
+                no_match_end2end_xywh = end2end_xywh_bboxes[no_match_end2end_indexes]
+                # sort the render no-match end2end bbox in row
+                (
+                    end2end_sorted_indexes_list,
+                    end2end_sorted_bboxes_list,
+                    sorted_groups,
+                    sorted_bboxes_groups,
+                ) = sort_bbox(no_match_end2end_xywh, no_match_end2end_indexes)
+                # make virtual master bboxes, and get matching with the no-match end2end bboxes.
+                extra_match_list = extra_match(
+                    end2end_sorted_indexes_list, len(structure_master_xywh_bboxes)
+                )
+                match_list_add_extra_match = copy.deepcopy(match_list)
+                match_list_add_extra_match.extend(extra_match_list)
+            else:
+                # no no-match end2end bboxes
+                match_list_add_extra_match = copy.deepcopy(match_list)
+                sorted_groups = []
+                sorted_bboxes_groups = []
+
+            match_result_dict = {
+                "match_list": match_list,
+                "match_list_add_extra_match": match_list_add_extra_match,
+                "sorted_groups": sorted_groups,
+                "sorted_bboxes_groups": sorted_bboxes_groups,
+            }
+
+            # format output
+            match_result_dict = self._format(match_result_dict, file_name)
+
+            match_results[file_name] = match_result_dict
+
+        return match_results
+
+    def _format(self, match_result, file_name):
+        """
+        Extend the master token(insert virtual master token), and format matching result.
+        :param match_result:
+        :param file_name:
+        :return:
+        """
+        end2end_info = self.end2end_results[file_name]
+        master_info = self.structure_master_results[file_name]
+        master_token = master_info["text"]
+        sorted_groups = match_result["sorted_groups"]
+
+        # creat virtual master token
+        virtual_master_token_list = []
+        for line_group in sorted_groups:
+            tmp_list = ["<tr>"]
+            item_nums = len(line_group)
+            for _ in range(item_nums):
+                tmp_list.append("<td></td>")
+            tmp_list.append("</tr>")
+            virtual_master_token_list.extend(tmp_list)
+
+        # insert virtual master token
+        master_token_list = master_token.split(",")
+        if master_token_list[-1] == "</tbody>":
+            # complete predict(no cut by max length)
+            # This situation insert virtual master token will drop TEDs score in val set.
+            # So we will not extend virtual token in this situation.
+
+            # fake extend virtual
+            master_token_list[:-1].extend(virtual_master_token_list)
+
+            # real extend virtual
+            # master_token_list = master_token_list[:-1]
+            # master_token_list.extend(virtual_master_token_list)
+            # master_token_list.append('</tbody>')
+
+        elif master_token_list[-1] == "<td></td>":
+            master_token_list.append("</tr>")
+            master_token_list.extend(virtual_master_token_list)
+            master_token_list.append("</tbody>")
+        else:
+            master_token_list.extend(virtual_master_token_list)
+            master_token_list.append("</tbody>")
+
+        # format output
+        match_result.setdefault("matched_master_token_list", master_token_list)
+        return match_result
+
+    def get_merge_result(self, match_results):
+        """
+        Merge the OCR result into structure token to get final results.
+        :param match_results:
+        :return:
+        """
+        merged_results = dict()
+
+        # break_token is linefeed token, when one master bbox has multiply end2end bboxes.
+        break_token = " "
+
+        for idx, (file_name, match_info) in enumerate(match_results.items()):
+            end2end_info = self.end2end_results[file_name]
+            master_token_list = match_info["matched_master_token_list"]
+            match_list = match_info["match_list_add_extra_match"]
+
+            match_dict = get_match_dict(match_list)
+            match_text_dict = get_match_text_dict(match_dict, end2end_info, break_token)
+            merged_result = insert_text_to_token(master_token_list, match_text_dict)
+            merged_result = deal_bb(merged_result)
+
+            merged_results[file_name] = merged_result
+
+        return merged_results
+
+
+class TableMasterMatcher(Matcher):
+    def __init__(self):
+        pass
+
+    def __call__(self, structure_res, dt_boxes, rec_res, img_name=1):
+        end2end_results = {img_name: []}
+        for dt_box, res in zip(dt_boxes, rec_res):
+            d = dict(
+                bbox=np.array(dt_box),
+                text=res[0],
+            )
+            end2end_results[img_name].append(d)
+
+        self.end2end_results = end2end_results
+
+        structure_master_result_dict = {img_name: {}}
+        pred_structures, pred_bboxes = structure_res
+        pred_structures = ",".join(pred_structures[3:-3])
+        structure_master_result_dict[img_name]["text"] = pred_structures
+        structure_master_result_dict[img_name]["bbox"] = pred_bboxes
+        self.structure_master_results = structure_master_result_dict
+
+        # match
+        match_results = self.match()
+        merged_results = self.get_merge_result(match_results)
+        pred_html = merged_results[img_name]
+        pred_html = "<html><body><table>" + pred_html + "</table></body></html>"
+        return pred_html
diff --git a/ppstructure/table/table_metric/__init__.py b/ppstructure/table/table_metric/__init__.py
new file mode 100755
index 0000000..70be931
--- /dev/null
+++ b/ppstructure/table/table_metric/__init__.py
@@ -0,0 +1,16 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+__all__ = ["TEDS"]
+from .table_metric import TEDS
diff --git a/ppstructure/table/table_metric/parallel.py b/ppstructure/table/table_metric/parallel.py
new file mode 100755
index 0000000..f706f35
--- /dev/null
+++ b/ppstructure/table/table_metric/parallel.py
@@ -0,0 +1,55 @@
+from tqdm import tqdm
+from concurrent.futures import ProcessPoolExecutor, as_completed
+
+
+def parallel_process(array, function, n_jobs=16, use_kwargs=False, front_num=0):
+    """
+    A parallel version of the map function with a progress bar.
+    Args:
+        array (array-like): An array to iterate over.
+        function (function): A python function to apply to the elements of array
+        n_jobs (int, default=16): The number of cores to use
+        use_kwargs (boolean, default=False): Whether to consider the elements of array as dictionaries of
+            keyword arguments to function
+        front_num (int, default=3): The number of iterations to run serially before kicking off the parallel job.
+            Useful for catching bugs
+    Returns:
+        [function(array[0]), function(array[1]), ...]
+    """
+    # We run the first few iterations serially to catch bugs
+    if front_num > 0:
+        front = [
+            function(**a) if use_kwargs else function(a) for a in array[:front_num]
+        ]
+    else:
+        front = []
+    # If we set n_jobs to 1, just run a list comprehension. This is useful for benchmarking and debugging.
+    if n_jobs == 1:
+        return front + [
+            function(**a) if use_kwargs else function(a)
+            for a in tqdm(array[front_num:])
+        ]
+    # Assemble the workers
+    with ProcessPoolExecutor(max_workers=n_jobs) as pool:
+        # Pass the elements of array into function
+        if use_kwargs:
+            futures = [pool.submit(function, **a) for a in array[front_num:]]
+        else:
+            futures = [pool.submit(function, a) for a in array[front_num:]]
+        kwargs = {
+            "total": len(futures),
+            "unit": "it",
+            "unit_scale": True,
+            "leave": True,
+        }
+        # Print out the progress as tasks complete
+        for f in tqdm(as_completed(futures), **kwargs):
+            pass
+    out = []
+    # Get the results from the futures.
+    for i, future in tqdm(enumerate(futures)):
+        try:
+            out.append(future.result())
+        except Exception as e:
+            out.append(e)
+    return front + out
diff --git a/ppstructure/table/table_metric/table_metric.py b/ppstructure/table/table_metric/table_metric.py
new file mode 100755
index 0000000..636fed4
--- /dev/null
+++ b/ppstructure/table/table_metric/table_metric.py
@@ -0,0 +1,249 @@
+# Copyright 2020 IBM
+# Author: peter.zhong@au1.ibm.com
+#
+# This is free software; you can redistribute it and/or modify
+# it under the terms of the Apache 2.0 License.
+#
+# This software is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# Apache 2.0 License for more details.
+
+from rapidfuzz.distance import Levenshtein
+from apted import APTED, Config
+from apted.helpers import Tree
+from collections import deque
+from .parallel import parallel_process
+from tqdm import tqdm
+from paddle.utils import try_import
+
+
+class TableTree(Tree):
+    def __init__(self, tag, colspan=None, rowspan=None, content=None, *children):
+        self.tag = tag
+        self.colspan = colspan
+        self.rowspan = rowspan
+        self.content = content
+        self.children = list(children)
+
+    def bracket(self):
+        """Show tree using brackets notation"""
+        if self.tag == "td":
+            result = '"tag": %s, "colspan": %d, "rowspan": %d, "text": %s' % (
+                self.tag,
+                self.colspan,
+                self.rowspan,
+                self.content,
+            )
+        else:
+            result = '"tag": %s' % self.tag
+        for child in self.children:
+            result += child.bracket()
+        return "{{{}}}".format(result)
+
+
+class CustomConfig(Config):
+    def rename(self, node1, node2):
+        """Compares attributes of trees"""
+        # print(node1.tag)
+        if (
+            (node1.tag != node2.tag)
+            or (node1.colspan != node2.colspan)
+            or (node1.rowspan != node2.rowspan)
+        ):
+            return 1.0
+        if node1.tag == "td":
+            if node1.content or node2.content:
+                # print(node1.content, )
+                return Levenshtein.normalized_distance(node1.content, node2.content)
+        return 0.0
+
+
+class CustomConfig_del_short(Config):
+    def rename(self, node1, node2):
+        """Compares attributes of trees"""
+        if (
+            (node1.tag != node2.tag)
+            or (node1.colspan != node2.colspan)
+            or (node1.rowspan != node2.rowspan)
+        ):
+            return 1.0
+        if node1.tag == "td":
+            if node1.content or node2.content:
+                # print('before')
+                # print(node1.content, node2.content)
+                # print('after')
+                node1_content = node1.content
+                node2_content = node2.content
+                if len(node1_content) < 3:
+                    node1_content = ["####"]
+                if len(node2_content) < 3:
+                    node2_content = ["####"]
+                return Levenshtein.normalized_distance(node1_content, node2_content)
+        return 0.0
+
+
+class CustomConfig_del_block(Config):
+    def rename(self, node1, node2):
+        """Compares attributes of trees"""
+        if (
+            (node1.tag != node2.tag)
+            or (node1.colspan != node2.colspan)
+            or (node1.rowspan != node2.rowspan)
+        ):
+            return 1.0
+        if node1.tag == "td":
+            if node1.content or node2.content:
+                node1_content = node1.content
+                node2_content = node2.content
+                while " " in node1_content:
+                    print(node1_content.index(" "))
+                    node1_content.pop(node1_content.index(" "))
+                while " " in node2_content:
+                    print(node2_content.index(" "))
+                    node2_content.pop(node2_content.index(" "))
+                return Levenshtein.normalized_distance(node1_content, node2_content)
+        return 0.0
+
+
+class TEDS(object):
+    """Tree Edit Distance basead Similarity"""
+
+    def __init__(self, structure_only=False, n_jobs=1, ignore_nodes=None):
+        assert isinstance(n_jobs, int) and (
+            n_jobs >= 1
+        ), "n_jobs must be an integer greater than 1"
+        self.structure_only = structure_only
+        self.n_jobs = n_jobs
+        self.ignore_nodes = ignore_nodes
+        self.__tokens__ = []
+
+    def tokenize(self, node):
+        """Tokenizes table cells"""
+        self.__tokens__.append("<%s>" % node.tag)
+        if node.text is not None:
+            self.__tokens__ += list(node.text)
+        for n in node:
+            self.tokenize(n)
+        if node.tag != "unk":
+            self.__tokens__.append("</%s>" % node.tag)
+        if node.tag != "td" and node.tail is not None:
+            self.__tokens__ += list(node.tail)
+
+    def load_html_tree(self, node, parent=None):
+        """Converts HTML tree to the format required by apted"""
+        global __tokens__
+        if node.tag == "td":
+            if self.structure_only:
+                cell = []
+            else:
+                self.__tokens__ = []
+                self.tokenize(node)
+                cell = self.__tokens__[1:-1].copy()
+            new_node = TableTree(
+                node.tag,
+                int(node.attrib.get("colspan", "1")),
+                int(node.attrib.get("rowspan", "1")),
+                cell,
+                *deque(),
+            )
+        else:
+            new_node = TableTree(node.tag, None, None, None, *deque())
+        if parent is not None:
+            parent.children.append(new_node)
+        if node.tag != "td":
+            for n in node:
+                self.load_html_tree(n, new_node)
+        if parent is None:
+            return new_node
+
+    def evaluate(self, pred, true):
+        """Computes TEDS score between the prediction and the ground truth of a
+        given sample
+        """
+        try_import("lxml")
+        from lxml import etree, html
+
+        if (not pred) or (not true):
+            return 0.0
+        parser = html.HTMLParser(remove_comments=True, encoding="utf-8")
+        pred = html.fromstring(pred, parser=parser)
+        true = html.fromstring(true, parser=parser)
+        if pred.xpath("body/table") and true.xpath("body/table"):
+            pred = pred.xpath("body/table")[0]
+            true = true.xpath("body/table")[0]
+            if self.ignore_nodes:
+                etree.strip_tags(pred, *self.ignore_nodes)
+                etree.strip_tags(true, *self.ignore_nodes)
+            n_nodes_pred = len(pred.xpath(".//*"))
+            n_nodes_true = len(true.xpath(".//*"))
+            n_nodes = max(n_nodes_pred, n_nodes_true)
+            tree_pred = self.load_html_tree(pred)
+            tree_true = self.load_html_tree(true)
+            distance = APTED(
+                tree_pred, tree_true, CustomConfig()
+            ).compute_edit_distance()
+            return 1.0 - (float(distance) / n_nodes)
+        else:
+            return 0.0
+
+    def batch_evaluate(self, pred_json, true_json):
+        """Computes TEDS score between the prediction and the ground truth of
+        a batch of samples
+        @params pred_json: {'FILENAME': 'HTML CODE', ...}
+        @params true_json: {'FILENAME': {'html': 'HTML CODE'}, ...}
+        @output: {'FILENAME': 'TEDS SCORE', ...}
+        """
+        samples = true_json.keys()
+        if self.n_jobs == 1:
+            scores = [
+                self.evaluate(pred_json.get(filename, ""), true_json[filename]["html"])
+                for filename in tqdm(samples)
+            ]
+        else:
+            inputs = [
+                {
+                    "pred": pred_json.get(filename, ""),
+                    "true": true_json[filename]["html"],
+                }
+                for filename in samples
+            ]
+            scores = parallel_process(
+                inputs, self.evaluate, use_kwargs=True, n_jobs=self.n_jobs, front_num=1
+            )
+        scores = dict(zip(samples, scores))
+        return scores
+
+    def batch_evaluate_html(self, pred_htmls, true_htmls):
+        """Computes TEDS score between the prediction and the ground truth of
+        a batch of samples
+        """
+        if self.n_jobs == 1:
+            scores = [
+                self.evaluate(pred_html, true_html)
+                for (pred_html, true_html) in zip(pred_htmls, true_htmls)
+            ]
+        else:
+            inputs = [
+                {"pred": pred_html, "true": true_html}
+                for (pred_html, true_html) in zip(pred_htmls, true_htmls)
+            ]
+
+            scores = parallel_process(
+                inputs, self.evaluate, use_kwargs=True, n_jobs=self.n_jobs, front_num=1
+            )
+        return scores
+
+
+if __name__ == "__main__":
+    import json
+    import pprint
+
+    with open("sample_pred.json") as fp:
+        pred_json = json.load(fp)
+    with open("sample_gt.json") as fp:
+        true_json = json.load(fp)
+    teds = TEDS(n_jobs=4)
+    scores = teds.batch_evaluate(pred_json, true_json)
+    pp = pprint.PrettyPrinter()
+    pp.pprint(scores)
diff --git a/ppstructure/table/tablepyxl/__init__.py b/ppstructure/table/tablepyxl/__init__.py
new file mode 100644
index 0000000..1d11e26
--- /dev/null
+++ b/ppstructure/table/tablepyxl/__init__.py
@@ -0,0 +1,13 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/ppstructure/table/tablepyxl/style.py b/ppstructure/table/tablepyxl/style.py
new file mode 100644
index 0000000..dfd0f24
--- /dev/null
+++ b/ppstructure/table/tablepyxl/style.py
@@ -0,0 +1,349 @@
+# This is where we handle translating css styles into openpyxl styles
+# and cascading those from parent to child in the dom.
+
+try:
+    from openpyxl.cell import cell
+    from openpyxl.styles import (
+        Font,
+        Alignment,
+        PatternFill,
+        NamedStyle,
+        Border,
+        Side,
+        Color,
+    )
+    from openpyxl.styles.fills import FILL_SOLID
+    from openpyxl.styles.numbers import FORMAT_CURRENCY_USD_SIMPLE, FORMAT_PERCENTAGE
+    from openpyxl.styles.colors import BLACK
+except:
+    import warnings
+
+    warnings.warn(
+        "Can not import openpyxl, some functions in the ppstructure may not work. Please manually install openpyxl before using ppstructure."
+    )
+
+FORMAT_DATE_MMDDYYYY = "mm/dd/yyyy"
+
+
+def colormap(color):
+    """
+    Convenience for looking up known colors
+    """
+    cmap = {"black": BLACK}
+    return cmap.get(color, color)
+
+
+def style_string_to_dict(style):
+    """
+    Convert css style string to a python dictionary
+    """
+
+    def clean_split(string, delim):
+        return (s.strip() for s in string.split(delim))
+
+    styles = [clean_split(s, ":") for s in style.split(";") if ":" in s]
+    return dict(styles)
+
+
+def get_side(style, name):
+    return {
+        "border_style": style.get("border-{}-style".format(name)),
+        "color": colormap(style.get("border-{}-color".format(name))),
+    }
+
+
+known_styles = {}
+
+
+def style_dict_to_named_style(style_dict, number_format=None):
+    """
+    Change css style (stored in a python dictionary) to openpyxl NamedStyle
+    """
+
+    style_and_format_string = str(
+        {
+            "style_dict": style_dict,
+            "parent": style_dict.parent,
+            "number_format": number_format,
+        }
+    )
+
+    if style_and_format_string not in known_styles:
+        # Font
+        font = Font(
+            bold=style_dict.get("font-weight") == "bold",
+            color=style_dict.get_color("color", None),
+            size=style_dict.get("font-size"),
+        )
+
+        # Alignment
+        alignment = Alignment(
+            horizontal=style_dict.get("text-align", "general"),
+            vertical=style_dict.get("vertical-align"),
+            wrap_text=style_dict.get("white-space", "nowrap") == "normal",
+        )
+
+        # Fill
+        bg_color = style_dict.get_color("background-color")
+        fg_color = style_dict.get_color("foreground-color", Color())
+        fill_type = style_dict.get("fill-type")
+        if bg_color and bg_color != "transparent":
+            fill = PatternFill(
+                fill_type=fill_type or FILL_SOLID,
+                start_color=bg_color,
+                end_color=fg_color,
+            )
+        else:
+            fill = PatternFill()
+
+        # Border
+        border = Border(
+            left=Side(**get_side(style_dict, "left")),
+            right=Side(**get_side(style_dict, "right")),
+            top=Side(**get_side(style_dict, "top")),
+            bottom=Side(**get_side(style_dict, "bottom")),
+            diagonal=Side(**get_side(style_dict, "diagonal")),
+            diagonal_direction=None,
+            outline=Side(**get_side(style_dict, "outline")),
+            vertical=None,
+            horizontal=None,
+        )
+
+        name = "Style {}".format(len(known_styles) + 1)
+
+        pyxl_style = NamedStyle(
+            name=name,
+            font=font,
+            fill=fill,
+            alignment=alignment,
+            border=border,
+            number_format=number_format,
+        )
+
+        known_styles[style_and_format_string] = pyxl_style
+
+    return known_styles[style_and_format_string]
+
+
+class StyleDict(dict):
+    """
+    It's like a dictionary, but it looks for items in the parent dictionary
+    """
+
+    def __init__(self, *args, **kwargs):
+        self.parent = kwargs.pop("parent", None)
+        super(StyleDict, self).__init__(*args, **kwargs)
+
+    def __getitem__(self, item):
+        if item in self:
+            return super(StyleDict, self).__getitem__(item)
+        elif self.parent:
+            return self.parent[item]
+        else:
+            raise KeyError("{} not found".format(item))
+
+    def __hash__(self):
+        return hash(tuple([(k, self.get(k)) for k in self._keys()]))
+
+    # Yielding the keys avoids creating unnecessary data structures
+    # and happily works with both python2 and python3 where the
+    # .keys() method is a dictionary_view in python3 and a list in python2.
+    def _keys(self):
+        yielded = set()
+        for k in self.keys():
+            yielded.add(k)
+            yield k
+        if self.parent:
+            for k in self.parent._keys():
+                if k not in yielded:
+                    yielded.add(k)
+                    yield k
+
+    def get(self, k, d=None):
+        try:
+            return self[k]
+        except KeyError:
+            return d
+
+    def get_color(self, k, d=None):
+        """
+        Strip leading # off colors if necessary
+        """
+        color = self.get(k, d)
+        if hasattr(color, "startswith") and color.startswith("#"):
+            color = color[1:]
+            if (
+                len(color) == 3
+            ):  # Premailers reduces colors like #00ff00 to #0f0, openpyxl doesn't like that
+                color = "".join(2 * c for c in color)
+        return color
+
+
+class Element(object):
+    """
+    Our base class for representing an html element along with a cascading style.
+    The element is created along with a parent so that the StyleDict that we store
+    can point to the parent's StyleDict.
+    """
+
+    def __init__(self, element, parent=None):
+        self.element = element
+        self.number_format = None
+        parent_style = parent.style_dict if parent else None
+        self.style_dict = StyleDict(
+            style_string_to_dict(element.get("style", "")), parent=parent_style
+        )
+        self._style_cache = None
+
+    def style(self):
+        """
+        Turn the css styles for this element into an openpyxl NamedStyle.
+        """
+        if not self._style_cache:
+            self._style_cache = style_dict_to_named_style(
+                self.style_dict, number_format=self.number_format
+            )
+        return self._style_cache
+
+    def get_dimension(self, dimension_key):
+        """
+        Extracts the dimension from the style dict of the Element and returns it as a float.
+        """
+        dimension = self.style_dict.get(dimension_key)
+        if dimension:
+            if dimension[-2:] in ["px", "em", "pt", "in", "cm"]:
+                dimension = dimension[:-2]
+            dimension = float(dimension)
+        return dimension
+
+
+class Table(Element):
+    """
+    The concrete implementations of Elements are semantically named for the types of elements we are interested in.
+    This defines a very concrete tree structure for html tables that we expect to deal with. I prefer this compared to
+    allowing Element to have an arbitrary number of children and dealing with an abstract element tree.
+    """
+
+    def __init__(self, table):
+        """
+        takes an html table object (from lxml)
+        """
+        super(Table, self).__init__(table)
+        table_head = table.find("thead")
+        self.head = (
+            TableHead(table_head, parent=self) if table_head is not None else None
+        )
+        table_body = table.find("tbody")
+        self.body = TableBody(
+            table_body if table_body is not None else table, parent=self
+        )
+
+
+class TableHead(Element):
+    """
+    This class maps to the `<th>` element of the html table.
+    """
+
+    def __init__(self, head, parent=None):
+        super(TableHead, self).__init__(head, parent=parent)
+        self.rows = [TableRow(tr, parent=self) for tr in head.findall("tr")]
+
+
+class TableBody(Element):
+    """
+    This class maps to the `<tbody>` element of the html table.
+    """
+
+    def __init__(self, body, parent=None):
+        super(TableBody, self).__init__(body, parent=parent)
+        self.rows = [TableRow(tr, parent=self) for tr in body.findall("tr")]
+
+
+class TableRow(Element):
+    """
+    This class maps to the `<tr>` element of the html table.
+    """
+
+    def __init__(self, tr, parent=None):
+        super(TableRow, self).__init__(tr, parent=parent)
+        self.cells = [
+            TableCell(cell, parent=self) for cell in tr.findall("th") + tr.findall("td")
+        ]
+
+
+def element_to_string(el):
+    return _element_to_string(el).strip()
+
+
+def _element_to_string(el):
+    string = ""
+
+    for x in el.iterchildren():
+        string += "\n" + _element_to_string(x)
+
+    text = el.text.strip() if el.text else ""
+    tail = el.tail.strip() if el.tail else ""
+
+    return text + string + "\n" + tail
+
+
+class TableCell(Element):
+    """
+    This class maps to the `<td>` element of the html table.
+    """
+
+    CELL_TYPES = {
+        "TYPE_STRING",
+        "TYPE_FORMULA",
+        "TYPE_NUMERIC",
+        "TYPE_BOOL",
+        "TYPE_CURRENCY",
+        "TYPE_PERCENTAGE",
+        "TYPE_NULL",
+        "TYPE_INLINE",
+        "TYPE_ERROR",
+        "TYPE_FORMULA_CACHE_STRING",
+        "TYPE_INTEGER",
+    }
+
+    def __init__(self, cell, parent=None):
+        super(TableCell, self).__init__(cell, parent=parent)
+        self.value = element_to_string(cell)
+        self.number_format = self.get_number_format()
+
+    def data_type(self):
+        cell_types = self.CELL_TYPES & set(self.element.get("class", "").split())
+        if cell_types:
+            if "TYPE_FORMULA" in cell_types:
+                # Make sure TYPE_FORMULA takes precedence over the other classes in the set.
+                cell_type = "TYPE_FORMULA"
+            elif cell_types & {"TYPE_CURRENCY", "TYPE_INTEGER", "TYPE_PERCENTAGE"}:
+                cell_type = "TYPE_NUMERIC"
+            else:
+                cell_type = cell_types.pop()
+        else:
+            cell_type = "TYPE_STRING"
+        return getattr(cell, cell_type)
+
+    def get_number_format(self):
+        if "TYPE_CURRENCY" in self.element.get("class", "").split():
+            return FORMAT_CURRENCY_USD_SIMPLE
+        if "TYPE_INTEGER" in self.element.get("class", "").split():
+            return "#,##0"
+        if "TYPE_PERCENTAGE" in self.element.get("class", "").split():
+            return FORMAT_PERCENTAGE
+        if "TYPE_DATE" in self.element.get("class", "").split():
+            return FORMAT_DATE_MMDDYYYY
+        if self.data_type() == cell.TYPE_NUMERIC:
+            try:
+                int(self.value)
+            except ValueError:
+                return "#,##0.##"
+            else:
+                return "#,##0"
+
+    def format(self, cell):
+        cell.style = self.style()
+        data_type = self.data_type()
+        if data_type:
+            cell.data_type = data_type
diff --git a/ppstructure/table/tablepyxl/tablepyxl.py b/ppstructure/table/tablepyxl/tablepyxl.py
new file mode 100644
index 0000000..92f6536
--- /dev/null
+++ b/ppstructure/table/tablepyxl/tablepyxl.py
@@ -0,0 +1,135 @@
+# Do imports like python3 so our package works for 2 and 3
+from __future__ import absolute_import
+
+
+from tablepyxl.style import Table
+from paddle.utils import try_import
+
+
+def string_to_int(s):
+    if s.isdigit():
+        return int(s)
+    return 0
+
+
+def get_Tables(doc):
+    try_import("lxml")
+    from lxml import etree, html
+
+    tree = html.fromstring(doc)
+    comments = tree.xpath("//comment()")
+    for comment in comments:
+        comment.drop_tag()
+    return [Table(table) for table in tree.xpath("//table")]
+
+
+def write_rows(worksheet, elem, row, column=1):
+    """
+    Writes every tr child element of elem to a row in the worksheet
+    returns the next row after all rows are written
+    """
+    try_import("openpyxl")
+    from openpyxl.cell.cell import MergedCell
+    from openpyxl.utils import get_column_letter
+
+    initial_column = column
+    for table_row in elem.rows:
+        for table_cell in table_row.cells:
+            cell = worksheet.cell(row=row, column=column)
+            while isinstance(cell, MergedCell):
+                column += 1
+                cell = worksheet.cell(row=row, column=column)
+
+            colspan = string_to_int(table_cell.element.get("colspan", "1"))
+            rowspan = string_to_int(table_cell.element.get("rowspan", "1"))
+            if rowspan > 1 or colspan > 1:
+                worksheet.merge_cells(
+                    start_row=row,
+                    start_column=column,
+                    end_row=row + rowspan - 1,
+                    end_column=column + colspan - 1,
+                )
+
+            cell.value = table_cell.value
+            table_cell.format(cell)
+            min_width = table_cell.get_dimension("min-width")
+            max_width = table_cell.get_dimension("max-width")
+
+            if colspan == 1:
+                # Initially, when iterating for the first time through the loop, the width of all the cells is None.
+                # As we start filling in contents, the initial width of the cell (which can be retrieved by:
+                # worksheet.column_dimensions[get_column_letter(column)].width) is equal to the width of the previous
+                # cell in the same column (i.e. width of A2 = width of A1)
+                width = max(
+                    worksheet.column_dimensions[get_column_letter(column)].width or 0,
+                    len(table_cell.value) + 2,
+                )
+                if max_width and width > max_width:
+                    width = max_width
+                elif min_width and width < min_width:
+                    width = min_width
+                worksheet.column_dimensions[get_column_letter(column)].width = width
+            column += colspan
+        row += 1
+        column = initial_column
+    return row
+
+
+def table_to_sheet(table, wb):
+    """
+    Takes a table and workbook and writes the table to a new sheet.
+    The sheet title will be the same as the table attribute name.
+    """
+    ws = wb.create_sheet(title=table.element.get("name"))
+    insert_table(table, ws, 1, 1)
+
+
+def document_to_workbook(doc, wb=None, base_url=None):
+    """
+    Takes a string representation of an html document and writes one sheet for
+    every table in the document.
+    The workbook is returned
+    """
+    try_import("premailer")
+    try_import("openpyxl")
+    from premailer import Premailer
+    from openpyxl import Workbook
+
+    if not wb:
+        wb = Workbook()
+        wb.remove(wb.active)
+
+    inline_styles_doc = Premailer(
+        doc, base_url=base_url, remove_classes=False
+    ).transform()
+    tables = get_Tables(inline_styles_doc)
+
+    for table in tables:
+        table_to_sheet(table, wb)
+
+    return wb
+
+
+def document_to_xl(doc, filename, base_url=None):
+    """
+    Takes a string representation of an html document and writes one sheet for
+    every table in the document. The workbook is written out to a file called filename
+    """
+    wb = document_to_workbook(doc, base_url=base_url)
+    wb.save(filename)
+
+
+def insert_table(table, worksheet, column, row):
+    if table.head:
+        row = write_rows(worksheet, table.head, row, column)
+    if table.body:
+        row = write_rows(worksheet, table.body, row, column)
+
+
+def insert_table_at_cell(table, cell):
+    """
+    Inserts a table at the location of an openpyxl Cell object.
+    """
+    ws = cell.parent
+    column, row = cell.column, cell.row
+    insert_table(table, ws, column, row)
diff --git a/ppstructure/utility.py b/ppstructure/utility.py
new file mode 100644
index 0000000..97e434d
--- /dev/null
+++ b/ppstructure/utility.py
@@ -0,0 +1,298 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import random
+import ast
+import PIL
+from PIL import Image, ImageDraw, ImageFont
+import numpy as np
+from tools.infer.utility import (
+    draw_ocr_box_txt,
+    str2bool,
+    str2int_tuple,
+    init_args as infer_args,
+)
+import math
+
+
+def init_args():
+    parser = infer_args()
+
+    # params for output
+    parser.add_argument("--output", type=str, default="./output")
+    # params for table structure
+    parser.add_argument("--table_max_len", type=int, default=488)
+    parser.add_argument("--table_algorithm", type=str, default="TableAttn")
+    parser.add_argument("--table_model_dir", type=str)
+    parser.add_argument("--merge_no_span_structure", type=str2bool, default=True)
+    parser.add_argument(
+        "--table_char_dict_path",
+        type=str,
+        default="../ppocr/utils/dict/table_structure_dict_ch.txt",
+    )
+    # params for formula recognition
+    parser.add_argument("--formula_algorithm", type=str, default="LaTeXOCR")
+    parser.add_argument("--formula_model_dir", type=str)
+    parser.add_argument(
+        "--formula_char_dict_path",
+        type=str,
+        default="../ppocr/utils/dict/latex_ocr_tokenizer.json",
+    )
+    parser.add_argument("--formula_batch_num", type=int, default=1)
+    # params for layout
+    parser.add_argument("--layout_model_dir", type=str)
+    parser.add_argument(
+        "--layout_dict_path",
+        type=str,
+        default="../ppocr/utils/dict/layout_dict/layout_publaynet_dict.txt",
+    )
+    parser.add_argument(
+        "--layout_score_threshold", type=float, default=0.5, help="Threshold of score."
+    )
+    parser.add_argument(
+        "--layout_nms_threshold", type=float, default=0.5, help="Threshold of nms."
+    )
+    # params for kie
+    parser.add_argument("--kie_algorithm", type=str, default="LayoutXLM")
+    parser.add_argument("--ser_model_dir", type=str)
+    parser.add_argument("--re_model_dir", type=str)
+    parser.add_argument("--use_visual_backbone", type=str2bool, default=True)
+    parser.add_argument(
+        "--ser_dict_path", type=str, default="../train_data/XFUND/class_list_xfun.txt"
+    )
+    # need to be None or tb-yx
+    parser.add_argument("--ocr_order_method", type=str, default=None)
+    # params for inference
+    parser.add_argument(
+        "--mode",
+        type=str,
+        choices=["structure", "kie"],
+        default="structure",
+        help="structure and kie is supported",
+    )
+    parser.add_argument(
+        "--image_orientation",
+        type=bool,
+        default=False,
+        help="Whether to enable image orientation recognition",
+    )
+    parser.add_argument(
+        "--layout",
+        type=str2bool,
+        default=True,
+        help="Whether to enable layout analysis",
+    )
+    parser.add_argument(
+        "--table",
+        type=str2bool,
+        default=True,
+        help="In the forward, whether the table area uses table recognition",
+    )
+    parser.add_argument(
+        "--formula",
+        type=str2bool,
+        default=False,
+        help="Whether to enable formula recognition",
+    )
+    parser.add_argument(
+        "--ocr",
+        type=str2bool,
+        default=True,
+        help="In the forward, whether the non-table area is recognition by ocr",
+    )
+    # param for recovery
+    parser.add_argument(
+        "--recovery",
+        type=str2bool,
+        default=False,
+        help="Whether to enable layout of recovery",
+    )
+    parser.add_argument(
+        "--recovery_to_markdown",
+        type=str2bool,
+        default=False,
+        help="Whether to enable layout of recovery to markdown",
+    )
+    parser.add_argument(
+        "--use_pdf2docx_api",
+        type=str2bool,
+        default=False,
+        help="Whether to use pdf2docx api",
+    )
+    parser.add_argument(
+        "--invert",
+        type=str2bool,
+        default=False,
+        help="Whether to invert image before processing",
+    )
+    parser.add_argument(
+        "--binarize",
+        type=str2bool,
+        default=False,
+        help="Whether to threshold binarize image before processing",
+    )
+    parser.add_argument(
+        "--alphacolor",
+        type=str2int_tuple,
+        default=(255, 255, 255),
+        help="Replacement color for the alpha channel, if the latter is present; R,G,B integers",
+    )
+
+    return parser
+
+
+def parse_args():
+    parser = init_args()
+    return parser.parse_args()
+
+
+def draw_structure_result(image, result, font_path):
+    if isinstance(image, np.ndarray):
+        image = Image.fromarray(image)
+    boxes, txts, scores = [], [], []
+
+    img_layout = image.copy()
+    draw_layout = ImageDraw.Draw(img_layout)
+    text_color = (255, 255, 255)
+    text_background_color = (80, 127, 255)
+    catid2color = {}
+    font_size = 15
+    font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
+
+    for region in result:
+        if region["type"] not in catid2color:
+            box_color = (
+                random.randint(0, 255),
+                random.randint(0, 255),
+                random.randint(0, 255),
+            )
+            catid2color[region["type"]] = box_color
+        else:
+            box_color = catid2color[region["type"]]
+        box_layout = region["bbox"]
+        draw_layout.rectangle(
+            [(box_layout[0], box_layout[1]), (box_layout[2], box_layout[3])],
+            outline=box_color,
+            width=3,
+        )
+
+        if int(PIL.__version__.split(".")[0]) < 10:
+            text_w, text_h = font.getsize(region["type"])
+        else:
+            left, top, right, bottom = font.getbbox(region["type"])
+            text_w, text_h = right - left, bottom - top
+
+        draw_layout.rectangle(
+            [
+                (box_layout[0], box_layout[1]),
+                (box_layout[0] + text_w, box_layout[1] + text_h),
+            ],
+            fill=text_background_color,
+        )
+        draw_layout.text(
+            (box_layout[0], box_layout[1]), region["type"], fill=text_color, font=font
+        )
+
+        if region["type"] == "table" or (
+            region["type"] == "equation" and "latex" in region["res"]
+        ):
+            pass
+        else:
+            for text_result in region["res"]:
+                boxes.append(np.array(text_result["text_region"]))
+                txts.append(text_result["text"])
+                scores.append(text_result["confidence"])
+
+                if "text_word_region" in text_result:
+                    for word_region in text_result["text_word_region"]:
+                        char_box = word_region
+                        box_height = int(
+                            math.sqrt(
+                                (char_box[0][0] - char_box[3][0]) ** 2
+                                + (char_box[0][1] - char_box[3][1]) ** 2
+                            )
+                        )
+                        box_width = int(
+                            math.sqrt(
+                                (char_box[0][0] - char_box[1][0]) ** 2
+                                + (char_box[0][1] - char_box[1][1]) ** 2
+                            )
+                        )
+                        if box_height == 0 or box_width == 0:
+                            continue
+                        boxes.append(word_region)
+                        txts.append("")
+                        scores.append(1.0)
+
+    im_show = draw_ocr_box_txt(
+        img_layout, boxes, txts, scores, font_path=font_path, drop_score=0
+    )
+    return im_show
+
+
+def cal_ocr_word_box(rec_str, box, rec_word_info):
+    """Calculate the detection frame for each word based on the results of recognition and detection of ocr"""
+
+    col_num, word_list, word_col_list, state_list = rec_word_info
+    box = box.tolist()
+    bbox_x_start = box[0][0]
+    bbox_x_end = box[1][0]
+    bbox_y_start = box[0][1]
+    bbox_y_end = box[2][1]
+
+    cell_width = (bbox_x_end - bbox_x_start) / col_num
+
+    word_box_list = []
+    word_box_content_list = []
+    cn_width_list = []
+    cn_col_list = []
+    for word, word_col, state in zip(word_list, word_col_list, state_list):
+        if state == "cn":
+            if len(word_col) != 1:
+                char_seq_length = (word_col[-1] - word_col[0] + 1) * cell_width
+                char_width = char_seq_length / (len(word_col) - 1)
+                cn_width_list.append(char_width)
+            cn_col_list += word_col
+            word_box_content_list += word
+        else:
+            cell_x_start = bbox_x_start + int(word_col[0] * cell_width)
+            cell_x_end = bbox_x_start + int((word_col[-1] + 1) * cell_width)
+            cell = (
+                (cell_x_start, bbox_y_start),
+                (cell_x_end, bbox_y_start),
+                (cell_x_end, bbox_y_end),
+                (cell_x_start, bbox_y_end),
+            )
+            word_box_list.append(cell)
+            word_box_content_list.append("".join(word))
+    if len(cn_col_list) != 0:
+        if len(cn_width_list) != 0:
+            avg_char_width = np.mean(cn_width_list)
+        else:
+            avg_char_width = (bbox_x_end - bbox_x_start) / len(rec_str)
+        for center_idx in cn_col_list:
+            center_x = (center_idx + 0.5) * cell_width
+            cell_x_start = max(int(center_x - avg_char_width / 2), 0) + bbox_x_start
+            cell_x_end = (
+                min(int(center_x + avg_char_width / 2), bbox_x_end - bbox_x_start)
+                + bbox_x_start
+            )
+            cell = (
+                (cell_x_start, bbox_y_start),
+                (cell_x_end, bbox_y_start),
+                (cell_x_end, bbox_y_end),
+                (cell_x_start, bbox_y_end),
+            )
+            word_box_list.append(cell)
+
+    return word_box_content_list, word_box_list
diff --git a/pyproject.toml b/pyproject.toml
new file mode 100644
index 0000000..184dec7
--- /dev/null
+++ b/pyproject.toml
@@ -0,0 +1,87 @@
+[build-system]
+requires = ["setuptools==72.1.0", "wheel", "setuptools_scm"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "paddleocr"
+# After each version release, the version number needs to be incremented
+dynamic = ["version"]
+description = "Awesome multilingual OCR and document parsing toolkits based on PaddlePaddle"
+authors = [
+    {name = "PaddlePaddle", email = "paddleocr@baidu.com"},
+]
+maintainers = [
+    {name = "PaddlePaddle", email = "paddleocr@baidu.com"},
+]
+readme = "README.md"
+requires-python = ">=3.8"
+keywords = [
+    "ocr",
+    "textdetection",
+    "textrecognition",
+    "paddleocr",
+    "chineseocr",
+    "pp-ocr",
+    "pp-structure",
+    "document-parsing",
+    "pdf2markdown",
+    "doc2markdown",
+    "docx2markdown",
+    "kie",
+    "pp-chatocr",
+    "document-translation",
+    "rag"
+]
+license = {text = "Apache License 2.0"}
+classifiers = [
+    "Intended Audience :: Developers",
+    "Operating System :: OS Independent",
+    "Natural Language :: Chinese (Simplified)",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.8",
+    "Programming Language :: Python :: 3.9",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+    "Topic :: Utilities",
+]
+dependencies = [
+    "paddlex[ocr-core]>=3.7.0,<3.8.0",
+    "PyYAML>=6",
+    "requests",
+    "aiohttp>=3.8.0",
+    "typing-extensions>=4.12",
+]
+
+[project.urls]
+homepage = "https://github.com/PaddlePaddle/PaddleOCR"
+documentation = "https://github.com/PaddlePaddle/PaddleOCR/blob/main/README.md"
+repository = "https://github.com/PaddlePaddle/PaddleOCR.git"
+issues = "https://github.com/PaddlePaddle/PaddleOCR/issues"
+
+[project.scripts]
+paddleocr = "paddleocr.__main__:console_entry"
+
+[project.optional-dependencies]
+doc-parser = ["paddlex[ocr,genai-client]>=3.7.0,<3.8.0"]
+ie = ["paddlex[ie]>=3.7.0,<3.8.0"]
+trans = ["paddlex[trans]>=3.7.0,<3.8.0"]
+doc2md = ["python-docx>=0.8.11", "python-pptx>=0.6.21", "openpyxl>=3.0.0", "pylatexenc>=2.10,<3"]
+all = ["paddlex[ocr,genai-client,ie,trans]>=3.7.0,<3.8.0", "python-docx>=0.8.11", "python-pptx>=0.6.21", "openpyxl>=3.0.0"]
+
+[tool.setuptools.packages.find]
+where = ["."]
+include = ["paddleocr", "paddleocr.*"]
+namespaces = false
+
+[tool.setuptools_scm]
+version_scheme = 'release-branch-semver'
+git_describe_command = ["git", "describe", "--dirty", "--tags", "--long", "--match", "v[0-9]*"]
+
+[tool.pytest.ini_options]
+markers = [
+    "resource_intensive: mark a test as resource intensive",
+    "py38_incompatible: test requires extras or deps that need Python 3.9+"
+]
+addopts = "-m 'not resource_intensive'"
diff --git a/readme/README_ar.md b/readme/README_ar.md
new file mode 100644
index 0000000..34c5f7e
--- /dev/null
+++ b/readme/README_ar.md
@@ -0,0 +1,305 @@
+
+
+<div align="center">
+  <p>
+      <img width="800" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Banner.png" alt="تاريخ النجوم">
+  </p>
+
+
+
+<h3>مجموعة أدوات التعرف الضوئي على الحروف (OCR) الرائدة عالمياً ومحرك الذكاء الاصطناعي للمستندات</h3>
+
+[English](../README.md) | [简体中文](./README_cn.md) | [繁體中文](./README_tcn.md) | [日本語](./README_ja.md) | [한국어](./README_ko.md) | [Français](./README_fr.md) | [Русский](./README_ru.md) | [Español](./README_es.md) | العربية
+
+<!-- icon -->
+
+[![تنزيلات PyPI](https://static.pepy.tech/badge/paddleocr)](https://pepy.tech/projects/paddleocr)
+[![مُستخدَم بواسطة](https://img.shields.io/badge/Used%20by-6k%2B%20repositories-blue)](https://github.com/PaddlePaddle/PaddleOCR/network/dependents)
+![python](https://img.shields.io/badge/python-3.8~3.12-aff.svg)
+![os](https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-pink.svg)
+![hardware](https://img.shields.io/badge/hardware-cpu%2C%20gpu%2C%20xpu%2C%20npu-yellow.svg)
+
+[![AI Studio](https://img.shields.io/badge/PaddleOCR-_Offiical_Website-1927BA?logo=data:image/png;base64,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&labelColor=white)](https://www.paddleocr.com)
+[![اسأل DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/PaddlePaddle/PaddleOCR)
+[![الرخصة](https://img.shields.io/badge/license-Apache_2.0-green)](../LICENSE)
+
+</div>
+
+
+
+
+
+
+**يحوّل PaddleOCR المستندات والصور إلى بيانات منظمة جاهزة للنماذج اللغوية الكبيرة (JSON/Markdown) بدقة رائدة في المجال. بأكثر من 70 ألف نجمة وثقة مشاريع رائدة مثل Dify وRAGFlow وCherry Studio، يُعد PaddleOCR الأساس المتين لبناء تطبيقات RAG والتطبيقات الوكيلية الذكية.**
+
+
+## 🚀 الميزات الرئيسية
+
+### 📄 تحليل ذكي للمستندات (جاهز للنماذج اللغوية الكبيرة)
+> *تحويل المرئيات المعقدة إلى بيانات منظمة لعصر النماذج اللغوية الكبيرة.*
+
+* **نموذج رؤية-لغة رائد للمستندات**: يتميز بنموذج **PaddleOCR-VL-1.6 (0.9B)**، النموذج خفيف الحجم الرائد في المجال للرؤية واللغة لتحليل المستندات. يحقق دقة 96.3% على OmniDocBench v1.6، ويتصدر في التعرف على النصوص والصيغ والجداول، مع تحسينات كبيرة في المستندات القديمة والأحرف النادرة والأختام والرسوم البيانية، مع مخرجات منظمة بصيغ **Markdown** و**JSON**.
+* **تحويل مدرك للبنية**: بدعم من **PP-StructureV3**، يتم تحويل ملفات PDF والصور المعقدة بسلاسة إلى **Markdown** أو **JSON**. على عكس نماذج سلسلة PaddleOCR-VL، يوفر معلومات إحداثية أدق تشمل إحداثيات خلايا الجداول وإحداثيات النصوص وغيرها.
+* **كفاءة جاهزة للإنتاج**: تحقيق دقة بمستوى تجاري مع حجم صغير للغاية. يتفوق على العديد من الحلول المغلقة المصدر في المعايير المرجعية العامة مع الحفاظ على كفاءة استخدام الموارد للنشر على الأجهزة الطرفية والسحابية.
+
+### 🔍 التعرف الشامل على النصوص (OCR للمشاهد)
+> *المعيار الذهبي العالمي للكشف السريع عن النصوص متعددة اللغات.*
+
+* **دعم أكثر من 100 لغة**: تعرف أصلي على مكتبة عالمية واسعة. **PP-OCRv6** يدعم 50 لغة بنموذج واحد موحد (الصينية، الإنجليزية، اليابانية، و46 لغة لاتينية) — لا حاجة لتبديل النماذج.
+* **إتقان العناصر المعقدة**: بالإضافة إلى التعرف القياسي على النصوص، ندعم **الكشف عن النصوص في المشاهد الطبيعية** عبر مجموعة واسعة من البيئات، بما في ذلك بطاقات الهوية، ومشاهد الشوارع، والكتب، والمكونات الصناعية.
+* **قفزة في الأداء**: يحقق PP-OCRv6 تحسيناً في دقة الكشف بنسبة **4.6%** والتعرف بنسبة **5.1%** مقارنة بـ PP-OCRv5، متفوقاً على النماذج اللغوية البصرية الرئيسية. تسريع 5.2× في الاستدلال على وحدة المعالجة المركزية.
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Arch.jpg" alt="هندسة PaddleOCR">
+  </p>
+</div>
+
+### 🛠️ منظومة مُركّزة على المطورين
+* **تكامل سلس**: الخيار الأمثل لمنظومة الوكلاء الذكيين — متكامل بعمق مع **Dify وRAGFlow وPathway وCherry Studio**.
+* **حلقة بيانات النماذج اللغوية الكبيرة**: خط أنابيب كامل لبناء مجموعات بيانات عالية الجودة، يوفر "محرك بيانات" مستداماً لضبط النماذج اللغوية الكبيرة.
+* **نشر بنقرة واحدة**: دعم لمختلف منصات العتاد (وحدات معالجة الرسومات NVIDIA، معالجات Intel، وحدات Kunlunxin XPU، ومسرّعات الذكاء الاصطناعي المتنوعة).
+
+
+## 📣 آخر التحديثات
+
+### 🔥 2026.06.11: إصدار PaddleOCR 3.7.0
+- أبرز مزايا PP-OCRv6:
+
+    - **تحسين الدقة**: يحقق الطراز medium تحسيناً +4.6% في الكشف و+5.1% في التعرف مقارنة بـ PP-OCRv5_server، متفوقاً على النماذج اللغوية البصرية الرئيسية بـ 34.5M معلمة فقط.
+    - **دعم 50 لغة موحد**: نموذج واحد يغطي الصينية والإنجليزية واليابانية و46 لغة لاتينية — لا حاجة لتبديل النماذج.
+    - **سيناريوهات متخصصة**: تحسينات كبيرة في الشاشات الرقمية وأحرف المصفوفة النقطية وبصمات الإطارات والنصوص الصناعية.
+    - **استدلال أسرع**: تسريع 5.2× على CPU (OpenVINO)، 6.1× على Apple M4 (tiny)، 0.13 ثانية على A100 GPU.
+    - **ثلاثة مستويات لجميع السيناريوهات**: tiny (1.5M) / small (7.7M) / medium (34.5M) للأجهزة الطرفية والمحمولة والخوادم.
+    - **توفر النماذج**: جميع النماذج متاحة على [HuggingFace](https://huggingface.co/collections/PaddlePaddle/pp-ocrv6) و [ModelScope](https://www.modelscope.cn/collections/PaddlePaddle/PP-OCRv6).
+
+<details>
+<summary><strong>2026.05.28: إصدار PaddleOCR 3.6.0</strong></summary>
+
+- أبرز مزايا PaddleOCR-VL-1.6:
+
+    - **دقة SOTA جديدة**: 96.3% على OmniDocBench v1.6، SOTA جديد على OmniDocBench v1.5 وReal5-OmniDocBench.
+    - **ترقية شاملة**: تحسينات كبيرة في الجداول والمستندات القديمة والأحرف النادرة.
+    - **ترحيل سلس**: بنية متوافقة تماماً مع PaddleOCR-VL-1.5.
+    - **جرّبه الآن**: [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.6) أو [الموقع الرسمي](https://www.paddleocr.com).
+
+</details>
+<details>
+<summary><strong>2026.04.21: إصدار PaddleOCR 3.5.0</strong></summary>
+
+* **مرونة أكبر في تبديل واجهات الاستدلال**: يدعم التبديل السلس بين الرسم البياني الثابت في Paddle، والرسم البياني الديناميكي في Paddle، وTransformers. أصبح PaddleOCR الآن متكاملاً بعمق مع منظومة Hugging Face، كما تدعم 20 من النماذج الرئيسية استخدام Transformers كواجهة للاستدلال.
+* **تحويل مستندات Office إلى Markdown**: يدعم تحويل صيغ المستندات الشائعة مثل Word وExcel وPowerPoint إلى Markdown.
+* **تصدير نتائج التحليل إلى DOCX**: أصبحت سلاسل `PaddleOCR-VL` و`PP-StructureV3` و`PP-DocTranslation` تدعم الآن تصدير نتائج التحليل إلى DOCX لسهولة العرض والتحرير في Microsoft Word.
+* **حزمة SDK الرسمية للاستدلال في المتصفح**: تم إصدار `PaddleOCR.js`، وهي حزمة SDK الرسمية للاستدلال في المتصفح، وتدعم تشغيل `PP-OCRv5` مباشرة داخل المتصفح.
+
+</details>
+
+<details>
+<summary><strong>2026.01.29: إصدار PaddleOCR 3.4.0</strong></summary>
+* PaddleOCR-VL-1.5 (نموذج VLM رائد بحجم 0.9B): نموذجنا الرائد الأحدث لتحليل المستندات متاح الآن!
+    * **دقة 94.5% على OmniDocBench**: يتفوق على النماذج الكبيرة العامة الرائدة ومحللات المستندات المتخصصة.
+    * **متانة في العالم الواقعي**: أول من يقدم خوارزمية **PP-DocLayoutV3** لتحديد موقع الأشكال غير المنتظمة، مع إتقان 5 سيناريوهات صعبة: *الميل، التشوه، المسح الضوئي، الإضاءة، وتصوير الشاشة*.
+    * **توسيع القدرات**: يدعم الآن **التعرف على الأختام**، و**الكشف عن النصوص**، ويتوسع ليشمل **111 لغة** (بما في ذلك الخط التبتي الصيني والبنغالية).
+    * **إتقان المستندات الطويلة**: يدعم الدمج التلقائي للجداول عبر الصفحات وتحديد العناوين الهرمية.
+    * **جرّبه الآن**: متاح على [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.5) أو [موقعنا الرسمي](https://www.paddleocr.com).
+
+</details>
+
+<details>
+<summary><strong>2025.10.16: إصدار PaddleOCR 3.3.0</strong></summary>
+
+- إصدار PaddleOCR-VL:
+    - **تقديم النموذج**:
+        - **PaddleOCR-VL** هو نموذج رائد وموفر للموارد مصمم خصيصاً لتحليل المستندات. مكونه الأساسي هو PaddleOCR-VL-0.9B، وهو نموذج رؤية-لغة (VLM) صغير الحجم لكنه قوي يدمج مشفر بصري ديناميكي الدقة بأسلوب NaViT مع النموذج اللغوي ERNIE-4.5-0.3B لتمكين التعرف الدقيق على العناصر. **يدعم هذا النموذج المبتكر 109 لغات بكفاءة ويتفوق في التعرف على العناصر المعقدة (مثل النصوص والجداول والصيغ الرياضية والرسوم البيانية)، مع الحفاظ على الحد الأدنى من استهلاك الموارد**. من خلال التقييمات الشاملة على المعايير المرجعية العامة المستخدمة على نطاق واسع والمعايير الداخلية، يحقق PaddleOCR-VL أداءً رائداً في كل من تحليل المستندات على مستوى الصفحة والتعرف على العناصر. يتفوق بشكل كبير على الحلول الحالية، ويُظهر تنافسية قوية أمام نماذج VLM الرائدة، ويقدم سرعات استدلال عالية. هذه المزايا تجعله مناسباً جداً للنشر العملي في السيناريوهات الواقعية. تم إصدار النموذج على [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL). نرحب بالجميع لتنزيله واستخدامه! يمكن العثور على مزيد من المعلومات في [PaddleOCR-VL](https://www.paddleocr.ai/latest/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL.html).
+
+    - **الميزات الأساسية**:
+        - **بنية VLM صغيرة لكنها قوية**: نقدم نموذج رؤية-لغة مبتكراً مصمماً خصيصاً للاستدلال الموفر للموارد، يحقق أداءً متميزاً في التعرف على العناصر. من خلال دمج مشفر بصري ديناميكي عالي الدقة بأسلوب NaViT مع النموذج اللغوي خفيف الحجم ERNIE-4.5-0.3B، نعزز بشكل كبير قدرات التعرف وكفاءة فك التشفير. يحافظ هذا الدمج على دقة عالية مع تقليل المتطلبات الحسابية، مما يجعله مناسباً تماماً لتطبيقات معالجة المستندات الفعالة والعملية.
+        - **أداء رائد في تحليل المستندات**: يحقق PaddleOCR-VL أداءً رائداً في كل من تحليل المستندات على مستوى الصفحة والتعرف على العناصر. يتفوق بشكل كبير على الحلول القائمة على خطوط الأنابيب ويُظهر تنافسية قوية أمام نماذج الرؤية-اللغة (VLMs) الرائدة في تحليل المستندات. علاوة على ذلك، يتفوق في التعرف على عناصر المستندات المعقدة، مثل النصوص والجداول والصيغ الرياضية والرسوم البيانية، مما يجعله مناسباً لمجموعة واسعة من أنواع المحتوى الصعبة، بما في ذلك النصوص المكتوبة بخط اليد والمستندات التاريخية. وهذا يجعله متعدد الاستخدامات ومناسباً لمجموعة واسعة من أنواع المستندات والسيناريوهات.
+        - **دعم متعدد اللغات**: يدعم PaddleOCR-VL 109 لغات، تغطي اللغات العالمية الرئيسية، بما في ذلك على سبيل المثال لا الحصر الصينية والإنجليزية واليابانية واللاتينية والكورية، بالإضافة إلى لغات ذات خطوط وبنى مختلفة مثل الروسية (الخط السيريلي) والعربية والهندية (خط ديفاناغاري) والتايلاندية. تعزز هذه التغطية اللغوية الواسعة بشكل كبير من قابلية تطبيق نظامنا في سيناريوهات معالجة المستندات متعددة اللغات والعالمية.
+
+- إصدار نموذج التعرف متعدد اللغات PP-OCRv5:
+    - تحسين دقة وتغطية التعرف على الخط اللاتيني؛ إضافة دعم للأنظمة الكتابية السيريلية والعربية والديفاناغارية والتيلوغية والتاميلية وغيرها، مع تغطية التعرف على 109 لغات. يحتوي النموذج على 2 مليون معامل فقط، وقد زادت دقة بعض النماذج بأكثر من 40% مقارنة بالجيل السابق.
+
+</details>
+
+
+<details>
+<summary><strong>2025.08.21: إصدار PaddleOCR 3.2.0</strong></summary>
+
+- **إضافات نموذجية مهمة:**
+    - تقديم التدريب والاستدلال والنشر لنماذج التعرف PP-OCRv5 باللغات الإنجليزية والتايلاندية واليونانية. **يقدم نموذج PP-OCRv5 الإنجليزي تحسيناً بنسبة 11% في السيناريوهات الإنجليزية مقارنة بنموذج PP-OCRv5 الرئيسي، مع تحقيق نماذج التعرف التايلاندي واليوناني دقة 82.68% و89.28% على التوالي.**
+
+- **ترقيات قدرات النشر:**
+    - **دعم كامل لإصداري إطار عمل PaddlePaddle 3.1.0 و3.1.1.**
+    - **ترقية شاملة لحل النشر المحلي بلغة C++ لـ PP-OCRv5، يدعم الآن كلاً من Linux وWindows، مع تكافؤ الميزات ودقة مطابقة لتنفيذ Python.**
+    - **يدعم الاستدلال عالي الأداء الآن CUDA 12، ويمكن إجراء الاستدلال باستخدام واجهة Paddle Inference أو ONNX Runtime.**
+    - **حل النشر الخدمي عالي الاستقرار أصبح الآن مفتوح المصدر بالكامل، مما يتيح للمستخدمين تخصيص صور Docker وحزم SDK حسب الحاجة.**
+    - يدعم حل النشر الخدمي عالي الاستقرار أيضاً الاستدعاء عبر طلبات HTTP المُنشأة يدوياً، مما يتيح تطوير الشيفرة البرمجية من جانب العميل بأي لغة برمجة.
+
+- **دعم المعايير المرجعية:**
+    - **تدعم جميع خطوط الإنتاج الآن قياس الأداء الدقيق، مما يتيح قياس زمن الاستدلال الشامل بالإضافة إلى بيانات زمن الاستجابة لكل طبقة ووحدة للمساعدة في تحليل الأداء. [إليك](docs/version3.x/pipeline_usage/instructions/benchmark.en.md) كيفية إعداد واستخدام ميزة قياس الأداء.**
+    - **تم تحديث الوثائق لتشمل المقاييس الرئيسية للتكوينات الشائعة الاستخدام على العتاد الرئيسي، مثل زمن الاستجابة واستخدام الذاكرة، لتوفير مراجع النشر للمستخدمين.**
+
+- **إصلاح الأخطاء:**
+    - حل مشكلة فشل حفظ السجلات أثناء تدريب النموذج.
+    - ترقية مكون تعزيز البيانات لنماذج الصيغ الرياضية للتوافق مع الإصدارات الأحدث من مكتبة albumentations، وإصلاح تحذيرات الجمود عند استخدام حزمة tokenizers في سيناريوهات متعددة العمليات.
+    - إصلاح التناقضات في سلوكيات المفاتيح (مثل `use_chart_parsing`) في ملفات تكوين PP-StructureV3 مقارنة بخطوط الأنابيب الأخرى.
+
+- **تحسينات أخرى:**
+    - **فصل التبعيات الأساسية والاختيارية. التبعيات الأساسية الدنيا فقط مطلوبة للتعرف الأساسي على النصوص؛ يمكن تثبيت التبعيات الإضافية لتحليل المستندات واستخراج المعلومات حسب الحاجة.**
+    - **تمكين دعم بطاقات الرسومات NVIDIA RTX من سلسلة 50 على Windows؛ يمكن للمستخدمين الرجوع إلى [دليل التثبيت](docs/version3.x/installation.en.md) لإصدارات إطار PaddlePaddle المقابلة.**
+    - **تدعم نماذج سلسلة PP-OCR الآن إرجاع إحداثيات الأحرف المفردة.**
+    - إضافة مصادر تنزيل النماذج من AIStudio وModelScope وغيرها، مما يتيح للمستخدمين تحديد مصدر تنزيل النماذج.
+    - إضافة دعم تحويل الرسوم البيانية إلى جداول عبر وحدة PP-Chart2Table.
+    - تحسين أوصاف الوثائق لتعزيز قابلية الاستخدام.
+</details>
+
+
+[سجل التحديثات](https://paddlepaddle.github.io/PaddleOCR/latest/en/update/update.html)
+
+
+## 🚀 بداية سريعة
+
+### الخطوة 1: جرّب عبر الإنترنت
+يوفر الموقع الرسمي لـ PaddleOCR **مركز تجربة** تفاعلي و**واجهات برمجة التطبيقات (APIs)** — لا حاجة لأي إعداد، فقط انقر لتجربة الخدمة.
+
+👉 [زيارة الموقع الرسمي](https://www.paddleocr.com)
+
+### الخطوة 2: النشر المحلي
+للاستخدام المحلي، يُرجى الرجوع إلى الوثائق التالية بناءً على احتياجاتك:
+
+- **سلسلة PP-OCR**: انظر [وثائق PP-OCR](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/OCR.html)
+- **سلسلة PaddleOCR-VL**: انظر [وثائق PaddleOCR-VL](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PaddleOCR-VL.html)
+- **PP-StructureV3**: انظر [وثائق PP-StructureV3](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PP-StructureV3.html)
+- **المزيد من القدرات**: انظر [وثائق المزيد من القدرات](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/pipeline_overview.html)
+
+
+## 🧩 المزيد من الميزات
+
+- تحويل النماذج إلى صيغة ONNX: [الحصول على نماذج ONNX](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/others/obtaining_onnx_models.html).
+- تسريع الاستدلال باستخدام محركات مثل OpenVINO وONNX Runtime وTensorRT، أو إجراء الاستدلال باستخدام نماذج بصيغة ONNX: [الاستدلال عالي الأداء](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/local_inference/high_performance_inference.html).
+- تسريع الاستدلال باستخدام وحدات GPU متعددة وعمليات متعددة: [الاستدلال المتوازي لخطوط الأنابيب](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/pipeline_usage/instructions/parallel_inference.html).
+- دمج PaddleOCR في تطبيقات مكتوبة بلغات C++ وC# وJava وغيرها: [الخدمة](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/serving/serving.html).
+
+## 🔄 نظرة سريعة على نتائج التنفيذ
+
+### PP-OCRv5
+
+<div align="center">
+  <p>
+       <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-OCRv5_demo.gif" alt="عرض توضيحي لـ PP-OCRv5">
+  </p>
+</div>
+
+
+
+### PP-StructureV3
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-StructureV3_demo.gif" alt="عرض توضيحي لـ PP-StructureV3">
+  </p>
+</div>
+
+### PaddleOCR-VL
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PaddleOCR-VL_demo.gif" alt="عرض توضيحي لـ PaddleOCR-VL">
+  </p>
+</div>
+
+
+## ✨ تابعنا
+
+⭐ **قم بتمييز هذا المستودع بنجمة لمتابعة التحديثات والإصدارات الجديدة المثيرة، بما في ذلك إمكانيات التعرف الضوئي على الحروف وتحليل المستندات القوية!** ⭐
+
+<div align="center">
+  <p>
+       <img width="1200" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/star_paddleocr2.en.gif" alt="تمييز المشروع بنجمة">
+  </p>
+</div>
+
+
+## 👩‍👩‍👧‍👦 المجتمع
+
+<div align="center">
+
+| حساب PaddlePaddle الرسمي على WeChat | انضم إلى مجموعة النقاش التقني |
+| :---: | :---: |
+| <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qrcode_for_paddlepaddle_official_account.jpg" width="150"> | <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qr_code_for_the_questionnaire.jpg" width="150"> |
+</div>
+
+
+## 😃 مشاريع رائعة تستفيد من PaddleOCR
+لم يكن PaddleOCR ليصل إلى ما هو عليه اليوم لولا مجتمعه المذهل! 💗 شكر جزيل لجميع شركائنا القدامى والمتعاونين الجدد وكل من بذل شغفه في PaddleOCR — سواء ذكرنا اسمه أم لا. دعمكم هو وقود حماسنا!
+
+<div align="center">
+
+| اسم المشروع | الوصف |
+| ------------ | ----------- |
+| [Dify](https://github.com/langgenius/dify) <a href="https://github.com/langgenius/dify"><img src="https://img.shields.io/github/stars/langgenius/dify"></a>|منصة جاهزة للإنتاج لتطوير سير العمل الوكيلي.|
+| [RAGFlow](https://github.com/infiniflow/ragflow) <a href="https://github.com/infiniflow/ragflow"><img src="https://img.shields.io/github/stars/infiniflow/ragflow"></a>|محرك RAG قائم على الفهم العميق للمستندات.|
+| [pathway](https://github.com/pathwaycom/pathway) <a href="https://github.com/pathwaycom/pathway"><img src="https://img.shields.io/github/stars/pathwaycom/pathway"></a>|إطار عمل Python ETL لمعالجة التدفقات والتحليلات الآنية وخطوط أنابيب النماذج اللغوية الكبيرة وRAG.|
+| [MinerU](https://github.com/opendatalab/MinerU) <a href="https://github.com/opendatalab/MinerU"><img src="https://img.shields.io/github/stars/opendatalab/MinerU"></a>|أداة تحويل المستندات متعددة الأنواع إلى Markdown.|
+| [Umi-OCR](https://github.com/hiroi-sora/Umi-OCR) <a href="https://github.com/hiroi-sora/Umi-OCR"><img src="https://img.shields.io/github/stars/hiroi-sora/Umi-OCR"></a>|برنامج OCR مجاني، مفتوح المصدر، للمعالجة الدفعية دون اتصال بالإنترنت.|
+| [cherry-studio](https://github.com/CherryHQ/cherry-studio) <a href="https://github.com/CherryHQ/cherry-studio"><img src="https://img.shields.io/github/stars/CherryHQ/cherry-studio"></a>|تطبيق سطح مكتب يدعم مزودي نماذج لغوية كبيرة متعددين.|
+| [haystack](https://github.com/deepset-ai/haystack)<a href="https://github.com/deepset-ai/haystack"><img src="https://img.shields.io/github/stars/deepset-ai/haystack"></a> |إطار عمل لتنظيم الذكاء الاصطناعي لبناء تطبيقات نماذج لغوية كبيرة قابلة للتخصيص وجاهزة للإنتاج.|
+| [OmniParser](https://github.com/microsoft/OmniParser)<a href="https://github.com/microsoft/OmniParser"><img src="https://img.shields.io/github/stars/microsoft/OmniParser"></a> |OmniParser: أداة تحليل الشاشة لوكيل واجهة المستخدم الرسومية القائم على الرؤية البحتة.|
+| [QAnything](https://github.com/netease-youdao/QAnything)<a href="https://github.com/netease-youdao/QAnything"><img src="https://img.shields.io/github/stars/netease-youdao/QAnything"></a> |الأسئلة والأجوبة المبنية على أي شيء.|
+| [تعرّف على المزيد من المشاريع](./awesome_projects.md) | [المزيد من المشاريع المبنية على PaddleOCR](./awesome_projects.md)|
+</div>
+
+## 👩‍👩‍👧‍👦 المساهمون
+
+<div align="center">
+<a href="https://github.com/PaddlePaddle/PaddleOCR/graphs/contributors">
+  <img src="https://contrib.rocks/image?repo=PaddlePaddle/PaddleOCR&max=400&columns=20"  width="800"/>
+</a>
+</div>
+
+## 🌟 النجوم
+
+<div align="center">
+  <p>
+      <img width="800" src="https://api.star-history.com/svg?repos=PaddlePaddle/PaddleOCR&type=Date" alt="تاريخ النجوم">
+  </p>
+</div>
+
+
+## 📄 الرخصة
+هذا المشروع مُصدر بموجب [رخصة Apache 2.0](LICENSE).
+
+## 🎓 الاستشهاد
+
+```bibtex
+@misc{cui2025paddleocr30technicalreport,
+      title={PaddleOCR 3.0 Technical Report},
+      author={Cheng Cui and Ting Sun and Manhui Lin and Tingquan Gao and Yubo Zhang and Jiaxuan Liu and Xueqing Wang and Zelun Zhang and Changda Zhou and Hongen Liu and Yue Zhang and Wenyu Lv and Kui Huang and Yichao Zhang and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2507.05595},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2507.05595},
+}
+
+@misc{cui2025paddleocrvlboostingmultilingualdocument,
+      title={PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model},
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Handong Zheng and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2510.14528},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2510.14528},
+}
+
+@misc{cui2026paddleocrvl15multitask09bvlm,
+      title={PaddleOCR-VL-1.5: Towards a Multi-Task 0.9B VLM for Robust In-the-Wild Document Parsing},
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2026},
+      eprint={2601.21957},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2601.21957},
+}
+```
diff --git a/readme/README_cn.md b/readme/README_cn.md
new file mode 100644
index 0000000..b2f6e35
--- /dev/null
+++ b/readme/README_cn.md
@@ -0,0 +1,306 @@
+<div align="center">
+  <p>
+      <img width="800" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Banner.png" alt="Star-history">
+  </p>
+
+
+
+<h3>全球领先的 OCR 工具包与文档 AI 引擎</h3>
+
+[English](../README.md) | 简体中文| [繁體中文](./README_tcn.md) | [日本語](./README_ja.md) | [한국어](./README_ko.md) | [Français](./README_fr.md) | [Русский](./README_ru.md) | [Español](./README_es.md) | [العربية](./README_ar.md)
+
+<!-- icon -->
+
+[![PyPI Downloads](https://static.pepy.tech/badge/paddleocr)](https://pepy.tech/projects/paddleocr)
+[![Used by](https://img.shields.io/badge/Used%20by-6k%2B%20repositories-blue)](https://github.com/PaddlePaddle/PaddleOCR/network/dependents)
+![python](https://img.shields.io/badge/python-3.8~3.12-aff.svg)
+![os](https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-pink.svg)
+![hardware](https://img.shields.io/badge/hardware-cpu%2C%20gpu%2C%20xpu%2C%20npu-yellow.svg)
+
+[![AI Studio](https://img.shields.io/badge/PaddleOCR-_Offiical_Website-1927BA?logo=data:image/png;base64,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&labelColor=white)](https://www.paddleocr.com)
+[![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/PaddlePaddle/PaddleOCR)
+[![License](https://img.shields.io/badge/license-Apache_2.0-green)](../LICENSE)
+
+</div>
+
+
+
+
+
+
+**PaddleOCR 以业界领先的精准度，将 PDF 文档和图像转换为结构化、LLM 友好的数据格式（JSON/Markdown）。凭借 70,000+ Stars 的成绩，PaddleOCR 已获得 Dify、RAGFlow、Cherry Studio 等顶级项目的广泛信赖，是构建智能 RAG 和 Agentic 应用的核心基础组件。**
+
+
+## 🚀 核心特性
+
+### 📄 智能文档解析（面向大模型）
+> *为大模型时代将杂乱的文档视觉信息转化为结构化数据。*
+
+* **SOTA 级文档视觉语言模型 (VLM)**: 业界领先的轻量级文档解析视觉语言模型 **PaddleOCR-VL-1.6 (0.9B)**。该模型以 96.3% 精度刷新 OmniDocBench v1.6，文本、公式、表格识别全面领先，并在古籍、生僻字、印章、图表等多场景能力显著增强，支持以 **Markdown** 和 **JSON** 格式输出结构化结果。
+* **版面结构分析**：由 **PP-StructureV3** 驱动，无缝将复杂的 PDF文档 和图像转换为 **Markdown** 或 **JSON** 格式。与 PaddleOCR-VL 系列模型不同，它提供更细粒度的坐标信息，包括表格单元格坐标、文本坐标等。
+* **生产级高效能**：以极小的模型体积实现商业级别的准确率。在公开基准测试中超越众多闭源解决方案，同时保持极高的资源利用率，完美适配边缘计算与云端部署。
+
+### 🔍 通用文本识别（场景 OCR）
+> *快速、精准的多语言文本检测与识别，被全球开发者广泛采用。*
+
+* **支持 100+ 种语言**：原生支持庞大丰富的全球语种库。**PP-OCRv6** 单模型统一支持 50 种语言（中、英、日及 46 种拉丁语系），无需切换模型即可应对多语言混合排版文档。
+* **复杂场景支持**：除了标准的文本识别，还支持在各种广泛的环境下进行**自然场景文本检测与识别**，涵盖身份证件、街景、书籍以及工业零部件等。
+* **性能提升**：PP-OCRv6 相比 PP-OCRv5 检测精度提升 **4.6%**、识别精度提升 **5.1%**，超越主流视觉语言大模型，CPU 推理加速 5.2×。
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Arch_cn.jpg" alt="PaddleOCR Architecture">
+  </p>
+</div>
+
+### 🛠️ 以开发者为中心的生态系统
+* **无缝集成**：AI 智能体生态系统的首选——与 **Dify、RAGFlow、Pathway和Cherry Studio** 深度集成。
+* **大语言模型数据飞轮**：完整的数据流水线,用于构建高质量数据集,为微调大语言模型提供可持续的"数据引擎"。
+* **一键部署**：支持多种硬件后端(NVIDIA GPU、Intel CPU、昆仑芯 XPU 和多种 AI 加速器)。
+
+
+## 📣 最新动态
+
+### 🔥 2026.06.11: PaddleOCR 3.7.0 发布
+- PP-OCRv6 核心亮点如下：
+
+    - **精度全面提升**：medium 档相比 PP-OCRv5_server 检测精度提升 4.6%、识别精度提升 5.1%，以仅 34.5M 参数超越 Qwen3-VL-235B、GPT-5.5 等主流视觉语言大模型。
+    - **50 种语言统一支持**：单一模型覆盖中文、英文、日文及 46 种拉丁语系语言，无需为不同语种切换模型。
+    - **专业场景增强**：数码显示屏、点阵字符、轮胎印字、工业字符等传统 VLM 难以覆盖的场景识别能力大幅提升。
+    - **推理速度更快**：medium 档 CPU OpenVINO 推理加速 5.2×，tiny 档 Apple M4 加速 6.1×，A100 上仅需 0.13s。
+    - **三档模型覆盖全场景**：tiny（1.5M）/ small（7.7M）/ medium（34.5M）分别面向端侧/移动端/服务端部署。
+    - **模型获取**：所有模型均可通过 [HuggingFace](https://huggingface.co/collections/PaddlePaddle/pp-ocrv6) 和 [ModelScope](https://www.modelscope.cn/collections/PaddlePaddle/PP-OCRv6) 获取。
+
+<details>
+<summary><strong>2026.05.28: PaddleOCR 3.6.0 发布</strong></summary>
+
+- PaddleOCR-VL-1.6核心亮点如下：
+
+    - **全新SOTA精度**：OmniDocBench v1.6 突破96.3%，在OmniDocBench v1.5、Real5-OmniDocBench上同样刷新SOTA，文本、公式、表格识别全面领先开源与闭源方案。
+    - **能力全面升级**：表格、古籍、生僻字识别大幅提升，印章、spotting、图表识别等多场景均有显著增强。
+    - **无缝迁移**：模型结构与PaddleOCR-VL-1.5完全一致，零成本适配，即换即用。
+    - **立即试用**：可在 [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.6) 或 [PaddleOCR 官方网站](https://www.paddleocr.com) 使用。
+
+</details>
+
+<details>
+<summary><strong>2026.04.21: PaddleOCR 3.5.0 发布</strong></summary>
+
+* **推理后端灵活切换**：支持在飞桨静态图、飞桨动态图和 Transformers 之间无缝切换。深度适配 Hugging Face 生态，20 个主要模型支持以 Transformers 作为推理后端。
+* **常见文档格式转 Markdown**：支持将 Word、Excel、Powerpoint 等常见文档格式转换为 Markdown。
+* **解析结果导出 DOCX**：`PaddleOCR-VL` 系列、`PP-StructureV3` 和 `PP-DocTranslation` 现已支持将解析结果导出为 DOCX，便于在 Microsoft Word 中查看和编辑。
+* **官方浏览器推理 SDK**：发布官方浏览器推理 SDK `PaddleOCR.js`，支持在浏览器中运行 `PP-OCRv5`。
+
+</details>
+
+<details>
+<summary><strong>2026.01.29: PaddleOCR 3.4.0 发布</strong></summary>
+* PaddleOCR-VL-1.5 (SOTA 0.9B VLM)：
+    * **OmniDocBench 94.5%准确率**：超越顶级通用大模型和专业文档解析模型。
+    * **现实5大场景文档解析的SOTA性能**：首次引入**PP-DocLayoutV3**算法进行异形框定位，可以解决5种真实场景: 倾斜、弯曲、扫描、光线变化和屏幕拍照。
+    * **能力拓展**：增加**印章识别**、**文本行检测/识别**，并扩展至**111种语言**(包括中国的藏文和孟加拉语)。
+    * **长文档跨页解析**：支持自动跨页表格合并和分层标题识别。
+    * **立即试用**：可在[HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.5)或[PaddleOCR 官方网站](https://www.paddleocr.com)使用。
+
+</details>
+
+<details>
+<summary><strong>2025.10.16: PaddleOCR 3.3.0 发布</strong></summary>
+
+- **发布PaddleOCR-VL**：
+    - **模型介绍**：
+        - **PaddleOCR-VL** 是一款先进、高效的文档解析模型，专为文档中的元素识别设计。其核心组件为 PaddleOCR-VL-0.9B，这是一种紧凑而强大的视觉语言模型（VLM），它由 NaViT 风格的动态分辨率视觉编码器与 ERNIE-4.5-0.3B 语言模型组成，能够实现精准的元素识别。**该模型支持 109 种语言，并在识别复杂元素（如文本、表格、公式和图表）方面表现出色，同时保持极低的资源消耗。通过在广泛使用的公开基准与内部基准上的全面评测，PaddleOCR-VL 在页级级文档解析与元素级识别均达到 SOTA 表现**。它显著优于现有的基于Pipeline方案和文档解析多模态方案以及先进的通用多模态大模型，并具备更快的推理速度。这些优势使其非常适合在真实场景中落地部署。模型已发布至[HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL)，欢迎大家下载使用！更多介绍内容请点击[PaddleOCR-VL](https://www.paddleocr.ai/latest/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL.html)。
+
+    - **特性**：
+        - **紧凑而强大的视觉语言模型架构**：我们提出了一种新的视觉语言模型，专为资源高效的推理而设计，在元素识别方面表现出色。通过将NaViT风格的动态高分辨率视觉编码器与轻量级的ERNIE-4.5-0.3B语言模型结合，我们显著增强了模型的识别能力和解码效率。这种集成在保持高准确率的同时降低了计算需求，使其非常适合高效且实用的文档处理应用。
+        - **文档解析的SOTA性能**：PaddleOCR-VL在页面级文档解析和元素级识别中达到了最先进的性能。它显著优于现有的基于流水线的解决方案，并在文档解析中展现出与领先的视觉语言模型（VLMs）竞争的强劲实力。此外，它在识别复杂的文档元素（如文本、表格、公式和图表）方面表现出色，使其适用于包括手写文本和历史文献在内的各种具有挑战性的内容类型。这使得它具有高度的多功能性，适用于广泛的文档类型和场景。
+        - **多语言支持**：PaddleOCR-VL支持109种语言，覆盖了主要的全球语言，包括但不限于中文、英文、日文、拉丁文和韩文，以及使用不同文字和结构的语言，如俄语（西里尔字母）、阿拉伯语、印地语（天城文）和泰语。这种广泛的语言覆盖大大增强了我们系统在多语言和全球化文档处理场景中的适用性。
+
+- **发布PP-OCRv5小语种识别模型**：
+    - 优化拉丁文识别的准度和广度，新增西里尔文、阿拉伯文、天城文、泰卢固语、泰米尔语等语系，覆盖109种语言文字的识别。模型参数量仅为2M，部分模型精度较上一代提升40%以上。
+
+</details>
+
+
+<details>
+<summary><strong>2025.08.21: PaddleOCR 3.2.0 发布</strong></summary>
+
+- **重要模型新增：**
+    - 新增 PP-OCRv5 英文、泰文、希腊文识别模型的训练、推理、部署。**其中 PP-OCRv5 英文模型较 PP-OCRv5 主模型在英文场景提升 11%，泰文识别模型精度 82.68%，希腊文识别模型精度 89.28%。**
+
+- **部署能力升级：**
+    - **全面支持飞桨框架 3.1.0 和 3.1.1 版本。**
+    - **全面升级 PP-OCRv5 C++ 本地部署方案，支持 Linux、Windows，功能及精度效果与 Python 方案保持一致。**
+    - **高性能推理支持 CUDA 12，可使用 Paddle Inference、ONNX Runtime 后端推理。**
+    - **高稳定性服务化部署方案全面开源，支持用户根据需求对 Docker 镜像和 SDK 进行定制化修改。**
+    - 高稳定性服务化部署方案支持通过手动构造HTTP请求的方式调用，该方式允许客户端代码使用任意编程语言编写。
+
+- **Benchmark支持**：
+    - **全部产线支持产线细粒度 benchmark，能够测量产线端到端推理时间以及逐层、逐模块的耗时数据，可用于辅助产线性能分析。可以参考[文档](../docs/version3.x/pipeline_usage/instructions/benchmark.md)来进行性能测试。**
+    - **文档中补充各产线常用配置在主流硬件上的关键指标，包括推理耗时和内存占用等，为用户部署提供参考。**
+
+- **Bug修复：**
+    - 修复模型训练时训练日志保存失败的问题。
+    - 对公式模型的数据增强部分进行了版本兼容性升级，以适应新版本的 albumentations 依赖，并修复了在多进程使用 tokenizers 依赖包时出现的死锁警告。
+    - 修复 PP-StructureV3 配置文件中的 `use_chart_parsing` 等开关行为与其他产线不统一的问题。
+
+- **其他升级：**
+    - **分离必要依赖与可选依赖。使用基础文字识别功能时，仅需安装少量核心依赖；若需文档解析、信息抽取等功能，用户可按需选择安装额外依赖。**
+    - **支持 Windows 用户使用英伟达 50 系显卡，可根据 [安装文档](../docs/version3.x/installation.md) 安装对应版本的 paddle 框架。**
+    - **PP-OCR 系列模型支持返回单文字坐标。**
+    - 模型新增 AIStudio、ModelScope 等下载源。可指定相关下载源下载对应的模型。
+    - 支持图表转表 PP-Chart2Table 单功能模块推理能力。
+    - 优化部分使用文档中的描述，提升易用性。
+</details>
+
+
+[历史日志](https://paddlepaddle.github.io/PaddleOCR/latest/en/update/update.html)
+
+
+## 🚀 快速开始
+
+### 步骤 1: 在线体验
+PaddleOCR官方网站提供交互式**体验中心**和**APIs**——无需设置，一键体验。
+
+👉 [访问官方网站](https://www.paddleocr.com)
+
+### 步骤 2: 本地部署
+
+对于本地使用,请根据您的需求参考以下文档:
+
+- **PP-OCR系列**：查看[PP-OCR文档](https://www.paddleocr.ai/latest/version3.x/pipeline_usage/OCR.html)
+- **PaddleOCR-VL系列**：查看[PaddleOCR-VL文档](https://www.paddleocr.ai/latest/version3.x/pipeline_usage/PaddleOCR-VL.html)
+- **PP-StructureV3**：查看[PP-StructureV3文档](https://www.paddleocr.ai/latest/version3.x/pipeline_usage/PP-StructureV3.html)
+- **更多能力**：查看[更多能力文档](https://www.paddleocr.ai/latest/version3.x/pipeline_usage/pipeline_overview.html)
+
+
+## 🧩 更多功能
+
+- 将模型转换为ONNX格式: [获取ONNX模型](https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/inference_deployment/others/obtaining_onnx_models.html)。
+- 使用OpenVINO、ONNX Runtime、TensorRT等引擎加速推理,或使用ONNX格式模型进行推理: [高性能推理](https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/inference_deployment/local_inference/high_performance_inference.html)。
+- 使用多GPU和多进程加速推理: [流水线并行推理](https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/pipeline_usage/instructions/parallel_inference.html)。
+- 将PaddleOCR集成到C++、C#、Java等语言编写的应用程序中: [服务化部署](https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/inference_deployment/serving/serving.html)。
+
+## 🔄 执行结果快速预览
+
+### PP-OCRv5
+
+<div align="center">
+  <p>
+       <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-OCRv5_demo.gif" alt="PP-OCRv5 Demo">
+  </p>
+</div>
+
+
+
+### PP-StructureV3
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-StructureV3_demo.gif" alt="PP-StructureV3 Demo">
+  </p>
+</div>
+
+### PaddleOCR-VL
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PaddleOCR-VL_demo.gif" alt="PP-StructureV3 Demo">
+  </p>
+</div>
+
+
+## ✨ 保持关注
+
+⭐ **收藏本仓库，持续关注最新动态与版本发布，包括强大的 OCR 及文档解析等新功能特性。** ⭐
+
+<div align="center">
+  <p>
+       <img width="1200" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/star_paddleocr2.en.gif" alt="Star-Project">
+  </p>
+</div>
+
+
+## 👩‍👩‍👧‍👦 社区
+
+<div align="center">
+
+| PaddlePaddle 微信公众号 |  加入技术讨论群 |
+| :---: | :---: |
+| <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qrcode_for_paddlepaddle_official_account.jpg" width="150"> | <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qr_code_for_the_questionnaire.jpg" width="150"> |
+</div>
+
+
+## 😃 使用 PaddleOCR 的优秀项目
+
+<div align="center">
+
+PaddleOCR 的发展离不开社区贡献！💗衷心感谢所有开发者、合作伙伴与贡献者！
+| 项目名称 | 简介 |
+| ------------ | ----------- |
+| [RAGFlow](https://github.com/infiniflow/ragflow) <a href="https://github.com/infiniflow/ragflow"><img src="https://img.shields.io/github/stars/infiniflow/ragflow"></a>|基于RAG的AI工作流引擎|
+| [pathway](https://github.com/pathwaycom/pathway) <a href="https://github.com/pathwaycom/pathway"><img src="https://img.shields.io/github/stars/pathwaycom/pathway"></a>|用于流处理、实时分析、LLM流水线和RAG的Python ETL框架|
+| [MinerU](https://github.com/opendatalab/MinerU) <a href="https://github.com/opendatalab/MinerU"><img src="https://img.shields.io/github/stars/opendatalab/MinerU"></a>|多类型文档转换Markdown工具|
+| [Umi-OCR](https://github.com/hiroi-sora/Umi-OCR) <a href="https://github.com/hiroi-sora/Umi-OCR"><img src="https://img.shields.io/github/stars/hiroi-sora/Umi-OCR"></a>|开源批量离线OCR软件|
+| [cherry-studio](https://github.com/CherryHQ/cherry-studio) <a href="https://github.com/CherryHQ/cherry-studio"><img src="https://img.shields.io/github/stars/CherryHQ/cherry-studio"></a>|一个支持多个LLM提供商的桌面客户端|
+| [OmniParser](https://github.com/microsoft/OmniParser)<a href="https://github.com/microsoft/OmniParser"><img src="https://img.shields.io/github/stars/microsoft/OmniParser"></a> |基于纯视觉的GUI智能体屏幕解析工具|
+| [QAnything](https://github.com/netease-youdao/QAnything)<a href="https://github.com/netease-youdao/QAnything"><img src="https://img.shields.io/github/stars/netease-youdao/QAnything"></a> |基于任意内容的问答系统|
+| [PDF-Extract-Kit](https://github.com/opendatalab/PDF-Extract-Kit) <a href="https://github.com/opendatalab/PDF-Extract-Kit"><img src="https://img.shields.io/github/stars/opendatalab/PDF-Extract-Kit"></a>|高效复杂PDF文档提取工具包|
+| [Dango-Translator](https://github.com/PantsuDango/Dango-Translator)<a href="https://github.com/PantsuDango/Dango-Translator"><img src="https://img.shields.io/github/stars/PantsuDango/Dango-Translator"></a> |屏幕实时翻译工具|
+| [更多项目](../awesome_projects.md) |  [更多基于PaddleOCR的项目](../awesome_projects.md) |
+</div>
+
+
+## 👩‍👩‍👧‍👦 贡献者
+
+<div align="center">
+<a href="https://github.com/PaddlePaddle/PaddleOCR/graphs/contributors">
+  <img src="https://contrib.rocks/image?repo=PaddlePaddle/PaddleOCR&max=400&columns=20"  width="800"/>
+</a>
+</div>
+
+## 🌟 Star历史
+
+<div align="center">
+  <p>
+      <img width="800" src="https://api.star-history.com/svg?repos=PaddlePaddle/PaddleOCR&type=Date" alt="Star-history">
+  </p>
+</div>
+
+
+## 📄 许可证
+本项目采用[Apache 2.0许可证](LICENSE)发布。
+
+## 🎓 引用
+
+```bibtex
+@misc{cui2025paddleocr30technicalreport,
+      title={PaddleOCR 3.0 Technical Report},
+      author={Cheng Cui and Ting Sun and Manhui Lin and Tingquan Gao and Yubo Zhang and Jiaxuan Liu and Xueqing Wang and Zelun Zhang and Changda Zhou and Hongen Liu and Yue Zhang and Wenyu Lv and Kui Huang and Yichao Zhang and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2507.05595},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2507.05595},
+}
+
+@misc{cui2025paddleocrvlboostingmultilingualdocument,
+      title={PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model},
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Handong Zheng and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2510.14528},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2510.14528},
+}
+
+@misc{cui2026paddleocrvl15multitask09bvlm,
+      title={PaddleOCR-VL-1.5: Towards a Multi-Task 0.9B VLM for Robust In-the-Wild Document Parsing},
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2026},
+      eprint={2601.21957},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2601.21957},
+}
+```
diff --git a/readme/README_es.md b/readme/README_es.md
new file mode 100644
index 0000000..4f21fbc
--- /dev/null
+++ b/readme/README_es.md
@@ -0,0 +1,302 @@
+<div align="center">
+  <p>
+      <img width="800" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Banner.png" alt="Star-history">
+  </p>
+
+
+
+<h3>Kit de OCR líder mundial y motor de IA para documentos</h3>
+
+[English](../README.md) | [简体中文](./README_cn.md) | [繁體中文](./README_tcn.md) | [日本語](./README_ja.md) | [한국어](./README_ko.md) | [Français](./README_fr.md) | [Русский](./README_ru.md) | Español | [العربية](./README_ar.md)
+
+<!-- icon -->
+
+[![PyPI Downloads](https://static.pepy.tech/badge/paddleocr)](https://pepy.tech/projects/paddleocr)
+[![Used by](https://img.shields.io/badge/Used%20by-6k%2B%20repositories-blue)](https://github.com/PaddlePaddle/PaddleOCR/network/dependents)
+![python](https://img.shields.io/badge/python-3.8~3.12-aff.svg)
+![os](https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-pink.svg)
+![hardware](https://img.shields.io/badge/hardware-cpu%2C%20gpu%2C%20xpu%2C%20npu-yellow.svg)
+
+[![AI Studio](https://img.shields.io/badge/PaddleOCR-_Offiical_Website-1927BA?logo=data:image/png;base64,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&labelColor=white)](https://www.paddleocr.com)
+[![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/PaddlePaddle/PaddleOCR)
+[![License](https://img.shields.io/badge/license-Apache_2.0-green)](../LICENSE)
+
+</div>
+
+
+
+
+
+
+**PaddleOCR convierte documentos e imágenes en datos estructurados y listos para LLM (JSON/Markdown) con una precisión líder en la industria. Con más de 70 000 estrellas y la confianza de proyectos de primer nivel como Dify, RAGFlow y Cherry Studio, PaddleOCR es la base para construir aplicaciones inteligentes de RAG y Agentic.**
+
+
+## 🚀 Características principales
+
+### 📄 Análisis inteligente de documentos (listo para LLM)
+> *Transformando contenido visual complejo en datos estructurados para la era de los LLM.*
+
+* **SOTA Document VLM**: Featuring **PaddleOCR-VL-1.6 (0.9B)**, the industry's leading lightweight vision-language model for document parsing. It achieves 96.3% accuracy on OmniDocBench v1.6, leads in text, formula, and table recognition, and shows significantly enhanced capabilities in ancient documents, rare characters, seals, and charts, with structured outputs in **Markdown** and **JSON** formats.
+* **Conversión con reconocimiento de estructura**: Impulsado por **PP-StructureV3**, convierte sin problemas PDFs e imágenes complejas en **Markdown** o **JSON**. A diferencia de los modelos de la serie PaddleOCR-VL, proporciona información de coordenadas más detallada, incluyendo coordenadas de celdas de tablas, coordenadas de texto y más.
+* **Eficiencia lista para producción**: Logra precisión de nivel comercial con una huella ultrapequeña. Supera a numerosas soluciones de código cerrado en benchmarks públicos, manteniéndose eficiente en recursos para despliegue en el borde o en la nube.
+
+### 🔍 Reconocimiento universal de texto (Scene OCR)
+> *El estándar de oro mundial para la detección de texto multilingüe de alta velocidad.*
+
+* **Compatibilidad con más de 100 idiomas**: Reconocimiento nativo de una amplia biblioteca global. **PP-OCRv6** soporta 50 idiomas con un único modelo unificado (chino, inglés, japonés y 46 idiomas latinos) — sin necesidad de cambiar modelos.
+* **Dominio de elementos complejos**: Más allá del reconocimiento de texto estándar, admitimos la **detección de texto en escenas naturales** en una amplia gama de entornos, incluyendo documentos de identidad, vistas de calles, libros y componentes industriales.
+* **Salto en rendimiento**: PP-OCRv6 logra **+4.6% en detección** y **+5.1% en reconocimiento** sobre PP-OCRv5, superando a los principales modelos de lenguaje visual. Aceleración 5.2× en inferencia CPU end-to-end.
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Arch.jpg" alt="PaddleOCR Architecture">
+  </p>
+</div>
+
+### 🛠️ Ecosistema centrado en el desarrollador
+* **Integración perfecta**: La opción preferida para el ecosistema de agentes de IA, con integración profunda en **Dify, RAGFlow, Pathway y Cherry Studio**.
+* **Motor de datos para LLM**: Un pipeline completo para construir conjuntos de datos de alta calidad, proporcionando un "Motor de Datos" sostenible para el ajuste fino de modelos de lenguaje grandes.
+* **Despliegue en un clic**: Compatible con diversos backends de hardware (GPU NVIDIA, CPU Intel, XPU Kunlunxin y diversos aceleradores de IA).
+
+
+## 📣 Actualizaciones recientes
+
+### 🔥 2026.06.11: Lanzamiento de PaddleOCR 3.7.0
+- Aspectos destacados de PP-OCRv6:
+
+    - **Mejora de precisión**: El nivel medium logra +4.6% en detección y +5.1% en reconocimiento sobre PP-OCRv5_server, superando a los principales VLMs (Qwen3-VL-235B, GPT-5.5) con solo 34.5M parámetros.
+    - **50 idiomas unificados**: Un solo modelo cubre chino, inglés, japonés y 46 idiomas latinos — sin necesidad de cambiar modelos.
+    - **Escenarios especializados**: Mejoras significativas en pantallas digitales, caracteres de matriz de puntos, impresiones de neumáticos y texto industrial.
+    - **Inferencia más rápida**: Aceleración 5.2× en CPU (OpenVINO), 6.1× en Apple M4 (tiny), 0.13s en A100 GPU.
+    - **Tres niveles para todos los escenarios**: tiny (1.5M) / small (7.7M) / medium (34.5M) para despliegue en edge, móvil y servidor.
+    - **Disponibilidad de modelos**: Todos los modelos están disponibles en [HuggingFace](https://huggingface.co/collections/PaddlePaddle/pp-ocrv6) y [ModelScope](https://www.modelscope.cn/collections/PaddlePaddle/PP-OCRv6).
+
+<details>
+<summary><strong>2026.05.28: Lanzamiento de PaddleOCR 3.6.0</strong></summary>
+
+- Aspectos destacados de PaddleOCR-VL-1.6:
+
+    - **Nueva precisión SOTA**: 96.3% en OmniDocBench v1.6, nuevo SOTA en OmniDocBench v1.5 y Real5-OmniDocBench.
+    - **Mejora integral**: mejoras significativas en tablas, documentos antiguos y caracteres raros.
+    - **Migración sin fisuras**: arquitectura totalmente compatible con PaddleOCR-VL-1.5.
+    - **Pruébalo**: [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.6) o [sitio oficial](https://www.paddleocr.com).
+
+</details>
+<details>
+<summary><strong>2026.04.21: Lanzamiento de PaddleOCR 3.5.0</strong></summary>
+
+* **Backends de inferencia más flexibles**: cambia sin problemas entre grafo estático de Paddle, grafo dinámico de Paddle y Transformers. PaddleOCR está ahora profundamente integrado con el ecosistema de Hugging Face, y 20 modelos principales admiten Transformers como backend de inferencia.
+* **Documentos de Office a Markdown**: convierte formatos de documentos comunes como Word, Excel y PowerPoint a Markdown.
+* **Exportación de resultados a DOCX**: las series `PaddleOCR-VL`, `PP-StructureV3` y `PP-DocTranslation` ahora admiten exportar los resultados de análisis a DOCX para verlos y editarlos cómodamente en Microsoft Word.
+* **SDK oficial de inferencia en navegador**: se lanzó `PaddleOCR.js`, el SDK oficial de inferencia en navegador, que permite ejecutar `PP-OCRv5` directamente en el navegador.
+
+</details>
+
+<details>
+<summary><strong>2026.01.29: Lanzamiento de PaddleOCR 3.4.0</strong></summary>
+* PaddleOCR-VL-1.5 (SOTA 0,9B VLM): ¡Nuestro último modelo insignia para el análisis de documentos ya está disponible!
+    * **94,5% de precisión en OmniDocBench**: Superando a los mejores modelos generales de gran escala y a los analizadores de documentos especializados.
+    * **Robustez en el mundo real**: El primero en introducir el algoritmo **PP-DocLayoutV3** para el posicionamiento de formas irregulares, dominando 5 escenarios difíciles: *inclinación, deformación, escaneo, iluminación y fotografía de pantalla*.
+    * **Expansión de capacidades**: Ahora admite **reconocimiento de sellos**, **detección de texto** y se expande a **111 idiomas** (incluyendo el tibetano de China y el bengalí).
+    * **Dominio de documentos largos**: Admite la fusión automática de tablas entre páginas e identificación jerárquica de encabezados.
+    * **Pruébalo ahora**: Disponible en [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.5) o en nuestro [Sitio web oficial](https://www.paddleocr.com).
+
+</details>
+
+<details>
+<summary><strong>2025.10.16: Lanzamiento de PaddleOCR 3.3.0</strong></summary>
+
+- Lanzamiento de PaddleOCR-VL:
+    - **Introducción al modelo**:
+        - **PaddleOCR-VL** es un modelo SOTA eficiente en recursos diseñado específicamente para el análisis de documentos. Su componente principal es PaddleOCR-VL-0.9B, un modelo de visión y lenguaje (VLM) compacto pero potente que integra un codificador visual de resolución dinámica al estilo NaViT con el modelo de lenguaje ERNIE-4.5-0.3B para permitir un reconocimiento preciso de elementos. **Este innovador modelo admite eficientemente 109 idiomas y sobresale en el reconocimiento de elementos complejos (p. ej., texto, tablas, fórmulas y gráficos), mientras mantiene un consumo mínimo de recursos**. A través de evaluaciones exhaustivas en benchmarks públicos ampliamente utilizados y benchmarks internos, PaddleOCR-VL logra un rendimiento SOTA tanto en el análisis de documentos a nivel de página como en el reconocimiento a nivel de elemento. Supera significativamente a las soluciones existentes, exhibe una fuerte competitividad frente a los VLM de primer nivel y ofrece velocidades de inferencia rápidas. Estas fortalezas lo hacen altamente adecuado para el despliegue práctico en escenarios del mundo real. El modelo ha sido publicado en [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL). ¡Todos son bienvenidos a descargarlo y usarlo! Más información de introducción se puede encontrar en [PaddleOCR-VL](https://www.paddleocr.ai/latest/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL.html).
+
+    - **Características principales**:
+        - **Arquitectura VLM compacta pero potente**: Presentamos un novedoso modelo de visión y lenguaje diseñado específicamente para una inferencia eficiente en recursos, logrando un rendimiento sobresaliente en el reconocimiento de elementos. Al integrar un codificador visual dinámico de alta resolución al estilo NaViT con el modelo de lenguaje ligero ERNIE-4.5-0.3B, mejoramos significativamente las capacidades de reconocimiento y la eficiencia de decodificación del modelo. Esta integración mantiene una alta precisión mientras reduce las demandas computacionales, lo que lo hace adecuado para aplicaciones de procesamiento de documentos eficientes y prácticas.
+        - **Rendimiento SOTA en análisis de documentos**: PaddleOCR-VL logra un rendimiento de vanguardia tanto en el análisis de documentos a nivel de página como en el reconocimiento a nivel de elemento. Supera significativamente a las soluciones basadas en pipeline existentes y exhibe una fuerte competitividad frente a los modelos de visión y lenguaje (VLM) líderes en el análisis de documentos. Además, sobresale en el reconocimiento de elementos de documentos complejos, como texto, tablas, fórmulas y gráficos, lo que lo hace adecuado para una amplia gama de tipos de contenido desafiantes, incluyendo texto manuscrito y documentos históricos. Esto lo hace altamente versátil y adecuado para una amplia gama de tipos de documentos y escenarios.
+        - **Soporte multilingüe**: PaddleOCR-VL admite 109 idiomas, cubriendo los principales idiomas globales, incluyendo, entre otros, chino, inglés, japonés, latín y coreano, así como idiomas con diferentes escrituras y estructuras, como el ruso (escritura cirílica), el árabe, el hindi (escritura devanagari) y el tailandés. Esta amplia cobertura de idiomas mejora sustancialmente la aplicabilidad de nuestro sistema a escenarios de procesamiento de documentos multilingües y globalizados.
+
+- Lanzamiento del modelo de reconocimiento multilingüe PP-OCRv5:
+    - Se mejoró la precisión y cobertura del reconocimiento de escritura latina; se añadió compatibilidad con sistemas de escritura cirílico, árabe, devanagari, telugu, tamil y otros, cubriendo el reconocimiento de 109 idiomas. El modelo tiene solo 2M de parámetros, y la precisión de algunos modelos ha aumentado más de un 40% en comparación con la generación anterior.
+
+</details>
+
+
+<details>
+<summary><strong>2025.08.21: Lanzamiento de PaddleOCR 3.2.0</strong></summary>
+
+- **Adiciones significativas de modelos:**
+    - Se introdujeron entrenamiento, inferencia y despliegue para modelos de reconocimiento PP-OCRv5 en inglés, tailandés y griego. **El modelo PP-OCRv5 en inglés ofrece una mejora del 11% en escenarios en inglés en comparación con el modelo principal PP-OCRv5, con los modelos de reconocimiento en tailandés y griego alcanzando precisiones del 82,68% y 89,28%, respectivamente.**
+
+- **Mejoras en las capacidades de despliegue:**
+    - **Compatibilidad total con las versiones 3.1.0 y 3.1.1 del framework PaddlePaddle.**
+    - **Actualización integral de la solución de despliegue local en C++ de PP-OCRv5, que ahora admite tanto Linux como Windows, con paridad de características y precisión idéntica a la implementación en Python.**
+    - **La inferencia de alto rendimiento ahora admite CUDA 12, y la inferencia puede realizarse utilizando los backends Paddle Inference u ONNX Runtime.**
+    - **La solución de despliegue orientado a servicios de alta estabilidad ahora es completamente de código abierto, lo que permite a los usuarios personalizar imágenes Docker y SDKs según sea necesario.**
+    - La solución de despliegue orientado a servicios de alta estabilidad también admite la invocación mediante solicitudes HTTP construidas manualmente, lo que permite el desarrollo de código del lado del cliente en cualquier lenguaje de programación.
+
+- **Soporte de benchmarks:**
+    - **Todas las líneas de producción ahora admiten benchmarking detallado, permitiendo medir el tiempo de inferencia de extremo a extremo, así como datos de latencia por capa y por módulo para ayudar en el análisis de rendimiento. [Aquí](docs/version3.x/pipeline_usage/instructions/benchmark.en.md) se explica cómo configurar y usar la función de benchmark.**
+    - **La documentación se ha actualizado para incluir métricas clave para configuraciones de uso común en hardware convencional, como la latencia de inferencia y el uso de memoria, proporcionando referencias de despliegue para los usuarios.**
+
+- **Corrección de errores:**
+    - Se resolvió el problema del guardado fallido de registros durante el entrenamiento del modelo.
+    - Se actualizó el componente de aumento de datos para modelos de fórmulas para compatibilidad con versiones más recientes de la dependencia albumentations, y se corrigieron advertencias de bloqueo al usar el paquete tokenizers en escenarios multiproceso.
+    - Se corrigieron inconsistencias en los comportamientos de los interruptores (p. ej., `use_chart_parsing`) en los archivos de configuración de PP-StructureV3 en comparación con otros pipelines.
+
+- **Otras mejoras:**
+    - **Se separaron las dependencias principales y opcionales. Solo se requieren dependencias principales mínimas para el reconocimiento de texto básico; las dependencias adicionales para el análisis de documentos y la extracción de información se pueden instalar según sea necesario.**
+    - **Se habilitó la compatibilidad con tarjetas gráficas NVIDIA RTX de la serie 50 en Windows; los usuarios pueden consultar la [guía de instalación](docs/version3.x/installation.en.md) para conocer las versiones correspondientes del framework PaddlePaddle.**
+    - **Los modelos de la serie PP-OCR ahora admiten la devolución de coordenadas de caracteres individuales.**
+    - Se añadieron fuentes de descarga de modelos de AIStudio, ModelScope y otras, permitiendo a los usuarios especificar la fuente para las descargas de modelos.
+    - Se añadió compatibilidad con la conversión de gráficos a tablas a través del módulo PP-Chart2Table.
+    - Se optimizaron las descripciones de la documentación para mejorar la usabilidad.
+</details>
+
+
+[Historial de cambios](https://paddlepaddle.github.io/PaddleOCR/latest/en/update/update.html)
+
+
+## 🚀 Inicio rápido
+
+### Paso 1: Pruébalo en línea
+El sitio web oficial de PaddleOCR ofrece un **Centro de experiencia** interactivo y **APIs** — sin necesidad de configuración, solo un clic para experimentar.
+
+👉 [Visitar el sitio web oficial](https://www.paddleocr.com)
+
+### Paso 2: Despliegue local
+Para uso local, consulte la siguiente documentación según sus necesidades:
+
+- **Serie PP-OCR**: Consulte la [Documentación de PP-OCR](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/OCR.html)
+- **Serie PaddleOCR-VL**: Consulte la [Documentación de PaddleOCR-VL](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PaddleOCR-VL.html)
+- **PP-StructureV3**: Consulte la [Documentación de PP-StructureV3](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PP-StructureV3.html)
+- **Más capacidades**: Consulte la [Documentación de más capacidades](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/pipeline_overview.html)
+
+
+## 🧩 Más características
+
+- Convertir modelos al formato ONNX: [Obtención de modelos ONNX](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/others/obtaining_onnx_models.html).
+- Acelerar la inferencia usando motores como OpenVINO, ONNX Runtime, TensorRT, o realizar inferencia usando modelos en formato ONNX: [Inferencia de alto rendimiento](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/local_inference/high_performance_inference.html).
+- Acelerar la inferencia usando múltiples GPU y múltiples procesos: [Inferencia paralela para pipelines](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/pipeline_usage/instructions/parallel_inference.html).
+- Integrar PaddleOCR en aplicaciones escritas en C++, C#, Java, etc.: [Serving](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/serving/serving.html).
+
+## 🔄 Resumen rápido de los resultados de ejecución
+
+### PP-OCRv5
+
+<div align="center">
+  <p>
+       <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-OCRv5_demo.gif" alt="PP-OCRv5 Demo">
+  </p>
+</div>
+
+
+
+### PP-StructureV3
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-StructureV3_demo.gif" alt="PP-StructureV3 Demo">
+  </p>
+</div>
+
+### PaddleOCR-VL
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PaddleOCR-VL_demo.gif" alt="PP-StructureV3 Demo">
+  </p>
+</div>
+
+
+## ✨ Mantente al día
+
+⭐ **¡Dale una estrella a este repositorio para estar al tanto de emocionantes actualizaciones y nuevos lanzamientos, incluyendo potentes capacidades de OCR y análisis de documentos!** ⭐
+
+<div align="center">
+  <p>
+       <img width="1200" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/star_paddleocr2.en.gif" alt="Star-Project">
+  </p>
+</div>
+
+
+## 👩‍👩‍👧‍👦 Comunidad
+
+<div align="center">
+
+| Cuenta oficial de PaddlePaddle en WeChat | Únete al grupo de discusión técnica |
+| :---: | :---: |
+| <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qrcode_for_paddlepaddle_official_account.jpg" width="150"> | <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qr_code_for_the_questionnaire.jpg" width="150"> |
+</div>
+
+
+## 😃 Proyectos destacados que utilizan PaddleOCR
+¡PaddleOCR no estaría donde está hoy sin su increíble comunidad! 💗 Un enorme agradecimiento a todos nuestros socios de larga data, nuevos colaboradores y a todos los que han volcado su pasión en PaddleOCR — los hayamos mencionado o no. ¡Su apoyo alimenta nuestro fuego!
+
+<div align="center">
+
+| Nombre del proyecto | Descripción |
+| ------------ | ----------- |
+| [Dify](https://github.com/langgenius/dify) <a href="https://github.com/langgenius/dify"><img src="https://img.shields.io/github/stars/langgenius/dify"></a>|Plataforma lista para producción para el desarrollo de flujos de trabajo agénticos.|
+| [RAGFlow](https://github.com/infiniflow/ragflow) <a href="https://github.com/infiniflow/ragflow"><img src="https://img.shields.io/github/stars/infiniflow/ragflow"></a>|Motor RAG basado en la comprensión profunda de documentos.|
+| [pathway](https://github.com/pathwaycom/pathway) <a href="https://github.com/pathwaycom/pathway"><img src="https://img.shields.io/github/stars/pathwaycom/pathway"></a>|Framework ETL de Python para procesamiento de flujos, análisis en tiempo real, pipelines de LLM y RAG.|
+| [MinerU](https://github.com/opendatalab/MinerU) <a href="https://github.com/opendatalab/MinerU"><img src="https://img.shields.io/github/stars/opendatalab/MinerU"></a>|Herramienta de conversión de documentos de múltiples tipos a Markdown.|
+| [Umi-OCR](https://github.com/hiroi-sora/Umi-OCR) <a href="https://github.com/hiroi-sora/Umi-OCR"><img src="https://img.shields.io/github/stars/hiroi-sora/Umi-OCR"></a>|Software de OCR offline por lotes, gratuito y de código abierto.|
+| [cherry-studio](https://github.com/CherryHQ/cherry-studio) <a href="https://github.com/CherryHQ/cherry-studio"><img src="https://img.shields.io/github/stars/CherryHQ/cherry-studio"></a>|Cliente de escritorio compatible con múltiples proveedores de LLM.|
+| [haystack](https://github.com/deepset-ai/haystack)<a href="https://github.com/deepset-ai/haystack"><img src="https://img.shields.io/github/stars/deepset-ai/haystack"></a> |Framework de orquestación de IA para construir aplicaciones LLM personalizables y listas para producción.|
+| [OmniParser](https://github.com/microsoft/OmniParser)<a href="https://github.com/microsoft/OmniParser"><img src="https://img.shields.io/github/stars/microsoft/OmniParser"></a> |OmniParser: herramienta de análisis de pantalla para agentes GUI basados únicamente en visión.|
+| [QAnything](https://github.com/netease-youdao/QAnything)<a href="https://github.com/netease-youdao/QAnything"><img src="https://img.shields.io/github/stars/netease-youdao/QAnything"></a> |Preguntas y respuestas basadas en cualquier cosa.|
+| [Ver más proyectos](./awesome_projects.md) | [Más proyectos basados en PaddleOCR](./awesome_projects.md)|
+</div>
+
+## 👩‍👩‍👧‍👦 Colaboradores
+
+<div align="center">
+<a href="https://github.com/PaddlePaddle/PaddleOCR/graphs/contributors">
+  <img src="https://contrib.rocks/image?repo=PaddlePaddle/PaddleOCR&max=400&columns=20"  width="800"/>
+</a>
+</div>
+
+## 🌟 Estrellas
+
+<div align="center">
+  <p>
+      <img width="800" src="https://api.star-history.com/svg?repos=PaddlePaddle/PaddleOCR&type=Date" alt="Star-history">
+  </p>
+</div>
+
+
+## 📄 Licencia
+Este proyecto se publica bajo la [licencia Apache 2.0](LICENSE).
+
+## 🎓 Cita
+
+```bibtex
+@misc{cui2025paddleocr30technicalreport,
+      title={PaddleOCR 3.0 Technical Report}, 
+      author={Cheng Cui and Ting Sun and Manhui Lin and Tingquan Gao and Yubo Zhang and Jiaxuan Liu and Xueqing Wang and Zelun Zhang and Changda Zhou and Hongen Liu and Yue Zhang and Wenyu Lv and Kui Huang and Yichao Zhang and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2507.05595},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2507.05595}, 
+}
+
+@misc{cui2025paddleocrvlboostingmultilingualdocument,
+      title={PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model}, 
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Handong Zheng and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2510.14528},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2510.14528}, 
+}
+
+@misc{cui2026paddleocrvl15multitask09bvlm,
+      title={PaddleOCR-VL-1.5: Towards a Multi-Task 0.9B VLM for Robust In-the-Wild Document Parsing}, 
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2026},
+      eprint={2601.21957},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2601.21957}, 
+}
diff --git a/readme/README_fr.md b/readme/README_fr.md
new file mode 100644
index 0000000..29815f3
--- /dev/null
+++ b/readme/README_fr.md
@@ -0,0 +1,302 @@
+<div align="center">
+  <p>
+      <img width="800" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Banner.png" alt="Star-history">
+  </p>
+
+
+
+<h3>Boîte à outils OCR de pointe mondiale & Moteur d'IA documentaire</h3>
+
+[English](../README.md) | [简体中文](./README_cn.md) | [繁體中文](./README_tcn.md) | [日本語](./README_ja.md) | [한국어](./README_ko.md) | Français | [Русский](./README_ru.md) | [Español](./README_es.md) | [العربية](./README_ar.md)
+
+<!-- icon -->
+
+[![PyPI Downloads](https://static.pepy.tech/badge/paddleocr)](https://pepy.tech/projects/paddleocr)
+[![Used by](https://img.shields.io/badge/Used%20by-6k%2B%20repositories-blue)](https://github.com/PaddlePaddle/PaddleOCR/network/dependents)
+![python](https://img.shields.io/badge/python-3.8~3.12-aff.svg)
+![os](https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-pink.svg)
+![hardware](https://img.shields.io/badge/hardware-cpu%2C%20gpu%2C%20xpu%2C%20npu-yellow.svg)
+
+[![AI Studio](https://img.shields.io/badge/PaddleOCR-_Site_Officiel-1927BA?logo=data:image/png;base64,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&labelColor=white)](https://www.paddleocr.com)
+[![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/PaddlePaddle/PaddleOCR)
+[![License](https://img.shields.io/badge/license-Apache_2.0-green)](../LICENSE)
+
+</div>
+
+
+
+
+
+
+**PaddleOCR convertit des documents et des images en données structurées prêtes pour les LLM (JSON/Markdown) avec une précision de pointe dans l'industrie. Avec plus de 70k étoiles et la confiance de projets de premier plan tels que Dify, RAGFlow et Cherry Studio, PaddleOCR est le socle fondamental pour construire des applications RAG intelligentes et des applications Agentiques.**
+
+
+## 🚀 Fonctionnalités clés
+
+### 📄 Analyse intelligente de documents (prêt pour les LLM)
+> *Transformer des visuels désordonnés en données structurées pour l'ère des LLM.*
+
+* **VLM documentaire de pointe** : Avec **PaddleOCR-VL-1.6 (0,9 milliard de paramètres)**, le modèle vision-langage léger de pointe de l'industrie pour l'analyse de documents. Il atteint 96,3 % de précision sur OmniDocBench v1.6, mène l'industrie en reconnaissance de texte, formules et tableaux, et améliore significativement les capacités pour les documents anciens, les caractères rares, les sceaux et les graphiques, avec des sorties structurées aux formats **Markdown** et **JSON**.
+* **Conversion avec conscience de la structure** : Propulsé par **PP-StructureV3**, convertissez sans effort des PDF et images complexes en **Markdown** ou **JSON**. Contrairement aux modèles de la série PaddleOCR-VL, il fournit des informations de coordonnées plus fines, incluant les coordonnées des cellules de tableau, les coordonnées du texte, et bien plus encore.
+* **Efficacité prête pour la production** : Atteignez une précision de niveau commercial avec une empreinte ultra-réduite. Surpasse de nombreuses solutions propriétaires sur les benchmarks publics tout en restant économe en ressources pour le déploiement en périphérie ou dans le cloud.
+
+### 🔍 Reconnaissance de texte universelle (OCR de scène)
+> *L'étalon-or mondial pour la détection de texte multilingue à haute vitesse.*
+
+* **Plus de 100 langues supportées** : Reconnaissance native pour une vaste bibliothèque mondiale. **PP-OCRv6** supporte 50 langues avec un seul modèle unifié (chinois, anglais, japonais et 46 langues latines) — sans changement de modèle.
+* **Maîtrise des éléments complexes** : Au-delà de la reconnaissance de texte standard, nous prenons en charge la **détection de texte en scène naturelle** dans une large gamme d'environnements, y compris les pièces d'identité, les vues de rue, les livres et les composants industriels.
+* **Bond en performance** : PP-OCRv6 atteint **+4.6% en détection** et **+5.1% en reconnaissance** par rapport à PP-OCRv5, surpassant les principaux modèles de langage visuel. Accélération 5.2× en inférence CPU de bout en bout.
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Arch.jpg" alt="Architecture PaddleOCR">
+  </p>
+</div>
+
+### 🛠️ Écosystème centré sur les développeurs
+* **Intégration transparente** : Le premier choix pour l'écosystème des agents IA — profondément intégré avec **Dify, RAGFlow, Pathway et Cherry Studio**.
+* **Volant de données pour LLM** : Un pipeline complet pour construire des jeux de données de haute qualité, fournissant un « Moteur de données » durable pour l'affinage des grands modèles de langage.
+* **Déploiement en un clic** : Prend en charge divers backends matériels (GPU NVIDIA, CPU Intel, XPU Kunlunxin et divers accélérateurs IA).
+
+
+## 📣 Mises à jour récentes
+
+### 🔥 2026.06.11 : Publication de PaddleOCR 3.7.0
+- Points forts de PP-OCRv6 :
+
+    - **Amélioration de la précision** : Le niveau medium atteint +4.6% en détection et +5.1% en reconnaissance par rapport à PP-OCRv5_server, surpassant les principaux VLMs (Qwen3-VL-235B, GPT-5.5) avec seulement 34.5M paramètres.
+    - **50 langues unifiées** : Un seul modèle couvre le chinois, l'anglais, le japonais et 46 langues latines — sans changement de modèle.
+    - **Scénarios spécialisés** : Améliorations majeures pour les écrans numériques, caractères matriciels, empreintes de pneus et texte industriel.
+    - **Inférence plus rapide** : Accélération 5.2× CPU (OpenVINO), 6.1× sur Apple M4 (tiny), 0.13s sur A100 GPU.
+    - **Trois niveaux pour tous les scénarios** : tiny (1.5M) / small (7.7M) / medium (34.5M) pour le déploiement edge, mobile et serveur.
+    - **Disponibilité des modèles** : Tous les modèles sont disponibles sur [HuggingFace](https://huggingface.co/collections/PaddlePaddle/pp-ocrv6) et [ModelScope](https://www.modelscope.cn/collections/PaddlePaddle/PP-OCRv6).
+
+<details>
+<summary><strong>2026.05.28 : Publication de PaddleOCR 3.6.0</strong></summary>
+
+- Points forts de PaddleOCR-VL-1.6 :
+
+    - **Nouvelle précision SOTA** : 96.3% sur OmniDocBench v1.6, nouveau SOTA sur OmniDocBench v1.5 et Real5-OmniDocBench.
+    - **Amélioration complète** : améliorations significatives des tableaux, documents anciens et caractères rares.
+    - **Migration transparente** : architecture entièrement compatible avec PaddleOCR-VL-1.5.
+    - **Essayez** : [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.6) ou [site officiel](https://www.paddleocr.com).
+
+</details>
+<details>
+<summary><strong>2026.04.21 : Publication de PaddleOCR 3.5.0</strong></summary>
+
+* **Backends d'inférence plus flexibles** : basculez en toute transparence entre les graphes statiques Paddle, les graphes dynamiques Paddle et Transformers. PaddleOCR est désormais profondément intégré à l'écosystème Hugging Face, et 20 modèles majeurs prennent en charge Transformers comme backend d'inférence.
+* **Conversion des documents bureautiques en Markdown** : conversion des formats courants comme Word, Excel et PowerPoint vers Markdown.
+* **Export DOCX des résultats d'analyse** : les séries `PaddleOCR-VL`, `PP-StructureV3` et `PP-DocTranslation` prennent désormais en charge l'export des résultats d'analyse au format DOCX pour une consultation et une édition pratiques dans Microsoft Word.
+* **SDK officiel d'inférence dans le navigateur** : sortie du SDK officiel d'inférence dans le navigateur `PaddleOCR.js`, qui permet d'exécuter `PP-OCRv5` directement dans le navigateur.
+
+</details>
+
+<details>
+<summary><strong>2026.01.29 : Publication de PaddleOCR 3.4.0</strong></summary>
+* PaddleOCR-VL-1.5 (VLM 0,9 milliard de paramètres, état de l'art) : Notre dernier modèle phare pour l'analyse de documents est désormais disponible !
+    * **94,5 % de précision sur OmniDocBench** : Surpasse les grands modèles généralistes de premier rang et les analyseurs de documents spécialisés.
+    * **Robustesse dans le monde réel** : Premier à introduire l'algorithme **PP-DocLayoutV3** pour le positionnement de formes irrégulières, maîtrisant 5 scénarios difficiles : *Inclinaison, Déformation, Numérisation, Éclairage et Photographie d'écran*.
+    * **Extension des capacités** : Prend désormais en charge la **Reconnaissance de sceaux**, la **Détection de texte**, et s'étend à **111 langues** (incluant le tibétain et le bengali).
+    * **Maîtrise des longs documents** : Prend en charge la fusion automatique de tableaux sur plusieurs pages et l'identification hiérarchique des titres.
+    * **Essayez-le maintenant** : Disponible sur [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.5) ou sur notre [Site officiel](https://www.paddleocr.com).
+
+</details>
+
+<details>
+<summary><strong>2025.10.16 : Publication de PaddleOCR 3.3.0</strong></summary>
+
+- Publication de PaddleOCR-VL :
+    - **Présentation du modèle** :
+        - **PaddleOCR-VL** est un modèle de pointe et économe en ressources, spécialement conçu pour l'analyse de documents. Son composant principal est PaddleOCR-VL-0.9B, un modèle vision-langage (VLM) compact mais puissant qui intègre un encodeur visuel à résolution dynamique de style NaViT avec le modèle de langage ERNIE-4.5-0.3B pour permettre une reconnaissance précise des éléments. **Ce modèle innovant prend en charge efficacement 109 langues et excelle dans la reconnaissance d'éléments complexes (par exemple, texte, tableaux, formules et graphiques), tout en maintenant une consommation minimale de ressources**. Grâce à des évaluations complètes sur des benchmarks publics largement utilisés et des benchmarks internes, PaddleOCR-VL atteint des performances de pointe à la fois dans l'analyse de documents au niveau de la page et dans la reconnaissance au niveau des éléments. Il surpasse significativement les solutions existantes, présente une forte compétitivité face aux VLM de premier plan, et offre des vitesses d'inférence rapides. Ces atouts le rendent très adapté au déploiement pratique dans des scénarios du monde réel. Le modèle a été publié sur [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL). Tout le monde est invité à le télécharger et à l'utiliser ! Plus d'informations d'introduction sont disponibles dans [PaddleOCR-VL](https://www.paddleocr.ai/latest/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL.html).
+
+    - **Fonctionnalités principales** :
+        - **Architecture VLM compacte mais puissante** : Nous présentons un nouveau modèle vision-langage spécialement conçu pour une inférence économe en ressources, atteignant des performances remarquables dans la reconnaissance d'éléments. En intégrant un encodeur visuel haute résolution dynamique de style NaViT avec le modèle de langage léger ERNIE-4.5-0.3B, nous améliorons considérablement les capacités de reconnaissance et l'efficacité du décodage du modèle. Cette intégration maintient une haute précision tout en réduisant les besoins de calcul, ce qui le rend bien adapté aux applications de traitement de documents efficaces et pratiques.
+        - **Performances de pointe en analyse de documents** : PaddleOCR-VL atteint des performances à l'état de l'art à la fois dans l'analyse de documents au niveau de la page et dans la reconnaissance au niveau des éléments. Il surpasse significativement les solutions existantes basées sur des pipelines et présente une forte compétitivité face aux principaux modèles vision-langage (VLM) en analyse de documents. De plus, il excelle dans la reconnaissance d'éléments documentaires complexes, tels que le texte, les tableaux, les formules et les graphiques, ce qui le rend adapté à une large gamme de types de contenu difficiles, y compris le texte manuscrit et les documents historiques. Cela le rend très polyvalent et adapté à une large gamme de types de documents et de scénarios.
+        - **Support multilingue** : PaddleOCR-VL prend en charge 109 langues, couvrant les principales langues mondiales, notamment le chinois, l'anglais, le japonais, le latin et le coréen, ainsi que les langues avec des scripts et des structures différents, tels que le russe (script cyrillique), l'arabe, l'hindi (script devanagari) et le thaï. Cette large couverture linguistique améliore considérablement l'applicabilité de notre système aux scénarios de traitement de documents multilingues et mondialisés.
+
+- Publication du modèle de reconnaissance multilingue PP-OCRv5 :
+    - Amélioration de la précision et de la couverture de la reconnaissance des scripts latins ; ajout de la prise en charge des systèmes cyrillique, arabe, devanagari, télougou, tamoul et d'autres systèmes linguistiques, couvrant la reconnaissance de 109 langues. Le modèle ne compte que 2 millions de paramètres, et la précision de certains modèles a augmenté de plus de 40 % par rapport à la génération précédente.
+
+</details>
+
+
+<details>
+<summary><strong>2025.08.21 : Publication de PaddleOCR 3.2.0</strong></summary>
+
+- **Ajouts significatifs de modèles :**
+    - Introduction de l'entraînement, de l'inférence et du déploiement pour les modèles de reconnaissance PP-OCRv5 en anglais, thaï et grec. **Le modèle PP-OCRv5 anglais apporte une amélioration de 11 % dans les scénarios en anglais par rapport au modèle principal PP-OCRv5, avec les modèles de reconnaissance thaï et grec atteignant des précisions de 82,68 % et 89,28 %, respectivement.**
+
+- **Améliorations des capacités de déploiement :**
+    - **Prise en charge complète des versions 3.1.0 et 3.1.1 du framework PaddlePaddle.**
+    - **Mise à niveau complète de la solution de déploiement local C++ de PP-OCRv5, prenant désormais en charge Linux et Windows, avec une parité de fonctionnalités et une précision identique à l'implémentation Python.**
+    - **L'inférence haute performance prend désormais en charge CUDA 12, et l'inférence peut être effectuée en utilisant le backend Paddle Inference ou ONNX Runtime.**
+    - **La solution de déploiement orientée services à haute stabilité est désormais entièrement open-source, permettant aux utilisateurs de personnaliser les images Docker et les SDK selon leurs besoins.**
+    - La solution de déploiement orientée services à haute stabilité prend également en charge l'invocation via des requêtes HTTP construites manuellement, permettant le développement de code client dans n'importe quel langage de programmation.
+
+- **Support des benchmarks :**
+    - **Toutes les lignes de production prennent désormais en charge des benchmarks granulaires, permettant la mesure du temps d'inférence de bout en bout ainsi que les données de latence par couche et par module pour faciliter l'analyse des performances. [Voici](docs/version3.x/pipeline_usage/instructions/benchmark.en.md) comment configurer et utiliser la fonctionnalité de benchmark.**
+    - **La documentation a été mise à jour pour inclure les métriques clés pour les configurations couramment utilisées sur le matériel grand public, telles que la latence d'inférence et l'utilisation de la mémoire, fournissant des références de déploiement pour les utilisateurs.**
+
+- **Corrections de bugs :**
+    - Résolution du problème de l'échec de sauvegarde des journaux lors de l'entraînement du modèle.
+    - Mise à niveau du composant d'augmentation de données pour les modèles de formules pour la compatibilité avec les nouvelles versions de la dépendance albumentations, et correction des avertissements de blocage lors de l'utilisation du package tokenizers dans des scénarios multi-processus.
+    - Correction des incohérences dans les comportements des commutateurs (par exemple, `use_chart_parsing`) dans les fichiers de configuration PP-StructureV3 par rapport aux autres pipelines.
+
+- **Autres améliorations :**
+    - **Séparation des dépendances principales et optionnelles. Seules les dépendances principales minimales sont requises pour la reconnaissance de texte de base ; des dépendances supplémentaires pour l'analyse de documents et l'extraction d'informations peuvent être installées selon les besoins.**
+    - **Activation de la prise en charge des cartes graphiques NVIDIA RTX série 50 sous Windows ; les utilisateurs peuvent consulter le [guide d'installation](docs/version3.x/installation.en.md) pour les versions correspondantes du framework PaddlePaddle.**
+    - **Les modèles de la série PP-OCR prennent désormais en charge le retour des coordonnées de chaque caractère.**
+    - Ajout de sources de téléchargement de modèles AIStudio, ModelScope et autres, permettant aux utilisateurs de spécifier la source pour les téléchargements de modèles.
+    - Ajout de la prise en charge de la conversion graphique en tableau via le module PP-Chart2Table.
+    - Optimisation des descriptions de documentation pour améliorer la facilité d'utilisation.
+</details>
+
+
+[Journal des modifications](https://paddlepaddle.github.io/PaddleOCR/latest/en/update/update.html)
+
+
+## 🚀 Démarrage rapide
+
+### Étape 1 : Essayer en ligne
+Le site officiel de PaddleOCR propose un **Centre d'expérience** interactif et des **API** — aucune configuration requise, un seul clic pour découvrir.
+
+👉 [Visiter le site officiel](https://www.paddleocr.com)
+
+### Étape 2 : Déploiement local
+Pour une utilisation locale, veuillez consulter la documentation suivante en fonction de vos besoins :
+
+- **Série PP-OCR** : Voir la [documentation PP-OCR](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/OCR.html)
+- **Série PaddleOCR-VL** : Voir la [documentation PaddleOCR-VL](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PaddleOCR-VL.html)
+- **PP-StructureV3** : Voir la [documentation PP-StructureV3](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PP-StructureV3.html)
+- **Autres capacités** : Voir la [documentation sur les autres capacités](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/pipeline_overview.html)
+
+
+## 🧩 Plus de fonctionnalités
+
+- Convertir des modèles au format ONNX : [Obtenir des modèles ONNX](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/others/obtaining_onnx_models.html).
+- Accélérer l'inférence à l'aide de moteurs tels qu'OpenVINO, ONNX Runtime, TensorRT, ou effectuer une inférence à l'aide de modèles au format ONNX : [Inférence haute performance](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/local_inference/high_performance_inference.html).
+- Accélérer l'inférence à l'aide de plusieurs GPU et plusieurs processus : [Inférence parallèle pour les pipelines](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/pipeline_usage/instructions/parallel_inference.html).
+- Intégrer PaddleOCR dans des applications écrites en C++, C#, Java, etc. : [Services](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/serving/serving.html).
+
+## 🔄 Aperçu rapide des résultats d'exécution
+
+### PP-OCRv5
+
+<div align="center">
+  <p>
+       <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-OCRv5_demo.gif" alt="Démo PP-OCRv5">
+  </p>
+</div>
+
+
+
+### PP-StructureV3
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-StructureV3_demo.gif" alt="Démo PP-StructureV3">
+  </p>
+</div>
+
+### PaddleOCR-VL
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PaddleOCR-VL_demo.gif" alt="Démo PP-StructureV3">
+  </p>
+</div>
+
+
+## ✨ Restez informé
+
+⭐ **Mettez une étoile à ce dépôt pour suivre les mises à jour passionnantes et les nouvelles versions, y compris les puissantes capacités d'OCR et d'analyse de documents !** ⭐
+
+<div align="center">
+  <p>
+       <img width="1200" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/star_paddleocr2.en.gif" alt="Star-Project">
+  </p>
+</div>
+
+
+## 👩‍👩‍👧‍👦 Communauté
+
+<div align="center">
+
+| Compte officiel WeChat de PaddlePaddle | Rejoindre le groupe de discussion technique |
+| :---: | :---: |
+| <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qrcode_for_paddlepaddle_official_account.jpg" width="150"> | <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qr_code_for_the_questionnaire.jpg" width="150"> |
+</div>
+
+
+## 😃 Projets remarquables utilisant PaddleOCR
+PaddleOCR n'en serait pas là aujourd'hui sans son incroyable communauté ! 💗 Un immense merci à tous nos partenaires de longue date, aux nouveaux collaborateurs et à tous ceux qui ont mis leur passion dans PaddleOCR — que nous vous ayons cités ou non. Votre soutien alimente notre feu !
+
+<div align="center">
+
+| Nom du projet | Description |
+| ------------ | ----------- |
+| [Dify](https://github.com/langgenius/dify) <a href="https://github.com/langgenius/dify"><img src="https://img.shields.io/github/stars/langgenius/dify"></a>|Plateforme prête pour la production pour le développement de flux de travail agentiques.|
+| [RAGFlow](https://github.com/infiniflow/ragflow) <a href="https://github.com/infiniflow/ragflow"><img src="https://img.shields.io/github/stars/infiniflow/ragflow"></a>|Moteur RAG basé sur la compréhension approfondie des documents.|
+| [pathway](https://github.com/pathwaycom/pathway) <a href="https://github.com/pathwaycom/pathway"><img src="https://img.shields.io/github/stars/pathwaycom/pathway"></a>|Framework Python ETL pour le traitement de flux, l'analytique en temps réel, les pipelines LLM et le RAG.|
+| [MinerU](https://github.com/opendatalab/MinerU) <a href="https://github.com/opendatalab/MinerU"><img src="https://img.shields.io/github/stars/opendatalab/MinerU"></a>|Outil de conversion de documents multi-types en Markdown.|
+| [Umi-OCR](https://github.com/hiroi-sora/Umi-OCR) <a href="https://github.com/hiroi-sora/Umi-OCR"><img src="https://img.shields.io/github/stars/hiroi-sora/Umi-OCR"></a>|Logiciel OCR hors ligne par lots, gratuit et open-source.|
+| [cherry-studio](https://github.com/CherryHQ/cherry-studio) <a href="https://github.com/CherryHQ/cherry-studio"><img src="https://img.shields.io/github/stars/CherryHQ/cherry-studio"></a>|Un client de bureau prenant en charge plusieurs fournisseurs de LLM.|
+| [haystack](https://github.com/deepset-ai/haystack)<a href="https://github.com/deepset-ai/haystack"><img src="https://img.shields.io/github/stars/deepset-ai/haystack"></a> |Framework d'orchestration IA pour construire des applications LLM personnalisables et prêtes pour la production.|
+| [OmniParser](https://github.com/microsoft/OmniParser)<a href="https://github.com/microsoft/OmniParser"><img src="https://img.shields.io/github/stars/microsoft/OmniParser"></a> |OmniParser : Outil d'analyse d'écran pour agent GUI basé sur la vision pure.|
+| [QAnything](https://github.com/netease-youdao/QAnything)<a href="https://github.com/netease-youdao/QAnything"><img src="https://img.shields.io/github/stars/netease-youdao/QAnything"></a> |Questions et réponses basées sur n'importe quoi.|
+| [En savoir plus sur les projets](./awesome_projects.md) | [Plus de projets basés sur PaddleOCR](./awesome_projects.md)|
+</div>
+
+## 👩‍👩‍👧‍👦 Contributeurs
+
+<div align="center">
+<a href="https://github.com/PaddlePaddle/PaddleOCR/graphs/contributors">
+  <img src="https://contrib.rocks/image?repo=PaddlePaddle/PaddleOCR&max=400&columns=20"  width="800"/>
+</a>
+</div>
+
+## 🌟 Étoiles
+
+<div align="center">
+  <p>
+      <img width="800" src="https://api.star-history.com/svg?repos=PaddlePaddle/PaddleOCR&type=Date" alt="Star-history">
+  </p>
+</div>
+
+
+## 📄 Licence
+Ce projet est publié sous la [licence Apache 2.0](LICENSE).
+
+## 🎓 Citation
+
+```bibtex
+@misc{cui2025paddleocr30technicalreport,
+      title={PaddleOCR 3.0 Technical Report}, 
+      author={Cheng Cui and Ting Sun and Manhui Lin and Tingquan Gao and Yubo Zhang and Jiaxuan Liu and Xueqing Wang and Zelun Zhang and Changda Zhou and Hongen Liu and Yue Zhang and Wenyu Lv and Kui Huang and Yichao Zhang and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2507.05595},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2507.05595}, 
+}
+
+@misc{cui2025paddleocrvlboostingmultilingualdocument,
+      title={PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model}, 
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Handong Zheng and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2510.14528},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2510.14528}, 
+}
+
+@misc{cui2026paddleocrvl15multitask09bvlm,
+      title={PaddleOCR-VL-1.5: Towards a Multi-Task 0.9B VLM for Robust In-the-Wild Document Parsing}, 
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2026},
+      eprint={2601.21957},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2601.21957}, 
+}
diff --git a/readme/README_ja.md b/readme/README_ja.md
new file mode 100644
index 0000000..4ff6c92
--- /dev/null
+++ b/readme/README_ja.md
@@ -0,0 +1,304 @@
+
+<div align="center">
+  <p>
+      <img width="800" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Banner.png" alt="Star履歴">
+  </p>
+
+
+
+<h3>世界をリードするOCRツールキット & ドキュメントAIエンジン</h3>
+
+[English](../README.md) | [简体中文](./README_cn.md) | [繁體中文](./README_tcn.md) | 日本語 | [한국어](./README_ko.md) | [Français](./README_fr.md) | [Русский](./README_ru.md) | [Español](./README_es.md) | [العربية](./README_ar.md)
+
+<!-- icon -->
+
+[![PyPI Downloads](https://static.pepy.tech/badge/paddleocr)](https://pepy.tech/projects/paddleocr)
+[![Used by](https://img.shields.io/badge/Used%20by-6k%2B%20repositories-blue)](https://github.com/PaddlePaddle/PaddleOCR/network/dependents)
+![python](https://img.shields.io/badge/python-3.8~3.12-aff.svg)
+![os](https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-pink.svg)
+![hardware](https://img.shields.io/badge/hardware-cpu%2C%20gpu%2C%20xpu%2C%20npu-yellow.svg)
+
+[![AI Studio](https://img.shields.io/badge/PaddleOCR-_Offiical_Website-1927BA?logo=data:image/png;base64,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&labelColor=white)](https://www.paddleocr.com)
+[![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/PaddlePaddle/PaddleOCR)
+[![License](https://img.shields.io/badge/license-Apache_2.0-green)](../LICENSE)
+
+</div>
+
+
+
+
+
+
+**PaddleOCRは、ドキュメントや画像を業界最高水準の精度で構造化されたLLM対応データ（JSON/Markdown）に変換します。70,000以上のStarを獲得し、Dify、RAGFlow、Cherry Studioなどの一流プロジェクトで採用されているPaddleOCRは、インテリジェントなRAGおよびエージェントアプリケーション構築の基盤です。**
+
+
+## 🚀 主な機能
+
+### 📄 インテリジェントドキュメント解析（LLM対応）
+> *LLM時代に向けて、雑然とした視覚データを構造化データに変換*
+
+* **最先端のドキュメントVLM**: 業界をリードする軽量視覚言語モデル **PaddleOCR-VL-1.6（0.9B）** を搭載。OmniDocBench v1.6 で 96.3% の精度を達成し、テキスト、数式、表の認識で業界をリード。古文書、稀な文字、印鑑、チャートなど多シーンの能力も大幅に強化され、**Markdown**および**JSON**形式の構造化出力に対応しています。
+* **構造認識型変換**: **PP-StructureV3**を活用し、複雑なPDFや画像を**Markdown**または**JSON**にシームレスに変換します。PaddleOCR-VLシリーズモデルとは異なり、テーブルセル座標、テキスト座標などのより詳細な座標情報を提供します。
+* **本番環境対応の効率性**: 超小型フットプリントで商用レベルの精度を実現。公開ベンチマークで多くのクローズドソースソリューションを凌駕しつつ、エッジ/クラウドデプロイメントに対してリソース効率を維持します。
+
+### 🔍 汎用テキスト認識（シーンOCR）
+> *高速・多言語テキスト検出のグローバルスタンダード*
+
+* **100以上の言語をサポート**: 広範なグローバル言語ライブラリのネイティブ認識。**PP-OCRv6**は単一モデルで50言語（中国語、英語、日本語、46のラテン語系言語）を統一サポートします。
+* **複雑な要素への対応力**: 標準的なテキスト認識を超え、身分証明書、街頭風景、書籍、産業部品など、幅広い環境での**自然シーンテキスト検出**をサポートします。
+* **性能の飛躍的向上**: PP-OCRv6はPP-OCRv5と比較して検出精度**+4.6%**、認識精度**+5.1%**を達成し、主要な視覚言語モデルを上回ります。CPU推論5.2×高速化。
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Arch.jpg" alt="PaddleOCR アーキテクチャ">
+  </p>
+</div>
+
+### 🛠️ 開発者中心のエコシステム
+* **シームレスな統合**: AIエージェントエコシステムの最良の選択肢 ── **Dify、RAGFlow、Pathway、Cherry Studio**と深く統合されています。
+* **LLMデータフライホイール**: 高品質データセットを構築する完全なパイプラインを提供し、大規模言語モデルのファインチューニングのための持続可能な「データエンジン」を実現します。
+* **ワンクリックデプロイ**: さまざまなハードウェアバックエンド（NVIDIA GPU、Intel CPU、Kunlunxin XPU、各種AIアクセラレータ）をサポートします。
+
+
+## 📣 最新情報
+
+### 🔥 2026.06.11: PaddleOCR 3.7.0リリース
+- PP-OCRv6の主な特徴：
+
+    - **精度向上**: mediumティアはPP-OCRv5_serverと比較して検出+4.6%、認識+5.1%を達成、34.5Mパラメータのみで主要VLM（Qwen3-VL-235B、GPT-5.5）を上回る。
+    - **50言語統一サポート**: 単一モデルで中国語、英語、日本語、46のラテン語系言語をカバー — モデル切替不要。
+    - **専門シナリオ強化**: デジタルディスプレイ、ドットマトリックス文字、タイヤ印字、工業文字の認識が大幅向上。
+    - **高速推論**: CPU 5.2×高速化（OpenVINO）、Apple M4 6.1×（tiny）、A100 GPUで0.13s。
+    - **全シナリオ対応3ティア**: tiny（1.5M）/ small（7.7M）/ medium（34.5M）でエッジからサーバーまで対応。
+    - **モデル入手**: 全モデルは [HuggingFace](https://huggingface.co/collections/PaddlePaddle/pp-ocrv6) および [ModelScope](https://www.modelscope.cn/collections/PaddlePaddle/PP-OCRv6) で公開中。
+
+<details>
+<summary><strong>2026.05.28: PaddleOCR 3.6.0リリース</strong></summary>
+
+- PaddleOCR-VL-1.6の主な特徴：
+
+    - **新SOTA精度**: OmniDocBench v1.6で96.3%、OmniDocBench v1.5とReal5-OmniDocBenchでも新SOTA達成。
+    - **総合的な強化**: テーブル、古文書、希少文字の認識が大幅向上。
+    - **シームレスな移行**: PaddleOCR-VL-1.5と完全互換のアーキテクチャ。
+    - **今すぐ試す**: [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.6)または[公式サイト](https://www.paddleocr.com)。
+
+</details>
+<details>
+<summary><strong>2026.04.21: PaddleOCR 3.5.0リリース</strong></summary>
+
+* **柔軟な推論バックエンド**: Paddleの静的グラフ、動的グラフ、Transformersをシームレスに切り替え可能。Hugging Face エコシステムに深く対応し、主要20モデルがTransformersを推論バックエンドとしてサポート。
+* **Office文書をMarkdownに変換**: Word、Excel、PowerPoint などの一般的な文書形式を Markdown に変換可能。
+* **解析結果の DOCX 出力**: `PaddleOCR-VL` シリーズ、`PP-StructureV3`、`PP-DocTranslation` で、解析結果を DOCX として出力できるようになり、Microsoft Word での閲覧・編集が容易に。
+* **公式ブラウザ推論 SDK**: 公式ブラウザ推論 SDK `PaddleOCR.js` を公開し、ブラウザ上で `PP-OCRv5` を実行可能。
+
+</details>
+
+<details>
+<summary><strong>2026.01.29: PaddleOCR 3.4.0リリース</strong></summary>
+* PaddleOCR-VL-1.5（最先端の0.9B VLM）: ドキュメント解析のための最新フラッグシップモデルが公開されました！
+    * **OmniDocBenchで94.5%の精度**: トップクラスの汎用大規模モデルや専門ドキュメントパーサーを凌駕。
+    * **実環境でのロバスト性**: 非定型形状位置決定のための**PP-DocLayoutV3**アルゴリズムを初めて導入し、*傾き、歪み、スキャン、照明、画面撮影*の5つの困難なシナリオに対応。
+    * **機能拡張**: **印鑑認識**、**テキスト検出**をサポートし、**111言語**（中国のチベット文字やベンガル文字を含む）に対応拡大。
+    * **長文ドキュメントへの対応**: ページをまたがるテーブルの自動結合および階層的な見出し識別をサポート。
+    * **今すぐ試す**: [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.5)または[公式ウェブサイト](https://www.paddleocr.com)で利用可能です。
+
+</details>
+
+<details>
+<summary><strong>2025.10.16: PaddleOCR 3.3.0リリース</strong></summary>
+
+- PaddleOCR-VLをリリース:
+    - **モデル紹介**:
+        - **PaddleOCR-VL**はドキュメント解析に特化した最先端かつリソース効率の高いモデルです。コアコンポーネントであるPaddleOCR-VL-0.9Bは、NaViTスタイルの動的解像度ビジュアルエンコーダとERNIE-4.5-0.3B言語モデルを統合したコンパクトながらも強力な視覚言語モデル（VLM）であり、正確な要素認識を実現します。**この革新的なモデルは109言語を効率的にサポートし、複雑な要素（テキスト、テーブル、数式、チャートなど）の認識に優れつつ、リソース消費を最小限に抑えます**。広く使用されている公開ベンチマークおよび社内ベンチマークでの包括的な評価を通じて、PaddleOCR-VLはページレベルのドキュメント解析と要素レベルの認識の両方で最先端の性能を達成しています。既存のソリューションを大幅に上回り、トップクラスのVLMに対して高い競争力を示し、高速な推論速度を提供します。これらの強みにより、実世界のシナリオへの実践的なデプロイメントに非常に適しています。モデルは[HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL)で公開されています。ぜひダウンロードしてお使いください！詳細情報は[PaddleOCR-VL](https://www.paddleocr.ai/latest/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL.html)をご覧ください。
+
+    - **主要機能**:
+        - **コンパクトかつ強力なVLMアーキテクチャ**: リソース効率の高い推論に特化した新しい視覚言語モデルを提案し、要素認識において卓越した性能を実現しました。NaViTスタイルの動的高解像度ビジュアルエンコーダと軽量なERNIE-4.5-0.3B言語モデルを統合することで、モデルの認識能力とデコード効率を大幅に向上させました。この統合により、計算負荷を削減しつつ高い精度を維持し、効率的で実用的なドキュメント処理アプリケーションに適しています。
+        - **ドキュメント解析における最先端性能**: PaddleOCR-VLはページレベルのドキュメント解析と要素レベルの認識の両方で最先端の性能を達成しています。既存のパイプラインベースのソリューションを大幅に上回り、ドキュメント解析における主要な視覚言語モデル（VLM）に対して高い競争力を示しています。さらに、テキスト、テーブル、数式、チャートなどの複雑なドキュメント要素の認識に優れ、手書きテキストや歴史文書を含む幅広い種類のコンテンツに対応可能です。これにより、多様なドキュメントタイプやシナリオに対して高い汎用性を発揮します。
+        - **多言語サポート**: PaddleOCR-VLは109言語をサポートし、中国語、英語、日本語、ラテン語、韓国語をはじめ、ロシア語（キリル文字）、アラビア語、ヒンディー語（デーヴァナーガリー文字）、タイ語など、異なる文字体系や構造を持つ言語を含む主要なグローバル言語をカバーしています。この幅広い言語対応により、多言語およびグローバルなドキュメント処理シナリオへの適用性が大幅に向上しています。
+
+- PP-OCRv5多言語認識モデルをリリース:
+    - ラテン文字認識の精度とカバレッジを改善。キリル文字、アラビア文字、デーヴァナーガリー文字、テルグ文字、タミル文字など他の文字体系のサポートを追加し、109言語の認識をカバー。モデルのパラメータ数はわずか2Mで、一部のモデルの精度は前世代と比較して40%以上向上しています。
+
+</details>
+
+
+<details>
+<summary><strong>2025.08.21: PaddleOCR 3.2.0リリース</strong></summary>
+
+- **モデルの大幅な追加:**
+    - PP-OCRv5認識モデルの英語、タイ語、ギリシャ語の学習、推論、デプロイメントを導入。**PP-OCRv5英語モデルは、主要なPP-OCRv5モデルと比較して英語シナリオで11%の改善を達成し、タイ語およびギリシャ語の認識モデルはそれぞれ82.68%および89.28%の精度を実現しています。**
+
+- **デプロイメント機能のアップグレード:**
+    - **PaddlePaddleフレームワークバージョン3.1.0および3.1.1を完全サポート。**
+    - **PP-OCRv5 C++ローカルデプロイメントソリューションを全面アップグレードし、LinuxとWindowsの両方をサポート。Python実装と同等の機能および精度を実現。**
+    - **高性能推論がCUDA 12をサポートし、Paddle InferenceまたはONNX Runtimeバックエンドを使用した推論が可能。**
+    - **高安定性サービス指向デプロイメントソリューションが完全にオープンソース化され、ユーザーが必要に応じてDockerイメージやSDKをカスタマイズ可能。**
+    - 高安定性サービス指向デプロイメントソリューションは、手動でHTTPリクエストを構築して呼び出すこともサポートしており、任意のプログラミング言語でクライアント側コードの開発が可能です。
+
+- **ベンチマークサポート:**
+    - **全プロダクションラインがきめ細かいベンチマークをサポートし、エンドツーエンドの推論時間だけでなく、レイヤーごとおよびモジュールごとのレイテンシデータの測定が可能となり、性能分析を支援します。ベンチマーク機能のセットアップと使用方法は[こちら](docs/version3.x/pipeline_usage/instructions/benchmark.en.md)をご覧ください。**
+    - **主要なハードウェアでよく使用される構成の推論レイテンシやメモリ使用量などの重要な指標を含むようドキュメントを更新し、ユーザーにデプロイメントの参考情報を提供。**
+
+- **バグ修正:**
+    - モデル学習時のログ保存失敗の問題を解決。
+    - 数式モデルのデータ拡張コンポーネントをアップグレードし、albumentations依存関係の新しいバージョンとの互換性を確保。マルチプロセスシナリオでtokenizersパッケージ使用時のデッドロック警告を修正。
+    - PP-StructureV3設定ファイルにおけるスイッチ動作（例：`use_chart_parsing`）の他のパイプラインとの不整合を修正。
+
+- **その他の改善:**
+    - **コア依存関係とオプション依存関係を分離。基本的なテキスト認識には最小限のコア依存関係のみが必要で、ドキュメント解析や情報抽出のための追加依存関係は必要に応じてインストール可能。**
+    - **WindowsでNVIDIA RTX 50シリーズグラフィックスカードのサポートを有効化。対応するPaddlePaddleフレームワークバージョンについては[インストールガイド](docs/version3.x/installation.en.md)を参照してください。**
+    - **PP-OCRシリーズモデルが単一文字座標の返却をサポート。**
+    - AIStudio、ModelScopeなどのモデルダウンロードソースを追加し、ユーザーがモデルダウンロードのソースを指定可能に。
+    - PP-Chart2Tableモジュールによるチャートからテーブルへの変換サポートを追加。
+    - ドキュメントの説明を最適化し、使いやすさを向上。
+</details>
+
+
+[更新履歴](https://paddlepaddle.github.io/PaddleOCR/latest/en/update/update.html)
+
+
+## 🚀 クイックスタート
+
+### ステップ1：オンラインで試す
+PaddleOCR公式ウェブサイトでは、インタラクティブな**体験センター**と**API**を提供しています。セットアップ不要、ワンクリックで体験できます。
+
+👉 [公式ウェブサイトへアクセス](https://www.paddleocr.com)
+
+### ステップ2：ローカルデプロイメント
+ローカルでの使用については、ニーズに応じて以下のドキュメントを参照してください：
+
+- **PP-OCRシリーズ**: [PP-OCRドキュメント](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/OCR.html)を参照
+- **PaddleOCR-VLシリーズ**: [PaddleOCR-VLドキュメント](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PaddleOCR-VL.html)を参照
+- **PP-StructureV3**: [PP-StructureV3ドキュメント](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PP-StructureV3.html)を参照
+- **その他の機能**: [その他の機能ドキュメント](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/pipeline_overview.html)を参照
+
+
+## 🧩 その他の機能
+
+- モデルをONNX形式に変換: [ONNXモデルの取得](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/others/obtaining_onnx_models.html)
+- OpenVINO、ONNX Runtime、TensorRTなどのエンジンを使用した推論高速化、またはONNX形式モデルによる推論: [高性能推論](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/local_inference/high_performance_inference.html)
+- マルチGPUおよびマルチプロセスによる推論高速化: [パイプラインの並列推論](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/pipeline_usage/instructions/parallel_inference.html)
+- PaddleOCRをC++、C#、Javaなどで書かれたアプリケーションに統合: [サービスデプロイメント](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/serving/serving.html)
+
+## 🔄 実行結果の概要
+
+### PP-OCRv5
+
+<div align="center">
+  <p>
+       <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-OCRv5_demo.gif" alt="PP-OCRv5 デモ">
+  </p>
+</div>
+
+
+
+### PP-StructureV3
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-StructureV3_demo.gif" alt="PP-StructureV3 デモ">
+  </p>
+</div>
+
+### PaddleOCR-VL
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PaddleOCR-VL_demo.gif" alt="PaddleOCR-VL デモ">
+  </p>
+</div>
+
+
+## ✨ 最新情報をチェック
+
+⭐ **このリポジトリにStarを付けて、強力なOCRおよびドキュメント解析機能を含むエキサイティングなアップデートや新リリースを見逃さないようにしましょう！** ⭐
+
+<div align="center">
+  <p>
+       <img width="1200" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/star_paddleocr2.en.gif" alt="プロジェクトにStarを付ける">
+  </p>
+</div>
+
+
+## 👩‍👩‍👧‍👦 コミュニティ
+
+<div align="center">
+
+| PaddlePaddle WeChat公式アカウント | 技術ディスカッショングループに参加 |
+| :---: | :---: |
+| <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qrcode_for_paddlepaddle_official_account.jpg" width="150"> | <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qr_code_for_the_questionnaire.jpg" width="150"> |
+</div>
+
+
+## 😃 PaddleOCRを活用した素晴らしいプロジェクト
+PaddleOCRが今日あるのは、素晴らしいコミュニティのおかげです！💗 長年のパートナー、新たな協力者、そしてPaddleOCRに情熱を注いでくださったすべての皆様に心から感謝いたします。名前を挙げきれなかった方も含めて、皆様のサポートが私たちの原動力です！
+
+<div align="center">
+
+| プロジェクト名 | 説明 |
+| ------------ | ----------- |
+| [Dify](https://github.com/langgenius/dify) <a href="https://github.com/langgenius/dify"><img src="https://img.shields.io/github/stars/langgenius/dify"></a>|エージェントワークフロー開発のためのプロダクション対応プラットフォーム。|
+| [RAGFlow](https://github.com/infiniflow/ragflow) <a href="https://github.com/infiniflow/ragflow"><img src="https://img.shields.io/github/stars/infiniflow/ragflow"></a>|深いドキュメント理解に基づくRAGエンジン。|
+| [pathway](https://github.com/pathwaycom/pathway) <a href="https://github.com/pathwaycom/pathway"><img src="https://img.shields.io/github/stars/pathwaycom/pathway"></a>|ストリーム処理、リアルタイム分析、LLMパイプライン、RAG向けのPython ETLフレームワーク。|
+| [MinerU](https://github.com/opendatalab/MinerU) <a href="https://github.com/opendatalab/MinerU"><img src="https://img.shields.io/github/stars/opendatalab/MinerU"></a>|マルチタイプドキュメントからMarkdownへの変換ツール。|
+| [Umi-OCR](https://github.com/hiroi-sora/Umi-OCR) <a href="https://github.com/hiroi-sora/Umi-OCR"><img src="https://img.shields.io/github/stars/hiroi-sora/Umi-OCR"></a>|無料・オープンソースのバッチオフラインOCRソフトウェア。|
+| [cherry-studio](https://github.com/CherryHQ/cherry-studio) <a href="https://github.com/CherryHQ/cherry-studio"><img src="https://img.shields.io/github/stars/CherryHQ/cherry-studio"></a>|複数のLLMプロバイダーをサポートするデスクトップクライアント。|
+| [haystack](https://github.com/deepset-ai/haystack)<a href="https://github.com/deepset-ai/haystack"><img src="https://img.shields.io/github/stars/deepset-ai/haystack"></a> |カスタマイズ可能なプロダクション対応LLMアプリケーションを構築するためのAIオーケストレーションフレームワーク。|
+| [OmniParser](https://github.com/microsoft/OmniParser)<a href="https://github.com/microsoft/OmniParser"><img src="https://img.shields.io/github/stars/microsoft/OmniParser"></a> |純粋なビジョンベースのGUIエージェント向け画面解析ツール。|
+| [QAnything](https://github.com/netease-youdao/QAnything)<a href="https://github.com/netease-youdao/QAnything"><img src="https://img.shields.io/github/stars/netease-youdao/QAnything"></a> |あらゆるものに基づく質問応答。|
+| [その他のプロジェクトを見る](./awesome_projects.md) | [PaddleOCRに基づくその他のプロジェクト](./awesome_projects.md)|
+</div>
+
+## 👩‍👩‍👧‍👦 コントリビューター
+
+<div align="center">
+<a href="https://github.com/PaddlePaddle/PaddleOCR/graphs/contributors">
+  <img src="https://contrib.rocks/image?repo=PaddlePaddle/PaddleOCR&max=400&columns=20"  width="800"/>
+</a>
+</div>
+
+## 🌟 Star
+
+<div align="center">
+  <p>
+      <img width="800" src="https://api.star-history.com/svg?repos=PaddlePaddle/PaddleOCR&type=Date" alt="Star履歴">
+  </p>
+</div>
+
+
+## 📄 ライセンス
+このプロジェクトは[Apache 2.0ライセンス](LICENSE)の下で公開されています。
+
+## 🎓 引用
+
+```bibtex
+@misc{cui2025paddleocr30technicalreport,
+      title={PaddleOCR 3.0 Technical Report},
+      author={Cheng Cui and Ting Sun and Manhui Lin and Tingquan Gao and Yubo Zhang and Jiaxuan Liu and Xueqing Wang and Zelun Zhang and Changda Zhou and Hongen Liu and Yue Zhang and Wenyu Lv and Kui Huang and Yichao Zhang and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2507.05595},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2507.05595},
+}
+
+@misc{cui2025paddleocrvlboostingmultilingualdocument,
+      title={PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model},
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Handong Zheng and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2510.14528},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2510.14528},
+}
+
+@misc{cui2026paddleocrvl15multitask09bvlm,
+      title={PaddleOCR-VL-1.5: Towards a Multi-Task 0.9B VLM for Robust In-the-Wild Document Parsing},
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2026},
+      eprint={2601.21957},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2601.21957},
+}
+```
diff --git a/readme/README_ko.md b/readme/README_ko.md
new file mode 100644
index 0000000..9c5025c
--- /dev/null
+++ b/readme/README_ko.md
@@ -0,0 +1,304 @@
+
+<div align="center">
+  <p>
+      <img width="800" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Banner.png" alt="스타 히스토리">
+  </p>
+
+
+
+<h3>세계 최고의 OCR 툴킷 & 문서 AI 엔진</h3>
+
+[English](../README.md) | [简体中文](./README_cn.md) | [繁體中文](./README_tcn.md) | [日本語](./README_ja.md) | 한국어 | [Français](./README_fr.md) | [Русский](./README_ru.md) | [Español](./README_es.md) | [العربية](./README_ar.md)
+
+<!-- icon -->
+
+[![PyPI Downloads](https://static.pepy.tech/badge/paddleocr)](https://pepy.tech/projects/paddleocr)
+[![Used by](https://img.shields.io/badge/Used%20by-6k%2B%20repositories-blue)](https://github.com/PaddlePaddle/PaddleOCR/network/dependents)
+![python](https://img.shields.io/badge/python-3.8~3.12-aff.svg)
+![os](https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-pink.svg)
+![hardware](https://img.shields.io/badge/hardware-cpu%2C%20gpu%2C%20xpu%2C%20npu-yellow.svg)
+
+[![AI Studio](https://img.shields.io/badge/PaddleOCR-_Offiical_Website-1927BA?logo=data:image/png;base64,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&labelColor=white)](https://www.paddleocr.com)
+[![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/PaddlePaddle/PaddleOCR)
+[![License](https://img.shields.io/badge/license-Apache_2.0-green)](../LICENSE)
+
+</div>
+
+
+
+
+
+
+**PaddleOCR는 문서와 이미지를 업계 최고 수준의 정확도로 구조화된 LLM 지원 데이터(JSON/Markdown)로 변환합니다. 70,000개 이상의 Star와 Dify, RAGFlow, Cherry Studio 등 최상위 프로젝트의 신뢰를 받는 PaddleOCR는 지능형 RAG 및 에이전트 기반 애플리케이션 구축의 핵심 기반입니다.**
+
+
+## 🚀 주요 기능
+
+### 📄 지능형 문서 파싱 (LLM 지원)
+> *복잡한 시각 자료를 LLM 시대에 맞는 구조화된 데이터로 변환합니다.*
+
+* **최첨단 문서 VLM**: 문서 파싱을 위한 업계 최고의 경량 비전-언어 모델인 **PaddleOCR-VL-1.6 (0.9B)**를 탑재하였습니다. OmniDocBench v1.6에서 96.3% 정확도를 달성했으며, 텍스트·수식·표 인식에서 업계를 선도합니다. 고문서, 희귀 문자, 인장, 차트 등 다양한 시나리오에서의 성능도 크게 향상되었으며, **Markdown** 및 **JSON** 형식의 구조화된 출력을 지원합니다.
+* **구조 인식 변환**: **PP-StructureV3**를 기반으로 복잡한 PDF와 이미지를 **Markdown** 또는 **JSON**으로 원활하게 변환합니다. PaddleOCR-VL 시리즈 모델과 달리 표 셀 좌표, 텍스트 좌표 등 더욱 세밀한 좌표 정보를 제공합니다.
+* **상용 수준의 효율성**: 초소형 모델로 상용 등급의 정확도를 달성합니다. 공개 벤치마크에서 다수의 비공개 솔루션을 능가하면서도 엣지/클라우드 배포에 적합한 자원 효율성을 유지합니다.
+
+### 🔍 범용 텍스트 인식 (장면 OCR)
+> *고속 다국어 텍스트 탐지의 글로벌 표준.*
+
+* **100개 이상의 언어 지원**: 방대한 글로벌 언어 라이브러리를 기본 지원합니다. **PP-OCRv6**는 단일 모델로 50개 언어(중국어, 영어, 일본어 및 46개 라틴 언어)를 통합 지원합니다.
+* **복잡한 요소 처리**: 표준 텍스트 인식을 넘어 신분증, 거리 풍경, 도서, 산업 부품 등 다양한 환경에서의 **자연 장면 텍스트 탐지**를 지원합니다.
+* **성능 도약**: PP-OCRv6는 PP-OCRv5 대비 검출 정확도 **+4.6%**, 인식 정확도 **+5.1%** 향상을 달성하며, 주류 비전 언어 모델을 능가합니다. CPU 추론 5.2배 가속.
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Arch.jpg" alt="PaddleOCR 아키텍처">
+  </p>
+</div>
+
+### 🛠️ 개발자 중심 생태계
+* **원활한 통합**: AI 에이전트 생태계의 최고 선택 - **Dify, RAGFlow, Pathway, Cherry Studio**와 깊이 통합되어 있습니다.
+* **LLM 데이터 플라이휠**: 고품질 데이터셋 구축을 위한 완전한 파이프라인으로, 대규모 언어 모델 파인튜닝을 위한 지속 가능한 "데이터 엔진"을 제공합니다.
+* **원클릭 배포**: 다양한 하드웨어 백엔드(NVIDIA GPU, Intel CPU, Kunlunxin XPU 및 다양한 AI 가속기)를 지원합니다.
+
+
+## 📣 최근 업데이트
+
+### 🔥 2026.06.11: PaddleOCR 3.7.0 출시
+- PP-OCRv6 주요 특징:
+
+    - **정확도 향상**: medium 티어가 PP-OCRv5_server 대비 검출 +4.6%, 인식 +5.1% 달성, 34.5M 파라미터만으로 주류 VLM(Qwen3-VL-235B, GPT-5.5)을 능가.
+    - **50개 언어 통합 지원**: 단일 모델로 중국어, 영어, 일본어 및 46개 라틴 언어를 지원 — 모델 전환 불필요.
+    - **전문 시나리오 강화**: 디지털 디스플레이, 도트 매트릭스 문자, 타이어 인쇄, 산업용 문자 인식 대폭 향상.
+    - **빠른 추론**: CPU 5.2배 가속(OpenVINO), Apple M4 6.1배(tiny), A100 GPU 0.13초.
+    - **전 시나리오 3단계 모델**: tiny(1.5M) / small(7.7M) / medium(34.5M)으로 엣지부터 서버까지 대응.
+    - **모델 다운로드**: 모든 모델은 [HuggingFace](https://huggingface.co/collections/PaddlePaddle/pp-ocrv6) 및 [ModelScope](https://www.modelscope.cn/collections/PaddlePaddle/PP-OCRv6)에서 제공됩니다.
+
+<details>
+<summary><strong>2026.05.28: PaddleOCR 3.6.0 출시</strong></summary>
+
+- PaddleOCR-VL-1.6 주요 특징:
+
+    - **새로운 SOTA 정확도**: OmniDocBench v1.6에서 96.3%, OmniDocBench v1.5 및 Real5-OmniDocBench에서도 새로운 SOTA.
+    - **종합 강화**: 테이블, 고문서, 희귀 문자 인식 대폭 향상.
+    - **원활한 마이그레이션**: PaddleOCR-VL-1.5와 완전 호환 아키텍처.
+    - **지금 사용해보기**: [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.6) 또는 [공식 사이트](https://www.paddleocr.com).
+
+</details>
+<details>
+<summary><strong>2026.04.21: PaddleOCR 3.5.0 출시</strong></summary>
+
+* **유연한 추론 백엔드 전환**: Paddle 정적 그래프, Paddle 동적 그래프, Transformers 사이를 원활하게 전환할 수 있습니다. Hugging Face 생태계에 깊이 통합되었으며, 주요 20개 모델이 Transformers를 추론 백엔드로 지원합니다.
+* **Office 문서를 Markdown으로 변환**: Word, Excel, PowerPoint 등 일반적인 문서 형식을 Markdown으로 변환할 수 있습니다.
+* **분석 결과 DOCX 내보내기**: `PaddleOCR-VL` 시리즈, `PP-StructureV3`, `PP-DocTranslation`이 이제 분석 결과를 DOCX로 내보내 Microsoft Word에서 편리하게 확인하고 편집할 수 있습니다.
+* **공식 브라우저 추론 SDK**: 공식 브라우저 추론 SDK `PaddleOCR.js`를 출시하여 브라우저에서 `PP-OCRv5`를 실행할 수 있습니다.
+
+</details>
+
+<details>
+<summary><strong>2026.01.29: PaddleOCR 3.4.0 출시</strong></summary>
+* PaddleOCR-VL-1.5 (최첨단 0.9B VLM): 문서 파싱을 위한 최신 플래그십 모델이 출시되었습니다!
+    * **OmniDocBench에서 94.5% 정확도**: 최상위 범용 대규모 모델 및 전문 문서 파서를 능가합니다.
+    * **실제 환경 강건성**: 비정형 형태 위치 지정을 위한 **PP-DocLayoutV3** 알고리즘을 최초 도입하여 *기울기, 왜곡, 스캔, 조명, 화면 촬영*이라는 5가지 까다로운 시나리오를 정복합니다.
+    * **기능 확장**: **인감 인식**, **텍스트 탐지**를 지원하며, **111개 언어**(중국 티베트 문자 및 벵골어 포함)로 확장되었습니다.
+    * **장문서 처리**: 자동 교차 페이지 표 병합 및 계층적 제목 식별을 지원합니다.
+    * **지금 사용해 보세요**: [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.5) 또는 [공식 웹사이트](https://www.paddleocr.com)에서 이용 가능합니다.
+
+</details>
+
+<details>
+<summary><strong>2025.10.16: PaddleOCR 3.3.0 출시</strong></summary>
+
+- PaddleOCR-VL 출시:
+    - **모델 소개**:
+        - **PaddleOCR-VL**은 문서 파싱에 특화된 최첨단 자원 효율적 모델입니다. 핵심 구성 요소인 PaddleOCR-VL-0.9B는 NaViT 스타일의 동적 해상도 비주얼 인코더와 ERNIE-4.5-0.3B 언어 모델을 통합하여 정확한 요소 인식을 가능하게 하는 컴팩트하면서도 강력한 비전-언어 모델(VLM)입니다. **이 혁신적인 모델은 109개 언어를 효율적으로 지원하며, 최소한의 자원 소비를 유지하면서 복잡한 요소(텍스트, 표, 수식, 차트 등) 인식에 탁월합니다**. 널리 사용되는 공개 벤치마크와 자체 벤치마크를 통한 종합 평가에서 PaddleOCR-VL은 페이지 수준 문서 파싱과 요소 수준 인식 모두에서 최첨단 성능을 달성합니다. 기존 솔루션을 크게 능가하고, 최상위 VLM과의 경쟁에서도 강한 경쟁력을 보이며, 빠른 추론 속도를 제공합니다. 이러한 강점은 실제 시나리오에서의 실용적 배포에 매우 적합합니다. 모델은 [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL)에 공개되었습니다. 누구나 다운로드하여 사용할 수 있습니다! 더 자세한 소개는 [PaddleOCR-VL](https://www.paddleocr.ai/latest/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL.html)에서 확인할 수 있습니다.
+
+    - **핵심 기능**:
+        - **컴팩트하면서도 강력한 VLM 아키텍처**: 자원 효율적 추론에 특화 설계된 새로운 비전-언어 모델을 제시하며, 요소 인식에서 탁월한 성능을 달성합니다. NaViT 스타일의 동적 고해상도 비주얼 인코더와 경량 ERNIE-4.5-0.3B 언어 모델을 통합하여 모델의 인식 능력과 디코딩 효율성을 크게 향상시킵니다. 이러한 통합은 높은 정확도를 유지하면서 연산 요구량을 줄여 효율적이고 실용적인 문서 처리 애플리케이션에 적합합니다.
+        - **문서 파싱 최첨단 성능**: PaddleOCR-VL은 페이지 수준 문서 파싱과 요소 수준 인식 모두에서 최첨단 성능을 달성합니다. 기존 파이프라인 기반 솔루션을 크게 능가하며, 문서 파싱에서 선도적인 비전-언어 모델(VLM)과 강한 경쟁력을 보입니다. 또한 텍스트, 표, 수식, 차트 등 복잡한 문서 요소 인식에 탁월하여 필기체 텍스트 및 역사 문서를 포함한 다양하고 까다로운 콘텐츠 유형에 적합합니다. 이로 인해 다양한 문서 유형과 시나리오에 폭넓게 활용할 수 있습니다.
+        - **다국어 지원**: PaddleOCR-VL은 109개 언어를 지원하며, 중국어, 영어, 일본어, 라틴어, 한국어를 포함한(이에 국한되지 않는) 주요 글로벌 언어와 러시아어(키릴 문자), 아랍어, 힌디어(데바나가리 문자), 태국어 등 다양한 문자 체계와 구조를 가진 언어를 포괄합니다. 이러한 광범위한 언어 지원은 다국어 및 글로벌 문서 처리 시나리오에서의 시스템 적용성을 크게 향상시킵니다.
+
+- PP-OCRv5 다국어 인식 모델 출시:
+    - 라틴 문자 인식의 정확도와 범위를 개선하였습니다. 키릴 문자, 아랍 문자, 데바나가리 문자, 텔루구 문자, 타밀 문자 및 기타 언어 체계 지원을 추가하여 109개 언어 인식을 지원합니다. 모델 파라미터는 2M에 불과하며, 일부 모델의 정확도는 이전 세대 대비 40% 이상 향상되었습니다.
+
+</details>
+
+
+<details>
+<summary><strong>2025.08.21: PaddleOCR 3.2.0 출시</strong></summary>
+
+- **주요 모델 추가:**
+    - PP-OCRv5 인식 모델의 영어, 태국어, 그리스어 학습, 추론 및 배포를 도입하였습니다. **PP-OCRv5 영어 모델은 기존 PP-OCRv5 메인 모델 대비 영어 시나리오에서 11% 향상을 달성했으며, 태국어 및 그리스어 인식 모델은 각각 82.68%와 89.28%의 정확도를 기록하였습니다.**
+
+- **배포 기능 업그레이드:**
+    - **PaddlePaddle 프레임워크 버전 3.1.0 및 3.1.1 완전 지원.**
+    - **PP-OCRv5 C++ 로컬 배포 솔루션 종합 업그레이드, Linux 및 Windows 모두 지원하며 Python 구현과 동일한 기능 및 정확도를 제공합니다.**
+    - **고성능 추론이 CUDA 12를 지원하며, Paddle Inference 또는 ONNX Runtime 백엔드를 사용하여 추론을 수행할 수 있습니다.**
+    - **고안정성 서비스 지향 배포 솔루션이 완전히 오픈소스화되어, 사용자가 필요에 따라 Docker 이미지와 SDK를 커스터마이징할 수 있습니다.**
+    - 고안정성 서비스 지향 배포 솔루션은 수동으로 구성한 HTTP 요청을 통한 호출도 지원하여, 모든 프로그래밍 언어에서 클라이언트 측 코드 개발이 가능합니다.
+
+- **벤치마크 지원:**
+    - **모든 프로덕션 라인이 세분화된 벤치마킹을 지원하여, 엔드투엔드 추론 시간은 물론 레이어별 및 모듈별 지연 시간 데이터를 측정하여 성능 분석을 지원합니다. 벤치마크 기능의 설정 및 사용 방법은 [여기](docs/version3.x/pipeline_usage/instructions/benchmark.en.md)를 참조하세요.**
+    - **문서에 주류 하드웨어에서의 일반 구성에 대한 주요 지표(추론 지연 시간 및 메모리 사용량 등)가 업데이트되어, 사용자에게 배포 참고 자료를 제공합니다.**
+
+- **버그 수정:**
+    - 모델 학습 중 로그 저장 실패 문제를 해결하였습니다.
+    - 수식 모델의 데이터 증강 구성 요소를 업그레이드하여 최신 버전의 albumentations 종속성과의 호환성을 확보하고, 멀티프로세스 시나리오에서 tokenizers 패키지 사용 시 발생하는 데드락 경고를 수정하였습니다.
+    - PP-StructureV3 구성 파일에서 스위치 동작(예: `use_chart_parsing`)이 다른 파이프라인과 불일치하는 문제를 수정하였습니다.
+
+- **기타 개선 사항:**
+    - **핵심 종속성과 선택적 종속성을 분리하였습니다. 기본 텍스트 인식에는 최소한의 핵심 종속성만 필요하며, 문서 파싱 및 정보 추출을 위한 추가 종속성은 필요에 따라 설치할 수 있습니다.**
+    - **Windows에서 NVIDIA RTX 50 시리즈 그래픽 카드 지원을 활성화하였습니다. 해당 PaddlePaddle 프레임워크 버전은 [설치 가이드](docs/version3.x/installation.en.md)를 참조하세요.**
+    - **PP-OCR 시리즈 모델이 단일 문자 좌표 반환을 지원합니다.**
+    - AIStudio, ModelScope 등 모델 다운로드 소스를 추가하여, 사용자가 모델 다운로드 소스를 지정할 수 있습니다.
+    - PP-Chart2Table 모듈을 통한 차트-표 변환 지원을 추가하였습니다.
+    - 사용성 향상을 위해 문서 설명을 최적화하였습니다.
+</details>
+
+
+[변경 이력](https://paddlepaddle.github.io/PaddleOCR/latest/en/update/update.html)
+
+
+## 🚀 빠른 시작
+
+### 1단계: 온라인 체험
+PaddleOCR 공식 웹사이트에서는 별도 설정 없이 클릭 한 번으로 체험할 수 있는 인터랙티브 **체험 센터**와 **API**를 제공합니다.
+
+👉 [공식 웹사이트 방문](https://www.paddleocr.com)
+
+### 2단계: 로컬 배포
+로컬 사용을 위해 필요에 따라 다음 문서를 참조하세요:
+
+- **PP-OCR 시리즈**: [PP-OCR 문서](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/OCR.html) 참조
+- **PaddleOCR-VL 시리즈**: [PaddleOCR-VL 문서](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PaddleOCR-VL.html) 참조
+- **PP-StructureV3**: [PP-StructureV3 문서](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PP-StructureV3.html) 참조
+- **추가 기능**: [추가 기능 문서](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/pipeline_overview.html) 참조
+
+
+## 🧩 추가 기능
+
+- 모델을 ONNX 형식으로 변환: [ONNX 모델 획득](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/others/obtaining_onnx_models.html).
+- OpenVINO, ONNX Runtime, TensorRT 등의 엔진을 사용한 추론 가속 또는 ONNX 형식 모델을 사용한 추론: [고성능 추론](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/local_inference/high_performance_inference.html).
+- 다중 GPU 및 다중 프로세스를 사용한 추론 가속: [파이프라인 병렬 추론](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/pipeline_usage/instructions/parallel_inference.html).
+- PaddleOCR를 C++, C#, Java 등으로 작성된 애플리케이션에 통합: [서빙](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/serving/serving.html).
+
+## 🔄 실행 결과 빠른 미리보기
+
+### PP-OCRv5
+
+<div align="center">
+  <p>
+       <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-OCRv5_demo.gif" alt="PP-OCRv5 데모">
+  </p>
+</div>
+
+
+
+### PP-StructureV3
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-StructureV3_demo.gif" alt="PP-StructureV3 데모">
+  </p>
+</div>
+
+### PaddleOCR-VL
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PaddleOCR-VL_demo.gif" alt="PaddleOCR-VL 데모">
+  </p>
+</div>
+
+
+## ✨ 소식을 받아보세요
+
+⭐ **이 저장소에 Star를 눌러 강력한 OCR 및 문서 파싱 기능을 포함한 흥미로운 업데이트와 새로운 릴리스를 확인하세요!** ⭐
+
+<div align="center">
+  <p>
+       <img width="1200" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/star_paddleocr2.en.gif" alt="프로젝트에 Star 누르기">
+  </p>
+</div>
+
+
+## 👩‍👩‍👧‍👦 커뮤니티
+
+<div align="center">
+
+| PaddlePaddle 위챗 공식 계정 | 기술 토론 그룹 참여 |
+| :---: | :---: |
+| <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qrcode_for_paddlepaddle_official_account.jpg" width="150"> | <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qr_code_for_the_questionnaire.jpg" width="150"> |
+</div>
+
+
+## 😃 PaddleOCR를 활용한 멋진 프로젝트
+PaddleOCR가 오늘날의 모습을 갖추기까지 놀라운 커뮤니티의 힘이 있었습니다! 💗 오랜 파트너, 새로운 협력자, 그리고 PaddleOCR에 열정을 쏟아주신 모든 분들께 진심으로 감사드립니다 - 여기에 이름이 언급되지 않은 분들까지 포함하여. 여러분의 지원이 우리의 원동력입니다!
+
+<div align="center">
+
+| 프로젝트 이름 | 설명 |
+| ------------ | ----------- |
+| [Dify](https://github.com/langgenius/dify) <a href="https://github.com/langgenius/dify"><img src="https://img.shields.io/github/stars/langgenius/dify"></a>|에이전트 워크플로우 개발을 위한 프로덕션 지원 플랫폼.|
+| [RAGFlow](https://github.com/infiniflow/ragflow) <a href="https://github.com/infiniflow/ragflow"><img src="https://img.shields.io/github/stars/infiniflow/ragflow"></a>|심층 문서 이해 기반 RAG 엔진.|
+| [pathway](https://github.com/pathwaycom/pathway) <a href="https://github.com/pathwaycom/pathway"><img src="https://img.shields.io/github/stars/pathwaycom/pathway"></a>|스트림 처리, 실시간 분석, LLM 파이프라인 및 RAG를 위한 Python ETL 프레임워크.|
+| [MinerU](https://github.com/opendatalab/MinerU) <a href="https://github.com/opendatalab/MinerU"><img src="https://img.shields.io/github/stars/opendatalab/MinerU"></a>|다양한 유형의 문서를 Markdown으로 변환하는 도구.|
+| [Umi-OCR](https://github.com/hiroi-sora/Umi-OCR) <a href="https://github.com/hiroi-sora/Umi-OCR"><img src="https://img.shields.io/github/stars/hiroi-sora/Umi-OCR"></a>|무료, 오픈소스, 일괄 오프라인 OCR 소프트웨어.|
+| [cherry-studio](https://github.com/CherryHQ/cherry-studio) <a href="https://github.com/CherryHQ/cherry-studio"><img src="https://img.shields.io/github/stars/CherryHQ/cherry-studio"></a>|여러 LLM 제공업체를 지원하는 데스크톱 클라이언트.|
+| [haystack](https://github.com/deepset-ai/haystack)<a href="https://github.com/deepset-ai/haystack"><img src="https://img.shields.io/github/stars/deepset-ai/haystack"></a> |커스터마이징 가능한 프로덕션 지원 LLM 애플리케이션 구축을 위한 AI 오케스트레이션 프레임워크.|
+| [OmniParser](https://github.com/microsoft/OmniParser)<a href="https://github.com/microsoft/OmniParser"><img src="https://img.shields.io/github/stars/microsoft/OmniParser"></a> |순수 비전 기반 GUI 에이전트를 위한 화면 파싱 도구.|
+| [QAnything](https://github.com/netease-youdao/QAnything)<a href="https://github.com/netease-youdao/QAnything"><img src="https://img.shields.io/github/stars/netease-youdao/QAnything"></a> |모든 것에 기반한 질의응답.|
+| [더 많은 프로젝트 보기](./awesome_projects.md) | [PaddleOCR 기반 추가 프로젝트](./awesome_projects.md)|
+</div>
+
+## 👩‍👩‍👧‍👦 기여자
+
+<div align="center">
+<a href="https://github.com/PaddlePaddle/PaddleOCR/graphs/contributors">
+  <img src="https://contrib.rocks/image?repo=PaddlePaddle/PaddleOCR&max=400&columns=20"  width="800"/>
+</a>
+</div>
+
+## 🌟 Star
+
+<div align="center">
+  <p>
+      <img width="800" src="https://api.star-history.com/svg?repos=PaddlePaddle/PaddleOCR&type=Date" alt="스타 히스토리">
+  </p>
+</div>
+
+
+## 📄 라이선스
+이 프로젝트는 [Apache 2.0 라이선스](LICENSE)로 배포됩니다.
+
+## 🎓 인용
+
+```bibtex
+@misc{cui2025paddleocr30technicalreport,
+      title={PaddleOCR 3.0 Technical Report},
+      author={Cheng Cui and Ting Sun and Manhui Lin and Tingquan Gao and Yubo Zhang and Jiaxuan Liu and Xueqing Wang and Zelun Zhang and Changda Zhou and Hongen Liu and Yue Zhang and Wenyu Lv and Kui Huang and Yichao Zhang and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2507.05595},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2507.05595},
+}
+
+@misc{cui2025paddleocrvlboostingmultilingualdocument,
+      title={PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model},
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Handong Zheng and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2510.14528},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2510.14528},
+}
+
+@misc{cui2026paddleocrvl15multitask09bvlm,
+      title={PaddleOCR-VL-1.5: Towards a Multi-Task 0.9B VLM for Robust In-the-Wild Document Parsing},
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2026},
+      eprint={2601.21957},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2601.21957},
+}
+```
diff --git a/readme/README_ru.md b/readme/README_ru.md
new file mode 100644
index 0000000..b53f168
--- /dev/null
+++ b/readme/README_ru.md
@@ -0,0 +1,303 @@
+<div align="center">
+  <p>
+      <img width="800" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Banner.png" alt="Star-history">
+  </p>
+
+
+
+<h3>Ведущий в мире инструментарий OCR и движок Document AI</h3>
+
+[English](../README.md) | [简体中文](./README_cn.md) | [繁體中文](./README_tcn.md) | [日本語](./README_ja.md) | [한국어](./README_ko.md) | [Français](./README_fr.md) | Русский | [Español](./README_es.md) | [العربية](./README_ar.md)
+
+<!-- icon -->
+
+[![PyPI Downloads](https://static.pepy.tech/badge/paddleocr)](https://pepy.tech/projects/paddleocr)
+[![Used by](https://img.shields.io/badge/Used%20by-6k%2B%20repositories-blue)](https://github.com/PaddlePaddle/PaddleOCR/network/dependents)
+![python](https://img.shields.io/badge/python-3.8~3.12-aff.svg)
+![os](https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-pink.svg)
+![hardware](https://img.shields.io/badge/hardware-cpu%2C%20gpu%2C%20xpu%2C%20npu-yellow.svg)
+
+[![AI Studio](https://img.shields.io/badge/PaddleOCR-_Offiical_Website-1927BA?logo=data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAMAAADDpiTIAAAABlBMVEU2P+X///+1KuUwAAAHKklEQVR42u3dS5bjOAwEwALvf2fMavZum6IAImI7b2yYSqU+1Zb//gAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADKCR/+fzly7rD92yVg69xh8zeLwOa5w+ZvFYHtc4ft3ykB++cOm79PAp6YO2z/Ngl4ZO5l+9+yT4QAvLqS748VF33Ylzdvzpl72f6z53YIGJ6SZdPeNHcIwOycaADdLgCSIgAIgCOAACAAykIAEAAEAAFAABCAT+WQuQVgeBqXhXQIQAAYegowLQBpbg3gZGFyAC6vgBQAMREA2/YfDPxyaDQNyTNz+3Zwn5J4ZG7PB2h0kHhi7plPCImmJwkPzO0RMa3OET0i5uGlzHFze0xcu0vE2Dq3J4U2vEPgSaHbFzPNDQAAAAAAAMBNovdw+cP/ny+uaf7w/+eYADy8kE+F4Offdjn6zZXhAXgiA78G4MNNsmnu1Xr7b3mbOL8T5Ja5bw/A35EC2LiWpzt1y9jRugBy30fLg3NvHPvnuZcC2NsCUXA/aRmA89V07Fwgt37uH8deCmBr6N44pP4UgaUATpdA7v/cMbIB8okliY65/SW5HhJ1ehPmM+8edwXgpbu4R88FayR32Y/P7oZZbOx13/Zr//ZHx27bAPnkFoyewYlbAhD3TvBobr95gaUAtr1EdNx1lgI4OcTTuR3z6+FZMEDRcu9ZCuDgGCdyGxMa4EgBRMvcjrkM7NgBZw5c0TwAUWUhZwRXA2xaya65Xa3jO2qYZ8bu2AD5w38tG5V8aZpoGN6Tz0bOfa9bceyWAciTO0jWyO1Tc5cLwJmF/JfPnXVyu3/slgHIg1n79O2O5fZv+1cHV7sC2HYqmUdHysNzX3sVkMcjUK5Gc+dMs28E5bGtm0V3gloBOP9vgZv+4sYn3RUaYFMCol5uN77g6lUApc8pWs69Zn7snS9Z9Q8G0S0AUTVUUTG3A54R1KSvo/diLAv5fKzynZeN6xogC75u93+AtBTA47OlAFSv6qY/vp3DAjD8iv2ZdFYJwKynMhTK1rInPfzaxW81LnvSgFP9KxrATaCLA3DxHpbFX31ZyNm5XRZyXG5bNkAWfP0rcrsUwOgC6NIAzgBcBiqAWwPgLrAGuGBP6jr2sifdfiJ6QQM4Bbw4AK4B3129ZSFn53ZZyA/GyFty27IBFMDFAXAG8PbyLQv5xULGPRl0K3h2AbwcgCZPhs+LD1zLnjS6AN4NwMU/DVFh7LyhASreTbvqrxdr/J4XT4Swz4FrTS+AGJ7bNbwAYkxuWzZAVljHrJfbjb9wviYXwFO/FJ8Vli4vaICsEMFyBbA3tmtsAUS0zG1c/bj4YwsZH2/+Whd0+1Nb+S7IE2sfPw4RL0XmsR8Nqvz7qFngmPHF34EqjP15AAofAkosZKPC/K6FVoeP02Ehi540NG6AK/4pYP3cLgVwXwHkDQ1QcSGb/uF4WwCmfX8u/+4vgLINcMUlQIfcLgXwXAF0+BGkpQDuuJx7/hwgpu//cWVuO3wxJOz/z8297vgYBwaIO3O7Kn+c194578ltywbIgu8fl+Z2lS+APvnLjnOv8hsgSqxjgwL4Ln9LAezaj98tgPzy7ZcC+GQzxrWxXQpgx370dm6/H7v6jaBoso5dY1swAFlwHWvfBf5pxVa93fCtdx64+1dsgCy4joWvAfPX9VoKYMs6Zse9/8Mlvv7LILlhAfKFFdsSutJXAdFkL3qlADJPrXFcXAC5KYaH586jO9mtAch9S3T0GQJ726ZWAE49kjP3rlDJuetdaL/1zeqZY9c7CRz7s0wCUPxienQBnAuAAtAAlxaAAAxfyBQABSAACkAAFIAAKAABUAACMEkKwL170oh7V8ueNLoAjgTAXWAN4BRwcABcA2oABTA4AApAAyiAwQFQABpAAQwOgALQADMWUgCuEmNyu15fSIY3gFPAiwPgFFADKIDBAVAAGkABCIACmBqAUAAaQAHMDUCMWkgBuMWw3K43F5LhDeAU8OIAuAmkARTA4AAoAA2gAARAAUwNgLvAGkABDA6Au8AaoKOJuV0vLSTDG8Ap4MUBcBNIAyiAwQFQABpAAQwOgALQAApAABTA1AC4C6wBOhqb23V+IRneAE4BLw6Aa0ANoAAGB0ABaAAFMDgACkADKAABUABTA+AusAboKATAQs4trjV+IYcfuJYCcA6gAATAQk69dFkKQANYyLkFcLIBFIDLQAVwawDsSRrAEWBwAJwCagAFMDgACkADKIDBAVAAGkABCIACmBoAzwXWAApgcADsSRrg0iNACoACEADXgAIwdCFTACykALgGFIAfl0kBAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAPBv/gN+IH8U6YveYgAAAABJRU5ErkJggg==&labelColor=white)](https://www.paddleocr.com)
+[![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/PaddlePaddle/PaddleOCR)
+[![License](https://img.shields.io/badge/license-Apache_2.0-green)](../LICENSE)
+
+</div>
+
+
+
+
+
+
+**PaddleOCR преобразует документы и изображения в структурированные данные, готовые для использования с LLM (JSON/Markdown), с точностью мирового уровня. Имея более 70 тысяч звёзд и доверие таких ведущих проектов, как Dify, RAGFlow и Cherry Studio, PaddleOCR является основой для создания интеллектуальных приложений RAG и Agentic.**
+
+
+## 🚀 Ключевые возможности
+
+### 📄 Интеллектуальный разбор документов (готово для LLM)
+> *Преобразование сложных визуальных данных в структурированные данные для эпохи LLM.*
+
+* **SOTA Document VLM**: Featuring **PaddleOCR-VL-1.6 (0.9B)**, the industry's leading lightweight vision-language model for document parsing. It achieves 96.3% accuracy on OmniDocBench v1.6, leads in text, formula, and table recognition, and shows significantly enhanced capabilities in ancient documents, rare characters, seals, and charts, with structured outputs in **Markdown** and **JSON** formats.
+* **Конвертация с учётом структуры**: На основе **PP-StructureV3** — бесшовное преобразование сложных PDF-файлов и изображений в **Markdown** или **JSON**. В отличие от моделей серии PaddleOCR-VL, предоставляет более детальную координатную информацию, включая координаты ячеек таблиц, координаты текста и многое другое.
+* **Эффективность промышленного уровня**: Коммерческая точность при минимальном объёме ресурсов. Превосходит многочисленные закрытые решения в публичных тестах, оставаясь ресурсоэффективным для развёртывания на периферийных устройствах и в облаке.
+
+### 🔍 Универсальное распознавание текста (Scene OCR)
+> *Мировой золотой стандарт высокоскоростного многоязычного обнаружения текста.*
+
+* **Поддержка 100+ языков**: Нативное распознавание обширной глобальной библиотеки. **PP-OCRv6** поддерживает 50 языков единой моделью (китайский, английский, японский и 46 латинских языков) — без переключения моделей.
+* **Мастерство работы со сложными элементами**: Помимо стандартного распознавания текста, поддерживается **обнаружение текста в естественных сценах** в широком диапазоне условий, включая удостоверения личности, уличные виды, книги и промышленные компоненты.
+* **Скачок производительности**: PP-OCRv6 достигает **+4.6% детекции** и **+5.1% распознавания** по сравнению с PP-OCRv5, превосходя ведущие визуально-языковые модели. Ускорение CPU-инференса в 5.2×.
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Arch.jpg" alt="PaddleOCR Architecture">
+  </p>
+</div>
+
+### 🛠️ Экосистема, ориентированная на разработчиков
+* **Бесшовная интеграция**: Первый выбор для экосистемы AI Agent — глубокая интеграция с **Dify, RAGFlow, Pathway и Cherry Studio**.
+* **Маховик данных для LLM**: Полный конвейер для создания высококачественных наборов данных, обеспечивающий устойчивый «Data Engine» для тонкой настройки больших языковых моделей.
+* **Развёртывание в один клик**: Поддержка различных аппаратных бэкендов (NVIDIA GPU, Intel CPU, Kunlunxin XPU и разнообразные AI-ускорители).
+
+
+## 📣 Последние обновления
+
+### 🔥 2026.06.11: Выпуск PaddleOCR 3.7.0
+- Ключевые особенности PP-OCRv6:
+
+    - **Повышение точности**: уровень medium достигает +4.6% детекции и +5.1% распознавания по сравнению с PP-OCRv5_server, превосходя ведущие VLM (Qwen3-VL-235B, GPT-5.5) с всего 34.5M параметрами.
+    - **50 языков в одной модели**: единая модель покрывает китайский, английский, японский и 46 латинских языков — без переключения моделей.
+    - **Специализированные сценарии**: значительные улучшения в распознавании цифровых дисплеев, матричных символов, шинных отпечатков и промышленного текста.
+    - **Быстрый инференс**: ускорение CPU 5.2× (OpenVINO), 6.1× на Apple M4 (tiny), 0.13с на A100 GPU.
+    - **Три уровня для всех сценариев**: tiny (1.5M) / small (7.7M) / medium (34.5M) для периферии, мобильных устройств и серверов.
+    - **Доступность моделей**: Все модели доступны на [HuggingFace](https://huggingface.co/collections/PaddlePaddle/pp-ocrv6) и [ModelScope](https://www.modelscope.cn/collections/PaddlePaddle/PP-OCRv6).
+
+<details>
+<summary><strong>2026.05.28: Выпуск PaddleOCR 3.6.0</strong></summary>
+
+- Ключевые особенности PaddleOCR-VL-1.6:
+
+    - **Новая точность SOTA**: 96.3% на OmniDocBench v1.6, новый SOTA на OmniDocBench v1.5 и Real5-OmniDocBench.
+    - **Всестороннее улучшение**: значительные улучшения в таблицах, древних документах и редких символах.
+    - **Бесшовная миграция**: архитектура полностью совместима с PaddleOCR-VL-1.5.
+    - **Попробуйте**: доступно на [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.6) или [официальном сайте](https://www.paddleocr.com).
+
+</details>
+<details>
+<summary><strong>2026.04.21: Выпуск PaddleOCR 3.5.0</strong></summary>
+
+* **Гибкое переключение бэкендов инференса**: поддерживается бесшовное переключение между статическим графом Paddle, динамическим графом Paddle и Transformers. PaddleOCR теперь глубоко интегрирован с экосистемой Hugging Face, а 20 ключевых моделей поддерживают Transformers в качестве бэкенда инференса.
+* **Преобразование офисных документов в Markdown**: поддерживается преобразование распространённых форматов документов, таких как Word, Excel и PowerPoint, в Markdown.
+* **Экспорт результатов разбора в DOCX**: серии `PaddleOCR-VL`, `PP-StructureV3` и `PP-DocTranslation` теперь поддерживают экспорт результатов разбора в формат DOCX для удобного просмотра и редактирования в Microsoft Word.
+* **Официальный браузерный SDK инференса**: выпущен официальный браузерный SDK инференса `PaddleOCR.js`, который поддерживает запуск `PP-OCRv5` прямо в браузере.
+
+</details>
+
+<details>
+<summary><strong>2026.01.29: Выпуск PaddleOCR 3.4.0</strong></summary>
+* PaddleOCR-VL-1.5 (SOTA 0.9B VLM): Наша новейшая флагманская модель для разбора документов уже доступна!
+    * **94,5% точность на OmniDocBench**: Превосходит ведущие универсальные большие модели и специализированные парсеры документов.
+    * **Устойчивость к реальным условиям**: Первая реализация алгоритма **PP-DocLayoutV3** для позиционирования нестандартных форм, освоившая 5 сложных сценариев: *перекос, деформация, сканирование, неравномерное освещение и фотосъёмка экрана*.
+    * **Расширение возможностей**: Теперь поддерживается **распознавание печатей**, **обнаружение текста** и расширение до **111 языков** (включая тибетское письмо Китая и бенгальский язык).
+    * **Работа с длинными документами**: Поддержка автоматического объединения таблиц на нескольких страницах и иерархической идентификации заголовков.
+    * **Попробуйте сейчас**: Доступно на [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.5) или на нашем [официальном сайте](https://www.paddleocr.com).
+
+</details>
+
+<details>
+<summary><strong>2025.10.16: Выпуск PaddleOCR 3.3.0</strong></summary>
+
+- Выпуск PaddleOCR-VL:
+    - **Описание модели**:
+        - **PaddleOCR-VL** — это SOTA-модель с эффективным использованием ресурсов, разработанная специально для разбора документов. Её ключевым компонентом является PaddleOCR-VL-0.9B — компактная, но мощная визуально-языковая модель (VLM), объединяющая динамический визуальный энкодер с переменным разрешением в стиле NaViT с языковой моделью ERNIE-4.5-0.3B для точного распознавания элементов. **Эта инновационная модель эффективно поддерживает 109 языков и превосходно справляется с распознаванием сложных элементов (например, текста, таблиц, формул и диаграмм), сохраняя минимальное потребление ресурсов**. По результатам комплексных оценок на широко используемых публичных тестах и внутренних тестах PaddleOCR-VL достигает SOTA-производительности как в разборе документов на уровне страниц, так и в распознавании элементов. Она значительно превосходит существующие решения, демонстрирует высокую конкурентоспособность по сравнению с ведущими VLM и обеспечивает высокую скорость вывода. Эти преимущества делают её высокопригодной для практического развёртывания в реальных сценариях. Модель опубликована на [HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL). Приглашаем всех скачать и использовать! Дополнительная информация доступна в разделе [PaddleOCR-VL](https://www.paddleocr.ai/latest/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL.html).
+
+    - **Основные возможности**:
+        - **Компактная, но мощная архитектура VLM**: Представлена новая визуально-языковая модель, специально разработанная для ресурсоэффективного вывода, достигающая выдающейся производительности в распознавании элементов. Благодаря интеграции динамического высокоразрешающего визуального энкодера в стиле NaViT с лёгкой языковой моделью ERNIE-4.5-0.3B мы значительно повысили возможности распознавания и эффективность декодирования модели. Эта интеграция сохраняет высокую точность при снижении вычислительных требований, что делает её хорошо подходящей для эффективной и практической обработки документов.
+        - **SOTA-производительность в разборе документов**: PaddleOCR-VL достигает передовой производительности как в разборе документов на уровне страниц, так и в распознавании элементов. Она значительно превосходит существующие конвейерные решения и демонстрирует высокую конкурентоспособность по сравнению с ведущими визуально-языковыми моделями (VLM) в разборе документов. Кроме того, она превосходно справляется с распознаванием сложных элементов документов, таких как текст, таблицы, формулы и диаграммы, что делает её пригодной для широкого спектра сложных типов контента, включая рукописный текст и исторические документы. Это делает её высоко универсальной и подходящей для широкого спектра типов и сценариев документов.
+        - **Многоязычная поддержка**: PaddleOCR-VL поддерживает 109 языков, охватывая основные мировые языки, включая, но не ограничиваясь китайским, английским, японским, латинским и корейским, а также языки с различными системами письма и структурами, такие как русский (кириллица), арабский, хинди (письмо деванагари) и тайский. Широкий охват языков существенно повышает применимость нашей системы к многоязычным и глобализированным сценариям обработки документов.
+
+- Выпуск PP-OCRv5 — многоязычной модели распознавания:
+    - Улучшена точность и охват распознавания латинского письма; добавлена поддержка кириллицы, арабского, деванагари, телугу, тамильского и других языковых систем, охватывающих распознавание 109 языков. Модель имеет всего 2 МБ параметров, а точность некоторых моделей выросла более чем на 40% по сравнению с предыдущим поколением.
+
+</details>
+
+
+<details>
+<summary><strong>2025.08.21: Выпуск PaddleOCR 3.2.0</strong></summary>
+
+- **Значительные дополнения моделей:**
+    - Введены обучение, вывод и развёртывание моделей распознавания PP-OCRv5 для английского, тайского и греческого языков. **Модель PP-OCRv5 для английского языка обеспечивает улучшение на 11% в английских сценариях по сравнению с основной моделью PP-OCRv5, при этом модели распознавания тайского и греческого языков достигают точности 82,68% и 89,28% соответственно.**
+
+- **Улучшения возможностей развёртывания:**
+    - **Полная поддержка версий фреймворка PaddlePaddle 3.1.0 и 3.1.1.**
+    - **Комплексное обновление решения для локального развёртывания PP-OCRv5 на C++, теперь поддерживающего как Linux, так и Windows, с полным соответствием функций и идентичной точностью реализации на Python.**
+    - **Высокопроизводительный вывод теперь поддерживает CUDA 12, а вывод может выполняться с использованием бэкендов Paddle Inference или ONNX Runtime.**
+    - **Решение для высоконадёжного сервисного развёртывания теперь полностью открыто, позволяя пользователям при необходимости настраивать образы Docker и SDK.**
+    - Решение для высоконадёжного сервисного развёртывания также поддерживает вызов через вручную сформированные HTTP-запросы, что позволяет разрабатывать клиентский код на любом языке программирования.
+
+- **Поддержка бенчмарков:**
+    - **Все производственные конвейеры теперь поддерживают детализированное бенчмаркирование, позволяя измерять сквозное время вывода, а также задержки на уровне отдельных слоёв и модулей для анализа производительности. [Здесь](docs/version3.x/pipeline_usage/instructions/benchmark.en.md) описано, как настроить и использовать функцию бенчмарка.**
+    - **Документация обновлена и включает ключевые метрики для часто используемых конфигураций на основном оборудовании, такие как задержка вывода и использование памяти, предоставляя справочные данные для развёртывания.**
+
+- **Исправления ошибок:**
+    - Устранена проблема с неудачным сохранением журналов во время обучения модели.
+    - Обновлён компонент аугментации данных для моделей формул для совместимости с более новыми версиями зависимости albumentations, а также исправлены предупреждения о взаимоблокировке при использовании пакета tokenizers в многопроцессорных сценариях.
+    - Исправлены несоответствия в поведении переключателей (например, `use_chart_parsing`) в файлах конфигурации PP-StructureV3 по сравнению с другими конвейерами.
+
+- **Прочие улучшения:**
+    - **Разделены основные и дополнительные зависимости. Для базового распознавания текста требуются только минимальные основные зависимости; дополнительные зависимости для разбора документов и извлечения информации могут быть установлены по мере необходимости.**
+    - **Включена поддержка видеокарт NVIDIA RTX серии 50 на Windows; пользователи могут обратиться к [руководству по установке](docs/version3.x/installation.en.md) для получения информации о соответствующих версиях фреймворка PaddlePaddle.**
+    - **Модели серии PP-OCR теперь поддерживают возврат координат отдельных символов.**
+    - Добавлены источники загрузки моделей AIStudio, ModelScope и другие, позволяющие пользователям указывать источник для загрузки моделей.
+    - Добавлена поддержка преобразования диаграмм в таблицы через модуль PP-Chart2Table.
+    - Оптимизированы описания в документации для улучшения удобства использования.
+</details>
+
+
+[История изменений](https://paddlepaddle.github.io/PaddleOCR/latest/en/update/update.html)
+
+
+## 🚀 Быстрый старт
+
+### Шаг 1: Попробуйте онлайн
+Официальный сайт PaddleOCR предоставляет интерактивный **Центр опыта** и **API** — без необходимости настройки, просто один клик для ознакомления.
+
+👉 [Посетить официальный сайт](https://www.paddleocr.com)
+
+### Шаг 2: Локальное развёртывание
+Для локального использования обратитесь к следующей документации в соответствии с вашими потребностями:
+
+- **Серия PP-OCR**: См. [Документацию PP-OCR](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/OCR.html)
+- **Серия PaddleOCR-VL**: См. [Документацию PaddleOCR-VL](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PaddleOCR-VL.html)
+- **PP-StructureV3**: См. [Документацию PP-StructureV3](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PP-StructureV3.html)
+- **Дополнительные возможности**: См. [Документацию по дополнительным возможностям](https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/pipeline_overview.html)
+
+
+## 🧩 Дополнительные возможности
+
+- Конвертация моделей в формат ONNX: [Получение моделей ONNX](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/others/obtaining_onnx_models.html).
+- Ускорение вывода с использованием движков OpenVINO, ONNX Runtime, TensorRT или выполнение вывода с использованием моделей в формате ONNX: [Высокопроизводительный вывод](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/local_inference/high_performance_inference.html).
+- Ускорение вывода с использованием нескольких GPU и многопроцессорной обработки: [Параллельный вывод для конвейеров](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/pipeline_usage/instructions/parallel_inference.html).
+- Интеграция PaddleOCR в приложения, написанные на C++, C#, Java и др.: [Сервисное развёртывание](https://paddlepaddle.github.io/PaddleOCR/latest/en/version3.x/inference_deployment/serving/serving.html).
+
+## 🔄 Краткий обзор результатов выполнения
+
+### PP-OCRv5
+
+<div align="center">
+  <p>
+       <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-OCRv5_demo.gif" alt="PP-OCRv5 Demo">
+  </p>
+</div>
+
+
+
+### PP-StructureV3
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-StructureV3_demo.gif" alt="PP-StructureV3 Demo">
+  </p>
+</div>
+
+### PaddleOCR-VL
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PaddleOCR-VL_demo.gif" alt="PP-StructureV3 Demo">
+  </p>
+</div>
+
+
+## ✨ Следите за обновлениями
+
+⭐ **Добавьте этот репозиторий в избранное, чтобы быть в курсе захватывающих обновлений и новых выпусков, включая мощные возможности OCR и разбора документов!** ⭐
+
+<div align="center">
+  <p>
+       <img width="1200" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/star_paddleocr2.en.gif" alt="Star-Project">
+  </p>
+</div>
+
+
+## 👩‍👩‍👧‍👦 Сообщество
+
+<div align="center">
+
+| Официальный аккаунт PaddlePaddle в WeChat | Присоединиться к группе технических обсуждений |
+| :---: | :---: |
+| <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qrcode_for_paddlepaddle_official_account.jpg" width="150"> | <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qr_code_for_the_questionnaire.jpg" width="150"> |
+</div>
+
+
+## 😃 Замечательные проекты, использующие PaddleOCR
+PaddleOCR не достиг бы своего нынешнего уровня без своего невероятного сообщества! 💗 Огромная благодарность всем нашим давним партнёрам, новым соавторам и всем, кто вложил свою душу в PaddleOCR — независимо от того, упомянуты вы здесь или нет. Ваша поддержка питает наш огонь!
+
+<div align="center">
+
+| Название проекта | Описание |
+| ------------ | ----------- |
+| [Dify](https://github.com/langgenius/dify) <a href="https://github.com/langgenius/dify"><img src="https://img.shields.io/github/stars/langgenius/dify"></a>|Готовая к производству платформа для разработки агентных рабочих процессов.|
+| [RAGFlow](https://github.com/infiniflow/ragflow) <a href="https://github.com/infiniflow/ragflow"><img src="https://img.shields.io/github/stars/infiniflow/ragflow"></a>|RAG-движок на основе глубокого понимания документов.|
+| [pathway](https://github.com/pathwaycom/pathway) <a href="https://github.com/pathwaycom/pathway"><img src="https://img.shields.io/github/stars/pathwaycom/pathway"></a>|Python ETL-фреймворк для потоковой обработки, аналитики в реальном времени, конвейеров LLM и RAG.|
+| [MinerU](https://github.com/opendatalab/MinerU) <a href="https://github.com/opendatalab/MinerU"><img src="https://img.shields.io/github/stars/opendatalab/MinerU"></a>|Инструмент для конвертации документов различных типов в Markdown.|
+| [Umi-OCR](https://github.com/hiroi-sora/Umi-OCR) <a href="https://github.com/hiroi-sora/Umi-OCR"><img src="https://img.shields.io/github/stars/hiroi-sora/Umi-OCR"></a>|Бесплатное программное обеспечение для пакетного офлайн-OCR с открытым исходным кодом.|
+| [cherry-studio](https://github.com/CherryHQ/cherry-studio) <a href="https://github.com/CherryHQ/cherry-studio"><img src="https://img.shields.io/github/stars/CherryHQ/cherry-studio"></a>|Настольный клиент с поддержкой нескольких провайдеров LLM.|
+| [haystack](https://github.com/deepset-ai/haystack)<a href="https://github.com/deepset-ai/haystack"><img src="https://img.shields.io/github/stars/deepset-ai/haystack"></a> |Фреймворк оркестрации AI для создания настраиваемых, готовых к производству приложений LLM.|
+| [OmniParser](https://github.com/microsoft/OmniParser)<a href="https://github.com/microsoft/OmniParser"><img src="https://img.shields.io/github/stars/microsoft/OmniParser"></a> |OmniParser: инструмент разбора экрана для агента GUI на основе чистого зрения.|
+| [QAnything](https://github.com/netease-youdao/QAnything)<a href="https://github.com/netease-youdao/QAnything"><img src="https://img.shields.io/github/stars/netease-youdao/QAnything"></a> |Вопросы и ответы на основе чего угодно.|
+| [Узнать о других проектах](./awesome_projects.md) | [Другие проекты на основе PaddleOCR](./awesome_projects.md)|
+</div>
+
+## 👩‍👩‍👧‍👦 Участники
+
+<div align="center">
+<a href="https://github.com/PaddlePaddle/PaddleOCR/graphs/contributors">
+  <img src="https://contrib.rocks/image?repo=PaddlePaddle/PaddleOCR&max=400&columns=20"  width="800"/>
+</a>
+</div>
+
+## 🌟 Звёзды
+
+<div align="center">
+  <p>
+      <img width="800" src="https://api.star-history.com/svg?repos=PaddlePaddle/PaddleOCR&type=Date" alt="Star-history">
+  </p>
+</div>
+
+
+## 📄 Лицензия
+Этот проект выпущен под лицензией [Apache 2.0](LICENSE).
+
+## 🎓 Цитирование
+
+```bibtex
+@misc{cui2025paddleocr30technicalreport,
+      title={PaddleOCR 3.0 Technical Report}, 
+      author={Cheng Cui and Ting Sun and Manhui Lin and Tingquan Gao and Yubo Zhang and Jiaxuan Liu and Xueqing Wang and Zelun Zhang and Changda Zhou and Hongen Liu and Yue Zhang and Wenyu Lv and Kui Huang and Yichao Zhang and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2507.05595},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2507.05595}, 
+}
+
+@misc{cui2025paddleocrvlboostingmultilingualdocument,
+      title={PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model}, 
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Handong Zheng and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2510.14528},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2510.14528}, 
+}
+
+@misc{cui2026paddleocrvl15multitask09bvlm,
+      title={PaddleOCR-VL-1.5: Towards a Multi-Task 0.9B VLM for Robust In-the-Wild Document Parsing}, 
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2026},
+      eprint={2601.21957},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2601.21957}, 
+}
+```
diff --git a/readme/README_tcn.md b/readme/README_tcn.md
new file mode 100644
index 0000000..2a1e37e
--- /dev/null
+++ b/readme/README_tcn.md
@@ -0,0 +1,303 @@
+<div align="center">
+  <p>
+      <img width="800" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Banner.png" alt="Star-history">
+  </p>
+
+
+
+<h3>全球領先的 OCR 工具包與文檔 AI 引擎</h3>
+
+
+[English](../README.md) | [简体中文](./README_cn.md)| 繁體中文 | [日本語](./README_ja.md) | [한국어](./README_ko.md) | [Français](./README_fr.md) | [Русский](./README_ru.md) | [Español](./README_es.md) | [العربية](./README_ar.md)
+
+<!-- icon -->
+
+[![PyPI Downloads](https://static.pepy.tech/badge/paddleocr)](https://pepy.tech/projects/paddleocr)
+[![Used by](https://img.shields.io/badge/Used%20by-6k%2B%20repositories-blue)](https://github.com/PaddlePaddle/PaddleOCR/network/dependents)
+![python](https://img.shields.io/badge/python-3.8~3.12-aff.svg)
+![os](https://img.shields.io/badge/os-linux%2C%20win%2C%20mac-pink.svg)
+![hardware](https://img.shields.io/badge/hardware-cpu%2C%20gpu%2C%20xpu%2C%20npu-yellow.svg)
+
+[![AI Studio](https://img.shields.io/badge/PaddleOCR-_Offiical_Website-1927BA?logo=data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAMAAADDpiTIAAAABlBMVEU2P+X///+1KuUwAAAHKklEQVR42u3dS5bjOAwEwALvf2fMavZum6IAImI7b2yYSqU+1Zb//gAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADKCR/+fzly7rD92yVg69xh8zeLwOa5w+ZvFYHtc4ft3ykB++cOm79PAp6YO2z/Ngl4ZO5l+9+yT4QAvLqS748VF33Ylzdvzpl72f6z53YIGJ6SZdPeNHcIwOycaADdLgCSIgAIgCOAACAAykIAEAAEAAFAABCAT+WQuQVgeBqXhXQIQAAYegowLQBpbg3gZGFyAC6vgBQAMREA2/YfDPxyaDQNyTNz+3Zwn5J4ZG7PB2h0kHhi7plPCImmJwkPzO0RMa3OET0i5uGlzHFze0xcu0vE2Dq3J4U2vEPgSaHbFzPNDQAAAAAAAMBNovdw+cP/ny+uaf7w/+eYADy8kE+F4Offdjn6zZXhAXgiA78G4MNNsmnu1Xr7b3mbOL8T5Ja5bw/A35EC2LiWpzt1y9jRugBy30fLg3NvHPvnuZcC2NsCUXA/aRmA89V07Fwgt37uH8deCmBr6N44pP4UgaUATpdA7v/cMbIB8okliY65/SW5HhJ1ehPmM+8edwXgpbu4R88FayR32Y/P7oZZbOx13/Zr//ZHx27bAPnkFoyewYlbAhD3TvBobr95gaUAtr1EdNx1lgI4OcTTuR3z6+FZMEDRcu9ZCuDgGCdyGxMa4EgBRMvcjrkM7NgBZw5c0TwAUWUhZwRXA2xaya65Xa3jO2qYZ8bu2AD5w38tG5V8aZpoGN6Tz0bOfa9bceyWAciTO0jWyO1Tc5cLwJmF/JfPnXVyu3/slgHIg1n79O2O5fZv+1cHV7sC2HYqmUdHysNzX3sVkMcjUK5Gc+dMs28E5bGtm0V3gloBOP9vgZv+4sYn3RUaYFMCol5uN77g6lUApc8pWs69Zn7snS9Z9Q8G0S0AUTVUUTG3A54R1KSvo/diLAv5fKzynZeN6xogC75u93+AtBTA47OlAFSv6qY/vp3DAjD8iv2ZdFYJwKynMhTK1rInPfzaxW81LnvSgFP9KxrATaCLA3DxHpbFX31ZyNm5XRZyXG5bNkAWfP0rcrsUwOgC6NIAzgBcBiqAWwPgLrAGuGBP6jr2sifdfiJ6QQM4Bbw4AK4B3129ZSFn53ZZyA/GyFty27IBFMDFAXAG8PbyLQv5xULGPRl0K3h2AbwcgCZPhs+LD1zLnjS6AN4NwMU/DVFh7LyhASreTbvqrxdr/J4XT4Swz4FrTS+AGJ7bNbwAYkxuWzZAVljHrJfbjb9wviYXwFO/FJ8Vli4vaICsEMFyBbA3tmtsAUS0zG1c/bj4YwsZH2/+Whd0+1Nb+S7IE2sfPw4RL0XmsR8Nqvz7qFngmPHF34EqjP15AAofAkosZKPC/K6FVoeP02Ehi540NG6AK/4pYP3cLgVwXwHkDQ1QcSGb/uF4WwCmfX8u/+4vgLINcMUlQIfcLgXwXAF0+BGkpQDuuJx7/hwgpu//cWVuO3wxJOz/z8297vgYBwaIO3O7Kn+c194578ltywbIgu8fl+Z2lS+APvnLjnOv8hsgSqxjgwL4Ln9LAezaj98tgPzy7ZcC+GQzxrWxXQpgx370dm6/H7v6jaBoso5dY1swAFlwHWvfBf5pxVa93fCtdx64+1dsgCy4joWvAfPX9VoKYMs6Zse9/8Mlvv7LILlhAfKFFdsSutJXAdFkL3qlADJPrXFcXAC5KYaH586jO9mtAch9S3T0GQJ726ZWAE49kjP3rlDJuetdaL/1zeqZY9c7CRz7s0wCUPxienQBnAuAAtAAlxaAAAxfyBQABSAACkAAFIAAKAABUAACMEkKwL170oh7V8ueNLoAjgTAXWAN4BRwcABcA2oABTA4AApAAyiAwQFQABpAAQwOgALQADMWUgCuEmNyu15fSIY3gFPAiwPgFFADKIDBAVAAGkABCIACmBqAUAAaQAHMDUCMWkgBuMWw3K43F5LhDeAU8OIAuAmkARTA4AAoAA2gAARAAUwNgLvAGkABDA6Au8AaoKOJuV0vLSTDG8Ap4MUBcBNIAyiAwQFQABpAAQwOgALQAApAABTA1AC4C6wBOhqb23V+IRneAE4BLw6Aa0ANoAAGB0ABaAAFMDgACkADKAABUABTA+AusAboKATAQs4trjV+IYcfuJYCcA6gAATAQk69dFkKQANYyLkFcLIBFIDLQAVwawDsSRrAEWBwAJwCagAFMDgACkADKIDBAVAAGkABCIACmBoAzwXWAApgcADsSRrg0iNACoACEADXgAIwdCFTACykALgGFIAfl0kBAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAPBv/gN+IH8U6YveYgAAAABJRU5ErkJggg==&labelColor=white)](https://www.paddleocr.com)
+[![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/PaddlePaddle/PaddleOCR)
+[![License](https://img.shields.io/badge/license-Apache_2.0-green)](../LICENSE)
+
+</div>
+
+
+
+
+**PaddleOCR 以業界領先的精準度，將 PDF 文件和圖像轉換為結構化、LLM 友好的資料格式（JSON/Markdown）。憑藉 70,000+ Stars 的成績，PaddleOCR 已獲得 Dify、RAGFlow、Cherry Studio 等頂級專案的廣泛信賴，是建構智慧 RAG 和 Agentic 應用的核心基礎元件。**
+
+
+## 🚀 核心特性
+
+### 📄 智能文檔解析（面向大模型）
+> *為大模型時代將雜亂的文檔視覺信息轉化為結構化數據。*
+
+* **SOTA 級文檔視覺語言模型 (VLM)**: 業界領先的輕量級文檔解析視覺語言模型 **PaddleOCR-VL-1.6 (0.9B)**。該模型以 96.3% 精度刷新 OmniDocBench v1.6，文本、公式、表格識別全面領先，並在古籍、生僻字、印章、圖表等多場景能力顯著增強，支持以 **Markdown** 和 **JSON** 格式輸出結構化結果。
+* **版面結構分析**：由**PP-StructureV3**驅動，無縫將複雜的PDF和圖像轉換為**Markdown**或**JSON**格式。與PaddleOCR-VL系列模型不同，它提供更細粒度的坐標信息,包括表格單元格坐標、文本坐標等，
+* **生產級高效能**：以極小的模型體積實現商業級別的準確率。在公開基準測試中超越眾多閉源解決方案，同時保持極高的資源利用率，完美適配邊緣計算與雲端部署。
+
+### 🔍 通用文本識別（場景 OCR）
+> *快速、多語言文本檢測與識別的全球黃金標準。*
+
+* **支持 100+ 種語言**：原生支持龐大豐富的全球語種庫。**PP-OCRv6** 單模型統一支持 50 種語言（中、英、日及 46 種拉丁語系），無需切換模型即可應對多語言混合排版文檔。
+* **複雜場景支持**：除了標準的文本識別，我們還支持在各種廣泛的環境下進行**自然場景文本檢測與識別**，涵蓋身份證件、街景、書籍以及工業零部件等。
+* **性能提升**：PP-OCRv6 相比 PP-OCRv5 檢測精度提升 **4.6%**、識別精度提升 **5.1%**，超越主流視覺語言大模型，CPU 推理加速 5.2×。
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/Arch_cn.jpg" alt="PaddleOCR Architecture">
+  </p>
+</div>
+
+### 🛠️ 以開發者為中心的生態系統
+* **無縫集成**：AI智能體生態系統的首選——與**Dify、RAGFlow、Pathway和Cherry Studio**深度集成。
+* **大語言模型數據飛輪**：完整的數據流水線,用於構建高質量數據集，為微調大語言模型提供可持續的"數據引擎"。
+* **一鍵部署**：支持多種硬件後端(NVIDIA GPU、Intel CPU、昆侖芯XPU和多種AI加速器)。
+
+
+## 📣 最新動態
+
+### 🔥 2026.06.11: PaddleOCR 3.7.0 發布
+- PP-OCRv6 核心亮點如下：
+
+    - **精度全面提升**：medium 檔相比 PP-OCRv5_server 檢測精度提升 4.6%、識別精度提升 5.1%，以僅 34.5M 參數超越 Qwen3-VL-235B、GPT-5.5 等主流視覺語言大模型。
+    - **50 種語言統一支持**：單一模型覆蓋中文、英文、日文及 46 種拉丁語系語言，無需為不同語種切換模型。
+    - **專業場景增強**：數碼顯示屏、點陣字符、輪胎印字、工業字符等傳統 VLM 難以覆蓋的場景識別能力大幅提升。
+    - **推理速度更快**：medium 檔 CPU OpenVINO 推理加速 5.2×，tiny 檔 Apple M4 加速 6.1×，A100 上僅需 0.13s。
+    - **三檔模型覆蓋全場景**：tiny（1.5M）/ small（7.7M）/ medium（34.5M）分別面向端側/移動端/服務端部署。
+    - **模型取得**：所有模型均可透過 [HuggingFace](https://huggingface.co/collections/PaddlePaddle/pp-ocrv6) 和 [ModelScope](https://www.modelscope.cn/collections/PaddlePaddle/PP-OCRv6) 取得。
+
+<details>
+<summary><strong>2026.05.28: PaddleOCR 3.6.0 發布</strong></summary>
+
+- PaddleOCR-VL-1.6 核心亮點：
+
+    - **全新 SOTA 精度**：OmniDocBench v1.6 突破 96.3%，OmniDocBench v1.5 和 Real5-OmniDocBench 同步刷新 SOTA。
+    - **能力全面升級**：表格、古籍、生僻字識別大幅提升。
+    - **無縫遷移**：模型結構與 PaddleOCR-VL-1.5 完全一致，零成本適配。
+    - **立即試用**：[HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.6) 或[官方網站](https://www.paddleocr.com)。
+
+</details>
+<details>
+<summary><strong>2026.04.21: PaddleOCR 3.5.0 發布</strong></summary>
+
+* **推理後端靈活切換**：支持在飛槳靜態圖、飛槳動態圖和 Transformers 之間無縫切換。深度適配 Hugging Face 生態，20 個主要模型支持以 Transformers 作為推理後端。
+* **常見文檔格式轉 Markdown**：支持將 Word、Excel、PowerPoint 等常見文檔格式轉換為 Markdown。
+* **解析結果導出 DOCX**：`PaddleOCR-VL` 系列、`PP-StructureV3` 和 `PP-DocTranslation` 現已支持將解析結果導出為 DOCX，便於在 Microsoft Word 中查看和編輯。
+* **官方瀏覽器推理 SDK**：發布官方瀏覽器推理 SDK `PaddleOCR.js`，支持在瀏覽器中運行 `PP-OCRv5`。
+
+</details>
+
+<details>
+<summary><strong>2026.01.29: PaddleOCR 3.4.0 發布</strong></summary>
+* PaddleOCR-VL-1.5 (SOTA 0.9B VLM)：我們最新的旗艦文檔解析模型現已上線!
+    * **OmniDocBench 94.5%準確率**：超越頂級通用大模型和專業文檔解析模型。
+    * **現實5大場景文檔解析的SOTA性能**：首次引入**PP-DocLayoutV3**算法進行不規則形狀定位，掌控5種艱難場景:傾斜、彎曲、掃描、光照和屏幕拍照。
+    * **能力拓展**：現已支持**印章識別**、**文本識別**，並擴展至**111種語言**(包括中國的藏文和孟加拉語)。
+    * **長文檔跨頁解析**：支持自動跨頁表格合併和分層標題識別。
+    * **立即試用**：可在[HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.5)或我們的[官方網站](https://www.paddleocr.com)使用。
+
+</details>
+
+<details>
+<summary><strong>2025.10.16: PaddleOCR 3.3.0 發布</strong></summary>
+
+- **發布PaddleOCR-VL**：
+    - **模型介紹**：
+        - **PaddleOCR-VL** 是一款先進、高效的文檔解析模型，專為文檔中的元素識別設計。其核心組件為 PaddleOCR-VL-0.9B，這是一種緊湊而強大的視覺語言模型（VLM），它由 NaViT 風格的動態分辨率視覺編碼器與 ERNIE-4.5-0.3B 語言模型組成，能夠實現精準的元素識別。**該模型支持 109 種語言，並在識別複雜元素（如文本、表格、公式和圖表）方面表現出色，同時保持極低的資源消耗。通過在廣泛使用的公開基準與內部基準上的全面評測，PaddleOCR-VL 在頁級級文檔解析與元素級識別均達到 SOTA 表現**。它顯著優於現有的基於Pipeline方案和文檔解析多模態方案以及先進的通用多模態大模型，並具備更快的推理速度。這些優勢使其非常適合在真實場景中落地部署。模型已發布至[HuggingFace](https://huggingface.co/PaddlePaddle/PaddleOCR-VL)，歡迎大家下載使用！更多介紹內容請點擊[PaddleOCR-VL](https://www.paddleocr.ai/latest/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL.html)。
+
+    - **特性**：
+        - **緊湊而強大的視覺語言模型架構**：我們提出了一種新的視覺語言模型，專為資源高效的推理而設計，在元素識別方面表現出色。通過將NaViT風格的動態高分辨率視覺編碼器與輕量級的ERNIE-4.5-0.3B語言模型結合，我們顯著增強了模型的識別能力和解碼效率。這種集成在保持高準確率的同時降低了計算需求，使其非常適合高效且實用的文檔處理應用。
+        - **文檔解析的SOTA性能**：PaddleOCR-VL在頁面級文檔解析和元素級識別中達到了最先進的性能。它顯著優於現有的基於流水線的解決方案，並在文檔解析中展現出與領先的視覺語言模型（VLMs）競爭的強勁實力。此外，它在識別複雜的文檔元素（如文本、表格、公式和圖表）方面表現出色，使其適用於包括手寫文本和歷史文獻在內的各種具有挑戰性的內容類型。這使得它具有高度的多功能性，適用於廣泛的文檔類型和場景。
+        - **多語言支持**：PaddleOCR-VL支持109種語言，覆蓋了主要的全球語言，包括但不限於中文、英文、日文、拉丁文和韓文，以及使用不同文字和結構的語言，如俄語（西里爾字母）、阿拉伯語、印地語（天城文）和泰語。這種廣泛的語言覆蓋大大增強了我們系統在多語言和全球化文檔處理場景中的適用性。
+
+- **發布PP-OCRv5小語種識別模型**：
+    - 優化拉丁文識別的準度和廣度，新增西里爾文、阿拉伯文、天城文、泰盧固語、泰米爾語等語系，覆蓋109種語言文字的識別。模型參數量僅為2M，部分模型精度較上一代提升40%以上。
+
+</details>
+
+
+<details>
+<summary><strong>2025.08.21: PaddleOCR 3.2.0 發布</strong></summary>
+
+- **重要模型新增：**
+    - 新增 PP-OCRv5 英文、泰文、希臘文識別模型的訓練、推理、部署。**其中 PP-OCRv5 英文模型較 PP-OCRv5 主模型在英文場景提升 11%，泰文識別模型精度 82.68%，希臘文識別模型精度 89.28%。**
+
+- **部署能力升級：**
+    - **全面支持飛槳框架 3.1.0 和 3.1.1 版本。**
+    - **全面升級 PP-OCRv5 C++ 本地部署方案，支持 Linux、Windows，功能及精度效果與 Python 方案保持一致。**
+    - **高性能推理支持 CUDA 12，可使用 Paddle Inference、ONNX Runtime 後端推理。**
+    - **高穩定性服務化部署方案全面開源，支持用戶根據需求對 Docker 鏡像和 SDK 進行定制化修改。**
+    - 高穩定性服務化部署方案支持通過手動構造HTTP請求的方式調用，該方式允許客戶端代碼使用任意編程語言編寫。
+
+- **Benchmark支持**：
+    - **全部產線支持產線細粒度 benchmark，能夠測量產線端到端推理時間以及逐層、逐模塊的耗時數據，可用於輔助產線性能分析。可以參考[文檔](../docs/version3.x/pipeline_usage/instructions/benchmark.md)來進行性能測試。**
+    - **文檔中補充各產線常用配置在主流硬件上的關鍵指標，包括推理耗時和內存佔用等，為用戶部署提供參考。**
+
+- **Bug修復：**
+    - 修復模型訓練時訓練日誌保存失敗的問題。
+    - 對公式模型的數據增強部分進行了版本兼容性升級，以適應新版本的 albumentations 依賴，並修復了在多進程使用 tokenizers 依賴包時出現的死鎖警告。
+    - 修復 PP-StructureV3 配置文件中的 `use_chart_parsing` 等開關行為與其他產線不統一的問題。
+
+- **其他升級：**
+    - **分離必要依賴與可選依賴。使用基礎文字識別功能時，僅需安裝少量核心依賴；若需文檔解析、信息抽取等功能，用戶可按需選擇安裝額外依賴。**
+    - **支持 Windows 用戶使用英偉達 50 系顯卡，可根據 [安裝文檔](../docs/version3.x/installation.md) 安裝對應版本的 paddle 框架。**
+    - **PP-OCR 系列模型支持返回單文字坐標。**
+    - 模型新增 AIStudio、ModelScope 等下載源。可指定相關下載源下載對應的模型。
+    - 支持圖表轉表 PP-Chart2Table 單功能模塊推理能力。
+    - 優化部分使用文檔中的描述，提升易用性。
+</details>
+
+
+[歷史日誌](https://paddlepaddle.github.io/PaddleOCR/latest/en/update/update.html)
+
+
+## 🚀 快速開始
+
+### 步驟 1: 在線體驗
+PaddleOCR官方網站提供交互式**體驗中心**和**APIs**——無需設置,一鍵體驗。
+
+👉 [訪問官方網站](https://www.paddleocr.com)
+
+### 步驟 2: 本地部署
+對於本地使用,請根據您的需求參考以下文檔:
+
+- **PP-OCR系列**：查看[PP-OCR文檔](https://www.paddleocr.ai/latest/version3.x/pipeline_usage/OCR.html)
+- **PaddleOCR-VL系列**：查看[PaddleOCR-VL文檔](https://www.paddleocr.ai/latest/version3.x/pipeline_usage/PaddleOCR-VL.html)
+- **PP-StructureV3**：查看[PP-StructureV3文檔](https://www.paddleocr.ai/latest/version3.x/pipeline_usage/PP-StructureV3.html)
+- **更多能力**：查看[更多能力文檔](https://www.paddleocr.ai/latest/version3.x/pipeline_usage/pipeline_overview.html)
+
+
+## 🧩 更多功能
+
+- 將模型轉換為ONNX格式: [獲取ONNX模型](https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/inference_deployment/others/obtaining_onnx_models.html)。
+- 使用OpenVINO、ONNX Runtime、TensorRT等引擎加速推理,或使用ONNX格式模型進行推理: [高性能推理](https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/inference_deployment/local_inference/high_performance_inference.html)。
+- 使用多GPU和多進程加速推理: [流水線並行推理](https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/pipeline_usage/instructions/parallel_inference.html)。
+- 將PaddleOCR集成到C++、C#、Java等語言編寫的應用程序中: [服務化部署](https://paddlepaddle.github.io/PaddleOCR/latest/version3.x/inference_deployment/serving/serving.html)。
+
+## 🔄 執行結果快速預覽
+
+### PP-OCRv5
+
+<div align="center">
+  <p>
+       <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-OCRv5_demo.gif" alt="PP-OCRv5 Demo">
+  </p>
+</div>
+
+
+
+### PP-StructureV3
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PP-StructureV3_demo.gif" alt="PP-StructureV3 Demo">
+  </p>
+</div>
+
+### PaddleOCR-VL
+
+<div align="center">
+  <p>
+      <img width="100%" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/PaddleOCR-VL_demo.gif" alt="PP-StructureV3 Demo">
+  </p>
+</div>
+
+
+## ✨ 保持關注
+
+⭐ **收藏本倉庫，持續關注最新動態與版本發布，包括強大的 OCR 及文檔解析等新功能特性。** ⭐
+
+<div align="center">
+  <p>
+       <img width="1200" src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/main/images/paddleocr/README/star_paddleocr2.en.gif" alt="Star-Project">
+  </p>
+</div>
+
+
+## 👩‍👩‍👧‍👦 社區
+
+<div align="center">
+
+| PaddlePaddle 微信公眾號 |  加入技術討論群 |
+| :---: | :---: |
+| <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qrcode_for_paddlepaddle_official_account.jpg" width="150"> | <img src="https://raw.githubusercontent.com/cuicheng01/PaddleX_doc_images/refs/heads/main/images/paddleocr/README/qr_code_for_the_questionnaire.jpg" width="150"> |
+</div>
+
+
+## 😃 使用 PaddleOCR 的優秀項目
+
+<div align="center">
+
+PaddleOCR 的發展離不開社區貢獻！💗衷心感謝所有開發者、合作夥伴與貢獻者！
+| 項目名稱 | 簡介 |
+| ------------ | ----------- |
+| [RAGFlow](https://github.com/infiniflow/ragflow) <a href="https://github.com/infiniflow/ragflow"><img src="https://img.shields.io/github/stars/infiniflow/ragflow"></a>|基於RAG的AI工作流引擎|
+| [pathway](https://github.com/pathwaycom/pathway) <a href="https://github.com/pathwaycom/pathway"><img src="https://img.shields.io/github/stars/pathwaycom/pathway"></a>|用於流處理、實時分析、LLM流水線和RAG的Python ETL框架|
+| [MinerU](https://github.com/opendatalab/MinerU) <a href="https://github.com/opendatalab/MinerU"><img src="https://img.shields.io/github/stars/opendatalab/MinerU"></a>|多類型文檔轉換Markdown工具|
+| [Umi-OCR](https://github.com/hiroi-sora/Umi-OCR) <a href="https://github.com/hiroi-sora/Umi-OCR"><img src="https://img.shields.io/github/stars/hiroi-sora/Umi-OCR"></a>|開源批量離線OCR軟件|
+| [cherry-studio](https://github.com/CherryHQ/cherry-studio) <a href="https://github.com/CherryHQ/cherry-studio"><img src="https://img.shields.io/github/stars/CherryHQ/cherry-studio"></a>|一個支持多個LLM提供商的桌面客戶端|
+| [OmniParser](https://github.com/microsoft/OmniParser)<a href="https://github.com/microsoft/OmniParser"><img src="https://img.shields.io/github/stars/microsoft/OmniParser"></a> |基於純視覺的GUI智能體屏幕解析工具|
+| [QAnything](https://github.com/netease-youdao/QAnything)<a href="https://github.com/netease-youdao/QAnything"><img src="https://img.shields.io/github/stars/netease-youdao/QAnything"></a> |基於任意內容的問答系統|
+| [PDF-Extract-Kit](https://github.com/opendatalab/PDF-Extract-Kit) <a href="https://github.com/opendatalab/PDF-Extract-Kit"><img src="https://img.shields.io/github/stars/opendatalab/PDF-Extract-Kit"></a>|高效複雜PDF文檔提取工具包|
+| [Dango-Translator](https://github.com/PantsuDango/Dango-Translator)<a href="https://github.com/PantsuDango/Dango-Translator"><img src="https://img.shields.io/github/stars/PantsuDango/Dango-Translator"></a> |屏幕實時翻譯工具|
+| [更多項目](../awesome_projects.md) |  [更多基於PaddleOCR的項目](../awesome_projects.md) |
+</div>
+
+
+## 👩‍👩‍👧‍👦 貢獻者
+
+<div align="center">
+<a href="https://github.com/PaddlePaddle/PaddleOCR/graphs/contributors">
+  <img src="https://contrib.rocks/image?repo=PaddlePaddle/PaddleOCR&max=400&columns=20"  width="800"/>
+</a>
+</div>
+
+## 🌟 Star歷史
+
+<div align="center">
+  <p>
+      <img width="800" src="https://api.star-history.com/svg?repos=PaddlePaddle/PaddleOCR&type=Date" alt="Star-history">
+  </p>
+</div>
+
+
+## 📄 許可證
+本項目採用[Apache 2.0許可證](LICENSE)發布。
+
+## 🎓 引用
+
+```bibtex
+@misc{cui2025paddleocr30technicalreport,
+      title={PaddleOCR 3.0 Technical Report},
+      author={Cheng Cui and Ting Sun and Manhui Lin and Tingquan Gao and Yubo Zhang and Jiaxuan Liu and Xueqing Wang and Zelun Zhang and Changda Zhou and Hongen Liu and Yue Zhang and Wenyu Lv and Kui Huang and Yichao Zhang and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2507.05595},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2507.05595},
+}
+
+@misc{cui2025paddleocrvlboostingmultilingualdocument,
+      title={PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model},
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Handong Zheng and Jing Zhang and Jun Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2025},
+      eprint={2510.14528},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2510.14528},
+}
+
+@misc{cui2026paddleocrvl15multitask09bvlm,
+      title={PaddleOCR-VL-1.5: Towards a Multi-Task 0.9B VLM for Robust In-the-Wild Document Parsing},
+      author={Cheng Cui and Ting Sun and Suyin Liang and Tingquan Gao and Zelun Zhang and Jiaxuan Liu and Xueqing Wang and Changda Zhou and Hongen Liu and Manhui Lin and Yue Zhang and Yubo Zhang and Yi Liu and Dianhai Yu and Yanjun Ma},
+      year={2026},
+      eprint={2601.21957},
+      archivePrefix={arXiv},
+      primaryClass={cs.CV},
+      url={https://arxiv.org/abs/2601.21957},
+}
+```
diff --git a/requirements.txt b/requirements.txt
new file mode 100644
index 0000000..5d4f463
--- /dev/null
+++ b/requirements.txt
@@ -0,0 +1,18 @@
+shapely
+scikit-image
+pyclipper
+lmdb; python_version >= "3.9"
+lmdb<1.5; python_version < "3.9"
+tqdm
+numpy
+rapidfuzz
+opencv-python
+opencv-contrib-python
+cython
+Pillow
+pyyaml
+requests
+albumentations
+# to be compatible with albumentations
+albucore
+packaging
diff --git a/setup.py b/setup.py
new file mode 100644
index 0000000..f1e3f1b
--- /dev/null
+++ b/setup.py
@@ -0,0 +1,18 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from setuptools import setup
+
+
+setup()
diff --git a/skills/README.md b/skills/README.md
new file mode 100644
index 0000000..8b9729f
--- /dev/null
+++ b/skills/README.md
@@ -0,0 +1,19 @@
+# PaddleOCR Agent Skills
+
+English | [中文](./README_cn.md)
+
+This directory contains the official PaddleOCR Agent Skills.
+
+## Included Skills
+
+- `paddleocr-text-recognition`: text recognition for images and PDF files
+- `paddleocr-doc-parsing`: document parsing and structured output
+
+## Documentation
+
+For installation, configuration, usage examples, and local verification, see the [documentation](../docs/version3.x/integrations/skills.en.md).
+
+## Directory Layout
+
+- `paddleocr-text-recognition/`: text recognition skill
+- `paddleocr-doc-parsing/`: document parsing skill
diff --git a/skills/README_cn.md b/skills/README_cn.md
new file mode 100644
index 0000000..8b373e8
--- /dev/null
+++ b/skills/README_cn.md
@@ -0,0 +1,19 @@
+# PaddleOCR Agent Skills
+
+中文 | [English](./README.md)
+
+本目录包含 PaddleOCR 官方提供的 Agent Skills。
+
+## Skills 列表
+
+- `paddleocr-text-recognition`：用于图片和 PDF 文字识别。
+- `paddleocr-doc-parsing`：用于文档解析与结构化输出。
+
+## 详细文档
+
+安装、配置、使用示例和本地验证说明，请参考 [文档](../docs/version3.x/integrations/skills.md)。
+
+## 目录说明
+
+- `paddleocr-text-recognition/`：文字识别 Skill
+- `paddleocr-doc-parsing/`：文档解析 Skill
diff --git a/skills/paddleocr-doc-parsing/SKILL.md b/skills/paddleocr-doc-parsing/SKILL.md
new file mode 100644
index 0000000..66e8cba
--- /dev/null
+++ b/skills/paddleocr-doc-parsing/SKILL.md
@@ -0,0 +1,138 @@
+---
+name: paddleocr-doc-parsing
+description: >-
+  Use this skill to extract structured Markdown/JSON from PDFs and document images—tables with
+  cell-level precision, formulas as LaTeX, figures, seals, charts, headers/footers, multi-column
+  layout and correct reading order.
+  Trigger terms: 文档解析, 版面分析, 版面还原, 表格提取, 公式识别, 多栏排版, 扫描件结构化,
+  发票, 财报, 复杂 PDF, PDF转Markdown, 图表, 阅读顺序; reading order, formula, LaTeX,
+  layout parsing, structure extraction, PP-StructureV3, PaddleOCR-VL.
+license: Apache-2.0
+metadata:
+  openclaw:
+    requires:
+      env:
+        - PADDLEOCR_ACCESS_TOKEN
+      bins:
+        - paddleocr
+    primaryEnv: PADDLEOCR_ACCESS_TOKEN
+    emoji: "📄"
+    install:
+      - kind: uv
+        package: paddleocr
+        bins: [paddleocr]
+---
+
+# PaddleOCR Document Parsing
+
+## When to Use This Skill
+
+**Use this skill for**:
+
+- Documents with tables (invoices, financial reports, spreadsheets)
+- Documents with mathematical formulas (academic papers, scientific documents)
+- Documents with charts and diagrams
+- Multi-column layouts (newspapers, magazines, brochures)
+- Complex document structures requiring layout analysis
+
+## Usage
+
+### Basic Document Parsing
+
+From URL:
+
+```bash
+paddleocr api \
+  --model_type doc_parsing \
+  --file_url "https://example.com/report.pdf"
+```
+
+From local file:
+
+```bash
+paddleocr api \
+  --model_type doc_parsing \
+  --file_path "./document.pdf"
+```
+
+### Common Options
+
+```bash
+# With specific model
+paddleocr api \
+  --model_type doc_parsing \
+  --model PP-StructureV3 \
+  --file_path "./report.pdf"
+
+# Disable preprocessing (faster, for flat/well-oriented images)
+paddleocr api \
+  --model_type doc_parsing \
+  --file_path "./document.pdf" \
+  --use_doc_unwarping False \
+  --use_doc_orientation_classify False
+
+# With page ranges
+paddleocr api \
+  --model_type doc_parsing \
+  --file_path "./large.pdf" \
+  --page_ranges "1-5,10,15-20"
+
+# Save result and resources
+paddleocr api \
+  --model_type doc_parsing \
+  --file_url "https://..." \
+  --output result.json \
+  --save_resources ./resources
+
+# Prettify markdown output
+paddleocr api \
+  --model_type doc_parsing \
+  --file_path "./document.pdf" \
+  --prettify_markdown True
+```
+
+### Output Format
+
+```json
+{
+  "jobId": "job-xxx",
+  "pages": [
+    {
+      "markdownText": "# Title\n\nContent...",
+      "markdownImages": {
+        "img1": "https://...",
+        "img2": "https://..."
+      },
+      "outputImages": {
+        "layout1": "https://..."
+      }
+    }
+  ]
+}
+```
+
+## Important Notes
+
+**Preprocessing options**: For flat, well-oriented images (screenshots, properly scanned documents), you can disable preprocessing for faster results:
+
+```bash
+paddleocr api --model_type doc_parsing --file_path "./document.pdf" --use_doc_unwarping False --use_doc_orientation_classify False
+```
+
+Keep preprocessing enabled when:
+- The input is a photo of a curved or folded document
+- The document has significant perspective distortion
+- Orientation is uncertain (rotated 90/180/270 degrees)
+
+**Display complete results**: Always show the full extracted content to users. Do not truncate with "..." unless content exceeds 10,000 characters. When multiple pages are processed, summarize if needed but provide complete results when explicitly requested.
+
+**Handle errors gracefully**: When the CLI returns an error, inform the user of the specific issue rather than silently failing. Common errors:
+- Authentication: `PADDLEOCR_ACCESS_TOKEN` invalid or missing
+- Quota: API rate limit exceeded
+- No content detected: Document may be blank or contain no extractable text
+
+## CLI Reference
+
+Run `paddleocr api --help` for all options.
+
+For full documentation, see: [PaddleOCR Official Documentation](https://www.paddleocr.ai/latest/en/version3.x/inference_deployment/serving/paddleocr_official_api/cli.html)
diff --git a/skills/paddleocr-text-recognition/SKILL.md b/skills/paddleocr-text-recognition/SKILL.md
new file mode 100644
index 0000000..f9f792c
--- /dev/null
+++ b/skills/paddleocr-text-recognition/SKILL.md
@@ -0,0 +1,130 @@
+---
+name: paddleocr-text-recognition
+description: >-
+  Use this skill whenever the user wants text extracted from images, photos, scans, screenshots,
+  or scanned PDFs. Returns exact machine-readable strings with line-level text and optional bbox
+  coordinates. Strong accuracy for CJK, small print, and handwritten text.
+  Trigger terms: OCR, 文字识别, 图片转文字, 截图识字, 提取图中文字, 扫描识字, 识字, 纯文字,
+  plain text extraction, 坐标, 检测框, bbox, bounding box, image to text, screenshot, photo scan,
+  recognize text.
+license: Apache-2.0
+metadata:
+  openclaw:
+    requires:
+      env:
+        - PADDLEOCR_ACCESS_TOKEN
+      bins:
+        - paddleocr
+    primaryEnv: PADDLEOCR_ACCESS_TOKEN
+    emoji: "🔤"
+    install:
+      - kind: uv
+        package: paddleocr
+        bins: [paddleocr]
+---
+
+# PaddleOCR Text Recognition
+
+## When to Use This Skill
+
+**Use this skill for**:
+
+- Extract text from images (screenshots, photos, scans)
+- Extract text from PDFs or document images when the goal is **line/box-level text**
+- Extract text from URLs or local files that point to images/PDFs
+
+**Do not use for**:
+
+- Documents with tables, formulas, charts, or complex layouts — use Document Parsing instead
+
+## Usage
+
+### Basic OCR
+
+From URL:
+
+```bash
+paddleocr api \
+  --model_type ocr \
+  --file_url "https://example.com/image.png"
+```
+
+From local file:
+
+```bash
+paddleocr api \
+  --model_type ocr \
+  --file_path "./document.pdf"
+```
+
+### Common Options
+
+```bash
+# With specific model
+paddleocr api \
+  --model_type ocr \
+  --model PP-OCRv5 \
+  --file_path "./report.pdf"
+
+# Disable preprocessing (faster, for flat/well-oriented images)
+paddleocr api \
+  --model_type ocr \
+  --file_path "./document.pdf" \
+  --use_doc_unwarping False \
+  --use_doc_orientation_classify False
+
+# Save result to file
+paddleocr api \
+  --model_type ocr \
+  --file_url "https://..." \
+  --output result.json
+
+# Page ranges
+paddleocr api \
+  --model_type ocr \
+  --file_path "./large.pdf" \
+  --page_ranges "1-5,10,15-20"
+```
+
+### Output Format
+
+```json
+{
+  "jobId": "job-xxx",
+  "pages": [
+    {
+      "prunedResult": {
+        "rec_texts": ["Line 1", "Line 2"],
+        "rec_scores": [0.98, 0.95]
+      },
+      "ocrImageUrl": "https://..."
+    }
+  ]
+}
+```
+
+## Important Notes
+
+**Preprocessing options**: By default, the API enables document preprocessing (unwarping and orientation classification). For flat, well-oriented images (screenshots, properly scanned documents), you can disable preprocessing for faster results:
+
+```bash
+paddleocr api --model_type ocr --file_path "./document.pdf" --use_doc_unwarping False --use_doc_orientation_classify False
+```
+
+Keep preprocessing enabled when:
+- The input is a photo of a curved or folded document
+- The document has significant perspective distortion
+- Orientation is uncertain (rotated 90/180/270 degrees)
+
+**Display complete results**: Always show the full extracted content to users. Do not truncate with "..." unless content exceeds 10,000 characters. When multiple pages are processed, summarize if needed but provide complete results when explicitly requested.
+
+**Handle errors gracefully**: When the CLI returns an error, inform the user of the specific issue rather than silently failing or falling back to your own vision capabilities. Common errors:
+- Authentication: `PADDLEOCR_ACCESS_TOKEN` invalid or missing
+- Quota: API rate limit exceeded
+- No content detected: Image may be blank or contain no text
+
+## CLI Reference
+
+Run `paddleocr api --help` for all options.
+
+For full documentation, see: [PaddleOCR Official Documentation](https://www.paddleocr.ai/latest/en/version3.x/inference_deployment/serving/paddleocr_official_api/cli.html)
diff --git a/test_tipc/benchmark_train.sh b/test_tipc/benchmark_train.sh
new file mode 100644
index 0000000..1668e41
--- /dev/null
+++ b/test_tipc/benchmark_train.sh
@@ -0,0 +1,294 @@
+#!/bin/bash
+source test_tipc/common_func.sh
+
+# run benchmark sh 
+# Usage:
+# bash run_benchmark_train.sh config.txt params
+# or 
+# bash run_benchmark_train.sh config.txt
+
+function func_parser_params(){
+    strs=$1
+    IFS="="
+    array=(${strs})
+    tmp=${array[1]}
+    echo ${tmp}
+}
+
+function set_dynamic_epoch(){
+    string=$1
+    num=$2
+    _str=${string:1:6}
+    IFS="C"
+    arr=(${_str})
+    M=${arr[0]}
+    P=${arr[1]}
+    ep=`expr $num \* $M \* $P`
+    echo $ep
+}
+
+function func_sed_params(){
+    filename=$1
+    line=$2
+    param_value=$3
+    params=`sed -n "${line}p" $filename`
+    IFS=":"
+    array=(${params})
+    key=${array[0]}
+    value=${array[1]}
+
+    new_params="${key}:${param_value}"
+    IFS=";"
+    cmd="sed -i '${line}s/.*/${new_params}/' '${filename}'"
+    eval $cmd
+}
+
+function set_gpu_id(){
+    string=$1
+    _str=${string:1:6}
+    IFS="C"
+    arr=(${_str})
+    M=${arr[0]}
+    P=${arr[1]}
+    gn=`expr $P - 1`
+    gpu_num=`expr $gn / $M`
+    seq=`seq -s "," 0 $gpu_num`
+    echo $seq
+}
+
+function get_repo_name(){
+    IFS=";"
+    cur_dir=$(pwd)
+    IFS="/"
+    arr=(${cur_dir})
+    echo ${arr[-1]}
+}
+
+FILENAME=$1
+# copy FILENAME as new
+new_filename="./test_tipc/benchmark_train.txt"
+cmd=`yes|cp $FILENAME $new_filename`
+FILENAME=$new_filename
+# MODE must be one of ['benchmark_train']
+MODE=$2
+PARAMS=$3
+
+to_static=""
+# parse "to_static" options and modify trainer into "to_static_trainer"
+if [[ $PARAMS =~ "dynamicTostatic" ]] ;then
+   to_static="d2sT_"
+   sed -i 's/trainer:norm_train/trainer:to_static_train/g' $FILENAME
+   # clear PARAM contents
+   if [ $PARAMS = "to_static" ] ;then
+    PARAMS=""
+   fi
+fi
+# bash test_tipc/benchmark_train.sh test_tipc/configs/det_mv3_db_v2_0/train_benchmark.txt  benchmark_train dynamic_bs8_fp32_DP_N1C8
+# bash test_tipc/benchmark_train.sh test_tipc/configs/det_mv3_db_v2_0/train_benchmark.txt  benchmark_train dynamicTostatic_bs8_fp32_DP_N1C8
+# bash test_tipc/benchmark_train.sh test_tipc/configs/det_mv3_db_v2_0/train_benchmark.txt  benchmark_train dynamic_bs8_null_DP_N1C1
+IFS=$'\n'
+# parser params from train_benchmark.txt
+dataline=`cat $FILENAME`
+# parser params
+IFS=$'\n'
+lines=(${dataline})
+model_name=$(func_parser_value "${lines[1]}")
+python_name=$(func_parser_value "${lines[2]}")
+
+# set env
+python=python
+export str_tmp=$(echo `pip list|grep paddlepaddle-gpu|awk -F ' ' '{print $2}'`)
+export frame_version=${str_tmp%%.post*}
+export frame_commit=$(echo `${python} -c "import paddle;print(paddle.version.commit)"`)
+
+# 获取benchmark_params所在的行数
+line_num=`grep -n -w "train_benchmark_params" $FILENAME  | cut -d ":" -f 1`
+# for train log parser
+batch_size=$(func_parser_value "${lines[line_num]}")
+line_num=`expr $line_num + 1`
+fp_items=$(func_parser_value "${lines[line_num]}")
+line_num=`expr $line_num + 1`
+epoch=$(func_parser_value "${lines[line_num]}")
+
+line_num=`expr $line_num + 1`
+profile_option_key=$(func_parser_key "${lines[line_num]}")
+profile_option_params=$(func_parser_value "${lines[line_num]}")
+profile_option="${profile_option_key}:${profile_option_params}"
+
+line_num=`expr $line_num + 1`
+flags_value=$(func_parser_value "${lines[line_num]}")
+# set flags
+IFS=";"
+flags_list=(${flags_value})
+for _flag in ${flags_list[*]}; do
+    cmd="export ${_flag}"
+    eval $cmd
+done
+
+# set log_name
+repo_name=$(get_repo_name )
+SAVE_LOG=${BENCHMARK_LOG_DIR:-$(pwd)}   # */benchmark_log
+mkdir -p "${SAVE_LOG}/benchmark_log/"
+status_log="${SAVE_LOG}/benchmark_log/results.log"
+# get benchmark profiling params : PROFILING_TIMER_ONLY=no|True|False
+PROFILING_TIMER_ONLY=${PROFILING_TIMER_ONLY:-"True"}
+# The number of lines in which train params can be replaced.
+line_python=3
+line_gpuid=4
+line_precision=6
+line_epoch=7
+line_batchsize=9
+line_profile=13
+line_eval_py=24
+line_export_py=30
+
+func_sed_params "$FILENAME" "${line_eval_py}" "null"
+func_sed_params "$FILENAME" "${line_export_py}" "null"
+func_sed_params "$FILENAME" "${line_python}"  "$python"
+
+# if params
+if  [ ! -n "$PARAMS" ] ;then
+    # PARAMS input is not a word.
+    IFS="|"
+    batch_size_list=(${batch_size})
+    fp_items_list=(${fp_items})
+    device_num_list=(N1C4)
+    run_mode="DP"
+elif [[ ${PARAMS} = "dynamicTostatic" ]];then
+    IFS="|"
+    model_type=$PARAMS
+    batch_size_list=(${batch_size})
+    fp_items_list=(${fp_items})
+    device_num_list=(N1C4)
+    run_mode="DP"
+else
+    # parser params from input: modeltype_bs${bs_item}_${fp_item}_${run_mode}_${device_num}
+    IFS="_"
+    params_list=(${PARAMS})
+    model_type=${params_list[0]}
+    batch_size=${params_list[1]}
+    batch_size=`echo  ${batch_size} | tr -cd "[0-9]" `
+    precision=${params_list[2]}
+    run_mode=${params_list[3]}
+    device_num=${params_list[4]}
+    IFS=";"
+
+    if [ ${precision} = "fp16" ];then
+        precision="amp"
+    fi
+
+    epoch=$(set_dynamic_epoch $device_num $epoch)
+    fp_items_list=($precision)
+    batch_size_list=($batch_size)
+    device_num_list=($device_num)
+fi
+
+IFS="|"
+for batch_size in ${batch_size_list[*]}; do 
+    for train_precision in ${fp_items_list[*]}; do
+        for device_num in ${device_num_list[*]}; do
+            # sed batchsize and precision
+            if [ ${train_precision} = "amp" ];then
+                precision="fp16"
+            else
+                precision="fp32"
+            fi
+
+            func_sed_params "$FILENAME" "${line_precision}" "$train_precision"
+            func_sed_params "$FILENAME" "${line_batchsize}" "$MODE=$batch_size"
+            func_sed_params "$FILENAME" "${line_epoch}" "$MODE=$epoch"
+            gpu_id=$(set_gpu_id $device_num)
+
+            if [ ${#gpu_id} -le 1 ];then
+                func_sed_params "$FILENAME" "${line_gpuid}" "0"  # sed used gpu_id 
+                if [[ ${PROFILING_TIMER_ONLY} != "no" ]];then
+                    echo "run profile"
+                    # The default value of profile_option's timer_only parameter is True
+                    if [[ ${PROFILING_TIMER_ONLY} = "False" ]];then
+                        profile_option="${profile_option};timer_only=False"
+                    fi
+                    log_path="$SAVE_LOG/profiling_log"
+                    mkdir -p $log_path
+                    log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_mode}_${device_num}_${to_static}profiling"
+                    # set profile_option params
+                    tmp=`sed -i "${line_profile}s/.*/\"${profile_option}\"/" "${FILENAME}"`
+                    # run test_train_inference_python.sh
+                    cmd="timeout 5m bash test_tipc/test_train_inference_python.sh ${FILENAME} benchmark_train > ${log_path}/${log_name} 2>&1 "
+                    echo $cmd
+                    eval ${cmd}
+                    eval "cat ${log_path}/${log_name}"
+                fi
+                echo "run without profile"
+                # without profile
+                log_path="$SAVE_LOG/train_log"
+                speed_log_path="$SAVE_LOG/index"
+                mkdir -p $log_path
+                mkdir -p $speed_log_path
+                log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_mode}_${device_num}_${to_static}log"
+                speed_log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_mode}_${device_num}_${to_static}speed"
+                func_sed_params "$FILENAME" "${line_profile}" "null"  # sed profile_id as null
+                cmd="bash test_tipc/test_train_inference_python.sh ${FILENAME} benchmark_train > ${log_path}/${log_name} 2>&1 "
+                echo $cmd
+                job_bt=`date '+%Y%m%d%H%M%S'`
+                eval $cmd
+                job_et=`date '+%Y%m%d%H%M%S'`
+                export model_run_time=$((${job_et}-${job_bt}))
+                eval "cat ${log_path}/${log_name}"
+
+                # parser log
+                _model_name="${model_name}_bs${batch_size}_${precision}_${run_mode}"
+                cmd="${python} ${BENCHMARK_ROOT}/scripts/analysis.py --filename ${log_path}/${log_name} \
+                        --speed_log_file '${speed_log_path}/${speed_log_name}' \
+                        --model_name ${_model_name} \
+                        --base_batch_size ${batch_size} \
+                        --run_mode ${run_mode} \
+                        --fp_item ${precision} \
+                        --keyword ips: \
+                        --skip_steps 2 \
+                        --device_num ${device_num} \
+                        --speed_unit samples/s \
+                        --convergence_key loss: "
+                echo $cmd
+                eval $cmd
+                last_status=${PIPESTATUS[0]}
+                status_check $last_status "${cmd}" "${status_log}"
+            else
+                IFS=";"
+                unset_env=`unset CUDA_VISIBLE_DEVICES`
+                log_path="$SAVE_LOG/train_log"
+                speed_log_path="$SAVE_LOG/index"
+                mkdir -p $log_path
+                mkdir -p $speed_log_path
+                log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_mode}_${device_num}_${to_static}log"
+                speed_log_name="${repo_name}_${model_name}_bs${batch_size}_${precision}_${run_mode}_${device_num}_${to_static}speed"
+                func_sed_params "$FILENAME" "${line_gpuid}" "$gpu_id"  # sed used gpu_id 
+                func_sed_params "$FILENAME" "${line_profile}" "null"  # sed --profile_option as null
+                cmd="bash test_tipc/test_train_inference_python.sh ${FILENAME} benchmark_train > ${log_path}/${log_name} 2>&1 "
+                echo $cmd
+                job_bt=`date '+%Y%m%d%H%M%S'`
+                eval $cmd
+                job_et=`date '+%Y%m%d%H%M%S'`
+                export model_run_time=$((${job_et}-${job_bt}))
+                eval "cat ${log_path}/${log_name}"
+                # parser log
+                _model_name="${model_name}_bs${batch_size}_${precision}_${run_mode}"
+                
+                cmd="${python} ${BENCHMARK_ROOT}/scripts/analysis.py --filename ${log_path}/${log_name} \
+                        --speed_log_file '${speed_log_path}/${speed_log_name}' \
+                        --model_name ${_model_name} \
+                        --base_batch_size ${batch_size} \
+                        --run_mode ${run_mode} \
+                        --fp_item ${precision} \
+                        --keyword ips: \
+                        --skip_steps 2 \
+                        --device_num ${device_num} \
+                        --speed_unit images/s \
+                        --convergence_key loss: "
+                echo $cmd
+                eval $cmd
+                last_status=${PIPESTATUS[0]}
+                status_check $last_status "${cmd}" "${status_log}"
+            fi
+        done
+    done
+done
diff --git a/test_tipc/build_server.sh b/test_tipc/build_server.sh
new file mode 100644
index 0000000..36d9789
--- /dev/null
+++ b/test_tipc/build_server.sh
@@ -0,0 +1,69 @@
+#使用镜像：
+#registry.baidubce.com/paddlepaddle/paddle:latest-dev-cuda10.1-cudnn7-gcc82
+
+#编译Serving Server：
+
+#client和app可以直接使用release版本
+
+#server因为加入了自定义OP，需要重新编译
+
+apt-get update
+apt install -y libcurl4-openssl-dev libbz2-dev
+wget https://paddle-serving.bj.bcebos.com/others/centos_ssl.tar && tar xf centos_ssl.tar && rm -rf centos_ssl.tar && mv libcrypto.so.1.0.2k /usr/lib/libcrypto.so.1.0.2k && mv libssl.so.1.0.2k /usr/lib/libssl.so.1.0.2k && ln -sf /usr/lib/libcrypto.so.1.0.2k /usr/lib/libcrypto.so.10 && ln -sf /usr/lib/libssl.so.1.0.2k /usr/lib/libssl.so.10 && ln -sf /usr/lib/libcrypto.so.10 /usr/lib/libcrypto.so && ln -sf /usr/lib/libssl.so.10 /usr/lib/libssl.so
+
+# 安装go依赖
+rm -rf /usr/local/go
+wget -qO- https://paddle-ci.cdn.bcebos.com/go1.17.2.linux-amd64.tar.gz | tar -xz -C /usr/local
+export GOROOT=/usr/local/go
+export GOPATH=/root/gopath
+export PATH=$PATH:$GOPATH/bin:$GOROOT/bin
+go env -w GO111MODULE=on
+go env -w GOPROXY=https://goproxy.cn,direct
+go install github.com/grpc-ecosystem/grpc-gateway/protoc-gen-grpc-gateway@v1.15.2
+go install github.com/grpc-ecosystem/grpc-gateway/protoc-gen-swagger@v1.15.2
+go install github.com/golang/protobuf/protoc-gen-go@v1.4.3
+go install google.golang.org/grpc@v1.33.0
+go env -w GO111MODULE=auto
+
+# 下载opencv库
+wget https://paddle-qa.bj.bcebos.com/PaddleServing/opencv3.tar.gz && tar -xvf opencv3.tar.gz && rm -rf opencv3.tar.gz
+export OPENCV_DIR=$PWD/opencv3
+
+# clone Serving
+git clone https://github.com/PaddlePaddle/Serving.git -b develop --depth=1
+cd Serving
+export Serving_repo_path=$PWD
+git submodule update --init --recursive
+python -m pip install -r python/requirements.txt
+
+
+export PYTHON_INCLUDE_DIR=$(python -c "import sysconfig; print(sysconfig.get_path('include'))")
+export PYTHON_LIBRARIES=$(python -c "import sysconfig; print(sysconfig.get_config_var('LIBDIR'))")
+export PYTHON_EXECUTABLE=`which python`
+
+export CUDA_PATH='/usr/local/cuda'
+export CUDNN_LIBRARY='/usr/local/cuda/lib64/'
+export CUDA_CUDART_LIBRARY='/usr/local/cuda/lib64/'
+export TENSORRT_LIBRARY_PATH='/usr/local/TensorRT6-cuda10.1-cudnn7/targets/x86_64-linux-gnu/'
+
+# cp 自定义OP代码
+cp -rf ../deploy/pdserving/general_detection_op.cpp ${Serving_repo_path}/core/general-server/op
+
+# 编译Server, export SERVING_BIN
+mkdir server-build-gpu-opencv && cd server-build-gpu-opencv
+cmake -DPYTHON_INCLUDE_DIR=$PYTHON_INCLUDE_DIR \
+            -DPYTHON_LIBRARIES=$PYTHON_LIBRARIES \
+            -DPYTHON_EXECUTABLE=$PYTHON_EXECUTABLE \
+            -DCUDA_TOOLKIT_ROOT_DIR=${CUDA_PATH} \
+            -DCUDNN_LIBRARY=${CUDNN_LIBRARY} \
+            -DCUDA_CUDART_LIBRARY=${CUDA_CUDART_LIBRARY} \
+            -DTENSORRT_ROOT=${TENSORRT_LIBRARY_PATH} \
+            -DOPENCV_DIR=${OPENCV_DIR} \
+            -DWITH_OPENCV=ON \
+            -DSERVER=ON \
+            -DWITH_GPU=ON ..
+make -j32
+
+python -m pip install python/dist/paddle*
+export SERVING_BIN=$PWD/core/general-server/serving
+cd ../../
diff --git a/test_tipc/common_func.sh b/test_tipc/common_func.sh
new file mode 100644
index 0000000..9ec22f0
--- /dev/null
+++ b/test_tipc/common_func.sh
@@ -0,0 +1,67 @@
+#!/bin/bash
+
+function func_parser_key(){
+    strs=$1
+    IFS=":"
+    array=(${strs})
+    tmp=${array[0]}
+    echo ${tmp}
+}
+
+function func_parser_value(){
+    strs=$1
+    IFS=":"
+    array=(${strs})
+    tmp=${array[1]}
+    echo ${tmp}
+}
+
+function func_set_params(){
+    key=$1
+    value=$2
+    if [ ${key}x = "null"x ];then
+        echo " "
+    elif [[ ${value} = "null" ]] || [[ ${value} = " " ]] || [ ${#value} -le 0 ];then
+        echo " "
+    else 
+        echo "${key}=${value}"
+    fi
+}
+
+function func_parser_params(){
+    strs=$1
+    MODE=$2
+    IFS=":"
+    array=(${strs})
+    key=${array[0]}
+    tmp=${array[1]}
+    IFS="|"
+    res=""
+    for _params in ${tmp[*]}; do
+        IFS="="
+        array=(${_params})
+        mode=${array[0]}
+        value=${array[1]}
+        if [[ ${mode} = ${MODE} ]]; then
+            IFS="|"
+            #echo $(func_set_params "${mode}" "${value}")
+            echo $value
+            break
+        fi
+        IFS="|"
+    done
+    echo ${res}
+}
+
+function status_check(){
+    last_status=$1   # the exit code
+    run_command=$2
+    run_log=$3
+    model_name=$4
+    log_path=$5
+    if [ $last_status -eq 0 ]; then
+        echo -e "\033[33m Run successfully with command - ${model_name} - ${run_command} - ${log_path} \033[0m" | tee -a ${run_log}
+    else
+        echo -e "\033[33m Run failed with command - ${model_name} - ${run_command} - ${log_path} \033[0m" | tee -a ${run_log}
+    fi
+}
diff --git a/test_tipc/compare_results.py b/test_tipc/compare_results.py
new file mode 100644
index 0000000..96a1f20
--- /dev/null
+++ b/test_tipc/compare_results.py
@@ -0,0 +1,145 @@
+import numpy as np
+import os
+import subprocess
+import json
+import argparse
+import glob
+
+
+def init_args():
+    parser = argparse.ArgumentParser()
+    # params for testing assert allclose
+    parser.add_argument("--atol", type=float, default=1e-3)
+    parser.add_argument("--rtol", type=float, default=1e-3)
+    parser.add_argument("--gt_file", type=str, default="")
+    parser.add_argument("--log_file", type=str, default="")
+    parser.add_argument("--precision", type=str, default="fp32")
+    return parser
+
+
+def parse_args():
+    parser = init_args()
+    return parser.parse_args()
+
+
+def run_shell_command(cmd):
+    p = subprocess.Popen(
+        cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, shell=True
+    )
+    out, err = p.communicate()
+
+    if p.returncode == 0:
+        return out.decode("utf-8")
+    else:
+        return None
+
+
+def parser_results_from_log_by_name(log_path, names_list):
+    if not os.path.exists(log_path):
+        raise ValueError("The log file {} does not exists!".format(log_path))
+
+    if names_list is None or len(names_list) < 1:
+        return []
+
+    parser_results = {}
+    for name in names_list:
+        cmd = "grep {} {}".format(name, log_path)
+        outs = run_shell_command(cmd)
+        outs = outs.split("\n")[0]
+        result = outs.split("{}".format(name))[-1]
+        try:
+            result = json.loads(result)
+        except:
+            result = np.array([int(r) for r in result.split()]).reshape(-1, 4)
+        parser_results[name] = result
+    return parser_results
+
+
+def load_gt_from_file(gt_file):
+    if not os.path.exists(gt_file):
+        raise ValueError("The log file {} does not exists!".format(gt_file))
+    with open(gt_file, "r") as f:
+        data = f.readlines()
+        f.close()
+    parser_gt = {}
+    for line in data:
+        image_name, result = line.strip("\n").split("\t")
+        image_name = image_name.split("/")[-1]
+        try:
+            result = json.loads(result)
+        except:
+            result = np.array([int(r) for r in result.split()]).reshape(-1, 4)
+        parser_gt[image_name] = result
+    return parser_gt
+
+
+def load_gt_from_txts(gt_file):
+    gt_list = glob.glob(gt_file)
+    gt_collection = {}
+    for gt_f in gt_list:
+        gt_dict = load_gt_from_file(gt_f)
+        basename = os.path.basename(gt_f)
+        if "fp32" in basename:
+            gt_collection["fp32"] = [gt_dict, gt_f]
+        elif "fp16" in basename:
+            gt_collection["fp16"] = [gt_dict, gt_f]
+        elif "int8" in basename:
+            gt_collection["int8"] = [gt_dict, gt_f]
+        else:
+            continue
+    return gt_collection
+
+
+def collect_predict_from_logs(log_path, key_list):
+    log_list = glob.glob(log_path)
+    pred_collection = {}
+    for log_f in log_list:
+        pred_dict = parser_results_from_log_by_name(log_f, key_list)
+        key = os.path.basename(log_f)
+        pred_collection[key] = pred_dict
+
+    return pred_collection
+
+
+def testing_assert_allclose(dict_x, dict_y, atol=1e-7, rtol=1e-7):
+    for k in dict_x:
+        np.testing.assert_allclose(
+            np.array(dict_x[k]), np.array(dict_y[k]), atol=atol, rtol=rtol
+        )
+
+
+if __name__ == "__main__":
+    # Usage:
+    # python3.7 tests/compare_results.py --gt_file=./tests/results/*.txt  --log_file=./tests/output/infer_*.log
+
+    args = parse_args()
+
+    gt_collection = load_gt_from_txts(args.gt_file)
+    key_list = gt_collection["fp32"][0].keys()
+
+    pred_collection = collect_predict_from_logs(args.log_file, key_list)
+    for filename in pred_collection.keys():
+        if "fp32" in filename:
+            gt_dict, gt_filename = gt_collection["fp32"]
+        elif "fp16" in filename:
+            gt_dict, gt_filename = gt_collection["fp16"]
+        elif "int8" in filename:
+            gt_dict, gt_filename = gt_collection["int8"]
+        else:
+            continue
+        pred_dict = pred_collection[filename]
+
+        try:
+            testing_assert_allclose(gt_dict, pred_dict, atol=args.atol, rtol=args.rtol)
+            print(
+                "Assert allclose passed! The results of {} and {} are consistent!".format(
+                    filename, gt_filename
+                )
+            )
+        except Exception as E:
+            print(E)
+            raise ValueError(
+                "The results of {} and the results of {} are inconsistent!".format(
+                    filename, gt_filename
+                )
+            )
diff --git a/test_tipc/configs/PP-OCRv3_mobile_rec/PP-OCRv3_mobile_rec_distillation.yml b/test_tipc/configs/PP-OCRv3_mobile_rec/PP-OCRv3_mobile_rec_distillation.yml
new file mode 100644
index 0000000..6336213
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv3_mobile_rec/PP-OCRv3_mobile_rec_distillation.yml
@@ -0,0 +1,206 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 800
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_ppocr_v3_distillation
+  save_epoch_step: 3
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: &max_text_length 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_ppocrv3_distillation.txt
+  d2s_train_image_shape: [3, 48, -1]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Piecewise
+    decay_epochs : [700]
+    values : [0.0005, 0.00005]
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+
+Architecture:
+  model_type: &model_type "rec"
+  name: DistillationModel
+  algorithm: Distillation
+  Models:
+    Teacher:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: SVTR_LCNet
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+        last_conv_stride: [1, 2]
+        last_pool_type: avg
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 64
+                depth: 2
+                hidden_dims: 120
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - SARHead:
+              enc_dim: 512
+              max_text_length: *max_text_length
+    Student:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: SVTR_LCNet
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+        last_conv_stride: [1, 2]
+        last_pool_type: avg
+      Head:
+        name: MultiHead
+        head_list:
+          - CTCHead:
+              Neck:
+                name: svtr
+                dims: 64
+                depth: 2
+                hidden_dims: 120
+                use_guide: True
+              Head:
+                fc_decay: 0.00001
+          - SARHead:
+              enc_dim: 512
+              max_text_length: *max_text_length
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationDMLLoss:
+      weight: 1.0
+      act: "softmax"
+      use_log: true
+      model_name_pairs:
+      - ["Student", "Teacher"]
+      key: head_out
+      multi_head: True
+      dis_head: ctc
+      name: dml_ctc
+  - DistillationDMLLoss:
+      weight: 0.5
+      act: "softmax"
+      use_log: true
+      model_name_pairs:
+      - ["Student", "Teacher"]
+      key: head_out
+      multi_head: True
+      dis_head: sar
+      name: dml_sar
+  - DistillationDistanceLoss:
+      weight: 1.0
+      mode: "l2"
+      model_name_pairs:
+      - ["Student", "Teacher"]
+      key: backbone_out
+  - DistillationCTCLoss:
+      weight: 1.0
+      model_name_list: ["Student", "Teacher"]
+      key: head_out
+      multi_head: True
+  - DistillationSARLoss:
+      weight: 1.0
+      model_name_list: ["Student", "Teacher"]
+      key: head_out
+      multi_head: True
+
+PostProcess:
+  name: DistillationCTCLabelDecode
+  model_name: ["Student", "Teacher"]
+  key: head_out
+  multi_head: True
+
+Metric:
+  name: DistillationMetric
+  base_metric_name: RecMetric
+  main_indicator: acc
+  key: "Student"
+  ignore_space: True
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    ext_op_transform_idx: 1
+    label_file_list:
+    - ./train_data/ic15_data/rec_gt_train.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecConAug:
+        prob: 0.5
+        ext_data_num: 2
+        image_shape: [48, 320, 3]
+    - RecAug:
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_workers: 4
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list:
+    - ./train_data/ic15_data/rec_gt_test.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - MultiLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 48, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label_ctc
+        - label_sar
+        - length
+        - valid_ratio
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 4
diff --git a/test_tipc/configs/PP-OCRv3_mobile_rec/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/PP-OCRv3_mobile_rec/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..4071cbe
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv3_mobile_rec/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:PP-OCRv3_mobile_rec
+use_opencv:True
+infer_model:./inference/PP-OCRv3_mobile_rec_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr --rec_img_h=48 --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference/
+null:null
+--benchmark:True
+--det:False
+--rec:True
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/PP-OCRv3_mobile_rec/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt b/test_tipc/configs/PP-OCRv3_mobile_rec/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
new file mode 100644
index 0000000..1b946f3
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv3_mobile_rec/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
@@ -0,0 +1,17 @@
+===========================paddle2onnx_params===========================
+model_name:PP-OCRv3_mobile_rec
+python:python3.7
+2onnx: paddle2onnx
+--det_model_dir:
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_save_file:
+--rec_model_dir:./inference/PP-OCRv3_mobile_rec_infer/
+--rec_save_file:./inference/rec_v3_onnx/model.onnx
+--opset_version:10
+--enable_onnx_checker:True
+inference:tools/infer/predict_rec.py --rec_image_shape="3,48,320"
+--use_gpu:True|False
+--det_model_dir:
+--rec_model_dir:
+--image_dir:./inference/rec_inference/
diff --git a/test_tipc/configs/PP-OCRv3_mobile_rec/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/PP-OCRv3_mobile_rec/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..a22da5d
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv3_mobile_rec/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:PP-OCRv3_mobile_rec
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:null
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:null
+--det_serving_client:null
+--rec_dirname:./inference/PP-OCRv3_mobile_rec_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v3_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v3_client/
+serving_dir:./deploy/pdserving
+web_service:web_service_rec.py --config=config.yml --opt op.rec.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py --det=False
+--image_dir:../../inference/rec_inference
diff --git a/test_tipc/configs/PP-OCRv3_mobile_rec/train_infer_python.txt b/test_tipc/configs/PP-OCRv3_mobile_rec/train_infer_python.txt
new file mode 100644
index 0000000..7899862
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv3_mobile_rec/train_infer_python.txt
@@ -0,0 +1,61 @@
+===========================train_params===========================
+model_name:PP-OCRv3_mobile_rec
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/PP-OCRv3_mobile_rec/PP-OCRv3_mobile_rec_distillation.yml -o Global.print_batch_step=1 Train.loader.shuffle=false
+pact_train:null
+fpgm_train:null
+distill_train:null
+to_static_train:Global.to_static=true
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/PP-OCRv3_mobile_rec/PP-OCRv3_mobile_rec_distillation.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/PP-OCRv3_mobile_rec_infer
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_image_shape="3,48,320"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,48,320]}]
+===========================train_benchmark_params==========================
+batch_size:64
+fp_items:fp32|fp16
+epoch:1
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
+===========================to_static_train_benchmark_params===========================
+to_static_train:Global.to_static=true
diff --git a/test_tipc/configs/PP-OCRv3_mobile_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/PP-OCRv3_mobile_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..4b7e7cb
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv3_mobile_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:PP-OCRv3_mobile_rec
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/PP-OCRv3_mobile_rec/PP-OCRv3_mobile_rec_distillation.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/PP-OCRv3_mobile_rec/PP-OCRv3_mobile_rec_distillation.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/PP-OCRv3_mobile_rec_infer
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_image_shape="3,48,320"
+--use_gpu:False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,48,320]}]
diff --git a/test_tipc/configs/PP-OCRv3_mobile_rec/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/PP-OCRv3_mobile_rec/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..8e7feef
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv3_mobile_rec/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:PP-OCRv3_mobile_rec
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/PP-OCRv3_mobile_rec/PP-OCRv3_mobile_rec_distillation.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/PP-OCRv3_mobile_rec/PP-OCRv3_mobile_rec_distillation.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/PP-OCRv3_mobile_rec_infer
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_image_shape="3,48,320"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,48,320]}]
diff --git a/test_tipc/configs/PP-OCRv3_mobile_rec/train_pact_infer_python.txt b/test_tipc/configs/PP-OCRv3_mobile_rec/train_pact_infer_python.txt
new file mode 100644
index 0000000..6239c95
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv3_mobile_rec/train_pact_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:PP-OCRv3_mobile_rec_PACT
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:pretrain_models/PP-OCRv3_mobile_rec_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:pact_train
+norm_train:null
+pact_train:deploy/slim/quantization/quant.py -c test_tipc/configs/PP-OCRv3_mobile_rec/PP-OCRv3_mobile_rec_distillation.yml -o 
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null
+quant_export:deploy/slim/quantization/export_model.py -c test_tipc/configs/PP-OCRv3_mobile_rec/PP-OCRv3_mobile_rec_distillation.yml -o 
+fpgm_export: null
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/PP-OCRv3_mobile_rec_slim_infer
+infer_export:null
+infer_quant:True
+inference:tools/infer/predict_rec.py --rec_image_shape="3,48,320"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,48,320]}]
diff --git a/test_tipc/configs/PP-OCRv3_mobile_rec/train_ptq_infer_python.txt b/test_tipc/configs/PP-OCRv3_mobile_rec/train_ptq_infer_python.txt
new file mode 100644
index 0000000..3aa42cb
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv3_mobile_rec/train_ptq_infer_python.txt
@@ -0,0 +1,21 @@
+===========================kl_quant_params===========================
+model_name:PP-OCRv3_mobile_rec_KL
+python:python3.7
+Global.pretrained_model:
+Global.save_inference_dir:null
+infer_model:./inference/PP-OCRv3_mobile_rec_infer/
+infer_export:deploy/slim/quantization/quant_kl.py -c test_tipc/configs/PP-OCRv3_mobile_rec/PP-OCRv3_mobile_rec_distillation.yml -o
+infer_quant:True
+inference:tools/infer/predict_rec.py --rec_image_shape="3,48,320"
+--use_gpu:False|True
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:int8
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+null:null
diff --git a/test_tipc/configs/PP-OCRv3_mobile_rec_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/PP-OCRv3_mobile_rec_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..9ae0023
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv3_mobile_rec_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:PP-OCRv3_mobile_rec_KL
+use_opencv:True
+infer_model:./inference/PP-OCRv3_mobile_rec_klquant_infer
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr --rec_img_h=48 --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference/
+null:null
+--benchmark:True
+--det:False
+--rec:True
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/PP-OCRv3_mobile_rec_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt b/test_tipc/configs/PP-OCRv3_mobile_rec_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..d418f49
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv3_mobile_rec_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================serving_params===========================
+model_name:PP-OCRv3_mobile_rec_KL
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_PP-OCRv3_det_klquant_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v3_kl_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v3_kl_client/
+--rec_dirname:./inference/PP-OCRv3_mobile_rec_klquant_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v3_kl_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v3_kl_client/
+serving_dir:./deploy/pdserving
+web_service:-m paddle_serving_server.serve
+--op:GeneralDetectionOp GeneralInferOp
+--port:8181
+--gpu_id:"0"|null
+cpp_client:ocr_cpp_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/PP-OCRv3_mobile_rec_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/PP-OCRv3_mobile_rec_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..92546b1
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv3_mobile_rec_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:PP-OCRv3_mobile_rec_KL
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:null
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:null
+--det_serving_client:null
+--rec_dirname:./inference/PP-OCRv3_mobile_rec_klquant_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v3_kl_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v3_kl_client/
+serving_dir:./deploy/pdserving
+web_service:web_service_rec.py --config=config.yml --opt op.rec.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py --det=False
+--image_dir:../../inference/rec_inference
diff --git a/test_tipc/configs/PP-OCRv3_mobile_rec_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/PP-OCRv3_mobile_rec_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..f325dd3
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv3_mobile_rec_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:PP-OCRv3_mobile_rec_PACT
+use_opencv:True
+infer_model:./inference/PP-OCRv3_mobile_rec_pact_infer
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr --rec_img_h=48 --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference/
+null:null
+--benchmark:True
+--det:False
+--rec:True
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/PP-OCRv3_mobile_rec_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt b/test_tipc/configs/PP-OCRv3_mobile_rec_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..1108d59
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv3_mobile_rec_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================serving_params===========================
+model_name:PP-OCRv3_mobile_rec_PACT
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_PP-OCRv3_det_pact_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v3_pact_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v3_pact_client/
+--rec_dirname:./inference/PP-OCRv3_mobile_rec_pact_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v3_pact_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v3_pact_client/
+serving_dir:./deploy/pdserving
+web_service:-m paddle_serving_server.serve
+--op:GeneralDetectionOp GeneralInferOp
+--port:8181
+--gpu_id:"0"|null
+cpp_client:ocr_cpp_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/PP-OCRv3_mobile_rec_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/PP-OCRv3_mobile_rec_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..948eac8
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv3_mobile_rec_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:PP-OCRv3_mobile_rec_PACT
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:null
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:null
+--det_serving_client:null
+--rec_dirname:./inference/PP-OCRv3_mobile_rec_pact_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v3_pact_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v3_pact_client/
+serving_dir:./deploy/pdserving
+web_service:web_service_rec.py --config=config.yml --opt op.rec.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py --det=False
+--image_dir:../../inference/rec_inference
diff --git a/test_tipc/configs/PP-OCRv3_mobile_rec_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/PP-OCRv3_mobile_rec_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..eda0833
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv3_mobile_rec_PACT/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:PP-OCRv3_mobile_rec_PACT
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:pretrain_models/PP-OCRv3_mobile_rec_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:pact_train
+norm_train:null
+pact_train:deploy/slim/quantization/quant.py -c test_tipc/configs/PP-OCRv3_mobile_rec/PP-OCRv3_mobile_rec_distillation.yml -o 
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null
+quant_export:deploy/slim/quantization/export_model.py -c test_tipc/configs/PP-OCRv3_mobile_rec/PP-OCRv3_mobile_rec_distillation.yml -o 
+fpgm_export: null
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/PP-OCRv3_mobile_rec_slim_quant_infer
+infer_export:null
+infer_quant:True
+inference:tools/infer/predict_rec.py --rec_image_shape="3,48,320"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,48,320]}]
diff --git a/test_tipc/configs/PP-OCRv4_mobile_rec/train_infer_python.txt b/test_tipc/configs/PP-OCRv4_mobile_rec/train_infer_python.txt
new file mode 100644
index 0000000..3730a5e
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv4_mobile_rec/train_infer_python.txt
@@ -0,0 +1,59 @@
+===========================train_params===========================
+model_name:PP-OCRv4_mobile_rec
+python:python
+gpu_list:0
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec.yml -o Global.cal_metric_during_train=False Global.print_batch_step=1 Train.loader.shuffle=false Train.dataset.data_dir=./train_data/ic15_data Train.dataset.label_file_list=[./train_data/ic15_data/rec_gt_train.txt] Eval.dataset.data_dir=./train_data/ic15_data Eval.dataset.label_file_list=[./train_data/ic15_data/rec_gt_test.txt] Train.loader.num_workers=16 Eval.loader.num_workers=16
+pact_train:null
+fpgm_train:null
+distill_train:null
+to_static_train:Global.to_static=true
+null:null
+##
+===========================eval_params===========================
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec.yml -o
+quant_export:
+fpgm_export:
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/PP-OCRv4_mobile_rec_infer
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_image_shape="3,48,320"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,48,320]}]
+===========================train_benchmark_params==========================
+batch_size:128
+fp_items:fp32|fp16
+epoch:1
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
diff --git a/test_tipc/configs/PP-OCRv4_mobile_rec_ampO2_ultra/train_infer_python.txt b/test_tipc/configs/PP-OCRv4_mobile_rec_ampO2_ultra/train_infer_python.txt
new file mode 100644
index 0000000..909f390
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv4_mobile_rec_ampO2_ultra/train_infer_python.txt
@@ -0,0 +1,61 @@
+===========================train_params===========================
+model_name:PP-OCRv4_mobile_rec
+python:python
+gpu_list:0
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec_ampO2_ultra.yml -o Global.cal_metric_during_train=False Global.print_batch_step=1 Train.loader.shuffle=false Train.dataset.data_dir=./train_data/ic15_data Train.dataset.label_file_list=[./train_data/ic15_data/rec_gt_train.txt] Eval.dataset.data_dir=./train_data/ic15_data Eval.dataset.label_file_list=[./train_data/ic15_data/rec_gt_test.txt]
+pact_train:null
+fpgm_train:null
+distill_train:null
+to_static_train:Global.to_static=true
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/PP-OCRv4_mobile_rec_infer
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_image_shape="3,48,320"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,48,320]}]
+===========================train_benchmark_params==========================
+batch_size:384
+fp_items:fp16
+epoch:1
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
+===========================disable_to_static_train_benchmark===========================
+to_static_train:Global.to_static=False
diff --git a/test_tipc/configs/PP-OCRv4_mobile_rec_fp32_ultra/train_infer_python.txt b/test_tipc/configs/PP-OCRv4_mobile_rec_fp32_ultra/train_infer_python.txt
new file mode 100644
index 0000000..eda3409
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv4_mobile_rec_fp32_ultra/train_infer_python.txt
@@ -0,0 +1,61 @@
+===========================train_params===========================
+model_name:PP-OCRv4_mobile_rec
+python:python
+gpu_list:0
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec_fp32_ultra.yml -o Global.cal_metric_during_train=False Global.print_batch_step=1 Train.loader.shuffle=false Train.dataset.data_dir=./train_data/ic15_data Train.dataset.label_file_list=[./train_data/ic15_data/rec_gt_train.txt] Eval.dataset.data_dir=./train_data/ic15_data Eval.dataset.label_file_list=[./train_data/ic15_data/rec_gt_test.txt]
+pact_train:null
+fpgm_train:null
+distill_train:null
+to_static_train:Global.to_static=true
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/PP-OCRv4_mobile_rec_infer
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_image_shape="3,48,320"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,48,320]}]
+===========================train_benchmark_params==========================
+batch_size:192
+fp_items:fp32
+epoch:1
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
+===========================disable_to_static_train_benchmark===========================
+to_static_train:Global.to_static=False
diff --git a/test_tipc/configs/PP-OCRv4_server_rec/train_infer_python.txt b/test_tipc/configs/PP-OCRv4_server_rec/train_infer_python.txt
new file mode 100644
index 0000000..8f12c5d
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv4_server_rec/train_infer_python.txt
@@ -0,0 +1,59 @@
+===========================train_params===========================
+model_name:PP-OCRv4_server_rec
+python:python
+gpu_list:0
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/rec/PP-OCRv4/PP-OCRv4_server_rec.yml -o Global.cal_metric_during_train=False Global.print_batch_step=1 Train.loader.shuffle=false Train.dataset.data_dir=./train_data/ic15_data Train.dataset.label_file_list=[./train_data/ic15_data/rec_gt_train.txt] Eval.dataset.data_dir=./train_data/ic15_data Eval.dataset.label_file_list=[./train_data/ic15_data/rec_gt_test.txt] Train.loader.num_workers=16 Eval.loader.num_workers=16
+pact_train:null
+fpgm_train:null
+distill_train:null
+to_static_train:Global.to_static=true
+null:null
+##
+===========================eval_params===========================
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/rec/PP-OCRv4/PP-OCRv4_server_rec.yml -o
+quant_export:
+fpgm_export:
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/PP-OCRv4_mobile_rec_infer
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_image_shape="3,48,320"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,48,320]}]
+===========================train_benchmark_params==========================
+batch_size:128
+fp_items:fp32|fp16
+epoch:1
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
diff --git a/test_tipc/configs/PP-OCRv4_server_rec_ampO2_ultra/train_infer_python.txt b/test_tipc/configs/PP-OCRv4_server_rec_ampO2_ultra/train_infer_python.txt
new file mode 100644
index 0000000..0789305
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv4_server_rec_ampO2_ultra/train_infer_python.txt
@@ -0,0 +1,60 @@
+===========================train_params===========================
+model_name:PP-OCRv4_server_rec
+python:python
+gpu_list:0
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/rec/PP-OCRv4/PP-OCRv4_server_rec_ampO2_ultra.yml -o Global.cal_metric_during_train=False Global.print_batch_step=1 Train.loader.shuffle=false Train.dataset.data_dir=./train_data/ic15_data Train.dataset.label_file_list=[./train_data/ic15_data/rec_gt_train.txt] Eval.dataset.data_dir=./train_data/ic15_data Eval.dataset.label_file_list=[./train_data/ic15_data/rec_gt_test.txt]
+fpgm_train:null
+distill_train:null
+to_static_train:Global.to_static=true
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/PP-OCRv4_mobile_rec_infer
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_image_shape="3,48,320"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,48,320]}]
+===========================train_benchmark_params==========================
+batch_size:256
+fp_items:fp16
+epoch:1
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
+===========================disable_to_static_train_benchmark===========================
+to_static_train:Global.to_static=False
diff --git a/test_tipc/configs/PP-OCRv4_server_rec_fp32_ultra/train_infer_python.txt b/test_tipc/configs/PP-OCRv4_server_rec_fp32_ultra/train_infer_python.txt
new file mode 100644
index 0000000..7221d78
--- /dev/null
+++ b/test_tipc/configs/PP-OCRv4_server_rec_fp32_ultra/train_infer_python.txt
@@ -0,0 +1,61 @@
+===========================train_params===========================
+model_name:PP-OCRv4_server_rec
+python:python
+gpu_list:0
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/rec/PP-OCRv4/PP-OCRv4_server_rec_fp32_ultra.yml -o Global.cal_metric_during_train=False Global.print_batch_step=1 Train.loader.shuffle=false Train.dataset.data_dir=./train_data/ic15_data Train.dataset.label_file_list=[./train_data/ic15_data/rec_gt_train.txt] Eval.dataset.data_dir=./train_data/ic15_data Eval.dataset.label_file_list=[./train_data/ic15_data/rec_gt_test.txt]
+pact_train:null
+fpgm_train:null
+distill_train:null
+to_static_train:Global.to_static=true
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/rec/PP-OCRv4/PP-OCRv4_mobile_rec.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/PP-OCRv4_mobile_rec_infer
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_image_shape="3,48,320"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,48,320]}]
+===========================train_benchmark_params==========================
+batch_size:256
+fp_items:fp32
+epoch:1
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
+===========================disable_to_static_train_benchmark===========================
+to_static_train:Global.to_static=False
diff --git a/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..2a432b3
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_PP-OCRv2
+use_opencv:True
+infer_model:./inference/ch_PP-OCRv2_det_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_img_h=32
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--rec_model_dir:./inference/ch_PP-OCRv2_rec_infer/
+--benchmark:True
+--det:True
+--rec:True
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..32b290a
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================ch_PP-OCRv2===========================
+model_name:ch_PP-OCRv2
+python:python3.7
+infer_model:./inference/ch_PP-OCRv2_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_system.py
+--use_gpu:False|True
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--rec_model_dir:./inference/ch_PP-OCRv2_rec_infer/
+--benchmark:True
+null:null
+null:null
diff --git a/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt b/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt
new file mode 100644
index 0000000..d03ee1e
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt
@@ -0,0 +1,13 @@
+===========================lite_params===========================
+inference:./ocr_db_crnn system
+runtime_device:ARM_CPU
+det_infer_model:ch_PP-OCRv2_det_infer|ch_PP-OCRv2_det_slim_quant_infer
+rec_infer_model:ch_PP-OCRv2_rec_infer|ch_PP-OCRv2_rec_slim_quant_infer
+cls_infer_model:ch_ppocr_mobile_v2.0_cls_infer|ch_ppocr_mobile_v2.0_cls_slim_infer
+--cpu_threads:1|4
+--det_batch_size:1
+--rec_batch_size:1
+--image_dir:./test_data/icdar2015_lite/text_localization/ch4_test_images/
+--config_dir:./config.txt
+--rec_dict_dir:./ppocr_keys_v1.txt
+--benchmark:True
diff --git a/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt b/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt
new file mode 100644
index 0000000..a9fcafb
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt
@@ -0,0 +1,13 @@
+===========================lite_params===========================
+inference:./ocr_db_crnn system
+runtime_device:ARM_GPU_OPENCL
+det_infer_model:ch_PP-OCRv2_det_infer|ch_PP-OCRv2_det_slim_quant_infer
+rec_infer_model:ch_PP-OCRv2_rec_infer|ch_PP-OCRv2_rec_slim_quant_infer
+cls_infer_model:ch_ppocr_mobile_v2.0_cls_infer|ch_ppocr_mobile_v2.0_cls_slim_infer
+--cpu_threads:1|4
+--det_batch_size:1
+--rec_batch_size:1
+--image_dir:./test_data/icdar2015_lite/text_localization/ch4_test_images/
+--config_dir:./config.txt
+--rec_dict_dir:./ppocr_keys_v1.txt
+--benchmark:True
diff --git a/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt b/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
new file mode 100644
index 0000000..6a85f9f
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
@@ -0,0 +1,17 @@
+===========================paddle2onnx_params===========================
+model_name:ch_PP-OCRv2
+python:python3.7
+2onnx: paddle2onnx
+--det_model_dir:./inference/ch_PP-OCRv2_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_save_file:./inference/det_v2_onnx/model.onnx
+--rec_model_dir:./inference/ch_PP-OCRv2_rec_infer/
+--rec_save_file:./inference/rec_v2_onnx/model.onnx
+--opset_version:10
+--enable_onnx_checker:True
+inference:tools/infer/predict_system.py --rec_image_shape="3,32,320"
+--use_gpu:True|False
+--det_model_dir:
+--rec_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/00008790.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..f0456b5
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv2
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_PP-OCRv2_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v2_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v2_client/
+--rec_dirname:./inference/ch_PP-OCRv2_rec_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v2_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v2_client/
+serving_dir:./deploy/pdserving
+web_service:-m paddle_serving_server.serve
+--op:GeneralDetectionOp GeneralInferOp
+--port:8181
+--gpu_id:"0"|null
+cpp_client:ocr_cpp_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..4ad64db
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv2
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_PP-OCRv2_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v2_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v2_client/
+--rec_dirname:./inference/ch_PP-OCRv2_rec_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v2_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v2_client/
+serving_dir:./deploy/pdserving
+web_service:web_service.py --config=config.yml --opt op.det.concurrency="1" op.rec.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..a723fff
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_PP-OCRv2_det
+use_opencv:True
+infer_model:./inference/ch_PP-OCRv2_det_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+--det:True
+--rec:False
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt
new file mode 100644
index 0000000..003d776
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt
@@ -0,0 +1,13 @@
+===========================lite_params===========================
+inference:./ocr_db_crnn det
+runtime_device:ARM_CPU
+det_infer_model:ch_PP-OCRv2_det_infer|ch_PP-OCRv2_det_slim_quant_infer
+null:null
+null:null
+--cpu_threads:1|4
+--det_batch_size:1
+null:null
+--image_dir:./test_data/icdar2015_lite/text_localization/ch4_test_images/
+--config_dir:./config.txt
+null:null
+--benchmark:True
diff --git a/test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt b/test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt
new file mode 100644
index 0000000..a0b5569
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt
@@ -0,0 +1,13 @@
+===========================lite_params===========================
+inference:./ocr_db_crnn det
+runtime_device:ARM_GPU_OPENCL
+det_infer_model:ch_PP-OCRv2_det_infer|ch_PP-OCRv2_det_slim_quant_infer
+null:null
+null:null
+--cpu_threads:1|4
+--det_batch_size:1
+null:null
+--image_dir:./test_data/icdar2015_lite/text_localization/ch4_test_images/
+--config_dir:./config.txt
+null:null
+--benchmark:True
diff --git a/test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
new file mode 100644
index 0000000..834aef4
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
@@ -0,0 +1,17 @@
+===========================paddle2onnx_params===========================
+model_name:ch_PP-OCRv2_det
+python:python3.7
+2onnx: paddle2onnx
+--det_model_dir:./inference/ch_PP-OCRv2_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_save_file:./inference/det_v2_onnx/model.onnx
+--rec_model_dir:
+--rec_save_file:
+--opset_version:10
+--enable_onnx_checker:True
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--det_model_dir:
+--rec_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
diff --git a/test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..587a7d7
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv2_det
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_PP-OCRv2_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v2_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v2_client/
+--rec_dirname:null
+--rec_serving_server:null
+--rec_serving_client:null
+serving_dir:./deploy/pdserving
+web_service:web_service_det.py --config=config.yml --opt op.det.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv2_det/train_infer_python.txt b/test_tipc/configs/ch_PP-OCRv2_det/train_infer_python.txt
new file mode 100644
index 0000000..7a20df7
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_det/train_infer_python.txt
@@ -0,0 +1,59 @@
+===========================train_params===========================
+model_name:ch_PP-OCRv2_det
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o Global.print_batch_step=1 Train.loader.shuffle=false
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o 
+quant_export:null
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/ch_PP-OCRv2_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
+===========================train_benchmark_params==========================
+batch_size:8
+fp_items:fp32|fp16
+epoch:2
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
diff --git a/test_tipc/configs/ch_PP-OCRv2_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..91a6288
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_PP-OCRv2_det
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o 
+quant_export:null
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/ch_PP-OCRv2_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_PP-OCRv2_det/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_det/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..85b0ebc
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_det/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_PP-OCRv2_det
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o 
+quant_export:null
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/ch_PP-OCRv2_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_PP-OCRv2_det/train_pact_infer_python.txt b/test_tipc/configs/ch_PP-OCRv2_det/train_pact_infer_python.txt
new file mode 100644
index 0000000..1a20f97
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_det/train_pact_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_PP-OCRv2_det_PACT
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=1|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:pact_train
+norm_train:null
+pact_train:deploy/slim/quantization/quant.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null
+quant_export:deploy/slim/quantization/export_model.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o 
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/ch_PP-OCRv2_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_PP-OCRv2_det/train_ptq_infer_python.txt b/test_tipc/configs/ch_PP-OCRv2_det/train_ptq_infer_python.txt
new file mode 100644
index 0000000..ccc9e5c
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_det/train_ptq_infer_python.txt
@@ -0,0 +1,21 @@
+===========================kl_quant_params===========================
+model_name:ch_PP-OCRv2_det_KL
+python:python3.7
+Global.pretrained_model:null
+Global.save_inference_dir:null
+infer_model:./inference/ch_PP-OCRv2_det_infer/
+infer_export:deploy/slim/quantization/quant_kl.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o
+infer_quant:True
+inference:tools/infer/predict_det.py
+--use_gpu:False|True
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:int8
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+null:null
diff --git a/test_tipc/configs/ch_PP-OCRv2_det_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_det_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..038223a
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_det_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_PP-OCRv2_det_KL
+use_opencv:True
+infer_model:./inference/ch_PP-OCRv2_det_klquant_infer
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+--det:True
+--rec:False
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_PP-OCRv2_det_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_det_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..e306b0a
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_det_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv2_det_KL
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_PP-OCRv2_det_klquant_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v2_kl_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v2_kl_client/
+--rec_dirname:./inference/ch_PP-OCRv2_rec_klquant_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v2_kl_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v2_kl_client/
+serving_dir:./deploy/pdserving
+web_service:-m paddle_serving_server.serve
+--op:GeneralDetectionOp GeneralInferOp
+--port:8181
+--gpu_id:"0"|null
+cpp_client:ocr_cpp_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv2_det_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_det_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..2c96d2b
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_det_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv2_det_KL
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_PP-OCRv2_det_klquant_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v2_kl_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v2_kl_client/
+--rec_dirname:null
+--rec_serving_server:null
+--rec_serving_client:null
+serving_dir:./deploy/pdserving
+web_service:web_service_det.py --config=config.yml --opt op.det.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv2_det_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_det_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..2106672
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_det_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_PP-OCRv2_det_PACT
+use_opencv:True
+infer_model:./inference/ch_PP-OCRv2_det_pact_infer
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+--det:True
+--rec:False
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_PP-OCRv2_det_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_det_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..b2d929b
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_det_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv2_det_PACT
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_PP-OCRv2_det_pact_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v2_pact_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v2_pact_client/
+--rec_dirname:./inference/ch_PP-OCRv2_rec_pact_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v2_pact_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v2_pact_client/
+serving_dir:./deploy/pdserving
+web_service:-m paddle_serving_server.serve
+--op:GeneralDetectionOp GeneralInferOp
+--port:8181
+--gpu_id:"0"|null
+cpp_client:ocr_cpp_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv2_det_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_det_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..d5d99ab
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_det_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv2_det_PACT
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_PP-OCRv2_det_pact_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v2_pact_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v2_pact_client/
+--rec_dirname:null
+--rec_serving_server:null
+--rec_serving_client:null
+serving_dir:./deploy/pdserving
+web_service:web_service_det.py --config=config.yml --opt op.det.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml b/test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml
new file mode 100644
index 0000000..43e14b8
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml
@@ -0,0 +1,159 @@
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 800
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_pp-OCRv2_distillation
+  save_epoch_step: 3
+  eval_batch_step: [0, 200000]
+  cal_metric_during_train: true
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: false
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: 25
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: ./output/rec/predicts_pp-OCRv2_distillation.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Piecewise
+    decay_epochs : [700]
+    values : [0.001, 0.0001]
+    warmup_epoch: 5
+  regularizer:
+    name: L2
+    factor: 2.0e-05
+
+Architecture:
+  model_type: &model_type "rec"
+  name: DistillationModel
+  algorithm: Distillation
+  Models:
+    Teacher:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: CRNN
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+      Neck:
+        name: SequenceEncoder
+        encoder_type: rnn
+        hidden_size: 64
+      Head:
+        name: CTCHead
+        mid_channels: 96
+        fc_decay: 0.00002
+    Student:
+      pretrained:
+      freeze_params: false
+      return_all_feats: true
+      model_type: *model_type
+      algorithm: CRNN
+      Transform:
+      Backbone:
+        name: MobileNetV1Enhance
+        scale: 0.5
+      Neck:
+        name: SequenceEncoder
+        encoder_type: rnn
+        hidden_size: 64
+      Head:
+        name: CTCHead
+        mid_channels: 96
+        fc_decay: 0.00002
+  
+
+Loss:
+  name: CombinedLoss
+  loss_config_list:
+  - DistillationCTCLoss:
+      weight: 1.0
+      model_name_list: ["Student", "Teacher"]
+      key: head_out
+  - DistillationDMLLoss:
+      weight: 1.0
+      act: "softmax"
+      use_log: true
+      model_name_pairs:
+      - ["Student", "Teacher"]
+      key: head_out
+  - DistillationDistanceLoss:
+      weight: 1.0
+      mode: "l2"
+      model_name_pairs:
+      - ["Student", "Teacher"]
+      key: backbone_out
+
+PostProcess:
+  name: DistillationCTCLabelDecode
+  model_name: ["Student", "Teacher"]
+  key: head_out
+
+Metric:
+  name: DistillationMetric
+  base_metric_name: RecMetric
+  main_indicator: acc
+  key: "Student"
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list:
+    - ./train_data/ic15_data/rec_gt_train.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - RecAug:
+    - CTCLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 32, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: true
+    batch_size_per_card: 128
+    drop_last: true
+    num_sections: 1
+    num_workers: 8
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list:
+    - ./train_data/ic15_data/rec_gt_test.txt
+    transforms:
+    - DecodeImage:
+        img_mode: BGR
+        channel_first: false
+    - CTCLabelEncode:
+    - RecResizeImg:
+        image_shape: [3, 32, 320]
+    - KeepKeys:
+        keep_keys:
+        - image
+        - label
+        - length
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 128
+    num_workers: 8
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_rec/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..c616e41
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_rec/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_PP-OCRv2_rec
+use_opencv:True
+infer_model:./inference/ch_PP-OCRv2_rec_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_img_h=32
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference/
+null:null
+--benchmark:True
+--det:False
+--rec:True
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_rec/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
new file mode 100644
index 0000000..003d593
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_rec/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
@@ -0,0 +1,17 @@
+===========================paddle2onnx_params===========================
+model_name:ch_PP-OCRv2_rec
+python:python3.7
+2onnx: paddle2onnx
+--det_model_dir:
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_save_file:
+--rec_model_dir:./inference/ch_PP-OCRv2_rec_infer/
+--rec_save_file:./inference/rec_v2_onnx/model.onnx
+--opset_version:10
+--enable_onnx_checker:True
+inference:tools/infer/predict_rec.py --rec_image_shape="3,32,320"
+--use_gpu:True|False
+--det_model_dir:
+--rec_model_dir:
+--image_dir:./inference/rec_inference/
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_rec/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..e9e90d3
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_rec/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv2_rec
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:null
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:null
+--det_serving_client:null
+--rec_dirname:./inference/ch_PP-OCRv2_rec_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v2_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v2_client/
+serving_dir:./deploy/pdserving
+web_service:web_service_rec.py --config=config.yml --opt op.rec.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py --det=False
+--image_dir:../../inference/rec_inference
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec/train_infer_python.txt b/test_tipc/configs/ch_PP-OCRv2_rec/train_infer_python.txt
new file mode 100644
index 0000000..6d20b2d
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_rec/train_infer_python.txt
@@ -0,0 +1,59 @@
+===========================train_params===========================
+model_name:ch_PP-OCRv2_rec
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o Global.print_batch_step=4 Train.loader.shuffle=false
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/ch_PP-OCRv2_rec_infer
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,320]}]
+===========================train_benchmark_params==========================
+batch_size:64
+fp_items:fp32|fp16
+epoch:1
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..5795bc2
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_PP-OCRv2_rec
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/ch_PP-OCRv2_rec_infer
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,320]}]
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_rec/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..1b8800f
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_rec/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_PP-OCRv2_rec
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/ch_PP-OCRv2_rec_infer
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,320]}]
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec/train_pact_infer_python.txt b/test_tipc/configs/ch_PP-OCRv2_rec/train_pact_infer_python.txt
new file mode 100644
index 0000000..0ac75ef
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_rec/train_pact_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_PP-OCRv2_rec_PACT
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:pretrain_models/ch_PP-OCRv2_rec_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:pact_train
+norm_train:null
+pact_train:deploy/slim/quantization/quant.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o 
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null
+quant_export:deploy/slim/quantization/export_model.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o 
+fpgm_export: null
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/ch_PP-OCRv2_rec_slim_quant_infer
+infer_export:null
+infer_quant:True
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,320]}]
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec/train_ptq_infer_python.txt b/test_tipc/configs/ch_PP-OCRv2_rec/train_ptq_infer_python.txt
new file mode 100644
index 0000000..c30e085
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_rec/train_ptq_infer_python.txt
@@ -0,0 +1,21 @@
+===========================kl_quant_params===========================
+model_name:ch_PP-OCRv2_rec_KL
+python:python3.7
+Global.pretrained_model:null
+Global.save_inference_dir:null
+infer_model:./inference/ch_PP-OCRv2_rec_infer/
+infer_export:deploy/slim/quantization/quant_kl.py -c test_tipc/configs/ch_PP-OCRv2_rec/ch_PP-OCRv2_rec_distillation.yml -o
+infer_quant:True
+inference:tools/infer/predict_rec.py --rec_image_shape="3,32,320"
+--use_gpu:False|True
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:int8
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+null:null
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_rec_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..cdf6cb3
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_rec_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_PP-OCRv2_rec_KL
+use_opencv:True
+infer_model:./inference/ch_PP-OCRv2_rec_klquant_infer
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_img_h=32
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference/
+null:null
+--benchmark:True
+--det:False
+--rec:True
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_rec_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..34d4007
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_rec_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv2_rec_KL
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_PP-OCRv2_det_klquant_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v2_kl_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v2_kl_client/
+--rec_dirname:./inference/ch_PP-OCRv2_rec_klquant_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v2_kl_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v2_kl_client/
+serving_dir:./deploy/pdserving
+web_service:-m paddle_serving_server.serve
+--op:GeneralDetectionOp GeneralInferOp
+--port:8181
+--gpu_id:"0"|null
+cpp_client:ocr_cpp_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_rec_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..3405f2b
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_rec_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv2_rec_KL
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:null
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:null
+--det_serving_client:null
+--rec_dirname:./inference/ch_PP-OCRv2_rec_klquant_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v2_kl_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v2_kl_client/
+serving_dir:./deploy/pdserving
+web_service:web_service_rec.py --config=config.yml --opt op.rec.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py --det=False
+--image_dir:../../inference/rec_inference
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_rec_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..73943d4
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_rec_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_PP-OCRv2_rec_PACT
+use_opencv:True
+infer_model:./inference/ch_PP-OCRv2_rec_pact_infer
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_img_h=32
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference/
+null:null
+--benchmark:True
+--det:False
+--rec:True
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_rec_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..2a174b9
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_rec_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv2_rec_PACT
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_PP-OCRv2_det_pact_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v2_pact_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v2_pact_client/
+--rec_dirname:./inference/ch_PP-OCRv2_rec_pact_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v2_pact_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v2_pact_client/
+serving_dir:./deploy/pdserving
+web_service:-m paddle_serving_server.serve
+--op:GeneralDetectionOp GeneralInferOp
+--port:8181
+--gpu_id:"0"|null
+cpp_client:ocr_cpp_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv2_rec_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv2_rec_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..2b7ed81
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv2_rec_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv2_rec_PACT
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:null
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:null
+--det_serving_client:null
+--rec_dirname:./inference/ch_PP-OCRv2_rec_pact_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v2_pact_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v2_pact_client/
+serving_dir:./deploy/pdserving
+web_service:web_service_rec.py --config=config.yml --opt op.rec.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py --det=False
+--image_dir:../../inference/rec_inference
diff --git a/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..97ee8d7
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_PP-OCRv3
+use_opencv:True
+infer_model:./inference/PP-OCRv3_mobile_det_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr --rec_img_h=48 --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--rec_model_dir:./inference/PP-OCRv3_mobile_rec_infer/
+--benchmark:True
+--det:True
+--rec:True
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..29b20a6
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================ch_PP-OCRv2===========================
+model_name:ch_PP-OCRv3
+python:python3.7
+infer_model:./inference/PP-OCRv3_mobile_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_system.py --rec_image_shape="3,48,320"
+--use_gpu:False|True
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--rec_model_dir:./inference/PP-OCRv3_mobile_rec_infer/
+--benchmark:True
+null:null
+null:null
diff --git a/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt b/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt
new file mode 100644
index 0000000..946ec08
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt
@@ -0,0 +1,13 @@
+===========================lite_params===========================
+inference:./ocr_db_crnn system
+runtime_device:ARM_CPU
+det_infer_model:PP-OCRv3_mobile_det_infer|ch_PP-OCRv3_det_slim_quant_infer
+rec_infer_model:PP-OCRv3_mobile_rec_infer|PP-OCRv3_mobile_rec_slim_quant_infer
+cls_infer_model:ch_ppocr_mobile_v2.0_cls_infer|ch_ppocr_mobile_v2.0_cls_slim_infer
+--cpu_threads:1|4
+--det_batch_size:1
+--rec_batch_size:1
+--image_dir:./test_data/icdar2015_lite/text_localization/ch4_test_images/
+--config_dir:./config.txt
+--rec_dict_dir:./ppocr_keys_v1.txt
+--benchmark:True
diff --git a/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt b/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt
new file mode 100644
index 0000000..61822d9
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt
@@ -0,0 +1,13 @@
+===========================lite_params===========================
+inference:./ocr_db_crnn system
+runtime_device:ARM_GPU_OPENCL
+det_infer_model:PP-OCRv3_mobile_det_infer|ch_PP-OCRv3_det_slim_quant_infer
+rec_infer_model:PP-OCRv3_mobile_rec_infer|PP-OCRv3_mobile_rec_slim_quant_infer
+cls_infer_model:ch_ppocr_mobile_v2.0_cls_infer|ch_ppocr_mobile_v2.0_cls_slim_infer
+--cpu_threads:1|4
+--det_batch_size:1
+--rec_batch_size:1
+--image_dir:./test_data/icdar2015_lite/text_localization/ch4_test_images/
+--config_dir:./config.txt
+--rec_dict_dir:./ppocr_keys_v1.txt
+--benchmark:True
diff --git a/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt b/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
new file mode 100644
index 0000000..0f4cef3
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
@@ -0,0 +1,17 @@
+===========================paddle2onnx_params===========================
+model_name:ch_PP-OCRv3
+python:python3.7
+2onnx: paddle2onnx
+--det_model_dir:./inference/PP-OCRv3_mobile_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_save_file:./inference/det_v3_onnx/model.onnx
+--rec_model_dir:./inference/PP-OCRv3_mobile_rec_infer/
+--rec_save_file:./inference/rec_v3_onnx/model.onnx
+--opset_version:10
+--enable_onnx_checker:True
+inference:tools/infer/predict_system.py --rec_image_shape="3,48,320"
+--use_gpu:True|False
+--det_model_dir:
+--rec_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/00008790.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..d5979ad
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv3
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/PP-OCRv3_mobile_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v3_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v3_client/
+--rec_dirname:./inference/PP-OCRv3_mobile_rec_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v3_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v3_client/
+serving_dir:./deploy/pdserving
+web_service:-m paddle_serving_server.serve
+--op:GeneralDetectionOp GeneralInferOp
+--port:8181
+--gpu_id:"0"|null
+cpp_client:ocr_cpp_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..c3cee90
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv3
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/PP-OCRv3_mobile_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v3_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v3_client/
+--rec_dirname:./inference/PP-OCRv3_mobile_rec_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v3_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v3_client/
+serving_dir:./deploy/pdserving
+web_service:web_service.py --config=config.yml --opt op.det.concurrency="1" op.rec.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv3_det/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_det/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..e266f2a
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3_det/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_PP-OCRv3_det
+use_opencv:True
+infer_model:./inference/PP-OCRv3_mobile_det_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+--det:True
+--rec:False
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_PP-OCRv3_det/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_det/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt
new file mode 100644
index 0000000..c0cc839
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3_det/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt
@@ -0,0 +1,13 @@
+===========================lite_params===========================
+inference:./ocr_db_crnn det
+runtime_device:ARM_CPU
+det_infer_model:PP-OCRv3_mobile_det_infer|ch_PP-OCRv3_det_slim_quant_infer
+null:null
+null:null
+--cpu_threads:1|4
+--det_batch_size:1
+null:null
+--image_dir:./test_data/icdar2015_lite/text_localization/ch4_test_images/
+--config_dir:./config.txt
+null:null
+--benchmark:True
diff --git a/test_tipc/configs/ch_PP-OCRv3_det/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt b/test_tipc/configs/ch_PP-OCRv3_det/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt
new file mode 100644
index 0000000..f8adb28
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3_det/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt
@@ -0,0 +1,13 @@
+===========================lite_params===========================
+inference:./ocr_db_crnn det
+runtime_device:ARM_GPU_OPENCL
+det_infer_model:PP-OCRv3_mobile_det_infer|ch_PP-OCRv3_det_slim_quant_infer
+null:null
+null:null
+--cpu_threads:1|4
+--det_batch_size:1
+null:null
+--image_dir:./test_data/icdar2015_lite/text_localization/ch4_test_images/
+--config_dir:./config.txt
+null:null
+--benchmark:True
diff --git a/test_tipc/configs/ch_PP-OCRv3_det/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_det/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
new file mode 100644
index 0000000..7dae93e
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3_det/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
@@ -0,0 +1,17 @@
+===========================paddle2onnx_params===========================
+model_name:ch_PP-OCRv3_det
+python:python3.7
+2onnx: paddle2onnx
+--det_model_dir:./inference/PP-OCRv3_mobile_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_save_file:./inference/det_v3_onnx/model.onnx
+--rec_model_dir:
+--rec_save_file:
+--opset_version:10
+--enable_onnx_checker:True
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--det_model_dir:
+--rec_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
diff --git a/test_tipc/configs/ch_PP-OCRv3_det/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_det/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..60111cb
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3_det/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv3_det
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/PP-OCRv3_mobile_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v3_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v3_client/
+--rec_dirname:null
+--rec_serving_server:null
+--rec_serving_client:null
+serving_dir:./deploy/pdserving
+web_service:web_service_det.py --config=config.yml --opt op.det.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv3_det/train_infer_python.txt b/test_tipc/configs/ch_PP-OCRv3_det/train_infer_python.txt
new file mode 100644
index 0000000..91d05df
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3_det/train_infer_python.txt
@@ -0,0 +1,61 @@
+===========================train_params===========================
+model_name:ch_PP-OCRv3_det
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml -o Global.print_batch_step=1 Train.loader.shuffle=false Global.eval_batch_step=[4000,400]
+pact_train:null
+fpgm_train:null
+distill_train:null
+to_static_train:Global.to_static=true
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml -o 
+quant_export:null
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/PP-OCRv3_mobile_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
+===========================train_benchmark_params==========================
+batch_size:8
+fp_items:fp32|fp16
+epoch:2
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
+===========================to_static_train_benchmark_params===========================
+to_static_train:Global.to_static=true
diff --git a/test_tipc/configs/ch_PP-OCRv3_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..40a44ff
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_PP-OCRv3_det
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml -o 
+quant_export:null
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/PP-OCRv3_mobile_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_PP-OCRv3_det/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_det/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..5b2b080
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3_det/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_PP-OCRv3_det
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml -o 
+quant_export:null
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/PP-OCRv3_mobile_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_PP-OCRv3_det/train_pact_infer_python.txt b/test_tipc/configs/ch_PP-OCRv3_det/train_pact_infer_python.txt
new file mode 100644
index 0000000..76d2148
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3_det/train_pact_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_PP-OCRv3_det_PACT
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=1|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:pact_train
+norm_train:null
+pact_train:deploy/slim/quantization/quant.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml -o
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null
+quant_export:deploy/slim/quantization/export_model.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml -o 
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/PP-OCRv3_mobile_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_PP-OCRv3_det/train_ptq_infer_python.txt b/test_tipc/configs/ch_PP-OCRv3_det/train_ptq_infer_python.txt
new file mode 100644
index 0000000..c193039
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3_det/train_ptq_infer_python.txt
@@ -0,0 +1,21 @@
+===========================kl_quant_params===========================
+model_name:ch_PP-OCRv3_det_KL
+python:python3.7
+Global.pretrained_model:null
+Global.save_inference_dir:null
+infer_model:./inference/PP-OCRv3_mobile_det_infer/
+infer_export:deploy/slim/quantization/quant_kl.py -c configs/det/PP-OCRv3/PP-OCRv3_det_cml.yml -o
+infer_quant:True
+inference:tools/infer/predict_det.py
+--use_gpu:False|True
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:int8
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+null:null
diff --git a/test_tipc/configs/ch_PP-OCRv3_det_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_det_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..afaa015
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3_det_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_PP-OCRv3_det_KL
+use_opencv:True
+infer_model:./inference/ch_PP-OCRv3_det_klquant_infer
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+--det:True
+--rec:False
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_PP-OCRv3_det_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_det_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..5376f85
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3_det_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv3_det_KL
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_PP-OCRv3_det_klquant_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v3_kl_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v3_kl_client/
+--rec_dirname:./inference/PP-OCRv3_mobile_rec_klquant_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v3_kl_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v3_kl_client/
+serving_dir:./deploy/pdserving
+web_service:-m paddle_serving_server.serve
+--op:GeneralDetectionOp GeneralInferOp
+--port:8181
+--gpu_id:"0"|null
+cpp_client:ocr_cpp_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv3_det_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_det_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..23dbc49
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3_det_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv3_det_KL
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_PP-OCRv3_det_klquant_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v3_kl_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v3_kl_client/
+--rec_dirname:null
+--rec_serving_server:null
+--rec_serving_client:null
+serving_dir:./deploy/pdserving
+web_service:web_service_det.py --config=config.yml --opt op.det.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv3_det_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_det_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..794e979
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3_det_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_PP-OCRv3_det_PACT
+use_opencv:True
+infer_model:./inference/ch_PP-OCRv3_det_pact_infer
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+--det:True
+--rec:False
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_PP-OCRv3_det_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_det_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..86f9c06
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3_det_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv3_det_PACT
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_PP-OCRv3_det_pact_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v3_pact_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v3_pact_client/
+--rec_dirname:./inference/PP-OCRv3_mobile_rec_pact_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_v3_pact_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_v3_pact_client/
+serving_dir:./deploy/pdserving
+web_service:-m paddle_serving_server.serve
+--op:GeneralDetectionOp GeneralInferOp
+--port:8181
+--gpu_id:"0"|null
+cpp_client:ocr_cpp_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv3_det_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_PP-OCRv3_det_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..4546644
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv3_det_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_PP-OCRv3_det_PACT
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_PP-OCRv3_det_pact_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_v3_pact_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_v3_pact_client/
+--rec_dirname:null
+--rec_serving_server:null
+--rec_serving_client:null
+serving_dir:./deploy/pdserving
+web_service:web_service_det.py --config=config.yml --opt op.det.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_PP-OCRv4_mobile_det/train_infer_python.txt b/test_tipc/configs/ch_PP-OCRv4_mobile_det/train_infer_python.txt
new file mode 100644
index 0000000..a52f9d1
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv4_mobile_det/train_infer_python.txt
@@ -0,0 +1,61 @@
+===========================train_params===========================
+model_name:PP-OCRv4_mobile_det
+python:python
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:pretrain_models/PP-OCRv4_mobile_det_pretrained.pdparams
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/PP-OCRv4/PP-OCRv4_mobile_det.yml -o Global.print_batch_step=1 Train.loader.shuffle=false Global.eval_batch_step=[4000,400]
+pact_train:null
+fpgm_train:null
+distill_train:null
+to_static_train:Global.to_static=true
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/PP-OCRv4/PP-OCRv4_mobile_det.yml -o 
+quant_export:null
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/PP-OCRv3_mobile_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
+===========================train_benchmark_params==========================
+batch_size:8
+fp_items:fp32|fp16
+epoch:2
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
+===========================to_static_train_benchmark_params===========================
+to_static_train:Global.to_static=true
diff --git a/test_tipc/configs/ch_PP-OCRv4_server_det/train_infer_python.txt b/test_tipc/configs/ch_PP-OCRv4_server_det/train_infer_python.txt
new file mode 100644
index 0000000..3974509
--- /dev/null
+++ b/test_tipc/configs/ch_PP-OCRv4_server_det/train_infer_python.txt
@@ -0,0 +1,61 @@
+===========================train_params===========================
+model_name:PP-OCRv4_server_det
+python:python
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:pretrain_models/PP-OCRv4_server_det_pretrained.pdparams
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/PP-OCRv4/PP-OCRv4_server_det.yml -o Global.print_batch_step=1 Train.loader.shuffle=false Global.eval_batch_step=[4000,400]
+pact_train:null
+fpgm_train:null
+distill_train:null
+to_static_train:Global.to_static=true
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/PP-OCRv4/PP-OCRv4_server_det.yml -o 
+quant_export:null
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+inference_dir:Student
+infer_model:./inference/PP-OCRv3_mobile_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
+===========================train_benchmark_params==========================
+batch_size:4
+fp_items:fp32|fp16
+epoch:2
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
+===========================to_static_train_benchmark_params===========================
+to_static_train:Global.to_static=true
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..65312a7
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_ppocr_mobile_v2_0
+use_opencv:True
+infer_model:./inference/ch_ppocr_mobile_v2.0_det_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_img_h=32
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--rec_model_dir:./inference/ch_ppocr_mobile_v2.0_rec_infer/
+--benchmark:True
+--det:True
+--rec:True
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..00373b6
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================ch_ppocr_mobile_v2.0===========================
+model_name:ch_ppocr_mobile_v2_0
+python:python3.7
+infer_model:./inference/ch_ppocr_mobile_v2.0_det_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_system.py
+--use_gpu:False|True
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--rec_model_dir:./inference/ch_ppocr_mobile_v2.0_rec_infer/
+--benchmark:True
+null:null
+null:null
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt
new file mode 100644
index 0000000..9594368
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt
@@ -0,0 +1,13 @@
+===========================lite_params===========================
+inference:./ocr_db_crnn system
+runtime_device:ARM_CPU
+det_infer_model:ch_ppocr_mobile_v2.0_det_infer|ch_ppocr_db_mobile_v2.0_det_quant_infer
+rec_infer_model:ch_ppocr_mobile_v2.0_rec_infer|ch_ppocr_mobile_v2.0_rec_slim_infer
+cls_infer_model:ch_ppocr_mobile_v2.0_cls_infer|ch_ppocr_mobile_v2.0_cls_slim_infer
+--cpu_threads:1|4
+--det_batch_size:1
+--rec_batch_size:1
+--image_dir:./test_data/icdar2015_lite/text_localization/ch4_test_images/
+--config_dir:./config.txt
+--rec_dict_dir:./ppocr_keys_v1.txt
+--benchmark:True
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt
new file mode 100644
index 0000000..c13b9be
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt
@@ -0,0 +1,13 @@
+===========================lite_params===========================
+inference:./ocr_db_crnn system
+runtime_device:ARM_GPU_OPENCL
+det_infer_model:ch_ppocr_mobile_v2.0_det_infer|ch_ppocr_db_mobile_v2.0_det_quant_infer
+rec_infer_model:ch_ppocr_mobile_v2.0_rec_infer|ch_ppocr_mobile_v2.0_rec_slim_infer
+cls_infer_model:ch_ppocr_mobile_v2.0_cls_infer|ch_ppocr_mobile_v2.0_cls_slim_infer
+--cpu_threads:1|4
+--det_batch_size:1
+--rec_batch_size:1
+--image_dir:./test_data/icdar2015_lite/text_localization/ch4_test_images/
+--config_dir:./config.txt
+--rec_dict_dir:./ppocr_keys_v1.txt
+--benchmark:True
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
new file mode 100644
index 0000000..ec6135f
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
@@ -0,0 +1,17 @@
+===========================paddle2onnx_params===========================
+model_name:ch_ppocr_mobile_v2_0
+python:python3.7
+2onnx: paddle2onnx
+--det_model_dir:./inference/ch_ppocr_mobile_v2.0_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_save_file:./inference/det_mobile_onnx/model.onnx
+--rec_model_dir:./inference/ch_ppocr_mobile_v2.0_rec_infer/
+--rec_save_file:./inference/rec_mobile_onnx/model.onnx
+--opset_version:10
+--enable_onnx_checker:True
+inference:tools/infer/predict_system.py --rec_image_shape="3,32,320"
+--use_gpu:True|False
+--det_model_dir:
+--rec_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..305882a
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================serving_params===========================
+model_name:ch_ppocr_mobile_v2_0
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_ppocr_mobile_v2.0_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_mobile_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_mobile_client/
+--rec_dirname:./inference/ch_ppocr_mobile_v2.0_rec_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_mobile_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_mobile_client/
+serving_dir:./deploy/pdserving
+web_service:-m paddle_serving_server.serve
+--op:GeneralDetectionOp GeneralInferOp
+--port:8181
+--gpu_id:"0"|null
+cpp_client:ocr_cpp_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..0c366b0
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_ppocr_mobile_v2_0
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_ppocr_mobile_v2.0_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_mobile_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_mobile_client/
+--rec_dirname:./inference/ch_ppocr_mobile_v2.0_rec_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_mobile_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_mobile_client/
+serving_dir:./deploy/pdserving
+web_service:web_service.py --config=config.yml --opt op.det.concurrency="1" op.rec.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..8dd7718
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_ppocr_mobile_v2_0_det
+use_opencv:True
+infer_model:./inference/ch_ppocr_mobile_v2.0_det_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+--det:True
+--rec:False
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_infer_python_jetson.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_infer_python_jetson.txt
new file mode 100644
index 0000000..5f9dfa5
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_infer_python_jetson.txt
@@ -0,0 +1,18 @@
+===========================infer_params===========================
+model_name:ch_ppocr_mobile_v2_0_det
+python:python
+infer_model:./inference/ch_ppocr_mobile_v2.0_det_infer
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt
new file mode 100644
index 0000000..e83534b
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt
@@ -0,0 +1,13 @@
+===========================lite_params===========================
+inference:./ocr_db_crnn det
+runtime_device:ARM_CPU
+det_infer_model:ch_ppocr_mobile_v2.0_det_infer|ch_ppocr_db_mobile_v2.0_det_quant_infer
+null:null
+null:null
+--cpu_threads:1|4
+--det_batch_size:1
+null:null
+--image_dir:./test_data/icdar2015_lite/text_localization/ch4_test_images/
+--config_dir:./config.txt
+null:null
+--benchmark:True
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt
new file mode 100644
index 0000000..2ce8aec
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt
@@ -0,0 +1,13 @@
+===========================lite_params===========================
+inference:./ocr_db_crnn det
+runtime_device:ARM_GPU_OPENCL
+det_infer_model:ch_ppocr_mobile_v2.0_det_infer|ch_ppocr_db_mobile_v2.0_det_quant_infer
+null:null
+null:null
+--cpu_threads:1|4
+--det_batch_size:1
+null:null
+--image_dir:./test_data/icdar2015_lite/text_localization/ch4_test_images/
+--config_dir:./config.txt
+null:null
+--benchmark:True
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
new file mode 100644
index 0000000..37ceec3
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
@@ -0,0 +1,17 @@
+===========================paddle2onnx_params===========================
+model_name:ch_ppocr_mobile_v2_0_det
+python:python3.7
+2onnx: paddle2onnx
+--det_model_dir:./inference/ch_ppocr_mobile_v2.0_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_save_file:./inference/det_mobile_onnx/model.onnx
+--rec_model_dir:
+--rec_save_file:
+--opset_version:10
+--enable_onnx_checker:True
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--det_model_dir:
+--rec_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..6dfd7e7
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_ppocr_mobile_v2_0_det
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_ppocr_mobile_v2.0_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_mobile_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_mobile_client/
+--rec_dirname:null
+--rec_serving_server:null
+--rec_serving_client:null
+serving_dir:./deploy/pdserving
+web_service:web_service_det.py --config=config.yml --opt op.det.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_infer_python.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_infer_python.txt
new file mode 100644
index 0000000..4112e64
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_infer_python.txt
@@ -0,0 +1,59 @@
+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2_0_det
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=100|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained Global.print_batch_step=2 Train.loader.shuffle=false
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o 
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:./inference/ch_ppocr_mobile_v2.0_det_train/best_accuracy
+infer_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
+===========================train_benchmark_params==========================
+batch_size:8
+fp_items:fp32|fp16
+epoch:2
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_linux_dcu_normal_normal_infer_python_dcu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_linux_dcu_normal_normal_infer_python_dcu.txt
new file mode 100644
index 0000000..e69de29
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..a3bcbe1
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2_0_det
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=100|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o 
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:./inference/ch_ppocr_mobile_v2.0_det_train/best_accuracy
+infer_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..537d15e
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2_0_det
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=100|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o 
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:./inference/ch_ppocr_mobile_v2.0_det_train/best_accuracy
+infer_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_mac_cpu_normal_normal_infer_python_mac_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_mac_cpu_normal_normal_infer_python_mac_cpu.txt
new file mode 100644
index 0000000..3f321a1
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_mac_cpu_normal_normal_infer_python_mac_cpu.txt
@@ -0,0 +1,51 @@
+===========================train_params===========================
+model_name:ocr_det
+python:python
+gpu_list:-1
+Global.use_gpu:False
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=5|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.pretrained_model:
+norm_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o 
+quant_export:null
+fpgm_export:null 
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:./inference/ch_ppocr_mobile_v2.0_det_train/best_accuracy
+infer_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:False
+--enable_mkldnn:False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_pact_infer_python.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_pact_infer_python.txt
new file mode 100644
index 0000000..ba880d1
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_pact_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2_0_det_PACT
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:pact_train
+norm_train:null
+pact_train:deploy/slim/quantization/quant.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null
+quant_export:deploy/slim/quantization/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:./inference/ch_ppocr_mobile_v2.0_det_prune_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_ptq_infer_python.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_ptq_infer_python.txt
new file mode 100644
index 0000000..45c4fd1
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_ptq_infer_python.txt
@@ -0,0 +1,21 @@
+===========================kl_quant_params===========================
+model_name:ch_ppocr_mobile_v2_0_det_KL
+python:python3.7
+Global.pretrained_model:null
+Global.save_inference_dir:null
+infer_model:./inference/ch_ppocr_mobile_v2.0_det_infer/
+infer_export:deploy/slim/quantization/quant_kl.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o
+infer_quant:True
+inference:tools/infer/predict_det.py
+--use_gpu:False|True
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:int8
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+null:null
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_windows_gpu_normal_normal_infer_python_windows_cpu_gpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_windows_gpu_normal_normal_infer_python_windows_cpu_gpu.txt
new file mode 100644
index 0000000..f338e02
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_windows_gpu_normal_normal_infer_python_windows_cpu_gpu.txt
@@ -0,0 +1,51 @@
+===========================train_params===========================
+model_name:ocr_det
+python:python
+gpu_list:0
+Global.use_gpu:True
+Global.auto_cast:fp32|amp
+Global.epoch_num:lite_train_lite_infer=5|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.pretrained_model:
+norm_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o 
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:./inference/ch_ppocr_mobile_v2.0_det_train/best_accuracy
+infer_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32|fp16|int8
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det_FPGM/train_infer_python.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_FPGM/train_infer_python.txt
new file mode 100644
index 0000000..8b6dfbe
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_FPGM/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2_0_det_FPGM
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=5|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:fpgm_train
+norm_train:null
+pact_train:null
+fpgm_train:deploy/slim/prune/sensitivity_anal.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model=./pretrain_models/det_mv3_db_v2.0_train/best_accuracy
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null
+quant_export:null
+fpgm_export:deploy/slim/prune/export_prune_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o 
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:null
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det_FPGM/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_FPGM/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..279b428
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_FPGM/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2_0_det_FPGM
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=5|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:fpgm_train
+norm_train:null
+pact_train:null
+fpgm_train:deploy/slim/prune/sensitivity_anal.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o Global.pretrained_model=./pretrain_models/det_mv3_db_v2.0_train/best_accuracy
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null
+quant_export:null
+fpgm_export:deploy/slim/prune/export_prune_model.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o 
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:null
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..6ee80f6
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_ppocr_mobile_v2_0_det_KL
+use_opencv:True
+infer_model:./inference/ch_ppocr_mobile_v2.0_det_klquant_infer
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+--det:True
+--rec:False
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det_KL/model_linux_gpu_normal_normal_infer_python_mac_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_KL/model_linux_gpu_normal_normal_infer_python_mac_cpu.txt
new file mode 100644
index 0000000..ea74b7f
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_KL/model_linux_gpu_normal_normal_infer_python_mac_cpu.txt
@@ -0,0 +1,17 @@
+===========================kl_quant_params===========================
+infer_model:./inference/ch_ppocr_mobile_v2.0_det_infer/
+infer_export:deploy/slim/quantization/quant_kl.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o
+infer_quant:True
+inference:tools/infer/predict_det.py
+--use_gpu:False
+--enable_mkldnn:False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:int8
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+null:null
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det_KL/model_linux_gpu_normal_normal_infer_python_windows_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_KL/model_linux_gpu_normal_normal_infer_python_windows_gpu_cpu.txt
new file mode 100644
index 0000000..ea74b7f
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_KL/model_linux_gpu_normal_normal_infer_python_windows_gpu_cpu.txt
@@ -0,0 +1,17 @@
+===========================kl_quant_params===========================
+infer_model:./inference/ch_ppocr_mobile_v2.0_det_infer/
+infer_export:deploy/slim/quantization/quant_kl.py -c configs/det/ch_ppocr_v2.0/ch_det_mv3_db_v2.0.yml -o
+infer_quant:True
+inference:tools/infer/predict_det.py
+--use_gpu:False
+--enable_mkldnn:False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:int8
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+null:null
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..c5dc525
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================serving_params===========================
+model_name:ch_ppocr_mobile_v2_0_det_KL
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_ppocr_mobile_v2.0_det_klquant_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_mobile_kl_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_mobile_kl_client/
+--rec_dirname:./inference/ch_ppocr_mobile_v2.0_rec_klquant_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_mobile_kl_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_mobile_kl_client/
+serving_dir:./deploy/pdserving
+web_service:-m paddle_serving_server.serve
+--op:GeneralDetectionOp GeneralInferOp
+--port:8181
+--gpu_id:"0"|null
+cpp_client:ocr_cpp_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..82d4db3
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_ppocr_mobile_v2_0_det_KL
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_ppocr_mobile_v2.0_det_klquant_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_mobile_kl_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_mobile_kl_client/
+--rec_dirname:null
+--rec_serving_server:null
+--rec_serving_client:null
+serving_dir:./deploy/pdserving
+web_service:web_service_det.py --config=config.yml --opt op.det.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..a8066dd
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_ppocr_mobile_v2_0_det_PACT
+use_opencv:True
+infer_model:./inference/ch_ppocr_mobile_v2.0_det_pact_infer
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+--det:True
+--rec:False
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..3be5395
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================serving_params===========================
+model_name:ch_ppocr_mobile_v2_0_det_PACT
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_ppocr_mobile_v2.0_det_pact_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_mobile_pact_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_mobile_pact_client/
+--rec_dirname:./inference/ch_ppocr_mobile_v2.0_rec_pact_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_mobile_pact_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_mobile_pact_client/
+serving_dir:./deploy/pdserving
+web_service:-m paddle_serving_server.serve
+--op:GeneralDetectionOp GeneralInferOp
+--port:8181
+--gpu_id:"0"|null
+cpp_client:ocr_cpp_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_det_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..63e7f8f
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_det_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_ppocr_mobile_v2_0_det_PACT
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_ppocr_mobile_v2.0_det_pact_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_mobile_pact_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_mobile_pact_client/
+--rec_dirname:null
+--rec_serving_server:null
+--rec_serving_client:null
+serving_dir:./deploy/pdserving
+web_service:web_service_det.py --config=config.yml --opt op.det.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..a8342c5
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_ppocr_mobile_v2_0_rec
+use_opencv:True
+infer_model:./inference/ch_ppocr_mobile_v2.0_rec_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_img_h=32
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference/
+null:null
+--benchmark:True
+--det:False
+--rec:True
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
new file mode 100644
index 0000000..79da03e
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
@@ -0,0 +1,17 @@
+===========================paddle2onnx_params===========================
+model_name:ch_ppocr_mobile_v2_0_rec
+python:python3.7
+2onnx: paddle2onnx
+--det_model_dir:
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_save_file:
+--rec_model_dir:./inference/ch_ppocr_mobile_v2.0_rec_infer/
+--rec_save_file:./inference/rec_mobile_onnx/model.onnx
+--opset_version:10
+--enable_onnx_checker:True
+inference:tools/infer/predict_rec.py --rec_image_shape="3,32,320"
+--use_gpu:True|False
+--det_model_dir:
+--rec_model_dir:
+--image_dir:./inference/rec_inference/
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..5c60903
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_ppocr_mobile_v2_0_rec
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:null
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:null
+--det_serving_client:null
+--rec_dirname:./inference/ch_ppocr_mobile_v2.0_rec_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_mobile_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_mobile_client/
+serving_dir:./deploy/pdserving
+web_service:web_service_rec.py --config=config.yml --opt op.rec.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py --det=False
+--image_dir:../../inference/rec_inference
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/train_infer_python.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/train_infer_python.txt
new file mode 100644
index 0000000..dc490cd
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/train_infer_python.txt
@@ -0,0 +1,59 @@
+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2_0_rec
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/rec/rec_icdar15_train.yml -o Global.print_batch_step=4 Train.loader.shuffle=false
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c configs/rec/rec_icdar15_train.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/rec/rec_icdar15_train.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/ch_ppocr_mobile_v2.0_rec_train/best_accuracy
+infer_export:tools/export_model.py -c configs/rec/rec_icdar15_train.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]
+===========================train_benchmark_params==========================
+batch_size:256
+fp_items:fp32|fp16
+epoch:3
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..2f919d1
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2_0_rec
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/rec/rec_icdar15_train.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c configs/rec/rec_icdar15_train.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/rec/rec_icdar15_train.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/ch_ppocr_mobile_v2.0_rec_train/best_accuracy
+infer_export:tools/export_model.py -c configs/rec/rec_icdar15_train.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..f60e279
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2_0_rec
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/rec/rec_icdar15_train.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c configs/rec/rec_icdar15_train.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/rec/rec_icdar15_train.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/ch_ppocr_mobile_v2.0_rec_train/best_accuracy
+infer_export:tools/export_model.py -c configs/rec/rec_icdar15_train.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/train_pact_infer_python.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/train_pact_infer_python.txt
new file mode 100644
index 0000000..9c1223f
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/train_pact_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2_0_rec_PACT
+python:python3.7
+gpu_list:0
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.checkpoints:null
+train_model_name:latest
+train_infer_img_dir:./train_data/ic15_data/test/word_1.png
+null:null
+##
+trainer:pact_train
+norm_train:null
+pact_train:deploy/slim/quantization/quant.py -c test_tipc/configs/ch_ppocr_mobile_v2_0_rec_PACT/rec_chinese_lite_train_v2.0.yml -o
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null
+quant_export:deploy/slim/quantization/export_model.py -c test_tipc/configs/ch_ppocr_mobile_v2_0_rec_PACT/rec_chinese_lite_train_v2.0.yml -o 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+infer_model:./inference/ch_ppocr_mobile_v2.0_rec_slim_infer/
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_image_shape="3,32,100"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,320]}]
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/train_ptq_infer_python.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/train_ptq_infer_python.txt
new file mode 100644
index 0000000..df47f32
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec/train_ptq_infer_python.txt
@@ -0,0 +1,21 @@
+===========================kl_quant_params===========================
+model_name:ch_ppocr_mobile_v2_0_rec_KL
+python:python3.7
+Global.pretrained_model:null
+Global.save_inference_dir:null
+infer_model:./inference/ch_ppocr_mobile_v2.0_rec_infer/
+infer_export:deploy/slim/quantization/quant_kl.py -c test_tipc/configs/ch_ppocr_mobile_v2_0_rec_KL/rec_chinese_lite_train_v2.0.yml -o
+infer_quant:True
+inference:tools/infer/predict_rec.py --rec_image_shape="3,32,320"
+--use_gpu:False|True
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:int8
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+null:null
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_FPGM/rec_chinese_lite_train_v2.0.yml b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_FPGM/rec_chinese_lite_train_v2.0.yml
new file mode 100644
index 0000000..ee42dbf
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_FPGM/rec_chinese_lite_train_v2.0.yml
@@ -0,0 +1,102 @@
+Global:
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_chinese_lite_v2.0
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  # for data or label process
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: True
+  save_res_path: ./output/rec/predicts_chinese_lite_v2.0.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0.00001
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: small
+    small_stride: [1, 2, 2, 2]
+    disable_se: True
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 48
+  Head:
+    name: CTCHead
+    fc_decay: 0.00001
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/ic15_data
+    label_file_list: ["train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - RecAug: 
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/ic15_data
+    label_file_list: ["train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_FPGM/train_infer_python.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_FPGM/train_infer_python.txt
new file mode 100644
index 0000000..d99ad00
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_FPGM/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2_0_rec_FPGM
+python:python3.7
+gpu_list:0
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/ic15_data/test/word_1.png
+null:null
+##
+trainer:fpgm_train
+norm_train:null
+pact_train:null
+fpgm_train:deploy/slim/prune/sensitivity_anal.py -c test_tipc/configs/ch_ppocr_mobile_v2_0_rec_FPGM/rec_chinese_lite_train_v2.0.yml -o Global.pretrained_model=./pretrain_models/ch_ppocr_mobile_v2.0_rec_train/best_accuracy
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null
+quant_export:null
+fpgm_export:deploy/slim/prune/export_prune_model.py -c test_tipc/configs/ch_ppocr_mobile_v2_0_rec_FPGM/rec_chinese_lite_train_v2.0.yml -o 
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:null
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,320]}]
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_FPGM/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_FPGM/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..6754f73
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_FPGM/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_mobile_v2_0_rec_FPGM
+python:python3.7
+gpu_list:0
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/ic15_data/test/word_1.png
+null:null
+##
+trainer:fpgm_train
+norm_train:null
+pact_train:null
+fpgm_train:deploy/slim/prune/sensitivity_anal.py -c test_tipc/configs/ch_ppocr_mobile_v2_0_rec_FPGM/rec_chinese_lite_train_v2.0.yml -o Global.pretrained_model=./pretrain_models/ch_ppocr_mobile_v2.0_rec_train/best_accuracy
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null
+quant_export:null
+fpgm_export:deploy/slim/prune/export_prune_model.py -c test_tipc/configs/ch_ppocr_mobile_v2_0_rec_FPGM/rec_chinese_lite_train_v2.0.yml -o 
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:null
+infer_export:null
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,320]}]
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..7669b44
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_KL/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_ppocr_mobile_v2_0_rec_KL
+use_opencv:True
+infer_model:./inference/ch_ppocr_mobile_v2.0_rec_klquant_infer
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_img_h=32
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference/
+null:null
+--benchmark:True
+--det:False
+--rec:True
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..d904e22
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_KL/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================serving_params===========================
+model_name:ch_ppocr_mobile_v2_0_rec_KL
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_ppocr_mobile_v2.0_det_klquant_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_mobile_kl_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_mobile_kl_client/
+--rec_dirname:./inference/ch_ppocr_mobile_v2.0_rec_klquant_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_mobile_kl_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_mobile_kl_client/
+serving_dir:./deploy/pdserving
+web_service:-m paddle_serving_server.serve
+--op:GeneralDetectionOp GeneralInferOp
+--port:8181
+--gpu_id:"0"|null
+cpp_client:ocr_cpp_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..de4f7ed
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_KL/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_ppocr_mobile_v2_0_rec_KL
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:null
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:null
+--det_serving_client:null
+--rec_dirname:./inference/ch_ppocr_mobile_v2.0_rec_klquant_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_mobile_kl_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_mobile_kl_client/
+serving_dir:./deploy/pdserving
+web_service:web_service_rec.py --config=config.yml --opt op.rec.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py --det=False
+--image_dir:../../inference/rec_inference
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_KL/rec_chinese_lite_train_v2.0.yml b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_KL/rec_chinese_lite_train_v2.0.yml
new file mode 100644
index 0000000..b06dafe
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_KL/rec_chinese_lite_train_v2.0.yml
@@ -0,0 +1,101 @@
+Global:
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_chinese_lite_v2.0
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  # for data or label process
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: True
+  save_res_path: ./output/rec/predicts_chinese_lite_v2.0.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0.00001
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: small
+    small_stride: [1, 2, 2, 2]
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 48
+  Head:
+    name: CTCHead
+    fc_decay: 0.00001
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/ic15_data
+    label_file_list: ["train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - RecAug: 
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/ic15_data
+    label_file_list: ["train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..d1f7c48
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_ppocr_mobile_v2_0_rec_PACT
+use_opencv:True
+infer_model:./inference/ch_ppocr_mobile_v2.0_rec_pact_infer
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_img_h=32
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference/
+null:null
+--benchmark:True
+--det:False
+--rec:True
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..5a30474
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_PACT/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================serving_params===========================
+model_name:ch_ppocr_mobile_v2_0_rec_PACT
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_ppocr_mobile_v2.0_det_pact_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_mobile_pact_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_mobile_pact_client/
+--rec_dirname:./inference/ch_ppocr_mobile_v2.0_rec_pact_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_mobile_pact_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_mobile_pact_client/
+serving_dir:./deploy/pdserving
+web_service:-m paddle_serving_server.serve
+--op:GeneralDetectionOp GeneralInferOp
+--port:8181
+--gpu_id:"0"|null
+cpp_client:ocr_cpp_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..5871199
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_PACT/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_ppocr_mobile_v2_0_rec_PACT
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:null
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:null
+--det_serving_client:null
+--rec_dirname:./inference/ch_ppocr_mobile_v2.0_rec_pact_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_mobile_pact_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_mobile_pact_client/
+serving_dir:./deploy/pdserving
+web_service:web_service_rec.py --config=config.yml --opt op.rec.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py --det=False
+--image_dir:../../inference/rec_inference
diff --git a/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_PACT/rec_chinese_lite_train_v2.0.yml b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_PACT/rec_chinese_lite_train_v2.0.yml
new file mode 100644
index 0000000..b06dafe
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_mobile_v2_0_rec_PACT/rec_chinese_lite_train_v2.0.yml
@@ -0,0 +1,101 @@
+Global:
+  use_gpu: true
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec_chinese_lite_v2.0
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  # for data or label process
+  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: True
+  save_res_path: ./output/rec/predicts_chinese_lite_v2.0.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0.00001
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: small
+    small_stride: [1, 2, 2, 2]
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 48
+  Head:
+    name: CTCHead
+    fc_decay: 0.00001
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/ic15_data
+    label_file_list: ["train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - RecAug: 
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/ic15_data
+    label_file_list: ["train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 320]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2_0/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..f742892
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_ppocr_server_v2_0
+use_opencv:True
+infer_model:./inference/ch_ppocr_server_v2.0_det_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_img_h=32
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--rec_model_dir:./inference/ch_ppocr_server_v2.0_rec_infer/
+--benchmark:True
+--det:True
+--rec:True
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2_0/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..53f8ab0
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0/model_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================ch_ppocr_server_v2.0===========================
+model_name:ch_ppocr_server_v2_0
+python:python3.7
+infer_model:./inference/ch_ppocr_server_v2.0_det_infer/
+infer_export:null
+infer_quant:True
+inference:tools/infer/predict_system.py
+--use_gpu:False|True
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--rec_model_dir:./inference/ch_ppocr_server_v2.0_rec_infer/
+--benchmark:True
+null:null
+null:null
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2_0/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
new file mode 100644
index 0000000..cd2d857
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
@@ -0,0 +1,17 @@
+===========================paddle2onnx_params===========================
+model_name:ch_ppocr_server_v2_0
+python:python3.7
+2onnx: paddle2onnx
+--det_model_dir:./inference/ch_ppocr_server_v2.0_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_save_file:./inference/det_server_onnx/model.onnx
+--rec_model_dir:./inference/ch_ppocr_server_v2.0_rec_infer/
+--rec_save_file:./inference/rec_server_onnx/model.onnx
+--opset_version:10
+--enable_onnx_checker:True
+inference:tools/infer/predict_system.py --rec_image_shape="3,32,320"
+--use_gpu:True|False
+--det_model_dir:
+--rec_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/00008790.jpg
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2_0/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..55b27e0
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,19 @@
+===========================serving_params===========================
+model_name:ch_ppocr_server_v2_0
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_ppocr_server_v2.0_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_server_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_server_client/
+--rec_dirname:./inference/ch_ppocr_server_v2.0_rec_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_server_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_server_client/
+serving_dir:./deploy/pdserving
+web_service:-m paddle_serving_server.serve
+--op:GeneralDetectionOp GeneralInferOp
+--port:8181
+--gpu_id:"0"|null
+cpp_client:ocr_cpp_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2_0/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..21b8c9a
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_ppocr_server_v2_0
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_ppocr_server_v2.0_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_server_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_server_client/
+--rec_dirname:./inference/ch_ppocr_server_v2.0_rec_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_server_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_server_client/
+serving_dir:./deploy/pdserving
+web_service:web_service.py --config=config.yml --opt op.det.concurrency="1" op.rec.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0_det/det_r50_vd_db.yml b/test_tipc/configs/ch_ppocr_server_v2_0_det/det_r50_vd_db.yml
new file mode 100644
index 0000000..6728703
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0_det/det_r50_vd_db.yml
@@ -0,0 +1,124 @@
+Global:
+  use_gpu: false
+  epoch_num: 5
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/db_mv3/
+  save_epoch_step: 1200
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 30000]
+  cal_metric_during_train: False
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_db/predicts_db.txt
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform:
+  Backbone:
+    name: ResNet  #MobileNetV3
+    layers: 50
+  Neck:
+    name: DBFPN
+    out_channels: 256
+  Head:
+    name: DBHead
+    k: 50
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5 #5
+  beta: 10 #10
+  ohem_ratio: 3
+
+Optimizer:
+  name: Adam #Momentum
+  #momentum: 0.9
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.001
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - Resize:
+          # size: [640, 640]
+      - MakeBorderMap:
+          shrink_ratio: 0.4
+          thresh_min: 0.3
+          thresh_max: 0.7
+      - MakeShrinkMap:
+          shrink_ratio: 0.4
+          min_text_size: 8
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1
+    num_workers: 0
+    use_shared_memory: False
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          image_shape: [736, 1280]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 0
+    use_shared_memory: False
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0_det/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2_0_det/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..53e6ab1
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0_det/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_ppocr_server_v2_0_det
+use_opencv:True
+infer_model:./inference/ch_ppocr_server_v2.0_det_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+--det:True
+--rec:False
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0_det/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2_0_det/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
new file mode 100644
index 0000000..2c8a9b3
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0_det/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
@@ -0,0 +1,17 @@
+===========================paddle2onnx_params===========================
+model_name:ch_ppocr_server_v2_0_det
+python:python3.7
+2onnx: paddle2onnx
+--det_model_dir:./inference/ch_ppocr_server_v2.0_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_save_file:./inference/det_server_onnx/model.onnx
+--rec_model_dir:
+--rec_save_file:
+--opset_version:10
+--enable_onnx_checker:True
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--det_model_dir:
+--rec_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/00008790.jpg
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0_det/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2_0_det/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..4d4f067
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0_det/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_ppocr_server_v2_0_det
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:./inference/ch_ppocr_server_v2.0_det_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:./deploy/pdserving/ppocr_det_server_serving/
+--det_serving_client:./deploy/pdserving/ppocr_det_server_client/
+--rec_dirname:null
+--rec_serving_server:null
+--rec_serving_client:null
+serving_dir:./deploy/pdserving
+web_service:web_service_det.py --config=config.yml --opt op.det.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py
+--image_dir:../../doc/imgs/1.jpg
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0_det/train_infer_python.txt b/test_tipc/configs/ch_ppocr_server_v2_0_det/train_infer_python.txt
new file mode 100644
index 0000000..90ed29f
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0_det/train_infer_python.txt
@@ -0,0 +1,59 @@
+===========================train_params===========================
+model_name:ch_ppocr_server_v2_0_det
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_lite_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/ch_ppocr_server_v2_0_det/det_r50_vd_db.yml -o
+quant_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:tools/eval.py -c test_tipc/configs/ch_ppocr_server_v2_0_det/det_r50_vd_db.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/ch_ppocr_server_v2_0_det/det_r50_vd_db.yml -o 
+quant_export:null 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/ch_ppocr_server_v2.0_det_train/best_accuracy
+infer_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_res18_db_v2.0.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--save_log_path:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
+===========================train_benchmark_params==========================
+batch_size:8
+fp_items:fp32|fp16
+epoch:2
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2_0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..e4e2f2e
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0_det/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_server_v2_0_det
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_lite_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/ch_ppocr_server_v2_0_det/det_r50_vd_db.yml -o 
+quant_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:tools/eval.py -c test_tipc/configs/ch_ppocr_server_v2_0_det/det_r50_vd_db.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/ch_ppocr_server_v2_0_det/det_r50_vd_db.yml -o 
+quant_export:null 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/ch_ppocr_server_v2.0_det_train/best_accuracy
+infer_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_res18_db_v2.0.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--save_log_path:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0_det/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2_0_det/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..c09c163
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0_det/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_server_v2_0_det
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_lite_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/ch_ppocr_server_v2_0_det/det_r50_vd_db.yml -o 
+quant_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:tools/eval.py -c test_tipc/configs/ch_ppocr_server_v2_0_det/det_r50_vd_db.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/ch_ppocr_server_v2_0_det/det_r50_vd_db.yml -o 
+quant_export:null 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/ch_ppocr_server_v2.0_det_train/best_accuracy
+infer_export:tools/export_model.py -c configs/det/ch_ppocr_v2.0/ch_det_res18_db_v2.0.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--save_log_path:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0_rec/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2_0_rec/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..f1f0a17
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0_rec/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:ch_ppocr_server_v2_0_rec
+use_opencv:True
+infer_model:./inference/ch_ppocr_server_v2.0_rec_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_img_h=32
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference/
+null:null
+--benchmark:True
+--det:False
+--rec:True
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0_rec/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2_0_rec/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
new file mode 100644
index 0000000..227b6fd
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0_rec/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
@@ -0,0 +1,17 @@
+===========================paddle2onnx_params===========================
+model_name:ch_ppocr_server_v2_0_rec
+python:python3.7
+2onnx: paddle2onnx
+--det_model_dir:
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_save_file:
+--rec_model_dir:./inference/ch_ppocr_server_v2.0_rec_infer/
+--rec_save_file:./inference/rec_server_onnx/model.onnx
+--opset_version:10
+--enable_onnx_checker:True
+inference:tools/infer/predict_rec.py --rec_image_shape="3,32,320"
+--use_gpu:True|False
+--det_model_dir:
+--rec_model_dir:
+--image_dir:./inference/rec_inference/
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0_rec/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2_0_rec/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..4133e96
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0_rec/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt
@@ -0,0 +1,23 @@
+===========================serving_params===========================
+model_name:ch_ppocr_server_v2_0_rec
+python:python3.7
+trans_model:-m paddle_serving_client.convert
+--det_dirname:null
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_serving_server:null
+--det_serving_client:null
+--rec_dirname:./inference/ch_ppocr_server_v2.0_rec_infer/
+--rec_serving_server:./deploy/pdserving/ppocr_rec_server_serving/
+--rec_serving_client:./deploy/pdserving/ppocr_rec_server_client/
+serving_dir:./deploy/pdserving
+web_service:web_service_rec.py --config=config.yml --opt op.rec.concurrency="1"
+op.det.local_service_conf.devices:gpu|null
+op.det.local_service_conf.use_mkldnn:False
+op.det.local_service_conf.thread_num:6
+op.det.local_service_conf.use_trt:False
+op.det.local_service_conf.precision:fp32
+op.det.local_service_conf.model_config:
+op.rec.local_service_conf.model_config:
+pipline:pipeline_http_client.py --det=False
+--image_dir:../../inference/rec_inference
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0_rec/rec_icdar15_train.yml b/test_tipc/configs/ch_ppocr_server_v2_0_rec/rec_icdar15_train.yml
new file mode 100644
index 0000000..5825c3e
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0_rec/rec_icdar15_train.yml
@@ -0,0 +1,99 @@
+Global:
+  use_gpu: true
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/ic15/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir: ./
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path: ppocr/utils/en_dict.txt
+  character_type: EN
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_ic15.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: ResNet
+    layers: 34
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 256
+  Head:
+    name: CTCHead
+    fc_decay: 0
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+    use_shared_memory: False
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
+    use_shared_memory: False
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0_rec/train_infer_python.txt b/test_tipc/configs/ch_ppocr_server_v2_0_rec/train_infer_python.txt
new file mode 100644
index 0000000..85741f9
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0_rec/train_infer_python.txt
@@ -0,0 +1,59 @@
+===========================train_params===========================
+model_name:ch_ppocr_server_v2_0_rec
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=5|whole_train_whole_infer=100
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/ch_ppocr_server_v2_0_rec/rec_icdar15_train.yml -o Global.print_batch_step=4 Train.loader.shuffle=false
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/ch_ppocr_server_v2_0_rec/rec_icdar15_train.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/ch_ppocr_server_v2_0_rec/rec_icdar15_train.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/ch_ppocr_server_v2.0_rec_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/ch_ppocr_server_v2_0_rec/rec_icdar15_train.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]
+===========================train_benchmark_params==========================
+batch_size:256
+fp_items:fp32|fp16
+epoch:2
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2_0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..d5f57ae
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_server_v2_0_rec
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=5|whole_train_whole_infer=100
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/ch_ppocr_server_v2_0_rec/rec_icdar15_train.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/ch_ppocr_server_v2_0_rec/rec_icdar15_train.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/ch_ppocr_server_v2_0_rec/rec_icdar15_train.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/ch_ppocr_server_v2.0_rec_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/ch_ppocr_server_v2_0_rec/rec_icdar15_train.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]
diff --git a/test_tipc/configs/ch_ppocr_server_v2_0_rec/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/ch_ppocr_server_v2_0_rec/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..20eb10b
--- /dev/null
+++ b/test_tipc/configs/ch_ppocr_server_v2_0_rec/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:ch_ppocr_server_v2_0_rec
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=5|whole_train_whole_infer=100
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/ch_ppocr_server_v2_0_rec/rec_icdar15_train.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/ch_ppocr_server_v2_0_rec/rec_icdar15_train.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/ch_ppocr_server_v2_0_rec/rec_icdar15_train.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/ch_ppocr_server_v2.0_rec_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/ch_ppocr_server_v2_0_rec/rec_icdar15_train.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1|6
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]
diff --git a/test_tipc/configs/det_mv3_db_v2_0/train_infer_python.txt b/test_tipc/configs/det_mv3_db_v2_0/train_infer_python.txt
new file mode 100644
index 0000000..f6ee9df
--- /dev/null
+++ b/test_tipc/configs/det_mv3_db_v2_0/train_infer_python.txt
@@ -0,0 +1,59 @@
+===========================train_params===========================
+model_name:det_mv3_db_v2_0
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/det_mv3_db.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained Global.print_batch_step=1 Train.loader.shuffle=false
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/det_mv3_db.yml -o 
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:./inference/det_mv3_db_v2.0_train/best_accuracy
+infer_export:tools/export_model.py -c configs/det/det_mv3_db.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
+===========================train_benchmark_params==========================
+batch_size:16
+fp_items:fp32|fp16
+epoch:4
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
diff --git a/test_tipc/configs/det_mv3_east_v2_0/det_mv3_east.yml b/test_tipc/configs/det_mv3_east_v2_0/det_mv3_east.yml
new file mode 100644
index 0000000..461179e
--- /dev/null
+++ b/test_tipc/configs/det_mv3_east_v2_0/det_mv3_east.yml
@@ -0,0 +1,109 @@
+Global:
+  use_gpu: true
+  epoch_num: 10000
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/east_mv3/
+  save_epoch_step: 1000
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [4000, 5000]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: 
+  save_res_path: ./output/det_east/predicts_east.txt
+
+Architecture:
+  model_type: det
+  algorithm: EAST
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+  Neck:
+    name: EASTFPN
+    model_name: small
+  Head:
+    name: EASTHead
+    model_name: small
+
+Loss:
+  name: EASTLoss
+  
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+  #  name: Cosine
+    learning_rate: 0.001
+  #  warmup_epoch: 0
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: EASTPostProcess
+  score_thresh: 0.8
+  cover_thresh: 0.1
+  nms_thresh: 0.2
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - EASTProcessTrain:
+          image_shape: [512, 512]
+          background_ratio: 0.125
+          min_crop_side_ratio: 0.1
+          min_text_size: 10
+      - KeepKeys:
+          keep_keys: ['image', 'score_map', 'geo_map', 'training_mask'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 16
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          limit_side_len: 2400
+          limit_type: max
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/test_tipc/configs/det_mv3_east_v2_0/train_infer_python.txt b/test_tipc/configs/det_mv3_east_v2_0/train_infer_python.txt
new file mode 100644
index 0000000..9c6d966
--- /dev/null
+++ b/test_tipc/configs/det_mv3_east_v2_0/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:det_mv3_east_v2_0
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=500
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:./pretrain_models/det_mv3_east_v2.0_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/det_mv3_east_v2_0/det_mv3_east.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/det_mv3_east_v2_0/det_mv3_east.yml -o 
+quant_export:null 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/det_mv3_east_v2.0_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/det_mv3_east_v2_0/det_mv3_east.yml  -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--save_log_path:null
+--benchmark:True
+--det_algorithm:EAST
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/det_mv3_pse_v2_0/det_mv3_pse.yml b/test_tipc/configs/det_mv3_pse_v2_0/det_mv3_pse.yml
new file mode 100644
index 0000000..27ec3eb
--- /dev/null
+++ b/test_tipc/configs/det_mv3_pse_v2_0/det_mv3_pse.yml
@@ -0,0 +1,135 @@
+Global:
+  use_gpu: true
+  epoch_num: 600
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/det_mv3_pse/
+  save_epoch_step: 600
+  # evaluation is run every 63 iterations
+  eval_batch_step: [ 0,1000 ]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+  checkpoints: #./output/det_r50_vd_pse_batch8_ColorJitter/best_accuracy
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_pse/predicts_pse.txt
+
+Architecture:
+  model_type: det
+  algorithm: PSE
+  Transform: null
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+  Neck:
+    name: FPN
+    out_channels: 96
+  Head:
+    name: PSEHead
+    hidden_dim: 96
+    out_channels: 7
+
+Loss:
+  name: PSELoss
+  alpha: 0.7
+  ohem_ratio: 3
+  kernel_sample_mask: pred
+  reduction: none
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Step
+    learning_rate: 0.001
+    step_size: 200
+    gamma: 0.1
+  regularizer:
+    name: 'L2'
+    factor: 0.0005
+
+PostProcess:
+  name: PSEPostProcess
+  thresh: 0
+  box_thresh: 0.85
+  min_area: 16
+  box_type: quad # 'quad' or 'poly'
+  scale: 1
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - ColorJitter:
+          brightness: 0.12549019607843137
+          saturation: 0.5
+      - IaaAugment:
+          augmenter_args:
+            - { 'type': Resize, 'args': { 'size': [ 0.5, 3 ] } }
+            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
+            - { 'type': Affine, 'args': { 'rotate': [ -10, 10 ] } }
+      - MakePseGt:
+          kernel_num: 7
+          min_shrink_ratio: 0.4
+          size: 640
+      - RandomCropImgMask:
+          size: [ 640,640 ]
+          main_key: gt_text
+          crop_keys: [ 'image', 'gt_text', 'gt_kernels', 'mask' ]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [ 0.485, 0.456, 0.406 ]
+          std: [ 0.229, 0.224, 0.225 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'image', 'gt_text', 'gt_kernels', 'mask' ] # the order of the dataloader list
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 16
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          limit_side_len: 736
+          limit_type: min
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [ 0.485, 0.456, 0.406 ]
+          std: [ 0.229, 0.224, 0.225 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'image', 'shape', 'polys', 'ignore_tags' ]
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 8
diff --git a/test_tipc/configs/det_mv3_pse_v2_0/train_infer_python.txt b/test_tipc/configs/det_mv3_pse_v2_0/train_infer_python.txt
new file mode 100644
index 0000000..525fdc7
--- /dev/null
+++ b/test_tipc/configs/det_mv3_pse_v2_0/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:det_mv3_pse_v2_0
+python:python3.7
+gpu_list:0
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=500
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/det_mv3_pse_v2_0/det_mv3_pse.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/det_mv3_pse_v2_0/det_mv3_pse.yml -o 
+quant_export:null 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/det_mv3_pse_v2.0_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/det_mv3_pse_v2_0/det_mv3_pse.yml  -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--save_log_path:null
+--benchmark:True
+--det_algorithm:PSE
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/det_r18_ct/train_infer_python.txt b/test_tipc/configs/det_r18_ct/train_infer_python.txt
new file mode 100644
index 0000000..e025312
--- /dev/null
+++ b/test_tipc/configs/det_r18_ct/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:det_r18_ct
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_lite_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/total_text/test/rgb/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/det_r18_vd_ct.yml -o Global.print_batch_step=1 Train.loader.shuffle=false
+quant_export:null
+fpgm_export:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:tools/eval.py -c configs/det/det_r18_vd_ct.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/det_r18_vd_ct.yml -o 
+quant_export:null 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/det_r18_vd_ct/best_accuracy
+infer_export:tools/export_model.py -c configs/det/det_r18_vd_ct.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--save_log_path:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/det_r50_db_plusplus/train_infer_python.txt b/test_tipc/configs/det_r50_db_plusplus/train_infer_python.txt
new file mode 100644
index 0000000..110b7f9
--- /dev/null
+++ b/test_tipc/configs/det_r50_db_plusplus/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:det_r50_db_plusplus
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/det_r50_db++_icdar15.yml -o Global.pretrained_model=./pretrain_models/ResNet50_dcn_asf_synthtext_pretrained
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/det_r50_db++_icdar15.yml -o 
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:./inference/det_r50_db++_train/best_accuracy
+infer_export:tools/export_model.py -c configs/det/det_r50_db++_icdar15.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py --det_algorithm="DB++"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/det_r50_db_v2_0/train_infer_python.txt b/test_tipc/configs/det_r50_db_v2_0/train_infer_python.txt
new file mode 100644
index 0000000..e5c1dd8
--- /dev/null
+++ b/test_tipc/configs/det_r50_db_v2_0/train_infer_python.txt
@@ -0,0 +1,59 @@
+===========================train_params===========================
+model_name:det_r50_db_v2_0
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_lite_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/det/det_r50_vd_db.yml -o Global.print_batch_step=1 Train.loader.shuffle=false
+quant_export:null
+fpgm_export:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:tools/eval.py -c configs/det/det_r50_vd_db.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/det/det_r50_vd_db.yml -o 
+quant_export:null 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/det_r50_vd_db_v2.0_train/best_accuracy
+infer_export:tools/export_model.py -c configs/det/det_r50_vd_db.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--save_log_path:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
+===========================train_benchmark_params==========================
+batch_size:8
+fp_items:fp32|fp16
+epoch:2
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
diff --git a/test_tipc/configs/det_r50_dcn_fce_ctw_v2_0/det_r50_vd_dcn_fce_ctw.yml b/test_tipc/configs/det_r50_dcn_fce_ctw_v2_0/det_r50_vd_dcn_fce_ctw.yml
new file mode 100644
index 0000000..cd355f1
--- /dev/null
+++ b/test_tipc/configs/det_r50_dcn_fce_ctw_v2_0/det_r50_vd_dcn_fce_ctw.yml
@@ -0,0 +1,139 @@
+Global:
+  use_gpu: true
+  epoch_num: 1500
+  log_smooth_window: 20
+  print_batch_step: 20
+  save_model_dir: ./output/det_r50_dcn_fce_ctw/
+  save_epoch_step: 100
+  # evaluation is run every 835 iterations
+  eval_batch_step: [0, 4000]
+  cal_metric_during_train: False
+  pretrained_model: pretrain_models/det_r50_dcn_fce_ctw_v2.0_train/best_accuracy.pdparams
+  checkpoints: 
+  save_inference_dir: 
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_fce/predicts_fce.txt
+
+
+Architecture:
+  model_type: det
+  algorithm: FCE
+  Transform:
+  Backbone:
+    name: ResNet_vd
+    layers: 50
+    dcn_stage: [False, True, True, True]
+    out_indices: [1,2,3]
+  Neck:
+    name: FCEFPN
+    out_channels: 256
+    has_extra_convs: False
+    extra_stage: 0
+  Head:
+    name: FCEHead
+    fourier_degree: 5
+Loss:
+  name: FCELoss
+  fourier_degree: 5
+  num_sample: 50
+  
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0001
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: FCEPostProcess
+  scales: [8, 16, 32]
+  alpha: 1.0
+  beta: 1.0
+  fourier_degree: 5
+  box_type: 'poly'
+
+Metric:
+  name: DetFCEMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+          ignore_orientation: True
+      - DetLabelEncode: # Class handling label
+      - ColorJitter: 
+          brightness: 0.142
+          saturation: 0.5
+          contrast: 0.5
+      - RandomScaling: 
+      - RandomCropFlip:
+          crop_ratio: 0.5
+      - RandomCropPolyInstances:
+          crop_ratio: 0.8
+          min_side_ratio: 0.3
+      - RandomRotatePolyInstances:
+          rotate_ratio: 0.5
+          max_angle: 30
+          pad_with_fixed_color: False
+      - SquareResizePad:
+          target_size: 800
+          pad_ratio: 0.6
+      - IaaAugment:
+          augmenter_args:
+            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
+      - FCENetTargets:
+          fourier_degree: 5
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'p3_maps', 'p4_maps', 'p5_maps'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 6
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+          ignore_orientation: True
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          limit_type: 'min'
+          limit_side_len: 736
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - Pad: 
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/test_tipc/configs/det_r50_dcn_fce_ctw_v2_0/train_infer_python.txt b/test_tipc/configs/det_r50_dcn_fce_ctw_v2_0/train_infer_python.txt
new file mode 100644
index 0000000..cc6080f
--- /dev/null
+++ b/test_tipc/configs/det_r50_dcn_fce_ctw_v2_0/train_infer_python.txt
@@ -0,0 +1,59 @@
+===========================train_params===========================
+model_name:det_r50_dcn_fce_ctw_v2_0
+python:python3.7
+gpu_list:0
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=500
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/det_r50_dcn_fce_ctw_v2_0/det_r50_vd_dcn_fce_ctw.yml -o Global.print_batch_step=1 Train.loader.shuffle=false
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/det_r50_dcn_fce_ctw_v2_0/det_r50_vd_dcn_fce_ctw.yml -o 
+quant_export:null 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/det_r50_dcn_fce_ctw_v2.0_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/det_r50_dcn_fce_ctw_v2_0/det_r50_vd_dcn_fce_ctw.yml  -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--save_log_path:null
+--benchmark:True
+--det_algorithm:FCE
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
+===========================train_benchmark_params==========================
+batch_size:6
+fp_items:fp32|fp16
+epoch:1
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
diff --git a/test_tipc/configs/det_r50_vd_east_v2_0/det_r50_vd_east.yml b/test_tipc/configs/det_r50_vd_east_v2_0/det_r50_vd_east.yml
new file mode 100644
index 0000000..ae7bd94
--- /dev/null
+++ b/test_tipc/configs/det_r50_vd_east_v2_0/det_r50_vd_east.yml
@@ -0,0 +1,108 @@
+Global:
+  use_gpu: true
+  epoch_num: 10000
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/east_r50_vd/
+  save_epoch_step: 1000
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [4000, 5000]
+  cal_metric_during_train: False
+  pretrained_model: 
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: 
+  save_res_path: ./output/det_east/predicts_east.txt
+
+Architecture:
+  model_type: det
+  algorithm: EAST
+  Transform:
+  Backbone:
+    name: ResNet_vd
+    layers: 50
+  Neck:
+    name: EASTFPN
+    model_name: large
+  Head:
+    name: EASTHead
+    model_name: large
+
+Loss:
+  name: EASTLoss
+  
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+  #  name: Cosine
+    learning_rate: 0.001
+  #  warmup_epoch: 0
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: EASTPostProcess
+  score_thresh: 0.8
+  cover_thresh: 0.1
+  nms_thresh: 0.2
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - EASTProcessTrain:
+          image_shape: [512, 512]
+          background_ratio: 0.125
+          min_crop_side_ratio: 0.1
+          min_text_size: 10
+      - KeepKeys:
+          keep_keys: ['image', 'score_map', 'geo_map', 'training_mask'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          Fa: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          limit_side_len: 2400
+          limit_type: max
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/test_tipc/configs/det_r50_vd_east_v2_0/train_infer_python.txt b/test_tipc/configs/det_r50_vd_east_v2_0/train_infer_python.txt
new file mode 100644
index 0000000..e3ef8c6
--- /dev/null
+++ b/test_tipc/configs/det_r50_vd_east_v2_0/train_infer_python.txt
@@ -0,0 +1,59 @@
+===========================train_params===========================
+model_name:det_r50_vd_east_v2_0
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=500
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:./pretrain_models/det_r50_vd_east_v2.0_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/det_r50_vd_east_v2_0/det_r50_vd_east.yml -o Global.pretrained_model=pretrain_models/det_r50_vd_east_v2.0_train/best_accuracy.pdparams Global.print_batch_step=1 Train.loader.shuffle=false
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/det_r50_vd_east_v2_0/det_r50_vd_east.yml -o 
+quant_export:null 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/det_r50_vd_east_v2.0_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/det_r50_vd_east_v2_0/det_r50_vd_east.yml  -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--save_log_path:null
+--benchmark:True
+--det_algorithm:EAST
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
+===========================train_benchmark_params==========================
+batch_size:8
+fp_items:fp32|fp16
+epoch:2
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
diff --git a/test_tipc/configs/det_r50_vd_pse_v2_0/det_r50_vd_pse.yml b/test_tipc/configs/det_r50_vd_pse_v2_0/det_r50_vd_pse.yml
new file mode 100644
index 0000000..c069d1f
--- /dev/null
+++ b/test_tipc/configs/det_r50_vd_pse_v2_0/det_r50_vd_pse.yml
@@ -0,0 +1,134 @@
+Global:
+  use_gpu: true
+  epoch_num: 600
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/det_r50_vd_pse/
+  save_epoch_step: 600
+  # evaluation is run every 125 iterations
+  eval_batch_step: [ 0,1000 ]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/ResNet50_vd_ssld_pretrained
+  checkpoints: #./output/det_r50_vd_pse_batch8_ColorJitter/best_accuracy
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_en/img_10.jpg
+  save_res_path: ./output/det_pse/predicts_pse.txt
+
+Architecture:
+  model_type: det
+  algorithm: PSE
+  Transform:
+  Backbone:
+    name: ResNet_vd
+    layers: 50
+  Neck:
+    name: FPN
+    out_channels: 256
+  Head:
+    name: PSEHead
+    hidden_dim: 256
+    out_channels: 7
+
+Loss:
+  name: PSELoss
+  alpha: 0.7
+  ohem_ratio: 3
+  kernel_sample_mask: pred
+  reduction: none
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Step
+    learning_rate: 0.0001
+    step_size: 200
+    gamma: 0.1
+  regularizer:
+    name: 'L2'
+    factor: 0.0005
+
+PostProcess:
+  name: PSEPostProcess
+  thresh: 0
+  box_thresh: 0.85
+  min_area: 16
+  box_type: quad # 'quad' or 'poly'
+  scale: 1
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - ColorJitter:
+          brightness: 0.12549019607843137
+          saturation: 0.5
+      - IaaAugment:
+          augmenter_args:
+            - { 'type': Resize, 'args': { 'size': [ 0.5, 3 ] } }
+            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
+            - { 'type': Affine, 'args': { 'rotate': [ -10, 10 ] } }
+      - MakePseGt:
+          kernel_num: 7
+          min_shrink_ratio: 0.4
+          size: 640
+      - RandomCropImgMask:
+          size: [ 640,640 ]
+          main_key: gt_text
+          crop_keys: [ 'image', 'gt_text', 'gt_kernels', 'mask' ]
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [ 0.485, 0.456, 0.406 ]
+          std: [ 0.229, 0.224, 0.225 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'image', 'gt_text', 'gt_kernels', 'mask' ] # the order of the dataloader list
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          limit_side_len: 736
+          limit_type: min
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [ 0.485, 0.456, 0.406 ]
+          std: [ 0.229, 0.224, 0.225 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'image', 'shape', 'polys', 'ignore_tags' ]
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 8
diff --git a/test_tipc/configs/det_r50_vd_pse_v2_0/train_infer_python.txt b/test_tipc/configs/det_r50_vd_pse_v2_0/train_infer_python.txt
new file mode 100644
index 0000000..6658905
--- /dev/null
+++ b/test_tipc/configs/det_r50_vd_pse_v2_0/train_infer_python.txt
@@ -0,0 +1,59 @@
+===========================train_params===========================
+model_name:det_r50_vd_pse_v2_0
+python:python3.7
+gpu_list:0
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=500
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/det_r50_vd_pse_v2_0/det_r50_vd_pse.yml -o Global.print_batch_step=1 Train.loader.shuffle=false
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/det_r50_vd_pse_v2_0/det_r50_vd_pse.yml -o 
+quant_export:null 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/det_r50_vd_pse_v2.0_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/det_r50_vd_pse_v2_0/det_r50_vd_pse.yml  -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--save_log_path:null
+--benchmark:True
+--det_algorithm:PSE
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
+===========================train_benchmark_params==========================
+batch_size:8
+fp_items:fp32|fp16
+epoch:2
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
diff --git a/test_tipc/configs/det_r50_vd_sast_icdar15_v2_0/det_r50_vd_sast_icdar2015.yml b/test_tipc/configs/det_r50_vd_sast_icdar15_v2_0/det_r50_vd_sast_icdar2015.yml
new file mode 100644
index 0000000..8055884
--- /dev/null
+++ b/test_tipc/configs/det_r50_vd_sast_icdar15_v2_0/det_r50_vd_sast_icdar2015.yml
@@ -0,0 +1,110 @@
+Global:
+  use_gpu: true
+  epoch_num: 5000
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/sast_r50_vd_ic15/
+  save_epoch_step: 1000
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [4000, 5000]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/ResNet50_vd_ssld_pretrained
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img:
+  save_res_path: ./output/sast_r50_vd_ic15/predicts_sast.txt
+
+
+Architecture:
+  model_type: det
+  algorithm: SAST
+  Transform:
+  Backbone:
+    name: ResNet_SAST
+    layers: 50
+  Neck:
+    name: SASTFPN
+    with_cab: True
+  Head:
+    name: SASTHead
+
+Loss:
+  name: SASTLoss
+  
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+  #  name: Cosine
+    learning_rate: 0.001
+  #  warmup_epoch: 0
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: SASTPostProcess
+  score_thresh: 0.5
+  sample_pts_num: 2
+  nms_thresh: 0.2
+  expand_scale: 1.0
+  shrink_ratio_of_width: 0.3
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - SASTProcessTrain:
+          image_shape: [512, 512]
+          min_crop_side_ratio: 0.3
+          min_crop_size: 24
+          min_text_size: 4
+          max_text_size: 512
+      - KeepKeys:
+          keep_keys: ['image', 'score_map', 'border_map', 'training_mask', 'tvo_map', 'tco_map'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 4
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          resize_long: 1536
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/test_tipc/configs/det_r50_vd_sast_icdar15_v2_0/train_infer_python.txt b/test_tipc/configs/det_r50_vd_sast_icdar15_v2_0/train_infer_python.txt
new file mode 100644
index 0000000..b01f192
--- /dev/null
+++ b/test_tipc/configs/det_r50_vd_sast_icdar15_v2_0/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:det_r50_vd_sast_icdar15_v2_0
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=500
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:./pretrain_models/det_r50_vd_sast_icdar15_v2.0_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/det_r50_vd_sast_icdar15_v2_0/det_r50_vd_sast_icdar2015.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/det_r50_vd_sast_icdar15_v2_0/det_r50_vd_sast_icdar2015.yml -o 
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:./inference/det_r50_vd_sast_icdar15_v2.0_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/det_r50_vd_sast_icdar15_v2_0/det_r50_vd_sast_icdar2015.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/00008790.jpg
+null:null
+--benchmark:True
+--det_algorithm:SAST
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/det_r50_vd_sast_totaltext_v2_0/det_r50_vd_sast_totaltext.yml b/test_tipc/configs/det_r50_vd_sast_totaltext_v2_0/det_r50_vd_sast_totaltext.yml
new file mode 100644
index 0000000..6c6a9f1
--- /dev/null
+++ b/test_tipc/configs/det_r50_vd_sast_totaltext_v2_0/det_r50_vd_sast_totaltext.yml
@@ -0,0 +1,108 @@
+Global:
+  use_gpu: true
+  epoch_num: 5000
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/sast_r50_vd_tt/
+  save_epoch_step: 1000
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [4000, 5000]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/ResNet50_vd_ssld_pretrained
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: 
+  save_res_path: ./output/sast_r50_vd_tt/predicts_sast.txt
+
+Architecture:
+  model_type: det
+  algorithm: SAST
+  Transform:
+  Backbone:
+    name: ResNet_SAST
+    layers: 50
+  Neck:
+    name: SASTFPN
+    with_cab: True
+  Head:
+    name: SASTHead
+
+Loss:
+  name: SASTLoss
+  
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+  #  name: Cosine
+    learning_rate: 0.001
+  #  warmup_epoch: 0
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: SASTPostProcess
+  score_thresh: 0.5
+  sample_pts_num: 6
+  nms_thresh: 0.2
+  expand_scale: 1.2
+  shrink_ratio_of_width: 0.2
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/total_text/train
+    label_file_list: [./train_data/total_text/train/train.txt]
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - SASTProcessTrain:
+          image_shape: [512, 512]
+          min_crop_side_ratio: 0.3
+          min_crop_size: 24
+          min_text_size: 4
+          max_text_size: 512
+      - KeepKeys:
+          keep_keys: ['image', 'score_map', 'border_map', 'training_mask', 'tvo_map', 'tco_map'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 4
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/
+    label_file_list: 
+      - ./train_data/total_text/test/test.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          resize_long: 768
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
diff --git a/test_tipc/configs/det_r50_vd_sast_totaltext_v2_0/train_infer_python.txt b/test_tipc/configs/det_r50_vd_sast_totaltext_v2_0/train_infer_python.txt
new file mode 100644
index 0000000..a47ad68
--- /dev/null
+++ b/test_tipc/configs/det_r50_vd_sast_totaltext_v2_0/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:det_r50_vd_sast_totaltext_v2_0
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=5000
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/det_r50_vd_sast_totaltext_v2_0/det_r50_vd_sast_totaltext.yml -o Global.pretrained_model=./pretrain_models/ResNet50_vd_ssld_pretrained
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/det_r50_vd_sast_totaltext_v2_0/det_r50_vd_sast_totaltext.yml -o 
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:./inference/det_r50_vd_sast_totaltext_v2.0_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/det_r50_vd_sast_totaltext_v2_0/det_r50_vd_sast_totaltext.yml -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+--det_algorithm:SAST
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/en_server_pgnetA/train_infer_python.txt b/test_tipc/configs/en_server_pgnetA/train_infer_python.txt
new file mode 100644
index 0000000..a9dd4e6
--- /dev/null
+++ b/test_tipc/configs/en_server_pgnetA/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:en_server_pgnetA
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=5|whole_train_whole_infer=500
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=14
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/total_text/test/rgb/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.pretrained_model=./pretrain_models/en_server_pgnetA/best_accuracy
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/e2e/e2e_r50_vd_pg.yml -o 
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+inference_dir:null
+train_model:./inference/en_server_pgnetA/best_accuracy
+infer_export:tools/export_model.py -c configs/e2e/e2e_r50_vd_pg.yml -o
+infer_quant:False
+inference:tools/infer/predict_e2e.py
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--e2e_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,640,640]}];[{float32,[3,960,960]}]
diff --git a/test_tipc/configs/en_table_structure/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/en_table_structure/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..4dd41e3
--- /dev/null
+++ b/test_tipc/configs/en_table_structure/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:en_table_structure
+use_opencv:True
+infer_model:./inference/en_ppocr_mobile_v2.0_table_structure_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr --rec_img_h=32 --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --limit_side_len=736 --limit_type=min --output=./output/table --merge_no_span_structure=False --type=structure --table=True
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:6
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:True
+--det:True
+--rec:True
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/en_table_structure/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt b/test_tipc/configs/en_table_structure/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
new file mode 100644
index 0000000..e81bae8
--- /dev/null
+++ b/test_tipc/configs/en_table_structure/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
@@ -0,0 +1,17 @@
+===========================paddle2onnx_params===========================
+model_name:en_table_structure
+python:python3.7
+2onnx: paddle2onnx
+--det_model_dir:./inference/en_ppocr_mobile_v2.0_table_structure_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_save_file:./inference/en_ppocr_mobile_v2.0_table_structure_infer/model.onnx
+--rec_model_dir:
+--rec_save_file:
+--opset_version:10
+--enable_onnx_checker:True
+inference:ppstructure/table/predict_structure.py --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt
+--use_gpu:True|False
+--det_model_dir:
+--rec_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
diff --git a/test_tipc/configs/en_table_structure/train_infer_python.txt b/test_tipc/configs/en_table_structure/train_infer_python.txt
new file mode 100644
index 0000000..8861ea8
--- /dev/null
+++ b/test_tipc/configs/en_table_structure/train_infer_python.txt
@@ -0,0 +1,59 @@
+===========================train_params===========================
+model_name:en_table_structure
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:./pretrain_models/en_ppocr_mobile_v2.0_table_structure_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./ppstructure/docs/table/table.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/table/table_mv3.yml -o Global.print_batch_step=1 Train.loader.shuffle=false Train.dataset.data_dir=./train_data/pubtabnet/train Train.dataset.label_file_list=[./train_data/pubtabnet/train.jsonl] Eval.dataset.data_dir=./train_data/pubtabnet/test Eval.dataset.label_file_list=[./train_data/pubtabnet/test.jsonl]
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/table/table_mv3.yml -o
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/en_ppocr_mobile_v2.0_table_structure_infer
+infer_export:null
+infer_quant:False
+inference:ppstructure/table/predict_table.py --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,488,488]}]
+===========================train_benchmark_params==========================
+batch_size:32
+fp_items:fp32|fp16
+epoch:2
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
diff --git a/test_tipc/configs/en_table_structure/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/en_table_structure/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..8e25b9d
--- /dev/null
+++ b/test_tipc/configs/en_table_structure/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:en_table_structure
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:./pretrain_models/en_ppocr_mobile_v2.0_table_structure_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./ppstructure/docs/table/table.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/table/table_mv3.yml -o Train.dataset.data_dir=./train_data/pubtabnet/train Train.dataset.label_file_list=[./train_data/pubtabnet/train.jsonl] Eval.dataset.data_dir=./train_data/pubtabnet/test Eval.dataset.label_file_list=[./train_data/pubtabnet/test.jsonl]
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/table/table_mv3.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/en_ppocr_mobile_v2.0_table_structure_infer
+infer_export:null
+infer_quant:False
+inference:ppstructure/table/predict_table.py --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
+--use_gpu:False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,488,488]}]
diff --git a/test_tipc/configs/en_table_structure/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/en_table_structure/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..a399e35
--- /dev/null
+++ b/test_tipc/configs/en_table_structure/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:en_table_structure
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:./pretrain_models/en_ppocr_mobile_v2.0_table_structure_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./ppstructure/docs/table/table.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/table/table_mv3.yml -o Train.dataset.data_dir=./train_data/pubtabnet/train Train.dataset.label_file_list=[./train_data/pubtabnet/train.jsonl] Eval.dataset.data_dir=./train_data/pubtabnet/test Eval.dataset.label_file_list=[./train_data/pubtabnet/test.jsonl]
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/table/table_mv3.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/en_ppocr_mobile_v2.0_table_structure_infer
+infer_export:null
+infer_quant:False
+inference:ppstructure/table/predict_table.py --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,488,488]}]
diff --git a/test_tipc/configs/en_table_structure/train_pact_infer_python.txt b/test_tipc/configs/en_table_structure/train_pact_infer_python.txt
new file mode 100644
index 0000000..0bb04c4
--- /dev/null
+++ b/test_tipc/configs/en_table_structure/train_pact_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:en_table_structure_PACT
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=2
+Global.pretrained_model:./pretrain_models/en_ppocr_mobile_v2.0_table_structure_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./ppstructure/docs/table/table.jpg
+null:null
+##
+trainer:pact_train
+norm_train:null 
+pact_train:deploy/slim/quantization/quant.py -c configs/table/table_mv3.yml -o Train.dataset.data_dir=./train_data/pubtabnet/train Train.dataset.label_file_list=[./train_data/pubtabnet/train.jsonl] Eval.dataset.data_dir=./train_data/pubtabnet/test Eval.dataset.label_file_list=[./train_data/pubtabnet/test.jsonl]
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null 
+quant_export:deploy/slim/quantization/export_model.py -c configs/table/table_mv3.yml -o 
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/en_ppocr_mobile_v2.0_table_structure_infer
+infer_export:null
+infer_quant:True
+inference:ppstructure/table/predict_table.py --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,488,488]}]
diff --git a/test_tipc/configs/en_table_structure/train_ptq_infer_python.txt b/test_tipc/configs/en_table_structure/train_ptq_infer_python.txt
new file mode 100644
index 0000000..aae0895
--- /dev/null
+++ b/test_tipc/configs/en_table_structure/train_ptq_infer_python.txt
@@ -0,0 +1,21 @@
+===========================train_params===========================
+model_name:en_table_structure_KL
+python:python3.7
+Global.pretrained_model:
+Global.save_inference_dir:null
+infer_model:./inference/en_ppocr_mobile_v2.0_table_structure_infer/
+infer_export:deploy/slim/quantization/quant_kl.py -c configs/table/table_mv3.yml -o
+infer_quant:True
+inference:ppstructure/table/predict_table.py --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:int8
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:False
+null:null
+null:null
diff --git a/test_tipc/configs/en_table_structure_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/en_table_structure_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..2f457dc
--- /dev/null
+++ b/test_tipc/configs/en_table_structure_PACT/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:en_table_structure_PACT
+use_opencv:True
+infer_model:./inference/en_ppocr_mobile_v2.0_table_structure_slim_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr --rec_img_h=32 --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --limit_side_len=736 --limit_type=min --output=./output/table --merge_no_span_structure=False --type=structure --table=True
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:6
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:True
+--det:True
+--rec:True
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/layoutxlm_ser/ser_layoutxlm_xfund_zh.yml b/test_tipc/configs/layoutxlm_ser/ser_layoutxlm_xfund_zh.yml
new file mode 100644
index 0000000..31e0ed4
--- /dev/null
+++ b/test_tipc/configs/layoutxlm_ser/ser_layoutxlm_xfund_zh.yml
@@ -0,0 +1,122 @@
+Global:
+  use_gpu: True
+  epoch_num: &epoch_num 200
+  log_smooth_window: 10
+  print_batch_step: 10
+  save_model_dir: ./output/ser_layoutxlm_xfund_zh
+  save_epoch_step: 2000
+  # evaluation is run every 10 iterations after the 0th iteration
+  eval_batch_step: [ 0, 187 ]
+  cal_metric_during_train: False
+  save_inference_dir:
+  use_visualdl: False
+  seed: 2022
+  infer_img: ppstructure/docs/kie/input/zh_val_42.jpg
+  save_res_path: ./output/ser_layoutxlm_xfund_zh/res
+
+Architecture:
+  model_type: kie
+  algorithm: &algorithm "LayoutXLM"
+  Transform:
+  Backbone:
+    name: LayoutXLMForSer
+    pretrained: True
+    checkpoints: 
+    num_classes: &num_classes 7
+
+Loss:
+  name: VQASerTokenLayoutLMLoss
+  num_classes: *num_classes
+  key: "backbone_out"
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Linear
+    learning_rate: 0.00005
+    epochs: *epoch_num
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 0.00000
+    
+PostProcess:
+  name: VQASerTokenLayoutLMPostProcess
+  class_path: &class_path train_data/XFUND/class_list_xfun.txt
+
+Metric:
+  name: VQASerTokenMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_train/image
+    label_file_list: 
+      - train_data/XFUND/zh_train/train.json
+    ratio_list: [ 1.0 ]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: &max_seq_len 512
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 16
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: train_data/XFUND/zh_val/image
+    label_file_list:
+      - train_data/XFUND/zh_val/val.json
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VQATokenLabelEncode: # Class handling label
+          contains_re: False
+          algorithm: *algorithm
+          class_path: *class_path
+      - VQATokenPad:
+          max_seq_len: *max_seq_len
+          return_attention_mask: True
+      - VQASerTokenChunk:
+          max_seq_len: *max_seq_len
+      - Resize:
+          size: [224,224]
+      - NormalizeImage:
+          scale: 1
+          mean: [ 123.675, 116.28, 103.53 ]
+          std: [ 58.395, 57.12, 57.375 ]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: [ 'input_ids', 'bbox', 'attention_mask', 'token_type_ids', 'image', 'labels'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 4
diff --git a/test_tipc/configs/layoutxlm_ser/train_infer_python.txt b/test_tipc/configs/layoutxlm_ser/train_infer_python.txt
new file mode 100644
index 0000000..d07daa9
--- /dev/null
+++ b/test_tipc/configs/layoutxlm_ser/train_infer_python.txt
@@ -0,0 +1,59 @@
+===========================train_params===========================
+model_name:layoutxlm_ser
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=17
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=4|whole_train_whole_infer=8
+Architecture.Backbone.checkpoints:null
+train_model_name:latest
+train_infer_img_dir:ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/layoutxlm_ser/ser_layoutxlm_xfund_zh.yml -o Global.print_batch_step=1 Global.eval_batch_step=[1000,1000] Train.loader.shuffle=false
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Architecture.Backbone.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/layoutxlm_ser/ser_layoutxlm_xfund_zh.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:null
+infer_export:null
+infer_quant:False
+inference:ppstructure/kie/predict_kie_token_ser.py  --kie_algorithm=LayoutXLM  --ser_dict_path=train_data/XFUND/class_list_xfun.txt --output=output
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--ser_model_dir:
+--image_dir:./ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,224,224]}]
+===========================train_benchmark_params==========================
+batch_size:8
+fp_items:fp32|fp16
+epoch:3
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98
diff --git a/test_tipc/configs/layoutxlm_ser/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/layoutxlm_ser/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..96b43ce
--- /dev/null
+++ b/test_tipc/configs/layoutxlm_ser/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:layoutxlm_ser
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=17
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=4|whole_train_whole_infer=8
+Architecture.Backbone.checkpoints:null
+train_model_name:latest
+train_infer_img_dir:ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/layoutxlm_ser/ser_layoutxlm_xfund_zh.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Architecture.Backbone.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/layoutxlm_ser/ser_layoutxlm_xfund_zh.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:null
+infer_export:null
+infer_quant:False
+inference:ppstructure/kie/predict_kie_token_ser.py  --kie_algorithm=LayoutXLM  --ser_dict_path=train_data/XFUND/class_list_xfun.txt --output=output
+--use_gpu:False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--ser_model_dir:
+--image_dir:./ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,224,224]}]
diff --git a/test_tipc/configs/layoutxlm_ser/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/layoutxlm_ser/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..8353957
--- /dev/null
+++ b/test_tipc/configs/layoutxlm_ser/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:layoutxlm_ser
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=17
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=4|whole_train_whole_infer=8
+Architecture.Backbone.checkpoints:null
+train_model_name:latest
+train_infer_img_dir:ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/layoutxlm_ser/ser_layoutxlm_xfund_zh.yml -o Global.print_batch_step=1 Global.eval_batch_step=[1000,1000] Train.loader.shuffle=false
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Architecture.Backbone.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/layoutxlm_ser/ser_layoutxlm_xfund_zh.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:null
+infer_export:null
+infer_quant:False
+inference:ppstructure/kie/predict_kie_token_ser.py  --kie_algorithm=LayoutXLM  --ser_dict_path=train_data/XFUND/class_list_xfun.txt --output=output
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--ser_model_dir:
+--image_dir:./ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,224,224]}]
diff --git a/test_tipc/configs/layoutxlm_ser/train_pact_infer_python.txt b/test_tipc/configs/layoutxlm_ser/train_pact_infer_python.txt
new file mode 100644
index 0000000..c19b4b7
--- /dev/null
+++ b/test_tipc/configs/layoutxlm_ser/train_pact_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:layoutxlm_ser_PACT
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=17
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=4|whole_train_whole_infer=8
+Architecture.Backbone.pretrained:pretrain_models/ser_LayoutXLM_xfun_zh
+train_model_name:latest
+train_infer_img_dir:ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+##
+trainer:pact_train
+norm_train:null
+pact_train:deploy/slim/quantization/quant.py -c test_tipc/configs/layoutxlm_ser/ser_layoutxlm_xfund_zh.yml -o Global.eval_batch_step=[2000,10]
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Architecture.Backbone.checkpoints:
+norm_export:null
+quant_export:deploy/slim/quantization/export_model.py -c test_tipc/configs/layoutxlm_ser/ser_layoutxlm_xfund_zh.yml -o 
+fpgm_export: null
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:null
+infer_export:null
+infer_quant:False
+inference:ppstructure/kie/predict_kie_token_ser.py  --kie_algorithm=LayoutXLM  --ser_dict_path=train_data/XFUND/class_list_xfun.txt --output=output
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--ser_model_dir:
+--image_dir:./ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,224,224]}]
diff --git a/test_tipc/configs/layoutxlm_ser/train_ptq_infer_python.txt b/test_tipc/configs/layoutxlm_ser/train_ptq_infer_python.txt
new file mode 100644
index 0000000..47e1e70
--- /dev/null
+++ b/test_tipc/configs/layoutxlm_ser/train_ptq_infer_python.txt
@@ -0,0 +1,21 @@
+===========================train_params===========================
+model_name:layoutxlm_ser_KL
+python:python3.7
+Global.pretrained_model:
+Global.save_inference_dir:null
+infer_model:./inference/ser_LayoutXLM_xfun_zh_infer/
+infer_export:deploy/slim/quantization/quant_kl.py -c test_tipc/configs/layoutxlm_ser/ser_layoutxlm_xfund_zh.yml -o Train.loader.batch_size_per_card=1 Eval.loader.batch_size_per_card=1
+infer_quant:True
+inference:ppstructure/kie/predict_kie_token_ser.py --kie_algorithm=LayoutXLM --ser_dict_path=./train_data/XFUND/class_list_xfun.txt
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:int8
+--ser_model_dir:
+--image_dir:./ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+--benchmark:False
+null:null
+null:null
diff --git a/test_tipc/configs/rec_d28_can/rec_d28_can.yml b/test_tipc/configs/rec_d28_can/rec_d28_can.yml
new file mode 100644
index 0000000..5501865
--- /dev/null
+++ b/test_tipc/configs/rec_d28_can/rec_d28_can.yml
@@ -0,0 +1,122 @@
+Global:
+  use_gpu: True
+  epoch_num: 240
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/can/
+  save_epoch_step: 1
+  # evaluation is run every 1105 iterations (1 epoch)(batch_size = 8)
+  eval_batch_step: [0, 1105]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/datasets/crohme_demo/hme_00.jpg
+  # for data or label process
+  character_dict_path: ppocr/utils/dict/latex_symbol_dict.txt
+  max_text_length: 36
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_can.txt
+
+Optimizer:
+  name: Momentum
+  momentum: 0.9
+  clip_norm_global: 100.0
+  lr:
+    name: TwoStepCosine
+    learning_rate: 0.01
+    warmup_epoch: 1
+  weight_decay: 0.0001
+
+Architecture:
+  model_type: rec
+  algorithm: CAN
+  in_channels: 1
+  Transform:
+  Backbone:
+    name: DenseNet 
+    growthRate: 24
+    reduction: 0.5
+    bottleneck: True
+    use_dropout: True
+    input_channel: 1 
+  Head:
+    name: CANHead
+    in_channel: 684
+    out_channel: 111
+    max_text_length: 36
+    ratio: 16
+    attdecoder:
+      is_train: True
+      input_size: 256
+      hidden_size: 256
+      encoder_out_channel: 684
+      dropout: True
+      dropout_ratio: 0.5
+      word_num: 111
+      counting_decoder_out_channel: 111
+      attention:
+        attention_dim: 512
+        word_conv_kernel: 1
+   
+Loss:
+  name: CANLoss
+
+PostProcess:
+  name: CANLabelDecode
+
+Metric:
+  name: CANMetric
+  main_indicator: exp_rate
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/CROHME_lite/training/images/
+    label_file_list: ["./train_data/CROHME_lite/training/labels.txt"]
+    transforms:
+      - DecodeImage:
+          channel_first: False
+      - NormalizeImage:
+          mean: [0,0,0]
+          std: [1,1,1]
+          order: 'hwc'
+      - GrayImageChannelFormat: 
+          inverse: True
+      - CANLabelEncode:
+          lower: False
+      - KeepKeys:
+          keep_keys: ['image', 'label']
+  loader:
+    shuffle: True
+    batch_size_per_card: 8
+    drop_last: False
+    num_workers: 4
+    collate_fn: DyMaskCollator
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/CROHME_lite/evaluation/images/
+    label_file_list: ["./train_data/CROHME_lite/evaluation/labels.txt"]
+    transforms: 
+      - DecodeImage:
+          channel_first: False
+      - NormalizeImage:
+          mean: [0,0,0]
+          std: [1,1,1]
+          order: 'hwc'
+      - GrayImageChannelFormat:
+          inverse: True
+      - CANLabelEncode:
+          lower: False
+      - KeepKeys:
+          keep_keys: ['image', 'label']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1
+    num_workers: 4
+    collate_fn: DyMaskCollator
diff --git a/test_tipc/configs/rec_d28_can/train_infer_python.txt b/test_tipc/configs/rec_d28_can/train_infer_python.txt
new file mode 100644
index 0000000..731d327
--- /dev/null
+++ b/test_tipc/configs/rec_d28_can/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_d28_can
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=240
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=8
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./doc/datasets/crohme_demo
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_d28_can/rec_d28_can.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_d28_can/rec_d28_can.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_d28_can/rec_d28_can.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_d28_can_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_d28_can/rec_d28_can.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/dict/latex_symbol_dict.txt --rec_algorithm="CAN"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./doc/datasets/crohme_demo
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[1,100,100]}]
diff --git a/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml b/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml
new file mode 100644
index 0000000..8118d58
--- /dev/null
+++ b/test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml
@@ -0,0 +1,103 @@
+Global:
+  use_gpu: True
+  epoch_num: 21
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/nrtr/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path: ppocr/utils/EN_symbol_dict.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_nrtr.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.99
+  clip_norm: 5.0
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 2
+  regularizer:
+    name: 'L2'
+    factor: 0.
+
+Architecture:
+  model_type: rec
+  algorithm: NRTR
+  in_channels: 1
+  Transform:
+  Backbone:
+    name: MTB
+    cnn_num: 2
+  Head:
+    name: Transformer
+    d_model: 512
+    num_encoder_layers: 6
+    beam_size: -1 # When Beam size is greater than 0, it means to use beam search when evaluation.
+    
+
+Loss:
+  name: CELoss
+  smoothing: True
+
+PostProcess:
+  name: NRTRLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - NRTRLabelEncode: # Class handling label
+      - GrayRecResizeImg:
+          image_shape: [100, 32]
+          resize_type: PIL # PIL or OpenCV
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 512
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - NRTRLabelEncode: # Class handling label
+      - GrayRecResizeImg:
+          image_shape: [100, 32]
+          resize_type: PIL # PIL or OpenCV
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
+    use_shared_memory: False
diff --git a/test_tipc/configs/rec_mtb_nrtr/train_infer_python.txt b/test_tipc/configs/rec_mtb_nrtr/train_infer_python.txt
new file mode 100644
index 0000000..4a8fda0
--- /dev/null
+++ b/test_tipc/configs/rec_mtb_nrtr/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_mtb_nrtr
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_mtb_nrtr_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_mtb_nrtr/rec_mtb_nrtr.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/EN_symbol_dict.txt --rec_image_shape="1,32,100" --rec_algorithm="NRTR"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[1,32,100]}]
diff --git a/test_tipc/configs/rec_mv3_none_bilstm_ctc_v2_0/rec_icdar15_train.yml b/test_tipc/configs/rec_mv3_none_bilstm_ctc_v2_0/rec_icdar15_train.yml
new file mode 100644
index 0000000..b0ba615
--- /dev/null
+++ b/test_tipc/configs/rec_mv3_none_bilstm_ctc_v2_0/rec_icdar15_train.yml
@@ -0,0 +1,97 @@
+Global:
+  use_gpu: True
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/mv3_none_bilstm_ctc/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 20000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_mv3_none_bilstm_ctc.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 96
+  Head:
+    name: CTCHead
+    fc_decay: 0
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
diff --git a/test_tipc/configs/rec_mv3_none_bilstm_ctc_v2_0/train_infer_python.txt b/test_tipc/configs/rec_mv3_none_bilstm_ctc_v2_0/train_infer_python.txt
new file mode 100644
index 0000000..22c29c9
--- /dev/null
+++ b/test_tipc/configs/rec_mv3_none_bilstm_ctc_v2_0/train_infer_python.txt
@@ -0,0 +1,59 @@
+===========================train_params===========================
+model_name:rec_mv3_none_bilstm_ctc_v2_0
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=5|whole_train_whole_infer=100
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_mv3_none_bilstm_ctc_v2_0/rec_icdar15_train.yml -o Global.print_batch_step=4 Train.loader.shuffle=false
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_mv3_none_bilstm_ctc_v2_0/rec_icdar15_train.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_mv3_none_bilstm_ctc_v2_0/rec_icdar15_train.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_mv3_none_bilstm_ctc_v2.0_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_mv3_none_bilstm_ctc_v2_0/rec_icdar15_train.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ic15_dict.txt --rec_image_shape="3,32,100"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]
+===========================train_benchmark_params==========================
+batch_size:256
+fp_items:fp32|fp16
+epoch:4
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
diff --git a/test_tipc/configs/rec_mv3_none_none_ctc_v2_0/rec_icdar15_train.yml b/test_tipc/configs/rec_mv3_none_none_ctc_v2_0/rec_icdar15_train.yml
new file mode 100644
index 0000000..06a2b71
--- /dev/null
+++ b/test_tipc/configs/rec_mv3_none_none_ctc_v2_0/rec_icdar15_train.yml
@@ -0,0 +1,96 @@
+Global:
+  use_gpu: True
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/mv3_none_none_ctc/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_mv3_none_none_ctc.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: Rosetta
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+    fc_decay: 0.0004
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/test_tipc/configs/rec_mv3_none_none_ctc_v2_0/train_infer_python.txt b/test_tipc/configs/rec_mv3_none_none_ctc_v2_0/train_infer_python.txt
new file mode 100644
index 0000000..0b66a97
--- /dev/null
+++ b/test_tipc/configs/rec_mv3_none_none_ctc_v2_0/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_mv3_none_none_ctc_v2_0
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=5|whole_train_whole_infer=100
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_mv3_none_none_ctc_v2_0/rec_icdar15_train.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_mv3_none_none_ctc_v2_0/rec_icdar15_train.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_mv3_none_none_ctc_v2_0/rec_icdar15_train.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_mv3_none_none_ctc_v2.0_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_mv3_none_none_ctc_v2_0/rec_icdar15_train.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ic15_dict.txt --rec_image_shape="3,32,100"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]
diff --git a/test_tipc/configs/rec_mv3_tps_bilstm_att_v2_0/rec_mv3_tps_bilstm_att.yml b/test_tipc/configs/rec_mv3_tps_bilstm_att_v2_0/rec_mv3_tps_bilstm_att.yml
new file mode 100644
index 0000000..2b14c04
--- /dev/null
+++ b/test_tipc/configs/rec_mv3_tps_bilstm_att_v2_0/rec_mv3_tps_bilstm_att.yml
@@ -0,0 +1,103 @@
+Global:
+  use_gpu: True
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/rec_mv3_tps_bilstm_att/
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_mv3_tps_bilstm_att.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0.00001
+
+Architecture:
+  model_type: rec
+  algorithm: RARE
+  Transform:
+    name: TPS
+    num_fiducial: 20
+    loc_lr: 0.1
+    model_name: small
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn 
+    hidden_size: 96
+  Head:
+    name: AttentionHead  
+    hidden_size: 96
+    
+
+Loss:
+  name: AttentionLoss
+
+PostProcess:
+  name: AttnLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - AttnLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - AttnLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 1
diff --git a/test_tipc/configs/rec_mv3_tps_bilstm_att_v2_0/train_infer_python.txt b/test_tipc/configs/rec_mv3_tps_bilstm_att_v2_0/train_infer_python.txt
new file mode 100644
index 0000000..77dc79c
--- /dev/null
+++ b/test_tipc/configs/rec_mv3_tps_bilstm_att_v2_0/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_mv3_tps_bilstm_att_v2_0
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_mv3_tps_bilstm_att_v2_0/rec_mv3_tps_bilstm_att.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_mv3_tps_bilstm_att_v2_0/rec_mv3_tps_bilstm_att.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_mv3_tps_bilstm_att_v2_0/rec_mv3_tps_bilstm_att.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_mv3_tps_bilstm_att_v2.0_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_mv3_tps_bilstm_att_v2_0/rec_mv3_tps_bilstm_att.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ic15_dict.txt --rec_image_shape="3,32,100" --rec_algorithm="RARE" --min_subgraph_size=5
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]
diff --git a/test_tipc/configs/rec_mv3_tps_bilstm_ctc_v2_0/rec_icdar15_train.yml b/test_tipc/configs/rec_mv3_tps_bilstm_ctc_v2_0/rec_icdar15_train.yml
new file mode 100644
index 0000000..e20ea00
--- /dev/null
+++ b/test_tipc/configs/rec_mv3_tps_bilstm_ctc_v2_0/rec_icdar15_train.yml
@@ -0,0 +1,101 @@
+Global:
+  use_gpu: True
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/mv3_tps_bilstm_ctc/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_mv3_tps_bilstm_ctc.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: STARNet
+  Transform:
+    name: TPS
+    num_fiducial: 20
+    loc_lr: 0.1
+    model_name: small
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 96
+  Head:
+    name: CTCHead
+    fc_decay: 0.0004
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
diff --git a/test_tipc/configs/rec_mv3_tps_bilstm_ctc_v2_0/train_infer_python.txt b/test_tipc/configs/rec_mv3_tps_bilstm_ctc_v2_0/train_infer_python.txt
new file mode 100644
index 0000000..f38c8d8
--- /dev/null
+++ b/test_tipc/configs/rec_mv3_tps_bilstm_ctc_v2_0/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_mv3_tps_bilstm_ctc_v2_0
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=5|whole_train_whole_infer=100
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_mv3_tps_bilstm_ctc_v2_0/rec_icdar15_train.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_mv3_tps_bilstm_ctc_v2_0/rec_icdar15_train.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_mv3_tps_bilstm_ctc_v2_0/rec_icdar15_train.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_mv3_tps_bilstm_ctc_v2.0_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_mv3_tps_bilstm_ctc_v2_0/rec_icdar15_train.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ic15_dict.txt --rec_image_shape="3,32,100" --rec_algorithm="StarNet"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]
diff --git a/test_tipc/configs/rec_parseq/rec_vit_parseq.yml b/test_tipc/configs/rec_parseq/rec_vit_parseq.yml
new file mode 100644
index 0000000..fc283ba
--- /dev/null
+++ b/test_tipc/configs/rec_parseq/rec_vit_parseq.yml
@@ -0,0 +1,118 @@
+Global:
+  use_gpu: True
+  epoch_num: 20
+  log_smooth_window: 20
+  print_batch_step: 5
+  save_model_dir: ./output/rec/parseq
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 500]
+  cal_metric_during_train: True
+  pretrained_model: 
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  character_type: en
+  max_text_length: 25
+  num_heads: 8
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_parseq.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: OneCycle
+    max_lr: 0.0007
+
+Architecture:
+  model_type: rec
+  algorithm: ParseQ
+  in_channels: 3
+  Transform:
+  Backbone:
+    name: ViTParseQ
+    img_size: [32, 128]
+    patch_size: [4, 8]
+    embed_dim: 384
+    depth: 12
+    num_heads: 6
+    mlp_ratio: 4
+    in_channels: 3
+  Head:
+    name: ParseQHead
+    # Architecture
+    max_text_length: 25
+    embed_dim: 384
+    dec_num_heads: 12
+    dec_mlp_ratio: 4
+    dec_depth: 1
+    # Training
+    perm_num: 6
+    perm_forward: true
+    perm_mirrored: true
+    dropout: 0.1
+    # Decoding mode (test)
+    decode_ar: true
+    refine_iters: 1
+
+Loss:
+  name: ParseQLoss
+
+PostProcess:
+  name: ParseQLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  is_filter: True
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ParseQRecAug:
+          aug_type: 0 # or 1
+      - ParseQLabelEncode:
+      - SVTRRecResizeImg:
+          image_shape: [3, 32, 128]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 192
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ParseQLabelEncode: # Class handling label
+      - SVTRRecResizeImg:
+          image_shape: [3, 32, 128]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] 
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 384
+    num_workers: 4
diff --git a/test_tipc/configs/rec_parseq/train_infer_python.txt b/test_tipc/configs/rec_parseq/train_infer_python.txt
new file mode 100644
index 0000000..2e441fc
--- /dev/null
+++ b/test_tipc/configs/rec_parseq/train_infer_python.txt
@@ -0,0 +1,61 @@
+===========================train_params===========================
+model_name:rec_parseq
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_parseq/rec_vit_parseq.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_parseq/rec_vit_parseq.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_parseq/rec_vit_parseq.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_parseq_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_parseq/rec_vit_parseq.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ic15_dict.txt --rec_image_shape="3,32,128" --rec_algorithm="ParseQ"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,128]}]
+===========================train_benchmark_params==========================
+batch_size:512
+fp_items:fp32|fp16
+epoch:2
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
+===========================to_static_train_benchmark_params===========================
+to_static_train:Global.to_static=true
diff --git a/test_tipc/configs/rec_r31_robustscanner/rec_r31_robustscanner.yml b/test_tipc/configs/rec_r31_robustscanner/rec_r31_robustscanner.yml
new file mode 100644
index 0000000..b5466d4
--- /dev/null
+++ b/test_tipc/configs/rec_r31_robustscanner/rec_r31_robustscanner.yml
@@ -0,0 +1,111 @@
+Global:
+  use_gpu: true
+  epoch_num: 5
+  log_smooth_window: 20
+  print_batch_step: 20
+  save_model_dir: ./output/rec/rec_r31_robustscanner/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: ./inference/rec_inference
+  # for data or label process
+  character_dict_path: ppocr/utils/dict90.txt
+  max_text_length: &max_text_length 40
+  infer_mode: False
+  use_space_char: False
+  rm_symbol: True
+  save_res_path: ./output/rec/predicts_robustscanner.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Piecewise
+    decay_epochs: [3, 4]
+    values: [0.001, 0.0001, 0.00001] 
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: RobustScanner
+  Transform:
+  Backbone:
+    name: ResNet31
+    init_type: KaimingNormal
+  Head:
+    name: RobustScannerHead
+    enc_outchannles: 128
+    hybrid_dec_rnn_layers: 2
+    hybrid_dec_dropout: 0
+    position_dec_rnn_layers: 2
+    start_idx: 91
+    mask: True
+    padding_idx: 92
+    encode_value: False
+    max_text_length: *max_text_length
+
+Loss:
+  name: SARLoss
+
+PostProcess:
+  name: SARLabelDecode
+
+Metric:
+  name: RecMetric
+  is_filter: True
+
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SARLabelEncode: # Class handling label
+      - RobustScannerRecResizeImg:
+          image_shape: [3, 48, 48, 160] # h:48 w:[48,160]
+          width_downsample_ratio: 0.25
+          max_text_length: *max_text_length
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'valid_ratio', 'word_positons'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 16
+    drop_last: True
+    num_workers: 0
+    use_shared_memory: False
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SARLabelEncode: # Class handling label
+      - RobustScannerRecResizeImg:
+          image_shape: [3, 48, 48, 160]
+          max_text_length: *max_text_length
+          width_downsample_ratio: 0.25
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'valid_ratio', 'word_positons'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 16
+    num_workers: 0
+    use_shared_memory: False
+  
diff --git a/test_tipc/configs/rec_r31_robustscanner/train_infer_python.txt b/test_tipc/configs/rec_r31_robustscanner/train_infer_python.txt
new file mode 100644
index 0000000..c2e2f2b
--- /dev/null
+++ b/test_tipc/configs/rec_r31_robustscanner/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_r31_robustscanner
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=5
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_r31_robustscanner/rec_r31_robustscanner.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_r31_robustscanner/rec_r31_robustscanner.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_r31_robustscanner/rec_r31_robustscanner.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_r31_robustscanner/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_r31_robustscanner/rec_r31_robustscanner.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/dict90.txt --rec_image_shape="3,48,48,160" --use_space_char=False --rec_algorithm="RobustScanner"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,48,160]}]
diff --git a/test_tipc/configs/rec_r31_sar/rec_r31_sar.yml b/test_tipc/configs/rec_r31_sar/rec_r31_sar.yml
new file mode 100644
index 0000000..36bc3c5
--- /dev/null
+++ b/test_tipc/configs/rec_r31_sar/rec_r31_sar.yml
@@ -0,0 +1,98 @@
+Global:
+  use_gpu: true
+  epoch_num: 5
+  log_smooth_window: 20
+  print_batch_step: 20
+  save_model_dir: ./sar_rec
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: 
+  # for data or label process
+  character_dict_path: ppocr/utils/dict90.txt
+  max_text_length: 30
+  infer_mode: False
+  use_space_char: False
+  rm_symbol: True
+  save_res_path: ./output/rec/predicts_sar.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Piecewise
+    decay_epochs: [3, 4]
+    values: [0.001, 0.0001, 0.00001] 
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: SAR
+  Transform:
+  Backbone:
+    name: ResNet31
+  Head:
+    name: SARHead
+
+Loss:
+  name: SARLoss
+
+PostProcess:
+  name: SARLabelDecode
+
+Metric:
+  name: RecMetric
+
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SARLabelEncode: # Class handling label
+      - SARRecResizeImg:
+          image_shape: [3, 48, 48, 160] # h:48 w:[48,160]
+          width_downsample_ratio: 0.25
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'valid_ratio'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 64
+    drop_last: True
+    num_workers: 8
+    use_shared_memory: False
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SARLabelEncode: # Class handling label
+      - SARRecResizeImg:
+          image_shape: [3, 48, 48, 160]
+          width_downsample_ratio: 0.25
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'valid_ratio'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 64
+    num_workers: 4
+    use_shared_memory: False
+  
diff --git a/test_tipc/configs/rec_r31_sar/train_infer_python.txt b/test_tipc/configs/rec_r31_sar/train_infer_python.txt
new file mode 100644
index 0000000..4acc622
--- /dev/null
+++ b/test_tipc/configs/rec_r31_sar/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_r31_sar
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_r31_sar/rec_r31_sar.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_r31_sar/rec_r31_sar.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_r31_sar/rec_r31_sar.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_r31_sar_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_r31_sar/rec_r31_sar.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/dict90.txt --rec_image_shape="3,48,48,160" --rec_algorithm="SAR"
+--use_gpu:True
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,48,160]}]
diff --git a/test_tipc/configs/rec_r32_gaspin_bilstm_att/rec_r32_gaspin_bilstm_att.yml b/test_tipc/configs/rec_r32_gaspin_bilstm_att/rec_r32_gaspin_bilstm_att.yml
new file mode 100644
index 0000000..21d56b6
--- /dev/null
+++ b/test_tipc/configs/rec_r32_gaspin_bilstm_att/rec_r32_gaspin_bilstm_att.yml
@@ -0,0 +1,117 @@
+Global:
+  use_gpu: True
+  epoch_num: 6
+  log_smooth_window: 50
+  print_batch_step: 50
+  save_model_dir: ./output/rec/rec_r32_gaspin_bilstm_att/
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model: pretrain_models/rec_r32_gaspin_bilstm_att_train/best_accuracy
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  # for data or label process
+  character_dict_path: ./ppocr/utils/dict/spin_dict.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_r32_gaspin_bilstm_att.txt
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Piecewise
+    decay_epochs: [3, 4, 5]
+    values: [0.001, 0.0003, 0.00009, 0.000027] 
+
+  clip_norm: 5
+
+Architecture:
+  model_type: rec
+  algorithm: SPIN
+  in_channels: 1
+  Transform:
+    name: GA_SPIN
+    offsets: True
+    default_type: 6
+    loc_lr: 0.1
+    stn: True
+  Backbone:
+    name: ResNet32
+    out_channels: 512
+  Neck:
+    name: SequenceEncoder
+    encoder_type: cascadernn 
+    hidden_size: 256
+    out_channels: [256, 512]
+    with_linear: True
+  Head:
+    name: SPINAttentionHead  
+    hidden_size: 256
+    
+
+Loss:
+  name: SPINAttentionLoss
+  ignore_index: 0
+
+PostProcess:
+  name: SPINLabelDecode
+  use_space_char: False
+
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+  is_filter: True
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SPINLabelEncode: # Class handling label
+      - SPINRecResizeImg:
+          image_shape: [100, 32]
+          interpolation : 2
+          mean: [127.5]
+          std: [127.5]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 128
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SPINLabelEncode: # Class handling label
+      - SPINRecResizeImg:
+          image_shape: [100, 32]
+          interpolation : 2
+          mean: [127.5]
+          std: [127.5]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1
+    num_workers: 1
diff --git a/test_tipc/configs/rec_r32_gaspin_bilstm_att/train_infer_python.txt b/test_tipc/configs/rec_r32_gaspin_bilstm_att/train_infer_python.txt
new file mode 100644
index 0000000..ac378b3
--- /dev/null
+++ b/test_tipc/configs/rec_r32_gaspin_bilstm_att/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_r32_gaspin_bilstm_att
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_r32_gaspin_bilstm_att/rec_r32_gaspin_bilstm_att.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_r32_gaspin_bilstm_att/rec_r32_gaspin_bilstm_att.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_r32_gaspin_bilstm_att/rec_r32_gaspin_bilstm_att.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_r32_gaspin_bilstm_att/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_r32_gaspin_bilstm_att/rec_r32_gaspin_bilstm_att.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/dict/spin_dict.txt --use_space_char=False --rec_image_shape="3,32,100" --rec_algorithm="SPIN"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]
diff --git a/test_tipc/configs/rec_r34_vd_none_bilstm_ctc_v2_0/rec_icdar15_train.yml b/test_tipc/configs/rec_r34_vd_none_bilstm_ctc_v2_0/rec_icdar15_train.yml
new file mode 100644
index 0000000..3e9ea48
--- /dev/null
+++ b/test_tipc/configs/rec_r34_vd_none_bilstm_ctc_v2_0/rec_icdar15_train.yml
@@ -0,0 +1,96 @@
+Global:
+  use_gpu: true
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/r34_vd_none_bilstm_ctc/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_r34_vd_none_bilstm_ctc.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: CRNN
+  Transform:
+  Backbone:
+    name: ResNet
+    layers: 34
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 256
+  Head:
+    name: CTCHead
+    fc_decay: 0
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
diff --git a/test_tipc/configs/rec_r34_vd_none_bilstm_ctc_v2_0/train_infer_python.txt b/test_tipc/configs/rec_r34_vd_none_bilstm_ctc_v2_0/train_infer_python.txt
new file mode 100644
index 0000000..b9c6b6f
--- /dev/null
+++ b/test_tipc/configs/rec_r34_vd_none_bilstm_ctc_v2_0/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_r34_vd_none_bilstm_ctc_v2_0
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=5|whole_train_whole_infer=100
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_r34_vd_none_bilstm_ctc_v2_0/rec_icdar15_train.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_r34_vd_none_bilstm_ctc_v2_0/rec_icdar15_train.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_r34_vd_none_bilstm_ctc_v2_0/rec_icdar15_train.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_r34_vd_none_bilstm_ctc_v2.0_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_r34_vd_none_bilstm_ctc_v2_0/rec_icdar15_train.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ic15_dict.txt --rec_image_shape="3,32,100"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]
diff --git a/test_tipc/configs/rec_r34_vd_none_none_ctc_v2_0/rec_icdar15_train.yml b/test_tipc/configs/rec_r34_vd_none_none_ctc_v2_0/rec_icdar15_train.yml
new file mode 100644
index 0000000..ce15e27
--- /dev/null
+++ b/test_tipc/configs/rec_r34_vd_none_none_ctc_v2_0/rec_icdar15_train.yml
@@ -0,0 +1,94 @@
+Global:
+  use_gpu: true
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/r34_vd_none_none_ctc/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_r34_vd_none_none_ctc.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: Rosetta
+  Backbone:
+    name: ResNet
+    layers: 34
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+    fc_decay: 0.0004
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
diff --git a/test_tipc/configs/rec_r34_vd_none_none_ctc_v2_0/train_infer_python.txt b/test_tipc/configs/rec_r34_vd_none_none_ctc_v2_0/train_infer_python.txt
new file mode 100644
index 0000000..cd336f0
--- /dev/null
+++ b/test_tipc/configs/rec_r34_vd_none_none_ctc_v2_0/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_r34_vd_none_none_ctc_v2_0
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=5|whole_train_whole_infer=100
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_r34_vd_none_none_ctc_v2_0/rec_icdar15_train.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_r34_vd_none_none_ctc_v2_0/rec_icdar15_train.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_r34_vd_none_none_ctc_v2_0/rec_icdar15_train.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_r34_vd_none_none_ctc_v2.0_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_r34_vd_none_none_ctc_v2_0/rec_icdar15_train.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ic15_dict.txt --rec_image_shape="3,32,100"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]
diff --git a/test_tipc/configs/rec_r34_vd_tps_bilstm_att_v2_0/rec_r34_vd_tps_bilstm_att.yml b/test_tipc/configs/rec_r34_vd_tps_bilstm_att_v2_0/rec_r34_vd_tps_bilstm_att.yml
new file mode 100644
index 0000000..5dd797b
--- /dev/null
+++ b/test_tipc/configs/rec_r34_vd_tps_bilstm_att_v2_0/rec_r34_vd_tps_bilstm_att.yml
@@ -0,0 +1,102 @@
+Global:
+  use_gpu: True
+  epoch_num: 400
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/b3_rare_r34_none_gru/
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_b3_rare_r34_none_gru.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0.00000
+
+Architecture:
+  model_type: rec
+  algorithm: RARE
+  Transform:
+    name: TPS
+    num_fiducial: 20
+    loc_lr: 0.1
+    model_name: large
+  Backbone:
+    name: ResNet  
+    layers: 34
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn 
+    hidden_size: 256 #96
+  Head:
+    name: AttentionHead  # AttentionHead
+    hidden_size: 256 #
+    l2_decay: 0.00001
+
+Loss:
+  name: AttentionLoss
+
+PostProcess:
+  name: AttnLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - AttnLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - AttnLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/test_tipc/configs/rec_r34_vd_tps_bilstm_att_v2_0/train_infer_python.txt b/test_tipc/configs/rec_r34_vd_tps_bilstm_att_v2_0/train_infer_python.txt
new file mode 100644
index 0000000..0910ff8
--- /dev/null
+++ b/test_tipc/configs/rec_r34_vd_tps_bilstm_att_v2_0/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_r34_vd_tps_bilstm_att_v2_0
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_r34_vd_tps_bilstm_att_v2_0/rec_r34_vd_tps_bilstm_att.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_r34_vd_tps_bilstm_att_v2_0/rec_r34_vd_tps_bilstm_att.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_r34_vd_tps_bilstm_att_v2_0/rec_r34_vd_tps_bilstm_att.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_r34_vd_tps_bilstm_att_v2.0_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_r34_vd_tps_bilstm_att_v2_0/rec_r34_vd_tps_bilstm_att.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ic15_dict.txt --rec_image_shape="3,32,100" --rec_algorithm="RARE" --min_subgraph_size=5
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]
diff --git a/test_tipc/configs/rec_r34_vd_tps_bilstm_ctc_v2_0/rec_icdar15_train.yml b/test_tipc/configs/rec_r34_vd_tps_bilstm_ctc_v2_0/rec_icdar15_train.yml
new file mode 100644
index 0000000..2dcba81
--- /dev/null
+++ b/test_tipc/configs/rec_r34_vd_tps_bilstm_ctc_v2_0/rec_icdar15_train.yml
@@ -0,0 +1,100 @@
+Global:
+  use_gpu: true
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/r34_vd_tps_bilstm_ctc/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_r34_vd_tps_bilstm_ctc.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    learning_rate: 0.0005
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: STARNet
+  Transform:
+    name: TPS
+    num_fiducial: 20
+    loc_lr: 0.1
+    model_name: large
+  Backbone:
+    name: ResNet
+    layers: 34
+  Neck:
+    name: SequenceEncoder
+    encoder_type: rnn
+    hidden_size: 256
+  Head:
+    name: CTCHead
+    fc_decay: 0
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 256
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - RecResizeImg:
+          image_shape: [3, 32, 100]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
diff --git a/test_tipc/configs/rec_r34_vd_tps_bilstm_ctc_v2_0/train_infer_python.txt b/test_tipc/configs/rec_r34_vd_tps_bilstm_ctc_v2_0/train_infer_python.txt
new file mode 100644
index 0000000..33144e6
--- /dev/null
+++ b/test_tipc/configs/rec_r34_vd_tps_bilstm_ctc_v2_0/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_r34_vd_tps_bilstm_ctc_v2_0
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=5|whole_train_whole_infer=100
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=128|whole_train_whole_infer=128
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_r34_vd_tps_bilstm_ctc_v2_0/rec_icdar15_train.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_r34_vd_tps_bilstm_ctc_v2_0/rec_icdar15_train.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_r34_vd_tps_bilstm_ctc_v2_0/rec_icdar15_train.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_r34_vd_tps_bilstm_ctc_v2.0_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_r34_vd_tps_bilstm_ctc_v2_0/rec_icdar15_train.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ic15_dict.txt --rec_image_shape="3,32,100" --rec_algorithm="StarNet"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,100]}]
diff --git a/test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml b/test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml
new file mode 100644
index 0000000..5b5890e
--- /dev/null
+++ b/test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml
@@ -0,0 +1,106 @@
+Global:
+  use_gpu: True
+  epoch_num: 10
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/r45_abinet/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  character_type: en
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_abinet.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.99
+  clip_norm: 20.0
+  lr:
+    name: Piecewise
+    decay_epochs: [6]
+    values: [0.0001, 0.00001] 
+  regularizer:
+    name: 'L2'
+    factor: 0.
+
+Architecture:
+  model_type: rec
+  algorithm: ABINet
+  in_channels: 3
+  Transform:
+  Backbone:
+    name: ResNet45
+
+  Head:
+    name: ABINetHead
+    use_lang: True
+    iter_size: 3
+    
+
+Loss:
+  name: CELoss
+  ignore_index: &ignore_index 100 # Must be greater than the number of character classes
+
+PostProcess:
+  name: ABINetLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - ABINetRecAug:
+      - ABINetLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - ABINetRecResizeImg:
+          image_shape: [3, 32, 128]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 96
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - ABINetLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - ABINetRecResizeImg:
+          image_shape: [3, 32, 128]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 4
+    use_shared_memory: False
diff --git a/test_tipc/configs/rec_r45_abinet/train_infer_python.txt b/test_tipc/configs/rec_r45_abinet/train_infer_python.txt
new file mode 100644
index 0000000..04fc188
--- /dev/null
+++ b/test_tipc/configs/rec_r45_abinet/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_abinet
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_r45_abinet_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_r45_abinet/rec_r45_abinet.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ic15_dict.txt --rec_image_shape="3,32,128" --rec_algorithm="ABINet"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,32,128]}]
diff --git a/test_tipc/configs/rec_r45_visionlan/rec_r45_visionlan.yml b/test_tipc/configs/rec_r45_visionlan/rec_r45_visionlan.yml
new file mode 100644
index 0000000..860e4f5
--- /dev/null
+++ b/test_tipc/configs/rec_r45_visionlan/rec_r45_visionlan.yml
@@ -0,0 +1,108 @@
+Global:
+  use_gpu: true
+  epoch_num: 8
+  log_smooth_window: 200
+  print_batch_step: 200
+  save_model_dir: ./output/rec/r45_visionlan
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words/en/word_2.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: &max_text_length 25
+  training_step: &training_step LA
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_visionlan.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  clip_norm: 20.0
+  group_lr: true
+  training_step: *training_step
+  lr:
+    name: Piecewise
+    decay_epochs: [6]
+    values: [0.0001, 0.00001] 
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+Architecture:
+  model_type: rec
+  algorithm: VisionLAN
+  Transform:
+  Backbone:
+    name: ResNet45
+    strides: [2, 2, 2, 1, 1]
+  Head:
+    name: VLHead
+    n_layers: 3
+    n_position: 256
+    n_dim: 512
+    max_text_length: *max_text_length
+    training_step: *training_step
+
+Loss:
+  name: VLLoss
+  mode: *training_step
+  weight_res: 0.5
+  weight_mas: 0.5
+
+PostProcess:
+  name: VLLabelDecode
+
+Metric:
+  name: RecMetric
+  is_filter: true
+
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - ABINetRecAug:
+      - VLLabelEncode: # Class handling label
+      - VLRecResizeImg:
+          image_shape: [3, 64, 256]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'label_res', 'label_sub', 'label_id', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 220
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: RGB
+          channel_first: False
+      - VLLabelEncode: # Class handling label
+      - VLRecResizeImg:
+          image_shape: [3, 64, 256]
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'label_res', 'label_sub', 'label_id', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 64
+    num_workers: 4
+  
diff --git a/test_tipc/configs/rec_r45_visionlan/train_infer_python.txt b/test_tipc/configs/rec_r45_visionlan/train_infer_python.txt
new file mode 100644
index 0000000..79618ed
--- /dev/null
+++ b/test_tipc/configs/rec_r45_visionlan/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_r45_visionlan
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=32|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_r45_visionlan/rec_r45_visionlan.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_r45_visionlan/rec_r45_visionlan.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_r45_visionlan/rec_r45_visionlan.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_r45_visionlan_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_r45_visionlan/rec_r45_visionlan.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ic15_dict.txt --rec_image_shape="3,64,256" --rec_algorithm="VisionLAN" --use_space_char=False
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,64,256]}]
diff --git a/test_tipc/configs/rec_r50_fpn_vd_none_srn/rec_r50_fpn_srn.yml b/test_tipc/configs/rec_r50_fpn_vd_none_srn/rec_r50_fpn_srn.yml
new file mode 100644
index 0000000..41e5252
--- /dev/null
+++ b/test_tipc/configs/rec_r50_fpn_vd_none_srn/rec_r50_fpn_srn.yml
@@ -0,0 +1,108 @@
+Global:
+  use_gpu: True
+  epoch_num: 72
+  log_smooth_window: 20
+  print_batch_step: 5
+  save_model_dir: ./output/rec/srn_new
+  save_epoch_step: 3
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 5000]
+  cal_metric_during_train: True
+  pretrained_model: 
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words/ch/word_1.jpg
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  num_heads: 8
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_srn.txt
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  clip_norm: 10.0
+  lr:
+    learning_rate: 0.0001
+
+Architecture:
+  model_type: rec
+  algorithm: SRN
+  in_channels: 1
+  Transform:
+  Backbone:
+    name: ResNetFPN
+  Head:
+    name: SRNHead
+    max_text_length: 25
+    num_heads: 8
+    num_encoder_TUs: 2
+    num_decoder_TUs: 4
+    hidden_dims: 512
+
+Loss:
+  name: SRNLoss
+
+PostProcess:
+  name: SRNLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SRNLabelEncode: # Class handling label
+      - SRNRecResizeImg:
+          image_shape: [1, 64, 256]
+      - KeepKeys:
+          keep_keys: ['image',
+                      'label',
+                      'length',
+                      'encoder_word_pos',
+                      'gsrm_word_pos',
+                      'gsrm_slf_attn_bias1',
+                      'gsrm_slf_attn_bias2'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    batch_size_per_card: 64
+    drop_last: False
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - SRNLabelEncode: # Class handling label
+      - SRNRecResizeImg:
+          image_shape: [1, 64, 256]
+      - KeepKeys:
+          keep_keys: ['image',
+                      'label',
+                      'length',
+                      'encoder_word_pos',
+                      'gsrm_word_pos',
+                      'gsrm_slf_attn_bias1',
+                      'gsrm_slf_attn_bias2'] 
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 32
+    num_workers: 4
diff --git a/test_tipc/configs/rec_r50_fpn_vd_none_srn/train_infer_python.txt b/test_tipc/configs/rec_r50_fpn_vd_none_srn/train_infer_python.txt
new file mode 100644
index 0000000..c1cfd1f
--- /dev/null
+++ b/test_tipc/configs/rec_r50_fpn_vd_none_srn/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_r50_fpn_vd_none_srn
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_r50_fpn_vd_none_srn/rec_r50_fpn_srn.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_r50_fpn_vd_none_srn/rec_r50_fpn_srn.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_r50_fpn_vd_none_srn/rec_r50_fpn_srn.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_r50_vd_srn_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_r50_fpn_vd_none_srn/rec_r50_fpn_srn.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ic15_dict.txt --rec_image_shape="1,64,256" --rec_algorithm="SRN" --use_space_char=False --min_subgraph_size=3
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[1,64,256]}]
diff --git a/test_tipc/configs/rec_resnet_rfl/rec_resnet_rfl.yml b/test_tipc/configs/rec_resnet_rfl/rec_resnet_rfl.yml
new file mode 100644
index 0000000..b4f18f5
--- /dev/null
+++ b/test_tipc/configs/rec_resnet_rfl/rec_resnet_rfl.yml
@@ -0,0 +1,111 @@
+Global:
+  use_gpu: True
+  epoch_num: 6
+  log_smooth_window: 20
+  print_batch_step: 50
+  save_model_dir: ./output/rec/rec_resnet_rfl/
+  save_epoch_step: 1
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [0, 5000]
+  cal_metric_during_train: False
+  pretrained_model:
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/rec_resnet_rfl.txt
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  weight_decay: 0.0
+  clip_norm_global: 5.0
+  lr:
+    name: Piecewise
+    decay_epochs : [3, 4, 5]
+    values : [0.001, 0.0003, 0.00009, 0.000027]
+
+Architecture:
+  model_type: rec
+  algorithm: RFL
+  in_channels: 1
+  Transform:
+    name: TPS
+    num_fiducial: 20
+    loc_lr: 1.0
+    model_name: large
+  Backbone:
+    name: ResNetRFL
+    use_cnt: True
+    use_seq: True
+  Neck:
+    name: RFAdaptor
+    use_v2s: True
+    use_s2v: True
+  Head:
+    name: RFLHead  
+    in_channels: 512
+    hidden_size: 256
+    batch_max_legnth: 25
+    out_channels: 38
+    use_cnt: True
+    use_seq: True
+
+Loss:
+  name: RFLLoss
+
+PostProcess:
+  name: RFLLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - RFLLabelEncode: # Class handling label
+      - RFLRecResizeImg:
+          image_shape: [1, 32, 100]
+          interpolation: 2
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length', 'cnt_label'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 64
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - RFLLabelEncode: # Class handling label
+      - RFLRecResizeImg:
+          image_shape: [1, 32, 100]
+          interpolation: 2
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length', 'cnt_label'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 8
diff --git a/test_tipc/configs/rec_resnet_rfl/train_infer_python.txt b/test_tipc/configs/rec_resnet_rfl/train_infer_python.txt
new file mode 100644
index 0000000..091e962
--- /dev/null
+++ b/test_tipc/configs/rec_resnet_rfl/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_resnet_rfl
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_resnet_rfl/rec_resnet_rfl.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_resnet_rfl/rec_resnet_rfl.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_resnet_rfl/rec_resnet_rfl.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_resnet_rfl_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_resnet_rfl/rec_resnet_rfl.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_image_shape="1,32,100" --rec_algorithm="RFL" --min_subgraph_size=5
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[1,32,100]}]
diff --git a/test_tipc/configs/rec_svtrnet/rec_svtrnet.yml b/test_tipc/configs/rec_svtrnet/rec_svtrnet.yml
new file mode 100644
index 0000000..6e22bc7
--- /dev/null
+++ b/test_tipc/configs/rec_svtrnet/rec_svtrnet.yml
@@ -0,0 +1,118 @@
+Global:
+  use_gpu: True
+  epoch_num: 20
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/svtr/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations after the 0th iteration
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path:
+  character_type: en
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_svtr_tiny.txt
+  d2s_train_image_shape: [3, 64, 256]
+
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.99
+  epsilon: 8.e-8
+  weight_decay: 0.05
+  no_weight_decay_name: norm pos_embed
+  one_dim_param_no_weight_decay: true
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 2
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR
+  Transform:
+    name: STN_ON
+    tps_inputsize: [32, 64]
+    tps_outputsize: [32, 100]
+    num_control_points: 20
+    tps_margins: [0.05,0.05]
+    stn_activation: none
+  Backbone:
+    name: SVTRNet
+    img_size: [32, 100]
+    out_char_num: 25
+    out_channels: 192
+    patch_merging: 'Conv'
+    embed_dim: [64, 128, 256]
+    depth: [3, 6, 3]
+    num_heads: [2, 4, 8]
+    mixer: ['Local','Local','Local','Local','Local','Local','Global','Global','Global','Global','Global','Global']
+    local_mixer: [[7, 11], [7, 11], [7, 11]]
+    last_stage: True
+    prenorm: false
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+
+Loss:
+  name: CTCLoss
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - SVTRRecResizeImg:
+          image_shape: [3, 64, 256]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 512
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - CTCLabelEncode: # Class handling label
+      - SVTRRecResizeImg:
+          image_shape: [3, 64, 256]
+          padding: False
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 2
diff --git a/test_tipc/configs/rec_svtrnet/train_infer_python.txt b/test_tipc/configs/rec_svtrnet/train_infer_python.txt
new file mode 100644
index 0000000..04ac03a
--- /dev/null
+++ b/test_tipc/configs/rec_svtrnet/train_infer_python.txt
@@ -0,0 +1,61 @@
+===========================train_params===========================
+model_name:rec_svtrnet
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_svtrnet/rec_svtrnet.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_svtrnet/rec_svtrnet.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_svtrnet/rec_svtrnet.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_svtrnet_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_svtrnet/rec_svtrnet.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/ic15_dict.txt --rec_image_shape="3,64,256" --rec_algorithm="SVTR"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,64,256]}]
+===========================train_benchmark_params==========================
+batch_size:512
+fp_items:fp32|fp16
+epoch:2
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
+===========================to_static_train_benchmark_params===========================
+to_static_train:Global.to_static=true
diff --git a/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml b/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml
new file mode 100644
index 0000000..a0aed48
--- /dev/null
+++ b/test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml
@@ -0,0 +1,104 @@
+Global:
+  use_gpu: True
+  epoch_num: 20
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/rec/vitstr_none_ce/
+  save_epoch_step: 1
+  # evaluation is run every 2000 iterations after the 0th iteration#
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_10.png
+  # for data or label process
+  character_dict_path: ppocr/utils/EN_symbol_dict.txt
+  max_text_length: 25
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/rec/predicts_vitstr.txt
+
+
+Optimizer:
+  name: Adadelta
+  epsilon: 1.e-8
+  rho: 0.95
+  clip_norm: 5.0
+  lr:
+    learning_rate: 1.0
+
+Architecture:
+  model_type: rec
+  algorithm: ViTSTR
+  in_channels: 1
+  Transform:
+  Backbone:
+    name: ViTSTR
+  Neck:
+    name: SequenceEncoder
+    encoder_type: reshape
+  Head:
+    name: CTCHead
+
+Loss:
+  name: CELoss
+  smoothing: False
+  with_all: True
+  ignore_index: &ignore_index 0 # Must be zero or greater than the number of character classes
+
+PostProcess:
+  name: ViTSTRLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data/
+    label_file_list: ["./train_data/ic15_data/rec_gt_train.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ViTSTRLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - GrayRecResizeImg:
+          image_shape: [224, 224] # W H
+          resize_type: PIL # PIL or OpenCV
+          inter_type: 'Image.BICUBIC'
+          scale: false
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    batch_size_per_card: 48
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/ic15_data
+    label_file_list: ["./train_data/ic15_data/rec_gt_test.txt"]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - ViTSTRLabelEncode: # Class handling label
+          ignore_index: *ignore_index
+      - GrayRecResizeImg:
+          image_shape: [224, 224] # W H
+          resize_type: PIL # PIL or OpenCV
+          inter_type: 'Image.BICUBIC'
+          scale: false
+      - KeepKeys:
+          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 256
+    num_workers: 2
diff --git a/test_tipc/configs/rec_vitstr_none_ce/train_infer_python.txt b/test_tipc/configs/rec_vitstr_none_ce/train_infer_python.txt
new file mode 100644
index 0000000..187c115
--- /dev/null
+++ b/test_tipc/configs/rec_vitstr_none_ce/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:rec_vitstr
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=2|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=64
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:tools/eval.py -c test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml -o
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/rec_vitstr_none_ce_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/rec_vitstr_none_ce/rec_vitstr_none_ce.yml -o
+infer_quant:False
+inference:tools/infer/predict_rec.py --rec_char_dict_path=./ppocr/utils/EN_symbol_dict.txt --rec_image_shape="1,224,224" --rec_algorithm="ViTSTR"
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--rec_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[1,224,224]}]
diff --git a/test_tipc/configs/slanet/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt b/test_tipc/configs/slanet/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
new file mode 100644
index 0000000..1829b21
--- /dev/null
+++ b/test_tipc/configs/slanet/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
@@ -0,0 +1,20 @@
+===========================cpp_infer_params===========================
+model_name:slanet
+use_opencv:True
+infer_model:./inference/ch_ppstructure_mobile_v2.0_SLANet_infer/
+infer_quant:False
+inference:./deploy/cpp_infer/build/ppocr --det_model_dir=./inference/PP-OCRv3_mobile_det_infer --rec_model_dir=./inference/PP-OCRv3_mobile_rec_infer --output=./output/table --type=structure --table=True --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict_ch.txt
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:6
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:True
+--det:True
+--rec:True
+--cls:False
+--use_angle_cls:False
diff --git a/test_tipc/configs/slanet/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt b/test_tipc/configs/slanet/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
new file mode 100644
index 0000000..99062b0
--- /dev/null
+++ b/test_tipc/configs/slanet/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
@@ -0,0 +1,17 @@
+===========================paddle2onnx_params===========================
+model_name:slanet
+python:python3.7
+2onnx: paddle2onnx
+--det_model_dir:./inference/ch_ppstructure_mobile_v2.0_SLANet_infer/
+--model_filename:inference.pdmodel
+--params_filename:inference.pdiparams
+--det_save_file:./inference/ch_ppstructure_mobile_v2.0_SLANet_infer/model.onnx
+--rec_model_dir:
+--rec_save_file:
+--opset_version:10
+--enable_onnx_checker:True
+inference:ppstructure/table/predict_structure.py --table_char_dict_path=./ppocr/utils/dict/table_structure_dict_ch.txt
+--use_gpu:True|False
+--det_model_dir:
+--rec_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
diff --git a/test_tipc/configs/slanet/train_infer_python.txt b/test_tipc/configs/slanet/train_infer_python.txt
new file mode 100644
index 0000000..e596e44
--- /dev/null
+++ b/test_tipc/configs/slanet/train_infer_python.txt
@@ -0,0 +1,61 @@
+===========================train_params===========================
+model_name:slanet
+python:python
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:./pretrain_models/ch_ppstructure_mobile_v2.0_SLANet_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./ppstructure/docs/table/table.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c configs/table/SLANet_ch.yml -o Global.cal_metric_during_train=False Global.print_batch_step=1 Train.loader.shuffle=false Train.dataset.data_dir=./train_data/pubtabnet Train.dataset.label_file_list=[./train_data/pubtabnet/train.txt] Eval.dataset.data_dir=./train_data/pubtabnet Eval.dataset.label_file_list=[./train_data/pubtabnet/val.txt] Global.d2s_train_image_shape='[3, -1, -1]'
+pact_train:null
+fpgm_train:null
+distill_train:null
+to_static_train:Global.to_static=true
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c configs/table/SLANet.yml -o
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/en_ppstructure_mobile_v2.0_SLANet_train
+infer_export:null
+infer_quant:False
+inference:ppstructure/table/predict_table.py --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,488,488]}]
+===========================train_benchmark_params==========================
+batch_size:64
+fp_items:fp32|fp16
+epoch:1
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
+===========================to_static_train_benchmark_params===========================
+to_static_train:Global.to_static=true
diff --git a/test_tipc/configs/slanet/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/slanet/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..4c9d8d6
--- /dev/null
+++ b/test_tipc/configs/slanet/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:slanet
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:./pretrain_models/en_ppstructure_mobile_v2.0_SLANet_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./ppstructure/docs/table/table.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/slanet/SLANet.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/slanet/SLANet.yml -o
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/en_ppstructure_mobile_v2.0_SLANet_train
+infer_export:null
+infer_quant:False
+inference:ppstructure/table/predict_table.py --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
+--use_gpu:False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,488,488]}]
diff --git a/test_tipc/configs/slanet/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/slanet/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..f8d3eff
--- /dev/null
+++ b/test_tipc/configs/slanet/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:slanet
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=3|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=128
+Global.pretrained_model:./pretrain_models/en_ppstructure_mobile_v2.0_SLANet_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./ppstructure/docs/table/table.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/slanet/SLANet.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/slanet/SLANet.yml -o
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/en_ppstructure_mobile_v2.0_SLANet_train
+infer_export:null
+infer_quant:False
+inference:ppstructure/table/predict_table.py --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:True
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,488,488]}]
diff --git a/test_tipc/configs/slanet/train_pact_infer_python.txt b/test_tipc/configs/slanet/train_pact_infer_python.txt
new file mode 100644
index 0000000..98546af
--- /dev/null
+++ b/test_tipc/configs/slanet/train_pact_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:slanet_PACT
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=50
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=2
+Global.pretrained_model:./pretrain_models/en_ppstructure_mobile_v2.0_SLANet_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./ppstructure/docs/table/table.jpg
+null:null
+##
+trainer:pact_train
+norm_train:null 
+pact_train:deploy/slim/quantization/quant.py -c test_tipc/configs/slanet/SLANet.yml  -o 
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:null 
+quant_export:deploy/slim/quantization/export_model.py -c test_tipc/configs/slanet/SLANet.yml  -o 
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:./inference/en_ppstructure_mobile_v2.0_SLANet_infer
+infer_export:null
+infer_quant:True
+inference:ppstructure/table/predict_table.py --det_model_dir=./inference/en_ppocr_mobile_v2.0_table_det_infer --rec_model_dir=./inference/en_ppocr_mobile_v2.0_table_rec_infer  --rec_char_dict_path=./ppocr/utils/dict/table_dict.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,488,488]}]
diff --git a/test_tipc/configs/slanet/train_ptq_infer_python.txt b/test_tipc/configs/slanet/train_ptq_infer_python.txt
new file mode 100644
index 0000000..05d148e
--- /dev/null
+++ b/test_tipc/configs/slanet/train_ptq_infer_python.txt
@@ -0,0 +1,21 @@
+===========================train_params===========================
+model_name:slanet_KL
+python:python3.7
+Global.pretrained_model:
+Global.save_inference_dir:null
+infer_model:./inference/en_ppstructure_mobile_v2.0_SLANet_infer/
+infer_export:deploy/slim/quantization/quant_kl.py -c test_tipc/configs/slanet/SLANet.yml -o
+infer_quant:True
+inference:ppstructure/table/predict_table.py --det_model_dir=./inference/PP-OCRv3_mobile_det_infer --rec_model_dir=./inference/PP-OCRv3_mobile_rec_infer  --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --table_char_dict_path=./ppocr/utils/dict/table_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --det_limit_side_len=736 --det_limit_type=min --output ./output/table
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:int8
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:True
+null:null
+null:null
diff --git a/test_tipc/configs/sr_telescope/sr_telescope.yml b/test_tipc/configs/sr_telescope/sr_telescope.yml
new file mode 100644
index 0000000..774c19d
--- /dev/null
+++ b/test_tipc/configs/sr_telescope/sr_telescope.yml
@@ -0,0 +1,83 @@
+Global:
+  use_gpu: true
+  epoch_num: 2
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/sr/sr_telescope/
+  save_epoch_step: 3
+  # evaluation is run every 2000 iterations
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: False
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:  ./output/sr/sr_telescope/infer
+  use_visualdl: False
+  infer_img: doc/imgs_words_en/word_52.png
+  # for data or label process
+  character_dict_path:
+  max_text_length: 100
+  infer_mode: False
+  use_space_char: False
+  save_res_path: ./output/sr/predicts_telescope.txt
+
+Optimizer:
+  name: Adam
+  beta1: 0.5
+  beta2: 0.999
+  clip_norm: 0.25
+  lr:
+    learning_rate: 0.0001
+
+Architecture:
+  model_type: sr
+  algorithm: Telescope
+  Transform:
+    name: TBSRN
+    STN: True
+    infer_mode: False
+
+Loss:
+  name: TelescopeLoss
+  confuse_dict_path: ./ppocr/utils/dict/confuse.pkl
+
+
+PostProcess:
+  name: None
+
+Metric:
+  name: SRMetric
+  main_indicator: all
+
+Train:
+  dataset:
+    name: LMDBDataSetSR
+    data_dir: ./train_data/TextZoom/test
+    transforms:
+      - SRResize:
+          imgH: 32
+          imgW: 128
+          down_sample_scale: 2
+      - KeepKeys:
+          keep_keys: ['img_lr', 'img_hr', 'label'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    batch_size_per_card: 16
+    drop_last: True
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: LMDBDataSetSR
+    data_dir: ./train_data/TextZoom/test
+    transforms:
+      - SRResize:
+          imgH: 32
+          imgW: 128
+          down_sample_scale: 2
+      - KeepKeys:
+          keep_keys: ['img_lr', 'img_hr', 'label'] # dataloader will return list in this order
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 16
+    num_workers: 4
diff --git a/test_tipc/configs/sr_telescope/train_infer_python.txt b/test_tipc/configs/sr_telescope/train_infer_python.txt
new file mode 100644
index 0000000..7235f07
--- /dev/null
+++ b/test_tipc/configs/sr_telescope/train_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:sr_telescope
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:null
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=300
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=16|whole_train_whole_infer=16
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./inference/rec_inference
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/sr_telescope/sr_telescope.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params===========================
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/sr_telescope/sr_telescope.yml -o
+quant_export:null
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/sr_telescope_train/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/configs/sr_telescope/sr_telescope.yml -o
+infer_quant:False
+inference:tools/infer/predict_sr.py --sr_image_shape="1,32,128" --rec_algorithm="Telescope" --min_subgraph_size=5
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--sr_model_dir:
+--image_dir:./inference/rec_inference
+--save_log_path:./test/output/
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[1,32,128]}]
diff --git a/test_tipc/configs/table_master/table_master.yml b/test_tipc/configs/table_master/table_master.yml
new file mode 100644
index 0000000..cc96fbc
--- /dev/null
+++ b/test_tipc/configs/table_master/table_master.yml
@@ -0,0 +1,133 @@
+Global:
+  use_gpu: true
+  epoch_num: 17
+  log_smooth_window: 20
+  print_batch_step: 100
+  save_model_dir: ./output/table_master/
+  save_epoch_step: 17
+  eval_batch_step: [0,  6259]
+  cal_metric_during_train: false
+  pretrained_model: null
+  checkpoints: 
+  save_inference_dir: output/table_master/infer
+  use_visualdl: false
+  infer_img: ppstructure/docs/table/table.jpg
+  save_res_path: ./output/table_master
+  character_dict_path: ppocr/utils/dict/table_master_structure_dict.txt
+  infer_mode: false
+  max_text_length: 500
+  d2s_train_image_shape: [3, 480, 480]
+
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: MultiStepDecay
+    learning_rate: 0.001
+    milestones: [12, 15]
+    gamma: 0.1
+    warmup_epoch: 0.02
+  regularizer:
+    name: L2
+    factor: 0.0
+
+Architecture:
+  model_type: table
+  algorithm: TableMaster
+  Backbone:
+    name: TableResNetExtra
+    gcb_config:
+      ratio: 0.0625
+      headers: 1
+      att_scale: False
+      fusion_type: channel_add
+      layers: [False, True, True, True]
+    layers: [1,2,5,3]
+  Head:
+    name: TableMasterHead
+    hidden_size: 512
+    headers: 8
+    dropout: 0
+    d_ff: 2024
+    max_text_length: 500
+
+Loss:
+  name: TableMasterLoss
+  ignore_index: 42 # set to len of dict + 3
+
+PostProcess:
+  name: TableMasterLabelDecode
+  box_shape: pad
+
+Metric:
+  name: TableMetric
+  main_indicator: acc
+  compute_bbox_metric: False
+
+Train:
+  dataset:
+    name: LMDBDataSetTableMaster
+    data_dir: train_data/StructureLabel_val_500/
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: False
+      - TableMasterLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: False
+          replace_empty_cell_token: True
+      - ResizeTableImage:
+          max_len: 480
+          resize_bboxes: True
+      - PaddingTableImage:
+          size: [480, 480]
+      - TableBoxEncode:
+          box_format: 'xywh'
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.5, 0.5, 0.5]
+          std: [0.5, 0.5, 0.5]
+          order: hwc
+      - ToCHWImage: null
+      - KeepKeys:
+          keep_keys: [image, structure, bboxes, bbox_masks, shape]
+  loader:
+    shuffle: True
+    batch_size_per_card: 10
+    drop_last: True
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: LMDBDataSetTableMaster
+    data_dir: train_data/StructureLabel_val_500/
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: False
+      - TableMasterLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: False
+          replace_empty_cell_token: True
+      - ResizeTableImage:
+          max_len: 480
+          resize_bboxes: True
+      - PaddingTableImage:
+          size: [480, 480]
+      - TableBoxEncode:
+          box_format: 'xywh'
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.5, 0.5, 0.5]
+          std: [0.5, 0.5, 0.5]
+          order: hwc
+      - ToCHWImage: null
+      - KeepKeys:
+          keep_keys: [image, structure, bboxes, bbox_masks, shape]
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 10
+    num_workers: 8
diff --git a/test_tipc/configs/table_master/train_infer_python.txt b/test_tipc/configs/table_master/train_infer_python.txt
new file mode 100644
index 0000000..a83a031
--- /dev/null
+++ b/test_tipc/configs/table_master/train_infer_python.txt
@@ -0,0 +1,61 @@
+===========================train_params===========================
+model_name:table_master
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=17
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:./pretrain_models/table_structure_tablemaster_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./ppstructure/docs/table/table.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/table_master/table_master.yml -o Global.print_batch_step=1
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/table_master/table_master.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:null
+infer_export:null
+infer_quant:False
+inference:ppstructure/table/predict_structure.py  --table_char_dict_path=./ppocr/utils/dict/table_master_structure_dict.txt --output ./output/table --table_algorithm=TableMaster --table_max_len=480 
+--use_gpu:True
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,480,480]}]
+===========================train_benchmark_params==========================
+batch_size:10
+fp_items:fp32|fp16
+epoch:2
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98;FLAGS_conv_workspace_size_limit=4096
+===========================to_static_train_benchmark_params===========================
+to_static_train:Global.to_static=true
diff --git a/test_tipc/configs/vi_layoutxlm_ser/train_infer_python.txt b/test_tipc/configs/vi_layoutxlm_ser/train_infer_python.txt
new file mode 100644
index 0000000..e304519
--- /dev/null
+++ b/test_tipc/configs/vi_layoutxlm_ser/train_infer_python.txt
@@ -0,0 +1,61 @@
+===========================train_params===========================
+model_name:vi_layoutxlm_ser
+python:python
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=17
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=4|whole_train_whole_infer=8
+Architecture.Backbone.checkpoints:null
+train_model_name:latest
+train_infer_img_dir:ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c ./configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o Global.print_batch_step=1 Global.eval_batch_step=[1000,1000] Train.loader.shuffle=false Train.loader.num_workers=32 Eval.loader.num_workers=32
+pact_train:null
+fpgm_train:null
+distill_train:null
+to_static_train:Global.to_static=true
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Architecture.Backbone.checkpoints:
+norm_export:tools/export_model.py -c ./configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:null
+infer_export:null
+infer_quant:False
+inference:ppstructure/kie/predict_kie_token_ser.py  --kie_algorithm=LayoutXLM  --ser_dict_path=train_data/XFUND/class_list_xfun.txt --output=output --ocr_order_method=tb-yx
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--ser_model_dir:
+--image_dir:./ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,224,224]}]
+===========================train_benchmark_params==========================
+batch_size:8
+fp_items:fp32|fp16
+epoch:3
+--profiler_options:batch_range=[10,20];state=GPU;tracer_option=Default;profile_path=model.profile
+flags:FLAGS_eager_delete_tensor_gb=0.0;FLAGS_fraction_of_gpu_memory_to_use=0.98
+===========================to_static_train_benchmark_params===========================
+to_static_train:Global.to_static=true
diff --git a/test_tipc/configs/vi_layoutxlm_ser/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/vi_layoutxlm_ser/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..74953ec
--- /dev/null
+++ b/test_tipc/configs/vi_layoutxlm_ser/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:vi_layoutxlm_ser
+python:python3.7
+gpu_list:192.168.0.1,192.168.0.2;0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=17
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=4|whole_train_whole_infer=8
+Architecture.Backbone.checkpoints:null
+train_model_name:latest
+train_infer_img_dir:ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c ./configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o 
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Architecture.Backbone.checkpoints:
+norm_export:tools/export_model.py -c ./configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:null
+infer_export:null
+infer_quant:False
+inference:ppstructure/kie/predict_kie_token_ser.py  --kie_algorithm=LayoutXLM  --ser_dict_path=train_data/XFUND/class_list_xfun.txt --output=output --ocr_order_method=tb-yx
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--ser_model_dir:
+--image_dir:./ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,224,224]}]
diff --git a/test_tipc/configs/vi_layoutxlm_ser/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt b/test_tipc/configs/vi_layoutxlm_ser/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
new file mode 100644
index 0000000..95daa02
--- /dev/null
+++ b/test_tipc/configs/vi_layoutxlm_ser/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:vi_layoutxlm_ser
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:amp
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=17
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=4|whole_train_whole_infer=8
+Architecture.Backbone.checkpoints:null
+train_model_name:latest
+train_infer_img_dir:ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c ./configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Architecture.Backbone.checkpoints:
+norm_export:tools/export_model.py -c ./configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:null
+infer_export:null
+infer_quant:False
+inference:ppstructure/kie/predict_kie_token_ser.py  --kie_algorithm=LayoutXLM  --ser_dict_path=train_data/XFUND/class_list_xfun.txt --output=output --ocr_order_method=tb-yx
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--ser_model_dir:
+--image_dir:./ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,224,224]}]
diff --git a/test_tipc/configs/vi_layoutxlm_ser/train_pact_infer_python.txt b/test_tipc/configs/vi_layoutxlm_ser/train_pact_infer_python.txt
new file mode 100644
index 0000000..c1fc48c
--- /dev/null
+++ b/test_tipc/configs/vi_layoutxlm_ser/train_pact_infer_python.txt
@@ -0,0 +1,53 @@
+===========================train_params===========================
+model_name:vi_layoutxlm_ser_PACT
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=17
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=4|whole_train_whole_infer=8
+Architecture.Backbone.pretrained:./pretrain_models/ser_vi_layoutxlm_xfund_pretrained/best_accuracy
+train_model_name:latest
+train_infer_img_dir:ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+##
+trainer:pact_train
+norm_train:null
+pact_train:deploy/slim/quantization/quant.py -c ./configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o Global.eval_batch_step=[2000,10]
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Architecture.Backbone.checkpoints:
+norm_export:null
+quant_export:deploy/slim/quantization/export_model.py -c ./configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o 
+fpgm_export: null
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:null
+infer_export:null
+infer_quant:False
+inference:ppstructure/kie/predict_kie_token_ser.py  --kie_algorithm=LayoutXLM  --ser_dict_path=train_data/XFUND/class_list_xfun.txt --output=output --ocr_order_method=tb-yx
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--ser_model_dir:
+--image_dir:./ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+--benchmark:True
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,224,224]}]
diff --git a/test_tipc/configs/vi_layoutxlm_ser/train_ptq_infer_python.txt b/test_tipc/configs/vi_layoutxlm_ser/train_ptq_infer_python.txt
new file mode 100644
index 0000000..faae871
--- /dev/null
+++ b/test_tipc/configs/vi_layoutxlm_ser/train_ptq_infer_python.txt
@@ -0,0 +1,21 @@
+===========================train_params===========================
+model_name:vi_layoutxlm_ser_KL
+python:python3.7
+Global.pretrained_model:
+Global.save_inference_dir:null
+infer_model:./inference/ser_vi_layoutxlm_xfund_infer/
+infer_export:deploy/slim/quantization/quant_kl.py -c ./configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yml -o Train.loader.batch_size_per_card=1 Eval.loader.batch_size_per_card=1
+infer_quant:True
+inference:ppstructure/kie/predict_kie_token_ser.py  --kie_algorithm=LayoutXLM  --ser_dict_path=train_data/XFUND/class_list_xfun.txt --output=output --ocr_order_method=tb-yx
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:int8
+--ser_model_dir:
+--image_dir:./ppstructure/docs/kie/input/zh_val_42.jpg
+null:null
+--benchmark:True
+null:null
+null:null
diff --git a/test_tipc/docs/benchmark_train.md b/test_tipc/docs/benchmark_train.md
new file mode 100644
index 0000000..3c22e42
--- /dev/null
+++ b/test_tipc/docs/benchmark_train.md
@@ -0,0 +1,76 @@
+
+# TIPC Linux端Benchmark测试文档
+
+该文档为Benchmark测试说明，Benchmark预测功能测试的主程序为`benchmark_train.sh`，用于验证监控模型训练的性能。
+
+# 1. 测试流程
+## 1.1 准备数据和环境安装
+运行`test_tipc/prepare.sh`，完成训练数据准备和安装环境流程。
+
+```shell
+# 运行格式：bash test_tipc/prepare.sh  train_benchmark.txt  mode
+bash test_tipc/prepare.sh test_tipc/configs/det_mv3_db_v2_0/train_infer_python.txt benchmark_train
+```
+
+## 1.2 功能测试
+执行`test_tipc/benchmark_train.sh`，完成模型训练和日志解析
+
+```shell
+# 运行格式：bash test_tipc/benchmark_train.sh train_benchmark.txt mode
+bash test_tipc/benchmark_train.sh test_tipc/configs/det_mv3_db_v2_0/train_infer_python.txt benchmark_train
+
+```
+
+`test_tipc/benchmark_train.sh`支持根据传入的第三个参数实现只运行某一个训练配置，如下：
+```shell
+# 运行格式：bash test_tipc/benchmark_train.sh train_benchmark.txt mode
+bash test_tipc/benchmark_train.sh test_tipc/configs/det_mv3_db_v2_0/train_infer_python.txt benchmark_train  dynamic_bs8_fp32_DP_N1C1
+```
+dynamic_bs8_fp32_DP_N1C1为test_tipc/benchmark_train.sh传入的参数，格式如下：
+`${modeltype}_${batch_size}_${fp_item}_${run_mode}_${device_num}`
+包含的信息有：模型类型、batchsize大小、训练精度如fp32,fp16等、分布式运行模式以及分布式训练使用的机器信息如单机单卡（N1C1）。
+
+
+## 2. 日志输出
+
+运行后将保存模型的训练日志和解析日志，使用 `test_tipc/configs/det_mv3_db_v2_0/train_infer_python.txt` 参数文件的训练日志解析结果是：
+
+```
+{"model_branch": "dygaph", "model_commit": "7c39a1996b19087737c05d883fd346d2f39dbcc0", "model_name": "det_mv3_db_v2_0_bs8_fp32_SingleP_DP", "batch_size": 8, "fp_item": "fp32", "run_process_type": "SingleP", "run_mode": "DP", "convergence_value": "5.413110", "convergence_key": "loss:", "ips": 19.333, "speed_unit": "samples/s", "device_num": "N1C1", "model_run_time": "0", "frame_commit": "8cc09552473b842c651ead3b9848d41827a3dbab", "frame_version": "0.0.0"}
+```
+
+训练日志和日志解析结果保存在benchmark_log目录下，文件组织格式如下：
+```
+train_log/
+├── index
+│   ├── PaddleOCR_det_mv3_db_v2_0_bs8_fp32_SingleP_DP_N1C1_speed
+│   └── PaddleOCR_det_mv3_db_v2_0_bs8_fp32_SingleP_DP_N1C4_speed
+├── profiling_log
+│   └── PaddleOCR_det_mv3_db_v2_0_bs8_fp32_SingleP_DP_N1C1_profiling
+└── train_log
+    ├── PaddleOCR_det_mv3_db_v2_0_bs8_fp32_SingleP_DP_N1C1_log
+    └── PaddleOCR_det_mv3_db_v2_0_bs8_fp32_SingleP_DP_N1C4_log
+```
+## 3. 各模型单卡性能数据一览
+
+*注：本节中的速度指标均使用单卡（1块Nvidia V100 16G GPU）测得。通常情况下。
+
+
+|模型名称|配置文件|大数据集 float32 fps |小数据集 float32 fps |diff |大数据集 float16 fps|小数据集 float16 fps| diff | 大数据集大小 | 小数据集大小 |
+|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|:-:|
+| ch_ppocr_mobile_v2.0_det |[config](../configs/ch_ppocr_mobile_v2.0_det/train_infer_python.txt) | 53.836 | 53.343 / 53.914 / 52.785 |0.020940758 | 45.574 | 45.57 / 46.292 / 46.213 | 0.015596647 | 10,000| 2,000|
+| ch_ppocr_mobile_v2.0_rec |[config](../configs/ch_ppocr_mobile_v2.0_rec/train_infer_python.txt) | 2083.311 | 2043.194  / 2066.372 / 2093.317 |0.023944295 | 2153.261 | 2167.561 /  2165.726 /  2155.614| 0.005511725 | 600,000| 160,000|
+| ch_ppocr_server_v2.0_det |[config](../configs/ch_ppocr_server_v2.0_det/train_infer_python.txt) | 20.716 | 20.739 /    20.807 /    20.755 |0.003268131 | 20.592 | 20.498 / 20.993 /    20.75| 0.023579288 | 10,000| 2,000|
+| ch_ppocr_server_v2.0_rec |[config](../configs/ch_ppocr_server_v2.0_rec/train_infer_python.txt) | 528.56 | 528.386 /   528.991 /   528.391 |0.001143687 | 1189.788 | 1190.007 /    1176.332 /  1192.084| 0.013213834 |  600,000| 160,000|
+| ch_PP-OCRv2_det    |[config](../configs/ch_PP-OCRv2_det/train_infer_python.txt) | 13.87 | 13.386 /    13.529 /    13.428 |0.010569887 | 17.847 | 17.746 / 17.908 /    17.96| 0.011915367 | 10,000| 2,000|
+| ch_PP-OCRv2_rec    |[config](../configs/ch_PP-OCRv2_rec/train_infer_python.txt) | 109.248 | 106.32 /  106.318 /   108.587 |0.020895687 | 117.491 | 117.62 /   117.757 /   117.726| 0.001163413 | 140,000| 40,000|
+| det_mv3_db_v2.0    |[config](../configs/det_mv3_db_v2_0/train_infer_python.txt) | 61.802 | 62.078 /   61.802 /    62.008 |0.00444602 | 82.947 | 84.294 /  84.457 /    84.005| 0.005351836 | 10,000| 2,000|
+| det_r50_vd_db_v2.0     |[config](../configs/det_r50_vd_db_v2.0/train_infer_python.txt) | 29.955 | 29.092 /    29.31 / 28.844 |0.015899011 | 51.097 |50.367 /  50.879 /    50.227| 0.012814717 | 10,000| 2,000|
+| det_r50_vd_east_v2.0   |[config](../configs/det_r50_vd_east_v2.0/train_infer_python.txt) | 42.485 | 42.624 /  42.663 /    42.561 |0.00239083 | 67.61 |67.825/     68.299/     68.51| 0.00999854 | 10,000| 2,000|
+| det_r50_vd_pse_v2.0    |[config](../configs/det_r50_vd_pse_v2.0/train_infer_python.txt) | 16.455 | 16.517 / 16.555 /  16.353 |0.012201752 | 27.02 |27.288 /   27.152 /    27.408| 0.009340339 | 10,000| 2,000|
+| rec_mv3_none_bilstm_ctc_v2.0   |[config](../configs/rec_mv3_none_bilstm_ctc_v2.0/train_infer_python.txt) | 2288.358 | 2291.906 /  2293.725 /  2290.05 |0.001602197 | 2336.17 |2327.042 /  2328.093 /  2344.915| 0.007622025 | 600,000| 160,000|
+| layoutxlm_ser  |[config](../configs/layoutxlm/train_infer_python.txt) | 18.001 | 18.114 / 18.107 /    18.307 |0.010924783 | 21.982 | 21.507 / 21.116 /    21.406| 0.018180127 | 1490 | 1490|
+| PP-Structure-table     |[config](../configs/en_table_structure/train_infer_python.txt) | 14.151 | 14.077 /    14.23 / 14.25 |0.012140351 | 16.285 | 16.595 /  16.878 /    16.531 | 0.020559308 | 20,000| 5,000|
+| det_r50_dcn_fce_ctw_v2.0   |[config](../configs/det_r50_dcn_fce_ctw_v2.0/train_infer_python.txt) | 14.057 | 14.029 /  14.02 / 14.014 |0.001069214 | 18.298 |18.411 /  18.376 /    18.331| 0.004345228 | 10,000| 2,000|
+| ch_PP-OCRv3_det    |[config](../configs/ch_PP-OCRv3_det/train_infer_python.txt) | 8.622 | 8.431 / 8.423 / 8.479|0.006604552 | 14.203 |14.346  14.468  14.23| 0.016450097 | 10,000| 2,000|
+| PP-OCRv3_mobile_rec    |[config](../configs/PP-OCRv3_mobile_rec/train_infer_python.txt) | 90.239 | 90.077 /   91.513 /    91.325|0.01569176 | | |  | 160,000| 40,000|
diff --git a/test_tipc/docs/compare_cpp_right.png b/test_tipc/docs/compare_cpp_right.png
new file mode 100644
index 0000000..f9d0ba8
Binary files /dev/null and b/test_tipc/docs/compare_cpp_right.png differ
diff --git a/test_tipc/docs/compare_cpp_wrong.png b/test_tipc/docs/compare_cpp_wrong.png
new file mode 100644
index 0000000..621d446
Binary files /dev/null and b/test_tipc/docs/compare_cpp_wrong.png differ
diff --git a/test_tipc/docs/compare_right.png b/test_tipc/docs/compare_right.png
new file mode 100644
index 0000000..3d74ef1
Binary files /dev/null and b/test_tipc/docs/compare_right.png differ
diff --git a/test_tipc/docs/compare_wrong.png b/test_tipc/docs/compare_wrong.png
new file mode 100644
index 0000000..26ad576
Binary files /dev/null and b/test_tipc/docs/compare_wrong.png differ
diff --git a/test_tipc/docs/guide.png b/test_tipc/docs/guide.png
new file mode 100644
index 0000000..319ac81
Binary files /dev/null and b/test_tipc/docs/guide.png differ
diff --git a/test_tipc/docs/install.md b/test_tipc/docs/install.md
new file mode 100644
index 0000000..d67637a
--- /dev/null
+++ b/test_tipc/docs/install.md
@@ -0,0 +1,116 @@
+## 1. 环境准备
+
+本教程适用于test_tipc目录下基础功能测试的运行环境搭建。
+
+推荐环境：
+- CUDA 10.1/10.2
+- CUDNN 7.6/cudnn8.1
+- TensorRT 6.1.0.5 / 7.1 / 7.2
+
+环境配置可以选择docker镜像安装，或者在本地环境Python搭建环境。推荐使用docker镜像安装，避免不必要的环境配置。
+
+## 2. Docker 镜像安装
+
+推荐docker镜像安装，按照如下命令创建镜像，当前目录映射到镜像中的`/paddle`目录下
+```
+nvidia-docker run --name paddle -it -v $PWD:/paddle paddlepaddle/paddle:latest-dev-cuda10.1-cudnn7-gcc82 /bin/bash
+cd /paddle
+
+# 安装带TRT的paddle
+pip3.7 install https://paddle-wheel.bj.bcebos.com/with-trt/2.1.3/linux-gpu-cuda10.1-cudnn7-mkl-gcc8.2-trt6-avx/paddlepaddle_gpu-2.1.3.post101-cp37-cp37m-linux_x86_64.whl
+```
+
+## 3 Python 环境构建
+
+非docker环境下，环境配置比较灵活，推荐环境组合配置：
+- CUDA10.1 + CUDNN7.6 + TensorRT 6
+- CUDA10.2 + CUDNN8.1 + TensorRT 7
+- CUDA11.1 + CUDNN8.1 + TensorRT 7
+
+下面以 CUDA10.2 + CUDNN8.1 + TensorRT 7 配置为例，介绍环境配置的流程。
+
+### 3.1 安装CUDNN
+
+如果当前环境满足CUDNN版本的要求，可以跳过此步骤。
+
+以CUDNN8.1 安装安装为例，安装步骤如下，首先下载CUDNN，从[Nvidia官网](https://developer.nvidia.com/rdp/cudnn-archive)下载CUDNN8.1版本，下载符合当前系统版本的三个deb文件，分别是：
+- cuDNN Runtime Library ，如：libcudnn8_8.1.0.77-1+cuda10.2_amd64.deb
+- cuDNN Developer Library ，如：libcudnn8-dev_8.1.0.77-1+cuda10.2_amd64.deb
+- cuDNN Code Samples，如：libcudnn8-samples_8.1.0.77-1+cuda10.2_amd64.deb
+
+deb安装可以参考[官方文档](https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#installlinux-deb)，安装方式如下
+```
+# x.x.x表示下载的版本号
+# $HOME为工作目录
+sudo dpkg -i libcudnn8_x.x.x-1+cudax.x_arm64.deb
+sudo dpkg -i libcudnn8-dev_8.x.x.x-1+cudax.x_arm64.deb
+sudo dpkg -i libcudnn8-samples_8.x.x.x-1+cudax.x_arm64.deb
+
+# 验证是否正确安装
+cp -r /usr/src/cudnn_samples_v8/ $HOME
+cd  $HOME/cudnn_samples_v8/mnistCUDNN
+
+# 编译
+make clean && make
+./mnistCUDNN
+```
+如果运行mnistCUDNN完后提示运行成功，则表示安装成功。如果运行后出现freeimage相关的报错，需要按照提示安装freeimage库:
+```
+sudo apt-get install libfreeimage-dev
+sudo apt-get install libfreeimage
+```
+
+### 3.2 安装TensorRT
+
+首先，从[Nvidia官网TensorRT板块](https://developer.nvidia.com/tensorrt-getting-started)下载TensorRT，这里选择7.1.3.4版本的TensorRT，注意选择适合自己系统版本和CUDA版本的TensorRT，另外建议下载TAR package的安装包。
+
+以Ubuntu16.04+CUDA10.2为例，下载并解压后可以参考[官方文档](https://docs.nvidia.com/deeplearning/tensorrt/archives/tensorrt-713/install-guide/index.html#installing-tar)的安装步骤，按照如下步骤安装:
+```
+# 以下安装命令中 '${version}' 为下载的TensorRT版本，如7.1.3.4
+# 设置环境变量，<TensorRT-${version}/lib> 为解压后的TensorRT的lib目录
+export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:<TensorRT-${version}/lib>
+
+# 安装TensorRT
+cd TensorRT-${version}/python
+pip3.7 install tensorrt-*-cp3x-none-linux_x86_64.whl
+
+# 安装graphsurgeon
+cd TensorRT-${version}/graphsurgeon
+```
+
+
+### 3.3 安装PaddlePaddle
+
+下载[链接](https://www.paddlepaddle.org.cn/install/quick)
+
+
+## 4. 安装PaddleOCR依赖
+```
+# 安装AutoLog
+git clone https://github.com/LDOUBLEV/AutoLog
+cd AutoLog
+pip3.7 install -r requirements.txt
+python3.7 setup.py bdist_wheel
+pip3.7 install ./dist/auto_log-1.0.0-py3-none-any.whl
+
+# 下载OCR代码
+cd ../
+git clone https://github.com/PaddlePaddle/PaddleOCR
+
+```
+
+安装PaddleOCR依赖：
+```
+cd PaddleOCR
+pip3.7 install -r requirements.txt
+```
+
+## FAQ :
+Q. You are using Paddle compiled with TensorRT, but TensorRT dynamic library is not found. Ignore this if TensorRT is not needed.
+
+A. 问题一般是当前安装paddle版本带TRT，但是本地环境找不到TensorRT的预测库，需要下载TensorRT库，解压后设置环境变量LD_LIBRARY_PATH;
+如：
+```
+export LD_LIBRARY_PATH=/usr/local/python3.7.0/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/paddle/package/TensorRT-6.0.1.5/lib
+```
+或者问题是下载的TensorRT版本和当前paddle中编译的TRT版本不匹配，需要下载版本相符的TensorRT重新安装。
diff --git a/test_tipc/docs/lite_auto_log.png b/test_tipc/docs/lite_auto_log.png
new file mode 100644
index 0000000..d5b6b45
Binary files /dev/null and b/test_tipc/docs/lite_auto_log.png differ
diff --git a/test_tipc/docs/lite_log.png b/test_tipc/docs/lite_log.png
new file mode 100644
index 0000000..2b3e40b
Binary files /dev/null and b/test_tipc/docs/lite_log.png differ
diff --git a/test_tipc/docs/mac_test_train_inference_python.md b/test_tipc/docs/mac_test_train_inference_python.md
new file mode 100644
index 0000000..f3ece93
--- /dev/null
+++ b/test_tipc/docs/mac_test_train_inference_python.md
@@ -0,0 +1,143 @@
+# Mac端基础训练预测功能测试
+
+Mac端基础训练预测功能测试的主程序为`test_train_inference_python.sh`，可以测试基于Python的模型CPU训练，包括裁剪、PACT在线量化、蒸馏训练，以及评估、CPU推理等基本功能。
+
+注：Mac端测试用法同linux端测试方法类似，但是无需测试需要在GPU上运行的测试。
+
+## 1. 测试结论汇总
+
+- 训练相关：
+
+| 算法名称 | 模型名称 | 单机单卡（CPU） | 单机多卡 | 多机多卡 | 模型压缩（CPU） |
+|  :----  |   :----  |    :----  |  :----   |  :----   |  :----   |
+|  DB  | ch_ppocr_mobile_v2.0_det| 正常训练 | - | - | 正常训练：FPGM裁剪、PACT量化 |
+
+
+- 预测相关：基于训练是否使用量化，可以将训练产出的模型可以分为`正常模型`和`量化模型`，这两类模型对应的预测功能汇总如下，
+
+| 模型类型 |device | batchsize | tensorrt | mkldnn | cpu多线程 |
+|  ----   |  ---- |   ----   |  :----:  |   :----:   |  :----:  |
+| 正常模型 | CPU | 1/6 | - | fp32 | 支持 |
+| 量化模型 | CPU | 1/6 | - | int8 | 支持 |
+
+
+## 2. 测试流程
+
+Mac端无GPU，环境准备只需要Python环境即可，安装PaddlePaddle等依赖参考下述文档。
+
+### 2.1 安装依赖
+- 安装PaddlePaddle >= 2.3
+- 安装PaddleOCR依赖
+    ```
+    pip install  -r ../requirements.txt
+    ```
+- 安装autolog（规范化日志输出工具）
+    ```
+    pip install https://paddleocr.bj.bcebos.com/libs/auto_log-1.2.0-py3-none-any.whl
+    ```
+- 安装PaddleSlim (可选)
+   ```
+   # 如果要测试量化、裁剪等功能，需要安装PaddleSlim
+   pip install paddleslim
+   ```
+
+
+### 2.2 功能测试
+
+先运行`prepare.sh`准备数据和模型，然后运行`test_train_inference_python.sh`进行测试，最终在```test_tipc/output```目录下生成`,model_name/lite_train_lite_infer/*.log`格式的日志文件。
+
+`test_train_inference_python.sh`包含基础链条的4种运行模式，每种模式的运行数据不同，分别用于测试速度和精度，分别是：
+
+- 模式1：lite_train_lite_infer，使用少量数据训练，用于快速验证训练到预测的走通流程，不验证精度和速度；
+```shell
+# 同linux端运行不同的是，Mac端测试使用新的配置文件mac_ppocr_det_mobile_params.txt，
+# 配置文件中默认去掉了GPU和mkldnn相关的测试链条
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_mac_cpu_normal_normal_infer_python_mac_cpu.txt 'lite_train_lite_infer'
+bash test_tipc/test_train_inference_python.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_mac_cpu_normal_normal_infer_python_mac_cpu.txt 'lite_train_lite_infer'
+```
+
+- 模式2：lite_train_whole_infer，使用少量数据训练，一定量数据预测，用于验证训练后的模型执行预测，预测速度是否合理；
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_mac_cpu_normal_normal_infer_python_mac_cpu.txt 'lite_train_whole_infer'
+bash test_tipc/test_train_inference_python.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_mac_cpu_normal_normal_infer_python_mac_cpu.txt  'lite_train_whole_infer'
+```
+
+- 模式3：whole_infer，不训练，全量数据预测，走通开源模型评估、动转静，检查inference model预测时间和精度;
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_mac_cpu_normal_normal_infer_python_mac_cpu.txt 'whole_infer'
+# 用法1:
+bash test_tipc/test_train_inference_python.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_mac_cpu_normal_normal_infer_python_mac_cpu.txt 'whole_infer'
+# 用法2: 指定GPU卡预测，第三个传入参数为GPU卡号
+bash test_tipc/test_train_inference_python.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_mac_cpu_normal_normal_infer_python_mac_cpu.txt 'whole_infer' '1'
+```
+
+- 模式4：whole_train_whole_infer，CE： 全量数据训练，全量数据预测，验证模型训练精度，预测精度，预测速度；（Mac端不建议运行此模式）
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_mac_cpu_normal_normal_infer_python_mac_cpu.txt 'whole_train_whole_infer'
+bash test_tipc/test_train_inference_python.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_mac_cpu_normal_normal_infer_python_mac_cpu.txt 'whole_train_whole_infer'
+```
+
+运行相应指令后，在`test_tipc/output`文件夹下自动会保存运行日志。如`lite_train_lite_infer`模式下，会运行训练+inference的链条，因此，在`test_tipc/output`文件夹有以下文件：
+```
+test_tipc/output/model_name/lite_train_lite_infer/
+|- results_python.log    # 运行指令状态的日志
+|- norm_train_gpus_-1_autocast_null/  # CPU上正常训练的训练日志和模型保存文件夹
+......
+|- python_infer_cpu_usemkldnn_False_threads_1_precision_fp32_batchsize_1.log  # CPU上关闭Mkldnn线程数设置为1，测试batch_size=1条件下的fp32精度预测运行日志
+......
+```
+
+其中`results_python.log`中包含了每条指令的运行状态，如果运行成功会输出：
+```
+Run successfully with command - python3.7 tools/train.py -c tests/configs/det_mv3_db.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained Global.use_gpu=False  Global.save_model_dir=./tests/output/norm_train_gpus_-1_autocast_null Global.epoch_num=1     Train.loader.batch_size_per_card=2   !
+Run successfully with command - python3.7 tools/export_model.py -c tests/configs/det_mv3_db.yml -o  Global.pretrained_model=./tests/output/norm_train_gpus_-1_autocast_null/latest Global.save_inference_dir=./tests/output/norm_train_gpus_-1_autocast_null!
+......
+```
+如果运行失败，会输出：
+```
+Run failed with command - python3.7 tools/train.py -c tests/configs/det_mv3_db.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained Global.use_gpu=Faslse  Global.save_model_dir=./tests/output/norm_train_gpus_-1_autocast_null Global.epoch_num=1     Train.loader.batch_size_per_card=2   !
+Run failed with command - python3.7 tools/export_model.py -c tests/configs/det_mv3_db.yml -o  Global.pretrained_model=./tests/output/norm_train_gpus_0_autocast_null/latest Global.save_inference_dir=./tests/output/norm_train_gpus_-1_autocast_null!
+......
+```
+可以很方便的根据`results_python.log`中的内容判定哪一个指令运行错误。
+
+### 2.3 精度测试
+
+使用compare_results.py脚本比较模型预测的结果是否符合预期，主要步骤包括：
+- 提取日志中的预测坐标；
+- 从本地文件中提取保存好的坐标结果；
+- 比较上述两个结果是否符合精度预期，误差大于设置阈值时会报错。
+
+#### 使用方式
+运行命令：
+```shell
+python test_tipc/compare_results.py --gt_file=./test_tipc/results/python_*.txt  --log_file=./test_tipc/output/python_*.log --atol=1e-3 --rtol=1e-3
+```
+
+参数介绍：
+- gt_file： 指向事先保存好的预测结果路径，支持*.txt 结尾，会自动索引*.txt格式的文件，文件默认保存在test_tipc/result/ 文件夹下
+- log_file: 指向运行test_tipc/test_train_inference_python.sh 脚本的infer模式保存的预测日志，预测日志中打印的有预测结果，比如：文本框，预测文本，类别等等，同样支持python_infer_*.log格式传入
+- atol: 设置的绝对误差
+- rtol: 设置的相对误差
+
+#### 运行结果
+
+正常运行效果如下：
+```
+Assert allclose passed! The results of python_infer_cpu_usemkldnn_False_threads_1_batchsize_1.log and ./test_tipc/results/python_ppocr_det_mobile_results_fp32.txt are consistent!
+```
+
+出现不一致结果时的运行输出：
+```
+......
+Traceback (most recent call last):
+  File "test_tipc/compare_results.py", line 140, in <module>
+    format(filename, gt_filename))
+ValueError: The results of python_infer_cpu_usemkldnn_False_threads_1_batchsize_1.log and the results of ./test_tipc/results/python_ppocr_det_mobile_results_fp32.txt are inconsistent!
+```
+
+
+## 3. 更多教程
+本文档为功能测试用，更丰富的训练预测使用教程请参考：
+[模型训练](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/training.md)
+[基于Python预测引擎推理](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/inference_ppocr.md)
diff --git a/test_tipc/docs/ssh_termux_ls.png b/test_tipc/docs/ssh_termux_ls.png
new file mode 100644
index 0000000..2df7802
Binary files /dev/null and b/test_tipc/docs/ssh_termux_ls.png differ
diff --git a/test_tipc/docs/termux.jpg b/test_tipc/docs/termux.jpg
new file mode 100644
index 0000000..cb87c4c
Binary files /dev/null and b/test_tipc/docs/termux.jpg differ
diff --git a/test_tipc/docs/termux_for_android.md b/test_tipc/docs/termux_for_android.md
new file mode 100644
index 0000000..3ae9d18
--- /dev/null
+++ b/test_tipc/docs/termux_for_android.md
@@ -0,0 +1,127 @@
+# 安卓手机通过Termux连接电脑
+
+由于通过adb方式连接手机后，很多linux命令无法运行，自动化测试受阻，所以此处特此介绍另外一种通过Termux的连接方式，不仅可以运行大部分linux命令，方便开发者在手机上在线调试，甚至还可以多实现台机器同时连接手机。Termux不是真实的Linux环境，但是Termux可以安装真实的Linux，而且不会损失性能，与此同时，Termux不需要root。在配置Termux之前，请确保电脑已经安装adb工具，安装方式请参考[Lite端部署](https://github.com/PaddlePaddle/PaddleOCR/blob/develop/deploy/lite/readme.md) 。在运行以下命令后确保可以显示安卓设备信息。
+
+```
+adb devices
+```
+连接成功信息提示：
+```
+List of devices attached
+744be294    device
+```
+
+## 1.安卓手机安装termux app
+
+### 1.1 下载termux apk文件
+
+由于目前该app目前各大商城暂无，所以可以直接下载如下apk文件。
+
+打开电脑终端，执行以下命令：
+
+```
+wget http://10.12.121.133:8911/cuicheng01/fullchain/termux-v1.0.3.apk
+```
+
+### 1.2 安装termux到手机上
+
+在手机端的开发者模式下，允许USB调试，允许USB安装。在电脑终端，执行如下命令，将termux app安装到手机上：
+
+```
+adb install termux-v1.0.3.apk
+```
+
+此处需要手机端确认安装，点击确认。
+
+### 1.3 验证是否安装成功
+
+打开手机，检验termux是否安装成功，如果没有，重新执行1.2，如果有相应的app，点击进入，会有如下显示。
+
+<img src="termux.jpg" width="300" height = "300">
+
+接下来的配置环境需要在手机上此终端运行相关命令。
+
+## 2.手机端配置termux
+
+首先将手机联网，最好可以连接外网，部分的配置需要外网。打开Termux终端，执行以下命令安装基础件`proot`，并使用`termux-chroot`命令可以模拟 root 环境与标准的 Linux 目录结构。
+
+```
+pkg i -y proot
+termux-chroot
+```
+
+Termux 默认只能访问自身内部的数据，如果要访问手机中其它的数据，输入下面的命令后，手机弹出对请求权限的窗口，允许即可（方便对部分运行出的结果在手机端可视化）。
+
+```
+termux-setup-storage
+```
+
+### 2.1 配置SSH
+
+作为 Linux 终端或者服务器，必须有SSH。不管你是 SSH 连接到 Termux还是使用Termux去连其它主机，都需要先安装openssh。如果安装失败，请重复执行命令。
+
+```
+pkg i -y openssh
+```
+
+启动 SSH 服务端，默认端口号为8022
+
+```
+sshd
+```
+
+
+### 2.2 电脑通过SSH方式连接手机
+
+1.保证手机和电脑处于同一局域网下
+手机端分别输入以下命令获得ip地址和当前用户：
+
+```
+# 获取ip地址
+ifconfig
+
+# 获取当前用户
+whoami
+```
+
+如获取到的ip地址和当前用户分别是`172.24.162.117`和`u0_a374`。
+
+2.电脑端通过SSH连接手机
+
+```
+#默认端口号为8022
+ssh u0_a374@172.24.162.117 -p 8022
+```
+
+3.运行ls命令后，会有如下显示：
+
+```
+ls
+```
+
+<img src="ssh_termux_ls.png" width="800">
+
+
+### 2.3 通过scp传输数据
+
+1.在当前目录上新建test目录
+
+```
+mkdir test
+```
+
+2.测试scp功能
+
+将电脑中的某个文件拷贝到手机上：
+```
+scp -P 8022 test.txt u0_a374@172.24.162.117:/home/storage/test
+```
+
+3.手机端查看
+
+打开手机终端，在`/home/storage/test`下查看是否存在`test.txt`
+
+
+## 3. 更多教程
+
+本教程可以完成Termux基本配置，更多关于Termux的用法，请参考：[Termux高级终端安装使用配置教程](https://www.sqlsec.com/2018/05/termux.html)。
diff --git a/test_tipc/docs/test.png b/test_tipc/docs/test.png
new file mode 100644
index 0000000..f99f23d
Binary files /dev/null and b/test_tipc/docs/test.png differ
diff --git a/test_tipc/docs/test_inference_cpp.md b/test_tipc/docs/test_inference_cpp.md
new file mode 100644
index 0000000..5e9e852
--- /dev/null
+++ b/test_tipc/docs/test_inference_cpp.md
@@ -0,0 +1,86 @@
+# C++预测功能测试
+
+C++预测功能测试的主程序为`test_inference_cpp.sh`，可以测试基于C++预测库的模型推理功能。
+
+## 1. 测试结论汇总
+
+基于训练是否使用量化，进行本测试的模型可以分为`正常模型`和`量化模型`，这两类模型对应的C++预测功能汇总如下：
+
+| 模型类型 |device | batchsize | tensorrt | mkldnn | cpu多线程 |
+|  ----   |  ---- |   ----   |  :----:  |   :----:   |  :----:  |
+| 正常模型 | GPU | 1/6 | fp32/fp16 | - | - |
+| 正常模型 | CPU | 1/6 | - | fp32 | 支持 |
+| 量化模型 | GPU | 1/6 | int8 | - | - |
+| 量化模型 | CPU | 1/6 | - | int8 | 支持 |
+
+## 2. 测试流程
+运行环境配置请参考[文档](./install.md)的内容配置TIPC的运行环境。
+
+### 2.1 功能测试
+先运行`prepare.sh`准备数据和模型，然后运行`test_inference_cpp.sh`进行测试，最终在```test_tipc/output/{model_name}/cpp_infer```目录下生成`cpp_infer_*.log`后缀的日志文件。
+
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_PP-OCRv2_rec/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt "cpp_infer"
+
+# 用法1:
+bash test_tipc/test_inference_cpp.sh test_tipc/configs/ch_PP-OCRv2_rec/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt
+# 用法2: 指定GPU卡预测，第三个传入参数为GPU卡号
+bash test_tipc/test_inference_cpp.sh test_tipc/configs/ch_PP-OCRv2_rec/model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt '1'
+```
+
+运行预测指令后，在`test_tipc/output`文件夹下自动会保存运行日志，包括以下文件：
+
+```shell
+test_tipc/output/
+|- results_cpp.log    # 运行指令状态的日志
+|- cpp_infer_cpu_usemkldnn_False_threads_6_precision_fp32_batchsize_6.log  # CPU上不开启Mkldnn，线程数设置为6，测试batch_size=6条件下的预测运行日志
+|- cpp_infer_gpu_usetrt_False_precision_fp32_batchsize_6.log # GPU上不开启TensorRT，测试batch_size=6的fp32精度预测日志
+......
+```
+其中results_cpp.log中包含了每条指令的运行状态，如果运行成功会输出：
+
+```
+[33m Run successfully with command - ch_PP-OCRv2_rec - ./deploy/cpp_infer/build/ppocr --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_img_h=32 --use_gpu=True --use_tensorrt=False --precision=fp32 --rec_model_dir=./inference/ch_PP-OCRv2_rec_infer/ --rec_batch_num=6 --image_dir=./inference/rec_inference/ --benchmark=True --det=False --rec=True --cls=False --use_angle_cls=False   > ./test_tipc/output/ch_PP-OCRv2_rec/cpp_infer/cpp_infer_gpu_usetrt_False_precision_fp32_batchsize_6.log 2>&1 !  
+ Run successfully with command - ch_PP-OCRv2_rec - ./deploy/cpp_infer/build/ppocr --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_img_h=32 --use_gpu=False --enable_mkldnn=False --cpu_threads=6 --rec_model_dir=./inference/ch_PP-OCRv2_rec_infer/ --rec_batch_num=6 --image_dir=./inference/rec_inference/ --benchmark=True --det=False --rec=True --cls=False --use_angle_cls=False   > ./test_tipc/output/ch_PP-OCRv2_rec/cpp_infer/cpp_infer_cpu_usemkldnn_False_threads_6_precision_fp32_batchsize_6.log 2>&1 !  
+......
+```
+如果运行失败，会输出：
+```
+Run failed with command - ch_PP-OCRv2_rec - ./deploy/cpp_infer/build/ppocr --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_img_h=32 --use_gpu=True --use_tensorrt=False --precision=fp32 --rec_model_dir=./inference/ch_PP-OCRv2_rec_infer/ --rec_batch_num=6 --image_dir=./inference/rec_inference/ --benchmark=True --det=False --rec=True --cls=False --use_angle_cls=False   > ./test_tipc/output/ch_PP-OCRv2_rec/cpp_infer/cpp_infer_gpu_usetrt_False_precision_fp32_batchsize_6.log 2>&1 !
+Run failed with command - ch_PP-OCRv2_rec - ./deploy/cpp_infer/build/ppocr --rec_char_dict_path=./ppocr/utils/ppocr_keys_v1.txt --rec_img_h=32 --use_gpu=False --enable_mkldnn=False --cpu_threads=6 --rec_model_dir=./inference/ch_PP-OCRv2_rec_infer/ --rec_batch_num=6 --image_dir=./inference/rec_inference/ --benchmark=True --det=False --rec=True --cls=False --use_angle_cls=False   > ./test_tipc/output/ch_PP-OCRv2_rec/cpp_infer/cpp_infer_cpu_usemkldnn_False_threads_6_precision_fp32_batchsize_6.log 2>&1 !
+......
+```
+可以很方便的根据results_cpp.log中的内容判定哪一个指令运行错误。
+
+
+### 2.2 精度测试
+
+使用compare_results.py脚本比较模型预测的结果是否符合预期，主要步骤包括：
+- 提取日志中的预测坐标；
+- 从本地文件中提取保存好的坐标结果；
+- 比较上述两个结果是否符合精度预期，误差大于设置阈值时会报错。
+
+#### 使用方式
+运行命令：
+```shell
+python3.7 test_tipc/compare_results.py --gt_file=./test_tipc/results/cpp_*.txt  --log_file=./test_tipc/output/cpp_*.log --atol=1e-3 --rtol=1e-3
+```
+
+参数介绍：
+- gt_file： 指向事先保存好的预测结果路径，支持*.txt 结尾，会自动索引*.txt格式的文件，文件默认保存在test_tipc/result/ 文件夹下
+- log_file: 指向运行test_tipc/test_inference_cpp.sh 脚本的infer模式保存的预测日志，预测日志中打印的有预测结果，比如：文本框，预测文本，类别等等，同样支持cpp_infer_*.log格式传入
+- atol: 设置的绝对误差
+- rtol: 设置的相对误差
+
+#### 运行结果
+
+正常运行效果如下图：
+<img src="compare_cpp_right.png" width="1000">
+
+出现不一致结果时的运行输出：
+<img src="compare_cpp_wrong.png" width="1000">
+
+
+## 3. 更多教程
+
+本文档为功能测试用，更详细的c++预测使用教程请参考：[服务器端C++预测](https://github.com/PaddlePaddle/PaddleOCR/tree/dygraph/deploy/cpp_infer)
diff --git a/test_tipc/docs/test_lite_arm_cpp.md b/test_tipc/docs/test_lite_arm_cpp.md
new file mode 100644
index 0000000..b84e069
--- /dev/null
+++ b/test_tipc/docs/test_lite_arm_cpp.md
@@ -0,0 +1,95 @@
+# Lite\_arm\_cpp预测功能测试
+
+Lite\_arm\_cpp预测功能测试的主程序为`test_lite_arm_cpp.sh`，可以在ARM上基于Lite预测库测试模型的C++推理功能。
+
+## 1. 测试结论汇总
+
+目前Lite端的样本间支持以方式的组合：
+
+**字段说明：**
+- 模型类型：包括正常模型（FP32）和量化模型（INT8）
+- batch-size：包括1和4
+- threads：包括1和4
+- predictor数量：包括单predictor预测和多predictor预测
+- 预测库来源：包括下载方式和编译方式
+- 测试硬件：ARM\_CPU/ARM\_GPU_OPENCL
+
+| 模型类型 | batch-size | threads | predictor数量 | 预测库来源 | 测试硬件 |
+|  :----:   |  :----:  | :----:  |  :----:  |  :----:  |  :----:  |
+| 正常模型/量化模型 | 1 | 1/4 |  单/多 | 下载方式/编译方式 | ARM\_CPU/ARM\_GPU_OPENCL |
+
+
+## 2. 测试流程
+运行环境配置请参考[文档](./install.md)的内容配置TIPC的运行环境。
+
+### 2.1 功能测试
+
+先运行`prepare_lite_cpp.sh`，运行后会在当前路径下生成`test_lite.tar`，其中包含了测试数据、测试模型和用于预测的可执行文件。将`test_lite.tar`上传到被测试的手机上，在手机的终端解压该文件，进入`test_lite`目录中，然后运行`test_lite_arm_cpp.sh`进行测试，最终在`test_lite/output`目录下生成`lite_*.log`后缀的日志文件。
+
+#### 2.1.1 基于ARM\_CPU测试
+
+```shell
+
+# 数据、模型、Paddle-Lite预测库准备
+#预测库为下载方式
+bash test_tipc/prepare_lite_cpp.sh ./test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt download
+#预测库为编译方式
+bash test_tipc/prepare_lite_cpp.sh ./test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt compile
+
+# 手机端测试:
+bash test_lite_arm_cpp.sh model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt
+
+```
+
+#### 2.1.2 基于ARM\_GPU\_OPENCL测试
+
+```shell
+
+# 数据、模型、Paddle-Lite预测库准备
+#预测库下载方式
+bash test_tipc/prepare_lite_cpp.sh ./test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt download
+#预测库编译方式
+bash test_tipc/prepare_lite_cpp.sh ./test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt compile
+
+# 手机端测试:
+bash test_lite_arm_cpp.sh model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt
+
+```
+
+
+**注意**：
+
+由于运行该项目需要bash等命令，传统的adb方式不能很好的安装。所以此处推荐通在手机上开启虚拟终端的方式连接电脑，连接方式可以参考[安卓手机termux连接电脑](./termux_for_android.md)。
+
+### 2.2 运行结果
+
+各测试的运行情况会打印在 `./output/` 中：
+运行成功时会输出：
+
+```
+Run successfully with command - ./ocr_db_crnn det ch_PP-OCRv2_det_infer_opt.nb ARM_CPU FP32 1 1  ./test_data/icdar2015_lite/text_localization/ch4_test_images/ ./config.txt True > ./output/lite_ch_PP-OCRv2_det_infer_opt.nb_runtime_device_ARM_CPU_precision_FP32_batchsize_1_threads_1.log 2>&1!
+Run successfully with command xxx
+...
+```
+
+运行失败时会输出：
+
+```
+Run failed with command - ./ocr_db_crnn det ch_PP-OCRv2_det_infer_opt.nb ARM_CPU FP32 1 1  ./test_data/icdar2015_lite/text_localization/ch4_test_images/ ./config.txt True > ./output/lite_ch_PP-OCRv2_det_infer_opt.nb_runtime_device_ARM_CPU_precision_FP32_batchsize_1_threads_1.log 2>&1!
+Run failed with command xxx
+...
+```
+
+在./output/文件夹下，会存在如下日志，每一个日志都是不同配置下的log结果：
+
+<img src="lite_log.png" width="1000">
+
+在每一个log中，都会调用autolog打印如下信息：
+
+<img src="lite_auto_log.png" width="1000">
+
+
+
+## 3. 更多教程
+
+本文档为功能测试用，更详细的Lite端预测使用教程请参考：[Lite端部署](https://github.com/PaddlePaddle/PaddleOCR/blob/develop/deploy/lite/readme.md)。
diff --git a/test_tipc/docs/test_paddle2onnx.md b/test_tipc/docs/test_paddle2onnx.md
new file mode 100644
index 0000000..9393388
--- /dev/null
+++ b/test_tipc/docs/test_paddle2onnx.md
@@ -0,0 +1,48 @@
+# Paddle2onnx预测功能测试
+
+PaddleServing预测功能测试的主程序为`test_paddle2onnx.sh`，可以测试Paddle2ONNX的模型转化功能，并验证正确性。
+
+## 1. 测试结论汇总
+
+基于训练是否使用量化，进行本测试的模型可以分为`正常模型`和`量化模型`，这两类模型对应的Paddle2ONNX预测功能汇总如下：
+
+| 模型类型 |device |
+|  ----   |  ---- |
+| 正常模型 | GPU |
+| 正常模型 | CPU |
+| 量化模型 | GPU |
+| 量化模型 | CPU |
+
+## 2. 测试流程
+### 2.1 功能测试
+先运行`prepare.sh`准备数据和模型，然后运行`test_paddle2onnx.sh`进行测试，最终在```test_tipc/output/{model_name}/paddle2onnx```目录下生成`paddle2onnx_infer_*.log`后缀的日志文件。
+
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt "paddle2onnx_infer"
+
+# 用法:
+bash test_tipc/test_paddle2onnx.sh ./test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_paddle2onnx_python_linux_cpu.txt
+```
+
+#### 运行结果
+
+各测试的运行情况会打印在 `test_tipc/output/{model_name}/paddle2onnx/results_paddle2onnx.log` 中：
+运行成功时会输出：
+
+```
+Run successfully with command - ch_PP-OCRv2_det -  paddle2onnx --model_dir=./inference/ch_PP-OCRv2_det_infer/ --model_filename=inference.pdmodel --params_filename=inference.pdiparams --save_file=./inference/det_v2_onnx/model.onnx --opset_version=10 --enable_onnx_checker=True!
+Run successfully with command - ch_PP-OCRv2_det - python3.7 tools/infer/predict_det.py --use_gpu=True --image_dir=./inference/ch_det_data_50/all-sum-510/ --det_model_dir=./inference/det_v2_onnx/model.onnx --use_onnx=True > ./test_tipc/output/ch_PP-OCRv2_det/paddle2onnx/paddle2onnx_infer_gpu.log 2>&1 !
+Run successfully with command - ch_PP-OCRv2_det - python3.7 tools/infer/predict_det.py --use_gpu=False --image_dir=./inference/ch_det_data_50/all-sum-510/ --det_model_dir=./inference/det_v2_onnx/model.onnx --use_onnx=True > ./test_tipc/output/ch_PP-OCRv2_det/paddle2onnx/paddle2onnx_infer_cpu.log 2>&1 !
+```
+
+运行失败时会输出：
+
+```
+Run failed with command - ch_PP-OCRv2_det -  paddle2onnx --model_dir=./inference/ch_PP-OCRv2_det_infer/ --model_filename=inference.pdmodel --params_filename=inference.pdiparams --save_file=./inference/det_v2_onnx/model.onnx --opset_version=10 --enable_onnx_checker=True!
+...
+```
+
+
+## 3. 更多教程
+
+本文档为功能测试用，更详细的Paddle2onnx预测使用教程请参考：[Paddle2ONNX](https://github.com/PaddlePaddle/Paddle2ONNX)
diff --git a/test_tipc/docs/test_ptq_inference_python.md b/test_tipc/docs/test_ptq_inference_python.md
new file mode 100644
index 0000000..b118b66
--- /dev/null
+++ b/test_tipc/docs/test_ptq_inference_python.md
@@ -0,0 +1,51 @@
+# Linux GPU/CPU KL离线量化训练推理测试
+
+Linux GPU/CPU KL离线量化训练推理测试的主程序为`test_ptq_inference_python.sh`，可以测试基于Python的模型训练、评估、推理等基本功能。
+
+## 1. 测试结论汇总
+- 训练相关：
+
+| 算法名称 | 模型名称 | 单机单卡 |
+|  :----: |   :----:  |    :----:  |
+|    | model_name | KL离线量化训练 |
+
+- 推理相关：
+
+| 算法名称 | 模型名称 | device_CPU | device_GPU | batchsize |
+|  :----:   |  :----: |   :----:   |  :----:  |   :----:   |
+|     |  model_name |  支持 | 支持 | 1 |
+
+## 2. 测试流程
+
+### 2.1 准备数据和模型
+
+先运行`prepare.sh`准备数据和模型，然后运行`test_ptq_inference_python.sh`进行测试，最终在```test_tipc/output/{model_name}/whole_infer```目录下生成`python_infer_*.log`后缀的日志文件。
+
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_PP-OCRv2_det/train_ptq_infer_python.txt "whole_infer"
+
+# 用法:
+bash test_tipc/test_ptq_inference_python.sh ./test_tipc/configs/ch_PP-OCRv2_det/train_ptq_infer_python.txt "whole_infer"
+```
+
+#### 运行结果
+
+各测试的运行情况会打印在 `test_tipc/output/{model_name}/paddle2onnx/results_paddle2onnx.log` 中：
+运行成功时会输出：
+
+```
+Run successfully with command - ch_PP-OCRv2_det_KL - python3.7 deploy/slim/quantization/quant_kl.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o Global.pretrained_model=./inference/ch_PP-OCRv2_det_infer/ Global.save_inference_dir=./inference/ch_PP-OCRv2_det_infer/_klquant > ./test_tipc/output/ch_PP-OCRv2_det_KL/whole_infer/whole_infer_export_0.log 2>&1 !
+Run successfully with command - ch_PP-OCRv2_det_KL - python3.7 tools/infer/predict_det.py --use_gpu=False --enable_mkldnn=False --cpu_threads=6 --det_model_dir=./inference/ch_PP-OCRv2_det_infer/_klquant --rec_batch_num=1   --image_dir=./inference/ch_det_data_50/all-sum-510/   --precision=int8   > ./test_tipc/output/ch_PP-OCRv2_det_KL/whole_infer/python_infer_cpu_usemkldnn_False_threads_6_precision_int8_batchsize_1.log 2>&1 !
+Run successfully with command - ch_PP-OCRv2_det_KL - python3.7 tools/infer/predict_det.py --use_gpu=True --use_tensorrt=False --precision=int8 --det_model_dir=./inference/ch_PP-OCRv2_det_infer/_klquant --rec_batch_num=1 --image_dir=./inference/ch_det_data_50/all-sum-510/       > ./test_tipc/output/ch_PP-OCRv2_det_KL/whole_infer/python_infer_gpu_usetrt_False_precision_int8_batchsize_1.log 2>&1 !
+```
+
+运行失败时会输出：
+
+```
+Run failed with command - ch_PP-OCRv2_det_KL - python3.7 deploy/slim/quantization/quant_kl.py -c configs/det/ch_PP-OCRv2/ch_PP-OCRv2_det_cml.yml -o Global.pretrained_model=./inference/ch_PP-OCRv2_det_infer/ Global.save_inference_dir=./inference/ch_PP-OCRv2_det_infer/_klquant > ./test_tipc/output/ch_PP-OCRv2_det_KL/whole_infer/whole_infer_export_0.log 2>&1 !
+...
+```
+
+## 3. 更多教程
+
+本文档为功能测试用，更详细的量化使用教程请参考：[量化](../../deploy/slim/quantization/README.md)
diff --git a/test_tipc/docs/test_serving.md b/test_tipc/docs/test_serving.md
new file mode 100644
index 0000000..1be7fa2
--- /dev/null
+++ b/test_tipc/docs/test_serving.md
@@ -0,0 +1,63 @@
+# PaddleServing预测功能测试
+
+PaddleServing预测功能测试的主程序为`test_serving_infer_python.sh`和`test_serving_infer_cpp.sh`，可以测试基于PaddleServing的部署功能。
+
+## 1. 测试结论汇总
+
+基于训练是否使用量化，进行本测试的模型可以分为`正常模型`和`量化模型`，这两类模型对应的Serving预测功能汇总如下：
+
+| 模型类型 |device | batchsize | tensorrt | mkldnn | cpu多线程 |
+|  ----   |  ---- |   ----   |  :----:  |   :----:   |  :----:  |
+| 正常模型 | GPU | 1/6 | fp32/fp16 | - | - |
+| 正常模型 | CPU | 1/6 | - | fp32 | 支持 |
+| 量化模型 | GPU | 1/6 | int8 | - | - |
+| 量化模型 | CPU | 1/6 | - | int8 | 支持 |
+
+## 2. 测试流程
+运行环境配置请参考[文档](./install.md)的内容配置TIPC的运行环境。
+
+### 2.1 功能测试
+**python serving**
+先运行`prepare.sh`准备数据和模型，然后运行`test_serving_infer_python.sh`进行测试，最终在```test_tipc/output/{model_name}/serving_infer/python```目录下生成`python_*.log`后缀的日志文件。
+
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt "serving_infer"
+
+# 用法:
+bash test_tipc/test_serving_infer_python.sh ./test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_serving_python_linux_gpu_cpu.txt "serving_infer"
+```
+**cpp serving**
+先运行`prepare.sh`准备数据和模型，然后运行`test_serving_infer_cpp.sh`进行测试，最终在```test_tipc/output/{model_name}/serving_infer/cpp```目录下生成`cpp_*.log`后缀的日志文件。
+
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt "serving_infer"
+
+# 用法:
+bash test_tipc/test_serving_infer_cpp.sh ./test_tipc/configs/ch_PP-OCRv2/model_linux_gpu_normal_normal_serving_cpp_linux_gpu_cpu.txt "serving_infer"
+```
+
+#### 运行结果
+
+各测试的运行情况会打印在 `test_tipc/output/{model_name}/serving_infer/python(cpp)/results_python(cpp)_serving.log` 中：
+运行成功时会输出：
+
+```
+Run successfully with command - ch_PP-OCRv2_rec - nohup python3.7 web_service_rec.py --config=config.yml --opt op.rec.concurrency="1" op.det.local_service_conf.devices= op.det.local_service_conf.use_mkldnn=False op.det.local_service_conf.thread_num=6 op.rec.local_service_conf.model_config=ppocr_rec_v2_serving > ./test_tipc/output/ch_PP-OCRv2_rec/serving_infer/python/python_server_cpu_usemkldnn_False_threads_6.log 2>&1 &!
+Run successfully with command - ch_PP-OCRv2_rec - python3.7 pipeline_http_client.py --det=False --image_dir=../../inference/rec_inference > ./test_tipc/output/ch_PP-OCRv2_rec/serving_infer/python/python_client_cpu_pipeline_http_usemkldnn_False_threads_6_batchsize_1.log 2>&1 !
+...
+```
+
+运行失败时会输出：
+
+```
+Run failed with command - ch_PP-OCRv2_rec - nohup python3.7 web_service_rec.py --config=config.yml --opt op.rec.concurrency="1" op.det.local_service_conf.devices= op.det.local_service_conf.use_mkldnn=False op.det.local_service_conf.thread_num=6 op.rec.local_service_conf.model_config=ppocr_rec_v2_serving > ./test_tipc/output/ch_PP-OCRv2_rec/serving_infer/python/python_server_cpu_usemkldnn_False_threads_6.log 2>&1 &!
+Run failed with command - ch_PP-OCRv2_rec - python3.7 pipeline_http_client.py --det=False --image_dir=../../inference/rec_inference > ./test_tipc/output/ch_PP-OCRv2_rec/serving_infer/python/python_client_cpu_pipeline_http_usemkldnn_False_threads_6_batchsize_1.log 2>&1 !
+...
+```
+
+详细的预测结果会存在 test_tipc/output/{model_name}/serving_infer/python(cpp)/ 文件夹下
+
+
+## 3. 更多教程
+
+本文档为功能测试用，更详细的Serving预测使用教程请参考：[PPOCR 服务化部署](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/deploy/pdserving/README_CN.md)
diff --git a/test_tipc/docs/test_train_fleet_inference_python.md b/test_tipc/docs/test_train_fleet_inference_python.md
new file mode 100644
index 0000000..af1e7e8
--- /dev/null
+++ b/test_tipc/docs/test_train_fleet_inference_python.md
@@ -0,0 +1,107 @@
+# Linux GPU/CPU 多机多卡训练推理测试
+
+Linux GPU/CPU 多机多卡训练推理测试的主程序为`test_train_inference_python.sh`，可以测试基于Python的模型训练、评估、推理等基本功能。
+
+## 1. 测试结论汇总
+
+- 训练相关：
+
+| 算法名称 | 模型名称 | 多机多卡 |
+|  :----: |   :----:  |    :----:  |
+|  PP-OCRv3      | PP-OCRv3_mobile_rec     | 分布式训练 |
+
+
+- 推理相关：
+
+| 算法名称 | 模型名称 | device_CPU | device_GPU | batchsize |
+|  :----:   |  :----: |   :----:   |  :----:  |   :----:   |
+|  PP-OCRv3   |  PP-OCRv3_mobile_rec |  支持 | - | 1/6 |
+
+
+## 2. 测试流程
+
+运行环境配置请参考[文档](./install.md)的内容配置TIPC的运行环境。
+
+### 2.1 功能测试
+
+#### 2.1.1 修改配置文件
+
+首先，修改配置文件中的`ip`设置:  假设两台机器的`ip`地址分别为`192.168.0.1`和`192.168.0.2`，则对应的配置文件`gpu_list`字段需要修改为`gpu_list:192.168.0.1,192.168.0.2;0,1`； `ip`地址查看命令为`ifconfig`。
+
+
+#### 2.1.2 准备数据
+
+运行`prepare.sh`准备数据和模型，以配置文件`test_tipc/configs/PP-OCRv3_mobile_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt`为例，数据准备命令如下所示。
+
+```shell
+bash test_tipc/prepare.sh test_tipc/configs/PP-OCRv3_mobile_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt lite_train_lite_infer
+```
+
+**注意：** 由于是多机训练，这里需要在所有的节点上均运行启动上述命令，准备数据。
+
+#### 2.1.3 修改起始端口并开始测试
+
+在多机的节点上使用下面的命令设置分布式的起始端口（否则后面运行的时候会由于无法找到运行端口而hang住），一般建议设置在`10000~20000`之间。
+
+```shell
+export FLAGS_START_PORT=17000
+```
+
+以配置文件`test_tipc/configs/PP-OCRv3_mobile_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt`为例，测试方法如下所示。
+
+```shell
+bash test_tipc/test_train_inference_python.sh  test_tipc/configs/PP-OCRv3_mobile_rec/train_linux_gpu_fleet_normal_infer_python_linux_gpu_cpu.txt lite_train_lite_infer
+```
+
+**注意：** 由于是多机训练，这里需要在所有的节点上均运行启动上述命令进行测试。
+
+
+#### 2.1.4 输出结果
+
+输出结果如下，表示命令运行成功。
+
+```bash
+ Run successfully with command - PP-OCRv3_mobile_rec - python3.7 -m paddle.distributed.launch --ips=192.168.0.1,192.168.0.2 --gpus=0,1 tools/train.py -c test_tipc/configs/PP-OCRv3_mobile_rec/PP-OCRv3_mobile_rec_distillation.yml -o  Global.use_gpu=True Global.save_model_dir=./test_tipc/output/PP-OCRv3_mobile_rec/lite_train_lite_infer/norm_train_gpus_0,1_autocast_fp32_nodes_2   Global.epoch_num=3 Global.auto_cast=fp32 Train.loader.batch_size_per_card=16    !
+ ......
+  Run successfully with command - PP-OCRv3_mobile_rec - python3.7 tools/infer/predict_rec.py --rec_image_shape="3,48,320" --use_gpu=False --enable_mkldnn=False --cpu_threads=6 --rec_model_dir=./test_tipc/output/PP-OCRv3_mobile_rec/lite_train_lite_infer/norm_train_gpus_0,1_autocast_fp32_nodes_2/Student --rec_batch_num=1   --image_dir=./inference/rec_inference --benchmark=True --precision=fp32   > ./test_tipc/output/PP-OCRv3_mobile_rec/lite_train_lite_infer/python_infer_cpu_usemkldnn_False_threads_6_precision_fp32_batchsize_1.log 2>&1 !
+```
+
+在开启benchmark参数时，可以得到测试的详细数据，包含运行环境信息（系统版本、CUDA版本、CUDNN版本、驱动版本），Paddle版本信息，参数设置信息（运行设备、线程数、是否开启内存优化等），模型信息（模型名称、精度），数据信息（batchsize、是否为动态shape等），性能信息（CPU,GPU的占用、运行耗时、预处理耗时、推理耗时、后处理耗时），内容如下所示：
+
+```
+[2022/06/02 22:53:35] ppocr INFO:
+
+[2022/06/02 22:53:35] ppocr INFO: ---------------------- Env info ----------------------
+[2022/06/02 22:53:35] ppocr INFO:  OS_version: Ubuntu 16.04
+[2022/06/02 22:53:35] ppocr INFO:  CUDA_version: 10.1.243
+[2022/06/02 22:53:35] ppocr INFO:  CUDNN_version: 7.6.5
+[2022/06/02 22:53:35] ppocr INFO:  drivier_version: 460.32.03
+[2022/06/02 22:53:35] ppocr INFO: ---------------------- Paddle info ----------------------
+[2022/06/02 22:53:35] ppocr INFO:  paddle_version: 2.3.0-rc0
+[2022/06/02 22:53:35] ppocr INFO:  paddle_commit: 5d4980c052583fec022812d9c29460aff7cdc18b
+[2022/06/02 22:53:35] ppocr INFO:  log_api_version: 1.0
+[2022/06/02 22:53:35] ppocr INFO: ----------------------- Conf info -----------------------
+[2022/06/02 22:53:35] ppocr INFO:  runtime_device: cpu
+[2022/06/02 22:53:35] ppocr INFO:  ir_optim: True
+[2022/06/02 22:53:35] ppocr INFO:  enable_memory_optim: True
+[2022/06/02 22:53:35] ppocr INFO:  enable_tensorrt: False
+[2022/06/02 22:53:35] ppocr INFO:  enable_mkldnn: False
+[2022/06/02 22:53:35] ppocr INFO:  cpu_math_library_num_threads: 6
+[2022/06/02 22:53:35] ppocr INFO: ----------------------- Model info ----------------------
+[2022/06/02 22:53:35] ppocr INFO:  model_name: rec
+[2022/06/02 22:53:35] ppocr INFO:  precision: fp32
+[2022/06/02 22:53:35] ppocr INFO: ----------------------- Data info -----------------------
+[2022/06/02 22:53:35] ppocr INFO:  batch_size: 1
+[2022/06/02 22:53:35] ppocr INFO:  input_shape: dynamic
+[2022/06/02 22:53:35] ppocr INFO:  data_num: 6
+[2022/06/02 22:53:35] ppocr INFO: ----------------------- Perf info -----------------------
+[2022/06/02 22:53:35] ppocr INFO:  cpu_rss(MB): 288.957, gpu_rss(MB): None, gpu_util: None%
+[2022/06/02 22:53:35] ppocr INFO:  total time spent(s): 0.4824
+[2022/06/02 22:53:35] ppocr INFO:  preprocess_time(ms): 0.1136, inference_time(ms): 79.5877, postprocess_time(ms): 0.6945
+```
+
+该信息可以在运行log中查看，以上面的`PP-OCRv3_mobile_rec`为例，log位置在`./test_tipc/output/PP-OCRv3_mobile_rec/lite_train_lite_infer/results_python.log`。
+
+如果运行失败，也会在终端中输出运行失败的日志信息以及对应的运行命令。可以基于该命令，分析运行失败的原因。
+
+**注意：** 由于分布式训练时，仅在`trainer_id=0`所在的节点中保存模型，因此其他的节点中在运行模型导出与推理时会报错，为正常现象。
diff --git a/test_tipc/docs/test_train_inference_python.md b/test_tipc/docs/test_train_inference_python.md
new file mode 100644
index 0000000..4578057
--- /dev/null
+++ b/test_tipc/docs/test_train_inference_python.md
@@ -0,0 +1,165 @@
+# Linux端基础训练预测功能测试
+
+Linux端基础训练预测功能测试的主程序为`test_train_inference_python.sh`，可以测试基于Python的模型训练、评估、推理等基本功能，包括PACT在线量化。
+
+- Mac端基础训练预测功能测试参考[链接](./mac_test_train_inference_python.md)
+- Windows端基础训练预测功能测试参考[链接](./win_test_train_inference_python.md)
+
+## 1. 测试结论汇总
+
+- 训练相关：
+
+| 算法名称 | 模型名称 | 单机单卡 | 单机多卡 | 多机多卡 | 模型压缩（单机多卡） |
+|  :----  |   :----  |    :----  |  :----   |  :----   |  :----   |
+|  DB  | ch_ppocr_mobile_v2_0_det| 正常训练 <br> 混合精度 | 正常训练 <br> 混合精度 | 正常训练 <br> 混合精度 | 正常训练：FPGM裁剪、PACT量化 |
+|  DB  | ch_ppocr_server_v2_0_det| 正常训练 <br> 混合精度 | 正常训练 <br> 混合精度 | 正常训练 <br> 混合精度 | 正常训练：FPGM裁剪、PACT量化 |
+| CRNN | ch_ppocr_mobile_v2_0_rec| 正常训练 <br> 混合精度 | 正常训练 <br> 混合精度 | 正常训练 <br> 混合精度 | 正常训练：PACT量化 |
+| CRNN | ch_ppocr_server_v2_0_rec| 正常训练 <br> 混合精度 | 正常训练 <br> 混合精度 | 正常训练 <br> 混合精度 | 正常训练：PACT量化 |
+|PP-OCR| ch_ppocr_mobile_v2_0| 正常训练 <br> 混合精度 | 正常训练 <br> 混合精度 | 正常训练 <br> 混合精度 | - |
+|PP-OCR| ch_ppocr_server_v2_0| 正常训练 <br> 混合精度 | 正常训练 <br> 混合精度 | 正常训练 <br> 混合精度 | - |
+|PP-OCRv2| ch_PP-OCRv2 | 正常训练 <br> 混合精度 | 正常训练 <br> 混合精度 | 正常训练 <br> 混合精度 | - |
+|PP-OCRv3| ch_PP-OCRv3 | 正常训练 <br> 混合精度 | 正常训练 <br> 混合精度 | 正常训练 <br> 混合精度 | - |
+
+
+- 预测相关：基于训练是否使用量化，可以将训练产出的模型可以分为`正常模型`和`量化模型`，这两类模型对应的预测功能汇总如下，
+
+| 模型类型 |device | batchsize | tensorrt | mkldnn | cpu多线程 |
+|  ----   |  ---- |   ----   |  :----:  |   :----:   |  :----:  |
+| 正常模型 | GPU | 1/6 | fp32/fp16 | - | - |
+| 正常模型 | CPU | 1/6 | - | fp32/fp16 | 支持 |
+| 量化模型 | GPU | 1/6 | int8 | - | - |
+| 量化模型 | CPU | 1/6 | - | int8 | 支持 |
+
+
+## 2. 测试流程
+
+运行环境配置请参考[文档](./install.md)的内容配置TIPC的运行环境。
+
+### 2.1 安装依赖
+- 安装PaddlePaddle >= 2.3
+- 安装PaddleOCR依赖
+    ```
+    pip3 install  -r ../requirements.txt
+    ```
+- 安装autolog（规范化日志输出工具）
+    ```
+    pip3 install https://paddleocr.bj.bcebos.com/libs/auto_log-1.2.0-py3-none-any.whl
+    ```
+- 安装PaddleSlim (可选)
+   ```
+   # 如果要测试量化、裁剪等功能，需要安装PaddleSlim
+   pip3 install paddleslim
+   ```
+
+
+### 2.2 功能测试
+#### 2.2.1 基础训练推理链条
+先运行`prepare.sh`准备数据和模型，然后运行`test_train_inference_python.sh`进行测试，最终在```test_tipc/output```目录下生成`,model_name/lite_train_lite_infer/*.log`格式的日志文件。
+
+
+`test_train_inference_python.sh`包含基础链条的4种运行模式，每种模式的运行数据不同，分别用于测试速度和精度，分别是：
+
+- 模式1：lite_train_lite_infer，使用少量数据训练，用于快速验证训练到预测的走通流程，不验证精度和速度；
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_infer_python.txt 'lite_train_lite_infer'
+bash test_tipc/test_train_inference_python.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_infer_python.txt 'lite_train_lite_infer'
+```
+
+- 模式2：lite_train_whole_infer，使用少量数据训练，一定量数据预测，用于验证训练后的模型执行预测，预测速度是否合理；
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_infer_python.txt  'lite_train_whole_infer'
+bash test_tipc/test_train_inference_python.sh ../test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_infer_python.txt 'lite_train_whole_infer'
+```
+
+- 模式3：whole_infer，不训练，全量数据预测，走通开源模型评估、动转静，检查inference model预测时间和精度;
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_infer_python.txt 'whole_infer'
+# 用法1:
+bash test_tipc/test_train_inference_python.sh ../test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_infer_python.txt 'whole_infer'
+# 用法2: 指定GPU卡预测，第三个传入参数为GPU卡号
+bash test_tipc/test_train_inference_python.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_infer_python.txt 'whole_infer' '1'
+```
+
+- 模式4：whole_train_whole_infer，CE： 全量数据训练，全量数据预测，验证模型训练精度，预测精度，预测速度；
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_infer_python.txt 'whole_train_whole_infer'
+bash test_tipc/test_train_inference_python.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_infer_python.txt 'whole_train_whole_infer'
+```
+
+运行相应指令后，在`test_tipc/output`文件夹下自动会保存运行日志。如'lite_train_lite_infer'模式下，会运行训练+inference的链条，因此，在`test_tipc/output`文件夹有以下文件：
+```
+test_tipc/output/model_name/lite_train_lite_infer/
+|- results_python.log    # 运行指令状态的日志
+|- norm_train_gpus_0_autocast_null/  # GPU 0号卡上正常单机单卡训练的训练日志和模型保存文件夹
+|- norm_train_gpus_0,1_autocast_null/  # GPU 0,1号卡上正常单机多卡训练的训练日志和模型保存文件夹
+......
+|- python_infer_cpu_usemkldnn_False_threads_6_precision_fp32_batchsize_1.log  # CPU上关闭Mkldnn线程数设置为6，测试batch_size=1条件下的fp32精度预测运行日志
+|- python_infer_gpu_usetrt_False_precision_fp32_batchsize_1.log # GPU上关闭TensorRT，测试batch_size=1的fp32精度预测日志
+......
+```
+
+其中`results_python.log`中包含了每条指令的运行状态，如果运行成功会输出：
+```
+[33m Run successfully with command - ch_ppocr_mobile_v2_0_det - python3.7 tools/train.py -c configs/det/ch_ppocr_v2_0/ch_det_mv3_db_v2_0.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained Global.use_gpu=True  Global.save_model_dir=./test_tipc/output/ch_ppocr_mobile_v2_0_det/lite_train_lite_infer/norm_train_gpus_0_autocast_null Global.epoch_num=100     Train.loader.batch_size_per_card=2     !  
+ Run successfully with command - ch_ppocr_mobile_v2_0_det - python3.7 tools/export_model.py -c configs/det/ch_ppocr_v2_0/ch_det_mv3_db_v2_0.yml -o  Global.checkpoints=./test_tipc/output/ch_ppocr_mobile_v2_0_det/lite_train_lite_infer/norm_train_gpus_0_autocast_null/latest Global.save_inference_dir=./test_tipc/output/ch_ppocr_mobile_v2_0_det/lite_train_lite_infer/norm_train_gpus_0_autocast_null > ./test_tipc/output/ch_ppocr_mobile_v2_0_det/lite_train_lite_infer/norm_train_gpus_0_autocast_null_nodes_1_export.log 2>&1 !  
+ Run successfully with command - ch_ppocr_mobile_v2_0_det - python3.7 tools/infer/predict_det.py --use_gpu=True --use_tensorrt=False --precision=fp32 --det_model_dir=./test_tipc/output/ch_ppocr_mobile_v2_0_det/lite_train_lite_infer/norm_train_gpus_0_autocast_null --rec_batch_num=1 --image_dir=./train_data/icdar2015/text_localization/ch4_test_images/ --benchmark=True     > ./test_tipc/output/ch_ppocr_mobile_v2_0_det/lite_train_lite_infer/python_infer_gpu_usetrt_False_precision_fp32_batchsize_1.log 2>&1 !  
+ Run successfully with command - ch_ppocr_mobile_v2_0_det - python3.7 tools/infer/predict_det.py --use_gpu=False --enable_mkldnn=False --cpu_threads=6 --det_model_dir=./test_tipc/output/ch_ppocr_mobile_v2_0_det/lite_train_lite_infer/norm_train_gpus_0_autocast_null --rec_batch_num=1   --image_dir=./train_data/icdar2015/text_localization/ch4_test_images/ --benchmark=True --precision=fp32   > ./test_tipc/output/ch_ppocr_mobile_v2_0_det/lite_train_lite_infer/python_infer_cpu_usemkldnn_False_threads_6_precision_fp32_batchsize_1.log 2>&1 !  
+......
+```
+如果运行失败，会输出：
+```
+Run failed with command - python3.7 tools/train.py -c tests/configs/det_mv3_db.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained Global.use_gpu=True  Global.save_model_dir=./tests/output/norm_train_gpus_0_autocast_null Global.epoch_num=1     Train.loader.batch_size_per_card=2   !
+Run failed with command - python3.7 tools/export_model.py -c tests/configs/det_mv3_db.yml -o  Global.pretrained_model=./tests/output/norm_train_gpus_0_autocast_null/latest Global.save_inference_dir=./tests/output/norm_train_gpus_0_autocast_null!
+......
+```
+可以很方便的根据`results_python.log`中的内容判定哪一个指令运行错误。
+
+#### 2.2.2 PACT在线量化链条
+此外，`test_train_inference_python.sh`还包含PACT在线量化模式，命令如下：
+以ch_PP-OCRv2_det为例，如需测试其他模型更换配置即可。
+
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_PP-OCRv2_det/train_pact_infer_python.txt 'lite_train_lite_infer'
+bash test_tipc/test_train_inference_python.sh ./test_tipc/configs/ch_PP-OCRv2_det/train_pact_infer_python.txt 'lite_train_lite_infer'
+```
+#### 2.2.3 混合精度训练链条
+此外，`test_train_inference_python.sh`还包含混合精度训练模式，命令如下：
+以ch_PP-OCRv2_det为例，如需测试其他模型更换配置即可。
+
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_PP-OCRv2_det/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt 'lite_train_lite_infer'
+bash test_tipc/test_train_inference_python.sh ./test_tipc/configs/ch_PP-OCRv2_det/train_linux_gpu_normal_amp_infer_python_linux_gpu_cpu.txt 'lite_train_lite_infer'
+```
+
+### 2.3 精度测试
+
+使用compare_results.py脚本比较模型预测的结果是否符合预期，主要步骤包括：
+- 提取日志中的预测坐标；
+- 从本地文件中提取保存好的坐标结果；
+- 比较上述两个结果是否符合精度预期，误差大于设置阈值时会报错。
+
+#### 使用方式
+运行命令：
+```shell
+python3.7 test_tipc/compare_results.py --gt_file=./test_tipc/results/python_*.txt  --log_file=./test_tipc/output/python_*.log --atol=1e-3 --rtol=1e-3
+```
+
+参数介绍：
+- gt_file： 指向事先保存好的预测结果路径，支持*.txt 结尾，会自动索引*.txt格式的文件，文件默认保存在test_tipc/result/ 文件夹下
+- log_file: 指向运行test_tipc/test_train_inference_python.sh 脚本的infer模式保存的预测日志，预测日志中打印的有预测结果，比如：文本框，预测文本，类别等等，同样支持python_infer_*.log格式传入
+- atol: 设置的绝对误差
+- rtol: 设置的相对误差
+
+#### 运行结果
+
+正常运行效果如下图：
+<img src="compare_right.png" width="1000">
+
+出现不一致结果时的运行输出：
+<img src="compare_wrong.png" width="1000">
+
+
+## 3. 更多教程
+本文档为功能测试用，更丰富的训练预测使用教程请参考：
+[模型训练](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/training.md)
+[基于Python预测引擎推理](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/inference_ppocr.md)
diff --git a/test_tipc/docs/win_test_train_inference_python.md b/test_tipc/docs/win_test_train_inference_python.md
new file mode 100644
index 0000000..5f80df1
--- /dev/null
+++ b/test_tipc/docs/win_test_train_inference_python.md
@@ -0,0 +1,146 @@
+# Windows端基础训练预测功能测试
+
+Windows端基础训练预测功能测试的主程序为`test_train_inference_python.sh`，可以测试基于Python的模型训练、评估、推理等基本功能，包括裁剪、量化、蒸馏。
+
+## 1. 测试结论汇总
+
+- 训练相关：
+
+| 算法名称 | 模型名称 | 单机单卡 | 单机多卡 | 多机多卡 | 模型压缩（单机多卡） |
+|  :----  |   :----  |    :----  |  :----   |  :----   |  :----   |
+|  DB  | ch_ppocr_mobile_v2_0_det| 正常训练 <br> 混合精度 | - | - | 正常训练：FPGM裁剪、PACT量化 |
+
+
+- 预测相关：基于训练是否使用量化，可以将训练产出的模型可以分为`正常模型`和`量化模型`，这两类模型对应的预测功能汇总如下：
+
+| 模型类型 |device | batchsize | tensorrt | mkldnn | cpu多线程 |
+|  ----   |  ---- |   ----   |  :----:  |   :----:   |  :----:  |
+| 正常模型 | GPU | 1/6 | fp32/fp16 | - | - |
+| 正常模型 | CPU | 1/6 | - | fp32/fp16 | 支持 |
+| 量化模型 | GPU | 1/6 | int8 | - | - |
+| 量化模型 | CPU | 1/6 | - | int8 | 支持 |
+
+
+## 2. 测试流程
+
+运行环境配置请参考[文档](./install.md)的内容配置TIPC的运行环境。
+
+另外，由于Windows上和linux的路径管理方式不同，可以在win上安装gitbash终端，在gitbash中执行指令的方式和在linux端执行指令方式相同，更方便tipc测试。gitbash[下载链接](https://git-scm.com/download/win)。
+
+
+### 2.1 安装依赖
+- 安装PaddlePaddle >= 2.3
+- 安装PaddleOCR依赖
+    ```
+    pip install  -r ../requirements.txt
+    ```
+- 安装autolog（规范化日志输出工具）
+    ```
+    pip install https://paddleocr.bj.bcebos.com/libs/auto_log-1.2.0-py3-none-any.whl
+    ```
+- 安装PaddleSlim (可选)
+   ```
+   # 如果要测试量化、裁剪等功能，需要安装PaddleSlim
+   pip install paddleslim
+   ```
+
+
+### 2.2 功能测试
+先运行`prepare.sh`准备数据和模型，然后运行`test_train_inference_python.sh`进行测试，最终在```test_tipc/output```目录下生成`,model_name/lite_train_lite_infer/*.log`格式的日志文件。
+
+
+`test_train_inference_python.sh`包含基础链条的4种运行模式，每种模式的运行数据不同，分别用于测试速度和精度，分别是：
+
+- 模式1：lite_train_lite_infer，使用少量数据训练，用于快速验证训练到预测的走通流程，不验证精度和速度；
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_windows_gpu_normal_normal_infer_python_windows_cpu_gpu.txt  'lite_train_lite_infer'
+bash test_tipc/test_train_inference_python.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_windows_gpu_normal_normal_infer_python_windows_cpu_gpu.txt  'lite_train_lite_infer'
+```
+
+- 模式2：lite_train_whole_infer，使用少量数据训练，一定量数据预测，用于验证训练后的模型执行预测，预测速度是否合理；
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_windows_gpu_normal_normal_infer_python_windows_cpu_gpu.txt  'lite_train_whole_infer'
+bash test_tipc/test_train_inference_python.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_windows_gpu_normal_normal_infer_python_windows_cpu_gpu.txt  'lite_train_whole_infer'
+```
+
+- 模式3：whole_infer，不训练，全量数据预测，走通开源模型评估、动转静，检查inference model预测时间和精度;
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_windows_gpu_normal_normal_infer_python_windows_cpu_gpu.txt  'whole_infer'
+# 用法1:
+bash test_tipc/test_train_inference_python.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_windows_gpu_normal_normal_infer_python_windows_cpu_gpu.txt  'whole_infer'
+# 用法2: 指定GPU卡预测，第三个传入参数为GPU卡号
+bash test_tipc/test_train_inference_python.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_windows_gpu_normal_normal_infer_python_windows_cpu_gpu.txt  'whole_infer' '1'
+```
+
+- 模式4：whole_train_whole_infer，CE： 全量数据训练，全量数据预测，验证模型训练精度，预测精度，预测速度；
+```shell
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_windows_gpu_normal_normal_infer_python_windows_cpu_gpu.txt  'whole_train_whole_infer'
+bash test_tipc/test_train_inference_python.sh ./test_tipc/configs/ch_ppocr_mobile_v2_0_det/train_windows_gpu_normal_normal_infer_python_windows_cpu_gpu.txt  'whole_train_whole_infer'
+```
+
+运行相应指令后，在`test_tipc/output`文件夹下自动会保存运行日志。如'lite_train_lite_infer'模式下，会运行训练+inference的链条，因此，在`test_tipc/output`文件夹有以下文件：
+```
+test_tipc/output/model_name/lite_train_lite_infer/
+|- results_python.log    # 运行指令状态的日志
+|- norm_train_gpus_0_autocast_null/  # GPU 0号卡上正常训练的训练日志和模型保存文件夹
+......
+|- python_infer_cpu_usemkldnn_False_threads_6_precision_fp32_batchsize_1.log  # CPU上关闭Mkldnn线程数设置为6，测试batch_size=1条件下的fp32精度预测运行日志
+|- python_infer_gpu_usetrt_False_precision_fp32_batchsize_1.log # GPU上关闭TensorRT，测试batch_size=1的fp32精度预测日志
+......
+```
+
+其中`results_python.log`中包含了每条指令的运行状态，如果运行成功会输出：
+```
+Run successfully with command - python3.7 tools/train.py -c tests/configs/det_mv3_db.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained Global.use_gpu=True  Global.save_model_dir=./tests/output/norm_train_gpus_0_autocast_null Global.epoch_num=1     Train.loader.batch_size_per_card=2   !
+Run successfully with command - python3.7 tools/export_model.py -c tests/configs/det_mv3_db.yml -o  Global.pretrained_model=./tests/output/norm_train_gpus_0_autocast_null/latest Global.save_inference_dir=./tests/output/norm_train_gpus_0_autocast_null!
+......
+```
+如果运行失败，会输出：
+```
+Run failed with command - python3.7 tools/train.py -c tests/configs/det_mv3_db.yml -o Global.pretrained_model=./pretrain_models/MobileNetV3_large_x0_5_pretrained Global.use_gpu=True  Global.save_model_dir=./tests/output/norm_train_gpus_0_autocast_null Global.epoch_num=1     Train.loader.batch_size_per_card=2   !
+Run failed with command - python3.7 tools/export_model.py -c tests/configs/det_mv3_db.yml -o  Global.pretrained_model=./tests/output/norm_train_gpus_0_autocast_null/latest Global.save_inference_dir=./tests/output/norm_train_gpus_0_autocast_null!
+......
+```
+可以很方便的根据`results_python.log`中的内容判定哪一个指令运行错误。
+
+
+### 2.3 精度测试
+
+使用compare_results.py脚本比较模型预测的结果是否符合预期，主要步骤包括：
+- 提取日志中的预测坐标；
+- 从本地文件中提取保存好的坐标结果；
+- 比较上述两个结果是否符合精度预期，误差大于设置阈值时会报错。
+
+#### 使用方式
+运行命令：
+```shell
+python test_tipc/compare_results.py --gt_file=./test_tipc/results/python_*.txt  --log_file=./test_tipc/output/python_*.log --atol=1e-3 --rtol=1e-3
+```
+
+参数介绍：
+- gt_file： 指向事先保存好的预测结果路径，支持*.txt 结尾，会自动索引*.txt格式的文件，文件默认保存在test_tipc/result/ 文件夹下
+- log_file: 指向运行test_tipc/test_train_inference_python.sh 脚本的infer模式保存的预测日志，预测日志中打印的有预测结果，比如：文本框，预测文本，类别等等，同样支持python_infer_*.log格式传入
+- atol: 设置的绝对误差
+- rtol: 设置的相对误差
+
+#### 运行结果
+
+正常运行效果如下：
+```
+Assert allclose passed! The results of python_infer_cpu_usemkldnn_False_threads_1_batchsize_1.log and ./test_tipc/results/python_ppocr_det_mobile_results_fp32.txt are consistent!
+```
+
+出现不一致结果时的运行输出：
+```
+......
+Traceback (most recent call last):
+  File "test_tipc/compare_results.py", line 140, in <module>
+    format(filename, gt_filename))
+ValueError: The results of python_infer_cpu_usemkldnn_False_threads_1_batchsize_1.log and the results of ./test_tipc/results/python_ppocr_det_mobile_results_fp32.txt are inconsistent!
+```
+
+
+## 3. 更多教程
+本文档为功能测试用，更丰富的训练预测使用教程请参考：
+[模型训练](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/training.md)
+[基于Python预测引擎推理](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/doc/doc_ch/inference_ppocr.md)
diff --git a/test_tipc/prepare.sh b/test_tipc/prepare.sh
new file mode 100644
index 0000000..7e58741
--- /dev/null
+++ b/test_tipc/prepare.sh
@@ -0,0 +1,912 @@
+#!/bin/bash
+source test_tipc/common_func.sh
+
+FILENAME=$1
+
+# MODE be one of ['lite_train_lite_infer' 'lite_train_whole_infer' 'whole_train_whole_infer',  
+#                 'whole_infer', 'klquant_whole_infer',
+#                 'cpp_infer', 'serving_infer']
+
+MODE=$2
+
+dataline=$(cat ${FILENAME})
+
+# parser params
+IFS=$'\n'
+lines=(${dataline})
+
+# The training params
+model_name=$(func_parser_value "${lines[1]}")
+
+trainer_list=$(func_parser_value "${lines[14]}")
+
+if [ ${MODE} = "benchmark_train" ];then
+    python_name_list=$(func_parser_value "${lines[2]}")
+    array=(${python_name_list}) 
+    python_name=python
+    ${python_name} -m pip install -r requirements.txt
+    if [[ ${model_name} =~ "ch_ppocr_mobile_v2_0_det" || ${model_name} =~ "det_mv3_db_v2_0" ]];then
+        wget -nc -P  ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/MobileNetV3_large_x0_5_pretrained.pdparams  --no-check-certificate
+        rm -rf ./train_data/icdar2015
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/icdar2015_benckmark.tar --no-check-certificate
+        cd ./train_data/ && tar xf icdar2015_benckmark.tar
+        ln -s ./icdar2015_benckmark ./icdar2015
+        cd ../
+        if [[ ${model_name} =~ "ch_ppocr_mobile_v2_0_det" ]];then
+            # expand gt.txt 2 times
+            cd ./train_data/icdar2015/text_localization
+            for i in `seq 2`;do cp train_icdar2015_label.txt dup$i.txt;done
+            cat dup* > train_icdar2015_label.txt && rm -rf dup*
+            cd ../../../
+        fi
+    fi
+    if [[ ${model_name} =~ "PP-OCRv4_mobile_det" ]];then
+        wget -nc -P  ./pretrain_models/ https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_mobile_det_pretrained.pdparams
+  --no-check-certificate
+        rm -rf ./train_data/icdar2015
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/icdar2015_benckmark.tar --no-check-certificate
+        cd ./train_data/ && tar xf icdar2015_benckmark.tar
+        ln -s ./icdar2015_benckmark ./icdar2015
+        cd ../
+    fi
+    if [[ ${model_name} =~ "PP-OCRv4_server_det" ]];then
+        wget -nc -P  ./pretrain_models/ https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv4_server_det_pretrained.pdparams  --no-check-certificate
+        rm -rf ./train_data/icdar2015
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/icdar2015_benckmark.tar --no-check-certificate
+        cd ./train_data/ && tar xf icdar2015_benckmark.tar
+        ln -s ./icdar2015_benckmark ./icdar2015
+        cd ../
+    fi
+    if [[ ${model_name} =~ "PP-OCRv4_mobile_rec" ]];then
+        rm -rf ./train_data/ic15_data
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/ic15_data_benckmark.tar --no-check-certificate
+        cd ./train_data/ && tar xf ic15_data_benckmark.tar
+        ln -s ./ic15_data_benckmark ./ic15_data
+        cd ic15_data
+        mv rec_gt_train4w.txt rec_gt_train.txt
+        cd ../
+        cd ../
+    fi
+    if [[ ${model_name} =~ "PP-OCRv4_server_rec" ]];then
+        rm -rf ./train_data/ic15_data
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/ic15_data_benckmark.tar --no-check-certificate
+        cd ./train_data/ && tar xf ic15_data_benckmark.tar
+        ln -s ./ic15_data_benckmark ./ic15_data
+        cd ic15_data
+        mv rec_gt_train4w.txt rec_gt_train.txt
+        cd ../
+        cd ../
+    fi
+    if [[ ${model_name} =~ "ch_ppocr_server_v2_0_det" || ${model_name} =~ "ch_PP-OCRv3_det" ]];then
+        rm -rf ./train_data/icdar2015
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/icdar2015_benckmark.tar --no-check-certificate
+        cd ./train_data/ && tar xf icdar2015_benckmark.tar
+        ln -s ./icdar2015_benckmark ./icdar2015
+        cd ../
+    fi
+    if [[ ${model_name} =~ "ch_PP-OCRv2_det" ]];then
+        wget  -nc -P  ./pretrain_models/  https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf ch_ppocr_server_v2.0_det_train.tar  && cd ../
+        rm -rf ./train_data/icdar2015
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/icdar2015_benckmark.tar --no-check-certificate
+        cd ./train_data/ && tar xf icdar2015_benckmark.tar
+        ln -s ./icdar2015_benckmark ./icdar2015
+        cd ../
+    fi
+    if [[ ${model_name} =~ "det_r50_vd_east_v2_0" ]]; then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf det_r50_vd_east_v2.0_train.tar && cd ../
+        rm -rf ./train_data/icdar2015
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/icdar2015_benckmark.tar --no-check-certificate
+        cd ./train_data/ && tar xf icdar2015_benckmark.tar
+        ln -s ./icdar2015_benckmark ./icdar2015
+        cd ../
+    fi
+    if [[ ${model_name} =~ "det_r50_db_v2_0" || ${model_name} =~ "det_r50_vd_pse_v2_0" ]];then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/ResNet50_vd_ssld_pretrained.pdparams --no-check-certificate
+        rm -rf ./train_data/icdar2015
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/icdar2015_benckmark.tar --no-check-certificate
+        cd ./train_data/ && tar xf icdar2015_benckmark.tar
+        ln -s ./icdar2015_benckmark ./icdar2015
+        cd ../
+    fi
+    if [[ ${model_name} =~ "det_r18_db_v2_0" ]];then
+        wget -nc -P ./pretrain_models/  https://paddleocr.bj.bcebos.com/pretrained/ResNet18_vd_pretrained.pdparams  --no-check-certificate
+        rm -rf ./train_data/icdar2015
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/icdar2015_benckmark.tar --no-check-certificate
+        cd ./train_data/ && tar xf icdar2015_benckmark.tar
+        ln -s ./icdar2015_benckmark ./icdar2015
+        cd ../
+    fi
+    if [[ ${model_name} =~ "ch_ppocr_mobile_v2_0_rec" || ${model_name} =~ "ch_ppocr_server_v2_0_rec" || ${model_name} =~ "ch_PP-OCRv2_rec" || ${model_name} =~ "rec_mv3_none_bilstm_ctc_v2_0" || ${model_name} =~ "PP-OCRv3_mobile_rec" ]];then
+        rm -rf ./train_data/ic15_data
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/ic15_data_benckmark.tar --no-check-certificate
+        cd ./train_data/ && tar xf ic15_data_benckmark.tar
+        ln -s ./ic15_data_benckmark ./ic15_data
+        cd ../
+    fi
+    if [[ ${model_name} =~ "ch_PP-OCRv2_rec" || ${model_name} =~ "PP-OCRv3_mobile_rec" || ${model_name} =~ "PP-OCRv4_mobile_rec" || ${model_name} =~ "PP-OCRv4_server_rec" ]];then
+        rm -rf ./train_data/ic15_data
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/ic15_data_benckmark.tar --no-check-certificate
+        cd ./train_data/ && tar xf ic15_data_benckmark.tar
+        ln -s ./ic15_data_benckmark ./ic15_data
+        cd ic15_data
+        mv rec_gt_train4w.txt rec_gt_train.txt
+        cd ../
+        cd ../
+    fi
+    if [[ ${model_name} == "en_table_structure" ]];then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_structure_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf en_ppocr_mobile_v2.0_table_structure_train.tar  && cd ../
+        rm -rf ./train_data/pubtabnet
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/pubtabnet_benckmark.tar --no-check-certificate
+        cd ./train_data/ && tar xf pubtabnet_benckmark.tar
+        ln -s ./pubtabnet_benckmark ./pubtabnet
+        cd ../
+    fi
+    if [[ ${model_name} == "slanet" ]];then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/ch_ppstructure_mobile_v2.0_SLANet_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf ch_ppstructure_mobile_v2.0_SLANet_train.tar  && cd ../
+        rm -rf ./train_data/pubtabnet
+        wget -nc -P ./train_data/ https://paddle-model-ecology.bj.bcebos.com/uapi/data/table_rec_dataset_examples.tar --no-check-certificate
+        cd ./train_data/ && tar xf table_rec_dataset_examples.tar
+        ln -s ./table_rec_dataset_examples ./pubtabnet
+        cd pubtabnet
+        for i in `seq 10`;do cp train.txt dup$i.txt;done
+        cat dup* > train.txt && rm -rf dup*
+        cd ../
+        cd ../
+    fi
+    if [[ ${model_name} == "det_r50_dcn_fce_ctw_v2_0" ]]; then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/contribution/det_r50_dcn_fce_ctw_v2.0_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf det_r50_dcn_fce_ctw_v2.0_train.tar && cd ../
+        rm -rf ./train_data/icdar2015
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/icdar2015_benckmark.tar --no-check-certificate
+        cd ./train_data/ && tar xf icdar2015_benckmark.tar
+        ln -s ./icdar2015_benckmark ./icdar2015
+        cd ../
+    fi
+    if [ ${model_name} == "layoutxlm_ser" ] || [ ${model_name} == "vi_layoutxlm_ser" ]; then
+        ${python_name} -m pip install -r ppstructure/kie/requirements.txt
+        ${python_name} -m pip install opencv-python -U
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/ppstructure/dataset/XFUND.tar --no-check-certificate
+        cd ./train_data/ && tar xf XFUND.tar
+        # expand gt.txt 10 times
+        cd XFUND/zh_train
+        for i in `seq 10`;do cp train.json dup$i.txt;done
+        cat dup* > train.json && rm -rf dup*
+        cd ../../
+        
+        cd ../
+    fi
+    if [ ${model_name} == "table_master" ];then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf table_structure_tablemaster_train.tar  && cd ../
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/StructureLabel_val_500.tar --no-check-certificate
+        cd ./train_data/ && tar xf StructureLabel_val_500.tar
+        cd ../
+    fi
+    if [ ${model_name} == "rec_svtrnet" ]; then
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/ic15_data_benckmark.tar --no-check-certificate
+        cd ./train_data/ && tar xf ic15_data_benckmark.tar
+        ln -s ./ic15_data_benckmark ./ic15_data
+        cd ic15_data
+        mv rec_gt_train4w.txt rec_gt_train.txt
+
+        for i in `seq 10`;do cp rec_gt_train.txt dup$i.txt;done
+        cat dup* > rec_gt_train.txt && rm -rf dup*
+
+        cd ../
+        cd ../
+    fi
+fi
+
+if [ ${MODE} = "lite_train_lite_infer" ];then
+    python_name_list=$(func_parser_value "${lines[2]}")
+    array=(${python_name_list}) 
+    python_name=${array[0]}
+    ${python_name} -m pip install -r requirements.txt
+    ${python_name} -m pip install https://paddleocr.bj.bcebos.com/libs/auto_log-1.2.0-py3-none-any.whl
+    ${python_name} -m pip install paddleslim
+    # pretrain lite train data
+    wget -nc -P  ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV3_large_x0_5_pretrained.pdparams  --no-check-certificate
+    wget -nc -P ./pretrain_models/  https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar  --no-check-certificate
+    cd ./pretrain_models/
+    tar xf det_mv3_db_v2.0_train.tar
+    cd ../
+    if [[ ${model_name} =~ "ch_PP-OCRv2_det" ]];then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_distill_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf ch_PP-OCRv2_det_distill_train.tar && cd ../
+    fi
+    if [[ ${model_name} =~ "PP-OCRv3_det" ]];then
+        wget -nc -P ./pretrain_models/ https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams --no-check-certificate
+    fi
+    if [[ ${model_name} =~ "PP-OCRv4_mobile_det" ]];then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/PPLCNetV3_x0_75_ocr_det.pdparams --no-check-certificate
+    fi
+    if [[ ${model_name} =~ "PP-OCRv4_server_det" ]];then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pretrained/PPHGNet_small_ocr_det.pdparams --no-check-certificate
+    fi
+    if [ ${model_name} == "en_table_structure" ] || [ ${model_name} == "en_table_structure_PACT" ];then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_structure_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf en_ppocr_mobile_v2.0_table_structure_train.tar  && cd ../
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar --no-check-certificate
+        cd ./inference/ && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar && cd ../
+    fi
+    if [[ ${model_name} =~ "slanet" ]];then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/en_ppstructure_mobile_v2.0_SLANet_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf en_ppstructure_mobile_v2.0_SLANet_train.tar  && cd ../
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar --no-check-certificate
+        cd ./inference/ && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar && cd ../
+    fi
+    if [[ ${model_name} =~ "det_r50_db_plusplus" ]];then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/ResNet50_dcn_asf_synthtext_pretrained.pdparams --no-check-certificate
+    fi
+    if [ ${model_name} == "table_master" ];then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf table_structure_tablemaster_train.tar  && cd ../
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/StructureLabel_val_500.tar --no-check-certificate
+        cd ./train_data/ && tar xf StructureLabel_val_500.tar && cd ../
+    fi
+    rm -rf ./train_data/icdar2015
+    rm -rf ./train_data/ic15_data
+    rm -rf ./train_data/pubtabnet
+    wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/icdar2015_lite.tar --no-check-certificate
+    wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ic15_data.tar --no-check-certificate
+    wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/pubtabnet.tar --no-check-certificate
+    wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar --no-check-certificate
+    wget -nc -P ./deploy/slim/prune https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/sen.pickle --no-check-certificate
+    
+    cd ./train_data/ && tar xf icdar2015_lite.tar && tar xf ic15_data.tar && tar xf pubtabnet.tar
+    ln -s ./icdar2015_lite ./icdar2015
+    wget -nc -P ./ic15_data/ https://paddleocr.bj.bcebos.com/dataset/rec_gt_train_lite.txt --no-check-certificate
+    wget -nc -P ./ic15_data/ https://paddleocr.bj.bcebos.com/dataset/rec_gt_test_lite.txt --no-check-certificate
+    mv ic15_data/rec_gt_train_lite.txt ic15_data/rec_gt_train.txt
+    mv ic15_data/rec_gt_test_lite.txt ic15_data/rec_gt_test.txt
+    cd ../
+    cd ./inference && tar xf rec_inference.tar && cd ../
+    if [ ${model_name} == "ch_PP-OCRv2_det" ] || [ ${model_name} == "ch_PP-OCRv2_det_PACT" ]; then
+        wget  -nc -P  ./pretrain_models/  https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf ch_ppocr_server_v2.0_det_train.tar  && cd ../
+    fi
+    if [ ${model_name} == "ch_PP-OCRv2_rec" ] || [ ${model_name} == "ch_PP-OCRv2_rec_PACT" ]; then
+        wget  -nc -P  ./pretrain_models/  https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf ch_PP-OCRv2_rec_train.tar && cd ../
+    fi
+    if [ ${model_name} == "PP-OCRv3_mobile_rec" ] || [ ${model_name} == "PP-OCRv3_mobile_rec_PACT" ]; then
+        wget  -nc -P  ./pretrain_models/  https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf PP-OCRv3_mobile_rec_train.tar && cd ../
+    fi
+    if [ ${model_name} == "det_r18_db_v2_0" ]; then
+        wget -nc -P ./pretrain_models/  https://paddleocr.bj.bcebos.com/pretrained/ResNet18_vd_pretrained.pdparams  --no-check-certificate
+    fi
+    if [ ${model_name} == "en_server_pgnetA" ]; then
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/total_text_lite.tar --no-check-certificate
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/en_server_pgnetA.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf en_server_pgnetA.tar && cd ../
+        cd ./train_data && tar xf total_text_lite.tar && ln -s total_text_lite total_text && cd ../
+    fi
+    if [ ${model_name} == "det_r50_vd_sast_icdar15_v2_0" ] || [ ${model_name} == "det_r50_vd_sast_totaltext_v2_0" ]; then
+        wget -nc -P ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet50_vd_ssld_pretrained.pdparams --no-check-certificate
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar --no-check-certificate
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/total_text_lite.tar --no-check-certificate
+        cd ./train_data && tar xf total_text_lite.tar && ln -s total_text_lite total_text  && cd ../
+        cd ./pretrain_models && tar xf det_r50_vd_sast_icdar15_v2.0_train.tar && cd ../
+    fi
+    if [ ${model_name} == "det_mv3_db_v2_0" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar  --no-check-certificate
+        cd ./inference/ && tar xf det_mv3_db_v2.0_train.tar && cd ../
+    fi
+    if [ ${model_name} == "det_r50_db_v2_0" ]; then
+        wget -nc -P ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet50_vd_ssld_pretrained.pdparams --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar  --no-check-certificate
+        cd ./inference/ && tar xf det_r50_vd_db_v2.0_train.tar && cd ../
+    fi
+    if [ ${model_name} == "ch_ppocr_mobile_v2_0_rec_FPGM" ]; then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf ch_ppocr_mobile_v2.0_rec_train.tar && cd ../
+        ${python_name} -m pip install paddleslim
+    fi
+    if [ ${model_name} == "ch_ppocr_mobile_v2_0_det_FPGM" ]; then
+        ${python_name} -m pip install paddleslim
+    fi
+    if [ ${model_name} == "det_r50_vd_pse_v2_0" ]; then
+        wget -nc -P ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet50_vd_ssld_pretrained.pdparams --no-check-certificate
+    fi
+    if [ ${model_name} == "det_mv3_east_v2_0" ]; then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_east_v2.0_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf det_mv3_east_v2.0_train.tar && cd ../
+    fi
+    if [ ${model_name} == "det_r50_vd_east_v2_0" ]; then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf det_r50_vd_east_v2.0_train.tar && cd ../
+    fi
+    if [ ${model_name} == "det_r50_dcn_fce_ctw_v2_0" ]; then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/contribution/det_r50_dcn_fce_ctw_v2.0_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf det_r50_dcn_fce_ctw_v2.0_train.tar & cd ../
+    fi
+    if [ ${model_name} == "rec_r32_gaspin_bilstm_att" ]; then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/rec_r32_gaspin_bilstm_att_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf rec_r32_gaspin_bilstm_att_train.tar && cd ../
+    fi
+    if [[ ${model_name} =~ "layoutxlm_ser" ]]; then
+        ${python_name} -m pip install -r ppstructure/kie/requirements.txt
+        ${python_name} -m pip install opencv-python -U
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/ppstructure/dataset/XFUND.tar --no-check-certificate
+        cd ./train_data/ && tar xf XFUND.tar
+        cd ../
+
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf ser_LayoutXLM_xfun_zh.tar  && cd ../
+    fi
+    if [[ ${model_name} =~ "vi_layoutxlm_ser" ]]; then
+        ${python_name} -m pip install -r ppstructure/kie/requirements.txt
+        ${python_name} -m pip install opencv-python -U
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/ppstructure/dataset/XFUND.tar --no-check-certificate
+        cd ./train_data/ && tar xf XFUND.tar
+        cd ../
+        if [ ${model_name} == "vi_layoutxlm_ser_PACT" ]; then
+            wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_pretrained.tar --no-check-certificate
+            cd ./pretrain_models/ && tar xf ser_vi_layoutxlm_xfund_pretrained.tar  && cd ../
+        fi
+    fi
+    if [ ${model_name} == "det_r18_ct" ]; then
+        wget -nc -P ./pretrain_models/  https://paddleocr.bj.bcebos.com/pretrained/ResNet18_vd_pretrained.pdparams  --no-check-certificate
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/ct_tipc/total_text_lite2.tar --no-check-certificate
+        cd ./train_data && tar xf total_text_lite2.tar && ln -s total_text_lite2 total_text && cd ../
+    fi
+    if [ ${model_name} == "sr_telescope" ]; then
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/TextZoom.tar --no-check-certificate
+        cd ./train_data/ && tar xf TextZoom.tar && cd ../
+    fi
+    if [ ${model_name} == "rec_d28_can" ]; then
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/CROHME_lite.tar --no-check-certificate
+        cd ./train_data/ && tar xf CROHME_lite.tar && cd ../
+    fi
+
+elif [ ${MODE} = "whole_train_whole_infer" ];then
+    wget -nc -P  ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV3_large_x0_5_pretrained.pdparams --no-check-certificate
+    rm -rf ./train_data/icdar2015
+    rm -rf ./train_data/ic15_data
+    rm -rf ./train_data/pubtabnet
+    wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/icdar2015.tar --no-check-certificate
+    wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ic15_data.tar --no-check-certificate
+    wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/pubtabnet.tar --no-check-certificate
+    cd ./train_data/ && tar xf icdar2015.tar && tar xf ic15_data.tar && tar xf pubtabnet.tar 
+    wget -nc -P ./ic15_data/ https://paddleocr.bj.bcebos.com/dataset/rec_gt_train_lite.txt --no-check-certificate
+    wget -nc -P ./ic15_data/ https://paddleocr.bj.bcebos.com/dataset/rec_gt_test_lite.txt --no-check-certificate
+    cd ../
+    if [ ${model_name} == "ch_PP-OCRv2_det" ]; then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_distill_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf ch_PP-OCRv2_det_distill_train.tar && cd ../
+    fi
+    if [ ${model_name} == "PP-OCRv3_det" ]; then
+        wget -nc -P ./pretrain_models/ https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams --no-check-certificate
+    fi
+    if [ ${model_name} == "en_server_pgnetA" ]; then
+        wget -nc -P ./train_data/  https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/total_text_lite.tar  --no-check-certificate
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/en_server_pgnetA.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf en_server_pgnetA.tar && cd ../
+        cd ./train_data && tar xf total_text.tar && ln -s total_text_lite total_text  && cd ../
+    fi
+    if [ ${model_name} == "det_r50_vd_sast_totaltext_v2_0" ]; then
+        wget -nc -P ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet50_vd_ssld_pretrained.pdparams --no-check-certificate
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/total_text_lite.tar  --no-check-certificate
+        cd ./train_data && tar xf total_text.tar && ln -s total_text_lite total_text  && cd ../
+    fi
+    if [[ ${model_name} =~ "en_table_structure" ]];then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_structure_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf en_ppocr_mobile_v2.0_table_structure_train.tar  && cd ../
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar --no-check-certificate
+        cd ./inference/ && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar && cd ../
+    fi
+elif [ ${MODE} = "lite_train_whole_infer" ];then
+    wget -nc -P  ./pretrain_models/ https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV3_large_x0_5_pretrained.pdparams --no-check-certificate
+    rm -rf ./train_data/icdar2015
+    rm -rf ./train_data/ic15_data
+    rm -rf ./train_data/pubtabnet
+    wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/icdar2015_infer.tar --no-check-certificate
+    wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ic15_data.tar --no-check-certificate
+    wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/pubtabnet.tar --no-check-certificate
+    cd ./train_data/ && tar xf icdar2015_infer.tar && tar xf ic15_data.tar && tar xf pubtabnet.tar
+    ln -s ./icdar2015_infer ./icdar2015
+    wget -nc -P ./ic15_data/ https://paddleocr.bj.bcebos.com/dataset/rec_gt_train_lite.txt --no-check-certificate
+    wget -nc -P ./ic15_data/ https://paddleocr.bj.bcebos.com/dataset/rec_gt_test_lite.txt --no-check-certificate
+    cd ../
+    if [ ${model_name} == "ch_PP-OCRv2_det" ]; then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_distill_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf ch_PP-OCRv2_det_distill_train.tar && cd ../
+    fi
+    if [ ${model_name} == "PP-OCRv3_det" ]; then
+        wget -nc -P ./pretrain_models/ https://paddle-model-ecology.bj.bcebos.com/paddlex/official_pretrained_model/PP-OCRv3_mobile_det_pretrained.pdparams --no-check-certificate
+    fi
+    if [[ ${model_name} =~ "en_table_structure" ]];then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.1/table/en_ppocr_mobile_v2.0_table_structure_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf en_ppocr_mobile_v2.0_table_structure_train.tar  && cd ../
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar --no-check-certificate
+        cd ./inference/ && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar && cd ../
+    fi
+elif [ ${MODE} = "whole_infer" ];then
+    python_name_list=$(func_parser_value "${lines[2]}")
+    array=(${python_name_list}) 
+    python_name=${array[0]}
+    ${python_name} -m pip install paddleslim
+    ${python_name} -m pip install -r requirements.txt
+    wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar --no-check-certificate
+    wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar --no-check-certificate
+    cd ./inference && tar xf rec_inference.tar  && tar xf ch_det_data_50.tar && cd ../
+    wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/ppstructure/dataset/XFUND.tar --no-check-certificate
+    wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/pubtabnet.tar --no-check-certificate
+    cd ./train_data/ && tar xf XFUND.tar && tar xf pubtabnet.tar && cd ../
+    head -n 2 train_data/XFUND/zh_val/val.json > train_data/XFUND/zh_val/val_lite.json
+    mv train_data/XFUND/zh_val/val_lite.json train_data/XFUND/zh_val/val.json
+    if [ ${model_name} = "ch_ppocr_mobile_v2_0_det" ]; then
+        eval_model_name="ch_ppocr_mobile_v2.0_det_train"
+        rm -rf ./train_data/icdar2015
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_train.tar --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar --no-check-certificate
+        cd ./inference && tar xf ${eval_model_name}.tar && tar xf ch_det_data_50.tar && tar xf ch_ppocr_mobile_v2.0_det_infer.tar && cd ../
+    elif [ ${model_name} = "ch_ppocr_mobile_v2_0_det_PACT" ]; then
+        eval_model_name="ch_ppocr_mobile_v2.0_det_prune_infer"
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/slim/ch_ppocr_mobile_v2.0_det_prune_infer.tar --no-check-certificate
+        cd ./inference && tar xf ${eval_model_name}.tar && tar xf ch_det_data_50.tar && cd ../
+    elif [ ${model_name} = "ch_ppocr_server_v2_0_det" ]; then
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_train.tar --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_server_v2.0_det_train.tar && tar xf ch_det_data_50.tar && cd ../
+    elif  [ ${model_name} = "ch_ppocr_mobile_v2_0" ]; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_det_infer.tar && tar xf ch_ppocr_mobile_v2.0_rec_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif  [ ${model_name} = "ch_ppocr_server_v2_0" ]; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_server_v2.0_det_infer.tar && tar xf ch_ppocr_server_v2.0_rec_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif [ ${model_name} = "ch_ppocr_mobile_v2_0_rec_PACT" ]; then
+        eval_model_name="ch_ppocr_mobile_v2.0_rec_slim_infer"
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_slim_infer.tar --no-check-certificate
+        cd ./inference && tar xf ${eval_model_name}.tar && cd ../
+    elif [ ${model_name} = "ch_ppocr_mobile_v2_0_rec_FPGM" ]; then
+        eval_model_name="ch_PP-OCRv2_rec_infer"
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar --no-check-certificate
+        cd ./inference && tar xf ${eval_model_name}.tar && cd ../
+    fi 
+    if [[ ${model_name} =~ "ch_PP-OCRv2" ]]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar --no-check-certificate
+        cd ./inference && tar xf ch_PP-OCRv2_det_infer.tar && tar xf ch_PP-OCRv2_rec_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    fi
+    if [[ ${model_name} =~ "ch_PP-OCRv3" ]]; then
+        wget -nc -P ./inference/ https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar --no-check-certificate
+        cd ./inference && tar xf PP-OCRv3_mobile_det_infer.tar && tar xf PP-OCRv3_mobile_rec_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    fi
+    if [[ ${model_name} =~ "ch_PP-OCRv2_det" ]]; then
+        eval_model_name="ch_PP-OCRv2_det_infer"
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer.tar --no-check-certificate
+        cd ./inference && tar xf ${eval_model_name}.tar && tar xf ch_det_data_50.tar && cd ../
+    fi
+    if [[ ${model_name} =~ "ch_PP-OCRv3_det" ]]; then
+        eval_model_name="PP-OCRv3_mobile_det_infer"
+        wget -nc -P ./inference/ https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar --no-check-certificate
+        cd ./inference && tar xf ${eval_model_name}.tar && tar xf ch_det_data_50.tar && cd ../
+    fi
+    if [[ ${model_name} =~ "ch_PP-OCRv2_rec" ]]; then
+        eval_model_name="ch_PP-OCRv2_rec_infer"
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_slim_quant_infer.tar --no-check-certificate
+        cd ./inference && tar xf ${eval_model_name}.tar && tar xf ch_PP-OCRv2_rec_slim_quant_infer.tar && cd ../
+    fi   
+    if [[ ${model_name} =~ "PP-OCRv3_mobile_rec" ]]; then
+        eval_model_name="PP-OCRv3_mobile_rec_infer"
+        wget -nc -P ./inference/ https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/PP-OCRv3_mobile_rec_slim_infer.tar --no-check-certificate
+        cd ./inference && tar xf ${eval_model_name}.tar && tar xf PP-OCRv3_mobile_rec_slim_infer.tar && cd ../
+    fi
+    if [[ ${model_name} == "PP-OCRv3_mobile_rec_PACT" ]]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/PP-OCRv3_mobile_rec_slim_infer.tar --no-check-certificate
+        cd ./inference && tar xf PP-OCRv3_mobile_rec_slim_infer.tar && cd ../
+    fi  
+    if [ ${model_name} == "en_server_pgnetA" ]; then
+        wget -nc -P ./inference/  https://paddleocr.bj.bcebos.com/dygraph_v2.0/pgnet/en_server_pgnetA.tar  --no-check-certificate
+        cd ./inference && tar xf en_server_pgnetA.tar && tar xf ch_det_data_50.tar && cd ../
+    fi
+    if [ ${model_name} == "det_r50_vd_sast_icdar15_v2_0" ]; then
+        wget -nc -P  ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_icdar15_v2.0_train.tar --no-check-certificate
+        cd ./inference/ && tar xf det_r50_vd_sast_icdar15_v2.0_train.tar && tar xf ch_det_data_50.tar && cd ../
+    fi
+    if [ ${model_name} == "rec_mv3_none_none_ctc_v2_0" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_none_ctc_v2.0_train.tar --no-check-certificate
+        cd ./inference/ && tar xf rec_mv3_none_none_ctc_v2.0_train.tar && cd ../
+    fi
+    if [ ${model_name} == "rec_r34_vd_none_none_ctc_v2_0" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_none_ctc_v2.0_train.tar --no-check-certificate
+        cd ./inference/ && tar xf rec_r34_vd_none_none_ctc_v2.0_train.tar && cd ../
+    fi
+    if [ ${model_name} == "rec_mv3_none_bilstm_ctc_v2_0" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_none_bilstm_ctc_v2.0_train.tar --no-check-certificate
+        cd ./inference/ && tar xf rec_mv3_none_bilstm_ctc_v2.0_train.tar && cd ../
+    fi
+    if [ ${model_name} == "rec_r34_vd_none_bilstm_ctc_v2_0" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_none_bilstm_ctc_v2.0_train.tar --no-check-certificate
+        cd ./inference/ && tar xf rec_r34_vd_none_bilstm_ctc_v2.0_train.tar && cd ../
+    fi
+    if [ ${model_name} == "rec_mv3_tps_bilstm_ctc_v2_0" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_tps_bilstm_ctc_v2.0_train.tar --no-check-certificate
+        cd ./inference/ && tar xf rec_mv3_tps_bilstm_ctc_v2.0_train.tar && cd ../
+    fi
+    if [ ${model_name} == "rec_r34_vd_tps_bilstm_ctc_v2_0" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_ctc_v2.0_train.tar --no-check-certificate
+        cd ./inference/ && tar xf rec_r34_vd_tps_bilstm_ctc_v2.0_train.tar && cd ../
+    fi
+    if [ ${model_name} == "ch_ppocr_server_v2_0_rec" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_train.tar --no-check-certificate
+        cd ./inference/ && tar xf ch_ppocr_server_v2.0_rec_train.tar && cd ../
+    fi
+    if [ ${model_name} == "ch_ppocr_mobile_v2_0_rec" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_train.tar --no-check-certificate
+        cd ./inference/ && tar xf ch_ppocr_mobile_v2.0_rec_train.tar && cd ../
+    fi
+    if [ ${model_name} == "rec_mtb_nrtr" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mtb_nrtr_train.tar --no-check-certificate
+        cd ./inference/ && tar xf rec_mtb_nrtr_train.tar && cd ../
+    fi
+    if [ ${model_name} == "rec_mv3_tps_bilstm_att_v2_0" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_mv3_tps_bilstm_att_v2.0_train.tar --no-check-certificate
+        cd ./inference/ && tar xf rec_mv3_tps_bilstm_att_v2.0_train.tar && cd ../
+    fi
+    if [ ${model_name} == "rec_r34_vd_tps_bilstm_att_v2_0" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r34_vd_tps_bilstm_att_v2.0_train.tar --no-check-certificate
+        cd ./inference/ && tar xf rec_r34_vd_tps_bilstm_att_v2.0_train.tar && cd ../
+    fi
+    if [ ${model_name} == "rec_r31_sar" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.1/rec/rec_r31_sar_train.tar --no-check-certificate
+        cd ./inference/ && tar xf rec_r31_sar_train.tar && cd ../
+    fi
+    if [ ${model_name} == "rec_r50_fpn_vd_none_srn" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/rec_r50_vd_srn_train.tar --no-check-certificate
+        cd ./inference/ && tar xf rec_r50_vd_srn_train.tar && cd ../
+    fi
+    
+    if [ ${model_name} == "det_r50_vd_sast_totaltext_v2_0" ]; then
+        wget -nc -P  ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_sast_totaltext_v2.0_train.tar --no-check-certificate
+        cd ./inference/ && tar xf det_r50_vd_sast_totaltext_v2.0_train.tar && cd ../
+    fi
+    if [ ${model_name} == "det_mv3_db_v2_0" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_db_v2.0_train.tar  --no-check-certificate
+        cd ./inference/ && tar xf det_mv3_db_v2.0_train.tar && tar xf ch_det_data_50.tar && cd ../
+    fi
+    if [ ${model_name} == "det_r50_db_v2_0" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_db_v2.0_train.tar  --no-check-certificate
+        cd ./inference/ && tar xf det_r50_vd_db_v2.0_train.tar && tar xf ch_det_data_50.tar && cd ../
+    fi
+    if [ ${model_name} == "det_mv3_pse_v2_0" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_mv3_pse_v2.0_train.tar --no-check-certificate
+        cd ./inference/ && tar xf det_mv3_pse_v2.0_train.tar & cd ../
+    fi
+    if [ ${model_name} == "det_r50_vd_pse_v2_0" ]; then
+        wget -nc -P ./inference/  https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/det_r50_vd_pse_v2.0_train.tar  --no-check-certificate
+        cd ./inference/ && tar xf det_r50_vd_pse_v2.0_train.tar & cd ../
+    fi
+    if [ ${model_name} == "det_mv3_east_v2_0" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_mv3_east_v2.0_train.tar --no-check-certificate
+        cd ./inference/ && tar xf det_mv3_east_v2.0_train.tar & cd ../
+    fi
+    if [ ${model_name} == "det_r50_vd_east_v2_0" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/en/det_r50_vd_east_v2.0_train.tar --no-check-certificate
+        cd ./inference/ && tar xf det_r50_vd_east_v2.0_train.tar & cd ../
+    fi
+    if [ ${model_name} == "det_r50_dcn_fce_ctw_v2_0" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/contribution/det_r50_dcn_fce_ctw_v2.0_train.tar --no-check-certificate
+        cd ./inference/ && tar xf det_r50_dcn_fce_ctw_v2.0_train.tar & cd ../
+    fi
+    if [[ ${model_name} =~ "en_table_structure" ]];then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar --no-check-certificate
+
+        cd ./inference/ && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar
+        if [ ${model_name} == "en_table_structure" ]; then
+            wget -nc https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_infer.tar --no-check-certificate
+            tar xf en_ppocr_mobile_v2.0_table_structure_infer.tar
+        elif [ ${model_name} == "en_table_structure_PACT" ]; then
+            wget -nc https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_slim_infer.tar --no-check-certificate
+            tar xf en_ppocr_mobile_v2.0_table_structure_slim_infer.tar
+        fi
+        cd ../
+    fi
+    if [[ ${model_name} =~ "slanet" ]];then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/en_ppstructure_mobile_v2.0_SLANet_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar --no-check-certificate
+        cd ./inference/ && tar xf en_ppstructure_mobile_v2.0_SLANet_infer.tar && tar xf PP-OCRv3_mobile_det_infer.tar && tar xf PP-OCRv3_mobile_rec_infer.tar && cd ../
+    fi
+    if [[ ${model_name} =~ "vi_layoutxlm_ser" ]]; then
+        ${python_name} -m pip install -r ppstructure/kie/requirements.txt
+        ${python_name} -m pip install opencv-python -U 
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/ppstructure/models/vi_layoutxlm/ser_vi_layoutxlm_xfund_infer.tar --no-check-certificate
+        cd ./inference/ && tar xf ser_vi_layoutxlm_xfund_infer.tar & cd ../
+    fi
+    if [[ ${model_name} =~ "layoutxlm_ser" ]]; then
+        ${python_name} -m pip install -r ppstructure/kie/requirements.txt
+        ${python_name} -m pip install opencv-python -U 
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh_infer.tar --no-check-certificate
+        cd ./inference/ && tar xf ser_LayoutXLM_xfun_zh_infer.tar & cd ../
+    fi
+fi
+
+if [[ ${model_name} =~ "KL" ]]; then
+    wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/icdar2015_lite.tar --no-check-certificate
+    cd ./train_data/ && tar xf icdar2015_lite.tar && rm -rf ./icdar2015 && ln -s ./icdar2015_lite ./icdar2015 && cd ../
+    if [ ${model_name} = "ch_ppocr_mobile_v2_0_det_KL" ]; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar --no-check-certificate
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_det_infer.tar && tar xf ch_det_data_50.tar && cd ../ 
+    fi
+    if [ ${model_name} = "ch_PP-OCRv2_rec_KL" ]; then
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar  --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar  --no-check-certificate
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ic15_data.tar --no-check-certificate
+        cd ./train_data/ && tar xf ic15_data.tar && cd ../
+        cd ./inference && tar xf rec_inference.tar && tar xf ch_PP-OCRv2_rec_infer.tar && cd ../
+    fi
+    if [ ${model_name} = "PP-OCRv3_mobile_rec_KL" ]; then
+        wget -nc -P ./inference https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar  --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar  --no-check-certificate
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ic15_data.tar --no-check-certificate
+        cd ./train_data/ && tar xf ic15_data.tar 
+        wget -nc -P ./ic15_data/ https://paddleocr.bj.bcebos.com/dataset/rec_gt_train_lite.txt --no-check-certificate
+        wget -nc -P ./ic15_data/ https://paddleocr.bj.bcebos.com/dataset/rec_gt_test_lite.txt --no-check-certificate
+        cd ../
+        cd ./inference && tar xf rec_inference.tar && tar xf PP-OCRv3_mobile_rec_infer.tar && cd ../
+    fi
+    if [ ${model_name} = "ch_PP-OCRv2_det_KL" ]; then
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_PP-OCRv2_det_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    fi 
+    if [ ${model_name} = "ch_PP-OCRv3_det_KL" ]; then
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+        wget -nc -P ./inference https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar  --no-check-certificate
+        cd ./inference && tar xf PP-OCRv3_mobile_det_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    fi 
+    if [ ${model_name} = "ch_ppocr_mobile_v2_0_rec_KL" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar  --no-check-certificate
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ic15_data.tar --no-check-certificate
+        cd ./train_data/ && tar xf ic15_data.tar && cd ../
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_rec_infer.tar && tar xf rec_inference.tar &&  cd ../ 
+    fi
+    if [ ${model_name} = "en_table_structure_KL" ];then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar --no-check-certificate
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/dataset/pubtabnet.tar --no-check-certificate
+        cd ./inference/ && tar xf en_ppocr_mobile_v2.0_table_structure_infer.tar && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar && cd ../
+        cd ./train_data/ && tar xf pubtabnet.tar && cd ../
+    fi
+    if [[ ${model_name} =~ "layoutxlm_ser_KL" ]]; then
+        wget -nc -P ./train_data/ https://paddleocr.bj.bcebos.com/ppstructure/dataset/XFUND.tar --no-check-certificate
+        cd ./train_data/ && tar xf XFUND.tar && cd ../
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/pplayout/ser_LayoutXLM_xfun_zh_infer.tar --no-check-certificate
+        cd ./inference/ && tar xf ser_LayoutXLM_xfun_zh_infer.tar & cd ../
+    fi
+fi
+
+if [ ${MODE} = "cpp_infer" ];then
+    if [ ${model_name} = "ch_ppocr_mobile_v2_0_det" ]; then
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_det_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif [ ${model_name} = "ch_ppocr_mobile_v2_0_det_KL" ]; then
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_ppocr_mobile_v2.0_det_klquant_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_det_klquant_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif [ ${model_name} = "ch_ppocr_mobile_v2_0_det_PACT" ]; then
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_ppocr_mobile_v2.0_det_pact_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_det_pact_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif [ ${model_name} = "ch_ppocr_mobile_v2_0_rec" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_rec_infer.tar && tar xf rec_inference.tar && cd ../
+    elif [ ${model_name} = "ch_ppocr_mobile_v2_0_rec_KL" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_ppocr_mobile_v2.0_rec_klquant_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_rec_klquant_infer.tar && tar xf rec_inference.tar && cd ../
+    elif [ ${model_name} = "ch_ppocr_mobile_v2_0_rec_PACT" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_ppocr_mobile_v2.0_rec_pact_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_rec_pact_infer.tar && tar xf rec_inference.tar && cd ../
+    elif [ ${model_name} = "ch_ppocr_server_v2_0_det" ]; then
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_server_v2.0_det_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif [ ${model_name} = "ch_ppocr_server_v2_0_rec" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_server_v2.0_rec_infer.tar && tar xf rec_inference.tar && cd ../
+    elif [ ${model_name} = "ch_PP-OCRv2_det" ]; then
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_PP-OCRv2_det_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif [ ${model_name} = "ch_PP-OCRv2_det_KL" ]; then
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_PP-OCRv2_det_klquant_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_PP-OCRv2_det_klquant_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif [ ${model_name} = "ch_PP-OCRv2_det_PACT" ]; then
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_PP-OCRv2_det_pact_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_PP-OCRv2_det_pact_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif [ ${model_name} = "ch_PP-OCRv2_rec" ]; then
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_PP-OCRv2_rec_infer.tar && tar xf rec_inference.tar && cd ../
+    elif [ ${model_name} = "ch_PP-OCRv2_rec_KL" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_PP-OCRv2_rec_klquant_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_PP-OCRv2_rec_klquant_infer.tar && tar xf rec_inference.tar && cd ../
+    elif [ ${model_name} = "ch_PP-OCRv2_rec_PACT" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_PP-OCRv2_rec_pact_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_PP-OCRv2_rec_pact_infer.tar && tar xf rec_inference.tar && cd ../
+    elif [ ${model_name} = "ch_PP-OCRv3_det" ]; then
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar  --no-check-certificate
+        cd ./inference && tar xf PP-OCRv3_mobile_det_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif [ ${model_name} = "ch_PP-OCRv3_det_KL" ]; then
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_PP-OCRv3_det_klquant_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_PP-OCRv3_det_klquant_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif [ ${model_name} = "ch_PP-OCRv3_det_PACT" ]; then
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_PP-OCRv3_det_pact_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_PP-OCRv3_det_pact_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif [ ${model_name} = "PP-OCRv3_mobile_rec" ]; then
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar  --no-check-certificate
+        cd ./inference && tar xf PP-OCRv3_mobile_rec_infer.tar && tar xf rec_inference.tar && cd ../
+    elif [ ${model_name} = "PP-OCRv3_mobile_rec_KL" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/PP-OCRv3_mobile_rec_klquant_infer.tar  --no-check-certificate
+        cd ./inference && tar xf PP-OCRv3_mobile_rec_klquant_infer.tar && tar xf rec_inference.tar && cd ../
+    elif [ ${model_name} = "PP-OCRv3_mobile_rec_PACT" ]; then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/PP-OCRv3_mobile_rec_pact_infer.tar  --no-check-certificate
+        cd ./inference && tar xf PP-OCRv3_mobile_rec_pact_infer.tar && tar xf rec_inference.tar && cd ../
+    elif  [ ${model_name} = "ch_ppocr_mobile_v2_0" ]; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar  --no-check-certificate
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_det_infer.tar && tar xf ch_ppocr_mobile_v2.0_rec_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif  [ ${model_name} = "ch_ppocr_server_v2_0" ]; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar  --no-check-certificate
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_server_v2.0_det_infer.tar && tar xf ch_ppocr_server_v2.0_rec_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif  [ ${model_name} = "ch_PP-OCRv2" ]; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer.tar  --no-check-certificate
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar  --no-check-certificate
+        cd ./inference && tar xf ch_PP-OCRv2_det_infer.tar && tar xf ch_PP-OCRv2_rec_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif  [ ${model_name} = "ch_PP-OCRv3" ]; then
+        wget -nc  -P ./inference https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar  --no-check-certificate
+        wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+        wget -nc  -P ./inference https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar  --no-check-certificate
+        cd ./inference && tar xf PP-OCRv3_mobile_det_infer.tar && tar xf PP-OCRv3_mobile_rec_infer.tar && tar xf ch_det_data_50.tar && cd ../
+    elif [[ ${model_name} =~ "en_table_structure" ]];then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_det_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_rec_infer.tar --no-check-certificate
+
+        cd ./inference/ && tar xf en_ppocr_mobile_v2.0_table_det_infer.tar && tar xf en_ppocr_mobile_v2.0_table_rec_infer.tar
+        if [ ${model_name} == "en_table_structure" ]; then
+            wget -nc https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_infer.tar --no-check-certificate
+            tar xf en_ppocr_mobile_v2.0_table_structure_infer.tar
+        elif [ ${model_name} == "en_table_structure_PACT" ]; then
+            wget -nc https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_slim_infer.tar --no-check-certificate
+            tar xf en_ppocr_mobile_v2.0_table_structure_slim_infer.tar
+        fi
+        cd ../
+    elif [[ ${model_name} =~ "slanet" ]];then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/ch_ppstructure_mobile_v2.0_SLANet_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar --no-check-certificate
+        wget -nc -P ./inference/ https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar --no-check-certificate
+        cd ./inference/ && tar xf ch_ppstructure_mobile_v2.0_SLANet_infer.tar && tar xf PP-OCRv3_mobile_det_infer.tar && tar xf PP-OCRv3_mobile_rec_infer.tar && cd ../
+    fi
+fi
+
+if [ ${MODE} = "serving_infer" ];then
+    # prepare serving env
+    python_name_list=$(func_parser_value "${lines[2]}")
+    IFS='|'
+    array=(${python_name_list})
+    python_name=${array[0]}
+    ${python_name} -m pip install paddle-serving-server-gpu
+    ${python_name} -m pip install paddle_serving_client
+    ${python_name} -m pip install paddle-serving-app
+    ${python_name} -m pip install https://paddleocr.bj.bcebos.com/libs/auto_log-1.2.0-py3-none-any.whl
+    # wget model
+    if [ ${model_name} == "ch_ppocr_mobile_v2_0_det_KL" ] || [ ${model_name} == "ch_ppocr_mobile_v2.0_rec_KL" ] ; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_ppocr_mobile_v2.0_det_klquant_infer.tar --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_ppocr_mobile_v2.0_rec_klquant_infer.tar --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_det_klquant_infer.tar && tar xf ch_ppocr_mobile_v2.0_rec_klquant_infer.tar && cd ../
+    elif [ ${model_name} == "ch_PP-OCRv2_det_KL" ] || [ ${model_name} == "ch_PP-OCRv2_rec_KL" ] ; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_PP-OCRv2_det_klquant_infer.tar --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_PP-OCRv2_rec_klquant_infer.tar --no-check-certificate
+        cd ./inference && tar xf ch_PP-OCRv2_det_klquant_infer.tar && tar xf ch_PP-OCRv2_rec_klquant_infer.tar && cd ../
+    elif [ ${model_name} == "ch_PP-OCRv3_det_KL" ] || [ ${model_name} == "PP-OCRv3_mobile_rec_KL" ] ; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_PP-OCRv3_det_klquant_infer.tar --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/PP-OCRv3_mobile_rec_klquant_infer.tar --no-check-certificate
+        cd ./inference && tar xf ch_PP-OCRv3_det_klquant_infer.tar && tar xf PP-OCRv3_mobile_rec_klquant_infer.tar && cd ../
+    elif [ ${model_name} == "ch_ppocr_mobile_v2_0_det_PACT" ] || [ ${model_name} == "ch_ppocr_mobile_v2.0_rec_PACT" ] ; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_ppocr_mobile_v2.0_det_pact_infer.tar --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_ppocr_mobile_v2.0_rec_pact_infer.tar --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_det_pact_infer.tar && tar xf ch_ppocr_mobile_v2.0_rec_pact_infer.tar && cd ../
+    elif [ ${model_name} == "ch_PP-OCRv2_det_PACT" ] || [ ${model_name} == "ch_PP-OCRv2_rec_PACT" ] ; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_PP-OCRv2_det_pact_infer.tar --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_PP-OCRv2_rec_pact_infer.tar --no-check-certificate
+        cd ./inference && tar xf ch_PP-OCRv2_det_pact_infer.tar && tar xf ch_PP-OCRv2_rec_pact_infer.tar && cd ../
+    elif [ ${model_name} == "ch_PP-OCRv3_det_PACT" ] || [ ${model_name} == "PP-OCRv3_mobile_rec_PACT" ] ; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/ch_PP-OCRv3_det_pact_infer.tar --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/tipc_fake_model/PP-OCRv3_mobile_rec_pact_infer.tar --no-check-certificate
+        cd ./inference && tar xf ch_PP-OCRv3_det_pact_infer.tar && tar xf PP-OCRv3_mobile_rec_pact_infer.tar && cd ../
+    elif [[ ${model_name} =~ "ch_ppocr_mobile_v2_0" ]]; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_det_infer.tar && tar xf ch_ppocr_mobile_v2.0_rec_infer.tar && cd ../
+    elif [[ ${model_name} =~ "ch_ppocr_server_v2_0" ]]; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_server_v2.0_det_infer.tar && tar xf ch_ppocr_server_v2.0_rec_infer.tar && cd ../
+    elif [[ ${model_name} =~ "ch_PP-OCRv2" ]]; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer.tar --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar --no-check-certificate
+        cd ./inference && tar xf ch_PP-OCRv2_det_infer.tar && tar xf ch_PP-OCRv2_rec_infer.tar && cd ../
+    elif [[ ${model_name} =~ "ch_PP-OCRv3" ]]; then
+        wget -nc  -P ./inference https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar --no-check-certificate
+        wget -nc  -P ./inference https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar --no-check-certificate
+        cd ./inference && tar xf PP-OCRv3_mobile_det_infer.tar && tar xf PP-OCRv3_mobile_rec_infer.tar && cd ../
+    fi
+    # wget data
+    wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar  --no-check-certificate
+    wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar  --no-check-certificate
+    cd ./inference && tar xf ch_det_data_50.tar && tar xf rec_inference.tar && cd ../
+fi
+
+if [ ${MODE} = "paddle2onnx_infer" ];then
+    # prepare serving env
+    python_name=$(func_parser_value "${lines[2]}")
+    ${python_name} -m pip install paddle2onnx onnxruntime onnx
+    # wget model
+    if [[ ${model_name} =~ "ch_ppocr_mobile_v2_0" ]]; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_det_infer.tar --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_infer.tar --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_mobile_v2.0_det_infer.tar && tar xf ch_ppocr_mobile_v2.0_rec_infer.tar && cd ../
+    elif [[ ${model_name} =~ "ch_ppocr_server_v2_0" ]]; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar --no-check-certificate
+        cd ./inference && tar xf ch_ppocr_server_v2.0_det_infer.tar && tar xf ch_ppocr_server_v2.0_rec_infer.tar && cd ../
+    elif [[ ${model_name} =~ "ch_PP-OCRv2" ]]; then
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_det_infer.tar --no-check-certificate
+        wget -nc  -P ./inference https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/ch_PP-OCRv2_rec_infer.tar --no-check-certificate
+        cd ./inference && tar xf ch_PP-OCRv2_det_infer.tar && tar xf ch_PP-OCRv2_rec_infer.tar && cd ../
+    elif [[ ${model_name} =~ "ch_PP-OCRv3" ]]; then
+        wget -nc  -P ./inference https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_det_infer.tar --no-check-certificate
+        wget -nc  -P ./inference https://paddle-model-ecology.bj.bcebos.com/paddlex/official_inference_model/paddle3.0.0/PP-OCRv3_mobile_rec_infer.tar --no-check-certificate
+        cd ./inference && tar xf PP-OCRv3_mobile_det_infer.tar && tar xf PP-OCRv3_mobile_rec_infer.tar && cd ../
+    elif [[ ${model_name} =~ "slanet" ]];then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/ppstructure/models/slanet/paddle3.0b2/ch_ppstructure_mobile_v2.0_SLANet_infer.tar --no-check-certificate
+        cd ./inference/ && tar xf ch_ppstructure_mobile_v2.0_SLANet_infer.tar && cd ../
+    elif [[ ${model_name} =~ "en_table_structure" ]];then
+        wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/table/en_ppocr_mobile_v2.0_table_structure_infer.tar --no-check-certificate
+        cd ./inference/ && tar xf en_ppocr_mobile_v2.0_table_structure_infer.tar && cd ../
+    fi
+    
+    # wget data
+    wget -nc -P ./inference https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/ch_det_data_50.tar
+    wget -nc -P ./inference/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/rec_inference.tar
+    cd ./inference && tar xf ch_det_data_50.tar && tar xf rec_inference.tar && cd ../
+    
+fi
diff --git a/test_tipc/prepare_lite_cpp.sh b/test_tipc/prepare_lite_cpp.sh
new file mode 100644
index 0000000..0d3a5ca
--- /dev/null
+++ b/test_tipc/prepare_lite_cpp.sh
@@ -0,0 +1,109 @@
+#!/bin/bash
+source ./test_tipc/common_func.sh
+FILENAME=$1
+dataline=$(cat ${FILENAME})
+# parser params
+IFS=$'\n'
+lines=(${dataline})
+IFS=$'\n'
+paddlelite_library_source=$2
+
+inference_cmd=$(func_parser_value "${lines[1]}")
+DEVICE=$(func_parser_value "${lines[2]}")
+det_lite_model_list=$(func_parser_value "${lines[3]}")
+rec_lite_model_list=$(func_parser_value "${lines[4]}")
+cls_lite_model_list=$(func_parser_value "${lines[5]}")
+
+if [[ $inference_cmd =~ "det" ]]; then
+    lite_model_list=${det_lite_model_list}
+elif [[ $inference_cmd =~ "rec" ]]; then
+    lite_model_list=(${rec_lite_model_list[*]} ${cls_lite_model_list[*]})
+elif [[ $inference_cmd =~ "system" ]]; then
+    lite_model_list=(${det_lite_model_list[*]} ${rec_lite_model_list[*]} ${cls_lite_model_list[*]})
+else
+    echo "inference_cmd is wrong, please check."
+    exit 1
+fi
+
+if [ ${DEVICE} = "ARM_CPU" ]; then
+    valid_targets="arm"
+    paddlelite_library_url="https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10-rc/inference_lite_lib.android.armv8.gcc.c++_shared.with_extra.with_cv.tar.gz"
+    end_index="66"
+    compile_with_opencl="OFF"
+elif [ ${DEVICE} = "ARM_GPU_OPENCL" ]; then
+    valid_targets="opencl"
+    paddlelite_library_url="https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10-rc/inference_lite_lib.armv8.clang.with_exception.with_extra.with_cv.opencl.tar.gz"
+    end_index="71"
+    compile_with_opencl="ON"
+else
+    echo "DEVICE only support ARM_CPU, ARM_GPU_OPENCL."
+    exit 2    
+fi
+
+# prepare paddlelite model
+pip install paddlelite==2.10-rc
+current_dir=${PWD}
+IFS="|"
+model_path=./inference_models
+
+for model in ${lite_model_list[*]}; do
+    if [[ $model =~ "PP-OCRv2" ]]; then
+        inference_model_url=https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/${model}.tar
+    elif [[ $model =~ "v2_0" ]]; then
+        inference_model_url=https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/${model}.tar
+    elif [[ $model =~ "PP-OCRv3" ]]; then
+        inference_model_url=https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/${model}.tar
+    else 
+        echo "Model is wrong, please check."
+        exit 3
+    fi
+    inference_model=${inference_model_url##*/}
+    wget -nc  -P ${model_path} ${inference_model_url}
+    cd ${model_path} && tar -xf ${inference_model} && cd ../
+    model_dir=${model_path}/${inference_model%.*}
+    model_file=${model_dir}/inference.pdmodel
+    param_file=${model_dir}/inference.pdiparams
+    paddle_lite_opt --model_dir=${model_dir} --model_file=${model_file} --param_file=${param_file} --valid_targets=${valid_targets} --optimize_out=${model_dir}_opt
+done
+
+# prepare test data
+data_url=https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/icdar2015_lite.tar
+data_file=${data_url##*/}
+wget -nc  -P ./test_data ${data_url}
+cd ./test_data && tar -xf ${data_file} && rm ${data_file} && cd ../
+
+# prepare paddlelite predict library
+if [[ ${paddlelite_library_source} = "download" ]]; then
+    paddlelite_library_zipfile=$(echo $paddlelite_library_url | awk -F "/" '{print $NF}')
+    paddlelite_library_file=${paddlelite_library_zipfile:0:${end_index}}
+    wget ${paddlelite_library_url} && tar -xf ${paddlelite_library_zipfile}
+    cd ${paddlelite_library_zipfile}
+elif [[ ${paddlelite_library_source} = "compile" ]]; then
+    git clone -b release/v2.10 https://github.com/PaddlePaddle/Paddle-Lite.git
+    cd Paddle-Lite
+    ./lite/tools/build_android.sh  --arch=armv8  --with_cv=ON --with_extra=ON --toolchain=clang --with_opencl=${compile_with_opencl}
+    cd ../
+    cp -r Paddle-Lite/build.lite.android.armv8.clang/inference_lite_lib.android.armv8/ .
+    paddlelite_library_file=inference_lite_lib.android.armv8
+else
+    echo "paddlelite_library_source only support 'download' and 'compile'"
+    exit 3
+fi
+
+# organize the required files  
+mkdir -p  ${paddlelite_library_file}/demo/cxx/ocr/test_lite
+cp -r ${model_path}/*_opt.nb test_data ${paddlelite_library_file}/demo/cxx/ocr/test_lite
+cp ppocr/utils/ppocr_keys_v1.txt deploy/lite/config.txt ${paddlelite_library_file}/demo/cxx/ocr/test_lite
+cp -r ./deploy/lite/* ${paddlelite_library_file}/demo/cxx/ocr/
+cp ${paddlelite_library_file}/cxx/lib/libpaddle_light_api_shared.so ${paddlelite_library_file}/demo/cxx/ocr/test_lite
+cp ${FILENAME} test_tipc/test_lite_arm_cpp.sh test_tipc/common_func.sh ${paddlelite_library_file}/demo/cxx/ocr/test_lite
+cd ${paddlelite_library_file}/demo/cxx/ocr/
+git clone https://github.com/cuicheng01/AutoLog.git
+
+# compile and do some postprocess
+make -j
+sleep 1
+make -j
+cp ocr_db_crnn test_lite && cp test_lite/libpaddle_light_api_shared.so test_lite/libc++_shared.so
+tar -cf test_lite.tar ./test_lite && cp test_lite.tar ${current_dir} && cd ${current_dir}
+rm -rf ${paddlelite_library_file}* && rm -rf ${model_path}
diff --git a/test_tipc/readme.md b/test_tipc/readme.md
new file mode 100644
index 0000000..540d9d9
--- /dev/null
+++ b/test_tipc/readme.md
@@ -0,0 +1,148 @@
+
+# 飞桨训推一体全流程（TIPC）
+
+## 1. 简介
+
+飞桨除了基本的模型训练和预测，还提供了支持多端多平台的高性能推理部署工具。本文档提供了PaddleOCR中所有模型的飞桨训推一体全流程（Training and Inference Pipeline Criterion(TIPC)）信息和测试工具，方便用户查阅每种模型的训练推理部署打通情况，并可以进行一键测试。
+
+<div align="center">
+    <img src="docs/guide.png" width="1000">
+</div>
+
+## 2. 汇总信息
+
+打通情况汇总如下，已填写的部分表示可以使用本工具进行一键测试，未填写的表示正在支持中。
+
+**字段说明：**
+- 基础训练预测：包括模型训练、Paddle Inference Python预测。
+- 更多训练方式：包括多机多卡、混合精度。
+- 模型压缩：包括裁剪、离线/在线量化、蒸馏。
+- 其他预测部署：包括Paddle Inference C++预测、Paddle Serving部署、Paddle-Lite部署等。
+
+更详细的mkldnn、Tensorrt等预测加速相关功能的支持情况可以查看各测试工具的[更多教程](#more)。
+
+| 算法论文 | 模型名称 | 模型类型 | 基础<br>训练预测 | 更多<br>训练方式 | 模型压缩 |  其他预测部署  |
+| :--- | :--- |  :----:  | :--------: |  :----  |   :----  |   :----  |
+| DB     |ch_ppocr_mobile_v2.0_det | 检测  | 支持 | 多机多卡 <br> 混合精度 | - | Paddle Inference: C++ <br> Paddle Serving: Python, C++ <br> Paddle-Lite: <br> (1) ARM CPU(C++) |
+| DB     |ch_ppocr_mobile_v2.0_det_FPGM | 检测  | 支持 | 多机多卡 <br> 混合精度 | FPGM裁剪 | Paddle Inference: C++ <br> Paddle Serving: Python, C++ <br> Paddle-Lite: <br> (1) ARM CPU(C++) |
+| DB     |ch_ppocr_mobile_v2.0_det_PACT | 检测  | 支持 | 多机多卡 <br> 混合精度 | PACT量化 | Paddle Inference: C++ <br> Paddle Serving: Python, C++ <br> Paddle-Lite: <br> (1) ARM CPU(C++) |
+| DB     |ch_ppocr_mobile_v2.0_det_KL | 检测  | 支持 | 多机多卡 <br> 混合精度 | 离线量化| Paddle Inference: C++ <br> Paddle Serving: Python, C++ <br> Paddle-Lite: <br> (1) ARM CPU(C++) |
+| DB     |ch_ppocr_server_v2.0_det | 检测  | 支持 | 多机多卡 <br> 混合精度 | - | Paddle Inference: C++ <br> Paddle Serving: Python, C++ |
+| DB     |ch_PP-OCRv2_det          | 检测  | 支持 | 多机多卡 <br> 混合精度 | - | Paddle Inference: C++ <br> Paddle Serving: Python, C++ |
+| CRNN   |ch_ppocr_mobile_v2.0_rec | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | Paddle Inference: C++ <br> Paddle Serving: Python, C++ <br> Paddle-Lite: <br> (1) ARM CPU(C++) |
+| CRNN   |ch_ppocr_server_v2.0_rec | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | Paddle Inference: C++ <br> Paddle Serving: Python, C++ |
+| CRNN   |ch_PP-OCRv2_rec          | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | Paddle Inference: C++ <br> Paddle Serving: Python, C++ |
+| PP-OCR |ch_ppocr_mobile_v2.0 | 检测+识别  | 支持 | 多机多卡 <br> 混合精度 | - | Paddle Inference: C++ <br> Paddle Serving: Python, C++ <br> Paddle-Lite: <br> (1) ARM CPU(C++) |
+| PP-OCR |ch_ppocr_server_v2.0 | 检测+识别  | 支持 | 多机多卡 <br> 混合精度 | - | Paddle Inference: C++ <br> Paddle Serving: Python, C++ |
+|PP-OCRv2|ch_PP-OCRv2 | 检测+识别  | 支持 | 多机多卡 <br> 混合精度 | - | Paddle Inference: C++ <br> Paddle Serving: Python, C++ |
+| DB     |det_mv3_db_v2.0                | 检测  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| DB     |det_r50_vd_db_v2.0             | 检测  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| EAST   |det_mv3_east_v2.0              | 检测  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| EAST   |det_r50_vd_east_v2.0           | 检测  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| PSENet |det_mv3_pse_v2.0               | 检测  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| PSENet |det_r50_vd_pse_v2.0            | 检测  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| SAST   |det_r50_vd_sast_totaltext_v2.0 | 检测  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| Rosetta|rec_mv3_none_none_ctc_v2.0     | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| Rosetta|rec_r34_vd_none_none_ctc_v2.0  | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| CAN    |rec_d28_can                    | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| CRNN   |rec_mv3_none_bilstm_ctc_v2.0   | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| CRNN   |rec_r34_vd_none_bilstm_ctc_v2.0| 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| StarNet|rec_mv3_tps_bilstm_ctc_v2.0    | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| StarNet|rec_r34_vd_tps_bilstm_ctc_v2.0 | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| RARE   |rec_mv3_tps_bilstm_att_v2.0    | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| RARE   |rec_r34_vd_tps_bilstm_att_v2.0 | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| SRN    |rec_r50fpn_vd_none_srn         | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| NRTR   |rec_mtb_nrtr                   | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| SAR    |rec_r31_sar                    | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| SPIN   |rec_r32_gaspin_bilstm_att      | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| RobustScanner |rec_r31_robustscanner   | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
+| PGNet  |rec_r34_vd_none_none_ctc_v2.0  | 端到端| 支持 | 多机多卡 <br> 混合精度 | - | - |
+| TableMaster  |table_structure_tablemaster_train  | 表格识别| 支持 | 多机多卡 <br> 混合精度 | - | - |
+
+
+
+## 3. 测试工具简介
+### 目录介绍
+
+```shell
+test_tipc/
+├── configs/  # 配置文件目录
+    ├── ch_ppocr_mobile_v2.0_det    # ch_ppocr_mobile_v2.0_det模型的测试配置文件目录
+        ├── train_infer_python.txt      # 测试Linux上python训练预测（基础训练预测）的配置文件
+        ├── model_linux_gpu_normal_normal_infer_cpp_linux_gpu_cpu.txt     # 测试Linux上c++预测的配置文件
+        ├── train_linux_gpu_fleet_amp_infer_python_linux_gpu_cpu.txt      # 测试Linux上多机多卡、混合精度训练和python预测的配置文件
+        ├── ...
+    ├── ch_ppocr_server_v2.0_det               # ch_ppocr_server_v2.0_det模型的测试配置文件目录
+        ├── ...
+    ├── ch_ppocr_mobile_v2.0_rec               # ch_ppocr_mobile_v2.0_rec模型的测试配置文件目录
+        ├── ...
+    ├── ch_ppocr_server_v2.0_det               # ch_ppocr_server_v2.0_det模型的测试配置文件目录
+        ├── ...
+    ├── ...
+├── results/   # 预先保存的预测结果，用于和实际预测结果进行精读比对
+    ├── python_ppocr_det_mobile_results_fp32.txt           # 预存的mobile版ppocr检测模型python预测fp32精度的结果
+    ├── python_ppocr_det_mobile_results_fp16.txt           # 预存的mobile版ppocr检测模型python预测fp16精度的结果
+    ├── cpp_ppocr_det_mobile_results_fp32.txt       # 预存的mobile版ppocr检测模型c++预测的fp32精度的结果
+    ├── cpp_ppocr_det_mobile_results_fp16.txt       # 预存的mobile版ppocr检测模型c++预测的fp16精度的结果
+    ├── ...
+├── prepare.sh                        # 完成test_*.sh运行所需要的数据和模型下载
+├── test_train_inference_python.sh    # 测试python训练预测的主程序
+├── test_inference_cpp.sh             # 测试c++预测的主程序
+├── test_serving.sh                   # 测试serving部署预测的主程序
+├── test_lite_arm_cpu_cpp.sh          # 测试lite在arm_cpu上部署的C++预测的主程序
+├── compare_results.py                # 用于对比log中的预测结果与results中的预存结果精度误差是否在限定范围内
+└── readme.md                         # 使用文档
+```
+
+### 测试流程概述
+
+使用本工具，可以测试不同功能的支持情况，以及预测结果是否对齐，测试流程概括如下：
+<div align="center">
+    <img src="docs/test.png" width="800">
+</div>
+
+1. 运行prepare.sh准备测试所需数据和模型；
+2. 运行要测试的功能对应的测试脚本`test_*.sh`，产出log，由log可以看到不同配置是否运行成功；
+3. 用`compare_results.py`对比log中的预测结果和预存在results目录下的结果，判断预测精度是否符合预期（在误差范围内）。
+
+测试单项功能仅需两行命令，**如需测试不同模型/功能，替换配置文件即可**，命令格式如下：
+```shell
+# 功能：准备数据
+# 格式：bash + 运行脚本 + 参数1: 配置文件选择 + 参数2: 模式选择
+bash test_tipc/prepare.sh  configs/[model_name]/[params_file_name]  [Mode]
+
+# 功能：运行测试
+# 格式：bash + 运行脚本 + 参数1: 配置文件选择 + 参数2: 模式选择
+bash test_tipc/test_train_inference_python.sh configs/[model_name]/[params_file_name]  [Mode]
+```
+
+例如，测试基本训练预测功能的`lite_train_lite_infer`模式，运行：
+```shell
+# 准备数据
+bash test_tipc/prepare.sh ./test_tipc/configs/ch_ppocr_mobile_v2.0_det/train_infer_python.txt 'lite_train_lite_infer'
+# 运行测试
+bash test_tipc/test_train_inference_python.sh ./test_tipc/configs/ch_ppocr_mobile_v2.0_det/train_infer_python.txt 'lite_train_lite_infer'
+```
+关于本示例命令的更多信息可查看[基础训练预测使用文档](https://github.com/PaddlePaddle/PaddleOCR/blob/dygraph/test_tipc/docs/test_train_inference_python.md#22-%E5%8A%9F%E8%83%BD%E6%B5%8B%E8%AF%95)。
+
+### 配置文件命名规范
+在`configs`目录下，**按模型名称划分为子目录**，子目录中存放所有该模型测试需要用到的配置文件，配置文件的命名遵循如下规范：
+
+1. 基础训练预测配置简单命名为：`train_infer_python.txt`，表示**Linux环境下单机、不使用混合精度训练+python预测**，其完整命名对应`train_linux_gpu_normal_normal_infer_python_linux_gpu_cpu.txt`，由于本配置文件使用频率较高，这里进行了名称简化。
+
+2. 其他带训练配置命名格式为：`train_训练硬件环境(linux_gpu/linux_dcu/…)_是否多机(fleet/normal)_是否混合精度(amp/normal)_预测模式(infer/lite/serving)_语言(cpp/python/java)_预测硬件环境(linux_gpu/mac/opencl_arm_gpu/...).txt`。如，linux gpu下多机多卡+混合精度链条测试对应配置 `train_linux_gpu_fleet_amp_infer_python_linux_gpu_cpu.txt`，linux dcu下基础训练预测对应配置 `train_linux_dcu_normal_normal_infer_python_linux_dcu.txt`。
+
+3. 仅预测的配置（如serving、lite等）命名格式：`model_训练硬件环境(linux_gpu/linux_dcu/…)_是否多机(fleet/normal)_是否混合精度(amp/normal)_(infer/lite/serving)_语言(cpp/python/java)_预测硬件环境(linux_gpu/mac/opencl_arm_gpu/...).txt`，即，与2相比，仅第一个字段从train换为model，测试时模型直接下载获取，这里的“训练硬件环境”表示所测试的模型是在哪种环境下训练得到的。
+
+**根据上述命名规范，可以直接从子目录名称和配置文件名找到需要测试的场景和功能对应的配置文件。**
+
+
+<a name="more"></a>
+## 4. 开始测试
+各功能测试中涉及混合精度、裁剪、量化等训练相关，及mkldnn、Tensorrt等多种预测相关参数配置，请点击下方相应链接了解更多细节和使用教程：
+- [test_train_inference_python 使用](docs/test_train_inference_python.md) ：测试基于Python的模型训练、评估、推理等基本功能，包括裁剪、量化、蒸馏。
+- [test_train_fleet_inference_python 使用](./docs/test_train_fleet_inference_python.md)：测试基于Python的多机多卡训练与推理等基本功能。
+- [test_inference_cpp 使用](docs/test_inference_cpp.md)：测试基于C++的模型推理。
+- [test_serving 使用](docs/test_serving.md)：测试基于Paddle Serving的服务化部署功能。
+- [test_lite_arm_cpp 使用](docs/test_lite_arm_cpp.md)：测试基于Paddle-Lite的ARM CPU端c++预测部署功能。
+- [test_paddle2onnx 使用](docs/test_paddle2onnx.md)：测试Paddle2ONNX的模型转化功能，并验证正确性。
diff --git a/test_tipc/results/cpp_ppocr_det_mobile_results_fp16.txt b/test_tipc/results/cpp_ppocr_det_mobile_results_fp16.txt
new file mode 100644
index 0000000..34cde25
--- /dev/null
+++ b/test_tipc/results/cpp_ppocr_det_mobile_results_fp16.txt
@@ -0,0 +1,50 @@
+../../inference/ch_det_data_50/all-sum-510/00008790.jpg	208 404 282 404 282 421 208 421 58 396 107 396 107 413 58 413 197 387 296 387 296 403 197 403 161 389 174 389 174 402 161 402 34 378 134 378 134 394 34 394 323 377 329 377 329 382 323 382 199 370 292 370 292 383 199 383 216 309 274 309 274 325 216 325 161 304 173 304 173 315 161 315 370 301 437 301 437 317 370 317 30 301 135 300 135 316 30 317 221 291 270 291 270 308 221 308 58 224 106 224 106 238 58 238 216 222 274 222 274 239 216 239 161 217 174 217 174 229 161 229 33 205 133 205 133 221 33 221 221 204 270 204 270 221 221 221 73 145 385 145 385 162 73 162 52 119 119 119 119 135 52 135 72 50 296 50 296 66 72 66 54 15 118 15 118 32 54 32 
+../../inference/ch_det_data_50/all-sum-510/00018946.jpg	439 327 476 327 476 341 439 341 85 284 142 284 142 308 85 308 300 278 380 278 380 299 300 299 195 262 287 275 284 299 192 286 196 196 454 218 452 244 194 222 343 182 376 182 376 193 343 193 198 162 341 169 340 195 197 188 176 130 381 145 380 165 175 150 176 100 417 118 415 148 174 130 
+../../inference/ch_det_data_50/all-sum-510/00034387.jpg	263 459 741 459 741 485 263 485 346 415 421 415 421 444 346 444 544 418 568 418 568 442 544 442 684 415 712 415 712 444 684 444 173 413 228 413 228 444 173 444 872 412 910 412 910 447 872 447 55 415 76 415 76 443 55 443 855 371 927 371 927 401 855 401 347 371 420 371 420 400 347 400 672 370 725 370 725 402 672 402 537 371 571 371 571 401 537 401 136 364 230 367 229 403 135 400 55 370 76 370 76 399 55 399 856 328 927 328 927 358 856 358 350 328 420 328 420 358 350 358 672 326 725 326 725 358 672 358 539 327 571 327 571 359 539 359 170 326 229 323 231 357 171 359 56 328 76 328 76 358 56 358 297 326 316 326 316 334 297 334 854 284 927 284 927 314 854 314 672 284 725 284 725 315 672 315 344 284 431 282 432 315 345 317 537 283 570 283 570 314 537 314 170 281 228 281 228 315 170 315 55 285 75 285 75 314 55 314 856 241 927 241 927 270 856 270 346 240 464 240 464 271 346 271 154 241 228 241 228 271 154 271 672 240 726 240 726 271 672 271 530 240 573 240 573 272 530 272 55 241 76 241 76 270 55 270 854 196 927 198 926 228 853 225 672 197 728 197 728 228 672 228 342 199 439 194 441 224 344 230 175 196 229 196 229 226 175 226 55 199 75 199 75 228 55 228 526 193 578 193 578 228 526 228 347 154 420 154 420 182 347 182 853 153 927 153 927 181 853 181 175 153 228 153 228 184 175 184 668 152 725 152 725 182 668 182 536 153 572 153 572 183 536 183 55 155 76 155 76 183 55 183 347 109 420 109 420 138 347 138 172 109 229 109 229 140 172 140 544 111 565 111 565 138 544 138 51 110 77 110 77 140 51 140 639 105 729 105 729 141 639 141 815 101 929 109 927 141 813 133 812 65 953 65 953 93 812 93 305 64 447 66 447 94 305 92 671 65 725 65 725 95 671 95 173 64 229 66 228 96 172 94 37 64 91 66 90 98 36 96 527 63 581 63 581 95 527 95 333 18 671 18 671 45 333 45 
+../../inference/ch_det_data_50/all-sum-510/00037951.jpg	432 973 552 977 552 994 432 991 431 931 554 931 554 970 431 970 29 520 101 520 101 546 29 546 29 441 146 441 146 465 29 465 233 333 328 331 328 356 233 358 121 250 439 250 439 287 121 287 180 205 380 205 380 229 180 229 255 104 323 121 307 184 239 166 35 57 147 57 147 82 35 82 
+../../inference/ch_det_data_50/all-sum-510/00044782.jpg	222 214 247 214 247 230 222 230 162 214 183 214 183 231 162 231 122 190 216 190 216 203 122 203 90 82 252 82 252 100 90 100 70 61 279 61 279 78 70 78 103 14 244 14 244 46 103 46 
+../../inference/ch_det_data_50/all-sum-510/00067516.jpg	139 806 596 807 596 824 139 823 46 782 699 782 699 800 46 800 577 749 669 749 669 766 577 766 353 748 397 748 397 769 353 769 220 749 261 749 261 767 220 767 475 748 502 748 502 769 475 769 68 746 134 749 133 766 67 763 574 680 670 680 670 700 574 700 474 680 519 680 519 701 474 701 352 680 397 680 397 701 352 701 68 679 134 682 133 700 67 697 219 678 245 681 242 702 216 698 575 614 669 614 669 633 575 633 66 612 135 614 135 633 66 631 474 613 501 613 501 633 474 633 353 613 379 613 379 634 353 634 219 612 245 612 245 633 219 633 576 546 669 546 669 566 576 566 474 545 519 545 519 566 474 566 351 544 381 544 381 567 351 567 219 545 245 545 245 566 219 566 67 541 134 544 133 565 66 562 67 477 134 480 133 501 66 498 584 479 666 479 666 499 584 499 474 478 519 478 519 500 474 500 352 478 397 478 397 500 352 500 218 477 246 477 246 502 218 502 579 424 666 427 665 451 578 448 344 428 410 428 410 449 344 449 66 425 151 427 151 451 66 449 473 427 515 427 515 450 473 450 218 427 259 427 259 450 218 450 282 396 479 397 479 420 282 419 83 316 667 316 667 335 83 335 64 277 666 277 666 292 64 292 456 209 585 209 585 226 456 226 311 208 373 208 373 227 311 227 163 208 227 208 227 227 163 227 504 150 541 150 541 168 504 168 264 47 485 47 485 69 264 69 
+../../inference/ch_det_data_50/all-sum-510/00088568.jpg	57 443 119 443 119 456 57 456 309 413 744 413 744 430 309 430 309 375 737 375 737 392 309 392 415 337 559 337 559 351 415 351 307 322 674 321 674 338 307 339 275 292 348 294 348 313 275 311 52 285 210 285 210 301 52 301 273 262 421 262 421 279 273 279 55 262 249 262 249 279 55 279 669 247 697 247 697 262 669 262 601 247 629 247 629 262 601 262 531 247 559 247 559 262 531 262 461 247 489 247 489 262 461 262 277 247 310 247 310 261 277 261 55 240 142 240 142 254 55 254 276 230 400 230 400 246 276 246 741 227 749 237 741 246 732 237 665 230 701 230 701 245 665 245 598 230 631 230 631 245 598 245 527 230 563 230 563 245 527 245 458 230 493 230 493 245 458 245 52 213 212 215 212 233 52 231 732 214 747 214 747 227 732 227 662 212 706 212 706 230 662 230 594 213 638 213 638 227 594 227 522 213 570 213 570 227 522 227 453 213 497 213 497 227 453 227 278 213 352 213 352 227 278 227 734 198 748 198 748 210 734 210 667 196 702 196 702 210 667 210 599 196 633 196 633 211 599 211 527 196 564 196 564 210 527 210 459 196 493 196 493 210 459 210 276 194 418 195 418 212 276 211 54 190 241 190 241 207 54 207 664 179 705 179 705 194 664 194 278 178 352 180 352 195 278 193 733 179 747 179 747 194 733 194 596 178 635 178 635 193 596 193 523 177 567 177 567 195 523 195 456 178 495 178 495 193 456 193 55 170 142 170 142 184 55 184 733 164 748 164 748 176 733 176 664 162 705 162 705 176 664 176 597 162 635 162 635 176 597 176 525 162 566 162 566 176 525 176 456 162 494 162 494 176 456 176 277 160 399 160 399 176 277 176 54 146 149 146 149 161 54 161 452 145 497 145 497 160 452 160 729 144 748 144 748 162 729 162 662 143 706 143 706 161 662 161 595 144 636 144 636 159 595 159 521 143 566 141 567 159 522 161 277 143 310 143 310 159 277 159 275 120 430 120 430 140 275 140 50 119 234 120 234 140 50 139 402 90 703 90 703 107 402 107 46 78 282 78 282 98 46 98 324 67 745 68 745 86 324 85 667 47 744 47 744 64 667 64 295 47 435 47 435 63 295 63 64 30 232 27 233 65 65 68 
+../../inference/ch_det_data_50/all-sum-510/00091741.jpg	46 335 87 335 87 360 46 360 98 209 258 209 258 232 98 232 101 189 258 190 258 206 101 205 87 99 268 97 269 184 88 186 92 45 266 53 263 117 89 109 89 10 258 12 258 38 89 36 
+../../inference/ch_det_data_50/all-sum-510/00105313.jpg	289 261 407 261 407 277 289 277 152 260 265 260 265 276 152 276 10 257 74 259 74 276 10 274 32 230 134 230 134 245 32 245 34 215 218 215 218 228 34 228 32 199 148 199 148 214 32 214 31 181 217 182 217 199 31 198 34 169 107 169 107 182 34 182 34 153 126 153 126 166 34 166 33 136 144 137 144 150 33 149 34 122 177 122 177 135 34 135 32 104 178 104 178 120 32 120 32 91 102 91 102 104 32 104 33 75 121 75 121 88 33 88 32 60 121 60 121 73 32 73 34 44 121 44 121 57 34 57 31 28 144 28 144 43 31 43 177 20 415 15 416 51 178 56 24 10 152 10 152 26 24 26 
+../../inference/ch_det_data_50/all-sum-510/00134770.jpg	386 645 457 645 457 658 386 658 406 618 486 616 486 634 406 636 111 533 272 530 272 550 111 553 110 501 445 496 445 516 110 521 110 469 445 465 445 485 110 489 110 438 446 433 446 453 110 458 109 407 445 403 445 423 109 427 151 375 443 372 443 392 151 395 183 336 371 334 371 358 183 360 73 96 517 101 516 220 72 215 
+../../inference/ch_det_data_50/all-sum-510/00145943.jpg	390 243 751 274 735 454 375 423 88 90 302 90 302 121 88 121 43 40 329 37 329 78 43 81 
+../../inference/ch_det_data_50/all-sum-510/00147605.jpg	800 613 878 613 878 627 800 627 514 605 786 604 786 629 514 630 116 521 226 521 226 561 116 561 252 522 309 522 309 558 252 558 713 500 902 503 902 539 713 536 254 501 296 501 296 519 254 519 345 479 475 479 475 517 345 517 251 483 296 483 296 501 251 501 350 456 447 456 447 471 350 471 143 442 203 442 203 469 143 469 727 370 880 370 880 422 727 422 526 369 684 369 684 421 526 421 140 367 490 367 490 423 140 423 742 313 872 313 872 338 742 338 798 155 888 155 888 192 798 192 272 140 457 140 457 161 272 161 737 114 895 118 894 158 736 155 107 110 206 110 206 131 107 131 268 92 464 94 464 134 268 131 
+../../inference/ch_det_data_50/all-sum-510/00150341.jpg	98 640 300 640 300 664 98 664 113 615 289 615 289 633 113 633 82 591 320 590 320 611 82 612 30 563 315 561 315 582 30 584 30 513 169 513 169 531 30 531 32 488 111 488 111 506 32 506 357 458 465 461 464 486 356 483 26 458 271 459 271 483 26 482 338 438 423 442 422 461 337 457 64 437 145 437 145 455 64 455 205 414 293 414 293 436 205 436 318 407 442 411 441 439 317 435 42 404 176 407 176 435 42 432 28 381 137 381 137 405 28 405 
+../../inference/ch_det_data_50/all-sum-510/00150669.jpg	647 698 683 698 683 718 647 718 515 684 551 684 551 721 515 721 650 687 680 687 680 702 650 702 920 673 938 673 938 686 920 686 518 670 548 670 548 690 518 690 785 670 808 670 808 688 785 688 590 670 608 670 608 688 590 688 732 665 745 679 732 692 718 679 652 668 680 668 680 689 652 689 271 665 423 665 423 690 271 690 45 666 110 666 110 688 45 688 130 664 205 664 205 690 130 690 781 628 812 628 812 663 781 663 643 626 687 626 687 666 643 666 514 627 550 627 550 665 514 665 654 617 673 617 673 629 654 629 521 617 541 617 541 629 521 629 858 617 868 617 868 628 858 628 727 617 736 617 736 628 727 628 920 614 940 614 940 631 920 631 785 614 807 614 807 631 785 631 371 603 421 603 421 620 371 620 83 600 216 603 216 624 83 620 46 602 72 602 72 623 46 623 780 569 817 573 813 610 776 606 922 559 936 559 936 575 922 575 856 559 869 559 869 575 856 575 61 552 411 552 411 569 61 569 61 531 117 533 117 547 61 545 859 527 868 527 868 539 859 539 923 525 936 525 936 542 923 542 787 524 807 524 807 540 787 540 526 526 536 526 536 536 526 536 261 511 396 511 396 528 261 528 120 512 246 512 246 526 120 526 47 512 120 512 120 527 47 527 753 491 829 491 829 508 753 508 636 491 712 491 712 508 636 508 517 491 593 491 593 508 517 508 84 448 125 448 125 463 84 463 221 448 238 448 238 462 221 462 682 444 869 444 869 461 682 461 561 444 667 444 667 461 561 461 489 445 545 445 545 459 489 459 183 437 209 437 209 459 183 459 52 429 73 437 64 464 42 456 222 430 278 430 278 445 222 445 86 430 145 430 145 445 86 445 505 382 617 381 617 398 505 399 701 380 758 380 758 398 701 398 307 371 365 371 365 386 307 386 90 371 168 371 168 386 90 386 686 334 821 334 821 352 686 352 496 333 659 333 659 350 496 350 207 314 245 314 245 333 207 333 497 287 642 287 642 304 497 304 670 286 804 286 804 304 670 304 668 239 817 239 817 257 668 257 495 239 644 239 644 257 495 257 668 193 816 193 816 209 668 209 496 192 644 192 644 208 496 208 668 144 816 144 816 161 668 161 497 144 646 144 646 161 497 161 488 102 546 102 546 121 488 121 845 21 900 21 900 43 845 43 25 18 702 18 702 39 25 39 896 10 997 14 996 46 895 42 
+../../inference/ch_det_data_50/all-sum-510/00152568.jpg	2 250 285 252 285 281 2 279 195 231 255 231 255 241 195 241 198 158 282 164 277 230 193 224 177 148 251 148 251 161 177 161 
+../../inference/ch_det_data_50/all-sum-510/00155628.jpg	147 898 506 901 506 925 147 922 519 892 562 894 561 912 518 910 59 884 83 884 83 895 59 895 148 877 505 881 505 902 148 897 523 833 641 837 640 858 522 854 68 832 187 834 187 855 68 853 245 554 468 554 468 570 245 570 307 506 405 508 405 526 307 523 243 481 460 483 460 504 243 502 250 420 460 422 460 454 250 452 193 377 518 379 518 410 193 408 473 194 625 194 625 212 473 212 70 127 643 129 643 163 70 161 478 39 599 35 602 101 481 105 67 23 136 14 140 44 71 54 
+../../inference/ch_det_data_50/all-sum-510/00173364.jpg	7 176 59 176 59 201 7 201 135 118 196 118 196 135 135 135 38 75 87 75 87 105 38 105 249 19 313 19 313 38 249 38 19 15 105 15 105 40 19 40 
+../../inference/ch_det_data_50/all-sum-510/00175503.jpg	39 256 503 252 504 362 40 366 49 198 351 175 357 253 55 276 
+../../inference/ch_det_data_50/all-sum-510/00193218.jpg	282 373 411 373 411 389 282 389 170 373 223 373 223 390 170 390 108 373 162 373 162 390 108 390 276 357 358 357 358 371 276 371 169 357 222 357 222 371 169 371 106 356 175 356 175 373 106 373 408 356 493 356 493 370 408 370 24 185 64 185 64 203 24 203 500 184 558 184 558 201 500 201 379 185 421 183 422 200 380 202 283 184 311 184 311 202 283 202 173 185 197 185 197 201 173 201 498 163 544 163 544 177 498 177 379 162 412 162 412 177 379 177 261 161 303 161 303 178 261 178 174 161 231 161 231 178 174 178 24 161 80 161 80 178 24 178 385 139 489 139 489 155 385 155 26 137 133 137 133 153 26 153 442 115 538 117 538 134 442 132 345 117 406 117 406 131 345 131 259 117 303 117 303 131 259 131 28 112 229 114 229 132 28 130 130 90 395 93 395 110 130 107 560 81 585 81 585 109 560 109 
+../../inference/ch_det_data_50/all-sum-510/00195033.jpg	221 302 240 302 240 309 221 309 487 262 534 264 533 282 486 280 125 249 194 249 194 285 125 285 336 248 364 248 364 268 336 268 317 221 381 223 381 240 317 238 431 224 450 224 450 236 431 236 360 202 539 202 539 218 360 218 87 199 148 201 148 218 87 216 371 181 450 181 450 195 371 195 327 180 354 180 354 194 327 194 94 178 241 178 241 195 94 195 431 159 559 159 559 175 431 175 128 148 289 149 289 166 128 165 35 145 75 148 74 163 34 160 487 146 501 146 501 153 487 153 100 143 122 143 122 154 100 154 370 127 505 126 505 140 370 141 98 125 194 125 194 139 98 139 320 125 338 125 338 136 320 136 35 121 78 121 78 135 35 135 322 104 338 104 338 116 322 116 371 101 503 101 503 117 371 117 348 103 362 103 362 115 348 115 37 101 81 101 81 114 37 114 97 98 207 99 207 116 97 115 305 89 317 89 317 97 305 97 346 86 364 86 364 97 346 97 319 85 342 85 342 100 319 100 357 82 515 80 515 96 357 98 40 81 90 81 90 94 40 94 92 77 242 78 242 95 92 94 312 65 394 65 394 79 312 79 240 64 290 64 290 78 240 78 183 52 222 52 222 66 183 66 468 47 547 47 547 61 468 61 422 34 438 34 438 55 422 55 464 29 551 29 551 43 464 43 206 19 330 21 330 42 206 40 
+../../inference/ch_det_data_50/all-sum-510/00208502.jpg	556 535 630 535 630 569 556 569 204 537 284 537 284 552 204 552 142 512 191 512 191 526 142 526 248 511 309 511 309 525 248 525 41 499 118 499 118 520 41 520 465 490 558 490 558 510 465 510 666 489 680 493 677 505 662 501 724 490 739 490 739 503 724 503 40 450 118 448 118 469 40 471 173 448 237 448 237 465 173 465 93 403 121 403 121 424 93 424 38 403 63 403 63 424 38 424 214 392 232 405 220 422 203 409 39 357 58 357 58 375 39 375 92 355 121 355 121 375 92 375 187 339 248 337 249 363 188 365 458 319 551 317 551 338 458 340 457 271 553 271 553 292 457 292 562 271 737 267 737 288 562 292 516 225 548 225 548 245 516 245 620 185 675 185 675 202 620 202 456 130 550 128 550 149 456 151 571 104 789 98 789 121 571 127 121 46 291 46 291 99 121 99 536 36 710 36 710 92 536 92 
+../../inference/ch_det_data_50/all-sum-510/00224225.jpg	135 426 157 426 157 449 135 449 199 402 480 408 479 461 198 455 200 225 474 225 474 394 200 394 130 264 174 264 174 281 130 281 343 205 458 205 458 232 343 232 197 186 349 194 346 242 194 234 7 41 160 39 161 115 8 117 
+../../inference/ch_det_data_50/all-sum-510/00227746.jpg	142 230 210 230 210 240 142 240 71 230 130 230 130 240 71 240 215 228 386 228 386 240 215 240 290 208 347 208 347 224 290 224 142 179 165 181 162 209 139 208 172 179 250 179 250 195 172 195 171 152 347 152 347 167 171 167 143 110 279 112 279 135 143 132 202 53 387 53 387 69 202 69 141 47 193 47 193 64 141 64 
+../../inference/ch_det_data_50/all-sum-510/00229605.jpg	742 528 882 528 882 545 742 545 232 497 590 496 590 524 232 525 5 496 229 496 229 524 5 524 733 494 884 497 884 522 733 519 605 493 718 488 719 517 606 522 2 242 865 227 866 291 3 305 477 26 884 26 884 77 477 77 
+../../inference/ch_det_data_50/all-sum-510/00233011.jpg	61 225 293 225 293 243 61 243 11 218 43 218 43 252 11 252 60 177 120 177 120 196 60 196 11 169 44 169 44 204 11 204 59 127 149 129 149 148 59 146 11 123 45 123 45 156 11 156 124 86 239 86 239 104 124 104 147 49 218 49 218 67 147 67 257 44 354 47 353 71 256 68 8 47 54 47 54 69 8 69 275 10 346 10 346 32 275 32 26 9 75 9 75 32 26 32 
+../../inference/ch_det_data_50/all-sum-510/00233625.jpg	370 395 635 397 635 445 370 443 67 210 935 204 936 325 68 331 
+../../inference/ch_det_data_50/all-sum-510/00233634.jpg	213 637 264 637 264 706 213 706 522 634 572 634 572 697 522 697 641 522 684 522 684 570 641 570 95 514 155 514 155 592 95 592 754 394 762 394 762 403 754 403 677 362 730 360 733 432 679 433 53 360 109 360 109 436 53 436 77 207 157 207 157 282 77 282 642 204 695 204 695 274 642 274 208 88 262 85 266 165 212 168 362 47 428 44 432 117 366 120 
+../../inference/ch_det_data_50/all-sum-510/00234400.jpg	156 419 739 419 739 439 156 439 157 393 653 393 653 412 157 412 38 390 129 390 129 413 38 413 156 339 307 342 307 365 156 362 36 342 125 342 125 363 36 363 519 293 705 293 705 316 519 316 393 290 485 288 485 316 393 318 156 291 271 291 271 315 156 315 35 291 127 291 127 315 35 315 155 242 360 242 360 269 155 269 34 242 83 242 83 270 34 270 27 150 159 150 159 177 27 177 280 96 507 96 507 113 280 113 313 44 477 47 476 90 312 87 516 50 664 52 664 68 516 67 485 17 708 15 708 45 485 47 
+../../inference/ch_det_data_50/all-sum-510/00234883.jpg	64 122 318 117 319 193 65 197 71 118 122 118 122 132 71 132 381 62 506 61 506 75 381 76 57 26 368 26 368 116 57 116 385 26 503 23 503 47 385 50 
+../../inference/ch_det_data_50/all-sum-510/test_add_0.jpg	311 521 391 521 391 534 311 534 277 499 426 499 426 516 277 516 259 445 438 445 438 461 259 461 210 426 487 426 487 443 210 443 244 385 460 385 460 411 244 411 220 327 476 327 476 373 220 373 205 204 494 208 493 279 204 275 264 163 423 165 423 198 264 196 15 17 203 15 203 45 15 47 
+../../inference/ch_det_data_50/all-sum-510/test_add_1.png	
+../../inference/ch_det_data_50/all-sum-510/test_add_10.png	155 123 187 123 187 174 155 174 160 105 184 105 184 131 160 131 116 45 155 44 158 176 119 176 63 30 102 31 99 172 60 171 
+../../inference/ch_det_data_50/all-sum-510/test_add_11.jpg	1388 755 1486 755 1486 794 1388 794 1011 752 1210 752 1210 802 1011 802 681 752 879 752 879 801 681 801 355 750 568 745 570 796 356 801 76 748 266 743 268 796 78 801 600 645 1155 645 1155 706 600 706 600 562 1151 553 1151 614 600 622 596 478 1070 470 1070 529 596 537 595 390 1095 385 1095 444 595 448 600 303 1061 303 1061 362 600 362 353 180 1521 180 1521 265 353 265 59 40 261 40 261 91 59 91 1303 39 1495 39 1495 90 1303 90 971 37 1173 32 1175 83 973 88 668 37 864 32 866 83 670 88 361 32 561 32 561 88 361 88 
+../../inference/ch_det_data_50/all-sum-510/test_add_12.jpg	9 590 140 592 140 615 9 613 107 520 908 524 908 571 107 566 632 448 905 445 905 481 632 484 110 445 468 447 468 487 110 485 580 303 682 301 683 351 581 353 368 257 568 262 565 361 364 355 61 83 856 85 856 164 61 162 
+../../inference/ch_det_data_50/all-sum-510/test_add_13.jpg	68 94 117 97 116 115 67 112 
+../../inference/ch_det_data_50/all-sum-510/test_add_14.jpg	28 94 238 92 238 130 28 132 27 50 241 48 241 88 27 90 
+../../inference/ch_det_data_50/all-sum-510/test_add_15.jpg	140 251 354 251 354 268 140 268 203 212 407 217 407 234 203 229 104 210 194 212 194 229 104 227 153 155 287 159 287 175 153 172 143 134 307 140 307 157 143 150 106 136 147 136 147 149 106 149 106 101 278 107 277 126 105 119 106 70 247 77 246 97 105 90 106 37 211 40 210 64 105 61 
+../../inference/ch_det_data_50/all-sum-510/test_add_16.jpg	380 740 750 740 750 780 380 780 360 700 472 700 472 728 360 728 1550 698 1580 698 1580 750 1550 750 1256 694 1444 694 1444 722 1256 722 1242 659 1452 659 1452 690 1242 690 384 643 672 643 672 682 384 682 1226 623 1474 621 1474 655 1226 657 356 599 582 599 582 631 356 631 1198 587 1496 587 1496 619 1198 619 1164 553 1534 553 1534 585 1164 585 378 549 642 549 642 589 378 589 354 500 520 500 520 540 354 540 772 258 1128 258 1128 303 772 303 372 208 508 208 508 303 372 303 774 208 1092 214 1092 260 774 254 
+../../inference/ch_det_data_50/all-sum-510/test_add_17.jpg	319 255 394 257 394 271 319 269 306 236 407 238 407 257 306 255 306 221 413 226 412 243 305 237 93 134 387 140 386 210 92 204 69 92 401 100 401 127 69 118 66 74 225 77 225 95 66 92 64 58 227 60 227 77 64 75 
+../../inference/ch_det_data_50/all-sum-510/test_add_18.jpg	153 908 616 914 616 935 153 930 464 786 718 788 718 816 464 813 552 750 666 755 665 792 551 788 117 538 190 538 190 572 117 572 115 472 676 484 675 530 114 518 119 427 670 439 670 471 119 459 119 374 676 379 676 411 119 406 555 261 677 262 677 280 555 279 164 258 336 258 336 275 164 275 342 194 457 196 457 221 342 219 307 172 490 172 490 190 307 190 252 125 540 129 540 171 252 168 345 90 488 92 488 110 345 108 283 40 569 48 567 84 282 76 235 30 268 30 268 64 235 64 
+../../inference/ch_det_data_50/all-sum-510/test_add_19.jpg	22 293 44 293 44 304 22 304 62 291 106 291 106 305 62 305 61 279 107 279 107 291 61 291 218 278 247 278 247 292 218 292 176 278 210 278 210 291 176 291 141 275 166 275 166 307 141 307 7 266 20 266 20 278 7 278 219 264 245 264 245 279 219 279 60 263 133 263 133 279 60 279 22 264 49 264 49 279 22 279 218 251 250 251 250 266 218 266 63 251 133 251 133 264 63 264 22 250 45 250 45 265 22 265 7 251 20 251 20 263 7 263 8 240 18 240 18 249 8 249 61 236 115 236 115 252 61 252 23 234 49 237 47 253 21 250 210 235 246 235 246 252 210 252 143 236 166 236 166 252 143 252 493 224 533 224 533 241 493 241 334 224 355 224 355 239 334 239 287 224 315 224 315 239 287 239 61 224 114 224 114 238 61 238 7 226 18 226 18 235 7 235 219 223 250 223 250 237 219 237 141 224 167 221 169 235 143 238 23 223 49 223 49 239 23 239 494 212 526 212 526 225 494 225 418 211 439 211 439 226 418 226 335 211 400 211 400 224 335 224 291 211 322 211 322 224 291 224 220 211 251 211 251 224 220 224 144 212 167 212 167 223 144 223 60 211 115 209 115 222 60 224 24 210 50 210 50 224 24 224 336 197 384 197 384 211 336 211 63 198 89 198 89 209 63 209 492 195 542 195 542 213 492 213 443 201 456 194 464 207 451 215 219 195 257 195 257 213 219 213 177 196 207 196 207 210 177 210 144 197 158 197 158 210 144 210 23 196 44 196 44 212 23 212 416 193 440 193 440 213 416 213 63 185 134 185 134 197 63 197 335 184 400 184 400 197 335 197 455 180 466 191 456 201 444 190 289 187 309 180 315 194 295 202 219 183 256 183 256 197 219 197 140 183 160 183 160 198 140 198 493 182 519 182 519 197 493 197 426 178 441 191 426 204 412 190 32 177 46 189 32 202 19 189 176 180 193 180 193 197 176 197 335 170 402 170 402 186 335 186 491 169 521 169 521 186 491 186 426 163 441 176 426 191 412 179 292 170 315 170 315 186 292 186 219 170 252 170 252 185 219 185 177 171 189 171 189 185 177 185 62 170 127 168 127 182 62 184 454 167 464 177 455 186 445 176 142 169 164 169 164 185 142 185 492 158 525 158 525 172 492 172 399 159 436 159 436 169 399 169 334 157 403 157 403 170 334 170 295 157 327 157 327 171 295 171 219 156 253 156 253 170 219 170 143 156 164 156 164 171 143 171 60 157 127 155 127 169 60 171 491 142 543 142 543 158 491 158 449 143 480 143 480 157 449 157 334 142 441 142 441 157 334 157 294 143 328 143 328 157 294 157 219 143 254 143 254 157 219 157 61 143 105 143 105 156 61 156 142 141 164 141 164 157 142 157 17 150 31 136 45 149 30 162 285 133 293 133 293 141 285 141 177 132 193 132 193 145 177 145 335 130 389 130 389 143 335 143 491 129 528 129 528 143 491 143 449 129 479 129 479 143 449 143 417 130 437 130 437 142 417 142 291 129 323 129 323 143 291 143 217 130 256 128 257 143 218 145 61 129 97 129 97 143 61 143 143 128 161 128 161 145 143 145 29 123 45 132 34 149 18 139 492 117 537 117 537 130 492 130 335 117 389 117 389 130 335 130 218 118 256 118 256 128 218 128 450 116 480 116 480 130 450 130 417 116 440 116 440 131 417 131 177 116 210 116 210 130 177 130 143 116 164 116 164 131 143 131 60 115 90 115 90 132 60 132 17 121 32 110 45 124 29 136 490 105 527 105 527 115 490 115 448 105 479 105 479 115 448 115 419 106 436 106 436 114 419 114 292 105 321 105 321 116 292 116 218 105 244 105 244 115 218 115 175 105 205 105 205 115 175 115 143 105 163 105 163 116 143 116 334 104 373 104 373 115 334 115 61 104 88 104 88 115 61 115 483 89 523 89 523 99 483 99 330 87 381 87 381 100 330 100 274 87 336 87 336 100 274 100 213 87 248 87 248 100 213 100 5 85 103 85 103 101 5 101 414 64 464 64 464 78 414 78 287 64 335 64 335 78 287 78 155 62 208 62 208 79 155 79 414 47 525 48 525 64 414 63 287 48 377 48 377 64 287 64 157 48 270 48 270 63 157 63 415 34 483 34 483 48 415 48 287 33 338 33 338 50 287 50 26 34 45 34 45 52 26 52 155 32 207 32 207 49 155 49 55 32 115 31 116 51 56 53 411 2 529 2 529 19 411 19 144 2 346 0 346 17 144 19 
+../../inference/ch_det_data_50/all-sum-510/test_add_2.jpg	251 404 535 404 535 430 251 430 302 339 483 339 483 385 302 385 302 303 482 303 482 326 302 326 573 217 693 217 693 240 573 240 331 216 455 214 455 240 331 242 108 212 182 214 181 244 107 242 313 98 672 99 672 121 313 120 311 60 585 61 585 87 311 86 
+../../inference/ch_det_data_50/all-sum-510/test_add_20.jpg	30 345 607 345 607 372 30 372 216 292 512 292 512 323 216 323 472 270 527 270 527 287 472 287 216 266 292 266 292 287 216 287 218 238 486 238 486 265 218 265 220 215 305 215 305 236 220 236 399 190 419 190 419 207 399 207 221 185 343 185 343 209 221 209 220 160 289 160 289 182 220 182 374 120 477 122 477 147 374 145 221 122 367 120 367 145 221 147 217 80 354 82 354 117 217 115 439 33 607 33 607 60 439 60 67 15 400 15 400 46 67 46 
+../../inference/ch_det_data_50/all-sum-510/test_add_3.jpg	168 326 339 324 339 341 168 343 169 286 309 288 309 314 169 312 169 219 324 219 324 235 169 235 339 219 451 216 451 232 339 235 168 200 373 200 373 216 168 216 168 180 418 180 418 197 168 197 169 147 417 147 417 165 169 165 170 117 419 117 419 141 170 141 325 62 480 62 480 93 325 93 170 62 310 59 311 91 171 94 
+../../inference/ch_det_data_50/all-sum-510/test_add_4.png	
+../../inference/ch_det_data_50/all-sum-510/test_add_5.png	47 162 109 162 109 176 47 176 51 119 170 119 170 136 51 136 49 100 166 100 166 119 49 119 51 83 166 83 166 102 51 102 50 66 169 66 169 85 50 85 49 47 149 46 149 68 49 69 5 9 81 9 81 43 5 43 
+../../inference/ch_det_data_50/all-sum-510/test_add_6.jpg	122 222 220 226 219 253 121 249 160 176 185 180 182 200 157 196 
+../../inference/ch_det_data_50/all-sum-510/test_add_7.jpg	47 937 175 933 176 964 48 967 224 870 632 873 632 955 224 952 53 743 640 743 640 793 53 793 148 673 546 676 546 723 148 720 71 502 636 502 636 604 71 604 54 264 660 274 657 446 51 436 59 173 534 173 534 241 59 241 502 173 646 173 646 239 502 239 
+../../inference/ch_det_data_50/all-sum-510/test_add_8.jpg	249 584 455 578 456 608 250 614 106 531 458 524 458 561 107 568 334 492 385 492 385 509 334 509 26 306 356 296 357 321 27 331 21 258 447 250 447 275 21 283 77 208 447 204 447 226 77 230 158 20 322 28 319 82 155 74 
+../../inference/ch_det_data_50/all-sum-510/test_add_9.png	264 684 486 684 486 697 264 697 194 666 556 666 556 682 194 682 152 595 600 595 600 608 152 608 211 577 542 577 542 590 211 590 131 558 616 558 616 571 131 571 84 540 665 540 665 553 84 553 95 521 654 521 654 536 95 536 361 448 390 448 390 461 361 461 236 375 515 375 515 391 236 391 174 353 575 353 575 369 174 369 342 279 409 281 409 298 342 296 254 203 493 203 493 220 254 220 
diff --git a/test_tipc/results/cpp_ppocr_det_mobile_results_fp32.txt b/test_tipc/results/cpp_ppocr_det_mobile_results_fp32.txt
new file mode 100644
index 0000000..fb33ce1
--- /dev/null
+++ b/test_tipc/results/cpp_ppocr_det_mobile_results_fp32.txt
@@ -0,0 +1,50 @@
+../../inference/ch_det_data_50/all-sum-510/00008790.jpg	208 404 282 404 282 421 208 421 58 396 107 396 107 413 58 413 197 387 296 387 296 403 197 403 161 389 174 389 174 402 161 402 34 378 134 378 134 394 34 394 323 377 329 377 329 382 323 382 199 370 292 370 292 383 199 383 216 309 274 309 274 325 216 325 161 304 173 304 173 315 161 315 370 301 437 301 437 317 370 317 30 301 135 300 135 316 30 317 221 291 270 291 270 308 221 308 58 224 106 224 106 238 58 238 216 222 274 222 274 239 216 239 161 217 174 217 174 229 161 229 33 205 133 205 133 221 33 221 221 204 270 204 270 221 221 221 73 145 385 145 385 162 73 162 52 119 119 119 119 135 52 135 72 50 296 50 296 66 72 66 54 15 118 15 118 32 54 32 
+../../inference/ch_det_data_50/all-sum-510/00018946.jpg	439 327 476 327 476 341 439 341 85 284 142 284 142 308 85 308 300 278 380 278 380 299 300 299 195 262 287 275 284 299 192 286 196 196 454 218 452 244 194 222 343 182 376 182 376 193 343 193 198 162 341 169 340 195 197 188 176 130 381 145 380 165 175 150 176 100 417 118 415 148 174 130 
+../../inference/ch_det_data_50/all-sum-510/00034387.jpg	263 459 741 459 741 485 263 485 346 415 421 415 421 444 346 444 544 418 568 418 568 442 544 442 684 415 712 415 712 444 684 444 173 413 228 413 228 444 173 444 872 412 910 412 910 447 872 447 55 415 76 415 76 443 55 443 855 371 927 371 927 401 855 401 347 371 420 371 420 400 347 400 672 370 725 370 725 402 672 402 537 371 571 371 571 401 537 401 136 364 230 367 229 403 135 400 55 370 76 370 76 399 55 399 856 328 927 328 927 358 856 358 350 328 420 328 420 358 350 358 672 326 725 326 725 358 672 358 539 327 571 327 571 359 539 359 170 326 229 323 231 357 171 359 56 328 76 328 76 358 56 358 297 326 316 326 316 334 297 334 854 284 927 284 927 314 854 314 672 284 725 284 725 315 672 315 344 284 431 282 432 315 345 317 537 283 570 283 570 314 537 314 170 281 228 281 228 315 170 315 55 285 75 285 75 314 55 314 856 241 927 241 927 270 856 270 346 240 464 240 464 271 346 271 154 241 228 241 228 271 154 271 672 240 726 240 726 271 672 271 530 240 573 240 573 272 530 272 55 241 76 241 76 270 55 270 854 196 927 198 926 228 853 225 672 197 728 197 728 228 672 228 342 199 439 194 441 224 344 230 175 196 229 196 229 226 175 226 55 199 75 199 75 228 55 228 526 193 578 193 578 228 526 228 347 154 420 154 420 182 347 182 853 153 927 153 927 181 853 181 175 153 228 153 228 184 175 184 668 152 725 152 725 182 668 182 536 153 572 153 572 183 536 183 55 155 76 155 76 183 55 183 347 109 420 109 420 138 347 138 172 109 229 109 229 140 172 140 544 111 565 111 565 138 544 138 51 110 77 110 77 140 51 140 639 105 730 105 730 141 639 141 815 101 929 109 927 141 813 133 812 65 953 65 953 93 812 93 305 64 447 66 447 94 305 92 671 65 725 65 725 95 671 95 173 64 229 66 228 96 172 94 37 64 91 66 90 98 36 96 527 63 581 63 581 95 527 95 333 18 671 18 671 45 333 45 
+../../inference/ch_det_data_50/all-sum-510/00037951.jpg	432 973 552 977 552 994 432 991 431 931 554 931 554 970 431 970 29 520 101 520 101 546 29 546 29 441 146 441 146 465 29 465 233 333 328 331 328 356 233 358 121 250 439 250 439 287 121 287 180 205 380 205 380 229 180 229 257 103 323 121 305 184 239 165 35 57 147 57 147 82 35 82 
+../../inference/ch_det_data_50/all-sum-510/00044782.jpg	222 214 247 214 247 230 222 230 162 214 183 214 183 231 162 231 122 190 216 190 216 203 122 203 90 82 252 82 252 100 90 100 70 61 279 61 279 78 70 78 103 14 244 14 244 46 103 46 
+../../inference/ch_det_data_50/all-sum-510/00067516.jpg	139 806 596 807 596 824 139 823 46 782 699 782 699 800 46 800 577 749 669 749 669 766 577 766 353 748 397 748 397 769 353 769 220 749 261 749 261 767 220 767 475 748 502 748 502 769 475 769 68 746 134 749 133 766 67 763 574 680 670 680 670 700 574 700 474 680 519 680 519 701 474 701 352 680 397 680 397 701 352 701 68 679 134 682 133 700 67 697 219 678 245 681 242 702 216 698 575 614 669 614 669 633 575 633 68 611 134 614 133 633 67 630 474 613 501 613 501 633 474 633 353 613 379 613 379 634 353 634 219 612 245 612 245 633 219 633 576 546 669 546 669 566 576 566 474 545 519 545 519 566 474 566 351 544 381 544 381 567 351 567 219 545 245 545 245 566 219 566 67 541 134 544 133 565 66 562 67 477 134 480 133 501 66 498 584 479 666 479 666 499 584 499 474 478 519 478 519 500 474 500 352 478 397 478 397 500 352 500 218 477 246 477 246 502 218 502 579 424 666 427 665 451 578 448 345 428 411 428 411 449 345 449 66 425 151 427 151 451 66 449 473 427 515 427 515 450 473 450 218 427 259 427 259 450 218 450 282 396 479 397 479 420 282 419 83 316 667 316 667 335 83 335 64 277 666 277 666 292 64 292 456 209 585 209 585 226 456 226 311 208 373 208 373 227 311 227 163 208 227 208 227 227 163 227 504 150 541 150 541 168 504 168 264 47 485 47 485 69 264 69 
+../../inference/ch_det_data_50/all-sum-510/00088568.jpg	57 443 119 443 119 456 57 456 309 413 744 413 744 430 309 430 309 375 737 375 737 392 309 392 415 337 559 337 559 351 415 351 307 322 674 321 674 338 307 339 275 292 349 294 349 313 275 311 52 285 210 285 210 301 52 301 273 262 420 262 420 279 273 279 55 262 249 262 249 279 55 279 669 247 697 247 697 262 669 262 601 247 629 247 629 262 601 262 531 247 559 247 559 262 531 262 461 247 489 247 489 262 461 262 277 247 310 247 310 261 277 261 55 240 142 240 142 254 55 254 276 230 400 230 400 246 276 246 741 227 749 237 741 246 732 237 665 230 701 230 701 245 665 245 598 230 631 230 631 245 598 245 527 230 563 230 563 245 527 245 458 230 493 230 493 245 458 245 52 213 212 215 212 233 52 231 732 214 747 214 747 227 732 227 662 212 706 212 706 230 662 230 594 213 638 213 638 227 594 227 522 213 570 213 570 227 522 227 453 213 497 213 497 227 453 227 278 213 352 213 352 227 278 227 734 198 748 198 748 210 734 210 667 196 702 196 702 210 667 210 599 196 633 196 633 211 599 211 527 196 564 196 564 210 527 210 459 196 493 196 493 210 459 210 276 194 418 195 418 212 276 211 54 190 241 190 241 207 54 207 664 179 705 179 705 194 664 194 278 178 352 180 352 195 278 193 733 179 747 179 747 194 733 194 596 178 635 178 635 193 596 193 523 177 567 177 567 195 523 195 456 178 495 178 495 193 456 193 55 170 142 170 142 184 55 184 733 164 748 164 748 176 733 176 664 162 705 162 705 176 664 176 597 162 635 162 635 176 597 176 525 162 566 162 566 176 525 176 456 162 494 162 494 176 456 176 277 160 399 160 399 176 277 176 54 146 149 146 149 161 54 161 452 145 497 145 497 160 452 160 729 144 748 144 748 162 729 162 662 143 706 143 706 161 662 161 595 144 636 144 636 159 595 159 521 143 566 141 567 159 522 161 277 143 310 143 310 159 277 159 275 120 430 120 430 140 275 140 50 119 234 120 234 140 50 139 402 90 703 90 703 107 402 107 46 78 282 78 282 98 46 98 324 67 745 68 745 86 324 85 666 47 743 47 743 64 666 64 295 47 435 47 435 63 295 63 64 30 232 27 233 65 65 68 
+../../inference/ch_det_data_50/all-sum-510/00091741.jpg	46 335 87 335 87 360 46 360 98 209 258 209 258 232 98 232 101 189 258 190 258 206 101 205 87 99 268 97 269 184 88 186 92 45 266 53 263 117 89 109 89 10 258 12 258 38 89 36 
+../../inference/ch_det_data_50/all-sum-510/00105313.jpg	289 261 407 261 407 277 289 277 152 260 265 260 265 276 152 276 10 257 74 259 74 276 10 274 32 230 134 230 134 245 32 245 34 215 218 215 218 228 34 228 32 199 148 199 148 214 32 214 31 181 217 182 217 199 31 198 34 169 107 169 107 182 34 182 34 153 126 153 126 166 34 166 33 136 144 137 144 150 33 149 34 122 177 122 177 135 34 135 32 104 178 104 178 120 32 120 32 91 102 91 102 104 32 104 33 75 121 75 121 88 33 88 32 60 121 60 121 73 32 73 34 44 121 44 121 57 34 57 31 28 144 28 144 43 31 43 177 20 415 15 416 51 178 56 24 10 152 10 152 26 24 26 
+../../inference/ch_det_data_50/all-sum-510/00134770.jpg	386 645 457 645 457 658 386 658 406 618 486 617 486 635 406 636 111 533 272 530 272 550 111 553 110 501 445 496 445 516 110 521 110 469 445 465 445 485 110 489 110 438 446 433 446 453 110 458 109 407 445 403 445 423 109 427 151 375 443 372 443 392 151 395 183 336 371 334 371 358 183 360 224 307 272 308 272 318 224 317 73 96 517 101 516 220 72 215 
+../../inference/ch_det_data_50/all-sum-510/00145943.jpg	390 243 751 274 735 454 375 423 88 90 302 90 302 121 88 121 43 40 329 37 329 78 43 81 
+../../inference/ch_det_data_50/all-sum-510/00147605.jpg	514 605 786 604 786 629 514 630 116 521 226 521 226 561 116 561 252 522 309 522 309 558 252 558 713 500 902 503 902 539 713 536 254 501 296 501 296 519 254 519 345 479 475 479 475 517 345 517 251 483 296 483 296 501 251 501 350 456 447 456 447 471 350 471 143 442 203 442 203 469 143 469 727 370 880 370 880 422 727 422 526 369 684 369 684 421 526 421 140 367 490 367 490 423 140 423 742 313 872 313 872 338 742 338 798 155 888 155 888 192 798 192 272 140 457 140 457 161 272 161 737 114 895 118 894 158 736 155 107 110 206 110 206 131 107 131 268 92 464 94 464 134 268 131 
+../../inference/ch_det_data_50/all-sum-510/00150341.jpg	99 643 300 643 300 664 99 664 113 615 289 615 289 633 113 633 82 591 320 590 320 611 82 612 30 563 315 561 315 582 30 584 30 513 169 513 169 531 30 531 32 488 111 488 111 506 32 506 357 458 465 461 464 486 356 483 26 458 271 459 271 483 26 482 338 438 423 442 422 461 337 457 64 437 145 437 145 455 64 455 205 414 293 414 293 436 205 436 318 407 442 411 441 439 317 435 42 404 176 407 176 435 42 432 28 381 137 381 137 405 28 405 
+../../inference/ch_det_data_50/all-sum-510/00150669.jpg	647 698 683 698 683 718 647 718 515 684 551 684 551 721 515 721 650 687 680 687 680 702 650 702 920 673 938 673 938 686 920 686 518 670 548 670 548 690 518 690 785 670 808 670 808 688 785 688 590 670 608 670 608 688 590 688 732 665 745 679 732 692 718 679 652 668 680 668 680 689 652 689 271 665 423 665 423 690 271 690 130 664 205 664 205 690 130 690 44 664 111 664 111 689 44 689 781 628 812 628 812 663 781 663 643 626 687 626 687 666 643 666 514 627 550 627 550 665 514 665 654 617 673 617 673 629 654 629 858 617 868 617 868 628 858 628 727 617 736 617 736 628 727 628 920 614 940 614 940 631 920 631 785 614 807 614 807 631 785 631 371 603 421 603 421 620 371 620 83 600 216 603 216 624 83 620 46 602 72 602 72 624 46 624 780 569 817 573 813 610 776 606 922 559 936 559 936 575 922 575 856 559 869 559 869 575 856 575 61 552 411 552 411 569 61 569 61 531 117 533 117 547 61 545 859 527 868 527 868 539 859 539 923 525 936 525 936 542 923 542 787 524 807 524 807 540 787 540 526 526 536 526 536 536 526 536 261 511 396 511 396 528 261 528 120 512 246 512 246 526 120 526 47 512 120 512 120 527 47 527 753 491 829 491 829 508 753 508 636 491 712 491 712 508 636 508 517 491 593 491 593 508 517 508 84 448 125 448 125 463 84 463 221 448 238 448 238 462 221 462 682 444 869 444 869 461 682 461 561 444 667 444 667 461 561 461 489 445 545 445 545 459 489 459 183 437 209 437 209 459 183 459 52 429 73 437 64 464 42 456 222 430 278 430 278 445 222 445 86 430 145 430 145 445 86 445 505 382 617 381 617 398 505 399 701 380 758 380 758 398 701 398 307 371 365 371 365 386 307 386 90 371 168 371 168 386 90 386 686 334 821 334 821 352 686 352 496 333 659 333 659 350 496 350 207 314 245 314 245 333 207 333 497 287 642 287 642 304 497 304 670 286 804 286 804 304 670 304 668 239 817 239 817 257 668 257 495 239 644 239 644 257 495 257 668 193 816 193 816 209 668 209 496 193 644 193 644 209 496 209 668 144 816 144 816 161 668 161 497 144 646 144 646 161 497 161 488 102 546 102 546 121 488 121 845 21 900 21 900 43 845 43 25 18 702 18 702 39 25 39 896 10 997 14 996 46 895 42 
+../../inference/ch_det_data_50/all-sum-510/00152568.jpg	2 250 285 252 285 281 2 279 195 231 255 231 255 241 195 241 198 158 282 164 277 230 193 224 177 148 251 148 251 161 177 161 
+../../inference/ch_det_data_50/all-sum-510/00155628.jpg	147 898 506 901 506 925 147 922 519 892 562 894 561 912 518 910 59 884 83 884 83 895 59 895 148 877 505 881 505 902 148 897 523 833 641 837 640 858 522 854 68 832 187 834 187 855 68 853 245 554 468 554 468 570 245 570 307 506 405 508 405 526 307 523 243 481 460 483 460 504 243 502 250 420 460 422 460 454 250 452 193 377 518 379 518 410 193 408 473 194 625 194 625 212 473 212 70 127 643 129 643 163 70 161 478 39 599 35 602 101 481 105 67 23 136 14 140 44 71 54 
+../../inference/ch_det_data_50/all-sum-510/00173364.jpg	7 176 58 176 58 201 7 201 135 118 196 118 196 135 135 135 38 75 87 75 87 105 38 105 249 19 313 19 313 38 249 38 19 15 105 15 105 40 19 40 
+../../inference/ch_det_data_50/all-sum-510/00175503.jpg	39 256 503 252 504 362 40 366 49 198 351 175 357 253 55 276 
+../../inference/ch_det_data_50/all-sum-510/00193218.jpg	282 373 411 373 411 389 282 389 170 373 223 373 223 390 170 390 108 373 162 373 162 390 108 390 276 357 358 357 358 371 276 371 169 357 222 357 222 371 169 371 106 356 175 356 175 373 106 373 408 356 493 356 493 370 408 370 24 185 64 185 64 203 24 203 500 184 558 184 558 201 500 201 379 185 421 183 422 200 380 202 283 184 311 184 311 202 283 202 173 185 197 185 197 201 173 201 498 163 544 163 544 177 498 177 379 162 412 162 412 177 379 177 261 161 303 161 303 178 261 178 174 161 231 161 231 178 174 178 24 161 80 161 80 178 24 178 385 139 489 139 489 155 385 155 26 137 133 137 133 153 26 153 442 115 538 117 538 134 442 132 345 117 406 117 406 131 345 131 259 117 303 117 303 131 259 131 28 112 229 114 229 132 28 130 130 90 395 93 395 110 130 107 560 81 585 81 585 109 560 109 
+../../inference/ch_det_data_50/all-sum-510/00195033.jpg	221 302 240 302 240 309 221 309 487 262 534 264 533 282 486 280 125 249 194 249 194 285 125 285 336 248 364 248 364 268 336 268 317 221 381 223 381 240 317 238 431 224 450 224 450 236 431 236 360 202 539 202 539 218 360 218 87 199 148 201 148 218 87 216 371 181 450 181 450 195 371 195 327 180 354 180 354 194 327 194 94 178 241 178 241 195 94 195 431 159 559 159 559 175 431 175 128 148 289 149 289 166 128 165 35 145 75 148 74 163 34 160 487 146 501 146 501 153 487 153 100 143 122 143 122 154 100 154 370 127 505 126 505 140 370 141 98 125 194 125 194 139 98 139 320 125 338 125 338 136 320 136 35 121 78 121 78 135 35 135 322 104 338 104 338 116 322 116 371 101 503 101 503 117 371 117 348 103 362 103 362 115 348 115 37 101 81 101 81 114 37 114 97 98 207 99 207 116 97 115 305 89 317 89 317 97 305 97 346 86 364 86 364 97 346 97 319 85 342 85 342 100 319 100 357 82 515 80 515 96 357 98 40 81 90 81 90 94 40 94 92 77 242 78 242 95 92 94 312 65 394 65 394 79 312 79 240 64 290 64 290 78 240 78 183 52 222 52 222 66 183 66 468 47 547 47 547 61 468 61 422 34 438 34 438 55 422 55 464 29 551 29 551 43 464 43 206 19 330 21 330 42 206 40 
+../../inference/ch_det_data_50/all-sum-510/00208502.jpg	556 535 630 535 630 569 556 569 204 537 284 537 284 552 204 552 142 512 191 512 191 526 142 526 248 511 309 511 309 525 248 525 41 499 118 499 118 520 41 520 465 490 558 490 558 510 465 510 666 489 680 493 677 505 662 501 724 490 739 490 739 503 724 503 40 450 118 448 118 469 40 471 173 448 237 448 237 465 173 465 93 403 121 403 121 424 93 424 38 403 63 403 63 424 38 424 214 392 232 405 220 422 203 409 39 357 58 357 58 375 39 375 92 355 121 355 121 375 92 375 187 339 248 337 249 363 188 365 458 319 551 317 551 338 458 340 457 271 553 271 553 292 457 292 562 271 737 267 737 288 562 292 516 225 548 225 548 245 516 245 620 185 675 185 675 202 620 202 456 130 550 128 550 149 456 151 571 104 789 98 789 121 571 127 121 46 291 46 291 99 121 99 536 36 710 36 710 92 536 92 
+../../inference/ch_det_data_50/all-sum-510/00224225.jpg	135 426 157 426 157 449 135 449 199 402 480 408 479 461 198 455 200 225 474 225 474 394 200 394 130 264 174 264 174 281 130 281 343 205 458 205 458 232 343 232 197 186 349 194 346 242 194 234 7 41 160 39 161 115 8 117 
+../../inference/ch_det_data_50/all-sum-510/00227746.jpg	142 230 210 230 210 240 142 240 72 230 130 230 130 240 72 240 215 228 386 228 386 240 215 240 290 208 347 208 347 224 290 224 142 179 165 181 162 209 139 208 171 152 347 152 347 167 171 167 143 110 279 112 279 135 143 132 202 53 387 53 387 69 202 69 141 47 193 47 193 64 141 64 
+../../inference/ch_det_data_50/all-sum-510/00229605.jpg	742 528 882 528 882 545 742 545 232 497 590 496 590 524 232 525 5 496 229 496 229 524 5 524 734 494 884 497 884 522 734 519 605 493 718 488 719 517 606 522 2 242 865 227 866 291 3 305 477 26 884 26 884 77 477 77 
+../../inference/ch_det_data_50/all-sum-510/00233011.jpg	61 225 293 225 293 243 61 243 11 218 43 218 43 252 11 252 60 177 120 177 120 196 60 196 11 169 44 169 44 204 11 204 59 127 149 129 149 148 59 146 11 123 45 123 45 156 11 156 124 87 239 87 239 105 124 105 147 49 218 49 218 67 147 67 257 44 354 47 353 71 256 68 8 47 54 47 54 69 8 69 275 10 346 10 346 32 275 32 26 9 75 9 75 32 26 32 
+../../inference/ch_det_data_50/all-sum-510/00233625.jpg	370 395 635 397 635 445 370 443 67 210 935 204 936 325 68 331 
+../../inference/ch_det_data_50/all-sum-510/00233634.jpg	213 637 264 637 264 706 213 706 522 634 572 634 572 697 522 697 641 522 684 522 684 570 641 570 95 514 155 514 155 592 95 592 754 394 762 394 762 403 754 403 677 362 730 360 733 432 679 433 53 360 109 360 109 436 53 436 77 207 157 207 157 282 77 282 642 204 695 204 695 274 642 274 208 88 262 85 266 165 212 168 362 47 428 44 432 117 366 120 
+../../inference/ch_det_data_50/all-sum-510/00234400.jpg	156 419 739 419 739 439 156 439 157 393 653 393 653 412 157 412 38 390 129 390 129 413 38 413 156 339 307 342 307 365 156 362 36 342 125 342 125 363 36 363 519 293 705 293 705 316 519 316 393 290 485 288 485 316 393 318 156 291 271 291 271 315 156 315 35 291 127 291 127 315 35 315 155 242 360 242 360 269 155 269 34 242 83 242 83 270 34 270 27 150 159 150 159 177 27 177 280 96 507 96 507 113 280 113 313 44 477 47 476 90 312 87 516 50 664 52 664 68 516 67 485 17 708 15 708 45 485 47 
+../../inference/ch_det_data_50/all-sum-510/00234883.jpg	64 122 318 117 319 193 65 197 71 118 122 118 122 132 71 132 381 62 506 61 506 75 381 76 54 25 369 23 369 116 54 118 385 26 503 23 503 47 385 50 
+../../inference/ch_det_data_50/all-sum-510/test_add_0.jpg	311 521 391 521 391 534 311 534 277 499 426 499 426 516 277 516 259 445 438 445 438 461 259 461 210 426 487 426 487 443 210 443 244 385 460 385 460 411 244 411 220 327 476 327 476 373 220 373 205 204 494 208 493 279 204 275 264 163 423 165 423 198 264 196 15 17 203 15 203 45 15 47 
+../../inference/ch_det_data_50/all-sum-510/test_add_1.png	
+../../inference/ch_det_data_50/all-sum-510/test_add_10.png	155 123 187 123 187 174 155 174 160 105 184 105 184 131 160 131 116 45 155 44 158 176 119 176 63 30 102 31 99 172 60 171 
+../../inference/ch_det_data_50/all-sum-510/test_add_11.jpg	1388 755 1486 755 1486 794 1388 794 1011 752 1210 752 1210 802 1011 802 681 752 879 752 879 801 681 801 355 750 568 745 570 796 356 801 76 748 266 743 268 796 78 801 600 645 1155 645 1155 706 600 706 600 562 1151 553 1151 614 600 622 596 478 1070 470 1070 529 596 537 595 390 1095 385 1095 444 595 448 600 303 1061 303 1061 362 600 362 353 180 1521 180 1521 265 353 265 59 40 261 40 261 91 59 91 1303 39 1495 39 1495 90 1303 90 971 37 1173 32 1175 83 973 88 668 37 864 32 866 83 670 88 361 32 561 32 561 88 361 88 
+../../inference/ch_det_data_50/all-sum-510/test_add_12.jpg	9 590 140 592 140 615 9 613 107 520 908 524 908 571 107 566 632 448 905 445 905 481 632 484 110 445 468 447 468 487 110 485 580 303 682 301 683 351 581 353 368 257 568 262 565 361 364 355 61 83 856 85 856 164 61 162 
+../../inference/ch_det_data_50/all-sum-510/test_add_13.jpg	68 93 118 96 116 116 66 113 
+../../inference/ch_det_data_50/all-sum-510/test_add_14.jpg	28 94 238 92 238 130 28 132 27 50 241 48 241 88 27 90 
+../../inference/ch_det_data_50/all-sum-510/test_add_15.jpg	140 251 354 251 354 268 140 268 203 212 407 217 407 234 203 229 104 210 194 212 194 229 104 227 153 155 287 159 287 175 153 172 143 134 307 140 307 157 143 150 106 136 147 136 147 149 106 149 106 101 278 107 277 126 105 119 106 70 247 77 246 97 105 90 106 37 211 40 210 64 105 61 
+../../inference/ch_det_data_50/all-sum-510/test_add_16.jpg	380 740 750 740 750 780 380 780 360 700 472 700 472 728 360 728 1550 698 1580 698 1580 750 1550 750 1256 694 1444 694 1444 722 1256 722 1242 659 1452 659 1452 690 1242 690 384 643 672 643 672 682 384 682 1226 623 1474 621 1474 655 1226 657 356 599 582 599 582 631 356 631 1198 587 1496 587 1496 619 1198 619 1164 553 1534 553 1534 585 1164 585 378 549 642 549 642 589 378 589 354 500 520 500 520 540 354 540 772 258 1128 258 1128 303 772 303 372 208 508 208 508 303 372 303 774 208 1092 214 1092 260 774 254 
+../../inference/ch_det_data_50/all-sum-510/test_add_17.jpg	319 255 394 257 394 271 319 269 306 236 407 238 407 257 306 255 306 221 413 226 412 243 305 237 93 135 387 140 386 209 92 204 69 92 401 100 401 127 69 118 66 74 225 77 225 95 66 92 64 58 227 60 227 77 64 75 
+../../inference/ch_det_data_50/all-sum-510/test_add_18.jpg	153 908 616 914 616 935 153 930 464 786 718 788 718 816 464 813 552 750 666 755 665 792 551 788 117 538 190 538 190 572 117 572 115 472 676 484 675 530 114 518 119 427 670 439 670 471 119 459 119 374 676 380 676 411 119 405 555 261 677 262 677 280 555 279 164 258 336 258 336 275 164 275 342 194 457 196 457 221 342 219 307 172 490 172 490 190 307 190 252 125 540 129 540 171 252 168 345 90 488 92 488 110 345 108 283 40 569 48 567 84 282 76 235 30 268 30 268 64 235 64 
+../../inference/ch_det_data_50/all-sum-510/test_add_19.jpg	22 293 44 293 44 304 22 304 62 291 106 291 106 305 62 305 61 279 107 279 107 291 61 291 218 278 247 278 247 292 218 292 176 278 210 278 210 291 176 291 141 275 166 275 166 307 141 307 7 266 20 266 20 278 7 278 219 264 245 264 245 279 219 279 60 263 133 263 133 279 60 279 22 264 49 264 49 279 22 279 218 251 250 251 250 266 218 266 63 251 133 251 133 264 63 264 22 250 45 250 45 265 22 265 7 251 20 251 20 263 7 263 8 240 18 240 18 249 8 249 61 236 115 236 115 252 61 252 23 234 49 237 47 253 21 250 210 235 246 235 246 252 210 252 143 236 166 236 166 252 143 252 493 224 533 224 533 241 493 241 334 224 355 224 355 239 334 239 287 224 315 224 315 239 287 239 61 224 114 224 114 238 61 238 7 226 18 226 18 235 7 235 219 223 250 223 250 237 219 237 141 224 167 221 169 235 143 238 23 223 49 223 49 239 23 239 494 212 526 212 526 225 494 225 418 211 439 211 439 226 418 226 335 211 400 211 400 224 335 224 291 211 322 211 322 224 291 224 220 211 251 211 251 224 220 224 144 212 167 212 167 223 144 223 60 211 115 209 115 222 60 224 24 210 50 210 50 224 24 224 336 197 384 197 384 211 336 211 63 198 89 198 89 209 63 209 492 195 542 195 542 213 492 213 219 195 257 195 257 213 219 213 177 196 207 196 207 210 177 210 144 197 158 197 158 210 144 210 23 196 44 196 44 212 23 212 416 193 440 193 440 213 416 213 63 185 134 185 134 197 63 197 335 184 400 184 400 197 335 197 455 180 466 191 456 201 444 190 289 187 309 180 315 194 295 202 219 183 256 183 256 197 219 197 140 183 160 183 160 198 140 198 493 182 519 182 519 197 493 197 426 178 441 191 426 204 412 190 32 177 46 189 32 202 19 189 176 180 193 180 193 197 176 197 335 170 402 170 402 186 335 186 491 169 521 169 521 186 491 186 426 163 441 176 426 191 412 179 292 170 315 170 315 186 292 186 219 170 252 170 252 185 219 185 177 171 189 171 189 185 177 185 62 170 127 168 127 182 62 184 454 167 464 177 455 186 445 176 142 169 164 169 164 185 142 185 492 158 525 158 525 172 492 172 399 159 436 159 436 169 399 169 334 157 403 157 403 170 334 170 295 157 327 157 327 171 295 171 219 156 253 156 253 170 219 170 143 156 164 156 164 171 143 171 60 157 127 155 127 169 60 171 491 142 543 142 543 158 491 158 449 143 480 143 480 157 449 157 334 142 441 142 441 157 334 157 294 143 328 143 328 157 294 157 219 143 254 143 254 157 219 157 61 143 105 143 105 156 61 156 142 141 164 141 164 157 142 157 17 150 31 136 45 149 30 162 285 133 293 133 293 141 285 141 177 132 193 132 193 145 177 145 335 130 389 130 389 143 335 143 491 129 528 129 528 143 491 143 449 129 479 129 479 143 449 143 291 129 323 129 323 143 291 143 217 130 256 128 257 143 218 145 61 129 97 129 97 143 61 143 416 128 439 128 439 143 416 143 143 128 161 128 161 145 143 145 29 123 45 132 34 149 18 139 492 117 537 117 537 130 492 130 335 117 389 117 389 130 335 130 218 118 256 118 256 128 218 128 450 116 480 116 480 130 450 130 417 116 440 116 440 131 417 131 177 116 210 116 210 130 177 130 143 116 164 116 164 131 143 131 60 115 90 115 90 132 60 132 17 121 32 110 45 124 29 136 490 105 527 105 527 115 490 115 448 105 479 105 479 115 448 115 419 106 436 106 436 114 419 114 292 105 321 105 321 116 292 116 218 105 244 105 244 115 218 115 175 105 205 105 205 115 175 115 143 105 163 105 163 116 143 116 334 104 373 104 373 115 334 115 61 104 88 104 88 115 61 115 483 89 523 89 523 99 483 99 330 87 381 87 381 100 330 100 274 87 336 87 336 100 274 100 213 87 248 87 248 100 213 100 5 85 103 85 103 101 5 101 414 64 464 64 464 78 414 78 287 64 335 64 335 78 287 78 155 62 208 62 208 79 155 79 414 47 525 48 525 64 414 63 287 48 377 48 377 64 287 64 157 48 270 48 270 63 157 63 415 34 483 34 483 48 415 48 287 33 338 33 338 50 287 50 26 34 45 34 45 52 26 52 155 32 207 32 207 49 155 49 55 32 115 31 116 51 56 53 411 2 529 2 529 19 411 19 144 2 346 0 346 17 144 19 
+../../inference/ch_det_data_50/all-sum-510/test_add_2.jpg	251 404 535 404 535 430 251 430 302 339 483 339 483 385 302 385 302 303 482 303 482 326 302 326 573 217 693 217 693 240 573 240 331 216 455 214 455 240 331 242 108 212 182 214 181 244 107 242 313 98 672 99 672 121 313 120 311 60 585 61 585 87 311 86 
+../../inference/ch_det_data_50/all-sum-510/test_add_20.jpg	30 345 607 345 607 372 30 372 216 292 512 292 512 323 216 323 472 270 527 270 527 287 472 287 216 266 292 266 292 287 216 287 218 238 486 238 486 265 218 265 220 215 305 215 305 236 220 236 221 185 343 185 343 209 221 209 220 160 289 160 289 182 220 182 374 120 477 122 477 147 374 145 221 122 367 120 367 145 221 147 217 80 354 82 354 117 217 115 439 33 607 33 607 60 439 60 67 15 400 15 400 46 67 46 
+../../inference/ch_det_data_50/all-sum-510/test_add_3.jpg	168 326 339 324 339 341 168 343 169 286 309 288 309 314 169 312 169 219 324 219 324 235 169 235 339 219 451 216 451 232 339 235 168 200 373 200 373 216 168 216 168 180 418 180 418 197 168 197 169 147 417 147 417 165 169 165 170 117 419 117 419 141 170 141 325 62 480 62 480 93 325 93 170 62 310 59 311 91 171 94 
+../../inference/ch_det_data_50/all-sum-510/test_add_4.png	
+../../inference/ch_det_data_50/all-sum-510/test_add_5.png	47 162 109 162 109 176 47 176 51 119 170 119 170 136 51 136 49 100 166 100 166 119 49 119 51 83 166 83 166 102 51 102 50 66 169 66 169 85 50 85 49 47 149 46 149 68 49 69 5 9 81 9 81 43 5 43 
+../../inference/ch_det_data_50/all-sum-510/test_add_6.jpg	122 222 220 226 219 253 121 249 160 176 185 180 182 200 157 196 
+../../inference/ch_det_data_50/all-sum-510/test_add_7.jpg	47 937 175 933 176 964 48 967 224 870 632 873 632 955 224 952 53 743 640 743 640 793 53 793 148 673 546 676 546 723 148 720 71 502 636 502 636 604 71 604 54 264 660 274 657 446 51 436 59 173 534 173 534 241 59 241 502 173 646 173 646 239 502 239 
+../../inference/ch_det_data_50/all-sum-510/test_add_8.jpg	249 584 455 578 456 608 250 614 106 531 458 524 458 561 107 568 334 492 385 492 385 509 334 509 26 306 356 296 357 321 27 331 21 258 447 250 447 275 21 283 77 208 447 204 447 226 77 230 158 20 322 28 319 82 155 74 
+../../inference/ch_det_data_50/all-sum-510/test_add_9.png	264 684 486 684 486 697 264 697 194 666 556 666 556 682 194 682 152 595 600 595 600 608 152 608 211 577 543 577 543 590 211 590 131 559 617 558 617 572 131 573 84 540 665 540 665 553 84 553 95 521 654 521 654 536 95 536 361 448 390 448 390 461 361 461 236 375 515 375 515 391 236 391 174 353 575 353 575 369 174 369 342 279 409 281 409 298 342 296 254 203 493 203 493 220 254 220 
diff --git a/test_tipc/results/python_ppocr_det_mobile_results_fp16.txt b/test_tipc/results/python_ppocr_det_mobile_results_fp16.txt
new file mode 100644
index 0000000..191bdaf
--- /dev/null
+++ b/test_tipc/results/python_ppocr_det_mobile_results_fp16.txt
@@ -0,0 +1,49 @@
+00008790.jpg	[[[209, 406], [280, 406], [280, 419], [209, 419]], [[60, 398], [105, 398], [105, 411], [60, 411]], [[198, 389], [291, 389], [291, 402], [198, 402]], [[162, 391], [173, 391], [173, 401], [162, 401]], [[35, 380], [133, 380], [133, 393], [35, 393]], [[199, 371], [292, 371], [292, 384], [199, 384]], [[218, 310], [272, 310], [272, 324], [218, 324]], [[162, 305], [172, 305], [172, 314], [162, 314]], [[371, 302], [436, 302], [436, 316], [371, 316]], [[31, 302], [134, 301], [134, 315], [31, 316]], [[223, 292], [269, 292], [269, 306], [223, 306]], [[60, 225], [104, 225], [104, 236], [60, 236]], [[218, 223], [272, 223], [272, 237], [218, 237]], [[162, 219], [173, 219], [173, 227], [162, 227]], [[33, 207], [131, 207], [131, 220], [33, 220]], [[223, 206], [269, 206], [269, 220], [223, 220]], [[74, 146], [383, 146], [383, 159], [74, 159]], [[54, 120], [117, 120], [117, 134], [54, 134]], [[74, 51], [296, 51], [296, 65], [74, 65]], [[56, 18], [116, 18], [116, 32], [56, 32]]]
+00018946.jpg	[[[441, 328], [474, 328], [474, 339], [441, 339]], [[86, 284], [141, 286], [140, 307], [85, 305]], [[302, 279], [377, 279], [377, 297], [302, 297]], [[197, 265], [281, 274], [279, 293], [195, 284]], [[198, 197], [452, 219], [450, 242], [196, 220]], [[343, 182], [376, 182], [376, 192], [343, 192]], [[199, 164], [340, 171], [339, 192], [198, 185]], [[177, 101], [415, 118], [413, 145], [175, 128]]]
+00034387.jpg	[[[265, 460], [740, 460], [740, 484], [265, 484]], [[348, 417], [420, 417], [420, 443], [348, 443]], [[545, 418], [568, 418], [568, 442], [545, 442]], [[685, 417], [710, 417], [710, 443], [685, 443]], [[175, 415], [226, 415], [226, 443], [175, 443]], [[874, 414], [908, 414], [908, 446], [874, 446]], [[56, 417], [74, 417], [74, 442], [56, 442]], [[856, 373], [925, 373], [925, 400], [856, 400]], [[348, 372], [418, 372], [418, 397], [348, 397]], [[674, 372], [723, 372], [723, 401], [674, 401]], [[539, 373], [570, 373], [570, 400], [539, 400]], [[151, 365], [228, 369], [226, 402], [149, 398]], [[56, 372], [74, 372], [74, 397], [56, 397]], [[857, 329], [925, 329], [925, 355], [857, 355]], [[351, 330], [419, 330], [419, 356], [351, 356]], [[674, 328], [723, 328], [723, 356], [674, 356]], [[541, 329], [570, 329], [570, 357], [541, 357]], [[171, 327], [227, 324], [229, 355], [173, 358]], [[57, 330], [74, 330], [74, 356], [57, 356]], [[298, 327], [316, 327], [316, 334], [298, 334]], [[855, 286], [925, 286], [925, 312], [855, 312]], [[674, 286], [723, 286], [723, 313], [674, 313]], [[346, 286], [426, 283], [427, 313], [347, 316]], [[540, 285], [569, 285], [569, 312], [540, 312]], [[172, 282], [226, 282], [226, 313], [172, 313]], [[56, 287], [73, 287], [73, 312], [56, 312]], [[857, 242], [925, 242], [925, 268], [857, 268]], [[348, 242], [460, 242], [460, 268], [348, 268]], [[156, 242], [227, 242], [227, 269], [156, 269]], [[674, 241], [724, 241], [724, 269], [674, 269]], [[531, 241], [572, 241], [572, 270], [531, 270]], [[56, 242], [74, 242], [74, 268], [56, 268]], [[855, 197], [925, 200], [924, 226], [854, 224]], [[674, 198], [726, 198], [726, 226], [674, 226]], [[344, 200], [430, 195], [432, 224], [346, 230]], [[176, 197], [227, 197], [227, 225], [176, 225]], [[56, 200], [73, 200], [73, 226], [56, 226]], [[527, 194], [576, 194], [576, 226], [527, 226]], [[349, 155], [419, 155], [419, 181], [349, 181]], [[854, 154], [925, 154], [925, 180], [854, 180]], [[176, 154], [226, 154], [226, 183], [176, 183]], [[670, 153], [723, 153], [723, 181], [670, 181]], [[538, 154], [571, 154], [571, 182], [538, 182]], [[56, 156], [74, 156], [74, 182], [56, 182]], [[349, 111], [419, 111], [419, 137], [349, 137]], [[174, 111], [227, 111], [227, 139], [174, 139]], [[546, 113], [564, 113], [564, 137], [546, 137]], [[52, 112], [75, 112], [75, 139], [52, 139]], [[639, 108], [727, 105], [728, 138], [640, 141]], [[817, 103], [927, 110], [925, 139], [815, 132]], [[814, 68], [951, 68], [951, 92], [814, 92]], [[307, 66], [446, 68], [446, 93], [306, 90]], [[673, 67], [723, 67], [723, 93], [673, 93]], [[175, 65], [228, 68], [226, 95], [174, 92]], [[39, 65], [90, 68], [88, 97], [37, 94]], [[528, 65], [580, 65], [580, 94], [528, 94]], [[334, 20], [670, 20], [670, 43], [334, 43]]]
+00037951.jpg	[[[434, 976], [551, 978], [550, 993], [434, 991]], [[433, 932], [553, 932], [553, 969], [433, 969]], [[30, 522], [98, 522], [98, 545], [30, 545]], [[31, 443], [145, 443], [145, 464], [31, 464]], [[234, 335], [326, 332], [327, 354], [235, 356]], [[124, 252], [436, 252], [436, 284], [124, 284]], [[182, 206], [378, 206], [378, 227], [182, 227]], [[258, 106], [320, 123], [304, 181], [242, 163]], [[28, 65], [33, 65], [33, 71], [28, 71]], [[37, 58], [147, 58], [147, 80], [37, 80]]]
+00044782.jpg	[[[104, 218], [115, 218], [115, 227], [104, 227]], [[223, 216], [246, 216], [246, 228], [223, 228]], [[163, 216], [182, 216], [182, 229], [163, 229]], [[124, 191], [164, 191], [164, 202], [124, 202]], [[91, 84], [251, 84], [251, 98], [91, 98]], [[73, 63], [278, 63], [278, 78], [73, 78]], [[104, 15], [243, 15], [243, 44], [104, 44]]]
+00067516.jpg	[[[141, 808], [594, 809], [594, 822], [141, 821]], [[49, 784], [696, 784], [696, 798], [49, 798]], [[579, 751], [667, 751], [667, 764], [579, 764]], [[355, 750], [395, 750], [395, 767], [355, 767]], [[221, 751], [260, 751], [260, 765], [221, 765]], [[477, 750], [501, 750], [501, 768], [477, 768]], [[69, 748], [133, 751], [132, 765], [68, 761]], [[576, 682], [668, 682], [668, 699], [576, 699]], [[476, 682], [518, 682], [518, 700], [476, 700]], [[354, 682], [395, 682], [395, 700], [354, 700]], [[69, 681], [133, 684], [132, 699], [68, 695]], [[220, 679], [243, 682], [241, 700], [218, 697]], [[577, 615], [667, 615], [667, 632], [577, 632]], [[68, 612], [134, 615], [133, 632], [67, 629]], [[476, 614], [500, 614], [500, 633], [476, 633]], [[354, 613], [378, 613], [378, 634], [354, 634]], [[219, 612], [245, 612], [245, 633], [219, 633]], [[578, 547], [667, 547], [667, 564], [578, 564]], [[476, 546], [518, 546], [518, 565], [476, 565]], [[353, 545], [379, 545], [379, 566], [353, 566]], [[219, 545], [245, 545], [245, 566], [219, 566]], [[68, 542], [133, 546], [132, 563], [67, 560]], [[68, 478], [133, 482], [132, 499], [67, 496]], [[586, 481], [664, 481], [664, 497], [586, 497]], [[476, 480], [518, 480], [518, 498], [476, 498]], [[354, 480], [395, 480], [395, 498], [354, 498]], [[219, 479], [245, 479], [245, 500], [219, 500]], [[580, 425], [665, 429], [664, 449], [580, 446]], [[346, 429], [410, 429], [410, 447], [346, 447]], [[68, 426], [150, 429], [149, 449], [67, 447]], [[474, 427], [515, 427], [515, 449], [474, 449]], [[218, 427], [259, 427], [259, 449], [218, 449]], [[283, 398], [478, 399], [478, 419], [283, 418]], [[86, 318], [664, 318], [664, 332], [86, 332]], [[65, 279], [665, 279], [665, 292], [65, 292]], [[458, 210], [584, 210], [584, 224], [458, 224]], [[313, 209], [372, 209], [372, 226], [313, 226]], [[164, 209], [225, 209], [225, 226], [164, 226]], [[505, 151], [539, 151], [539, 166], [505, 166]], [[266, 48], [483, 48], [483, 68], [266, 68]]]
+00088568.jpg	[[[341, 446], [371, 446], [371, 453], [341, 453]], [[58, 445], [117, 445], [117, 455], [58, 455]], [[552, 433], [571, 433], [571, 440], [552, 440]], [[583, 431], [740, 431], [740, 442], [583, 442]], [[311, 415], [743, 415], [743, 428], [311, 428]], [[311, 377], [736, 377], [736, 390], [311, 390]], [[425, 340], [551, 340], [551, 350], [425, 350]], [[287, 324], [294, 332], [289, 337], [281, 330]], [[276, 294], [348, 296], [347, 311], [276, 309]], [[54, 288], [210, 288], [210, 301], [54, 301]], [[275, 265], [421, 265], [421, 278], [275, 278]], [[56, 264], [248, 264], [248, 277], [56, 277]], [[671, 248], [695, 248], [695, 261], [671, 261]], [[602, 248], [628, 248], [628, 261], [602, 261]], [[533, 248], [557, 248], [557, 261], [533, 261]], [[463, 248], [487, 248], [487, 261], [463, 261]], [[278, 248], [309, 248], [309, 260], [278, 260]], [[55, 240], [142, 240], [142, 254], [55, 254]], [[277, 231], [398, 231], [398, 244], [277, 244]], [[741, 228], [749, 237], [742, 245], [733, 236]], [[665, 230], [700, 230], [700, 244], [665, 244]], [[598, 230], [631, 230], [631, 244], [598, 244]], [[528, 230], [562, 230], [562, 244], [528, 244]], [[459, 230], [492, 230], [492, 244], [459, 244]], [[54, 215], [211, 217], [211, 231], [54, 229]], [[739, 211], [749, 221], [740, 229], [731, 220]], [[663, 214], [704, 214], [704, 228], [663, 228]], [[595, 215], [637, 215], [637, 226], [595, 226]], [[524, 215], [568, 215], [568, 226], [524, 226]], [[454, 215], [495, 215], [495, 226], [454, 226]], [[279, 215], [351, 215], [351, 226], [279, 226]], [[736, 199], [747, 199], [747, 208], [736, 208]], [[668, 197], [700, 197], [700, 208], [668, 208]], [[599, 196], [633, 196], [633, 210], [599, 210]], [[529, 197], [562, 197], [562, 208], [529, 208]], [[461, 197], [491, 197], [491, 208], [461, 208]], [[277, 195], [417, 196], [417, 211], [277, 209]], [[55, 192], [239, 192], [239, 205], [55, 205]], [[665, 181], [703, 181], [703, 192], [665, 192]], [[279, 180], [351, 181], [350, 192], [279, 191]], [[734, 180], [747, 180], [747, 193], [734, 193]], [[597, 180], [634, 180], [634, 191], [597, 191]], [[525, 179], [566, 179], [566, 193], [525, 193]], [[458, 180], [493, 180], [493, 191], [458, 191]], [[55, 170], [142, 170], [142, 184], [55, 184]], [[735, 165], [747, 165], [747, 175], [735, 175]], [[665, 163], [703, 163], [703, 175], [665, 175]], [[598, 163], [634, 163], [634, 175], [598, 175]], [[527, 163], [565, 163], [565, 175], [527, 175]], [[458, 163], [492, 163], [492, 175], [458, 175]], [[279, 162], [398, 162], [398, 176], [279, 176]], [[54, 146], [148, 146], [148, 159], [54, 159]], [[453, 147], [495, 147], [495, 158], [453, 158]], [[731, 143], [748, 146], [745, 161], [728, 158]], [[663, 145], [704, 145], [704, 159], [663, 159]], [[596, 146], [635, 146], [635, 157], [596, 157]], [[522, 145], [566, 142], [567, 157], [523, 159]], [[277, 144], [310, 144], [310, 158], [277, 158]], [[276, 121], [428, 121], [428, 139], [276, 139]], [[52, 120], [232, 121], [232, 139], [52, 138]], [[404, 91], [701, 91], [701, 106], [404, 106]], [[48, 79], [280, 79], [280, 97], [48, 97]], [[325, 69], [744, 70], [744, 84], [325, 83]], [[668, 48], [743, 48], [743, 63], [668, 63]], [[297, 48], [433, 48], [433, 62], [297, 62]]]
+00091741.jpg	[[[47, 336], [83, 336], [83, 358], [47, 358]], [[98, 211], [257, 209], [257, 229], [98, 231]], [[103, 190], [257, 191], [257, 205], [103, 204]], [[89, 101], [266, 99], [267, 181], [90, 184]], [[94, 48], [262, 55], [260, 114], [91, 107]], [[91, 12], [257, 14], [257, 37], [90, 35]]]
+00105313.jpg	[[[291, 262], [406, 262], [406, 275], [291, 275]], [[153, 262], [264, 262], [264, 274], [153, 274]], [[11, 258], [73, 261], [72, 274], [11, 272]], [[33, 231], [132, 231], [132, 244], [33, 244]], [[35, 217], [216, 217], [216, 227], [35, 227]], [[33, 200], [146, 200], [146, 213], [33, 213]], [[32, 183], [215, 184], [215, 197], [32, 196]], [[35, 170], [105, 170], [105, 181], [35, 181]], [[35, 155], [124, 155], [124, 164], [35, 164]], [[34, 137], [142, 138], [142, 149], [34, 148]], [[35, 123], [176, 123], [176, 133], [35, 133]], [[33, 106], [176, 106], [176, 119], [33, 119]], [[34, 92], [102, 92], [102, 102], [34, 102]], [[34, 77], [119, 77], [119, 87], [34, 87]], [[32, 60], [120, 60], [120, 73], [32, 73]], [[35, 46], [119, 46], [119, 55], [35, 55]], [[32, 29], [142, 29], [142, 42], [32, 42]], [[25, 12], [147, 12], [147, 24], [25, 24]]]
+00134770.jpg	[[[388, 646], [456, 646], [456, 656], [388, 656]], [[407, 620], [484, 619], [485, 633], [408, 634]], [[112, 534], [270, 531], [270, 549], [113, 551]], [[111, 502], [443, 497], [443, 514], [112, 519]], [[111, 471], [443, 467], [443, 484], [112, 488]], [[111, 439], [444, 434], [444, 451], [112, 457]], [[111, 409], [442, 405], [442, 421], [112, 425]], [[153, 376], [441, 373], [441, 390], [153, 394]], [[184, 338], [369, 336], [369, 356], [185, 358]], [[75, 98], [515, 104], [513, 218], [74, 212]]]
+00145943.jpg	[[[394, 248], [746, 279], [731, 449], [379, 418]], [[90, 92], [300, 92], [300, 119], [90, 119]], [[46, 41], [326, 39], [326, 75], [46, 77]]]
+00147605.jpg	[[[805, 616], [874, 616], [874, 627], [805, 627]], [[516, 607], [784, 605], [784, 628], [516, 629]], [[118, 522], [224, 522], [224, 560], [118, 560]], [[253, 524], [307, 524], [307, 557], [253, 557]], [[715, 501], [900, 505], [900, 538], [714, 534]], [[255, 502], [295, 502], [295, 517], [255, 517]], [[347, 481], [473, 481], [473, 515], [347, 515]], [[252, 484], [295, 484], [295, 499], [252, 499]], [[350, 456], [447, 456], [447, 470], [350, 470]], [[145, 444], [201, 444], [201, 467], [145, 467]], [[728, 371], [878, 371], [878, 420], [728, 420]], [[528, 369], [681, 369], [681, 418], [528, 418]], [[143, 369], [488, 369], [488, 420], [143, 420]], [[744, 315], [871, 315], [871, 336], [744, 336]], [[799, 157], [886, 154], [887, 188], [800, 191]], [[274, 142], [455, 142], [455, 160], [274, 160]], [[738, 116], [894, 119], [893, 157], [737, 153]], [[108, 112], [204, 112], [204, 130], [108, 130]], [[270, 92], [463, 96], [462, 132], [270, 129]]]
+00150341.jpg	[[[100, 644], [297, 644], [297, 661], [100, 661]], [[115, 617], [288, 617], [288, 631], [115, 631]], [[84, 593], [319, 592], [319, 609], [84, 610]], [[31, 565], [313, 562], [314, 580], [31, 582]], [[444, 560], [461, 560], [461, 569], [444, 569]], [[390, 557], [446, 557], [446, 572], [390, 572]], [[31, 515], [168, 515], [168, 529], [31, 529]], [[33, 490], [110, 490], [110, 504], [33, 504]], [[358, 459], [464, 463], [463, 485], [357, 481]], [[28, 459], [268, 460], [268, 481], [28, 480]], [[339, 439], [421, 444], [421, 460], [338, 455]], [[65, 439], [143, 439], [143, 453], [65, 453]], [[207, 416], [292, 416], [292, 434], [207, 434]], [[319, 408], [441, 413], [440, 438], [318, 433]], [[44, 405], [175, 409], [174, 434], [43, 430]], [[31, 383], [137, 383], [137, 404], [31, 404]]]
+00150669.jpg	[[[649, 700], [681, 700], [681, 716], [649, 716]], [[517, 685], [549, 685], [549, 720], [517, 720]], [[651, 688], [678, 688], [678, 701], [651, 701]], [[862, 687], [876, 687], [876, 695], [862, 695]], [[922, 675], [938, 675], [938, 685], [922, 685]], [[785, 671], [807, 671], [807, 687], [785, 687]], [[592, 672], [606, 672], [606, 686], [592, 686]], [[722, 679], [732, 669], [742, 678], [731, 688]], [[651, 680], [667, 665], [681, 679], [666, 695]], [[273, 667], [422, 667], [422, 688], [273, 688]], [[47, 668], [108, 668], [108, 686], [47, 686]], [[136, 666], [203, 666], [203, 688], [136, 688]], [[782, 629], [810, 629], [810, 661], [782, 661]], [[645, 627], [685, 627], [685, 665], [645, 665]], [[516, 628], [548, 628], [548, 664], [516, 664]], [[655, 619], [672, 619], [672, 627], [655, 627]], [[598, 617], [605, 624], [599, 629], [592, 622]], [[523, 619], [540, 619], [540, 627], [523, 627]], [[858, 618], [868, 618], [868, 627], [858, 627]], [[727, 618], [735, 618], [735, 627], [727, 627]], [[919, 620], [932, 611], [942, 624], [929, 633]], [[786, 616], [805, 616], [805, 629], [786, 629]], [[373, 604], [420, 604], [420, 619], [373, 619]], [[85, 603], [215, 605], [214, 621], [84, 619]], [[48, 603], [71, 603], [71, 622], [48, 622]], [[788, 561], [806, 561], [806, 572], [788, 572]], [[923, 560], [935, 560], [935, 574], [923, 574]], [[856, 560], [869, 560], [869, 574], [856, 574]], [[62, 554], [410, 554], [410, 568], [62, 568]], [[63, 532], [116, 535], [115, 545], [62, 543]], [[859, 527], [868, 527], [868, 539], [859, 539]], [[925, 526], [934, 526], [934, 540], [925, 540]], [[794, 520], [807, 529], [798, 542], [785, 533]], [[526, 526], [535, 526], [535, 536], [526, 536]], [[262, 513], [395, 513], [395, 526], [262, 526]], [[122, 514], [245, 514], [245, 524], [122, 524]], [[49, 514], [119, 514], [119, 525], [49, 525]], [[755, 492], [828, 492], [828, 507], [755, 507]], [[638, 492], [710, 492], [710, 507], [638, 507]], [[519, 492], [592, 492], [592, 507], [519, 507]], [[85, 450], [123, 450], [123, 461], [85, 461]], [[220, 450], [236, 447], [238, 459], [223, 462]], [[683, 445], [868, 445], [868, 459], [683, 459]], [[562, 445], [666, 445], [666, 459], [562, 459]], [[491, 446], [544, 446], [544, 458], [491, 458]], [[183, 437], [208, 437], [208, 459], [183, 459]], [[52, 431], [72, 438], [64, 462], [44, 455]], [[224, 432], [276, 432], [276, 443], [224, 443]], [[88, 432], [144, 432], [144, 443], [88, 443]], [[506, 383], [616, 382], [616, 397], [506, 398]], [[702, 381], [758, 381], [758, 399], [702, 399]], [[308, 373], [364, 373], [364, 384], [308, 384]], [[92, 373], [167, 373], [167, 384], [92, 384]], [[688, 335], [820, 335], [820, 350], [688, 350]], [[498, 335], [657, 335], [657, 350], [498, 350]], [[208, 316], [244, 316], [244, 331], [208, 331]], [[499, 289], [641, 289], [641, 302], [499, 302]], [[671, 287], [801, 287], [801, 301], [671, 301]], [[670, 241], [816, 241], [816, 255], [670, 255]], [[497, 241], [643, 241], [643, 255], [497, 255]], [[670, 194], [815, 194], [815, 208], [670, 208]], [[498, 194], [643, 194], [643, 208], [498, 208]], [[670, 145], [815, 145], [815, 160], [670, 160]], [[499, 145], [645, 145], [645, 160], [499, 160]], [[489, 103], [546, 103], [546, 120], [489, 120]], [[56, 89], [95, 89], [95, 97], [56, 97]], [[845, 26], [887, 20], [889, 39], [848, 44]], [[26, 20], [700, 20], [700, 37], [26, 37]], [[898, 11], [996, 16], [995, 45], [896, 40]]]
+00152568.jpg	[[[3, 252], [284, 254], [284, 280], [3, 278]], [[196, 233], [254, 233], [254, 240], [196, 240]], [[49, 229], [90, 229], [90, 240], [49, 240]], [[200, 159], [281, 165], [276, 229], [195, 222]]]
+00155628.jpg	[[[149, 901], [503, 903], [503, 922], [149, 920]], [[520, 893], [561, 896], [560, 911], [519, 908]], [[61, 885], [81, 885], [81, 894], [61, 894]], [[150, 878], [503, 882], [503, 900], [149, 896]], [[524, 834], [640, 839], [639, 856], [524, 852]], [[70, 834], [185, 835], [185, 853], [69, 852]], [[246, 555], [466, 555], [466, 569], [246, 569]], [[308, 507], [403, 509], [403, 524], [308, 522]], [[244, 482], [459, 484], [459, 502], [244, 500]], [[252, 422], [459, 424], [458, 452], [251, 450]], [[195, 378], [517, 380], [516, 408], [195, 406]], [[474, 194], [624, 196], [624, 210], [473, 208]], [[73, 129], [641, 131], [641, 160], [73, 158]], [[483, 41], [597, 37], [599, 98], [486, 102]], [[68, 25], [135, 16], [139, 43], [72, 52]]]
+00173364.jpg	[[[8, 178], [56, 178], [56, 200], [8, 200]], [[137, 120], [194, 120], [194, 133], [137, 133]], [[39, 76], [86, 76], [86, 105], [39, 105]], [[249, 20], [310, 20], [310, 36], [249, 36]], [[21, 16], [104, 16], [104, 39], [21, 39]]]
+00175503.jpg	[[[43, 260], [500, 255], [501, 358], [44, 363]], [[52, 200], [349, 178], [354, 251], [58, 273]]]
+00193218.jpg	[[[283, 375], [410, 375], [410, 388], [283, 388]], [[172, 375], [221, 375], [221, 389], [172, 389]], [[110, 375], [161, 375], [161, 389], [110, 389]], [[276, 358], [357, 358], [357, 371], [276, 371]], [[171, 359], [220, 359], [220, 370], [171, 370]], [[409, 357], [492, 357], [492, 370], [409, 370]], [[26, 187], [62, 187], [62, 202], [26, 202]], [[501, 185], [557, 185], [557, 199], [501, 199]], [[381, 187], [420, 185], [421, 199], [382, 201]], [[284, 186], [310, 186], [310, 201], [284, 201]], [[174, 186], [196, 186], [196, 201], [174, 201]], [[499, 165], [540, 165], [540, 176], [499, 176]], [[381, 164], [409, 164], [409, 177], [381, 177]], [[262, 163], [302, 163], [302, 177], [262, 177]], [[176, 163], [230, 163], [230, 177], [176, 177]], [[26, 163], [79, 163], [79, 177], [26, 177]], [[387, 140], [488, 140], [488, 153], [387, 153]], [[28, 139], [131, 139], [131, 152], [28, 152]], [[443, 117], [537, 119], [537, 133], [443, 132]], [[346, 119], [405, 119], [405, 130], [346, 130]], [[261, 119], [303, 119], [303, 130], [261, 130]], [[30, 113], [228, 116], [228, 131], [30, 129]], [[131, 91], [394, 94], [394, 109], [131, 105]], [[562, 82], [583, 82], [583, 107], [562, 107]]]
+00195033.jpg	[[[488, 263], [533, 265], [532, 280], [487, 278]], [[126, 250], [192, 250], [192, 283], [126, 283]], [[338, 249], [362, 249], [362, 266], [338, 266]], [[319, 222], [380, 225], [380, 238], [319, 236]], [[431, 224], [450, 224], [450, 235], [431, 235]], [[365, 203], [538, 203], [538, 216], [365, 216]], [[89, 200], [146, 203], [146, 217], [89, 214]], [[329, 201], [354, 201], [354, 212], [329, 212]], [[371, 181], [449, 181], [449, 194], [371, 194]], [[329, 181], [352, 181], [352, 192], [329, 192]], [[96, 179], [240, 179], [240, 193], [96, 193]], [[456, 162], [555, 162], [555, 175], [456, 175]], [[129, 150], [287, 151], [287, 165], [129, 164]], [[36, 145], [73, 149], [72, 163], [35, 159]], [[527, 146], [552, 146], [552, 155], [527, 155]], [[102, 145], [120, 145], [120, 153], [102, 153]], [[371, 129], [503, 128], [503, 139], [371, 140]], [[99, 126], [193, 126], [193, 139], [99, 139]], [[322, 127], [337, 127], [337, 135], [322, 135]], [[37, 123], [77, 123], [77, 134], [37, 134]], [[324, 106], [337, 106], [337, 115], [324, 115]], [[309, 107], [315, 107], [315, 112], [309, 112]], [[372, 103], [501, 103], [501, 116], [372, 116]], [[349, 105], [360, 105], [360, 114], [349, 114]], [[38, 103], [80, 103], [80, 113], [38, 113]], [[99, 100], [205, 101], [205, 115], [99, 114]], [[306, 90], [317, 90], [317, 97], [306, 97]], [[347, 88], [362, 88], [362, 96], [347, 96]], [[321, 87], [340, 87], [340, 99], [321, 99]], [[358, 84], [513, 82], [513, 95], [358, 97]], [[41, 83], [89, 83], [89, 93], [41, 93]], [[94, 79], [241, 80], [241, 94], [94, 93]], [[313, 66], [394, 66], [394, 79], [313, 79]], [[242, 66], [288, 66], [288, 77], [242, 77]], [[185, 54], [220, 54], [220, 65], [185, 65]], [[469, 48], [547, 48], [547, 61], [469, 61]], [[423, 36], [436, 36], [436, 54], [423, 54]], [[465, 30], [551, 30], [551, 43], [465, 43]], [[207, 21], [329, 23], [328, 41], [207, 39]]]
+00208502.jpg	[[[247, 566], [282, 566], [282, 573], [247, 573]], [[558, 534], [629, 539], [627, 570], [556, 565]], [[205, 540], [284, 540], [284, 552], [205, 552]], [[143, 513], [189, 513], [189, 525], [143, 525]], [[249, 512], [307, 512], [307, 524], [249, 524]], [[44, 500], [118, 500], [118, 519], [44, 519]], [[467, 491], [556, 491], [556, 508], [467, 508]], [[667, 490], [678, 494], [675, 503], [664, 499]], [[788, 489], [794, 495], [789, 499], [783, 494]], [[726, 491], [737, 491], [737, 501], [726, 501]], [[42, 452], [117, 450], [117, 469], [42, 470]], [[175, 450], [236, 450], [236, 464], [175, 464]], [[614, 407], [638, 407], [638, 422], [614, 422]], [[95, 405], [119, 405], [119, 422], [95, 422]], [[49, 399], [64, 414], [50, 427], [36, 413]], [[209, 401], [226, 401], [226, 415], [209, 415]], [[40, 357], [58, 357], [58, 374], [40, 374]], [[94, 356], [119, 356], [119, 373], [94, 373]], [[188, 341], [246, 339], [247, 361], [189, 364]], [[459, 321], [549, 319], [549, 337], [460, 339]], [[459, 273], [551, 273], [551, 290], [459, 290]], [[563, 272], [735, 269], [735, 286], [564, 289]], [[517, 225], [547, 225], [547, 245], [517, 245]], [[459, 226], [480, 226], [480, 244], [459, 244]], [[621, 187], [673, 187], [673, 201], [621, 201]], [[457, 132], [548, 130], [548, 147], [458, 149]], [[572, 106], [787, 99], [787, 120], [573, 126]], [[122, 48], [290, 48], [290, 97], [122, 97]], [[539, 39], [708, 39], [708, 89], [539, 89]]]
+00224225.jpg	[[[134, 429], [153, 426], [157, 445], [138, 448]], [[202, 404], [478, 411], [476, 459], [201, 452]], [[205, 230], [469, 230], [469, 390], [205, 390]], [[131, 265], [172, 265], [172, 279], [131, 279]], [[345, 207], [456, 207], [456, 231], [345, 231]], [[199, 189], [346, 196], [344, 239], [197, 232]], [[10, 44], [157, 41], [158, 112], [11, 115]]]
+00227746.jpg	[[[190, 232], [258, 232], [258, 238], [190, 238]], [[160, 232], [183, 232], [183, 238], [160, 238]], [[123, 232], [150, 232], [150, 238], [123, 238]], [[290, 208], [345, 206], [346, 222], [291, 224]], [[172, 181], [249, 181], [249, 194], [172, 194]], [[143, 178], [165, 180], [162, 208], [140, 206]], [[142, 164], [157, 162], [160, 177], [145, 180]], [[173, 157], [203, 157], [203, 164], [173, 164]], [[200, 154], [347, 154], [347, 167], [200, 167]], [[144, 111], [277, 114], [277, 134], [144, 131]], [[201, 52], [387, 53], [386, 69], [201, 68]], [[141, 46], [192, 46], [192, 63], [141, 63]], [[40, 26], [61, 26], [61, 42], [40, 42]]]
+00229605.jpg	[[[743, 529], [881, 529], [881, 544], [743, 544]], [[236, 499], [589, 498], [589, 522], [236, 523]], [[6, 498], [227, 498], [227, 522], [6, 522]], [[735, 496], [883, 499], [883, 520], [734, 517]], [[606, 495], [716, 489], [718, 515], [608, 521]], [[4, 245], [863, 230], [864, 288], [5, 303]], [[478, 28], [883, 28], [883, 76], [478, 76]]]
+00233011.jpg	[[[63, 227], [291, 227], [291, 242], [63, 242]], [[12, 219], [41, 219], [41, 250], [12, 250]], [[61, 177], [119, 177], [119, 195], [61, 195]], [[11, 173], [40, 169], [44, 200], [14, 203]], [[61, 129], [147, 131], [147, 147], [61, 144]], [[12, 124], [43, 124], [43, 154], [12, 154]], [[125, 89], [238, 89], [238, 103], [125, 103]], [[148, 51], [216, 51], [216, 65], [148, 65]], [[258, 46], [353, 50], [352, 69], [257, 65]], [[9, 49], [52, 49], [52, 68], [9, 68]], [[277, 12], [345, 12], [345, 31], [277, 31]], [[28, 11], [74, 11], [74, 31], [28, 31]]]
+00233625.jpg	[[[375, 397], [632, 399], [632, 443], [375, 440]], [[71, 214], [932, 207], [933, 321], [71, 328]]]
+00233634.jpg	[[[215, 639], [261, 639], [261, 703], [215, 703]], [[523, 635], [570, 635], [570, 695], [523, 695]], [[643, 523], [682, 523], [682, 568], [643, 568]], [[97, 516], [152, 516], [152, 589], [97, 589]], [[755, 395], [760, 395], [760, 401], [755, 401]], [[26, 395], [32, 395], [32, 400], [26, 400]], [[678, 364], [728, 362], [731, 430], [681, 432]], [[54, 361], [107, 361], [107, 434], [54, 434]], [[78, 208], [155, 208], [155, 280], [78, 280]], [[643, 205], [693, 205], [693, 272], [643, 272]], [[210, 88], [260, 86], [263, 164], [213, 166]], [[363, 48], [426, 45], [430, 115], [367, 118]]]
+00234400.jpg	[[[446, 421], [738, 421], [738, 438], [446, 438]], [[157, 421], [454, 421], [454, 438], [157, 438]], [[158, 394], [652, 394], [652, 411], [158, 411]], [[40, 391], [127, 391], [127, 412], [40, 412]], [[158, 342], [304, 345], [304, 363], [158, 360]], [[38, 344], [123, 344], [123, 362], [38, 362]], [[520, 295], [703, 295], [703, 314], [520, 314]], [[394, 292], [483, 290], [484, 314], [394, 317]], [[157, 293], [270, 293], [270, 313], [157, 313]], [[37, 293], [125, 293], [125, 313], [37, 313]], [[156, 243], [358, 243], [358, 267], [156, 267]], [[36, 243], [82, 243], [82, 269], [36, 269]], [[29, 152], [158, 152], [158, 175], [29, 175]], [[282, 98], [507, 98], [507, 111], [282, 111]], [[315, 46], [475, 50], [474, 88], [314, 85]], [[518, 51], [663, 53], [662, 67], [517, 65]], [[487, 19], [706, 17], [706, 43], [487, 45]]]
+00234883.jpg	[[[344, 145], [355, 145], [355, 153], [344, 153]], [[66, 125], [316, 120], [317, 190], [67, 195]], [[79, 138], [109, 141], [108, 152], [78, 148]], [[72, 120], [120, 120], [120, 130], [72, 130]], [[383, 63], [504, 62], [504, 74], [383, 75]], [[58, 29], [365, 26], [366, 112], [59, 115]], [[387, 28], [501, 26], [501, 45], [387, 47]]]
+test_add_0.jpg	[[[311, 521], [391, 521], [391, 534], [311, 534]], [[277, 500], [424, 500], [424, 514], [277, 514]], [[261, 446], [437, 446], [437, 459], [261, 459]], [[212, 428], [485, 428], [485, 441], [212, 441]], [[247, 388], [457, 388], [457, 409], [247, 409]], [[222, 328], [474, 328], [474, 372], [222, 372]], [[208, 207], [492, 211], [490, 277], [207, 272]], [[266, 164], [422, 166], [421, 197], [265, 195]], [[18, 20], [201, 18], [201, 43], [18, 45]]]
+test_add_1.png	[]
+test_add_10.png	[[[157, 124], [186, 124], [186, 172], [157, 172]], [[65, 117], [95, 117], [95, 168], [65, 168]], [[161, 106], [183, 106], [183, 127], [161, 127]], [[69, 100], [94, 100], [94, 128], [69, 128]], [[117, 46], [154, 45], [157, 174], [121, 175]], [[66, 34], [97, 34], [97, 112], [66, 112]]]
+test_add_11.jpg	[[[1525, 773], [1564, 756], [1575, 780], [1536, 798]], [[1390, 757], [1483, 757], [1483, 791], [1390, 791]], [[1013, 754], [1207, 754], [1207, 800], [1013, 800]], [[685, 755], [875, 755], [875, 796], [685, 796]], [[356, 753], [566, 747], [567, 793], [358, 798]], [[78, 751], [264, 745], [265, 793], [79, 798]], [[602, 647], [1152, 647], [1152, 703], [602, 703]], [[601, 564], [1148, 555], [1149, 611], [602, 620]], [[598, 480], [1066, 472], [1067, 526], [599, 535]], [[598, 393], [1090, 388], [1090, 439], [599, 444]], [[603, 306], [1057, 306], [1057, 357], [603, 357]], [[357, 184], [1517, 184], [1517, 261], [357, 261]], [[60, 43], [257, 37], [259, 83], [61, 89]], [[1305, 41], [1492, 41], [1492, 87], [1305, 87]], [[973, 40], [1171, 34], [1172, 80], [974, 86]], [[670, 40], [862, 34], [864, 80], [671, 86]], [[363, 34], [558, 34], [558, 85], [363, 85]]]
+test_add_12.jpg	[[[11, 592], [136, 594], [135, 613], [11, 611]], [[109, 521], [907, 526], [907, 569], [109, 565]], [[635, 451], [902, 448], [903, 478], [635, 481]], [[112, 447], [466, 449], [466, 486], [112, 483]], [[582, 306], [680, 304], [681, 348], [583, 351]], [[369, 261], [565, 266], [563, 357], [367, 353]], [[64, 85], [853, 88], [853, 161], [64, 159]]]
+test_add_13.jpg	[[[68, 94], [117, 97], [116, 114], [67, 111]]]
+test_add_14.jpg	[[[30, 97], [235, 95], [236, 127], [31, 129]], [[30, 52], [239, 50], [239, 86], [30, 87]]]
+test_add_15.jpg	[[[141, 253], [353, 253], [353, 266], [141, 266]], [[205, 214], [406, 219], [406, 232], [204, 227]], [[106, 212], [193, 213], [193, 227], [106, 226]], [[154, 156], [286, 161], [286, 174], [154, 170]], [[148, 136], [305, 142], [305, 156], [147, 150]], [[108, 137], [144, 137], [144, 148], [108, 148]], [[108, 102], [275, 109], [275, 125], [107, 117]], [[107, 72], [245, 79], [245, 96], [106, 88]], [[107, 39], [209, 42], [209, 62], [106, 59]]]
+test_add_16.jpg	[[[398, 842], [408, 842], [408, 852], [398, 852]], [[382, 742], [746, 742], [746, 776], [382, 776]], [[362, 703], [468, 703], [468, 725], [362, 725]], [[1552, 701], [1576, 701], [1576, 746], [1552, 746]], [[1256, 695], [1442, 695], [1442, 721], [1256, 721]], [[1244, 661], [1448, 661], [1448, 687], [1244, 687]], [[386, 645], [668, 645], [668, 679], [386, 679]], [[1228, 625], [1470, 623], [1470, 651], [1228, 653]], [[360, 604], [580, 604], [580, 629], [360, 629]], [[1202, 592], [1494, 592], [1494, 617], [1202, 617]], [[1166, 556], [1530, 556], [1530, 582], [1166, 582]], [[380, 552], [638, 552], [638, 586], [380, 586]], [[356, 502], [516, 502], [516, 536], [356, 536]], [[774, 260], [1124, 260], [1124, 300], [774, 300]], [[374, 210], [504, 210], [504, 300], [374, 300]], [[776, 212], [1088, 217], [1088, 252], [776, 248]]]
+test_add_17.jpg	[[[321, 255], [393, 258], [392, 271], [320, 269]], [[307, 222], [411, 228], [411, 241], [306, 236]], [[96, 136], [385, 143], [384, 208], [94, 201]], [[72, 95], [399, 103], [398, 124], [71, 117]], [[68, 76], [224, 79], [223, 93], [67, 90]], [[66, 59], [226, 62], [225, 76], [65, 74]]]
+test_add_18.jpg	[[[466, 788], [715, 790], [715, 813], [466, 811]], [[553, 752], [665, 757], [663, 791], [552, 786]], [[119, 539], [189, 539], [189, 570], [119, 570]], [[116, 473], [674, 486], [673, 528], [115, 516]], [[121, 429], [669, 441], [668, 470], [121, 457]], [[121, 375], [674, 381], [674, 410], [121, 404]], [[556, 262], [675, 264], [675, 278], [556, 277]], [[164, 259], [334, 259], [334, 273], [164, 273]], [[344, 195], [456, 197], [455, 220], [343, 217]], [[309, 175], [490, 175], [490, 190], [309, 190]], [[255, 128], [537, 131], [537, 169], [254, 165]], [[347, 92], [486, 94], [486, 109], [346, 107]], [[285, 41], [567, 49], [566, 82], [284, 74]], [[236, 32], [266, 32], [266, 60], [236, 60]]]
+test_add_19.jpg	[[[24, 294], [42, 294], [42, 302], [24, 302]], [[64, 293], [105, 293], [105, 303], [64, 303]], [[145, 287], [163, 287], [163, 304], [145, 304]], [[63, 280], [106, 280], [106, 290], [63, 290]], [[9, 281], [26, 281], [26, 288], [9, 288]], [[220, 279], [245, 279], [245, 291], [220, 291]], [[177, 279], [208, 279], [208, 290], [177, 290]], [[23, 279], [51, 279], [51, 290], [23, 290]], [[145, 278], [162, 278], [162, 292], [145, 292]], [[8, 267], [18, 267], [18, 276], [8, 276]], [[221, 265], [243, 265], [243, 277], [221, 277]], [[24, 265], [47, 265], [47, 277], [24, 277]], [[142, 263], [163, 263], [163, 279], [142, 279]], [[218, 252], [249, 252], [249, 265], [218, 265]], [[65, 253], [131, 253], [131, 263], [65, 263]], [[24, 252], [43, 252], [43, 264], [24, 264]], [[8, 253], [18, 253], [18, 262], [8, 262]], [[8, 240], [17, 240], [17, 249], [8, 249]], [[63, 237], [114, 237], [114, 251], [63, 251]], [[25, 236], [47, 239], [45, 251], [23, 249]], [[144, 234], [166, 237], [163, 253], [142, 249]], [[494, 226], [531, 226], [531, 239], [494, 239]], [[335, 226], [354, 226], [354, 237], [335, 237]], [[288, 226], [314, 226], [314, 237], [288, 237]], [[63, 226], [113, 226], [113, 236], [63, 236]], [[7, 227], [17, 227], [17, 234], [7, 234]], [[221, 225], [248, 225], [248, 235], [221, 235]], [[143, 225], [165, 222], [167, 234], [145, 237]], [[24, 224], [48, 224], [48, 238], [24, 238]], [[495, 213], [524, 213], [524, 224], [495, 224]], [[420, 212], [437, 212], [437, 225], [420, 225]], [[336, 212], [398, 212], [398, 223], [336, 223]], [[292, 212], [320, 212], [320, 223], [292, 223]], [[222, 212], [249, 212], [249, 223], [222, 223]], [[145, 212], [166, 212], [166, 223], [145, 223]], [[61, 211], [113, 209], [114, 222], [62, 224]], [[26, 211], [48, 211], [48, 223], [26, 223]], [[337, 199], [383, 199], [383, 209], [337, 209]], [[65, 200], [87, 200], [87, 207], [65, 207]], [[493, 197], [541, 197], [541, 211], [493, 211]], [[445, 202], [455, 196], [462, 206], [452, 212]], [[178, 198], [205, 198], [205, 208], [178, 208]], [[146, 199], [157, 199], [157, 208], [146, 208]], [[32, 194], [43, 204], [33, 214], [22, 203]], [[422, 193], [440, 201], [432, 215], [415, 207]], [[65, 186], [132, 186], [132, 196], [65, 196]], [[337, 185], [399, 185], [399, 196], [337, 196]], [[445, 190], [456, 182], [465, 191], [454, 200]], [[292, 188], [308, 182], [313, 193], [297, 200]], [[220, 183], [255, 183], [255, 197], [220, 197]], [[142, 184], [158, 184], [158, 197], [142, 197]], [[493, 182], [518, 182], [518, 197], [493, 197]], [[425, 180], [437, 191], [427, 202], [414, 190]], [[32, 179], [42, 189], [32, 199], [22, 189]], [[182, 179], [195, 185], [188, 198], [175, 192]], [[335, 172], [400, 169], [400, 183], [336, 185]], [[492, 170], [519, 170], [519, 185], [492, 185]], [[412, 177], [428, 164], [440, 178], [425, 190]], [[293, 171], [315, 171], [315, 185], [293, 185]], [[220, 170], [251, 170], [251, 184], [220, 184]], [[178, 172], [188, 172], [188, 183], [178, 183]], [[64, 172], [125, 170], [125, 181], [64, 182]], [[454, 168], [464, 176], [454, 185], [445, 176]], [[142, 172], [159, 168], [163, 180], [145, 185]], [[30, 165], [43, 174], [34, 186], [20, 177]], [[493, 160], [523, 160], [523, 170], [493, 170]], [[402, 161], [435, 161], [435, 168], [402, 168]], [[335, 159], [401, 159], [401, 169], [335, 169]], [[296, 159], [325, 159], [325, 170], [296, 170]], [[221, 158], [251, 158], [251, 169], [221, 169]], [[174, 161], [183, 156], [190, 167], [181, 172]], [[145, 158], [162, 158], [162, 170], [145, 170]], [[61, 158], [125, 157], [125, 168], [62, 169]], [[20, 161], [33, 154], [40, 167], [28, 174]], [[492, 143], [542, 143], [542, 157], [492, 157]], [[450, 144], [479, 144], [479, 157], [450, 157]], [[335, 143], [439, 143], [439, 156], [335, 156]], [[294, 143], [327, 143], [327, 157], [294, 157]], [[220, 143], [253, 143], [253, 157], [220, 157]], [[178, 145], [187, 145], [187, 156], [178, 156]], [[63, 144], [104, 144], [104, 155], [63, 155]], [[144, 140], [164, 145], [160, 159], [141, 154]], [[31, 137], [44, 149], [31, 162], [17, 149]], [[286, 135], [291, 135], [291, 140], [286, 140]], [[177, 133], [193, 133], [193, 144], [177, 144]], [[336, 132], [388, 132], [388, 141], [336, 141]], [[492, 131], [525, 131], [525, 141], [492, 141]], [[450, 131], [477, 131], [477, 141], [450, 141]], [[292, 131], [321, 131], [321, 141], [292, 141]], [[218, 132], [255, 130], [256, 141], [219, 144]], [[63, 131], [95, 131], [95, 141], [63, 141]], [[417, 130], [437, 130], [437, 141], [417, 141]], [[145, 130], [159, 130], [159, 143], [145, 143]], [[30, 124], [43, 133], [32, 147], [19, 138]], [[493, 118], [535, 118], [535, 129], [493, 129]], [[336, 118], [388, 118], [388, 129], [336, 129]], [[218, 118], [255, 118], [255, 128], [218, 128]], [[451, 117], [478, 117], [478, 129], [451, 129]], [[418, 117], [438, 117], [438, 130], [418, 130]], [[177, 116], [209, 116], [209, 130], [177, 130]], [[145, 117], [162, 117], [162, 130], [145, 130]], [[62, 116], [88, 116], [88, 131], [62, 131]], [[19, 121], [33, 111], [43, 124], [29, 134]], [[491, 107], [523, 107], [523, 113], [491, 113]], [[449, 107], [477, 107], [477, 113], [449, 113]], [[420, 107], [436, 107], [436, 113], [420, 113]], [[295, 107], [319, 107], [319, 114], [295, 114]], [[220, 107], [242, 107], [242, 113], [220, 113]], [[176, 107], [203, 107], [203, 113], [176, 113]], [[145, 107], [161, 107], [161, 114], [145, 114]], [[334, 105], [372, 105], [372, 114], [334, 114]], [[63, 106], [86, 106], [86, 113], [63, 113]], [[483, 89], [522, 89], [522, 99], [483, 99]], [[331, 88], [380, 88], [380, 99], [331, 99]], [[276, 88], [325, 88], [325, 99], [276, 99]], [[214, 88], [246, 88], [246, 99], [214, 99]], [[411, 86], [474, 86], [474, 100], [411, 100]], [[6, 86], [102, 86], [102, 100], [6, 100]], [[415, 66], [461, 66], [461, 77], [415, 77]], [[288, 66], [333, 66], [333, 77], [288, 77]], [[157, 64], [206, 64], [206, 78], [157, 78]], [[416, 48], [523, 49], [523, 63], [415, 62]], [[288, 49], [375, 49], [375, 63], [288, 63]], [[159, 49], [269, 49], [269, 62], [159, 62]], [[24, 53], [36, 46], [45, 59], [33, 67]], [[416, 36], [481, 36], [481, 46], [416, 46]], [[25, 38], [39, 32], [46, 46], [33, 52]], [[157, 34], [205, 34], [205, 47], [157, 47]], [[412, 4], [527, 4], [527, 17], [412, 17]], [[146, 4], [345, 2], [345, 15], [146, 17]]]
+test_add_20.jpg	[[[31, 346], [605, 346], [605, 370], [31, 370]], [[217, 294], [510, 294], [510, 322], [217, 322]], [[473, 271], [525, 271], [525, 286], [473, 286]], [[220, 267], [287, 267], [287, 286], [220, 286]], [[219, 239], [484, 239], [484, 263], [219, 263]], [[221, 217], [303, 217], [303, 234], [221, 234]], [[402, 192], [417, 192], [417, 205], [402, 205]], [[222, 187], [341, 187], [341, 207], [222, 207]], [[221, 162], [287, 162], [287, 180], [221, 180]], [[375, 122], [475, 124], [475, 146], [375, 143]], [[222, 124], [356, 122], [356, 143], [222, 146]], [[218, 81], [352, 84], [352, 116], [218, 113]], [[440, 35], [605, 35], [605, 60], [440, 60]], [[72, 16], [398, 16], [398, 44], [72, 44]]]
+test_add_3.jpg	[[[169, 327], [337, 326], [337, 341], [169, 342]], [[170, 288], [307, 290], [307, 312], [170, 310]], [[171, 221], [323, 221], [323, 234], [171, 234]], [[340, 221], [449, 217], [449, 231], [341, 234]], [[169, 201], [372, 201], [372, 214], [169, 214]], [[170, 183], [418, 183], [418, 196], [170, 196]], [[170, 149], [416, 149], [416, 163], [170, 163]], [[171, 119], [418, 119], [418, 140], [171, 140]], [[326, 64], [478, 64], [478, 91], [326, 91]], [[173, 64], [306, 60], [306, 89], [174, 93]]]
+test_add_4.png	[]
+test_add_5.png	[[[48, 164], [108, 164], [108, 174], [48, 174]], [[52, 121], [169, 121], [169, 134], [52, 134]], [[50, 102], [165, 102], [165, 118], [50, 118]], [[52, 83], [164, 83], [164, 100], [52, 100]], [[51, 68], [166, 68], [166, 84], [51, 84]], [[51, 50], [145, 47], [145, 64], [52, 67]]]
+test_add_6.jpg	[[[123, 223], [219, 227], [218, 251], [122, 247]], [[172, 172], [186, 186], [172, 200], [158, 186]]]
+test_add_7.jpg	[[[48, 938], [174, 936], [174, 962], [48, 964]], [[227, 873], [629, 876], [628, 953], [226, 949]], [[56, 745], [638, 745], [638, 790], [56, 790]], [[150, 674], [545, 678], [544, 721], [150, 718]], [[73, 504], [633, 504], [633, 601], [73, 601]], [[59, 270], [655, 279], [652, 441], [56, 432]], [[513, 193], [553, 193], [553, 223], [513, 223]], [[61, 175], [532, 175], [532, 239], [61, 239]], [[533, 178], [642, 178], [642, 236], [533, 236]]]
+test_add_8.jpg	[[[251, 586], [454, 580], [454, 606], [252, 613]], [[107, 533], [457, 527], [457, 560], [108, 566]], [[336, 494], [384, 494], [384, 507], [336, 507]], [[27, 307], [355, 297], [356, 320], [28, 330]], [[22, 259], [445, 251], [445, 274], [23, 282]], [[78, 209], [445, 205], [445, 225], [78, 229]], [[160, 23], [319, 30], [317, 79], [158, 72]]]
+test_add_9.png	[[[266, 687], [486, 687], [486, 696], [266, 696]], [[196, 668], [554, 668], [554, 681], [196, 681]], [[154, 596], [597, 596], [597, 606], [154, 606]], [[215, 578], [541, 578], [541, 588], [215, 588]], [[134, 560], [615, 560], [615, 570], [134, 570]], [[85, 543], [665, 543], [665, 553], [85, 553]], [[96, 522], [653, 522], [653, 535], [96, 535]], [[362, 449], [389, 449], [389, 460], [362, 460]], [[238, 376], [513, 376], [513, 389], [238, 389]], [[177, 356], [574, 356], [574, 368], [177, 368]], [[344, 281], [408, 283], [407, 297], [343, 294]], [[257, 205], [493, 205], [493, 219], [257, 219]]]
diff --git a/test_tipc/results/python_ppocr_det_mobile_results_fp32.txt b/test_tipc/results/python_ppocr_det_mobile_results_fp32.txt
new file mode 100644
index 0000000..e370b49
--- /dev/null
+++ b/test_tipc/results/python_ppocr_det_mobile_results_fp32.txt
@@ -0,0 +1,49 @@
+00008790.jpg	[[[209, 406], [280, 406], [280, 419], [209, 419]], [[60, 398], [105, 398], [105, 411], [60, 411]], [[198, 389], [291, 389], [291, 402], [198, 402]], [[162, 391], [173, 391], [173, 401], [162, 401]], [[35, 380], [133, 380], [133, 393], [35, 393]], [[199, 371], [292, 371], [292, 384], [199, 384]], [[218, 310], [272, 310], [272, 324], [218, 324]], [[162, 305], [172, 305], [172, 314], [162, 314]], [[371, 302], [436, 302], [436, 316], [371, 316]], [[31, 302], [134, 301], [134, 315], [31, 316]], [[223, 292], [269, 292], [269, 306], [223, 306]], [[60, 225], [104, 225], [104, 236], [60, 236]], [[218, 223], [272, 223], [272, 237], [218, 237]], [[162, 219], [173, 219], [173, 227], [162, 227]], [[33, 207], [131, 207], [131, 220], [33, 220]], [[223, 206], [269, 206], [269, 220], [223, 220]], [[74, 146], [383, 146], [383, 159], [74, 159]], [[54, 120], [117, 120], [117, 134], [54, 134]], [[74, 51], [296, 51], [296, 65], [74, 65]], [[56, 17], [116, 17], [116, 31], [56, 31]]]
+00018946.jpg	[[[441, 328], [474, 328], [474, 339], [441, 339]], [[86, 284], [141, 286], [140, 307], [85, 305]], [[302, 279], [377, 279], [377, 297], [302, 297]], [[197, 265], [281, 274], [279, 293], [195, 284]], [[198, 197], [452, 219], [450, 242], [196, 220]], [[343, 182], [376, 182], [376, 192], [343, 192]], [[199, 164], [340, 171], [339, 192], [198, 185]], [[177, 101], [415, 118], [413, 145], [175, 128]]]
+00034387.jpg	[[[265, 460], [740, 460], [740, 484], [265, 484]], [[348, 417], [420, 417], [420, 443], [348, 443]], [[545, 418], [568, 418], [568, 442], [545, 442]], [[685, 417], [710, 417], [710, 443], [685, 443]], [[175, 415], [226, 415], [226, 443], [175, 443]], [[874, 414], [908, 414], [908, 446], [874, 446]], [[56, 417], [74, 417], [74, 442], [56, 442]], [[856, 373], [925, 373], [925, 400], [856, 400]], [[348, 372], [418, 372], [418, 397], [348, 397]], [[674, 372], [723, 372], [723, 401], [674, 401]], [[539, 373], [570, 373], [570, 400], [539, 400]], [[151, 365], [228, 369], [226, 402], [149, 398]], [[56, 372], [74, 372], [74, 397], [56, 397]], [[857, 329], [925, 329], [925, 355], [857, 355]], [[351, 330], [419, 330], [419, 356], [351, 356]], [[674, 328], [723, 328], [723, 356], [674, 356]], [[541, 329], [570, 329], [570, 357], [541, 357]], [[171, 327], [227, 324], [229, 355], [173, 358]], [[57, 330], [74, 330], [74, 356], [57, 356]], [[298, 327], [316, 327], [316, 334], [298, 334]], [[855, 286], [925, 286], [925, 312], [855, 312]], [[674, 286], [723, 286], [723, 313], [674, 313]], [[346, 286], [426, 283], [427, 313], [347, 316]], [[540, 285], [569, 285], [569, 312], [540, 312]], [[172, 282], [226, 282], [226, 313], [172, 313]], [[56, 287], [73, 287], [73, 312], [56, 312]], [[857, 242], [925, 242], [925, 268], [857, 268]], [[348, 242], [460, 242], [460, 268], [348, 268]], [[156, 242], [227, 242], [227, 269], [156, 269]], [[674, 241], [724, 241], [724, 269], [674, 269]], [[531, 241], [572, 241], [572, 270], [531, 270]], [[56, 242], [74, 242], [74, 268], [56, 268]], [[855, 197], [925, 200], [924, 226], [854, 224]], [[674, 198], [726, 198], [726, 226], [674, 226]], [[344, 200], [434, 195], [436, 223], [346, 228]], [[176, 197], [227, 197], [227, 225], [176, 225]], [[56, 200], [73, 200], [73, 226], [56, 226]], [[527, 194], [576, 194], [576, 226], [527, 226]], [[349, 155], [419, 155], [419, 181], [349, 181]], [[854, 154], [925, 154], [925, 180], [854, 180]], [[176, 154], [226, 154], [226, 183], [176, 183]], [[670, 153], [723, 153], [723, 181], [670, 181]], [[538, 154], [571, 154], [571, 182], [538, 182]], [[56, 156], [74, 156], [74, 182], [56, 182]], [[349, 111], [419, 111], [419, 137], [349, 137]], [[174, 111], [227, 111], [227, 139], [174, 139]], [[546, 113], [564, 113], [564, 137], [546, 137]], [[52, 112], [75, 112], [75, 139], [52, 139]], [[639, 108], [727, 105], [728, 138], [640, 141]], [[817, 103], [927, 110], [925, 139], [815, 132]], [[814, 68], [951, 68], [951, 92], [814, 92]], [[307, 66], [446, 68], [446, 93], [306, 90]], [[673, 67], [723, 67], [723, 93], [673, 93]], [[175, 65], [228, 68], [226, 95], [174, 92]], [[39, 65], [90, 68], [88, 97], [37, 94]], [[528, 65], [580, 65], [580, 94], [528, 94]], [[334, 20], [670, 20], [670, 43], [334, 43]]]
+00037951.jpg	[[[434, 976], [551, 978], [550, 993], [434, 991]], [[433, 932], [553, 932], [553, 969], [433, 969]], [[30, 522], [98, 522], [98, 545], [30, 545]], [[31, 443], [145, 443], [145, 464], [31, 464]], [[234, 335], [326, 332], [327, 354], [235, 356]], [[124, 252], [436, 252], [436, 284], [124, 284]], [[182, 206], [378, 206], [378, 227], [182, 227]], [[258, 106], [320, 123], [304, 181], [242, 163]], [[28, 65], [33, 65], [33, 71], [28, 71]], [[37, 58], [147, 58], [147, 80], [37, 80]]]
+00044782.jpg	[[[104, 218], [115, 218], [115, 227], [104, 227]], [[223, 216], [246, 216], [246, 228], [223, 228]], [[163, 216], [182, 216], [182, 229], [163, 229]], [[124, 191], [164, 191], [164, 202], [124, 202]], [[91, 84], [251, 84], [251, 98], [91, 98]], [[73, 63], [278, 63], [278, 78], [73, 78]], [[104, 15], [243, 15], [243, 44], [104, 44]]]
+00067516.jpg	[[[141, 808], [594, 809], [594, 822], [141, 821]], [[49, 784], [695, 784], [695, 798], [49, 798]], [[579, 751], [667, 751], [667, 764], [579, 764]], [[355, 750], [395, 750], [395, 767], [355, 767]], [[221, 751], [260, 751], [260, 765], [221, 765]], [[477, 750], [501, 750], [501, 768], [477, 768]], [[69, 748], [133, 751], [132, 765], [68, 761]], [[576, 682], [668, 682], [668, 699], [576, 699]], [[476, 682], [518, 682], [518, 700], [476, 700]], [[354, 682], [395, 682], [395, 700], [354, 700]], [[69, 681], [133, 684], [132, 699], [68, 695]], [[220, 679], [243, 682], [241, 700], [218, 697]], [[577, 615], [667, 615], [667, 632], [577, 632]], [[68, 612], [134, 615], [133, 632], [67, 629]], [[476, 614], [500, 614], [500, 633], [476, 633]], [[354, 613], [378, 613], [378, 634], [354, 634]], [[219, 612], [245, 612], [245, 633], [219, 633]], [[578, 547], [667, 547], [667, 564], [578, 564]], [[476, 546], [518, 546], [518, 565], [476, 565]], [[353, 545], [379, 545], [379, 566], [353, 566]], [[219, 545], [245, 545], [245, 566], [219, 566]], [[68, 542], [133, 546], [132, 563], [67, 560]], [[68, 478], [133, 482], [132, 499], [67, 496]], [[586, 481], [664, 481], [664, 497], [586, 497]], [[476, 480], [518, 480], [518, 498], [476, 498]], [[354, 480], [395, 480], [395, 498], [354, 498]], [[219, 479], [245, 479], [245, 500], [219, 500]], [[580, 425], [665, 429], [664, 449], [580, 446]], [[346, 429], [410, 429], [410, 447], [346, 447]], [[68, 426], [150, 429], [149, 449], [67, 447]], [[474, 427], [515, 427], [515, 449], [474, 449]], [[218, 427], [259, 427], [259, 449], [218, 449]], [[283, 398], [478, 399], [478, 419], [283, 418]], [[86, 318], [664, 318], [664, 332], [86, 332]], [[65, 279], [665, 279], [665, 292], [65, 292]], [[458, 210], [584, 210], [584, 224], [458, 224]], [[312, 209], [371, 209], [371, 226], [312, 226]], [[164, 209], [225, 209], [225, 226], [164, 226]], [[505, 151], [539, 151], [539, 166], [505, 166]], [[266, 48], [483, 48], [483, 68], [266, 68]]]
+00088568.jpg	[[[341, 446], [371, 446], [371, 453], [341, 453]], [[58, 445], [117, 445], [117, 455], [58, 455]], [[552, 433], [571, 433], [571, 440], [552, 440]], [[583, 431], [740, 431], [740, 442], [583, 442]], [[311, 415], [743, 415], [743, 428], [311, 428]], [[310, 376], [735, 376], [735, 389], [310, 389]], [[425, 340], [551, 340], [551, 350], [425, 350]], [[287, 324], [294, 332], [289, 337], [281, 330]], [[276, 294], [348, 296], [347, 311], [276, 309]], [[54, 288], [210, 288], [210, 301], [54, 301]], [[275, 265], [421, 265], [421, 278], [275, 278]], [[56, 264], [248, 264], [248, 277], [56, 277]], [[671, 248], [695, 248], [695, 261], [671, 261]], [[602, 248], [628, 248], [628, 261], [602, 261]], [[533, 248], [557, 248], [557, 261], [533, 261]], [[463, 248], [487, 248], [487, 261], [463, 261]], [[278, 248], [309, 248], [309, 260], [278, 260]], [[55, 240], [142, 240], [142, 254], [55, 254]], [[277, 231], [398, 231], [398, 244], [277, 244]], [[741, 228], [749, 237], [742, 245], [733, 236]], [[665, 230], [700, 230], [700, 244], [665, 244]], [[598, 230], [631, 230], [631, 244], [598, 244]], [[528, 230], [562, 230], [562, 244], [528, 244]], [[459, 230], [492, 230], [492, 244], [459, 244]], [[54, 215], [211, 217], [211, 231], [54, 229]], [[739, 211], [749, 221], [740, 229], [731, 220]], [[663, 214], [704, 214], [704, 228], [663, 228]], [[595, 215], [637, 215], [637, 226], [595, 226]], [[524, 215], [569, 215], [569, 226], [524, 226]], [[454, 215], [495, 215], [495, 226], [454, 226]], [[279, 215], [351, 215], [351, 226], [279, 226]], [[736, 199], [747, 199], [747, 208], [736, 208]], [[668, 197], [700, 197], [700, 208], [668, 208]], [[599, 196], [633, 196], [633, 210], [599, 210]], [[529, 197], [562, 197], [562, 208], [529, 208]], [[461, 197], [491, 197], [491, 208], [461, 208]], [[277, 195], [417, 196], [417, 211], [277, 209]], [[55, 192], [239, 192], [239, 205], [55, 205]], [[665, 181], [703, 181], [703, 192], [665, 192]], [[279, 180], [351, 181], [350, 192], [279, 191]], [[734, 180], [747, 180], [747, 193], [734, 193]], [[597, 180], [634, 180], [634, 191], [597, 191]], [[525, 179], [566, 179], [566, 193], [525, 193]], [[458, 180], [493, 180], [493, 191], [458, 191]], [[55, 170], [142, 170], [142, 184], [55, 184]], [[735, 165], [747, 165], [747, 175], [735, 175]], [[665, 163], [703, 163], [703, 175], [665, 175]], [[598, 163], [634, 163], [634, 175], [598, 175]], [[527, 163], [565, 163], [565, 175], [527, 175]], [[458, 163], [492, 163], [492, 175], [458, 175]], [[279, 162], [398, 162], [398, 176], [279, 176]], [[54, 146], [148, 146], [148, 159], [54, 159]], [[453, 147], [495, 147], [495, 158], [453, 158]], [[731, 143], [748, 146], [745, 161], [728, 158]], [[663, 145], [704, 145], [704, 159], [663, 159]], [[596, 146], [635, 146], [635, 157], [596, 157]], [[522, 145], [566, 142], [567, 157], [523, 159]], [[277, 144], [310, 144], [310, 158], [277, 158]], [[276, 121], [428, 121], [428, 139], [276, 139]], [[52, 120], [232, 121], [232, 139], [52, 138]], [[404, 91], [701, 91], [701, 106], [404, 106]], [[48, 79], [280, 79], [280, 97], [48, 97]], [[325, 69], [744, 70], [744, 84], [325, 83]], [[668, 48], [743, 48], [743, 63], [668, 63]], [[297, 48], [433, 48], [433, 62], [297, 62]]]
+00091741.jpg	[[[47, 336], [83, 336], [83, 358], [47, 358]], [[99, 211], [257, 211], [257, 230], [99, 230]], [[103, 190], [257, 191], [257, 205], [103, 204]], [[89, 101], [266, 99], [267, 181], [90, 184]], [[94, 48], [262, 55], [260, 114], [91, 107]], [[91, 12], [257, 14], [257, 37], [90, 35]]]
+00105313.jpg	[[[291, 262], [406, 262], [406, 275], [291, 275]], [[153, 262], [264, 262], [264, 274], [153, 274]], [[11, 258], [73, 261], [72, 274], [11, 272]], [[33, 231], [132, 231], [132, 244], [33, 244]], [[35, 217], [216, 217], [216, 227], [35, 227]], [[32, 200], [145, 200], [145, 213], [32, 213]], [[32, 183], [215, 184], [215, 197], [32, 196]], [[35, 170], [105, 170], [105, 181], [35, 181]], [[35, 155], [124, 155], [124, 164], [35, 164]], [[34, 137], [142, 138], [142, 149], [34, 148]], [[35, 123], [176, 123], [176, 133], [35, 133]], [[33, 106], [176, 106], [176, 119], [33, 119]], [[34, 92], [102, 92], [102, 102], [34, 102]], [[34, 77], [119, 77], [119, 87], [34, 87]], [[32, 60], [120, 60], [120, 73], [32, 73]], [[35, 46], [119, 46], [119, 55], [35, 55]], [[32, 29], [142, 29], [142, 42], [32, 42]], [[25, 12], [147, 12], [147, 24], [25, 24]]]
+00134770.jpg	[[[388, 645], [456, 645], [456, 655], [388, 655]], [[407, 620], [484, 619], [485, 633], [408, 634]], [[112, 534], [270, 531], [270, 549], [113, 551]], [[111, 502], [443, 497], [443, 514], [112, 519]], [[111, 471], [443, 467], [443, 484], [112, 488]], [[111, 439], [444, 434], [444, 451], [112, 457]], [[111, 409], [442, 405], [442, 421], [112, 425]], [[153, 376], [441, 373], [441, 390], [153, 394]], [[184, 338], [369, 336], [369, 356], [185, 358]], [[75, 98], [515, 104], [513, 218], [74, 212]]]
+00145943.jpg	[[[394, 248], [746, 279], [731, 449], [379, 418]], [[90, 92], [300, 92], [300, 119], [90, 119]], [[46, 41], [326, 39], [326, 75], [46, 77]]]
+00147605.jpg	[[[804, 615], [874, 616], [874, 627], [804, 626]], [[516, 607], [784, 605], [784, 628], [516, 629]], [[118, 522], [224, 522], [224, 560], [118, 560]], [[253, 524], [307, 524], [307, 557], [253, 557]], [[715, 501], [900, 505], [900, 538], [714, 534]], [[255, 502], [295, 502], [295, 517], [255, 517]], [[347, 481], [473, 481], [473, 515], [347, 515]], [[252, 484], [295, 484], [295, 499], [252, 499]], [[350, 456], [447, 456], [447, 470], [350, 470]], [[145, 444], [201, 444], [201, 467], [145, 467]], [[728, 371], [878, 371], [878, 420], [728, 420]], [[528, 369], [681, 369], [681, 418], [528, 418]], [[142, 368], [486, 368], [486, 419], [142, 419]], [[744, 315], [871, 315], [871, 336], [744, 336]], [[799, 157], [886, 154], [887, 188], [800, 191]], [[274, 142], [455, 142], [455, 160], [274, 160]], [[738, 116], [894, 119], [893, 157], [737, 153]], [[108, 112], [204, 112], [204, 130], [108, 130]], [[270, 92], [463, 96], [462, 132], [270, 129]]]
+00150341.jpg	[[[100, 645], [298, 645], [298, 662], [100, 662]], [[115, 617], [288, 617], [288, 631], [115, 631]], [[84, 593], [319, 592], [319, 609], [84, 610]], [[31, 565], [313, 562], [314, 580], [31, 582]], [[444, 560], [461, 560], [461, 569], [444, 569]], [[390, 557], [446, 557], [446, 572], [390, 572]], [[31, 515], [168, 515], [168, 529], [31, 529]], [[33, 490], [110, 490], [110, 504], [33, 504]], [[358, 459], [464, 463], [463, 485], [357, 481]], [[28, 459], [268, 460], [268, 481], [28, 480]], [[339, 439], [421, 444], [421, 460], [338, 455]], [[64, 439], [143, 439], [143, 453], [64, 453]], [[207, 416], [292, 416], [292, 434], [207, 434]], [[319, 408], [441, 413], [440, 438], [318, 433]], [[44, 405], [175, 409], [174, 434], [43, 430]], [[31, 383], [137, 383], [137, 404], [31, 404]]]
+00150669.jpg	[[[649, 700], [681, 700], [681, 716], [649, 716]], [[517, 685], [549, 685], [549, 720], [517, 720]], [[651, 688], [678, 688], [678, 701], [651, 701]], [[862, 687], [876, 687], [876, 695], [862, 695]], [[922, 675], [938, 675], [938, 685], [922, 685]], [[785, 671], [807, 671], [807, 687], [785, 687]], [[592, 672], [606, 672], [606, 686], [592, 686]], [[722, 679], [732, 669], [742, 678], [731, 688]], [[651, 680], [667, 665], [681, 679], [666, 695]], [[273, 667], [422, 667], [422, 688], [273, 688]], [[136, 666], [203, 666], [203, 688], [136, 688]], [[46, 666], [109, 666], [109, 687], [46, 687]], [[782, 629], [810, 629], [810, 661], [782, 661]], [[645, 627], [685, 627], [685, 665], [645, 665]], [[516, 628], [548, 628], [548, 664], [516, 664]], [[655, 619], [672, 619], [672, 627], [655, 627]], [[598, 617], [605, 624], [599, 629], [592, 622]], [[523, 619], [540, 619], [540, 627], [523, 627]], [[858, 618], [868, 618], [868, 627], [858, 627]], [[727, 618], [735, 618], [735, 627], [727, 627]], [[918, 621], [932, 611], [942, 624], [928, 634]], [[786, 616], [805, 616], [805, 629], [786, 629]], [[373, 604], [420, 604], [420, 619], [373, 619]], [[85, 603], [215, 605], [214, 621], [84, 619]], [[48, 603], [71, 603], [71, 622], [48, 622]], [[788, 561], [806, 561], [806, 572], [788, 572]], [[923, 560], [935, 560], [935, 574], [923, 574]], [[856, 560], [869, 560], [869, 574], [856, 574]], [[62, 553], [409, 553], [409, 567], [62, 567]], [[63, 532], [116, 535], [115, 545], [62, 543]], [[859, 527], [868, 527], [868, 539], [859, 539]], [[925, 526], [934, 526], [934, 540], [925, 540]], [[794, 520], [807, 529], [798, 542], [785, 533]], [[526, 526], [535, 526], [535, 536], [526, 536]], [[262, 513], [395, 513], [395, 526], [262, 526]], [[122, 514], [245, 514], [245, 524], [122, 524]], [[49, 514], [119, 514], [119, 525], [49, 525]], [[755, 492], [828, 492], [828, 507], [755, 507]], [[638, 492], [710, 492], [710, 507], [638, 507]], [[519, 492], [592, 492], [592, 507], [519, 507]], [[85, 450], [123, 450], [123, 461], [85, 461]], [[220, 450], [236, 447], [238, 459], [223, 462]], [[682, 445], [867, 445], [867, 459], [682, 459]], [[562, 445], [666, 445], [666, 459], [562, 459]], [[491, 446], [544, 446], [544, 458], [491, 458]], [[183, 437], [208, 437], [208, 459], [183, 459]], [[52, 431], [72, 438], [64, 462], [44, 455]], [[224, 432], [276, 432], [276, 443], [224, 443]], [[88, 432], [144, 432], [144, 443], [88, 443]], [[506, 383], [616, 382], [616, 397], [506, 398]], [[702, 381], [758, 381], [758, 399], [702, 399]], [[308, 373], [364, 373], [364, 384], [308, 384]], [[91, 372], [166, 372], [166, 383], [91, 383]], [[688, 335], [820, 335], [820, 350], [688, 350]], [[498, 335], [657, 335], [657, 350], [498, 350]], [[208, 316], [244, 316], [244, 331], [208, 331]], [[499, 289], [641, 289], [641, 302], [499, 302]], [[671, 287], [801, 287], [801, 301], [671, 301]], [[670, 241], [816, 241], [816, 255], [670, 255]], [[497, 241], [643, 241], [643, 255], [497, 255]], [[669, 193], [814, 193], [814, 207], [669, 207]], [[498, 193], [643, 193], [643, 207], [498, 207]], [[670, 145], [815, 145], [815, 160], [670, 160]], [[499, 145], [645, 145], [645, 160], [499, 160]], [[489, 103], [546, 103], [546, 120], [489, 120]], [[56, 89], [95, 89], [95, 97], [56, 97]], [[845, 26], [887, 20], [889, 39], [848, 44]], [[26, 20], [700, 20], [700, 37], [26, 37]], [[898, 11], [996, 16], [995, 45], [896, 40]]]
+00152568.jpg	[[[3, 252], [284, 254], [284, 280], [3, 278]], [[196, 233], [254, 233], [254, 240], [196, 240]], [[49, 229], [90, 229], [90, 240], [49, 240]], [[200, 159], [281, 165], [276, 229], [195, 222]]]
+00155628.jpg	[[[149, 901], [503, 903], [503, 922], [149, 920]], [[520, 893], [561, 896], [560, 911], [519, 908]], [[61, 885], [81, 885], [81, 894], [61, 894]], [[150, 878], [503, 882], [503, 900], [149, 896]], [[524, 834], [640, 839], [639, 856], [524, 852]], [[70, 834], [185, 835], [185, 853], [69, 852]], [[246, 555], [466, 555], [466, 569], [246, 569]], [[308, 507], [403, 509], [403, 524], [308, 522]], [[244, 482], [459, 484], [459, 502], [244, 500]], [[252, 422], [459, 424], [458, 452], [251, 450]], [[195, 378], [517, 380], [516, 408], [195, 406]], [[474, 194], [624, 196], [624, 210], [473, 208]], [[73, 129], [641, 131], [641, 160], [73, 158]], [[483, 41], [597, 37], [599, 98], [486, 102]], [[68, 25], [135, 16], [139, 43], [72, 52]]]
+00173364.jpg	[[[8, 178], [57, 178], [57, 200], [8, 200]], [[137, 120], [194, 120], [194, 133], [137, 133]], [[39, 76], [86, 76], [86, 105], [39, 105]], [[250, 20], [311, 20], [311, 36], [250, 36]], [[21, 16], [104, 16], [104, 39], [21, 39]]]
+00175503.jpg	[[[43, 260], [500, 255], [501, 358], [44, 363]], [[52, 200], [349, 178], [354, 251], [58, 273]]]
+00193218.jpg	[[[283, 375], [410, 375], [410, 388], [283, 388]], [[172, 375], [221, 375], [221, 389], [172, 389]], [[110, 375], [161, 375], [161, 389], [110, 389]], [[276, 358], [357, 358], [357, 371], [276, 371]], [[171, 359], [220, 359], [220, 370], [171, 370]], [[409, 357], [492, 357], [492, 370], [409, 370]], [[26, 187], [62, 187], [62, 202], [26, 202]], [[501, 185], [557, 185], [557, 199], [501, 199]], [[381, 187], [420, 185], [421, 199], [382, 201]], [[284, 186], [310, 186], [310, 201], [284, 201]], [[174, 186], [196, 186], [196, 201], [174, 201]], [[499, 165], [540, 165], [540, 176], [499, 176]], [[381, 164], [409, 164], [409, 177], [381, 177]], [[262, 163], [302, 163], [302, 177], [262, 177]], [[176, 163], [230, 163], [230, 177], [176, 177]], [[26, 163], [79, 163], [79, 177], [26, 177]], [[387, 140], [488, 140], [488, 153], [387, 153]], [[28, 139], [131, 139], [131, 152], [28, 152]], [[443, 117], [537, 119], [537, 133], [443, 132]], [[346, 119], [405, 119], [405, 130], [346, 130]], [[261, 119], [302, 119], [302, 130], [261, 130]], [[30, 113], [228, 116], [228, 131], [30, 129]], [[131, 91], [394, 94], [394, 109], [131, 105]], [[562, 82], [583, 82], [583, 107], [562, 107]]]
+00195033.jpg	[[[488, 263], [533, 265], [532, 280], [487, 278]], [[126, 250], [192, 250], [192, 283], [126, 283]], [[338, 249], [362, 249], [362, 266], [338, 266]], [[319, 222], [380, 225], [380, 238], [319, 236]], [[431, 224], [450, 224], [450, 235], [431, 235]], [[365, 203], [538, 203], [538, 216], [365, 216]], [[89, 200], [146, 203], [146, 217], [89, 214]], [[329, 201], [354, 201], [354, 212], [329, 212]], [[371, 181], [449, 181], [449, 194], [371, 194]], [[329, 181], [352, 181], [352, 192], [329, 192]], [[96, 179], [240, 179], [240, 193], [96, 193]], [[456, 161], [555, 161], [555, 174], [456, 174]], [[129, 150], [287, 151], [287, 165], [129, 164]], [[36, 145], [73, 149], [72, 163], [35, 159]], [[527, 146], [552, 146], [552, 155], [527, 155]], [[102, 145], [120, 145], [120, 153], [102, 153]], [[371, 129], [503, 128], [503, 139], [371, 140]], [[99, 126], [193, 126], [193, 139], [99, 139]], [[322, 127], [337, 127], [337, 135], [322, 135]], [[37, 123], [77, 123], [77, 134], [37, 134]], [[324, 106], [337, 106], [337, 115], [324, 115]], [[309, 107], [315, 107], [315, 112], [309, 112]], [[372, 103], [501, 103], [501, 116], [372, 116]], [[349, 105], [360, 105], [360, 114], [349, 114]], [[38, 103], [80, 103], [80, 113], [38, 113]], [[99, 100], [205, 101], [205, 115], [99, 114]], [[306, 90], [317, 90], [317, 97], [306, 97]], [[347, 88], [362, 88], [362, 96], [347, 96]], [[321, 87], [340, 87], [340, 99], [321, 99]], [[358, 84], [513, 82], [513, 95], [358, 97]], [[41, 83], [89, 83], [89, 93], [41, 93]], [[94, 79], [241, 80], [241, 94], [94, 93]], [[313, 66], [394, 66], [394, 79], [313, 79]], [[242, 66], [288, 66], [288, 77], [242, 77]], [[185, 54], [220, 54], [220, 65], [185, 65]], [[469, 48], [547, 48], [547, 61], [469, 61]], [[423, 36], [436, 36], [436, 54], [423, 54]], [[465, 30], [551, 30], [551, 43], [465, 43]], [[207, 21], [329, 23], [328, 41], [207, 39]]]
+00208502.jpg	[[[247, 566], [282, 566], [282, 573], [247, 573]], [[558, 534], [629, 539], [627, 570], [556, 565]], [[205, 540], [284, 540], [284, 552], [205, 552]], [[143, 513], [189, 513], [189, 525], [143, 525]], [[249, 512], [307, 512], [307, 524], [249, 524]], [[44, 500], [118, 500], [118, 519], [44, 519]], [[467, 491], [556, 491], [556, 508], [467, 508]], [[667, 490], [678, 494], [675, 503], [664, 499]], [[788, 489], [794, 495], [789, 499], [783, 494]], [[726, 491], [737, 491], [737, 501], [726, 501]], [[42, 452], [117, 450], [117, 469], [42, 470]], [[175, 450], [236, 450], [236, 464], [175, 464]], [[614, 407], [638, 407], [638, 422], [614, 422]], [[95, 405], [119, 405], [119, 422], [95, 422]], [[49, 399], [63, 413], [49, 427], [35, 414]], [[209, 401], [226, 401], [226, 415], [209, 415]], [[40, 357], [58, 357], [58, 374], [40, 374]], [[94, 356], [119, 356], [119, 373], [94, 373]], [[188, 341], [246, 339], [247, 361], [189, 364]], [[459, 321], [549, 319], [549, 337], [460, 339]], [[459, 273], [551, 273], [551, 290], [459, 290]], [[563, 272], [735, 269], [735, 286], [564, 289]], [[517, 225], [547, 225], [547, 245], [517, 245]], [[459, 226], [480, 226], [480, 244], [459, 244]], [[621, 187], [673, 187], [673, 201], [621, 201]], [[457, 132], [548, 130], [548, 147], [458, 149]], [[572, 106], [787, 99], [787, 120], [573, 126]], [[122, 48], [290, 48], [290, 97], [122, 97]], [[539, 39], [708, 39], [708, 89], [539, 89]]]
+00224225.jpg	[[[134, 429], [153, 426], [157, 445], [138, 448]], [[202, 404], [478, 411], [476, 459], [201, 452]], [[205, 230], [469, 230], [469, 390], [205, 390]], [[131, 265], [172, 265], [172, 279], [131, 279]], [[344, 207], [455, 207], [455, 231], [344, 231]], [[199, 189], [346, 196], [344, 239], [197, 232]], [[10, 44], [157, 41], [158, 112], [11, 115]]]
+00227746.jpg	[[[190, 232], [258, 232], [258, 238], [190, 238]], [[160, 232], [183, 232], [183, 238], [160, 238]], [[123, 232], [150, 232], [150, 238], [123, 238]], [[290, 209], [345, 207], [346, 221], [291, 223]], [[172, 181], [249, 181], [249, 194], [172, 194]], [[143, 178], [165, 180], [162, 208], [140, 206]], [[142, 164], [157, 162], [160, 177], [145, 180]], [[173, 157], [203, 157], [203, 164], [173, 164]], [[200, 154], [347, 154], [347, 167], [200, 167]], [[144, 111], [277, 114], [277, 134], [144, 131]], [[201, 52], [387, 53], [386, 69], [201, 68]], [[139, 47], [191, 45], [192, 62], [140, 64]], [[40, 26], [61, 26], [61, 42], [40, 42]]]
+00229605.jpg	[[[743, 529], [881, 529], [881, 544], [743, 544]], [[236, 499], [589, 498], [589, 522], [236, 523]], [[6, 498], [227, 498], [227, 522], [6, 522]], [[736, 496], [883, 499], [883, 520], [735, 517]], [[606, 495], [716, 489], [718, 515], [608, 521]], [[4, 245], [863, 230], [864, 288], [5, 303]], [[478, 28], [883, 28], [883, 76], [478, 76]]]
+00233011.jpg	[[[63, 227], [291, 227], [291, 242], [63, 242]], [[12, 219], [41, 219], [41, 250], [12, 250]], [[61, 177], [119, 177], [119, 195], [61, 195]], [[11, 173], [40, 169], [44, 200], [14, 203]], [[61, 129], [147, 131], [147, 147], [61, 144]], [[12, 124], [43, 124], [43, 154], [12, 154]], [[125, 89], [238, 89], [238, 103], [125, 103]], [[148, 51], [216, 51], [216, 65], [148, 65]], [[258, 46], [353, 50], [352, 69], [257, 65]], [[9, 49], [52, 49], [52, 68], [9, 68]], [[277, 12], [345, 12], [345, 31], [277, 31]], [[28, 11], [73, 11], [73, 31], [28, 31]]]
+00233625.jpg	[[[375, 397], [632, 399], [632, 443], [375, 440]], [[71, 214], [932, 207], [933, 321], [71, 328]]]
+00233634.jpg	[[[215, 639], [261, 639], [261, 703], [215, 703]], [[523, 635], [570, 635], [570, 695], [523, 695]], [[643, 523], [682, 523], [682, 568], [643, 568]], [[97, 516], [152, 516], [152, 589], [97, 589]], [[755, 395], [760, 395], [760, 401], [755, 401]], [[26, 395], [32, 395], [32, 400], [26, 400]], [[678, 364], [728, 362], [731, 430], [681, 432]], [[54, 361], [107, 361], [107, 434], [54, 434]], [[78, 208], [155, 208], [155, 280], [78, 280]], [[643, 205], [693, 205], [693, 272], [643, 272]], [[210, 88], [260, 86], [263, 164], [213, 166]], [[363, 48], [426, 45], [430, 115], [367, 118]]]
+00234400.jpg	[[[446, 421], [738, 421], [738, 438], [446, 438]], [[157, 421], [454, 421], [454, 438], [157, 438]], [[158, 394], [652, 394], [652, 411], [158, 411]], [[40, 391], [127, 391], [127, 412], [40, 412]], [[158, 342], [304, 345], [304, 363], [158, 360]], [[38, 344], [123, 344], [123, 362], [38, 362]], [[520, 295], [703, 295], [703, 314], [520, 314]], [[394, 292], [483, 290], [484, 314], [394, 317]], [[157, 293], [270, 293], [270, 313], [157, 313]], [[37, 293], [125, 293], [125, 313], [37, 313]], [[156, 243], [358, 243], [358, 267], [156, 267]], [[36, 243], [82, 243], [82, 269], [36, 269]], [[28, 151], [157, 151], [157, 174], [28, 174]], [[282, 98], [507, 98], [507, 111], [282, 111]], [[315, 46], [475, 50], [474, 88], [314, 85]], [[518, 51], [663, 53], [662, 67], [517, 65]], [[487, 19], [706, 17], [706, 43], [487, 45]]]
+00234883.jpg	[[[66, 125], [316, 120], [317, 190], [67, 195]], [[79, 138], [109, 140], [108, 151], [78, 148]], [[72, 120], [120, 120], [120, 130], [72, 130]], [[383, 63], [504, 62], [504, 74], [383, 75]], [[58, 29], [365, 26], [366, 112], [59, 115]], [[387, 28], [501, 26], [501, 45], [387, 47]]]
+test_add_0.jpg	[[[311, 521], [391, 521], [391, 534], [311, 534]], [[277, 500], [424, 500], [424, 514], [277, 514]], [[261, 446], [437, 446], [437, 459], [261, 459]], [[212, 428], [485, 428], [485, 441], [212, 441]], [[247, 388], [457, 388], [457, 409], [247, 409]], [[222, 328], [474, 328], [474, 372], [222, 372]], [[208, 207], [492, 211], [490, 277], [207, 272]], [[266, 164], [422, 166], [421, 197], [265, 195]], [[18, 20], [201, 18], [201, 43], [18, 45]]]
+test_add_1.png	[]
+test_add_10.png	[[[157, 124], [186, 124], [186, 172], [157, 172]], [[65, 117], [95, 117], [95, 168], [65, 168]], [[161, 106], [183, 106], [183, 127], [161, 127]], [[69, 100], [94, 100], [94, 128], [69, 128]], [[117, 46], [154, 45], [157, 174], [121, 175]], [[66, 34], [97, 34], [97, 112], [66, 112]]]
+test_add_11.jpg	[[[1525, 773], [1564, 756], [1575, 780], [1536, 798]], [[1390, 757], [1483, 757], [1483, 791], [1390, 791]], [[1013, 754], [1207, 754], [1207, 800], [1013, 800]], [[685, 755], [875, 755], [875, 796], [685, 796]], [[356, 753], [566, 747], [567, 793], [358, 798]], [[78, 751], [264, 745], [265, 793], [79, 798]], [[602, 647], [1152, 647], [1152, 703], [602, 703]], [[601, 564], [1148, 555], [1149, 611], [602, 620]], [[598, 480], [1066, 472], [1067, 526], [599, 535]], [[598, 393], [1090, 388], [1090, 439], [599, 444]], [[603, 306], [1057, 306], [1057, 357], [603, 357]], [[357, 184], [1515, 184], [1515, 261], [357, 261]], [[60, 43], [257, 37], [259, 83], [61, 89]], [[1305, 41], [1492, 41], [1492, 87], [1305, 87]], [[973, 40], [1171, 34], [1172, 80], [974, 86]], [[670, 40], [862, 34], [864, 80], [671, 86]], [[363, 34], [558, 34], [558, 85], [363, 85]]]
+test_add_12.jpg	[[[11, 592], [136, 594], [135, 613], [11, 611]], [[109, 521], [907, 526], [907, 569], [109, 565]], [[635, 451], [902, 448], [903, 478], [635, 481]], [[112, 447], [466, 449], [466, 486], [112, 483]], [[582, 306], [680, 304], [681, 348], [583, 351]], [[369, 261], [565, 266], [563, 357], [367, 353]], [[64, 85], [853, 88], [853, 161], [64, 159]]]
+test_add_13.jpg	[[[68, 94], [117, 97], [116, 114], [67, 111]]]
+test_add_14.jpg	[[[30, 97], [235, 95], [236, 127], [31, 129]], [[30, 52], [239, 50], [239, 86], [30, 87]]]
+test_add_15.jpg	[[[141, 253], [353, 253], [353, 266], [141, 266]], [[205, 214], [406, 219], [406, 232], [204, 227]], [[106, 212], [193, 213], [193, 227], [106, 226]], [[154, 156], [286, 161], [286, 174], [154, 170]], [[148, 136], [305, 142], [305, 156], [147, 150]], [[108, 137], [145, 137], [145, 148], [108, 148]], [[108, 102], [275, 109], [275, 125], [107, 117]], [[107, 72], [245, 79], [245, 96], [106, 88]], [[107, 39], [209, 42], [209, 62], [106, 59]]]
+test_add_16.jpg	[[[398, 842], [408, 842], [408, 852], [398, 852]], [[382, 741], [746, 741], [746, 774], [382, 774]], [[362, 703], [468, 703], [468, 725], [362, 725]], [[1552, 701], [1576, 701], [1576, 746], [1552, 746]], [[1256, 695], [1442, 695], [1442, 721], [1256, 721]], [[1244, 661], [1448, 661], [1448, 687], [1244, 687]], [[386, 645], [668, 645], [668, 679], [386, 679]], [[1226, 625], [1470, 623], [1470, 651], [1226, 653]], [[360, 604], [580, 604], [580, 629], [360, 629]], [[1202, 592], [1494, 592], [1494, 617], [1202, 617]], [[1166, 556], [1530, 556], [1530, 582], [1166, 582]], [[380, 552], [638, 552], [638, 586], [380, 586]], [[356, 502], [516, 502], [516, 536], [356, 536]], [[774, 260], [1124, 260], [1124, 300], [774, 300]], [[374, 210], [504, 210], [504, 300], [374, 300]], [[776, 212], [1088, 217], [1088, 252], [776, 248]]]
+test_add_17.jpg	[[[321, 255], [393, 258], [392, 271], [320, 269]], [[307, 222], [411, 228], [411, 241], [306, 236]], [[96, 137], [385, 143], [384, 206], [94, 201]], [[72, 95], [399, 103], [398, 124], [71, 117]], [[68, 76], [224, 79], [223, 93], [67, 90]], [[66, 59], [226, 62], [225, 76], [65, 74]]]
+test_add_18.jpg	[[[466, 788], [715, 790], [715, 813], [466, 811]], [[553, 752], [665, 757], [663, 791], [552, 786]], [[119, 539], [189, 539], [189, 570], [119, 570]], [[116, 473], [674, 486], [673, 528], [115, 516]], [[121, 429], [669, 441], [668, 470], [121, 457]], [[122, 376], [673, 383], [673, 409], [122, 402]], [[556, 262], [675, 264], [675, 278], [556, 277]], [[165, 259], [335, 259], [335, 273], [165, 273]], [[344, 195], [456, 197], [455, 220], [343, 217]], [[309, 175], [490, 175], [490, 190], [309, 190]], [[255, 128], [537, 131], [537, 169], [254, 165]], [[347, 92], [486, 94], [486, 109], [346, 107]], [[285, 41], [567, 49], [566, 82], [284, 74]], [[236, 32], [266, 32], [266, 60], [236, 60]]]
+test_add_19.jpg	[[[24, 294], [42, 294], [42, 302], [24, 302]], [[64, 293], [105, 293], [105, 303], [64, 303]], [[145, 287], [163, 287], [163, 304], [145, 304]], [[63, 280], [106, 280], [106, 290], [63, 290]], [[9, 281], [26, 281], [26, 288], [9, 288]], [[220, 279], [245, 279], [245, 291], [220, 291]], [[177, 279], [208, 279], [208, 290], [177, 290]], [[23, 279], [51, 279], [51, 290], [23, 290]], [[145, 278], [162, 278], [162, 292], [145, 292]], [[8, 267], [18, 267], [18, 276], [8, 276]], [[221, 265], [243, 265], [243, 277], [221, 277]], [[24, 265], [47, 265], [47, 277], [24, 277]], [[142, 263], [163, 263], [163, 279], [142, 279]], [[218, 252], [249, 252], [249, 265], [218, 265]], [[65, 253], [131, 253], [131, 263], [65, 263]], [[24, 252], [43, 252], [43, 264], [24, 264]], [[8, 253], [18, 253], [18, 262], [8, 262]], [[8, 240], [17, 240], [17, 249], [8, 249]], [[63, 237], [114, 237], [114, 251], [63, 251]], [[25, 236], [47, 239], [45, 251], [23, 249]], [[144, 234], [166, 237], [163, 253], [142, 249]], [[494, 226], [531, 226], [531, 239], [494, 239]], [[335, 226], [354, 226], [354, 237], [335, 237]], [[288, 226], [314, 226], [314, 237], [288, 237]], [[63, 226], [113, 226], [113, 236], [63, 236]], [[7, 227], [17, 227], [17, 234], [7, 234]], [[221, 225], [248, 225], [248, 235], [221, 235]], [[143, 225], [165, 222], [167, 234], [145, 237]], [[24, 224], [48, 224], [48, 238], [24, 238]], [[495, 213], [524, 213], [524, 224], [495, 224]], [[420, 212], [437, 212], [437, 225], [420, 225]], [[336, 212], [398, 212], [398, 223], [336, 223]], [[292, 212], [320, 212], [320, 223], [292, 223]], [[222, 212], [249, 212], [249, 223], [222, 223]], [[145, 212], [166, 212], [166, 223], [145, 223]], [[61, 211], [113, 209], [114, 222], [62, 224]], [[26, 211], [48, 211], [48, 223], [26, 223]], [[337, 199], [383, 199], [383, 209], [337, 209]], [[65, 200], [87, 200], [87, 207], [65, 207]], [[493, 197], [541, 197], [541, 211], [493, 211]], [[445, 202], [455, 196], [462, 206], [452, 212]], [[178, 198], [205, 198], [205, 208], [178, 208]], [[146, 199], [157, 199], [157, 208], [146, 208]], [[32, 194], [43, 204], [33, 214], [22, 203]], [[422, 193], [440, 201], [432, 215], [415, 207]], [[64, 186], [131, 186], [131, 196], [64, 196]], [[337, 185], [399, 185], [399, 196], [337, 196]], [[445, 190], [456, 182], [465, 191], [454, 200]], [[292, 188], [308, 182], [313, 193], [297, 200]], [[220, 183], [255, 183], [255, 197], [220, 197]], [[142, 184], [158, 184], [158, 197], [142, 197]], [[493, 182], [518, 182], [518, 197], [493, 197]], [[425, 180], [437, 191], [427, 202], [414, 190]], [[32, 179], [42, 189], [32, 199], [22, 189]], [[182, 179], [196, 185], [188, 199], [175, 193]], [[335, 172], [400, 169], [400, 183], [336, 185]], [[492, 170], [519, 170], [519, 185], [492, 185]], [[412, 177], [428, 164], [440, 178], [425, 190]], [[293, 171], [315, 171], [315, 185], [293, 185]], [[220, 170], [251, 170], [251, 184], [220, 184]], [[178, 172], [188, 172], [188, 183], [178, 183]], [[64, 172], [125, 170], [125, 181], [64, 182]], [[454, 168], [464, 176], [454, 185], [445, 176]], [[145, 168], [163, 172], [159, 185], [142, 180]], [[30, 165], [43, 174], [34, 186], [20, 177]], [[493, 160], [523, 160], [523, 170], [493, 170]], [[402, 161], [435, 161], [435, 168], [402, 168]], [[335, 159], [401, 159], [401, 169], [335, 169]], [[296, 159], [325, 159], [325, 170], [296, 170]], [[221, 158], [251, 158], [251, 169], [221, 169]], [[174, 161], [183, 156], [190, 167], [181, 172]], [[145, 158], [162, 158], [162, 170], [145, 170]], [[61, 158], [125, 157], [125, 168], [62, 169]], [[20, 161], [33, 154], [40, 167], [28, 174]], [[492, 143], [542, 143], [542, 157], [492, 157]], [[450, 144], [479, 144], [479, 157], [450, 157]], [[335, 143], [439, 143], [439, 156], [335, 156]], [[294, 143], [327, 143], [327, 157], [294, 157]], [[220, 143], [253, 143], [253, 157], [220, 157]], [[178, 145], [187, 145], [187, 156], [178, 156]], [[63, 144], [104, 144], [104, 155], [63, 155]], [[144, 140], [164, 145], [160, 159], [141, 154]], [[31, 137], [44, 149], [31, 162], [17, 149]], [[286, 135], [291, 135], [291, 140], [286, 140]], [[177, 133], [193, 133], [193, 144], [177, 144]], [[336, 132], [388, 132], [388, 141], [336, 141]], [[492, 131], [525, 131], [525, 141], [492, 141]], [[450, 131], [477, 131], [477, 141], [450, 141]], [[292, 131], [321, 131], [321, 141], [292, 141]], [[218, 132], [255, 130], [256, 141], [219, 144]], [[63, 131], [95, 131], [95, 141], [63, 141]], [[417, 130], [437, 130], [437, 141], [417, 141]], [[145, 130], [159, 130], [159, 143], [145, 143]], [[30, 124], [43, 133], [32, 147], [19, 138]], [[493, 118], [535, 118], [535, 129], [493, 129]], [[335, 118], [387, 118], [387, 129], [335, 129]], [[218, 118], [255, 118], [255, 128], [218, 128]], [[451, 117], [478, 117], [478, 129], [451, 129]], [[418, 117], [438, 117], [438, 130], [418, 130]], [[177, 116], [209, 116], [209, 130], [177, 130]], [[145, 117], [162, 117], [162, 130], [145, 130]], [[62, 116], [88, 116], [88, 131], [62, 131]], [[19, 121], [33, 111], [43, 124], [29, 134]], [[491, 107], [523, 107], [523, 113], [491, 113]], [[449, 107], [477, 107], [477, 113], [449, 113]], [[420, 107], [436, 107], [436, 113], [420, 113]], [[295, 107], [319, 107], [319, 114], [295, 114]], [[220, 107], [242, 107], [242, 113], [220, 113]], [[176, 107], [203, 107], [203, 113], [176, 113]], [[145, 107], [161, 107], [161, 114], [145, 114]], [[334, 105], [372, 105], [372, 114], [334, 114]], [[63, 106], [86, 106], [86, 113], [63, 113]], [[483, 89], [522, 89], [522, 99], [483, 99]], [[331, 88], [380, 88], [380, 99], [331, 99]], [[276, 88], [325, 88], [325, 99], [276, 99]], [[214, 88], [246, 88], [246, 99], [214, 99]], [[411, 86], [474, 86], [474, 100], [411, 100]], [[6, 86], [102, 86], [102, 100], [6, 100]], [[415, 66], [461, 66], [461, 77], [415, 77]], [[288, 66], [333, 66], [333, 77], [288, 77]], [[157, 64], [206, 64], [206, 78], [157, 78]], [[33, 61], [43, 66], [37, 77], [27, 72]], [[416, 48], [523, 49], [523, 63], [415, 62]], [[288, 49], [375, 49], [375, 63], [288, 63]], [[159, 49], [269, 49], [269, 62], [159, 62]], [[24, 53], [36, 46], [45, 59], [33, 67]], [[416, 36], [481, 36], [481, 46], [416, 46]], [[25, 38], [39, 32], [46, 45], [32, 52]], [[157, 34], [205, 34], [205, 47], [157, 47]], [[412, 4], [527, 4], [527, 17], [412, 17]], [[146, 4], [345, 2], [345, 15], [146, 17]]]
+test_add_20.jpg	[[[31, 346], [605, 346], [605, 370], [31, 370]], [[217, 294], [510, 294], [510, 322], [217, 322]], [[473, 271], [525, 271], [525, 286], [473, 286]], [[220, 267], [287, 267], [287, 286], [220, 286]], [[219, 239], [484, 239], [484, 263], [219, 263]], [[221, 217], [303, 217], [303, 234], [221, 234]], [[402, 192], [417, 192], [417, 205], [402, 205]], [[222, 187], [341, 187], [341, 207], [222, 207]], [[221, 162], [287, 162], [287, 180], [221, 180]], [[375, 122], [475, 124], [475, 146], [375, 143]], [[222, 124], [356, 122], [356, 143], [222, 146]], [[218, 81], [352, 84], [352, 116], [218, 113]], [[440, 34], [605, 34], [605, 59], [440, 59]], [[72, 16], [398, 16], [398, 44], [72, 44]]]
+test_add_3.jpg	[[[169, 327], [337, 326], [337, 341], [169, 342]], [[170, 288], [307, 290], [307, 312], [170, 310]], [[171, 221], [323, 221], [323, 234], [171, 234]], [[340, 221], [449, 217], [449, 231], [341, 234]], [[169, 201], [372, 201], [372, 214], [169, 214]], [[170, 183], [418, 183], [418, 196], [170, 196]], [[170, 149], [416, 149], [416, 163], [170, 163]], [[171, 119], [418, 119], [418, 140], [171, 140]], [[326, 64], [478, 64], [478, 91], [326, 91]], [[173, 64], [306, 60], [306, 89], [174, 93]]]
+test_add_4.png	[]
+test_add_5.png	[[[48, 164], [108, 164], [108, 174], [48, 174]], [[52, 121], [169, 121], [169, 134], [52, 134]], [[50, 102], [165, 102], [165, 118], [50, 118]], [[52, 83], [164, 83], [164, 100], [52, 100]], [[51, 68], [166, 68], [166, 84], [51, 84]], [[51, 50], [145, 47], [145, 64], [52, 67]]]
+test_add_6.jpg	[[[123, 223], [219, 227], [218, 251], [122, 247]], [[172, 172], [186, 186], [172, 200], [158, 186]]]
+test_add_7.jpg	[[[48, 938], [174, 936], [174, 962], [48, 964]], [[227, 873], [629, 876], [628, 953], [226, 949]], [[56, 744], [637, 744], [637, 789], [56, 789]], [[150, 674], [545, 678], [544, 721], [150, 718]], [[73, 504], [633, 504], [633, 601], [73, 601]], [[59, 270], [655, 279], [652, 441], [56, 432]], [[513, 193], [553, 193], [553, 223], [513, 223]], [[61, 175], [532, 175], [532, 239], [61, 239]], [[533, 178], [642, 178], [642, 236], [533, 236]]]
+test_add_8.jpg	[[[251, 586], [454, 580], [454, 606], [252, 613]], [[107, 533], [457, 527], [457, 560], [108, 566]], [[336, 494], [384, 494], [384, 507], [336, 507]], [[27, 307], [355, 297], [356, 320], [28, 330]], [[22, 259], [445, 251], [445, 274], [23, 282]], [[78, 209], [445, 205], [445, 225], [78, 229]], [[160, 23], [319, 30], [317, 79], [158, 72]]]
+test_add_9.png	[[[266, 687], [486, 687], [486, 696], [266, 696]], [[196, 668], [554, 668], [554, 681], [196, 681]], [[153, 596], [597, 596], [597, 606], [153, 606]], [[215, 578], [541, 578], [541, 588], [215, 588]], [[85, 542], [664, 542], [664, 552], [85, 552]], [[96, 522], [653, 522], [653, 535], [96, 535]], [[362, 449], [389, 449], [389, 460], [362, 460]], [[238, 376], [513, 376], [513, 389], [238, 389]], [[177, 356], [574, 356], [574, 368], [177, 368]], [[344, 281], [408, 283], [407, 297], [343, 294]], [[256, 205], [492, 205], [492, 219], [256, 219]]]
diff --git a/test_tipc/supplementary/__init__.py b/test_tipc/supplementary/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/test_tipc/supplementary/config.py b/test_tipc/supplementary/config.py
new file mode 100644
index 0000000..635a2b3
--- /dev/null
+++ b/test_tipc/supplementary/config.py
@@ -0,0 +1,138 @@
+import numpy as np
+import os
+import sys
+import platform
+import yaml
+import time
+import shutil
+import paddle
+import paddle.distributed as dist
+from tqdm import tqdm
+from argparse import ArgumentParser, RawDescriptionHelpFormatter
+from utils import get_logger, print_dict
+
+
+class ArgsParser(ArgumentParser):
+    def __init__(self):
+        super(ArgsParser, self).__init__(formatter_class=RawDescriptionHelpFormatter)
+        self.add_argument("-c", "--config", help="configuration file to use")
+        self.add_argument("-o", "--opt", nargs="+", help="set configuration options")
+        self.add_argument(
+            "-p",
+            "--profiler_options",
+            type=str,
+            default=None,
+            help='The option of profiler, which should be in format "key1=value1;key2=value2;key3=value3".',
+        )
+
+    def parse_args(self, argv=None):
+        args = super(ArgsParser, self).parse_args(argv)
+        assert args.config is not None, "Please specify --config=configure_file_path."
+        args.opt = self._parse_opt(args.opt)
+        return args
+
+    def _parse_opt(self, opts):
+        config = {}
+        if not opts:
+            return config
+        for s in opts:
+            s = s.strip()
+            k, v = s.split("=")
+            config[k] = yaml.load(v, Loader=yaml.SafeLoader)
+        return config
+
+
+class AttrDict(dict):
+    """Single level attribute dict, NOT recursive"""
+
+    def __init__(self, **kwargs):
+        super(AttrDict, self).__init__()
+        super(AttrDict, self).update(kwargs)
+
+    def __getattr__(self, key):
+        if key in self:
+            return self[key]
+        raise AttributeError("object has no attribute '{}'".format(key))
+
+
+global_config = AttrDict()
+
+default_config = {
+    "Global": {
+        "debug": False,
+    }
+}
+
+
+def load_config(file_path):
+    """
+    Load config from yml/yaml file.
+    Args:
+        file_path (str): Path of the config file to be loaded.
+    Returns: global config
+    """
+    merge_config(default_config)
+    _, ext = os.path.splitext(file_path)
+    assert ext in [".yml", ".yaml"], "only support yaml files for now"
+    merge_config(yaml.load(open(file_path, "rb"), Loader=yaml.SafeLoader))
+    return global_config
+
+
+def merge_config(config):
+    """
+    Merge config into global config.
+    Args:
+        config (dict): Config to be merged.
+    Returns: global config
+    """
+    for key, value in config.items():
+        if "." not in key:
+            if isinstance(value, dict) and key in global_config:
+                global_config[key].update(value)
+            else:
+                global_config[key] = value
+        else:
+            sub_keys = key.split(".")
+            assert (
+                sub_keys[0] in global_config
+            ), "the sub_keys can only be one of global_config: {}, but get: {}, please check your running command".format(
+                global_config.keys(), sub_keys[0]
+            )
+            cur = global_config[sub_keys[0]]
+            for idx, sub_key in enumerate(sub_keys[1:]):
+                if idx == len(sub_keys) - 2:
+                    cur[sub_key] = value
+                else:
+                    cur = cur[sub_key]
+
+
+def preprocess(is_train=False):
+    FLAGS = ArgsParser().parse_args()
+    profiler_options = FLAGS.profiler_options
+    config = load_config(FLAGS.config)
+    merge_config(FLAGS.opt)
+    profile_dic = {"profiler_options": FLAGS.profiler_options}
+    merge_config(profile_dic)
+
+    if is_train:
+        # save_config
+        save_model_dir = config["save_model_dir"]
+        os.makedirs(save_model_dir, exist_ok=True)
+        with open(os.path.join(save_model_dir, "config.yml"), "w") as f:
+            yaml.dump(dict(config), f, default_flow_style=False, sort_keys=False)
+        log_file = "{}/train.log".format(save_model_dir)
+    else:
+        log_file = None
+    logger = get_logger(log_file=log_file)
+
+    # check if set use_gpu=True in paddlepaddle cpu version
+    use_gpu = config["use_gpu"]
+
+    print_dict(config, logger)
+
+    return config, logger
+
+
+if __name__ == "__main__":
+    config, logger = preprocess(is_train=False)
+    # print(config)
diff --git a/test_tipc/supplementary/custom_op/custom_relu_op.cc b/test_tipc/supplementary/custom_op/custom_relu_op.cc
new file mode 100644
index 0000000..6c48da7
--- /dev/null
+++ b/test_tipc/supplementary/custom_op/custom_relu_op.cc
@@ -0,0 +1,106 @@
+// Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// reference from :
+// https://github.com/PaddlePaddle/Paddle-Inference-Demo/blob/master/python/custom-operator/custom_relu_op.cc
+#include <iostream>
+#include <vector>
+
+#include "paddle/extension.h"
+
+template <typename data_t>
+void relu_cpu_forward_kernel(const data_t *x_data, data_t *out_data,
+                             int64_t x_numel) {
+  for (int i = 0; i < x_numel; ++i) {
+    out_data[i] = std::max(static_cast<data_t>(0.), x_data[i]);
+  }
+}
+
+template <typename data_t>
+void relu_cpu_backward_kernel(const data_t *grad_out_data,
+                              const data_t *out_data, data_t *grad_x_data,
+                              int64_t out_numel) {
+  for (int i = 0; i < out_numel; ++i) {
+    grad_x_data[i] =
+        grad_out_data[i] * (out_data[i] > static_cast<data_t>(0) ? 1. : 0.);
+  }
+}
+
+std::vector<paddle::Tensor> relu_cpu_forward(const paddle::Tensor &x) {
+  auto out = paddle::Tensor(paddle::PlaceType::kCPU);
+
+  out.reshape(x.shape());
+  PD_DISPATCH_FLOATING_TYPES(
+      x.type(), "relu_cpu_forward", ([&] {
+        relu_cpu_forward_kernel<data_t>(
+            x.data<data_t>(), out.mutable_data<data_t>(x.place()), x.size());
+      }));
+
+  return {out};
+}
+
+std::vector<paddle::Tensor> relu_cpu_backward(const paddle::Tensor &x,
+                                              const paddle::Tensor &out,
+                                              const paddle::Tensor &grad_out) {
+  auto grad_x = paddle::Tensor(paddle::PlaceType::kCPU);
+  grad_x.reshape(x.shape());
+
+  PD_DISPATCH_FLOATING_TYPES(out.type(), "relu_cpu_backward", ([&] {
+                               relu_cpu_backward_kernel<data_t>(
+                                   grad_out.data<data_t>(), out.data<data_t>(),
+                                   grad_x.mutable_data<data_t>(x.place()),
+                                   out.size());
+                             }));
+
+  return {grad_x};
+}
+
+std::vector<paddle::Tensor> relu_cuda_forward(const paddle::Tensor &x);
+std::vector<paddle::Tensor> relu_cuda_backward(const paddle::Tensor &x,
+                                               const paddle::Tensor &out,
+                                               const paddle::Tensor &grad_out);
+
+std::vector<paddle::Tensor> ReluForward(const paddle::Tensor &x) {
+  // TODO(chenweihang): Check Input
+  if (x.place() == paddle::PlaceType::kCPU) {
+    return relu_cpu_forward(x);
+  } else if (x.place() == paddle::PlaceType::kGPU) {
+    return relu_cuda_forward(x);
+  } else {
+    throw std::runtime_error("Not implemented.");
+  }
+}
+
+std::vector<paddle::Tensor> ReluBackward(const paddle::Tensor &x,
+                                         const paddle::Tensor &out,
+                                         const paddle::Tensor &grad_out) {
+  // TODO(chenweihang): Check Input
+  if (x.place() == paddle::PlaceType::kCPU) {
+    return relu_cpu_backward(x, out, grad_out);
+  } else if (x.place() == paddle::PlaceType::kGPU) {
+    return relu_cuda_backward(x, out, grad_out);
+  } else {
+    throw std::runtime_error("Not implemented.");
+  }
+}
+
+PD_BUILD_OP(custom_relu)
+    .Inputs({"X"})
+    .Outputs({"Out"})
+    .SetKernelFn(PD_KERNEL(ReluForward));
+
+PD_BUILD_GRAD_OP(custom_relu)
+    .Inputs({"X", "Out", paddle::Grad("Out")})
+    .Outputs({paddle::Grad("X")})
+    .SetKernelFn(PD_KERNEL(ReluBackward));
diff --git a/test_tipc/supplementary/custom_op/custom_relu_op.cu b/test_tipc/supplementary/custom_op/custom_relu_op.cu
new file mode 100644
index 0000000..fc2792e
--- /dev/null
+++ b/test_tipc/supplementary/custom_op/custom_relu_op.cu
@@ -0,0 +1,71 @@
+// Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// reference
+// https://github.com/PaddlePaddle/Paddle-Inference-Demo/blob/master/python/custom-operator/custom_relu_op.cu
+
+#include "paddle/extension.h"
+
+template <typename data_t>
+__global__ void relu_cuda_forward_kernel(const data_t *x, data_t *y,
+                                         const int num) {
+  int gid = blockIdx.x * blockDim.x + threadIdx.x;
+  for (int i = gid; i < num; i += blockDim.x * gridDim.x) {
+    y[i] = max(x[i], static_cast<data_t>(0.));
+  }
+}
+
+template <typename data_t>
+__global__ void relu_cuda_backward_kernel(const data_t *dy, const data_t *y,
+                                          data_t *dx, const int num) {
+  int gid = blockIdx.x * blockDim.x + threadIdx.x;
+  for (int i = gid; i < num; i += blockDim.x * gridDim.x) {
+    dx[i] = dy[i] * (y[i] > 0 ? 1. : 0.);
+  }
+}
+
+std::vector<paddle::Tensor> relu_cuda_forward(const paddle::Tensor &x) {
+  auto out = paddle::Tensor(paddle::PlaceType::kGPU);
+
+  out.reshape(x.shape());
+  int numel = x.size();
+  int block = 512;
+  int grid = (numel + block - 1) / block;
+  PD_DISPATCH_FLOATING_TYPES(
+      x.type(), "relu_cuda_forward_kernel", ([&] {
+        relu_cuda_forward_kernel<data_t><<<grid, block, 0, x.stream()>>>(
+            x.data<data_t>(), out.mutable_data<data_t>(x.place()), numel);
+      }));
+
+  return {out};
+}
+
+std::vector<paddle::Tensor> relu_cuda_backward(const paddle::Tensor &x,
+                                               const paddle::Tensor &out,
+                                               const paddle::Tensor &grad_out) {
+  auto grad_x = paddle::Tensor(paddle::PlaceType::kGPU);
+  grad_x.reshape(x.shape());
+
+  int numel = out.size();
+  int block = 512;
+  int grid = (numel + block - 1) / block;
+  PD_DISPATCH_FLOATING_TYPES(
+      out.type(), "relu_cuda_backward_kernel", ([&] {
+        relu_cuda_backward_kernel<data_t><<<grid, block, 0, x.stream()>>>(
+            grad_out.data<data_t>(), out.data<data_t>(),
+            grad_x.mutable_data<data_t>(x.place()), numel);
+      }));
+
+  return {grad_x};
+}
diff --git a/test_tipc/supplementary/custom_op/test.py b/test_tipc/supplementary/custom_op/test.py
new file mode 100644
index 0000000..df8d939
--- /dev/null
+++ b/test_tipc/supplementary/custom_op/test.py
@@ -0,0 +1,77 @@
+import paddle
+import paddle.nn as nn
+from paddle.vision.transforms import Compose, Normalize
+from paddle.utils.cpp_extension import load
+from paddle.inference import Config
+from paddle.inference import create_predictor
+import numpy as np
+
+EPOCH_NUM = 4
+BATCH_SIZE = 64
+
+# jit compile custom op
+custom_ops = load(
+    name="custom_jit_ops", sources=["custom_relu_op.cc", "custom_relu_op.cu"]
+)
+
+
+class LeNet(nn.Layer):
+    def __init__(self):
+        super(LeNet, self).__init__()
+        self.conv1 = nn.Conv2D(
+            in_channels=1, out_channels=6, kernel_size=5, stride=1, padding=2
+        )
+        self.max_pool1 = nn.MaxPool2D(kernel_size=2, stride=2)
+        self.conv2 = nn.Conv2D(in_channels=6, out_channels=16, kernel_size=5, stride=1)
+        self.max_pool2 = nn.MaxPool2D(kernel_size=2, stride=2)
+        self.linear1 = nn.Linear(in_features=16 * 5 * 5, out_features=120)
+        self.linear2 = nn.Linear(in_features=120, out_features=84)
+        self.linear3 = nn.Linear(in_features=84, out_features=10)
+
+    def forward(self, x):
+        x = self.conv1(x)
+        x = custom_ops.custom_relu(x)
+        x = self.max_pool1(x)
+        x = custom_ops.custom_relu(x)
+        x = self.conv2(x)
+        x = self.max_pool2(x)
+        x = paddle.flatten(x, start_axis=1, stop_axis=-1)
+        x = self.linear1(x)
+        x = custom_ops.custom_relu(x)
+        x = self.linear2(x)
+        x = custom_ops.custom_relu(x)
+        x = self.linear3(x)
+        return x
+
+
+# set device
+paddle.set_device("gpu")
+
+# model
+net = LeNet()
+loss_fn = nn.CrossEntropyLoss()
+opt = paddle.optimizer.Adam(learning_rate=0.001, parameters=net.parameters())
+
+# data loader
+transform = Compose([Normalize(mean=[127.5], std=[127.5], data_format="CHW")])
+train_dataset = paddle.vision.datasets.MNIST(mode="train", transform=transform)
+train_loader = paddle.io.DataLoader(
+    train_dataset, batch_size=BATCH_SIZE, shuffle=True, drop_last=True, num_workers=2
+)
+
+# train
+for epoch_id in range(EPOCH_NUM):
+    for batch_id, (image, label) in enumerate(train_loader()):
+        out = net(image)
+        loss = loss_fn(out, label)
+        loss.backward()
+
+        if batch_id % 300 == 0:
+            print(
+                "Epoch {} batch {}: loss = {}".format(
+                    epoch_id, batch_id, np.mean(loss.numpy())
+                )
+            )
+
+        opt.step()
+        opt.clear_grad()
diff --git a/test_tipc/supplementary/data.py b/test_tipc/supplementary/data.py
new file mode 100644
index 0000000..02c814d
--- /dev/null
+++ b/test_tipc/supplementary/data.py
@@ -0,0 +1,133 @@
+import numpy as np
+import paddle
+import os
+import cv2
+import glob
+
+
+def transform(data, ops=None):
+    """transform"""
+    if ops is None:
+        ops = []
+    for op in ops:
+        data = op(data)
+        if data is None:
+            return None
+    return data
+
+
+def create_operators(op_param_list, global_config=None):
+    """
+    create operators based on the config
+    Args:
+        params(list): a dict list, used to create some operators
+    """
+    assert isinstance(op_param_list, list), "operator config should be a list"
+    ops = []
+    for operator in op_param_list:
+        assert isinstance(operator, dict) and len(operator) == 1, "yaml format error"
+        op_name = list(operator)[0]
+        param = {} if operator[op_name] is None else operator[op_name]
+        if global_config is not None:
+            param.update(global_config)
+        op = eval(op_name)(**param)
+        ops.append(op)
+    return ops
+
+
+class DecodeImage(object):
+    """decode image"""
+
+    def __init__(self, img_mode="RGB", channel_first=False, **kwargs):
+        self.img_mode = img_mode
+        self.channel_first = channel_first
+
+    def __call__(self, data):
+        img = data["image"]
+        assert type(img) is bytes and len(img) > 0, "invalid input 'img' in DecodeImage"
+        img = np.frombuffer(img, dtype="uint8")
+        img = cv2.imdecode(img, 1)
+        if img is None:
+            return None
+        if self.img_mode == "GRAY":
+            img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
+        elif self.img_mode == "RGB":
+            assert img.shape[2] == 3, "invalid shape of image[%s]" % (img.shape)
+            img = img[:, :, ::-1]
+
+        if self.channel_first:
+            img = img.transpose((2, 0, 1))
+
+        data["image"] = img
+        data["src_image"] = img
+        return data
+
+
+class NormalizeImage(object):
+    """normalize image such as subtract mean, divide std"""
+
+    def __init__(self, scale=None, mean=None, std=None, order="chw", **kwargs):
+        if isinstance(scale, str):
+            scale = eval(scale)
+        self.scale = np.float32(scale if scale is not None else 1.0 / 255.0)
+        mean = mean if mean is not None else [0.485, 0.456, 0.406]
+        std = std if std is not None else [0.229, 0.224, 0.225]
+
+        shape = (3, 1, 1) if order == "chw" else (1, 1, 3)
+        self.mean = np.array(mean).reshape(shape).astype("float32")
+        self.std = np.array(std).reshape(shape).astype("float32")
+
+    def __call__(self, data):
+        img = data["image"]
+        from PIL import Image
+
+        if isinstance(img, Image.Image):
+            img = np.array(img)
+        assert isinstance(img, np.ndarray), "invalid input 'img' in NormalizeImage"
+        data["image"] = (img.astype("float32") * self.scale - self.mean) / self.std
+        return data
+
+
+class ToCHWImage(object):
+    """convert hwc image to chw image"""
+
+    def __init__(self, **kwargs):
+        pass
+
+    def __call__(self, data):
+        img = data["image"]
+        from PIL import Image
+
+        if isinstance(img, Image.Image):
+            img = np.array(img)
+        data["image"] = img.transpose((2, 0, 1))
+
+        src_img = data["src_image"]
+        from PIL import Image
+
+        if isinstance(img, Image.Image):
+            src_img = np.array(src_img)
+        data["src_image"] = img.transpose((2, 0, 1))
+
+        return data
+
+
+class SimpleDataset(nn.Dataset):
+    def __init__(self, config, mode, logger, seed=None):
+        self.logger = logger
+        self.mode = mode.lower()
+
+        data_dir = config["Train"]["data_dir"]
+
+        imgs_list = self.get_image_list(data_dir)
+
+        self.ops = create_operators(cfg["transforms"], None)
+
+    def get_image_list(self, img_dir):
+        imgs = glob.glob(os.path.join(img_dir, "*.png"))
+        if len(imgs) == 0:
+            raise ValueError(f"not any images founded in {img_dir}")
+        return imgs
+
+    def __getitem__(self, idx):
+        return None
diff --git a/test_tipc/supplementary/data_loader.py b/test_tipc/supplementary/data_loader.py
new file mode 100644
index 0000000..2e40662
--- /dev/null
+++ b/test_tipc/supplementary/data_loader.py
@@ -0,0 +1,60 @@
+import numpy as np
+from paddle.vision.datasets import Cifar100
+from paddle.vision.transforms import Normalize
+import signal
+import os
+from paddle.io import Dataset, DataLoader, DistributedBatchSampler
+
+
+def term_mp(sig_num, frame):
+    """kill all child processes"""
+    pid = os.getpid()
+    pgid = os.getpgid(os.getpid())
+    print("main proc {} exit, kill process group " "{}".format(pid, pgid))
+    os.killpg(pgid, signal.SIGKILL)
+    return
+
+
+def build_dataloader(mode, batch_size=4, seed=None, num_workers=0, device="gpu:0"):
+    normalize = Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5], data_format="HWC")
+
+    if mode.lower() == "train":
+        dataset = Cifar100(mode=mode, transform=normalize)
+    elif mode.lower() in ["test", "valid", "eval"]:
+        dataset = Cifar100(mode="test", transform=normalize)
+    else:
+        raise ValueError(f"{mode} should be one of ['train', 'test']")
+
+    # define batch sampler
+    batch_sampler = DistributedBatchSampler(
+        dataset=dataset, batch_size=batch_size, shuffle=False, drop_last=True
+    )
+
+    data_loader = DataLoader(
+        dataset=dataset,
+        batch_sampler=batch_sampler,
+        places=device,
+        num_workers=num_workers,
+        return_list=True,
+        use_shared_memory=False,
+    )
+
+    # support exit using ctrl+c
+    signal.signal(signal.SIGINT, term_mp)
+    signal.signal(signal.SIGTERM, term_mp)
+
+    return data_loader
+
+
+# cifar100 = Cifar100(mode='train', transform=normalize)
+
+# data = cifar100[0]
+
+# image, label = data
+
+# reader = build_dataloader('train')
+
+# for idx, data in enumerate(reader):
+#     print(idx, data[0].shape, data[1].shape)
+#     if idx >= 10:
+#         break
diff --git a/test_tipc/supplementary/load_cifar.py b/test_tipc/supplementary/load_cifar.py
new file mode 100644
index 0000000..35d0ffa
--- /dev/null
+++ b/test_tipc/supplementary/load_cifar.py
@@ -0,0 +1,40 @@
+import pickle as p
+import numpy as np
+from PIL import Image
+
+
+def load_CIFAR_batch(filename):
+    """load single batch of cifar"""
+    with open(filename, "rb") as f:
+        datadict = p.load(f, encoding="bytes")
+        # 以字典的形式取出数据
+        X = datadict[b"data"]
+        Y = datadict[b"fine_labels"]
+        try:
+            X = X.reshape(10000, 3, 32, 32)
+        except:
+            X = X.reshape(50000, 3, 32, 32)
+        Y = np.array(Y)
+        print(Y.shape)
+        return X, Y
+
+
+if __name__ == "__main__":
+    mode = "train"
+    imgX, imgY = load_CIFAR_batch(f"./cifar-100-python/{mode}")
+    with open(f"./cifar-100-python/{mode}_imgs/img_label.txt", "a+") as f:
+        for i in range(imgY.shape[0]):
+            f.write("img" + str(i) + " " + str(imgY[i]) + "\n")
+
+    for i in range(imgX.shape[0]):
+        imgs = imgX[i]
+        img0 = imgs[0]
+        img1 = imgs[1]
+        img2 = imgs[2]
+        i0 = Image.fromarray(img0)
+        i1 = Image.fromarray(img1)
+        i2 = Image.fromarray(img2)
+        img = Image.merge("RGB", (i0, i1, i2))
+        name = "img" + str(i) + ".png"
+        img.save(f"./cifar-100-python/{mode}_imgs/" + name, "png")
+    print("save successfully!")
diff --git a/test_tipc/supplementary/loss.py b/test_tipc/supplementary/loss.py
new file mode 100644
index 0000000..f139a71
--- /dev/null
+++ b/test_tipc/supplementary/loss.py
@@ -0,0 +1,131 @@
+import paddle
+import paddle.nn.functional as F
+
+
+class Loss(object):
+    """
+    Loss
+    """
+
+    def __init__(self, class_dim=1000, epsilon=None):
+        assert class_dim > 1, "class_dim=%d is not larger than 1" % (class_dim)
+        self._class_dim = class_dim
+        if epsilon is not None and epsilon >= 0.0 and epsilon <= 1.0:
+            self._epsilon = epsilon
+            self._label_smoothing = True
+        else:
+            self._epsilon = None
+            self._label_smoothing = False
+
+    def _labelsmoothing(self, target):
+        if target.shape[-1] != self._class_dim:
+            one_hot_target = F.one_hot(target, self._class_dim)
+        else:
+            one_hot_target = target
+        soft_target = F.label_smooth(one_hot_target, epsilon=self._epsilon)
+        soft_target = paddle.reshape(soft_target, shape=[-1, self._class_dim])
+        return soft_target
+
+    def _crossentropy(self, input, target, use_pure_fp16=False):
+        if self._label_smoothing:
+            target = self._labelsmoothing(target)
+            input = -F.log_softmax(input, axis=-1)
+            cost = paddle.sum(target * input, axis=-1)
+        else:
+            cost = F.cross_entropy(input=input, label=target)
+        if use_pure_fp16:
+            avg_cost = paddle.sum(cost)
+        else:
+            avg_cost = paddle.mean(cost)
+        return avg_cost
+
+    def __call__(self, input, target):
+        return self._crossentropy(input, target)
+
+
+def build_loss(config, epsilon=None):
+    class_dim = config["class_dim"]
+    loss_func = Loss(class_dim=class_dim, epsilon=epsilon)
+    return loss_func
+
+
+class LossDistill(Loss):
+    def __init__(self, model_name_list, class_dim=1000, epsilon=None):
+        assert class_dim > 1, "class_dim=%d is not larger than 1" % (class_dim)
+        self._class_dim = class_dim
+        if epsilon is not None and epsilon >= 0.0 and epsilon <= 1.0:
+            self._epsilon = epsilon
+            self._label_smoothing = True
+        else:
+            self._epsilon = None
+            self._label_smoothing = False
+
+        self.model_name_list = model_name_list
+        assert len(self.model_name_list) > 1, "error"
+
+    def __call__(self, input, target):
+        losses = {}
+        for k in self.model_name_list:
+            inp = input[k]
+            losses[k] = self._crossentropy(inp, target)
+        return losses
+
+
+class KLJSLoss(object):
+    def __init__(self, mode="kl"):
+        assert mode in [
+            "kl",
+            "js",
+            "KL",
+            "JS",
+        ], "mode can only be one of ['kl', 'js', 'KL', 'JS']"
+        self.mode = mode
+
+    def __call__(self, p1, p2, reduction="mean"):
+        p1 = F.softmax(p1, axis=-1)
+        p2 = F.softmax(p2, axis=-1)
+
+        loss = paddle.multiply(p2, paddle.log((p2 + 1e-5) / (p1 + 1e-5) + 1e-5))
+
+        if self.mode.lower() == "js":
+            loss += paddle.multiply(p1, paddle.log((p1 + 1e-5) / (p2 + 1e-5) + 1e-5))
+            loss *= 0.5
+        if reduction == "mean":
+            loss = paddle.mean(loss)
+        elif reduction == "none" or reduction is None:
+            return loss
+        else:
+            loss = paddle.sum(loss)
+        return loss
+
+
+class DMLLoss(object):
+    def __init__(self, model_name_pairs, mode="js"):
+        self.model_name_pairs = self._check_model_name_pairs(model_name_pairs)
+        self.kljs_loss = KLJSLoss(mode=mode)
+
+    def _check_model_name_pairs(self, model_name_pairs):
+        if not isinstance(model_name_pairs, list):
+            return []
+        elif isinstance(model_name_pairs[0], list) and isinstance(
+            model_name_pairs[0][0], str
+        ):
+            return model_name_pairs
+        else:
+            return [model_name_pairs]
+
+    def __call__(self, predicts, target=None):
+        loss_dict = dict()
+        for pairs in self.model_name_pairs:
+            p1 = predicts[pairs[0]]
+            p2 = predicts[pairs[1]]
+
+            loss_dict[pairs[0] + "_" + pairs[1]] = self.kljs_loss(p1, p2)
+
+        return loss_dict
+
+
+# def build_distill_loss(config, epsilon=None):
+#     class_dim = config['class_dim']
+#     loss = LossDistill(model_name_list=['student', 'student1'], )
+#     return loss_func
diff --git a/test_tipc/supplementary/metric.py b/test_tipc/supplementary/metric.py
new file mode 100644
index 0000000..4851afc
--- /dev/null
+++ b/test_tipc/supplementary/metric.py
@@ -0,0 +1,60 @@
+import paddle
+import paddle.nn.functional as F
+from collections import OrderedDict
+
+
+def create_metric(
+    out,
+    label,
+    architecture=None,
+    topk=5,
+    classes_num=1000,
+    use_distillation=False,
+    mode="train",
+):
+    """
+    Create measures of model accuracy, such as top1 and top5
+
+    Args:
+        out(variable): model output variable
+        feeds(dict): dict of model input variables(included label)
+        topk(int): usually top5
+        classes_num(int): num of classes
+        use_distillation(bool): whether to use distillation training
+        mode(str): mode, train/valid
+
+    Returns:
+        fetches(dict): dict of measures
+    """
+    # if architecture["name"] == "GoogLeNet":
+    #     assert len(out) == 3, "GoogLeNet should have 3 outputs"
+    #     out = out[0]
+    # else:
+    #     # just need student label to get metrics
+    #     if use_distillation:
+    #         out = out[1]
+    softmax_out = F.softmax(out)
+
+    fetches = OrderedDict()
+    # set top1 to fetches
+    top1 = paddle.metric.accuracy(softmax_out, label=label, k=1)
+    # set topk to fetches
+    k = min(topk, classes_num)
+    topk = paddle.metric.accuracy(softmax_out, label=label, k=k)
+
+    # multi cards' eval
+    if mode != "train" and paddle.distributed.get_world_size() > 1:
+        top1 = (
+            paddle.distributed.all_reduce(top1, op=paddle.distributed.ReduceOp.SUM)
+            / paddle.distributed.get_world_size()
+        )
+        topk = (
+            paddle.distributed.all_reduce(topk, op=paddle.distributed.ReduceOp.SUM)
+            / paddle.distributed.get_world_size()
+        )
+
+    fetches["top1"] = top1
+    topk_name = "top{}".format(k)
+    fetches[topk_name] = topk
+
+    return fetches
diff --git a/test_tipc/supplementary/mv3.py b/test_tipc/supplementary/mv3.py
new file mode 100644
index 0000000..f891b86
--- /dev/null
+++ b/test_tipc/supplementary/mv3.py
@@ -0,0 +1,518 @@
+# copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+import paddle
+from paddle import ParamAttr
+import paddle.nn as nn
+import paddle.nn.functional as F
+from paddle.nn.functional import hardswish, hardsigmoid
+from paddle.nn import Conv2D, BatchNorm, Linear, Dropout
+from paddle.nn import AdaptiveAvgPool2D, MaxPool2D, AvgPool2D
+from paddle.regularizer import L2Decay
+import math
+
+from paddle.utils.cpp_extension import load
+
+# jit compile custom op
+custom_ops = load(
+    name="custom_jit_ops",
+    sources=["./custom_op/custom_relu_op.cc", "./custom_op/custom_relu_op.cu"],
+)
+
+
+def make_divisible(v, divisor=8, min_value=None):
+    if min_value is None:
+        min_value = divisor
+    new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
+    if new_v < 0.9 * v:
+        new_v += divisor
+    return new_v
+
+
+class MobileNetV3(nn.Layer):
+    def __init__(
+        self,
+        scale=1.0,
+        model_name="small",
+        dropout_prob=0.2,
+        class_dim=1000,
+        use_custom_relu=False,
+    ):
+        super(MobileNetV3, self).__init__()
+        self.use_custom_relu = use_custom_relu
+
+        inplanes = 16
+        if model_name == "large":
+            self.cfg = [
+                # k, exp, c,  se,     nl,  s,
+                [3, 16, 16, False, "relu", 1],
+                [3, 64, 24, False, "relu", 2],
+                [3, 72, 24, False, "relu", 1],
+                [5, 72, 40, True, "relu", 2],
+                [5, 120, 40, True, "relu", 1],
+                [5, 120, 40, True, "relu", 1],
+                [3, 240, 80, False, "hardswish", 2],
+                [3, 200, 80, False, "hardswish", 1],
+                [3, 184, 80, False, "hardswish", 1],
+                [3, 184, 80, False, "hardswish", 1],
+                [3, 480, 112, True, "hardswish", 1],
+                [3, 672, 112, True, "hardswish", 1],
+                [5, 672, 160, True, "hardswish", 2],
+                [5, 960, 160, True, "hardswish", 1],
+                [5, 960, 160, True, "hardswish", 1],
+            ]
+            self.cls_ch_squeeze = 960
+            self.cls_ch_expand = 1280
+        elif model_name == "small":
+            self.cfg = [
+                # k, exp, c,  se,     nl,  s,
+                [3, 16, 16, True, "relu", 2],
+                [3, 72, 24, False, "relu", 2],
+                [3, 88, 24, False, "relu", 1],
+                [5, 96, 40, True, "hardswish", 2],
+                [5, 240, 40, True, "hardswish", 1],
+                [5, 240, 40, True, "hardswish", 1],
+                [5, 120, 48, True, "hardswish", 1],
+                [5, 144, 48, True, "hardswish", 1],
+                [5, 288, 96, True, "hardswish", 2],
+                [5, 576, 96, True, "hardswish", 1],
+                [5, 576, 96, True, "hardswish", 1],
+            ]
+            self.cls_ch_squeeze = 576
+            self.cls_ch_expand = 1280
+        else:
+            raise NotImplementedError(
+                "mode[{}_model] is not implemented!".format(model_name)
+            )
+
+        self.conv1 = ConvBNLayer(
+            in_c=3,
+            out_c=make_divisible(inplanes * scale),
+            filter_size=3,
+            stride=2,
+            padding=1,
+            num_groups=1,
+            if_act=True,
+            act="hardswish",
+            name="conv1",
+            use_custom_relu=self.use_custom_relu,
+        )
+
+        self.block_list = []
+        i = 0
+        inplanes = make_divisible(inplanes * scale)
+        for k, exp, c, se, nl, s in self.cfg:
+            block = self.add_sublayer(
+                "conv" + str(i + 2),
+                ResidualUnit(
+                    in_c=inplanes,
+                    mid_c=make_divisible(scale * exp),
+                    out_c=make_divisible(scale * c),
+                    filter_size=k,
+                    stride=s,
+                    use_se=se,
+                    act=nl,
+                    name="conv" + str(i + 2),
+                    use_custom_relu=self.use_custom_relu,
+                ),
+            )
+            self.block_list.append(block)
+            inplanes = make_divisible(scale * c)
+            i += 1
+
+        self.last_second_conv = ConvBNLayer(
+            in_c=inplanes,
+            out_c=make_divisible(scale * self.cls_ch_squeeze),
+            filter_size=1,
+            stride=1,
+            padding=0,
+            num_groups=1,
+            if_act=True,
+            act="hardswish",
+            name="conv_last",
+            use_custom_relu=self.use_custom_relu,
+        )
+
+        self.pool = AdaptiveAvgPool2D(1)
+
+        self.last_conv = Conv2D(
+            in_channels=make_divisible(scale * self.cls_ch_squeeze),
+            out_channels=self.cls_ch_expand,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            weight_attr=ParamAttr(),
+            bias_attr=False,
+        )
+
+        self.dropout = Dropout(p=dropout_prob, mode="downscale_in_infer")
+
+        self.out = Linear(
+            self.cls_ch_expand,
+            class_dim,
+            weight_attr=ParamAttr(),
+            bias_attr=ParamAttr(),
+        )
+
+    def forward(self, inputs, label=None):
+        x = self.conv1(inputs)
+
+        for block in self.block_list:
+            x = block(x)
+
+        x = self.last_second_conv(x)
+        x = self.pool(x)
+
+        x = self.last_conv(x)
+        x = hardswish(x)
+        x = self.dropout(x)
+        x = paddle.flatten(x, start_axis=1, stop_axis=-1)
+        x = self.out(x)
+        return x
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(
+        self,
+        in_c,
+        out_c,
+        filter_size,
+        stride,
+        padding,
+        num_groups=1,
+        if_act=True,
+        act=None,
+        use_cudnn=True,
+        name="",
+        use_custom_relu=False,
+    ):
+        super(ConvBNLayer, self).__init__()
+        self.if_act = if_act
+        self.act = act
+        self.conv = Conv2D(
+            in_channels=in_c,
+            out_channels=out_c,
+            kernel_size=filter_size,
+            stride=stride,
+            padding=padding,
+            groups=num_groups,
+            weight_attr=ParamAttr(),
+            bias_attr=False,
+        )
+        self.bn = BatchNorm(
+            num_channels=out_c,
+            act=None,
+            param_attr=ParamAttr(regularizer=L2Decay(0.0)),
+            bias_attr=ParamAttr(regularizer=L2Decay(0.0)),
+        )
+        # moving_mean_name=name + "_bn_mean",
+        # moving_variance_name=name + "_bn_variance")
+
+        self.use_custom_relu = use_custom_relu
+
+    def forward(self, x):
+        x = self.conv(x)
+        x = self.bn(x)
+        if self.if_act:
+            if self.act == "relu":
+                if self.use_custom_relu:
+                    x = custom_ops.custom_relu(x)
+                else:
+                    x = F.relu(x)
+            elif self.act == "hardswish":
+                x = hardswish(x)
+            else:
+                print("The activation function is selected incorrectly.")
+                exit()
+        return x
+
+
+class ResidualUnit(nn.Layer):
+    def __init__(
+        self,
+        in_c,
+        mid_c,
+        out_c,
+        filter_size,
+        stride,
+        use_se,
+        act=None,
+        name="",
+        use_custom_relu=False,
+    ):
+        super(ResidualUnit, self).__init__()
+        self.if_shortcut = stride == 1 and in_c == out_c
+        self.if_se = use_se
+
+        self.use_custom_relu = use_custom_relu
+
+        self.expand_conv = ConvBNLayer(
+            in_c=in_c,
+            out_c=mid_c,
+            filter_size=1,
+            stride=1,
+            padding=0,
+            if_act=True,
+            act=act,
+            name=name + "_expand",
+            use_custom_relu=self.use_custom_relu,
+        )
+        self.bottleneck_conv = ConvBNLayer(
+            in_c=mid_c,
+            out_c=mid_c,
+            filter_size=filter_size,
+            stride=stride,
+            padding=int((filter_size - 1) // 2),
+            num_groups=mid_c,
+            if_act=True,
+            act=act,
+            name=name + "_depthwise",
+            use_custom_relu=self.use_custom_relu,
+        )
+        if self.if_se:
+            self.mid_se = SEModule(mid_c, name=name + "_se")
+        self.linear_conv = ConvBNLayer(
+            in_c=mid_c,
+            out_c=out_c,
+            filter_size=1,
+            stride=1,
+            padding=0,
+            if_act=False,
+            act=None,
+            name=name + "_linear",
+            use_custom_relu=self.use_custom_relu,
+        )
+
+    def forward(self, inputs):
+        x = self.expand_conv(inputs)
+        x = self.bottleneck_conv(x)
+        if self.if_se:
+            x = self.mid_se(x)
+        x = self.linear_conv(x)
+        if self.if_shortcut:
+            x = paddle.add(inputs, x)
+        return x
+
+
+class SEModule(nn.Layer):
+    def __init__(self, channel, reduction=4, name=""):
+        super(SEModule, self).__init__()
+        self.avg_pool = AdaptiveAvgPool2D(1)
+        self.conv1 = Conv2D(
+            in_channels=channel,
+            out_channels=channel // reduction,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            weight_attr=ParamAttr(),
+            bias_attr=ParamAttr(),
+        )
+        self.conv2 = Conv2D(
+            in_channels=channel // reduction,
+            out_channels=channel,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            weight_attr=ParamAttr(),
+            bias_attr=ParamAttr(),
+        )
+
+    def forward(self, inputs):
+        outputs = self.avg_pool(inputs)
+        outputs = self.conv1(outputs)
+        outputs = F.relu(outputs)
+        outputs = self.conv2(outputs)
+        outputs = hardsigmoid(outputs, slope=0.2, offset=0.5)
+        return paddle.multiply(x=inputs, y=outputs)
+
+
+def MobileNetV3_small_x0_35(**args):
+    model = MobileNetV3(model_name="small", scale=0.35, **args)
+    return model
+
+
+def MobileNetV3_small_x0_5(**args):
+    model = MobileNetV3(model_name="small", scale=0.5, **args)
+    return model
+
+
+def MobileNetV3_small_x0_75(**args):
+    model = MobileNetV3(model_name="small", scale=0.75, **args)
+    return model
+
+
+def MobileNetV3_small_x1_0(**args):
+    model = MobileNetV3(model_name="small", scale=1.0, **args)
+    return model
+
+
+def MobileNetV3_small_x1_25(**args):
+    model = MobileNetV3(model_name="small", scale=1.25, **args)
+    return model
+
+
+def MobileNetV3_large_x0_35(**args):
+    model = MobileNetV3(model_name="large", scale=0.35, **args)
+    return model
+
+
+def MobileNetV3_large_x0_5(**args):
+    model = MobileNetV3(model_name="large", scale=0.5, **args)
+    return model
+
+
+def MobileNetV3_large_x0_75(**args):
+    model = MobileNetV3(model_name="large", scale=0.75, **args)
+    return model
+
+
+def MobileNetV3_large_x1_0(**args):
+    model = MobileNetV3(model_name="large", scale=1.0, **args)
+    return model
+
+
+def MobileNetV3_large_x1_25(**args):
+    model = MobileNetV3(model_name="large", scale=1.25, **args)
+    return
+
+
+class DistillMV3(nn.Layer):
+    def __init__(
+        self,
+        scale=1.0,
+        model_name="small",
+        dropout_prob=0.2,
+        class_dim=1000,
+        args=None,
+        use_custom_relu=False,
+    ):
+        super(DistillMV3, self).__init__()
+
+        self.student = MobileNetV3(
+            model_name=model_name,
+            scale=scale,
+            class_dim=class_dim,
+            use_custom_relu=use_custom_relu,
+        )
+
+        self.student1 = MobileNetV3(
+            model_name=model_name,
+            scale=scale,
+            class_dim=class_dim,
+            use_custom_relu=use_custom_relu,
+        )
+
+    def forward(self, inputs, label=None):
+        predicts = dict()
+        predicts["student"] = self.student(inputs, label)
+        predicts["student1"] = self.student1(inputs, label)
+        return predicts
+
+
+def distillmv3_large_x0_5(**args):
+    model = DistillMV3(model_name="large", scale=0.5, **args)
+    return model
+
+
+class SiameseMV3(nn.Layer):
+    def __init__(
+        self,
+        scale=1.0,
+        model_name="small",
+        dropout_prob=0.2,
+        class_dim=1000,
+        args=None,
+        use_custom_relu=False,
+    ):
+        super(SiameseMV3, self).__init__()
+
+        self.net = MobileNetV3(
+            model_name=model_name,
+            scale=scale,
+            class_dim=class_dim,
+            use_custom_relu=use_custom_relu,
+        )
+        self.net1 = MobileNetV3(
+            model_name=model_name,
+            scale=scale,
+            class_dim=class_dim,
+            use_custom_relu=use_custom_relu,
+        )
+
+    def forward(self, inputs, label=None):
+        # net
+        x = self.net.conv1(inputs)
+        for block in self.net.block_list:
+            x = block(x)
+
+        # net1
+        x1 = self.net1.conv1(inputs)
+        for block in self.net1.block_list:
+            x1 = block(x1)
+        # add
+        x = x + x1
+
+        x = self.net.last_second_conv(x)
+        x = self.net.pool(x)
+
+        x = self.net.last_conv(x)
+        x = hardswish(x)
+        x = self.net.dropout(x)
+        x = paddle.flatten(x, start_axis=1, stop_axis=-1)
+        x = self.net.out(x)
+        return x
+
+
+def siamese_mv3(class_dim, use_custom_relu):
+    model = SiameseMV3(
+        scale=0.5,
+        model_name="large",
+        class_dim=class_dim,
+        use_custom_relu=use_custom_relu,
+    )
+    return model
+
+
+def build_model(config):
+    model_type = config["model_type"]
+    if model_type == "cls":
+        class_dim = config["MODEL"]["class_dim"]
+        use_custom_relu = config["MODEL"]["use_custom_relu"]
+        if "siamese" in config["MODEL"] and config["MODEL"]["siamese"] is True:
+            model = siamese_mv3(class_dim=class_dim, use_custom_relu=use_custom_relu)
+        else:
+            model = MobileNetV3_large_x0_5(
+                class_dim=class_dim, use_custom_relu=use_custom_relu
+            )
+
+    elif model_type == "cls_distill":
+        class_dim = config["MODEL"]["class_dim"]
+        use_custom_relu = config["MODEL"]["use_custom_relu"]
+        model = distillmv3_large_x0_5(
+            class_dim=class_dim, use_custom_relu=use_custom_relu
+        )
+
+    elif model_type == "cls_distill_multiopt":
+        class_dim = config["MODEL"]["class_dim"]
+        use_custom_relu = config["MODEL"]["use_custom_relu"]
+        model = distillmv3_large_x0_5(class_dim=100, use_custom_relu=use_custom_relu)
+    else:
+        raise ValueError("model_type should be one of ['']")
+
+    return model
diff --git a/test_tipc/supplementary/mv3_distill.yml b/test_tipc/supplementary/mv3_distill.yml
new file mode 100644
index 0000000..57b8130
--- /dev/null
+++ b/test_tipc/supplementary/mv3_distill.yml
@@ -0,0 +1,30 @@
+
+class_dim: 100
+total_images: 50000
+epochs: 1000
+topk: 5
+save_model_dir: ./output/
+use_gpu: True
+model_type: cls_distill
+
+LEARNING_RATE:
+    function: 'Cosine'
+    params:
+        lr: 0.001
+        warmup_epoch: 5
+
+OPTIMIZER:
+    function: 'Momentum'
+    params:
+        momentum: 0.9
+    regularizer:
+        function: 'L2'
+        factor: 0.00002
+
+TRAIN:
+    batch_size: 1280
+    num_workers: 4
+
+VALID:
+    batch_size: 64
+    num_workers: 4
diff --git a/test_tipc/supplementary/mv3_large_x0_5.yml b/test_tipc/supplementary/mv3_large_x0_5.yml
new file mode 100644
index 0000000..fd7d308
--- /dev/null
+++ b/test_tipc/supplementary/mv3_large_x0_5.yml
@@ -0,0 +1,48 @@
+
+class_dim: 100
+total_images: 50000
+epoch: 1000
+topk: 5
+save_model_dir: ./output/
+use_gpu: True
+model_type: cls
+use_custom_relu: false
+pretrained_model: 
+checkpoints: 
+save_model_dir: ./output/cls/
+
+# slim
+quant_train: false
+prune_train: false
+
+MODEL:
+    class_dim: 100
+    use_custom_relu: False
+    siamese: False
+
+AMP:
+    use_amp: False
+    scale_loss: 1024.0
+    use_dynamic_loss_scale: True
+
+LEARNING_RATE:
+    function: 'Cosine'
+    params:
+        lr: 0.001
+        warmup_epoch: 5
+
+OPTIMIZER:
+    function: 'Momentum'
+    params:
+        momentum: 0.9
+    regularizer:
+        function: 'L2'
+        factor: 0.00002
+
+TRAIN:
+    batch_size: 1280
+    num_workers: 4
+
+VALID:
+    batch_size: 64
+    num_workers: 4
diff --git a/test_tipc/supplementary/optimizer.py b/test_tipc/supplementary/optimizer.py
new file mode 100644
index 0000000..cb7ca3a
--- /dev/null
+++ b/test_tipc/supplementary/optimizer.py
@@ -0,0 +1,330 @@
+import sys
+import math
+from paddle.optimizer.lr import LinearWarmup
+from paddle.optimizer.lr import PiecewiseDecay
+from paddle.optimizer.lr import CosineAnnealingDecay
+from paddle.optimizer.lr import ExponentialDecay
+import paddle
+import paddle.regularizer as regularizer
+from copy import deepcopy
+
+
+class Cosine(CosineAnnealingDecay):
+    """
+    Cosine learning rate decay
+    lr = 0.05 * (math.cos(epoch * (math.pi / epochs)) + 1)
+    Args:
+        lr(float): initial learning rate
+        step_each_epoch(int): steps each epoch
+        epochs(int): total training epochs
+    """
+
+    def __init__(self, lr, step_each_epoch, epochs, **kwargs):
+        super(Cosine, self).__init__(
+            learning_rate=lr,
+            T_max=step_each_epoch * epochs,
+        )
+
+        self.update_specified = False
+
+
+class Piecewise(PiecewiseDecay):
+    """
+    Piecewise learning rate decay
+    Args:
+        lr(float): initial learning rate
+        step_each_epoch(int): steps each epoch
+        decay_epochs(list): piecewise decay epochs
+        gamma(float): decay factor
+    """
+
+    def __init__(self, lr, step_each_epoch, decay_epochs, gamma=0.1, **kwargs):
+        boundaries = [step_each_epoch * e for e in decay_epochs]
+        lr_values = [lr * (gamma**i) for i in range(len(boundaries) + 1)]
+        super(Piecewise, self).__init__(boundaries=boundaries, values=lr_values)
+
+        self.update_specified = False
+
+
+class CosineWarmup(LinearWarmup):
+    """
+    Cosine learning rate decay with warmup
+    [0, warmup_epoch): linear warmup
+    [warmup_epoch, epochs): cosine decay
+    Args:
+        lr(float): initial learning rate
+        step_each_epoch(int): steps each epoch
+        epochs(int): total training epochs
+        warmup_epoch(int): epoch num of warmup
+    """
+
+    def __init__(self, lr, step_each_epoch, epochs, warmup_epoch=5, **kwargs):
+        assert (
+            epochs > warmup_epoch
+        ), "total epoch({}) should be larger than warmup_epoch({}) in CosineWarmup.".format(
+            epochs, warmup_epoch
+        )
+        warmup_step = warmup_epoch * step_each_epoch
+        start_lr = 0.0
+        end_lr = lr
+        lr_sch = Cosine(lr, step_each_epoch, epochs - warmup_epoch)
+
+        super(CosineWarmup, self).__init__(
+            learning_rate=lr_sch,
+            warmup_steps=warmup_step,
+            start_lr=start_lr,
+            end_lr=end_lr,
+        )
+
+        self.update_specified = False
+
+
+class ExponentialWarmup(LinearWarmup):
+    """
+    Exponential learning rate decay with warmup
+    [0, warmup_epoch): linear warmup
+    [warmup_epoch, epochs): Exponential decay
+    Args:
+        lr(float): initial learning rate
+        step_each_epoch(int): steps each epoch
+        decay_epochs(float): decay epochs
+        decay_rate(float): decay rate
+        warmup_epoch(int): epoch num of warmup
+    """
+
+    def __init__(
+        self,
+        lr,
+        step_each_epoch,
+        decay_epochs=2.4,
+        decay_rate=0.97,
+        warmup_epoch=5,
+        **kwargs,
+    ):
+        warmup_step = warmup_epoch * step_each_epoch
+        start_lr = 0.0
+        end_lr = lr
+        lr_sch = ExponentialDecay(lr, decay_rate)
+
+        super(ExponentialWarmup, self).__init__(
+            learning_rate=lr_sch,
+            warmup_steps=warmup_step,
+            start_lr=start_lr,
+            end_lr=end_lr,
+        )
+
+        # NOTE: hac method to update exponential lr scheduler
+        self.update_specified = True
+        self.update_start_step = warmup_step
+        self.update_step_interval = int(decay_epochs * step_each_epoch)
+        self.step_each_epoch = step_each_epoch
+
+
+class LearningRateBuilder:
+    """
+    Build learning rate variable
+    https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/layers_cn.html
+    Args:
+        function(str): class name of learning rate
+        params(dict): parameters used for init the class
+    """
+
+    def __init__(
+        self, function="Linear", params={"lr": 0.1, "steps": 100, "end_lr": 0.0}
+    ):
+        self.function = function
+        self.params = params
+
+    def __call__(self):
+        mod = sys.modules[__name__]
+        lr = getattr(mod, self.function)(**self.params)
+        return lr
+
+
+class L1Decay(object):
+    """
+    L1 Weight Decay Regularization, which encourages the weights to be sparse.
+    Args:
+        factor(float): regularization coeff. Default:0.0.
+    """
+
+    def __init__(self, factor=0.0):
+        super(L1Decay, self).__init__()
+        self.factor = factor
+
+    def __call__(self):
+        reg = regularizer.L1Decay(self.factor)
+        return reg
+
+
+class L2Decay(object):
+    """
+    L2 Weight Decay Regularization, which encourages the weights to be sparse.
+    Args:
+        factor(float): regularization coeff. Default:0.0.
+    """
+
+    def __init__(self, factor=0.0):
+        super(L2Decay, self).__init__()
+        self.factor = factor
+
+    def __call__(self):
+        reg = regularizer.L2Decay(self.factor)
+        return reg
+
+
+class Momentum(object):
+    """
+    Simple Momentum optimizer with velocity state.
+    Args:
+        learning_rate (float|Variable) - The learning rate used to update parameters.
+            Can be a float value or a Variable with one float value as data element.
+        momentum (float) - Momentum factor.
+        regularization (WeightDecayRegularizer, optional) - The strategy of regularization.
+    """
+
+    def __init__(
+        self, learning_rate, momentum, parameter_list=None, regularization=None, **args
+    ):
+        super(Momentum, self).__init__()
+        self.learning_rate = learning_rate
+        self.momentum = momentum
+        self.parameter_list = parameter_list
+        self.regularization = regularization
+
+    def __call__(self):
+        opt = paddle.optimizer.Momentum(
+            learning_rate=self.learning_rate,
+            momentum=self.momentum,
+            parameters=self.parameter_list,
+            weight_decay=self.regularization,
+        )
+        return opt
+
+
+class RMSProp(object):
+    """
+    Root Mean Squared Propagation (RMSProp) is an unpublished, adaptive learning rate method.
+    Args:
+        learning_rate (float|Variable) - The learning rate used to update parameters.
+            Can be a float value or a Variable with one float value as data element.
+        momentum (float) - Momentum factor.
+        rho (float) - rho value in equation.
+        epsilon (float) - avoid division by zero, default is 1e-6.
+        regularization (WeightDecayRegularizer, optional) - The strategy of regularization.
+    """
+
+    def __init__(
+        self,
+        learning_rate,
+        momentum,
+        rho=0.95,
+        epsilon=1e-6,
+        parameter_list=None,
+        regularization=None,
+        **args,
+    ):
+        super(RMSProp, self).__init__()
+        self.learning_rate = learning_rate
+        self.momentum = momentum
+        self.rho = rho
+        self.epsilon = epsilon
+        self.parameter_list = parameter_list
+        self.regularization = regularization
+
+    def __call__(self):
+        opt = paddle.optimizer.RMSProp(
+            learning_rate=self.learning_rate,
+            momentum=self.momentum,
+            rho=self.rho,
+            epsilon=self.epsilon,
+            parameters=self.parameter_list,
+            weight_decay=self.regularization,
+        )
+        return opt
+
+
+class OptimizerBuilder(object):
+    """
+    Build optimizer
+    Args:
+        function(str): optimizer name of learning rate
+        params(dict): parameters used for init the class
+        regularizer (dict): parameters used for create regularization
+    """
+
+    def __init__(self, function="Momentum", params={"momentum": 0.9}, regularizer=None):
+        self.function = function
+        self.params = params
+        # create regularizer
+        if regularizer is not None:
+            mod = sys.modules[__name__]
+            reg_func = regularizer["function"] + "Decay"
+            del regularizer["function"]
+            reg = getattr(mod, reg_func)(**regularizer)()
+            self.params["regularization"] = reg
+
+    def __call__(self, learning_rate, parameter_list=None):
+        mod = sys.modules[__name__]
+        opt = getattr(mod, self.function)
+        return opt(
+            learning_rate=learning_rate, parameter_list=parameter_list, **self.params
+        )()
+
+
+def create_optimizer(config, parameter_list=None):
+    """
+    Create an optimizer using config, usually including
+    learning rate and regularization.
+
+    Args:
+        config(dict):  such as
+        {
+            'LEARNING_RATE':
+                {'function': 'Cosine',
+                 'params': {'lr': 0.1}
+                },
+            'OPTIMIZER':
+                {'function': 'Momentum',
+                 'params':{'momentum': 0.9},
+                 'regularizer':
+                    {'function': 'L2', 'factor': 0.0001}
+                }
+        }
+
+    Returns:
+        an optimizer instance
+    """
+    # create learning_rate instance
+    lr_config = config["LEARNING_RATE"]
+    lr_config["params"].update(
+        {
+            "epochs": config["epoch"],
+            "step_each_epoch": config["total_images"] // config["TRAIN"]["batch_size"],
+        }
+    )
+    lr = LearningRateBuilder(**lr_config)()
+
+    # create optimizer instance
+    opt_config = deepcopy(config["OPTIMIZER"])
+
+    opt = OptimizerBuilder(**opt_config)
+    return opt(lr, parameter_list), lr
+
+
+def create_multi_optimizer(config, parameter_list=None):
+    """ """
+    # create learning_rate instance
+    lr_config = config["LEARNING_RATE"]
+    lr_config["params"].update(
+        {
+            "epochs": config["epoch"],
+            "step_each_epoch": config["total_images"] // config["TRAIN"]["batch_size"],
+        }
+    )
+    lr = LearningRateBuilder(**lr_config)()
+
+    # create optimizer instance
+    opt_config = deepcopy.copy(config["OPTIMIZER"])
+    opt = OptimizerBuilder(**opt_config)
+    return opt(lr, parameter_list), lr
diff --git a/test_tipc/supplementary/readme.md b/test_tipc/supplementary/readme.md
new file mode 100644
index 0000000..8e822fa
--- /dev/null
+++ b/test_tipc/supplementary/readme.md
@@ -0,0 +1,74 @@
+
+# TIPC Linux端补充训练功能测试
+
+Linux端基础训练预测功能测试的主程序为test_train_python.sh，可以测试基于Python的模型训练、评估等基本功能，包括裁剪、量化、蒸馏训练。
+
+![](./test_tipc/tipc_train.png)
+
+测试链条如上图所示，主要测试内容有带共享权重，自定义OP的模型的正常训练和slim相关功能训练流程是否正常。
+
+
+# 2. 测试流程
+
+本节介绍补充链条的测试流程
+
+## 2.1 安装依赖
+
+- 安装PaddlePaddle >= 2.2
+- 安装其他依赖
+
+```
+pip3 install -r requirements.txt
+```
+
+## 2.2 功能测试
+
+`test_train_python.sh`包含2种运行模式，每种模式的运行数据不同，分别用于测试训练是否正常，分别是：
+
+- 模式1：lite_train_lite_infer，使用少量数据训练，用于快速验证训练到预测的走通流程，不验证精度和速度；
+
+```
+bash test_tipc/test_train_python.sh ./test_tipc/train_infer_python.txt 'lite_train_lite_infer'
+```
+
+- 模式2：whole_train_whole_infer，使用全量数据训练，用于快速验证训练到预测的走通流程，验证模型最终训练精度；
+
+```
+bash test_tipc/test_train_python.sh ./test_tipc/train_infer_python.txt 'whole_train_whole_infer'
+```
+
+如果是运行量化裁剪等训练方式，需要使用不同的配置文件。量化训练的测试指令如下：
+```
+bash test_tipc/test_train_python.sh ./test_tipc/train_infer_python_PACT.txt 'lite_train_lite_infer'
+```
+
+同理，FPGM裁剪的运行方式如下：
+```
+bash test_tipc/test_train_python.sh ./test_tipc/train_infer_python_FPGM.txt 'lite_train_lite_infer'
+```
+
+多机多卡的运行配置文件分别为 `train_infer_python_fleet.txt`, `train_infer_python_FPGM_fleet.txt` 和 `train_infer_python_PACT_fleet.txt`。
+运行时，需要修改配置文件中的 `gpu_list:xx.xx.xx.xx,yy.yy.yy.yy;0,1`。 将 `xx.xx.xx.xx` 替换为具体的 `ip` 地址，各个`ip`地址之间用`,`分隔。 另外，和单机训练
+不同，启动多机多卡训练需要在多机的每个节点上分别运行命令。以多机多卡量化训练为例，指令如下：
+```
+bash test_tipc/test_train_python.sh ./test_tipc/train_infer_python_PACT_fleet.txt 'lite_train_lite_infer'
+```
+
+运行相应指令后，在`test_tipc/output`文件夹下自动会保存运行日志。如'lite_train_lite_infer'模式运行后，在test_tipc/extra_output文件夹有以下文件：
+
+```
+test_tipc/output/
+|- results_python.log    # 运行指令状态的日志
+```
+
+其中results_python.log中包含了每条指令的运行状态，如果运行成功会输出：
+
+```
+Run successfully with command - python3.7 train.py -c mv3_large_x0_5.yml -o  use_gpu=True     epoch=20       AMP.use_amp=True TRAIN.batch_size=1280  use_custom_relu=False model_type=cls MODEL.siamese=False  !
+Run successfully with command - python3.7 train.py -c mv3_large_x0_5.yml -o  use_gpu=True     epoch=2       AMP.use_amp=True TRAIN.batch_size=1280  use_custom_relu=False model_type=cls MODEL.siamese=False  !
+Run successfully with command - python3.7 train.py -c mv3_large_x0_5.yml -o  use_gpu=True     epoch=2       AMP.use_amp=True TRAIN.batch_size=1280  use_custom_relu=False model_type=cls MODEL.siamese=True  !
+Run successfully with command - python3.7 train.py -c mv3_large_x0_5.yml -o  use_gpu=True     epoch=2       AMP.use_amp=True TRAIN.batch_size=1280  use_custom_relu=False model_type=cls_distill MODEL.siamese=False  !
+Run successfully with command - python3.7 train.py -c mv3_large_x0_5.yml -o  use_gpu=True     epoch=2       AMP.use_amp=True TRAIN.batch_size=1280  use_custom_relu=False model_type=cls_distill MODEL.siamese=True  !
+Run successfully with command - python3.7 train.py -c mv3_large_x0_5.yml -o  use_gpu=True     epoch=2       AMP.use_amp=True TRAIN.batch_size=1280  use_custom_relu=False model_type=cls_distill_multiopt MODEL.siamese=False  !
+
+```
diff --git a/test_tipc/supplementary/requirements.txt b/test_tipc/supplementary/requirements.txt
new file mode 100644
index 0000000..c55500a
--- /dev/null
+++ b/test_tipc/supplementary/requirements.txt
@@ -0,0 +1 @@
+paddleslim==2.2.1
diff --git a/test_tipc/supplementary/slim/__init__.py b/test_tipc/supplementary/slim/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/test_tipc/supplementary/slim/slim_fpgm.py b/test_tipc/supplementary/slim/slim_fpgm.py
new file mode 100644
index 0000000..0304744
--- /dev/null
+++ b/test_tipc/supplementary/slim/slim_fpgm.py
@@ -0,0 +1,21 @@
+import paddleslim
+import paddle
+import numpy as np
+
+from paddleslim.dygraph import FPGMFilterPruner
+
+
+def prune_model(model, input_shape, prune_ratio=0.1):
+    flops = paddle.flops(model, input_shape)
+    pruner = FPGMFilterPruner(model, input_shape)
+
+    params_sensitive = {}
+    for param in model.parameters():
+        if "transpose" not in param.name and "linear" not in param.name:
+            # set prune ratio as 10%. The larger the value, the more convolution weights will be cropped
+            params_sensitive[param.name] = prune_ratio
+
+    plan = pruner.prune_vars(params_sensitive, [0])
+
+    flops = paddle.flops(model, input_shape)
+    return model
diff --git a/test_tipc/supplementary/slim/slim_quant.py b/test_tipc/supplementary/slim/slim_quant.py
new file mode 100644
index 0000000..41db3eb
--- /dev/null
+++ b/test_tipc/supplementary/slim/slim_quant.py
@@ -0,0 +1,48 @@
+import paddle
+import numpy as np
+import os
+import paddle.nn as nn
+import paddleslim
+
+
+class PACT(paddle.nn.Layer):
+    def __init__(self):
+        super(PACT, self).__init__()
+        alpha_attr = paddle.ParamAttr(
+            name=self.full_name() + ".pact",
+            initializer=paddle.nn.initializer.Constant(value=20),
+            learning_rate=1.0,
+            regularizer=paddle.regularizer.L2Decay(2e-5),
+        )
+
+        self.alpha = self.create_parameter(shape=[1], attr=alpha_attr, dtype="float32")
+
+    def forward(self, x):
+        out_left = paddle.nn.functional.relu(x - self.alpha)
+        out_right = paddle.nn.functional.relu(-self.alpha - x)
+        x = x - out_left + out_right
+        return x
+
+
+quant_config = {
+    # weight preprocess type, default is None and no preprocessing is performed.
+    "weight_preprocess_type": None,
+    # activation preprocess type, default is None and no preprocessing is performed.
+    "activation_preprocess_type": None,
+    # weight quantize type, default is 'channel_wise_abs_max'
+    "weight_quantize_type": "channel_wise_abs_max",
+    # activation quantize type, default is 'moving_average_abs_max'
+    "activation_quantize_type": "moving_average_abs_max",
+    # weight quantize bit num, default is 8
+    "weight_bits": 8,
+    # activation quantize bit num, default is 8
+    "activation_bits": 8,
+    # data type after quantization, such as 'uint8', 'int8', etc. default is 'int8'
+    "dtype": "int8",
+    # window size for 'range_abs_max' quantization. default is 10000
+    "window_size": 10000,
+    # The decay coefficient of moving average, default is 0.9
+    "moving_rate": 0.9,
+    # for dygraph quantization, layers of type in quantizable_layer_type will be quantized
+    "quantizable_layer_type": ["Conv2D", "Linear"],
+}
diff --git a/test_tipc/supplementary/test_tipc/common_func.sh b/test_tipc/supplementary/test_tipc/common_func.sh
new file mode 100644
index 0000000..844536a
--- /dev/null
+++ b/test_tipc/supplementary/test_tipc/common_func.sh
@@ -0,0 +1,65 @@
+#!/bin/bash
+
+function func_parser_key(){
+    strs=$1
+    IFS=":"
+    array=(${strs})
+    tmp=${array[0]}
+    echo ${tmp}
+}
+
+function func_parser_value(){
+    strs=$1
+    IFS=":"
+    array=(${strs})
+    tmp=${array[1]}
+    echo ${tmp}
+}
+
+function func_set_params(){
+    key=$1
+    value=$2
+    if [ ${key}x = "null"x ];then
+        echo " "
+    elif [[ ${value} = "null" ]] || [[ ${value} = " " ]] || [ ${#value} -le 0 ];then
+        echo " "
+    else 
+        echo "${key}=${value}"
+    fi
+}
+
+function func_parser_params(){
+    strs=$1
+    MODE=$2
+    IFS=":"
+    array=(${strs})
+    key=${array[0]}
+    tmp=${array[1]}
+    IFS="|"
+    res=""
+    for _params in ${tmp[*]}; do
+        IFS="="
+        array=(${_params})
+        mode=${array[0]}
+        value=${array[1]}
+        if [[ ${mode} = ${MODE} ]]; then
+            IFS="|"
+            #echo $(func_set_params "${mode}" "${value}")
+            echo $value
+            break
+        fi
+        IFS="|"
+    done
+    echo ${res}
+}
+
+function status_check(){
+    last_status=$1   # the exit code
+    run_command=$2
+    run_log=$3
+    if [ $last_status -eq 0 ]; then
+        echo -e "\033[33m Run successfully with command - ${run_command}!  \033[0m" | tee -a ${run_log}
+    else
+        echo -e "\033[33m Run failed with command - ${run_command}!  \033[0m" | tee -a ${run_log}
+    fi
+}
diff --git a/test_tipc/supplementary/test_tipc/test_train_python.sh b/test_tipc/supplementary/test_tipc/test_train_python.sh
new file mode 100644
index 0000000..53e1cc9
--- /dev/null
+++ b/test_tipc/supplementary/test_tipc/test_train_python.sh
@@ -0,0 +1,114 @@
+#!/bin/bash
+source test_tipc/common_func.sh
+
+FILENAME=$1
+# MODE be one of ['lite_train_lite_infer' 'lite_train_whole_infer']
+MODE=$2
+
+dataline=$(awk 'NR==1, NR==51{print}'  $FILENAME)
+
+# parser params
+IFS=$'\n'
+lines=(${dataline})
+
+model_name=$(func_parser_value "${lines[1]}")
+python=$(func_parser_value "${lines[2]}")
+gpu_list=$(func_parser_value "${lines[3]}")
+train_use_gpu_key=$(func_parser_key "${lines[4]}")
+train_use_gpu_value=$(func_parser_value "${lines[4]}")
+autocast_list=$(func_parser_value "${lines[5]}")
+autocast_key=$(func_parser_key "${lines[5]}")
+epoch_key=$(func_parser_key "${lines[6]}")
+epoch_num=$(func_parser_params "${lines[6]}" "${MODE}")
+save_model_key=$(func_parser_key "${lines[7]}")
+train_batch_key=$(func_parser_key "${lines[8]}")
+train_batch_value=$(func_parser_params "${lines[8]}" "${MODE}")
+pretrain_model_key=$(func_parser_key "${lines[9]}")
+pretrain_model_value=$(func_parser_value "${lines[9]}")
+checkpoints_key=$(func_parser_key "${lines[10]}")
+checkpoints_value=$(func_parser_value "${lines[10]}")
+use_custom_key=$(func_parser_key "${lines[11]}")
+use_custom_list=$(func_parser_value "${lines[11]}")
+model_type_key=$(func_parser_key "${lines[12]}")
+model_type_list=$(func_parser_value "${lines[12]}")
+use_share_conv_key=$(func_parser_key "${lines[13]}")
+use_share_conv_list=$(func_parser_value "${lines[13]}")
+run_train_py=$(func_parser_value "${lines[14]}")
+
+LOG_PATH="./test_tipc/extra_output"
+mkdir -p ${LOG_PATH}
+status_log="${LOG_PATH}/results_python.log"
+
+if [ ${MODE} = "lite_train_lite_infer" ] || [ ${MODE} = "whole_train_whole_infer" ]; then
+    IFS="|"
+    export Count=0
+    USE_GPU_KEY=(${train_use_gpu_value})
+    # select cpu\gpu\distribute training
+    for gpu in ${gpu_list[*]}; do
+        train_use_gpu=${USE_GPU_KEY[Count]}
+        Count=$(($Count + 1))
+        ips=""
+        if [ ${gpu} = "-1" ];then
+            env=""
+        elif [ ${#gpu} -le 1 ];then
+            env="export CUDA_VISIBLE_DEVICES=${gpu}"
+            eval ${env}
+        elif [ ${#gpu} -le 15 ];then
+            IFS=","
+            array=(${gpu})
+            env="export CUDA_VISIBLE_DEVICES=${array[0]}"
+            IFS="|"
+        else
+            IFS=";"
+            array=(${gpu})
+            ips=${array[0]}
+            gpu=${array[1]}
+            IFS="|"
+            env=" "
+        fi
+        for autocast in ${autocast_list[*]}; do 
+            # set amp
+            if [ ${autocast} = "amp" ]; then
+                set_amp_config="AMP.use_amp=True"
+            else
+                set_amp_config=" "
+            fi
+
+            if [ ${run_train_py} = "null" ]; then
+                continue
+            fi
+
+            set_autocast=$(func_set_params "${autocast_key}" "${autocast}")
+            set_epoch=$(func_set_params "${epoch_key}" "${epoch_num}")
+            set_pretrain=$(func_set_params "${pretrain_model_key}" "${pretrain_model_value}")
+            set_checkpoints=$(func_set_params "${checkpoints_key}" "${checkpoints_value}")
+            set_batchsize=$(func_set_params "${train_batch_key}" "${train_batch_value}")
+            set_use_gpu=$(func_set_params "${train_use_gpu_key}" "${train_use_gpu}")
+
+            for custom_op in ${use_custom_list[*]}; do 
+                for model_type in ${model_type_list[*]}; do
+                    for share_conv in ${use_share_conv_list[*]}; do
+                        set_use_custom_op=$(func_set_params "${use_custom_key}" "${custom_op}")
+                        set_model_type=$(func_set_params "${model_type_key}" "${model_type}")
+                        set_use_share_conv=$(func_set_params "${use_share_conv_key}" "${share_conv}")
+
+                        set_save_model=$(func_set_params "${save_model_key}" "${save_log}")
+                        if [ ${#gpu} -le 2 ];then  # train with cpu or single gpu
+                            cmd="${python} ${run_train_py} ${set_use_gpu}  ${set_save_model}  ${set_epoch}  ${set_pretrain} ${set_checkpoints}  ${set_autocast} ${set_batchsize}  ${set_use_custom_op} ${set_model_type} ${set_use_share_conv} ${set_amp_config}"
+                        elif [ ${#ips} -le 26 ];then  # train with multi-gpu
+                            cmd="${python} -m paddle.distributed.launch --gpus=${gpu} ${run_train_py} ${set_use_gpu}  ${set_save_model}  ${set_epoch}  ${set_pretrain} ${set_checkpoints}  ${set_autocast} ${set_batchsize}  ${set_use_custom_op} ${set_model_type} ${set_use_share_conv} ${set_amp_config}"
+                        else
+                            cmd="${python} -m paddle.distributed.launch --ips=${ips} --gpus=${gpu} ${run_train_py} ${set_use_gpu}  ${set_save_model}  ${set_epoch}  ${set_pretrain} ${set_checkpoints}  ${set_autocast} ${set_batchsize}  ${set_use_custom_op} ${set_model_type} ${set_use_share_conv} ${set_amp_config}"
+                        fi
+
+                        # run train
+                        eval "unset CUDA_VISIBLE_DEVICES"
+                        # echo $cmd
+                        eval $cmd
+                        status_check $? "${cmd}" "${status_log}"
+                    done
+                done
+            done
+        done
+    done
+fi
diff --git a/test_tipc/supplementary/test_tipc/tipc_train.png b/test_tipc/supplementary/test_tipc/tipc_train.png
new file mode 100644
index 0000000..9ca124e
Binary files /dev/null and b/test_tipc/supplementary/test_tipc/tipc_train.png differ
diff --git a/test_tipc/supplementary/test_tipc/train_infer_python.txt b/test_tipc/supplementary/test_tipc/train_infer_python.txt
new file mode 100644
index 0000000..99028c0
--- /dev/null
+++ b/test_tipc/supplementary/test_tipc/train_infer_python.txt
@@ -0,0 +1,17 @@
+===========================train_params===========================
+model_name:ch_PPOCRv2_det
+python:python3.7
+gpu_list:0|0,1
+use_gpu:True|True
+AMP.use_amp:True|False
+epoch:lite_train_lite_infer=2|whole_train_whole_infer=1000
+save_model_dir:./output/
+TRAIN.batch_size:lite_train_lite_infer=1280|whole_train_whole_infer=1280
+pretrained_model:null
+checkpoints:null
+use_custom_relu:False|True
+model_type:cls|cls_distill|cls_distill_multiopt
+MODEL.siamese:False|True
+norm_train:train.py -c mv3_large_x0_5.yml -o 
+quant_train:False
+prune_train:False
diff --git a/test_tipc/supplementary/test_tipc/train_infer_python_FPGM.txt b/test_tipc/supplementary/test_tipc/train_infer_python_FPGM.txt
new file mode 100644
index 0000000..ccbd27f
--- /dev/null
+++ b/test_tipc/supplementary/test_tipc/train_infer_python_FPGM.txt
@@ -0,0 +1,17 @@
+===========================train_params===========================
+model_name:ch_PPOCRv2_det
+python:python3.7
+gpu_list:0|0,1
+use_gpu:True|True
+AMP.use_amp:True|False
+epoch:lite_train_lite_infer=2|whole_train_whole_infer=1000
+save_model_dir:./output/
+TRAIN.batch_size:lite_train_lite_infer=1280|whole_train_whole_infer=1280
+pretrained_model:null
+checkpoints:null
+use_custom_relu:False|True
+model_type:cls|cls_distill|cls_distill_multiopt
+MODEL.siamese:False|True
+norm_train:train.py -c mv3_large_x0_5.yml -o prune_train=True
+quant_train:False
+prune_train:False
diff --git a/test_tipc/supplementary/test_tipc/train_infer_python_FPGM_fleet.txt b/test_tipc/supplementary/test_tipc/train_infer_python_FPGM_fleet.txt
new file mode 100644
index 0000000..be2b211
--- /dev/null
+++ b/test_tipc/supplementary/test_tipc/train_infer_python_FPGM_fleet.txt
@@ -0,0 +1,17 @@
+===========================train_params===========================
+model_name:ch_PPOCRv2_det
+python:python3.7
+gpu_list:xx.xx.xx.xx,yy.yy.yy.yy;0,1
+use_gpu:True
+AMP.use_amp:True|False
+epoch:lite_train_lite_infer=2|whole_train_whole_infer=1000
+save_model_dir:./output/
+TRAIN.batch_size:lite_train_lite_infer=1280|whole_train_whole_infer=1280
+pretrained_model:null
+checkpoints:null
+use_custom_relu:False|True
+model_type:cls|cls_distill|cls_distill_multiopt
+MODEL.siamese:False|True
+norm_train:train.py -c mv3_large_x0_5.yml -o prune_train=True
+quant_train:False
+prune_train:False
diff --git a/test_tipc/supplementary/test_tipc/train_infer_python_PACT.txt b/test_tipc/supplementary/test_tipc/train_infer_python_PACT.txt
new file mode 100644
index 0000000..24d291b
--- /dev/null
+++ b/test_tipc/supplementary/test_tipc/train_infer_python_PACT.txt
@@ -0,0 +1,17 @@
+===========================train_params===========================
+model_name:ch_PPOCRv2_det
+python:python3.7
+gpu_list:0|0,1
+use_gpu:True|True
+AMP.use_amp:True|False
+epoch:lite_train_lite_infer=2|whole_train_whole_infer=1000
+save_model_dir:./output/
+TRAIN.batch_size:lite_train_lite_infer=1280|whole_train_whole_infer=1280
+pretrained_model:null
+checkpoints:null
+use_custom_relu:False|True
+model_type:cls|cls_distill|cls_distill_multiopt
+MODEL.siamese:False|True
+norm_train:train.py -c mv3_large_x0_5.yml -o quant_train=True
+quant_train:False
+prune_train:False
diff --git a/test_tipc/supplementary/test_tipc/train_infer_python_PACT_fleet.txt b/test_tipc/supplementary/test_tipc/train_infer_python_PACT_fleet.txt
new file mode 100644
index 0000000..93f06d7
--- /dev/null
+++ b/test_tipc/supplementary/test_tipc/train_infer_python_PACT_fleet.txt
@@ -0,0 +1,17 @@
+===========================train_params===========================
+model_name:ch_PPOCRv2_det
+python:python3.7
+gpu_list:xx.xx.xx.xx,yy.yy.yy.yy;0,1
+use_gpu:True
+AMP.use_amp:True|False
+epoch:lite_train_lite_infer=2|whole_train_whole_infer=1000
+save_model_dir:./output/
+TRAIN.batch_size:lite_train_lite_infer=1280|whole_train_whole_infer=1280
+pretrained_model:null
+checkpoints:null
+use_custom_relu:False|True
+model_type:cls|cls_distill|cls_distill_multiopt
+MODEL.siamese:False|True
+norm_train:train.py -c mv3_large_x0_5.yml -o quant_train=True
+quant_train:False
+prune_train:False
diff --git a/test_tipc/supplementary/test_tipc/train_infer_python_fleet.txt b/test_tipc/supplementary/test_tipc/train_infer_python_fleet.txt
new file mode 100644
index 0000000..00b9e82
--- /dev/null
+++ b/test_tipc/supplementary/test_tipc/train_infer_python_fleet.txt
@@ -0,0 +1,17 @@
+===========================train_params===========================
+model_name:ch_PPOCRv2_det
+python:python3.7
+gpu_list:xx.xx.xx.xx,yy.yy.yy.yy;0,1
+use_gpu:True
+AMP.use_amp:True|False
+epoch:lite_train_lite_infer=2|whole_train_whole_infer=1000
+save_model_dir:./output/
+TRAIN.batch_size:lite_train_lite_infer=1280|whole_train_whole_infer=1280
+pretrained_model:null
+checkpoints:null
+use_custom_relu:False|True
+model_type:cls|cls_distill|cls_distill_multiopt
+MODEL.siamese:False|True
+norm_train: train.py -c mv3_large_x0_5.yml -o
+quant_train:False
+prune_train:False
diff --git a/test_tipc/supplementary/train.py b/test_tipc/supplementary/train.py
new file mode 100644
index 0000000..ece8d5b
--- /dev/null
+++ b/test_tipc/supplementary/train.py
@@ -0,0 +1,494 @@
+import paddle
+import numpy as np
+import os
+import paddle.nn as nn
+import paddle.distributed as dist
+
+dist.get_world_size()
+dist.init_parallel_env()
+
+from loss import build_loss, LossDistill, DMLLoss, KLJSLoss
+from optimizer import create_optimizer
+from data_loader import build_dataloader
+from metric import create_metric
+from mv3 import MobileNetV3_large_x0_5, distillmv3_large_x0_5, build_model
+from config import preprocess
+import time
+
+from paddleslim.dygraph.quant import QAT
+from slim.slim_quant import PACT, quant_config
+from slim.slim_fpgm import prune_model
+from utils import load_model
+
+
+def _mkdir_if_not_exist(path, logger):
+    """
+    mkdir if not exists, ignore the exception when multiprocess mkdir together
+    """
+    if not os.path.exists(path):
+        try:
+            os.makedirs(path)
+        except OSError as e:
+            if e.errno == errno.EEXIST and os.path.isdir(path):
+                logger.warning(
+                    "be happy if some process has already created {}".format(path)
+                )
+            else:
+                raise OSError("Failed to mkdir {}".format(path))
+
+
+def save_model(
+    model, optimizer, model_path, logger, is_best=False, prefix="ppocr", **kwargs
+):
+    """
+    save model to the target path
+    """
+    _mkdir_if_not_exist(model_path, logger)
+    model_prefix = os.path.join(model_path, prefix)
+    paddle.save(model.state_dict(), model_prefix + ".pdparams")
+    if type(optimizer) is list:
+        paddle.save(optimizer[0].state_dict(), model_prefix + ".pdopt")
+        paddle.save(optimizer[1].state_dict(), model_prefix + "_1" + ".pdopt")
+
+    else:
+        paddle.save(optimizer.state_dict(), model_prefix + ".pdopt")
+
+    # # save metric and config
+    # with open(model_prefix + '.states', 'wb') as f:
+    #     pickle.dump(kwargs, f, protocol=2)
+    if is_best:
+        logger.info("save best model is to {}".format(model_prefix))
+    else:
+        logger.info("save model in {}".format(model_prefix))
+
+
+def amp_scaler(config):
+    if "AMP" in config and config["AMP"]["use_amp"] is True:
+        AMP_RELATED_FLAGS_SETTING = {
+            "FLAGS_cudnn_batchnorm_spatial_persistent": 1,
+        }
+        paddle.set_flags(AMP_RELATED_FLAGS_SETTING)
+        scale_loss = config["AMP"].get("scale_loss", 1.0)
+        use_dynamic_loss_scaling = config["AMP"].get("use_dynamic_loss_scaling", False)
+        scaler = paddle.amp.GradScaler(
+            init_loss_scaling=scale_loss,
+            use_dynamic_loss_scaling=use_dynamic_loss_scaling,
+        )
+        return scaler
+    else:
+        return None
+
+
+def set_seed(seed):
+    paddle.seed(seed)
+    np.random.seed(seed)
+
+
+def train(config, scaler=None):
+    EPOCH = config["epoch"]
+    topk = config["topk"]
+
+    batch_size = config["TRAIN"]["batch_size"]
+    num_workers = config["TRAIN"]["num_workers"]
+    train_loader = build_dataloader(
+        "train", batch_size=batch_size, num_workers=num_workers
+    )
+
+    # build metric
+    metric_func = create_metric
+
+    # build model
+    # model = MobileNetV3_large_x0_5(class_dim=100)
+    model = build_model(config)
+
+    # build_optimizer
+    optimizer, lr_scheduler = create_optimizer(
+        config, parameter_list=model.parameters()
+    )
+
+    # load model
+    pre_best_model_dict = load_model(config, model, optimizer)
+    if len(pre_best_model_dict) > 0:
+        pre_str = "The metric of loaded metric as follows {}".format(
+            ", ".join(["{}: {}".format(k, v) for k, v in pre_best_model_dict.items()])
+        )
+        logger.info(pre_str)
+
+    # about slim prune and quant
+    if "quant_train" in config and config["quant_train"] is True:
+        quanter = QAT(config=quant_config, act_preprocess=PACT)
+        quanter.quantize(model)
+    elif "prune_train" in config and config["prune_train"] is True:
+        model = prune_model(model, [1, 3, 32, 32], 0.1)
+    else:
+        pass
+
+    # distribution
+    model.train()
+    model = paddle.DataParallel(model)
+    # build loss function
+    loss_func = build_loss(config)
+
+    data_num = len(train_loader)
+
+    best_acc = {}
+    for epoch in range(EPOCH):
+        st = time.time()
+        for idx, data in enumerate(train_loader):
+            img_batch, label = data
+            img_batch = paddle.transpose(img_batch, [0, 3, 1, 2])
+            label = paddle.unsqueeze(label, -1)
+
+            if scaler is not None:
+                with paddle.amp.auto_cast():
+                    outs = model(img_batch)
+            else:
+                outs = model(img_batch)
+
+            # cal metric
+            acc = metric_func(outs, label)
+
+            # cal loss
+            avg_loss = loss_func(outs, label)
+
+            if scaler is None:
+                # backward
+                avg_loss.backward()
+                optimizer.step()
+                optimizer.clear_grad()
+            else:
+                scaled_avg_loss = scaler.scale(avg_loss)
+                scaled_avg_loss.backward()
+                scaler.minimize(optimizer, scaled_avg_loss)
+
+            if not isinstance(lr_scheduler, float):
+                lr_scheduler.step()
+
+            if idx % 10 == 0:
+                et = time.time()
+                strs = f"epoch: [{epoch}/{EPOCH}], iter: [{idx}/{data_num}], "
+                strs += f"loss: {float(avg_loss)}"
+                strs += (
+                    f", acc_topk1: {float(acc['top1'])}, acc_top5: {float(acc['top5'])}"
+                )
+                strs += f", batch_time: {round(et-st, 4)} s"
+                logger.info(strs)
+                st = time.time()
+
+        if epoch % 10 == 0:
+            acc = eval(config, model)
+            if len(best_acc) < 1 or float(acc["top5"]) > best_acc["top5"]:
+                best_acc = acc
+                best_acc["epoch"] = epoch
+                is_best = True
+            else:
+                is_best = False
+            logger.info(
+                f"The best acc: acc_topk1: {float(best_acc['top1'])}, acc_top5: {float(best_acc['top5'])}, best_epoch: {best_acc['epoch']}"
+            )
+            save_model(
+                model,
+                optimizer,
+                config["save_model_dir"],
+                logger,
+                is_best,
+                prefix="cls",
+            )
+
+
+def train_distill(config, scaler=None):
+    EPOCH = config["epoch"]
+    topk = config["topk"]
+
+    batch_size = config["TRAIN"]["batch_size"]
+    num_workers = config["TRAIN"]["num_workers"]
+    train_loader = build_dataloader(
+        "train", batch_size=batch_size, num_workers=num_workers
+    )
+
+    # build metric
+    metric_func = create_metric
+
+    # model = distillmv3_large_x0_5(class_dim=100)
+    model = build_model(config)
+
+    # pact quant train
+    if "quant_train" in config and config["quant_train"] is True:
+        quanter = QAT(config=quant_config, act_preprocess=PACT)
+        quanter.quantize(model)
+    elif "prune_train" in config and config["prune_train"] is True:
+        model = prune_model(model, [1, 3, 32, 32], 0.1)
+    else:
+        pass
+
+    # build_optimizer
+    optimizer, lr_scheduler = create_optimizer(
+        config, parameter_list=model.parameters()
+    )
+
+    # load model
+    pre_best_model_dict = load_model(config, model, optimizer)
+    if len(pre_best_model_dict) > 0:
+        pre_str = "The metric of loaded metric as follows {}".format(
+            ", ".join(["{}: {}".format(k, v) for k, v in pre_best_model_dict.items()])
+        )
+        logger.info(pre_str)
+
+    model.train()
+    model = paddle.DataParallel(model)
+
+    # build loss function
+    loss_func_distill = LossDistill(model_name_list=["student", "student1"])
+    loss_func_dml = DMLLoss(model_name_pairs=["student", "student1"])
+    loss_func_js = KLJSLoss(mode="js")
+
+    data_num = len(train_loader)
+
+    best_acc = {}
+    for epoch in range(EPOCH):
+        st = time.time()
+        for idx, data in enumerate(train_loader):
+            img_batch, label = data
+            img_batch = paddle.transpose(img_batch, [0, 3, 1, 2])
+            label = paddle.unsqueeze(label, -1)
+            if scaler is not None:
+                with paddle.amp.auto_cast():
+                    outs = model(img_batch)
+            else:
+                outs = model(img_batch)
+
+            # cal metric
+            acc = metric_func(outs["student"], label)
+
+            # cal loss
+            avg_loss = (
+                loss_func_distill(outs, label)["student"]
+                + loss_func_distill(outs, label)["student1"]
+                + loss_func_dml(outs, label)["student_student1"]
+            )
+
+            # backward
+            if scaler is None:
+                avg_loss.backward()
+                optimizer.step()
+                optimizer.clear_grad()
+            else:
+                scaled_avg_loss = scaler.scale(avg_loss)
+                scaled_avg_loss.backward()
+                scaler.minimize(optimizer, scaled_avg_loss)
+
+            if not isinstance(lr_scheduler, float):
+                lr_scheduler.step()
+
+            if idx % 10 == 0:
+                et = time.time()
+                strs = f"epoch: [{epoch}/{EPOCH}], iter: [{idx}/{data_num}], "
+                strs += f"loss: {float(avg_loss)}"
+                strs += (
+                    f", acc_topk1: {float(acc['top1'])}, acc_top5: {float(acc['top5'])}"
+                )
+                strs += f", batch_time: {round(et-st, 4)} s"
+                logger.info(strs)
+                st = time.time()
+
+        if epoch % 10 == 0:
+            acc = eval(config, model._layers.student)
+            if len(best_acc) < 1 or float(acc["top5"]) > best_acc["top5"]:
+                best_acc = acc
+                best_acc["epoch"] = epoch
+                is_best = True
+            else:
+                is_best = False
+            logger.info(
+                f"The best acc: acc_topk1: {float(best_acc['top1'])}, acc_top5: {float(best_acc['top5'])}, best_epoch: {best_acc['epoch']}"
+            )
+
+            save_model(
+                model,
+                optimizer,
+                config["save_model_dir"],
+                logger,
+                is_best,
+                prefix="cls_distill",
+            )
+
+
+def train_distill_multiopt(config, scaler=None):
+    EPOCH = config["epoch"]
+    topk = config["topk"]
+
+    batch_size = config["TRAIN"]["batch_size"]
+    num_workers = config["TRAIN"]["num_workers"]
+    train_loader = build_dataloader(
+        "train", batch_size=batch_size, num_workers=num_workers
+    )
+
+    # build metric
+    metric_func = create_metric
+
+    # model = distillmv3_large_x0_5(class_dim=100)
+    model = build_model(config)
+
+    # build_optimizer
+    optimizer, lr_scheduler = create_optimizer(
+        config, parameter_list=model.student.parameters()
+    )
+    optimizer1, lr_scheduler1 = create_optimizer(
+        config, parameter_list=model.student1.parameters()
+    )
+
+    # load model
+    pre_best_model_dict = load_model(config, model, optimizer)
+    if len(pre_best_model_dict) > 0:
+        pre_str = "The metric of loaded metric as follows {}".format(
+            ", ".join(["{}: {}".format(k, v) for k, v in pre_best_model_dict.items()])
+        )
+        logger.info(pre_str)
+
+    # quant train
+    if "quant_train" in config and config["quant_train"] is True:
+        quanter = QAT(config=quant_config, act_preprocess=PACT)
+        quanter.quantize(model)
+    elif "prune_train" in config and config["prune_train"] is True:
+        model = prune_model(model, [1, 3, 32, 32], 0.1)
+    else:
+        pass
+
+    model.train()
+
+    model = paddle.DataParallel(model)
+
+    # build loss function
+    loss_func_distill = LossDistill(model_name_list=["student", "student1"])
+    loss_func_dml = DMLLoss(model_name_pairs=["student", "student1"])
+    loss_func_js = KLJSLoss(mode="js")
+
+    data_num = len(train_loader)
+    best_acc = {}
+    for epoch in range(EPOCH):
+        st = time.time()
+        for idx, data in enumerate(train_loader):
+            img_batch, label = data
+            img_batch = paddle.transpose(img_batch, [0, 3, 1, 2])
+            label = paddle.unsqueeze(label, -1)
+
+            if scaler is not None:
+                with paddle.amp.auto_cast():
+                    outs = model(img_batch)
+            else:
+                outs = model(img_batch)
+
+            # cal metric
+            acc = metric_func(outs["student"], label)
+
+            # cal loss
+            avg_loss = (
+                loss_func_distill(outs, label)["student"]
+                + loss_func_dml(outs, label)["student_student1"]
+            )
+            avg_loss1 = (
+                loss_func_distill(outs, label)["student1"]
+                + loss_func_dml(outs, label)["student_student1"]
+            )
+
+            if scaler is None:
+                # backward
+                avg_loss.backward(retain_graph=True)
+                optimizer.step()
+                optimizer.clear_grad()
+
+                avg_loss1.backward()
+                optimizer1.step()
+                optimizer1.clear_grad()
+            else:
+                scaled_avg_loss = scaler.scale(avg_loss)
+                scaled_avg_loss.backward()
+                scaler.minimize(optimizer, scaled_avg_loss)
+
+                scaled_avg_loss = scaler.scale(avg_loss1)
+                scaled_avg_loss.backward()
+                scaler.minimize(optimizer1, scaled_avg_loss)
+
+            if not isinstance(lr_scheduler, float):
+                lr_scheduler.step()
+            if not isinstance(lr_scheduler1, float):
+                lr_scheduler1.step()
+
+            if idx % 10 == 0:
+                et = time.time()
+                strs = f"epoch: [{epoch}/{EPOCH}], iter: [{idx}/{data_num}], "
+                strs += f"loss: {float(avg_loss)}, loss1: {float(avg_loss1)}"
+                strs += (
+                    f", acc_topk1: {float(acc['top1'])}, acc_top5: {float(acc['top5'])}"
+                )
+                strs += f", batch_time: {round(et-st, 4)} s"
+                logger.info(strs)
+                st = time.time()
+
+        if epoch % 10 == 0:
+            acc = eval(config, model._layers.student)
+            if len(best_acc) < 1 or float(acc["top5"]) > best_acc["top5"]:
+                best_acc = acc
+                best_acc["epoch"] = epoch
+                is_best = True
+            else:
+                is_best = False
+            logger.info(
+                f"The best acc: acc_topk1: {float(best_acc['top1'])}, acc_top5: {float(best_acc['top5'])}, best_epoch: {best_acc['epoch']}"
+            )
+            save_model(
+                model,
+                [optimizer, optimizer1],
+                config["save_model_dir"],
+                logger,
+                is_best,
+                prefix="cls_distill_multiopt",
+            )
+
+
+def eval(config, model):
+    batch_size = config["VALID"]["batch_size"]
+    num_workers = config["VALID"]["num_workers"]
+    valid_loader = build_dataloader(
+        "test", batch_size=batch_size, num_workers=num_workers
+    )
+
+    # build metric
+    metric_func = create_metric
+
+    outs = []
+    labels = []
+    for idx, data in enumerate(valid_loader):
+        img_batch, label = data
+        img_batch = paddle.transpose(img_batch, [0, 3, 1, 2])
+        label = paddle.unsqueeze(label, -1)
+        out = model(img_batch)
+
+        outs.append(out)
+        labels.append(label)
+
+    outs = paddle.concat(outs, axis=0)
+    labels = paddle.concat(labels, axis=0)
+    acc = metric_func(outs, labels)
+
+    strs = f"The metric are as follows: acc_topk1: {float(acc['top1'])}, acc_top5: {float(acc['top5'])}"
+    logger.info(strs)
+    return acc
+
+
+if __name__ == "__main__":
+    config, logger = preprocess(is_train=False)
+
+    # AMP scaler
+    scaler = amp_scaler(config)
+
+    model_type = config["model_type"]
+
+    if model_type == "cls":
+        train(config)
+    elif model_type == "cls_distill":
+        train_distill(config)
+    elif model_type == "cls_distill_multiopt":
+        train_distill_multiopt(config)
+    else:
+        raise ValueError("model_type should be one of ['']")
diff --git a/test_tipc/supplementary/train.sh b/test_tipc/supplementary/train.sh
new file mode 100644
index 0000000..a2c7c90
--- /dev/null
+++ b/test_tipc/supplementary/train.sh
@@ -0,0 +1,5 @@
+# single GPU
+python3.7 train.py  -c mv3_large_x0_5.yml
+
+# distribute training
+python3.7 -m paddle.distributed.launch --log_dir=./debug/ --gpus '0,1'  train.py  -c mv3_large_x0_5.yml
diff --git a/test_tipc/supplementary/utils.py b/test_tipc/supplementary/utils.py
new file mode 100644
index 0000000..ffd61fa
--- /dev/null
+++ b/test_tipc/supplementary/utils.py
@@ -0,0 +1,172 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import sys
+import logging
+import functools
+import paddle.distributed as dist
+
+logger_initialized = {}
+
+
+def print_dict(d, logger, delimiter=0):
+    """
+    Recursively visualize a dict and
+    indenting acrrording by the relationship of keys.
+    """
+    for k, v in sorted(d.items()):
+        if isinstance(v, dict):
+            logger.info("{}{} : ".format(delimiter * " ", str(k)))
+            print_dict(v, logger, delimiter + 4)
+        elif isinstance(v, list) and len(v) >= 1 and isinstance(v[0], dict):
+            logger.info("{}{} : ".format(delimiter * " ", str(k)))
+            for value in v:
+                print_dict(value, logger, delimiter + 4)
+        else:
+            logger.info("{}{} : {}".format(delimiter * " ", k, v))
+
+
+@functools.lru_cache()
+def get_logger(name="root", log_file=None, log_level=logging.DEBUG):
+    """Initialize and get a logger by name.
+    If the logger has not been initialized, this method will initialize the
+    logger by adding one or two handlers, otherwise the initialized logger will
+    be directly returned. During initialization, a StreamHandler will always be
+    added. If `log_file` is specified a FileHandler will also be added.
+    Args:
+        name (str): Logger name.
+        log_file (str | None): The log filename. If specified, a FileHandler
+            will be added to the logger.
+        log_level (int): The logger level. Note that only the process of
+            rank 0 is affected, and other processes will set the level to
+            "Error" thus be silent most of the time.
+    Returns:
+        logging.Logger: The expected logger.
+    """
+    logger = logging.getLogger(name)
+    if name in logger_initialized:
+        return logger
+    for logger_name in logger_initialized:
+        if name.startswith(logger_name):
+            return logger
+
+    formatter = logging.Formatter(
+        "[%(asctime)s] %(name)s %(levelname)s: %(message)s", datefmt="%Y/%m/%d %H:%M:%S"
+    )
+
+    stream_handler = logging.StreamHandler(stream=sys.stdout)
+    stream_handler.setFormatter(formatter)
+    logger.addHandler(stream_handler)
+    if log_file is not None and dist.get_rank() == 0:
+        log_file_folder = os.path.split(log_file)[0]
+        os.makedirs(log_file_folder, exist_ok=True)
+        file_handler = logging.FileHandler(log_file, "a")
+        file_handler.setFormatter(formatter)
+        logger.addHandler(file_handler)
+    if dist.get_rank() == 0:
+        logger.setLevel(log_level)
+    else:
+        logger.setLevel(logging.ERROR)
+    logger_initialized[name] = True
+    return logger
+
+
+def load_model(config, model, optimizer=None):
+    """
+    load model from checkpoint or pretrained_model
+    """
+    logger = get_logger()
+    checkpoints = config.get("checkpoints")
+    pretrained_model = config.get("pretrained_model")
+    best_model_dict = {}
+    if checkpoints:
+        if checkpoints.endswith(".pdparams"):
+            checkpoints = checkpoints.replace(".pdparams", "")
+        assert os.path.exists(
+            checkpoints + ".pdparams"
+        ), "The {}.pdparams does not exists!".format(checkpoints)
+
+        # load params from trained model
+        params = paddle.load(checkpoints + ".pdparams")
+        state_dict = model.state_dict()
+        new_state_dict = {}
+        for key, value in state_dict.items():
+            if key not in params:
+                logger.warning(
+                    "{} not in loaded params {} !".format(key, params.keys())
+                )
+                continue
+            pre_value = params[key]
+            if list(value.shape) == list(pre_value.shape):
+                new_state_dict[key] = pre_value
+            else:
+                logger.warning(
+                    "The shape of model params {} {} not matched with loaded params shape {} !".format(
+                        key, value.shape, pre_value.shape
+                    )
+                )
+        model.set_state_dict(new_state_dict)
+
+        if optimizer is not None:
+            if os.path.exists(checkpoints + ".pdopt"):
+                optim_dict = paddle.load(checkpoints + ".pdopt")
+                optimizer.set_state_dict(optim_dict)
+            else:
+                logger.warning(
+                    "{}.pdopt is not exists, params of optimizer is not loaded".format(
+                        checkpoints
+                    )
+                )
+
+        if os.path.exists(checkpoints + ".states"):
+            with open(checkpoints + ".states", "rb") as f:
+                states_dict = pickle.load(f, encoding="latin1")
+            best_model_dict = states_dict.get("best_model_dict", {})
+            if "epoch" in states_dict:
+                best_model_dict["start_epoch"] = states_dict["epoch"] + 1
+        logger.info("resume from {}".format(checkpoints))
+    elif pretrained_model:
+        load_pretrained_params(model, pretrained_model)
+    else:
+        logger.info("train from scratch")
+    return best_model_dict
+
+
+def load_pretrained_params(model, path):
+    logger = get_logger()
+    if path.endswith(".pdparams"):
+        path = path.replace(".pdparams", "")
+    assert os.path.exists(
+        path + ".pdparams"
+    ), "The {}.pdparams does not exists!".format(path)
+
+    params = paddle.load(path + ".pdparams")
+    state_dict = model.state_dict()
+    new_state_dict = {}
+    for k1 in params.keys():
+        if k1 not in state_dict.keys():
+            logger.warning("The pretrained params {} not in model".format(k1))
+        else:
+            if list(state_dict[k1].shape) == list(params[k1].shape):
+                new_state_dict[k1] = params[k1]
+            else:
+                logger.warning(
+                    "The shape of model params {} {} not matched with loaded params {} {} !".format(
+                        k1, state_dict[k1].shape, k1, params[k1].shape
+                    )
+                )
+    model.set_state_dict(new_state_dict)
+    logger.info("load pretrain successful from {}".format(path))
+    return model
diff --git a/test_tipc/test_inference_cpp.sh b/test_tipc/test_inference_cpp.sh
new file mode 100644
index 0000000..c84f225
--- /dev/null
+++ b/test_tipc/test_inference_cpp.sh
@@ -0,0 +1,240 @@
+#!/bin/bash
+source test_tipc/common_func.sh
+
+FILENAME=$1
+dataline=$(awk 'NR==1, NR==20{print}'  $FILENAME)
+
+# parser params
+IFS=$'\n'
+lines=(${dataline})
+
+# parser cpp inference model 
+model_name=$(func_parser_value "${lines[1]}")
+use_opencv=$(func_parser_value "${lines[2]}")
+cpp_infer_model_dir_list=$(func_parser_value "${lines[3]}")
+cpp_infer_is_quant=$(func_parser_value "${lines[4]}")
+# parser cpp inference 
+inference_cmd=$(func_parser_value "${lines[5]}")
+cpp_use_gpu_key=$(func_parser_key "${lines[6]}")
+cpp_use_gpu_list=$(func_parser_value "${lines[6]}")
+cpp_use_mkldnn_key=$(func_parser_key "${lines[7]}")
+cpp_use_mkldnn_list=$(func_parser_value "${lines[7]}")
+cpp_cpu_threads_key=$(func_parser_key "${lines[8]}")
+cpp_cpu_threads_list=$(func_parser_value "${lines[8]}")
+cpp_batch_size_key=$(func_parser_key "${lines[9]}")
+cpp_batch_size_list=$(func_parser_value "${lines[9]}")
+cpp_use_trt_key=$(func_parser_key "${lines[10]}")
+cpp_use_trt_list=$(func_parser_value "${lines[10]}")
+cpp_precision_key=$(func_parser_key "${lines[11]}")
+cpp_precision_list=$(func_parser_value "${lines[11]}")
+cpp_infer_model_key=$(func_parser_key "${lines[12]}")
+cpp_image_dir_key=$(func_parser_key "${lines[13]}")
+cpp_infer_img_dir=$(func_parser_value "${lines[13]}")
+cpp_infer_key1=$(func_parser_key "${lines[14]}")
+cpp_infer_value1=$(func_parser_value "${lines[14]}")
+cpp_benchmark_key=$(func_parser_key "${lines[15]}")
+cpp_benchmark_value=$(func_parser_value "${lines[15]}")
+cpp_det_key=$(func_parser_key "${lines[16]}")
+cpp_det_value=$(func_parser_value "${lines[16]}")
+cpp_rec_key=$(func_parser_key "${lines[17]}")
+cpp_rec_value=$(func_parser_value "${lines[17]}")
+cpp_cls_key=$(func_parser_key "${lines[18]}")
+cpp_cls_value=$(func_parser_value "${lines[18]}")
+cpp_use_angle_cls_key=$(func_parser_key "${lines[19]}")
+cpp_use_angle_cls_value=$(func_parser_value "${lines[19]}")
+
+LOG_PATH="./test_tipc/output/${model_name}/cpp_infer"
+mkdir -p ${LOG_PATH}
+status_log="${LOG_PATH}/results_cpp.log"
+
+
+function func_cpp_inference(){
+    IFS='|'
+    _script=$1
+    _model_dir=$2
+    _log_path=$3
+    _img_dir=$4
+    _flag_quant=$5
+    # inference 
+    for use_gpu in ${cpp_use_gpu_list[*]}; do
+        if [ ${use_gpu} = "False" ] || [ ${use_gpu} = "cpu" ]; then
+            for use_mkldnn in ${cpp_use_mkldnn_list[*]}; do
+                if [ ${use_mkldnn} = "False" ] && [ ${_flag_quant} = "True" ]; then
+                    continue
+                fi
+                for threads in ${cpp_cpu_threads_list[*]}; do
+                    for batch_size in ${cpp_batch_size_list[*]}; do
+                        precision="fp32"
+                        if [ ${use_mkldnn} = "False" ] && [ ${_flag_quant} = "True" ]; then
+                            precision="int8"
+                        fi
+                        _save_log_path="${_log_path}/cpp_infer_cpu_usemkldnn_${use_mkldnn}_threads_${threads}_precision_${precision}_batchsize_${batch_size}.log"
+                        set_infer_data=$(func_set_params "${cpp_image_dir_key}" "${_img_dir}")
+                        set_benchmark=$(func_set_params "${cpp_benchmark_key}" "${cpp_benchmark_value}")
+                        set_batchsize=$(func_set_params "${cpp_batch_size_key}" "${batch_size}")
+                        set_mkldnn=$(func_set_params "${cpp_use_mkldnn_key}" "${use_mkldnn}")
+                        set_cpu_threads=$(func_set_params "${cpp_cpu_threads_key}" "${threads}")
+                        set_model_dir=$(func_set_params "${cpp_infer_model_key}" "${_model_dir}")
+                        set_infer_params1=$(func_set_params "${cpp_infer_key1}" "${cpp_infer_value1}")
+                        set_det=$(func_set_params "${cpp_det_key}" "${cpp_det_value}")
+                        set_rec=$(func_set_params "${cpp_rec_key}" "${cpp_rec_value}")
+                        set_cls=$(func_set_params "${cpp_cls_key}" "${cpp_cls_value}")
+                        set_use_angle_cls=$(func_set_params "${cpp_use_angle_cls_key}" "${cpp_use_angle_cls_value}")
+                        command="${_script} ${cpp_use_gpu_key}=${use_gpu} ${set_mkldnn} ${set_cpu_threads} ${set_model_dir} ${set_batchsize} ${set_infer_data} ${set_benchmark} ${set_det} ${set_rec} ${set_cls} ${set_use_angle_cls} ${set_infer_params1} > ${_save_log_path} 2>&1 "
+                        eval $command
+                        last_status=${PIPESTATUS[0]}
+                        eval "cat ${_save_log_path}"
+                        status_check $last_status "${command}" "${status_log}" "${model_name}" "${_save_log_path}"
+                    done
+                done
+            done
+        elif [ ${use_gpu} = "True" ] || [ ${use_gpu} = "gpu" ]; then
+            for use_trt in ${cpp_use_trt_list[*]}; do
+                for precision in ${cpp_precision_list[*]}; do
+                    if [[ ${_flag_quant} = "False" ]] && [[ ${precision} =~ "int8" ]]; then
+                        continue
+                    fi 
+                    if [[ ${precision} =~ "fp16" || ${precision} =~ "int8" ]] && [ ${use_trt} = "False" ]; then
+                        continue
+                    fi
+                    if [[ ${use_trt} = "False" || ${precision} =~ "int8" ]] && [ ${_flag_quant} = "True" ]; then
+                        continue
+                    fi
+                    for batch_size in ${cpp_batch_size_list[*]}; do
+                        _save_log_path="${_log_path}/cpp_infer_gpu_usetrt_${use_trt}_precision_${precision}_batchsize_${batch_size}.log"
+                        set_infer_data=$(func_set_params "${cpp_image_dir_key}" "${_img_dir}")
+                        set_benchmark=$(func_set_params "${cpp_benchmark_key}" "${cpp_benchmark_value}")
+                        set_batchsize=$(func_set_params "${cpp_batch_size_key}" "${batch_size}")
+                        set_tensorrt=$(func_set_params "${cpp_use_trt_key}" "${use_trt}")
+                        set_precision=$(func_set_params "${cpp_precision_key}" "${precision}")
+                        set_model_dir=$(func_set_params "${cpp_infer_model_key}" "${_model_dir}")
+                        set_infer_params1=$(func_set_params "${cpp_infer_key1}" "${cpp_infer_value1}")
+                        set_det=$(func_set_params "${cpp_det_key}" "${cpp_det_value}")
+                        set_rec=$(func_set_params "${cpp_rec_key}" "${cpp_rec_value}")
+                        set_cls=$(func_set_params "${cpp_cls_key}" "${cpp_cls_value}")
+                        set_use_angle_cls=$(func_set_params "${cpp_use_angle_cls_key}" "${cpp_use_angle_cls_value}")
+                        command="${_script} ${cpp_use_gpu_key}=${use_gpu} ${set_tensorrt} ${set_precision} ${set_model_dir} ${set_batchsize} ${set_infer_data} ${set_benchmark} ${set_det} ${set_rec} ${set_cls} ${set_use_angle_cls} ${set_infer_params1} > ${_save_log_path} 2>&1 "
+                        eval $command
+                        last_status=${PIPESTATUS[0]}
+                        eval "cat ${_save_log_path}"
+                        status_check $last_status "${command}" "${status_log}" "${model_name}" "${_save_log_path}"
+                        
+                    done
+                done
+            done
+        else
+            echo "Does not support hardware other than CPU and GPU Currently!"
+        fi
+    done
+}
+
+
+cd deploy/cpp_infer
+if [ ${use_opencv} = "True" ]; then
+    if [ -d "opencv-3.4.7/opencv3/" ] && [ $(md5sum opencv-3.4.7.tar.gz | awk -F ' ' '{print $1}') = "faa2b5950f8bee3f03118e600c74746a" ];then
+        echo "################### build opencv skipped ###################"
+    else
+        echo "################### build opencv ###################"
+        rm -rf opencv-3.4.7.tar.gz opencv-3.4.7/
+        wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/opencv-3.4.7.tar.gz
+        tar -xf opencv-3.4.7.tar.gz
+
+        cd opencv-3.4.7/
+        install_path=$(pwd)/opencv3
+
+        rm -rf build
+        mkdir build
+        cd build
+
+        cmake .. \
+            -DCMAKE_INSTALL_PREFIX=${install_path} \
+            -DCMAKE_BUILD_TYPE=Release \
+            -DBUILD_SHARED_LIBS=OFF \
+            -DWITH_IPP=OFF \
+            -DBUILD_IPP_IW=OFF \
+            -DWITH_LAPACK=OFF \
+            -DWITH_EIGEN=OFF \
+            -DCMAKE_INSTALL_LIBDIR=lib64 \
+            -DWITH_ZLIB=ON \
+            -DBUILD_ZLIB=ON \
+            -DWITH_JPEG=ON \
+            -DBUILD_JPEG=ON \
+            -DWITH_PNG=ON \
+            -DBUILD_PNG=ON \
+            -DWITH_TIFF=ON \
+            -DBUILD_TIFF=ON
+
+        make -j
+        make install
+        cd ../..
+        echo "################### build opencv finished ###################"
+    fi
+fi
+
+
+echo "################### build PaddleOCR demo ####################"
+if [ ${use_opencv} = "True" ]; then
+    OPENCV_DIR=$(pwd)/opencv-3.4.7/opencv3/
+else
+    OPENCV_DIR=''
+fi
+if [ -d "paddle_inference/" ] ;then
+    echo "################### download paddle inference skipped ###################"
+else
+    echo "################### download paddle inference ###################"
+    PADDLEInfer=$3
+    if [ "" = "$PADDLEInfer" ];then
+        wget -nc https://paddle-inference-lib.bj.bcebos.com/2.3.0/cxx_c/Linux/GPU/x86-64_gcc8.2_avx_mkl_cuda10.1_cudnn7.6.5_trt6.0.1.5/paddle_inference.tgz --no-check-certificate
+    else
+        wget -nc $PADDLEInfer --no-check-certificate
+    fi
+    tar zxf paddle_inference.tgz
+    if [ ! -d "paddle_inference" ]; then
+        ln -s paddle_inference_install_dir paddle_inference
+    fi
+    echo "################### download paddle inference finished ###################"
+fi
+LIB_DIR=$(pwd)/paddle_inference/
+CUDA_LIB_DIR=$(dirname `find /usr -name libcudart.so`)
+CUDNN_LIB_DIR=$(dirname `find /usr -name libcudnn.so`)
+
+BUILD_DIR=build
+rm -rf ${BUILD_DIR}
+mkdir ${BUILD_DIR}
+cd ${BUILD_DIR}
+cmake .. \
+    -DPADDLE_LIB=${LIB_DIR} \
+    -DWITH_MKL=ON \
+    -DWITH_GPU=OFF \
+    -DWITH_STATIC_LIB=OFF \
+    -DWITH_TENSORRT=OFF \
+    -DOPENCV_DIR=${OPENCV_DIR} \
+    -DCUDNN_LIB=${CUDNN_LIB_DIR} \
+    -DCUDA_LIB=${CUDA_LIB_DIR} \
+    -DTENSORRT_DIR=${TENSORRT_DIR} \
+
+make -j
+cd ../../../
+echo "################### build PaddleOCR demo finished ###################"
+
+
+# set cuda device
+GPUID=$2
+if [ ${#GPUID} -le 0 ];then
+    env="export CUDA_VISIBLE_DEVICES=0"
+else
+    env="export CUDA_VISIBLE_DEVICES=${GPUID}"
+fi
+eval $env
+
+
+echo "################### run test ###################"
+export Count=0
+IFS="|"
+infer_quant_flag=(${cpp_infer_is_quant})
+for infer_model in ${cpp_infer_model_dir_list[*]}; do
+    #run inference
+    is_quant=${infer_quant_flag[Count]}
+    func_cpp_inference "${inference_cmd}" "${infer_model}" "${LOG_PATH}" "${cpp_infer_img_dir}" ${is_quant}
+    Count=$(($Count + 1))
+done
diff --git a/test_tipc/test_inference_python.sh b/test_tipc/test_inference_python.sh
new file mode 100644
index 0000000..a127358
--- /dev/null
+++ b/test_tipc/test_inference_python.sh
@@ -0,0 +1,169 @@
+#!/bin/bash
+source test_tipc/common_func.sh
+#source test_tipc/test_train_inference_python.sh
+
+FILENAME=$1
+# MODE be one of ['whole_infer']
+MODE=$2
+
+dataline=$(awk 'NR==1, NR==20{print}'  $FILENAME)
+
+# parser params
+IFS=$'\n'
+lines=(${dataline})
+
+model_name=$(func_parser_value "${lines[1]}")
+python=$(func_parser_value "${lines[2]}")
+
+infer_model_dir_list=$(func_parser_value "${lines[3]}")
+infer_export_list=$(func_parser_value "${lines[4]}")
+infer_is_quant=$(func_parser_value "${lines[5]}")
+# parser inference 
+inference_py=$(func_parser_value "${lines[6]}")
+use_gpu_key=$(func_parser_key "${lines[7]}")
+use_gpu_list=$(func_parser_value "${lines[7]}")
+use_mkldnn_key=$(func_parser_key "${lines[8]}")
+use_mkldnn_list=$(func_parser_value "${lines[8]}")
+cpu_threads_key=$(func_parser_key "${lines[9]}")
+cpu_threads_list=$(func_parser_value "${lines[9]}")
+batch_size_key=$(func_parser_key "${lines[10]}")
+batch_size_list=$(func_parser_value "${lines[10]}")
+use_trt_key=$(func_parser_key "${lines[11]}")
+use_trt_list=$(func_parser_value "${lines[11]}")
+precision_key=$(func_parser_key "${lines[12]}")
+precision_list=$(func_parser_value "${lines[12]}")
+infer_model_key=$(func_parser_key "${lines[13]}")
+image_dir_key=$(func_parser_key "${lines[14]}")
+infer_img_dir=$(func_parser_value "${lines[14]}")
+rec_model_key=$(func_parser_key "${lines[15]}")
+rec_model_value=$(func_parser_value "${lines[15]}")
+benchmark_key=$(func_parser_key "${lines[16]}")
+benchmark_value=$(func_parser_value "${lines[16]}")
+infer_key1=$(func_parser_key "${lines[17]}")
+infer_value1=$(func_parser_value "${lines[17]}")
+
+
+
+LOG_PATH="./test_tipc/output/${model_name}/${MODE}"
+mkdir -p ${LOG_PATH}
+status_log="${LOG_PATH}/results_python.log"
+
+
+function func_inference(){
+    IFS='|'
+    _python=$1
+    _script=$2
+    _model_dir=$3
+    _log_path=$4
+    _img_dir=$5
+    _flag_quant=$6
+    # inference 
+    for use_gpu in ${use_gpu_list[*]}; do
+        if [ ${use_gpu} = "False" ] || [ ${use_gpu} = "cpu" ]; then
+            for use_mkldnn in ${use_mkldnn_list[*]}; do
+                if [ ${use_mkldnn} = "False" ] && [ ${_flag_quant} = "True" ]; then
+                    continue
+                fi
+                for threads in ${cpu_threads_list[*]}; do
+                    for batch_size in ${batch_size_list[*]}; do
+                        for precision in ${precision_list[*]}; do
+                            if [ ${use_mkldnn} = "False" ] && [ ${precision} = "fp16" ]; then
+                                continue
+                            fi # skip when enable fp16 but disable mkldnn
+                            if [ ${_flag_quant} = "True" ] && [ ${precision} != "int8" ]; then
+                                continue
+                            fi # skip when quant model inference but precision is not int8
+                            set_precision=$(func_set_params "${precision_key}" "${precision}")
+                            
+                            _save_log_path="${_log_path}/python_infer_cpu_usemkldnn_${use_mkldnn}_threads_${threads}_precision_${precision}_batchsize_${batch_size}.log"
+                            set_infer_data=$(func_set_params "${image_dir_key}" "${_img_dir}")
+                            set_benchmark=$(func_set_params "${benchmark_key}" "${benchmark_value}")
+                            set_batchsize=$(func_set_params "${batch_size_key}" "${batch_size}")
+                            set_mkldnn=$(func_set_params "${use_mkldnn_key}" "${use_mkldnn}")
+                            set_cpu_threads=$(func_set_params "${cpu_threads_key}" "${threads}")
+                            set_model_dir=$(func_set_params "${infer_model_key}" "${_model_dir}")
+                            set_infer_params0=$(func_set_params "${rec_model_key}" "${rec_model_value}")
+                            set_infer_params1=$(func_set_params "${infer_key1}" "${infer_value1}")
+                            command="${_python} ${_script} ${use_gpu_key}=${use_gpu} ${set_mkldnn} ${set_cpu_threads} ${set_model_dir} ${set_batchsize} ${set_infer_params0} ${set_infer_data} ${set_benchmark} ${set_precision} ${set_infer_params1} > ${_save_log_path} 2>&1 "
+                            eval $command
+                            last_status=${PIPESTATUS[0]}
+                            eval "cat ${_save_log_path}"
+                            status_check $last_status "${command}" "${status_log}" "${model_name}" "${_save_log_path}"
+                        done
+                    done
+                done
+            done
+        elif [ ${use_gpu} = "True" ] || [ ${use_gpu} = "gpu" ]; then
+            for use_trt in ${use_trt_list[*]}; do
+                for precision in ${precision_list[*]}; do
+                    if [[ ${_flag_quant} = "False" ]] && [[ ${precision} =~ "int8" ]]; then
+                        continue
+                    fi 
+                    if [[ ${precision} =~ "fp16" || ${precision} =~ "int8" ]] && [ ${use_trt} = "False" ]; then
+                        continue
+                    fi
+                    if [[ ${use_trt} = "False" || ${precision} =~ "int8" ]] && [ ${_flag_quant} = "True" ]; then
+                        continue
+                    fi
+                    for batch_size in ${batch_size_list[*]}; do
+                        _save_log_path="${_log_path}/python_infer_gpu_usetrt_${use_trt}_precision_${precision}_batchsize_${batch_size}.log"
+                        set_infer_data=$(func_set_params "${image_dir_key}" "${_img_dir}")
+                        set_benchmark=$(func_set_params "${benchmark_key}" "${benchmark_value}")
+                        set_batchsize=$(func_set_params "${batch_size_key}" "${batch_size}")
+                        set_tensorrt=$(func_set_params "${use_trt_key}" "${use_trt}")
+                        set_precision=$(func_set_params "${precision_key}" "${precision}")
+                        set_model_dir=$(func_set_params "${infer_model_key}" "${_model_dir}")
+                        set_infer_params0=$(func_set_params "${rec_model_key}" "${rec_model_value}")
+                        set_infer_params1=$(func_set_params "${infer_key1}" "${infer_value1}")
+                        command="${_python} ${_script} ${use_gpu_key}=${use_gpu} ${set_tensorrt} ${set_precision} ${set_model_dir} ${set_batchsize} ${set_infer_data} ${set_benchmark} ${set_infer_params1} ${set_infer_params0} > ${_save_log_path} 2>&1 "
+                        eval $command
+                        last_status=${PIPESTATUS[0]}
+                        eval "cat ${_save_log_path}"
+                        status_check $last_status "${command}" "${status_log}" "${model_name}" "${_save_log_path}"
+                        
+                    done
+                done
+            done
+        else
+            echo "Does not support hardware other than CPU and GPU Currently!"
+        fi
+    done
+}
+
+if [ ${MODE} = "whole_infer" ]; then
+    GPUID=$3
+    if [ ${#GPUID} -le 0 ];then
+        env=" "
+    else
+        env="export CUDA_VISIBLE_DEVICES=${GPUID}"
+    fi
+    # set CUDA_VISIBLE_DEVICES
+    eval $env
+    export Count=0
+    IFS="|"
+    infer_run_exports=(${infer_export_list})
+    infer_quant_flag=(${infer_is_quant})
+    for infer_model in ${infer_model_dir_list[*]}; do
+        # run export
+        if [ ${infer_run_exports[Count]} != "null" ];then
+            _save_log_path="${_log_path}/python_infer_gpu_usetrt_${use_trt}_precision_${precision}_batchsize_${batch_size}_infermodel_${infer_model}.log"
+            save_infer_dir=$(dirname $infer_model)
+            set_export_weight=$(func_set_params "${export_weight}" "${infer_model}")
+            set_save_infer_key=$(func_set_params "${save_infer_key}" "${save_infer_dir}")
+            export_cmd="${python} ${infer_run_exports[Count]} ${set_export_weight} ${set_save_infer_key} > ${_save_log_path} 2>&1 "
+            echo ${infer_run_exports[Count]} 
+            eval $export_cmd
+            status_export=$?
+            status_check $status_export "${export_cmd}" "${status_log}" "${model_name}" "${_save_log_path}"
+        else
+            save_infer_dir=${infer_model}
+        fi
+        #run inference
+        is_quant=${infer_quant_flag[Count]}
+        if [ ${MODE} = "klquant_infer" ]; then
+            is_quant="True"
+        fi
+        func_inference "${python}" "${inference_py}" "${save_infer_dir}" "${LOG_PATH}" "${infer_img_dir}" ${is_quant}
+        Count=$(($Count + 1))
+    done
+fi
diff --git a/test_tipc/test_lite_arm_cpp.sh b/test_tipc/test_lite_arm_cpp.sh
new file mode 100644
index 0000000..c071a23
--- /dev/null
+++ b/test_tipc/test_lite_arm_cpp.sh
@@ -0,0 +1,159 @@
+#!/bin/bash
+source ./common_func.sh
+export LD_LIBRARY_PATH=${PWD}:$LD_LIBRARY_PATH
+
+FILENAME=$1
+dataline=$(cat $FILENAME)
+# parser params
+IFS=$'\n'
+lines=(${dataline})
+
+# parser lite inference
+inference_cmd=$(func_parser_value "${lines[1]}")
+runtime_device=$(func_parser_value "${lines[2]}")
+det_model_list=$(func_parser_value "${lines[3]}")
+rec_model_list=$(func_parser_value "${lines[4]}")
+cls_model_list=$(func_parser_value "${lines[5]}")
+cpu_threads_list=$(func_parser_value "${lines[6]}")
+det_batch_size_list=$(func_parser_value "${lines[7]}")
+rec_batch_size_list=$(func_parser_value "${lines[8]}")
+infer_img_dir_list=$(func_parser_value "${lines[9]}")
+config_dir=$(func_parser_value "${lines[10]}")
+rec_dict_dir=$(func_parser_value "${lines[11]}")
+benchmark_value=$(func_parser_value "${lines[12]}")
+
+if [[ $inference_cmd =~ "det" ]]; then
+    lite_model_list=${det_lite_model_list}
+elif [[ $inference_cmd =~ "rec" ]]; then
+    lite_model_list=(${rec_lite_model_list[*]} ${cls_lite_model_list[*]})
+elif [[ $inference_cmd =~ "system" ]]; then
+    lite_model_list=(${det_lite_model_list[*]} ${rec_lite_model_list[*]} ${cls_lite_model_list[*]})
+else
+    echo "inference_cmd is wrong, please check."
+    exit 1
+fi
+
+LOG_PATH="./output"
+mkdir -p ${LOG_PATH}
+status_log="${LOG_PATH}/results.log"
+
+
+function func_test_det(){
+    IFS='|'
+    _script=$1
+    _det_model=$2
+    _log_path=$3
+    _img_dir=$4
+    _config=$5
+    if [[ $_det_model =~ "slim" ]]; then
+        precision="INT8"
+    else
+        precision="FP32"
+    fi
+
+    # lite inference
+    for num_threads in ${cpu_threads_list[*]}; do
+	for det_batchsize in ${det_batch_size_list[*]}; do
+            _save_log_path="${_log_path}/lite_${_det_model}_runtime_device_${runtime_device}_precision_${precision}_det_batchsize_${det_batchsize}_threads_${num_threads}.log"
+            command="${_script} ${_det_model} ${runtime_device} ${precision} ${num_threads} ${det_batchsize}  ${_img_dir} ${_config} ${benchmark_value} > ${_save_log_path} 2>&1"
+            eval ${command}
+            status_check $? "${command}" "${status_log}"
+        done
+    done
+}
+
+function func_test_rec(){
+    IFS='|'
+    _script=$1
+    _rec_model=$2
+    _cls_model=$3
+    _log_path=$4
+    _img_dir=$5
+    _config=$6
+    _rec_dict_dir=$7
+
+    if [[ $_det_model =~ "slim" ]]; then
+        _precision="INT8"
+    else
+        _precision="FP32"
+    fi
+
+    # lite inference
+    for num_threads in ${cpu_threads_list[*]}; do
+	for rec_batchsize in ${rec_batch_size_list[*]}; do
+            _save_log_path="${_log_path}/lite_${_rec_model}_${cls_model}_runtime_device_${runtime_device}_precision_${_precision}_rec_batchsize_${rec_batchsize}_threads_${num_threads}.log"
+            command="${_script} ${_rec_model} ${_cls_model} ${runtime_device} ${_precision} ${num_threads} ${rec_batchsize}  ${_img_dir} ${_config} ${_rec_dict_dir} ${benchmark_value} > ${_save_log_path} 2>&1"
+            eval ${command}
+            status_check $? "${command}" "${status_log}"
+        done
+    done
+}
+
+function func_test_system(){
+    IFS='|'
+    _script=$1
+    _det_model=$2
+    _rec_model=$3
+    _cls_model=$4
+    _log_path=$5
+    _img_dir=$6
+    _config=$7
+    _rec_dict_dir=$8
+    if [[ $_det_model =~ "slim" ]]; then
+        _precision="INT8"
+    else
+        _precision="FP32"
+    fi
+
+    # lite inference
+    for num_threads in ${cpu_threads_list[*]}; do
+	for det_batchsize in ${det_batch_size_list[*]}; do
+	   for rec_batchsize in ${rec_batch_size_list[*]}; do
+                _save_log_path="${_log_path}/lite_${_det_model}_${_rec_model}_${_cls_model}_runtime_device_${runtime_device}_precision_${_precision}_det_batchsize_${det_batchsize}_rec_batchsize_${rec_batchsize}_threads_${num_threads}.log"
+                command="${_script} ${_det_model} ${_rec_model} ${_cls_model} ${runtime_device} ${_precision} ${num_threads} ${det_batchsize}  ${_img_dir} ${_config} ${_rec_dict_dir} ${benchmark_value} > ${_save_log_path} 2>&1"
+               eval ${command}
+               status_check $? "${command}" "${status_log}"
+	    done
+        done
+    done
+}
+
+
+echo "################### run test ###################"
+
+if [[ $inference_cmd =~ "det" ]]; then
+    IFS="|"
+    det_model_list=(${det_model_list[*]})
+
+    for i in {0..1}; do
+        #run lite inference
+        for img_dir in ${infer_img_dir_list[*]}; do
+            func_test_det "${inference_cmd}" "${det_model_list[i]}_opt.nb" "${LOG_PATH}" "${img_dir}" "${config_dir}"
+        done
+    done
+
+elif [[ $inference_cmd =~ "rec" ]]; then
+    IFS="|"
+    rec_model_list=(${rec_model_list[*]})
+    cls_model_list=(${cls_model_list[*]})
+
+    for i in {0..1}; do
+        #run lite inference
+        for img_dir in ${infer_img_dir_list[*]}; do
+            func_test_rec "${inference_cmd}" "${rec_model}_opt.nb" "${cls_model_list[i]}_opt.nb" "${LOG_PATH}" "${img_dir}" "${rec_dict_dir}" "${config_dir}"
+        done
+    done
+
+elif [[ $inference_cmd =~ "system" ]]; then
+    IFS="|"
+    det_model_list=(${det_model_list[*]})
+    rec_model_list=(${rec_model_list[*]})
+    cls_model_list=(${cls_model_list[*]})
+
+    for i in {0..1}; do
+	#run lite inference
+        for img_dir in ${infer_img_dir_list[*]}; do
+            func_test_system "${inference_cmd}" "${det_model_list[i]}_opt.nb" "${rec_model_list[i]}_opt.nb" "${cls_model_list[i]}_opt.nb" "${LOG_PATH}" "${img_dir}" "${config_dir}" "${rec_dict_dir}"
+        done
+    done
+fi
diff --git a/test_tipc/test_paddle2onnx.sh b/test_tipc/test_paddle2onnx.sh
new file mode 100644
index 0000000..f035e6b
--- /dev/null
+++ b/test_tipc/test_paddle2onnx.sh
@@ -0,0 +1,174 @@
+#!/bin/bash
+source test_tipc/common_func.sh 
+
+FILENAME=$1
+
+dataline=$(cat ${FILENAME})
+lines=(${dataline})
+# common params
+model_name=$(func_parser_value "${lines[1]}")
+python=$(func_parser_value "${lines[2]}")
+
+
+# parser params
+dataline=$(awk 'NR==1, NR==17{print}'  $FILENAME)
+IFS=$'\n'
+lines=(${dataline})
+
+# parser paddle2onnx
+model_name=$(func_parser_value "${lines[1]}")
+python=$(func_parser_value "${lines[2]}")
+padlle2onnx_cmd=$(func_parser_value "${lines[3]}")
+det_infer_model_dir_key=$(func_parser_key "${lines[4]}")
+det_infer_model_dir_value=$(func_parser_value "${lines[4]}")
+model_filename_key=$(func_parser_key "${lines[5]}")
+model_filename_value=$(func_parser_value "${lines[5]}")
+params_filename_key=$(func_parser_key "${lines[6]}")
+params_filename_value=$(func_parser_value "${lines[6]}")
+det_save_file_key=$(func_parser_key "${lines[7]}")
+det_save_file_value=$(func_parser_value "${lines[7]}")
+rec_infer_model_dir_key=$(func_parser_key "${lines[8]}")
+rec_infer_model_dir_value=$(func_parser_value "${lines[8]}")
+rec_save_file_key=$(func_parser_key "${lines[9]}")
+rec_save_file_value=$(func_parser_value "${lines[9]}")
+opset_version_key=$(func_parser_key "${lines[10]}")
+opset_version_value=$(func_parser_value "${lines[10]}")
+enable_onnx_checker_key=$(func_parser_key "${lines[11]}")
+enable_onnx_checker_value=$(func_parser_value "${lines[11]}")
+# parser onnx inference 
+inference_py=$(func_parser_value "${lines[12]}")
+use_gpu_key=$(func_parser_key "${lines[13]}")
+use_gpu_list=$(func_parser_value "${lines[13]}")
+det_model_key=$(func_parser_key "${lines[14]}")
+rec_model_key=$(func_parser_key "${lines[15]}")
+image_dir_key=$(func_parser_key "${lines[16]}")
+image_dir_value=$(func_parser_value "${lines[16]}")
+
+LOG_PATH="./test_tipc/output/${model_name}/paddle2onnx"
+mkdir -p ${LOG_PATH}
+status_log="${LOG_PATH}/results_paddle2onnx.log"
+
+
+function func_paddle2onnx(){
+    IFS='|'
+    _script=$1
+
+    # paddle2onnx
+    if [ ${model_name} = "ch_PP-OCRv2" ] || [ ${model_name} = "ch_PP-OCRv3" ] || [ ${model_name} = "ch_ppocr_mobile_v2_0" ] || [ ${model_name} = "ch_ppocr_server_v2_0" ]; then
+        # trans det
+        set_dirname=$(func_set_params "--model_dir" "${det_infer_model_dir_value}")
+        set_model_filename=$(func_set_params "${model_filename_key}" "${model_filename_value}")
+        set_params_filename=$(func_set_params "${params_filename_key}" "${params_filename_value}")
+        set_save_model=$(func_set_params "--save_file" "${det_save_file_value}")
+        set_opset_version=$(func_set_params "${opset_version_key}" "${opset_version_value}")
+        set_enable_onnx_checker=$(func_set_params "${enable_onnx_checker_key}" "${enable_onnx_checker_value}")
+        trans_det_log="${LOG_PATH}/trans_model_det.log"
+        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker} --enable_dev_version=False > ${trans_det_log} 2>&1 "
+        eval $trans_model_cmd
+        last_status=${PIPESTATUS[0]}
+        status_check $last_status "${trans_model_cmd}" "${status_log}" "${model_name}" "${trans_det_log}"
+        # trans rec
+        set_dirname=$(func_set_params "--model_dir" "${rec_infer_model_dir_value}")
+        set_model_filename=$(func_set_params "${model_filename_key}" "${model_filename_value}")
+        set_params_filename=$(func_set_params "${params_filename_key}" "${params_filename_value}")
+        set_save_model=$(func_set_params "--save_file" "${rec_save_file_value}")
+        set_opset_version=$(func_set_params "${opset_version_key}" "${opset_version_value}")
+        set_enable_onnx_checker=$(func_set_params "${enable_onnx_checker_key}" "${enable_onnx_checker_value}")
+        trans_rec_log="${LOG_PATH}/trans_model_rec.log"
+        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker} --enable_dev_version=False > ${trans_rec_log} 2>&1 "
+        eval $trans_model_cmd
+        last_status=${PIPESTATUS[0]}
+        status_check $last_status "${trans_model_cmd}" "${status_log}" "${model_name}" "${trans_rec_log}"
+    elif [[ ${model_name} =~ "det" ]]; then
+        # trans det
+        set_dirname=$(func_set_params "--model_dir" "${det_infer_model_dir_value}")
+        set_model_filename=$(func_set_params "${model_filename_key}" "${model_filename_value}")
+        set_params_filename=$(func_set_params "${params_filename_key}" "${params_filename_value}")
+        set_save_model=$(func_set_params "--save_file" "${det_save_file_value}")
+        set_opset_version=$(func_set_params "${opset_version_key}" "${opset_version_value}")
+        set_enable_onnx_checker=$(func_set_params "${enable_onnx_checker_key}" "${enable_onnx_checker_value}")
+        trans_det_log="${LOG_PATH}/trans_model_det.log"
+        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker} --enable_dev_version=False > ${trans_det_log} 2>&1 "
+        eval $trans_model_cmd
+        last_status=${PIPESTATUS[0]}
+        status_check $last_status "${trans_model_cmd}" "${status_log}" "${model_name}" "${trans_det_log}" 
+    elif [[ ${model_name} =~ "rec" ]]; then
+        # trans rec
+        set_dirname=$(func_set_params "--model_dir" "${rec_infer_model_dir_value}")
+        set_model_filename=$(func_set_params "${model_filename_key}" "${model_filename_value}")
+        set_params_filename=$(func_set_params "${params_filename_key}" "${params_filename_value}")
+        set_save_model=$(func_set_params "--save_file" "${rec_save_file_value}")
+        set_opset_version=$(func_set_params "${opset_version_key}" "${opset_version_value}")
+        set_enable_onnx_checker=$(func_set_params "${enable_onnx_checker_key}" "${enable_onnx_checker_value}")
+        trans_rec_log="${LOG_PATH}/trans_model_rec.log"
+        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker} --enable_dev_version=False > ${trans_rec_log} 2>&1 "
+        eval $trans_model_cmd
+        last_status=${PIPESTATUS[0]}
+        status_check $last_status "${trans_model_cmd}" "${status_log}" "${model_name}" "${trans_rec_log}"
+    elif [ ${model_name} = "slanet" ] || [ ${model_name} = "en_table_structure" ]; then
+        # trans det
+        set_dirname=$(func_set_params "--model_dir" "${det_infer_model_dir_value}")
+        set_model_filename=$(func_set_params "${model_filename_key}" "${model_filename_value}")
+        set_params_filename=$(func_set_params "${params_filename_key}" "${params_filename_value}")
+        set_save_model=$(func_set_params "--save_file" "${det_save_file_value}")
+        set_opset_version=$(func_set_params "${opset_version_key}" "${opset_version_value}")
+        set_enable_onnx_checker=$(func_set_params "${enable_onnx_checker_key}" "${enable_onnx_checker_value}")
+        trans_det_log="${LOG_PATH}/trans_model_det.log"
+        trans_model_cmd="${padlle2onnx_cmd} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_save_model} ${set_opset_version} ${set_enable_onnx_checker} --enable_dev_version=True > ${trans_det_log} 2>&1 "
+        eval $trans_model_cmd
+        last_status=${PIPESTATUS[0]}
+        status_check $last_status "${trans_model_cmd}" "${status_log}" "${model_name}" "${trans_det_log}" 
+    fi
+
+    # python inference
+    for use_gpu in ${use_gpu_list[*]}; do
+        if [ ${use_gpu} = "False" ] || [ ${use_gpu} = "cpu" ]; then
+            _save_log_path="${LOG_PATH}/paddle2onnx_infer_cpu.log"
+            set_gpu=$(func_set_params "${use_gpu_key}" "${use_gpu}")
+            set_img_dir=$(func_set_params "${image_dir_key}" "${image_dir_value}")
+            if [ ${model_name} = "ch_PP-OCRv2" ] || [ ${model_name} = "ch_PP-OCRv3" ] || [ ${model_name} = "ch_ppocr_mobile_v2_0" ] || [ ${model_name} = "ch_ppocr_server_v2_0" ]; then
+                set_det_model_dir=$(func_set_params "${det_model_key}" "${det_save_file_value}")
+                set_rec_model_dir=$(func_set_params "${rec_model_key}" "${rec_save_file_value}")
+                infer_model_cmd="${python} ${inference_py} ${set_gpu} ${set_img_dir} ${set_det_model_dir} ${set_rec_model_dir} --use_onnx=True > ${_save_log_path} 2>&1 "
+            elif [[ ${model_name} =~ "det" ]] || [ ${model_name} = "slanet" ] || [ ${model_name} = "en_table_structure" ]; then
+                set_det_model_dir=$(func_set_params "${det_model_key}" "${det_save_file_value}")
+                infer_model_cmd="${python} ${inference_py} ${set_gpu} ${set_img_dir} ${set_det_model_dir} --use_onnx=True > ${_save_log_path} 2>&1 "
+            elif [[ ${model_name} =~ "rec" ]]; then
+                set_rec_model_dir=$(func_set_params "${rec_model_key}" "${rec_save_file_value}")
+                infer_model_cmd="${python} ${inference_py} ${set_gpu} ${set_img_dir} ${set_rec_model_dir} --use_onnx=True > ${_save_log_path} 2>&1 "
+            fi
+            eval $infer_model_cmd
+            last_status=${PIPESTATUS[0]}
+            eval "cat ${_save_log_path}"
+            status_check $last_status "${infer_model_cmd}" "${status_log}" "${model_name}" "${_save_log_path}"
+        elif [ ${use_gpu} = "True" ] || [ ${use_gpu} = "gpu" ]; then
+            _save_log_path="${LOG_PATH}/paddle2onnx_infer_gpu.log"
+            set_gpu=$(func_set_params "${use_gpu_key}" "${use_gpu}")
+            set_img_dir=$(func_set_params "${image_dir_key}" "${image_dir_value}")
+            if [ ${model_name} = "ch_PP-OCRv2" ] || [ ${model_name} = "ch_PP-OCRv3" ] || [ ${model_name} = "ch_ppocr_mobile_v2_0" ] || [ ${model_name} = "ch_ppocr_server_v2_0" ]; then
+                set_det_model_dir=$(func_set_params "${det_model_key}" "${det_save_file_value}")
+                set_rec_model_dir=$(func_set_params "${rec_model_key}" "${rec_save_file_value}")
+                infer_model_cmd="${python} ${inference_py} ${set_gpu} ${set_img_dir} ${set_det_model_dir} ${set_rec_model_dir} --use_onnx=True > ${_save_log_path} 2>&1 "
+            elif [[ ${model_name} =~ "det" ]]|| [ ${model_name} = "slanet" ] || [ ${model_name} = "en_table_structure" ]; then
+                set_det_model_dir=$(func_set_params "${det_model_key}" "${det_save_file_value}")
+                infer_model_cmd="${python} ${inference_py} ${set_gpu} ${set_img_dir} ${set_det_model_dir} --use_onnx=True > ${_save_log_path} 2>&1 "
+            elif [[ ${model_name} =~ "rec" ]]; then
+                set_rec_model_dir=$(func_set_params "${rec_model_key}" "${rec_save_file_value}")
+                infer_model_cmd="${python} ${inference_py} ${set_gpu} ${set_img_dir} ${set_rec_model_dir} --use_onnx=True > ${_save_log_path} 2>&1 "
+            fi
+            eval $infer_model_cmd
+            last_status=${PIPESTATUS[0]}
+            eval "cat ${_save_log_path}"
+            status_check $last_status "${infer_model_cmd}" "${status_log}" "${model_name}" "${_save_log_path}"
+        else
+            echo "Does not support hardware other than CPU and GPU Currently!"
+        fi
+    done
+}
+
+
+echo "################### run test ###################"
+
+export Count=0
+IFS="|"
+func_paddle2onnx 
diff --git a/test_tipc/test_ptq_inference_python.sh b/test_tipc/test_ptq_inference_python.sh
new file mode 100644
index 0000000..f9cd829
--- /dev/null
+++ b/test_tipc/test_ptq_inference_python.sh
@@ -0,0 +1,157 @@
+#!/bin/bash
+source test_tipc/common_func.sh
+
+FILENAME=$1
+# MODE be one of [''whole_infer']
+MODE=$2
+
+IFS=$'\n'
+# parser klquant_infer params
+
+dataline=$(awk 'NR==1, NR==17{print}'  $FILENAME)
+lines=(${dataline})
+model_name=$(func_parser_value "${lines[1]}")
+python=$(func_parser_value "${lines[2]}")
+export_weight=$(func_parser_key "${lines[3]}")
+save_infer_key=$(func_parser_key "${lines[4]}")
+# parser inference model 
+infer_model_dir_list=$(func_parser_value "${lines[5]}")
+infer_export_list=$(func_parser_value "${lines[6]}")
+infer_is_quant=$(func_parser_value "${lines[7]}")
+# parser inference 
+inference_py=$(func_parser_value "${lines[8]}")
+use_gpu_key=$(func_parser_key "${lines[9]}")
+use_gpu_list=$(func_parser_value "${lines[9]}")
+use_mkldnn_key=$(func_parser_key "${lines[10]}")
+use_mkldnn_list=$(func_parser_value "${lines[10]}")
+cpu_threads_key=$(func_parser_key "${lines[11]}")
+cpu_threads_list=$(func_parser_value "${lines[11]}")
+batch_size_key=$(func_parser_key "${lines[12]}")
+batch_size_list=$(func_parser_value "${lines[12]}")
+use_trt_key=$(func_parser_key "${lines[13]}")
+use_trt_list=$(func_parser_value "${lines[13]}")
+precision_key=$(func_parser_key "${lines[14]}")
+precision_list=$(func_parser_value "${lines[14]}")
+infer_model_key=$(func_parser_key "${lines[15]}")
+image_dir_key=$(func_parser_key "${lines[16]}")
+infer_img_dir=$(func_parser_value "${lines[16]}")
+save_log_key=$(func_parser_key "${lines[17]}")
+save_log_value=$(func_parser_value "${lines[17]}")
+benchmark_key=$(func_parser_key "${lines[18]}")
+benchmark_value=$(func_parser_value "${lines[18]}")
+infer_key1=$(func_parser_key "${lines[19]}")
+infer_value1=$(func_parser_value "${lines[19]}")
+
+
+LOG_PATH="./test_tipc/output/${model_name}/${MODE}"
+mkdir -p ${LOG_PATH}
+status_log="${LOG_PATH}/results_python.log"
+
+
+function func_inference(){
+    IFS='|'
+    _python=$1
+    _script=$2
+    _model_dir=$3
+    _log_path=$4
+    _img_dir=$5
+    _flag_quant=$6
+    # inference 
+    for use_gpu in ${use_gpu_list[*]}; do
+        if [ ${use_gpu} = "False" ] || [ ${use_gpu} = "cpu" ]; then
+            for use_mkldnn in ${use_mkldnn_list[*]}; do
+                for threads in ${cpu_threads_list[*]}; do
+                    for batch_size in ${batch_size_list[*]}; do
+                        for precision in ${precision_list[*]}; do
+                            if [ ${use_mkldnn} = "False" ] && [ ${precision} = "fp16" ]; then
+                                continue
+                            fi # skip when enable fp16 but disable mkldnn
+                            if [ ${_flag_quant} = "True" ] && [ ${precision} != "int8" ]; then
+                                continue
+                            fi # skip when quant model inference but precision is not int8
+                            set_precision=$(func_set_params "${precision_key}" "${precision}")
+                            
+                            _save_log_path="${_log_path}/python_infer_cpu_usemkldnn_${use_mkldnn}_threads_${threads}_precision_${precision}_batchsize_${batch_size}.log"
+                            set_infer_data=$(func_set_params "${image_dir_key}" "${_img_dir}")
+                            set_benchmark=$(func_set_params "${benchmark_key}" "${benchmark_value}")
+                            set_batchsize=$(func_set_params "${batch_size_key}" "${batch_size}")
+                            set_mkldnn=$(func_set_params "${use_mkldnn_key}" "${use_mkldnn}")
+                            set_cpu_threads=$(func_set_params "${cpu_threads_key}" "${threads}")
+                            set_model_dir=$(func_set_params "${infer_model_key}" "${_model_dir}")
+                            set_infer_params0=$(func_set_params "${save_log_key}" "${save_log_value}")
+                            set_infer_params1=$(func_set_params "${infer_key1}" "${infer_value1}")
+                            command="${_python} ${_script} ${use_gpu_key}=${use_gpu} ${set_mkldnn} ${set_cpu_threads} ${set_model_dir} ${set_batchsize} ${set_infer_params0} ${set_infer_data} ${set_benchmark} ${set_precision} ${set_infer_params1} > ${_save_log_path} 2>&1 "
+                            eval $command
+                            last_status=${PIPESTATUS[0]}
+                            eval "cat ${_save_log_path}"
+                            status_check $last_status "${command}" "${status_log}" "${model_name}" "${_save_log_path}"
+                        done
+                    done
+                done
+            done
+        elif [ ${use_gpu} = "True" ] || [ ${use_gpu} = "gpu" ]; then
+            for use_trt in ${use_trt_list[*]}; do
+                for precision in ${precision_list[*]}; do
+                    if [ ${_flag_quant} = "True" ] && [ ${precision} != "int8" ]; then
+                        continue
+                    fi # skip when quant model inference but precision is not int8
+                    for batch_size in ${batch_size_list[*]}; do
+                        _save_log_path="${_log_path}/python_infer_gpu_usetrt_${use_trt}_precision_${precision}_batchsize_${batch_size}.log"
+                        set_infer_data=$(func_set_params "${image_dir_key}" "${_img_dir}")
+                        set_benchmark=$(func_set_params "${benchmark_key}" "${benchmark_value}")
+                        set_batchsize=$(func_set_params "${batch_size_key}" "${batch_size}")
+                        set_tensorrt=$(func_set_params "${use_trt_key}" "${use_trt}")
+                        set_precision=$(func_set_params "${precision_key}" "${precision}")
+                        set_model_dir=$(func_set_params "${infer_model_key}" "${_model_dir}")
+                        set_infer_params0=$(func_set_params "${save_log_key}" "${save_log_value}")
+                        set_infer_params1=$(func_set_params "${infer_key1}" "${infer_value1}")
+                        command="${_python} ${_script} ${use_gpu_key}=${use_gpu} ${set_tensorrt} ${set_precision} ${set_model_dir} ${set_batchsize} ${set_infer_data} ${set_benchmark} ${set_infer_params1} ${set_infer_params0} > ${_save_log_path} 2>&1 "
+                        eval $command
+                        last_status=${PIPESTATUS[0]}
+                        eval "cat ${_save_log_path}"
+                        status_check $last_status "${command}" "${status_log}" "${model_name}" "${_save_log_path}"
+                        
+                    done
+                done
+            done
+        else
+            echo "Does not support hardware other than CPU and GPU Currently!"
+        fi
+    done
+}
+
+if [ ${MODE} = "whole_infer" ]; then
+    GPUID=$3
+    if [ ${#GPUID} -le 0 ];then
+        env=" "
+    else
+        env="export CUDA_VISIBLE_DEVICES=${GPUID}"
+    fi
+    # set CUDA_VISIBLE_DEVICES
+    eval $env
+    export Count=0
+    IFS="|"
+    infer_run_exports=(${infer_export_list})
+    infer_quant_flag=(${infer_is_quant})
+    for infer_model in ${infer_model_dir_list[*]}; do
+        # run export
+        if [ ${infer_run_exports[Count]} != "null" ];then
+            save_infer_dir="${infer_model}_klquant"
+            set_export_weight=$(func_set_params "${export_weight}" "${infer_model}")
+            set_save_infer_key=$(func_set_params "${save_infer_key}" "${save_infer_dir}")
+            export_log_path="${LOG_PATH}/${MODE}_export_${Count}.log"
+            export_cmd="${python} ${infer_run_exports[Count]} ${set_export_weight} ${set_save_infer_key} > ${export_log_path} 2>&1 "
+            echo ${infer_run_exports[Count]} 
+            echo $export_cmd
+            eval $export_cmd
+            status_export=$?
+            status_check $status_export "${export_cmd}" "${status_log}" "${model_name}" "${export_log_path}"
+        else
+            save_infer_dir=${infer_model}
+        fi
+        #run inference
+        is_quant="True"
+        func_inference "${python}" "${inference_py}" "${save_infer_dir}" "${LOG_PATH}" "${infer_img_dir}" ${is_quant}
+        Count=$(($Count + 1))
+    done
+fi
diff --git a/test_tipc/test_serving_infer_cpp.sh b/test_tipc/test_serving_infer_cpp.sh
new file mode 100644
index 0000000..6de6856
--- /dev/null
+++ b/test_tipc/test_serving_infer_cpp.sh
@@ -0,0 +1,142 @@
+#!/bin/bash
+source test_tipc/common_func.sh
+
+function func_parser_model_config(){
+    strs=$1
+    IFS="/"
+    array=(${strs})
+    tmp=${array[-1]}
+    echo ${tmp}
+}
+
+FILENAME=$1
+dataline=$(awk 'NR==1, NR==19{print}'  $FILENAME)
+MODE=$2
+
+# parser params
+IFS=$'\n'
+lines=(${dataline})
+
+# parser serving
+model_name=$(func_parser_value "${lines[1]}")
+python_list=$(func_parser_value "${lines[2]}")
+trans_model_py=$(func_parser_value "${lines[3]}")
+det_infer_model_dir_key=$(func_parser_key "${lines[4]}")
+det_infer_model_dir_value=$(func_parser_value "${lines[4]}")
+model_filename_key=$(func_parser_key "${lines[5]}")
+model_filename_value=$(func_parser_value "${lines[5]}")
+params_filename_key=$(func_parser_key "${lines[6]}")
+params_filename_value=$(func_parser_value "${lines[6]}")
+det_serving_server_key=$(func_parser_key "${lines[7]}")
+det_serving_server_value=$(func_parser_value "${lines[7]}")
+det_serving_client_key=$(func_parser_key "${lines[8]}")
+det_serving_client_value=$(func_parser_value "${lines[8]}")
+rec_infer_model_dir_key=$(func_parser_key "${lines[9]}")
+rec_infer_model_dir_value=$(func_parser_value "${lines[9]}")
+rec_serving_server_key=$(func_parser_key "${lines[10]}")
+rec_serving_server_value=$(func_parser_value "${lines[10]}")
+rec_serving_client_key=$(func_parser_key "${lines[11]}")
+rec_serving_client_value=$(func_parser_value "${lines[11]}")
+det_server_value=$(func_parser_model_config "${lines[7]}")
+det_client_value=$(func_parser_model_config "${lines[8]}")
+rec_server_value=$(func_parser_model_config "${lines[10]}")
+rec_client_value=$(func_parser_model_config "${lines[11]}")
+serving_dir_value=$(func_parser_value "${lines[12]}")
+web_service_py=$(func_parser_value "${lines[13]}")
+op_key=$(func_parser_key "${lines[14]}")
+op_value=$(func_parser_value "${lines[14]}")
+port_key=$(func_parser_key "${lines[15]}")
+port_value=$(func_parser_value "${lines[15]}")
+gpu_key=$(func_parser_key "${lines[16]}")
+gpu_value=$(func_parser_value "${lines[16]}")
+cpp_client_py=$(func_parser_value "${lines[17]}")
+image_dir_key=$(func_parser_key "${lines[18]}")
+image_dir_value=$(func_parser_value "${lines[18]}")
+
+LOG_PATH="$(pwd)/test_tipc/output/${model_name}/${MODE}/cpp"
+mkdir -p ${LOG_PATH}
+status_log="${LOG_PATH}/results_cpp_serving.log"
+
+function func_serving(){
+    IFS='|'
+    _python=$1
+    _script=$2
+    _model_dir=$3
+    # pdserving
+    set_model_filename=$(func_set_params "${model_filename_key}" "${model_filename_value}")
+    set_params_filename=$(func_set_params "${params_filename_key}" "${params_filename_value}")
+    # trans det
+    set_dirname=$(func_set_params "--dirname" "${det_infer_model_dir_value}")
+    set_serving_server=$(func_set_params "--serving_server" "${det_serving_server_value}")
+    set_serving_client=$(func_set_params "--serving_client" "${det_serving_client_value}")
+    python_list=(${python_list})
+    trans_det_log="${LOG_PATH}/cpp_trans_model_det.log"
+    trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_det_log} 2>&1 "
+    eval $trans_model_cmd
+    cp "deploy/pdserving/serving_client_conf.prototxt" ${det_serving_client_value}
+    # trans rec
+    set_dirname=$(func_set_params "--dirname" "${rec_infer_model_dir_value}")
+    set_serving_server=$(func_set_params "--serving_server" "${rec_serving_server_value}")
+    set_serving_client=$(func_set_params "--serving_client" "${rec_serving_client_value}")
+    python_list=(${python_list})
+    trans_rec_log="${LOG_PATH}/cpp_trans_model_rec.log"
+    trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_rec_log} 2>&1 "
+    eval $trans_model_cmd
+    last_status=${PIPESTATUS[0]}
+    status_check $last_status "${trans_model_cmd}" "${status_log}" "${model_name}" "${trans_rec_log}"
+    set_image_dir=$(func_set_params "${image_dir_key}" "${image_dir_value}")
+    python_list=(${python_list})
+    cd ${serving_dir_value}
+    
+    # cpp serving
+    for gpu_id in ${gpu_value[*]}; do
+        if [ ${gpu_id} = "null" ]; then
+            server_log_path="${LOG_PATH}/cpp_server_cpu.log"
+            web_service_cpp_cmd="nohup ${python_list[0]} ${web_service_py} --model ${det_server_value} ${rec_server_value} ${op_key} ${op_value} ${port_key} ${port_value} > ${server_log_path} 2>&1 &"
+            eval $web_service_cpp_cmd
+            last_status=${PIPESTATUS[0]}
+            status_check $last_status "${web_service_cpp_cmd}" "${status_log}" "${model_name}" "${server_log_path}"
+            sleep 5s
+            _save_log_path="${LOG_PATH}/cpp_client_cpu.log"
+            cpp_client_cmd="${python_list[0]} ${cpp_client_py} ${det_client_value} ${rec_client_value} > ${_save_log_path} 2>&1"
+            eval $cpp_client_cmd
+            last_status=${PIPESTATUS[0]}
+            eval "cat ${_save_log_path}"
+            status_check $last_status "${cpp_client_cmd}" "${status_log}" "${model_name}" "${_save_log_path}"
+            #ps ux | grep -i ${port_value} | awk '{print $2}' | xargs kill -s 9
+            ${python_list[0]} ${web_service_py} stop
+            sleep 5s
+        else
+            server_log_path="${LOG_PATH}/cpp_server_gpu.log"
+            web_service_cpp_cmd="nohup ${python_list[0]} ${web_service_py} --model ${det_server_value} ${rec_server_value} ${op_key} ${op_value} ${port_key} ${port_value} ${gpu_key} ${gpu_id} > ${server_log_path} 2>&1 &"
+            eval $web_service_cpp_cmd
+            sleep 5s
+            _save_log_path="${LOG_PATH}/cpp_client_gpu.log"
+            cpp_client_cmd="${python_list[0]} ${cpp_client_py} ${det_client_value} ${rec_client_value} > ${_save_log_path} 2>&1"
+            eval $cpp_client_cmd
+            last_status=${PIPESTATUS[0]}
+            eval "cat ${_save_log_path}" 
+            status_check $last_status "${cpp_client_cmd}" "${status_log}" "${model_name}" "${_save_log_path}"
+            #ps ux | grep -i ${port_value} | awk '{print $2}' | xargs kill -s 9
+            ${python_list[0]} ${web_service_py} stop
+        fi
+    done
+}
+
+
+#set cuda device
+GPUID=$3
+if [ ${#GPUID} -le 0 ];then
+    env="export CUDA_VISIBLE_DEVICES=0"
+else
+    env="export CUDA_VISIBLE_DEVICES=${GPUID}"
+fi
+eval $env
+echo $env
+
+
+echo "################### run test ###################"
+
+export Count=0
+IFS="|"
+func_serving "${web_service_cpp_cmd}"
diff --git a/test_tipc/test_serving_infer_python.sh b/test_tipc/test_serving_infer_python.sh
new file mode 100644
index 0000000..c7d305d
--- /dev/null
+++ b/test_tipc/test_serving_infer_python.sh
@@ -0,0 +1,229 @@
+#!/bin/bash
+source test_tipc/common_func.sh
+
+function func_parser_model_config(){
+    strs=$1
+    IFS="/"
+    array=(${strs})
+    tmp=${array[-1]}
+    echo ${tmp}
+}
+
+FILENAME=$1
+dataline=$(awk 'NR==1, NR==23{print}'  $FILENAME)
+MODE=$2
+
+# parser params
+IFS=$'\n'
+lines=(${dataline})
+
+# parser serving
+model_name=$(func_parser_value "${lines[1]}")
+python_list=$(func_parser_value "${lines[2]}")
+trans_model_py=$(func_parser_value "${lines[3]}")
+det_infer_model_dir_key=$(func_parser_key "${lines[4]}")
+det_infer_model_dir_value=$(func_parser_value "${lines[4]}")
+model_filename_key=$(func_parser_key "${lines[5]}")
+model_filename_value=$(func_parser_value "${lines[5]}")
+params_filename_key=$(func_parser_key "${lines[6]}")
+params_filename_value=$(func_parser_value "${lines[6]}")
+det_serving_server_key=$(func_parser_key "${lines[7]}")
+det_serving_server_value=$(func_parser_value "${lines[7]}")
+det_serving_client_key=$(func_parser_key "${lines[8]}")
+det_serving_client_value=$(func_parser_value "${lines[8]}")
+rec_infer_model_dir_key=$(func_parser_key "${lines[9]}")
+rec_infer_model_dir_value=$(func_parser_value "${lines[9]}")
+rec_serving_server_key=$(func_parser_key "${lines[10]}")
+rec_serving_server_value=$(func_parser_value "${lines[10]}")
+rec_serving_client_key=$(func_parser_key "${lines[11]}")
+rec_serving_client_value=$(func_parser_value "${lines[11]}")
+serving_dir_value=$(func_parser_value "${lines[12]}")
+web_service_py=$(func_parser_value "${lines[13]}")
+web_use_gpu_key=$(func_parser_key "${lines[14]}")
+web_use_gpu_list=$(func_parser_value "${lines[14]}")
+web_use_mkldnn_key=$(func_parser_key "${lines[15]}")
+web_use_mkldnn_list=$(func_parser_value "${lines[15]}")
+web_cpu_threads_key=$(func_parser_key "${lines[16]}")
+web_cpu_threads_list=$(func_parser_value "${lines[16]}")
+web_use_trt_key=$(func_parser_key "${lines[17]}")
+web_use_trt_list=$(func_parser_value "${lines[17]}")
+web_precision_key=$(func_parser_key "${lines[18]}")
+web_precision_list=$(func_parser_value "${lines[18]}")
+det_server_key=$(func_parser_key "${lines[19]}")
+det_server_value=$(func_parser_model_config "${lines[7]}")
+det_client_value=$(func_parser_model_config "${lines[8]}")
+rec_server_key=$(func_parser_key "${lines[20]}")
+rec_server_value=$(func_parser_model_config "${lines[10]}")
+rec_client_value=$(func_parser_model_config "${lines[11]}")
+pipeline_py=$(func_parser_value "${lines[21]}")
+image_dir_key=$(func_parser_key "${lines[22]}")
+image_dir_value=$(func_parser_value "${lines[22]}")
+
+LOG_PATH="$(pwd)/test_tipc/output/${model_name}/${MODE}/python"
+mkdir -p ${LOG_PATH}
+status_log="${LOG_PATH}/results_python_serving.log"
+
+function func_serving(){
+    IFS='|'
+    _python=$1
+    _script=$2
+    _model_dir=$3
+    # pdserving
+    set_model_filename=$(func_set_params "${model_filename_key}" "${model_filename_value}")
+    set_params_filename=$(func_set_params "${params_filename_key}" "${params_filename_value}")
+    if [ ${model_name} = "ch_PP-OCRv2" ] || [ ${model_name} = "ch_PP-OCRv3" ] || [ ${model_name} = "ch_ppocr_mobile_v2_0" ] || [ ${model_name} = "ch_ppocr_server_v2_0" ]; then
+        # trans det
+        set_dirname=$(func_set_params "--dirname" "${det_infer_model_dir_value}")
+        set_serving_server=$(func_set_params "--serving_server" "${det_serving_server_value}")
+        set_serving_client=$(func_set_params "--serving_client" "${det_serving_client_value}")
+        python_list=(${python_list})
+        trans_det_log="${LOG_PATH}/python_trans_model_det.log"
+        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_det_log} 2>&1 "
+        eval $trans_model_cmd
+        # trans rec
+        set_dirname=$(func_set_params "--dirname" "${rec_infer_model_dir_value}")
+        set_serving_server=$(func_set_params "--serving_server" "${rec_serving_server_value}")
+        set_serving_client=$(func_set_params "--serving_client" "${rec_serving_client_value}")
+        python_list=(${python_list})
+        trans_rec_log="${LOG_PATH}/python_trans_model_rec.log"
+        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_rec_log} 2>&1 "
+        eval $trans_model_cmd
+    elif [[ ${model_name} =~ "det" ]]; then
+        # trans det
+        set_dirname=$(func_set_params "--dirname" "${det_infer_model_dir_value}")
+        set_serving_server=$(func_set_params "--serving_server" "${det_serving_server_value}")
+        set_serving_client=$(func_set_params "--serving_client" "${det_serving_client_value}")
+        python_list=(${python_list})
+        trans_det_log="${LOG_PATH}/python_trans_model_det.log"
+        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_det_log} 2>&1 "
+        eval $trans_model_cmd
+    elif [[ ${model_name} =~ "rec" ]]; then
+        # trans rec
+        set_dirname=$(func_set_params "--dirname" "${rec_infer_model_dir_value}")
+        set_serving_server=$(func_set_params "--serving_server" "${rec_serving_server_value}")
+        set_serving_client=$(func_set_params "--serving_client" "${rec_serving_client_value}")
+        python_list=(${python_list})
+        trans_rec_log="${LOG_PATH}/python_trans_model_rec.log"
+        trans_model_cmd="${python_list[0]} ${trans_model_py} ${set_dirname} ${set_model_filename} ${set_params_filename} ${set_serving_server} ${set_serving_client} > ${trans_rec_log} 2>&1 "
+        eval $trans_model_cmd
+    fi
+    set_image_dir=$(func_set_params "${image_dir_key}" "${image_dir_value}")
+    python_list=(${python_list})
+    
+    cd ${serving_dir_value}
+    python=${python_list[0]}
+    
+    # python serving
+    for use_gpu in ${web_use_gpu_list[*]}; do
+        if [ ${use_gpu} = "null" ]; then
+            for use_mkldnn in ${web_use_mkldnn_list[*]}; do
+                for threads in ${web_cpu_threads_list[*]}; do
+                    set_cpu_threads=$(func_set_params "${web_cpu_threads_key}" "${threads}")
+                    server_log_path="${LOG_PATH}/python_server_cpu_usemkldnn_${use_mkldnn}_threads_${threads}.log"
+                    if [ ${model_name} = "ch_PP-OCRv2" ] || [ ${model_name} = "ch_PP-OCRv3" ] || [ ${model_name} = "ch_ppocr_mobile_v2_0" ] || [ ${model_name} = "ch_ppocr_server_v2_0" ]; then
+                        set_det_model_config=$(func_set_params "${det_server_key}" "${det_server_value}")
+                        set_rec_model_config=$(func_set_params "${rec_server_key}" "${rec_server_value}")
+                        web_service_cmd="nohup ${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_det_model_config} ${set_rec_model_config} > ${server_log_path} 2>&1 &"
+                        eval $web_service_cmd
+                        last_status=${PIPESTATUS[0]}
+                        status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}" "${server_log_path}"
+                    elif [[ ${model_name} =~ "det" ]]; then
+                        set_det_model_config=$(func_set_params "${det_server_key}" "${det_server_value}")
+                        web_service_cmd="nohup ${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_det_model_config} > ${server_log_path} 2>&1 &"
+                        eval $web_service_cmd
+                        last_status=${PIPESTATUS[0]}
+                        status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}" "${server_log_path}"
+                    elif [[ ${model_name} =~ "rec" ]]; then
+                        set_rec_model_config=$(func_set_params "${rec_server_key}" "${rec_server_value}")
+                        web_service_cmd="nohup ${python} ${web_service_py} ${web_use_gpu_key}="" ${web_use_mkldnn_key}=${use_mkldnn} ${set_cpu_threads} ${set_rec_model_config} > ${server_log_path} 2>&1 &"
+                        eval $web_service_cmd
+                        last_status=${PIPESTATUS[0]}
+                        status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}" "${server_log_path}"
+                    fi
+                    sleep 2s
+                    for pipeline in ${pipeline_py[*]}; do
+                        _save_log_path="${LOG_PATH}/python_client_cpu_${pipeline%_client*}_usemkldnn_${use_mkldnn}_threads_${threads}_batchsize_1.log"
+                        pipeline_cmd="${python} ${pipeline} ${set_image_dir} > ${_save_log_path} 2>&1 "
+                        eval $pipeline_cmd
+                        last_status=${PIPESTATUS[0]}
+                        eval "cat ${_save_log_path}"
+                        status_check $last_status "${pipeline_cmd}" "${status_log}" "${model_name}" "${_save_log_path}"
+                        sleep 2s
+                    done
+                    ps ux | grep -E 'web_service' | awk '{print $2}' | xargs kill -s 9
+                done
+            done
+        elif [ ${use_gpu} = "gpu" ]; then
+            for use_trt in ${web_use_trt_list[*]}; do
+                for precision in ${web_precision_list[*]}; do
+                    server_log_path="${LOG_PATH}/python_server_gpu_usetrt_${use_trt}_precision_${precision}.log"
+                    if [[ ${_flag_quant} = "False" ]] && [[ ${precision} =~ "int8" ]]; then
+                        continue
+                    fi
+                    if [[ ${precision} =~ "fp16" || ${precision} =~ "int8" ]] && [ ${use_trt} = "False" ]; then
+                        continue
+                    fi
+                    if [[ ${use_trt} = "False" || ${precision} =~ "int8" ]] && [[ ${_flag_quant} = "True" ]]; then
+                        continue
+                    fi
+                    set_tensorrt=$(func_set_params "${web_use_trt_key}" "${use_trt}")
+                    if [ ${use_trt} = True ]; then
+                        device_type=2
+                    fi
+                    set_precision=$(func_set_params "${web_precision_key}" "${precision}")
+                    if [ ${model_name} = "ch_PP-OCRv2" ] || [ ${model_name} = "ch_PP-OCRv3" ] || [ ${model_name} = "ch_ppocr_mobile_v2_0" ] || [ ${model_name} = "ch_ppocr_server_v2_0" ]; then
+                        set_det_model_config=$(func_set_params "${det_server_key}" "${det_server_value}")
+                        set_rec_model_config=$(func_set_params "${rec_server_key}" "${rec_server_value}")
+                        web_service_cmd="nohup ${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_det_model_config} ${set_rec_model_config} > ${server_log_path} 2>&1 &"
+                        eval $web_service_cmd
+                        last_status=${PIPESTATUS[0]}
+                        status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}" "${server_log_path}"
+                    elif [[ ${model_name} =~ "det" ]]; then
+                        set_det_model_config=$(func_set_params "${det_server_key}" "${det_server_value}")
+                        web_service_cmd="nohup ${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_det_model_config} > ${server_log_path} 2>&1 &"
+                        eval $web_service_cmd
+                        last_status=${PIPESTATUS[0]}
+                        status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}" "${server_log_path}"
+                    elif [[ ${model_name} =~ "rec" ]]; then
+                        set_rec_model_config=$(func_set_params "${rec_server_key}" "${rec_server_value}")
+                        web_service_cmd="nohup ${python} ${web_service_py} ${set_tensorrt} ${set_precision} ${set_rec_model_config} > ${server_log_path} 2>&1 &"
+                        eval $web_service_cmd
+                        last_status=${PIPESTATUS[0]}
+                        status_check $last_status "${web_service_cmd}" "${status_log}" "${model_name}" "${server_log_path}"
+                    fi
+                    sleep 2s
+                    for pipeline in ${pipeline_py[*]}; do
+                        _save_log_path="${LOG_PATH}/python_client_gpu_${pipeline%_client*}_usetrt_${use_trt}_precision_${precision}_batchsize_1.log"
+                        pipeline_cmd="${python} ${pipeline} ${set_image_dir}> ${_save_log_path} 2>&1"
+                        eval $pipeline_cmd
+                        last_status=${PIPESTATUS[0]}
+                        eval "cat ${_save_log_path}"
+                        status_check $last_status "${pipeline_cmd}" "${status_log}" "${model_name}" "${_save_log_path}"
+                        sleep 2s
+                    done
+                    ps ux | grep -E 'web_service' | awk '{print $2}' | xargs kill -s 9
+                done
+            done
+        else
+            echo "Does not support hardware other than CPU and GPU Currently!"
+        fi
+    done
+}
+
+
+#set cuda device
+GPUID=$3
+if [ ${#GPUID} -le 0 ];then
+    env="export CUDA_VISIBLE_DEVICES=0"
+else
+    env="export CUDA_VISIBLE_DEVICES=${GPUID}"
+fi
+eval $env
+echo $env
+
+
+echo "################### run test ###################"
+
+export Count=0
+IFS="|"
+func_serving "${web_service_cmd}"
diff --git a/test_tipc/test_train_inference_python.sh b/test_tipc/test_train_inference_python.sh
new file mode 100644
index 0000000..95a6bd8
--- /dev/null
+++ b/test_tipc/test_train_inference_python.sh
@@ -0,0 +1,343 @@
+#!/bin/bash
+source test_tipc/common_func.sh
+
+FILENAME=$1
+# MODE be one of ['lite_train_lite_infer' 'lite_train_whole_infer' 'whole_train_whole_infer', 'whole_infer']
+MODE=$2
+
+dataline=$(awk 'NR>=1{print}'  $FILENAME)
+
+# parser params
+IFS=$'\n'
+lines=(${dataline})
+
+# The training params
+model_name=$(func_parser_value "${lines[1]}")
+python=$(func_parser_value "${lines[2]}")
+gpu_list=$(func_parser_value "${lines[3]}")
+train_use_gpu_key=$(func_parser_key "${lines[4]}")
+train_use_gpu_value=$(func_parser_value "${lines[4]}")
+autocast_list=$(func_parser_value "${lines[5]}")
+autocast_key=$(func_parser_key "${lines[5]}")
+epoch_key=$(func_parser_key "${lines[6]}")
+epoch_num=$(func_parser_params "${lines[6]}" "${MODE}")
+save_model_key=$(func_parser_key "${lines[7]}")
+train_batch_key=$(func_parser_key "${lines[8]}")
+train_batch_value=$(func_parser_params "${lines[8]}" "${MODE}")
+pretrain_model_key=$(func_parser_key "${lines[9]}")
+pretrain_model_value=$(func_parser_value "${lines[9]}")
+train_model_name=$(func_parser_value "${lines[10]}")
+train_infer_img_dir=$(func_parser_value "${lines[11]}")
+train_param_key1=$(func_parser_key "${lines[12]}")
+train_param_value1=$(func_parser_value "${lines[12]}")
+
+trainer_list=$(func_parser_value "${lines[14]}")
+trainer_norm=$(func_parser_key "${lines[15]}")
+norm_trainer=$(func_parser_value "${lines[15]}")
+pact_key=$(func_parser_key "${lines[16]}")
+pact_trainer=$(func_parser_value "${lines[16]}")
+fpgm_key=$(func_parser_key "${lines[17]}")
+fpgm_trainer=$(func_parser_value "${lines[17]}")
+distill_key=$(func_parser_key "${lines[18]}")
+distill_trainer=$(func_parser_value "${lines[18]}")
+trainer_key1=$(func_parser_key "${lines[19]}")
+trainer_value1=$(func_parser_value "${lines[19]}")
+trainer_key2=$(func_parser_key "${lines[20]}")
+trainer_value2=$(func_parser_value "${lines[20]}")
+
+eval_py=$(func_parser_value "${lines[23]}")
+eval_key1=$(func_parser_key "${lines[24]}")
+eval_value1=$(func_parser_value "${lines[24]}")
+
+save_infer_key=$(func_parser_key "${lines[27]}")
+export_weight=$(func_parser_key "${lines[28]}")
+norm_export=$(func_parser_value "${lines[29]}")
+pact_export=$(func_parser_value "${lines[30]}")
+fpgm_export=$(func_parser_value "${lines[31]}")
+distill_export=$(func_parser_value "${lines[32]}")
+export_key1=$(func_parser_key "${lines[33]}")
+export_value1=$(func_parser_value "${lines[33]}")
+export_key2=$(func_parser_key "${lines[34]}")
+export_value2=$(func_parser_value "${lines[34]}")
+inference_dir=$(func_parser_value "${lines[35]}")
+
+# parser inference model 
+infer_model_dir_list=$(func_parser_value "${lines[36]}")
+infer_export_list=$(func_parser_value "${lines[37]}")
+infer_is_quant=$(func_parser_value "${lines[38]}")
+# parser inference 
+inference_py=$(func_parser_value "${lines[39]}")
+use_gpu_key=$(func_parser_key "${lines[40]}")
+use_gpu_list=$(func_parser_value "${lines[40]}")
+use_mkldnn_key=$(func_parser_key "${lines[41]}")
+use_mkldnn_list=$(func_parser_value "${lines[41]}")
+cpu_threads_key=$(func_parser_key "${lines[42]}")
+cpu_threads_list=$(func_parser_value "${lines[42]}")
+batch_size_key=$(func_parser_key "${lines[43]}")
+batch_size_list=$(func_parser_value "${lines[43]}")
+use_trt_key=$(func_parser_key "${lines[44]}")
+use_trt_list=$(func_parser_value "${lines[44]}")
+precision_key=$(func_parser_key "${lines[45]}")
+precision_list=$(func_parser_value "${lines[45]}")
+infer_model_key=$(func_parser_key "${lines[46]}")
+image_dir_key=$(func_parser_key "${lines[47]}")
+infer_img_dir=$(func_parser_value "${lines[47]}")
+save_log_key=$(func_parser_key "${lines[48]}")
+benchmark_key=$(func_parser_key "${lines[49]}")
+benchmark_value=$(func_parser_value "${lines[49]}")
+infer_key1=$(func_parser_key "${lines[50]}")
+infer_value1=$(func_parser_value "${lines[50]}")
+
+line_num=`grep -n -w "to_static_train_benchmark_params" $FILENAME  | cut -d ":" -f 1`
+to_static_key=$(func_parser_key "${lines[line_num]}")
+to_static_trainer=$(func_parser_value "${lines[line_num]}")
+
+LOG_PATH="./test_tipc/output/${model_name}/${MODE}"
+mkdir -p ${LOG_PATH}
+status_log="${LOG_PATH}/results_python.log"
+
+function func_inference(){
+    IFS='|'
+    _python=$1
+    _script=$2
+    _model_dir=$3
+    _log_path=$4
+    _img_dir=$5
+    _flag_quant=$6
+    _gpu=$7
+    # inference 
+    for use_gpu in ${use_gpu_list[*]}; do
+        if [ ${use_gpu} = "False" ] || [ ${use_gpu} = "cpu" ]; then
+            for use_mkldnn in ${use_mkldnn_list[*]}; do
+                # if [ ${use_mkldnn} = "False" ] && [ ${_flag_quant} = "True" ]; then
+                #     continue
+                # fi
+                for threads in ${cpu_threads_list[*]}; do
+                    for batch_size in ${batch_size_list[*]}; do
+                        for precision in ${precision_list[*]}; do
+                            if [ ${use_mkldnn} = "False" ] && [ ${precision} = "fp16" ]; then
+                                continue
+                            fi # skip when enable fp16 but disable mkldnn
+                            if [ ${_flag_quant} = "True" ] && [ ${precision} != "int8" ]; then
+                                continue
+                            fi # skip when quant model inference but precision is not int8
+                            set_precision=$(func_set_params "${precision_key}" "${precision}")
+                            
+                            _save_log_path="${_log_path}/python_infer_cpu_gpus_${_gpu}_usemkldnn_${use_mkldnn}_threads_${threads}_precision_${precision}_batchsize_${batch_size}.log"
+                            set_infer_data=$(func_set_params "${image_dir_key}" "${_img_dir}")
+                            set_benchmark=$(func_set_params "${benchmark_key}" "${benchmark_value}")
+                            set_batchsize=$(func_set_params "${batch_size_key}" "${batch_size}")
+                            set_mkldnn=$(func_set_params "${use_mkldnn_key}" "${use_mkldnn}")
+                            set_cpu_threads=$(func_set_params "${cpu_threads_key}" "${threads}")
+                            set_model_dir=$(func_set_params "${infer_model_key}" "${_model_dir}")
+                            set_infer_params0=$(func_set_params "${save_log_key}" "${save_log_value}")
+                            set_infer_params1=$(func_set_params "${infer_key1}" "${infer_value1}")
+                            command="${_python} ${_script} ${use_gpu_key}=${use_gpu} ${set_mkldnn} ${set_cpu_threads} ${set_model_dir} ${set_batchsize} ${set_infer_params0} ${set_infer_data} ${set_benchmark} ${set_precision} ${set_infer_params1} > ${_save_log_path} 2>&1 "
+                            eval $command
+                            last_status=${PIPESTATUS[0]}
+                            eval "cat ${_save_log_path}"
+                            status_check $last_status "${command}" "${status_log}" "${model_name}" "${_save_log_path}"
+                        done
+                    done
+                done
+            done
+        elif [ ${use_gpu} = "True" ] || [ ${use_gpu} = "gpu" ]; then
+            for use_trt in ${use_trt_list[*]}; do
+                for precision in ${precision_list[*]}; do
+                    if [[ ${_flag_quant} = "False" ]] && [[ ${precision} =~ "int8" ]]; then
+                        continue
+                    fi 
+                    if [[ ${precision} =~ "fp16" || ${precision} =~ "int8" ]] && [ ${use_trt} = "False" ]; then
+                        continue
+                    fi
+                    if [[ ${use_trt} = "False" && ${precision} =~ "int8" ]] && [ ${_flag_quant} = "True" ]; then
+                        continue
+                    fi
+                    for batch_size in ${batch_size_list[*]}; do
+                        _save_log_path="${_log_path}/python_infer_gpu_gpus_${_gpu}_usetrt_${use_trt}_precision_${precision}_batchsize_${batch_size}.log"
+                        set_infer_data=$(func_set_params "${image_dir_key}" "${_img_dir}")
+                        set_benchmark=$(func_set_params "${benchmark_key}" "${benchmark_value}")
+                        set_batchsize=$(func_set_params "${batch_size_key}" "${batch_size}")
+                        set_tensorrt=$(func_set_params "${use_trt_key}" "${use_trt}")
+                        set_precision=$(func_set_params "${precision_key}" "${precision}")
+                        set_model_dir=$(func_set_params "${infer_model_key}" "${_model_dir}")
+                        set_infer_params0=$(func_set_params "${save_log_key}" "${save_log_value}")
+                        set_infer_params1=$(func_set_params "${infer_key1}" "${infer_value1}")
+                        command="${_python} ${_script} ${use_gpu_key}=${use_gpu} ${set_tensorrt} ${set_precision} ${set_model_dir} ${set_batchsize} ${set_infer_data} ${set_benchmark} ${set_infer_params1} ${set_infer_params0} > ${_save_log_path} 2>&1 "
+                        eval $command
+                        last_status=${PIPESTATUS[0]}
+                        eval "cat ${_save_log_path}"
+                        status_check $last_status "${command}" "${status_log}" "${model_name}" "${_save_log_path}"
+                        
+                    done
+                done
+            done
+        else
+            echo "Does not support hardware other than CPU and GPU Currently!"
+        fi
+    done
+}
+
+if [ ${MODE} = "whole_infer" ]; then
+    GPUID=$3
+    if [ ${#GPUID} -le 0 ];then
+        env=" "
+    else
+        env="export CUDA_VISIBLE_DEVICES=${GPUID}"
+    fi
+    # set CUDA_VISIBLE_DEVICES
+    eval $env
+    export Count=0
+    gpu=0
+    IFS="|"
+    infer_run_exports=(${infer_export_list})
+    infer_quant_flag=(${infer_is_quant})
+    for infer_model in ${infer_model_dir_list[*]}; do
+        # run export
+        if [ ${infer_run_exports[Count]} != "null" ];then 
+            save_infer_dir="${infer_model}"
+            set_export_weight=$(func_set_params "${export_weight}" "${infer_model}")
+            set_save_infer_key=$(func_set_params "${save_infer_key}" "${save_infer_dir}")
+            export_log_path="${LOG_PATH}_export_${Count}.log"
+            export_cmd="${python} ${infer_run_exports[Count]} ${set_export_weight} ${set_save_infer_key} > ${export_log_path} 2>&1 "
+            echo ${infer_run_exports[Count]} 
+            echo $export_cmd
+            eval $export_cmd
+            status_export=$?
+            status_check $status_export "${export_cmd}" "${status_log}" "${model_name}" "${export_log_path}"
+        else
+            save_infer_dir=${infer_model}
+        fi
+        #run inference
+        is_quant=${infer_quant_flag[Count]}
+        func_inference "${python}" "${inference_py}" "${save_infer_dir}" "${LOG_PATH}" "${infer_img_dir}" ${is_quant} "${gpu}"
+        Count=$(($Count + 1))
+    done
+else
+    IFS="|"
+    export Count=0
+    USE_GPU_KEY=(${train_use_gpu_value})
+    for gpu in ${gpu_list[*]}; do
+        train_use_gpu=${USE_GPU_KEY[Count]}
+        Count=$(($Count + 1))
+        ips=""
+        if [ ${gpu} = "-1" ];then
+            env=""
+        elif [ ${#gpu} -le 1 ];then
+            env="export CUDA_VISIBLE_DEVICES=${gpu}"
+        elif [ ${#gpu} -le 15 ];then
+            IFS=","
+            array=(${gpu})
+            env="export CUDA_VISIBLE_DEVICES=${array[0]}"
+            IFS="|"
+        else
+            IFS=";"
+            array=(${gpu})
+            ips=${array[0]}
+            gpu=${array[1]}
+            IFS="|"
+            env=" "
+        fi
+        for autocast in ${autocast_list[*]}; do 
+            if [ ${autocast} = "amp" ]; then
+                set_amp_config="Global.use_amp=True Global.scale_loss=1024.0 Global.use_dynamic_loss_scaling=True"
+            else
+                set_amp_config=" "
+            fi          
+            for trainer in ${trainer_list[*]}; do 
+                flag_quant=False
+                if [ ${trainer} = ${pact_key} ]; then
+                    run_train=${pact_trainer}
+                    run_export=${pact_export}
+                    flag_quant=True
+                elif [ ${trainer} = "${fpgm_key}" ]; then
+                    run_train=${fpgm_trainer}
+                    run_export=${fpgm_export}
+                elif [ ${trainer} = "${distill_key}" ]; then
+                    run_train=${distill_trainer}
+                    run_export=${distill_export}
+                elif [ ${trainer} = "${to_static_key}" ]; then
+                    run_train="${norm_trainer}  ${to_static_trainer}"
+                    run_export=${norm_export}
+                elif [[ ${trainer} = ${trainer_key2} ]]; then
+                    run_train=${trainer_value2}
+                    run_export=${export_value2}
+                else
+                    run_train=${norm_trainer}
+                    run_export=${norm_export}
+                fi
+
+                if [ ${run_train} = "null" ]; then
+                    continue
+                fi
+
+                set_epoch=$(func_set_params "${epoch_key}" "${epoch_num}")
+                set_pretrain=$(func_set_params "${pretrain_model_key}" "${pretrain_model_value}")
+                set_batchsize=$(func_set_params "${train_batch_key}" "${train_batch_value}")
+                set_train_params1=$(func_set_params "${train_param_key1}" "${train_param_value1}")
+                set_use_gpu=$(func_set_params "${train_use_gpu_key}" "${train_use_gpu}")
+                # if length of ips >= 15, then it is seen as multi-machine
+                # 15 is the min length of ips info for multi-machine: 0.0.0.0,0.0.0.0
+                if [ ${#ips} -le 15 ];then
+                    save_log="${LOG_PATH}/${trainer}_gpus_${gpu}_autocast_${autocast}"
+                    nodes=1
+                else
+                    IFS=","
+                    ips_array=(${ips})
+                    IFS="|"
+                    nodes=${#ips_array[@]}
+                    save_log="${LOG_PATH}/${trainer}_gpus_${gpu}_autocast_${autocast}_nodes_${nodes}"
+                fi
+
+
+                set_save_model=$(func_set_params "${save_model_key}" "${save_log}")
+                if [ ${#gpu} -le 2 ];then  # train with cpu or single gpu
+                    cmd="${python} ${run_train} ${set_use_gpu}  ${set_save_model} ${set_epoch} ${set_pretrain}  ${set_batchsize} ${set_amp_config} ${set_train_params1}"
+                elif [ ${#ips} -le 15 ];then  # train with multi-gpu
+                    cmd="${python} -m paddle.distributed.launch --gpus=${gpu} ${run_train} ${set_use_gpu} ${set_save_model} ${set_epoch} ${set_pretrain}  ${set_batchsize} ${set_amp_config} ${set_train_params1}"
+                else     # train with multi-machine
+                    cmd="${python} -m paddle.distributed.launch --ips=${ips} --gpus=${gpu} ${run_train} ${set_use_gpu} ${set_save_model} ${set_pretrain} ${set_epoch}  ${set_batchsize} ${set_amp_config} ${set_train_params1}"
+                fi
+                # run train
+                eval $cmd
+                eval "cat ${save_log}/train.log >> ${save_log}.log"
+                status_check $? "${cmd}" "${status_log}" "${model_name}" "${save_log}.log"
+
+                set_eval_pretrain=$(func_set_params "${pretrain_model_key}" "${save_log}/${train_model_name}")
+
+                # run eval 
+                if [ ${eval_py} != "null" ]; then
+                    eval ${env}
+                    set_eval_params1=$(func_set_params "${eval_key1}" "${eval_value1}")
+                    eval_log_path="${LOG_PATH}/${trainer}_gpus_${gpu}_autocast_${autocast}_nodes_${nodes}_eval.log"
+                    eval_cmd="${python} ${eval_py} ${set_eval_pretrain} ${set_use_gpu} ${set_eval_params1} > ${eval_log_path} 2>&1 " 
+                    eval $eval_cmd
+                    status_check $? "${eval_cmd}" "${status_log}" "${model_name}" "${eval_log_path}"
+                fi
+                # run export model
+                if [ ${run_export} != "null" ]; then 
+                    # run export model
+                    save_infer_path="${save_log}"
+                    export_log_path="${LOG_PATH}/${trainer}_gpus_${gpu}_autocast_${autocast}_nodes_${nodes}_export.log"
+                    set_export_weight=$(func_set_params "${export_weight}" "${save_log}/${train_model_name}")
+                    set_save_infer_key=$(func_set_params "${save_infer_key}" "${save_infer_path}")
+                    export_cmd="${python} ${run_export} ${set_export_weight} ${set_save_infer_key} > ${export_log_path} 2>&1 "
+                    eval $export_cmd
+                    status_check $? "${export_cmd}" "${status_log}" "${model_name}" "${export_log_path}"
+
+                    #run inference
+                    eval $env
+                    save_infer_path="${save_log}"
+                    if [[ ${inference_dir} != "null" ]] && [[ ${inference_dir} != '##' ]]; then
+                        infer_model_dir="${save_infer_path}/${inference_dir}"
+                    else
+                        infer_model_dir=${save_infer_path}
+                    fi
+                    func_inference "${python}" "${inference_py}" "${infer_model_dir}" "${LOG_PATH}" "${train_infer_img_dir}" "${flag_quant}" "${gpu}"
+                    
+                    eval "unset CUDA_VISIBLE_DEVICES"
+                fi
+            done  # done with:    for trainer in ${trainer_list[*]}; do 
+        done      # done with:    for autocast in ${autocast_list[*]}; do 
+    done          # done with:    for gpu in ${gpu_list[*]}; do
+fi  # end if [ ${MODE} = "infer" ]; then
diff --git a/test_tipc/test_train_inference_python_npu.sh b/test_tipc/test_train_inference_python_npu.sh
new file mode 100644
index 0000000..4b122b0
--- /dev/null
+++ b/test_tipc/test_train_inference_python_npu.sh
@@ -0,0 +1,66 @@
+#!/bin/bash
+source test_tipc/common_func.sh
+
+function readlinkf() {
+    perl -MCwd -e 'print Cwd::abs_path shift' "$1";
+}
+
+function func_parser_config() {
+    strs=$1
+    IFS=" "
+    array=(${strs})
+    tmp=${array[2]}
+    echo ${tmp}
+}
+
+BASEDIR=$(dirname "$0")
+REPO_ROOT_PATH=$(readlinkf ${BASEDIR}/../)
+
+FILENAME=$1
+
+# disable mkldnn on non x86_64 env
+arch=$(uname -i)
+if [ $arch != 'x86_64' ]; then
+    sed -i 's/--enable_mkldnn:True|False/--enable_mkldnn:False/g' $FILENAME
+    sed -i 's/--enable_mkldnn:True/--enable_mkldnn:False/g' $FILENAME
+fi
+
+# change gpu to npu in tipc txt configs
+sed -i 's/use_gpu/use_npu/g' $FILENAME
+# disable benchmark as AutoLog required nvidia-smi command
+sed -i 's/--benchmark:True/--benchmark:False/g' $FILENAME
+# python has been updated to version 3.9 for npu backend
+sed -i "s/python3.7/python3.9/g" $FILENAME
+dataline=`cat $FILENAME`
+
+# parser params
+IFS=$'\n'
+lines=(${dataline})
+
+modelname=$(echo ${lines[1]} | cut -d ":" -f2)
+if  [ $modelname == "rec_r31_sar" ] || [ $modelname == "rec_mtb_nrtr" ]; then
+    sed -i "s/Global.epoch_num:lite_train_lite_infer=2/Global.epoch_num:lite_train_lite_infer=1/g" $FILENAME
+    sed -i "s/gpu_list:0|0,1/gpu_list:0,1/g" $FILENAME
+    sed -i "s/Global.use_npu:True|True/Global.use_npu:True/g" $FILENAME
+fi
+if [ $modelname == "ch_ppocr_mobile_v2_0_rec_FPGM" ]; then
+    sed -i '18s/$/ -o Global.use_gpu=False/' $FILENAME 
+    sed -i '32s/$/ Global.use_gpu=False/' $FILENAME
+fi
+
+# replace training config file
+grep -n 'tools/.*yml' $FILENAME  | cut -d ":" -f 1 \
+| while read line_num ; do
+    train_cmd=$(func_parser_value "${lines[line_num-1]}")
+    trainer_config=$(func_parser_config ${train_cmd})
+    sed -i 's/use_gpu/use_npu/g' "$REPO_ROOT_PATH/$trainer_config"
+    sed -i 's/use_sync_bn: True/use_sync_bn: False/g' "$REPO_ROOT_PATH/$trainer_config"
+done
+
+# change gpu to npu in execution script
+sed -i 's/\"gpu\"/\"npu\"/g' test_tipc/test_train_inference_python.sh
+
+# pass parameters to test_train_inference_python.sh
+cmd='bash test_tipc/test_train_inference_python.sh ${FILENAME} $2'
+echo -e '\033[1;32m Started to run command: ${cmd}!  \033[0m'
+eval $cmd
diff --git a/test_tipc/test_train_inference_python_xpu.sh b/test_tipc/test_train_inference_python_xpu.sh
new file mode 100644
index 0000000..88015e4
--- /dev/null
+++ b/test_tipc/test_train_inference_python_xpu.sh
@@ -0,0 +1,66 @@
+#!/bin/bash
+source test_tipc/common_func.sh
+
+function readlinkf() {
+    perl -MCwd -e 'print Cwd::abs_path shift' "$1";
+}
+
+function func_parser_config() {
+    strs=$1
+    IFS=" "
+    array=(${strs})
+    tmp=${array[2]}
+    echo ${tmp}
+}
+
+BASEDIR=$(dirname "$0")
+REPO_ROOT_PATH=$(readlinkf ${BASEDIR}/../)
+
+FILENAME=$1
+
+# disable mkldnn on non x86_64 env
+arch=$(uname -i)
+if [ $arch != 'x86_64' ]; then
+    sed -i 's/--enable_mkldnn:True|False/--enable_mkldnn:False/g' $FILENAME
+    sed -i 's/--enable_mkldnn:True/--enable_mkldnn:False/g' $FILENAME
+fi
+
+# change gpu to xpu in tipc txt configs
+sed -i 's/use_gpu/use_xpu/g' $FILENAME
+# disable benchmark as AutoLog required nvidia-smi command
+sed -i 's/--benchmark:True/--benchmark:False/g' $FILENAME
+# python has been updated to version 3.9 for xpu backend
+sed -i "s/python3.7/python3.9/g" $FILENAME
+dataline=`cat $FILENAME`
+
+# parser params
+IFS=$'\n'
+lines=(${dataline})
+
+modelname=$(echo ${lines[1]} | cut -d ":" -f2)
+if  [ $modelname == "rec_r31_sar" ] || [ $modelname == "rec_mtb_nrtr" ]; then
+    sed -i "s/Global.epoch_num:lite_train_lite_infer=2/Global.epoch_num:lite_train_lite_infer=1/g" $FILENAME
+    sed -i "s/gpu_list:0|0,1/gpu_list:0,1/g" $FILENAME
+    sed -i "s/Global.use_xpu:True|True/Global.use_xpu:True/g" $FILENAME
+fi
+if [ $modelname == "ch_ppocr_mobile_v2_0_rec_FPGM" ]; then
+    sed -i '18s/$/ -o Global.use_gpu=False/' $FILENAME 
+    sed -i '32s/$/ Global.use_gpu=False/' $FILENAME
+fi
+
+# replace training config file
+grep -n 'tools/.*yml' $FILENAME  | cut -d ":" -f 1 \
+| while read line_num ; do
+    train_cmd=$(func_parser_value "${lines[line_num-1]}")
+    trainer_config=$(func_parser_config ${train_cmd})
+    sed -i 's/use_gpu/use_xpu/g' "$REPO_ROOT_PATH/$trainer_config"
+    sed -i 's/use_sync_bn: True/use_sync_bn: False/g' "$REPO_ROOT_PATH/$trainer_config"
+done
+
+# change gpu to xpu in execution script
+sed -i 's/\"gpu\"/\"xpu\"/g' test_tipc/test_train_inference_python.sh
+
+# pass parameters to test_train_inference_python.sh
+cmd='bash test_tipc/test_train_inference_python.sh ${FILENAME} $2'
+echo -e '\033[1;32m Started to run command: ${cmd}!  \033[0m'
+eval $cmd
diff --git a/test_tipc/web/expect.json b/test_tipc/web/expect.json
new file mode 100644
index 0000000..a60c80a
--- /dev/null
+++ b/test_tipc/web/expect.json
@@ -0,0 +1,20 @@
+{
+    "text":  [
+        "纯臻营养护发素",
+        "产品信息/参数",
+        "（45元/每公斤，100公斤起订）",
+        "每瓶22元，1000瓶起订）",
+        "【品牌】：代加工方式/OEMODM",
+        "【品名】：纯臻营养护发素",
+        "【产品编号】：YM-X-3011",
+        "ODMOEM",
+        "【净含量】：220ml",
+        "【适用人群】：适合所有肤质",
+        "【主要成分】：鲸蜡硬脂醇、燕麦β-葡聚",
+        "糖、椰油酰胺丙基甜菜碱、泛醌",
+        "（成品包材）",
+        "【主要功能】：可紧致头发磷层，从而达到",
+        "即时持久改善头发光泽的效果，给干燥的头",
+        "发足够的滋养"
+    ]
+}
diff --git a/test_tipc/web/index.html b/test_tipc/web/index.html
new file mode 100644
index 0000000..2111155
--- /dev/null
+++ b/test_tipc/web/index.html
@@ -0,0 +1,13 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <meta http-equiv="X-UA-Compatible" content="ie=edge">
+    <title>ocr test</title>
+</head>
+<body>
+    <img id="ocr" src="./test.jpg" />
+</body>
+<script src="./node_modules/@paddlejs-models/ocr/lib/index.js"></script>
+</html>
diff --git a/test_tipc/web/index.test.js b/test_tipc/web/index.test.js
new file mode 100644
index 0000000..e07aed8
--- /dev/null
+++ b/test_tipc/web/index.test.js
@@ -0,0 +1,82 @@
+const expectData = require('./expect.json');
+
+describe('e2e test ocr model', () => {
+
+    beforeAll(async () => {
+        await page.goto(PATH);
+    });
+
+    it('ocr infer and diff test', async () => {
+        page.on('console', msg => console.log('PAGE LOG:', msg.text()));
+
+        const text = await page.evaluate(async () => {
+            const $ocr = document.querySelector('#ocr');
+            const ocr = paddlejs['ocr'];
+            await ocr.init('./models/ch_PP-OCRv2_det_infer', './models/ch_PP-OCRv2_rec_infer');
+            const res = await ocr.recognize($ocr);
+            return res.text;
+        });
+        // 模型文字识别结果与预期结果diff的字符数
+        let diffNum = 0;
+        // 文本框字符串相似度
+        let similarity = 0;
+        // 预期字符diff数
+        const expectedDiffNum = 10;
+        // 预期文本框字符串相似度
+        const expecteSimilarity = 0.9;
+        // 预期文本内容
+        const expectResult = expectData.text;
+
+        expectResult && expectResult.forEach((item, index) => {
+            const word = text[index];
+            // 逐字符对比
+            for(let i = 0; i < item.length; i++) {
+                if (item[i] !== word[i]) {
+                    console.log('expect: ', item[i], ' word: ', word[i]);
+                    diffNum++;
+                }
+            }
+            // 文本框字符串相似度对比
+            const s = similar(item, word);
+            similarity += s;
+        });
+
+        similarity = similarity / expectResult.length;
+
+        expect(diffNum).toBeLessThanOrEqual(expectedDiffNum);
+
+        expect(similarity).toBeGreaterThanOrEqual(expecteSimilarity);
+
+        function similar(string, expect) {
+            if (!string || !expect) {
+                return 0;
+            }
+            const length = string.length > expect.length ? string.length : expect.length;
+            const n = string.length;
+            const m = expect.length;
+            let data = [];
+            const min = (a, b, c) => {
+                return a < b ? (a < c ? a : c) : (b < c ? b : c);
+            };
+            let i, j, si, ej, cost;
+            if (n === 0) return m;
+            if (m === 0) return n;
+            for (i = 0; i <= n; i++) {
+                data[i] = [];
+                [i][0] = i
+            }
+            for (j = 0; j <= m; j++) {
+                data[0][j] = j;
+            }
+            for (i = 1; i <= n; i++) {
+                si = string.charAt(i - 1);
+                for (j = 1; j <= m; j++) {
+                    ej = expect.charAt(j - 1);
+                    cost = si === ej ? 0 : 1;
+                    data[i][j] = min(data[i - 1][j] + 1, data[i][j - 1] + 1, data[i - 1][j - 1] + cost);
+                }
+            }
+            return (1 - data[n][m] / length);
+        }
+    });
+});
diff --git a/test_tipc/web/jest-puppeteer.config.js b/test_tipc/web/jest-puppeteer.config.js
new file mode 100644
index 0000000..ac60eea
--- /dev/null
+++ b/test_tipc/web/jest-puppeteer.config.js
@@ -0,0 +1,14 @@
+// jest-puppeteer.config.js
+module.exports = {
+    launch: {
+        headless: false,
+        product: 'chrome'
+    },
+    browserContext: 'default',
+    server: {
+        command: 'python3 -m http.server 9811',
+        port: 9811,
+        launchTimeout: 10000,
+        debug: true
+    }
+};
diff --git a/test_tipc/web/jest.config.js b/test_tipc/web/jest.config.js
new file mode 100644
index 0000000..aed1573
--- /dev/null
+++ b/test_tipc/web/jest.config.js
@@ -0,0 +1,111 @@
+// For a detailed explanation regarding each configuration property and type check, visit:
+// https://jestjs.io/docs/en/configuration.html
+
+module.exports = {
+    preset: 'jest-puppeteer',
+    // All imported modules in your tests should be mocked automatically
+    // automock: false,
+
+    // Automatically clear mock calls and instances between every test
+    clearMocks: true,
+
+    // An object that configures minimum threshold enforcement for coverage results
+    // coverageThreshold: undefined,
+
+    // A set of global variables that need to be available in all test environments
+    globals: {
+        PATH: 'http://localhost:9811'
+    },
+
+    // The maximum amount of workers used to run your tests. Can be specified as % or a number. E.g. maxWorkers: 10% will use 10% of your CPU amount + 1 as the maximum worker number. maxWorkers: 2 will use a maximum of 2 workers.
+    // maxWorkers: "50%",
+
+    // An array of directory names to be searched recursively up from the requiring module's location
+    // moduleDirectories: [
+    //   "node_modules"
+    // ],
+
+    // An array of file extensions your modules use
+    moduleFileExtensions: [
+        'js',
+        'json',
+        'jsx',
+        'ts',
+        'tsx',
+        'node'
+    ],
+
+
+    // The root directory that Jest should scan for tests and modules within
+    // rootDir: undefined,
+
+    // A list of paths to directories that Jest should use to search for files in
+    roots: [
+        '<rootDir>'
+    ],
+
+    // Allows you to use a custom runner instead of Jest's default test runner
+    // runner: "jest-runner",
+
+    // The paths to modules that run some code to configure or set up the testing environment before each test
+    // setupFiles: [],
+
+    // A list of paths to modules that run some code to configure or set up the testing framework before each test
+    // setupFilesAfterEnv: [],
+
+    // The number of seconds after which a test is considered as slow and reported as such in the results.
+    // slowTestThreshold: 5,
+
+    // A list of paths to snapshot serializer modules Jest should use for snapshot testing
+    // snapshotSerializers: [],
+
+    // The test environment that will be used for testing
+    // testEnvironment: 'jsdom',
+
+    // Options that will be passed to the testEnvironment
+    // testEnvironmentOptions: {},
+
+    // An array of regexp pattern strings that are matched against all test paths, matched tests are skipped
+    testPathIgnorePatterns: [
+        '/node_modules/'
+    ],
+
+    // The regexp pattern or array of patterns that Jest uses to detect test files
+    testRegex: '.(.+)\\.test\\.(js|ts)$',
+
+    // This option allows the use of a custom results processor
+    // testResultsProcessor: undefined,
+
+    // This option allows use of a custom test runner
+    // testRunner: "jest-circus/runner",
+
+    // This option sets the URL for the jsdom environment. It is reflected in properties such as location.href
+    testURL: 'http://localhost:9898/',
+
+    // Setting this value to "fake" allows the use of fake timers for functions such as "setTimeout"
+    // timers: "real",
+
+    // A map from regular expressions to paths to transformers
+    transform: {
+        '^.+\\.js$': 'babel-jest'
+    },
+
+    // An array of regexp pattern strings that are matched against all source file paths, matched files will skip transformation
+    transformIgnorePatterns: [
+        '/node_modules/',
+        '\\.pnp\\.[^\\/]+$'
+    ],
+
+    // An array of regexp pattern strings that are matched against all modules before the module loader will automatically return a mock for them
+    // unmockedModulePathPatterns: undefined,
+
+    // Indicates whether each individual test should be reported during the run
+    verbose: true,
+
+    // An array of regexp patterns that are matched against all source file paths before re-running tests in watch mode
+    // watchPathIgnorePatterns: [],
+
+    // Whether to use watchman for file crawling
+    // watchman: true,
+    testTimeout: 50000
+};
diff --git a/test_tipc/web/test.jpg b/test_tipc/web/test.jpg
new file mode 100644
index 0000000..60682be
Binary files /dev/null and b/test_tipc/web/test.jpg differ
diff --git a/tests/__init__.py b/tests/__init__.py
new file mode 100644
index 0000000..290f972
--- /dev/null
+++ b/tests/__init__.py
@@ -0,0 +1,13 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/tests/api_client/test_cli.py b/tests/api_client/test_cli.py
new file mode 100644
index 0000000..225fe67
--- /dev/null
+++ b/tests/api_client/test_cli.py
@@ -0,0 +1,82 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import argparse
+from pathlib import Path
+
+from paddleocr._api_client import cli
+from paddleocr._api_client.results import OCRPage, OCRResult
+
+
+def test_cli_save_resources_uses_result_specific_helper(tmp_path, monkeypatch, capsys):
+    class FakeClient:
+        def __init__(self, **kwargs):
+            self.kwargs = kwargs
+
+        def ocr(self, **kwargs):
+            return OCRResult(
+                job_id="job-1",
+                pages=[
+                    OCRPage(
+                        pruned_result={}, ocr_image_url="https://example.test/a.png"
+                    )
+                ],
+            )
+
+        def save_ocr_result_resources(self, result, destination, overwrite=False):
+            assert result.job_id == "job-1"
+            assert destination == str(tmp_path)
+            assert overwrite is True
+            return [str(Path(destination) / "ocr-page-1.png")]
+
+        def close(self):
+            return None
+
+    monkeypatch.setattr(cli, "PaddleOCRClient", FakeClient)
+
+    args = argparse.Namespace(
+        token="token",
+        client_platform=None,
+        base_url=None,
+        request_timeout=300.0,
+        poll_timeout=600.0,
+        model=None,
+        model_type="ocr",
+        file_url="https://example.test/input.pdf",
+        file_path=None,
+        output=None,
+        save_resources=str(tmp_path),
+        overwrite_resources=True,
+        page_ranges=None,
+        batch_id=None,
+        use_doc_orientation_classify=None,
+        use_doc_unwarping=None,
+        use_textline_orientation=None,
+        use_chart_recognition=None,
+        use_seal_recognition=None,
+        use_table_recognition=None,
+        use_formula_recognition=None,
+        use_layout_detection=None,
+        text_det_limit_side_len=None,
+        text_det_limit_type=None,
+        text_rec_score_thresh=None,
+        prettify_markdown=None,
+        visualize=None,
+    )
+
+    cli._execute_api(args)
+
+    captured = capsys.readouterr()
+    assert '"jobId": "job-1"' in captured.out
+    assert "Resources saved to:" in captured.err
diff --git a/tests/api_client/test_core.py b/tests/api_client/test_core.py
new file mode 100644
index 0000000..913f668
--- /dev/null
+++ b/tests/api_client/test_core.py
@@ -0,0 +1,174 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+from paddleocr._api_client._core import (
+    default_payload,
+    job_id_for_task,
+    resolve_document_model,
+    resolve_document_options,
+    resolve_ocr_model,
+    validate_input_source,
+)
+from paddleocr._api_client.errors import InvalidRequestError
+from paddleocr._api_client.models import (
+    Model,
+    OCROptions,
+    PaddleOCRVLOptions,
+    PPStructureV3Options,
+)
+from paddleocr._utils.naming import snake_to_camel
+from paddleocr._api_client._poller import parse_doc_parsing_result, parse_ocr_result
+from paddleocr._api_client.results import Job
+
+
+def test_ppstructurev3_options_to_payload_uses_official_camel_case_keys():
+    assert snake_to_camel("top_p") == "topP"
+    assert snake_to_camel("use_e2e_wired_table_rec_model") == "useE2eWiredTableRecModel"
+
+    options = PPStructureV3Options(
+        use_e2e_wired_table_rec_model=True,
+        format_block_content=True,
+        markdown_ignore_labels=["image"],
+        text_det_limit_side_len=1280,
+        extra_options={"futureOption": "enabled"},
+    )
+    assert options.to_payload() == {
+        "useE2eWiredTableRecModel": True,
+        "formatBlockContent": True,
+        "markdownIgnoreLabels": ["image"],
+        "textDetLimitSideLen": 1280,
+        "futureOption": "enabled",
+    }
+
+
+def test_paddleocr_vl_options_include_service_parameters_and_extra_options():
+    options = PaddleOCRVLOptions(
+        use_ocr_for_image_block=True,
+        format_block_content=True,
+        markdown_ignore_labels=["image"],
+        vlm_extra_args={"temperature": 0.1},
+        return_markdown_images=False,
+        output_formats=["docx"],
+        extra_options={"futureOption": "enabled"},
+    )
+
+    assert options.to_payload() == {
+        "useOcrForImageBlock": True,
+        "formatBlockContent": True,
+        "markdownIgnoreLabels": ["image"],
+        "vlmExtraArgs": {"temperature": 0.1},
+        "returnMarkdownImages": False,
+        "outputFormats": ["docx"],
+        "futureOption": "enabled",
+    }
+
+
+def test_result_parsers_preserve_raw_fields_and_data_info():
+    ocr_line = {
+        "result": {
+            "dataInfo": {"numPages": 1},
+            "ocrResults": [
+                {
+                    "prunedResult": {"rec_texts": ["hello"]},
+                    "ocrImage": "ocr.png",
+                    "docPreprocessingImage": "pre.png",
+                    "inputImage": "input.png",
+                }
+            ],
+        }
+    }
+    ocr_result = parse_ocr_result("job-ocr", [ocr_line])
+    assert ocr_result.data_info == {"numPages": 1}
+    assert ocr_result.pages[0].doc_preprocessing_image_url == "pre.png"
+    assert ocr_result.pages[0].input_image_url == "input.png"
+    assert ocr_result.pages[0].raw["ocrImage"] == "ocr.png"
+
+    doc_page = {
+        "prunedResult": {"blocks": [{"label": "text", "content": "hello"}]},
+        "markdown": {
+            "text": "hello",
+            "images": {"figure.png": "figure-url"},
+            "isStart": True,
+        },
+        "outputImages": {"page.png": "page-url"},
+        "inputImage": "input.png",
+        "exports": {"docx": "docx-url"},
+    }
+    doc_line = {
+        "result": {
+            "dataInfo": {"numPages": 1},
+            "layoutParsingResults": [doc_page],
+        }
+    }
+    doc_result = parse_doc_parsing_result("job-doc", [doc_line])
+    assert doc_result.data_info == {"numPages": 1}
+    assert doc_result.pages[0].pruned_result == doc_page["prunedResult"]
+    assert doc_result.pages[0].markdown == doc_page["markdown"]
+    assert doc_result.pages[0].exports == {"docx": "docx-url"}
+    assert doc_result.pages[0].raw == doc_page
+
+
+def test_core_resolves_models_and_default_payloads():
+    assert resolve_ocr_model("PP-OCRv5") is Model.PP_OCRV5
+    assert resolve_ocr_model("PP-OCRv5-latin") is Model.PP_OCRV5_LATIN
+    assert resolve_ocr_model("PP-OCRv6") is Model.PP_OCRV6
+    assert resolve_document_model("PaddleOCR-VL-1.6") is Model.PADDLE_OCR_VL_16
+
+    assert default_payload(Model.PP_OCRV5) == OCROptions().to_payload()
+    assert default_payload(Model.PP_OCRV5_LATIN) == OCROptions().to_payload()
+    assert default_payload(Model.PP_OCRV6) == OCROptions().to_payload()
+    assert default_payload(Model.PADDLE_OCR_VL_16) == PaddleOCRVLOptions().to_payload()
+
+    with pytest.raises(InvalidRequestError):
+        resolve_ocr_model("PP-StructureV3")
+
+    with pytest.raises(InvalidRequestError):
+        resolve_document_model("unknown-model")
+
+
+def test_core_selects_document_options_by_model():
+    assert isinstance(
+        resolve_document_options(Model.PP_STRUCTURE_V3, None),
+        PPStructureV3Options,
+    )
+    assert isinstance(
+        resolve_document_options(Model.PADDLE_OCR_VL_16, None),
+        PaddleOCRVLOptions,
+    )
+
+    with pytest.raises(InvalidRequestError):
+        resolve_document_options(Model.PP_STRUCTURE_V3, PaddleOCRVLOptions())
+
+    with pytest.raises(InvalidRequestError):
+        resolve_document_options(Model.PADDLE_OCR_VL_16, PPStructureV3Options())
+
+
+def test_core_validates_input_source_and_job_task():
+    validate_input_source("https://example.test/file.pdf", None)
+    validate_input_source(None, "/tmp/file.pdf")
+
+    with pytest.raises(InvalidRequestError):
+        validate_input_source(None, None)
+
+    with pytest.raises(InvalidRequestError):
+        validate_input_source("https://example.test/file.pdf", "/tmp/file.pdf")
+
+    job = Job(job_id="job-1", model=Model.PP_OCRV5.value, task="ocr")
+    assert job_id_for_task(job, "ocr") == "job-1"
+    assert job_id_for_task("job-2", "ocr") == "job-2"
+
+    with pytest.raises(InvalidRequestError):
+        job_id_for_task(job, "document_parsing")
diff --git a/tests/api_client/test_http.py b/tests/api_client/test_http.py
new file mode 100644
index 0000000..8d3f38d
--- /dev/null
+++ b/tests/api_client/test_http.py
@@ -0,0 +1,264 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import json
+from http.server import BaseHTTPRequestHandler, HTTPServer
+from threading import Thread
+
+import pytest
+
+from paddleocr._api_client._http import API_PATH, DEFAULT_BASE_URL, HTTPClient
+from paddleocr._api_client.client import PaddleOCRClient
+from paddleocr._api_client.errors import (
+    APIError,
+    AuthError,
+    InvalidRequestError,
+    NetworkError,
+    RateLimitError,
+    RequestTimeoutError,
+    ServiceUnavailableError,
+)
+from paddleocr._api_client.models import Model
+
+
+class _MockHandler(BaseHTTPRequestHandler):
+    """Records requests and responds with canned JSON."""
+
+    requests_log = []
+    response_body = {"code": 0, "data": {"jobId": "job-1"}}
+    response_status = 200
+
+    def do_POST(self):
+        content_length = int(self.headers.get("Content-Length", 0))
+        body = self.rfile.read(content_length) if content_length else b""
+        self.__class__.requests_log.append(
+            {
+                "method": "POST",
+                "path": self.path,
+                "headers": dict(self.headers),
+                "body": body,
+            }
+        )
+        self.send_response(self.__class__.response_status)
+        self.send_header("Content-Type", "application/json")
+        self.end_headers()
+        self.wfile.write(json.dumps(self.__class__.response_body).encode())
+
+    def do_GET(self):
+        self.__class__.requests_log.append(
+            {"method": "GET", "path": self.path, "headers": dict(self.headers)}
+        )
+        self.send_response(self.__class__.response_status)
+        self.send_header("Content-Type", "application/json")
+        self.end_headers()
+        self.wfile.write(json.dumps(self.__class__.response_body).encode())
+
+    def log_message(self, format, *args):
+        pass
+
+
+@pytest.fixture()
+def mock_server():
+    _MockHandler.requests_log = []
+    _MockHandler.response_body = {"code": 0, "data": {"jobId": "job-1"}}
+    _MockHandler.response_status = 200
+
+    server = HTTPServer(("127.0.0.1", 0), _MockHandler)
+    port = server.server_address[1]
+    thread = Thread(target=server.serve_forever, daemon=True)
+    thread.start()
+    yield f"http://127.0.0.1:{port}", _MockHandler
+    server.shutdown()
+
+
+def test_base_url_appends_api_path(mock_server):
+    base_url, handler = mock_server
+    client = HTTPClient("test-token", base_url, timeout=5.0)
+
+    client.submit_url("PP-OCRv5", "https://example.test/file.pdf", {})
+
+    assert len(handler.requests_log) == 1
+    assert handler.requests_log[0]["path"] == API_PATH
+
+
+def test_get_job_status_url(mock_server):
+    base_url, handler = mock_server
+    handler.response_body = {"code": 0, "data": {"jobId": "job-1", "state": "running"}}
+    client = HTTPClient("test-token", base_url, timeout=5.0)
+
+    client.get_job_status("job-123")
+
+    assert handler.requests_log[0]["path"] == f"{API_PATH}/job-123"
+
+
+def test_get_batch_status_url(mock_server):
+    base_url, handler = mock_server
+    handler.response_body = {"code": 0, "data": {"batchId": "b-1", "jobs": []}}
+    client = HTTPClient("test-token", base_url, timeout=5.0)
+
+    client.get_batch_status("batch-abc")
+
+    assert handler.requests_log[0]["path"] == f"{API_PATH}/batch/batch-abc"
+
+
+def test_base_url_trailing_slashes_normalized(mock_server):
+    base_url, handler = mock_server
+    client = HTTPClient("test-token", base_url + "///", timeout=5.0)
+
+    client.submit_url("PP-OCRv5", "https://example.test/file.pdf", {})
+
+    assert handler.requests_log[0]["path"] == API_PATH
+
+
+def test_authorization_header_sent(mock_server):
+    base_url, handler = mock_server
+    client = HTTPClient("my-secret-token", base_url, timeout=5.0)
+
+    client.submit_url("PP-OCRv5", "https://example.test/file.pdf", {})
+
+    assert (
+        handler.requests_log[0]["headers"]["Authorization"] == "Bearer my-secret-token"
+    )
+
+
+def test_client_platform_header(mock_server):
+    base_url, handler = mock_server
+    client = HTTPClient("token", base_url, timeout=5.0, client_platform="sdk-test")
+
+    client.submit_url("PP-OCRv5", "https://example.test/file.pdf", {})
+
+    assert handler.requests_log[0]["headers"]["Client-Platform"] == "sdk-test"
+
+
+def test_submit_url_posts_correct_body(mock_server):
+    base_url, handler = mock_server
+    client = HTTPClient("token", base_url, timeout=5.0)
+
+    job_id = client.submit_url(
+        "PP-OCRv5",
+        "https://example.test/file.pdf",
+        {"visualize": True},
+        page_ranges="1-3",
+        batch_id="batch-1",
+    )
+
+    assert job_id == "job-1"
+    body = json.loads(handler.requests_log[0]["body"])
+    assert body == {
+        "fileUrl": "https://example.test/file.pdf",
+        "model": "PP-OCRv5",
+        "optionalPayload": {"visualize": True},
+        "pageRanges": "1-3",
+        "batchId": "batch-1",
+    }
+
+
+def test_error_mapping_401(mock_server):
+    base_url, handler = mock_server
+    handler.response_status = 401
+    handler.response_body = {"code": 401, "msg": "Unauthorized"}
+    client = HTTPClient("bad-token", base_url, timeout=5.0)
+
+    with pytest.raises(AuthError):
+        client.submit_url("PP-OCRv5", "https://example.test/file.pdf", {})
+
+
+def test_error_mapping_400(mock_server):
+    base_url, handler = mock_server
+    handler.response_status = 400
+    handler.response_body = {"code": 400, "msg": "Bad request"}
+    client = HTTPClient("token", base_url, timeout=5.0)
+
+    with pytest.raises(InvalidRequestError):
+        client.submit_url("PP-OCRv5", "https://example.test/file.pdf", {})
+
+
+def test_error_mapping_429(mock_server):
+    base_url, handler = mock_server
+    handler.response_status = 429
+    handler.response_body = {"code": 429, "msg": "Rate limited"}
+    client = HTTPClient("token", base_url, timeout=5.0)
+
+    with pytest.raises(RateLimitError):
+        client.submit_url("PP-OCRv5", "https://example.test/file.pdf", {})
+
+
+def test_error_mapping_503(mock_server):
+    base_url, handler = mock_server
+    handler.response_status = 503
+    handler.response_body = {"code": 503, "msg": "Service unavailable"}
+    client = HTTPClient("token", base_url, timeout=5.0)
+
+    with pytest.raises(ServiceUnavailableError):
+        client.submit_url("PP-OCRv5", "https://example.test/file.pdf", {})
+
+
+def test_error_mapping_500(mock_server):
+    base_url, handler = mock_server
+    handler.response_status = 500
+    handler.response_body = {"code": 500, "msg": "Internal error"}
+    client = HTTPClient("token", base_url, timeout=5.0)
+
+    with pytest.raises(APIError):
+        client.submit_url("PP-OCRv5", "https://example.test/file.pdf", {})
+
+
+def test_paddleocr_client_requires_token(monkeypatch):
+    monkeypatch.delenv("PADDLEOCR_ACCESS_TOKEN", raising=False)
+    with pytest.raises(AuthError):
+        PaddleOCRClient()
+
+
+def test_paddleocr_client_reads_env_token(monkeypatch):
+    monkeypatch.setenv("PADDLEOCR_ACCESS_TOKEN", "env-token")
+    client = PaddleOCRClient()
+    client.close()
+
+
+def test_default_base_url_is_host_only():
+    assert "/api/" not in DEFAULT_BASE_URL
+    assert DEFAULT_BASE_URL == "https://paddleocr.aistudio-app.com"
+
+
+def test_paddleocr_client_client_platform_header(monkeypatch):
+    class FakeResponse:
+        status_code = 200
+        text = ""
+
+        def json(self):
+            return {"code": 0, "data": {"jobId": "job-1"}}
+
+    class FakeSession:
+        def __init__(self):
+            self.headers = {}
+            self.captured_headers = None
+
+        def post(self, *args, **kwargs):
+            self.captured_headers = dict(self.headers)
+            return FakeResponse()
+
+        def close(self):
+            return None
+
+    fake_session = FakeSession()
+    monkeypatch.setattr(
+        "paddleocr._api_client._http.requests.Session",
+        lambda: fake_session,
+    )
+
+    client = PaddleOCRClient(token="token", client_platform="my-app")
+    job = client.submit_ocr(file_url="https://example.test/input.pdf")
+
+    assert job.job_id == "job-1"
+    assert fake_session.captured_headers["Client-Platform"] == "my-app"
diff --git a/tests/api_client/test_resources.py b/tests/api_client/test_resources.py
new file mode 100644
index 0000000..a4ab8f7
--- /dev/null
+++ b/tests/api_client/test_resources.py
@@ -0,0 +1,133 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from pathlib import Path
+
+import pytest
+
+from paddleocr._api_client.client import PaddleOCRClient
+from paddleocr._api_client.errors import InvalidRequestError
+from paddleocr._api_client.results import (
+    DocParsingPage,
+    DocParsingResult,
+    OCRPage,
+    OCRResult,
+)
+
+
+class _Response:
+    def __init__(self, content: bytes):
+        self.content = content
+
+    def raise_for_status(self):
+        return None
+
+
+def test_save_resource_writes_single_url(tmp_path, monkeypatch):
+    calls = []
+
+    def fake_get(url, timeout):
+        calls.append((url, timeout))
+        return _Response(b"content")
+
+    monkeypatch.setattr("paddleocr._api_client._resources.requests.get", fake_get)
+
+    client = PaddleOCRClient(token="token", request_timeout=12.0)
+    saved_path = client.save_resource(
+        "https://example.test/path/image.png", str(tmp_path)
+    )
+
+    assert saved_path == str(tmp_path / "image.png")
+    assert (tmp_path / "image.png").read_bytes() == b"content"
+    assert calls == [("https://example.test/path/image.png", 12.0)]
+
+
+def test_save_ocr_result_resources_uses_stable_page_names(tmp_path, monkeypatch):
+    contents = [b"one", b"two"]
+
+    def fake_get(url, timeout):
+        return _Response(contents.pop(0))
+
+    monkeypatch.setattr("paddleocr._api_client._resources.requests.get", fake_get)
+
+    client = PaddleOCRClient(token="token")
+    result = OCRResult(
+        job_id="job-1",
+        pages=[
+            OCRPage(pruned_result={}, ocr_image_url="https://example.test/a.png?sig=1"),
+            OCRPage(pruned_result={}, ocr_image_url="https://example.test/b.jpg"),
+        ],
+    )
+
+    saved_paths = client.save_ocr_result_resources(result, str(tmp_path))
+
+    assert saved_paths == [
+        str(tmp_path / "ocr-page-1.png"),
+        str(tmp_path / "ocr-page-2.jpg"),
+    ]
+    assert (tmp_path / "ocr-page-1.png").read_bytes() == b"one"
+    assert (tmp_path / "ocr-page-2.jpg").read_bytes() == b"two"
+
+
+def test_save_document_parsing_result_resources_sorts_and_validates_keys(
+    tmp_path, monkeypatch
+):
+    calls = []
+
+    def fake_get(url, timeout):
+        calls.append(url)
+        return _Response(url.encode())
+
+    monkeypatch.setattr("paddleocr._api_client._resources.requests.get", fake_get)
+
+    client = PaddleOCRClient(token="token")
+    result = DocParsingResult(
+        job_id="job-1",
+        pages=[
+            DocParsingPage(
+                markdown_text="",
+                markdown_images={
+                    "b.png": "https://example.test/b",
+                    "a.png": "https://example.test/a",
+                },
+                output_images={"rendered.jpg": "https://example.test/rendered"},
+            )
+        ],
+    )
+
+    saved_paths = client.save_document_parsing_result_resources(result, str(tmp_path))
+
+    assert saved_paths == [
+        str(tmp_path / "a.png"),
+        str(tmp_path / "b.png"),
+        str(tmp_path / "rendered.jpg"),
+    ]
+    assert calls == [
+        "https://example.test/a",
+        "https://example.test/b",
+        "https://example.test/rendered",
+    ]
+
+    for unsafe_key in ("../escape.png", "nested\\escape.png"):
+        bad_result = DocParsingResult(
+            job_id="job-2",
+            pages=[
+                DocParsingPage(
+                    markdown_text="",
+                    markdown_images={unsafe_key: "https://example.test/escape"},
+                )
+            ],
+        )
+        with pytest.raises(InvalidRequestError):
+            client.save_document_parsing_result_resources(bad_result, str(tmp_path))
diff --git a/tests/conftest.py b/tests/conftest.py
new file mode 100644
index 0000000..22d5161
--- /dev/null
+++ b/tests/conftest.py
@@ -0,0 +1,28 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Shared pytest configuration for the PaddleOCR repository test suite."""
+
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+REPO_ROOT = Path(__file__).resolve().parent.parent
+MCP_SERVER_ROOT = REPO_ROOT / "mcp_server"
+
+if str(REPO_ROOT) not in sys.path:
+    sys.path.insert(0, str(REPO_ROOT))
+if str(MCP_SERVER_ROOT) not in sys.path:
+    sys.path.insert(0, str(MCP_SERVER_ROOT))
diff --git a/tests/models/__init__.py b/tests/models/__init__.py
new file mode 100644
index 0000000..290f972
--- /dev/null
+++ b/tests/models/__init__.py
@@ -0,0 +1,13 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/tests/models/image_classification_common.py b/tests/models/image_classification_common.py
new file mode 100644
index 0000000..c547bea
--- /dev/null
+++ b/tests/models/image_classification_common.py
@@ -0,0 +1,24 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+def check_result_item_keys(result_item):
+    assert result_item.keys() == {
+        "input_path",
+        "page_index",
+        "input_img",
+        "class_ids",
+        "scores",
+        "label_names",
+    }
diff --git a/tests/models/object_detection_common.py b/tests/models/object_detection_common.py
new file mode 100644
index 0000000..7e59eaa
--- /dev/null
+++ b/tests/models/object_detection_common.py
@@ -0,0 +1,22 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+def check_result_item_keys(result_item):
+    assert result_item.keys() == {
+        "input_path",
+        "page_index",
+        "input_img",
+        "boxes",
+    }
diff --git a/tests/models/test_doc_img_orientation_classification.py b/tests/models/test_doc_img_orientation_classification.py
new file mode 100644
index 0000000..7a7a4da
--- /dev/null
+++ b/tests/models/test_doc_img_orientation_classification.py
@@ -0,0 +1,37 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+from paddleocr import DocImgOrientationClassification
+from ..testing_utils import TEST_DATA_DIR, check_simple_inference_result
+from .image_classification_common import check_result_item_keys
+
+
+@pytest.fixture(scope="module")
+def doc_img_orientation_classification_predictor():
+    return DocImgOrientationClassification()
+
+
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "book_rot180.jpg",
+    ],
+)
+def test_predict(doc_img_orientation_classification_predictor, image_path):
+    result = doc_img_orientation_classification_predictor.predict(str(image_path))
+
+    check_simple_inference_result(result)
+    check_result_item_keys(result[0])
diff --git a/tests/models/test_doc_vlm.py b/tests/models/test_doc_vlm.py
new file mode 100644
index 0000000..303653a
--- /dev/null
+++ b/tests/models/test_doc_vlm.py
@@ -0,0 +1,69 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+pytestmark = pytest.mark.py38_incompatible
+
+from paddleocr import DocVLM
+from ..testing_utils import (
+    TEST_DATA_DIR,
+    check_simple_inference_result,
+    check_wrapper_simple_inference_param_forwarding,
+)
+
+
+@pytest.fixture(scope="module")
+def doc_vlm_predictor():
+    return DocVLM()
+
+
+@pytest.mark.resource_intensive
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "medal_table.png",
+    ],
+)
+def test_predict(doc_vlm_predictor, image_path):
+    result = doc_vlm_predictor.predict(str(image_path))
+
+    check_simple_inference_result(result)
+    assert result[0].keys() == {
+        "input_path",
+        "page_index",
+        "input_img",
+        "result",
+    }
+
+
+@pytest.mark.resource_intensive
+@pytest.mark.parametrize(
+    "params",
+    [
+        {},
+    ],
+)
+def test_predict_params(
+    monkeypatch,
+    doc_vlm_predictor,
+    params,
+):
+    check_wrapper_simple_inference_param_forwarding(
+        monkeypatch,
+        doc_vlm_predictor,
+        "paddlex_predictor",
+        "dummy_path",
+        params,
+    )
diff --git a/tests/models/test_formula_recognition.py b/tests/models/test_formula_recognition.py
new file mode 100644
index 0000000..1ffa4bd
--- /dev/null
+++ b/tests/models/test_formula_recognition.py
@@ -0,0 +1,67 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+pytestmark = pytest.mark.py38_incompatible
+
+from paddleocr import FormulaRecognition
+from ..testing_utils import (
+    TEST_DATA_DIR,
+    check_simple_inference_result,
+    check_wrapper_simple_inference_param_forwarding,
+)
+
+
+@pytest.fixture(scope="module")
+def formula_recognition_predictor():
+    return FormulaRecognition()
+
+
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "formula.png",
+    ],
+)
+def test_predict(formula_recognition_predictor, image_path):
+    result = formula_recognition_predictor.predict(str(image_path))
+
+    check_simple_inference_result(result)
+    assert result[0].keys() == {
+        "input_path",
+        "page_index",
+        "input_img",
+        "rec_formula",
+    }
+
+
+@pytest.mark.parametrize(
+    "params",
+    [
+        {},
+    ],
+)
+def test_predict_params(
+    monkeypatch,
+    formula_recognition_predictor,
+    params,
+):
+    check_wrapper_simple_inference_param_forwarding(
+        monkeypatch,
+        formula_recognition_predictor,
+        "paddlex_predictor",
+        "dummy_path",
+        params,
+    )
diff --git a/tests/models/test_layout_detection.py b/tests/models/test_layout_detection.py
new file mode 100644
index 0000000..7fd03d5
--- /dev/null
+++ b/tests/models/test_layout_detection.py
@@ -0,0 +1,67 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+pytestmark = pytest.mark.py38_incompatible
+
+from paddleocr import LayoutDetection
+from ..testing_utils import (
+    TEST_DATA_DIR,
+    check_simple_inference_result,
+    check_wrapper_simple_inference_param_forwarding,
+)
+from .object_detection_common import check_result_item_keys
+
+
+@pytest.fixture(scope="module")
+def layout_detection_predictor():
+    return LayoutDetection()
+
+
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "doc_with_formula.png",
+    ],
+)
+def test_predict(layout_detection_predictor, image_path):
+    result = layout_detection_predictor.predict(str(image_path))
+
+    check_simple_inference_result(result)
+    check_result_item_keys(result[0])
+
+
+@pytest.mark.parametrize(
+    "params",
+    [
+        {"img_size": 640},
+        {"threshold": 0.5},
+        {"layout_nms": True},
+        {"layout_unclip_ratio": True},
+        {"layout_merge_bboxes_mode": True},
+    ],
+)
+def test_predict_params(
+    monkeypatch,
+    layout_detection_predictor,
+    params,
+):
+    check_wrapper_simple_inference_param_forwarding(
+        monkeypatch,
+        layout_detection_predictor,
+        "paddlex_predictor",
+        "dummy_path",
+        params,
+    )
diff --git a/tests/models/test_seal_text_detection.py b/tests/models/test_seal_text_detection.py
new file mode 100644
index 0000000..6b8687d
--- /dev/null
+++ b/tests/models/test_seal_text_detection.py
@@ -0,0 +1,71 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+pytestmark = pytest.mark.py38_incompatible
+
+from paddleocr import SealTextDetection
+from ..testing_utils import (
+    TEST_DATA_DIR,
+    check_simple_inference_result,
+    check_wrapper_simple_inference_param_forwarding,
+)
+
+
+@pytest.fixture(scope="module")
+def seal_text_detection_predictor():
+    return SealTextDetection()
+
+
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "seal.png",
+    ],
+)
+def test_predict(seal_text_detection_predictor, image_path):
+    result = seal_text_detection_predictor.predict(str(image_path))
+
+    check_simple_inference_result(result)
+    assert result[0].keys() == {
+        "input_path",
+        "page_index",
+        "input_img",
+        "dt_polys",
+        "dt_scores",
+    }
+
+
+@pytest.mark.parametrize(
+    "params",
+    [
+        {"limit_side_len": 640, "limit_type": "min"},
+        {"thresh": 0.5},
+        {"box_thresh": 0.3},
+        {"unclip_ratio": 3.0},
+    ],
+)
+def test_predict_params(
+    monkeypatch,
+    seal_text_detection_predictor,
+    params,
+):
+    check_wrapper_simple_inference_param_forwarding(
+        monkeypatch,
+        seal_text_detection_predictor,
+        "paddlex_predictor",
+        "dummy_path",
+        params,
+    )
diff --git a/tests/models/test_table_cells_detection.py b/tests/models/test_table_cells_detection.py
new file mode 100644
index 0000000..b7e8f88
--- /dev/null
+++ b/tests/models/test_table_cells_detection.py
@@ -0,0 +1,64 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+pytestmark = pytest.mark.py38_incompatible
+
+from paddleocr import TableCellsDetection
+from ..testing_utils import (
+    TEST_DATA_DIR,
+    check_simple_inference_result,
+    check_wrapper_simple_inference_param_forwarding,
+)
+from .object_detection_common import check_result_item_keys
+
+
+@pytest.fixture(scope="module")
+def table_cells_detection_predictor():
+    return TableCellsDetection()
+
+
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "table.jpg",
+    ],
+)
+def test_predict(table_cells_detection_predictor, image_path):
+    result = table_cells_detection_predictor.predict(str(image_path))
+
+    check_simple_inference_result(result)
+    check_result_item_keys(result[0])
+
+
+@pytest.mark.parametrize(
+    "params",
+    [
+        {"img_size": 640},
+        {"threshold": 0.5},
+    ],
+)
+def test_predict_params(
+    monkeypatch,
+    table_cells_detection_predictor,
+    params,
+):
+    check_wrapper_simple_inference_param_forwarding(
+        monkeypatch,
+        table_cells_detection_predictor,
+        "paddlex_predictor",
+        "dummy_path",
+        params,
+    )
diff --git a/tests/models/test_table_classification.py b/tests/models/test_table_classification.py
new file mode 100644
index 0000000..f7dc4c1
--- /dev/null
+++ b/tests/models/test_table_classification.py
@@ -0,0 +1,39 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+pytestmark = pytest.mark.py38_incompatible
+
+from paddleocr import TableClassification
+from ..testing_utils import TEST_DATA_DIR, check_simple_inference_result
+from .image_classification_common import check_result_item_keys
+
+
+@pytest.fixture(scope="module")
+def table_classification_predictor():
+    return TableClassification()
+
+
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "table.jpg",
+    ],
+)
+def test_predict(table_classification_predictor, image_path):
+    result = table_classification_predictor.predict(str(image_path))
+
+    check_simple_inference_result(result)
+    check_result_item_keys(result[0])
diff --git a/tests/models/test_table_structure_recognition.py b/tests/models/test_table_structure_recognition.py
new file mode 100644
index 0000000..9a7c3bd
--- /dev/null
+++ b/tests/models/test_table_structure_recognition.py
@@ -0,0 +1,69 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+pytestmark = pytest.mark.py38_incompatible
+
+from paddleocr import TableStructureRecognition
+from ..testing_utils import (
+    TEST_DATA_DIR,
+    check_simple_inference_result,
+    check_wrapper_simple_inference_param_forwarding,
+)
+
+
+@pytest.fixture(scope="module")
+def table_structure_recognition_predictor():
+    return TableStructureRecognition()
+
+
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "table.jpg",
+    ],
+)
+def test_predict(table_structure_recognition_predictor, image_path):
+    result = table_structure_recognition_predictor.predict(str(image_path))
+
+    check_simple_inference_result(result)
+    assert result[0].keys() == {
+        "input_path",
+        "page_index",
+        "input_img",
+        "bbox",
+        "structure",
+        "structure_score",
+    }
+
+
+@pytest.mark.parametrize(
+    "params",
+    [
+        {},
+    ],
+)
+def test_predict_params(
+    monkeypatch,
+    table_structure_recognition_predictor,
+    params,
+):
+    check_wrapper_simple_inference_param_forwarding(
+        monkeypatch,
+        table_structure_recognition_predictor,
+        "paddlex_predictor",
+        "dummy_path",
+        params,
+    )
diff --git a/tests/models/test_text_detection.py b/tests/models/test_text_detection.py
new file mode 100644
index 0000000..21ea8c0
--- /dev/null
+++ b/tests/models/test_text_detection.py
@@ -0,0 +1,72 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+from paddleocr import TextDetection
+from ..testing_utils import (
+    TEST_DATA_DIR,
+    check_simple_inference_result,
+    check_wrapper_simple_inference_param_forwarding,
+)
+
+
+pytestmark = pytest.mark.py38_incompatible
+
+
+@pytest.fixture(scope="module")
+def text_detection_predictor():
+    return TextDetection()
+
+
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "table.jpg",
+    ],
+)
+def test_predict(text_detection_predictor, image_path):
+    result = text_detection_predictor.predict(str(image_path))
+
+    check_simple_inference_result(result)
+    assert result[0].keys() == {
+        "input_path",
+        "page_index",
+        "input_img",
+        "dt_polys",
+        "dt_scores",
+    }
+
+
+@pytest.mark.parametrize(
+    "params",
+    [
+        {"limit_side_len": 640, "limit_type": "min"},
+        {"thresh": 0.5},
+        {"box_thresh": 0.3},
+        {"unclip_ratio": 3.0},
+    ],
+)
+def test_predict_params(
+    monkeypatch,
+    text_detection_predictor,
+    params,
+):
+    check_wrapper_simple_inference_param_forwarding(
+        monkeypatch,
+        text_detection_predictor,
+        "paddlex_predictor",
+        "dummy_path",
+        params,
+    )
diff --git a/tests/models/test_text_image_unwarping.py b/tests/models/test_text_image_unwarping.py
new file mode 100644
index 0000000..ab6ada5
--- /dev/null
+++ b/tests/models/test_text_image_unwarping.py
@@ -0,0 +1,65 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+from paddleocr import TextImageUnwarping
+from ..testing_utils import (
+    TEST_DATA_DIR,
+    check_simple_inference_result,
+    check_wrapper_simple_inference_param_forwarding,
+)
+
+
+@pytest.fixture(scope="module")
+def text_image_unwarping_predictor():
+    return TextImageUnwarping()
+
+
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "book.jpg",
+    ],
+)
+def test_predict(text_image_unwarping_predictor, image_path):
+    result = text_image_unwarping_predictor.predict(str(image_path))
+
+    check_simple_inference_result(result)
+    assert result[0].keys() == {
+        "input_path",
+        "page_index",
+        "input_img",
+        "doctr_img",
+    }
+
+
+@pytest.mark.parametrize(
+    "params",
+    [
+        {},
+    ],
+)
+def test_predict_params(
+    monkeypatch,
+    text_image_unwarping_predictor,
+    params,
+):
+    check_wrapper_simple_inference_param_forwarding(
+        monkeypatch,
+        text_image_unwarping_predictor,
+        "paddlex_predictor",
+        "dummy_path",
+        params,
+    )
diff --git a/tests/models/test_text_recognition.py b/tests/models/test_text_recognition.py
new file mode 100644
index 0000000..b47bd43
--- /dev/null
+++ b/tests/models/test_text_recognition.py
@@ -0,0 +1,46 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+from paddleocr import TextRecognition
+from ..testing_utils import TEST_DATA_DIR, check_simple_inference_result
+
+
+pytestmark = pytest.mark.py38_incompatible
+
+
+@pytest.fixture(scope="module")
+def text_recognition_predictor():
+    return TextRecognition()
+
+
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "textline.png",
+    ],
+)
+def test_predict(text_recognition_predictor, image_path):
+    result = text_recognition_predictor.predict(str(image_path))
+
+    check_simple_inference_result(result)
+    assert result[0].keys() == {
+        "input_path",
+        "page_index",
+        "input_img",
+        "rec_text",
+        "rec_score",
+        "vis_font",
+    }
diff --git a/tests/models/test_textline_orientation_classification.py b/tests/models/test_textline_orientation_classification.py
new file mode 100644
index 0000000..ca4c5e7
--- /dev/null
+++ b/tests/models/test_textline_orientation_classification.py
@@ -0,0 +1,37 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+from paddleocr import TextLineOrientationClassification
+from ..testing_utils import TEST_DATA_DIR, check_simple_inference_result
+from .image_classification_common import check_result_item_keys
+
+
+@pytest.fixture(scope="module")
+def textline_orientation_classification_predictor():
+    return TextLineOrientationClassification()
+
+
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "textline_rot180.jpg",
+    ],
+)
+def test_predict(textline_orientation_classification_predictor, image_path):
+    result = textline_orientation_classification_predictor.predict(str(image_path))
+
+    check_simple_inference_result(result)
+    check_result_item_keys(result[0])
diff --git a/tests/pipelines/__init__.py b/tests/pipelines/__init__.py
new file mode 100644
index 0000000..290f972
--- /dev/null
+++ b/tests/pipelines/__init__.py
@@ -0,0 +1,13 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/tests/pipelines/test_doc_preprocessor.py b/tests/pipelines/test_doc_preprocessor.py
new file mode 100644
index 0000000..d6b44d6
--- /dev/null
+++ b/tests/pipelines/test_doc_preprocessor.py
@@ -0,0 +1,71 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+from paddleocr import DocPreprocessor
+from ..testing_utils import (
+    TEST_DATA_DIR,
+    check_simple_inference_result,
+    check_wrapper_simple_inference_param_forwarding,
+)
+
+
+@pytest.fixture(scope="module")
+def ocr_engine() -> DocPreprocessor:
+    return DocPreprocessor()
+
+
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "book_rot180.jpg",
+    ],
+)
+def test_predict(ocr_engine: DocPreprocessor, image_path: str) -> None:
+    """
+    Test PaddleOCR's doc preprocessor functionality.
+
+    Args:
+        ocr_engine: An instance of `DocPreprocessor`.
+        image_path: Path to the image to be processed.
+    """
+    result = ocr_engine.predict(str(image_path))
+
+    check_simple_inference_result(result)
+    res = result[0]
+    assert res["angle"] in {0, 90, 180, 270, -1}
+    assert res["rot_img"] is not None
+    assert res["output_img"] is not None
+
+
+@pytest.mark.parametrize(
+    "params",
+    [
+        {"use_doc_orientation_classify": False},
+        {"use_doc_unwarping": False},
+    ],
+)
+def test_predict_params(
+    monkeypatch,
+    ocr_engine: DocPreprocessor,
+    params: dict,
+) -> None:
+    check_wrapper_simple_inference_param_forwarding(
+        monkeypatch,
+        ocr_engine,
+        "paddlex_pipeline",
+        "dummy_path",
+        params,
+    )
diff --git a/tests/pipelines/test_doc_understanding.py b/tests/pipelines/test_doc_understanding.py
new file mode 100644
index 0000000..a02ce6a
--- /dev/null
+++ b/tests/pipelines/test_doc_understanding.py
@@ -0,0 +1,59 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+pytestmark = pytest.mark.py38_incompatible
+
+from paddleocr import DocUnderstanding
+from ..testing_utils import (
+    TEST_DATA_DIR,
+    check_simple_inference_result,
+    check_wrapper_simple_inference_param_forwarding,
+)
+
+
+@pytest.fixture(scope="module")
+def ocr_engine() -> DocUnderstanding:
+    return DocUnderstanding()
+
+
+@pytest.mark.resource_intensive
+@pytest.mark.parametrize(
+    "input",
+    [
+        {
+            "image": str(TEST_DATA_DIR / "medal_table.png"),
+            "query": "识别这份表格的内容",
+        },
+        {
+            "image": str(TEST_DATA_DIR / "table.jpg"),
+            "query": "识别这份表格的内容",
+        },
+    ],
+)
+def test_predict(ocr_engine: DocUnderstanding, input: dict) -> None:
+    """
+    Test PaddleOCR's doc understanding functionality.
+
+    Args:
+        ocr_engine: An instance of `DocUnderstanding`.
+        input: Input dict to be processed.
+    """
+    result = ocr_engine.predict(input)
+
+    check_simple_inference_result(result)
+    res = result[0]
+    assert res["result"] is not None
+    assert isinstance(res["result"], str)
diff --git a/tests/pipelines/test_formula_recognition.py b/tests/pipelines/test_formula_recognition.py
new file mode 100644
index 0000000..389477c
--- /dev/null
+++ b/tests/pipelines/test_formula_recognition.py
@@ -0,0 +1,84 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+pytestmark = pytest.mark.py38_incompatible
+
+from paddleocr import FormulaRecognitionPipeline
+from ..testing_utils import (
+    TEST_DATA_DIR,
+    check_simple_inference_result,
+    check_wrapper_simple_inference_param_forwarding,
+)
+
+
+@pytest.fixture(scope="module")
+def formula_recognition_engine() -> FormulaRecognitionPipeline:
+    return FormulaRecognitionPipeline()
+
+
+# TODO: Should we separate unit tests and integration tests?
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "doc_with_formula.png",
+    ],
+)
+def test_predict(
+    formula_recognition_engine: FormulaRecognitionPipeline, image_path: str
+) -> None:
+    """
+    Test FormulaRecognitionPipeline's formula_recognition functionality.
+
+    Args:
+        formula_recognition_engine: An instance of `FormulaRecognitionPipeline`.
+        image_path: Path to the image to be processed.
+    """
+    result = formula_recognition_engine.predict(str(image_path))
+
+    check_simple_inference_result(result)
+    res = result[0]
+    assert isinstance(res["formula_res_list"], list)
+    assert len(res["formula_res_list"]) > 0
+
+
+# TODO: Also check passing `None`
+@pytest.mark.parametrize(
+    "params",
+    [
+        {"use_doc_orientation_classify": False},
+        {"use_doc_unwarping": False},
+        {"use_layout_detection": False},
+        {"layout_threshold": 0.5},
+        {"layout_nms": True},
+        {"layout_unclip_ratio": 1.5},
+        {"layout_merge_bboxes_mode": "large"},
+    ],
+)
+def test_predict_params(
+    monkeypatch,
+    formula_recognition_engine: FormulaRecognitionPipeline,
+    params: dict,
+) -> None:
+    check_wrapper_simple_inference_param_forwarding(
+        monkeypatch,
+        formula_recognition_engine,
+        "paddlex_pipeline",
+        "dummy_path",
+        params,
+    )
+
+
+# TODO: Test init params
diff --git a/tests/pipelines/test_ocr.py b/tests/pipelines/test_ocr.py
new file mode 100644
index 0000000..4a0a675
--- /dev/null
+++ b/tests/pipelines/test_ocr.py
@@ -0,0 +1,188 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+from paddleocr import PaddleOCR
+from ..testing_utils import (
+    TEST_DATA_DIR,
+    check_simple_inference_result,
+    check_wrapper_simple_inference_param_forwarding,
+)
+
+
+@pytest.fixture(scope="module")
+def ocr_engine() -> PaddleOCR:
+    return PaddleOCR(
+        text_detection_model_name="PP-OCRv5_server_det",
+        text_recognition_model_name="PP-OCRv5_server_rec",
+    )
+
+
+# TODO: Should we separate unit tests and integration tests?
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "table.jpg",
+    ],
+)
+def test_predict(ocr_engine: PaddleOCR, image_path: str) -> None:
+    """
+    Test PaddleOCR's OCR functionality.
+
+    Args:
+        ocr_engine: An instance of `PaddleOCR`.
+        image_path: Path to the image to be processed.
+    """
+    result = ocr_engine.predict(str(image_path))
+
+    check_simple_inference_result(result)
+    res = result[0]
+    assert len(res["dt_polys"]) > 0
+    assert isinstance(res["rec_texts"], list)
+    assert len(res["rec_texts"]) > 0
+    for text in res["rec_texts"]:
+        assert isinstance(text, str)
+
+
+# TODO: Also check passing `None`
+@pytest.mark.parametrize(
+    "params",
+    [
+        {"use_doc_orientation_classify": False},
+        {"use_doc_unwarping": False},
+        {"use_textline_orientation": False},
+        {"text_det_limit_side_len": 640, "text_det_limit_type": "min"},
+        {"text_det_thresh": 0.5},
+        {"text_det_box_thresh": 0.3},
+        {"text_det_unclip_ratio": 3.0},
+        {"text_rec_score_thresh": 0.5},
+    ],
+)
+def test_predict_params(
+    monkeypatch,
+    ocr_engine: PaddleOCR,
+    params: dict,
+) -> None:
+    check_wrapper_simple_inference_param_forwarding(
+        monkeypatch,
+        ocr_engine,
+        "paddlex_pipeline",
+        "dummy_path",
+        params,
+    )
+
+
+# TODO: Test init params
+
+
+def test_lang_and_ocr_version():
+    ocr_engine = PaddleOCR(lang="ch", ocr_version="PP-OCRv5")
+    assert ocr_engine._params["text_detection_model_name"] == "PP-OCRv5_server_det"
+    assert ocr_engine._params["text_recognition_model_name"] == "PP-OCRv5_server_rec"
+    ocr_engine = PaddleOCR(lang="chinese_cht", ocr_version="PP-OCRv5")
+    assert ocr_engine._params["text_detection_model_name"] == "PP-OCRv5_server_det"
+    assert ocr_engine._params["text_recognition_model_name"] == "PP-OCRv5_server_rec"
+    ocr_engine = PaddleOCR(lang="en", ocr_version="PP-OCRv5")
+    assert ocr_engine._params["text_detection_model_name"] == "PP-OCRv5_server_det"
+    assert ocr_engine._params["text_recognition_model_name"] == "en_PP-OCRv5_mobile_rec"
+    ocr_engine = PaddleOCR(lang="japan", ocr_version="PP-OCRv5")
+    assert ocr_engine._params["text_detection_model_name"] == "PP-OCRv5_server_det"
+    assert ocr_engine._params["text_recognition_model_name"] == "PP-OCRv5_server_rec"
+    ocr_engine = PaddleOCR(lang="ch", ocr_version="PP-OCRv4")
+    assert ocr_engine._params["text_detection_model_name"] == "PP-OCRv4_mobile_det"
+    assert ocr_engine._params["text_recognition_model_name"] == "PP-OCRv4_mobile_rec"
+    ocr_engine = PaddleOCR(lang="en", ocr_version="PP-OCRv4")
+    assert ocr_engine._params["text_detection_model_name"] == "PP-OCRv4_mobile_det"
+    assert ocr_engine._params["text_recognition_model_name"] == "en_PP-OCRv4_mobile_rec"
+    ocr_engine = PaddleOCR(lang="ch", ocr_version="PP-OCRv3")
+    assert ocr_engine._params["text_detection_model_name"] == "PP-OCRv3_mobile_det"
+    assert ocr_engine._params["text_recognition_model_name"] == "PP-OCRv3_mobile_rec"
+    ocr_engine = PaddleOCR(lang="en", ocr_version="PP-OCRv3")
+    assert ocr_engine._params["text_detection_model_name"] == "PP-OCRv3_mobile_det"
+    assert ocr_engine._params["text_recognition_model_name"] == "en_PP-OCRv3_mobile_rec"
+    ocr_engine = PaddleOCR(lang="fr", ocr_version="PP-OCRv3")
+    assert ocr_engine._params["text_detection_model_name"] == "PP-OCRv3_mobile_det"
+    assert (
+        ocr_engine._params["text_recognition_model_name"] == "latin_PP-OCRv3_mobile_rec"
+    )
+    ocr_engine = PaddleOCR(lang="ar", ocr_version="PP-OCRv3")
+    assert ocr_engine._params["text_detection_model_name"] == "PP-OCRv3_mobile_det"
+    assert (
+        ocr_engine._params["text_recognition_model_name"]
+        == "arabic_PP-OCRv3_mobile_rec"
+    )
+    ocr_engine = PaddleOCR(lang="ru", ocr_version="PP-OCRv3")
+    assert ocr_engine._params["text_detection_model_name"] == "PP-OCRv3_mobile_det"
+    assert (
+        ocr_engine._params["text_recognition_model_name"]
+        == "cyrillic_PP-OCRv3_mobile_rec"
+    )
+    ocr_engine = PaddleOCR(lang="hi", ocr_version="PP-OCRv3")
+    assert ocr_engine._params["text_detection_model_name"] == "PP-OCRv3_mobile_det"
+    assert (
+        ocr_engine._params["text_recognition_model_name"]
+        == "devanagari_PP-OCRv3_mobile_rec"
+    )
+    ocr_engine = PaddleOCR(lang="japan", ocr_version="PP-OCRv3")
+    assert ocr_engine._params["text_detection_model_name"] == "PP-OCRv3_mobile_det"
+    assert (
+        ocr_engine._params["text_recognition_model_name"] == "japan_PP-OCRv3_mobile_rec"
+    )
+
+
+def test_pp_ocrv6_model_names():
+    engine = object.__new__(PaddleOCR)
+    for lang in ("ch", "chinese_cht", "en", "japan", "fr", "de", "vi", "ku", "az"):
+        det, rec = engine._get_ocr_model_names(lang, "PP-OCRv6")
+        assert det == "PP-OCRv6_medium_det"
+        assert rec == "PP-OCRv6_medium_rec"
+    det, rec = engine._get_ocr_model_names("pi", "PP-OCRv6")
+    assert det is None
+    assert rec is None
+    det, rec = engine._get_ocr_model_names("ru", "PP-OCRv6")
+    assert det is None
+    assert rec is None
+
+
+def test_default_ocr_model_names():
+    engine = object.__new__(PaddleOCR)
+    det, rec = engine._get_ocr_model_names(None, None)
+    assert det == "PP-OCRv6_medium_det"
+    assert rec == "PP-OCRv6_medium_rec"
+    det, rec = engine._get_ocr_model_names("ch", None)
+    assert det == "PP-OCRv6_medium_det"
+    assert rec == "PP-OCRv6_medium_rec"
+    det, rec = engine._get_ocr_model_names("fr", None)
+    assert det == "PP-OCRv6_medium_det"
+    assert rec == "PP-OCRv6_medium_rec"
+    det, rec = engine._get_ocr_model_names("ru", None)
+    assert det == "PP-OCRv5_server_det"
+    assert rec == "eslav_PP-OCRv5_mobile_rec"
+    det, rec = engine._get_ocr_model_names("az", None)
+    assert det == "PP-OCRv6_medium_det"
+    assert rec == "PP-OCRv6_medium_rec"
+    det, rec = engine._get_ocr_model_names("ku", None)
+    assert det == "PP-OCRv6_medium_det"
+    assert rec == "PP-OCRv6_medium_rec"
+    det, rec = engine._get_ocr_model_names("pi", None)
+    assert det == "PP-OCRv5_server_det"
+    assert rec == "latin_PP-OCRv5_mobile_rec"
+
+
+def test_unsupported_lang_version_raises():
+    with pytest.raises(ValueError, match="No models are available"):
+        PaddleOCR(lang="pi", ocr_version="PP-OCRv6")
+    with pytest.raises(ValueError, match="No models are available"):
+        PaddleOCR(lang="ru", ocr_version="PP-OCRv6")
diff --git a/tests/pipelines/test_pp_chatocrv4_doc.py b/tests/pipelines/test_pp_chatocrv4_doc.py
new file mode 100644
index 0000000..2f4b70a
--- /dev/null
+++ b/tests/pipelines/test_pp_chatocrv4_doc.py
@@ -0,0 +1,96 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+pytestmark = pytest.mark.py38_incompatible
+
+from paddleocr import PPChatOCRv4Doc
+from ..testing_utils import TEST_DATA_DIR
+
+
+@pytest.fixture(scope="module")
+def pp_chatocrv4_doc_pipeline():
+    return PPChatOCRv4Doc()
+
+
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "doc_with_formula.png",
+    ],
+)
+def test_visual_predict(pp_chatocrv4_doc_pipeline, image_path):
+    result = pp_chatocrv4_doc_pipeline.visual_predict(str(image_path))
+
+    assert result is not None
+    assert isinstance(result, list)
+    assert len(result) == 1
+    res = result[0]
+    assert isinstance(res, dict)
+    assert res.keys() == {"visual_info", "layout_parsing_result"}
+    assert isinstance(res["visual_info"], dict)
+    assert isinstance(res["layout_parsing_result"], dict)
+
+
+@pytest.mark.parametrize(
+    "params",
+    [
+        {"use_doc_orientation_classify": False},
+        {"use_doc_unwarping": False},
+        {"use_table_recognition": False},
+        {"layout_threshold": 0.88},
+        {"layout_threshold": [0.45, 0.4]},
+        {"layout_threshold": {0: 0.45, 2: 0.48, 7: 0.4}},
+        {"layout_nms": False},
+        {"layout_unclip_ratio": 1.1},
+        {"layout_unclip_ratio": [1.2, 1.5]},
+        {"layout_unclip_ratio": {0: 1.2, 2: 1.5, 7: 1.8}},
+        {"layout_merge_bboxes_mode": "large"},
+        {"layout_merge_bboxes_mode": {0: "large", 2: "small", 7: "union"}},
+        {"text_det_limit_side_len": 640, "text_det_limit_type": "min"},
+        {"text_det_thresh": 0.5},
+        {"text_det_box_thresh": 0.3},
+        {"text_det_unclip_ratio": 3.0},
+        {"text_rec_score_thresh": 0.5},
+    ],
+)
+def test_visual_predict_params(
+    monkeypatch,
+    pp_chatocrv4_doc_pipeline,
+    params,
+):
+    def _dummy_visual_predict(input, **params):
+        yield {"visual_info": {}, "layout_parsing_result": params}
+
+    monkeypatch.setattr(
+        pp_chatocrv4_doc_pipeline.paddlex_pipeline,
+        "visual_predict",
+        _dummy_visual_predict,
+    )
+
+    result = pp_chatocrv4_doc_pipeline.visual_predict(
+        input,
+        **params,
+    )
+
+    assert isinstance(result, list)
+    assert len(result) == 1
+    res = result[0]
+    res = res["layout_parsing_result"]
+    for k, v in params.items():
+        assert res[k] == v
+
+
+# TODO: Test constructor and other methods
diff --git a/tests/pipelines/test_pp_doctranslation.py b/tests/pipelines/test_pp_doctranslation.py
new file mode 100644
index 0000000..1ba060e
--- /dev/null
+++ b/tests/pipelines/test_pp_doctranslation.py
@@ -0,0 +1,96 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+pytestmark = pytest.mark.py38_incompatible
+
+from paddleocr import PPDocTranslation
+from ..testing_utils import TEST_DATA_DIR
+
+
+@pytest.fixture(scope="module")
+def pp_doctranslation_pipeline():
+    return PPDocTranslation()
+
+
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "book.jpg",
+    ],
+)
+def test_visual_predict(pp_doctranslation_pipeline, image_path):
+    result = pp_doctranslation_pipeline.visual_predict(str(image_path))
+
+    assert result is not None
+    assert isinstance(result, list)
+    assert len(result) == 1
+    res = result[0]
+    assert isinstance(res, dict)
+    assert res.keys() == {"layout_parsing_result"}
+    assert isinstance(res["layout_parsing_result"], dict)
+
+
+@pytest.mark.parametrize(
+    "params",
+    [
+        {"use_doc_orientation_classify": False},
+        {"use_doc_unwarping": False},
+        {"use_table_recognition": False},
+        {"use_formula_recognition": False},
+        {"layout_threshold": 0.88},
+        {"layout_threshold": [0.45, 0.4]},
+        {"layout_threshold": {0: 0.45, 2: 0.48, 7: 0.4}},
+        {"layout_nms": False},
+        {"layout_unclip_ratio": 1.1},
+        {"layout_unclip_ratio": [1.2, 1.5]},
+        {"layout_unclip_ratio": {0: 1.2, 2: 1.5, 7: 1.8}},
+        {"layout_merge_bboxes_mode": "large"},
+        {"layout_merge_bboxes_mode": {0: "large", 2: "small", 7: "union"}},
+        {"text_det_limit_side_len": 640, "text_det_limit_type": "min"},
+        {"text_det_thresh": 0.5},
+        {"text_det_box_thresh": 0.3},
+        {"text_det_unclip_ratio": 3.0},
+        {"text_rec_score_thresh": 0.5},
+    ],
+)
+def test_visual_predict_params(
+    monkeypatch,
+    pp_doctranslation_pipeline,
+    params,
+):
+    def _dummy_visual_predict(input, **params):
+        yield {"layout_parsing_result": params}
+
+    monkeypatch.setattr(
+        pp_doctranslation_pipeline.paddlex_pipeline,
+        "visual_predict",
+        _dummy_visual_predict,
+    )
+
+    result = pp_doctranslation_pipeline.visual_predict(
+        input,
+        **params,
+    )
+
+    assert isinstance(result, list)
+    assert len(result) == 1
+    res = result[0]
+    res = res["layout_parsing_result"]
+    for k, v in params.items():
+        assert res[k] == v
+
+
+# TODO: Test constructor and other methods
diff --git a/tests/pipelines/test_pp_structurev3.py b/tests/pipelines/test_pp_structurev3.py
new file mode 100644
index 0000000..c85b81e
--- /dev/null
+++ b/tests/pipelines/test_pp_structurev3.py
@@ -0,0 +1,88 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+pytestmark = pytest.mark.py38_incompatible
+
+from paddleocr import PPStructureV3
+from ..testing_utils import (
+    TEST_DATA_DIR,
+    check_simple_inference_result,
+    check_wrapper_simple_inference_param_forwarding,
+)
+
+
+@pytest.fixture(scope="module")
+def pp_structurev3_pipeline():
+    return PPStructureV3()
+
+
+@pytest.mark.resource_intensive
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "doc_with_formula.png",
+    ],
+)
+def test_predict(pp_structurev3_pipeline, image_path):
+    result = pp_structurev3_pipeline.predict(str(image_path))
+
+    check_simple_inference_result(result)
+    res = result[0]
+    overall_ocr_res = res["overall_ocr_res"]
+    assert len(overall_ocr_res["dt_polys"]) > 0
+    assert len(overall_ocr_res["rec_texts"]) > 0
+    assert len(overall_ocr_res["rec_polys"]) > 0
+    assert len(overall_ocr_res["rec_boxes"]) > 0
+
+
+@pytest.mark.parametrize(
+    "params",
+    [
+        {"use_doc_orientation_classify": False},
+        {"use_doc_unwarping": False},
+        {"use_table_recognition": False},
+        {"use_formula_recognition": False},
+        {"layout_threshold": 0.88},
+        {"layout_threshold": [0.45, 0.4]},
+        {"layout_threshold": {0: 0.45, 2: 0.48, 7: 0.4}},
+        {"layout_nms": False},
+        {"layout_unclip_ratio": 1.1},
+        {"layout_unclip_ratio": [1.2, 1.5]},
+        {"layout_unclip_ratio": {0: 1.2, 2: 1.5, 7: 1.8}},
+        {"layout_merge_bboxes_mode": "large"},
+        {"layout_merge_bboxes_mode": {0: "large", 2: "small", 7: "union"}},
+        {"text_det_limit_side_len": 640, "text_det_limit_type": "min"},
+        {"text_det_thresh": 0.5},
+        {"text_det_box_thresh": 0.3},
+        {"text_det_unclip_ratio": 3.0},
+        {"text_rec_score_thresh": 0.5},
+    ],
+)
+def test_predict_params(
+    monkeypatch,
+    pp_structurev3_pipeline,
+    params,
+):
+    check_wrapper_simple_inference_param_forwarding(
+        monkeypatch,
+        pp_structurev3_pipeline,
+        "paddlex_pipeline",
+        "dummy_path",
+        params,
+    )
+
+
+# TODO: Test constructor and other methods
diff --git a/tests/pipelines/test_seal_rec.py b/tests/pipelines/test_seal_rec.py
new file mode 100644
index 0000000..145fdd9
--- /dev/null
+++ b/tests/pipelines/test_seal_rec.py
@@ -0,0 +1,86 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+pytestmark = pytest.mark.py38_incompatible
+
+from paddleocr import SealRecognition
+from ..testing_utils import (
+    TEST_DATA_DIR,
+    check_simple_inference_result,
+    check_wrapper_simple_inference_param_forwarding,
+)
+
+
+@pytest.fixture(scope="module")
+def ocr_engine() -> SealRecognition:
+    return SealRecognition()
+
+
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "seal.png",
+    ],
+)
+def test_predict(ocr_engine: SealRecognition, image_path: str) -> None:
+    """
+    Test PaddleOCR's seal recognition functionality.
+
+    Args:
+        ocr_engine: An instance of `SealRecognition`.
+        image_path: Path to the image to be processed.
+    """
+    result = ocr_engine.predict(str(image_path))
+
+    check_simple_inference_result(result)
+    res = result[0]["seal_res_list"][0]
+    assert len(res["dt_polys"]) > 0
+    assert isinstance(res["rec_texts"], list)
+    assert len(res["rec_texts"]) > 0
+    for text in res["rec_texts"]:
+        assert isinstance(text, str)
+
+
+@pytest.mark.parametrize(
+    "params",
+    [
+        {"use_doc_orientation_classify": False, "use_doc_unwarping": False},
+        {"use_layout_detection": False},
+        {"layout_det_res": None},
+        {"layout_threshold": 0.5},
+        {"layout_nms": False},
+        {"layout_unclip_ratio": 1.0},
+        {"layout_merge_bboxes_mode": "large"},
+        {"seal_det_limit_side_len": 736},
+        {"seal_det_limit_type": "min"},
+        {"seal_det_thresh": 0.5},
+        {"seal_det_box_thresh": 0.6},
+        {"seal_det_unclip_ratio": 0.5},
+        {"seal_rec_score_thresh": 0.05},
+    ],
+)
+def test_predict_params(
+    monkeypatch,
+    ocr_engine: SealRecognition,
+    params: dict,
+) -> None:
+    check_wrapper_simple_inference_param_forwarding(
+        monkeypatch,
+        ocr_engine,
+        "paddlex_pipeline",
+        "dummy_path",
+        params,
+    )
diff --git a/tests/pipelines/test_table_recognition_v2.py b/tests/pipelines/test_table_recognition_v2.py
new file mode 100644
index 0000000..58fdbda
--- /dev/null
+++ b/tests/pipelines/test_table_recognition_v2.py
@@ -0,0 +1,80 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import pytest
+
+pytestmark = pytest.mark.py38_incompatible
+
+from paddleocr import TableRecognitionPipelineV2
+from ..testing_utils import (
+    TEST_DATA_DIR,
+    check_simple_inference_result,
+    check_wrapper_simple_inference_param_forwarding,
+)
+
+
+@pytest.fixture(scope="module")
+def table_recognition_v2_pipeline():
+    return TableRecognitionPipelineV2()
+
+
+@pytest.mark.parametrize(
+    "image_path",
+    [
+        TEST_DATA_DIR / "table.jpg",
+    ],
+)
+def test_predict(table_recognition_v2_pipeline, image_path):
+    result = table_recognition_v2_pipeline.predict(
+        str(image_path), use_doc_orientation_classify=False, use_doc_unwarping=False
+    )
+
+    check_simple_inference_result(result)
+    res = result[0]
+    assert len(res["table_res_list"]) > 0
+    assert isinstance(res["table_res_list"][0], dict)
+    assert len(res["table_res_list"][0]["cell_box_list"]) > 0
+    assert isinstance(res["table_res_list"][0]["pred_html"], str)
+    assert isinstance(res["table_res_list"][0]["table_ocr_pred"], dict)
+
+
+@pytest.mark.parametrize(
+    "params",
+    [
+        {"use_doc_orientation_classify": False},
+        {"use_doc_unwarping": False},
+        {"use_layout_detection": False},
+        {"use_ocr_model": False},
+        {"text_det_limit_side_len": 640, "text_det_limit_type": "min"},
+        {"text_det_thresh": 0.5},
+        {"text_det_box_thresh": 0.3},
+        {"text_det_unclip_ratio": 3.0},
+        {"text_rec_score_thresh": 0.5},
+    ],
+)
+def test_predict_params(
+    monkeypatch,
+    table_recognition_v2_pipeline,
+    params,
+):
+    check_wrapper_simple_inference_param_forwarding(
+        monkeypatch,
+        table_recognition_v2_pipeline,
+        "paddlex_pipeline",
+        "dummy_path",
+        params,
+    )
+
+
+# TODO: Test constructor and other methods
diff --git a/tests/ppocr/test_cls_postprocess.py b/tests/ppocr/test_cls_postprocess.py
new file mode 100644
index 0000000..f65039c
--- /dev/null
+++ b/tests/ppocr/test_cls_postprocess.py
@@ -0,0 +1,72 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import numpy as np
+import paddle
+import pytest
+
+from ppocr.postprocess.cls_postprocess import ClsPostProcess
+
+
+# Fixtures for common test inputs
+@pytest.fixture
+def preds_tensor():
+    return paddle.to_tensor(np.array([[0.1, 0.7, 0.2], [0.3, 0.3, 0.4]]))
+
+
+@pytest.fixture
+def label_list():
+    return {0: "class0", 1: "class1", 2: "class2"}
+
+
+# Parameterize tests to cover multiple scenarios
+@pytest.mark.parametrize(
+    "label_list, expected",
+    [
+        ({0: "class0", 1: "class1", 2: "class2"}, [("class1", 0.7), ("class2", 0.4)]),
+        (None, [(1, 0.7), (2, 0.4)]),
+    ],
+)
+def test_cls_post_process_with_and_without_label_list(
+    preds_tensor, label_list, expected
+):
+    post_process = ClsPostProcess(label_list=label_list)
+    result = post_process(preds_tensor)
+    assert isinstance(result, list), "Result should be a list"
+    assert result == expected, f"Expected {expected}, got {result}"
+
+
+# Test with a key in the prediction dictionary
+def test_cls_post_process_with_key(preds_tensor, label_list):
+    preds_dict = {"key": preds_tensor}
+    post_process = ClsPostProcess(label_list=label_list, key="key")
+    result = post_process(preds_dict)
+    expected = [("class1", 0.7), ("class2", 0.4)]
+    assert isinstance(result, list), "Result should be a list"
+    assert result == expected, f"Expected {expected}, got {result}"
+
+
+# Test with label input
+def test_cls_post_process_with_label(preds_tensor, label_list):
+    labels = [2, 0]
+    post_process = ClsPostProcess(label_list=label_list)
+    result, label_result = post_process(preds_tensor, labels)
+    expected_result = [("class1", 0.7), ("class2", 0.4)]
+    expected_label_result = [("class2", 1.0), ("class0", 1.0)]
+    assert isinstance(result, list), "Result should be a list"
+    assert result == expected_result, f"Expected {expected_result}, got {result}"
+    assert isinstance(label_result, list), "Label result should be a list"
+    assert (
+        label_result == expected_label_result
+    ), f"Expected {expected_label_result}, got {label_result}"
diff --git a/tests/ppocr/test_formula_model.py b/tests/ppocr/test_formula_model.py
new file mode 100644
index 0000000..8705573
--- /dev/null
+++ b/tests/ppocr/test_formula_model.py
@@ -0,0 +1,202 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle
+import pytest
+
+from ppocr.modeling.backbones.rec_donut_swin import DonutSwinModel, DonutSwinModelOutput
+from ppocr.modeling.backbones.rec_pphgnetv2 import PPHGNetV2_B4_Formula
+from ppocr.modeling.backbones.rec_vary_vit import Vary_VIT_B_Formula
+from ppocr.modeling.heads.rec_unimernet_head import UniMERNetHead
+from ppocr.modeling.heads.rec_ppformulanet_head import PPFormulaNet_Head
+
+
+@pytest.fixture
+def sample_image():
+    return paddle.randn([1, 1, 192, 672])
+
+
+@pytest.fixture
+def sample_image_ppformulanet_s():
+    return paddle.randn([1, 1, 384, 384])
+
+
+@pytest.fixture
+def sample_image_ppformulanet_l():
+    return paddle.randn([1, 1, 768, 768])
+
+
+@pytest.fixture
+def encoder_feat():
+    encoded_feat = paddle.randn([1, 126, 1024])
+    return DonutSwinModelOutput(
+        last_hidden_state=encoded_feat,
+    )
+
+
+@pytest.fixture
+def encoder_feat_ppformulanet_s():
+    encoded_feat = paddle.randn([1, 144, 2048])
+    return DonutSwinModelOutput(
+        last_hidden_state=encoded_feat,
+    )
+
+
+@pytest.fixture
+def encoder_feat_ppformulanet_l():
+    encoded_feat = paddle.randn([1, 144, 1024])
+    return DonutSwinModelOutput(
+        last_hidden_state=encoded_feat,
+    )
+
+
+def test_unimernet_backbone(sample_image):
+    """
+    Test UniMERNet backbone.
+
+    Args:
+        sample_image: sample image to be processed.
+    """
+    backbone = DonutSwinModel(
+        hidden_size=1024,
+        num_layers=4,
+        num_heads=[4, 8, 16, 32],
+        add_pooling_layer=True,
+        use_mask_token=False,
+    )
+    backbone.eval()
+    with paddle.no_grad():
+        result = backbone(sample_image)
+        encoder_feat = result[0]
+        assert encoder_feat.shape == [1, 126, 1024]
+
+
+def test_unimernet_head(encoder_feat):
+    """
+    Test UniMERNet head.
+
+    Args:
+        encoder_feat: encoder feature from unimernet backbone.
+    """
+    head = UniMERNetHead(
+        max_new_tokens=5,
+        decoder_start_token_id=0,
+        temperature=0.2,
+        do_sample=False,
+        top_p=0.95,
+        encoder_hidden_size=1024,
+        is_export=False,
+        length_aware=True,
+    )
+
+    head.eval()
+    with paddle.no_grad():
+        result = head(encoder_feat)
+        assert result.shape == [1, 6]
+
+
+def test_ppformulanet_s_backbone(sample_image_ppformulanet_s):
+    """
+    Test PP-FormulaNet-S backbone.
+
+    Args:
+        sample_image_ppformulanet_s: sample image to be processed.
+    """
+    backbone = PPHGNetV2_B4_Formula(
+        class_num=1024,
+    )
+    backbone.eval()
+    with paddle.no_grad():
+        result = backbone(sample_image_ppformulanet_s)
+        encoder_feat = result[0]
+        assert encoder_feat.shape == [1, 144, 2048]
+
+
+def test_ppformulanet_s_head(encoder_feat_ppformulanet_s):
+    """
+    Test PP-FormulaNet-S head.
+
+    Args:
+        encoder_feat_ppformulanet_s: encoder feature from PP-FormulaNet-S backbone.
+    """
+    head = PPFormulaNet_Head(
+        max_new_tokens=6,
+        decoder_start_token_id=0,
+        decoder_ffn_dim=1536,
+        decoder_hidden_size=384,
+        decoder_layers=2,
+        temperature=0.2,
+        do_sample=False,
+        top_p=0.95,
+        encoder_hidden_size=2048,
+        is_export=False,
+        length_aware=True,
+        use_parallel=True,
+        parallel_step=3,
+    )
+
+    head.eval()
+    with paddle.no_grad():
+        result = head(encoder_feat_ppformulanet_s)
+        assert result.shape == [1, 9]
+
+
+def test_ppformulanet_l_backbone(sample_image_ppformulanet_l):
+    """
+    Test PP-FormulaNet-L backbone.
+
+    Args:
+        sample_image_ppformulanet_l: sample image to be processed.
+    """
+    backbone = Vary_VIT_B_Formula(
+        image_size=768,
+        encoder_embed_dim=768,
+        encoder_depth=12,
+        encoder_num_heads=12,
+        encoder_global_attn_indexes=[2, 5, 8, 11],
+    )
+    backbone.eval()
+    with paddle.no_grad():
+        result = backbone(sample_image_ppformulanet_l)
+        encoder_feat = result[0]
+        assert encoder_feat.shape == [1, 144, 1024]
+
+
+def test_ppformulanet_l_head(encoder_feat_ppformulanet_l):
+    """
+    Test PP-FormulaNet-L head.
+
+    Args:
+        encoder_feat_ppformulanet_l: encoder feature from PP-FormulaNet-L Head.
+    """
+    head = PPFormulaNet_Head(
+        max_new_tokens=6,
+        decoder_start_token_id=0,
+        decoder_ffn_dim=2048,
+        decoder_hidden_size=512,
+        decoder_layers=8,
+        temperature=0.2,
+        do_sample=False,
+        top_p=0.95,
+        encoder_hidden_size=1024,
+        is_export=False,
+        length_aware=False,
+        use_parallel=False,
+        parallel_step=0,
+    )
+
+    head.eval()
+    with paddle.no_grad():
+        result = head(encoder_feat_ppformulanet_l)
+        assert result.shape == [1, 7]
diff --git a/tests/ppocr/test_iaa_augment.py b/tests/ppocr/test_iaa_augment.py
new file mode 100644
index 0000000..250da1a
--- /dev/null
+++ b/tests/ppocr/test_iaa_augment.py
@@ -0,0 +1,211 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import random
+import sys
+
+import numpy as np
+import pytest
+
+pytestmark = pytest.mark.py38_incompatible
+# py3.8 lmdb wheel exposes an undefined Py_SET_REFCNT symbol and fails to
+# load. Skip collection only on py3.8; on py3.9+ a lmdb import failure is
+# a real regression and must surface rather than be silently skipped.
+if sys.version_info < (3, 9):
+    pytest.importorskip("lmdb")
+
+from ppocr.data.imaug.iaa_augment import IaaAugment
+
+# Set a fixed random seed to ensure test reproducibility
+np.random.seed(42)
+random.seed(42)
+
+
+# Fixture to provide a sample image for tests
+@pytest.fixture
+def sample_image():
+    # Create a 100x100 pixel dummy image with 3 color channels (RGB)
+    return np.random.randint(0, 256, (100, 100, 3), dtype=np.uint8)
+
+
+# Fixture to provide sample polygons for tests
+@pytest.fixture
+def sample_polys():
+    # Create dummy polygons as sample data
+    polys = [
+        np.array([[10, 10], [20, 10], [20, 20], [10, 20]], dtype=np.float32),
+        np.array([[30, 30], [40, 30], [40, 40], [30, 40]], dtype=np.float32),
+    ]
+    return polys
+
+
+# Helper function to create a data dictionary for testing
+def create_data(sample_image, sample_polys):
+    return {
+        "image": sample_image.copy(),
+        "polys": [poly.copy() for poly in sample_polys],
+    }
+
+
+# Test the default behavior of the augmenter (without specified arguments)
+def test_iaa_augment_default(sample_image, sample_polys):
+    data = create_data(sample_image, sample_polys)
+    augmenter = IaaAugment()
+    transformed_data = augmenter(data)
+
+    # Check the data types and structure of the transformed image and polygons
+    assert isinstance(
+        transformed_data["image"], np.ndarray
+    ), "Image should be a numpy array"
+    assert isinstance(
+        transformed_data["polys"], np.ndarray
+    ), "Polys should be a numpy array"
+    assert transformed_data["image"].ndim == 3, "Image should be 3-dimensional"
+
+    # Verify that the polygons have been transformed
+    polys_changed = any(
+        not np.allclose(orig_poly, trans_poly)
+        for orig_poly, trans_poly in zip(sample_polys, transformed_data["polys"])
+    )
+    assert polys_changed, "Polygons should have been transformed"
+
+
+# Test the augmenter with empty arguments, meaning no transformations should occur
+def test_iaa_augment_none(sample_image, sample_polys):
+    data = create_data(sample_image, sample_polys)
+    augmenter = IaaAugment(augmenter_args=[])
+    transformed_data = augmenter(data)
+
+    # Check that the image and polygons remain unchanged
+    assert np.array_equal(
+        data["image"], transformed_data["image"]
+    ), "Image should be unchanged"
+    for orig_poly, transformed_poly in zip(data["polys"], transformed_data["polys"]):
+        assert np.array_equal(
+            orig_poly, transformed_poly
+        ), "Polygons should be unchanged"
+
+
+# Parameterized test to check various augmenter arguments and expected image shapes
+@pytest.mark.parametrize(
+    "augmenter_args, expected_shape",
+    [
+        ([], (100, 100, 3)),
+        ([{"type": "Resize", "args": {"size": [0.5, 0.5]}}], (50, 50, 3)),
+        ([{"type": "Resize", "args": {"size": [2.0, 2.0]}}], (200, 200, 3)),
+    ],
+)
+def test_iaa_augment_resize(sample_image, sample_polys, augmenter_args, expected_shape):
+    data = create_data(sample_image, sample_polys)
+    augmenter = IaaAugment(augmenter_args=augmenter_args)
+    transformed_data = augmenter(data)
+
+    # Verify that the transformed image has the expected shape
+    assert (
+        transformed_data["image"].shape == expected_shape
+    ), f"Expected image shape {expected_shape}, got {transformed_data['image'].shape}"
+
+
+# Test custom augmenter arguments with specific transformations
+def test_iaa_augment_custom(sample_image, sample_polys):
+    data = create_data(sample_image, sample_polys)
+    augmenter_args = [
+        {"type": "Affine", "args": {"rotate": [45, 45]}},  # Apply 45-degree rotation
+        {"type": "Resize", "args": {"size": [0.5, 0.5]}},
+    ]
+    augmenter = IaaAugment(augmenter_args=augmenter_args)
+    transformed_data = augmenter(data)
+
+    # Check the expected image dimensions after resizing
+    expected_height = int(sample_image.shape[0] * 0.5)
+    expected_width = int(sample_image.shape[1] * 0.5)
+    assert (
+        transformed_data["image"].shape[0] == expected_height
+    ), "Image height should be scaled by 0.5"
+    assert (
+        transformed_data["image"].shape[1] == expected_width
+    ), "Image width should be scaled by 0.5"
+
+    # Verify that the polygons have been transformed
+    polys_changed = any(
+        not np.allclose(orig_poly, trans_poly)
+        for orig_poly, trans_poly in zip(sample_polys, transformed_data["polys"])
+    )
+    assert polys_changed, "Polygons should have been transformed"
+
+
+def test_iaa_augment_affine_fit_output(sample_image, sample_polys):
+    data = create_data(sample_image, sample_polys)
+    augmenter_args = [
+        {"type": "Affine", "args": {"rotate": [45, 45], "fit_output": True}},
+    ]
+    augmenter = IaaAugment(augmenter_args=augmenter_args)
+    transformed_data = augmenter(data)
+
+    assert transformed_data["image"].shape[0] > sample_image.shape[0]
+    assert transformed_data["image"].shape[1] > sample_image.shape[1]
+
+
+# Test that an unknown transformation type raises an AttributeError
+def test_iaa_augment_unknown_transform():
+    augmenter_args = [{"type": "UnknownTransform", "args": {}}]
+    with pytest.raises(AttributeError):
+        IaaAugment(augmenter_args=augmenter_args)
+
+
+# Test that an invalid resize size parameter raises a ValueError
+def test_iaa_augment_invalid_resize_size(sample_image, sample_polys):
+    augmenter_args = [{"type": "Resize", "args": {"size": "invalid_size"}}]
+    with pytest.raises(ValueError) as exc_info:
+        IaaAugment(augmenter_args=augmenter_args)
+    assert "'size' must be a list or tuple of two numbers" in str(exc_info.value)
+
+
+# Test that polygons are transformed as expected
+def test_iaa_augment_polys_transformation(sample_image, sample_polys):
+    data = create_data(sample_image, sample_polys)
+    augmenter_args = [
+        {"type": "Affine", "args": {"rotate": [90, 90]}},  # Apply 90-degree rotation
+    ]
+    augmenter = IaaAugment(augmenter_args=augmenter_args)
+    transformed_data = augmenter(data)
+
+    # Verify that the polygons have been transformed
+    polys_changed = any(
+        not np.allclose(orig_poly, trans_poly)
+        for orig_poly, trans_poly in zip(sample_polys, transformed_data["polys"])
+    )
+    assert polys_changed, "Polygons should have been transformed"
+
+
+# Test multiple transformations applied to the augmenter
+def test_iaa_augment_multiple_transforms(sample_image, sample_polys):
+    augmenter_args = [
+        {"type": "Fliplr", "args": {"p": 1.0}},  # Always apply horizontal flip
+        {"type": "Affine", "args": {"shear": 10}},
+    ]
+    data = create_data(sample_image, sample_polys)
+    augmenter = IaaAugment(augmenter_args=augmenter_args)
+    transformed_data = augmenter(data)
+
+    # Ensure the image has been transformed
+    images_different = not np.array_equal(transformed_data["image"], sample_image)
+    assert images_different, "Image should be transformed"
+
+    # Ensure the polygons have been transformed
+    polys_changed = any(
+        not np.allclose(orig_poly, trans_poly)
+        for orig_poly, trans_poly in zip(sample_polys, transformed_data["polys"])
+    )
+    assert polys_changed, "Polygons should have been transformed"
diff --git a/tests/ppocr/test_rec_postprocess.py b/tests/ppocr/test_rec_postprocess.py
new file mode 100644
index 0000000..41b951e
--- /dev/null
+++ b/tests/ppocr/test_rec_postprocess.py
@@ -0,0 +1,149 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import importlib.util
+from pathlib import Path
+
+import numpy as np
+import pytest
+
+REPO_ROOT = Path(__file__).resolve().parents[2]
+
+
+def _load_base_rec_label_decode():
+    spec = importlib.util.spec_from_file_location(
+        "ppocr.postprocess.rec_postprocess",
+        REPO_ROOT / "ppocr" / "postprocess" / "rec_postprocess.py",
+    )
+    module = importlib.util.module_from_spec(spec)
+    assert spec.loader is not None
+    spec.loader.exec_module(module)
+    return module.BaseRecLabelDecode
+
+
+BaseRecLabelDecode = _load_base_rec_label_decode()
+
+
+class TestBaseRecLabelDecode:
+    """Tests for BaseRecLabelDecode.get_word_info()."""
+
+    @pytest.fixture
+    def decoder(self):
+        return BaseRecLabelDecode()
+
+    def test_get_word_info_with_german_accented_chars(self, decoder):
+        text = "Grüßen"
+        selection = np.ones(len(text), dtype=bool)
+        word_list, _, state_list = decoder.get_word_info(text, selection)
+        assert len(word_list) == 1
+        assert "".join(word_list[0]) == "Grüßen"
+        assert state_list[0] == "en&num"
+
+    def test_get_word_info_with_longer_german_word(self, decoder):
+        text = "ungewöhnlichen"
+        selection = np.ones(len(text), dtype=bool)
+        word_list, _, state_list = decoder.get_word_info(text, selection)
+        assert len(word_list) == 1
+        assert "".join(word_list[0]) == "ungewöhnlichen"
+        assert state_list[0] == "en&num"
+
+    def test_get_word_info_with_french_accented_chars(self, decoder):
+        text = "café"
+        selection = np.ones(len(text), dtype=bool)
+        word_list, _, state_list = decoder.get_word_info(text, selection)
+        assert len(word_list) == 1
+        assert "".join(word_list[0]) == "café"
+
+    def test_get_word_info_underscore_as_splitter(self, decoder):
+        text = "hello_world"
+        selection = np.ones(len(text), dtype=bool)
+        word_list, _, state_list = decoder.get_word_info(text, selection)
+        assert len(word_list) == 2
+        assert "".join(word_list[0]) == "hello"
+        assert "".join(word_list[1]) == "world"
+
+    def test_get_word_info_with_mixed_content(self, decoder):
+        text = "Grüßen Sie"
+        selection = np.ones(len(text), dtype=bool)
+        word_list, _, state_list = decoder.get_word_info(text, selection)
+        assert len(word_list) == 2
+        assert "".join(word_list[0]) == "Grüßen"
+        assert "".join(word_list[1]) == "Sie"
+
+    def test_get_word_info_with_french_apostrophe(self, decoder):
+        text = "n'êtes"
+        selection = np.ones(len(text), dtype=bool)
+        word_list, _, state_list = decoder.get_word_info(text, selection)
+        assert len(word_list) == 1
+        assert "".join(word_list[0]) == "n'êtes"
+
+    @pytest.mark.parametrize(
+        "text,expected_word_count,expected_joined_words",
+        [
+            ("été", 1, ["été"]),
+            ("français", 1, ["français"]),
+            ("élève", 1, ["élève"]),
+            ("à demain", 2, ["à", "demain"]),
+        ],
+    )
+    def test_get_word_info_french_accented_words(
+        self, decoder, text, expected_word_count, expected_joined_words
+    ):
+        selection = np.ones(len(text), dtype=bool)
+        word_list, _, state_list = decoder.get_word_info(text, selection)
+        assert len(word_list) == expected_word_count
+        assert ["".join(word) for word in word_list] == expected_joined_words
+        assert all(state == "en&num" for state in state_list)
+
+    def test_get_word_info_french_complex_sentence(self, decoder):
+        text = "C'était très français"
+        selection = np.ones(len(text), dtype=bool)
+        word_list, _, state_list = decoder.get_word_info(text, selection)
+        assert ["".join(word) for word in word_list] == [
+            "C'était",
+            "très",
+            "français",
+        ]
+        assert state_list == ["en&num", "en&num", "en&num"]
+
+    def test_get_word_info_with_ascii_only(self, decoder):
+        text = "hello world"
+        selection = np.ones(len(text), dtype=bool)
+        word_list, _, state_list = decoder.get_word_info(text, selection)
+        assert len(word_list) == 2
+        assert "".join(word_list[0]) == "hello"
+        assert "".join(word_list[1]) == "world"
+
+    def test_get_word_info_with_numbers(self, decoder):
+        text = "VGG-16"
+        selection = np.ones(len(text), dtype=bool)
+        word_list, _, state_list = decoder.get_word_info(text, selection)
+        assert len(word_list) == 1
+        assert "".join(word_list[0]) == "VGG-16"
+
+    def test_get_word_info_with_floating_point(self, decoder):
+        text = "price 3.14"
+        selection = np.ones(len(text), dtype=bool)
+        word_list, _, state_list = decoder.get_word_info(text, selection)
+        assert len(word_list) == 2
+        assert "".join(word_list[0]) == "price"
+        assert "".join(word_list[1]) == "3.14"
+
+    def test_get_word_info_with_chinese(self, decoder):
+        text = "你好啊"
+        selection = np.ones(len(text), dtype=bool)
+        word_list, _, state_list = decoder.get_word_info(text, selection)
+        assert len(word_list) == 1
+        assert "".join(word_list[0]) == "你好啊"
+        assert state_list[0] == "cn"
diff --git a/tests/security/test_latexocr_pickle.py b/tests/security/test_latexocr_pickle.py
new file mode 100644
index 0000000..59fe54a
--- /dev/null
+++ b/tests/security/test_latexocr_pickle.py
@@ -0,0 +1,138 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import importlib.util
+import io
+import pickle
+import sys
+import types
+from pathlib import Path
+
+import numpy as np
+import pytest
+
+
+def _load_latexocr_module(monkeypatch):
+    module_name = "ppocr.data.latexocr_dataset"
+    module_path = (
+        Path(__file__).resolve().parents[2] / "ppocr" / "data" / "latexocr_dataset.py"
+    )
+
+    ppocr_pkg = types.ModuleType("ppocr")
+    ppocr_pkg.__path__ = []
+    data_pkg = types.ModuleType("ppocr.data")
+    data_pkg.__path__ = []
+    imaug_pkg = types.ModuleType("ppocr.data.imaug")
+    imaug_pkg.__path__ = []
+    label_ops_mod = types.ModuleType("ppocr.data.imaug.label_ops")
+
+    class _Tokenizer:
+        def __init__(self, *_args, **_kwargs):
+            pass
+
+    label_ops_mod.LatexOCRLabelEncode = _Tokenizer
+    imaug_pkg.transform = lambda data, ops: data
+    imaug_pkg.create_operators = lambda transforms, global_config: []
+
+    paddle_mod = types.ModuleType("paddle")
+    paddle_io_mod = types.ModuleType("paddle.io")
+
+    class _Dataset:
+        pass
+
+    paddle_io_mod.Dataset = _Dataset
+    paddle_mod.io = paddle_io_mod
+    paddle_mod.arange = lambda n: list(range(n))
+    paddle_mod.randperm = lambda n: list(range(n))
+
+    for name, module in {
+        "ppocr": ppocr_pkg,
+        "ppocr.data": data_pkg,
+        "ppocr.data.imaug": imaug_pkg,
+        "ppocr.data.imaug.label_ops": label_ops_mod,
+        "paddle": paddle_mod,
+        "paddle.io": paddle_io_mod,
+        "cv2": types.ModuleType("cv2"),
+    }.items():
+        monkeypatch.setitem(sys.modules, name, module)
+
+    spec = importlib.util.spec_from_file_location(module_name, module_path)
+    module = importlib.util.module_from_spec(spec)
+    assert spec.loader is not None
+    monkeypatch.setitem(sys.modules, module_name, module)
+    spec.loader.exec_module(module)
+    return module
+
+
+class _Payload:
+    def __init__(self, marker_path: Path):
+        self.marker_path = marker_path
+
+    def __reduce__(self):
+        expression = (
+            f"(__import__('pathlib').Path({self.marker_path.as_posix()!r}).write_text('pwned'), "
+            "{(32, 32): [('eq', 'img.png')]})[1]"
+        )
+        return (eval, (expression,))
+
+
+@pytest.fixture
+def latexocr_config(tmp_path):
+    return {
+        "Global": {
+            "max_seq_len": 16,
+            "rec_char_dict_path": str(tmp_path / "dict.txt"),
+        },
+        "Train": {
+            "dataset": {
+                "data": str(tmp_path / "dataset.pkl"),
+                "data_dir": str(tmp_path),
+                "min_dimensions": [1, 1],
+                "max_dimensions": [128, 128],
+                "batch_size_per_pair": 1,
+                "keep_smaller_batches": True,
+                "transforms": [],
+            },
+            "loader": {"shuffle": False},
+        },
+    }
+
+
+def test_latexocr_dataset_rejects_executable_pickle_payload(
+    tmp_path, latexocr_config, monkeypatch
+):
+    module = _load_latexocr_module(monkeypatch)
+    marker_path = tmp_path / "marker.txt"
+    payload_path = Path(latexocr_config["Train"]["dataset"]["data"])
+    payload_path.write_bytes(pickle.dumps(_Payload(marker_path)))
+
+    with pytest.raises(pickle.UnpicklingError):
+        module.LaTeXOCRDataSet(latexocr_config, "Train", logger=types.SimpleNamespace())
+
+    assert not marker_path.exists()
+
+
+def test_latexocr_dataset_accepts_basic_dict_payload(
+    tmp_path, latexocr_config, monkeypatch
+):
+    module = _load_latexocr_module(monkeypatch)
+    payload_path = Path(latexocr_config["Train"]["dataset"]["data"])
+    payload_path.write_bytes(pickle.dumps({(32, 32): [("eq", "img.png")]}))
+
+    dataset = module.LaTeXOCRDataSet(
+        latexocr_config, "Train", logger=types.SimpleNamespace()
+    )
+
+    assert dataset.size == 1
+    assert dataset.data == {(32, 32): [("eq", "img.png")]}
diff --git a/tests/security/test_lmdb_pickle.py b/tests/security/test_lmdb_pickle.py
new file mode 100644
index 0000000..de5231d
--- /dev/null
+++ b/tests/security/test_lmdb_pickle.py
@@ -0,0 +1,150 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import importlib.util
+import os
+import pickle
+import sys
+import types
+from pathlib import Path
+
+import pytest
+
+REPO_ROOT = Path(__file__).resolve().parents[2]
+
+
+def _stub_module(monkeypatch, name, **attrs):
+    module = types.ModuleType(name)
+    for key, value in attrs.items():
+        setattr(module, key, value)
+    monkeypatch.setitem(sys.modules, name, module)
+    return module
+
+
+def _load_lmdb_dataset_module(monkeypatch, lmdb_open):
+    _stub_module(monkeypatch, "cv2")
+    _stub_module(
+        monkeypatch, "numpy", random=types.SimpleNamespace(randint=lambda upper: 0)
+    )
+    _stub_module(monkeypatch, "lmdb", open=lmdb_open)
+    _stub_module(monkeypatch, "PIL")
+    _stub_module(monkeypatch, "PIL.Image")
+
+    paddle_module = _stub_module(monkeypatch, "paddle")
+    paddle_io_module = _stub_module(
+        monkeypatch, "paddle.io", Dataset=type("Dataset", (), {})
+    )
+    paddle_module.io = paddle_io_module
+
+    ppocr_module = _stub_module(monkeypatch, "ppocr")
+    ppocr_data_module = _stub_module(monkeypatch, "ppocr.data")
+    ppocr_data_module.__path__ = []
+    ppocr_module.data = ppocr_data_module
+    _stub_module(
+        monkeypatch,
+        "ppocr.data.imaug",
+        transform=lambda data, ops: data,
+        create_operators=lambda *args, **kwargs: [],
+    )
+
+    spec = importlib.util.spec_from_file_location(
+        "ppocr.data.lmdb_dataset", REPO_ROOT / "ppocr" / "data" / "lmdb_dataset.py"
+    )
+    module = importlib.util.module_from_spec(spec)
+    monkeypatch.setitem(sys.modules, "ppocr.data.lmdb_dataset", module)
+    spec.loader.exec_module(module)
+    return module
+
+
+class _Exploit:
+    def __reduce__(self):
+        return (os.system, ("echo SHOULD_NOT_RUN > /tmp/paddleocr_lmdb_pickle_test",))
+
+
+class _Txn:
+    def __init__(self, values):
+        self._values = values
+
+    def get(self, key):
+        return self._values.get(key)
+
+
+class _Env:
+    def __init__(self, txn):
+        self._txn = txn
+
+    def begin(self, write=False):
+        return self._txn
+
+
+@pytest.fixture(autouse=True)
+def _cleanup_marker():
+    marker = Path("/tmp/paddleocr_lmdb_pickle_test")
+    marker.unlink(missing_ok=True)
+    yield
+    marker.unlink(missing_ok=True)
+
+
+def test_tablemaster_sample_info_rejects_pickle_rce_payload(monkeypatch):
+    module = _load_lmdb_dataset_module(
+        monkeypatch, lmdb_open=lambda *args, **kwargs: None
+    )
+    payload = pickle.dumps(_Exploit())
+    txn = _Txn({b"1": payload})
+    dataset = module.LMDBDataSetTableMaster.__new__(module.LMDBDataSetTableMaster)
+
+    assert dataset.get_lmdb_sample_info(txn, 1) is None
+    assert not Path("/tmp/paddleocr_lmdb_pickle_test").exists()
+
+
+def test_tablemaster_sample_info_accepts_expected_basic_pickle_data(monkeypatch):
+    module = _load_lmdb_dataset_module(
+        monkeypatch, lmdb_open=lambda *args, **kwargs: None
+    )
+    safe_payload = pickle.dumps(("sample.png", b"img-bytes", "raw-name\ntext\n1,2,3,4"))
+    txn = _Txn({b"1": safe_payload})
+    dataset = module.LMDBDataSetTableMaster.__new__(module.LMDBDataSetTableMaster)
+
+    sample = dataset.get_lmdb_sample_info(txn, 1)
+
+    assert sample["file_name"] == "sample.png"
+    assert sample["image"] == b"img-bytes"
+    assert sample["structure"] == ["text"]
+    assert sample["cells"] == [{"bbox": [1, 2, 3, 4], "tokens": ["1", "2"]}]
+
+
+def test_tablemaster_length_metadata_rejects_pickle_rce_payload(monkeypatch):
+    payload = pickle.dumps(_Exploit())
+    txn = _Txn({b"__len__": payload})
+    module = _load_lmdb_dataset_module(
+        monkeypatch, lmdb_open=lambda *args, **kwargs: _Env(txn)
+    )
+    dataset = module.LMDBDataSetTableMaster.__new__(module.LMDBDataSetTableMaster)
+
+    with pytest.raises(pickle.UnpicklingError):
+        dataset.load_hierarchical_lmdb_dataset("/tmp/ignored")
+
+    assert not Path("/tmp/paddleocr_lmdb_pickle_test").exists()
+
+
+def test_tablemaster_length_metadata_accepts_expected_integer_pickle(monkeypatch):
+    txn = _Txn({b"__len__": pickle.dumps(7)})
+    module = _load_lmdb_dataset_module(
+        monkeypatch, lmdb_open=lambda *args, **kwargs: _Env(txn)
+    )
+    dataset = module.LMDBDataSetTableMaster.__new__(module.LMDBDataSetTableMaster)
+
+    lmdb_sets = dataset.load_hierarchical_lmdb_dataset("/tmp/ignored")
+
+    assert lmdb_sets[0]["num_samples"] == 7
diff --git a/tests/test_files/book.jpg b/tests/test_files/book.jpg
new file mode 100644
index 0000000..05897ee
Binary files /dev/null and b/tests/test_files/book.jpg differ
diff --git a/tests/test_files/book_rot180.jpg b/tests/test_files/book_rot180.jpg
new file mode 100644
index 0000000..64f8af1
Binary files /dev/null and b/tests/test_files/book_rot180.jpg differ
diff --git a/tests/test_files/doc_with_formula.png b/tests/test_files/doc_with_formula.png
new file mode 100644
index 0000000..a7e7f22
Binary files /dev/null and b/tests/test_files/doc_with_formula.png differ
diff --git a/tests/test_files/formula.png b/tests/test_files/formula.png
new file mode 100644
index 0000000..26a271a
Binary files /dev/null and b/tests/test_files/formula.png differ
diff --git a/tests/test_files/medal_table.png b/tests/test_files/medal_table.png
new file mode 100644
index 0000000..d4f6b3f
Binary files /dev/null and b/tests/test_files/medal_table.png differ
diff --git a/tests/test_files/seal.png b/tests/test_files/seal.png
new file mode 100644
index 0000000..b814e41
Binary files /dev/null and b/tests/test_files/seal.png differ
diff --git a/tests/test_files/table.jpg b/tests/test_files/table.jpg
new file mode 100644
index 0000000..07059ca
Binary files /dev/null and b/tests/test_files/table.jpg differ
diff --git a/tests/test_files/textline.png b/tests/test_files/textline.png
new file mode 100644
index 0000000..6e4efd2
Binary files /dev/null and b/tests/test_files/textline.png differ
diff --git a/tests/test_files/textline_rot180.jpg b/tests/test_files/textline_rot180.jpg
new file mode 100644
index 0000000..c28fafa
Binary files /dev/null and b/tests/test_files/textline_rot180.jpg differ
diff --git a/tests/test_ppstructure.py b/tests/test_ppstructure.py
new file mode 100644
index 0000000..290f972
--- /dev/null
+++ b/tests/test_ppstructure.py
@@ -0,0 +1,13 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/tests/testing_utils.py b/tests/testing_utils.py
new file mode 100644
index 0000000..9e6683a
--- /dev/null
+++ b/tests/testing_utils.py
@@ -0,0 +1,50 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from pathlib import Path
+
+TEST_DATA_DIR = Path(__file__).parent / "test_files"
+
+
+def check_simple_inference_result(result, *, expected_length=1):
+    assert result is not None
+    assert isinstance(result, list)
+    assert len(result) == expected_length
+    for res in result:
+        assert isinstance(res, dict)
+
+
+def check_wrapper_simple_inference_param_forwarding(
+    monkeypatch,
+    wrapper,
+    wrapped_obj_attr_name,
+    input,
+    params,
+):
+    def _dummy_predict(input, **params):
+        yield params
+
+    monkeypatch.setattr(
+        getattr(wrapper, wrapped_obj_attr_name), "predict", _dummy_predict
+    )
+
+    result = getattr(wrapper, "predict")(
+        input,
+        **params,
+    )
+
+    assert isinstance(result, list)
+    assert len(result) == 1
+    for k, v in params.items():
+        assert result[0][k] == v
diff --git a/tests/tools/test_docs_github_links.py b/tests/tools/test_docs_github_links.py
new file mode 100644
index 0000000..2ebf10d
--- /dev/null
+++ b/tests/tools/test_docs_github_links.py
@@ -0,0 +1,91 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import importlib.util
+from pathlib import Path
+
+
+REPO_ROOT = Path(__file__).resolve().parents[2]
+
+
+def _load_tool(name):
+    spec = importlib.util.spec_from_file_location(
+        name, REPO_ROOT / "tools" / f"{name}.py"
+    )
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+    return module
+
+
+def test_check_docs_github_links_rejects_moving_source_refs(tmp_path):
+    checker = _load_tool("check_docs_github_links")
+    doc_path = tmp_path / "doc.md"
+    doc_path.write_text(
+        "[config](https://github.com/PaddlePaddle/PaddleOCR/blob/main/deploy/config.yaml)\n",
+        encoding="utf-8",
+    )
+
+    violations = checker.find_forbidden_links(
+        tmp_path,
+        repo_slug="PaddlePaddle/PaddleOCR",
+        forbidden_refs={"main", "master"},
+    )
+
+    assert len(violations) == 1
+    assert violations[0].path == doc_path
+    assert violations[0].ref == "main"
+    assert violations[0].line_number == 1
+
+
+def test_check_docs_github_links_allows_versioned_and_placeholder_refs(tmp_path):
+    checker = _load_tool("check_docs_github_links")
+    doc_path = tmp_path / "doc.md"
+    doc_path.write_text(
+        "\n".join(
+            [
+                "[release](https://github.com/PaddlePaddle/PaddleOCR/blob/release/3.5/deploy/config.yaml)",
+                "[placeholder](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/config.yaml)",
+            ]
+        ),
+        encoding="utf-8",
+    )
+
+    violations = checker.find_forbidden_links(
+        tmp_path,
+        repo_slug="PaddlePaddle/PaddleOCR",
+        forbidden_refs={"main", "master"},
+    )
+
+    assert violations == []
+
+
+def test_resolve_doc_github_refs_replaces_placeholders(tmp_path):
+    resolver = _load_tool("resolve_doc_github_refs")
+    doc_path = tmp_path / "doc.md"
+    doc_path.write_text(
+        "[config](https://github.com/PaddlePaddle/PaddleOCR/blob/{{PADDLEOCR_GITHUB_REF}}/deploy/config.yaml)\n",
+        encoding="utf-8",
+    )
+
+    changed = resolver.resolve_placeholders(
+        tmp_path,
+        placeholder="{{PADDLEOCR_GITHUB_REF}}",
+        source_ref="release/3.5",
+    )
+
+    assert changed == [doc_path]
+    assert (
+        doc_path.read_text(encoding="utf-8")
+        == "[config](https://github.com/PaddlePaddle/PaddleOCR/blob/release/3.5/deploy/config.yaml)\n"
+    )
diff --git a/tests/tools/test_program_safe_yaml.py b/tests/tools/test_program_safe_yaml.py
new file mode 100644
index 0000000..6adf3e4
--- /dev/null
+++ b/tests/tools/test_program_safe_yaml.py
@@ -0,0 +1,95 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import importlib.util
+import sys
+import types
+from pathlib import Path
+
+import pytest
+import yaml
+
+REPO_ROOT = Path(__file__).resolve().parents[2]
+
+
+def _stub_module(monkeypatch, name, **attrs):
+    module = types.ModuleType(name)
+    for key, value in attrs.items():
+        setattr(module, key, value)
+    monkeypatch.setitem(sys.modules, name, module)
+    return module
+
+
+def _load_tools_program_module(monkeypatch):
+    _stub_module(monkeypatch, "paddle")
+    _stub_module(monkeypatch, "paddle.distributed")
+    _stub_module(monkeypatch, "cv2")
+    _stub_module(monkeypatch, "numpy")
+    _stub_module(monkeypatch, "tqdm", tqdm=lambda *args, **kwargs: None)
+    _stub_module(monkeypatch, "ppocr")
+    _stub_module(monkeypatch, "ppocr.utils")
+    _stub_module(monkeypatch, "ppocr.utils.stats", TrainingStats=object)
+    _stub_module(
+        monkeypatch, "ppocr.utils.save_load", save_model=lambda *args, **kwargs: None
+    )
+    _stub_module(
+        monkeypatch,
+        "ppocr.utils.utility",
+        print_dict=lambda *args, **kwargs: None,
+        AverageMeter=object,
+    )
+    _stub_module(
+        monkeypatch, "ppocr.utils.logging", get_logger=lambda *args, **kwargs: None
+    )
+    _stub_module(monkeypatch, "ppocr.utils.loggers", WandbLogger=object, Loggers=object)
+    _stub_module(monkeypatch, "ppocr.utils.profiler")
+    _stub_module(
+        monkeypatch, "ppocr.data", build_dataloader=lambda *args, **kwargs: None
+    )
+    _stub_module(
+        monkeypatch, "ppocr.utils.export_model", export=lambda *args, **kwargs: None
+    )
+
+    module_name = "paddleocr_tools_program"
+    spec = importlib.util.spec_from_file_location(
+        module_name, REPO_ROOT / "tools" / "program.py"
+    )
+    module = importlib.util.module_from_spec(spec)
+    monkeypatch.setitem(sys.modules, module_name, module)
+    spec.loader.exec_module(module)
+    return module
+
+
+def test_tools_program_load_config_rejects_python_object_tags(tmp_path, monkeypatch):
+    module = _load_tools_program_module(monkeypatch)
+    payload = (
+        '!!python/object/apply:os.system ["echo SHOULD_NOT_RUN > '
+        '/tmp/paddleocr_tools_program_test"]\n'
+    )
+    config_path = tmp_path / "malicious.yml"
+    config_path.write_text(payload, encoding="utf-8")
+
+    with pytest.raises(yaml.constructor.ConstructorError):
+        module.load_config(str(config_path))
+
+
+def test_tools_program_parse_opt_rejects_python_object_tags(monkeypatch):
+    parser = _load_tools_program_module(monkeypatch).ArgsParser()
+    malicious_opt = [
+        'Global.debug=!!python/object/apply:os.system ["echo SHOULD_NOT_RUN > '
+        '/tmp/paddleocr_opt_test"]'
+    ]
+
+    with pytest.raises(yaml.constructor.ConstructorError):
+        parser._parse_opt(malicious_opt)
diff --git a/tests/unit/test_patch_layout_parsing.py b/tests/unit/test_patch_layout_parsing.py
new file mode 100644
index 0000000..cfd383b
--- /dev/null
+++ b/tests/unit/test_patch_layout_parsing.py
@@ -0,0 +1,151 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Tests for the layout parsing patches (issue #17503).
+
+These tests verify the fixed functions directly without importing the
+full paddleocr package (which requires paddlex).
+"""
+
+from pathlib import Path
+
+import importlib.util
+import numpy as np
+import pytest
+
+REPO_ROOT = Path(__file__).resolve().parents[2]
+
+
+# Import the patch module directly to avoid triggering the full paddleocr
+# import chain which requires paddlex
+def _import_patch_module():
+    """Import _patch_layout_parsing without triggering paddleocr.__init__."""
+    spec = importlib.util.spec_from_file_location(
+        "paddleocr._pipelines._patch_layout_parsing",
+        REPO_ROOT / "paddleocr" / "_pipelines" / "_patch_layout_parsing.py",
+    )
+    mod = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(mod)
+    return mod
+
+
+_patch_mod = _import_patch_module()
+_fixed_calculate_overlap_ratio = _patch_mod._fixed_calculate_overlap_ratio
+_fixed_calculate_minimum_enclosing_bbox = (
+    _patch_mod._fixed_calculate_minimum_enclosing_bbox
+)
+
+
+class TestFixedCalculateOverlapRatio:
+    """Tests for the overflow-safe calculate_overlap_ratio."""
+
+    def test_normal_overlap(self):
+        bbox1 = [0, 0, 100, 100]
+        bbox2 = [50, 50, 150, 150]
+        ratio = _fixed_calculate_overlap_ratio(bbox1, bbox2, mode="union")
+        # Intersection: 50x50 = 2500
+        # Union: 10000 + 10000 - 2500 = 17500
+        expected = 2500.0 / 17500.0
+        assert abs(ratio - expected) < 1e-6
+
+    def test_no_overlap(self):
+        bbox1 = [0, 0, 50, 50]
+        bbox2 = [100, 100, 200, 200]
+        ratio = _fixed_calculate_overlap_ratio(bbox1, bbox2, mode="union")
+        assert ratio == 0.0
+
+    def test_complete_overlap(self):
+        bbox1 = [0, 0, 100, 100]
+        bbox2 = [0, 0, 100, 100]
+        ratio = _fixed_calculate_overlap_ratio(bbox1, bbox2, mode="union")
+        assert abs(ratio - 1.0) < 1e-6
+
+    def test_large_coordinates_no_overflow(self):
+        """Verify no integer overflow with large coordinate values.
+
+        This is the primary bug from issue #17503. With int32 arithmetic,
+        inter_width * inter_height would overflow for large images
+        (e.g. after document unwarping).
+        """
+        # Coordinates large enough to cause int32 overflow when multiplied
+        bbox1 = np.array([0, 0, 50000, 50000], dtype=np.int32)
+        bbox2 = np.array([0, 0, 50000, 50000], dtype=np.int32)
+        ratio = _fixed_calculate_overlap_ratio(bbox1, bbox2, mode="union")
+        # Should be 1.0 (identical boxes), not corrupted by overflow
+        assert abs(ratio - 1.0) < 1e-6
+
+    def test_large_coordinates_partial_overlap(self):
+        """Verify correct overlap ratio with large coordinates."""
+        bbox1 = np.array([0, 0, 60000, 60000], dtype=np.int32)
+        bbox2 = np.array([30000, 30000, 90000, 90000], dtype=np.int32)
+        ratio = _fixed_calculate_overlap_ratio(bbox1, bbox2, mode="small")
+        # Intersection: 30000x30000 = 900_000_000
+        # Small box area: 60000*60000 = 3_600_000_000
+        expected = 900_000_000.0 / 3_600_000_000.0
+        assert abs(ratio - expected) < 1e-6
+
+    def test_mode_small(self):
+        bbox1 = [0, 0, 100, 100]
+        bbox2 = [0, 0, 200, 200]
+        ratio = _fixed_calculate_overlap_ratio(bbox1, bbox2, mode="small")
+        # Intersection = 100*100 = 10000, small area = 10000
+        assert abs(ratio - 1.0) < 1e-6
+
+    def test_mode_large(self):
+        bbox1 = [0, 0, 100, 100]
+        bbox2 = [0, 0, 200, 200]
+        ratio = _fixed_calculate_overlap_ratio(bbox1, bbox2, mode="large")
+        # Intersection = 100*100 = 10000, large area = 40000
+        expected = 10000.0 / 40000.0
+        assert abs(ratio - expected) < 1e-6
+
+    def test_zero_area_bbox(self):
+        bbox1 = [0, 0, 0, 0]
+        bbox2 = [0, 0, 100, 100]
+        ratio = _fixed_calculate_overlap_ratio(bbox1, bbox2, mode="union")
+        assert ratio == 0.0
+
+    def test_invalid_mode_raises(self):
+        bbox1 = [0, 0, 100, 100]
+        bbox2 = [50, 50, 150, 150]
+        with pytest.raises(ValueError, match="Invalid mode"):
+            _fixed_calculate_overlap_ratio(bbox1, bbox2, mode="invalid")
+
+
+class TestFixedCalculateMinimumEnclosingBbox:
+    """Tests for the empty-safe calculate_minimum_enclosing_bbox."""
+
+    def test_single_bbox(self):
+        result = _fixed_calculate_minimum_enclosing_bbox([[10, 20, 30, 40]])
+        np.testing.assert_array_equal(result, [10, 20, 30, 40])
+
+    def test_multiple_bboxes(self):
+        bboxes = [[10, 20, 30, 40], [5, 15, 35, 45]]
+        result = _fixed_calculate_minimum_enclosing_bbox(bboxes)
+        np.testing.assert_array_equal(result, [5, 15, 35, 45])
+
+    def test_empty_list_returns_degenerate_bbox(self):
+        """Verify empty list returns a degenerate bbox instead of raising.
+
+        This is the secondary fix from issue #17503. The original code
+        raises ValueError("The list of bounding boxes is empty.").
+        """
+        result = _fixed_calculate_minimum_enclosing_bbox([])
+        assert result is not None
+        np.testing.assert_array_equal(result, [0, 0, 0, 0])
+
+    def test_none_input_returns_degenerate_bbox(self):
+        result = _fixed_calculate_minimum_enclosing_bbox(None)
+        assert result is not None
+        np.testing.assert_array_equal(result, [0, 0, 0, 0])
diff --git a/tests/utils/test_naming.py b/tests/utils/test_naming.py
new file mode 100644
index 0000000..7a2d225
--- /dev/null
+++ b/tests/utils/test_naming.py
@@ -0,0 +1,33 @@
+# Copyright (c) 2026 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddleocr._utils.naming import snake_keys_to_camel, snake_to_camel
+
+
+def test_snake_to_camel():
+    assert snake_to_camel("visualize") == "visualize"
+    assert snake_to_camel("top_p") == "topP"
+    assert snake_to_camel("use_e2e_wired_table_rec_model") == "useE2eWiredTableRecModel"
+    assert (
+        snake_to_camel("use_ocr_results_with_table_cells")
+        == "useOcrResultsWithTableCells"
+    )
+    assert snake_to_camel("") == ""
+
+
+def test_snake_keys_to_camel():
+    assert snake_keys_to_camel({"top_p": 1, "visualize": True}) == {
+        "topP": 1,
+        "visualize": True,
+    }
diff --git a/tools/__init__.py b/tools/__init__.py
new file mode 100644
index 0000000..d56c9db
--- /dev/null
+++ b/tools/__init__.py
@@ -0,0 +1,14 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+# Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
diff --git a/tools/check_docs_github_links.py b/tools/check_docs_github_links.py
new file mode 100644
index 0000000..306ef0a
--- /dev/null
+++ b/tools/check_docs_github_links.py
@@ -0,0 +1,81 @@
+#!/usr/bin/env python3
+import argparse
+import re
+from pathlib import Path
+from typing import NamedTuple
+
+
+class Violation(NamedTuple):
+    path: Path
+    line_number: int
+    ref: str
+    url: str
+
+
+def _compile_link_pattern(repo_slug):
+    escaped_repo = re.escape(repo_slug)
+    return re.compile(
+        rf"https://github\.com/{escaped_repo}/(?:blob|tree)/(?P<ref>[^/\s)\]\"'<>]+)"
+        rf"(?P<rest>/[^\s)\]\"'<>]*)?"
+    )
+
+
+def find_forbidden_links(root, repo_slug, forbidden_refs):
+    root = Path(root)
+    forbidden_refs = set(forbidden_refs)
+    link_pattern = _compile_link_pattern(repo_slug)
+    violations = []
+
+    for path in sorted(root.rglob("*.md")):
+        if not path.is_file():
+            continue
+        for line_number, line in enumerate(
+            path.read_text(encoding="utf-8").splitlines(), 1
+        ):
+            for match in link_pattern.finditer(line):
+                ref = match.group("ref")
+                if ref in forbidden_refs:
+                    violations.append(
+                        Violation(
+                            path=path,
+                            line_number=line_number,
+                            ref=ref,
+                            url=match.group(0),
+                        )
+                    )
+    return violations
+
+
+def main(argv=None):
+    parser = argparse.ArgumentParser(
+        description="Reject docs links that point to moving GitHub source refs."
+    )
+    parser.add_argument("--root", default="docs", help="Directory to scan.")
+    parser.add_argument(
+        "--repo-slug", required=True, help="Example: PaddlePaddle/PaddleOCR"
+    )
+    parser.add_argument(
+        "--forbidden-ref",
+        action="append",
+        required=True,
+        help="Moving source ref to reject. Can be passed multiple times.",
+    )
+    args = parser.parse_args(argv)
+
+    violations = find_forbidden_links(
+        args.root,
+        repo_slug=args.repo_slug,
+        forbidden_refs=set(args.forbidden_ref),
+    )
+    if violations:
+        for violation in violations:
+            print(
+                f"{violation.path}:{violation.line_number}: "
+                f"forbidden GitHub ref '{violation.ref}' in {violation.url}"
+            )
+        return 1
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/tools/end2end/convert_ppocr_label.py b/tools/end2end/convert_ppocr_label.py
new file mode 100644
index 0000000..08cea77
--- /dev/null
+++ b/tools/end2end/convert_ppocr_label.py
@@ -0,0 +1,103 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import numpy as np
+import json
+import os
+
+
+def poly_to_string(poly):
+    if len(poly.shape) > 1:
+        poly = np.array(poly).flatten()
+
+    string = "\t".join(str(i) for i in poly)
+    return string
+
+
+def convert_label(label_dir, mode="gt", save_dir="./save_results/"):
+    if not os.path.exists(label_dir):
+        raise ValueError(f"The file {label_dir} does not exist!")
+
+    assert label_dir != save_dir, "hahahhaha"
+
+    label_file = open(label_dir, "r")
+    data = label_file.readlines()
+
+    gt_dict = {}
+
+    for line in data:
+        try:
+            tmp = line.split("\t")
+            assert len(tmp) == 2, ""
+        except:
+            tmp = line.strip().split("    ")
+
+        gt_lists = []
+
+        if tmp[0].split("/")[0] is not None:
+            img_path = tmp[0]
+            anno = json.loads(tmp[1])
+            gt_collect = []
+            for dic in anno:
+                # txt = dic['transcription'].replace(' ', '')  # ignore blank
+                txt = dic["transcription"]
+                if "score" in dic and float(dic["score"]) < 0.5:
+                    continue
+                if "\u3000" in txt:
+                    txt = txt.replace("\u3000", " ")
+                # while ' ' in txt:
+                #    txt = txt.replace(' ', '')
+                poly = np.array(dic["points"]).flatten()
+                if txt == "###":
+                    txt_tag = 1  ## ignore 1
+                else:
+                    txt_tag = 0
+                if mode == "gt":
+                    gt_label = (
+                        poly_to_string(poly) + "\t" + str(txt_tag) + "\t" + txt + "\n"
+                    )
+                else:
+                    gt_label = poly_to_string(poly) + "\t" + txt + "\n"
+
+                gt_lists.append(gt_label)
+
+            gt_dict[img_path] = gt_lists
+        else:
+            continue
+
+    if not os.path.exists(save_dir):
+        os.makedirs(save_dir)
+
+    for img_name in gt_dict.keys():
+        save_name = img_name.split("/")[-1]
+        save_file = os.path.join(save_dir, save_name + ".txt")
+        with open(save_file, "w") as f:
+            f.writelines(gt_dict[img_name])
+
+    print("The convert label saved in {}".format(save_dir))
+
+
+def parse_args():
+    import argparse
+
+    parser = argparse.ArgumentParser(description="args")
+    parser.add_argument("--label_path", type=str, required=True)
+    parser.add_argument("--save_folder", type=str, required=True)
+    parser.add_argument("--mode", type=str, default=False)
+    args = parser.parse_args()
+    return args
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    convert_label(args.label_path, args.mode, args.save_folder)
diff --git a/tools/end2end/draw_html.py b/tools/end2end/draw_html.py
new file mode 100644
index 0000000..c894f44
--- /dev/null
+++ b/tools/end2end/draw_html.py
@@ -0,0 +1,72 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import argparse
+
+
+def str2bool(v):
+    return v.lower() in ("true", "t", "1")
+
+
+def init_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--image_dir", type=str, default="")
+    parser.add_argument("--save_html_path", type=str, default="./default.html")
+    parser.add_argument("--width", type=int, default=640)
+    return parser
+
+
+def parse_args():
+    parser = init_args()
+    return parser.parse_args()
+
+
+def draw_debug_img(args):
+    html_path = args.save_html_path
+
+    err_cnt = 0
+    with open(html_path, "w") as html:
+        html.write("<html>\n<body>\n")
+        html.write('<table border="1">\n')
+        html.write(
+            '<meta http-equiv="Content-Type" content="text/html; charset=utf-8" />'
+        )
+        image_list = []
+        path = args.image_dir
+        for i, filename in enumerate(sorted(os.listdir(path))):
+            if filename.endswith("txt"):
+                continue
+            # The image path
+            base = "{}/{}".format(path, filename)
+            html.write("<tr>\n")
+            html.write(f"<td> {filename}\n GT")
+            html.write(f'<td>GT\n<img src="{base}" width={args.width}></td>')
+
+            html.write("</tr>\n")
+        html.write("<style>\n")
+        html.write("span {\n")
+        html.write("    color: red;\n")
+        html.write("}\n")
+        html.write("</style>\n")
+        html.write("</table>\n")
+        html.write("</html>\n</body>\n")
+    print(f"The html file saved in {html_path}")
+    return
+
+
+if __name__ == "__main__":
+    args = parse_args()
+
+    draw_debug_img(args)
diff --git a/tools/end2end/eval_end2end.py b/tools/end2end/eval_end2end.py
new file mode 100644
index 0000000..3795c3c
--- /dev/null
+++ b/tools/end2end/eval_end2end.py
@@ -0,0 +1,191 @@
+# Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import re
+import sys
+import shapely
+from shapely.geometry import Polygon
+import numpy as np
+from collections import defaultdict
+import operator
+import editdistance
+
+
+def strQ2B(ustring):
+    rstring = ""
+    for uchar in ustring:
+        inside_code = ord(uchar)
+        if inside_code == 12288:
+            inside_code = 32
+        elif inside_code >= 65281 and inside_code <= 65374:
+            inside_code -= 65248
+        rstring += chr(inside_code)
+    return rstring
+
+
+def polygon_from_str(polygon_points):
+    """
+    Create a shapely polygon object from gt or dt line.
+    """
+    polygon_points = np.array(polygon_points).reshape(4, 2)
+    polygon = Polygon(polygon_points).convex_hull
+    return polygon
+
+
+def polygon_iou(poly1, poly2):
+    """
+    Intersection over union between two shapely polygons.
+    """
+    if not poly1.intersects(poly2):  # this test is fast and can accelerate calculation
+        iou = 0
+    else:
+        try:
+            inter_area = poly1.intersection(poly2).area
+            union_area = poly1.area + poly2.area - inter_area
+            iou = float(inter_area) / union_area
+        except shapely.geos.TopologicalError:
+            # except Exception as e:
+            #     print(e)
+            print("shapely.geos.TopologicalError occurred, iou set to 0")
+            iou = 0
+    return iou
+
+
+def ed(str1, str2):
+    return editdistance.eval(str1, str2)
+
+
+def e2e_eval(gt_dir, res_dir, ignore_blank=False):
+    print("start testing...")
+    iou_thresh = 0.5
+    val_names = os.listdir(gt_dir)
+    num_gt_chars = 0
+    gt_count = 0
+    dt_count = 0
+    hit = 0
+    ed_sum = 0
+
+    for i, val_name in enumerate(val_names):
+        with open(os.path.join(gt_dir, val_name), encoding="utf-8") as f:
+            gt_lines = [o.strip() for o in f.readlines()]
+        gts = []
+        ignore_masks = []
+        for line in gt_lines:
+            parts = line.strip().split("\t")
+            # ignore illegal data
+            if len(parts) < 9:
+                continue
+            assert len(parts) < 11
+            if len(parts) == 9:
+                gts.append(parts[:8] + [""])
+            else:
+                gts.append(parts[:8] + [parts[-1]])
+
+            ignore_masks.append(parts[8])
+
+        val_path = os.path.join(res_dir, val_name)
+        if not os.path.exists(val_path):
+            dt_lines = []
+        else:
+            with open(val_path, encoding="utf-8") as f:
+                dt_lines = [o.strip() for o in f.readlines()]
+        dts = []
+        for line in dt_lines:
+            # print(line)
+            parts = line.strip().split("\t")
+            assert len(parts) < 10, "line error: {}".format(line)
+            if len(parts) == 8:
+                dts.append(parts + [""])
+            else:
+                dts.append(parts)
+
+        dt_match = [False] * len(dts)
+        gt_match = [False] * len(gts)
+        all_ious = defaultdict(tuple)
+        for index_gt, gt in enumerate(gts):
+            gt_coors = [float(gt_coor) for gt_coor in gt[0:8]]
+            gt_poly = polygon_from_str(gt_coors)
+            for index_dt, dt in enumerate(dts):
+                dt_coors = [float(dt_coor) for dt_coor in dt[0:8]]
+                dt_poly = polygon_from_str(dt_coors)
+                iou = polygon_iou(dt_poly, gt_poly)
+                if iou >= iou_thresh:
+                    all_ious[(index_gt, index_dt)] = iou
+        sorted_ious = sorted(all_ious.items(), key=operator.itemgetter(1), reverse=True)
+        sorted_gt_dt_pairs = [item[0] for item in sorted_ious]
+
+        # matched gt and dt
+        for gt_dt_pair in sorted_gt_dt_pairs:
+            index_gt, index_dt = gt_dt_pair
+            if gt_match[index_gt] == False and dt_match[index_dt] == False:
+                gt_match[index_gt] = True
+                dt_match[index_dt] = True
+                if ignore_blank:
+                    gt_str = strQ2B(gts[index_gt][8]).replace(" ", "")
+                    dt_str = strQ2B(dts[index_dt][8]).replace(" ", "")
+                else:
+                    gt_str = strQ2B(gts[index_gt][8])
+                    dt_str = strQ2B(dts[index_dt][8])
+                if ignore_masks[index_gt] == "0":
+                    ed_sum += ed(gt_str, dt_str)
+                    num_gt_chars += len(gt_str)
+                    if gt_str == dt_str:
+                        hit += 1
+                    gt_count += 1
+                    dt_count += 1
+
+        # unmatched dt
+        for tindex, dt_match_flag in enumerate(dt_match):
+            if dt_match_flag == False:
+                dt_str = dts[tindex][8]
+                gt_str = ""
+                ed_sum += ed(dt_str, gt_str)
+                dt_count += 1
+
+        # unmatched gt
+        for tindex, gt_match_flag in enumerate(gt_match):
+            if gt_match_flag == False and ignore_masks[tindex] == "0":
+                dt_str = ""
+                gt_str = gts[tindex][8]
+                ed_sum += ed(gt_str, dt_str)
+                num_gt_chars += len(gt_str)
+                gt_count += 1
+
+    eps = 1e-9
+    print("hit, dt_count, gt_count", hit, dt_count, gt_count)
+    precision = hit / (dt_count + eps)
+    recall = hit / (gt_count + eps)
+    fmeasure = 2.0 * precision * recall / (precision + recall + eps)
+    avg_edit_dist_img = ed_sum / len(val_names)
+    avg_edit_dist_field = ed_sum / (gt_count + eps)
+    character_acc = 1 - ed_sum / (num_gt_chars + eps)
+
+    print("character_acc: %.2f" % (character_acc * 100) + "%")
+    print("avg_edit_dist_field: %.2f" % (avg_edit_dist_field))
+    print("avg_edit_dist_img: %.2f" % (avg_edit_dist_img))
+    print("precision: %.2f" % (precision * 100) + "%")
+    print("recall: %.2f" % (recall * 100) + "%")
+    print("fmeasure: %.2f" % (fmeasure * 100) + "%")
+
+
+if __name__ == "__main__":
+    # if len(sys.argv) != 3:
+    #     print("python3 ocr_e2e_eval.py gt_dir res_dir")
+    #     exit(-1)
+    # gt_folder = sys.argv[1]
+    # pred_folder = sys.argv[2]
+    gt_folder = sys.argv[1]
+    pred_folder = sys.argv[2]
+    e2e_eval(gt_folder, pred_folder)
diff --git a/tools/end2end/readme.md b/tools/end2end/readme.md
new file mode 100644
index 0000000..636ee76
--- /dev/null
+++ b/tools/end2end/readme.md
@@ -0,0 +1,63 @@
+
+# 简介
+
+`tools/end2end`目录下存放了文本检测+文本识别pipeline串联预测的指标评测代码以及可视化工具。本节介绍文本检测+文本识别的端对端指标评估方式。
+
+
+## 端对端评测步骤
+
+**步骤一：**
+
+运行`tools/infer/predict_system.py`，得到保存的结果:
+
+```
+python3 tools/infer/predict_system.py  --det_model_dir=./ch_PP-OCRv2_det_infer/ --rec_model_dir=./ch_PP-OCRv2_rec_infer/  --image_dir=./datasets/img_dir/ --draw_img_save_dir=./ch_PP-OCRv2_results/ --is_visualize=True
+```
+
+文本检测识别可视化图默认保存在`./ch_PP-OCRv2_results/`目录下，预测结果默认保存在`./ch_PP-OCRv2_results/system_results.txt`中，格式如下：
+```
+all-sum-510/00224225.jpg        [{"transcription": "超赞", "points": [[8.0, 48.0], [157.0, 44.0], [159.0, 115.0], [10.0, 119.0]], "score": "0.99396634"}, {"transcription": "中", "points": [[202.0, 152.0], [230.0, 152.0], [230.0, 163.0], [202.0, 163.0]], "score": "0.09310734"}, {"transcription": "58.0m", "points": [[196.0, 192.0], [444.0, 192.0], [444.0, 240.0], [196.0, 240.0]], "score": "0.44041982"}, {"transcription": "汽配", "points": [[55.0, 263.0], [95.0, 263.0], [95.0, 281.0], [55.0, 281.0]], "score": "0.9986651"}, {"transcription": "成总店", "points": [[120.0, 262.0], [176.0, 262.0], [176.0, 283.0], [120.0, 283.0]], "score": "0.9929402"}, {"transcription": "K", "points": [[237.0, 286.0], [311.0, 286.0], [311.0, 345.0], [237.0, 345.0]], "score": "0.6074794"}, {"transcription": "88：-8", "points": [[203.0, 405.0], [477.0, 414.0], [475.0, 459.0], [201.0, 450.0]], "score": "0.7106863"}]
+```
+
+
+**步骤二：**
+
+将步骤一保存的数据转换为端对端评测需要的数据格式：
+
+修改 `tools/end2end/convert_ppocr_label.py`中的代码，convert_label函数中设置输入标签路径，Mode，保存标签路径等，对预测数据的GTlabel和预测结果的label格式进行转换。
+
+```
+python3 tools/end2end/convert_ppocr_label.py --mode=gt --label_path=path/to/label_txt --save_folder=save_gt_label
+
+python3 tools/end2end/convert_ppocr_label.py --mode=pred --label_path=path/to/pred_txt --save_folder=save_PPOCRV2_infer
+```
+
+得到如下结果：
+```
+├── ./save_gt_label/
+├── ./save_PPOCRV2_infer/
+```
+
+**步骤三：**
+
+执行端对端评测，运行`tools/eval_end2end.py`计算端对端指标，运行方式如下：
+
+```
+python3 tools/eval_end2end.py "gt_label_dir"  "predict_label_dir"
+```
+
+比如：
+
+```
+python3 tools/eval_end2end.py ./save_gt_label/ ./save_PPOCRV2_infer/
+```
+将得到如下结果，fmeasure为主要关注的指标：
+```
+hit, dt_count, gt_count 1557 2693 3283
+character_acc: 61.77%
+avg_edit_dist_field: 3.08
+avg_edit_dist_img: 51.82
+precision: 57.82%
+recall: 47.43%
+fmeasure: 52.11%
+```
diff --git a/tools/eval.py b/tools/eval.py
new file mode 100755
index 0000000..4a59f2c
--- /dev/null
+++ b/tools/eval.py
@@ -0,0 +1,181 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.insert(0, __dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "..")))
+
+import paddle
+from ppocr.data import build_dataloader, set_signal_handlers
+from ppocr.modeling.architectures import build_model
+from ppocr.postprocess import build_post_process
+from ppocr.metrics import build_metric
+from ppocr.utils.save_load import load_model
+import tools.program as program
+
+
+def main():
+    global_config = config["Global"]
+    # build dataloader
+    set_signal_handlers()
+    valid_dataloader = build_dataloader(config, "Eval", device, logger)
+
+    # build post process
+    post_process_class = build_post_process(config["PostProcess"], global_config)
+
+    # build model
+    # for rec algorithm
+    if hasattr(post_process_class, "character"):
+        char_num = len(getattr(post_process_class, "character"))
+        if config["Architecture"]["algorithm"] in [
+            "Distillation",
+        ]:  # distillation model
+            for key in config["Architecture"]["Models"]:
+                if (
+                    config["Architecture"]["Models"][key]["Head"]["name"] == "MultiHead"
+                ):  # for multi head
+                    out_channels_list = {}
+                    if config["PostProcess"]["name"] == "DistillationSARLabelDecode":
+                        char_num = char_num - 2
+                    if config["PostProcess"]["name"] == "DistillationNRTRLabelDecode":
+                        char_num = char_num - 3
+                    out_channels_list["CTCLabelDecode"] = char_num
+                    out_channels_list["SARLabelDecode"] = char_num + 2
+                    out_channels_list["NRTRLabelDecode"] = char_num + 3
+                    config["Architecture"]["Models"][key]["Head"][
+                        "out_channels_list"
+                    ] = out_channels_list
+                else:
+                    config["Architecture"]["Models"][key]["Head"][
+                        "out_channels"
+                    ] = char_num
+        elif config["Architecture"]["Head"]["name"] == "MultiHead":  # for multi head
+            out_channels_list = {}
+            if config["PostProcess"]["name"] == "SARLabelDecode":
+                char_num = char_num - 2
+            if config["PostProcess"]["name"] == "NRTRLabelDecode":
+                char_num = char_num - 3
+            out_channels_list["CTCLabelDecode"] = char_num
+            out_channels_list["SARLabelDecode"] = char_num + 2
+            out_channels_list["NRTRLabelDecode"] = char_num + 3
+            config["Architecture"]["Head"]["out_channels_list"] = out_channels_list
+        else:  # base rec model
+            config["Architecture"]["Head"]["out_channels"] = char_num
+
+    model = build_model(config["Architecture"])
+    extra_input_models = [
+        "SRN",
+        "NRTR",
+        "SAR",
+        "SEED",
+        "SVTR",
+        "SVTR_LCNet",
+        "VisionLAN",
+        "RobustScanner",
+        "SVTR_HGNet",
+    ]
+    extra_input = False
+    if config["Architecture"]["algorithm"] == "Distillation":
+        for key in config["Architecture"]["Models"]:
+            extra_input = (
+                extra_input
+                or config["Architecture"]["Models"][key]["algorithm"]
+                in extra_input_models
+            )
+    else:
+        extra_input = config["Architecture"]["algorithm"] in extra_input_models
+    if "model_type" in config["Architecture"].keys():
+        if config["Architecture"]["algorithm"] == "CAN":
+            model_type = "can"
+        elif config["Architecture"]["algorithm"] == "LaTeXOCR":
+            model_type = "latexocr"
+            config["Metric"]["cal_bleu_score"] = True
+        elif config["Architecture"]["algorithm"] == "UniMERNet":
+            model_type = "unimernet"
+            config["Metric"]["cal_bleu_score"] = True
+        elif config["Architecture"]["algorithm"] in [
+            "PP-FormulaNet-S",
+            "PP-FormulaNet-L",
+            "PP-FormulaNet_plus-S",
+            "PP-FormulaNet_plus-M",
+            "PP-FormulaNet_plus-L",
+        ]:
+            model_type = "pp_formulanet"
+            config["Metric"]["cal_bleu_score"] = True
+        else:
+            model_type = config["Architecture"]["model_type"]
+    else:
+        model_type = None
+
+    # build metric
+    eval_class = build_metric(config["Metric"])
+    # amp
+    use_amp = config["Global"].get("use_amp", False)
+    amp_level = config["Global"].get("amp_level", "O2")
+    amp_custom_black_list = config["Global"].get("amp_custom_black_list", [])
+    if use_amp:
+        AMP_RELATED_FLAGS_SETTING = {
+            "FLAGS_cudnn_batchnorm_spatial_persistent": 1,
+        }
+        paddle.set_flags(AMP_RELATED_FLAGS_SETTING)
+        scale_loss = config["Global"].get("scale_loss", 1.0)
+        use_dynamic_loss_scaling = config["Global"].get(
+            "use_dynamic_loss_scaling", False
+        )
+        scaler = paddle.amp.GradScaler(
+            init_loss_scaling=scale_loss,
+            use_dynamic_loss_scaling=use_dynamic_loss_scaling,
+        )
+        if amp_level == "O2":
+            model = paddle.amp.decorate(
+                models=model, level=amp_level, master_weight=True
+            )
+    else:
+        scaler = None
+
+    best_model_dict = load_model(
+        config, model, model_type=config["Architecture"]["model_type"]
+    )
+    if len(best_model_dict):
+        logger.info("metric in ckpt ***************")
+        for k, v in best_model_dict.items():
+            logger.info("{}:{}".format(k, v))
+
+    # start eval
+    metric = program.eval(
+        model,
+        valid_dataloader,
+        post_process_class,
+        eval_class,
+        model_type,
+        extra_input,
+        scaler,
+        amp_level,
+        amp_custom_black_list,
+    )
+    logger.info("metric eval ***************")
+    for k, v in metric.items():
+        logger.info("{}:{}".format(k, v))
+
+
+if __name__ == "__main__":
+    config, device, logger, vdl_writer = program.preprocess()
+    main()
diff --git a/tools/export_center.py b/tools/export_center.py
new file mode 100644
index 0000000..e79c2b8
--- /dev/null
+++ b/tools/export_center.py
@@ -0,0 +1,77 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+import pickle
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.append(os.path.abspath(os.path.join(__dir__, "..")))
+
+from ppocr.data import build_dataloader, set_signal_handlers
+from ppocr.modeling.architectures import build_model
+from ppocr.postprocess import build_post_process
+from ppocr.utils.save_load import load_model
+from ppocr.utils.utility import print_dict
+import tools.program as program
+
+
+def main():
+    global_config = config["Global"]
+    # build dataloader
+    config["Eval"]["dataset"]["name"] = config["Train"]["dataset"]["name"]
+    config["Eval"]["dataset"]["data_dir"] = config["Train"]["dataset"]["data_dir"]
+    config["Eval"]["dataset"]["label_file_list"] = config["Train"]["dataset"][
+        "label_file_list"
+    ]
+    set_signal_handlers()
+    eval_dataloader = build_dataloader(config, "Eval", device, logger)
+
+    # build post process
+    post_process_class = build_post_process(config["PostProcess"], global_config)
+
+    # build model
+    # for rec algorithm
+    if hasattr(post_process_class, "character"):
+        char_num = len(getattr(post_process_class, "character"))
+        config["Architecture"]["Head"]["out_channels"] = char_num
+
+    # set return_features = True
+    config["Architecture"]["Head"]["return_feats"] = True
+
+    model = build_model(config["Architecture"])
+
+    best_model_dict = load_model(config, model)
+    if len(best_model_dict):
+        logger.info("metric in ckpt ***************")
+        for k, v in best_model_dict.items():
+            logger.info("{}:{}".format(k, v))
+
+    # get features from train data
+    char_center = program.get_center(model, eval_dataloader, post_process_class)
+
+    # serialize to disk
+    with open("train_center.pkl", "wb") as f:
+        pickle.dump(char_center, f)
+    return
+
+
+if __name__ == "__main__":
+    config, device, logger, vdl_writer = program.preprocess()
+    main()
diff --git a/tools/export_model.py b/tools/export_model.py
new file mode 100755
index 0000000..da1bcbb
--- /dev/null
+++ b/tools/export_model.py
@@ -0,0 +1,37 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "..")))
+
+import argparse
+
+from tools.program import load_config, merge_config, ArgsParser
+from ppocr.utils.export_model import export
+
+
+def main():
+    FLAGS = ArgsParser().parse_args()
+    config = load_config(FLAGS.config)
+    config = merge_config(config, FLAGS.opt)
+    # export model
+    export(config)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/tools/infer/predict_cls.py b/tools/infer/predict_cls.py
new file mode 100755
index 0000000..043e645
--- /dev/null
+++ b/tools/infer/predict_cls.py
@@ -0,0 +1,167 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "../..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+
+import cv2
+import copy
+import numpy as np
+import math
+import time
+import traceback
+
+import tools.infer.utility as utility
+from ppocr.postprocess import build_post_process
+from ppocr.utils.logging import get_logger
+from ppocr.utils.utility import get_image_file_list, check_and_read
+
+logger = get_logger()
+
+
+class TextClassifier(object):
+    def __init__(self, args):
+        if os.path.exists(f"{args.cls_model_dir}/inference.yml"):
+            model_config = utility.load_config(f"{args.cls_model_dir}/inference.yml")
+            model_name = model_config.get("Global", {}).get("model_name", "")
+            if model_name and model_name not in [
+                "PP-LCNet_x1_0_textline_ori",
+                "PP-LCNet_x0_25_textline_ori",
+            ]:
+                raise ValueError(
+                    f"{model_name} is not supported. Please check if the model is supported by the PaddleOCR wheel."
+                )
+
+        self.cls_image_shape = [int(v) for v in args.cls_image_shape.split(",")]
+        self.cls_batch_num = args.cls_batch_num
+        self.cls_thresh = args.cls_thresh
+        postprocess_params = {
+            "name": "ClsPostProcess",
+            "label_list": args.label_list,
+        }
+        self.postprocess_op = build_post_process(postprocess_params)
+        (
+            self.predictor,
+            self.input_tensor,
+            self.output_tensors,
+            _,
+        ) = utility.create_predictor(args, "cls", logger)
+        self.use_onnx = args.use_onnx
+
+    def resize_norm_img(self, img):
+        imgC, imgH, imgW = self.cls_image_shape
+        h = img.shape[0]
+        w = img.shape[1]
+        ratio = w / float(h)
+        if math.ceil(imgH * ratio) > imgW:
+            resized_w = imgW
+        else:
+            resized_w = int(math.ceil(imgH * ratio))
+        resized_image = cv2.resize(img, (resized_w, imgH))
+        resized_image = resized_image.astype("float32")
+        if self.cls_image_shape[0] == 1:
+            resized_image = resized_image / 255
+            resized_image = resized_image[np.newaxis, :]
+        else:
+            resized_image = resized_image.transpose((2, 0, 1)) / 255
+        resized_image -= 0.5
+        resized_image /= 0.5
+        padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
+        padding_im[:, :, 0:resized_w] = resized_image
+        return padding_im
+
+    def __call__(self, img_list):
+        img_list = copy.deepcopy(img_list)
+        img_num = len(img_list)
+        # Calculate the aspect ratio of all text bars
+        width_list = []
+        for img in img_list:
+            width_list.append(img.shape[1] / float(img.shape[0]))
+        # Sorting can speed up the cls process
+        indices = np.argsort(np.array(width_list))
+
+        cls_res = [["", 0.0]] * img_num
+        batch_num = self.cls_batch_num
+        elapse = 0
+        for beg_img_no in range(0, img_num, batch_num):
+            end_img_no = min(img_num, beg_img_no + batch_num)
+            norm_img_batch = []
+            max_wh_ratio = 0
+            starttime = time.time()
+            for ino in range(beg_img_no, end_img_no):
+                h, w = img_list[indices[ino]].shape[0:2]
+                wh_ratio = w * 1.0 / h
+                max_wh_ratio = max(max_wh_ratio, wh_ratio)
+            for ino in range(beg_img_no, end_img_no):
+                norm_img = self.resize_norm_img(img_list[indices[ino]])
+                norm_img = norm_img[np.newaxis, :]
+                norm_img_batch.append(norm_img)
+            norm_img_batch = np.concatenate(norm_img_batch)
+            norm_img_batch = norm_img_batch.copy()
+
+            if self.use_onnx:
+                input_dict = {}
+                input_dict[self.input_tensor.name] = norm_img_batch
+                outputs = self.predictor.run(self.output_tensors, input_dict)
+                prob_out = outputs[0]
+            else:
+                self.input_tensor.copy_from_cpu(norm_img_batch)
+                self.predictor.run()
+                prob_out = self.output_tensors[0].copy_to_cpu()
+                self.predictor.try_shrink_memory()
+            cls_result = self.postprocess_op(prob_out)
+            elapse += time.time() - starttime
+            for rno in range(len(cls_result)):
+                label, score = cls_result[rno]
+                cls_res[indices[beg_img_no + rno]] = [label, score]
+                if "180" in label and score > self.cls_thresh:
+                    img_list[indices[beg_img_no + rno]] = cv2.rotate(
+                        img_list[indices[beg_img_no + rno]], 1
+                    )
+        return img_list, cls_res, elapse
+
+
+def main(args):
+    image_file_list = get_image_file_list(args.image_dir)
+    text_classifier = TextClassifier(args)
+    valid_image_file_list = []
+    img_list = []
+    for image_file in image_file_list:
+        img, flag, _ = check_and_read(image_file)
+        if not flag:
+            img = cv2.imread(image_file)
+        if img is None:
+            logger.info("error in loading image:{}".format(image_file))
+            continue
+        valid_image_file_list.append(image_file)
+        img_list.append(img)
+    try:
+        img_list, cls_res, predict_time = text_classifier(img_list)
+    except Exception as E:
+        logger.info(traceback.format_exc())
+        logger.info(E)
+        exit()
+    for ino in range(len(img_list)):
+        logger.info(
+            "Predicts of {}:{}".format(valid_image_file_list[ino], cls_res[ino])
+        )
+
+
+if __name__ == "__main__":
+    main(utility.parse_args())
diff --git a/tools/infer/predict_det.py b/tools/infer/predict_det.py
new file mode 100755
index 0000000..3d9e59b
--- /dev/null
+++ b/tools/infer/predict_det.py
@@ -0,0 +1,504 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "../..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+
+import cv2
+import numpy as np
+import time
+import sys
+
+import tools.infer.utility as utility
+from ppocr.utils.logging import get_logger
+from ppocr.utils.utility import get_image_file_list, check_and_read
+from ppocr.data import create_operators, transform
+from ppocr.postprocess import build_post_process
+import json
+
+
+class TextDetector(object):
+    def __init__(self, args, logger=None):
+        if os.path.exists(f"{args.det_model_dir}/inference.yml"):
+            model_config = utility.load_config(f"{args.det_model_dir}/inference.yml")
+            model_name = model_config.get("Global", {}).get("model_name", "")
+            if model_name and model_name not in [
+                "PP-OCRv5_mobile_det",
+                "PP-OCRv5_server_det",
+                "PP-OCRv6_tiny_det",
+                "PP-OCRv6_small_det",
+                "PP-OCRv6_medium_det",
+            ]:
+                raise ValueError(
+                    f"{model_name} is not supported. Please check if the model is supported by the PaddleOCR wheel."
+                )
+
+        if logger is None:
+            logger = get_logger()
+        self.args = args
+        self.det_algorithm = args.det_algorithm
+        self.use_onnx = args.use_onnx
+        pre_process_list = [
+            {
+                "DetResizeForTest": {
+                    "limit_side_len": args.det_limit_side_len,
+                    "limit_type": args.det_limit_type,
+                }
+            },
+            {
+                "NormalizeImage": {
+                    "std": [0.229, 0.224, 0.225],
+                    "mean": [0.485, 0.456, 0.406],
+                    "scale": "1./255.",
+                    "order": "hwc",
+                }
+            },
+            {"ToCHWImage": None},
+            {"KeepKeys": {"keep_keys": ["image", "shape"]}},
+        ]
+        postprocess_params = {}
+        if self.det_algorithm == "DB":
+            postprocess_params["name"] = "DBPostProcess"
+            postprocess_params["thresh"] = args.det_db_thresh
+            postprocess_params["box_thresh"] = args.det_db_box_thresh
+            postprocess_params["max_candidates"] = 1000
+            postprocess_params["unclip_ratio"] = args.det_db_unclip_ratio
+            postprocess_params["use_dilation"] = args.use_dilation
+            postprocess_params["score_mode"] = args.det_db_score_mode
+            postprocess_params["box_type"] = args.det_box_type
+        elif self.det_algorithm == "DB++":
+            postprocess_params["name"] = "DBPostProcess"
+            postprocess_params["thresh"] = args.det_db_thresh
+            postprocess_params["box_thresh"] = args.det_db_box_thresh
+            postprocess_params["max_candidates"] = 1000
+            postprocess_params["unclip_ratio"] = args.det_db_unclip_ratio
+            postprocess_params["use_dilation"] = args.use_dilation
+            postprocess_params["score_mode"] = args.det_db_score_mode
+            postprocess_params["box_type"] = args.det_box_type
+            pre_process_list[1] = {
+                "NormalizeImage": {
+                    "std": [1.0, 1.0, 1.0],
+                    "mean": [0.48109378172549, 0.45752457890196, 0.40787054090196],
+                    "scale": "1./255.",
+                    "order": "hwc",
+                }
+            }
+        elif self.det_algorithm == "EAST":
+            postprocess_params["name"] = "EASTPostProcess"
+            postprocess_params["score_thresh"] = args.det_east_score_thresh
+            postprocess_params["cover_thresh"] = args.det_east_cover_thresh
+            postprocess_params["nms_thresh"] = args.det_east_nms_thresh
+        elif self.det_algorithm == "SAST":
+            pre_process_list[0] = {
+                "DetResizeForTest": {"resize_long": args.det_limit_side_len}
+            }
+            postprocess_params["name"] = "SASTPostProcess"
+            postprocess_params["score_thresh"] = args.det_sast_score_thresh
+            postprocess_params["nms_thresh"] = args.det_sast_nms_thresh
+
+            if args.det_box_type == "poly":
+                postprocess_params["sample_pts_num"] = 6
+                postprocess_params["expand_scale"] = 1.2
+                postprocess_params["shrink_ratio_of_width"] = 0.2
+            else:
+                postprocess_params["sample_pts_num"] = 2
+                postprocess_params["expand_scale"] = 1.0
+                postprocess_params["shrink_ratio_of_width"] = 0.3
+
+        elif self.det_algorithm == "PSE":
+            postprocess_params["name"] = "PSEPostProcess"
+            postprocess_params["thresh"] = args.det_pse_thresh
+            postprocess_params["box_thresh"] = args.det_pse_box_thresh
+            postprocess_params["min_area"] = args.det_pse_min_area
+            postprocess_params["box_type"] = args.det_box_type
+            postprocess_params["scale"] = args.det_pse_scale
+        elif self.det_algorithm == "FCE":
+            pre_process_list[0] = {"DetResizeForTest": {"rescale_img": [1080, 736]}}
+            postprocess_params["name"] = "FCEPostProcess"
+            postprocess_params["scales"] = args.scales
+            postprocess_params["alpha"] = args.alpha
+            postprocess_params["beta"] = args.beta
+            postprocess_params["fourier_degree"] = args.fourier_degree
+            postprocess_params["box_type"] = args.det_box_type
+        elif self.det_algorithm == "CT":
+            pre_process_list[0] = {"ScaleAlignedShort": {"short_size": 640}}
+            postprocess_params["name"] = "CTPostProcess"
+        else:
+            logger.info("unknown det_algorithm:{}".format(self.det_algorithm))
+            sys.exit(0)
+
+        self.preprocess_op = create_operators(pre_process_list)
+        self.postprocess_op = build_post_process(postprocess_params)
+        (
+            self.predictor,
+            self.input_tensor,
+            self.output_tensors,
+            self.config,
+        ) = utility.create_predictor(args, "det", logger)
+
+        if self.use_onnx:
+            img_h, img_w = self.input_tensor.shape[2:]
+            if isinstance(img_h, str) or isinstance(img_w, str):
+                pass
+            elif img_h is not None and img_w is not None and img_h > 0 and img_w > 0:
+                pre_process_list[0] = {
+                    "DetResizeForTest": {"image_shape": [img_h, img_w]}
+                }
+        self.preprocess_op = create_operators(pre_process_list)
+
+        if args.benchmark:
+            import auto_log
+
+            pid = os.getpid()
+            gpu_id = utility.get_infer_gpuid()
+            self.autolog = auto_log.AutoLogger(
+                model_name="det",
+                model_precision=args.precision,
+                batch_size=1,
+                data_shape="dynamic",
+                save_path=None,  # not used if logger is not None
+                inference_config=self.config,
+                pids=pid,
+                process_name=None,
+                gpu_ids=gpu_id if args.use_gpu else None,
+                time_keys=["preprocess_time", "inference_time", "postprocess_time"],
+                warmup=2,
+                logger=logger,
+            )
+
+    def order_points_clockwise(self, pts):
+        rect = np.zeros((4, 2), dtype="float32")
+        s = pts.sum(axis=1)
+        rect[0] = pts[np.argmin(s)]
+        rect[2] = pts[np.argmax(s)]
+        tmp = np.delete(pts, (np.argmin(s), np.argmax(s)), axis=0)
+        diff = np.diff(np.array(tmp), axis=1)
+        rect[1] = tmp[np.argmin(diff)]
+        rect[3] = tmp[np.argmax(diff)]
+        return rect
+
+    def pad_polygons(self, polygon, max_points):
+        padding_size = max_points - len(polygon)
+        if padding_size == 0:
+            return polygon
+        last_point = polygon[-1]
+        padding = np.repeat([last_point], padding_size, axis=0)
+        return np.vstack([polygon, padding])
+
+    def clip_det_res(self, points, img_height, img_width):
+        for pno in range(points.shape[0]):
+            points[pno, 0] = int(min(max(points[pno, 0], 0), img_width - 1))
+            points[pno, 1] = int(min(max(points[pno, 1], 0), img_height - 1))
+        return points
+
+    def filter_tag_det_res(self, dt_boxes, image_shape):
+        img_height, img_width = image_shape[0:2]
+        dt_boxes_new = []
+        for box in dt_boxes:
+            if type(box) is list:
+                box = np.array(box)
+            box = self.order_points_clockwise(box)
+            box = self.clip_det_res(box, img_height, img_width)
+            rect_width = int(np.linalg.norm(box[0] - box[1]))
+            rect_height = int(np.linalg.norm(box[0] - box[3]))
+            if rect_width <= 3 or rect_height <= 3:
+                continue
+            dt_boxes_new.append(box)
+        dt_boxes = np.array(dt_boxes_new)
+        return dt_boxes
+
+    def filter_tag_det_res_only_clip(self, dt_boxes, image_shape):
+        img_height, img_width = image_shape[0:2]
+        dt_boxes_new = []
+        for box in dt_boxes:
+            if type(box) is list:
+                box = np.array(box)
+            box = self.clip_det_res(box, img_height, img_width)
+            dt_boxes_new.append(box)
+
+        if len(dt_boxes_new) > 0:
+            max_points = max(len(polygon) for polygon in dt_boxes_new)
+            dt_boxes_new = [
+                self.pad_polygons(polygon, max_points) for polygon in dt_boxes_new
+            ]
+
+        dt_boxes = np.array(dt_boxes_new)
+        return dt_boxes
+
+    def predict(self, img):
+        ori_im = img.copy()
+        data = {"image": img}
+
+        st = time.time()
+
+        if self.args.benchmark:
+            self.autolog.times.start()
+
+        data = transform(data, self.preprocess_op)
+        img, shape_list = data
+        if img is None:
+            return None, 0
+        img = np.expand_dims(img, axis=0)
+        shape_list = np.expand_dims(shape_list, axis=0)
+        img = img.copy()
+
+        if self.args.benchmark:
+            self.autolog.times.stamp()
+        if self.use_onnx:
+            input_dict = {}
+            input_dict[self.input_tensor.name] = img
+            outputs = self.predictor.run(self.output_tensors, input_dict)
+        else:
+            self.input_tensor.copy_from_cpu(img)
+            self.predictor.run()
+            outputs = []
+            for output_tensor in self.output_tensors:
+                output = output_tensor.copy_to_cpu()
+                outputs.append(output)
+            if self.args.benchmark:
+                self.autolog.times.stamp()
+
+        preds = {}
+        if self.det_algorithm == "EAST":
+            preds["f_geo"] = outputs[0]
+            preds["f_score"] = outputs[1]
+        elif self.det_algorithm == "SAST":
+            preds["f_border"] = outputs[0]
+            preds["f_score"] = outputs[1]
+            preds["f_tco"] = outputs[2]
+            preds["f_tvo"] = outputs[3]
+        elif self.det_algorithm in ["DB", "PSE", "DB++"]:
+            preds["maps"] = outputs[0]
+        elif self.det_algorithm == "FCE":
+            for i, output in enumerate(outputs):
+                preds["level_{}".format(i)] = output
+        elif self.det_algorithm == "CT":
+            preds["maps"] = outputs[0]
+            preds["score"] = outputs[1]
+        else:
+            raise NotImplementedError
+
+        post_result = self.postprocess_op(preds, shape_list)
+        dt_boxes = post_result[0]["points"]
+
+        if self.args.det_box_type == "poly":
+            dt_boxes = self.filter_tag_det_res_only_clip(dt_boxes, ori_im.shape)
+        else:
+            dt_boxes = self.filter_tag_det_res(dt_boxes, ori_im.shape)
+
+        if self.args.benchmark:
+            self.autolog.times.end(stamp=True)
+        et = time.time()
+        return dt_boxes, et - st
+
+    def __call__(self, img, use_slice=False):
+        # For image like poster with one side much greater than the other side,
+        # splitting recursively and processing with overlap to enhance performance.
+        MIN_BOUND_DISTANCE = 50
+        dt_boxes = np.zeros((0, 4, 2), dtype=np.float32)
+        elapse = 0
+        if (
+            img.shape[0] / img.shape[1] > 2
+            and img.shape[0] > self.args.det_limit_side_len
+            and use_slice
+        ):
+            start_h = 0
+            end_h = 0
+            while end_h <= img.shape[0]:
+                end_h = start_h + img.shape[1] * 3 // 4
+                subimg = img[start_h:end_h, :]
+                if len(subimg) == 0:
+                    break
+                sub_dt_boxes, sub_elapse = self.predict(subimg)
+                offset = start_h
+                # To prevent text blocks from being cut off, roll back a certain buffer area.
+                if (
+                    len(sub_dt_boxes) == 0
+                    or img.shape[1] - max([x[-1][1] for x in sub_dt_boxes])
+                    > MIN_BOUND_DISTANCE
+                ):
+                    start_h = end_h
+                else:
+                    sorted_indices = np.argsort(sub_dt_boxes[:, 2, 1])
+                    sub_dt_boxes = sub_dt_boxes[sorted_indices]
+                    bottom_line = (
+                        0
+                        if len(sub_dt_boxes) <= 1
+                        else int(np.max(sub_dt_boxes[:-1, 2, 1]))
+                    )
+                    if bottom_line > 0:
+                        start_h += bottom_line
+                        sub_dt_boxes = sub_dt_boxes[
+                            sub_dt_boxes[:, 2, 1] <= bottom_line
+                        ]
+                    else:
+                        start_h = end_h
+                if len(sub_dt_boxes) > 0:
+                    if dt_boxes.shape[0] == 0:
+                        dt_boxes = sub_dt_boxes + np.array(
+                            [0, offset], dtype=np.float32
+                        )
+                    else:
+                        dt_boxes = np.append(
+                            dt_boxes,
+                            sub_dt_boxes + np.array([0, offset], dtype=np.float32),
+                            axis=0,
+                        )
+                elapse += sub_elapse
+        elif (
+            img.shape[1] / img.shape[0] > 3
+            and img.shape[1] > self.args.det_limit_side_len * 3
+            and use_slice
+        ):
+            start_w = 0
+            end_w = 0
+            while end_w <= img.shape[1]:
+                end_w = start_w + img.shape[0] * 3 // 4
+                subimg = img[:, start_w:end_w]
+                if len(subimg) == 0:
+                    break
+                sub_dt_boxes, sub_elapse = self.predict(subimg)
+                offset = start_w
+                if (
+                    len(sub_dt_boxes) == 0
+                    or img.shape[0] - max([x[-1][0] for x in sub_dt_boxes])
+                    > MIN_BOUND_DISTANCE
+                ):
+                    start_w = end_w
+                else:
+                    sorted_indices = np.argsort(sub_dt_boxes[:, 2, 0])
+                    sub_dt_boxes = sub_dt_boxes[sorted_indices]
+                    right_line = (
+                        0
+                        if len(sub_dt_boxes) <= 1
+                        else int(np.max(sub_dt_boxes[:-1, 1, 0]))
+                    )
+                    if right_line > 0:
+                        start_w += right_line
+                        sub_dt_boxes = sub_dt_boxes[sub_dt_boxes[:, 1, 0] <= right_line]
+                    else:
+                        start_w = end_w
+                if len(sub_dt_boxes) > 0:
+                    if dt_boxes.shape[0] == 0:
+                        dt_boxes = sub_dt_boxes + np.array(
+                            [offset, 0], dtype=np.float32
+                        )
+                    else:
+                        dt_boxes = np.append(
+                            dt_boxes,
+                            sub_dt_boxes + np.array([offset, 0], dtype=np.float32),
+                            axis=0,
+                        )
+                elapse += sub_elapse
+        else:
+            dt_boxes, elapse = self.predict(img)
+        return dt_boxes, elapse
+
+
+if __name__ == "__main__":
+    args = utility.parse_args()
+    image_file_list = get_image_file_list(args.image_dir)
+    total_time = 0
+    draw_img_save_dir = args.draw_img_save_dir
+    os.makedirs(draw_img_save_dir, exist_ok=True)
+
+    # logger
+    log_file = args.save_log_path
+    if os.path.isdir(args.save_log_path) or (
+        not os.path.exists(args.save_log_path) and args.save_log_path.endswith("/")
+    ):
+        log_file = os.path.join(log_file, "benchmark_detection.log")
+    logger = get_logger(log_file=log_file)
+
+    # create text detector
+    text_detector = TextDetector(args, logger)
+
+    if args.warmup:
+        img = np.random.uniform(0, 255, [640, 640, 3]).astype(np.uint8)
+        for i in range(2):
+            res = text_detector(img)
+
+    save_results = []
+    for idx, image_file in enumerate(image_file_list):
+        img, flag_gif, flag_pdf = check_and_read(image_file)
+        if not flag_gif and not flag_pdf:
+            img = cv2.imread(image_file)
+        if not flag_pdf:
+            if img is None:
+                logger.debug("error in loading image:{}".format(image_file))
+                continue
+            imgs = [img]
+        else:
+            page_num = args.page_num
+            if page_num > len(img) or page_num == 0:
+                page_num = len(img)
+            imgs = img[:page_num]
+        for index, img in enumerate(imgs):
+            st = time.time()
+            dt_boxes, _ = text_detector(img)
+            elapse = time.time() - st
+            total_time += elapse
+            if len(imgs) > 1:
+                save_pred = (
+                    os.path.basename(image_file)
+                    + "_"
+                    + str(index)
+                    + "\t"
+                    + str(json.dumps([x.tolist() for x in dt_boxes]))
+                    + "\n"
+                )
+            else:
+                save_pred = (
+                    os.path.basename(image_file)
+                    + "\t"
+                    + str(json.dumps([x.tolist() for x in dt_boxes]))
+                    + "\n"
+                )
+            save_results.append(save_pred)
+            logger.info(save_pred)
+            if len(imgs) > 1:
+                logger.info(
+                    "{}_{} The predict time of {}: {}".format(
+                        idx, index, image_file, elapse
+                    )
+                )
+            else:
+                logger.info(
+                    "{} The predict time of {}: {}".format(idx, image_file, elapse)
+                )
+
+            src_im = utility.draw_text_det_res(dt_boxes, img)
+
+            if flag_gif:
+                save_file = image_file[:-3] + "png"
+            elif flag_pdf:
+                save_file = image_file.replace(".pdf", "_" + str(index) + ".png")
+            else:
+                save_file = image_file
+            img_path = os.path.join(
+                draw_img_save_dir, "det_res_{}".format(os.path.basename(save_file))
+            )
+            cv2.imwrite(img_path, src_im)
+            logger.info("The visualized image saved in {}".format(img_path))
+
+    with open(os.path.join(draw_img_save_dir, "det_results.txt"), "w") as f:
+        f.writelines(save_results)
+        f.close()
+    if args.benchmark:
+        text_detector.autolog.report()
diff --git a/tools/infer/predict_e2e.py b/tools/infer/predict_e2e.py
new file mode 100755
index 0000000..95eb792
--- /dev/null
+++ b/tools/infer/predict_e2e.py
@@ -0,0 +1,178 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "../..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+
+import cv2
+import numpy as np
+import time
+import sys
+
+import tools.infer.utility as utility
+from ppocr.utils.logging import get_logger
+from ppocr.utils.utility import get_image_file_list, check_and_read
+from ppocr.data import create_operators, transform
+from ppocr.postprocess import build_post_process
+
+logger = get_logger()
+
+
+class TextE2E(object):
+    def __init__(self, args):
+        if os.path.exists(f"{args.e2e_model_dir}/inference.yml"):
+            model_config = utility.load_config(f"{args.e2e_model_dir}/inference.yml")
+            model_name = model_config.get("Global", {}).get("model_name", "")
+            if model_name:
+                raise ValueError(
+                    f"{model_name} is not supported. Please check if the model is supported by the PaddleOCR wheel."
+                )
+
+        self.args = args
+        self.e2e_algorithm = args.e2e_algorithm
+        self.use_onnx = args.use_onnx
+        pre_process_list = [
+            {"E2EResizeForTest": {}},
+            {
+                "NormalizeImage": {
+                    "std": [0.229, 0.224, 0.225],
+                    "mean": [0.485, 0.456, 0.406],
+                    "scale": "1./255.",
+                    "order": "hwc",
+                }
+            },
+            {"ToCHWImage": None},
+            {"KeepKeys": {"keep_keys": ["image", "shape"]}},
+        ]
+        postprocess_params = {}
+        if self.e2e_algorithm == "PGNet":
+            pre_process_list[0] = {
+                "E2EResizeForTest": {
+                    "max_side_len": args.e2e_limit_side_len,
+                    "valid_set": "totaltext",
+                }
+            }
+            postprocess_params["name"] = "PGPostProcess"
+            postprocess_params["score_thresh"] = args.e2e_pgnet_score_thresh
+            postprocess_params["character_dict_path"] = args.e2e_char_dict_path
+            postprocess_params["valid_set"] = args.e2e_pgnet_valid_set
+            postprocess_params["mode"] = args.e2e_pgnet_mode
+        else:
+            logger.info("unknown e2e_algorithm:{}".format(self.e2e_algorithm))
+            sys.exit(0)
+
+        self.preprocess_op = create_operators(pre_process_list)
+        self.postprocess_op = build_post_process(postprocess_params)
+        (
+            self.predictor,
+            self.input_tensor,
+            self.output_tensors,
+            _,
+        ) = utility.create_predictor(
+            args, "e2e", logger
+        )  # paddle.jit.load(args.det_model_dir)
+        # self.predictor.eval()
+
+    def clip_det_res(self, points, img_height, img_width):
+        for pno in range(points.shape[0]):
+            points[pno, 0] = int(min(max(points[pno, 0], 0), img_width - 1))
+            points[pno, 1] = int(min(max(points[pno, 1], 0), img_height - 1))
+        return points
+
+    def filter_tag_det_res_only_clip(self, dt_boxes, image_shape):
+        img_height, img_width = image_shape[0:2]
+        dt_boxes_new = []
+        for box in dt_boxes:
+            box = self.clip_det_res(box, img_height, img_width)
+            dt_boxes_new.append(box)
+        dt_boxes = np.array(dt_boxes_new)
+        return dt_boxes
+
+    def __call__(self, img):
+        ori_im = img.copy()
+        data = {"image": img}
+        data = transform(data, self.preprocess_op)
+        img, shape_list = data
+        if img is None:
+            return None, 0
+        img = np.expand_dims(img, axis=0)
+        shape_list = np.expand_dims(shape_list, axis=0)
+        img = img.copy()
+        starttime = time.time()
+
+        if self.use_onnx:
+            input_dict = {}
+            input_dict[self.input_tensor.name] = img
+            outputs = self.predictor.run(self.output_tensors, input_dict)
+            preds = {}
+            preds["f_border"] = outputs[0]
+            preds["f_char"] = outputs[1]
+            preds["f_direction"] = outputs[2]
+            preds["f_score"] = outputs[3]
+        else:
+            self.input_tensor.copy_from_cpu(img)
+            self.predictor.run()
+            outputs = []
+            for output_tensor in self.output_tensors:
+                output = output_tensor.copy_to_cpu()
+                outputs.append(output)
+
+            preds = {}
+            if self.e2e_algorithm == "PGNet":
+                preds["f_border"] = outputs[0]
+                preds["f_char"] = outputs[1]
+                preds["f_direction"] = outputs[2]
+                preds["f_score"] = outputs[3]
+            else:
+                raise NotImplementedError
+        post_result = self.postprocess_op(preds, shape_list)
+        points, strs = post_result["points"], post_result["texts"]
+        dt_boxes = self.filter_tag_det_res_only_clip(points, ori_im.shape)
+        elapse = time.time() - starttime
+        return dt_boxes, strs, elapse
+
+
+if __name__ == "__main__":
+    args = utility.parse_args()
+    image_file_list = get_image_file_list(args.image_dir)
+    text_detector = TextE2E(args)
+    count = 0
+    total_time = 0
+    draw_img_save = "./inference_results"
+    if not os.path.exists(draw_img_save):
+        os.makedirs(draw_img_save)
+    for image_file in image_file_list:
+        img, flag, _ = check_and_read(image_file)
+        if not flag:
+            img = cv2.imread(image_file)
+        if img is None:
+            logger.info("error in loading image:{}".format(image_file))
+            continue
+        points, strs, elapse = text_detector(img)
+        if count > 0:
+            total_time += elapse
+        count += 1
+        logger.info("Predict time of {}: {}".format(image_file, elapse))
+        src_im = utility.draw_e2e_res(points, strs, image_file)
+        img_name_pure = os.path.split(image_file)[-1]
+        img_path = os.path.join(draw_img_save, "e2e_res_{}".format(img_name_pure))
+        cv2.imwrite(img_path, src_im)
+        logger.info("The visualized image saved in {}".format(img_path))
+    if count > 1:
+        logger.info("Avg Time: {}".format(total_time / (count - 1)))
diff --git a/tools/infer/predict_rec.py b/tools/infer/predict_rec.py
new file mode 100755
index 0000000..b53f58d
--- /dev/null
+++ b/tools/infer/predict_rec.py
@@ -0,0 +1,916 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import sys
+from PIL import Image
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "../..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+
+import cv2
+import numpy as np
+import math
+import time
+import traceback
+import paddle
+
+import tools.infer.utility as utility
+from ppocr.postprocess import build_post_process
+from ppocr.utils.logging import get_logger
+from ppocr.utils.utility import get_image_file_list, check_and_read
+
+logger = get_logger()
+
+
+class TextRecognizer(object):
+    def __init__(self, args, logger=None):
+        if os.path.exists(f"{args.rec_model_dir}/inference.yml"):
+            model_config = utility.load_config(f"{args.rec_model_dir}/inference.yml")
+            model_name = model_config.get("Global", {}).get("model_name", "")
+            if model_name and model_name not in [
+                "PP-OCRv5_mobile_rec",
+                "PP-OCRv5_server_rec",
+                "korean_PP-OCRv5_mobile_rec",
+                "eslav_PP-OCRv5_mobile_rec",
+                "latin_PP-OCRv5_mobile_rec",
+                "en_PP-OCRv5_mobile_rec",
+                "th_PP-OCRv5_mobile_rec",
+                "el_PP-OCRv5_mobile_rec",
+                "PP-OCRv6_tiny_rec",
+                "PP-OCRv6_small_rec",
+                "PP-OCRv6_medium_rec",
+            ]:
+                raise ValueError(
+                    f"{model_name} is not supported. Please check if the model is supported by the PaddleOCR wheel."
+                )
+
+            if args.rec_char_dict_path == "./ppocr/utils/ppocr_keys_v1.txt":
+                rec_char_list = model_config.get("PostProcess", {}).get(
+                    "character_dict", []
+                )
+                if rec_char_list:
+                    new_rec_char_dict_path = f"{args.rec_model_dir}/ppocr_keys.txt"
+                    with open(new_rec_char_dict_path, "w", encoding="utf-8") as f:
+                        f.writelines([char + "\n" for char in rec_char_list])
+                    args.rec_char_dict_path = new_rec_char_dict_path
+
+        if logger is None:
+            logger = get_logger()
+        self.rec_image_shape = [int(v) for v in args.rec_image_shape.split(",")]
+        self.rec_batch_num = args.rec_batch_num
+        self.rec_algorithm = args.rec_algorithm
+        postprocess_params = {
+            "name": "CTCLabelDecode",
+            "character_dict_path": args.rec_char_dict_path,
+            "use_space_char": args.use_space_char,
+        }
+        if self.rec_algorithm == "SRN":
+            postprocess_params = {
+                "name": "SRNLabelDecode",
+                "character_dict_path": args.rec_char_dict_path,
+                "use_space_char": args.use_space_char,
+            }
+        elif self.rec_algorithm == "RARE":
+            postprocess_params = {
+                "name": "AttnLabelDecode",
+                "character_dict_path": args.rec_char_dict_path,
+                "use_space_char": args.use_space_char,
+            }
+        elif self.rec_algorithm == "NRTR":
+            postprocess_params = {
+                "name": "NRTRLabelDecode",
+                "character_dict_path": args.rec_char_dict_path,
+                "use_space_char": args.use_space_char,
+            }
+        elif self.rec_algorithm == "SAR":
+            postprocess_params = {
+                "name": "SARLabelDecode",
+                "character_dict_path": args.rec_char_dict_path,
+                "use_space_char": args.use_space_char,
+            }
+        elif self.rec_algorithm == "VisionLAN":
+            postprocess_params = {
+                "name": "VLLabelDecode",
+                "character_dict_path": args.rec_char_dict_path,
+                "use_space_char": args.use_space_char,
+                "max_text_length": args.max_text_length,
+            }
+        elif self.rec_algorithm == "ViTSTR":
+            postprocess_params = {
+                "name": "ViTSTRLabelDecode",
+                "character_dict_path": args.rec_char_dict_path,
+                "use_space_char": args.use_space_char,
+            }
+        elif self.rec_algorithm == "ABINet":
+            postprocess_params = {
+                "name": "ABINetLabelDecode",
+                "character_dict_path": args.rec_char_dict_path,
+                "use_space_char": args.use_space_char,
+            }
+        elif self.rec_algorithm == "SPIN":
+            postprocess_params = {
+                "name": "SPINLabelDecode",
+                "character_dict_path": args.rec_char_dict_path,
+                "use_space_char": args.use_space_char,
+            }
+        elif self.rec_algorithm == "RobustScanner":
+            postprocess_params = {
+                "name": "SARLabelDecode",
+                "character_dict_path": args.rec_char_dict_path,
+                "use_space_char": args.use_space_char,
+                "rm_symbol": True,
+            }
+        elif self.rec_algorithm == "RFL":
+            postprocess_params = {
+                "name": "RFLLabelDecode",
+                "character_dict_path": None,
+                "use_space_char": args.use_space_char,
+            }
+        elif self.rec_algorithm == "SATRN":
+            postprocess_params = {
+                "name": "SATRNLabelDecode",
+                "character_dict_path": args.rec_char_dict_path,
+                "use_space_char": args.use_space_char,
+                "rm_symbol": True,
+            }
+        elif self.rec_algorithm in ["CPPD", "CPPDPadding"]:
+            postprocess_params = {
+                "name": "CPPDLabelDecode",
+                "character_dict_path": args.rec_char_dict_path,
+                "use_space_char": args.use_space_char,
+                "rm_symbol": True,
+            }
+        elif self.rec_algorithm == "PREN":
+            postprocess_params = {"name": "PRENLabelDecode"}
+        elif self.rec_algorithm == "CAN":
+            self.inverse = args.rec_image_inverse
+            postprocess_params = {
+                "name": "CANLabelDecode",
+                "character_dict_path": args.rec_char_dict_path,
+                "use_space_char": args.use_space_char,
+            }
+        elif self.rec_algorithm == "LaTeXOCR":
+            postprocess_params = {
+                "name": "LaTeXOCRDecode",
+                "rec_char_dict_path": args.rec_char_dict_path,
+            }
+        elif self.rec_algorithm == "ParseQ":
+            postprocess_params = {
+                "name": "ParseQLabelDecode",
+                "character_dict_path": args.rec_char_dict_path,
+                "use_space_char": args.use_space_char,
+            }
+        self.postprocess_op = build_post_process(postprocess_params)
+        self.postprocess_params = postprocess_params
+        (
+            self.predictor,
+            self.input_tensor,
+            self.output_tensors,
+            self.config,
+        ) = utility.create_predictor(args, "rec", logger)
+        self.benchmark = args.benchmark
+        self.use_onnx = args.use_onnx
+        if args.benchmark:
+            import auto_log
+
+            pid = os.getpid()
+            gpu_id = utility.get_infer_gpuid()
+            self.autolog = auto_log.AutoLogger(
+                model_name="rec",
+                model_precision=args.precision,
+                batch_size=args.rec_batch_num,
+                data_shape="dynamic",
+                save_path=None,  # not used if logger is not None
+                inference_config=self.config,
+                pids=pid,
+                process_name=None,
+                gpu_ids=gpu_id if args.use_gpu else None,
+                time_keys=["preprocess_time", "inference_time", "postprocess_time"],
+                warmup=0,
+                logger=logger,
+            )
+        self.return_word_box = args.return_word_box
+
+    def resize_norm_img(self, img, max_wh_ratio):
+        imgC, imgH, imgW = self.rec_image_shape
+        if self.rec_algorithm == "NRTR" or self.rec_algorithm == "ViTSTR":
+            img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+            # return padding_im
+            image_pil = Image.fromarray(np.uint8(img))
+            if self.rec_algorithm == "ViTSTR":
+                img = image_pil.resize([imgW, imgH], Image.BICUBIC)
+            else:
+                img = image_pil.resize([imgW, imgH], Image.Resampling.LANCZOS)
+            img = np.array(img)
+            norm_img = np.expand_dims(img, -1)
+            norm_img = norm_img.transpose((2, 0, 1))
+            if self.rec_algorithm == "ViTSTR":
+                norm_img = norm_img.astype(np.float32) / 255.0
+            else:
+                norm_img = norm_img.astype(np.float32) / 128.0 - 1.0
+            return norm_img
+        elif self.rec_algorithm == "RFL":
+            img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+            resized_image = cv2.resize(img, (imgW, imgH), interpolation=cv2.INTER_CUBIC)
+            resized_image = resized_image.astype("float32")
+            resized_image = resized_image / 255
+            resized_image = resized_image[np.newaxis, :]
+            resized_image -= 0.5
+            resized_image /= 0.5
+            return resized_image
+
+        assert imgC == img.shape[2]
+        imgW = int((imgH * max_wh_ratio))
+        if self.use_onnx:
+            w = self.input_tensor.shape[3:][0]
+            if isinstance(w, str):
+                pass
+            elif w is not None and w > 0:
+                imgW = w
+        h, w = img.shape[:2]
+        ratio = w / float(h)
+        if math.ceil(imgH * ratio) > imgW:
+            resized_w = imgW
+        else:
+            resized_w = int(math.ceil(imgH * ratio))
+        if self.rec_algorithm == "RARE":
+            if resized_w > self.rec_image_shape[2]:
+                resized_w = self.rec_image_shape[2]
+            imgW = self.rec_image_shape[2]
+        resized_image = cv2.resize(img, (resized_w, imgH))
+        resized_image = resized_image.astype("float32")
+        resized_image = resized_image.transpose((2, 0, 1)) / 255
+        resized_image -= 0.5
+        resized_image /= 0.5
+        padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
+        padding_im[:, :, 0:resized_w] = resized_image
+        return padding_im
+
+    def resize_norm_img_vl(self, img, image_shape):
+        imgC, imgH, imgW = image_shape
+        img = img[:, :, ::-1]  # bgr2rgb
+        resized_image = cv2.resize(img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
+        resized_image = resized_image.astype("float32")
+        resized_image = resized_image.transpose((2, 0, 1)) / 255
+        return resized_image
+
+    def resize_norm_img_srn(self, img, image_shape):
+        imgC, imgH, imgW = image_shape
+
+        img_black = np.zeros((imgH, imgW))
+        im_hei = img.shape[0]
+        im_wid = img.shape[1]
+
+        if im_wid <= im_hei * 1:
+            img_new = cv2.resize(img, (imgH * 1, imgH))
+        elif im_wid <= im_hei * 2:
+            img_new = cv2.resize(img, (imgH * 2, imgH))
+        elif im_wid <= im_hei * 3:
+            img_new = cv2.resize(img, (imgH * 3, imgH))
+        else:
+            img_new = cv2.resize(img, (imgW, imgH))
+
+        img_np = np.asarray(img_new)
+        img_np = cv2.cvtColor(img_np, cv2.COLOR_BGR2GRAY)
+        img_black[:, 0 : img_np.shape[1]] = img_np
+        img_black = img_black[:, :, np.newaxis]
+
+        row, col, c = img_black.shape
+        c = 1
+
+        return np.reshape(img_black, (c, row, col)).astype(np.float32)
+
+    def srn_other_inputs(self, image_shape, num_heads, max_text_length):
+        imgC, imgH, imgW = image_shape
+        feature_dim = int((imgH / 8) * (imgW / 8))
+
+        encoder_word_pos = (
+            np.array(range(0, feature_dim)).reshape((feature_dim, 1)).astype("int64")
+        )
+        gsrm_word_pos = (
+            np.array(range(0, max_text_length))
+            .reshape((max_text_length, 1))
+            .astype("int64")
+        )
+
+        gsrm_attn_bias_data = np.ones((1, max_text_length, max_text_length))
+        gsrm_slf_attn_bias1 = np.triu(gsrm_attn_bias_data, 1).reshape(
+            [-1, 1, max_text_length, max_text_length]
+        )
+        gsrm_slf_attn_bias1 = np.tile(gsrm_slf_attn_bias1, [1, num_heads, 1, 1]).astype(
+            "float32"
+        ) * [-1e9]
+
+        gsrm_slf_attn_bias2 = np.tril(gsrm_attn_bias_data, -1).reshape(
+            [-1, 1, max_text_length, max_text_length]
+        )
+        gsrm_slf_attn_bias2 = np.tile(gsrm_slf_attn_bias2, [1, num_heads, 1, 1]).astype(
+            "float32"
+        ) * [-1e9]
+
+        encoder_word_pos = encoder_word_pos[np.newaxis, :]
+        gsrm_word_pos = gsrm_word_pos[np.newaxis, :]
+
+        return [
+            encoder_word_pos,
+            gsrm_word_pos,
+            gsrm_slf_attn_bias1,
+            gsrm_slf_attn_bias2,
+        ]
+
+    def process_image_srn(self, img, image_shape, num_heads, max_text_length):
+        norm_img = self.resize_norm_img_srn(img, image_shape)
+        norm_img = norm_img[np.newaxis, :]
+
+        [
+            encoder_word_pos,
+            gsrm_word_pos,
+            gsrm_slf_attn_bias1,
+            gsrm_slf_attn_bias2,
+        ] = self.srn_other_inputs(image_shape, num_heads, max_text_length)
+
+        gsrm_slf_attn_bias1 = gsrm_slf_attn_bias1.astype(np.float32)
+        gsrm_slf_attn_bias2 = gsrm_slf_attn_bias2.astype(np.float32)
+        encoder_word_pos = encoder_word_pos.astype(np.int64)
+        gsrm_word_pos = gsrm_word_pos.astype(np.int64)
+
+        return (
+            norm_img,
+            encoder_word_pos,
+            gsrm_word_pos,
+            gsrm_slf_attn_bias1,
+            gsrm_slf_attn_bias2,
+        )
+
+    def resize_norm_img_sar(self, img, image_shape, width_downsample_ratio=0.25):
+        imgC, imgH, imgW_min, imgW_max = image_shape
+        h = img.shape[0]
+        w = img.shape[1]
+        valid_ratio = 1.0
+        # make sure new_width is an integral multiple of width_divisor.
+        width_divisor = int(1 / width_downsample_ratio)
+        # resize
+        ratio = w / float(h)
+        resize_w = math.ceil(imgH * ratio)
+        if resize_w % width_divisor != 0:
+            resize_w = round(resize_w / width_divisor) * width_divisor
+        if imgW_min is not None:
+            resize_w = max(imgW_min, resize_w)
+        if imgW_max is not None:
+            valid_ratio = min(1.0, 1.0 * resize_w / imgW_max)
+            resize_w = min(imgW_max, resize_w)
+        resized_image = cv2.resize(img, (resize_w, imgH))
+        resized_image = resized_image.astype("float32")
+        # norm
+        if image_shape[0] == 1:
+            resized_image = resized_image / 255
+            resized_image = resized_image[np.newaxis, :]
+        else:
+            resized_image = resized_image.transpose((2, 0, 1)) / 255
+        resized_image -= 0.5
+        resized_image /= 0.5
+        resize_shape = resized_image.shape
+        padding_im = -1.0 * np.ones((imgC, imgH, imgW_max), dtype=np.float32)
+        padding_im[:, :, 0:resize_w] = resized_image
+        pad_shape = padding_im.shape
+
+        return padding_im, resize_shape, pad_shape, valid_ratio
+
+    def resize_norm_img_spin(self, img):
+        img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+        # return padding_im
+        img = cv2.resize(img, tuple([100, 32]), cv2.INTER_CUBIC)
+        img = np.array(img, np.float32)
+        img = np.expand_dims(img, -1)
+        img = img.transpose((2, 0, 1))
+        mean = [127.5]
+        std = [127.5]
+        mean = np.array(mean, dtype=np.float32)
+        std = np.array(std, dtype=np.float32)
+        mean = np.float32(mean.reshape(1, -1))
+        stdinv = 1 / np.float32(std.reshape(1, -1))
+        img -= mean
+        img *= stdinv
+        return img
+
+    def resize_norm_img_svtr(self, img, image_shape):
+        imgC, imgH, imgW = image_shape
+        max_wh_ratio = imgW * 1.0 / imgH
+        h, w = img.shape[0], img.shape[1]
+        ratio = w * 1.0 / h
+        max_wh_ratio = min(max(max_wh_ratio, ratio), max_wh_ratio)
+        imgW = int(imgH * max_wh_ratio)
+        if math.ceil(imgH * ratio) > imgW:
+            resized_w = imgW
+        else:
+            resized_w = int(math.ceil(imgH * ratio))
+        resized_image = cv2.resize(img, (resized_w, imgH))
+        resized_image = resized_image.astype("float32")
+        resized_image = resized_image.transpose((2, 0, 1)) / 255
+        resized_image -= 0.5
+        resized_image /= 0.5
+        padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
+        padding_im[:, :, 0:resized_w] = resized_image
+        return padding_im
+
+    def resize_norm_img_cppd_padding(
+        self, img, image_shape, padding=True, interpolation=cv2.INTER_LINEAR
+    ):
+        imgC, imgH, imgW = image_shape
+        h = img.shape[0]
+        w = img.shape[1]
+        if not padding:
+            resized_image = cv2.resize(img, (imgW, imgH), interpolation=interpolation)
+            resized_w = imgW
+        else:
+            ratio = w / float(h)
+            if math.ceil(imgH * ratio) > imgW:
+                resized_w = imgW
+            else:
+                resized_w = int(math.ceil(imgH * ratio))
+            resized_image = cv2.resize(img, (resized_w, imgH))
+        resized_image = resized_image.astype("float32")
+        if image_shape[0] == 1:
+            resized_image = resized_image / 255
+            resized_image = resized_image[np.newaxis, :]
+        else:
+            resized_image = resized_image.transpose((2, 0, 1)) / 255
+        resized_image -= 0.5
+        resized_image /= 0.5
+        padding_im = np.zeros((imgC, imgH, imgW), dtype=np.float32)
+        padding_im[:, :, 0:resized_w] = resized_image
+
+        return padding_im
+
+    def resize_norm_img_abinet(self, img, image_shape):
+        imgC, imgH, imgW = image_shape
+
+        resized_image = cv2.resize(img, (imgW, imgH), interpolation=cv2.INTER_LINEAR)
+        resized_image = resized_image.astype("float32")
+        resized_image = resized_image / 255.0
+
+        mean = np.array([0.485, 0.456, 0.406])
+        std = np.array([0.229, 0.224, 0.225])
+        resized_image = (resized_image - mean[None, None, ...]) / std[None, None, ...]
+        resized_image = resized_image.transpose((2, 0, 1))
+        resized_image = resized_image.astype("float32")
+
+        return resized_image
+
+    def norm_img_can(self, img, image_shape):
+        img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)  # CAN only predict gray scale image
+
+        if self.inverse:
+            img = 255 - img
+
+        if self.rec_image_shape[0] == 1:
+            h, w = img.shape
+            _, imgH, imgW = self.rec_image_shape
+            if h < imgH or w < imgW:
+                padding_h = max(imgH - h, 0)
+                padding_w = max(imgW - w, 0)
+                img_padded = np.pad(
+                    img,
+                    ((0, padding_h), (0, padding_w)),
+                    "constant",
+                    constant_values=(255),
+                )
+                img = img_padded
+
+        img = np.expand_dims(img, 0) / 255.0  # h,w,c -> c,h,w
+        img = img.astype("float32")
+
+        return img
+
+    def pad_(self, img, divable=32):
+        threshold = 128
+        data = np.array(img.convert("LA"))
+        if data[..., -1].var() == 0:
+            data = (data[..., 0]).astype(np.uint8)
+        else:
+            data = (255 - data[..., -1]).astype(np.uint8)
+        data = (data - data.min()) / (data.max() - data.min()) * 255
+        if data.mean() > threshold:
+            # To invert the text to white
+            gray = 255 * (data < threshold).astype(np.uint8)
+        else:
+            gray = 255 * (data > threshold).astype(np.uint8)
+            data = 255 - data
+
+        coords = cv2.findNonZero(gray)  # Find all non-zero points (text)
+        a, b, w, h = cv2.boundingRect(coords)  # Find minimum spanning bounding box
+        rect = data[b : b + h, a : a + w]
+        im = Image.fromarray(rect).convert("L")
+        dims = []
+        for x in [w, h]:
+            div, mod = divmod(x, divable)
+            dims.append(divable * (div + (1 if mod > 0 else 0)))
+        padded = Image.new("L", dims, 255)
+        padded.paste(im, (0, 0, im.size[0], im.size[1]))
+        return padded
+
+    def minmax_size_(
+        self,
+        img,
+        max_dimensions,
+        min_dimensions,
+    ):
+        if max_dimensions is not None:
+            ratios = [a / b for a, b in zip(img.size, max_dimensions)]
+            if any([r > 1 for r in ratios]):
+                size = np.array(img.size) // max(ratios)
+                img = img.resize(tuple(size.astype(int)), Image.BILINEAR)
+        if min_dimensions is not None:
+            # hypothesis: there is a dim in img smaller than min_dimensions, and return a proper dim >= min_dimensions
+            padded_size = [
+                max(img_dim, min_dim)
+                for img_dim, min_dim in zip(img.size, min_dimensions)
+            ]
+            if padded_size != list(img.size):  # assert hypothesis
+                padded_im = Image.new("L", padded_size, 255)
+                padded_im.paste(img, img.getbbox())
+                img = padded_im
+        return img
+
+    def norm_img_latexocr(self, img):
+        # CAN only predict gray scale image
+        shape = (1, 1, 3)
+        mean = [0.7931, 0.7931, 0.7931]
+        std = [0.1738, 0.1738, 0.1738]
+        scale = np.float32(1.0 / 255.0)
+        min_dimensions = [32, 32]
+        max_dimensions = [672, 192]
+        mean = np.array(mean).reshape(shape).astype("float32")
+        std = np.array(std).reshape(shape).astype("float32")
+
+        im_h, im_w = img.shape[:2]
+        if (
+            min_dimensions[0] <= im_w <= max_dimensions[0]
+            and min_dimensions[1] <= im_h <= max_dimensions[1]
+        ):
+            pass
+        else:
+            img = Image.fromarray(np.uint8(img))
+            img = self.minmax_size_(self.pad_(img), max_dimensions, min_dimensions)
+            img = np.array(img)
+            im_h, im_w = img.shape[:2]
+            img = np.dstack([img, img, img])
+        img = (img.astype("float32") * scale - mean) / std
+        img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+        divide_h = math.ceil(im_h / 16) * 16
+        divide_w = math.ceil(im_w / 16) * 16
+        img = np.pad(
+            img, ((0, divide_h - im_h), (0, divide_w - im_w)), constant_values=(1, 1)
+        )
+        img = img[:, :, np.newaxis].transpose(2, 0, 1)
+        img = img.astype("float32")
+        return img
+
+    def __call__(self, img_list):
+        img_num = len(img_list)
+        # Calculate the aspect ratio of all text bars
+        width_list = []
+        for img in img_list:
+            width_list.append(img.shape[1] / float(img.shape[0]))
+        # Sorting can speed up the recognition process
+        indices = np.argsort(np.array(width_list))
+        rec_res = [["", 0.0]] * img_num
+        batch_num = self.rec_batch_num
+        st = time.time()
+        if self.benchmark:
+            self.autolog.times.start()
+        for beg_img_no in range(0, img_num, batch_num):
+            end_img_no = min(img_num, beg_img_no + batch_num)
+            norm_img_batch = []
+            if self.rec_algorithm == "SRN":
+                encoder_word_pos_list = []
+                gsrm_word_pos_list = []
+                gsrm_slf_attn_bias1_list = []
+                gsrm_slf_attn_bias2_list = []
+            if self.rec_algorithm == "SAR":
+                valid_ratios = []
+            imgC, imgH, imgW = self.rec_image_shape[:3]
+            max_wh_ratio = imgW / imgH
+            wh_ratio_list = []
+            for ino in range(beg_img_no, end_img_no):
+                h, w = img_list[indices[ino]].shape[0:2]
+                wh_ratio = w * 1.0 / h
+                max_wh_ratio = max(max_wh_ratio, wh_ratio)
+                wh_ratio_list.append(wh_ratio)
+            for ino in range(beg_img_no, end_img_no):
+                if self.rec_algorithm == "SAR":
+                    norm_img, _, _, valid_ratio = self.resize_norm_img_sar(
+                        img_list[indices[ino]], self.rec_image_shape
+                    )
+                    norm_img = norm_img[np.newaxis, :]
+                    valid_ratio = np.expand_dims(valid_ratio, axis=0)
+                    valid_ratios.append(valid_ratio)
+                    norm_img_batch.append(norm_img)
+                elif self.rec_algorithm == "SRN":
+                    norm_img = self.process_image_srn(
+                        img_list[indices[ino]], self.rec_image_shape, 8, 25
+                    )
+                    encoder_word_pos_list.append(norm_img[1])
+                    gsrm_word_pos_list.append(norm_img[2])
+                    gsrm_slf_attn_bias1_list.append(norm_img[3])
+                    gsrm_slf_attn_bias2_list.append(norm_img[4])
+                    norm_img_batch.append(norm_img[0])
+                elif self.rec_algorithm in ["SVTR", "SATRN", "ParseQ", "CPPD"]:
+                    norm_img = self.resize_norm_img_svtr(
+                        img_list[indices[ino]], self.rec_image_shape
+                    )
+                    norm_img = norm_img[np.newaxis, :]
+                    norm_img_batch.append(norm_img)
+                elif self.rec_algorithm in ["CPPDPadding"]:
+                    norm_img = self.resize_norm_img_cppd_padding(
+                        img_list[indices[ino]], self.rec_image_shape
+                    )
+                    norm_img = norm_img[np.newaxis, :]
+                    norm_img_batch.append(norm_img)
+                elif self.rec_algorithm in ["VisionLAN", "PREN"]:
+                    norm_img = self.resize_norm_img_vl(
+                        img_list[indices[ino]], self.rec_image_shape
+                    )
+                    norm_img = norm_img[np.newaxis, :]
+                    norm_img_batch.append(norm_img)
+                elif self.rec_algorithm == "SPIN":
+                    norm_img = self.resize_norm_img_spin(img_list[indices[ino]])
+                    norm_img = norm_img[np.newaxis, :]
+                    norm_img_batch.append(norm_img)
+                elif self.rec_algorithm == "ABINet":
+                    norm_img = self.resize_norm_img_abinet(
+                        img_list[indices[ino]], self.rec_image_shape
+                    )
+                    norm_img = norm_img[np.newaxis, :]
+                    norm_img_batch.append(norm_img)
+                elif self.rec_algorithm == "RobustScanner":
+                    norm_img, _, _, valid_ratio = self.resize_norm_img_sar(
+                        img_list[indices[ino]],
+                        self.rec_image_shape,
+                        width_downsample_ratio=0.25,
+                    )
+                    norm_img = norm_img[np.newaxis, :]
+                    valid_ratio = np.expand_dims(valid_ratio, axis=0)
+                    valid_ratios = []
+                    valid_ratios.append(valid_ratio)
+                    norm_img_batch.append(norm_img)
+                    word_positions_list = []
+                    word_positions = np.array(range(0, 40)).astype("int64")
+                    word_positions = np.expand_dims(word_positions, axis=0)
+                    word_positions_list.append(word_positions)
+                elif self.rec_algorithm == "CAN":
+                    norm_img = self.norm_img_can(img_list[indices[ino]], max_wh_ratio)
+                    norm_img = norm_img[np.newaxis, :]
+                    norm_img_batch.append(norm_img)
+                    norm_image_mask = np.ones(norm_img.shape, dtype="float32")
+                    word_label = np.ones([1, 36], dtype="int64")
+                    norm_img_mask_batch = []
+                    word_label_list = []
+                    norm_img_mask_batch.append(norm_image_mask)
+                    word_label_list.append(word_label)
+                elif self.rec_algorithm == "LaTeXOCR":
+                    norm_img = self.norm_img_latexocr(img_list[indices[ino]])
+                    norm_img = norm_img[np.newaxis, :]
+                    norm_img_batch.append(norm_img)
+                else:
+                    norm_img = self.resize_norm_img(
+                        img_list[indices[ino]], max_wh_ratio
+                    )
+                    norm_img = norm_img[np.newaxis, :]
+                    norm_img_batch.append(norm_img)
+            norm_img_batch = np.concatenate(norm_img_batch)
+            norm_img_batch = norm_img_batch.copy()
+            if self.benchmark:
+                self.autolog.times.stamp()
+
+            if self.rec_algorithm == "SRN":
+                encoder_word_pos_list = np.concatenate(encoder_word_pos_list)
+                gsrm_word_pos_list = np.concatenate(gsrm_word_pos_list)
+                gsrm_slf_attn_bias1_list = np.concatenate(gsrm_slf_attn_bias1_list)
+                gsrm_slf_attn_bias2_list = np.concatenate(gsrm_slf_attn_bias2_list)
+
+                inputs = [
+                    norm_img_batch,
+                    encoder_word_pos_list,
+                    gsrm_word_pos_list,
+                    gsrm_slf_attn_bias1_list,
+                    gsrm_slf_attn_bias2_list,
+                ]
+                if self.use_onnx:
+                    input_dict = {}
+                    input_dict[self.input_tensor.name] = norm_img_batch
+                    outputs = self.predictor.run(self.output_tensors, input_dict)
+                    preds = {"predict": outputs[2]}
+                else:
+                    input_names = self.predictor.get_input_names()
+                    for i in range(len(input_names)):
+                        input_tensor = self.predictor.get_input_handle(input_names[i])
+                        input_tensor.copy_from_cpu(inputs[i])
+                    self.predictor.run()
+                    outputs = []
+                    for output_tensor in self.output_tensors:
+                        output = output_tensor.copy_to_cpu()
+                        outputs.append(output)
+                    if self.benchmark:
+                        self.autolog.times.stamp()
+                    preds = {"predict": outputs[2]}
+            elif self.rec_algorithm == "SAR":
+                valid_ratios = np.concatenate(valid_ratios)
+                inputs = [
+                    norm_img_batch,
+                    np.array([valid_ratios], dtype=np.float32).T,
+                ]
+                if self.use_onnx:
+                    input_dict = {}
+                    input_dict[self.input_tensor.name] = norm_img_batch
+                    outputs = self.predictor.run(self.output_tensors, input_dict)
+                    preds = outputs[0]
+                else:
+                    input_names = self.predictor.get_input_names()
+                    for i in range(len(input_names)):
+                        input_tensor = self.predictor.get_input_handle(input_names[i])
+                        input_tensor.copy_from_cpu(inputs[i])
+                    self.predictor.run()
+                    outputs = []
+                    for output_tensor in self.output_tensors:
+                        output = output_tensor.copy_to_cpu()
+                        outputs.append(output)
+                    if self.benchmark:
+                        self.autolog.times.stamp()
+                    preds = outputs[0]
+            elif self.rec_algorithm == "RobustScanner":
+                valid_ratios = np.concatenate(valid_ratios)
+                word_positions_list = np.concatenate(word_positions_list)
+                inputs = [norm_img_batch, valid_ratios, word_positions_list]
+
+                if self.use_onnx:
+                    input_dict = {}
+                    input_dict[self.input_tensor.name] = norm_img_batch
+                    outputs = self.predictor.run(self.output_tensors, input_dict)
+                    preds = outputs[0]
+                else:
+                    input_names = self.predictor.get_input_names()
+                    for i in range(len(input_names)):
+                        input_tensor = self.predictor.get_input_handle(input_names[i])
+                        input_tensor.copy_from_cpu(inputs[i])
+                    self.predictor.run()
+                    outputs = []
+                    for output_tensor in self.output_tensors:
+                        output = output_tensor.copy_to_cpu()
+                        outputs.append(output)
+                    if self.benchmark:
+                        self.autolog.times.stamp()
+                    preds = outputs[0]
+            elif self.rec_algorithm == "CAN":
+                norm_img_mask_batch = np.concatenate(norm_img_mask_batch)
+                word_label_list = np.concatenate(word_label_list)
+                inputs = [norm_img_batch, norm_img_mask_batch, word_label_list]
+                if self.use_onnx:
+                    input_dict = {}
+                    input_dict[self.input_tensor.name] = norm_img_batch
+                    outputs = self.predictor.run(self.output_tensors, input_dict)
+                    preds = outputs
+                else:
+                    input_names = self.predictor.get_input_names()
+                    input_tensor = []
+                    for i in range(len(input_names)):
+                        input_tensor_i = self.predictor.get_input_handle(input_names[i])
+                        input_tensor_i.copy_from_cpu(inputs[i])
+                        input_tensor.append(input_tensor_i)
+                    self.input_tensor = input_tensor
+                    self.predictor.run()
+                    outputs = []
+                    for output_tensor in self.output_tensors:
+                        output = output_tensor.copy_to_cpu()
+                        outputs.append(output)
+                    if self.benchmark:
+                        self.autolog.times.stamp()
+                    preds = outputs
+            elif self.rec_algorithm == "LaTeXOCR":
+                inputs = [norm_img_batch]
+                if self.use_onnx:
+                    input_dict = {}
+                    input_dict[self.input_tensor.name] = norm_img_batch
+                    outputs = self.predictor.run(self.output_tensors, input_dict)
+                    preds = outputs
+                else:
+                    input_names = self.predictor.get_input_names()
+                    input_tensor = []
+                    for i in range(len(input_names)):
+                        input_tensor_i = self.predictor.get_input_handle(input_names[i])
+                        input_tensor_i.copy_from_cpu(inputs[i])
+                        input_tensor.append(input_tensor_i)
+                    self.input_tensor = input_tensor
+                    self.predictor.run()
+                    outputs = []
+                    for output_tensor in self.output_tensors:
+                        output = output_tensor.copy_to_cpu()
+                        outputs.append(output)
+                    if self.benchmark:
+                        self.autolog.times.stamp()
+                    preds = outputs
+            else:
+                if self.use_onnx:
+                    input_dict = {}
+                    input_dict[self.input_tensor.name] = norm_img_batch
+                    outputs = self.predictor.run(self.output_tensors, input_dict)
+                    preds = outputs[0]
+                else:
+                    self.input_tensor.copy_from_cpu(norm_img_batch)
+                    self.predictor.run()
+                    outputs = []
+                    for output_tensor in self.output_tensors:
+                        output = output_tensor.copy_to_cpu()
+                        outputs.append(output)
+                    if self.benchmark:
+                        self.autolog.times.stamp()
+                    if len(outputs) != 1:
+                        preds = outputs
+                    else:
+                        preds = outputs[0]
+            if self.postprocess_params["name"] == "CTCLabelDecode":
+                rec_result = self.postprocess_op(
+                    preds,
+                    return_word_box=self.return_word_box,
+                    wh_ratio_list=wh_ratio_list,
+                    max_wh_ratio=max_wh_ratio,
+                )
+            elif self.postprocess_params["name"] == "LaTeXOCRDecode":
+                preds = [p.reshape([-1]) for p in preds]
+                rec_result = self.postprocess_op(preds)
+            else:
+                rec_result = self.postprocess_op(preds)
+            for rno in range(len(rec_result)):
+                rec_res[indices[beg_img_no + rno]] = rec_result[rno]
+            if self.benchmark:
+                self.autolog.times.end(stamp=True)
+        return rec_res, time.time() - st
+
+
+def main(args):
+    image_file_list = get_image_file_list(args.image_dir)
+    valid_image_file_list = []
+    img_list = []
+
+    # logger
+    log_file = args.save_log_path
+    if os.path.isdir(args.save_log_path) or (
+        not os.path.exists(args.save_log_path) and args.save_log_path.endswith("/")
+    ):
+        log_file = os.path.join(log_file, "benchmark_recognition.log")
+    logger = get_logger(log_file=log_file)
+
+    # create text recognizer
+    text_recognizer = TextRecognizer(args)
+
+    logger.info(
+        "In PP-OCRv3, rec_image_shape parameter defaults to '3, 48, 320', "
+        "if you are using recognition model with PP-OCRv2 or an older version, please set --rec_image_shape='3,32,320"
+    )
+
+    # warmup 2 times
+    if args.warmup:
+        img = np.random.uniform(0, 255, [48, 320, 3]).astype(np.uint8)
+        for i in range(2):
+            res = text_recognizer([img] * int(args.rec_batch_num))
+
+    for image_file in image_file_list:
+        img, flag, _ = check_and_read(image_file)
+        if not flag:
+            img = cv2.imread(image_file)
+        if img is None:
+            logger.info("error in loading image:{}".format(image_file))
+            continue
+        valid_image_file_list.append(image_file)
+        img_list.append(img)
+    try:
+        rec_res, _ = text_recognizer(img_list)
+
+    except Exception as E:
+        logger.info(traceback.format_exc())
+        logger.info(E)
+        exit()
+    for ino in range(len(img_list)):
+        logger.info(
+            "Predicts of {}:{}".format(valid_image_file_list[ino], rec_res[ino])
+        )
+    if args.benchmark:
+        text_recognizer.autolog.report()
+
+
+if __name__ == "__main__":
+    main(utility.parse_args())
diff --git a/tools/infer/predict_sr.py b/tools/infer/predict_sr.py
new file mode 100755
index 0000000..389fc24
--- /dev/null
+++ b/tools/infer/predict_sr.py
@@ -0,0 +1,173 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import sys
+from PIL import Image
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.insert(0, __dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "../..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+
+import cv2
+import numpy as np
+import math
+import time
+import traceback
+import paddle
+
+import tools.infer.utility as utility
+from ppocr.postprocess import build_post_process
+from ppocr.utils.logging import get_logger
+from ppocr.utils.utility import get_image_file_list, check_and_read
+
+logger = get_logger()
+
+
+class TextSR(object):
+    def __init__(self, args):
+        if os.path.exists(f"{args.sr_model_dir}/inference.yml"):
+            model_config = utility.load_config(f"{args.sr_model_dir}/inference.yml")
+            model_name = model_config.get("Global", {}).get("model_name", "")
+            if model_name:
+                raise ValueError(
+                    f"{model_name} is not supported. Please check if the model is supported by the PaddleOCR wheel."
+                )
+
+        self.sr_image_shape = [int(v) for v in args.sr_image_shape.split(",")]
+        self.sr_batch_num = args.sr_batch_num
+
+        (
+            self.predictor,
+            self.input_tensor,
+            self.output_tensors,
+            self.config,
+        ) = utility.create_predictor(args, "sr", logger)
+        self.benchmark = args.benchmark
+        if args.benchmark:
+            import auto_log
+
+            pid = os.getpid()
+            gpu_id = utility.get_infer_gpuid()
+            self.autolog = auto_log.AutoLogger(
+                model_name="sr",
+                model_precision=args.precision,
+                batch_size=args.sr_batch_num,
+                data_shape="dynamic",
+                save_path=None,  # args.save_log_path,
+                inference_config=self.config,
+                pids=pid,
+                process_name=None,
+                gpu_ids=gpu_id if args.use_gpu else None,
+                time_keys=["preprocess_time", "inference_time", "postprocess_time"],
+                warmup=0,
+                logger=logger,
+            )
+
+    def resize_norm_img(self, img):
+        imgC, imgH, imgW = self.sr_image_shape
+        img = img.resize((imgW // 2, imgH // 2), Image.BICUBIC)
+        img_numpy = np.array(img).astype("float32")
+        img_numpy = img_numpy.transpose((2, 0, 1)) / 255
+        return img_numpy
+
+    def __call__(self, img_list):
+        img_num = len(img_list)
+        batch_num = self.sr_batch_num
+        st = time.time()
+        st = time.time()
+        all_result = [] * img_num
+        if self.benchmark:
+            self.autolog.times.start()
+        for beg_img_no in range(0, img_num, batch_num):
+            end_img_no = min(img_num, beg_img_no + batch_num)
+            norm_img_batch = []
+            imgC, imgH, imgW = self.sr_image_shape
+            for ino in range(beg_img_no, end_img_no):
+                norm_img = self.resize_norm_img(img_list[ino])
+                norm_img = norm_img[np.newaxis, :]
+                norm_img_batch.append(norm_img)
+
+            norm_img_batch = np.concatenate(norm_img_batch)
+            norm_img_batch = norm_img_batch.copy()
+            if self.benchmark:
+                self.autolog.times.stamp()
+            self.input_tensor.copy_from_cpu(norm_img_batch)
+            self.predictor.run()
+            outputs = []
+            for output_tensor in self.output_tensors:
+                output = output_tensor.copy_to_cpu()
+                outputs.append(output)
+            if len(outputs) != 1:
+                preds = outputs
+            else:
+                preds = outputs[0]
+            all_result.append(outputs)
+        if self.benchmark:
+            self.autolog.times.end(stamp=True)
+        return all_result, time.time() - st
+
+
+def main(args):
+    image_file_list = get_image_file_list(args.image_dir)
+    text_recognizer = TextSR(args)
+    valid_image_file_list = []
+    img_list = []
+
+    # warmup 2 times
+    if args.warmup:
+        img = np.random.uniform(0, 255, [16, 64, 3]).astype(np.uint8)
+        for i in range(2):
+            res = text_recognizer([img] * int(args.sr_batch_num))
+
+    for image_file in image_file_list:
+        img, flag, _ = check_and_read(image_file)
+        if not flag:
+            img = Image.open(image_file).convert("RGB")
+        if img is None:
+            logger.info("error in loading image:{}".format(image_file))
+            continue
+        valid_image_file_list.append(image_file)
+        img_list.append(img)
+    try:
+        preds, _ = text_recognizer(img_list)
+        for beg_no in range(len(preds)):
+            sr_img = preds[beg_no][1]
+            lr_img = preds[beg_no][0]
+            for i in range(sr_img.shape[0]):
+                fm_sr = (sr_img[i] * 255).transpose(1, 2, 0).astype(np.uint8)
+                fm_lr = (lr_img[i] * 255).transpose(1, 2, 0).astype(np.uint8)
+                img_name_pure = os.path.split(
+                    valid_image_file_list[beg_no * args.sr_batch_num + i]
+                )[-1]
+                cv2.imwrite(
+                    "infer_result/sr_{}".format(img_name_pure), fm_sr[:, :, ::-1]
+                )
+                logger.info(
+                    "The visualized image saved in infer_result/sr_{}".format(
+                        img_name_pure
+                    )
+                )
+
+    except Exception as E:
+        logger.info(traceback.format_exc())
+        logger.info(E)
+        exit()
+    if args.benchmark:
+        text_recognizer.autolog.report()
+
+
+if __name__ == "__main__":
+    main(utility.parse_args())
diff --git a/tools/infer/predict_system.py b/tools/infer/predict_system.py
new file mode 100755
index 0000000..bcb0758
--- /dev/null
+++ b/tools/infer/predict_system.py
@@ -0,0 +1,326 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import sys
+import subprocess
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "../..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+
+import cv2
+import copy
+import numpy as np
+import json
+import time
+import logging
+from PIL import Image
+import tools.infer.utility as utility
+import tools.infer.predict_rec as predict_rec
+import tools.infer.predict_det as predict_det
+import tools.infer.predict_cls as predict_cls
+from ppocr.utils.utility import get_image_file_list, check_and_read
+from ppocr.utils.logging import get_logger
+from tools.infer.utility import (
+    draw_ocr_box_txt,
+    get_rotate_crop_image,
+    get_minarea_rect_crop,
+    slice_generator,
+    merge_fragmented,
+)
+
+logger = get_logger()
+
+
+class TextSystem(object):
+    def __init__(self, args):
+        if not args.show_log:
+            logger.setLevel(logging.INFO)
+
+        self.text_detector = predict_det.TextDetector(args)
+        self.text_recognizer = predict_rec.TextRecognizer(args)
+        self.use_angle_cls = args.use_angle_cls
+        self.drop_score = args.drop_score
+        if self.use_angle_cls:
+            self.text_classifier = predict_cls.TextClassifier(args)
+
+        self.args = args
+        self.crop_image_res_index = 0
+
+    def draw_crop_rec_res(self, output_dir, img_crop_list, rec_res):
+        os.makedirs(output_dir, exist_ok=True)
+        bbox_num = len(img_crop_list)
+        for bno in range(bbox_num):
+            cv2.imwrite(
+                os.path.join(
+                    output_dir, f"mg_crop_{bno+self.crop_image_res_index}.jpg"
+                ),
+                img_crop_list[bno],
+            )
+            logger.debug(f"{bno}, {rec_res[bno]}")
+        self.crop_image_res_index += bbox_num
+
+    def __call__(self, img, cls=True, slice={}):
+        time_dict = {"det": 0, "rec": 0, "cls": 0, "all": 0}
+
+        if img is None:
+            logger.debug("no valid image provided")
+            return None, None, time_dict
+
+        start = time.time()
+        ori_im = img.copy()
+        if slice:
+            slice_gen = slice_generator(
+                img,
+                horizontal_stride=slice["horizontal_stride"],
+                vertical_stride=slice["vertical_stride"],
+            )
+            elapsed = []
+            dt_slice_boxes = []
+            for slice_crop, v_start, h_start in slice_gen:
+                dt_boxes, elapse = self.text_detector(slice_crop, use_slice=True)
+                if dt_boxes.size:
+                    dt_boxes[:, :, 0] += h_start
+                    dt_boxes[:, :, 1] += v_start
+                    dt_slice_boxes.append(dt_boxes)
+                    elapsed.append(elapse)
+            dt_boxes = np.concatenate(dt_slice_boxes)
+
+            dt_boxes = merge_fragmented(
+                boxes=dt_boxes,
+                x_threshold=slice["merge_x_thres"],
+                y_threshold=slice["merge_y_thres"],
+            )
+            elapse = sum(elapsed)
+        else:
+            dt_boxes, elapse = self.text_detector(img)
+
+        time_dict["det"] = elapse
+
+        if dt_boxes is None:
+            logger.debug("no dt_boxes found, elapsed : {}".format(elapse))
+            end = time.time()
+            time_dict["all"] = end - start
+            return None, None, time_dict
+        else:
+            logger.debug(
+                "dt_boxes num : {}, elapsed : {}".format(len(dt_boxes), elapse)
+            )
+        img_crop_list = []
+
+        dt_boxes = sorted_boxes(dt_boxes)
+
+        for bno in range(len(dt_boxes)):
+            tmp_box = copy.deepcopy(dt_boxes[bno])
+            if self.args.det_box_type == "quad":
+                img_crop = get_rotate_crop_image(ori_im, tmp_box)
+            else:
+                img_crop = get_minarea_rect_crop(ori_im, tmp_box)
+            img_crop_list.append(img_crop)
+        if self.use_angle_cls and cls:
+            img_crop_list, angle_list, elapse = self.text_classifier(img_crop_list)
+            time_dict["cls"] = elapse
+            logger.debug(
+                "cls num  : {}, elapsed : {}".format(len(img_crop_list), elapse)
+            )
+        if len(img_crop_list) > 1000:
+            logger.debug(
+                f"rec crops num: {len(img_crop_list)}, time and memory cost may be large."
+            )
+
+        rec_res, elapse = self.text_recognizer(img_crop_list)
+        time_dict["rec"] = elapse
+        logger.debug("rec_res num  : {}, elapsed : {}".format(len(rec_res), elapse))
+        if self.args.save_crop_res:
+            self.draw_crop_rec_res(self.args.crop_res_save_dir, img_crop_list, rec_res)
+        filter_boxes, filter_rec_res = [], []
+        for box, rec_result in zip(dt_boxes, rec_res):
+            text, score = rec_result[0], rec_result[1]
+            if score >= self.drop_score:
+                filter_boxes.append(box)
+                filter_rec_res.append(rec_result)
+        end = time.time()
+        time_dict["all"] = end - start
+        return filter_boxes, filter_rec_res, time_dict
+
+
+def sorted_boxes(dt_boxes):
+    """
+    Sort text boxes in order from top to bottom, left to right
+    args:
+        dt_boxes(array):detected text boxes with shape [4, 2]
+    return:
+        sorted boxes(array) with shape [4, 2]
+    """
+    num_boxes = dt_boxes.shape[0]
+    sorted_boxes = sorted(dt_boxes, key=lambda x: (x[0][1], x[0][0]))
+    _boxes = list(sorted_boxes)
+
+    for i in range(num_boxes - 1):
+        for j in range(i, -1, -1):
+            if abs(_boxes[j + 1][0][1] - _boxes[j][0][1]) < 10 and (
+                _boxes[j + 1][0][0] < _boxes[j][0][0]
+            ):
+                tmp = _boxes[j]
+                _boxes[j] = _boxes[j + 1]
+                _boxes[j + 1] = tmp
+            else:
+                break
+    return _boxes
+
+
+def main(args):
+    image_file_list = get_image_file_list(args.image_dir)
+    image_file_list = image_file_list[args.process_id :: args.total_process_num]
+    text_sys = TextSystem(args)
+    is_visualize = True
+    font_path = args.vis_font_path
+    drop_score = args.drop_score
+    draw_img_save_dir = args.draw_img_save_dir
+    os.makedirs(draw_img_save_dir, exist_ok=True)
+    save_results = []
+
+    logger.info(
+        "In PP-OCRv3, rec_image_shape parameter defaults to '3, 48, 320', "
+        "if you are using recognition model with PP-OCRv2 or an older version, please set --rec_image_shape='3,32,320"
+    )
+
+    # warm up 10 times
+    if args.warmup:
+        img = np.random.uniform(0, 255, [640, 640, 3]).astype(np.uint8)
+        for i in range(10):
+            res = text_sys(img)
+
+    total_time = 0
+    cpu_mem, gpu_mem, gpu_util = 0, 0, 0
+    _st = time.time()
+    count = 0
+    for idx, image_file in enumerate(image_file_list):
+        img, flag_gif, flag_pdf = check_and_read(image_file)
+        if not flag_gif and not flag_pdf:
+            img = cv2.imread(image_file)
+        if not flag_pdf:
+            if img is None:
+                logger.debug("error in loading image:{}".format(image_file))
+                continue
+            imgs = [img]
+        else:
+            page_num = args.page_num
+            if page_num > len(img) or page_num == 0:
+                page_num = len(img)
+            imgs = img[:page_num]
+        for index, img in enumerate(imgs):
+            starttime = time.time()
+            dt_boxes, rec_res, time_dict = text_sys(img)
+            elapse = time.time() - starttime
+            total_time += elapse
+            if len(imgs) > 1:
+                logger.debug(
+                    str(idx)
+                    + "_"
+                    + str(index)
+                    + "  Predict time of %s: %.3fs" % (image_file, elapse)
+                )
+            else:
+                logger.debug(
+                    str(idx) + "  Predict time of %s: %.3fs" % (image_file, elapse)
+                )
+            for text, score in rec_res:
+                logger.debug("{}, {:.3f}".format(text, score))
+
+            res = [
+                {
+                    "transcription": rec_res[i][0],
+                    "points": np.array(dt_boxes[i]).astype(np.int32).tolist(),
+                }
+                for i in range(len(dt_boxes))
+            ]
+            if len(imgs) > 1:
+                save_pred = (
+                    os.path.basename(image_file)
+                    + "_"
+                    + str(index)
+                    + "\t"
+                    + json.dumps(res, ensure_ascii=False)
+                    + "\n"
+                )
+            else:
+                save_pred = (
+                    os.path.basename(image_file)
+                    + "\t"
+                    + json.dumps(res, ensure_ascii=False)
+                    + "\n"
+                )
+            save_results.append(save_pred)
+
+            if is_visualize:
+                image = Image.fromarray(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
+                boxes = dt_boxes
+                txts = [rec_res[i][0] for i in range(len(rec_res))]
+                scores = [rec_res[i][1] for i in range(len(rec_res))]
+
+                draw_img = draw_ocr_box_txt(
+                    image,
+                    boxes,
+                    txts,
+                    scores,
+                    drop_score=drop_score,
+                    font_path=font_path,
+                )
+                if flag_gif:
+                    save_file = image_file[:-3] + "png"
+                elif flag_pdf:
+                    save_file = image_file.replace(".pdf", "_" + str(index) + ".png")
+                else:
+                    save_file = image_file
+                cv2.imwrite(
+                    os.path.join(draw_img_save_dir, os.path.basename(save_file)),
+                    draw_img[:, :, ::-1],
+                )
+                logger.debug(
+                    "The visualized image saved in {}".format(
+                        os.path.join(draw_img_save_dir, os.path.basename(save_file))
+                    )
+                )
+
+    logger.info("The predict total time is {}".format(time.time() - _st))
+    if args.benchmark:
+        text_sys.text_detector.autolog.report()
+        text_sys.text_recognizer.autolog.report()
+
+    with open(
+        os.path.join(draw_img_save_dir, "system_results.txt"), "w", encoding="utf-8"
+    ) as f:
+        f.writelines(save_results)
+
+
+if __name__ == "__main__":
+    args = utility.parse_args()
+    if args.use_mp:
+        p_list = []
+        total_process_num = args.total_process_num
+        for process_id in range(total_process_num):
+            cmd = (
+                [sys.executable, "-u"]
+                + sys.argv
+                + ["--process_id={}".format(process_id), "--use_mp={}".format(False)]
+            )
+            p = subprocess.Popen(cmd, stdout=sys.stdout, stderr=sys.stdout)
+            p_list.append(p)
+        for p in p_list:
+            p.wait()
+    else:
+        main(args)
diff --git a/tools/infer/utility.py b/tools/infer/utility.py
new file mode 100644
index 0000000..fb48334
--- /dev/null
+++ b/tools/infer/utility.py
@@ -0,0 +1,1042 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import argparse
+import os
+import sys
+import cv2
+import numpy as np
+import paddle
+import PIL
+from PIL import Image, ImageDraw, ImageFont
+import math
+from paddle import inference
+import random
+import yaml
+from ppocr.utils.logging import get_logger
+
+
+def str2bool(v):
+    return v.lower() in ("true", "yes", "t", "y", "1")
+
+
+def str2int_tuple(v):
+    return tuple([int(i.strip()) for i in v.split(",")])
+
+
+def init_args():
+    parser = argparse.ArgumentParser()
+    # params for prediction engine
+    parser.add_argument("--use_gpu", type=str2bool, default=True)
+    parser.add_argument("--use_xpu", type=str2bool, default=False)
+    parser.add_argument("--use_npu", type=str2bool, default=False)
+    parser.add_argument("--use_mlu", type=str2bool, default=False)
+    parser.add_argument("--use_metax_gpu", type=str2bool, default=False)
+    parser.add_argument(
+        "--use_gcu",
+        type=str2bool,
+        default=False,
+        help="Use Enflame GCU(General Compute Unit)",
+    )
+    parser.add_argument("--ir_optim", type=str2bool, default=True)
+    parser.add_argument("--use_tensorrt", type=str2bool, default=False)
+    parser.add_argument("--min_subgraph_size", type=int, default=15)
+    parser.add_argument("--precision", type=str, default="fp32")
+    parser.add_argument("--gpu_mem", type=int, default=500)
+    parser.add_argument("--gpu_id", type=int, default=0)
+
+    # params for text detector
+    parser.add_argument("--image_dir", type=str)
+    parser.add_argument("--page_num", type=int, default=0)
+    parser.add_argument("--det_algorithm", type=str, default="DB")
+    parser.add_argument("--det_model_dir", type=str)
+    parser.add_argument("--det_limit_side_len", type=float, default=960)
+    parser.add_argument("--det_limit_type", type=str, default="max")
+    parser.add_argument("--det_box_type", type=str, default="quad")
+
+    # DB params
+    parser.add_argument("--det_db_thresh", type=float, default=0.3)
+    parser.add_argument("--det_db_box_thresh", type=float, default=0.6)
+    parser.add_argument("--det_db_unclip_ratio", type=float, default=1.5)
+    parser.add_argument("--max_batch_size", type=int, default=10)
+    parser.add_argument("--use_dilation", type=str2bool, default=False)
+    parser.add_argument("--det_db_score_mode", type=str, default="fast")
+
+    # EAST params
+    parser.add_argument("--det_east_score_thresh", type=float, default=0.8)
+    parser.add_argument("--det_east_cover_thresh", type=float, default=0.1)
+    parser.add_argument("--det_east_nms_thresh", type=float, default=0.2)
+
+    # SAST params
+    parser.add_argument("--det_sast_score_thresh", type=float, default=0.5)
+    parser.add_argument("--det_sast_nms_thresh", type=float, default=0.2)
+
+    # PSE params
+    parser.add_argument("--det_pse_thresh", type=float, default=0)
+    parser.add_argument("--det_pse_box_thresh", type=float, default=0.85)
+    parser.add_argument("--det_pse_min_area", type=float, default=16)
+    parser.add_argument("--det_pse_scale", type=int, default=1)
+
+    # FCE params
+    parser.add_argument("--scales", type=list, default=[8, 16, 32])
+    parser.add_argument("--alpha", type=float, default=1.0)
+    parser.add_argument("--beta", type=float, default=1.0)
+    parser.add_argument("--fourier_degree", type=int, default=5)
+
+    # params for text recognizer
+    parser.add_argument("--rec_algorithm", type=str, default="SVTR_LCNet")
+    parser.add_argument("--rec_model_dir", type=str)
+    parser.add_argument("--rec_image_inverse", type=str2bool, default=True)
+    parser.add_argument("--rec_image_shape", type=str, default="3, 48, 320")
+    parser.add_argument("--rec_batch_num", type=int, default=6)
+    parser.add_argument("--max_text_length", type=int, default=25)
+    parser.add_argument(
+        "--rec_char_dict_path", type=str, default="./ppocr/utils/ppocr_keys_v1.txt"
+    )
+    parser.add_argument("--use_space_char", type=str2bool, default=True)
+    parser.add_argument("--vis_font_path", type=str, default="./doc/fonts/simfang.ttf")
+    parser.add_argument("--drop_score", type=float, default=0.5)
+
+    # params for e2e
+    parser.add_argument("--e2e_algorithm", type=str, default="PGNet")
+    parser.add_argument("--e2e_model_dir", type=str)
+    parser.add_argument("--e2e_limit_side_len", type=float, default=768)
+    parser.add_argument("--e2e_limit_type", type=str, default="max")
+
+    # PGNet params
+    parser.add_argument("--e2e_pgnet_score_thresh", type=float, default=0.5)
+    parser.add_argument(
+        "--e2e_char_dict_path", type=str, default="./ppocr/utils/ic15_dict.txt"
+    )
+    parser.add_argument("--e2e_pgnet_valid_set", type=str, default="totaltext")
+    parser.add_argument("--e2e_pgnet_mode", type=str, default="fast")
+
+    # params for text classifier
+    parser.add_argument("--use_angle_cls", type=str2bool, default=False)
+    parser.add_argument("--cls_model_dir", type=str)
+    parser.add_argument("--cls_image_shape", type=str, default="3, 48, 192")
+    parser.add_argument("--label_list", type=list, default=["0", "180"])
+    parser.add_argument("--cls_batch_num", type=int, default=6)
+    parser.add_argument("--cls_thresh", type=float, default=0.9)
+
+    parser.add_argument("--enable_mkldnn", type=str2bool, default=None)
+    parser.add_argument("--cpu_threads", type=int, default=10)
+    parser.add_argument("--use_pdserving", type=str2bool, default=False)
+    parser.add_argument("--warmup", type=str2bool, default=False)
+
+    # SR params
+    parser.add_argument("--sr_model_dir", type=str)
+    parser.add_argument("--sr_image_shape", type=str, default="3, 32, 128")
+    parser.add_argument("--sr_batch_num", type=int, default=1)
+
+    #
+    parser.add_argument("--draw_img_save_dir", type=str, default="./inference_results")
+    parser.add_argument("--save_crop_res", type=str2bool, default=False)
+    parser.add_argument("--crop_res_save_dir", type=str, default="./output")
+
+    # multi-process
+    parser.add_argument("--use_mp", type=str2bool, default=False)
+    parser.add_argument("--total_process_num", type=int, default=1)
+    parser.add_argument("--process_id", type=int, default=0)
+
+    parser.add_argument("--benchmark", type=str2bool, default=False)
+    parser.add_argument("--save_log_path", type=str, default="./log_output/")
+
+    parser.add_argument("--show_log", type=str2bool, default=True)
+    parser.add_argument("--use_onnx", type=str2bool, default=False)
+    parser.add_argument("--onnx_providers", nargs="+", type=str, default=False)
+    parser.add_argument("--onnx_sess_options", type=list, default=False)
+
+    # extended function
+    parser.add_argument(
+        "--return_word_box",
+        type=str2bool,
+        default=False,
+        help="Whether return the bbox of each word (split by space) or chinese character. Only used in ppstructure for layout recovery",
+    )
+
+    return parser
+
+
+def parse_args():
+    parser = init_args()
+    return parser.parse_args()
+
+
+def create_predictor(args, mode, logger):
+    if mode == "det":
+        model_dir = args.det_model_dir
+    elif mode == "cls":
+        model_dir = args.cls_model_dir
+    elif mode == "rec":
+        model_dir = args.rec_model_dir
+    elif mode == "table":
+        model_dir = args.table_model_dir
+    elif mode == "ser":
+        model_dir = args.ser_model_dir
+    elif mode == "re":
+        model_dir = args.re_model_dir
+    elif mode == "sr":
+        model_dir = args.sr_model_dir
+    elif mode == "layout":
+        model_dir = args.layout_model_dir
+    else:
+        model_dir = args.e2e_model_dir
+
+    if model_dir is None:
+        logger.info("not find {} model file path {}".format(mode, model_dir))
+        sys.exit(0)
+    if args.use_onnx:
+        import onnxruntime as ort
+
+        model_file_path = model_dir
+        if not os.path.exists(model_file_path):
+            raise ValueError("not find model file path {}".format(model_file_path))
+
+        sess_options = args.onnx_sess_options or None
+
+        if args.onnx_providers and len(args.onnx_providers) > 0:
+            sess = ort.InferenceSession(
+                model_file_path,
+                providers=args.onnx_providers,
+                sess_options=sess_options,
+            )
+        elif args.use_gpu:
+            sess = ort.InferenceSession(
+                model_file_path,
+                providers=[
+                    (
+                        "CUDAExecutionProvider",
+                        {
+                            "device_id": args.gpu_id,
+                            "cudnn_conv_algo_search": "HEURISTIC",
+                        },
+                    )
+                ],
+                sess_options=sess_options,
+            )
+        else:
+            sess = ort.InferenceSession(
+                model_file_path,
+                providers=["CPUExecutionProvider"],
+                sess_options=sess_options,
+            )
+        inputs = sess.get_inputs()
+        return (
+            sess,
+            inputs[0] if len(inputs) == 1 else [vo.name for vo in inputs],
+            None,
+            None,
+        )
+
+    else:
+        file_names = ["model", "inference"]
+        for file_name in file_names:
+            params_file_path = f"{model_dir}/{file_name}.pdiparams"
+            if os.path.exists(params_file_path):
+                break
+
+        if not os.path.exists(params_file_path):
+            raise ValueError(f"not find {file_name}.pdiparams in {model_dir}")
+
+        if not (
+            os.path.exists(f"{model_dir}/{file_name}.pdmodel")
+            or os.path.exists(f"{model_dir}/{file_name}.json")
+        ):
+            raise ValueError(
+                f"neither {file_name}.json nor {file_name}.pdmodel was found in {model_dir}."
+            )
+
+        if os.path.exists(f"{model_dir}/{file_name}.json"):
+            model_file_path = f"{model_dir}/{file_name}.json"
+        else:
+            model_file_path = f"{model_dir}/{file_name}.pdmodel"
+
+        config = inference.Config(model_file_path, params_file_path)
+
+        if hasattr(args, "precision"):
+            if args.precision == "fp16" and args.use_tensorrt:
+                precision = inference.PrecisionType.Half
+            elif args.precision == "int8":
+                precision = inference.PrecisionType.Int8
+            else:
+                precision = inference.PrecisionType.Float32
+        else:
+            precision = inference.PrecisionType.Float32
+
+        if args.use_gpu:
+            gpu_id = get_infer_gpuid()
+            if gpu_id is None:
+                logger.warning(
+                    "GPU is not found in current device by nvidia-smi. Please check your device or ignore it if run on jetson."
+                )
+            config.enable_use_gpu(args.gpu_mem, args.gpu_id)
+            if args.use_tensorrt:
+                if ".json" in model_file_path:
+                    trt_dynamic_shapes = {}
+                    trt_dynamic_shape_input_data = {}
+                    if os.path.exists(f"{model_dir}/inference.yml"):
+                        model_config = load_config(f"{model_dir}/inference.yml")
+                        trt_dynamic_shapes = (
+                            model_config.get("Hpi", {})
+                            .get("backend_configs", {})
+                            .get("paddle_infer", {})
+                            .get("trt_dynamic_shapes", {})
+                        )
+                        trt_dynamic_shape_input_data = (
+                            model_config.get("Hpi", {})
+                            .get("backend_configs", {})
+                            .get("paddle_infer", {})
+                            .get("trt_dynamic_shapes_input_data", {})
+                        )
+
+                    if not trt_dynamic_shapes:
+                        raise RuntimeError(
+                            "Configuration Error: 'trt_dynamic_shapes' must be defined in 'inference.yml' for Paddle Inference TensorRT."
+                        )
+
+                    trt_save_path = f"{model_dir}/.cache/trt/{file_name}"
+                    trt_model_file_path = trt_save_path + ".json"
+                    trt_params_file_path = trt_save_path + ".pdiparams"
+                    if not os.path.exists(trt_model_file_path) or not os.path.exists(
+                        trt_params_file_path
+                    ):
+                        _convert_trt(
+                            {},
+                            model_file_path,
+                            params_file_path,
+                            trt_save_path,
+                            args.gpu_id,
+                            trt_dynamic_shapes,
+                            trt_dynamic_shape_input_data,
+                        )
+                    config = inference.Config(model_file_path, params_file_path)
+                    config.exp_disable_mixed_precision_ops({"feed", "fetch"})
+                    config.enable_use_gpu(args.gpu_mem, args.gpu_id)
+                else:
+                    config.enable_tensorrt_engine(
+                        workspace_size=1 << 30,
+                        precision_mode=precision,
+                        max_batch_size=args.max_batch_size,
+                        min_subgraph_size=args.min_subgraph_size,  # skip the minimum trt subgraph
+                        use_calib_mode=False,
+                    )
+
+                    # collect shape
+                    trt_shape_f = os.path.join(
+                        model_dir, f"{mode}_trt_dynamic_shape.txt"
+                    )
+
+                    if not os.path.exists(trt_shape_f):
+                        config.collect_shape_range_info(trt_shape_f)
+                        logger.info(f"collect dynamic shape info into : {trt_shape_f}")
+                    try:
+                        config.enable_tuned_tensorrt_dynamic_shape(trt_shape_f, True)
+                    except Exception as E:
+                        logger.info(E)
+                        logger.info("Please keep your paddlepaddle-gpu >= 2.3.0!")
+
+        elif args.use_npu:
+            config.enable_custom_device("npu")
+        elif args.use_mlu:
+            config.enable_custom_device("mlu")
+        elif args.use_metax_gpu:
+            if args.precision == "fp16":
+                config.enable_custom_device(
+                    "metax_gpu", 0, paddle.inference.PrecisionType.Half
+                )
+
+            else:
+                config.enable_custom_device("metax_gpu")
+        elif args.use_xpu:
+            config.enable_xpu(10 * 1024 * 1024)
+        elif args.use_gcu:  # for Enflame GCU(General Compute Unit)
+            assert paddle.device.is_compiled_with_custom_device("gcu"), (
+                "Args use_gcu cannot be set as True while your paddle "
+                "is not compiled with gcu! \nPlease try: \n"
+                "\t1. Install paddle-custom-gcu to run model on GCU. \n"
+                "\t2. Set use_gcu as False in args to run model on CPU."
+            )
+            import paddle_custom_device.gcu.passes as gcu_passes
+
+            gcu_passes.setUp()
+            if args.precision == "fp16":
+                config.enable_custom_device(
+                    "gcu", 0, paddle.inference.PrecisionType.Half
+                )
+                gcu_passes.set_exp_enable_mixed_precision_ops(config)
+            else:
+                config.enable_custom_device("gcu")
+
+            if paddle.framework.use_pir_api():
+                config.enable_new_ir(True)
+                config.enable_new_executor(True)
+            else:
+                pass_builder = config.pass_builder()
+                gcu_passes.append_passes_for_legacy_ir(pass_builder, "PaddleOCR")
+        else:
+            config.disable_gpu()
+            if args.enable_mkldnn is not None:
+                if args.enable_mkldnn:
+                    # cache 10 different shapes for mkldnn to avoid memory leak
+                    config.set_mkldnn_cache_capacity(10)
+                    config.enable_mkldnn()
+                    if args.precision == "fp16":
+                        config.enable_mkldnn_bfloat16()
+                else:
+                    if hasattr(config, "disable_mkldnn"):
+                        config.disable_mkldnn()
+
+            if hasattr(args, "cpu_threads"):
+                config.set_cpu_math_library_num_threads(args.cpu_threads)
+            else:
+                # default cpu threads as 10
+                config.set_cpu_math_library_num_threads(10)
+
+            if hasattr(config, "enable_new_ir"):
+                config.enable_new_ir()
+            if hasattr(config, "enable_new_executor"):
+                config.enable_new_executor()
+
+        # enable memory optim
+        config.enable_memory_optim()
+        config.disable_glog_info()
+        if not args.use_gcu:  # for Enflame GCU(General Compute Unit)
+            config.delete_pass("conv_transpose_eltwiseadd_bn_fuse_pass")
+        config.delete_pass("matmul_transpose_reshape_fuse_pass")
+        if mode == "rec" and args.rec_algorithm == "SRN":
+            config.delete_pass("gpu_cpu_map_matmul_v2_to_matmul_pass")
+        if mode == "re":
+            config.delete_pass("simplify_with_basic_ops_pass")
+        if mode == "table":
+            config.delete_pass("fc_fuse_pass")  # not supported for table
+        config.switch_use_feed_fetch_ops(False)
+        config.switch_ir_optim(True)
+
+        # create predictor
+        predictor = inference.create_predictor(config)
+        input_names = predictor.get_input_names()
+        if mode in ["ser", "re"]:
+            input_tensor = []
+            for name in input_names:
+                input_tensor.append(predictor.get_input_handle(name))
+        else:
+            for name in input_names:
+                input_tensor = predictor.get_input_handle(name)
+        output_tensors = get_output_tensors(args, mode, predictor)
+        return predictor, input_tensor, output_tensors, config
+
+
+def _convert_trt(
+    trt_cfg_setting,
+    pp_model_file,
+    pp_params_file,
+    trt_save_path,
+    device_id,
+    dynamic_shapes,
+    dynamic_shape_input_data,
+):
+    from paddle.tensorrt.export import Input, TensorRTConfig, convert
+
+    def _set_trt_config():
+        for attr_name in trt_cfg_setting:
+            assert hasattr(
+                trt_config, attr_name
+            ), f"The `{type(trt_config)}` don't have the attribute `{attr_name}`!"
+            setattr(trt_config, attr_name, trt_cfg_setting[attr_name])
+
+    def _get_predictor(model_file, params_file):
+        # HACK
+        config = inference.Config(str(model_file), str(params_file))
+        config.enable_use_gpu(100, device_id)
+        # NOTE: Disable oneDNN to circumvent a bug in Paddle Inference
+        config.disable_mkldnn()
+        config.disable_glog_info()
+        return inference.create_predictor(config)
+
+    dynamic_shape_input_data = dynamic_shape_input_data or {}
+
+    predictor = _get_predictor(pp_model_file, pp_params_file)
+    input_names = predictor.get_input_names()
+    for name in dynamic_shapes:
+        if name not in input_names:
+            raise ValueError(
+                f"Invalid input name {repr(name)} found in `dynamic_shapes`"
+            )
+    for name in input_names:
+        if name not in dynamic_shapes:
+            raise ValueError(f"Input name {repr(name)} not found in `dynamic_shapes`")
+    for name in dynamic_shape_input_data:
+        if name not in input_names:
+            raise ValueError(
+                f"Invalid input name {repr(name)} found in `dynamic_shape_input_data`"
+            )
+
+    trt_inputs = []
+    for name, candidate_shapes in dynamic_shapes.items():
+        # XXX: Currently we have no way to get the data type of the tensor
+        # without creating an input handle.
+        handle = predictor.get_input_handle(name)
+        dtype = _pd_dtype_to_np_dtype(handle.type())
+        min_shape, opt_shape, max_shape = candidate_shapes
+        if name in dynamic_shape_input_data:
+            min_arr = np.array(dynamic_shape_input_data[name][0], dtype=dtype).reshape(
+                min_shape
+            )
+            opt_arr = np.array(dynamic_shape_input_data[name][1], dtype=dtype).reshape(
+                opt_shape
+            )
+            max_arr = np.array(dynamic_shape_input_data[name][2], dtype=dtype).reshape(
+                max_shape
+            )
+        else:
+            min_arr = np.ones(min_shape, dtype=dtype)
+            opt_arr = np.ones(opt_shape, dtype=dtype)
+            max_arr = np.ones(max_shape, dtype=dtype)
+
+        # refer to: https://github.com/PolaKuma/Paddle/blob/3347f225bc09f2ec09802a2090432dd5cb5b6739/test/tensorrt/test_converter_model_resnet50.py
+        trt_input = Input((min_arr, opt_arr, max_arr))
+        trt_inputs.append(trt_input)
+
+    # Create TensorRTConfig
+    trt_config = TensorRTConfig(inputs=trt_inputs)
+    _set_trt_config()
+    trt_config.save_model_dir = trt_save_path
+    pp_model_path = pp_model_file.split(".")[0]
+    convert(pp_model_path, trt_config)
+
+
+def _pd_dtype_to_np_dtype(pd_dtype):
+    if pd_dtype == inference.DataType.FLOAT64:
+        return np.float64
+    elif pd_dtype == inference.DataType.FLOAT32:
+        return np.float32
+    elif pd_dtype == inference.DataType.INT64:
+        return np.int64
+    elif pd_dtype == inference.DataType.INT32:
+        return np.int32
+    elif pd_dtype == inference.DataType.UINT8:
+        return np.uint8
+    elif pd_dtype == inference.DataType.INT8:
+        return np.int8
+    else:
+        raise TypeError(f"Unsupported data type: {pd_dtype}")
+
+
+def load_config(file_path):
+    _, ext = os.path.splitext(file_path)
+    if ext not in [".yml", ".yaml"]:
+        raise ValueError(f"only support yaml files for now, got {file_path}")
+    with open(file_path, "rb") as file:
+        config = yaml.load(file, Loader=yaml.SafeLoader)
+    return config
+
+
+def get_output_tensors(args, mode, predictor):
+    output_names = predictor.get_output_names()
+    output_tensors = []
+    if mode == "rec" and args.rec_algorithm in ["CRNN", "SVTR_LCNet", "SVTR_HGNet"]:
+        output_name = "softmax_0.tmp_0"
+        if output_name in output_names:
+            return [predictor.get_output_handle(output_name)]
+        else:
+            for output_name in output_names:
+                output_tensor = predictor.get_output_handle(output_name)
+                output_tensors.append(output_tensor)
+    else:
+        for output_name in output_names:
+            output_tensor = predictor.get_output_handle(output_name)
+            output_tensors.append(output_tensor)
+    return output_tensors
+
+
+def get_infer_gpuid():
+    """
+    Get the GPU ID to be used for inference.
+
+    Returns:
+        int: The GPU ID to be used for inference.
+    """
+    logger = get_logger()
+    if not paddle.device.is_compiled_with_rocm:
+        gpu_id_str = os.environ.get("CUDA_VISIBLE_DEVICES", "0")
+    else:
+        gpu_id_str = os.environ.get("HIP_VISIBLE_DEVICES", "0")
+
+    gpu_ids = gpu_id_str.split(",")
+    logger.warning(
+        "The first GPU is used for inference by default, GPU ID: {}".format(gpu_ids[0])
+    )
+    return int(gpu_ids[0])
+
+
+def draw_e2e_res(dt_boxes, strs, img_path):
+    src_im = cv2.imread(img_path)
+    for box, str in zip(dt_boxes, strs):
+        box = box.astype(np.int32).reshape((-1, 1, 2))
+        cv2.polylines(src_im, [box], True, color=(255, 255, 0), thickness=2)
+        cv2.putText(
+            src_im,
+            str,
+            org=(int(box[0, 0, 0]), int(box[0, 0, 1])),
+            fontFace=cv2.FONT_HERSHEY_COMPLEX,
+            fontScale=0.7,
+            color=(0, 255, 0),
+            thickness=1,
+        )
+    return src_im
+
+
+def draw_text_det_res(dt_boxes, img):
+    for box in dt_boxes:
+        box = np.array(box).astype(np.int32).reshape(-1, 2)
+        cv2.polylines(img, [box], True, color=(255, 255, 0), thickness=2)
+    return img
+
+
+def resize_img(img, input_size=600):
+    """
+    resize img and limit the longest side of the image to input_size
+    """
+    img = np.array(img)
+    im_shape = img.shape
+    im_size_max = np.max(im_shape[0:2])
+    im_scale = float(input_size) / float(im_size_max)
+    img = cv2.resize(img, None, None, fx=im_scale, fy=im_scale)
+    return img
+
+
+def draw_ocr(
+    image,
+    boxes,
+    txts=None,
+    scores=None,
+    drop_score=0.5,
+    font_path="./doc/fonts/simfang.ttf",
+):
+    """
+    Visualize the results of OCR detection and recognition
+    args:
+        image(Image|array): RGB image
+        boxes(list): boxes with shape(N, 4, 2)
+        txts(list): the texts
+        scores(list): txxs corresponding scores
+        drop_score(float): only scores greater than drop_threshold will be visualized
+        font_path: the path of font which is used to draw text
+    return(array):
+        the visualized img
+    """
+    if scores is None:
+        scores = [1] * len(boxes)
+    box_num = len(boxes)
+    for i in range(box_num):
+        if scores is not None and (scores[i] < drop_score or math.isnan(scores[i])):
+            continue
+        box = np.reshape(np.array(boxes[i]), [-1, 1, 2]).astype(np.int64)
+        image = cv2.polylines(np.array(image), [box], True, (255, 0, 0), 2)
+    if txts is not None:
+        img = np.array(resize_img(image, input_size=600))
+        txt_img = text_visual(
+            txts,
+            scores,
+            img_h=img.shape[0],
+            img_w=600,
+            threshold=drop_score,
+            font_path=font_path,
+        )
+        img = np.concatenate([np.array(img), np.array(txt_img)], axis=1)
+        return img
+    return image
+
+
+def draw_ocr_box_txt(
+    image,
+    boxes,
+    txts=None,
+    scores=None,
+    drop_score=0.5,
+    font_path="./doc/fonts/simfang.ttf",
+):
+    h, w = image.height, image.width
+    img_left = image.copy()
+    img_right = np.ones((h, w, 3), dtype=np.uint8) * 255
+    random.seed(0)
+
+    draw_left = ImageDraw.Draw(img_left)
+    if txts is None or len(txts) != len(boxes):
+        txts = [None] * len(boxes)
+    for idx, (box, txt) in enumerate(zip(boxes, txts)):
+        if scores is not None and scores[idx] < drop_score:
+            continue
+        color = (random.randint(0, 255), random.randint(0, 255), random.randint(0, 255))
+        draw_left.polygon(box, fill=color)
+        img_right_text = draw_box_txt_fine((w, h), box, txt, font_path)
+        pts = np.array(box, np.int32).reshape((-1, 1, 2))
+        cv2.polylines(img_right_text, [pts], True, color, 1)
+        img_right = cv2.bitwise_and(img_right, img_right_text)
+    img_left = Image.blend(image, img_left, 0.5)
+    img_show = Image.new("RGB", (w * 2, h), (255, 255, 255))
+    img_show.paste(img_left, (0, 0, w, h))
+    img_show.paste(Image.fromarray(img_right), (w, 0, w * 2, h))
+    return np.array(img_show)
+
+
+def draw_box_txt_fine(img_size, box, txt, font_path="./doc/fonts/simfang.ttf"):
+    box_height = int(
+        math.sqrt((box[0][0] - box[3][0]) ** 2 + (box[0][1] - box[3][1]) ** 2)
+    )
+    box_width = int(
+        math.sqrt((box[0][0] - box[1][0]) ** 2 + (box[0][1] - box[1][1]) ** 2)
+    )
+
+    if box_height > 2 * box_width and box_height > 30:
+        img_text = Image.new("RGB", (box_height, box_width), (255, 255, 255))
+        draw_text = ImageDraw.Draw(img_text)
+        if txt:
+            font = create_font(txt, (box_height, box_width), font_path)
+            draw_text.text([0, 0], txt, fill=(0, 0, 0), font=font)
+        img_text = img_text.transpose(Image.ROTATE_270)
+    else:
+        img_text = Image.new("RGB", (box_width, box_height), (255, 255, 255))
+        draw_text = ImageDraw.Draw(img_text)
+        if txt:
+            font = create_font(txt, (box_width, box_height), font_path)
+            draw_text.text([0, 0], txt, fill=(0, 0, 0), font=font)
+
+    pts1 = np.float32(
+        [[0, 0], [box_width, 0], [box_width, box_height], [0, box_height]]
+    )
+    pts2 = np.array(box, dtype=np.float32)
+    M = cv2.getPerspectiveTransform(pts1, pts2)
+
+    img_text = np.array(img_text, dtype=np.uint8)
+    img_right_text = cv2.warpPerspective(
+        img_text,
+        M,
+        img_size,
+        flags=cv2.INTER_NEAREST,
+        borderMode=cv2.BORDER_CONSTANT,
+        borderValue=(255, 255, 255),
+    )
+    return img_right_text
+
+
+def create_font(txt, sz, font_path="./doc/fonts/simfang.ttf"):
+    font_size = int(sz[1] * 0.99)
+    font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
+    if int(PIL.__version__.split(".")[0]) < 10:
+        length = font.getsize(txt)[0]
+    else:
+        length = font.getlength(txt)
+
+    if length > sz[0]:
+        font_size = int(font_size * sz[0] / length)
+        font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
+    return font
+
+
+def str_count(s):
+    """
+    Count the number of Chinese characters,
+    a single English character and a single number
+    equal to half the length of Chinese characters.
+    args:
+        s(string): the input of string
+    return(int):
+        the number of Chinese characters
+    """
+    import string
+
+    count_zh = count_pu = 0
+    s_len = len(s)
+    en_dg_count = 0
+    for c in s:
+        if c in string.ascii_letters or c.isdigit() or c.isspace():
+            en_dg_count += 1
+        elif c.isalpha():
+            count_zh += 1
+        else:
+            count_pu += 1
+    return s_len - math.ceil(en_dg_count / 2)
+
+
+def text_visual(
+    texts, scores, img_h=400, img_w=600, threshold=0.0, font_path="./doc/simfang.ttf"
+):
+    """
+    create new blank img and draw txt on it
+    args:
+        texts(list): the text will be draw
+        scores(list|None): corresponding score of each txt
+        img_h(int): the height of blank img
+        img_w(int): the width of blank img
+        font_path: the path of font which is used to draw text
+    return(array):
+    """
+    if scores is not None:
+        assert len(texts) == len(
+            scores
+        ), "The number of txts and corresponding scores must match"
+
+    def create_blank_img():
+        blank_img = np.ones(shape=[img_h, img_w], dtype=np.uint8) * 255
+        blank_img[:, img_w - 1 :] = 0
+        blank_img = Image.fromarray(blank_img).convert("RGB")
+        draw_txt = ImageDraw.Draw(blank_img)
+        return blank_img, draw_txt
+
+    blank_img, draw_txt = create_blank_img()
+
+    font_size = 20
+    txt_color = (0, 0, 0)
+    font = ImageFont.truetype(font_path, font_size, encoding="utf-8")
+
+    gap = font_size + 5
+    txt_img_list = []
+    count, index = 1, 0
+    for idx, txt in enumerate(texts):
+        index += 1
+        if scores[idx] < threshold or math.isnan(scores[idx]):
+            index -= 1
+            continue
+        first_line = True
+        while str_count(txt) >= img_w // font_size - 4:
+            tmp = txt
+            txt = tmp[: img_w // font_size - 4]
+            if first_line:
+                new_txt = str(index) + ": " + txt
+                first_line = False
+            else:
+                new_txt = "    " + txt
+            draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
+            txt = tmp[img_w // font_size - 4 :]
+            if count >= img_h // gap - 1:
+                txt_img_list.append(np.array(blank_img))
+                blank_img, draw_txt = create_blank_img()
+                count = 0
+            count += 1
+        if first_line:
+            new_txt = str(index) + ": " + txt + "   " + "%.3f" % (scores[idx])
+        else:
+            new_txt = "  " + txt + "  " + "%.3f" % (scores[idx])
+        draw_txt.text((0, gap * count), new_txt, txt_color, font=font)
+        # whether add new blank img or not
+        if count >= img_h // gap - 1 and idx + 1 < len(texts):
+            txt_img_list.append(np.array(blank_img))
+            blank_img, draw_txt = create_blank_img()
+            count = 0
+        count += 1
+    txt_img_list.append(np.array(blank_img))
+    if len(txt_img_list) == 1:
+        blank_img = np.array(txt_img_list[0])
+    else:
+        blank_img = np.concatenate(txt_img_list, axis=1)
+    return np.array(blank_img)
+
+
+def base64_to_cv2(b64str):
+    import base64
+
+    data = base64.b64decode(b64str.encode("utf8"))
+    data = np.frombuffer(data, np.uint8)
+    data = cv2.imdecode(data, cv2.IMREAD_COLOR)
+    return data
+
+
+def draw_boxes(image, boxes, scores=None, drop_score=0.5):
+    if scores is None:
+        scores = [1] * len(boxes)
+    for box, score in zip(boxes, scores):
+        if score < drop_score:
+            continue
+        box = np.reshape(np.array(box), [-1, 1, 2]).astype(np.int64)
+        image = cv2.polylines(np.array(image), [box], True, (255, 0, 0), 2)
+    return image
+
+
+def get_rotate_crop_image(img, points):
+    """
+    img_height, img_width = img.shape[0:2]
+    left = int(np.min(points[:, 0]))
+    right = int(np.max(points[:, 0]))
+    top = int(np.min(points[:, 1]))
+    bottom = int(np.max(points[:, 1]))
+    img_crop = img[top:bottom, left:right, :].copy()
+    points[:, 0] = points[:, 0] - left
+    points[:, 1] = points[:, 1] - top
+    """
+    assert len(points) == 4, "shape of points must be 4*2"
+    img_crop_width = int(
+        max(
+            np.linalg.norm(points[0] - points[1]), np.linalg.norm(points[2] - points[3])
+        )
+    )
+    img_crop_height = int(
+        max(
+            np.linalg.norm(points[0] - points[3]), np.linalg.norm(points[1] - points[2])
+        )
+    )
+    pts_std = np.float32(
+        [
+            [0, 0],
+            [img_crop_width, 0],
+            [img_crop_width, img_crop_height],
+            [0, img_crop_height],
+        ]
+    )
+    M = cv2.getPerspectiveTransform(points, pts_std)
+    dst_img = cv2.warpPerspective(
+        img,
+        M,
+        (img_crop_width, img_crop_height),
+        borderMode=cv2.BORDER_REPLICATE,
+        flags=cv2.INTER_CUBIC,
+    )
+    dst_img_height, dst_img_width = dst_img.shape[0:2]
+    if dst_img_height * 1.0 / dst_img_width >= 1.5:
+        dst_img = np.rot90(dst_img)
+    return dst_img
+
+
+def get_minarea_rect_crop(img, points):
+    bounding_box = cv2.minAreaRect(np.array(points).astype(np.int32))
+    points = sorted(list(cv2.boxPoints(bounding_box)), key=lambda x: x[0])
+
+    index_a, index_b, index_c, index_d = 0, 1, 2, 3
+    if points[1][1] > points[0][1]:
+        index_a = 0
+        index_d = 1
+    else:
+        index_a = 1
+        index_d = 0
+    if points[3][1] > points[2][1]:
+        index_b = 2
+        index_c = 3
+    else:
+        index_b = 3
+        index_c = 2
+
+    box = [points[index_a], points[index_b], points[index_c], points[index_d]]
+    crop_img = get_rotate_crop_image(img, np.array(box))
+    return crop_img
+
+
+def slice_generator(image, horizontal_stride, vertical_stride, maximum_slices=500):
+    if not isinstance(image, np.ndarray):
+        image = np.array(image)
+
+    image_h, image_w = image.shape[:2]
+    vertical_num_slices = (image_h + vertical_stride - 1) // vertical_stride
+    horizontal_num_slices = (image_w + horizontal_stride - 1) // horizontal_stride
+
+    assert (
+        vertical_num_slices > 0
+    ), f"Invalid number ({vertical_num_slices}) of vertical slices"
+
+    assert (
+        horizontal_num_slices > 0
+    ), f"Invalid number ({horizontal_num_slices}) of horizontal slices"
+
+    if vertical_num_slices >= maximum_slices:
+        recommended_vertical_stride = max(1, image_h // maximum_slices) + 1
+        assert (
+            False
+        ), f"Too computationally expensive with {vertical_num_slices} slices, try a higher vertical stride (recommended minimum: {recommended_vertical_stride})"
+
+    if horizontal_num_slices >= maximum_slices:
+        recommended_horizontal_stride = max(1, image_w // maximum_slices) + 1
+        assert (
+            False
+        ), f"Too computationally expensive with {horizontal_num_slices} slices, try a higher horizontal stride (recommended minimum: {recommended_horizontal_stride})"
+
+    for v_slice_idx in range(vertical_num_slices):
+        v_start = max(0, (v_slice_idx * vertical_stride))
+        v_end = min(((v_slice_idx + 1) * vertical_stride), image_h)
+        vertical_slice = image[v_start:v_end, :]
+        for h_slice_idx in range(horizontal_num_slices):
+            h_start = max(0, (h_slice_idx * horizontal_stride))
+            h_end = min(((h_slice_idx + 1) * horizontal_stride), image_w)
+            horizontal_slice = vertical_slice[:, h_start:h_end]
+
+            yield (horizontal_slice, v_start, h_start)
+
+
+def calculate_box_extents(box):
+    min_x = box[0][0]
+    max_x = box[1][0]
+    min_y = box[0][1]
+    max_y = box[2][1]
+    return min_x, max_x, min_y, max_y
+
+
+def merge_boxes(box1, box2, x_threshold, y_threshold):
+    min_x1, max_x1, min_y1, max_y1 = calculate_box_extents(box1)
+    min_x2, max_x2, min_y2, max_y2 = calculate_box_extents(box2)
+
+    if (
+        abs(min_y1 - min_y2) <= y_threshold
+        and abs(max_y1 - max_y2) <= y_threshold
+        and abs(max_x1 - min_x2) <= x_threshold
+    ):
+        new_xmin = min(min_x1, min_x2)
+        new_xmax = max(max_x1, max_x2)
+        new_ymin = min(min_y1, min_y2)
+        new_ymax = max(max_y1, max_y2)
+        return [
+            [new_xmin, new_ymin],
+            [new_xmax, new_ymin],
+            [new_xmax, new_ymax],
+            [new_xmin, new_ymax],
+        ]
+    else:
+        return None
+
+
+def merge_fragmented(boxes, x_threshold=10, y_threshold=10):
+    merged_boxes = []
+    visited = set()
+
+    for i, box1 in enumerate(boxes):
+        if i in visited:
+            continue
+
+        merged_box = [point[:] for point in box1]
+
+        for j, box2 in enumerate(boxes[i + 1 :], start=i + 1):
+            if j not in visited:
+                merged_result = merge_boxes(
+                    merged_box, box2, x_threshold=x_threshold, y_threshold=y_threshold
+                )
+                if merged_result:
+                    merged_box = merged_result
+                    visited.add(j)
+
+        merged_boxes.append(merged_box)
+
+    if len(merged_boxes) == len(boxes):
+        return np.array(merged_boxes)
+    else:
+        return merge_fragmented(merged_boxes, x_threshold, y_threshold)
+
+
+def check_gpu(use_gpu):
+    if use_gpu and (
+        not paddle.is_compiled_with_cuda() or paddle.device.get_device() == "cpu"
+    ):
+        use_gpu = False
+    return use_gpu
+
+
+if __name__ == "__main__":
+    pass
diff --git a/tools/infer_cls.py b/tools/infer_cls.py
new file mode 100755
index 0000000..6c26ff4
--- /dev/null
+++ b/tools/infer_cls.py
@@ -0,0 +1,84 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+
+import paddle
+
+from ppocr.data import create_operators, transform
+from ppocr.modeling.architectures import build_model
+from ppocr.postprocess import build_post_process
+from ppocr.utils.save_load import load_model
+from ppocr.utils.utility import get_image_file_list
+import tools.program as program
+
+
+def main():
+    global_config = config["Global"]
+
+    # build post process
+    post_process_class = build_post_process(config["PostProcess"], global_config)
+
+    # build model
+    model = build_model(config["Architecture"])
+
+    load_model(config, model)
+
+    # create data ops
+    transforms = []
+    for op in config["Eval"]["dataset"]["transforms"]:
+        op_name = list(op)[0]
+        if "Label" in op_name:
+            continue
+        elif op_name == "KeepKeys":
+            op[op_name]["keep_keys"] = ["image"]
+        elif op_name == "SSLRotateResize":
+            op[op_name]["mode"] = "test"
+        transforms.append(op)
+    global_config["infer_mode"] = True
+    ops = create_operators(transforms, global_config)
+
+    model.eval()
+    for file in get_image_file_list(config["Global"]["infer_img"]):
+        logger.info("infer_img: {}".format(file))
+        with open(file, "rb") as f:
+            img = f.read()
+            data = {"image": img}
+        batch = transform(data, ops)
+
+        images = np.expand_dims(batch[0], axis=0)
+        images = paddle.to_tensor(images)
+        preds = model(images)
+        post_result = post_process_class(preds)
+        for rec_result in post_result:
+            logger.info("\t result: {}".format(rec_result))
+    logger.info("success!")
+
+
+if __name__ == "__main__":
+    config, device, logger, vdl_writer = program.preprocess()
+    main()
diff --git a/tools/infer_det.py b/tools/infer_det.py
new file mode 100755
index 0000000..3c001c5
--- /dev/null
+++ b/tools/infer_det.py
@@ -0,0 +1,143 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+
+import cv2
+import json
+import paddle
+
+from ppocr.data import create_operators, transform
+from ppocr.modeling.architectures import build_model
+from ppocr.postprocess import build_post_process
+from ppocr.utils.save_load import load_model
+from ppocr.utils.utility import get_image_file_list
+import tools.program as program
+
+
+def draw_det_res(dt_boxes, config, img, img_name, save_path):
+    import cv2
+
+    src_im = img
+    for box in dt_boxes:
+        box = np.array(box).astype(np.int32).reshape((-1, 1, 2))
+        cv2.polylines(src_im, [box], True, color=(255, 255, 0), thickness=2)
+    if not os.path.exists(save_path):
+        os.makedirs(save_path)
+    save_path = os.path.join(save_path, os.path.basename(img_name))
+    cv2.imwrite(save_path, src_im)
+    logger.info("The detected Image saved in {}".format(save_path))
+
+
+@paddle.no_grad()
+def main():
+    global_config = config["Global"]
+
+    # build model
+    model = build_model(config["Architecture"])
+
+    load_model(config, model)
+    # build post process
+    post_process_class = build_post_process(config["PostProcess"])
+
+    # create data ops
+    transforms = []
+    for op in config["Eval"]["dataset"]["transforms"]:
+        op_name = list(op)[0]
+        if "Label" in op_name:
+            continue
+        elif op_name == "KeepKeys":
+            op[op_name]["keep_keys"] = ["image", "shape"]
+        transforms.append(op)
+
+    ops = create_operators(transforms, global_config)
+
+    save_res_path = config["Global"]["save_res_path"]
+    if not os.path.exists(os.path.dirname(save_res_path)):
+        os.makedirs(os.path.dirname(save_res_path))
+
+    model.eval()
+    infer_img = config["Global"]["infer_img"]
+    # Support txt file containing one image path per line
+    if infer_img.endswith(".txt") and os.path.isfile(infer_img):
+        with open(infer_img, "r") as f:
+            img_file_list = [line.strip() for line in f if line.strip()]
+    else:
+        img_file_list = get_image_file_list(infer_img)
+    with open(save_res_path, "wb") as fout:
+        for file in img_file_list:
+            logger.info("infer_img: {}".format(file))
+            with open(file, "rb") as f:
+                img = f.read()
+                data = {"image": img}
+            batch = transform(data, ops)
+
+            images = np.expand_dims(batch[0], axis=0)
+            shape_list = np.expand_dims(batch[1], axis=0)
+            images = paddle.to_tensor(images)
+            preds = model(images)
+            post_result = post_process_class(preds, shape_list)
+
+            src_img = cv2.imread(file)
+
+            dt_boxes_json = []
+            # parser boxes if post_result is dict
+            if isinstance(post_result, dict):
+                det_box_json = {}
+                for k in post_result.keys():
+                    boxes = post_result[k][0]["points"]
+                    dt_boxes_list = []
+                    for box in boxes:
+                        tmp_json = {"transcription": ""}
+                        tmp_json["points"] = np.array(box).tolist()
+                        dt_boxes_list.append(tmp_json)
+                    det_box_json[k] = dt_boxes_list
+                    save_det_path = os.path.dirname(
+                        config["Global"]["save_res_path"]
+                    ) + "/det_results_{}/".format(k)
+                    draw_det_res(boxes, config, src_img, file, save_det_path)
+            else:
+                boxes = post_result[0]["points"]
+                dt_boxes_json = []
+                # write result
+                for box in boxes:
+                    tmp_json = {"transcription": ""}
+                    tmp_json["points"] = np.array(box).tolist()
+                    dt_boxes_json.append(tmp_json)
+                save_det_path = (
+                    os.path.dirname(config["Global"]["save_res_path"]) + "/det_results/"
+                )
+                draw_det_res(boxes, config, src_img, file, save_det_path)
+            otstr = file + "\t" + json.dumps(dt_boxes_json) + "\n"
+            fout.write(otstr.encode())
+
+    logger.info("success!")
+
+
+if __name__ == "__main__":
+    config, device, logger, vdl_writer = program.preprocess()
+    main()
diff --git a/tools/infer_e2e.py b/tools/infer_e2e.py
new file mode 100755
index 0000000..02cd68a
--- /dev/null
+++ b/tools/infer_e2e.py
@@ -0,0 +1,170 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+
+import cv2
+import json
+import paddle
+
+from ppocr.data import create_operators, transform
+from ppocr.modeling.architectures import build_model
+from ppocr.postprocess import build_post_process
+from ppocr.utils.save_load import load_model
+from ppocr.utils.utility import get_image_file_list
+import tools.program as program
+from PIL import Image, ImageDraw, ImageFont
+import math
+
+
+def draw_e2e_res_for_chinese(
+    image, boxes, txts, config, img_name, font_path="./doc/simfang.ttf"
+):
+    h, w = image.height, image.width
+    img_left = image.copy()
+    img_right = Image.new("RGB", (w, h), (255, 255, 255))
+
+    import random
+
+    random.seed(0)
+    draw_left = ImageDraw.Draw(img_left)
+    draw_right = ImageDraw.Draw(img_right)
+    for idx, (box, txt) in enumerate(zip(boxes, txts)):
+        box = np.array(box)
+        box = [tuple(x) for x in box]
+        color = (random.randint(0, 255), random.randint(0, 255), random.randint(0, 255))
+        draw_left.polygon(box, fill=color)
+        draw_right.polygon(box, outline=color)
+        font = ImageFont.truetype(font_path, 15, encoding="utf-8")
+        draw_right.text([box[0][0], box[0][1]], txt, fill=(0, 0, 0), font=font)
+    img_left = Image.blend(image, img_left, 0.5)
+    img_show = Image.new("RGB", (w * 2, h), (255, 255, 255))
+    img_show.paste(img_left, (0, 0, w, h))
+    img_show.paste(img_right, (w, 0, w * 2, h))
+
+    save_e2e_path = os.path.dirname(config["Global"]["save_res_path"]) + "/e2e_results/"
+    if not os.path.exists(save_e2e_path):
+        os.makedirs(save_e2e_path)
+    save_path = os.path.join(save_e2e_path, os.path.basename(img_name))
+    cv2.imwrite(save_path, np.array(img_show)[:, :, ::-1])
+    logger.info("The e2e Image saved in {}".format(save_path))
+
+
+def draw_e2e_res(dt_boxes, strs, config, img, img_name):
+    if len(dt_boxes) > 0:
+        src_im = img
+        for box, str in zip(dt_boxes, strs):
+            box = box.astype(np.int32).reshape((-1, 1, 2))
+            cv2.polylines(src_im, [box], True, color=(255, 255, 0), thickness=2)
+            cv2.putText(
+                src_im,
+                str,
+                org=(int(box[0, 0, 0]), int(box[0, 0, 1])),
+                fontFace=cv2.FONT_HERSHEY_COMPLEX,
+                fontScale=0.7,
+                color=(0, 255, 0),
+                thickness=1,
+            )
+        save_det_path = (
+            os.path.dirname(config["Global"]["save_res_path"]) + "/e2e_results/"
+        )
+        if not os.path.exists(save_det_path):
+            os.makedirs(save_det_path)
+        save_path = os.path.join(save_det_path, os.path.basename(img_name))
+        cv2.imwrite(save_path, src_im)
+        logger.info("The e2e Image saved in {}".format(save_path))
+
+
+def main():
+    global_config = config["Global"]
+
+    # build model
+    model = build_model(config["Architecture"])
+
+    load_model(config, model)
+
+    # build post process
+    post_process_class = build_post_process(config["PostProcess"], global_config)
+
+    # create data ops
+    transforms = []
+    for op in config["Eval"]["dataset"]["transforms"]:
+        op_name = list(op)[0]
+        if "Label" in op_name:
+            continue
+        elif op_name == "KeepKeys":
+            op[op_name]["keep_keys"] = ["image", "shape"]
+        transforms.append(op)
+
+    ops = create_operators(transforms, global_config)
+
+    save_res_path = config["Global"]["save_res_path"]
+    if not os.path.exists(os.path.dirname(save_res_path)):
+        os.makedirs(os.path.dirname(save_res_path))
+
+    model.eval()
+    with open(save_res_path, "wb") as fout:
+        for file in get_image_file_list(config["Global"]["infer_img"]):
+            logger.info("infer_img: {}".format(file))
+            with open(file, "rb") as f:
+                img = f.read()
+                data = {"image": img}
+            batch = transform(data, ops)
+            images = np.expand_dims(batch[0], axis=0)
+            shape_list = np.expand_dims(batch[1], axis=0)
+            images = paddle.to_tensor(images)
+            preds = model(images)
+            post_result = post_process_class(preds, shape_list)
+            points, strs = post_result["points"], post_result["texts"]
+            # write result
+            dt_boxes_json = []
+            for poly, str in zip(points, strs):
+                tmp_json = {"transcription": str}
+                tmp_json["points"] = poly.tolist()
+                dt_boxes_json.append(tmp_json)
+            otstr = file + "\t" + json.dumps(dt_boxes_json) + "\n"
+            fout.write(otstr.encode())
+            src_img = cv2.imread(file)
+            if global_config["infer_visual_type"] == "EN":
+                draw_e2e_res(points, strs, config, src_img, file)
+            elif global_config["infer_visual_type"] == "CN":
+                src_img = Image.fromarray(cv2.cvtColor(src_img, cv2.COLOR_BGR2RGB))
+                draw_e2e_res_for_chinese(
+                    src_img,
+                    points,
+                    strs,
+                    config,
+                    file,
+                    font_path="./doc/fonts/simfang.ttf",
+                )
+
+    logger.info("success!")
+
+
+if __name__ == "__main__":
+    config, device, logger, vdl_writer = program.preprocess()
+    main()
diff --git a/tools/infer_kie.py b/tools/infer_kie.py
new file mode 100755
index 0000000..f3efaf9
--- /dev/null
+++ b/tools/infer_kie.py
@@ -0,0 +1,186 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+import paddle.nn.functional as F
+
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+
+import cv2
+import paddle
+
+from ppocr.data import create_operators, transform
+from ppocr.modeling.architectures import build_model
+from ppocr.utils.save_load import load_model
+import tools.program as program
+import time
+
+
+def read_class_list(filepath):
+    ret = {}
+    with open(filepath, "r") as f:
+        lines = f.readlines()
+        for idx, line in enumerate(lines):
+            ret[idx] = line.strip("\n")
+    return ret
+
+
+def draw_kie_result(batch, node, idx_to_cls, count):
+    img = batch[6].copy()
+    boxes = batch[7]
+    h, w = img.shape[:2]
+    pred_img = np.ones((h, w * 2, 3), dtype=np.uint8) * 255
+    max_value, max_idx = paddle.max(node, -1), paddle.argmax(node, -1)
+    node_pred_label = max_idx.numpy().tolist()
+    node_pred_score = max_value.numpy().tolist()
+
+    for i, box in enumerate(boxes):
+        if i >= len(node_pred_label):
+            break
+        new_box = [
+            [box[0], box[1]],
+            [box[2], box[1]],
+            [box[2], box[3]],
+            [box[0], box[3]],
+        ]
+        Pts = np.array([new_box], np.int32)
+        cv2.polylines(
+            img, [Pts.reshape((-1, 1, 2))], True, color=(255, 255, 0), thickness=1
+        )
+        x_min = int(min([point[0] for point in new_box]))
+        y_min = int(min([point[1] for point in new_box]))
+
+        pred_label = node_pred_label[i]
+        if pred_label in idx_to_cls:
+            pred_label = idx_to_cls[pred_label]
+        pred_score = "{:.2f}".format(node_pred_score[i])
+        text = pred_label + "(" + pred_score + ")"
+        cv2.putText(
+            pred_img,
+            text,
+            (x_min * 2, y_min),
+            cv2.FONT_HERSHEY_SIMPLEX,
+            0.5,
+            (255, 0, 0),
+            1,
+        )
+    vis_img = np.ones((h, w * 3, 3), dtype=np.uint8) * 255
+    vis_img[:, :w] = img
+    vis_img[:, w:] = pred_img
+    save_kie_path = os.path.dirname(config["Global"]["save_res_path"]) + "/kie_results/"
+    if not os.path.exists(save_kie_path):
+        os.makedirs(save_kie_path)
+    save_path = os.path.join(save_kie_path, str(count) + ".png")
+    cv2.imwrite(save_path, vis_img)
+    logger.info("The Kie Image saved in {}".format(save_path))
+
+
+def write_kie_result(fout, node, data):
+    """
+    Write infer result to output file, sorted by the predict label of each line.
+    The format keeps the same as the input with additional score attribute.
+    """
+    import json
+
+    label = data["label"]
+    annotations = json.loads(label)
+    max_value, max_idx = paddle.max(node, -1), paddle.argmax(node, -1)
+    node_pred_label = max_idx.numpy().tolist()
+    node_pred_score = max_value.numpy().tolist()
+    res = []
+    for i, label in enumerate(node_pred_label):
+        pred_score = "{:.2f}".format(node_pred_score[i])
+        pred_res = {
+            "label": label,
+            "transcription": annotations[i]["transcription"],
+            "score": pred_score,
+            "points": annotations[i]["points"],
+        }
+        res.append(pred_res)
+    res.sort(key=lambda x: x["label"])
+    fout.writelines([json.dumps(res, ensure_ascii=False) + "\n"])
+
+
+def main():
+    global_config = config["Global"]
+
+    # build model
+    model = build_model(config["Architecture"])
+    load_model(config, model)
+
+    # create data ops
+    transforms = []
+    for op in config["Eval"]["dataset"]["transforms"]:
+        transforms.append(op)
+
+    data_dir = config["Eval"]["dataset"]["data_dir"]
+
+    ops = create_operators(transforms, global_config)
+
+    save_res_path = config["Global"]["save_res_path"]
+    class_path = config["Global"]["class_path"]
+    idx_to_cls = read_class_list(class_path)
+    os.makedirs(os.path.dirname(save_res_path), exist_ok=True)
+
+    model.eval()
+
+    warmup_times = 0
+    count_t = []
+    with open(save_res_path, "w") as fout:
+        with open(config["Global"]["infer_img"], "rb") as f:
+            lines = f.readlines()
+            for index, data_line in enumerate(lines):
+                if index == 10:
+                    warmup_t = time.time()
+                data_line = data_line.decode("utf-8")
+                substr = data_line.strip("\n").split("\t")
+                img_path, label = data_dir + "/" + substr[0], substr[1]
+                data = {"img_path": img_path, "label": label}
+                with open(data["img_path"], "rb") as f:
+                    img = f.read()
+                    data["image"] = img
+                st = time.time()
+                batch = transform(data, ops)
+                batch_pred = [0] * len(batch)
+                for i in range(len(batch)):
+                    batch_pred[i] = paddle.to_tensor(np.expand_dims(batch[i], axis=0))
+                st = time.time()
+                node, edge = model(batch_pred)
+                node = F.softmax(node, -1)
+                count_t.append(time.time() - st)
+                draw_kie_result(batch, node, idx_to_cls, index)
+                write_kie_result(fout, node, data)
+        fout.close()
+    logger.info("success!")
+    logger.info(
+        "It took {} s for predict {} images.".format(np.sum(count_t), len(count_t))
+    )
+    ips = len(count_t[warmup_times:]) / np.sum(count_t[warmup_times:])
+    logger.info("The ips is {} images/s".format(ips))
+
+
+if __name__ == "__main__":
+    config, device, logger, vdl_writer = program.preprocess()
+    main()
diff --git a/tools/infer_kie_token_ser.py b/tools/infer_kie_token_ser.py
new file mode 100755
index 0000000..8fe25ac
--- /dev/null
+++ b/tools/infer_kie_token_ser.py
@@ -0,0 +1,178 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+import cv2
+import json
+import paddle
+
+from ppocr.data import create_operators, transform
+from ppocr.modeling.architectures import build_model
+from ppocr.postprocess import build_post_process
+from ppocr.utils.save_load import load_model
+from ppocr.utils.visual import draw_ser_results
+from ppocr.utils.utility import get_image_file_list, load_vqa_bio_label_maps
+import tools.program as program
+
+
+def to_tensor(data):
+    import numbers
+    from collections import defaultdict
+
+    data_dict = defaultdict(list)
+    to_tensor_idxs = []
+
+    for idx, v in enumerate(data):
+        if isinstance(v, (np.ndarray, paddle.Tensor, numbers.Number)):
+            if idx not in to_tensor_idxs:
+                to_tensor_idxs.append(idx)
+        data_dict[idx].append(v)
+    for idx in to_tensor_idxs:
+        data_dict[idx] = paddle.to_tensor(data_dict[idx])
+    return list(data_dict.values())
+
+
+class SerPredictor(object):
+    def __init__(self, config):
+        global_config = config["Global"]
+        self.algorithm = config["Architecture"]["algorithm"]
+
+        # build post process
+        self.post_process_class = build_post_process(
+            config["PostProcess"], global_config
+        )
+
+        # build model
+        self.model = build_model(config["Architecture"])
+
+        load_model(config, self.model, model_type=config["Architecture"]["model_type"])
+
+        from paddleocr import PaddleOCR
+
+        self.ocr_engine = PaddleOCR(
+            use_angle_cls=False,
+            show_log=False,
+            rec_model_dir=global_config.get("kie_rec_model_dir", None),
+            det_model_dir=global_config.get("kie_det_model_dir", None),
+            use_gpu=global_config["use_gpu"],
+        )
+
+        # create data ops
+        transforms = []
+        for op in config["Eval"]["dataset"]["transforms"]:
+            op_name = list(op)[0]
+            if "Label" in op_name:
+                op[op_name]["ocr_engine"] = self.ocr_engine
+            elif op_name == "KeepKeys":
+                op[op_name]["keep_keys"] = [
+                    "input_ids",
+                    "bbox",
+                    "attention_mask",
+                    "token_type_ids",
+                    "image",
+                    "labels",
+                    "segment_offset_id",
+                    "ocr_info",
+                    "entities",
+                ]
+
+            transforms.append(op)
+        if config["Global"].get("infer_mode", None) is None:
+            global_config["infer_mode"] = True
+        self.ops = create_operators(
+            config["Eval"]["dataset"]["transforms"], global_config
+        )
+        self.model.eval()
+
+    def __call__(self, data):
+        with open(data["img_path"], "rb") as f:
+            img = f.read()
+        data["image"] = img
+        batch = transform(data, self.ops)
+        batch = to_tensor(batch)
+        preds = self.model(batch)
+
+        post_result = self.post_process_class(
+            preds, segment_offset_ids=batch[6], ocr_infos=batch[7]
+        )
+        return post_result, batch
+
+
+if __name__ == "__main__":
+    config, device, logger, vdl_writer = program.preprocess()
+    os.makedirs(config["Global"]["save_res_path"], exist_ok=True)
+
+    ser_engine = SerPredictor(config)
+
+    if config["Global"].get("infer_mode", None) is False:
+        data_dir = config["Eval"]["dataset"]["data_dir"]
+        with open(config["Global"]["infer_img"], "rb") as f:
+            infer_imgs = f.readlines()
+    else:
+        infer_imgs = get_image_file_list(config["Global"]["infer_img"])
+
+    with open(
+        os.path.join(config["Global"]["save_res_path"], "infer_results.txt"),
+        "w",
+        encoding="utf-8",
+    ) as fout:
+        for idx, info in enumerate(infer_imgs):
+            if config["Global"].get("infer_mode", None) is False:
+                data_line = info.decode("utf-8")
+                substr = data_line.strip("\n").split("\t")
+                img_path = os.path.join(data_dir, substr[0])
+                data = {"img_path": img_path, "label": substr[1]}
+            else:
+                img_path = info
+                data = {"img_path": img_path}
+
+            save_img_path = os.path.join(
+                config["Global"]["save_res_path"],
+                os.path.splitext(os.path.basename(img_path))[0] + "_ser.jpg",
+            )
+
+            result, _ = ser_engine(data)
+            result = result[0]
+            fout.write(
+                img_path
+                + "\t"
+                + json.dumps(
+                    {
+                        "ocr_info": result,
+                    },
+                    ensure_ascii=False,
+                )
+                + "\n"
+            )
+            img_res = draw_ser_results(img_path, result)
+            cv2.imwrite(save_img_path, img_res)
+
+            logger.info(
+                "process: [{}/{}], save result to {}".format(
+                    idx, len(infer_imgs), save_img_path
+                )
+            )
diff --git a/tools/infer_kie_token_ser_re.py b/tools/infer_kie_token_ser_re.py
new file mode 100755
index 0000000..a9589ca
--- /dev/null
+++ b/tools/infer_kie_token_ser_re.py
@@ -0,0 +1,226 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+import cv2
+import json
+import paddle
+import paddle.distributed as dist
+
+from ppocr.data import create_operators, transform
+from ppocr.modeling.architectures import build_model
+from ppocr.postprocess import build_post_process
+from ppocr.utils.save_load import load_model
+from ppocr.utils.visual import draw_re_results
+from ppocr.utils.logging import get_logger
+from ppocr.utils.utility import get_image_file_list, load_vqa_bio_label_maps, print_dict
+from tools.program import ArgsParser, load_config, merge_config
+from tools.infer_kie_token_ser import SerPredictor
+
+
+class ReArgsParser(ArgsParser):
+    def __init__(self):
+        super(ReArgsParser, self).__init__()
+        self.add_argument(
+            "-c_ser", "--config_ser", help="ser configuration file to use"
+        )
+        self.add_argument(
+            "-o_ser", "--opt_ser", nargs="+", help="set ser configuration options "
+        )
+
+    def parse_args(self, argv=None):
+        args = super(ReArgsParser, self).parse_args(argv)
+        assert (
+            args.config_ser is not None
+        ), "Please specify --config_ser=ser_configure_file_path."
+        args.opt_ser = self._parse_opt(args.opt_ser)
+        return args
+
+
+def make_input(ser_inputs, ser_results):
+    entities_labels = {"HEADER": 0, "QUESTION": 1, "ANSWER": 2}
+    batch_size, max_seq_len = ser_inputs[0].shape[:2]
+    entities = ser_inputs[8][0]
+    ser_results = ser_results[0]
+    assert len(entities) == len(ser_results)
+
+    # entities
+    start = []
+    end = []
+    label = []
+    entity_idx_dict = {}
+    for i, (res, entity) in enumerate(zip(ser_results, entities)):
+        if res["pred"] == "O":
+            continue
+        entity_idx_dict[len(start)] = i
+        start.append(entity["start"])
+        end.append(entity["end"])
+        label.append(entities_labels[res["pred"]])
+
+    entities = np.full([max_seq_len + 1, 3], fill_value=-1, dtype=np.int64)
+    entities[0, 0] = len(start)
+    entities[1 : len(start) + 1, 0] = start
+    entities[0, 1] = len(end)
+    entities[1 : len(end) + 1, 1] = end
+    entities[0, 2] = len(label)
+    entities[1 : len(label) + 1, 2] = label
+
+    # relations
+    head = []
+    tail = []
+    for i in range(len(label)):
+        for j in range(len(label)):
+            if label[i] == 1 and label[j] == 2:
+                head.append(i)
+                tail.append(j)
+
+    relations = np.full([len(head) + 1, 2], fill_value=-1, dtype=np.int64)
+    relations[0, 0] = len(head)
+    relations[1 : len(head) + 1, 0] = head
+    relations[0, 1] = len(tail)
+    relations[1 : len(tail) + 1, 1] = tail
+
+    entities = np.expand_dims(entities, axis=0)
+    entities = np.repeat(entities, batch_size, axis=0)
+    relations = np.expand_dims(relations, axis=0)
+    relations = np.repeat(relations, batch_size, axis=0)
+
+    # remove ocr_info segment_offset_id and label in ser input
+    if isinstance(ser_inputs[0], paddle.Tensor):
+        entities = paddle.to_tensor(entities)
+        relations = paddle.to_tensor(relations)
+    ser_inputs = ser_inputs[:5] + [entities, relations]
+
+    entity_idx_dict_batch = []
+    for b in range(batch_size):
+        entity_idx_dict_batch.append(entity_idx_dict)
+    return ser_inputs, entity_idx_dict_batch
+
+
+class SerRePredictor(object):
+    def __init__(self, config, ser_config):
+        global_config = config["Global"]
+        if "infer_mode" in global_config:
+            ser_config["Global"]["infer_mode"] = global_config["infer_mode"]
+
+        self.ser_engine = SerPredictor(ser_config)
+
+        #  init re model
+
+        # build post process
+        self.post_process_class = build_post_process(
+            config["PostProcess"], global_config
+        )
+
+        # build model
+        self.model = build_model(config["Architecture"])
+
+        load_model(config, self.model, model_type=config["Architecture"]["model_type"])
+
+        self.model.eval()
+
+    def __call__(self, data):
+        ser_results, ser_inputs = self.ser_engine(data)
+        re_input, entity_idx_dict_batch = make_input(ser_inputs, ser_results)
+        if self.model.backbone.use_visual_backbone is False:
+            re_input.pop(4)
+        preds = self.model(re_input)
+        post_result = self.post_process_class(
+            preds, ser_results=ser_results, entity_idx_dict_batch=entity_idx_dict_batch
+        )
+        return post_result
+
+
+def preprocess():
+    FLAGS = ReArgsParser().parse_args()
+    config = load_config(FLAGS.config)
+    config = merge_config(config, FLAGS.opt)
+
+    ser_config = load_config(FLAGS.config_ser)
+    ser_config = merge_config(ser_config, FLAGS.opt_ser)
+
+    logger = get_logger()
+
+    # check if set use_gpu=True in paddlepaddle cpu version
+    use_gpu = config["Global"]["use_gpu"]
+
+    device = "gpu:{}".format(dist.ParallelEnv().dev_id) if use_gpu else "cpu"
+    device = paddle.set_device(device)
+
+    logger.info("{} re config {}".format("*" * 10, "*" * 10))
+    print_dict(config, logger)
+    logger.info("\n")
+    logger.info("{} ser config {}".format("*" * 10, "*" * 10))
+    print_dict(ser_config, logger)
+    logger.info("train with paddle {} and device {}".format(paddle.__version__, device))
+    return config, ser_config, device, logger
+
+
+if __name__ == "__main__":
+    config, ser_config, device, logger = preprocess()
+    os.makedirs(config["Global"]["save_res_path"], exist_ok=True)
+
+    ser_re_engine = SerRePredictor(config, ser_config)
+
+    if config["Global"].get("infer_mode", None) is False:
+        data_dir = config["Eval"]["dataset"]["data_dir"]
+        with open(config["Global"]["infer_img"], "rb") as f:
+            infer_imgs = f.readlines()
+    else:
+        infer_imgs = get_image_file_list(config["Global"]["infer_img"])
+
+    with open(
+        os.path.join(config["Global"]["save_res_path"], "infer_results.txt"),
+        "w",
+        encoding="utf-8",
+    ) as fout:
+        for idx, info in enumerate(infer_imgs):
+            if config["Global"].get("infer_mode", None) is False:
+                data_line = info.decode("utf-8")
+                substr = data_line.strip("\n").split("\t")
+                img_path = os.path.join(data_dir, substr[0])
+                data = {"img_path": img_path, "label": substr[1]}
+            else:
+                img_path = info
+                data = {"img_path": img_path}
+
+            save_img_path = os.path.join(
+                config["Global"]["save_res_path"],
+                os.path.splitext(os.path.basename(img_path))[0] + "_ser_re.jpg",
+            )
+
+            result = ser_re_engine(data)
+            result = result[0]
+            fout.write(img_path + "\t" + json.dumps(result, ensure_ascii=False) + "\n")
+            img_res = draw_re_results(img_path, result)
+            cv2.imwrite(save_img_path, img_res)
+
+            logger.info(
+                "process: [{}/{}], save result to {}".format(
+                    idx, len(infer_imgs), save_img_path
+                )
+            )
diff --git a/tools/infer_rec.py b/tools/infer_rec.py
new file mode 100755
index 0000000..be00880
--- /dev/null
+++ b/tools/infer_rec.py
@@ -0,0 +1,232 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+
+import os
+import sys
+import json
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+
+import paddle
+
+from ppocr.data import create_operators, transform
+from ppocr.modeling.architectures import build_model
+from ppocr.postprocess import build_post_process
+from ppocr.utils.save_load import load_model
+from ppocr.utils.utility import get_image_file_list
+import tools.program as program
+
+
+def main():
+    global_config = config["Global"]
+    if config["Architecture"].get("algorithm") in [
+        "UniMERNet",
+        "PP-FormulaNet-S",
+        "PP-FormulaNet-L",
+        "PP-FormulaNet_plus-S",
+        "PP-FormulaNet_plus-M",
+        "PP-FormulaNet_plus-L",
+    ]:
+        config["PostProcess"]["is_infer"] = True
+    # build post process
+    post_process_class = build_post_process(config["PostProcess"], global_config)
+
+    # build model
+    if hasattr(post_process_class, "character"):
+        char_num = len(getattr(post_process_class, "character"))
+        if config["Architecture"]["algorithm"] in [
+            "Distillation",
+        ]:  # distillation model
+            for key in config["Architecture"]["Models"]:
+                if (
+                    config["Architecture"]["Models"][key]["Head"]["name"] == "MultiHead"
+                ):  # multi head
+                    out_channels_list = {}
+                    if config["PostProcess"]["name"] == "DistillationSARLabelDecode":
+                        char_num = char_num - 2
+                    if config["PostProcess"]["name"] == "DistillationNRTRLabelDecode":
+                        char_num = char_num - 3
+                    out_channels_list["CTCLabelDecode"] = char_num
+                    out_channels_list["SARLabelDecode"] = char_num + 2
+                    out_channels_list["NRTRLabelDecode"] = char_num + 3
+                    config["Architecture"]["Models"][key]["Head"][
+                        "out_channels_list"
+                    ] = out_channels_list
+                else:
+                    config["Architecture"]["Models"][key]["Head"][
+                        "out_channels"
+                    ] = char_num
+        elif config["Architecture"]["Head"]["name"] == "MultiHead":  # multi head
+            out_channels_list = {}
+            char_num = len(getattr(post_process_class, "character"))
+            if config["PostProcess"]["name"] == "SARLabelDecode":
+                char_num = char_num - 2
+            if config["PostProcess"]["name"] == "NRTRLabelDecode":
+                char_num = char_num - 3
+            out_channels_list["CTCLabelDecode"] = char_num
+            out_channels_list["SARLabelDecode"] = char_num + 2
+            out_channels_list["NRTRLabelDecode"] = char_num + 3
+            config["Architecture"]["Head"]["out_channels_list"] = out_channels_list
+        else:  # base rec model
+            config["Architecture"]["Head"]["out_channels"] = char_num
+
+    if config["Architecture"].get("algorithm") in ["LaTeXOCR"]:
+        config["Architecture"]["Backbone"]["is_predict"] = True
+        config["Architecture"]["Backbone"]["is_export"] = True
+        config["Architecture"]["Head"]["is_export"] = True
+
+    model = build_model(config["Architecture"])
+
+    load_model(config, model)
+
+    # create data ops
+    transforms = []
+    for op in config["Eval"]["dataset"]["transforms"]:
+        op_name = list(op)[0]
+        if "Label" in op_name:
+            continue
+        elif op_name in ["RecResizeImg"]:
+            op[op_name]["infer_mode"] = True
+        elif op_name == "KeepKeys":
+            if config["Architecture"]["algorithm"] == "SRN":
+                op[op_name]["keep_keys"] = [
+                    "image",
+                    "encoder_word_pos",
+                    "gsrm_word_pos",
+                    "gsrm_slf_attn_bias1",
+                    "gsrm_slf_attn_bias2",
+                ]
+            elif config["Architecture"]["algorithm"] == "SAR":
+                op[op_name]["keep_keys"] = ["image", "valid_ratio"]
+            elif config["Architecture"]["algorithm"] == "RobustScanner":
+                op[op_name]["keep_keys"] = ["image", "valid_ratio", "word_positions"]
+            else:
+                op[op_name]["keep_keys"] = ["image"]
+        transforms.append(op)
+    global_config["infer_mode"] = True
+    ops = create_operators(transforms, global_config)
+
+    save_res_path = config["Global"].get(
+        "save_res_path", "./output/rec/predicts_rec.txt"
+    )
+    if not os.path.exists(os.path.dirname(save_res_path)):
+        os.makedirs(os.path.dirname(save_res_path))
+
+    model.eval()
+
+    infer_imgs = config["Global"]["infer_img"]
+    infer_list = config["Global"].get("infer_list", None)
+    with open(save_res_path, "w") as fout:
+        for file in get_image_file_list(infer_imgs, infer_list=infer_list):
+            logger.info("infer_img: {}".format(file))
+            with open(file, "rb") as f:
+                img = f.read()
+                if config["Architecture"]["algorithm"] in [
+                    "UniMERNet",
+                    "PP-FormulaNet-S",
+                    "PP-FormulaNet-L",
+                    "PP-FormulaNet_plus-S",
+                    "PP-FormulaNet_plus-M",
+                    "PP-FormulaNet_plus-L",
+                ]:
+                    data = {"image": img, "filename": file}
+                else:
+                    data = {"image": img}
+            batch = transform(data, ops)
+            if config["Architecture"]["algorithm"] == "SRN":
+                encoder_word_pos_list = np.expand_dims(batch[1], axis=0)
+                gsrm_word_pos_list = np.expand_dims(batch[2], axis=0)
+                gsrm_slf_attn_bias1_list = np.expand_dims(batch[3], axis=0)
+                gsrm_slf_attn_bias2_list = np.expand_dims(batch[4], axis=0)
+
+                others = [
+                    paddle.to_tensor(encoder_word_pos_list),
+                    paddle.to_tensor(gsrm_word_pos_list),
+                    paddle.to_tensor(gsrm_slf_attn_bias1_list),
+                    paddle.to_tensor(gsrm_slf_attn_bias2_list),
+                ]
+            if config["Architecture"]["algorithm"] == "SAR":
+                valid_ratio = np.expand_dims(batch[-1], axis=0)
+                img_metas = [paddle.to_tensor(valid_ratio)]
+            if config["Architecture"]["algorithm"] == "RobustScanner":
+                valid_ratio = np.expand_dims(batch[1], axis=0)
+                word_positions = np.expand_dims(batch[2], axis=0)
+                img_metas = [
+                    paddle.to_tensor(valid_ratio),
+                    paddle.to_tensor(word_positions),
+                ]
+            if config["Architecture"]["algorithm"] == "CAN":
+                image_mask = paddle.ones(
+                    (np.expand_dims(batch[0], axis=0).shape), dtype="float32"
+                )
+                label = paddle.ones((1, 36), dtype="int64")
+            images = np.expand_dims(batch[0], axis=0)
+            images = paddle.to_tensor(images)
+            if config["Architecture"]["algorithm"] == "SRN":
+                preds = model(images, others)
+            elif config["Architecture"]["algorithm"] == "SAR":
+                preds = model(images, img_metas)
+            elif config["Architecture"]["algorithm"] == "RobustScanner":
+                preds = model(images, img_metas)
+            elif config["Architecture"]["algorithm"] == "CAN":
+                preds = model([images, image_mask, label])
+            else:
+                preds = model(images)
+            post_result = post_process_class(preds)
+            info = None
+            if isinstance(post_result, dict):
+                rec_info = dict()
+                for key in post_result:
+                    if len(post_result[key][0]) >= 2:
+                        rec_info[key] = {
+                            "label": post_result[key][0][0],
+                            "score": float(post_result[key][0][1]),
+                        }
+                info = json.dumps(rec_info, ensure_ascii=False)
+            elif isinstance(post_result, list) and isinstance(post_result[0], int):
+                # for RFLearning CNT branch
+                info = str(post_result[0])
+            elif config["Architecture"]["algorithm"] in [
+                "LaTeXOCR",
+                "UniMERNet",
+                "PP-FormulaNet-S",
+                "PP-FormulaNet-L",
+                "PP-FormulaNet_plus-S",
+                "PP-FormulaNet_plus-M",
+                "PP-FormulaNet_plus-L",
+            ]:
+                info = str(post_result[0])
+            else:
+                if len(post_result[0]) >= 2:
+                    info = post_result[0][0] + "\t" + str(post_result[0][1])
+
+            if info is not None:
+                logger.info("\t result: {}".format(info))
+                fout.write(file + "\t" + info + "\n")
+    logger.info("success!")
+
+
+if __name__ == "__main__":
+    config, device, logger, vdl_writer = program.preprocess()
+    main()
diff --git a/tools/infer_sr.py b/tools/infer_sr.py
new file mode 100755
index 0000000..4285625
--- /dev/null
+++ b/tools/infer_sr.py
@@ -0,0 +1,101 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+
+import os
+import sys
+import json
+from PIL import Image
+import cv2
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.insert(0, __dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+
+import paddle
+
+from ppocr.data import create_operators, transform
+from ppocr.modeling.architectures import build_model
+from ppocr.postprocess import build_post_process
+from ppocr.utils.save_load import load_model
+from ppocr.utils.utility import get_image_file_list
+import tools.program as program
+
+
+def main():
+    global_config = config["Global"]
+
+    # build post process
+    post_process_class = build_post_process(config["PostProcess"], global_config)
+
+    # sr transform
+    config["Architecture"]["Transform"]["infer_mode"] = True
+
+    model = build_model(config["Architecture"])
+
+    load_model(config, model)
+
+    # create data ops
+    transforms = []
+    for op in config["Eval"]["dataset"]["transforms"]:
+        op_name = list(op)[0]
+        if "Label" in op_name:
+            continue
+        elif op_name in ["SRResize"]:
+            op[op_name]["infer_mode"] = True
+        elif op_name == "KeepKeys":
+            op[op_name]["keep_keys"] = ["img_lr"]
+        transforms.append(op)
+    global_config["infer_mode"] = True
+    ops = create_operators(transforms, global_config)
+
+    save_visual_path = config["Global"].get("save_visual", "infer_result/")
+    if not os.path.exists(os.path.dirname(save_visual_path)):
+        os.makedirs(os.path.dirname(save_visual_path))
+
+    model.eval()
+    for file in get_image_file_list(config["Global"]["infer_img"]):
+        logger.info("infer_img: {}".format(file))
+        img = Image.open(file).convert("RGB")
+        data = {"image_lr": img}
+        batch = transform(data, ops)
+        images = np.expand_dims(batch[0], axis=0)
+        images = paddle.to_tensor(images)
+
+        preds = model(images)
+        sr_img = preds["sr_img"][0]
+        lr_img = preds["lr_img"][0]
+        fm_sr = (sr_img.numpy() * 255).transpose(1, 2, 0).astype(np.uint8)
+        fm_lr = (lr_img.numpy() * 255).transpose(1, 2, 0).astype(np.uint8)
+        img_name_pure = os.path.split(file)[-1]
+        cv2.imwrite(
+            "{}/sr_{}".format(save_visual_path, img_name_pure), fm_sr[:, :, ::-1]
+        )
+        logger.info(
+            "The visualized image saved in infer_result/sr_{}".format(img_name_pure)
+        )
+
+    logger.info("success!")
+
+
+if __name__ == "__main__":
+    config, device, logger, vdl_writer = program.preprocess()
+    main()
diff --git a/tools/infer_table.py b/tools/infer_table.py
new file mode 100644
index 0000000..c386cef
--- /dev/null
+++ b/tools/infer_table.py
@@ -0,0 +1,120 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import numpy as np
+
+import os
+import sys
+import json
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "..")))
+
+os.environ["FLAGS_allocator_strategy"] = "auto_growth"
+
+import paddle
+from paddle.jit import to_static
+
+from ppocr.data import create_operators, transform
+from ppocr.modeling.architectures import build_model
+from ppocr.postprocess import build_post_process
+from ppocr.utils.save_load import load_model
+from ppocr.utils.utility import get_image_file_list
+from ppocr.utils.visual import draw_rectangle
+from tools.infer.utility import draw_boxes
+import tools.program as program
+import cv2
+
+
+@paddle.no_grad()
+def main(config, device, logger, vdl_writer):
+    global_config = config["Global"]
+
+    # build post process
+    post_process_class = build_post_process(config["PostProcess"], global_config)
+
+    # build model
+    if hasattr(post_process_class, "character"):
+        config["Architecture"]["Head"]["out_channels"] = len(
+            getattr(post_process_class, "character")
+        )
+
+    model = build_model(config["Architecture"])
+    algorithm = config["Architecture"]["algorithm"]
+
+    load_model(config, model)
+
+    # create data ops
+    transforms = []
+    for op in config["Eval"]["dataset"]["transforms"]:
+        op_name = list(op)[0]
+        if "Encode" in op_name:
+            continue
+        if op_name == "KeepKeys":
+            op[op_name]["keep_keys"] = ["image", "shape"]
+        transforms.append(op)
+
+    global_config["infer_mode"] = True
+    ops = create_operators(transforms, global_config)
+
+    save_res_path = config["Global"]["save_res_path"]
+    os.makedirs(save_res_path, exist_ok=True)
+
+    model.eval()
+    with open(
+        os.path.join(save_res_path, "infer.txt"), mode="w", encoding="utf-8"
+    ) as f_w:
+        for file in get_image_file_list(config["Global"]["infer_img"]):
+            logger.info("infer_img: {}".format(file))
+            with open(file, "rb") as f:
+                img = f.read()
+                data = {"image": img}
+            batch = transform(data, ops)
+            images = np.expand_dims(batch[0], axis=0)
+            shape_list = np.expand_dims(batch[1], axis=0)
+
+            images = paddle.to_tensor(images)
+            preds = model(images)
+            post_result = post_process_class(preds, [shape_list])
+
+            structure_str_list = post_result["structure_batch_list"][0]
+            bbox_list = post_result["bbox_batch_list"][0]
+            structure_str_list = structure_str_list[0]
+            structure_str_list = (
+                ["<html>", "<body>", "<table>"]
+                + structure_str_list
+                + ["</table>", "</body>", "</html>"]
+            )
+            bbox_list_str = json.dumps(bbox_list.tolist())
+
+            logger.info("result: {}, {}".format(structure_str_list, bbox_list_str))
+            f_w.write("result: {}, {}\n".format(structure_str_list, bbox_list_str))
+
+            if len(bbox_list) > 0 and len(bbox_list[0]) == 4:
+                img = draw_rectangle(file, bbox_list)
+            else:
+                img = draw_boxes(cv2.imread(file), bbox_list)
+            cv2.imwrite(os.path.join(save_res_path, os.path.basename(file)), img)
+            logger.info("save result to {}".format(save_res_path))
+        logger.info("success!")
+
+
+if __name__ == "__main__":
+    config, device, logger, vdl_writer = program.preprocess()
+    main(config, device, logger, vdl_writer)
diff --git a/tools/naive_sync_bn.py b/tools/naive_sync_bn.py
new file mode 100644
index 0000000..50da685
--- /dev/null
+++ b/tools/naive_sync_bn.py
@@ -0,0 +1,120 @@
+# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import paddle.distributed as dist
+import math
+import paddle
+import paddle.nn as nn
+
+
+class _AllReduce(paddle.autograd.PyLayer):
+    @staticmethod
+    def forward(ctx, input):
+        input_list = [paddle.zeros_like(input) for k in range(dist.get_world_size())]
+        # Use allgather instead of allreduce since I don't trust in-place operations ..
+        dist.all_gather(input_list, input, sync_op=True)
+        inputs = paddle.stack(input_list, axis=0)
+        return paddle.sum(inputs, axis=0)
+
+    @staticmethod
+    def backward(ctx, grad_output):
+        dist.all_reduce(grad_output, sync_op=True)
+        return grad_output
+
+
+def differentiable_all_reduce(input):
+    """
+    Differentiable counterpart of `dist.all_reduce`.
+    """
+    if (
+        not dist.is_available()
+        or not dist.is_initialized()
+        or dist.get_world_size() == 1
+    ):
+        return input
+    return _AllReduce.apply(input)
+
+
+class NaiveSyncBatchNorm(nn.BatchNorm2D):
+    def __init__(self, *args, stats_mode="", **kwargs):
+        super().__init__(*args, **kwargs)
+        assert stats_mode in ["", "N"]
+        self._stats_mode = stats_mode
+
+    def forward(self, input):
+        if dist.get_world_size() == 1 or not self.training:
+            return super().forward(input)
+
+        B, C = input.shape[0], input.shape[1]
+
+        mean = paddle.mean(input, axis=[0, 2, 3])
+        meansqr = paddle.mean(input * input, axis=[0, 2, 3])
+
+        if self._stats_mode == "":
+            assert (
+                B > 0
+            ), 'SyncBatchNorm(stats_mode="") does not support zero batch size.'
+            vec = paddle.concat([mean, meansqr], axis=0)
+            vec = differentiable_all_reduce(vec) * (1.0 / dist.get_world_size())
+            mean, meansqr = paddle.split(vec, [C, C])
+            momentum = (
+                1 - self._momentum
+            )  # NOTE: paddle has reverse momentum definition
+        else:
+            if B == 0:
+                vec = paddle.zeros([2 * C + 1], dtype=mean.dtype)
+                vec = vec + input.sum()  # make sure there is gradient w.r.t input
+            else:
+                vec = paddle.concat(
+                    [
+                        mean,
+                        meansqr,
+                        paddle.ones([1], dtype=mean.dtype),
+                    ],
+                    axis=0,
+                )
+            vec = differentiable_all_reduce(vec * B)
+
+            total_batch = vec[-1].detach()
+            momentum = total_batch.clip(max=1) * (
+                1 - self._momentum
+            )  # no update if total_batch is 0
+            mean, meansqr, _ = paddle.split(
+                vec / total_batch.clip(min=1), [C, C, int(vec.shape[0] - 2 * C)]
+            )  # avoid div-by-zero
+
+        var = meansqr - mean * mean
+        invstd = paddle.rsqrt(var + self._epsilon)
+        scale = self.weight * invstd
+        bias = self.bias - mean * scale
+        scale = scale.reshape([1, -1, 1, 1])
+        bias = bias.reshape([1, -1, 1, 1])
+
+        tmp_mean = self._mean + momentum * (mean.detach() - self._mean)
+        self._mean.set_value(tmp_mean)
+        tmp_variance = self._variance + (momentum * (var.detach() - self._variance))
+        self._variance.set_value(tmp_variance)
+        ret = input * scale + bias
+        return ret
+
+
+def convert_syncbn(model):
+    for n, m in model.named_children():
+        if isinstance(m, nn.layer.norm._BatchNormBase):
+            syncbn = NaiveSyncBatchNorm(
+                m._num_features, m._momentum, m._epsilon, m._weight_attr, m._bias_attr
+            )
+            setattr(model, n, syncbn)
+        else:
+            convert_syncbn(m)
diff --git a/tools/program.py b/tools/program.py
new file mode 100755
index 0000000..5db1ddd
--- /dev/null
+++ b/tools/program.py
@@ -0,0 +1,1010 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import gc
+import sys
+import platform
+import yaml
+import time
+import datetime
+import paddle
+import paddle.distributed as dist
+from tqdm import tqdm
+import cv2
+import numpy as np
+import copy
+from argparse import ArgumentParser, RawDescriptionHelpFormatter
+
+from ppocr.utils.stats import TrainingStats
+from ppocr.utils.save_load import save_model
+from ppocr.utils.utility import print_dict, AverageMeter
+from ppocr.utils.logging import get_logger
+from ppocr.utils.loggers import WandbLogger, Loggers
+from ppocr.utils import profiler
+from ppocr.data import build_dataloader
+from ppocr.utils.export_model import export
+
+
+class ArgsParser(ArgumentParser):
+    def __init__(self):
+        super(ArgsParser, self).__init__(formatter_class=RawDescriptionHelpFormatter)
+        self.add_argument("-c", "--config", help="configuration file to use")
+        self.add_argument("-o", "--opt", nargs="+", help="set configuration options")
+        self.add_argument(
+            "-p",
+            "--profiler_options",
+            type=str,
+            default=None,
+            help="The option of profiler, which should be in format "
+            '"key1=value1;key2=value2;key3=value3".',
+        )
+
+    def parse_args(self, argv=None):
+        args = super(ArgsParser, self).parse_args(argv)
+        assert args.config is not None, "Please specify --config=configure_file_path."
+        args.opt = self._parse_opt(args.opt)
+        return args
+
+    def _parse_opt(self, opts):
+        config = {}
+        if not opts:
+            return config
+        for s in opts:
+            s = s.strip()
+            k, v = s.split("=")
+            config[k] = yaml.load(v, Loader=yaml.SafeLoader)
+        return config
+
+
+def load_config(file_path):
+    """
+    Load config from yml/yaml file.
+    Args:
+        file_path (str): Path of the config file to be loaded.
+    Returns: global config
+    """
+    _, ext = os.path.splitext(file_path)
+    assert ext in [".yml", ".yaml"], "only support yaml files for now"
+    config = yaml.load(open(file_path, "rb"), Loader=yaml.SafeLoader)
+    return config
+
+
+def merge_config(config, opts):
+    """
+    Merge config into global config.
+    Args:
+        config (dict): Config to be merged.
+    Returns: global config
+    """
+    for key, value in opts.items():
+        if "." not in key:
+            if isinstance(value, dict) and key in config:
+                config[key].update(value)
+            else:
+                config[key] = value
+        else:
+            sub_keys = key.split(".")
+            assert sub_keys[0] in config, (
+                "the sub_keys can only be one of global_config: {}, but get: "
+                "{}, please check your running command".format(
+                    config.keys(), sub_keys[0]
+                )
+            )
+            cur = config[sub_keys[0]]
+            for idx, sub_key in enumerate(sub_keys[1:]):
+                if idx == len(sub_keys) - 2:
+                    cur[sub_key] = value
+                else:
+                    cur = cur[sub_key]
+    return config
+
+
+def check_device(
+    use_gpu,
+    use_xpu=False,
+    use_npu=False,
+    use_mlu=False,
+    use_gcu=False,
+    use_iluvatar_gpu=False,
+    use_metax_gpu=False,
+):
+    """
+    Log error and exit when set use_gpu=true in paddlepaddle
+    cpu version.
+    """
+    err = (
+        "Config {} cannot be set as true while your paddle "
+        "is not compiled with {} ! \nPlease try: \n"
+        "\t1. Install paddlepaddle to run model on {} \n"
+        "\t2. Set {} as false in config file to run "
+        "model on CPU"
+    )
+
+    try:
+        if use_gpu and use_xpu:
+            print("use_xpu and use_gpu can not both be true.")
+        if use_gpu and not paddle.is_compiled_with_cuda():
+            print(err.format("use_gpu", "cuda", "gpu", "use_gpu"))
+            sys.exit(1)
+        if use_xpu and not paddle.device.is_compiled_with_xpu():
+            print(err.format("use_xpu", "xpu", "xpu", "use_xpu"))
+            sys.exit(1)
+        if use_npu:
+            if (
+                int(paddle.version.major) != 0
+                and int(paddle.version.major) <= 2
+                and int(paddle.version.minor) <= 4
+            ):
+                if not paddle.device.is_compiled_with_npu():
+                    print(err.format("use_npu", "npu", "npu", "use_npu"))
+                    sys.exit(1)
+            # is_compiled_with_npu() has been updated after paddle-2.4
+            else:
+                if not paddle.device.is_compiled_with_custom_device("npu"):
+                    print(err.format("use_npu", "npu", "npu", "use_npu"))
+                    sys.exit(1)
+        if use_mlu and not paddle.device.is_compiled_with_mlu():
+            print(err.format("use_mlu", "mlu", "mlu", "use_mlu"))
+            sys.exit(1)
+        if use_gcu and not paddle.device.is_compiled_with_custom_device("gcu"):
+            print(err.format("use_gcu", "gcu", "gcu", "use_gcu"))
+            sys.exit(1)
+        if use_metax_gpu and not paddle.device.is_compiled_with_custom_device(
+            "metax_gpu"
+        ):
+            print(
+                err.format("use_metax_gpu", "metax_gpu", "metax_gpu", "use_metax_gpu")
+            )
+            sys.exit(1)
+
+    except Exception as e:
+        pass
+
+
+def to_float32(preds):
+    if isinstance(preds, dict):
+        for k in preds:
+            if isinstance(preds[k], dict) or isinstance(preds[k], list):
+                preds[k] = to_float32(preds[k])
+            elif isinstance(preds[k], paddle.Tensor):
+                preds[k] = preds[k].astype(paddle.float32)
+    elif isinstance(preds, list):
+        for k in range(len(preds)):
+            if isinstance(preds[k], dict):
+                preds[k] = to_float32(preds[k])
+            elif isinstance(preds[k], list):
+                preds[k] = to_float32(preds[k])
+            elif isinstance(preds[k], paddle.Tensor):
+                preds[k] = preds[k].astype(paddle.float32)
+    elif isinstance(preds, paddle.Tensor):
+        preds = preds.astype(paddle.float32)
+    return preds
+
+
+def train(
+    config,
+    train_dataloader,
+    valid_dataloader,
+    device,
+    model,
+    loss_class,
+    optimizer,
+    lr_scheduler,
+    post_process_class,
+    eval_class,
+    pre_best_model_dict,
+    logger,
+    step_pre_epoch,
+    log_writer=None,
+    scaler=None,
+    amp_level="O2",
+    amp_custom_black_list=[],
+    amp_custom_white_list=[],
+    amp_dtype="float16",
+    wd_scheduler=None,
+    ema=None,
+):
+    cal_metric_during_train = config["Global"].get("cal_metric_during_train", False)
+    calc_epoch_interval = config["Global"].get("calc_epoch_interval", 1)
+    log_smooth_window = config["Global"]["log_smooth_window"]
+    epoch_num = config["Global"]["epoch_num"]
+    print_batch_step = config["Global"]["print_batch_step"]
+    eval_batch_step = config["Global"]["eval_batch_step"]
+    eval_batch_epoch = config["Global"].get("eval_batch_epoch", None)
+    profiler_options = config["profiler_options"]
+    print_mem_info = config["Global"].get("print_mem_info", True)
+    uniform_output_enabled = config["Global"].get("uniform_output_enabled", False)
+
+    global_step = 0
+    if "global_step" in pre_best_model_dict:
+        global_step = pre_best_model_dict["global_step"]
+    start_eval_step = 0
+    if isinstance(eval_batch_step, list) and len(eval_batch_step) >= 2:
+        start_eval_step = eval_batch_step[0] if not eval_batch_epoch else 0
+        eval_batch_step = (
+            eval_batch_step[1]
+            if not eval_batch_epoch
+            else step_pre_epoch * eval_batch_epoch
+        )
+        if len(valid_dataloader) == 0:
+            logger.info(
+                "No Images in eval dataset, evaluation during training "
+                "will be disabled"
+            )
+            start_eval_step = 1e111
+        logger.info(
+            "During the training process, after the {}th iteration, "
+            "an evaluation is run every {} iterations".format(
+                start_eval_step, eval_batch_step
+            )
+        )
+    save_epoch_step = config["Global"]["save_epoch_step"]
+    save_model_dir = config["Global"]["save_model_dir"]
+    if not os.path.exists(save_model_dir):
+        os.makedirs(save_model_dir)
+    main_indicator = eval_class.main_indicator
+    best_model_dict = {main_indicator: 0}
+    best_model_dict.update(pre_best_model_dict)
+    train_stats = TrainingStats(log_smooth_window, ["lr"])
+    model_average = False
+    model.train()
+
+    use_srn = config["Architecture"]["algorithm"] == "SRN"
+    extra_input_models = [
+        "SRN",
+        "NRTR",
+        "SAR",
+        "SEED",
+        "SVTR",
+        "SVTR_LCNet",
+        "SPIN",
+        "VisionLAN",
+        "RobustScanner",
+        "RFL",
+        "DRRG",
+        "SATRN",
+        "SVTR_HGNet",
+        "ParseQ",
+        "CPPD",
+    ]
+    extra_input = False
+    if config["Architecture"]["algorithm"] == "Distillation":
+        for key in config["Architecture"]["Models"]:
+            extra_input = (
+                extra_input
+                or config["Architecture"]["Models"][key]["algorithm"]
+                in extra_input_models
+            )
+    else:
+        extra_input = config["Architecture"]["algorithm"] in extra_input_models
+    try:
+        model_type = config["Architecture"]["model_type"]
+    except:
+        model_type = None
+
+    algorithm = config["Architecture"]["algorithm"]
+
+    start_epoch = (
+        best_model_dict["start_epoch"] if "start_epoch" in best_model_dict else 1
+    )
+
+    total_samples = 0
+    train_reader_cost = 0.0
+    train_batch_cost = 0.0
+    reader_start = time.time()
+    eta_meter = AverageMeter()
+
+    max_iter = (
+        len(train_dataloader) - 1
+        if platform.system() == "Windows"
+        else len(train_dataloader)
+    )
+
+    for epoch in range(start_epoch, epoch_num + 1):
+        if train_dataloader.dataset.need_reset:
+            # Update index mapping + shared epoch (no disk I/O, no worker restart)
+            train_dataloader.dataset.reset_data_lines(seed=epoch, epoch=epoch)
+            max_iter = (
+                len(train_dataloader) - 1
+                if platform.system() == "Windows"
+                else len(train_dataloader)
+            )
+            # Match original behavior: fresh DistributedBatchSampler always
+            # starts with self.epoch=0 for its internal np shuffle seed
+            if hasattr(train_dataloader.batch_sampler, "set_epoch"):
+                train_dataloader.batch_sampler.set_epoch(0)
+
+        for idx, batch in enumerate(train_dataloader):
+            model.train()
+            profiler.add_profiler_step(profiler_options)
+            train_reader_cost += time.time() - reader_start
+            if idx >= max_iter:
+                break
+            lr = optimizer.get_lr()
+            images = batch[0]
+            if use_srn:
+                model_average = True
+            # use amp
+            if scaler:
+                with paddle.amp.auto_cast(
+                    level=amp_level,
+                    custom_black_list=amp_custom_black_list,
+                    custom_white_list=amp_custom_white_list,
+                    dtype=amp_dtype,
+                ):
+                    if model_type == "table" or extra_input:
+                        preds = model(images, data=batch[1:])
+                    elif model_type in ["kie"]:
+                        preds = model(batch)
+                    elif algorithm in ["CAN"]:
+                        preds = model(batch[:3])
+                    elif algorithm in [
+                        "LaTeXOCR",
+                        "UniMERNet",
+                        "PP-FormulaNet-S",
+                        "PP-FormulaNet-L",
+                        "PP-FormulaNet_plus-S",
+                        "PP-FormulaNet_plus-M",
+                        "PP-FormulaNet_plus-L",
+                    ]:
+                        preds = model(batch)
+                    else:
+                        preds = model(images)
+                preds = to_float32(preds)
+                loss = loss_class(preds, batch)
+                avg_loss = loss["loss"]
+                scaled_avg_loss = scaler.scale(avg_loss)
+                scaled_avg_loss.backward()
+                scaler.minimize(optimizer, scaled_avg_loss)
+            else:
+                if model_type == "table" or extra_input:
+                    preds = model(images, data=batch[1:])
+                elif model_type in ["kie", "sr"]:
+                    preds = model(batch)
+                elif algorithm in ["CAN"]:
+                    preds = model(batch[:3])
+                elif algorithm in [
+                    "LaTeXOCR",
+                    "UniMERNet",
+                    "PP-FormulaNet-S",
+                    "PP-FormulaNet-L",
+                    "PP-FormulaNet_plus-S",
+                    "PP-FormulaNet_plus-M",
+                    "PP-FormulaNet_plus-L",
+                ]:
+                    preds = model(batch)
+                else:
+                    preds = model(images)
+                loss = loss_class(preds, batch)
+                avg_loss = loss["loss"]
+                avg_loss.backward()
+                optimizer.step()
+
+            optimizer.clear_grad()
+
+            if ema is not None:
+                ema.update(model)
+
+            if (
+                cal_metric_during_train and epoch % calc_epoch_interval == 0
+            ):  # only rec and cls need
+                batch = [item.numpy() for item in batch]
+                if model_type in ["kie", "sr"]:
+                    eval_class(preds, batch)
+                elif model_type in ["table"]:
+                    post_result = post_process_class(preds, batch)
+                    eval_class(post_result, batch)
+                elif algorithm in ["CAN"]:
+                    model_type = "can"
+                    eval_class(preds[0], batch[2:], epoch_reset=(idx == 0))
+                elif algorithm in ["LaTeXOCR"]:
+                    model_type = "latexocr"
+                    post_result = post_process_class(preds, batch[1], mode="train")
+                    eval_class(post_result[0], post_result[1], epoch_reset=(idx == 0))
+                elif algorithm in ["UniMERNet"]:
+                    model_type = "unimernet"
+                    post_result = post_process_class(preds[0], batch[1], mode="train")
+                    eval_class(post_result[0], post_result[1], epoch_reset=(idx == 0))
+                elif algorithm in [
+                    "PP-FormulaNet-S",
+                    "PP-FormulaNet-L",
+                    "PP-FormulaNet_plus-S",
+                    "PP-FormulaNet_plus-M",
+                    "PP-FormulaNet_plus-L",
+                ]:
+                    model_type = "pp_formulanet"
+                    post_result = post_process_class(preds[0], batch[1], mode="train")
+                    eval_class(post_result[0], post_result[1], epoch_reset=(idx == 0))
+                else:
+                    if config["Loss"]["name"] in [
+                        "MultiLoss",
+                        "MultiLoss_v2",
+                    ]:  # for multi head loss
+                        post_result = post_process_class(
+                            preds["ctc"], batch[1]
+                        )  # for CTC head out
+                    elif config["Loss"]["name"] in ["VLLoss"]:
+                        post_result = post_process_class(preds, batch[1], batch[-1])
+                    else:
+                        post_result = post_process_class(preds, batch[1])
+                    eval_class(post_result, batch)
+                metric = eval_class.get_metric()
+                train_stats.update(metric)
+
+            train_batch_time = time.time() - reader_start
+            train_batch_cost += train_batch_time
+            eta_meter.update(train_batch_time)
+            global_step += 1
+            total_samples += len(images)
+
+            if not isinstance(lr_scheduler, float):
+                lr_scheduler.step()
+
+            if wd_scheduler is not None:
+                wd_scheduler.step()
+
+            # logger and visualdl
+            stats = {
+                k: float(v) if v.shape == [] else v.numpy().mean()
+                for k, v in loss.items()
+            }
+            stats["lr"] = lr
+            if wd_scheduler is not None:
+                stats["wd"] = wd_scheduler.get_wd()
+            train_stats.update(stats)
+
+            if log_writer is not None and dist.get_rank() == 0:
+                log_writer.log_metrics(
+                    metrics=train_stats.get(), prefix="TRAIN", step=global_step
+                )
+
+            if (global_step > 0 and global_step % print_batch_step == 0) or (
+                idx >= len(train_dataloader) - 1
+            ):
+                logs = train_stats.log()
+
+                eta_sec = (
+                    (epoch_num + 1 - epoch) * len(train_dataloader) - idx - 1
+                ) * eta_meter.avg
+                eta_sec_format = str(datetime.timedelta(seconds=int(eta_sec)))
+                max_mem_reserved_str = ""
+                max_mem_allocated_str = ""
+                if paddle.device.is_compiled_with_cuda() and print_mem_info:
+                    max_mem_reserved_str = f", max_mem_reserved: {paddle.device.cuda.max_memory_reserved() // (1024 ** 2)} MB,"
+                    max_mem_allocated_str = f" max_mem_allocated: {paddle.device.cuda.max_memory_allocated() // (1024 ** 2)} MB"
+                strs = (
+                    "epoch: [{}/{}], global_step: {}, {}, avg_reader_cost: "
+                    "{:.5f} s, avg_batch_cost: {:.5f} s, avg_samples: {}, "
+                    "ips: {:.5f} samples/s, eta: {}{}{}".format(
+                        epoch,
+                        epoch_num,
+                        global_step,
+                        logs,
+                        train_reader_cost / print_batch_step,
+                        train_batch_cost / print_batch_step,
+                        total_samples / print_batch_step,
+                        total_samples / train_batch_cost,
+                        eta_sec_format,
+                        max_mem_reserved_str,
+                        max_mem_allocated_str,
+                    )
+                )
+                logger.info(strs)
+
+                total_samples = 0
+                train_reader_cost = 0.0
+                train_batch_cost = 0.0
+            # eval
+            if (
+                global_step > start_eval_step
+                and (global_step - start_eval_step) % eval_batch_step == 0
+                and dist.get_rank() == 0
+            ):
+                if model_average:
+                    Model_Average = paddle.incubate.ModelAverage(
+                        0.15,
+                        parameters=model.parameters(),
+                        min_average_window=10000,
+                        max_average_window=15625,
+                    )
+                    Model_Average.apply()
+                # Apply EMA weights for eval and save
+                _ema_train_state = None
+                if ema is not None:
+                    _ema_train_state = copy.deepcopy(model.state_dict())
+                    model.set_state_dict(ema.apply())
+                cur_metric = eval(
+                    model,
+                    valid_dataloader,
+                    post_process_class,
+                    eval_class,
+                    model_type,
+                    extra_input=extra_input,
+                    scaler=scaler,
+                    amp_level=amp_level,
+                    amp_custom_black_list=amp_custom_black_list,
+                    amp_custom_white_list=amp_custom_white_list,
+                    amp_dtype=amp_dtype,
+                )
+                cur_metric_str = "cur metric, {}".format(
+                    ", ".join(["{}: {}".format(k, v) for k, v in cur_metric.items()])
+                )
+                logger.info(cur_metric_str)
+
+                # logger metric
+                if log_writer is not None:
+                    log_writer.log_metrics(
+                        metrics=cur_metric, prefix="EVAL", step=global_step
+                    )
+
+                if cur_metric[main_indicator] >= best_model_dict[main_indicator]:
+                    best_model_dict.update(cur_metric)
+                    best_model_dict["best_epoch"] = epoch
+                    prefix = "best_accuracy"
+                    if uniform_output_enabled:
+                        export(
+                            config,
+                            model,
+                            os.path.join(save_model_dir, prefix, "inference"),
+                        )
+                        gc.collect()
+                        model_info = {"epoch": epoch, "metric": best_model_dict}
+                    else:
+                        model_info = None
+                    save_model(
+                        model,
+                        optimizer,
+                        (
+                            os.path.join(save_model_dir, prefix)
+                            if uniform_output_enabled
+                            else save_model_dir
+                        ),
+                        logger,
+                        config,
+                        is_best=True,
+                        prefix=prefix,
+                        ema=ema,
+                        train_state=_ema_train_state,
+                        save_model_info=model_info,
+                        best_model_dict=best_model_dict,
+                        epoch=epoch,
+                        global_step=global_step,
+                    )
+                best_str = "best metric, {}".format(
+                    ", ".join(
+                        ["{}: {}".format(k, v) for k, v in best_model_dict.items()]
+                    )
+                )
+                logger.info(best_str)
+                # logger best metric
+                if log_writer is not None:
+                    log_writer.log_metrics(
+                        metrics={
+                            "best_{}".format(main_indicator): best_model_dict[
+                                main_indicator
+                            ]
+                        },
+                        prefix="EVAL",
+                        step=global_step,
+                    )
+
+                    log_writer.log_model(
+                        is_best=True, prefix="best_accuracy", metadata=best_model_dict
+                    )
+
+                # Restore training weights after eval/save
+                if _ema_train_state is not None:
+                    model.set_state_dict(_ema_train_state)
+
+            reader_start = time.time()
+        if dist.get_rank() == 0:
+            prefix = "latest"
+            # Apply EMA weights for save
+            _ema_train_state_latest = None
+            if ema is not None:
+                _ema_train_state_latest = copy.deepcopy(model.state_dict())
+                model.set_state_dict(ema.apply())
+            if uniform_output_enabled:
+                export(config, model, os.path.join(save_model_dir, prefix, "inference"))
+                gc.collect()
+                model_info = {"epoch": epoch, "metric": best_model_dict}
+            else:
+                model_info = None
+            save_model(
+                model,
+                optimizer,
+                (
+                    os.path.join(save_model_dir, prefix)
+                    if uniform_output_enabled
+                    else save_model_dir
+                ),
+                logger,
+                config,
+                is_best=False,
+                prefix=prefix,
+                ema=ema,
+                train_state=_ema_train_state_latest,
+                save_model_info=model_info,
+                best_model_dict=best_model_dict,
+                epoch=epoch,
+                global_step=global_step,
+            )
+            # Restore training weights
+            if _ema_train_state_latest is not None:
+                model.set_state_dict(_ema_train_state_latest)
+
+            if log_writer is not None:
+                log_writer.log_model(is_best=False, prefix="latest")
+
+        if dist.get_rank() == 0 and epoch > 0 and epoch % save_epoch_step == 0:
+            prefix = "iter_epoch_{}".format(epoch)
+            # Apply EMA weights for save
+            _ema_train_state_iter = None
+            if ema is not None:
+                _ema_train_state_iter = copy.deepcopy(model.state_dict())
+                model.set_state_dict(ema.apply())
+            if uniform_output_enabled:
+                export(config, model, os.path.join(save_model_dir, prefix, "inference"))
+                gc.collect()
+                model_info = {"epoch": epoch, "metric": best_model_dict}
+            else:
+                model_info = None
+            save_model(
+                model,
+                optimizer,
+                (
+                    os.path.join(save_model_dir, prefix)
+                    if uniform_output_enabled
+                    else save_model_dir
+                ),
+                logger,
+                config,
+                is_best=False,
+                prefix=prefix,
+                ema=ema,
+                train_state=_ema_train_state_iter,
+                save_model_info=model_info,
+                best_model_dict=best_model_dict,
+                epoch=epoch,
+                global_step=global_step,
+                done_flag=epoch == config["Global"]["epoch_num"],
+            )
+            # Restore training weights
+            if _ema_train_state_iter is not None:
+                model.set_state_dict(_ema_train_state_iter)
+            if log_writer is not None:
+                log_writer.log_model(
+                    is_best=False, prefix="iter_epoch_{}".format(epoch)
+                )
+
+        # Reset reader_start so next epoch's first batch doesn't include
+        # save_model time in avg_reader_cost
+        reader_start = time.time()
+
+    best_str = "best metric, {}".format(
+        ", ".join(["{}: {}".format(k, v) for k, v in best_model_dict.items()])
+    )
+    logger.info(best_str)
+    if dist.get_rank() == 0 and log_writer is not None:
+        log_writer.close()
+    return
+
+
+def eval(
+    model,
+    valid_dataloader,
+    post_process_class,
+    eval_class,
+    model_type=None,
+    extra_input=False,
+    scaler=None,
+    amp_level="O2",
+    amp_custom_black_list=[],
+    amp_custom_white_list=[],
+    amp_dtype="float16",
+):
+    model.eval()
+    with paddle.no_grad():
+        total_frame = 0.0
+        total_time = 0.0
+        pbar = tqdm(
+            total=len(valid_dataloader), desc="eval model:", position=0, leave=True
+        )
+        max_iter = (
+            len(valid_dataloader) - 1
+            if platform.system() == "Windows"
+            else len(valid_dataloader)
+        )
+        sum_images = 0
+        for idx, batch in enumerate(valid_dataloader):
+            if idx >= max_iter:
+                break
+            images = batch[0]
+            start = time.time()
+
+            # use amp
+            if scaler:
+                with paddle.amp.auto_cast(
+                    level=amp_level,
+                    custom_black_list=amp_custom_black_list,
+                    dtype=amp_dtype,
+                ):
+                    if model_type == "table" or extra_input:
+                        preds = model(images, data=batch[1:])
+                    elif model_type in ["kie"]:
+                        preds = model(batch)
+                    elif model_type in ["can"]:
+                        preds = model(batch[:3])
+                    elif model_type in ["latexocr"]:
+                        preds = model(batch)
+                    elif model_type in ["sr"]:
+                        preds = model(batch)
+                        sr_img = preds["sr_img"]
+                        lr_img = preds["lr_img"]
+                    else:
+                        preds = model(images)
+                preds = to_float32(preds)
+            else:
+                if model_type == "table" or extra_input:
+                    preds = model(images, data=batch[1:])
+                elif model_type in ["kie"]:
+                    preds = model(batch)
+                elif model_type in ["can"]:
+                    preds = model(batch[:3])
+                elif model_type in ["latexocr", "unimernet", "pp_formulanet"]:
+                    preds = model(batch)
+                elif model_type in ["sr"]:
+                    preds = model(batch)
+                    sr_img = preds["sr_img"]
+                    lr_img = preds["lr_img"]
+                else:
+                    preds = model(images)
+
+            batch_numpy = []
+            for item in batch:
+                if isinstance(item, paddle.Tensor):
+                    batch_numpy.append(item.numpy())
+                else:
+                    batch_numpy.append(item)
+            # Obtain usable results from post-processing methods
+            total_time += time.time() - start
+            # Evaluate the results of the current batch
+            if model_type in ["table", "kie"]:
+                if post_process_class is None:
+                    eval_class(preds, batch_numpy)
+                else:
+                    post_result = post_process_class(preds, batch_numpy)
+                    eval_class(post_result, batch_numpy)
+            elif model_type in ["sr"]:
+                eval_class(preds, batch_numpy)
+            elif model_type in ["can"]:
+                eval_class(preds[0], batch_numpy[2:], epoch_reset=(idx == 0))
+            elif model_type in ["latexocr", "unimernet", "pp_formulanet"]:
+                post_result = post_process_class(preds, batch[1], "eval")
+                eval_class(post_result[0], post_result[1], epoch_reset=(idx == 0))
+            else:
+                post_result = post_process_class(preds, batch_numpy[1])
+                eval_class(post_result, batch_numpy)
+
+            pbar.update(1)
+            total_frame += len(images)
+            sum_images += 1
+        # Get final metric，eg. acc or hmean
+        metric = eval_class.get_metric()
+
+    pbar.close()
+    model.train()
+    # Avoid ZeroDivisionError
+    if total_time > 0:
+        metric["fps"] = total_frame / total_time
+    else:
+        metric["fps"] = 0  # or set to a fallback value
+    return metric
+
+
+def update_center(char_center, post_result, preds):
+    result, label = post_result
+    feats, logits = preds
+    logits = paddle.argmax(logits, axis=-1)
+    feats = feats.numpy()
+    logits = logits.numpy()
+
+    for idx_sample in range(len(label)):
+        if result[idx_sample][0] == label[idx_sample][0]:
+            feat = feats[idx_sample]
+            logit = logits[idx_sample]
+            for idx_time in range(len(logit)):
+                index = logit[idx_time]
+                if index in char_center.keys():
+                    char_center[index][0] = (
+                        char_center[index][0] * char_center[index][1] + feat[idx_time]
+                    ) / (char_center[index][1] + 1)
+                    char_center[index][1] += 1
+                else:
+                    char_center[index] = [feat[idx_time], 1]
+    return char_center
+
+
+def get_center(model, eval_dataloader, post_process_class):
+    pbar = tqdm(total=len(eval_dataloader), desc="get center:")
+    max_iter = (
+        len(eval_dataloader) - 1
+        if platform.system() == "Windows"
+        else len(eval_dataloader)
+    )
+    char_center = dict()
+    for idx, batch in enumerate(eval_dataloader):
+        if idx >= max_iter:
+            break
+        images = batch[0]
+        start = time.time()
+        preds = model(images)
+
+        batch = [item.numpy() for item in batch]
+        # Obtain usable results from post-processing methods
+        post_result = post_process_class(preds, batch[1])
+
+        # update char_center
+        char_center = update_center(char_center, post_result, preds)
+        pbar.update(1)
+
+    pbar.close()
+    for key in char_center.keys():
+        char_center[key] = char_center[key][0]
+    return char_center
+
+
+def preprocess(is_train=False):
+    FLAGS = ArgsParser().parse_args()
+    profiler_options = FLAGS.profiler_options
+    config = load_config(FLAGS.config)
+    config = merge_config(config, FLAGS.opt)
+    profile_dic = {"profiler_options": FLAGS.profiler_options}
+    config = merge_config(config, profile_dic)
+
+    if is_train:
+        # save_config
+        save_model_dir = config["Global"]["save_model_dir"]
+        os.makedirs(save_model_dir, exist_ok=True)
+        with open(os.path.join(save_model_dir, "config.yml"), "w") as f:
+            yaml.dump(dict(config), f, default_flow_style=False, sort_keys=False)
+        log_file = "{}/train.log".format(save_model_dir)
+    else:
+        log_file = None
+
+    log_ranks = config["Global"].get("log_ranks", "0")
+    logger = get_logger(log_file=log_file, log_ranks=log_ranks)
+
+    # check if set use_gpu=True in paddlepaddle cpu version
+    use_gpu = config["Global"].get("use_gpu", False)
+    use_xpu = config["Global"].get("use_xpu", False)
+    use_npu = config["Global"].get("use_npu", False)
+    use_mlu = config["Global"].get("use_mlu", False)
+    use_gcu = config["Global"].get("use_gcu", False)
+    use_metax_gpu = config["Global"].get("use_metax_gpu", False)
+
+    use_iluvatar_gpu = config["Global"].get("use_iluvatar_gpu", False)
+
+    alg = config["Architecture"]["algorithm"]
+    assert alg in [
+        "EAST",
+        "DB",
+        "SAST",
+        "Rosetta",
+        "CRNN",
+        "STARNet",
+        "RARE",
+        "SRN",
+        "CLS",
+        "PGNet",
+        "Distillation",
+        "NRTR",
+        "TableAttn",
+        "SAR",
+        "PSE",
+        "SEED",
+        "SDMGR",
+        "LayoutXLM",
+        "LayoutLM",
+        "LayoutLMv2",
+        "PREN",
+        "FCE",
+        "SVTR",
+        "SVTR_LCNet",
+        "ViTSTR",
+        "ABINet",
+        "DB++",
+        "TableMaster",
+        "SPIN",
+        "VisionLAN",
+        "Gestalt",
+        "SLANet",
+        "RobustScanner",
+        "CT",
+        "RFL",
+        "DRRG",
+        "CAN",
+        "Telescope",
+        "SATRN",
+        "SVTR_HGNet",
+        "ParseQ",
+        "CPPD",
+        "LaTeXOCR",
+        "UniMERNet",
+        "SLANeXt",
+        "PP-FormulaNet-S",
+        "PP-FormulaNet-L",
+        "PP-FormulaNet_plus-S",
+        "PP-FormulaNet_plus-M",
+        "PP-FormulaNet_plus-L",
+    ]
+
+    if use_xpu:
+        device = "xpu:{0}".format(os.getenv("FLAGS_selected_xpus", 0))
+    elif use_npu:
+        device = "npu:{0}".format(os.getenv("FLAGS_selected_npus", 0))
+    elif use_mlu:
+        device = "mlu:{0}".format(os.getenv("FLAGS_selected_mlus", 0))
+    elif use_gcu:  # Use Enflame GCU(General Compute Unit)
+        device = "gcu:{0}".format(os.getenv("FLAGS_selected_gcus", 0))
+    elif use_metax_gpu:  # Use Enflame GCU(General Compute Unit)
+        device = "metax:{0}".format(dist.ParallelEnv().dev_id)
+    elif use_iluvatar_gpu:
+        device = "iluvatar_gpu:{0}".format(dist.ParallelEnv().dev_id)
+    else:
+        device = "gpu:{}".format(dist.ParallelEnv().dev_id) if use_gpu else "cpu"
+    check_device(
+        use_gpu, use_xpu, use_npu, use_mlu, use_gcu, use_iluvatar_gpu, use_metax_gpu
+    )
+
+    device = paddle.set_device(device)
+
+    config["Global"]["distributed"] = dist.get_world_size() != 1
+
+    loggers = []
+
+    if "use_visualdl" in config["Global"] and config["Global"]["use_visualdl"]:
+        logger.warning(
+            "You are using VisualDL, the VisualDL is deprecated and "
+            "removed in ppocr!"
+        )
+        log_writer = None
+    if (
+        "use_wandb" in config["Global"] and config["Global"]["use_wandb"]
+    ) or "wandb" in config:
+        save_dir = config["Global"]["save_model_dir"]
+        wandb_writer_path = "{}/wandb".format(save_dir)
+        if "wandb" in config:
+            wandb_params = config["wandb"]
+        else:
+            wandb_params = dict()
+        wandb_params.update({"save_dir": save_dir})
+        log_writer = WandbLogger(**wandb_params, config=config)
+        loggers.append(log_writer)
+    else:
+        log_writer = None
+    print_dict(config, logger)
+
+    if loggers:
+        log_writer = Loggers(loggers)
+    else:
+        log_writer = None
+
+    logger.info("train with paddle {} and device {}".format(paddle.__version__, device))
+    return config, device, logger, log_writer
diff --git a/tools/resolve_doc_github_refs.py b/tools/resolve_doc_github_refs.py
new file mode 100644
index 0000000..3bde8d9
--- /dev/null
+++ b/tools/resolve_doc_github_refs.py
@@ -0,0 +1,49 @@
+#!/usr/bin/env python3
+import argparse
+from pathlib import Path
+
+
+TEXT_SUFFIXES = {".md", ".yml", ".yaml"}
+
+
+def resolve_placeholders(root, placeholder, source_ref):
+    if (
+        not source_ref
+        or source_ref.strip() != source_ref
+        or any(c.isspace() for c in source_ref)
+    ):
+        raise ValueError("source_ref must be a non-empty ref without whitespace")
+
+    root = Path(root)
+    changed = []
+    for path in sorted(root.rglob("*")):
+        if not path.is_file() or path.suffix not in TEXT_SUFFIXES:
+            continue
+        content = path.read_text(encoding="utf-8")
+        if placeholder not in content:
+            continue
+        path.write_text(content.replace(placeholder, source_ref), encoding="utf-8")
+        changed.append(path)
+    return changed
+
+
+def main(argv=None):
+    parser = argparse.ArgumentParser(
+        description="Resolve docs GitHub source-ref placeholders before building docs."
+    )
+    parser.add_argument("--root", default="docs", help="Directory to rewrite.")
+    parser.add_argument("--placeholder", required=True)
+    parser.add_argument("--source-ref", required=True)
+    args = parser.parse_args(argv)
+
+    changed = resolve_placeholders(
+        args.root,
+        placeholder=args.placeholder,
+        source_ref=args.source_ref,
+    )
+    print(f"Resolved {len(changed)} file(s) under {args.root}.")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/tools/test_hubserving.py b/tools/test_hubserving.py
new file mode 100755
index 0000000..5480107
--- /dev/null
+++ b/tools/test_hubserving.py
@@ -0,0 +1,162 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.append(os.path.abspath(os.path.join(__dir__, "..")))
+
+from ppocr.utils.logging import get_logger
+
+logger = get_logger()
+
+import cv2
+import numpy as np
+import time
+from PIL import Image
+from ppocr.utils.utility import get_image_file_list
+from tools.infer.utility import draw_ocr, draw_boxes, str2bool
+from ppstructure.utility import draw_structure_result
+from ppstructure.predict_system import to_excel
+
+import requests
+import json
+import base64
+
+
+def cv2_to_base64(image):
+    return base64.b64encode(image).decode("utf8")
+
+
+def draw_server_result(image_file, res):
+    img = cv2.imread(image_file)
+    image = Image.fromarray(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
+    if len(res) == 0:
+        return np.array(image)
+    keys = res[0].keys()
+    if "text_region" not in keys:  # for ocr_rec, draw function is invalid
+        logger.info("draw function is invalid for ocr_rec!")
+        return None
+    elif "text" not in keys:  # for ocr_det
+        logger.info("draw text boxes only!")
+        boxes = []
+        for dno in range(len(res)):
+            boxes.append(res[dno]["text_region"])
+        boxes = np.array(boxes)
+        draw_img = draw_boxes(image, boxes)
+        return draw_img
+    else:  # for ocr_system
+        logger.info("draw boxes and texts!")
+        boxes = []
+        texts = []
+        scores = []
+        for dno in range(len(res)):
+            boxes.append(res[dno]["text_region"])
+            texts.append(res[dno]["text"])
+            scores.append(res[dno]["confidence"])
+        boxes = np.array(boxes)
+        scores = np.array(scores)
+        draw_img = draw_ocr(image, boxes, texts, scores, draw_txt=True, drop_score=0.5)
+        return draw_img
+
+
+def save_structure_res(res, save_folder, image_file):
+    img = cv2.imread(image_file)
+    excel_save_folder = os.path.join(save_folder, os.path.basename(image_file))
+    os.makedirs(excel_save_folder, exist_ok=True)
+    # save res
+    with open(os.path.join(excel_save_folder, "res.txt"), "w", encoding="utf8") as f:
+        for region in res:
+            if region["type"] == "Table":
+                excel_path = os.path.join(
+                    excel_save_folder, "{}.xlsx".format(region["bbox"])
+                )
+                to_excel(region["res"], excel_path)
+            elif region["type"] == "Figure":
+                x1, y1, x2, y2 = region["bbox"]
+                print(region["bbox"])
+                roi_img = img[y1:y2, x1:x2, :]
+                img_path = os.path.join(
+                    excel_save_folder, "{}.jpg".format(region["bbox"])
+                )
+                cv2.imwrite(img_path, roi_img)
+            else:
+                for text_result in region["res"]:
+                    f.write("{}\n".format(json.dumps(text_result)))
+
+
+def main(args):
+    image_file_list = get_image_file_list(args.image_dir)
+    is_visualize = False
+    headers = {"Content-type": "application/json"}
+    cnt = 0
+    total_time = 0
+    for image_file in image_file_list:
+        img = open(image_file, "rb").read()
+        if img is None:
+            logger.info("error in loading image:{}".format(image_file))
+            continue
+        img_name = os.path.basename(image_file)
+        # seed http request
+        starttime = time.time()
+        data = {"images": [cv2_to_base64(img)]}
+        r = requests.post(url=args.server_url, headers=headers, data=json.dumps(data))
+        elapse = time.time() - starttime
+        total_time += elapse
+        logger.info("Predict time of %s: %.3fs" % (image_file, elapse))
+        res = r.json()["results"][0]
+        logger.info(res)
+
+        if args.visualize:
+            draw_img = None
+            if "structure_table" in args.server_url:
+                to_excel(res["html"], "./{}.xlsx".format(img_name))
+            elif "structure_system" in args.server_url:
+                save_structure_res(res["regions"], args.output, image_file)
+            else:
+                draw_img = draw_server_result(image_file, res)
+            if draw_img is not None:
+                if not os.path.exists(args.output):
+                    os.makedirs(args.output)
+                cv2.imwrite(
+                    os.path.join(args.output, os.path.basename(image_file)),
+                    draw_img[:, :, ::-1],
+                )
+                logger.info(
+                    "The visualized image saved in {}".format(
+                        os.path.join(args.output, os.path.basename(image_file))
+                    )
+                )
+        cnt += 1
+        if cnt % 100 == 0:
+            logger.info("{} processed".format(cnt))
+    logger.info("avg time cost: {}".format(float(total_time) / cnt))
+
+
+def parse_args():
+    import argparse
+
+    parser = argparse.ArgumentParser(description="args for hub serving")
+    parser.add_argument("--server_url", type=str, required=True)
+    parser.add_argument("--image_dir", type=str, required=True)
+    parser.add_argument("--visualize", type=str2bool, default=False)
+    parser.add_argument("--output", type=str, default="./hubserving_result")
+    args = parser.parse_args()
+    return args
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    main(args)
diff --git a/tools/train.py b/tools/train.py
new file mode 100755
index 0000000..fc2b49b
--- /dev/null
+++ b/tools/train.py
@@ -0,0 +1,301 @@
+# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
+import os
+import sys
+
+__dir__ = os.path.dirname(os.path.abspath(__file__))
+sys.path.append(__dir__)
+sys.path.insert(0, os.path.abspath(os.path.join(__dir__, "..")))
+
+import yaml
+import paddle
+import paddle.distributed as dist
+
+from ppocr.data import build_dataloader, set_signal_handlers
+from ppocr.modeling.architectures import build_model
+from ppocr.losses import build_loss
+from ppocr.optimizer import build_optimizer
+from ppocr.postprocess import build_post_process
+from ppocr.metrics import build_metric
+from ppocr.utils.save_load import load_model
+from ppocr.utils.utility import set_seed
+from ppocr.modeling.architectures import apply_to_static
+import tools.program as program
+import tools.naive_sync_bn as naive_sync_bn
+
+dist.get_world_size()
+
+
+def main(config, device, logger, vdl_writer):
+    # init dist environment
+    if config["Global"]["distributed"]:
+        dist.init_parallel_env()
+
+    global_config = config["Global"]
+
+    # build dataloader
+    # NOTE: Do NOT pass seed here. The seed parameter in build_dataloader is used
+    # as epoch number by set_epoch_as_seed (for adaptive shrink_ratio), not as
+    # random seed. First construction should use epoch=0 (i.e., seed=None).
+    # The epoch loop in program.train() handles subsequent updates via
+    # reset_data_lines(seed=epoch).
+    set_signal_handlers()
+    train_dataloader = build_dataloader(config, "Train", device, logger)
+    if len(train_dataloader) == 0:
+        logger.error(
+            "No Images in train dataset, please ensure\n"
+            + "\t1. The images num in the train label_file_list should be larger than or equal with batch size.\n"
+            + "\t2. The annotation file and path in the configuration file are provided normally."
+        )
+        return
+
+    if config["Eval"]:
+        valid_dataloader = build_dataloader(config, "Eval", device, logger)
+    else:
+        valid_dataloader = None
+    step_pre_epoch = len(train_dataloader)
+
+    # build post process
+    post_process_class = build_post_process(config["PostProcess"], global_config)
+
+    # build model
+    # for rec algorithm
+    if hasattr(post_process_class, "character"):
+        char_num = len(getattr(post_process_class, "character"))
+        if config["Architecture"]["algorithm"] in [
+            "Distillation",
+        ]:  # distillation model
+            for key in config["Architecture"]["Models"]:
+                if (
+                    config["Architecture"]["Models"][key]["Head"]["name"] == "MultiHead"
+                ):  # for multi head
+                    if config["PostProcess"]["name"] == "DistillationSARLabelDecode":
+                        char_num = char_num - 2
+                    if config["PostProcess"]["name"] == "DistillationNRTRLabelDecode":
+                        char_num = char_num - 3
+                    out_channels_list = {}
+                    out_channels_list["CTCLabelDecode"] = char_num
+                    # update SARLoss params
+                    if (
+                        list(config["Loss"]["loss_config_list"][-1].keys())[0]
+                        == "DistillationSARLoss"
+                    ):
+                        config["Loss"]["loss_config_list"][-1]["DistillationSARLoss"][
+                            "ignore_index"
+                        ] = (char_num + 1)
+                        out_channels_list["SARLabelDecode"] = char_num + 2
+                    elif any(
+                        "DistillationNRTRLoss" in d
+                        for d in config["Loss"]["loss_config_list"]
+                    ):
+                        out_channels_list["NRTRLabelDecode"] = char_num + 3
+
+                    config["Architecture"]["Models"][key]["Head"][
+                        "out_channels_list"
+                    ] = out_channels_list
+                else:
+                    config["Architecture"]["Models"][key]["Head"][
+                        "out_channels"
+                    ] = char_num
+        elif config["Architecture"]["Head"]["name"] == "MultiHead":  # for multi head
+            if config["PostProcess"]["name"] == "SARLabelDecode":
+                char_num = char_num - 2
+            if config["PostProcess"]["name"] == "NRTRLabelDecode":
+                char_num = char_num - 3
+            out_channels_list = {}
+            out_channels_list["CTCLabelDecode"] = char_num
+            # update SARLoss params
+            if list(config["Loss"]["loss_config_list"][1].keys())[0] == "SARLoss":
+                if config["Loss"]["loss_config_list"][1]["SARLoss"] is None:
+                    config["Loss"]["loss_config_list"][1]["SARLoss"] = {
+                        "ignore_index": char_num + 1
+                    }
+                else:
+                    config["Loss"]["loss_config_list"][1]["SARLoss"]["ignore_index"] = (
+                        char_num + 1
+                    )
+                out_channels_list["SARLabelDecode"] = char_num + 2
+            elif list(config["Loss"]["loss_config_list"][1].keys())[0] == "NRTRLoss":
+                out_channels_list["NRTRLabelDecode"] = char_num + 3
+            config["Architecture"]["Head"]["out_channels_list"] = out_channels_list
+        else:  # base rec model
+            config["Architecture"]["Head"]["out_channels"] = char_num
+
+        if config["PostProcess"]["name"] == "SARLabelDecode":  # for SAR model
+            config["Loss"]["ignore_index"] = char_num - 1
+
+    model = build_model(config["Architecture"])
+
+    use_sync_bn = config["Global"].get("use_sync_bn", False)
+    if use_sync_bn:
+        if config["Global"].get("use_npu", False) or config["Global"].get(
+            "use_xpu", False
+        ):
+            naive_sync_bn.convert_syncbn(model)
+        else:
+            model = paddle.nn.SyncBatchNorm.convert_sync_batchnorm(model)
+        logger.info("convert_sync_batchnorm")
+
+    model = apply_to_static(model, config, logger)
+
+    # build loss
+    loss_class = build_loss(config["Loss"])
+
+    # build optim
+    optimizer, lr_scheduler, wd_scheduler = build_optimizer(
+        config["Optimizer"],
+        epochs=config["Global"]["epoch_num"],
+        step_each_epoch=len(train_dataloader),
+        model=model,
+    )
+
+    # build metric
+    eval_class = build_metric(config["Metric"])
+
+    logger.info("train dataloader has {} iters".format(len(train_dataloader)))
+    if valid_dataloader is not None:
+        logger.info("valid dataloader has {} iters".format(len(valid_dataloader)))
+
+    use_amp = config["Global"].get("use_amp", False)
+    amp_level = config["Global"].get("amp_level", "O2")
+    amp_dtype = config["Global"].get("amp_dtype", "float16")
+    amp_custom_black_list = config["Global"].get("amp_custom_black_list", [])
+    amp_custom_white_list = config["Global"].get("amp_custom_white_list", [])
+    if os.path.exists(
+        os.path.join(config["Global"]["save_model_dir"], "train_result.json")
+    ):
+        try:
+            os.remove(
+                os.path.join(config["Global"]["save_model_dir"], "train_result.json")
+            )
+        except:
+            pass
+    if use_amp:
+        AMP_RELATED_FLAGS_SETTING = {}
+        if paddle.is_compiled_with_cuda():
+            AMP_RELATED_FLAGS_SETTING.update(
+                {
+                    "FLAGS_cudnn_batchnorm_spatial_persistent": 1,
+                    "FLAGS_gemm_use_half_precision_compute_type": 0,
+                }
+            )
+        paddle.set_flags(AMP_RELATED_FLAGS_SETTING)
+        scale_loss = config["Global"].get("scale_loss", 1.0)
+        use_dynamic_loss_scaling = config["Global"].get(
+            "use_dynamic_loss_scaling", False
+        )
+        scaler = paddle.amp.GradScaler(
+            init_loss_scaling=scale_loss,
+            use_dynamic_loss_scaling=use_dynamic_loss_scaling,
+        )
+        if amp_level == "O2":
+            model, optimizer = paddle.amp.decorate(
+                models=model,
+                optimizers=optimizer,
+                level=amp_level,
+                master_weight=True,
+                dtype=amp_dtype,
+            )
+    else:
+        scaler = None
+
+    # build EMA (after AMP decorate, before load_model)
+    ema = None
+    if config["Global"].get("use_ema", False):
+        from ppocr.utils.ema import ModelEMA
+
+        ema = ModelEMA(
+            model,
+            decay=config["Global"].get("ema_decay", 0.9998),
+            gamma=config["Global"].get("ema_gamma", 2000),
+            ema_decay_type=config["Global"].get("ema_decay_type", "threshold"),
+            ema_filter_no_grad=config["Global"].get("ema_filter_no_grad", False),
+        )
+        logger.info(
+            "EMA enabled: decay={}, gamma={}, type={}, filter_no_grad={}".format(
+                ema.decay,
+                ema.gamma,
+                ema.ema_decay_type,
+                config["Global"].get("ema_filter_no_grad", False),
+            )
+        )
+
+    # load pretrain model
+    pre_best_model_dict = load_model(
+        config, model, optimizer, config["Architecture"]["model_type"], ema=ema
+    )
+
+    if config["Global"]["distributed"]:
+        find_unused_parameters = config["Global"].get("find_unused_parameters", False)
+        model = paddle.DataParallel(
+            model, find_unused_parameters=find_unused_parameters
+        )
+    # start train
+    program.train(
+        config,
+        train_dataloader,
+        valid_dataloader,
+        device,
+        model,
+        loss_class,
+        optimizer,
+        lr_scheduler,
+        post_process_class,
+        eval_class,
+        pre_best_model_dict,
+        logger,
+        step_pre_epoch,
+        vdl_writer,
+        scaler,
+        amp_level,
+        amp_custom_black_list,
+        amp_custom_white_list,
+        amp_dtype,
+        wd_scheduler=wd_scheduler,
+        ema=ema,
+    )
+
+
+def test_reader(config, device, logger):
+    loader = build_dataloader(config, "Train", device, logger)
+    import time
+
+    starttime = time.time()
+    count = 0
+    try:
+        for data in loader():
+            count += 1
+            if count % 1 == 0:
+                batch_time = time.time() - starttime
+                starttime = time.time()
+                logger.info(
+                    "reader: {}, {}, {}".format(count, len(data[0]), batch_time)
+                )
+    except Exception as e:
+        logger.info(e)
+    logger.info("finish reader: {}, Success!".format(count))
+
+
+if __name__ == "__main__":
+    config, device, logger, vdl_writer = program.preprocess(is_train=True)
+    seed = config["Global"]["seed"] if "seed" in config["Global"] else 1024
+    set_seed(seed)
+    main(config, device, logger, vdl_writer)
+    # test_reader(config, device, logger)
diff --git a/train.sh b/train.sh
new file mode 100644
index 0000000..0a64150
--- /dev/null
+++ b/train.sh
@@ -0,0 +1,2 @@
+# recommended paddle.__version__ == 3.0.0
+python3 -m paddle.distributed.launch --log_dir=./log/ --gpus '0,1,2,3,4,5,6,7'  tools/train.py -c configs/rec/PP-OCRv5/PP-OCRv5_server_rec.yml